hep-th/0303060

AEI 2003-028

The Dilatation Operator of Conformal

N = 4 Super Yang-Mills Theory
arXiv:hep-th/0303060v4 3 Sep 2004

N. Beisert, C. Kristjansen and M. Staudacher

Max-Planck-Institut fur Gravitationsphysik

Albert-Einstein-Institut
Am Muhlenberg 1, D-14476 Golm, Germany
nbeisert,kristjan,matthias@aei.mpg.de

Abstract

We argue that existing methods for the perturbative computation of anomalous

dimensions and the disentanglement of mixing in N = 4 gauge theory can be
considerably simplified, systematized and extended by focusing on the theorys
dilatation operator. The efficiency of the method is first illustrated at the one-
loop level for general non-derivative scalar states. We then go on to derive, for
pure scalar states, the two-loop structure of the dilatation operator. This allows us
to obtain a host of new results. Among these are an infinite number of previously
unknown two-loop anomalous dimensions, new subtleties concerning t Hoofts
large N expansion due to mixing effects of degenerate single and multiple trace
states, two-loop tests of various protected operators, as well as two-loop non-
planar results for two-impurity operators in BMN gauge theory. We also put to
use the recently discovered integrable spin chain description of the planar one-loop
dilatation operator and show that the associated Yang-Baxter equation explains
the existence of a hitherto unknown planar axial symmetry between infinitely
many gauge theory states. We present evidence that this integrability can be
extended to all loops, with intriguing consequences for gauge theory, and that
it leads to a novel integrable deformation of the XXX Heisenberg spin chain.
Assuming that the integrability structure extends to more than two loops, we
determine the planar three-loop contribution to the dilatation operator.
1 Introduction and overview
Conformal quantum field theories have been fascinating theoretical physicists for a long
time. In two dimensions they are arguably the most important class of field theories:
Mathematically, they are quite tractable once one puts to full use the fact that the
conformal algebra has an infinite number of generators. In consequence many exact
results on the spectrum of operators and the structure of correlation functions may
be derived. Apart form their pivotal importance to string theory, they are physically
of great value due to their relationship with critical phenomena and integrable models
of statistical mechanics. For example, in many cases the representation theory of 2D
conformal field theories fixes the scaling dimensions of local operators, which in turn
are often related to critical exponents of experimentally relevant systems of solid states
physics.
In four dimensions conformal symmetry was long believed to play only a minor role.
The group has only finitely many generators, the QFTs relevant to particle physics
are certainly not conformal, and the only, trivial, example seemed to be free, massless
field theory. However, after the discovery of supersymmetry, it has become clear that
supersymmetric gauge theories can be exactly conformally invariant on the quantum
level, and in many cases the phase diagram of such gauge theories contained conformal
points or regions, quite analogously to phase diagrams in two dimensions. There is
therefore obvious theoretical interest in increasing our understanding of these phases.
In particular, the 4D gauge theory with the maximum possible number N = 4 of rigid
supersymmetries, discovered in 1976 [1, 2], has a superconformal phase [35].
Supersymmetric gauge theories are intimately connected to superstring theory. In
fact, the N = 4 action was originally discovered [1] by considering the low-energy limit
of superstrings. Surprisingly, with the advent of the AdS/CFT correspondence (see [6,7]
for comprehensive reviews) it was argued that the N = 4 gauge theory is, via duality,
a superstring theory in a particular background. This conjecture woke the Sleeping
Beauty and resulted in a large number of investigations into the structure of N = 4. In
particular the representation theory of the superconformal symmetry group SU(2, 2|4) [8]
was investigated more closely [9, 10], and numerous non-renormalization theorems were
derived, see e.g. [11]. In addition, some unexpected non-renormalization theorems that
do not follow from SU(2, 2|4) representation theory were found [12]. Once thought to
be somewhat boring, it gradually became clear that conformal N = 4 is an extremely
rich and non-trivial theory with many hidden secrets. We will see more examples for its
intricate structure in the present paper.
While many of the recently obtained new results on N = 4 in principle could have
been derived a long time ago, the AdS/CFT correspondence clearly helped immensely
in formulating the right questions. The latest chapter in exploiting this useful duality
involved taking a certain limit on both sides of the correspondence: On the string side,
after a process termed Penrose contraction of the AdS space, one obtains a new max-
imally supersymmetric background for the IIB superstring [13, 14] which has the great
advantage that the spectrum of free, massive string modes can be found exactly [15, 16].
On the gauge theory side, one considers the so-called BMN limit [17]. It involves studying
infinite sequences of operators containing an increasing number of fields. The relation-

1
ship to plane wave string excitations is then made by comparing the conformal scaling
dimension of these high-dimension operators to the masses of string excitations [17].
One therefore had to develop techniques to efficiently compute in perturbation theory
the scaling dimensions of operators containing an arbitrary number of fields. This was
done in various papers, on the planar [1721] and non-planar level [2228], extending
earlier work on protected half BPS [2932] and quarter BPS operators [33]. In [34] it was
realized, following important insights by [35,36] (see also [37]) that these well-established
techniques can be considerably simplified and extended, as we will now explain.
The standard way to find the scaling dimensions of a set of conformal fields O ,
employed by all of the just mentioned papers, is to consider the two-point functions


O (x) O (0) = 2 . (1.1)
|x|

Here the form of the two-point function is determined by conformal symmetry alone.
Finding the scaling dimension of a conformal field O is more subtle and, in general,
requires an understanding of the dynamics of the theory. In the free field limit (i.e. zero
coupling constant gYM = 0) equals the naive classical (tree level) scaling dimension ob-
tained by standard power counting. For weak coupling gYM , we can compute corrections
to the naive dimension by perturbation theory. In practice, extracting these corrections
from eq.(1.1) is somewhat tedious and fraught with various technical complications. For
one, while the corrections are perfectly finite numbers, the perturbative computation of
two-point functions such as eq.(1.1) leads to spurious infinities, requiring renormaliza-
tion1 . Secondly, the definition of the scaling dimensions tacitly assumes that we have
already found the correct conformal operators O . However, if one starts with a set
of naive operators O with the same engineering (tree-level) dimension one generically
encounters the phenomenon of mixing: The two-point function is not diagonal in , ,
and one rather has a matrix hO (x)O (0)i since a generic field does not have a definite
scaling dimension. It therefore seems that we have to diagonalize the two-point functions,
after renormalization, order by order in perturbation theory.
Here we would like to shift the attention away from the two-point functions eq.(1.1)
and rather focus on the dilatation operator acting on states at the origin of space-time
(in a radial quantization scheme) as was already done, for one-loop, in [37] (on the planar
level) and in [34] (in the BMN limit). Its eigensystem consists of the eigenvalues and
the eigenstates O . One thus has

D O = O . (1.2)

By way of example we will restrict the discussion to fields solely composed of the six
(a)
scalar fields n = n T a (n = 1, . . . , 6, a = 1, . . . , N 2 1) of the N = 4 SU(N) gauge
theory, but the extension to a larger class of fields is clearly possible. (Our notation and
conventions are explained in App. A.) The classical dilatation operator D0 then simply
1
An alternative method consists in extracting the anomalous dimensions from four-point functions.
This is how e.g. the two-loop anomalous dimension of the Konishi field was first found [38,39]. However,
this is not really different, as one can look at these four-point functions, via a double operator product
expansion, as a generator of two-point functions regulated by a different method, namely point-splitting.

2
counts the total number of scalar fields, as the engineering dimension of scalars is one.
This can be formally written as

D0 = Tr m m , (1.3)
2
where we introduced the notation for the variation with respect to the field

m = = T a (a) . (1.4)
m m
In an interacting theory the scaling dimensions and therefore also the dilatation generator
depend on the coupling constant. In perturbation theory the dilatation generator can
be expanded in powers of the coupling constant
 2 k
X gYM
D= D2k , (1.5)
k=0
16 2

and we shall denote D2k as the k-loop dilatation generator. The full one-loop contribution
D2 was first worked out, in the disguise of an effective vertex, in appendix C of [25]
(but see also [22, 23, 26, 20, 37, 34]): 3

D2 = :Tr[m , n ][m , n ]: 12 :Tr[m , n ][m , n ]:. (1.6)

However, it was shown in [34] that one can bypass the consideration of two-point functions
and directly diagonalize D2 in any convenient, linearly independent and complete basis.
In eq.(1.6) the normal ordering notation indicates that the derivatives do not act on the
fields enclosed by : :. To see how D2 acts, let us consider the simplest scalar fields of
the theory, which have classical dimension two. One has the protected chiral primaries
Qnm as well as the Konishi field K

Qnm = Tr n m 16 nm Tr k k and K = Tr k k . (1.7)

One immediately verifies4 , using eqs.(1.3),(1.4),(1.5),(1.6) that, to one loop, as D2 Qnm =

0 and D2 K = 12N K,
2
3gYM N
Q = 2 and K = 2 + 2
, (1.8)
4
that is the one-loop anomalous dimension of Qnm vanishes as expected, while the anoma-
lous dimension of the Konishi scalar agrees with the well-known result [40].
2
In the language of canonical quantization the field and the variation correspond to creation
and annihilation operators, respectively. The dilatation generator corresponds to the Hamiltonian and
a vacuum expectation value is obtained by setting the fields to zero, . . . |=0 .
3
Note that this expression is perfectly valid also for SO(N ), Sp(N ) and exceptional gauge groups.
4
The fission rule Tr Am Bn = mn Tr A Tr B and the fusion rule Tr Am Tr n B = mn Tr AB are
useful when calculating the action of D2 . These are the U(N ) rules but due to the commutators in D2
there is no difference when using the SU(N ) rules.

3
 Let us next look at a less trivial example and consider the following set of unprotected
SO(6) invariant operators of classical dimension four, spanning a four-dimensional space
of states:
Tr m n Tr m n
Tr m m Tr n n
O= Tr m n m n .
(1.9)
Tr m m n n
For the convenience of the reader we are optically separating double and single trace
sectors by solid lines. This case was studied in detail in [41], using the standard procedure
(see also [32]) of directly analyzing, employing N = 1 superspace Feynman rules, the set
of two-point functions of the four composite fields, cf. our above discussion surrounding
eq.(1.1). Using eq.(1.6), we find the following action of the one-loop dilatation generator

4 20 20

0 N N
0 24 24 24
D2 = N N N . (1.10)
24 4 8 4
N N
N4 14
N
4 18
The matrix notation indicates the fields that are generated by the action of D2 from the
ones in eq.(1.9), e.g.

D2 Tr m n Tr m n = 0 Tr m n Tr m n + 4N Tr m m Tr n n
20 Tr m n m n + 20 Tr m m n n . (1.11)

We see that the states in eq.(1.9) are not eigenstates of the dilatation operator: The
latter linearly mixes all operators in the four-dimensional vector space [41]. However,
diagonalizing the dilatation matrix D2 (we will throughout this paper take the freedom
of using the same notation D2k for both the dilatation operator and its matrix represen-
tations) in eq.(1.10) will in one go resolve the mixing problem and yield the anomalous
dimensions: The conformal operators are given, up to normalization, by the eigenvectors
of eq.(1.10) (in the basis eq.(1.9)), and their respective one-loop anomalous dimensions
correspond to the eigenvalues of eq.(1.10). The latter are obtained by finding the roots
of the characteristic polynomial det( 21 D2 ):
   
4 3 160 2 1760 7680
25 + 188 2 384 2
= 0. (1.12)
N N N2

The dimensions of the four conformal operators are

2
gYM N
=4+ , (1.13)
8 2
with being the four different roots of the above quartic equation. They are easily seen
to agree with the ones obtained in [41]. Comparison with the methodology of [41] clearly
demonstrates the great simplifications resulting from the direct use of the dilatation

4
operator5 .
In this paper we will find the two-loop contribution D4 to the scalar dilatation op-
erator D eq.(1.5), with the restriction to operators in the SO(6) representations [p, q, p]
with tree-level scaling dimension 2p + q. These fields may always be constructed out
of two complex scalars, say Z = 12 (5 + i6 ) and = 12 (1 + i2 ). We will argue
that operators composed by these two fields form a closed set under dilatation. The
idea is then to inspect all Feynman diagrams contributing to the dilatation operator and
write down the possible linear combinations of terms contributing to D4 . We fix some
unknown constants by using gauge invariance, and proven two-loop non-renormalization
theorems [30, 31]. To save us extra work, we finally use a result conjectured in the con-
text of BMN gauge theory in [42] and derived in [18, 43] to determine the remaining
coefficients. All this is done without explicitly working out a single Feynman diagram!
One finds

D0 = Tr Z Z + Tr ,
D2 = 2 :Tr[, Z][, Z]:,
  
D4 = 2 :Tr [, Z], Z [, Z], Z :
  
2 :Tr [, Z], [, Z], :
2 :Tr [, Z], T a [, Z], T a :,
  
(1.14)

where we have also explicitly rewritten the action of D0 and D2 on two complex scalars.
Let us check eq.(1.14) and apply our result to work out the two-loop dimension of the
Konishi field. We cannot directly apply D4 to K in eq.(1.7) as we have not yet worked
out the terms relevant to scalar fields containing SO(6) traces. However, it is well-known
that the following field is a Konishi descendant, with identical anomalous correction to
its classical dimension four:
K = Tr[, Z][, Z]. (1.15)
Applying eq.(1.14) to K we find
2 4
3gYM N 3gYM N2
K = 4 + . (1.16)
4 2 16 4
This agrees with the results of [38, 39].
Recently it was discovered by Minahan and Zarembo that properties of pure scalar
states in the planar, one-loop sector of N = 4 Super Yang-Mills theory can be described
by an integrable SO(6) spin chain [37]: The planar version of the one-loop dilatation
operator maps onto the spin chain Hamiltonian. Interestingly, we are able to derive
from this spin chain picture new results for the gauge theory. The integrability ensures,
via the existence of an R-matrix satisfying the Yang-Baxter equation, the existence of
5
Incidentally, one easily verifies that our dilatation matrix D2 in eq.(1.10) is, up to an overall constant,
precisely given by the product of the one-loop mixing matrix eq.(5.4) of [41] times the inverse of the
tree-level mixing matrix eq.(5.3) of [41], in agreement with the results of [34], where the relation between
the traditional method and the dilatation operator method was explained in detail. This furnishes a
nice, explicit example illustrating that much of the combinatorial complications of the usual method
cancel out from the physics.

5
further conserved charges in addition to the Hamiltonian. Here we shall find that the
first non-trivial conserved charge of the spin chain translates into an axial symmetry
of N = 4 Super Yang-Mills theory at N = . The latter transforms between states of
opposite parity with respect to reflection of color traces.
What is more, this symmetry turns out to be conserved by the two loop interactions
as well. This might indicate that N = 4 gauge theory can be used as a source for new
integrable spin chain Hamiltonians with interactions more subtle than just the standard
nearest neighbor ones: At one-loop, we have the standard XXX1/2 Heisenberg spin chain,
which is exactly integrable. Higher loop quantum corrections constitute deformations
2 2k
of the spin chain with a smooth deformation parameter gYM . At O(gYM ) the spin chain
2k
interactions involve k nearest neighbors. When all interactions up to O(gYM ) are included,
the models charges commute only up to terms of higher order. Therefore, in order for
the deformed model to still be exactly integrable, all quantum corrections (any k) have
to be taken into account. Our integrable deformation is thus, for finite gYM , non-local.
In a certain sense locality is recovered as gYM is tuned to zero.
Assuming the symmetry to hold at the three-loop level allows us to fix the planar
version of the three-loop dilatation operator for scalar operators of the type discussed
above. E.g. the resulting dilatation operator implies the following result for the three-
loop planar contribution to K :
2 4
3gYM N 3gYM N 2 21gYM
6
N3
K = 4 + + . (1.17)
4 2 16 4 256 6
Our paper is organized as follows. We start by recalling the derivation of the one-loop
dilatation generator in Sec. 2 and move on to determine its two-loop version in Sec. 3.
Section 4 is devoted to the application of the resulting dilatation generator to lower
dimensional operators and contains in particular a discussion of certain peculiarities of
the large-N expansion related to degeneracy and mixing of single and multiple trace
states. Next follows in Sec. 5 a treatment of two-impurity operators of arbitrary finite
dimension in the spirit of reference [20], leading in particular to an exact expression
for the planar two-loop anomalous dimension for any such operator. As a corollary we
show in Sec. 6 that the two-loop contribution to the quantum mechanical Hamiltonian
describing BMN gauge theory [34] can be simply expressed in terms of its one-loop
counterpart which suggests an all genera version of the celebrated all-loop BMN square
root formula [17, 18, 43]. In Sec. 7 we discuss the spin chain description of N = 4 Super
Yang-Mills theory [37] and point out certain consequences thereof, cf. the discussion
above. Finally, in Sec. 8 we discuss possible implications for arbitrarily high loop orders.
In particular, we speculate about how fully exploiting the integrability might lead to
exact planar anomalous dimensions. In addition we conjecture the existence of novel
integrable deformations of the standard Heisenberg spin chain. We furthermore call the
attention of the reader to App. D which contains a collection of previously unknown
anomalous dimensions for lower dimensional operators.

6
2 One-loop revisited
The generator of dilatations on scalar fields at one-loop is given by eq.(1.6). At one-
loop mixing between operators with different kinds of fields is irrelevant. Therefore (1.6)
alone determines the one-loop anomalous dimensions of operators whose leading piece is
purely made out of scalar fields.
In the following we will rederive this result to make the reader familiar with the
methods and transformations in a rather simple context. All these steps will reappear at
two-loops in a more involved fashion. We start by introducing the notation at tree-level.
We sketch the renormalization procedure at one-loop and apply it to obtain the one-loop
dilatation generator.

2.1 Tree-level
We obtain the anomalous dimensions of a set of operators O by formally evaluating
their two-point functions. For that we distinguish between fields at points x and 0 by
the superscript
+m = m (x),
m = m (0). (2.1)
The operators O are constructed as traces of scalar fields

O () = Tr m n m . . . Tr n . . . , O = O ( ), (2.2)

where enumerates the operators. The tree-level two-point function can formally be
written as
+ 
O O tree = exp W0 (x, + , ) O+ O =0 ,


(2.3)
where W0 is the tree-level Green function

W0 (x, + , ) = I0x Tr +
m m . (2.4)

The scalar propagator Ixy is defined in (B.1). In order for the result to be non-vanishing
all the scalar fields in O need to be contracted to fields + in O+ with the propagator
I0x . In particular the number of fields of the two operators must be equal.

2.2 Renormalization
For the one-loop correlator we insert the (connected) one-loop Green function W2 into
the tree-level correlator (2.3)

+ 
O O oneloop = exp W0 (x, + , ) 1 + g 2W2 (x, + , ) O+ O =0 ,



(2.5)

1
where g 2 = gYM2
/16 2 . We now change the argument + of W2 to I0x . This can be
done because the result vanishes unless every is contracted with some + before the

fields are set to zero. Here, the only possibility is to contract with a term in W0 which
effectively changes the argument back to + . In doing that we need to make sure that

7
no new contractions appear between the arguments and of W2 . Formally, this is
achieved by normal ordering : :. The correlator becomes

+ 
O O oneloop = exp W0 (x, + , ) 1 + g 2V2 (x) O+ O =0 ,



(2.6)

with the one-loop effective vertex

1
V2 (x) = :W2 (x, I0x , ):. (2.7)

Instead of replacing + we could have replaced . This shows that in (2.6) V2 is

equivalent to V2+ . In other words V2 is self-adjoint with respect to the tree-level scalar
product.
We renormalize the operators according to

O = 1 12 g 2 V2 (x0 ) O,


(2.8)

and find

+ 
O O oneloop = exp W0 (x, + , ) 1 + g 2V2 (x) g 2 V2 (x0 ) O+ O =0 .



(2.9)

In the present renormalizable field theory in dimensional regularization the dependence

of V2 on x is fixed, we write
(1 )
V2 (x) =  1  V. (2.10)
 2 x2  2
2
We send the regulator to zero and find

lim V2 (x) V2 (x0 ) = log x20 /x2 D2 ,

(2.11)
0

with
D2 = lim V2 . (2.12)
0

The final answer for the renormalized correlator at = 0 is

+ 
O O oneloop = exp(W0 ) exp log(x20 /x2 )g 2 D2 O+ O =0 ,


(2.13)

in agreement with conformal field theory. D2 is the one-loop correction to the dilatation
generator. Furthermore, the coefficient of the correlator is given by its tree-level value as
D2 multiplies the logarithm only. Although we are interested in correlators of renormal-
ized operators as on the left-hand side of (2.13), we can work with bare operators as on
the right hand side of (2.13). We will not distinguish between the two types of operators
O and O. It will be understood that an operator in a correlator is renormalized.

2.3 Evaluation of diagrams

The one-loop connected Green functions are depicted in Fig. 1. To evaluate them we
make use of the N = 4 SYM action in (A.1) with the coupling constant g which is

8
 a b c d1 d2 d3 e
Figure 1: One-loop Green functions. The diagrams display the combinatorial structure with
respect to the gauge group rather than their space-time configuration. The solid, wiggly, dashed
lines represent scalars, gluons, fermions, respectively. The dotted lines correspond to a local
interaction of four scalars/gluons.

related to the usual coupling constant gYM by g 2 = gYM

2
/16 2. The one-loop Green
functions evaluate to

W2,a = 41 X00xx Tr[+ +

m , n ][m , n ],
W2,b = 14 X00xx Tr[+ + + +


m , n ][m , n ] + Tr[m , n ][m , n ] ,

W2,c = 21 H0x,0x Y00x I0x + 41 X00xx Tr[+ +

m , m ][n , n ],
W2,d = Y00x Tr[+ a a
m , T ][T , m ]. (2.14)

The functions X, Y, H are defined in (B.2). Diagram e vanishes by antisymmetry, its

structure is Tr[+ +
m , m ][n , n ] = 0. We use a Jacobi-Identity to transform the second
structure in W2,b

Tr[+ + + + + +
m , n ][m , n ] = Tr[m , n ][m , n ] Tr[m , m ][n , n ]. (2.15)

We shuffle some terms around and get

1
Tr[+ + 1 + +


W2,A = X00xx 2 m , n ][m , n ] + 4 Tr[m , n ][m , n ] ,

W2,B = 21 H0x,0x Tr[+ +

m , m ][n , n ],
W2,C = Y00x I0x Tr[+ + + a a


m , m ][n , n ] + Tr[m , T ][T , m ] . (2.16)

According to (2.7) we write

1
2
:Tr[m , n ][m , n ]: + 14 :Tr[m , n ][m , n ]:


V2,A = X00xx I0x 2
,
V2,B = 21 H0x,0x I0x
2
:Tr[m , m ][n , n ]:,
1
:Tr[m , m ][n , n ]: + :Tr[m , T a ][T a , m ]: .


V2,C = Y00x I0x (2.17)

The last term can now be written as

1
V2,C = Y00x I0x Tr[m , m ][n , n ]
= Y00x I0x Tr [T a , m ]m Tr [T a , n ]n
1



1 a a
= Y00x I0x G G . (2.18)

The generator Ga = Tr[T a , m ]m is the generator of gauge transformations for scalar

fields. Therefore V2,C does not act on operators which are color singlets, it is irrelevant.

9
The remaining functions X and H have the following expansion in , see (B.3),
 
X00xx 2 2 (1 )
= + 2 + O( )  1  ,
2
I0x  2 x2 
2
H0x,0x
(1 )
= 48(3) + O(2 )  1  . (2.19)
2
I0x  2 x2 
2

When inserted in (2.12) this proves our initial statement about the one-loop dilatation
generator

D2 = D2,A = :Tr[m , n ][m , n ]: 12 :Tr[m , n ][m , n ]:. (2.20)

3 Two-loop
In this chapter we will derive the central technical result of the paper: We will find the
two-loop correction to the dilatation generator, as announced in eq.(1.14). Written in
SO(6) notation, it is given by

D4 = 2:Tr m [n , [m , [p , [n , p ]]]]: + :Tr m [n , [T a , [T a , [m , n ]]]]:. (3.1)

This result is valid for pure scalar operators. Here pure means those composite operators,
made from the six fundamental scalar fields of the model, that do not mix with other,
non-scalar types of fields (or fields containing covariant derivatives). These pure scalars
are given by the SO(6) representations [p, q, p] with 0 = 2p + q. We start with a group-
theoretical argument why the dilatation generator closes on this class of operators. We
proceed by extending the renormalization scheme to two-loops. Then we investigate the
allowed structures in D4 by considering the relevant Feynman diagrams. Finally we fix
the coefficients of the structures to obtain (3.1).

3.1 Group-theoretical constraints on mixing

Restricting oneself to operators constructed from scalar fields m only is in general not
possible due to operator mixing: Scalar operators can also contain scalar combinations
of fermions A , field strengths F and derivatives D . All these operators are space-
time singlets and they must be treated on equal footing. Nevertheless, one can isolate
certain classes of operators by employing certain representations of the flavor symmetry
group SO(6) SU(4). For example the representations (a, b, c) [b + c, a b, b c]
(Young tableau/Dynkin labels) with classical scaling dimension 0 = a + b + |c| do not
admit derivatives or field strengths. If furthermore c = 0, fermions are excluded and the
operator must consist of only scalar fields. We will therefore consider operators in the
SO(6) representation

(a, b, 0) [b, a b, b] with 0 = a + b. (3.2)

These operators need not be superconformal primaries. If they are, they belong to the
protected C series of unitary irreducible representations (UIR) of SU(2, 2|4). In that

10
 a b c d e f g
Figure 2: Sample two-loop Green functions.

case they are quarter BPS operators (for b 6= 0) or half BPS (for b = 0). Otherwise
they belong the A series unprotected scalar UIR (a 2, b 2, 0) [b 2, a b, b 2]
with 0 = a + b 2. This UIR is at both unitarity bounds and splits into four UIRs at
gYM = 0. The operators under consideration are primary operators of the highest-lying
submultiplet.
We will now prove the above two statements concerning absence of mixing. The
SO(6) weights of the six scalars are (1, 0, 0), (0, 1, 0), (0, 0, 1), the weights of the
eight space-time pairs of fermions are ( 21 , 12 , 21 ) and derivatives as well as field
strengths have trivial weight (0, 0, 0). We see that the sum of labels a + b + c for all of
these is bounded from above by the corresponding scaling dimension of the field. For
a composite operator the labels of the constituents simply are summed and therefore
a + b + c is bounded from above by the scaling dimension 0 . If this bound is to be
achieved, all constituent fields must be on the bound as well. This is true only for
the three scalars with the weights (1, 0, 0), (0, 1, 0), (0, 0, 1) and a pair of fermions with
weight ( 12 , 21 , 12 ). This proves the first statement. Furthermore if c = 0 for the composite
operator the third scalar and the fermions are excluded as constituents. This proves the
second statement.
As advertised, an operator with a + b = 0 , c = 0 is constructed from two scalars
only. These scalars are charged with respect to different U(1) subgroups of SO(6). We
choose
Z = 12 (5 + i6 ), = 12 (1 + i2 ). (3.3)
We identify a, b with the number of Z, fields and assume a b. It may be shown
that any combination of the fields Z, is traceless with respect to the original SO(6)
symmetry. Written in terms of Z, the dilatation generator (1.3), (1.6), (3.1) is

D0 = Tr Z Z + Tr ,
D2 = 2 :Tr[, Z][, Z]:,
  
D4 = 2 :Tr [, Z], Z [, Z], Z :
  
2 :Tr [, Z], [, Z], :
2 :Tr [, Z], T a [, Z], T a :.
  
(3.4)

For the derivation of the two-loop result, however, we keep the SO(6) symmetry manifest
and consider the correlator of two pure scalar SO(6)-traceless operators.

11
 wo
of
woof

aa ab ac

ba bb bc d1 d2 d3

ca cb cc
Figure 3: Two-loop Green functions with three pairs of legs.

3.2 Three pairs of legs

At two loops the interaction can be disconnected (Fig. 2a) or connected. The discon-
nected diagrams can be shown to effectively yield (log)2 terms, necessary for the con-
formal structure of the correlator, plus further connected diagrams, see App. C. We
classify the connected interactions by their number of legs. We can have three pairs of
legs (Fig. 2b), two pairs of legs with one bulk loop (Fig. 2c,d,e,f) or one pair of legs with
two bulk loops (Fig. 2g). We start by analyzing the diagrams with three pairs of legs.
The interactions with three pairs of legs are depicted in Fig. 3. Most of the diagrams
correspond to two one-loop interactions (see Fig. 1) connected at one leg. We start by
subtracting the connected part of two one-loop vertices, see (C.6). This gives rise to
the same structures but with different coefficient functions. As in the one-loop case the
type b sub-diagrams consist of two parts, see (2.14). The first part contains local SO(6)
traces, + +
m m . We have restricted ourselves to traceless operators and drop the term.
The second term can be transformed by a Jacobi identity to sub-diagrams of type a and
c, see (2.15). This effectively doubles the sub-diagrams of type a. Then we notice that all
type c sub-diagrams have a scalar line that is connected to the rest of the diagram only
by one gluon-line. By removing some normal ordering we can write those diagrams as
Ga X a plus diagrams with two pairs of legs in analogy to (2.18). As Ga is the generator of
gauge transformations and the operators are color singlets, we can drop Ga X a . The same
holds for the d1 diagram. The d2 diagram has local SO(6) traces and will be dropped.
The d3 diagram (as well as a number of diagrams with vertical gluon lines as in Fig. 1e,
which have been omitted in Fig. 3) vanishes by symmetry: The diagram is proportional
to :Ga Gb Gc : Tr[T a , T b]T c .
The only diagram with three pairs of legs that effectively remains is Fig. 3aa. It is
multiplied by four due to contributions from ab, ba and bb. Its combinatorial structure

12
is given by
:Tr m [n , [m , [p , [n , p ]]]]:. (3.5)

3.3 One and two legs

There are many diagrams with one or two pairs of legs, see Fig. 2c,d,e,f,g. Additionally,
the connected piece of two one-loop diagrams and some reductions of diagrams with
three pairs of legs have to be taken into account. Instead of considering all these we
investigate the possible algebraic structures. A generic diagram consists of one trace and
four commutators. This can always be brought into the form

:Tr . . . [. . . , [. . . , [. . . , [. . . , . . .]]]]:. (3.6)

For diagrams with two pairs of legs we need to distribute m , n , m , n and two T a on
the six slots. We start by distributing the T a . If both T a are in the first two or the last
two slots, the structure vanishes by antisymmetry. If we place a T a in the first or the
last slot, we can interchange it with the second or second-but-last. There are identities
when the two T a are next to each other (corresponding to Fig. 2c)

[T a , [T a , [X, Y ]]] = [X, [T a , [T a , Y ]]] = 2 (Tr T a T a / Tr 1) [X, Y ], (3.7)

or one step apart (corresponding to Fig. 2d)

[T a , [X, [T a , Y ]]] = (Tr T a T a / Tr 1) [X, Y ]. (3.8)

The combination Tr T a T a / Tr 1 depends only on the gauge group, for U(N) it equals N.
With these we can remove two commutators along with the T a and get two independent
structures
Tr m [n , [T a , [T a , [m , n ]]]] N Tr[m , n ][m , n ], (3.9)
and
:Tr m [m , [T a , [T a , [n , n ]]]]: N:Tr[m , m ][n , n ]:. (3.10)
By use of gauge invariance, see (2.18), we can write (3.10) as a one pair of legs diagram.
The remaining structures have the two T a two steps apart (corresponding to Fig. 2e,f).
We write them as
:Tr T a [X1 , [X2 , [X3 , [X4 , T a ]]]]:. (3.11)
Using Jacobi identities on the nested commutators it is easy to see that we can inter-
change any of the Xi at the cost of terms like (3.8). The only remaining independent
structure can thus be chosen as

:Tr T a [m , [m , [n , [n , T a ]]]]:. (3.12)

By use of the identities (3.7) and (3.8) a diagram with one pair of legs (Fig. 2g) can
always be written as

Tr m [T a , [T a , [T b, [T b , m ]]]] N 2 Tr m m N Tr m Tr m . (3.13)

13
3.4 Determining the coefficients
All in all there are four independent structures (3.5), (3.9), (3.12) and (3.13). The
dilatation generator D4 at two-loops must be a linear combination of these. By using
known results we determine this linear combination.
We start by applying the structures to the half BPS operator Tr Z J . This operator
is protected, it does not receive corrections to its scaling dimension. Consequently D4
must annihilate it. Indeed the structures (3.5) and (3.9) do so. The structure (3.13)
gives JN 2 Tr Z J while (3.12) gives
J1
X
N Tr Z p Tr Z Jp + Tr Z 2 Tr Z p1 Tr Z J1p ,


2J (3.14)
p=1

(up to terms involving Tr Z). Thus in order for D4 to annihilate Tr Z J the coefficients of
(3.13) and (3.12) must vanish. This vanishing was explicitly confirmed in references [30,
31]. Our two independent structures correspond directly to those appearing there.
We are now left with
D4 = :Tr m [n , [m , [p , [n , p ]]]]: + Tr m [n , [T a , [T a , [m , n ]]]]. (3.15)
For the BMN-operators, see Sec. 6, (3.15) produces the two-loop planar anomalous di-
mension
4
(2 + )gYM 4
N 2 n2 gYM N 2 n4
+ . (3.16)
16 2 J 2 8J 4
By the pp-wave/BMN correspondence [17] we would expect the first term to vanish in
order for the BMN limit (6.1) to be well-defined. The second term should equal
4
gYM N 2 n4
14 2 n4 = . (3.17)
4J 4
This is also the result of an explicit calculation [18] and an investigation using supercon-
formal symmetry [43]. We find
= 2, = 1. (3.18)
This concludes our derivation of the two-loop dilatation generator (3.1).

4 Applications to lower dimensional operators

In this section we will apply the two-loop dilatation generator to a set of lower dimen-
sional operators. We consider only those representations where mixing with fermionic
and derivative insertions is prohibited, see Sec. 3.1. More explicitly, these are space-time
scalars in the SO(6) representations [p, 0 2p, p]. Furthermore, we will consider SU(N)
as the gauge group. This is equivalent to dropping all traces of single fields. For low
dimensions the calculations are easily performed by hand; as the dimension increases it
is useful to use an appropriate code in order to quickly and reliably work out the action
of the dilatation operator6 .
6
Many results of this section and Sec. 8 have been obtained using Mathematica. Computing e.g.
the exact dilatation matrix at two-loops for the 48 operators of dimension 8 took 90 sec. on a 366 MHz

14
 dim SO(6) BPS + dim SO(6) BPS +
2 [0,2,0] 1 - - 8 [0,8,0] 7 - -
3 [0,3,0] 1 - - [1,6,1] 4 - -
4 [0,4,0] 2 - - [2,4,2] 8 8 -
[2,0,2] 1 1 - [3,2,3] 3 4 3
5 [0,5,0] 2 - - [4,0,4] 4 7 -
[1,3,1] 1 - - 9 [0,9,0] 8 - -
[2,1,2] 1 1 - [1,7,1] 7 - -
6 [0,6,0] 4 - - [2,5,2] 11 11 -
[1,4,1] 1 - - [3,3,3] 9 11 5
[2,2,2] 3 3 - [4,1,4] 5 11 2
[3,0,3] - - 1 10 [0,10,0] 12 - -
7 [0,7,0] 4 - - [1,8,1] 10 - -
[1,5,1] 3 - - [2,6,2] 19 19 -
[2,3,2] 4 4 - [3,4,3] 13 18 9
[3,1,3] 2 2 1 [4,2,4] 15 33 5
[5,0,5] 1 4 8
Table 1: BPS and unprotected operators up to dimension 10. The unprotected operators of
representation 0 , [p, q, p] belong to the SU(2, 2|4) UIR [p2, q, p2] of bare dimension 0 2.
They are separated into two groups, + and according to their parity, cf. Sec. 7.

4.1 Quarter BPS operators

Quarter BPS operators belong to the representations [p, q, p] with p 1 and have classical
scaling dimension 0 = 2p + q. (For p = 0 the operators are half BPS.) In reference [33]
the explicit construction of lower dimensional 1/4 BPS operators was initiated. The
strategy consisted in first writing down all linearly independent local polynomial scalar
composite operators in a given representation and next diagonalizing the corresponding
matrix of one-loop two-point functions in the inner product given by the matrix of tree-
level two-point functions. The former part of the strategy was very recently systematized
and completed in [44]. Our idea of focusing on the dilatation operator of the theory allows
us to significantly simplify and extend the analysis of [33,44]. As explained in Sec. 1 one
does not need to work with the tree-level inner product but can by use of the effective
vertices (2.20) and (3.1) directly write down the action of the dilatation generator in
any convenient basis of operators. The (possible) 1/4 BPS operators are then simply
determined as the kernel of the dilatation operator. By our method we have confirmed
the BPS nature of the operators found in [33, 44] up to two-loop order (without any
2
modification of mixing coefficients at O(gYM )). Furthermore, we have determined all 1/4
BPS operators of dimension 8 (to two-loop order) and of dimension 9 and 10 (one-loop
order). Needless to say that our method also gives access to a host of previously unknown
two-loop anomalous dimensions of unprotected operators. Our results on BPS operators
are summarized in Tab. 1 and a collection of anomalous dimensions can be found in
PC. Obviously it is also useful and straightforward to use Mathematica or similar to diagonalize the
dilatation matrices and perform the N1 expansion of the eigensystem.

15
the following sections and in App. D. It is known that operators in the representations
[1, p, 1] are always 1/4 BPS. For the operators considered in Tab. 1 it appears that exactly
half of those belonging to representations of the type [2, p, 2] are 1/4 BPS, see also Sec. 5.
In the following two sub-sections we shall discuss some further insights gained from
our analysis.

4.2 Exact results for lower dimensional operators

From Tab. 1 we see that there is exactly one unprotected operator in each of the repre-
sentations (0 = 4, [2, 0, 2]), (0 = 5, [2, 1, 2]) and (0 = 6, [3, 0, 3]). These operators
can be written as

K = Oa = Tr[, Z][, Z],

Ob = Tr[, Z][, Z]Z,
Oc = Tr[, Z][, Z][, Z], (4.1)

and descend from the operators

K = Oa,0 = Tr m m ,
Ob,0 = Tr m m n ,
Oc,0 = Tr m m [n , p ], (4.2)

with (0 = 2, [0, 0, 0]), (0 = 3, [0, 1, 0]) and (0 = 4, [1, 0, 1]). For the operators (4.1)
the dilatation generator yields
2 4
3gYM N 3gYM N2
K = a = 4 + ,
4 2 16 4
g2 N 3g 4 N 2
b = 5 + YM 2 YM 4 ,
2 32
2
3g N 9g 4 N 2
c = 6 + YM2 YM 4 , (4.3)
4 64
and these values do not receive any non-planar corrections at the present loop order.
Note that the anomalous dimension of the Konishi descendant K agrees with previous
results [45,38,46] and that K and b are in agreement with the general formula (5.17),
corresponding respectively to the cases (J, n) = (2, 1) and (J, n) = (3, 1). The operator
Oc is the first in an infinite sequence of operators with identical quantum anomalous
dimensions, to be discussed in Sec. 8.4.

4.3 Peculiarities of the large N expansion

Most of the other single-trace operators in Tab. 1 cease to be eigenvectors of the one-
loop dilatation generator when we include non-planar corrections, due to mixing with
multiple trace operators. This mixing has important consequences for the very nature
of the anomalous dimension of these operators as we shall illustrate by the examples of
(0 = 6, [2, 2, 2]) and (0 = 7, [2, 3, 2]) from Tab. 1.

16
 The operators in the first example are descendants of the operators with (0 = 4,
[0, 2, 0]) studied in [47]. The three unprotected operators can be chosen as

Tr[, Z][, Z]Z 2

O = Tr[, Z]Z[, Z]Z , (4.4)

2
Tr Z Tr[, Z][, Z]

the commutators conveniently projecting out the three protected ones. In the basis given
by (4.4) the one- and two-loop dilatation generator take the form

+12 2 + N6

D2 = N 8 +8 N4 , (4.5)
16 16
+ N N +12

and
52 N242 +4 + N242 52

N
D4 = N 2 +48 + N162 16 N162 + 48
N
. (4.6)
144 80 64
N +N 48 N 2
Similarly, for 0 = 7 a convenient basis of unprotected operators consists of the set
of operators given by
Tr[, Z][, Z]Z 3

Tr[, Z]Z[, Z]Z 2
O= Tr Z 2 Tr[, Z][, Z]Z ,
(4.7)
Tr Z 3 Tr[, Z][, Z]
and in this basis the one- and two-loop dilatation matrices read

0 + N6

+12 0
4 +4 + 4 6
D2 = N N N , (4.8)
+ 8 8 +8 0
N N
+ 24
N
24
N
0 +12

and
48 N362 12 + N362 48

0 N
+20 + 42 N42 16 + 20
D4 = N 2

N N N . (4.9)
54 16 8 24
N +N 24 + N 2 N2
192 72 24 72
N +N + N2 48 N 2
Notice that the dilatation matrices (4.5), (4.6), (4.8) and (4.9) are exact to the given
loop order. Not surprisingly, these matrices have a form compatible with the existence
of a t Hooft expansion, trace splitting and trace joining processes being suppressed by
a power of N1 compared to trace conserving ones.
At the classical level all the operators in respectively (4.4) and (4.7) have conformal
dimension respectively 6 and 7. Including quantum effects at one-loop and two-loop
order the proper scaling operators consist of some operators which are mainly single trace
and some which are mainly multiple trace. The corresponding quantum corrections to
the conformal dimensions are the roots of the characteristic polynomial of the matrix
g 2 D2 + g 4 D4 . It is easy to show that this characteristic polynomial has a well-defined t

17
 2
Hooft expansion with the dependence of gYM and N organizing into the t Hooft coupling
= gYM N and the genus counting parameter N12 . Furthermore, its topological expansion
2

terminates at a finite genus. However, the roots of the characteristic polynomial need
not have a terminating genus expansion7 or even a well defined t Hooft expansion. Let
us illustrate this by considering in more detail the operators (4.4), (4.7).
For 0 = 6 the relevant characteristic polynomial is of cubic order
   
3 2 10 10
8 + 20 2 15 2 (4.10)
N N
2
     
gYM N 26 2 54 100
+ 29 + 2 141 2 + 150 2 ,
16 2 N N N

and the scaling dimensions up to two-loops are exactly

2
gYM N
=6+ , (4.11)
4 2
with a root of the corresponding cubic equation. The discriminant of the cubic equation
does not take the form of a perfect square but is easily seen to be expressible as
X (g 2N)i
D= ci,j , (4.12)
i,j
N 2j

with the ci,j some constants and c0,0 6= 0, implying that the roots will have a well-defined
t Hooft expansion which however involves a non-terminating sum over genera. To three
leading orders in the genus expansion the conformal dimension of the two mainly single
trace operators read
2
!
gYM N 5 5 5 2 5 75 34 5
single = 6 + + (4.13)
2 8 2N 2 N4
4
!
gYM N2 17 5 5 131 57 5 1675 751 5
+ + ,
4 128 64N 2 16N 4

whereas that of the mainly double trace one equals

2 4
N2
   
gYM N 3 5 150 gYM 3 59 1675
double = 6 + + 4 + + .
2 4 N2 N 4 16 16N 2 8N 4

The one-loop contributions to these conformal dimensions agree with those obtained
in reference [47] and the planar versions of single are in accordance with the general
formula (5.17) for J = 4 and n = 1 and n = 2 respectively. Furthermore the planar part
of the quantum correction to double is identical to that of the operator K in (4.1) as it
should be.
7
This was also noticed in [33, 47, 41].

18
 In the case 0 = 7 the situation is more involved. Here the relevant characteristic
polynomial is of degree four. The discriminant of the associated cubic polynomial does
also not here take the form of a perfect square but can be expressed as
X (g 2N)i
D= ci,j , (4.14)
i,j
N 2j

where the ci,j are again some constants. However, this time c0,0 = c1,0 = 0. This implies
that not all the wished for conformal dimensions have well defined t Hooft expansions.
(Of course they still exist as the exact roots of the characteristic polynomial.) Two of the
conformal dimensions, associated with respectively a mostly single trace and a mostly
double trace operator, do have well defined t Hooft expansions and for these we find for
the first few orders
2 4
N2
   
gYM N 1 11 253 gYM 3 103 5541
single = 7 + + + + ,
2 4 16N 2 64N 4 4 128 1024N 2 8192N 4
2 4
N2
   
gYM N 1 1 17 gYM 3 7 129
double = 7 + + + + . (4.15)
2 2 2N 2 2N 4 4 32 32N 2 32N 4

Here we notice that the planar contribution to single is in accordance with the general
formula (5.17) for (J, n) = (5, 1) and that the planar quantum correction to double as
expected is the same as that of the operator Ob in equation (4.1). We then remove the
roots associated to these scaling dimensions and obtain from the quartic polynomial a
quadratic one. This quadratic polynomial has the roots

g2 N

remaining = 7 + YM2 M D , (4.16)

2
   
3 3 291 gYM N 45 537 27483
M= + + ,
4 32N 2 128N 4 2 256 2048N 2 16384N 4
2 4
N2
   
27 531 gYM N 3123 29295 9gYM
D= + + + .
32N 2 1024N 4 2 4096N 2 32768N 4 65536 4

As we have determined the two well-defined scaling dimensions in a perturbative fashion,

the coefficients M, D of the new polynomial are series in N1 and gYM 2
N. (Here we notice
that the last coefficient in D is indeed determined by D2 and D4 alone and as we shall
see it plays a crucial role for the analysis.) From the form of D it follows that we have a
square root singularity in the (, N1 ) plane with a branch point at the origin. Obviously,

we can not expand D in powers of and N1 at the same time. Thus for the remaining
two operators it only makes sense to speak of either a singular anomalous dimension
(4.16), a planar anomalous dimension (N ) or a one-loop anomalous dimension
(g 0). For the planar anomalous dimension we find
2 4
3gYM N 21gYM N2
single,planar = 7 + ,
4 2 128 4
3g 2 N 3g 4 N 2
double,planar = 7 + YM2 YM 4 . (4.17)
4 16
19
The first of these anomalous dimensions correctly reproduces the result encoded in the
general formula (5.17) and the quantum correction of the second agrees with that of
the Konishi operator K. For the one-loop anomalous dimension one gets to four leading
orders in the large-N expansion
2
!
gYM N 3 3 6 3 59 6
oneloop = 7 + . (4.18)
2 4 8N 32N 2 512N 3

The absence of a well-defined t Hooft expansion for the conformal dimension of certain
operators is a degeneracy effect and the appearance of odd powers of N1 is a particular
manifestation of this. From the form of the one-loop dilatation matrix (4.8) it follows that
at the planar level a single trace state and a double trace state are degenerate. Treating
the N1 effects as a perturbation thus calls for the use of degenerate perturbation theory
which exactly gives rise to the N1 corrections in equation (4.18), lifting the degeneracy
present at the one-loop level. The degeneracy is also lifted if instead the two-loop effects
are treated perturbatively and it is this simultaneous lift of degeneracy by two non-
commuting perturbations which is the origin of the problems with the t Hooft expansion.
Note that these problems also renders problematic the perturbative evaluation of higher
point functions in conjunction with the N1 expansion. This question is however beyond
the scope of this paper.

5 Two impurity operators

Berenstein, Maldacena and Nastase suggested to investigate operators of a large dimen-
sion 0 and a nearly equally large charge J under a subgroup SO(2) of SO(6). We
identify the SO(2) subgroup with the group U(1) corresponding to the phase of the field
Z defined in (3.3). Then the relevant operators constitute long strings of Z-fields with
0 J impurities scattered in. As particular examples BMN investigated operators
with zero, one and two impurities of type . As shown in [20] it makes perfect sense to
consider these operators also for arbitrary finite values of J. The primary single-trace
operators with up to two impurities are in one-to-one correspondence with the SO(6)
representations [0, 0 , 0] and [0, 0 2, 0]. These primary operators have supersymme-
try descendants that can be written in terms of J fields Z and two impurities of type .
They belong to the representation [2, 0 4, 2] for which we may apply the dilatation
generator up to two-loops as given by (3.4). A generic multi-trace operator with two
impurities is written as8
k
Y
OpJ0 ;J1 ,...,Jk = Tr(Z p Z J0 p ) Tr Z Ji , (5.1)
i=1
k
Y
QJ0 ,J1;J2 ,...,Jk = Tr(Z J0 ) Tr(Z J1 ) Tr Z Ji , (5.2)
i=2
8
We observe that there seem to be as many unprotected operators of the type (5.1) as there are BPS
operators of the type (5.2) in the SO(6) representation [2, 0 4, 2], see also Tab. 1.

20
with ki=0 Ji = J. Both series of operators are symmetric under the interchange of sizes
P
Jk of traces Tr Z Jk , O is symmetric under p J0 p and Q is symmetric under J0 J1 .
The quantum dilatation generator (3.4) can be seen to act as
    
2 4 O 0 O
(g D2 + g D4 ) = , (5.3)
Q 0 Q

i.e. operators of type Q are never produced. This follows from the fact that all produced
objects will contain a commutator [, Z] in some trace and this trace will vanish unless
it contains another . It immediately shows that for every Q there is one protected
(quarter BPS) operator. Its leading part is given by Q itself, plus a N1 correction from
the operators O [33, 44, 25, 26]. On the other hand the operators O are in general
not protected and we will investigate their spectrum of anomalous dimensions in the
following. From the form of the dilatation matrix we infer that operators of type O do
not receive corrections from operators of type Q; the latter therefore completely decouple
as far as the consideration of the Os is concerned.

5.1 The action of the dilatation generator

When acting on states of the type given in (5.1) the one- and two-loop dilatation operator
take the form given in equation (3.4), i.e.

D2 = 2 :Tr[Z, ][Z, ]:,

D4 = 2 :Tr[Z, ][Z, [Z, [Z, ]]]: (5.4)
+ 4N :Tr[Z, ][Z, ]: (5.5)
+ 2 :Tr[Z, ][, [, [Z, ]]]:.

Here we shall prove a further simplification occurring when D4 acts on (5.1), namely

D4 = 14 (D2 )2 + 2 :Tr[Z, ][, [, [Z, ]]]: (5.6)

2
14 (D2 ) + D4 . (5.7)

This relation explicitly shows that D2 and D4 do not commute due to the extra piece D4 ,
involving an interaction of both impurities that was not present at the one-loop level.
Eq. (5.6) is shown as follows: When applying D2 to an operator Op of the type (5.1)
there are two possible ways of contracting the in D2 with a in Op . Each of these
possibilities gives the same contribution to D2 Op 9 . The operator D2 Op is again a sum
of operators of the type (5.1). Therefore, when evaluating D2 (D2 Op ) we get the same
contribution from the two possible contractions of the in the leftmost D2 with a in
(D2 Op ). We can thus determine D2 (D2 Op ) by contracting in the leftmost D2 with
9
This fact can be confirmed by direct computation. It can also be proven by use of the parity
operation defined in Sec. 7. Let us tag one of the impurities, , D2 will only contract to this impurity.
Whatever the outcome of D2 Tr Z p Z Jp will be, it is again of the type (5.1) and thus has positive
parity. Therefore D2 Tr Z p Z Jp = P D2 Tr Z p Z Jp = D2 P Tr Z p Z Jp = D2 Tr Z Jp Z p =
D2 Tr Z p Tr Z Jp .

21
the in the already applied D2 and multiplying the result by two. Next, the Z of the
leftmost D2 must be contracted with one of the Zs of (D2 Op ). This can either be a
Z of the original Op or the Z appearing in the already applied D2 . These two types
of contractions are, up to an overall factor, described respectively by the vertices (5.4)
and (5.5).
It is easy to write down the exact expression for D2 Op . Let us define

D2 ND2;0 + D2;+ + D2; , (5.8)

where D2;0 is trace conserving and D2;+ and D2; respectively increases and decreases
the trace number by one. Then we find

J0 ;J1 ,...,Jk J0 ;J1 ,...,Jk
D2;0 Op = 4 p6=J0 Op+1 (p6=J0 + p6=0 ) OpJ0 ;J1 ,...,Jk (5.9)

J0 ;J1 ,...,Jk
+ p6=0 Op1 ,
p1  
J J ;J1 ,...,Jk+1 J J ;J ,...,Jk+1
X
D2;+ OpJ0 ;J1 ,...,Jk = 4 0
OpJ k+1
k+1
0 k+1 1
Op1J k+1
Jk+1 =1
J0 p1
X  J Jk+1 ;J1 ,...,Jk+1

4 0
Op+1 OpJ0 Jk+1 ;J1 ,...,Jk+1 ,
Jk+1 =1
k
+Ji ;J1 ,..., +Ji ;J1 ,...,
X  
D2; OpJ0 ;J1 ,...,Jk = 4 Ji OJJi0+p Ji ,...,Jk
OJJi0+p1 Ji ,...,Jk

i=1
k
J0 +Ji ;J1 ,...,
OpJ0 +Ji ;J1 ,...,
X  
Ji ,...,Jk Ji ,...,Jk
4 Ji Op+1 .
i=1

In analogy with (5.8) one can write D4 as

D4 = N 2 D4;0 + N D4;+ + N D4; + D4;++ + D4; + D4;+ , (5.10)

where D4;+ stands for the genus one trace number conserving part. For D4;0 we get
(for J0 > 0)
 
J0 ;J1 ,...,Jk J0 ;J1 ,...,Jk J0 ;J1 ,...,Jk
D4;0 Op = 4(p,0 + p,J0 p,1 p,J01 ) O1 O0 . (5.11)

The remaining contributions to D4 likewise always produce operators in the combination

O1 O0 . We defer the exact expressions for D4; , D4;+ , D4;++ and D4;+ to App. E.
The contribution D4; is easily seen to vanish.

22
5.2 The planar limit
In the basis of symmetric two-impurity single trace states of the type (5.1) with k = 0
the planar one-loop dilatation matrix takes the form (for J 1)

+1 1

1 +2 1
.. .. ..

. . .
D2;0 =4 . (5.12)

.. .. ..

. . .

1 +2 1
1 +1

Notice the special form of the boundary contributions which follows from the presence
of the -functions in equation (5.9). As exploited in reference [20] the matrix (5.12) has
the following exact eigenvectors
J  
1 X n(2p + 1)
OnJ = cos OpJ . (5.13)
J + 1 p=0 J +1

We emphasize that equation (5.13) thus gives for any value of J our one-loop planar
eigenoperators. Note that OnJ = On
J J
= OJ+1n . Thus the set of independent modes
can be chosen as given by the mode numbers 0 n [J/2]. The inverse transform is
[J/2]  
X n(2p + 1)
OpJ = J
On=0 +2 cos OnJ . (5.14)
n=1
J +1

The corresponding exact planar one-loop anomalous dimension was found in [20]. Moving
on to the two-loop analysis we infer from (5.7) and (5.10) that the matrix D4,0 can be
expressed as the square of the one-loop matrix (5.12) plus contact-term contributions
from (5.11)

1 +1
+1 1


0

D4;0 = 14 (D2;0 )2 + 4 .. . (5.15)
.



0

1 +1
+1 1

We now face the problem that the states (5.13) are no longer eigenstates of D4;0 , since
D2;0 and D4;0 do not commute. We find
[J/2]
256 2 n n X 2 m m
D4;0 OnJ = sin cos sin cos OJ . (5.16)
J +1 J +1 J + 1 m=1 J +1 J +1 m

23
However, we can treat D4;0 as a perturbation and thus find that the two-loop part of the
planar anomalous dimension is the diagonal piece of D4;0 proportional to OnJ . Exploiting
the relation (5.7), for the operators whose one-loop form is given by (5.13) we obtain the
following planar anomalous dimension exact to two loops
!
2 n
g 2
N n g 4
N 2
n 1 cos J+1
Jn = J + 2 + YM2 sin2 YM4 sin4 + . (5.17)
J +1 J +1 4 J +1

This formula comprises the results K , b in (4.3) as well as the planar piece in single
in (4.13), (4.14) and (4.17). In (8.10) we present a conjecture for the planar three-loop
contribution to Jn . Furthermore, using standard perturbation theory we can also find
the (planar) perturbative correction to the eigenstates (5.13): They involve the coupling
constant dependent redefinition
[J/2]
2
gYM N 1 X sin2 n
J+1
n
cos J+1 sin2 m
J+1
m
cos J+1
OnJ OnJ m6=n J
Om . (5.18)
2 J + 1 m=1 sin2 n
J+1
sin2 J+1
m

This gYM -dependent remixing is a complicating feature that we can expect at each further
quantum loop order; remarkably, it is absent in the small J limit (i.e. for J = 2 (Konishi)
and J = 3) and in the large J (BMN) limit as discussed in the next section.

6 The BMN limit of two impurity operators

In the BMN limit one is supposed to send J to infinity and consider only operators for
which D0 J is finite [17]. More precisely, the BMN limit is defined as the double scaling
limit [22, 23]
2
gYM N J2
J , N , = , g 2 = fixed. (6.1)
J2 N
In reference [34] it was shown how the one-loop dilatation operator could be identified
as a certain quantum mechanical Hamiltonian in the BMN limit. Since the same quan-
tum mechanical Hamiltonian turns out to contain all information about the two-loop
dilatation operator in the BMN limit we shall briefly recall its derivation.

6.1 The one-loop Hamiltonian

With J being very large we can view x Jp and ri JJi as continuum variables and we
replace the discrete set of states in equation (5.1) by a set of continuum states

OpJ0 ;J1 ,...,Jk |x; r1 , . . . , rk i = |r0 x; r1 , . . . , rk i, (6.2)

where
x [0, r0 ], r0 , ri [0, 1] and r0 = 1 (r1 + . . . + rk ), (6.3)
and where it is understood that |x; r1 , . . . , rk i = |x; r(1) , . . . , r(k) i with Sk an arbi-
trary permutation of k elements.

24
 Defining
2
g D2 = 2 d 2 , (6.4)
4
we can write d2 = d2;0 + g2 d2;+ + g2 d2; and imposing the BMN limit (6.1) we get a
continuum version of the equations (5.9)
d2;0 |x; r1 , . . . , rk i = x2 |x; r1 , . . . , rk i, (6.5)
Z x
d2;+ |x; r1 , . . . , rk i = drk+1 x |x rk+1 ; r1 , . . . , rk+1 i
0
Z r0 x
drk+1 x |x; r1 , . . . , rk+1i,
0
k
X
d2; |x; r1 , . . . , rk i = ri x |x + ri ; r1 , . . . ,
ri , . . . , rk i
i=1
k
X
ri x |x; r1 , . . . ,
ri, . . . , rk i.
i=1

The (k + 1)-trace eigenstates at g2 = 0 are (with n integer)10

Z r0
1
|n; r1 , . . . , rk i = dx cos (2nx/r0 ) |x; r1 , . . . , rk i. (6.6)
r0 0
This is of course in accordance with the nature of the exact eigenstates at finite J, cf.
eqn. (5.13). The inverse transform of (6.6) reads

X
|x; r1 , . . . , rk i = |0; r1 , . . . , rk i + 2 cos (2mx/r0 ) |m; r1 , . . . , rk i. (6.7)
m=1

In the basis (6.6) the action of the operator d2 reads

2
d2;0 |n; r1 , . . . , rk i = 2n
r0
|n; r1 , . . . , rk i, (6.8)
2
2m
sin2 n rk+1




8 r0
Z
r0 rk+1 r0
X
d2;+ |n; r1 , . . . , rk i = drk+1
2
2 |m; r1 , . . . , rk+1i,
r0 0 m=1
2m
r02n
r0 rk+1
k 2m

2 2 ri


X ri X r0 +ri sin m r0 +r
d2; |n; r1 , . . . , rk i = 8
i |m; r1 , . . . ,
ri , . . . , rk i.
r 2m 2 2n 2


i=1 0 m=1 r0 +ri
r0

Now, the scene is set for determining the spectrum of the full one-loop Hamiltonian order
by order in g2 by standard quantum mechanical perturbation theory and in reference [34]
we carried out this program for single trace operators to three leading orders in g2 . Note
the great simplicity of the expressions eq.(6.8) (no contact terms etc.), emerging directly
from the study of the dilatation operator. Rather than comparing specific consequences
of the above equations, and turning around the suggestion in [35, 48, 49], we feel that it
would be important to derive the above Hamiltonian in the string field formulation of
plane wave strings [5056] or the (interpolating) string bit formulation of [5759].
10
Notice that as opposed to in reference [34] we use symmetrized operators and therefore the cosine
transform appears.

25
6.2 Two loops
Proceeding to two loops we define
2


4
g D4 = d4 , (6.9)
4 2

and we see from equation (5.7) that the only new element in the analysis is the deter-
mination of the BMN limit of the operator D4 . Now, it follows from the equations in
App. E, that acting with D4 on a state of the type (5.1) produces at the discrete level
only states in the combination O1 O0 . Expressing this quantity in terms of the exact
eigenstates at g2 = 0, (5.13), it is easily seen that in the BMN limit we have11
1
O1J0 ;J1,...,Jk O0J0 ;J1 ,...,Jk . (6.10)
J02

Using this result a straightforward scaling analysis of the relations in App. E shows that
the operator D4 becomes irrelevant in the BMN limit. Thus, in this limit we have

d4 = 14 (d2 )2 , (6.11)

and at any given order in the genus expansion, the eigenstates of the one-loop Hamil-
tonian are also eigenstates of the two-loop Hamiltonian12 . For the BMN limit of D J
one gets up to and including two loops
  2
1


D J 2 + d2 (d2 )2 . (6.12)
4 2 4 4 2

Writing the BMN double expansion of the conformal dimension of an operator as

 k

X
( , g2 ) 0 = 2
k (g2 ), (6.13)
k=1
4

with 0 the tree-level conformal dimension, what we learned above can be expressed as

2
2 (g2 ) = 41 1 (g2 ) , g2 . (6.14)

6.3 Degeneracies
From equation (6.8) it follows that at the planar, one-loop level we have a degeneracy

between states |n; r1 , . . . , rk i and |n ; r1 , . . . , rk i for which rn0 = rn . Focusing on the
0
simplest case a single trace state |ni is degenerate with double trace states of the type
m
|m; ri for which n = 1r , with triple trace states |k; r1 , r2 i for which n = 1rk1 r2 and so
11
Notice that O1 O0 is down by a factor of J1 compared to Op Op1 for general p. In the continuum
language this is reflected by the fact that x |x; r1 , . . . , rk i|x=0 = 0 which follows from the relation (6.7).
12
This proves that the O( ) contribution to the three-point function of symmetric-traceless operators
J
O(ij),n in [25] is indeed correct. For the singlet or antisymmetric operators this remains an open question.

26
on [26]. Notice that degeneracy is excluded for the special case |ni = |1i. Determining the
energy shift of the state |ni with n > 1 to genus one by treating the term g2 (d2,+ + d2, )
as a perturbation thus in principle requires the use of degenerate perturbation theory.
It can be shown that at order g2 matrix elements between single trace states and double
trace states vanish [25,26]. In contrast, matrix elements between states |ni and |k; r1, r2 i
for n = 1rk1 r2 are finite [34] at order g22 . So far these degeneracy effects have not been
rigorously taken into account due to the technical obstacles caused by the fact that a
single trace state is degenerate with a continuum of triple trace states13 . Bearing in
mind the implications of the finite-J degeneracies, cf. Sec. 4.3, it is most desirable to
stringently carry through analysis in the large-J case as well.

6.4 All loop conjecture

Assuming that the effective vertex idea works at any loop order one would expect that d2
encodes all information about the BMN limit of the dilatation operator in the basis (5.1).
In the BMN limit we single out those terms of the dilatation operator which produce
the maximum number of J factors. These are the terms with the maximum number of
Z constituents. Therefore, only terms of the dilatation operator which have only one
impurity will survive the limit. Any such term will appear in some higher power of d2
and it has to appear with a weight which ensures that the planar BMN limit [17, 18, 43]
p
J + 2 1 + n2 , (6.15)
is conserved. Thus, it is tempting to conjecture that the all loop version of the for-
mula (6.12) reads s  

D D0 2 1 + d2 . (6.16)
4 2
We now construct a possible all-loop expression for the one impurity part of the full
dilatation generator which manifestly gives rise to the above formula in the case of two
impurity operators. Iterating the argument given in the beginning of Sec. 5.1 we have
the following simplification when (D2 )l acts on two-impurity operators
a a
(D2 )l 2l1 (a0 ) D2a0 a1 D2a1 a2 . . . D2 l1 l (al ) , (6.17)
where
D2ab = 2:Tr[Z, T a][Z, T b ]:. (6.18)
Thus, a possible form for the one-impurity part of the full l-loop dilatation generator is
(2l 2)! (a0 ) a0 a1 a1 a2 a a
D2l = (2)1l D2 D2 . . . D2 l1 l (al ) + . . . , (6.19)
(l 1)!l!
up to terms irrelevant in the BMN limit. There might, however, be other combinations
which also give rise to (6.16). At the planar level (6.19) is equivalent to the conjecture
13
One way to treat the degeneracy might be to consider the finite J operators discussed in Sec. 5 as
regularized BMN operators and then ensure that all relevant degeneracies are resolved before taking J
to infinity. For single-trace operators this would require working with n being a divisor of J + 1.

27
of [17, 18] and the result of [43]. However, our possible formula would apply to arbitrary
genus. If it was correct, one would be able to perform all-loop computations not only for
two-impurity BMN operators, but also for more than two impurities. There, we do not
find a reason why D2l should be proportional to (D2 )l and an individual genus expansion
would be required each loop level.
The formula (6.19), if true, would also have implications for N = 4 SYM beyond
the BMN limit. It successfully predicts the dilatation generator at two-loops completely:
D4 consists of three parts which can be labelled as (1, 1), (2, 1) and (1, 2), the numbers
representing the number of interacting background and impurity fields Z, . The parts
with one impurity are predicted directly, the part (1, 2) is related to (2, 1) by symmetry.
At the three-loop level all terms except the terms with two impurities and two background
fields, (2, 2), can be predicted. As this represents a three-loop interaction between four
fields, the corresponding Feynman-diagrams do not have bulk loops. The complexity of
such diagrams is comparatively low and a direct computation of coefficients may still be
feasible.

7 Symmetry enhancement at the planar level and

integrability
Let us define a parity operation P acting on traces of group generators T a by inverting
the order of generators within a trace or, alternatively, by transposing all generators T a .
For example
P Tr Z 3 2 Z = Tr Z2Z 3 = Tr Z 3 Z2 . (7.1)
It is easily seen that the dilatation generators D0 , D2 , D4 commute with the parity
operation P D2l P = D2l . In fact, parity is related to complex conjugation on group
generators, P T a = (T a )T = (T a ) . As the dilatation generator D is real it conserves
parity in general
P D = DP. (7.2)
This implies that there is no mixing between operators of different parity14 , which was
also noticed in [44]. This parity operation is specific to the SU(N) series. In the SO(N)
and Sp(N) series the parity of a trace of n generators T a is given by (1)n due to the
identity P T a = (T a )T = T a 15 .
As parity is conserved one should a priori expect that the spectra of positive and
negative parity operators are not related. An example of a system with conserved parity
is the infinite potential well. There, the even and odd modes have distinct spectra which
are not related to each other in any obvious way. This is what happens in N = 4 SYM
in general. If we however consider the strict planar limit, N = , we find that the two
sectors are related. We observe that, whenever parity even and odd states have equal
quantum numbers (except parity), they form pairs of operators with degenerate scaling
14
The reason why the separation of parities has not been an issue so far is that at least three impurities
are required to allow both parities within the same SO(6) representation.
15
A consequence of this is that many of the operators discussed so far are incompatible with the
SO(N ) and Sp(N ) series.

28
dimension. As a systematic degeneracy almost inevitably indicates a symmetry there
seems to be a symmetry enhancement in the planar limit of N = 4 SYM.
We start by discussing the most simple example of such a planar parity pair and
afterwards we explain the degeneracy at the one-loop level by means of an integrable
spin chain [37].

7.1 Planar parity pairs

There are three unprotected operators in the representation [3, 1, 3] with 0 = 7. One
has negative parity
O = Tr[, Z][, Z][, Z]Z, (7.3)
and two have positive parity

2 Tr Z 4 3 + 2 Tr Z 2 Z 2 2 + 2 Tr Z 2 ZZ 3 Tr Z 3 {, Z}
 
O+ = . (7.4)
Tr Z Tr Z 2 [, Z] Tr Z 2 Tr Z[, Z]2

The scaling dimension for the first is readily obtained

2 4
5gYM N 15gYM N2
= 7 + . (7.5)
8 2 128 4
The dilatation generator acts on the other two as

10 N8 30 N402 40
   
2 N
D2 = N 20 , D4 = N , (7.6)
N
8 140
N
24 N642

corresponding to the scaling dimensions (the one loop part of which has been found
in [33])

2 4 2 948
3+ 2
 
g N g N 52
q
+ = 7 + YM 2 9 1 + 160 N2
YM 4 27 + 2 q N . (7.7)
16 256 N 1 + 160 N2

Intriguingly we find that the scaling dimension of the two single-trace operators, O and
O+,1 approach each other for N at one-loop as well as at two-loop order. We hence
find a pair of operators with opposite parity which have degenerate scaling dimensions.
This is actually not an exception, it is rather the rule. Among the operators in the
representations [3, 0 6, 3] with 6 0 10 we find 8 pairs of operators and only 3
unpaired ones, cf. Tab. 2, Sec. 5 and App. D. And they are not the only examples, we
find them in the representations (0 = 9, [4, 1, 4]), (0 = 10, [4, 2, 4]), (0 = 6, [0, 2, 0])
and (0 = 6, [1, 0, 1]). In fact, every representation that admits both parities seems to
have such pairs of operators.
This phenomenon is a signal of an enhanced symmetry in the planar limit of the
SU(N) series16 . It cannot be supersymmetry (this time) for a simple reason: Supersym-
metry is independent of the choice of gauge group whereas at finite N the operators O+
16
The groups SO(N ), Sp(N ), although admitting a planar limit, do not show this behavior: One of
the parity partners is always absent.

29
 dim SO(6) SU(2) + dim SO(6) SU(2) +
4 [2,0,2] 1 1 - 10 [2,6,2] 7 4 -
5 [2,1,2] 2 1 - [3,4,3] 5 3 4
6 [2,2,2] 3 2 - [4,2,4] 3 8 2
[3,0,3] 1 - 1 [5,0,5] 1 - 4
7 [2,3,2] 4 2 - 11 [2,7,2] 8 4 -
[3,1,3] 2 1 1 [3,5,3] 6 5 5
8 [2,4,2] 5 3 - [4,3,4] 4 10 5
[3,2,3] 3 1 2 [5,1,5] 2 6 6
[4,0,4] 1 3 - 12 [2,8,2] 9 5 -
9 [2,5,2] 6 3 - [3,6,3] 7 6 7
[3,3,3] 4 3 3 [4,4,4] 5 17 7
[4,1,4] 2 3 1 [5,2,5] 3 9 14
[6,0,6] 1 12 2
Table 2: Single-trace pure scalar operators of dimension 0 12. The operators are distin-
guished by their representation and parity. The protected operators of representation [0, 0 , 0]
were omitted. The operators correspond to states of a SU(2) spin chain of length L = 0 .

mix and have modified anomalous dimensions. Furthermore, for gauge groups other than
SU(N) the pairs cannot even exist. Therefore the operators O+,1 and O cannot be part
of the same supermultiplet. As the multiplets of this symmetry seem to be either singlets
or doublets the symmetry is most likely abelian. In the next subsection we proceed by
proving the degeneracy at the one-loop level.

7.2 A conserved charge

In order to prove the existence of an abelian symmetry in the planar sector we will
find a generator U and show that it commutes with the generator of dilatations D.
Furthermore, we demand that U anticommutes with the parity operation P so that it
may relate operators of opposite parities.

[D, U] = {P, U} = 0. (7.8)

We will restrict ourselves to the one-loop level, where we can allow all six scalar fields;
mixing with other fields is a higher-order effect. The degeneracy at the two-loop level
will be investigated in the next section.
We represent a single-trace operator by a cyclic SO(6)-vector spin chain of length
L = 0 with zero (angular) momentum as suggested by Minahan and Zarembo [37].
The one-loop dilatation operator is then given by
L
X
D2 = N D2,k(k+1) , (7.9)
k=1

where D2,k(k+1) is a local interaction linking sites k and k + 1 (the sites are periodically

30
 D2 = 2 2 +

Figure 4: Graphical representation of the planar one-loop interaction D2 .

D2 D2
U2 =
D2 D2

Figure 5: Graphical representation of the generator U2 .

identified). The local interaction D2,k(k+1) is given by [37]

D2,k(k+1) = 2Ik(k+1) 2Pk(k+1) + Kk(k+1) . (7.10)

Here, Ik(k+1) is the identity, Pk(k+1) exchanges sites k, k + 1 and Kk(k+1) is a SO(6) trace
over sites k, k + 1. Graphically this may be represented as in Fig. 4.
We introduce a generator
L
X
U2 = U2,k(k+1)(k+2) , U2,k(k+1)(k+2) = [D2,(k+1)(k+2) , D2,k(k+1) ]. (7.11)
k=1

Graphically it may be represented as in Fig. 5. This generator is easily seen to anticom-

mute with parity (it has negative mirror symmetry w.r.t. the vertical axis, see Fig. 5).
A straightforward but tedious calculation shows that U2 indeed commutes with D2 . We
will explain this fact in terms of integrability in the next subsection, cf. (7.17).
We note that U2 interchanges O of (7.3) with O+,1 of (7.4),

U2 O+,1 = 60O , U2 O = +4O+,1, (7.12)

so the generator U2 is indeed responsible for the degeneracy of their scaling dimensions.
The same is true for all the other pairs of operators we observed. The unpaired operators,
for example (4.1), are annihilated by U2 . As an aside we note thatthe eigenoperators
are 160 O+,1 2i O 17 . The corresponding eigenvalues of U2 are 4i 15 in this case.
Thus we have proven the existence of an additional abelian symmetry in the planar
sector at the one-loop level. The symmetry generated by U2 , however, cannot be compact:
The eigenvalues of U2 are not integer multiples of a common number. Nevertheless,
one is led to believe that there exists a SO(2) symmetry (not generated by U2 ) which
has uncharged singlets and charged doublets. Together with the parity operation P
it would form the group O(2). In this scenario, when N1 corrections are included, the
group O(2) breaks to the parity Z2 . It would be very desirable to understand this
17
This could also be inferred by demanding that the eigenoperators be orthonormal and related by P .

31
degeneracy/symmetry better, in terms of N = 4 SYM as well as in terms of the AdS/CFT
correspondence. Another peculiarity of the planar sector is a recently found degeneracy in
four-point functions [60]. There it was observed that for N a four-point function at
one-loop could be described by a single function, although by superconformal symmetry,
two functions would be allowed. Four-point functions are related to three-point functions
and anomalous dimensions by means of the operator product expansion. Therefore these
issues might be related in some way. It would also be of interest to study whether the
degeneracy can be observed as a symmetry of planar n-point functions.

7.3 Higher charges of the spin chain

A much more enlightening way to prove [D2 , U2 ] = 0 is to make use of integrability of
the spin chain. In [37] it was shown that the interaction (7.10) has just the right relative
coefficients to exhibit integrability. The corresponding R matrix is
R0k (u) = P0k (1 32 u + 21 u2 )I0k + u(1 32 u) 21 D2,0k + 21 u2 ( 12 D2,0k )2 ,


(7.13)
where D2,0k is the local dilatation generator (7.10) acting on an auxiliary site 0, see [37]
for details. This R matrix satisfies the Yang-Baxter equation for the SO(6) case [61,62]
R12 (u)R13 (u + v)R23 (v) = R23 (v)R13 (u + v)R12 (u), (7.14)
and thus gives rise to an integrable spin chain. Integrability predicts the existence of
a tower of commuting charges tn . The zeroth charge is the cyclic shift, it equals the
identity in the zero (angular) momentum sector. The first charge is identified with the
dilatation generator
1
t1 = 2N D2 23 L, (7.15)
up to a constant. Using the expressions in [37] we find for the second charge
t2 = 81 U2 + 12 (t1 )2 85 L. (7.16)
From the fact that t1 and t2 commute we easily derive
[D2 , U2 ] = 0. (7.17)
We have seen that the second charge of the integrable spin chain has important
consequences. It is thus natural to investigate the third charge t3 . Up to polynomials
in t1 , t2 and L we find some generator Q3,2 which commutes with D2 , U2 and parity
P . In contrast to U2 we find that Q3,2 does not pair up operators, it simply assigns a
number (charge) to each operator. This is in fact what is to be expected. The reason
why U2 was interesting is that it anticommutes with P while D2 commutes thus giving
rise to pairs. The next charge, t4 , will again give rise to some new generator, Q4,2 , that
anticommutes with parity. This generator will relate the same pairs, only with different
coefficients (charges). Due to (7.15) we know that the spectrum of t1 is related to the
spectrum of one-loop anomalous dimensions. A natural question to ask is whether the
spectra of the higher charges Q2,2 = U2 , Q3,2 , . . . , have a physical meaning in the gauge
theory. Except for special classes of operators, the two-impurity operators for example,
we find no obvious relation between the spectra of D2 and Q3,2 . Thus, there might be
some non-trivial information contained in the higher charges. Clearly, the deep question
is why integrability emerges from the planar gauge theory.

32
8 Integrability at higher loops
In the previous section we have found planar parity pairs and justified their existence
using integrability at the one-loop level. At the two-loop level integrability has not been
established yet. An obstacle in doing so is that two-loop interactions are interactions
of next-to-nearest neighbors, whereas an integrable spin chain usually involves nearest
neighbor interactions only. Although some next-to-nearest neighbor interactions are
included in the tower of higher charges, these cannot be related to the two-loop dilatation
generator, because D2 and D4 do not commute while the spin chain charges do. In order
to construct an integrable spin chain with non-nearest neighbor interactions we cannot
make direct use of the R-matrix formalism and the Yang-Baxter equation. However, we
may use our discovery of degenerate pairs in the previous section as the starting point.
Integrability at one-loop gives rise to a conserved charge U2 . The charge anticommutes
with parity {P, U2 } = 0 and thus pairs up operators. By promoting D2 and U2 to their
full counterparts D and U we may generalize the integrable spin chain to higher loops
or non-nearest neighbor interactions. The observed degeneracy at the two-loop level is
a clear indication that a conserved U exists up to two-loops.
By going to the two-loop level we again restrict ourselves to the operators described
in Sec. 3.1. Effectively this means we consider a XXX1/2 (SU(2) spin 21 ) spin chain
which does not have the trace term. We can now represent all interactions by means of
permutations of adjacent sites. A generic term will be written as
L
X
{n1 , n2 , . . .} = Pk+n1,k+n1 +1 Pk+n2 ,k+n2+1 . . . (8.1)
k=1

In this notation the one-loop dilatation generator (we will drop factors of N in this
section) and the charge U2 are given by



D2 = 2 {} {0} , U2 = 4 {1, 0} {0, 1} . (8.2)

We start by showing that the degeneracy of parity pairs holds at two-loop. From the
assumption that planar N = 4 SYM is integrable at three-loops we derive the cor-
responding dilatation generator. Then we investigate the constraints on the four-loop
dilatation operator and conclude with an analysis of a tower of special three-impurity
operators.

8.1 Two-loops
At the two-loop level we have found that the dilatation generator D receives the correc-
tion

D4 = 2 4{} + 6{0} ({0, 1} + {1, 0}) , (8.3)
and we expect that also U receives some correction U4 . For the degeneracy to hold at
the two-loop level we have to find a U4 such that

[D4 , U2 ] + [D2 , U4 ] = 0, and {P, U4 } = 0. (8.4)

33
We find that this is satisfied by


U4 = 8 {2, 1, 0} {0, 1, 2} . (8.5)
This proves the degeneracy of anomalous dimensions at two-loops. In performing the
commutator we were required to make use of the impossibility of three antisymmetric
spins in SU(2). This indicates that for spin chains with next-to-nearest neighbor in-
teractions, integrability requires that for each site there can only be two states, i.e. a
SU(2) spin chain. This, however, does not exclude that integrability extends to all six
scalars of SYM. When the other scalars are included, we have to consider fermions and
derivative insertions as well. This might be, roughly speaking, a SU(2, 2|4) spin chain.
The interactions involving all relevant fields might then be integrable. At least for the
superpartners of the operators under consideration this must indeed be the case due to
supersymmetry.
In addition we have found the extension of the third charge Q3 to two-loops, see
App. F. This suggests that also the higher charges will generalize to at least two-loops.
If this is indeed so it justifies our claim that planar N = 4 SYM at two-loops is integrable.
It is thus natural to conjecture that integrability holds at all-loops. This would mean
there exist infinitely many commuting charges Qk with loop expansion
 2 l
X gYM
Qk = 2
Qk,2l , (8.6)
l=0
16

where Q1 = D and Q2 = U. Considering the structure of the first few known terms
we assume that Qk,2l is composed of up to k + l 1 permutations involving up to k + l
adjacent sites. For k odd the interactions are symmetric and parity-conserving, and for
k even they are antisymmetric and parity-inverting. Note that in the latter case this
implies a vanishing constant piece.

8.2 Three-loops
In this subsection we investigate the constraints on the planar dilatation generator at
three-loops due to integrability. This requirement together with the correct behavior in
the BMN limit, (6.15), fixes D6 completely. This is not supposed to imply that the D6
we derive is the correct planar dilatation generator of N = 4 SYM. It may well be that
the degeneracy of pairs is broken at three-loops.
As building blocks for the three-loop interaction we allow symmetric (with respect to
the order of permutations {n1 , . . . , nk } 7 {nk , . . . , n1 } to ensure a real spectrum) and
parity-conserving (positive symmetry under {n1 , n2 , . . .} 7 {n1 , n2 , . . .}) structures.
These should act on only four adjacent sites and have no more than three permutations.
This follows from the general structure of a connected three-loop vertex. We find exactly
six structures that satisfy these constraints. Using the planar BMN limit (6.15) we can fix
the coefficients of four of these structures. The requirement of pairing for three-impurity
operators at dimension 7 and 8, see Sec. 7.1 and (D.13) fixes the remaining two. We find

D6 = 4 15{} 26{0} + 6 ({0, 1} + {1, 0}) + {0, 2} ({0, 1, 2} + {2, 1, 0}) . (8.7)

34
The associated correction to U can be found in much the same way analyzing all possible
antisymmetric, parity-inverting structures and requiring that D and U commute. This
fixes U6 up to a contribution proportional to the fourth charge at one-loop Q4,2 . Acting
on the Konishi descendant Tr[, Z][, Z] we find the three-loop planar contribution to
its scaling dimension
2 4
3gYM N 3gYM N 2 21gYM
6
N3
K = 4 + + , (8.8)
4 2 16 4 256 6
subject to the assumption made at the beginning of this subsection. It is also worth
noting the contribution to a general two-impurity operator. Up to a piece proportional
to (D2 )3 we find some contribution in the upper-left 3 3 block (and equivalently in the
lower-right 3 3 block)

+9 10 +1
10 +10

D6 = 18 (D2 )3 + 4 +1
1 ,

(8.9)

0

..
.

in analogy to (5.15). Considering now both D4 and D6 as perturbations of D2 (cf.

Sec. 5.2) we can obtain the planar three-loop contribution to the anomalous dimension
of the operators (5.13). We get one contribution from the diagonal part of D6 and one
contribution from the off-diagonal part of D4 as encoded in the usual formula for second
order non-degenerate perturbation theory. The final result reads
2 n  !
6
g N 3
n 1 cos J+1 n
Jn = YM6 sin6 + 2
3J + 2(J + 6) cos2 , (8.10)
J + 1 8 4(J + 1) J +1

and it produces the correct result for all values of J, n, even for the Konishi (8.8) at
J = 2, n = 1.
Using the planar three-loop dilatation generator it might be possible derive the non-
planar version by analyzing all possible diagrams and fitting their coefficients to non-
renormalization theorems and the planar version in analogy to Sec. 3.

8.3 Four-loops
Having convinced ourselves of the usefulness of the constraints from the pairing and the
BMN limit we cheerfully proceed to four-loops, just to find that all coefficients but one
are fixed. Unfortunately, the left-over coefficient does influence most four-loop anomalous
dimensions, in particular the one of the Konishi descendant K 18 . The consideration of
higher charges might fix this remaining parameter. Nevertheless, we will see below that
we are able to find special operators for which the four-loop scaling dimensions turn out
to be independent of the so far undetermined parameter.
18
It is not even clear how to treat K , because the interaction is longer than the spin chain in this
case.

35
 At four-loops we find in total twelve independent structures. Five coefficients can be
fixed using the planar BMN limit (6.15). Further five coefficients are fixed by demanding
degeneracy of pairs. An expression for D8 involving the remaining two coefficients ,
can be found in App. F. Evaluating some eigenvalues of D up to four-loops one observes
that they do not depend on the parameter . This behavior is however expected. At
four loops we have the freedom to rotate the space of operators with the orthogonal
transformation generated by the antisymmetric generator [D2 , D4 ]. This gives rise to the
following similarity transformation

D = (1 + g 6 [D2 , D4 ])1 D (1 + g 6 [D2 , D4 ]). (8.11)

The first term in D due to the transformation is

g 8 D2 , [D2 , D4 ] ,
 
(8.12)

and this is proportional to what multiplies in D8 . Now, as D and D are related by a

similarity transformation, their eigenvalues are equal and only affects the eigenvectors,
but not the eigenvalues.

8.4 A tower of three impurity operators

For 0 odd we observe that there are as many even operators in the representation
[3, 0 6, 3] as there are odd ones. For a complete table of spin chain states, up to
dimension 12, see Tab. 2. If 0 is even the same is true except for one additional
operator with negative parity. This additional operator is

X 0 4

O= (1)k Tr Z k Z 0 3k , (8.13)
k=1

which is an exact planar eigenoperator of D2 with eigenvalue 12N. The lowest dimen-
sional example at 0 = 6 was discussed in Sec. 4.2. It is annihilated by the extra
symmetry generator U2 , i.e. it is unpaired. Note, that for all of these operators two of
the impurities are always next to each other.
The operators O are not exact planar eigenoperators of D4 . Nevertheless, we can
project D4 O to the piece proportional to O and find 36N 2 as the coefficient. Conse-
quently we have found a sequence of operator with planar dimensions
2 4
3gYM N 9gYM N2
= 0 + . (8.14)
4 2 64 4
Interestingly, the anomalous dimension does not depend on the bare dimension 0 . In
that sense, this operator behaves much like the highest mode in the series of two impurity
operators discussed in Sec. 5. Assuming the highest mode number were n = (J + 1)/2
19 2
, the anomalous dimension (5.17) would equal gYM N/ 2 gYM
4
N 2 /4 4 , which is also
19
The corresponding operator does not exist. The highest mode number is [J/2], for which, however,
the anomalous dimension does not differ considerably.

36
independent of 0 . Furthermore, the frequency of the phase factor in (5.13) would also
be extremal as in (8.13).
For a few of the lower dimensional operators we can also evaluate the anomalous
dimensions up to four-loops (subject to the assumptions in deriving the corresponding
vertices). Here, the yet undetermined coefficient does not contribute. Intriguingly we
find
3g 2 N 9g 4 N 2 66gYM6
N 3 645gYM 8
N4
= 0 + YM2 YM 2 + , (8.15)
4 64 1024 6 16384 8
for 0 = 10, 12, 14. For 0 = 8 the four-loop contribution differs
2 4
3gYM N 9gYM N 2 66gYM
6
N 3 648gYM
8
N4
=8+ + , (8.16)
4 2 64 2 1024 6 16384 8
while for 0 = 6 also the three-loop contribution is modified
2 4
3gYM N 9gYM N 2 63gYM
6
N 3 621gYM
8
N4
=6+ + . (8.17)
4 2 64 2 1024 6 16384 8
So the following picture emerges: The k-loop anomalous dimensions for the operators of
dimensions 0 = 2 + 2k + n, n 0 seem to be equal.

9 Conclusions and outlook

The main message of this paper is the proposal that perturbative scaling dimensions in
4D conformal gauge theories should not be computed on a case-by-case basis, using the
standard, very laborious procedure. The latter consists in first working out classical and
quantum two-point functions of a set of fields, subsequently recursively diagonalizing and
renormalizing the fields at each order in the gauge coupling gYM in order to find good
conformal fields satisfying the expected diagonal form in eq.(1.1) to the desired order,
and finally extracting the scaling dimensions from eq.(1.1). Instead one should focus on
the dilatation operator relevant to the general (i.e. possessing an arbitrary engineering
dimension) class of operators under study. Once it is found, to the desired order in gYM ,
the computation of the dilatation matrix becomes a straightforward, purely algebraic
exercise. The subsequent calculation of the eigenvalues of the matrix then yields the
scaling dimensions, while the eigenvectors resolve the mixing problem.
In the present work we illustrated this proposal in a specific example: We obtained the
4
dilatation operator up to O(gYM ) of N = 4 SU(N) Yang-Mills theory for arbitrary trace-
less pure scalar fields, cf. eqs.(1.14). However, we certainly believe that our methodology
is rather general and should therefore be equally applicable to other four-dimensional
conformal gauge theories with less supersymmetry.
In order to save us a significant amount of additional work we did use non-re-
normalization theorems as well as recent results on the scaling dimensions of the so-called
BMN two-impurity operators [17,18,43] to fix constants in our two-loop dilatation oper-
ator. Incidentally, this may serve as an interesting illustration how the AdS/CFT corre-
spondence (here in its latest reincarnation, i.e. the plane wave/BMN correspondence) can
lead to new insights into gauge theory. We then went on to apply our example to a large

37
number of situations of interest, obtaining with ease a host of novel specific results for
anomalous dimensions and for the resolution of mixing of scalar operators. This yields
important new information on the generic structure of N = 4: E.g. the degeneracy of
certain single and double-trace operators leads to unexpected N1 terms in the associated
anomalous dimensions, and to the breakdown of a well-defined double expansion in the
t Hooft coupling = gYM 2
N and the genus counting parameter N12 . One surely very
interesting issue we did not yet address is the interpretation of these effects in the light
of the AdS/CFT correspondence.
Our new results on N = 4 SYM already lead to new insights into the plane wave
strings/BMN correspondence. We were able to extend our quantum mechanical de-
scription of BMN gauge theory [34], deriving from the two-loop dilatation operator the
two-loop contribution to the quantum Hamiltonian. This contribution strongly hinted at
an all genera version of the celebrated all loop BMN square root formula, cf. eqn. (6.16).
Whether this generalization withstands closer scrutiny remains to be seen, of course.
It would also be very helpful to develop techniques similar to the ones presented
here for the efficient evaluation of correlation functions of more than two local fields.
It is e.g. well known that conformal invariance completely fixes the space-time form of
three-point functions (xij = xi xj )

 C
O (x1 )O (x2 )O (x3 ) = +
. (9.1)
|x12 | |x23 | + |x31 | +

Therefore the quantities of interest are the finite structure constants C alone. These
constants appear in the operator product expansion and apart from the scaling dimen-
sions they are the other central quantity in a conformal field theory of local operators20 .
However, in order to really obtain this form one has to use the correct eigenstates of
the dilatation operator. Furthermore, standard perturbative computations are contam-
inated by useless finite and divergent contributions from the perturbative expansion of
the weights . One therefore wonders whether one may generalize our methodology
and develop purely algebraic techniques for directly finding the structure constants. Of
equal interest are four-point functions, which also seem to possess many unexpected,
simplifying features waiting to be explored (see e.g. [25, 60]).
By exploiting the recently discovered description of the planar limit of the one-loop
dilatation operator as the Hamiltonian of an integrable spin chain [37] we pointed out the
existence of a new axial symmetry of N = 4 Super Yang-Mills theory at N = , linking
fields of opposite parity (w.r.t. reversing the order of fields inside SU(N) traces). We were
able to prove the symmetry as a direct consequence of the presence of non-trivial charges
commuting with the Hamiltonian. We furthermore derived this to follow immediately
from the existence of an R-matrix satisfying the Yang-Baxter equation, i.e. from the
integrability discovered in [37]. An exciting open question is the interpretation of this
symmetry (and of the integrability in general) from the point of view of the planar gauge
theory, or possibly from the point of view of its dual string description. This is even more
20
A number of structure constants for two and three impurity BMN operators [23, 63, 25, 26, 64] have
been obtained so far. In terms of the BMN correspondence it would be important to understand their
dual on the string theory side [65].

38
pressing as we found much evidence that this integrable structure extends to two loops
(we did prove that at least two further charges commute with the two loop dilatation
operator) and it is obviously tempting to conjecture that it holds to all loops. One could
therefore hope that a proper understanding of the integrability might lead to the exact
construction of the all-order planar dilatation generator, see also the discussion below.
It would be important to extend the methods of this work to fix the remaining terms
in the dilatation operator eq.(1.5) pertaining to further classes of operators such as scalars
with SO(6) traces, fermions, field strengths or covariant derivatives. The latter would
be interesting in order to study the high spin limit of [66], see also [67]. Here we expect
the superconformal symmetry of the model to be helpful. It would also be fascinating
to investigate whether the reformulation of the planar theory as an integrable spin chain
can be extended to include the other fields.
It is striking that, once the dilatation operator is found, the calculations of anoma-
lous dimension matrices become purely algebraic. However, in order to fully justify its
derivation we did need to take a look at two-point functions (cf. Sec. 2,3). It is natural
to wonder whether there are further shortcuts that leads to the determination of the
terms in the perturbative expansion of the dilatation operator eq.(1.5). For the N = 4
model this is not an unreasonable expectation: First of all, the theory is scale-invariant
on the quantum level and therefore possesses a finite dilatation operator. Thus one could
expect that the renormalization procedure of Sec. 2,3 is only a scaffold that one might
be able to avoid. Secondly, the theorys action is unique, and entirely determined by
the maximal superconformal symmetry SU(2, 2|4). Is there a way to use the symmetry
algebra, possibly paired with some further insights into the N = 4 model, to completely
fix the structure of the dilatation operator (1.5)?
We would certainly like to push our procedure to higher loops. This is not a pointless
exercise. E.g. little seems to be known about the analytic structure of the exact anoma-
lous dimension of low dimensional fields such as Konishi. All we currently have is the
two-loop result eq.(1.16), the planar three-loop conjecture eq.(1.17) and the conjectured
2
(from AdS/CFT) strong coupling behavior (gYM N)1/4 . Knowing further terms in
the perturbative expansion might give essential clues about the convergence structure of
the series (is the radius of convergence zero or finite?) and might allow us to estimate
the strong coupling result by Pade approximants. Incidentally, the Konishi field eq.(1.7)
is particularly interesting as it cannot mix with any other fields and is therefore known
to be an exact eigenstate of the dilatation operator to all orders in perturbation theory.
We have taken first steps towards constraining the complete dilatation operator,
building directly on the consequences of the spin chain picture: Assuming the above
mentioned axial symmetry to hold also beyond two loops we were able to derive a planar
version of the three-loop dilatation operator which in turn allowed us to obtain further
results on anomalous dimensions. A preliminary study of the possible planar four-loop
structures showed that the information which completely fixed the three-loop case leaves
one coefficient of the planar four-loop dilatation operator undetermined. A speculation
of, potentially, tremendous importance would be that imposing the full integrability
structure completely fixes, at each loop order, the exact planar dilatation operator.

39
Acknowledgments
We would like to thank Natan Andrei, Gleb Arutyunov, David Berenstein, Volodya
Kazakov, Thomas Klose, Stefano Kovacs, Ari Pankiewicz, Jan Plefka and Thomas Quella
for interesting discussions. M.S. thanks the Rutgers Physics department for hospitality
while working on parts of this manuscript. N.B. dankt der Studienstiftung des deutschen
Volkes fur die Unterstutzung durch ein Promotionsforderungsstipendium.

A Conventions
We use the following N = 4 supersymmetric action in components
Z 42
1 d x 
S= Tr 1
F F + 12 D m D m 41 g 2 2 [m , n ][m , n ]
4
2 (2)2

+ 12 T D 2i g T m [m , ] ,
D X = X ig [A , X],
F = A A ig [A , A ]. (A.1)

The coupling constant g is related to the common coupling constant of N = 4 SYM by

2 2
gYM gYM
g2 = . (A.2)
4(2)2 16 2

This normalization turns out convenient when evaluating space-time integrals and when
operators with fermions and derivative insertions are considered.
(a)
The fields carry a color structure, m = T a m . We will consider the gauge group
U(N) whose generators T a we have normalized such that
X
Tr T a T b = ab , (T a ) (T a ) = . (A.3)
a

This implies the U(N) fusion and fission rules

Tr T a A Tr T a B = Tr AB, Tr T a AT a B = Tr A Tr B. (A.4)

In this work we make extensive use of variations with respect to a field m , which we
will denote by

m = = T a (a) . (A.5)
m m
It is therefore understood when the variation hits a field, both the variation symbol and
the field are replaced by a color generator T a . Note that the variation symbols m act
only to the right. In a normal ordered word of fields and variations, : . . . :, it is
understood that the variations do not contract to any of the fields within the normal
ordering.

40
B Scalar space-time integrals
We have normalized the scalar propagator to
(1 )
Ixy =   , (B.1)
 (x y)21
1
2

it is a rather convenient normalization when derivatives are taken (e.g. for fermions).
The following integrals are required at one-loop
Z 42
2 d z
Y x1 x2 x3 = 2
Ix1 z Ix2 z Ix3 z ,
(2)
Z 42
2 d z
Xx 1 x 2 x 3 x 4 = 2
Ix1 z Ix2 z Ix3 z Ix4 z , (B.2)
(2)
2 Z 42
z1 d42 z2

1 2 d
Hx1 x2 ,x3x4 = 2 + Ix1 z1 Ix2 z1 Iz1 z2 Iz2 x3 Iz2 x4 .
x1 x3 (2)42
When evaluated in two-point functions they yield [68]
Y00x 1
= ,
I0x (1 2)
X00xx 2(1 3)
2
= , (B.3)
I0x (1 2)2
H0x,0x 2(1 3)( 1)
2
= ,
I0x 2 (1 2)
where
(1 ) (1 ) (1 + )2 (1 3)
= 1  , = = 1 + 6(3)3 + O(4 ). (B.4)
 2 x2  (1 2)2 (1 + 2)
2

C Renormalization at higher loops

To obtain the arbitrary loop correlator we insert all l-loop connected Green functions
W2l in the correlator

+  P 2l +

+ 
O O = exp(W0 ) exp l=1 g W2l (x, , ) O O =0 .
 (C.1)
1
In analogy to (2.6) we change the argument + of W2l (x, + , ) to I0x

+  P 2l 1

+ 
O O = exp(W0 ):exp l=1 g W2l (x, I0x , ) :O O =0 .
 (C.2)
1 +
Alternatively, we could change the argument to I0x . This does not make a differ-
ence as W2l (x, , ) is symmetric in the arguments + and . We would then like
+

to rewrite (C.2) in a convenient form for the conformal structure of the correlator:

+ 
O O = exp(W0 ) exp V (x) O+ O =0 ,


(C.3)

41
where
X
g 2l V2l (x) 1 8
 
V (x) = 48
g V2 (x), [V2 (x), V4 (x)] + . . . . (C.4)
l=1

The terms V2l are defined by the equality of (C.2) and (C.3)

P 2l 1


exp V (x) = :exp l=1 g W 2l (x, I0x , ) :, (C.5)

which will have to be solved perturbatively. All the terms that arise due to normal
ordering of the exponential and the commutator terms in (C.4) need to be absorbed into
the definition of higher order vertices. For example, the two-loop effective vertex is
1
, ): 12 V2 (x)V2 (x) :V2 (x)V2 (x): .


V4 (x) = :W4 (x, I0x (C.6)

The commutator terms in (C.4) were included for convenience, we will explain this issue
below. The symmetry of W2l is translated into the effective equality

exp(W0 )V2l (x) = exp(W0 )V2l+ (x). (C.7)

In fact we can introduce a transpose operation on a generator X by the definition

exp(W0 ) X = exp(W0 ) X T + , (C.8)

In other words, letting X act on is equivalent to letting X T act on + . The symmetry

of the vertices translates to
V2lT (x) = V2l (x). (C.9)
We renormalize the operators according to

O = exp 12 Z(x0 ) O,


(C.10)

with
X
g 2l V2l (x0 ) 1 6
 
Z(x0 ) = 12
g V2 (x0 ), V4 (x0 ) + . . . (C.11)
l=1

This gives

+ 
O O = exp(W0 ) exp V (x) exp 21 Z + (x0 ) O+ exp 21 Z (x0 ) O =0 .




(C.12)

We can commute objects with a + and a index freely and use the transpose operation
(C.8) to make Z + act on instead. We get

+ 
O O = exp(W0 ) exp 12 Z T (x0 ) exp V (x) exp 12 Z (x0 ) O+ O =0 . (C.13)





The vertices V2l (x0 ) in Z(x0 ) are symmetric, (C.9), only the commutator in (C.11) re-
quires special care, because V2 and V4 need to be transformed consecutively. This effec-
tively inverts their order and flips the sign of the commutator:

X
g 2l V2l (x0 ) + 1 6
T
 
Z (x0 ) = 12
g V2 (x0 ), V4 (x0 ) + . . . (C.14)
l=1

42
In a renormalizable field theory the dependence of V2l on x is determined, we write

(1 )
V2l (x) = l V2l , = 1  . (C.15)
 2 x2 
2

We combine the last three exponentials in (C.13) into one with exponent

X
( l 0l )g 2l V2l 1 8
0 )4 V2 , [V2 , V4 ] + . . .
 
48
g ( (C.16)
l=1

The l-loop Green function W2l is expected to have multiple poles at = 0. In a conformal
field theory, however, these poles must have cancelled in the combination V2l as given by
(C.4), (C.5). If so, we can finally send the regulator to zero and find

O O = exp(W0 ) exp log(x20 /x2 )

+  P 2l

+ 
l=1 g D2l O O =0 , (C.17)


with
D2l = l lim V2l . (C.18)
0

Note that the commutator term in (C.16) vanishes due to four powers of from ( 0 )4
opposed to only three powers of 1/ from the V2l . For this cancellation to happen the
commutator terms in (C.4) and (C.11) are necessary21 .
Some comments about the renormalization program are in order. The effective ver-
tices V2l are connected diagrams. They are generated from the Green functions W2l by
removing the normal ordering of an exponential (C.5) and adding commutators (C.4).
One can easily convince oneself that these operations produce connected diagrams. The
same is then true also for the dilatation generator D. Secondly, the program ensures
that the coefficient of the two-point function is given by free-contractions of the unrenor-
malized operators. Thirdly, the effective vertices V2l are symmetric with respect to the
scalar product induced by free contractions, see (C.7). The same holds for the dilatation
generator which consequently has real eigenvalues22 .

D A collection of anomalous dimensions

D.1 Away from the unitarity bounds
The following operators have a form similar to the ones discussed in Sec. 1. As they
contain all six scalar fields and SO(6) traces they mix with operators containing fermions
and derivatives, see Sec. 3.1. This mixing, however, only becomes relevant beyond one-
loop and we can determine their one-loop anomalous dimensions using (1.6).
21
We have investigated all possible terms that arise in a four-loop computation. We found that exactly
the commutator structure in (C.4) was required to obtain a finite, conformally covariant correlator.
22
Some eigenvalues may appear to be complex. This can only happen if the corresponding eigenvector
has zero norm. In this case the operator in fact does not exist. This happens if the rank of the group
is small compared to the size of the operator. Then group identities make some operators linearly
dependent.

43
D.1.1 Dimension 5, [0, 1, 0]
We find one single trace operator with dimension
2
5gYM N
=5+ 2
, (D.0)
4
three single-trace and three double-trace operators with dimensions
2
gYM N
=5+ , (D.1)
4 2
where is a root of the sixth-order equation
     
6 5 50 4 565 3 2440 400
17 + 110 2 335 2 + 475 + 4 2
N N N2 N
   
4850 1600 3750 4000
250 2
+ 4
= 0. (D.2)
N N N2 N4
Two of these operators, a single-trace and a double-trace one, are degenerate in the
planar limit. The degeneracy is lifted by N1 corrections and the t Hooft expansion of
their scaling dimension is subject to the issues discussed in Sec. 4.3.

D.1.2 Dimension 6, [0, 2, 0], planar

We find one pair of operators with dimension
2
7gYM N
=6+ 2
, (D.3)
8
two operators with dimension
2
gYM N
=6+ 2
, (D.4)

and six operators with dimensions
2
gYM N
=6+ , (D.5)
8 2
where is a root of the sixth-order equation

6 43 5 + 731 4 6238 3 + 27936 2 61776 + 52272 = 0. (D.6)

D.1.3 Dimension 6, [1, 0, 1], planar

We find two pairs of operators with dimension
2
gYM N
=6+ 2
, (D.7)

two operators with dimension
2
5gYM N
=6+ 2
, (D.8)
4
44
and three operators with dimensions
2
gYM N
=6+ , (D.9)
8 2
where is a root of the cubic equation

3 23 2 + 158 308 = 0. (D.10)

D.2 Further away from the unitarity bounds

D.2.1 Dimension 6, [0, 0, 0], planar
We find five operators with dimensions
2
gYM N
=6+ , (D.11)
8 2
where is a root of the quintic equation

5 43 4 + 701 3 5338 2 + 18480 21960 = 0. (D.12)

D.3 Three impurities

The following operators are in the SO(6) representations for which mixing with fermions
and derivative insertions is prohibited, see Sec. 3.1. We may apply the dilatation gener-
ator at two-loops (3.1).

D.3.1 Dimension 8, [3, 2, 3], planar

We find a pair of operators with dimension
2 4
gYM N 5gYM N2
=8+ , (D.13)
2 2 64 4
and a single operator with dimension (see Sec. 8.4)
2 4
3gYM N 9gYM N2
=8+ . (D.14)
4 2 64 4

D.3.2 Dimension 9, [3, 3, 3], planar

We find three pairs of operators with dimensions
2
gYM N
=9+ , (D.15)
16 2
where is a root of the cubic equation
2
gYM N
3 34 2 + 360 1176 + (102 2 2100 + 9912) = 0. (D.16)
16 2

45
D.3.3 Dimension 10, [3, 4, 3], planar
We find one operator with dimension (see Sec. 8.4)
2 4
3gYM N 9gYM N2
= 10 + , (D.17)
4 2 64 4
and three pairs of operators with dimensions
2
gYM N
= 10 + , (D.18)
16 2
where is a root of the cubic equation

3 30 2 + 276 768 + gYM

2
N(86 2 1516 + 5984) = 0. (D.19)

D.4 Four impurities

D.4.1 Dimension 8, [4, 0, 4], planar
We find three operators with dimensions
2
gYM N
=8+ , (D.20)
16 2
where is a root of the cubic equation
2
gYM N
3 40 2 + 464 1600 + 2
(128 2 2720 + 12800) = 0. (D.21)
16

D.4.2 Dimension 9, [4, 1, 4], planar

We find a pair of operators with dimension23
2 4
5gYM N 15gYM N2
=9+ , (D.22)
8 2 128 4
and two operators with dimensions



g2 N 3 3 4
gYM N 2 18 9 3
= 9 + YM (D.23)
4 2 128 4

D.4.3 Dimension 10, [4, 2, 4], planar

We find two pairs of operators with dimension
2

4


gYM N 11 5 gYM N 2 31 3 5
= 10 + , (D.24)
16 2 256 4
23
This dimension matches (7.5).

46
and six operators with dimensions
2
gYM N
= 10 + , (D.25)
4 2
where is a root of the sixth order equation

6 21 5 + 173 4 711 3 + 1525 2 1603 + 637 (D.26)

2
g N
+ YM 2 67 5 1074 4 + 6409 3 17623 2 + 22078 9947 = 0.


16

D.5 Five Impurities

D.5.1 Dimension 10, [5, 0, 5], planar
We find four operators with dimensions
2
gYM N
= 10 + , (D.27)
4 2
where is a root of the quartic equation

4 15 3 + 78 2 165 + 120 (D.28)

g2 N
+ YM 2 47 3 472 2 + 1430 1305 = 0.


16

E The operator D4
Here we list the exact contributions to the generator D4 acting on OpJ0 ;J1 ,...,Jk , cf. equa-
tions (5.7) and (5.10). We see that in strong contrast to (D2 )2 , the operator D4 only
creates states in the combination O1 O0 ; we write in short
J0 ;J1 ,...,Jk
O10 = O1J0 ;J1 ,...,Jk O0J0 ;J1 ,...,Jk . (E.1)

The operators with J0 = 0 are always annihilated, we assume J0 > 1. The planar part
is
J0 ;J1 ,...,Jk
D4;0 OpJ0 ;J1 ,...,Jk = 4(p,0 + p,J0 p,1 p,J0 1 ) O10 . (E.2)
The operator can split off one trace
p;J1 ,...,Jk ,J0 p J0 p;J1 ,...,Jk ,p
D4;+ OpJ0 ;J1 ,...,Jk = 4p6=0,p6=J0 O10


+ O10
J0 p+1;J1 ,...,Jk ,p1
8p>1 O10
p+1;J1 ,...,Jk ,J0 p1
8p<J0 1 O10
JX
0 1
J Jk+1 ;J1 ,...,Jk+1
0
+ 4(p,0 + p,J0 ) O10 , (E.3)
Jk+1 =1

47
or two traces
J0 p1
J pJk+1 ;J1 ,...,Jk+1 ,p
X
D4;++ OpJ0 ;J1,...,Jk = 4p6=0 0
O10
Jk+1 =1
p1
pJ ;J1 ,...,Jk+1 ,J0 p
X
+ 4p6=J0 O10 k+1
Jk+1 =1
J0 p2
p+1;J1 ,...,Jk+1 ,J0 pJk+1 1
X
4 O10
Jk+1 =1
p2
J p+1;J1 ,...,Jk+1 ,pJk+11
X
0
4 O10 , (E.4)
Jk+1 =1

it can join two traces

k
J0 +Ji ;J1 ,...,
X Ji ,...,Jk
D4; OpJ0 ;J1 ,...,Jk = 4(p,0 + p,J0 ) Ji O10 , (E.5)
i=1

or it can let two traces interact without changing the number of traces
k
J0 +Ji p;J1 ,...,
X Ji ,...,Jk ,p
D4;+ OpJ0 ;J1,...,Jk = 4p6=0 Ji O10
i=1
k
Ji +p;J1 ,...,
X Ji ,...,Jk ,J0 p
+ 4p6=J0 Ji O10
i=1
k
p+1;J1 ,...,
X Ji ,...,Jk ,J0 +Ji p1
4 Ji O10
i=1
k
J0 p+1;J1 ,...,
X Ji ,...,Jk ,Ji +p1
4 Ji O10 . (E.6)
i=1

F Charges of the spin chain

In this appendix we present some of the commuting charges of the non-nearest neighbor
SU(2) spin chain investigated in Sec. 8. We use the notation
L
X
{n1 , n2 , . . .} = Pk+n1,k+n1 +1 Pk+n2 ,k+n2+1 . . . (F.1)
k=1

We make extensive use of the identity

{. . . , n, n 1, n, . . .} = {. . . , . . .} {. . . , n, . . .} {. . . , n 1, . . .}
+ {. . . , n, n 1, . . .} + {. . . , n 1, n, . . .}, (F.2)

48
due the impossibility of antisymmetrizing three sites in SU(2). The following expression
for the higher charges are unique up lower charges multiplied by powers of the coupling
constant.

The first charge D = Q1

D0 = {},
D2 = 2{} 2{0},
D4 = 8{} + 12{0} 2 ({0, 1} + {1, 0}) ,
D6 = 60{} 104{0} + 24 ({0, 1} + {1, 0}) + 4{0, 2} 4 ({0, 1, 2} + {2, 1, 0}) ,
D8 = +(572 + 4){} + (1072 12 + 4){0}
+ (278 + 4 4)({0, 1} + {1, 0}) + (84 + 6 2){0, 2} 4{0, 3}
+ 4({0, 1, 3} + {0, 2, 3} + {0, 3, 2} + {1, 0, 3})
+ (78 + 2)({0, 1, 2} + {2, 1, 0}) + (6 4 + 2)({0, 2, 1} + {1, 0, 2})
+ (1 )({0, 1, 3, 2} + {0, 3, 2, 1} + {1, 0, 2, 3} + {2, 1, 0, 3})
+ (2 2){1, 0, 2, 1} + 2({0, 2, 1, 3} + {1, 0, 3, 2})
10({0, 1, 2, 3} + {3, 2, 1, 0}). (F.3)
The second charge U = Q2
U2 = 4 ({1, 0} {0, 1}) ,
U4 = 8 ({2, 1, 0} {0, 1, 2}) ,
U6 = 8 ({0, 1, 3} + {0, 2, 3} {0, 3, 2} {1, 0, 3})
+ 40 ({0, 1, 2} {2, 1, 0}) + 16 ({3, 2, 1, 0} {0, 1, 2, 3}) . (F.4)
The third charge Q3
Q3,2 = 2{0} + ({0, 1} + {1, 0})
+ ({0, 2, 1} + {1, 0, 2}) ({0, 1, 2} + {2, 1, 0}) ,
Q3,4 = 2{0} + ({0, 1} + {1, 0}) 4{0, 2}
3 ({0, 1, 2} + {2, 1, 0}) + 5 ({0, 2, 1} + {1, 0, 2}) + 2{1, 0, 2, 1}
3 ({0, 1, 2, 3} + {3, 2, 1, 0}) + ({0, 2, 1, 3} + {1, 0, 3, 2})
+ ({0, 1, 3, 2} + {2, 1, 0, 3} + {0, 3, 2, 1} + {1, 0, 2, 3}) . (F.5)
The fourth charge Q4
Q4,2 = 2 ({0, 1, 2} {2, 1, 0})
({0, 2, 1, 3} {1, 0, 3, 2}) + ({0, 1, 2, 3} {3, 2, 1, 0})
+ ({0, 3, 2, 1} {0, 1, 3, 2} {1, 0, 2, 3} + {2, 1, 0, 3}) . (F.6)

References
[1] F. Gliozzi, J. Scherk and D. I. Olive, Supersymmetry, Supergravity theories and the dual
spinor model, Nucl. Phys. B122 (1977) 253.

49
 [2] L. Brink, J. H. Schwarz and J. Scherk, Supersymmetric Yang-Mills Theories,
Nucl. Phys. B121 (1977) 77.
[3] M. F. Sohnius and P. C. West, Conformal invariance in N = 4 Supersymmetric
Yang-Mills Theory, Phys. Lett. B100 (1981) 245.
[4] P. S. Howe, K. S. Stelle and P. K. Townsend, Miraculous ultraviolet cancellations in
supersymmetry made manifest, Nucl. Phys. B236 (1984) 125.
[5] L. Brink, O. Lindgren and B. E. W. Nilsson, N = 4 Yang-Mills theory on the light
cone, Nucl. Phys. B212 (1983) 401.
[6] O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri and Y. Oz, Large N field
theories, string theory and gravity, Phys. Rept. 323 (2000) 183, hep-th/9905111.
[7] E. DHoker and D. Z. Freedman, Supersymmetric gauge theories and the AdS/CFT
correspondence, hep-th/0201253.
[8] V. K. Dobrev and V. B. Petkova, All positive energy unitary irreducible representations
of extended conformal supersymmetry, Phys. Lett. B162 (1985) 127.
[9] L. Andrianopoli and S. Ferrara, Short and long SU(2,2/4) multiplets in the AdS/CFT
correspondence, Lett. Math. Phys. 48 (1999) 145, hep-th/9812067.
[10] L. Andrianopoli, S. Ferrara, E. Sokatchev and B. Zupnik, Shortening of primary
operators in N-extended SCFT4 and harmonic-superspace analyticity,
Adv. Theor. Math. Phys. 3 (1999) 1149, hep-th/9912007.
[11] S.-M. Lee, S. Minwalla, M. Rangamani and N. Seiberg, Three-point functions of chiral
operators in D = 4, N = 4 SYM at large N, Adv. Theor. Math. Phys. 2 (1998) 697,
hep-th/9806074.
[12] G. Arutyunov, S. Frolov and A. C. Petkou, Operator product expansion of the lowest
weight CPOs in N = 4 SYM4 at strong coupling, Nucl. Phys. B586 (2000) 547,
hep-th/0005182.
[13] M. Blau, J. Figueroa-OFarrill, C. Hull and G. Papadopoulos, A new maximally
supersymmetric background of IIB superstring theory, JHEP 0201 (2002) 047,
hep-th/0110242.
[14] M. Blau, J. Figueroa-OFarrill, C. Hull and G. Papadopoulos, Penrose limits and
maximal supersymmetry, Class. Quant. Grav. 19 (2002) L87, hep-th/0201081.
[15] R. R. Metsaev, Type IIB Green-Schwarz superstring in plane wave Ramond-Ramond
background, Nucl. Phys. B625 (2002) 70, hep-th/0112044.
[16] R. R. Metsaev and A. A. Tseytlin, Exactly solvable model of superstring in plane wave
Ramond-Ramond background, Phys. Rev. D65 (2002) 126004, hep-th/0202109.
[17] D. Berenstein, J. M. Maldacena and H. Nastase, Strings in flat space and pp waves
from N = 4 Super Yang Mills, JHEP 0204 (2002) 013, hep-th/0202021.
[18] D. J. Gross, A. Mikhailov and R. Roiban, Operators with large R charge in N = 4
Yang-Mills theory, Annals Phys. 301 (2002) 31, hep-th/0205066.
[19] A. Parnachev and A. V. Ryzhov, Strings in the near plane wave background and
AdS/CFT, JHEP 0210 (2002) 066, hep-th/0208010.
[20] N. Beisert, BMN Operators and Superconformal Symmetry,
Nucl. Phys. B659 (2003) 79, hep-th/0211032.

50
[21] T. Klose, Conformal Dimensions of Two-Derivative BMN Operators,
JHEP 0303 (2003) 012, hep-th/0301150.
[22] C. Kristjansen, J. Plefka, G. W. Semenoff and M. Staudacher, A new double-scaling
limit of N = 4 super Yang-Mills theory and PP-wave strings,
Nucl. Phys. B643 (2002) 3, hep-th/0205033.
[23] N. R. Constable, D. Z. Freedman, M. Headrick, S. Minwalla, L. Motl, A. Postnikov and
W. Skiba, PP-wave string interactions from perturbative Yang-Mills theory,
JHEP 0207 (2002) 017, hep-th/0205089.
[24] U. Gursoy, Vector operators in the BMN correspondence, JHEP 0307 (2003) 048,
hep-th/0208041.
[25] N. Beisert, C. Kristjansen, J. Plefka, G. W. Semenoff and M. Staudacher, BMN
correlators and operator mixing in N = 4 super Yang-Mills theory,
Nucl. Phys. B650 (2003) 125, hep-th/0208178.
[26] N. R. Constable, D. Z. Freedman, M. Headrick and S. Minwalla, Operator mixing and
the BMN correspondence, JHEP 0210 (2002) 068, hep-th/0209002.
[27] B. Eynard and C. Kristjansen, BMN correlators by loop equations,
JHEP 0210 (2002) 027, hep-th/0209244.
[28] U. Gursoy, Predictions for pp-wave string amplitudes from perturbative SYM,
JHEP 0310 (2003) 027, hep-th/0212118.
[29] E. DHoker, D. Z. Freedman and W. Skiba, Field theory tests for correlators in the
AdS/CFT correspondence, Phys. Rev. D59 (1999) 045008, hep-th/9807098.
[30] S. Penati, A. Santambrogio and D. Zanon, Two-point functions of chiral operators in
N = 4 SYM at order g4 , JHEP 9912 (1999) 006, hep-th/9910197.
[31] S. Penati, A. Santambrogio and D. Zanon, More on correlators and contact terms in
N = 4 SYM at order g4 , Nucl. Phys. B593 (2001) 651, hep-th/0005223.
[32] S. Penati and A. Santambrogio, Superspace approach to anomalous dimensions in
N = 4 SYM, Nucl. Phys. B614 (2001) 367, hep-th/0107071.
[33] A. V. Ryzhov, Quarter BPS operators in N = 4 SYM, JHEP 0111 (2001) 046,
hep-th/0109064.
[34] N. Beisert, C. Kristjansen, J. Plefka and M. Staudacher, BMN gauge theory as a
quantum mechanical system, Phys. Lett. B558 (2003) 229, hep-th/0212269.
[35] D. J. Gross, A. Mikhailov and R. Roiban, A calculation of the plane wave string
Hamiltonian from N = 4 super-Yang-Mills theory, JHEP 0305 (2003) 025,
hep-th/0208231.
[36] R. A. Janik, BMN operators and string field theory, Phys. Lett. B549 (2002) 237,
hep-th/0209263.
[37] J. A. Minahan and K. Zarembo, The Bethe-ansatz for N = 4 super Yang-Mills,
JHEP 0303 (2003) 013, hep-th/0212208.
[38] M. Bianchi, S. Kovacs, G. Rossi and Y. S. Stanev, Anomalous dimensions in N = 4
SYM theory at order g4 , Nucl. Phys. B584 (2000) 216, hep-th/0003203.

51
[39] G. Arutyunov, B. Eden, A. C. Petkou and E. Sokatchev, Exceptional
non-renormalization properties and OPE analysis of chiral four-point functions in N = 4
SYM4 , Nucl. Phys. B620 (2002) 380, hep-th/0103230.
[40] D. Anselmi, M. T. Grisaru and A. Johansen, A Critical Behaviour of Anomalous
Currents, Electric-Magnetic Universality and CFT4 , Nucl. Phys. B491 (1997) 221,
hep-th/9601023.
[41] G. Arutyunov, S. Penati, A. C. Petkou, A. Santambrogio and E. Sokatchev,
Non-protected operators in N = 4 SYM and multiparticle states of AdS5 SUGRA,
Nucl. Phys. B643 (2002) 49, hep-th/0206020.
[42] D. Berenstein and H. Nastase, On lightcone string field theory from super Yang-Mills
and holography, hep-th/0205048.
[43] A. Santambrogio and D. Zanon, Exact anomalous dimensions of N = 4 Yang-Mills
operators with large R charge, Phys. Lett. B545 (2002) 425, hep-th/0206079.
[44] E. DHoker, P. Heslop, P. Howe and A. V. Ryzhov, Systematics of quarter BPS
operators in N = 4 SYM, JHEP 0304 (2003) 038, hep-th/0301104.
[45] M. Bianchi, S. Kovacs, G. Rossi and Y. S. Stanev, On the logarithmic behavior in
N = 4 SYM theory, JHEP 9908 (1999) 020, hep-th/9906188.
[46] M. Bianchi, S. Kovacs, G. Rossi and Y. S. Stanev, Properties of the Konishi multiplet
in N = 4 SYM theory, JHEP 0105 (2001) 042, hep-th/0104016.
[47] M. Bianchi, B. Eden, G. Rossi and Y. S. Stanev, On operator mixing in N = 4 SYM,
Nucl. Phys. B646 (2002) 69, hep-th/0205321.
[48] J. Gomis, S. Moriyama and J. Park, SYM description of SFT Hamiltonian in a
pp-wave background, Nucl. Phys. B659 (2003) 179, hep-th/0210153.
[49] J. Gomis, S. Moriyama and J. Park, SYM description of pp-wave string interactions:
Singlet sector and arbitrary impurities, Nucl. Phys. B665 (2003) 49, hep-th/0301250.
[50] M. Spradlin and A. Volovich, Superstring interactions in a pp-wave background,
Phys. Rev. D66 (2002) 086004, hep-th/0204146.
[51] I. R. Klebanov, M. Spradlin and A. Volovich, New effects in gauge theory from pp-wave
superstrings, Phys. Lett. B548 (2002) 111, hep-th/0206221.
[52] M. Spradlin and A. Volovich, Superstring interactions in a pp-wave background. II,
JHEP 0301 (2003) 036, hep-th/0206073.
[53] A. Pankiewicz, More comments on superstring interactions in the pp-wave background,
JHEP 0209 (2002) 056, hep-th/0208209.
[54] A. Pankiewicz and B. Stefanski, Jr., PP-Wave Light-Cone Superstring Field Theory,
Nucl. Phys. B657 (2003) 79, hep-th/0210246.
[55] Y.-H. He, J. H. Schwarz, M. Spradlin and A. Volovich, Explicit formulas for Neumann
coefficients in the plane-wave geometry, Phys. Rev. D67 (2003) 086005,
hep-th/0211198.
[56] R. Roiban, M. Spradlin and A. Volovich, On light-cone SFT contact terms in a plane
wave, JHEP 0310 (2003) 055, hep-th/0211220.
[57] H. Verlinde, Bits, matrices and 1/N, JHEP 0312 (2003) 052, hep-th/0206059.

52
[58] D. Vaman and H. Verlinde, Bit strings from N = 4 gauge theory,
JHEP 0311 (2003) 041, hep-th/0209215.
[59] J. Pearson, M. Spradlin, D. Vaman, H. Verlinde and A. Volovich, Tracing the String:
BMN correspondence at Finite J 2 /N , JHEP 0305 (2003) 022, hep-th/0210102.
[60] G. Arutyunov and E. Sokatchev, On a large N degeneracy in N = 4 SYM and the
AdS/CFT correspondence, Nucl. Phys. B663 (2003) 163, hep-th/0301058.
[61] N. Y. Reshetikhin, A method of functional equations in the theory of exactly solvable
quantum system, Lett. Math. Phys. 7 (1983) 205.
[62] N. Y. Reshetikhin, Integrable models of quantum one-dimensional magnets with O(N)
and Sp(2K) symmetry, Theor. Math. Phys. 63 (1985) 555.
[63] C.-S. Chu, V. V. Khoze and G. Travaglini, Three-point functions in N = 4 Yang-Mills
theory and pp-waves, JHEP 0206 (2002) 011, hep-th/0206005.
[64] G. Georgiou and V. V. Khoze, BMN operators with three scalar impurites and the
vertex- correlator duality in pp-wave, JHEP 0304 (2003) 015, hep-th/0302064.
[65] C.-S. Chu and V. V. Khoze, Correspondence between the 3-point BMN correlators and
the 3-string vertex on the pp-wave, JHEP 0304 (2003) 014, hep-th/0301036.
[66] S. S. Gubser, I. R. Klebanov and A. M. Polyakov, A semi-classical limit of the
gauge/string correspondence, Nucl. Phys. B636 (2002) 99, hep-th/0204051.
[67] A. V. Kotikov, L. N. Lipatov and V. N. Velizhanin, Anomalous dimensions of Wilson
operators in N = 4 SYM theory, Phys. Lett. B557 (2003) 114, hep-ph/0301021.
[68] D. I. Kazakov, Calculation of Feynman integrals by the method of uniqueness,
Theor. Math. Phys. 58 (1984) 223.

53