LPT-Orsay-07-74

Cusp anomalous dimension in maximally supersymmetric

Yang-Mills theory at strong coupling

B. Basso1 , G.P. Korchemsky1, J. Kotanski1,2

1
Laboratoire de Physique Theorique, Universite de Paris XI and CNRS (UMR 8627), 91405 Orsay Cedex, France and
2
Institute of Physics, Jagellonian University, Reymonta 4, PL-30-059 Cracow, Poland

We construct an analytical solution to the integral equation which is believed to describe logarith-
mic growth of the anomalous dimensions of high spin operators in planar N = 4 super Yang-Mills
theory and use it to determine the strong coupling expansion of the cusp anomalous dimension.

1. Introduction: The cusp anomalous dimension is an cusp (g) in powers of g 2 which agrees with the known
important observable in four dimensional gauge theories four-loop result [4]. At intermediate coupling, the inte-
arXiv:0708.3933v3 [hep-th] 24 Jan 2008

ranging from QCD to maximally supersymmetric N = 4 gral equation (2) was solved numerically in [16]. The
Yang-Mills theory (SYM) since it governs the scaling be- obtained solution for cusp (g) was found to be a smooth
havior of various gauge invariant quantities like logarith- function of g that matches for g > 1 the string theory
mic growth of the anomalous dimensions of high-spin prediction (1) with high accuracy. Moreover, an exact
Wilson operators, Sudakov asymptotics of elastic form analytical solution to equation (2) in the limit g
factors, the gluon Regge trajectory, infrared singularities was recently constructed in [17] leading to cusp (g) =
of on-shell scattering amplitudes etc. By definition [1, 2], 2g + O(g 0 ), in agreement with (1) and with the numeri-
cusp (g) measures the anomalous dimension of a Wil- cal analysis of [16]. Equation (2) was analyzed further in
son loop evaluated over a closed contour with a light-like [18] but it resisted an analytical solution so far. In paral-
cusp in Minkowski space-time. It is a function of the lel development, the computation of the two-loop O(1/g)
gauge coupling only and its expansion at weak coupling corrections to the string theory prediction (1) was initi-
is known in QCD to three loops [3] and in N = 4 SYM to ated in [19]. Also, the result (1) was reproduced [20]
four loops [4]. Recently, a significant progress has been from the quantum string Bethe ansatz for a folded string
achieved in determining cusp (g) at strong coupling in rotating in AdS3 S1 .
planar N = 4 SYM. Within the AdS/CFT correspon- In this Letter, we describe an approach to finding a
dence [5], cusp (g) at strong coupling is related to the strong coupling expansion of the solution to equation (2).
semiclassical expansion of the energy of folded string ro- It allows us to determine exact analytical expressions for
tating in AdS3 part of the target space [6, 7] (see also [8]) the coefficients in the 1/gexpansion of the cusp anoma-
lous dimension (1) to any desired order.
3 ln 2 Let us introduce two even/odd functions (t) =
cusp (g) = 2g + O(1/g) , g= , (1)
2 4 (t)
with = gYM 2
Nc being t Hooft coupling. On the gauge et 1 + (2gt) (2gt)

b(t) = + . (3)
theory side, the Bethe ansatz approach to calculating t 2gt 2gt
cusp (g) in the weak coupling limit was developed in
Following [17], we expand (t) into the Bessel function
[9] based on integrability symmetry of planar Yang-Mills
Neumann series
theory to one loop [10]. This approach was recently ex- P k+1
tended to all loops in planar N = 4 SYM. Comparing in- + (t) = k1 (1) (2k)J2k (t)2k , (4)
tegrable structures present on both sides of the AdS/CFT P k+1
correspondence, an all-loop asymptotic Bethe ansatz was (t) = k1 (1) (2k 1)J2k1 (t)2k1 ,
proposed in [11]. It involves a nontrivial scattering phase R
with the expansion coefficients k 0 dt /t Jk (t ) (t )
satisfying the crossing symmetry [12] whose explicit form
( = +/ for k = even/odd) and sign factors introduced
was found in [13]. This led to an integral equation for the
for the later convenience. Substituting (3) into equation
all-loop cusp anomalous dimension [14, 15], the Beisert-
(2) and separating even/odd in t parts, we find after some
Eden-Staudacher (BES) equation,
algebra that (2) is equivalent to the (infinite) system of
 Z 
t 2
equations

b(t) = t K(2gt, 0) 4g dt K(2gt, 2gt )b (t ) , Z  
e 1 0 dt + (t) (t)
(2) t/2g J2n (t) = 0 , (5)
0 t 1 et/2g e 1
with cusp (g) = 8g 2 b(0). Here, the kernel K(t, t ) Z  
dt (t) + (t) 1
is
P expressed in terms of Bessel functions, K(t, t ) = + t/2g J2n1 (t) = n,1 ,
0 t 1 e t/2g e 1 2
n,m=1 znm (g)Jn (t)Jm (t )/(tt ), and its explicit form
can be found in [14]. At weak coupling, the Neumann with n 1. The cusp anomalous dimension can be read
series solution to (2) yields perturbative expansion of from smallt expansion (t) = t cusp (g)/(8g 2 )+O(t2 ).
 2

At weak coupling, one finds from (5) that (t) = J1 (t)+ c

p coefficients we can determine the cusp anomalous di-
O(g 2 ) leading to cusp (g) = 4g 2 + O(g 4 ) in agreement mension cusp (g) = 2g+ (0) = 2g0 as
with the known one-loop result [1, 2].  

2. Exact solution: The system (5) has the following re- X 1 2c
p 3
2c+
p 1
cusp (g) = 2g+ (2p 2 )+ (2p 2 ) .
markable property. Introducing two even/odd functions g p1
t p=1
(t) = (t) (t) coth 4g , or equivalently (11)
  Let us now establish the relation between the coeffi-
t t cients n and n . To this end, we return to the relation
2 (t) = 1 sech (t) tanh (t) , (6)
2g 2g (6) and apply the identities
P
we find from (5) sech t 1 = n1 (1)n a2n t2n , (12)
Z   P n 2n1
dt tanh t = n1 (1) a2n1 t ,
+ (t) + (t) J2n (t) = 0 , (7)
0 t
Z   where acoefficients with even/odd indices are related to
dt
(t) + (t) J2n1 (t) = n,1 , the Euler/Bernoulli numbers, respectively. This leads to
0 t
X
(with n 1) and the cusp anomalous dimension is now 2 (t) = (1)n+1 (13)
n1
given by cusp (g) = 2g+ (0). Here, in comparison with  
(5), the dependence on g only resides in (t). a2n 2n a2n1 2n1
(t/2) (t) 2n1 (t/2) (t) .
At large g, we expect from (6) that the functions (t) g 2n g
admit expansion in the Bessel function Neumann series
Replacing (t) and (t) by the series (4) and (8), re-
P k+1 spectively, we make use of the Bessel function Neumann
+ (t) = k0 (1) J2k (t)2k , (8)
series for (t/2)p Jm (t) in the r.h.s. of (13) to obtain
P k+1
(t) = k0 (1) J2k1 (t)2k1 . m mn+1
X X  
2n1 2n
2m = 2j1 K2m,2j1 + 2j K2m,2j , (14)
In distinction to (4), the first series involves J0 (t) term
n=1 j=0
which ensures that + (0) 6= 0. Indeed, for t 0 one
m mn
X X
gets from (4) and (6) that + (t) t2 and (t) t and, 2n 2n1

2m1 = 2j1 K2m1,2j1 + 2j K2m1,2j .
therefore, + (t) t0 and (t) t. Also, in virtue of
n=1 j=0
J1 (t) = J1 (t), the coefficient in front of J1 (t) is given
by (1 +1 ) so that it is only the sum of two coefficients Here the notation was introduced for the coefficients
that is uniquely defined. We make use of this ambiguity
to choose 1 = 1. n an /g n
Km,j = (15)
Substitution of (8) into (7) yields an infinite system 2(n)
of finite-difference equations for the coefficients k . Ap- ( 12 (m + j + n))( 12 (m j + n))
plying standard methods, we were able to construct its ,
1
( 2 (m + j n) + 1)( 12 (m j n) + 1)
solution for k (with k 1) in the following form (de-
tailed analysis will be published elsewhere) which are given by a product of two identical even/odd
polynomials of degree (n 1) in variables (m j)
1 (0) X 1 h (2p1) i

+ (2p)
k = k +
2
c
gp p k
+ c
p k , (9)
n 2an /g n h n1
ih
n1
i
p=1 Km,j = (m + j) + . . . (m j) + . . . ,
4n (n)
(p)
(16)
where k are basis functions independent on g with ellipses denoting terms with smaller nonnegative
power of (m j). Replacing j in (14) by their explicit
(p) (m + p 12 ) (1)p (m 12 )
(p) expressions, Eqs.(9) and (10), we express 2m and 2m1
2m = , 2m1 = ,
(m + 1)( 21 ) (m + 1 p)( 12 ) in terms of yet unknown coefficients c p . In particular,
(10) the first two terms of their large-g expansion look as
and the expansion coefficients c
p given
P by series in in-
verse powers of the coupling, c = g r
c . The 1 (0) 1 (1), (1),+ 
p r0 p,r k = + 2 c k + c+ k + O(1/g 3 ), (17)
sum over p in the r.h.s. of (9) describes the contri- g k g
bution of zero modes of (7). Their dependence on g (0,1)
1 + +
is fixed by the additional condition of scaling behavior with c = c 16
1 + 2 c1 , c+ = 3 c1 + 2c1 and k given in
(p) (0) (0) 1 (1) (1),
of (t) (see Eqs.(18) and (19) below). Knowing the terms of 2m , Eq.(10), 2m+1 = 2m = 2 2m , 2m1 =
 3

(1), (2) (1),+ (1),+ (3)

2m = 2(m1) and 2m+1 = 2m = 21 2(m1) . The As a function of s, it contains an infinite number of both
k(0) term in (17) is in agreement with findings of [17]. poles and zerosR on the real saxis. Requiring that the

3. Quantization conditions: In our approach, the coef- integrals zp 0 dx xp z+ (x) should be finite for p 0,
R
ficients cp are determined from the behavior of 2m and
we find that, firstly, 0 dx ex/s z+ (x) is an analytical
2m1 , Eq. (14), at large m. To this end, we introduce function of s for Re s > 0 and, secondly, it scales at large
the functions z (x) 2m1 2m and examine their s as z0 z1 /s + O(1/s2 ). To satisfy these conditions in
asymptotic behavior in the double-scaling limit the r.h.s. of (21), it proves sufficient to take
s


X 1 2
m, g , x = (m 41 )2 /g = fixed . (18) p + 1
s cp (p 4 ) = + 3 s

+ O(1/g), (23)
p0
4 2
Employing the approach of [16] and going through nu-
merical analysis of z (x), we found that in the limit (18) with c+ 1
0 = 2 and + the normalization factor. Putting

the solutions to (2) have the following remarkable scaling s = 0 in both sides of (23), we get + = 2[( 34 )]2 . Cal-
R
behavior culating the Laplace transform 0 dx ex/s z (x) in the
" # similar manner and imposing the same conditions as for
(1) z+ (x) we obtain the second quantization condition
(gx)1/4 (0) z+ (x)
z+ (x) = z+ (x) + + O(1/g 2 ) , (19) X 

g gx sp c 3 + 1 1
p (p 4 ) + 2cp p 4 (p + 4 ) (24)
" (1)
#
p0
(gx)3/4 (0) z (x) 2

z (x) = z (x) + + O(1/g ) , s
1 2
4g gx =
+ O(1/g),
s
41 2
(r)
where the functions z (x) (with r 0) do not depend with c +
0 = 0 and c0 = 2 .
1
In comparison with
on g and have faster-than-power decrease at large x. For (21),Pthe Laplace transform of z (x) contains the fac-
(r) n
s

3 s
x 0, smallx expansion of z (x) runs in integer tor a
n=1 n (s) = 2 sin( 2 )/sin( 4 + 2 ) that leads
positive powers of x only. Indeed, matching (17) into to (24). As before, putting s = 0 in both sides of (24)
(0) (0)
(19) we find z+ (x) = 1 + c+ x + O(x2 ) and z (x) = we fix the normalization factor = 41 [( 14 )]2 . Then,
1 + (8c 3c+ )x + O(x2 ). For x , asymptotic be- expanding both sides of the quantization conditions (23)
(r) and (24) around s = 0 and matching the coefficients in
havior of z (x) is controlled by the coefficients c p . The
quantization conditions for c follow from the require- front of powers of s, we determine the coefficients c p
R p
ment 0 dx xp z+ (r)
(x) = finite for any given p , r 0. (with p 1) to the leading order in 1/g. In this way,
(0)
Let us start with the leading term z (x) in the ex- 3 ln 2 1 3 ln 2 1
c+
1 = + + O(1/g) , c
1 = + O(1/g).
pansion (19). From (14) and (9), we evaluate z (x) = 2 4 4
(25)
2m1 2m in the scaling limit (18) and find that the
Substituting these relations into (11), we obtain cusp (g)
sums in (14) receive dominant contribution from large j.
n which coincides with the string theory prediction (1).
This allows us to substitute the Km,j kernel in (14) by
To calculate subleading strong coupling corrections
its leading asymptotic behavior (16)
P andR evaluate sum to cusp (g), or equivalently to determine the coeffi-
over large j in (14) by integration j 7 dj, leading to
cients c p , we expand further the Laplace transforms
R x/s
5 0 dx e z (x) in powers of 1/g and require each term
g 5/4 X + X an xn+p 4
z+ (x) = cp (p 41 ) + ... of the expansion to verify the same analyticity conditions
2
p0 n1
(n + p 41 ) as the leading term. This can be done systematically by
(20) applying the Euler-Maclaurin formula to the sums over
where ellipses denote terms suppressed by powers of 1/g. j in the r.h.s. of (14). In this manner, we obtain the
Then, taking the Laplace transform w.r.t. x we obtain following all-order quantization conditions
Z X   
1 1
 
1

(gs)1/4 sp c+ Q
p p
+
+ c p Q p 1 (26)
dx ex/s z+ (x) = (21) gs gs 2 gs
p0
0 2g

X  X  1 2 s X
sp c+
p (p 1
) a n s n
+ ... = 3 s

(gs)k k+ (1/g),
4 4 2 k0
p0 n1
X   
1
 
1
p + +
The sum over n can be evaluated with a help of (12) as s cp Qp 1 + cp Q p
2 gs gs
p0
s


X
n 2 sin( 2s ) 1 2 X
an s = (1 sec s + tan s) = . (22) =
(gs)k k (1/g),
sin( 4 + 2s ) 41 2s
n1 k0
 4
P r
where k (1/g) = r0 k,r g and the (gindependent) in even powers of g and the expansion coefficients only

functions Qp (x) are of the form involve products of functions of both even and odd
arguments such that the sum of their arguments equals
X X
Q+ xk+l Q2l Q xk+l Q2l+1 the order in g [14, 21].
p (x) = k,p , p (x) = k,p .
k, l0 k, l0 We found that, up to order O(1/g 40 ), all expansion
(27) coefficients of cusp (g) except the first one are negative.
Explicit expressions for the coefficients Qlk,p follow univo- In addition, at large orders in 1/g, they grow factorially
cally from the Euler-Maclaurin summation formula and and the asymptotic expansion is not Borel summable
they are too lengthy to present them here. For g , X (k 1 ) Z
the relations (26) coincide with (23) and (24) for du u1/2 eu

0 (0)1=
2
cusp (g) g = g , (30)
+ 1 1
, Qp (0) = (p 4 ) and Qp (0) = 2 p 4 (p + 4 ). (2g)k 0 u 2g
k
Expanding both sides of (26) in powers of 1/g and s
and matching the expansion coefficients, we can deter- with the Stieltjes integral having a pole at u = 2g. This
mine the functions k (g) and c indicates that cusp (g) receives nonperturbative correc-
p (g) to arbitrary order
in 1/g. Substitution of the resulting expression for c tion g 1/2 e2g proportional to the residue at the pole.
p (g)
into (11) yields the strong coupling expansion of the cusp Our prediction for the cusp anomalous dimension (28)
anomalous dimension. relies on the strong coupling expansion of the solution
4. Strong coupling expansion: Solving the quantization to the BES equation (2). Eventual verification of (28)
conditions (26), we calculated cusp (g) to order O(1/g 40 ). remains a challenge for the string theory. We would
The first few terms of the expansion are like to mention that our result for c2 = K/(4)2 is in a
 structural agreement with the (revised) two-loop super-

string result of [19] and in precise agreement with a new
cusp (g + c1 ) = 2g 1 c2 g 2 c3 g 3 c4 + 2 c22 g 4
superstring computation (R. Roiban and A.A. Tseytlin,

arXiv:0709.0681 [hep-th]).
(c5 + 23 c2 c3 )g 5 c6 +c2 c4 + 54 c23 + 25 c32 g 6
166
7 We are most grateful to A. Gorsky for collaboration
2

7 at an early stage of the project. We would like to thank
c7 + 1721 1431
29 c2 c5 + 7 c3 c4 + 457 c2 c3 g A. Belitsky, J. Maldacena, A. Manashov, A. Tseytlin and
c8 + 6352
c2 c6 + 12606
c3 c5 + 7916 2 6864 2 A. Turbiner for interesting discussions and correspon-
107 29 49 c4 + 7 c2 c4
 dence. This research has been supported in part by the

8

2 4
+4563 c2c3 + 374 c2 g + O g 9
, (28) French Agence Nationale de la Recherche, contract ANR-
06-BLAN-0142 (B.B., G.K.), by the Foundation for Pol-
ish Science and by the MEiN grant P03B-024-27 (J.K.).
where the expansion coefficients are given by
Note added: After this paper was submitted to Arxiv,
3 ln 2 1 27 we learned (J. Maldacena, private communication) that
c1 = , c2 = K, c3 = 11 3 (3), our result for nonperturbative corrections to cusp (g) is
4 16 2 2
21 43065 1605 in a perfect agreement with the findings of Ref. [22].
c4 = 10 4 (4), c5 = 21 5 (5), c6 = 15 6 (6), As was shown in [22], cusp (g), has the interpretation
2 2 2
of an energy density of a certain flux configuration and,
101303055 1317645 as such, it receives correction proportional to m2 with
c7 = (7), c8 = (8), (29)
230 7 222 8 m g 1/4 eg being the mass gap in the O(6) sigma
with (x) the Riemann zeta function, (x) = model.
P n
n0 (1) (2n + 1)x the Dirichlet beta function and
K = (2) the Catalans constant. We verified that the
coefficients (29) are in an excellent agreement with the
numerical values obtained within the approach of [16]. [1] A.M. Polyakov, Nucl. Phys. B 164 (1980) 171.
[2] G.P. Korchemsky and A.V. Radyushkin, Phys. Lett. B
The following concluding remarks are in order. 171 (1986) 459; G.P. Korchemsky, Mod. Phys. Lett. A 4
The reason why in (28) we expanded cusp (g + c1 ) in- (1989) 1257.
stead of cusp (g) is that the c1 dependent terms inside [3] A. Vogt, S. Moch and J.A.M. Vermaseren, Nucl. Phys. B
cusp (g) can be resummed to all orders in 1/g by simply 691 (2004) 129.
replacing g g + c1 . This suggests to use g g c1 as [4] Z. Bern, M. Czakon, L.J. Dixon, D.A. Kosower and
a new parameter of the strong coupling expansion. V.A. Smirnov, Phys. Rev. D 75 (2007) 085010; F. Cac-
hazo, M. Spradlin and A. Volovich, Phys. Rev. D 75
A distinguished feature of the series (28) is that the
(2007) 105011.
coefficients in front of 1/g n are given by a linear combi- [5] J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231;
nation of the product of (2p+1) and (2r) such that sum S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett.
of their arguments equals n. Let us compare this with B 428 (1998) 105; E. Witten, Adv. Theor. Math. Phys.
the weak coupling expansion of cusp (g). The latter runs 2 (1999) 253.

[6] S.S. Gubser, I.R. Klebanov and A.M. Polyakov, Nucl.
Phys. B 636 (2002) 99.
[7] S. Frolov and A.A. Tseytlin, JHEP 0206 (2002) 007.
[8] M. Kruczenski, JHEP 0212 (2002) 024.
[9] G.P. Korchemsky, Nucl. Phys. B 462 (1996) 333;
A.V. Belitsky, A.S. Gorsky and G.P. Korchemsky, Nucl.
Phys. B 748 (2006) 24.
[10] For reviews, see A.V. Belitsky, V.M. Braun, A.S. Gorsky
and G.P. Korchemsky, Int. J. Mod. Phys. A 19 (2004)
4715; N. Beisert, Phys. Rept. 407 (2004) 1.
[11] G. Arutyunov, S. Frolov and M. Staudacher, JHEP 0410
(2004) 016; M. Staudacher, JHEP 0505 (2005) 054;
N. Beisert and M. Staudacher, Nucl. Phys. B 727 (2005)
1.
[12] R. A. Janik, Phys. Rev. D 73 (2006) 086006.
[13] N. Beisert and A. A. Tseytlin, Phys. Lett. B 629 (2005)
102; R. Hernandez and E. Lopez, JHEP 0607 (2006)
004; N. Beisert, R. Hernandez and E. Lopez, JHEP 0611
(2006) 070.
5
[14] B. Eden and M. Staudacher, J. Stat. Mech. 0611 (2006)
P014; N. Beisert, B. Eden and M. Staudacher, J. Stat.
Mech. 0701 (2007) P021.
[15] A.V. Belitsky, Phys. Lett. B 643 (2006) 354.
[16] M.K. Benna, S. Benvenuti, I.R. Klebanov and A. Scardic-
chio, Phys. Rev. Lett. 98 (2007) 131603.
[17] L.F. Alday, G. Arutyunov, M.K. Benna, B. Eden and
I.R. Klebanov, JHEP 0704 (2007) 082.
[18] I. Kostov, D. Serban and D. Volin, hep-th/0703031;
M. Beccaria, G.F. De Angelis and V. Forini, JHEP 0704
(2007) 066.
[19] R. Roiban, A. Tirziu and A.A. Tseytlin, 0704.3638v4
[hep-th] 30 Aug 2007.
[20] P.Y. Casteill and C. Kristjansen, 0705.0890 [hep-th].
[21] A.V. Kotikov and L.N. Lipatov, Nucl. Phys. B 769
(2007) 217.
[22] L.F. Alday and J.M. Maldacena, 0708.0672 [hep-th].