WIS/06/08-MAR-DPP

Holographic Duals of Long Open Strings

arXiv:0803.3547v2 [hep-th] 18 May 2008

Ofer Aharonya and David Kutasovb

a
Department of Particle Physics
The Weizmann Institute of Science, Rehovot 76100, Israel
[email protected]
b
EFI and Department of Physics
University of Chicago, 5640 S. Ellis Ave., Chicago, IL 60637, USA
[email protected]

We study the holographic map between long open strings, which stretch between
D-branes separated in the bulk space-time, and operators in the dual boundary
theory. We focus on a generalization of the Sakai-Sugimoto holographic model
of QCD, where the simplest chiral condensate involves an operator of this type.
Its expectation value is dominated by a semi-classical string worldsheet, as for
Wilson loops. We also discuss the deformation of the model by this operator,
and in particular its effect on the meson spectrum. This deformation can be
thought of as a generalization of a quark mass term to strong coupling. It
leads to the first top-down holographic model of QCD with a non-Abelian chiral
symmetry which is both spontaneously and explicitly broken, as in QCD. Other
examples we study include half-supersymmetric open Wilson lines, and systems
of D-branes ending on N S5-branes, which can be analyzed using worldsheet
methods.

March 2008
1. Introduction and summary

Holographic dualities (generalizing the AdS/CFT correspondence [1]) have proven

to be very useful, both for studying quantum gravity in backgrounds with appropriate
boundaries, and for studying the dual theories living on these boundaries. However, the
dictionary between the boundary theories and the corresponding quantum gravity duals is
not yet complete.
There are two types of objects which we know how to translate between the bulk and
boundary theories (at least in the limit in which the bulk geometry is weakly curved, and
thus well described by supergravity). Local fields in the bulk map to local operators in
the dual boundary theory; sources for these fields map to sources for the corresponding
operators [2,3]. Extended p-dimensional branes in the bulk can end on closed (p − 1)-
dimensional surfaces on the boundary. They correspond to operators in the boundary
theory that are associated with these surfaces.
For example, when the boundary theory is a large N gauge theory, a Euclidean closed
fundamental string worldsheet in the bulk, which ends on a closed loop C on the boundary,
maps to (a locally supersymmetric version of) a Wilson loop in the dual field theory [4-6].
The latter can be thought of as associated with external (infinitely massive) W -bosons in
the gauge theory.
In this note we add another entry to this dictionary. When the bulk background
includes branes extending to the boundary, it is possible for other branes to end on these
branes, and give additional observables in the theory. We will focus on the case where the
bulk contains D-branes, and the branes ending on them are fundamental strings, but the
discussion can be generalized to other systems.
String theory in the bulk contains in this case operators corresponding to open strings
stretched between D-branes near the boundary. There are two qualitatively different
classes of such operators. One corresponds to strings which can shrink to zero size (“short
strings”). These are very similar to the closed string operators mentioned above; their
duals in the boundary theory are local operators, which contain the degrees of freedom
associated with the D-branes. The second class corresponds to “long strings,” that are
stretched between D-branes which are separated by a finite amount near the boundary.
Such operators depend on the choice of an open contour C̃, which ends on the two D-
branes on the boundary. We propose that their duals in the boundary theory are certain
“line operators.”
In the case of large N gauge theories, D-branes ending on the boundary are associated
with fields in the fundamental representation of SU (N ), and the line operators are open
Wilson lines starting from a field in the fundamental representation associated with the first
D-brane, and ending on a field in the anti-fundamental representation associated with the
second one. We propose that an insertion of such an open Wilson line in the field theory

1
corresponds in the bulk to an insertion of an open string ending on the corresponding
contour on the boundary, as in the Wilson loop case. As there, some correlation functions
of these operators are dominated by semi-classical string worldsheets with the appropriate
boundary conditions, and can thus be computed in the supergravity limit.
A case in which long open string operators play an important role is the Sakai-
Sugimoto holographic model of QCD [7], and its generalizations studied in [8,9]. This
model shares with QCD the phenomena of confinement and non-Abelian chiral symmetry
breaking. As we discuss below, open Wilson lines play an important role in understanding
the latter. Previous attempts to study them in this model appeared in [10,11], but our
methods are different.
The Sakai-Sugimoto model describes a 4 + 1 dimensional SU (Nc ) maximally super-
symmetric Yang-Mills (SYM) theory, with ’t Hooft coupling λ5 , compactified on a circle
of radius R (x4 ≡ x4 + 2πR) with anti-periodic boundary conditions for the fermions, and
coupled to Nf left and right-handed fermions in the fundamental representation of SU (Nc )
localized at x4 = −L/2 and x4 = L/2, respectively.
The three parameters with dimensions of length, λ5 , R, and L, can be thought of as
providing an overall scale and two dimensionless couplings on which the dynamics depends
[8,9]. In the region of parameter space λ5 ≪ L ∼ R, the 4 + 1 dimensional gauge theory
is weakly coupled at the scale L ∼ R, and the model is equivalent at long distances (much
larger than L, R, which can be viewed from this perspective as a UV cutoff), to massless
3 + 1 dimensional QCD.
For large λ5 , the 4+1 dimensional gauge theory is strongly coupled and needs to be UV
completed. In string theory this is achieved by realizing the gauge theory as a low energy
theory on a stack of Nc D4-branes wrapped around the twisted x4 circle, intersecting Nf
D8 and D̄8-branes along an IR3,1 , at x4 = ±L/2.
At strong coupling (and large Nc ), one can replace the D4-branes by their near-horizon
geometry, and study the dynamics of the eightbranes in this geometry. One finds that in
the vacuum the U (Nf )L ×U (Nf )R global chiral symmetry associated with the D8 and D̄8-
branes is spontaneously broken to the diagonal U (Nf ), due to the fact that the eightbranes
connect in the bulk. To study this breaking in more detail, one would like to identify an
operator in the field theory that transforms non-trivially under U (Nf )L × U (Nf )R , and
has a non-zero vacuum expectation value (VEV) that preserves the diagonal U (Nf ), i.e.
an order parameter for the symmetry breaking.
Since the left and right-handed fermions are separated in x4 , there are no local gauge-
invariant operators in the D4-brane theory that are charged under both U (Nf ) groups.
The simplest operators with the desired flavor quantum numbers are open Wilson lines of

2
the type discussed above, such as (for a specific choice of the contour C̃)
"Z #
L/2
L L
OWij (xµ ) = ψL
†j µ
(x , x4 = − )P exp (iA4 + Φ) dx4 ψRi (xµ , x4 = ), (1.1)
2 −L/2 2

where Φ is one of the scalar fields of the SYM theory, and P denotes path-ordering.
In the weak coupling regime, the gauge field A4 and scalar Φ are weakly coupled at the
scale L, and the Wilson line in (1.1) can be neglected. Thus, the operator OWij reduces
†j
in this case to the local operator ψL ψRi , which is the familiar order parameter of chiral
symmetry breaking in field theory. In the QCD regime, its VEV is expected to be of order
Λ3QCD , where the QCD scale ΛQCD also sets the scale of masses of mesons and glueballs
in the theory.
At strong coupling, the Wilson line can not be neglected, since the 4 + 1 dimensional
gauge theory degrees of freedom are strongly interacting at the scale L. We will compute
the expectation value of OWij (1.1) below and find that it is exponentially large. For
example, in the original model of [7] (in which L = πR), it scales like exp(λ5 /18πR). We
interpret this exponentially large value as associated with the Wilson line contribution to
(1.1), rather than with the fermions, since such exponentially large values do not appear in
the effective action of the Nambu-Goldstone bosons (the “pions”) and of the other mesons.
Moreover, we will see that the expectation value depends strongly on the choice of contour
C̃ connecting the two intersections.
Another interesting question in the Sakai-Sugimoto model is how to give a mass to
the quarks.1 As explained above, local operators which couple the left and right-handed
fermions are not gauge-invariant in this model. The best we can do is to add to the
Lagrangian the non-local operator (1.1). This breaks the chiral symmetry explicitly, and
in the region in parameter space in which the model reduces to QCD, becomes equivalent
to the quark mass deformation.
On the other hand, at strong coupling where we can use supergravity, this deformation
is highly non-local and irrelevant (i.e. it grows in the UV). At low energies it leads to a
change in the masses of the mesons, and in particular to a non-zero mass for the Nambu-
Goldstone bosons associated with the symmetry breaking. We will study this deformation
to leading order in the deformation parameter, and comment briefly on higher order effects.
In addition to this main example, we present two other examples of long open string
operators. One involves a system of k N S5-branes, with Nf Dp and D̄p-branes a distance
L apart ending on them. For a critical value of the distance, the branes and anti-branes can
connect, and form a single curved D-brane, the hairpin brane of [12-14]. In the process,

1
So far there are no top-down holographic examples of quark masses in theories with a non-
Abelian chiral symmetry.

3
the U (Nf )L × U (Nf )R symmetry acting on the D-branes breaks to the diagonal subgroup,
as before. The long open string stretched between the branes and anti-branes near the
boundary can again be viewed as an order parameter for the breaking. It has a non-zero
VEV that can be computed in the same way as for the generalized Sakai-Sugimoto model,
and one can again study the deformation that breaks the symmetry explicitly.
The main advantage of this example compared to the previous one is its tractability.
The near-horizon limit of the N S5-branes is described by a solvable worldsheet theory
(the linear dilaton conformal field theory (CFT)), and the hairpin boundary state gives
rise to a solvable boundary CFT, due to the fact that it preserves N = 2 superconformal
symmetry on the worldsheet. One can write down explicitly the open string vertex operator
corresponding to the long string, and compare the results of our semi-classical analysis to
those obtained from the effective action of the stretched open strings, and to the exact
solution of the worldsheet CFT.
A second example which we present briefly is of a long open string operator in type
IIB string theory on AdS5 × S 5 with D-brane defects, which preserves half of the super-
symmetry, and is analogous to the circular closed Wilson loop in the d = 4 N = 4 SYM
theory. It is easy to construct many other examples of supersymmetric open string opera-
tors, and it would be interesting to study them in more detail, generalizing the studies of
supersymmetric closed string operators. It would also be interesting to understand if there
is any relation (along the lines of [15-17]) between open Wilson lines of the type studied
here and scattering amplitudes of quarks and gluons.
The organization of this paper is as follows. We begin in section 2 with a general
discussion of open Wilson line operators, and their holographic description. In section 3
we discuss some holographic computations of their correlation functions in the D4 − D8 −
D̄8 system. In section 4 we study the deformation of the Lagrangian of this system by
the operator (1.1), which explicitly breaks the chiral symmetry, to leading order in the
deformation. In section 5 we describe the system of D-branes ending on N S5-branes.
Finally, in section 6 we present a simple example of a supersymmetric long open string
operator, and discuss cusp-like divergences which occur in the computation of correlation
functions of generic long open string operators (both at weak and at strong coupling).

2. Holographic open Wilson lines

As mentioned in the introduction, in this paper we will discuss certain non-local

observables in the context of the AdS/CFT correspondence and its generalizations [1-3].
A class of such observables that has been widely studied is Wilson loops in large N gauge
theories with only adjoint fields. Locally supersymmetric Wilson loops in the fundamental

4
representation dressed with scalar fields Φi ,
 I 
ν µ i ν


W [C] = tr P exp ds iAµ (x (s))ẋ (s) + n (s)Φi (x (s))|ẋ|(s) , (2.1)
C

have been found [4-6] to correspond to strings ending on the closed boundary contour C
(parameterized by xν (s) in the non-compact space-time, and by the unit vector ~n(s) in the
compact space). Thus, an insertion of the operator W [C] on the boundary corresponds in
the bulk path integral to summing over configurations which include a string worldsheet2
ending on the loop C on the boundary. Wilson loops with generic (or no) couplings
to scalar fields are more subtle; in particular, their correlation functions have perimeter
divergences (unlike (2.1)) that need to be regularized. Nevertheless, the operators (2.1)
already give a large amount of information about the theory. For instance, they can be
used as a diagnostic for confinement.
When the boundary theory is a large N gauge theory with a finite number of fields
in the fundamental representation, the corresponding bulk description involves adding D-
branes to the gravity background created by the adjoint fields.3 The gauge symmetry on
the D-branes corresponds to a global flavor symmetry in the dual field theory (which may
or may not be a symmetry of the vacuum). From the (bosonic or fermionic) fundamental
and anti-fundamental fields ψi (x), ψ̄ j (x), one can form local gauge invariant operators
such as ψ̄ j (x)ψi (x). Such operators typically map under holography to local fields in the
bulk, arising from short open strings stretching from the i’th to the j’th D-brane [22].
The situation is different when the D-branes are localized in some of the dimensions
in which the gauge theory lives, and thus give rise to defects. Examples include the
D4 − D8 − D̄8 (generalized Sakai-Sugimoto) model [7-9] mentioned in the previous section,
the closely related intersecting brane systems described in [23,24], and the D3 −D5 system
that corresponds to adding 2 + 1-dimensional hypermultiplets to N = 4 SYM [25,26].
In these cases there are no local gauge-invariant operators that involve fundamental
fields from different brane intersections (separated in space-time). The best one can do is
to consider generalizations of (1.1),
Z 
j j ν µ k ν


OWi [C̃] = ψ̄ (xj )P exp ds iAµ (x (s))ẋ (s) + n (s)Φk (x (s))|ẋ|(s) ψi (xi ) , (2.2)
C̃

where C̃ is a contour between the point xj in the intersection at which the field ψ̄ j lives,
and the point xi in the intersection at which the field ψi lives. This contour is topologically
2
For Wilson loops in higher dimensional representations of the gauge group, the dominant
configurations do not look like strings but rather like other branes carrying the same charges
[18-20].
3
This follows from ’t Hooft’s [21] mapping of Feynman diagrams to string worldsheets, in
which loops of fields in the fundamental representation correspond to holes in the worldsheet.

5
a line segment; thus, we will refer to operators of the form (2.2) as Open Wilson Lines, or
OWLs.
Since locally along the contour C̃ the operator (2.2) looks just like (2.1), when ~n is
a unit vector this operator is locally supersymmetric and its correlation functions do not
exhibit divergences proportional to the length of the path C̃. The holographic dual of
(2.2) must involve a string worldsheet ending on the open contour C̃ on the boundary.
Thus, we propose that an insertion of the operator (2.2) into the path integral of the
boundary gauge theory corresponds in the bulk to summing over configurations which
include an open string worldsheet which approaches the contour C̃ at the boundary of the
bulk space-time, and near the boundary looks like a strip whose ends lie on the i’th and
j’th D-branes.
As we will see, in some cases the computation of correlation functions of these opera-
tors is dominated by a saddle point corresponding to a semi-classical string worldsheet, just
like for many holographic closed Wilson loop computations. In particular, the one-point
function hOWij i is given to leading order in the semi-classical expansion by exp(−A/2πα′ ),
where A is the minimal area of the worldsheet of such a string. If a finite area string world-
sheet does not exist, the one-point function of the OWL vanishes.
A few comments about the preceding discussion are in order:
(1) Just like for other holographic operators, in order to obtain finite correlation functions
one needs to introduce a UV cutoff, and renormalize the OWL operators described
above. In particular, the string worldsheet that enters the calculation of the one-point
function must only have finite area for finite UV cutoff.
(2) When performing the bulk path integral in the presence of the open string worldsheet,
one has to include all the couplings of the string to the background fields, such as the
NS-NS Bµν field, and the gauge fields that live on the D-branes.
(3) When the i’th and/or j’th D-branes give rise to more than one fundamental field
in the gauge theory, the distinction between the corresponding bulk operators in the
semi-classical calculation described above arises from quantization of zero modes on
the worldsheet of the string.
The example that motivated this investigation is the Sakai-Sugimoto model of holographic
QCD. In this model, the large N gauge theory lives on D4-branes in type IIA string
theory, and the fundamental fields are left and right-handed fermions, ψL and ψR , which
are localized at 3 + 1 dimensional defects – the intersections of the D4-branes with Nf D8
and D̄8-branes, respectively. The D8 and D̄8-branes are separated by a distance L in the
direction x4 along the D4-branes. The model has a U (Nf )L × U (Nf )R global symmetry
corresponding to the gauge symmetry on the D8 and D̄8-branes.
In the strongly coupled regime λ5 ≫ L, R, the vacuum of this model corresponds
to a brane configuration in which the D8 and D̄8-branes are connected, and the chiral

6
U (Nf )L × U (Nf )R symmetry is dynamically broken to the diagonal U (Nf ). Most of the
work on the model involved light open 8 − 8 strings, such as the translational modes of the
eightbranes and their worldvolume gauge fields. The latter correspond in the boundary
† †
theory to the U (Nf )L × U (Nf )R chiral symmetry currents ψL (x)σ µ ψL (x), ψR (x)σ̄ µ ψR (x).
These “short string” operators are useful for analyzing the low lying spectrum of
mesons, but in order to study chiral symmetry breaking it is better to consider operators
such as (1.1), which transform as (Nf , N̄f ) under the chiral symmetry. In the next section
we will use holography to show that the expectation value of these operators is non-zero
at strong coupling; thus, they are natural order parameters for chiral symmetry breaking.
In QCD one can break the chiral symmetry explicitly by adding a mass term for the
quarks. The closest analog of this at strong coupling is to add (1.1) to the Lagrangian.
We will describe some results about this deformation in section 4 below.
OWL operators of the form (2.2) can in principle be also defined for theories in which
the fundamental fields are not localized at defects, but they seem to be less useful in such
cases. Consider, for example, the D3 − D7 system, which corresponds to adding to N = 4
SYM a massless hypermultiplet in the fundamental representation of the gauge group. In
the dual description this corresponds [22,27] to adding a D7-brane wrapping AdS5 × S 3 to
type IIB string theory on AdS5 × S 5 , where the S 3 is a maximal three-sphere inside S 5 .
The one-point function of an open Wilson line operator (2.2) involves in this case a
configuration with a string ending on the contour C̃ connecting the points xi and xj in
IR3,1 . The worldsheet of such a string can always reduce its area by contracting towards the
boundary. Therefore, the corresponding one-point function depends on the UV regulator,
and does not appear to be well-behaved.4 This is reasonable from the general perspective
described in the introduction. The open string in question is really a short D7 − D7 string
that does not shrink only because its two ends are held fixed at two different points in
IR3,1 . A more natural basis for describing such strings is in terms of excited perturbative
D7 − D7 strings, rather than the OWL basis (2.2).
The last remark is also applicable to long open strings connecting widely separated D-
branes. For a given pair of branes there is a preferred contour C̃ that has minimal length,
and it is natural to study the OWL operator associated with it. For the D4 − D8 − D̄8
system this is the operator (1.1). One can consider other contours that connect different
points in IR3,1 and/or vary non-trivially in the interior, as in (2.2), but these are less
natural. They can be alternatively described by adding string oscillators to the operator
corresponding to the minimal contour (1.1).

4
This is also true at weak coupling, due to divergences associated with the screening by the
fundamental representation fields.

7
3. Open Wilson lines in the D4 − D8 − D̄8 system
To demonstrate the general discussion of the previous section, we will consider here
the following intersecting brane system in type IIA string theory. We start with Nc D4-
branes stretched in the IR4,1 labeled by (x0 , x1 , x2 , x3 , x4 ), and add to them Nf D8-branes
localized at x4 = −L/2, as well as Nf D̄8-branes localized at x4 = +L/2.
This leads [8] to a non-confining theory of massless left and right-handed fermions, ψL ,
ψR , which are localized at the 4 − 8 and 4 − 8̄ intersections, respectively, and interact via
exchange of modes living on the D4-branes.5 The strength of the interaction is determined
by the ’t Hooft coupling λ5 = (2π)2 gs Nc ls . When the interaction at the scale L is strong
(λ5 ≫ L), one can replace the D4-branes by their near-horizon geometry [29,28]. The
metric is given by
3/2 " 4
 #  3/2
u X RD4
ds2 = 0 2 i 2
 2
du + u2 dΩ24 ,

−(dx ) + (dx ) + (3.1)
RD4 u
i=1

3
where RD4 ≡ πgs Nc ls3 . The RR four-form and dilaton are
 3/4
2πNc Φ u
F(4) = ǫ4 , e = gs . (3.2)
Vol(S 4 ) RD4

The dynamics of the fermions is determined by the shape of the eightbranes in this back-
ground. It was found in [8] that in the lowest energy configuration the D8 and D̄8-branes
are connected by a tube and form a single stack of Nf connected eightbranes. They are
extended in the IR3,1 labeled by (x0 , x1 , x2 , x3 ), wrap the four-sphere labeled by Ω4 , and
form a curve u(x4 ) in the (u, x4 ) plane, which is a solution of the first order differential
equation
u4
q = u40 . (3.3)
RD4

3
1+ u u ′2

The solution of (3.3) is a U -shaped brane, with the distance between the two arms ap-
proaching L at large u. The minimal value of u to which the D8-branes extend, u0 , is
determined by L,
1 3/2 −1/2 9 1
L = RD4 u0 B( , ) . (3.4)
4 16 2
For strong coupling, the curvature of the metric (3.1) near the D8-branes is small. The
string coupling (3.2) diverges as u → ∞, but in the ’t Hooft large Nc limit there is a

5
The Sakai-Sugimoto model [7] is obtained by compactifying x4 on a circle, with twisted
boundary conditions for the fermions on the D4-branes [28]. We will comment on the generaliza-
tion of our considerations to this case below.

8
parameterically wide region in u in which it is small, and we can restrict attention to that
region by placing a UV cutoff on u, u ≤ umax .
Since the D8 and D̄8-branes are connected in the vacuum, the chiral U (Nf )L ×U (Nf )R
symmetry acting on them is spontaneously broken to its diagonal subgroup. The fermions
ψL , ψR , which correspond in the brane picture to strings stretching from the bottom of
the curved D8-branes towards u = 0, obtain a dynamically generated “constituent mass”
m = u0 /2πα′ ∼ λ5 /L2 .
One can use the effective action for the D8-branes (which includes the DBI, Wess-
Zumino, and fermionic terms) to study the low-lying excitations of the model, and in
particular to verify the existence of Nf2 massless Nambu-Goldstone mesons corresponding
to the breaking of the chiral symmetry. The rest of the spectrum is massive; the masses of
the lowest lying mesons are of order 1/L. They are much lighter than the fermions, and
can be thought of as tightly bound states of two fermions.
As explained above, the simplest operator which can serve as an order parameter
for chiral symmetry breaking in this theory is the OWL operator (1.1). We next turn
to a calculation of its one-point function at strong coupling, the chiral condensate, and
comment on more general operators of the form (2.2).

3.1. One-point functions of open Wilson lines

In order to compute the expectation value of the OWL operator (1.1) at strong cou-
pling, we need to perform the gravitational path integral in the closed string background
(3.1), (3.2), in the presence of the curved D8-branes described around (3.3), and of a Eu-
clidean fundamental string worldsheet which near the (regularized) boundary at u = umax
stretches along a straight line in the x4 direction between the two arms of the curved
D8-branes.
This path integral is dominated by a semi-classical contribution of a Euclidean worldsheet,
which is localized at a point in IR3,1 × S 4 , and fills the region in the (u, x4 ) plane between
the boundary at u = umax and the D8-branes (see figure 1).
To leading order in α′ the action of such a string is proportional to its area,
Z umax
1 √ 1
Z Z
4
dx4 umax − u(x4 ) .
 
Sstr = ′
dx du guu g44 = ′
(3.5)
2πα u(x4 ) 2πα

Performing the integral one finds

3/2 1/2
umax L RD4 u0 7 1 umax L λ5
Sstr = ′
− ′
B( , ) = ′
− C1 , (3.6)
2πα 8πα 16 2 2πα L
7 1 9 1
where the constant C1 is given by C1 = B( 16 , 2 )B( 16 , 2 )/128π 2 ≃ 0.0079, and we ne-
glected corrections that go to zero in the limit umax → ∞. The term proportional to

9
 u max

u0 x4
D8

L
Figure 1: The semi-classical worldsheet which gives the chiral condensate in the
D4 − D8 − D̄8 model is drawn in green.

the UV cutoff umax on the right-hand side is independent of the coupling λ5 , and can be
absorbed in the definition of the operator (1.1).6
Thus, we conclude that at strong coupling the expectation value of the operator (1.1)
is given by
hOWij i ≃ δij exp (−Sstr ) ≃ δij exp(C1 λ5 /L) . (3.7)

The calculation above captures the leading behavior of this one-point function at strong
coupling. The first subleading corrections come from quadratic fluctuations around the
Euclidean worldsheet of figure 1, and from the coupling of the string to the varying dilaton.
They are expected to give a polynomial pre-factor in front of the exponential (3.7).
The non-vanishing expectation value (3.7) exhibits the expected pattern of chiral sym-
metry breaking, U (Nf )L ×U (Nf )R → U (Nf )diag , in agreement with our earlier discussion.
It grows exponentially with the coupling λ5 /L in the region λ5 ≫ L in which our calcula-
tion is reliable. At first sight this might seem surprising, since in the weakly coupled field
theory regime the chiral condensate is closely related to the dynamically generated fermion
mass, whereas here this is not the case – the fermion mass scales like λ5 /L2 , those of the
mesons scale like 1/L, while the condensate (3.7) is exponentially large.7 The difference
between the two regimes is that for strong coupling most of the contribution to (3.7) ap-
pears to be due to the Wilson line in (1.1) rather than to the fermion bilinear part of the
operator, while for weak coupling this Wilson line gives a negligible contribution.

6
As explained in [6], this term is naturally canceled by a Legendre transform which is part of
the definition of locally supersymmetric Wilson line operators.
7
The chiral condensate we find is also widely separated from the pion decay constant fπ ,
which will be discussed in the next section.

10
 The fact that the exponential behavior of (3.7) at strong coupling is due to the Wilson
line rather than to the fermions can be seen more quantitatively by studying its dependence
on the contour C̃. Consider, for instance, the one point function of an open Wilson line
(2.2) connecting two points in IR3,1 , xµ0 and xµ1 , (space-like) separated by a distance much
larger than L. A class of contours connecting these points that is useful for our purposes
involves moving first in x4 (at a fixed value of xµ , xµ = xµ0 ) from −L/2 to some x40 , then
varying xµ from xµ0 to xµ1 at fixed x4 , and finally moving again in x4 to L/2. Such contours
have cusps, but these can be smoothed out (and in any case the divergences they lead to
are well understood and can be subtracted out).
Finding the precise shape of the string worldsheet which minimizes the action with
these boundary condition is rather complicated. However, when the two points x0 and x1
are widely separated, we expect the main contribution to this expectation value to come
from the part of the worldsheet at x4 = x40 . In the special case x40 = 0, this part of the
worldsheet is easy to analyze. Its contribution to the regularized action is given by
u0 |~x0 − ~x1 | λ5
Sstr = − ′
∝ − 2 |~x0 − ~x1 | . (3.8)
2πα L
The proportionality constant on the right-hand side can be read off from (3.4). On the other
hand, when x40 approaches (say) L/2, the regularized action turns out to be proportional
to −λ5 |~x0 − ~x1 |/(x40 − L/2)2 .
We see that the expectation values of these operators, proportional to exp(−Sstr ),
grow exponentially with the distance between the endpoints of the contour C̃ in IR3,1 ,
and the coefficient of the distance in the exponent depends on the precise contour we
choose. We conclude that this exponential growth is a property of the contour rather than
of the fermion bilinear at its ends. This also explains why the expectation value under
consideration does not decay exponentially with the distance in IR3,1 , |~x0 − ~x1 |, as one
might have expected due to the fact that the fermions are massive.
So far we have discussed the computation of the chiral condensate for the extremal
D4-brane background, but it is easy to generalize the discussion to the case where x4 lives
on a circle of radius R, with anti-periodic boundary conditions for the fermions. The near-
horizon D4-brane geometry is in this case a Wick-rotated Euclidean black hole geometry,
in which the radius of the x4 circle varies between its asymptotic value R at large u and
zero at u = uΛ = λ5 α′ /9πR2 [28].
One can again analyze the shape of the D8-branes as a function of L and R and
calculate the expectation value of OWL operators such as (1.1), by evaluating the area of
the corresponding Euclidean string worldsheet. There are some small differences in the
precise form of the solutions for the D8-branes and for the strings, but the qualitative
properties are not modified. There are now two independent operators of the form (1.1),
one with the Wilson line going in the positive x4 direction and the other in the negative

11
x4 direction. For generic L, R, the worldsheets that determine the expectation values of
the two operators have different areas, so one of the operators has a larger VEV. In the
special anti-podal case L = πR considered in [7] the one-point functions of both operators
are given by

hOWij i ≃ δij exp(λ5 /18πR) . (3.9)

For general L ≪ λ5 one finds a result that smoothly interpolates between (3.9) for L = πR
and (3.7) for L ≪ R, where the modification of the background at the location of the
D8-branes due to the compactness of x4 is negligible.
The discussion above was restricted to the strong coupling regime λ5 ≫ L. In the
opposite limit, λ5 ≪ L ∼ R, at energies much smaller than 1/R one expects the model
to reduce to QCD with massless quarks. In this limit the gauge field A4 and the scalar
fields are expected to decouple [28], so (1.1) should go over to the usual chiral condensate
of QCD, which scales as Λ3QCD ≃ R13 exp(−16π 3 R/λ5 ). If there is no phase transition as
one varies λ5 /R, we expect a smooth interpolation between this result and (3.9).
For L ≪ R, and in particular in the limit R → ∞ with fixed L, the situation is not
completely clear. At strong coupling (λ5 ≫ L), one finds in this limit a theory which
breaks chiral symmetry but does not confine, which can be thought of as a particular UV
completion of the Nambu-Jona-Lasinio model [8]. Field theoretic intuition suggests that
at weak coupling (λ5 ≪ L) chiral symmetry is not broken, and thus the theory undergoes
a phase transition at some critical value of the coupling λ5 /L, but this has not been
conclusively established yet.
In other closely related brane systems, discussed in [23,24], which give rise to 1 + 1
dimensional intersections, such as the D4 −D6 − D̄6 system, one can analyze the dynamics
for both weak and strong coupling, and in particular calculate the expectation value of
(1.1) in both limits (for any value of R). The strong coupling computation is very similar
to that described above, and gives hOW i ∼ exp(C̃1 λ5 /L) with some calculable constant
C̃1 that depends on L/R and approaches a finite value as L/R → 0.
For weak coupling and infinite R, one obtains in this case a generalized Gross-Neveu
model which can be analyzed using field theoretic methods and gives hOW i ∼ exp(−L/λ5 ).
For finite R one gets a generalization of the ’t Hooft model of two dimensional QCD that
includes four-Fermi interactions, and is solvable at large Nc , like its two extreme limits –
the ’t Hooft and Gross-Neveu models. It would be interesting to compute the chiral
condensate in this model as a function of L/R, and compare it to the strong coupling
calculation described above. We expect a smooth interpolation between the strong and
weak coupling limits as one varies the parameters λ5 /L, λ5 /R that govern chiral symmetry
breaking and confinement, respectively.

12
3.2. Correlation functions of open Wilson lines

The computation of the expectation value of a product of several OWLs (2.2) is also
straightforward in principle, but in practice it is more difficult to find the appropriate semi-
classical string worldsheets (if they exist). As in correlation functions of closed Wilson
loops, in some cases a correlation function of a product of OWLs is dominated by a
single semi-classical worldsheet; in other cases it is dominated by several semi-classical
worldsheets connected by propagators in the bulk (at leading order in 1/Nc they must be
connected by propagators of open string fields); in yet other cases there may be no semi-
classical contribution at all. In the supergravity limit, there can be sharp phase transitions
between the first two possibilities, as in closed Wilson loop correlators [30].

11111
00000 x0 x1 000000000000000
111111111111111
000000000000000
111111111111111
x0 x1

00000
11111 000000
111111 000000000000000
111111111111111
00000
11111
00000
11111 000000
111111 000000000000000
111111111111111
00000
11111 000000
111111
000000
111111 000000000000000
111111111111111
00000
11111 000000
111111 000000000000000
111111111111111
000000000000000
111111111111111
00000
11111 000000
111111 000000000000000
111111111111111
00000
11111
00000
11111 000000
111111
00000
11111 000000
111111
000000
111111
1111111111
0000000000
Figure 2: The two semi-classical configurations that dominate the computation
of F2 . On the left we have the two-string configuration, with one string (as in
figure 1) ending on a D8-brane at x0 and the other at x1 , connected by an open
string propagator inside the D8-brane. On the right we have the single string
configuration. The string worldsheets are filled with diagonal lines, and the D8-
brane lives everywhere but was only drawn at x0 and x1 .

A case where the dominant worldsheets are easy to describe is the correlation function
F2 ≡ hOWii (xµ0 )(OWii )† (xµ1 )i (no sum over i implied) in the D4 − D8 − D̄8 system. There
are two distinct semi-classical contributions to this correlation function. One involves
the worldsheets that appear in the computation of the one-point functions of OWii and
(OWii )† (the worldsheet corresponding to OW † is the same as the one for OW , but with an
opposite orientation), connected by a propagator of an open string field on the D8-brane.
It is depicted in the left part of figure 2. The leading order contribution at large distances
is due to massless pion exchange, and should be proportional to

†
hOWii ihOWii i exp(2C1 λ5 /L)
F2 ≃ 2
≃ . (3.10)
|~x0 − ~x1 | |~x0 − ~x1 |2

The second semi-classical worldsheet smoothly connects the two OWL’s at xµ0 and xµ1 , by
extending into the bulk, as in the right part of figure 2.

13
 If the D8 and D̄8-branes were localized at fixed values of x4 , this configuration would
be precisely the one that appears in the computation of the energy of a quark and an anti-
quark separated by a distance |~x0 − ~x1 |, with x4 playing the role of time (the worldsheet
would simply stretch in this direction and end on the D8-branes at x4 = ±L/2). In the
D4-brane background (3.1), (3.2), this energy is given by [31] (−λ5 /|~x0 − ~x1 |2 ), so in this
case we would obtain
F2 ≃ exp(λ5 L/|~x0 − ~x1 |2 ) . (3.11)

In the actual configuration we are interested in, the D8-branes bend in x4 , and (3.11)
should be modified by taking their shape into account. When the extent of the string in
the radial direction becomes comparable to u0 , this modification is significant. However,
for short distances, (3.11) is still reliable.8
As the distance increases, the area of the worldsheet in the right part of figure 2
increases, and at some point it becomes larger than that in the left part. At that point,
the correlation function in question makes a phase transition from (3.11) to (3.10). This
transition is expected to occur at |~x0 − ~x1 | ≃ L.
An example of a correlator for which there is no obvious smooth worldsheet configu-
ration at short distances is hOWii (xµ0 )OWii (xµ1 )i. The string ending at xµ0 would have to
change its orientation in the bulk before coming back to end at xµ1 . Thus, in this case it
seems likely that the two-string configuration on the left of figure 2 always dominates and
gives the behavior (3.10).
Another interesting correlator is hdet(OWij (xµ ))i, which is a singlet of the non-Abelian
SU (Nf )L × SU (Nf )R but carries axial U (1) charge. In the case of finite R, due to the
axial anomaly, this should be non-zero even in phases where the chiral symmetry is not
spontaneously broken and the D8-branes and D̄8-branes do not connect. The computation
of hdet(OWij )i involves in this case Nf strings ending on the boundary, on the D8-branes
and on the D̄8-branes. Naively it vanishes when the D8-branes and D̄8-branes do not
connect, since the strings have nowhere to end in the IR. However, there are contributions
from Euclidean D0-branes wrapped around the x4 circle (which are instantons from the
point of view of the 4 + 1 dimensional gauge theory). Such Euclidean D0-branes should
have Nf fundamental strings ending on them between the D8-branes and the D̄8-branes9
[32]. These strings can extend to the boundary and thus contribute to hdet(OWij )i. Being
D-instanton effects, such contributions are exponentially suppressed in the ’t Hooft large
Nc limit, but they are the leading contribution to hdet(OWij )i in phases where the non-
Abelian chiral symmetry is unbroken.

8
Note that F2 diverges as ~
x0 → ~
x1 (when the cutoff is sent to infinity).
9
Recall that the D8-branes generate a ten-form flux, which couples to the gauge field on the
D0-branes.

14
4. Deforming by open Wilson lines

In the previous section we computed the expectation value of the OWL operator (1.1)
in the generalized Sakai-Sugimoto model. In this section we will study a deformation of
the model that corresponds to adding this operator to the Lagrangian,

Z Nf
OWjj (x) + c.c. .
X
4
δS = κ d x (4.1)
j=1

This leads to explicit breaking of the U (Nf )L × U (Nf )R chiral symmetry to the diagonal
U (Nf ), in addition to the spontaneous breaking present at κ = 0. The deformation
(4.1) can be thought of as a generalization to strong coupling of a “current mass” for the
P †j
fermions, δS = κ d4 x j ψL
R
ψRj + c.c., that plays a role in QCD. The generalization to
non-equal “masses” κj for different quark flavors is straightforward.
We will study the deformed theory semi-classically at strong coupling, in the hope that
the strong coupling results are smoothly related to large Nc QCD with massive quarks.
We will work to first order in the mass parameter κ; this involves a single insertion of the
perturbation, for which we can use our results from the previous section. In QCD this is
a good approximation for the u and d quarks, whose current mass is much smaller than
the QCD scale. It would be interesting to go beyond first order in κ. For this, one needs
to evaluate n > 1 point functions of the operators OWii (x), which are complicated, as
discussed in the previous section.
To first order in κ, the deformation (4.1) can be described by adding to the space-time
action the term
κ
Z Z X (i)
4 4
δS = 4
d x d Ω e−Sstr + c.c. , (4.2)
Vol(S ) i

(i)
where Sstr is the action of the string ending on the i’th D8-brane discussed in the previous
section,10 and the integral over the four-sphere implements an average over the scalar field
that enters the definition of the operator (1.1), restoring the SO(5) symmetry of the model.
(i)
The deformation (4.2) is non-local11 , since the action Sstr depends on the position
of the D8-branes everywhere in the radial coordinate (it also includes a coupling to the
gauge field on the D8-branes, and to closed string fields). This is not surprising, since the
field theory deformation (4.1) is non-local. In the dual string description, in addition to

10
Of course, this only makes sense in the phase in which the D8 and D̄8-branes are connected.
In phases like the high-temperature phase of the Sakai-Sugimoto model in which the branes are
not connected, hOW i vanishes, and we do not have a semi-classical description of the deformation.
11
Similar non-local mass terms were also recently considered in [33].

15
the explicit non-locality in the direction of the Wilson line, we also have non-locality in
the radial direction.12
The deformation (4.1), (4.2), is of order Nc (or 1/gs ), like any other open string
deformation, so it is expected to influence open string fields (like the position of the D8-
branes) at leading order, while the corrections to closed string fields (like the metric) are
suppressed by a power of gs . One thing that is relatively easy to compute is the mass of
the Nambu-Goldstone bosons (the “pions”) due to the deformation (4.1) at leading order
in κ.
For κ = 0 we have a U (Nf )L × U (Nf )R global symmetry spontaneously broken to
U (Nf ). In the effective field theory on the D8-branes the order parameter for this breaking
can be taken to be the holonomy matrix
Z L/2
U ≡ P exp(i dx4 Ãx4 ) , (4.3)
−L/2

where à is the gauge field on the D8-branes. The matrix U transforms as a bifundamental
of U (Nf )L × U (Nf )R .13 It is precisely the matrix appearing in the low-energy chiral
Lagrangian,14 which is usually written in terms of pion fields as U (x) = exp(iπ(x)/fπ ).
Its low-energy effective Lagrangian is given by

Lef f = (fπ2 /4)Tr(∂µ U ∂ µ U † ) , (4.4)

12
This non-locality could be avoided if instead of deforming by (OW +c.c.) we would deform by
(ln(OW )+c.c.), since this would just shift the action by a multiple of the action Sstr of the stretched
string, which is an integral of a local function of the D8-brane fields (a similar perturbation for
closed Wilson loops was recently considered in [34]). However, such a deformation does not have
the same symmetry properties as (4.1) (in particular it does not break the axial U (1) symmetry),
and it is not obvious that hln(OW )i = ln(hOW i) semi-classically, so we will not consider it further
here.
13
Parameterizing the position of the D8-branes in the (u, x4 ) plane by a variable z which goes
from minus infinity at one boundary of the branes to plus infinity at the other boundary, we can
R∞
write U = P exp(i −∞
dz Ãz ).
14
Naively one might think that the holonomy matrix U could serve as an order parameter for
the chiral symmetry breaking in the full string theory as well. However, while the holonomy is
gauge-invariant in the D8-brane gauge theory, it is not gauge-invariant in the full string theory
under gauge transformations of the NS-NS B field. In order to obtain a gauge-invariant object we
R
must multiply U by exp(i B) where the integral is over a surface bounded by the D8-branes. The
only way to construct an operator containing this phase in string theory is to put in a fundamental
string (or another object with the same charges) ending on the D8-branes, giving precisely the
OWL operators discussed above. Thus, one can think of OW as a completion of U to the full
string theory; in the Nf = 1 case U is the phase of OW .

16
with [7,35]
fπ2 ≃ λ5 Nc /L3 . (4.5)
The Nf2 pion fields in U are massless Nambu-Goldstone bosons.15
The deformation (4.1) explicitly breaks the chiral symmetry to the diagonal subgroup,
and is expected to give a mass to all the Nambu-Goldstone bosons. Indeed, the perturba-
(i)
tion exp(−Sstr ) in (4.2) includes a coupling to the gauge field on the D8-branes, of the form
R L/2
(P exp[−i −L/2 dx4 Ãx4 ])ii . This coupling did not play a role in our evaluation of hOW i,
since we assumed that we were expanding around a configuration in which the gauge field
on the D8-branes vanishes, but it is important in analyzing the perturbed theory (4.2).
The effective Lagrangian for U , (4.4), is deformed at first order in κ by
δLef f = |hOW i|κtr(U ) + c.c. . (4.6)
When κ is real and positive, this is precisely the same as the change in the low-energy
effective action of QCD when we add to the theory a quark mass proportional to κ (the
proportionality constant depends on the chiral condensate). It leads to a pion mass equal
to (this is sometimes called the Gell-Mann-Oakes-Renner relation [38])
4κ|hOW i|
m2π = . (4.7)
fπ2
Note that κ and fπ have dimensions of mass, while OW has the dimension of a mass cubed.
When κ has an imaginary part (or is negative), the minimum of the pion potential is
no longer at π(x) = 0 and the deformation (4.1) leads to a change in the phase of the chiral
condensate. We will assume that κ is positive from here on (the other cases are classically
equivalent to this, since they are related by the axial U (1) symmetry).
In addition to giving a mass to the pions, the perturbation (4.2) changes the masses
of the massive mesons as well. To calculate their mass shifts, one needs to determine the
shape of the D8-branes in the presence of the perturbation. This shape is obtained by
minimizing the deformed Lagrangian for the eightbranes,
s  3  
RD4 1
Z Z
4 4 4
L = TD8 dx u 1 + u′2 + 2 κB exp dx u , (4.8)
u 2πα′
where B is the coefficient of the exponent in the computation of hOW i, which is necessary
to give L the appropriate dimension and to make the second term have the same scaling
O(Nc ) as the first term.16 In the second term in (4.8) we used (3.5).
15
The axial U (1) symmetry is anomalous, and the corresponding pion obtains a mass at order
1/Nc [36,37,7,32].
16
This coefficient also depends on u(x4 ) through the coupling of the worldsheet to the varying
dilaton in our background. However, this dependence is suppressed by a power of α′ in the small
curvature limit we are working in.

17
 The equation of motion corresponding to (4.8) is
" 3 #
3
  
TD8 RD4 11 u κB 1
Z
4
i3/2 uu′′ − u′2 − 4 = exp dx u . (4.9)
h
RD4 3

2 RD4 πα′ 2πα′
1+ u
u′2

The right-hand side is independent of x4 , so the left-hand side is a constant. Denoting this
constant by A ≡ κ|hOW i|/πα′ , (4.9) is equivalent to the first order differential equation

TD8 u4
H=q − Au = constant , (4.10)
RD4 3 ′2


1+ u u

associated with the symmetry {x4 → x4 +constant} of (4.8). The constant value of H may
be determined by requiring that u goes to the UV cutoff u = umax at x4 = ±L/2. It is
related to the minimal position u0 of the D8-branes in the u direction by H = TD8 u40 −Au0 .
Equation (4.10) enables us to compute the deformation in the position of the D8-
branes at leading order in κ|hOW i|. This may then be used to determine the shift of the
meson masses, by analyzing the quadratic fluctuations of the deformed action around this
new solution. It would be interesting to understand how to go to higher orders in κ.
Note that, unlike the QCD mass deformation, the deformation (4.2) in the strongly
coupled D4 − D8 − D̄8 theory is irrelevant, and its effect grows in the UV region; this is
clear from (4.10). Thus, as for other irrelevant deformations, the perturbation expansion
in the deformation is only meaningful if we put in a finite UV cutoff u = umax , and demand
that the deformation is small at the cutoff scale.
It is easy to generalize the computations above to the Sakai-Sugimoto model in which
the x4 direction is compactified. One interesting difference is that, in the special case of
L = πR, it seems natural to deform by the sum of the OWL operator (1.1) corresponding
to the contour connecting the D8 and D̄8-branes in the positive x4 direction, and the one
connecting them in the negative x4 direction. In this case the shape of the D8-branes is
not modified by the deformation, since the two semi-classical strings pull the D8-branes in
opposite directions. Thus, in this special case, adding the “quark mass deformation” does
not change the shape of the D8-branes, but it does give a mass to the “pions” as discussed
above. In all other cases, the shape of the D8-branes is also modified; they are pulled to
larger values of u by the string. In this model the distance between the minimal position
of the D8-branes, u0 , and the minimal value of the u coordinate, uΛ , may be interpreted
as a constituent quark mass (at least in the context of high-spin mesons [39-41]). We find
generically (except for the special case discussed above) that increasing the bare quark
mass increases also the constituent quark mass, as expected.

18
5. D-branes in the background of N S5-branes
In this section we study another example of a holographic description of operators
corresponding to long strings stretched between two D-branes. This example is of interest
for the study of D-brane dynamics near singularities of the bulk geometry. It also has
the advantage that the relevant classical string background is under control, and can be
analyzed exactly in α′ .
Consider the following brane configuration in type II string theory. We start with k
N S5-branes stretched in IR5,1 labeled by (x0 , x1 , x2 , x3 , x4 , x5 ), and located at the origin
in the transverse IR4 . As is well known from the brane construction of gauge theories (see
[42] for a review), Dp-branes which have one direction transverse to the fivebranes can end
on them. Thus, we add a Dp-brane stretched in the directions (x0 , x1 , x2 , · · · , xp−1 ), and
semi-infinite in the x6 direction (i.e. it has x6 ≥ 0 and ends on the fivebranes at x6 = 0).

x6
Dp Dp
x

k NS

L
Figure 3: The brane configuration : a Dp-brane and an D̄p-brane ending on
N S5-branes.

The above D-brane is localized in the IR6−p labeled by (xp , xp+1 , · · · , x5 ). We can add a
second D-brane, which is parallel to the first one, but is displaced from it by a distance L
in IR6−p , and has the opposite orientation, i.e. it is a D̄p-brane. We will label the direction
along which the D and D̄-brane are separated by x, with x(D) = − L2 and x(D̄) = + L2 .
The brane configuration is depicted in figure 3.
We will be primarily interested in the physics associated with the two brane intersec-
tions in figure 3. As reviewed in [42], each of the two intersections separately preserves 8
supercharges,17 and carries no localized massless modes. One way to see this is to com-
pactify some of the directions along the fivebranes, and use U-duality to turn each of the
intersections in the system in question to k D5-branes stretched in (012345) intersecting a
D3-brane stretched in (0126) along an IR2,1 . If the D3-brane is fully extended in x6 , 3 − 5
strings give a massless hypermultiplet in the fundamental representation of the low-energy
17
The system with both branes and anti-branes of course does not preserve any supersymmetry.

19
U (k) gauge symmetry on the fivebranes, localized at the intersection. To reach the config-
uration of interest to us, one needs to separate the two halves of the D3-brane (those with
positive and negative x6 ) along the fivebranes, and to send the lower half to infinity. This
corresponds to giving an infinite mass to the hypermultiplet.
The endpoint of the Dp-brane on the fivebranes looks like a charged object in the
fivebrane theory. For example, for p = 1, the D1-brane ending on the N S5-branes gives
rise to a static quark in the fundamental representation of the low-energy U (k) gauge
theory of k N S5-branes in type IIB string theory. For p = 3, the D-brane is extended in
two of the directions along the fivebranes (12), and looks like a magnetic monopole in the
remaining three.
While the system with just one intersection is uninteresting in the infrared,18 when
both branes and anti-branes are present, as in figure 3, the situation is richer. Since we
are interested in the physics near the intersections, we can replace the fivebranes by their
near-horizon geometry, the CHS geometry [43]:

ds2 = dxµ dxµ + dφ2 + dΩ2 , (5.1)

where φ is related to the radial coordinate in the transverse IR4 as follows:

√
 
φ
r = gs kα′ exp √ , (5.2)
kα′
√
and Ω parameterizes the angular three-sphere in IR4 , whose radius is given by kα′ . More
precisely, the angular degrees of freedom are described by a supersymmetric SU (2) WZW
model at level k. gs is the asymptotic string coupling, far from the fivebranes. The
geometry (5.1) is obtained from the full fivebrane geometry by taking gs → 0 with φ held
fixed; in this limit it describes a “little string theory” (LST) (see [44,45] for reviews). The
dilaton behaves in this limit like
φ
Φ = −√ . (5.3)
kα′
A Dp-brane ending on the fivebranes corresponds in the geometry (5.1) – (5.3) to a brane
stretched in (x0 , x1 , · · · , xp−1 , φ), and localized on the three-sphere and in IR6−p [46]; the
D̄p-brane is described similarly. As in the full geometry, the D and D̄-branes are a distance
L apart in IR6−p . Note that unlike the previous cases we discussed, here this distance does
not grow as we move out in the radial direction.
The Dp and D̄p-branes attract each other via exchange of closed string modes, but we
will ignore this effect, and work just at leading order in the string coupling. We will view
18
In brane constructions of gauge theories, such systems do give interesting infrared physics
when embedded in richer brane configurations; this will not play a role in our discussion below.

20
the distance between the Dp and D̄p-branes at φ → ∞, L, as a fixed (= non-normalizable)
boundary condition. Our focus here will be on the classical dynamics of normalizable open
string modes.
It turns out that for L larger than a certain critical value,
√
Lcrit = π kα′ , (5.4)

all such modes are massive. As L → Lcrit , a light mode appears. For L < Lcrit this mode
becomes tachyonic and destabilizes the brane configuration of figure 3. A heuristic way of
understanding this instability is the following. The endpoints of the Dp and D̄p-branes on
the N S5-branes attract each other via exchange of modes localized on the fivebranes (LST
modes). However, since the tension of the Dp-branes goes like the inverse string coupling,
while the attractive force due to exchange of a particular fivebrane mode is of order one,
this is a subleading effect in (the local) gs .
A classical instability can only occur if the sum over the exchanges of all modes of the
LST diverges. Such a divergence can only be due to the contributions of arbitrarily heavy
LST states. The contribution to the attractive force of a given mode of mass m decreases
at large mass like exp(−mL), while the density of LST states is well known to behave like
√
ρ(m) ∼ exp(2π α′ km). Thus, superficially it seems that the sum over states diverges for
L < 2Lcrit (5.4).
This factor of two discrepancy is familiar from another, closely related, context –
closed string emission from accelerating branes in LST. It was argued in [47] that it is
natural to expect that the density of states that can be emitted by D-branes in LST in
p
fact goes like ρ(m). This would certainly be the case in ordinary (critical) string theory,
since a D-brane can only emit left-right symmetric closed string states. Assuming that this
is the case in LST as well, we conclude that the exchange of LST modes by the D-branes
diverges precisely for L < Lcrit .
In the regime k ≫ 1, Lcrit is large in string units, and the above light mode is best
described as a translational mode of the D-brane configuration (which will be described in
detail below). For k ∼ 1 or smaller19 , a better description of this mode is as a fundamental
string stretched between the D and D̄-branes. We will consider the geometric regime
k ≫ 1, but will return to this stretched string below.
To exhibit the geometric massless mode for k ≫ 1, consider the projection of the
D-branes of figure 3 on the two dimensional space labeled by (φ, x). This corresponds to
a D-string described by a curve x = x(φ). The configuration of figure 3 corresponds to
x = ±L/2; the light mode corresponds to deformations to a more general x(φ). The DBI

19
Such values of k in (5.3) cannot arise in the near-horizon limit of flat N S5-branes, but they
can arise in other systems.

21
action for such a D-brane is given by
 p
φ
Z
S = −C dx exp √ 1 + φ′2 . (5.5)
kα′

Here φ = φ(x), φ′ = ∂x φ, and C is a known constant whose value will not be needed below.
The fact that the Lagrangian (5.5) does not depend explicitly on x implies that one
can integrate the Euler-Lagrange equation once. After squaring the resulting equation one
gets  
2φ
exp √ = 1 + φ′2 , (5.6)
kα ′

where we fixed a constant that appears in the integration to a particular value by shifting
φ. The solution of (5.6) is
   
φ x
exp − √ = cos √ . (5.7)
kα′ kα′

φ
Dp
x

k NS

Lcrit
Figure 4: The hairpin D-brane (5.7).

It describes a U-shaped connected brane, the hairpin brane of [12] (or, more precisely, its
generalization to the fermionic string discussed in [13,14] and other papers). As φ → ∞,
it approaches a brane and anti-brane a distance Lcrit (5.4) apart. As φ decreases, the two
D-branes bend towards each other; they smoothly connect at φ = 0 (see figure 4).
As mentioned above, the position of the bottom of the brane depicted in figure 4 is a
free parameter of the solution, as is clear from the form of the action (5.5). Moreover, the
energy of the brane is independent of this parameter. Thus, when the distance between
the D and D̄-branes at infinity is equal to the critical one (5.4), the mode corresponding

22
to fluctuations of the bottom of the hairpin brane is massless and has a flat potential (at
leading order in the string coupling).
When L > Lcrit , this mode is massive, and the hairpin tends to collapse back to the
brane-antibrane configuration of figure 3. For L < Lcrit it is tachyonic and the bottom
of the U-shape tends to run to large φ. The resulting time-dependent solutions can be
described using techniques similar to those of [48], who studied a closed string analog of
this problem.
The original brane configuration of figure 3 has a U (1) × U (1) symmetry associated
with the two Dp-branes. This is a local symmetry on the D-branes, but from the point of
view of the LST it is a global one. This symmetry is broken to the diagonal U (1) when the
branes connect. It is interesting to ask whether there is an operator that is charged under
the broken U (1) and has a non-zero expectation value in the configuration of figure 4.
Such an operator could serve as an order parameter for the symmetry breaking described
above, as in our discussion of the D4 − D8 − D̄8 system in section 3.
A natural candidate for such an operator is a string stretched between the Dp and
D̄p-branes in figure 3. The lowest lying state of such a string is the open string “tachyon”
stretched between the two branes. From studies of the hairpin brane, which turns out to
be described by an exactly solvable boundary conformal theory, it is known that such an
operator is indeed turned on in the vacuum. In the bosonic string this was discussed in
[12,49], while in the fermionic case of interest to us here in [50].
Asymptotically, at large φ, the worldsheet Lagrangian contains a term corresponding
to a boundary N = 2 superpotential, which behaves like
" r #
k
Z
δSws = µ dtdθ exp − 12 (φ + ix̃) + c.c. . (5.8)
α′

Here x̃ = xL − xR is the T-dual of x, and the coupling µ is determined by φIR , the location
of the bottom of the hairpin brane. The dependence can be determined by a scaling
argument of the kind familiar from Liouville theory. For the hairpin shape (5.7), one has
φIR = 0, which corresponds to some particular µ = µ(∗) . If we replace φ → φ − φIR , such
that the bottom of the hairpin is at φ = φIR , we see from (5.8) that
r !
k
µ = µ(∗) exp 12 φIR . (5.9)
α′

When φIR → −∞, the bottom of the hairpin brane descends into the strong coupling
region and one smoothly approaches the parallel brane-antibrane configuration of figure 3.
µ (5.9) also goes to zero in this limit.20
20
When φIR becomes too small, we cannot trust the shape of the bottom of the hairpin due
to strong quantum effects, but there is no reason to expect non-smooth behavior there.

23
 As mentioned above, the large symmetry of the problem (N = 2 worldsheet supercon-
formal symmetry) allows one to solve the boundary conformal field theory corresponding
to the hairpin brane exactly, and in particular one can deduce the presence of the boundary
N = 2 superpotential (5.8). Thus, it is interesting to study this case in detail, in the hope
of developing techniques which could be useful also in more general circumstances where
the worldsheet theory is not solvable, such as backgrounds with Ramond-Ramond fields
turned on.
In particular, we would like to understand the origin of (5.8) at large k, where both
the closed string background (5.1) – (5.3), and the shape of the D-brane (5.7), are slowly
varying, and we can expect semi-classical techniques to be valid. To do that it is useful
to note that the boundary superpotential (5.8) is a normalizable operator at large φ. As
is familiar from holography in general, µ is proportional to the expectation value of the
non-normalizable operator that creates a string stretched between the D and D̄-branes at
the boundary. This operator, which is analogous to the OWL operators described in the
previous sections, behaves at large φ like
" r ! r #
1 k 1 i k
T ≃ exp ′
−√ φ− x̃ . (5.10)
2 α kα′ 2 α′

Thus, we need to calculate the expectation value of (5.10) in the hairpin state. A scaling
argument similar to that described above implies that if this expectation value is non-zero,
it is indeed proportional to µ (5.9).
To calculate this expectation value it is useful to note that the tachyon background
(5.8) is a non-perturbative effect in the worldsheet theory, whose loop expansion parameter
√
is 1/ k (the curvature of the D-brane). Thus, it is natural to expect that it is due to
a worldsheet instanton effect, involving an open string ending on the boundary; this also
follows from our general discussion in the previous sections of the holographic dual of long
open strings. The instanton in question is a map from the worldsheet disk |z| ≤ 1 to the
part of the two dimensional (x, φ) plane bounded by the hairpin,
   
φ x
exp − √ ≤ cos √ . (5.11)
kα′ kα′

Near the boundary φ → ∞ this worldsheet looks like a string stretched between the D-
branes, which implies that this configuration contributes to the one point function of the
stretched string operator (5.10).
The instanton configuration can be constructed as follows. Start with the worldsheet
action
1
Z
S= d2 z(∂z φ∂z̄ φ + ∂z x∂z̄ x) . (5.12)
πα′

24
It is convenient to parameterize the (x, φ)-plane by the coordinate

φ − φIR + ix
 
U = exp √ , (5.13)
kα′

in terms of which the hairpin shape (5.7) takes the simple form

U + U∗ = 2 , (5.14)

or, equivalently, Re(U − 1) = 0.

The worldsheet action (5.12) now takes the form

k 1
Z
S= d2 z (∂z U ∂z̄ U ∗ + ∂z̄ U ∂z U ∗ ) . (5.15)
π |U |2

The disk instanton we are looking for is a holomorphic map from the disk to the half-plane
bounded by (5.14), and is easy to write down:

1+z
U −1 = . (5.16)
1−z

Its action is proportional to the area A of the Euclidean string worldsheet (5.16):

A
Sinst = . (5.17)
2πα′
This area is infinite, since as φ → ∞ the hairpin looks like two D-strings a distance Lcrit
(5.4) apart, so there is a divergence from that region. This divergence can be regulated by
introducing an upper bound on φ, φUV , which can be thought of as a UV cutoff.
In any case, we are only interested in the dependence of the area on the position of
the bottom of the hairpin, φIR , discussed around (5.9). We can isolate this dependence
by differentiating the area with respect to φIR . A short calculation leads to

∂A
= −Lcrit (5.18)
∂φIR

in the limit φUV → ∞. Therefore, after rescaling the operator (5.10) by a factor which
depends on the UV cutoff, we conclude that
 
Lcrit φIR
hT i ∼ exp (−Sinst ) ∼ exp ∼µ, (5.19)
2πα′

where we used (5.4), (5.9), (5.17), (5.18). We see that indeed the instanton contribution
scales in the right way with φIR to give a non-zero one-point function to the long open string
operator (5.10). Note that we have only computed the leading exponential contribution to

25
the one point function. The pre-exponential factor involves contributions from the dilaton
coupling in the worldsheet action, and the determinant of small fluctuations around the
instanton (5.16). These are subleading in the large k limit, and are expected to give rise
to a constant contribution to (5.19) (independent of φIR ).
So far we discussed the spontaneous breaking of the U (1)×U (1) symmetry of the brane
configuration of figure 3 to the diagonal U (1), by the brane configuration of figure 4. We
have seen that the order parameter for this breaking can be taken to be the stretched string
operator (5.10), and it indeed has a non-zero expectation value in the hairpin state (5.19).
It is natural to ask what happens if we deform the system by adding to the worldsheet
Lagrangian the non-normalizable operator T (5.10),

κ
Z
δSws = dtdθT (φ, x) + c.c. . (5.20)
2

This deformation breaks the U (1) × U (1) symmetry explicitly. It also breaks the N = 2
superconformal symmetry of the hairpin brane; therefore we do not expect the resulting
theory to be exactly solvable. However, one can still ask how the shape of the D-brane
and its low-lying spectrum change in the presence of this deformation.
To first order in κ and in the semi-classical regime k ≫ 1 one can answer this question
by adding to the DBI action (5.5) the exponential of the Nambu-Goto action for the
instanton string discussed above,
 p
φ
Z
S = −C dx exp √ 1 + φ′2 − κBe−SNG . (5.21)
kα′
A
Here, as before (see (5.17)), SNG = 2πα ′ , where A is the area of a minimal worldsheet

enclosed by the deformed hairpin, and B is the pre-exponential factor in the expectation
value of T above. It depends on the shape of the deformed hairpin, but for the purpose of
the calculation below, to leading order in the 1/k expansion we can neglect this dependence.
To calculate the shape of the deformed hairpin to first order in κ we need to solve
the equation of motion of φ(x) with the deformed action (5.21). For this we need the
dependence of SNG on the shape φ(x). It is easy to see that it is given by

1
Z
SNG = − dxφ(x) + · · · , (5.22)
2πα′

where the ellipsis stand for terms that depend on the UV cutoff φUV , but not on the shape
φ(x). Varying (5.21) with respect to φ(x) and integrating once, we find the first order
equation
√φ
e kα′ κhT i
C + φ=D, (5.23)
2πα′
p
1 + φ′2

26
where D is a function of φIR (or, equivalently, of the separation between the brane and
anti-brane at some UV cutoff φUV ). This equation generalizes (5.6) to non-zero κ, and it
can be solved by expanding φ as φ = φ0 + κφ1 + · · ·, and keeping only first order terms in
κ. For example, at large φ, the leading deformation of the hairpin from its original form
is given by
κhT i
 
− √φ
C∂φ x = D − ′
φ e kα′ . (5.24)
2πα
Of course, when φ becomes too large, one has to go beyond the linear approximation in κ
described above.
It is interesting to compare the deformed shape of the hairpin (5.24) which we found
above, to the deformed shape implied by the effective action on the Dp-brane coupled to
the “tachyon” field T . In curved space and for curved D-branes (as in the discussion of
the previous sections) it is not known how to write down such an effective action, but for
flat D-branes in flat space we know how to write it down, and this is the situation in the
asymptotic region of the hairpin. In this region we know that, if we denote the distance
between the brane and the anti-brane by Lcrit − 2x(φ) (where x(φ) is small in the UV),
the mass of the open string ground state is given by
2 r
Lcrit − 2x

2 1 2 k x
m (x(φ)) = − ′ + ′
≃ m0 − . (5.25)
2α 2πα α′3 π

The effective action of the “tachyon” stretched between the D and D̄-branes in figure 4 is
given to quadratic order by
 
φ
Z
(∂φ x)2 + (∂φ T )2 + m2 (x)T 2 .
 
S = −C dφ exp √ (5.26)
kα′
We are looking for a configuration where the normalizable mode of the “tachyon” (5.8) is
turned on with a coefficient hT i, and the non-normalizable mode (5.10) is turned on with
a coefficient κ, such that at leading order in κ the tachyon field behaves asymptotically as
φ
T 2 ≃ βκhT i exp(− √ ), (5.27)
kα′
where β is a constant coming from carefully normalizing the normalizable and non-
normalizable modes of the “tachyon”.
The equation of motion of x(φ) with this “tachyon” source, at leading order in κ (and
in the UV region where x is small), then takes the form
r
βκhT i k
   
φ
∂φ exp √ ∂φ x = − . (5.28)
kα′ 2πα′ α′
p
For 1/β = C k/α′ , this precisely agrees with (5.24) above.

27
6. Additional issues
6.1. A supersymmetric example
The main examples we focused on so far were non-supersymmetric, but one can also
construct interesting examples of OWL operators (2.2) in supersymmetric theories, includ-
ing examples which preserve some of the supersymmetry. We will describe here just one
example, leaving a further investigation to future work.
Consider the d = 4 N = 4 SU (Nc ) SYM theory coupled to Nf three dimensional
massless hypermultiplets living on the surface x3 = 0. In the ’t Hooft large Nc limit
with ’t Hooft coupling λ4 and with fixed Nf , this is described by type IIB string theory
on AdS5 × S 5 , with Nf D5-branes filling an AdS4 × S 2 subspace [25,26]; if we use the
Poincaré coordinates of AdS5 (with a boundary at z → 0),

p
2 ′
dx2µ + dz 2
ds = λ4 α , (6.1)
z2
then the D5-branes are simply located at x3 = 0 (and wrap some maximal S 2 inside the
S 5 ). This theory breaks half of the supersymmetry of the N = 4 SYM theory; it preserves
a d = 3 N = 4 superconformal symmetry.
Now, consider an OWL starting at a hypermultiplet at x0 = x1 = x2 = x3 = 0 and
stretching to infinity in the x3 direction. Such an operator is analogous to the “straight
Wilson line” in the N = 4 SYM theory; it is well-defined if we put appropriate boundary
conditions at infinity. In the holographic dual description, the computation of the one-
point function of this operator is dominated by a string sitting at x0 = x1 = x2 = 0 and
filling the z axis and the positive x3 axis in (6.1) (we assume that the OWL couples to
a scalar such that the string lives at a point in the S 2 filled by the D5-branes). This
operator breaks half of the supersymmetry (leaving 8 unbroken supercharges, including
both regular supercharges and superconformal charges), and the holographic computation
of its VEV gives one, since the regularized area of the surface vanishes (just like for the
“straight Wilson line”).
This case is not very interesting, but suppose that we now perform a conformal trans-
formation involving an inversion around a point x0 = x1 = x3 = 0, x2 = a. This trans-
formation leaves the field theory described above invariant. However, the contour in the
OWL now maps to a semi-circle
1 2 1
(x2 − a + ) + x23 = 2 , x3 ≥ 0 . (6.2)
2a 4a
This is a standard OWL connecting two hypermultiplets of the form (2.2), with a semi-
circular contour (6.2) between the two points (x2 = a, x3 = 0) and (x2 = a − 1/a, x3 = 0).
Our derivation of this configuration by a conformal transformation ensures that this OWL

28
still preserves 8 supercharges, though these are now combinations of standard supersym-
metries and superconformal symmetries.
The holographic computation of the one-point function of this OWL is straightforward;
the dominant solution is just half of the solution for the circular Wilson line [51,6], with a
string worldsheet at

1 2 1
(x2 − a + ) + x23 + z 2 = 2 , x3 ≥ 0 . (6.3)
2a 4a
√
Its area is thus half of that corresponding to the circular Wilson line, which is λ4 , so the
√
VEV of the OWL (at leading order in the α′ expansion) is equal to exp( λ4 /2).
In the case of the closed circular Wilson line case it has been conjectured [52,53] and
recently proven [54] that the result is given by a zero-dimensional matrix model, since the
conformal transformation can only change the result because a point is brought in from
infinity. Similar arguments imply that the semi-circular open Wilson line hOW i described
above should also be computable by a zero-dimensional model of matrices and vectors; it
would be interesting to verify this.

6.2. Divergences in open Wilson line computations

Closed supersymmetric Wilson loops are known to have divergences at cusps, which
can be computed both perturbatively and at strong coupling (with a qualitatively similar
behavior found in both limits [6]). Similarly, in the case that the fields in the fundamental
representation are localized on some subspace, the correlation functions of the open Wilson
line observables (2.2) have a divergence whenever the contour C̃ ends on that subspace at
an angle which is not a straight angle. In this section we describe this divergence both at
weak coupling (using perturbation theory) and at strong coupling (using the mapping to
string worldsheets).
Let us consider a D-dimensional large N gauge theory, in which some fields in the
fundamental representation are localized on a d-dimensional subspace; without loss of
generality we can take this subspace to be

xd+1 = xd+2 = · · · = xD = 0 . (6.4)

When we consider an open Wilson line operator of the form (2.2), starting at a fundamental
field located at x = 0, there is now an angle associated with this operator, which is the
angle θ between the direction of the Wilson line (near x = 0) and the subspace that the
fundamental fields live on. For instance, again without loss of generality, we can assume
that near x = 0 the Wilson line (parameterized by t) looks like

xd+1 = t sin(θ), xd = t cos(θ), t≥0, (6.5)

29
implying that the Wilson line couples to the gauge field components sin(θ)Ad+1 +cos(θ)Ad .
On the other hand, the fields in the fundamental representation couple just to Ad and
they do not couple to Ad+1 . The one-loop diagram involving the exchange of a gauge field
between the Wilson line and the propagator of the field in the fundamental representation
then has a divergence as t → 0, proportional (near θ = π/2) to cos2 (θ), going as dt/tD−3 .
R

For D = 4 we have a logarithmic divergence (as for a cusp in a closed Wilson line), and
for D = 5 a linear divergence. The only case in which there is no divergence is when the
Wilson line intersects the surface (6.4) at a straight angle θ = π/2.
When the fundamental representation fields couple also to scalar fields (note that this
is not the case in the D4 − D8 − D̄8 system), then, for a specific choice of the scalar field
couplings of the open Wilson line, it may be possible to cancel this divergence. However,
generally this divergence is present even for locally supersymmetric OWL operators (as is
the case for the cusp divergence).
On the strong coupling side, for the purposes of computing the divergence in the open
Wilson line correlators we can concentrate just on the region near the boundary, where
the flavor D-brane just sits at xd+1 = · · · = xD = 0 and stretches in the radial direction.
We need to find a minimal worldsheet ending on the contour (6.5) at the boundary and
transverse to the D-brane. It is easy to convince oneself that such a worldsheet is the same
as half of the closed string worldsheet ending on the contour
xd+1 = t sin(θ), xd = |t| cos(θ) , (6.6)
that we obtain by joining to (6.5) its reflection around the subspace that the fundamental
fields live on (a similar trick was recently used in [17]). This contour has a cusp at t = 0
with an angle of 2θ, so it leads to a divergence which is similar to the cusp divergence
occurring in closed Wilson loops (whenever θ 6= π/2). For the case of D = 4 this is a
logarithmic divergence, just as in the previous paragraph, but its precise dependence on
the angle is different from the one found at weak coupling (this is also true for the closed
Wilson loop cusp divergence) [6].
Note that in this computation we assumed that the end of the open Wilson line is at
the same position as the D-brane in the compact directions (otherwise there is no semi-
classical worldsheet contributing to the computation of correlation functions of OW ). If
the D-brane is partially localized in the compact directions (so that the fundamental fields
couple to some of the scalar fields of the gauge theory) then this implies that near the end of
the open Wilson line, the Wilson line couples to different scalar fields than the ones which
the fundamental fields couple to. Thus, for such Wilson lines there is no contribution from
the scalar fields at leading order in perturbation theory, and their one-loop computation
diverges as described above.
In any case, we showed that both at weak coupling and at strong coupling, when
the fundamental representation fields are localized on a subspace, one has to choose the

30
Wilson line operators (2.2) such that the direction of the open Wilson line is transverse
to that subspace at its beginning and end, in order to avoid cusp-like divergences in the
computation.

Acknowledgements
We would like to thank O. Bergman, N. Drukker, S. Hartnoll, Z. Komargodski, O.
Lunin, D. Reichmann, A. Schwimmer, J. Sonnenschein, and S. Yankielowicz for useful
discussions. The work of OA was supported in part by the Israel-U.S. Binational Science
Foundation, by a center of excellence supported by the Israel Science Foundation (grant
number 1468/06), by a grant (DIP H52) of the German Israel Project Cooperation, by
the European network MRTN-CT-2004-512194, and by Minerva. DK is supported in part
by DOE grant DE-FG02-90ER40560, by the National Science Foundation under Grant
0529954, and by the Israel-U.S. Binational Science Foundation. DK thanks the Weizmann
Institute for hospitality during part of this work.

31
 References

[1] J. M. Maldacena, “The large N limit of superconformal field theories and supergrav-
ity,” Adv. Theor. Math. Phys. 2, 231 (1998) [Int. J. Theor. Phys. 38, 1113 (1999)]
[arXiv:hep-th/9711200].
[2] S. S. Gubser, I. R. Klebanov and A. M. Polyakov, “Gauge theory correlators from
non-critical string theory,” Phys. Lett. B 428, 105 (1998) [arXiv:hep-th/9802109].
[3] E. Witten, “Anti-de Sitter space and holography,” Adv. Theor. Math. Phys. 2, 253
(1998) [arXiv:hep-th/9802150].
[4] J. M. Maldacena, “Wilson loops in large N field theories,” Phys. Rev. Lett. 80, 4859
(1998) [arXiv:hep-th/9803002].
[5] S. J. Rey and J. T. Yee, “Macroscopic strings as heavy quarks in large N gauge
theory and anti-de Sitter supergravity,” Eur. Phys. J. C 22, 379 (2001) [arXiv:hep-
th/9803001].
[6] N. Drukker, D. J. Gross and H. Ooguri, “Wilson loops and minimal surfaces,” Phys.
Rev. D 60, 125006 (1999) [arXiv:hep-th/9904191].
[7] T. Sakai and S. Sugimoto, “Low energy hadron physics in holographic QCD,” Prog.
Theor. Phys. 113, 843 (2005) [arXiv:hep-th/0412141].
[8] E. Antonyan, J. A. Harvey, S. Jensen and D. Kutasov, “NJL and QCD from string
theory,” arXiv:hep-th/0604017.
[9] O. Aharony, J. Sonnenschein and S. Yankielowicz, “A holographic model of deconfine-
ment and chiral symmetry restoration,” Annals Phys. 322, 1420 (2007) [arXiv:hep-
th/0604161].
[10] O. Bergman, S. Seki and J. Sonnenschein, “Quark mass and condensate in HQCD,”
JHEP 0712, 037 (2007) [arXiv:0708.2839 [hep-th]].
[11] A. Dhar and P. Nag, “Sakai-Sugimoto model, Tachyon Condensation and Chiral sym-
metry Breaking,” JHEP 0801, 055 (2008) [arXiv:0708.3233 [hep-th]].
[12] S. L. Lukyanov, E. S. Vitchev and A. B. Zamolodchikov, “Integrable model of
boundary interaction: The paperclip,” Nucl. Phys. B 683, 423 (2004) [arXiv:hep-
th/0312168].
[13] D. Kutasov, “D-brane dynamics near NS5-branes,” arXiv:hep-th/0405058.
[14] Y. Nakayama, Y. Sugawara and H. Takayanagi, “Boundary states for the rolling D-
branes in NS5 background,” JHEP 0407, 020 (2004) [arXiv:hep-th/0406173].
[15] L. F. Alday and J. M. Maldacena, “Gluon scattering amplitudes at strong coupling,”
JHEP 0706, 064 (2007) [arXiv:0705.0303 [hep-th]].
[16] Z. Komargodski and S. S. Razamat, “Planar quark scattering at strong coupling and
universality,” JHEP 0801, 044 (2008) [arXiv:0707.4367 [hep-th]].
[17] J. McGreevy and A. Sever, “Quark scattering amplitudes at strong coupling,”
arXiv:0710.0393 [hep-th].

32
[18] N. Drukker and B. Fiol, “All-genus calculation of Wilson loops using D-branes,” JHEP
0502, 010 (2005) [arXiv:hep-th/0501109].
[19] S. Yamaguchi, “Wilson loops of anti-symmetric representation and D5-branes,” JHEP
0605, 037 (2006) [arXiv:hep-th/0603208].
[20] J. Gomis and F. Passerini, “Holographic Wilson loops,” JHEP 0608, 074 (2006)
[arXiv:hep-th/0604007].
[21] G. ’t Hooft, “A planar diagram theory for strong interactions,” Nucl. Phys. B 72, 461
(1974).
[22] O. Aharony, A. Fayyazuddin and J. M. Maldacena, “The large N limit of N = 2,1
field theories from three-branes in F-theory,” JHEP 9807, 013 (1998) [arXiv:hep-
th/9806159].
[23] E. Antonyan, J. A. Harvey and D. Kutasov, “The Gross-Neveu model from string
theory,” Nucl. Phys. B 776, 93 (2007) [arXiv:hep-th/0608149].
[24] E. Antonyan, J. A. Harvey and D. Kutasov, “Chiral symmetry breaking from inter-
secting D-branes,” Nucl. Phys. B 784, 1 (2007) [arXiv:hep-th/0608177].
[25] A. Karch and L. Randall, “Open and closed string interpretation of SUSY CFT’s on
branes with boundaries,” JHEP 0106, 063 (2001) [arXiv:hep-th/0105132].
[26] O. DeWolfe, D. Z. Freedman and H. Ooguri, “Holography and defect conformal field
theories,” Phys. Rev. D 66, 025009 (2002) [arXiv:hep-th/0111135].
[27] A. Karch and E. Katz, “Adding flavor to AdS/CFT,” JHEP 0206, 043 (2002)
[arXiv:hep-th/0205236].
[28] E. Witten, “Anti-de Sitter space, thermal phase transition, and confinement in gauge
theories,” Adv. Theor. Math. Phys. 2, 505 (1998) [arXiv:hep-th/9803131].
[29] N. Itzhaki, J. M. Maldacena, J. Sonnenschein and S. Yankielowicz, “Supergravity and
the large N limit of theories with sixteen supercharges,” Phys. Rev. D 58, 046004
(1998) [arXiv:hep-th/9802042].
[30] D. J. Gross and H. Ooguri, “Aspects of large N gauge theory dynamics as seen by
string theory,” Phys. Rev. D 58, 106002 (1998) [arXiv:hep-th/9805129].
[31] A. Brandhuber, N. Itzhaki, J. Sonnenschein and S. Yankielowicz, “Wilson loops, con-
finement, and phase transitions in large N gauge theories from supergravity,” JHEP
9806, 001 (1998) [arXiv:hep-th/9803263].
[32] O. Bergman and G. Lifschytz, “Holographic U (1)A and string creation,” JHEP 0704,
043 (2007) [arXiv:hep-th/0612289].
[33] C. D. Carone, J. Erlich and M. Sher, “Extra Gauge Invariance from an Extra Dimen-
sion,” arXiv:0802.3702 [hep-ph].
[34] M. Headrick, “Hedgehog black holes and the Polyakov loop at strong coupling,”
arXiv:0712.4155 [hep-th].

33
[35] O. Aharony, K. Peeters, J. Sonnenschein and M. Zamaklar, “Rho meson condensation
at finite isospin chemical potential in a holographic model for QCD,” JHEP 0802,
071 (2008) [arXiv:0709.3948 [hep-th]].
[36] E. Witten, “Current Algebra Theorems For The U (1) Goldstone Boson,” Nucl. Phys.
B 156, 269 (1979).
[37] G. Veneziano, “U (1) Without Instantons,” Nucl. Phys. B 159, 213 (1979).
[38] M. Gell-Mann, R. J. Oakes and B. Renner, “Behavior of current divergences under
SU(3) x SU(3),” Phys. Rev. 175, 2195 (1968).
[39] M. Kruczenski, L. A. P. Zayas, J. Sonnenschein and D. Vaman, “Regge trajectories
for mesons in the holographic dual of large-N(c) QCD,” JHEP 0506, 046 (2005)
[arXiv:hep-th/0410035].
[40] K. Peeters, J. Sonnenschein and M. Zamaklar, “Holographic decays of large-spin
mesons,” JHEP 0602, 009 (2006) [arXiv:hep-th/0511044].
[41] F. Bigazzi and A. L. Cotrone, “New predictions on meson decays from string splitting,”
JHEP 0611, 066 (2006) [arXiv:hep-th/0606059].
[42] A. Giveon and D. Kutasov, “Brane dynamics and gauge theory,” Rev. Mod. Phys. 71,
983 (1999) [arXiv:hep-th/9802067].
[43] C. G. Callan, J. A. Harvey and A. Strominger, “Supersymmetric string solitons,”
arXiv:hep-th/9112030.
[44] O. Aharony, “A brief review of ’little string theories’,” Class. Quant. Grav. 17, 929
(2000) [arXiv:hep-th/9911147].
[45] D. Kutasov, “Introduction To Little String Theory,” Prepared for ICTP Spring School
on Superstrings and Related Matters, Trieste, Italy, 2-10 Apr 2001
[46] S. Elitzur, A. Giveon, D. Kutasov, E. Rabinovici and G. Sarkissian, “D-branes in the
background of NS fivebranes,” JHEP 0008, 046 (2000) [arXiv:hep-th/0005052].
[47] D. A. Sahakyan, “Comments on D-brane dynamics near NS5-branes,” JHEP 0410,
008 (2004) [arXiv:hep-th/0408070].
[48] N. Itzhaki, D. Kutasov and N. Seiberg, “Non-supersymmetric deformations of non-
critical superstrings,” JHEP 0512, 035 (2005) [arXiv:hep-th/0510087].
[49] S. L. Lukyanov and A. B. Zamolodchikov, “Dual form of the paperclip model,” Nucl.
Phys. B 744, 295 (2006) [arXiv:hep-th/0510145].
[50] D. Kutasov, “Accelerating branes and the string / black hole transition,” arXiv:hep-
th/0509170.
[51] D. E. Berenstein, R. Corrado, W. Fischler and J. M. Maldacena, “The operator prod-
uct expansion for Wilson loops and surfaces in the large N limit,” Phys. Rev. D 59,
105023 (1999) [arXiv:hep-th/9809188].
[52] J. K. Erickson, G. W. Semenoff and K. Zarembo, “Wilson loops in N = 4 supersym-
metric Yang-Mills theory,” Nucl. Phys. B 582, 155 (2000) [arXiv:hep-th/0003055].

34
[53] N. Drukker and D. J. Gross, “An exact prediction of N = 4 SUSYM theory for string
theory,” J. Math. Phys. 42, 2896 (2001) [arXiv:hep-th/0010274].
[54] V. Pestun, “Localization of gauge theory on a four-sphere and supersymmetric Wilson
loops,” arXiv:0712.2824 [hep-th].

35