UT-11-14

IPMU11-0080

NSF-KITP-11-075

Investigating Generalized Parton Distribution in Gravity Dual

Ryoichi Nishio and Taizan Watari

Department of Physics, University of Tokyo, Tokyo 113-0033, Japan

[2mm]

Institute for the Physics and Mathematics of the Universe, University of Tokyo,

Chiba 277-8583, Japan

## 1 Introduction

AdS/CFT correspondence and its extension to non-conformal theories have been exploited for study of non-perturbative aspects of strongly coupled gauge theories. Hadron spectra, coupling constants among them and chiral symmetry breaking have been studied intensively in the literature by using gravitational dual descriptions with smooth infra-red non-conformal geometries. The gravitational dual approach can be used, however, to study not just static properties of strongly coupled gauge theories, but also scattering of hadrons. Indeed, string theory or dual resonance model was originally constructed to describe scattering of hadrons. Qualitative aspects of hadron scattering can be obtained in gravitational dual descriptions, if the background geometry (target space) of string theory is chosen properly [1, 2, 3].

In this article, we will study 2-body to 2-body scattering of a hadron and
a virtual photon at high energy in gravitational dual descriptions.
This process is called double deeply virtual Compton scattering (DDVCS). When
the final sate photon is on-shell, it is called deeply virtual Compton
scattering (DVCS), and is accessible in experiments [4].
Because of QCD factorization theorem [5],
the DVCS or DDVCS amplitude
is obtained as a convolution of generalized parton distribution (GPD)
[6] and a hard kernel, the latter of which can be calculated
in perturbative QCD.
GPD itself (at a certain factorization scale), however, is a
non-perturbative object in nature, and cannot be calculated in
perturbative QCD. Even in determining it by using experimental data,
its profile needs to be parametrized^{1}^{1}1See [7] for
review articles, which also have extensive list of literatures.
based on proper understanding on non-perturbative dynamics behind
confinement. We thus use gravitational dual descriptions to extract
theoretical understanding on the GPD profile.

It is not that we just use a well-developed technique to calculate a specific scattering amplitude (or GPD) in this article, however. This article clarifies structure of Pomeron “exchange” amplitudes, how to organize them, as well as their field-theory interpretation. We find that a saddle point value of the scattering amplitude in complex spin -plane representation is a key concept in organizing Pomeron amplitudes and in understanding kinematical variable dependence of the scattering amplitude. Based on this understanding, sharp cross-over behavior is expected in the photon-hadron 2-to-2 scattering amplitude in small limit.

This article is meant to be a brief summary of reference [8]. To keep this letter short enough, we extracted material mainly from §5 of [8], and only minimum from other sections, imagining people in perturbative QCD community as primary readers of this letter. More theoretical aspects of the scattering amplitude in gravity dual, as well as more detailed account of the materials in this letter, are found in [8].

## 2 Amplitude in Gravity Dual

In order to calculate hadron–virtual photon scattering amplitude in gravitational dual, one needs to adopt a certain holographic model. Since the real world QCD turns from weak coupling at high energy into strong coupling at infrared, it is desirable to have a holographic model that is faithful to string theory where AdS curvature becomes larger than string scale toward UV boundary. Such a model becomes even more realistic, if spontaneous chiral symmetry breaking is implemented in it. Our primary goal in this article, however, is not in pursuing precision in numerical calculation (as lattice QCD does) by setting up a perfectly realistic gravitational dual description. An appropriate set-up that suits the best for one’s purpose should depend on the purpose.

We will focus on qualitative aspects of hadron–virtual photon scattering amplitude at small (at high center-of-mass energy). Since small physics is dominated by gluon, not by quarks and anti-quarks, we do not find it a crucial element to implement flavor in the gravitational set up for the purpose of this article. For explicit calculation, we adopt the hard wall model [2], which is type IIB string theory on for some 5-dimensional manifold with cut off at finite radius at infrared. Such a crude treatment of infrared geometry is sufficient for our qualitative study [2], and the choice of becomes irrelevant (at least directly) for sufficiently small [8]. Since it is almost straightforward to see how the curvature and running of dilaton expectation value affects various observables in explicit calculations based on the hard wall model, one can also learn what happens in gravitational dual models that are asymptotically conformal or asymptotically free without carrying out calculations separately on these models.

As an analogy of the electromagnetic global symmetry of QCD,
we take a global symmetry of in the gravitational dual. Since we are
interested in the Compton tensor^{2}^{2}2In our convention, .
Let us remark that
the Compton tensor in this letter is defined differently from one in [8];
the Lorentz indices are interchanged.
of QCD,

(1) |

we use the bulk-to-boundary propagator of an vector field
associated with a Killing vector of in calculating the matrix
element involving the global symmetry current. As for the target
hadron in the gravity dual, we use a Kaluza–Klein state of a dilaton,
whose wavefunction is given by a Bessel function in the hard wall model.
Thus, the leading order contribution in expansion is given by
a closed string sphere amplitude with four NS–NS string vertex operator
insertions [2].^{3}^{3}3
The target hadron which is dual to a Kaluza–Klein state of a dilaton is a glueball.
The case of a meson target can also be studied in the same way
if we use open strings.
For the case of a baryon target, we should use -brane in the gravity dual.
We will see
that the saddle point value and singularities in the complex -plane
representation are important in describing the amplitude.
Because they do not depend on the target hadron wavefunctions,
they are expected to be unchanged even if the species of target hadron is replaced.

As we consider cases where the initial state “photon” or both the initial and final state “photons” are highly virtual, that is, or , the “photon”–hadron scattering amplitude can be decomposed into various contributions through operator product expansion of and in QCD language. Such a decomposition still holds true in strongly coupled gauge theories (and hence in gravitational dual), except that the anomalous dimensions of operators in the expansion may be quite different from what one expects in the weak coupling regime. Reference [2] noted that the operators that are twist-2 in the weakly coupled regime still appear in the operator product expansion even in the strongly coupled regime, and their contributions to the Compton tensor dominate at sufficiently small ; this is because the “twist-2” contribution corresponds to exchange of leading Regge trajectory containing graviton in gravity dual language [9, 3]. We will thus focus on small hadron–virtual photon scattering in gravity dual to study non-perturbative behavior of the “twist-2” contribution.

Before writing down the Pomeron contribution to the scattering amplitude explicitly, let us note that the Compton tensor is described by five structure functions as in [10],

(2) |

for a scalar target hadron, because of gauge invariance. In parity-preserving theory, . In the limit of purely forward scattering, the two structure functions of deep inelastic scattering are restored from and . Here, we introduced a convenient notation

(3) |

In this article, we will use the following notations,

(4) |

and and .

In the generalized Bjorken limit, , and for much smaller than unity, the Pomeron contribution to the five structure functions are given by and [8] as in

(5) |

and are given for vanishing skewedness in the form of

(6) |

For vanishing skewedness,
the Pomeron kernel
is [3]^{4}^{4}4
More careful discussion on the choice of integration contour is given in [11, 8].
A pedagogical explanation of the origin of factor is also given in [8].

(7) |

the integration contour in the complex -plane encircles the pole , and once the residue of this pole is picked up, a relation

(8) |

sets the (analytically continued) relation between spin and anomalous dimension of “twist-2” operators in the large ’t Hooft coupling regime [3]. is the warp factor in the part of the metric in the hard wall model,

(9) |

and in (6) is that of this metric of 5-dimensional spacetime. is the AdS radius, and the infrared cut off of the hard wall model at sets the confinement scale . , and is the slope parameter of the Type IIB string theory. in (7) is the Pomeron wavefunction in the spin channel, which is given by

(10) |

in the hard wall model.^{5}^{5}5Dirichlet boundary condition was
imposed at the infrared boundary , just to make expressions
simpler.

The impact factor of the target hadron side is given by the normalizable mode wavefunction of the target hadron, as in . On the virtual “photon” side, the bulk-to-boundary propagator (non-normalizable wavefunction) of the graviton associated with the Killing vector of is used; in the hard wall model, they are

(11) | |||||

(12) |

for and , respectively. is a constant of a theory of mass dimension and is proportional to . , and are dimensionless constants of order unity. See [8] for their definitions.

## 3 Structure and Behavior of the Amplitude

### 3.1 Complex -plane amplitude, Pomeron vertex and form factor

Before discussing kinematical parameter () dependence of the DDVCS amplitude in gravity dual, let us clarify a couple of conceptual issues associated with Pomerons. Using the explicit form of the Pomeron kernel (7) and Pomeron wavefunctions (10), amplitudes () in (6) can be rewritten (see [8] for details) as

(13) |

where

(14) | |||||

(15) | |||||

a parameter of mass dimension is introduced in (14, 15) in a way the observables are unaffected. One can change the integration variable of (13) from to ; now the amplitudes are given by integration over the complex -plane, and the contour becomes the one in Figure 1 (a).

(a) | (b) | (c) | (d) |

The factor is now regarded as a function of , and also depends on and , but not on or . Its asymptotic form for is given by

(16) |

with a dimensionless constant of order unity that depends only on . Here, , and is the inverse function of (8). dependence and dependence of the amplitudes come from the other factor . It can be rewritten as

(17) |

where is a dimensionless function of and . For the final expression, we used which holds at small . Combining both, one finds that

(18) |

This is in the form of inverse Mellin transformation, and the
integration variable is identified with the complex angular
momentum (complex spin).^{6}^{6}6Since we restrict ourselves to the scattering at
, total derivative operators in field-theory language do not
contribute to the OPE of the scattering amplitude. Thus, there is no
subtleties in what this is here.

Now, physical meaning of the separation between and (or ) is clear. By changing the integration contour in the -plane, (13, 18) can be rewritten as

(19) |

This is regarded as an OPE form of . The first factor in , which comes from , is regarded as the Wilson coefficient of OPE for a spin operator; the parameter is now identified with the renormalization scale, because of its appropriate scaling behavior determined by the anomalous dimension of the “twist-2” spin operator. The second factor in is identified with the spin form factor, which is the coefficient of the term of the hadron matrix element of the spin operator renormalized at the scale . The gravity dual expression (15) justifies such an interpretation [12].

Knowing physical meaning of these factors in the scattering
amplitude (13) in a gravity dual model, one can define
a GPD even in the model, which corresponds to a strongly coupled
gauge theory. GPD as a function of and (we only consider the
case in this letter) is defined as an inverse Mellin
transform of form factors of twist-2
spin operators
(the second factor in [] of (19)).
The scattering amplitude is given by
convolution of this GPD, inverse Mellin transform of the Wilson
coefficient and that of the signature factor
, just like in perturbative QCD
factorization formula. The inverse Mellin transform of the signature
factor gives rise to a light-cone singularity of a propagating parton
(like the one in [13]), even in the gravity dual description.
The GPD determined in this way is essentially^{7}^{7}7Since GPD
is defined as the inverse Mellin transform of form factors of
“twist-2” spin operators, it would become different when the
normalization of the operators were changed in a -dependent
manner. We do not pay such a careful attention in this article.
We claim similarity between and GPD after replacement of
by only at this level of precision. the same as
, with of replaced by the
renormalization scale ; thus, various statements on
in the rest of this section are also applied to
the GPD after is replaced by .

Now that the field theory OPE interpretation of the gravity dual amplitude (13) is clarified, let us go back to the amplitude (13) and explicit expressions (14, 15) once again. We will now clarify how this string theory amplitude on a warped background is related to the traditional Regge phenomenology ansatz. It should be noted that the DDVCS amplitudes in gravity dual (18) do not have a Pomeron pole like in their -plane representation apparently. There was once a pole at the stage of (7), but it is gone in (18), after picking the residue to evaluate an integral in (7). Nevertheless, one can see that the expression (18) may have, in fact, many poles in the -plane, rather than a single pole or none.

To see this, we can use Kneser–Sommerfeld expansion of Bessel functions in the hard wall model to rewrite as [8]

(20) | |||||

(21) |

The Pomeron trajectory (that contains graviton) of the Type IIB string theory on 10-dimensions (or on after dimensional reduction on ) gives rise to a Kaluza–Klein tower of infinitely many Pomeron trajectories in hadron scattering on 3+1 dimensions. These trajectories are labeled by the Kaluza–Klein excitation level ; the masses of spin hadrons are , and their wavefunctions on are . The factor in (20) becomes a dependent pole in the -plane, the Pomeron pole, for any one of ’s. In the hard wall model, the Pomeron trajectory ( relation set by ) and the Pomeron wavefunction for the -th trajectory are obtained holomorphically in (not just for ) as in

(22) |

where, is the -th zero of Bessel function . is read out from the denominator in (15); the wavefunction satisfies an equation of motion of a spin field on , just like (10) does. Although explicit expressions above rely heavily on the hard wall model, conceptual understanding itself is quite general, and is applicable at least to any asymptotically conformal gravity dual models.

Therefore, gravity dual descriptions of strongly coupled gauge theories come up with a following picture of Pomeron exchange amplitude. Individual Pomerons in the Kaluza–Klein tower couple to the target hadron with a coupling in (21), which do not show any power-law fall-off behavior in large negative . Only after all the Pomeron couplings and Pomeron propagators are combined as in (20), do we obtain what we might call a “Pomeron form factor” , which has a power-law behavior in (see (26)). Such a relation between a form factor of a conserved current and a combination of a Kaluza–Klein tower of hadrons, three point couplings and decay constants has been known for fixed spins (such as and ) [14]. The relation (20) is regarded as an analytic continuation in of the one for graviton (spin ).

### 3.2 Saddle point in the -plane

It was a conventional wisdom of traditional Regge phenomenology that behavior of hadron scattering amplitudes at high energy are governed by the position of singularities in the complex -plane. The same is true in gravity dual description of strongly coupled gauge theories. Singularities in scattering amplitudes in the complex -plane representation depend on choice of gravity dual models. In case of the hadron–virtual “photon” scattering, however, the scattering amplitude can be approximated at saddle point in the -plane (within a certain kinematical region which we call “saddle point phase” in §3.3). In this case, the expression of the amplitude becomes not directly dependent on the singularities, and hence, detail of gravitational dual is irrelevent. In this subsection, we employ the hard wall model, and study the behavior of this scattering amplitude.

In the hard wall model, there are no isolated poles in the complex -plane for negative —physical kinematics—except the branch
cut that extends to negative along the real axis,
Figure 1(a); never vanishes for .
Thus, the integral of (18) along the contour
in Figure 1 (a) is evaluated by the saddle point method
for small [11].
For , the saddle point
value of is given by^{8}^{8}8
The integrand of the Pomeron kernel (7) is
reliable at , but not at
[9].
Therefore, we note that kinematical variables (, and )
are required to be consistent with .

(23) |

and

(24) |

the form factor has a
dimensionless non-zero limit of order unity when ;
it begins to fall off in power-law^{9}^{9}9Thus, for ,
the saddle point becomes

(26) |

for larger momentum transfer . Here, is the scaling dimension of the scalar field on containing the target hadron , and is a -independent (but -dependent) constant of order unity. Note, in particular, that the -dependence of the scattering amplitude is given by the form factor that is once analytically continued to complex -plane and then evaluated at the saddle point. The Regge factor of string theory amplitude justifies focusing on a small range of (or ) around the saddle point value at high-energy scattering; the power-law behavior in follows from the power-law wavefunction of the target hadron and exponential cut-off of the Pomeron wavefunction, in (15) in particular, in the limited range of .

The saddle point method provides a good approximation to the scattering amplitude for . It should be noted, however, that it allows us to keep all-order contributions in , which is not necessarily small and can be as large as . Thus, amplitudes and observables are expressed as functions of (or ). This makes easy to understand their dependence of kinematical variables ().

Equation (24) clearly shows the importance of the value of the saddle point of the -plane amplitude. To see this more explicitly, let us define

(27) |

It is straightforward to see that they are given by

(28) |

These effective exponents and depend on kinematical variables and only through the saddle point value . The ratio is also related directly to the saddle point value .

We can see from (23, 25) that becomes large for large and small for small . Thus, at a given renormalization scale (replace in ), GPD in gravity dual still increases in the DGLAP evolution (that is, ) for small enough such that the saddle point value is still less than 2. Even at such a small value of , however, GPD eventually begins to decrease (that is, becomes positive) for large enough . Such a behavior of GPD—qualitatively the same as in the real world QCD—in the DGLAP evolution was anticipated in [2]; this is indeed realized for finite in gravity dual, when both and dependence are included in the saddle point approximation. The other parameter characterizing the evolution is known to increase gradually for larger in the real world QCD [15]. As already seen in [16], it does follow from gravity dual as well; we understand that this phenomenon is also essentially due to the increase of the saddle point value for larger . The same behavior is also obtained in perturbative QCD (See [17]).

The dependence of the scattering amplitude is characterized by the slope parameter of the forward peak (also known as -slope parameter), which we define for non-skewed scattering as

(29) |

The -dependence (and hence the slope parameter) comes entirely from the form factor for the physical kinematical region in the hard wall model. The -slope parameter at in such a case can be regarded as the charge radius square of the hadron under “spin- probe”. Explicit expressions for the form factor in the hard wall model allow us to calculate the -slope parameter; see Figure 2.

The larger the spin (and hence ), the smaller the slope. Therefore, through (23), the slope parameter decreases for larger , a prediction of a gravity dual model which cannot be made within perturbative QCD.

We can also see from (23, 25) that the saddle point value depends weakly on or than on for small , and the dependence is in the opposite direction. Thus, the dependence (or dependence) of the slope parameter must be weaker than its dependence. This property of , shared by , and , is an immediate consequence of the fact that the scattering amplitude is well approximated by the saddle point method on the -plane integral. This is a fairly robust feature of the saddle point approximation, and does not rely on specific details of the hard wall model.

The saddle point turns out to be an important concept also in the scattering amplitude in the impact parameter space, which is obtained by taking a Fourier transform in the transverse direction of the momentum transfer . The -dependent parton density profile in the transverse direction obtained in this way [18] in gravity dual shows Gaussian profile at large impact parameter , but is larger than the simple Gaussian form for smaller ([8]; see also [16]; deviation from the simple Gaussian profile is an immediate consequence of the fact that the 4D leading trajectory is not perfectly linear). This core of larger parton density has approximately a linear exponential profile, . The effective mass scale gradually changes as a function of , and the linear exponential form smoothly turns into the Gaussian form for larger , when becomes of order unity. See [8] for more.

### 3.3 Pole–Saddle Point Crossover

Although we saw that the saddle point method well approximated the DDVCS amplitude for physical kinematical region in the hard-wall model, it does not in general. Even in small , whether or not the scattering amplitude is well approximated by the saddle point method, depends on singularities of the amplitude in the -plane representation, and hence on the gravity dual model one considers, and also the values of the kinematical variables and . Although all the asymptotically conformal gravity dual models have a branch cut that is stretched to large negative , there may also be some isolated poles in the -plane as in Figure 1 (b). The hard wall model does not have such a pole for physical kinematical region (there are for sufficiently positive ), but there may be some for other UV conformal models that have different (and faithful to string theory construction) infrared geometry. Even more interesting are gravity dual models that are asymptotically free, where the cut is replaced by isolated singularities (Figure 1 (c, d)) [3].

When the saddle point (open circle in the figure) has a larger real part than any one of the singularities in the complex -plane, then the integration contour in the -plane should simply be chosen so that it passes through the saddle point, as in Figure 1 (a, c). When some of the singularities have larger real parts than the saddle point value , however, it is more convenient to take the contour as in Figure 1 (b, d), so that the scattering amplitude is given by contributions from finite number of isolated Pomeron poles () and by a continuous integration over a contour passing through the saddle point. We refer to the two situations as saddle point phase and leading pole phase (or leading singularity phase), respectively. Such observables as , , and exhibit totally different dependence on the kinematical variables and in the two phases. In a given theory (i.e., in a given gravity dual model), one always enters into the saddle point phase for sufficiently large or sufficiently negative . In asymptotically free theories, it is likely that the leading singularity phase also exists for sufficiently small and not so large negative , even in the physical kinematical region .

The transition between the two phases is not singular but is a (smooth) crossover for finite . This is because the saddle point approximation is never exact, and the “saddle point” should be thought of as a sort of diffuse object for finite . Subleading singularities may also give rise to significant corrections to the amplitude simply given by the leading pole for finite , too. The transition becomes a singular phase transition only in the extreme small limit.

## 4 Lessons to Learn

It is true that gravity dual calculation employs a background that corresponds to large ’t Hooft coupling even at energy scale much larger than the hadronic scale . Still, there are surprisingly many qualitative features in the gravity dual hadron–virtual “photon” scattering amplitude that are in common with the scattering amplitude in the real world QCD. Scattering amplitudes in gravity dual have and scaling governed by and in the saddle point phase, and this is the same qualitatively as the prediction of the saddle point method in perturbative QCD, as we have already seen in §3.2. The only difference between gravity dual and real world QCD is in the choice of anomalous dimension, . Qualitative features are shared by both, and are controlled by the saddle point value .

Qualitative features in -dependence also show agreements. The gravity dual amplitude continues to the power-law fall-off behavior at large momentum transfer . This property, which is expected to hold in the real world QCD theoretically [19] and confirmed experimentally, was difficult to be consistent with the traditional Regge phenomenology, but this problem is now overcome in gravity dual on warped spacetime (cf. [1, 2]). Moreover, the -slope parameter of (29) and its result in Figure 2 for in gravity dual at saddle point phase nicely agrees with that in DVCS differential cross section [20], in that the slope parameter decreases for larger , and is less sensitive to or . Such observation suggests the (analytically continued) spin form factors in both a gravity dual model and the real QCD are similar to each other.

With so many basic qualitative features that gravity dual shares with the real world QCD, it is thus tempting to try to extract some lessons from the hadron–virtual “photon” amplitude in gravity dual. The origin of such similarity at the qualitative level becomes clear in the complex -plane representation, where GPD is given by inverse Mellin transformation:

(30) |

Indeed, it is always possible to describe scattering amplitude by the -plane integral in any theories, independent of whether the scattering is based on the real QCD or on the strongly coupled gauge theory studied in gravity dual; this is because Mellin transformation is only a mathematical transformation. This -plane integral also comes form OPE, notion of which is well-defined even in strongly coupled theories [2]. GPD in the -plane representation (30) is given by dropping the Wilson coefficient of OPE from the scattering amplitude (13), so the spin form factor (reduced matrix element of twist-2 spin operator), which is the content of [] in (30), determines GPD. The spin form factor is decomposed into two parts: RG evolution part , and form factor at renormalization scale , ; both show common properties in the real QCD and in gravity dual. The anomalous dimensions of the twist-2 spin operator in both theories are qualitatively similar [3], and has the power-law fall-off behavior at in common.

The behavior of GPD is determined by the saddle point, or alternatively,
by the leading singularity
depending on which phase a set of parameters sits in.
This classification is applicable in any theories, not just in gravity dual.
Then it is important to know which phase a given set sits in.
As we have pointed out, the behaviors of
and
observed in HERA for DIS [15]
are successfully explained by the predictions of the saddle point phase
and are inconsistent with the prediction of the leading pole phase
(or the leading singularity phase)
[8].
Therefore,
it is very likely that
the (most of) kinematical region of DVCS
that has been explored in HERA measurements
is in the saddle point phase,^{10}^{10}10
In the standard parametrization of DVCS cross section
,
the parameters are given by
and
in the saddle point phase.
Thus, the saddle point phase implies rise of for larger .
HERA measurement [21] gives
for GeV,
for GeV,
for GeV,
for GeV, and
for GeV
in ZEUS,
and
for GeV,
for GeV, and
for GeV,
in H1. and GPD is approximately
given by

(31) |

Most of the observed properties of the -slope parameter of DVCS in HERA [20]
can be understood only from the fact that the kinematical region
is in the saddle point phase (see [8]).
A GPD model with a specific choice of
in [22] belongs to this category.^{11}^{11}11Reference
[22] introduces an ansatz ,
inspired by a leading Pomeron pole and a power-law fall-off for .
Our result (20) is conceptually different
from this model;
each Pomeron pole term with a Kaluza–Klein excitation level
does not show the behavior of power-law fall-off,
but the power-law (26) appears only after summing all the
Kaluza-Klein tower of Pomeron pole terms.

One can also see that the saddle point expression (31) automatically satisfies a requirement that GPD should be consistent with DGLAP evolution, because -evolution is correctly taken into account in the -plane expression (30). This is a nontrivial requirement on GPD modeling in general. One can consider, for example, a GPD profile given by PDF (GPD at ) multiplied by some form factor at a given renormalization scale [23]:

(32) |

where , and and are parameters. The profile of GPD like this are not stable under DGLAP evolution. On the other hand, the GPD under the saddle point approximation (31) is given by the PDF multiplied by a spin form factor evaluated at the saddle point value . The saddle point value depends on and the factorization/renormalization scale . This result obviously takes into account renormalization effects, and hence is stable/reliable at any renormalization scale.

The remaining task is to determine the spin form factor at renormalized point , as a holomorphic function of . This is along the line of the collinear factorization approach (dual parametrization) to the modeling of GPD[24]. Derivation of from the first principle is an impossible task in perturbative QCD, because of the non-perturbative origin of the form factor, and this is also hard in lattice simulation, because there is practically no way of finding analytic continuation of integer spin matrix elements into complex . An alternative is to use predictions from the gauge/string duality, and a crude way is to use the prediction of the hard wall model derived in §3.1 as it is. Indeed, as we saw, the hard wall model can explain decreasing slope parameter of DVCS for large , observed in HERA [20]. It is also possible to use more realistic gravity dual models for similar calculation, where at least we might want to require the model to have asymptotic free running for certain energy range (as in [25]) still with large ’t Hooft coupling.

If one wants to consider a gravity dual model that is truly dual to
the real world QCD (if there is any),
then it should run into a problem in its UV region
of the geometry because of large curvature.
This problem of gravitational description, however, may be alleviated
by borrowing the understanding of perturbative QCD.
Such strategy may not be totally nonsense.
We saw that
the singularities of the form factor in the -plane are important in
determining GPD,
and gravity dual with asymptotic free running suggests that
the singularities are infinitely many poles^{12}^{12}12
These poles correspond to trajectories of Kaluza-Klein modes
in radial direction of a single graviton trajectory in 10 dimensions
(or on ).
On top of this tower structure, there is yet another
tower structure of trajectories associated with the daughter trajectories
of stringy excitations on 10 dimensions.
[3].
The BFKL theory in perturbative QCD with a running coupling effect
also suggests infinitely many poles in the -plane [26].
Now, let us examine how sensitive the position of the poles
predicted from gravity
dual are to
the unreliable large curvature geometry in the UV region.
In gravitational descriptions,
each Pomeron pole has its wavefunction on the holographic coordinate,
and the Pomeron wavefunction becomes localized more and more
into the IR region of the holographic radius when of the pole
increases.
Therefore, the poles in large are determined
mainly by IR physics, and
position of poles predicted by gravity dual should be reliable,
while the poles in small are
quite sensitive to the unreliable geometry in the UV region.
As for such smaller poles, however,
the position of the poles predicted by the BFKL theory
(with asymptotically free running)
will be reliable.
Thus, by using both predictions from the gravity dual and the BFKL theory,
the poles in the -plane may be properly determined.

In order to determine GPD completely, not only the position of the poles but also complete profile of the spin form factor are required. The spin form factor is given by integrating Pomeron wavefunction and impact factor in gravity dual, and in fact, also in the BFKL theory; the integration is carried out over the holographic radius in gravity dual, whereas it is done over gluon transverse momentum in the BFKL theory. The similar structure in the factorization formula and the gravity dual scattering amplitude (6) has been pointed out, and identification of gluon transverse momentum in the BFKL theory with holographic radius in the gravity dual is suggested [27, 3]. Thus, one can retain the integration over the holographic radius in gravity dual in the IR (large ) region. The integration in the UV region may be replaced by that over coordinate in the BFKL theory; this large region is where perturbative QCD is reliable.

## Acknowledgments

Part of this work was carried out during long term programs “Branes, Strings and Black Holes” at YITP, 2009 (TW), “Strings at the LHC and in the Early Universe” at KITP, 2010 (TW), “High Energy Strong Interactions 2010” at YITP, 2010 (RN, TW) and also during a stay at Caltech theory group of TW. This work is supported by JSPS Research Fellowships for Young Scientists (RN), by WPI Initiative, MEXT, Japan (RN, TW) and National Science Foundation under Grant No. PHY05-51164 (TW).

## References

- [1] J. Polchinski and M. J. Strassler, Phys. Rev. Lett. 88 (2002) 031601, [hep-th/0109174].
- [2] J. Polchinski and M. J. Strassler, JHEP 0305 (2003) 012, [hep-th/0209211].
- [3] R. C. Brower, J. Polchinski, M. J. Strassler, and C.-I. Tan, JHEP 0712 (2007) 005, [hep-th/0603115].
- [4] HERMES Collaboration, A. Airapetian et al., Phys.Rev.Lett. 87 (2001) 182001, [hep-ex/0106068]; HERMES Collaboration, A. Airapetian et al., Phys.Rev. D75 (2007) 011103, [hep-ex/0605108]; CLAS Collaboration, S. Stepanyan et al., Phys.Rev.Lett. 87 (2001) 182002, [hep-ex/0107043]; CLAS Collaboration, S. Chen et al., Phys.Rev.Lett. 97 (2006) 072002, [hep-ex/0605012]; Jefferson Lab Hall A Collaboration and Hall A DVCS Collaboration, C. Munoz Camacho et al., Phys.Rev.Lett. 97 (2006) 262002, [nucl-ex/0607029]; Jefferson Lab Hall A Collaboration, M. Mazouz et al., Phys.Rev.Lett. 99 (2007) 242501, [0709.0450 [nucl-ex]]; ZEUS Collaboration, S. Chekanov et al., Phys. Lett. B573 (2003) 46–62, [hep-ex/0305028]; ZEUS Collaboration, S. Chekanov et al., JHEP 0905, 108 (2009) [arXiv:0812.2517 [hep-ex]]; H1 Collaboration, C. Adloff et al., Phys. Lett. B517 (2001) 47–58, [hep-ex/0107005]; and the three papers by H1 in [20]. See also review articles [7].
- [5] S. J. Brodsky, L. Frankfurt, J. F. Gunion, A. H. Mueller, M. Strikman, Phys. Rev. D50, 3134-3144 (1994) [hep-ph/9402283]; J. C. Collins, L. Frankfurt, M. Strikman, Phys. Rev. D56, 2982-3006 (1997) [hep-ph/9611433].
- [6] D. Mueller, D. Robaschik, B. Geyer, F. M. Dittes, J. Horejsi, Fortsch. Phys. 42, 101 (1994) [hep-ph/9812448]; X. -D. Ji, Phys. Rev. Lett. 78, 610-613 (1997) [hep-ph/9603249]; X. -D. Ji, Phys. Rev. D55, 7114-7125 (1997) [hep-ph/9609381]; A. V. Radyushkin, Phys. Rev. D56, 5524-5557 (1997) [hep-ph/9704207].
- [7] X. -D. Ji, J. Phys. G G24, 1181-1205 (1998) [hep-ph/9807358]; A. V. Radyushkin, [hep-ph/0101225]; K. Goeke, M. V. Polyakov, and M. Vanderhaeghen, Prog. Part. Nucl. Phys. 47 (2001) 401–515, [hep-ph/0106012]; M. Diehl, Phys.Rept. 388 (2003) 41–277, [hep-ph/0307382]; A. V. Belitsky and A. V. Radyushkin, Phys. Rept. 418 (2005) 1–387, [hep-ph/0504030]; S. Boffi and B. Pasquini, Riv. Nuovo Cim. 30 (2007) 387, [0711.2625 [hep-ph]].
- [8] R. Nishio and T. Watari, “High-energy Photon–Hadron Scattering in Holographic QCD,” UT-11-02/IPMU11-0003/NSF-KITP-11-028, [1105.2999 [hep-ph]].
- [9] S. S. Gubser, I. R. Klebanov, A. M. Polyakov, Nucl. Phys. B636, 99-114 (2002) [hep-th/0204051].
- [10] C. Marquet, C. Roiesnel, and S. Wallon, JHEP 04 (2010) 051, [1002.0566 [hep-ph]].
- [11] Y. Hatta, E. Iancu, A. H. Mueller, JHEP 0801, 026 (2008) [0710.2148 [hep-th]].
- [12] E. Witten, Adv. Theor. Math. Phys. 2, 253-291 (1998) [hep-th/9802150], S. S. Gubser, I. R. Klebanov, A. M. Polyakov, Phys. Lett. B428, 105-114 (1998) [hep-th/9802109], W. Mueck, K. S. Viswanathan, Phys. Rev. D58, 041901 (1998) [hep-th/9804035].
- [13] The 3rd reference of [6].
- [14] S. Hong, S. Yoon, and M. J. Strassler, JHEP 04 (2004) 046, [hep-th/0312071]; S. Hong, S. Yoon, and M. J. Strassler, JHEP04 (2006) 003, [hep-th/0409118]; S. Hong, S. Yoon, and M. J. Strassler, [hep-ph/0501197].
- [15] E665 Collaboration, M. R. Adams et al., Phys. Rev. D54(1996) 3006–3056; ZEUS Collaboration, J. Breitweg et al., Eur. Phys. J. C7(1999) 609–630, [hep-ex/9809005].
- [16] R. C. Brower, M. Djuric, I. Sarcevic, and C.-I. Tan, [1007.2259 [hep-ph]].
- [17] R. D. Ball, S. Forte, Phys. Lett. B335,(1994) 77-86 [hep-ph/9405320]; R. D. Ball, S. Forte, Phys. Lett. B336, (1994) 77–79. [hep-ph/9406385].
- [18] M. Burkardt, Phys. Rev. D62, 071503 (2000) [hep-ph/0005108], Phys. Rev. D 66, 119903(E) (2002); J. P. Ralston and B. Pire, Phys. Rev. D 66, 111501 (2002) [hep-ph/0110075]; M. Burkardt, Int. J. Mod. Phys. A18 (2003) 173–208, [hep-ph/0207047]; M. Diehl, Eur. Phys. J. C25, 223-232 (2002), [hep-ph/0205208], Eur. Phys. J. C31, 277(E) (2003).
- [19] G. P. Lepage, S. J. Brodsky, Phys. Rev. D22, 2157 (1980).
- [20] H1 Collaboration, A. Aktas et al., Eur. Phys. J. C44 (2005) 1–11, [hep-ex/0506061]; H1 Collaboration, F. D. Aaron et al., Phys. Lett. B659 (2008) 796–806, [0709.4114 [hep-ex]]; H1 Collaboration, F. D. Aaron et al., Phys. Lett. B681 (2009) 391–399, [0907.5289 [hep-ex]]. See also A. Freund, M. McDermott, M. Strikman, Phys. Rev. D67, 036001 (2003) [hep-ph/0208160], and the references therein.
- [21] The 8th reference of [4] by ZEUS, and the 3rd reference of [20] by H1.
- [22] D. Mueller, [hep-ph/0605013].
- [23] M. Diehl, Nucl. Phys. Proc. Suppl. 161, 49 (2006) [hep-ph/0510221].
- [24] A. V. Belitsky, B. Geyer, D. Mueller, and A. Schafer, Phys. Lett. B421 (1998) 312–318, [hep-ph/9710427]; M. V. Polyakov, Nucl. Phys. B555 (1999) 231, [hep-ph/9809483]; M. V. Polyakov and A. G. Shuvaev, [hep-ph/0207153]; D. Mueller, A. Schafer, Nucl. Phys. B739, 1-59 (2006) [hep-ph/0509204], K. Kumericki, D. Mueller, K. Passek-Kumericki, Nucl. Phys. B794, 244-323 (2008) [hep-ph/0703179].
- [25] O. Aharony, A. Fayyazuddin, and J. M. Maldacena, JHEP 9807 (1998) 013, [hep-th/9806159].
- [26] J. R. Forshaw, D. A. Ross, Quantum chromodynamics and the pomeron, Cambridge Lect. Notes Phys. 9, 1-248 (1997).
- [27] R. C. Brower, C.-I. Tan, Nucl. Phys. B662, 393-405 (2003) [hep-th/0207144]; S. J. Brodsky, G. F. de Teramond, Phys. Lett. B582, 211-221 (2004) [hep-th/0310227].