HIP200913/TH
Revised 10 September 2009
{centering}
Spatial string tension of finite temperature QCD matter in gauge/gravity duality
J. Alanen^{*}^{*}*janne.alanen@helsinki.fi, K. Kajantie^{†}^{†}†keijo.kajantie@helsinki.fi, V. SuurUski^{‡}^{‡}‡ville.suuruski@helsinki.fi
Department of Physics, P.O.Box 64, FI00014 University of Helsinki,
Finland
Helsinki Institute of Physics, P.O.Box 64, FI00014 University of
Helsinki, Finland
The computation of the spatial string tension of finite temperature QCD is discussed in QCD and in a gravity+scalar model of gauge/gravity duality.
September 2009
1 Introduction
The spatial string tension in finite temperature QCD is determined with lattice Monte Carlo techniques [1, 2, 3, 4, 5] by studying in asymmetric lattices expectation values of large Wilson loops with both sides in spatial directions. It is a nonperturbative quantity and thus not analytically calculable. However, if it is determined numerically in 3 dimensional (3d) SU(3) gauge theory [6], one can use the well established equations of perturbative high temperature dimensional reduction to compute it in full 4d finite QCD and compare it with the lattice result [7, 8]. The agreement is surprisingly good.
Since is so well under control in QCD, it is a good laboratory for testing various models for AdS/QCD [9, 10, 11]. The heart of QCD computations is conformal invariance breaking and running of the coupling constant while the heart of well established AdS/CFT duality is conformal invariance of CFT. The essential part of AdS/QCD models is thus how conformal invariance breaking is modelled. We shall in this article apply the AdS/QCD model of Kiritsis and coworkers [12, 13, 14, 15], approximating it in a form which permits an analytic solution.
2 QCD discussion
A firstprinciple method for determining the spatial string tension in hot QCD matter is to measure rectangular Wilson loops of size in the plane. A potential is then defined by:
(2.1) 
and then finding from the large behavior
Measurements are done with full 4d QCD action on a finite lattice and they can be done for pure SU(3) YangMills theory or for QCD with dynamical quarks.
There is also a natural way to interpret the result and to derive it by splitting the problem in an analytic and a numerical nonperturbative part. The spatial string tension lives in the 3d spatial space and one can also determine it for 3d SU(3) YangMills theory. Since the only dimensionful parameter in this theory is its coupling constant of dimension mass, one knows that the string tension must be a number times . In fact [6, 24],
(2.2) 
On a schematic level, one knows that , where the 4d coupling constant has a simple logarithmic expansion. Thus it is natural to choose to plot
(2.3) 
where the standard 2loop expansion is inserted (with ). The scale is unknown but can be fitted to the data. The data for finite SU(3) YangMills theory [3] is as shown below in Figs. 4 and 5.
To obtain a controlled theoretical result one at least has to compute the in (2.3) and not fit it to data. To do this one can start from the nonperturbative magnetic sector 3d number (2.2) and work out backwards, first to the electric sector of hot QCD by restoring the electric mass by the 2loop relation [16]
(2.4) 
and finally the hard scale by the 2loop relations [7] ()
(2.5)  
(2.6)  
(2.7) 
Here is the running coupling in which we, for concreteness, optimized the scale so that the term in (2.5) vanishes, i.e., . A thorough analysis of the optimisation scale dependence is given in [7]. Noting that [7] the logarithmic factors in could be written as , i.e., in (2.3) has been evaluated to be .
The reliable QCD prediction then is obtained by inserting to (2.2) from (2.4) with and as given by (2.5) and (2.6)^{1}^{1}1Note that in a 3d gauge theory the couplings are scale independent while has both a linear 1loop and a logarithmic 2loop divergence. It so happens that the coefficient of the logarithmic divergence cancels for [17] and above is scale independent to the order shown, .. The prediction is shown as the continuous curve plotted over the range in Figs. 4 and 5 below. The lattice data, shown in the same figures, extends over the range .
One also observes that the correction terms in (2.4) and the term in (2.5) almost cancel each other so that to a good accuracy the result is just given by the 2loop expression in (2.3).
The QCD computation is an entirely controlled perturbative computation, the nonperturbative part is isolated in the number (2.2). It works surprisingly well, perhaps even too well. The computation is, namely, based on integrating out the large scales and and one does not expect it to work down to . For the pressure a similar computation shows clear deviations from the lattice results for below [18].
Clearly, to assess the reliability of the result it would be useful to further compute the 3loop contribution – a formidable task.
3 AdS/QCD model
Consider then the computation of the spatial string tension in AdS/QCD models. In bottomup models conformal invariance can be broken by putting by hand some structure in the extra dimension : a hard wall or some soft function of the type [19]. We shall use a model [14] the idea of which is to add a scalar field and to create this scale dynamically via the equations of motion. Related work on vacuum potentials is in [20].
3.1 Defining the AdS/QCD model
One starts from the gravity + scalar action (in the Einstein frame and in standard notation)
(3.1) 
or, in the string frame, writing
(3.2) 
(3.3) 
One now assumes a metric ansatz
(3.4) 
The four functions in the metric, the scalar field and the potential are then determined as the solutions of the three field equations following from (3.1):
(3.5)  
(3.6)  
(3.7) 
(, etc.) and from a fourth equation,
(3.8) 
where is the beta function of the field theory one is seeking the gravity dual for. This is the crucial assumption of the model. The logic here is that the energy scale of the coupling is identified by , so that () corresponds to the UV (IR). Any similar monotonic function would do and actually in the definition (3.8) one uses . Eq.(3.8) in the context of a black hole horizon at some is discussed below (see Fig. 1).
For given one can from the fourth equation solve , then from the third equation and finally from the second equation so that also and are determined, in terms of a number of integration constants. The role of the first equation then simply is to fix the scalar potential. Defining
(3.9) 
the answer simply is
(3.10) 
where from the independent Eq. (3.6) (which can be written in the form ) and (3.8)
(3.11) 
The normalisation follows from the requirement that the boundary be asymptotically AdS, .
In spite of appearances, the potential in Eq.(3.10) is expressible in terms of only. A thorough discussion of the choice of the potential has been given in [12][15]. The outcome is that SU(3) thermodynamics is well described by the above model with the potential^{2}^{2}2The numerical computation includes in (3.12) also two more terms which enforce QCD asymptotic freedom in the UV. These terms are totally irrelevant for how the model in [15] describes thermodynamics of hot SU(3) matter. In fact, they dominate for which, using (3.18), corresponds to
(3.12) 
To find predictions of the model one must solve the Einstein equations numerically, as discussed in detail on [14, 15]. For concreteness, the parameter values and the normalisation of used there are
(3.13) 
To supplement numerical computations it is very useful to have an analytic approximation. Clearly in the UV is small, but the outcome of the numerical computation is that even at is small, in fact . Assuming is small, , one has
(3.14) 
One observes from (3.10) that a function
(3.15) 
leads to
(3.16) 
Thus the approximate form (3.15) with
(3.17) 
gives the approximate potential in (3.14). We shall base our analysis on this approximate beta function.
Analytical solutions for the metric and the scalar for the model beta function (3.15) are given in the Appendix. Both small or large expansions and bulk thermodynamics can be derived from them. Physically, they are accurate enough for , but clearly the full potential (3.12) is required to describe the phase transition region.
A numerical solution [15, 21] of Eqs.(3.5)  (3.8) gives for the functions and the value of for each value of at the horizon: . Some properties of the functions computed are as follows (”numerics” refers to [15]):

when and, related to this, . In numerics .

There is a horizon at , . In numerics there is a different value of for each value of .

increases monotonically with from so that the leading term at small is
(3.18) where (=1/174 in numerics) is a constant fixed by the normalisation (3.13).

The numerically computed solutions reproduce the vacuum beta function via the relation only when . In fact, if there is a horizon at some , the functions terminate at and thus also . The outcome of a computation of a ”thermal beta function” defined by the relation is shown in Figs. 1 and 2. One should emphasize that these are computed using the full potential (3.12). One also sees how well the approximate beta function (3.15) reproduces the beta function corresponding to the potential (3.12).
3.2 Deriving the spatial string tension
The equations giving the extremal string configurations and the corresponding potential are well known. There are two cases: the spatial string tension at finite is determined from a spacespace Wilson loop, the usual string tension from a timespace Wilson loop at . We shall summarize the equations separately for these two cases. The end points of the string are always at and the string hangs in the fifth dimension so that the maximum depth it reaches at is . We abbreviate
(3.20) 
1. Spacespace loop:
The relation between and is given by
(3.21) 
and the potential is
(3.22)  
(3.23) 
In (3.23) the first term separates a term proportional to , the second term is finite when and (one can put in it) and the third term contains the singularity when ; its divergence is cancelled when the energy of two independent (anti)quarks is included [22]. Numerical examples are given in Fig. 3 for .
One observes from Fig. 3 that the large domain is reached for . To see this analytically, expand the integrand in (3.21) near : , with . Then
(3.24) 
Thus diverges when provided that (always at ). Since the inequality demands . This is true as is seen from Figs. 1 and 2, even down to . From (3.23) with the final result is
(3.25) 
This is the component of the metric (3.4) in the string frame (3.2) multiplied by the string tension [22].
2. Timespace loop.
In this case the relations are ()
(3.26) 
and
(3.27)  
(3.28) 
where the divergence in can be regulated as for an loop. If does not diverge, the equilibrium state is that of two separate quarks [23, 19]. However, taking and expanding near one finds that so that will diverge for that value of satisfying
(3.29) 
From (3.28) the string tension then is given by
(3.30) 
The functions in (3.25) and (3.30) are different, in the former they are computed for , in the latter .
3.3 Evaluating the result for spatial string tension
The lattice results for in finite SU(3) YangMills we want to compare with [3] are measured for . The QCD computation in Section 2 should be a good estimate of the data even at higher temperatures, the better the larger is, see Fig. 5.
It is rather automatic to evaluate the expression (3.25) numerically [21], but to control various effects contributing to it we want to see how far we can get with the approximate potential (3.14) and the approximate beta function in (3.15). The overall normalisation will be fixed from the string tension , measured as [24], using (3.30). This has the important property that the arbitrary normalisation of cancels.
The result (3.25) can be analytically approximated as follows:
(3.31)  
(3.32)  
(3.33)  
(3.34) 
where we first used the small expansions in the Appendix on the leading log level, writing in the arguments of the logarithms , then introduced the explicit values used in the numerics [15] and finally and most subtly, used from the numerics.
To get the normalisation we shall use the lattice data for [24] and its prediction (3.30) in this model. To have an analytic approximation for we again use the beta function , although the accuracy deteriorates with increasing . Since implies one immediately obtains and then from Eq.(A.1) . Thus, with ,
(3.35) 
Numerical computation () with the unapproximated potential (3.12) gives ; (3.35) with parameter values (3.17) gives 0.00146. From (3.30) then
(3.36) 
With the lattice result , the numerical result , the value of and the correct numerical value of (reduction of the RHS of (3.36) by 0.00113/0.00146) this can be converted to
(3.37) 
This result depends on the arbitrary normalisation of .
A comparison of the model with lattice data is shown in Figs. 4 and 5. We comment on the result as follows:

The QCD result (2.3) was given as a numerically computed independent 3d quantity times a perturbatively computed dependence. In analogy, here the result is given as the numerically computed 4d quantity times a dependence computed from a gaugegravity duality model.

Within the range where lattice data exists, it is fitted well by both by the QCD result and the AdS/QCD model discussed here. For larger the QCD prediction varies more rapidly with , , than the AdS/QCD model, . The exponent here is with , as followed from the small expansion of the potential (3.12). In the extreme UV one meets with the QCD beta function with , corresponding to the result behaving . If the QCDlike behavior persists, the present AdS/QCD model will require some modification for .
4 Conclusions
We have in this article considered a particularly simple quantity in finite quarkless QCD, the spatial string tension, both in QCD and in a model for gauge/gravity duality, AdS/QCD. The QCD result is perfectly under control: after one nonperturbative number is determined numerically, the rest follows analytically via symbolic perturbative computations, which may be technically very demanding. For the AdS/QCD model the result needs solutions of Einstein’s gravity equations and a particular contribution of this article has been developing analytically tractable approximations to the model in [14].
There are some interesting parallels and differences in the QCD and AdS/QCD results. Both start from a independent nonperturbative quantity, for QCD the 3 dimensional YangMills string tension, for AdS/QCD from a 4 dimensional computation of the string tension, expressed as . The dependence in QCD comes from perturbatively reintroducing the scales and , in AdS/QCD from the extra dimensional coordinate dependence of the metric and dilaton in the model used to break conformal invariance.
In the range where lattice data exists, Fig. 4, both calculations agree with the data within reasonable estimates of errors. At higher , Fig. 5, the AdS/QCD result, with used here, increases more slowly than the QCD result and it should be possible to distinguish between the two.
There are puzzling features in both approaches. The QCD approach adds perturbative corrections to a nonperturbative number and it is unexpected that it works so well down to . The AdS/QCD model fits very well and elegantly bulk thermodynamics [15] using the potential (3.12), but the power of is quite different from that in the QCD running coupling (2.7). This is not surprising near but one would expect that at least at some high the matter would probe distances small enough to see QCD asymptotic freedom.
The spatial string tension has also been measured for QCD [5]. An interesting further project would be to study this in AdS/QCD by developing the AdS/QCD model used here to also include fundamental quarks.
Acknowledgements. We thank E. Kiritsis, Mikko Laine, York Schröder and Patta Yogendran for discussions and correspondence. We are particularly indebted to Francesco Nitti for the Mathematica code used to solve the Einstein equations of this model and for advice concerning its use.
Appendix A Appendix
We summarize here the solution of Eqs.(3.5)(3.8) for the beta function (3.15), . First, from Eq.(3.8),
(A.1) 
where is a constant, the analogue of , the scale at which diverges. From (3.11)
(A.2) 
The second equation can then be written in the form
(A.3) 
from which by integration
(A.4) 
where
(A.5) 
Correct dimensions are here given by
(A.6) 
Using (A.3) one can replace as a fifth coordinate by , or . For example,
(A.7)  
If the potential satisfies the constraint (3.10), there is a horizon with the temperature
(A.8)  
Using (A.5) one can derive various small (, small , large ) approximations. One has, for example,
(A.9)  
(A.10)  
(A.11)  
(A.12) 
Finally, as in [14, 15] one can obtain the entire bulk thermodynamics by integrating , .
References
 [1] E. Manousakis and J. Polonyi, “Nonperturbative length scale in high temperature QCD,” Phys. Rev. Lett. 58, 847 (1987).
 [2] G. S. Bali, J. Fingberg, U. M. Heller, F. Karsch and K. Schilling, “The Spatial string tension in the deconfined phase of the (3+1)dimensional SU(2) gauge theory,” Phys. Rev. Lett. 71, 3059 (1993) [arXiv:heplat/9306024].
 [3] G. Boyd, J. Engels, F. Karsch, E. Laermann, C. Legeland, M. Lutgemeier and B. Petersson, “Thermodynamics of SU(3) Lattice Gauge Theory,” Nucl. Phys. B 469, 419 (1996) [arXiv:heplat/9602007].
 [4] J. Liddle [RBCBielefeld Collaboration], “The spatial string tension and dimensional reduction in QCD,” PoS LAT2007, 204 (2007) [arXiv:0710.2011 [heplat]].
 [5] M. Cheng et al., “The Spatial String Tension and Dimensional Reduction in QCD,” Phys. Rev. D 78, 034506 (2008) [arXiv:0806.3264 [heplat]].
 [6] B. Lucini and M. Teper, “SU(N) gauge theories in 2+1 dimensions: Further results,” Phys. Rev. D 66, 097502 (2002) [arXiv:heplat/0206027].
 [7] M. Laine and Y. Schroder, “Twoloop QCD gauge coupling at high temperatures,” JHEP 0503, 067 (2005) [arXiv:hepph/0503061].
 [8] Y. Schroder and M. Laine, “Spatial string tension revisited,” PoS LAT2005, 180 (2006) [arXiv:heplat/0509104].
 [9] O. Andreev, “The Spatial String Tension in the Deconfined Phase of SU(N) Gauge Theory and Gauge/String Duality,” Phys. Lett. B 659, 416 (2008) [arXiv:0709.4395 [hepph]].
 [10] O. Andreev and V. I. Zakharov, “The Spatial String Tension, Thermal Phase Transition, and AdS/QCD,” Phys. Lett. B 645, 437 (2007) [arXiv:hepph/0607026].
 [11] O. Andreev, “Some MultiQuark Potentials, PseudoPotentials and AdS/QCD,” Phys. Rev. D 78, 065007 (2008) [arXiv:0804.4756 [hepph]].
 [12] U. Gursoy and E. Kiritsis, “Exploring improved holographic theories for QCD: Part I,” JHEP 0802, 032 (2008) [arXiv:0707.1324 [hepth]].
 [13] U. Gursoy, E. Kiritsis and F. Nitti, “Exploring improved holographic theories for QCD: Part II,” JHEP 0802, 019 (2008) [arXiv:0707.1349 [hepth]].
 [14] U. Gursoy, E. Kiritsis, L. Mazzanti and F. Nitti, “Holography and Thermodynamics of 5D Dilatongravity,” arXiv:0812.0792 [hepth].
 [15] U. Gursoy, E. Kiritsis, L. Mazzanti and F. Nitti, “Improved Holographic YangMills at Finite Temperature: Comparison with Data,” arXiv:0903.2859 [hepth].
 [16] P. Giovannangeli, “Two loop renormalization of the magnetic coupling in hot QCD,” Phys. Lett. B 585, 144 (2004) [arXiv:hepph/0312307].
 [17] K. Farakos, K. Kajantie, K. Rummukainen and M. E. Shaposhnikov, “3D physics and the electroweak phase transition: Perturbation theory,” Nucl. Phys. B 425, 67 (1994) [arXiv:hepph/9404201].
 [18] A. Hietanen, K. Kajantie, M. Laine, K. Rummukainen and Y. Schroder, “Threedimensional physics and the pressure of hot QCD,” Phys. Rev. D 79, 045018 (2009) [arXiv:0811.4664 [heplat]].
 [19] K. Kajantie, T. Tahkokallio and J. T. Yee, “Thermodynamics of AdS/QCD,” JHEP 0701, 019 (2007) [arXiv:hepph/0609254].
 [20] Dingfang Zeng, “Heavy quark potentials in some renormalization group revised AdS/QCD models,” Phys. Rev. D 78, 126006 (2008) [arXiv:0805.2733 [hepth]].
 [21] Francesco Nitti, private communication.
 [22] Y. Kinar, E. Schreiber and J. Sonnenschein, “Q antiQ potential from strings in curved spacetime: Classical results,” Nucl. Phys. B 566, 103 (2000) [arXiv:hepth/9811192].
 [23] S. J. Rey, S. Theisen and J. T. Yee, “WilsonPolyakov loop at finite temperature in large N gauge theory and antide Sitter supergravity,” Nucl. Phys. B 527, 171 (1998) [arXiv:hepth/9803135]. A. Brandhuber, N. Itzhaki, J. Sonnenschein and S. Yankielowicz, “Wilson loops in the large N limit at finite temperature,” Phys. Lett. B 434, 36 (1998) [arXiv:hepth/9803137].
 [24] M. Teper, “Large N,” arXiv:0812.0085 [heplat].