Pressure and compressibility of conformal field theories from the AdS/CFT correspondence

The equation of state associated with ${\cal N}=4$ supersymmetric Yang-Mills in 4 dimensions, for $SU(N)$ in the large $N$ limit, is investigated using the AdS/CFT correspondence. An asymptotically AdS black-hole on the gravity side provides a thermal background for the Yang-Mills theory on the boundary in which the cosmological constant is equivalent to a volume. The thermodynamic variable conjugate to the cosmological constant is a pressure and the $P-V$ diagram is studied. It is known that there is a critical point where the heat capacity diverges and this is reflected in the isothermal compressibility. Critical exponents are derived and found to be mean field in the large $N$ limit. The same analysis applied to 3 and 6 dimensional conformal field theories again yields mean field exponents associated with the compressibility at the critical point.


Introduction
Ever since Bekenstein's discovery of the relation between entropy and the area of a black hole's event horizon [1] and Hawking's subsequent observation that black holes have an intrinsic temperature associated with them [2], the subject of black hole thermodynamics has been a fascinating area of research. Indeed black hole thermodynamics now has potentially great usefulness in providing insights into the thermodynamics of conformal theories via the AdS/CFT correspondence [3].
An interesting aspect of black hole thermodynamics is the role of the cosmological constant, Λ, as a thermodynamic variable: an idea originally considered in [4] and re-visited in [5]. Since Λ has the natural physical interpretation of pressure in the bulk, the conjugate variable is a volume and this has led to some intriguing recent results for the pressure and volume of black holes, with phase transitions and critical points showing very similar properties to liquid-gas phase transitions [6,7]. For Einstein gravity in the bulk, the critical exponents are mean field for all black hole solutions studied to date, leading to an analogy with the liquid-gas transition of a van der Waals fluid.
In the context of AdS/CFT it was suggested in [5,8] that varying the cosmological constant in the bulk should be associated with varying the number of colors in the boundary CFT and this proposal was investigated in more detail in [9], where the thermodynamically conjugate variable to Λ was interpreted as a kind of chemical potential for color.
An alternative approach is followed here based on a different interpretation of the thermodynamic significance of Λ which provides a length scale in the CFT. When global AdS co-ordinates are used the CFT on the boundary has a finite volume, determined by Λ, hence varying Λ in the bulk corresponds to varying the volume of the CFT on the boundary. If Λ gives the volume of the CFT then the thermodynamically conjugate variable is the pressure and one is led to the construction of P − V diagrams for the CFT, in which the pressure and volume are interchanged relative to their roles in the bulk. This can be achieved while keeping the number of colors fixed provided the higher dimensional Newton's constant is adjusted as Λ is varied, as suggested in [10]. In 10-dimensional string theory, for example, Newton's constant is related to the string coupling constant and tension by G 10 = 8π 6 g 2 s (α ′ ) 4 so varying G 10 can be thought of as varying g s with the tension fixed.
In this paper the P − V diagram associated with 4-dimensional N = 4 SUSY SU(N) Yang-Mills theory at large N is investigated using the AdS/CFT correspondence. When the bulk contains a black hole there is a first order phase transition [11] which is associated with the deconfining transition for the quark-gluon plasma on the boundary [12]. When the black hole carries a U(1) charge, corresponding to R-charge in the Yang-Mills theory, this phase transition is one end of a line of first order phase transitions while the other end terminates at a second order transition where the heat capacity at constant volume for the Yang-Mills theory diverges [13]. The critical exponents in the bulk gravitational theory, with charge as the order parameter and temperature as the control parameter, are known to be mean field, [14], and so C V is finite at the critical point for the black hole in the bulk. At first sight this is at odds with the statement in [13] that there is a critical point in the CFT at which C V diverges. We shall show that there is no contradiction here and, when interpreted correctly, the boundary CFT has mean field exponents. In terms of pressure and volume there is a number of aspects of the phase transition that make it different from more usual cases. Firstly there is a single phase at low temperatures and the two phase regime exists for temperatures above the critical point [13]; secondly, above the critical point, along the line of first order phase transitions, it is the pressure that jumps across the phase transition, not the volume, so it is more appropriate to use the pressure as an order parameter rather than the more usual volume; thirdly the conformal symmetry dictates that volume and temperature are not fully independent and it is better to use charge as the control parameter rather than the temperature. This last point of view is more in keeping with the notion of the phase transition being a quantum phase transition rather than a thermal phase transition. With this interpretation the critical exponents for pressure and volume of the Yang-Mills theory are calculated in the large N limit and shown to be mean field.
The general structure is the same for the 3-dimensional and 6-dimensional CFT's considered in [15], in particular the critical exponents at large N are mean field in all three conformal field theories.
In section 2 the black hole thermodynamics of the relevant bulk solution is summarized and is related to that of the boundary CFT in section 3. Section 4 analyzes the case of N = 4 SUSY Yang-Mills in detail, the P − V diagram is constructed and critical exponents for the deconfining phase transition in the large N limit are calculated. Finally section 5 summarizes the results. Some technical details are in two appendices.

The bulk
In D space-time dimensions the Lagrangian for Einstein gravity, with a cosmological constant Λ, coupled to a U(1) gauge field, is The normalization of F is the same as in [13], it has dimensions of inverse length.
A solution of the equations of motion arising from (1), corresponding a spherical charged asymptotically AdS black hole in D space-time dimensions is easily written down. In global co-coordinates the gauge potential is where is the volume of a unit (D − 2)-sphere. The constant c 0 accommodates some freedom in the choice of gauge. The normalization of the U(1) charge Q here is determined by requiring that the gauge field F = dA satisfies Gauss law where S D−2 is a sphere containing the charge. The line element is where and d 2 Ω D−2 is the line element on a unit (D − 2) sphere. There is an event horizon and the largest root of f (r) = 0 will be denoted by r h . It is natural to chose a gauge in which the potential above vanishes at the outer horizon, c 0 = 1 The parameters q and µ are then related to Q and the mass M of the black hole by and (we use units with c = 1, but keep the D-dimensional Newton's constant explicit). From (5), with f (r h ) = 0, and (7) The area of the event horizon is Ω D−2 r D−2 h and the Bekenstein-Hawking entropy is From the point of view of black hole thermodynamics, the mass is interpreted as the internal energy of the system, M = U(S, Q), and the Bekenstein-Hawking temperature is It will prove useful to define dimensionless variables, x := r h L and y := q L D−3 which can be used in lieu of S and the charge. In terms of these variables and In the context of superstring theory, the D-dimensional Newton's constant G D is descended from Newton's constant in the full D-dimensional theory (D = 10 or 11). Compactifying the higher dimensional theory on a (D − D)-dimensional compact space K D−D , with size L , allows for a direct where C is a dimensionless number determined by the metric on K D−D .
The AdS length scale L is not intrinsic to the D-dimensional action, but is merely a parameter in a classical solution of the D-dimensional theory and it is perfectly reasonable to consider varying L in the solution.

Thermodynamics of the boundary field theory
In the AdS/CFT correspondence the boundary CFT theory has d = D − 2 space dimensions. With the global co-oordinates used in the previous section, the (d + 1)-dimensional space-time metric at fixed r is which is conformal to a (d + 1)-dimensional space-time with constant time spatial volume and this can be interpreted as the spatial volume of the boundary conformal field theory. The dimensionless ratio of the AdS length scale L to the D-dimensional Planck length in the bulk determines the number of degrees of freedom in the boundary field theory [15].
The most studied case is D = 10 compactified on K 5 = S 5 , with D = 5. the boundary CFT is then N = 4 SUSY SU(N) Yang-Mills theory, when the classical solution in the bulk (4) is relevant for large N. This has 16N 2 degrees of freedom (8N 2 bosonic and 8N 2 fermionic) and the large N limit is the weak gravity limit: hG 10 is of course related to the 10-dimensional Planck length,hG 10 = l 8 P l and, as usual, large N is related to the classical limit of the bulk theory, in which L >> l P l . The CFT is on a 3-sphere with volume and the inverse of the 5-dimensional Planck length (16) is related to the number of degrees of freedom per unit volume of the CFT so varying the volume of the CFT, keeping N 2 fixed, is completely equivalent to varying the 5-dimensional Planck length. The entropy is proportional to N 2 (with x := r h L as before) as is the dimensionless charge where y := q L 2 . On dimensional grounds the internal energy U(S, V, Q) = M is of the form with u(S, Q) a dimensionless function of dimensionless variables. The thermodynamics of the CFT is determined by the functional dependence of the mass on S, Q and V (or equivalently, x, y and L), explicitly Two other interesting cases are the 3-dimensional CFT obtained from M-theory in D = 11 compactified on AdS 4 × S 7 (more generally S 7 /k for integral k, [3,13]) and the 6-dimensional CFT obtained from M-theory in D = 11 compactified on AdS 7 × S 4 . For these two cases the N 2 behavior of the "extensive" quantities M, S and Q is replaced with N 3/2 and N 3 respectively and the powers of x change as in equations (11) and (12) but otherwise the analysis of the thermodynamics is similar to the case D = 10 and D = 5 studied in §4 below.

The thermodynamics of N = SUSY Yang-Mills
To explore the thermodynamics of the CFT fully we wish to fix N and allow S, Q and V to vary. This means we must vary L, but at the same time vary the D-dimensional Newton constant in such a way that N 2 in is fixed in the first equation of (16), [10].
For concreteness we shall focus on D = D = 5, the general structure is the same the other two cases, in particular the critical exponents are the same. Thermodynamically we interpret M = U(S, Q, V ) in (22) as the internal energy of the large N Yang-Mills theory at strong coupling, Note that, while the mass in the bulk is classical, the internal energy of the CFT is quantum mechanical and vanishes ash → 0 with x and y fixed. The temperature is then in the deconfined phase this is interpreted as the temperature of the quarkgluon plasma [12,13]. The pressure is [10] where ǫ = U V is the energy density. Hence and the speed of sound is given by simple consequences of the fact that dimensionally, Lastly the chemical potential is These expressions simplify at high T (large x) when the event horizon curvature and the charge are negligible. This is deep into the de-confined phase of the quark-gluon plasma and in this limit Since T ∝h we see that the pressure and the chemical potential here are of quantum mechanical origin. At fixed charge and temperature there is a phase transition [13], an extension of the Q = 0 Hawking-Page phase transition in the bulk [11], which is interpreted as the deconfining phase transition in the boundary Yang-Mills theory [12]. The heat capacity diverges when ∂T ∂S V,Q = ∂ 2 U ∂S 2 V,Q vanishes and there is a critical point when = 0. This happens at a critical point that was first found in [13]. The critical temperature is given by .
Conformal invariance dictates that only the combination V 1 3 T is determined at critical point, V and T are not fixed separately. 2 2 Finite temperature field theory is associated with periodicity in Euclidean time with period 1/T and Euclidean time parameterizes a circle S 1 of radius 1 2πT . The spatial geometry is S 3 with radius L and, because the theory is conformal, the physics can only depend on the ratio of the radii of these two spheres, namely 2πT L. The physics depends on V 1 3 T , not on V and T separately, [12].
The specific heat critical exponent α can easily be determined by fixing y = y * and expanding 1/C V,Q around x * . With α can be extracted by similarly expanding the reduced temperature at constant volume, The critical exponent follows from This singularity in the heat capacity was not found in the bulk thermodynamics of higher dimensional asymptotically AdS charged black holes studied in [14,16], these authors found a critical point in the bulk but the exponents were mean field and C V,Q is finite there. To understand this apparent contradiction we write the full equation of state in terms of Q and Φ. Using (20) and (28) to eliminate x in favor of Q and Φ in (24) gives the equation of state Expanding about the critical point (31) with y = y * (1 + η), T = T * (1 + t) and Φ = Φ * (1 + ϕ), where LΦ * := 1 2 √ 5 , leads to Aside from a change in the sign of t this has the same analytic structure as the equation of state for a van der Waals gas [14] which, in reduced variables p = P −P * P * The van der Waals equation of state has mean field exponents, so one expects α = 0. The previous calculation gave α = 2 3 because Q was held fixed (fixed η) but the order parameter for the liquid-gas phase transition in the van der Waals gas is v, which would be analogous to ϕ in (38). Indeed C V,Φ is perfectly finite, [13], so α is indeed zero for fixed Φ, which is the analogue of the van der Waals case.
If Q is the order parameter the usual exponents are defined by and the response function, behaves as

. The equation of state (38) results in
While these exponents, together with α = 2 3 , do satisfy the Rushbrooke and Widom scaling relations there is a difficulty in that γ is negative so the response function vanishes at critically rather than diverging. As indicated in (42) the response function χ T can be negative near the critical point, and this is the instability found in [17]. 3 It was also shown in [17] that Φ jumps across the phase transition in the two phase regime, suggesting that Φ is the order parameter for this transition.
If Q is just fixed from the start and not varied then there is no problem, C V,Q diverges with exponent 2 3 and β, γ and δ are not relevant. But if we wish to probe the system by adjusting Q then Φ should be viewed as the order parameter, not Q. In that case C V,Q diverges as |t| −1 rather than |t| − 2 3 because it should be calculated for ϕ = 0 and not η = 0.
The other exponents, ϕ ∼ |t| β and ϕ ∼ |η| δ , are easily obtained from (38) in the usual way and are found to be mean field In particular the heat capacity C V,Φ is finite (and negative) at the critical point [13]. Mean field behavior with Φ as the order parameter for the black hole in the bulk was first found in [14].
Alternatively the equation of state can be written in terms of the pressure rather than the chemical potential to study the compressibility. The adiabatic compressibility of the plasma follows easily from (25) and is but the isothermal properties are more interesting. It is shown in appendix A that the isothermal compressibility, κ T,Q , is given by Hence κ T,Q is vanishes at the critical point and is negative along Q = Q * close to the critical point when C V,Q * > 0 diverges. Thus the P -V response function, κ T,Q , behaves the same way as the Q-Φ response function χ T . This suggests that the order parameter here is P rather than V . If this is a valid point of view we should focus on the heat capacity at constant pressure, C P,Q , rather than C V,Q . Using the standard relation equations (47) and (48) give which is finite and negative when C V,Q diverges, just as C V,Φ is. In particular at the critical point Thus, with P as the order parameter, α = 0 is mean field. The negative value of C P at the critical point is a reflection of the instability found above in χ T . There is a subtlety in trying to extract β and δ as there is no independent definition of a critical volume, V and T are linked due to conformal invariance. In particular it would be wrong to assume the usual scaling relations to derive β and δ from α and γ above as conformal invariance imposes an extra constraint.
A reduced volume can be defined as but this is not a new variable since conformal invariance relates it directly to t, As a consequence the usual definitions of the exponents β and δ do not apply.
Since v and t are not independent it is better, when discussing pressure and volume, to use η to probe the physics near the critical point rather than t. Lines of equal charge in the P -V plane are plotted in figure 1 where P and V are rendered dimensionless by multiplying by appropriate powers of T . The isothermal compressibility is positive when the slope of the plotted curves is negative and there are two regions in the P -V plane where this is the case, the critical point lies on the boundary of the rightmost of these regions which corresponds to Q < Q * and v > 0 (hence t > 0, i.e. the high temperature regime). The pressure of the system jumps across the phase transition.
To calculate critical exponents first define a reduced pressure at fixed volume. 4 Then, with p as the order parameter and Q as the control variable, β and δ are defined by in analogy with the usual definition of β, but with t replaced by η and p as the order parameter (so p and v are interchanged). Similarly Lastly the relevant response function is We leave the details to an appendix and here quote the result that β, γ and δ are indeed mean field. We conclude that, with the proper identification of the order parameter as being the thermodynamic variable that jumps across the line of first order phase transformations, the exponents are always mean field in the large N limit, both in the Φ-Q plane and the P -V plane.

Summary
The critical point of N = 4 SUSY SU(N) Yang-Mills in the large N limit, first found in [13,17], is also visible in the P − V diagram of the quark-gluon plasma. The volume here is provided by the cosmological constant in the 5-D bulk, Λ = − 6 L 2 , which is related to the volume of 3-space, S 3 , in the boundary Yang-Mills theory by V = 2π 2 L 3 . The pressure is then defined through the thermodynamic relation P = − ∂U ∂V , where U is the internal energy of the thermodynamic system -identified with the mass of the black hole in the bulk. This is in contrast to the thermodynamics of the black hole itself, where the roles of pressure and volume are reversed so that, on the gravity side, the cosmological constant provides the pressure and the thermodynamically 4 One can also use the temperature to render P dimensionless and define the reduced pressure as p = P/T 4 −(P/T 4 ) * (P/T 4 ) * . This does not change the critical behavior, in particular it gives the same critical exponents. conjugate variable is a volume and the black hole mass is the enthalpy of the system [5].
There is a constraint on the thermodynamic variables, due to the fact that the boundary field theory is conformal, and only the combination V T 3 is relevant to the phase structure, not V and T separately. As a consequence it seems better to use the charge rather than the temperature as the control parameter in discussing the thermodynamics. Then the pressure jumps across the phase transition in the two phase regime. In the approximation in which the fluid is treated as a gas of non-interacting particles -quarks and gluons in the plasma and non-interacting hadrons in the confined phase -this jump may be viewed as simply due to a change in the number of degrees of freedom: each degree of freedom contributes equally to the pressure any change across the phase transition in the number of effective degrees of freedom contributes to a change in pressure.
With pressure as the order parameter in the P -V plane the exponents are mean field and the phase transition is in the same universality class as the van der Waals gas. Mean field exponents also characterize the phase transition in the Φ-Q plane.
An unsatisfactory feature of the analysis is the instability associated with the negative value of C P at the critical point, implying that the system is unstable there when the pressure is fixed. Although κ T is positive in the region enclosed by the black curve to the right of the critical point in figure  1 both C P and C V are negative there. The only region of the P − V plane that has all three of C P , C V and κ T positive is the region to the left of the left-hand black curve in the figure, the single phase region. C P < 0 to the right of this curve.
Instability in black hole thermodynamics is not new and indeed lies behind the phenomenon of black hole evaporation -an asymptotically flat Schwarzschild black hole also has negative C P . Whether or not negative C P is a problem is a question of time-scales, for Hawking radiation the time-scale for black hole evaporation can be very large, so large that the system is essentially quasi-static equilibrium and thermodynamic principles can still be applied, at least at early times before the evaporation process has gone too far. It is not clear whether the same thing can be said for the quark-gluon plasma, or what the physical interpretation of the instability should be in this case.
Although the analysis here has focused on the case of 4-dimensional N = 4 SUSY SU(N) Yang-Mills, associated with string theory in D = 10 compactified on AdS 5 × S 5 , the thermodynamics of the 3-dimensional and 6-dimensional CFTs, obtained from M-theory in D = 11 compactified on AdS 4 × S 7 /k and on AdS 7 × S 4 respectively, is essentially similar -the critical exponents are mean field in all three cases.
It would be of great interest to evaluate 1/N corrections to the exponents but the method employed here relies on a representation of the thermodynamic potentials arising from an exact solution of the Einstein equations in the bulk. The paucity of known exact solutions relevant to the problem is an obstacle to realizing 1/N corrections with this method. However it may be possible to make some progress in studying 1/N corrections by adding a Gauss-Bonnet term to the 5-D Einstein action [20].
Acknowledgment: This article is based upon work from COST Action MP1405 QSPACE, supported by COST (European Cooperation in Science and Technology).

A Free energy calculations
To study the theory at fixed T in more detail consider the free energy which is of the form where and F is a dimensionless function of dimensionless variables. The free energy is most easily expressed in parametric form using (17) and (24), and τ = 1 (4π) In terms of the functions F (x, y) and τ (x, y) in equations (62) and (63), whereḞ = ∂F ∂τ , and Note also that ∂S From this follows Now, since P = U 3V , we have Thus the isothermal compressibility, κ T,Q vanishes at the critical point.
Expanding this around the critical point one finds that κ T,Q vanishes like where (69) has been used to eliminate η in the second equation. Now we can set p = 0 in (70) and equation (69) then implies ε ≈ 1 2 t and η ∼ −24ε, so Hence (κ T,Q ) −1 ∼ 1/|η| ∼ 1/t at the critical pressure and, with pressure as the order parameter, γ = 1. In the 2-phase regime (η < 0) the isothermal compressibility is positive, but it is negative for η = 0. Lastly setting η = 0 in the parametric equation of state (72) and (73) implies that p ≈ 16 7 ε and v ≈ 5 2 ε 3 so and hence δ = 3. p v v 0 p < p > Figure 2: Maxwell construction in the P -V plane along a curve of constant charge with P as the order parameter (P and V are rendered dimensionless using appropriate powers of T as in figure 1).