Shear viscosity from thermal fluctuations in relativistic conformal fluid dynamics J. Peralta-Ramos1,2,∗ and E. Calzetta2,† 1Instituto de F´ısica Te´orica, Universidade Estadual Paulista, Rua Doutor Bento Teobaldo Ferraz 271 - Bloco II, 01140-070 Sa˜o Paulo, Brazil 2 2Departamento de F´ısica, FCEyN-UBA, 1 0 and IFIBA-CONICET, Ciudad Universitaria, 2 n Pabell´on I, Buenos Aires 1428, Argentina a J (Dated: February 1, 2012) 1 3 Abstract ] h Within the framework of relativistic fluctuating hydrodynamics we compute the contribution p - of thermal fluctuations to the effective infrared shear viscosity of a conformal fluid, focusing on p e h quadratic(influctuations),secondorder(invelocitygradients)termsintheconservationequations. [ Our approach is based on the separation of hydrodynamic fields in soft and ultrasoft sectors, in 3 v which the effective shear viscosity arises due to the action of the soft modes on the evolution 3 3 of the ultrasoft ones. We find that for a strongly coupled fluid with small shear viscosity–to– 8 3 . entropy ratio η/s the contribution of thermal fluctuations to the effective shear viscosity is small 9 0 but significant. Using realistic estimates for the strongly coupled quark–gluon plasma created in 1 1 heavy ion collisions, we find that for η/s close to the AdS/CFT lower bound 1/(4π) the correction : v i is positive and at most amounts to 10% in the temperature range 200–300 MeV, whereas for larger X r a values η/s 2/(4π) the correction is negligible. For weakly coupled theories the correction is very ∼ small even for η/s = 0.08 and can be neglected. ∗Electronic address: [email protected] †Electronic address: [email protected] 1 I. INTRODUCTION The heavy ion collisions experiments performed at RHIC and LHC create a hot and dense medium, the so-called Quark-Gluon plasma (QGP), that is believed to be strongly coupled. The most compelling evidence supporting this idea comes from relativistic viscous fluid dynamics simulations that reproduce the momentum anisotropy patterns measured by experiment (see, e.g. Refs. [1–12]). The momentum anisotropy is the translation to momentum space of the initial spatial eccentricity of non central collisions. A weakly in- teracting system of particles can not convert spatial anisotropy into momentum anisotropy in an efficient way, but this translation can occur quite efficiently if the particles interact strongly. The observed large magnitude of the momentum anisotropy indicates that the QGP is strongly coupled. Another indication of the strongly coupled nature of the QGP is the strong quenching of high-p probes measured at RHIC [1–4]. T Transport coefficients such as the shear (η) and bulk (ζ) viscosities are crucial inputs in fluid dynamics simulations attempting to describe the evolution of matter created in heavy ion collisions. Great efforts are currently focused on developing new theoretical tools to compute these transport coefficients more accurately from microscopic models [13–28], and also on extracting them more precisely from RHIC and LHC measurements [5–11]. Results obtained from Lattice QCD calculations [13, 19, 29, 30] indicate that η/s and ζ/s, where s is the entropy density, depend significantly on temperature, showing a minimum and maxi- mum, respectively, at the critical temperature corresponding to the QGP–hadron crossover. Note that although it is expected that the transport coefficients of the QGP depend on temperature, the impact of a temperature–dependent η/s on momentum anisotropies as obtained from simulations has been investigated only recently [31–35]. It is possible to extract an average value of η/s by matching the particle spectra and elliptic flow obtained from fluid dynamics [8] or hybrid fluid–kinetic simulations [9–11] to data. See also Refs. [36–38] for a description of other approaches to extract values of η/s from data. The optimal value of η/s that comes out of all these fits is close to the lower bound η/s = 1/(4π) that was found by Kovtun, Son and Starinets [39] (KSS bound from now on) via the Anti de Sitter/Conformal Field Theory (AdS/CFT) correspondence. However, deviations from the KSS bound by a factor of two or more are still possible due to uncertainties stemming from diverse sources such as the initial conditions, the dynamics of 2 chiral fields coupled to the quark fluid, and the freeze out stage, to mention a few – see Refs. [5–12, 35, 40–45]. Very low values of η/s are also favored by studies of the overall entropy production during a collision – see e.g. [46] – and, albeit more indirectly, by the fact that flow structures such as cones and ridges are able to survive until freeze out. If dissipative effects were too large, such collective flow structures would be washed away much earlier in the evolution of the fireball [47–50]. The system created in heavy ion collisions consists of a fireball of quarks and gluons that expands and cools very rapidly under its own pressure. Since the fireball is formed out of a finite number of particles and has a small size in the range 5–10 fm, it is natural to expect that thermal fluctuations may have an observable impact on its evolution. In spite of this expectation, the role of thermal fluctuations and its impact on hydrodynamic evolution has only very recently been discussed in the context of heavy ion collisions [51–53]. It was shown in Ref. [52] that for the QGP with small values of η/s near the KSS bound, thermal fluctuations on top of long–lived sound and shear waves can contribute significantly to the effective value of η/s as would be measured on large scales. In contrast, for larger values η/s 2/(4π) the correction coming from thermal fluctuations on sound and shear ∼ waves was found to be negligible. Specifically, in [52] the authors used Kubo’s formula together with second order hydrodynamics and focused on corrections to η/s coming from the zeroth order (in gradients), quadratic term (ρ + P )uµuν in the energy momentum 0 0 tensor Tµν, where ρ and P are the equilibrium energy density and pressure (P = ρ /3 0 0 0 0 for a conformal fluid), respectively, and uµ is the flow velocity fluctuation ([52] considers a vanishing background velocity). A related study on the effective shear viscosity of the strongly coupled QGP was carried out in [53] using the AdS/CFT correspondence, where corrections to η arising from higher order velocity gradients were computed using the idea of resumming those corrections into an effective η(ω,k) depending on both frequency and momentum. In this work we compute the correction to the shear viscosity that come from the effect of relatively small scale fluctuations of the hydrodynamic modes on their evolution on larger scales. For this purpose, we use the equations of relativistic fluctuating fluid dynamics [51, 54–58] and focus on terms which are second order in velocity gradients (in Tµν these termsarefirst order)andquadraticinvelocity fluctuations. Thedevelopment presented here is based on a division of hydrodynamic modes in ultrasoft and soft sectors, both of them 3 subjected to the influence of stochastic noise coming from hard (i.e. particle–like) modes. It is, therefore, inspired in the hard thermal loops approach to thermal field theory [60–62] and particularly in earlier studies of the Langevin dynamics of soft and ultrasoft modes [63–67]. As this approach does not rely on Kubo’s formula it is different from the one adopted in [52]. We anticipate that the correction computed here is of order χ2 p2 η2/(sT)2 or higher, ≡ max where p is the value of momentum beyond which the second order gradient expansion max breaks down, while the correction obtained in [52] is of order χ. Since χ must be small for a fluid description to be reliable, the correction to η/s obtained here is suppressed with respect to the one of [52], but as we will show it is still significant. We will compare our results with those of [52] in Sections III and IIIA. Itisappropriateat thispoint toemphasize thatour approachispurely phenomenological. It is not our intention to give a precise value of the correction to the microscopic shear viscosity arising from second order terms, but rather to provide a reasonable upper bound for it under realistic conditions for the QGP. We note that the correction that we obtain, as well as that obtained in [52], is inversely proportional to η/s, and therefore diverges for vanishing η/s. As pointed out recently by Torrieri [68] in the context of ideal quantum hydrodynamics, the cure for this divergence may lie in quantum corrections that set in for small η/s and therefore “correct the correction”. Although this is an important issue with strong implications forthestudy ofmatter created inheavy ioncollisions andsurely deserves further attention, here we shall not dwelve further into it since we focus on the correction to η/s arising at finite values of this ratio. The paper is organized as follows. In Section II we describe the theoretical setup and obtain the equations for the ultrasoft and soft modes. In Section III we compute the correc- tion to the shear viscosity arising from second order soft terms, and then provide numerical estimates for a weakly and a strongly coupled theory, in particular focusing on the QGP created in heavy ion collisions. Finally, we conclude with a brief summary and outlook in Section IV. Details on the calculation of the correction to η/s are given in Appendix A, while a description of the procedure used to compute the frequency that separates ultrasoft and soft modes is given in Appendix B. 4 II. FLUCTUATING CONFORMAL HYDRODYNAMICS A. Soft and ultrasoft hydrodynamic modes Weconsideraconformalfluidinflatspacetimewithsignature(+, , , ). Thecomoving − − − time and space derivatives are D = u ∂µ and µ = ∆µν∂ , respectively. The hydrodynamic µ ν ∇ equations at second order in velocity gradients read 4 1 ρDuµ µρ+η∆µ∂ σαβ = fµ 3 − 3∇ α β (1) 4 Dρ+ ρ uµ ησµνσ = f µ µν 3 ∇ − where we have made the approximation Πµν = ησµν = η <µuν> which is enough for our ∇ purposes here. The brackets <> around indices denote the symmetric and traceless pro- jection orthogonal to uµ, while round brackets denote symmetrization. η is the microscopic shear viscosity as derived e.g. by Kubo’s formula, and (f,fµ) are stochastic noises. For recent reviews on relativistic viscous fluid dynamics as applied to heavy ion collisions see Refs. [5–7]. Next we divide the hydrodynamic fields into “ultrasoft” and “soft” parts, ρ = ρ +ρ and uµ = uµ +uµ (2) < > < > Ontopofthisseparationintoultrasoftandsoftsectors, weperfomanexpansionindeviations from equilibrium characterized by a small parameter, ǫ, staying at linear order, thus ρ = ρ +ǫρ +... < 0,< 1,< ρ = ρ +ǫρ +... > 0,> 1,> (3) uµ = uµ +ǫuµ +... < 0,< 1,< uµ = uµ +ǫuµ +... > 0,> 1,> In this way (ρ + ρ ,uµ + uµ ) correspond to thermal equilibrium. For notational 0,< 0,> 0,< 0,> simplicity is it convenient to define ρ ρ , ρ ρ 0 0,< 1 1,< ≡ ≡ δρ ρ , δρ ρ 0 0,> 1 1,> ≡ ≡ (4) Vµ uµ , Wµ u ≡ 0,< ≡ 1,< sµ uµ , tµ uµ ≡ 0,> ≡ 1,> 5 It will be shown in the next section that Vµ = (V0,0,0,0), with V0 close to but not exactly unity, and that V sµ = V Wµ = 0, which imply that sµ = (0,~s) and Wµ = (0,W~ ) in the µ µ local rest frame (LRF). We note that V tµ = 0. Additional constrains arising from the µ 6 normalization of uµ are discussed in the next section. Note that our treatment is purely hydrodynamic, i.e. assuming momenta to be small, so no attempt is made to actually derive the noise as coming from the hard sector of a micro- scopic field which should be described by transport equations such as Boltzmann equation [56, 63–67]. So, the ultrasoft and soft modes of the hydrodynamic fields are described by the conservation equations of fluid dynamics, while the effect of the hard modes on the hydrodynamic evolution is included through the noises (f,fµ). This is not a limitation to our purpose here, since we intend to determine the correction to η coming from long–lived hydrodynamic modes (shear and sound waves). Areasonable estimate for the valueof the momentum separating soft andhard modes will be given in Section III. This estimate is essentially based on the requirement that second order terms in derivatives must be smaller than first order terms, i.e. that the gradient expansion does not break down. The value of momentum that separates ultrasoft from soft modes can be computed by requiring that the equations of the soft modes be linearizable (see, e.g., [56, 59]). This is done in Appendix B. Due to the phenomenological nature of our approach, and taking into account that we intend to provide an upper bound to the correction to η/s, we also compute this correction for different values of the soft–ultrasoft separation frequency. We will come back to this point in Section III. For the moment, let us leave this separation implicit. In what follows we will use Latin indices (i,j,k) to denote spatial coordinates, i.e. sµ = (0,si) with i = 1,2,3. At zeroth order in ǫ the conservation equations for the ultrasoft modes read 4 4 δD δρ + δρ si ρ V s ∂0si η sijs = 0 0 0 0 (0)i 0 0 i ij h i 3 ∇ − 3 − (cid:10) (cid:11) (cid:10) (cid:11) (cid:10) (cid:11) (5) 4 4 2 ρ δD sµ + δρ D sµ + V(µsφ)∂ δρ η Vµsi∂js = 0 0 0 0 0 φ 0 ij 3 h i 3 h i 3 − (cid:10) (cid:11) (cid:10) (cid:11) where is a thermal average. The zeroth order linearized equations for the soft modes are h·i 4 D δρ + ρ si = f 0 0 0 (0)i 3 ∇ (6) 4 1 ρ D si i δρ +η∆i ∂ sjk = fi . 3 0 0 − 3∇(0) 0 (0)j k 6 Note that the noises are determined by the equilibrium state, as dictated by the fluctuation– dissipation theorem. Their role is to enforce thermal equilibrium at zeroth order in ǫ, constraining the expression for the thermal correlator of velocity fluctuations which will be given in Section III. At first order we get for the ultrasoft modes 4 4 δD δρ + δD δρ + ρ Wi + δρ si 0 1 1 0 0 (0)i 1 (0)i h i h i 3 ∇ 3 ∇ (cid:10) (cid:11) 8 4 4 ρ V s ∂φtµ + δρ tµ δρ V s ∂0Wi (7) 0 (µ φ) 0 (0)µ 0 0 i − 3 3 ∇ − 3 (cid:10) (cid:11) (cid:10) (cid:11) (cid:10) (cid:11) 4 8 4 δρ V W ∂0si δρ W s ∂isj ρ V t ∂0si 2η s tij = 0 0 0 i 0 (i j) 0 0 i ij − 3 − 3 − 3 − (cid:10) (cid:11) (cid:10) (cid:11) (cid:10) (cid:11) (cid:10) (cid:11) and 4 4 4 ρ D Wµ + ρ δD tµ + ρ δD sµ 0 0 0 0 0 1 3 3 h i 3 h i 4 4 4 + δρ D tµ + δρ δD Wµ + δρ D sµ 0 0 0 0 0 1 3 h i 3 h i 3 h i (8) 4 2 2 + δρ D sµ + V(µsφ)∂ δρ + W(µsφ)∂ δρ 1 0 φ 1 φ 0 3 h i 3 3 (cid:10) (cid:11) (cid:10) (cid:11) 2 + V(µtφ)∂ δρ +ηCµ = 0 φ 0 3 (cid:10) (cid:11) with Cµ = ∆µ ∂ Wαβ 2 ∆µ ∂ (W<(αsφ)∂ sβ>) 2g V(µsγ)∂ tαβ (0)α β − (0)α β φ − γα β D E (9) (cid:10) (cid:11) 2g W(µsγ)∂ sαβ 2g V(µtγ)∂ sαβ . γα β γα β − − (cid:10) (cid:11) (cid:10) (cid:11) For the soft modes we get 4 D δρ +D δρ ρ V W ∂0si 0 1 1 0 0 0 i − 3 (10) 4 4 + ρ tµ + δρ Wi 2ηs Wij = 0 0 (0)µ 0 (0)i ij 3 ∇ 3 ∇ − and 4 4 1 2 ρ D tµ + δρ D Wµ µ δρ + V(µWφ)∂ δρ +ηEµ = 0 (11) 3 0 0 3 0 0 − 3∇(0) 1 3 φ 0 with Eµ = ∆µ ∂ tαβ 2g V(µsγ)∂ Wαβ 2g V(µWγ)∂ sαβ . (12) (0)α β − γα β − γα β In deriving Eqs. (5)–(12) we have used that µ = µ 2V(µsφ)∂ 2ǫV(µWφ)∂ 2ǫW(µsφ)∂ 2ǫV(µtφ)∂ (13) ∇ ∇(0) − φ − φ − φ − φ 7 where µ = (gµν VµVν)∂ and A(µν) = (Aµν +Aνµ)/2, ∇(0) − ν D = D +δD +ǫD +ǫδD (14) 0 0 1 1 where D = Vµ∂ , δD = sµ∂ , D = Wµ∂ and δD = tµ∂ , and put 0 µ 0 µ 1 µ 1 µ sµν <µsν> ≡ ∇(0) tµν <µtν> (15) ≡ ∇(0) Wµν <µWν> ≡ ∇(0) which by definition are traceless and orthogonal to V . Note that we have neglected terms µ O(δ3) or higher in the thermal averages. Moreover, we have set ρ = 0, which is a reasonable 1 approximation since, as it will be seen later, the term we are interested in is already linear in ǫ. Note that, since we stay at first order and O(δ3), those terms containing V<µ that would appear in Eqs. (5)–(12) actually vanish because in these terms we can put ∆µν = ∆µν (see (0) Eq. (13)). B. Constraints on four velocity We will now discuss the normalization of the fluid velocity. For consistency, the velocity should be normalized both in the mean and at linear order in fluctuations, i.e. uµu = 1, µ because it is uµ who satisfies the conservation equations (1). We also require that (Vµ + sµ)2 = 1 which implies VµV + sµs = 1 and µ µ h i (16) Vµs = 0 µ so at first order we get VµW = 0 and µ (17) Vµt +Wµs = 0 . µ µ The transversality and tracelessness of σµν must be satisfied both in the mean and at linear orderinfluctuations. Theconstraints coming fromthetransversality andtracelessness 8 of σµν then read sµs = 0 µν h i V(µsφ)∂ s = 0 φ µ (cid:10) (cid:11) tµs = 0 µν h i (18) V(µsφ)∂ t + W(µsφ)∂ s + V(µtφ)∂ s = 0 φ µ φ µ φ µ (cid:10) (cid:11) (cid:10) (cid:11) (cid:10) (cid:11) sµW +Wµs = 0 µν µν V(µsφ)∂ W +V(µWφ)∂ s = 0 φ µ φ µ These relations were used in deriving Eqs. (5)–(12) and will be used in what follows. III. SHEAR VISCOSITY INDUCED BY THE SOFT MODES The idea is now to solve the soft mode equations to compute the induced viscosity arising in the ultrasoft equations. We will solve the soft mode equations in Fourier space. For any given quantity R(xµ) we have R˜(kµ) = F[R(x)](kµ) = R(xµ)eikµxµ d4x (19) Z where F[g(x)](k ) g˜(k ) denotes the Fourier transform. We split the wavevector kµ = µ µ ≡ ωnµ + pµ, where nµ = Vµ/ V2 . Since Vµ = (V0,0) in the LRF and Vµs = 0 we have µ | | sµ = (0,~s), and similarly for Wµ (but not for tµ). Note that pµ = (0,pi) in the LRF. The zeroth order equations then read 4 ωδρ˜ + ρ p s˜j = if˜ and 0 0 j 3 (20) 4 1 iη 1 ρ ωs˜i piδρ˜ pi(p s˜j)+s˜ipjp = if˜i, i = 1,2,3 0 0 j j 3 − 3 − 2 (cid:20)3 (cid:21) Without loss of generality we can set p = p = 0. The solution is (p p ) 2 3 1 ≡ iηf˜p2 +2f˜ρ ω +2f˜ρ p 0 1 0 δρ˜ = 3i 0 A ˜ ˜ 3f ω +fp 1 s˜ = 3i 1 2A (21) ˜ 6f 2 s˜ = 2 B ˜ 6f 3 s˜ = 3 B 9 where 1 A = 3iηωp2+6ρ (ω2 p2) and 0 − 3 (22) B = 3ηp2 8iρ ω 0 − The first order soft mode equations are 4 ωδρ˜ + ρ p t˜j = iG[Wσ] 1 0 j 3 − 1 4 iη 1 piδρ˜ + ρ ωt˜i pi(p t˜j)+t˜ipjp = iHi[Wσ] (23) 1 0 j j −3 3 − 2 (cid:20)3 (cid:21) − 4 ρ ωt˜0 = iH0[Wσ] 0 3 where 8 G[Wµ] = eikσxσ Wµ∂ δρ ρ V W ∂φsµ µ 0 0 (µ φ) Z (cid:20) − 3 4 + δρ Wµ 2ηs Wµν d4x 0 (0)µ µν 3 ∇ − (cid:21) (24) 4 2 Hµ[Wµ] = eikσxσ δρ D Wµ + V(µWφ)∂ δρ 0 0 φ 0 Z (cid:20)3 3 2ηg V(µsγ)∂ Wαβ 2ηg V(µWγ)∂ sαβ d4x γα β γα β − − (cid:21) whose solution is iηGp2 +2Gρ ω +2H ρ p 0 1 0 δρ˜ = 3i 1 A 3H ω +Gp t˜ = 3i 1 1 2A 6H t˜ = 2 (25) 2 B 6H t˜ = 3 3 B 3 t˜ = i H 0 0 4ρ ω 0 We are interested in those terms appearing in Eq. (11) containing two (orthogonal to Vµ) derivatives of Wγ. Moreover, since the bare term ∆µ ∂ Wαβ is orthogonal to V , the (0)α β µ induced terms that provide the correction to η must be orthogonal to V as well. The only µ term in Hµ proportional to ∂ Wαβ is β 2ηg V(µ eikσxσsγ)∂ Wαβ d4x = ηVµ eikσxσs ∂ Wαβ d4x (26) γα β α β − Z − Z 10