Relativistic viscous hydrodynamics for heavy-ion collisions: A comparison between

the Chapman-Enskog and Grad methods

Rajeev S. Bhalerao, Amaresh Jaiswal, Subrata Pal, and V. Sreekanth
Tata Institute of Fundamental Research, Homi Bhabha Road, Mumbai 400005, India
(Dated: October 30, 2018)
Derivations of relativistic second-order dissipative hydrodynamic equations have relied almost
exclusively on the use of Grad’s 14-moment approximation to write f (x, p), the nonequilibrium
distribution function in the phase space. Here we consider an alternative Chapman-Enskog-like
method, which, unlike Grad’s, involves a small expansion parameter. We derive an expression for
f (x, p) to second order in this parameter. We show
√ analytically that while Grad’s method leads to
the violation of the experimentally observed 1/ mT scaling of the longitudinal femtoscopic radii,
arXiv:1312.1864v2 [nucl-th] 30 May 2014

the alternative method does not exhibit such an unphysical behavior. We compare numerical re-
sults for hadron transverse-momentum spectra and femtoscopic radii obtained in these two methods,
within the one-dimensional scaling expansion scenario. Moreover, we demonstrate a rapid conver-
gence of the Chapman-Enskog-like expansion up to second order. This leads to an expression for
δf (x, p) which provides a better alternative to Grad’s approximation for hydrodynamic modeling of
relativistic heavy-ion collisions.

PACS numbers: 25.75.-q, 24.10.Nz, 47.75+f

I. INTRODUCTION approximated by means of a Taylor-like series expansion

The standard model of relativistic heavy-ion colli-
sions relies on relativistic hydrodynamics to simulate the
intermediate-stage evolution of the high-energy-density
fireball formed in these collisions [1]. Recent simula-
tions generally make use of some version of the Müller-
Israel-Stewart second-order theory of causal dissipative
hydrodynamics [2, 3]. Hydrodynamics has achieved re-
markable success in explaining, for example, the observed
mass ordering of the elliptic flow [4–6], higher harmon-
ics of the azimuthal anisotropic flow [7, 8], and the ridge
and shoulder structure in long-range rapidity correlations
[9]. The recently measured correlators between event
planes of different harmonics [10] too can be understood
qualitatively within event-by-event hydrodynamics [11].
Notwithstanding these successes, the basic formulation
of the dissipative hydrodynamic equations continues to
be an area of considerable activity, largely because of the
ambiguities arising due to the variety of ways in which
these equations can be derived [12–18].
For a system that is out of equilibrium, the existence
of thermodynamic gradients results in thermodynamic
forces, which give rise to various transport phenomena.
To quantify these nonequilibrium effects, it is convenient
to first specify the nonequilibrium phase-space distribu-
tion function f (x, p) and then calculate the various trans-
port coefficients. In the context of hydrodynamics, two
most commonly used methods to determine the form of
the distribution function close to local thermodynamic
equilibrium are (1) Grad’s 14-moment approximation [19]
and (2) the Chapman-Enskog method [20]. Although
both the methods involve expanding f (x, p) around the
equilibrium distribution function f0 (x, p), there are im-
portant differences.
In the relativistic version of Grad’s 14-moment approx-
imation, the small deviation from equilibrium is usually
in momenta truncated at quadratic order [2, 17]. Fur-
ther, the 14 coefficients in this expansion are assumed to
be linear in dissipative fluxes. However, it is not appar-
ent why a power series in momenta should be convergent
and whether one is justified in making such an ansatz,
without a small expansion parameter.
The Chapman-Enskog method, on the other hand,
aims at obtaining a perturbative solution of the Boltz-
mann transport equation using the Knudsen number (ra-
tio of mean free path to a typical macroscopic length)
as a small expansion parameter. This is equivalent to
making a gradient expansion about the local equilibrium
distribution function [21]. This method of obtaining the
form of the nonequilibrium distribution function is con-
sistent [16] with dissipative hydrodynamics, which is also
formulated as a gradient expansion.
The above two methods have been compared and
shortcomings of Grad’s approximation have been pointed
out in the literature [22–24]. In spite of these short-
comings, the derivations of relativistic second-order dis-
sipative hydrodynamic equations, as well as particle-
production prescriptions, rely almost exclusively on
Grad’s approximation. The Chapman-Enskog method,
on the other hand, has seldom been employed in the hy-
drodynamic modeling of the relativistic heavy-ion colli-
sions. The focus of the present work is to explore the
applicability of the latter method.
In this paper, the Boltzmann equation in the
relaxation-time approximation is solved iteratively, which
results in a Chapman-Enskog-like expansion of the
nonequilibrium distribution function. Truncating the ex-
pansion at the second order, we derive an explicit ex-
pression for the viscous correction to the equilibrium
distribution function. We compare the hadronic spec-
tra and longitudinal Hanbury-Brown-Twiss (HBT) radii
obtained using the form of the viscous correction derived

here and Grad’s 14-moment approximation, within a one-
dimensional scaling expansion. We find that at large
transverse momenta, the present method yields smaller
hadron multiplicities. We also show analytically that
while Grad’s approximation leads to the violation of the
√ ∇α β = −β u̇α −
experimentally observed 1/ mT scaling of HBT radii
[25–29], the viscous correction obtained here does not ex-
hibit such unphysical behavior. Finally, we demonstrate
the rapid convergence of the Chapman-Enskog-like ex-
pansion up to second order.
II. RELATIVISTIC VISCOUS
HYDRODYNAMICS
Within the framework of relativistic hydrodynamics,
the variables that characterize the macroscopic state of
a system are the energy-momentum tensor, T µν , parti-
cle four-current, N µ , and entropy four-current, S µ . The
local conservation of net charge (∂µ N µ = 0) and energy-
momentum (∂µ T µν = 0) lead to the equations of motion
of a relativistic fluid, whereas the second law of thermo-
dynamics requires ∂µ S µ ≥ 0. For a system with no net
conserved charges, hydrodynamic evolution is governed
only by the conservation equations for energy and mo-
mentum.
The energy-momentum tensor of a macroscopic system
can be expressed in terms of a single-particle phase-space
distribution function and can be tensor decomposed into
hydrodynamic degrees of freedom [21]. Here we restrict
ourselves to a system of massless particles (ultrarelativis-
tic limit) for which the bulk viscosity vanishes, leading
to
Z
T µν = dp pµ pν f (x, p) = ǫuµ uν − P ∆µν + π µν . (1)
Here dp ≡ gdp/[(2π)3 |p|], where g is the degeneracy fac-
tor, pµ is the particle four-momentum, and f (x, p) is the
phase-space distribution function. In the tensor decom-
position, ǫ, P , and π µν are energy density, thermody-
namic pressure, and shear stress tensor, respectively. The
projection operator ∆µν ≡ g µν − uµ uν is orthogonal to
the hydrodynamic four-velocity uµ defined in the Landau
frame: T µν uν = ǫuµ . The metric tensor is Minkowskian,
g µν ≡ diag(+, −, −, −).
The evolution equations for ǫ and uµ ,
ǫ̇ + (ǫ + P )θ − π µν ∇(µ uν) = 0,
(ǫ + P )u̇α − ∇α P + ∆α µν
ν ∂µ π = 0, (2)
are obtained from the conservation of the energy-
momentum tensor. We use the standard notation Ȧ ≡
uµ ∂µ A for comoving derivative, θ ≡ ∂µ uµ for expansion
scalar, A(α B β) ≡ (Aα B β + Aβ B α )/2 for symmetrization,
and ∇α ≡ ∆µα ∂µ for spacelike derivatives. In the ultra-
relativistic limit, the equation of state relating energy
density and pressure is ǫ = 3P ∝ β −4 . The inverse tem-
perature, β ≡ 1/T , is determined by the Landau match-
ing condition ǫ = ǫ0 where ǫ0 is the equilibrium energy
2
density. In this limit, the derivatives of β,
β β ργ
β̇ = θ− π σργ , (3)
3 12P
β α
∆ ∂γ π ργ , (4)
4P ρ
can be obtained from Eq. (2), where σ ργ ≡ ∇(ρ uγ) −
(θ/3)∆ργ is the velocity stress tensor [30]. The above
identities are used later in the derivations of viscous cor-
rections to the distribution function and shear evolution
equation.
For a system close to local thermodynamic equilibrium,
the phase-space distribution function can be written as
f = f0 + δf , where the deviation from equilibrium is as-
sumed to be small (δf ≪ f ). Here f0 represents the equi-
librium distribution function of massless Boltzmann par-
ticles at vanishing chemical potential, f0 = exp(−β u · p),
where u · p ≡ uµ pµ . From Eq. (1), the shear stress ten-
sor, π µν , can be expressed in terms of the nonequilibrium
part of the distribution function, δf , as [17]
Z
µν µν
π = ∆αβ dp pα pβ δf, (5)
where ∆µν µ ν µν
αβ ≡ ∆(α ∆β) − (1/3)∆ ∆αβ is a traceless sym-
metric projection operator orthogonal to uµ . To make
further progress, the form of δf has to be determined. In
the following, we adopt a Chapman-Enskog-like expan-
sion for the distribution function, to obtain δf order-by-
order in gradients, by solving the Boltzmann equation
iteratively in the relaxation-time approximation.
III. CHAPMAN-ENSKOG EXPANSION
Determination of the nonequilibrium phase-space dis-
tribution function is one of the central problems in sta-
tistical mechanics. This can be achieved by solving a
kinetic equation such as the Boltzmann equation. The
relativistic Boltzmann equation with the relaxation-time
approximation for the collision term is given by [31],
δf
pµ ∂µ f = C[f ] = − (u·p) , (6)
τR
where τR is the relaxation time. We recall that the ze-
roth and first moments of the collision term, C[f ], should
vanish to ensure the conservation of particle current and
energy-momentum tensor [21]. This requires that τR is
independent of momenta, and uµ is defined in the Lan-
dau frame [31]. Therefore, within the relaxation-time
approximation, Landau frame is mandatory and not a
choice. Momentum-dependent τR was considered in Ref.
[32] where the authors also studied the consequences of
different momentum dependencies of δf for the heavy-ion
observables.
Exact solutions of the Boltzmann equation are possible
only in rare circumstances. The most common technique
 3

of generating an approximate solution to the Boltzmann V. CORRECTIONS TO THE DISTRIBUTION

equation is the Chapman-Enskog expansion, where the FUNCTION
distribution function is expanded about its equilibrium
value in powers of space-time gradients [20] In this section, we derive the expression for the
nonequilibrium part of the distribution function, δf , up
f = f0 + δf, δf = δf (1) + δf (2) + · · · , (7) to second order in gradients of uµ . For this purpose, we
where δf (n) is nth-order in derivatives. The Boltzmann employ Eqs. (9) and (10), which were obtained using
equation can be solved iteratively by rewriting Eq. (6) a Chapman-Enskog-like expansion. We then recall the
in the form f = f0 − (τR /u · p) pµ ∂µ f [16, 33, 34]. We derivation of the standard Grad’s 14-moment approxi-
obtain mation for δf , and compare these two expressions.
Using Eqs. (3) and (4) for the derivatives of β, and Eq.
τR µ τR µ
f1 = f0 − p ∂µ f0 , f2 = f0 − p ∂µ f1 , · · · (8) (13) for σ µν , in Eqs. (9) and (10), we arrive at the form
u·p u·p of the second-order viscous correction to the distribution
function:
where fn = f0 + δf (1) + δf (2) + · · · + δf (n) . To first- and
second-orders in derivatives, we have

f0 β α β f0 β τπ α β γ
δf = p p παβ − p p πα ωβγ
τR µ 2βπ (u·p) βπ u·p
δf (1) = − p ∂µ f0 , (9) 5 τπ
u·p − pα pβ παγ πβγ + pα pβ παβ θ
τR µ ν  τR 
14βπ (u·p) 3(u·p)
δf (2) = p p ∂µ ∂ν f0 . (10)
u·p u·p 6τπ α β (u·p) αβ τπ α
p ∇β παβ


− p u̇ παβ + π παβ +
In the next section, the above expressions for δf along 5 70βπ 5
with Eq. (5) are used in the derivation of the evolution 3τπ α β γ τπ
− p p p παβ u̇γ + pα pβ pγ (∇γ παβ )
equation for the shear stress tensor. (u·p)2 2(u·p)2
β + (u·p)−1 α β


2
− p p παβ + O(δ 3 ), (14)
4(u·p)2 βπ
IV. VISCOUS EVOLUTION EQUATION
≡ δf1 + δf2 + O(δ 3 ). (15)
In order to complete the set of hydrodynamic equa- The first term on the right-hand side of Eq. (14) cor-
tions, Eq. (2), we need to derive an expression for the responds to the first-order correction, δf1 , whereas the
shear stress tensor, π µν . The first-order expression for terms within square brackets are of second order, δf2 (see
π µν can be obtained from Eq. (5) using δf = δf (1) from Appendix A). Note that δf1 6= δf (1) and δf2 6= δf (2) , due
Eq. (9), to the nonlinear nature of Eqs. (3), (4), and (13). It is
Z 
τR γ
 straightforward to show that the form of δf in Eq. (14)
µν µν α β
π = ∆αβ dp p p − p ∂γ f0 . (11) is consistent with the definition of the shear stress tensor,
u·p
Eq. (5), and satisfies the matching condition ǫ = ǫ0 and
Using Eqs. (3) and (4) and keeping only those terms the Landau frame definition uν T µν = ǫuµ [21], i.e.,
which are first-order in gradients, the integral in the Z Z
above equation reduces to dp (u · p)2 δf = 0, dp ∆µα uβ pα pβ δf = 0, (16)
π µν = 2τR βπ σ µν , (12)
order-by-order in gradients (see Appendix A).
where βπ = 4P/5 [16]. On the other hand, Grad’s 14-moment approximation
The second-order evolution equation for shear stress for δf can be obtained from a Taylor-like expansion in
tensor can also be obtained in a similar way by using the powers of momenta [2, 17]
δf = δf (1) + δf (2) from Eqs. (9) and (10) in Eq. (5).
δfG = f0 ε(x) + εα (x)pα + εαβ (x)pα pβ ,
 
(17)
Performing the integrations and using Eqs. (3), (4) and
(12), we get [16, 30] where ε’s are the momentum-independent coefficients in
π µν
10 4 the expansion, which, however, may depend on ther-
π̇ hµνi + = 2βπ σ µν + 2πγhµ ω νiγ − πγhµ σ νiγ − π µν θ, modynamic and dissipative quantities. For a system of
τR 7 3 massless particles with no net conserved charges, i.e., in
(13)
the absence of bulk viscosity and charge diffusion current,
where ω µν ≡ (∇µ uν − ∇ν uµ )/2 is the vorticity tensor,
the above equation reduces to
and we have used Eq. (12). It is clear from the form
of the above equation that the relaxation time τR can f0 β 2 α β
be identified with the shear relaxation time τπ . By com- δfG = p p παβ , (18)
10βπ
paring the first-order evolution Eq. (12) with the rela-
tivistic Navier-Stokes equation π µν = 2ησ µν , we obtain where the coefficient is obtained using Eq. (5). We ob-
τπ = η/βπ , where η is the coefficient of shear viscosity. serve that unlike Eq. (14) for the Chapman-Enskog case,
 4

Eq. (18) for Grad’s is linear in shear stress tensor. How-
ever, it is important to note that both the forms of δf ,
i.e., δf1 and δfG , lead to identical evolution equations
for the shear stress tensor, Eq. (13), with the same coef-
ficients [13, 30].
the nonequilibrium distribution function [36]. Hydrody-
namic evolution and the nonequilibrium corrections to
the distribution function were considered in the previ-
ous sections; in the following sections, we focus on two
observables, namely transverse-momentum spectra and
HBT radii of hadrons.

VI. BJORKEN SCENARIO

VII. HADRONIC SPECTRA
In order to model the hydrodynamical evolution of the
matter formed in the heavy-ion collision experiments, we The hadron spectra can be obtained using the Cooper-
use the Bjorken prescription [35] for one-dimensional ex- Frye freezeout prescription [37]
pansion. We consider the evolution of a system of mass-
dN g
Z
less particles (ǫ = 3P ) at vanishing net baryon number = pµ dΣµ f (x, p), (24)
density. In √ terms of the Milne d2 pT dy (2π)3
p coordinates (τ, r, ϕ, ηs ),
where τ = t2 − z 2 , r = x2 + y 2 , ϕ = tan−1 (y/x), where pµ is the particle four-momentum, dΣµ represents
and ηs = tanh−1 (z/t), and with uµ = (1, 0, 0, 0), evolu- the element of the three-dimensional freezeout hypersur-
tion equations for ǫ and Φ ≡ −τ 2 π ηs ηs become face, and f (x, p) represents the phase-space distribution
function at freezeout.
dǫ 1 For the ideal freezeout case (f = f0 ), we get
= − (ǫ + P − Φ) , (19)
dτ τ
dΦ Φ 4 Φ dN (0) g
= − + βπ −λ . (20) 2
= mT τ A⊥ K1 , (25)
dτ τπ 3τ τ d pT dy 4π 3

The transport coefficients appearing in the above equa- where A⊥ denotes the transverse area of the overlap
tion reduce to [16] zone of colliding nuclei and Kn ≡ Kn (zm ) are the mod-
ified Bessel functions of the second kind with argument
η 4P 38 zm ≡ mT /T . In Eq. (25) and hereafter, the hydrody-
τπ = , βπ = , λ= . (21) namical quantities such as T, τ, Φ, P , etc., correspond
βπ 5 21
to their values at freezeout. The expression for hadron
In (τ, r, ϕ, ηs ) coordinates, the components of particle production up to first order (f = f0 + δf1 ) is obtained as
four-momenta are given by
dN (1) dN (0)
  
Φ 2 K0
= 1 + z − 2z m , (26)
pτ = mT cosh(y − ηs ), pr = pT cos(ϕp − ϕ), (22) 2
d pT dy 4βπ zm p
K1 d2 pT dy
ϕ ηs
p = pT sin(ϕp − ϕ)/r, p = mT sinh(y − ηs )/τ,
where zp ≡ pT /T . Here we have used the recurrence rela-
tion Kn+1 (z) = 2nKn (z)/z + Kn−1(z). The derivation of
where m2T = p2T 2
+ m , pT is the transverse momen-
the hadron spectra up to second order, dN (2) /d2 pT dy (by
tum, y is the particle rapidity, and ϕp is the azimuthal
setting f = f0 + δf1 + δf2 ), is presented in the Appendix
angle in the momentum space. We note that for the
B.
Bjorken expansion, θ = 1/τ , u̇µ = 0, ω µν = 0 and
For comparison, we also present the result for hadron
pµ dΣµ = mT cosh(y − ηs )τ dηs rdrdϕ. In this scenario,
production obtained using Grad’s 14-moment approxi-
the nonvanishing factors appearing in Eq. (14) reduce to
mation (f = f0 + δfG ) [36, 38]
u · p = mT cosh(y − ηs ), παβ π αβ = 3Φ2 /2, and
dN (G) dN (0)
  
Φ 2 K2
Φ 2 = 1 + z − 2z m . (27)
pα pβ παβ = p − Φ m2T sinh2 (y − ηs ), d2 pT dy 20βπ p
K1 d2 pT dy
2 T
Φ2 2 We solve the evolution equations (19) and (20) with
pα pβ παγ πγβ =− p − Φ2 m2T sinh2 (y − ηs ),
4 T initial temperature T0 = 360 MeV, time τ0 = 0.6 fm/c,
Φ and isotropic pressure configuration Φ0 = 0, correspond-
pα pβ pγ ∇α πβγ = 2 m3T sinh2 (y − ηs ) cosh(y − ηs ), ing to central (b = 0) Au-Au collisions at the Relativistic
τ
Φ Heavy-Ion Collider. The system is evolved with shear vis-
pα ∇β παβ = − mT cosh(y − ηs ). (23) cosity to entropy density ratio η/s = 1/4π corresponding
τ
to the Kovtun-Son-Starinets (KSS) lower bound [39], un-
Within the framework of the relativistic hydrodynam- til the freezeout temperature T = 150 MeV is reached.
ics, observables pertaining to heavy-ion collisions are in- In order to study the effects of the various forms of δf via
fluenced by viscosity in two ways: first through the vis- the freezeout prescription, Eq. (24), we evolve the system
cous hydrodynamic evolution of the system and second using the second-order viscous hydrodynamic equations
through corrections to the particle production rate via (19) and (20) in all the cases.
 5

persurface and is given by

10
2 Au+Au: T0 = 360 MeV
g
Z
τ0 = 0.6 fm/c S(x, K) = pµ dΣµ (x′ )f (x′ , p)δ 4 (x − x′ ). (29)
10
1
(2π)3
dN/(d pT dy) [c /GeV ]
2

Freezeout conditions
10
0 Ideal At relatively small momenta, certain space-time vari-
Pions Grad’s approx. ances of the source function can be obtained, to a good
2

-1 First-order CE approximation, from the correlation between particle

10 Second-order CE
pairs [41]. Space-time averages with respect to the source
2.0
-2
10 function are defined as
2

1.6 R 4
Kµ dΣµ f (x, K)α
R
d x S(x, K)α
Ratio

-3
10 hαiK ≡ R 4 = R , (30)
1.2 d x S(x, K) Kµ dΣµ f (x, K)
-4 0.8
10 0 1 2 3 where Kµ is the pair four-momentum.
pT (GeV/c)
-5 The longitudinal HBT radius, RL , is calculated in
10 0 0.5 1 1.5 2 2.5 3 terms of the transverse momentum, KT , of the identical-
pT (GeV/c) particle pair [41]:
Kµ dΣµ f (x, K)z 2
R
FIG. 1: (Color online) Pion spectra as a function of the trans- 2
verse momentum pT , obtained with the second-order hydro- RL (KT ) = R . (31)
Kµ dΣµ f (x, K)
dynamic evolution, followed by freezeout in various scenar-
ios: ideal, Grad’s 14-moment approximation, and first- and In the central-rapidity region, the pair four-momentum
second-order Chapman-Enskog. Inset: Pion yields in the is given by K µ = (K τ , K r , K ϕ , K ηs ) = (mT , KT , 0, 0).
above four cases scaled by the corresponding values in the The integration measure is given µ
ideal case. q by Kµ dΣ =
mT cosh(ηs )τ dηs rdrdϕ with mT = KT2 + m2p , mp be-
ing the particle mass. Using the relation z = τ sinh(ηs ),
In Fig. 1, we present the pion transverse-momentum we get
spectra for the four freezeout conditions discussed above, "R #
namely ideal, first- and second-order Chapman-Enskog, 2 2 Kµ dΣµ f (x, K)cosh2 (ηs )
and Grad’s 14-moment approximation. We observe that RL (KT ) = τ R −1 ,
Kµ dΣµ f (x, K)
nonideal freezeout conditions tend to increase the high-  
pT particle production. While the Chapman-Enskog cor- 2 N [f ]
≡τ −1 . (32)
rections are small, Grad’s 14-moment approximation re- D[f ]
sults in rather large corrections to the ideal case. This
is clearly evident in the inset where we show the pion Note that the integral, D[f ], in the denominator in the
yields in the four cases scaled by the values in the ideal above equation is the same as that occurring in the
case. These features can be easily understood from Eqs. Cooper-Frye prescription for particle production, Eq.
(26) and (27): The first-order Chapman-Enskog correc- (24), and was already calculated in the previous section.
tion is essentially linear in pT whereas that due to Grad We next calculate the integral, N [f ], in the numerator.
is quadratic. The second-order Chapman-Enskog correc- In the ideal case, f = f0 , we have
tion is small, indicating rapid convergence of the expan- 2A⊥ τ zm
sion up to second order. N [f0 ] = (K3 + 3K1 ) . (33)
4β
This leads to the well-known result of Hermann and
VIII. HBT RADII Bertsch [42]

2 (0) τ 2 K2
HBT interferometry provides a powerful tool to un- (RL ) = , (34)
zm K 1
ravel the space-time structure of the particle-emitting
sources in heavy-ion collisions, because of its ability to which for large values of zm results in the Makhlin-
2 (0)
measure source sizes, lifetimes, and particle emission du- Sinyukov formula (RL ) = τ 2 T /mT [43, 44]. Thus in
(0) √
rations [40]. The source function, S(x, K), for on-shell the ideal case, (RL ) exhibits the so-called 1/ mT scal-
particle emission is defined such that it satisfies ing.
The first-order calculation requires N [δf1 ], which is
dN
Z
given by
≡ d4 x S(x, K). (28)
d2 KT dy
2A⊥ τ Φ h i
2zp2 + zm
2
K0 + 2zp2 K2 − zm
2


N [δf1 ] = K4 .
By comparing the above equation with Eq. (24), we see 16ββπ
that the source function is restricted to the freezeout hy- (35)
 6

The second-order calculation requires N [δf2 ], which is T0 = 360 MeV, τ0 = 0.6 fm/c
given in the Appendix B. For comparison we also calcu- 12
late RL in Grad’s 14-moment approximation. This re- 1
quires N [δfG ], which we obtain as 10 0.9

RL/ RL(0)
0.8
2A⊥ τ Φzm h
2zp2 − 6zm2 0.7


N [δfG ] = K1 8
160ββπ

RL (fm)
0.6
i 0.5
+ 2zp2 − zm
2 2


K 3 − zm K5 . (36) 0.2 0.4 0.6 0.8 1
6 mT (GeV/c)
Pions
In the following, we show that the viscous correction to
RL due to Grad’s 14-moment approximation violates the 4 Freezeout conditions
√ Ideal
experimentally observed 1/ mT scaling [25–29], whereas
Grad’s approx.
it is preserved in the Chapman-Enskog case. To this 2 First-order CE
end, we calculate the first-order viscous correction to RL Second-order CE
in both the cases. Expanding the RL in Eq. (31) to 0
first order in δf and using the relation z = τ sinh(ηs ) we 0.2 0.4 0.6 0.8 1
obtain the ideal contribution KT (GeV/c)
K dΣµ f0 τ 2 sinh2 (ηs )
R µ
(RL2 (0)
) = R , (37) FIG. 2: (Color online) Longitudinal HBT radius as a function
K µ dΣµ f0 of the transverse momentum KT of the pion pair, obtained
with the second-order hydrodynamic evolution, followed by
and the first viscous correction in the two cases freezeout in various scenarios: ideal, Grad’s 14-moment ap-
dN (1,G) dN (0) . dN (0)
 
2 (1,G)

2 (0) proximation, and first- and second-order Chapman-Enskog.
δRL = − (RL ) − 2 Inset: HBT radius in the above cases scaled by the corre-
d2 KT d KT d2 KT
2
sponding values in the ideal case.
K dΣµ τ 2 sinh (ηs ) δf1,G
R µ
+ R . (38)
K µ dΣµ f0
2 (0)
It is clear from the above two equations that the viscous
The ideal radius (RL ) was obtained in Eq. (34). Vis- correction to RL in the Chapman-Enskog case preserves
cous corrections due to the Chapman-Enskog method √
the 1/ mT scaling, whereas in Grad’s 14-moment ap-
and Grad’s 14-moment approximation can be obtained proximation it grows as mT /T relative to the ideal result,
similarly. By substituting the viscous correction, δf1 , and thus violates the scaling [36].
from Eq. (14) into Eq. (38), using the results for the Results for the longitudinal HBT radius, RL , for
particle spectra, Eqs. (25) and (26), and the ideal radius, identical-pion pairs in central Au-Au collisions, for the
Eq. (34), and performing the ηs integrals, we obtain four cases discussed above, are displayed in Fig. 2. We
2 (1)

"
4zp2 K0
# note that while there is no noticeable difference between
δRL

Φ K1
=− 16 + − . (39) first- and second-order Chapman-Enskog results com-
(0) 16βπ zm K1 K2
(R2 )
L pared to the ideal case, they predict a slightly smaller
value for RL . On the other hand, RL corresponding
Similarly, for Grad’s approximation, Eq. (18), we obtain
to Grad’s approximation exhibits a qualitatively differ-
2 (G) ent behavior and even becomes imaginary for KT > ∼ 0.9


δRL
   
Φ K0 K1 K1
=− 20 − 2zm − + 4zm . GeV/c, which is clearly unphysical. More importantly,
(RL2 )(0) 20βπ K1 K2 K2 (0)
the ratio RL√/RL shown in the inset of Fig. 2 illustrates
(40)
that the 1/ mT scaling, which is violated in Grad’s ap-
Using the asymptotic expansion of modified Bessel
proximation, survives in the Chapman-Enskog case.
functions of the second kind [45],
 21
4n2 − 1
  
π
Kn (zm ) = e−zm 1 + + · · · , (41) IX. SUMMARY AND CONCLUSIONS
2zm 8zm
for large zm , we have We derived the form of the viscous correction to the
equilibrium distribution function, up to second order in
 
K0 K1 1 1
− = +O 2
. (42) gradients, by employing a Chapman-Enskog-like iterative
K1 K2 zm zm
solution of the Boltzmann equation in the relaxation-time
Hence, for large values of zm , we find approximation. This approach is in accordance with the

(1) 5τ 2 T Φ formulation of hydrodynamics, which is also a gradient
2
δRL =− , (43) expansion. We used this form of the viscous correction
4βπ mT
to calculate the hadronic transverse-momentum spectra
2

(G) τ 2T Φ  mT  and longitudinal Hanbury-Brown-Twiss radii and com-
δRL =− 3+ . (44)
5βπ mT T pared them with those obtained in Grad’s 14-moment
 7

approximation within the one-dimensional scaling expan- whereas the second-order correction is
sion. These results demonstrate the rapid convergence of 
the Chapman-Enskog expansion up to second order, and f0 β τπ α β γ 5
δf2 = − p p πα ωβγ − pα pβ παγ πβγ
thus it is sufficient to retain only the first-order correc- βπ u·p 14βπ (u·p)
tion in the freezeout prescription. We found that the τπ 6τπ α β (u·p) αβ
Chapman-Enskog method results in softer hadron spec- + pα pβ παβ θ − p u̇ παβ + π παβ
3(u·p) 5 70βπ
tra compared with Grad’s approximation. We further τπ 3τπ α β γ τπ
√
+ pα ∇β παβ −


showed that the experimentally observed 1/ mT scaling 2
p p p παβ u̇γ +
of HBT radii, which is also seen in the ideal freezeout cal- 5 (u·p) 2(u·p)2
β +(u·p)−1 α β

culation, is maintained in the Chapman-Enskog method.
2
×pα pβ pγ (∇γ παβ) − p p παβ .
In contrast, the Grad’s 14-moment approximation leads 4(u·p)2 βπ
to the violation of this scaling as well as an imaginary (A2)
value for RL at large momenta. For initial conditions
typical of heavy-ion collisions at the Large Hadron Col- In the following, we show that the δfi given in Eqs.
lider (T0 = 500 MeV and τ0 = 0.4 fm/c), we have found (A1) and (A2) satisfies the conditions
that the above conclusions remain unchanged. Z
We conclude by recalling the well-known form of the L1 [δfi ] ≡ dp (u · p)2 δfi = 0, (A3)
viscous correction due to Grad’s 14-moment approxima-
tion,
corresponding to ǫ = ǫ0 , and

f0 f˜0
Z
δfG = pα pβ παβ , (45) L2 [δfi ] ≡ dp ∆µα uβ pα pβ δfi = 0, (A4)
2(ǫ + P )T 2
corresponding to uν T µν = ǫuµ .
and the alternate form due to Chapman-Enskog method At first order, we obtain
proposed here,
β αβγ β αβγ
L1 [δf1 ] = παβ uγ I(0) , L2 [δf1 ] = παβ ∆µγ I(0) ,
5f0 f˜0 2βπ 2βπ
δfCE = pα pβ παβ , (46) (A5)
8P T (u·p) where we define the integral

dp µ1 µ2
Z
where f˜0 ≡ 1 − rf0 , with r = 1, −1, 0 for Fermi, Bose, µ1 µ2 ···µn
I(r) ≡ p p · · · pµn f0 . (A6)
and Boltzmann gases, respectively. In view of the ar- (u·p)r
guments presented in this paper, we advocate that the
form of δfCE proposed here should be a better alterna- The above momentum integral can be decomposed into
tive for hydrodynamic modeling of relativistic heavy-ion hydrodynamic tensor degrees of freedom as
collisions. µ1 µ2 ···µn (r) (r)
I(r) = In0 uµ1 uµ2 · · · uµn + In1 ∆µ1 µ2 uµ3 · · · uµn


+ perms + · · · , (A7)

(0) (0)
Appendix A: CONSTRAINTS ON THE VISCOUS where we readily identify I20 = ǫ and I21 = −P . Using
CORRECTION TO THE DISTRIBUTION αβγ
the above tensor decomposition for I(0) in Eq. (A5), we
FUNCTION
obtain

In this appendix, we show that the form of the vis- L1 [δf1 ] = 0, L2 [δf1 ] = 0. (A8)
cous correction to the distribution function, δf , given in
Eq. (14) satisfies the matching condition ǫ = ǫ0 and the Similarly, for second-order corrections given in Eq.
Landau frame definition uν T µν = ǫuµ , at each order in (A2), we obtain
gradients [21]. We also show that δf is consistent with
the definition of the shear stress tensor, Eq. (5). 5β (0) β (0)
L1 [δf2 ] = 0 + 2
παβ π αβI31 + 0 + 0 − παβ π αβI30
14βπ 70βπ2
The first- and second-order viscous corrections to the
distribution function can be written separately using Eq. βτπ α (0) βτπ (0)
− (∇ παβ )I30 uβ + 0 − (∇γ παβ )I31
(14). The first-order correction is given by 5βπ βπ
β 
(0) (1)

×u(α ∆β)γ + 2 παβ π αβ βI42 + I42 .
f0 β 2βπ
δf1 = pα pβ παβ , (A1) (A9)
2βπ (u·p)
 8

Using the identities Hence L3 [δf ] = L3 [δf1 ] + L3 [δf2 ] = π µν . This result

was expected because no second-order term (e.g., ππ,
(r) 1 (r−1)
Inq =− I , (A10) πω, etc.) or their linear combinations, when substituted
2q + 1 n−1,q−1 in Eq. (5), can result in a first-order term (π) which
(0) 1 h (0) (0)
i
we have on the left-hand side of Eq. (5). In fact, each
Inq = −In−1,q−1 + (n − 2q)In−1,q , (A11)
β higher-order correction (δfn ) when substituted in Eq. (5)
will vanish. The fact that δf given in Eq. (14) satisfies
and Eq. (12), we obtain the constraints, as demonstrated in this Appendix, shows
25 3 12 that our method of obtaining the viscous corrections to
L1 [δf2 ] = − παβ π αβ − παβ π αβ + παβ π αβ the distribution function is quite robust.
14βπ 14βπ 8βπ
5 3
− παβ π αβ + παβ π αβ
2βπ βπ Appendix B: SECOND-ORDER VISCOUS
= 0. (A12) CORRECTIONS TO HADRON SPECTRA AND
HBT RADII
A similar calculation leads to
6βτπ (0) α β Within the one-dimensional scaling expansion, u̇ =
L2 [δf2 ] = 0 + 0 + 0 + I ∆ u̇ παβ + 0 0 = ω µν , which reduces the number of terms in Eq. (A2).
5βπ 31 µ
βτπ (0) α β
6βτπ (0) α β The nonvanishing terms can be simplified using Eq. (23)
− I31 ∆µ ∇ παβ − I ∆ u̇ παβ as
5βπ 5βπ 31 µ "
5Φ2 mT p2T /(4m2T ) + sinh2 (y − ηs )

βτπ (1) α β f0 β
δf2 = −


− I42 ∆µ ∇ παβ + 0
βπ βπ 14βπ cosh(y − ηs )
= 0. (A13) τπ Φ mT pT /(2m2T ) − sinh2 (y − ηs )
 2
−
To obtain the second equality, we have used Eq. (A10) 3τ cosh(y − ηs )
(1) (0) 2
to replace I42 = −I31 /5. 3Φ mT cosh(y − ηs ) τπ Φ mT cosh(y − ηs )
− +
Next we show that the form of the viscous correction 140βπ 5τ
to the distribution function, δf = δf1 + δf2 given in Eqs. 2
τπ Φ mT sinh (y − ηs ) Φ2 β
(A1) and (A2), is consistent with the definition of the − +
shear stress tensor given in Eq. (5). In other words, we
τ cosh(y − ηs ) 4βπ cosh2 (y − ηs )
2 #
show that π µν = L3 [δf1 ] + L3 [δf2 ], where (βmT )−1 p2T
 
× 1+ − sinh2 (y − ηs ) .
Z cosh(y − ηs ) 2m2T
L3 [δfi ] ≡ ∆µν
αβ dp pα pβ δfi . (A14) (B1)

At first order, we get The contribution to the hadronic spectra resulting

from these second-order terms is calculated using Eq.
β (24) as
L3 [δf1 ] = ∆µν πγδ I(1)
αβγδ
. (A15)
2βπ αβ
δdN (2) g
Z
αβγδ ≡ mT cosh(y − ηs )τ dηs rdrdϕ δf2
Using the tensor decomposition for I(1) in the above d2 pT dy (2π)3
equation, we obtain
"
g τ A⊥ −5Φ2 2

= 3
zp K0 + 4zm K1
β (1) µν 4π ββπ 56βπ
L3 [δf1 ] = I π = π µν . (A16)
βπ 42 Φτπ 2
3Φ2 zm2
− zp K0 − 2zm K1 − (K0 +K2 )
(1)
Here we have used I42 = βπ /β, obtained by employing 6τ 280βπ
2
the recursion relations, Eqs. (A10) and (A11). Φτπ zm Φτπ zm Φ2 z m
2
+ (K0 + K2 ) − K1 +
Similarly, for the second-order correction δf2 given in 10τ τ 4βπ
Eq. (A2), we obtain n
2 zm
× zm X I1 − 2zm XK1 + (K3 + 3K1 )
4
5 hµ νiγ 2
L3 [δf2 ] = − 2τπ πγhµ ω νiγ + − τπ π µν θ + 0
#
π π 1
7βπ γ
o
3 2
+ X I2 − 2XK0 + (K0 + K2 ) , (B2)
 1 2
hµ νiγ
+0+0+0+ π π + 2τπ πγhµ ω νiγ
βπ γ
where X ≡ zp2 /(2zm2
)+1, Kn (zm ) are the modified Bessel
2  12 hµ νiγ functions of the second kind
+ τπ π µν θ − π π
3 7βπ γ Z ∞
= 0. (A17) Kn (z) ≡ dt e−z cosh(t) cosh(nt), (B3)
0
 9

and In are the integrals defined as

Z ∞
In (z) ≡ dt e−z cosh(t) sechn (t), (B4)
0

with the following properties:

dn In (z)
= (−1)n K0 (z), I0 (z) = K0 (z). (B5)
dz n
The expression for hadron spectra up to second order, by
setting f = f0 + δf1 + δf2 in the freezeout prescription,
Eq. (24), becomes

dN (2) dN (1) δdN (2)

2
= 2 + 2 . (B6)
d pT dy d pT dy d pT dy
Similarly, within the Bjorken model, one can calculate
the longitudinal HBT radii by including the second-order
viscous corrections in Eq. (32) using Eq. (B1). To this
end, we calculate N [δf2 ] by setting f = f0 + δf1 + δf2 in
Eq. (32) and performing the integrations
Z
N [δf2 ] = mT cosh3 (y − ηs )τ dηs rdrdϕ δf2
"
2A⊥ τ −5Φ2 n 2 2
K0 + zp2 K2


= zp − zm
ββπ 112βπ
o Φτ n
2 π
2zp2 + zm2
K0 + 2zp2 K2


+ zm K4 −
24τ
o 3Φ2 z 2
2 m
− zm K4 − (3K0 + 4K2 + K4 )
1120βπ
2 2
Φτπ zm Φτπ zm
+ (3K0 + 4K2 + K4 ) − K4
40τ  8τ 
2 2


Φ zm 3
− K0 + X2 − X + K 0 + zm X 2
4βπ 8
   
3 5 1 5
− zm X + zm K 1 + − X K2 + zm
2 8 2 16
 #
1 1 1
− zm X K3 + K4 + zm K 5 . (B7)
2 8 16

[1] U. Heinz and R. Snellings, Ann. Rev. Nucl. Part. Sci. 63,
123 (2013).
[2] W. Israel and J. M. Stewart, Annals Phys. (N.Y.) 118,
341 (1979).
[3] A. Muronga, Phys. Rev. C 69, 034903 (2004).
[4] J. Adams et al. [STAR Collaboration], Nucl. Phys. A
757, 102 (2005).
[5] K. Adcox et al. [PHENIX Collaboration], Nucl. Phys. A
757, 184 (2005).
[6] H. Song, S. Bass and U. W. Heinz, Phys. Rev. C 89,
034919 (2014).
[7] A. Adare et al. [PHENIX Collaboration], Phys. Rev.
Lett. 107, 252301 (2011).
[8] S. Chatrchyan et al. [CMS Collaboration], Phys. Rev. C
89, 044906 (2014).
[9] G. Aad et al. [ATLAS Collaboration], Phys. Rev. C 86,
014907 (2012).
[10] J. Jia [ATLAS Collaboration], Nucl. Phys. A 910-911,
276 (2013).
[11] Z. Qiu and U. Heinz, Phys. Lett. B 717, 261 (2012).
[12] A. El, Z. Xu and C. Greiner, Phys. Rev. C 81, 041901
(2010).
[13] G. S. Denicol, H. Niemi, E. Molnar and D. H. Rischke,
Phys. Rev. D 85, 114047 (2012).
[14] A. Jaiswal, R. S. Bhalerao and S. Pal, Phys. Lett. B 720,
347 (2013); J. Phys. Conf. Ser. 422, 012003 (2013).
[15] A. Jaiswal, R. S. Bhalerao and S. Pal, Phys. Rev. C 87,
021901(R) (2013).

[16] A. Jaiswal, Phys. Rev. C 87, 051901 (2013).
[17] P. Romatschke, Int. J. Mod. Phys. E 19, 1 (2010), and
references therein.
[18] D. Bazow, U. W. Heinz and M. Strickland,
arXiv:1311.6720 [nucl-th].
[19] H. Grad, Comm. Pure Appl. Math. 2, 331 (1949).
[20] S. Chapman and T. G. Cowling, The Mathematical The-
ory of Non-uniform Gases (Cambridge University Press,
Cambridge, 1970), 3rd ed.
[21] S.R. de Groot, W.A. van Leeuwen, and Ch.G. van Weert,
Relativistic Kinetic Theory: Principles and Applications
(North-Holland, Amsterdam, 1980).
[22] R. M. Velasco, F. J. Uribe, and L. S. Garcia-Colin, Phys.
Rev. E 66, 032103 (2002).
[23] E. Calzetta, arXiv:1311.1845 [hep-ph].
[24] K. Tsumura and T. Kunihiro, arXiv:1311.7059
[physics.flu-dyn].
[25] H. Beker et al. [NA44 Collaboration], Phys. Rev. Lett.
74, 3340 (1995).
[26] I. G. Bearden et al. [NA44 Collaboration], Phys. Rev. C
58, 1656 (1998).
[27] I. G. Bearden et al. [NA44 Collaboration], Phys. Rev.
Lett. 87, 112301 (2001).
[28] K. Adcox et al. [PHENIX Collaboration], Phys. Rev.
Lett. 88, 192302 (2002).
[29] M. Lopez Noriega et al. [STAR Collaboration], Nucl.
10
Phys. A 715, 623c (2003).
[30] A. Jaiswal, Phys. Rev. C 88, 021903 (2013).
[31] J. L. Anderson and H. R. Witting Physica 74, 466 (1974).
[32] K. Dusling, G. D. Moore and D. Teaney, Phys. Rev. C
81, 034907 (2010).
[33] P. Romatschke, Phys. Rev. D 85, 065012 (2012).
[34] D. Teaney and L. Yan, Phys. Rev. C 89, 014901 (2014).
[35] J. D. Bjorken, Phys. Rev. D 27, 140 (1983).
[36] D. Teaney, Phys. Rev. C 68, 034913 (2003).
[37] F. Cooper and G. Frye, Phys. Rev. D 10, 186 (1974).
[38] R. S. Bhalerao, A. Jaiswal, S. Pal and V. Sreekanth,
Phys. Rev. C 88, 044911 (2013).
[39] P. Kovtun, D. T. Son and A. O. Starinets, Phys. Rev.
Lett. 94, 111601 (2005).
[40] M. A. Lisa, S. Pratt, R. Soltz and U. Wiedemann, Ann.
Rev. Nucl. Part. Sci. 55, 357 (2005).
[41] U. A. Wiedemann and U. W. Heinz, Phys. Rept. 319,
145 (1999).
[42] M. Herrmann and G. F. Bertsch, Phys. Rev. C 51, 328
(1995).
[43] A. N. Makhlin and Y. .M. Sinyukov, Z. Phys. C 39, 69
(1988).
[44] T. Csorgo and B. Lorstad, Phys. Rev. C 54, 1390 (1996).
[45] M. Abramowitz and I. A. Stegun, eds., Handbook of
Mathematical Functions (Dover, New York, 1970), p. 378.