Arxiv:1202.2224V2 [Cond-Mat.Stat-Mech] 1 Jul 2012 Ls,Adi ES Fvrossses Serf.[,3] [2, Refs

Home , Shake (unit)

arXiv:1202.2224v2 [cond-mat.stat-mech] 1 Jul 2012 ls,adi ES fvrossses SeRf.[,3] [2, Refs. (See systems. reviews.) various is recent as of such FDR for NESSs systems, an in ageing how and in Re- modified glass, in be interest should much or perturbation. violated been without has spontaneous system there of cently, the correlation in time small fluctuations the a to with system FDR equilibrium perturbation An an of vibration. response external the to connects subject fluids (FDR) granular relation in fluctuation-dissipation the of validity the nature. in fluctuations NESSs of of property variety common wide the a descrip- identify in thermodynamic-like to goal any fundamental or find a tion to been years has many sta- It for non-equilibrium (NESSs). of vi- states example tionary archetypal of an state structures as spatial serves stationary complex pattern- without plain fluids these granular Besides the brated (see instabilities, therein). segregation convection, references forming and as and surface, such [1] the phenomena Ref. on of ex- formation variety They pattern rich fluids. in- a as non-equilibrium hibit studied of been have examples matter teresting Flu- granular of vibration. states external by idized energy sufficient supplied are † ∗ lcrncades [email protected] address: Electronic lcrncades [email protected] address: Electronic lcuto-ispto eain o oin fCne o Center of Motions for Relations Fluctuation-Dissipation o rnlrsses Dshv ensuidfrsev- for studied been have FDRs systems, granular For is paper this in addressed issues important the of One they when behavior fluid-like show materials Granular ASnmes 57.n 77.d 54.a 05.70.Ln 05.40.-a, d 47.70.Nd, the 45.70.-n, of numbers: scales PACS time relationshi the the and describe region we time) high-freque Finally, (long he the low-frequency mass parameters. in the of system especially that shown of center satisfied, is is t the It system a of granular predictions. with function theoretical system with response compared model and the spectrum for perfo power we simulations modifica center relation, dynamics a the fluctuation-dissipation molecular by with the defined equilibrium, of temperature at the validity effective The an satisfied wall. by be dimensiona thermal temperature to a any by known to f replaced relation generalized is constructed was plate originally bottom plate, was vibrating which bottom height, vibrating indepe Langevin-t mass a two phenomenological of by center A gravity of under simulations. media numerical granular and fluidized of state eivsiae h aiiyo utaindsiainre fluctuation-dissipation of validity the investigated We .INTRODUCTION I. 1 iaoooNtoa olg fTcnlg,Miyakonojo-sh Technology, of College National Miyakonojo 3 rdaeSho fEgneig aoaIsiueo Techno of Institute Nagoya Engineering, of School Graduate 2 eateto hsc,Kuh nvriy3,Fkoa812-8 Fukuoka 33, University Kyushu Physics, of Department u’ciWakou Jun’ichi lisudrGravity under Fluids Dtd ac ,2018) 7, March (Dated: 1 , 2 ∗ n aaauIsobe Masaharu and rvngaua ytm aebe tde nRf.[9– Refs. in uniformly studied Puglisi such been in have 13]. FDRs systems bath granular particle. thermal every driven by a of to using grains effect modeled couples all the often that to is wall, wall vibrating energy vibrating vibrat- (see, the the inject (or wall with shaking to collisions the bottom enough where frequent a strong case vibrate is the In energy or ing) injecting [8]). container Ref. of a e.g., means shake experimental to by typical achieved is forc- external A be of can means by ing. cooling NESS outside freely from gas a energy granular for supplying state a [4– stationary gas, calculated no granular is be there can While coefficients relation 7]. The Green-Kubo transport result (modified) interactions. which a a grain derive from as to the been develops of “cools” has gas nature aim and dissipative the forces the where of external gas, case without granular the to freely cooling devoted been freely has of work Much situations. eral oe fuiomydie rnlrgsadsoe that showed and gas granular driven uniformly of the model to Bunin distri- due velocity function. impurity by violated bution the replaced is of is impurity, behavior gas the non-Maxwellian rela- the of Einstein of temperature the temperature the of the which form analytically. in modified un- forces tion, a driving gas that uniform granular showed of a They influence in the the immersed der studied impurities Garzó [10] of velocity. diffusion grain mean-square the the replaced of as was fluctuation defined equilibrium temperature, at granular system the a tempera- by for equilibrium FDR observed the stud- the if They in and satisfied ture were observables. gas FDRs different granular the two that driven for uniformly FDRs of ied model a of mdetnieadacrt event-driven accurate and extensive rmed c sottm)rgo,frawd range wide a for region, time) (short ncy ain ntenneulbimstationary non-equilibrium the in lations ewe ytmtcdvain nthe in deviations systematic between p utaindsiainrlto o the for relation fluctuation-dissipation raoedmninlgaua a on gas granular one-dimensional fluctuation a the or describing theory ype ie rnlrsystem. granular riven r rdcsafluctuation-dissipation a predicts ory iy vnfrtecs nwihthe which in case the for even lity, dn prahs ae ntheory on based approaches, ndent fms iei nry ots the test To energy. kinetic mass of inta elcsteequilibrium the replaces that tion emlwl ttebto.The bottom. the at wall hermal ,Myzk,8586,Japan 885-8567, Miyazaki, i, tal et gtwr esrdadclosely and measured were ight oy 6-55 Japan 466-8555, logy, 9 are u ueia simulations numerical out carried [9] . 3 † asi rvnGranular Driven in Mass f 8,Japan 581, tal et 1]aaye mean-field a analyzed [12] . 2 the effective temperature defined by an FDR depends inelastic hard disks on a thermal wall. on the frequency. In the case where the shaking (or vi- Our main result is that an FDR with an effective tem- brating) is not strong enough to be regarded as uniform perature holds within statistical uncertainty for simula- driving, energy injection through a boundary has to be tions in a high-frequency (short time) region, while it is explicitly considered. Brey et al. [13] studied the vol- violated in a low-frequency (long time) region. The effec- ume fluctuations of a vibrated low-density granular gas tive temperature is defined by the COM kinetic energy. confined at the top by a mobile piston numerically. In We observed in our simulations that the ratio between this system, energy is supplied from the vibrating bot- the effective temperature and the global granular tem- tom wall. They discussed the interpretation of an effec- perature increases with inelasticity; the former can be tive temperature defined by requiring the same relation more than four times larger than the latter for the high- between fluctuations of volume and compressibility as in est inelasticity case. equilibrium systems. The FDRs and effective temperatures in much denser systems have also been studied by This paper is organized as follows. In Sec. II, we de- several authors [14–17]. Among these studies, we refer scribe a model granular system and discuss important to an experimental study by D’Anna et al. [17], because time scales in the system. In Sec. III, the Langevin their theory based on a Langevin equation formally has equation is introduced, and analytical expressions for the same form as ours, although their experimental setup the power spectrum and response function of the COM was very different. They performed an experiment to height are described briefly. We also remark on the FDR observe the fluctuating motion of a torsion oscillator im- between these two functions. The complete derivation of mersed in vibration-fluidized granular matter and found the Langevin equation and detailed calculation for the that it can be described to first approximation by the power spectrum and the response function are summa- formalism for Brownian motion in equilibrium, and an rized in Appendix A and B, respectively. A compari- FDR with an effective temperature approximately holds. son between the theoretical predictions and an extensive event-driven MD are shown in Sec. IV. Finally, in Sec. V, We investigated the fluctuating motion of the center of we summarize the main results for FDR validity and com- mass (COM) in an NESS of granular matter fluidized by ment on the relation between the systematic deviations in an external energy source located at a bottom wall, un- the low-frequency (long time) region and the time scales der the influence of gravity. Instead of using macroscopic of the driven granular system. probes such as a piston [13] or torsion oscillator [17], we focused on the position of the COM, which is observable using digital high-speed photography in experiments [18]. Our major motivation for studying fluctuations of the COM is that a simple (or universal) law might hold as a result of the following properties. First, the fluctuations II. THE MODEL SYSTEM of macrovariables such as the COM position often possess the largest time scales in the system. Second, they are A. System expected to be Gaussian in a similar sense to the central limit theorem. (In the case of a Markovian stochastic process, the Gaussian property of macrovariables fluctu- As a model of grains bouncing on a vibrating bottom ations can indeed be derived from a master equation of plate under gravity, we consider a d-dimensional system the Markovian process [19] .) With this expectation, we of N inelastic particles on a “thermal” bottom wall in a proposed a phenomenological theory based on a simple constant gravitational field g. The particles in the system formalism for Brownian motion that describes the motion have diameter σ and mass m; the total mass of particles of the COM height in the NESS of a one-dimensional vi- is denoted by M (= Nm). The thermal wall is kept brated granular fluid [20]. We found that the important at a constant temperature T0, which plays the role of a qualitative features of the dynamics of the COM in event- heat source supplying sufficient translational energy to driven molecular dynamics (MD) simulations were all ac- the particles to fluidize them. The z-direction is cho- counted for by the theory. The theory was extended to sen to be opposite to the direction of gravity, and the a two-dimensional granular fluid on a thermal wall [21]. thermal wall is fixed at z = 0. For simplicity, we adopt Here, we show that when we apply the phenomenologi- periodic boundary conditions in horizontal directions, so cal theory to granular fluids in higher dimensions, care- as to ignore the boundary (side-wall) effects. Collisions ful consideration of time scales in granular hydrodynam- between particles are inelastic; inelasticity of the par- ics [22, 23] is necessary. Within the time range for which ticle collisions is characterized by a normal restitution our theory is valid, it predicts the existence of an FDR. coefficient r. To avoid any pattern-forming instability in However, the equilibrium temperature in the FDR for the horizontal directions, we chose both the inelasticity an equilibrium system must be modified by the effective and linear scales of the system in horizontal directions to temperature of the COM velocity fluctuation. To test our be sufficiently small that the system remained homoge- prediction, we performed extensive and accurate event- neous in the horizontal directions. These conditions are driven MD simulations for a two-dimensional system of discussed in more detail in Sec. IV. 3

B. Time scales with time scaled by c/g are independent of g. We utilize this fact later to obtain a frequency response function in Before discussing the important time scales in the an efficient way. system, we define several quantities that character- There are three dimensionless parameters, obtained as ize the macroscopic properties of the system. We the ratios between two of these three time scales. The first define the kinetic energy per particle as K(t) first is τtherm/τosc Nz. The second is τosc/τdiss ≡ 2 ∼ 2 2 ∼ (1/N) N mv (t)2/2 and the long time average of K(t) Nz(1 r ). The third is τtherm/τdiss Nz (1 r ). The i=1 i first and− third parameters are the governing∼ parameters− in an NESS as K lim (1/T ) T K(t)dt (here- P T →∞ 0 for the hydrodynamic description of the system, as in- after, the overline on≡ a quantity represents its long time R troduced by Bromberg et al. [23]. They showed that the average in an NESS). We also define the global gran- steady-state profile is governed only by the parameter ular temperature T as k T (2/d)K, where k is B ≡ B the Boltzmann constant, and the thermal velocity as √π 2 1/2 1/2 1/2 Λ Nz(1 r ) , (2) c (dkB T/m) = (2K/m) . A characteristic length ≡ 2 − scale≡ of the system in the vertical direction l is then de- 2 1/2 fined as l c /g. which is proportional to (τtherm/τdiss) . If1 r 1, the ≡ − ≪ Bromberg et al. [23] have suggested that there are three second parameter τosc/τdiss is related to X Nz(1 r). ≡ − important time scales in this system at the hydrodynam- It plays the role of the governing transition parameter ical level: the macroscopic oscillation time τosc (referred from a condensed to fluidized state in a one-dimensional to as the “fast time scale” in Ref. [23]), the relaxation column of beads on a vibrating bottom plate [26]. In our time for thermal conduction τ , and the relaxation study, we consider the case Nz 1 and assume τtherm therm ≫ ≫ time for collisional dissipation τdiss. τosc in the following theoretical analysis. For simplicity, we assume that the system is nearly homogeneous, although this is not true for small r and III. THEORETICAL DERIVATION OF THE large N. The time scale τosc represents the period of the slowest oscillation in the vertical direction, that is, FLUCTUATION-DISSIPATION RELATION the period of the sound mode with the longest wave- length. Thus, τ l/c , where c is the sound velocity. Here, we summarize the theoretical derivation of (i) osc ∼ s s Assuming cs c, which is satisfied for a normal gas, the power spectrum, (ii) the frequency response func- τ can be estimated∼ as τ c/g. Because l/c also tion, and (iii) the FDR between (i) and (ii). First, we osc osc ∼ s characterizes the pressure relaxation time τp, we can re- introduce a Langevin equation as a first approximation gard τ and τ as on the same order, τ τ c/g. that describes the fluctuating motion of the COM on the p osc p ∼ osc ∼ The relaxation time for thermal conduction τtherm is es- fast time scales τosc and τp. Note that the derivation of timated as τ l2/(κ/ρc ), where κ is the thermal our theory has already been published in Ref. [20]. We therm ∼ p conductivity, ρ is the mass density, and c is the spe- assume τtherm τosc, as mentioned above, and focus on p ≫ cific heat at constant pressure [24]. ρ can be estimated the dynamics of the COM on the time scale τosc, ignoring d−1 as ρ M/(lA) mNz/(lσ ), where A represents the the significant slow relaxation process of fluctuations of area∼ of the bottom∼ plate in three dimensions (A repre- global granular temperature around its stationary value sents the length of the bottom plate in two dimensions (2/d)K/kB. The effect of this slow dynamics of granular and A = 1 in one dimension) and Nz represents the temperature and validity of our time scale assumption number of monolayers at rest. κ and cp are obtained are discussed later. from kinetic theory for elastic spheres and disks [25]: We summarize the details of the derivation of our d−1 κ kB c/σ and cp kB/m. Substituting these re- Langevin formalism in Appendix A and show the final sults,∼ we obtain τ ∼N c/g. The relaxation time for result here. We denote the height of the COM of granu- therm ∼ z collisional dissipation τdiss can be estimated as the inverse lar fluids at time t as Z(t), the time average of Z(t) over of (1 r2)ν, where ν is the collision frequency between a long time interval in an NESS as Z, and small devia- two particles.− Substituting the lowest order estimation tions of Z(t) from Z as δZ(t) Z(t) Z. The Langevin d−1 ≡ − of ν based on kinetic theory, ν ρσ c/m Nzc/l, we equation for fluctuating motion of δZ(t) is given by (see 2 −1 ∼ ∼ obtain τdiss (1 r ) c/(Nzg). Eq. (A5) in Appendix A) The time∼ scales− estimated above are summarized as d2δZ dδZ R(t) follows: = Ω2δZ µ + , (3) c c dt2 − − dt M τosc τp , τtherm Nz , ∼ ∼ g ∼ g where R(t) represents a random force, which is assumed −1 c τ N (1 r2) . (1) to be a Gaussian white noise: diss ∼ z − g ′ ′ R(t) =0, R(t)R(t ) = Iδ(t t ). (4) It is important to note that all time scales, τosc, τp, τtherm, h i h i − and τdiss, are proportional to c/g. This means that for a The brackets denote an average over the random system with given N and r, the macroscopic dynamics force. In NESS,h···i it is reasonable to assume Z(t) = Z, z h ist 4

where st represents the average in a stationary state. NESS against a small external force εf(t) can be defined The constanth· · · i I represents the intensity of the random as force, which is related to the second moment of the COM ˜ velocity fluctuations. This relation can be obtained by χ(ω) lim δZ˜(ω) /εf(ω), (9) ≡ ε→0 calculating the average kinetic energy of the COM motion D E 2 ˜ ˜ in z-direction KCOMz MVz(t) /2 st, where Vz is the where δZ(ω) and f(ω) are the Fourier transform of δZ(t) ≡ dZ(t) and f(t), respectively. The analytical expression of χ(ω) z-component of the velocity of the COM, Vz(t) dt . Using an analytical solution Eq. (B1) of the Langevin≡ is given as equation, we obtain KCOMz = I/4Mµ. Hence, the con- 1 1 χ(ω)= . (10) stant I is identified as M Ω2 ω2 + iµω − I 4MµKCOMz. (5) A detailed derivation is given in Appendix B. According ≡ to conventional definition, χ(ω) can be decomposed into This is the same procedure used to determine the noise real χ′(ω) and imaginary χ′′(ω) parts as χ(ω)= χ′(ω) intensity I when the Langevin equation describes fluctu- iχ′′(ω). Thus, we obtain the expression − ations in equilibrium at temperature T . In equilibrium, 2 2 equipartition of energy implies KCOMz = kBT/2, that is, 1 Ω ω χ′(ω) = − , (11) the mean kinetic energy of the COM in the z-direction M (Ω2 ω2)2 + (µω)2 KCOMz and the mean kinetic energy of a particle in one − direction kBT/2 are the same. Thus, we obtain the well- ′′ 1 µω known result I = 2MµkBT . In the case of an NESS χ (ω) = . (12) M (Ω2 ω2)2 + (µω)2 of granular fluids, the violation of equipartition of en- − ergy is observed in various systems. A heated binary Fluctuation Dissipation Relation Comparing granular system (see Ref. [27] and references therein) is Eqs. (8) and (12), we obtain the FDR one notable example in which non-equipartition between the mean kinetic energies of two species has been stud- ωS(ω) = χ′′(ω), (13) ied. Later, we present numerical simulations that clearly 2kBTeff show violation of equipartition, KCOMz = kBT/2, when 6 where T is an effective temperature defined as T we recognize T as the global granular temperature. eff eff ≡ The coefficients Ω and µ describe an angular frequency 2KCOMz/kB. This has the same form as the FDR in an of the slowest oscillation of the COM height and frictional equilibrium system except for Teff , which replaces the coefficient with respect to relative motion of the COM equilibrium temperature. height against the bottom wall, respectively. According −1 to the time scales we consider here, we assume Ω τosc and µ τ −1 and write them as ∼ IV. NUMERICAL SIMULATIONS ∼ p Ω= Ωˆg/c, µ =µg/c. ˆ (6) Here, we compare the three theoretical predictions described in the previous section with results of the numer- Because values of the coefficients Ωˆ andµ ˆ cannot be es- ical simulation of a two-dimensional granular gas system. timated in our phenomenological theory, they are fixed The predictions are the power spectrum Eq. (8), the fre- as fitting parameters when we compare results of simu- quency response function Eq. (10), and the fluctuation- lations with the theoretical predictions. dissipation relation Eq. (13). Our system consisted of N Power Spectrum The power spectrum S(ω) that rep- inelastic hard disks of mass m and diameter σ moving resents the fluctuations of Z around the NESS is defined in two dimensions on a thermal wall with a fixed tem- as the Fourier transform of the time correlation function, perature T0. Here, the x- and z-axes represent the horizontal and vertical directions of the system, respectively. ∞ −iωt The system width is denoted as L, and periodic bound- S(ω) dt e δZ(0)δZ(t) st . (7) ≡ −∞ h i ary conditions were adopted in the horizontal direction Z at x = 0 and x = L. The bottom wall was located at The derivation of S(ω) using the analytic solution of the z = 0, and there was no top wall. Gravitational force Langevin equation is straightforward. The final expres- was exerted on each disk along the negative z-direction. sion of S(ω) in this system is Inelastic collisions between hard disks were considered by the normal restitution coefficient r. When a disk collided 1 4µK S(ω) = COMz . (8) with a thermal wall at the bottom, it left with a value of 2 2 2 2 M (Ω ω ) + (µω) z-component of velocity vz sampled from the probability − density See Appendix B for a detailed derivation. 2 Response Function The frequency response function mvz mv p(v )= exp z . (14) χ(ω) that characterizes the linear response of Z in the z k T −2k T B 0 B 0 5

The horizontal component of velocity did not change dur- 2000 2000 300 ing the collision. Numerical simulations were performed with an event- driven algorithm devised to enhance the speed of calculation in dense hard sphere systems [28]. In the following, all simulation data are presented with mass, 1/2 length, and time in units of m, σ, and σ/(kBT0/m) ,

respectively. This corresponds to choosing kBT0 = 1. Height We set N = 5000 and L = 100 (these parameters are unchanged throughout this paper). For our main results, r = 0.99 0.999 and g = 10−3 were used un- − 300 - less otherwise mentioned. These correspond to Nz = 50, 0.05 X 0.5 and 1.98 Λ < 6.25. A system of width ≤ ≤ ≤ 0 0 0 L = 100 for r 0.99 is small enough to prevent any hor- (a) r=0.999 (b) r=0.992 izontal pattern≥ formation (e.g., ripples or undulations). The global temperature T and the thermal velocity c were 2000 300 1/2 calculated using T = K/kB and c = (2K/m) , where r=0.999 r=0.992 K is the long time average of the kinetic energy per disk. 1500 200 A. Macroscopic properties in the NESS 1000 Height In Fig. 1 (top), we show typical snapshots of particle 100 configurations in the system of N = 5000 and g = 10−3 500 for r =0.999 and 0.992. The corresponding area-fraction profiles are plotted in Fig. 1 (bottom). For a nearly elas- 0 0 tic case (r = 0.999), the profile had one peak around 0 0.05 0 0.5 the height z 350. However, the area fraction was rel- Area fraction atively dilute≃ (less than 0.06), even at the height of the peak. Many inelastic particles were raised up relatively FIG. 1: Top: Snapshots of the two-dimensional simulation 3 high, like the equilibrium profile of the Boltzmann distri- with N = 5000, L = 100, and g = 10− for different values of bution. In contrast, for r =0.992, the profile drastically the restitution coefficient (a) r = 0.999; (b) r = 0.992. Bot- changed. Most particles condensed at a relatively low tom: Area-fraction profiles averaged over a long time period level in a cluster; the area-fraction profile showed a clear for r = 0.999 and r = 0.992. peak above the low-density region around the thermal wall. This state is known as density inversion state and has been observed in many experiments [29, 30] and simulations [31, 32] of vibrofluidized granular matter. much smaller systems has been reported by us [21]. In In accordance with the theoretical study by Bromberg Ref. [23], it was shown that the density inversion appears above the threshold Λ (Λ > Λ ), where Λ 1.06569. et al. [23], which showed that the steady state is charac- c c c ≃ terized by a single parameter Λ, defined in Eq. (2), we In the density inversion state, which becomes pronounced plotted the average kinetic energy per disk K and the for Λ > 2, as shown in Fig. 1, a low-density and high- temperature gaseous region near the bottom can cause average kinetic energy of the COM KCOMz as a function of Λ in Fig. 2. The statistical error bars with stan- large fluctuations of the dense cluster on top. There- dard deviation were also plotted in all figures through- fore, this violation of the equipartition of energy should out the paper. Λ = 0 (that is, r = 1) corresponds be closely connected to development of the density inver- to the equilibrium state in which equipartition of en- sion. ergy 2KCOMz = K = kB T0 = 1 is satisfied. The fac- The relation between the long time average of the tor 2 comes from the fact that KCOMz is defined using COM height Z and the kinetic energy per particle is only the z-component of the COM velocity. The hori- given in Fig. 3. A linear relation Z = K/mg + const. zontal component of the velocity of the COM vanished was satisfied for K > 0.1, even when the system had in our simulations because the horizontal component of a density inversion with a relatively high density clus- disk velocity was unchanged on collision with the bot- ter. In an equilibrium system of dilute gases, the relation tom wall. While K systematically decreased following Z = K/mg + const. holds as a result of statistical me- −1.48 a power law Λ , KCOMz reached a minimum at chanics. The fact that K characterizes Z in the same way Λ 2 (r =0.999)∼ and increased with Λ for Λ > 2. Fig. 2 as in equilibrium suggests that the global granular tem- clearly≃ indicates that equipartition of energy breaks down perature T in the inhomogeneous non-equilibrium state when Λ > 2 (that is, 2K = K). Similar behavior in still retains the same meaning as the equilibrium temper- COMz 6 6

1 0.5 K r=0.999 2 K r=0.998 0.8 COMz 0.4 r=0.996 r=0.994 r=0.992 0.6 r=0.99

COMz 0.3 K K 0.4 0.2 K, 2 0.2 −1.48 0.1 1.04 Λ 0 0 1 2 3 4 5 6 7 0 1/2 2 1/2 0 0.005 0.01 Λ = 25π (1 - r ) g

FIG. 2: (Color online) Kinetic energy per particle FIG. 4: (Color online) Kinetic energy per particle K as a K (circles) and kinetic energy of the COM KCOMz function of g. The error bars are smaller than the sizes of the √π − 2 1/2 (squares), plotted versus Λ = 2 50(1 r ) for marks. r = 0.9999, 0.9996, 0.999, 0.998, 0.996, 0.994, 0.992, and 0.99 from left to right. The solid line gives a numerical ﬁt of the × 1.48 0 form 1.04 Λ− . 10

-1 ature, at least in a macroscopic sense. 10

500 -2 10 K/mg + const. P(C) r=0.999 r=0.998 400 (const. = 39) r=0.996 -3 10 r=0.994 300 r=0.992 r=0.99 Z -4 200 10 -4 -2 0 2 4 C

100 FIG. 5: (Color online) Probability distribution of the scaled 1/2 COM velocity C ≡ Vz/(2KCOMz/M) . The solid line is 0 Gaussian with unity dispersion. 0 0.1 0.2 0.3 0.4 K

FIG. 3: (Color online) The average height of the center of with our theory based on a linear Langevin equation with mass Z versus K for r = 0.999, 0.998, 0.996, 0.994, 0.992, additive Gaussian noise. and 0.99 from right to left. The error bars are smaller than the size of the marks. The solid line gives a linear ﬁt with the 1 3 slope (mg)− , where m = 1 and g = 10− . B. Power spectrum of the COM height

In Fig. 4, K is plotted as a function of the gravitational We first tested the theoretical prediction Eq. (8) for the acceleration g. The dependence of K on g turned out to power spectrum of the COM height. Using the relation be rather weak. We utilized this fact to measure the Eq. (6), Eq. (8) can be rewritten as response function from simulations in an efficient way (see Sec. IV C). c 3 K In Fig. 5, we plotted the probability distribution P (C) Sˆ(ˆω) S(ˆωg/c)/ 4 COMz of the scaled COM velocity C V /(2K /M)1/2. ≡ " g M # ≡ z COMz The data were fitted sufficiently by a Gaussian for all µˆ cases studied in this paper, as expected from the central = 2 , (15) Ωˆ 2 ωˆ2 +(ˆµωˆ)2 limit theorem. This Gaussian property was consistent − 7 whereω ˆ is the scaled angular frequency, defined byω ˆ ≡ r=0.999 ωc/g. This expression suggests that if we scale the power r=0.998 spectrum and the angular frequency as in Eq. (15), it × 5 r=0.996 shows a universal behavior independent of any system 1.0 10 r=0.994 parameters. r=0.992

) r=0.99 In Fig. 6 (top), the power spectrum S(ω) is plotted ω

for different values of r. Two sharp peaks were observed; S( 4 one is near zero angular frequency (ω = 0), the other 5.0×10 one is at the angular frequency of the macroscopic oscillation (ω = ωosc), which increased as r decreased. The heights of both these peaks decreased with r. Fig- ure 6 (bottom) shows the scaled power spectrum Sˆ(ˆω) 0.0 obtained by scaling S(ω) in Fig. 6 (top), according to 0 0.002 0.004 0.006 Eq. (15) using c and KCOMz calculated from simulation ω data. The theoretical prediction Eq. (15) with fitting numerical parametersµ ˆ = 0.50, Ω=1ˆ .7 is presented 1.5 r=0.999 as a thick solid line. It is consistent with the results r=0.998 of simulations for the range 0.99 r 0.996 in this ≤ ≤ r=0.996 region near the peak atω ˆ =ω ôsc ωoscc/g, where r=0.994 we expect our theory to serve as a first-order≡ approxi- 1 r=0.992 r=0.99 mation. We found large deviations from the theoretical ) ^ ω

prediction in the regionω ˆ < ωôsc (the sharp peak near ( S ωˆ = 0). As we illustrate below, the peak nearω ˆ = 0 ^ could be associated with slow fluctuations of global gran- 0.5 ular temperature due to thermal conduction and collisional dissipation. Because τtherm/τosc Nz 1 and τ /τ [N (1 r2)]−1 1.0 for our simulations∼ ≫ with diss osc ∼ z − ≥ Nz = 50 and r 0.99, the contributions of these two 0 ≥ 0 1 2 3 4 processes should appear atω ˆ < ωôsc. We also found that ω^ for r 0.998, the simulation data in Fig. 6 (bottom) de- viated≥ from our theory, even in the region near the peak FIG. 6: (Color online) Top: Power spectrum for the COM atω ˆ =ω ôsc. These deviations nearω ôsc could be attributed to the drastic change in density profiles shown height versus angular frequency ω. Averages were taken over 400 realizations. Bottom: Scaled power spectrum for the in Fig. 1 as r is varied. Concerning our theory, the change ˆ COM height. The solid line depicts the theoretical predic- in density profiles may affect the numerical coefficients Ω tion given in Eq. (15) withµ ˆ = 0.50 and Ωˆ = 1.7. andµ ˆ in Eq. (6). Furthermore, in the regionω ˆ < ωôsc, the effect of global temperature fluctuations mentioned above could become pronounced for r 0.998, because ≥ both τtherm and τdiss became much larger than τosc, and hence the fluctuations had long lifetimes. Nonetheless, a satisfactory explanation of these deviations for r 0.998 ≥ has not yet been given. power spectrum of δK′(t)/mg, where δK′(t)= K′(t) K. Here, we show simulations suggesting that the behav- In Fig. 7, we show the power spectrum of δK(t)/mg− , ior of S(ω) in the region near ω = 0 can be described by where δK(t) = K(t) K, and S(ω) for r = 0.999 and taking into account the slow dynamics of K(t). We de- 0.992. The figure shows− that the curves around the peak note the slowly varying part of K(t) as K′(t) and suppose in S(ω) near ω = 0 and the peak in the power spectrum it fluctuates on a much longer time scale than τosc, due of δK(t)/mg near ω = 0 are consistent. The consistency to thermal conduction and collisional dissipation. Then, between the two curves is also observed for the other r ′ K (t)/kB can be regarded as a time-dependent global values. This result indicates that the peak in S(ω) near granular temperature. Similarly, we let Z′(t) denote the ω = 0 can be accounted for by slow dynamics of K(t) slowly varying part of Z(t) on the same time scale as due to thermal conduction and collisional dissipation. It K′(t). We assume here that in this long time scale, should be emphasized that in our present theory, fluctu- K′(t) and Z′(t) play the same role as their long time ations of granular temperature in both space and time averages K and Z. That is, they satisfy the same linear are ignored and only the global granular temperature T relation as their long time averages observed in Fig. 3: is defined, using the long time average of K(t). Further Z′(t) = K′(t)/mg + const. with the same constant fac- investigation is necessary to construct a theory that fully tor. If this is the case, the power spectrum of δZ(t), describes the behavior of S(ω), taking into account the S(ω), in the region near ω = 0 should be given by the effect of slow fluctuations of granular temperature. 8

were non-negligible. This can be seen from the fact that 5 (a) δ 1×10 PS of Z(t) the relevant time scales shown in Eq. (1) all depended on PS of δK(t)/mg g and that exerting a constant force in the direction of gravity was equivalent to changing g. Therefore, a lin- K(t)/mg

δ ear response could be defined only in the limit of small 4 5×10 external force. This shows that our Langevin-type theory is different from the well-known Langevin theory for Brownian motion in a fixed harmonic potential, where

Z(t) and the response of a Brownian particle is linear against a δ 0 ﬁnite external force. Consequently, we had to exert an external force that was much smaller than the gravita- 4 (b) 1×10 tional force in our system, in order to measure the linear response of the COM height. Because the ﬂuctuation of the COM height of 5000 particles was typically much larger than the response against such a small constant 3 5×10 force, we needed to perform the response function measurement for a large number of systems with the same

Power spectrum of parameters Nz, r, and g but different initial conditions and take an average of the response functions over all 0 0 0.002 0.004 0.006 realizations. As shown later, in the case of a constant ω force that is 1% of the gravitational force, we needed more than 104 realizations to obtain sufficient statistics for clear response functions. This required relatively long FIG. 7: (Color online) Power spectrum (PS) of δK(t)/mg and CPU times that impeded long simulations with a wide of δZ(t), S(ω), for (a) r = 0.999 and (b) r = 0.992. range of parameters r, Nz, and g. We therefore optimized the method by choosing an ap- propriate parameter to approximately evaluate the re- C. Response functions sponse function from a small number of realizations, which could be provided in an acceptable time with our Next, we test the theoretical prediction Eqs. (11) and computational facilities. Suppose a system with gravi- (12) for the frequency response functions of the COM. tational field g is initially in an NESS and the gravita- By scaling these functions in the same way as the power tional acceleration is increased at t = 0 from g to g +∆g. spectrum, we can derive universal equations This is equivalent to exerting a step function external force M∆gθ(t) on the COM height, where θ(t) is the ˆ 2 2 − ′ ′ 2 2 Ω ωˆ Heaviside unit step function. Now we define the function χˆ (ˆω) χ (ˆωg/c)Mg /c = −2 , (16) ≡ 2 χ(t; g,g + ∆g) as Ωˆ 2 ωˆ2 +(ˆµωˆ) − d δZ χ(t; g,g + ∆g) h it /M∆g, (18) ′′ ′′ 2 2 µˆωˆ ≡− dt χˆ (ˆω) χ (ˆωg/c)Mg /c = 2 .(17) ≡ 2 Ωˆ 2 ωˆ2 +(ˆµωˆ) where t represents the average taken over the en- − sembleh· of · · realizations i δZ at time t. This is a function of The frequency response function was measured using ∆g in our system due to the nonlinear effects mentioned numerical simulations via the following procedure. First, above; it would equal the response function only if δZ h it we prepared for a system in the stationary state with were linear in ∆g. We denote the Fourier transform of a given Nz, r, and g after a sufficiently long relaxation χ(t; g,g + ∆g) as χ(ω; g,g + ∆g). According to Eq. (9), time from the initial state of particles with randomly dis- the frequency response function χ(ω; g) for the system in tributed positions and velocities. At t = 0, we exerted the stationary state with g is given by a small constant external force on all particles in the di- χ(ω; g) = lim χ(ω; g,g + ∆g). (19) rection of gravity and measured the height of the COM ∆g→0 at t> 0; from this COM relaxation process, we deduced a response function by the standard procedure given in We now consider the time scales that we introduced textbooks (see, e.g., Ref. [33]). In other words, we mea- in Sec. II B, which characterize macroscopic dynamics at sured a response function against a step functional exter- t> 0. As we discussed in Sec. II B, all these time scales nal force. The frequency response function was obtained in the NESS depend on g in the form τ = c(g)/g const., as the Fourier transform of the response function. where we wrote the g-dependence of c explicitly× for the It is important to note that in the response of the COM sake of clarity. Based on our observations in Fig. 4 that height against a small but finite external force in our sys- c(g) changed a few percent as g was increased by 10%, tem, nonlinear effects resulting from time scale changes we assumed that the thermal velocity at t > 0 is given 9 by c(g) if ∆g is sufficiently small. Thus, these time scales 30 at t> 0 have the form τ = c(g)/(g + ∆g) const., where -1 × 20 ∆g/g=10 g + ∆g is the gravitational acceleration at t > 0. This -2 dependence of all the characteristic time scales on ∆g ∆g/g=10 leads us to the scaling relation 10 )

2 ω

1 c(g) c(g) ´( 0 χ(ω; g,g + ∆g)= χˆ ω , (20) χ M g + ∆g g + ∆g -10 whereχ ˆ is a non-dimensional function. As long as Eq. (20) holds, we can estimate the limit in -20 Eq. (19) as -30 0 0.005 0.01 0.015 g + ∆g 2 g + ∆g ω χ(ω; g)= χ ω ; g,g + ∆g . (21) g g -1 To verify the validity of the scaling relation Eq. (20), 40 ∆g/g=10 -2 we performed two series of simulations for r = 0.992. ∆g/g=10 First, we measured the function χ(ω; g, g + ∆g) in −2

Eq. (18) for ∆g/g = 10 , taking the average over 41000 ) realizations. Second, we measured the frequency re- ω 20 −1 "( sponse function χ(ω; g) using Eq. (21) for ∆g/g = 10 , χ taking the average over 800 realizations. In Figs. 8, we compare the frequency response functions obtained from these two series of simulations. We found that they 0 were consistent, although there were some discrepancies in χ′(ω) near ω = 0. 0 0.01 More evidence of validity of the scaling relation comes 0.005 ω 0.015 from the fact that an FDR in an equilibrium system is satisfied when we measured the frequency response func- FIG. 8: (Color online) The real part χ′ (top) and the imag- tion using Eq. (21). This is discussed further later (see inary part χ′′ (bottom) of the frequency response function Fig. 10). versus angular frequency ω for r = 0.992. Circles show the 1 The frequency response functions presented below data for ∆g/g = 10− using Eq. (21) with averages obtained 2 were obtained using the scaling relation Eq. (20) (and over 800 realizations. Squares are for ∆g/g = 10− without Eq. (21)) by averaging over 800 realizations. In Figs. 9, using Eq. (21); averages were obtained over 41000 realizations. we show the real (top) and imaginary (bottom) parts of the scaled response functionsχ ˆ′ andχ ˆ′′ as functions ofω ˆ. Here, values of c in Eq. (16) and (17) were calculated in an NESS without perturbation. Theoretical predictions plotted as a function of ω for different r values. For all r Eqs. (16) and (17) with the same (universal) fitting pa- (0.99 r 0.999), we found that the FDR held within rameter as estimated in Fig. 6,µ ˆ =0.50 and Ω=1ˆ .7, are the error≤ bounds≤ in the higher frequency range of ω, in- shown by thick lines. It appears thatχ ˆ(ω) is consistent cluding a region near the highest peak at ω = ωp. Thean- with the theoretical predictions if r 0.996. gular frequency of the highest peak ωp was close to ωosc, ≤ defined as the angular frequency of a peak in S(ω). We stress here that we defined Teff as Teff = 2KCOMz/kB D. Fluctuation-dissipation relation in the FDR. The quantitative agreement in Fig. 11 sup- ports this definition of Teff , using KCOMz instead of us- To test the FDR Eq. (13) predicted by our theory, ing the global granular temperature T , because Teff is more than three times larger than T for r 0.996 (see we evaluate the left- and right-hand sides independently ≤ using the results of simulations on S(ω) (Sec. II B) and Fig. 2). ′′ χ (ω) (Sec. II C) presented in previous subsections. Note We found systematic deviations in the region ω<ωp that KCOMz were measured in the NESS where S(ω) was for r = 0.999 and r 0.994. These deviations were measured. related to the fact that≤ there was a peak near ω = 0 in First, we confirmed that the FDR held within the error S(ω) (shown in Fig. 6), while no corresponding peak near bounds of the simulation result in the whole range of ω ω = 0 appeared in χ′′(ω). As we discussed in Sec. IV B, given r = 1 in Fig. 10. The stationary state is just the the peak near ω = 0 could have been connected with equilibrium state of elastic particles on a thermal wall. slow fluctuations of granular temperature due to thermal In Fig. 11, the left- and right-hand sides of the FDR are conduction and collisional dissipation. For r = 0.999, 10

1 r=0.999 5 r=0.998 LHS r=0.996 RHS 0.5 r=0.994 4 r=0.992

) r=0.99 3 ^ ω 0 ( ´ ^ χ 2 -0.5 1

-1 LHS and RHS of FDR 0 0 1 2 3 4 5 ω^ 0 0.01 0.02ω 0.03 0.04 2 r=0.999 FIG. 10: (Color online) Left-hand side ωS(ω)/2kBTeff and right-hand side χ′′(ω) of Eq. (13) for r = 1. N = 5000, r=0.998 2 1.5 r=0.996 L = 100, and g = 10− , with averages over 400 realizations r=0.994 for S(ω) and over 800 realizations for χ′′(ω). r=0.992

) r=0.99 ^ ω 1 " ( ^ χ 300 200 0.5 250 (a) (b) LHS 150 200 RHS 0 150 100 100 0 1 2 3 4 5 50 ω^ 50 0 0 0 0.002 0.004 0 0.005 FIG. 9: (Color online) Real (top) and imaginary (bottom) parts of the scaled frequency response function versus the 100 60 scaled angular frequencyω ˆ = ωc/g. Averages were taken over 80 (c) 50 (d) 800 realizations. The thick lines are the theoretical prediction 40 ˆ 60 Eqs. (16) and (17) with fitting parametersµ ˆ = 0.50, Ω = 1.7. 30 40 20 20 10 the time scales of these two processes (τtherm and τdiss) 0 0 became much larger than τosc. Hence, the fluctuations LHS and RHS of FDR 0 0.005 0 0.005 0.01 with long lifetimes might be responsible for the devia- 50 40 tions at small ω. For r 0.994, the deviation appeared to increase as r decreased.≤ Because the system had lower 40 (e)30 (f) granular temperature for smaller r, a larger heat cur- 30 20 rent from the thermal wall was induced, causing larger 20 fluctuations in global granular temperature. Further in- 10 vestigation is necessary to understand this violation of 10 the FDR more precisely. 0 0 0 0.005 0.01 0 0.005 0.01 ω ω V. CONCLUSION FIG. 11: (Color online) Left-hand side ωS(ω)/2kBTeff and right-hand side χ′′(ω) of Eq. (13) for (a) r = 0.999, (b) r = We studied the validity of the fluctuation-dissipation 0.998, (c) r = 0.996, (d) r = 0.994, (e) r = 0.992, and (f) 3 relation with regard to the COM motion in an NESS r = 0.99. N = 5000, L = 100, and g = 10− , with averages of a driven granular fluid under gravity. By neglecting over 400 realizations for S(ω) and over 800 realizations for the fluctuations of global temperature caused by ther- χ′′(ω). mal conduction and collisional dissipation, which change much slower than the macroscopic oscillation of the fluid, 11 we derived a Langevin equation for the COM height using ify the physical meaning of the effective temperature. phenomenological considerations. This equation predicts functional forms of the correlation and response functions for the COM height that contain two phenomenological Acknowledgments numerical constantsµ ˆ and Ω,ˆ which are used as fitting parameters. It also gives a fluctuation-dissipation rela- J. W. is grateful to H. Nakanishi, T. Sakaue, T. Saito, tion accompanied by an effective temperature Teff that and C. Nakajima for their hospitality during his stay at characterizes the agitating motion of the COM height by Kyushu University, where part of this study was done. Teff =2KCOMz/kB. This study was supported by the Grant-in-Aid for Scien- To test the fluctuation-dissipation relation, we per- tific Research from the Ministry of Education, Culture, formed event-driven MD simulations and measured the Sports, Science and Technology No. 23740293. Some power spectrum and response function for the COM of the computations for this study were performed using height. While the power spectrum was consistent with the facilities of the Supercomputer Center, the Institute our theory for r 0.996 and ω around the angular fre- for Solid State Physics, the University of Tokyo, and the quency of the slowest≤ oscillation of the COM, it also Research Center for Computational Science (RCCS) in showed large deviations from the theoretical predictions Okazaki, Japan. near ω = 0 for all r (0.99 r 0.999) and in the whole range of ω for r > 0.996.≤ The≤ response function agreed closely with our theory for r 0.996 but showed devia- Appendix A: Langevin equation tions for r > 0.996. Furthermore,≤ we compared the left- and right-hand sides of the FDR. The results showed that In this section, we summarize the derivation [20, 21] of the FDR held in a region of ω near the highest peak for the Langevin equation that describes the motion of the all cases of r we tested. It was violated near ω = 0 for COM of grains. small r, r 0.994, and for r close to unity, r = 0.999. The equation of motion for the COM of the grains in ≤ For r 0.994, the violation became more pronounced as the model described in Sec. II can be written as r decreased.≤ The violation of the FDR was attributed to 2 a peak near ω = 0 in the power spectrum for the COM d Z M = Mg + Fb. (A1) height, which was absent in the imaginary part of the dt2 − frequency response function. The peak near ω = 0 in the The right-hand side of the equation of motion for the power spectrum cannot be described by our theory. COM must be, in general, the sum of the external forces We showed that these deviations near ω = 0 could acting on the grains. In our model, these are the grav- be attributed to slow fluctuations of global temperature, itational force Mg and the z-component of the force defined as the slowly varying part of the kinetic energy − exerted by the bottom wall Fb. Thus, it is essential to per particle K(t) due to thermal conduction and colli- understand the properties of Fb for the study of COM sional dissipation. These fluctuations of global temper- motion. ature were neglected in our theory. The deviations in ′ Let us consider the reaction force Fb(= Fb): the force the power spectrum and resulting violation of the FDR exerted by grains against the bottom wall.− A snapshot are expected to be accounted for by a theory that de- of the granular fluid is sketched in Fig. 12 (a). Suppose scribes both Z(t) and K(t), which we will investigate in the COM is at a height Z and is moving downward at the future. velocity V . We now change the frame of reference to In Ref. [34] a formula that connects the violation of the the center of mass frame (see Fig. 12 [b]); the bottom FDR in an NESS with the energy dissipation, or equiv- wall that lies a distance Z away from the COM in the alently the energy input from outside, was proposed. A z-direction is moving upward with velocity V . Now the theory extended to include the slow dynamics of K(t) ′ − problem is how to determine Fb, the force acting on the and direct measurement of energy input in our simula- bottom wall as a result of frequent collisions of granular tions might give some insight into the generality of their particles, in the situation shown in Fig. 12 (b). There, the formula. bottom wall is moving upward with velocity V against Finally, the basic question of whether the effective the macroscopically static fluid. − temperature Teff obtained here has any physical mean- This problem is similar to the problem of determining ing in terms of thermodynamics remains. The defini- the force acting on a one-dimensional Brownian particle tion of effective temperature in a system that relaxes in (the Rayleigh piston [19]) moving in the z-direction with several time scales, typically glass, has been debated in a velocity V (see Fig. 12 [c]). We create an expression ′ − ′ Refs. [2, 35, 36]. It would be interesting to apply their for Fb on the basis of this analogy and assume that Fb theories to our problem with three time scales τtherm, consists of three components. The first is a systematic τdiss, and τosc. It would also be interesting to investigate force fP (t) that equals the pressure multiplied by the via simulation what happens if two systems with differ- area of the bottom wall. Because the local density near ent effective temperatures are in contact with each other. the bottom wall changes according to the motion of the Measuring the direction of heat flow directly might clar- COM, this force may depend on time. Apparently, the 12

Appendix B: Derivation of the power spectrum and (a) (b) (c) the response function Center of mass -V Derivation of the power spectrum and the response V Z Z Fb’ function from the Langevin equation describing Brown- -V ian motion in a harmonic potential is given in textbooks

Fb’ Fb’ (see e.g., Ref. [37]). We therefore present only essential steps in their calculation. First, we consider the power spectrum of the ﬂuctuating motion of the COM obey- ing the Langevin equation (A5). The formal solution of FIG. 12: (a) Schematic of the system observed in the labo- Eq. (A5) is written as ratory frame of reference, (b) the same system observed in t ′ the center of mass frame, (c) the Rayleigh piston: a piston ′ R(t ) ′ Z(t) Z = G(t t ) dt + Fini(t), (B1) that undergoes random collisions with a one-dimensional heat − −∞ − M bath of particles. Z where the function G(t) is given by

µ e− 2 t long time average of f (t), that is, f , must be equal G(t)= sin (ω0t) , (B2) P P ω to Mg, the gravitational force acting on all particles. 0 The− simplest assumption for the time-dependent part of and ω is defined by ω (Ω2 (µ/2)2)1/2. The last 0 0 ≡ − fP (t) is that it is proportional to the deviation of the term Fini(t) in Eq. (B1) consists of those that depend on COM height from its stationary value, Z(t) Z. This is the initial conditions and vanish after a sufficient amount because the change in local density near the− bottom wall of time. Thus, the term is negligible when calculating is proportional to (Z(t) Z) if the change in the height long time averages of physical quantities in the stationary of the COM is sufficiently− − small: Z(t) Z /Z 1. The state. second component is a frictional| force.− We| assume≪ here Using this formal solution, we can calculate the two- the simplest form of the frictional force: linear in the time correlation function φ(t) in an NESS defined by ′ ′ relative velocity V (t) of the bottom wall to the COM. φ(t) limt′→∞ δZ(t )δZ(t + t) , where the brackets The third component− is a random force. We assume ≡indicate anh average over thei random force R(t). Weh· · · tooki the limit t′ to ensure that the system is in F ′(t) = Mg + MΩ2 Z(t) Z + MµV (t)+ R′(t) the stationary state.→ ∞ b − − The power spectrum of δZ(t) can be obtained using = Fb(t), (A2) − the Winner-Khinchin theorem: where Ω is a coefficient that specifies the angular fre- ∞ −iωt quency of the slowest oscillation of the COM, and µ is S(ω) = dte φ(t) (B3) the frictional coefficient. According to the discussion of Z−∞ I 1 characteristic time scales in Sec. II B, the time scales for = . (B4) macroscopic oscillation τ and that for pressure relax- M 2 (Ω2 ω2)2 + (µω)2 osc − ation τ are τ τ c/g. Thus, we assume p osc ∼ p ∼ Next, we consider the response function for the COM, which describes the linear response of the COM with re- ˆ ˆ Ω= Ω/τosc = Ωg/c, µ =µ/τ ˆ p =µg/c. ˆ (A3) gard to a small external force εf(t). The Langevin equation in this case is written as For the random force, we assume stationary Gaussian d2δZ dδZ εf(t) R(t) white noise in the same way as for the Rayleigh piston: +Ω2δZ + µ =0. (B5) dt2 dt − M − M R′(t) =0, R′(t)R′(t′) = Iδ(t t′). (A4) h i h i − Taking the average over the random force, we obtain where I represents the intensity of the random force. d2 δZ d δZ εf(t) h i +Ω2 δZ + µ h i =0. (B6) Substituting the Fb obtained in (A2) into the equation dt2 h i dt − M of motion of the COM (A1), we obtain The response function χ(t) is defined as d2δZ dδZ R(t) t = Ω2δZ µ + , (A5) δZ(t) = dt′χ(t t′)εf(t′). (B7) dt2 − − dt M h i − Z−∞ where δZ Z(t) Z and R(t) = R′(t). The random Here, the external force εf(t) is assumed to be infinitely force R(t)≡ has exactly− the same property− described in small. The Fourier transform of this relation yields (A4) as R′(t). Note that the Langevin equation (A5) has δZ˜(ω) = χ(ω)εf˜(ω), and hence h i the same form as that describing Brownian motion in a χ(ω) = lim δZ˜(ω) /εf˜(ω). (B8) harmonic potential. ε→0 D E 13

Performing the Fourier transform of the relation (B6) and comparing it with Eq. (B8), we obtain the frequency response function (complex admittance) 1 1 χ(ω)= . (B9) M Ω2 ω2 + iµω −

[1] I. S. Aranson and L. S. Tsimring, Rev. Mod. Phys. 78, 77, 034402 (2008). 641 (2006). [21] J. Wakou and M. Isobe, AIP Conf. Proc. 1227, 135 [2] L. F. Cugliandolo, J. Phys. A: Math. Theor. 44, 483001 (2010). (2011). [22] J. J. Brey, M. J. Ruiz-Montero, and F. Moreno, Phys. [3] U. M. B. Marconi, A. Puglisi, L. Rondoni, and A. Vulpi- Rev. E 63, 061305 (2001). ani, Phys. Rep. 461, 111 (2008). [23] Y. Bromberg, E. Livne, and B. Meerson, in Granular [4] I. Goldhirsch and T. P. C. van Noije, Phys. Rev. E 61, Gas Dynamics, edited by T. Pöschel and N. Brilliantov 3241 (2000). (Springer-Verlag, Berlin, 2003), pp.251-266. [5] J. W. Dufty and V. Garzó, J. Stat. Phys. 105, 723 (2001). [24] L. D. Landau and E. M. Lifshitz, Fluid mechanics (Perg- [6] J. W. Dufty and J. J. Brey, J. Stat. Phys. 109, 433 amon Press, New York, 1987). (2002). [25] S. Chapman and T. G. Cowling, The Mathematical The- [7] J. Dufty, A. Baskaran, and J. J. Brey, J. Stat. Mech. ory of Non-Uniform Gases, (Cambridge University Press, Theory Exp. L08002 (2006). Cambridge, 1970). [8] J. Duran, Sands, Powders, and Grains: An Introduction [26] S. Luding, E. Clément, A. Blumen, J. Rajchenbach, and to the Physics of Granular Materials (Springer Verlag, J. Duran, Phys. Rev. E 49, 1634 (1994). New York, 2000). [27] H.-Q. Wang and N. Menon, Phys. Rev. Lett. 100, 158001 [9] A. Puglisi, A. Baldassarri, and V. Loreto, Phys. Rev. E (2008). 66, 061305 (2002). [28] M. Isobe, Int. J. Mod. Phys. C10, 1281 (1999). [10] V. Garzó, Physica A 343, 105 (2004). [29] A. Kudrolli, M. Wolpert, and J. P. Gollub, Phys. Rev. [11] A. Puglisi, A. Baldassarri, and A. Vulpiani, J. Stat. Lett. 78, 1383 (1997). Mech. Theory Exp. P08016 (2007). [30] R. D. Wildman, J. M. Huntley, and J.-P. Hansen, in [12] G. Bunin, Y. Shokef, and D. Levine, Phys. Rev. E 77, Granular Gases, edited by T. Pöschel and S. Luding 051301 (2008). (Springer-Verlag, Berlin, 2001), pp. 215-232. [13] J. J. Brey and M. J. Ruiz-Montero, Phys. Rev. E 81, [31] Y. Lan and A. D. Rosato, Phys. Fluids 7, 1818 (1995). 021304 (2010). [32] M. Isobe and H. Nakanishi, J. Phys. Soc. Jpn. 68, 2882 [14] H. A. Makse and J. Kurchan, Nature 415, 614 (2002). (1999). [15] A. Barrat, V. Colizza, and V. Loreto, Phys. Rev. E 66, [33] R. Kubo, M. Toda, and N. Hashitsume, Statistical 011310 (2002). Physics II, (Springer-Verlag, Berlin, 1991). [16] C. S. O’Hern, A. J. Liu, and S. R. Nagel, Phys. Rev. [34] T. Harada and S.-I Sasa, Phys. Rev. Lett. 95, 130602 Lett. 93, 165702 (2004). (2005). [17] G. D’Anna, P. Mayor, A. Barrat, V. Loreto, and F. Nori, [35] L. F. Cugliandolo, J. Kurchan, and L. Peliti, Phys. Rev. Nature 424, 909 (2003). E 55, 3898 (1997). [18] S. Warr, J. M. Huntley, and G. T. H. Jacques, Phys. Rev. [36] L. Berthier and J.-L. Barrat, J. Chem. Phys. 116, 6228 E 52, 5583 (1995). (2002). [19] N. G. van Kampen, Stochastic Processes in Physics and [37] P. Résibois and M. De Leener, Classical Kinetic Theory Chemistry, (Elsevier Science, Amsterdam, 1992). of Fluids, (Jhon Wiley & Sons Inc, New York, 1977). [20] J. Wakou, A. Ochiai, and M. Isobe, J. Phys. Soc. Jpn.