Available online at www.sciencedirect.com Powder Technology 182 (2008) 130 – 136 www.elsevier.com/locate/powtec Introduction to granular temperature ☆ Isaac Goldhirsch School of Mechanical Engineering, Faculty of Engineering, Tel Aviv University, Ramat-Aviv, Tel Aviv 69978, Israel Available online 14 December 2007 Abstract The goal of this article is to provide a somewhat critical introduction to the concept of granular temperature and some of its applications. A brief history of the concept is followed by a presentation of several of its major properties and implications thereof. A number of misconceptions concerning this concept is presented and discussed. Certain open questions are described and some recent developments are briefly outlined. © 2007 Elsevier B.V. All rights reserved. Keywords: Temperature; Granular matter; Granular gases; Kinetic theory PACS: 05.20.Dd; 47.45.Ab; 45.70.Mg; 83.80.Fg; 62.20.Dc 1. Introduction In his classic paper on Brownian motion [1] Einstein notes that “... a dissolved molecule is differentiated from a suspended body solely by its dimensions” and claims that the properties of a collection of suspended macroscopic particles are basically the same as those of suspended molecules. Put in modern terms, if molecules are replaced by macroscopic particles, one can use all the tools and concepts that pertain to molecular assemblies (though quantum mechanics is not directly relevant in this case) to obtain similar properties albeit on larger spatial and temporal scales. It is for this reason that one may find it strange that the concept of granular temperature as a measure of the velocity fluctuations in a fluidized granular system (now known as a granular gas) was put forward only as late as 1978 [2,3]. The importance of energy fluctuations in granular gases was also appreciated by a few others in the late seventies, cf .e.g., [4–7]. Perhaps the reason for this is that most prior work on granular matter had focused on dense and nearly-static granular materials (in particular, soils) where one has little a-priori reason to define a (granular) temperature characterizing the random parts of the velocities of the grains. The idea expressed in [2,3] was almost immediately put to use (in the early eighties) in constructing ☆ Support from the Israel Science Foundation (ISF), grant. no. 689/04, the German Israel Science Foundation (GIF), grant. no. 795/2003 and the US-Israel Binational Science Foundation (BSF), grant. no. 2004391, is gratefully acknowledged. E-mail address: isaac@eng.tau.ac.il. 0032-5910/$ - see front matter © 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.powtec.2007.12.002 kinetic-based theories for granular gases, whose states were then referred to as “rapid granular flows”. Early work can be found in [8,9]. Even some of the phenomenology of molecular gases, like mean field and mean free path theories, was adapted to granular matter [10,11]. A summary of these early results, their successes and failures, can found in the review article [12]. Granular kinetic theory was perfected and made systematic in the nineties of the twentieth century [13,14]. Some of the differences between the pioneering approaches to granular gases and the more modern treatments, as well as new insights and results obtained from the latter are presented below, see also the review paper [15]. While Einstein's theory of Brownian motion (and its extensions) is now widely accepted, one must bear in mind that Einstein studied an equilibrium system, and that Brownian motion is a manifestation of equilibrium fluctuations. It is not apriori clear in general that a concept derived from equilibrium thermodynamics or statistical mechanics is directly relevant to granular assemblies, even granular gases. The reason is that unlike molecules, grains collide inelastically (spontaneous retrieval of the energy taken up by the grains can only happen on huge time scales, not relevant to the present discussion) and therefore one needs to pump energy into a granular gas in order to maintain it fluidized. Consequently, granular gases are always in nonequilibrium states. Furthermore, grains undergo attrition, breakup, coagulation and other processes, see e.g., [16], which are rare for molecular gases at not-too-high temperatures. Therefore it is not immediately evident that a characterization of the motion of grains by a granular temperature (field) and the I. Goldhirsch / Powder Technology 182 (2008) 130–136 related applications in kinetic and hydrodynamic theories are useful. Indeed, some researchers expressed doubts concerning the applicability of kinetic theory and hydrodynamics (both of which employ the granular temperature as a basic entity) to granular gases [17–19] whereas others ignored the granular temperature in their theories, see e.g., [20]. One must realize though that in addition to the physical and mathematical arguments (part of which are presented below) which support the use of the granular temperature and (the related) kinetic theories, the successes of the kinetic and hydrodynamic theories of granular matter, even beyond their nominal ranges of applicability [21–24], are impressive and speak for themselves. As is well known, the macroscopic velocity of a fluid, or its velocity field, does not represent the velocities of the molecular motions: whereas the speeds of the molecules are typically of the order of the speed of sound in a fluid, the fluid may exhibit much slower macroscopic (also known as hydrodynamic or averaged or coarse-grained) speeds. The difference between the velocity of a molecule that resides (more accurately, it's center of mass does) at point r at time t and the value of the hydrodynamic velocity at that point is known as the peculiar or fluctuating velocity of the molecule and is basically a random entity (note that ‘random’ does not imply e.g., ‘directionally isotropic’). Since by definition the average (whose precise nature must be specified) of the peculiar velocity is zero, a good measure of its magnitude is the average of the square of this entity. This average is proportional to the local temperature of the fluid (a field in its own right), by definition. A similar argument can be used to justify the definition of a granular temperature. One can define a velocity field as an averaged or coarse-grained entity (see the paper by Serero et al in this issue for formal definitions) and distinguish it from the actual velocities of the individual grains. Let r i(t) denote the center of mass position of grain i at time t and v i(t) its velocity. Let V(r,t) represent the velocity field of the system one considers. The macroscopic velocity field at the center of mass position of particle i is given by V(ri(t),t). The fluctuating (or peculiar) velocity of the grain is defined as u(ri(t),t) ≡ v i(t) − V(ri(t),t) (for a more general definition that takes into account the finite resolution involved in coarse-graining see the paper by Serero et al in this issue). Clearly (as in the molecular case) the average of the peculiar velocity vanishes. Therefore it is convenient to define the average of the square of the peculiar velocity as a measure of the velocity fluctuations. It is common to define the granular temperature for a monodisperse system as D1 bu2 N, where D denotes the spatial dimension and bXN is the average of the entity X (over space or time or both, or an ensemble; see more below). A slightly different definition is usually employed for polydisperse systems (see the paper by Serero et al). The fact that one can define an entity such as the granular temperature (defining objects is usually not difficult) does not prove that this definition begets useful results; only such results can provide a justification for the definition. Indeed this is the case, as shown below and as this entire special issue hopefully demonstrates. Though temperature has intuitive and everyday significance, ∂S 1 its (equilibrium) thermodynamic definition is: T u ∂U , 131 where S denotes the entropy of a system and U its energy (all other independent thermodynamic variables being kept fixed). Statistical mechanics shows that the thermodynamic temperature of a classical system is the same as that defined using the peculiar velocity (up to a factor that includes the mass of a particle and, when one wishes to measure the temperature in degrees, another factor of Boltzmann's constant). The advantage of the statistical mechanical definition is that it makes possible to define temperature as a statistical property without the need to refer to a state of equilibrium. Moreover, one does not even need to define an ensemble of states since one can average u2 (as is often done in simulations) over (usually small) volumes and time ranges instead. Therefore the temperature is usually well defined even for far-from-equilibrium systems, but it does not necessarily possess (all of) the same properties as the equilibrium temperature. The reason that the concept of temperature is so useful in most molecular systems is not only the fact that it comprises an important characterization of the velocity fluctuations, but the strong scale separation that characterizes these systems. Even for “large” temperature gradients, say 100 K/cm, the change of temperature over a typical interatomic distance in a solid (about 3 Å) is 3 × 10− 6 K; in air under standard (STP) conditions the corresponding change over a mean free path (about 0.1 μm) is still only 10− 3 K. Therefore on scales of many interatomic distances or mean free paths the temperature (as well as other characterizations) can be considered to be nearly constant, justifying the notion of local equilibrium. This way, a thermometer of macroscopic size can measure a meaningful temperature. One should be reminded that local equilibrium is not equilibrium since e.g., a system subject to a temperature gradient supports a heat flux, unlike a state of equilibrium. This implies that the distribution function of a system in local equilibrium cannot be Maxwellian since the heat flux corresponding to a Maxwellian distribution vanishes; however the distribution function can be close to Maxwellian, thereby further justifying the notion of local equilibrium. Returning to granular systems, it turns out that they do not usually possess the strong scale separation that characterizes molecular systems, except in the case of nearly-elastic interactions. This fact has numerous consequences, one of which is the sizeable normal stress differences in granular gases (see below). Still, the use of granular temperature in the realm of granular gases, and to some extent in dense granular fluids, has proven highly successful [15]. Fluid temperatures can be measured using thermometers. There is no known granular thermometer [see though the suggestion proposed in [25]] or direct way to measure the granular temperature. The granular temperature of a system (or parts of it) can be found either by detailed measurements of the grains’ velocities, numerically (in simulations) or experimentally (see e.g., the paper by Wildman and Huntley in this issue) or indirectly through its consequences. The fact that a granular thermometer is not available cannot be taken as an argument against the concept itself, since its consequences are measurable and important. Another argument against the usefulness of the concept of granular temperature is the fact that equipartition does not hold in 132 I. Goldhirsch / Powder Technology 182 (2008) 130–136 granular systems. This fact is fully consistent with results obtained from kinetic theory and is by itself not truly surprising since equipartition is not supposed to be valid in any nonequilibrium system. An interesting manifestation of the lack of equipartition is the fact that the granular temperatures of the components of a mixture do not equal each other (see the paper by Serero et al in this issue). The rotational and translational temperatures of granular gases that experience tangential restitution (all real granular gases do) do not equal each other either. The only relatively strong argument against the use of the concept of granular temperature or kinetic theory is the fact that granular gases (and solids, as a matter of fact) lack strong scale separation. This issue is taken up below. The structure of this paper is as follows. Section 2 relates granular temperature to kinetic theory and mentions some of the results of this theory. It also explains the difference between kinetic-based theories and systematic kinetic approaches. Section 3 is devoted to the lack of strong scale separation and some of its consequences. Section 4 describes one of the important consequences of the inelasticity of the granular collisions, namely the clustering phenomenon and presents some of the consequences of this effect. It also presents the collapse phenomenon. Section 5 is devoted to the description of some of the misconceptions that are still abound concerning granular temperatures. Finally, Section 6 provides a brief summary and some additional comments, and briefly outlines some recent developments and extensions of the concept of granular temperature. 2. Granular temperature and kinetic theory One of the main applications of the concept of granular temperature is the construction of kinetic and (from them) hydrodynamic theories for granular gases. In the pioneering papers (as well as some recent ones) in the field, the pertinent Boltzmann equation [26] was not directly employed. Instead, a set of relations among the low order moments of the (single particle) distribution function, f (r,v,t)(that represents the number density of particles at point r and time t whose velocity is v), known as the Enskog equations, see e.g., [8,9,27] and references therein, was used (these equations can be obtained from the Boltzmann equation, and, more generally from the BBGKY hierarchy). Since the Enskog equations are not closed (the equation for a given moment depends on higher order moments), closure was obtained by conjecturing a form of the distribution function, typically a Maxwellian multiplied by terms that depend on the local gradients of the hydrodynamic fields. These guesses missed certain terms that systematic theories [13,14] obtain. For instance, when the gradients vanish these guesses would entail the assumption that the distribution function is Maxwellian. However, in the absence of gradients (or forcing) a granular gas is in a “homogeneous cooling state” (HCS) whose distribution is known not to be Maxwellian, see [15,28] and references therein. The missed corrections are important when the grain collisions are not close to elastic and therefore the “old” theories are at best relevant in the nearelastic case. The above theories also account for finite density effects by using a correction to the Boltzmann equation proposed by Enskog. The hydrodynamic equations of motion resulting from the application of these theories resemble the Navier–Stokes equations with one (very important) difference: the equation for the energy density (or granular temperature) contains a “sink term” that represents the loss of energy due to the inelasticity of the collisions (as a matter of fact, the form of these equations can be shown to be very general, much beyond the realm of validity of kinetic theory). This term is “responsible” for the existence of steady granular shear flows (else the work by shear would have “heated” the granular system indefinitely) and for many other phenomena that characterize granular gases. The agreement of the early theories with experiments is reasonable given their approximate nature, see e.g., [12]. There are two main systematic approaches to the kinetics of granular gases. The first is based on the observation that in the limit of elastic collisions and when the gradients vanish, the distribution function of the grain velocities is Maxwellian, corresponding to a state of (local) equilibrium. This limit is not singular [13] and therefore one can expand the solution in two small parameters around this state of equilibrium: the Knudsen number (a non-dimensional measure of gradients, defined as the ratio of the mean free path and the scale over which the hydrodynamic fields vary) and the degree of inelasticity, defined as ϵ ≡ 1 − e2, where e is the coefficient of normal restitution (when more complex interaction are allowed, the definition of the degree of inelasticity needs to be modified; also, additional small parameters appear when tangential restitution is allowed). This theory is restricted to the case of nearly-elastic collisions. The second approach [14] is based on an expansion in the Knudsen number (or gradients) without any other small parameters. The zeroth order in the latter expansion is the HCS, for any value of the coefficient (or coefficients) of restitution. The respective expansions in both methods are extensions of the classical Chapman– Enskog expansion of kinetic theory [26]. Both approaches produce constitutive relations, which agree with each other in the common domain of validity (nearly-elastic collisions). Accurate results for the transport coefficients for granular gases have been recently obtained [28] on the basis of the method proposed in [14]. Among the novel results obtained using the systematic approaches we would like to mention the discovery of a new contribution to the heat flux, which is proportional to the gradient of the number density (of course, the Fourier term, which is proportional to the gradient of the granular temperature is there as well). This term has been shown to strongly affect e.g., the temperature profile in a vertically vibrated granular bed [29], predicting a nonmonotonic dependence of the temperature on the height. Another important result [13] is that the normal stress differences observed in simulations and experiments [12,15] stem from contributions of the Burnett terms, i.e. terms in the constitutive relations which are of second-order in the gradients and therefore beyond the Navier–Stokes level of description. The latter result quantitatively agrees with simulations. The reason that the Burnett terms are of importance for granular gases is discussed below in Section (3). While the pioneering studies of granular kinetics did employ the Enskog correction, the first systematic studies did not. This I. Goldhirsch / Powder Technology 182 (2008) 130–136 seems to have led some people to believe that the systematic studies are only good for very low volume fractions. This was rectified by papers in which moderately dense granular gases were studied [30] on the basis of the Enskog corrected Boltzmann equation. Another approach to the study of the dynamics of moderately dense granular gases involves the derivation of corresponding Green–Kubo relations [31,32], whose validity is not limited in density (see also the article by Brey and Prados in this issue). This problem is currently under active research. Yet another method to extend the validity of kinetic theory for granular gases is to go to higher orders in the BBGKY hierarchy and include the “ring contributions” [33] that account for events in which a particle recollides with another particle after the other particle collides with at least one more (third) particle. These events are known to contribute to the buildup of precollisional correlations. A problem that has not been yet satisfactorily addressed in the kinetic theories of granular gases is that of the “molecular chaos” assumption. The only assumption underlying the derivation of the Boltzmann equation for (molecular or granular) gases is that the velocities of particles about to collide are not correlated. This assumption has been validated in simulations of elastically colliding particles at low density but inelastically colliding particles do not possess this property (except for the near elastic case), cf. e.g., [34]. One of the basic reasons for this difference is that in inelastic collisions the normal component of the relative velocity of the colliding particles is decreased. The Enskog correction accounts for part of these correlations but not all (e.g., their directional dependence). The successes of the existing granular flow models suggest that in most cases these correlations do not strongly affect the hydrodynamic constitutive relations, but the matter deserves further study. The mentioned derivations of Green–Kubo relations and studies of higher order equations in the BBGKY hierarchy should contribute to the elucidation of the role of precollisional correlations. 3. Lack of strong scale separation in granular gases In molecular fluids under non-extreme conditions the mean free paths are much smaller than the scales on which gradients change (small Knudsen numbers) and the mean free times are much smaller than typical macroscopic times scales. It turns out that in granular gases (as a matter of fact, also in solids) there is no strong scale separation, except in the near-elastic case [35]. This fact and some of its consequences seem to underlie a certain confusion in part of the community, and has led to “hasty” conclusions that the concept of granular temperature as well as kinetic theory may not be relevant to granular gases. The reason for the lack of strong scale separation is presented immediately below. Consider a simply sheared (monodisperse) granular gas, i.e. one whose density and granular temperature are constant and whose velocity field is given by: V(r,t) = γyx̂, where γ denotes the shear rate and x̂ is a unit vector. In the absence of gravity (for simplicity) the only ‘input’ parameter in this problem that has dimensions of time is γ− 1. The only relevant length scale is the mean free path ℓ (which equals nr1T , where n is the number 133 density and σ T the total collisional cross section). It follows from dimensional analysis that the granular temperature, T, must satisfy: T∝γ2ℓ2. Next, assume that the collisions are characterized by a fixed coefficient of normal restitution, e. The degree of inelasticity is defined by ϵ ≡ 1 − e2. Clearly, in the absence of inelasticity the granular temperature steadily increases because of the shear work. Therefore the steady state temperature diverges2 when ϵ is set to zero. One can therefore assume that 2 T ¼ C g ℓ , where C is a constant. Phenomenological models [10] as well as accurate kinetic theoretical studies [15] corroborate this result and determine C to be about 1 in two dimensions and 3 in three dimensions, i.e. O(1). The mean free time, τ, is given by the ratio of the mean free path and the thermal speed: s ¼ pℓffiffiTffi. The macroscopic time scale in this problem is the inverse shear rate. Therefore the ratio of the microscopic time scale, τ, and the s macroscopic time scale is: 1=g ¼ sg ¼ pℓγffiffiTffi. Substituting the above pffi s formula for T one obtains: 1=g ¼ pffiffiffi. Thus, unless the collisions C are nearly-elastic, the mean free time is of the same order as the macroscopic time scale, though smaller than unity. Furthermore, the change of velocity over a distance that equals the mean free path in the y direction is γℓ, i.e. of the order of the thermal speed, hence the gradient of the velocity is “large” compared to the pffi ℓ ∂Vx molecular case: p∂yffiffiTffi ¼ pffiffiCffi. The lack of strong scale separation demonstrated above is of very general nature in granular materials and has numerous consequences, part of which are mentioned below. First, since the (non-dimensionalized by the mean free path) gradients of the fields are typically not as tiny as in molecular gases, one may have to consider higher order terms in the gradient expansion. Returning to the example of simple shear flow, the ratio of the stress component, σxx, and the pressure, p (proportional to the trace of the stress tensor) is dimensionless, and must be a function of the dimensionless shear rate, pgℓffiffiTffi. Since the stress is invariant to the transformation γ→−γ (and assuming analyticity) 2 2 one obtains: rxx ¼ p 1 þ Cx gTℓ , to leadingnonvanishing order g2 ℓ 2 in γ, where Cx is a number. Similarly, ryy ¼ p 1 Cx T , since the average stress is the pressure. Note that the corrections are of second order in the gradients, hence they pertain to the Burnett order. The value of Cx is of the order of unity, e.g., for two dimensional disks Cx = 0.679 (see [15] and refs therein). Therefore: rrxxyy ¼ 22 1þCx g Tℓ 22 1Cx g Tℓ 2 2 . In air at STP conditions and a shear rate of 1 s− 1 the ratio gTℓ is of the order of 10−19 and thus the normal stress can justifiably be considered to be isotropic. In contrast, in a granular gas one obtains (after substituting the above formula for T ): pffi pffiffi 1þC x rxx ffi pC ryy ¼ 1Cx pffiffi , an O(1) entity that is typically significantly different C from unity. Indeed, normal stress differences have been observed in flowing granular materials and the above formula has been verified against simulations [15]. A curious consequence of the lack of scale separation is the finding (O. Herbst and I.G., unpublished) that granular shear flows have a sizeable streamwise heat flux (of the order of the spanwise heat flux), again – a Burnett contribution. This result has been quantitatively corroborated by molecular dynamics simulations. 134 I. Goldhirsch / Powder Technology 182 (2008) 130–136 The Burnett coefficients for molecular fluids of finite density (see [15] and references therein) are known to diverge (the higher order corrections to Navier–Stokes are non-analytic in the gradients) and even if one ignores this fact (it is ignored in the gas dynamic community that manages to use Burnett order hydrodynamics to model shocks) the Burnett equations are ill posed and do not have physically based boundary conditions. Some of these problems can be overcome (see the work by Jin and Slemrod cited in [15]) whereas others are still open. Whether the same problems afflict granular gases is yet unknown. Accepting the proven relevance of the Burnett contributions and the lack of strong scale separation as the reason for their importance, one may ask if even higher order terms in the gradient expansion (the next order after Burnett is known as super-Burnett) are important. Indeed such a case exists [36], but usually Burnett order hydrodynamics suffices for the description of the dynamics of granular gases, and often even Navier–Stokes hydrodynamics does well. While systems in which scale separation is weak or nonexistent are known (e.g., turbulent fluids, most rheological systems), their description often requires the use of scaledependent transport coefficients (such as the “eddy viscosity” of turbulent flows). Furthermore, the fields themselves (including the granular temperature) can be scale dependent, since they may significantly change on the scale of a mean free path. An analysis of this problem as it pertains to granular gases can be found in [37]. It seems that the problems that emanate from the lack of scale separation in granular gases are not as severe as those encountered in e.g., fluid mechanical turbulence. The reason is that granular gases do possess scale separation, weak as it may be. The description of steady states of granular gases may require, as mentioned, the use of corrections to the Navier– Stokes constitutive relations but this is merely a solvable technical problem. More serious problems may be encountered in the realm of the transient dynamics of granular gases, since in the latter case, due to the weak scale separation, the velocity distribution function may not “have sufficient time” to saturate to its hydrodynamic value; in particular, the hydrodynamic constitutive relations may not saturate. Consider, for instance, the heat flux. When there is strong scale separation (e.g, when the collisions are nearly-elastic) the heat flux induced by a temperature gradient is given by the corresponding hydrodynamic constitutive relation, since even when the external conditions (e.g., imposed gradients) change in time, the collisions are sufficiently frequent to establish a local distribution function that depends on the external parameters on time scales that are much shorter than the (typical) macroscopic time scales. When the temperature gradient changes “too rapidly” the heat flux cannot follow the change and need not be given by the hydrodynamic constitutive relation. One way of overcoming this problem is to take the heat flux as a dynamical variable (same for the stress tensor), as is done in some Grad-based theories, see e.g., [38] and references therein, or use an even more extended set of hydrodynamic fields. In this case one needs to modify the Chapman–Enskog expansion since the unresolved degrees of freedom (e.g., peculiar velocity fluctuations) are assumed to be enslaved to the extended (rather than the original) set of hydrodynamic fields. Another consequence of the weak scale separation becomes evident when one analyzes numerical simulations. As there need not be a significant plateau of scales for which the various fields are nearly scale-independent one may obtain results that are manifestly scale dependent [37]. This again does not pose a problem for the study of steady states, since then one can use small coarse graining volumes and large coarse graining times to obtain steady (and convergent) statistics. However, the description of transients requires special care and the establishment of connections with descriptions by ensembles is less trivial and not satisfactorily solved. It is possible that one may need to extend granular hydrodynamics in this case by using stochastic rather then deterministic equations (in addition to the extension of the set of fields). All of this really reinforces the notion that granular gases are rheological fluids (but not in the extreme sense of the term). 4. Clustering and collapse in granular gases The appearance of density microstructures in granular gases has been known since 1991 [12,15]. The clustering mechanism proposed in [39] has been verified in simulations, detailed theories and experiments [15]. This mechanism, often referred to as “collisional cooling”, is explained next. Consider a mass density fluctuation in a granular gas (as a many-body system, a granular gas experiences fluctuations of every density). In a region where the density is higher than in its ambient the rate of collisions (proportional to the square of the number density) is larger. Since the collisions are inelastic the energy density (or granular temperature) decreases in this region at a relatively rapid rate, causing a corresponding decrease in the pressure. The result is that mass flows from the relatively dilute ambient into the dense region further increasing its density until the process is stopped by other mechanisms. Clustering has been observed in granular shear flows, vibrated and other granular systems. One of the consequences of the clustering mechanism is that (at least) two different temperatures can coexist in the same system (in stark contrast to molecular systems where the temperature is usually a continuous field, except for the case of shocks). Numerous other consequences of clustering are mentioned in [15] including a curious phenomenon coined the “Maxwell demon” of granular gases. Since some clusters may be quite small in size they may be sub-resolution objects in coarse grained theories and contribute to the temperature and density fluctuations. A related but distinct phenomenon that characterizes (ideal) hard grains is that of collapse. Consider e.g., a ball dropped from rest at height h above a floor and let e be the coefficient of restitution between the ball and the floor. As shown in practically any elementary text on mechanics, the ball will reach a maximal height e2h after its collision with the floor. The sequence of heights, hn (the maximal height after hitting the floor n times) satisfies hn = e2hn − 1. Similarly the time that elapses between the instant at which the height of the ball is hn and that in which it is hn + 1 satisfies the recursion relation: τn +P 1 = eτn. Therefore the total time it takes the ball to fully relax s0 is: l n¼0 sn ¼ 1e. Since the coefficient of restitution is smaller I. Goldhirsch / Powder Technology 182 (2008) 130–136 than unity this time is finite but the number of collisions of the ball with the floor is infinite. A similar collapse phenomenon can happen in a granular gas (see [15] and references therein). This fact leads to some problems in simulations of hard (but not soft) sphere gases. The collapse phenomenon can be preceded by clustering [15] and it is one of the reasons that Kadanoff expressed doubts about kinetic descriptions of granular matter. As mentioned, collapse does not truly occur in soft sphere systems and it basically does not occur in forced three dimensional granular gases either. Therefore it does not give rise to a serious problem in the application of granular kinetics. 5. Some misconceptions concerning granular temperature The following comprise a few prevalent misconceptions concerning the granular temperature: (1) The velocity distribution needs to be normal or Maxwellian for the granular temperature to “make sense”. This is not true since the distribution is only Maxwellian in equilibrium. As shown above, granular gases are always in nonequilibrium states. Furthermore, both measured velocity distributions and calculated ones (e.g., in [28]) are not Maxwellian. (2) The distribution function needs to be near-Maxwellian for the kinetic approach to be valid. This is not correct either. It has been shown in [28] that one could use kinetic theory to obtain accurate results even when the distribution function is far from Maxwellian or when one expands around a Maxwellian corresponding to a “wrong” temperature (i.e. one that does not equal the actual temperature at the considered position). (3) The “correct” temperature of a granular fluid is really a tensor and the temperature is usually anisotropic. This misconception seems to have its roots in a model [40] for the distribution of the granular peculiar velocities of the 1 form e2uKu where K is a matrix, following a similar usage by astrophysicists. It is easy to see that (since the kinetic stress tensor σkin is proportional to K−1) all that this statement says is that the kinetic stress is a tensor and it is usually anisotropic. This anisotropy is obtained from kinetic theories without the need to model the distribution function as a “generalized Maxwellian” and it does not imply that one cannot employ the “usual” scalar granular temperature even when the stress tensor is “very anisotropic”. (4) Some kinetic theories account for equipartition whereas others don't. This is true for the old kinetic theories in which the form of the distribution functions was conjectured. It is not true for the modern systematic kinetic approaches in which the distribution function is derived, as explained above. (5) The concept of granular temperature or kinetic theory does not apply to dense granular gases. Not true. The Enskog–Boltzmann equation has been used to obtain hydrodynamic constitutive relations for moderately dense granular gases and these have been successfully applied even to “very dense” granular gases [22–24]. 135 (6) Fluctuations in granular systems are large and there is a lack of scale separation, therefore describing them using kinetic theory is useless. Scale separation is weak in granular gases but it does exist. Therefore Burnett contributions are important, as explained above, but this does not disqualify the kinetic approach. (7) “Granular hydrodynamics does not describe some experiments/simulations”. Though no theory comes with a warranty, there seems to be a “rich man's problem” in this field, namely that due to the fact that the grains are macroscopic and one can often measure the dynamics of each and every grain, hydrodynamic theories are often expected to describe scales that they are not designed to model or resolve. There are though (surprising) examples in which granular hydrodynamics provides good descriptions of systems that are only five grains deep (see e.g. the paper by Forterre and Pouliquen cited in [15]). In addition the reader should be warned that many (usually approximate) statistical approaches to granular gases call themselves kinetic theoretical. While there is nothing wrong with attempts to derive simplified theories (mostly of the mean field variety), the possible failings of some of these models should not be used as arguments against kinetic theory. 6. Conclusion The granular temperature comprises an important characterization of granular fluids and is central to their kinetic and hydrodynamic descriptions. Due to the lack of strong scale separation one needs to go beyond the Navier–Stokes level of description and this raises certain problems that require further study. A good description of the transient dynamics of granular fluids is yet to be developed though it is possible that the presently available hydrodynamic equations are good enough for the description of (at least) part of the transient dynamics. An improved description of transient granular dynamics will most likely require the extension of the set of hydrodynamic fields. This problem clearly needs further study as well. The description of dense granular fluids is far from complete and requires, among other things, a better elucidation of the nature of precollisional correlations. The possibility that one may need to employ stochastic equations to describe granular flows has been raised. The issue of granular fluids in which persistent contacts are prevalent (and the interactions are not collisional) has not been discussed here. This is an important problem of much current interest. Finally, it is worth mentioning the existence of relatively recent theories that are based on a proposal by Edwards and co-workers, which define an effective temperature for static granular assemblies (see the paper by Serero et al in this issue and references therein). While these theories have not been related heretofore to the now classical granular temperature (which is kinetic in nature), experiments indicate that the Edwards temperature can be used in fluctuation–dissipation relations, thereby providing a link between this new temperature and granular dynamics. Whether this approach will lead to a uniform description of granular dynamics and/or a universal definition of granular temperature remains to be seen. 136 I. Goldhirsch / Powder Technology 182 (2008) 130–136 References [1] A. Einstein, Investigations on the theory of the Brownian movement, Dover, NY, 1956, p. 3. [2] S. Ogawa, Multitemperature theory of granular materials, in: S.C. Cowin, M. Satake (Eds.), Proc. U. S.-Japan Symp. on Continuum Mechanics and Statistical Approaches in the Mechanics of Granular Materials, Gakujutsu Bunken Fukyu-kai, 1978, pp. 208–217. [3] S. Ogawa, A. Umemura, N. Oshima, On the equations of fully fluidized granular materials, ZAMP 31 (1980) 483–493. [4] D.F. McTigue, A model for stresses in shear flow of granular material, in: S.C. Cowin, M. Satake (Eds.), Proc. U. S.-Japan Symp. on Continuum Mechanics and Statistical Approaches in the Mechanics of Granular Materials, Gakujutsu Bunken Fukyu-kai, 1978, pp. 266–271. [5] A. Blinowski, On the dynamic flow of granular media, Arch. Mech. 30 (1978) 27–34. [6] J.T. Jenkins, S.C. Cowin, Theories for flowing granular materials, in: S.C. Cowin (Ed.), Mechanics Applied to the Transport of Bulk Materials, A. S. M. E. AMD-31, 1979, pp. 79–89. [7] N.L. Ackerman, H.H. Shen, Flow of granular material as a two-component system, in: S.C. Cowin, M. Satake (Eds.), Proc. U. S.-Japan Symp. on Continuum Mechanics and Statistical Approaches in the Mechanics of Granular Materials, Gakujutsu Bunken Fukyu-kai, 1978, pp. 258–265. [8] S.B. Savage, D.J. Jeffrey, The stress tensor in a granular flow at high shear rates, J. Fluid Mech. 110 (1981) 255–272. [9] J.T. Jenkins, S.B. Savage, A theory for the rapid flow of identical, smooth, nearly elastic, spherical particles, J. Fluid Mech. 130 (1983) 187–202. [10] P.K. Haff, Grain flow as a fluid-mechanical phenomenon, J. Fluid Mech. 134 (1983) 401–430. [11] M. Babic, H.H. Shen, A simple mean free path theory for the stresses in a rapidly deforming granular material, J. Eng. Mech. ASCE 115 (1989) 1262–1282. [12] C.S. Campbell, Rapid granular flows, Annu. Rev. Fluid Mech. 22 (1990) 57–92. [13] N. Sela, I. Goldhirsch, Hydrodynamic equations for rapid flows of smooth inelastic spheres, to Burnett order, J. Fluid Mech. 361 (1998) 41–74. [14] J.J. Brey, J.W. Dufty, C.S. Kim, A. Santos, Hydrodynamics of granular flow at low density, Phys. Rev., E Stat. Phys. Plasmas Fluids Relat. Interdiscip. Topics 58 (1998) 4638–4653. [15] I. Goldhirsch, Rapid granular flows, Annu. Rev. Fluid Mech. 35 (2003) 267–293. [16] A. Levy, H. Kalman (Eds.), Handbook of conveying and handling of particulate solids, Elsevier, 2001. [17] L.P. Kadanoff, Built upon sand: theoretical ideas inspired by granular flows, Rev. Mod. Phys. 71 (1999) 435–444. [18] C.S. Campbell, Granular material flows – an overview, Powder Technol. 162 (3) (2006) 208–229. [19] P. Evesque, Are temperature and other thermodynamic variables efficient concepts for describing granular gases and/or flows? Poudres & Grains 13 (2) (2002) 20–26. [20] K. Hutter, K.R. Rajagopal, On flows of granular materials, Contin. Mech. Thermodyn. 6 (1994) 81–139. [21] I. Goldhirsch, Granular gases: probing the boundaries of hydrodynamics, in: T. Pöschel, S. Luding (Eds.), Granular Gases, Springer, 2001, pp. 79–99. [22] E. Khain, B. Meerson, Shear induced crystallization of a dense granular flow: hydrodynamics beyond the melting point, Phys. Rev., E Stat. Phys. Plasmas Fluids Relat. Interdiscip. Topics 73 (6) (2006) article no. 061301. [23] E. Khain, Hydrodynamics of fluid-solid coexistence in dense shear granular flow, Phys. Rev., E Stat. Phys. Plasmas Fluids Relat. Interdiscip. Topics 75 (5) (2007) article no. 051310. [24] V. Kumaran, The constitutive relation for the granular flow of rough particles, and its application to the flow down an inclined plane, J. Fluid Mech. 561 (2006) 1–42. [25] G. D'Anna, P. Mayor, A. Barrat, V. Loreto, F. Nori, Observing Brownian motion in vibration fluidized granular matter, Nature 424 (2003) 909–912. [26] S. Chapman, T.G. Cowling, The mathematical theory of nonuniform gases, Cambridge University Press, 1970. [27] D.K. Yoon, J.T. Jenkins, Kinetic theory for identical, frictional, nearly elastic disks, Phys. Fluids 17 (8) (2005) article no. 083301. [28] S.H. Noskowicz, O. Bar-Lev, D. Serero, I. Goldhirsch, Computer-aided kinetic theory, Europhys. Lett. 79 (2007) article no. 60001 and unpublished results on two and three dimensional dilute and dense granular gases. [29] J.J. Brey, M.J. Ruiz-Montero, F. Moreno, Hydrodynamics of an open vibrated granular system, Phys. Rev., E Stat. Phys. Plasmas Fluids Relat. Interdiscip. Topics 63 (6) (2001) article no. 061305. [30] V. Garzó, J.W. Dufty, Dense fluid transport for inelastic hard spheres, Phys. Rev., E Stat. Phys. Plasmas Fluids Relat. Interdiscip. Topics 59 (5) (1999) 5895–5911. [31] I. Goldhirsch, T.P.C. van Noije, Green–Kubo relations for granular fluids, Phys. Rev., E Stat. Phys. Plasmas Fluids Relat. Interdiscip. Topics 61 (3) (2000) 3241–3244. [32] J.J. Brey, M.J. Ruiz-Montero, P. Maynar, M.I.G. de Soria, Hydrodynamic modes, Green–Kubo relations, and velocity correlations in dilute granular gases, J. Phys., Condens. Matter 17 (24) (2005) 2489–2502. [33] T.P.C. van Noije, M.H. Ernst, R. Brito, Ring kinetic theory for an idealized granular gas, Physica A 251 (1–2) (1998) 266–283. [34] R. Soto, M. Mareschal, Statistical mechanics of fluidized granular media: short-range velocity correlations, Phys. Rev., E Stat. Phys. Plasmas Fluids Relat. Interdiscip. Topics 63 (2001) 1–9. [35] M.-L. Tan, I. Goldhirsch, Rapid granular flows as mesoscopic systems, Phys. Rev. Lett. 81 (1998) 3022–3025. [36] C.M. Hrenya, J.E. Galvin, R.D. Wildman, Evidence of higher-order effects in thermally-driven, rapid granular flows, J. Fluid Mech.,(in press). [37] B.J. Glasser, I. Goldhirsch, Scale dependence, correlations and fluctuations of stresses in rapid granular flows, Phys. Fluids 13 (2001) 407–420. [38] R. Ramirez, D. Risso, R. Soto, P. Cordero, Hydrodynamic theory for granular gases, Phys. Rev., E Stat. Phys. Plasmas Fluids Relat. Interdiscip. Topics 62 (2) (2000) 2521–2530. [39] I. Goldhirsch, G. Zanetti, Clustering instability in dissipative gases, Phys. Rev. Lett. 70 (1993) 1619–1622. [40] J.T. Jenkins, M.W. Richman, Plane simple shear of smooth inelastic circular disks: the anisotropy of the second moment in the dilute and dense limits, J. Fluid Mech. 192 (1988) 313–328.