1: \documentclass[pre,twocolumn,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
2:
3: \usepackage{graphicx}% Include figure files
4: \usepackage{dcolumn}% Align table columns on decimal point
5: \usepackage{bm}% bold math
6:
7: \begin{document}
8: \title{Phase separation of a driven granular gas in
9: annular geometry}% Force line breaks with \\
10: \author{Manuel D{\'{\i}}ez-Minguito}
11: \affiliation{Institute `Carlos I' for Theoretical and Computational
12: Physics, University of Granada, E-18071 - Granada, Spain}
13: \author{Baruch Meerson}
14: \affiliation{Racah Institute of Physics, Hebrew University of
15: Jerusalem, Jerusalem 91904, Israel}
16: \date{\today}
17:
18: \begin{abstract}
19: This work investigates phase separation of a monodisperse gas of inelastically
20: colliding hard disks confined in a two-dimensional annulus, the inner circle of
21: which represents a ``thermal wall". When described by granular hydrodynamic
22: equations, the basic steady state of this system is an azimuthally symmetric
23: state of increased particle density at the exterior circle of the annulus. When
24: the inelastic energy loss is sufficiently large, hydrodynamics predicts
25: spontaneous symmetry breaking of the annular state, analogous to the van der
26: Waals-like phase separation phenomenon previously found in a driven granular gas
27: in rectangular geometry. At a fixed aspect ratio of the annulus, the phase
28: separation involves a ``spinodal interval" of particle area fractions, where the
29: gas has negative compressibility in the azimuthal direction. The heat conduction
30: in the azimuthal direction tends to suppress the instability, as corroborated by
31: a marginal stability analysis of the basic steady state with respect to small
32: perturbations. To test and complement our theoretical predictions we performed
33: event-driven molecular dynamics (MD) simulations of this system. We clearly
34: identify the transition to phase separated states in the MD simulations, despite
35: large fluctuations present, by measuring the probability distribution of the
36: amplitude of the fundamental Fourier mode of the azimuthal spectrum of the
37: particle density. We find that the instability region, predicted from
38: hydrodynamics, is always located within the phase separation region observed in
39: the MD simulations. This implies the presence of a binodal (coexistence) region,
40: where the annular state is metastable. The phase separation persists when the
41: driving and elastic walls are interchanged, and also when the elastic wall is
42: replaced by weakly inelastic one.
43: \end{abstract}
44:
45: \pacs{45.70.Qj}
46:
47: \maketitle
48:
49: \section{\label{SecI}Introduction}
50: Flows of granular materials %, an ensemble of macroscopic particles colliding inelastically,
51: are ubiquitous in nature and technology \cite{jaeger}. Examples are numerous and
52: range from Saturn's rings to powder processing. Being dissipative and therefore
53: intrinsically far from thermal equilibrium, granular flows exhibit a plethora of
54: pattern forming instabilities \cite{ristow,aranson+tsimring}. In spite of a
55: surge of recent interest in granular flows, their quantitative modeling remains
56: challenging, and the pattern forming instabilities provide sensitive tests to
57: the models. This work focuses on the simple model of \textit{rapid} granular
58: flows, also referred to as granular gases: large assemblies of inelastically
59: colliding hard spheres
60: \cite{campbell,kadanoff,thorsten1,thorsten2,goldhirsch1,brilliantov}. In the
61: simplest version of this model the only dissipative effect taken into account is
62: a reduction in the relative normal velocity of the two colliding particles,
63: modeled by the coefficient of normal restitution, see below. Under some
64: additional assumptions a hydrodynamic description of granular gases becomes
65: possible. The Molecular Chaos assumption allows for a description in terms of
66: the Boltzmann or Enskog equations, properly generalized to account for the
67: inelasticity of particle collisions, followed by a systematic derivation of
68: hydrodynamic equations \cite{sela,brey1,lutsko}. For inhomogeneous (and/or
69: unsteady) flows hydrodynamics demands scale separation: the mean free path of
70: the particles (respectively, the mean time between two consecutive collisions)
71: must be much less than any characteristic length (respectively, time) scale that
72: the hydrodynamic theory attempts to describe. The implications of these
73: conditions can be usually seen only \textit{a posteriori}, after the
74: hydrodynamic problem in question is solved, and the hydrodynamic length/time
75: scales are determined. We will restrict ourselves in this work to nearly
76: elastic collisions and moderate gas densities where, based on previous studies,
77: hydrodynamics is expected to be an accurate leading order theory
78: \cite{campbell,kadanoff,thorsten1,thorsten2,goldhirsch1,brilliantov}. These
79: assumptions allow for a detailed quantitative study (and prediction) of a
80: variety of pattern formation phenomena in granular gases. One of these phenomena
81: is the phase separation instability, first predicted in Ref. \cite{livne1} and
82: further investigated in Refs.
83: \cite{argentina,brey2,khain1,livne2,baruch2,khain2}. This instability arises
84: already in a very simple, indeed prototypical setting: a monodisperse granular
85: gas at zero gravity confined in a rectangular box, one of the walls of which is
86: a ``thermal" wall. The basic state of this system is the stripe state. In the
87: hydrodynamic language it represents a laterally uniform stripe of increased
88: particle density at the wall opposite to the driving wall. The stripe state was
89: observed in experiment \cite{kudrolli}, and this and similar settings have
90: served for testing the validity of quantitative modeling
91: \cite{kadanoff2,esipov,grossman}. It turns out that (i) within a ``spinodal"
92: interval of area fractions and (ii) if the system is sufficiently wide in the
93: lateral direction, the stripe state is unstable with respect to small density
94: perturbations in the lateral direction \cite{livne1,brey2,khain1}. Within a
95: broader ``binodal" (or coexistence) interval the stripe state is stable to small
96: perturbations, but unstable to sufficiently large ones \cite{argentina,khain2}.
97: In both cases the stripe gives way, usually via a coarsening process, to
98: coexistence of dense and dilute regions of the granulate (granular ``droplets"
99: and ``bubbles") along the wall opposite to the driving wall
100: \cite{argentina,livne2,khain2}. This far-from-equilibrium phase separation
101: phenomenon is strikingly similar to a gas-liquid transition as described by the
102: classical van der Waals model, except for large fluctuations observed in a broad
103: region of aspect ratios around the instability threshold \cite{baruch2}. The
104: large fluctuations have not yet received a theoretical explanation.
105:
106: This work addresses a phase separation process in a different geometry. We will
107: deal here with an assembly of hard disks at zero gravity, colliding
108: inelastically inside a two-dimensional annulus. The interior wall of the annulus
109: drives the granulate into a non-equilibrium steady state with a
110: (hydrodynamically) zero mean flow. Particle collisions with the exterior wall
111: are assumed elastic. The basic steady state of this system, as predicted by
112: hydrodynamics, is the \textit{annular} state: an azimuthally symmetric state of
113: increased particle density at the exterior wall. The phase separation
114: instability manifests itself here in the appearance of dense clusters with
115: broken azimuthal symmetry along the exterior wall. Our main objectives are to
116: characterize the instability and compute the phase diagram by using granular
117: hydrodynamics (or, more precisely, granular hydrostatics, see below) and event
118: driven molecular dynamics simulations. By focusing on the annular geometry, we
119: hope to motivate experimental studies of the granular phase separation which may
120: be advantageous in this geometry. The annular setting avoids lateral side walls
121: (with an unnecessary/unaccounted for energy loss of the particles). Furthermore,
122: driving can be implemented here by a rapid rotation of the (slightly eccentric
123: and possibly rough) interior circle.
124:
125: We organized the paper as follows. Section \ref{SecII} deals with a hydrodynamic
126: description of the annular state of the gas. As we will be dealing only with
127: states with a zero mean flow, we will call the respective equations hydrostatic.
128: A marginal stability analysis predicts a spontaneous symmetry breaking of the
129: annular state. We compute the marginal stability curves and compare them to the
130: borders of the spinodal (negative compressibility) interval of the system. In
131: Section \ref{SecIII} we report event-driven molecular dynamics (MD) simulations
132: of this system and compare the simulation results with the hydrostatic theory.
133: In Section \ref{SecIV} we discuss some modifications of the model, while Section
134: \ref{SecV} contains a summary of our results.
135:
136:
137: \section{\label{SecII} Particles in an annulus and granular hydrostatics}
138: \textit{The density equation.} Let $N$ hard disks of diameter $d$ and mass $m=1$
139: move, at zero gravity, inside an annulus of aspect ratio $\Omega=
140: R_{\text{ext}}/R_{\text{int}}$, where $R_{\text{ext}}$ is the exterior radius
141: and $R_{\text{int}}$ is the interior one. The disks undergo inelastic collisions
142: with a constant coefficient of normal restitution
143: $\mu$. For simplicity, we neglect the rotational degree of freedom of the particles. %In each
144: %inter-particle collision the kinetic energy is continuously transferred into
145: %heat, while the momentum is conserved.
146: The (driving) interior wall is modeled by a thermal wall kept at temperature
147: $T_{0}$, whereas particle collisions with the exterior wall are considered
148: elastic. The energy transferred from the thermal wall to the granulate
149: dissipates in the particle inelastic collisions, and we assume that the system
150: reaches a (non-equilibrium) steady state with a zero mean flow. We restrict
151: ourselves to the nearly elastic limit by assuming a restitution coefficient
152: close to, but less than, unity: $1-\mu \ll 1$. This allows us to safely use
153: granular hydrodynamics \cite{goldhirsch1}. For a zero mean flow steady state
154: the continuity equation is obeyed trivially, while the momentum and energy
155: equations yield two \textit{hydrostatic} relations:
156: \begin{equation}
157: \nabla \cdot \mathbf{q} (\mathbf{r})+I=0\,, \; \; \; p=const\,,
158: \label{en_balance_prev}
159: \end{equation}
160: where $\mathbf{q}$ is the local heat flux, $I$ is the energy loss term due to
161: inelastic collisions, and $P=P(n,T)$ is the gas pressure that depends on the
162: number density $n(\mathbf{r})$ and granular temperature $T(\mathbf{r})$. We
163: adopt the classical Fourier relation for the heat flux
164: $\mathbf{q}(\mathbf{r})=-\kappa \nabla T(\mathbf{r})$ (where $\kappa$ is the
165: thermal conductivity), omitting a density gradient term. In the dilute limit
166: this term was derived in Ref. \cite{brey1}. It can be neglected in the nearly
167: elastic limit which is assumed throughout this paper.
168:
169: The momentum and energy balance equations read
170: \begin{equation}
171: \nabla \cdot \left[ \kappa \nabla T(\mathbf{r}) \right] =I\,, \; \; \;
172: p=const\,, \label{en_balance}
173: \end{equation}
174: To get a closed formulation, we need constitutive relations for $p(n,T)$,
175: $\kappa(n,T)$ and $I(n,T)$. We will employ the widely used semi-empiric
176: transport coefficients derived by Jenkins and Richman \cite{jenkins} for
177: moderate densities:
178: \begin{equation}
179: \begin{split}
180: \kappa=\frac{2d n T^{1/2} \tilde{G}}{\pi^{1/2}}
181: \left[ 1+\frac{9\pi}{16} \left( 1+\frac{2}{3\tilde{G}}\right)^{2}\right],\\
182: I=\frac{8(1-\mu)n T^{3/2}\tilde{G}}{d\sqrt{\pi}}\,,
183: \end{split}
184: \label{JR}
185: \end{equation}
186: and the equation of state first proposed by Carnahan and Starling \cite{carnahan}
187: \begin{equation}
188: p=n T(1+2\tilde{G})\,,
189: \label{CS}
190: \end{equation}
191: where $\tilde{G}=\nu (1-\frac{7\nu}{16})/(1-\nu)^{2}$ and $\nu=n \left( \pi
192: d^{2}/4 \right) $ is the solid fraction. Let us rescale the radial coordinate by
193: $R_{\text{int}}$ and introduce the rescaled inverse density
194: $Z(r,\theta)=n_{c}/n(r,\theta)$, where $n_{c}=2/\left( \sqrt{3}d^2 \right)$ is
195: the hexagonal close packing density. The rescaled radial coordinate $r$ now
196: changes between $1$ and $\Omega\equiv R_{\text{ext}}/R_{\text{int}}$, the aspect
197: ratio of the annulus. As in the previous work \cite{khain1}, Eqs.
198: (\ref{en_balance}), (\ref{CS}) and (\ref{JR}) can be transformed into a single
199: equation for the inverse density $Z(r)$:
200: \begin{equation}
201: \nabla \cdot \left[ \mathcal{F}(Z)\nabla Z\right] =\Lambda \mathcal{Q}(Z)\,,
202: \label{eq:density_eq}
203: \end{equation}%
204: where
205: \begin{equation}
206: \begin{split}
207: \mathcal{F}(Z)=\mathcal{F}_{1}(Z)\mathcal{F}_{2}(Z),\\
208: \mathcal{Q}(Z)=\frac{6}{\pi} \frac{Z^{1/2}\mathcal{G}}{(1+2\mathcal{G})^{3/2}},\\
209: \mathcal{F}_{1}(Z)=\frac{\mathcal{G}(Z)\left[ 1+{\frac{9\pi}{16}}
210: \left( 1+\frac{2}{3\mathcal{G}}\right)^2\right]}{Z^{1/2}(1+2\mathcal{G})^{5/2}},\\
211: \mathcal{F}_{2}(Z)=1+2\mathcal{G}+ \frac{\pi}{\sqrt{3}}\frac{Z\left(
212: Z+\frac{\pi}{16\sqrt{3}}\right) }
213: {\left( Z-\frac{\pi}{2\sqrt{3}}\right)^3},\\
214: \mathcal{G}(Z)=\frac{\pi}{2\sqrt{3}}\frac{\left(
215: Z-\frac{7\pi}{32\sqrt{3}}\right)} {\left( Z-\frac{\pi}{2\sqrt{3}}\right) ^2}.
216: \end{split}
217: \label{funct}
218: \end{equation}
219: The dimensionless parameter $\Lambda \equiv (2\pi/3) (1-\mu)
220: \left(R_{\text{int}}/d\right)^2$ is the hydrodynamic inelastic loss parameter.
221: The boundary conditions for Eq.~(\ref{eq:density_eq}) are
222: \begin{equation}
223: \partial Z(1,\theta)/\partial\theta=0 \quad \text{and} \quad \nabla_{n} Z(\Omega,\theta)=0\,,
224: \label{eq:boundary}
225: \end{equation}
226: The first of these follows from the constancy of the temperature at the
227: (thermal) interior wall which, in view of the constancy of the pressure in a
228: steady state, becomes constancy of the density. The second condition demands a
229: zero normal component of the heat flux at the elastic wall. Finally, working
230: with a fixed number of particles, we demand the normalization condition
231: \begin{equation}
232: \int_{0}^{2\pi} \int_{1}^{\Omega} Z^{-1} (r,\theta) r dr d\theta = \pi f
233: (\Omega^2-1)\,, \label{eq:norma}
234: \end{equation}
235: where $$f=\frac{N}{\pi n_{c}R_{\text{int}}^{2}(\Omega^2-1)}$$ is the area
236: fraction of the grains in the annulus. Equations
237: (\ref{eq:density_eq})-(\ref{eq:norma}) determine all possible steady state
238: density profiles, governed by three dimensionless parameters: $f$, $\Lambda$,
239: and $\Omega$.\\
240:
241: \begin{figure}
242: \includegraphics[width=7.5278cm]{fig1.eps}% Here is how to import EPS art
243: \caption{\label{fig1} The normalized density profiles obtained from the MD
244: simulations (the dots) and hydrostatics (the line) for $\Omega=2$,
245: $\Lambda=81.09$, and $f=0.356$ (equivalently, $z_{\Omega}=2.351$). The
246: simulations were carried out with $N=1250$ particles, $\mu=0.92$, and
247: $R_{\text{int}}=22.0$. Also shown is a typical snapshot of the system at the
248: steady state as observed in the MD simulation.}
249: \end{figure}
250:
251: \textit{Annular state.} The simplest solution of the density equation
252: (\ref{eq:density_eq}) is azimuthally symmetric ($\theta$-independent): $Z=
253: z(r)$. Henceforth we refer to this basic state of the system as the
254: \textit{annular state}. It is determined by the following equations:
255: \begin{equation}
256: \begin{split}
257: \left[ r \mathcal{F}(z) z^{\prime} \right]^{\prime} =
258: r \Lambda \mathcal{Q}(z),\; z^{\prime}(\Omega)=0, \; \text{and} \\
259: \int_{1}^{\Omega} z^{-1} r dr=(\Omega^{2}-1)f/2 \,,
260: \end{split}
261: \label{eq:annular}
262: \end{equation}
263: where the primes denote $r$-derivatives. In order to solve the second order
264: equation (\ref{eq:annular}) numerically, one can prescribe the inverse density
265: at the elastic wall $z_{\Omega}\equiv z(\Omega)$. Combined with the no-flux condition at
266: $r=\Omega$, this condition define a Cauchy problem for $z(r)$
267: \cite{livne2,khain1}. Solving the Cauchy problem, one can compute the respective
268: value of $f$ from the normalization condition in Eq.~(\ref{eq:annular}). At
269: fixed $\Lambda$ and $\Omega$, %the function $z_{\Omega}$ is a monotonically decreasing function of $f$.
270: there is a one-to-one relation between $z_{\Omega}$ and $f$. Therefore, an alternative parameterization of the
271: annular state is given by the scaled numbers $z_{\Omega}$, $\Lambda$, and
272: $\Omega$. The same is true for the marginal stability analysis performed in the
273: next subsection.
274:
275: Figure~\ref{fig1} depicts an example of annular state that we found numerically.
276: One can see that the gas density increases with the radial coordinate, as
277: expected from the temperature decrease via inelastic losses, combined with the
278: constancy of the pressure throughout the system. The hydrodynamic density
279: profile agrees well with the one found in our MD simulations, see below.
280:
281: \begin{figure}
282: \includegraphics[width=7.5278cm]{fig2.eps}% Here is how to import EPS art
283: \caption{\label{fig2} The main graph: the marginal stability curves $k=k(f)$
284: (where $k$ is an integer) for $\Omega=1.5$ and $\Lambda=10^4$ (circles),
285: $\Lambda=5\times 10^4$ (squares), and $\Lambda=10^5$ (triangles). For a given
286: $\Lambda$ the annular state is stable above the respective curve and unstable
287: below it, as indicated for $\Lambda=10^4$. As $\Lambda$ increases the
288: marginal stability interval shrinks. %The lines connecting dots serve to guide the eye.
289: The inset: the dependence of $k_{max}$ on $\Lambda^{1/2}$. The straight line
290: shows that, at large $\Lambda$, $k_{max}\propto \Lambda^{1/2}$.}
291: \end{figure}
292:
293: \textit{Phase separation.} Mathematically, phase separation manifests itself in
294: the existence of \textit{additional} solutions to
295: Eqs.~(\ref{eq:density_eq})-(\ref{eq:norma}) in some region of the parameter
296: space $f$, $\Lambda$, and $\Omega$. These additional solutions are \textit{not}
297: azimuthally symmetric. Solving Eqs.~(\ref{eq:density_eq})-(\ref{eq:norma}) for
298: fully two-dimensional solutions is not easy \cite{livne1}. One class of such
299: solutions, however, bifurcate continuously from the annular state, so they can
300: be found by linearizing Eq.~(\ref{eq:density_eq}), as in rectangular geometry
301: \cite{livne1,khain1}. In the framework of a time-dependent hydrodynamic
302: formulation, this analysis corresponds to a \textit{marginal stability} analysis
303: which involves a small perturbation to the annular state. For a single azimuthal
304: mode $\sim \sin(k\theta)$ (where $k$ is integer) we can write
305: $Z(r,\theta)=z(r)+\varepsilon\; \Xi(r) \sin(k\theta)$, where $\Xi(r)$ is a
306: smooth function, and $\varepsilon \ll 1$ a small parameter. Substituting this
307: into Eq.~(\ref{eq:density_eq}) and linearizing the resulting equation yields a
308: $k$-dependent second order differential equation for the function
309: $\Gamma(r)\equiv\mathcal{F}[Z(r)]\,\Xi(r)$:
310: \begin{equation}
311: \Gamma^{\prime\prime}_k+\frac{1}{r}\Gamma^{\prime}_k- \left(
312: \frac{k^{2}}{r^{2}}+\frac{\Lambda \mathcal{Q}^{\prime}(Z)}
313: {\mathcal{F}(Z)}\right) \Gamma_{k}=0\,. \label{eq:MSA}
314: \end{equation}%
315: This equation is complemented by the boundary conditions
316: \begin{equation}
317: \Gamma(1)=0 \;\;\; \mbox{and} \;\;\; \Gamma^{\prime}(\Omega)=0\,.
318: \label{eq:MSA_BC}
319: \end{equation}%
320: For fixed values of the scaled parameters $f$, $\Lambda$, and $\Omega$,
321: Eqs.~(\ref{eq:MSA}) and (\ref{eq:MSA_BC}) determine a linear eigenvalue problem
322: for $k$. Solving this eigenvalue problem numerically, one obtains the marginal
323: stability hypersurface $k=k(f,\Lambda,\Omega)$. For fixed $\Lambda$ and
324: $\Omega$, we obtain a marginal stability curve $k=k(f)$. Examples of such
325: curves, for a fixed $\Omega$ and three different $\Lambda$ are shown in
326: Fig.~\ref{fig2}. Each $k=k(f)$ curve has a maximum $k_{max}$, so that a density
327: modulation with the azimuthal wavenumber larger than $k_{max}$ is stable. As
328: expected, the instability interval is the largest for the fundamental mode
329: $k=1$. The inset in Fig.~\ref{fig2} shows the dependence of $k_{max}$ on
330: $\Lambda^{1/2}$. The straight line shows that, at large $\Lambda$,
331: $k_{max}\propto \Lambda^{1/2}$, as in rectangular geometry \cite{khain1}.
332:
333: Two-dimensional projections of the ($f$, $\Lambda$, $\Omega$)-phase diagram at
334: three different $\Omega$ are shown in Fig.~\ref{fig3} for the fundamental mode.
335: The annular state is unstable in the region bounded by the marginal stability
336: curve, and stable elsewhere. Therefore, the marginal stability analysis predicts
337: loss of stability of the annular state within a finite interval of $f$, that is
338: at $f_{min}(\Lambda,\Omega)<f<f_{max}(\Lambda,\Omega)$.
339:
340: \begin{figure}
341: \includegraphics[width=7.5278cm]{fig3.eps}% Here is how to import EPS art
342: \caption{\label{fig3} Two-dimensional projections on the $[\Lambda$,
343: $f\,(\Omega^{2}-1)/2]$ plane of the phase diagram at $\Omega=1.5$ (the solid
344: line), $\Omega=3$ (the dotted line), and $\Omega=5$ (the dashed line). The inset
345: shows more clearly, for $\Omega=3$, that the marginal stability curve (the solid
346: line) lies within the negative compressibility region (bounded by the dashed
347: line).}
348: \end{figure}
349:
350: The physical mechanism of this phase separation \textit{instability} is the
351: negative compressibility of the granular gas in the azimuthal direction, caused
352: by the inelastic energy loss. To clarify this point, let us compute the
353: \textit{pressure} of the annular state, given by Eq. (\ref{CS}). First we
354: introduce a rescaled pressure $P=p/(n_{c}T_{0})$ and, in view of the pressure
355: constancy in the annular state, compute it at the thermal wall, where $T=T_{0}$
356: is prescribed and $z(1)$ is known from our numerical solution for the annular
357: state. We obtain
358: $$
359: P(f,\Lambda,\Omega)=\frac{1+2\mathcal{G}(z(1))}{z(1)}\,.
360: $$
361: The spinodal (negative compressibility) region is determined by the necessary
362: condition for the \textit{instability}: $\left(
363: \partial P/\partial f\right)_{\Lambda,\Omega}<0$, whereas the borders of the
364: spinodal region are defined by $\left( \partial P/\partial
365: f\right)_{\Lambda,\Omega}=0$. Typical $P(f)$ curves for a fixed $\Omega$ and
366: several different $\Lambda$ are shown in Fig.~\ref{fig4}. One can see that, at
367: sufficiently large $\Lambda$, the rescaled pressure $P$ goes down with an
368: increase of $f$ at an interval $f_{1}<f<f_{2}$. That is, the effective
369: compressibility of the gas with respect to a redistribution of the material in
370: the azimuthal direction is negative on this interval of area fractions. By
371: joining the spinodal points $f_1$ and (separately) $f_2$ at different $\Lambda$,
372: we can draw the spinodal line for a fixed $\Omega$. As $\Lambda$ goes down, the
373: spinodal interval shrinks and eventually becomes a point at a critical point
374: $(P_{c},f_c)$, or $(\Lambda_{c},f_c)$ (where all the critical values are
375: $\Omega$-dependent). For $\Lambda<\Lambda_c$ $P(f)$ monotonically increases and
376: there is no instability.
377:
378: What is the relation between the spinodal interval $(f_{1},f_{2})$ and the
379: marginal stability interval $(f_{min},f_{max})$? These intervals would coincide
380: were the azimuthal wavelength of the perturbation infinite (or, equivalently,
381: $k\rightarrow 0$), so that the azimuthal heat conduction would vanish. Of
382: course, this is not possible in annular geometry, where $k\ge 1$. As a result,
383: the negative compressibility interval must include in itself the marginal
384: stability interval $(f_{min},f_{max})$. This is what our calculations indeed
385: show, see the inset of Fig.~\ref{fig3}. That is, a negative compressibility is
386: necessary, but not sufficient, for instability, similarly to what was found in
387: rectangular geometry \cite{khain1}.
388:
389: Importantly, the instability region of the parameter space is by no means not
390: the \textit{whole} region the region where phase separation is expected to
391: occur. Indeed, in analogy to what happens in rectangular geometry
392: \cite{argentina,khain2}, phase separation is also expected in a \textit{binodal}
393: (or coexistence) region of the area fractions, where the annular state is stable
394: to small perturbations, but unstable to sufficiently large ones. The whole
395: region of phase separation should be larger than the instability region, and it
396: should of course \textit{include} the instability region. Though we did not
397: attempt to determine the binodal region of the system from the hydrostatic
398: equations (this task has not been accomplished yet even for rectangular
399: geometry, except in the close vicinity of the critical point \cite{khain2}), we
400: determined the binodal region from our MD simulations reported in the next
401: section.
402:
403:
404: \begin{figure}
405: \includegraphics[width=7.5278cm]{fig4.eps}% Here is how to import EPS art
406: \caption{\label{fig4} The scaled steady state granular pressure $P$ versus the
407: grain area fraction $f$ for $\Omega=1.5$ and $\Lambda=1.1\times 10^{3}$ (the
408: dotted line), $\Lambda=1.5\times 10^{3}$ (the dash-dotted line),
409: $\Lambda=3.5\times 10^{3}$ (dashed line), and $\Lambda=5\times 10^{4}$ (the
410: solid line). The inset shows a zoom-in for $\Lambda=3.5\times 10^{3}$. The
411: borders $f_1$ and $f_2$ of the spinodal interval are determined from the
412: condition $\left(
413: \partial P/\partial f \right)_{\Lambda,\Omega}=0$. The thick solid line
414: encloses the spinodal balloon where the effective azimuthal compressibility of
415: the gas is negative.}
416: \end{figure}
417:
418: \section{\label{SecIII}MD Simulations}
419: \textit{Method}. We performed a series of event-driven MD simulations of this
420: system using an algorithm described by P\"{o}schel and Schwager
421: \cite{thorsten3}. Simulations involved $N$ hard disks of diameter $d=1$ and mass
422: $m=1$. After each collision of particle $i$ with particle $j$, their relative
423: velocity is updated according to
424: \begin{equation}
425: \vec{v}_{ij}^{\,\prime}=\vec{v}_{ij} - \left( 1+\mu \right) \left(
426: \vec{v}_{ij}\cdot \hat{r}_{ij}\right)\hat{r}_{ij}\,, \label{eq:velocs}
427: \end{equation}
428: where $\vec{v}_{ij}$ is the precollisional relative velocity, and
429: $\hat{r}_{ij}\equiv \vec{r}_{ij}/\left|\vec{r}_{ij}\right|$ is a unit vector
430: connecting the centers of the two particles. Particle collisions with the
431: exterior wall $r=R_{\text{ext}}$ are assumed elastic. The interior wall is kept
432: at constant temperature $T_{0}$ that we set to unity. This is implemented as
433: follows. When a particle collides with the wall it forgets its velocity and
434: picks up a new one from a proper Maxwellian distribution with temperature
435: $T_{0}$, see \textit{e.g}. Ref. \cite{thorsten3}, pages 173-177, for detail.
436: The time scale is therefore $d(m/T_{0})^{1/2}=1$. The initial condition is a
437: uniform distribution of non-overlapping particles inside the annular box. Their
438: initial velocities are taken randomly from a Maxwellian distribution at
439: temperature $T_{0}=1$. In all simulations the coefficient of normal restitution
440: $\mu=0.92$ and the interior radius $R_{\text{int}}/d=22.0$ were fixed, whereas
441: the the number of particles $527\leq N\leq 7800$ and the aspect ratio
442: $1.5\leq\Omega\leq 6$ were varied. In terms of the three scaled hydrodynamic
443: parameters the heat loss parameter $\Lambda=81.09$ was fixed whereas $f$ and
444: $\Omega$ varied.
445:
446: To compare the simulation results with predictions of our hydrostatic theory,
447: all the measurements were performed once the system reached a steady state. This
448: was monitored by the evolution of the total kinetic energy $(1/2)
449: \sum_{i=1}^{N}\vec{v}_{i}^{\;2}$, which first decays and then, on the
450: average, stays constant.\\
451:
452: \begin{figure}
453: \includegraphics[width=7.5278cm]{fig5.eps}% Here is how to import EPS art
454: \caption{\label{fig5} Typical steady state snapshots for $N=1250$ and $\Omega=6$
455: (a); $N=5267$ and $\Omega=3$ (b), and $N=6320$ and $\Omega=6$ (c). Panels (a)
456: and (b) correspond to annular states of the hydrostatic theory, whereas panel
457: (c) shows a broken-symmetry (phase separated) state.}
458: \end{figure}
459:
460: \textit{Steady States}. Typical steady state snapshots of the system, observed
461: in our MD simulation, are displayed in Fig.~\ref{fig5}. Panel (a) shows a dilute
462: state where the radial density inhomogeneity, though actually present, is not
463: visible by naked eye. Panels (b) and (c) do exhibit a pronounced radial density
464: inhomogeneity. Apart from visible density fluctuations, panels (a) and (b)
465: correspond to annular states. Panel (c) depicts a broken-symmetry (phase
466: separated) state. When an annular state is observed, its density profile agrees
467: well with the solution of the hydrostatic equations
468: (\ref{eq:density_eq})-(\ref{eq:norma}). A typical example of such a comparison
469: is shown in Fig.~\ref{fig1}.
470:
471: \begin{figure}
472: \includegraphics[width=7.5278cm]{fig6.eps}% Here is how to import EPS art
473: \caption{\label{fig6} Typical steady state snapshots (the left column) and the
474: temporal evolution of the COM (the middle column) and of the squared amplitude
475: of the fundamental Fourier mode (the right column). The temporal data are
476: sampled each $150$ collisions per particle. Each row corresponds to one
477: simulation with the indicated parameters. The vertical scale of panels a and b
478: was stretched for clarity.}
479: \end{figure}
480:
481: Let us fix the aspect ratio $\Omega$ of the annulus at not too a small value and
482: vary the number of particles $N$. First, what happens on a qualitative level?
483: The simulations show that, at small $N$, dilute annular states, similar to
484: snapshot (a) in Fig.~\ref{fig5}, are observed. As $N$ increases,
485: broken-symmetric states start to appear. Well within the unstable region, found
486: from hydrodynamics, a high density cluster appears, like the one shown in
487: Fig.~\ref{fig5}c, and performs an erratic motion along the exterior wall. As $N$
488: is increased still further, well beyond the high-$f$ branch of the unstable
489: region, an \textit{annular state} reappears, as in Fig.~\ref{fig5}b. This time,
490: however, the annular state is denser, while its local structure varies from a
491: solid-like (with imperfections such as voids and line defects) to a liquid-like.
492:
493: To characterize the spatio-temporal behavior of the granulate at a steady state,
494: we followed the position of the center of mass (COM) of the granulate. Several
495: examples of the COM trajectories are displayed in Fig.~\ref{fig6}. Here cases
496: (a) and (b) correspond, in the hydrodynamic language, to annular states. There
497: are, however, significant fluctuations of the COM around the center of the
498: annulus. These fluctuations are of course not accounted for by hydrodynamic
499: theory. In case (b), where the dense cluster develops, the fluctuations are much
500: weaker that in case (a). More interesting are cases (c) and (d). They correspond
501: to broken-symmetry states: well within the phase separation region of the
502: parameter space (case c) and close to the phase separation border (case d). The
503: COM trajectory in case (c) shows that the granular ``droplet" performs random
504: motion in the azimuthal direction, staying close to the exterior wall. This is
505: in contrast with case (d), where fluctuations are strong both in the azimuthal
506: and in the radial directions. Following the actual snapshots of the simulation,
507: one observes here a very complicated motion of the ``droplet", as well as its
508: dissolution into more ``droplets", mergers of the droplets \textit{etc}.
509: Therefore, as in the case of granular phase separation in rectangular geometry
510: \cite{baruch2}, the granular phase separation in annular geometry is accompanied
511: by considerable spatio-temporal fluctuations. In this situation a clear
512: distinction between a phase-separated state and an annular state, and a
513: comparison between the MD simulations and hydrodynamic theory, demand proper
514: diagnostics. We found that such diagnostics are provided by the azimuthal
515: spectrum of the particle density and its probability distribution.
516:
517: \begin{figure}
518: \includegraphics[width=7.5278cm]{fig7.eps}% Here is how to import EPS art
519: \caption{\label{fig7} The normalized probability distribution functions
520: $P_{1}\left(A_{1}^{2}/a_{0}^{2} \right)$ for $\Omega=3$ (the left column) and
521: $\Omega=5$ (the right column) for different numbers of particles.}
522: \end{figure}
523:
524: \textit{Azimuthal Density Spectrum}. Let us consider the (time-dependent)
525: rescaled density field $\nu(r,\theta,t)=n(r,\theta,t)/n_c $ (where $r$ is
526: rescaled to the interior wall radius as before), and introduce the integrated
527: field $\hat{\nu}(\theta, t)$:
528: \begin{equation}
529: \hat{\nu}(\theta,t) = \int_{1}^{\Omega}\,\nu(r,\theta,t)\, r \, dr \,.
530: \end{equation}
531: In a system of $N$ particles, $\hat{\nu}(\theta,t)$ is normalized so that
532: \begin{equation}\label{nunorm}
533: \int_{0}^{2 \pi} \hat{\nu} (\theta,t)\, d\theta=\frac{N}{n_c\,R^2_{\text{int}}}\,.
534: \end{equation}
535: %$N=n_c\,R^2_{\text{int}} \int_{0}^{2 \pi} \hat{\nu} \, d\theta\,$.
536: Because of the periodicity in $\theta$ the function $\hat{\nu}(\theta,t)$ can be
537: expanded in a Fourier series:
538: \begin{equation}
539: \hat{\nu}(\theta,t) = a_0+\sum_{k=1}^{\infty} \left[ a_k (t) \cos (k\theta) +
540: b_k(t) \sin (k\theta)\right] \,,
541: \end{equation}
542: where $a_0$ is independent of time because of the normalization condition
543: (\ref{nunorm}). We will work with the quantities
544: \begin{equation}
545: A_k^{2}(t)\equiv a_k^{2}(t) + b_k^{2}(t) \,, \;\;\;k\ge1\,. \label{eq:spectral}
546: \end{equation}
547: For the (deterministic) annular state one has $A_k=0$ for all $k\ge 1$, while
548: for a symmetry-broken state $A_k>0$. The relative quantities $A_k^2(t)/a_{0}^2$
549: can serve as measures of the azimuthal symmetry breaking. As is shown in
550: Table~\ref{tab:Ak}, $A_1^2(t)$ is usually much larger (on the average) that the
551: rest of $A_{k}^{2}(t)$. Therefore, the quantity $A_1^2(t)/a_{0}^2$ is sufficient
552: for our purposes.
553:
554: \begin{table}
555: \caption{\label{tab:Ak}The averaged squared relative amplitudes $\langle
556: A_{k}^2(t)\rangle/a_{0}^2$ for the first three modes $k=1,2$ and $3$. (a)
557: $N=2634$, $\Omega=3$, (b) $N=5267$, $\Omega=4$, (c) $N=1000$, $\Omega=2.25$, and
558: (d) $N=1250$, $\Omega=3$.}
559: \begin{ruledtabular}
560: \begin{tabular}{ccccc}
561: $k$&(a)&(b)&(c)&(d)\\
562: \hline
563: $1$ & $0.66\pm 0.05$ & $0.39\pm 0.04$ & $0.30\pm 0.08$ & $0.77\pm 0.05$\\
564: $2$ & $0.04\pm 0.02$ & $0.05\pm 0.02$ & $0.07\pm 0.01$ & $0.28\pm 0.09$\\
565: $3$ & $0.03\pm 0.02$ & $0.03\pm 0.03$ & $0.02\pm 0.02$ & $0.11\pm 0.08$\\
566: \end{tabular}
567: \end{ruledtabular}
568: \end{table}
569:
570: Once the system relaxed to a steady state, we followed the temporal evolution of
571: the quantity $A_{1}^{2}/a_0^2$. Typical results are shown in the right column of
572: Fig.~\ref{fig6}. One observes that, for annular states, this quantity is usually
573: small, as is the cases (a) and (b) in Fig.~\ref{fig6}. For broken-symmetry states
574: $A_{1}^{2}$ is larger, and it increases as one moves deeper into the phase
575: separation region. (Notice that in Fig.~\ref{fig6} the averaged value
576: of $A_{1}^{2}/a_0^2$ in (c) is
577: larger than in (d), which means that (c) is deeper in the phase separation
578: region.) Another characteristics of $A_{1}^{2}(t)/a_0^2$ is the magnitude of
579: fluctuations. One can notice that, in the vicinity of the phase separation
580: border the fluctuations are stronger (as in case (d) in Fig.~\ref{fig6}).
581:
582: All these properties are encoded in the \textit{probability distribution}
583: $P_{1}$ of the values of $\left(A_{1}/a_{0}\right)^{2}$: the ultimate tool of
584: our diagnostics. Figure~\ref{fig7} shows two series of measurements of this
585: quantity at different $N$: for $\Omega=3$ and $\Omega=5$. By following the
586: position of the maximum of $P_{1}$ we were able to to sharply discriminate
587: between the annular states and phase separated states and therefore to locate
588: the phase separation border. When the maximum of $P_1$ occurs at the zero value
589: of $\left(A_{1}/a_{0}\right)^{2}$ (as in cases (a) and (d) and, respectively, (e) and
590: (h) in Fig.~\ref{fig7}), an annular state is observed. On the contrary, when the
591: maximum of $P_1$ occurs at a non-zero value of $\left(A_{1}/a_{0}\right)^{2}$
592: (as in cases (b) and (c) and, respectively, (f) and (g) in Fig.~\ref{fig7}), a phase
593: separated state is observed. In each case, the width of the probability
594: distribution (measured, for example, at the half-maximum) yields a direct
595: measure of the magnitude of fluctuations. Near the phase separation border,
596: strong fluctuations (that is, broader distributions) are observed, as in case (c)
597: of Fig.~\ref{fig7}.
598:
599: Using the position of the maximum of $P_1$ as a criterion for phase separation,
600: we show, in Fig.~\ref{fig8}, the $\Omega-f$ diagram obtained from the MD
601: simulations. The same figure also depicts the hydrostatic prediction of the
602: \textit{instability} region. One can see that the instability region is located
603: within the phase separation region, as expected.
604:
605: \begin{figure}
606: \includegraphics[width=7.5278cm]{fig8.eps}% Here is how to import EPS art
607: \caption{\label{fig8} The $\Omega$-$f$ phase diagram for $\Lambda=81.09$. The
608: solid curve is given by the granular hydrostatics: it shows the borders of the
609: region where the annular state is unstable with respect to small perturbations.
610: The filled symbols depict the parameters in which phase separated states are
611: observed, whereas the hollow symbols show the parameters at which annular states
612: are observed. The dashed line is an estimated binodal line of the system.}
613: \end{figure}
614:
615: \begin{figure}
616: \includegraphics[width=7.5278cm]{fig9.eps}% Here is how to import EPS art
617: \caption{\label{fig9} The marginal stability lines for our main setting (the
618: solid line) and for an alternative setting in which the thermal wall is at
619: $r=R_{\textit{ext}}$ and the elastic wall is at $r=R_{\textit{int}}$ (the dashed
620: line).}
621: \end{figure}
622:
623: \section{\label{SecIV}Some modifications of the model}
624: We also investigated an alternative setting in which the exterior wall is the
625: driving wall, while the interior wall is elastic. The corresponding hydrostatic
626: problem is determined by the same three scaled parameters $f$, $\Lambda$ and
627: $\Omega$, but the boundary conditions must be changed accordingly. Here
628: azimuthally symmetric clusters appear near the (elastic) interior wall. Symmetry
629: breaking instability occurs here as well. We found very similar marginal
630: stability curves here, but they are narrower (as shown in Fig.~\ref{fig9}) than
631: those obtained for the original setting.
632:
633: \begin{figure}
634: \includegraphics[width=7.5278cm]{fig10.eps}% Here is how to import EPS art
635: \caption{\label{fig10} Typical steady state snapshots (the left column) and the
636: normalized probability distribution functions $P_{1}\left(A_{1}^{2}/a_{0}^{2}
637: \right)$ for an inelastic exterior wall, $\mu_{wall}=0.99$ (the right column)
638: for different numbers of particles.}
639: \end{figure}
640:
641: Finally, we returned to our original setting and performed several MD
642: simulations, replacing the perfectly elastic exterior wall by a weakly inelastic
643: one. The inelastic particle collisions with the exterior wall were modeled in
644: the same way as the inelastic collisions between particles. Typical results of
645: these simulations are shown in Fig.~\ref{fig10}. It can be seen that, for the
646: right choice of parameters, the phase separation persists. This result is
647: important for a possible experimental test of our theory.
648:
649:
650: \section{\label{SecV}Summary}
651: We combined equations of granular hydrostatics and event-driven MD simulations
652: to investigate spontaneous phase separation of a monodisperse gas of
653: inelastically colliding hard disks in a two-dimensional annulus, the inner
654: circle of which serves as a ``thermal wall". A marginal stability analysis
655: yields a region of the parameter space where the annular state -- the basic,
656: azimuthally symmetric steady state of the system -- is unstable with respect to
657: small perturbations which break the azimuthal symmetry. The physics behind the
658: instability is negative effective compressibility of the gas in the azimuthal
659: direction, which results from the inelastic energy loss. MD simulations of this
660: system show phase separation, but it is masked by large spatio-temporal
661: fluctuations. By measuring the probability distribution of the amplitude of the
662: fundamental Fourier mode of the azimuthal spectrum of the particle density we
663: have been able to clearly identify the transition to phase separated states in
664: the MD simulations. We have found that the instability region of the parameter
665: space, predicted from hydrostatics, is located within the phase separation
666: region observed in the MD simulations. This implies the presence of a binodal
667: (coexistence) region, where the annular state is \textit{metastable}, similar to
668: what was found in rectangular geometry \cite{argentina,khain2}. The instability
669: persists in an alternative setting (a driving exterior wall and an elastic
670: interior wall), and also when the elastic wall is replaced by a weakly inelastic
671: one. We hope our results will stimulate experimental work on the phase
672: separation instability.
673:
674:
675: \section{Acknowledgments}
676: This work grew out of a student research project at the 2003 Summer School
677: ``From pattern formation to granular physics and soft condensed matter"
678: sponsored by the European Union under FP5 High Level Scientific Conferences, and
679: by the NATO Advanced Study Institute. We are grateful to Igor Aranson, Pavel
680: Sasorov and Thomas Schwager for advice. MDM acknowledges financial support from
681: MEyC and FEDER (project FIS2005-00791). BM acknowledges financial support from
682: the Israel Science Foundation (grant No. 107/05) and from the German-Israel
683: Foundation for Scientific Research and Development (Grant I-795-166.10/2003).
684:
685: \begin{thebibliography}{99}
686: \bibitem{jaeger} H. R. Jaeger, S. Nagel, and R. P. Behringer, Rev.
687: Mod. Phys. \textbf{68}, 1259 (1996); Phys. Today \textbf{49}(4), 32 (1996).
688: \bibitem{ristow} G.H. Ristow, {\it Pattern Formation in Granular
689: Materials (Springer Tracts in Modern Physics)} (Springer, Berlin, 2000).
690: \bibitem{aranson+tsimring} I.S. Aranson and L.S. Tsimring, Rev. Mod. Phys. \textbf{78},
691: 641 (2006).
692: \bibitem{campbell} C. S. Campbell, Annu. Rev. Fluid Mech. \textbf{22}, 57 (1990).
693: \bibitem{kadanoff} L. P. Kadanoff, Rev. Mod. Phys. \textbf{71}, 435 (1999).
694: \bibitem{thorsten1} \textit{Granular Gases}, edited by T. P\"oschel and S. Luding (Springer, Berlin, 2001).
695: \bibitem{thorsten2} \textit{Granular Gas Dynamics}, edited by T. P\"oschel and N.
696: Brilliantov (Springer, Berlin, 2001).
697: \bibitem{goldhirsch1} I. Goldhirsch, Annu. Rev. Fluid Mech. \textbf{35}, 267 (2003).
698: \bibitem{brilliantov} N.V. Brilliantov N.V. and T. P\"{o}schel, \textit{Kinetic Theory of Granular
699: Gases}, (Oxford University Press, Oxford, 2004).
700: \bibitem{sela} N. Sela and I. Goldhirsch, J. Fluid Mech. \textbf{361}, 41 (1998).
701: \bibitem{brey1} J. J. Brey, J. W. Dufty, C. S. Kim, and A. Santos, Phys. Rev. E \textbf{58}, 4638 (1998).
702: \bibitem{lutsko} J. F. Lutsko, Phys. Rev. E \textbf{72}, 021306 (2005).
703: \bibitem{livne1} E. Livne, B. Meerson, and P. V. Sasorov, Phys. Rev. E \textbf{65}, 021302 (2002);
704: cond-mat/0008301 (2000).
705: \bibitem{argentina} M. Argentina, M.G. Clerc, and R. Soto, Phys.
706: Rev. Lett. \textbf{89}, 044301 (2002).
707: \bibitem{brey2} J. J. Brey, M. J. Ruiz-Montero, F. Moreno, and R. Garc\'ia-Rojo, Phys. Rev. E \textbf{65}, 061302 (2002).
708: \bibitem{khain1} E. Khain and B. Meerson, Phys. Rev. E \textbf{66}, 021306 (2002).
709: \bibitem{livne2} E. Livne, B. Meerson, and P.V. Sasorov, Phys. Rev. E \textbf{66}, 050301(R) (2002).
710: \bibitem{baruch2} B. Meerson, T. P\"oschel, P. V. Sasorov, and T. Schwager, Phys. Rev. E \textbf{69}, 021302 (2004).
711: \bibitem{khain2} E. Khain, B. Meerson, and P. V. Sasorov, Phys. Rev. E \textbf{70}, 051310 (2004).
712: \bibitem{kudrolli} A. Kudrolli, M. Wolpert, and J. P. Gollub, Phys. Rev. Lett. \textbf{78}, 1383 (1997).
713: \bibitem{kadanoff2} Y. Du, H. Li, and L. P. Kadanoff, Phys. Rev. Lett. \textbf{74}, 1268 (1995).
714: \bibitem{esipov} S. E. Esipov and T. P\"oschel, J. Stat. Phys. \textbf{86}, 1385 (1997).
715: \bibitem{grossman} E. L. Grossman, T. Zhou, and E. Ben-Naim, Phys. Rev. E \textbf{55}, 4200 (1997).
716: \bibitem{jenkins} J. T. Jenkins and M. W. Richman, Phys. Fluids \textbf{28}, 3485 (1985).
717: \bibitem{carnahan} N. F. Carnahan and K. E. Starling, J. Chem. Phys. \textbf{51}, 635 (1969).
718: \bibitem{thorsten3} T. P\"oschel and T. Schwager, \textit{Computational Granular Dynamics: Models and Algorithms}. (Springer, Berlin, 2005).
719:
720: \end{thebibliography}
721:
722: \end{document}
723: