1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
3: \documentclass[pre,twocolumn,superscriptaddress]{revtex4}
4: \usepackage{epsfig,amsmath,revsymb}
5: \begin {document}
6: \title {Exponential velocity tails in a driven inelastic Maxwell model}
7: \author{T. Antal} \email{Tibor.Antal@physics.unige.ch}
8: \affiliation{Department of Physics, University of Geneva, CH 1211
9: Geneva 4, Switzerland}
10: \affiliation{Institute for Theoretical Physics, E\"otv\"os University,
11: 1117 Budapest, P\'azm\'any s\'et\'any 1/a, Hungary}
12: \author{Michel Droz}\email{Michel.Droz@physics.unige.ch}
13: \affiliation{Department of Physics, University of Geneva, CH 1211
14: Geneva 4, Switzerland}
15: \author{Adam Lipowski}\email{lipowski@amu.edu.pl}
16: \affiliation{Department of Physics, University of Geneva, CH 1211
17: Geneva 4, Switzerland}
18: \affiliation{Department of Physics, Adam Mickiewicz University,
19: 61-614 Pozna\'{n}, Poland}
20: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
21: \pacs{}
22: \begin {abstract}
23: The problem of the steady-state velocity distribution in a driven inelastic Maxwell
24: model of shaken granular material is revisited.
25: Numerical solution of the master equation and analytical arguments
26: show that the model has
27: bilateral exponential velocity tails ($P(v)\sim e^{-|v|/\sqrt D}$), where $D$ is
28: the amplitude of the noise.
29: Previous study of this model predicted Gaussian tails
30: ($P(v)\sim e^{-av^2}$).
31: \end{abstract}
32: \maketitle
33: %------------------------------------------------------------------
34: Recently, granular systems have been intensively studied~\mbox{\cite{SWINNEY,MUZZIO}}.
35: One reason of such an interest is the fact that in many respects these
36: systems are highly unconventional and even exotic~\cite{KADANOFF}.
37: As a result, very often even standard laws of statistical physics require certain
38: modification when applied to granular systems.
39:
40: As an example, let us consider the Maxwell-Boltzmann law for velocity
41: distribution $P(v)$ of atoms or molecules in a gaseous state, which states that
42: $P(v)\sim {\rm e}^{-av^2}$.
43: Under certain experimental conditions, a granular system can be considered as a
44: gas and a natural question is what is its velocity distribution.
45: Numerous theoretical and experimental studies do not provide a simple answer to
46: this question.
47: On the contrary, they show that $P(v)$ depends on certain details of the
48: system as for example how energy is transferred by a thermostat into the system,
49: in order to balance the energy loss during inelastic collisions.
50: Theoretical works have shown that $P(v)$ might be of the form
51: ${\rm e}^{-av^2}$~\cite{SWINNEY1,BIBEN},
52: ${\rm e}^{-av^{3/2}}$~\cite{NOIJE,BARRAT} or
53: ${\rm e}^{-av}$~\cite{MONTANERO}.
54: Under certain conditions experiments show clear deviations from the
55: Maxwell-Boltzmann law, but it is still rather difficult to decide what is the form
56: of $P(v)$ in real granular systems~\cite{OLAFSEN}.
57:
58: Very often granular systems are described using the so-called inelastic
59: hard-sphere models, which could be then analysed using corresponding Boltzmann
60: equations.
61: These off-lattice, two- or three-dimensional systems are however very difficult to
62: study, especially when we want to explore large-velocity regions of the phase space.
63: A possible alternative is to construct simplified models for which more
64: accurate calculations are possible.
65: One class of such models are Maxwell models, for which the collision term
66: in the corresponding Boltzmann equation is velocity independent.
67: Recent studies show that for Maxwell models $P(v)$ might take either
68: exponential (${\rm e}^{-av}$)~\cite{ERNST} or Gaussian
69: (${\rm e}^{-av^2}$) form~\cite{BEN,CARILLO}.
70:
71: A particularly interesting Maxwell model was proposed by Ben-Naim and
72: Krapivsky (BN-K)~\cite{BEN}.
73: These authors presented an elegant solution of the master equation of their model using a
74: Fourier transform method.
75: However, their conclusion that $P(v)$ has a Gaussian decay is based on certain
76: approximation whose validity is difficult to assess.
77: Namely, they infer the large-velocity behaviour of $P(v)$ from the low-$k$ behaviour of
78: the Fourier transform $\hat P(k)$.
79: Since BN-K model is one of the very few models for which exact or numerical but
80: very precise calculations can be made, it would be desirable to clarify the validity
81: of this approach.
82:
83: In the present paper we reexamine the BN-K model.
84: Analysing numerically the solution of the master equation of the model we obtain the velocity
85: distribution $P(v)$.
86: Asymptotically ($v\rightarrow\infty$), this quantity shows bilateral exponential tails
87: ($P(v)\sim {\rm e}^{-a|v|}$) and such a behaviour is seen over more than 10 decades.
88: We also present analytical arguments which support existence of
89: bilateral exponential tails in this model.
90: From our analysis it follows that $a=\frac{1}{\sqrt D}$, where $D$ is the
91: amplitude of noise which simulates the input of energy into the system.
92:
93: To introduce the model, we consider a collection of particles that are
94: characterized by a single parameter, their velocity $v$.
95: In this model positions of particles are not specified hence the model neglects any
96: spatial correlations.
97: Particles undergo two-body inelastic collisions that change their velocities
98: according to $(v_1,v_2)\rightarrow(v_1',v_2')$, where
99: \begin{equation}
100: \begin{pmatrix} v_1' \\ v_2'\end{pmatrix} =
101: \begin{pmatrix} \gamma & 1-\gamma \\ 1-\gamma & \gamma \end{pmatrix}
102: \begin{pmatrix} v_1 \\ v_2\end{pmatrix},
103: \label{e1}
104: \end{equation}
105: and $\gamma$ is the inelasticity parameter ($0<\gamma<1$).
106: Particles which participate in a collision are chosen at random.
107: In addition to the collision rule (\ref{e1}), particles are subjected to the
108: uncorrelated white noise of strength $D$.
109: Existence of noise ensures that the model has a well-defined nontrivial steady
110: state.
111: In the steady-state the velocity distribution $P(v)$ satisfies
112: the following equation~\cite{BEN}
113: \begin{equation}
114: -DP''(v)=-P(v)+
115: \frac{1}{1-\gamma}\int_{-\infty}^{+\infty} duP(u)
116: P\left(\frac{v-\gamma u}{1-\gamma}\right),
117: \label{e2}
118: \end{equation}
119: where double prime $''$ denotes the second derivative with respect to velocity.
120: To solve eq.~(\ref{e2}) BN-K introduced the Fourier transform of the velocity
121: distribution $\hat P(k) = \int dv e^{ikv} P(v)$, which
122: in the steady state satisfies
123: \begin{equation}
124: (1+Dk^2)\hat P(k) = \hat P[\gamma k] \hat P[(1-\gamma)k].
125: \label{pksteady}
126: \end{equation}
127: This equation admits the following solution
128: \begin{equation}
129: \hat P(k) = \prod_{i=0}^{\infty} \prod_{j=0}^i [1+\gamma^{2j}(1-\gamma)^{2(i-j)}Dk^2].
130: \label{pkrecur}
131: \end{equation}
132: After some transformations eq.~(\ref{pkrecur}) can be written as
133: \begin{equation}
134: \hat P(k) = {\rm exp} \left[ \sum_{n=1}^\infty
135: \frac{(-Dk^2)^n}{na_{2n}(\gamma)} \right] ~,
136: \label{final}
137: \end{equation}
138: where $a_n(\gamma) = 1 - (1-\gamma)^n - \gamma^n$.
139: To obtain a large velocity behaviour of $P(v)$ BN-K truncate the series in
140: eq.~(\ref{final}) keeping only the first, quadratic in $k$, term
141: \begin{equation}
142: \hat P(k)\approx {\rm exp}\left[\frac{-Dk^2}{a_2(\gamma)}\right].
143: \label{app}
144: \end{equation}
145: Subsequently, taking the inverse Fourier transform of this Gaussian function they obtain
146: $P(v)\sim {\rm exp}(\frac{-v^2}{2v_0^2})$, where $v_0^2=2D/a_2(\gamma)$.
147:
148: In our opinion, the above procedure of truncating the series and inverting the resulting
149: function is not well justified.
150: The approximation (\ref{app}) agrees with the exact
151: expression (\ref{final}) only up to the $k^2$ term of the Taylor expansion.
152: Equally consistent approximation can be given as
153: \begin{equation}
154: \hat P(k)\approx \frac{c^2}{c^2+k^2},
155: \label{app1}
156: \end{equation}
157: where $c=\sqrt{\frac{a_2(\gamma)}{D}}$.
158: But the inverse Fourier transform of~(\ref{app1}) is $\frac{c}{2}{\rm exp}(-c|v|)$, which
159: is qualitatively different than the Gaussian decay obtained by BN-K.
160: Both (\ref{app}) and (\ref{app1}) are consistent with the exact solution (\ref{final}) up
161: to the $k^2$ order but differ at the higher orders.
162: Apparently, these higher order terms qualitatively affect the large-velocity limit of
163: $P(v)$.
164: Consequently, without more detailed arguments such approximations are not justified.
165: Indeed, as we will show below, $P(v)$ has a bilateral exponential decay but with a
166: different coefficient $c$.
167:
168: To check the validity of the BN-K approach we calculate $P(v)$ numerically as an inverse
169: Fourier transform of $\hat P(k)$.
170: One way to compute $\hat P(k)$ is to evaluate the infinite sums in the logarithm of
171: eq.~(\ref{pkrecur}).
172: However, it turns out that using eq.~(\ref{final}) leads to a better precision.
173: Let us notice, however, that the series in eq.~(\ref{final}) converges but only for
174: $|k|<k_c=\frac{1}{\sqrt D}$.
175: To calculate $\hat P(k)$ for $|k|>k_c$ we can then use eq.~(\ref{pksteady}),
176: provided that $\gamma k$ and $(1-\gamma)k$ fall within the range of convergence.
177: If not, we have to calculate $\hat P(\gamma k)$ and $\hat P((1-\gamma)k)$
178: referring again to eq.~(\ref{pksteady}).
179: Implementing this recursive procedure, we calculated $\hat P(k)$ and then using the
180: Fast Fourier Transformation algorithm we obtained $P(v)$.
181: Our results for $D=1$ are shown in Fig.~\ref{fft}.
182: This figure clearly shows that $P(v)$ decays exponentially.
183: For $\gamma=0.5$ and 0.8 such a behaviour is seen for more than 10 decades.
184: For $\gamma$ close to unity there is a $\exp (-av^2)$ decay for small velocities which
185: asymptotically is replaced by the bilateral exponential decay ($\exp (-c|v|)$).
186: Moreover, as shown in Fig.~\ref{fft} the asymptotic slope $c$ is independent of
187: $\gamma$ and is approximately one.
188: As we will show below this slope only depends on the amplitude of noise $D$.
189: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
190: \begin{figure}
191: \centerline{\epsfxsize=9cm
192: \epsfbox{fig1.ps}
193: }
194: %\figspace
195: \caption{
196: The velocity distribution $P(v)$ as a function of $v$ calculated for $D=1$.
197: Due to symmetry only $v>0$ part is shown.
198: The thick line represents the exponential function $\exp (-v)$.
199: }
200: \label{fft}
201: \end{figure}
202: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
203: Before doing that let us notice that the $D$-dependence of the velocity distribution
204: $P(v)$ can be easily inferred from the fact that $D$ and $k$ enter its Fourier transform
205: only through $Dk^2$ terms (see eqs.~(\ref{pkrecur}) and (\ref{final})).
206: From this property one can easily obtain that
207: \begin{equation}
208: P(v, D) = \frac{1}{\sqrt{D}} P\left(\frac{v}{\sqrt{D}},D=1 \right),
209: \label{rescal}
210: \end{equation}
211: where we explicitly indicated the dependence on the noise $D$.
212:
213: In the following we provide some analytical arguments showing
214: that $P(v)$ decays exponentially.
215: Let us assume that in eq.~(\ref{e2}) the second term (gain) can be neglected.
216: Then eq.~(\ref{e2}) simplifies to $DP''(v)=P(v)$ and the normalizable solution reads
217: \begin{equation}
218: P(v)\sim e^{-\frac{|v|}{\sqrt D}}.
219: \label{e3}
220: \end{equation}
221: Let us notice that the exponential part is $\gamma$-independent
222: (since $\gamma$ enters the master equation only through the neglected
223: gain term).
224: Moreover, such a solution is in a very good agreement with numerical
225: calculations (see Fig.~\ref{fft}).
226:
227: For more general models it is known that neglecting the gain term is justified
228: for $v\rightarrow\infty$ when the resulting solution decays faster than
229: exponentially~\cite{NOIJE,BARRAT}.
230: Our solution (\ref{e3}) is thus a marginal case.
231: However, we can show that for our model the gain term in the limit
232: $v\rightarrow\infty$ indeed can be neglected.
233: First, let us evaluate the gain term for the solution (\ref{e3}).
234: Elementary integration for $D=1$ and $v>0$ gives
235: \begin{multline}
236: \frac{1}{1-\gamma}\int_{-\infty}^{+\infty} du
237: {\rm exp}(-|u|)
238: {\rm exp}\Bigl[-\Big|\frac{v-\gamma u}{1-\gamma}\Big|\Bigr] \\ =
239: \frac{2}{2\gamma-1}\left[\gamma {\rm exp} \left(\frac{-v}{\gamma}\right)
240: +(\gamma-1){\rm exp} \left(\frac{-v}{1-\gamma}\right)\right]
241: \label{e4}
242: \end{multline}
243: One can see that since $0<\gamma<1$, the gain term decays exponentially
244: with $v$ but faster than the solution (\ref{e3}) (i.e., a non-neglected loss term
245: in (\ref{e2})).
246: We expect that eq.~(\ref{e3}) is only an asymptotic ($v\rightarrow\infty$)
247: solution of the master equation (\ref{e2}).
248: Thus, for velocities $u\sim O(1)$ the distribution $P(u)$ deviates
249: from the asymptotic form (\ref{e3}).
250: This will modify the integral (\ref{e4}) but only in the vicinity of
251: $u=0$ and $u=\frac{v}{\gamma}$.
252: As we argue below, such a modification of $P(u)$ still leads to the gain term decaying
253: faster than the loss term.
254:
255: Indeed, the contribution around $u=0$ is a product of $P(u)$ and of the
256: exponential term ${\rm exp}\bigl(\frac{-v-\gamma u}{1-\gamma}\bigr)$.
257: Thus, for large $v$ a modification of $P(u)$ around $u=0$ will change our
258: estimation (\ref{e4}) but only at the order of
259: ${\rm exp}(\frac{-v}{1-\gamma})$.
260: Similarly, one can show that a modification of $P(u)$ around
261: $u=\frac{v}{\gamma}$ will change (\ref{e4}) by a factor of the order of
262: ${\rm exp}(\frac{-v}{\gamma})$.
263: Consequently, the gain term again decays faster than the solution $P(v)$,
264: which justifies its neglect.
265: Let us also notice that when $\gamma$ approaches 0 or 1 the model becomes
266: energy conserving and the bilateral exponential distribution (\ref{e3}) is no longer
267: expected to hold.
268: But it is easy to notice that in such cases the decay of the gain term
269: matches the decay of the solution $P(v)$ and it cannot be neglected.
270:
271: As a comment let us notice that knowing the second moment of $P(v)$~\cite{BEN}
272: enables us to calculate the effective temperature $T$ of our systems
273: defined as an averaged square velocity
274: \begin{equation}
275: T=\langle v^2\rangle = \frac{D}{\gamma(1-\gamma)}.
276: \label{temp}
277: \end{equation}
278: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
279: \begin{figure}
280: \centerline{\epsfxsize=9cm
281: \epsfbox{fig2.ps}
282: }
283: %\figspace
284: \caption{
285: Velocity distribution $P(v)$ calculated using Monte Carlo simulations.
286: Simulations were made for $N=10^5$ particles.
287: Continuous lines are the results obtained from Fourier inversion.
288: The inset shows our data in the semi-logarithmic scale.
289: }
290: \label{monte}
291: \end{figure}
292: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
293: To check the validity of our calculations we also made Monte Carlo simulations of this
294: model.
295: Results, which are shown in Fig.~\ref{monte} confirm the bilateral exponential
296: decay of $P(v)$
297: although the accuracy is this time much lower.
298: Our Monte Carlo data are rescaled in such a way that their variance matches that
299: obtained using the Fourier transform method and a very good agreement is seen even on
300: the logarithmic scale.
301:
302: In conclusion, we have shown that the Maxwell model proposed by Ben-Naim and Krapivsky
303: has velocity distribution decaying as a bilateral exponential.
304: Together with the recent results by Ernst and Brito~\cite{ERNST} it indicates that
305: such a decay might
306: be of more generic nature for this class of models.
307: As a possible extension, it would be desirable to examine some other Maxwell models
308: where velocities are not scalars but rather $d$-dimensional vectors.
309: Actually, such models were already studied and the analysis indicates that, for
310: increasing
311: $d$, correlations between velocities and deviation from the pure Gaussian distribution
312: decrease~\cite{BEN1}.
313: One possibility is that there might be a critical dimension $d_c$ and such that
314: for $d<d_c$ the velocity distribution has a bilateral exponential decay (as in the
315: present model) while it it has Gaussian
316: decay for $d\geq d_c$.
317: Analysis of such models is however left for the future.
318: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
319: \begin{acknowledgments}
320: We acknowledge interesting discussion with Fran\c{c}ois Coppex.
321: This work was partially supported by the Swiss National Science Foundation
322: and the project OFES 00-0578 "COSYC OF SENS" and
323: the Hungarian Academy of Sciences (Grant No. OTKA T029792).
324: \end{acknowledgments}
325: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
326: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
327: \begin{thebibliography}{}
328: \bibitem{SWINNEY} P.~B.~Umbanhowar, F.~Melo, and H.~L.~Swinney, Nature
329: {\bf 382}, 793 (1996).
330: \bibitem{MUZZIO} T.~Shinbrot and F.~J.~Muzzio, Nature {\bf 410}, 251 (2001).
331: \bibitem{KADANOFF} L.~Kadanoff, Rev.~Mod.~Phys.~{\bf 71}, 435 (1999).
332: \bibitem{SWINNEY1} C.~Bizon, M.~D.~Shattuck, J.~B.~Swift, and H.~L.~Swinney,
333: Phys.~Rev.~E {\bf 60}, 4340 (1999).
334: \bibitem{BIBEN} T.~Biben, Ph.~A.~Martin, and J.~Piasecki, Physica A~{\bf 310}, 308
335: (2002).
336: \bibitem{NOIJE} T.~P.~C.~Noije and M.~H.~Ernst, Granular Matter {\bf 1}, 57 (1998).
337: \bibitem{BARRAT} A.~Barrat, T.~Biben, Z.~R\'acz, E.~Trizac, and F.~van Wijland,
338: J.~Phys.~A {\bf 35}, 463 (2002).
339: \bibitem{MONTANERO} J.~M.~Montanero and A.~Santos, Granular Matter {\bf 2}, 53
340: (2000).
341: \bibitem{OLAFSEN} J.~S.~Olafsen and J.~S.~Urbach, Phys.~Rev.~Lett.~{\bf 81}, 4369
342: (1998).
343: J.~S.~Olafsen and J.~S.~Urbach, Phys.~Rev.~E {\bf 60}, R2468 (1999).
344: \bibitem{ERNST} M.~H.~Ernst and R.~Brito, Phys.~Rev.~E {\bf 65}, 040301 (2002).
345: \bibitem{CARILLO} J.~A.~Carillo, C.~Cercignani, and I.~M.~Gamba, Phys.~Rev.~E
346: {\bf 62}, 7700 (2000).
347: \bibitem{BEN} E.~Ben-Naim and P.~L.~Krapivsky, Phys.~Rev.~E {\bf 61}, R5 (2000).
348: \bibitem{BEN1} E.~Ben-Naim and P.~L.~Krapivsky, e-print: cond-mat/0202332.
349: \end{thebibliography}
350: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
351: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
352: \end {document}
353: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
354: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
355: