1: %% Template for a preprint Letter or Article for submission
2: %% to the journal Nature.
3: %% Written by Peter Czoschke, 26 February 2004
4: %%
5:
6: \documentclass{nature}
7: \usepackage{graphicx}
8:
9: %% make sure you have the nature.cls and naturemag.bst files where
10: %% LaTeX can find them
11:
12: \bibliographystyle{naturemag}
13:
14: \title{Scale invariance and universality of force networks in static granular
15: matter}
16:
17: %% Notice placement of commas and superscripts and use of &
18: %% in the author list
19:
20: \author{Srdjan Ostojic$^{1}$, Ell\'ak Somfai$^2$ \& Bernard Nienhuis$^1$}
21:
22:
23: \begin{document}
24:
25: \maketitle
26:
27: \begin{affiliations}
28: \item Institute for Theoretical Physics, Universiteit van Amsterdam,
29: Valckenierstraat 65, 1018 XE Amsterdam, Netherlands
30: \item Universiteit Leiden, Instituut-Lorentz, PO Box 9506, 2300 RA Leiden,
31: Netherlands
32: \end{affiliations}
33:
34: \begin{abstract}
35: Force networks form the skeleton of static granular
36: matter\cite{pg2005, jaeger:rev}. They are the key ingredient to
37: mechanical properties, such as stability\cite{daerr99},
38: elasticity\cite{makse99, resp:goldenberg05} and sound
39: transmission\cite{jia99,somfai05}, which are of utmost importance for
40: civil engineering and industrial processing. Previous studies have
41: focused on the global structure of external forces\cite{vanel99,
42: wittmer96, resp:geng1,resp:reydellet} (the boundary condition), and on
43: the probability distribution of individual contact
44: forces\cite{makse99,majmudar05}. The disordered spatial structure of
45: the force network, however, has remained elusive so far. Here we
46: report evidence for scale invariance of clusters of particles that
47: interact via relatively strong forces. We analyzed granular packings
48: generated by molecular dynamics simulations mimicking real granular
49: matter; despite the visual variation, force networks for various
50: values of the confining pressure and other parameters have identical
51: scaling exponents and scaling function, and thus determine a
52: universality class. Remarkably, the flat ensemble of force
53: configurations\cite{edwards1, jam:makse1,fluct:jacco1}---a simple
54: generalization of equilibrium statistical mechanics---belongs to the
55: same universality class, while some widely studied simplified
56: models\cite{qmodel0,qmodel, raj} do not.
57: \end{abstract}
58:
59:
60: Sand dunes, piles of apples in a supermarket, and coffee beans in a
61: jar are all examples of static granular matter. In these examples the
62: particles (sand grains, apples, coffee beans) interact with their
63: neighbors with repulsive contact forces forming a network. Due to
64: disorder (even in a regular pile each apple is slightly different) the
65: force network is inhomogeneous in space\cite{fluct:radjai1,
66: fluct:makse}. The simplest characterization of the force network is
67: the probability distribution of individual contact
68: forces\cite{fluct:mueth, fluct:blair1, fluct:erikson}: $P(F)$. In
69: typical granular packings $P(F)$ has an exponential tail at large $F$,
70: and a plateau at small $F$. While $P(F)$ is an important and well
71: studied quantity, it tells nothing about the spatial structure of the
72: force network, which is so striking visually, as can be seen on
73: Fig.~\ref{fig1}~a. As a large contact force on one side of a particle
74: is typically balanced by another large force on the opposite side,
75: large forces tend to align in filamentary structures, called
76: \emph{force chains} though they are not completely linear. The force
77: chains are stabilized by weaker side contacts, and form a disordered
78: branching network resembling a fractal. Typical fractals are
79: invariant under a change of length scale. A quantitative analysis of
80: that scale-invariance is however difficult, in part because there is
81: no sharp distinction between the force chains and the background
82: contacts.
83:
84: In this Letter we show that discriminating between forces of different
85: magnitudes provides a quantitative characterization of force chains.
86: Via an analogy with equilibrium critical phenomena, such a description
87: naturally captures scale invariance: a set of scaling exponents
88: (fractal dimensions) and a scaling function are extracted from a given
89: ensemble of force networks. We applied this approach to
90: two-dimensional granular packings under isotropic pressure generated
91: by molecular dynamics simulations. Although the appearance of the
92: force network varies significantly with pressure and polydispersity,
93: we found that the scaling exponents and the scaling function are
94: independent of these and other parameters, and thus capture universal
95: properties of force networks. We also show that a theoretical model based
96: on the extension of equilibrium statistical mechanics proposed by
97: Edwards faithfully represents this universality class, while some
98: other simple models do not.
99:
100:
101: The force network in a granular packing can be represented as a set of
102: bonds connecting grains in contact. Each bond carries a scalar given
103: by the magnitude of the repulsive force between the grains (see
104: Fig.~\ref{fig1}). To quantify the, visually obvious, force chains in such
105: a network, we choose a threshold $f$, and consider only chains formed
106: by bonds carrying forces larger than $f$. Rather than select a fixed
107: arbitrary value of $f$, we study such structures at different scales
108: by varying the threshold. For small values, most of the grains remain
109: connected, but as the threshold is increased the packing breaks up
110: into disconnected clusters (see Fig.~\ref{fig2}). The extent of each
111: force chain can then be characterized by the size of the corresponding
112: cluster, i.e. the number of mutually connected bonds.
113:
114: The force network varies totally between packings created under the
115: same external conditions, so that a statistical approach is needed.
116: Following the above procedure, force chains in an ensemble of packings
117: can be described by the probability $P(s,f)$ that, at a threshold $f$,
118: a given random grain belongs to a cluster of size $s$. Similar
119: statistical descriptions in terms of clusters are common in the theory
120: of equilibrium phase transitions \cite{fort-kast}, where the threshold
121: $f$ plays the role of temperature. As in a percolation model
122: \cite{stauffer}, a phase transition occurs at the critical threshold
123: $f_c$ above which no cluster connects the opposite boundaries of the
124: system. Around this value, the system exhibits scale invariance, in
125: particular the probability distribution of cluster sizes can be
126: expressed as $P(s,f)\sim s^{-\tau}\rho (s/(f-f_c)^\sigma)$. This
127: scale invariance is {\it universal}: the scaling exponents $\tau$ and
128: $\sigma$, as well as the scaling function $\rho$ are determined only
129: by the global symmetries of the system rather than the microscopic
130: details of the interactions. Equilibrium critical phenomena can
131: therefore be classified in a discrete number of { universality
132: classes} according to the values of the exponents.
133:
134:
135:
136: \begin{figure}
137: \begin{center}
138: \begin{minipage}{0.32\linewidth}
139: \includegraphics [width=0.95\linewidth]{fig1.eps}
140: \end{minipage}
141: \begin{minipage}{0.32\linewidth}
142: \includegraphics [width=0.95\linewidth]{fig2.eps}
143: \end{minipage}
144: \begin{minipage}{0.32\linewidth}
145: \includegraphics [width=0.95\linewidth]{fig3.eps}
146: \end{minipage}
147: \caption{Dependence of the force network on pressure and
148: polydispersity. Grains are represented as gray disks and
149: forces as bonds . The thickness of each bond
150: is proportional to the magnitude of the force.
151: Packing under (a) pressure $p=10^{-4}$ (in rescaled units) and
152: polydispersity $d=20\% $ (b) $p=10^{-1}$ and $d=20\% $, (c) $p=10^{-2}$ and $d=5\% $ \label{fig1} }
153: \end{center}
154: \end{figure}
155:
156:
157: \begin{figure}
158: \begin{center}
159: \begin{minipage}{0.32\linewidth}
160: \includegraphics [width=0.95\linewidth]{fig4}
161: \end{minipage}
162: \begin{minipage}{0.32\linewidth}
163: \includegraphics [width=0.95\linewidth]{fig5.eps}
164: \end{minipage}
165: \begin{minipage}{0.32\linewidth}
166: \includegraphics [width=0.95\linewidth]{fig6.eps}
167: \end{minipage}
168: \caption{ Force chains at different scales. The packing of
169: Fig.~\ref{fig1}~a is repeated here, showing as bold lines forces
170: larger then a threshold $f$. The threshold values are (a)
171: $f=0.5f_c$, (b) $f=f_c$ and (c) $f=1.5f_c$, where $f_c$ is the
172: critical threshold.\label{fig2}}
173: \end{center}
174: \end{figure}
175:
176:
177:
178: Such an analogy with equilibrium critical phenomena suggests the
179: existence of scale invariance around the threshold $f_c$ below which
180: one of the clusters spans the entire packing. The associated scaling
181: exponents and scaling function would provide a novel characterization
182: of the spatial organization of forces in granular matter. To study
183: the existence of scale invariance and universality, we examine the
184: behavior of the $n^{th}$ moment $m_n$ of $P(s,f)$ as function of the
185: threshold $f$ and the system size $N$ (number of contacts in the
186: packing). We look for scaling as function of the system size, which in
187: case of equilibrium critical phenomena takes the form
188: \begin{equation}
189: m_n(f,N)\sim N^{\phi_n}{M_n}\left( [f-f_c]{N}^{1/2\nu}\right),
190: \label{fss}
191: \end{equation}
192: where the
193: scaling function $M_n$ is related via integration to $\rho$, and
194: the scaling exponents are given by $\phi_n=(n+1-\tau)/(\tau-1)$ and
195: $\nu=(\tau-1)/2\sigma$. Here we present a study for $n=2$ because it is the lowest moment to diverge, but the results hold also for higher moments.
196:
197:
198: We have carried out a scaling analysis on packings created by
199: molecular dynamics (MD) simulations (also called discrete elements
200: methods) of a system of polydisperse spheres in a periodic
201: two-dimensional cell subject to an isotropic pressure. The particles
202: are deformable and interact via nonlinear Hertz-Mindlin forces (see
203: Methods: Molecular Dynamics). We first consider strongly polydisperse
204: frictionless packings at low pressure, leading to strongly disordered
205: force networks with clearly visible force chains shown in
206: Fig.~\ref{fig1}~a. For different system sizes, we determined the
207: force-chain clusters at a large number of threshold values and thus
208: computed $m_2(f, N)$, the second moment of clusters sizes (leaving out
209: the largest cluster in each sample). If $m_2$ scales according to
210: Eq.~(\ref{fss}), then plotting $N^{-\phi}m_2$ as function of
211: $(f-f_c){N}^{1/2\nu}$ should lead to a collapse of data for different
212: system sizes on a single curve. Varying $\phi$, $\nu$ and $f_c$, a
213: good collapse is indeed obtained for $\phi=0.89 \pm 0.01$ and $\nu=1.6
214: \pm 0.1$, clearly confirming scale invariance.
215:
216:
217:
218: In order to test the dependence on various parameters, we repeated the
219: procedure for packings created at different pressures,
220: polydispersities, coefficients of friction and force-laws. As the
221: pressure is increased, the grains deform and the number of contacts
222: per particle rises (see Fig. \ref{fig2} a). The force network becomes
223: increasingly uniform in appearance, and the distribution of force
224: magnitudes narrower (see Supplementary Information). Computing
225: $m_2(f)$ for values of pressure spanning three orders of magnitude,
226: we nevertheless find that the optimal data collapse is obtained in
227: the same range $\phi=0.89 \pm 0.01$ and $\nu=1.6 \pm 0.1$
228: independently of the pressure. Decreasing the polydispersity leads
229: to crystallization of the grains (see Fig. \ref{fig2} b), however
230: such ordering of the contact network also leaves the values of the
231: scaling exponents unchanged. Moreover, although friction leads to
232: smaller coordination number in the packing, it does not influence the
233: scaling properties. In this case the forces are not normal to the
234: grains, but tangential forces are typically much smaller then $f_c$
235: (see Supplementary Information), so that they do not affect the
236: scaling of cluster sizes. Finally, particles interacting with
237: harmonic forces instead of non-linear Hertz forces lead once again to
238: the same exponents.
239:
240:
241:
242:
243:
244:
245:
246: The independence of the exponents on pressure, polydispersity,
247: friction and force law is clear evidence of universality. On the
248: other hand, as expected the value of the critical threshold is not
249: universal, it varies between $1.3$ and $1.6$ times the average normal
250: force, and depends on all the parameters. Moreover the width and
251: height of the scaling function vary with the parameters: the width
252: decreases with increasing pressure, while the height increases with
253: increasing polydispersity. A linear rescaling of both axes
254: nevertheless leads to a single collapse for all considered values of
255: parameters, displayed in Fig. \ref{fig3}: similarly to equilibrium
256: critical phenomena, the full scaling function appears to be
257: universal.
258:
259:
260: \begin{figure}
261: \begin{center}
262: \includegraphics[width=0.75\linewidth]{fig7.eps}
263: \caption{The universal scaling function. The mean square of the
264: cluster sizes, omitting the largest cluster, $m_2$, is rescaled as
265: $BN^{-\phi}m_2$ and plotted as function of the rescaled threshold
266: $A(f-f_c)N^{1/2\nu}$. The figure shows the collapse on a single
267: curve of 50 datasets corresponding to various packing parameters and
268: system sizes. A color is associated with each of six different sets
269: of parameters (pressure $p$, polydispersity $d$) used in MD
270: simulations. For each set of parameters, five system sizes
271: (increasing by factors of two) are represented by symbols $\circ,
272: \diamond, \triangle, \times, +$. The collapse was obtained for
273: $\phi=0.89\pm0.1$ and $\nu=1.6\pm0.1$ for all datasets, while the
274: value of $f_c, A$ and $B$ depends on the parameters (but not on the
275: system size). In black we show Monte Carlo results of the Edwards'
276: ensemble on the ``snooker packing'', which gives the same values of
277: $\phi$ and $\nu$.\label{fig3}}
278: \end{center}
279: \end{figure}
280:
281:
282: As the scaling exponents and the scaling function are universal, they
283: can be obtained from simplified models in the appropriate universality
284: class. Years ago, Edwards proposed to generalize the micro canonical
285: principle of thermodynamics to jammed systems such as granular matter
286: \cite{edwards1}. The key idea is to ignore history and treat as
287: equally likely all stable configurations of grains. The approach was
288: found to be successful in slowly flowing granular matter
289: \cite{jam:makse1}, but so far a quantitative confirmation in the case
290: of static granular matter has been lacking.
291:
292: A simple extension of this micro-canonical ensemble to a packing of
293: grains in a fixed geometry was named the ``snooker packing''
294: \cite{fluct:jacco1}. It consists of a hexagonal arrangement of rigid,
295: frictionless and mono-disperse spheres confined within a triangular
296: domain, with the same pressure applied to all sides of the packing.
297: In such a system, force balance alone does not completely determine
298: the forces. Here it is natural to generalize Edwards' approach by
299: treating as equally likely all arrangements of repulsive forces in
300: balance on each grain, while keeping fixed the geometry of contacts.
301: Using Monte Carlo simulations (see Methods: Monte Carlo), we generated
302: force networks on the snooker packing with uniform probability, and
303: examined the scaling of the second moment of cluster size distribution
304: for different system sizes. The optimal collapse is obtained for
305: values of scaling exponents within error bars equal to those found
306: from MD simulations. Moreover, the full scaling function appears to
307: be identical to the one found in MD simulations as shown in
308: Fig. \ref{fig3}.
309:
310: Some other simple models of granular matter lead to different scaling
311: exponents. For example, the extensively studied $q$-model
312: \cite{qmodel0, qmodel, raj}, which implements balance between vertical
313: forces only, leads to $\phi=0.69\pm 0.01$ and $\nu=3.1\pm 0.1$.
314: Moreover, the universality class depends on whether a top-to-bottom
315: propagation of forces is assumed (S.O and B.N., manuscript in
316: preparation). The agreement between the Edwards ensemble and the
317: results of molecular dynamics simulations is thus rather remarkable.
318: In particular it shows that the elasticity of the grains and the
319: details of the force it generates are irrelevant for the shape of the
320: scaling function. More surprisingly, the geometrical randomness of
321: the grain arrangements does not have any bearing on the universal
322: mechanical properties. We expect that the key ingredients are the
323: vectorial balance of forces on each grain, as well as the isotropy of
324: the confining pressure: static packings under shear may lead to
325: different scaling exponents. We conjecture that in three dimensions
326: the exponents and scaling function are equally universal, though
327: different from their two-dimensional equivalent.
328:
329: Our results establish an unexpected connection between static granular
330: matter and equilibrium critical phenomena. The observed scale
331: invariance is in all aspects comparable to equilibrium criticality,
332: and defines a novel universality class of isotropic force networks.
333: This universality class is for example distinct from that of
334: percolation, where the same analysis can be applied. This result
335: implies the existence of long-range correlations between forces, a
336: characteristic of structures such as force chains. We expect force
337: networks in other jammed systems, such as foams and emulsions, to
338: belong to the same universality class. We hope that recent progress
339: in the measurement of inter-grain and inter-bubble forces
340: \cite{majmudar05, fluct:brujic2} will lead
341: to an experimental evaluation of the scaling exponents.
342:
343:
344: \begin{methods}
345:
346: \subsection{Molecular dynamics.}
347: We use molecular dynamics simulations (also known as discrete element
348: method\cite{cundall79} in the engineering literature) to create static
349: granular packings which mimic real granular matter. Starting from a
350: dilute gas phase, we solve Newton's equations where the particles
351: interact with a frictionless repulsive contact force, called the
352: Hertzian interaction\cite{johnson85}. The centers of the particles
353: are restricted to a two-dimensional plane; however, the Hertzian
354: repulsive force is that of three-dimensional elastic spheres:
355: $F=\frac{2}{3}E/(1-\nu^2) R^{1/2}n^{3/2}$, where $E$ and $\nu$ are the
356: Young modulus and Poisson ratio of the particle's material, and
357: $R=R_1R_2/(R_1+R_2)$. The overlap $n=R_1+R_2-r_{12}$ is calculated
358: from the particle's radii $R_1$ and $R_2$ and their centers' distance
359: $r_{12}$. In some of the analysis, we used the frictional
360: Hertz-Mindlin\cite{johnson85} force law. During the evolution we
361: shrink the volume of the periodic box with rate controlled by a
362: feedback loop to achieve a target pressure. As we include some
363: dissipation in the particle interactions, eventually all motion stops
364: and a static granular packing is obtained in mechanical equilibrium.
365: For each set of parameters we use 100 independent packings of 10,000
366: particles in two dimensions, with polydispersity ranging 5\%-20\%
367: (maximum deviation from the mean, using flat distribution of
368: radii). By extracting recursively subsystems with half the number of
369: grains, in total five system sizes were obtained. The pressures are
370: expressed in units of the Young modulus of the particle's
371: material. For more details see ref.~\onlinecite{somfai05}.
372:
373: \subsection{Monte Carlo.}
374: The Edwards ensemble for a granular heap is a uniform probability
375: measure on the forces subject to equations for the balance of each
376: particle and inequalities for the positivity of each contact force.
377: In an appropriately reduced space of forces this represents a constant
378: non-zero probability density inside a convex polyhedron. We sample
379: this distribution by the following stochastic process. Starting with
380: a point inside the polyhedron we select a line in a random direction
381: (from a discrete set of directions). In this direction we calculate
382: the intersections with the polyhedron. A new point is chosen on the
383: line with uniform probability between the intersection
384: points. Averages calculated from such a process converge to unweighted
385: averages over the whole polyhedron. For more details see
386: ref.~\onlinecite{resp:epl}.
387: \end{methods}
388:
389: %% Put the bibliography here, most people will use BiBTeX in
390: %% which case the environment below should be replaced with
391: %% the \bibliography{} command.
392:
393: \bibliography{gran_stat}
394:
395:
396: %% Here is the endmatter stuff: Supplementary Info, etc.
397: %% Use \item's to separate, default label is "Acknowledgements"
398:
399: \begin{addendum}
400: \item We thank B.~ Behringer, W.~Ellenbroek, M.~van Hecke,
401: C.~Goldenberg, W.~van Saarloos and K.~Shundyak for discussions. SO is financially supported by the Dutch research organization FOM (Fundamenteel Onderzoek der Materie).
402: ES is supported by the PHYNECS training network of the
403: European Commission.
404: \item[Competing Interests] The authors declare that they have no
405: competing financial interests.
406: \item[Correspondence] Correspondence and requests for materials
407: should be addressed to B.N.~(email: nienhuis@science.uva.nl).
408: \end{addendum}
409:
410:
411: \begin{figure*}
412: \begin{center}
413: \includegraphics[width=0.95\linewidth]{fig8.eps}
414: \caption*{{\bf Supplementary Figure:} Probability distributions $P(F)$ of force magnitudes in
415: all the
416: systems studied. All the forces are rescaled by the mean normal
417: force $<F_n>$. (a-c), Dependence of $P(F)$ on the
418: parameters varied in the molecular dynamics simulations: (a)
419: pressure, (b) polydispersity and (c) friction. In (c), the
420: distribution of normal forces is represented by the symbols
421: $\bullet$, and the tangential forces by $+$. (d), $P(F)$ in the
422: three
423: lattice models mentioned in the text. }
424: \end{center}
425: \end{figure*}
426:
427:
428:
429:
430:
431:
432: \end{document}
433: