1: \documentclass[12pt]{iopart}
2: \usepackage{graphicx}
3:
4: \newcommand{\beq}{\begin{equation}}
5: \newcommand{\eeq}{\end{equation}}
6: \newcommand{\bea}{\begin{eqnarray}}
7: \newcommand{\eea}{\end{eqnarray}}
8: \renewcommand{\a}{\alpha}
9: \newcommand{\abs}[1]{\vert#1\vert}
10: \newcommand{\all}{{\rm all}}
11: \renewcommand{\b}{{\rm b}}
12: \renewcommand{\d}{{\rm d}}
13: \newcommand{\dpar}{\partial}
14: \renewcommand{\e}{{\rm e}}
15: \newcommand{\eps}{\varepsilon}
16: \newcommand{\eq}{{\rm eq}}
17: \newcommand{\imi}{{\rm i}}
18: \renewcommand{\lim}{{\rm lim}}
19: \newcommand{\lin}{{\rm lin}}
20: \newcommand{\m}{{\rm m}}
21: \newcommand{\mean}[1]{\langle#1\rangle}
22: \newcommand{\s}{\sigma}
23: \newcommand{\prob}{{\rm Prob}}
24: \newcommand{\thr}{{\rm th}}
25: \newcommand{\tot}{{\rm tot}}
26: \newcommand{\w}{\widehat}
27: \newcommand{\D}{\Delta}
28: \newcommand{\F}{{\cal F}}
29: \newcommand{\R}{{\rm R}}
30: \eqnobysec
31:
32: \begin{document}
33: \title[Statistics of bridge geometries]
34: {Cooperativity in sandpiles:\\ statistics of bridge geometries}
35: \author{Anita Mehta\dag, G C Barker\ddag, and J M Luck\S}
36:
37: \address{\dag\ S N Bose National Centre for Basic Sciences, Block JD, Sector 3,
38: Salt Lake, Calcutta 700098, India}
39:
40: \address{\ddag\ Institute of Food Research, Colney Lane, Norwich NR4 7UA, UK}
41:
42: \address{\S\ Service de Physique Th\'eorique\footnote{URA 2306 of CNRS},
43: CEA Saclay, 91191~Gif-sur-Yvette cedex, France}
44:
45: \begin{abstract}
46: Bridges form dynamically in granular media
47: as a result of spatiotemporal inhomogeneities.
48: We classify bridges as linear and complex,
49: and analyse their geometrical characteristics.
50: In particular,
51: we find that the length distribution of linear bridges is exponential.
52: We then turn to the analysis of the orientational distribution
53: of linear bridges and find that, in three dimensions,
54: they are {\it vertically diffusive but horizontally superdiffusive};
55: thus, when they exist, long linear bridges form `domes'.
56: Our results are in good accord with Monte Carlo simulations
57: of bridge structure;
58: we make predictions for quantities that are experimentally accessible,
59: and suggest that bridges are very closely related to force chains.
60: \end{abstract}
61: \pacs{45.70.-n, 61.43.Gt, 89.75.Fb, 05.65.+b, 05.40.-a}
62: \eads{\mailto{anita@bose.res.in},
63: \mailto{barker@bbsrc.ac.uk},
64: \mailto{luck@spht.saclay.cea.fr}}
65: \maketitle
66:
67: \section{Introduction}
68:
69: The surface of a sandpile cloaks within it a vast array
70: of complex structures -- networks of grains whose stability
71: is interconnected, surrounded by pores and necks of void space.
72: Bridges -- arch-like structures, where mutual stabilisation
73: is a principal ingredient -- are prime among these, spanning
74: all manner of shapes and sizes through a granular medium.
75: They can be stable for arbitrarily long times, since the Brownian motion
76: that would dissolve them away in a liquid is absent in sandpiles
77: -- grains are simply too large for the ambient temperature
78: to have any effect.
79: As a result, they can affect
80: the ensuing dynamics of the sandpile; a major mechanism of compaction is
81: the gradual collapse of long-lived bridges in weakly
82: vibrated granular media, resulting
83: in the disappearance of the voids that were earlier enclosed~\cite{usbr}.
84: Bridges are also responsible for the `jamming'~\cite{refs} that
85: occurs, for example, as grains flow out of a hopper.
86:
87: The difficulty of even identifying, leave alone analysing,
88: structures as complex as bridges
89: in a three-dimensional assembly should not be underestimated;
90: such an algorithm now exists, and
91: its use has resulted in the identification
92: and classification of a panoply of bridge configurations
93: generated via numerical simulations~\cite{I,II}.
94: However, even given the presence of such data,
95: it is a far from obvious task to cast it in theoretical terms.
96: In the following, we present a theory for the formation of bridges,
97: which, among other things, is able to explain at least
98: some of the salient features of our numerical data.
99:
100: We first define a bridge.
101: Consider a stable packing of hard spheres under gravity, in three dimensions.
102: Each particle typically rests on three others, which stabilise it,
103: in the sense that downward motion is impeded.
104: {\it A bridge is a configuration of particles in which
105: the three-point stability conditions of two or more particles
106: are linked; that is, two or more particles are mutually stabilised}.
107: They are thus structures which cannot be formed by
108: the {\it sequential} placement of individual particles;
109: they are, however, frequently formed
110: by natural processes such as shaking and pouring,
111: where cooperative effects arise naturally.
112: In a typical packing, upto 70 percent of particles
113: are involved in bridge configurations.
114: The detailed history of a stabilisation process identifies a unique
115: set of bridges, since the stabilising triplet for each sphere is well defined.
116: While it is impossible to determine bridge distributions
117: uniquely from an array of coordinates representing particle positions,
118: we are able via our algorithm to obtain ones
119: that are the most likely to be the result of a given
120: stabilisation process~\cite{I,II}.
121:
122: We now distinguish between {\it linear} and {\it complex} bridges
123: via a comparison of Figures~\ref{fig1} and~\ref{fig2}.
124: Figure~\ref{fig1} illustrates a {\it complex} bridge,
125: i.e., a mutually stabilised cluster of five particles (shown in green),
126: where the stability is provided by six stable base particles (shown in blue).
127: Of course the whole is embedded in a stable network of grains within
128: the sandpile.
129: Also shown is the network of contacts for the particles in the bridge:
130: we see clearly that three of the particles each have two mutual stabilisations.
131: Figure~\ref{fig2} illustrates a seven particle linear bridge with nine
132: base particles.
133: This is an example of a {\it linear} bridge.
134: The contact network shows that this bridge
135: has a simpler topology than that in Figure~\ref{fig1}.
136: Here, all of the mutually stabilised particles are in sequence, as in a string.
137: A linear bridge made of $n$ particles therefore always rests
138: on $n_\b=n+2$ base particles,
139: whereas the number $n_\b$ of base particles of a complex bridge of size $n$
140: varies from one bridge to another, and always obeys $n_\b<n+2$,
141: because of the presence of loops in the contact network of complex bridges.
142:
143: \begin{figure}[htb]
144: \begin{center}
145: \includegraphics[angle=0,width=.38\linewidth]{b5+61z.eps}
146: \includegraphics[angle=0,width=.38\linewidth]{b5+61cnz.eps}
147: \caption{\small
148: A five particle {\it complex bridge}, with six base particles (left),
149: and the corresponding contact network (right).
150: Thus $n=5$ and $n_\b=6<5+2$.}
151: \label{fig1}
152: \end{center}
153: \end{figure}
154:
155: \begin{figure}[htb]
156: \begin{center}
157: \includegraphics[angle=0,width=.38\linewidth]{b7+9z.eps}
158: \includegraphics[angle=0,width=.4\linewidth]{b7+9cnz.eps}
159: \caption{\small
160: A seven particle {\it linear bridge} with nine base particles (left),
161: and the corresponding contact network (right).
162: Thus $n=7$ and $n_\b=9=7+2$.}
163: \label{fig2}
164: \end{center}
165: \end{figure}
166:
167: An important point to note is that bridges can only be formed sustainably
168: in the presence of friction; the mutual stabilisations needed would
169: be unstable otherwise! Although we use Monte Carlo
170: simulations (described below) which do not contain friction
171: explicitly, the configurations we generate correspond to those
172: in nature that include frictional effects.
173: In particular, they generate coordination numbers in a range that is consistent
174: with the presence of friction~\cite{usbr,sam,silbert}.
175: Additionally, we have analysed the configurations of molecular dynamics
176: simulations in the limit of high friction\footnote{We thank
177: Leo Silbert for letting us use configurations generated from his
178: molecular dynamics simulations, where we have found and analysed
179: the distribution of bridges.};
180: these generate distributions of bridges very similar to our own.
181: Our simulations suggest strong analogies
182: between bridge structure and force distributions in granular media,
183: which have been measured both in experiments~\cite{mueth}
184: and in simulations~\cite{sid}.
185: This indicates that bridges might be com\-ple\-men\-tary
186: (and experimentally easier) probes of inhomogeneities in these systems.
187:
188: \section{Simulation details}
189:
190: We have examined bridge structures in hard assemblies that are
191: generated by an established, non-sequential,
192: restructuring algorithm~\cite{usbr,I,II},
193: whose main modelling ingredients involve {\it stochastic} grain displacements
194: and {\it collective} relaxation from them.
195:
196: This algorithm restructures a stable
197: hard sphere deposit in three distinct stages.
198:
199: \begin{itemize}
200:
201: \item[(1)]
202: The assembly is dilated in a vertical direction (with free volume
203: being introduced homogeneously throughout the system), and each
204: particle is given a random horizontal displacement; this models
205: the dilation phase of a vibrated granular medium.
206:
207: \item[(2)]
208: The packing is compressed in a uniaxial external field representing gravity,
209: using a low-temperature Monte Carlo process.
210:
211: \item[(3)]
212: The spheres are stabilised using a steepest descent `drop and roll' dynamics to
213: find a local minimum of the potential energy.
214:
215: \end{itemize}
216: Steps (2) and (3) model the quench phase of the vibration,
217: where particles relax to locally stable positions in the presence of gravity.
218: Crucially, during the third phase, the spheres
219: are able to roll in contact with others; {\it mutual stabilisations} are thus
220: allowed to arise, mimicking collective effects.
221: The final configuration has
222: a well-defined network of contacts and each sphere is supported,
223: in its locally stable position, through point contacts, by a set
224: of three other spheres uniquely defined.
225: In practice, the final
226: configuration may include a few non-stabilised particles.
227:
228: The simulation method recalled above builds a sequence of static packings.
229: Each new packing is built from its predecessor by a random process and
230: the sequence achieves a steady state.
231: In the steady state thus obtained, structural descriptors such as the
232: mean packing fraction and the mean coordination number
233: fluctuate about well-defined mean values.
234: The steady-state mean volume fraction $\Phi$
235: is found to lie typically in the range $\Phi\sim 0.55\hbox{ -- } 0.61$.
236: This volume fraction depends on all the parameters of the simulation,
237: and can therefore be adjusted to any desired value in the above range.
238: The mean coordination number is always $Z\approx 4.6\pm 0.1$.
239: Since for frictional packings the minimal coordination number is
240: $Z=d+1$~\cite{sam}, this confirms that our 3D configurations
241: correspond to those generated in the presence of friction.
242: In fact, the mean coordination number
243: of molecular dynamics configurations of frictional sphere packings
244: is slightly above 4.5, in the limit of a large
245: friction coefficient~\cite{silbert}.
246: We recall that, in the absence of friction, the coordination number
247: is that of an isostatic system, $Z=2d$~\cite{donev},
248: hence $Z=6$ in the present three-dimensional situation.
249:
250: To be more specific, simulations were performed in a rectangular cell
251: with lateral periodic boundaries, and a hard disordered base.
252: The simulations, performed
253: serially on a desktop workstation, have a very time-consuming
254: Monte Carlo compression phase with each simulation taking
255: several days' worth of CPU time.
256: This is necessary in order to have simulation results which are
257: reproducible, without appreciable dependence on
258: Monte Carlo parameters or system size.
259: Each of our configurations includes about $N_\tot=2200$ particles.
260: We have examined ap\-pro\-xi\-ma\-tely $100$ configurations from the
261: steady states of the reorganisation process
262: for the following two values of the packing fraction:
263: $\Phi=0.56$ and $\Phi=0.58$.
264: Segregation is avoided by choosing monodisperse particles; a rough
265: base prevents ordering.
266: A large number of restructuring cycles is needed to
267: reach the steady state for a given shaking amplitude;
268: about 100 stable configurations (picked every 100
269: cycles in order to avoid correlation effects) are saved for future analysis.
270: {}From these configurations, and following
271: specific prescriptions, our algorithm identifies bridges
272: as clusters of mutually stabilised particles~\cite{I,II}.
273:
274: Figure~\ref{figbig} illustrates two characteristic descriptors
275: of bridges used in this work.
276: Along the lines of~\cite{II}, the {\it main axis} of a bridge
277: is defined using a triangulation of its base particles.
278: Triangles are constructed by choosing all possible connected triplets
279: of base particles: the vector sum of their normals
280: is defined to be the direction of the main axis of the bridge.
281: The orientation angle $\Theta$ is defined as the angle
282: between the main axis and the $z$-axis.
283: The {\it base extension} $b$ is defined as the radius of gyration of
284: the base particles about the $z$-axis (which is thus distinct
285: from the radius of gyration about the main axis of the bridge).
286:
287: \begin{figure}[htb]
288: \begin{center}
289: \includegraphics[angle=0,width=.6\linewidth]{b5+6LabelsX2.eps}
290: \caption{\small
291: Definition of the angle $\Theta$ and the base extension $b$ of a bridge.
292: The main axis makes an angle $\Theta$ with the $z$-axis;
293: the base extension $b$ is the radius of gyration of the base particles
294: about the $z$-axis.}
295: \label{figbig}
296: \end{center}
297: \end{figure}
298:
299: \section{Geometrical characteristics.~Size and diameter distribution}
300:
301: In the following, we present statistics
302: for both linear and complex bridges; however it is the former that we analyse
303: in greater detail, both because they are conceptually simpler and
304: because they are in fact more numerous in our simulations.
305: The latter clearly show that as a linear bridge
306: gets longer, it is increasingly likely to develop
307: branches, whereupon it becomes complex~\cite{I,II}.
308:
309: The formation of either type of bridge in shaken granular
310: media is, of course, a dynamical process; we, however,
311: adopt an ergodic viewpoint~\cite{edwards} here, inspired by the
312: ensemble statistics of bridges generated by our simulations.
313: As in polymer theory~\cite{doi}, we differentiate between
314: linear and complex geometries; in our phenomenology, we regard a linear bridge
315: as a random chain which
316: grows in arc length $s$ by the sequential addition of links to its end,
317: while complex bridges, being branched structures, cannot grow sequentially.
318: This replacement of what is in reality a collective phenomenon by
319: a random walk is somewhat analogous to the `tube model'
320: of linear polymers~\cite{doi}: both are simple but efficient effective
321: pictures of very complex problems.
322:
323: In this spirit, we first address the question of the length distribution
324: of linear bridges.
325: We define the length distribution $f_n$ as the probability that a linear bridge
326: consists of exactly $n$ spheres.
327: We make the simplest and the most natural assumption
328: that a bridge of size $n$ remains linear with some probability $p<1$
329: if an $(n+1)^{\rm th}$ sphere is added to it, in the above sense.
330: This simple effective picture therefore leads to the exponential distribution
331: \beq
332: f_n=(1-p)p^n.
333: \label{fexp}
334: \eeq
335: This exponential law is a common feature of all models
336: for random `strings' made of~$n$ units, irrespective of the way they are formed.
337: An illuminating example is provided by percolation clusters
338: in one dimension, which are the prototype of random linear objects~\cite{stah}.
339:
340: The above argument leading to an exponential distribution
341: can alternatively be reformulated as follows, by means of a continuum approach.
342: This will give us the opportunity
343: to introduce a formalism that will be extensively used later.
344: A linear bridge is now viewed as a continuous random curve or `string',
345: parametrised by the arc length~$s$ from one of its endpoints.
346: Within the above picture of a sequential process,
347: we assume that a linear bridge
348: disappears at a constant rate $\a$ per unit length,
349: either by changing from linear to complex or by collapsing.
350: The probability $S(s)$ that a given bridge survives
351: at least up to length~$s$ thus obeys the rate equation
352: $\dot S=-\a S$, and therefore falls off exponentially,
353: according to $S(s)=\exp(-\a s)$.
354: The probability distribution of the length $s$ of linear bridges
355: therefore reads $f(s)=-\dot S(s)=\a\,\exp(-\a s)$.
356: This is the continuum analogue of~(\ref{fexp}).
357:
358: Figure~\ref{figflin} shows a logarithmic plot of numerical data
359: for the length distribution $f_n$ of linear bridges.
360: The data exhibit an exponential fall off of the form~(\ref{fexp}),
361: with $p\approx0.37$, i.e., $\a\approx0.99$\footnote{Despite its proximity
362: to unity, this value of $\a$ has no special significance, as can
363: be seen from the corresponding value of $p$.}.
364: This exponential decay of the distribution of linear bridges is
365: very accurately observed until $n\approx12$.
366:
367: \begin{figure}[htb]
368: \begin{center}
369: \includegraphics[angle=90,width=.58\linewidth]{figflin.eps}
370: \caption{\small
371: Logarithmic plot of the length distribution $f_n$ of linear bridges,
372: against $n$, for $\Phi=0.58$.
373: Full line: least-square fit yielding $p\approx0.37$, i.e., $\a\approx0.99$.}
374: \label{figflin}
375: \end{center}
376: \end{figure}
377:
378: Complex bridges begin to be dominant around $n\approx8$.
379: For a complex bridge, the number $n$ is now referred to
380: as the `size' of the bridge.
381: As complex bridges are branched objects,
382: it is natural to expect that their size distribution resembles
383: that of generic randomly branched objects,
384: such as e.g.~lattice animals or critical percolation clusters~\cite{stah}.
385: These prototypical examples have a power-law fall-off
386: in their size distribution:
387: \beq
388: f_n\sim n^{-\tau}.
389: \label{fpower}
390: \eeq
391: The numerical data shown in Figure~\ref{figfall}
392: clearly show that the size distribution of bridges obeys
393: a power-law behaviour of the form~(\ref{fpower}),
394: and thus behave as generic randomly branched objects.
395: The measured value of the exponent $\tau$
396: seems to coincide with the limiting value $\tau=2$,
397: at and below which the mean size $\mean{n}=\sum n\,f_n$ is divergent.
398:
399: \begin{figure}[htb]
400: \begin{center}
401: \includegraphics[angle=90,width=.6\linewidth]{figfall.eps}
402: \caption{\small
403: Log-log plot of the size distribution $f_n$ of all bridges,
404: against $n$, for two volume fractions.
405: This
406: distribution is dominated by complex bridges, since
407: these predominate after $n\sim 8$.
408: Full line: least-square fit yielding $\tau\approx2.00$.}
409: \label{figfall}
410: \end{center}
411: \end{figure}
412:
413: We now turn to the diameter of linear and complex bridges,
414: which is one of the most important of their geometrical characteristics.
415: The typical diameter $R_n$ of a bridge of size $n$
416: is such that $R_n^2$ is the mean squared end-to-end distance
417: over all the bridges of size $n$.
418: We examine this quantity
419: separately for linear bridges, as well as the ensemble of all bridges.
420: In both cases we expect and find a power-law behaviour
421: \beq
422: R_n\sim n^\nu.
423: \label{nudef}
424: \eeq
425:
426: For linear bridges, $D_\lin=1/\nu_\lin$ can be interpreted as their
427: fractal dimension.
428: The observed value (see Figure~\ref{figrlin}) $\nu_\lin\approx0.66$,
429: i.e., $D_\lin\approx1.51$,
430: lies between those for a self-avoiding walk in two ($\nu_2=3/4$)
431: and three ($\nu_3\approx0.59$) dimensions~\cite{saw}.
432: This seems entirely reasonable, since linear bridges
433: are likely to behave as typical self-avoiding curves.
434: Furthermore, in three dimensions, their effective dimensionality
435: is expected to lie between 2 and 3: they may
436: start off confined to a plane, and then collapse onto each other, due to the
437: effects of vibration.
438:
439: \begin{figure}[htb]
440: \begin{center}
441: \includegraphics[angle=90,width=.54\linewidth]{figrlin.eps}
442: \caption{\small
443: Log-log plot of the typical diameter $R_n$ of linear bridges,
444: against $n-1$, the number of mutually stabilising bonds, for $\Phi=0.58$.
445: Full line: least-square fit yielding $\nu_\lin\approx0.66$,
446: i.e., $D_\lin\approx1.51$.}
447: \label{figrlin}
448: \end{center}
449: \end{figure}
450:
451: For complex bridges, we again find a power law of the form~(\ref{nudef}).
452: The geometrical exponent $\nu_\all\approx0.74$ (see Figure~\ref{figrall})
453: is now naturally compared with the value
454: $\nu\approx0.875$ of 3D percolation~\cite{stah}.
455: (Although the figure shows statistics for `all' bridges,
456: these mostly reflect the behaviour of complex bridges,
457: which dominate beyond about $n\approx8$.)
458:
459: \begin{figure}[htb]
460: \begin{center}
461: \includegraphics[angle=90,width=.55\linewidth]{figrall.eps}
462: \caption{\small
463: Log-log plot of the typical diameter $R_n$ of all bridges, against size $n$,
464: for $\Phi=0.58$.
465: These statistics mainly derive from complex
466: bridges, which dominate beyond about $n\approx8$.
467: Full line: least-square fit to points with $n>5$,
468: yielding $\nu_\all\approx0.74$.}
469: \label{figrall}
470: \end{center}
471: \end{figure}
472:
473: Lastly, we plot in Figure~\ref{fig8} the probability distribution function of
474: the base extension~$b$, for linear bridges; this
475: is clearly a measure of the spanning, and hence the jamming,
476: potential of a bridge (see Figure~\ref{figbig}).
477: We plot also in Figure~\ref{fig8} distribution functions
478: that are conditional on the bridge size $n$.
479: This was done to compare our results with those of~\cite{pak},
480: bearing in mind of course that our results
481: refer to sphere packings in three dimensions whereas
482: theirs refer to disk packings in two.
483: In our case, the conditional distributions are sharply peaked,
484: and only extend over a finite range of~$b$;
485: in theirs the peaks are smaller and the distributions broader.
486: Our results indicate that at least in three dimensions, and upto
487: the lengths of linear bridges probed, bridges of a given length
488: $n$ have a fairly characteristic horizontal extension; we can predict
489: fairly reliably the dimension of the orifice that they might
490: be expected to jam.
491: The results of~\cite{pak} indicate,
492: on the contrary, that there is a wide range of orifices
493: which would be jammed by disk packings of a given length, in two dimensions.
494: Also, for our three-dimensional bridges,
495: the cumulative distribution has a long tail at
496: large extensions, reflecting chiefly the existence of larger bridges
497: than in the data of~\cite{pak}.
498: (We note that this long tail
499: is characteristic of three-dimensional experiments
500: on force chains~\cite{mueth,ohern} which we will discuss
501: at greater length below.)
502: All of this seems entirely reasonable
503: given the more complex topology of three-dimensional space,
504: as well as the greater diversity of possible sphere packings within it.
505:
506: \begin{figure}[htb]
507: \begin{center}
508: \vskip 10pt
509: \includegraphics[angle=0,width=.6\linewidth]{fig6aX.eps}
510: \vskip 10pt
511: \includegraphics[angle=0,width=.6\linewidth]{fig6bX.eps}
512: \caption{\small
513: Distribution of base extensions of bridges, for $\Phi=0.58$.
514: In the top panel, the distributions
515: conditional on the bridge size $n$, for $n=2,4,6$ are marked in blue,
516: while the cumulative distribution is drawn in black.
517: The bottom panel shows the logarithm of the normalised probability distribution,
518: as a function of the normalised variable $b/\mean{b}$.}
519: \label{fig8}
520: \end{center}
521: \end{figure}
522:
523: In the lower part of Figure~\ref{fig8} we have plotted the logarithm
524: of the normalised probability distribution of base extensions,
525: against the normalised variable
526: $b/\mean{b}$, where~$\mean{b}$ is the mean extension of bridge bases.
527: This figure emphasises the exponential tail of the distribution function,
528: and also shows that bridges with small base extensions are unfavoured.
529: It is interesting to note that the form of the normalised distribution
530: resembles that found in force distributions of emulsions~\cite{jasna},
531: and that there are also similarities with normal force distributions
532: in MD simulations of particle packings (see~e.g.~\cite{sid,ohern}).
533: In our simulations, the sharp drop at the origin
534: with a very tangible peak strongly resembles that obtained
535: in the latter case, in the limit of strong
536: deformations~\cite{sid}.
537: Realising that the measured
538: forces~\cite{sid,jasna} propagate through chains of particles,
539: we use this similarity to suggest that our bridges are
540: really just long-lived force chains, which have survived
541: despite strong deformations.
542: We suggest also that with the current availability of 3D
543: visualisation techniques such as NMR~\cite{fukushima},
544: bridge configurations might be an easily measurable,
545: and effective, tool to probe inhomogeneous force networks in shaken sand.
546:
547: \section{Orientational distribution of linear bridges.
548: Theory and simulation}
549:
550: Linear bridges predominate at small $n$, which is
551: after all the regime of interest for finite-sized simulations and experiments.
552: The theory we present below focuses on such bridges; the distribution
553: of their orientation is predicted, and then compared
554: with our simulation results.
555:
556: As before, a linear bridge is modelled as a continuous curve,
557: parametrised by the arc length~$s$ from one of its endpoints.
558: We propose the following effective description of the local
559: and global orientation of this curve.
560: For the sake of simplicity,
561: we choose to focus on the most important degree of freedom,
562: which is the tilt with respect to the horizontal; the azimuthal degree
563: of freedom is therefore neglected.
564: Accordingly, we define the local or `link' angle $\theta(s)$
565: between the direction of the tangent to the bridge at point~$s$
566: and the horizontal,
567: and the mean angle made by the bridge from its origin up to point~$s$,
568: \beq
569: \Theta(s)=\frac{1}{s}\int_0^s\theta(u)\,\d u.
570: \label{Thdef}
571: \eeq
572: The local angle $\theta(s)$ so defined may be either
573: positive and negative; it can even change sign
574: along the random curve which represents a linear bridge.
575: Of course, the orientation angle $\Theta$ measured in our numerical simulations
576: pertains to the three-dimensional realm of real sphere packings.
577: It is there positive by construction,
578: being defined at the end of Section~2 as the angle between the main
579: bridge axis and the $z$-axis.
580:
581: Our simulations show that
582: the angle $\Theta(s)$ typically becomes smaller and smaller
583: as the length $s$ of the bridge increases.
584: Small linear bridges are almost never flat~\cite{I,II}; as they get
585: longer, assuming that they still stay linear, they get `weighed down', arching
586: over as at the mouth of a hopper~\cite{br}.
587: Thus, in addition to our earlier claim that long
588: linear bridges are rare, we claim further here that
589: (if and) when they exist, they typically have flat bases, becoming `domes'.
590:
591: We use these insights to write down equations
592: to investigate the angular distribution of linear bridges.
593: These couple the evolution of the local angle $\theta(s)$
594: with local density fluctuations $\phi(s)$ at point~$s$:
595: \bea
596: &&\dot\theta=-a\theta-b\phi^2+\D_1\eta_1(s),
597: \label{cou1}\\
598: &&\dot\phi=-c\phi+\D_2\eta_2(s).
599: \label{cou2}
600: \eea
601: The effects of vibration on each of $\theta$ and $\phi$ are represented by
602: two independent white noises $\eta_1(s)$, $\eta_2(s)$, such that
603: \beq
604: \mean{\eta_i(s)\eta_j(s')}=2\,\delta_{ij}\,\delta(s-s'),
605: \label{white}
606: \eeq
607: whereas the parameters $a$,~...,~$\D_2$ are assumed to be constant.
608:
609: The phenomenology behind the above equations is the following:
610: the evolution of~$\theta(s)$ is caused by the addition to the bridge
611: of single particles to the bridge,
612: along the lines of our effective sequential picture.
613: The motion of particles within their cages gives rise to the fluctuations
614: of local density at a point $s$; thus $\phi$ can be regarded
615: as an effective {\it collective} coordinate,
616: with~$\theta$ seen as an {\it independent-particle} coordinate~\cite{mln}.
617:
618: The individual dynamics are simple: the
619: first terms on the right-hand side of~(\ref{cou1}),~(\ref{cou2})
620: say that neither $\theta$ nor $\phi$ is allowed to be arbitrarily large.
621: Their coupling via the second
622: term in~(\ref{cou1}) arises as follows: large link
623: angles $\theta$ will be increasingly unstable in the
624: presence of density fluctuations $\phi^2$ of large magnitude,
625: which would to a first approximation `weigh the bridge down', i.e.,
626: decrease the angle $\theta$ locally.
627: Density fluctuations (otherwise known as `dilatancy')
628: also play a key role in the dynamics of
629: the angle of repose of a sandpile, which are described in a concurrent
630: paper~\cite{angle}.
631: There, sandpile collapse is described
632: in terms of an activated process, where an effective temperature
633: competes against configurational barriers generated by dilatancy.
634:
635: Reasoning as above, we therefore anticipate that for low-intensity vibrations
636: and stable bridges, both density fluctuations $\phi(s)$
637: and link angles $\theta(s)$ will be small.
638: Accordingly, we linearise~(\ref{cou1}),
639: obtaining thus an Ornstein-Uhlenbeck equation~\cite{ou,vank}
640: \beq
641: \dot\theta=-a\theta+\D_1\eta_1(s).
642: \label{ou}
643: \eeq
644:
645: Let us make the additional assumption that the initial angle $\theta_0$,
646: i.e., that observed for very small bridges,
647: is itself Gaussian with variance $\s_0^2=\mean{\theta_0^2}$.
648: The angle $\theta(s)$ is then a Gaussian process with zero mean
649: for any value of the length $s$.
650: Its correlation function can be easily evaluated to be
651: \beq
652: \mean{\theta(s)\theta(s')}
653: =\s_\eq^2\,\e^{-a\abs{s-s'}}+(\s_0^2-\s_\eq^2)\e^{-a(s+s')}.
654: \label{thcor}
655: \eeq
656: The characteristic length for the decay of orientation
657: correlations is $\xi=1/a$, and the variance of the link angle reads
658: \beq
659: \mean{\theta^2}(s)=\s_\eq^2+(\s_0^2-\s_\eq^2)\e^{-2as},
660: \eeq
661: where
662: \beq
663: \s_\eq^2=\frac{\D_1^2}{a}
664: \eeq
665: is the `equilibrium' value of the variance of the link angle.
666: Thus
667: as the chain gets longer, the variance of the link angle relaxes
668: from $\s_0^2$ (that for the initial link) to $\s_\eq^2$, in the limit
669: of an infinite bridge.
670:
671: We predict from the above that the mean angle $\Theta(s)$
672: will have a Gaussian distribution, for any fixed length $s$.
673: By inserting~(\ref{thcor}) into~(\ref{Thdef}), we derive its variance:
674: \beq
675: \mean{\Theta^2}(s)=2\s_\eq^2\,\frac{as-1+\e^{-as}}{a^2s^2}
676: +(\s_0^2-\s_\eq^2)\frac{(1-\e^{-as})^2}{a^2s^2}.
677: \label{Thvar}
678: \eeq
679: The asymptotic result
680: \beq
681: \mean{\Theta^2}(s)\approx\frac{2\s_\eq^2}{as}\approx\frac{2\D_1^2}{a^2s}
682: \label{thas}
683: \eeq
684: confirms our earlier statement (see below~(\ref{Thdef}))
685: that a typical long bridge has a base that is almost
686: flat.
687: It can be viewed as consisting of a large number $as=s/\xi\gg1$ of
688: independent `blobs'\footnote{This is qualitatively reminiscent
689: of a similar problem in polymers,
690: where the de Gennes picture of `blob' dynamics in dilute solutions
691: gives way to rigid rod-like behaviour -- see e.g.~\cite{doi}.},
692: each of length $\xi$.
693:
694: The result~(\ref{thas}) has another interpretation.
695: As $\Theta(s)$ is small with high probability for a very long bridge,
696: its extension in the vertical direction reads approximately
697: \beq
698: Z=z(s)-z(0)\approx s\,\Theta(s),
699: \label{zed}
700: \eeq
701: so that $\mean{Z^2}\approx s^2\mean{\Theta^2}(s)\approx2(\D_1/a)^2\,s$.
702: Going back to the discrete formalism, we have therefore
703: \beq
704: Z_n\sim n^{1/2}.
705: \eeq
706: The vertical extension of a linear bridge is thus found to grow with the usual
707: random-walk exponent $1/2$, whereas its horizontal extension
708: exhibits the non-trivial exponent $\nu_\lin\approx0.66$ of
709: Figure~\ref{figrlin}.
710: Thus, {\it long linear bridges are domelike; they are
711: vertically diffusive but horizontally superdiffusive}.
712: We recall
713: that the two-dimensional arches
714: found in~\cite{pak} were diffusive in a vertical
715: direction; our present results show that in three
716: dimensions, the vertical diffusivities of bridge structures
717: remain, and are enriched by a horizontal superdiffusivity.
718: Evidently, jamming in a three-dimensional hopper would be caused
719: by the planar projection of such a {\it dome}.
720:
721: We now compare these predictions with data from our simulations.
722: The numerical data shown in Figure~\ref{figgauss}
723: confirm that the mean angle has, to a good approximation,
724: a Gaussian distribution in this typical situation of bridges of size $n=4$.
725: We recall that the angle $\Theta$ measured in simulations
726: pertains to the three-dimensional world.
727: Hence only the positive half of the Gaussian is used for comparison
728: with results of simulations, and the $\sin\Theta$
729: three-dimensional Jacobian is divided out.
730: It has also been checked to high accuracy
731: (less than 2 percent for a bin size $\delta\varphi=5$ degrees)
732: that the azimuthal angle $\varphi$
733: of the main bridge axis has the expected flat distribution.
734: Figure~\ref{figvar} shows the measured size dependence
735: of the variance $\mean{\Theta^2}(s)$, for both volume fractions.
736: The numerical data are found to agree well with a common fit
737: to the first (stationary) term of~(\ref{Thvar}),
738: with a common value of the parameters $\s_\eq^2=0.093$ and $a=0.55$.
739: The `transient' effects of the second term of~(\ref{Thvar}) are invisible
740: with the present accuracy (see below).
741: This observed agreement therefore provides a first confirmation
742: of the validity of our theory, quite remarkable given
743: that it is entirely independent of the simulations.
744: We conclude that in spite of its simplicity,
745: our theory captures the main features of linear bridges.
746:
747: \begin{figure}[htb]
748: \begin{center}
749: \includegraphics[angle=90,width=.57\linewidth]{figgauss.eps}
750: \caption{\small
751: Plot of the normalised distribution of the mean angle $\Theta$
752: (in radians) of linear bridges of size $n=4$, for both volume fractions.
753: The $\sin\Theta$ Jacobian has been duly divided out,
754: explaining thus the larger statistical errors at small angles.
755: Full lines: common fit to (half) a Gaussian law.}
756: \label{figgauss}
757: \end{center}
758: \end{figure}
759:
760: \begin{figure}[htb]
761: \begin{center}
762: \includegraphics[angle=90,width=.62\linewidth]{figvar.eps}
763: \caption{\small
764: Plot of the variance of the mean angle of a linear bridge,
765: against size~$n$, for both volume fractions.
766: Full line: common fit to the first (stationary) term of~(\ref{Thvar}),
767: yielding $\s_\eq^2=0.093$ and $a=0.55$.
768: The `transient' effects of the second term of~(\ref{Thvar}) are invisible
769: with the present accuracy.}
770: \label{figvar}
771: \end{center}
772: \end{figure}
773:
774: We end this section with the following remarks.
775: First, more subtle effects,
776: including the effects of transients via the second term of~(\ref{Thvar}),
777: and the dependence of the parameters $\s_\eq^2$ and~$a$
778: on the packing fraction $\Phi$,
779: are deserving of further investigation.
780: As a preface to our second remark,
781: we recall that in our phenomenological picture
782: linear bridge formation proceeds sequentially,
783: while complex bridges with their branched structures
784: form collectively, in space.\footnote{We reiterate once again
785: that this effective picture replaces the complexity
786: of the full dynamical picture, where {\it all} bridges form cooperatively.}
787: We might expect that with increasing density~$\Phi$, branched structures
788: would become more and more common; linear bridge formation, with its
789: `sequential' progressive attachment of independent blobs would then
790: become more and more rare.
791: Our theory should therefore cease to hold
792: at a limit packing fraction~$\Phi_\lim$,
793: which is reminiscent of the {\it single-particle
794: relaxation threshold density}~\cite{johannes},
795: at which the dynamics of granular compaction crosses over from being
796: single-particle (sequential) to~collective.
797: Finally, while we have
798: allowed for the local variability of density fluctuations
799: $\phi$ via~(\ref{cou2}), we have so far not explicitly coupled
800: these to bridge orientations.
801: All of these issues form part of ongoing work, since much more numerical data
802: is needed before a detailed comparison with theory is possible.
803:
804: \section{Discussion}
805:
806: In the above, we have classified bridge structures
807: in granular media as either linear or complex,
808: analysed their structures as obtained in our simulations,
809: and obtained values for some of their more important geometrical exponents.
810: We have also presented a theory for their orientational correlations,
811: which is in reasonably good agreement with our simulations,
812: and more importantly, makes predictions that can
813: be observed experimentally.
814: Do linear bridges really predominate at short lengths? Do they really
815: cross over to being domelike at large lengths,
816: if they survive? The answers to these and other
817: questions, if obtained experimentally, will not just
818: satisfy what is arguably mere scientific curiosity.
819: Our investigations above suggest that {\it long-lived bridges
820: are natural indicators of sustained inhomogeneities in granular systems}.
821: Consequently, experimental and theoretical
822: explorations of geometry and dynamics in stable bridge networks
823: will have immediate implications for issues such as
824: hysteresis and cooperative stability, which are some
825: of the most visible manifestations of complexity in granular media.
826:
827: \ack
828:
829: AM warmly thanks SMC-INFM
830: (Research and Development Center for Statistical Mechanics and Complexity,
831: Rome, Italy), and the Service de Physique Th\'eorique, CEA Saclay,
832: where parts of this work were done.
833: GCB acknowledges support from the Biotechnology
834: and Biological Sciences Research Council UK and thanks Luis Pugnaloni
835: for help with the figures.
836: Kirone Mallick is gratefully acknowledged for fruitful discussions.
837:
838: \Bibliography{99}
839:
840: \bibitem{usbr}
841: Mehta A and Barker G C 1991 Phys. Rev. Lett. {\bf 67} 394
842: \nonum
843: Barker G C and Mehta A 1992 Phys. Rev. A {\bf 45} 3435
844:
845: \bibitem{refs}
846: Jaeger H M, Nagel S R, and Behringer R P 1996 Rev. Mod. Phys. {\bf 68} 1259
847: \nonum
848: de Gennes P G 1999 Rev. Mod. Phys. {\bf 71} S374
849:
850: \bibitem{I}
851: Pugnaloni L A, Barker G C, and Mehta A 2001 Adv. Complex Syst. {\bf 4} 289
852:
853: \bibitem{II}
854: Pugnaloni L A and Barker G C 2004 Physica A {\bf 337} 428
855:
856: \bibitem{sam}
857: Edwards S F 1998 Physica {\bf 249} 226
858:
859: \bibitem{silbert}
860: Silbert L E \etal 2002 Phys. Rev. E {\bf 65} 031304
861:
862: \bibitem{mueth}
863: Liu C H \etal 1995 Science {\bf 269} 513
864: \nonum
865: Mueth D M, Jaeger H M, and Nagel S R 1998 Phys. Rev. E {\bf 57} 3164
866:
867: \bibitem{sid}
868: Erikson J M \etal 2002 Phys. Rev. E {\bf 66} 040301
869: \nonum
870: O'Hern C S \etal 2002 Phys. Rev. Lett. {\bf 88} 075507
871:
872: \bibitem{donev}
873: Donev A \etal 2004 Science {\bf 303} 990
874:
875: \bibitem{edwards}
876: Edwards S F 2003 in {\it Challenges in Granular Physics} edited by Mehta A
877: and Halsey T C (Singapore: World Scientific)
878:
879: \bibitem{doi}
880: Doi M and Edwards S F 1986
881: {\it The Theory of Polymer Dynamics} (Oxford: Clarendon)
882:
883: \bibitem{stah}
884: Stauffer D and Aharony A 1992 {\it Introduction to Percolation Theory}
885: 2nd ed (London: Taylor and Francis)
886:
887: \bibitem{saw}
888: des Cloizeaux J and Jannink G 1990 {\it Polymers in Solution.
889: Their Modelling and Structure} (Oxford: Clarendon)
890:
891: \bibitem{pak}
892: To K, Lai P Y, and Pak H K 2001 Phys. Rev. Lett. {\bf 86} 71
893:
894: \bibitem{ohern}
895: O'Hern C S \etal 2001 Phys. Rev. Lett. {\bf 86} 111
896: \nonum
897: Landry J W \etal 2003 Phys. Rev. E {\bf 67} 041303
898:
899: \bibitem{jasna}
900: Brujic J \etal 2003 Physica A {\bf 327} 201
901:
902: \bibitem{fukushima}
903: see chapters by Fukushima E and Seidler G T \etal
904: 2003 in {\it Challenges in Granular Physics} edited by Mehta A
905: and Halsey T C (Singapore: World Scientific)
906:
907: \bibitem{br}
908: Brown R L and Richards J C 1966 {\it Principles of Powder Mechanics}
909: (Oxford: Pergamon)
910:
911: \bibitem{mln}
912: Mehta A, Needs R J, and Dattagupta S 1992 J. Stat. Phys. {\bf 68} 1131
913: \nonum
914: Mehta A, Luck J M, and Needs R J 1996 Phys. Rev. E {\bf 53} 92
915: \nonum
916: Hoyle R B and Mehta A 1999 Phys. Rev. Lett. {\bf 83} 5170
917:
918: \bibitem{angle}
919: Luck J M and Mehta A 2004 preprint cond-mat/0407201
920:
921: \bibitem{ou}
922: Uhlenbeck G E and Ornstein L S 1930 Phys. Rev. {\bf 36} 823
923: \nonum
924: Wang M C and Uhlenbeck G E 1945 Rev. Mod. Phys. {\bf 17} 323
925:
926: \bibitem{vank}
927: van Kampen N G 1992 {\it Stochastic Processes in Physics and Chemistry}
928: (Amsterdam: North-Holland)
929:
930: \bibitem{johannes}
931: Berg J and Mehta A 2001 Europhys. Lett. {\bf 56} 784
932:
933: \endbib
934: \end{document}
935: