1: \documentclass[aps,preprint,epsfig,multicol,nofootinbib,showpacs]{revtex4}
2: %\documentclass[aps,epsfig,floatfix,twocolumn,nofootinbib,showpacs]{revtex4}
3: \usepackage{graphicx}
4: \usepackage{latexsym}
5:
6: % 6 LINES PER INCH
7: \renewcommand{\baselinestretch}{.87}
8: % DOUBLE SPACE
9: %\renewcommand{\baselinestretch}{2}
10: % TRIPLE SPACE
11: %\renewcommand{\baselinestretch}{3}
12:
13: \begin{document}
14:
15: \title{Dry Friction Avalanches: Experiment and Robin Hood model}
16:
17: \author{Sergey V. Buldyrev$^1$, John Ferrante$^2$, and Fredy R. Zypman$^1$}
18: \affiliation{
19: $^1$Department of Physics, Yeshiva University
20: 2495 Amsterdam Avenue, New York, NY 10033\\
21: $^2$NASA-Glenn Research Center,
22: 21000 Brookpark Road, Cleveland, OH 44135, USA\\}
23: \date{bfz.tex ~~~8~September~2005}
24:
25: \begin{abstract}
26:
27: This paper presents experimental evidence and theoretical models supporting
28: that dry friction stick-slip is described by self-organized criticality. We
29: use the data, obtained with a pin-on-disc tribometer set to measure lateral
30: force to examine the variation of the friction force as a function of time.
31: We study nominally flat surfaces of aluminum and steel. The probability
32: distribution of force jumps follows a power law with exponents $\mu$ in the
33: range 2.2 -- 5.4. The frequency power spectrum follows a $1/{f^\alpha}$
34: pattern with $\alpha$ in the range 1 -- 2.6. In addition, we present an
35: explanation of these power-laws observed in the dry friction experiments
36: based on the Robin Hood model of self organized criticality. We relate the
37: values of the exponents characterizing these power laws to the critical
38: exponents $D$ an $\nu$ of the Robin Hood model. Furthermore, we numerically
39: solve the equation of motion of a block pulled by a spring and show that at
40: certain spring constant values the motion is characterized by the same power
41: law spectrum as in experiments. We propose a physical picture relating the
42: fluctuations of the force with the microscopic geometry of the surface.
43:
44: \end{abstract}
45:
46: \pacs{05.65.+b,46.55.+d,64.60.Ht}
47:
48: \maketitle
49:
50: \section{Introduction}
51:
52: There are experimental and theoretical studies suggesting that certain far
53: from equilibrium systems with many degrees of freedom naturally organize in a
54: critical state, releasing energy through rapid relaxation events (avalanches)
55: of different sizes, these sizes being distributed according to a power law
56: probability density. Examples of such behavior are found in earthquakes
57: \cite{Hallgass, Elmer}, biological systems \cite{Sole}, the stock market
58: \cite{Plerou}, rainfall \cite{Peters}, and friction
59: \cite{Slanina,Ciliberto,Vallette}. All these phenomena share the features of
60: the prototypical sand pile model \cite {Bak} for which the concepts of
61: self-organized criticality (SOC) were first proposed. Recently, the
62: possibility of SOC \cite{Janson} in systems presenting stick-slip due to dry
63: friction has been under scrutiny \cite{Turcotte}. In particular, Slanina
64: \cite{Slanina} presented theoretical attempts to explain dry friction in
65: terms of SOC. However, the central question remains unanswered: to what
66: extent is dry friction stick-slip a manifestation of SOC? The clarification
67: of this issue has practical as well as fundamental implications. From the
68: practical point of view, the power law exponents could be used as parameters
69: to characterize friction and wear of surfaces. From the fundamental point of
70: view, there is a growing interest to understand systems driven far away from
71: equilibrium from a single unifying principle. In addition, there is not yet
72: a full understanding of the dissipation mechanisms in friction. Particularly,
73: an overall description of the topography of the interface would be useful.
74:
75: In the present study, we first present experimental results on stick slip in
76: dry friction using a pin-on-disc arrangement, set to measure lateral forces.
77: The probability distributions of force jumps sizes and the corresponding
78: frequency power spectra for aluminum and steel are examined for evidence of
79: SOC. Second, we present a theoretical explanation of the observed power laws
80: based on the Robin Hood model \cite{Zaitsev,Slanina}, which has been
81: successfully used in the past to study dislocation motion and friction.
82:
83: \section{Experiment}
84:
85: The pin-on-disk tribometer used for these experiments is
86: shown in Fig.~\ref{X1}. This configuration was chosen because it
87: allows for easy replacement of the contacting surfaces, and it is the
88: standard method for measuring friction and wear in unlubricated and
89: lubricated contacts. The apparatus uses a $2.54$ cm diameter disk and a
90: spherical pin with a $0.95$ cm radius machined on its end. The pin is attached
91: to a load arm that is mounted on a gimbal supported at the center
92: through which a load applied at the end of the arm is transferred to
93: the contact zone. A strain-gauge is mounted at the end of the arm to
94: monitor tangential friction force. The tangential force is monitored
95: at a sample rate of 1,000 scans/sec and conversion is done using a
96: 16-bit data acquisition card controlled by LabVIEW. Data is recorded to a text
97: file for later analysis. Frictional force measurements are done on
98: matching aluminum and steel (M50) pin-and-disc tribometers. The
99: signal is collected at 1KHz during 16min, thus collecting $10^6$ points.
100: The first quarter of each data set, or about 4 minutes, is
101: discarded to assure that steady state is reached.
102: In order to drive the system very slowly away from static equilibrium,
103: we select slow rotational speeds in the range 10-20 RPM.
104: Each disk is used for up to
105: four tests by changing the radial position of the pin on the disk.
106: Loads for aluminum range from 250 g to 1000 g. Steel is studied with a 1000 g normal load
107: between pin and disk. Figure~\ref{X2} shows a
108: typical tangential friction force time series. It shows force jumps of
109: various sizes. We first construct the
110: probability distribution of force jumps. Force jumps are taken as those events
111: corresponding to negative changes in the tangential force.
112:
113: We next obtain the probability distributions of the tangential friction
114: forces corresponding to aluminum under various loads and M50 steel.
115: Results of such analysis are presented in Figs. \ref{X3} through
116: \ref{X6}. We observe an approximate linear behavior on
117: the double logarithmic plots suggesting the power law behavior of the
118: distributions $P(F)\sim F^{\mu}$ with the exponents $\mu$ in the range
119: between 2.2 and 5.4.
120:
121:
122: We also compute the power spectra of the tangential friction force time
123: series. We divide the original data ($2^{19}$) into $2^8$ statistically
124: independent non-overlapping data sets of $2^{11}$ points each. Next, we
125: calculate $2^8$ individual power spectra and finally average them to obtain
126: the resulting power spectrum. Specific double-logarithmic plots are shown in
127: Figs.~\ref{X7}-\ref{X10}. The power spectra follow power laws with
128: exponents $1.0\leq\alpha \leq 2.6$. The results of the probability
129: distribution and power spectra analysis are summarized in Table I.
130:
131: \section{Theory}
132:
133: Friction is believed to occur on the atomic scale due to asperities on the
134: surfaces in contacts. The simplest way to model such asperities is to use
135: lattice models in the spirit of the invasion percolation \cite{Guyon}, the
136: sandpile model \cite{Bak}, Bak-Sneppen evolution model \cite{Sneppen}, and
137: Zaitsev's Robin Hood model \cite{Zaitsev}. In general, these models are too
138: crude to provide quantitative agreement with all experimental statistical
139: quantities. However, quantities such as distribution of avalanche sizes are
140: usually described by power laws characterized by exponents belonging to a few
141: distinct universality classes. These exponents may be compared with
142: experimentally found ones. If several models belong to the same universality
143: class, it is reasonable to study the simplest among them, since it will
144: provide the clearest understanding of the physical mechanisms of the
145: phenomenon under study. One such simple and elegant example is the Robin Hood
146: model, which was originally proposed for dislocation movement \cite{Zaitsev}
147: and later was adopted for modeling dry friction \cite{Slanina} by Slanina who
148: added to the original Robin-Hood model several parameters aimed to better
149: capture the dry friction mechanism, but essentially obtained the same type of
150: behavior as the original model had. Here we return to the original Robin
151: Hood model due to its simplicity.
152:
153:
154: The model consist of a $d$-dimensional lattice. Each site $i$ on this
155: lattice at any time step $n$ is characterized by the height $h_i(n)$
156: which we assume to be the height of an atomic scale asperity at a given point
157: of the interface between two bodies in contact. Here we present the model for
158: $d=1$, which is an appropriate choice to model slide friction but the
159: analytical treatment is the same in any dimension, although the physically
160: relevant cases are only $d=1$ and $d=2$. As the bodies slide against each
161: other, the asperity with the maximal height is destroyed and some random
162: number of atoms from this asperity is distributed among the neighboring
163: asperities. To be specific, at each time step the site $i$ with maximal
164: height $h_m(n)=\max h_i(n)$ is found and the new heights are determined
165: according to the following rule: $h_m(n+1)=h_m(n)-r(n)$ and $h_{m\pm1}(n+1)=
166: h_{m\pm1}(n)+r(n)/2$, where $r(n)$ are independent random variables uniformly
167: distributed between 0 and 1. (Robin Hood determines the richest merchant in
168: the market, robs him by a random amount $r(n)$ and distributes it equally
169: among the neighbors without leaving anything for himself). If we assume
170: periodic boundary conditions so that the sites with $i=0$ and $i=L$ are
171: equivalent, the total amount of matter $\sum_{i=0}^L h_i(n)$ is conserved and
172: we can assume it to be zero. The distance between the surfaces at a given
173: site $i$ can be determined as $h_m(n)-h_i(n)$.
174:
175: The particular details of the model such as the
176: distribution of $r(n)$ or the rule of dividing it among neighbors can vary, but
177: the model still retains its SOC behavior . The critical exponents appear to be
178: sensitive to the details of dividing $r(n)$, for example, an exactly
179: solvable asymmetric model (in which all the profit is given to the site on
180: the left) \cite{Maslov95} belongs to a different universality class.
181:
182: It has been shown\cite{Maslov,Paczuski} that in a wide class of
183: depinning SOC models, all the critical exponents can be expressed in
184: terms of the two main exponents : avalanche dimension $D$ and
185: correlation exponent $\nu$. The avalanche of threshold $h_0$ is
186: defined as sequence of time steps during which the hight of the
187: maximal asperity is above $h_0$. Namely, if $h_m(n_0)\geq h_0$ and
188: $h_m(n_0+s)\geq h_0$ while for $n_0<n<n_0+s$, $h_m(n)<h_0$, the sequence
189: $n=n_0+1,.., n_0+s$ is called a punctuating avalanche of threshold
190: $h_0$ and mass $s$. The avalanche dimension
191: describes how the avalanche mass $s$, scales with the horizontal
192: dimension of the avalanche $R$. To be more precise, the mass
193: distribution of forward avalanches with threshold $h_0$, scales as
194: %
195: \begin{equation}
196: P_s(s)\sim s^{-\tau_s}g_s(s(h_0-h_c)^{D\nu})
197: \label{P_s}
198: \end{equation}
199: %
200: and the distribution of the avalanche horizontal size, $R$, scales as
201: %
202: \begin{equation}
203: P_R(R)\sim R^{-\tau_R}g_R(R(h_0-h_c)^{\nu}),
204: \label{P_R}
205: \end{equation}
206: %
207: where $h_c \approx 0.114$ is the critical height,
208: %
209: \begin{equation}
210: \tau_s=1+(d-1/\nu)/D
211: \label{tau_s}
212: \end{equation}
213: %
214: and
215: %
216: \begin{equation}
217: \tau_R=1+d-1/\nu
218: \label{tau_R}
219: \end{equation}
220: %
221: are Fisher exponents first introduced to characterized cluster distributions in
222: percolation theory \cite{Stauffer}, while
223: $g_s$ and $g_R$ are exponentially decreasing cutoff functions.
224: It has been suggested \cite{Roux}, that the Robin Hood model belongs
225: to the same universality class as the linear interface model, for which
226: the values ($D=2.23$ and $\tau_s=1.13$ in $d=1$; $D=2.725$ and $\tau_s=1.29$
227: in $d=2$) are given in Ref.\cite{Paczuski}. Using these values
228: and Eq. (\ref{tau_s}), one gets $\nu=1.41$ for $d=1$ and $\nu=0.83$ for $d=2$.
229:
230: It can be shown that after the initial equilibration number of time
231: steps $T\sim L^D$,
232: any initial shape of the interface $h_i(0)$ reaches a steady state
233: such that very few $N(L)$ ``rich'' sites have $h_i(n)>h_c\approx 0.114$, where
234: %
235: \begin{equation}
236: N(L)\sim L^{d_f}
237: \label{N_L}
238: \end{equation}
239: %
240: and
241: %
242: \begin{equation}
243: d_f=d-1/\nu
244: \label{d_f}
245: \end{equation}
246: %
247: plays the role of fractal dimension of rich sites. Only those few
248: rich sites have a chance to be robbed. The chance $P_m(h_m)$ that at a
249: given time step, the maximal height is equal to $h_m$ decreases
250: for an infinite system \cite{Maslov} as
251: %
252: \begin{equation}
253: P_m(h_m)=(h_m-h_c)^{\gamma-1},
254: \label{P_r}
255: \end{equation}
256: %
257: where the exponent
258: %
259: \begin{equation}
260: \gamma=1+\nu(D-d)
261: \label{gamma}
262: \end{equation}
263: %
264: characterizes the dependence of the average avalanche size on its
265: threshold $h_0$:
266: $\langle s\rangle\sim (h_0-h_c)^{-\gamma}$.
267:
268: The distribution of
269: heights of the poor sites converges to a smooth distribution on
270: the interval $[h_c-1,h_c]$, while the distribution of the rich sites
271: converges to the distribution with a power law singularity
272: %
273: \begin{equation}
274: P_h(h)\sim(h-h_c)^{-d\nu}.
275: \label{P_h}
276: \end{equation}
277: %
278: This result is not presented in Refs.
279: \cite{Maslov,Paczuski} but can be justified by the following
280: heuristic arguments. Indeed, the number of sites with $h>h_0$ scales as
281: the number of the active sites in an avalanche of threshold $h_0$,
282: and thus scales as $R^{d_f}(h_0)$, where $R(h_0)$ is the cutoff
283: of the avalanche distribution (\ref{P_R}) which scales as
284: %
285: \begin{equation}
286: R(h_0)\sim(h_0-h_c)^{-\nu}.
287: \label{R_h_0}
288: \end{equation}
289: %
290: Thus the probability that $h>h_0$ scales as
291: $(h_0-h_c)^{-d_f\nu}$
292: and the probability density of $h=h_0$ scales as
293: %
294: \begin{equation}
295: P_h(h_0)\sim(h_0-h_c)^{-d_f\nu-1}= (h_0-h_c)^{-d\nu}.
296: \label{P_h1}
297: \end{equation}
298:
299: We can assume that the
300: friction force, $F(n)$, at a given time step is proportional to the
301: number $P_h(h_m(n))\Delta h$ of asperities with heights between
302: $h_m(n)-\Delta h$ and $h_m(n)$:
303: %
304: \begin{equation}
305: F(n)= F_1 P_h(h_m(n))\Delta h,
306: \label{F_h}
307: \end{equation}
308: %
309: where $\Delta h$ is the interaction distance of atomic forces acting between
310: the two surfaces and $F_1$ is a proportionality coefficient, corresponding
311: to the surfaces interaction force at the asperity. Accordingly, the
312: distribution of the friction forces $P(F)$ satisfies the equation $P(F)d
313: F=P_m(h_m)d h_m$, where the random variables $F$ and $h_m$ are linked by
314: Eq. (\ref{F_h}). Taking into account Eqs. (\ref{P_h}) and (\ref{F_h}) we have
315: $dh_m/dF\sim d F^{-1/d\nu}/dF \sim F^{-1/d\nu -1}$. Finally Eq. (\ref{P_r})
316: yields
317: %
318: \begin{equation}
319: P(F)=
320: P_m(h_m(F)){d h_m\over d F}\sim F^{-(\gamma-1)/d\nu}
321: F^{-1/d\nu-1}=F^{-\mu},
322: \label{P_F}
323: \end{equation}
324: %
325: where
326: %
327: \begin{equation}
328: \mu=(D+1/\nu)/d.
329: \label{mu}
330: \end{equation}
331: %
332: For $d=1$, using
333: values of Ref.\cite{Paczuski} we have $P(F)=F^{\mu}$ with $\mu=2.94$ which is
334: consistent with the experimental observations of the density of jump
335: sizes presented here and in Ref.\cite{Zypman}. For $d=2$ we have $\mu=1.96$.
336:
337: In order to test this theoretical predictions, we perform
338: simulations of the one dimensional Robin Hood model. Starting at $n=0$ with
339: a flat interface $h_i(0)=0$, and selecting the first site to rob at
340: random, after $T$ steps we get all $L$ sites of the interface updated
341: at least once. Measuring the average $\langle T\rangle$ for many
342: independent runs for different system sizes, and plotting it versus $L$
343: in a double-logarithmic scale (Fig. \ref{T-L}), we can obtain the
344: avalanche dimension $D$ as the limit of the successive slopes of this
345: graph for $L\to\infty$.
346:
347: A typical shape of the interface at time $n>T$ is presented in
348: Fig. \ref{h-i}. One can see that the height of the majority of sites
349: do not exceed the critical value $h_c\approx 0.114$. Interestingly,
350: the majority of rich sites with heights above the critical barrier are
351: localized in the vicinity of the richest site.
352:
353:
354: Figure \ref{hist}
355: shows the histogram of all the interface heights $P_h(h)$ collected over
356: many time steps after the system has reached the steady state and the
357: histogram of the heights of the robbed sites $P_m(h_m)$. One can see that
358: while $P_h(h)$ dramatically increases as $h\to h_c^+$, no sites below
359: the critical value are robbed and $P_m(h_m)\to 0$ as $h_m\to h_c^+$.
360: In order to find the exponents governing the behavior of these distributions
361: near the critical point, we plot these quantities in a double logarithmic
362: scale as functions of $h-h_c$ (Fig.~\ref{hist}b).
363:
364: Finally we determine the time series of forces, defined as the number
365: of heights between $h_m(n)$ and $h_m(n)-\Delta h$ as function
366: of time. (Fig.~\ref{F_t}).
367: The histogram of this time series is presented
368: in Fig.~\ref{fP_F} in a double logarithmic scale.
369: The slope of this plot is $\mu=3.0$, which is consistent with the theoretical
370: prediction (\ref{mu}).
371:
372: Note that the time series $F(n)$ is slightly correlated, which can be
373: demonstrated by the negative slope of its power spectrum $S_F(f)\sim
374: f^{-\alpha}$ in the log-log scale (Fig.\ref{power}). The explanation of this
375: phenomenon is based on the fact that values $h_m(n)$ fluctuate
376: in the vicinity of
377: $h_c$ in a non-trivial way, so that $h_m(n)$ become less than
378: $h_c+\epsilon$ at
379: time steps $n$ separated by intervals distributed according to
380: Eq. (\ref{P_s}). This is because these intervals coincide with
381: avalanches for the threshold $h_0=h_c+\epsilon$.
382: The values of $h_m(n)$ below $h_c+\epsilon$
383: correspond to the large values forces $F(n)$ and thus the intervals between the
384: forces $F(n)$ above certain threshold
385: are also distributed according to Eq. (\ref{P_s}). It can be shown
386: \cite{Paczuski,Lowen}that the exponent $\alpha$ of a time series generated by
387: peaks separated by intervals of zero signal distributed according to a
388: power law as in Eq. (\ref{P_s}) is equal to $\tau_s-1$ for $1<\tau_s<2$.
389: Thus according to Eq. (\ref{tau_s}) $\alpha=(d-1/\nu)/D\approx 0.13$.
390: Indeed, the numerical data of Fig.~\ref{power} give $\alpha\approx 0.14$ in a
391: very good agreement to the above theoretical prediction. However, this value
392: of spectral exponent is much smaller than the values observed experimentally
393: which are in the range between 1 and 2.6.
394:
395: This difference is to be expected since the materials in contact as well as
396: the strain gauge have finite elastic constants and inertia which produce a
397: time delay between the applied force and the displacement record by the
398: tribometer and lead to an effective integration of the input force time
399: series. We would expect that if the materials were infinitely stiff then the
400: experimental force power spectrum should agree with the theoretical
401: predictions. Therefore, we construct a mechanical model of a tribometer,
402: that accounts for these effects.
403:
404: We assume that the pin of the tribometer contacts the sample at time $t$ at a
405: point with coordinate $x(t)$ and it is dragged along the sample by the strain
406: gauge spring with spring constant $k$ attached to the body of the instrument
407: moving along the sample with constant velocity $v_0$, which is equivalent to
408: the rotational speed of the disk. The force measured by the tribometer is
409: thus $k[v_0 t-x(t)]$, which fluctuates as the pin moves against the sample with
410: velocity $v(t)=dx/dt$ and acceleration $a(t)=d^2x/dt^2$. The equation of
411: motion of the pin is thus
412: \begin{equation}
413: m a= (v_0 t-x)k -F(t,v),
414: \label{ma}
415: \end{equation}
416: where $F(t,v)$ is the friction force generated by the highest
417: asperity of the sample and $m$ is the mass of the pin.
418:
419: Now our goal is to relate $F(t,v$) with the input from the Robin Hood
420: model. Note, that the physical time $t$ is not directly proportional to the
421: time step $n$ of the Robin Hood model, but is equal to the sum of the
422: durations of each time step
423: \begin{equation}
424: t=\sum_{i=1}^n t_i,
425: \label{time}
426: \end{equation}
427: where the durations $t_i$ are the times needed for the pin to travel a
428: characteristic distance $\Delta x$, which is the linear size of each
429: asperity. We assume that if the pin moves along the sample by $\Delta x$ ,
430: the current asperity is destroyed and the landscape of the contact between
431: the pin and the sample is rearranged according to the rules of the
432: model. Thus the time step $n_t$ of the Robin Hood model, corresponding to a
433: given moment of time $t$ can be determined as $n_t={\rm int}[x(t)/\Delta x]$,
434: where ${\rm int}[...]$ denotes the integer part of the expression in the
435: brackets.
436:
437: If $v=0$, $F(t,v)={\rm sign}(v_0 t -x)\min [b F(n_t),k(v_0 t -x)]$, where
438: $F(n_t)$ is the input from the Robin Hood model, and $b$ is some material and
439: load-dependent constant. If $v\ne 0$, $F(x,v)= {\rm sign}(v) [b F(n_t)+ \eta
440: v]$, where $\eta>0$ is some dissipative constant. Constant $b$ is
441: proportional to the load and depends on the elastic properties of the
442: material. Introducing dimensionless variables by $x'=x/\Delta x$ and
443: $t'=tv_0/\Delta x$, we arrive to a dimensionless equation
444: \begin{equation}
445: a'=(t'-x')k'-F'(t',v'),
446: \label{ma'}
447: \end{equation}
448: where $k'=k \Delta x^2/m v_0^2$ and $F'(t',v')$ is the same as $F(t,v)$ but
449: the constants $b$ and $\eta$ are changed by $b'=b\Delta x/m v_0^2$, $\eta'=\eta
450: \Delta x/mv_0$.
451:
452: Thus, there are three independent dimensionless parameters
453: of the model: $k'$, $b'$ and $\eta'$. Varying these parameters, we found a
454: wide region in the parameter space in which the power spectrum of the model
455: resembles the experimental one. A typical example of the spectrum for
456: $k'=0.001$, $b'=0.3$ and $\eta'=0.01$ is shown on Fig.(\ref{Sf}). The
457: frequency of the resonance peak is determined by $\sqrt{k'}/2\pi \approx
458: 5\cdot 10^{-3}$. The peak becomes more pronounced as we decrease $\eta'$. The
459: increase in $\eta'$ also increases the slope of the spectrum. The increase of
460: $b'$ at given $k'$ increases the frequency region in which the power spectrum
461: follows the power law but it also increases the absolute value of the slope
462: closer to $2$, a characteristic value of the Brownian motion. In general, an
463: integration of the time series corresponds to the increase of the spectral
464: exponent by 2, so the integration of the white noise produces the Brownian
465: noise. The observed spectral exponent $\alpha=1.45$ suggests that in a
466: certain range of parameters, our model acts as the fractional integrator of
467: the input signal. For a very stiff spring and large dissipation ($k=1$,
468: $b=0.1$, $\eta=1$) the output signal of our equation is not much different
469: from the input time series $F(n_t)$ and we recover the small value of the
470: spectral exponent $\alpha=0.14$.
471:
472: \section{Conclusions}
473:
474: We present experimental results and theoretical arguments that support the
475: presence of self organized criticality in dry sliding friction. The
476: experiments are pin-on-disk friction force traces of aluminum-aluminum and
477: steel-steel systems. In both cases and for a variety of normal loads, the
478: distribution of the friction force jumps and the frequency power spectra are
479: power laws. The theoretical arguments are based on the application of the
480: Robin Hood model to the friction problem. This model provides rules by which
481: the surface profile changes as a function of time. The model introduces a
482: height $h$ that we interpret physically as the height of the asperity. At
483: each time step, atoms from the highest asperity are distributed among
484: neighboring sites. We use the known distribution of heights and of maximal
485: heights $h_m$ of the Robin Hood model to obtain the time series of the
486: friction forces created by the asperities. As the maximum height fluctuates
487: near the critical value, the number of smaller asperities whose heights are
488: within the reach of atomic forces also fluctuates, diverging as $h_m$ comes
489: close to the critical value $h_c$. These smaller asperities correspond to
490: the contact sites and are responsible for the friction force. Specifically,
491: the friction force is proportional to the number of contacts. Thus we
492: propose that the friction force at a given time step is proportional to the
493: probability density of the interface heights at the current value of the
494: maximal height. The statistical distribution of the friction forces is
495: studied both numerically and analytically.
496:
497: We also find that the large forces are bunched in time. This is due to
498: fluctuations of the maximal heights above a constant critical
499: height. When the maximum heights return to the critical value, the forces
500: become large. Thus, the surface waxes and wanes between a situation of large
501: force due to many asperities acting, and a situation of smaller force in
502: which only few asperities are in contact.
503:
504: In addition, we use the time series of
505: forces as an input to the Newton's equation which describes
506: the kinematics of the
507: pin. For stiff or massless materials, the experimental distribution
508: of force jumps should coincide with the theoretical distribution of forces.
509: However, materials have finite mass and elasticity and thus the experimentally
510: measured friction forces differ from the actual
511: forces at the contact. To investigate these effects, we
512: solve this equation numerically for different values of the parameters
513: and find good agreement with the experiment.
514:
515: \section{Acknowledgments}.
516:
517: SVB thanks Yeshiva University for providing the high performance computer
518: cluster that made this work possible. FRZ acknowledges support by Research
519: Corporation through grant CC5786. FRZ thanks Phillip Abel,
520: Mark Jansen, Kathleen Scanlon of NASA-Glenn Tribology Group for
521: collaboration in the initial stages of this project.
522:
523:
524: \begin{thebibliography}{99}
525:
526:
527: \bibitem{Hallgass}
528: R. Hallgass, V. Loreto, O. Mazzella, G. Paladin, L. Pietronero, Phys. Rev. E {\bf 56}, 1346 (1997).
529: \bibitem{Elmer}
530: F.-J. Elmer, Phys. Rev. E {\bf 56}, R6225 (1997).
531: \bibitem{Sole}
532: R.V. Sole, S. C. Manrubia, Phys. Rev. E {\bf 54}, R42 (1996).
533: \bibitem{Plerou}
534: V. Plerou, P. Gopikrishnan, L.A. Nunes Amaral, M. Meyer, and H.E. Stanley, Phys. Rev. E {\bf 60} 6519 (1999).
535: \bibitem{Peters}
536: O. Peters, C. Hertlein, K. Christensen, Phys. Rev. Lett. {\bf 88}, 018701-1 (2002).
537: \bibitem{Slanina}
538: F. Slanina, Phys. Rev. E {\bf 59}, 3947 (1999).
539: \bibitem{Ciliberto}
540: S. Ciliberto, C. Laroche, J. Phys. I France {\bf 4}, 223 (1994)
541: \bibitem{Vallette}
542: D.P. Vallette, J.P. Gollub, Phys. Rev. E {\bf 47}, 820 (1993)
543: \bibitem{Bak}
544: P. Bak, C. Tang, and K. Wiesenfeld, Phys. Rev. Lett. {\bf 59} 381 (1987).
545: \bibitem{Janson}
546: H. J. Jensen , {\it Self-Organized Criticality: Emergent Complex Behavior in
547: Physical and Biological Systems, Cambridge Lecture Notes in Physics (No. 10)} (London, 1998).
548: \bibitem{Turcotte}
549: D.L. Turcotte, Rep. Prog. Phys. {\bf 62}, 1377 (1999).
550: \bibitem{Zaitsev}
551: S. I. Zaitsev, Physica A {\bf 189} 411 (1992).
552: \bibitem{Guyon}
553: S. Raux and E. Guyon, J. Phys. A {\bf 22} 3693 (1989)
554: \bibitem{Sneppen}
555: P. Bak and K. Sneppen, Phys. Rev. Lett. {\bf 71}, 4083 (1993)
556: \bibitem{Maslov95}
557: S. Maslov and Y.-C. Zhang, Phys. Rev. Lett. {\bf 75} 1550 (1995)
558: \bibitem{Maslov}
559: S. Maslov, Phys. Rev. Lett. {\bf 74} 562 (1995)
560: \bibitem{Paczuski}
561: M. Paczuski, S. Maslov, and P. Bak , Phys. Rev. E {\bf 53} 414 (1996).
562: \bibitem{Stauffer} D. Stauffer and A. Aharony, {\it Introduction to
563: Percolation Theory\/} (Taylor \& Francis, Philadelphia, 1994).
564: \bibitem{Roux}
565: S. Raux and A. Hansen, J. Physique I {\bf 4} 515 (1994)
566: \bibitem{Zypman}
567: F. R. Zypman,J. Ferrante, M. Jansen, K. Scanlon, and P. Abel,
568: J. Phys.-Cond. Mat.{\bf 15} L191 (2003)
569: \bibitem{Lowen}
570: S. B. Lowen and M. C. Teich, Phys. Rev. E, {\bf 47}, 992 (1993).
571:
572: \end{thebibliography}
573: \eject
574: \begin{figure}[htb]
575: %\centerline
576: \includegraphics[width=12.0cm,height=12.0cm,angle=0]{ExperimentalSetup.eps}
577: %}
578: \caption{Pin-on-disk tribometer. The arrow points at the
579: location where the spherical pin and the disk touch. The disk lies
580: horizontally while the pin attached to the arm, rests above it.
581: \label{X1}}
582: \end{figure}
583:
584: \begin{figure}[htb]
585: %\centerline
586: \includegraphics[width=12.0cm,height=12.0cm,angle=0]{Trace.eps}
587: %}
588: \caption{Typical signal from the tribometer. The effective
589: spring constant of the apparatus is $~1$g/${\mu}$m, giving the
590: largest force jumps as a few hundred ${\mu}m$.
591: \label{X2}}
592: \end{figure}
593:
594: \begin{table}[htb]
595: \caption{\label{lattice} The values of exponents $\mu$ characterizing the
596: power law behavior of the distribution of the force jump sizes and spectral
597: exponents $\alpha$ characterizing the power spectrum of the friction force
598: time series for different materials and loads.}
599: \begin{ruledtabular}
600: \begin{tabular}{llll}
601: Material & Load & $\mu$ & $\alpha$ \\
602: \hline
603: M50 & 1000g. & 3.5 & 3.0 \\
604: Al & 250g. & 5.4 & 1.0 \\
605: Al & 750g. & 2.2 & 1.5 \\
606: Al & 1000g. & 3.2 & 1.3 \\
607: \end{tabular}
608: \end{ruledtabular}
609: \end{table}
610:
611: \begin{figure}[htb]
612: %\centerline
613: \includegraphics[width=12.0cm,height=12.0cm,angle=0]{Fig03.eps}
614: %}
615: \caption{Probability density for jump size distribution on
616: steel M50.
617: \label{X3}}
618: \end{figure}
619:
620: \begin{figure}[htb]
621: %\centerline
622: \includegraphics[width=12.0cm,height=12.0cm,angle=0]{Fig04.eps}
623: %}
624: \caption{Probability density for jump size distribution
625: on aluminum with a normal load of 250g.
626: \label{X4}}
627: \end{figure}
628:
629: \begin{figure}[htb]
630: %\centerline
631: \includegraphics[width=12.0cm,height=12.0cm,angle=0]{Fig05.eps}
632: %}
633: \caption{As in Fig.~\ref{X4} with normal load of 750g.
634: \label{X5}}
635: \end{figure}
636:
637: \begin{figure}[htb]
638: %\centerline
639: \includegraphics[width=12.0cm,height=12.0cm,angle=0]{Fig06.eps}
640: %}
641: \caption{As in Fig.~\ref{X4} with normal load of 1000g.
642: \label{X6}}
643: \end{figure}
644:
645: \begin{figure}[htb]
646: %\centerline
647: \includegraphics[width=12.0cm,height=12.0cm,angle=0]{Fig07.eps}
648: %}
649: \caption{Power spectrum of jump sizes for a
650: steel M50 sample.
651: \label{X7}}
652: \end{figure}
653:
654: \begin{figure}[htb]
655: %\centerline
656: \includegraphics[width=12.0cm,height=12.0cm,angle=0]{Fig08.eps}
657: %}
658: \caption{Power spectrum of jump sizes for
659: an aluminum sample with a normal load of 250g.
660: \label{X8}}
661: \end{figure}
662:
663: \begin{figure}[htb]
664: %\centerline
665: \includegraphics[width=12.0cm,height=12.0cm,angle=0]{Fig09.eps}
666: %}
667: \caption{As in Fig.~\ref{X8} with a normal load of 750g.
668: \label{X9}}
669: \end{figure}
670:
671: \begin{figure}[htb]
672: %\centerline
673: \includegraphics[width=12.0cm,height=12.0cm,angle=0]{Fig10.eps}
674: %}
675: \caption{As in Fig.~\ref{X8} with a normal load of 1000g.
676: \label{X10}}
677: \end{figure}
678:
679: \begin{figure}[htb]
680: \includegraphics[width=12.0cm,height=12.0cm,angle=270]{T-L.eps}
681: \caption{ Double logarithmic plot of the average equilibration time
682: $\langle T\rangle$ versus system size $L=2^3,2^4, ..., 2^{12}$.
683: The inset shows the successive slopes of the main graph versus $1/L$.
684: The intercept $D=2.23$, agrees with the data of Ref.\cite{Paczuski}.
685: \label{T-L}}
686: \end{figure}
687:
688: \begin{figure}[htb]
689: \includegraphics[width=12.0cm,height=12.0cm,angle=270]{landscape.eps}
690: \caption{ A typical shape of the steady state interface. The horizontal
691: line shows the critical height.
692: \label{h-i}}
693: \end{figure}
694:
695: \begin{figure}[htb]
696: \includegraphics[height=9.0cm,angle=270]{hist.eps}
697: \includegraphics[height=9.0cm,angle=270]{hist-log.eps}
698: \caption{ (a) The semi-logarithmic plot of the distribution of heights
699: $P_h(h)$ of all sites (solid line) and the distribution of heights
700: $P_m(h_m)$ of robbed sites (dashed bold line).
701: Vertical dotted line shows the position of the critical height $h_c=0.114$.
702: (b) Double logarithmic plot of the same quantities plotted as functions of
703: $h-h_c$. The slopes of the curves in the fitted regions are
704: in agreement with Eq.(\ref{P_h}) $-d\nu=-1.4$ and Eq. (\ref{P_r}) $\nu(D-1)=1.72$.
705: \label{hist}}
706: \end{figure}
707:
708: \begin{figure}[htb]
709: %\centerline
710: \includegraphics[width=12.0cm,height=12.0cm,angle=270]{F_t.eps}
711: %}
712: \caption{Time series of $F(t)$ for L=4096, $\Delta h=2^{-10}$
713: \label{F_t}}
714: \end{figure}
715:
716: \begin{figure}[htb]
717: %\centerline
718: \includegraphics[width=12.0cm,height=12.0cm]{P_F.eps}
719: %}
720: \caption{Distribution of forces $P_F(F)$ for L=8192, $\Delta h=2^{-10}$.
721: The slope of the straight line fit is $-\mu=-3$.
722: \label{fP_F}}
723: \end{figure}
724:
725: \begin{figure}[htb]
726: %\centerline
727: \includegraphics[width=12.0cm,height=12.0cm,angle=270]{power.eps}
728: %}
729: \caption{Power spectrum of the time series F(t) presented in Fig.(\ref{F_t}).
730: The slope of the straight line fit is $-\alpha -0.14$.
731: \label{power}}
732: \end{figure}
733:
734: \begin{figure}[htb]
735: %\centerline
736: \includegraphics[width=12.0cm,height=12.0cm,angle=0]{Sf.eps}
737: %}
738: \caption{Power spectrum of the time series produced by Eq.(\ref{ma'}) for
739: $k'=0.001$, $b'=0.3$ end $\eta'=0.01$.
740: \label{Sf}}
741: \end{figure}
742:
743:
744: \end{document}
745:
746:
747:
748:
749:
750:
751:
752:
753:
754: