PHYSICAL REVIEW E 77, 036123 共2008兲

Dependence of friction on roughness, velocity, and temperature

Yi Sang,1 Martin Dubé,2 and Martin Grant1

1
Physics Department, McGill University, Rutherford Building, 3600 rue University, Montréal, Québec, Canada H3A 2T8
2
Centre Intégré en Pâtes et Papiers, Université du Québec a Trois-Rivières,
Case Postale 500, Trois-Rivières, Québec, Canada G9A 5H7
共Received 25 July 2007; published 27 March 2008兲
We study the dependence of friction on surface roughness, sliding velocity, and temperature. Expanding on
the classic treatment of Greenwood and Williamson, we show that the fractal nature of a surface has little
influence on the real area of contact and the static friction coefficient. A simple scaling argument shows that the
static friction exhibits a weak anomaly ␮ ⬃ A−0 ␹/4, where A0 is the apparent area and ␹ is the roughness
exponent of the surface. We then develop a method to calculate atomic-scale friction between a microscopic
asperity, such as the tip of a friction force microscope 共FFM兲 and a solid substrate. This method, based on the
thermal activation of the FFM tip, allows a quantitative extraction of all the relevant microscopic parameters
and reveals a universal scaling behavior of atomic friction on velocity and temperature. This method is
extended to include a soft atomic substrate in order to simulate FFM scans more realistically. The tip is
connected with the support of the cantilever by an ideal spring and the substrate is simulated with a ball-spring
model. The tip and substrate are coupled with repulsive potentials. Simulations are done at different tempera-
tures and scanning velocities on substrates with different elastic moduli. Stick-slip motion of the tip is ob-
served, and the numerical results of the friction force and distribution of force maxima match the theoretical
framework.

DOI: 10.1103/PhysRevE.77.036123 PACS number共s兲: 81.40.Pq, 46.55.⫹d, 07.79.⫺v, 83.50.Lh

I. INTRODUCTION on the length scale of observation. Since many real surfaces

The empirical laws of friction between macroscopic sur-
faces were discovered long ago by Da Vinci, Amonton, and
Coulomb. They found that friction is 共i兲 independent of the
apparent contact area, 共ii兲 proportional to the normal load,
and 共iii兲 independent of the sliding velocity 关1兴. Although
these laws are modified by plasticity 关2兴, they remain essen-
tially valid, and friction at the macroscopic level is now well
understood, for both dry rough 关3兴 and lubricated surfaces
关4兴. Nevertheless, the subject continues to be reconsidered,
particularly in connection to developments in the field of
nanotribology 关5,6兴, to the new experimental possibilities of
studying friction at the atomistic level opened by the atomic
force microscope 关7–13兴, and to the fractal structure inherent
in several surfaces 关5,14–24兴. The contact problem between
a plane and a rough surface is also of fundamental impor-
tance in several processes, such as printing on paper and
board 关25,26兴.
Macroscopic solids sliding against each other interact
mainly through many atomic-scale asperities present on their
surface. Although it is in essence a macroscopic phenom-
enon, the physics of friction thus extends over several length
scales and depends crucially on the microscopic aspects of
the problem. In this paper, we study separately two aspects,
microscopic and macroscopic, of friction.
The macroscopic collective effect of all the asperities is
measured through the friction coefficient ␮, defined as the
ratio between the tangential force F needed to slide surfaces
at velocity v and the normal load Fn applied on them: ␮
= F / Fn. From Amonton’s law, the friction coefficient is inde-
pendent of the apparent contact area of the sliding solids.
This holds for nominally rough surfaces 关3兴 but not neces-
sarily for self-affine surfaces, where the roughness depends
show self-affine behavior over some length scales 关23,24兴, it
is important to know if Amonton’s law is modified in this
case. Several very detailed studies on this topic have shown
that the fractal character of the surface has very little effect,
or none at all, on the real area of contact. We show that the
earlier treatment of Greenwood and Williamson 关3兴 can be
used even in the case of self-affine surfaces and calculate the
contact area between rough self-affine surfaces. We show
how both the area of real contact, and the friction coefficient
between them, depend on the roughness exponent ␹ of the
surfaces. We find that the real area of contact Ar between a
plane and a self-affine fractal surface scales as Fn ⬃ Ar␤ where
␤ = 共4 + 3␹兲 / 共4 + 2␹兲. For a self-affine surface, 0 ⬍ ␹ ⬍ 1,
leading to 1 ⬍ ␤ ⬍ 7 / 6, in good agreement with recent nu-
merical results of Batrouni et al. 关17兴, and in line with other
recent work 关16,18–21兴 predicting also an almost linear re-
lationship between the normal force and the real contact area
of the surface. We also find that the friction coefficient has a
weak anomalous dependence on the apparent contact area A0
through the roughening exponent, namely, ␮ ⬀ A−0 ␹/4, which
in principle is measurable experimentally.
The study of friction at the microscopic level of the as-
perity has gained importance over the years, particularly with
respect to the field of nanotechnology. An understanding of
friction at this level is also necessary to make the link to the
global behavior of the asperities. Friction at the single-
asperity level is ideally explored by the friction force micro-
scope 共FFM兲, whose tip consists of only a few atoms
关7–9,27,28兴. At low scanning velocities, FFM experiments
found that the tip can exhibit stick-slip behavior commensu-
rate with the periodicity of the substrate’s underlying lattice.
As the FFM measures friction force F under a constant load,
this gives the effective microscopic friction coefficient ␮micro

1539-3755/2008/77共3兲/036123共11兲 036123-1 ©2008 The American Physical Society

SANG, DUBÉ, AND GRANT
as ␮micro ⬀ F under those conditions. That is, the FFM essen-
tially provides experimental access to the friction of a single
asperity, under weak loading, linear response conditions.
The situation can be studied using a simplified Tomlinson
model 关29–32兴, and a detailed analysis of thermal effects on
the stick-slip behavior, treated as a dynamic critical phenom-
enon 关33–38兴, shows that the friction force F decreases from
its optimal value of Fc at T = 0, with a correction of the form
关6,39,40兴
冉 冊
F − Fc ⬀ − T2/3 ln2/3 B
T
v
, 共1兲
where F is the friction force, T is temperature, and v is the
scanning velocity. As noted above, the microscopic friction
coefficient ␮micro ⬀ F. The nonuniversal constant B depends
on the exact details of the setup 关39兴. We have shown before
that this result compares well to numerical simulations when
the FFM tip can be described by a simple Langevin equation
with a static periodic potential coming from the substrate.
We show in the present paper that this result also holds when
elasticity of the substrate, modeled as a ball-spring system, is
included, provided the substrate is sufficiently rigid.
The rest of the paper is organized as follows. Section II
shows the friction coefficient’s dependence on surface rough-
ness. In Sec. III theoretical and numerical calculations of the
friction force between a FFM and a substrate are presented.
Conclusions are summarized in Sec. IV.
II. CONTACT AND FRICTION BETWEEN ROUGH
SURFACES
Several experimental studies 共see, e.g., Chap. 4 of Ref.
关5兴. or Refs. 关23,24兴兲 have shown that surfaces often present
a self-affine character over several length scales. Hence, the
contact mechanics between rough surfaces has been the sub-
ject of several recent studies 关15–22兴. Knowledge of the con-
tact area between rough surfaces is the first step in the study
of friction on rough surfaces. However, it is often argued that
a fractal surface possesses asperities on all length scales so
that the usual treatments, based on the contact mechanics of
individual asperities with a surface 关3兴, are not valid. In con-
trast, here we consider the realistic case of a standard rough
surface. Such a surface is assuredly self-affine on large
length scales, but on small length scales such behavior is
modified by the presence of a small-length-scale ultraviolet
cutoff.
Consider a rough surface in contact with a flat plane. The
rough surface, of height h共x兲, is assumed to be a self-affine
fractal in the range a ⬍ x ⬍ L. The length a is a physical
microscopic cutoff length, typically of the order of a few
nanometers, at which the fractal behavior breaks down. The
length scale L ⬃ A1/2
0 is the lateral size of observation, with
A0 is the apparent area of the surface. It is also possible that
the self-affine character is present only up to an upper length
scale ␰, of the order of a few micrometers. The surface is
characterized by a power spectrum in d dimensions
036123-2
PHYSICAL REVIEW E 77, 036123 共2008兲
TABLE I. Numerical value of the length l␦ for various values of
the total width w for an apparent area Aa = 10⫻ 10 ␮m2, a rough-
ness exponent ␹ = 1 / 2, and molecular cutoff length a = 0.5 nm
w 共nm兲 l␦ 共nm兲 ␬ 共109 m−1兲 A
5 0.0025 0.14 400 ␮m2
10 0.01 0.28 1.56 ␮m2
50 0.25 1.41 4 nm2
具兩hk兩2典 = ⌬2k−d−2␹ ,
where hk is the Fourier transform of the surface, ␹ is the
共2兲
roughness exponent, and 具¯典 represents an average over the
statistical distribution of the interface 关41兴. In real space, the
self-affine structure is determined by the correlation function
具关h共r + l兲 − h共r兲兴典1/2 ⬃ l␹ . 共3兲
For ␹ ⬍ 1, the amplitude of the power spectrum can be ex-
pressed in terms of a length ⌬ = 共l␦兲1−␹ 关42兴. The total width
of the interface,
w = ⌬L␹ ⬃ l␦1−␹A0␹/2 , 共4兲
depends on the apparent area of observation. The roughness
exponent must satisfy ␹ ⬍ 1 for the surface to be well de-
fined. Its value depends on the universality class of the phe-
nomenon 关41兴. For example, in the case of simple linear
roughening, ␹ = 共3 − d兲 / 2 in d 共1 ⬍ d ⬍ 3兲 dimensions and
logarithmic roughness in three dimensions. For several other
common surfaces, the roughness exponent may be quite
large, 0.5⬍ ␹ ⬍ 0.9, and w typically is of the order of a few
tens of nanometers for a lateral length scale of the order of 1
␮m 关5兴. In most cases, however, the length l␦
a, as shown
in Table I.
A possible starting point to obtain the contact area be-
tween the rough surface and the plane is the general formula
关43兴
u共r兲 = 冕 dr⬘ Gik共r − r⬘兲␴k共r⬘兲, 共5兲
where u are the displacements induced by the normal stress
␴k over the surface r = 共x , y兲 and the Green’s function G共r兲
⬃ 1 / r asymptotically. An early scaling analysis of Eq. 共5兲,
using a self-affine fractal surface on all length scales, pre-
dicted that the normal force Fn and the area of real contact
would be related as Fn ⬃ Ar共1+␹兲/2 关14兴. However, in marked
contrast, a numerical solution of Eq. 共5兲 by Batrouni et al.
关17兴 found Fn ⬃ Ar1.1 for both ␹ = 0.6 and ␹ = 0.9. A detailed
analysis by Persson 关15兴, starting essentially from Eq. 共5兲,
but for a self-affine surface only over a limited range, shows
Fn ⬃ Ar unless the load is so large that the area of real contact
becomes similar to the nominal area, probably far above the
plastic threshold. The very weak dependence of the area of
real contact on the roughness exponent is further confirmed
by the multiscale analysis of Hyun et al. 关18,19兴 and modi-
fied Green’s function methods 关20,21兴. These results point to
the fact that the self-affine nature of the surface has only
DEPENDENCE OF FRICTION ON ROUGHNESS… PHYSICAL REVIEW E 77, 036123 共2008兲

little influence on the contact mechanics, contrary to natural Fn = N0Eⴱ共w3/␬兲1/2e−u/w = Eⴱ共w␬兲1/2Ar . 共11兲
expectations.
This small effect of the self-affine surface is easily under- Since the curvature is a well-defined quantity, the number of
stood by realizing that in contact mechanics the important asperities is related to the apparent area of contact via N0
quantity is not necessarily the height itself or the height- ⬃ A0␬2, and the relation between the real area of contact and
height correlations 关19兴, but rather the mass density, which is the normal force can be established for different values of u.
essentially In the limit u
w, the exponential term can be neglected and
N ⬇ N0. Using Eq. 共4兲 to express the width w in term of the
␳共z兲 = ␳0具␪„z − h共r兲…典, 共6兲 apparent area A0, the apparent and real areas of contact are
related through Eq. 共10兲 as
where ␳0 is the bulk density, and ␪共u ⬎ 0兲 = 1 and ␪共u ⬍ 0兲
= 0. This mass density should be used, rather than the asper-
ity height distribution usually considered in traditional treat-
A0 = A 冉冊
Ar
A
2/2+␹
, 共12兲
ments of the problem.
For a self-affine surface with roughness exponent ␹ ⬍ 1, where the area
the ratio w共L兲 / L → 0 as L → 0, meaning that the surface is
asymptotically flat and that there cannot be any long-range
correlations in the mass density 共see also Ref. 关22兴兲. For an
A = l␦2 冉冊
a
l␦
2共2/x−1兲
共13兲

interface with Gaussian fluctuations, it is straightforward to ranges between a few nm2 and several ␮m2 共see Table I兲. In
show that ␳共z兲 = erfc共z / w共A0兲兲 关44兴, where w共A0兲 ⬃ A0␹/2 is the this limit, the relation between the normal force and the real
surface standard deviation of heights. In a general case, the area of contact is

冉冊
mass density, or, equivalently, the height distribution, is well
␤
approximated asymptotically by the exponential form Ar
F n = E ⴱA , 共14兲
A
␳共z兲 ⯝ ␳0e −z/w
. 共7兲
where the exponent ␤
The important point is of course that it has a finite range in
the direction perpendicular to the surface. 4 + 3␹
␤= , 共15兲
The average curvature of the rough interface is obtained 4 + 2␹
from ␬ = 具共ⵜ2h兲2典1/2, written explicitly as 关44兴
For 0 ⬍ ␹ ⬍ 1, the exponent 1 ⬍ ␤ ⬍ 7 / 6.
␬2 = 冕
L−1
a−1
dk k3+d具兩hk兩2典 = a−2 冉冊
l␦
a
2共1−␹兲
. 共8兲
The width is itself related to the real contact area, so the
range of validity of this power-law regime is obtained self-
consistently as
Even though it depends on the short-distance cutoff, this is
also a well-defined quantity. For typical surfaces, 10−2 ⬍ ␬a
ⱕ O共1兲 and we can therefore think of the surface as a set of
u
w
= u␬
A
冉冊
Ar
x/2+␹

1, 共16兲

N0 ⬃ A0 / ␬2 asperities with radius of curvature ␬−1 ranging
from a few nanometers to a fraction of a micrometer.
It is then straightforward to use previous theories of con-
tact mechanics due to Greenwood and Williamson 关3兴 for
rough surfaces. In this theory, the rough surface is composed
of many distinct asperities. The contact of each asperity with
the flat surface is treated within Hertz theory 关45,46兴, which
states that the contact area between the surface and the as-
perity is proportional to the ratio of the normal displacement
to the curvature 共z − u兲 / ␬, where u is the separation between
the two surfaces, z is the height of the asperity, and ␬ the
curvature of the asperity. The normal force is proportional to
Eⴱ␬−1/2共z − u兲3/2, where Eⴱ is the effective elastic modulus of
the surfaces 关47兴. With an exponent distribution of density,
all the standard results of the theory follow. If the total num-
ber of asperities on the surface is N0, the number of asperities
in contact is
N = N0e−u/w ,
and the total real contact area Ar and normal load Fn of all
the asperities in contact are
Ar = N0␬−1we−u/w = 共w/␬兲N,
meaning that large apparent and real surfaces of contact are
necessary. The parametric plot of Fig. 1 shows that the
power-law relationship between the real area and the force is
actually valid over a wide range of parameters. Deviations
from Eq. 共14兲 occur only when u␬共a0 / Ar兲2/共2+␹兲  1, in
which case we obtain Fn ⬃ Ar␤/2.
The weak dependence of ␤ on ␹, and the range of values
for ␤, are in good agreement with numerical results of Ba-
trouni et al. 关17兴. Those authors found that Fn ⯝ Ar1.1—that is,
␤ ⬇ 1.1, for both ␹ = 0.6 and ␹ = 0.9. It is, however, not in
agreement with the linear behavior obtained by Persson 关16兴
as well as by Hyun et al. 关18,19兴 and Campana 关20,21兴. This
disagreement may come from the fact that the transverse
displacement of the surface must be carefully taken into ac-
count during the contact process. Large system sizes must
also be considered in order to deal with such small expo-
nents. Nevertheless, our result, which is obtained from the
finite range of the asperity height distribution, again points to
共9兲
the weak effect of a fractal surface on the real area of con-
tact.
A simple argument may then be used to estimate the static
coefficient of friction. Assuming that most of the asperities
共10兲 are in a state of incipient plastic flow, and that the previous

036123-3
SANG, DUBÉ, AND GRANT PHYSICAL REVIEW E 77, 036123 共2008兲

F
10

5
10

4
10
Fm
3
10

2
10
Fn/E*A

1
10

0
10

10
−1 uκ=15

−2
10
Rc − a Rc R
−3 uκ=0.001
10

FIG. 2. Idealized friction force measured during a stick-slip

−3 −2 −1 0 1 2 3 4 5 6
10 10 10 10 10 10 10 10 10 10 event at zero temperature. As the support moves away, the force
Ar/A
increases linearly until a position Rc 共to which corresponds a maxi-
mal force Fm兲. At this point, the energy barrier opposing the motion
FIG. 1. Relation between the normal force and the real area of
of the tip vanishes, the tip is released, and the force decreases sud-
contact for a surface characterized by ␹ = 0.5 and values of the
denly. At nonzero temperature, thermal activation can cause transi-
dimensionless parameters u␬ = 10−3 共lowest curve兲 10−2, 1, 5, 10,
tions for positions R ⱕ Rc. The position at which this transition oc-
and 15 共highest curve兲. The slope of the lines corresponds to the
curs as well as the maximal force registered between the events are
exponent ␤ 关Eq. 共15兲兴, with only very weak deviations at small
then stochastic quantities whose distribution is related to Eq. 共28兲 of
loads.
the text.

result, based on linear elasticity is not drastically modified,

the lateral force necessary to displace the asperities is F
⬃ ␴cAr where the yield stress is ␴c 关4兴. Then, using Eqs. 共9兲
and 共10兲, we find that the friction coefficient ␮ ⬅ F / Fn de-
pends on the width of the interface: ␮ ⬀ w−1/2. Immediately,
then, from the dependence of the width on the apparent area
of contact in Eq. 共3兲, we obtain the anomalous dependence of
the friction coefficient on the apparent area of contact,
␮共A0兲 ⬀ A−0 ␹/4 .
This transport-coefficient anomaly is measurable experimen-
tally, as well as numerically, provided that the load depen-
dence on the area of contact is also taken into account 关20兴.
In any case, this is a very weak dependence; as mentioned
above, on general grounds ␹ ⬍ 1.
At fairly large loads, plasticity leads to an increase in the
contact area over time. Heslot et al. 关2兴 have shown that for
large load values 共⬃1 N兲 but very small pulling velocities
共v ⬍ 1 ␮m / s兲, the friction coefficient ␮ = ␮共v兲, together with
stick-slip motion caused by the plastic relaxation of the con-
tact. On the other hand, at very small loads, the stick-slip
phenomenon, caused solely by interatomic forces between
the asperities and the surface periodic potential, occurs 关8兴.
We now concentrate on this case.
III. STICK-SLIP BEHAVIOR AND THERMAL
ACTIVATION
A friction force microscope provides direct access at the
atomic level to friction. As such, it essentially probes the
friction of a single asperity, under weak loading, linear re-
sponse conditions. In this section, we present the general
stick-slip phenomenon in atomic friction and introduce a
simple one-dimensional model to calculate atomic-scale fric-
tion. For the measurement probe, we refer explicitly to the
FFM, modeled as a tip coupled elastically to a moving can-
tilever 关7–9兴.
In a typical experimental situation, the FFM is dragged in
contact mode across a surface at constant velocity, and the
friction force F needed for this continuous displacement is
recorded as a function of R, the displacement of the cantile-
ver. The occurrence of stick-slip behavior is then seen as a
共17兲 series of triangular force jumps 共Fig. 2兲 with the periodicity
of the substrate lattice spacing. As noted above, this force is
proportional to the microscopic friction coefficient ␮micro.
This behavior can be understood from an analysis of the
energetics of the tip. If the tip of a FFM, put close to a
substrate, its global energy is the sum of two terms: the lat-
tice surface potential and the elastic coupling between the tip
and the cantilever. A typical superimposition of these two
potentials is shown in Fig. 3. Most of the time, the tip re-
mains static at potential minima 关see Fig. 3共a兲兴. As the can-
tilever is displaced, the elastic force F increases while the
potential barrier decreases 关see Fig. 3共b兲兴. At low tempera-
tures, the tip barely moves until the potential barrier van-
ishes, at which point it jumps to the next potential minimum,
thus causing a sudden decrease in the friction force 关see Fig.
3共c兲兴. At nonzero temperature, thermal fluctuations make it
possible for a jump to occur before the barrier vanishes,
when it becomes comparable to kBT.
A. Analysis
For commensurate surfaces, we treat the tip as a point
particle, with coordinate r = 共x , z兲 关48兴, subject to an effective
substrate surface potential U共r兲, which possesses the period-
icity of the substrate, U共x + a , z兲 = U共x , z兲, where a is the lat-
tice constant of the surface. Sinusoidal functions are com-

036123-4
DEPENDENCE OF FRICTION ON ROUGHNESS… PHYSICAL REVIEW E 77, 036123 共2008兲

(a)
F= 冏 ⳵ K共R,x兲
⳵x
冏 xeq
⬵ k̃R. 共20兲

R=0
v The second part of Eq. 共20兲 is a linear approximation to the
E(R, x)
friction force. It introduces the effective elastic constant k̃
which accounts for the effect of the potential U共x兲 on the
elastic force of the FFM 关39兴. The effective constant k̃ cor-
responds to the slope of the friction force in the sticking part
x of a typical sawtooth figure 共see Fig. 2兲.
(b) The critical point at which the energy barrier vanishes
corresponds to an inflection point of the global potential. For
T>0K
0 < R < Rc this position of the support Rc 共to which corresponds an equi-
v librium position of the tip xc兲, both the first and second de-
E(R, x)
rivatives of E共R , x兲 at xc are zero,

⳵ E共Rc,xc兲
=0 共21兲
⳵x
x < xc
and
(c)
⳵2E共Rc,xc兲
T=0K R = Rc = 0. 共22兲
v ⳵ x2
E(R, x)
At nonzero temperature, the tip is thermally activated over
the energy barrier to the next potential minimum before R
reaches Rc. Since we expect this actual jump to take place on
x = xc
time scales much faster than the typical time spent in a po-
tential well, we can describe this transition by the Kramers
rate 关49兴 in the potential corresponding to the instantaneous
position R,
FIG. 3. Schematic plots of the potential function E共R , x兲 vs x,
the position of the tip, for different positions R of the support. ⍀2 −⌬E/k T
␶−1共R兲 = e B , 共23兲
2␲␥
monly used to model this part of the surface potential
关11,30,31,39兴. In the special case of a FFM operating under where ⍀ and ⌬E correspond, respectively, to the effective
constant load, the normal force is fixed and the z coordinate oscillation frequency and barrier height, and ␥ is the dissipa-
drops out of the problem. The second potential is the elastic tion coefficient. Due to this exponential behavior, the great-
coupling between the tip and the cantilever, K共R , x兲, where R est probability for a transition to occur is in the region
is the coordinate of the cantilever support 关30,31,39兴. ⌬E
kBT. In this case, Eq. 共A6兲 shows that, to lowest order
The global energy of the tip is thus 关see Fig. 3共a兲兴 in the bias f ⬅ 1 − 共R / Rc兲
1, the energy barrier and oscilla-
tion frequency of a general potential of the form E共R , x兲 are
E共R,x兲 = U共x兲 + K共R,x兲. 共18兲
⌬E = Eeff f 3/2 共24兲

In this section, we keep U共r兲 and K共R , x兲 as general func- and

tions. Specific forms appropriate to the case of a FFM on a
surface will be given in the next section. ⍀ = ⍀eff f 1/4 , 共25兲
For a given support position R, the equilibrium position of
the tip, denoted xeq共R兲, is obtained by where the constants Eeff and ⍀eff, given in Eqs. 共A7兲 and
共A11兲, are expressed in terms of derivatives of E共R , x兲 at the
critical position. The effective energy barrier is in general a
⳵ E关R,xeq共R兲兴 complex interplay between the substrate elastic energy, the
= 0, 共19兲 structure of the tip, and the normal load applied to the tip and
⳵x
can be measured experimentally 关9–11兴.
For a support moving at constant velocity, R共t兲 = vt. The
and the friction force measured experimentally corresponds probability that a transition has not taken place at time t is
to 关35,37,39兴

036123-5
SANG, DUBÉ, AND GRANT PHYSICAL REVIEW E 77, 036123 共2008兲

W„R共t兲… = exp − 冕 t0
t
␶−1关R共t⬘兲兴dt⬘ . 共26兲

Changing the variable from t⬘ to R, the distribution of the

support’s position at which a transition occurs, i.e., Rm, is
then obtained as
dW共R兲
P共Rm兲 = − 共27兲 FIG. 4. Schematic diagram of the numerical model. FFM tip is
dR
modeled with a tip coupled elastically to a support and the substrate
is modeled with a ball-spring model.
3 ⴱ3/2
= Xf ⴱ1/2 exp共− f ⴱ3/2 − Xe−f 兲, 共28兲
2
The maximal force Fm between tip and cantilever before a
where f ⴱ = 共Eeff / kBT兲2/3 f and transition occurs decreases from its optimal value at T = 0 and
becomes a stochastic quantity 关39兴. Assuming Fm ⬃ k̃Rm, its
2 ⍀eff
2
R c k BT distribution is obtained directly from Eq. 共28兲. We show in
X= . 共29兲
3 2␲␥ v Eeff Sec. III B that there is a good agreement between Eqs. 共31兲
and 共28兲 and the numerical simulations. The exponent 2/3 in
From the distribution of the transition positions P共Rm兲, we Eq. 共31兲 comes from the nonlinear characteristic of Eqs. 共24兲
calculate the average transition position 关37兴 and 共25兲. As we show in the Appendix, those nonlinear char-

具Rm典 = Rc − Rc 冉 冊
k BT
Eeff
2/3
g共X兲, 共30兲
acteristics are universal for problems with transition close to
the critical position, and do not depend on any particular
form of the global energy.
where the function g共X兲 ⬃ ln2/3 X + O共1 / ln X兲 in the limit
X  1 关35,37兴 and g共X兲 ⬃ X + O共X2兲 in the limit X
1. B. Numerical model
With the linear approximation Eq. 共20兲 between F and R, To test the method developed in the last section, we set up
it is straightforward to calculate the average lateral force as an atomistic model to simulate the operation of a FFM. A
the integral of the instantaneous force over a cycle of the schematic graph of the model is shown in Fig. 4. The tip,
stick-slip motion, which yields 关51兴 with coordinate r = 共x , z兲, and its coupling to the cantilever
F = Fc − ⌬F共T兲关ln X共v,T兲兴2/3 , 共31兲 are modeled by a single atom attached to a moving support.
The support of the cantilever has coordinates R = 共R , Z兲. For
where the constants operation under constant load, the separation between the

冉 冊
Fc = k̃ Rc −
a
2
, 共32兲
support and the tip in the normal direction, Z − z, remains
constant at all times. The support moves horizontally at con-
stant velocity v, R = vt. The support and the tip are connected

冉 冊
2/3
k BT
⌬F = k̃Rc , 共33兲
Eeff
and X共v , T兲 is presented in Eq. 共29兲. As noted above, this is
proportional to ␮micro, and is, in essence, the friction of a
single asperity—under weak loading and linear response
conditions. The incoherent average of many such asperities
gives the friction coefficient, albeit under those quite re-
stricted conditions.
As the velocity of the support is increased, the tip spends
less and less time in the minima of the potential wells. When
a velocity vⴱ, defined by X共vⴱ , T兲, is reached, thermal fluc-
tuations do not act on a time scale sufficiently fast to activate
the tip and the friction force reaches the static zero-
temperature limit as Fc − F ⬃ T5/3 / v. Once F = Fc, the friction
force remains at this plateau, as already seen experimentally
关9兴. The behavior continues until viscous friction Fv = M ␥v
dominates. Comparing the viscous and potential forces
yields the velocity vc at which this occurs:
by a spring with spring constant k, and the elastic energy
between the tip and the support is
1
K共R,x兲 = k共R − x兲2 共35兲
2
We set the numerical value k = 0.93 N / m, typical of experi-
mental setups 关8,9,39兴.
The substrate is modeled using a ball-spring model 关50兴.
Twenty atoms with coordinates ri = 共xi , zi兲 and equilibrium
position ri,0 = 共xi,0 , z0兲 are connected with springs and satisfy
periodic boundary conditions. The original positions of these
atoms are xi,0 and zi,0 and they are connected to a rigid sec-
ond layer. The atoms are held in their original positions with
harmonic potentials described by spring constants kxz and kzz,
while the spring constant of the lateral springs between the
atoms is kxx. The elastic energy of the substrate atoms is
1
Vsub共兵ri其兲 = 兺 kxz共xi − xi,0兲2
i 2

⳵ U共xc兲/⳵ x 1 1
+ 兺 kzz共zi − zi,0兲2 + 兺 kxx共xi+1 − xi − a兲2 ,
vc = . 共34兲 共36兲
M␥ i 2 i 2

Note that the ratio vⴱ / vc is independent of ␥. where the lattice constant a = 0.4 nm. Different values of kxx,

036123-6
DEPENDENCE OF FRICTION ON ROUGHNESS… PHYSICAL REVIEW E 77, 036123 共2008兲

kxz, and kzz correspond to substrates with different elastic TABLE II. Values of load and elastic constants chosen for the
moduli. In the x direction 共the scan direction兲 the shear simulations. The basic unit of the elastic constant is k = 0.93 N / m,
modulus G = kxz / a, while the compressibility is described by related to the elasticity of the FFM 关Eq. 共35兲兴. The shear strain ⑀
the modulus Bx = 共2kxx + kxz兲. The bulk compressibility in the corresponds to the maximal displacement of the substrate atoms
z direction Bz = kzz / a. It is not our intention to describe a under the motion of the tip and characterizes the softness of the
surface in all its details, but rather to consider the effect of substrate.
the elasticity of a substrate on the theoretical prediction of
Eq. 共1兲. Equation 共36兲 then defines the simplest model that ⑀ FN 共nN兲 kxx共kxx / k兲 kxz 共kxz / k兲 kzz 共kzz / k兲
allows us to interpolate between a completely rigid and a 0.01 2.2 120 160 400
very soft substrate.
0.05 1.6 12 16 40
The interaction potential between the tip and substrate
0.1 1.6 4 4 12
atoms is modeled using an exponential repulsive potential,

Vint共r,兵ri其兲 = 兺 V0共er0/兩r−ri兩 − 1兲, 共37兲 d 2x dx

i M 2 + M␥ − k共R − x兲 − f x = ␰共t兲, 共40兲
dt dt
where 兩r − ri兩 = 冑共x − xi兲2 + 共z − zi兲2 is the displacement between where M is the effective mass of the tip 共chosen to be M
the tip and the substrate atom i. We set V0 = 0.049 eV and = 8.7⫻ 10−12 kg兲. The fast degrees of freedom of the sub-
r0 = 2a. strate are present through the dissipation coefficient ␥ 共␥
A common alternative for the interaction potential is the = 8.9⫻ 105 sec−1兲 and the random noise ␰, satisfying the
Lennard-Jones potential 关50兴. Since FFMs are typically op- fluctuation-dissipation relation 具␰共t兲␰共t⬘兲典 = 2M ␥kBT␦共t − t⬘兲,
erated in contact mode, the tip and substrate separation is where the angular brackets denote an average, and kB is Bolt-
small and the force between them is repulsive. Long-range zmann’s constant. The total lateral force applied on the tip by
van der Waals attractive forces are trivial in this situation. the substrate is
The choice between the exponential repulsive potential and
⳵ Vint r0 x − xi
Lennard-Jones potential will not alter the final results. An- fx = − = 兺 V0er0/ri 2 . 共41兲
other potential in the system is the potential due to the load, ⳵x i ri ri
which is Fn共z − z0兲 关50兴. The total potential energy of the
The slow motion of the substrate is obtained by letting the
whole system is then
substrate totally relax to equilibrium at each time step, i.e.,
for a given position r of the tip, the position of each atom of
E = K共R,x兲 + V共r,兵ri其兲 + Fn共z − z0兲 + Vsub共r,兵ri其兲. 共38兲 the substrate ri is set by

The operation of the FFM under constant load uses a feed-

⳵ Vint ⳵ Ksub
+ = 0,
back loop to keep the separation between the tip and canti- ⳵ xi ⳵ xi
lever constant. We thus allow the tip to follow the surface
smoothly in the vertical direction by adjusting the coordinate ⳵ Vint ⳵ Ksub
z such that the normal force + = 0. 共42兲
⳵ zi ⳵ zi
The effect of the thermal fluctuations of substrate atoms
⳵ Vsub
Fn = − 共39兲 around their equilibrium positions was tested and we found
⳵z that its impact is negligible 关54兴.

remains fixed at all times.

The local deformations of the substrate spread at the
speed of sound 共⬃104 m / sec兲, and so the typical time scales
associated with the substrate ⬃10−13 sec. In contrast, the tip
of the mass is several orders of magnitude greater that the
mass of the atoms in the substrate, with much larger time
scales 关⬃⍀−1; see Eq. 共25兲兴. Typical scans are over several
lattice spacings, at velocities ranging from a few nm/sec to
the range of ␮m / sec, and a full molecular dynamics simula-
tion, extending over a time ⬃10−3 sec that would encompass
both the time scales of the substrate and tip is thus unrealis-
tic. However, on the time scales relevant to the tip motion,
the energy transferred from the tip to the substrate is dissi-
pated almost instantly, and the fast motion of the substrate
can be incorporated phenomenologically into a Langevin
equation for the lateral motion of the tip alone 关52,53兴:
C. Comparison of analysis and numerical results
We study the influence of the elasticity of the substrate,
with comparison to the theoretical model of Sec. III A, by
adjusting the elastic constants of the substrate kxx, kxz, and kzz
to produce different values of shear strain ⑀ = 0.01, 0.05, and
0.1. This last value is just below typical criteria for surface
melting 关55兴.
Equation 共40兲 is simulated using Ermak’s algorithm 关52兴.
The parameters used in the calculations are shown in Table
II. For each temperature and velocity, the scanning distance
is 50 lattice spacings and time steps are between 0.01/ ␥ and
0.07/ ␥. The simulations are done at several temperatures and
velocities for each substrate. For the hard substrate 关⑀ = 0.01;
see Fig. 5共b兲兴, we use four different temperatures, 133, 213,
293, and 373 K, with scanning velocities 5 nm/secⱕ v

036123-7
SANG, DUBÉ, AND GRANT PHYSICAL REVIEW E 77, 036123 共2008兲

(a) (b) 1
0.6 1.2

0.5 1

0.4 0.8 0.5

F(R) (nN)

F (nN)
0.3 0.6

fx(x) (nN)
0.2 0.4
0
0.1 0.2

0 0
5 6 7 8 9 0 2 4 6 8 10 12 14
R (nm) ln(v)
−0.5
(c) (d)
0.6 0.08
−1
0.5 4 4.4 4.8
0.06
0.4 x (nm)
F (nN)
F (nN)

0.3 0.04

0.2 FIG. 6. Resistance force applied on the tip by the substrate

0.02
0.1 atoms for three different substrates. Solid, dashed, and dot-dashed
0 0 lines correspond, respectively, to hard 共⑀ = 0.01兲, mid-soft
0 2 4 6 8 10 12 0 2 4 6 8
ln(v) ln(v) 共⑀ = 0.05兲, and soft 共⑀ = 0.1兲 substrates.

FIG. 5. 共a兲 Typical stick-slip behavior of the instantaneous fric- Eq. 共40兲 at T = 0 共see Fig. 6兲. Once f x is known, higher-order
tion force as the support is moved at velocity v = 25 nm/ sec for a derivatives of U共x兲 are obtained by numerical differentiation.
temperature T = 293 K on hard substrate 共⑀ = 0.01兲. 共b兲 shows aver-
The critical position xc is found from Eq. 共21兲,
age friction force for different velocities at different temperatures
for the same substrate. Squares, diamonds, up triangles, and down ⳵2E共Rc,xc兲 d2U共xc兲 df x
triangles correspond, respectively, to temperatures T = 133, 213, = = = − k, 共44兲
⳵ x2 dx2 dx
293, and 373 K. 共c兲 shows average friction force for different ve-
locities at different temperatures for mid-soft substrate 共⑀ = 0.05兲. while the critical position of the support is obtained from Eq.
Circles, squares, diamonds, and up triangles correspond, respec- 共22兲,
tively, to temperatures T = 53, 133, 213, and 293 K. 共d兲 shows av-
erage friction force for different velocities at different temperatures ⳵ E共Rc,xc兲 dU共xc兲
= − k共Rc − xc兲 = 0. 共45兲
for soft substrate 共⑀ = 0.1兲. Circles, pluses, stars, and squares corre- ⳵x dx
spond, respectively, to temperatures T = 53, 73, 93, and 133 K. The
units of velocity v are nm/sec. The effective elastic constant between the tip and the canti-
lever, k̃ is extracted from the sawtooth graphs 关for instance,
Fig. 5共a兲兴, and Fc can be then be obtained from Eq. 共32兲
ⱕ 256 ␮m / sec. For the softer substrates, the simulations
Finally, the numerical evaluation of the third derivative of
must be performed at lower temperatures. For the ⑀ = 0.05
U共x兲 yields Eeff and ⍀eff. Then the coefficients in X and ⌬F
substrate 关Fig. 5共c兲兴, we use 53, 133, 213, and 293 K with
can be calculated.
scanning velocities 5 nm/ secⱕ v ⱕ 64 ␮m / sec. Finally, for
We define
the softest substrate 关⑀ = 0.1; Fig. 5共d兲兴, we use 53, 73, 93,
and 133 K, with 5 nm/ secⱕ v ⱕ ␮m / sec. Smaller loads are
used for the softer substrate, so as to obtain equivalent in-
dentation of the surface.
A ⬅ k̃Rc 冉 冊
kB
Eeff
2/3
共46兲

The general results show an increase of the friction force and

with velocity and a decrease with temperature. At high ve-
locities, stick-slip behavior disappears and linear friction F 2 ⍀eff
2
kB
B⬅ Rc , 共47兲
⬃ M ␥v starts to dominate. This corresponds to the sharp in- 3 2␲␥ Eeff
crease in Figs. 5共b兲–5共d兲.
and rewrite Eq. 共31兲 as
In order to apply the method developed in Sec. III A to
the numerical model of Sec. III B, we need to know the
effective surface potential U共x兲, introduced in Eq. 共18兲. This
is not trivial since the interaction potential Vint is a function
F = Fc − AT2/3 ln2/3 B 冉 冊 T
v
, 共48兲

of both x and z, with z constantly changing to maintain a which is proportional to the microscopic friction coefficient.
constant load. However, the collective effect of the substrate The numerical results obtained in this way for the various
and load is included in Eq. 共40兲, and we can associate substrates are shown in Table III. This allows us to scale the
data according to Eq. 共31兲, with the result shown in Fig. 7.
⳵ U共x兲 For these figures, we keep only temperatures and velocities
⬅ − f x„x,z共x兲,兵ri共x兲其…, 共43兲
⳵x that are clearly within the stick-slip regime. The collapse of
all data shows clearly that Eq. 共31兲 gives an adequate de-
where the transverse position of the tip and the position of scription of the stick-slip process even when local elasticity
the substrate are obtained from Eqs. 共39兲 and 共42兲. This of the substrate is present. Note that the data points for the
quantity can be obtained numerically from a simulation of soft substrate 共⑀ = 0.1兲 at 133 K and 5 and 10 nm/sec do not

036123-8
DEPENDENCE OF FRICTION ON ROUGHNESS…
TABLE III. Numerical values of the parameters obtained from
the scaling analysis of Fig. 7.
Fc 共nN兲 A 共10−3 nN/ K2/3兲
0.01 0.57 2.0
0.05 0.26 1.5
0.1 0.045 0.79
scale well with the rest of the data. This is because soft
substrates require a considerably smaller normal load than
harder substrates. The small normal load leads to a small
effective potential barrier. Starting from high temperatures
where kBT is comparable to the effective potential barrier, the
tip is no longer confined to the potential minimum. So for
temperatures higher than 93 K the transition is not confined
to the small regime around the critical position, which means
Eq. 共31兲 does not apply.
We also studied the distribution of force maxima from Eq.
共28兲, under the assumption Fm = k̃Rm. For hard 共⑀ = 0.1 strain兲
substrate at 293 K, we calculated the force maxima distribu-
tion for velocities v1 = 25 nm/ sec and v2 = 1 ␮m / sec. For
these choices, X共v1兲 = 3.6⫻ 102 and X共v2兲 = 8.9. Other param-
eters obtained from the numerical simulations and used in
the theoretical calculations are Eeff = 0.6 eV and Rc
= 0.86 nm. The comparison of theoretical and numerical re-
sults is shown in Fig. 8. The theoretical and numerical results
agree very well at low velocities although deviations are seen
at larger velocities 共1 ␮m / sec兲. These deviations are be-
cause linear friction 共F = M ␥v兲 becomes comparable to the
stick-slip friction 关F given by Eq. 共31兲兴 at large velocity.
IV. CONCLUSIONS
In this paper, we studied the dependence of the friction
coefficient ␮ on the surface roughness. We pointed out the
importance of using the mass density and explained previous
numerical results concerning the dependence of forces on the
real area of contact for rough surfaces. As well, we have
8 the relationship between the friction and normal load 关9兴,
7
6 would be particularly valuable to consider an extended range
[(Fc−F)/A] /T
5
3/2
4
3 tracted from such comparisons provide direct, feasible, and
2 simple access to the fundamental description of friction at
1
0
−1 0 1 2 3 4 5 6 7
ln (B T/v)
FIG. 7. 共Color online兲 Scaling for the data of three different
substrates. Black, green, and red symbols correspond to hard
共⑀ = 0.01兲, mid-soft 共⑀ = 0.05兲, and soft 共⑀ = 0.1兲 substrates
respectively.
PHYSICAL REVIEW E 77, 036123 共2008兲
0.12
0.1
B 共m / sec K兲
30.4 0.08
14.9
4.3
P(Fm)
0.06
0.04
0.02
0
0.25 0.35 0.45 0.55 0.65 0.75
Fm (nN)
FIG. 8. Normalized distribution, with statistical error bar, of the
maxima of the friction force Fm for hard 共⑀ = 0.01兲 substrate at 293
K with v = 2 共diamonds兲 and 1 ␮m / sec 共circles兲. Solid lines are
theoretical results.
made a prediction of a weak anomaly in the friction coeffi-
cient, which is dependent on the apparent area of the rough
surface. We also developed a method to calculate the atomic
friction based on thermally activated stochastic processes. A
numerical model was set up to test our analytical results. The
results in this paper and our previous work 关39兴 show that
this nonlinearly logarithmic dependence on scanning veloc-
ity and temperature is universal. It does not depend on the
particular form of surface potential in the model nor on the
substrate’s elastic properties, and should be experimentally
observable on different substrates. In order to study thermal
effects, we used a Langevin equation to describe the ther-
mally activated motion of the FFM tip. This method allowed
us to simulate time scales ranging from a few seconds to a
few minutes, which match the experiment time scales. Fi-
nally, the simplified Tomlinson model we used in this paper
is very successful in probing problems related to stick-slip
motion in the lateral direction and can also be used to study
allowing one to calculate the effective microscopic friction
coefficient ␮micro ⬀ F. To test those results experimentally, it
of temperatures and velocities, obtaining not only the depen-
dence of the friction force on those quantities, but the distri-
bution function of such forces as well. The parameters ex-
the atomic scale.
8
ACKNOWLEDGMENTS
We thank Peter Grütter, Roland Bennewitz, François Drolet,
and Lyne Cormier for interesting discussions. This work was
supported by the Natural Sciences and Engineering Research
036123-9
SANG, DUBÉ, AND GRANT PHYSICAL REVIEW E 77, 036123 共2008兲

Council of Canada, the Canada Research Foundation on where ␦x = x − xc. The first term of the expansion as well as
Value-Added Paper, and le Fonds Québécois de la Recherche the term of order O共␦x兲 vanish from the definition of the
sur la Nature et les Technologies. spinodal point and terms of higher order than O共Rc f兲 can be
omitted from the calculation. The displacement ␦x ⬃ O共f 1/2兲,
APPENDIX or more precisely
In this appendix, we show how Eqs. 共24兲 and 共25兲, lead-
ing to Eq. 共1兲, can be understood in terms of a generic po- ␦x2 = 2fRc 冉 ⳵2E共Rc,xc兲
⳵R ⳵ x
冊冉 ⳵3E共Rc,xc兲
⳵ x3
冊 −1
, 共A4兲
tential E共R , x兲. We only assume that the potential consists of
a metastable minimum and that R is a parameter that controls or, in the specific case of a potential of the form Eq. 共18兲,
the height of the potential barrier. At R = 0, the local mini-
mum and maximum of the potential are denoted by x⫾
tained from the solution of
0
ob- ␦x2 = 2fRc 冉 ⳵2K共Rc,xc兲
⳵R ⳵ x
冊冉 ⳵3U共xc兲
⳵ x3
冊 −1
. 共A5兲

冏 ⳵ E共0,x兲
⳵x
冏 0
x⫾
= 0. 共A1兲
The barrier height is thus
⌬E共f兲 = E„Rc共1 − f兲,xc + ␦x… − E„Rc共1 − f兲,xc − ␦x…

The barrier height for thermal activation is then ⌬E0 ⳵2E共Rc,xc兲 1 3 ⳵3E共Rc,xc兲
⯝ − 2fRc␦x + ␦x ⬅ Eeff f 3/2 .
= E共0 , x+兲 − E共0 , x−兲. For nonzero but small values of R, the ⳵R ⳵ x 3 ⳵ x3
barrier height can be found from a simple Taylor expansion 共A6兲
0
around R = 0 and x⫾ .
Again specializing to the case of Eqs. 共18兲 and 共35兲
⌬E共R兲 = ⌬E共0兲 + R
⳵R
冉
⳵ E共0,x+0兲 ⳵ E共0,x−0兲
−
⳵R
冊 共A2兲
Eeff =
4冑2 3/2 3/2 ⳵3U共xc兲
k Rc 冉 冊 −1/2
. 共A7兲
3 ⳵ x3
is linear in the control parameter R. However, this expansion
is valid only if the second term in Eq. 共A2兲 is much smaller The oscillation frequency in the local minimum and in the
than the unperturbed barrier height ⌬E0. As a renormaliza- “inverted” potential are also needed for a complete determi-
tion group analysis clearly shows 关33兴, this “linear response” nation of the Kramers rate 关49兴. Up to O共f 1/4兲, the oscillation
is not valid in the cases where the thermally activated behav- frequency of the potential, ⍀共f , xc + ␦x兲 = ⍀共f , xc − ␦ f兲 = ⍀共f兲,
ior takes place when the barrier is extremely small. with
Instead, the expansion of the potential must be made with
1 ⳵2E„Rc共1 − f兲,xc − ␦x…
respect to the point where ⌬E = 0. A vanishing energy barrier ⍀2 = 共A8兲
corresponds to an inflection point of the potential E共R , x兲; M ⳵ x2
this point is located at the position xc, given by Eqs. 共21兲 and
共22兲. The spinodal point of the potential, Rc, corresponds to
the of the control parameter for which ⌬E共Rc兲 = 0. Denoting =
M
冉
1 ⳵2E共Rc,xc兲
⳵ x2
− fR c
⳵3E共Rc,xc兲
⳵ R ⳵ x2
− ␦ x
⳵3E共Rc,xc兲
⳵ x3
+¯ . 冊
f = 1 − R / Rc, we look for the new minimum and maximum by
Taylor expansion in the proximity of Rc and xc: 共A9兲
However, since ␦x ⬃ O共f 1/2
兲, to lowest order,
⳵
0 = E„Rc共1 − f兲,xc ⫾ ␦x…
⳵x ␦x ⳵3E共Rc,xc兲
⍀2 ⯝ − 共A10兲
⳵ E共Rc,xc兲 ⳵2E共Rc,xc兲 M ⳵ x3
= − fRc
⳵x ⳵R ⳵ x and ⍀ ⬅ ⍀eff f 1/4, where

⫾ ␦x
⳵2E共Rc,xc兲 1 2 ⳵3E共Rc,xc兲
⳵ x2
+ ␦x
2 ⳵ x3
+¯ 共A3兲 ⍀eff
2
=−
1
M
共2k兲1/2R1/2
c
⳵3U共xc兲
⳵ x3
冉 冊 1/2
. 共A11兲

关1兴 E. Meyer, R. M. Overney, K. Dransfeld, and T. Gyalog, Nano-
science: Friction and Rheology on the Nanometer Scale
共World Scientific, Singapore, 1998兲.
关2兴 F. Heslot, T. Baumberger, B. Perrin, B. Caroli, and C. Caroli,
Phys. Rev. E 49, 4973 共1994兲; C. Caroli and B. Velicki, J.
Phys. I 7, 1391 共1997兲.
关3兴 J. A. Greenwood and J. B. P. Williamson, Proc. R. Soc. Lon-
don, Ser. A 295, 300 共1966兲. They find that the behavior of the
normal load as a function of the contact area is almost con-
stant. They nevertheless do find some power-law-like behavior
共with effective exponent of 0.9兲 in the electrical resistivity. We
expect the latter is not an asymptotic power law, but rather a
slow crossover to linear behavior.
关4兴 B. J. Persson, Sliding Friction 共Springer, Berlin, 1998兲.
关5兴 Handbook of Micro/Nanotribology, edited by B. Bhushan
共CRC Press, Boca Raton, FL, 1999兲.

036123-10
DEPENDENCE OF FRICTION ON ROUGHNESS…
关6兴 B. N. J. Persson, O. Albohr, F. Mancosu, V. Peveri, V. N.
Samoilov, and I. M. Sivebaek, Wear 254, 835 共2003兲.
关7兴 C. M. Mate, G. M. McClelland, R. Erlandsson, and S. Chiang,
Phys. Rev. Lett. 59, 1942 共1987兲.
关8兴 E. Gnecco, R. Bennewitz, T. Gyalog, Ch. Loppacher, M. Bam-
merlin, E. Meyer, and H.-J. Guntherodt, Phys. Rev. Lett. 84,
1172 共2000兲.
关9兴 E. Riedo, E. Gnecco, R. Bennewitz, E. Meyer, and H. Brune,
Phys. Rev. Lett. 91, 084502 共2003兲.
关10兴 S. Maier, Y. Sang, T. Filleter, M. Grant, R. Bennewitz, E.
Gnecco, and E. Meyer, Phys. Rev. B 72, 245418 共2005兲.
关11兴 A. Schirmeisen, D. Weiner, and H. Fuchs, Phys. Rev. Lett. 97,
136101 共2006兲.
关12兴 M. Evstigneev, A. Schirmeisen, L. Jansen, H. Fuchs, and P.
Reimann, Phys. Rev. Lett. 97, 240601 共2006兲.
关13兴 B. Luan and M. O. Robbins, Phys. Rev. E 74, 026111 共2006兲.
关14兴 S. Roux, J. Schmittbuhl, J. P. Vilotte, and A. Hansen, Euro-
phys. Lett. 23, 277 共1993兲.
关15兴 B. N. J. Persson, J. Chem. Phys. 115, 3840 共2001兲.
关16兴 B. N. J. Persson, Phys. Rev. Lett. 87, 116101 共2001兲.
关17兴 G. G. Batrouni, A. Hansen, and J. Schmittbuhl, Europhys. Lett.
60, 724 共2002兲.
关18兴 S. Hyun, L. Pei, J.-F. Molinari, and M. O. Robbins, Phys. Rev.
E 70, 026117 共2004兲.
关19兴 S. Hyun and M. O. Robbins, Tribol. Int. 40, 1413 共2007兲.
关20兴 C. Campana, Phys. Rev. B 75, 155419 共2007兲.
关21兴 C. Campana and M. H. Muser, Europhys. Lett. 77, 38005
共2007兲.
关22兴 J. A. Greenwood and J. J. Wu, Meccanica 36, 617 共2001兲.
关23兴 S. Ganti and B. Bhushan, Wear 180, 17 共1995兲.
关24兴 A. Majumdar and B. Bhushan, J. Tribol.. 113, 1 共1991兲.
关25兴 P. J. Mangin, M. C. Béland, and L. M. Cormier, in Products of
Papermaking, edited by C. F. Barker, Transactions of the 10th
Fundamental Research Symposium, Oxford, England 共Pira In-
ternational, Leatherhead, 1993兲, p. 1397.
关26兴 F. Drolet and T. Uesaka, in Advances in Paper Science and
Technology, edited by J. L’Anson, Transactions of the 13th
Fundamental Research Symposium 共Bury, United Kingdom,
2005兲, Vol. 2, p. 1139–1154.
关27兴 S. Morita, S. Fujisawa, and Y. Sugawara, Surf. Sci. Rep. 23, 1
共1996兲.
关28兴 T. Bouhacina, J. P. Aime, S. Gauthier, D. Michel, and V.
Heroguez, Phys. Rev. B 56, 7694 共1997兲.
关29兴 G. A. Tomlinson, Philos. Mag. 7, 905 共1929兲.
关30兴 D. Tomanek et al., Europhys. Lett. 15, 887 共1991兲; T. Gyalog
et al., ibid. 31, 269 共1995兲; T. Gyalog and H. Thomas, Z. Phys.
B: Condens. Matter 104, 669 共1997兲.
关31兴 H. Hölscher, U. D. Schwarz, and R. Wiesendanger, Europhys.
Lett. 36, 19 共1996兲; Surf. Sci. 375, 395 共1997兲; O. Zwörner et
al., Appl. Phys. A: Mater. Sci. Process. 66, S263 共1998兲.
关32兴 M. H. Muser, Phys. Rev. Lett. 89, 224301 共2002兲.
036123-11
PHYSICAL REVIEW E 77, 036123 共2008兲
关33兴 J. D. Gunton and M. C. Yalabik, Phys. Rev. B 18, 6199
共1978兲.
关34兴 C. Unger and W. Klein, Phys. Rev. B 29, 2698 共1984兲.
关35兴 J. Kurkijarvi, Phys. Rev. B 6, 832 共1972兲.
关36兴 L. D. Jackel, W. W. Webb, J. E. Lukens, and S. S. Lei, Phys.
Rev. B 9, 115 共1974兲.
关37兴 A. Garg, Phys. Rev. B 51, 15592 共1995兲.
关38兴 D. S. Fisher, Phys. Rev. B 31, 1396 共1985兲.
关39兴 Y. Sang, M. Dube, and M. Grant, Phys. Rev. Lett. 87, 174301
共2001兲.
关40兴 O. K. Dudko, A. E. Filippov, J. Kalfter, and M. Urbakh, Chem.
Phys. Lett. 352, 499 共2002兲.
关41兴 A. L. Barab’asi and H. E. Stanley, Fractal Concepts in Surface
Growth 共Cambridge University Press, Cambridge, U.K.,
1995兲.
关42兴 The length l␦ can be obtained from a microscopic model of
roughness; see e.g., J. Krug, Adv. Phys. 46, 139 共1997兲.
关43兴 L. D. Landau and E. M. Lifshitz, Theory of Elasticity 共Perga-
mon, Oxford, 1970兲.
关44兴 R. Kant, Phys. Rev. E 53, 5749 共1996兲.
关45兴 H. J. Hertz, J. Reine Angew. Math. 92, 156 共1881兲.
关46兴 K. L. Johnson, Contact Mechanics 共Cambridge University
Press, Cambridge, U.K., 1985兲.
关47兴 In principle, the combined product of density and curvature
should be averaged over the fluctuations of the surface:
具ⵜ2h␪共z − h兲典. A separate average is appropriate since each
quantity depends on different regions of the frequency of fluc-
tuations. The curvature is dominated by the large-wavelength
part of the spectrum while the density is dominated by the
low-wavelength components. Again, it is straightforward to
show this for a Gaussian interface.
关48兴 The neglect of the y coordinate has no influence on the results
presented in this paper.
关49兴 P. Hanggi, P. Talkner, and M. Borkocev, Rev. Mod. Phys. 62,
251 共1990兲.
关50兴 A. Buldum and S. Ciraci, Phys. Rev. B 55, 2606 共1997兲.
关51兴 Note that, at very low velocities, thermally activated events
can occur before the spinodal limit considered here. However,
this corresponds to velocities below 1 nm/sec, well below the
experimental range 关39兴.
关52兴 M. P. Allen and D. J. Tildesley, Computer Simulations of Liq-
uids 共Clarendon, Oxford, 1990兲.
关53兴 D. Forster, Hydrodynamic Fluctuations, Broken Symmetry and
Correlation Functions 共Addison-Wesley, Reading, MA, 1983兲.
关54兴 A detailed analysis shows that the dissipation coefficient ␥
may itself depend on the position of the atoms. This effect is
neglected here. See, e.g., L. N. Kantorovich, Phys. Rev. B 64,
245409 共2001兲.
关55兴 P. M. Chaikin and T. C. Lubensky, Principles of Condensed
Matter Physics 共Cambridge University Press, Cambridge,
U.K., 1995兲.