Download as pdf or txt
Download as pdf or txt
You are on page 1of 23

Downloaded from rspa.royalsocietypublishing.

org on June 11, 2012

Adhesion of elastic spheres


J. A. Greenwood
Proc. R. Soc. Lond. A 1997 453, 1277-1297
doi: 10.1098/rspa.1997.0070

References Article cited in:


http://rspa.royalsocietypublishing.org/content/453/1961/1277#related-urls

Email alerting service Receive free email alerts when new articles cite this article - sign up in the box at the top right-hand
corner of the article or click here

To subscribe to Proc. R. Soc. Lond. A go to: http://rspa.royalsocietypublishing.org/subscriptions


Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

Adhesion of elastic spheres


B y J. A. G r e e n w o o d
Department of Engineering , University of Cambridge, Trumpington Street,
Cambridge CB2 1PZ, UK

Bradley (1932) showed that if two rigid spheres of radii R1 and R2 are placed in
contact, they will adhere with a force 2πR∆γ, where R is the equivalent radius
R1 R2 /(R1 +R2 ) and ∆γ is the surface energy or ‘work of adhesion’ (equal to γ1 +γ2 −
γ12 ). Subsequently Johnson et al. (1971) (JKR theory) showed by a Griffith energy
argument (assuming that contact over a circle of radius a introduces a surface energy
−πa2 ∆γ ) how the Hertz equations for the contact of elastic spheres are modifed
by surface energy, and showed that the force needed to separate the spheres is equal
to (3/2)πR∆γ, which is independent of the elastic modulus and so appears to be
universally applicable and therefore to conflict with Bradley’s answer.
The discrepancy was explained by Tabor (1977), who identified a parameter µ ≡
R1/3 ∆γ 2/3 /E ∗2/3 ε governing the transition from the Bradley pull-off force 2πR∆γ to
the JKR value (3/2)πR∆γ. Subsequently Muller et al. (1980) performed a complete
numerical solution in terms of surface forces rather than surface energy, (combining
the Lennard–Jones law of force between surfaces with the elastic equations for a half-
space), and confirmed that Tabor’s parameter does indeed govern the transition.
The numerical solution is repeated more accurately and in greater detail, confirm-
ing the results, but showing also that the load–approach curves become S-shaped
for values of µ greater than one, leading to jumps into and out of contact. The JKR
equations describe the behaviour well for values of µ of 3 or more, but for low val-
ues of µ the simple Bradley equation better describes the behaviour under negative
loads.

1. Introduction
Bradley (1932) showed that if two rigid spheres of radii R1 and R2 are placed in
contact, they will adhere with a force 2πR∆γ, where R is the equivalent radius
R1 R2 /(R1 +R2 ) and ∆γ is the surface energy or ‘work of adhesion’ (equal to γ1 +γ2 −
γ12 ). Bradley obtained this result from an exact solution for the potential between two
spheres, in which he showed that if the potential between individual pairs of atoms
varies with distance x as x−n , then that between spheres varies as d−n+5 (where d is
the gap). If instead of assuming a law of force between atoms, we begin by assuming
a law of force between surfaces, and use the approximation of replacing the spheres
by paraboloids so that the gap near the contact point is exactly h = r2 /2R, the
result is elementary: for if the force between surfaces separated by a distance h is
σ(h) per unit area, then the force between the spheres will be
Z ∞ Z ∞
T = 2πrσ(h) dr = 2πR σ(h) dh = 2πR∆γ, (1.1 a)
0 0
Proc. R. Soc. Lond. A (1997) 453, 1277–1297 c 1997 The Royal Society

Printed in Great Britain 1277 TEX Paper
Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

1278 J. A. Greenwood
since the surface energy ∆γ is just the work required to move the surfaces from
contact to infinity. This argument was first introduced by Derjaguin (1934) in terms
of energy rather than forces, and the Derjaguin approximation that the interaction
energy between small areas of solids, which may be curved and slightly inclined to
each other, is the same as the energy per unit area between infinite plane solids, is
perhaps more immediately acceptable than the equivalent approximation for surface
forces: either may be regarded as justified by the agreement with Bradley’s answer.
The argument may readily be extended to give the force between two rigid spheres
separated by a distance h0 :
Z ∞
T (h0 ) = 2πR σ(h) dh. (1.1 b)
h0

For the Lennard–Jones law of force (Muller et al. 1980, 1983; Pashley 1984; Attard
& Parker 1992)
 3  9 
8∆γ ε ε
σ(h) = − , (1.2)
3ε h h
(where ε is the interatomic spacing) the force between two spheres becomes
  2  8 
4 ε 1 ε
T (h0 ) = 2πR∆γ − . (1.3)
3 h0 3 h0
Bradley only takes h0 to be the atomic separation ε while Derjaguin envisages it
to be a variable: so equation (1.3) should perhaps be referred to as the Bradley–
Derjaguin equation; but to avoid confusion with the better known Derjaguin (or
DMT) equation (Derjaguin et al. 1975) it will be convenient to call (1.3) the Bradley
equation.
What happens when the rigid spheres are replaced by elastic spheres? In the
absence of surface forces, it is well-known (e.g. Johnson 1985) that under an applied
load W the spheres deform to make contact over a circle of radius a given by Hertz’s
equation
3 WR
a3 = , (1.4)
4 E∗
where E ∗ is the ‘contact modulus’ given by 1/E ∗ = (1 − ν12 )/E1 + (1 − ν22 )/E2
(see below), and that the centres of the spheres ‘approach’ by a distance α = a2 /R
with respect to their position at first contact. Johnson et al. (1971) (‘JKR theory’)
showed by a Griffith energy argument, assuming that contact over a circle of radius
a introduces a surface energy −πa2 ∆γ, that the effect of surface energy is to modify
the Hertz result to
4E ∗ a3 p
= W + 3πR∆γ + 6πR∆γW + (3πR∆γ)2 . (1.5)
3R
(Subsequently Maugis & Barquins (1978) showed that the same result may be ob-
tained readily by application of the fracture mechanics√principle pthat the stress in-
tensity factor at the edge of the contact is N ≡ K/ 2π = E ∗ ∆γ/π, see also
Greenwood & Johnson (1980).)
From equation (1.5) the minimum value of the load W , i.e. the pull-off force Tmax
is given by Tmax = −W = 32 πR∆γ, which, rather unexpectedly, is independent of
the elastic modulus, and so appears to be equally applicable to rigid spheres and to
conflict with Bradley’s answer.
Proc. R. Soc. Lond. A (1997)
Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

Adhesion of elastic spheres 1279


The discrepancy was explained by Tabor (1977), who noted the existence in the
JKR theory of a neck around the contact area, with a height of order (R∆γ 2 /E 2 )1/3 ,
and argued that a surface energy analysis can only be valid when this height is
greater than the range of action of the surface forces, so that separated surfaces are
truly separated. Thus a parameter µ = (R1/3 ∆γ 2/3 /E ∗2/3 ε) will determine whether
or not the sphere may be treated as rigid. We shall refer to µ as the Tabor parameter
although Tabor used Young’s modulus E in the original. Subsequently Derjaguin and
his collaborators (Muller et al. 1980) performed a complete numerical solution, using
the Lennard–Jones law of force and the elastic half-space equations, and showed that
Tabor’s parameter, with a different numerical factor, does indeed govern the tran-
sition from the Bradley pull-off force 2πR∆γ to the JKR pull-off force (3/2)πR∆γ.
They also gave the detailed shape and pressure distribution for a high value of µ, and
demonstrated almost perfect agreement with the predictions of JKR theory. How-
ever, the paper is rather lacking in details of the calculation. There is no information
about which cases were evaluated, and integrals apparently evaluated numerically
have singular integrands; and no general results are given other than the variation
of the pull-off force: there are, for example, no load–approach or load–area curves.
The present work was originally envisaged as filling these gaps, and in particular as
investigating jumps into and out of contact. But in a recent paper (Attard & Parker
1993) the numerical solution was repeated, more accurately (using up to 1500 nodes
instead of 90, but apparently evaluating the same integrals numerically), with very
different results: although as µ increases the pull-off force initially decreases from the
Bradley value, following the Muller curve, it reaches a minimum of about 1.6πR∆γ
when µ ≈ 1 and then appears to increase indefinitely, the highest value quoted be-
ing 3.4πR∆γ i.e. 1.7× Bradley’s value at µ = 6.7, the largest µ studied†. Attard &
Parker also claim that for µ > 1 the pull-off force is variable, depending on the max-
imum load which has been applied. However, another contribution is that of Maugis
(1992), who gives an analytical analysis using a Dugdale model (i.e. the surface force
is a fixed value σmax when the separation is less than ∆γ/σmax ) and demonstrates
that for this case there is again a continuous decrease from the Bradley value to
the JKR value as a parameter corresponding to the Tabor parameter increases from
zero to ∞. Maugis also finds that load–approach and load–area curves are in good
agreement with those predicted by the JKR theory.

2. Analysis
We are concerned only with small contacts between much larger spheres, so that
the gap between the spheres, if they were rigid and just in contact, would be
r2 r2 r2 1 1 1
h= + ≡ , where ≡ + . (2.1)
2R1 2R2 2R R R1 R 2
It is often convenient to think of the contact as occurring between a sphere of radius
R and a plane.
The Hertzian approximation is to assume that the elastic deflections of the bodies
can be calculated using the equations for a half-space, since the curvature of the

† Attard & Parker use a parameter σ ≡ 0.5µ3/2 ; σ 6 8.7. See Appendix A for a comparison of the
different parameters in use.

Proc. R. Soc. Lond. A (1997)


Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

1280 J. A. Greenwood
spheres is insignificant on the scale of the contact. The deflection of a half-space
with elastic constants E1 and ν1 under a normal point load P is simply
P (1 − ν12 )
w(r) = ,
πE1 r
so that the combined deflection of two bodies due to a contact force P is
P
w(r) = ,
πE ∗ r
where E ∗ is the contact modulus,
1 1 − ν12 1 − ν22
= + . (2.2)
E∗ E1 E2
It is convenient to think of one body as rigid, and the other as elastic: its plane
strain modulus must then be taken to be E ∗ . Thus, the Hertz equation governing
the radius of the circle of contact under a load W between two spheres of different
radii and elasticities
3 WR
a3 = , (2.3)
4 E∗
may be thought of as describing the contact between a rigid sphere and an elastic
half-space.
For a general axisymmetric pressure distribution p(r) the deflection is
Z
1
w(r) = ∗ p(s)G(r, s)s ds, where
E
4 s 4 r
G(r, s) = K , s < r; K , r<s (2.4)
πr r πs s
and K(k) is an elliptic integral of modulus k†.
In the present model the sphere does not touch the plane: we assume that between
two bodies separated by a distance h there is a tensile stress
 
8∆γ  ε 3  ε 9
σ(h) = − , (2.5)
3ε h h
where ε is the interatomic spacing; only when the separation is equal to ε does
the force vanish.
√ (The maximum tensile stress occurs when h = 31/6 ε and is equal
to 16∆γ/9 3ε.) We use the Derjaguin approximation of assuming that this force
law can be applied between small areas of surfaces, even when these are inclined or
curved. The separation between sphere and plane is taken to be
r2
h = −α + ε + + w(r), (2.6)
2R

† The equation is ’well-known’ and easily derived (see Greenwood & Tripp 1967; Hills et al. 1993).
Attard & Parker put the factor π in the numerator and confuses the modulus k with the parameter k2 :
but these are surely misprints not affecting their calculation. (The π-misprint is corrected by Attard &
Parker 1994). Applying Landen’s transformation gives the form
Z  √ 
4 s 2 rs
w(r) = p(s) K ds,
πE ∗ s+r r+s

quoted by Johnson (1985) and used by Muller et al. (1980).

Proc. R. Soc. Lond. A (1997)


Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

Adhesion of elastic spheres 1281


where α is the approach of the centre of the sphere with respect to the zero force
position h = ε. The problem is therefore to find a pressure distribution p(r) ≡ −σ(h)
satisfying equations (2.4), (2.5) and (2.6).

(a ) Evaluation of the elastic integral


The kernel G(r, s) of the deflection integral is singular when r = s, so that direct
numerical evaluation is impossible. This is commonplace of contact problems and is
treated in the usual way by assuming that over discrete elements the pressure takes
a simple form for which an analytical answer is known. Here the pressure is assumed
to increase linearly from zero at ri−1 to pi at ri and then fall linearly to zero at
ri+1 (the method of overlapping triangles, JohnsonP(1985)). Hence (see Appendix B)
equation (2.4) is approximated by wi = (1/E ∗ ) Gij pj where the matrix Gij is
calculated once for all. Note that if the radial intervals are equal, Gij is directly
proportional to the interval size.

(b ) Non-dimensionalization
We set h − ε = εH, α = εα∗ , r = βu and p = P ∆γ/µε. Then defining
 2
R2 ∆γ R∆γ
β3 = and µ3
= , (2.7)
E∗ E ∗2 ε3
the equations become
X
Hi = −α∗ + 12 µu2i + µ G0ij Pj (2.8)

and
 
8 1 1
Pi = 3
µ − . (2.9)
(Hi + 1)3 (Hi + 1)9
Note that the dimensionless radial distance and the pressure are independent of the
interatomic distance ε, and so can be directly compared with the same quantities in
the JKR theory in which no interatomic distance exists.

(c ) Numerical procedure

For large separations (α large and negative) the surface forces are weak (tensile)
and barely affect the shape. A simple iteration, starting with the rigid body shape
or that of a neighbouring case, calculating the forces from (2.9) and recalculating
the shape using (2.8) rapidly converges. For small values of the Tabor parameter µ
the same method is adequate for all values of the approach studied (i.e. until the
load became compressive) provided a ‘convergence factor’ is used: that is, the new
values of the gap were taken as H new = H old + c(H 0 − H old ) with sometimes rather
small values of c (as low as 0.005) being necessary to avoid instability. For larger
values of µ this failed. Instead the residual error ei was found at each point as the
difference between the initial value of Pi used in (2.8) and the final value from (2.9);
and a Newton–Raphson estimate made of the correct change to Pi , subject to the
condition that with each change to Pi , the immediate neighbours Pi−1 and Pi+1 were
changed by half the amount in the opposite direction, so that the effect on the shape
is purely local. Here too, under-relaxation was sometimes needed; changes from 0.5
to 1.5 times the proper value were used.
Proc. R. Soc. Lond. A (1997)
Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

1282 J. A. Greenwood

Figure 1. (a) Variation of load with approach for low values of µ. (b) Numerical results for
µ = 0.05 compared with the Bradley curve.

3. Results
Figure 1a shows the variation of load with approach for low values of µ, together
with the Bradley equation (equation (1.3)). It appears that this is the limit as µ → 0:
the results for µ = 0.05 (figure 1b) are almost indistinguishable from the Bradley
curve. In all cases the maximum tensile force is slightly below the Bradley value, and
occurs at a small positive separation.
Using the Bradley equation for h < ε does perhaps need some justification: cer-
tainly Bradley tacitly assumes that his (rigid) spheres can do no more than touch
each other (‘since the spheres are touching in the experiment, d is the molecular
diameter’). But if the compressive stresses are to be included as one component of
the surface forces between two solids, values of h < ε must certainly occur. The
author finds it helpful to regard the two spheres as non-interacting bodies to which
Proc. R. Soc. Lond. A (1997)
Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

Adhesion of elastic spheres 1283

Figure 1. (c) Numerical results for µ = 0.02 for compressive loads, compared with Bradley,
JKR and DMT–M theories.

are bonded a layer of equilibrium thickness ε of a material possessing the remarkable


elastic stress–strain law (1.2): the term ‘æther layer’ seems appropriate. Certainly
equation (1.3) appears to be the asymptote of the numerical solutions as µ → 0, as
we should expect when the elastic modulus of the spheres becomes sufficiently large.
However, figure 1c shows that this is not the whole story. For large compressive
loads (h appreciably less than ε) the æther layer ultimately becomes stiffer than
the elastic solid, and the load increases in an elastic way, that is, more-or-less as
α3/2 , though with an offset. For comparison, two other theoretical curves are shown:
Maugis’ approximation to the DMT theory (see below) and the JKR curve. The first
of these consists in adding a fictitious load 2πR∆γ to the actual load and then using
the Hertz equation to find the approach; it may be shown that the JKR curve, for
high loads, is almost equivalent to this, except that the fictitious load is πR∆γ.
Figure 2a shows that at µ ∼ 1 a new effect appears. The curves have been growing
steadily steeper, and near this point have a vertical tangent. For µ > 1 the computed
points lie on two separate curves; and there is a range of (−α∗ ) for which there are two
possible values of the load. As the surfaces are brought together (in the computation),
the load follows the lower branch until point A (figure 2a): but when the shape at
point A is used as the initial guess for a slightly smaller value of (−α∗ ), the solution
converges on point B. If the separation is now increased, the load follows the upper
curve as far as point C but with a further increase the load drops to D and then
continues along the lower branch.
For larger values of µ it is convenient to change the independent variable from α∗
to α∗ /µ so that a single JKR curve may be added to the figure†. Figure 2b shows
that as µ increases, the maximum tensile force continues to fall, and to occur at

† In these variables δ ∗ ≡ α∗ /µ, W † ≡ W/2πR∆γ, s ≡ a/a0 where a30 ≡ 92 π(R2 ∆γ/E ∗ ) the JKR
√ √
curve is δ ∗ = ( 92 π)2/3 ( 23 s − s2 ), W † = 3s(s2 − s) with a largest tensile force of W † = −0.75 when
s=2 −2/3 and a maximum separation of δ = −0.75π 2/3 when s = 6−2/3 .

Proc. R. Soc. Lond. A (1997)


Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

1284 J. A. Greenwood

Figure 2. (a) At µ ∼= 1 the curves develop a vertical tangent. As the separation is reduced the
load follows the lower branch to point A but then jumps to point B. If the separation is now
increased the load follows the upper branch as far as point C but then drops to D and continues
along the lower branch. In an apparatus of finite stiffness, the jumps will be from A0 to B0 and
from C0 to D0 . (b) Variation of load with approach for higher values of µ, together with JKR
theory.

steadily increasing values of separation (i.e. at almost constant values of α∗ /µ). The
high-load branches of the family of curves appear to condense on to the JKR solution
as a limit. In particular, for µ = 5 the pull-off force is within 2% of the JKR value
and occurs at the JKR separation, while the jump-off separation is within 1% of the
Proc. R. Soc. Lond. A (1997)
Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

Adhesion of elastic spheres 1285

Figure 3. Variation of load with approach when the central gap h(0), is specified: (a) µ = 1; (b)
µ = 2.

Proc. R. Soc. Lond. A (1997)


Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

1286 J. A. Greenwood

Figure 3. Variation of load with approach when the central gap h(0), is specified: (c) µ = 3.

JKR value (the load is perhaps slightly higher than the JKR load). Inevitably the
low-load branch is not related to the JKR theory: on the low branch there is nothing
identifiable as a contact area, and the very concept of surface energy is inapplicable‡.
It seemed likely that the two branches formed part of a single S-curve, so a modified
method of solution was tried in the hope of establishing this. Instead of specifying
the approach, α∗ , the central gap, h(0) was fixed. (This is equivalent to specifying
the central pressure, p(0).) Figures 3a–c show the results. The curve relating load to
approach is indeed a single, continuous curve: for µ = 1 or 2 the curve is S-shaped,
but for larger values of µ the shape is even more sinuous, first crossing the JKR curve
but then twisting round to recross and envelop this. Clearly in an ideal experiment
jumps on or off will occur when the curve is vertical: the calculated points found by
prescribing α∗ occur slightly earlier because of computing incompetence.
The jumps will be real features of a fixed-load device, so that the value at which
they first appear, µ ∼ 1, is of some significance.. For lower values of µ the loading
is smooth and completely reversible, and presumably involves no energy dissipation.
For higher values there will be jumps and energy losses. Note that in a fixed-grips
device there is no ‘pull-off force’ for µ < 1; the (tensile) force simply decreases steadily
to zero as the separation increases. However, for µ > 1 we can identify a ‘pull-off’
force in a fixed-grips device, though this is a slight misnomer since the force drops
to a lower value rather than to zero. Equally, of course, there will be a jump into
contact, when, in a fixed grips rig, the load jumps from its value at A to that at B

‡ The surface energy is defined as the work needed to separate the surfaces from the equilibrium
position to infinity: but the shape of the surfaces before jumping-on depends on only part of the surface
force curve—typically from h = 2ε → ∞.

Proc. R. Soc. Lond. A (1997)


Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

Adhesion of elastic spheres 1287

Figure 4. Details of the ‘jump into contact’ for µ = 5. (a) Pressures, (b) shape.

(figure 2a), or when at constant load the separation jumps across to a point in the
second quadrant and becomes an ‘approach’.
Figure 4 shows details of the jump into contact for µ = 5. Before the jump,
the minimum separation (i.e. the thickness (h − ε) of the ‘æther layer’) is 2.554ε,
(P = −0.2968) while the jumped-on state has a minimum separation −0.0036ε
(P = +0.2943) The shape before jumping-on cannot depend on the law relating force
to separation for separations less than this minimum value (h − ε) = 2.554ε and so
cannot be related to the surface energy ∆γ; it is less certain but seems probable that
the condition for jumping-on is equally unrelated to ∆γ.
(a ) Area of contact
There is much to be said for the view expressed by Attard & Parker, that contact
radius is an ill-defined concept best left alone: certainly attempts to measure it
Proc. R. Soc. Lond. A (1997)
Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

1288 J. A. Greenwood
optically run into serious difficulties over how small a gap can be detected, while
a logical definition seems impossible when the surfaces never actually contact but
merely compress the æther layer. However, experimenters are successfully correlating
friction measurements with the calculated JKR contact area so the question must
be faced.
It is clear that the traditional contact mechanics definition—that the contact area
is where there is a positive pressure—will not do: contacts under negative loads
are well established. In fracture mechanics, which geometrically is very close to a
contact problem, no-one would suggest that the crack tip is where the stress changes
sign; and the contact problem must be treated in a consistent way. Once we have
entered the tensile region, it is clear that the only distinctive point is the maximum
tensile stress, and that we must regard our solids as in contact (or uncracked) when
the stress is increasing with separation, and separated (or cracked) when the stress
decreases. (The same conclusion is reached by Maugis (1992) in his discussion of
realistic force–separation laws, although his calculations, based on a constant tensile
stress, necessarily use a slightly modified form, that the contact edge is where the
stress first reaches the maximum tensile stress.) Conveniently, we find that for large
values of µ, the tensile maximum approximates to the tensile singularity of the JKR
solution, and, as shown earlier by Muller et al., occurs at the same location.
Thus, for the Lennard–Jones force law, we define the contact radius as the radius at
which the thickness of the æther layer becomes 1.2009ε, i.e. where the gap thickness is
0.2009ε. (This is much more readily found than the precise location of the maximum
stress.) Figure 5 shows how the contact radius varies with load. Above the nose of
the JKR curve, the numerical results lie close to, but slightly below, the JKR curve,
very much as found by Maugis. Below the nose the results again approximate to
the JKR curve, but are distinctly lower, resembling the behaviour found in Maugis’
model but for a given value of µ somewhat closer to JKR than his answers.

4. Discussion
(a ) The DMT theory
The knowledgeable reader will have noted with surprise the absence of any refer-
ence to the ‘DMT theory’ (Derjaguin et al. 1975) which is generally quoted as the
antithesis of the JKR theory: Bradley’s value for the pull-off force, 2πR∆γ, is often
referred to as the DMT value. The DMT theory does not recognize the possibil-
ity of penetrating the ‘æther layer’, the layer of initial thickness ε which represents
the surface forces, which means that compressive surface forces are excluded. Thus,
the forces are arbitrarily divided into compressive elastic forces, unrelated to the
Lennard–Jones law, and tensile surface forces, which produce no elastic deformation
When the approach α is negative and there are no compressive forces, the elastic
sphere retains its undeformed shape, so that the (tensile) load follows the Bradley
equation (1.3): when α becomes positive a contact of radius a given by the Hertz
equation a2 = αR is formed, and the load then consists of two terms: the elastic com-
pressive Hertzian load (4/3)(E ∗ a3 /R) and a tensile load due to the surface forces
acting across the gap outside the Hertzian contact.
Derjaguin et al. (1975) calculated this tensile load by a ‘thermodynamic’ method,
and found that as α increased, the load steadily decreased from the Bradley value
2πR∆γ; it follows that the maximum pull-off force still has the Bradley value. How-
ever, Derjaguin and his colleagues (Muller et al. 1983) later demonstrated, somewhat
Proc. R. Soc. Lond. A (1997)
Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

Adhesion of elastic spheres 1289

Figure 5. Variation of contact radius with load.

Table 1. Surface forces according to DMT model

α∗ 0.001 0.002 0.005 0.01 0.02 0.05 0.1 0.2 0.5 1


f (α∗ )/α∗3/2 1.612 1.530 1.386 1.246 1.082 0.840 0.652 0.477 0.286 0.180

grudgingly, that the DMT calculation is wrong (‘the force method should be pre-
ferred’); a much clearer exposition of the error and the implications has been given
by Pashley (1984). Both papers give a plot of the corrected tensile load and these
roughly agree: as the approach, α, increases, the tensile force is always greater than
the Bradley value (and as Pashley shows, this is self evident). For small values of the
approach, α, Muller et al. state that the contribution of the surface forces is
Ts ∼ 2πR∆γ(1 + 1.84(α∗ )3/2 + O((α∗ )5/2 )), where α∗ ≡ α/ε > 0 (4.1)
but this appeared to conflict somewhat with Pashley’s curve, so the calculation was
repeated, giving the values of f (α∗ ) ≡ Ts /2πR∆γ −1 shown in table 1. These support
Pashley’s numerical integration, as far as can be determined from his graph, rather
than that of Muller et al., but for smaller values of α∗ our calculations confirm the
analytical result (4.1), fitting f (α∗ ) = 1.839(α∗ )3/2 −7.7(α∗ )2 for α∗ between 0.000 02
and 0.0001.
Adding the Hertz load gives the total load as
  ∗ 3/2 
2 α
W = (2πR∆γ) − 1 − f (α∗ ) . (4.2 a)
3π µ
Proc. R. Soc. Lond. A (1997)
Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

1290 J. A. Greenwood
This also increases with approach when α ≈ 0, except for µ < 0.237†. These curves
are shown in figure 6a, and are clearly unrelated to the numerical results, shown on
the same axes in figure 6b. When (α∗ /µ) is used as the abscissa, the effect of the
term in α∗ in equation (4.2) is to spread the curves by an inadequate amount and
in the wrong order . Note also that in the numerical solutions, pull-off never occurs
with α < 0.
The curve labelled DMT-M needs some comment. For a given value of the sepa-
ration variable (α∗ /µ), the value of α∗ becomes vanishingly small as µ → 0, so that
f (α∗ ) equals zero and the adhesive load Ts remains at the Bradley value 2πR∆γ (cf.
Pashley or Maugis). Using this value for µ ∼ 0 and α∗ 6= 0, equation (4.2) becomes
  ∗ 3/2 
2 α
W = (2πR∆γ) − 1 , α∗ > 0 (4.2 b)
3π µ
which conveniently is a single curve and so may be compared with the single curve
found by JKR theory; indeed, Maugis (somewhat misleadingly) labels this curve
‘DMT’ in his work. It has been included in figure 6a, where it is the proper limit of
the DMT solutions as µ → 0; but is hardly representative of non-zero values of µ.
Figure 6b shows that it does not agree with the numerical calculations either; but
the fundamental objection to it is that since, by virtue of the dependence on µ it
still depends on E ∗ , it is not a rigid body limit.
In fact, the numerical demonstration was unnecessary, for it is clear from the form
of equation (4.2) that the DMT theory does not possess a rigid-body limit‡, that is,
a form independent of the elastic modulus E ∗ . We conclude that the division of the
forces between solids into surface forces and elastic contact pressures is unjustifiable,
and the behaviour near α ∼ 0 as the solids become stiffer and µ → 0 is not the DMT
theory but, as shown in figures 1b and 1c, the simple Bradley theory. (As described
above (figure 1c), when the load becomes compressive (α  0) none of the theories
is adequate.)

(b ) Attard & Parker


Figure 7 shows the calculated values of the pull-off force. These are unique values
corresponding to the maximum of the force–separation curves; the only variation
being if unloading begins immediately after jumping-on, before the maximum is
attained, when a slightly lower value would be found. The curve is close to that
plotted by Muller et al. There is qualitative agreement with the results of Maugis’
analysis, which, it will be recalled, are for a simplified surface force curve, so that
no exact correspondence exists between the Maugis parameter λ and the Tabor
parameter µ: all three solutions show a steady decrease from the Bradley value to a
value close to the JKR value.
In contrast, Attard & Parker’s results initially follow the same curve, but then
increase to values considerably larger than the Bradley value, while becoming vari-
able. Attard & Parker appear to have a more accurate solution than that described
here, in that they used ‘200 to 1500’ nodes compared with at most 200 in the present

† Pashley asserts that the maximum force is at a = α = 0 for µP < 0.8, i.e. for µ < 0.27, but here it
seems that the MYD answer is more accurate.
‡ The fundamental problem is that the gap outside a Herzian contact of given radius does not depend
on E ∗ and so is independent of µ and in a sense is ‘rigid’. So varying the elasticity does not convert one
‘rigid’ shape into the other.

Proc. R. Soc. Lond. A (1997)


Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

Adhesion of elastic spheres 1291

Figure 6. (a) Recalculated DMT theory plotted against (α∗ /µ). (b) Numerical results for low
values of µ plotted against (α∗ /µ) together with the limiting case of DMT theory.

Proc. R. Soc. Lond. A (1997)


Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

1292 J. A. Greenwood

Figure 7. Variation of the pull-off force with Tabor parameter. The present values steadily
decrease from the Bradley value towards the JKR value as found by MYD. Maugis’ analytical
solution behaves similarly: note that there is no direct equivalence between λ and µ. In contrast,
Attard & Parker’s values increase and are variable for µ > 2.

work; their (uniform) node spacing was varied from ‘10 to 100 nm’. This suggests
that the tensile spike was adequately defined; and the authors claim that ‘the results
were not very sensitive to the choice of grid or spacing, except for compliant bodies
with large adhesions’. The present results were distinctly sensitive to the choice of
interval for µ > 2. In particular, early results for µ = 5 gave a higher value for the
pull-off force than was found for µ = 3, though a much smaller increase than that
found by Attard & Parker. Inspection of plots of these early pressure distributions
gave no reason for concern : the interval appeared completely adequate. The only
unsatisfactory feature appeared to be (very) small discontinuities in the force separa-
tion curve corresponding to movements of the maximum tension with respect to the
grid points, suggesting that the spikes were inadequately defined. When the interval
was reduced sufficiently (halved) to give a smooth force–approach curve considerably
lower values of the load were obtained. The change made no detectable difference to
the pressure distributions: so it is clear that the problem was only in the adequate
definition of the spike, and clear also that for large values of µ the spike makes a
major contribution to the total load. (It is, of course, in these cases, at a relatively
large radial distance from the centre and so is weighted heavily.)
We are completely unable to suggest how, if we could obtain adequate definition
of the spike with 200 points, Attard & Parker might not have done so with 1500.
An alternative possibility for the source of the discrepancy lies in the method used
to calculate the elastic deformation. Attard & Parker appear to use equation (2.4)
directly, despite the fact that the integrand has a logarithmic singularity at s = r,
about which no comment is made! The author has known a research student, faced
with the same integral, who simply omitted the singular point from his summation;
one would hesitate to suggest this could occur in a mathematics department, but it
would certainly be of interest to know how the singularity was treated. It should be
Proc. R. Soc. Lond. A (1997)
Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

Adhesion of elastic spheres 1293


Table 2. Jumping-on and off

µ 1 2 3 5 6

jumping-on
−α∗ 1.628 2.547 3.218 4.241 4.671
−W/R∆γ 2.140 1.156 0.814 0.525 0.450
H(0) 0.647 1.263 1.699 2.361 2.632
(H(0) + 1)/µ3/7 1.647 1.681 1.686 1.686 1.685
−W/R∆γ (A & P) 1.907 1.053 0.744 0.480 0.411

jumping-off
−α∗ 1.742 3.374 4.980 8.178 —
−α∗ /µ 1.742 1.687 1.660 1.637 —
−W/R∆γ 3.233 2.856 2.751 2.708 —
H(0) 0.0706 0.0201 0.01172 0.00630 —
a∗ 0.421 0.596 0.650 0.700 —

recalled that Muller et al., also using a singular integrand without comment, obtained
results close to ours, using at most 90 points (though with a variable spacing).
Another possibility is more mundane. Our method had an appalling convergence
rate, many runs taking 20 000 iterations to achieve convergence (justifiable only by
the underuse of departmental computing facilities in the vacation). Even accepting
the effectiveness of the methods used by Attard & Parker to accelerate their con-
vergence, one wonders whether their solutions were really fully converged after 1000
iterations?
(c ) Jumping-on and off
Experimentally, one of the clearer features is whether contact and separation occur
smoothly or whether there is a sudden jump into or out of contact. The calculations
show that the load–separation curve becomes S-shaped when µ ∼ = 1, so that in a
fixed-grips experiment, jumps will occur when the tangent becomes vertical. For
an apparatus with a finite stiffness, the location of the jumps can be found by a
standard ‘load-line’ construction, determining where lines of the appropriate slope
(namely minus the stiffness of the apparatus) touch this curve (see figure 2a). Note
that this implies that in a particular apparatus there may well be jumps even for
a contact with µ = 0.5 or less; the criterion µ ∼ = 1 holds only for an infinitely stiff
apparatus.
The ‘nose’ of the S-bend will occur at a well defined value of separation, but,
without computing an unjustifiable number of additional solutions, at a less well
defined load. Table 2 gives the best values found.
Attard & Parker suggest an alternative criterion for jumping: that the response to
a perturbation of the shape should be larger than the perturbation. By considering a
perturbation of a particular form, they obtain a sufficient condition for instability†
w(0)σ 0 (h(0))/σ(h(0)) > −1 (4.3)

† A factor two has been omitted corresponding to the change in notation.

Proc. R. Soc. Lond. A (1997)


Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

1294 J. A. Greenwood
(their text seems somewhat confused between sufficient conditions for stability and
sufficient conditions for instability). By neglecting the short-range repulsive forces,
this leads to a simple criterion

(H0 + 1)7/2 = 3π 2µ3/2 . (4.4)
An alternative suggested by Pethica & Sutton is (H0 + 1)7/2 √ = (16π/3)µ3/2 , and in
Appendix C a third criterion is derived: (H0 + 1) = (5π/ 2)3/2 . Table 2 shows
7/2

that (H0 + 1) is indeed proportional to µ3/7 but that these numerical factors are all
substantially too large. Curve-fitting gives an equation
(H0 + 1)7/2 = 6.233µ3/2 [1 − 1.6/(H0 + 1)6 ], (4.5)
though the second term is of little importance except for µ = 1 (H0 = 1.62).
Attard & Parker’s criterion for the critical approach leads to an approximate
equation which can be used to predict the critical load, and the values are given in
table 2; the agreement is merely fair. We may also test their fundamental criterion
for jumps, equation (4.2), using the data found in the calculations. For jumping-off
their criterion is simply wrong, as is perhaps predictable since they regard the jump
as initiated at the centre of the contact, while it seems clear that jumping-off actually
takes place by peeling from the edge of the contact. At jumping-on, typical values
of the critical parameter seem to be around 0.6–0.8, somewhat below the suggested
value 1.0, so although the criterion can be used as a rough guide it does not seem to
be well founded: indeed it is not clear how a criterion based only on values at the
centre could consistently agree with our condition related to the overall load.

5. Conclusion
The force needed to separate two spheres decreases steadily from the Bradley
value 2πR∆γ to the JKR value (3/2)πR∆γ as the Tabor parameter µ = (R1/3
∆γ 2/3 /E ∗2/3 ε) increases. The force–separation curves for µ > 2 approximate to a
hybrid consisting of the JKR curve when the approach α is positive or only slightly
negative, with the Bradley curve for large separations {(−α) large}, while for µ 6 0.2
they approximate to the Bradley curve in the tensile region. For values of µ greater
than about unity, the approach curves are S-shaped or even more convoluted, so that
in a fixed-grips apparatus there will be jumps in the load when the tangents to the
curve are vertical; in an apparatus of finite stiffness jumps will occur even for lower
values of µ.
For values of µ > 3 the shapes and pressure distributions in the jumped-on state
are close to the predictions of the JKR theory, and the radius of contact, defined
as the location of the maximum tensile stress, agrees well with the JKR radius of
contact, as found previously by Muller et al.
The DMT theory, both in its original and corrected forms, is wrong both in prin-
ciple and in practice. Instead, the Bradley–Derjaguin equation
  2  8 
4 ε 1 ε
T (h) = 2πR∆γ − ,
3 h 3 h
which for all values of µ describes how the load varies with approach for large positive
separations is found for small values of µ to accurately describe the tensile region
even for negative separations.
Proc. R. Soc. Lond. A (1997)
Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

Adhesion of elastic spheres 1295


I thank Professor K. L. Johnson for his interest and advice, and especially for his well founded
belief in the JKR theory. I thank Dr D. Maugis for his concern which led me to re-examine
the almost-Hertzian case of low surface energy and substantial compressive load. The method
of obtaining solutions for higher values of µ was suggested by Professor A. A. Lubrecht, and
I regret that I have not taken his further advice and adopted a multigrid method of solution.
The concept required to obtain solutions on the unstable branch of the load–approach curve
was explained to me by Dr C. H. Venner.

Appendix A.
There is no agreement on the best form for Tabor’s parameter, so a conversion
table is needed. The subscripts do not appear in the original works, but may help
clarity.
 1/3
R∆γ 2
µ= (this work: also Pashley et al. (1984)),
E ∗2 ε3
 1/3 
R∆γ 2
µ= Tabor: amplified by MYD to
E 2 ε3


2 1/3 R
1/3
∆γ 2/3
µT = 6π , µT = 3.898µ ,
E ∗2/3 ε
 1/3
32 2R∆γ 2
µD = (MYD, Pashley, µD = 2.9208µ),
3π πE ∗2 ε3
 1/3  
9 σ03 R
λ= Maugis matching σmax , i.e. taking
2π E ∗2 ∆γ

16 ∆γ
σ0 = √ , gives λ = 1.1570µ ,
9 3 ε
 1/2
∆γ R
σ= (Attard & Parker, σ = 0.5(µ)3/2 ).
2E ∗ ε3

Appendix B. Evaluation of the elastic integral


Z
1
w(r) = p(s)G(r, s)s ds
E∗
where
4 s 4 r
G(r, s) = K , s < r; K , r < s. (2.4)
πr r πs s
For a constant pressure p̄ acting over the ring c1 < s < c2 , the displacement at r
of a single half-space is
4p̄
w(r) = [F (c2 , r) − F (c1 , r)] ,
πE 0
where r
Z min(c,r)
c2 − s2
F (c, r) = ds,
0 r2 − s2
Proc. R. Soc. Lond. A (1997)
Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

1296 J. A. Greenwood
which is equal to rE(c/r) or (r2 /c)B(r/c) depending on whether c < r or c > r:
both elliptic integrals are finite for all r. Early results were obtained by approximat-
ing the pressure by uniform bands and using this formulation, but suspicion of the
representation of the pressure spike led to its replacement by the piecewise linear
approximation. For this, no analytical form was found; instead the effect at r of a
pressure pj at cj falling to zero at cj−1 , cj+1 was found by numerical integration, and
so again involves consideration of the singularity at r = cj . (At r = cj−1 , r = cj+1
the pressure is zero, cancelling the singularity.) The method used was to treat nu-
merically a pressure equal to −pj at cj−1 , cj+1 , rising linearly to zero at cj and to
add the effect of a uniform pressure pj over the whole band found using the equation
above.

Appendix C. Jumping-on
Empirically, an excellent fit for the values of central gap at jump-on is
(H0 + 1)7/2 = 6.233µ3/2 [1 − 1.6/(H0 + 1)6 ]
(the second term is of little importance except for {H0 = 1.62}). The corresponding
values of the separation obey (|α∗ | + 1)7/2 = (3π 2 )µ3/2 , where, as yet, the (3π 2 ) is
purely empirical. Note that the introduction of the surface energy ∆γ in the guise
of µ is misleading: as explained earlier, at the jump, only the law of force for the
relevant range of separations is operative. Any attempt to transfer these conditions
to a different law of force could be disastrous.
A further, equally unsuccessful, attempt to predict the gap at the jump is as
follows.
Like Pethica & Sutton and Attard & Parker, the central displacement is calculated
on the assumption that the shape is the rigid body shape but with the correct central
gap, so that h = h0 + r2 /2R. If the force law is σ(h) = C/hn+1 this gives
Z ∞ √
2 2πR 1 (n − 12 )
w(0) = ∗ σ(r) dr = C ,
E 0 E ∗ hn+1/2
0
n
or, for n = 2,

2R 1 3π
w(0) = C ∗ 5/2 .
E h0 8
Assuming this law of force to be the leading term of the Lennard–Jones law, we can
set C = 83 ∆γε2 to give
 √  
π 2R ∆γε2
w(0) = .
E∗ 5/2
h0
The separation |−α| is equal to w(0) + h0 and the condition that changing the
load does not change |−α| implies that changing h0 does not change |−α|, so that
(dα/ dh0 ) = 0, i.e. (dw(0)/ dh0 ) = −1. √
Hence, at the jump, h0 = (5/2)w(0), leading to (H0 + 1)7/2 = (5π/ 2)µ3/2 on
returning to the film thickness datum of the paper.
Note that the other criteria may also be written in the form
√ h0 = kw(0): Attard
& Parker’s is h0 = 3w(0), while Pethica & Sutton’s is h0 = (8 2/3)w(0). The result
of the numerical calculations is much lower than any of these: in terms of the central
displacement w(0) found numerically, which is appreciably larger than that from the
rigid body equation above, h0 = 1.403w(0).
Proc. R. Soc. Lond. A (1997)
Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

Adhesion of elastic spheres 1297


References
Attard, P. & Parker, J. L. 1992 Deformation and adhesion of elastic bodies in contact. Phys.
Rev. A 46, 7959–7971.
Attard, P. & Parker, J. L. 1994 Deformation and adhesion of elastic bodies in contact: correction.
Phys. Rev. E 50, 5145.
Bradley, R. S. 1932 The cohesive force between solid surfaces and the surface energy of solids.
Phil. Mag. 13, 853–862.
Derjaguin, B. V. 1934 Theorie des Anhaftens kleiner Teilchen. Koll. Z. 69, 155–164.
Derjaguin, B. V., Muller, V. M. & Toporov, Yu. P. 1975 Effect of contact deformations on the
adhesion of particles. J. Coll. Interf. Sci. 53, 314–326.
Greenwood, J. A. & Johnson, K. L. 1981 The mechanics of adhesion of viscoelastic solids. Phil.
Mag. 43, 697–711.
Greenwood, J. A. & Tripp, J. H. 1967 The elastic contact of rough spheres. J. Appl. Mech 34,
153–159.
Hills, D. A., Nowell, D. & Sackfield, A. 1993 Mechanics of elastic contacts. Oxford: Butterworth-
Heineman.
Johnson, K. L., Kendall, K. & Roberts, A. D. 1971 Proc. R. Soc. Lond. A 324, 301–313.
Johnson, K. L. 1985 Contact mechanics. Cambridge University Press.
Maugis, D. & Barquins, M. 1978 Fracture mechanics and the adherence of viscoelastic bodies.
J. Phys. D 11, 1989–2023.
Maugis, D. 1992 Adhesion of spheres: the JKR–DMT transition using a Dugdale model. J. Coll.
Interf. Sci. 150, 243–269.
Muller, V. M., Yuschenko, V. S. & Derjaguin, B. V. 1980 On the influence of molecular forces
on the deformation of an elastic sphere and its sticking to a rigid plane. J. Coll. Interf. Sci.
77, 91–101.
Muller, V. M., Derjaguin, B. V. & Toporov, Yu. P. 1983 On two methods of calculation of the
force of sticking of an elastic sphere to a rigid plane. Coll. Surf. 7, 251–259.
Pashley, M. D., Pethica, J. B. & Tabor, D. 1984 Adhesion and micromechanical properties of
metal surfaces. Wear 100, 7–31.
Pashley, M. D. 1984 Further consideration of the DMT model for elastic contact. Coll. Surf. 12,
69–77.
Pethica, J. B. & Sutton, A. P. 1988 On the stability of a tip and flat at very small separations.
J. Vac. Sci. Technol. A 6, 2490–2498.
Tabor, D. 1977 Surface forces and surface interactions. J. Coll. Interf. Sci. 58, 2–13.

Received 16 April 1996; accepted 15 August 1996

Proc. R. Soc. Lond. A (1997)


Downloaded from rspa.royalsocietypublishing.org on June 11, 2012

You might also like