Вы находитесь на странице: 1из 9

Coefficient of restitution of colliding viscoelastic spheres

Rosa Ramı́rez1 , Thorsten Pöschel2 , Nikolai V. Brilliantov2,3 , and Thomas Schwager2


1
Dpto. de Fisica. F.C.F.M., Universidad de Chile, Casilla 487-3, Santiago, Chile
2
Humboldt-Universität zu Berlin, Institut für Physik, Invalidenstr. 110, D-10115 Berlin, Germany
arXiv:cond-mat/9905382v1 [cond-mat.stat-mech] 26 May 1999

http://summa.physik.hu-berlin.de/∼kies/
3
Moscow State University, Physics Department, Moscow 119899, Russia
(February 1, 2008)
We perform a dimension analysis for colliding viscoelastic spheres to show that the coefficient of
normal restitution ǫ depends on the impact velocity g as ǫ = 1 − γ1 g 1/5 + γ2 g 2/5 ∓ . . ., in accordance
with recent findings. We develop a simple theory to find explicit expressions for coefficients γ1 and
γ2 . Using these and few next expansion coefficients for ǫ(g) we construct a Padé-approximation
for this function which may be used for a wide range of impact velocities where the concept of
the viscoelastic collision is valid. The obtained expression reproduces quite accurately the existing
experimental dependence ǫ(g) for ice particles.

I. INTRODUCTION the radii and the positions of the spheres), Y and ν are
respectively the Young modulus and the Poisson ratio of
The change of relative velocity of inelastically colliding the particle material, Reff ≡ R1 R2 /(R1 + R2 ), and the
particles can be characterized by the coefficient of resti- dissipative parameter A reads [25,26]
tution ǫ. The normal component of the relative velocity
1 (3η2 − η1 )2 (1 − ν 2 )(1 − 2ν)
 
after a collision g ′ = ~v12′
· ~e follows from that before the A= . (4)
collision g = ~v12 · ~e via 3 (3η2 + 2η1 ) Y ν2
The viscous constants η1 , η2 relate the dissipative stress
g ′ = −ǫg (1) tensor to the deformation rate tensor  [25,26,28].
 The
where ~v1 , ~v2 and ~v1′ , ~v2′ are respectively the veloci- ˙
same functional dependence of Fdiss ξ, ξ has been ob-
ties before and after the collision, while the unit vector tained in [29–31] using a different approach.
~e ≡ ~r12 / |~r12 | gives the direction of the inter-particle vec- The equation of motion for inelastically colliding
tor ~r12 = ~r1 − ~r2 at the instant of the collision. spheres reads, therefore,
From experiments as well as from theory it is well  
known that the coefficient of normal restitution ǫ is not a ¨ ρ 3/2 3 p
ξ + eff ξ + A ξ ξ̇ = 0 , (5)
constant but it depends sensitively on the impact velocity m 2
[1–11]. Although most of the results in the field of gran- with ξ(0) = 0, ξ̇(0) = g ,
ular gases have been derived neglecting this dependence
but using a velocity-independent coefficient of restitution and with meff ≡ m1 m2 /(m1 + m2 ) (m1 , m2 are the
(e.g. [12–18]) it has been shown that the impact-velocity masses of the colliding particles). To obtain the depen-
dependence of the coefficient of restitution has serious dence of the restitution coefficient on the impact veloc-
consequences for various problems in granular gas dy- ity for 3D-spheres the equation of motion (5) was solved
namics [19–24]. numerically [10,24–26] and analytically [32], where the
The equation of motion for inelastically colliding 3D- velocity-dependent restitution coefficient has been ob-
spheres has been addressed in [24–26], where the Hertz tained as a series in powers of g 1/5 :
contact law [27]  
3 ρ 2/5 1/5
√ ǫ = 1 − C1 A g (6)
2Y 2 meff
Fel = ρ ξ 3/2 , ρ≡ 2
Reff (2)
3(1 − ν ) 
3
2 
ρ 4/5 2/5
+ C2 A g ∓ ···
for the elastic inter-particle force has been extended 2 meff
to account for the viscoelasticity of the material which
The first coefficients C1 = 1.15344 and C2 = 0.79826
causes the dissipative part of the force
were evaluated analytically and then confirmed by nu-
3 merical simulations [32].
A ρ ξ ξ˙ .
p
Fdiss = (3) Although in [32] a general method of derivation of all
2
coefficients of the expansion (6) has been proposed, to ob-
Here, ξ is the compression of the particles during the col- tain these, extensive calculations have to be performed.
lision ξ = R1 + R2 − |~r1 − ~r2 | (R1 , R2 and ~r1 , ~r2 are This approach does not provide closed-form expressions

1
for the coefficients, but rather gives them in terms of con- In the last Eq. (10) we supplemented the pre-collisional
vergent series which are to be evaluated up to the desired initial conditions at τ = 0 with the after-collisional con-
precision. ditions at τ = τc (τ is the dimensionless time and τc is
In the present study we show that a dimension anal- the dimensionless duration of the collision). These follow
ysis allows to obtain the functional form of the ǫ(g)- just from the definition of the restitution coefficient. We
dependence for the elastic and dissipative forces. Within point out that all dependence on the initial impact ve-
the framework of this analysis we reproduce the depen- locity on any quantity of the problem, including ǫ (this
dence (6) up to numerical values of coefficients Ck . A is just the dimensionless after-collisional velocity) comes
similar approach has been used by Tanaka et al. [33] to only through the constant δ, which reads
prove that the constant coefficient of restitution is not
1+γ
consistent with physical reality (see also [10,34]). We also 
1+α
 1+α
2(γ−α)
develop a simple approximative theory, which gives a con- δ(g) = D2 g 1+α +β . (11)
tinuum fraction representation for ǫ(g) and a closed-form 2 D1
expressions for C1 and C2 with the same numerical values
as above. Using then coefficients C1 , . . . C4 (with C3 and Hence, ǫ(g) = ǫ (δ(g)). A similar result for ǫ → 0, β = 1
C4 evaluated in Appendix in accordance with the general and α = 3/2 has been obtained in [36].
scheme of Ref. [32]) we construct a Padé-approximation, If we assume that the restitution coefficient does not
which reproduces fairly well the experimental data for depend on the impact velocity g then follows
colliding ice-particles [5].
2 (γ − α) + β (1 + α) = 0 . (12)

II. THE DIMENSIONAL ANALYSIS For a linear dependence of the dissipative force on the ve-
locity, i.e. for β = 1 (this seems to be the most realistic
To perform the general dimensional analysis we adopt for small ξ)˙ one obtains a constant restitution coefficient
the following form for the elastic and dissipative forces: for
Fel = meff D1 ξ α
Fdiss = meff D2 ξ γ ξ˙β . • the linear elastic force, Fel ∼ ξ, i.e. α = 1. The
condition (12) implies γ = 0, and thus the linear
˙ follows from
This general form (at least for small ξ and ξ) ˙
dissipative force Fdiss ∼ ξ.
the fact that both elastic and dissipative forces vanish at
ξ = 0 and ξ˙ = 0, respectively. With these notations the
• the Hertz law for 3D-spheres (2) α = 3/2, there-
equation of motion for colliding particles reads
fore γ = 41 and Fdiss ∼ ξ˙ ξ 1/4 provides a constant
ξ̈ + D1 ξ α + D2 ξ γ ξ̇ β = 0 , (7) restitution coefficient.
with ξ(0) = 0, ξ(0) ˙ = g,
We now ask the question: What kind of ǫ(g) depen-
where g has already been introduced. Now we choose as dence corresponds to the forces which act during colli-
the characteristic length ξ0 of the problem the maximal sions of viscoelastic particles? It may be generally shown
compression for the elastic case. It may be found from [25,26,37] that the relation
the condition that the initial kinetic energy meff g 2 /2 [35]
equals the maximal elastic energy meff D1 ξ0 α+1 /(α + 1), ∂
which yields Fdiss = A ξ̇ Fel (ξ) (13)
∂ξ
  1
α + 1 1+α 1+α 2
ξ0 = g , (8) between the dissipative and elastic forces with the dissi-
2 D1
pative constant A given in eq. (4) holds, provided three
Choosing then the characteristic time of the problem as conditions are met [38]:
τ0 = ξ0 /g, we construct new dimensionless variables
2 (i) The elastic part of the stress tensor depends lin-
˙ ¨ˆ g
ξ̂ = ξ/ξ0 , ξˆ = ξ/g
˙ , ξ = ξ̈ (9) early on the deformation tensor [28].
ξ0
and recast the equation of motion into dimensionless (ii) The dissipative part of the stress tensor depends
form: linearly on the deformation rate tensor [28].
¨ ˙ 1 + α ˆα
ξ̂ + δ(g) ξˆγ ξ̂ β + ξ = 0 with (10)
2 (iii) The conditions of quasistatic motion are provided,
˙ˆ i.e. g ≪ c, τvis ≪ τc [25,26] (here c is the speed of
ξ̂(0) = 0, ξ(0) =1
sound in the material of particles, τvis is relaxation
˙
ξ̂(τc ) = 0, ξ̂(τc ) = −ǫ time of viscous processes in its bulk).

2
q
From this follows that β = 1, γ = α − 1, and thus ˙ˆ ˆ ˆ ξˆ0 )5/2
ξ( ξ) ≈ 1 − (ξ/ (20)
the constant restitution coefficient may be observed only
for collisions of cubic particles with surfaces normal to ˙
the direction of collision. We wish to emphasize that this which also gives vanishing velocity ξˆ at the turning point
conclusion comes from the general analysis of viscoelastic ξ̂0 . Integration in Eq. (18) may be performed yielding
collisions.
Let us discuss now collisions between spheres with elas- 1 ˆ5/2 1
− = −δ d ξˆ0
3/2
ξ (21)
tic and dissipative forces as given by (2) and (3), respec- 2 0 2
tively. For these we have meff D1 = ρ, α = 3/2 and
where we take into account that E(ξˆ0 ) = 21 ξ̂0 , E(0) =
5/2
meff D2 = 32 A ρ, γ = 1/2, β = 1 which yields the func-
tional dependence for δ(g) and ǫ(g) respectively: 1˙ 1
2 ξ̂(0) = 2 and introduce a constant
 3/5 
3 5 ρ 2/5 1/5 √
1
π Γ 53
Z 
p
δ= A g (14) d≡ 1/2
x 1−x5/2 =  . (22)
2 4 meff 5 Γ 21
   0 10
ρ 2/5 1/5
ǫ=ǫ A g (15) Consider now the inverse collision, which is defined as a
meff
collision which starts with velocity ǫ g and ends with ve-
(skipping the prefactor of δ(g) in the last equation) in locity g. According to the concept of the inverse collision
accordance with (6) as found previously. introduced in [32] (which is a useful auxiliary model),
it is characterized by a negative damping (the energy
“is pumped” into the system during the collision). The
III. THE RESTITUTION COEFFICIENT FOR maximal compression ξˆ0 is the same in both collisions,
SPHERES the direct and the inverse.
Rescaling equation of motion for the inverse collision
˙
d
Using dt = ξˆddξ̂ it is convenient to write the equation in the very same way as for the direct collision yields
of motion for a collision in the form
ˆ
dE(ξ) ˙

d 1 ˙ 2 1 ˆ5/2
 ˆ
dE(ξ) = +δ ξˆξˆ1/2
˙ ˆ

ξ̂ + ξ = −δ ξˆξˆ1/2 =
dξˆ 2 2 dξˆ ˙ˆ
ξ̂(0) = 0 , ξ(0) = ǫ. (23)
ˆ = 0; ˙ˆ
ξ(0) ξ(0) =1 (16)
This suggests the following approximative relation for
where we introduce the mechanical energy ˙ ˆ
ξ̂(ξ) during the inverse collision:
1 ˙ 2 1 5/2
E≡ ξ̂ + ξ̂ . (17)
q
˙ˆ ˆ ˆ ξˆ0 )5/2 ,
2 2 ξ(ξ) ≈ ǫ 1 − (ξ/ (24)
To find the energy losses in the first stage of the colli- with the additional prefactor ǫ, which is the initial veloc-
sion, which starts with zero compression and ends in the ity in the inverse collision.
turning point with maximal compression ξˆ0 Integration of the energy gain for the first stage of the
ξ̂0 ξ̂0 inverse collision (which equals up to its sign the energy
dE ˆ
Z Z
˙ loss in the second stage of the direct collision [32]) may be
dξ = −δ ξ̂ ξ̂ 1/2 dξˆ (18)
0 dξˆ 0 performed just in the same way as for the direct collision,
yielding the result
one needs to know the dependence of the compression
˙ 1 5/2 ǫ2
rate ξ̂ as a function of the compression ξ̂.
= +ǫ δ d ξˆ0 ,
3/2
ξ̂ − (25)
For the case of elastic collisions, the maximal compres- 2 0 2
sion is ξ̂0 = 1, according to the definition of our dimen-
where we again use E(ξˆ0 ) = 21 ξˆ0 and E(0) = 12 ǫ2 . Mul-
5/2
˙ ˆ
sionless variables. Hence, the dependence ξ̂(ξ) follows
from the conservation of energy: tiplying Eq. (21) by ǫ and summing it up with Eq. (25)
we obtain a simple approximative relation between the
restitution coefficient and the (dimensionless) maximal
q
˙ ˆ
ξ̂(ξ) = 1 − ξˆ5/2 . (19) compression:
˙
The velocity ξˆ vanishes at the turning point ξ̂ = 1. For ǫ = ξˆ0
5/2
(26)
inelastic collisions the maximal compression ξ̂0 is smaller
than 1, therefore, one can write an approximation rela- Substituting this into Eq. (21) we arrive at an equation
tion for the inelastic case: for the restitution coefficient

3
ǫ + 2δ d ǫ3/5 = 1 . (27) from the correct ones, see below). Comparing (32) with
(6) we conclude that our simple approximative theory re-
The formal solution to this equation may be written as a produces exactly the coefficients C1 and C2 , which were
continuum fraction (which does not diverge in the limit found before using extensive analysis [32].
g → ∞): We also performed rigorous but elaborated calculations
 2/5 according to the general scheme of [32] to find the exact
ǫ−1 = 1 + 2δ d 1 + 2δd (1 + · · ·)2/5 · · · . (28) coefficients (details are given in the Appendix)

Another way of representation of the restitution coef- a3 = 0.315119 , a4 = 0.161167 , (33)


ficient ǫ is a series expansion in terms of δ. For practi-
cal applications it is convenient to return to dimensional or, respectively,
units. We define the characteristic velocity g ∗ such that
C3 = −0.483582 , C4 = 0.285279 . (34)
 1/5
1 g
δ≡ , (29) Hence, we observe that while the first two coefficients
2d g∗ a1 = 1 and a2 = 3/5 are correctly obtained from the ap-
proximative theory, the next approximated coefficients
with d beeing defined in Eq. (22). Using the definition
a3 , a4 differ from the exact ones.
(14) together with Eq. (22) we find for the characteristic
For practical applications such as Molecular Dynam-
velocity
ics simulations, however, the expansion (32) is of limited
√   value, due to its divergence for high impact velocities,
∗ −1/5 π Γ (3/5) 3 ρ 2/5
(g ) = 1/5 2/5 A . (30) g → ∞. According to the velocity distribution function
2 5 Γ (21/10) 2 meff
there is a certain probability that the relative velocity g
Evaluating the numerical prefactor finally yields of colliding particles exceeds the limit of applicability of
  (32). Therefore, we use the obtained coefficients to con-
−1/5 3 ρ 2/5 struct a Padé-approximation for ǫ(g) which reveals the
(g ∗ ) = 1.15344 A . (31)
2 meff correct limits of the boundary conditions, ǫ(0) = 1 and
ǫ(∞) = 0. Since the dependence ǫ(g) is expected to be a
Note that the numerical constant 1.15344 has to be equal smooth, monotonically decreasing function, we choose a
to C1 in Eq. (6). “1-4” Padé-approximation
With this new notation the restitution coefficient reads
  51
  51
g
  25
g
  35
g 1 + d1 gg∗
ǫ = 1 − a1 + a2 − a3 ǫ=   54 .
g∗ g∗ g∗   51   52   35
 4/5 1 + d2 gg∗ + d3 gg∗ + d4 gg∗ + d5 gg∗
g
+ a4 + ··· , (32)
g∗ (35)

with a1 = 1, a2 = 3/5 (which are exact values), a3 = Standard analysis yields the coefficients dk in terms of
6/25 = 0.24, a4 = 7/125 = 0.056, · · · (which deviate the coefficients ak [39] (see Table I)

d0 = a4 − 2a3 − a22 + 3a2 − 1


d1 = [1 − a2 + a3 − 2a4 + (a2 − 1)(3a2 − 2a3 )] d−1 0 = 2.5839
−1
d2 = [(a
 3 − a22 )(1 − 2a2 ) − a4 ] d0  −1 = 3.5839
d3 = a3 + a2 (a2 − 1) − a4 (a2 + 1) d0  = 2.9839
d4 = a4 (a3 − 1) + (a3 − a2 )(a22 − 2a3 ) d−1
0 = 1.1487
= 2(a3 − a2 )(a4 − a2 a3 ) − (a4 − a22 )2 − a3 (a3 − a22 ) d−1

d5 0 = 0.3265
TABLE I. The coefficients of the Padé formula (35) as derived from the coefficients ak .

Using the characteristic velocity g ∗ = 0.32 cm/s for ice trapolation expression, ǫ = 0.32/g 0.234 used in [5] has an
as a fitting parameter we could reproduce fairly well the unphysical divergence at g → 0 and does not imply the
experimental dependence of the restitution coefficient of fail of the theory for this region. The scattering of the ex-
ice as a function of the impact velocity g in the whole perimental data presented in [5] is large for small impact
range of g (Fig. 1). The discrepancy with the experi- velocity according to experimental complications, hence
mental data at small g follows from the fact that the ex- the fit formula of [5] cannot be expected to be accurate

4
enough for too small velocities. Moreover in the region V. APPENDIX
of very small velocity possibly other than viscoelastic in-
teractions influence the collision behavior, e.g. adhesion.
The general method of derivation of the expansion co-
0.6 efficients Ck has been given in [32]. Here, we briefly
sketch the main lines of derivation, and provide some
0.55
1 - Bridges et al. (1984)
details for the particular case of C3 and C4 . Since the
0.5 1 2 - Pade approximant method of derivation is based on the collection of terms
Normal Restitution Coefficient

with different dependence on the initial velocity g, it is


0.45
convenient to use a scaling, somewhat different from that
0.4 used before for the dimensional analysis. Namely, we
2
0.35
rescale the time as t′ = (ρ/meff )2/5 g 1/5 t and the length
as x = (ρ/meff )2/5 ξ to recast Eq. (5) into the form [40]
0.3

0.25 x′′ + αg −1/5 x′ x + g −2/5 x3/2 = 0 , (36)
0.2
0.5 1 1.5 2 2.5 3 3.5
Normal Impact Velocity (cm/s)
4 4.5 5
with α ≡ 23 A(ρ/meff )2/5 and using all the notations intro-
FIG. 1. Dependence of the normal restitution coef- duced previously. The initial conditions for the rescaled
ficient on the impact velocity for ice particles. The Eq. (36) now read x(0) = 0 and x′ (0) = g 4/5 . For sim-
solid line – experimental data of [5], dashed line – the plicity of notations we will keep in what follows t for the
Padé-approximation (35) with the constants given in rescaled time. As it was shown √ in [32], the trajectory
the Table and with the characteristic velocity for ice may be expanded in terms of t as
g ∗ = 0.32 cm/s.
x(t) = b1 t1/2 + b2 t + b3 t3/2 + b4 t2 (37)
IV. CONCLUSION + b5 t5/2 + b6 t3 + b7 t7/2 + . . . .

We developed a dimensional analysis for the inelas- Clearly, both, b1 and b3 should be zero to avoid diver-
tic collision of spherical particles. We could show that gence of velocity and acceleration at t = 0. At the same
for 3D-spheres the functional form for ǫ(g) agrees with time b2 = g 4/5 and b4 = 0 due to the equation of motion
that derived previously [32] using a much more compli- at vanishing compression. This yields
cated approach. Using a simple approximative theory we
found a continuum-fraction representation for ǫ(g) and x(t) = g 4/5 t + b5 t5/2 + b6 t3 + b7 t7/2 + . . . . (38)
obtained explicit expressions for the coefficients of the se-
ries expansion of the restitution coefficient in terms of the From (38) one obtains x′ (t) and x′′ (t) which are to be
impact velocity. The first two coefficients in this series substituted
√ into the equation of motion (36). One also
coincide with that found previously by numerical evalu- needs
√ x and x3/2 ; the expansions for these in terms of
ation. We report also a few next coefficients which we t read
have derived within the general approach of a previous
study [32]. Using the first four coefficients of this series √ b5 b6 b7
expansion we constructed a Padé-approximation for ǫ(g). x = g 2/5 t1/2 + 2/5 t2 + 2/5 t5/2 + 2/5 t3 + . . . (39)
2g 2g 2g
It reproduces fairly well the experimental data for collid-
ing ice particles. The obtained relation for the restitution and
coefficient may be used for a wide range of the impact ve-
locities, provided that the energy loss during a collision 3 3
is attributed to viscoelasticity and that all the other dis- x3/2 = g 6/5 t3/2 + g 2/5 b5 t3 + g 2/5 b6 t7/2 + . . . . (40)
2 2
sipative processes (plastic deformation, fragmentation of
particles) may be ignored. √
Inserting the expansions for x′ (t), x′′ (t), x and x3/2
into (36) and collecting the orders of t we obtain
ACKNOWLEDGMENTS    
15 1/5 1/2 35
0= b5 + αg t + 6 b6 t + b7 + 1 t3/2
4 4
The authors want to thank W. Ebeling and L.  

1/5

2 63 7 1/5
Schimansky-Geier for discussion. The work was sup- + 12 b8 + 3αg b5 t + b9 + αg b6 t5/2 . (41)
ported by Deutsche Forschungsgemeinschaft through 4 2
grant Po 472/3-2 and by FONDECYT Chile, through
project 02960021. This suggests the coefficients:

5
4 x′max (tmax ) = 0 (50)
b5 = − αg 1/5 (42)
15
b6 = 0 (43) holds at this point. With the above expression for the
4 trajectory Eqs. (48,49), the last Eq. (50) is an equation
b7 = − (44)
35 to determine tmax , which may be then used to find xmax .
1 2 2/5 This√equation, however is a high-order algebraic equation
b8 = α g (45) for tmax , which is not generally solvable. On the other
15
b9 = 0 , (46) hand, for the undamped collision tmax equals one-half of
the collision duration t0c and both quantities of interest
so that the solution for the trajectory finally reads are known [28]:
4 4
x(t) = g 4/5 t − αgt5/2 − g 4/5 t7/2
 3/5
Γ 52 Γ 12
 
15 35 t0c 4
t0max
= = 9

1 2 6/5 4 2 5 2 Γ 10
+ α g t + ... . (47)
15  0   2/5
t 5
x0 c = (51)
In order to get the higher orders, which is conception- 2 4
ally simple but requires extensive calculus, we wrote a
program [41], which by formula manipulations performs For a viscoelastic collision tmax certainly differs from
exactly the steps we described above and which is able t0c /2, so that tmax = t0c /2 + δt. If the dissipation param-
to find the trajectory up to any desired order. eter α is not large, the deviation δtis presumably small,
Generally, it is convenient to write the solution as a 0

′ ′ tc
therefore we expand x (tmax ) = x 2 + δt in terms of
series
  δt:
x(t) = g 4/5 x0 (t) + αg 1/5 x1 (t) + α2 g 2/5 x2 (t) + . . . .
g −4/5 x′ (tmax ) = (52)
(48)   0
t
 0
tc 2
 0
δt ′′′ tc

= x′0 c + δtx′′0 + x + ...
Here x0 (t) is a “zero-order” trajectory, which refers 2 2 2 0 2
  0  0
to the case of undamped collision, the “first-order” tra- δt2 ′′′ t0c
  
1/5 ′ tc ′′ tc
jectory, x1 (t), accounts for damping in linear (with re- +αg x1 + δtx1 + x + ...
2 2 2 1 2
spect to α) approximation, the “second-order” trajectory,   0  0 
t tc
x2 (t), corresponds to the next approximation ∼ α2 , etc. +α2 g 2/5 x′2 c + δtx′′2 + ...
Here we give our results for these “n-order” trajectories 2 2
  0 
up to n = 3, obtained using the above mentioned pro- t
gram up to the order t11 : +α3 g 3/5 x′3 c + . . . + . . . = 0 ,
2
4 7/2 1 6 22 17/2 52
x0 = t − t + t − t + t11 where we use representation (48) for the trajectory. The
35 175 104125 8017625
4 3 713 15/2 61216 10 deviation δt, vanishes at α = 0 and, thus, suggests the
x1 = − t5/2 + t5 − t + t expansion in terms of α:
15 70 238875 42639187
1 4 937 13/2 871 9
x2 = t − t + t (49) δt = τ1 α + τ2 α2 + τ3 α3 . . . (53)
15 75075 808500
38 11/2 43943 8 1184627 21/2
x3 = − t + t − t Substituting δt, given by Eq. (53), into (52) and collect-
2475 13513500 3594591000 ing terms of the same order of α yields
To proceed we need to find the maximal compression
xmax , which is reached at time tmax . The point of max- Y0 + αY1 + α2 Y2 + α3 Y3 + · · · = 0 , (54)
imal compression is a turning point of the trajectory,
where the velocity is zero. Therefore the condition with the abbreviations
 0
t
Y0 = x′0 c (55)
2
 0  0
tc t
′′
Y1 = τ1 x0 + g x1 c
1/5 ′
2 2
 0 2
 0  0  0
tc τ1 ′′′ tc tc t
′′
Y2 = τ2 x0 + x0 1/5
+ g τ1 x1 ′′
+ g x2 c
2/5 ′
2 2 2 2 2

6

! ′
!
t0c t0c τ3 t0c t0c 2
t0c t0c t0c
         
1/5 τ1
Y3 = τ3 x′′0 + τ1 τ2 x′′′
0 + 1 x′′′′ +g 1/5
τ2 x′′1 +g x′′′ +g 2/5
τ1 x′′2 + g 3/5 x′3 .
2 2 6 0 2 2 2 1
2 2 2

The conditions Yk = 0 for k = 0, . . . 3 together with Eq. (55) allows to express the constants τ1 , τ2 , τ3 , etc., in terms
of functions x1 (t), x2 (t), x3 (t), etc., and their time derivatives taken at time t0c /2 :
 0
t
x′1 2c
1/5
τ1 = −g   (56)
t0
x′′0 2c
 0  0  0  0  0
tc tc t t t

x′2
1 2 x′′′
0 2 x′1 2c x′′1 2c x′2 2c
2/5 
τ2 = g −  0 +  0 −  0 
tc ′′2 tc ′′ tc
2 x′′3
0 2 x 0 2 x0 2

We do not  write the expression for τ3 , since due to the special properties of the problem, i.e. due to the fact that
x′0 t0c /2 = 0 the value τ3 is not needed for calculation of ǫ up to fourth order of α. The functions x1 (t), x2 (t), x3 (t)
are known and given by Eqs. (49), so that the constants τ1 and τ2 may be found explicitly.
Writing the maximal compression as
  0   0   0   0 
t t t t
xmax = g 4/5 x0 c + δt + αg 1/5 x1 c + δt + αg 2/5 x2 c + δt + α3 g 3/5 x3 c + δt , (57)
2 2 2 2

and expanding this in terms of δt, using then representation of δt as δt = ατ1 + α2 τ2 + · · ·, with τ1 , τ2 from (56) and
collecting terms of the same order of α we obtain
 
xmax = g 4/5 y0 + αg 1/5 y1 + α2 g 2/5 y2 + α3 g 3/5 y3 , (58)

where y0 , . . . y3 are pure numbers:


 0
t
y0 = x0 c = 1.093362 (59)
2
 0
t
y1 = x1 c = −0.504455 (60)
2
 0 
′2 tc

 0
t 1 x 1 2
y2 = x2 c −    = 0.260542 (61)
2 2 x′′ t0c
0 2
 0  0  0  0  0
t
 0  x′ c x′ tc tc t tc
 
t 1 2 2 2 1 x′2
1 2 x′′1 2c x′3
1 2 x′′′
0
y3 = x3 c −  0 +  0 − ′′3 0  = −0.136769 (62)
2 x′′ c
t 2 t
x′′ c x0 (tc /2)
0 2 0 2

and where we use expressions (49) for x1 (t), x2 (t) and x3 (t).
To calculate the coefficient of restitution one has to use the concept of inverse collision, as was introduced in [32]
and discussed in previous chapters of the present study. One obtains the solution of this inverse collision by replacing
g → ǫg for the initial velocity and α → −α for the dissipative coefficient. In particular, this applies to the maximal
compression of the inverse collision xinv
max = xmax (g → ǫg, α → −α). For consistency one has to require the maximum
compressions for direct and inverse collision to be equal, i.e.

xinv
max = xmax , (63)

or using (58),
4 4
 1 1 2 2 3 3
 4
 1 2 3

ǫ 5 g 5 y0 − αǫ 5 g 5 y1 + α2 ǫ 5 g 5 y2 − α3 ǫ 5 g 5 y3 + · · · = g 5 y0 + αg 5 y1 + α2 g 5 y2 + α3 g 5 y3 + · · · . (64)

Eq. (64) is in fact an algebraic equation for ǫ1/5 , which may not be generally solved. For this reason we write ǫ as
an expansion of αg 1/5 , which is the only combination in which both parameters appear

7
 2  3  4
ǫ = 1 + C1 αg 1/5 + C2 αg 1/5 + C3 αg 1/5 + C4 αg 1/5 + . . . (65)

and substitute (65) into (64). Collecting orders we find


    
4 4 2
− y0 C1 + 2y1 αg 1/5 + − C2 + C12 y0 + y1 C1 α2 g 2/5 + (66)
5 5 25
  
4 4 4 3 6
− C3 + C1 C2 − C1 y0 + y1 C2 − y2 C1 + 2y3 α3 g 3/5 +
5 25 125 5
    
4 2 12 11 4 6 3 7
C22 + 2C1 C3 − C12 C2 + C1 y0 + y1 C3 + − C2 − C12 y2 + y3 C1 α4 g 4/5 = 0 .

− C4 +
5 25 125 625 5 25 5

The last Eq. (66) yields the final result for the coeffi- [7] C. V. Raman, Phys. Rev. 12, 442 (1918).
cients: [8] S. F. Foerster, M. Y. Louge, H. Chang, and Kh. Allia,
Phys. Fluids 6, 1108 (1994).
5 y1 [9] S. Hatzes, F. G. Bridges, and D. N. C. Lin, Mon. Not.
C1 = = −1.153449 (67)
2 y2 R. Astr. Soc. 231, 1191 (1988).
 2
15 y1 3 [10] S. Luding, E. Clément, A. Blumen, J. Rajchenbach, and
C2 = = C12 = 0.798267 J. Duran, Phys. Rev. E 50, 4113 (1994).
4 y2 5 [11] S. Luding, E. Clément, J. Rajchenbach, and J. Duran,
 3
95 y1 15 y2 y1 5 y3 Europhys. Lett. 36, 247 (1996).
C3 = − + = −0.483582 [12] I. Goldhirsch and G. Zanetti, Phys. Rev. Lett. 70, 1619
16 y2 4 y02 2 y0
 4 (1993).
315 y1 105 y2 y12 35 y3 y1 [13] Y. Du, H. Li, and L. P. Kadanoff, Phys. Rev. Lett. 74,
C4 = − + = 0.285279
32 y2 8 y03 4 y02 1268 (1995).
[14] T. Zhou and L. P. Kadanoff, Phys. Rev. E 54, 623 (1996).
Using (g ∗ )−1/5 = C1 α, we obtain for coefficients ak in [15] T. P. C. van Noije, M. H. Ernst, and R. Brito, Phys. Rev.
expansion (32): E 57, R4891 (1998).
[16] J. A. C. Orza, R. Brito, T. P. C. van Noije, and M. H.
a1 =1 (68) Ernst, Int. J. Mod. Phys. C 8, 953 (1997).
[17] T. C. P. van Noije, M. H. Ernst, R. Brito, and J. A. G.
a2 = C2 /C12 = 3/5 (69) Orza, Phys. Rev. Lett. 79, 411 (1997).
a3 = C3 /C13 = 0.315119 (70) [18] J. J. Brey and D. Cubero, Phys. Rev. E 57, 2019 (1998).
a4 = C4 /C14 = 0.161167 . (71) [19] H. Salo, J. Lukkari, and J. Hanninen, Earth, Moon, and
Planets 43, 33 (1988).
Note that although the general method given in Ap- [20] K. A. Hämeen-Anttila and J. Lukkari, Astrophys. Space.
pendix allows to evaluate up to a desired precision all, in Sci. 71, 475 (1980).
principle, coefficients Ck , it does not provide the closed [21] F. Spahn, U. Schwarz, and J. Kurths, Phys. Rev. Lett.
form expression for C1 as the simple approximate ap- 78, 1596 (1997).
[22] T. Pöschel and T. Schwager, Phys. Rev. Lett. 80, 5708
proach given in the main text does.
(1998).
[23] N. V. Brilliantov and T. Pöschel, cond-mat/9803387.
[24] F. Spahn, J.-M. Hertzsch, and N. V. Brilliantov, Chaos,
Solitons and Fractals, 5, 1945 (1995).
[25] N. V. Brilliantov, F. Spahn, J.-M. Hertzsch, and T.
Pöschel, Phys. Rev. E 53, 5382 (1996).
[26] J.-M. Hertzsch, F. Spahn, and N. V. Brilliantov,
[1] R. M. Brach, J. Appl. Mech. 56, 133 (1989). J.Phys.II (France), 5, 1725 (1995).
[2] S. Wall, W. John, H. C. Wang, and S. L. Goren, Aerosol
[27] H. Hertz, J. f. reine u. angewandte Math. 92, 156 (1882).
Sci. Tech. 12, 926 (1990).
[28] L. D. Landau and E. M. Lifschitz, Theory of Elasticity,
[3] W. Goldsmith, Impact: The Theory and Physical Be-
Oxford University Press (Oxford, 1965).
haviour of Colliding Solids, Edward Arnold (London, [29] G. Kuwabara and K. Kono, Jpn. J. Appl. Phys. 26, 1230
1960).
(1987).
[4] P. F. Luckham, Pow. Tech. 58, 75 (1989).
[30] W. A. M. Morgado and I. Oppenheim, Phys. Rev. E 55,
[5] F. G. Bridges, A. Hatzes, and D. N. C. Lin, Nature 309, 1940 (1997).
333 (1984).
[31] W. A. M. Morgado and I. Oppenheim, Physica A 246,
[6] E. Hodgkinson, Report of the 4th Meeting of the British
547 (1997).
Association for the Advancement of Science, London [32] T. Schwager and T. Pöschel, Phys. Rev. E 57, 650 (1998).
(1835).

8
[33] T. Tanaka, Kiron 49-441B, 1020 (1983). [39] Obviously, the coefficients of the Padé-approximation
[34] Y. Taguchi J. Phys. (Paris) 2, 2103 (1992). may be chosen up to an arbitrary factor to multiply nu-
[35] As usual for collision (e.g. [3,28]) the two-particle prob- merator and denominator; we chose it to have unity as a
lem is reduced to scattering problem of a single-particle leading term for both of these.
with an effective mass meff . [40] Note that, in difference to the calculations in the main
[36] E. Falcon, C. Laroche, S. Fauve, and C. Coste, Eur. Phys. part of the article the quantities x, x′ and x′′ do have
J. B 3, 45 (1998). units, namely (m/sec)4/5 . The rescaled time is dimen-
[37] Derivation of the dissipative force given in [25,26] for col- sionless. The purpose of this scaling was only to simplify
liding spheres may be straightforwardly generalized to the dependence of the problem on the material parame-
obtain the relation (13) (or (A17) in [25,26]) for colliding ters, it was necessary to keep the explicit dependence of
bodies of any shape, provided that displacement field in the problem on the initial velocity.
the bulk of the material of bodies in contact is a one- [41] The maple-program is available at
valued function of the compression (see also [38]). http://www.summa.physik.hu-berlin.de/
[38] N. V. Brilliantov and T.Pöschel, in preparation. ∼ kies/papers/DimAnalysis/epsilon simple.txt

Вам также может понравиться