PHYSICAL REVIEW E 85, 041306 (2012)
Negative coefficient of normal restitution
Patric M¨ uller, Dominik Krengel, and Thorsten P¨ oschel
Institute for Multiscale Simulation, Universit ¨ at Erlangen-N¨ urnberg, N¨ agelsbachstraße 49b, 91052 Erlangen, Germany
(Received 15 February 2012; published 27 April 2012)
This paper shows that negative coefficients of normal restitution occur inevitably when the interaction force
between colliding particles is finite. We derive an explicit criterion showing that for any set of material properties
there is always a collision geometry leading to negative restitution coefficients. While from a phenomenological
point of view, negative coefficients of normal restitution appear rather artificial, this phenomenon is generic and
implies an important overlooked limitation of the widely used hard sphere model. The criterion is explicitly
applied to two paradigmatic situations: for the linear dashpot model and for viscoelastic particles. In addition,
we show that for frictional particles the phenomenon is less pronounced than for smooth spheres.
DOI: 10.1103/PhysRevE.85.041306 PACS number(s): 45.70.−n, 45.50.Tn
I. INTRODUCTION
Both, kinetic theory of granular matter, based on the
Boltzmann equation [1–3], as well as highly efficient event-
driven molecular dynamics (eMD) simulation of granular
matter [4–6] are based on the hard sphere model of particle
collisions. Hard sphere collisions are characterized by δ-
shaped interaction forces. Therefore, in a collision the particles
instantaneously exchange momentum, while their positions
stay invariant. Due to the instantaneous character of the
collisions, the dynamics of any finite hard sphere system is
represented by a sequence of binary collisions (events), leading
to the main concept of eMD. As only momentum is transferred
during a collision, each event is characterized by only two
scalar values: The coefficient of normal restitution ε
n
relating
the post- and precollisional normal component of the particles’
relative velocity and the corresponding coefficient of tangential
restitution for the tangential component. For colliding hard
spheres located at r
1
and r
2
, traveling at velocities
˙
r
1
and
˙
r
2
,
the coefficient of normal restitution ε
HS
n
is, thus, defined by
(
˙
r
′
1
−
˙
r
′
2
) · ˆ e
0
r
=−ε
HS
n
(
˙
r
0
1
−
˙
r
0
2
)
· ˆ e
0
r
, (1)
where for each quantity X, the symbol X
0
denotes the value of
X at the beginning of the impact at time t
0
and X
′
is the value
at time t
0
+ τ , when the collision terminates. Equation (1)
addresses hard spheres implying instantaneous collisions, τ →
0. Note that the unit vector ˆ e
r
≡ ( r
2
− r
1
)/ | r
2
− r
1
| remains
unchanged during a collision, ˆ e
′
r
= ˆ e
0
r
, due to the instantaneous
character of hard sphere collisions. Therefore, ˆ e
0
r
appears on
both sides of Eq. (1). For soft sphere collisions characterized
by finite interaction forces and contact durations τ , this may
be invalid: Oblique impacts may lead to variations of the unit
vector ˆ e
′
r
= ˆ e
0
r
. Consequently, for soft spheres the coefficient
of normal restitution ε
n
is defined by
(
˙
r
′
1
−
˙
r
′
2
)
· ˆ e
′
r
=−ε
n
(
˙
r
0
1
−
˙
r
0
2
)
· ˆ e
0
r
, (2)
which reduces to Eq. (1) in the limit τ → 0. The variation of
the unit vector ˆ e
r
during a collision may be expressed by the
angle
ϕ
′
≡ cos
−1
(
ˆ e
0
r
· ˆ e
′
r
)
. (3)
For both, kinetic theory based on the Boltzmann equation as
well as eMD, it is essential to assume that ϕ
′
is negligible.
Then, Eq. (1) allows for the computation of the post-collisional
particle velocities from the impact velocities, that is, the
collision rule needed in eMD simulations and also the
Jacobian ∂ ( v
1
, v
2
) /∂
(
v
′
1
, v
′
2
)
needed for the integration of
the Boltzmann equation. It was shown, however, that the
assumption ϕ
′
≈ 0 is not always justified [7].
In textbooks, the coefficient of normal restitution is fre-
quently assumed to be a material constant [8], 0 ε
n
1.
While from experimental (e.g., [9–13]) and theoretical (e.g.,
[14–18]) studies it is known that the coefficient of normal
restitution depends on the impact velocity or is a fluctuating
quantity (e.g., [19]) it was still assumed to not fall below zero.
However, for the case of oblique impacts of nanoclusters
it was shown recently [20] that the rotation of the unit vector
ϕ
′
may lead to negative values of the coefficient of normal
restitution defined by Eq. (1). This surprising result was
attributed to the softness of nanoclusters leading to relatively
long contact durations which, in turn, may lead to a significant
reorientation of the particles’ normal vector during a collision.
While the investigation in [20] refers to nanoclusters
colliding at very large impact velocity (1850 m/s), here we
show that negative coefficients of normal restitution are a
much more general phenomenon. We show that negative values
occur inevitably for any type of collision whose dynamics is
governed by finite interaction forces leading to finite duration
of contact: For any set of material properties there is always a
collision geometry leading to negative values of ε
HS
n
.
We wish to mention that there are alternative definitions of
the coefficient of restitution (e.g., [21–25]), which avoid the
problem addressed here. However, only the definition Eq. (1)
or (2), respectively, referred to as Newton’s model [26], allows
for the computation of the post-collisional vectorial velocities
which are needed for both event-driven simulations and kinetic
theory.
In Sec. II we introduce our notation. In Sec. III we
derive a general condition for negative coefficients, while in
Secs. IV and V we specify our findings to particles which
interact via a linear dashpot model and to smooth viscoelastic
spheres, respectively. Section VI discusses the role of frictional
forces between the colliding particles. Finally, in Sec. VII we
summarize our results and present some outlook.
II. COLLISION OF SMOOTH SPHERES
The collision of a pair of smooth spheres of masses m
1
and
m
2
, located at r
1
and r
2
, is governed by Newton’s equation
041306-1 1539-3755/2012/85(4)/041306(8) ©2012 American Physical Society