IMA Preprint 2473
IMA Preprint 2473
By
(August 2016)
Feifei Wang
Department of Mathematics, Iowa State University
Ames, Iowa 50011
wangff@iastate.edu
Scott W. Hansen
Department of Mathematics, Iowa State University
Ames, Iowa 50011
shansen@iastate.edu
Key words: Signorini problem, elastic rod, coefficient of restitution, viscous wave
equation
Abstract
1 Introduction
The classical problem of determining the height of rebound of a bouncing object has
been investigated in a huge number of works going back Newton (1687) who defined the
coefficient of restitution (COR) as a ratio of relative velocities after and before a collision,
Hertz, 1882 [5], who developed the foundations of contact mechanics (see e.g., [8]) based
on stresses in spherical elastic bodies, and Love 1905 [12] who investigated the dynamics
of colliding elastic rods governed by one-dimensional wave equations.
Although there are numerous applications where impact problems are studied; e.g.,
sports science, robotics and earthquake dynamics (see e.g., [4], [6], [11], [18], [20], [19],
[21], [10]), there are very few situations where a precise knowledge of motion during the
impact has been determined. Consequently in most applications, an impact is modeled
using a COR. While this methodology has become quite developed, (see e.g., Stronge [17]
for a detailed description of various theories of COR and application), there are numerous
drawbacks and inconsistencies with this approach, some of which are described in [16].
1
A COR less than 1 describes an inelastic collision, and hence describes some amount of
dissipation (at least in terms of velocity) that occurs during the impact. It has long been
thought [16] that dissipation during an impact results from 1) localized plastic deformation;
2) viscous damping in the material; and 3) energy transfer to vibrational energy. Therefore
a better knowledge (even in specialized situations) of the portion of translational kinetic
energy that is transferred to vibrational energy during the impact, could improve how the
COR approach is applied.
One situation where it is possible to write down an explicit solution formula (see e.g.,
Love [12]) for an impact is the one-dimensional elastic rod
q ρ so that g ≥ 0 is propor-
where in comparison to (1.1), we have divided the equation by
tional to the gravitational constant, and the wave speed c = σρ is positive. The viscous
damping term 2rvt (r ≥ 0) describes a frictional force proportional to the velocity.
The rod is assumed to be stress-free at the end x = L and we impose Signorini boundary
conditions at the end x = 0:
In our formulation of this problem, we assume that the rod impacts the ground at time 0
with constant initial velocity −µ:
v(x, 0) = 0, x ∈ ΩL , (1.7)
vt (x, 0) = −µ, x ∈ ΩL , (1.8)
2
The Signorini boundary conditions (1.4)–(1.6) (also called unilateral constraints) essen-
tially state that the bottom of the rod should be stress-free while not in contact with the
ground and the Dirichlet condition v(0, t) = 0 should be imposed when in contact with
the ground.
Signorini problems have been considered in a wide variety of contact problems for plates
plates, beams, strings etc., see e.g., [2], [15], [1], and can be expressed as a variational in-
equality, for which there is a corresponding definition of a weak solution. For the damped
problem (1.2)–(1.6), but with general initial conditions of finite energy, Rivera and Oquen-
do [13] proved global existence of weak solutions and obtained asymptotic decay results.
Lebeau and Schatzman [9] studied an analogous multidimensional undamped wave equa-
tion on a half space, and proved existence and uniqueness of solutions in a suitably defined
functions space. Controllability in the context of a string equation was considered in [3].
For the present paper, we only need to consider the time interval [0, tb + ), where [0, tb ]
is the contact interval and is sufficiently small so that there is at most one bounce.
In this situation, solutions are classical except for jumps in the derivatives along certain
characteristic rays due to the bounce. Here tb (possibly = ∞) is defined by
Based on the Method of Descent and D’Alembert’s formula, we obtain explicit solution
formulas for v(x, t) over the contact interval [0, tb ]. The contact time tb is determined in
terms of a computable formula in Theorem 1. In particular, we show tb ≥ t1 := 2L/c. In
the case r = 0, an explicit formula in terms of parameters µ, L, c, and g for tb is derived,
which in particular shows that t1 ≤ tb ≤ 2t1 . We also find expressions for the motion of
the center of mass over the time interval [0, tb + ] for small . At the time tb , the energy of
the elastic rod has an orthogonal decomposition into translational energy, corresponding
to the motion of the center of mass and and vibrational energy, corresponding to energy
of vibrations. We are able to calculate this decomposition at the time of impact, and
define an associated COR based on the ratio of translational energy after and before the
collision. Based on this definition of COR, we are able to make some observations about
the dependence of internal vibrational energy as a function of the µ, the velocity of impact;
see Section 4.
3
Setting v̂(x, t) = e−rt w(x, t) in (2.12) results in
2 2 rt
wtt − c wxx = r w − ge ,
(x, t) ∈ Ω2L × (0, tb ),
w(x, 0) = 0, wt (x, 0) = −µ, x ∈ Ω2L , (2.13)
w(0, t) = 0, w(2L, t) = 0, t ∈ (0, tb ).
We see below that (2.11) can be solved explicitly using the method of descent together
with the reflection principle. For a first step, consider the homogeneous problem on the
infinite domain:
(
utt − c2 uxx − c2 λ2 u = 0, x ∈ R, t ∈ (0, +∞),
(2.14)
u(x, 0) = 0, ut (x, 0) = ϕ(x). x ∈ R.
Proof: We apply the “method of descent”; see e.g., [7]. Let u be the solution of (2.14)
and define û(x1 , x2 , t) = cosh(λx2 )u(x1 , t). Then it is easy to verify that
(
ûtt − c2 (ûx1 x1 + ûx2 x2 ) = 0 x1 , x2 ∈ R, t ∈ (0, +∞),
(2.17)
û(x1 , x2 , 0) = 0, ût (x1 , x2 , 0) = cosh(λx2 )ϕ(x1 ), x1 , x2 ∈ R.
u(x1 , t) = û(x1 , 0, t)
Z x1 +ct Z √c2 t2 −(y1 −x1 )2
1 ϕ(y1 ) cosh(λy2 )
= √ p dy2 dy1
2πc x1 −ct − c2 t2 −(y1 −x1 )2 c2 t2 − (y1 − x1 )2 − y22
Z x1 +ct Z s
1 cosh(λy2 ) p
= ϕ(y1 ) p dy 2 dy 1 ; (s = c2 t2 − (y1 − x1 )2 )
2πc x1 −ct −s s2 − y22
Z x1 +ct Z π/2
1
= ϕ(y1 ) cosh(λs sin θ) dθ dy1
2πc x1 −ct −π/2
1 x1 +ct
Z
= ϕ(y1 )I0 (λs) dy1 ,
2c x1 −ct
4
as claimed. The proof is complete.
For the moment, consider problem (2.13) in the more general form:
2 2
utt − c uxx − r u = h(x, t),
x ∈ Ω2L , t ∈ (0, +∞),
u(x, 0) = 0, ut (x, 0) = ϕ(x), x ∈ Ω2L , (2.18)
u(0, t) = u(2L, t) = 0 t ∈ (0, +∞).
We will use symmetry to extend the data in (2.18) to all of R. To this end, given a
function f : (0, 2L) → R, let f˜ : (−2L, 2L) \ {0} → R be the odd extension of f . Then
define the symmetric extension f e : R \ 2LZ → R as the periodic extension of f˜.
Proposition 1. Assume that ϕ ∈ C[0, 2L], h and ht ∈ C([0, 2L] × [0, T ]). Then (2.18)
has a unique weak solution u in C([0, T ]; H01 [0, 2L]), and ut in C([0, T ]; L2 (0, 2L)), which
is given pointwise by
1 x+ct
Z p
u(x, t) = I0 (r t2 − (x − y)2 /c2 )ϕe (y) dy (2.19)
2c x−ct
1 t x+cτ
Z Z p
+ I0 (r (τ 2 − (x − y)2 /c2 )he (y, t − τ ) dy dτ.
2c 0 x−cτ
The solution above is continuous and continuously differentiable except on the character-
istics ct = ±(x − 2kL), k ∈ Z, where the partial derivatives ux and ut could have jump
discontinuities.
Proof: Under the assumptions on ϕ and h, it is well known that (2.18) has a unique
weak solution as described in Proposition 1. Let u1 denote solution to the problem (2.18)
with h = 0. It is easily verified that the reflection principal applies to formula (2.15) and
problem (2.14) exactly the same way that D’Alembert’s formula applies to the undamped
(r = 0) wave equation (2.18). Hence by the reflection principle u1 coincides with the
solution ue1 of the problem (2.14) with initial velocity ϕe (and h ≡ 0). Thus
Z x+ct
1 p
u1 (x, t) = I0 (r t2 − (x − y)2 /c2 )ϕe (y) dy. (2.20)
2c x−ct
5
2.1 Calculation of jumps along characteristics
Let u(x, t) be the solution of (2.18) as given by (2.19) with λ = r/c. A calculation gives
1 e
ux (x, t) = (ϕ (x + ct) − ϕe (x − ct))
2c
1 x+ct (x − y)λ e
Z
− I1 (λs(t, x − y)) ϕ (y) dy
2c x−ct s(t, x − y)
1 t e
Z
+ (h (x + cτ, t − τ ) − he (x − cτ, t − τ )) dτ
2c 0
1 t x+cτ −(x − y) e
Z Z
+ λI1 (λs(τ, x − y)) h (y, t − τ ) dy dτ, (2.22)
2c 0 x−cτ s(τ, x − y)
√
where s(t, r) = c2 t2 − r2 , and
Z π
1
I1 (z) = I00 (z) = cos θez cos θ dθ
π 0
Similarly,
1 e
ut (x, t) = (ϕ (x + ct) + ϕe (x − ct))
2
1 x+ct rtI1 (λs(t, x − y)) e
Z
+ ϕ (y) dy
2 x−ct s(t, x − y)
1 x+ct
Z
+ I0 (λs(t, x − y))he (y, 0) dy
2c x−ct
1 t x+cτ
Z Z
+ I0 (λs(τ, x − y))het (y, t − τ ) dy dτ (2.23)
2c 0 x−cτ
6
We again can easily calculate [ut ], the jump in ut along the characteristics. If (x, t) ∈
Ξ− \ Ξ+ , then
1 1
[ut (x, t)] = [ut (x0 − ct, t)] = ([ϕe (x0 − 2ct)] + [ϕe (x0 )]) = [ϕe (x0 )].
2 2
If (x, t) ∈ Ξ+ \ Ξ− , then
−1 e 1
[ut (x, t)] = [ut (x0 + ct, t)] = ([ϕ (x0 )] + [ϕe (x0 + 2ct)]) = − [ϕe (x0 )].
2 2
We summarize this calculation in the following.
Proposition 2. The solutions u(x, t) given in Proposition 1 satisfy the following proper-
ties:
(i) If (x, t) = (x0 − ct, t) ∈ Ξ− \ Ξ+ , then
1 1
[ux (x, t)] = [ϕe (x0 )], [ut (x, t)] = [ϕe (x0 )].
2c 2
In the case of interest, namely ϕ(x) = −µ is a constant and h(x, t) = −gert with x ∈ Ω2L .
Therefore ϕe (x) = −µ(x)e and h(x, t)e = −g(x)e ert , where g(x)e and µ(x)e denote the
symmetric extensions defined on R \ 2LZ of the constant functions g, µ (originally defined
on (0, 2L)). With λ = rc , we see from (2.13) and Proposition 1 that the solution to the
original problem (2.11) is
v(x, t) = e−rt u(x, t) = e−rt w(x, t), where (2.24)
Z x+ct
1 p
w(x, t) = I0 (λs(t, (x − y))(−µe (y)) dy; (s(t, r) := c2 t2 − r2 ),
2c x−ct
1 t x+cτ
Z Z
+ I0 (λs(τ, x − y))(−g e (y)er(t−τ ) ) dy dτ.
2c 0 x−cτ
Since vx (0, t) = e−rt wx (0, t), the boundary condition (1.4) holds if and only if wx (0, t) 6 0.
Therefore, we compute wx (0, t) = ux (0, t) where u(x, t) is given by (2.19) with ϕe = −µe
and he = −g e ert . We obtain
1 e
wx (0, t) = (ϕ (ct) − ϕe (−ct))
2c
1 ct
Z
(−y)λ e
− I1 (λs(t, y)) ϕ (y) dy
2c −ct s(t, y)
1 t e
Z
+ h (cτ, t − τ ) − he (−cτ, t − τ ) dτ
2c 0
1 t cτ
Z Z
y
+ λI1 (λs(τ, y)) he (y, t − τ ) dy dτ. (2.25)
2c 0 −cτ s(τ, y)
7
Assume for some k = k(t) ∈ N0 = {0, 1, 2...} that
or equivalently
t1 k < t < t1 (k + 1). (2.26)
Using that (i) s(t, r) is even with respect to r, (ii) φe = −µe is an odd function that is
constant on (2jL, 2(j + 1)L), ∀j ∈ Z, (iii) s(t, ±ct) = 0, (iv) λ = r/c, we compute the
first two terms in (2.25),
1 ct
Z
1 e (−y)λ e
(ϕ (ct) − ϕe (−ct)) − I1 (λs(t, y)) ϕ (y) dy
2c 2c −ct s(t, y)
1 ct d
Z
1 e
= (2ϕ (ct)) − I0 (λs(t, y)) ϕe (y) dy
2c 2c −ct dy
k−1 Z
ϕe (ct) 1 X 2L(j+1) d
= − I0 (λs(t, y)) φe (y) dy
c 2c 2Lj dy
j=−k
Z −2Lk Z ct
1 d
− + I0 (λs(t, y)) φe (y) dy
2c −ct 2Lk dy
e k−1
ϕ (ct) 1 X 2L(j+1)
−2Lk
= − { I0 (λs(t, y))φe (y) 2Lj + I0 (λs(t, y))φe (y) −ct
c 2c
j=−k
ct
+I0 (λs(t, y))φe (y) 2Lk }
1 e +
= 2ϕ (0 ){I0 (λs(t, 0)) − ( I0 (λs(t, 2L)) + I0 (λs(t, −2L)) )
2c
+ ( I0 (λs(t, 4L)) + I0 (λs(t, −4L)) ) − ...... + (−1)k (I0 (λs(t, 2kL)) + I0 (λs(t, −2kL)))}
(1 e +
2c (2ϕ
(0 )I 0 (rt)) k = 0,
= 1 e +
P k j
p
2 2
2c 2ϕ (0 ) I0 (rt) + 2 j=1 (−1) I0 (r t − (jt1 ) ) k > 1.
−µ
= Hr (t), (2.27)
c
where on each interval t ∈ (t1 k, t1 (k + 1)),
(
I0 (rt), t ∈ (0, t1 := 2L
c ),
Hr (t) :=
I0 (rt) + 2 kj=1 (−1)j I0 (r t2 − (jt1 )2 )
P p
t ∈ (kt1 , (k + 1)t1 ).
and k is defined in terms of t as in (2.26). Similar calculations can be used to simplify the
last two terms in (2.25). Hence we obtain
−µ g t r(t−τ )
Z
wx (0, t) = Ψr (t) := Hr (t) − e Hr (τ )dτ. (2.28)
c c 0
Returning to the original system (1.2)–(1.8), we see that inequality (1.4) is maintained if
and only if Ψr (t) 6 0. Therefore, we have the following result that defines time tb of the
first bounce in equation (1.10).
Theorem 1. Let set S = {t : Ψr (t) > 0}, where Ψr (t) is defined by (2.28).
(i) If S = ∅, there is no bounce.
8
(ii) If S 6= ∅, then the first bounce is given by tb = inf S > t1 = 2L/c.
(iii) tb = t1 if and only if Ψr (t−
1 ) + 2µ/c > 0, i.e.
Z t1
−µI0 (rt1 ) + 2µ − g er(t1 −τ ) I0 (rτ ) dτ > 0, (2.29)
0
Proof. Part (i) is clear. Hence assume S is not empty, and tb ∈ R. Note that Ψr (0) < 0
and is strictly decreasing for t ∈ (0, t1 ). Therefore tb > t1 , which proves part (ii). For part
(iii), in order for a bounce to occur at time t1 , the stress function Ψr (t) has to change
sign at time t1 . By Proposition 2, [ux (0, t1 )] = 2µ/c. Hence, the condition for a bounce is
Ψr (t−1 ) + 2µ/c > 0, which simplifies to (2.29).
Since the right hand side is positive, a necessary condition for solving (2.30) is
2 − I0 (rt1 ) > 0.
Since the graph of I0 (z) is monotonically increasing for z > 0, the above condition is
equivalent to
rt1 < I0−1 (2) ' 1.8. (2.31)
Then (2.30) can be simplified as
R t1
0 er(t1 −τ ) I0 (rτ ) dτ
µ/g > . (2.32)
2 − I0 (rt1 )
Therefore, in order for a bounce to occur at time t1 , r must be sufficiently small so that
(2.31) holds. If also µ/g is large enough such that (2.32) holds, then there will be a bounce
at time t1 .
We examine in further details for two special cases.
The case r = 0 and g > 0: To apply Theorem 1, in case of r = 0, we know that the
stress function Ψr (t) becomes
Z t
µ g
Ψ0 (t) = − H0 (t) − H0 (τ ) dτ,
c c 0
where (
1, t ∈ (0, t1 ),
H0 (t) :=
1 + 2 kj=1 (−1)j
P
t ∈ (kt1 , (k + 1)t1 ).
Set S in Theorem 1 is defined by t for which Ψ0 (t) > 0.
The function Ψ0 (t) is piecewise linear and 2t1 -periodic. One easily deduces the following,
which is more precise than Theorem 1.
9
Corollary 1. Assume r = 0 and g > 0 and µ > 0. There is a first bounce at time tb > t1
for which
(i) if µ > gt1 , the first bounce time is tb = t1
(ii) 0 < µ < gt1 , the first bounce time is tb = 2t1 − µg .
We mention that Corollary 1 was also obtained in [14], although his formulation is
different than ours.
The case g = 0, r > 0: The condition (2.29) for a bounce at time tb = t1 = 2L/c reduces
to
I0 (rt1 ) < 2;
or equivalently,
rt1 < I0−1 (2) ' 1.8. (2.33)
Figure 1: Plot for Ψr (t) with different r values of 2.6, 2.9 and 3.0, when g = 0, t1 = 1, µ = 1,
and c = 1, the time tb of first bounce for each case is indicated by a dot.
First we derive formulas for the center of mass and momentum of the rod.
10
3.1 Momentum and Center of Mass
We consider the system (2.11). The momentum of the rod for t > 0 is given by
Z L
P (t) = vt (x, t) dx. (3.34)
0
P 0 (t) + 2rP (t) = −c2 e−rt Ψr (t) − gL, 0 6 t 6 tb ; P (0) = −µL. (3.37)
Let h(t) denote the position at time t of center of mass of the rod relative to the midpoint
RL
of the rod at time 0. Thus h(0) = 0. For any t > 0, h(t) = L1 0 X(x, t) dx − L/2, where
X(x, t) denotes the position of the point with reference position x at time t defined in
(1.9). Then h(t) is given by
1 L 1 L
Z Z
h(t) = X(x, t) dx − L/2 = v(x, t) dx
L 0 L 0
1 L
Z Z t
= v(x, 0) + vt (x, τ ) dτ dx
L 0 0
1 t L
Z Z
= vt (x, τ ) dx dτ
L 0 0
1 t
Z
= P (τ ) dτ. (3.39)
L 0
Define
v 0 (x) = lim v(x, t), v 1 (x) = lim vt (x, t).
t→t−
b t→t−
b
These limits exist since by Proposition 1, v(x, tb ) is continuous and vt (x, tb ) is piecewise
continuous for 0 < x < L. Let
1 L 0 1 tb
Z Z
h0 = h(tb ) = v (x) dx = P (τ ) dτ, (3.40)
L 0 L 0
1 L 1
Z
1 0 1
h = h (tb ) = v (x) dx = P (tb ). (3.41)
L 0 L
Note that v 0 (x) and v 1 (x) do not need to be known in order to compute h0 and h1 since
P (t) is given explicitly by equation (3.38).
11
Let t2 = sup {t : v(0, t) > 0 on (tb , t)}. We can also solve h(t) and P (t) on time interval
(tb , t2 ). The PDE on this interval is:
2
vtt + 2rvt − c vxx = −g,
(x, t) ∈ ΩL × (tb , t2 ),
v(x, t1 ) = v 0 (x), vt (x, t1 ) = v 1 (x), x ∈ ΩL , (3.42)
vx (0, t) = vx (L, t) = 0, t ∈ (tb , t2 ).
Since the bottom of the rod is stress free on this interval, it follows that Ψr (t) = 0 on
(t1 , tb ). Thus (3.35) changes to
1 t
Z
0
h(t) = h + P (τ ) dτ
L tb
A − g(t − tb ) − h1 e−2r(t−tb ) g
= + (1 − e−2r(t−tb ) ), (3.44)
2r (2r)2
where
( Rt
−c2 e−2rt 0 erτ Ψ(τ ) dτ − gL
2r (1 − e
−2rt ) − µLe−2rt , t ∈ [0, t ],
b
P (t) = gL 1 gL −2r(t−tb ) (3.46)
− 2r + (h L + 2r )e , t ∈ [tb , t2 ],
Consequently
Z t
1 1
h(t) = P (τ ) dτ = (gct3 /6 + (µc − gL)t2 /2 − µLt), t ∈ [0, t1 ]. (3.48)
L 0 L
Therefore
2gL2 P (t1 ) µL
h0 = h(t1 ) = − , h1 = = = µ. (3.49)
3c2 L L
12
Thus, in the pure elastic case without damping, the momentum immediately after the
collision is the same in magnitude as before the collision. However, when g > 0, internal
stresses remain present.
For t ∈ (t1 , t2 ): Take limit as r → 0+ from equation (3.43) and (3.44), one obtains
and (
1 3
L (gct /6 + (µc − gL)t2 /2 − µLt), t ∈ [0, t1 ],
h(t) = 2 gt2 2L(µ−gt) (3.51)
− 8gL
3c2
+ µt − 2 − c , t ∈ [t1 , t2 ].
and
gt2 µct2 gct3
−µt − 2 + 2L + 6L , t ∈ [0, t1 ],
(4L−ct)(3µc(−2L+ct)+g(4L2 −5cLt+c2 t2 ))
h(t) = 6c2 L
, t ∈ [t1 , tb ], (3.53)
µ3 c3 +(ct−4L)(3c2 gµ2 −6cg 2 Lµ)−3g 3 L(ct−4L)2
, t ∈ [tb , t2 ].
6c2 g 2 L
Remark 2. One can use (3.50)-(3.53) to compute the maximum height hmax = h(t∗ )
by setting P (t∗ ) = 0. One obtains
2rh1
1
t∗ < t2 ,
tb +
2r ln(1 + g ), r 6= 0,
t∗ = 2gL+µc
, r = 0, µ > gt1 ,
8g2cg
L2 −2cgLµ+c2 µ2
, r = 0, 0 < µ < gt1 ,
2cg 2 L
13
3.2 Energy decomposition
The energy of the rod consists of the sum of the kinetic energy, the strain energy and the
gravitational potential energy and is given by
1 L
Z
(vt )2 + c2 vx2 dx + Lgh(t),
E(t) = (3.54)
2 0
for any t ∈ (0, t2 ). At points where E is differentiable one finds
Z L
d
E(t) = (vt vtt + c2 vx vxt ) dx + Lgh0 (t)
dt 0
Z L
2 L
= c vx vt |0 + vt (vtt − c2 vxx ) dx + P (t)g
0
Z L
= vt (−2rvt − g) dx + P (t)g
0
Z L
= −2r vt2 dx. (3.55)
0
The boundary value problem satisfied on (tb , t2 ) is (3.42). Due to the stress free bound-
ary conditions in (3.42), for t ∈ (tb , t2 ), the eigenfunctions corresponding to (3.42) are
{cos kπx/L}∞ k=0 . In particular the constant component is orthogonal to all other eigen-
functions. Hence if we decompose the initial data:
1 L 0
Z Z L
0 0 0 0
v (x) = v (s) ds + z (x) = h + z (x); z 0 (s) ds = 0 (3.56)
L 0 0
1 L 1
Z Z L
1 1 1 1
v (x) = v (s) ds + z (x) = h + z (x); z 1 (s) ds = 0, (3.57)
L 0 0
where h0 and h1 are given in (3.40) and (3.41), then the solution v(x, t) of (3.42) on [tb , t2 ]
has the orthogonal decomposition
and z satisfies
2
ztt + 2rzt − c zxx = 0,
(x, t) ∈ ΩL × (tb , t2 ),
z(x, tb ) = z 0 , zt (x, tb ) = z 1 , x ∈ ΩL , (3.60)
zx (0, t) = zx (L, t) = 0, t ∈ (tb , t2 ).
14
where for t ∈ [tb , t2 ],
1 L 2
Z
Ez (t) = z + c2 zx2 dx (3.62)
2 0 t
1 L 0
Z
1 2
Eh (t) = (h (t))2 dx + Lgh(t) = P (t) + Lgh(t). (3.63)
2 0 2L
Ez (t) is the vibrational energy, and Eh (t) is the energy corresponding to the motion of the
center of mass, which we refer to as the translational energy.
The goal of this subsection is to determine or approximate this decomposition at the
time of bounce tb . From equation (3.55), (3.58), (3.62), and (3.63), these energies have
respective decay rates given by (for t ∈ [tb , t2 ])
Z L
d 2r
Eh (t) = −2r (h0 )2 dx = − P (t)2 , (3.64)
dt 0 L
Z L
d
Ez (t) = −2r (zt )2 dx. (3.65)
dt 0
When r = 0, one can use the conservation of energy law on [0, t2 ] to obtain
1 L µ2 L
Z
E(t) = E(0) = (vt (x, 0)2 + c2 vx (x, 0)2 ) dx + Lgh(0) = := Etot . (3.66)
2 0 2
Furthermore, Eh (t) can be computed with (3.50)-(3.53) and (3.63). Explicitly, one obtains
for the case µ > µ0 (thus t1 = tb ), for t ∈ [t1 , t2 ]
1 2
Eh (t) = L (−g(t − 2L/c) + µ)2 + Lgh(t)
2L
2
8gL2 gt2 2L(µ − gt)
L 2L
= −g(t − ) + µ + Lg(− 2 + µt − − )
2 c 3c 2 c
2g 2 L3 Lµ2
= − +
3c2 2
L 2 µ20
2gL
= (µ − ) µ0 = . (3.67)
2 3 c
Similarly, one can compute for t ∈ [0, t1 ],
(µ(ct − L) + gt(ct/2 − L))2 gct3 /6 + (µc − gL)t2 /2 − µLt
Eh (t) = + Lg
2L L
12L2 µ2 + 3c2 t2 (2µ + gt)2 − 8cLt(3µ2 + 3gµt + g 2 t2 )
= .
24L
By similar computation, for the case 0 < µ < µ0 (thus tb > t1 ), using (3.52), (3.53) and
(3.66), one obtains
2 2 22
12L µ +3c t (2µ+gt)2 −8cLt(3µ2 +3gµt+g 2 t2 )
24L , t ∈ [0, t1 ],
g 2 (4L−3ct)(4L−ct)3 +12c2 µ2 (ct−3L)2 +12cgµ(ct−4L)2 (ct−2L)
Eh (t) = 24c2 L
, t ∈ [t1 , tb ], (3.68)
2 µ2 Lµ 2 (3µ2 −4µµ +3µ2 )
8cµ 3c
1
µ2 (12L − 0
g + g2 L ) =
0
, t ∈ [tb , t2 ].
24 6µ20
15
Using the energy decomposition (3.61), conservation of energy (3.66), together with
previous expressions for Eh , we have the following result.
Thus,
− µ(3µ−4µ0)
, 0 < µ < µ0 = 2gL/c,
3µ2
Ez (tb )/Etot = µ20
0 (3.70)
3µ2
, µ > µ0 .
Figure 2: Plot for Ez (tb ) when r = 0, c = 1.0, L = 1.0, g = 1.0 change with µ, the red dot
corresponds to µ = µ0 .
4 Coefficient of restitution
The standard definition for a coefficient of restitution (COR) for an impact of a particle
against a fixed surface normal to the motion of the particle is the ratio of velocities, or
equivalently ratio of momenta, immediately after impact (v1 or P1 ) to the immediately
before the impact (v0 or P0 ) i.e,
P1 v1
e=− =− .
P0 v0
16
Figure 3: Plot for Eh (tb )/E(0) and Ez (tb )/E(0), when r = 0, c = 1.0, L = 1.0, g = 1.0
change with µ, the red dot corresponds to µ = µ0 .
Thus a purely elastic collision is described by e = 1 and purely inelastic, where the motion
of the particle stops at impact is described by e = 0. Intermediate values of e describe
the degree of inelasticity of the impact. The idea of applying a COR to bouncing objects
goes back to Isaac Newton, who suggested (see [17]) for example that glass should have
e = 15/16 and steel should have e = 5/9.
Analogously, we can define a momentum coefficient of restitution for the bouncing elastic
rod as the ratio of momenta after and before the impact:
P (tb )
eP = − ,
P (0)
where tb is the bounce time. Equivalently, eP = −vCM (tb )/vCM (0), where vCM denotes
the velocity of the center of mass.
Thus, we have (
1, µ > gt1 ,
eP = µc µ
2gL = µ0 , 0 < µ < gt1 .
A flaw in this definition is that when g > 0, the position of center of mass at the end of
collision is lower than the initial center of mass. Hence, even when eP = 1, the elastic
rod will not bounce as high as it was dropped from (to generate the initial momentum
17
P (0)). Conservation of energy is not violated since after the bounce, internal energy
remains within the rod. This suggests an adjustment of the definition of COR based on
translational energy Eh which we propose next.
Since the internal energy Ez of the flexible rod does not contribute to the motion of the
center of mass, we define the energy coefficient of restitution eE as the square root of the
ratio of the energies Eh after, and before the impact:
1
Eh (tb ) 2
eE = . (4.71)
Eh (0)
Thus, 2
Eh (tb ) 3µ −4µµ20 +3µ20 , 0 6 µ < µ0 ,
3µ0
e2E = = µ20
(4.72)
Eh (0) 1 − , µ > µ0 .
3µ2
Thus the relative amount of internal energy in the bouncing elastic rod is maximized
at the intermediate value µ = 2µ0 /3 and as µ increases to infinity, the impact approaches
perfectly elastic.
A limitation in application of this definition of COR is that when one attempts to ap-
ply this definition to repeated bounces, subsequent initial conditions of the impact would
include some internal vibrations, and hence would not be of the same form (uniform veloc-
ity) as considered here. On the other hand, for some applications it could be appropriate
to assume that any internal vibrations remaining after an impact are converted to heat,
or otherwise dissipated before the next impact.
The damped case r 6= 0: Determination of the COR (using either the momentum or
energy definition) is much more complex in the damped case since all formulas involve
computing the complicated stress function Ψr (t) from (2.28). Therefore we focus on the
energy COR eE .
The energy COR eE can be computed numerically by the following steps:
18
2. Determine the bounce time tb from Theorem 1. This involves finding the first root
of the stress function Ψr (t), as in Figure 1.
For the parameters we considered, we were able to compute eE and related graphs using
Mathematica (with about 100 time steps) in about 2 minutes computational time.
Eh (t1 )
Figure 4: Plot for e2E = E(0) with c = 1, L = 1, g = 1, r = 0, 0.1, 0.2, 0.3 as a function of
µ.
For the parameters considered in Figure 4, the two conditions (2.31) and (2.32) are
satisfied for values of µ to the right of the black dot in each of the cases r = 0, 0.1, 0.2, 0.3.
(These values of µ correspond to those for which tb = t1 . In each case the curves continue
to the left of the black dots, but are significantly more difficult to compute since tb depends
on µ and r in a more complicated way.) Thus in each case tb = t1 = 2. The red curve is
the undamped case r = 0 and coincides with the right-most portion of the graph of the
corresponding function in Figure 3.
We mention that when r > 0, it is possible that Eh (tb ) can become negative. This
happens when at the time of the bounce, h(tb ) < 0 and the kinetic energy P (tb )2 /2L is
smaller in magnitude than the drop in potential energy |Lgh(tb )|; see eq. (3.63). This
implies that the center of mass can never return to the initial value of L/2 (corresponding
to h = 0) and thus in terms of the center of mass, there is not really a “bounce”. Therefore
we leave eE undefined in these cases.
In the case r = 0 we noted earlier that limµ→∞ eE = 1. For r > 0 however the graph
suggests a different limit as µ tends to infinity. In order to numerically investigate this
limiting behavior more systematically, we note that
Eh (tb ) E(tb )
e2E = · . (4.73)
E(tb ) E(0)
19
The second is a ratio of energy at time tb to the initial energy, hence is directly related to
the decay rate of energy.
In order to analyze the terms EE(t
h (tb )
b)
and E(tb)
E(0) separately, it is necessary to compute E(tb ).
Unfortunately there is no simple way to compute this and one must directly compute
E(tb ) from (3.54) by computing vt (x, tb ) and vx (x, tb ) for 0 < x < L. This computation is
described in the Appendix.
The ratios Eh (t1 )/E(t1 ) and E(t1 )/E(0) versus µ are graphed separately in Figures 5 and
6 below. We picked the same parameter values that were used in Figure 4; i.e., g = 1.0,
L = 1.0, c = 1.0. The four curves correspond to the r values as 0.0, 0.1, 0.2 and 0.3. The
values of µ changes from 3.2 to 12.5.
The graphs are linear interpolations of data based on 20 time steps. (This computation
took hours to run, so we did not attempt a finer time step.)
Figure 5: Plot for Eh (t1 )/E(t1 ) when c = 1.0, L = 1.0, g = 1.0 change with µ.
Figure 5 indicates a very similar behavior with respect to µ of the ratio Eh (t1 )/E(t1 )
compared to the undamped case, where energy is conserved, and hence Eh (t1 )/E(t1 ) =
Eh (t1 )/E(0) = e2E . Since limµ→∞ eE = 1. Figure 5 suggests that
Eh (t1 )
lim = 1. (4.74)
µ→∞ E(t1 )
Figure 6 suggests that the ratios E(t1 )/E(0) are nearly constant functions of µ for large
values of µ. This nearly constant value varies with respect to the damping r in a way that
is related to the behavior of the decay rate of the Fourier series solution corresponding to
the given initial data. This involves infinitely many modes since the initial data does not
correspond to a finite combination of eigenfunctions for boundary conditions v(0, t) = 0,
vx (L, t) = 0. However the energy decay in (3.55) suggests that the ratios E(t1 )/E(0) might
be proportionate to e−2rtb . This suggests investigating the quantity E(te−2rt
1 )/E(0)
1
which is
graphed in Figure 7.
20
Figure 6: Plot for E(t1 )/E(0) when c = 1.0, L = 1.0, g = 1.0 change with µ.
We were able to verify this numerically for the curves in Figure 4 to within 0.001 for each
curve. This required 100 time steps in the numerical integrations.
E(t1 )/E(0)
Figure 7: Plot for e−2rt1
when c = 1.0, L = 1.0, g = 1.0 change with µ.
5 Conclusions
In this paper we have analyzed the motion of a linear elastic rod with viscous damping
over the duration of an impact with the ground, which is assumed to be rigid.
21
First, we derived an explicit solution formula (2.19) for the motion of a bouncing elastic
rod during the impact. Based on this formula we were able to find an expression for the
stress function Ψ(t) at the bottom of the rod during the impact, and correspondingly obtain
a characterization of the bounce time tb in Theorem 1. In the undamped case (r = 0),
there is a closed form expression for tb described in Corollary 1, where in particular,
t1 := 2L/c ≤ tb < 2t1 . In the damped case we were not able to prove that Ψ must have a
root (so that there is a bounce), but were able to show that if there is a root, then tb ≥ t1 ,
and moreover found examples where tb > 3t1 (see Fig. 1).
We also obtained closed form expressions (3.45), (3.46) in terms of Ψ(t) and physical
constants c, µ, g, L for the relative center of mass h(t) and momentum P (t) for the time
interval [0, t2 ]. If r = 0, these expressions become explicit in terms parameters c, µ, g, L;
see (3.50)-(3.53).
At the time tb , the total energy E(t) has an orthogonal decomposition into translational
energy Eh (t) and vibrational energy Ez (t). Since Eh can be expressed in terms of h and
P , we obtained explicit expressions in terms of Ψ(t) for Eh (t) for t ∈ [0, t2 ]. This led to a
natural energy-based definition for the COR eE defined in (4.71). With this definition, eE =
.9, for example, would indicate that 81 percent of the original energy was in translational
form at the end of the impact, and 19 percent was either dissipated by damping or in
vibrational form. In the undamped case, eE is given explicitly in terms of parameters c,
µ, g, L in (4.72) and is graphed as a function of µ in Fig. 3.
In the damped case, the expression for eE depends on the stress function Ψ(tb ), and
hence the behavior with respect to µ is more complex when bounce times need to be
calculated. Figures 4 describes the behavior with respect to µ when µ is large enough so
that tb = t1 (so that the computation is not difficult). Part of the value of eE is due to
damping and part is due to the ratio of vibrational energy to total energy at the end of the
contact period, as indicated in (4.73). Figures 5, 6, 7 characterize, up to numerical error,
the behavior with respect to µ of the decomposition (4.73) of eE . Based on these graphs,
we suspect that the limiting behavior with respect to µ is described by (4.75), which at
least for the examples considered, seems to be correct.
6 Acknowledgements
This research was supported in part by the Institute for Mathematics and its Applications
with funds provided by the National Science Foundation. In addition, the second author
was supported by the National Science Foundation under grant DMS-1312952.
7 Appendix
In this appendix, we sketch the steps that were taken to numerically compute E(t1 ),
particularly when r 6= 0 and tb = t1 . Reviewing equation (3.54),
1 L
Z
vt (x, t1 )2 + c2 vx (x, t1 )2 dx + Lgh(t1 ).
E(t1 ) =
2 0
h(t1 ) can be calculated using (3.38) and (3.39) as shown in Section 3.1. Alternatively,
(3.34) can be used once vt (x, t1 ) has been computed numerically. Then again use (3.39)
22
to get h(t1 ). We need to compute vt (x, t1 ) and vx (x, t1 ). The idea is to apply equations
(2.22), (2.23) and (2.24) with v(x, t) = e−rt w(x, t), φe (y) = −µe (y) and he (y, t − τ ) =
−g e (y)er(t−τ ) .
Thus, vt (x, t1 ) and vx (x, t1 ) can be computed from w(x, t1 ), wt (x, t1 ) and wx (x, t1 ) as
below:
vt (x, t1 ) = −re−rt1 w(x, t1 ) + e−rt1 wt (x, t1 ),
vx (x, t1 ) = e−rt1 wx (x, t1 ).
The integrands for the terms w(x, t1 ), wt (x, t1 ) and wx (x, t1 ) have discontinuities along
characteristic lines ct = ±(x − 2kL), k ∈ Z. Consequently some care should be taken to
evaluate the integrals correctly in the separate regions, which are defined by the charac-
teristics, see Fig. 8. For example, in order to compute wt (x, t1 ), after using the√definition
of the extended functions µe and g e , and recall that λ = r/c and s(t, r) = c2 t2 − r2 ,
(2.23) becomes the following,
Z 0 Z 2L
1 2 I1 (λs(t1 , x − y)) I1 (λs(t1 , x − y))
wt (x, t1 ) = c t1 µλ dy−c2 t1 µλ dy
2c x−ct1 s(t1 , x − y) 0 s(t1 , x − y)
Z x+ct1
I1 (λs(t1 , x − y))
+ c2 t1 µλ dy + 2cµ
2L s(t1 , x − y)
Z 0 Z 2L
+ g( I0 (λs(t1 , x−y))s(t1 , x−y) dy− I0 (λs(t1 , x−y))s(t1 , x−y) dy
x−ct1 0
Z x+ct1
+ I0 (λs(t1 , x − y))s(t1 , x − y) dy)
2L
Z t1Z x+cτ
e r(t1 −τ )
+ r I0 (λs(τ, x − y))(−g (y)e ) dydτ . (7.76)
0 x−cτ
We handle the last double integral as follows. Given a point (x, t1 ), where x ∈ (0, L), the
region of integration is the region in Fig. 8 determined by characteristic lines emanating
from point (x, t1 ). We write this integral as a sum of three integrals, based on their region
of integration. In Fig. 8, the purple regions correspond to g e (x) = −g, while the green
area corresponds to g e (x) = g. Taking this geometry into consideration, the last term can
be written as
Z t Z x+cτ
r I0 (λs(τ, x − y))(−g e (y)er(t−τ ) ) dydτ
0 x−cτ
Z 0 Z t1 + y−x
c
= 2(−r (−g)I0 (λs(t1 − τ, x − y))erτ dτ dy)
x−ct1 0
Z t1 Z x+c(t1 −τ )
+ (−r (g)I0 (λs(t1 − τ, x − y))erτ dydτ )
0 x−c(t1 −τ )
Z x+ct1 Z t1 − y−x
c
+ 2(−r (−g)I0 (λs(t1 − τ, x − y))erτ dτ dy).
2L 0
Similar explicit expressions for w(x, t1 ) and wx (x, t1 ) can likewise be written down.
Therefore, E(t1 ) can be expressed in terms of integrals that can be computed numerically
using Mathematica.
As a benchmark, we computed E(t1 ) the way in the conservative case r = 0, and
obtained the red curves in Fig. 5, 6, 7, which has about 1% error using Mathematica 10.3.
23
Figure 8: Integration regions A, B and C.
References
[1] J. Ahn, A vibrating string with dynamic frictionless impact, Applied numerical math-
ematics, 57 (2007), pp. 861–884.
[2] J. Ahn and D. E. Stewart, Dynamic frictionless contact in linear viscoelasticity,
IMA journal of numerical analysis, 29 (2009), pp. 43–71.
[3] F. Ammar-Khodja, S. Micu, and A. Münch, Controllability of a string submit-
ted to unilateral constraint, in Annales de l’Institut Henri Poincare (C) Non Linear
Analysis, vol. 27, Elsevier, 2010, pp. 1097–1119.
[4] R. Cross, Mechanics of swinging a bat, American Journal of Physics, 77 (2009),
pp. 36–43.
[5] H. Hertz, Über die berührung fester elastischer körper, (1882).
[6] Y.-B. Jia, M. Mason, and M. Erdmann, A state transition diagram for simul-
taneous collisions with application in billiard shooting, in Algorithmic Foundation of
Robotics VIII, Springer, 2009, pp. 135–150.
[7] F. John, Partial differential equations, volume 1 of applied mathematical sciences,
1982.
[8] K. L. Johnson and K. L. Johnson, Contact mechanics, Cambridge university
press, 1987.
[9] G. Lebeau and M. Schatzman, A wave problem in a half-space with a unilateral
constraint at the boundary, Journal of differential equations, 53 (1984), pp. 309–361.
[10] P. Lipscombe and S. Pellegrino, Free rocking of prismatic blocks, Journal of
engineering mechanics, 119 (1993), pp. 1387–1410.
[11] S. Liu, L. Wu, and Z. Lu, Impact dynamics and control of a flexible dual-arm
space robot capturing an object, Applied mathematics and computation, 185 (2007),
pp. 1149–1159.
24
[12] A. E. H. Love, A treatise on the mathematical theory of elasticity, vol. 1, Cambridge
University Press, 2013.
[13] J. Rivera and H. P. Oquendo, Exponential decay for a contact problem with local
damping, FUNKCIALAJ EKVACIOJ SERIO INTERNACIA, 42 (1999), pp. 371–388.
[14] P. Shi, The restitution coefficient for a linear elastic rod, Mathematical and computer
modelling, 28 (1998), pp. 427–435.
[16] D. E. Stewart, Rigid-body dynamics with friction and impact, SIAM review, 42
(2000), pp. 3–39.
[18] I. D. Walker, Impact configurations and measures for kinematically redundant and
multiple armed robot systems, IEEE transactions on robotics and automation, 10
(1994), pp. 670–683.
[19] F. Wang, H. Lin, and Y.-B. Jia, Computational modeling of n-body collisions,
in Intelligent Robots and Systems (IROS), 2015 IEEE/RSJ International Conference
on, IEEE, 2015, pp. 5376–5381.
[20] Y. Wang and M. Mason, Modeling impact dynamics for robotic operations, in
Robotics and Automation. Proceedings. 1987 IEEE International Conference on,
vol. 4, IEEE, 1987, pp. 678–685.
25