Pulses, fronts and oscillations of an elastic rod

Joceline Lega & Alain Goriely

Department of Mathematics and Program in Applied Mathematics,
University of Arizona, Building #89, Tucson, AZ 85721
e-mail: lega,goriely@ math.arizona.edu
March 9, 1999
Abstract
Two coupled nonlinear Klein-Gordon equations modeling the three-dimensional dynamics of a twisted
elastic rod near its rst bifurcation threshold are analyzed. First, it is shown that these equations are
Hamiltonian and that they admit a 2-parameter family of traveling wave solutions. Second, special
solutions corresponding to simple deformations of the elastic rod are considered. The stability of such
congurations is analyzed by means of two coupled Nonlinear Schrodinger equations, which are derived
from the nonlinear Klein-Gordon equations in the limit of small deformations. In particular, it is shown
that periodic solutions are modulationally unstable, which is consistent with the looping process observed
in the writhing instability of elastic laments. Third, numerical simulations of the nonlinear Klein-Gordon
equations suggesting that traveling pulses are stable, are presented.
KEYWORDS: Elastic Rods, Amplitude equations, Nonlinear Klein-Gordon equations, Nonlinear Schro-
dinger equations.
1 Introduction
The study of elastic deformations in rods has a long tradition in mathematics, physics and engineering, dating
back to Euler. The main challenge is to understand the dierent possible shapes elastic rods can conform
into under a variety of external stresses. The classical approach to solve such problems is to consider the
dierent static equilibria of rods and study through variational principles the bifurcation between them [1, 2].
The dynamical problem, namely how dierent congurations evolve in time or change to one another is not
usually addressed. Another, closely related, classical problem in the theory of elasticity is the propagation of
waves in elastic media. In thin elastic rods, it is known that two types of waves can exist: they are exural
and torsional waves and propagate variation of curvature and twist along the rod. Roughly speaking, the
former can be obtained by suddenly moving one end of a long straight rod while the latter can result from
a sudden twist of one end of the lament. Both aspects, structural deformations under stress and wave
propagation, can be studied together by considering the dynamics of elastic rods. A convenient setting for
such a study is provided by the Kirchho model for thin elastic laments.
In this paper we study some aspects of the dynamics of elastic rods. As shown in [3], the Kirchho model
can be used to analyze the stability (with respect to change of tension or twist) of straight twisted rods
under tension. Near the rst bifurcation point, that is when straight twisted laments lose their stability,
the rod dynamics may be described by two amplitude equations, which were derived by Goriely and Tabor in
[3]. These equations take the form of two nonlinear Klein-Gordon equation describing the coupling of twist
density to amplitude of deformation. This paper is devoted to an analysis of these amplitude equations,
and is organized as follows. In Section 2, we briey describe the Kirchho model. The amplitude equations
for the near-threshold dynamics are given in Section 3, where we also discuss their scalings, symmetries
and Hamiltonian nature. Section 4 is devoted to the analysis of traveling wave solutions and to simple
congurations of the elastic lament. Section 5 discusses the reduction of the coupled Klein-Gordon equations
to two coupled nonlinear Schr odinger equations, which are valid in the limit of small deformations. Solutions
to the latter system are then compared to those of the former, and the stability of some simple traveling
1
wave solutions is thereby inferred. Section 6 shows numerical simulations of the amplitude equations, which
suggest that pulses may propagate in a stable fashion. Conclusions are given in Section 7.
2 The Kirchho model
The Kirchho model describes the dynamics of a free elastic lament in three dimensions. Assuming that the
lament is thin (its cross section is much smaller than its length) and weakly bent (its curvature radius is, at
all points, much shorter than its length), a one-dimensional theory can be derived where forces and moments
are averaged over cross sections perpendicular to the central axis of the lament. This allows to describe the
dynamics in terms of two independent variables, namely time and arc-length (respectively, x and t). The
forces and moments exerted on a cross section of the lament are then written in a local basis, attached
to the central axis XXX = XXX(x, t), which is similar to the Frenet frame. This director basis {d
1
, d
2
, d
3
} is
a right-handed orthonormal basis built by taking d
3
as the tangent vector to the axis, d
3
(x, t) = dXXX/dt,
and d
1
as the vector following the elastic twist in the plane normal to the central axis. Since the curve
XXX = XXX(x, t) can be obtained by integrating the tangent vector d
3
(x, t), i.e. XXX(x, t) =
_
s
d
3
(s, t)ds, the
space and time evolution of the director basis species the kinematics of the lament.
As a consequence of the orthonormality of the director basis {d
1
, d
2
, d
3
}, we can write
d

i
= d
i
,

d
i
= d
i
i = 1, 2, 3. (1)
where

( ) and ( )

stand, respectively, for derivatives with respect to time and arc-length. The vectors and
are called the twist and spin vectors. They can be written as =

3
i=1

i
d
i
and =

3
i=1

i
d
i
. The
dynamics of the lament depends on the total force and moment, F = F(s, t) and M = M(s, t), experienced
by the elastic rod. In terms of the director basis, these vectors read F =

3
i=1
f
i
d
i
, M =

3
i=1
M
i
d
i
. The
Kirchho model combines conservation of linear and angular momentum with the constitutive relationship of
linear elasticity, which relates moments to strains, characterized by the twist vector. For a naturally straight
rod (i.e., without intrinsic curvature or twist) with circular cross-section, the scaled Kirchho equations read
[4, 5]:
F

=

d
3
, (2.a)
M

+d
3
F = d
1

d
1
+d
2

d
2
, (2.b)
M =
1
d
1
+
2
d
2
+
3
d
3
, (2.c)
where = 1/(1+) ( is the Poisson ratio) measures the ratio between the bending and twisting coecients
of the rod. In typical elastic material the Poisson ratio varies between
1
2
and 0, that is, varies between
2
3
and 1. Equations (2), together with (1), can be reduced to a set of 9 second order equations in space and
time, for the 9 unknowns (, , F).
3 Amplitude equations, scalings and Hamiltonian formalism
3.1 Amplitude equations
The stability of a given stationary lament parameterized by (
(0)
,
(0)
, f
(0)
) can be obtained by expanding
all quantities up to rst order in a small perturbation parameter and requiring that the basis remains
orthornormal to each order [6]:
d
i
= d
(0)
i
+ d
(0)
i
+O(
2
), (3.a)
=
(0)
+
_
(
(1)
)

+
(0)

(1)
_
+O(
2
), (3.b)
=
(1)
+O(
2
), (3.c)
F =

i
_
f
(0)
i
+
_
f
(1)
i
+ ( f
(0)
)
i
__
d
(0)
i
+O(
2
). (3.d)
Inserting this expressions in the Kirchho equations and collecting the terms to rst order in , a linear
equation can be obtained for m = (
(1)
, f
(1)
):
L
E
(
(0)
, f
(0)
).m = 0, (4)
2
where L
E
is a linear, second-order dierential operator in s and t, whose coecients depend on s through the
unperturbed solution (
(0)
, f
(0)
). Its explicit form can be found in [7]. For the case of the twisted straight
rod under tension, the stationary solution is simply given by

(0)
= (0, 0, ) , f
(0)
=
_
0, 0, P
2
_
. (5)
Solutions to (4) can be expressed as the sum of fundamental modes of the form:
m = e
t
_
A
n
e
inx
+A

n
e
inx
_
, (6)
where C
6
and the growth rate is related to the mode n by the dispersion relation, (, n) = 0,
obtained by substituting (6) into (4). The neutral curves are the curves in parameter space for which
stationary solutions exist. They are obtained by solving (0, n) = 0, which reads:
_

2
n
2
_
_
_

2
( 1) P
2
n
2
_
2

2
( 2)
2
n
2
_
= 0. (7)
The analysis of this dispersion relation reveals that the straight innite twisted pulled rod becomes unstable
for
c
= 2P/. This relation is known in elasticity theory as Loves criterion. For this value, a stationary
solution can be written as
m = B
0
+A
nc
e
incx
+A

nc
e
incx
+. . . , (8)
where the dots stand for higher order corrections and n
c
= P(2 )/,

0
= (0, 0, 1, 0, 0, 1) ,
nc
=
_
1, i, 0, iP
2
, P
2
, 0
_
. (9)
For xed P, the straight rod becomes unstable at the critical twist
c
and evolves towards a conguation
which, to rst order, is a helical lament and reads
XXX =
_
x,
2A
P
cos xP,
2A
P
sin xP
_
+. . . . (10)
Near threshold, the nonlinear dynamics of the lament is described by the following amplitude equations
[7]:
P
2
+ 1
P
2

2
A
t
2


2
A
x
2
= PA
_
E 2P|A|
2
+
B
x
_
, (11.a)
2

2
B
t
2


2
B
x
2
= 2P
|A|
2
x
. (11.b)
In this nonlinear analysis, B = B(x, t) and A = A(x, t) represent, respectively, the slowly-varying amplitudes
of the axial twist and the unstable helical mode. For given solutions A and B of (11), the shape and motion
of the lament in space is obtained by integrating the tangent vector. To second order, the lament shape
is given by:
XXX(x, t) =
_
_
_
_
_
_
_
_
(1 2 |A|
2
)dx
_
[cos(Px) (2 Re(A) +B Im(A)) sin(Px) (2 Im(A) B Re(A))] dx
_
[sin(Px) (2 Re(A) +B Im(A)) + cos(Px) (2 Im(A) B Re(A))] dx
_
_
_
_
_
_
_
. (12)
The rest of this paper is devoted to the analysis of these solutions and its consequences for the dynamics of
rods.
3.2 Scalings
To keep the discussion as general as possible and to reduce the number of free parameters, it is convenient
to re-write the amplitude equations in terms of scaled quantities. If we make the changes of variables
3
t =
_
2

L t

, x = Lx

, A =

P
2
+ 1
(2LP
2
)
A

, B =
(P
2
+ 1)
(2P
3
L)
B

, (13)
Eqns. (11) become, after dropping the primes,

2
A
t
2
c
2
0

2
A
x
2
= A|A|
2
A+
B
x
A, (14.a)

2
B
t
2


2
B
x
2
=
|A|
2
x
, (14.b)
where
c
2
0
=
2P
2
(P
2
+ 1)
and =
2EP
3
L
2
P
2
+ 1
. (15)
Only two parameters are left. The rst one, , measures the linear growth rate of perturbations to the
solution A=0 in the absence of B. The second one, c
0
, is the ratio v
A
/v
B
of the group velocities v
A
and
v
B
describing the speed at which perturbations of A and B propagate. These scaled equations provide a
convenient way of analyzing the dynamics and the solutions of Eqns. (11), and most of the discussion below
will refer to them. The new space scale depends on L, a characteristic length scale of the problem. For a
nite rod, we take L as the total length of the elastic lament. In what follows, we refer to these equations as
the coupled nonlinear Klein-Gordon equations (CNLKG). Similar coupled nonlinear Klein-Gordon equations
appear in a variety of elds such as plasma physics (See [8]; p.318), partial dierence equations [9], baroclinic
instabilities [10], or more generally as amplitude equations for dispersive systems [11].
3.3 Hamiltonian form
From the scaled form (14), it is straightforward to see that the coupled Klein-Gordon equations can be
written in Hamiltonian form
A
t
=
H
u
,
u
t
=
H

A
, (16.a)
B
t
=
H
v
,
v
t
=
H
B
, (16.b)
where refers to the Frechet derivative,
u =
A
t
, v =
B
t
, (17)
and H =
_
h(x, t) dx with
h(x, t) = c
2
0

A
x

2
|A|
2
+
1
2
|A|
4
|A|
2
B
x
+

A
t

2
+
1
2
_
B
x
_
2
+
1
2
_
B
t
_
2
. (18)
The integral is taken over the real line if h(x, t) is L
1
[(, +)]. If A does not vanish at innity, the
integration is taken on a suitable box (for instance, one period of the solution for periodic solutions).
3.4 Symmetries
Equations (14) have a few symmetries that can be used to simplify our analysis of special solutions. Below,
we assume that a
0
(s, ) and b
0
(s, ) are solutions to (14). The solutions A(x, t) and B(x, t) are obtained by
applying the indicated symmetries to a
0
and b
0
.
1. Space translation invariance:
A(x, t) = a
0
(x +x
0
, t), B(x, t) = b
0
(x +x
0
, t), (19)
where x
0
is a real arbitrary constant.
4
2. Time translation invariance:
A(x, t) = a
0
(x, t +t
0
), B(x, t) = b
0
(x, t +t
0
), (20)
where t
0
is a real arbitrary constant.
3. Uniform shift on B
A(x, t) = a
0
(x, t), B(x, t) = b
0
(x, t) +, (21)
where is a real arbitrary constant. This transformation corresponds to a uniform rotation of the local
frame of reference (d
1
, d
2
, d
3
) about the tangent vector to the elastic rod d
3
, and is not physically
relevant.
4. Gauge invariance:
A(x, t) = a
0
(x, t) exp(i), B(x, t) = b
0
(x, t), (22)
where is a real arbitrary constant. This transformation corresponds to a uniform translation of the
periodic solution of envelope A and is due to the invariance of the Kirchho equations with respect to
space-time translations before bifurcation.
5. Space reection symmetry
A(x, t) = a
0
(x, t), B(x, t) = b
0
(x, t). (23)
6. Time reection symmetry
A(x, t) = a
0
(x, t), B(x, t) = b
0
(x, t), (24)
which is due to the fact that the original system is time reversible.
7. Dilation invariance:
A(x, t) = a
0
(x, t) , B(x, t) = b
0
(x, t) +

2
1
x, (25)
where is a real arbitrary constant. This transformation is due to the fact that adding a linear term
in x to B renormalizes the growth rate , which in turn can be changed back to its initial value by
scaling space, time, and the envelopes A and B.
Whereas the Hamiltonian is preserved by transformations 1 to 6, the dilation 7 does not conserve H.
Since the amplitude equations (11) have been derived by assuming that the solution m dened in Section 2
remains bounded at innity, we will set so that B tends to a constant at innity. This condition turns out
to be necessary for H to remain bounded.
4 Traveling waves solutions of CNLKG
We now look for a 2-parameter family of solutions of the form:
A = a()e
it
, B = b(), (26)
where = x ct. This family includes periodic stationary solutions (c = = 0), oscillatory periodic
solutions (c = 0), homogeneous solutions (a, b constant, c = = 0), plane waves (b, and |A| constant),
non-oscillating traveling waves solutions and pulses ( = 0). The general case where both c and do not
vanish correspond to traveling waves. These solutions periodically oscillate in time as they travel at constant
speed c. Together with (26), Eqns. (14) read (where primes denotes -derivatives)
(c
2
c
2
0
)a

+ 2ica

= a(
2
+ |a|
2
+b

), (27.a)
(c
2
1)b

= (|a|
2
)

. (27.b)
5
The last equations can be integrated once to yield (c
2
1)(b

K
1
) = |a|
2
. We then use the expression for
b

to simplify (27.a)
(c
2
c
2
0
)a

+ 2ica

= a
_

2
+ +K
1

c
2
(c
2
1)
|a|
2
_
(28)
In order to obtain a set of real dierential equations, we set a = R()e
i()
, which gives
(c
2
c
2
0
)R

= R
_

2
+ +K
1
2c + (c
2
c
0
)
2

c
2
(c
2
1)
R
2
_
, (29.a)
2
_

c
(c
2
c
2
0
)
_
R

+R

= 0, (29.b)
where

= . The system for (R, ) has two rst integrals:

H =
1
2
(R

)
2
+
1
2
_

( +K
1
+
2
)
c

_
R
2
+
c
2
R
4
4(c
2
1)(c
2
c
2
0
)
, (30.a)
G =
_
(c
2
c
2
0
)c
_
R
2
. (30.b)
The rst integral G is reminiscent of Keplers second law for the motion of a particle in a central force eld.
It can be used to simplify H as follows:
H
1
= H +G

2
(c
2
+c
2
0
) ( +K
1
)(c
2
c
2
0
)
2c(c
2
c
2
0
)
2
=
1
2
(R

)
2
+
G
2
2(c
2
c
2
0
)
2
R
2
+

2
c
2
0
( +K
1
)(c
2
c
2
0
)
2(c
2
c
2
0
)
2
R
2
+
c
2
4(c
2
1)(c
2
c
2
0
)
R
4
.
(31)
Thus, H
1
is an Hamiltonian describing the motion of a particle in a one-dimensional eective potential
V
e
= H
1
(R

)
2
/2 = v
2
R
2
+v
2
R
2
+v
4
R
4
. This reduced Hamiltonian system can be solved through the
change of variables
= , z = v
4
R
2
, (32)
which transforms the Hamiltonian H
1
into the standard form:
_
dz
dt
_
2
=
4
z
3
z
1
(z z
1
)(z
2
z)(z
3
z), (33)
where
1
=
_
2(z
3
z1) and
z
1
+z
2
+z
3
= v
2
, z
1
z
2
+z
2
z
3
+z
3
z
1
= v
4
H
1
, z
1
z
2
z
3
= v
2
4
v
2
2
. (34)
The general solution of (33) is z = z
1
+ (z
2
z
1
)sn
2
(|k) where k
2
= (z
2
z
1
)/(z
3
z
1
). That is,
R =

z
1
+ (z
2
z
1
)sn
2
(|k)
v
4
. (35)
The conditions for the roots z
1
, z
2
, z
3
to be real is:
27v
4
2
v
2
4
+ 2H
1
(9v
2
v
2
2
2H
2
1
)v
4
v
2
2
H
2
1
+ 4v
3
2
v
2
2
< 0. (36)
The rst integral G can then be used to give an explicit form of the angular coordinate in terms of
incomplete elliptic integrals of the third kind. Explicit forms for the periods of given periodic orbits can be
obtained in a similar fashion.
To understand the dierent types of orbits that the system H
1
may exhibit it is easier to study the
potential V
e
directly. There are two main cases that can be distinguished depending on whether G vanishes
6
v
2
>0
v
4
>0
v
2
<0
v
4
>0
v
2
>0
v
4
<0
v
2
< 0
v
4
< 0
R
0
a. b.
c. d.
Figure 1: Dierent shapes of the eective potential V
e
as a function of R when G = 0
v
2
>0
v
4
>0
a. b.
c. d.
v
2
<0
v
4
>0
v
2
>0
v
4
<0
v
2
< 0
v
4
< 0
R
0
R
0
=0
Figure 2: Dierent shapes of the eective potential V
e
as a function of R when G = 0
7
0.7
0.8
0.9
1
1.1
1.2
1.3
-3 -2 -1 0 1 2 3
R
Figure 3: A traveling hole solution (v
2
= 4, v
4
= 1, v
2
= 5, c
0
= 3, = 0)
or not. In each cases, four dierent potential shapes can be obtained (depending on the sign of the coecients
v
2
, v
4
). These potentials are shown, respectively, on Fig. 1 and Fig. 2.
We now focus on localized solutions, that is solutions for which the amplitudes A and B tend asymp-
totically to a constant and vary rapidly in a nite region. At the level of the Hamiltonian system with
Hamiltonian H
1
, we look for homoclinic and heteroclinic orbits connecting the xed point R
0
to himself
(homoclinic case) or to R
0
(heteroclinic case). In both cases, we choose the constant K
1
to ensure that
lim

= 0, that is K
1
= R
2
0
/(c
2
1). Analytic expressions will be given with a minimum number of
free parameters. More general solutions can be obtained by using the symmetries discussed in Section 3.4.
4.1 Traveling holes
If G = 0, the only possible localized solutions occur for v
2
> 0 and v
4
< 0. These orbits are, for the
Hamiltonian H
1
, homoclinic to the xed point R
0
(See Fig. 1c). In this case, one sets 0 < z
1
< z
2
= z
3
=
v
4
R
2
0
in (35) to obtain:
R
2
= R
2
0

2
sech
2
(), (37)
where

2
= 6v
4
R
2
0
2v
2
, and
2
=
(3v
4
R
2
0
+v
2
)
v
4
. (38)
The xed point R
0
is the largest root of 2z
6
+ v
2
z
4
v
2
2
= 0. The homoclinic orbit exists when H
1
=
R
2
0
(2v
2
+3v
4
R
2
0
). Explicit forms of = () and b = b() can be obtained in terms of elementary functions.
A typical traveling hole amplitude prole is shown in Fig. 3.
8
Re(A)
B
| A|
x
Figure 4: A traveling pulse ( = 1/4, c = 2, c
0
=

2, = 1/4, = 3/4)
c
Figure 5: Filament corresponding to the homoclinic solution of Figure 3.
4.2 Traveling pulses
Traveling pulses exist when G = 0, v
2
< 0 and v
4
> 0 (Fig. 2b). They correspond to orbits of the reduced
Hamiltonian system which are homoclinic to the xed point R
0
= 0, at which H
1
= 0. They read
a = sech() exp
_
i
c
c
2
c
2
0

_
. (39.a)
b =

2
(1 c
2
)
tanh(), (39.b)
where

2
=
2(c
2
1)
c
2
(c
2
c
2
0
)
_
(c
2
c
2
0
)
2
c
2
0
_
and
2
=
(c
2
c
2
0
)
2
c
2
0
(c
2
c
2
0
)
2
. (40)
This solution exists whenever
2
> 0 and
2
> 0. Typical proles of A and B are shown in Fig. 4, and the
corresponding lament is depicted in Fig. 5.
4.3 Fronts
There exists yet another localized solution when G = 0. If v
2
> 0 and v
4
< 0 the Hamiltonian H
1
accommodates heteroclinic orbits (See Fig. 2d), given by
a = tanh() exp
_
i
c
c
2
c
2
0

_
, (41.a)
b =

2
(c
2
1)
tanh(), (41.b)
9
Re(A)
B
| A|
x
Figure 6: A heteroclinic solution ( = 1/2, c = 6/5, c
0
= 3/2, = 10, = 3/4).
c
Figure 7: The lament corresponding to the front solution of Figure 5.
where

2
=
(c
2
1)((c
2
c
2
0
)
2
c
2
0
)
(2c
2
1)(c
2
c
2
0
)
and
2
=
c
2
(
2
c
2
0
(c
2
c
2
0
))
2(2c
2
1)(c
2
c
2
0
)
2
. (42)
Such a solution, referred to as a front, is depicted in Figures 6 and 7.
4.4 Stationary periodic solutions
We now consider a simple class of stationary solutions to Eqns. (14), which are periodic in x. They read
A = sin(kx), B = sin(2kx), (43)
where

2
= 2(k
2
c
2
0
), =
1
2k
(k
2
c
2
0
). (44)
Since (43) solves the scaled coupled Klein-Gordon equations, x varies between 0 and 1. We also assume that
k is a multiple of n in order to have the same values of A and B at x = 0 and x = 1. In terms of solutions to
the Kirchho equations, (43) corresponds to a helix whose envelope is modulated and vanishes periodically
with period 2L/k. Typical amplitude proles are given in Fig. 8, and the corresponding lament is shown
in Fig. 9.
5 Analysis of CNLKG in the limit of small deformations
In the limit of small deformations, the coupled nonlinear Klein-Gordon equations (14) can be reduced to
two coupled nonlinear Schr odinger (NLS) equations, which may then be used to investigate the stability of
some of the localized solutions discussed above. In what follows, we derive these coupled NLS equations and
compare some of their solutions to those of CNLKG.
10
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
-3 -2 -1 1 2 3
B
A
Figure 8: A Periodic solution ( = 6/10, k = 1, P = 10, = 3/4)
Figure 9: The stationary periodic solution corresponding to Figure 7.
5.1 Reduction to two coupled nonlinear Schr odinger equations
The translational and gauge invariances of equations (14) may lead to large-scale instabilities (also called
modulational instabilities), which play an important role in the dynamics of the system. As was shown in
[12] for the case of a single wave equation and for solutions which are periodic in time, these instabilities can
be analyzed in terms of envelope equations. The latter are derived from the coupled Klein-Gordon equations
(14) by multiple scales analysis. To understand the scalings involved in the derivation, consider the periodic
solutions of CNLKG given by
A = Re
ikx
, k
2
=

c
2
0

2
2c
2
0
. (45)
Near threshold ( 0), the wave vector of solution (45) can be written as
k =

c
0
+q, (46)
where q is small compared to

/c
0
. To order q, the coecients and satisfy

2
= 4qc
0

+O(q
2
), = qc
2
0
+O(q
2
). (47)
This suggests to develop a multiple scales analysis where q and are of order
2
, and is of order . The
scale of q implies that the slow space scale is given by X
2
=
2
x. We take the slow time scale as T
1
= t and
write
A = A
1
(x, T, X
2
) +
2
A
2
(x, T, X
2
) +
3
A
3
(x, T, X
2
) +. . . (48.a)
B = B
1
(x, T, X
2
) +
2
B
2
(x, T, X
2
) +
3
B
3
(x, T, X
2
) +. . . (48.b)
11
where T = t, X
2
=
2
x, and is a small parameter. As shown in Appendix A, the amplitudes A and B can
be written, to order
3
, as
A = F
1
(x, t)e
_
i

c
0
x
_
+F
2
(x, t)e
_
i

c
0
x
_
, (49.a)
B = +
ic
0
2

_
F
1
(x, t)F

2
(x, t)e
_
2i

c
0
x
_
F
2
(x, t)F

1
(x, t)e
_
2i

c
0
x
_
_
. (49.b)
The envelopes F
1
and F
2
satisfy two solvability conditions obtained at order
3
(see Appendix A). They take
the form of two coupled nonlinear Schr odinger equations for F
1
and F
2
, which read

2
F
1
t
2
+ 2ic
0

F
1
x
+|F
1
|
2
F
1
+|F
2
|
2
F
1
= 0, (50.a)

2
F
2
t
2
2ic
0

F
2
x
+|F
2
|
2
F
2
+|F
1
|
2
F
2
= 0. (50.b)
These equations, known as the Manakov equations [13, 14], have the symmetry F
1
F

2
and F
2
F

1
.
As shown in [14], they are completely integrable and admit a Lax Pair. Their solitary wave solutions are
classied in [15] and various perturbations of these equations have been considered for the description of
light propagation in birefrengent bers [16] or counterpropagation in Kerr media [17, 18] (see also [19] and
[20] as well as references therein).
5.2 Special solutions of the reduced equations
We now consider particular solutions to the two coupled nonlinear Schr odinger equations (50) and compare
them to similar solutions of the coupled Klein-Gordon equations.
5.2.1 Plane waves
Plane wave solutions to (50) of the form
F
1
=
_

2
+ 2c
0

p exp[i(px +t)], F
2
= 0, (51)
correspond to plane wave solutions of CNLKG
A = Rexp
_
i
_
px +t

x
c
0
__
, B = 0. (52)
Indeed, R
2
in (52) is given by R
2
=
2
+ 2c
0
p

c
2
0
p
2
=
2
+ 2c
0
p

+ O(p
2
), which is consistent with
the scaling p = O(
2
) and the fact that expressions (48) are valid at order
3
.
Since plane wave solutions of (50) are modulationally unstable, plane wave solutions of CNLKG are also
unstable. As a consequence, traveling holes and fronts of CNLKG are unstable since their asymptotic states,
which are plane wave solutions, are unstable.
5.2.2 Stationary solutions
Stationary solutions to (50) with F
1
= Rexp(iqx) and F
2
= F

1
are such that R
2
= c
0
q

. In terms of
A and B, and assuming = 0 (which can always be done by virtue of Symmetry 3 of Section 3.4), these
solutions read
A = cos
__

c
0
q
_
x
_
+O(
4
), (53.a)
B = sin
_
2
_

c
0
q
_
x
_
+O(
4
), (53.b)
where

2
= 4 c
0
q

, = c
2
0
q. (54)
12
If we translate this solution by a quarter of its period (which again is allowed by Symmetry 1 of Section
3.4), we obtain
A = cos
__

c
0
q
_
x +

2
_
+O(
4
) = sin
__

c
0
q
_
x
_
+O(
4
), (55.a)
B = sin
_
2
_

c
0
q
_
x +
_
+O(
4
) = sin
_
2
_

c
0
q
_
x
_
+O(
4
). (55.b)
This is exactly the near-threshold expansion (46, 47) of the periodic solutions (45). Therefore, the latter
correspond to solutions of constant amplitude of Eqns. (50), with F
2
= F

1
. Since, with F
2
= F

1
, the two
coupled equations (50) reduce to the self-focusing nonlinear Schr odinger equation

2
F
1
t
2
+ 2ic
0

F
1
x
+ 2|F
1
|
2
F
1
= 0, (56)
the periodic solutions (45) are modulationally unstable near threshold. This suggests that amplitude defor-
mations will focalize in nite time, and is consistent with the results obtained in [21] for the linear stability
analysis of the innite helix which appears above the writhing bifurcation of a twisted elastic rod.
5.2.3 Traveling pulses
Soliton solutions of (50) of the form
F
1
=
_
2
c
0
c
sech
_
c
0

c
2
(x ct)
_
exp
_
i
c
0

c
t
_
, F
2
= 0, (57)
correspond to solutions of (48) with
A = sech
_

(x ct)
_
exp
_
i
_

k(x ct) + t
__
, B = 0, (58)
and
=
c
0

2
c
,

=

c
0
c
2
,

k =

c
0
, =

c
2
0
c
2
c
0
c
. (59)
Since |A|
_
2
c
0
c
must be small for the coupled nonlinear Schr odinger equations to be valid, solution (57)
are valid in the limit of large c. More precisely, since c
0
and

are nite, c should be such that c

1
= O().
We now compare this solution to traveling pulse solutions of CNLKG. The latter are of the form:
A = sech ((x ct)) exp [i (k(x ct) +t)] , (60.a)
B =

2
(c
2
1)
tanh [(x ct)] , (60.b)
where

2
= 2
(c
2
1)
c
2
2

2
(c
2
1)c
2
0
c
2
(c
2
c
2
0
)
,
2
=

c
2
c
2
0

c
2
0

2
(c
2
c
2
0
)
2
(61.a)
k =
c
c
2
c
2
0
. (61.b)
With = =

c
2
0
c
2
c
0
c
, the last equation becomes k =

c
0
=

k, and
2
and
2
read

2
= 2
c
2
0
c
2
_
1
1
c
2
_
=
2
+O(
1
c
4
) =
2
+O(
4
), (62.a)

2
=
c
2
0

c
4
+O(
1
c
6
) =

2
+O(
6
). (62.b)
13
Figure 10: Shape of the lament corresponding to a double soliton solution of the coupled nonlinear
Schr odinger equations. The parameters used to plot this gure, given in terms of the unscaled equations,
are = 0.75, P = 3, E= 0.1, L = 40, = 0, and c = 2.
In other words, the expressions of
2
,
2
and
2
and

2
are consistent to lowest order. The fact that the
nonlinear Schr odinger equations give B = 0 instead of B =

2
(c
2
1)
tanh [(x ct)] can be understood as
follows. Traveling pulse solutions of CNLKG are found by solving a dierential equation for A and then by
solving Eqn. (27.b) for b. This amounts to solving
B

=
dB
d
=
|A|
2
c
2
1
=

2
c
2
1
1
cosh
2
()
, (63)
where = x ct. But with the expression of
2
, we have

2
c
2
1
=
2c
2
0
c
4
= O(
4
), (64)
so that B

= 0 at order
3
, i.e. B = = constant at order
3
. Therefore, we conclude that soliton solutions
(57) of the coupled nonlinear Schr odinger equations correspond to traveling pulses (39) of CNLKG.
5.2.4 Double solitons
The coupled nonlinear Schr odinger equations (50) also admit solutions of the form
F
1
=

c
0
c
sech
_
c
0

c
2
(x ct)
_
exp
_
i
c
0

c
t
_
, F
2
= F
1
, (65)
which correspond, in terms of A and B, to
A = 2

c
0
c
sech
_
c
0

c
2
(x ct)
_
cos
_

c
0
_
x
c
2
0
c
t
__
, (66.a)
B =

c
3
0
c
2
sech
_
c
0

c
2
(x ct)
_
sin
_
2

c
0
_
x
c
2
0
c
t
__
. (66.b)
Figure 10 shows a reconstructed lament with A and B given above.
6 Numerical results
Following the stability analysis presented in this paper, we conclude that traveling holes, fronts and periodic
solutions of the coupled Klein-Gordon equations are unstable, at least in systems which are large enough
for the modulational instability to develop. On the other hand, traveling pulses are unstable if is positive
since their asymptotic states (A = 0, B = ) experience exponential growth, but may be stable for negative
values of . In this section, we show results of numerical simulations of the original (unscaled) coupled Klein-
Gordon equations (11), which suggest that traveling pulses are stable in some parameter regimes. In order to
14
Figure 11: Numerical simulation of the original (unscaled) nonlinear Klein-Gordon equations, showing the
stable propagation of a traveling pulse. Parameters used in the simulation are = 0.75, P = 3, E = 0.1,
L = 40, = 0, and c = 0.6124.
study traveling waves, we designed a code with non-reecting [22] boundary conditions, which minimize the
amount of reection at the boundaries. The code is second order in time (a typical time-step is dt = 0.01)
and uses spectral-like nite dierences [23] (a typical box length value is L = 40, which corresponds to a
mesh size dx = 0.1). Although no symplectic structure is implemented in the numerics, we checked that the
Hamiltonian
H =
_
dx
_
P
2
P
2
+ 1

A
x

2
+
P
4
P
2
+ 1
|A|
4

P
3
P
2
+ 1
|A|
2
_
B
x
+E
_
+

A
t

2
+
P
2
4(P
2
+ 1)
_
B
x
+E
_
2
+
P
2
2(P
2
+ 1)
_
B
t
_
2
_
, (67)
is conserved to within 0.001% of its value. The simulation has a graphical interface, built with the Advanced
Visual Software AVS 5, which allows for interactive change of the parameters.
Figure 11 illustrates the stable propagation of a traveling pulse (see (39)), which, in terms of unscaled
15
variables, reads
A =

P
2
+ 1
2LP
2
sech
_

L
_
x c
_

2
t
__
exp
_
i
_
k
L
_
x c
_

2
t
_
+

L
_

2
t
__
, (68.a)
B =

2
(c
2
1)
P
2
+ 1
2LP
3
tanh
_

L
_
x c
_

2
t
__
, (68.b)
where

2
=
2(c
2
1)
c
2
(c
2
c
2
0
)
_
(c
2
c
2
0
)
2
c
2
0
_
,
2
=
(c
2
c
2
0
)
2
c
2
0
(c
2
c
2
0
)
2
, (69.a)
c
2
0
=
2

P
2
P
2
+ 1
, =
2EP
3
L
2
P
2
+ 1
, k =
c
c
2
c
2
0
. (69.b)
The parameter values are given in the caption of Fig. 11. We only show the reconstructed lament, whose
shape is computed from the numerical values of A and B according to formula (12) which is accurate to
second order in the perturbation parameter.
Our numerical results therefore suggest that traveling pulses are likely to propagate in a stable fashion, at
least in some parameter regime. A complete answer to this question would require an analysis of the space-
and time-dependent Klein-Gordon equations linearized about a traveling pulse, and is beyond the scope of
this paper.
7 Conclusions
This paper gives a detailed description of the dynamics of the coupled nonlinear Klein-Gordon equations,
which model an elastic lament subjected to external twist in the vicinity of the rst bifurcation threshold.
It is shown rst that these equations have only two relevant parameters, namely the linear growth rate of
amplitude deformations and the ratio of the group velocities of amplitude and twist modulations, and second,
that they are Hamiltonian. A complete description of traveling wave solutions to these equations is presented,
and localized solutions are identied. Because such solutions are of physical interest, the question of their
stability must be addressed. This is accomplished by means of two coupled nonlinear Schrodinger equations,
valid in the limit of small amplitude deformations. It turns out that these equations, known as Manakovs
equations, are integrable and arise as a limiting case of model equations describing propagation of light in
birefringent bers or counterpropagating beams in Kerr media. Moreover, because Manakovs equation are
defocusing, their plane wave solutions are modulationally unstable and, as a consequence, traveling holes
and fronts of the coupled Klein-Gordon equations have unstable asymptotic states. Interestingly, periodic
solutions, which correspond to modulated helical laments, are also proven to be unstable, and this self-
focusing instability is consistent with the looping process observed in elastic rods subject to excessive twist.
It was shown in [24] that, when made unstable, innite twisted rods tend to rst localize deformations and
then form twisted loops (eventually giving rise to a braided structures). The analysis of the periodic solutions
of CNLKG therefore suggests that the formation of loops in nite rods results from a self-focusing instability.
Finally, our numerical simulations show that traveling pulses are stable in some parameter regime, but only
if the system is kept below threshold.
Acknowledgments
The authors would like to thank N. Ercolani and M. Tabor for many interesting discussions. This work is
partly supported by Nato Collaborative Research Grant NATO-CRG 97/037.
A Derivation of the nonlinear Schr odinger equations
This section is devoted to a multiple scales analysis of time-independent periodic solutions to the scaled
coupled Klein-Gordon equations (14). As explained in Section 5, we set
A(x, t) = A
1
(x, T, X
2
) +
2
A
2
(x, T, X
2
) +
3
A
3
(x, T, X
2
) +. . . (70.a)
B(x, t) = B
1
(x, T, X
2
) +
2
B
2
(x, T, X
2
) +
3
B
3
(x, T, X
2
) +. . . (70.b)
16
where T = t, X
2
=
2
x, is a small parameter, and the dots stand for higher order corrections. After
substituting in (14) and equating like powers in , we get a hierarchy of partial dierential equations which
can be solved at each order. At order , we obtain
c
0
2

2
x
2
A
1
(x, T, X
2
) +A
1
(x, T, X
2
) = 0, (71.a)

2
x
2
B
1
(x, T, X
2
) = 0, (71.b)
which gives
A
1
(x, T, X
2
) = F
1
(T, X
2
)e

x
c
0
+F
2
(T, X
2
)e
i

x
c
0
, (72.a)
B
1
(x, T, X
2
) = , (72.b)
where F
1
and F
2
are two complex envelopes and is an arbitrary constant. Only the terms which do not
lead to divergence at innity have been kept in B
1
.
At order
2
, we get
A
2
(x, T, X
2
) +c
0
2

2
x
2
A
2
(x, T, X
2
) =
_

x
B
1
(x, T, X
2
)
_
A
1
(x, T, X
2
) = 0, (73.a)

2
x
2
B
2
(x, T, X
2
) =
_

x
A
1
(x, T, X
2
)
_
A

1
(x, T, X
2
) +A
1
(x, T, X
2
)

x
A

1
(x, T, X
2
). (73.b)
After substituting in the expressions for A
1
and its complex conjugate, we can solve for A
2
and B
2
and
obtain
A
2
(x, T, X
2
) = 0, (74.a)
B
2
(x, T, X
2
) =
ic
0

2
_
e
2
i

x
c
0
F

2
(T, X
2
)F
1
(T, X
2
) e
2
i

x
c
0
F
2
(T, X
2
)F

1
(T, X
2
)
_
. (74.b)
We do not include in A
2
and B
2
terms which are in the kernel of the linear operator obtained at order .
These terms have already been taken into account in the expressions of A
1
and B
1
.
At order
3
, the Klein-Gordon equations read
c
0
2

2
x
2
A
3
(x, T, X
2
) +A
3
(x, T, X
2
) (75.a)
=
_

2
T
2
F
1
(T, X
2
) +|F
1
(T, X
2
)|
2
F
1
(T, X
2
) + 2ic
0

X
2
F
1
(T, X
2
) +|F
2
(T, X
2
)|
2
F
1
(T, X
2
)
_
e
i

x
c
0
+
_

2
T
2
F
2
(T, X
2
) +|F
2
(T, X
2
)|
2
F
2
(T, X
2
) 2ic
0

X
2
F
2
(T, X
2
) +|F
1
(T, X
2
)|
2
F
2
(T, X
2
)
_
e
i

x
c
0
,

2
x
2
B
3
(x, T, X
2
) = 0. (75.b)
In order to avoid secular terms in A
3
, we need to impose the two solvability conditions

2
T
2
F
1
(T, X
2
) + |F
2
(T, X
2
)|
2
F
1
(T, X
2
) + 2 ic
0

X
2
F
1
(T, X
2
)
+ |F
1
(T, X
2
)|
2
F
1
(T, X
2
) = 0 (76.a)

2
T
2
F
2
(T, X
2
) + |F
1
(T, X
2
)|
2
F
2
(T, X
2
) 2 ic
0

X
2
F
2
(T, X
2
)
+ |F
2
(T, X
2
)|
2
F
2
(T, X
2
) = 0, (76.b)
which are two coupled nonlinear Schr odinger equations for the envelopes F
1
and F
2
. If we re-write these equa-
tions in terms of the original variables x and t and dene F

1
(x, t) = F
1
(t,
2
x) and F

2
(x, t) = F
2
(t,
2
x),
we get, after dropping the primes,

2
F
1
t
2
+ 2ic
0

F
1
x
+|F
1
|
2
F
1
+|F
2
|
2
F
1
= 0 (77.a)

2
F
2
t
2
2ic
0

F
2
x
+|F
2
|
2
F
2
+|F
1
|
2
F
2
= 0. (77.b)
17
If these equations are satised, we can then choose A
3
(x, T, X
2
) = 0 and B
3
(x, T, X
2
) = 0. To order
3
, the
solutions A and B are then given by
A(x, t) = F
1
(t, x)e

x
c
0
+F
2
(t, x)e
i

x
c
0
+O(
4
) (78.a)
B(x, t) = +
ic
0
2

_
e
2
i

x
c
0
F

2
(t, x)F
1
(t, x) e
2
i

x
c
0
F
2
(t, x)F

1
(t, x)
_
+O(
4
), (78.b)
where we have set

= and then dropped the prime.

References
[1] A. E. H. Love, A treatise on the mathematical theory of elasticity, Cambridge University Press, Cam-
bridge, 1892.
[2] S. S. Antman, Nonlinear problems of elasticity, Springer Verlag, New York, 1995.
[3] A. Goriely and M. Tabor, New amplitude equations for thin elastic rods, Phys. Rev. Lett. 77, 3537-3540
(1996).
[4] E. H. Dill, Kirchhos theory of rods, Arch. Hist. Exact. Sci. 44, 2-23 (1992).
[5] B. D. Coleman, E. H. Dill, M. Lembo, Z. Lu, and I. Tobias, On the dynamics of rods in the theory of
Kirchho and Clebsch, Arch. Rational Mech. Anal. 121, 339359 (1993).
[6] A. Goriely and M. Tabor, Nonlinear dynamics of laments I: Dynamical instabilities, Physica D 105,
20-44 (1997).
[7] A. Goriely and M. Tabor, Nonlinear dynamics of laments II: Nonlinear analysis, Physica D 105, 45-61
(1997).
[8] V. E. Zakharov, S. L. Musher and A. M. Rubenchik, Hamiltonian approach to the description of non-
linear plasma phenomena, Phys. Rep. 129, 286-366 (1985).
[9] A.C. Newell, Finite amplitude instabilities of partial dierence equations, SIAM J. Appl. Math. 33,
133-160 (1977).
[10] J. Pedlosky, Finite amplitude baroclinic disturbances in downstream varying currents, J. Phys. Oceanog-
raphy 6, 335-344 (1976).
[11] J. D. Gibbon and M. J. McGuiness, Amplitude equations at the critical points of unstable dispersive
physical systems, Proc. Roy. Soc. Lond. A 377, 185-219 (1981).
[12] E.J. Parkes, The modulational instability of the nonlinear Klein-Gordon equation, Wave Motion 13,
261-275 (1991).
[13] D.J. Benney and A.C. Newell, The propagation of nonlinear wave envelopes, J. Math. and Phys. 46,
133-139 (1967).
[14] S.V. Manakov, On the theory of two-dimensional stationary self-focusing of electromagnetic waves, Zh.
Eksp. Teor. Fiz. 65, 505-516 (1973) [Sov. Phys. JETP 38, 248-253 (1974)].
[15] J. Yang, Classication of the solitary waves in coupled nonlinear Schr odinger equations, Physica D 108,
92-112 (1997).
[16] D.J. Muraki and W.L. Kath, Hamiltonian dynamics of solitons in optical bers, Physica D 48, 53-64
(1991).
[17] W.J. Firth, A. Fitzgerald and C. Pare, Transverse instabilities due to counterpropagation in Kerr media,
J. Opt. Soc. Am. B 7, 1087-1097 (1990).
18
[18] M. Haelterman, A.P. Sheppard, and A.W. Snyder, Bimodal counterpropagating spatial solitary-waves,
Optics Communications 103, 145-152 (1993).
[19] O.C. Wright, Modulational instability in a defocusing coupled nonlinear Schr odinger system, Physica
D 82, 1-10 (1995).
[20] Y.A. Li and K. Promislow, Stability and Structural stability of non-ground state traveling waves of
coupled nonlinear Schr odinger equations, submitted to Physica D (1998).
[21] A. Goriely and M. Tabor, Spontaneous helix-hand reversal and tendril perversion in climbing plants,
Phys. Rev. Lett. 80, 1564-1567 (1998).
[22] R.L. Higdon, Radiation boundary conditions for dispersive waves, SIAM J. Numer. Anal. 31, 64-100
(1994).
[23] S.K. Lele, Compact nite dierence schemes with spectral-like resolution, J. Computational Physics
103, 16-42 (1992).
[24] A. Goriely and M. Tabor, Nonlinear dynamics of laments IV: Spontaneous looping of twisted elastic
rods, Proc. Roy. Soc. London (A) 45, 31833202 1998.
19