(SpringerBriefs in Physics) Christopher Chong, Panayotis G. Kevrekidis - Coherent Structures in Granular Crystals-Springer International Publishing (2018) PDF

SPRINGER BRIEFS IN PHYSICS
Christopher Chong
Panayotis G. Kevrekidis
Coherent Structures
in Granular Crystals
From Experiment
and Modelling to
Computation and
Mathematical Analysis
SpringerBriefs in Physics
Series editors
B. Ananthanarayan, Centre for High Energy Physics, Indian Institute of Science,
Bangalore, India
Egor Babaev, Amherst, MA, USA
Malcolm Bremer, Bristol, UK
Xavier Calmet, Department of Physics and Astronomy, University of Sussex,
Brighton, UK
Francesca Di Lodovico, London, UK
Pablo D. Esquinazi, Institute for Experimental Physics II, University of Leipzig,
Leipzig, Germany
Maarten Hoogerland, Auckland, New Zealand
Eric Le Ru, School of Chemical and Physical Sciences, Victoria University of
Wellington, Kelburn, Wellington, New Zealand
Hans-Joachim Lewerenz, Pasadena, CA, USA
James Overduin, Towson, MD, USA
Vesselin Petkov, Montreal, QC, Canada
Charles H.-T. Wang, Department of Physics, The University of Aberdeen,
Aberdeen, UK
Andrew Whitaker, Belfast, UK
Stefan Theisen, Max-Planck-Institut für Gravitationsphys, Potsdam, Germany
More information about this series at http://www.springer.com/series/8902
Christopher Chong Panayotis G. Kevrekidis
•
Coherent Structures
in Granular Crystals
From Experiment and Modelling
to Computation and Mathematical Analysis
123
Christopher Chong Panayotis G. Kevrekidis
Bowdoin College University of Massachusetts
Brunswick, ME Amherst, MA
USA USA
ISSN 2191-5423 ISSN 2191-5431 (electronic)

SpringerBriefs in Physics
ISBN 978-3-319-77883-9 ISBN 978-3-319-77884-6 (eBook)
https://doi.org/10.1007/978-3-319-77884-6
Library of Congress Control Number: 2018934912
© The Author(s) 2018

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
Printed on acid-free paper
This Springer imprint is published by the registered company Springer International Publishing AG
part of Springer Nature
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To my family for their continued support: For
my parents and sister, who helped me in an
uncountable number of ways throughout my
life. For my wife, who has given me an
unwavering support and so much happiness
and guidance since we met. And for my
children, who give me inspiration everyday.
Christopher Chong
To my father, George, for wanting (and

expecting) this book. This one is for you, dad.
Panayotis G. Kevrekidis
Preface
The topic of granular crystals has enjoyed a rich history over the past 30 years,
starting with the pioneering work of V. F. Nesterenko and his collaborators in the
1980s. A number of significant developments both physically and mathematically
took place in the 1990s, especially as regards the prototypical excitation of such
chains of beads, namely, traveling waves. The early 2000s led to a further booming
of interest fueled to a large degree by the development of novel experimental
techniques that could more controllably and reliably provide quantitative informa-
tion about the system (via suitable piezosensors). The 2001 authoritative book by
Nesterenko constitutes, still to this day, an excellent summary of the early stages
of the research on granular crystals. A 2008 Physics Reports review article by S. Sen,
another of the key contributors in this field, gave a concise summary of the later
developments. Around 2008, a new wave of contributions on the subject came along
providing a deeper understanding of different types of excitations (such as discrete
breathers and shock waves), and different types of setups (including heterogeneous
and random ones, ones involving mass-in-mass, mass-with-mass, branching,
intruder-based, and other types of configurations). Novel, noninvasive experimental
techniques (such as laser Doppler vibrometry) also allowed researchers to probe
granular crystals in unprecedented ways.
As this decade of significant and rapid developments is coming to a close, many
of the newer topics are becoming more well understood. Consequently, we felt
there was an opportunity to provide a brief, somewhat more mathematically minded
(with a broad array of computational touches) summary of a select, coherent set
of these developments, what motivated them, as well as where they stand presently
from the point of view of theory, computation, and experiment. In that vein, there
are two main goals of the present volume: One is to offer novices in granular
crystals (but also to some degree more seasoned researchers) a reference point for
the current state of the art (admittedly, with a considerable touch of personal flavor).
The other goal is to offer some insights and hopefully pave the way toward some
of the important future developments on this theme. We thus hope that it will be of
vii
viii Preface
use to advanced undergraduates, graduate students, and young researchers, but also
to our colleagues interested in the subject of nonlinear coherent structures and their
applications in materials science more broadly. Finally, we would much welcome
their thoughts and comments toward further improving its exposition in the future.
Brunswick, ME, USA Christopher Chong

Amherst, MA, USA Panayotis G. Kevrekidis
September 2017
Acknowledgements
This book would not have been possible without the help of our many collaborators
on this theme: Chiara Daraio, Mason Porter, George Theocharis, Nick Boechler, JK
Yang, Yannis Kevrekidis, Guillaume James, Alex Vakakis, Mark Ablowitz, Anna
Vainchtein, Atanas Stefanov, Surajit Sen, and many others, as well as of students
and young researchers that worked with us (most notably Stathis Charalampidis,
Chris Hoogeboom, and Haitao Xu), and importantly also numerous students and
young researchers working with our collaborators. We thank them tremendously for
teaching us all that we know about this subject and for inspiring us to consider
many of its physical and mathematical ramifications. We also thank the National
Science Foundation for its support of our program (through DMS-1615037 and
CMMI-1000337) as well as the US-AFOSR and Dr. Federica Darema for her
support via the DDDAS program (FA 9550-12-1-0332) and the US-ARO and
Dr. David Stepp for his support (via W911NF-15-1-0604).
ix
Contents
1 Introduction and Motivation of Models . . . . . . . . . . . . . . . . ...... 1

1.1 Dynamics of Hertzian Beads . . . . . . . . . . . . . . . . . . . . . ...... 1
1.2 The Linear Discrete Wave Equation . . . . . . . . . . . . . . . . ...... 6
1.3 The Nonlinear States: Dispersive Shock Waves, Traveling
Solitary Waves, and Discrete Breathers . . . . . . . . . . . . . . ...... 8
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...... 10
2 Dispersive Shock Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Theoretical Analysis of Shock Waves . . . . . . . . . . . . . . . . . . . . . 14
2.3 Prototypical Numerical Computations . . . . . . . . . . . . . . . . . . . . . 18
2.4 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3 Traveling Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 23
3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 23
3.2 Traveling Waves With Precompression: The KdV
and Toda Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 25
3.3 Traveling Waves Without Precompression I: The Continuum
Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 31
3.4 Traveling Waves Without Precompression II: Asymptotic
and Numerically Exact Results . . . . . . . . . . . . . . . . . . . . . . .... 33
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 38
4 Discrete (Dark) Breathers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1 Discrete Breathers: Overview and Theoretical Analysis . . . . . . . . 41
4.2 Connection with Experimental Results . . . . . . . . . . . . . . . . . . . . 46
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
xi
xii Contents
5 Heterogeneous Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...... 51

5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...... 51
5.2 Traveling Solitary Waves in Heterogeneous Chains . . . . . ...... 51
5.3 From Isolated Defect Modes to Bright Breathers in Dimer
Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...... 56
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...... 65
6 Media with Onsite Forces: The Newton’s Cradle and Beyond ..... 69
6.1 Breathers in the Newton’s Cradle . . . . . . . . . . . . . . . . . . . ..... 69
6.2 Solitary Waves and Breathers in Locally Resonant
Granular Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 71
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..... 75
7 Higher Dimensional Lattices . . . . . . . . . . . . . . . . . . . . . . ........ 77
7.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ........ 77
7.2 Conical Diffraction in the Hexagonal Packing . . . . . . . ........ 77
7.3 Other Work in Higher Dimensional Lattices and Future
Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ........ 81
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ........ 82
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Chapter 1
Introduction and Motivation of Models
1.1 Dynamics of Hertzian Beads
In the present short volume our aim will be to explore a variety of nonlinear wave
structures that are found in a particular physical system, i.e., granular crystals. Many
of the readers have almost certainly encountered a variant of the granular crystal.
For example, have you ever come across the intriguing toy shown in Fig. 1.1? This is
referred to as the Newton’s cradle. Besides being an interesting desktop decoration, it
is occasionally utilized in introductory physics classes to demonstrate the principle
of momentum conservation (since upon excitation of an incoming bead from one
side, the outgoing bead on the other side will appear to be symmetrically excited and
vice versa).
While we will return to this system later in the text (specifically Chap. 6), for
now we will explore a seemingly simpler variant thereof. In particular, we will part
with the “strings” inducing the pendular motion of the beads of Fig. 1.1 and we will
instead tackle a straight chain of beads. To simplify the system further, as is done
typically in related experiments, we will assume that the beads are confined by some
support system (such as rods) into a one-dimensional (1D) geometry, so that we can
minimize the role of transverse effects and the possibility of rotational motion. As
a result, we will consider the configuration of Fig. 1.2. Here, it is clear that each
bead interacts with its nearest neighbors, i.e., the nth bead incurs a force from its
interaction with beads n − 1 and n + 1. In light of that, we should be able to write
down the Newtonian dynamics for the evolution of its acceleration provided that
we can characterize its pairwise interaction with these neighbors. The key task then
amounts to modeling this interaction. It is important to appreciate firstly that the beads
deform elastically and thus the relevant interaction amounts to a compression force
(while the beads do not exert a force on each other when not in contact). Secondly,
this interaction depends on the geometric nature of the contact between the beads.
We will primarily consider case examples where the beads are spherical, in which
case, from a geometric perspective, the local radius of curvature near the contact
© The Author(s) 2018 1

C. Chong and P. G. Kevrekidis, Coherent Structures in Granular Crystals,
SpringerBriefs in Physics, https://doi.org/10.1007/978-3-319-77884-6_1
2 1 Introduction and Motivation of Models
Fig. 1.1 This figure showcases the classical Newton’s cradle toy, motivating many of the studies
to be presented herein. Figure credit: Dominique Toussaint/CC-BY-SA-3.0
point is the key determining feature of the force. In general, a network of particles
that interact elastically is called a granular crystal, see Fig. 1.2a. A granular crystal
confined to a 1D network will be called a granular chain.
The seminal work of H. Hertz in 1881 details the contact interaction between two
elastic bodies [2]. His setup consisted of two particles (e.g., spheres) that were in
point contact with each other. Upon being compressed, the particles deform slightly
resulting in repelling forces that push them. In Fig. 1.2b, we represent such a scenario
for identical spheres that are being pressed against each other. This squishing results
in an “overlap” length of δ. The interparticle force F that results is provided by the
expression for Hertz’s law [2]:
√
3/2 E 2R
F(δ) = A[δ]+ , A= , (1.1)
3(1 − ν 2 )
where the elastic (Young’s) modulus E indicates the material stiffness (analogous
to the spring constant in an ordinary linear spring), the Poisson ratio ν measures the
material tendency for expansion in other directions upon compression in a particular
direction, and the geometry enters through the sphere’s radius R. Such an interaction
among particles arises upon contact, with each particle feeling a force from its nearest
1.1 Dynamics of Hertzian Beads 3
(a)
(b)
Fig. 1.2 a Digital image of the granular crystal in its one-dimensional realization that will be of
principal interest in the present volume. The beads can consist of a wide array of materials but will
be assumed to generally be confined to a one-dimensional geometry via a support system of rods.
b Schematic of a precompressed 1D granular crystal, see Eq. (1.3). In the absence of precompression,
the distance between the centers of two adjacent particles is 2R, where R is the bead radius. If an
external (pre)compression force F0 is applied to both sides of the chain, the equilibrium positions
are modified. They are indicated by the open circles in the figure. The distance between adjacent
open circles is 2R − δ0 , where δ0 is the displacement caused from the precompression force, i.e.
3/2
F0 = Aδ0 . The black solid circles indicate the displacement from the equilibrium positions (i.e.
displacement from the open circles). The difference between the black circle and the open circle
in particle n is u n . Thus, the total overlap between particle u n and u n+1 is δ0 + u n − u n+1 . Figure
credits: Panel a was created by Joesph Lydon. Panel b figure is based on Fig. 1.1 of [1]; used with
permission
neighbors when there exists an overlap between them. This is the reason for the
bracket expression
δ, if δ > 0
[δ]+ = (1.2)
0, if δ ≤ 0
It is straightforward to generalize (1.1) for the case of two non-identical spheres [3].
The modeling of dissipation in these systems is presently an area of active research
(see discussion below). However, for periodic arrangements of grains composed of
very stiff materials, it is reasonable to ignore dissipative effects as a first approxima-
tion in the model. A convenient type of particle to use in experiments is stainless steel
spheres, such as ones with a radius of about R = 5 − 10 mm that are commonly used
in ball bearings (e.g., aluminum, bronze, glass, teflon, and many other materials have
also been used [4]). Steel’s elastic modulus is roughly E = 193 GPa, and ν = 0.3
is approximately the corresponding Poisson ratio. Remarkably, for granular crys-
tals consisting of steel spheres the linear wave speed is particularly low. Although
the sound speed (maximum linear wave speed) in bulk stainless steel is roughly
6000 m/s, the wave speed in a chain of steel spheres is typically below 100 m/s [1].
The nonlinearity in the Hertzian force stems from the above-mentioned bracket (see
Eq. (1.2)), as well as from the exponent of 3/2 (rather than a linear exponent of
unity) in Eq. (1.1), relevant to spherical contact interactions. Exponents other than
3/2 can be achieved by means of varying the geometry of the contact. For example,
O-rings interact via a double-power law, cylindrical particles feature a potential that
varies depending on the orientation of their axis, while for hollow spheres the particle
thickness affects the nature of the resulting interaction [3].
Using the considerations described above, in a one-dimensional granular chain of
finite length setting, the equations of motion are written as:
An p An+1 p
ü n = [δ0,n + u n−1 − u n ]+ − [δ0,n+1 + u n − u n+1 ]+ . (1.3)
Mn Mn
In this expression, the displacement is denoted by u n for the nth particle, with n
running from 1 to N for the chain. This is measured as a deviation from equilibrium
for the mass Mn . δ0,n = (F0 /An )1/ p represents the displacement that results for each
particle from the static compression force F0 = const. As indicated previously, for
spheres p = 3/2; see Fig. 1.2 for an illustration. The parameter An in the case of
non-identical neighboring beads is

4E n E n+1 (RRnn+R
Rn+1
n+1 )
An = , (1.4)
3E n+1 1 − νn2 + 3E n 1 − νn+1 2
where E n , νn , and Rn are the Young’s modulus, Poisson’s ratio and the radius,
respectively, of the nth bead. Cases of particular interest within Eq. (1.3) are the
monomer chain1 (which consists of identical beads with An = A, Mn = M, δ0,n =
δ0 ), the period 2 dimer, or just dimer, (which consists of alternating beads of two
types, i.e., An = A, δ0,n = δ0 , and e.g., Mn = M0 for n even and Mn = M1 for n odd;
here M1 /M0 is the sole additional tunable parameter in comparison to the case of
equal masses/radii), and chains with impurities (monomer chains with a few particles
of different type embedded).
To mathematically analyze the model, it is helpful to use the strain variable
yn = u n−1 − u n . In the case of the monomer chain the equations of motion become
1A note on nomenclature: In the literature, a monomer chain is also called a monatomic chain,
uniform chain, or homogenous chain.
1.1 Dynamics of Hertzian Beads 5
A 3/2 3/2 3/2

ÿn = − 2[δ0 + yn ]+ − [δ0 + yn−1 ]+ − [δ0 + yn+1 ]+ . (1.5)
M

Note that some authors define the strain as u n+1 − u n . Upon rescaling t → t A
M
we
can write Eq. (1.5) as

3/2 3/2 3/2
ÿn = − 2[δ0 + yn ]+ − [δ0 + yn−1 ]+ − [δ0 + yn+1 ]+ . (1.6)
1/2
If δ0 = 0, one can also use yn → δ0 yn , subsequently absorbing a factor of δ0 to the
time rescaling to render the model dimensionless. In the case of an infinite lattice,
the Hamiltonian associated with Eq. (1.3) reads:
1
H= Mn u̇ 2n + Vn (u n+1 − u n )
2
n∈ZZ
where the potential function is defined as
2
Vn (yn ) = An+1 [δ0,n+1 − yn ]5/2 + φn ,
5 (1.7)
2
φn = − An+1 [δ0,n+1 ]5/2 + An+1 [δ0,n+1 ]3/2 yn
5
Note that φn in Eq. (1.7) implies that Vn (0) = 0 and Vn (0) ≥ 0. This is needed to
ensure that u n = u̇ n = 0 is a minimum of H [5]. In particular, granular lattices with
vanishing or nonvanishing precompression were discussed in [6]. These cases had
Vn (0) = 0 and Vn (0) > 0, respectively.
This prototypical model does not include a number of features that can potentially
lead to disparities with respect to experimental results. An example of this type
concerns particle rotation [7–9], and another is the absence (for various practical
reasons) of perfect alignment of the particles’ contact points [10]. This can lead to
dynamic buckling in the case where high amplitudes are developed in the particle
motion. These features are considerably less pronounced and more well controlled in
1D settings, but this is far less so in two-dimensional (2D) ones where, e.g., rotation
effects are rather significant [7, 11–14]. We return to this point in Chap. 7.
Moreover, the Hamiltonian nature of Eq. (1.3) simplifies considerations facil-
itating an understanding of the main phenomena in such chains, yet in realistic,
experimentally tractable scenarios, dissipativity can play a significant role decel-
erating the coherent structure and modifying their dynamics in comparison to the
Hamiltonian ones. Arguably, one of the simplest approaches towards incorporating
the effect of dissipation involves including a dashpot term in the dynamics of Eq. (1.3)
in the form of Fd(1) = −u̇ n Mn /τn (where τn is a characteristic time scale). This is
qualitatively (and, with parameter fitting, even quantitatively in some cases) useful
as a description [15, 16]. A more recent proposal [17, 18] is to explore a form of
a dissipation variant incorporating a discrete velocity Laplacian. This idea is par-

tially inspired by hydrodynamic damping as implemented in a discrete variant of the
Navier-Stokes equations; see, e.g., the discussion in [19] where this has been used in
the context of “lattice turbulence.” The specific form employed in the granular setting
for the relevant force was Fd(2) = Mn [[u̇ n−1 − u̇ n ]+ /τn−1,n − [u̇ n − u̇ n+1 ]+ /τn,n+1 ]
for monomer chains, assuming τn−1,n is constant (as n varies). A mathematical
validation of this approach by means of perturbative methods can be found in [20].
Further efforts considered relative velocity power laws [21] and functional forms
such as [22]

Fd(3) ∝ Mn [u n−1 − u n ]+ (u̇ n−1 − u̇ n ) − [u n − u n+1 ]+ (u̇ n − u̇ n+1 ) ,
which are motivated by viscoelasticity theory and contains both displacements and
velocities. Moreover, finite-element approaches such as that of [23] have been used
to obtain yet another model for the dissipative contributions. To summarize, despite
numerous proposals, one can perhaps note that there does not seem to exist a uni-
versally acceptable functional form for incorporating dissipative effects (even in the
simplest scenario of a monomer chain). Traits like plastic deformation [24–29] and
rotation-induced damping may be partially responsible for the wave attenuation seen
in experiments [30] and the potential disparity with numerical output. Thus, devel-
oping a widely accepted formulation for incorporating dissipative effects remains
a key future challenge in the context of granular chains. In the present volume, we
mostly consider Hamiltonian models for this setting (and especially Eq. (1.3)).
1.2 The Linear Discrete Wave Equation
Before we examine the features of the fully nonlinear granular chain, we will outline
the features of its linear analogue with N − 1 linear oscillators.2 This is a considerably
simpler problem in comparison to the nonlinear ones that are focal points of interest in
this text. It will thus partially serve to contradistinguish the simplicity and generality
of linear theory in comparison to the more complex and suitably tailored nature of
the nonlinear analysis in the chapters that follow.
We select equal masses (denoted by M) coupled to each other through Hookian
(linear) springs/interactions. Then, applying Newton’s second law to each one, we
describe their deviations from equilibrium, as follows:
d 2un
M = Fright − Fleft (1.8)
dt 2
= −K 2 (u n − u n+1 ) + K 2 (u n−1 − u n ) (1.9)
= K 2 (u u+1 + u u−1 − 2u n ) (1.10)
2 The choice of N − 1 is made purely for reasons of notational convenience.

1.2 The Linear Discrete Wave Equation 7
where K 2 is the spring constant. Denoting ω02 = K 2 /M, we have a coupled set of
linear differential equations, i.e., a linear dynamical lattice.
To solve the linear system we can perform a separation of variables according to
u n = X n T (t) to obtain
T (t)X n = ω02 (X n+1 + X n−1 − 2X n )T (t) ⇒ (1.11)

T (t) X n+1 + X n−1 − 2X n
− 2 =− = β 2 = const. (1.12)
ω0 T (t) Xn
The solution to the temporal part, T = −(ω0 β)2 T , is a simple harmonic oscillator,
T (t) = A sin(βω0 t) + B cos(βω0 t), A, B ∈ IR. (1.13)
while, essentially by analogy, the spatial part, X n+1 + X n−1 − 2X n = −β 2 X n , will

have the solution
X n = C sin(qn) + D cos(qn) C, D ∈ IR, (1.14)
where A, B, C, D are arbitrary constants. To determine the relationship between q

and β, it is simpler to use the complex form of Eq. (1.14), namely X n ∼ eiqn , which
yields
eiqn+iq + eiqn−iq − 2eiqn

2 q
− = β 2
⇒ β 2
= 2 − 2 cos(q) = 4 sin (1.15)
eiqn 2
Since β = ±2| sin( q2 )|, the corresponding temporal frequencies are ω = βω0 =
±2ω0 | sin( q2 )| and hence will extend between −2ω0 and 2ω0 .
If we now impose the boundary conditions u 0 = u N = 0 (assuming that the end
beads are immobile walls), we obtain from X 0 = 0 that D = 0 in Eq. (1.14). From
X N = 0 we have that sin(q N ) = 0 (to avoid the trivial solution X n = 0 for all n, we
require C = 0), which is satisfied for q = mπ/N where m is an integer. To avoid
aliasing we only consider 1 ≤ m ≤ N − 1, since values of m outside this interval
only yield repetition of the modes already included. When we combine the spatial
and temporal parts (which are nothing more than the Fourier modes) into the full
solution for u n , we obtain that
mπ
X n T (t) = sin n Ã sin(βω0 t) + B̃ cos(βω0 t) , (1.16)
N
where Ã = AC and B̃ = BC. Now, employing linear superposition to obtain the

most general solution of the problem leads to:
(a) (b)
1 50
0.6
40
0.5
0.4
30
0 0.2
20
-0.5 0
10
-0.2
-1 0
0 5 10 15 20 10 20 30 40 50
Fig. 1.3 Solution to the linear mass-spring system in Eq. (1.8) with the mass and spring constant set
to unity (M = K 2 = 1) and all particle displacements and velocities set to zero, with the exception
of the first velocity (u̇ 1 = 1). This causes a traveling front to propagate through the lattice. However,
the front radiates energy as it travels due to the dispersive nature of the system. As we will see in
later chapters, nonlinearity can balance the dispersion, allowing for a perfect traveling wave. a Plot
of the velocity for the first three particles versus time. b Space-time contour plot of the solution.
Color intensity indicates displacement u. The checkered pattern indicates the radiated away energy
of the front due to dispersion
N −1
mπ
u n (t) = sin n Ãm sin(βm ω0 t) + B̃m cos(βm ω0 t) , (1.17)
m=1
N

mπ
βm = 2
sin
(1.18)
2N
This is, in principle, the most general solution of the original linear dynamical lattice,
as supplemented with the Dirichlet boundary conditions of vanishing field at sites n =
0 and n = N . The problem retains two sets of constant coefficients to be determined
Ãm and B̃m , per the initial conditions for the positions u n (t = 0) and velocities
u̇ n (t = 0) of the masses. See Fig. 1.3 for an example solution.
1.3 The Nonlinear States: Dispersive Shock Waves,

Traveling Solitary Waves, and Discrete Breathers
Having addressed the linear chain, the main focus of the present volume is to sum-
marize some of the principal developments of the last decade or so on the nonlinear
modes that are possible in this system. Although these have been discussed in the
recent review [4], the present work is a lengthier and more pedagogically minded
description. We will devote a separate chapter to each of the three principal excita-
tions that can arise in 1D granular crystals. These will be the dispersive shock wave,
the traveling solitary wave, and the discrete breather. Before discussing those, it
is important to point out other summaries on granular crystals for those looking to
obtain the broadest picture possible. V. F. Nesterenko was one of the early pioneers of
1.3 The Nonlinear States: Dispersive Shock Waves, Traveling Solitary … 9
the subject of granular crystals, with the key book contribution [1] being an excellent
summary of the early work on the subject. A significant fraction of the efforts until
2008 was also comprehensively summarized in review article [31] by S. Sen et al.
More topical reviews have since been published in the form of book chapters [32,
33] and a popular article in Physics Today was published recently [34]. The present
volume is complemented by the recent book [35], which covers topics with a different
emphasis than this book (although naturally an overlap between the two is present).
Traditionally, a shock wave is characterized as a state with an abrupt, nearly dis-
continuous change in the wave [36, 37]. Understanding such sudden changes in
the media of interest is relevant from an applied engineering perspective. Particular
examples include shock-absorbing armor, sports helmets, and other structures to be
protected against large-amplitude blasts and impacts [34]. To study shock waves in
continuous media (a subject that goes under the broad umbrella of so-called con-
servation laws [38]), one can study the evolution of jump initial data, the simplest
example of which is a step-function [38]. In this case, the constant low and high
amplitude states are “connected” by a line with infinite slope (at least in the ideal-
ized, non-dissipative case). If dispersion is present in the media, then the low and
high amplitude states can be connected by an oscillating wave train. The latter is
an example of a dispersive shock wave [37]. How such solutions can form in the
granular chain is described in Chap. 2. We begin this volume with the dispersive
shock waves not for historical reasons (the natural choice in that case would be the
solitary wave) but rather for pedagogical reasons; the analytical approach to study
the formation of dispersive shocks is the least technical of those that appear in this
book. Many of the other approaches will build upon this simpler case.
Chapter 3 deals with the traveling solitary wave, which is by far the most generic
and ubiquitous excitation in the granular chain. In an experimental setting, these
typically arise out of striking excitations (by another mass or by an actuator [1]).
Although in the engineering literature they are often referred to as compact (meaning
the support of the wave is finite), we will mathematically show that these waves bear
a very rapidly decaying (doubly exponential) tail when the precompression is zero
(δ0 = 0); yet, they are not genuinely compact, a misconception that is worthwhile
to finally do away with! In the presence of precompression, the decay of their tail
becomes exponential. The existence of traveling waves in granular crystals as a
means of stress propagation increases their attractiveness from a materials science
perspective (in the context of focusing and of guiding waves). This is important
because such waves are far less susceptible to dispersion in comparison to linear
waves. Through their low speeds, they are also considered to be useful candidates
for other applications such as impact mitigation [34].
The third principal type of excitation supported by granular crystals consists of
the so-called discrete breathers. These are time periodic, exponentially localized in
space excitations that also allow the trapping and potential transfer of energy. Such
excitations have been recognized in a wide array of physical systems ranging from
magnetic solids and superconducting junctions to nonlinear optics, and water waves,
among numerous other themes [39, 40]. In granular systems, different types of such
breathers (so-called dark or bright ones) have been theoretically argued, numerically
demonstrated, and experimentally corroborated as existing, depending on the nature

of the chain (e.g., homogeneous or heterogeneous) and the particular band (of the
underlying linear periodic spectrum) from which they bifurcate. We will see relevant
details in Chap. 4.
As a brief overview of what will follow: Chaps. 2–4 are each dedicated, respec-
tively, to dispersive shock waves, traveling solitary waves, and discrete breathers
in homogeneous granular chains. Examples of such excitations in heterogenous
media and media with onsite forces are explored in Chaps. 5 and 6. These are vari-
ations on the original 1D homogeneous granular chain theme that have in recent
years been of relevance to numerous experiments, and hence have also drawn
considerable theoretical attention. Finally, Chap. 7 touches on states that can arise
in higher dimensional crystals. This is a topic of higher complexity and model-
ing/mathematical/computational difficulty which is progressively emerging in the
forefront of considerations. In that light, in the relevant chapter we only give a brief
flavor of some recent developments, touching also in the final sections upon some of
the challenges that lie ahead. The appendices serve to explain a number of aspects
of the computational toolbox used in order to perform some of the computations
presented herein, as well as to expand upon some of the technical derivation details.
References
1. V.F. Nesterenko, Dynamics of Heterogeneous Materials (Springer, New York, 2001)

2. H. Hertz, Über die Berührung fester elastischer Körper. J. Reine. Angew. Math. 92, 156 (1881)
3. K.L. Johnson, Contact Mechanics (Cambridge University Press, Cambridge, 1985)
4. C. Chong, M.A. Porter, P.G. Kevrekidis, C. Daraio, Nonlinear coherent structures in granular
crystals. J. Phys. Condens. Matter 29, 413003 (2017)
5. G. Friesecke, J.A.D. Wattis, Existence theorem for solitary waves on lattices. Commun. Math.
Phys. 161, 391 (1994)
6. R.S. MacKay, Solitary waves in a chain of beads under Hertz contact. Phys. Lett. A 251, 191
(1999)
7. A. Merkel, V. Tournat, V. Gusev, Experimental evidence of rotational elastic waves in granular
phononic crystals. Phys. Rev. Lett. 107, 225502 (2011)
8. J. Cabaret, P. Béquin, G. Theocharis, V. Andreev, V.E. Gusev, V. Tournat, Nonlinear hysteretic
torsional waves. Phys. Rev. Lett. 115, 054301 (2015)
9. F. Allein, V. Tournat, V. Gusev, G. Theocharis, Tunable magneto-granular phononic crystals.
Appl. Phys. Lett. 108, 161903 (2016)
10. W. Lin, C. Daraio, Wave propagation in one-dimensional microscopic granular chains. Phys.
Rev. E 94, 052907 (2016)
11. J. Yang, S. Dunatunga, C. Daraio, Amplitude-dependent attenuation of compressive waves in
curved granular crystals constrained by elastic guides. Acta Mech. 223, 549 (2012)
12. J. Yang, M. Sutton, Nonlinear wave propagation in a hexagonally packed granular channel
under rotational dynamics. Int. J. Solids Struct. 77, 65 (2015)
13. P.A. Cundall, O.D.L. Strack, A discrete numerical model for granular assemblies. Géotechnique
29, 47 (1979)
14. D. Zabulionis, R. Kačianauskas, D. Markauskas, J. Rojek, An investigation of nonlinear tan-
gential contact behaviour of a spherical particle under varying loading. Bull. Pol. Acad. Sci.
Tech. Sci. 60, 265 (2012)
References 11
15. N. Boechler, G. Theocharis, C. Daraio, Bifurcation based acoustic switching and rectification.
Nat. Mater. 10, 665 (2011)
16. C. Hoogeboom, Y. Man, N. Boechler, G. Theocharis, P.G. Kevrekidis, I.G. Kevrekidis, C.
Daraio, Hysteresis loops and multi-stability: From periodic orbits to chaotic dynamics (and
back) in diatomic granular crystals. Euro. Phys. Lett. 101, 44003 (2013)
17. A. Rosas, A.H. Romero, V.F. Nesterenko, K. Lindenberg, Observation of two-wave structure
in strongly nonlinear dissipative granular chains. Phys. Rev. Lett. 98, 164301 (2007)
18. A. Rosas, A.H. Romero, V.F. Nesterenko, K. Lindenberg, Short-pulse dynamics in strongly
nonlinear dissipative granular chains. Phys. Rev. E 78, 051303 (2008)
19. M. Peyrard, I. Daumont, Statistical properties of one-dimensional "turbulence". Europhys. Lett.
59, 834 (2002)
20. N.V. Brilliantov, A.V. Pimenova, D.S. Goldobin, A dissipative force between colliding vis-
coelastic bodies: rigorous approach. EPL (Europhys. Lett.) 109, 14005 (2015)
21. R. Carretero-González, D. Khatri, M.A. Porter, P.G. Kevrekidis, C. Daraio, Dissipative solitary
waves in granular crystals. Phys. Rev. Lett. 102, 024102 (2009)
22. L. Vergara, Model for dissipative highly nonlinear waves in dry granular systems. Phys. Rev.
Lett. 104, 118001 (2010)
23. M. Gonzalez, J. Yang, C. Daraio, M. Ortiz, Mesoscopic approach to granular crystal dynamics.
Phys. Rev. E 85, 016604 (2012)
24. R.K. Pal, J. Morton, E. Wang, J. Lambros, P.H. Geubelle, Impact response of elasto-plastic
granular chains containing an intruder particle. J. Appl. Mech. 82, 38 (2015)
25. H.A. Burgoyne, C. Daraio, Elastic-plastic wave propagation in uniform and periodic granular
chains. J. Appl. Mech. 82, 081002 (2015)
26. H. Burgoyne, C. Daraio, Strain-rate-dependent model for the dynamic compression of elasto-
plastic spheres. Phys. Rev. E 89, 032203 (2014)
27. H.A. Burgoyne, Dynamics of granular crystals with elastic–plastic contacts, 2016. Ph.D. Dis-
sertation, California Institute of Technology (2016). https://doi.org/10.7907/Z9J38QG6
28. T. On, P.A. LaVigne, J. Lambros, Development of plastic nonlinear waves in one-dimensional
ductile granular chains under impact loading. Mech. Mater. 68, 29 (2014)
29. T. On, E. Wang, J. Lambros, Plasticwaves in one-dimensional heterogeneous granular chains
under impact loading: single intruders and dimer chains. Int. J. Solids Struct. 62, 81 (2015)
30. J. Yang, M. Gonzalez, E. Kim, C. Agbasi, M. Sutton, Attenuation of solitary waves and local-
ization of breathers in 1D granular crystals visualized via high speed photography. Exp. Mech.
54, 1043 (2014)
31. S. Sen, J. Hong, J. Bang, E. Avalos, R. Doney, Solitary waves in the granular chain. Phys. Rep.
462, 21 (2008)
32. G. Theocharis, N. Boechler, C. Daraio, Nonlinear Phononic Periodic Structures and Granular
Crystals, Acoustic Metamaterials, Phononic Crystals (Springer, Berlin, 2013), pp. 217–251
33. A.F. Vakakis, Analytical methodologies for nonlinear periodic media, Wave Propagation in
Linear and Nonlinear Periodic Media, (International Center for Mechanical Sciences (CISM)
Courses and Lectures) (Springer, Berlin, 2012), p. 257
34. M.A. Porter, P.G. Kevrekidis, C. Daraio, Granular crystals: nonlinear dynamics meets materials
engineering. Phys. Today 68, 44 (2015)
35. Y. Starosvetsky, K. Jayaprakash, M.A. Hasan, A. Vakakis, Dynamics and Acoustics of Ordered
Granular Media (World Scientific, Singapore, 2017)
36. M.J. Ablowitz, M. Hoefer, Dispersive shock waves. Scholarpedia 4, 5562 (2009)
37. G.A. El, M.A. Hoefer, M. Shearer, Dispersive and diffusive-dispersive shock waves for non-
convex conservation laws. SIAM Rev. 59, 3–61 (2017)
38. J. Smoller, Shock Waves and Reaction-Diffusion Equations (Springer, New York, 1983)
39. S. Flach, A. Gorbach, Discrete breathers: advances in theory and applications. Phys. Rep. 467,
1 (2008)
40. S. Aubry, Discrete breathers: localization and transfer of energy in discrete Hamiltonian non-
linear systems. Phys. D 216, 1 (2006)
Chapter 2
Dispersive Shock Waves
2.1 Overview
Shock waves are nonlinear structures characterized by a discontinuous jump in the

wave profile [1]. There exist prototypical examples of partial differential equations
(PDEs), including the well-known Burgers’ equation (without dissipation) [2], where
the relevant waveforms feature infinite derivatives arising in finite time [3]. To study
shock waves in equations like the inviscid Burgers’ equations one can study the evo-
lution of Riemann initial conditions [3] involving a jump in the initial data. Alterna-
tively, one can study the emergence of shocks by examining the evolution of smooth
bell-shaped initial data. In the latter scenario, the wave’s larger amplitude part prop-
agates faster than the lower amplitudes, and thus the large amplitude part “catches
up” with the small amplitude part, see Fig. 2.1a. The point of multi-valuedness of the
structure would represent the time instance when the shock wave forms.
In this chapter, we are concerned with the formation of structures like the shock
waves described above. However, the granular chain is spatially discrete, and thus
we cannot define a solution with an infinite spatial derivative. Since under external
compression force, dispersive effects can be controllably introduced in the chain,
it is appropriate to compare what forms in the granular chain with what forms in
other dispersive media, such as the Korteweg-de Vries (KdV) equation. It is well
known that in media described by the KdV equation (such as waves in shallow
water [4]) a modulated wave train may arise [1, 5], see Fig. 2.1b. Such solutions
are called dispersive shocks. Thus, the focus of this Chapter is their formation. It
should be noted that in the context of lattices of nonlinear springs, i.e., in the well-
known Fermi-Pasta-Ulam-Tsingou (FPUT) model [6] (of which the granular chain
is a specific example), such structures were numerically examined [7].
In granular media, shock-like waves have been induced experimentally [8] through
a continuous velocity drive of one boundary of a chain. In some of the relevant
realizations, it has been shown that when the dissipation plays a more significant role

14 2 Dispersive Shock Waves
(a) (b)
Y
Y
X X
Fig. 2.1 a A nonlinear wave Y = Y (X, T ) versus spatial coordinate X for a fixed value of time
T . The wave was obtained by solving the Burgers’ equation ∂T Y = (1 + Y )∂ X Y using the method
of characteristics with the initial data Y (X, 0) = sech(X ). The profile is shown for a time T > Ts ,
where Ts is the shock formation time (i.e., the first time the characteristic lines intersect, see text).
The wave is triple valued. It can be made single valued by inserting a discontinuity and ignoring the
“smooth” lobes, see the discussion in [4]. The resulting single-valued solution is a shock wave. b
Dispersive-shock solution of the KdV equation. The profile Y = Y (X, T ) versus spatial coordinate
X for a fixed value of time T is shown. Note the constant amplitude parts of the solution are
connected by an oscillatory wave train, rather than a discontinuity (as is the case for a shock wave).
[Figure credit: We use panel (b) with permission from [1]]
than the dispersion, it can completely “wash away” the oscillations of the dispersive
shock wave. This has been shown in [9, 10].
Despite the important differences between the traditional shock waves found in
Burgers’ equation and the dispersive shocks of the granular chain, there is important
insight that one gains through examining the connection of the two. This connection
will be explored in the present chapter. In particular, we will summarize the steps
of the recent work of [11] and the earlier one of [12]. The relevant methodology
will connect the granular model through a continuum reduction to a generalized
form of the prototypical equation that supports shock waves, namely the inviscid
Burgers’ model. The generality will result from the fact that we will examine arbitrary
exponents in Eq. (3). This is not purely a mathematical exercise though; as discussed
partly also in the introductory chapter, while the exponent p = 3/2 is applicable
to spheres, numerous other exponents have recently been argued to be realizable in
mechanical metamaterials where it is possible to realize case examples with 0 <
p < 1 [13–15].
2.2 Theoretical Analysis of Shock Waves
One of the most canonical ways of analyzing lattice dynamical systems of the form of
Eq. (1.6) involves resorting—in the so-called long wavelength limit—to their contin-
uum analogue, i.e., a corresponding PDE. The long wavelength limit here means that
the spatial extent of the excitation/structure of interest is much larger than the lattice
scale and hence a continuum approach is a reasonable thing to do. To obtain such
2.2 Theoretical Analysis of Shock Waves 15
a limit, however, a small parameter ε must be introduced (essentially characterizing

the scale of the lattice, or alternatively, if we think of the lattice spacing as unity, 1/ε
characterizes the scale of the excitation). This was done for nonlinear variants of the
wave equation (recall that for p = 1 this is the discrete wave equation, hence p = 1
renders it a nonlinear variant thereof) in a rigorous fashion in [16]. By rigorous we
mean that the approximations were mathematically developed with suitable error
estimates. We direct the interested reader to the latter work for more details and here
only use the main conclusion of that work.
In order to pass, at a phenomenological level, to a continuum limit, we introduce
an (artificial) lattice scale , tantamount to the well-known in numerical analysis
lattice spacing x [17]. It is then instructive to consider the excitations of interest as
extending over a spatial domain of scale 1/. This motivates the ansatz (or “guess”)
yn (t) ≈ Y (X, T ), where the (long) spatial and (slow) temporal scales are defined
as, respectively, X = εn and T = εt. The function Y (X, T ) will be determined in a
way that yields an approximation of an actual solution yn (t). The difference between
the actual solution yn (t) and the approximation Y (X, T ) should decrease as the
parameter ε tends to zero. In order to find a particular functional form of Y (X, T ),
we simply insert Y (X, T ) into Eq. (1.6). This will lead to an equation involving
Y (X, T ), that should be easier to analyze (or even solve) than the original problem
Eq. (1.6). With this plan in mind, we now insert Y (X, T ) in place of yn (t) in Eq. (1.6)
and compute the necessary partial derivatives and differences to arrive at,
{δ0 + Y (X + ε)} p + {δ0 + Y (X − ε)} p − 2{δ0 + Y (X )} p

∂T2 Y = . (2.1)
ε2
Then, taking the limit of ε → 0, in which case the second order difference on the
right-hand side tends to the second (partial) derivative, the continuum model follows:

∂T2 Y = ∂ X2 (δ0 + Y ) p . (2.2)
Clearly, if Y (X, T ) is a solution to Eq. (2.2), it will only represent a good approxi-
mation to yn (t) if ε > 0 is small, since Eq. (2.2) is only valid in the limit ε → 0.
Inspired by the work of [12], we examine a nonlinear transport (first order) PDE
that yields the same second order PDE dynamics as Eq. (2.2). I.e., we use here the
notion that one-dimensional (linear) wave equations bear left and right propagating
waves [2] to decompose this nonlinear wave equation into constituents describing
its left and right traveling waves. Thus, we seek a model of the form

∂T Y ± α∂ X (δ0 + Y )c = 0, (2.3)
where ± indicates the two directions of propagation. The parameters α and c are
intended to be chosen in a way such that the solutions of Eq. (2.3) are also solutions
of Eq. (2.2). We thus obtain:
α 2 c2 2
∂T2 Y = ∂T ∓α∂ X (δ0 + Y )c = ∓ ∂ X (δ0 + Y )2c−1 . (2.4)
2c − 1
By comparison of Eq. (2.2) and of Eq. (2.4), we conclude that a suitable choice is
c = ( p + 1)/2 and α 2 = 2c−1
c2
. Then Eq. (2.3) is rewritten as:
√ p−1
∂T Y ± p(δ0 + Y ) 2 ∂ X Y = 0. (2.5)
Equation (2.5) is a generalized form of the well-known Burgers’ equation [2] that is
ubiquitous in the study of hyperbolic conservation laws [3]. To obtain an intuition
regarding this system, notice that Eq. (2.5) looks like the linear transport equation
∂T Y ± φ∂ X Y = 0, φ ∈ IR (2.6)
which simply transports the initial data with the velocity φ, resulting in a solution
that has the form Y (X, T ) = Y0 (X ∓ φT ), where Y0 is the initial data. However,
in the case of (2.5) the velocity φ depends on the amplitude of the wave: φ =
√ p−1
p(δ0 + Y ) 2 . This suggests that the solution of (2.5) is given implicitly as
√ p−1
Y (X, T ) = Y0 (X ∓ p(δ0 + Y ) 2 T ). (2.7)
This heuristic argument can be made more precise by making use of the well-known
characteristics technique (see, e.g., [2] for a brief exposition). In particular, this
technique is based on the realization that for this (nonlinear) transport equation, the
characteristic lines defined by:
dX
= φ (Y (X, T )) (2.8)
dT
are ones where the solution (Y ) is constant. These are thus straight lines. Here we
note that
√ p−1
φ (Y (X, T )) = p (δ0 + Y (X, T )) 2 . (2.9)
From the constant slope of these characteristic lines, we can write
dX X − X0
= = φ (Y (X, T )) = φ (Y (X 0 , 0)) . (2.10)
dT T
We can rearrange Eq. (2.10) to obtain the following relation:
√ p−1

X = X0 + p (δ0 + Y (X 0 , 0)) 2 T, (2.11)
which implies the solution at time T is simply the initial profile Y0 displaced by
√ p−1
the amount ( p (δ0 + Y0 ) 2 )T . This is equivalent to Eq. (2.7). Another way to say
this is that since the solution will be constant along these (characteristic) lines, it
should depend on the only combination of X and T that amounts to a constant and,
2.2 Theoretical Analysis of Shock Waves 17
√ p−1

according to Eq. (2.11), this combination is X − p (δ0 + Y (X 0 , 0)) 2 T , hence
the solution will be a function of this argument.
From Eq. (2.7) we can infer that the velocity dependence of the solution’s ampli-
tude Y (with larger amplitudes corresponding to higher speeds for p > 1), causes
bell-shaped waveforms to steepen as shown in Fig. 2.1a. This is in line with the well-
established feature of generalized Burgers’ models (without dissipation) [2, 3] that
smooth profiles of Y at t = 0 develop wave breaking and form shocks (featuring
infinite derivatives) within a finite time. This shock wave reflects the resulting multi-
valuedness at the intersection point of the characteristic lines. To identify the shock
wave time Ts , we follow two arbitrary characteristics
X (T ) = φ (X 1 ) T + X 1 (2.12)
X (T ) = φ (X 2 ) T + X 2 (2.13)
where X 1 , X 2 ∈ R and X 2 = X 1 + h. At the time of their intersection, we obtain
X2 − X1 h
T =− =− . (2.14)
φ (X 2 ) − φ (X 1 ) φ (X 1 + h) − φ (X 1 )
The shock wave is first formed when the characteristic lines first intersect. Thus,
minimizing the above expression leads to an estimate for the shock formation time
(Ts ) as:

h
Ts = min −
φ (X 1 + h) − φ (X 1 )
1
=
.
min − φ(X 1 +h)−φ(X
h
1)
In the limit of h → 0, we have
1
Ts = −
. (2.15)
dφ(X )
min dX
)
In the work of [11], the semi-analytical prediction based on evaluating dφ(X
dX
and
upon numerical identification of the corresponding minimum, it was compared with
direct numerical simulations.
Before turning to numerical simulations, we briefly describe an alternative
approach that also leads to the conclusion that Eq. (2.2) has wave solutions with
amplitude-dependent wave speeds. One rewrites Eq. (2.2) as a system of conserva-
tion laws [16]:
∂T Y − ∂ X V =0
, (2.16)
∂T V − ∂ X [(δ0 + Y ) p ] = 0
This is termed a “p-system” [3, 18]. Using

Y −V
Y= and F(Y) = ,
V −(δ0 + Y ) p
we can rewrite Eq. (2.16) as
∂T Y + DF(Y)∂ X Y = 0 , (2.17)
where
0 −1
DF(Y) =
− p(δ0 + Y ) p−1 0
is the Jacobian of F. This equation looks like the vector valued transport equation
∂T Y + A∂ X Y = 0 , A ∈ IR2x2 (2.18)
which has solutions that are superpositions of traveling waves with velocity given by
the eigenvalues of the matrix A. The eigenvalues of the matrix DF(Y) are λ± (Y) =
√ p−1
± p (δ0 + Y (X, T )) 2 , which indicates that the solutions of Eq. (2.16) travel with
√ p−1
velocity p (δ0 + Y (X, T )) 2 . This is consistent with what we concluded when
considering Eq. (2.5).
2.3 Prototypical Numerical Computations
For an initial condition (at the strain level) of Y0 (X ), the corresponding velocity is
given by

d yn d
= [Y (X, T )]t=0 (2.19)
dt dt
t=0

√ p−1

= ± p(δ0 + Y0 ) 2 ∂ X Y0 . (2.20)
Following the discussion of [11], we use an initial strain given by
Y0 (X 0 ) = asech (bX 0 ) , (2.21)
choosing a = 0.01 and b = 0.3. Comparisons are made between the discrete simu-
lations and the continuum approximations. The latter predicted profile can be found
by solving the implicit Eq. (2.7), or equivalently, by modifying the positions of the
initial amplitudes using the relation, Eq. (2.11).
Figure 2.2 represents the strain propagation for p = 3/2 [10, 19], while Fig. 2.3,
as per the relevance of values p < 1 [14, 15], explores the value of p = 1/2. In these
2.3 Prototypical Numerical Computations 19
Fig. 2.2 Strain wave propagation of the precompressed monomer chain upon a boundary excitation
for p = 3/2 and a δ0 = 0, b 0.001, and c 0.01. The left figures show temporal plots of strain waves.
The black solid line is obtained from the direct simulation of the discrete model, and the red dashed
line is the prediction based on Eq. (2.7) (i.e., the continuum model). Strain curves are offset to
ease visualization. The inset shows the magnified view of the leading part at the predicted shock
time. The right figures show space-time contour plots of strain wave propagation. The gray solid
line indicates the predicted analytical shock wave time. [Figure credit: Used with permission from
Ref. [11]. Copyrighted (2017) by the American Physical Society]
Fig. 2.3 Strain wave propagation of the precompressed monomer chain upon a boundary excitation
for p = 1/2 and a δ0 = 0, b 0.001, and c 0.01. The left figures show temporal plots of strain waves.
The black solid line shows the discrete model numerical results as compared to the red dashed line
of the continuum theory of Eq. (2.7). Both these panels and the strain contour panels on the right are
similar to those shown in the previous figure. [Figure credit: Used with permission from Ref. [11].
Copyrighted (2017) by the American Physical Society]
two figures, the left panels illustrate four snapshots of the wave, while in the right,
the space-time contours for the strain evolution are shown. The no-precompression
(fully nonlinear) case is shown in the top, while the following rows correspond to
2.3 Prototypical Numerical Computations 21
cases of increasing precompression (with the problem becoming less nonlinear). This
is because the lower order terms in the Taylor expansion of (δ0 + Y ) p will dominate
over the higher order (more nonlinear) ones. For p = 3/2, the wave breaking occurs
at the front (see Fig. 2.2), whereas for p = 1/2 it emerges at the tail in Fig. 2.3. This
stems from the amplitude dependence of the speed of propagation φ(Y (X, T )) as per
Eq. (2.9). For p > 1, the faster traveling of large amplitudes leads them to overtake
the smaller amplitudes, resulting in a shock formation in the front and the emergence
of a rarefaction in the back. The reverse happens for p < 1, with the rarefaction
arising in the front, and the wave breaking discerned in the rear part of the wave.
In Figs. 2.2 and 2.3, the computation of the dynamical evolution is shown together
with the corresponding PDE as per Eq. (2.7). The two are in close agreement up to
the point in time where the discontinuity begins to form. Then, the discreteness of
the model leads to the emergence of oscillatory dynamics, a trait reminiscent of
shock waves of dispersive media (which are called dispersive shocks), such as the
KdV equation [1, 5] (compare Fig. 2.1b with the zoomed-in panels of Fig. 2.2). For
this reason, we call the shock waves studied here dispersive shocks. Note that this
feature is not present in the long wavelength model. A quantity used in order to
compare the discrete and continuum cases is the shock formation time. Numerically,
this was identified in [11] by examining the difference of sign of the strains of
adjacent particles: in the presence of multiple oscillations of this quantity, the time
(Tds ) of emergence of the shock was defined. The resulting comparison of Ts (shock
formation time in the continuum approximation) and Tds was a key result in the
work of [11]. There, the dependence of the shock time on both the exponent ( p) and
the value of the precompression (δ0 ) was examined. It was found that the disparity
between theoretical and numerical results was enhanced as linearity is approached
with p → 1, in which case no shocks would arise (and thus the shock formation time
diverges). On the other hand, as the precompression was increased (i.e., as the role of
the nonlinearity became progressively less significant and the system approached the
linear regime), the disparity between analytical and numerical findings also increases.
2.4 Outlook
It should be noted that in addition to the early work of [9] that laid the foundation
for the examination of shock waves, some of the most important works in this theme
have been those of [8] and of [12]. The former has paved the way for generalizing
shock waves in heterogeneous settings (in particular, dimers), also focusing on the
exploration of the traveling wave at the leading edge of the shock. On the other
hand, [12] spearheaded the connection of the continuum, nonlinear wave equation
analogue (see also [16]) of the system with a generalized variant of the Burgers’
equation. Using the formulation and results of [11] as a guide, we have discussed
this perspective here due to its intuitive nature and analytical tractability. It was found
that there are crucial differences as the nonlinearity exponent changes from p > 1
to p < 1. In the former case, and for bell-shaped initial conditions, a shock forms in
the front and a rarefaction in the back, while in the latter the features are reversed.
In the future, it will be especially interesting to explore whether further connec-
tions and understanding into the shock waves of the granular crystals can be obtained
from that of the well-known shock waves of continuum models such as the Korteweg-
de Vries equation (see, e.g., [4] and also the next chapter), or discrete ones such as the
Toda lattice (an integrable lattice with exponential pairwise interactions [20] instead
of the power law ones considered here), or from the discrete variant of Burgers’
equations [21]. The latter is especially interesting, as it may suggest an analogue of
the transition from a second order lattice to a first order one (in analogy to what was
previously considered at the PDE level) and also because of the existence of fully
solvable models of this type, as discussed in [21].
References
1. M.J. Ablowitz, M. Hoefer, Dispersive shock waves. Scholarpedia 4, 5562 (2009)

2. W.A. Strauss, Partial Differential Equations: An Introduction (Wiley, Hoboken, 2008)
3. J. Smoller, Shock Waves and Reaction-Diffusion Equations (Springer-Verlag, New York, 1983)
4. P.G. Drazin, R.S. Johnson, Solitons: An Introduction (Cambridge University Press, Cambridge,
UK, 1989)
5. G.A. El, M.A. Hoefer, M. Shearer, Dispersive and diffusive-dispersive shock waves for non-
convex conservation laws. SIAM Review 59, 3–61 (2017)
6. E. Fermi, J. Pasta, S. Ulam, Studies of Nonlinear Problems. I., (Los Alamos National Labora-
tory, Los Alamos, NM, USA), Technical Report (1955), pp. LA–1940
7. M. Herrmann, J.D.M. Rademacher, Riemann solvers and undercompressive shocks of convex
FPU chains. Nonlinearity 23, 277 (2010)
8. A. Molinari, C. Daraio, Stationary shocks in periodic highly nonlinear granular chains. Phys.
Rev. E 80, 056602 (2009)
9. E.B. Herbold, V.F. Nesterenko, Shock wave structure in a strongly nonlinear lattice with viscous
dissipation. Phys. Rev. E 75, 021304 (2007)
10. V.F. Nesterenko, Dynamics of Heterogeneous Materials (Springer-Verlag, New York, 2001)
11. H. Yasuda, C. Chong, J. Yang, P.G. Kevrekidis, Emergence of dispersive shocks and rarefaction
waves in power-law contact models. Phys. Rev. E 95, 062216 (2017)
12. B.E. McDonald, D. Calvo, Simple waves in Hertzian chains. Phys. Rev. E 85, 066602 (2012)
13. Propagation of rarefaction pulses in discrete materials with strain-softening behavior. Phys.
Rev. Lett. 110 144101 (2013)
14. F. Fraternali, G. Carpentieri, A. Amendola, R.E. Skelton, V.F. Nesterenko, Multiscale tunability
of solitary wave dynamics in tensegrity metamaterials. Appl. Phys. Lett. 105, 201903 (2014)
15. H. Yasuda, C. Chong, E.G. Charalampidis, P.G. Kevrekidis, J. Yang, Formation of rarefaction
waves in origami-based metamaterials. Phys. Rev. E 93, 043004 (2016)
16. C. Chong, P.G. Kevrekidis, G. Schneider, Justification of leading order quasicontinuum approx-
imations of strongly nonlinear lattices. Disc. Cont. Dyn. Sys. A 34, 3403 (2014)
17. K. Atkinson, An Introduction to Numerical Analysis (Wiley, Hoboken, 1989)
18. R.J. LeVeque, Numerical Methods for Conservation Laws (Birkhauser, Basel, 1992)
462, 21 (2008)
20. M. Toda, Theory of Nonlinear Lattices (Springer-Verlag, Heidelberg, 1989)
21. C.V. Turner, R.R. Rosales, The small dispersion limit for a nonlinear semidiscrete system of
equations. Stud. Appl. Math. 99, 205 (1997)
Chapter 3
Traveling Waves
3.1 Overview
Arguably, the most prototypical nonlinear wave structure that can arise in granular
crystals consists of traveling waves. While many kinds of traveling waves exist (such
as periodic ones [1]), in this chapter we are interested in ones that are spatially
localized, i.e., traveling solitary waves (which we will simply call traveling waves
when the distinction is clear). In fact, in the first “phase” of research on this topic,
as covered extensively by the quintessential references of [2, 3], shock waves (the
subject of the previous chapter) had barely been touched upon, while breathers (the
subject of the next chapter) were altogether absent. It was thus the highly robust, and
(comparatively) straightforward to excite experimentally, traveling waves that were
at the center of the theoretical, numerical, and experimental attention.
The prevalence of traveling waves is easy to understand given that prototypical
evolutions both in the absence (i.e., for δ0 = 0) and even in the presence (i.e., for
δ0 = 0) of precompression lead to their spontaneous formation. Importantly, in this
system there is no distributed control of the initial conditions in the experiments,
i.e., it is not straightforwardly possible to tune the displacement and velocity of
each bead. Instead, the walls/boundaries of the system are amenable to controllable
excitation and the resulting evolution can then be observed. While the ability to drive
the boundaries essentially at will can be used to “design” excitations that will result
in other forms of nonlinear waves, such as breathers (some principal examples of that
type will be given in the next chapter), a typical boundary “kick” (i.e., striking a single
particle at one end of the chain) generically leads to traveling wave structures. The
first such experiment was conducted by Lazaridi and Nesterenko in 1985 [4]; see also
Sect. 3.3. Figure 3.1 shows various traveling waves in the absence of precompression
of the monomer granular chain model of Eq. (1.3). In the presence of precompression
one difference is that additionally dispersive linear wave packets are also present;
in the absence of precompression, the sonic speed vanishes (the so-called sonic
vacuum [2]) and hence no linear waves can propagate.

24 3 Traveling Waves
(a) (b)
0.1 200
0.08 150
V (velocity)
0.06
100
t
0.04
0.02 50
0
0
0 5 10 15 20 0 5 10 15
t n
(c)
40
20
-20
-40
-10 -5 0 5 10
Fig. 3.1 Examples of traveling waves in a monomer chain without precompression illustrated in
various forms. a A compression wave travels through the chain after the configuration is struck at
one end. After the wave has traveled a few lattice sites, it appears as though a constant shape profile
forms and propagates through the chain (this is the traveling solitary wave). Each bell-shaped curve
corresponds to the velocity profile (V ) of a single particle. b Contour plot of a solitary wave solution.
Space (indexing bead numbers) and time in dimensional units (ms) are shown with the contour plot
denoting the strain. The constant slope (in space-time) clearly depicts the existence of a constant
speed traveling wave. c In this chapter we will present the theory involved to show that genuine
traveling solitary waves do indeed exist in the granular chain. The profile shown in this panel is the
solution to Eq. (3.28). The solid curves pertain to the strain φ(ξ)—symmetric curve—and velocity
φξ (ξ) –antisymmetric curve–. The variable ξ = n − ct is the traveling coordinate, where c is the
wave speed. Note, in the linear chain, such a traveling wave is not possible (compare panels (a) and
(b) of this figure with Fig. 1.3). [Figure credit: Used with permission from Ref. [5]. Copyrighted
(2017) by the Institute of Physics]
Our presentation of these waves will be structured as follows. In the next section,
we will consider the technically simpler case of precompression. There, we will see
that the reduction to the well-known Korteweg-de Vries (KdV) equation will enable
a road map for the characterization of both single traveling waves, and also more
complicated solutions such as pairs of traveling waves. Of similar usefulness is the
well-known Toda lattice example [6]. Then, we will turn to the technically more
demanding case without precompression. We will first explore continuum approxi-
mations yielding quantitative (yet somewhat uncontrollable) insights for the traveling
waves. Finally, we will close this chapter by presenting some methods for the exact
computation of traveling waves and their impact on the understanding of the cases
with and even without precompression.
3.2 Traveling Waves With Precompression: The KdV and Toda Limits 25
3.2 Traveling Waves With Precompression: The KdV

and Toda Limits
We start our discussions from Eq. (1.3). Given that we consider, at the moment,
the case of the monomer chain with equal precompressions at the two ends, all the
prefactors An /Mn are equal. Assuming now that the relative displacements satisfy
u n−1 − u n δ0 , we can Taylor expand Eq. (1.3) as:

d 2un p u n−1 − u n p( p − 1) u n−1 − u n 2
M 2 = Aδ0 1 + p + + ...
dt δ0 2 δ0
+
2
p u n − u n+1 p( p − 1) u n − u n+1
− Aδ0 1 + p + + ... (3.1)
δ0 2 δ0
+
Recognizing that the leading terms vanish, we can see that the model acquires the
form of an FPUT model [7] according to:
d 2un
= c2 (u n+1 + u n−1 − 2u n ) (3.2)
dt 2
c2 ( p − 1)
+ (u n−1 − u n+1 ) (u n+1 + u n−1 − 2u n )
2δ0
p−1
In this expression, we have already defined cs 2 = Apδ0 /M, with c carrying the
units of inverse seconds (note, since this setting is discrete, cs can be considered a
speed, i.e., lattice sites traveled per second). As per our discussion of Sect. 1.2, this
is the maximal speed available to the lattice setting.
To pass to a continuum model (similar to what was done in Chap. 2 for shock
waves) we once again start with the ansatz of the form u n (t) = U (X, T ), where
X = n and T = t. Substitution of this ansatz into Eq. 3.2 leads to

2 ∂T2 U = cs 2 U (X + ) + U (X − ) − 2U (X ) (3.3)
cs 2( p − 1)
+ U (X − ) − U (X + ) U (X + ) + U (X − ) − 2U (X )
2δ0
where we have dropped the explicit T dependence of U to simplify the notation.

Motivated by finite difference based Taylor expansions from numerical analysis [8],
we Taylor expand the new variable U ,
2 3 4
U (X ± ) = U (X ) ± ∂ X U (X ) + ∂ X2 U (X ) ± ∂ X3 U (X ) + ∂ X4 U (X ) + · · · ,
2 3! 4!
(3.4)
j
where ∂ X denotes the jth partial derivative with respect to the variable X . From this,
we infer the following finite difference approximations (where has assumed the
place of the “usual” Δx),
U (X + ) + U (X − ) − 2U (X ) 2
≈ ∂ 2
U (X ) + ∂ 4
U (X ) (3.5)
2 X X
4!
3
U (X + ) − U (X − ) ≈ 2∂ X U (X ) + ∂ X3 U (X ) (3.6)
3
Substituting Eqs. (3.5) and (3.6) into Eq. (3.3) leads to the partial differential equation
(long wavelength formal limit):
( p − 1)cs 2 c2 2
∂T2 U = cs 2 ∂ X2 U + ∂ X U ∂ X2 U + s ∂ X4 U. (3.7)
δ0 12
Since we are seeking traveling waves, it is natural to go to a co-traveling frame.

This is a standard practice when exploring traveling waves in partial differential
equations. This is because in this way, the traveling is “factored out” and the solution
appears to be steady. In this frame we will thus arrive at an ordinary differential equa-
tion rather than a partial differential one. This simplifies the analysis considerably,
as we will see explicitly in our example below. To go to a variable traveling with the
solution we define
ξ = X − cs T, τ = γT (3.8)
The second of these variables introduces an arbitrary (by a factor γ) rescaling of time
that will be determined momentarily (by the need for a self-consistent, simplifying
formulation). We thus obtain in the new frame U (X, T ) → U (ξ, τ ):
( p − 1)cs 2 2 cs 2 4
cs 2 ∂ξ2 U + γ 2 ∂τ2 U − 2cs γ∂ξτ U = cs 2 ∂ξ2 U + ∂ξ U ∂ξ2 U + ∂ U. (3.9)
δ0 12 ξ
The first terms on both sides drop out. Selecting γ = 2 renders the last term on each
side of the same order, i.e. O(2 ) (and the second term on the left, a higher order
ignorable contribution, i.e., O(4 )), while to render the second term on the right to
be O(2 ), we define the small parameter through δ0 = 1/. Canceling one power
of c and defining the (continuum) strain variable ∂ξ U = v we obtain
c 3
2∂τ v + c( p − 1)v∂ξ v + ∂ v = 0. (3.10)
12 ξ
Rescaling cs τ /24 → τ̃ and 2( p − 1)v → ṽ and dropping the tildes for simplicity,
we arrive at the celebrated KdV equation in the form:
∂τ v + 6v∂ξ v + ∂ξ3 v = 0. (3.11)

While any solution v of Eq. (3.11) will result in a traveling wave of Eq. (3.2) when
returning to the original variables u n (t), we are interested in traveling solitary waves.
The soliton solution of Eq. (3.11) will provide us with the desired solution with
spatially decaying tails [9]. To compute the soliton solution of the KdV equation, we
first seek a traveling wave of Eq. (3.11) in the frame v(ξ, τ ) → v(ξ − Cτ ) = v(Ξ ),
and obtain an ordinary differential equation −Cv = −v − 6vv , where denotes
the derivative with respect to Ξ that can be integrated once to yield −Cv = −v −
3v 2 (assuming vanishing strains at infinity). Then, multiplying both sides by v ,
and integrating again (this is the standard quadrature trick of classical mechanical
systems [9]), leads to
1 2 C
v = v2 − v3 (3.12)
2 2
and then finally upon integration we arrive at,
√
C C
v(Ξ ) = sech2 Ξ =
2 2
√
1 C C c
s
Uξ = sech2 X − T cs + 2 C (3.13)
2( p − 1) 2 2 24

p−1
Recall that here cs = Apδ0 /M is the maximum wave speed of the linear lattice,
while C > 0 is a free parameter characterizing the speed of the nonlinear wave. Notice
also that the overall speed of the nonlinear wave cs (1 + 2 C/24) is by necessity faster
than the speed of sound, c in these systems, i.e., the nonlinear waves are supersonic.
In addition to this simpler case example of the single traveling wave approx-
imation, it is possible to obtain expressions from the KdV limit of Eq. (3.11) for
two-soliton solutions. These have the form:
k12 f 1 + k22 f 2 + 2(k2 − k1 )2 f 1 f 2 + m(k22 f 12 f 2 + k12 f 1 f 22 )

v=8 . (3.14)
(1 + f 1 + f 2 + m f 1 f 2 )2
Here, f i = e2ki (4ki τ −ξ+si ) , and m = [(k2 − k1 )/(k2 + k1 )]2 ; see e.g., [10, 11] as well
2
as the more recent work of [12], for more details on multi-soliton solutions of the KdV
equation. In the recent work of [13], this solution was compared with the dynamics
of the original granular chain for a scenario where the initial positions of the two
solitons satisfied s1 < s2 , in which case it is necessary to require k1 > k2 for the two
solitons to collide. A relevant example from this work is shown in Fig. 3.2 where
select snapshots of the profile of the two waves in the strain variable are given in
comparison to the analytical KdV approximation.
The waves of the KdV equation bear the constraint that they are unidirectional. As
such, while it is possible to obtain a qualitative (and even semi-quantitative) character-
ization of co-propagating wave collisions, it is not straightforward to formulate at the
(a) (b)
(c) (d)
(e) (f)
Fig. 3.2 (Color Online) We consider the comparison of the collision√ of two co-propagating solitary
waves for the case of parameters chosen as: p = 23 , δ0 = 1, = 0.1. Here, the KdV approximation
is valid as the perturbation parameter tends to zero. The initial condition consists of a two-soliton
solution containing waves of amplitude of 0.1 and 0.05 centered at −20 and 20, respectively. Snap-
shots at different times of the collisional evolution are shown, namely: a t = 0, b 50, c 750, d 5000, e
7500, f 15000. The solid (red) line represents the actual (non-integrable) granular lattice numerical
evolution results (the displacement u n is shown, which has arbitrary units), while the dashed (blue)
line represents the KdV two-soliton approximation. [Figure credit: Used with permission from
Ref. [13]. Copyrighted (2014) by the American Physical Society]
level of the KdV analysis counter-propagating solitary wave collisions/interactions.

Moreover, the KdV formulation requires a continuum approximation of the original
discrete granular problem. In that light, using a genuinely discrete, and if possible
bidirectional, approximate description of the granular chain, is an approach that may
feature some inherent advantages. This can be done by utilizing the Toda lattice [6]
of the form:
ü n = e[u n−1 −u n ] − e[u n −u n+1 ]

1 1
= [1 + (u n−1 − u n ) + (u n−1 − u n )2 + (u n−1 − u n )3 + · · · ] − (3.15)
2 6
1 1
[1 + (u n − u n+1 ) + (u n − u n+1 ) + (u n − u n+1 )3 + · · · ]
2
2 6
1
= (u n−1 − 2u n + u n+1 ) + [(u n−1 − u n ) − (u n − u n+1 )2 ] + · · · (3.16)
2
2
In the 2nd and 3rd lines, we have expanded the lattice into an FPUT type form
maintaining the leading order nonlinear term. This is to be compared with the for-
mulation

of Eq. (3.2). In particular, rescaling time and displacements according to
p−1
τ =t pδ0 and ũ n = p−1
δ0
un , the granular dynamics reads:
1
ũ n = (ũ n−1 − 2u n + ũ n+1 ) + [(ũ n−1 − ũ n )2 − (ũ n − ũ n+1 )2 ] + · · · (3.17)
2
where is the derivative with respect to τ . Hence, Eqs. (3.17) and (3.16) agree up to
second order, and thus the leading order error in our granular chain approximation
by the Toda lattice will stem from the cubic term (for which it is straightforward to
show that it cannot be matched between the two models).
In the case of the Toda lattice, the single soliton has the form:

1 + exp[−2kn ± 2(sinh k)t]
u n = − ln . (3.18)
1 + exp[−2k(n − 1) ± 2(sinh k)t]
and considerably more complex (not given here) two-soliton solutions exist. Once
again, one can utilize the analytically available expressions for the evolution of
multi-solitons of the Toda lattice, comparing their interaction to that of the granular
crystal chain. Following [13], we give also an example of this type in Fig. 3.3. The
Toda lattice appears to be capturing the counter-propagating soliton dynamics of our
granular chain, both before, during, and after the collision.
It is clear from the above analysis as well as numerical observations that for the
weakly nonlinear (highly precompressed) case, the granular chain dynamics can be
adequately approximated by suitable integrable (and analytically tractable) models.
However, this is far less evident to be a priori true in the strongly nonlinear (weakly
precompressed) case. It is to that latter case that we now turn.
(a) (b)
(c) (d)
(e) (f)
Fig. 3.3 (Color Online) In this example, the parameter values p = 23 , δ0 = 1, k = 0.3 are used.
k is the sole parameter of the soliton solution of the Toda lattice. The initial condition consists
of two solitons of the same amplitude at −50 and 50. Here, the granular crystal (non-integrable)
dynamics is also compared to the addition of two one-soliton solutions of the Toda lattice. The
snapshots shown are at a t = 0, b 30, c 37, d 40, e 50, f 150. The (red) solid line is for the actual
(numerical) granular lattice dynamics (the displacement u n is shown, which has arbitrary units),
the (blue) dashed line is the plain superposition of two Toda one-soliton solutions of Eq. (3.18),
and the (green) dash-dotted line represents the numerical evolution of the Toda chain. The three
curves are nearly coincident for the snapshots considered. [Figure credit: Used with permission
from Ref. [13]. Copyrighted (2014) by the American Physical Society]
3.3 Traveling Waves Without Precompression I: The Continuum Limit 31
3.3 Traveling Waves Without Precompression I: The

Continuum Limit
Historically, the subject of traveling waves in granular crystals was arguably seeded
by the ground-breaking work of Lazaridi and Nesterenko [4] (see also [2] for a
more detailed treatment) who introduced a new type of solitary wave. This stemmed
from striking one side of an uncompressed monoatomic granular chain with another
particle. The resulting wave exhibited much faster spatial decay than the typically (in
one dimension) exponential one of solitary waves, and it was thus pondered whether
the waves might even have compact support [2]. Such waves have been obtained in
a wide array of equations bearing features of nonlinear dispersion [14] and are now
collectively known as compactons; see, e.g., the recent review of [15].
In this case, given that the precompression is absent, i.e., δ0 = 0, Eq. (1.3) is purely
nonlinear, and it is very challenging to study it analytically because the nonlinearity
is not smooth (V (0) is undefined in the absence of precompression). The seminal
analytical attempt in this direction was due to Nesterenko [16], who derived an
explicitly solvable PDE as an approximation to Eq. (1.3). This follows the standard
procedure in discrete models by using a long-wavelength limit, as done in the previous
section, to derive the KdV equation. In particular, assuming δ0 = 0 and a uniform
(absorbable in a rescaling of time) factor of A/M in Eq. (1.3), Nesterenko used (what
is equivalent to) the ansatz u n (t) = U (X, T ), where X = n, T = t and substituted
directly into Eq. (1.3). Using once again the Taylor expansion given in Eq. (3.4) the
resulting long-wavelength dynamical equation for U (X, T ) reads:

2 p( p − 1)
∂T2 U = −∂ X p
(−∂ X U ) + 2 p
∂ X (−∂ X U ) − (−∂ X U ) p−2 2
∂X U 2 . (3.19)
12 2
One can go a step further and evaluate the corresponding equation for the strain
Y = ∂ X U , by differentiating Eq. (3.19) with respect to X , obtaining the PDE:

2 p( p − 1)
∂T2 Y = −∂ X2 (−Y ) +
p
∂ X (−Y ) −
2 p
(−Y ) ∂ X Y
p−2 2
(3.20)
12 2
Equations (3.19)–(3.20) have a number of advantages and disadvantages. On the

one hand, one can seek explicit compacton solutions in the form of trigonometric
functions in the co-traveling frame, i.e., dependent only on ξ = X − cT . Upon this
reduction ∂ X Y = c∂ξ Y , ∂ X2 Y = c2 ∂ξ2 Y , etc., and hence the PDE of, say, Eq. (3.20)
becomes an ODE which turns out to have exact solutions of the form [17]:
Y (X − cT ) = cm A1 cosm (B(X − cT )), (3.21)

1−1 p
2
m = 2/( p − 1), A1 = , B = 6( p − 1)2 /( p( p + 1)). (3.22)
1+ p
While these are periodic solutions, one can envision a process of truncating a period
of these solutions and “gluing” it with zeros on both sides, forming a compactly
supported entity. Such entities are remarkably close (see also the comparison below)
to the findings of the experiments for the Hertzian chains with p = 3/2 [2].
On the other hand, there are clear pathologies to such a long-wavelength pro-
cedure. On the one hand, the expansion process is rather ad hoc given that there
is no genuine small parameter in the system. For this reason, the derived contin-
uum models are often called quasi-continuum models [2]. The width of the resulting
entity (characterized, roughly, by 1/B in the expression of Eq. (3.21)) is of O(1)
and these solitary waves are known to extend over only about 5 sites. Hence, there
is no naturally small scale (such as the lattice scale over the spatial extent of the
wave) permitting us to apply such a procedure rigorously. A related feature is that,
considering the potential extension of this procedure to the linear case of p = 1,
one immediately is met with nontrivial problems. More specifically, using a plane
wave u ∼ ei(kx+ωt) to explore the background modes of the linear system ( p = 1),
we obtain the dispersion relation:
2 4
ω 2 (k) = k 2 − k (3.23)
12
which leads to complex eigenfrequencies and,√hence, unstable growth (even at this
linear level) for sufficiently large k (i.e., |k| > 12/). This is naturally problematic
for the evolutionary dynamics of the system. While mathematically, it is an open
problem whether the case of p > 1 is ill-posed, to the best of our knowledge, it
seems far less promising as an evolutionary dynamical model.
Lastly, we show a related (long-wavelength) model used in works like [17–20].
Using the strain formulation of Eq. (1.6), obtaining a continuum limit is, in fact,
simpler. Using the ansatz Y (X, T ) where X = n and T = t, and the identity given
in Eq. (3.5) (where one applies the formula for Y p instead of U ), one can obtain the
long-wavelength PDE at the strain level directly:
2 4 p
∂T2 Y = ∂ X2 (Y p ) + ∂ (Y ). (3.24)
12 X
Once again, there are advantages and disadvantages to the approach. Upon truncation,
there is once again an explicit compacton solution that can be constructed in the form:
Y (X − cT ) = cm A1 cosm ( B̃(X − cT )), (3.25)

1
p + 1 1− p √
m = 2/( p − 1), A1 = , B̃ = 3( p − 1)/ p. (3.26)
2p
As can be seen in Fig. 3.4, both the result of Eq. (3.21) and that of Eq. (3.25) yield
a very good approximation to the numerically exact traveling wave (which we will
explore how to compute in the next section). Moreover, these expressions (see the
right panel of the figure) yield the correct scaling in connection with experimental
results of [21]. Nevertheless, it should be remarked that the approximation shown in
the panel of Fig. 3.4a for p = 3/2 becomes less adequate as p is increased.
3.3 Traveling Waves Without Precompression I: The Continuum Limit 33
(a) (b)
1.6
1.4
1.2
1
Y (ξ)
0.8
0.6
0.4
0.2
0
-4 -2 0 2 4
Fig. 3.4 a Genuine traveling solitary wave (given by the line connecting the points) compared with
continuum approximations. The approximations of Eqs. (3.22) and (3.26) are shown by dashed
and dotted lines, respectively. The results are shown for p = 3/2 where the agreement is fairly
reasonable, yet the long-wavelength, continuum approximations become worse as p increases [17].
b Scaling properties of the velocity of the solution from experiments versus its maximal force. The
theoretical scaling accurately capturing the experiments is shown by the solid line. [Figure credit:
Panel (a) used with permission from [17], Copyright (2009) by the American Physical Society.
Panel (b) used with permission from [21], Copyright (1997) by the American Physical Society]
Moreover, the model of Eq. (3.24) suffers from similar pathologies (includ-
ing short wavelength—or, equivalently, large wavenumber—instabilities as per
Eq. (3.23)). In fact, both long-wavelength models share the same dispersion relation
and the associated issues. It may be considered as an additional source of concern
that the processes that give the relevant evolutionary equation going towards the con-
tinuum limit yield distinct results depending on whether the limit is first taken for
displacements and then one transforms into strains, or whether one first goes to the
discrete equation for strains and then takes the continuum limit. This lack of com-
mutation of the two processes (transforming to strains and taking the limit) produces
distinct equations with similar, yet distinct solutions and hence naturally begs the
question of which, if any, of the two should be considered as the “right” one. Lastly,
as we will see in the next section, both processes suggest that the solution of the
traveling wave form is one which is compactly supported. As the detailed analysis of
the discrete problem will now show, neither of the two is accurate in this prediction.
3.4 Traveling Waves Without Precompression II:

Asymptotic and Numerically Exact Results
Since the inception of this field of granular crystals [4] (and throughout its early
accounts [2]), perhaps the most remarkable, highly studied, and structurally robust
feature of these lattices has been the stable traveling solitary wave solutions in a
homogeneous lattice. We now turn to the examination of the fundamental mathemat-
ical issue of the existence of such genuine traveling wave solutions (in a mathematical
sense). Genuine traveling waves are ones that travel through a granular chain without
any disturbance to their original shape. We will explore these in the most prototypical
case here where the precompression is absent. Working directly at the level of the
strain variables of Eq. (1.6) and substituting yn (t) = Φ(n − ct, t) = Φ(ξ, t) leads to
∂t2 Φ(ξ, t) = −c2 ∂ξ2 Φ(ξ, t) + 2c∂ξ t Φ(ξ, t)+

3/2 3/2 3/2
(Φ(ξ − 1, t))+ − 2 (Φ(ξ, t))+ + (Φ(ξ + 1, t))+ .
(3.27)
Traveling waves of Eq. (1.3) correspond to stationary (i.e., time-independent) solu-

tions Φ(ξ, t) = φ(ξ) of Eq. (3.27). These satisfy
3/2 3/2 3/2
0 = −c2 ∂ξ2 φ + {[φ(ξ − 1)]+ − 2 [φ(ξ)]+ + [φ(ξ + 1)]+ } . (3.28)
See Fig. 3.1c for an example. By definition, traveling solitary waves are solutions
of Eq. (3.28) that satisfy limξ→±∞ φ(ξ) = 0. The existence of such solutions fol-
lows from the general result of Friesecke and Wattis [22], which holds for traveling
wave solutions of general FPUT lattice equations. In [23], this result was applied to
Eq. (1.6). While this proves the existence of a traveling solitary wave, it does not pro-
vide further insight on its form. It was argued heuristically in [24] that the tails follow
a doubly exponential decay. This is in contrast to the prediction of a compact wave-
form implied by exact solutions, such as Eqs. (3.22) or (3.26), of formally derived
continuum models. The double exponential nature of the decay was subsequently
proved in [15] by formulating the advance–delay
∞ equation (3.28) as a fixed-point
problem. Using the transform φ(ξ) = −∞ φ̂(k)eikξ dk leads from Eq. (3.28) to

1 k
φ̂(k) = 2 sinc2 φ3/2 . (3.29)
c 2
The hat denotes the (wavenumber-dependent) Fourier transform of the spatial

variable. Fourier transforming back to (real) ξ-space and using the properties of the
convolution leads to
∞
φ(ξ) = Λ ∗ φ =
3/2
Λ(ξ − y)φ3/2 (y)dy (3.30)
−∞
where Λ(ξ) = (1/c2 )[ 1 − |ξ| ]+ is the so-called tent function. Without loss of gener-
ality (given the scaling properties of Eq. 1.6), we can assume c = 1. Using z = ξ − y
and changing variables in (3.30) then yields
1
Λ(z)φ3/2 (ξ + 1 − z)dz ≤ φ3/2 (ξ) ⇒ φ(ξ + n) ≤ φ(ξ)( 2 ) ,
3 n
φ(ξ + 1) =
−1
(3.31)
3.4 Traveling Waves Without Precompression II: Asymptotic … 35
where we used only the fact that sup y∈[−1,1] φ3/2 (ξ + 1 − y) = φ3/2 (ξ) for a positive,
1
monotonically decreasing traveling wave (and the fact that −1 Λ(z)dz = 1).
From the above, we obtain the doubly-exponential decay for traveling solitary
waves in granular chains with a Hertzian potential p = 3/2, while for general expo-
pn
nents p, the result is that φn ∼ φ0 as n → ∞. An important additional feature of
this calculation is that Eq. (3.30) provides an efficient numerical algorithm for the
computation of such waves through a fixed-point iteration scheme. This can be done
either at the level of Eq. (3.29), i.e., in Fourier space, or at the level of real space
through Eq. (3.30).
A typical result of the numerical implementation of this iterative procedure is
shown in Fig. 3.5. Panel (a) shows the profile of the solution and its momentum
(notice also the inset showcasing its rapid decay) and panel (b) confirms the exact
traveling nature of the solution, when seeded to the lattice, as well as its unit speed.
(a) (b)
(c)
(d)
Fig. 3.5 a Result of the iterative procedure (in real space, see Eq. (3.30)) providing the strain
profile (solid) and its time derivative (dashed line) in all of space; the points denote the lattice
nodes. The inset in semi-log suggests the faster than exponential decay. b Initialization of this wave
in the original strain lattice of Eq. (1.6) confirming its robust traveling with the prescribed speed of
c = 1. c The continuation of the spatial strain profile over precompression (see the discussion in
the text). d This panel separates between cases without precompression (red solid) and those with
precompression (black dashed and blue dash-dotted). The insets of the former and of one of the
latter in semi-log clearly distinguish between their distinct spatial decay
As an aside, it is worthwhile to note that the same procedure can also be used in
the case where the precompression is non-vanishing. In the latter case, however,
the waves do not exist for arbitrary speeds. As per our considerations in Sect. 3.2
(see also [2] for relevant discussion), the coherent structures can all be identified
when not resonating with the linear spectrum, i.e., they are supersonic and exist for
√ p−1
speeds beyond the speed of sound, namely c > cs = pδ0 2 . In this latter case,
the precompression absorbs most of the decay in the estimate of Eq. (3.31), leaving
φ(ξ + 1) bounded above by φ(ξ) multiplied by a suitable prefactor. Decay in this
case is thus exponential, as observed both numerically and experimentally [2], and
implied by the connection to the KdV and Toda lattice coherent structures in Sect. 3.2,
see Eqs. (3.13) and (3.18). A continuation in the precompression parameter shown
in panel (c) of Fig. 3.5 illustrates how as the precompression increases (and the
speed of sound also increases for p = 3/2), the wave eventually disappears for fixed
speed c (as the sound speed cs increases towards c). Even more meaningful perhaps
is panel (d) which distinguishes the solitary waves in the lattice with and without
precompression. The left inset in (d) is the case without precompression, which
shows the wave in a semi-log scale with a decay faster than linear (hence the overall
decay being faster than exponential, which is in line with the doubly exponential
expectation for δ0 = 0). On the other hand, the right inset of (d) is the case with
precompression, which shows a linear decay in the semi-log scale, reflecting the
exponential decay of the waveform in that case.
It is especially relevant to note here that the progress reported in the applied/
computational analysis of such waveforms is also paralleled by corresponding recent
steps towards the experimental visualization of these traveling waves, and also their
rigorous proof of existence. In this vein, Fig. 3.6 illustrates recent experiments from
the work of [26]. The use of a laser Doppler vibrometer enables the full space-time
resolution of the motion of a traveling wave, its reflection from the wall in the short
experimental chain and return propagation in the opposite direction. The result is all
the more remarkable when realizing that it has been reconstructed on the basis of
the scoping and measuring of the velocity of each of the lattice beads (in individual
realizations of the experiment for each one of them)! These signal the remarkable
advances in experiments, since the only way that they can be distinguished from the
numerical evolution of the velocities within the traveling wave of the right panel is
because the former have a hint of residual vibration after the wave has passed, while
the exact numerical traveling wave leaves no trace of residual excitation behind it.
On the rigorous side, which we briefly touch upon here, the (square root of)
operator of relevance to the fixed point iteration Q f (ξ) = sin(πξ)
πξ
fˆ(ξ) has a Fourier
x+1/2
transform as Q f (x) = x−1/2 f (y)dy. In the recent work of [27], this was used to
set up a constrained optimization problem of the following form:

J (v) = ε−1−1 |Qv(x)|2 d x → max
ε
subject to−εI (v) = v q (x)d x = 1, (3.32)

v ≥ 0, v bell − shaped,
3.4 Traveling Waves Without Precompression II: Asymptotic … 37
Fig. 3.6 a The space-time evolution of an experimentally measured (in terms of the velocities of
the beads) traveling wave. b Speeds of beads 15–19 within the chain in the experiment and c in
numerical simulation [Figure credit: Experimental results and figure obtained by the group of Prof.
J. Yang, University of Washington cf. Ref. [26]. Used with permission]
where q = 1 + p ∈ (1, 2), and is an appropriately chosen small parameter. It was

1
proved in [27] that the Euler-Lagrange equation associated with this constrained
extremization is equivalent to obtaining the solution of Eq. (3.30) and, equally impor-
tantly, that this problem of calculus of variations has a solution which corresponds
indeed to the bell-shaped traveling wave solution of the monoatomic problem in
the strain variables. A similar path was followed in the case with precompression
in [28]. Thus, at this stage, we possess a fairly complete characterization of the rigor-
ous existence, asymptotic decay, and experimental characterization of such traveling
waves.
However, a front that is far more open for future investigations is that of the stability
of traveling solitary waves. A natural starting point is to investigate the spectral
stability of stationary solutions φ0 of Eq. (3.28). More specifically, one substitutes the
linearization ansatz Φ(ξ, t) = φ0 + εa(ξ)eλt into Eq. (3.27) to obtain the eigenvalue
problem
λ2 a(ξ) = −c2 ∂ξ2 a(ξ) + 2λc∂ξ a(ξ)+

3 0 1/2 1/2 1/2
φ (ξ − 1) a(ξ − 1) − 2 φ0 (ξ) a(ξ) + φ0 (ξ + 1) a(ξ + 1) . (3.33)
2 + + +
The solution is called spectrally stable if all eigenvalues are non-positive and other-
wise unstable. However, understanding the spectrum of eigenvalues λ and eigenfunc-
tions a of this linearized operator is a non-trivial (and open) task. It is of particular
interest in its own right. See Appendix A.1 for details. A related recent perspec-
tive [29] (see also the earlier work of [30]) is that of viewing the problem as a
periodic orbit computation. Namely, a traveling wave on a lattice can be viewed as
a periodic orbit upon integrating the time evolution for a period T = h/c and then
shifting the solution back by one lattice site. This paves the way for the use of the tools
of Floquet theory to address the stability of traveling waves. Moreover, it enables the
identification of stability criteria, which have already been established for periodic
orbits [31]. In particular, if the energy of the wave changes monotonicity as a function
of its speed it can be proved that it must transition from stability to instability (or vice
versa). However, the relevant understanding of stability properties is still far from
becoming systematic and we will not attempt to address it here. Rather, we suggest
it as an important field in need of significant advances through future work.
References
1. Y. Starosvetsky, K. Jayaprakash, M.A. Hasan, A. Vakakis, Dynamics and Acoustics of Ordered

Granular Media (World Scientific, Singapore, 2017)
462, 21 (2008)
4. A.N. Lazaridi, V.F. Nesterenko, Observation of a new type of solitary waves in one-dimensional
granular medium. J. Appl. Mech. Tech. Phys. 26, 405 (1985)
6. M. Toda, Theory of nonlinear lattices (Springer, Heidelberg, 1989)
9. P.G. Drazin, R.S. Johnson, Solitons: An Introduction (Cambridge University Press, Cambridge,
1989)
10. R. Hirota, Exact solution of the Korteweg-de Vries equation for multiple collisions of solitons.
Phys. Rev. Lett. 27, 1192–1194 (1971)
11. H.D. Wahlquist, F.B. Estabrook, Bäcklund transformation for solutions of the Korteweg-de
Vries equation. Phys. Rev. Lett. 31, 1386–1390 (1973)
12. T.R. Marchant, Asymptotic solitons of the extended Korteweg-de Vries equation. Phys. Rev.
E 59, 3745–3748 (1999)
13. Y. Shen, P.G. Kevrekidis, S. Sen, A. Hoffman, Characterizing traveling-wave collisions in
granular chains starting from integrable limits: the case of the Korteweg-de Vries equation and
the Toda lattice. Phys. Rev. E 90, 022905 (2014)
14. P. Rosenau, J.M. Hyman, Compactons: solitons with finite wavelength. Phys. Rev. Lett. 70,
564–567 (1993)
15. B. Dey, Compactons, Preprint (2017)
16. V.F. Nesterenko, Propagation of nonlinear compression pulses in granular media. J. Appl.
Mech. Tech. Phys. 24, 733 (1983)
References 39
17. K. Ahnert, A. Pikovsky, Compactons and chaos in strongly nonlinear lattices. Phys. Rev. E 79,
026209 (2009)
18. M.A. Collins, A quasicontinuum approximation for solitons in an atomic chain. Chem. Phys.
Lett. 77, 342 (1981)
19. D. Hochstrasser, F. Mertens, H. Büttner, An iterative method for the calculation of narrow
solitary excitations on atomic chains. Physica D 35, 259 (1989)
20. J.A.D. Wattis, Approximations to solitary waves on lattices. II. Quasi-continuum methods for
fast and slow waves. J. Phys. A Math. Gen. 26, 1193 (1993)
21. C. Coste, E. Falcon, S. Fauve, Solitary waves in a chain of beads under Hertz contact. Phys.
Rev. E 56, 6104 (1997)
22. G. Friesecke, J.A.D. Wattis, Existence theorem for solitary waves on lattices. Commun. Math.
Phys. 161, 391 (1994)
23. R.S. MacKay, Solitary waves in a chain of beads under Hertz contact. Phys. Lett. A 251, 191
(1999)
24. A. Chatterjee, Asymptotic solution for solitary waves in a chain of elastic spheres. Phys. Rev.
E 59, 5912 (1999)
25. J.M. English, R.L. Pego, On the solitary wave pulse in a chain of beads. Proc. AMS 133, 1763
(2005)
26. E. Kim, R. Chaunsali, H. Xu, J. Castillo, J. Yang, P.G. Kevrekidis, A.F. Vakakis, Nonlinear
low-to-high frequency energy cascades in diatomic granular crystals. Phys. Rev. E 92, 062201
(2015)
27. A. Stefanov, P.G. Kevrekidis, On the existence of solitary traveling waves for generalized
Hertzian chains. J. Nonlinear Sci. 22, 327 (2012)
28. A. Stefanov, P.G. Kevrekidis, Traveling waves for monomer chains with pre-compression.
Nonlinearity 26, 539 (2013)
29. J. Cuevas-Maraver, P.G. Kevrekidis, A. Vainchtein, H. Xu, Unifying perspective: solitary trav-
eling waves as discrete breathers in Hamiltonian lattices and energy criteria for their stability.
Phys. Rev. E 96, 032214 (2017)
30. G. Friesecke, R.L. Pego, Solitary waves on Fermi-Pasta-Ulam lattices: III Howland-type Flo-
quet theory. Nonlinearity 17, 207 (2004)
31. P.G. Kevrekidis, J. Cuevas-Maraver, D.E. Pelinovsky, Energy criterion for the spectral stability
of discrete breathers. Phys. Rev. Lett. 117, 094101 (2016)
Chapter 4
Discrete (Dark) Breathers
4.1 Discrete Breathers: Overview and Theoretical Analysis
As indicated in the introductory chapter, discrete breathers are one of the most
prototypical excitations that can arise in nonlinear dynamical lattices. They are peri-
odic in time and strongly localized in space [2, 11]. In integrable systems such as the
sine-Gordon equation, such solutions were long known [4, 5] but were also believed
to be rather special. However, a series of investigations by Sievers, Takeno, Page, and
others [6, 7] in the late 1980s and early 1990s suggested that they might be generic for
nonlinear lattices of the Klein–Gordon (i.e., onsite nonlinearity) and FPUT (i.e., inter-
site nonlinearity) types. Then, the work of [8] provided a mathematical foundation
for this expectation illustrating that under (fairly generic) non-resonance conditions,
discrete breathers should be expected to persist in such nonlinear, spatially extended
discrete dynamical systems.
Given the widespread appeal of discrete breathers in systems ranging from the
DNA double strand denaturation to electrical lattices, and from arrays of coupled tor-
sion pendula to superconducting Josephson junctions and nonlinear optics or atomic
Bose-Einstein condensates [2], it seems perhaps surprising that no attempt was made
to identify such structures in granular crystals prior to 2010 [9, 10]. This first observa-
tion of discrete breathers in granular chains took place, however, in a heterogeneous
(dimer) system. In view of that, we will consider it in the next chapter which will be
dedicated to the heterogeneous granular setting.
The breathers just discussed are also called bright breathers, since the solution is
localized in space (i.e., the tails vanish at infinity). The term bright is motivated by
studies of breathers in optical systems, where such solutions have the light intensity
greatest at the center, and hence appear to be “brightest” there [11]. In this chapter, we
will explore a different (but related) excitation that is the only breather type relevant
for homogeneous granular chains, namely, the dark breather [1]. For a dark breather,
the localization occurs somewhat “in reverse” in comparison to a bright breather; the

42 4 Discrete (Dark) Breathers
tails approach a nonzero state and the displacement is (nearly or exactly) zero at its
center.
To help understand why a dark breather is the only relevant breather in the homo-
geneous granular chain, we start with some basic analytical considerations. Numer-
ical considerations will be discussed later, see also Appendix A.2. We begin with
the assumption of small strain |yn | δ0 , so rather than working directly with the
equations in the strain variable formulation, Eq. (1.5), we will work with its Taylor
expansion:

M ÿn = VFPUT (yn+1 ) − 2VFPUT (yn ) + VFPUT (yn−1 ). (4.1)
3/2 1/2
where VFPUT (x) = K 1 + K 2 x + K 3 x 2 + K 4 x 3 with K 1 = Aδ0 , K 2 = 23 Aδ0 ,
−1/2 3 −3/2
K 3 = −A 38 δ0 , K 4 = −A 48 δ0 is the Taylor expansion of the Hertzian force
3/2
A [δ0 − x]+ . In this Section we will assume M = 1 (indeed, we can reduce the
number of parameters to one through a scaling of time and amplitude). Note that this
is similar to what we did to arrive at Eq. (3.2), but now at the level of strain, rather
than the level of displacement.
We once again now deploy a technique that is particularly popular in the realm
of nonlinear wave equations, namely that of multiple-scale expansions [12]. The
fundamental premise is relatively simple. We look for a solution with a partic-
ular (potential breathing, as is the case here) structure. We work in appropriate
scales of space and time that will convert the original dynamical system of interest
into one of the “prototypical” models (such as the nonlinear Schrödinger equation
(NLS)) bearing such characteristics. Then, examining the dynamics order by order,
we retrieve at the suitable order (for NLS, this is typically the cubic order) such a
PDE. Then, its explicitly available analytical solutions give us a guideline of how to
approximate –admittedly, within the proper, already prescribed scales of space and
time– the waveforms of the original system. We have seen this approach already in
Sects. 2.1 and 3.2, where we derived the Burgers’ and KdV equations, respectively.
To derive the NLS equation, we will use the ansatz
yn (t) = Y (X, T )ei(kn+ωt) + c.c., (4.2)
where c.c. is the complex conjugate and the long and slow variables are X = (n − ct)
and T = 2 t, respectively. This choice may appear to be somewhat unexpected at
first, however, it is possible to convince ourselves that it is the most natural one.
We scale space by , in which case the dispersive scaling of NLS’ space and time
derivatives necessitates that we use an 2 for the time rescaling. As regards the
ansatz of the solution, we seek a solution that explicitly includes a breathing in time
(with frequency ω) and a spatial modulation with wavenumber k. The motivation
here is that we have in mind a rapidly oscillating plane wave that is modulated by
a slow envelope provided by Y (X, T ). A result of the substitution of this ansatz
in Eq. (4.1) is that at O(), we retrieve the dispersion relation of the system, i.e.,
4.1 Discrete Breathers: Overview and Theoretical Analysis 43
ω = 4K 2 sin2 ( k2 ). At the second order O(2 ), we are able to determine the group
velocity c to be c = −ω (k), where ω = dω
dk
. Subsequently, at the cubic order O(3 ),
the NLS equation is retrieved in the form:
i∂T Y (X, T ) + ν2 ∂ X2 Y (X, T ) + ν3 Y (X, T )|Y (X, T )|2 = 0. (4.3)
See Appendix A.3 for the details of the full derivation. In this expression, the
coefficient of the dispersion typically (in such calculations) reflects the (opposite
of the) concavity of the dispersion relation, i.e., ν2 = −ω (k)/2, while the non-
linearity coefficient is given by a considerably more elaborate expression account-
ing for the specifics of the potential of interest, see Appendix A.3. The expres-
sion simplifies considerably when the wavenumber is at the band edge, ν3 |k=π =
−3/2
K 2 (3K 2 K 4 − 4K 32 ) = B, which will be the wavenumber of interest for standing
breathers (see below). In the case of the granular
√ chain, the dispersion relation extends
from vanishing frequencies up to ω0 = 2 K 2 . It is well-known in the theory of dis-
crete breathers [2, 3] that to seek breathers, one needs to search near the edges of the
passbands. The passbands give a natural place from which the breathers can bifurcate.
Indeed, in the granular chain with no precompression (and hence there is no passband
to speak of since that system is non-linearizable) it was shown in [13] that the only
time-periodic state is the trivial (zero) one. In the homogeneous granular chain with
precompression, the edge at ω = 0 is not an option for the breathers to bifurcate from,
hence it is natural to search in the vicinity of ω = ω0 , i.e., near k = π at the edge of the
Brillouin zone. There the group velocity satisfies c = ω (π) = 0 (i.e., the breather
is standing) and ν3 assumes a familiar form (for researchers that have worked in
−3/2
such systems; see, e.g., the discussion of [2]): ν3 |k=π = K 2 (3K 2 K 4 − 4K 32 ) = B
which satisfies B < 0 for our problem. At this same band edge, the concave down
form of the dispersion relation gives ν2 > 0. This combination of coefficients ensures
that we are in the realm of the so-called defocusing NLS equation [14] for which
solutions that are heteroclinic connections in the form of hyperbolic tangent profiles
are possible. For Eq. (4.3), the so-called dark soliton has the form,

−κ
Y (X, T ) = κ/B tanh (X − X 0 ) eiκT , κ < 0. (4.4)
2ν2
Thus, the corresponding reconstruction of the approximate solution with k = π of

the original FPUT problem Eq. (4.1) is of the form:

κ −κ
yn (t) = 2(−1) n
tanh (n − x0 ) cos(ωb t), (4.5)
B 2ν2
where ωb = ω0 + κ2 . Here the (−1)n ≡ eiπn represents the spatial alternation
between adjacent sites at the edge of the Brillouin zone; the cos(ω0 t) is the tem-
poral vibration associated with the breathing of the wave form. The x0 is an arbitrary
shift parameter at this level of analysis. The value of x0 = 0 corresponds to an onsite
waveform and x0 = 0.5 corresponds to an intersite waveform.1 Finally, the tanh

captures the solitary wave of the defocusing NLS equation and κ parametrizes the
mono-parametric family of solitons of this PDE.
If the nonlinear coefficient satisfies B > 0, then the corresponding NLS equation
would be focusing, which has bright soliton solutions. In that case, Eq. (4.2) would
yield an approximate bright breather. However, since B < 0 for any combination
of parameters of the monoatomic granular chain (assuming that p > 1), this leads
us to the conclusion that dark breathers are the only breather waveform possible in
the homogeneous granular chain. The nonexistence of bright breathers in the FPUT
lattice (and hence granular chain for small strains) result has been proved rigorously
in [16] (using techniques different than multiple-scale analysis).
Some prototypical case examples of such a dark breather waveform, illustrating
both the alternating spatial structure and the tanh envelope, are shown in Fig. 4.1,
drawn from the work of [1]. Interestingly, there are two case examples of a dark
breather that can be identified in the Hamiltonian lattice of Eq. (4.1). Panel (a) shows
the onsite centered one, while panel (b) the intersite centered one. Such a distinction
is customarily known to arise in discrete systems [2]. As we explain in more detail
in the relevant Appendix A.2, the breathers in Fig. 4.1 are identified through an
iterative process where a Newton-Raphson, fixed point iteration step is performed
every time upon return of the solution to the so-called Poincaré section (associated
with its period). An additional advantage of the process is that the matrix used for
the Newton iteration is directly related to the so-called monodromy matrix whose
eigenvalues indicate whether small perturbations grow/decay (or oscillate) over the
course of a period. These so-called Floquet multipliers determine the stability of the
(a) (b)
Fig. 4.1 Hamiltonian discrete breathers analyzed in this section via the multiple-scale approxima-
tion in the strain variable yn = u n−1 − u n . a The center of the onsite waveform is located at a lattice
node (n = 0 in this example). b The center of the intersite waveform is located between two lattice
nodes (between n = 0 and n = 1 in this example). [Figure credit: Used with permission from [1],
Copyright (2011) by the American Physical Society]
1 Ata much more refined level of analysis involving exponential asymptotics, it can be seen that
only these two cases (x0 = 0 and x0 = 0.5) persist as solutions of the genuinely discrete problem.
A relevant qualitative discussion for a different model can be found, e.g., in [15].
4.1 Discrete Breathers: Overview and Theoretical Analysis 45
periodic solution, in accordance with Floquet theory; more details on the specifics of
this are provided in Appendix A.2. In Fig. 4.2a,c, we see the branch of the relevant
breather solutions for different frequencies ωb in the vicinity of the band edge of the
dispersion relation. The Floquet multipliers are shown in the in Fig. 4.2b,d. Whenever
they feature an absolute value |λ| > 1, they are associated with instability. Hence,
we can see that the dark breathers are stable in the immediate vicinity of the point
where they bifurcate from the band edge of the linear spectrum. As their frequency
deviates from this limit, they may be more robust as is the case for onsite centered
configurations in the top panel, or less robust as is the case with intersite centered
ones (yet no longer genuinely stable).
(a) (b)
(c) (d)
Fig. 4.2 Bifurcation diagrams (left panels) and corresponding stability diagrams (right panels)
for the onsite (top panels) and intersite (bottom panels) Hamiltonian discrete breathers shown in
Fig. 4.1. In a and c the renormalized l 2 norm of the solution as a function of the breather frequency
is shown. The renormalized l 2 norm is the l 2 norm of the solution with the background subtracted.
The solid (blue) refers to segments of the branch without pronounced instabilities (associated with
real Floquet multipliers), while the dashed (red) refers to ones where such instabilities are present.
In b and d the modulus of the Floquet multipliers is plotted vs breather frequency ω. The Floquet
multipliers indicate whether the solutions grow (|λ| > 1), decay (|λ| < 1) or oscillate and do not
decay (|λ| = 1). [Figure credit: Used with permission from [1], Copyright (2011) by the American
Physical Society]
4.2 Connection with Experimental Results
A natural question that comes to mind regarding these elaborate structures (i.e.,
the dark breathers) is whether there is a realistic possibility to observe them in
physical experiments. While such experiments can be devised (as discussed in the
two previous chapters) for the more straightforward traveling wave and even shock-
like excitations, the dark breathers bear a much more complex wave pattern involving
both the alternation between adjacent sites and the tanh profile. One thing that is
important to remind the reader here is that, despite tremendous versatility in the
experimental setup (different materials, different geometric shapes of the elements,
different masses, etc. are all natural possibilities), it does not have the potential
of distributed actuation. I.e., up to this point in time, granular crystals constitute
systems that can be accurately (and in a non-autonomous fashion) driven from the
boundaries. However, they cannot be initialized at will; the latter is an especially
intriguing challenge for future work. Hence, a natural question arises about how to
induce a dark breather in both theory and experiment in the absence of such an ability
to initialize.
To bypass this constraint, we revert to the “dark” nature of the dark breather. Such
a state can be thought of as arising from the destructive interference of two waves
(near the center of the lattice). Hence, we can use this intuition to drive with opposite
phase two plane waves at frequencies very near the upper band edge where we saw
that dark breathers may exist. In this way, we enable the transmission of the plane
waves (which only occurs within the linear band of the dispersion relation). Yet, as
these waves propagate towards the center of the domain, the judicious selection of
an anti-phase pattern (i.e., the π-phase difference) produces an “interference fringe”
at the center, which is preserved by the parity of the evolution dynamics. Around
this vanishing center, a nonlinear pattern will emerge in the presence of the Hertzian
nonlinearity (which will also respect the “staggered” i.e., proportional to (−1)n ,
nature of the plane waves). However, which is the pattern that has a nodal point at
the center, vibrates at frequency ωb (near the band edge ω0 ), has a staggered spatial
structure and is fundamentally nonlinear? It is, indeed, the dark breather. As a result,
it would seem rather natural that this process produces a dark breather state in this
driven (through the boundaries), yet intrinsically damped (in realistic experiments)
system.
A relevant prototypical example that showcases that this idea works experimen-
tally is shown in Fig. 4.3, drawn from the work of [17]. Panel (a) gives a glimpse of the
dark breather as formed in the simulation (red dots) versus the numerical exact dark
breather solution (blue circles) and the experimental evolution result (green lines).
Panel (b) shows the exact (obtained via Newton-Raphson) numerical time-periodic
solution compared to the one obtained in direct numerical simulations (middle) and
in experiments (right) when following the “protocol” discussed in the previous para-
graph. Notice the formation and persistence of the nodal site at the middle and the
spontaneous emergence around it of the dark breather which survives throughout the
duration of the experiment (and of the corresponding numerical simulation).
4.2 Connection with Experimental Results 47
(a) (b)
(c)
Fig. 4.3 a Theoretical prediction for the velocity profile of a dark breather (its numerically exact
profile is given by blue circles), as well as the evolution (green lines) of such a waveform over time in
the experiment. The result of performing the same process as the experiment numerically produces
the (red) dots. b Comparison of the detailed time evolution of the numerically exact (up to prescribed
tolerance) dark breather (left) versus the numerical simulation (middle) and experimental process
(right) output. Color intensity corresponds to velocity. c Comparison of the bifurcation diagram of
the dark breather solution in the driven (and damped) variant of the Hertzian model (solid blue lines
denote the stable and dashed red the unstable segments) versus the corresponding experimental
results (given by black points). The diagnostic used is the velocity of the 10th bead (i.e., well within
the bulk) of the chain. A zoom of the area boxed by a gray dashed line is shown in the inset, showing
a pitchfork bifurcation loop. The vertical dashed lines divide the diagram into zones indicating how
many density dips are present in the solution, which is either (0), (i), (iii), (v), or (vii). See also the
labels at the top of the panel indicating the zone. [Figure credit: Used with permission from [17],
Copyright (2014) by the American Physical Society]
It is important to note that in this case, the solution is amenable to the variation
of a number of experimental “knobs”. One of them is the frequency. One can drive
the boundaries at different frequencies near the edge of the linear spectral band.
This was also done in [17] and the relevant results are shown in Fig. 4.3c. There
we can see one of the most prototypical examples of what is known as a snaking
bifurcation diagram; see, e.g., relevant theoretical examples in [18]. We can see that
the computational analysis of the model with the driving at the boundary reveals
a series of solution branches. The stable ones among them are denoted by solid
(blue) lines and the unstable ones by dashed (red) lines. As the frequency is varied
the experiment appears to track the stable branches; as these become dynamically
(a) (b)
(c)
(d)
Fig. 4.4 Panels (a)–(c) represent waveforms obtained from the different branches of solutions
of the bifurcation diagram of Fig. 4.3, with three, five, and seven density dips, respectively. See
e.g. zones (iii), (v) and (vii) of Fig. 4.3. It can be clearly observed that in all three cases, the
numerically exact (up to prescribed tolerance), the experimental, and the numerical simulation
emulating the experimental procedure results are in close correspondence. Panel d illustrates the
potential variation, in addition to the frequency parameter, of the amplitude of the external boundary
drive for a fixed drive frequency f b = 7.14 kHz. In this case, we can observe that the solution presents
multiple stable branches for the same amplitude drive a (bistability), thus creating the possibility
of hysteretic response of the system as the amplitude is varied. As usual the stable portions of the
solution branches are shown by solid (blue) lines and the unstable ones by red (dashed) lines. The
diagnostic shown here is the maximum velocity of the 10th bead (v10 ). [Figure credit: Used with
permission from [17], Copyright (2014) by the American Physical Society]
unstable, the experiment appears to “jump” to a subsequent branch and then to a

further one and so on.
A natural question then is what type of waveform is represented by these additional
branches. The answer to this question is given by Fig. 4.4. Intriguingly, the relevant
solutions can be thought of as a 3-dark-breather, a 5-dark-breather, and a 7-dark-
breather waveform (even ones are excluded by parity considerations). Another way
to look at the relevant profiles is as realizations of periodic or elliptic function-type
solutions in this setting (with different periods). In any case, they constitute canonical
4.2 Connection with Experimental Results 49
generalizations of the single dark breather, producing in all cases very reasonable
agreement between the numerically exact (up to prescribed accuracy) waveform, the
direct numerical simulation, and the experimentally obtained result. It is relevant to
point out, finally, that in addition to varying the frequency of the drive at the domain
boundaries, another dimension of variation consists of varying the amplitude of the
drive. In this case, results like those of Fig. 4.4d are obtained, where it can be seen
that multi-stability can arise between solutions with, say, a single and a triple dark
breather. This paves the way for the observation of hysteretic phenomena that were
indeed observed in [17].
References
1. C. Chong, P.G. Kevrekidis, G. Theocharis, C. Daraio, Dark breathers in granular crystals. Phys.
Rev. E 87, 042202 (2013)
1 (2008)
linear systems. Physica D 216, 1 (2006)
4. R. Dodd, J. Eilbeck, J. Gibbon, H. Morris, Solitons and Nonlinear Wave Equations (Academic
Press, San Diego, 1982)
5. J. Cuevas, P. Kevrekidis, F. Williams, The sine-Gordon Model and its Applications: From
Pendula and Josephson Junctions to Gravity and High Energy Physics (Springer, Heidelberg,
2014)
6. A.J. Sievers, S. Takeno, Intrinsic localized modes in anharmonic crystals. Phys. Rev. Lett. 61,
970–973 (1988)
7. J.B. Page, Asymptotic solutions for localized vibrational modes in strongly anharmonic peri-
odic systems. Phys. Rev. B 41, 7835–7838 (1990)
8. R.S. MacKay, S. Aubry, Proof of existence of breathers for time-reversible or Hamiltonian
networks of weakly coupled oscillators. Nonlinearity 7, 1623 (1994)
9. N. Boechler, G. Theocharis, S. Job, P.G. Kevrekidis, M.A. Porter, C. Daraio, Discrete breathers
in one-dimensional diatomic granular crystals. Phys. Rev. Lett. 104, 244302 (2010)
10. G. Theocharis, N. Boechler, P.G. Kevrekidis, S. Job, M.A. Porter, C. Daraio, Intrinsic energy
localization through discrete gap breathers in one-dimensional diatomic granular crystals. Phys.
Rev. E 82, 056604 (2010)
11. Y.S. Kivshar, G. Agrawal, Optical Solitons: From Fibers to Photonic Crystals (Academic Press,
San Diego, 2003)
12. M. Remoissenet, Waves Called Solitons (Springer, Berlin, 1999)
13. G. James, P.G. Kevrekidis, J. Cuevas, Breathers in oscillator chains with Hertzian interactions.
Physica D 251, 39 (2013)
14. P.G. Kevrekidis, D.J. Frantzeskakis, R. Carretero-González, The Defocusing Nonlinear
Schrödinger Equation (SIAM, Philadelphia, 2015)
15. Y.S. Kivshar, D.K. Campbell, Peierls-Nabarro potential barrier for highly localized nonlinear
modes. Phys. Rev. E 48, 3077–3081 (1993)
16. G. James, Centre manifold reduction for quasilinear discrete systems. J. Nonlin. Sci. 13, 27
(2003)
17. C. Chong, F. Li, J. Yang, M.O. Williams, I.G. Kevrekidis, P.G. Kevrekidis, C. Daraio, Damped-
driven granular chains: An ideal playground for dark breathers and multibreathers. Phys. Rev.
E 89, 032924 (2014)
18. M. Beck, J. Knobloch, D.J.B. Lloyd, B. Sandstede, T. Wagenknecht, Snakes, ladders, and isolas
of localized patterns. SIAM J. Math. Anal. 41, 936–972 (2009)
Chapter 5
Heterogeneous Media
5.1 Overview
Of the three primary structures of interest in this book, solitary waves and breathers
have been studied in heterogeneous granular chains, while dispersive shock waves
have not. The latter represents an area of open and active research and the former
is the focus of this chapter. While we will focus on traveling solitary waves and
breathers (each with its own section) in this chapter, we briefly mention other work
that has been done in heterogeneous chains. For example, traveling waves in granular
chains with impurities can lead to transmitted and reflected waves [1]. In chains
with multiple impurities and precompression, resonant interactions can result in
perfect transmission of waves with small amplitude through the impurities [2]. This
is similar to the Ramsauer–Townsend resonance in quantum physics. Chains with
special heterogeneities can be used as impact mitigators [3]. Tapered chains have also
been thoroughly studied [4–7], which are chains where some feature, such as the mass
of the particles, changes gradually along the chain. One possible analytical approach
to study such heterogeneous chains is the binary collision approximation [8]. Here,
one studies the dynamics between two adjacent nodes for certain intervals of time.
Disorder in granular chains has also been studied, where the tuning of the balance of
disorder to nonlinearity strength can lead to subdiffusive or superdiffusive transport
of energy through the lattice [9, 10] (see also, e.g., [11, 12]).
5.2 Traveling Solitary Waves in Heterogeneous Chains
One of the early studies at the level of heterogeneous chains concerned the evolution
of the solitary waves that we explored in Chap. 3, but in the context of dimers, rather
than monomers. A typical example is shown in Fig. 5.1. The early experiments (and
supporting numerical simulations, as well as continuum analysis) of [13] and [14]

52 5 Heterogeneous Media
Fig. 5.1 Force versus time in chains of dimers with one (in (a)–(b)) or two (in (c)–(d)) stainless steel
beads alternating with one teflon bead. The top panels show experimental results, while the bottom
ones show the corresponding numerical data. The arrows with numbers point to the corresponding
particles in the chain. The first and third curve correspond to a steel bead, while the middle one to
a teflon bead. [Figure credit: Panel (a, b) used with permission from [13], Copyright (2008) by
the American Physical Society. Panel (c, d) used with permission from [14], Copyright (2009) by
Elsevier]
seemed to show the propagation of a solitary wave, albeit with modified character-
istics (such as a width dependent on the nature of the heterogeneity, e.g., the ratio
of the masses). While it is true that some decay of the principal traveling wave was
observed in these experiments, it was assumed to be due to the inherent dissipation
of the experimental setup rather than some other factor (e.g., the heterogeneity). It
is also worthwhile to mention that these experiments were of the “early type” in that
they involved measurements at individual sensors (in particular beads) rather than
the opportunity to monitor the entire space-time configuration as is possible, e.g.,
with a laser Doppler vibrometer.
However, these studies were subsequently superseded by the work of [15]. In this
work, a monoparametric study was conducted (since this heterogeneous problem
has as its key parameter the ratio of the masses ρ = M1 /M0 ) to explore the effects
of heterogeneity in the propagation of traveling waves. In particular, the authors
measured the fraction of force transmitted over the chain for different values of ρ
(and same input force). They normalized the result to the case of the monomer (i.e.,
the case of ρ = 1), see Fig. 5.2. There, the curious feature was identified that if the
ratio of the masses dropped from 1 to 0.59, not only would the initial energy not be
transmitted to the right part of the chain in the form of a solitary wave, but in fact less
than 30% of the energy in comparison to the monomer case would be transferred to
the right. The authors of [15] termed this phenomenon resonance. Perhaps even more
5.2 Traveling Solitary Waves in Heterogeneous Chains 53
Fig. 5.2 Normalized 1
FORCE TRANSMITTED
transmitted force as function 0.9
of the normalized mass ratio 0.8
(NORMALIZED)
parameter ρ for a dimer that 0.7
is excited with an impulse. 0.6
(Note in the original
0.5
publication, the variable
0.4
was used in place of ρ.)
[Figure credit: Used with 0.3
permission from [15], 0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Copyright (2011) by the ρ
American Physical Society]
curiously, they identified also an “anti-resonance”, a case example for even smaller
ρ where nearly the entire energy is transmitted. They showed that there is, indeed,
a cascade of resonances and anti-resonances for different ρ values [16, 17]. On this
basis, they developed a multiple-scale perturbation theoretic treatment, where they
examined separately the fast oscillatory dynamics of the small masses, as well as the
slow dynamics of the large ones. The resulting WKB analysis of the rapid motion
of the small masses led to an identification of these critical points. The interested
reader is referred to [15, 18] for details. The attenuation of waves in the dimer chain
near resonant mass ratios was verified experimentally in [19].
Now we will demonstrate the fundamental qualitative features of the (anti-) reso-
nance phenomena in dimer chains by utilizing the more recent experimental obser-
vations of [20]. In particular, in Fig. 5.3 we can observe the experimental (top) and
theoretical (bottom) demonstration of the resonance at ρ = 0.59. Panels (a) and (d)
correspond to the weakly vibrating –after the wave has passed– large beads, while
the strongly vibrating –after the wave has passed– small beads are shown in panels
(b) and (e). The intrinsic heterogeneity of the beads does not permit all the energy
to be transferred from large to small bead and results in the residual radiation shown
in Fig. 5.3. This is ultimately the reason for the decay of the energy of the principal
solitary pulse structure (and hence the decay is not due solely to dissipation). Panel
(c) shows a very characteristic experimental observation of the propagation, where
not only the dispersive radiation can be discerned, but also it can be seen that the
wave destructively interferes with its own tail and therefore eventually completely
disappears.
On the other hand, in Figs. 5.4 and 5.5, we examine the examples of the anti-
resonances for ρ = 0.3428 (the first one after the monomer case of ρ = 1), as well
as the next one of ρ = 0.1548. Here, too, why these particular ρ values are so special
becomes rather transparent. The large beads of the left panels always perform a
single oscillation as the traveling wave passes through. However, the smaller beads,
ringing back and forth between the bigger ones possess special properties at these
particular values of ρ. Namely, for ρ = 0.3428, the smaller beads execute precisely
two oscillations such that at the end of the second one, they transmit all of their
Fig. 5.3 Velocity versus time response for the resonant mass ratio case of ρ = 0.59 upon impulse
excitation. a Experimentally measured vibration of a number of large masses and b small masses. c
Space-time view of the evolution of the solitary wave, as it is inferred from collating the experimental
results for each bead obtained by the laser Doppler vibrometer. Panels d and e are the numerical
simulations corresponding to panels a and b, respectively. [Figure credit: Experimental results and
figure obtained by the group of Prof. J. Yang, University of Washington cf. Ref. [20]. Used with
permission]
Fig. 5.4 Same as the previous figure, but now for the anti-resonance occurring at ρ = 0.3428. Here
it can be clearly seen that the big beads in the left panels execute a single vibration, while the smaller
beads of the middle panels perform two. The nearly perfect transfer of the energy from the small
to the next big bead leads to nearly undistorted propagation of a traveling wave for such isolated
values of ρ. [Figure credit: Experimental results and figure obtained by the group of Prof. J. Yang,
University of Washington cf. Ref. [20]. Used with permission]
5.2 Traveling Solitary Waves in Heterogeneous Chains 55
Fig. 5.5 Similar case example as in the previous figure, but now for the next anti-resonance, occur-
ring at ρ = 0.1548. Here, the small beads execute three vibrations. [Figure credit: Experimental
results and figure obtained by the group of Prof. J. Yang, University of Washington cf. Ref. [20].
Used with permission]
energy (exactly in the numerical computation of the bottom panels, approximately

in the experiment of the top panels) to the next large bead. Normally (i.e., for generic
values of ρ), this does not happen. There is, instead, a residual “ringing” of the small
beads that draws energy from the primary propagating pulse gradually degrading its
energy content. However, at these non-generic values of ρ, the anti-resonances, the
phase of the smaller bead oscillation is perfectly tuned so as for it to transmit all of
its vibrating energy to the next large one, enabling the wave to propagate undistorted
over the lattice. In the case of ρ = 0.1548, the small beads execute precisely three
vibrations and then transfer their energy to the next larger neighbor; for the next
anti-resonance, this number will increase to four and so on. It is thus clear that anti-
resonances are a special feature that is identified at unique, isolated values of the ratio
of the masses in the dimer chain. Unlike solitary waves in a homogeneous chain, the
solitary wave tails in a dimer feature oscillations of small amplitude that do not decay
to zero [14, 15].
It is an open problem if there exist genuine solitary waves with tails that decay
exponentially to zero in dimer granular chains. Studies on solitary waves were also
conducted on the related problems of the dimer Toda lattice [21], and the dimer
FPUT problem with a polynomial potential [22].
5.3 From Isolated Defect Modes to Bright Breathers

in Dimer Chains
We now turn to the examination of breather modes in heterogeneous chains. As dis-

cussed in the previous section, for traveling waves there is a remarkable modification
of the behavior in such lattices, given that only for isolated values of the heterogeneity
parameter can such solutions be identified. Our aim is to explore whether similarly
novel features arise in heterogeneous chains for discrete breathers. In Chap. 4 we
realized that breathers do not exist in purely nonlinear granular chains, and thus, in
what follows in this section, it is assumed the chain is precompressed (and hence
linearizable). In homogeneous chains with precompression, as shown in Chap. 4, we
utilized a multiple-scale expansion to reach a defocusing NLS equation, which, in
turn, enabled us to formulate dark breather solutions.
A fundamental result that we will see in heterogeneous chains is that bright
breathers, i.e., breathers on a vanishing background, are possible. To illustrate this,
we will start from discussing so-called impurity modes. These are modes induced
by defects in an otherwise homogeneous chain. Such modes were explored system-
atically (including using experimental observations) in the work of [23]. In the case
of a single defect, the equations of motion are identical to the monomer chain, with
the exception of the lattice node n = k, which is where the defect is located and its
immediate vicinity. The equations of motion for the defect node k and its two nearest
neighbors are,
3/2 3/2
M ü k−1 = A1 [δ0 + u k−2 − u k−1 ]+ − A2 [δ1 + u k−1 − u k ]+ ,
3/2 3/2
m ü k = A2 [δ1 + u k−1 − u k ]+ − A2 [δ1 + u k − u k+1 ]+ ,
3/2 3/2
M ü k+1 = A2 [δ1 + u k − u k+1 ]+ − A1 [δ0 + u k+1 − u k+2 ]+ ,
2E ( R2 ) 2E ( R+r )
1/2 Rr 1/2
A1 = 3(1−ν 2 )
, A2 = 3(1−ν 2 )
. (5.1)
where m is the mass of the defect particle, which has radius r , and M is the mass
of the particles in the bulk (i.e., the non-defect particles), which have radius R. It
is relevant to point out that the equations (5.1) do not only contain a mass defect
(as is typically considered in the context of lattices, see e.g. the classical work [24]
and also more recent work [25]). They additionally contain an elastic factor defect,
as the smaller mass interacts differently with its nearest neighbors (coefficient A2 )
than two large masses interact between themselves (coefficient A1 ). Furthermore, a
key feature to appreciate is that the presence of a lighter mass in a chain of heavier
masses will introduce an isolated mode above the edge of the acoustic band. In this
3/2
case, and for precompression F0 = A1 δ0 , the frequency of this impurity mode was
found in [23] to be approximated by:
√ 1/3 1/6
3 A2 F0
fa ≈ . (5.2)
2π m 1/2
5.3 From Isolated Defect Modes to Bright Breathers in Dimer Chains 57
On the other hand, if the mass m > M, instead of the mode yielding an isolated
defect frequency above the band, it will lead to a so-called “embedded mode” inside
of the linear spectrum. This analysis in the work of [23] was a motivation to explain
that even in the case where F0 = 0 (i.e., the purely nonlinear no-precompression
case), there is localization, although in that case it is transient when a wave passes
through the defect (creating a transient, effective precompression) that in turn may
lead to an effective breathing.
The work of [26], on the other hand, had as its starting point the case with pre-
compression and the presence of a lighter mass defect which creates the isolated
defect mode frequency. Then, using techniques of periodic orbit identification (such
as a Newton method in the Poincaré section, as discussed in the Appendix A.2), it is
possible to identify a monoparametric family of bright breathers whose frequencies
extend from the linear mode frequency to that of the upper edge of the acoustic spec-
trum. The result of the continuation of such a mode is shown in Fig. 5.6a, where the
frequency dependence on the maximal amplitude of displacement in the central bead
of the chain was computed. The periodic breathing of the energy is not only evident
in the exact solution evolution in panel (b), but also in the localization persisting –in
the case of localized initial data– for different amplitudes (by an order of magnitude)
in panels (c) and (d). It is clear that while some dispersive wave radiation is emitted
especially at the early stages of the simulation, the solution eventually settles into a
vibrating waveform which has a near perfect periodicity, representing an essentially
exact breather.
Our aim is to gradually build a more heterogeneous setting, such as a granular
dimer as an extension of the above defect setup. The natural next step in this program
is to examine the case of two defects. In the linear variant of this problem (i.e., either
linearizing and keeping the leading term, or assuming the nonlinearity exponent to
be unity), Fig. 5.7a shows that there are two modes for different distances between
the defects. These are the in- and out-of-phase vibration modes of the two defects.
Naturally as the defects are separated more and more, these two modes essentially
become isolated and their frequencies become equal. Now, we separately consider
the nonlinear analogue of the case where the defects have a distance of two sites.
Then each of the two branches, both the symmetric and the anti-symmetric ones, have
a nonlinear continuation. However, for the symmetric branch, the nonlinear periodic
orbit that results from it does not feature any bifurcation. On the contrary, for the anti-
symmetric branch its continuation becomes subject to a pitchfork (i.e., symmetry-
breaking) bifurcation which is showcased in Fig. 5.7b. Here, the difference between
the force acting on the two defect sites naturally illustrates the pitchfork character
of the bifurcation since for the “parent” (anti-symmetric) branch this difference
vanishes, while for the two asymmetric waveforms that arise from it, it is different
than zero (and assumes equal/opposite values).
The Floquet multipliers corresponding to the solutions of Fig. 5.7b are shown in
Fig. 5.7c, which illustrates the substantial instability growth rate of the real pair of the
destabilized anti-symmetric branch. On the other hand, the asymmetric branches that
emerge only feature weak oscillatory instabilities (i.e.,complex Floquet multipliers
with modulus greater than unity) with relatively small growth rates. The dynamical
(a) (b)
4 −6
x 10
x 10 1 1
2.5
−6
Im(λ) 0 x 10
n
0
u
−6
x 10
2.4 2 1
−1
n
u
−1 0 1 −11 40 79
Re(λ) 0
n
2.3 0
u (m)
−6
x 10
b
1 2
f
−2
n
1 40 79
n
−1
Im(λ)
2.2 0 1
0
n
u
Im(λ)
−2
0
−1 −2
2.1 −1 0 1 −41 40 79 50 5
Re(λ) n −1
−1 0 1 40 −5
0 0.1 0.2 0.3 0.4 0.5 0.6 Re(λ) x 10
30 0 t (sec)
max(u40)/δ1
(c) −7
(d) −6
x 10 x 10
220 −7 200 −6
x 10 x 10
2 2
1 1
u (m)
u (m)
230 0 0
1.5 220 1.5
k
k
−1 −1
n site index
n site index
240 1.8 1.9 2 2.1 1 1.9 2 2.1 1

t (sec) x 10
−3 240 t(sec) x 10
−3
250 0.5 0.5

240 0 260 240 0
260
250
n
n
250 −0.5 −0.5
280
270 260
1.7 1.8 1.9 2 2.1 −1 1.6 1.8 2 −1
t(sec) x 10
−3 t(sec) x 10
−3
280 300
0 0.5 1 1.5 2 0 0.5 1 1.5 2
−3 −3
t(sec) x 10 t(sec) x 10
Fig. 5.6 a Bifurcation diagram of the branch of defect breathers localized at the light mass defect site
of a granular crystal (bead 40 in the middle of the chain). Breather frequency versus the normalized
displacement at node 40 (max(u 40 /δ1 )) is shown. Here δ1 is the equilibrium displacement between
a bulk particle and a defect particle. The insets show at different points in the diagram the form of the
spatial profile of the defect breather and its Floquet multipliers, ensuring the stability of the pertinent
branch. b The breathing evolution of a member of this branch. c–d Two case examples where energy
was input at the middle of the chain and was let to evolve. The result was a localization of the energy
at the central (defect) site such that after the dispersion manifest at the early dynamical stages a
very clear periodic pattern (see the insets) was distilled. [Figure credit: Used with permission from
[26], Copyright (2009) by the American Physical Society]
evolution of the instability is manifested in Fig. 5.7d. There, we can see, via the
space-time evolution of the displacements, how the energy is partitioned between
the two defects. In particular, the energy starts out being equally partitioned between
the two defect beads. Over time, however, this configuration turns out to be unstable
leading some of the energy to be redistributed from one defect to the other. This
is strongly reminiscent of dynamics in double-well potentials e.g. in atomic and
optical physics [27]. Due to the Hamiltonian nature of the evolution system at hand,
recurrence events subsequently arise as a result of which the energy then returns to a
roughly equal partition between the wells, then it asymmetrizes again and the cycle
restarts. This kind of “tunneling” dynamics is a characteristic manifestation of the
symmetry-breaking phenomenology. It should be noted in the work of [26], variants
of the bifurcation in which the masses of the two defects were slightly asymmetric
were also considered. In this case, it is well-known from standard bifurcation theory
(a) (b)
(c) (d)
Fig. 5.7 a Eigenfrequencies of the linear problem (for the case of two defects) as a function of the
separation between the defects (impurities). The ratio of the smaller radius r to the larger radius R is
r/R = 0.8. Here, k and are the site indices of the defects, such that the distance between the defects
is − k. The relevant modes are the symmetric and anti-symmetric normal modes of vibration of the
two (defect) sites. The insets show example profiles. b Prototypical form of the pitchfork bifurcation
by measuring the difference in force between the defect sites. Hence, symmetric branches (A1 ) will
be associated with a zero value while asymmetric ones with a non-zero value (A2 and A3 ). c
Floquet multipliers corresponding to panel (b). When the bifurcation occurs, the symmetric branch
acquires a (maximal) Floquet multiplier which is real and thus becomes exponentially unstable. The
asymmetric branches only inherit oscillatory (weak) instabilities of the principal (anti-symmetric)
branch. d Evolution of the displacement of the granular chain in space-time illustrating the transition
(as a result of the instability manifestation) from a symmetric to an asymmetric distribution of the
energy in the chain
that the relevant pitchfork is no longer symmetric, but rather an “imperfect” pitchfork,
i.e., it breaks into an isolated (always stable) branch and a pair of branches involving
an unstable saddle and a stable center.
It is worthwhile to note that such examples of both a single light mass defect and
even of a double light mass defect have been explored at the linear level in the study
of [28]. As can be seen in Fig. 5.8a, the relevant mode associated with the light mass
defect can be identified in experiments in close correspondence (especially at the
qualitative level) with the theoretical prediction of Eq. (5.2) in [23]. In Fig. 5.8b this
type of mode is followed as a function of the ratio of the light to the heavy mass.
(a)
(b) (c)
Fig. 5.8 a The (experimental) linear spectrum without (red) and with (blue) a light mass defect.
Experimentally observed band edge is denoted by the upside down triangle and the observed defect
mode is denoted by the rhombus. Theoretical prediction of the band edge is denoted by dashed
vertical line and the prediction of the defect mode is denoted by a dashed-dot vertical line. b A
systematic following of the relevant mode as a function of the mass ratio m/M (red is theoretical,
while blue is experimental). The inset shows the displacement of the solution for the mass ratio
0.20. c The general case of two defects (again red is without and blue is with defects). Here, we can
also see the two defect modes and their frequencies’ separation as a function of the distance between
the defects. [Figure credit: Used with permission from [28], Copyright (2012) by the American
Physical Society]
Finally, in Fig. 5.8c the case of two defects is touched upon showing how the relevant
–in this case two– defect modes can be observed in experimental settings.
In the last phase of this program which aims to explore the genuine heterogeneous
chain, namely the case of the dimer, we now perform what we refer to as the “1, 2, 3,
. . . , ∞” step. Namely, we now move away from the case of isolated defects (one or
two), and think of the case where every other site is a defect. In this way, we recover
order in the form of a heterogeneous dimer chain! Now, every contact point will be
one of a lighter mass particle with a heavier mass particle, and thus An = A and
δ0,n = δ0 . We now write m n = M0 if n is even and m n = M1 if n is odd. In the dimer
chain there is one more parameter than in the monomer chain, which is the mass
ratio ρ = M1 /M0 . If we introduce new variables wn to represent the displacement
of the even particles (e.g. wn = u 2k ) and vn to represent the displacement of the odd
particles (e.g. vn = u 2k+1 ) then we can rewrite Eq. (1.3) as
3/2 3/2
ρv̈n = [1 + wn−1 − vn ]+ − [1 + vn − wn ]+ ,
3/2 3/2 (5.3)
ẅn = [1 + vn − wn ]+ − [1 + wn − vn+1 ]+
where we have rescaled time and amplitude (the latter rescaling is by dividing by the
precompression displacement). We assume that M0 > M1 , such that the mass ratio
ρ < 1. One may inquire then about what this does to the linear spectrum. Having
one (light mass) defect, as we saw, adds one mode that lies above the acoustic band
of the spectrum. Two such defects lead to two isolated modes above the acoustic
band, three defects to three modes, and so on and so forth. As more defects lead to
more defect modes, eventually the restoration of order leads to the formation of an
additional band, an optical band as it is referred to, with frequencies higher than the
acoustic ones. It is then relevant to use linear theory, which is possible under the
assumption of small strain,
|wn−1 − vn | 1, |wn − vn | 1
in which case the equations of motion reduce to:
ρv̈n = K 2 (wn − vn ) − K 2 (vn − wn−1 )

(5.4)
ẅn = K 2 (vn+1 − wn ) − K 2 (wn − vn )
1/2
where K 2 = 23 Aδ0 . These linear equations are solved by

vn a i(kn−ωt)
= e .
wn b
Substituting this into Eq. (5.4), one obtains an algebraic system for (a, b). The con-
dition that the determinant should vanish leads to the well-known, e.g., in solid state
physics [29], dimer frequency spectrum:
2
K2 K2 K2 k
ω(k)2± = K 2 + ± K2 + − 4 2 sin2 (5.5)
ρ ρ ρ 2
where the (−) sign corresponds to the lower frequencies, i.e., the acoustic band, while
the (+) sign to the higher frequencies, i.e., the optical band. The left plot of Fig. 5.9
shows this dispersion relation for the parameter values taken from [30], while the
right panel is the corresponding experimental identification of these branches. This
analysis is drawn from the work of [30] and the supporting more detailed exposition
of [31].
(a) (b)
Fig. 5.9 Dispersion in the dimer granular chain considered in [30]. a Analytically derived dispersion
curves. b Transfer function obtained from the experiment, which is computed from frequencies
detected at one end of the chain after white noise is sent through the opposite end. (Note that the
experimental and theoretical cutoff frequencies differ somewhat) [Figure credit: We adapted panel
b with permission from Ref. [30]. Copyrighted (2010) by the American Physical Society]
A key feature of the optical band is that it has the opposite concavity of the acoustic
band at the wavenumber k = π. Now, delving again into the multiple-scale expansion
of the same type as in the previous chapter, we realize that it is the concavity of the
band that determines the coefficient of the second derivative (Laplacian) term, while
the nonlinearity preserves its definite prefactor sign. As a result, at the top of the
acoustic band there will still be a defocusing NLS approximation (since the sign of
the linear and nonlinear terms will be opposite) and a dark breather entity that could
be generated. This will no longer be true at the bottom of the optical band. There,
the concavity changes and we will thus obtain a focusing NLS equation (since now
the sign of the linear and nonlinear terms will be identical). In particular, to derive
the NLS equation from the dimer granular crystal, we use the following ansatz [32]

vn (t) = Uw (X, T ) + (Uv (X, T )E(n, t; 0, ω+ (π)) + c.c.) (5.6)
wn (t) = Uw (X, T ) (5.7)
where
X = n, T = 2 t, E(n, t; k0 , ω0 ) = ei(k0 n+ω0 t)
Note that we have already selected the plane wave at the bottom of the optical band
to be modulated by the envelope function Uv , since the notation is less cumbersome
than in the general wavenumber case. Substitution of this ansatz into Eq. (5.3) leads
us to obtain the focusing NLS equation at order E3
i∂T Uv (X, T ) + ν2 ∂ X2 Uv (X, T ) + ν3 Uv (X, T )|Uv (X, T )|2 = 0, (5.8)

ω (π) K 22 ω+ (π)
ν2 = − +2 , ν3 = 2
(3K 2 K 4 − 4K 32 )

Note that, since 3K 2 K 4 − 4K 32 < 0 and −ω+ (π)/2 < 0 both ν2 and ν3 are nega-
tive, such that Eq. (5.8) is focusing. The function Uw (X, T ) is defined explicitly in
terms of Uv (X, T ) [32]. A detailed explanation of the derivation of the NLS equa-
tion from a monomer chain is given in Appendix A.3. The derivation in the case of
a dimer chain can be found in [32]. It is well-known that a focusing NLS equation
has plane wave solutions which, however, are unstable due to the so-called modu-
lational instability (MI) [33]. In particular, the plane wave Uv (X, T ) = aei(k X +ωT )
is a solution of the NLS equation (5.8) if ω = k 2 − ν3 a 2 , where we have chosen
parameters such that ν2 = −1. The stability of this plane wave solution can be stud-
ied by perturbing it, Uv (X, T ) = (a + 1 (X, T ))ei(k X +ωT )+i2 (X,T ) , where 1 (X, T )
and 2 (X, T ) are perturbations to the amplitude and phase. Assuming 1 (X, T ) and
2 (X, T ) are small leads to a linear differential equation that has the dispersion
relation (−Ω + 2k Q)2 ∼ Q 2 (Q 2 − 2ν3 a 2 ) where Ω and Q are the frequency and
wavenumber of the perturbed solution, respectively. This implies that for sufficiently
small perturbation wavenumbers Q, the solutions will become unstable (i.e., the per-
turbation growth rate will become complex) and will manifest a dynamical instability.
It can be shown that this instability will result in the formation of (bright) solitary
waves at the level of the NLS equation [33]. Since the NLS equation describes the
modulation of plane waves (see ansatz Eq. (5.6)), the MI will result in the formation
of bright discrete breathers at the level of the original granular crystal. These are
solutions that are breathing over time, but which are now existing against a vanish-
ing background (contrary, e.g., to what was the case for the dark breathers of the
previous chapter).
An example of the manifestation of the MI is shown in Fig. 5.10. Here, a plane
wave initial condition (notice how the energy is essentially equidistributed initially)
becomes unstable and spontaneously chooses to focus the energy towards the for-
mation of highly localized waveforms, namely the bright discrete breather periodic
orbits; it is relevant to point out the clearly periodic in time, highly localized in
space form of the structure in the insets. Motivated by this dynamical evolution, it is
possible to now seek periodic orbits that, due to the modulational instability of the
bottom of the optical band of the dimer spectrum (given the focusing NLS nature
of the multiple-scale expansion), bifurcate as bright breathers from the bottom edge
of the optical band. As these waveforms bifurcate, they resemble the extended (out-
of-phase) linear mode from which they emerge, see Fig. 5.11a–b. As the frequency
of the periodic orbits is decreased and one moves deeper inside the spectral gap, the
bright breather modes become progressively more localized, see Fig. 5.11c–d. It is
interesting to note that similarly to what happens typically in discrete systems bear-
ing breather-like excitations, there exist two types of such waveforms. One of these
Fig. 5.10 Near the bottom of the optical band, plane waves are subject to modulational instability.
The space-time evolution of a normalized displacement is shown while the insets show that one
relaxes towards a localized (bottom), time-periodic (top) state. [Figure credit: Used with permission
from Ref. [30]. Copyrighted (2010) by the American Physical Society]
(a) (b)
(c) (d)
Fig. 5.11 a Profile of a breather with a frequency near the bottom of the optical band. The FMs
are shown in the inset. b Corresponding space-time evolution. c–d are the same as a–b, but for a
frequency deeper in the band gap
is a (principally) onsite excitation, where the energy is chiefly localized on a central

site. The other is a principally intersite excitation with the energy partitioned chiefly
between two adjacent sites. The intersite branch always possesses a pair of real Flo-
quet multipliers (see Appendix A.2 for details on stability computations of discrete
breathers) and the two branches nearly coincide as the linear limit is approached,
i.e., they bifurcate from the same linear mode.
Period-3 (also called “trimer”) chains and period-4 (also called “tetramer”) chains
have been studied as well [34]. Such chains with larger spatial period involve
sequences where all particles are distinct or include multiple instances of a particular
type of particle (such as a trimer consisting of 1 steel and 2 aluminum particles). The
dispersion curves in trimer and tetramer chains will have two and three band gaps,
respectively. Breathers in chains with higher order spatial periodicity that bifurcate
from the higher bands are observed to be more robust (in terms of their stability)
when compared to breathers that bifurcate from the lower bands. This surprising
feature might be due to the fact that the former avoid resonances with linear modes
of higher frequencies, although a systematic investigation of their stability remains
an open problem.
Finally, breathers in heterogeneous granular chain models that also account for
driving and damping have been studied. For example, upon variation of the drive
amplitude, hysteretic dynamics connecting limit cycles and quasiperiodic or chaotic
dynamics have been explored in damped in driven chains with a defect [35] and in
periodic chains [36, 37]. Upon variation of the driving frequency, a nonlinear bending
of resonant peaks is possible [37–39]. Breathers in damped-driven systems without
precompression have also been explored [40, 41]. For a more detailed summary of
breathers in damped-driven granular crystals, see [42].
References
462, 21 (2008)
2. A.J. Martínez, H. Yasuda, E. Kim, P.G. Kevrekidis, M.A. Porter, J. Yang, Scattering of waves
by impurities in precompressed granular chains. Phys. Rev. E 93, 052224 (2016)
3. F. Fraternali, M.A. Porter, C. Daraio, Optimal design of composite granular protectors. Mech.
Adv. Mat. Struct. 17(1), 1 (2010)
4. R. Doney, S. Sen, Decorated, tapered, and highly nonlinear granular chain. Phys. Rev. Lett. 97,
155502 (2006)
5. R.L. Doney, J.H. Agui, S. Sen, Energy partitioning and impulse dispersion in the decorated,
tapered, strongly nonlinear granular alignment: a system with many potential applications. J.
Appl. Phys. 106, 064905 (2009)
6. U. Harbola, A. Rosas, A.H. Romero, M. Esposito, K. Lindenberg, Pulse propagation in deco-
rated granular chains: an analytical approach. Phys. Rev. E 80, 051302 (2009)
7. U. Harbola, A. Rosas, M. Esposito, K. Lindenberg, Pulse propagation in tapered granular
chains: an analytic study. Phys. Rev. E 80, 031303 (2009)
8. K. Lindenberg, U. Harbola, A.H. Romero, A. Rosas, Pulse propagation in granular chains,
in International Conference on Applications in Nonlinear Dynamics — ICAND 2010 (AIP
Conference Proceedings 1339) (American Institute of Physics, 2011), p. 97
9. A.J. Martínez, P.G. Kevrekidis, M.A. Porter, Superdiffusive transport and energy localization
in disordered granular crystals. Phys. Rev. E 93, 022902 (2016)
10. V. Achilleos, G. Theocharis, C. Skokos, Energy transport in one-dimensional disordered gran-
ular solids. Phys. Rev. E 93, 022903 (2016)
11. A. Sokolow, S. Sen, Exact solution to the problem of nonlinear pulse propagation through
random layered media and its connection with number triangles. Ann. Phys. 322, 2104 (2007)
12. L. Ponson, N. Boechler, Y.M. Lai, M.A. Porter, P.G. Kevrekidis, C. Daraio, Nonlinear waves
in disordered diatomic granular chains. Phys. Rev. E 82, 021301 (2010)
13. M.A. Porter, C. Daraio, E.B. Herbold, I. Szelengowicz, P.G. Kevrekidis, Highly nonlinear
solitary waves in periodic dimer granular chains. Phys. Rev. E 77, 015601(R) (2008)
14. M.A. Porter, C. Daraio, I. Szelengowicz, E.B. Herbold, P.G. Kevrekidis, Highly nonlinear
solitary waves in heterogeneous periodic granular media. Physica D 238, 666 (2009)
15. K.R. Jayaprakash, Y. Starosvetsky, A.F. Vakakis, New family of solitary waves in granular
dimer chains with no precompression. Phys. Rev. E 83, 036606 (2011)
16. K.R. Jayaprakash, A.F. Vakakis, Y. Starosvetsky, Solitary waves in a general class of granular
dimer chains. J. App. Phys. 112, 034903 (2012)
17. K.R. Jayaprakash, A. Shiffer, Y. Starosvetsky, Traveling waves in trimer granular lattice I:
Bifurcation structure of traveling waves in the unit-cell model. Commun. Nonlinear Sci. Numer.
Simul. 38, 8–22 (2016)
18. K.R. Jayaprakash, Y. Starosvetsky, A.F. Vakakis, O.V. Gendelman, Nonlinear resonances lead-
ing to strong pulse attenuation in granular dimer chains. J. Nonlinear Sci. 23, 363 (2013)
19. S.Y. Wang, V.F. Nesterenko, Attenuation of short strongly nonlinear stress pulses in dissipative
granular chains. Phys. Rev. E 91, 062211 (2015)
20. E. Kim, R. Chaunsali, H. Xu, J. Castillo, J. Yang, P.G. Kevrekidis, A.F. Vakakis, Nonlinear
low-to-high frequency energy cascades in diatomic granular crystals. Phys. Rev. E 92, 062201
(2015)
21. A. Vainchtein, Y. Starosvetsky, J.D. Wright, R. Perline, Solitary waves in diatomic chains.
Phys. Rev. E 93, 042210 (2016)
22. T.E. Faver, J.D. Wright, Exact diatomic Fermi–Pasta–Ulam–Tsingou solitary waves with opti-
cal band ripples at infinity (2015), arXiv:1511.00942
23. S. Job, F. Santibanez, F. Tapia, F. Melo, Wave localization in strongly nonlinear Hertzian chains
with mass defect. Phys. Rev. E 80, 025602 (2009)
24. R.F. Wallis, Effect of free ends on the vibration frequencies of one-dimensional lattices. Phys.
Rev. 105, 540 (1957)
25. M. Makwana, R.V. Craster, Localised point defect states in asymptotic models of discrete
lattices. Q. J. Mech. Appl. Math. 66, 289 (2013)
26. G. Theocharis, M. Kavousanakis, P.G. Kevrekidis, C. Daraio, M.A. Porter, I.G. Kevrekidis,
Localized breathing modes in granular crystals with defects. Phys. Rev. E 80, 066601 (2009)
27. T. Zibold, E. Nicklas, C. Gross, M.K. Oberthaler, Classical bifurcation at the transition from
Rabi to Josephson dynamics. Phys. Rev. Lett. 105, 204101 (2010)
28. Y. Man, N. Boechler, G. Theocharis, P.G. Kevrekidis, C. Daraio, Defect modes in one-
dimensional granular crystalsboe. Phys. Rev. E 85, 037601 (2012)
29. C. Kittel, Introduction to Solid State Physics (Wiley, Hoboken, 2005)
30. N. Boechler, G. Theocharis, S. Job, P.G. Kevrekidis, M.A. Porter, C. Daraio, Discrete breathers
in one-dimensional diatomic granular crystals. Phys. Rev. Lett. 104, 244302 (2010)
31. G. Theocharis, N. Boechler, P.G. Kevrekidis, S. Job, M.A. Porter, C. Daraio, Intrinsic energy
localization through discrete gap breathers in one-dimensional diatomic granular crystals. Phys.
Rev. E 82, 056604 (2010)
32. G. Huang, B. Hu, Asymmetric gap soliton modes in diatomic lattices with cubic and quartic
nonlinearity. Phys. Rev. B 57, 5746 (1998)
33. Y.S. Kivshar, G. Agrawal, Optical Solitons: From Fibers to Photonic Crystals (Academic Press,
San Diego, 2003)
34. C. Hoogeboom, P. Kevrekidis, Breathers in periodic granular chains with multiple band gaps.
Phys. Rev. E 86, 061305 (2012)
35. N. Boechler, G. Theocharis, C. Daraio, Bifurcation based acoustic switching and rectification.
Nat. Mater. 10, 665 (2011)
36. C. Hoogeboom, Y. Man, N. Boechler, G. Theocharis, P.G. Kevrekidis, I.G. Kevrekidis, C.
Daraio, Hysteresis loops and multi-stability: from periodic orbits to chaotic dynamics (and
back) in diatomic granular crystals. Eur. Phys. Lett. 101, 44003 (2013)
37. E.G. Charalampidis, F. Li, C. Chong, J. Yang, P.G. Kevrekidis, Time-periodic solutions of
driven-damped trimer granular crystals. Math. Probl. Eng. 2015, 830978 (2015)
References 67
38. J. Lydon, G. Theocharis, C. Daraio, Nonlinear resonances and energy transfer in finite granular
chains. Phys. Rev. E 91, 023208 (2015)
39. C. Chong, F. Li, J. Yang, M.O. Williams, I.G. Kevrekidis, P.G. Kevrekidis, C. Daraio, Damped-
driven granular chains: an ideal playground for dark breathers and multibreathers. Phys. Rev.
E 89, 032924 (2014)
40. D. Pozharskiy, Y. Zhang, M.O. Williams, D.M. McFarland, P.G. Kevrekidis, A.F. Vakakis, I.G.
Kevrekidis, Nonlinear resonances and antiresonances of a forced sonic vacuum. Phys. Rev. E
92, 063203 (2015)
41. Y. Zhang, D. Pozharskiy, D.M. McFarland, P.G. Kevrekidis, I.G. Kevrekidis, A.F. Vakakis,
Experimental study of nonlinear resonances and anti-resonances in a forced, ordered granular
chain. Exp. Mech. 57, 505 (2017)
Chapter 6
Media with Onsite Forces:
The Newton’s Cradle and Beyond
6.1 Breathers in the Newton’s Cradle
In this section, we finally return to the Newton’s cradle shown in Fig. 1.1 and Fig. 6.1a.
When the chain of balls is struck at one end, the ball on the opposite end is ejected from
the chain, providing a nice illustration of the principle of conservation of momen-
tum. In reality, not all of the energy is transferred to the ejected particle [1]. While
dissipation will account for some energy loss, it is also important to account for
forces between the particles due to their elastic deformation. In other words, the
Hertzian contact between the particles plays an important role in the dynamics of the
Newton’s cradle. As we have seen throughout this volume, the dynamics of a chain
of particles interacting through Hertzian contact is extremely rich and complex. We
already have a model accounting for the Hertzian contact in Eq. (1.3). To account for
the strings suspending the particles in the cradle, we simply recall the equations of
motion describing a pendulum, θ̈ = − g sin(θ) where θ is the deviation angle from its
rest position, g is the gravitational constant and is the length of the string. While this
description accounts for two-dimensional motion, if we assume the displacements
from the equilibrium angle are very small we can linearize the equation and obtain
a one-dimensional equation for the horizontal displacement u n . In the small angle
approximation we have u n = sin(θn ) ≈ θn . Thus, the equations of motion for the
dissipation-free Newton’s cradle are:
3/2 3/2
M ü n = A[u n−1 − u n ]+ − A[u n − u n+1 ]+ − κu n , (6.1)
where the gravitational spring constant is κ = Mg/, and we have assumed that all
particles are identical. Equation (6.1) also describes an array of clamped cantilever
beams decorated with spheres, see Fig. 6.1b. Dispersive Shock waves and traveling
waves have not been systematically studied in Eq. (6.1), although some hints about
the possibility of propagating waveforms were given in the dynamical simulations

70 6 Media with Onsite Forces: The Newton’s Cradle and Beyond
Fig. 6.1 a Prototypical Newton’s cradle consisting of a chain of beads attached to pendula. b Array
of clamped cantilevers decorated by spherical beads. The lower image shows a different perspective
of a single cantilever coupled to a sphere on each side. [Figure credit: Used with permission from
[2], Copyright (2013) by Elsevier.]
of [2]. In this section, we will thus focus on breather solutions (and traveling variants
thereof) of Eq. (6.1).
As discussed in Sect. 4.1, there are no breathers in the monomer granular chain
without precompression [2]. One missing ingredient is a linear spectrum from which
breathers can bifurcate from. In the Newton’s cradle model Eq. (6.1), the presence
of the onsite force κu n introduces a linear term. Linearization results √ in a system
of uncoupled linear oscillators that oscillate at the frequency ω0 = κ/M. While
this system does not possess a linear passband, breathers with frequency ω > ω0 can
bifurcate from this point [3].
One can study Eq. (6.1) in the small amplitude limit using a multiple-scale
analysis [3]. With the ansatz u n (t) = Un (T )ei T + c.c. one can derive the following
spatially discrete equation,
2iτ0 U̇n = (Un+1 − Un )|Un+1 − Un | p−1 − (Un − Un−1 )|Un − Un−1 | p−1 (6.2)
where τ0 is a constant depending on the system parameters and p = 3/2 for the
usual Hertzian interaction; see [2–4] for details. This equation is called the discrete p-
Schrödinger (DpS) equation since it is a discretization of the p-Schrödinger equation,
which is the usual linear Schrödinger equation with the Laplacian term replaced with
a p-Laplacian term [3]. As in the case of the standard discrete nonlinear Schrödinger
equation [5], a stationary equation can be obtained from the DpS equation by insertion
of the ansatz Un = Rn eiμt ,
μRn = (Rn+1 − Rn )|Rn+1 − Rn | p−1 − (Rn − Rn−1 )|Rn − Rn−1 | p−1
where a rescaling has been applied, see [3, Sec. 2.2]. Homoclinic orbits connecting
the origin of the above second order difference equation will yield a solution of
the DpS equation that is localized in space. Such solutions, if they exist, can be
approximated numerically via Newton iterations, see Fig. 6.2. At the level of the
6.1 Breathers in the Newton’s Cradle 71
10 4
8 10 5
yn (0)
yn (0)
0 0
8 10 5
10 4
5 10 15 20 25 30 5 10 15 20 25 30
n n
Fig. 6.2 Examples of standing bright breather solutions of the Newton’s cradle problem Eq. (6.1)
found via Newton iterations. As in previous examples (of dark and bright breathers), we can observe
the existence of intersite (left panel) and onsite (right panel) solutions. [Figure credit: Used with
permission from [2], Copyright (2013) by Elsevier.]
Newton’s cradle problem, the solution will be spatially localized and periodic in
time, i.e., it will be a breather [2, 3, 6]. The multiple-scale analysis leading to the
DpS equation has been justified rigorously (through suitable error bounds) in [4].
Larger amplitude breathers and traveling breathers of Eq. (6.1) with harmonic and
anharmonic onsite forces have been studied numerically in [2].
6.2 Solitary Waves and Breathers in Locally Resonant

Granular Chains
We now turn to a different situation that has been of considerable interest to recent
experimental investigations. In this case, instead of vibrating beads embedded within
the chain (e.g., in the dimer), vibrating (or resonating) elements are connected to each
particle (or embedded in the particles themselves). Such chains are called locally
resonant granular chains. For example, assuming that the attached resonator is linear,
the rescaled equations of motion for a monomer chain with no precompression are:
3/2 3/2
ü n = [u n−1 − u n ]+ − [u n − u n+1 ]+ − κ(u n − vn ) ,
(6.3)
ν v̈n = κ(u n − vn ) ,
where vn is the displacement of the attached resonator. Direct physical representa-

tions of a locally resonant granular chain include “mass-in-mass” systems (in which
each particle contains another vibrating particle) [7, 8] or “mass-with-mass” sys-
tems (in which the resonators are external rather than internal) [8, 9]. Effective
locally resonant chains are also possible to derive. For example, the “woodpile”
phononic structure, which consists of a stacked array of cylinders (i.e., rods), shown
in Fig. 6.3a can be modeled by Eq. (6.3), see Fig. 6.3b. Each bending mode in a
Fig. 6.3 a A woodpile chain, which can be modeled as b a locally resonant granular crystal. Here,
βs , βc and βw are the spring coefficients of the spring of the location of the striker, bulk of the
chain and wall, respectively. k1 is the linear spring constant of the local resonator which has mass
m 1 and displacement given by vi . V0 is the velocity of the striker particle. c The first four bending
modes of a single rod. [Figure Credits: We adapted panels (a) and (b) with permission from Ref. [11].
Copyrighted (2015) by the American Physical Society. Panel (c) is used with permission from [10].]
single rod can be excited by vibrating the rod at the appropiate frequency, see
Fig. 6.3c. Thus, each odd-index—in the enumeration of Fig. 6.3— mode1 can be
modeled as a harmonic oscillator with a stiffness determined by the frequency that
excites that particular mode. An optimization procedure can then be used to deter-
mine the effective mass [10].
With these physical motivations in hand, we now summarize a few studies con-
cerning locally resonant media, and in particular Eq. (6.3). Although Eq. (6.3) rep-
resents a chain with no precompression, the presence of the resonator makes the
1 Since these will affect the dynamics in the direction of the wave propagation along the array (i.e.,
the vertical direction when considering the orientation in Fig. 6.3a).
6.2 Solitary Waves and Breathers in Locally Resonant Granular Chains 73
equation linearizable. In the linearized equations of motion, one finds that √ the rela-
tive motion of the particles u n − vn oscillates with the frequency ω0 = 1 + 1/ν.
This suggests that a structure that is spatially localized must have tails that oscillate
with a frequency given by ω0 . If that localized structure were to travel perfectly undis-
turbed through the lattice (i.e., a solitary wave), then its oscillating tails should travel
with the core at the same speed c. In other words, its tail should have a wavenumber k0
satisfying k0 = ω0 /c. This represents a different kind of solitary wave than discussed
in Chap. 3. It consists of a highly localized solitary waveform on top of an extended,
small-amplitude periodic tail. Such a solution, satisfying all the other requirements
of a solitary wave except that it asymptotes not to zero but to a small-amplitude
oscillation at infinity, has been long termed a nanopteron [12]. See the book [13] for
a summary of results on nonlocal solitary waves.
To understand why a solitary wave of Eq. (6.3) would have an oscillating tail, we
follow the pattern detailed in Sect. 3.4 for the study of solitary waves, see also [11,
14]. We define the strain variable of each field, namely yn = u n−1 − u n and sn =
vn−1 − vn . Within these strain variables, we seek traveling waves therein as yn (t) =
φ(n − ct) = φ(ξ) and sn = ψ(n − ct) = ψ(ξ), leading to the equations
3/2 3/2 3/2
c2 ∂ξ2 φ = [φ(ξ − 1)]+ − 2 [φ(ξ)]+ + [φ(ξ + 1)]+ − κ(φ − ψ) (6.4)
c2 ∂ξ2 ψ = κ(φ − ψ) (6.5)
∞ ∞
Using the Fourier transform φ(ξ) = −∞ φ̂(k)eikξ dk and ψ(ξ) = −∞ ψ̂(k)eikξ dk
leads from Eqs. (6.4)–(6.5) to

1 2 k κ
φ̂ = sinc φ3/2 + 2 2 (φ̂ − ψ̂), (6.6)
c2 2 k c
κ
ψ̂ = φ̂. (6.7)
κ − c2 k 2 ν
Substituting Eq. (6.7) into Eq. (6.6) and reshaping the relevant expression yields

1 k 1 κ 2 k
φ̂ = 2 sinc2 + 4 2 sinc φ3/2 . (6.8)
c 2 c k − k02 2
√
Recall that sinc(x) = sin(x)/x and k0 = ω0 /c with ω0 = 1 + 1/ν being the fre-
quency of the relative motion of the main particles and their respective local res-
onators. Using properties of the convolution leads to
∞
φ(x) = K ∗ φ 3/2
= K (x − y)φ3/2 (y)dy, (6.9)
−∞
(a) (b)
Fig. 6.4 Examples of nanoptera profile φ(ξ) versus the traveling coordinate ξ a and zoom-in of
their tails b in the locally resonant granular chain. The dashed lines correspond to non-anti-resonant
values of the wavenumber k0 , whereas the solid line corresponds to an anti-resonant value, leading
to vanishing oscillations in the tail. [Figure credit: Used with permission from [4], Copyright (2015)
by the Institute of Physics.]
where K (x) = Λ(x) + M(x), where Λ(x) = (1/c2 )[1 − |x|]+ . Note Λ(x) appears
in the corresponding calculation for the granular chain without internal resonators;
see Sect. 3.4 and [15]. For M(x) we find:
(2c4 k03 /κ)M(x) = |1 − x|(sinc(k0 (1 − x)) − k0 ) (6.10)

−2|x|(sinc(k0 x) − k0 ) + |x + 1|(sinc(k0 (x + 1)) − k0 ).
In Sect. 3.4 this formulation was useful, since one can infer the decay properties
of the solitary wave (recall that it was double exponential in the case of no
precompression) [15], it provided a framework to prove the existence of traveling
solitary waves [16], and naturally lent itself to an algorithm for the computation of
such waves [4, 15]. In the case of Eq. (6.3), the formulation in Eq. (6.9) demonstrates
that the sinusoidal dependence with the periodicity dictated by k0 within M(x) is
directly responsible for the formation of the nanopteronic tails; cf. also the resonant
term in the Fourier space expression of Eq. (6.8). In the granular chain without the
resonators, the presence of solely the Λ term in Eq. (6.8) produces a monotonically
decaying solitary wave according to a double exponential law [15, 16]. We should
note here that direct iterations of Eq. (6.8) allow one to construct the numerically
exact form of the nanopteron solution which is supported against the backdrop of a
potentially non-vanishing background [14, 18]. For special values of the wavenumber
k0 = 2nπ (the so-called anti-resonances), the oscillations in the tail can vanish [18],
see Fig. 6.4. However, generically, the relevant structures will be present. Nanoptera
were found experimentally in woodpile phononic crystals in [11].
6.2 Solitary Waves and Breathers in Locally Resonant Granular Chains 75
As in the Newton’s cradle problem, breathing states have also been explored in
locally resonant homogeneous granular chains. Periodic traveling waves and dark
breathers can exist [19]. The analysis is once again based on the derivation of a
DpS equation, like Eq. (6.2), however, the derived modulation equation relies on
averaging over the breather period, and has a considerably more complicated form
than Eq. (6.2). If one were to add precompression, the relevant modulation equation
(obtained via a multiple-scale analysis) is once again the NLS equation, which is
either of the defocusing or the focusing variety, depending on the particular location
in the acoustic or optical band. This, in turn, enables the construction of bright or
dark breathers [20]; one can find traveling bright breathers where a focusing NLS is
derived or stationary and traveling dark breathers where a defocusing NLS is derived.
References
1. S. Hutzler, G. Delaney, D. Weaire, F. MacLeod, Rocking Newton’s cradle. Am. J. Phys. 72,
1508–1516 (2004)
2. G. James, P.G. Kevrekidis, J. Cuevas, Breathers in oscillator chains with Hertzian interactions.
Physica D 251, 39 (2013)
3. G. James, Nonlinear waves in Newton’s cradle and the discrete p-Schrödinger equation. Math.
Models Meth. Appl. Sci. 21, 2335 (2011)
4. B. Bidégaray-Fesquet, E. Dumas, G. James, From Newton’s cradle to the discrete p-Schrödinger
equation. SIAM J. Math. Anal. 45, 3404–3430 (2013)
5. P.G. Kevrekidis, The Discrete Nonlinear Schrödinger Equation: Mathematical Analysis Numer-
ical Computations and Physical Perspectives (Springer, New York, 2009)
6. G. James, Y. Starosvetsky, Breather solutions of the discrete p-Schrödinger equation, Localized
Excitations in Nonlinear Complex Systems (Nonlinear Systems and Complexity 7) (Springer,
Berlin, Germany, 2014), p. 77
7. L. Bonanomi, G. Theocharis, C. Daraio, Wave propagation in granular chains with local reso-
nances. Phys. Rev. E 91, 033208 (2015)
8. P.G. Kevrekidis, A. Vainchtein, M. Serra-Garcia, C. Daraio, Interaction of traveling waves with
mass-with-mass defects within a Hertzian chain. Phys. Rev. E 87, 042911 (2013)
9. G. Gantzounis, M. Serra-Garcia, K. Homma, J.M. Mendoza, C. Daraio, Granular metamaterials
for vibration mitigation. J. Appl. Phys. 114, 093514 (2013)
10. E. Kim, J. Yang, Wave propagation in single column woodpile phononic crystals: Formation
of tunable band gaps. J. Mech. Phys. Solids 71, 33–45 (2014)
11. E. Kim, F. Li, C. Chong, G. Theocharis, J. Yang, P.G. Kevrekidis, Highly nonlinear wave
propagation in elastic woodpile periodic structures. Phys. Rev. Lett. 114, 118002 (2015)
12. J. Boyd, Weakly non-local solitary waves, in Nonlinear Topics in Ocean Physics: Proceedings
of the Fermi School, (Elsevier, Amsterdam, 1991), pp. 51–97
13. J.P. Boyd, Weakly Nonlocal Solitary Waves and Beyond-All-Orders Asymptotics: Generalized
Solitons and Hyperasymptotic Perturbation Theory (Springer, New York, 1998)
14. K. Vorotnikov, Y. Starosvetsky, G. Theocharis, P. Kevreidis, Wave propagation in a strongly
nonlinear locally resonant granular crystal. Preprint, (2017)
15. J.M. English, R.L. Pego, On the solitary wave pulse in a chain of beads. Proc. AMS 133, 1763
(2005)
16. A. Stefanov, P.G. Kevrekidis, On the existence of solitary traveling waves for generalized
Hertzian chains. J. Nonlinear Sci. 22, 327 (2012)
4. H. Xu, P.G. Kevrekidis, A. Stefanov, Traveling waves and their tails in locally resonant granular
systems. J. Phys. A Math. Theor. 48, 195204 (2015)
18. A. Stefanov, P.G. Kevrekidis, Traveling waves for monomer chains with pre-compression.
Nonlinearity 26, 539 (2013)
19. L. Liu, G. James, P. Kevrekidis, A. Vainchtein, Strongly nonlinear waves in locally resonant
granular chains. Nonlinearity 29, 3496 (2016)
20. L. Liu, G. James, P.G. Kevrekidis, A. Vainchtein, Breathers in a locally resonant granular chain
with precompression. (2016). arXiv:1603.06033
Chapter 7
Higher Dimensional Lattices
7.1 Overview
Two-dimensional granular crystals have been investigated far less than 1D configu-
rations. In two spatial dimensions, the nodes can be arranged in a number of ways.
While the role of disorder is of particular interest for engineering applications [1, 2],
we will focus on ordered configurations, such as hexagonal (see Figs. 7.1a,b) or square
packings.
The analytical, computational, and experimental study of higher dimensional
granular crystals is far more challenging. Indeed, at the time of writing, there is
a very limited set of considerations (especially so from a mathematical perspective)
as regards the three main structures considered in this volume, namely dispersive
shocks, genuine solitary waves, and breathers. Despite this, we have included a
chapter on higher dimensional systems to touch upon some of the emerging dynam-
ical features, outline preliminary work on breathers, and to highlight challenging
open problems for the mathematics, physics, and engineering communities. We start
by examining a trait not found in 1D systems, namely, conical diffraction.
7.2 Conical Diffraction in the Hexagonal Packing
While the square lattice seems to be the most natural extension of the 1D chain,
there are disadvantages from an experimental perspective. The particles can buckle
easily since adjacent particles on the diagonals are not in contact. One way around
this problem is to insert “intruder” particles into the vacant spaces between particles
[3–5].
Another option to obtain structural robustness is to arrange the configuration using
a hexagonal (also called triangular) packing, see Fig. 7.1. This is the case considered
in this section.

78 7 Higher Dimensional Lattices
n (b)
(a) F0
(m-1,n+1) (m,n+1)
(m-1,n) (m,n) (m+1,n)

m
(m,n-1) (m+1,n-1)
Fig. 7.1 a Diagram of lattice and indexing. The angle between the m axis and the n axis is θ = π/3.
b A hexagonal lattice compressed on each boundary with a force F0 will result in a static overlap,
which is denoted as δ in the text. The amount of compression is exaggerated in this figure for clarity.
This lattice features a single defect particle at the center
To write down the equations √ of motion, we first define the lattice basis vectors
e1 = (1, 0) and e2 = (1/2, 3/2). The motion will be two dimensional, so each
particle will have a vector to represent its displacement. For example, we let qm,n (t) =
(xm,n (t), ym,n (t)) ∈ IR2 represent the equilibrium displacement of the bead found at
position pm,n = d(me1 + ne2 ) in the plane, where d is the bead diameter. Here m
counts the index along the usual horizontal axis, while n counts the index along a
line that is diagonal (with angle θ = π/3 from the m axis), see Fig. 7.1a. If a static
force is applied to the boundary (see Fig. 7.1b) an overlap δ is induced between the
centers of each pair of adjacent particles. In this case, the positions of the beads in
equilibrium are
pm,n = (d − δ)(me1 + ne2 ) (7.1)
If the deformations are small relative to the bead diameter, then the Hertz law Eq. (1.1)
can be used to describe the magnitude of the force that results from the deformation.
The overlap between two adjacent particles is given as d − ρ, where d is the bead
diameter, and ρ is the distance between their centers. Thus, the force would have
3/2
magnitude A [d − ρ ]+ . Given the symmetry of the lattice, there will be three dif-
ferent force functions. For example, to determine the force between the node with
index (m, n) and (m + 1, n) one first computes the distance between their centers:
ρ = |(qm+1,n + pm+1,n ) − (qm,n + pm,n )| (7.2)

= (d − δ + xm+1,n − xm,n )2 + (ym+1,n − ym,n )2 . (7.3)
7.2 Conical Diffraction in the Hexagonal Packing 79
Repeating this exercise for the other contact points, one finds the force functions
have magnitude given by
3/2
|F j (x, y)| = A d − ((d − δ) cos(θ j ) + x)2 + ((d − δ) sin(θ j ) + y)2
+
where each angle θ1 = 0, θ2 = π/3, and θ3 = −π/3, accounts for two neighbors,
making a total of six neighbors. Now that we have accounted for the six points of
contact in the hexagonal configuration we can write down the equations of motion
M q̈m,n = F1 (qm,n − qm−1,n ) + F2 (qm,n − qm,n−1 ) − F3 (qm+1,n−1 − qm,n )

−F1 (qm+1,n − qm,n ) − F2 (qm,n+1 − qm,n ) + F3 (qm,n − qm−1,n+1 )
(7.4)
These equations ignore many other forces (plasticity, viscous damping, rotation
dynamics) but have shown to be in reasonable agreement with experiments, see [6].
In the presence of precompression, one can linearize the equations of motion. The
analysis of the linear equations is similar to the 1D situation described in Sect. 1.2.
In particular, the linear equations have the plane wave solution

a i(km+ n2 (k+√3)+ωt)
qm,n (t) = e , (7.5)
b
where k, are the spatial wavenumbers and the frequency ω is given explicitly in
terms of k, and the entries of the Jacobian matrices of the force functions D F j , j =
1, 2, 3, see [7] for details. An example of a dispersion surface is shown in Fig. 7.2.
By inspection of Fig. 7.2a, one sees that there are six points in the dispersion sur-
faces where upward and downward facing cones meet. These are the celebrated
Dirac points [8, 9]; see also the zoom of such a point in Fig. 7.2b. In the case
(a) (b)
ω ω
k k

Fig. 7.2 a Dispersion relationship of Eq. (7.4). The wavenumber vector is (k, ) and the angular
frequency is ω. b Zoom-in of the dispersion surface near the Dirac point (kd , d ) = (4π/3, 0). The
transparent layers correspond to the Taylor expansion of ω(k, ) centered at a Dirac point. [Figure
credit: Used with permission from [7], Copyright (2016) by the American Physical Society.]
of Eq. (7.4), there are six nontrivial ones, which can√be computed explicitly as
(kd , d ) = (± 4π/3, 0) and (kd , d ) = (± 2π/3, ± 2π/ 3). Taylor expansion of the
dispersion relation ω(k, ) about one of the Dirac points verifies the conical nature
of the dispersion around those points [7], as seen also in the transparent surfaces
of Fig. 7.2b. Graphene is another notable example where Dirac points are found in
the dispersion. Graphene, which is a monolayer of graphite where extremely high
electron mobility is possible [10, 11], has gained significant recent attention due to
the groundbreaking work of Novoselov and Geim (earning them the 2010 Nobel
Prize in Physics). This has resulted in a large amount of work concerning Dirac
points in other settings, such as photonics, see [12] and references therein, which
motivates the term “photonic graphene”. The hexagonally packed granular network
can then be considered as a “phononic” version of graphene, see also [13–15]. The
nature of the dynamics will strongly depend on which Fourier modes are excited
(i.e., which ones are most dominant). This is particularly evident in the case of the
Dirac points. To develop a semi-quantitative characterization based on an analytical
approximation, we once again return to a continuum model approach. We consider a
linear plane wave solution that is modulated by a slowly varying envelope function
for our ansatz,
qm,n (t) = φ P (T )E + c.c., (7.6)

√
√ n n 3
E = ei(km+ 2 (k+
n
3)+ωt)
P= m+ , , T = t
2 2
where φ P (T ) ∈ C2 , 1. One can use this ansatz to derive the linear Dirac system,

∂ X −∂Y
∂T φ P (T ) = ±β φ P (T ), (7.7)
−∂Y −∂ X
assuming the wavenumbers (k, ) are near a Dirac point [7]. The coefficient β depends
on entries of the Jacobian matrices of the force functions D F j , j = 1, 2, 3. If the
spatial structure of a solution appears to be an expanding ring as time evolves, then
the dynamics is called conical diffraction. Conical diffraction has been found in the
Dirac system [16], which underscores the importance of the derivation of Eq. (7.7). To
confirm that conical diffraction is possible in the hexagonal granular crystal under
small-amplitude excitations, numerical simulations of Eq. (7.4) were carried out,
see Fig. 7.4a. The initial studies of conical diffraction (or refraction) took place as
early as 1837 [8, 9]. Conical diffraction has been identified in “photonic graphene”
as well [17, 18]. Nonlinearity can effect the dynamics quite drastically, and can
even lead to a breakdown of the conical diffraction, as shown within the context of
honeycomb lattices in [16, 19]. In [7], it was shown that for significantly nonlinear
responses conical diffraction can break down in the hexagonal granular lattice. While
the derivation of a nonlinear Dirac equation may shed light on this issue, the exact
role of nonlinearity on conical diffraction in the granular lattice remains an open
problem.
7.3 Other Work in Higher Dimensional Lattices and Future Directions 81
0.07
0.2
0.06
Particle velocity (m/s)

0.1
Y−position (m)
0.05
0.04
0
0.03
−0.1
0.02
0.01
−0.2 time=0.275ms time=0.525ms
0
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
X−position (m)
Fig. 7.3 Wave front shape in a hexagonal granular lattice upon impact with a striker bead. Colors
correspond to the particle velocity magnitude in m/s. [Figure credit: Used with permission from
[6], Copyright (2014) by Springer.]
7.3 Other Work in Higher Dimensional Lattices

and Future Directions
In a square or hexagonal lattice a single particle is in contact with multiple particles.

Thus, if energy is applied to a single particle, it will be distributed to an expanding
array of neighbors, see Fig. 7.3 and also Ref. [6] for numerical and experimental
demonstrations of this process. This spreading of energy should prevent the pos-
sibility of genuine 2D traveling solitary waves in such lattices. In a square lattice,
quasi-1D traveling solitary waves are possilbe by exciting the lattice in the vertical
or horizontal directions. However, this does not represent a new type of wave, since
each line of particles is acting like a separate 1D chain that does not interact with its
neighboring arrays.1 Even though an impact to the lattice does not lead to a genuine
2D solitary wave (unless it is a quasi-1D wave), there have been several related stud-
ies. For example, many consider the dynamics after a granular lattice is impacted
[6, 20–31]; this is particularly interesting for applications such as impact mitiga-
tion. It was shown in [32] that the continuum modeling approach can be applied to
2D lattices with rotation. Networks of 1D chains that live in 3D space (i.e. struc-
tured granular composites) can be used to achieve exponential stress decay [33]
and can be used effectively for energy-dispersion [34]. “Sound bullets” have also
been explored, which are waves that can be guided to preselected lattice sites within
the crystal [35]. While the geometry of the higher dimensional granular crystals
seems to prohibit genuine traveling solitary waves (and only permits some form of
self-similarly expanding wave fronts which are of interest in their own right [6]),
it also affords the possibility of breathers. Based on what we know about breathers
1 Notice, however, that the potential presence of intruders would enable this “cross-communication”
between adjacent chains, a feature that has been utilized to showcase phenomena such as energy
redistribution and equipartition [3–5].
(a) (b)
Fig. 7.4 a Conical diffraction observed in a 2D hexagonal granular lattice, see Eq. (7.4). Colors
correspond to the quantity |qm,n |. Panel (a) was obtained by exciting one of the linear modes in
the vicinity of a Dirac point. b A (defect-induced) breather found numerically in a 2D hexagonal
granular lattice with an impurity at the center of the lattice. The quantity |qm,n | is shown. [Figure
credit: Panel (a) used with permission from [7], Copyright (2016) by the American Physical Society.]
in one-dimensional granular chains, we need precompression and also some form

of heterogeneity (e.g., defects, periodicity, or local resonators). In such scenarios,
open problems include the existence and stability properties (and more generally the
dynamics) of breathers and their variants (such as defect modes, as detailed for 1D
chains in Chap. 5). For example, in a hexagonal lattice with a single impurity at the
center (see Fig. 7.1b) our preliminary numerical computations suggest that breathers
exist, see Fig. 7.4b. A more in-depth study of breathers in higher dimensional granular
crystals is an open subject requiring further research.
References
1. R.P. Behringer, Jamming in granular materials. C. R. Phys. 16, 10 (2015)

2. A. Liu, S.R. Nagel, The jamming transition and the marginally jammed solid. Annu. Rev. Cond.
Matter Phys. 1, 347 (2010)
3. I. Szelengowicz, M.A. Hasan, Y. Starosvetsky, A. Vakakis, C. Daraio, Energy equipartition in
two-dimensional granular systems with spherical intruders. Phys. Rev. E 87, 032204 (2013)
4. I. Szelengowicz, P.G. Kevrekidis, C. Daraio, Wave propagation in square granular crystals with
spherical interstitial intruders. Phys. Rev. E 86, 061306 (2012)
5. Y. Starosvetsky, M.A. Hasan, A.F. Vakakis, Nonlinear pulse equipartition in weakly coupled
ordered granular chains with no precompression. J. Comp. Nonlin. Dyn. 8, 034504 (2013)
6. A. Leonard, C. Chong, P.G. Kevrekidis, C. Daraio, Traveling waves in 2D hexagonal granular
crystal lattices. Granul. Matter 16, 531 (2014)
7. C. Chong, P.G. Kevrekidis, M.J. Ablowitz, Y.-P. Ma, Conical wave propagation and diffraction
in 2D hexagonally packed granular lattices. Phys. Rev. E 93, 012909 (2016)
8. W.R. Hamilton, Third supplement to an essay on the theory of systems of rays. Trans. R. Ir.
Acad. 17, 1 (1837)
9. H. Lloyd, On the phenomena presented by light in its passage along the axes of biaxial crystals.
Trans. R. Ir. Acad. 17, 145 (1837)
10. K.S. Novoselov, A.K. Geim, S.V. Morozov, D. Jiang, Y. Zhang, S.V. Dubonos, I.V. Grigorieva,
A.A. Firsov, Electric field effect in atomically thin carbon films. Science 306, 666 (2004)
References 83
11. K.S. Novoselov, A.K. Geim, S.V. Morozov, D. Jiang, M.I. Katsnelson, I.V. Grigorieva, S.V.
Dubonos, A.A. Firsov, Two-dimensional gas of massless Dirac fermions in graphene. Nature
438, 197 (2005)
12. M.J. Ablowitz, C.W. Curtis, Y.-P. Ma, Linear and nonlinear traveling edge waves in optical
honeycomb lattices. Phys. Rev. A 90, 023813 (2014)
13. D. Torrent, J. Sánchez-Dehesa, Acoustic analogue of graphene: observation of Dirac cones in
acoustic surface waves. Phys. Rev. Lett. 108, 174301 (2012)
14. D. Torrent, D. Mayou, J. Sánchez-Dehesa, Elastic analog of graphene: Dirac cones and edge
states for flexural waves in thin plates. Phys. Rev. B 87, 115143 (2013)
15. T. Antonakakis, R.V. Craster, S. Guenneau, High-frequency homogenization of zero-frequency
stop band photonic and phononic crystals. New J. Phys. 15, 103014 (2013)
16. M.J. Ablowitz, Y. Zhu, Nonlinear dynamics of bloch wave packets in honeycomb lattices, in
Spontaneous Symmetry Breaking, Self-Trapping, and Josephson Oscillations (Springer, Ger-
many, 2013), p. 1
17. O. Peleg, G. Bartal, B. Freedman, O. Manela, M. Segev, D.N. Christodoulides, Conical diffrac-
tion and gap solitons in honeycomb photonic lattices. Phys. Rev. Lett. 98, 103901 (2007)
18. M.J. Ablowitz, S.D. Nixon, Y. Zhu, Conical diffraction in honeycomb lattices. Phys. Rev. A
79, 053830 (2009)
19. O. Bahat-Treidel, O. Peleg, M. Segev, H. Buljan, Breakdown of Dirac dynamics in honeycomb
lattices due to nonlinear interactions. Phys. Rev. A 82, 013830 (2010)
20. S.G. Bardenhagen, J.U. Brackbill, Dynamic stress bridging in granular materials. J. Appl.
Physics 83, 5732 (1998)
21. K.M. Roessig, J.C. Foster, S.G. Bardenhagen, Dynamic stress chain formation in a two-
dimensional particle bed. Exp. Mech. 42, 329 (2002)
22. A. Leonard, C. Daraio, Stress wave anisotropy in centered square highly nonlinear granular
systems. Phys. Rev. Lett. 108, 214301 (2012)
23. A. Leonard, C. Daraio, A. Awasthi, P. Geubelle, Effects of weak disorder on stress wave
anisotropy in centered square nonlinear granular crystals. Phys. Rev. E 86, 031305 (2012)
24. A.P. Awasthi, K.J. Smith, P.H. Geubelle, J. Lambros, Propagation of solitary waves in 2D
granular media: a numerical study. Mech. Mater. 54, 100 (2012)
25. M.S. Abd-Elhady, S. Abd-Elhady, C.C.M. Rindt, A.A. van Steenhoven, Force propagation
speed in a bed of particles due to an incident particle impact. Adv. Powder Tech. 21, 150
(2010)
26. M. Nishida, Y. Tanaka, DEM simulations and experiments for projectile impacting two-
dimensional particle packings including dissimilar material layers. Gran. Matter 12, 357 (2010)
27. M. Nishida, K. Tanaka, T. Ishida, DEM simulation of wave propagation in two-dimensional
ordered array of particles, in Shock Waves (Springer, Germany, 2009), p. 815
28. C. Coste, B. Gilles, Sound propagation in a constrained lattice of beads: high-frequency behav-
ior and dispersion relation. Phys. Rev. E 77, 021302 (2008)
29. B. Gilles, C. Coste, Low-frequency behavior of beads constrained on a lattice. Phys. Rev. Lett.
90, 174302 (2003)
30. O. Mouraille, A.W. Mulder, S. Luding, Sound wave acceleration in granular materials. J. Stat.
Mech: Theory Exp 7, 07023 (2006)
31. A. Leonard, F. Fraternali, C. Daraio, Directional wave propagation in a highly nonlinear square
packing of spheres. Exp. Mech. 53, 327 (2013)
32. V.I. Erofeev, V.V. Kazhaev, I.S. Pavlov, Nonlinear Localized Strain Waves in a 2D Medium
with Microstructure (Springer, Germany, 2013), p. 91
33. A. Leonard, L. Ponson, C. Daraio, Exponential stress mitigation in structured granular com-
posites. Extrem. Mech. Lett. 1, 23 (2015)
34. W.J. Falls, S. Sen, Solitary wave propagation through two-dimensional treelike structures. Phys.
Rev. E 89, 023209 (2014)
35. A. Spadoni, C. Daraio, Generation and control of sound bullets with a nonlinear acoustic lens.
Proc. Natl. Acad. Sci. USA 107, 7230 (2010)
Appendix
A.1 Numerical Computation of Traveling Solitary Waves

and Their Stability
A reasonable approximation to a genuine solitary wave solution of Eq. (1.3) can

be obtained by simply imparting velocity to the first particle and simulating suffi-
ciently long to let the solitary wave form [1]. A more accurate solitary wave can be
obtained by a Newton iteration scheme. We summarize the discussion of [2] which
details the procedure. First, we recall that genuine traveling waves of Eq. (1.6) with
precompression satisfy the equation
A p p
0 = −c2 φξξ + {[δ0 + φ(ξ − 1)]+ − 2 [δ0 + φ(ξ)]+
M
p
+ [δ0 + φ(ξ + 1)]+ }. (A.1)
where ξ = n − ct and φ(ξ) = yn (n − ct). To obtain numerical solutions of

Eq. (A.1), we can employ a uniform spatial discretization of ξ consisting of L
points ξk (k = − L−1 2
, . . . , 0, . . . , L−1
2
) with lattice spacing Δξ chosen such that
q = 1/Δξ is an integer. Then, the field φ(ξ) is replaced by its discrete counter-
part, i.e., φk := φ(ξk ) = φ(kΔξ). The second-order spatial derivative appearing in
Eq. (A.1) is replaced by a modified central difference approximation (φk−2 − 2φk +
φk+2 )/(4Δξ 2 ). The reason for this choice of central difference is connected to the sta-
bility calculation to be discussed below. Using this discretization, Eq. (A.1) becomes
the following root-finding problem,
φk−2 − 2φk + φk+2 A p

0 = −c2 + { δ0 + φk−q +
4Δξ 2 M
p p
− 2 [δ0 + φk ]+ + δ0 + φk+q + } (A.2)

SpringerBriefs in Physics, https://doi.org/10.1007/978-3-319-77884-6
86 Appendix
which can be solved via a Newton iteration scheme for the variables φk , see [3] for
details on Newton iteration schemes. We employ periodic boundary conditions at the
edges of the spatial grid. This method was employed to study solitary waves of the
granular chain in [4] and rarefaction waves in lattices with strain-softening potentials
(i.e., Eq. (1.3) with p < 1) [2], see also [5, 6] for other examples.
To investigate the spectral stability of a solution φ0 , we substitute the linearization
ansatz Φ(ξ, t) = φ0 + εa(ξ)eλt to arrive at Eq. (3.33), see Sect. 3.4. If φ0 is localized
in ξ, it can be seen as a compact perturbation of the eigenvalue problem in which
φ0 = 0. Thus, we expect the continuous spectra of these eigenvalue problems to
coincide [7]. If φ0 = 0, then we can solve Eq. (3.33) analytically with the ansatz
a(ξ) = eiξ , where the eigenvalues are given by

p−1
p Aδ0
λ() = ic ± 2i sin . (A.3)
2 M
This indicates that the continuous spectrum around the uniform steady state is
purely on the imaginary axis. Thus, any instability of the wave will be due to point
spectrum, which we must calculate numerically. We chose a discretization such that
the spectra of the zero solution (see Eq. (A.3)) corresponds to the spectra of the zero
solution of the discrete variant of Eq. (3.33). Using the standard central differences
for the spatial derivatives appearing in (3.33) will lead to eigenvalues with a real part
in the case of φ0 = 0. Hence, particular care needs to be taken to discretize the first
and second derivative operators in a “compatible” way. Indeed, in order to avoid this
“spurious instability,” we used a modified central difference formula for the second
spatial derivative, leading to the following eigenvalue problem
λak = bk , (A.4)
a k+2 − 2a k + a k−2 bk+1 − bk−1
λbk = −c2 +c + (A.5)
4Δξ 2 Δξ
A p−1 p−1 p−1
p δ0 + φ0k−q + ak−q − 2 δ0 + φ0k + ak + δ0 + φ0k+q + ak+q
M
which has been cast as a linear system in λ through the definition aλ = b (such that
standard eigenvalue solvers can be used), and where ak := a(kΔξ). In the case of
φ0 = 0, Eq. (A.4) with k ∈ ZZ is solved by ak = eikΔξ , where the eigenvalues are
given by
p−1
sin(Δξ) p Aδ0
λ() = i c ± 2i sin , (A.6)
Δξ 2 M
which is consistent with Eq. (A.3) in the limit Δξ → 0 and perhaps even more impor-
tantly is also purely on the imaginary axis, suggesting neutral stability. While this
choice of discretization seems reasonable, additional inconstancies may arise due to
the finite nature of the computations and introduction of boundary conditions, which
Appendix 87
are detailed in [2]. Thus, what the best discretization scheme is remains an open
problem, although recent work suggests that spectral collocation schemes may be
more well suited towards such a scope [8].
A.2 Numerical Computation of Breathers and Their

Stability
We now provide a brief summary of how to compute time-periodic solutions in

granular crystals (as well as more generally) and how to assess their stability by
computing Floquet multiplers. We begin by writing Eq. (1.3) as a system of first
order ODEs:
u
ẋ = F (t, u, v) , x= (A.7)
v

T
where u = u 1 , . . . , u p and v = u̇, respectively, represent the N -dimensional posi-
tion and velocity vectors.
In order to find periodic solutions to this system, we are searching for solutions x
so that x(0) = x(Tb ) where Tb is the fixed period of the solution. This suggests we
define the Poincaré map: P(x(0) ) = x(0) − x(Tb ), where x(0) is the initial condition
and x(Tb ) is the result of integrating Eq. (A.7) forward in time until t = Tb using
standard ODE integrators [9]. A periodic solution with period Tb (frequency 1/Tb )
will be a root of P. To obtain an approximation of this root, one can apply Newton’s
method to the map [10, Sect. 3],

x(0,k+1) = x(0,k) − [J ]−1

x(0,k)
P x(0,k) , (A.8)
where k is the index of the Newton iteration and x 0 is the desired root of P. The
Jacobian of the map P is J = I − V (Tb ), where I is the 2N × 2N identity matrix;
V is the solution to the variational problem
V̇ = [DF] (t) V , (A.9)
with initial data V (0) = I; and DF is the Jacobian of the equations of motion
Eq. (A.7) evaluated at the point x(0,k) (t). For each step, one simultaneously solves
Eq. (A.7) with initial data x(0,k) and Eq. (A.9). This iteration scheme is applied until a
user-prescribed tolerance criterion is satisfied. This method only ensures the solution
is periodic in time, with period Tb , and does not, for example, yield necessarily a
spatially localized structure. This highlights the importance of the initial step x(0,0) ,
which should have the desired spatial structure. The initial step (or guess) can be a
linear mode or a solution to a continuum nonlinear approximation.
There is a rich theory for the stability properties of breathers in Hamiltonian
systems, see [10, 4.2] and [11, 12]. A linear stability problem is determined in the
88 Appendix
standard way: A small perturbation V (t) is added to a solution x(t) that is time-
periodic. The result x(t) + V (t) is substituted into Eq. (A.7) to obtain an equation
describing the evolution of V . Keeping only linear terms in V will result in Eq. (A.9).
Thus, using Newton iterations for computation of time-periodic solutions immedi-
ately yields stability information of that solution, as explained below.
Since [DF] (t) will be periodic in time with period Tb , Eq. (A.9) represents a Hill’s
equation. It is well known within Floquet theory [13], that the fundamental solutions
of Eq. (A.9) have the property V (t + Tb ) = ρV (t), where ρ is a so-called Floquet
multiplier (FM). The Floquet multipliers are the eigenvalues of the matrix V (Tb ),
where V (Tb ) is the solution of Eq. (A.9) with initial data given by the identity matrix.
Thus, the perturbation will exhibit exponential growth if there is at least one Floquet
multiplier with |ρ| > 1. In this case, the corresponding solution x is deemed unstable.
Otherwise, the solution is called spectrally stable. Due to the Hamiltonian structure
of the system, all Floquet multipliers must lie on the unit circle for the solution to be
spectrally stable, otherwise, the solution is unstable, see [10, Sect. 4] for more details.
There are continuous arcs of spectrum on the unit circle (in the infinite-lattice limit),
and one can compute these arcs from the linear spectral bands of Eq. (1.3). In general,
the isolated multipliers (i.e., the “point spectrum”) must be computed numerically.
For Hamiltonian systems, there is always a pair of FMs at the point +1 (i.e., at the
point (1, 0) of the unit circle). These correspond to the invariance of the system under
time translation—a feature responsible for the conservation of the total energy. By
exploiting information about this FM pair, one can extract a stability criterion for
breathers [14] that is reminiscent of the well-known Vakhitov–Kolokolov criterion
for the stability of solitary waves, stating that when the monotonicity of the breather
energy as a function of its frequency changes, its stability will change too. Equally
remarkably, the fact that traveling waves are periodic orbits (modulo shifts) on the
lattice has allowed to extend such a criterion to traveling waves in a criterion stating
that when the monotonicity of the traveling wave energy as a function of its speed
changes, then its stability should also change. This topic presently constitutes an
active field of research [8].
Additional conservation laws of the system will lead to additional pairs of FMs
at the point +1. Such FMs also complicate the computation of breathers using the
Newton’s method described above, since the Jacobian J = I − V (Tb ) is not invert-
ible if there is a unit FM. To overcome this issue, one must break the degenerate
nature of J by introducing additional constraints, such as a vanishing time average
T
(i.e., 0 y1 (t)dt = 0) or a pinning condition (e.g., y1 (t) = 0). In practice, it can also
be effective to take a pseudoinverse of J .
Finally, it is important to note that spectral stability does not exclude the possi-
bility of algebraic growth of perturbations [15], or nonlinear instabilities [10]. For
breathers in granular crystals, such cases have been checked through direct numerical
simulations. See [10, 11] for numerous practical methods for computing the relevant
periodic orbits and their stability.
Appendix 89
A.3 Derivation of the Nonlinear Schrödinger Equation
The derivation of the nonlinear Schrödinger (NLS) equation from the granular chain
model will follow a similar pattern as the derivation of the KdV equation that appears
in Sect. 3.2. However, there is quite a bit of technical detail involved, even in the case
of the monomer chain with precompression. The derivation of the NLS equation from
a monomer FPUT lattice is given in [16], however not many details of the calculations
are provided in that work. The derivation of the FPUT lattice with an onsite potential
is considered in [17] (where also a rigorous justification is also provided), which
is accompanied by a great level of detail. In [18], the NLS is derived in the strain
variable formulation of the FPUT lattice. In [19], the NLS equation is derived from
the strain variable formulation of the FPUT lattice in Fourier space and rigorously
justified. The derivation presented here will be similar to [17], but will be modified
since no onsite potential is present in granular chain.
We start the discussions from Eq. (1.3) in the case of the monomer, where all the
prefactors An /Mn are equal. We write Eq. (1.3) in a slightly different form, which
will make the notation for our derivation less cumbersome,
d 2un
= −V (−∂n+ u) + V (−∂n− u) (A.10)
dt 2
where the notation for the forward and backward differences are given by ∂n+ =
u n+1 − u n and ∂n− = u n − u n−1 , respectively. Here we have defined,
A[δ0 + x] p
V (x) =
M
which has absorbed the mass M. Assuming now that the relative displacements
satisfy
u n−1 − u n δ0 (A.11)
we can approximate −V (−x) with its third order Taylor polynomial,

p
Aδ0
− V (−x) ≈ W (x) = J1 + J2 x + J3 x 2 + J4 x 3 , J1 =
M
p−1 p−2 p−3
Apδ0 Ap( p − 1)δ0 Ap( p − 1)( p − 2)δ0
J2 = , J3 = − , J4 =
M 2M 6M
(A.12)
Note that Js is defined simply as Js = K s /M for s = 1...3. We use this notation to

avoid dividing by M throughout the derivation. Recognizing that the leading terms
vanish when using the approximation W (x) instead of −V (−x) in Eq. (A.10), we
can see that our new model acquires the form of an FPUT lattice [20] according to:
90 Appendix
d 2un
= W (∂n+ u) − W (∂n− u) (A.13)
dt 2
Note, we defined W (x) through the Taylor expansion of −V (−x) rather than V (x)
in order to obtain an FPUT model in the standard form shown in Eq. (A.13). The
remainder of the analysis will be on Eq. (A.13).
The linear part of Eq. (A.13) (namely when J3 = J4 = 0) has the form
d 2u
= Lu = J2 (∂n+ u − ∂n− u) (A.14)
dt 2
and was analyzed in Sect. 1.2. It was shown that the linearized equation has plane
wave solutions
E n (t) = ei(kn+ω(k)t)
where ω(k) is given by the dispersion relation,
ω(k)2 = 2J2 (1 − cos(k)).
We will study the modulation of planes waves under the evolution of the nonlinear
equation Eq. (A.13) using the ansatz
u n (t) =
U (X, T )E n (t) + c.c + h.o.t. X =
(n − ct), T =
2 t (A.15)
where c.c. is the complex conjugate and h.o.t are higher order terms (see below).
This choice may appear to be somewhat unexpected at first, however, it is possible to
convince ourselves that it is the most natural one. We scale space by
, in which case
the dispersive scaling of NLS’ space and time derivatives necessitates that we use an
2 for the time rescaling. As regards the ansatz of the solution, we seek a solution that
explicitly includes a breathing in time (with frequency ω(k)) and a spatial modulation
with wavenumber k. The motivation here is that we have in mind a rapidly oscillating
plane wave that is modulated by a slow envelope provided by U (X, T ). The factor of
for the amplitude ensures that the solution is of small amplitude, making the Taylor
expansion used previously valid. Direct substitution of ansatz (A.15) into Eq. (A.13)
will generate higher order harmonics (i.e., terms like E s ) and thus one must include
an ansatz incorporating all possible harmonics. Thus, we will work with the ansatz,

3
s
u n (t) ≈ ψn (t) =
s Us, j (X, T )E n (t) j (A.16)
s=1 j=−s
where Us, j ∈ C and Us,− j = Us, j . Note, this implies Us,0 ∈ IR for s = 1, 2, 3. Insert-
ing this ansatz into Eq. (A.13) will produce a fairly large expression. To manage this,
Appendix 91
we will collect terms according to the power in

, but also their power in E to create
a hierarchy of solvability conditions. Moreover, within each of these orders, we will
organize terms in three different parts, namely those coming from the left-hand side
of Eq. (A.13) (the time derivative part), the linear part Lψ, and the nonlinear part
N (ψ), which is defined as,
N (ψ) = J3 [(∂n+ ψ)2 − (∂n− ψ)2 ] + J4 [(∂n+ ψ)3 − (∂n− ψ)3 ]
As we will see below, the NLS equation appears at the order

3 E. This explains
why we only include powers up to s = 3 in our ansatz. The forward and backward
differences of the ansatz have the form

3
s

∂n± ψn = ±
s Us, j (X ±
)e±ik j − Us, j (X ) E j
s=1 j=−s
where we have dropped the subscript and arguments of E n (t) and the T argument
from Us, j (X, T ) for notational simplicity. Rather than substituting this expression
directly into the right-hand side of Eq. (A.13), we make use of the expansion

2
U (X ±
) = U (X ) ± ∂ X U (X )
+ ∂ X2 U (X ) + O(
3 ) (A.17)
2
Note that since the lowest order term in our ansatz has order
, we need not keep
the O(
3 ) terms in the expansion given in Eq. (A.17). This is because the solvability
conditions at O(
4 ) will not be needed to derive the NLS equation. Below, we list the
solvability conditions appearing at each order. All terms appearing on the left-hand
side of the equality will be those resulting from the time derivative part ddtψ2 . All terms
2
appearing on the right-hand side of the equality without an underline will be those
resulting from the linear term Lψ. Finally, all terms appearing on the right-hand
side of the equality with an underline will be those resulting from the nonlinear part
N (ψ).
We start with the O(
) terms, in which there is only one solvability condition:

1 E 1 : −ω 2 U1,1 = −2J2 (1 − cos(k))U1,1
This equation is satisfied if ω obeys the linear dispersion relation ω 2 = −2J2 (1 −

cos(k)). Moving on to the O(
2 ) terms, we have two solvability conditions:

2 E 1 : −ω 2 U2,1 − 2iωc∂ X U1,1 = −ω 2 U2,1 + 2iωω ∂ X U1,1

2 E 2 : −4ω 2 U2,2 = −ω 2 (2k)U2,2 − J3 s1 c1 U1,1
2
92 Appendix
Here we used the fact that ω 2 = 2J2 (1 − cos(k)) and ωω = J2 sin(k). Later, we
will also use the fact (ω )2 + ωω = J2 cos(k). The definitions s1 = 2i sin(k) and
c1 = 2(1 − cos(k)) also simplify the relevant expressions. The equation for
2 E 1
yields the group velocity relation: c = −ω (k). The equation for
2 E 2 yields an
expression for U2,2
J3 s1 c1
U2,2 = U2 (A.18)
4ω 2 − ω 2 (2k) 1,1
Note that 4ω 2 = ω 2 (2k) for k = 0. Finally, moving on to the O(

3 ) terms, we have
the solvability conditions:

3 E 0 : c2 ∂ X2 U1,0 = (ω (0))2 ∂ X2 U1,0 + 2J3 c1 (U1,1 ∂ X U1,1 + U1,1 ∂ X U1,1 )

3 E 1 : −ω 2 U3,1 − 2iωc∂ X U2,1 + c2 ∂ X2 U1,1 + 2iω∂T U1,1
= −ω 2 U3,1 + 2iωω ∂ X U2,1 + [(ω )2 + ωω ]∂ X2 U1,1
+2J3 s1 c1 U1,1 U2,2 − 2J3 c1 ∂ X U1,0 U1,1 − 3J4 c12 |U1,1 |2 U1,1

3 E 2 : −4ω 2 U3,2 − 4iωcU2,2
= −ω 2 (2k)U3,2 + 2iω(2k)ω (2k)∂ X U2,2
+2J3 c1 (c1 − 3)U1,1 ∂ X U1,1 − 2J3 s1 c1 U1,1 U2,1

E : −9ω 2 U3,3 = −ω 2 (3k)U3,3
3 3
+2J3 s1 (c1 + s12 )U1,1 U2,1 + J4 c12 (3 − c1 )U1,1

3
The equation for

3 E 0 yields,
2J3 c1
∂ X2 U1,0 = ∂ X |U1,1 |2
c2 − (ω (0))2
Integrating both sides, and setting the integration constant to zero, yields
2J3 c1
∂ X U1,0 = |U1,1 |2 (A.19)
c2 − (ω (0))2
Using Eqs. (A.18) and (A.19), and the fact that c = −ω (k), the equation for
3 E 1
simplifies to

(J3 s1 c1 )2 (2J3 c1 )2
2iω∂T U1,1 = ωω ∂ 2X U1,1 + 2 − − 3J4 c 1 |U1,1 | U1,1 (A.20)
2 2
4ω 2 − ω 2 (2k) c2 − (ω (0))2
Simplifying once more, we arrive at the NLS equation:

Appendix 93

ω 2 c12 (J3 s1 )2 (2J3 )2
i∂T U1,1 = ∂ U1,1 + 2 − − 3J4 |U1,1 |2 U1,1
2 X 2ω 4ω 2 − ω 2 (2k) c2 − (ω (0))2
(A.21)
The functions U3,2 and U3,3 can be calculated from U1,1 using the equations for
3 E 2 and
3 E 3 respectively. If one wants to express the solution in the strain variable
yn = ∂n+ u, we simply compute the forward difference of our ansatz. To first order,
this is

yn = ∂n+ ψ =
(U1,0 (X +
) − U1,0 ) +
(U1,1 (X +
)eik − U1,1 )E + c.c. + O(
2 )
Using once again the expansion Eq. (A.17) the approximation in the strain becomes
∂n+ ψ =
(U1,1 (eik − 1)E + c.c.) + O(
2 )
Note, the terms without the harmonic factors, U1,0 have vanished. Now, we define
an envelope function for the strain Y (X, T ) = U1,1 (X, T )(eik − 1), such that our
ansatz for the strain equation becomes (up to first order)
yn (t) =
Y (X, T )E + c.c.
If we substitute U1,1 = Y (X, T )(eik − 1)−1 into Eq. (A.21) and simply, we obtain,

ω 2 c1 (J3 s1 )2 (2J3 )2
i∂T Y − ∂ Y− 2 − − 3J4 |Y |2 Y = 0
2 X 2ω 4ω 2 − ω 2 (2k) c2 − (ω (0))2
(A.22)
where we used the fact that |eik − 1|2 = c1 . Note, this is the same as the NLS equation
in the displacement formulation, except for the fact that the nonlinear coefficient is
scaled by the factor c1 (compare Eqs. (A.21) and (A.22)). If we let k = π, then we
have s1 = 0, c = 0, c1 = 4 and so the nonlinear coefficient

c1 (J3 s1 )2 (2J3 )2
ν3 = − 2 2 − − 3J4
2ω 4ω − ω 2 (2k) c2 − (ω (0))2
reduces to
1
B= √ (3J4 J2 − 4J32 )
J2 J2
√
where we used the fact that (ω (0))2 = J2 and ω(π) = 2 J2 . This is the value of the
nonlinear coefficient of the NLS equation appearing in Sect. 4.1.
94 Appendix
References

2. H. Yasuda, C. Chong, E.G. Charalampidis, P.G. Kevrekidis, J. Yang, Formation of rarefaction
waves in origami-based metamaterials. Phys. Rev. E 93, 043004 (2016)
4. H. Xu, P.G. Kevrekidis, A. Stefanov, Traveling waves and their tails in locally resonant granular
systems. J. Phys. A Math. Theor. 48, 195204 (2015)
5. M. Duanmu, N. Whitaker, P.G. Kevrekidis, A. Vainchtein, J. Rubin, Traveling wave solutions
in a chain of periodically forced couple nonlinear oscillators. Physica D 325, 25 (2016)
6. N. Nadkarni, A.F. Arrieta, C. Chong, D.M. Kochmann, C. Daraio, Unidirectional transition
waves in bistable lattices. Phys. Rev. Lett. 116, 244501 (2016)
7. T. Kato, Perturbation Theory for Linear Operators (Springer, Heidelberg, 1976)
8. J. Cuevas-Maraver, P.G. Kevrekidis, A. Vainchtein, H. Xu, Unifying perspective: solitary trav-
eling waves as discrete breathers in Hamiltonian lattices and energy criteria for their stability.
Phys. Rev. E 96, 032214 (2017)
9. E. Hairer, S. Nørsett, G. Wanner, Solving Ordinary Differential Equations I (Springer-Verlag,
Berlin, Germany, 1993)
1 (2008)
linear systems. Physica D 216, 1 (2006)
12. S. Flach, A.V. Gorbach, Discrete breathers in Fermi-Pasta-Ulam lattices. Chaos 15, 015112
(2005)
13. C. Chicone, Ordinary Differential Equations with Applications (Springer, New York, 1999)
14. P.G. Kevrekidis, J. Cuevas-Maraver, D.E. Pelinovsky, Energy criterion for the spectral stability
of discrete breathers. Phys. Rev. Lett. 117, 094101 (2016)
15. C. Chong, P.G. Kevrekidis, G. Theocharis, C. Daraio, Dark breathers in granular crystals. Phys.
Rev. E 87, 042202 (2013)
16. G. Huang, Z.-P. Shi, Z. Xu, Asymmetric intrinsic localized modes in a homogeneous lattice
with cubic and quartic anharmonicity. Phys. Rev. B 47, 14561 (1993)
17. J. Giannoulis, A. Mielke, The nonlinear schrödinger equation as a macroscopic limit for an
oscillator chain with cubic nonlinearities. Nonlinearity 17, 551 (2004)
18. I. A Butt , J. A. D Wattis. Asymptotic analysis of combined breather-kink modes in a Fermi-
Pasta-Ulam chain, Physica D, 231, 165–179, (2007)
19. G. Schneider, Bounds for the nonlinear Schrödinger approximation of the Fermi-Pasta-Ulam
system. Appl. Anal. 89, 1523 (2010)

(SpringerBriefs in Physics) Christopher Chong, Panayotis G. Kevrekidis - Coherent Structures in Granular Crystals-Springer International Publishing (2018) PDF

Enviado por

Dados do documento

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

(SpringerBriefs in Physics) Christopher Chong, Panayotis G. Kevrekidis - Coherent Structures in Granular Crystals-Springer International Publishing (2018) PDF

Enviado por

Direitos autorais:

Formatos disponíveis

SPRINGER BRIEFS IN PHYSICS

ISSN 2191-5423 ISSN 2191-5431 (electronic)

© The Author(s) 2018

Printed on acid-free paper

To my father, George, for wanting (and

Brunswick, ME, USA Christopher Chong

1 Introduction and Motivation of Models . . . . . . . . . . . . . . . . ...... 1

5 Heterogeneous Media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...... 51

1.1 Dynamics of Hertzian Beads

© The Author(s) 2018 1

A  3/2 3/2 3/2

where the potential function is defined as

a dissipation variant incorporating a discrete velocity Laplacian. This idea is par-

1.2 The Linear Discrete Wave Equation

2 The choice of N − 1 is made purely for reasons of notational convenience.

T  (t)X n = ω02 (X n+1 + X n−1 − 2X n )T (t) ⇒ (1.11)

T (t) = A sin(βω0 t) + B cos(βω0 t), A, B ∈ IR. (1.13)

while, essentially by analogy, the spatial part, X n+1 + X n−1 − 2X n = −β 2 X n , will

X n = C sin(qn) + D cos(qn) C, D ∈ IR, (1.14)

where A, B, C, D are arbitrary constants. To determine the relationship between q

eiqn+iq + eiqn−iq − 2eiqn  

where Ã = AC and B̃ = BC. Now, employing linear superposition to obtain the

1.3 The Nonlinear States: Dispersive Shock Waves,

demonstrated, and experimentally corroborated as existing, depending on the nature

1. V.F. Nesterenko, Dynamics of Heterogeneous Materials (Springer, New York, 2001)

Shock waves are nonlinear structures characterized by a discontinuous jump in the

© The Author(s) 2018 13

2.2 Theoretical Analysis of Shock Waves

a limit, however, a small parameter ε must be introduced (essentially characterizing

{δ0 + Y (X + ε)} p + {δ0 + Y (X − ε)} p − 2{δ0 + Y (X )} p

From the constant slope of these characteristic lines, we can write

where X 1 , X 2 ∈ R and X 2 = X 1 + h. At the time of their intersection, we obtain

In the limit of h → 0, we have

This is termed a “p-system” [3, 18]. Using

we can rewrite Eq. (2.16) as

2.3 Prototypical Numerical Computations

Following the discussion of [11], we use an initial strain given by

Y0 (X 0 ) = asech (bX 0 ) , (2.21)

1. M.J. Ablowitz, M. Hoefer, Dispersive shock waves. Scholarpedia 4, 5562 (2009)

© The Author(s) 2018 23

3.2 Traveling Waves With Precompression: The KdV

where we have dropped the explicit T dependence of U to simplify the notation.

Since we are seeking traveling waves, it is natural to go to a co-traveling frame.

∂τ v + 6v∂ξ v + ∂ξ3 v = 0. (3.11)

k12 f 1 + k22 f 2 + 2(k2 − k1 )2 f 1 f 2 + m(k22 f 12 f 2 + k12 f 1 f 22 )

level of the KdV analysis counter-propagating solitary wave collisions/interactions.

ü n = e[u n−1 −u n ] − e[u n −u n+1 ]

3.3 Traveling Waves Without Precompression I: The

Equations (3.19)–(3.20) have a number of advantages and disadvantages. On the

Y (X − cT ) = cm A1 cosm (B(X − cT )), (3.21)

Y (X − cT ) = cm A1 cosm ( B̃(X − cT )), (3.25)

3.4 Traveling Waves Without Precompression II:

∂t2 Φ(ξ, t) = −c2 ∂ξ2 Φ(ξ, t) + 2c∂ξ t Φ(ξ, t)+

Traveling waves of Eq. (1.3) correspond to stationary (i.e., time-independent) solu-

The hat  denotes the (wavenumber-dependent) Fourier transform of the spatial

where q = 1 + p ∈ (1, 2), and  is an appropriately chosen small parameter. It was

λ2 a(ξ) = −c2 ∂ξ2 a(ξ) + 2λc∂ξ a(ξ)+

1. Y. Starosvetsky, K. Jayaprakash, M.A. Hasan, A. Vakakis, Dynamics and Acoustics of Ordered

4.1 Discrete Breathers: Overview and Theoretical Analysis

© The Author(s) 2018 41

yn (t) = Y (X, T )ei(kn+ωt) + c.c., (4.2)

i∂T Y (X, T ) + ν2 ∂ X2 Y (X, T ) + ν3 Y (X, T )|Y (X, T )|2 = 0. (4.3)

Thus, the corresponding reconstruction of the approximate solution with k = π of

A 3/2 3/2 3/2

T (t)X n = ω02 (X n+1 + X n−1 − 2X n )T (t) ⇒ (1.11)

eiqn+iq + eiqn−iq − 2eiqn

The hat denotes the (wavenumber-dependent) Fourier transform of the spatial

where q = 1 + p ∈ (1, 2), and is an appropriately chosen small parameter. It was

qm,n (t) = φ P (T )E + c.c., (7.6)

Note that 4ω 2 = ω 2 (2k) for k = 0. Finally, moving on to the O(