Mechanical Metamaterials

Spring Semester 2018

Sebastian Huber
ETH Zürich
Email: sebastian.huber@phys.ethz.ch

Lecture Website:
http://www.cmt-qo.phys.ethz.ch

1
Learning goals
! You can formulate the principle of Bragg scattering.
! You can relate the dimension of a phononic crystal to the frequency of the band gap.
! You know the effect of the coupling of two near-identical oscillators.
! You know the wave equation.
! You know that concepts and differences between the group and phase velocity.
! You know the concept of a Brillouin zone.
! You know what an evanescent wave is.
! You know the relation of the number of elements in a unit cell and the number of bands.
! You know the effect of a local resonance for wave propagation.
! You know the concept of reciprocal space.
! You know how to find the first Brillouin zone.
! You can find the phonon spectrum of a d-dimensional discrete structure.
! You can explain why we map our designs to an elastic theory with effective parameters.
! You know what strain and stress are and how they are related.
! You know the Helmholtz decomposition.
! You know that doubly negative materials require some thoughts related to power flow.
! You know Snell’s generalized law in doubly negative materials.
! You know how to define effective parameters for a discrete system.
! You know how to find the effective mass for a simple setup.
! You know how to determine the effective spring constant.
! You can sketch the history of topological bandstructures.
! You know the Su-Schrieffer-Heeger model.
! You know what a Chern insulator is.
! You know why a Chern insulator cannot be built from passive mechanical components.
! You know how to do it anyway.

2
Contents

1 Bragg scattering vs. local resonances 6

1.1 Bragg scattering vs. local resonances . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.1 Bragg scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.2 Local resonances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Waves in solids 11
2.1 The wave equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 Traveling wave solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.2 Eigenmodes of the wave equation . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.3 The phase velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.4 The group velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.5 The wave number k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.1.6 A note on waves in solids . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3 Discrete one dimensional systems 15

3.1 Discrete systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 The monoatomic chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2.1 Evanescent waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2.2 The diatomic chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2.3 Locally resonant structures . . . . . . . . . . . . . . . . . . . . . . . . . . 20

4 Two and three dimensional discrete systems 23

4.1 Two dimensional discrete systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.2 A simple example: the square lattice . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2.1 Springs and pre-stress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.2.2 Flexural waves on the square lattice . . . . . . . . . . . . . . . . . . . . . 27
4.3 A honeycomb lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5 Effective negative parameters 31

5.1 Elasticity in one hour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.1.1 The strain tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.1.2 The stress tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.1.3 Hooke’s law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.1.4 The Poynting vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.2 The elastic wave equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.3 Negative E or ρ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.4 Doubly negative metamaterials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.4.1 Longitudinal waves in thin rods . . . . . . . . . . . . . . . . . . . . . . . . 34
5.4.2 The Veselago lens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.4.3 Superlensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3
 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

6 The continuum limit 41

6.1 Mass in mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.2 Effective spring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.3 Double negativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

7 Topological mechanical metamaterials 46

7.1 Mechanical waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
7.1.1 Surface acoustic waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
7.1.2 Total internal reflection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
7.1.3 Band gap wave guides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
7.2 A short history of topology in condensed matter physics . . . . . . . . . . . . . . 49
7.3 Simple examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.3.1 The Su-Schrieffer-Heeger model . . . . . . . . . . . . . . . . . . . . . . . . 51
7.3.1.1 Bulk topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7.3.2 Chern insulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
7.3.2.1 Edge states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.3.3 Application in mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
7.4 Time reversal invariant topological systems . . . . . . . . . . . . . . . . . . . . . 59
7.4.1 Doubling the degrees of freedom . . . . . . . . . . . . . . . . . . . . . . . 59
7.4.2 The doubled 13 -flux model . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
7.5 Gyroscope systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

4
List of Figures

1.1 Traveling waves. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.2 Band gap of a phononic crystal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Two coupled oscillators. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Local resonances. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.1 Waves in a thin rod. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2 Dispersion relation of a simple wave. . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.1 A discrete one dimensional chain. . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.2 Dispersion relation of a monoatomic chain. . . . . . . . . . . . . . . . . . . . . . 17
3.3 Diatomic chain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Dispersion of a diatomic chain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.5 Resonantor-in-resonator setup. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.6 Dispersion for a chain made of local resonators. . . . . . . . . . . . . . . . . . . . 22

4.1 Two dimensional lattice. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.2 Dispersion of the square lattice. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.3 Beaming. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.4 Honeycomb lattice. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.5 Dispersion of a honeycomb lattice. . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5.1 Surface element. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

5.2 Doubly negative rod. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.3 Veselago boundary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.4 Snell’s law. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.5 A Veselago lens. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

6.1 Effective coordinates of a more complicated body. . . . . . . . . . . . . . . . . . . 41

6.2 Mass in mass system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.3 Effective mass. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.4 Effective sping. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.5 Effective spring constant. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

7.1 Spectrum for waveguide plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5
Chapter 1

Bragg scattering vs. local resonances

Learning goals
• You can formulate the principle of Bragg scattering.
• You can relate the dimension of a phononic crystal to the frequency of a potential band gap.
• You know the effect of the coupling of two near-identical oscillators.

1.1 Bragg scattering vs. local resonances

We start our lecture on mechanical metamaterials with the discussion of metamaterials with a
target functionality at finite frequencies. In other words, we want to understand how we can
control or manipulate the flow of mechanical energy in the form of vibrations, or equivalently
waves.
To understand how vibrations propagate through materials we will have to introduce the wave
equation and familiarize ourselves with its properties. However, before we do so, we introduce
two generic building blocks that are used to achieve the goal of manipulating wave propagation
in solids: Bragg scattering and the use of locally resonant structures. Once we know how these
two principles work, we will understand most modern metamaterial designs.

1.1.1 Bragg scattering

For now, all we need to know about waves is that they can be described by a traveling modulation
of some property f (x, t) such as

f (x, t) = cos(ωt − kx + ϕ), (1.1)

where ω = 2π/T is the angular frequency related to the period T , k = 2π/λ is the wave number
related to the wave-length λ, and ϕ is a phase.

!
!t kx = 0 ) v = T = k

Figure 1.1: Traveling waves.

6
One of the most important aspects of waves is their ability to interfere. In particular:

⇒ destructive interference for ∆ϕ = π.

⇒ constructive interference for ∆ϕ = 0.

Let us now consider a wave incident on a scatterer:

1.) Wave is incident on scatterer 2.) Scatterer is excited

3.) Scatterer radiates waves

The “time lag”, i.e., the phase difference ∆ϕ, between the incident and the radiated wave is
given by the details of the interaction between the wave and the scatterer and the properties of
the excitation of the scatterer. In other words, we have nothing too generic to say about this
event. However, consider the effect of two identical scatterers at a distance ∆x:

cos(kx !t + '0 )

cos(kx !t)

cos(kx !t + '0 + k x)
x
We see that for

k∆x = π : the scattered waves interfere destructively.

λ
⇒ ∆x = : leads to a strong destructive effect on the propagation of waves.
2
This phenomena is called “Bragg scattering” if we deal with periodic arrays of scatterers. Bragg
scattering is at the heart of phononic crystals. It is not hard to imagine that periodic arrays
with a typical length-scale ∆L can have dramatic effects on the wave propagation at frequencies
where the Bragg condition
λ
∆L = (1.2)
2
is fulfilled.

7
 !(k) = ck k = ⇡/ L

Frequency !

Frequency !
Wave number k Wave number k

Figure 1.2: Band gap of a phononic crystal.

If I want to change the propagation of waves at a frequency ν = ω/2π, I have to design

structures with dimension ∆L ≈ c/ν.

In particular, at the frequencies where the Bragg condition is met, a window opens where no
waves can propagate. We call this a band gap, see Fig. 1.2.
Question: I live in California and want to protect my house from earthquakes with a phononic
crystal. How big is it going to be?

1.1.2 Local resonances

We have seen that periodic structures can have an influence on the propagation of waves. Here,
we want to present a building block to achieve a similar effect, however, without the limit
imposed by the size of the structure. In order to understand the concept of local resonances, we
need to understand the coupling of two oscillators.
Let us assume the following system illustrated in Fig. 1.3:
The oscillators x(t) and y(t) are described by

ẍ(t) = −ω02 x(t) + γ 2 y(t), (1.3)

ÿ(t) = −ω02 y(t) + γ x(t),2
(1.4)

where we want to assume that γ ≪ ω0 . For γ = 0, we know that

x(t) = x0 eiω0 t and y(t) = y0 eiω0 t , (1.5)

solve the equations (1.3) and (1.4). If the two oscillators are coupled (γ ̸= 0), we have to find
combined solutions
! " ! "
x(t) x0 iωt
= e ⇒ (1.6)
y(t) y0
! " ! 2 "! "
2 x0 −ω0 γ 2 x0
−ω = . (1.7)
y0 γ 2 −ω02 y0

In other words, we are dealing with a eigenvalue problem for ω. We have two routes to solve
this problem.
Route 1: We solve the characteristic equation:
! "
λ − ω02 γ2
det = (λ − ω02 )2 − γ 4 = 0 (1.8)
γ2 λ − ω02
#
2 2
⇒ λ± = ω 0 ± γ ⇒ ω± = ω02 ± γ 2 . (1.9)

8
 x(t) y(t)

Figure 1.3: Two coupled oscillators.

Route 2: Basically the same, but once and for all:

! 2 " ! "
−ω0 γ 2 2 2 0 1
= −ω0 1 + γ (1.10)
γ 2 −ω02 1 0
3
$
= −ω02 1 + d i σi with (1.11)
i=1
! " ! " ! "
0 1 0 −i 1 0
d = (γ 2 , 0, 0) and σ1 = , σ2 = , σ3 = . (1.12)
1 0 i 0 0 −1

From this we immediately read

λ± = −ω02 ± |d| = −ω02 ± γ 2 . (1.13)

We can now easily generalize that to two coupled oscillators that have different frequencies:
! 2 " % &
−ωx γ 2 ωx2 + ωy2 ω 2 − ω2
x y ω 2 − ω2
x y
=− 1 + γ 2 σ1 − σ3 ⇒ d = γ 2 , 0, − . (1.14)
γ 2 −ωy2 2 2 2

And the new eigenfrequencies are

' *! +2
(!
( ω 2 + ω 2 "2 ω 2 − ω 2 "2
ω± = )
x y x y
± + γ2 . (1.15)
2 2

As illustrated in Fig. 1.4(a), the effect of a coupling between two oscillators is strongest if they
are degenerate, i.e., have the same frequency. Moreover, the main effect of γ is to split the
frequency of the two-oscillator system. In Fig. 1.4(b) & (c) we see what effect we can expect
from a local resonance onto the dispersion of waves: The coupling will split the degenerate
system of the wave and the local oscillator. This effect is strongest where the two frequencies
cross. Again, like in the Bragg scattering effect, a frequency window is opening where no waves
propagate. However, this time the frequency is not dictated by the spacing of the periodic array
but by the frequency of the local oscillator!

We can modify the propagation of wave in the vicinity of a frequency ν by coupling the wave
to a local resonance with frequency ν0 ≈ ν.

Question: Why is the argument that we freed ourselves from length-scale constraints potentially
a lie?

9
 a) b) c)
non-degenerate

Frequency !±

Frequency !

Frequency !
!0 !0

degenerate

Coupling Wavenumber k Wavenumber k

Figure 1.4: (a) Effect of the coupling γ on the eigenfrequency of two oscillators. (b) Wave
dispersion (black) and the location of a local resonance (blue). (c) Coupling the local resonance
to the wave leads to the opening of a band gap at the frequency ω0 of the local resonance.

References
1. Ashcroft, N. W. & Mermin, N. D. Solid State Physics (Harcourt, Orlando, 1987).
2. Landau, L. D. & Lifshitz, E. M. Theory of Elasticity (Butterworth-Heinemann, London,
1986).
3. Fetter, A. L. & Walecka, J. D. Theoretical Mechanics of Particles and Continua (Dover,
1980).
4. Meyers, M. A. Dynamic Behavior of Materials (Wiley, 1994).
5. Acoustic Metamaterials and Phononic Crystals (ed Deymier, P. A.) (Springer, 2013).
6. Acoustic Metamaterials (eds Craster, R. V. & Guenneau, S.) (Springer, 2013).

10
Chapter 2

Waves in solids

Learning goals
• You know the wave equation.
• You know the concept and differences between the group and the phase velocity.

2.1 The wave equation

To familiarize ourselves with waves and their propagation, we study a simple example of longi-
tudinal waves in thin rods. Let us consider the forces acting on a small section of the rod as
shown in Fig. 2.1. The net force on a small segment is given by the difference of the force acting
on the left and right cross-section
F = −Aσ(x) + Aσ(x + δx) (2.1)
′ ′
= −A[σ(x) − σ(x) + δxσ (x)] = Aδxσ (x), (2.2)
where σ(x) is the stress field, or in other words, the force per area A, i.e., the pressure. δx is the
infinitesimal thickness of the segment and the prime on σ indicates the derivative with respect
to x.
We know that Newton’s equations of motion govern the behavior of mechanical systems. Hence
ma = F, (2.3)
which for our cross-section means
ma = Aδxρ ü(x, t), (2.4)
, -. / , -. /
m a
where u(x, t) is the displacement field and ρ the density. We therefore find
Aδρü(x, t) = Aδxσ ′ (x, t). (2.5)
In order to close this equation, we need a relation between the stress σ(x, t) and the displacement
u(x, t). As all segments of a finite rod can take up some of the applied force, the change in length
of an elastic object depends on its size:
∆L
σ=E or σ(x, t) = Eu′ (x, t) (2.6)
L
with a proportionality factor E called Young’s modulus (this is nothing but the “spring constant”
k for a continuous system. Note that u′ or ∆L/L is unitless, so E takes the units of σ, i.e.,
pressure.). Inserting this relation (2.5) into (2.6) we obtain
E ′′
ü(x, t) = u (x, t). (2.7)
ρ
This is our sought after wave equation. Let us analyze a few properties of this equation.

11
 Figure 2.1: Waves in a thin rod.

2.1.1 Traveling wave solutions

Any (twice differentiable) function
0
E
u(x − vt) with v= (2.8)
ρ
is a solution of (2.7). To see this, let us insert this ansatz into the wave equation
ρ 2 ′′
v u (x − vt) = u′′ (x − vt). (2.9)
E
1
Therefore, for v = E/ρ, u(x−vt) is indeed as solution. We read off a few interesting properties
form this solution
1. Any traveling wave-form is preserved under the evolution of time.
2. The stiffer the material (E larger), the faster the wave.
3. The lighter the material (ρ smaller), the faster the wave.
While these observations are important and useful for the design of metamaterials, we profit
from another analysis in terms of “modes”.

2.1.2 Eigenmodes of the wave equation

An eigenmode is a natural vibration of the system where all parts oscillate at the same frequency.
It is useful t know these modes, as we can construct all solutions from a super-position of such
eigenmodes.
For the wave equation these are simply traveling waves
u(x, t) = eiϕ(x,t) = ei(kx−ωt) . (2.10)
Inserted into (2.7) we find
− c2 k 2 ei(kx−ωt) = −ω 2 ei(jx−ωt) , (2.11)
which means that we have a relation between the wave number k and the angular frequency ω
given by
0
E
ω(k) = c|k| with c= . (2.12)
ρ

12
 Frequency !
!(k) = c|k|

Wave number k

Figure 2.2: Dispersion relation of a simple wave.

This connection between k and ω is called the dispersion relation, see Fig. 2.2. Most of what we
are going to do in this lecture is trying to manipulate ω(k) to achieve our design goals. Let us
therefore understand a few more properties of the dispersion relation.

2.1.3 The phase velocity

We have seen in the first lecture that the phase of a wave is a very important property. The
point in space of constant phase ϕ0 is moving in time

ω(k)
0 = ϕ0 (x, t) = kx − ω(k)t ⇒ x0 = t, (2.13)
k /
, -.
vϕ

where we defined the phase velocity vϕ .

The phase velocity will become particularly relevant for two and three-dimensional systems,
where a non-isotropic ω(k) (i.e., ω(k) ̸= c|k|) can lead to distortions of the wave front.

2.1.4 The group velocity

As we are often dealing with wave-packets (like the one depicted in the beginning where we
wrote u(x − vt)), the phase velocity is not the only important quantity. Imagine a wave-packet
made from modes that are close to some base frequency ω:
∆ω ∆k ∆ω ∆k
u(x, t) = ei(ω+ 2
)t
ei(k+ 2 )x + ei(ω− 2 )t ei(k− 2
)x
(2.14)
! "
ikx−iωt ∆k ∆ω
= 2e cos x− t . (2.15)
2 2

2⇡/ k

2⇡/k

The envelope cos(∆kx/2 − ∆ωt/2) travels with the velocity vg = ∆ω/∆k. If we take the limit
of ∆k → 0, we find the group velocity

∂ω(k)
vg = . (2.16)
∂k
In the exercises we will see when vg = vϕ and what the significance of the group velocity is!

13
2.1.5 The wave number k
We have seen that exp[ikx − ω(k)t] are solutions to the wave equation (2.7). Note that the wave
number k encodes the wavelength λ = 2π/k. Moreover, k controls how the phase of the wave
changes when we move in space:
2 3
arg[u(x, t0 )] = arg eikx−ω(k)t = kx − ω(k)t. (2.17)
arg[u(x + ∆x, t0 )] = k(x + ∆x) − ω(k)t, (2.18)

and therefore

∆ϕ = k∆x. (2.19)

And this is the only thing that changes in uk (x, t) when advancing by ∆x. On solutions char-
acterized by “k” we can think that translations act by multiplying by:

T∆x = ei∆xk . (2.20)

As we will see in the next chapter, this simple property might be lost if our medium is not
translational invariant. And this is what we typically do in a metamaterial design: we structure
a material in a way that breaks continuous translation-symmetry, e.g., by drilling holes.

2.1.6 A note on waves in solids

In the above example we studied one simple type of elastic waves that occur in solids. Of course,
there are many more types of waves like shear-waves, torsional waves, flexural waves, waves that
propagate on surfaces like Love or Rayleigh waves. For the purpose of this lecture, however, we
focus much more on wave in discrete systems, as for our metamaterial concepts, they are more
instructive.

References
1. Landau, L. D. & Lifshitz, E. M. Theory of Elasticity (Butterworth-Heinemann, London,
1986).
2. Fetter, A. L. & Walecka, J. D. Theoretical Mechanics of Particles and Continua (Dover,
1980).

14
Chapter 3

Discrete one dimensional systems

Learning goals
• You know the concept of a Brillouin zone.
• You know what an evanescent wave is.
• You know the relation between the number of elements in a unit cell and the number of
bands.
• You know the effect of a local resonance for wave propagation.

3.1 Discrete systems

So far, we have been considering the wave equation,
ρ 2
∂x2 u = ∂ u, (3.1)
E t
in a homogeneous and isotropic medium. As a first step towards understanding (periodically)
structured metamaterials we study discrete systems. To prepare ourselves for this step, we first
look at the translation “operator” T∆x that we encountered in the last section. So far, our
system was “living” on the real axis

u(x, t) with x ∈ R. (3.2)

We found that an eigenmode labelled by the wave-number k has the property that

u(x, t) = eik−iωt ⇒ u(x + ∆x, t) = eik∆x u(x, t). (3.3)

Let us now consider the discrete system depicted in Fig. 3.1. Each local discrete element is

m1 m2 m3 m4 m5
f1,2 f2,3 f3,4 f4,5

u1 (t) u2 (t) u3 (t) u4 (t) u5 (t)

a

Figure 3.1: A discrete one dimensional chain.

15
considered rigid, having a mass mi and it is connected to its neighbors via springs with spring
constants fn,n+1 and fn,n−1 . Each element, or mass, can be displaced by un (t). But now, the
displacement field un (t) is actually living on the set of integer numbers n ∈ Z and not R as we
have only a discrete number of displacements, one for each element located at xn = na, where
a is the lattice constant. (Of course, u itself can still take values in R, or C, respectively).
This has immediate consequences for the possible wave-numbers that can appear. Assume again
that plane waves are good solutions:
un (t) = eikan−iωt . (3.4)
Therefore, we again have that the solution at n and at n + 1 differ by
eika . (3.5)
But now, kna = 2π is really the same as kna = 0. Actually, all k = k + 2πm with m ∈ Z are
equivalent. Hence, we can confine the possible values of the wave-number to
2 π π3
k∈ − , , (3.6)
a a
which is known as the first Brillouin zone. Equivalently, k > 2π
a would correspond to a wave with
2π
wavelength λ < k = a. But as we have a mass only every a, there is nothing to be described
with a wavelength smaller than the lattice spacing a.

3.2 The monoatomic chain

Let us now solve the problem sketched in Fig. 3.1. In a first step, we assume all masses to be
the same mi = m and all springs to be identical with fi,j = f . The equations of motion are
then given by
mün (t) = f {[un−1 (t) − un (t)] − [un (t) − un+1 (t)]} (3.7)
When assuming solutions of the form
un (t) = eikna−iωt (3.8)
we find that the equation of motion reduces to
2 3
−mω 2 einak−iωt = f einak−iωt e−ika − 1 − 1 + eika (3.9)
! "
2 ka
= −4f sin einak−iωt . (3.10)
2
With this we find the dispersion relation shown in Fig. 3.2
4 5 ! "5
4f 5
5 ka 5 5
ω(k) = 5sin . (3.11)
m 2 5

Again plane waves are good solutions! However, our dispersion relation changed. A few impor-
tant observations
1. The dispersion ω(k) is periodic in k with period 2π/a.
2. Around k = 0, the waves seem to still linear disperse with
4
f
c=a . (3.12)
m
Where the spring constant f [N/m] replaced the Young’s modulus E [N/m2 ] and the mass
m [kg] the mass density ρ [kg/m]. The fact that around k = 0 we recover the result
for a continuous system is easily explained. At wavelengths much larger than the lattice
spacing, the waves don’t feel the granularity of the individual masses.

16
 homogeneous material

Frequency !
discrete chain

⇡/a ⇡/a
Wave number k

Figure 3.2: Dispersion relation of a monoatomic chain.

3. Owing to the non-linearity of ω(k), group and phase velocity are not identical anymore.

There is another interesting property: The spectrum ω(k) of possible traveling waves is now
bounded form above. That means, for frequencies
4
4f
ω > ωedge = ω(π/a) = , (3.13)
m
there are no propagating solutions.

3.2.1 Evanescent waves

The fact that there are no propagating solutions does not mean that the monoatomic chain is
totally inert if we try to excite it with frequencies ω > ωedge . Let us assume solutions
! "
i π+i 1ξ na−ωt
uξn (t) =e . (3.14)

Inserted into (3.7), we find

6 ! " !7 "
i π+i 1ξ −i π+i 1ξ
−mω 2 uξn (t) = f uξn (t)eiωt e +e −2 (3.15)
2 3 6 ! " 7
−1 1 1
−mω 2 = f −e ξ − e ξ − 2 = −f 2 cosh +2 . (3.16)
ξ

And therefore '

( 8 9
4 ( cosh 1 + 1
4f ) ξ
ω(ξ) = . (3.17)
m 2
, -. /
≥1

This is a solution only for ω > ωedge . We found that in this case, vibrations penetrate into the
monoatomic chain with a decay length ξ given by
1
ξ=4 8 9. (3.18)
mω 2
arccosh 2f −1

We summarize these findings in Fig. 3.3

17
 ⇡/a
Frequency !
⇡/a

p
4f /m

Wave number k

Inverse localization length 1/⇠

#
Figure 3.3: Summary of the monoatomic chain: Up to a frequency ω = 4f m waves propagate
with a real wavenumber k. Above this cut-off frequency, evanescent waves with a decay length
ξ can survive at the endpoints of the chain.

3.2.2 The diatomic chain

We are now in the position to embark on the first interesting example. Let us consider a system
similar to the one above, but with alternating masses:

m1 m2 m1 m2 m1
f f f f

What of the analysis of the last section can we carry over to this problem? Remember that we
knew ukn (t) if we had ukm (t) just by multiplying
ukn (t) = eik(m−n)a ukm (t). (3.19)
This simple phase relation was due to the fact that our problem was symmetric under a shift
by one lattice site. Now, that m1 ̸= m2 , we have no reason to assume that a simple relation
as in (3.19) should hold. However, every second mass is identical! Therefore, let us assume the
following structure for the eigenmodes
:
k ikna−iωt ak n = 2s, s ∈ Z,
un (t) = e × (3.20)
bk n = 2s + 1, s ∈ Z.
In other words, we assume that from even to even and from odd to odd site we again have a
simple phase factor
eik2a , (3.21)
but that the structure within one unit cell is given by the relation between ak and bk . Note,
that by virtue of this ansatz we have that
e2ika = e2i(k+rπ)a with r ∈ Z. (3.22)

18
By doubling (a → 2a) our unit cell size, we halved the sized of the first Brillouin zone.1
The equations of motion read

m1 ü2n = f (u2n+1 + u2n−1 − 2u2n ), (3.23)

m2 ü2n+1 = f (u2n + u2n+2 − 2u2n+1 ). (3.24)

Inserting ansatz (3.20) into the above equation we find

−m1 ω 2 ak = f (bk + e−2iak bk − 2ak ), (3.25)

2 2iak
−m2 ω bk = f (ak + e ak − 2bk ). (3.26)

We deal with a problem of the form

! "! " ! ; <" ! "
2 m1 ak ; 2f < −f 1 + e−2iak ak
ω = 2iak (3.27)
m2 bk −f 1 + e 2f bk

This is a generalized eigenvalue problem of the type αBx = Ax with B positive definite and
both A = A† and B = B † . We could do

αx = B −1 Ax, (3.28)

but we would not deal with a hermitian problem anymore. Let us write B = LL† with
!√ "
m1
L= √ . (3.29)
m2

With this we obtain

αLL† x = A(L† )−1 L† x (3.30)

L† x = L−1 A(L† )−1 ,-./
α ,-./ L† x . (3.31)
, -. /
y y
Ã

Therefore we deal with a regular eigenvalue problem

αy = Ãy, (3.32)

with the property

Æ = [(L† )−1 ]† A† (L−1 )† = L−1 A(L† )−1 = Ã, (3.33)
i.e., a hermitian eigenvalue problem. For us this means
% 1 &! ; <" % √ 1 &
√ 2f −f 1 + e −2iak
m1 ; < m1
à = √1 √1
= (3.34)
m2 −f 1 + e2iak 2f m2
% 2f ; <&
− √ f 1 + e −2iak
m1 m1 m2
= ; < (3.35)
− √mf1 m2 1 + e2iak 2f
m2
! " ! "
1 1 f f 1 1
=f + 1 − (1 + cos(ϕk )) σ1 − sin(ϕk ) σ2 + f − σ3 (3.36)
m1 m2 ,m -. / ,m -. / m1 m2
, -. /
d1 (k) d2 (k) d3 (k)
! " 3
$
1 1
=f + 1+ di (k)σi . (3.37)
m1 m2
i=1
1
You could argue that now there is a life below the wavelength λ = 2a as there is “something” at a distance a
rather than 2a. But this degree of freedom is taken care of by the relation between ak and bk !

19
 √
We introduced m = m1 m2 and ϕk = 2ak. We can now capitalize on our knowledge of 2 × 2
matrices written with Pauli matrices to immediately read off the eigenvalues
! " 0! "
1 1 f m m 2
α± = f + ± − + (1 + cos(ϕk ))2 + sin(ϕk )2 (3.38)
m1 m2 m m1 m2
!4 " 0!4 "2
4 4
f m2 m1 f m2 m1
= + ± − + 2 + 2 cos(2ka). (3.39)
m m1 m2 m m1 m2

With this, we arrive at the dispersion

'
4 (! " 0! "
f() 1 1 2
ω± (k) = r+ ± r− + 4 cos2 (ka), (3.40)
m r r

√ 1
with m = m1 m2 and r = m2 /m1 . Let us inspect how this dispersion behaves for various
π
values of the reduced mass m and ratio r. We see that we opened a band gap for k = 2a with
# #
magnitude 4f m r − 1r , which goes to zero for r = 1. In fact, for r = 1, we chose a unit cell
which is too large, or equivalently, a too small Brillouin zone. We see in the Fig. 3.4. that this
leads to a “folding” of the spectrum.

q q
Frequency !

r=1 4f 1
m r r

r = 0.6

Wave number k

Figure 3.4: Dispersion of a diatomic chain.

We have now seen that by structuring our chain at the scale 2a, we opened a band gap at
k = π/2a, and the frequency was hence dictated by ωk=π/a of the undisturbed (mono-atomic)
chain. We would like to see, if another design-principle exists that can beat this limit.

3.2.3 Locally resonant structures

We want to study the system shown in Fig. 3.5. The block system of masses m and spring
constants f alone defines a regular monoatomic chain with no band gap. We want to understand
the effect of an added local resonator:

müi = f (ui−1 + ui+1 − 2ui ) + F (ui − wi ), (3.41)

M ẅi = −F (wi − ui ). (3.42)

As usual, we assume plane waves and write the above equation in matrix form
! " ! " ! "! "
m 2 uk −(2f + F ) + 2f cos(ka) F uk
− ω = , (3.43)
M wk F −F wk

20
 m m m m m
f FM f FM f FM f FM f FM f

Figure 3.5: Resonantor-in-resonator setup.

where we assumed

un = uk eikna−iωt , (3.44)
ikna−iωt
wn = wk e . (3.45)

We again transform A → Ã wiht LL† = B to obtain

% (2f +F )−2f cos(ka) & !1 "
1 2 2 2
−1 † −1 m − √mM F 2 2 (1 − cos(k)) + γ ω̃L −γ ω̃L
à = L A(L ) = = ω0 ,
− √mM1
F F
M
−γ ω̃L2 ω̃L2

where we introduced
0
1 mF
ω̃L = : local resonance frequency in units of ω0 , (3.46)
2 f M
4
M
γ= : “coupling strength”, (3.47)
m
4
4f
ω0 = : overall frequency scale. (3.48)
m
With these abbreviations we can write
= > 3
$
1 1 1 2 2
2 Ã = [1 − cos(ka)] + ω̃L (γ + 1) 1 + di (k)σi , (3.49)
ω0 2 2
i=1

with ; ? @<
d(k) = −γ 2 ω̃L2 , 0, 12 12 [1 − cos(ka)] + ω̃L2 (γ 2 − 1) . (3.50)
And again, we just read off the dispersion relation
' ⎧ ⎫
( 0 6 72 ⎬
( ⎨
(1 1 1
ω± (k) = ω0 ) [1 − cos(ka)] + ω̃L2 (1 + γ 2 ) ± 4(γ ω̃L2 )2 + [1 − cos(ka)] + ω̃L2 (γ 2 − 1)
2 ⎩2 2 ⎭

The dispersion is displayed in Fig. 3.6. We see that for the case of a local resonance a band
gap is opening not at a specific k but at a specific ω̃L . Moreover, the stronger the resonance is
coupled, the bigger the band gap.

21
 !
˜ L = 0.1 !
˜ L = 0.3 !
˜ L = 0.5
frequency !

= 0.4
frequency !

= 0.8
frequency !

= 1.2

wavenumber k wavenumber k wavenumber k

Figure 3.6: Dispersion for a chain made of local resonators.

22
Chapter 4

Two and three dimensional discrete

systems

Learning goals
• You know the concept of reciprocal space.
• You know how to find the first Brillouin zone.
• You can find the phonon spectrum of a d-dimensional discrete structure.

4.1 Two dimensional discrete systems

Up to now, we have seen systems with interesting band structures in one dimension. We saw
how in a system with periodicity a, the good wave numbers k are confined to the interval
k ∈ [−π/a, π/a], the first Brillouin zone. This arose as waves with a wavelength λ = 2π/k < a
don’t encode sensible information.
How do these considerations generalize to higher dimensions? Let us start with the simple,
non-trivial, two-dimensional lattice in Fig. 4.1. Every point on the lattice is described by two
coordinates (ξ1 , ξ2 ) with respect to the basis vectors d1 and d2

rij = ξi d1 + ξj d2 . (4.1)

Let us now consider a plane wave propagating through that lattice (for now, we assume a scalar
wave, which might describe the pressure pij or a displacement field u on only one direction at
the location rij ).
ψ = e−ik·r+iωt . (4.2)
Clearly, the phase k · r − ωt is constant in space along directions perpendicular to k, where
k · r⊥ = 0.
Moreover, the wavelength is given by λ = 2π/|k|. Again, we only evaluate ψ at discrete lattice
points! Let us make this explicit:

k · rij = (k · d1 )ξi + (k · d2 )ξj = k1 ξi + k2 ξj . (4.3)

Therefore, we may write

ψ = eiωt−k1 ξi −k2 ξj . (4.4)
As ξi denote the integer coordinates with respect to the basis {di }, we can again add multiples
of 2π to the ki ’s
ki′ = ki + 2πmi with mi ∈ Z (4.5)

23
 (0,3)

(1,3)

(2,3)

(3,3)
(0,2)

(1,2)

(2,2)

(3,2)
(0,1)

(1,1)

(2,1)

(3,1)
d2
(1,0)

(2,0)

(3,0)
d1

Figure 4.1: Two dimensional lattice.

without changing the value of ψ at the discrete lattice points. Let us see how we changed the
original k by adding the mi ’s. We introduce the reciprocal lattice vectors
:
i = 1, 2
bi · dj = 2πδij (4.6)
j = 1, 2

and δij = 1 if i = j and zero otherwise. So b1 is perpendicular to d2 and has length 2π/|d1 |
and b2 is perpendicular to d1 and has length 2π/|d2 |.

d2
b2 /(2⇡)
d1

b1 /(2⇡)

The claim is now that we changed

k′ = k + m1 b1 + m2 b2 . (4.7)

24
This is easy to see as

k1′ = k′ · d1 = k · d1 + m1 b1 · d1 +m2 b2 · d1 (4.8)

, -. / , -. /
2π 0
= k1 + 2πm1 , (4.9)
k2′ ′
= k · d1 = k · d2 + m1 b1 · d2 +m2 b2 · d2 (4.10)
, -. / , -. /
0 2π
= k2 + 2πm2 , (4.11)

which proves (4.7). From (4.7) we learn that arbitrary additions of integer multiples of either
b1 or b2 do not change ψ evaluated on the lattice defined by d1 and d2 .
In one dimension it was easy to construct the first Brillouin zone by choosing [−π/a, π/a]. How-
ever, any interval [k0 , k0 + 2π/a] would have been possible. We only chose the most symmetric
around k = 0 for convenience. In two dimensions we follow the same principle: We define the
first Brillouin zone to be the area in reciprocal space where points are closer to the origin than
to any other reciprocal lattice vector m1 b1 + m2 b2 with mi ∈ Z.

4.2 A simple example: the square lattice

We consider a concrete example, the square lattice:
:
d1 = (a, 0)
rij = ξi d1 + ξj d2 with and ξi/j ∈ Z. (4.12)
d2 = (0, a)

The reciprocal lattice is trivially also a square lattice

! "
2π
b1 = ,0 , (4.13)
a
! "
2π
b2 = 0, , (4.14)
a

and the first Brillouin zone is simply given by

[−π/a, π/a] × [−π/a, π/a]. (4.15)

We also introduce the standard notations for special points in the Brillouin zone

M = (⇡/a, ⇡/a)

X = (⇡/a, 0)
= (0, 0)

As three-dimensional plots are typically hard to read, we often show the dispersion curve ωα (k)
(α labels the different bands) along the high-symmetry line Γ − X − M − Γ shown in red.
We now consider a concrete model:

25
 f
m m m f

m m m
a
m m m

For now, we keep the displacements uij three dimensional. In other words, we think of a
planar sheet that can be deformed in three dimensions. Before we go on, we quickly discuss the
energetics of springs.

4.2.1 Springs and pre-stress

Let us consider two points in space connected by a spring

uy2

ux2
uy1 l0
ay f

ux1

ax

where l0 is the rest length of the spring#

and ax and ay are the rest separation in x and y direction.
Moreover, we write ∆ = l0 − l = l0 − a2x + a2y . The energy of the spring is given by1
6# 72
f
E= (ux1 − ux2 − ax )2 + (uy1 − uy2 − ay )2 − l0 . (4.16)
2

The force on uαi is now given by

∂E
Fiα = − . (4.17)
∂uαi
We further linearize this around uαi = 0. From these considerations we obtain
% &
f ax ∆ a2x a2y ∆ f ax ay (l + ∆) y
x
−F2 = F1 =x
2
+f 2
− 3 (ux1 − ux2 ) + (u1 − uy2 ), (4.18)
l l l l3
% &
y y f ay ∆ a2y a2x ∆ f ax ay (l + ∆) x
−F2 = F1 = +f − 3 (uy1 − uy2 ) + (u1 − ux2 ). (4.19)
l2 l2 l l3

We learn that
1
This generalizes trivially to three dimensions.

26
 • If the spring is along x, i.e., ay = 0, ⇒ ux and uy are decoupled.

• If the spring is along x and free of pre-stress, i.e., ∆ = 0, the uy -degrees of freedom are
not coupled.
Note that for our square lattice, this means that flexural waves where uzij ̸= 0 decouple from
the in-plane waves. Moreover, they require the springs to be under tension! To simplify our
discussion, we therefore discuss such out-of-plane waves in the following. Complications arising
from the vectorial nature of elastic waves are not of interest to us for now.

4.2.2 Flexural waves on the square lattice

Using the above analysis, we set ax = ay = a, az = 0 and hence the out-of-plane motions are
coupled with strength
feff = f [1 − (a2 ∆/l3 )], (4.20)
which we will simply denote by f in the following. With this we arrive at the equations of
motion ψij = uzij
mψ̈i,j = −f [4ψi,j − ψi,j+1 − ψi,j−1 − ψi+1,j − ψi−1,j ]. (4.21)
In the usual way, we introduce
k
ψi,j = e−ik·(ξ1 d1 +ξ2 d2 )+iωt (4.22)

and we find 0 = >

4f 1 1
ω= [1 − cos(akx )] + [1 − cos(aky )] . (4.23)
m 2 2

The dispersion is shown in Fig. 4.2.

a) b) c)
⇡/a ⇡/a
Frequency

ky

ky

⇡/a ⇡/a
M X M ⇡/a kx ⇡/a ⇡/a kx ⇡/a
Momentum

Figure 4.2: (a) Dispersion along high symmetry lines. (b) Contour plot of the dispersion. (c)
Map of the group velocity.

Let us analyze this dispersion a bit further. The group velocity

! "
1 sin(kx )
vg = ∇ω(k) ∝ (4.24)
2ω(k) sin(ky )
has a very interesting behavior shown in Fig. 4.2c. We see that along the red contour line in the
Brillouin zone, the group velocity points only in four distinct directions! In other words, waves
can transport only along the diagonals of the square lattice. This leads to the phenomenon
called beaming. To illustrate this, we numerically simulate Eq. (4.21) with a harmonic driving
force at the origin

mψ0,0 = −f [4ψ0,0 − ψ0,1 − ψ1,0 − ψ0,−1 − ψ0,1 ] + F cos(ωt). (4.25)

27
 p p
! = 0.2 4f /m != 4f /m

1 1
Figure 4.3: Time evolution for a drive at the origin for (a) ω = 4f /m and (b) ω = 0.2 4f /m.

To reach a steady state, we also add a small damping 1 term −γ ψ̇i,j with ωγ ≪ f . The resulting
|ψi,j (t ≫ 1)|2 is shown in Fig. 4.3. We see that at ω = 4f /m, i.e., where the isofrequency con-
tour is a square, the vibrations can only propagate in four directions. In the left graph, we excite
at much lower frequencies, where the lattice effects are weak and waves essentially propagate
isotropically. Note, however, that this beaming effect is extremely narrow in frequency!

4.3 A honeycomb lattice

Let us look at another structure which offers two new insights. First, it has a slightly more
complicated geometry and thus a more interesting Brillouin zone. Second, we deal with a lattice
with more than one degree of freedom in the unit-cell, and hence more than one band.
We introduce the honeycomb lattice, cf. Fig. 4.4:

d1 = a(1, 0), (4.26)

√
d2 = a(1/2, 3/2), (4.27)
√
τ = a(0, 1/ 3). (4.28)

The lattice of gray unit cells is spanned by d1 and d2 . However, there are two inequivalent sites
A and B. The coordinates of the A sites are given by

rA
ij = ξ1 d1 + ξ2 d2 with ξi ∈ Z. (4.29)

The coordinates of the B sites on the other hand

rB
ij = ξ1 d1 + ξ2 d2 + τ . (4.30)

28
 A A A
b2
B B B
( 1, 1) (0, 1)
K
A A A
M
B B
(0, 0) d2
d1 K0
A A

B B B
b1
A A A

B B B

Figure 4.4: Honeycomb lattice. Left: Real space lattice. Right: First Brillouin zone with high
symmetry points.

We first construct the Brillouin zone. The two reciprocal vectors are given by
2π
b1 · d1 = and b1 · d2 = 0 ⇒ (4.31)
a ! "
2π 1
b1 = 1, − √ (4.32)
a 3
2π
b2 · d1 = 0 and b2 · d2 = ⇒ (4.33)
! " a
2π 2
b2 = 0, √ . (4.34)
a 3
In other words, the first Brillouin zone is a hexagon! We derive the equations of motion in
A/B
reciprocal space by assuming that the (out-of-plane) displacement fields ψi,j are given by

A/B A/B −ik·ri,j −iωt

ψi,j = ψk e (4.35)

A/B
(again, we do not add the τ to the exponent and let ψk take care of the relative phase between
site A and B).
To construct the equations of motion we use the drawing again.

• We label all unit cells with their coordinates (ξ1 , ξ2 ).

• We realize that springs only connect A with B sites.

• For every spring in and out of the unit cell we need a term in the dynamical matrix h(k)
! A" ! " ! A"
2 ψk h(k) ψk
−ω = . (4.36)
ψkB h∗ (k) ψkB
, -. /
A(k)

• If the spring lies within the unit cell, we add simply f to the block h(k) connecting A and
B.

29
 • For all springs between different unit cells we add

f eik·∆ , (4.37)

where ∆ is the vector connecting the respective unit cell. Remember how we got to this
logic: If we would write the equations of motion in real space, we would factor out a
common exponent and we are left with the term above.

• Add −3f on the diagonal because each mass is connected to three spings.

With these considerations we arrive at

! "
−3 1 + e−ik·d2 + e−ik·(d1 −d2 )
A(k) = f (4.38)
1 + eik·d2 + eik·(d1 −d2 ) −3

And we immediately write

A(k) = −3f 1 + {1 + cos(k · d2 ) + cos[k · (d1 − d2 )]}σx (4.39)

+ {sin(k · d2 ) + sin[k · (d1 − d2 )]}σy . (4.40)

And therefore
4 4 #
f √
ω± (k) = 3± 3 + 2 cos(kx ) + 4 cos(kx /2) cos( 3ky /2). (4.41)
m
√
The dispersion is characterized by a conical band touching around the K = 2π a (1/3, 1/ 3) and
K ′ = 2π
a (2/3, 0) points. A lot of active research in acoustic, optical and electronic materials is
centered around these so-called “Dirac” or “Weyl” points.
Frequency

K M
Momentum

Figure 4.5: Dispersion of a honeycomb lattice.

30
Chapter 5

Effective negative parameters

Learning goals
• You can explain why we map our designs to an elastic theory with effective parameters.
• You know what strain and stress are and how they are related.
• You know the Helmholtz decomposition.
• You know that doubly negative materials require some thought related to power flow.
• You know Snell’s law in doubly negative materials.

So far, we have seen several discrete models of metamaterials to control wave-propagation. In

order to better understand what one can achieve with metamaterials, we should connect what
we have done so far with the standard literature. For this, however, we need a bit of elasticity
theory.

5.1 Elasticity in one hour

5.1.1 The strain tensor
First, we introduce the strain tensor. Imagine a piece of material that we deform:

r r0 = r + u
0
r

For elastic properties, it is important to know if we deform the material. In other words, a
constant u does not lead to any deformations but only translates the whole object. Hence, the
quantity of interest is the linear strain tensor
! "
1 ∂ui ∂uj
ϵij = + , (5.1)
2 ∂xj ∂xi

31
which captures relative length changes which deform the material.1 We note that ϵij has dimen-
sionless entries. Now we need to connect the strain tensor ϵij to forces in the material that try
to restore the original shape.

5.1.2 The stress tensor

Forces acting on a body are most easily captured by a traction vector t. Let us make a imaginary
cut through our material:
t(r)

With this we might free forces on the cut that were balanced by the other half. We write for
the force F acting on a surface element dS

F = tdS. (5.2)

Therefore, the traction t has units of N/m2 or pressure. The issue with the traction is that it
depends on the cut we take. Think of a piece of liquid. If we cut in the xy-plane, pressure will
exert on that plane a force in z-direction. Had we cut along the xz-plane, would the pressure
give rise to a force in y-direction, and so on. This means, we need again a tensor to capture the
relevant physics. The stress tensor τij encodes the traction in the three principle axis
⎛ ⎞
τxx τxy τxz
τ = ⎝τyx τyy τyz ⎠ , (5.3)
τzx τzy τzz

where tx = (τxx , τyx , τzx ) stands for the traction for a cut normal to x, etc. From this follows
immediately that the traction for an arbitrary cut normal to n̂ is given by

tn̂ = τ · n̂. (5.4)

5.1.3 Hooke’s law

We now need a relation between relative deformations and the stress tensor

τij = cijkl ϵlm . (5.5)

Here, we defined the 4th-order tensor cijkl . While this looks scary, for our discussion we assume
a homogeneous and isotropic medium, where

τij = λδij ϵll + 2µϵij . (5.6)

1
Why do we symmetrize? Imagine two close-by points with distance dr. After deformation dr′ = dr + u. Let
us see how the distance changes (using the Einstein summation convention)
! "! "
′2 2 ′ ′ ∂ui ∂ui
dr − dr = dxi dxi − dxi dxi = (dxi + dui )(dxi + dui ) − dxi dxi = dxi + dxj dxi + dxl − dxi dxi
∂xj ∂xl
#! "2 $
∂ui
= 2ϵij dxi dxi + O .
∂xj

In other words, only the symmetric combination in ϵij leads to length-changes. It can be shown that the anti-
symmetric counter-part corresponds to rigid rotations.

32
The coefficient λ is called Lamé coefficient and µ is the shear modulus. Note that we made use
of the Einstein summation convention where repeated indices are summed over
$
ϵll = ϵll = ϵxx + ϵyy + ϵzz . (5.7)
l

We often use other names, such as

τxx
E= : Youngs modulus. (5.8)
ϵxx
ϵyy
ν=− : Poisson ratio. (5.9)
ϵxx
with the relations
λ
ν= , (5.10)
2(λ + µ)
νE
λ= , (5.11)
(1 + ν)(1 − 2ν)
E
µ= . (5.12)
2(1 + ν)

5.1.4 The Poynting vector

dSBefore we move on, it is useful to introduce the power flux in an elastic

medium. We know that a force F acting on a particle with velocity v
delivers a power F · v (it has units of Nm
s ). Consider the surface element
Figure 5.1: Surface dS. The force acting on it is given by t · dS = τ · n̂dS = τ dS. Therefore,
element. the power delivered to the cut is given by

P = vτ dS. (5.13)
Note that dS points out of the volume. To get the power delivered to a volume, we need to
invert the sign and we write for the power flux J(r)

J(r) = −τ · u̇. (5.14)

c
This is nothing but the elastic counter-part to the Poynting vector 4π E∧B in electromagnetism.

5.2 The elastic wave equation

The Newton’s equations of motion are given by
∂τmk
ρüm = . (5.15)
∂xk
Using ! "
∂uj ∂um ∂uk
τmk = λδmk +µ + (5.16)
∂xj ∂xk ∂xm
we find that
ρü = (λ + 2µ)∇(∇ · u) + µ∇ ∧ (∇ ∧ u). (5.17)
We see that things are slightly more complicated than what we did so far. Waves in solids have
a complicated vectorial structure (paralleling the tensor nature of stress and strain). But by
introducing the longitudinal and transverse potential (the Helmholtz decomposition)2
u = ∇ϕ + ∇ ∧ ψ. (5.18)
2 ik·x ik·x
If we write ϕ = Ae and ψ = Ae , we see that u = kA and u = k ∧ A, respectively. Hence the name
longitudinal and transverse potential.

33
Using this, we arrive at [making use of ∇2 u = ∇(∇ · u) − ∇ ∧ (∇ ∧ u)]
ρϕ̈ = (λ + 2µ)∇2 ϕ, (5.19)
2
ρψ̈ = µ∇ ψ. (5.20)
ν−1
By writing λ + 2µ = E (ν+1)(2ν−1) = EL∗ and µ = ES∗ we recover the same expressions for the
wave equations as we had before.

5.3 Negative E or ρ
Imagine that for some reason either ρ pr E acquire negative values. While this is impossible
for ω → 0 (masses are intrinsically zero and E < 0 would lead to a mechanical collapse), such
effective material parameters might arise from metamaterial engineering. This is indeed what we
are after here: We try to describe an emergent behavior arising from a (discrete) metamaterial
model by reducing it to a simple effective elasticity problem, albeit with material parameters
E, ν, ρ, etc. that can take values otherwise unattainable.
This approach has the benefit of enabling a simple description in terms of standard elasticity
theory, while we can incorporate complicated material designs.
Let us check what happens if either E or ρ take negative values. The wave equation (for
longitudinal waves)
sρ |ρ|ϕ̈ = sE |E|∇2 ϕ. (5.21)
Here, sρ = ±1 and sE = ±1 encode the sign of ρ and E, respectively. Assuming
ϕ = Aeik·x−iωt (5.22)
we find
− sρ |ρ|ω 2 = −sE |E|(kx2 + ky 2 + kz2 ). (5.23)
1
Solving for k = kx2 + ky 2 + kz2 we obtain
4 0
sρ |ρ|
k= × ω. (5.24)
sE |E|
We observe that for sρ /sE = −1, i.e., if only one of the two parameters is negative, we have
a k ∈ C. In other words we deal with evanescent weaves! We also see, that in the case
sρ = sE = −1 wave propagation seem to be unaffected by the negativity of E and ρ. We will
see, however, that this is not the case if we deal with interfaces between different materials.

5.4 Doubly negative metamaterials

We have seen that for doubly negative materials waves can propagate. Here, we study the effect
of double negativity on two examples.

5.4.1 Longitudinal waves in thin rods

We solve for solutions of the form u = ∇ϕ, i.e., longitudinal waves, in a thin rod shown in
Fig. 5.2. :
ϕin ik1 z−iωt ϕout e−ik1 z−iωt z < 0,
1e 1
ϕ(z) = ik2 z−iωt
(5.25)
ϕout
2 e z > 0.
This corresponds to an incoming wave ϕin out out
1 , a reflected wave ϕ1 and a transmitted wave ϕ2 .
At the boundary at z = 0 we need
u1 (z = 0) = u2 (z = 0), (5.26)
τ1 (z = 0) · z = τ2 (z = 0) · z. (5.27)

34
 z=0

Figure 5.2: A thin rod with a boundary between a regular and a doubly negative metamaterial.

For each segment we have

4
Ei ρi
ki2 =ω 2
⇒ ki = ± ω, (5.28)
ρi Ei
where we have to decide on the sign of ki . Let us first match the boundary conditions
−iωt −iωt
uz1 (z = 0) = ik1 ϕin
1e − ik1 ϕout
1 e , (5.29)
−iωt
uz2 (z = 0) = ik2 ϕ2 e out
. (5.30)

For the stress tensors, we need

5
∂uz1 5
5 = −k12 ϕin −iωt −iωt
1e − k12 ϕout
1 e , (5.31)
∂z 5 z=0
5
∂uz2 5
5 = −k22 ϕout −iωt
2 e . (5.32)
∂z 5 z=0

From this, we obtain

k1 (ϕin out out

1 − ϕ1 ) = k 2 ϕ2 , (5.33)
E1 k12 (ϕin
1
out
+ ϕ1 ) = E2 k22 ϕout
2 . (5.34)

We set ϕin out

1 = 1 and solve for ϕ1/2 :

E1 k1 − E2 k2 2E1 k12
ϕout
1 =− , ϕout
2 = . (5.35)
E1 k1 + E2 k2 k2 (E1 k1 + E2 k2 )
Now is a good moment to take care of the signs. We want ϕin
1 to be an incoming wave. Let us
calculate the Poynting vector for this wave

Jin,1 = −τ · u̇ = E1 ωk13 (ϕin 2

1 ) ẑ. (5.36)
1
⇒ for k1 = + ρ1 /E1 we have indeed an incoming wave described by ϕin
1 . Going through the
same calculation for ϕout
2 we find

Jout,2 = E2 ωk23 (ϕout 2

2 ) ẑ. (5.37)

For this to be an out-going wave we need

4 4
ρ2 ρ2
k2 = − = sign(E2 ) . (5.38)
E2 E2

35
Let us introduce a few helpful quantities. First the elastic impedance
1
zα = ρα Eα , (5.39)

and the elastic index of refraction

4
ρα
nα = sign(Eα ) . (5.40)
Eα

Using these definitions we can summarize this solution by writing

k1 n1
= , (5.41)
k2 n2
z2
z2 − z1 1− z1
ϕout
1 = =− z2 , (5.42)
z1 + z2 1+ z1
n1 2
ϕout
2 = . (5.43)
n2 1 + zz21

What do we learn from this exercise?

• If we match impedances, i.e., z1 = z2 , there is no reflected wave.

• For E < 0 ⇒ n < 0 and the sign of k is inverted to have a causal energy flow.

5.4.2 The Veselago lens

We finish this chapter with an application of a doubly negative material in two dimensions.
Imagine the following situation depicted in Fig. 5.3.

Figure 5.3: Boundary between two materials for a Veselago lens.

We consider shear waves of the form

⎛ ⎞
Ax
u=∇∧ψ with ψ = ⎝ 0 ⎠ ei(kx ,0,kz )·x−iωt , (5.44)
Az

which means ⎛ ⎞
0
u = (Ax ikz + Az ikx )ei(kx ,0,kz )·x−iωt ⎝1⎠ , (5.45)
0
or in other words shear waves traveling in (kx , 0, kz ) direction with a deformation in the ŷ-
direction. Again, the wave equation dictates

ρi ψ̈ = µi ∇2 ψ or ρi ω 2 = µi (kx2 + ky2 + kz2 ). (5.46)

36
Let us again assume an incoming wave form the left and a transmitted and reflected wave
⎧ ⎛ ⎞ ⎛ ⎞
⎪
⎪ Ain Aout
⎪
⎪ x
⎜ ⎟ ik1 z+ikx1 x−iωt ⎜
x
⎟
⎪
⎪
1 1
+ ⎝ 0 ⎠ e−ikz z+ikx x−iωt z < 0,
⎪ψ 1 (x, z) = ⎝ 0 ⎠ e z
⎪
⎪
⎨ Ain Aout
ψ(x, z) = ⎛ z ⎞ z
(5.47)
⎪
⎪ B out
⎪
⎪ ⎜
x
⎟ 2
⎪
⎪
2
ψ 2 (x, z) = ⎝ 0 ⎠ eikz z+ikx x−iωt z > 0.
⎪
⎪
⎪
⎩ B outz

The boundary conditions are given by

u1 (z = 0) = u2 (z = 0), (5.48)
τ1 (z = 0) · ẑ = τ2 (z = 0) · ẑ. (5.49)

For the first equation we have to assure that

1 1 ikx x1 1 ikx x 1 ! 2
(Ain in
x ikz − Az ikx )e + (−Aout out
x ikz − Az ikx )e = (Bxout ikz2 − Bzout ikx2 )eikx x . (5.50)

If this shall hold for all x we need

kx1 = kx2 . (5.51)
This was to be expected as we do not break translational symmetry in x-direction. Moreover, we
certainly need that all waves have the same frequency ω. That means that the above equation
together with
ρi ω 2 = µi [(kxi )2 + (kzi )2 ] (5.52)

fixes |kz |2 . To determine the sign of kz we go through

Figure 5.4: Snell’s law.
the Poynting argument again. It is easy to see that a
negative µ2 < 0 again reverses the sign of the power
flux with respect to µ1 > 0. We therefore again have to
take sign(kz2 ) = −1. From these considerations we can
determine Snell’s law

kx1
sin(ϑ) = 1 (5.53)
(kx1 )2 + (kz1 )2
kx2
sin(ϑ′ ) = − 1 (5.54)
(kx2 )2 + (kz2 )2

Hence, we have
1
sin(ϑ) sign(µ2 ) ρ2 /µ2 n2
′
= 1 = . (5.55)
sin(ϑ ) sign(µ1 ) ρ1 /µ1 n1

For the full scattering solution we need to solve the compatibility conditions above. However,
here we are not interested in the amount of power transferred, but only in the direction of in and
out-going waves. Veselago realized that [1] (in the context of electromagnetic waves) the above
Snell’s law for doubly negative materials leads to a perfect flat lens. He considered a situation
shown in Fig. 5.5. We see that this sandwich gives rise to a perfect lens.

37
 Figure 5.5: A Veselago lens.

5.4.3 Superlensing
Pendry realized in 2000 that a slab of a doubly negative material not only acts as a planar lens
but can also enhance features that are normally suppressed due to the diffraction limit.
The diffraction limit arises from the following consideration. We know that

ω 2 /c2 = kx2 + ky2 + kz2 . (5.56)

Let us assume that we have an object we want to image in the xy-plane at z = 0 of the form
f (x, y) = Θ(|x| − a/2). The Fourier transform of this object is given by
ˆ ˆ
ˆ
f (kx , ky ) = dx e ikx x
Θ(|x| − a/2) dy eiky y = (5.57)
5a/2
1 ikx x 55
=⟨ e 5 ⟩2πδ(ky ) (5.58)
ikx x=−a/2
sin(kx a/2)
= 4π δ(ky ). (5.59)
kx
We see that to reproduce f (x, y) we need kx to be arbitrarily large. However, we can write for
#
kz (kx , ky , ω) = ± (ω/c)2 − kx2 − ky2 . (5.60)

We immediately see that for kx2 + ky2 > (ω/c)2 , kz will become purely imaginary and features
small than c/ω will not propagate!
For the example above, we get the acoustic image at a distance d by calculating
1
ˆ
fd (x, y) = dkx dky fˆ(kx , ky )eikz (kx ,ky ,ω)d e−i(kx x+ky y) . (5.61)
(2π)2

We see that Fourier components with a large wave number are damped way faster, as these
components turn into evanescent waves at the frequency we probe the system.
We investigate how this changes if we add again a slab of a doubly negative material. For
simplicity we look at the evanescent waves, where kx2 > µρ ω 2 . Moreover, we constrain ourselves
to the situation where we have
We first concentrate on the left boundary and write for the potential
⎧ ⎛ ⎞ ⎛ ⎞
⎪
⎪ 1 r
⎪
⎪ ⎜ ⎟ ikz z+ikx x−iωt ⎜ ⎟ −ikz z+ikx x−ωt
⎪
⎪ ψ 1 (x, z) = ⎝0⎠ e + ⎝0⎠ e z < 0,
⎪
⎪
⎪
⎨ 0 0
ψ= ⎛ ⎞ (5.62)
⎪
⎪ t
⎪
⎪ ⎜ ⎟ ik′ z+ikx x−ωt
⎪
⎪
⎪ψ 2 (x, z) = ⎝0⎠ e z
⎪
z > 0.
⎪
⎩ 0

38
Clearly the wave vectors have to fulfill
1
kz = +i kx2 − ω 2 , (5.63)
1
kz′ = +i kx2 − ω 2 /|µ|. (5.64)

How did we choose those signs? As we are dealing with evanescent waves, Jz = 0 and no power-
flux argument can be invoked. However, we consider a decaying field from the left and therefore
the field on the right z > 0 should also decay in the positive z-direction to preserve causality.
We can now match the boundary conditions

u1 (z = 0) = u2 (z = 0), (5.65)
τ1 (z = 0) · ẑ = τ2 (z = 0) · ẑ. (5.66)

The first line is easy

kz − rkz = tkz′ . (5.67)
For the second we need to calculate τ from u. We only need the three components ταz for
α = x, y, z as we match τ · ẑ. Moreover, we only have uy which depends on x and z. Therefore
ϵll = ϵxx + ϵyy + ϵzz = 0. This leaves us with
! " :
1 ∂uα ∂uz ∂uy −kz2 (1 + r)eikz z+ikx x−iωt z < 0,
ταz = λδαz ϵll +2µϵαz = 2µ + =µ = ′
, -. / 2 ∂z ∂xα ∂z −kz′2 teikz z+ikx x−iωt z > 0.
=0 ,-./
=0

And with this we need

kz2 (1 + r) = µkz′2 t. (5.68)
Solving for t and r we find

kz2 1 µkz′ − kz
t=2 ; r= . (5.69)
kz kz + µkz′
′ kz + µkz′

Note that for µ → −1 we have kz′ → kz and both r and t diverge! In particular t2 + r2 diverges
as well. This is only possible because we deal with evanescent waves that carry no power-flux!
If we now try to move towards the description of a finite slab with thickness d, we also need the
same expressions for the right boundary

kz′ (1 − r) = tkz , (5.70)

µkz′2 (1 + r) = tkz2 . (5.71)

From which we obtain

kz′2 µ kz − µkz′
t′ = 2 ; r′ = . (5.72)
kz kz + µkz′ kz + µkz′
Naively, one could expect the transmission function to be
′
Tnaive (d) = teikz d t′ , (5.73)

where the first t is the left boundary, the exponential describes the transmission inside the slab
and t′ encodes the effects of the second boundary. However, the reflected wave inside the slab
hits the left boundary again, etc. So the full transmission is given by
′ ′ ′ ′
T (d) = tt′ eikz d + tt′ r′2 e3ikz d + tt′ r′4 e5ikz d + tt′ r′6 e7ikz d + . . . (5.74)
2 3 ′
tt′ eikz d
′ ikz′ d ′2 2ikz′ d ′4 4ikz′ d
= tt e 1+r e +r e + ... = . (5.75)
1 − r′2 e2ikz′ d

39
 1
Let us analyze this expression. kz′ = i kx2 − ω 2 /|µ| with kx2 > ω 2 /|µ| ⇒ both exponential factors
are much smaller than one for large enough d. If r′2 and tt′ would be well behaved, we could
neglect r′2 exp[−2ikz′ d] with respect to 1 and we would obtain the naive result. However, for
µ → −1, we need to be more careful
′
µ2 eikz d
lim T (d) = lim 4kz kz′ 8 92 (5.76)
µ→−1 µ→−1 (kz + µkz′ )2 ′
kz −µkz
1− kz +µkz′ e−2ikz′ d
′ 1
= lim 4kz kz′ eikz d (5.77)
µ→−1 (kz + µkz′ )2 − (kz − µkz′ )2 e−2ikz′ d
1 kz →kz′ −ikz d
= 4kz2 eikz d ′ = e . (5.78)
(kz − kz′ )2 − (kz + kz′ )2 e−2ikz d

We found an astonishing result: Evanescent waves are exponentially enhanced while passing
through the doubly negative material! This famous results by Pendry [2] established the concept
of a superlens built from metamaterials.

References
1. Veselago, V. G. “The electrodynamics of substances with simultaneously negative values of
ϵ and µ”. Sov. Phys. Usp. 10, 509 (1968).
2. Pendry, J. B. “Negative Refraction Makes a Perfect Lens”. Phys. Rev. Lett. 85, 3966 (2000).

40
Chapter 6

The continuum limit

Learning goals
• You know how to define effective parameters for a discrete system.
• You know how to find the effective mass for a simple setup.
• You know how to determine the effective spring constant.

In the chapters 3 and 4 we have seen how we can shape the propagation of waves by using
periodic structures or by employing local resonances. In the last chapter, on the other hand,
we have seen how one can obtain interesting phenomena within the simple framework of a
wave-equation, albeit with “unnatural” parameters E < 0 and ρ < 0.
How to map a structured metamaterial to a simple wave equation with effective parameters is
a highly non-trivial task. We can group approaches to do so into two classes. First, if we deal
with a continuous medium to which we add resonances of periodic array of scatterers one can
do the following

• Solve how a circular object with E, ρ, ν scatters a wave.

• Solve the problem of inclusions in a circular metamaterial and calcu-

late how it scatters an incoming wave.

• By comparing the two results, fix the effective E, ρ, ν of this circular

material.

This program is called homogenization. Here, we want to follow a different path. We would
like to know how to map a discrete model with an interesting band structure to an effective
“continuum” theory. In this case we introduce “natural” coordinates ξn and think of the blue
bodies in Fig. 6.1 as rigid, despite the fact that they might have an interesting inner life. The
effective mass meff and the effective spring constant are then defined via
Fn nm Fnm
meff = and feff = , (6.1)
∂t2 ξn ξn − ξm

where Fn are the (external) forces on block n and Fnm the forces between n and m.

Figure 6.1: Effective coordinates of a more complicated body.

41
 ⇠
m

F M

Figure 6.2: Mass in mass system.

6.1 Mass in mass

We start by revisiting the mass-in-mass system, cf. Fig. 6.2. Imagine that we would not see the
interior mass M . So we should find an effective description for the “outer” mass only. Let us
write
M v̈ = F (ξ − v). (6.2)
We assume harmonic motion

v = v0 eiωt , (6.3)
iωt
ξ = ξ0 e . (6.4)

Inserted into (6.2) we get

− ω 2 M v0 = F (ξ0 − v0 ) (6.5)
and therefore
F ξ0
v0 = . (6.6)
F − ω2M
1
By writing ωM = F/M , we get
2
ωM
v0 = 2 − ω 2 ξ0 . (6.7)
ωM
Let us now introduce the effective mass. We know that
dP
= ∂t (m∂t ξ + M ∂t v) = Fext . (6.8)
dt
As we do not “know” about the mass M we would write

meff ∂t2 ξ = Fext . (6.9)

As Eq. (6.7) is independent of time, we have

2
ωM
∂t v = 2 − ω 2 ∂t ξ. (6.10)
ωM

Using this in Eq. (6.8), we find

! 2 "
ωM
∂t m∂t ξ + M 2 ∂t ξ = Fext (6.11)
ωM − ω 2
6 ! 2 " 7
M ωM
∂t m 1 + 2 − ω 2 ∂t ξ = Fext (6.12)
m ωM
meff ∂t2 ξ = Fext . (6.13)

42
 50

0 1 2 3

Figure 6.3: Effective mass.

In the last line we defined ! "

2
M ωM
meff =m 1+ 2 − ω2 . (6.14)
m ωM
We observe:

• for meff < 0: You push the mass and it moves in your direction.

• meff is strongly frequency dependent!

• ω → 0: meff = m + M .

Let us now use this effective mass description to solve the chain of such local resonators

meff (ω)ξ¨n = −f (2ξn − ξn+1 − ξn−1 ). (6.15)

As usual we assume ξn = exp ikna to obtain

meff (ω)ω 2 = 2f [1 − cos(ka)] = 4f sin2 (ka/2). (6.16)

When we solve for k we find * 0 +

2 1 meff (ω)
k = arcsin ω . (6.17)
a 2 f
From this expression we see that k becomes imaginary if

1. meff < 0 ⇐ Band gap

1
2. 12 meff /f ω > 1 ⇐ above band edge

6.2 Effective spring

We have seen that a “hidden” mass can give rise to an effective negative mass. What about
negative springs? Let us consider the system shown in Fig. 6.4. The equations of motion are

mÿ = F2 , (6.18)
F2 = 2(F1 − 2κ2 x̄) tan(α), (6.19)
F1 = κ1 (x − x̄). (6.20)

43
 F2
y(t)
m

F1 ↵ F1
1 2 1
x(t) x̄(t)

Figure 6.4: Effective sping.

From the last line we infer that x̄ = x − F1 /κ1 . Inserted into the middle line we get F2 =
2[F1 − 2κ2 (x − F1 /κ1 )] tan(α). Moreover, we can use that x̄ = tan(α)y which yields y =
(x − F1 /κ1 )/ tan(α). Using all of this in the first line we find
! " 6 ! "7
F1 F1
−mω 2 1 − = 2 F1 − 2κ2 x − tan2 (α) (6.21)
κ1 κ1
mω 2 − 2(κ1 + 2κ2 ) tan2 (α)
⇒ x = F1 . (6.22)
κ1 [mω 2 − 4κ2 tan2 (α)]
We can now define
F1
x= (6.23)
2κeff
and write *; <+
4κ2
1 ω2 − 2
m tan (α)κ1
κeff = , (6.24)
2 ω 2 − ω02
2(κ1 +2κ2 )
with ω02 = m tan2 (α). We check for the sanity of this results by taking
1 1 2
lim = + , (6.25)
ω→0 κeff κ2 κ1
which is what we should obtain! Analogous to the effective mass, the effective spring constant
can be negative, cf. Fig. 6.5.
Again, we can bunch such elements together to find the equations of motion for a chain made
from effective springs
mξ¨n = −κeff (2ξn − ξn−1 − ξn+1 ) (6.26)

1/( 12 + 2
1 )

Figure 6.5: Effective spring constant.

44
Not so surprisingly the wave number
!4 "
2 m
k = arcsin ω (6.27)
a κeff (ω)

is again complex in the region where κeff is negative and above the upper band edge.

6.3 Double negativity

We now combine the two ingredients, an effective mass and an effective spring as shown in Fig. .
We immediately find
meff ω 2 = 4κeff sin2 (ka/2). (6.28)
Task: Write meff and κeff in a way that transparently captures the negative sections. Then
solve the above equations for ω and plot the solutions as a function of k and vary the tuning
parameters. Convince yourself that indeed for meff κeff < 0 you and up in a band gap and for
meff < 0 and κeff < 0 you have indeed ∂k ω(k) < 0.

References
1. Zhou, X., Liu, X. & Hu, G. “Elastic metamaterials with local resonances: an overview”.
Theor. Appl. Mech. Lett. 2, 041001 (2012).
2. Lee, S. H. & Wright, O. B. “Origin of negative density and modulus in acoustic metamate-
rials”. Phys. Rev. B 93, 024302 (2016).

45
Chapter 7

Topological mechanical
metamaterials

Learning goals
• You can sketch the history of topological bandstructures.
• You know the Su-Schrieer-Heeger model.
• You know what a Chern insulator is.

In this chapter we introduce a new approach to mechanical metamaterials which is inspired by

a topic known in modern solid state physics: topological bandstructures. Before we present the
main concepts we discuss the key application where topology might have a considerable impact
in the near future: wave-guiding.

7.1 Mechanical waveguides

The precise control of the flow of mechanical energy is an important task for applications ranging
from energy harvesting or vibration isolation to the design of mechanical signal and information
processing, both classical or quantum information. Let us introduce a few design principles for
wave-guiding.

7.1.1 Surface acoustic waves

One way to control the flow of vibrations is simply to take a half-space filled with some elastic
material. The wave equation
! "
1 ∂2ϕ ∂2 ∂2 ∂2
= + + ϕ (7.1)
c2 ∂ 2 t ∂x2 ∂y2 ∂z2
permits solutions of the form

ω2 1
ϕ(x) = eikx x+iky y−z/ξ−iωt ⇒ − 2
= −kx2 − ky2 + . (7.2)
c ξ
In other words, waves can be confined to the surface around z = 0
but they propagate in the x − y-plane. To really match the boundary conditions, one needs to
have either transverse (Love) or mixed (Rayleigh) polarizations of the full elastic waves.
What is important for our purposes is, that one can use surfaces as mechanical waveguides,
where the the energy is falling off only with 1/r rather than 1/r2 for a propagation in three
dimensions.

46
7.1.2 Total internal reflection
Even more spatial confinement is offered by using total internal reflection. Remember that for
a boundary we had to match the parallel wave number
∥ ∥
kin = ktrans . (7.3)

On top of that, we also have

0! "2
⊥ ω ∥
ktrans = ± − ktrans . (7.4)
c2
⊥
We see that if c2 > c1 and the angle of incident ϑ is too large, ktrans will
become imaginary. That means, that above a certain critical angle

ϑcrit = arcsin(n2 /n1 ) (7.5)

the waves experience total internal reflection.

7.1.3 Band gap wave guides

We know another way to get evanescent waves! Let us design a phononic crystal in a two-
dimensional sheet of Aluminum. If we take a unit cell of the form

and extend it in both directions, we get a Brillouin zone of the form

The spectrum along the high-symmetry line Γ-X-M -Γ is shown in Fig. 7.1.
Let us quickly check if the frequencies make sense. The speed of sound in Aluminum is about
v ∼ 6000 m/s and we choose a lattice constant of a = 5 mm. This should yield a gap frequency
of about νgap ∼ av ∼ 106 Hz.
From the above figure we learn that from about 680-850 kHz, no waves can travel through our
structure. This observation enables the following wave-guide design.

47
 Figure 7.1: Spectrum for waveguide plate

We expect that for frequencies around the band-gap, the middle strip without the clamped
regions acts as a perfect wave guide: The absence of the periodic structuring removes locally
the band-gap. The following finite-element mode analysis supports this expectation:

We have now seen a collection of possible wave-guiding structures. While some can be fairly
small and maybe even reliable for straight segments, they all suffer from the same problem:
Imperfections in the fabrication will induce disorder or the design requires bends in the wave
guide. Both will lead to backscattering, where part of the injected energy is reflected back to
where it came from. For a reliable waveguide, we might profit from an entirely new design
principle: topological band structures.

48
7.2 A short history of topology in condensed matter physics
An early instance of topological effects relevant for transport physics was discovered in 1980 in
the context of electron transport in a strong magnetic field.
For a two-dimensional electron system, a magnetic field exerts a force
on moving charges with is perpendicular to their velocity

FL = −|e|v ∧ B. (7.6)

This leads to a build-up of charges at the edges of the sample and

introduces a voltage across the sample. This effect is quantified by
the Hall conductance
! "
Ix 1
σH = . RH = . (7.7)
Vy σH

Clearly, σH will depend on the parameters of the system such as density ne , the magnetic field
strength B, etc.
In his famous measurement in 9180, Klaus von Klitzing discovered the following behavior:

It turns out, that

h
6453 Ω = . (7.8)
4e2
and the flatness of the plateau allows to determine the von Klitzing constant h/e2 with a precision
of 10−9 !
The Nobel prize for physics of 2016 was given to Thouless, Kosterlitz, and Haldane for uncovering
the topological origin of this precision. Thouless and co-workers could should that σH is given
by the Chern number, a topological quantity that depends on the system as a whole, takes only
integer values and can therefore only be changed in steps

e2
σH = m m ∈ Z. (7.9)
h
Underlying their calculation is the existence of a spectral gap. If we now think of a finite smaple
with an edge, it is easy to show that m = 0 in vacuum. If now m ̸= 0 inside the sample, it has
to change across the boundary from m ̸= 0 → m = 0. This is only possible if the spectral gap
is closing on the edge!

49
In fact, one can assign a Chern number to each band. The bulk-edge correspondence tells us
that the difference in Chern number above and below the gap ∆m = mabove − mbelow dictates
the signed number of surface channels. “Signed” means, that we count each channel times the
sign of its average group velocity.

The fact that the Chern number m = 1 (left picture) predicts a uni-directional channel is
responsible for the high degree of precision of the flatness of the plateau in σH . If we add
disorder to the system, electrons traveling on the red branch cannot scatter and turn around!
The electrons travel in a strict one-way road.
Such stable features that make transport of electrons extremely robust is what we seek for
vibrations in metamaterials.
For a wide use both in electronics as well as in phononics the strong magnetic field needed for
the quantum Hall effect is a bit of a problem. In electronics it is annoying, in phononics almost
impossible to achieve.
A discovery by Kane and Mele and a subsequent experiment led by Zhang and Molenkamp in
2005–2007 came as a cure to this problem.
Kane and Mele realized that under the right circumstances a topological band structure can
exist also for time-reversal symmetric systems, i.e., in the absence of a magnetic field. In what
is called a topological insulator, counter propagating edge channels exist, but scattering between
them is forbidden by a symmetry of the system.

50
For the case of topological insulators the topological quantum number is not an integer m ∈ Z,
but
m ∈ Z2 = {0, 1}. (7.10)
This indicates that there is an odd (even) nubmer of pairs of counter-propagating surface states
for m = 1 (m = 0).
In the following we are going to discuss a few simple lattice models that illustrate how the bulk
of a system can dictate what happens on the surface.

7.3 Simple examples

7.3.1 The Su-Schrieffer-Heeger model
The goal of this section is to realize two things. First, we want to see how a bulk quantity is
linked to an edge feature. Second, we try to understand how symmetries can be important.
Let us consider the following chain:

The equations of motion look like

! " ! "! "
2 uk f + f′ −f ′ − f eika uk
−ω = . (7.11)
vk −f ′ − f e−ika f + f′ vk
, -. /
h(k)

The matrix h(k) can be written as

h(k) = f + f ′ −[f ′ + cos(ka)] σx −f sin(ka) σy . (7.12)
, -. / , -. /
dx k dy (k)

We see that dz (k) ≡ 0, or equivalently

{σz , h(k)} = σz h(k) + h(k)σz = 0. (7.13)
Let us write
f = f0 (1 + ϵ), (7.14)
f = f0 (1 − ϵ). (7.15)
With this parameterization we obtain for the spectrum
2 1 3
ω 2 (k) = f0 2 ± 1 + ϵ2 + (1 − ϵ2 ) cos(k) . (7.16)

51
We see that for ϵ ̸= 0, we deal with two bands separated by a spectral gap. Let us check what
happens at a surface.

We find that we can construct a solution where there is no force on any of the vi ’s and they
stand still
mv̈i = f ′ (ui − vi ) − f (vi − ui+1 ) = 0 (7.17)
with vi = 0. We find that (7.17) requires

f ′ ui = −f ui+1 . (7.18)

This means that we have

f′
ui+1 = − ui , (7.19)
f
which we solve with ! "
1+ϵ
ui = e−ri /ξ , ξ = log . (7.20)
1−ϵ
What do we learn form that? We find an exponentially localized mode at the surface if ξ > 0.
In other words, if ϵ > 0! At ϵ = 0 we find that ξ → ∞ and also the band gap closes.
We obtain the frequency of the edge state by solving
1
− üi = −f ′ (ui − vi ) − f (ui − vi−1 ) = −2f0 ui ⇒ ω = 2f0 . (7.21)

52
We conclude that the edge state lies in the “middle” of the gap. We realize that we have a
band structure that changes from ϵ < 0 to ϵ > 0 via a gap-closing and only on one side of this
transition we have a surface state.

7.3.1.1 Bulk topology

Is there something in the bulk that lets us predict this? Let us look at the evolution of the
normalized d(k)–vector
d(k)
d̂(k) = . (7.22)
|d(k)|
First of all, this can be done if |d(k)| ̸= 0 for all k-values. This is exactly the gap condition.
Second, the Brillouin zone has the topology of a unit circle S 1 as we can identify k = −π/a with
k = +π/a.
Moreover, as d(k) has only two components [remember the symme-
try {σz , h(k)}], it is constrained to the equatorial plane.
In other words, our equation of motion for the bulk define a map-
ping from S 1 → S 1 . Such mappings are characterized by a winding
number:How many times is d̂(k) running around the unit circle if k
is goes around the unit circle (the Brillouin zone) once. For our case
the situation is shown below

We observe that for ϵ > 0, d̂(k) indeed wraps around the equator! The transition happens when
the circle defined by d(k) touches the origin, exactly where the gap is closing. We can cast this
observation into a formula. The infinitesimal increase in angle ϕ is given by

Ω(k) = e−iϕk ∂k eiϕk , (7.23)

where we wrote the two-dimensional, normalized d̂(k) as a complex number. The topological
index can now be written as
1
˛
C= dk Ω(k) ∈ Z. (7.24)
2π

53
The symmetry {σz , h(k)} was important in the definition of this index! If the image of [−π/a, π/a[
can be anywhere on the S 2 sphere, we can deform it without ever closing the gap.

This model has recently been implemented in a acoustic crystal [1]. This model is the simplest
example of the following corner-stones of topological crystals:
1. An integer valued index of the bulk can be defined. (Here, the winding number C).
2. The value of this index makes a statement about the physics on the edge of the system.
(Here, localized mid-gap states).
3. A symmetry is essential for the index to be well defined. (Here, dz (k) ≡ 0).
In the following we want to go one dimension higher. In two dimensions, the edge of the system
is a one dimensional. This means, the possible edge states are not single, isolated mid-gap
modes, but dispersion one-dimensional channels.

7.3.2 Chern insulators

In this chapter we introduce the work-horse of two-dimensional topological systems: the lattice
Dirac-model. We will see that in its simple form, one cannot implement it in a passive metama-
terial. However, a large amount of our understanding is based in this example, which makes it
a valuable model to know.
We start with simple Dirac fermions described by
3
$
H= di (k)σi with d(k) = (kx , ky , m). (7.25)
i=1

We immediately see that the spectrum of H is given by

1
ω(k) = ± |k|2 + m2 . (7.26)

We realize that H describes quite generically a touching (m = 0) or near-touching point (m ̸= 0)

of two adjacent bands. However, in its form above, H does not describe a lattice system as it is
not periodic in kx and ky . To turn it into a periodic model, we simply write
(kx , ky , m) → [sin(kx ), sin(ky ), m + 2 − cos(kx ) − cos(ky )]. (7.27)
We see that for kx , ky → 0m the two models are identical.
For the Su-Schrieffer-Heeger (SSH) model, the normalized d̂(k)-vector encoded the bulk topol-
ogy. What do we deal with in the present case? The first observation is that the Brillouin zone
has now the structure of a torus T 2 :

54
Here, however, we do not have any symmetry that constrains the target space to the circle S 1 .
Therefore, the target space is the sphere S 2 . In other words, the family of matrices (one for
each k) defines a mapping

H(k) : T 2 → S 2
d(kx , ky )
(kx , ky ) 4→ d̂(kx , ky ) = . (7.28)
|d(kx , ky )|

What can now be the topologically distinct classes of Hamiltonians? It is the question if the
d̂-vector maps the whole sphere!

Clearly, these two situations are very different and not smoothly deformable into each other.
Moreover, it is easy to see, that this wrapping of S 2 is nothing bu the generalization of our
S 1 → S 1 mapping in the SSH model. Note, that the winding number of the SSH model could
be both ±C, depending on how the d̂-vector rotated. The same is true here.
Let us look at our model. We first concentrate on the in-plane d-vector [dx (k), dy (k)]. This is
like looking at S 2 from a star above the north pole.

55
Note, that [dx (k), dy (k)] is zero only at the red points! In other words, gap closing can occur
only at these high-symmetry points.
We see that certainly the d-vector smoothly points in all directions on the equatorial plan. In
order to know if we wrap the whole sphere, we need to check if we reach both the south- and
north-pole when we go through the whole Brillouin zone.
Let us start at m < −4. Then

dz (k) = m + [2 − cos(kx ) − cos(ky )] (7.29)

, -. /
∈ [0,4]

is positive for all values of kx , ky and we never reach the south pole.
At m = −4 we have that dz (k) = 0 at (kx , ky ) = (±π, ±π). At these points also the in-plane
part vanishes and there we deal with a gap-closing.
After we passed through m = −4, the normalized d̂ is pointing to the north pole at (0, 0) and
to the south pole at (±π, ππ).1

⇒ We wrap the sphere once!

At m = −2 we close the gap at (0, ±π) and (±π, 0). Note that these are two different points.
After that we still point both to the south as well as to the north pole once, but we changed the
orientation of the map (This is admittedly hard to see without an actual calculation).
Finally, at m = 0, we have a gap-closing at (0, 0) and after that, the d̂-vector does not reach
the north-pole anymore. From this analysis, we conclude that the topological index evolves in
the following way:

1
Note that this is only one point!

56
Of course, this can be formalized by writing
1
ˆ
C= dk ϵαβγ dˆα ∂kx dˆβ ∂ky dˆγ , (7.30)
2π T 2
which is known as the skyrmion number. In fact, the skyrmion number is nothing but the special
case of the Chern number if we only deal with two bands. For a general problem we define

Aα = iψ † (kx , ky )∂kα ψ(kx , ky ), (7.31)

Fαβ = ∂α Aβ − ∂β Aα (7.32)
1
ˆ
C= dk Fαβ . (7.33)
2π T 2

7.3.2.1 Edge states

We have now established two cornerstones of a topological band structure: (i) a discrete topo-
logical index C, that (ii) can only change when we close a spectral gap.
Let us now check for the third property: non-trivial edge modes for C ̸= 0. After all, it is those
modes that we try to make use of for wave guiding applications.
For now, our model is written in k-space. To check for edge-states,
we want to study H on a cylinder. This implies that we keep one
direction, say x, periodic and eikx x transports our solution in x-
direction. However, in y-direction, we need to go to direct space.
Let us analyze what terms we deal with. First of all, we have a two-
band model. Moreover, we now need to include L sites in y-direction.
Therefore H will be of the form
! "
a γ
H= , (7.34)
γ† b

where a describes the L × L matrix of internal state a, b the L × L matrix of the internal state
b and γ the mixing between them. In other words, dz encodes a = −b, dx and dy encode γ.
We now analyze the different terms. In dz (k) we have

m + 2 − cos(kx ) − cos(ky ). (7.35)

Therefore, ±[m + 2 − cos(kx )] stands on the diagonal of a (b) as these terms do not change the
y-site. However
1 1
− cos(ky ) = − eiky − e−iky (7.36)
2 2
encode the hopping by ±1 in y-direction. We hence find
⎛ ⎞
m + 2 − cos(kx ) − 12
⎜ − 12 m + 2 − cos(kx ) − 12 ⎟
⎜ ⎟
⎜ −2 1
m + 2 − cos(kx ) −21 ⎟
a=⎜ ⎟ (7.37)
⎜ − 1
m + 2 − cos(k ) − 1⎟
⎝ 2 x 2⎠
1 ..
− 2
.

and b = −a. We now turn to dx (k) = sin(kx ). Clearly this just leads to the diagonal entry
sin(kx ) in γ. dy (k) is harder.
1 8 iky 9
sin(ky ) = e − e−iky (7.38)
2i
connects again y to ±1. Remember that
! "
0 −i
σy = (7.39)
i 0

57
That means that for γ we need to multiply (7.38) with the right pattern of ±i to obtain the γ
matrix ⎛ ⎞
sin(kx ) − 12
⎜ 1 sin(kx ) − 12 ⎟
⎜ 2 ⎟
⎜ 1
sin(kx ) −2 1 ⎟
γ=⎜ 2 ⎟. (7.40)
⎜ 1
sin(k ) − 1⎟
⎝ 2 x 2⎠
1 ..
2
.
With this we can now diagonalize ! "
a γ
H= (7.41)
γ T −a
to find

7.3.3 Application in mechanics

We set out to establish a method to construct stable waveguides for vibrations. In its current
form, H does not describe a valid dynamical matrix!

58
 1. Eigenvalues of H are smaller than zero. ⇒ in
ẍ = Hx → −ω 2 x = Hx (7.42)
we will have to take the square root of negative eigenvalues. This can be cured by adding
µ1. (7.43)
to the matrix H. If µ is large enough we get rid of all negative eigenvalues.
2. We have term of the from i sin(ky ) and sin(kx ). They become
1 1
äix ,iy = − bix ,iy +1 + bix ,iy −1 (7.44)
2 2
and
i i
äix ,iy = bix +1,iy − bix −1,iy . (7.45)
2 2
The first problem is in principle doable. The second we cannot implement with a mass-spring
model as we would need complex springs. This observation tells us that we cannot have truly
uni-directional channels in passive mass-spring systems. In the last part of this chapter we will
learn how to circumvent this problem!

7.4 Time reversal invariant topological systems

We have seen that a simple Chern insulator with d(k) = [sin(kx ), sin(ky ), m+2−cos(kx )−cos(ky )]
necessarily leads to complex coupling elements. For electrons in solids, such a complex hopping
can be obtained via an orbital magnetic field. However, for mechanical degrees of freedom this
is not possible. To cure this problem we have two options. (i) We can go beyond the simple
equations of motion in the form of
ẍ = Dx (7.46)
by using gyroscopes. (ii) A second variant is to take a Chern insulator and double the degrees
of freedom in a suitable manner. We cover the second strategy in this lecture.

7.4.1 Doubling the degrees of freedom

We had a system with dynamics described by a matrix H which contained imaginary couplings.
One way to turn a complex number real is by adding its complex conjugate
a + a∗ ∈ R. (7.47)
We follow this idea by doubling the degrees of freedom by writing
! "
H 0
D= . (7.48)
0 H∗
Let us see what happens if we transform to a new basis
! "
1 1 −i
U=√ (7.49)
2 1 i
We transform the dynamical matrix
! "! "! "
1 1 1 H 0 1 −i
D̃ = (7.50)
2 i −i 0 H∗ 1 i
! "! " ! 1 ∗ i
"
1 1 1 H −iH 2 (H + H ) 2 (H − H ∗)
= ∗ ∗ = i 1 (7.51)
2 i −i H iH − 2 (H − H ∗ ) 2 (H + H ∗)
! "
Re H Im H
= ∈ R! (7.52)
−Im H Re H

59
We found a way to render all couplings real by doubling the degrees of freedom.
Note that the Chern number for H ∗ is the negative of the Chern number of H. This implies
that our new edge spectrum looks like

This now means that our surface states are stable, if and only if there are no terms mixing the
two sectors in a way that breaks time reversal symmetry. We therefore deal with a symmetry
protected topological state. This is analogous to the situation in the SSH chain discussed before.

7.4.2 The doubled 13 -flux model

Doubling the Chern insulator discussed above does not lead to a simple mechanical model. we
present here a simpler version corresponding to electrons hopping an a lattice in the presence of
a magnetic field. This models is explained in the following picture.

If we hop in the direction of the arrow, we pick up a phase ϕ, if we hop against it we pick up
−ϕ. Note, that for the pattern of arrows above, whenever you hop around a plaquette, you pick
up a net phase of ϕ. Note that there is a distinct difference to the phase induced by a wave
number k which is −kx − ky + kx + ky = 0 for a roundtrip around a plaquette.

h 2
This non-vanishing flux ϕ corresponds to a magnetic field B = ec ϕa , where a is the lattice
constants. We specialize to ϕ = 2π
3 and hence we find

60
One sees that we have a unit cell with three sites and the spectrum on a cylinder has three
bands and both band gaps have a non-vanishing Chern number C = ±1.

This model now constitutes H, which has obviously imaginary entries. To construct the full
model we write ! "
H
D= (7.53)
H∗
or in pictures
Using the above logic, we obtain a good dynamical matrix via
! " ! "! "
xr,s 1 1 1 ψ+
=√ (7.54)
yr,s 2 i −i ψ−

on every site labelled by (r, s). These degrees of freedom can now be implemented in the
mechanical domain! We need two (x and y) degrees per site which we implement with pendulums.

61
For a 9 × 15 lattice with a total of 270 pendulums the resulting measured spectrum looks like

A few comments

• We indeed find counter-propagating modes that differ by “color”.

• ψ+ ∝ x + iy and ψ− ∝ x − iy. In other words, interpreted as a two-dimensional pendulum

with coordinates (x, y), the two modes are the two circularly polarized modes.

• By building a transducer that couples only to left or right circular polarization, one can
turn the edge channels essentially uni-directional.

7.5 Gyroscope systems

For setups implementing true Chern insulators with mechanical degrees of freedom we refer to
the Refs. [2–4].

References
1. Xiao, M. et al. “Geometric phase and band inversion in periodic acoustic systems”. Nature
Phys. 11, 240 (2015).
2. Nash, L. M. et al. “Topological mechanics of gyroscopic metamaterials”. Proc. Natl. Acad.
Sci. USA 112, 14495 (2015).
3. Wang, P., Lu, L. & Bertoldi, K. “Topological Phononic Crystals with One-Way Elastic Edge
Waves”. Phys. Rev. Lett. 115, 104302 (2015).
4. Ssstrunk, R. & Huber, S. D. “Classification of topological phonons in linear mechanical
metamaterials”. Proc. Natl. Acad. Sci. USA 113, E4767 (2016).

62