Tutorial: Acoustic Metamaterials

Jensen Li and Ping Sheng

Department of Physics, HKUST, Clear Water Bay
Kowloon, Hong Kong, China

Abstract
Waves are generally characterized by angular frequency and wavevector k. Accordingly, this
tutorial is structured into two sections, one on resonance-based acoustic metamaterials, in the
frequency domain, and one on topological acoustics, based in the wavevector domain as
topological structures inherently involve spatial configurations that are a step beyond the
periodic lattices. Each section will begin with a brief introduction of the basic principles,
followed by two examples described in some detail. In the first section we present decorated
membrane resonator and the broadband optimal acoustic absorption structures, the latter being
crucial to the potential applications of acoustic metamaterials. In the second section we present
Jensen, your additions here.
I. Introduction
Acoustic waves in air satisfy the scalar wave equation

, (1)

where denotes the wave speed, which has the value m/s in air, where
Pa is the atmospheric pressure, kg/m3 is the air density, and is the
adiabatic index of air. In Eq. (1), represents the pressure modulation of sound, where P is
the instantaneous pressure. Solution to Eq. (1) has the general functional form with the
dispersion relation

, (2)
as can be easily verified through differentiation. Among all the possible wave functional forms,
the plane wave is the most common. To supplement Eq. (1) and (2), we also
have Newton’s law that relates the pressure modulation and displacement (particle) velocity
, where denotes the air displacement of the sound wave:

1
 , (3)
and the continuity equation:

, (4)

where denotes the instantaneous density. In what follows, we will treat those cases in which
the metamaterial sample has a planar interface with air. Hence, we will regard both the
displacement and displacement velocity as complex scalar quantities, and , that represent the
components of the two vectors normal to the interface.

The and k characterization of the acoustic waves foreshadows the development of acoustic
metamaterials, initially in the realm of resonance-based structures that can display sample
characteristics such as the subwavelength sample size, together with their wave manipulation
characteristics not found in naturally-occurring materials. More recently, the advent of
topological quantum phenomena has propelled a new direction based on novel spatial structures,
i.e., in the wavevector domain, that have topological features distinct from the periodic
structures. This tutorial is accordingly divided into two sections, each with two examples. For the
resonance-based acoustic metamaterials [1-17], membrane-type acoustic metamaterial will be
used as the first example since it involves several general concepts that can be applied in many
other contexts. The second example is the optimal acoustic absorption structure, which can
display tunable absorption spectrum over arbitrarily broad frequency band and thereby
eliminates one of the major drawbacks of metamaterials for potential applications.
Jensen’s inputs here.
Rather than trying to cover a broad swath of this growing field, our intention here is to use
examples to elucidate the basic concepts underlying the acoustic metamaterials, hopefully to lay
the groundwork for the readers to further explore this rapidly evolving area.
II. Resonance-based metamaterials
Resonance-based metamaterials can manipulate waves with subwavelength sample sizes, in
contrast to phononic crystals whose wave characteristics are only apparent when sample’s
structural periodicity is comparable to the relevant wavelength. In this section we detail two
examples with emphasis on explaining the relevant concepts together with the pertinent
mathematics involved.
2.1 Preliminaries

In an acoustically resonant system with resonance frequency , the magnitude of displacement

response of the system to a pressure modulation has the general form

2
 , (5)
where p is the actual pressure on the sample that includes the pressure from both the incident and
reflected waves, A and m denote the relevant area and inertial mass of the resonator, respectively,
and is a dissipation coefficient in unit of frequency. The Green function is defined as

. (6)
The quantity in the bracket has the functional form denoted as a “Lorentzian,” which
characterizes almost all the resonances. In acoustics, an important parameter is the impedance:

. (7)

For a plane wave in air, . The real part of the Green function is non-dissipative; its
dissipative imaginary part corresponds to the real part of impedance. Equation (7) is generally
true for samples with subwavelength lateral dimension so that diffraction is not a concern. The
inverse relationship between the Green function and impedance is especially convenient for
evaluating the impedance of resonance-based metamaterials since the Green function can always
be written as the sum of Lorentzians for resonators with different resonance frequencies, which
can always be obtained, in conjunction with the eigenfunction configurations, either analytically
or through numerical simulations for resonators with complex geometries.
We note here some special properties of the Lorentzian that can be useful for later developments.
In the limit of vanishing , we have

, (8)

where denotes taking the principal value of the quantity that follows. It might appear puzzling
why there can be an imaginary part, which represents dissipation, even when the dissipation
coefficient is zero. The reason is that as diminishes, the resonance peak becomes sharper,
implying an increasing Q factor of the resonator. As Q approaches infinity, the lifetime of the
resonance also approaches infinity; hence even with an infinitesimal dissipation coefficient there
can be finite dissipation over an infinite duration. This is the meaning of the delta function
imaginary part, which can also be regarded as the density of resonant mode per unit frequency,
localized at .
II.2 Decorated membrane resonator

3
A decorated membrane resonator (DMR) consists of a thin polymeric membrane, such as Latex,
slightly stretched over a fixed boundary with one or multiple pieces of solid weight glued on top
[6]. The solid piece(s) can be in different shapes. Here we consider the simplest case of a
circular membrane with a circular weight in the centre. Such a resonator can have multiple
resonances; here we will be concerned with only the two lowest frequency resonance
eigenmodes whose profiles, , are depicted schematically in Fig. 1. Here denotes the
coordinate in the plane of the membrane, whereas the displacement is normal to the plane as
noted earlier and generally much smaller than the wavelength. The resonance frequencies of the
DMR can be especially low because the restoring force of the membrane is weak. Hence the
relevant wavelength can be order(s) of magnitude larger than the size of the DMR.

Figure 1 A schematic illustration of the lowest frequency eigenmode, in blue color, and the second lowest
frequency eigenmode, in red color, for the decorated membrane resonator (DMR). The rectangle
indicates the solid platelet. Here x denotes the coordinate axis in middle of the DMR’s plane, and ξ 1 ,2
denotes the displacement normal to the plane for the two eigenmodes. Adapted from ref. [13].

A resonator can be classified to have dipolar symmetry if its resonance involves centre of mass
motion. DMR is a dipolar resonator. In contrast, Helmholtz resonator [5] is monopolar in
symmetry since at its resonances involves no centre of mass movement.
Anti-resonance and the effective dynamic mass density

If we denote and as the two lowest resonance frequencies of a DMR, then the sample area-
averaged displacement response can be written as

(9)

by neglecting the higher order resonant modes. Here are positive constants and the angular

brackets denote area averaging. For an incident acoustic wave with frequency that is
intermediate between and the first term on the right hand side of Eq. (9) is negative in its
real part, whereas the second term is positive. The net behaviour is illustrated in Fig. 2. It is

4
inevitable that there is always a point where vanishes, denoted the anti-resonance
frequency. If we define a dynamic mass density [1-3, 7-9] as

, (10)

where denotes the averaged force density, and the denominator is recognized to be the area-
averaged acceleration. The behavior of is also illustrated in Fig. 2, seen to display a nearly
divergent behavior followed by a negative dynamic mass density region in which the
acceleration is opposite to the force.

Figure 2 An illustration of the displacement plotted as a function of frequency for the DMR. At a
frequency intermediate between the two resonance frequencies, there is a point at which the displacement
is zero. That point is denoted the anti-resonance frequency. The dynamic mass density as defined by Eq.
(10) is plotted as the solid red line.

Can the divergence of the dynamic mass density have a real effect, in the sense of exhibiting
consequences that imitate the effect of a large mass at the anti-resonance frequency? We show
below that indeed, the dynamic mass density effect is real.
Coupling with propagating waves and evanescent waves—total reflection at anti-resonance
Anti-resonance is associated with a special wave manipulation functionality of DMR [6]. To
understand this functionality, we have to examine DMR’s coupling to the incident and reflected
waves. The basis of this consideration is a propagating wave’s dispersion relation:

, (11)

where denotes the wavelength in air, and subscripts || and denote wavevector’s components
parallel and perpendicular to DMR’s surface, respectively. The basis of our considerations is that
the lateral dimension d of the DMR is subwavelength, i.e., . For the resonance mode

5
profiles depicted in Fig. 1, if we Fourier transform the two-dimensional mode profile to the
transverse wavevector domain, then it is clear that except for one component of the transverse
wavevector component that will be selected out for special note below, the other components
must satisfy the inequality

, (12)
because all the lateral features of the resonance profile are smaller than d. That means in order to
satisfy Eq. (11), the associated must be purely imaginary, i.e., decaying away from the DMR
surface exponentially and hence decoupled from the propagating waves. The one exception is the

component associated with the center of mass motion. Since delineates a mode in
which the whole DMR moves up and down in unison, we denote such a mode as the “piston
mode.” Hence for the DMR with subwavelength lateral dimensions, only the piston mode
couples to the propagating modes since in that case we have .

Figure 3 Results of a full-waveform simulation on the DMR, with transmission coefficient shown as the
black solid line. Full transmission is seen at the two resonances, with near-zero transmission at anti-
resonance. The divergence of the dynamic mass density at anti-resonance is shown by solid red circles,
as evaluated from Eq. (10) using simulated values for ⟨ ξ ⟩ . Adapted from ref. [8].

Because at anti-resonance , there is no piston mode and hence the DMR is decoupled
from the propagating mode, i.e., it acts like a rigid wall and can totally reflect the acoustic wave
at this frequency even though the DMR is a flimsy membrane. This effect has been
experimentally demonstrated [5] as well as numerically simulated as shown in Fig. 3.
The total reflection behavior is consistent with the divergence of dynamic mass density as
required by Eq. (10), i.e., at anti-resonance the DMR seems to acquire a very large mass density.

6
In accordance to the mass density law, the transmission amplitude T of an acoustic wave through
a solid wall with mass density and thickness h is given by

, (13)
independent of solid wall’s bulk modulus. Hence DMR’s near-total reflection at anti-resonance
is in agreement with the mass density law, by replacing the static mass density by the dynamic
mass density in Eq. (13). Another manifestation of the large dynamic mass density at anti-
resonance is the use of a very light-weight DMR in being able to successfully suppress vibration
that is tuned to the antiresonance frequency [4]. As to the negative dynamic mass density, its
effect was demonstrated in the first publication on locally resonant sonic materials [1-3, 7].
Acoustic absorption by DMR and its upper bound
Polymeric membranes usually have a low dissipation coefficient. However, absorption per unit
volume is the product of energy density multiplied by the dissipation coefficient. In acoustic
metamaterials, high absorption relies on high energy density as the result of resonance, rather
than the magnitude of the dissipation coefficient. In fact, large dissipation coefficient can cause
impedance mis-match with air, leading to increased reflection and reduced absorption as a result.
In this context a low dissipation coefficient may offer an advantage if properly utilized.
For the DMR, the largest energy density at resonance is located at the perimeter(s) of the solid
mass, where there is a kink in the mode profile due to the large rigidity contrast between the
solid and the membrane, as seen in Fig. 1. In this case there can be a very large curvature energy
density, localized at the perimeter(s), that is proportional to the square of the second derivative

along the lateral direction, i.e., . Since a kink implies a discontinuity in the first
derivative, the second derivative would resemble a delta function. Whereas the delta function can
be integrated to a finite value, the square of the delta function implies a very large value even
when integrated. A detailed finite element simulation has indeed confirmed this fact [11]. Hence
significant acoustic absorption by the DMR occurs at resonances through the curvature energy
density enhancement by the resonant mode profile.

Figure 4 An illustration of the incoming and outgoing acoustic waves on the two sides of the DMR, in
terms of the pressure modulation. Here k 0=ω /v 0.

7
While the absorption by DMR can be significant, there exists an upper bound owing to the thin
membrane thickness [12]. In order to see the inevitability of such a bound, let us consider two
counter-propagating incoming waves incident from two sides on the DMR with pressure

modulation amplitudes and . Upon reflection, they are converted into two outgoing waves

with the relevant amplitudes and . The subscripts – and + denote the left and right-hand
side regions, respectively, and superscripts i and o stand for incoming and outgoing waves,
respectively. In Fig. 4 we illustrate this incident and scattering configuration. At the surface of
the DMR, the net pressure modulation is the sum of the incoming and outgoing pressure

modulations. The time-averaged acoustic energy flux in air is given by ,

opposite in direction for the incident and outgoing waves. Owing to the small thickness of the

membrane, we must have , i.e., the membrane thickness does not change since the
membrane’s thickness-varying resonance has a much higher frequency and therefore “frozen” in
the relevant low frequency regime. As only the piston mode couples to the propagating waves
and the displacement velocity is continuous between the coupling mode and air, we must have:

. (14)

As a note on the side, for a stationary solid boundary so that the displacement velocity
has a node, such as the case at anti-resonance. The total pressure on the solid boundary, on the
other hand, is given by . Hence for a stationary wall the pressure exerted on the boundary
is twice that of the incident wave.
Conservation law in analogy to momentum conservation

Equation (14) leads directly to the conservation law for the mean pressure modulation [12]:

. (15)
Equation (15) is completely similar to the momentum conservation law before and after the

collision between two identical particles. Here we treat as complex numbers, or phasors,
i.e., 2D vectors in the complex plane. The analogue to the kinetic energy is the energy flux
.

It follows from analogy to classical mechanics that a wave incident from one side is just like the
collision between a moving particle and a stationary one, with equal mass. In that case half of the
incident kinetic energy is the conserved center of mass energy and therefore not available for
dissipation. Hence the upper bound for DMR dissipation is 50%.

8
Hybrid structure and total absorption
From the simple argument that leads to the absorption upper bound for incidence from one side,
it also becomes clear that if there are incident waves from both sides of the membrane, then
100% absorption is entirely possible. However, even if we have incidence only from one side, it
is still possible to have backscattered wave from a reflecting surface on the other side. Hybrid
resonance is the result of such considerations when the reflecting surface is placed in the near-
field region (i.e., at a very subwavelength distance) to the DMR so that the resonance pattern of
the DMR is altered as well [13]. In Fig. 5(a) we show such a configuration, denoted the hybrid
structure, in which the DMR is subject to incident and scattered waves on the right, and , as
well as similar waves on the left, and ,. However, the two waves on the right are subject to
the reflecting boundary condition imposed by the hard wall placed at a distance s away from the
DMR. Due to the hard wall boundary condition, the displacement velocity must have a node at
the reflecting boundary. That means at the position of the DMR, the two waves on the right must
have equal amplitude, i.e., , with a phase difference , where
. The phase of can be taken to be zero as the reference.

Figure 5 (a) A pictorial illustration of the hybrid resonator structure. (b) A phasor diagram that
illustrates Eq. (16) geometrically. Figure (a) adapted from ref. [13].

Imposition of the no-reflection condition

We wish to see if it is possible to achieve total absorption by the hybrid structure. Total
absorption implies . Hence from Eq. (15), we have

, (16)
which can be represented by a phasor diagram as shown in Fig. 5(b) in which Eq. (16) is
regarded as a two dimensional vector equation in the complex plane, forming a closed triangle.
Here the phase is treated as giving the relative angle(s) between the vectors.

9
The net area-averaged pressure on the DMR is given by the difference between the two sides of

the DMR: . From Fig. 5(b), it is clear that . Hence

, and it follows that with the impedance condition imposed by , i.e.,

no reflection, the DMR impedance for total absorption needs to be

. (17)
At the same time, we recognize that the impedance at the + side of the DMR, representative of
the air cavity impedance that is in series with the DMR impedance, is precisely given by

. (18)

When , in agreement with that of a hard reflecting wall. It follows from Eqs.
(17) and (18) that the total impedance of the hybrid structure is given by

. (19)
Since the reflection coefficient at normal incidence is given by

, (20)
hence consistency is achieved with the imposed no-reflection condition. We also note that from
energy conservation the absorption is given by when there is no transmission, it
follows that implies total absorption.
Emergence of hybrid resonance

A special situation occurs at , with n=0,1,- -, or . This is the

“drum resonance” condition. However, if the reflecting wall is very close to the DMR, then it
may seem impossible for total absorption to occur, unless DMR can exhibit an impedance that
has a very large negative imaginary part. It turns out that precisely such an impedance can occur
for the DMR at a frequency that is slightly less than the anti-resonance frequency. Just to make
this point intuitively obvious, we note that from Eq. (7) and (10), near the anti-resonance the
impedance of the DMR can be expressed as

, (21)

10
apart from some constants. Since the dynamic mass density at frequencies less than the anti-
resonance frequency can be very large and positive, it follows that the condition imposed by the
total absorption may indeed be realized close to the anti-resonance frequency. Such total
absorption can be realized with significantly subwavelength sample thickness, thereby realizing
the advantage of resonance-type acoustic metamaterials.
Mathematics of the hybrid mode
In a hybrid structure, the vibration modes of the DMR are altered in a manner such that the
original modes are replaced by the hybrid resonances [13]. To see how the hybrid resonance
actually emerge, we recall that the Green function can be expanded in terms of the
eigenfunctions of the system. In the vicinity of the anti-resonance the Green function can be
expressed as the sum of two Lorentzians:

, (22)

where denotes the displacement profile of the nth eigenfunction,

, (23)

denotes the area of the membrane, and is the area mass density, i.e., mass per unit area
as a function of the planar coordinate. Consider the case where in the relevant
frequency range, and for simplicity let . Then we can expand Eq. (22) as

. (24)

This expansion is to facilitate the consideration that at the anti-resonance frequency ,

. If we express the hybrid resonance frequency to be at , where

, then

, (25)
where

. (26)

11
From Eq. (7), we obtain from Eq. (25)

. (27)

If , then the impedance diverges towards negative infinity when . What we want,
however, is for the real and imaginary parts of Eq. (27) to agree with those of Eq. (17):

, (28a)

. (28b)

These two conditions can certainly be satisfied since there are three adjustable parameters: ,
, and . Since is a material parameter set by the material used, usually the tuning is carried
out by adjusting the other two.

Figure 6 (a) Numerical simulation of the hybrid mode, shown as the black line, plotted together with the laser
vibrometer measurement of the hybrid mode’s displacement, shown as empty red circles. It is seen that the ratio of
the displacement to the incident wave’s particle displacement, ξ s , can be more than a factor of 20. Since hybrid
resonance mode is impedance-matched to that of air, ξ s is approximately equal to the piston mode component of
the hybrid resonance’s displacement profile. (b) Measured (plotted as red circles) and simulated (plotted as black
line) absorption as a function of frequency. Total absorption is seen at 152 Hz. The solid arrow on the left indicates
the frequency of the lowest resonance, the arrow with dashed line on the right indicates the anti-resonance
frequency. Adapted from ref. [13].

Profile of the hybrid resonance mode

12
At anti-resonance frequency we have the condition . Since the hybrid
resonance frequency is only slightly less than , at the hybrid resonance we must have

but . If in addition we impose the condition that the hybrid

resonance’s piston component amplitude must match that of the incident wave, , in order to
achieve impedance matching, then the following condition should hold:

. (29)

As the right hand side of Eq. (29) is a small number, it follows that , , the
piston mode amplitude of the hybrid mode. The implication is that the evanescent component of
the hybrid resonance has a much larger displacement amplitude than that of the piston mode. In
other words, for the hybrid resonance the variance of the displacement can be very large, i.e.,

. This has indeed been verified experimentally as seen in Fig. 6(a),

together with the total absorption at hybrid resonance frequency of 152 Hz as shown in Fig. 6(b),
both in excellent agreement with the theory simulation predictions [13].
In the experiment, the sample thickness is noted to be more than two orders of magnitude smaller
than the wavelength at 152 Hz, 2.3 m.

II.3 Broadband optimal acoustic absorption structures

Resonance is inherently a dispersive, narrow frequency-width effect. In considering the practical
applications of resonance-type acoustic metamaterials, this feature is a drawback since for most
applications broadband functionality is a necessity. Therefore how to achieve broadband
functionality by integrating resonators with different resonance frequencies is a crucial concern
for the development of acoustic metamaterials into useful products. Associated with this issue is
the question whether acoustic metamaterials can still retain its advantage of subwavelength
sample size in the broadband regime. We will see in this section that in regard to broadband
acoustic absorption functionality, the definition of metamaterial’s advantage, as compared to
conventional absorption materials, is altered from the previous “subwavelength sample size.” We
elaborate below.
Noise absorption is one of the most important application areas for acoustic metamaterials [16].
A thick-enough conventional acoustic absorbing materials, such as those used in anechoic
chambers, can absorb acoustic waves of all frequencies; but the bulky unwieldiness limits their
broad applications. In this regard an optimally integrated acoustic resonator structure can display
the advantage of being able to target the spectrum range of peak noise, with a minimum sample

13
thickness allowed by natural law—the causality constraint on minimum sample thickness
associated with a given absorption spectrum.
Causality constraint on minimum sample thickness
Reflection and transmission of an incident wave are denoted the sample response to an incident
wave. Since the incident wave varies as a function of time, the sample response will similarly
acquire a time dependence. Causality is defined to mean that the response at any given moment
can only depend on what transpired prior to that moment, but not on anything that happens after
that moment, i.e., the future cannot affect what is happening now. When translated into
mathematical language, this simple and intuitive statement led to some marvellous results. The
most famous one is the Kramers-Kronig relation, taught in almost all the standard
electrodynamics textbooks, that relates the real and imaginary parts of the electromagnetic
dielectric constant in the frequency domain. A less-known consequence is the inequality
constraint that relates the absorption spectrum to a minimum sample thickness. Since its
derivation can be found in the literature [14] and is beyond the scope of this tutorial, we just state
it below as a simple inequality:

, (30)

where d is the sample thickness, is the absorption spectrum, denotes the effective bulk
modulus of the sound absorbing structure in the static limit, and is the bulk modulus of air.
Two implications of Eq. (30) are (a) low frequency absorption inherently requires a larger
sample thickness as compared to absorption at higher frequencies, and (b) any given absorption
spectrum is associated with a minimum sample thickness. We denote “optimal” those absorption
structures or materials that can nearly attain equality with .

The causality constraint serves as the reference for judging the “degree of success” of a
broadband absorption material/structure. It also tells us that for a given sample thickness, there is
only a finite amount of wave absorption resources available. Hence how to allocate the
absorption resources, in the form of an absorption spectrum, is an issue of importance. If the aim
is to absorb noise, then the best allocation strategy is naturally to match the absorption spectrum
with the noise spectrum. The freedom offered by an integrated array of acoustic resonators in
realizing a target absorption spectrum is therefore the main advantage over the conventional
absorption materials, which are otherwise excellent in their acoustic absorption capabilities.
Since the low frequency noise in the range of less than 500-1000 Hz is the most difficult to
absorb with a reasonable sample thickness, it is in this range that acoustic metamaterial can be
most effective.
Fabry-Perot resonators

14
An acoustic Fabry-Perot (FP) resonator is just a hollow pipe with subwavelength cross sectional
dimension. One end of the pipe is closed with a hard reflecting boundary; the other end is open.
The boundary condition at the closed end—displacement velocity node—means that the
displacement velocity has the following functional form:

, (31a)

where is the length of the FP resonator, is the amplitude of incident wave’s displacement
velocity, and z=0 is taken to be the position of the open mouth end of the resonator. Here the
time dependence, , is always implied. From Eq. (31a) and Eq. (3), we obtain the
pressure modulation as

. (31b)
The impedance at the mouth of the FP resonator as perceived by the incident wave, is given by

, (32a)

which is identical to Eq. (18) with . Here we use the lower-case z to denote the
impedance of a single FP resonator, in anticipation of later development below where we use Z
to denote the impedance of a sample comprising an array of integrated FP resonators. Similarly,
we use lower-case g to denote the Green function of a single resonator so that

. (32b)
The upper-case G will be used for the Green function of an integrated array of FP resonators.
The fact that a hollow pipe can be a resonator, i.e., with the Lorentzian form, becomes clear with
the following mathematical identity:

. (33)
By comparing the arguments of cotangent in Eq. (32) with the tangent in Eq. (33), we obtain
. In Eq. (33) indicates the condition for a resonance, hence
are the FP resonance frequencies. The lowest resonance
frequency , with m=1, corresponds with the condition , i.e.,

15
 . (34)
By adding a vanishingly small imaginary part in the denominator of Eq. (33), we obtain after
some re-arrangment the following expression:

, (35)
which is in full accordance with the Lorentzian expression for multiple resonances in a FP
resonator. From Eq. (8), Eq. (35) can be further reduced to the form

. (36)
Integration scheme with a continuum of FP resonators to achieve tunable absorption spectrum
In order to obtain broadband absorption, integration by using an array of FP resonators is a
necessity. However, using FP resonators with equally spaced (lowest order) frequencies is not
the best strategy to obtain, for example, a flat absorption spectrum. Instead, there is an
integration strategy [14, 15] that can best achieve a target absorption spectrum, shown below.

Consider a continuum of in an idealized FP resonator array. The lateral size of the array will be
assumed to be smaller than the relevant wavelength in the following discussion. Since these
resonators are arranged in parallel, the array impedance is the sum of the inverse of the
individual impedances. In addition, we assume that the FP resonators are separated by hard
reflecting surfaces that occupy a fraction of the total surface area exposed to the incident
wave. Since the inverse of the hard reflecting surface impedance is zero, hence the FP resonator
array has a real part of the inverse impedance given by

, (37)

where is a lower cut-off frequency as required by the causality constraint in order for the
array to have a finite sample thickness. Here we have inserted a mode density

, (38a)

, (38b)

in the integral. Here n is treated as a continuous real number which can be viewed as the index
of N resonators to span a finite, fixed frequency range. As N approaches infinity, N . We

16
will see that the determination of the mode density is the central task of the integration scheme,
with the goal of attaining the target absorption spectrum.
In Eq. (37) we have purposely ignored the imaginary part of the inverse impedance because that
part is oscillatory as a function of and therefore its integrated effect will be minimal.
However, the imaginary part of the inverse impedance will be fully taken into account after the
mode density is determined, so that the full integration of the Lorentzian form can be carried out
with the complex denominator. It will be seen that the effect of including the imaginary part is to
introduce a smooth transition region above the cutoff that tends exponentially to the designed
absorption spectrum.
By carrying out the integration in Eq. (37), we obtain a simple expression:

. (39)
At this point it is instructive to first illustrate the determination of mode density from the target
absorption spectrum by ignoring all terms on the right-hand side of Eq. (39) with m , i.e.,
retaining only the lowest-order FP resonances. From Eq. (38a) and the condition , we have

, (40)

where the quantity ( ) is treated as the known input, determined from the target absorption

spectrum . Here we would like to treat the simplest case of

total absorption above the cutoff frequency, by letting . In that case the solution of Eq.
(40) is an exponential:

, (41)

where we used the initial condition at n=0. It should be noted that in accordance to the
inverse relationship between and , Eq. (41) or its improved version by including all the
higher-order FP resonances (see below), determines FP resonators’ length distribution provided
is known. In anticipation of later development, here we mention that there is indeed an optimal
value of , fixed by using the causality constraint’s minimum sample thickness .
It should be noted that if the target absorption spectrum is not total absorption, then Eq. (40)
represents a somewhat more difficult differential equation to solve, but should be able to be
solved numerically if the given is frequency dependent.

17
Correction to the resonator frequency distribution by including higher-order FP resonances
Equation (39) can be written in the form of

, (42a)
where the left hand side is treated as known. By comparison with Eq. (39), it can be seen that

, (42b)
and since there is no mode density below , we have the condition

, if . (42c)

Provided that the left-hand side of Eq. (42b) can be solved in terms of the input , then the
distribution of resonator frequencies can be solved in terms of a first order differential equation.
Below we illustrate the solution of this problem in the case of total absorption, i.e., .

Figure 7 The value of solved a (Ω) plotted as a function of Ω/ Ω c.

In view of the factor in the series, Eq. (42a), the solution proceeds in discrete frequency
segments. For , we have 1= since for with m>1
in accordance to Eq. (42c). Therefore for , denoted as
. For , we have .
Therefore for , denoted as . For we have
. Therefore
for , denoted as . In every frequency segment all terms of the
series except the first one, m=1, are either zero as required by the condition set by Eq. (42c), or
fixed by the previous frequency segments. Since for m=1 we have , hence the value of

18
 for all frequencies can be uniquely fixed. This is shown in Fig. 7. Since in the mth
frequency segment is a constant, the differential equation has the simple form

. (43)
The solution for the total absorption case is therefore given by

, , (44)

where max[n]= .

Figure 8 Logarithm of the resonator frequency plotted as a function of . The dashed line indicates the exponential
relation obtained by ignoring all the higher order FP resonances. The solid red circles indicate the best choice for
the (lowest order) FP resonator frequencies for a 3 by 3 array.

In Fig. 8, we plot as a function of n for an arbitrary value of 0.98. It is seen that

the correction by including all the higher order FP resonances has the effect of making
resonator’s frequency distribution super-linear in the logarithmic plot. Here the red dots indicate
the best discrete choice of the resonator frequencies to achieve total absorption.
Again, if the target absorption spectrum differs from total absorption, then the solution to Eq.
(42) and (43) needs to be modified to take into account the frequency dependence of the input
.
Inclusion of the imaginary part of the impedance

19
Figure 9 (a) The real (in red) and imaginary (in blue) parts of the impedance, plotted as a function of frequency. (b)
Absorption coefficient evaluated from the impedance shown in (a). Imaginary part of the impedance is shown to
have a minor effect above the cutoff frequency. Adapted from ref. [14].

In the pursuit of the best design for the resonator integration we have so far ignored the
imaginary part of the impedance as pointed out above. Here we return to Eq. (35) to carry out the
integration over :

, (45)

where and by the definition of a lower cutoff, . For the total absorption
case, the numerator inside the integral is equal to 1 in accordance to the designed resonance
frequency distribution. By using Eq. (8), the resulting integration can be done exactly to yield the
impedance expression

, (46a)

. (46b)

In Eq. (46b) we have taken in order to cancel out the real part of the impedance as it
should vanish, since there is no mode density below the cutoff frequency. The real and
imaginary parts of the impedance are plotted in Fig. 9(a). The resulting absorption spectrum is

plotted in Fig. 9(b), where . It is seen that indeed the

effect of the imaginary part decays quickly above the cutoff frequency as stated earlier but
induces a smooth exponential approach to the target absorption spectrum.

Circle of consistency and the determination of optimal

20
 Figure 10 The circle of consistency that closes the loop. Adapted from ref. [15].

In the above, sample thickness is only implicitly implied by the distribution of , as the FP
resonator’s length is specified by Eq. (34). If the longer FP resonators can be folded so that the
overall shape of the sample is a compact cuboid, then its thickness should be given by the
average of all the FP resonator lengths as dictated by volume conservation. We denote the
average as . As is present in the distribution, Eq. (41) or (44), is therefore a function of
. By equating , the value of can be explicitly determined. We illustrate below this
process in the simple case of Eq. (41) where only the lowest-order FP resonances are taken into
account. From Eq. (34) and (41), we obtain from the the following condition:

. (47)

On the other hand, we can use Eq. (46) and (30) to calculate , where we have
used the effective medium expression . This leads to the optimal value

. (48)

A similar optimal value of may be obtained by considering the correction due to the
higher order FP resonances. This final piece of design strategy completes that “circle of
consistency” that may be depicted by Fig. 10.

Self-energy correction

Consider a square array of a finite number of FP resonators, each with a different . Let
denote the Green function of the th resonator. The pressure at the mouth of the resonators must
be different in accordance with Eq. (31b) when excited, e.g., at a frequency intermediate between

21
two (lowest order) resonances. With a pressure difference, there must be air flow. This implies
the different resonators to be essentially interacting with each other, i.e., they constitute an
interacting system. When that happens, the Green function of each resonator will be
renormalized with the consequence that the resonance frequency of each resonator is shifted
downward somewhat. The kernel of such interaction renormalization is called self-energy [14],
adapted from the Dyson equation terminology for a multi-entity interacting system. Also, the
lateral oscillating air flow implies evanescent waves that decay away from the array surface, and
such evanescent waves with their attendant lateral air flows can be utilized to smooth out the
absorption spectrum of an array with an insufficient, finite number of resonators; e.g., by putting
a thin layer of acoustic sponge over the surface of the array.

To obtain an explicit expression for the self-energy, we observe that the displacement velocity
at the mouth of the th resonator can be expressed as

, (50)

where is the modulation pressure averaged over the array surface, and denotes the local

deviation from the average, arising from the evanescent modes. Hence . Since
collectively ’s are associated with evanescent waves that exponentially decay away from the

array surface, one should be able to express in the wavevector domain by using

the normalized Fourier basis . Effectively, that means

. Here the wavevectors are discretized by the condition that

integrated over the array surface must vanish, with index ,

. That means , where L denotes the side length of the square array. The
reason for expressing in the wavevector domain is to obtain the z variation of the evanescent
mode. Since the lateral size of the array is smaller than the relevant incident wavelength, we

must have > . Hence it follows from Eq. (11) that is

indicative of the exponential decay length of the evanescent wave. It follows that

. (51)
The z variation of pressure modulation implies that from Eq. (3), we have at z=0:

, (52)

22
where . The Fourier component can be solved in terms of

by using the orthogonality property of the Fourier basis. The result is

. (53)

We observe that the integral of must vanish, owing to the condition imposed on

the Fourier basis, hence can be replaced by in Eq. (53). By substituting Eq. (53)
into Eq. (51) and setting z=0, we obtain

. (54)
Equation (54) can be easily expressed in discretized form as

, (55)

with , being the discrete indices of the resonators in the array. Just to be self-contained, we
give below the expression for :

, (56)

where is the total number of resonators in the array, and the integrals are to be carried out over

the cross sectional areas of the mouths for resonators and , with , being their
respective center positions. Here denotes the mouth area of resonator , and we have assumed
. By carrying out the integrations, we obtain

23
 . (57)

Equations (50) and (55) are the starting point of an infinite series as can be seen as follows. By
substituting Eq. (55) into Eq. (50) one obtains

. (58)
But then one can substitute Eq. (50) into the right hand side of Eq. (58) to obtain:

. (59)
At this point it is already clear that by repeated, alternating substitutions, one can obtain the
following series:

, (60)

where we have separated out the diagonal terms in the matrix. The off-diagonal terms are
grouped into the term , with not equal to . It is found that is order(s) of magnitude
smaller than the diagonal terms, hence it can be neglected. By retaining only the diagonal terms,
we obtain

. (61)

For the incident wave, the perceived effective impedance for the th oscillator is given by

, (62)

24
where we have defined to be the renormalized effective Green function for the th oscillator.
Comparison of Eq. (62) to Eq. (61) leads to the Dyson equation:

, (63)

from which we identify the self-energy correction to be . Since the resonance frequency

is generally identified by the vanishing of the real part of , it follows that the self-energy
renormalization has the effect of shifting the resonance frequency to where the real part of

vanishes. It should be noted that in accordance to Eq. (57), if , then .

Therefore the self-energy correction is inherently an effect associated with a finite number of
resonators.
Since the resonators’ impedances are in parallel, the sample impedance for the array is given by

, (64)

where is the area fraction occupied by the mouth area of the resonator, with . If the
resonators have the same cross sectional area, then so that

. (65)
Experimental verification

Figure 11 (a) A pictorial illustration of the sample (left panel) and a photo image of the actual 3D-printed sample
with four arrays. (b) Measured (shown by empty red circles) and theory (shown by black solid line) absorption
spectrum of the sample shown in (a). The dashed red line indicates the target absorption spectrum used to design

25
the sample. Deviation from the target spectrum can be attributed to the finite number (16 in the present case) of FP
resonators. The sample thickness, 10.9 cm, is only ~3 mm over the causality minimum. Adapted from ref. [15].

In Fig. 11(a) we show a schematic picture and a photo image of the sample. The array has 16
resonators whose ’s are designed in accordance with the total absorption target spectrum with a
cutoff frequency set at 345 Hz. The photo image actually shows 4 such arrays, each one in the
shape of a compact cuboid that is 10. 6 cm in thickness and approximately 2.3 cm by 2.3 cm in
cross sectional dimension. The longer FP channels were folded so that the cuboid thickness is
very close to . Since 16 resonators were insufficient to produce a flat absorption spectrum, a 3
mm thick sponge was put on top of the array surface in order to utilize the lateral air flows of the
evanescent modes for filling the dips in the absorption spectrum. The absorption results
measured by using the impedance tube are shown as open circles in Fig. 11(b). The idealized
target spectrum is shown by the red dashed line, and the theory prediction from Eq. (65), by
using parameter values detailed in Ref. ( ), is shown as the solid red line. Excellent agreement is
seen. From the experimental absorption spectrum, the minimum sample thickness predicted by
Eq. (30) is 10.55 cm (by using the effective medium expression for ), which is only
about 3 mm less than the actual sample thickness (including the acoustic sponge). It should be
noted that while the ideal value of is 0.982 as mentioned previously, the actual value for the
sample shown is 0.8. However, in the design of the FP channel lengths the 0.982 value was used,
and this turned out to be the most important. The lower actual value of only has the effect of
reducing the reflection loss from 20 dB to 15 to 17 dB in the frequency range of 1500 to 3000
Hz.

Figure 12 Measured (shown by empty red circles) and theory (shown by black solid line) absorption spectrum
design in accordance to the target absorption spectrum indicated by the red dashed line, using 16 FP resonators.
Here the sample thickness, 9.33 cm, is only 3 mm over the causality minimum. Adapted from ref. [15].

As a further example to demonstrate the tunability of the absorption spectrum, we select a target
spectrum as shown by the dashed line in Fig. 12, with a reflection window extending from 560 to
1000 Hz. The FP resonator length is now designed by solving the Eq. (42) and (43) with an input
that is different from that for total absorption. This particular sample also has 16 FP resonators.
Experimentally measured results and theory predictions by Eq. (65) are shown in Fig. 12 by open
circles and solid lines, respectively. It is apparent that 16 resonators in an array are insufficient to

26
closely mimic the target spectrum, but the resemblance is clear. In the present case the designed
, and the causal minimum thickness is 9 cm, which is 3 mm less than the actual sample
thickness of 9.33 mm.
Heuristic understanding of total absorption by resonator array
In the experimental results presented above, the array has only 16 resonators, far from the
idealized continuum model. In this array a single resonator occupies a small fraction of the total
area exposed to the incident wave. It follows that at its resonance frequency, only that area
fraction is absorbing. The question is: If that is the case, how can total absorption be achieved?
The answer to this question can have two different aspects. In Fig. 13 we show the physical
picture aspect of the answer. Similar to the flow streamline in a water sink, the Poynting vector
of the incident acoustic wave energy starts to deviate from the incident wavevector at a distance
above the array’s surface and converges toward the resonant FP resonator. The underlying reason
for this physical picture is that a discrete resonance always has high density of states to
accommodate the wave energy at that frequency. Below we show the second, heuristic
mathematical aspect of the answer to this question.

Figure 13 A pictorial illustration of the Poynting vector of the incident wave at the resonance frequency of one of
the resonators in the array. Close to the surface of the array, the power flow converges onto the resonating unit,
leading to total absorption.

Figure 14 A pictorial illustration of the slight frequency shift that can lead to the impedance matching condition as
described in the text.

27
In accordance to Eq. (64), total absorption at the frequency of the resonator is given by the

condition , where C denotes some constant. The imaginary part of the Green
function has a peaked shape as illustrated schematically in Fig. 14, where the red dashed line
indicates the impedance matching condition for total absorption. If the frequency of the

intersection between and is denoted by as delineated by a cross, then due to the

presence of the factor total absorption obviously can only be achieved at another frequency

indicated by an open circle, so that when the magnitude of at is multiplied by ,

the result exactly meets the red dashed line. One can see that obviously if is too small, i.e., the
number of resonators is very large so that each occupies only a very small fraction of the total
exposed area, then such a “mechanism” of frequency shift (from to ) would not work.
However, in that case the frequency difference between the resonators becomes small. Hence
when one resonator is excited by the incident wave, it is inevitable that some other resonators
with resonance frequencies in the vicinity of the incident wave frequency will also be excited in
accordance to the Lorentzian form. When that happens, total absorption at a particular frequency
is no longer shouldered by a single resonator. Rather, it is achieved by a collection of resonators,
each with some degree of “frequency shift.”
Jensen, please insert your part here.
References
1. Z. Liu, X. Zhang, Y. Mao, Y. Y. Zhu, Z. Yang, C. T. Chan, P. Sheng, Locally Resonant Sonic Materials.
Science 289, 1734-1736 (2000). It should be noted that in this initial publication, the novel effect
was mis-attributed to the negative modulus, rather than to the dynamic mass dispersion. It was
subsequently corrected in Ref. (2) and (3).
2. Ping Sheng, X. X. Zhang, Z. Liu and C. T. Chan, Locally Resonant Sonic Materials, Physica B 338, 201-
205 (2003).
3. Z. Liu, C. T. Chan, P. Sheng, Analytic Model of Phononic Crystals with Local Resonances. Phys. Rev. B
71, 014103 (2005).
4. L. Sun, K. Y. Au-Yeung, M. Yang, S. T. Tang, Z. Y. Yang and Ping Sheng, Membrane-type Resonator as
an Effective Miniaturized Tuned Vibration Mass Damper, AIP Adv. 6, 085212 (2016).
5. N. Fang, D. Xi, J. Xu, M. Ambati, W. Srituravanich, C. Sun, X. Zhang, Ultrasonic Metamaterials with
Negative Modulus. Nature Mater. 5, 452-456 (2006).
6. Z. Yang, J. Mei, M. Yang, N. H. Chan, P. Sheng, Membrane-Type Acoustic Metamaterial with
Negative Dynamic Mass. Phys. Rev. Lett. 101, 204301 (2008).
7. S. H. Lee, C. M. Park, Y. M. Seo, Z. G. Wang, C. K. Kim, Acoustic Metamaterial with Negative Density.
Phys. Lett. A 373, 4464-4469 (2009).
8. Ping Sheng, J. Mei, Z. Liu and W. Wen, Dynamic Mass Density and Acoustic Metamaterials, Physica B
394, 256-261 (2007).

28
9. J. Mei, Z. Liu, W. Wen, P. Sheng, Effective Dynamic Mass Density of Composites. Phys. Rev. B 76,
134205 (2007).
10. D.-Y. Maa, Potential of Microperforated Panel Absorber. J. Acoust. Soc. Am. 104, 2861-2866 (1998).
11. J. Mei, G. Ma, M. Yang, Z. Yang, W. Wen, P. Sheng, Dark Acoustic Metamaterials as Super Absorber
for Low-Frequency Sound. Nature Commun. 3, 756 (2012).
12. M. Yang, Y. Li, C. Meng, C. X. Fu, J. Mei, Z. Y. Yang, Ping Sheng, Sound Absorption by Subwavelength
Membrane Structures: A Geometric Perspective, C. R. Mecanique, 343, 635-644 (2015).
13. G. Ma, M. Yang, S. Xiao, Z. Yang, P. Sheng, Acoustic Metasurface with Hybrid Resonances. Nature
Mater. 13, 873-878 (2014).
14. M. Yang, S. Y. Chen, C. X. Fu and Ping Sheng, Optimal Sound-absorbing Structures, Mater. Horiz., 4,
673-680 (2017).
15. M. Yang and Ping Sheng, An Integration Strategy for Acoustic Metamaterialsto Achieve Absorption
by Design, Appl. Sci., 8, 1247 (2018).
16. M. Yang and Ping Sheng, Sound Absorption Structures: From Porous Media to Acoustic
Metamaterials, Annu. Rev. Mater. Res. 47, 83–114 (2017).
17. G.C. Ma and Ping Sheng, Acoustic Metamaterials: From Local Resonances to Broad Horizons, Sci.
Adv. 2, 83-114 (2016).

29