betaFeedback

WorksPhysics and Astronomy

Isospectral open cavities and gratings

Abstract

Open cavities are often an essential component in the design of ultra-thin subwavelength metasurfaces and a typical requirement is that cavities have precise, often low frequency, resonances whilst simultaneously being physically compact. To aid this design challenge we develop a methodology to allow isospectral twinning of reference cavities with either smaller or larger ones, enforcing their spectra to coincide so that open resonators are identical in terms of their complex eigenfrequencies. For open systems the spectrum is not purely discrete and real, and we pay special attention to the accurate twinning of leaky modes associated with complex valued eigenfrequencies with an imaginary part orders of magnitude lower than the real part. We further consider twinning of 2D gratings, and model these with Floquet-Bloch conditions along one direction and perfectly matched layers in the other one; complex eigenfrequencies of special interest are located in the vicinity of the positive real line and further depend upon the Bloch wavenumber. The isospectral behaviour is illustrated, and quantified, throughout by numerical simulation using finite element analysis.

Introduction

Acoustic metasurfaces, which have subwavelength thickness often constructed from local resonating structures, have found wide application to sound absorbing surfaces and surfaces that manipulate sound in often remarkable ways [1, 2]; the performance of these surfaces is due to the resonators embedded in the surfaces and the resonance frequencies are directly connected to the geometry and volume of the resonant cavity with low frequencies requiring large volume. Many approaches have been taken to ensure metasurfaces remain ultra thin, even at low frequencies and approaches such as spiral or labyrinthine resonators are popular to save space and give added functionality [3]. An approach, explored here, is to design smaller cavities that have matching wave frequency spectrum to larger reference cavities by designing the material properties within the resonator accordingly. We show that one can go further than simply match a single resonant frequency and one can match (i.e. twin) the entire wave frequency spectrum and hence match the entire behaviour of the cavities at all frequencies. Cavities with this perfect matching are called isospectral cavities and in a mathematical sense are defined as cavities with different shapes that exhibit identical eigenfrequencies, and their design presents an intriguing challenge in the field of wave physics. The ability to design such twinned cavities opens up new possibilities across various disciplines, including acoustics, electromagnetism, or water waves. By achieving isospectrality, two distinct resonators can exhibit indistinguishable wave behaviour, enabling applications such as creating rooms or auditoria with different geometries that possess identical sound characteristics or engineering elastic components with shared vibrational eigenfrequencies.

The history of isospectral problems can be traced back to the question famously posed by Mark Kac regarding whether one can hear the shape of a drum [4]. The investigation of isospectral drums, where the Laplacian operator within closed domains with Dirichlet boundary conditions yields identical spectra for distinct regions sharing the same area, has produced significant results for specific cases and subsets of the problem [5]. These isospectral problems in bounded domains are closely linked to inverse problems in open space [6], forming a rich area of research in the past. Furthermore, there has been a recent interest in the study of isospectral or quasi-isospectral potentials inspired by supersymmetric transformations as applied to electromagnetism [7]. We proceed to utilise Transformation Acoustics (TA), which has been extensively applied for manipulating wave propagation in diverse physical fields sharing the same analytical structure, like electromagnetism [8], acoustics [9], elasticity [10, 11, 12], and many others fields [13]. Perhaps the most striking and well-studied effect enabled by TA is cloaking, allowing perfect concealment of an arbitrary object using a singular transformation. However, another strategy proposed in the work by Li et al. [14], the so-called carpet cloak, requires non-singular transformations, making this route more suited for the practical implementation of the equivalent properties. Twinning closed cavities through TA has been recently developed by Lenz et al. in [15], where the discrete spectrum of a closed domain with Dirichlet boundary conditions is successfully matched. Here we consider unbounded open domains; this is not straightforward as spectral problems for open cavities allow for the leakage of energy into the unbounded medium, have complex eigenspectra and further complications for both theoretical and numerical aspects; we are unaware of attempts to achieve isospectral domains in open systems in wave physics and this opens the way to, for instance, twinning optical waveguides.

In primis, the eigenvalue problem of an unbounded open domain gives rise to complex valued eigenfrequencies ω\omega, whose real part gives the resonant frequency and imaginary part describes the radiation losses through the unbounded region. One can distinguish eigenmodes in two categories: on one hand, the ones linked to a localised resonance within the cavity are called trapped modes and they are characterised by a high quality factor Q=ℜ(ω)/2∣ℑ(ω)∣Q = \Re(\omega)/2|\Im(\omega)|. On the other hand, the so-called leaky modes or quasi-normal modes [16] exhibit higher damping in time, which gives rise to an exponential divergence in space, making them more challenging to compute numerically [17]. In practice, it is not clear how sensitive the matching of such modes can be in terms of both real and imaginary part, particularly when those are of different magnitude. Moreover, the eigenspectra of open cavities show both continuous and discrete branches [18, 19] and a non-singular transformation is required so that their topology is preserved [20, 21].

Finally, from a numerical standpoint, the computation of resonances in open systems is complicated by the reflection of the leaky waves at the necessarily finite-grid boundaries. To tackle this issue, we use Perfectly Matched Layers (PML), whose analysis and performance for spectral problems have been extensively documented in the literature, see, e.g., [22, 23, 24, 25, 26].

The manuscript is organised as follows: § 2 briefly outlines the main transformation acoustics tools and the notation adopted. In order to highlight the main effects and underlying ideas of the twinning process in a simple setting, we then give the closed form solution for two one-dimensional domains. In § 3, we study the spectrum of an open cavity whose geometry allows both leaky and resonant modes. The transformed region is then further discretised using a layered medium that approximates the required anisotropic properties to demonstrate that conceptually one can simpler building blocks to achieve the desired effects. Section 4 extends the method to the periodic case: we show that a saw-tooth grating can be matched by a flat, thinner metasurface by comparing the dispersion diagrams for the original and transformed elementary cell. Finally, in Section 5 we draw some conclusions and discuss potential extensions of the present work.

Twinning through transformation

We take a domain filled with an acoustic fluid supporting a pressure field, for time-harmonic waves, that satisfies Helmholtz’s equation, where the filling fluid is characterised by a density ρ0\rho_0 and a bulk modulus κ0\kappa_0. The same procedure can be extended to other areas of wave physics sharing the same analytical structure i.e. anti-plane shear waves in elasticity and polarised waves in electromagnetism; we choose to use the setting of pressure acoustics without loss of generality.

Transformation Acoustics [8, 10, 9] shows how to project a subdomain Σc\Sigma_c into a deformed subdomain σc\sigma_c using the map χ:X↦x\chi: X \mapsto x while preserving the acoustic behaviour in the remaining subdomains, if particular properties are chosen inside σc\sigma_c, and we follow this approach here. We indicate with capital letters XX, PP, and Ω\Omega the set of coordinates, the pressure field, and the frequencies respectively, defined on the undeformed (or material) domain Σ\Sigma; lowercase letters xx, pp, and ω\omega that refer to the deformed (or spatial) domain σ\sigma.

The generalised eigenvalue problem

−∇X⋅(ρ0−1∇XP)=Ω2κ0−1P,(1)-\nabla_X \cdot\left(\rho_0^{-1}\nabla_X P\right) = \Omega^2\kappa_0^{-1}P, \tag*{(1)}

written with respect to the material domain Σ\Sigma, is equivalent to

−∇x⋅(ρc−1∇xp)=ω2κc−1p,(2)-\nabla_x \cdot\left(\rho_c^{-1}\nabla_x p\right) = \omega^2\kappa_c^{-1}p, \tag*{(2)}

that describes the pressure pp with respect to the spatial domain σ\sigma. Here, ρc\rho_c is a tensor describing the equivalent anisotropic density and κc\kappa_c is the equivalent bulk modulus after transformation. They are defined as

ρc:=ρ0JJTdet⁡J,κc:=κ0det⁡J,(3)\rho_c := \frac{\rho_0\mathbf{J}\mathbf{J}^{\mathsf{T}}}{\det\mathbf{J}}, \qquad\kappa_c := \kappa_0\det\mathbf{J}, \tag*{(3)}

where J:=∂Xχ(X)\mathbf{J} := \partial_X\chi(X) is the Jacobian matrix associated with the transformation (a push-forward) that describes the local deformation of the geometry. In the following we use a non-singular transformation to define two open cavities with different shapes but sharing the same spectrum.

1D analytical example: twinning a semi-infinite pipe

We outline the implications of this approach by considering a simple one dimensional problem that admits a closed form solution.

Let us consider the 1D semi-infinite spaces Σ\Sigma and σ\sigma as shown in Figure 1a, they contain either one or two slabs of fluids different from the matrix. Namely, fluid ii is characterised by a density ρi∈R\rho_i \in\mathbb{R} and bulk modulus κi∈R\kappa_i \in\mathbb{R}, and it is contained in the geometric interval Σi\Sigma_i (or σi\sigma_i), for i={0,1,2}i = \{0,1,2\}. Note that a domain with only one slab can be easily handled as the special case of two slabs having the same properties. As a consequence, we can solve the eigenvalue problem for the most general configuration with two slabs and then readily analyse the pipe with only one slab. Note that this configuration is of interest because the slabs, having acoustic properties different from the matrix, behave like a cavity: this similarity leads to eigenmodes that are partially trapped inside the slab and that leak energy through the unbounded matrix.

Figure 1. (a) A schematic of the reference leaky resonant cavity (top) with domain Σ\Sigma and the transformed (bottom) domain σ\sigma. (b) A plot of the real part of the first four eigenmodes: (top) the reference configuration, (bottom) the transformed one. The modes inside [−b,−δ][-b,-\delta] of the reference configuration are compressed into [−a,−δ][-a,-\delta] for the transformed domain whilst they match for [−δ,+∞)[-\delta,+\infty).

Let us consider the domain σ\sigma in Figure 1a. The eigenvalue problem describing the complex valued pressure p(x):R→Cp(x):\mathbb{R}\to\mathbb{C} on the three acoustic domains reads as:

−∂xxpi(x)=ω2ci2pi(x),x∈σi, i={1,2,3}(4)-\partial_{xx}p_i(x)=\frac{\omega^2}{c_i^2}p_i(x),\quad x\in\sigma_i,\ i=\{1,2,3\} \tag*{(4)}

where we introduce the sound speed ci=κi/ρic_i=\sqrt{\kappa_i/\rho_i}, and the intervals σ0:=[−a,−δ)\sigma_0:=[-a,-\delta), σ1:=[−δ,0)\sigma_1:=[-\delta,0) and σ2:=[0,+∞)\sigma_2:=[0,+\infty), for i={1,2,3}i=\{1,2,3\}. The boundary value problem is completed by the rigid boundary condition on the left end and the Sommerfeld condition at infinity

−∂xp0(−a)=0,lim⁡x→+∞(∂xp2(x)+jk2p2(x))=0,(5)-\partial_xp_0(-a)=0,\quad\lim_{x\to+\infty}\left(\partial_xp_2(x)+jk_2p_2(x)\right)=0, \tag*{(5)}

where we have adopted the time-harmonic convention ejωte^{j\omega t}, being jj the imaginary unit, i.e. j2=−1j^2=-1. Moreover, the equilibrium and compatibility conditions hold on the interfaces at x=−δx=-\delta and x=0x=0

p0(−δ−)=p1(−δ+),1ρ0∂xp0(−δ−)=1ρ1∂xp2(−δ+),p1(0−)=p2(0+),1ρ1∂xp1(0−)=1ρ2∂xp2(0+).(6)\begin{aligned} p_0(-\delta^-)&=p_1(-\delta^+), & \frac{1}{\rho_0}\partial_xp_0(-\delta^-)&=\frac{1}{\rho_1}\partial_xp_2(-\delta^+),\\ p_1(0^-)&=p_2(0^+), & \frac{1}{\rho_1}\partial_xp_1(0^-)&=\frac{1}{\rho_2}\partial_xp_2(0^+). \tag*{(6)} \end{aligned}

By considering the dispersion relations ω=kici\omega=k_ic_i, i={0,1,2}i=\{0,1,2\} and choosing the ansatz p0(x)=A0cos⁡(k0x+φ0)p_0(x)=A_0\cos(k_0x+\varphi_0), p1(x)=A1cos⁡(k1x+φ1)p_1(x)=A_1\cos(k_1x+\varphi_1) that account for the resonances trapped in the slabs and p2(x)=A2e−jk2x+B2e+jk2xp_2(x)=A_2e^{-jk_2x}+B_2e^{+jk_2x} for the leakage through the matrix, the eigenvalue problem is solved if the boundary conditions are applied. We thus obtain the following equalities:

φ0=k0a,A0cos⁡(k0(a−δ))=A1cos⁡(φ1−k1δ),A1cos⁡(k1b)=A2,B2=0,k0ρ0tan⁡(k0(a−δ))=k1ρ1tan⁡(φ1−k1δ),k1ρ1tan⁡φ1=−jk2ρ2.(7)\begin{aligned} \varphi_0&=k_0a,\quad A_0\cos\left(k_0(a-\delta)\right)=A_1\cos\left(\varphi_1-k_1\delta\right),\quad A_1\cos(k_1b)=A_2,\\ B_2&=0,\quad\frac{k_0}{\rho_0}\tan\left(k_0(a-\delta)\right)=\frac{k_1}{\rho_1}\tan\left(\varphi_1-k_1\delta\right),\quad\frac{k_1}{\rho_1}\tan\varphi_1=-j\frac{k_2}{\rho_2}. \tag*{(7)} \end{aligned}

If ρ0=ρ1\rho_0=\rho_1 and κ0=κ1\kappa_0=\kappa_1, the two slabs merge together and the domain Σ\Sigma shown in Figure 1a can be considered. In that case, equalities (7) simplify to

φ0=φ1=k1b,A0=A1,B2=0,A2=A1cos⁡(k1b),k1ρ1tan⁡(k1b)=−jk2ρ2.(8)\varphi_0=\varphi_1=k_1b,\quad A_0=A_1,\quad B_2=0,\quad A_2=A_1\cos(k_1b),\quad\frac{k_1}{\rho_1}\tan(k_1b)=-\mathrm{j}\frac{k_2}{\rho_2}. \tag*{(8)}

This defines the set of eigenvalues

Ω(n)=c1b[nπ+j2ln⁡(z2+z1z2−z1)],n∈Z(9)\Omega^{(n)}=\frac{c_1}{b}\left[n\pi+\frac{\mathrm{j}}{2}\ln\left(\frac{z_2+z_1}{z_2-z_1}\right)\right],\qquad n\in\mathbb{Z} \tag*{(9)}

where zi=κiρiz_i=\sqrt{\kappa_i\rho_i} is the impedance of medium ii and Ω(n)=K1(n)c1=K2(n)c2\Omega^{(n)}=K_1^{(n)}c_1=K_2^{(n)}c_2. In the field of optics, this corresponds to Fabry-Perot resonances [27, 28] at which light exhibits constructive interference after one round trip, with the imaginary part giving the the lifetime of a photon inside the cavity τn=2π/ℑΩ(n)\tau_n=2\pi/\Im\Omega^{(n)}. The corresponding eigenmodes are given by

P^1(n)=cos⁡(K1(n)(X+b)),P^2(n)=cos⁡(K1(n)b)e−jK2(n)X.(10)\hat{P}_1^{(n)}=\cos\left(K_1^{(n)}(X+b)\right),\qquad\hat{P}_2^{(n)}=\cos\left(K_1^{(n)}b\right)e^{-\mathrm{j}K_2^{(n)}X}. \tag*{(10)}

Note that, since the one slab configuration is referred as the material frame, capital letters are adopted. Figure 1b shows the shape of the first four eigenmodes, where one can remark the exponential growth which is an evidence of the leakage.

In the following, two twins of the pipe are obtained first by compressing a portion of the slab 1 and then by folding the space at the extremity of the pipe; both approaches give rise to a second slab of fluid, referred to with the index 0. We then compute the properties of the fluid filling the transformed domain σ\sigma such that the cavities share the same eigenvalues and eigenvectors.

Let us choose a linear map χ:Σ→σ\chi:\Sigma\to\sigma, X↦xX\mapsto x such that Σc:=[−b,−δ]→σc:=[−a,−δ]\Sigma_c:=[-b,-\delta]\to\sigma_c:=[-a,-\delta] and Σf:=(−δ,+∞)→σf:=(−δ,+∞)\Sigma_f:=(-\delta,+\infty)\to\sigma_f:=(-\delta,+\infty). That is

χ(X):={a−δb−δX+δa−bb−δ,if X∈[−b;−δ]X,if X>−δ(11)\chi(X):= \begin{cases} \frac{a-\delta}{b-\delta}X+\delta\frac{a-b}{b-\delta}, & \text{if }X\in[-b;-\delta]\\ X, & \text{if }X>-\delta \end{cases} \tag*{(11)}

In this case, the deformation gradient J=∂Xx=a−δb−δJ=\partial_Xx=\frac{a-\delta}{b-\delta} is a scalar, so we obtain the following transformed properties

ρ0=b−δa−δρ1,κ0=a−δb−δκ1,c0=a−δb−δc1,z0=z1(12)\rho_0=\frac{b-\delta}{a-\delta}\rho_1,\qquad\kappa_0=\frac{a-\delta}{b-\delta}\kappa_1,\qquad c_0=\frac{a-\delta}{b-\delta}c_1,\qquad z_0=z_1 \tag*{(12)}

by usage of the formulae (3). The resulting acoustic system is the one of a shorter pipe having two slabs of homogeneous properties, depicted in Figure 1a. Its eigenmodes are readily computed through equalities (7):

p^0(n)=cos⁡(k0(n)(x+a)),p^1(n)=cos⁡(k1(n)(x+b)),p^2(n)=cos⁡(k1(n)b)e−jk2(n)x,(13)\hat{p}_0^{(n)}=\cos\left(k_0^{(n)}(x+a)\right),\qquad\hat{p}_1^{(n)}=\cos\left(k_1^{(n)}(x+b)\right),\qquad\hat{p}_2^{(n)}=\cos\left(k_1^{(n)}b\right)e^{-\mathrm{j}k_2^{(n)}x}, \tag*{(13)}

where k0(n)=J−1K1(n)k_0^{(n)}=J^{-1}K_1^{(n)}, k1(n)=K1(n)k_1^{(n)}=K_1^{(n)}, and the eigenvalues ω(n)=Ω(n)\omega^{(n)}=\Omega^{(n)} are the same of the original cavity given by Eq. (9). Note that the eigenmodes are such that p^(n)(x)=P^(n)(χ−1(x))\hat{p}^{(n)}(x)=\hat{P}^{(n)}(\chi^{-1}(x)).

As just illustrated, it is possible to transform the space and obtain a shorter or a longer twin cavity. However, as highlighted by Eq. (12), the properties change proportionally to the geometric stretch: if ρ\rho is increased, κ\kappa should decrease accordingly. In general, a material with such properties is not readily attainable, so in the following we define a slightly different strategy for twinning a cavity relying on space folding.

It is well known that space folding transformations [29] lead to Negative Index Materials (NIM) [30, 31, 32, 33, 34, 35], whose effective properties are the consequence of localised resonances. Hence, we can leverage this mathematical abstraction to unlock alternative transformations and create another route for the design of isospectral open cavities.

Let us consider the schematic on top of Figure 2a, the space [−c,−b)[-c,-b) beyond the rigid extremity of the pipe is considered as the overlap of two complementary materials [32] whose acoustic behaviours annihilate each other. Through the transformation χ\chi, such that

Figure 2. (a) A schematic of the reference leaky resonant cavity (top) and the transformed (bottom) domain. (b) The plot of the real part of the first four eigenmodes: (top) the reference configuration, and (bottom) the transformed one. The modes inside [−d,−b][-d,-b] of the transformed configuration are symmetric with respect to the point −c-c and, if folded into [−c,−b][-c,-b], they annihilate each other; the eigenmodes are preserved in the half space [−b,+∞)[-b,+\infty).

X=χ−1(x)={x,x≥−c−x−2c,−d≤x<−c(14)X=\chi^{-1}(x)= \begin{cases} x, & x\ge-c\\ -x-2c, & -d\le x<-c \end{cases} \tag*{(14)}

they are unfolded onto a straight line and, like the negative of a photograph, their properties are equal but with opposite signs: ρ0=−ρ1\rho_0=-\rho_1, κ0=−κ1\kappa_0=-\kappa_1 (doubly negative acoustic parameters can be achieved in practice with single resonance metamaterials [36], that constrains twinning to a small portion of the spectrum). This arrangement is highlighted in Figure 2 using complementary colours. Note that we chose c=(b+d)/2c=(b+d)/2 such that only unitary (positive or negative) stretches arise, however different choices are possible.

By applying once again the equalities (7), the closed form of the twin cavity eigenvalues is achieved:

p^0(n)=cos⁡(k0(n)(x+d)),p^1(n)=cos⁡(k1(n)(x+b)),p^2(n)=cos⁡(k1(n)b)e−jk2(n)x,(15)\hat{p}_0^{(n)}=\cos\left(k_0^{(n)}(x+d)\right),\qquad\hat{p}_1^{(n)}=\cos\left(k_1^{(n)}(x+b)\right),\qquad\hat{p}_2^{(n)}=\cos\left(k_1^{(n)}b\right)e^{-jk_2^{(n)}x}, \tag*{(15)}

where the wavenumbers ki(n)k_i^{(n)} are obtained using the dispersion relation ωi(n)=ki(n)ci\omega_i^{(n)}=k_i^{(n)}c_i, and the eigenvalues are the same as Eq. (9); thus the cavity is a twin.

These trivial 1D cases share many features with several practical applications, for instance a duct where pressure waves propagate, or a string under tension sustaining transverse waves, etc. One can therefore substitute a portion of the domain with a shorter/longer one with equivalent parameters whilst keeping the same eigenfrequencies and alter the eigenmodes inside the transformed region only.

Manipulating a Helmholtz resonator

We now consider a 2D open cavity and cross-validate computations of its spectrum against numerical simulations of an ideal anisotropic fluid and also against an effective medium using layers [37], each having constant material properties, which illustrates an approach towards experimental realization.

Let us consider the geometry depicted in Figure 3a, the cavity is assumed to have sound hard boundary Γ\Gamma and it communicates with the exterior medium through a small aperture; an artificial boundary is introduced to truncate the computational domain and it is surrounded by a Perfectly Matched Layer (PML) that mimics an infinite domain, see, e.g., [22]. The cavity is large compared to the aperture such that trapped modes are encouraged to arise, while the small aperture allows energy to leak out from the system.

Figure 3. (a) Virtual geometry of the reference cavity, both Σf\Sigma_f and Σc\Sigma_c are filled with a homogeneous fluid; (b) ideally twinned cavity, σf\sigma_f is filled with a homogeneous fluid while σc\sigma_c is filled with the anisotropic fluid; (c) the anisotropic fluid is approximated by a graded layered metamaterial. The geometry is defined by h=l=1 mh=l=1\,\mathrm{m}, t=1.2lt=1.2l, a=3la=3l, b=1.5ab=1.5a and δ=0.5a\delta=0.5a.

The cavity is filled by a fluid of density ρ0=1 kg m−3\rho_0=1\,\mathrm{kg\,m^{-3}} and bulk modulus κ0=1 Pa\kappa_0=1\,\mathrm{Pa}; the computational domain Σ⊂R2\Sigma\subset\mathbb{R}^2 is partitioned in two non-overlapping sets Σf\Sigma_f and Σc\Sigma_c such that Σ=Σf∪Σc\Sigma=\Sigma_f\cup\Sigma_c. The Helmholtz’s equation is supplied with Neumann boundary conditions:

{−∇X⋅(ρ0−1∇XP(X))=Ω2κ0−1P(X)in Σ,−ρ0−1∇XP(X)⋅N=0on Γ(16)\begin{cases} -\nabla_{\mathbf{X}}\cdot\left(\rho_0^{-1}\nabla_{\mathbf{X}}P(\mathbf{X})\right)=\Omega^2\kappa_0^{-1}P(\mathbf{X}) & \text{in }\Sigma,\\ -\rho_0^{-1}\nabla_{\mathbf{X}}P(\mathbf{X})\cdot\mathbf{N}=0 & \text{on }\Gamma \tag*{(16)} \end{cases}

where P(X)P(\mathbf{X}) is the complex valued pressure defined on the material domain Σ\Sigma. Note that the eigenvalue problem is obtained considering time-harmonic solutions of the form P^(X,t)=ℜ(P(X)ejωt)\hat{P}(\mathbf{X},t)=\Re(P(\mathbf{X})e^{j\omega t}).

A smaller twin cavity is defined by applying Transformation Acoustics, see § 2, and defining a portion of anisotropic fluid, similar to a carpet cloak [14]. Note that a smaller cavity simply filled with the ambient fluid would have higher resonance frequencies.

Figure 3a-Figure 3(b) illustrates the geometry we consider and the result of the transformation χ:Σ→σ\chi:\Sigma\to\sigma that maps the cavity geometry into a smaller one. In particular, the circular annulus Σc\Sigma_c of radii δ\delta and bb is mapped to a smaller one of radii δ\delta and a<ba < b, referred as the deformed domain σc\sigma_c. The transformation is the identity inside the remaining domain Σf\Sigma_f, that is simply mapped into σf≡Σf\sigma_f \equiv\Sigma_f. Also note that the boundary Γ\Gamma is smoothly mapped onto γ\gamma. Figure 3c shows the transformed medium with the anisotropic fluid replaced by an effective medium made from layers of homogeneous material.

Since the cavity is semi-circular, it is convenient to transform the geometry adopting a polar reference system centred in OO and described by the set of coordinates (R,Θ)(R,\Theta). Then χ:(R,Θ)↦(r,θ)\chi:(R,\Theta) \mapsto(r,\theta) is such that

r=f−1(R),θ=Θ,(17)r=f^{-1}(R), \qquad\theta=\Theta, \tag*{(17)}

where the only requirement is that f(a)=bf(a)=b and f(δ)=δf(\delta)=\delta. We emphasise here that f(r)f(r) does not have to be monotonic, meaning situations implying space folding are encapsulated by the present method. This would require Negative Index Materials (NIM) [30, 31, 34] whose properties can be approximated by locally resonant material. For the sake of clarity, the effect of such a choice is discussed at the end of this section. The transformation gradient J:=∇Xx\mathbf{J}:=\nabla_{\mathbf{X}}\mathbf{x} is

J=[1f′(r)00rf(r)],det⁡J=rf′(r)f(r).(18)\mathbf{J}= \begin{bmatrix} \frac{1}{f'(r)} & 0 \\ 0 & \frac{r}{f(r)} \end{bmatrix}, \qquad \det\mathbf{J}=\frac{r}{f'(r)f(r)}. \tag*{(18)}

Note that the mixed tensor J\mathbf{J} is expressed with respect to the canonical contravariant base (er,eθ)(\mathbf{e}_r,\mathbf{e}_\theta) and covariant base (ER,EΘ)(\mathbf{E}^R,\mathbf{E}^\Theta). Note also that no rotation is implied by the transformation, hence the deformation gradient is symmetric. This turns out to be useful if, for instance, the anisotropic transformed fluid is attained through a pentamode material, the reader is referred to [9] for a complete discussion.

Twinning via monotonic transform

For simplicity, we choose the linear function

f(r)=b−δa−δr−δb−aa−δ(19)f(r)=\frac{b-\delta}{a-\delta}r-\delta\frac{b-a}{a-\delta} \tag*{(19)}

so the bulk modulus κc\kappa_c and the tensor density ρc\boldsymbol{\rho}_c of the transformed domain σc\sigma_c are computed according to formulae (3)

ρc=ρ0[rf′(r)f(r)00f(r)rf′(r)]=[ρr(r)00ρθ(r)],κc=κ0rf′(r)f(r).(20)\boldsymbol{\rho}_c=\rho_0 \begin{bmatrix} \frac{r f'(r)}{f(r)} & 0 \\ 0 & \frac{f(r)}{r f'(r)} \end{bmatrix} = \begin{bmatrix} \rho_r(r) & 0 \\ 0 & \rho_\theta(r) \end{bmatrix}, \qquad \kappa_c=\kappa_0\frac{r}{f'(r)f(r)}. \tag*{(20)}

and, because of the axisymmetric geometry, these properties depend on the radial coordinate only.

In view of a real implementation of the anisotropic fluid through a metamaterial, it is fundamental to discuss how it affects the spectrum of the cavity. The stretch of the geometry induces an anisotropic density whose principal components are aligned with the radial and the tangential directions of the cavity, similarly to the invisibility cloak for axisymmetric obstacles see, e.g., [38]. This situation lends itself to the use of a layered arrangement of two homogeneous and isotropic fluids for achieving the effective properties we need. The following analytical relations hold [38] for the effective properties:

ρeff=[⟨ρ⟩00⟨1ρ⟩−1],κeff=⟨1κ⟩−1.(21)\boldsymbol{\rho}_{\mathrm{eff}}= \begin{bmatrix} \langle\rho\rangle& 0 \\ 0 & \langle\frac{1}{\rho}\rangle^{-1} \end{bmatrix}, \qquad \kappa_{\mathrm{eff}}=\left\langle\frac{1}{\kappa}\right\rangle^{-1}. \tag*{(21)}
ii0123456789
κAi=κBi\kappa_A^i = \kappa_B^i [Pa]0.3310.3260.3210.3180.3140.3110.3080.3060.3030.301
ρAi\rho_A^i [kg m−3\mathrm{kg}\,\mathrm{m}^{-3}]2.1882.1382.0942.0542.0171.9841.9531.9251.8991.875
ρBi\rho_B^i [kg m−3\mathrm{kg}\,\mathrm{m}^{-3}]0.4570.4680.4780.4870.4960.5040.5120.5200.5270.533

Table 1. Bulk moduli and densities of each layer composing the metamaterial inside the cavity.

For our illustration, the anisotropic fluid is discretised by 10 layers of the same thickness δr=(a−δ)/10\delta_r = (a-\delta)/10 each one composed by two materials A and B of thickness ε=δr/2\varepsilon= \delta_r/2, such that they approximate the anisotropic medium if homogenised two by two. A schematic of such arrangement is shown in Figure 3c.

We set ri:=(2i+1)εr^i := (2i+1)\varepsilon, i∈{0,…,9}i \in\{0,\ldots,9\} the discrete radii pointing in the middle of each layer and κi:=κc(ri)\kappa^i := \kappa_c(r^i), ρri:=ρr(ri)\rho_r^i := \rho_r(r^i), and ρθi:=ρθ(ri)\rho_\theta^i := \rho_\theta(r^i) the transformed properties evaluated at rir^i. The two materials AA and BB in the ithi^{\mathrm{th}} layer are chosen to have the same bulk modulus such that

κAi=κBi=κi,(22)\kappa_A^i=\kappa_B^i=\kappa^i, \tag*{(22)}

while the densities are chosen as

ρAi=ρri−(ρri)2−ρθiρri,ρBi=ρri+(ρri)2−ρθiρri;(23)\rho_A^i=\rho_r^i-\sqrt{(\rho_r^i)^2-\rho_\theta^i\rho_r^i},\qquad \rho_B^i=\rho_r^i+\sqrt{(\rho_r^i)^2-\rho_\theta^i\rho_r^i}; \tag*{(23)}

one can easily verify that

⟨ρ⟩=ρAi+ρBi2=ρri,⟨1ρ⟩−1=2(1ρAi+1ρBi)−1=ρθi.(24)\langle\rho\rangle=\frac{\rho_A^i+\rho_B^i}{2}=\rho_r^i,\qquad \left\langle\frac{1}{\rho}\right\rangle^{-1}=2\left(\frac{1}{\rho_A^i}+\frac{1}{\rho_B^i}\right)^{-1}=\rho_\theta^i. \tag*{(24)}

Table 1 describes the properties of each layer used in the discretized version of the twinned cavity.

Twinning via non-monotonic transform

Lastly, we briefly consider the transformation enforced by a non-monotonous function f(r)f(r) that, as anticipated § 2, leads to space folding. To show that the spectrum is matched, let us consider the annular region beyond the curved boundary of the cavity as an overlapped portion of space that can be unfolded into a plane space filled with complementary materials. As suggested by [34], the function

f(r)={r,r≤cc2/r,r>c(25)f(r)= \begin{cases} r, & r\le c\\ c^2/r, & r>c \end{cases} \tag*{(25)}

is a natural choice for this circular geometry since it leads to an isotropic yet inhomogeneous medium, whose properties are ρθ=ρr=−ρ0\rho_\theta=\rho_r=-\rho_0 and κc=−r4/c4\kappa_c=-r^4/c^4. Where c=bdc=\sqrt{bd} and d=2ad=2a describe the depth of the twin cavity equipped with a NIM layer, as for the 1D arrangement of § 2.1.

Twinning efficiency via modal assurance criterion

We now turn to a quantitative comparison of the spectrum of the reference cavity with the three configurations just defined (ideal, layer and NIM) and these are all modelled using a finite element method (COMSOL Multiphysics®). To ensure a fair comparison between the cases, the mesh characteristic size is set to h<δr/4h < \delta_r/4 in order to account for the spatially varying properties in the transformed domain and pressure fluctuations, for higher order modes in particular. In addition, the finite element problems share the same mesh in order to minimise the numerical errors due to the discretization.

The accuracy of the matching between the eigenfrequencies is expected to deteriorate for increasing frequencies, thus the 20 eigenvalues with real part larger than and closer to c0/5l=0.2c_0/5l = 0.2 Hz are computed in order to show this trend. Table 2 shows their values and highlights the good agreement between the reference and the ideal cases, and a strong match with the layered and NIM configurations. Please note that, in order to improve the numerical convergence for the NIM cavity, a small fictitious damping has been added to the negative material properties such that ρc→(1+i×10−4)ρc\rho_c \rightarrow(1 + i \times10^{-4})\rho_c and κc→(1+i×10−4)κc\kappa_c \rightarrow(1 + i \times10^{-4})\kappa_c; this inherently leads to small differences in the solution of the eigenvalue problem for that case.

Modefreff_{ref} [Hz]fidealf_{ideal} [Hz]flayerf_{layer} [Hz]fNIMf_{NIM} [Hz]
10.2269+8.0i e−150.2269 + 8.0i\,e-150.2269+4.9i e−150.2269 + 4.9i\,e-150.2219+5.8i e−150.2219 + 5.8i\,e-150.2269+1.3i e−60.2269 + 1.3i\,e-6
20.2371+0.0170i0.2371 + 0.0170i0.2371+0.0170i0.2371 + 0.0170i0.2389+0.0179i0.2389 + 0.0179i0.2371+0.0170i0.2371 + 0.0170i
30.2373+1.5i e−50.2373 + 1.5i\,e-50.2373+1.5i e−50.2373 + 1.5i\,e-50.2377+1.8i e−50.2377 + 1.8i\,e-50.2373+1.6i e−50.2373 + 1.6i\,e-5
40.2653+5.6i e−140.2653 + 5.6i\,e-140.2653+5.4i e−140.2653 + 5.4i\,e-140.2587+6.5i e−150.2587 + 6.5i\,e-150.2653+1.1i e−60.2653 + 1.1i\,e-6
50.2798+0.1049i0.2798 + 0.1049i0.2798+0.1049i0.2798 + 0.1049i0.2798+0.1049i0.2798 + 0.1049i0.2798+0.1049i0.2798 + 0.1049i
60.2835+8.1i e−100.2835 + 8.1i\,e-100.2835+8.1i e−100.2835 + 8.1i\,e-100.2833+8.1i e−100.2833 + 8.1i\,e-100.2835+6.2i e−70.2835 + 6.2i\,e-7
70.3023+0.0260i0.3023 + 0.0260i0.3023+0.0260i0.3023 + 0.0260i0.3017+0.0244i0.3017 + 0.0244i0.3022+0.0260i0.3022 + 0.0260i
80.3034+1.7i e−130.3034 + 1.7i\,e-130.3034+3.5i e−130.3034 + 3.5i\,e-130.2949+1.5i e−140.2949 + 1.5i\,e-140.3034+7.9i e−70.3034 + 7.9i\,e-7
90.3057+3.8i e−60.3057 + 3.8i\,e-60.3057+3.8i e−60.3057 + 3.8i\,e-60.3059+3.5i e−60.3059 + 3.5i\,e-60.3057+3.9i e−60.3057 + 3.9i\,e-6
100.3283+3.5i e−80.3283 + 3.5i\,e-80.3283+3.5i e−80.3283 + 3.5i\,e-80.3267+3.2i e−80.3267 + 3.2i\,e-80.3283+7.7i e−80.3283 + 7.7i\,e-8
110.3412+4.1i e−130.3412 + 4.1i\,e-130.3412+1.2i e−140.3412 + 1.2i\,e-140.3307+5.0i e−150.3307 + 5.0i\,e-150.3412+3.3i e−70.3412 + 3.3i\,e-7
120.3501+0.1016i0.3501 + 0.1016i0.3501+0.1016i0.3501 + 0.1016i0.3498+0.1016i0.3498 + 0.1016i0.3501+0.1017i0.3501 + 0.1017i
130.3530+1.9i e−50.3530 + 1.9i\,e-50.3530+1.9i e−50.3530 + 1.9i\,e-50.3526+1.5i e−50.3526 + 1.5i\,e-50.3529+1.8i e−50.3529 + 1.8i\,e-5
140.3721+3.7i e−120.3721 + 3.7i\,e-120.3721+3.6i e−120.3721 + 3.6i\,e-120.3692+3.1i e−120.3692 + 3.1i\,e-120.3721+4.1i e−70.3721 + 4.1i\,e-7
150.3788+3.7i e−130.3788 + 3.7i\,e-130.3788+7.3i e−140.3788 + 7.3i\,e-140.3661+1.8i e−140.3661 + 1.8i\,e-140.3788+1.7i e−70.3788 + 1.7i\,e-7
160.3930+0.0186i0.3930 + 0.0186i0.3930+0.0186i0.3930 + 0.0186i0.3958+0.0202i0.3958 + 0.0202i0.3930+0.0186i0.3930 + 0.0186i
170.4013+2.6i e−80.4013 + 2.6i\,e-80.4013+2.6i e−80.4013 + 2.6i\,e-80.4111+2.5i e−100.4111 + 2.5i\,e-100.4013+0.9i e−60.4013 + 0.9i\,e-6
180.4150+0.9i e−100.4150 + 0.9i\,e-100.4150+6.2i e−90.4150 + 6.2i\,e-90.4010+4.4i e−120.4010 + 4.4i\,e-120.4150+0.9i e−60.4150 + 0.9i\,e-6
190.4163+1.7i e−90.4163 + 1.7i\,e-90.4163+1.4i e−80.4163 + 1.4i\,e-80.4022+2.8i e−80.4022 + 2.8i\,e-80.4163+6.3i e−70.4163 + 6.3i\,e-7
200.4190+4.0i e−50.4190 + 4.0i\,e-50.4190+4.0i e−50.4190 + 4.0i\,e-50.4225+4.4i e−50.4225 + 4.4i\,e-50.4190+3.8i e−50.4190 + 3.8i\,e-5

Table 2. The first 20 computed eigenfrequencies are displayed; the first column being the reference cavity, the second column is the ideal transformation and these agree very closely. The final two columns show small discrepancies; the third because of the layer discretization, and the fourth due to fictitious damping that mainly affects the imaginary parts. The frequency highlighted in red is related to a PML resonance.

As an example, some eigenmodes are displayed in Figure 4 and notably we do not show the ideal case as the eigenvalues are identical to the reference case, in so far as digits shown, and the eigenfields are visually indistinguishable from each other. Since the precise comparison of the fields is not straightforward, a measure of the twinning is given by applying the so-called Modal Assurance Criterion (MAC) [39] which allows us to compare two eigenmodes by computing the following matrix:

Absolute value of the pressure for three eigenmodes of the original cavity, the cavity cloaked with a layered metamaterial, and the enlarged cavity equipped with a layer of NIM, shown in three columns labeled Reference, Layer, and NIM.

Figure 4. The three columns show the absolute value of the pressure for three eigenmodes of the original cavity (left), the cavity cloaked with a layered metamaterial (centre), and the enlarged cavity equipped with a layer of NIM (right). (a), (b) and (c) show the twinning of a mode mainly characterized by leakage; (d), (e) and (f) show a trapped resonance with some leakage; (g), (h) and (i) show a pure trapped mode inside the cavity. The dashed blue lines highlights the PML, the undeformed domain, and the metamaterial layers.

MACijideal:=∣∫Σfpi,idealpj,ref∗∣2∫Σfpi,idealpi,ideal∗∫Σfpj,refpj,ref∗(26)MAC_{ij}^{ideal} := \frac{\left|\int_{\Sigma_f} p_{i,ideal}p_{j,ref}^{*}\right|^{2}}{\int_{\Sigma_f}p_{i,ideal}p_{i,ideal}^{*}\int_{\Sigma_f}p_{j,ref}p_{j,ref}^{*}} \tag*{(26)}

where Σf≡σf\Sigma_f \equiv\sigma_f is the portion of the geometry that does not undergo any transformation, p∗p^* is the complex conjugate of pp, and the subscripts refref, idealideal, layerlayer, and NIMNIM indicate respectively the reference cavity, the ideally twinned cavity, and the two cavities equipped with the layered and the NIM gratings. The indexes ii and jj specify the eigenmodes under comparison. Note that the MAC matrix is real valued and it assumes unitary values if and only if the two fields under comparison are identical inside σf\sigma_f, except for an arbitrary scaling factor. So one can state that the twinning is satisfactory for all eigenmodes if the main diagonal terms of the MACMAC matrix are close to the unity.

Thus the idealideal, layerlayer, and NIMNIM twin cavities are compared to the reference one. Figure 5 displays the matrices 1−MACijideal1-MAC_{ij}^{ideal}, 1−MACijlayer1-MAC_{ij}^{layer}, and 1−MACijNIM1-MAC_{ij}^{NIM} using a logarithmic colour scale, such that good twinning is highlighted by values close to 0. Figure 5a, relative to the idealideal twin, shows a complete agreement for each and every mode, except the 19th. This mode is very localized in the transformed region σc\sigma_c, so the agreement cannot be captured by comparing σf\sigma_f only. Figure 5b proves that the layered arrangement approximates the ideal behaviour because most of the modes show excellent agreement with the reference ones. Lastly, Figure 5c shows a weaker match, due to the fictitious damping. Some of the off-diagonal terms are close to 0 even if the corresponding modes have very different frequencies. Indeed, those modes have a similar shape inside the test region σf\sigma_f, but since they are far in frequency, they can be easily disregarded.

Figure 5. The computed eigenmodes of the original cavity are compared with (a) the ideal,(b) the layer, and the (c) NIM twin cavities. We use a logarithmic scale to allow the twinning to be more easily seen; more accurate twinning corresponds to smaller values.

Finally, the spectra of the four cavities are compared by computing the first 100 eigenfrequencies. Figure 6a shows that they superpose in the complex plane and the relative errors in Figure 6b are small: below 1×10−51 \times10^{-5} for the ideal twin, below 1×10−21 \times10^{-2} for the layered twin, and below 5×10−25 \times10^{-2} for the NIM twin. Thus we obtain, numerically, that the spectrum is matched for both real and imaginary parts and hence that our design methodology is validated for open systems.

Figure 6. The first 100 eigenfrequencies of the four cavities are compared in (a) and the relative errors with respect to the reference are shown in (b). On the complex plane, we can distinguish an oblique straight line collecting the resonances induced by the PML and a cloud of points close to the real axis that contains the trapped cavity resonances.

Flattening a blazed grating

We now illustrate our approach for twinning open cavities by considering a class of diffraction gratings. Structured surfaces, such as the saw-tooth one depicted in Figure 7, usually called blazed or echelette gratings, are widely used in electromagnetism, for instance in high-resolution spectroscopy and spectrometers [40, 41], for their ability to diffract incident light into a given direction [42]. We now create a metasurface having the same behaviour as the echelette grating, but characterised by a completely flat profile, and use the formalism of acoustics.

Figure 7. Sketch of the metasurface and transformation. Left: the saw-tooth grating described by its depth hh and period ll; middle: a portion of the domain Σc\Sigma_c is transformed to a rectangle called σc\sigma_c using a projection with respect to the rotated reference systems XOY=xOyXOY = xOy; right: flattened metasurface having the same acoustic behaviour as the saw-tooth grating. l=1 ml = 1\,\mathrm{m}, h=l/2h = l/2, t=2lt = 2l.

Each unit cell of the echelette is considered as an open cavity, and a twinned cell is defined to enable us to build a flat metasurface thinner than the reference grating. Finally, we test the twinning showing a complete agreement between the dispersion diagrams of the grating and the flat metasurface. Using the periodicity along the horizontal direction the acoustic behaviour of the metasurface, displayed in Figure 7, is analysed by considering a single unit cell of width ll and imposing the Floquet-Bloch periodicity on the two vertical cell boundaries. We now focus on the transformation applied to the geometry of each cell such that the saw-tooth metasurface is mapped into a flat one.

As in section 4, the elementary cell Σ\Sigma (filled with a fluid having ρ0=1 kg/m3\rho_0 = 1\,\mathrm{kg}/\mathrm{m}^3 and κ=1 Pa\kappa= 1\,\mathrm{Pa}) is partitioned into two domains: Σc\Sigma_c subjected to the transformation and Σf\Sigma_f kept unaltered. The transformation we use involves the entire triangular cavity and an arbitrary portion of the fluid outside which is necessary in order to avoid singularities. The rigid boundary Γ\Gamma is continuously mapped onto a flat boundary; thus, for example, the point CC cannot be mapped on BB or DD, but is mapped to a point between them. The transformation we choose projects the point CC on the line BDBD along the bisector of the angle BCD^\widehat{BCD}, as shown in Figure 7.

This transformation is not unique and one can transform by stretching the domain in many different ways, controlling the amount of stretch imposed to the geometry or even imposing a space folding. This flexibility is potentially valuable in practical terms as according to the precice choice, the transformed properties can scale towards values smaller or greater than those of the fluid, and can be influenced by the availability of materials for implementation.

For the transformation we take the orthonormal rotated reference systems XOY≡xOyXOY \equiv xOy displayed in Figure 7, where one axis is parallel to the bisector of the angle BCD^\widehat{BCD}; this simplifies the analytical expression of the transformation. Following [15] we define three curves Y0(X)Y_0(X), Y1(X)Y_1(X) and Y2(X)Y_2(X) that correspond to the lines ACDACD, ABDABD, and AEDAED respectively. The transformation χ\chi is such that the points between Y0Y_0 and Y2Y_2 are linearly mapped to points between Y1Y_1 and Y2Y_2. That is:

x=X,y=α(X)Y+β(X),(27)\begin{aligned} x &= X,\\ y &= \alpha(X)Y + \beta(X), \tag*{(27)} \end{aligned}

where α(X):=Y2(X)−Y1(X)Y2(X)−Y0(X)\alpha(X) := \frac{Y_2(X)-Y_1(X)}{Y_2(X)-Y_0(X)}, β(X):=Y1(X)−Y0(X)α(X)\beta(X) := Y_1(X) - Y_0(X)\alpha(X). The deformation gradient is then given by:

J=[10α′Y+β′α],(28)\mathbf{J} = \begin{bmatrix} 1 & 0\\ \alpha'Y + \beta' & \alpha \end{bmatrix}, \tag*{(28)}

and the transformed properties ρc\rho_c and κc\kappa_c are computed according to Eq. (3). The tensor ρc\boldsymbol{\rho}_c is related to the rotated reference system xOyxOy, thus it has been rotated such that its components are referred to the original Cartesian system; the plots of Figure 8a show the properties obtained on a colour scale.

Figure 8. Transformed properties normalised with respect to the fluid properties κ0\kappa_0 and ρ0\rho_0. The transformation changes the material properties according to the stretch imposed. Values are displayed in colour scale, for the ideal (a) and discretized (b) twins. Note that the transformation keeps the properties unaltered in a triangle close to the upper left vertex since the curves Y0Y_0 and Y1Y_1 are superimposed along ABAB.

Numerical simulations are performed to find the dispersion curves, i.e. the eigenspectra, that relate frequency to Bloch wavenumber, kk, and these use the original and transformed unit cells in the finite element simulations that are augmented with a cartesian PML that truncates the semi-infinite medium. From periodicity the dispersion diagrams only require variation of the wavenumber kk in the first Brillouin zone [0,π/l][0,\pi/l]. Figure 9 compares the dispersion diagrams of the twinned cavities: both real and imaginary parts of the frequency show a very good agreement, indicating the ability to twin all the dispersion properties for these periodic structures. We note that a similar approach was implemented in [43] to quantify the cloaking efficiency of carpet cloaks. However, dispersion diagrams were an aside in a study addressing scattering problems.

Figure 9. The two dispersion diagrams of the metasurface and its ideal twin superimposed; (a) real and (b) imaginary parts of the frequency.

Turning now to creating an effective medium approximation using discrete elements made from materials with homogeneous properties. Noting that significant inhomogeneity can arise in the transformed medium, so the layered arrangement adopted in § 3 is not possible due to the rapid spatial variations, we introduce an intermediate discretization: a graded metamaterial, constructed from simple components, is designed by dividing the transformed domain into small cells with homogeneous properties, such that each cell can be constructed using, e.g., a layered medium with

MAC comparing the discretized twin against the reference echelette; errors of the complex valued eigenfrequencies for the ideal and the discrete twinsVector (PDF)

caption: (a) MAC comparing the discretized twin against the reference echelette; (b) errors of the complex valued eigenfrequencies for the ideal and the discrete twins.

:::

different orientations with anisotropic effective properties.

To evaluate this discretized approach, the transformed domain is divided into square cells of size ℓ/10\ell/10 and the properties are considered piecewise constant such that each pixel has the values given by ρc(x)\rho_c(\mathbf{x}) and κc(x)\kappa_c(\mathbf{x}) evaluated in its centre. The discretized arrangement is showed in Figure 8b. Numerical simulations enable us to evaluate the twinning numerically and we show results for the Floquet-Bloch ansatz with k=0 rad m−1\mathbf{k}=0\ \mathrm{rad}\ \mathrm{m}^{-1} and the eigenmodes and the eigenfrequencies of the twins are compared through the MAC and the errors shown in Figure 10. In particular, Figure 10a highlights that eigenmodes corresponding to higher frequencies are more affected by the discretization than the low frequency modes; higher frequencies are associated with shorter wavelengths that are more sensitive to a discrete arrangement with similar lengthscales. Figure 10b shows the errors introduced on the eigenmodes by the ideal and the discrete twinning; while the former comes from numerical approximations only, the latter mainly suffers from the coarse discretization used.

Conclusions

The scheme validated through this work suggests a viable way to design the shape of twin open cavities, i.e. those cavities intended to resonate with the same spectrum of a reference cavity, but having a different geometry. We have chosen examples to display the versatility of the approach, i.e. a resonant Helmholtz cavity, which is a good test of the approach having both trapped and leaky modes to reproduce and then a case from periodic systems, which is non-resonant but now has sharp corners. The upshot is that one could take designs from the literature that use constant parameter acoustic fluids as the reference and then transform then to other more convenient geometries. The resultant cavity/surface would have some region with anisotropic fluid within it but, as demonstrated, this could be replaced by layered or discrete homogeneous media as effective replacements.

We show that unbounded domains can be twinned by applying transformation theory: this reveals multiple ways of resizing an open cavity, for instance, the inner wall of a small cavity is covered with an ad hoc layered medium so that it has a spectrum identical to a larger one (or a smaller one for a space folding transform). Both the discrete and the continuous branches of the complex valued spectrum are restored within very good tolerance regardless of the quality factor of the eigenmodes.

We also chose a more challenging case consisting in flattening a blazed grating, and the method is shown to be robust even with strong inhomogeneities. The accuracy of the twinning remains within a good tolerance even when a coarse discretization oriented to realise the metasurface is taken into account. Via the Floquet-Bloch theorem, the assessment is performed comparing the dispersion diagrams of the grating and its twins.

In the context of metasurfaces based around resonators we anticipate that the mathematical approach taken here will allow thinner surfaces to be designed that have low frequency resonators taking up less space. The twinning works almost perfectly for an ideal anisotropic fluid and still provides isospectrality even when an effective medium, made of homogeneous layers, is used to mimic the anisotropic fluid. Our theoretical and numerical results hold in the context of anti-plane shear elasticity and transverse electromagnetism. This provides confidence that such twinned cavities could be fabricated and motivates further experimental work.

References

  1. [1]Cummer SA, Christensen J, Alù A. 2016 Controlling sound with acoustic metamaterials. Nature Reviews Materials 1, 1–13.
  2. [2]Romero-García V, Jimenez N, Theocharis G, Achilleos V, Merkl A, Richoux O, Tournat V, Groby JP, Pagneux V. 2020 Design of acoustic metamaterials made of Helmholtz resonators for perfect absorption by using the complex frequency plane. Comptes Rendus. Physique 21, 713–749.
  3. [3]Liang Z, Li J. 2012 Extreme acoustic metamaterial by coiling up space. Physical review letters 108, 114301.
  4. [4]Kac M. 1966 Can One Hear the Shape of a Drum? The American Mathematical Monthly 73, 1–23.
  5. [5]Gordon C, Webb DL, Wolpert S. 1992 One Cannot Hear the Shape of a Drum. Bulletin of the American Mathematical Society 27, 134–138.
  6. [6]Sleeman BD. 1982 The Inverse Problem of Acoustic Scattering. IMA Journal of Applied Mathematics 29, 113–142.
  7. [7]Park S, Lee I, Kim J, Park N, Yu S. 2022 Hearing the Shape of a Drum for Light: Isospectrality in Photonics. Nanophotonics 11, 2763–2778.DOI
  8. [8]Pendry JB, Schurig D, Smith DR. 2006 Controlling electromagnetic fields. Science 312, 1780–1782.
  9. [9]Norris AN. 2008 Acoustic cloaking theory. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 464, 2411–2434.
  10. [10]Milton GW, Briane M, Willis JR. 2006 On cloaking for elasticity and physical equations with a transformation invariant form. New Journal of Physics 8, 248.DOI
  11. [11]Brun M, Guenneau S, Movchan AB. 2009 Achieving control of in-plane elastic waves. Applied physics letters 94.DOI
  12. [12]Norris AN, Shuvalov AL. 2011 Elastic cloaking theory. Wave Motion 48, 525–538.DOI
  13. [13]Kadic M, Bückmann T, Schittny R, Wegener M. 2015 Experiments on cloaking in optics, thermodynamics and mechanics. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 373, 20140357.DOI
  14. [14]Li J, Pendry JB. 2008 Hiding under the carpet: a new strategy for cloaking. Physical Review Letters 101, 203901.
  15. [15]Lenz SV, Guenneau S, Drinkwater BW, Craster RV, Holderied MW. 2023 Transformation Twinning to Create Isospectral Cavities. Physical Review B 108, 064209.
  16. [16]Hu J, Menyuk CR. 2009 Understanding leaky modes: slab waveguide revisited. Advances in Optics and Photonics 1, 58–106.
  17. [17]Georgiades E, Lowe MJS, Craster RV. 2022 Leaky wave characterisation using spectral methods. The Journal of the Acoustical Society of America 152, 1487–1497.DOI
  18. [18]Bamberger A, Bonnet A. 1990 Mathematical analysis of the guided modes of an optical fiber. SIAM Journal on Mathematical Analysis 21, 1487–1510.DOI
  19. [19]Zolla F, Reverserz G, Nicolet A, Kuhlmey B, Guenneau SR, Felbacq D. 2005 Foundations of photonic crystal fibers. World Scientific.
  20. [20]Greenleaf A, Kurylev Y, Lassas M, Uhlmann G. 2009 Cloaking devices, electromagnetic wormholes, and transformation optics. SIAM review 51, 3–33.DOI
  21. [21]Kohn RV, Onofrei D, Vogelius MS, Weinstein MI. 2010 Cloaking via change of variables for the Helmholtz equation. Communications on Pure and Applied Mathematics: A Journal Issued by the Courant Institute of Mathematical Sciences 63, 973–1016.DOI
  22. [22]Hein S, Hohage T, Koch W. 2004 On resonances in open systems. Journal of Fluid Mechanics 506, 255–284.
  23. [23]Harari I, Slavutin M, Turkel E. 2000 Analytical and numerical studies of a finite element PML for the Helmholtz equation. Journal of Computational Acoustics 8, 121–137.
  24. [24]Olyslager F. 2004 Discretization of Continuous Spectra Based on Perfectly Matched Layers. SIAM Journal on Applied Mathematics 64, 1408–1433.DOI
  25. [25]Dhia ASBB, Goursaud B, Hazard C, Prieto A. 2009 Finite Element Computation of Leaky Modes in Stratified Waveguides. In Leger A, Deschamps M, editors, Ultrasonic Wave Propagation in Non Homogeneous Media Springer Proceedings in Physics pp. 73–86 Berlin, Heidelberg. Springer.DOI
  26. [26]Vial B, Zolla F, Nicolet A, Commandré M. 2014 Quasimodal Expansion of Electromagnetic Fields in Open Two-Dimensional Structures. Physical Review A 89, 023829.DOI
  27. [27]Perot A, Fabry C. 1899 On the Application of Interference Phenomena to the Solution of Various Problems of Spectroscopy and Metrology. Astrophysical Journal 9, 87.DOI
  28. [28]Ismail N, Kores CC, Geskus D, Pollnau M. 2016 Fabry-Perot resonator: spectral line shapes, generic and related Airy distributions, linewidths, finesses, and performance at low or frequency-dependent reflectivity. Opt. Express 24, 16366–16389.
  29. [29]Leonhardt U, Philbin TG. 2006 General relativity in electrical engineering. New Journal of Physics 8, 247.
  30. [30]Veselago VG. 1967 Electrodynamics of substances with simultaneously negative and. Usp. fiz. nauk 92, 517.
  31. [31]Shelby RA, Smith DR, Schultz S. 2001 Experimental verification of a negative index of refraction. Science 292, 77–79.
  32. [32]Pendry J, Ramakrishna SA. 2003 Focusing light using negative refraction. Journal of Physics: Condensed Matter 15, 6345.
  33. [33]Ramakrishna SA. 2005 Physics of negative refractive index materials. Reports on progress in physics 68, 449.
  34. [34]Lai Y, Chen H, Zhang ZQ, Chan CT. 2009 Complementary media invisibility cloak that cloaks objects at a distance outside the cloaking shell. Physical Review Letters 102, 093901.
  35. [35]Ramakrishna SA, Pendry J. 2004 Spherical perfect lens: Solutions of Maxwell’s equations for spherical geometry. Physical Review B 69, 115115.DOI
  36. [36]Li J, Chan CT. 2004 Double-negative acoustic metamaterial. Physical Review E 70, 055602.
  37. [37]Torrent D, Sánchez-Dehesa J. 2008 Anisotropic mass density by two-dimensional acoustic metamaterials. New journal of physics 10, 023004.
  38. [38]Antonakakis T, Baida FI, Belkhir A, Cherednichenko K, Cooper S, Craster R, Demésy G, Desanto J, Granet G, Gralak B et al.. 2014 Gratings: Theory and numeric applications. .
  39. [39]Allemang RJ. 2003 The modal assurance criterion–twenty years of use and abuse. Journal of Sound and vibration 37, 14–23.
  40. [40]Probst RA, Steinmetz T, Wu Y, Grupp F, Udem T, Holzwarth R. 2017 A compact echelle spectrograph for characterization of astro-combs. Applied Physics B 123, 1–8.
  41. [41]Barnard TW, Crockett MI, Ivaldi JC, Lundberg PL. 1993 Design and evaluation of an echelle grating optical system for ICP-OES. Analytical Chemistry 65, 1225–1230.
  42. [42]Loewen EG, Popov E. 2018 Diffraction gratings and applications. CRC Press.
  43. [43]Chatzopoulos Z, Palermo A, Guenneau S, Marzani A. 2022 Cloaking strategy for Love waves. Extreme Mechanics Letters 50, 101564.

Paper details

vlnc:W_bt8ecfx9
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents