Perspective on Non-Hermitian Elastodynamics

Abstract

The manipulation of mechanical waves is a long-standing challenge for scientists and engineers, as numerous devices require their control. The current forefront of research in the control of classical waves has emerged from a seemingly unrelated field, namely, non-Hermitian quantum mechanics. By drawing analogies between this theory and those of classical systems, researchers have discovered phenomena that defy conventional intuition and have exploited them to control light, sound, and elastic waves. Here, we provide a brief perspective on recent developments, challenges and intricacies that distinguish non-Hermitian elastodynamics from optics and acoustics. We close this perspective with an outlook on potential directions such as topological phases in non-Hermitian elastodynamics and broken Hermitian symmetry in materials with electromomentum couplings.

I Introduction

The past two decades have shown that the properties of artificial materials can be tailored to exhibit extraordinary dynamic behavior and properties by cleverly engineering their composition and structure. The development of such metamaterials is a prominent thrust in engineering today Cummer, Christensen, and Alù (2016); Haberman and Guild (2016); Kochmann and Bertoldi (2017); Kadic et al. (2019); Bigoni et al. (2013); Christensen et al. (2015); Celli and Gonella (2015); Nadkarni, Daraio, and Kochmann (2014); Parnell, Norris, and Shearer (2012). The principle focus of metamaterial research is tailored wave control based on the design of subwavelength structure. Elastic waves are of particular interest given the numerous mechanical applications that require their control, such as vibration isolation, impact mitigation, ultrasonography, energy harvesting, and stealth, to name just a few.

Currently, the forefront of research in the control of classical wave motion emerged from a seemingly unrelated field, namely, quantum mechanics, with the development of its non-Hermitian formalismBender and Boettcher (1998); Mostafazadeh (2010); Moiseyev (2011). This formalism describes open quantum systems that exchange energy with their environment, resulting with non-orthogonal or even colinear natural modes and degenerate eigenvalues in contrast with Hermitian systems whose natural modes are orthogonal. By drawing analogies between this formalism and those of classical systemsFeng, El-Ganainy, and Ge (2017); El-Ganainy et al. (2018); Özdemir et al. (2019); Miri and Alù (2019); Pile (2017); Midya, Zhao, and Feng (2018); El-Ganainy et al. (2019), researchers have discovered phenomena that defy conventional intuition and have used them to control light El-Ganainy et al. (2007); Klaiman, Günther, and Moiseyev (2008); Guo et al. (2009); Rüter et al. (2010); Longhi (2017a); Goldzak, Mailybaev, and Moiseyev (2018); Zhao and Feng (2018), sound Zhu et al. (2014); Fleury, Sounas, and Alù (2015); Cummer, Christensen, and Alù (2016); Shi et al. (2016); Xu et al. (2020); Thevamaran et al. (2019), and elastic waves Christensen et al. (2016); Hou and Assouar (2018); Merkel, Willatzen, and Christensen (2018); Hou, Ni, and Assouar (2018); Psiachos and Sigalas (2018); Merkel et al. (2018). Of all the branches of classical physics, these concepts were least studied in elastodynamics, in spite of (or potentially due to) the fact that elastodynamics exhibits a distinct tensorial richness. In this perspective, we discuss the intricacies and challenges that distinguish non-Hermitian elastodynamics from optics and acoustics, review some of the recent developments in this field, and present an outlook to potential directions from the current state-of-the-art.

II Background

Eigenvalue problems are ubiquitous in physics: they describe the natural modes of the relevant system using a suitable operator. Formally, they may be written as $\mathcal{H}\phi=\alpha\phi$ where $\mathcal{H}$ is an operator acting on the natural mode $\phi$, generating its multiplication by the (possibly complex) eigenvalue $\alpha$. In quantum mechanics, the operator is called the Hamiltonian, which operates on the wave function in the Schrödinger equation. One of the fundamental postulates in quantum mechanics is that the Hamiltonian exhibits a mathematical symmetry termed ‘Hermiticity’ which ensures real eigenvalues and is associated with energy conservationChew (2008); Srivastava (2015); Pernas-Salomón and Shmuel (2020). On a basic level, Hermiticity implies that for any two functions $\phi$ and $\varphi$ in a function space endowed with an inner product $\langle\cdot,\cdot\rangle$, the Hermitian operator $\mathcal{H}$ satisfies

$$\langle\mathcal{H}\phi,\varphi\rangle=\langle\phi,\mathcal{H}\varphi\rangle.$$ (1)

Strikingly, Bender and Boettcher (1998) discovered that Hamiltonians with so-called $\mathcal{PT}$ symmetry may also support real spectra. Unlike the eigenmodes of Hermitian operators, the eigenmodes of $\mathcal{PT}$-symmetric operators are no longer orthogonal one to another. In the extreme case, these modes may even coalesce together with their eigenvalues at the so-called ‘exceptional points’ (EPs) Zhao, Longhi, and Feng (2015); Longhi (2017b, a). These non-Hermitian degeneracies arise when suitable parameters of non-Hermitian systems are appropriately tuned. Near EPs, the functional dependency of the eigenvalues on these parameters defines self-intersecting Riemann sheets, as shown in Fig. 1.

[figure]style=plain,subcapbesideposition=top

[]
[]

Figure 1: The (a) real and (b) imaginary parts of the eigenvalues as functions of two suitable parameters of a non-Hermitian operator. For critical values of the two parameters, the Riemann sheets degenerate to a so-called exceptional point. The gray curves depict a loop around the EP which results in a different state, owing to the multi-valued nature of the Riemann sheet. Conventional research on non-Hermitian systems considers the natural frequencies as the eigenvalues and requires material gain (or loss) to be one of the parameters. In a recent paradigm shiftLustig et al. (2019); Mokhtari et al. (2020); Fishman et al. (2024), the EPs are designed in the wavevector space, using the unique coupling of elastic wave polarizations. Thus, the EPs are formed without material gain or loss, and the parameters that control the wavevector are the tuning parameters.

EPs and the unique topology around them are the source of fascinating counterintuitive phenomena such as chiral modesPeng et al. (2016), supersensitivityHodaei et al. (2017) and unidirectional zero reflection Shen et al. (2018). While the bulk of the research (and the above discussion) considers $\mathcal{PT}$-symmetric Hamiltonians, they belong to a larger class of non-Hermitian HamiltoniansMostafazadeh (2002, 2010); Suchkov et al. (2016); Lustig et al. (2019) that can exhibit real eigenvalues and EPs at critical values of a suitable parameter set. Understanding that these phenomena rely upon the nature of eigenvalue problems associated with the wave operator has led to their dissemination in other wave physics. We summarize next the fundamental operators in the different classical systems in order to highlight the uniqueness of non-Hermiticity in elastodynamics.

In photonics, Maxwell’s equations can be expressed in the form of an eigenproblem Griffiths (1999); Joannopoulos et al. (2008) for the magnetic eigenmodes $\mathbf{H}$ and their frequencies $\omega^{2}$ (the electric field can be determined subsequently). The corresponding wave operator is a function of the second-order dielectric tensor $\bm{\epsilon}$ (since the permeability of most materials is very close to the vacuum permeabilityJoannopoulos et al. (2008)). Magnetic waves are subjected to the constraint $\nabla\cdot\mathbf{H}=0,$ hence do not support longitudinal modes and are classified as transverse waves.

In acoustics, i.e., in the study of waves in gases and liquids without resistance to shear deformationKinsler et al. (2000), the eigenvalue problem can be expressed as a scalar equation for the pressure field, whose operator depends on the mass density¸ $\rho$, and bulk modulus, $K$, of the fluid in which waves propagateKinsler et al. (2000). The velocity field $\mathbf{v}$ of acoustic waves is subjected to the constraint $\nabla\times\mathbf{v}=\mathbf{0}$, and thus do not support transverse modes. Acoustic waves are longitudinal pressure waves, which are also referred to as dilatational or volumetric wavesGraff (1975), since they are accompanied with volume change, in contrast to transverse waves, i.e., shear waves which are isochoric.

In solid mechanics, the elastodynamics equations can be formulated as an eigenvalue problem for the time- and space- dependent displacement vector field $\mathbf{u}(\mathbf{x},t)$ of material points (see Appendix A). The wave operator is a function of the mass density and the elasticity tensor of the solid, $\mathbf{C}$. Its simplest form is the Christoffel equationAuld (1973); Nayfeh (1995) for bulk plane waves in the $\mathbf{n}$ direction: $k^{2}\rho^{-1}\bm{\Gamma}(\mathbf{u})=\omega^{2}\mathbf{u}$ where $k$ is the wavenumber and the components of the Christoffel operator are $\Gamma_{il}=C_{ijkl}n_{j}n_{l}$. The first source of richness originates from the dimension of $\mathbf{C}$, which is a fourth order tensor and thus constitutes a larger design space in comparison with photonics and acoustics. Second, unlike the magnetic field (subjected to $\nabla\cdot\mathbf{H}=0$) and the velocity field in acoustics (subjected to $\nabla\times\mathbf{v}=\mathbf{0}$) the displacement field $\mathbf{u}\left(\mathbf{x},t\right)$ is not subjected to any differential constraint. Thus, even the simplest solids (conservative, isotropic and homogeneous) support both transverse and longitudinal wave polarizations. These inherent modes can be coupled one with the other, a coupling that gives rise to unique physics Long, Ren, and Chen (2018).

Most works on non-Hermitian physics focus on problems of the type mentioned above, in which the frequency (squared) is the eigenvalue and corresponds to the propagation of free waves owing to some impulsive excitation, i.e., in the absence of a constant excitation. Assuming real wavenumbers (no radiation losses), the breaking of Hermitian symmetry (i.e., accessing complex frequencies and nonorthogonal eigenmodes) in these problems requires material loss or gain. While optical gain is well-established using stimulated emission, generating and controlling elastic gain remains a challenge. We list the recent advances for tackling this challenge in the first part of Sec. III, together with their applications.

Less explored from the perspective of non-Hermitian physics, a dual class of problems involves a formulation where the wavenumber is the eigenvalue and the frequency is a prescribed real quantity (see Appendix A). This formulation pertains to conservative media with open boundaries and sustained driving sources. Recent progress has utilized the unique coexistence and coupling of wave polarizations in elastodynamics to engineer EPs in such conservative solids, thus circumventing the challenges associated with the design of material gain and lossLustig et al. (2019); Mokhtari et al. (2020); Fishman et al. (2024). The second part of Sec. III expands on the works that introduced this approach and how they capitalized on emergent non-Hermitian features to manipulate elastic waves.

III Recent developments

As discussed in Sec. II, the common approach for breaking Hermitian symmetry requires the careful design of material loss and gain. Piezoelectric materials are therefore natural candidates to achieve non-Hermitian behavior in elastodynamic systems due to their ability to convert electric energy to mechanical energy and vice versa. In a series of works Willatzen and Christensen (2014); Christensen and Willatzen (2015); Christensen (2016); Gao, Willatzen, and Christensen (2020), Christensen and collaborators developed a procedure that exploits the acousto-electric effect in piezoelectric semiconductors for material gain, thereby synthesizing $\mathcal{PT}$-symmetric elastic media Christensen et al. (2016) and breaking elastodynamic reciprocity Merkel, Willatzen, and Christensen (2018). When an acoustic field impinges on a piezoelectric semiconductor slab a coherently oscillating electric charge is created. Superposing a sufficiently high DC electric field corresponding to a supersonic carrier drift speed leads to sound amplification by virtue of phonon emission, an effect known as acoustic Cherenkov radiation. In particular, this acousto-electric effect has been employed for two demonstrations. First, stacking multiple piezoelectric semiconductors was shown to generate a system with balanced loss and gain with associated non-Hermitian properties, if the layers are adequately electrically loaded Christensen et al. (2016). Next, a theoretical non-Hermitian Su-Schrieffer-Heeger demonstration was made using the same approach. In that case, the non-Hermitian skin effect and the ensuing failure of both the Bloch band topology were demonstrated Gao, Willatzen, and Christensen (2020). We note, however that experimental demonstrations of the latter theoretical prediction has not yet been provided.

A more conventional approach to use piezoelectric materials is to shunt them through electrical circuits of resistors, capacitors, and inductors Gardonio and Casagrande (2017), which collectively generate tunable elastic gain lossHou and Assouar (2018); Gardonio and Casagrande (2017). Assouar’s group proposed a tunable $\mathcal{PT}$-symmetric elastic beam based on such shunted piezoelectric elementsHou and Assouar (2018). They later employed this concept for negative refraction of flexural wavesHou, Ni, and Assouar (2018). Similarly, Huang’s group employed shunted piezoelectric patches attached to a $\mathcal{PT}$-symmetric elastic beam that displayed asymmetric flexural wave scatteringWu, Chen, and Huang (2019).

Ruzzene and colleagues have also used shunted piezoelectric arrays to generate spatiotemporal modulation of the elastic modulus, thereby breaking reciprocal elastic wave transport for flexural wavesMarconi et al. (2020); Xia et al. (2021). This approach was later employed by Erturk and colleagues.Thomes, Rosa, and Erturk (2024) to demonstrate the formation of a resonant EPs. Such materials that violate time invariance can exhibit a wide array of nonreciprocal wave phenomena such as one-way wave amplification and attenuationNassar et al. (2017); Torrent, Parnell, and Norris (2018); Nassar et al. (2020). From a modeling perspective, the constitutive parameters and phase velocity become time-dependent. These works specifically considered modulations in the form of progressive periodic waves (’pump waves’), which create a spatiotemporal bias and nonreciprocal wave motion as a function of the modulation. Small-amplitude medium-speed modulations lead to a Bragg scattering regime; faster modulations can be described by an equivalent medium with effective properties; and slower but higher-amplitude modulations lead to an adiabatic regimeNassar et al. (2020). This nonreciprocal wave propagation is fundamentally different from one-way Bragg reflection, but can still be leveraged for the purposes of one-way mirroring, especially at low frequencies where wide bandgaps are seldom available. In this scenario, Hooke’s law no longer applies and must be replaced by constitutive relations of the Willis formWillis (1981, 2011, 2012), which state that the stress and linear momentum depend both the strain and velocity fields. The Willis parameters in the case of spatiotemporal modulations capture the nonreciprocal nature of the modulated microstructuresNassar et al. (2017). Indeed, Willis materials offer another platform for breaking the Hermitian nature, as we discuss later.

Collectively, the works above have tackled the important challenge of generating gain in elastic materials in order to break temporal hermiticity. However, gain is not a necessity for realizing EPs and some of their associated features are accessible using loss alone, an inherent feature of viscoelastic materials. Indeed, Shmuel and Moiseyev (2020) showed that the introduction of a viscoelastic part in an elastic slab may generate EP of two modes (EP2) in the spectrum of the assembly. They applied non-Hermitian perturbation theory previously developed in quantum mechanics to determine the conditions under which this occurs and employed the resultant topology around the EP for mass sensing with enhanced sensitivity. Similarly, Domínguez-Rocha et al. (2020) used differential loss instead of gain and loss to form EP2. While the modes they analyzed were structural modes (torsional modes of pillars) rather than elastodynamics modes, their work is one of the few experimental demonstrations of the sensitive frequency splitting near EP of a mechanical system.

The nonconservative nature of the media presented so far manifests itself through time-dependent constitutive equations and hence complex elastic moduli. These, in turn, yield complex frequencies for the Christoffel equation. Another type of nonconservative media are Cauchy-elastic mediaOgden (1997): the work done by the stress in such media generally depends on the deformation path, and therefore cannot be derived from a strain energy function. As a result, the elasticity tensor no longer exhibits the major symmetry that is needed to ensure real frequencies in the Christoffel equation. Under the name ’odd elasticity’, Vitelli and colleaguesScheibner et al. (2020); Scheibner, Irvine, and Vitelli (2020) set forth a model of active material with nonconservative microscopic interactions that generate odd elastic moduli, and studied the non-Hermitian elastodynamics of such materials with odd elasticity (top circle in Fig. 2). They demonstrated that in isotropic solids with odd elasticity, the wave polarizations (eigenmodes) are no longer orthogonal, and may even become colinear for a threshold value of the elastic moduli. This EP marks the transition to odd-elastic waves with circular polarizationScheibner et al. (2020).

Figure 2: Possible non-Hermitian constitutive operators in elastodynamics. Elastic materials may exhibit odd elastic tensor ($\mathbf{C}=-\mathbf{C}^{\text{T}}$), Willis materials may exhibit odd Willis- ($\hat{\mathbf{S}}=-\mathbf{S}^{\text{T}}$) and density ($\bm{\rho}=-\bm{\rho}^{\text{T}}$) tensors, and electromomentum materials may exhibit odd electromomentum tensors ($\hat{\mathbf{W}}=-\mathbf{W}^{\text{T}}$).

Active material systems displaying odd elasticity have been experimentally demonstrated by introducing piezoelectric elements and motors controlled by electrical circuits, or spinning networks into host media Brandenbourger et al. (2019); Chen et al. (2021). Chen et al. (2021) 1D non-Hermitian metamaterial in which each unit cell consists of three piezoelectric patches mounted on a steel beam. The metabeam was assumed to have two modes of deformation including bending and shearing. These modes of deformation in turn induce a shear stress $\sigma$ and a bending moment, $M$. The crucial difference between this non-Hermitian beam and a traditional beam is the presence of internal energy sources that violate energy conservation. By designing the feedback to create non-reciprocal coupling between the elongation $s$ and shear $b$, the constitutive relation of the metabeam takes the form of odd elasticity

$$\left[\begin{array}[]{c}\sigma\\ M\end{array}\right]=\left[\begin{array}[]{ll}\mu&P\\ 0&B\end{array}\right]\left[\begin{array}[]{l}s\\ b\end{array}\right],$$ (2)

i.e., the electronic feedback between the piezoelectric patches induces a new modulus $P$, in addition to the shear and bending moduli $\mu$ and $B$, respectively. Because the energy differential is $\ \delta\psi=\sigma\delta s+M\delta b$, the asymmetric part of the matrix corresponds to a violation of Maxwell–Betti reciprocity. The parity-violating, nonconservative modulus $P$ also induces unidirectional wave amplification. For finite structures, these non-Hermitian systems with odd elasticity exhibit non-Hermitian 1D and 2D skin effect and non-Hermitian Rayleigh wave propagation Chen et al. (2021); Gao, Qu, and Christensen (2022). The skin effect in the non-Hermitian system are defined as the case where bulk modes behave as the skin modes collapsing at the open boundaries. Recently, the design of active microstructure for a 2D non-Hermitian odd plate with odd density through feed-forward interactions was suggested to explore a series of unconventional wave phenomena when conventional Timoshenko plate mechanics meets with non-HermiticityWang, Meng, and Chen (2023); Wu et al. (2023). Those works numerically and experimentally demonstrated nonreciprocal wave amplification and attenuation phenomena along with direction-dependent dispersion control of flexural waves in 2D odd plates.

The approaches mentioned so far to design non-Hermitian elastodynamics can be generalized using Willis materialsWillis (1981, 1997, 2011), which exhibit a different conversion of strain energy and kinetic energy relative to conventional materialsSieck, Alù, and Haberman (2017). This distinct mechanism is reflected by the Willis tensors $\mathbf{S}$ and $\mathbf{\hat{S}}$ and tensorial mass density $\bm{\rho}$ appearing in nonlocal constitutive equations, analogous to the bianisotropic equations in electromagnetismMilton, Briane, and Willis (2006); Sieck, Alù, and Haberman (2017). In the spatially local limit, the constitutive (Milton-Briane-WillisMilton, Briane, and Willis (2006); Milton (2020)) equations take the form $\sigma_{ij}=C_{ijkl}u_{k,l}+S_{ijk}\dot{u}_{k}$ and $p_{i}=\rho_{ij}\dot{u}_{j}+\hat{S}_{ijk}u_{j,k}$. Extending the concept of odd elasticity to odd mass density and odd Willis tensors thus breaks Hermitian symmetry in the Christoffel equation, giving rise to complex frequencies. This translates to designing a Willis material such that one of the symmetries $\ \rho_{ij}=\rho_{ji},\ \hat{S}_{ijk}=S_{kij}$ is broken (left circle in Fig. 2). Huang and collaboratorsWu et al. (2023) designed and experimentally realized such active microstructure that gives rise to odd mass density. They further demonstrated the formation of EPs of transition between stable and unstable waves, directional wave amplification and non-Hermitian skin effect in the medium. In another collaborationChen et al. (2020), they experimentally realized odd Willis couplings in active flexural media using piezoelectric sensor–actuator pairs controlled with digital circuits, giving rise to nonreciprocal wave propagation. Li and collaboratorsLiu et al. (2019) introduced loss to a Willis beam, and showed the formation of EP in the scattering matrix for flexural waves, at which there is unidirectional zero reflection occurs.

Building upon these works probing gain and loss in elastic materials, Psiachos and Sigalas (2018) showed that coupled transverse-longitudinal elastic waves scatter asymmetrically from alternating layers with gain and loss. Importantly, they later realized that the coupling between transverse and longitudinal waves is sufficient to generate asymmetric scattering even without gain and lossPsiachos and Sigalas (2019).

All the developments that are listed above rely upon material gain, loss or their combination in order to access non-Hermitian features in elastodynamics. Recently, Shmuel’s groupLustig et al. (2019) introduced a unique paradigm to eliminate the need for material gain or loss by breaking spatial Hermitian symmetry using the tensorial nature that is unique to elastodynamics. From a mathematical perspective, they designed the non-Hermitian part of the operator for the wavenumbers instead of the operator for the natural frequencies (see Appendix A). From a mechanical perspective, the analysis is of a conservative solid with open boundaries and sustained driving source. More specifically, they revisited the canonical scattering problem of monochromatic plane waves impinging on a semi-infinite periodic laminate made of two conservative isotropic materialsJoseph and Craster (2015). By designing the unit cell, they tuned the energy transfer between the two elastic polarizations at the material interfaces such that two of the forward Bloch modes coalesceLustig et al. (2019) (Fig. 3a). [figure]style=plain

[]

Figure 3: (a) The coalescence of two Bloch modes of the exemplary conservative laminate considered by Lustig et al. (2019). The EPs are shown in a cut in the frequency-Bloch wavenumber $(f,k_{\mathrm{B}})$ diagram for normalized vertical wavenumber $k_{2}h=0.5$ about 181.2 kHz. Re$\,k_{\mathrm{B}}$ and Im$\,k_{\mathrm{B}}$ are shown in solid and dashed curves, respectively. The real and imaginary parts of a certain branch are plotted using the same color. (b) One of the Riemann sheets associated with this EP in the wavevector space. The solid and dashed curves illustrate the asymmetric nature of a loop around the EP. One direction (solid curve) is adiabatic and the other (dashed curve) is non-adiabatic, resulting in a different state at the end of the loop. (c) Full-wave finite element simulations of the axial energy flux in the pertinent scattering problem near the EP. The energy flux is preserved in the laminate and exhibits negative refraction. See also supplementary video 1.

The EPs that are associated with this non-Hermitian eigenmode degeneracy are formed in the complex wavevector space, in contrast with the gain and complex frequency parameter space as in the works discussed above. These non-resonant EPs are surrounded by a self-intersecting Riemann surface (Fig. 3b), whose curvature is the related to the energy flow and therefore the unique topology of these surfaces can be leveraged for anomalous energy transport. For example, Lustig et al. (2019) showed that these spatial EPs give rise to negative refraction (Fig. 3c), as they are the crossing of branches of opposite slopes. The key point is that, contrary to previous works discussed above, the parameters of the Riemann surface about the non-resonant EPs do not correspond to material gain or loss, but, e.g., the angle of the incident wave as depicted in the left half-space in Fig. 3c). Subsequently, Srivastava’s groupMokhtari et al. (2020) developed designated tools to further analyze the non-Hermitian operator, its EPs and scattering spectrum, suggesting its connection to resonance trapping.

In a subsequent workFishman et al. (2024), Shmuel’s group designed the coalescence of three Bloch modes (EP3) by replacing one of isotropic materials in the unit cell with anisotropic one. This non-resonant higher degeneracy further expands the toolbox for elastic wave shaping by forming waves with zero axial group velocity and finite transmittance (known as ’axially frozen modes’Figotin and Vitebskiy (2003)), which can even reach unity (Fig. 4b and supplementary video 2). Notably, these modes, which were previously discovered in 3D dielectric laminatesFigotin and Vitebskiy (2003), are now accessible in simpler, planar settings in elastodynamics, thanks to its distinct tensorial richness.

[figure]style=plain

[]

[]

Figure 4: (a) Coalescence of three Bloch modes of the exemplary conservative laminate considered by Fishman et al. (2024). This EP3 is shown in a cut in the $(f,k_{\mathrm{B}})$ diagram for normalized vertical wavenumber $k_{2}h=0.15$ about 4.15 kHz. Re$\,k_{\mathrm{B}}$ and Im$\,k_{\mathrm{B}}$ are shown in solid and dashed curves, respectively. The real and imaginary parts of a certain branch are plotted using the same color. (b) Full-wave finite element simulations of the axial energy flux in the pertinent scattering problem near the EP, giving rise to axially frozen modes with unity transmittance (see also Supplementary video 2).

Additional useful phenomena are accessible by combining non-resonant EPs with gain and loss. For example, such combination enables the encirclement of non-resonant EPs in a suitable parameter space, a process that is known to be asymmetrically non-adiabaticHeiss (2016); Doppler et al. (2016); Xu et al. (2016). This means that the physical system will follow its instantaneous eigenmode when subjected to a loop in the parameter space, only along one of the two possible orientations of the loop. The solid (counterclockwise) and dashed (clockwise) curves in Fig. 3b demonstrate such adiabatic and non-adiabatic state evolution, respectively. Elbaz et al. (2022) carried out such an encirclement around non-resonant EP2 of forward and backward longitudinal Bloch waves in elastic laminates, using spatial modulation of gain and loss. They discovered that the starting point of the loop governs several unusual features. For example, the laminate may act as a source or a sink of energy, exhibit reflectance greater than unity, and accommodate spatial asymmetry in the energy flow with respect to the incidence direction, depending on that starting point. Encircling EPs provides another unexpected benefit in the form of a highly efficient algorithm for sorting eigenvalue bands, as observed by Lu and Srivastava (2018), who exploited EPs to distinguish real crossings from level repulsion zones in the real phononic spectrum.

IV Future directions

A future avenue that has been thoroughly explored in acoustics and photonics concerns non-Hermitian topology Zhang et al. (2023). Non-Hermitian topology refers to the study of topological properties in systems that do not satisfy the Hermitian symmetry. In other words, non-Hermitian topology explores the emergence of topological phenomena in systems containing lossy and amplifying components. One of the key concepts in this field is the non-Hermitian skin effect, which refers to the localization of eigenstates at the boundary of a system. As we know, the Bloch wave description is a powerful concept used to describe periodic states and resonances. However, in non-Hermitian systems, where gain and loss mechanisms are present, the Bloch wave description encounters challenges. Using planar mechanical metamaterials, several numericalChen et al. (2019) and experimental Wang, Wang, and Ma (2022); Wang, Meng, and Chen (2023) works demonstrated how the eigenstates become localized at the boundary rather than uniformly distributed throughout the bulk. Targeting elastodynamic non-Hermitian bulk states at interfaces beyond the plane remain an intriguing area of research to pursue. Other exciting directions include combining lattice symmetry and non-Hermiticity and other phenomena that are elusive in elastodynamics such as non-Hermitian Weyl exceptional rings, non-Hermitian higher-order topology with vibrating complex corner states, and non-Abelian permutationsZhang et al. (2023) of mechanical states.

A completely uncharted direction is the breaking of Hermitian symmetry in materials that display electromomentum couplingsPernas-Salomón and Shmuel (2020, 2020); Pernas-Salomón et al. (2021); Muhafra et al. (2022); Kosta et al. (2022); Muhafra, Haberman, and Shmuel (2023). The electromomentum couplings were theoretically discovered by Pernas-Salomón and Shmuel (2020) using a generalization of Willis’ dynamic homogenization methodWillis (2011) to composites made of constituents that mechanically respond to non-mechanical stimuli. For the case of piezoelectric constituents, the method revealed that the electric displacement field ($\mathbf{D}$) constitutively depends on velocity and that the macroscopic linear momentum constitutively depends on the electric field ($-\nabla\phi$). These two couplings are captured in terms of two second-order tensors ($\mathbf{W}$ and $\hat{\mathbf{W}}$) called the electromomentum tensors, in direct analogy with the Willis tensors. These coupling tensors capture macroscale effects resulting from spatial symmetry-breaking on the microscale and nonlocal interactionsPernas-Salomón and Shmuel (2020, 2020); Pernas-Salomón et al. (2021). In symbolic matrix notation, the spatially local limit of these constitutive equations is

$$\left(\begin{array}[]{c}\bm{\sigma}\\ \mathbf{D}\\ \mathbf{p}\end{array}\right)=\left(\begin{array}[]{ccc}\mathbf{C}&\hat{\mathbf{B}}&\hat{\mathbf{S}}\\ \mathbf{B}&-\mathbf{A}&\mathbf{W}\\ \mathbf{S}&\hat{\mathbf{W}}&\bm{\rho}\end{array}\right)\left(\begin{array}[]{c}\nabla\mathbf{u}\\ \nabla\phi\\ \dot{\mathbf{u}}\end{array}\right),$$ (12)

where $\mathbf{A}$ is the permittivity tensor, $\mathbf{B}$ and $\hat{\mathbf{B}}$ capture the direct- and inverse piezoelectric effect; collectively, these equations take a trianisotropic formShmuel et al. (2022). Note that the constituents of such composites are piezoelectric and do not display Willis nor electromomentum coupling. To date, studiesDanawe and Tol (2023); Wallen et al. (2022); Lee, Zhang, and Gu (2023); Huynh et al. (2023) have been restricted to materials with electromomentum tensors that are Hermitian adjointsPernas-Salomón and Shmuel (2020), which implies that the local electromomentum tensors satisfy $\hat{W}_{ij}=W_{ji}$. However, considering the works listed in Sec. III which exploit piezoelectricity to generate material gain, this platform naturally lends itself to non-Hermitian physics, the simplest of which is of odd electromomentum tensors ($W_{ij}=-\hat{W}_{ji}$, right circle in Fig. 2). The mplications of breaking this symmetry on elastic waves and how to realize designs with odd electromomentum tensors are yet to be explored, and constitute excellent candidates to extend this approach for elastic wave control.

Appendix A Non-Hermitian operators in elastodynamics

The dynamics of a solid occupying some domain $\Omega$ and subjected to the body force density $\mathbf{f}$ is governed by the balance of linear momentum $\sigma_{ij,j}+f_{i}=\dot{p}_{i}$, where $\bm{\sigma}$ and $\mathbf{p}$ are the Cauchy stress tensor and linear momentum density, respectively. In standard linear lossless and passive solids, the stress and linear momentum density constitutively depend on the displacement gradient $\nabla\mathbf{u}$ and velocity $\dot{\mathbf{u}}$ via the relations $\sigma_{ij}=C_{ijkl}u_{k,l}$ and $p_{i}=\rho\dot{u}_{i}$, where $\mathbf{C}$ is the (possibly anisotropic) elasticity tensor with major symmetry (owing to energy conservationOgden (1997)) and $\rho$ is the mass density. When $\Omega$ is infinite (in 1,2, or 3 dimensions) and homogeneous, the equations simplify to $C_{ijkl}u_{k,lj}+f_{i}=\rho\ddot{u}_{i}$ and admit homogeneous solutions of the form $\mathbf{u}=\tilde{\mathbf{u}}\exp(ikn_{j}x_{j}-\omega t)$ where $\omega$ is the frequency and $\bm{k}=kn_{j}\mathbf{e}_{j}$ is the wavevector with wavenumber $k$ and $\mathbf{e}_{j}$ denotes the direction of propagation. On substituting this ansatz in the equations of motion, one can formulate various generalized parametric eigenvalue problems which can be represented by linear operators. The simplest of these is $\mathbf{A}(\mathbf{k})\tilde{\mathbf{u}}=\lambda\mathbf{B}\tilde{\mathbf{u}}$ where in index notation the operators are $\lambda=-\omega^{2}$ and $A_{ij}=-k^{2}C_{ijkl}n_{j}n_{l},\,B_{ij}=\rho\delta_{ij}$. In the absence of loss and gain and if $\mathbf{k}$ is real, this represents a Hermitian (or self-adjoint) problem in the following sense: if an inner product is defined as $\langle\tilde{\mathbf{u}},\tilde{\mathbf{v}}\rangle=\tilde{u}_{i}\tilde{v}_{i}^{*}$ then $\langle\mathbf{A}\tilde{\mathbf{u}},\tilde{\mathbf{v}}\rangle=\langle\tilde{\mathbf{u}},\mathbf{A}\tilde{\mathbf{v}}\rangle$ where we note that $\mathbf{B}$ is clearly Hermitian in this sense. This gives rise to exclusively real $\omega$ solutions for the eigenvalue problem considered. In standard elastodynamics literatureAuld (1973); Nayfeh (1995); Carcione (2001), this standard eigenvalue problem for $\lambda=\omega^{2}$ is known as the Christoffel matrix equation $[k^{2}\mathsf{\Gamma}]\{\mathsf{u}\}=\rho\omega^{2}\{\mathsf{u}\}$ where the components of the acoustic (or Christoffel) matrix are $\Gamma_{il}=C_{ijkl}n_{j}n_{l}$. The major symmetry of $\mathbf{C}$ implies that the real square matrix $\mathsf{\Gamma}$ is symmetric, hence has real eigenvalues $\omega^{2}$ and orthogonal eigenvectors. As discussed in Secs. II and III, most works on non-Hermitian physics focus on this type of problems in which the eigenvalue is the natural frequency (squared), and use gain (or loss) to break Hermitian symmetry. This temporal energy transfer is modeled by complex $\mathbf{C}$ (or $\rho$), in which case $\mathbf{A}$ (equivalently, $\mathsf{\Gamma}$) is no longer Hermitian and therefore the solutions for $\omega$ are complex-valued in general even for real-valued $\mathbf{k}$.

In a dual class of problems, the balance of linear momentum is formulated as an eigenvalue problem for the wavenumber and the frequency is a prescribed real quantity. Mathematically, $\mathbf{A}$ is considered as a linear operator that depends upon a set of complex parameters $(\mathbf{k},\omega)$ and in this formulation the possibly complex wavevectors $\mathbf{k}$ are viewed as those resulting from the restriction of $\mathbf{A}(\omega);\omega\in\mathbb{C}$ to $\mathbf{A}(\omega);\,\omega\in\mathbb{R}$. Physically, this formulation models structured conservative media subjected to sustained driving sources and open boundaries, such that there is no temporal energy loss or gain (real frequencies); the complex nature of the wavenumbers models spatially evanescent modes that accommodate boundary (or continuity) conditions at material interfacesGraff (1975); Srivastava and Willis (2017), and implies that the corresponding operator is non-Hermitian even in the absence of gain or lossMokhtari, Lu, and Srivastava (2019). It is sufficient to illustrate this by way of a simple planar example. Consider a medium that is $h$-periodic along $x_{1}$ such that the Bloch ansatz applies: $\mathbf{u}=\mathbf{\tilde{u}}(x_{1})\exp(ikn_{j}x_{j}-\omega t)$ where $\mathbf{\tilde{u}},\,\rho,\,\mathbf{C}$ are $h$-periodic. Before we discuss the non-Hermitian problem, it is imperative to recall that the mainstay of phononic crystals research was the classical band structure $\omega(\mathbf{k})$ emanating from the eigenvalue problem $\mathbf{A}(\mathbf{k})\tilde{\mathbf{u}}=\lambda\mathbf{B}\tilde{\mathbf{u}}$ with $\lambda=-\omega^{2}$ at real $\mathbf{k}$, where now $\mathbf{A}$ and $\mathbf{B}$ are linear differential operators. The restriction $\mathbf{k}\in\mathbb{R}$ applies to infinite periodic media, whose translational invariance prohibits complex-valued $\mathbf{k}$, and results with real $\omega$ solutions in the absence of gain or lossMokhtari, Lu, and Srivastava (2019). This Hermitian problem has orthogonal eigenmodes and complete basis with respect to the inner product $\langle\tilde{\mathbf{u}},\tilde{\mathbf{v}}\rangle=\int_{h}\tilde{u}_{i}\tilde{v}_{i}^{*}dh$.

The operator loses its Hermitian nature once $\mathbf{k}$ is extended to the complex plane. As we discussed above, complex $\mathbf{k}$ has a clear physical meaning of evanescent waves in interface problems. For example, complex $\mathbf{k}$ captures the modes that would evolve in the periodic medium if it is truncated at some position $x_{1}$, and subjected to incoming waves from a homogeneous half-space occupying the volume $x_{1}<0$. Thus, we further specialize our problem to a scenario where $k_{2}$ is real and prescribed (physically, by defining the incident wave), to obtain the generalized eigenvalue problem $\mathbf{A}(\omega,k_{2})\tilde{\mathbf{u}}=k_{1}\mathbf{B}(\omega,k_{2})\tilde{\mathbf{u}}$. As discussed in Sec. III, Lustig et al. (2019) highlighted the non-Hermitian nature of this classical problem, identifying EPs inside the Brillouin zone (Fig. 3a) and the Riemann sheets around them in the complex wavevector space (Fig. 3b). Mokhtari et al. (2020) later highlighted another aspect of the non-Hermitian nature, namely, the coalescence of the spatial eigenmodes at these EPs. Again, this problem is non-Hermitian in spite of the fact that the media is conservative (real symmetric elasticity tensor and real mass density). We reemphasize that of all the classical systems, EPs between forward waves in such simple settings (planar layered medium made of isotropic conservative materials) can form only in elastodynamicsLustig et al. (2019), owing to its distinct co-existence of transverse and longitudinal wave polarizations. Recent works in elastodynamicsKhodavirdi, Mokhtari, and Srivastava (2022); Khodavirdi, Ong, and Srivastava (2023) and acoustic-elastodynamic interactions leading to scatteringZhou et al. (2024) address such problems using open system formalism which leads to a non-Hermitian effective Hamiltonian.

We complete this Appendix recalling that Chauchy-elastic materials, Willis materials and electromomentum materials offer a richer platform to break Hermitian symmetry, even in the first type of eigenvalues problems for $\omega^{2}$ in infinite homogeneous media. If the major symmetry of $\mathbf{C}$ is broken, then the operator for $\omega$ is no longer Hermitian and complex solutions will form. Similarly, if a Willis material is designed such that one of the symmetries $C_{ijkl}=C_{klij},\ \rho_{ij}=\rho_{ji},\ \hat{S}_{ijk}=S_{kij}$ is broken, then the operator in the first type of problems is non-Hermitian, giving rise to complex $\omega$ solutions. In direct analogy, electromomentum couplings constitute an additional degree of freedom to break Hermitian symmetry: if the microstructure is designed such that $W_{ij}\neq\hat{W}_{ji}$, then the operator in the first eigenvalue problem is non-Hermitian and its eigen-frequencies are complex.