Harnessing normal-shear coupling in metabarriers for deep sub-wavelength underwater noise control Authors: Vinícius F. Dal Poggetto, Marco Miniaci Published: 2026 ABSTRACT Anthropogenic underwater noise poses a significant threat to marine ecosystems, disrupting key biological functions. Common mitigation strategies include enclosing noise sources within acoustic barriers. Current designs include locally resonant absorbers, which offer narrow-band performance, and reflective systems with limited effectiveness at low frequencies. In this work, we propose an approach to design thin anisotropic metamaterial-based acoustic barriers for broadband underwater noise attenuation at deep sub-wavelength scales using topology optimization to maximize the coupling between normal stresses and shear strains. Unlike conventional methods, the proposed optimization is formulated in the static regime, relying solely on the homogenized elastic properties of the structured material and not on the characteristics of the surrounding fluid. The resulting metabarriers achieve a high sound transmission loss (STL, 100 dB peak) above 2 kHz, while maintaining a thickness-to-wavelength ratio as low as 1/70 below 1 kHz and STL of approximately 20–30 dB. The influence of hydrostatic pressure on performance is also evaluated, and structural modifications for practical deployment are proposed. The results demonstrate the potential of anisotropy-driven metamaterials as compact and efficient solutions for the control of underwater noise, offering a promising avenue for future acoustic insulation technologies. Introduction The rise of marine renewable energy has re-sparked the interest in compact, pressure-resistant devices that can reduce underwater noise. Particularly at low frequencies, human activities such as offshore wind farms, shipping, and military operations increasingly disrupt marine life, interfering with vital behaviors in aquatic species, especially mammals1. Whenever directly addressing the noise source is not possible, an alternative solution is to place barriers between the source and the areas requiring protection. However, reducing noise with ultra-thin, sub-wavelength devices remains a major challenge due to the longer wavelengths (compared to airborne acoustics) and complex fluid-structure interactions taking place in water2,3. Thus, the development of low-frequency, robust noise barriers to protect marine ecosystems is more crucial than ever. The primary approaches for underwater acoustic insulation materials involve designing the impedance match (or mismatch) of these barriers with respect to the acoustic impedance of water, resulting in (i) absorptive or (ii) reflective behavior4. Common design strategies include fluid or solid-filled cavities, backings, resonators, periodic structures, and metamaterials—composite materials made of periodic arrangements of sub-wavelength structures5–7. When impedance matching is exploited, the use of barriers made of polymeric materials enhanced with viscoelastic micro- or macro-inclusions and coatings embedded with periodic air pockets is a widely used absorptive approach4,8,9. In this case, insulation performance results from the polymeric matrix material’s molecular chain relaxation, providing frequency-dependent vibration characteristics (e.g., mode shapes) and good performance at high frequencies. Low-frequency absorption, however, remains limited8. Another possible approach is the conversion of acoustic waves into shear waves, exploiting the high shear damping characteristic of certain materials and improving performance10. Whenever inclusions in the matrix are considered, the operating frequency bandwidth is constrained by their resonance frequencies11. To address this limitation, researchers have proposed optimized geometries and material compositions for resonating inclusions, along with tailored properties of the host material to improve attenuation efficiency and broaden the operational frequency range12. An alternative strategy to tuning mechanical local resonators, whose resonance frequency remains fixed over time, is the use of coupled electro-mechanical devices to achieve tunable local resonance mechanisms through shunted piezoelectric patches. In ref. 13, the authors demonstrate that the sound absorption of underwater metamaterials can be adjusted with the use of a negative capacitance in a shunt circuit to achieve quasi-perfect absorption at low frequencies. This concept was later expanded in ref. 14, with an internal acoustic cavity coupled to the shunt circuit, enabling better acoustic absorption performance through optimization of the cavity shape and shunt circuit parameters. Although relatively well-performing at mid and high frequencies, these approaches perform poorly below 1 kHz. Ensuring impedance matching between resonating elements (such as Helmholtz-like resonators) and water has shown potential as an alternative, at least in the proximity of resonance frequencies. Widely studied in the context of airborne acoustics15, their application in underwater environments has been less explored. Duan et al.16 numerically proved that an underwater quasi-Helmholtz resonator that includes a rubber coating inside its resonant cavity can lead to quasi-perfect absorption (>90%) in the 306–921 Hz frequency range. In another work, Duan et al.17 developed a theoretical model to predict the sound absorption performance of a sub-wavelength quasi-Helmholtz absorber using a fluid-filled square honeycomb core with an interior rubber coating, numerically reporting a semi-absorption band (α > 0.5) for the 279–478 Hz frequency range. Furthermore, Qu et al.5 proposed a structured array of Fabry–Pèrot resonators with an average thickness of 8.9 mm whose impedance matching to water achieves high absorption (α > 0.9) over a broad frequency range (4–20 kHz). Another explored strategy to increase the absorption spectra of underwater metamaterials is the use of multi-mode coupling. Gu et al.18 presented an underwater sound-absorbing metasurface whose modal coupling between cavity and plate structure enables high underwater sound absorption (α > 0.8) over a broadband frequency range (470 Hz–10 kHz) using a 50-mm thick metasurface. Zhang et al.19 presented a meta-absorber with hexagonal unit cells consisting of steel plate scatterers of various diameters attached to a center support and embedded into an elastomer host material. Results show that the various resonances are combined to yield quasi-perfect absorption in a narrow frequency range (α > 0.94 from 500 to 700 Hz), with a high average sound absorption (α = 0.91) in the 1–10 kHz frequency range using a 70-mm thick sample. Other absorption-driven approaches include air or air/water-filled Helmholtz resonators, multilayered locally resonant coated plates20, viscoelastic polymer slabs embedded with locally resonant scatterers using hard or air-filled backings21, and hydrosound dampers (nets with air-filled elastic balloons)22. In particular, in offshore installations, such as pile driving for wind farms23,24, a collapsible framework with air-filled Helmholtz resonators was proposed by Wochner et al.25, exhibiting a maximum insertion loss (IL) of 37 dB around 100 Hz (measured using average one-third octave bands). However, drawbacks associated with these approaches include narrow-band performance and the elastic deformation of the resonators induced by hydrostatic pressure or by the oblique incidence of the acoustic wave, which can affect the resonant behavior of the finite system26. Alternatively, underwater noise barriers can be designed to exhibit a strong impedance mismatch with surrounding water, reflecting acoustic waves to prevent their transmission and presenting a more broadband effectiveness compared to locally resonant-based materials. Typical solutions include (i) arrays of one fluid immersed in another with contrasting properties, such as air bubbles in water, and (ii) mechanical casings that enclose the noise source to confine the energy of the sound waves between the noise source and the barrier. Arrays of bubbles in water can be produced using a perforated hosepipe resting on the seafloor (unconfined air bubble curtain) and encircling the source of noise. Compressed air is then pumped through this channel, and streams of bubbles are created, generating air curtains with a high impedance contrast. For instance, Lucke et al.27 reported a reduction in the order of 10–20 dB, from 500 to 2000 Hz, after the installation of an air bubble curtain to protect porpoises during the installation of a wooden pile. Würsig et al.28 reported ILs of 8–10 dB in the 400–800 Hz bands and 15–20 dB in the 1.6–6.4 kHz bands, indicating significantly poorer performance at lower frequencies. Although typically inexpensive, these systems are not effective in areas with high tidal currents, which can disrupt the air bubbles, causing them to be swept away from the noise source. Some of the other practical problems associated with these systems include the control of bubble size distribution and the generation of a sufficient volume of bubbles with large radii to achieve low-frequency noise attenuation29. For instance, to achieve significant attenuation below 600 Hz, air curtains would required bubbles radii ranging from 8 mm near the surface to 50 mm at a depth of 30 m30. This poses a significant limitation for their use due to the challenges associated with generating and maintaining air bubbles in a controlled manner in an offshore environment, particularly in the low-frequency range. In addition, a significant advantage of purely passive approaches with respect to air bubble curtains is that they do not require energy input, air supply, or active regulation once installed. In this case, passive solutions may offer advantages such as long-term reliability, maintenance, and environmental footprint in applications where continuous operation over extended periods is required. Architected acoustic panels represent another potential solution to reduce underwater noise, acting as mechanical casings placed around noise sources. However, water belongs to high-impedance media (Zw ≈ 1.5 × 106 N⋅s/m3), implying that barriers aimed at reflections due to a higher impedance with respect to water cannot efficiently isolate sound at low frequencies. To solve this issue, Jansen et al.31 proposed noise mitigation screens made of a double-walled steel cylinder filled with air submerged to 25 m of depth. The method combines (i) water or air/water between the pile and the casing and (ii) water or air inside the casing, producing a frequency-dependent acoustic IL with a slope of approximately 2 dB/octave, exceeding 20 dB above 4 kHz, but limited to 8–11 dB when averaged over a large frequency range (50 Hz to 40 kHz). The attenuation is less pronounced in the low-frequency range, where most of the energy is concentrated (100–250 Hz). Other technological solutions for underwater casing systems have also been reported, although with a lesser degree of maturity in terms of practical deployment32. Wang et al.33 combined topology optimization and homogenization methods to design lattice materials with low acoustic impedance compared to water. The results revealed that minimal acoustic impedance depends on both the degree of anisotropy and specific structural features that facilitate deformation. Two designs relying on a bi-mode lattice material have been proposed, with a thickness of 17 mm, yielding an average underwater sound transmission loss (STL) of 17 dB from 500–2000 Hz. Chen et al.34 explored the reflection of underwater sound waves on orthotropic solids whose acoustic impedance can be customized through mechanical anisotropy and velocity ratio of the quasi-transverse and quasi-longitudinal waves supported by these materials. Results have shown that a hexagonal lattice with a thickness of 21 mm can yield an average STL of 18.7 dB in the 1.5–3.5 kHz frequency range. Wang et al.35 proposed a metamaterial with a thickness of 54 mm, measuring an average STL of approximately 16 dB in the 400–1200 Hz frequency range. The mutual exclusion of low acoustic impedance and high mechanical properties was overcome by regulating the lattice orientation and incorporating a hierarchical morphology in an anisotropic metamaterial. Wang et al.36 proposed reverse engineering chiral metamaterials with optimized low acoustic impedance and appropriate stiffness using a topology optimization approach, demonstrating an average STL of 13.5 dB within the 1–5 kHz frequency range (for a sample with a thickness of 32 mm), as well as sufficient stiffness to ensure stable acoustic performance under moderate hydrostatic pressure. We summarize the reported literature review in Table 1, including also an additional line for some of the results that we present in the following sections.Table 1Summary of literature reviewReferenceDesign strategyAbsorption, STL, or ILThicknessFreq. range16q-Hrα > 0.950 mm306–921 Hz17q-Hrα > 0.550 mm279–478 Hz5FPrα > 0.98.9 mm (avg.)4–20 kHz18MCα > 0.850 mm0.47–10 kHz19MCα = 0.91 (avg.)70 mm1–10 kHz25HrIL = 37 dB (max.)N/A100 Hz27Rf (air bubbles)STL, 10–20 dBN/A500–2000 Hz28Rf (air bubbles)IL, 8–10 dBN/A400–800 Hz15–20 dBN/A1.6–6.4 kHz31Rf (double-wall casing)IL = 20 dB (max.)55 mm> 4 kHz33Rf (lattice material)IL = 17 dB (avg.)17 mm500–2000 Hz34Rf (lattice material)STL = 19 dB (avg.)21 mm1.5–3.5 kHz35Rf (lattice material)STL = 16 dB (avg.)54 mm400–1200 Hz36Rf (lattice material)STL = 13.5 dB (avg.)32 mm1–5 kHzThis workRf (lattice material)STL, 23.5–29.0 dB (max.)10–30 mm0.57–2.07 kHzq-Hr quasi-Helmholtz resonance, FPr Fabry–Pèrot resonance, MC mode coupling, Hr Helmholtz resonance, Rf reflection, avg. average, max. maximum, N/A not applicable. Table 1Summary of literature reviewReferenceDesign strategyAbsorption, STL, or ILThicknessFreq. range16q-Hrα > 0.950 mm306–921 Hz17q-Hrα > 0.550 mm279–478 Hz5FPrα > 0.98.9 mm (avg.)4–20 kHz18MCα > 0.850 mm0.47–10 kHz19MCα = 0.91 (avg.)70 mm1–10 kHz25HrIL = 37 dB (max.)N/A100 Hz27Rf (air bubbles)STL, 10–20 dBN/A500–2000 Hz28Rf (air bubbles)IL, 8–10 dBN/A400–800 Hz15–20 dBN/A1.6–6.4 kHz31Rf (double-wall casing)IL = 20 dB (max.)55 mm> 4 kHz33Rf (lattice material)IL = 17 dB (avg.)17 mm500–2000 Hz34Rf (lattice material)STL = 19 dB (avg.)21 mm1.5–3.5 kHz35Rf (lattice material)STL = 16 dB (avg.)54 mm400–1200 Hz36Rf (lattice material)STL = 13.5 dB (avg.)32 mm1–5 kHzThis workRf (lattice material)STL, 23.5–29.0 dB (max.)10–30 mm0.57–2.07 kHzq-Hr quasi-Helmholtz resonance, FPr Fabry–Pèrot resonance, MC mode coupling, Hr Helmholtz resonance, Rf reflection, avg. average, max. maximum, N/A not applicable. Summary of literature review q-Hr quasi-Helmholtz resonance, FPr Fabry–Pèrot resonance, MC mode coupling, Hr Helmholtz resonance, Rf reflection, avg. average, max. maximum, N/A not applicable. Despite the described approaches, achieving broadband and efficient underwater noise reduction at low frequencies (below 1 kHz) remains a challenge, especially when using highly sub-wavelength structures that must also withstand hydrostatic pressure. In this work, we address this issue by introducing a periodic metamaterial with a unit cell designed to enable a tailored interaction between normal and shear behavior along predefined directions. A similar idea has been introduced in ref. 34 using a single geometrical realization (honeycomb lattices), investigating the influence of selected dimensionless variables on the acoustic impedance of the resulting structure. Here, we propose an elasticity-based approach to ensure efficient underwater acoustic wave reflection in broad frequency ranges. We start by demonstrating the effect that a coupled normal–shear behavior—as quantified by a single dimensionless parameter—has on fluid-immersed homogeneous anisotropic metabarriers. Then, these results are utilized to formulate a topology optimization problem that maximizes such coupling and whose versatility allows a wide design space to be explored. Unlike previous approaches, we formulate our optimization problem in the static regime (zero frequency), relying exclusively on the homogenized material properties of the structured material, but not on the properties of the surrounding fluid. The obtained results are shown to be valid for broad frequency ranges for various thickness values, also withstanding increasing hydrostatic pressures. The paper is organized as follows. We begin by presenting the mathematical models and numerical methods used to investigate the STL curves of fluid-immersed anisotropic metabarriers, as well as the formulation of the topology optimization problem used. Then, we present the results of the optimization problem and its corresponding STL curves for an increasing number of unit cells, including suitable modifications to account for the potential exposure of the metabarrier to hydrostatic pressure. Finally, two-dimensional simulations to assess the acoustic attenuation effects for a point source are carried out. Conclusions and an outlook for future work are then discussed. Methods Normal-shear coupled behavior in continuous homogeneous anisotropic media Figure 1 a shows a noise source that generates underwater acoustic waves with wavelength λ (top panel) and our proposed metamaterial-based acoustic barrier (‘Metabarrier’) designed to reduce energy transmission at selected frequencies (bottom panel). The metabarrier consists of a finite periodic repetition of a rationally designed unit cell along the x direction (of length h ≪ λ) and is considered infinitely repeated along y. Energy absorption within the structure is assumed to be negligible. Anisotropy within the architected medium is introduced to achieve a significant STL. We consider a two-dimensional system governed by the constitutive relation σij = Cijklεkl, where σij is the Cauchy stress tensor, εkl is the strain tensor, and Cijkl is the fourth-order elasticity tensor37. In a fixed xy Cartesian frame, these relations can be expressed in matrix form as:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{\begin{array}{c}{\sigma }_{x}\\ {\sigma }_{y}\\ {\tau }_{xy}\end{array}\right\}=\left[\begin{array}{ccc}{C}_{11} & {C}_{12} & {C}_{16}\\ {C}_{21} & {C}_{22} & {C}_{26}\\ {C}_{61} & {C}_{62} & {C}_{66}\end{array}\right]\left\{\begin{array}{c}{\varepsilon }_{x}\\ {\varepsilon }_{y}\\ {\gamma }_{xy}\end{array}\right\},$$\end{document}σxσyτxy=C11C12C16C21C22C26C61C62C66εxεyγxy,where σx = σxx and σy = σyy are normal stresses, τxy is the shear stress, Cij = Cji, {i, j} = {1, 2, 6} represents the terms of the constitutive matrix C, εx = εxx and εy = εyy are normal strains, and γxy = 2ϵxy is the in-plane shear strain. For a coordinate system \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}^{{\prime} }{y}^{{\prime} }$$\end{document}x′y′ rotated with respect to xy about an angle β (Fig. 1b, top left panel), the corresponding constitutive matrix in the new coordinate system is obtained by an adequate coordinate transformation matrix T(β) of the form \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{C}}}^{{\prime} }(\beta )={{\bf{T}}}^{T}(\beta )\,{\bf{C}}\,{\bf{T}}(\beta )$$\end{document}C′(β)=TT(β)CT(β) (see Supplementary Information for details). As a consequence, each component \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{ij}^{{\prime} }(\beta )$$\end{document}Cij′(β) of the direction-dependent constitutive matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{C}}}^{{\prime} }(\beta )$$\end{document}C′(β) can be analyzed separately.Fig. 1Statement of the problem, schematic representation of the underwater metabarrier, and normal-shear coupling in anisotropic media concept.a Schematic of an underwater noise source, exemplifying a wind turbine pile, generating acoustic waves with wavelength λ (top panel). The wave fronts are illustrated by blue lines propagating in the fluid domain. The bottom panel shows the effect of enclosing the noise source within a thin metamaterial-based acoustic barrier (shown in green) of thickness h much smaller than the wavelength (h ≪ λ): a portion of the incident acoustic waves interacting with the metabarrier is reflected back, reducing transmitted noise on the opposite side. b Representative unit cell geometries (top row), corresponding anisotropic material symmetries reflected in their constitutive behavior (middle row), and polar plot of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }^{{\prime} }(\beta )$$\end{document}δ′(β) (bottom row). The left column depicts an isotropic material with direction-independent properties, showing a circular (constant) polar plot of the normalized coupling parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }^{{\prime} }(\beta )$$\end{document}δ′(β). The center column illustrates an anisotropic material with four-fold rotational symmetry (π/2-rotational invariance), yielding a polar plot with four maxima. The right column shows an anisotropic material with rotational symmetry (π-rotational invariance), resulting in a polar plot with two maxima. In the bottom panels, dashed green and red lines mark the π/2- and π-rotational symmetry axes, respectively. The green and red points identify the angular positions of maximum values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }^{{\prime} }$$\end{document}δ′, while the blue points indicate the reference value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }^{{\prime} }(\beta =0)$$\end{document}δ′(β=0). Statement of the problem, schematic representation of the underwater metabarrier, and normal-shear coupling in anisotropic media concept. a Schematic of an underwater noise source, exemplifying a wind turbine pile, generating acoustic waves with wavelength λ (top panel). The wave fronts are illustrated by blue lines propagating in the fluid domain. The bottom panel shows the effect of enclosing the noise source within a thin metamaterial-based acoustic barrier (shown in green) of thickness h much smaller than the wavelength (h ≪ λ): a portion of the incident acoustic waves interacting with the metabarrier is reflected back, reducing transmitted noise on the opposite side. b Representative unit cell geometries (top row), corresponding anisotropic material symmetries reflected in their constitutive behavior (middle row), and polar plot of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }^{{\prime} }(\beta )$$\end{document}δ′(β) (bottom row). The left column depicts an isotropic material with direction-independent properties, showing a circular (constant) polar plot of the normalized coupling parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }^{{\prime} }(\beta )$$\end{document}δ′(β). The center column illustrates an anisotropic material with four-fold rotational symmetry (π/2-rotational invariance), yielding a polar plot with four maxima. The right column shows an anisotropic material with rotational symmetry (π-rotational invariance), resulting in a polar plot with two maxima. In the bottom panels, dashed green and red lines mark the π/2- and π-rotational symmetry axes, respectively. The green and red points identify the angular positions of maximum values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }^{{\prime} }$$\end{document}δ′, while the blue points indicate the reference value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }^{{\prime} }(\beta =0)$$\end{document}δ′(β=0). The degree of coupling between normal stresses and shear strains (and vice versa) can be quantified by defining a mode coupling factor δ as the ratio between the terms C16 = C61 and the product C11C66 of the constitutive matrix, as follows:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta =\frac{{C}_{16}}{\sqrt{{C}_{11}{C}_{66}}},$$\end{document}δ=C16C11C66,yielding δ ∈ (− 1, 1). In the case of an isotropic medium, C16 = C61 = 0 and one obtains δ = 0. This parameter is typically used in the context of longitudinal-shear mode conversion38, since it quantifies the normal-shear coupling for a specific state of stress. This coupling is typical of highly anisotropic unit cells with specific off-axis features, thus not representing fundamental material constants, but orientation-dependent coupling stiffness values. It is also associated with the degree of structural instability39, which we further clarify in the next subsection. For the case of direction-dependent homogeneous properties, this relation also depends on the orientation angle β, following \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }^{{\prime} }(\beta )={C}_{16}^{{\prime} }(\beta )/\sqrt{{C}_{11}^{{\prime} }(\beta ){C}_{66}^{{\prime} }(\beta )}$$\end{document}δ′(β)=C16′(β)/C11′(β)C66′(β) (i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta ={\delta }^{{\prime} }(\beta =0)$$\end{document}δ=δ′(β=0)). Usual homogeneous structures employed in typical barriers present isotropic constitutive behavior32, resulting in decoupled normal stresses and shear strains (and vice versa) for every direction. An example of a continuous homogeneous structure is shown in Fig. 1b (top left panel). In this case, for the direction β = 0, the terms C16 = C61 = C26 = C62 = 0, while other terms are non-zero, as represented schematically in Fig. 1b (middle left panel). We highlight, in particular, the terms associated with the propagation of normal and shear waves using yellow squares, while other non-zero terms are indicated as black squares. In this case, the elasticity tensor components are invariable with respect to the orientation angle β and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{16}^{{\prime} }(\beta )={C}_{61}^{{\prime} }(\beta )=0$$\end{document}C16′(β)=C61′(β)=0, indicating that normal-shear coupling is zero for any angle. This behavior results in the direction independence of the mode coupling factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }^{{\prime} }(\beta )$$\end{document}δ′(β) (bottom left panel). In the case of structures with an architected unit cell, however, an anisotropic behavior may be observed. Consider, for instance, the unit cell with four-fold symmetry shown in Fig. 1b (top center panel). In this case, the corresponding constitutive matrix may present non-zero components for the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{16}^{{\prime} }(\beta )={C}_{61}^{{\prime} }(\beta )$$\end{document}C16′(β)=C61′(β) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{26}^{{\prime} }(\beta )={C}_{62}^{{\prime} }(\beta )$$\end{document}C26′(β)=C62′(β) terms for some preferential directions β ≠ 0 (see middle center panel, where possible non-zero terms are represented as squares). In particular, terms \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{16}^{{\prime} }={C}_{61}^{{\prime} }$$\end{document}C16′=C61′, which are associated with the coupling of normal and shear waves in the x-direction, are highlighted in blue. As a result, the mode coupling factor presents four-fold symmetry, displaying preferential angles (π/8 + nπ/2, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n\in {\mathcal{N}}$$\end{document}n∈N, illustrated by the green dashed lines, bottom center panel) for the coupling between normal and shear stresses (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{16}^{{\prime} }(\beta )\ne 0$$\end{document}C16′(β)≠0). This implies, however, that no coupling is present at an arbitrary angle (e.g., for β = 0). Finally, by removing the condition of four-fold symmetry, it is possible to design a unit cell that presents a non-zero value of normal-shear coupling at an arbitrary direction (e.g., for β = 0). An example of a unit cell with such a condition is shown in Fig. 1b (top right panel). Through a systematic approach, the unit cell can be designed to achieve C16 = C61 ≠ 0 at arbitrary wave propagation angles, yielding non-zero values even for β = 0 (highlighted by the red squares in the middle right panel of Fig. 1b). The corresponding behavior shows π-rotational symmetry for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\delta }^{{\prime} }(\beta )$$\end{document}δ′(β) (dashed red and yellow lines, bottom right panel of Fig. 1b), and as a consequence, non-zero normal-shear coupling for the angle β = 0. This allows for the excitation of modes with a shear component due to the incidence of longitudinal (acoustic) modes from the fluid, thus hindering the efficiency of longitudinal wave mode propagation in the solid medium and reducing the transmission of energy between incident and transmitted acoustic waves. Let us now verify the effect of a coupled normal-shear behavior on the STL curves for homogeneous anisotropic structures. Computation of STL for a homogeneous anisotropic medium To investigate the effects of coupled normal-shear behavior, we derive a general expression that relates acoustic pressures on both edges of an acoustic barrier with homogeneous anisotropic material properties and a thickness h in the x-direction immersed in a fluid (see Fig. 2a), allowing to obtain the corresponding STL curves.Fig. 2Investigation on the STL properties of a fluid-immersed homogeneous anisotropic metabarrier under normal incidence of acoustic waves.a Acoustic metabarrier with thickness h subjected to an incident wave (Pi), emitting reflected (Pr) and transmitted (Pt) acoustic waves. The internal propagation of elastic waves is described by two pairs of wave modes with positive- (U1, U2) and negative-going (U3, U4) components with distinct complex amplitudes. b Computed STL in the normalized frequency range \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega \overline{h}/{c}_{0}\in [0,1]$$\end{document}ωh¯/c0∈[0,1] for the thickness \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h=\overline{h}=1{0}^{-2}$$\end{document}h=h¯=10−2 m and coupling factor δ = {0.9, 0.95, 0.995} (upper panel). The specific acoustic impedance of the metabarrier normalized with respect to the fluid impedance, ∣z∣/ρ0c0, is shown in the same frequency range (lower panel). The frequencies of STL maxima correspond to minimum impedance values (e.g., see δ = 0.995), indicating a large impedance mismatch. c Same as (b) for δ = 0.999 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h/\overline{h}=\{1{0}^{-2},1{0}^{-1},1{0}^{0}\}$$\end{document}h/h¯={10−2,10−1,100}. The first and second peaks (dips) are represented by orange and purple stars (diamonds), respectively, for the case \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h/\overline{h}=1{0}^{0}$$\end{document}h/h¯=100. d Amplitude of normalized longitudinal (vx) and transverse (vy) velocities for the peaks and dips indicated in (c). Investigation on the STL properties of a fluid-immersed homogeneous anisotropic metabarrier under normal incidence of acoustic waves. a Acoustic metabarrier with thickness h subjected to an incident wave (Pi), emitting reflected (Pr) and transmitted (Pt) acoustic waves. The internal propagation of elastic waves is described by two pairs of wave modes with positive- (U1, U2) and negative-going (U3, U4) components with distinct complex amplitudes. b Computed STL in the normalized frequency range \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega \overline{h}/{c}_{0}\in [0,1]$$\end{document}ωh¯/c0∈[0,1] for the thickness \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h=\overline{h}=1{0}^{-2}$$\end{document}h=h¯=10−2 m and coupling factor δ = {0.9, 0.95, 0.995} (upper panel). The specific acoustic impedance of the metabarrier normalized with respect to the fluid impedance, ∣z∣/ρ0c0, is shown in the same frequency range (lower panel). The frequencies of STL maxima correspond to minimum impedance values (e.g., see δ = 0.995), indicating a large impedance mismatch. c Same as (b) for δ = 0.999 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h/\overline{h}=\{1{0}^{-2},1{0}^{-1},1{0}^{0}\}$$\end{document}h/h¯={10−2,10−1,100}. The first and second peaks (dips) are represented by orange and purple stars (diamonds), respectively, for the case \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h/\overline{h}=1{0}^{0}$$\end{document}h/h¯=100. d Amplitude of normalized longitudinal (vx) and transverse (vy) velocities for the peaks and dips indicated in (c). The equations of motion for a two-dimensional medium free of external forces can be written in tensor notation as403\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }_{ji,j}=\rho {\ddot{u}}_{i},$$\end{document}σji,j=ρu¨i,where σij is the stress tensor, ui is the displacement in the i-th direction, the partial derivatives with respect to the spatial coordinate j and time coordinate t are denoted, respectively, as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${(\circ )}_{,j}=\frac{\partial (\circ )}{\partial j}$$\end{document}(∘),j=∂(∘)∂j and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\ddot{\circ })=\frac{{\partial }^{2}(\circ )}{\partial {t}^{2}}$$\end{document}(∘¨)=∂2(∘)∂t2, and ρ is the material specific mass density. Equations (1) and (3) can be combined and rewritten by considering an infinitesimal strain tensor (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varepsilon }_{kl}={\varepsilon }_{lk}=\frac{1}{2}({u}_{k,l}+{u}_{l,k})$$\end{document}εkl=εlk=12(uk,l+ul,k)), yielding4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{C}}}_{x}{{\bf{u}}}_{,xx}+{{\bf{C}}}_{xy}{{\bf{u}}}_{,xy}+{{\bf{C}}}_{y}{{\bf{u}}}_{,yy}=\rho \ddot{{\bf{u}}},$$\end{document}Cxu,xx+Cxyu,xy+Cyu,yy=ρu¨,where the matrix and vector components are described as5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{C}}}_{x}=\left[\begin{array}{cc}{C}_{11} & {C}_{16}\\ {C}_{61} & {C}_{66}\end{array}\right],\,{{\bf{C}}}_{xy}=\left[\begin{array}{cc}{C}_{16}+{C}_{61} & {C}_{12}+{C}_{66}\\ {C}_{21}+{C}_{66} & {C}_{26}+{C}_{62}\end{array}\right],\,{{\bf{C}}}_{y}=\left[\begin{array}{cc}{C}_{66} & {C}_{62}\\ {C}_{26} & {C}_{22}\end{array}\right],\,{\bf{u}}=\left\{\begin{array}{c}{u}_{x}\\ {u}_{y}\end{array}\right\}.$$\end{document}Cx=C11C16C61C66,Cxy=C16+C61C12+C66C21+C66C26+C62,Cy=C66C62C26C22,u=uxuy. Consider now a Bloch solution for the displacements in the form416\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bf{u}}({\bf{r}},t)={\bf{U}}{e}^{-{\rm{i}}\omega t}{e}^{{\rm{i}}{\bf{k}}\cdot {\bf{r}}},$$\end{document}u(r,t)=Ue−iωteik⋅r,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bf{u}}={\{{u}_{x},{u}_{y}\}}^{T}$$\end{document}u={ux,uy}T is the displacement vector in Cartesian coordinates, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bf{U}}={\{{U}_{x},{U}_{y}\}}^{T}$$\end{document}U={Ux,Uy}T is the wave mode, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{i}}=\sqrt{-1}$$\end{document}i=−1 is the imaginary unit number, ω is the circular frequency, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bf{k}}={\{{k}_{x},{k}_{y}\}}^{T}$$\end{document}k={kx,ky}T is the two-dimensional wave vector, and r = {x, y}T is the two-dimensional coordinate vector. Substituting Eqs. (6) in (4) yields7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({k}_{x}^{2}{{\bf{C}}}_{x}+{k}_{x}{k}_{y}{{\bf{C}}}_{xy}+{k}_{y}^{2}{{\bf{C}}}_{y}-{\omega }^{2}\rho {\bf{I}}){\bf{U}}={\bf{0}},$$\end{document}(kx2Cx+kxkyCxy+ky2Cy−ω2ρI)U=0,where I a order-2 identity matrix. Rewriting the wavenumber components \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{x}=k\cos \theta$$\end{document}kx=kcosθ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{y}=k\sin \theta$$\end{document}ky=ksinθ for a given wavenumber k and propagation direction θ leads to the eigenproblem8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({k}^{2}{\bf{D}}-{\omega }^{2}\rho {\bf{I}}){\bf{U}}={\bf{0}},$$\end{document}(k2D−ω2ρI)U=0,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bf{D}}={\bf{D}}({\bf{C}},\theta )={{\bf{C}}}_{x}{\cos }^{2}\theta +{{\bf{C}}}_{xy}\sin \theta \cos \theta +{{\bf{C}}}_{y}{\sin }^{2}\theta$$\end{document}D=D(C,θ)=Cxcos2θ+Cxysinθcosθ+Cysin2θ is a direction-dependent stiffness matrix. Particularly, for the case θ = 0, D = Cx, yielding 4 solutions that correspond to two pairs of different types of waves propagating in the + x and − x directions. For the simplified case of a four-fold symmetric unit cell (i.e., with respect to both x and y axes), one has C16 = C61 = 0; considering waves propagating in the x-direction (θ = 0), the well-known solutions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{l}(\omega )=\omega \sqrt{\rho /{C}_{11}}$$\end{document}kl(ω)=ωρ/C11 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${k}_{s}(\omega )=\omega \sqrt{\rho /{C}_{66}}$$\end{document}ks(ω)=ωρ/C66 for longitudinal and shear waves, respectively, are retrieved. The corresponding wave modes, described as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{U}}}_{l}={\left\{1,0\right\}}^{T}$$\end{document}Ul=1,0T and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{U}}}_{s}={\left\{0,1\right\}}^{T}$$\end{document}Us=0,1T, are thus fully decoupled. This simplification is no longer valid for θ = 0 whenever C16 ≠ 0. In the case of structured media (as in the case of metamaterials), a constitutive matrix equivalent to a homogenized medium can be obtained using numerical procedures42, which are valid in the long wavelength limit (k → 0). Here we make an additional comment regarding structural instability as quantified by the coupling factor δ (Eq. 2). For waves propagating in the x-direction, D = Cx. The determinant of this matrix is given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\det ({{\bf{C}}}_{x})={C}_{11}{C}_{66}-{C}_{16}^{2}$$\end{document}det(Cx)=C11C66−C162. For a value of δ = 1, this yields \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\det ({{\bf{C}}}_{x})=0$$\end{document}det(Cx)=0, representing an upper limit for the strength of normal-shear coupling before the material itself becomes non-physical. This is caused by the loss of positive definiteness of Cx, thereby producing a non-zero strain mode with zero elastic energy (i.e., infinitesimal perturbations can cause large deformations without resistance). Let now the acoustic pressure waves (see Fig. 2a) be denoted as9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{rcl}{P}_{i}(x,t) & = & {e}^{-{\rm{i}}\omega t}{\hat{P}}_{i}{e}^{+{\rm{i}}{k}_{0}x},\\ {P}_{r}(x,t) & = & {e}^{-{\rm{i}}\omega t}{\hat{P}}_{r}{e}^{-{\rm{i}}{k}_{0}x},\\ {P}_{t}(x,t) & = & {e}^{-{\rm{i}}\omega t}{\hat{P}}_{t}{e}^{+{\rm{i}}{k}_{0}(x-h)},\end{array}$$\end{document}Pi(x,t)=e−iωtP^ie+ik0x,Pr(x,t)=e−iωtP^re−ik0x,Pt(x,t)=e−iωtP^te+ik0(x−h),where Pi is the incident wave propagating in the +x direction, Pr and Pt are, respectively, reflected and transmitted waves, k0 = ω/c0 is the acoustic wavenumber, and c0 is the speed of sound in the fluid. We conveniently choose x = 0 to be set at the fluid-structure interface between the metabarrier and incident acoustic waves. The shift (x − h) in the exponential term of the transmitted wave is adopted for analytical convenience. Considering the possible solutions for Eq. (8) in the θ = 0 direction, the displacements in the material slab can be written as10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bf{u}}(x,t)={e}^{-{\rm{i}}\omega t}\mathop{\sum }\limits_{n=1}^{4}{\hat{u}}_{n}{{\bf{U}}}_{n}{e}^{{\rm{i}}{k}_{n}x},$$\end{document}u(x,t)=e−iωt∑n=14u^nUneiknx,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{u}}_{n}$$\end{document}u^n are complex wave amplitudes, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{U}}}_{n}={\{{U}_{xn},{U}_{yn}\}}^{T}$$\end{document}Un={Uxn,Uyn}T are wave modes, and kn are x-direction wavenumbers corresponding to the 4 solutions obtained in Eq. (8). The complex amplitudes in Eq. (10) can be calculated by setting appropriate interface conditions corresponding to the immersion of the structure in the fluid. The continuity of accelerations at the left (x = 0) and right (x = h) boundaries of the slab are expressed as11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left.{({P}_{i}+{P}_{r})}_{,x}\right|}_{x=0}={\left.{\rho }_{0}{\omega }^{2}{u}_{x}\right|}_{x=0},{\left.{({P}_{t})}_{,x}\right|}_{x=h}={\left.{\rho }_{0}{\omega }^{2}{u}_{x}\right|}_{x=h},$$\end{document}(Pi+Pr),xx=0=ρ0ω2uxx=0,(Pt),xx=h=ρ0ω2uxx=h,which can be combined with Eq. (10) to yield12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widehat{P}}_{r}={\widehat{P}}_{i}+\frac{{\rm{i}}{\rho }_{0}{\omega }^{2}}{{k}_{0}}\mathop{\sum }\limits_{n=1}^{4}{\widehat{u}}_{n}\,{U}_{xn},{\widehat{P}}_{t}=-\frac{{\rm{i}}{\rho }_{0}{\omega }^{2}}{{k}_{0}}\mathop{\sum }\limits_{n=1}^{4}{\widehat{u}}_{n}\,{U}_{xn}{e}^{{\rm{i}}{k}_{n}h}.$$\end{document}P^r=P^i+iρ0ω2k0∑n=14u^nUxn,P^t=−iρ0ω2k0∑n=14u^nUxneiknh. Additionally, at the fluid-structure interfaces, shear stresses vanish and the continuity of normal stresses is ensured43, such that13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-{\left.({P}_{i}+{P}_{r})\right|}_{x=0}={\sigma }_{x}{| }_{x=0},\,{\left.-{P}_{t}\right|}_{x=h}={\sigma }_{x}{| }_{x=h},\,{\left.{\tau }_{xy}\right|}_{x=0}=0,\,{\tau }_{xy}{| }_{x=h}=0.$$\end{document}−(Pi+Pr)x=0=σx∣x=0,−Ptx=h=σx∣x=h,τxyx=0=0,τxy∣x=h=0. Normal and shear stresses can be expressed using Eq. (1) for θ = 0, respectively, as14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{rcl}{\sigma }_{x} & = & {C}_{11}{u}_{x,x}+{C}_{16}{u}_{y,x},\\ {\tau }_{xy} & = & {C}_{61}{u}_{x,x}+{C}_{66}{u}_{y,x},\end{array}$$\end{document}σx=C11ux,x+C16uy,x,τxy=C61ux,x+C66uy,x,which can be combined with Eqs. (10) and (13) to yield the linear system expressed as15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{rcl}\mathop{\sum }\limits_{n=1}^{4}\left[\left({C}_{11}{k}_{n}+\frac{{\rho }_{0}{\omega }^{2}}{{k}_{0}}\right){U}_{xn}+\left({C}_{16}{k}_{n}\right){U}_{yn}\right]{\hat{u}}_{n} & = & 2{\rm{i}}{\hat{P}}_{i}\\ \mathop{\sum }\limits_{n=1}^{4}{e}^{{\rm{i}}{k}_{n}h}\left[\left({C}_{11}{k}_{n}-\frac{{\rho }_{0}{\omega }^{2}}{{k}_{0}}\right){U}_{xn}+\left({C}_{16}{k}_{n}\right){U}_{yn}\right]{\hat{u}}_{n} & = & 0,\\ \mathop{\sum }\limits_{n=1}^{4}\left[\left({C}_{61}{k}_{n}\right){U}_{xn}+\left({C}_{66}{k}_{n}\right){U}_{yn}\right]{\hat{u}}_{n} & = & 0,\\ \mathop{\sum }\limits_{n=1}^{4}{e}^{{\rm{i}}{k}_{n}h}\left[\left({C}_{61}{k}_{n}\right){U}_{xn}+\left({C}_{66}{k}_{n}\right){U}_{yn}\right]{\hat{u}}_{n} & = & 0,\end{array}$$\end{document}∑n=14C11kn+ρ0ω2k0Uxn+C16knUynu^n=2iP^i∑n=14eiknhC11kn−ρ0ω2k0Uxn+C16knUynu^n=0,∑n=14C61knUxn+C66knUynu^n=0,∑n=14eiknhC61knUxn+C66knUynu^n=0,from which each complex constant \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{u}}_{n}$$\end{document}u^n can be computed for a given frequency ω, thus allowing to obtain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{P}}_{t}$$\end{document}P^t using Eq. (12). Although the analytical expression relating \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{P}}_{t}$$\end{document}P^t and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{P}}_{i}$$\end{document}P^i is too long to be conveniently written, we note that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{\hat{P}}_{t}}{{\hat{P}}_{i}}\propto {C}_{11}{C}_{66}-{C}_{16}{C}_{61}$$\end{document}P^tP^i∝C11C66−C16C61, indicating that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{P}}_{t}\to 0$$\end{document}P^t→0 for C16C61 → C11C66, i.e., noise transmission decreases as δ → 1. Finally, the STL is obtained as4416\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{STL}(\omega )=10\log {\left(\frac{{\hat{P}}_{i}}{{\hat{P}}_{t}}\right)}^{2}.$$\end{document}STL(ω)=10logP^iP^t2. We include here some remarks concerning the relationship between the reflection coefficient, R = Pr/Pi, and the specific acoustic impedance of the solid structure, denoted by z. These quantities are related at the fluid-structure interface located at x = x0 through the well-known expression4417\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R(x={x}_{0})={\left.\frac{z-{\rho }_{0}{c}_{0}}{z+{\rho }_{0}{c}_{0}}\right|}_{x={x}_{0}},$$\end{document}R(x=x0)=z−ρ0c0z+ρ0c0x=x0,where ρ0c0 denotes the specific acoustic impedance of the surrounding fluid. This expression indicates that ideal cases of perfect reflection (i.e., zero transmission) can theoretically be achieved in two limiting cases: (i) R = + 1 for z → ∞, or (ii) R = − 1 for z → 0. The positive and negative signs correspond, respectively, to phase differences of 0 and π between incident and reflected acoustic waves at the interface. In practice, however, the specific impedance z is finite, which limits the reflection coefficient to ∣R∣ < 1. Achieving extremely large values of z in a water medium is particularly challenging, as it would require unrealistically thick materials due to the relatively high acoustic impedance of water. Consequently, a more practical strategy consists in designing an acoustically “soft” material such that z → 0, thereby creating a strong impedance mismatch through the reduction of the acoustic impedance relative to that of water. This approach is explored in the following section. Effect of normal-shear coupling on STL curves We now present an illustrative example, computing the STL curves of a homogeneous two-dimensional structure whose constitutive matrix C presents components with values C11 = 1 GPa, C66/C11 = 0.9, and specific mass density ρ = 1000 kg/m3. The ratio between the acoustic wave speed in the fluid and the solid is assumed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${c}_{0}/\sqrt{{C}_{11}/\rho }=0.9$$\end{document}c0/C11/ρ=0.9 and the ratio between their specific mass densities as ρ0/ρ = 0.9, thus similar to the ratio between the material properties of water and typical polymers. We compute and investigate the STL curves for a barrier with a finite thickness h over the normalized frequency range \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\omega }^{* }=\omega \overline{h}/{c}_{0}\in [0,1]$$\end{document}ω*=ωh¯/c0∈[0,1], for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{h}=1{0}^{-2}$$\end{document}h¯=10−2 m. The STL curves computed for h = 10−2 m and δ = {0.9, 0.95, 0.995} are shown in Fig. 2b (top panel). For the value δ = 0.9, the presence of a local maximum of 2.6 is noticed at ω* = 0.92. For δ = 0.95, we notice the presence of a maximum STL of 4.7 located at ω* = 0.61, which suggests that increasing the value of δ presents the effects of (i) increasing the first peak STL value and (ii) red-shifting its frequency. Finally, for δ = 0.995, a total of 4 STL peaks is noticed, also with an increase in the number of dips (total of 4) in the same frequency range. Noticeably, the first peak is now located at ω* = 0.19 with an STL value of 13.8. Thus, increasing δ from 0.9 to 0.995 leads to a (i) five-fold increase in the first STL peak and (ii) 80% reduction in its corresponding normalized frequency. As a consequence, an infinite STL value would be achieved close to the zero frequency, for δ → 1−. We also report the value of the specific acoustic impedance of the metabarrier at the incident boundary (x = 0) using the relation z = ρ0c0(1 + R)/(1 − R), where R is computed from Eq. (12). The results are shown in Fig. 2b (bottom panel). We observe that the peaks in the STL curves correspond to minima of ∣z∣, with the peak values increasing as the acoustic impedance decreases. This correlation indicates that the STL enhancement arises from the large impedance mismatch between the metabarrier and the surrounding fluid, reaching values close to 1% of the acoustic impedance of water for δ = 0.995. We also investigate the variation of the STL curves computed for the fixed value of δ = 0.999 and increasing values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h/\overline{h}\in [1{0}^{-2},1{0}^{0}]$$\end{document}h/h¯∈[10−2,100], as shown in Fig. 2c (top panel). For \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h/\overline{h}=1{0}^{-2}$$\end{document}h/h¯=10−2, a maximum STL value of 6.3 is computed at ω* = 1, which suggests that ultra-thin structures may present significant STL values, with the first peak located at higher frequencies. It is important to notice that remaining in the sub-wavelength regime (thickness much smaller than typical wavelength) is of interest, which then requires small values of h. On the other hand, sufficiently large values of h are required to ensure that the homogenization hypotheses are valid, which may pose conflicting design objectives (sub-wavelength regime vs. homogenized material properties). For \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h/\overline{h}=1{0}^{-1}$$\end{document}h/h¯=10−1, a maximum STL value of 20.7 is achieved at ω* = 0.83, indicating the red-shift of STL peaks for an increase in the thickness. Finally, for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h/\overline{h}=1{0}^{0}$$\end{document}h/h¯=100, a total of 10 STL peaks and dips are observed in this frequency range, with a first STL peak of 20.7 dB at ω* = 0.083. Interestingly, this represents a maintenance in the STL level of the first peak with respect to the case \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h/\overline{h}=1{0}^{-1}$$\end{document}h/h¯=10−1, however with a ten-fold reduction in its frequency for a ten-fold thickness increase, thus suggesting that the frequency of the first peak of the STL curve can be tuned using the thickness of the metabarrier. Finally, we indicate the first two peaks (dips) in the STL curve for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h/\overline{h}=1{0}^{0}$$\end{document}h/h¯=100 using orange and purple stars (diamonds), respectively. For completeness, we also report the corresponding specific acoustic impedance of the metabarrier in Fig. 2c (bottom panel). A pronounced dip is observed at the first STL peak for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h/\overline{h}=1{0}^{0}$$\end{document}h/h¯=100, where the specific acoustic impedance of the metabarrier is reduced to approximately 0.2% of that of water. The profiles of the absolute normalized velocities (and equivalently, displacements and accelerations) indicated in Fig. 2c are shown in Fig. 2d. The modes corresponding to the selected peaks (orange and purple stars) exhibit a significant decrease in the longitudinal velocities (vx, black lines) between the left (x = 0) and right (x = h) edges of the metabarrier. This implies a decrease in the velocity (and acceleration) transmissibility between the input and transmitted edges, which justifies the sharp peaks in the STL curves. On the other hand, the modes corresponding to the dips in the STL curves (orange and purple diamonds), which correspond to Fabry–Pérot resonances45, present symmetric profiles of longitudinal and transverse absolute velocities, resulting in unit transmissibility between the left and right edges of the metabarrier. The n-th dip presents a phase variation of nπ, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n\in {\mathcal{N}}$$\end{document}n∈N, resulting in complete in-phase or out-of-phase behavior between both edges of the metabarrier. As a consequence, the transmitted and incident acoustic waves present the same amplitude, differing at most from a phase shift. Anisotropic heterogeneous material design: optimization problem of the unit cell The previous discussion suggests that a suitable optimization objective to obtain the periodic unit cell that forms a metabarrier yielding maximum STL at a minimal operating frequency is the maximization of the mode coupling factor δ. The desired value of δ → 1− would yield an infinite STL at zero frequency. However, this value cannot be achieved, as it represents a structural singularity. Thus, in order to obtain the periodic metamaterial unit cell, we propose a topology optimization problem which aims to minimize the objective function φ, given by18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{c}\mathop{\mathrm{minimize}}\limits_{{\bf{q}}\in \{0,1\}^{n_q}} \varphi ({\bf{q}})={\left(|{C}_{16}|/\sqrt{{C}_{11}{C}_{66}}-1\right)}^{2},\\ \,{\mathrm{subjected}}\; {\mathrm{to}} {\omega }_{1} > {\omega }_{\min},\end{array}$$\end{document}minimizeq∈{0,1}nqφ(q)=∣C16∣/C11C66−12,subjectedtoω1>ωmin,where q is a binary vector composed of nq design variables representing the presence of material or voids in the unit cell and ω1 is the first resonant frequency of the system for free boundary conditions excluding rigid body motion, which must be above a certain threshold (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\omega }_{\min }$$\end{document}ωmin). The proposed objective function aims to minimize the distance between ∣δ∣ and 1, while the first resonance frequency restriction is imposed to ensure connectivity between the pixels of the candidate unit cell, thus discarding solutions that present disconnected (i.e., “floating”) regions of solid material. We note that the objective function is formulated to consider only the effective static properties of the unit cell. On the other hand, polymers typically exhibit viscoelastic behavior, which becomes more pronounced at higher frequencies46. We recall, however, that the complex constitutive relations describing viscoelastic behavior are typically frequency-dependent47. Thus, since the objective of this work is to focus on the investigation of the effect of normal-shear coupling (i.e., a relation typical of the static regime), we do not consider the possible effects of viscoelasticity in our optimization function. Each square unit cell used in the topology optimization process is described by a distribution of 2q × 2q pixels. The continuity between adjacent unit cells is ensured by the correspondence of pixels at opposing edges. Additionally, a π-rotational symmetry is enforced in the internal region of the unit cell to ensure normal-shear coupling. This scheme is represented in Fig. 3a (top panel), where the continuity between the pixels of contiguous unit cells and the internal π-rotational symmetry are represented, respectively, as “Matching boundaries” (green contour) and “Anti-symmetry” (region inside the dashed red contour). This yields a total of nq = 2q2 − 2q − 1 pixels, which can then be encoded into a corresponding binary vector q (Fig. 3a, middle panel). The represented unit cell can then be discretized using regular square finite elements (FEs) under the assumption of plane strain48, allowing for the computation of effective constitutive matrix of the homogenized unit using the method described in ref. 42. The dynamic behavior of the homogenized unit cell is valid in the long wavelength limit (k = 2π/λ → 0, i.e., its validity decreases for increasing frequency values).Fig. 3Optimization of the metabarrier unit cell.a Periodic unit cell (top panel) with a discretized geometry indicating 2q × 2q pixels with either material (1) or void (0), π-rotational symmetry in the internal region ("Anti-symmetry''), and continuous boundaries ("Matching boundaries''). This description is encoded into a binary vector q (middle panel), which is utilized in a genetic algorithm optimization (bottom panel). b Result of the optimization procedure (top panel), obtained for q = 5, yielding a mode coupling ratio δ = 0.9879. After obtaining a smooth approximate representation (bottom panel), this value is decreased to δ = 0.9852. c Dispersion diagram (continuous lines) for the obtained unit cell (length a = 10 mm), with polarization values p indicating either completely longitudinal (p = 0) or transverse motion (p = 1). The wave modes are computed and shown at the X-point (k = π/a) of the first branch (2.67 kHz) and at 5 kHz on the second branch (k ≈ 0.14π/a). The dashed lines indicate the dispersion curves corresponding to the medium with effective homogenized properties for the first (k1) and second branches (k2). Optimization of the metabarrier unit cell. a Periodic unit cell (top panel) with a discretized geometry indicating 2q × 2q pixels with either material (1) or void (0), π-rotational symmetry in the internal region ("Anti-symmetry''), and continuous boundaries ("Matching boundaries''). This description is encoded into a binary vector q (middle panel), which is utilized in a genetic algorithm optimization (bottom panel). b Result of the optimization procedure (top panel), obtained for q = 5, yielding a mode coupling ratio δ = 0.9879. After obtaining a smooth approximate representation (bottom panel), this value is decreased to δ = 0.9852. c Dispersion diagram (continuous lines) for the obtained unit cell (length a = 10 mm), with polarization values p indicating either completely longitudinal (p = 0) or transverse motion (p = 1). The wave modes are computed and shown at the X-point (k = π/a) of the first branch (2.67 kHz) and at 5 kHz on the second branch (k ≈ 0.14π/a). The dashed lines indicate the dispersion curves corresponding to the medium with effective homogenized properties for the first (k1) and second branches (k2). We also note that, as the homogenization procedure only concerns the static properties of the unit cell, evaluating binary vectors in the optimization problem (see Eq. 18) presents a low computational cost, which allows for an efficient use of genetic algorithms49 to determine an optimal solution (Fig. 3a, bottom panel). More specifically, we use MATLAB’s Global Optimization toolbox (GA function), which enables the use of discrete variables. As such, no penalization scheme for continuous variables is considered (e.g., SIMP50). We also do not choose a set of solutions as the initial population for the optimization scheme, relying on the auto-initialization offered by the toolbox. We note that different runs of the genetic algorithm might yield different results, which is inherent to the stochastic behavior of the method in its search for a global optimum. Results Result of optimization problem We apply the optimization procedure considering a structure composed of a plastic material with Young’s modulus E = 3 GPa, Poisson’s ratio ν = 0.3, and specific mass density ρ = 1150 kg/m3. The unit cell has a length of a = 10 mm, which is amenable to fabrication using 3D-printing techniques. The resulting unit cell obtained using q = 5 (yielding a total of 2q × 2q = 100 pixels, i.e., nq = 39 optimization variables and 239 = 5.5 × 1011 possible designs) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\omega }_{\min }=0.1$$\end{document}ωmin=0.1 rad/s is presented in Fig. 3b (top panel), displaying the discretized geometry of the unit cell and the obtained δ value (δ = 0.9879). Similar solutions were obtained for different runs of the algorithm, so we chose the one with the best computed objective function. The obtained geometry then undergoes a smoothing procedure to yield a geometry amenable to fabrication (Fig. 3b, bottom panel). As a result of the geometrical modifications, a slightly smaller value of δ is obtained (δ = 0.9852). We then apply two-dimensional periodic Bloch-Floquet conditions to the resulting unit cell51 and compute its dispersion diagram in the ΓX direction, with Γ located at k = 0 and X located at k = π/a. For each wavenumber, we also compute a polarization metric given by p(ux, uy) = ∫S∣uy∣2 dS/∫S∣ux∣2 + ∣uy∣2 dS, where ux and uy are, respectively, the x- and y-direction displacement components of a given wave mode and S is the unit cell two-dimensional domain. As a result, the values p = 0 and p = 1 represent, respectively, purely longitudinal and transverse motion. The band diagram computed using the FE method is shown in Fig. 3c in the [0, 5] kHz frequency range using continuous lines, with polarization values indicated in a color scale ranging from p = 0 (dark blue) to p = 1 (dark green). The first obtained branch presents p ∈ [0.403, 0.414], demonstrating a predominant longitudinal behavior (p < 0.5), although with a high degree of hybridization between longitudinal and transverse motion. The wave mode corresponding to this branch at the point X is also shown with the colorbar representing displacement values (ux and uy), illustrating this coupled behavior at 2.67 kHz. Above this frequency (indicated with a horizontal red dashed line), the first branch is no longer present in the considered frequency range. Likewise, the second branch presents p ∈ [0.585, 0.589], exhibiting a predominant transverse behavior (p > 0.5) with a significant hybridization between longitudinal and transverse motion. A wave mode corresponding to this branch is shown at 5 kHz, displaying a rigid-body-like vertical motion due to the small wavenumber at this point (k ≈ 0.14π/a). Therefore, it is possible to observe that even though the coupling between normal/shear stresses/strains was optimized in the static regime (i.e., ω = 0), this also implies in coupled longitudinal-transverse motion. It is also interesting to note that the behavior predicted by the homogenized structure with effective properties (see Section “Computation of STL for a homogeneous anisotropic medium”), represented by the dashed green and purple lines in Fig. 3c, correlates well with the FE-based solution up to nearly 2.3 kHz, when the first branch starts to present significant dispersive behavior. We now proceed to evaluate the STL properties of a metabarrier composed of repetitions of the unit cell obtained in the optimization procedure. Metabarrier STL curves The unit cell resulting from the optimization process (Fig. 3b) presents inconvenient gaps in its left and right edges, resulting in finite structures with cavities at the borders. To remedy this issue, we may swap the left and right halves of the unit cell, keeping their orientation (regions inside the red and green contours in Fig. 4a), which does not affect the dispersion relation of the material due to its periodicity. This unit cell is then repeated along the x-direction to obtain finite structures with a thickness equal to h = Na, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N\in {{\mathcal{N}}}^{* }$$\end{document}N∈N* is the number of utilized unit cells. Infinite periodicity is considered in the y-direction.Fig. 4Metabarrier STL curves.a Redefinition of the unit cell (swapping regions inside the red and green contours, left to right) to obtain edges without gaps. b Sequential arrangement of unit cells filled with air and submerged into water. The medium is theoretically infinite in the y-direction, which allows to enforce periodic boundary conditions at the bottom and top edges of the unit cell. The ratio between the power of the transmitted and incident acoustic waves is then computed to obtain the STL curves. c STL curves for an increasing number of unit cells (N = 1, 2, …, 6). The peak (dip) displacement profiles are marked from i-p to vi-p (i-d to vi-d), shown in a normalized scale with respect to the largest absolute displacement. d Summary of values for STL peaks (left panel) and their corresponding frequencies (right panel). Metabarrier STL curves. a Redefinition of the unit cell (swapping regions inside the red and green contours, left to right) to obtain edges without gaps. b Sequential arrangement of unit cells filled with air and submerged into water. The medium is theoretically infinite in the y-direction, which allows to enforce periodic boundary conditions at the bottom and top edges of the unit cell. The ratio between the power of the transmitted and incident acoustic waves is then computed to obtain the STL curves. c STL curves for an increasing number of unit cells (N = 1, 2, …, 6). The peak (dip) displacement profiles are marked from i-p to vi-p (i-d to vi-d), shown in a normalized scale with respect to the largest absolute displacement. d Summary of values for STL peaks (left panel) and their corresponding frequencies (right panel). The resulting finite structures are utilized to evaluate the STL considering water as the surrounding fluid, as illustrated in Fig. 4b. The numerical procedure for computing the STL curves using the FE method is described in the Supplementary Information (see Fig. S1), which we perform considering a solid structure filled with air fluid. The effect of considering air-filled cavities is a slight blue-shift in the STL curves with respect to the case without fluid (this comparison is presented in Fig. S2 in the Supplementary Information). The material properties of water (air) are specific mass density ρw = 998 kg/m3 (ρa = 1.2 kg/m3) and speed of sound cw = 1481 m/s (ca = 343 m/s). The resulting STL curves are shown in Fig. 4c for an increasing number of unit cells, N = 1, 2, …, 6. We note that for the maximum frequency where a valid homogenization limit is considered (f = 2.3 kHz), the acoustic wavelength in water is equal to λ = 644 mm. Thus, for the largest considered metabarrier, we obtain a thickness h = 6a = 60 mm, indicating that a sub-wavelength regime is still in place (h/λ < 1/10). Displacement profiles corresponding to selected peaks and dips are also shown considering longitudinal and transverse displacements (ux and uy, respectively), normalized with respect to the largest computed displacement (i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$| u| =\sqrt{| {u}_{x}{| }^{2}+| {u}_{y}{| }^{2}}$$\end{document}∣u∣=∣ux∣2+∣uy∣2). The vertical dashed red line indicates the start of the frequency region where a single wave mode is present (corresponding to the horizontal dashed red line in Fig. 3c). We begin our analysis with N = 1, a particularly interesting case due to its ultra-thin profile. In the shown frequency range, this structure presents a maximum STL of 29.0 dB at 2.07 kHz (displacement profile i-p). At this frequency, the structure presents a deep sub-wavelength performance, since the ratio between the thickness h = a and the associated wavelength λ = cw/f is equal to h/λ ≈ 1/70. The STL curve reaches a first dip at 4.7 kHz, presenting a symmetric displacement profile (displacement i-d), which justifies a unit ratio between displacements and also acoustic pressures, yielding a zero STL. However, no visible correlation can be described with respect to the dispersion diagram shown in Fig. 3c due to the lack of periodicity in this case, which considers a single unit cell. For the case N = 2, a first STL peak of 24.4 dB is computed at 0.83 kHz, with a displacement profile (ii-p) indicating a behavior equivalent to that of the first peak computed at N = 1 (i-p), however with a longer wavelength for both ux and uy (i.e., the displacement profiles become distributed along two unit cells, instead of one). The second STL peak presents an attenuation of 31.8 dB at 2.63 kHz. We do not show this displacement profile since the formation of modes around 2.67 kHz will be discussed ahead. Also, a single dip is formed between these peaks at 1.47 kHz, displaying a symmetric displacement profile with respect to the y-axis (ii-d), which has the same type of symmetry displayed by the first STL dip with N = 1 (i-d). Thus, it is possible to notice that similar behaviors are observed for the first peak and first dip for an increasing number of unit cells (N = 1, 2), indicating the preservation of the behavior of equivalent peaks and dips, however, with a significant red-shift in frequency. For the case N = 3, three STL peaks are computed with attenuations of 23.5, 25.2, and 89.8 dB, at the frequencies 0.57, 1.49, and 4.00 kHz, respectively. The first two peaks (not shown here for the sake of brevity) present a similar behavior to the previously computed peaks for the cases N = 1 and N = 2, however with a longer wavelength (3 unit cells instead of 1 or 2). Interestingly, the third peak, presenting a much more considerable attenuation, displays a displacement profile similar to a wave mode associated with Bragg-scattering (iii-p), which can be explained due to the frequency range where it occurs (i.e., above 2.67 kHz). We note that the peak appears at a frequency outside the considered valid homogenization limit, which suggests a different explanation for the formation of this high STL value. We hypothesize that this peak is associated with the occurrence of a partial band gap with respect to the first propagative branch (see Fig. 3c). The only propagative branch in this frequency range corresponds to a mode tending to transverse motion, which in theory could hinder the efficient propagation of longitudinal (acoustic) modes. In this case, the dips are located at 1.03 and 1.93 kHz. It is interesting to notice that the displacement profile associated with the second dip (iii-d) displays a similar behavior to the wave mode computed at the X-point (see Fig. 3c) at its center unit cell. For the remaining considered cases (N = 4, 5, 6), we only show the dips that are closest to the frequency corresponding to the wave mode at the X-point (2.67 kHz) and the first peak observed immediately above this frequency. The corresponding dips (iv-d, v-d, and vi-d, respectively) present similar displacement profiles, with a behavior explained by the wave mode at the X-point of the first propagative branch (see Fig. 3c). Indeed, for a sufficient number of unit cells, a vibration mode is expected precisely at the frequency corresponding to this wave mode, in the case of unit cells with vacuum in their cavities (situation corresponding to the computed band diagrams). We have verified this by computing the STL curve for a total of N = 10 unit cells, checking the existence of the last dip at this region at 2.6 kHz. It is also interesting to note that the displacement profiles corresponding to the peaks (iv-p, v-p, and vi-p, respectively) present similar behaviors to that of the dips, however with a spatial decay, which suggests a Bragg-scattering behavior considering the first branch of the dispersion diagram. The existence of the second branch at the same frequency range, however, impedes the formation of a complete band gap in this direction. We also note here the STL values and frequencies for the sequential peaks occurring at each case, respectively, (i) N = 4: 23.2, 22.8, 29.1, and 42.4 dB at 0.44, 1.12, 1.93, and 3.14 kHz; (ii) N = 5: 23.0, 21.4, 26.9, 30.4, and 102.3 dB at 0.37, 0.90, 1.52, 2.20, and 3.17 kHz; and (iii) N = 6: 22.9, 20.5, 25.5, 27.9, 32.4, and 49.5 dB at 0.31, 0.76, 1.25, 1.81, 2.37, and 3.06 kHz. At each case, the dips are formed at (i) N = 4: 0.81, 1.42, and 2.31 kHz; (ii) N = 5: 0.67, 1.13, 1.84, and 2.46 kHz; and (iii) N = 6: 0.57, 0.95, 1.53, 2.06, and 2.56 kHz. For the sake of completeness, we point out here that several studies have reported a rather good agreement between numerical predictions and experimental STL measurements in comparable configurations, albeit at higher frequency ranges. In refs. 33,35, and36, consistent agreement is observed between computed and measured STLs over the ranges 500–2000 Hz, 400–1200 Hz, and 1–5 kHz, respectively. Moreover, in ref. 34, numerical STL predictions in the 1.5–3.5 kHz range exhibit peaks exceeding 20 dB, with experimentally measured average values of 18.7 dB. Although exact peak values are not always explicitly reported, the agreement in terms of trends and average attenuation levels provides a meaningful benchmark. These results indicate a strong correlation between numerical and experimental STLs in comparable metamaterial-based barriers, thereby supporting the reliability of the numerical framework adopted in the present work and the validity of the predicted low-frequency performance. We summarize the data concerning the attenuation and frequency of each maximum for the computed STL curves in Fig. 4d for an increasing number of unit cells. It is interesting to notice that, although for certain cases some peaks present a pronounced attenuation behavior similar to an anti-resonance (e.g., third peak for N = 3 and fifth peak for N = 5, left panel in Fig. 4d), the general trend concerning attenuation levels presents a practically constant value. On the other hand, for a sufficient number of unit cells, the frequency at which these peaks occur presents an almost linear behavior starting from the first peak (e.g., see N = 6, right panel in Fig. 4d). At this point, the question may arise whether the STL properties in this type of medium are only due to the high degree of hybridization observed (p ≈ 0.5) in the wave modes shown by the dispersion diagram (Fig. 3c). To verify that this is not the case, we have performed an additional optimization procedure resulting in a structure with a perfect coupling between longitudinal and shear motion (p ≈ 0.5 for all wave vectors in the [0, 5] kHz frequency range). In fact, the STL curves computed for this type of structure (see Fig. S3 in the Supplementary Information) present smaller peak values. This demonstrates that the fundamental ingredient to guarantee large STL values is the mode coupling factor δ, as previously verified in Section “Effect of normal-shear coupling on STL curves”. Influence of hydrostatic pressure on the design of the metabarrier After presenting the underlying physical phenomena responsible for generating significant STL levels, we now turn to a practical implementation concerning a fixed number of unit cells. To this end, we proceed with the design of our metabarrier by choosing a fixed number of unit cells with considerable sub-wavelength thickness as well as remarkable STL features. We choose N = 3 (see Fig. 4c, middle left panel), whose thickness-to-wavelength ratio is h/λ ≈ 1/50 in the operating frequency of 1 kHz, also presenting a considerable STL (≈90 dB) at 4 kHz. An immediate concern considering the applicability of the metabarrier is the increase in stress levels due to the hydrostatic pressure after its submersion in water. In particular, prohibitive stress concentrations might occur due to the geometrical variations presented by the optimized unit cell. To this end, we propose a solution which consists in increasing the unit cell overall thickness by adding layers of a homogeneous material with length hw (Fig. 5a), resulting in a total length h = Na + 2hw (in this case, N = 3). We note that this is a common solution for underwater lattice-based metamaterials33,35,36 with the dual purpose of (i) reducing maximum stress levels at the left and right faces of the unit cell and (ii) simplifying the geometry at the fluid-structure interface, allowing for plane waves to be directly applied at the metabarrier’s input face.Fig. 5Effects of hydrostatic pressure on the stress distribution and STL curves.a Finite structure obtained by adding two layers with a thickness hw made of a homogeneous isotropic material to the previously obtained unit cell configuration, chosen with N = 3. This structure is submerged into water, undergoing a hydrostatic pressure p0 on its lateral walls. b The value of hw is increased from 0 to 1.0 mm, indicating the decrease of the maximum von Mises stress from 108.6 (left panel) to 27.4 MPa (right panel). c Variation of average von Mises stress and average longitudinal strain as the depth d increases. The curve indicates a linear behavior of the sample under compression from d = 0 to d = 50 m. d STL curves for increasing values of hw, indicating a blue-shift (red-shift) in the frequency range below (above) 2.7 kHz. This difference in behavior is owed to the diverse attenuation formation mechanisms in each frequency range. e STL curves for the case hw = 1.0 mm for increasing depth values d = {0, 10, 20, 30, 40, 50} m, indicating a red-shift in the STL curves due to the development of compressive stresses. f STL variation with respect to increasing incidence angle (θ), showing an improvement in performance in the low-frequency range and a blue-shift of the highest STL peak for increasing incidence angles, followed by a decreased attenuation. Effects of hydrostatic pressure on the stress distribution and STL curves. a Finite structure obtained by adding two layers with a thickness hw made of a homogeneous isotropic material to the previously obtained unit cell configuration, chosen with N = 3. This structure is submerged into water, undergoing a hydrostatic pressure p0 on its lateral walls. b The value of hw is increased from 0 to 1.0 mm, indicating the decrease of the maximum von Mises stress from 108.6 (left panel) to 27.4 MPa (right panel). c Variation of average von Mises stress and average longitudinal strain as the depth d increases. The curve indicates a linear behavior of the sample under compression from d = 0 to d = 50 m. d STL curves for increasing values of hw, indicating a blue-shift (red-shift) in the frequency range below (above) 2.7 kHz. This difference in behavior is owed to the diverse attenuation formation mechanisms in each frequency range. e STL curves for the case hw = 1.0 mm for increasing depth values d = {0, 10, 20, 30, 40, 50} m, indicating a red-shift in the STL curves due to the development of compressive stresses. f STL variation with respect to increasing incidence angle (θ), showing an improvement in performance in the low-frequency range and a blue-shift of the highest STL peak for increasing incidence angles, followed by a decreased attenuation. To determine a suitable value of hw, we perform quasi-static analyses by imposing a pressure using consistent nodal loads in the x-direction facing inwards the unit cell at the lateral walls of the metabarrier (see Fig. 5a) given by p0 = ρwgd, where g = 9.81 m/s2 is the gravitational acceleration and d is the depth at which the structure is submerged. For the quasi-static FE analyses, the vertical degrees-of-freedom of the bottom and top edges (y = − a/2 and y = + a/2, respectively) present opposing values (i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${u}_{y}^{{\rm{top}}}=-{u}_{y}^{{\rm{bottom}}}$$\end{document}uytop=−uybottom, for each node) and horizontal degrees-of-freedom at these edges present equal values (i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${u}_{x}^{{\rm{top}}}={u}_{x}^{{\rm{bottom}}}$$\end{document}uxtop=uxbottom, for each node). Additionally, as we consider a plane strain state, the structure is theoretically infinite in the z-direction. Thus, we consider that additional homogeneous covers can be included in the finite structure (z = ± ∞) to impede water penetration inside the metabarrier, only allowing air pockets inside the structure. A non-linear FE analysis is performed considering large displacements with a Total Lagrangian formulation52, since a priori linearity is not guaranteed and geometrical buckling might occur. We then evaluate the stress distribution in the structure for the maximum depth of d = 50 m53 (i.e., p0 = 0.49 MPa). For the initial case hw = 0 (i.e., no additional thickness), the maximum computed von Mises stress is 108.6 MPa (Fig. 5b, left panel). We then increase the thickness hw until this maximum von Mises stress reaches a value close to 50% of the material’s tensile strength (55 MPa, no yield strength available from the manufacturer’s datasheet). We then end up with a final thickness of hw = 1.0 mm for a maximum von Mises stress of 27.4 MPa (Fig. 5b, right panel). We also verify that the maximum von Mises stress is indeed highly concentrated, while the overall mean von Mises stress levels are considerably lower. It is also worth to notice that the maximum lateral displacements at the left and right edges, computed for the values hw equal to 0, 0.1, 0.5, and 1.0 mm, are negligible with respect to the unit cell total length, presenting, respectively, the values 0.49, 0.38, 0.22, and 0.03 mm. The maximum vertical displacements at the bottom and top edges of the metabarrier unit cell are equal to 94.5, 82.8, 49.9, and 6.2 μm, respectively (i.e., negligible with respect to the unit cell dimensions). For the case of increased thickness hw = 1.0 mm, we also plot the computed average von Mises stress level (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{{\sigma }_{v}}$$\end{document}σv¯) versus average longitudinal strain (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{{\epsilon }_{x}}$$\end{document}ϵx¯) for increasing levels of depth d, as shown in Fig. 5c. The resulting curve shows an excellent correlation with a straight line (i.e., constant slope), thus indicating that, up to the selected depth of d = 50 m, no buckling or other geometrical softening mechanisms occur. Next, we verify the influence of the additional thickness on the STL curves by re-computing these for increasing values of hw with d = 0 (i.e., only the influence of hw is assessed at these analyses). In this case, we still consider the y-direction unit cell length equal to a. The results, shown in Fig. 5d, show that increasing the lateral thickness hw has a two-fold effect on the STL curves, which present (i) a blue-shift effect in the region below 2.67 kHz and (ii) a red-shift effect in the region above 2.67 kHz. This observation is in agreement with the previous statement that the formation mechanism for the STL peaks and dips is different for the regions divided by the frequency corresponding to the wave mode formed at the X-point (see Fig. 3c), namely, due to (i) the effective medium properties for lower frequencies and (ii) Bragg-scattering-like behavior for higher frequencies. This is also verified by re-computing the effective constitutive matrix for each case, from which we obtain the values of δ equal to 0.9851, 0.9848, and 0.9847, respectively, for hw equal to 0.1, 0.5, and 1.0 mm. Such a decrease in the value of δ explains both the blue-shift and the slight decrease in the STL peaks below 2.67 kHz. At the low-frequency region, it is also possible to verify that the behavior of the cases 0.5 and 1.0 mm are similar, which is explained by their approximate δ value. On the other hand, the STL curves obtained at higher frequencies present the opposed behavior, which however cannot be directly correlated to the band diagram presented in Fig. 3c due to the broken periodicity of the system caused by the introduction of the additional thickness. We also note that the STL peaks concerning the case hw = 1.0 mm now present the values 23.5, 21.5, and 84.3 dB, respectively, at 0.73, 1.71, and 3.27 kHz. We also investigate the influence of the hydrostatic pressure on the STL curves. To this end, we consider the inclusion of an additional stiffness matrix48 caused by the stresses induced by hydrostatic pressure. Figure 5e presents this variation with respect to the depth values d = 0, 10, 20, …, 50 m. It is possible to observe a red-shift effect concerning the whole STL curve, which is explained by the induced compressive stresses causing an overall softening of the metabarrier that decreases its characteristic frequencies. We note that for the case d = 50 m, the highest peak is shifted down to 2.86 kHz (12.5% reduction with respect to 3.27 kHz for d = 0), presenting a mild variation in the peak value (smaller than 2.7%). Finally, the performance of the metabarrier with respect to varying incidence angles is computed. The results, shown in Fig. 5f, indicate that in the first frequency region (below 2.67 kHz), the STL curves present a mild blue-shift, especially at higher frequencies (close to 2 kHz), however presenting also a significant increase in attenuation for larger incidence angles (close to 40 dB at θ = 80o). Also, for the second frequency region (above 2.67 kHz), the STL peak presents a decreased attenuation and a much more significant blue-shift effect; however, remaining above 60 dB for all computed cases. Once again, the different behaviors displayed by the structure when considering these frequency ranges are due to the different mechanisms that produce attenuation in each case. Finite two-dimensional circular structure To numerically validate our findings, we now verify the proposed unit cell’s performance in protecting an underwater region from noise generated by a point source. To this end, we dispose the unit cell in a circular manner, enclosing a point C at the center of a two-dimensional square space of length L, as shown in Fig. 6a. Additional perfectly matched layers (PMLs) of length ΔL are included to minimize reflections from the system boundaries, thus approximating an infinite medium in the xy plane. The considered medium is square due to the chosen PML FE implementation54. The distance from any point in the two-dimensional system to point C is indicated as r.Fig. 6Two-dimensional numerical experiment to verify the performance of the circular metabarrier.a Two-dimensional square medium with side length L, where a curved metabarrier, centered at C, is immersed in water. The radial distance r is measured with respect to C. Additional PML layers (side length ΔL) are included at the external boundary of the fluid medium. b The x- (y-)coordinates of the initially rectangular unit cell (top panel) are used to generate corresponding radial r (angular θ) coordinates and obtain a section of a circular shape, yielding an internal radius ri and an external radius re (bottom panel). c Manufactured structure with ri = 88 mm and re = 120 mm. d Input time-transient signal applied at the center C. e Normalized pressures obtained along the radial distance r without the metabarrier (left panel) and with the metabarrier (right panel). Two-dimensional numerical experiment to verify the performance of the circular metabarrier. a Two-dimensional square medium with side length L, where a curved metabarrier, centered at C, is immersed in water. The radial distance r is measured with respect to C. Additional PML layers (side length ΔL) are included at the external boundary of the fluid medium. b The x- (y-)coordinates of the initially rectangular unit cell (top panel) are used to generate corresponding radial r (angular θ) coordinates and obtain a section of a circular shape, yielding an internal radius ri and an external radius re (bottom panel). c Manufactured structure with ri = 88 mm and re = 120 mm. d Input time-transient signal applied at the center C. e Normalized pressures obtained along the radial distance r without the metabarrier (left panel) and with the metabarrier (right panel). In order to comply with the curved geometry, the metabarrier unit cell must be slightly modified with respect to its original (rectangular) design. Hence, we position the left (right) edge of the unit cell at x = ri (x = re) and its x-centerline aligned with y = 0. We then compute the angle α described by C with respect to a pair of points located at the center of the bottom (y = − a/2) and top edges (y = + a/2) of the unit cell, whose x coordinate is given by x = (ri + re)/2, as depicted in Fig. 6b. This angle is thus defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha =2{\tan }^{-1}(a/({r}_{i}+{r}_{e}))$$\end{document}α=2tan−1(a/(ri+re)), yielding a total number of nc = 2π/α, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${n}_{c}\in {{\mathcal{N}}}^{* }$$\end{document}nc∈N*, unit cells in the angular direction, creating a structure with cyclic symmetry. Due to fabrication constraints, we set the outer radius of the structure as re = 120 mm. The thickness of the metabarrier is fixed from the previous design as h = 32 mm, yielding an inner radius equal to ri = re − h = 88 mm. From these quantities, we compute α = 0.0961 rad, leading to a total number of nc = 65.4 unit cells, which cannot be implemented (non-integer number). We therefore approximate this number to nc = 64 unit cells to ensure a π/2-rotational symmetry on the circular metabarrier (to alleviate directional biasing associated with meshing), thus yielding a corrected angle α = 2π/64 = 0.0982 rad (≈2% deviation). The initial geometry of the unit cell, described in the (x, y) coordinates, is then mapped into a polar coordinate system (r, θ) following (x → r, y → θ), thus obtaining the final configuration \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({x}^{{\prime} }=r\cos \theta ,{y}^{{\prime} }=r\sin \theta )$$\end{document}(x′=rcosθ,y′=rsinθ). The manufactured structure is presented in Fig. 6c. We recall that acoustic wavelengths in water are significantly longer than in air (approximately 1.5 m for a 1 kHz plane wave), implying that a peak around 500 Hz would correspond to a wavelength of nearly 3 m. This would require experimental facilities of considerable dimensions, which, to the best of the authors’ knowledge, are not currently available in a standardized form, nor commonly accessible in research laboratories to date. Moreover, impedance-tube measurements may be affected by coupling between fluid-borne and structural elastic waves, to carefully account for during the measurements. Therefore, to consider real operation conditions and frequencies well below 1 kHz (thus, avoiding scale-related issues), the authors are currently progressing toward the implementation of dedicated low-frequency measurements in a natural lake environment, which represents the most suitable experimental configuration for the frequency range of interest. However, this approach requires careful experimental design and involves solving substantial technical challenges and fabrication. These efforts are part of ongoing work and will be reported separately. We proceed to perform a FE analysis of the resulting metabarrier embedded in an underwater medium with L = 1000 mm, considering a PML with length ΔL = 250 mm. The structural response of the metabarrier and the pressure distribution in the cavity (region encircled by the metabarrier) and external domain are computed using the FE model described in55. The pressure distribution in the fluid, p(r, ω), r ∈ [re, L/2], is obtained by imposing a unit acoustic force at the center point C, while zero Dirichlet boundary conditions are assigned to the outer edges of the PML (required for numerical accuracy). Using this approach, we compute a set of frequency response functions (FRFs) between the input point C and the output points distributed along a radius r, taken as directed along the x-axis of the two-dimensional domain, for simplicity. We then consider a 4-pulse sinusoidal signal centered at 500 Hz and modulated with a Hanning window (see Fig. 6d), whose frequency content is mostly concentrated between 0 and 1000 Hz. The pressure signals at the output points are then obtained via the product between the computed FRFs and the fast Fourier transform (FFT) relative to the input signal, which is transformed back into the time domain using inverse FFTs. The computed transient pressure signals are shown in Fig. 6e for the case of the absent metabarrier (left panel) using normalized values, revealing a weak dependence with respect to r. In order to quantify the amount of transmitted energy, we compute the integral \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\int }_{0}^{\infty }\,{p}^{2}(t)\,dt$$\end{document}∫0∞p2(t)dt, which in this case, is equal to 1.6 × 10−3 (in arbitrary units). We then proceed analogously for the case with the metabarrier. The corresponding results, shown in Fig. 6e (right panel), indicate a considerable reduction in the transmitted pressures between the region protected by the metabarrier and its exterior. In this case, the time integral yields \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\int }_{0}^{\infty }\,{p}^{2}(t)\,dt=3.3\times 1{0}^{-5}$$\end{document}∫0∞p2(t)dt=3.3×10−5, thus indicating a reduction of nearly 98% in terms of transmitted energy. Details on the computed IL, pressure distribution, and displacement fields are given in the Supplementary Information (see Fig. S4). Discussion In conclusion, we have proposed a design strategy to exploit structural anisotropy in metamaterial-based acoustic barriers, yielding efficient sub-wavelength underwater noise reduction. We have calculated the STL curves considering an anisotropic homogeneous medium immersed in a fluid. The first peak of these curves presents (i) an increasing maximum value and (ii) a decreasing frequency as a conveniently defined mode coupling factor between normal stresses and shear strains (and vice versa) approaches unity. We have also shown that the thickness of the anisotropic barrier has a lesser influence on the maximum STL value, suggesting that this can be used to tune the frequency of the STL peaks. This effect is contrary to Bragg-scattering band gaps, whose efficiency in wave attenuation is proportional to the number of unit cells (and therefore thickness) of the structure. These results were then used to propose a topology optimization problem with the objective of maximizing the absolute value of the mode coupling factor. The dispersion relation obtained using the optimized unit cell indicates a high degree of longitudinal-transverse polarization, which, however, was demonstrated not to be responsible for increasing maximum STL values. A partial band gap was also identified with respect to one of the propagative branches. The STL curves computed for an increasing number of unit cells immersed in water confirm the previous conclusions regarding the effect of increasing thickness values on the STL curves. Interestingly, a deep sub-wavelength behavior is reported for a single unit cell (thickness-to-wavelength ratio circa 1/70). Furthermore, the effect of the partial band gap is confirmed, achieving an almost 90 dB attenuation using three unit cells (30 mm). The effects of hydrostatic pressure on the configuration of three unit cells were also evaluated, indicating the necessity of including additional layers of homogeneous material on each side of the metabarrier (1 mm each), which is sufficient to reduce stress levels by a fourfold. Including these layers leads to (i) a blue-shift of the STL curves for low frequencies and (ii) a red-shift of the STL curves for higher frequencies, without, however, significantly degrading STL levels. Likewise, the hydrostatic-induced stresses lead to a red-shift in STL curves due to their compressive nature. The STL curves also present (i) increases in peaks for low frequencies and (ii) blue-shifts for higher frequencies for increasing incidence angles. These conclusions demonstrate that the designed metabarrier is robust with respect to variations in external stresses and incidence angles. Finally, we proposed a convenient transformation for the unit cell, morphing its shape from a rectangular to a circular design. We have also shown the manufacturability of the structure for an external radius of 120 mm and a thickness of 32 mm. The computed time-transient signals indicate a reduction of nearly 98% in energy for an input signal centered at 500 Hz with the use of the metabarrier. Thus, not only does the proposed structure seem suitable for technological exploitation, but its rationally designed anisotropy also indicates a fruitful direction for further investigation in underwater acoustics. Supplementary information Supplementary Information Supplementary Information