1. Introduction
Wave propagation at the interface of two immiscible fluids is ubiquitous in nature. Observable at spatial scales ranging from micrometres to kilometres (Whitham Reference Whitham2008), these surface waves convey information on the state of both the fluid dynamics via their interactions with flow and fluid dynamical structures (such as vortices or swells Craik Reference Craik1988; Buhler Reference Bühler2014) as well as its surrounding boundaries (Falkovich Reference Falkovich2011).
A distinctive feature of wave propagation (and in particular of surface water waves) is its ability to transport energy, which is of fundamental importance for basic and applied research, specifically for harvesting surface wave energy (Anton & Sodano Reference Anton and Sodano2007; Chen et al. Reference Chen2015; Wang, Jiang & Xu Reference Wang, Jiang and Xu2017). To this end, controlling surface wave propagation becomes of paramount importance. It is within this context that wave guiding becomes an appealing research avenue, that is, the localisation of wave energy in a confined region of space (Collin Reference Collin1990). For surface water waves, wave guiding has been theoretically predicted to develop along edges, ridges or coasts (Longuet-Higgins Reference Longuet-Higgins1967, Reference Longuett-Higgins1968; Buchwald Reference Buchwald1968, Reference Buchwald and Adams1969; Huthnance Reference Huthnance1975; Renardy Reference Renardy1983; Brink Reference Brink1991; Zhevandrov Reference Zhevandrov1991; Evans & Linton Reference Evans and Linton1993; Bryan & BowenReference Bryan and Bowen1993; Merzon & Zhevandrov Reference Merzon and Zhevandrov1998; Chamberlain & Porter Reference Chamberlain and Porter1999; Kurkin & Pelinovsky Reference Kurkin and Pelinovsky2002; Johnson Reference Johnson2005; Brink Reference Brink2006; Johnson Reference Johnson2007; Dodd et al. Reference Dodd, Stoker, Calvete and Sriariyawat2008; Wang et al. Reference Wang, Liang, Shi and Zheng2021). This feature has also been numerically simulated (Zamudio & Lopez Reference Zamudio and López1994; Hsiao, Shih & Wu Reference Hsiao, Shih and Wu2018), observed in the laboratory (Caldwell, Cutchin & Longuet-Higgins Reference Caldwell, Cutchin and Longuet-Higgins1972; Stocks Reference Stocks1976; Pite Reference Pite1977; Liu et al. Reference Liu, Cho, Kostense and Dingemans1992; Long et al. Reference Long, Lai, Huang and Spedding1993; Codiga, Renouard & Fincham Reference Codiga, Renouard and Fincham1999; Drivdal, Weber & Debernard Reference Drivdal, Weber and Debernard2016; Liao & Wang Reference Liao and Wang2018; DeFreitas et al. Reference De Freitas, Paiva, Cirano, Mill, da Costa, Gabioux and França2021; Wang et al. Reference Wang, Zhang, Zheng, Fu and Tao2023) and even environmentally recorded at large scales (Boyd & Smith Reference Boyd and Smith1983; Merrifield Reference Merrifield1992; Elgar, Herbers & Guza Reference Elgar, Herbers and Guza1994; Ralph et al. Reference Ralph, Armi, Bane, Dorman, Neff, Neiman, Nuss and Persson1998; Pizarro & Shaffer Reference Pizarro and Shaffer1998; Pruszak et al. Reference Pruszak, Róźyński, Szmytkiewicz and SkajaField2007; Igeta, Yamazaki & Watanabe Reference Igeta, Yamazaki and Watanabe2014; Inall et al. Reference Inall, Nilsen, Cottier and Daae2015; Weber & Isachsen Reference Weber and Isachsen2023). Wave guides for surface water waves can also be generated by large-scale currents (Howd, Bowen & Holman Reference Howd, Bowen and Holman1992; Tang & Grimshaw Reference Tang and Grimshaw1995; Miller, Matano & Palma Reference Miller, Matano and Palma2011; Shrira & Slunyaev Reference Shrira and Slunyaev2014a ,Reference Shrira and Slunyaevb ; Slunyaev Reference Slunyaev2023), vortices (Shelton & Trinh Reference Shelton and Trinh2023; Zuccoli, Brambley & Barkley Reference Zuccoli, Brambley and Barkley2024), artificially constructed structures (Yang, Gao & Zhang Reference Yang, Gao and Zhang2016; Wu, Wu & Mei Reference Wu, Wu and Mei2018; Laforge et al. Reference Laforge2019; Zhu et al. Reference Zhu, Zhao, Han, Zi, Hu and Chen2024) and even by topological constraints of the wave-guiding mechanism (Venaille & Delplace Reference Tauber, Delplace and Venaille2020).
The theoretical tools to study wave guiding for surface water waves have initially been borrowed (and later perfected) from optics. Initial work on surface water wave guiding using ray tracing (Arthur, Munk & Issacs Reference Arthur, Munk and Isaacs1955) suggested a direct analogy to its optical counterpart (Collin Reference Collin1990), which can be used to understand the effect of flow and bathymetry changes on the wave-guide properties. Longuet-Higgins (Reference Longuett-Higgins1968) theorised that waves with evanescent profiles can be trapped along depth discontinuities in a rotating system. A couple of months later, Buchwald (Reference Buchwald1968) expanded Longuet–Higgins’s calculations along oceanic ridges, finding an infinite number of guided wave modes. Experimental wave-guiding evidence was soon presented by Caldwell, Cutchin & Longuet-Higgins (Reference Caldwell, Cutchin and Longuet-Higgins1972) following Longuet-Higgins’s (Reference Longuett-Higgins1968) work within a rotating basin, and later confirmed by Stocks (Reference Stocks1976) in the shallow water limit for straight and curved shelf bathymetries via point-like measurements. Since these initial studies, theoretical, numerical and experimental techniques have evolved, allowing an accurate description of surface water wave guiding. In particular, experimental two-dimensional profilometry techniques (Takeda & Muthoh Reference Takeda and Mutoh1983; Grant, Stewart & Padilla-Perez Reference Grant, Stewart and Padilla-Perez1990; Wright Budakian & Putterman Reference Wright, Budakian and Putterman1997; Maurel et al. Reference Maurel, Cobelli, Pagneux and Petitjeans2009; Moisy, Rabaud, & Salsac Reference Moisy, Rabaud and Salsac2009; Przadka et al. Reference Przadka, Cabane, Pagneux, Maurel and Petitjeans2012) have allowed measurements of surface deformations with high spatial and temporal resolution, and have already been applied to the study of surface wave guiding by Sepulveda et al. (Reference Sepúlveda-Soto, Guzmán-Silva, Rosas, Vicencio and Falcón2020).
Most of the cases described above deal with surface water wave guides aligned with a given constant axis, and thus curvature effects on the wave guide are not taken into account. In the case of optical wave guides, curvature effects have been shown to dampen the amplitude of propagating modes within a wave guide (Marcuse Reference Marcuse1971; Thyagarajan, Shenoy & Ghatak, Reference Thyagarajan, Shenoy and Ghatak1987; Lui et al. Reference Lui, Xu, Hirono, Yokoyama and Huang1998; Berglund & Gonipath Reference Berglund and Gonipath2000; Krause Reference Krause2011) and even completely negate the wave-guiding properties of a structure by radiating wave energy away from the structure. Although curvature effects on optical wave guiding have been extensively studied (specifically within fibre optics) due to their promising applications for modal decomposition (Accatino & Bertin Reference Accatino and Bertin1998), mixing (Vinogradov & Denisov Reference Vinogradov and Denisov1990), filtering (Little et al. Reference Little, Chu, Haus, Foresi and Laine1997, Reference Little, Chu, Absil, Hryniewicz, Johnson, Sieferth, Gill, Van, King and Trakalo2004) and even amplification via resonance (Cai et al. Reference Cai, Lu, Chen, Lee, Lin and Yen2015) of optically guided waves, similar focus on curvature effects within continuous mechanics (such as the ones describe for elastic plates by Demma Cawley & Lowe (Reference Demma, Cawley and Lowe2002) and Fong & Lowe (Reference Fong and Lowe2004), and the work of Tam (Reference Tam1976) and Cho & Galan (Reference Cho and Galan2004) for acoustic waves) is lacking. It is within this context that the present work is framed: we study experimentally and theoretically shallow water surface waves of wavelength
$\lambda$
propagating over a layer of depth
$h_o\ll \lambda$
with a step-like wave guide (with a given width
$a$
, height
$h_w\lt h_o$
and constant radius of curvature
$R$
) which display partial wave guiding due to curvature effects.
The paper is structured as follows: in § 2 we describe the experimental set-up, the surface profilometry method used in this work and some results of guided modes along a straight wave guide for shallow water waves. Section 3 deals with the theoretical description of shallow water equations over step-like wave guides, both straight and curved. In § 4 we describe our main experimental results of shallow water waves propagating over curved wave guides. Lastly, in § 5, we will discuss the main results of our work, and present our conclusions with some perspectives of possible new research avenues regarding shallow water wave guiding along more complex wave-guiding structures.
2. Experimental set-up and measuring techniques
2.1. Experimental wave tank and wave guides
The experimental set-up to study water wave guiding is similar to the one described in Sepulveda et al. (Reference Sepúlveda-Soto, Guzmán-Silva, Rosas, Vicencio and Falcón2020) and depicted schematically in figure 1. Water waves propagate along the surface of a layer of distilled water (thickness
$h_o=$
2.5 cm) within a large wave tank (200 cm long, 60 cm wide and 20 cm deep). Reflections at the wave tank boundaries are strongly attenuated by placing sponges along the container’s inner walls. A total of 20 cm away from one of the tank’s ends, a plunging wave maker oscillates along the
${{\hat {\mathbf z}}}$
direction generating monochromatic surface waves. The motion of the wave maker is controlled by an electro-mechanical shaker (B
$\&$
K 4810) driven by the sinusoidal voltage generated by a function generator (Agilent 33521) in a frequency range
$f\in$
[3.0 Hz, 5.0 Hz], which is amplified by a power amplifier (SKA Pro Max 720) to achieve initial wave amplitudes
$\eta _o\in$
[0.5 mm, 2.0 mm]. A 10 cm long, 10 cm wide Plexiglass channel is positioned in front of the generated surface waves effectively creating a one-dimensional wave train (cf. figure 1). For our frequency range, the wave train displays typical wavelengths
$\lambda \in$
[6 cm, 16 cm] (wavenumbers
$k\in$
[0.4 cm
$^{-1}$
, 1.0 cm
$^{-1}$
]). Using these experimental parameters, we find that the non-dimensional numbers values
$\epsilon =k h_o\ll 1$
(shallowness),
$\delta =\eta _o/h_o\ll 1$
(aspect ratio) and
$\mathcal{U}=\delta /\epsilon ^{2}\ll 1$
(Ursell number) justify reasonably the linear wave assumption for our wave-guiding problem (Whitham Reference Whitham2008; Falkovich Reference Falkovich2011).
At the exit of this channel, the propagating wave train encounters a wave guide: a Plexiglass beam placed at the bottom of the wave tank which changes the local depth of the fluid. The wave guides were constructed by laser cutting three 5 mm thick Plexiglass sheets with a fixed width
$a=2.0$
cm and gluing them together to achieve a thickness
$h_g$
= 1.5 cm. We have fixed
$a$
to a single value to study only single propagating mode configurations over the wave guide, as increasing
$a$
allows more propagating modes (Sepulveda et al. Reference Sepúlveda-Soto, Guzmán-Silva, Rosas, Vicencio and Falcón2020). The shape of the wave guide is chosen to be a circular section with a given curvature radius
$R\in$
[24 cm, 160 cm], including the limiting case of a straight wave guide (
$R \rightarrow \infty$
). The surface wave train propagating along this wave guide is found to be spatially localised over the beam region with a characteristic transverse profile which decays in amplitude over the distance travelled along the wave guide.
Scheme of an experimental configuration for shallow water gravity wave guiding where a curved wave guide is depicted.

Results for straight wave guide showing the guiding different modes propagating over a guide with
$a=$
2.0 cm. (a,b) Correspond to the stroboscopic
$\textrm{Re}(\hat {\zeta }(x,y))$
and (c,d) to the envelope field
$|(\hat {\zeta }(x,y))|$
, both obtained via Fourier transform profilometry (FTP). These data were taken for a frequency
$f = 5.0$
Hz. The dotted lines over the right panels correspond to the segment cut plotted on the left panels and the solid lines correspond to the borders of the step-like wave-guide profile positioned on the bottom.

2.2. Fourier transform profilometry of guided waves
The surface deformations of the flat air–water interface are acquired using an optical technique based on fringe projection (Takeda & Mutoh Reference Takeda and Mutoh1983; Grant et al. Reference Grant, Stewart and Padilla-Perez1990) called Fourier transform profilometry (FTP) based on the work of Maurel et al. (Reference Maurel, Cobelli, Pagneux and Petitjeans2009). A sample of these deformations along a straight wave guide are depicted in figure 2. This technique, adapted specifically for fast free surface deformations, uses a highly diffusive interface as a background to project sinusoidal fringes which are acquired by a sensitive high-speed camera. To achieve such diffusive interface, titanium oxide (Kronos 1001 from Kronos Inc.) was added to distilled water (5 g l−1), dyeing the fluid white in the process while creating a canvas for fringe projection. It must be noted that, at this dye concentration, the physical properties of distilled water (mainly viscosity and surface tension) have been found to show negligible deviations from the pure fluid (Przadka et al. Reference Przadka, Cabane, Pagneux, Maurel and Petitjeans2012). To set up the FTP technique, a Norcan structure was constructed to support vertically a high definition light projector (Epson 8350) and a high-speed camera (Phantom V641) in a parallel optical axis configuration (Maurel et al. Reference Maurel, Cobelli, Pagneux and Petitjeans2009). The distance between both optical axes was set at 50 cm. Care was taken to set the 50 mm f 1.4 Nikon lens attached to the high-speed camera at the same horizontal level as the projector lens. In this configuration, the distance from the projector lens to the flat air–water interface was set at 180 cm and the observation window was set at
$100\times 60$
cm
$^{2}$
, which sets the horizontal and vertical sensitivity at 0.5 mm per pixel. The entire set-up is surrounded by black curtains to suppress other light sources in the image acquisition.
As the FTP deals with the demodulation of an intensity profile from a high-frequency carrier signal, the fringe pattern wavelength must be chosen to be much smaller than the typical surface deformation scales. We chose a 0.5 cm wavelength projected at the flat air–water interface (10 pixels per wavelength). In the same fashion, the acquisition frequency
$f_a$
is set at 30 Hz (i.e. 30 frames per second, much larger than the typical working frequencies for the experiments) and the acquisition time is set at 60 s. From the acquired image set for a given
$f_o$
and
$\eta _o$
, the surface deformation for a pixel at
$(x,y)$
within the observation window of image
$n=f_a \times t$
,
$\zeta (x,y,t)$
is computed. This deformation is time filtered around
$f_o$
with a bandwidth of 0.1 Hz. We have checked that second and third harmonics are at least two orders of magnitude smaller than the fundamental mode (Schmessane Reference Schmessane2016), and thus we can write
$\zeta (x,y,t)=\textrm{Re}(\hat {\zeta }(x,y)e^{i\omega _o t})$
with
$\omega _o=2\pi f_o$
. This stroboscopic component was computed by averaging 10 frames with the same temporal phase which is used to describe the spatial oscillations of the guided wave. We also computed the envelope of the guided wave
$|\hat {\zeta }(x,y)|$
which is used to track the modal profile and spatial decay rate of the guided wave. To wit, we depict the stroboscopic
$\textrm{Re}(\hat {\zeta }(x,y))$
and
$|\hat {\zeta }(x,y)|$
in figure 3 within a
$a=10$
cm straight wave guide for
$f_o=3.0$
for two different frequencies. This features the possibility of multimode wave guiding as a function of the guide’s properties, as in the case of its width
$a$
(Sepulveda et al. Reference Sepúlveda-Soto, Guzmán-Silva, Rosas, Vicencio and Falcón2020).
Guided modes over a straight wave guide with
$a=$
10 cm. Envelope field
$|\hat {\zeta }(x,y)|$
(a)
$f=$
3.0 Hz and (b)
$f=$
5.0 Hz. Stroboscopic field
$\textrm{Re}(\hat {\zeta }(x,y))$
for (c)
$f=$
3.0 Hz and (d)
$f=$
5.0 Hz. Solid lines correspond to the borders of the step-like wave-guide profile.

3. Theoretical framework
We begin by setting the problem’s conditions, variables and dynamical evolution equations for the problem of shallow water wave guiding.
3.1. Shallow water equations
In the shallow water regime, i.e.
$kh_o\ll 1$
, we consider an incompressible Newtonian fluid (density
$\rho$
and viscosity
$\nu$
) of undisturbed depth
$h_o$
in the presence of a uniform gravitational acceleration
$\boldsymbol{g}= -g\boldsymbol{\hat {z}}$
. The free surface elevation is described by
$h(\boldsymbol{r}_\perp , t) = h_o + \zeta (\boldsymbol{r}_\perp , t)$
, where
$\zeta (\boldsymbol{r}_\perp , t)$
is the local surface displacement while
$\boldsymbol{u}=\boldsymbol{u}(\boldsymbol{r}_\perp , t)$
describes the in-plane velocity field. Here,
$\boldsymbol{r}_\perp$
stands for in-plane coordinates These fields are coupled via the depth-averaged viscous shallow water equations
where we follow the initial prescription of Bresch & Desjardins (Reference Bresch and Desjardins2003), later specialised by Marche (Reference Marche2007), which has been successfully applied to describe the complex shallow water dynamics (Di Leoni, Cobelli & Mininni Reference di Leoni, Cobelli and Mininni2014; Reyes, Torrejon & Falcon Reference Reyes, Torrejón and Falcón2020; Liu, Hara & Shneider Reference Liu, Hara and Shneider2023). Here,
$\nu _o$
is a shallow water viscous coefficient which we can relate to a depth-average (Rayleigh) viscosity (Falkovich Reference Falkovich2011). This set of equations can be simply re-written in the limit of
$\epsilon ,\delta ,\mathcal{U} \ll 1$
and negligible viscosity as a wave equation
The layer depth can in principle depend on space
$h_o= h_o(\boldsymbol{r}_\perp )$
, which will change wave propagation in different regions of the domain. In our case the layer depth will be constant
$h=h_g$
inside the wave guide and
$h=h_0$
outside of the wave guide (cf. figure 1), that is
$h_m = \{h_g, h_0\}$
. Assuming monochromatic surface waves
$\zeta (\boldsymbol{r}_\perp , t) = \textrm{Re}(\hat {\zeta }(\boldsymbol{r}_\perp )e^{-i\omega t})$
, the former wave equation reads
with
$k_m = \omega /\sqrt {gh_m}$
the wavenumber inside and outside the wave guide. In what follows,
$\omega$
will be thought of as a control parameter of the monochromatic linear problem. Mass and momentum conservations impose continuity of
$\zeta$
and
$h\nabla _\perp \zeta$
at the boundaries of the wave guide. It must be stressed that (3.5) is coordinate-independent, which means that shallow water waves can be expressed in an appropriate coordinate system depending on the geometry of the problem. This will be of particular importance when dealing with curved shallow water wave guides.
3.1.1. Straight wave-guide limit
In the case of straight wave guides (
$R\to \infty$
) we can set a Cartesian coordinates system with
$\boldsymbol{r}_\perp = (x, y)$
, assuming the guide is set along the
$y$
direction we write
$\hat {\zeta }(\boldsymbol{r}_\perp ) = \textrm{Re}[\psi (x)e^{i\beta y}]$
(Sepulveda et al. Reference Sepúlveda-Soto, Guzmán-Silva, Rosas, Vicencio and Falcón2020). Here,
$\beta$
will be thought of as the eigenvalue of the problem in the straight wave-guide limit. Using this variable separation one obtains a reduced transverse equation
for
$\psi (x)$
which sets its evolution depending on the sign of
$k_m^2-\beta ^2$
. For guided solutions we impose
$k_g^2-\beta ^2\gt 0$
, and
$k_0^2-\beta ^2\lt 0$
which sets an oscillatory solution for
$\psi (x)$
inside the wave guide and a exponentially decaying one outside the wave guide. Thus, the transverse solutions for the guided wave, assuming finite amplitude solutions, are
\begin{equation} \displaystyle \psi (x)= \left \{ \begin{array}{ll} A\exp \left(\sqrt {\beta ^2-k_0^2}x\right)& x \lt -\tfrac {a}{2}, \\ \\ B\cos \left(\sqrt {k_g^2-\beta ^2}x\right) + C\sin \left(\sqrt {k_g^2-\beta ^2}x\right) & |x|\lt \tfrac {a}{2},\\ \\ D\exp \left(-\sqrt {\beta ^2-k_0^2}x\right)& x\gt \tfrac {a}{2}, \end{array} \right . \end{equation}
with
$A,B,C,D$
constants. Here, the wave-guide edges are set at
$\pm a/2$
. Imposing the continuity of mass and momentum via
$\psi (x)$
and
$\partial _x\psi (x)$
at
$\pm {a}/{2}$
, one finds relations between the constants allowing the calculation of the propagation constant
$\beta$
along the wave guide which can be reduced to solving the eigenvalue problem for
$A,B,C,D$
represented by
\begin{equation} \begin{bmatrix} \exp \left(-\dfrac {a}{2\lambda _0}\right)& -\!\cos \left(\dfrac {a}{2\lambda _g}\right)&+\!\sin \left(\dfrac {a}{2\lambda _g}\right)&0\\ \lambda _0^{-1}\exp \left(-\dfrac {a}{2\lambda _0}\right) & -\lambda _g^{-1} \sin \left(\dfrac {a}{2\lambda _g}\right)& -\lambda _g^{-1}\cos \left(\dfrac {a}{2\lambda _g}\right)& 0\\ 0&-\!\cos \left(\dfrac {a}{2\lambda _g}\right)& -\!\sin \left(\dfrac {a}{2\lambda _g}\right) &\exp \left(-\dfrac {a}{2\lambda _0}\right)\\ 0& -\lambda _g^{-1}\sin \left(\dfrac {a}{2\lambda _g}\right)&+\lambda _g^{-1}\cos \left(\dfrac {a}{2\lambda _g}\right)&\lambda _0^{-1}\exp \left(-\dfrac {a}{2\lambda _0}\right) \end{bmatrix}\! \begin{bmatrix} A\\B\\C\\D \end{bmatrix}\! =\! \begin{bmatrix} 0\\0\\0\\0 \end{bmatrix}, \end{equation}
where we have defined
$\lambda _0^{-1} = \sqrt {\beta ^2-k_0^2}$
and
$\lambda _g^{-1} = \sqrt {k_g^2-\beta ^2}$
. There are 2 families of propagating solutions to the eigenvalue problem in (3.8): the symmetric or anti-symmetric propagation modes (
$\psi _s(x)$
or
$\psi _a(x)$
, respectively), characterised by the symmetric or anti-symmetric propagation constant (
$\beta _{s}$
or
$\beta _a$
, respectively). Here,
$\beta _s$
and
$\beta _a$
satisfy the transcendental equations
\begin{align} \frac {\sqrt {k_g^2-\beta _s^2}}{\sqrt {\beta _s^2-k_0^2}}\tan \left ( \frac {a}{2}\sqrt {k_g^2-\beta _s^2}\right )-1 &=0, \end{align}
\begin{align} \frac {\sqrt {k_g^2-\beta _a^2}}{\sqrt {\beta _a^2 -k_0^2}}\cot \left ( \frac {a}{2}\sqrt {k_g^2-\beta _a^2}\right )+1 &=0, \end{align}
which set the corresponding eigenvectors
Using these definitions, one can write symmetric
$\psi _s(x)$
\begin{equation} \displaystyle \psi _s(x)= \left \{ \begin{array}{ll} \delta _s\cos \left(\dfrac {a}{2\lambda _g}\right)e^{-\tfrac {1}{\lambda _0}\left(|x|-\tfrac {a}{2}\right)}&|x| \gt \dfrac {a}{2}, \\ \\ \delta _s\cos \left(\dfrac {x}{\lambda _g}\right) & |x|\lt \dfrac {a}{2}, \end{array} \right . \end{equation}
and anti-symmetric
$\psi _a(x)$
\begin{equation} \psi _a(x)= \left \{ \begin{array}{ll} +\delta _a\sin \left(\dfrac {a}{2\lambda _g}\right)e^{\frac {1}{\lambda _0}\left(x+\tfrac {a}{2}\right)}& x \lt -\dfrac {a}{2}, \\ \\ \delta _a\sin \left(\dfrac {x}{\lambda _g}\right) & |x|\lt \dfrac {a}{2},\\ \\ -\delta _a\sin \left(\dfrac {a}{2\lambda _g}\right)e^{-\frac {1}{\lambda _0}\left(x-\frac {a}{2}\right)}& x \gt \dfrac {a}{2}, \end{array} \right . \end{equation}
guided propagative solutions with
$\delta _s$
and
$\delta _a$
the symmetric and anti-symmetric wave amplitude of the symmetric and anti-symmetric mode, respectively. We normalise the transverse profiles as
\begin{equation} \int \limits _{-\infty }^{+ \infty } |\psi _j(x)| {\textrm d}x = I_j, \end{equation}
with
$j=a,s$
. Here,
$I_j$
is computed numerically for a given frequency. To compare profiles we set all the
$I_j$
equal to the ones computed for the largest frequency in the range
$f\in$
[2.0, 5,5] Hz.
For
$a = 2$
cm: (a) symmetric propagation constant
$\beta _s$
(blue line) in between
$k_g$
(- - - -) and
$k_0$
(
$\cdots\cdots$
); (b)
$\psi _s(x)$
and
$\psi _a(x)$
for
$f=[2.5,3.5,4.5,5.5]$
Hz; (c) surface deformation Re
$(\hat {\zeta }_s(x, y))$
for
$f = 5.0$
Hz.

For
$a = 10$
cm: (a) first symmetric propagation constant
$\beta _s$
(blue line) and first anti-symmetric propagation constant
$\beta _a$
(red line) in between
$k_g$
(- - - -) and
$k_0$
(
$\cdots\cdots$
); (b)
$\psi _s(x)$
and (c)
$\psi _a(x)$
for
$f=[2.5,3.5,4.5,5.5]$
Hz; (d) surface deformation Re
$(\hat {\zeta }_s(x, y))$
; and (e) Re
$(\hat {\zeta }_a(x, y))$
for
$f = 5.0$
Hz.

For
$a = 30$
cm: (a) first symmetric propagation constants
$\beta _s$
(blue line) and first anti-symmetric propagation constants
$\beta _a$
(red line) in between
$k_g$
(- - - -) and
$k_0$
(
$\cdots\cdots$
); (b)
$\psi _s(x)$
and (c)
$\psi _a(x)$
for
$f=[2.5,5,5.5]$
Hz; (d) surface deformation Re
$(\hat {\zeta }_s(x, y))$
; and (e) Re
$(\hat {\zeta }_a(x, y))$
for
$f = 5.0$
Hz.

For small
$a$
only one mode (the fundamental one) develops in the frequency range
$f\in$
[2.5 Hz, 5.5 Hz], displaying large tails outside the wave-guide region as shown in figure 4. As
$a$
increases, other modes are allowed to propagate along the wave guide, as new solutions arise from (3.9a
)–(3.9b
).The second propagating mode is an anti-symmetric one, which is followed by a symmetric one as
$a$
increases. This tendency continues as the new propagating modes appear interleaved. Also, as
$a$
increases the localisation of the propagating modes over the wave-guide region also increases, as shown in figures 5 and 6. The same behaviour is observed as
$f$
increases, i.e. the number of allowed modes and the localisation of them increase as
$f$
increases for a fixed value of
$a$
.
3.1.2. Curved wave guides
The constant wave-guide case above changes when curvature is added to the problem. The configuration in mind is described schematically in figure 1. We will set polar coordinates
$\boldsymbol{r}_\perp = (r \cos \varphi , r\sin \varphi )$
, and write
$\hat {\zeta }(r, \varphi ) = \textrm{Re}(\psi (r)e^{i\gamma \varphi })$
. Here,
$\gamma$
will be thought of as the eigenvalue for the curved wave-guide problem, which can be a real (or even complex, i.e.
$\gamma =\textrm{Re}(\gamma )+i{Im}(\gamma )$
) number and not necessarily an integer. In this description, the operator
$\nabla _\perp ^2$
must be written in polar coordinates, which reduces (3.5) to a Bessel-type equation for
$\psi$
Notice that, as
$r\rightarrow 0$
, the solutions for (3.14) are strictly growing and asymptotically behave as
$r^\gamma$
. When
$(k_m^2r^2- \gamma ^2)$
changes sign, it gives rise to the characteristic oscillations displayed by the Bessel functions. Therefore, we impose
$k_0^2b^2- \gamma ^2 \lt 0$
, where
$b$
is the inner radius of the wave guide. This ensures that there are no oscillations in
$0 \lt r \lt b$
. On the other hand, it is impossible to prevent oscillations for
$r \gt c$
, with
$c$
being the outer radius of the wave guide, since there will always be some
$r \gt c$
such that
$k_0^2r^2- \gamma ^2 \gt 0$
, causing oscillations. Finally, since we want oscillations to occur inside the wave guide, we impose that
$k_g^2b^2- \gamma ^2 \gt 0$
. This leads to
$k_0 \lt \gamma /b\lt k_g$
, meaning that
$ \gamma /b$
plays a role similar to
$\beta$
in straight wave guides. An orthogonal pair of functions that solve (3.14) are
$J_\gamma (k_mr)$
and
$Y_\gamma (k_mr)$
(Bessel and Neumann functions, respectively). Since the latter diverges at
$r=0$
, the solution must be of the form
\begin{equation} \displaystyle \psi (r) = \left \{ \begin{array}{ll} AJ_\gamma (k_0r)&r \lt b, \\ \\ BJ_\gamma (k_gr) + CY_\gamma (k_gr) & b\lt r\lt c,\\ \\ DH_\gamma (k_0r)& r\gt c, \end{array} \right . \end{equation}
where,
$H_\gamma (x) = H^{(1)}_\gamma (x)=J_\gamma (k_mr)+i Y_\gamma (k_mr)$
is the Hankel function of the first kind.
Assuming mass and momentum continuity encoded in
$\psi (r)$
and
$h\partial _r \psi (r)$
at
$r = \{b, c\}$
, the problem of finding the propagation modes
$\gamma$
reduces to solving the eigenvalue problem given by
\begin{equation} \begin{bmatrix} J_\gamma (k_0b) & -J_\gamma (k_g b) & -Y_\gamma (k_g b) & 0 \\[6pt] k_0 J_\gamma ^{\prime}(k_0b) &- k_g J_\gamma ^{\prime}(k_g b) & -k_g Y_\gamma ^{\prime}(k_gb)& 0 \\[6pt]0& -J_\gamma (k_g c) & - Y_\gamma (k_g c) & H_\gamma (k_0 c)\\[6pt] 0 & - k_g J_\gamma ^{\prime}(k_g c) & -k_gY_\gamma ^{\prime}(k_gc) & k_0 H_\gamma ^{\prime}(k_0c) \end{bmatrix}\begin{bmatrix} A\\B\\C\\D \end{bmatrix} = \begin{bmatrix} 0\\0\\0\\0 \end{bmatrix}, \end{equation}
where
$\{J_\gamma ^{\prime}(x), Y^{\prime}_\gamma (x), H_\gamma ^{\prime}(x)\} = (d/\textrm{d}x)\{J_\gamma (x), Y_\gamma (x), H_\gamma (x)\}$
. For the previous matrix to be singular, the propagation modes
$\gamma$
must satisfy the equation
\begin{equation} \dfrac {k_gJ_\gamma ^{\prime} (k_gb)J_\gamma (k_0b)\! - \! k_0J_\gamma ^{\prime}(k_0b) J_\gamma (k_gb)}{k_0J_\gamma ^{\prime}(k_0b)Y_\gamma (k_gb)-k_g Y^{\prime}_\gamma (k_gb)J_\gamma (k_0b)} \!-\! \dfrac {k_gJ_\gamma ^{\prime}(k_gc)H_\gamma (k_0c)-k_0H^{\prime}_\gamma (k_0c)J_\gamma (k_gc)}{k_0H_\gamma ^{\prime}(k_0c)Y_\gamma (k_gc)\! -\! k_g Y^{\prime}_\gamma (k_gc)H_\gamma (k_0c)} = 0. \end{equation}
All special functions
$J_{\gamma }(z)$
,
$Y_{\gamma }(z)$
and
$H^{(1)}_{\gamma }(z)$
are evaluated with Mathematica (Wolfram Language 14.0). We solve (3.17) for the complex order
$\gamma$
using FindRoot, i.e. Newton’s method in the complex plane, implemented in Mathematica choosing the most stable representation at run time:
-
(i) power-series expansion for
$\lvert k_0 b\rvert \ll 1$
; -
(ii) hyperasymptotic/Debye series for
$\lvert k_0 b\rvert \gg 1$
; -
(iii) uniform Airy-type expansions when
$\lvert \gamma \rvert$
is large; -
(iv) Miller recurrences for high-order stability.
All branches and Stokes sectors are handled internally and the user may specify arbitrary precision; here, we requested WorkingPrecision = 80, guaranteeing
$\geqslant 60$
correct decimal digits.
The eigenvector associated with this solution has the form
$\boldsymbol{v} = [v_A, v_B, v_C, v_D]$
, where each component is respectively given by the expressions
\begin{align} v_A &= k_g[Y^{\prime}_\gamma (k_gb)J_\gamma (k_gb) - J^{\prime}_\gamma (k_gb)Y_\gamma (k_gb)]\nonumber\\&\quad\times[k_gY^{\prime}_\gamma (k_gc)H_\gamma (k_0c) - k_0H_\gamma ^{\prime}(k_0c)Y_\gamma (k_gc)], \end{align}
\begin{align} v_B & = [k_g Y^{\prime}_\gamma (k_gb)J_\gamma (k_0b) \!-\! k_0J^{\prime}_\gamma (k_0b)Y_\gamma (k_gb)]\nonumber\\&\quad\times[k_gY^{\prime}_\gamma (k_gc)H_\gamma (k_0c) \!-\! k_0H_\gamma ^{\prime}(k_0c)Y_\gamma (k_gc)], \end{align}
\begin{align} v_C &= [k_g Y^{\prime}_\gamma (k_gb)J_\gamma (k_0b)\!-\! k_0J^{\prime}_\gamma (k_0b)Y_\gamma (k_gb)]\nonumber\\&\quad\times[k_0H_\gamma ^{\prime}(k_0c)J_\gamma (k_gc)\!-\!k_gJ_\gamma ^{\prime}(k_gc)H_\gamma ^{\prime}(k_0c)], \end{align}
\begin{align} v_D &=k_g[k_g Y^{\prime}_\gamma (k_gb)J_\gamma (k_0b) - k_0J^{\prime}_\gamma (k_0b)Y_\gamma (k_gb)]\nonumber\\&\quad\times[Y^{\prime}_\gamma (k_gc)J_\gamma (k_gc)-J^{\prime}_\gamma (k_gc)Y_\gamma (k_gc)]. \end{align}
In this way, it follows that the profile
$\psi (r)$
is given by the expression
\begin{equation} \displaystyle \psi (r) = \left \{ \begin{array}{ll} \delta v_AJ_\gamma (k_0r)&r \lt b, \\ \\ \delta [v_BJ_\gamma (k_gr) + v_CY_\gamma (k_gr)] & b\lt r\lt c,\\ \\ \delta v_DY_\gamma (k_0r)& r\gt c, \end{array} \right . \end{equation}
where
$\delta$
corresponds to the amplitude of the perturbation. In order to compare with experimental results, the change of variables
$(x,y) = (r \cos \varphi , r \sin \varphi )$
is made, thus defining
$\zeta (x, y) = \psi (\sqrt {x^2+y^2})e^{i\gamma \arctan (y/x)}$
. Computed profiles following the prescription of (3.19) are depicted in figures 7–8 for
$f=3$
and
$f=5$
Hz, respectively.
Theoretical solutions of shallow water propagation over curved wave guides for
$f=3.5$
Hz for
$R=160, 80, 40, 24$
cm. Stroboscopic (a–d) wave fields and their corresponding envelope (e–h) display both wave guiding and wave leaking along the wave guide. Plots for
$\hat {\zeta }(s)$
(i–l) show how the wave amplitude decays along the centre line of the wave guide. The transverse profile of the wave (m–p) is depicted to showcase how wave energy is radiated outside of the wave guide.

Theoretical solutions of shallow water propagation over curved wave guides for
$f=5.0$
Hz for
$R=160, 80, 40, 24$
. Stroboscopic (a–d) wave fields and their corresponding envelope (e–h) display both wave guiding and wave leaking along the wave guide. Plots for
$\hat {\zeta }(s)$
(i–l) show how the wave amplitude decays along the centre line of the wave guide. The transverse profile of the wave (m–p) is depicted to showcase how wave energy is radiated outside of the wave guide.

It is important to note that curvature induces a region of parameters for which there is no guided propagation; this occurs at low frequencies and low curvature radii. However, similar to straight wave guides, it holds that for a given width
$c-b=a$
(following the straight wave-guide construction), there exists a frequency
$f\gt 0$
for which there are
$N$
propagation modes. Regarding the perturbation profiles
$\psi (r)$
, it is observed that when
$b$
is small with respect to
$\lambda$
, there are oscillations outside the wave guide comparable to the perturbation that exists within, that is, a large amount of radiation (cf. figures 7–8, panels (a)–(h)). This radiation decreases as the curvature radius increases and becomes negligible when the width of the wave guide is increased. It is also relevant to see that the modes lack symmetry with respect to the centre line of the wave guide (cf. figures 7–8, panels (m)–(p)). Furthermore, as the curvature radius and/or the width of the wave guide increases, these guided modes become increasingly symmetric, as the wave guide in question becomes more similar to a straight one.
Experimental data of shallow water propagation over curved wave guides for
$f=3.5$
Hz for
$R = 160, 80, 40$
and
$24$
cm. Stroboscopic (a–d) wave fields and their corresponding envelope (e–h) display both wave guiding and wave leaking along the wave guide. Plots for
$\hat {\zeta }(s)$
(i–l) show how the wave amplitude decays along the centre line of the wave guide. The amplitude of the transverse profile (m–p) is depicted to showcase how wave energy is radiated outside of the wave guide.

Experimental data of shallow water propagation over curved wave guides for
$f=5.0$
Hz for
$R= 160, 80, 40$
and
$2$
4 cm. Stroboscopic (a–d) wave fields and their corresponding envelope (e–h) display both wave guiding and wave leaking along the wave guide. Plots for
$\hat {\zeta }(s)$
(i–l) show how the wave amplitude decays along the centre line of the wave guide. The amplitude of the transverse profile (m–p) is depicted to showcase how wave energy is radiated outside of the wave guide.

4. Experimental results and analysis for curved wave guides
With the theoretical basis described above for both straight and curved wave guiding within the shallow water regime, we now show the experimental data for wave propagation on curved wave guides
As explained above, experiments on shallow water waveguiding over straight and curved wave guides of width
$a = 2.0$
cm in the frequency range
$f \in$
[3.0, 5.0] Hz were performed. We compute the stroboscopic wave profiles
$\zeta (r=r_c)e^{i\gamma \varphi }$
measured along the centre line of wave guide
$r_c=(b+c)/2$
for different values of
$R$
, including the straight wave guide
$R\to \infty$
. Thus, one can follow the arc length coordinate
$s=R\varphi$
which starts at the beginning of the wave guide and follows the centre line of the wave guide. From this profile, its envelope
$|\hat {\zeta }(s)|$
is computed via a Hilbert transform which is shown to decay along
$s$
. An exponential fit
$|\hat {\zeta }(s)| \sim \exp {( -s/\Lambda )}$
is used to access the wave decay rate
$\Lambda =\Lambda (R)$
. The decay rate for the straight wave guide
$\Lambda =\Lambda (R\to \infty )$
is computed to separate the effect of viscosity from the curvature ones from
$\Lambda (R)$
.
The transverse profiles of the guided waves along curved wave guides are also computed from the two-dimensional envelope
$|\hat {\zeta }(r,\varphi )|$
, and compared between them as a function of curvature. Transverse cuts along
$\varphi _o=10^{\circ }$
of
$|\hat {\zeta }(r,\varphi _o)|$
as a function of
$r$
are shown in figures 9 and 10 (m–p). The experimental behaviour of the curved wave-guiding process shown in these figures is in good agreement with the theoretical profiles shown in figures 7 and 8, which are based on the framework presented above. To qualitatively compare the experimentally observed behaviour along the curved wave guides with the theoretical description proposed in § 3, theoretical solutions were compared with the radial part of the experimental results at different angles, normalising the amplitude for each one. A comparison is depicted in figure 11 where experimental and theoretical radial profiles are plotted for different frequencies, curvature radii and angles. For small angles and large curvature radii the theoretical solutions are in good agreement with the experimentally measured profile, while for larger angles and/or shorter radii the agreement is worse as the width of the transverse mode increases. This disagreement can be related to dissipative effects which shift slightly the wavelength along the wave guide, allowing oscillations and thus leakage outside the wave guide, which are observed experimentally. This strengthens the notion that the curvature of the wave guide plays a paramount role on the decay rate and thus the radiation process outside the wave guide. It must be pointed out that the observed wavelength of guided waves is, at most, one order of magnitude smaller than the curvature radius (for
$f=$
5 Hz and
$R=160$
cm), which means that curvature effects on the damping and radiation of waves are dominant in this regime.
Comparison between the radial component of the theoretical solution for the wave-guide modes and experimental results for different angles, which have been previously normalised. Panels (a), (b) and (c) correspond to 3.5 Hz and (d),(e) and (f) correspond to 5.0 Hz. Panels (a) and (d) correspond to
$R$
= 80 cm, (b) and (e) correspond to
$R$
= 40 cm and (c) and (f) correspond to
$R$
= 24 cm.

(a) Values of
$\textrm{Re}(\zeta (s))$
(continuous line) and
$|\zeta (s)|$
(dashed line) for
$f=$
5.0 Hz and
$R=80$
cm. An exponential fit (dotted line) is also depicted. From these curves the decay rate and wavelength are computed and used to construct
$\alpha$
. (b) Semilogarithmic plot of
$ \alpha /\alpha _0 - 1$
as function of frequency for
$R$
= 24 (
$\circ$
), 40 (
$\triangle$
), 80 (
$\square$
) and 160 (
$\triangle$
) cm. Inset:
$\alpha _0$
(
$\diamond$
) as a function of frequency.

Curvature effects on wave guiding are taken into account by comparing the wavelength
$\lambda$
along the centre line of the wave guide and the decay length
$\Lambda (R)$
via
$\alpha = \lambda (R)/\Lambda (R)$
(see figure 12
a). To disentangle viscous effects from curvature ones, we compute
$\alpha _o=\lambda (R\to \infty )/\Lambda (R\to \infty )$
along a straight wave guide, which will be subtracted from the experimentally measured
$\alpha$
. Thus, the quantity
$(\alpha - \alpha _{0})$
will serve as a proxy of the curvature-induced wave dissipation. We show in figure 12(b) the normalised ratio
$(\alpha -\alpha _0)/\alpha _0$
as a function of frequency for different values of
$R$
, which gives us an understanding of the importance of curvature effects in relation to viscous effects. On the one hand, for a given
$R$
, we can observe that increasing
$f$
decreases
$(\alpha -\alpha _0)/\alpha _0$
, which means that
$\Lambda (R)$
increases and thus the decay rate is larger for smaller wavelengths. On the other hand, for a given
$f$
, decreasing
$R$
increases
$(\alpha -\alpha _0)/\alpha _0$
, which means that
$\Lambda (R)$
decreases and thus the decay rate is larger for larger curvature radii. The exhibited behaviour is in good agreement with the solutions of figures 7 and 8, which shows for a fixed frequency an increasing decay rate with increasing radii, until the wave cannot really be guided for small
$R$
. To enforce this fact we show in figure 13 the ratio between the experimentally computed
$\alpha -\alpha _0$
and the computed decay rate
$\alpha _t=\text{Im}(\gamma )$
from the solutions of (3.14) for the different wavelengths used in our experiments as a function of
$R$
. The data show that for different frequencies both decay rates are roughly equal (close to 1), except for very small
$R$
, where experimental decay rates are much larger than the theoretical ones.
Experimental to theoretical decay rate ratio
$(\alpha -\alpha _0)/\alpha _t$
as a function of
$R$
for
$f=$
3.0 Hz (
$\circ$
) and
$f=$
5.0 Hz (
$\square$
).

5. Discussion and perspectives
In this work, we have shown experimentally and theoretically the mechanics of wave guiding for shallow water waves over curved wave guides in the limit where the curvature radius is similar to the propagating wavelength. Using a two-dimensional profilometry technique, we have accessed the spatio-temporal surface deformations and compared them with the shallow water wave propagation over a curved wave guide constructed via a step-like bathymetry jump in the inviscid limit. Theoretical solutions of the inviscid shallow water wave problem display remarkable similarities to the experimental data. To wit, quantitative similarities are observed: i) between theoretical (cf. figures 7 and 8) and experimental (cf. figures 9 and 10) wavenumbers along the centre line of curved wave guides, ii) between experimental and theoretical guided transverse profiles for small
$\varphi _0$
regardless of curvature radii or forcing frequency (cf. figure 11) and iii) between the experimentally measured decay rate (compensated by the viscous wave damping along straight wave guides)
$\alpha -\alpha _0$
and the theoretically computed imaginary part of the eigenvalue
$\alpha _t=\text{Im}(\gamma )$
for large curvature radii (cf. figure 13). Furthermore, numerically computed profiles from (3.19) and the corresponding experimental data show that the water wave guide for finite curvature radii shows consistent wave leakage, which is quantified by the wave damping coefficient
$\alpha$
along the centre line of the curved wave guide. Compensating the viscous wave damping along straight wave guides encoded in
$\alpha _0$
(Sepulveda et al. Reference Sepúlveda-Soto, Guzmán-Silva, Rosas, Vicencio and Falcón2020), we show that as the wavelength increases (forcing frequency
$f$
decreases)
$\alpha$
grows monotonically. Although quantitative similarities between experimental and theoretical results are observed, certain qualitative differences appear systematically. For instance, experimental transverse wave profiles (cf. figures 9, 10 and 11) display radial oscillations that are not displayed by their theoretical counterparts as they propagate along the curved wave guide. In the same trend, theoretical decay rates for small curvature radii (with respect to the wavelength) (cf. figure 13) are found to be an order of magnitude smaller than the experimentally measured one (even after compensation with respect to the viscous dissipation along a straight wave guide). The nature and origin of these discrepancies, and their relation to wave leakage, will be the focus of future work.
Colour maps of
$\textrm{Re}(\hat {\zeta }(x,y))$
for (a)
$f = 3.0$
Hz and (b)
$f=5.0$
Hz of a complex wave guide constructed by two closely positioned straight wave guides. The solid lines over the colour maps correspond to the borders of the step-like wave-guide profile positioned on the bottom. In both cases the upper wave guide was the only one excited.

Stroboscopic (a–d) and envelope (e–h) wave fields for
$f=3.5$
Hz (a),(b),(e),(f) and
$f=5.0$
Hz (c),(d),(g),(h). For each frequency, the comparison between different interferometers can be observed.

Behind the idea of a single curved wave guide with a single curvature radius
$R$
lies the idea of more complex structures to transport wave energy along a given path. Single water wave guides can be used to construct compound wave-guiding structures which display interesting wave phenomena. For example, two straight wave guides can be coupled via the transverse profiles of their localised guided modes, generating oscillations of the wave field from one wave guide to another, as shown in figure 14. Due to the small separation between both wave guides, the interaction appears almost immediately. The wave energy migrates from the excited wave guide to the other, which is observed for two different frequencies. Thus, as the wave amplitude on the initially excited wave guide decreases, the amplitude on the other wave guide increases. This spatio-temporal dynamics must be taken into account if multimodal or coupled single wave guides are to be used to control and localise surface water waves in the shallow water limit. The dynamics of coupled wave guides is crucial when considering more complex curved wave-guiding structures. This can be showcased via surface water wave interferometry, which can be used to localise and control wave energy via the phase mismatch of two (or more) interfering waves. To wit, we conducted experiments where we measure the interaction of two waves propagating along curved wave guides, such as the ones described in § 4 with the same width and height, but different curvature radii, which converge to a single straight wave guide. Both wave guides are excited at the same frequency with no phase mismatch, creating an interferometer for shallow water waves as both the left and right ‘arms’ of the interferometer differ in their lengths.
These interferometers were built using curved wave guides of width
$a=2.0$
cm and height
$h=1.5$
cm and different curvature radii. The first one used a left arm with curvature radius
$R=80$
cm and a right arm of
$R=160$
cm converging to a single straight wave guide with the same width and height (cf. figure 15). A second interferometer has a left arm of
$R=40$
cm and a right arm of
$R=160$
cm, converging again to a straight wave guide with the same width and height (cf. figure 15). Due to the change in
$R$
for both arms, the phase difference of both propagating waves along the curved arms allows them to interfere. More interestingly, one can observe that the excited waves propagating along the curved arms of both interferometers interact between themselves, similarly to the process shown in figure 14.
Funding
This work was supported in part by Programa ICM Millenium Nucleus for Soft Smart Mechanical Metamaterials and Millennium Institute for Research in Optics (MIRO), and FONDECYT Grants Nos. 1190005, 1191205 and 3190601.
Declaration of interests
The authors report no conflict of interest.
Data availability statement
The data that support the findings of this study are openly available upon request to C.F.
Author contributions
F.S.-S. and C.F. performed the experiments, E.R. and F.S.-S. computed the analytical solutions, C.F. conceived the experiments and analysed the results and E.R. performed the analytical calculations for mode analysis. All authors reviewed the manuscript.
























































































