1 Introduction
Michael Faraday (Reference Faraday1831) gave a first experimental description of the instability that would be named after him. The Faraday instability occurs when a container filled with liquid is vertically vibrated with an acceleration
$\unicode[STIX]{x1D6E4}g\cos (\unicode[STIX]{x1D6FA}t)$
, where
$g$
is the gravitational acceleration and
$\unicode[STIX]{x1D6FA}/2\unicode[STIX]{x1D70B}$
is the shaking frequency. When the dimensionless amplitude
$\unicode[STIX]{x1D6E4}$
is higher than a threshold
$\unicode[STIX]{x1D6E4}_{F}$
, the forcing is sufficient to counterbalance viscous dissipation (Kumar & Tuckerman Reference Kumar and Tuckerman1994), the initially flat free surface destabilises and beautiful patterns of standing waves spontaneously appear. The resonance is often subharmonic so the frequency of the standing waves is
$\unicode[STIX]{x1D6FA}/4\unicode[STIX]{x03C0}$
. The Faraday wavelength
$\unicode[STIX]{x1D706}_{F}$
, defined as the characteristic wavelength of the standing waves, is usually close to the wavelength prescribed by the dispersion relation of gravity–capillary waves with a frequency
$\unicode[STIX]{x1D6FA}/4\unicode[STIX]{x03C0}$
. In its original and now most studied configuration (Benjamin & Ursell Reference Benjamin and Ursell1954), the depth and lateral extension of the liquid bath are respectively much deeper and much larger than the wavelength, which ensures that the resonance is subharmonic.
Below the Faraday instability threshold, all surface waves are damped and, without any perturbation, the liquid surface remains flat. A perturbation can be added in the form of a bouncing droplet. If the vertical vibration of the bath is of sufficient amplitude, this droplet can bounce on the bath periodically and indefinitely (Couder et al.
Reference Couder, Fort, Gautier and Boudaoud2005a
; Gilet et al.
Reference Gilet, Terwagne, Vandewalle and Dorbolo2008; Molacek & Bush Reference Molacek and Bush2013). Couder et al. (Reference Couder, Protiere, Fort and Boudaoud2005b
) showed that such droplets may excite Faraday waves, especially if they can bounce once every two forcing periods (i.e. at frequency
$\unicode[STIX]{x1D6FA}/4\unicode[STIX]{x03C0}$
). If the forcing amplitude is slightly below the instability threshold (
$\unicode[STIX]{x1D6E4}\lesssim \unicode[STIX]{x1D6E4}_{F}$
), these waves are slowly damped and they are still present when the droplet impacts again (Eddi et al.
Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011). Consequently, the droplet impacts on a non-flat surface and some horizontal momentum can be transferred from the wave to the droplet. The latter starts to ‘walk’ on the vibrated liquid surface. The combination of a walking droplet and the underlying surface waves is called a walker. A typical walker involves a droplet of radius
$R_{d}=0.4~\text{mm}$
and a bath of depth 6 mm, both made of silicone oil of density
$\unicode[STIX]{x1D70C}=956~\text{kg}~\text{m}^{-3}$
, kinematic viscosity
$\unicode[STIX]{x1D708}=20~\text{cS}$
and surface tension
$\unicode[STIX]{x1D70E}=20.6~\text{mN}~\text{m}^{-1}$
. The bath is typically shaken at a frequency
$\unicode[STIX]{x1D6FA}/(2\unicode[STIX]{x03C0})=80~\text{Hz}$
with an amplitude
$\unicode[STIX]{x1D6E4}\sim 4$
(table 1).
Table of notations used in the paper. Subscript
$_{0}$
(respectively
$_{F}$
and
$_{i}$
) refers to the solution for a null viscosity (respectively to the solution at the Faraday threshold and to the quantities relative to the impact).

Walkers form a unique macroscopic analogue of the pilot-wave theory imagined by de Broglie (Reference de Broglie1987) as an interpretation of quantum mechanics. In a series of configurations recently summarised by Bush (Reference Bush2015), the behaviour of walkers is strongly reminiscent of that of quantum particles. Quantisation of bound states is observed in a central force field (Perrard et al. Reference Perrard, Labousse, Fort and Couder2014a ,Reference Perrard, Labousse, Miskin, Fort and Couder b ; Labousse et al. Reference Labousse, Perrard, Couder and Fort2014, Reference Labousse, Oza, Perrard and Bush2016; Tambasco et al. Reference Tambasco, Harris, Oza, Rosales and Bush2016), in a rotating frame (Fort et al. Reference Fort, Eddi, Boudaoud, Moukhtar and Couder2010; Eddi et al. Reference Eddi, Moukhtar, Perrard, Fort and Couder2012; Harris & Bush Reference Harris and Bush2014; Oza et al. Reference Oza, Harris, Rosales and Bush2014a ,Reference Oza, Wind-Willassen, Harris, Rosales and Bush b ), or with two-body interactions (Protiere, Boudaoud & Couder Reference Protiere, Boudaoud and Couder2006; Protière, Bohn & Couder Reference Protière, Bohn and Couder2008; Borghesi et al. Reference Borghesi, Moukhtar, Labousse, Eddi, Fort and Couder2014; Filoux, Hubert & Vandewalle Reference Filoux, Hubert and Vandewalle2015; Durey & Milewski Reference Durey and Milewski2017; Oza et al. Reference Oza, Siéfert, Harris, Molác̆ek and Bush2017). Walkers interact with submerged boundaries through non-specular reflection (Pucci et al. Reference Pucci, Sáenz, Faria and Bush2016) and tunnelling (Eddi et al. Reference Eddi, Fort, Moisy and Couder2009; Carmigniani et al. Reference Carmigniani, Lapointe, Symon and McKeon2014; Nachbin, Milewski & Bush Reference Nachbin, Milewski and Bush2017). More remarkably, when they are confined in a cavity, they experience chaotic trajectories for which the corresponding statistics is close to the quantum statistics predicted by the Schrödinger equation (Harris et al. Reference Harris, Moukhtar, Fort, Couder and Bush2013; Gilet Reference Gilet2014, Reference Gilet2016). They also seem to be diffracted when they pass one by one above a submerged slit (Couder & Fort Reference Couder and Fort2006, Reference Couder and Fort2012; Dubertrand et al. Reference Dubertrand, Hubert, Schlagheck, Vandewalle, Bastin and Martin2016), although it is not yet clear that the underlying statistics is really similar to that of a diffracted quantum particle (Andersen et al. Reference Andersen, Madsen, Reichelt, Ahl, Lautrup, Ellegaard, Levinsen and Bohr2015). Walkers therefore represent a unique opportunity to probe the boundary between the macroscopic world and the quantum realm. They allow us to identify which specific behaviours of quantum particles could be explained or not through the deterministic coupling of a wave and a particle.
An appropriate model of the damped Faraday waves emitted by the bouncing droplet is necessary to rationalise the dynamics of walkers. The first model has been proposed by Benjamin & Ursell (Reference Benjamin and Ursell1954) for an inviscid fluid. Since there is no dissipation, the instability threshold is then identically zero. The Navier–Stokes equations simplify into a Matthieu equation for the interface elevation
$\unicode[STIX]{x1D701}(x,y,t)$
. In the limit of infinite depth, it reads
where
$k$
is the wavenumber. There is a subharmonic resonance for the Faraday wavenumber
$k_{F}=2\unicode[STIX]{x03C0}/\unicode[STIX]{x1D706}_{F}$
such that:
In order to account for a finite viscosity
$\unicode[STIX]{x1D708}$
, a phenomenological linear damping term is added to this Matthieu equation (e.g. Kumar & Tuckerman (Reference Kumar and Tuckerman1994), Eddi et al. (Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011), Moláček & Bush (Reference Moláček and Bush2013), Blanchette (Reference Blanchette2016)):
where
$\unicode[STIX]{x1D6FC}$
is a dimensionless factor. A slightly different model is derived by Milewski et al. (Reference Milewski, Galeano-Rios, Nachbin and Bush2015) from the Navier–Stokes equations in the limit of small viscosity, without any additional phenomenological term (Durey & Milewski Reference Durey and Milewski2017):
The wave field generated by one impact of the bouncing droplet at position
$\boldsymbol{r}_{n}$
at time
$t_{n}$
is then approximately (Moláček & Bush Reference Moláček and Bush2013; Milewski et al.
Reference Milewski, Galeano-Rios, Nachbin and Bush2015)
where
$Me=\unicode[STIX]{x1D6FA}\unicode[STIX]{x1D6E4}_{F}/[(8\unicode[STIX]{x03C0}\unicode[STIX]{x1D708}k_{F}^{2})(\unicode[STIX]{x1D6E4}_{F}-\unicode[STIX]{x1D6E4})]$
is defined as the memory parameter,
$A$
is a constant amplitude that represents the strength of the impact and
$\unicode[STIX]{x1D6F7}$
is linked (but not strictly equal) to the impact phase relative to the shaker. In the presence of obstacles, this equation (1.5) can be replaced by a more elaborate Green’s function that takes the wave reflection on boundaries into account (Dubertrand et al.
Reference Dubertrand, Hubert, Schlagheck, Vandewalle, Bastin and Martin2016). It has also been suggested in Fort et al. (Reference Fort, Eddi, Boudaoud, Moukhtar and Couder2010), Eddi et al. (Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011) to add a spatial damping factor
$\text{e}^{-|\boldsymbol{r}-\boldsymbol{r}_{n}|/l}$
, where
$l\sim 1.6\unicode[STIX]{x1D706}_{F}$
is a characteristic damping length.
The wave field generated by a walker is assumed linear, so it can be computed by summing wave contributions from successive impacts:
A walker that does not experience any interaction with either boundaries or other walkers, moves at constant horizontal speed and impacts the bath at regular times
$t_{n}=t_{i}+4\unicode[STIX]{x03C0}n/\unicode[STIX]{x1D6FA}$
. This is referred as the ‘stroboscopic approximation’ and it is implemented in all currently existing models of walkers, with the recent exception of that of Oza et al. (Reference Oza, Siéfert, Harris, Molác̆ek and Bush2017). Some models assume an instantaneous interaction of the droplet with the bath for the sake of simplicity (Eddi et al.
Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011; Gilet Reference Gilet2016; Durey & Milewski Reference Durey and Milewski2017), whereas more sophisticated models take into account a finite time of interaction of the droplet with the bath at each impact (Moláček & Bush Reference Moláček and Bush2013; Milewski et al.
Reference Milewski, Galeano-Rios, Nachbin and Bush2015). When the ‘stroboscopic approximation’ is combined with the additional assumption that the horizontal translation of the walker during one rebound is much less than the Faraday wavelength
$\unicode[STIX]{x1D706}_{F}$
, the sum of impacts can be replaced by a time integral, which yields an integro-differential equation for the walker trajectory (Oza, Rosales & Bush Reference Oza, Rosales and Bush2013; Bush, Oza & Moláček Reference Bush, Oza and Moláček2014).
There are several issues with (1.5) and its variants. First, the elevation
$\unicode[STIX]{x1D701}$
of waves created by a single impact diverges in
$t\rightarrow t_{n}$
, which is unphysical. Second, the viscosity is taken into account through the damping factor
$\unicode[STIX]{x1D6FC}$
and the spatial damping length
$l$
, both of which are empirical corrections assumed to be valid only when viscous damping is weak. However, for the typical experimental conditions described above, the damping time
$1/(4\unicode[STIX]{x1D708}k_{F}^{2})\sim 0.007~\text{s}$
is actually smaller than the Faraday period
$4\unicode[STIX]{x03C0}/\unicode[STIX]{x1D6FA}=0.025~\text{s}$
. So viscous damping is definitely not weak, and one needs to check how much error is made by using these low-viscosity empirical approximations. Third, Faraday waves are assumed in phase with the forcing signal, and of constant amplitude
$A$
, without any proof or experimental evidence. The dependence of
$A$
to the phase shift between the impact and the forcing signal has never been evidenced experimentally.
Without perturbations, a walker moves horizontally along a straight line, with a perfectly periodic bouncing motion. Therefore, these three limitations of (1.5) may induce discrepancies between models and experiments as soon as the walker’s bouncing is perturbed, either by the presence of boundaries, other walkers or waves that it left previously. Indeed, there is then no guarantee that the bouncing motion will remain periodic, and any perturbation of the impact phase would induce a different weight of each rebound to the wave field. This weight is also strongly dependent on temporal and spatial damping factors. The latter basically defines the distance below which a significant interaction between multiple walkers is expected. A wave field constructed with different weights for each impact will necessarily be qualitatively different, and so will the resulting walker trajectory be. The aforementioned perturbations by boundaries, other walkers and self-interferences, all correspond to configurations where quantum-like phenomena were observed. They seem to be key to the analogy between walkers and quantum particles.
The main aim of this article is to obtain a quantitatively more accurate understanding of the surface waves emitted by walkers. The Faraday instability has been the subject of much theoretical work since Benjamin & Ursell (Reference Benjamin and Ursell1954). Most notably, Kumar & Tuckerman (Reference Kumar and Tuckerman1994) solved the full linearised Navier–Stokes equations (without any assumption on viscosity or depth) with a spectral method based on Floquet’s theory. They showed, by comparing curves of marginal stability, that the damped Matthieu equation (1.3) misses an important additional dissipation in the boundary layer at the liquid/air interface that only the full model can capture. However, because of the parametric forcing, the harmonics of the Floquet series are coupled to each other. Consequently, the dispersion relation can only be obtained by solving a discouraging eigenvalue problem for a matrix of infinite dimension.
The truncation of the Floquet series of Kumar & Tuckerman (Reference Kumar and Tuckerman1994) to four harmonics (
$\pm \unicode[STIX]{x1D6FA}/2$
and
$\pm 3\unicode[STIX]{x1D6FA}/2$
) has been investigated by Kumar (Reference Kumar1996). With this truncated model, they could show that right above the threshold (
$\unicode[STIX]{x1D6E4}\gtrsim \unicode[STIX]{x1D6E4}_{F}$
), Faraday waves are always subharmonic when both the depth
$h$
and the Faraday wavelength
$\unicode[STIX]{x1D706}_{F}$
are much larger than the size of the boundary layer
$\sqrt{2\unicode[STIX]{x1D708}/\unicode[STIX]{x1D6FA}}$
. This is the case for walkers since
$\sqrt{2\unicode[STIX]{x1D708}/\unicode[STIX]{x1D6FA}}\simeq 300~\unicode[STIX]{x03BC}\text{m}\ll h\simeq 6~\text{mm}$
and
$\unicode[STIX]{x1D706}_{F}\simeq 4.7~\text{mm}$
. Special wave patterns, including solitons, appear at the bicritical point where the boundary layer thickness is of the same order of magnitude as either the depth
$h$
or the wavelength
$\unicode[STIX]{x1D706}_{F}$
(Lioubashevski, Arbell & Fineberg Reference Lioubashevski, Arbell and Fineberg1996; Wagner, Müller & Knorr Reference Wagner, Müller and Knorr2003). An extreme truncation to only the subharmonic response
$\pm \unicode[STIX]{x1D6FA}/2$
has been performed by Beyer & Friedrich (Reference Beyer and Friedrich1995) as well as Müller et al. (Reference Müller, Wittmer, Wagner, Albers and Knorr1997). Müller et al. (Reference Müller, Wittmer, Wagner, Albers and Knorr1997) could then derive an analytical expression for the Faraday threshold
$\unicode[STIX]{x1D6E4}_{F}$
in the limit of low viscosity (i.e.
$\unicode[STIX]{x1D708}k^{2}/\unicode[STIX]{x1D6FA}\ll 1$
) and large depth.
In this paper, we follow the approach of Müller et al. (Reference Müller, Wittmer, Wagner, Albers and Knorr1997) to describe the Faraday waves emitted by walkers. In § 2, we first derive an analytical model of slightly damped Faraday waves, while keeping control on the approximations that are made (e.g. the approximation of low viscosity). We then calculate the wave field generated by a series of impacts. In § 3, we report the results of experiments on Faraday waves excited by an impacting bead. We show that our model goes beyond the identified limitations of (1.5) currently used to describe walkers. In § 4, we finally discuss the implication of this wave model for the development of the analogy between walkers and quantum particles.
2 Surface wave theory for walkers
2.1 The surface wave equation
We consider an infinitely extended and infinitely deep Newtonian liquid with density
$\unicode[STIX]{x1D70C}$
, surface tension
$\unicode[STIX]{x1D70E}$
and viscosity
$\unicode[STIX]{x1D708}$
(figure 1). This liquid is incompressible, i.e.
$\unicode[STIX]{x1D735}\boldsymbol{\cdot }\boldsymbol{v}(\boldsymbol{r},t)=0$
, where
$\boldsymbol{v}$
is the velocity field within the liquid. It is subjected to a periodic vertical vibration of frequency
$\unicode[STIX]{x1D6FA}$
and acceleration amplitude
$\unicode[STIX]{x1D6E4}g$
, where
$g=9.81~\text{m}~\text{s}^{-2}$
is the gravitational acceleration. Within the comoving frame, the liquid is then effectively experiencing a time-periodic modulation of gravity according to
A liquid bath of infinite depth is vibrated vertically. The liquid experiences a time varying effective gravity
$g(t)$
. The elevation of the liquid surface is denoted
$\unicode[STIX]{x1D701}(x,y,t)=\unicode[STIX]{x1D701}(\boldsymbol{r},t)$
. The velocity and pressure fields inside the liquid bath are
$\boldsymbol{v}(\boldsymbol{r},t)$
and
$P(\boldsymbol{r},t)$
, respectively.

The spatio-temporal evolution of the liquid is governed by the Navier–Stokes equation,
where
$\boldsymbol{r}=[x,y,z]$
and
$P(\boldsymbol{r},t)$
denotes the pressure within the liquid relative to the atmospheric pressure. Without loss of generality, we choose a coordinate system whose origin is located at the surface of the liquid at rest, with the
$z$
-axis oriented upwards (i.e. opposite to the direction of the gravitational force). We consider small deviations from the stationary equilibrium state
$\boldsymbol{v}^{(eq)}(\boldsymbol{r},t)=0$
and
$P^{(eq)}(\boldsymbol{r},t)=-\unicode[STIX]{x1D70C}g(t)z$
of the liquid. Under the assumption that the amplitude of the resulting elevation
$\unicode[STIX]{x1D701}$
of the free surface is much smaller than both the thickness of the viscous boundary layer
$\sqrt{2\unicode[STIX]{x1D708}/\unicode[STIX]{x1D6FA}}$
and the Faraday wavelength
$\unicode[STIX]{x1D706}$
, we may linearise the Navier–Stokes equation (2.2):
where
$p(\boldsymbol{r},t)=P(\boldsymbol{r},t)+\unicode[STIX]{x1D70C}g(t)z$
denotes the pressure perturbation. As discussed in appendix F, the conditions of linearisation are satisfied for experimental realisations of walkers. Indeed, the measured wave amplitude is smaller than
$12~\unicode[STIX]{x03BC}\text{m}$
(Damiano et al.
Reference Damiano, Brun, Harris, Galeano-Rios and Bush2016), which is at least one order of magnitude smaller than both the thickness of the viscous boundary layer (
${\sim}300~\unicode[STIX]{x03BC}\text{m}$
) and the Faraday wavelength (
${\sim}4.7~\text{mm}$
). Owing to the incompressibility of the liquid,
$p$
satisfies the Laplace equation
$\unicode[STIX]{x1D6FB}^{2}p(\boldsymbol{r},t)=0$
.
The boundary conditions at the surface
$z=0$
of the liquid are also obtained under the assumption that the liquid is only weakly perturbed: the wave amplitude is assumed much smaller than both the characteristic wavelength and the boundary layer thickness. The absence of tangential shear stress at the surface can then be expressed as
with
$u=x$
or
$y$
, which together with the incompressibility condition yields the boundary condition
for the vertical component of the liquid velocity. The pressure, on the other hand, satisfies the boundary condition
at the surface of the liquid. Here
$\unicode[STIX]{x1D701}(x,y,t)$
denotes the surface wave profile, i.e. the local elevation of the surface with respect to the equilibrium level
$z=0$
. It satisfies the kinematic condition
The pressure
$P^{ext}(x,y,t)$
represents the spatio-temporal pressure profile that is induced from above the liquid, e.g. by the impact of a droplet on the surface (Milewski et al.
Reference Milewski, Galeano-Rios, Nachbin and Bush2015). This gives rise to the boundary condition
$$\begin{eqnarray}\displaystyle p(x,y,0,t) & = & \displaystyle 2\unicode[STIX]{x1D708}\unicode[STIX]{x1D70C}\frac{\unicode[STIX]{x2202}v_{z}}{\unicode[STIX]{x2202}z}(x,y,0,t)-\unicode[STIX]{x1D70E}\left(\frac{\unicode[STIX]{x2202}^{2}}{\unicode[STIX]{x2202}x^{2}}+\frac{\unicode[STIX]{x2202}^{2}}{\unicode[STIX]{x2202}y^{2}}\right)\unicode[STIX]{x1D701}(x,y,t)+\unicode[STIX]{x1D70C}g(t)\unicode[STIX]{x1D701}(x,y,t)\nonumber\\ \displaystyle & & \displaystyle +\,P^{ext}(x,y,t)\end{eqnarray}$$
for the local pressure perturbation.
Since the lateral extension of the bath is here supposed infinite, the above equations can be simplified by means of a Fourier ansatz defined through
where
$X$
is any of the variables
$\unicode[STIX]{x1D701}$
,
$p$
,
$v_{z}$
or
$P^{\text{ext}}$
, and where the horizontal wave vector
$\boldsymbol{k}=(k_{x},k_{y})$
. Defining
$k=|\boldsymbol{k}|$
, the Laplace equation for the pressure perturbation is then expressed as
This equation is straightforwardly solved as
$p_{\boldsymbol{k}}(z,t)=p_{\boldsymbol{k}}(0,t)\text{e}^{kz}$
using the boundary conditions
The Fourier transform of the linearised Navier–Stokes equation (2.3) then yields
$$\begin{eqnarray}\displaystyle \left(\frac{\unicode[STIX]{x2202}}{\unicode[STIX]{x2202}t}-\unicode[STIX]{x1D708}\frac{\unicode[STIX]{x2202}^{2}}{\unicode[STIX]{x2202}z^{2}}+\unicode[STIX]{x1D708}k^{2}\right)v_{z,\boldsymbol{k}}(z,t) & = & \displaystyle -\!\left\{2\unicode[STIX]{x1D708}\frac{\unicode[STIX]{x2202}v_{z,\boldsymbol{k}}}{\unicode[STIX]{x2202}z}(0,t)+\left[g(t)+\frac{\unicode[STIX]{x1D70E}}{\unicode[STIX]{x1D70C}}k^{2}\right]\unicode[STIX]{x1D701}_{\boldsymbol{k}}(t)\right\}k\text{e}^{kz}\nonumber\\ \displaystyle & & \displaystyle -\,\frac{k}{\unicode[STIX]{x1D70C}}P_{\boldsymbol{k}}^{ext}(t)\text{e}^{kz},\end{eqnarray}$$
with boundary conditions
and
We now normalise the main variables with the characteristic time scale
$2/\unicode[STIX]{x1D6FA}$
, and we define:
With this set of notations, equations (2.12)–(2.14) become,
$$\begin{eqnarray}\displaystyle \!\!\!\!\left.\begin{array}{@{}c@{}}\displaystyle \left(\frac{\unicode[STIX]{x2202}}{\unicode[STIX]{x2202}\unicode[STIX]{x1D70F}}-\frac{\unicode[STIX]{x1D6FE}_{k}}{2k^{2}}\frac{\unicode[STIX]{x2202}^{2}}{\unicode[STIX]{x2202}z^{2}}+\frac{\unicode[STIX]{x1D6FE}_{k}}{2}\!\right)w_{\boldsymbol{k}}(z,\unicode[STIX]{x1D70F})=-\!\left\{\frac{\unicode[STIX]{x1D6FE}_{k}}{k}\frac{\unicode[STIX]{x2202}w_{\boldsymbol{k}}}{\unicode[STIX]{x2202}z}(0,\unicode[STIX]{x1D70F})+\left[\unicode[STIX]{x1D714}_{k}^{2}+\frac{4\unicode[STIX]{x1D6E4}gk}{\unicode[STIX]{x1D6FA}^{2}}\cos (2\unicode[STIX]{x1D70F})\!\right]\unicode[STIX]{x1D701}_{\boldsymbol{k}}(\unicode[STIX]{x1D70F})\!\right\}\text{e}^{kz}\\ \displaystyle -\unicode[STIX]{x1D6F1}_{\boldsymbol{k}}(\unicode[STIX]{x1D70F})\text{e}^{kz},\\ \displaystyle \frac{\unicode[STIX]{x2202}}{\unicode[STIX]{x2202}\unicode[STIX]{x1D70F}}\unicode[STIX]{x1D701}_{\boldsymbol{k}}(\unicode[STIX]{x1D70F})=w_{\boldsymbol{k}}(0,\unicode[STIX]{x1D70F})\quad \text{and}\quad \frac{\unicode[STIX]{x2202}^{2}w_{\boldsymbol{ k}}}{\unicode[STIX]{x2202}z^{2}}(0,\unicode[STIX]{x1D70F})=-k^{2}w_{\boldsymbol{k}}(0,\unicode[STIX]{x1D70F}).\end{array}\right\} & & \displaystyle \nonumber\\ \displaystyle & & \displaystyle\end{eqnarray}$$
We now define the Laplace transform on the dimensionless time
$\unicode[STIX]{x1D70F}$
with zero initial condition (bath surface at rest),
and apply it to the three equations (2.16):
$$\begin{eqnarray}\displaystyle \!\!\!\left.\begin{array}{@{}c@{}}\displaystyle \left(s-\frac{\unicode[STIX]{x1D6FE}_{k}}{2k^{2}}\frac{\unicode[STIX]{x2202}^{2}}{\unicode[STIX]{x2202}z^{2}}+\frac{\unicode[STIX]{x1D6FE}_{k}}{2}\right)w_{\boldsymbol{k},s}(z)=-\!\left\{\frac{\unicode[STIX]{x1D6FE}_{k}}{k}\frac{\unicode[STIX]{x2202}w_{\boldsymbol{k},s}}{\unicode[STIX]{x2202}z}(0)+\unicode[STIX]{x1D714}_{k}^{2}\unicode[STIX]{x1D701}_{\boldsymbol{k},s}+\frac{2\unicode[STIX]{x1D6E4}gk}{\unicode[STIX]{x1D6FA}^{2}}(\unicode[STIX]{x1D701}_{\boldsymbol{k},s+2\text{i}}+\unicode[STIX]{x1D701}_{\boldsymbol{k},s-2\text{i}})\right\}\text{e}^{kz}\\ \displaystyle -\unicode[STIX]{x1D6F1}_{\boldsymbol{k},s}\text{e}^{kz},\\ \displaystyle s\unicode[STIX]{x1D701}_{\boldsymbol{k},s}=w_{\boldsymbol{k},s}(0)\quad \text{and}\quad \frac{\unicode[STIX]{x2202}^{2}w_{\boldsymbol{ k},s}}{\unicode[STIX]{x2202}z^{2}}(0)=-k^{2}w_{\boldsymbol{k},s}(0).\end{array}\right\} & & \displaystyle \nonumber\\ \displaystyle & & \displaystyle\end{eqnarray}$$
The solution to the above differential equation can be expressed as
provided that
$\unicode[STIX]{x1D701}_{\boldsymbol{k},s}$
satisfies the relation
where we define the function
This function and all its derivatives are self-conjugated, i.e.
$f_{k}(z^{\ast })=(f_{k}(z))^{\ast }$
. Similarly to Müller et al. (Reference Müller, Wittmer, Wagner, Albers and Knorr1997), we finally obtain the equation for surface waves of a shaken bath in the time domain, by back Laplace transforming the previous equation,
$$\begin{eqnarray}\displaystyle & & \displaystyle \left[\frac{\unicode[STIX]{x2202}^{2}\cdot }{\unicode[STIX]{x2202}\unicode[STIX]{x1D70F}^{2}}+2\unicode[STIX]{x1D6FE}_{k}\frac{\unicode[STIX]{x2202}\cdot }{\unicode[STIX]{x2202}\unicode[STIX]{x1D70F}}+\unicode[STIX]{x1D6FE}_{k}^{2}+\unicode[STIX]{x1D714}_{k}^{2}\right]\unicode[STIX]{x1D701}_{\boldsymbol{k}}(\unicode[STIX]{x1D70F})-\frac{\unicode[STIX]{x1D6FE}_{k}^{3/2}}{\sqrt{2\unicode[STIX]{x03C0}}}\int _{-\infty }^{\unicode[STIX]{x1D70F}}\frac{\text{e}^{-\unicode[STIX]{x1D6FE}_{k}(\unicode[STIX]{x1D70F}-u)/2}}{\sqrt{\unicode[STIX]{x1D70F}-u}}\left(\unicode[STIX]{x1D6FE}_{k}\unicode[STIX]{x1D701}_{\boldsymbol{k}}(u)+2\frac{\unicode[STIX]{x2202}\unicode[STIX]{x1D701}_{\boldsymbol{k}}}{\unicode[STIX]{x2202}u}(u)\right)\text{d}u\nonumber\\ \displaystyle & & \displaystyle \quad +\,\frac{4\unicode[STIX]{x1D6E4}gk}{\unicode[STIX]{x1D6FA}^{2}}\cos (2\unicode[STIX]{x1D70F})\unicode[STIX]{x1D701}_{\boldsymbol{k}}(\unicode[STIX]{x1D70F})+\unicode[STIX]{x1D6F1}_{\boldsymbol{k}}(\unicode[STIX]{x1D70F})=0,\end{eqnarray}$$
given that
${\mathcal{L}}^{-1}\{1/\sqrt{\unicode[STIX]{x1D6FE}_{k}+2s}\}=\text{e}^{-\unicode[STIX]{x1D6FE}_{k}\unicode[STIX]{x1D70F}/2}/\sqrt{2\unicode[STIX]{x03C0}\unicode[STIX]{x1D70F}}$
. Equation (2.22) can be integrated numerically to obtain the time dependency of each Fourier component
$k$
of the surface wave elevation. Several terms of this equation are identical to terms of the equation (1.4) obtained by Milewski et al. (Reference Milewski, Galeano-Rios, Nachbin and Bush2015) in the limit of low viscosity. However, the convolution integral in (2.22), which is an additional signature of viscous shear, is missing in (1.4).
2.1.1 Waves on an unshaken bath in the absence of external perturbation
In the absence of any external perturbation of the liquid (i.e. for
$\unicode[STIX]{x1D6E4}=0$
and
$\unicode[STIX]{x1D6F1}_{\boldsymbol{k},s}=0$
for all
$\boldsymbol{k}$
), the surface wave equation (2.20) becomes
$f_{k}(s)\unicode[STIX]{x1D701}_{\boldsymbol{k},s}=0$
. The condition of existence of waves
$\unicode[STIX]{x1D701}_{\boldsymbol{k},s}\neq 0$
leads to the dispersion relation of free viscous gravity–capillary waves:
Equation (2.23) can be solved for
$s$
in the low-viscosity limit
$\unicode[STIX]{x1D6FE}_{k}\ll 1$
:
$$\begin{eqnarray}\displaystyle \text{Re}(s) & = & \displaystyle -\unicode[STIX]{x1D6FE}_{k}\left\{1-\frac{1}{2}\left(\frac{\unicode[STIX]{x1D6FE}_{k}}{\unicode[STIX]{x1D714}_{k}}\right)^{1/2}-\frac{1}{8}\left(\frac{\unicode[STIX]{x1D6FE}_{k}}{\unicode[STIX]{x1D714}_{k}}\right)^{3/2}-\frac{1}{64}\left(\frac{\unicode[STIX]{x1D6FE}_{k}}{\unicode[STIX]{x1D714}_{k}}\right)^{5/2}\right.\nonumber\\ \displaystyle & & \displaystyle \left.+\,O\left(\left(\frac{\unicode[STIX]{x1D6FE}_{k}}{\unicode[STIX]{x1D714}_{k}}\right)^{9/2}\right)\right\},\end{eqnarray}$$
$$\begin{eqnarray}\displaystyle \text{Im}(s) & = & \displaystyle \unicode[STIX]{x1D714}_{k}\left\{1-\frac{1}{2}\left(\frac{\unicode[STIX]{x1D6FE}_{k}}{\unicode[STIX]{x1D714}_{k}}\right)^{3/2}+\frac{1}{8}\left(\frac{\unicode[STIX]{x1D6FE}_{k}}{\unicode[STIX]{x1D714}_{k}}\right)^{5/2}-\frac{1}{64}\left(\frac{\unicode[STIX]{x1D6FE}_{k}}{\unicode[STIX]{x1D714}_{k}}\right)^{7/2}+\frac{1}{8}\left(\frac{\unicode[STIX]{x1D6FE}_{k}}{\unicode[STIX]{x1D714}_{k}}\right)^{4}\right.\nonumber\\ \displaystyle & & \displaystyle \left.+\,O\left(\left(\frac{\unicode[STIX]{x1D6FE}_{k}}{\unicode[STIX]{x1D714}_{k}}\right)^{9/2}\right)\right\},\end{eqnarray}$$
where
$\text{Re}(s)$
and
$\text{Im}(s)$
represent respectively the damping and the frequency of the waves. The coefficients
$\unicode[STIX]{x1D6FE}_{k}$
and
$\unicode[STIX]{x1D714}_{k}$
therefore correspond to the first-order damping and frequency of gravity–capillary waves in the limit of low viscosity.
2.2 Faraday waves, just below the Faraday threshold
We now consider Faraday waves on a bath that is shaken just below the Faraday threshold
$\unicode[STIX]{x1D6E4}_{F}$
, i.e. in the conditions experienced by walkers. The goal of this section is to determine the damping and the phase shift of these waves. When they are subharmonic (as observed experimentally), they can be obtained by solving (2.22) with the Floquet ansatz
where
$\unicode[STIX]{x1D6FF}_{k}$
is a small complex perturbation from the subharmonic resonance frequency that should vanish when the instability threshold is reached. Equivalently in the Laplace domain,
Obtaining the exact solution to (2.20) from this ansatz requires solving an infinite system of algebraic equations where all the Floquet components
$\unicode[STIX]{x1D701}_{\boldsymbol{k}}^{(l)}$
are coupled to each other (Kumar & Tuckerman Reference Kumar and Tuckerman1994). As already suggested by Müller et al. (Reference Müller, Wittmer, Wagner, Albers and Knorr1997), we can simplify this ansatz by restricting it to the two most relevant Floquet components
$\unicode[STIX]{x1D701}_{\boldsymbol{k}}^{(0)}$
and
$\unicode[STIX]{x1D701}_{\boldsymbol{k}}^{(1)}$
of the wave, namely those that are associated with angular frequencies in the vicinity of
$\pm \unicode[STIX]{x1D6FA}/2$
. This truncated Floquet ansatz can be written as
$$\begin{eqnarray}\displaystyle \left.\begin{array}{@{}c@{}}\displaystyle \unicode[STIX]{x1D701}_{\boldsymbol{k}}(\unicode[STIX]{x1D70F})\simeq (\unicode[STIX]{x1D701}_{\boldsymbol{k}}^{(1)}\text{e}^{\text{i}\unicode[STIX]{x1D70F}}+\unicode[STIX]{x1D701}_{\boldsymbol{k}}^{(0)}\text{e}^{-\text{i}\unicode[STIX]{x1D70F}})\text{e}^{\unicode[STIX]{x1D6FF}_{k}\unicode[STIX]{x1D70F}},\\ \displaystyle \unicode[STIX]{x1D701}_{\boldsymbol{k},s}\simeq \frac{\unicode[STIX]{x1D701}_{\boldsymbol{k}}^{(1)}}{s-\text{i}-\unicode[STIX]{x1D6FF}_{k}}+\frac{\unicode[STIX]{x1D701}_{\boldsymbol{k}}^{(0)}}{s+\text{i}-\unicode[STIX]{x1D6FF}_{k}}.\end{array}\right\} & & \displaystyle\end{eqnarray}$$
It is shown in appendix A that this truncation to the two first terms is valid as long as
$|f_{k}(\text{i})|^{2}/|f_{k}(3\text{i})|^{2}\ll 1$
, which is true for wavenumbers
$k\sim k_{F}$
in the low-viscosity limit.
In the absence of external pressure applied on the bath,
$\unicode[STIX]{x1D6F1}_{\boldsymbol{k}}(\unicode[STIX]{x1D70F})=0$
, substitution of this ansatz into (2.20) and evaluation in
$s=-\text{i}+\unicode[STIX]{x1D6FF}_{k}$
and
$s=\text{i}+\unicode[STIX]{x1D6FF}_{k}$
yields
These two equations admit non-trivial solutions if and only if
2.2.1 Faraday instability
The special choice
$\unicode[STIX]{x1D6FF}_{k}=0$
in (2.31) yields the condition for the emergence of self-sustained Faraday waves at the subharmonic resonance. The dimensionless driving acceleration
$\unicode[STIX]{x1D6E4}_{k}$
necessary to make the wavenumber
$k$
unstable is then given by
since
$f_{k}(s)$
is self-conjugated. This function
$\unicode[STIX]{x1D6E4}_{k}$
is represented in figure 2(a). There is always a global minimum of
$\unicode[STIX]{x1D6E4}_{k}$
that corresponds to the most unstable wavenumber,
$k_{F}$
:
The corresponding acceleration threshold is denoted
$\unicode[STIX]{x1D6E4}_{F}=\unicode[STIX]{x1D6E4}_{k}(k_{F})$
. The effect of the truncation is seen in figure 2(a), where (2.32) is compared to the solution of the Floquet ansatz with a truncation to higher-order terms (appendix A). Close to the Faraday threshold
$k_{F}$
, the difference between the two truncations becomes negligible.
(a) Acceleration threshold
$\unicode[STIX]{x1D6E4}_{k}$
required to sustain waves of wavenumber
$k$
. The minimum of
$\unicode[STIX]{x1D6E4}_{k}$
gives the most unstable wavenumber,
$k_{F}$
. (b) Damping of the Faraday waves
$\text{Re}(\unicode[STIX]{x1D6FF}_{k})$
. (c) Frequency of the Faraday wave
$\text{Im}(\text{i}+\unicode[STIX]{x1D6FF}_{k})$
. (d) Phase shift
$\unicode[STIX]{x1D703}_{k}$
of the Faraday waves with respect to the forcing signal. In (a), the blue curve represents the solution of the Floquet’s ansatz with truncation to 4 harmonics (A 10), while the black curve corresponds to the truncation to 2 harmonics (2.32). In (b,c) the dotted line and the thick solid line represent equations (2.38) and (2.39), respectively. In (d), they represent (2.47). In (a,b) the red dashed curve is the low-viscosity approximation of the truncation to two harmonics (B 17) and (B 22). In (b,c) the damping and frequency of unforced waves ((2.24) and (2.25), respectively) are represented in light grey. In (a–d) all equations have been evaluated with numerical values of table 1. In particular, the distance to threshold
${\mathcal{M}}=(\unicode[STIX]{x1D6E4}_{F}-\unicode[STIX]{x1D6E4})/\unicode[STIX]{x1D6E4}_{F}=10$
. The grey shade indicates the Faraday window, i.e. the range of wavenumbers
$k$
that produce waves locked to half the driving frequency.

2.2.2 Waves just below the Faraday threshold
When the driving acceleration
$\unicode[STIX]{x1D6E4}$
is tuned slightly below the Faraday threshold
$\unicode[STIX]{x1D6E4}_{F}$
, some small complex detuning
$\unicode[STIX]{x1D6FF}_{k}$
should arise for the wavenumbers
$k\sim k_{F}$
, as prescribed by (2.31). A good approximation of this detuning can be captured by expanding the function
$f_{k}(\pm \text{i}+\unicode[STIX]{x1D6FF}_{k})$
to the second order in
$\unicode[STIX]{x1D6FF}_{k}$
:
where
${\dot{f}}_{k}=\unicode[STIX]{x2202}f_{k}/\unicode[STIX]{x2202}s$
. Substitution in (2.31) yields
$$\begin{eqnarray}\displaystyle & & \displaystyle \frac{\unicode[STIX]{x1D6FF}_{k}^{2}}{2}(f_{k}(\text{i})\ddot{f}_{k}(-\text{i})+f_{k}(-\text{i})\ddot{f}_{k}(\text{i})+2{\dot{f}}_{k}(\text{i}){\dot{f}}_{k}(-\text{i}))\nonumber\\ \displaystyle & & \displaystyle \quad +\,\unicode[STIX]{x1D6FF}_{k}(f_{k}(\text{i}){\dot{f}}_{k}(-\text{i})+f_{k}(-\text{i}){\dot{f}}_{k}(\text{i}))+\left(\frac{2gk}{\unicode[STIX]{x1D6FA}^{2}}\right)^{2}(\unicode[STIX]{x1D6E4}_{k}^{2}-\unicode[STIX]{x1D6E4}^{2})=0.\end{eqnarray}$$
The coefficients of this second-order polynomial in
$\unicode[STIX]{x1D6FF}_{k}$
are all real, because
$f_{k}$
and its derivatives are self-conjugated. If the corresponding discriminant is positive, i.e. if
then both solutions
$\unicode[STIX]{x1D6FF}_{k}\in \text{Re}$
and the resulting waves (cf. (2.28)) are locked at half the driving frequency, i.e. at
$\unicode[STIX]{x1D6FA}/2$
(figure 2
c). For a given
$\unicode[STIX]{x1D6E4}\lesssim \unicode[STIX]{x1D6E4}_{F}$
, equation (2.36) is valid for a small range of
$k$
around
$k_{F}$
that we call the Faraday window (shaded regions in figure 2). Outside this range, waves are travelling waves with group velocity different from 0, and their frequency is detuned from the forcing frequency since the corresponding
$\unicode[STIX]{x1D6FF}_{k}$
is complex. The latter waves compound the capillary wave packet by opposition to standing Faraday waves.
Equation (2.35) admits two solutions
where coefficients
$a$
,
$b$
and
$c$
are obtained by identification of (2.35) with
$a\unicode[STIX]{x1D6FF}_{k}^{2}+b\unicode[STIX]{x1D6FF}_{k}+c(\unicode[STIX]{x1D6E4}_{k}^{2}-\unicode[STIX]{x1D6E4}^{2})=0$
. Since
$\unicode[STIX]{x1D6E4}_{k}\simeq \unicode[STIX]{x1D6E4}$
inside the Faraday window, these solutions can be expanded in Taylor series for
$|4ac(\unicode[STIX]{x1D6E4}_{k}^{2}-\unicode[STIX]{x1D6E4}^{2})/b^{2}|\ll 1$
. To the first order,
and
The real and imaginary parts of these solutions
$\unicode[STIX]{x1D6FF}_{k}^{\pm }$
are represented in figures 2(b) and 2(c), respectively. The imaginary part
$\text{Im}(\unicode[STIX]{x1D6FF}_{k}^{\pm })$
corresponds to some detuning from the Faraday frequency
$\unicode[STIX]{x1D6FA}/2$
, and it is identically 0 in the Faraday window around
$k_{F}$
. The corresponding waves are standing. Outside from this Faraday window,
$\text{Im}(\unicode[STIX]{x1D6FF}_{k}^{\pm }+\text{i})$
increases with
$k$
, so the group velocity is positive and the solution corresponds to travelling waves. Inside the Faraday window, the real part of
$\unicode[STIX]{x1D6FF}_{k}^{-}$
remains largely negative so it describes Faraday waves that are strongly damped and that vanish quickly. By contrast, when
$\unicode[STIX]{x1D6E4}$
approaches
$\unicode[STIX]{x1D6E4}_{F}$
, the real part of
$\unicode[STIX]{x1D6FF}_{k}^{+}$
can be arbitrarily close to 0 for
$k\simeq k_{F}$
, in the Faraday window. Consequently,
$\unicode[STIX]{x1D6FF}_{k}^{+}$
describes some long-lived standing Faraday waves. These latter waves are those observed to travel with the walker, and they are crucial to the description of its dynamics since their amplitude remains significant during many rebounds.
The memory
$Me$
of the waves, initially described by Eddi et al. (Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011), is defined as the damping time of these long-lived Faraday waves, expressed in the number of rebounds of the droplet. Since the bouncing period corresponds to the Faraday period
$T_{F}=4\unicode[STIX]{x03C0}/\unicode[STIX]{x1D6FA}$
, the memory is given by
where
$\unicode[STIX]{x1D6FF}_{F}=\lim _{k\rightarrow k_{F}}\text{Re}(\unicode[STIX]{x1D6FF}_{k}^{+})$
and where
is called the ‘distance to the Faraday threshold’. This latter parameter has often been used as a proxy for the memory in experimental studies of walkers (Eddi et al. Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011).
The expression of
$\unicode[STIX]{x1D6FF}_{k}^{+}$
in (2.39) may be particularised to the wavenumbers around
$k_{F}$
, which will likely compose the major part of the walker’s wave field. For
$\unicode[STIX]{x1D6E4}\lesssim \unicode[STIX]{x1D6E4}_{F}$
(i.e. for
${\mathcal{M}}\gg 1$
), we can expand
$\unicode[STIX]{x1D6FF}_{k}^{+}$
in a Taylor series around
$k_{F}$
. Recalling that
$\unicode[STIX]{x2202}\unicode[STIX]{x1D6E4}_{k}/\unicode[STIX]{x2202}k=0$
in
$k=k_{F}$
, we obtain
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D6FF}_{k}^{+} & = & \displaystyle -\!\left[\frac{\unicode[STIX]{x1D6E4}_{F}(2gk_{F}/\unicode[STIX]{x1D6FA}^{2})^{2}}{(f_{F}(\text{i}){\dot{f}}_{F}(-\text{i})+f_{F}(-\text{i}){\dot{f}}_{F}(\text{i}))}\left(\frac{\unicode[STIX]{x2202}^{2}\unicode[STIX]{x1D6E4}_{k}}{\unicode[STIX]{x2202}k^{2}}\right)_{F}(k-k_{F})^{2}+O((k-k_{F})^{3})\right]\nonumber\\ \displaystyle & & \displaystyle -\,\frac{2}{{\mathcal{M}}}\left[\frac{\unicode[STIX]{x1D6E4}_{F}^{2}(2gk_{F}/\unicode[STIX]{x1D6FA}^{2})^{2}}{f_{F}(\text{i}){\dot{f}}_{F}(-\text{i})+f_{F}(-\text{i}){\dot{f}}_{F}(\text{i})}+O(k-k_{F})\right]+O({\mathcal{M}}^{-2}),\end{eqnarray}$$
where
$f_{F}=f_{k_{F}}$
and
${\dot{f}}_{F}={\dot{f}}_{k_{F}}$
. This approximation of
$\unicode[STIX]{x1D6FF}_{k}^{+}$
is plotted with a red dashed line in figure 2(b). We identify this expression with
$\unicode[STIX]{x1D6FF}_{k}^{+}=\unicode[STIX]{x1D6FF}_{F}-D(k-k_{F})^{2}+O((k-k_{F})^{3})+O({\mathcal{M}}^{-1})$
and define the diffusion coefficient
$D$
as
Since
and
$k^{2}\unicode[STIX]{x1D6E4}_{k}^{2}=\unicode[STIX]{x1D6FA}^{4}|f_{k}(\text{i})|^{2}/(2g)^{2}$
,
$D$
can be expressed in terms of
$f_{k}$
as
The amplitude
$\unicode[STIX]{x1D701}_{k}^{+}$
of the long-lived Faraday waves (respectively
$\unicode[STIX]{x1D701}_{k}^{-}$
for short-lived waves) is given from (2.29):
We introduce
in order to express the total wave field
$\unicode[STIX]{x1D701}_{k}(\unicode[STIX]{x1D70F})$
as
where
$A_{k}^{+}$
and
$A_{k}^{-}$
stand for the relative amplitude of long-lived and short-lived waves, respectively. The angles
$\unicode[STIX]{x1D703}_{k}^{\pm }$
then correspond to the phase shift between these waves and the driving acceleration. They are represented in figure 2(d). Outside the Faraday window, they are both close to 0, while inside the Faraday window they strongly vary with
$k$
in a non-trivial way. Remarkably around
$k_{F}$
, these phase shifts are
$\unicode[STIX]{x1D703}_{F}^{\pm }\simeq \pm \unicode[STIX]{x03C0}/4$
.
2.3 Single impact of a droplet
We now consider the impact of a droplet at the surface of the liquid at a time
$\unicode[STIX]{x1D70F}_{i}$
. While sophisticated theoretical frameworks have been developed in order to accurately calculate the interaction of a droplet with the surface of a liquid (Moláček & Bush Reference Moláček and Bush2012; Molacek & Bush Reference Molacek and Bush2013; Milewski et al.
Reference Milewski, Galeano-Rios, Nachbin and Bush2015), we shall, for illustration purposes, employ a rather simplified description of the droplet impact. Namely, we model its temporal profile with a delta function. The corresponding dimensionless pressure exerted by the droplet on the liquid bath is then
By integrating (2.22) across the delta kick, it is shown that
$v_{k}$
corresponds to the negative change of velocity of
$\unicode[STIX]{x1D701}_{\boldsymbol{k}}$
following impact. If the surface is perfectly flat and at rest before the impact, the wave profile is axisymmetric, i.e.
$\unicode[STIX]{x1D701}_{\boldsymbol{k}}(\unicode[STIX]{x1D70F})=\unicode[STIX]{x1D701}_{k}(\unicode[STIX]{x1D70F})$
. The initial conditions in
$\unicode[STIX]{x1D70F}\rightarrow \unicode[STIX]{x1D70F}_{i}^{+}$
are
$\unicode[STIX]{x1D701}_{k}=0$
and
$\unicode[STIX]{x2202}\unicode[STIX]{x1D701}_{k}/\unicode[STIX]{x2202}\unicode[STIX]{x1D70F}=-v_{k}$
. They determine the coefficients
$A_{k}^{\pm }$
of (2.48):
Transforming back to the spatial domain according to (2.9) then yields the radially symmetric surface wave profile
that is generated by a delta kick, where
$\text{J}_{0}$
is the Bessel function of the first kind of order zero.
At large times
$\unicode[STIX]{x1D70F}\gg \unicode[STIX]{x1D70F}_{i}$
for
$\unicode[STIX]{x1D6E4}\lesssim \unicode[STIX]{x1D6E4}_{F}$
, the wave field is dominated by the contribution
$\unicode[STIX]{x1D701}_{k}^{+}$
, so
where
$B_{k}^{+}(\unicode[STIX]{x1D70F}_{i})=A_{k}^{+}\text{e}^{\unicode[STIX]{x1D6FF}_{k}^{+}\unicode[STIX]{x1D70F}_{i}}$
. With the expansion
$\unicode[STIX]{x1D6FF}_{k}^{+}=\unicode[STIX]{x1D6FF}_{F}-D(k-k_{F})^{2}+O((k-k_{F})^{3})+O({\mathcal{M}}^{-1})$
, we can then approximate this integral thanks to the saddle-point method in the formal limit
$\unicode[STIX]{x1D70F}\rightarrow \infty$
. Calculations detailed in appendix C yield:
which is at least valid when
$k_{F}r\ll 1$
and
$k_{F}r\gg 1$
. Equation (2.53) is the impulse response of the Faraday waves in the long-time limit. According to Duhamel’s principle, the response to any sophisticated impact signal
$\unicode[STIX]{x1D6F1}_{k}(\unicode[STIX]{x1D70F})$
can be inferred by convoluting this impulse response with
$\unicode[STIX]{x1D6F1}_{k}(\unicode[STIX]{x1D70F})$
.
Equation (2.53) may be compared to the main previous model of waves generated by a single impact of a droplet (1.5), which can be rewritten as:
The main additions of (2.53) to (2.54) are (i) the phase shift
$\unicode[STIX]{x1D703}_{F}^{+}$
, (ii) a diffusive spreading in the exponential decay (diffusion coefficient
$D$
) and (iii) a well-defined amplitude
$B_{F}^{+}$
that depends on the impact phase
$\unicode[STIX]{x1D70F}_{i}$
through (2.50). Modifications similar to (ii) and (iii) can be deduced with additional calculations, starting from the theory of Moláček & Bush (Reference Moláček and Bush2013).
2.4 Surface waves emitted by a walker
The long-lived Faraday wave generated by a walker is obtained by superimposing impulse responses corresponding to each droplet impact at time
$\unicode[STIX]{x1D70F}_{n}$
and position
$\boldsymbol{r}_{n}$
. In the absence of other walkers, submerged boundaries or body forces, a walker bounces periodically every Faraday period and moves in a straight line at constant speed
$V_{w}$
. Consequently,
where
$v=2V_{w}/\unicode[STIX]{x1D6FA}$
is the normalised walking speed. The origin
$\boldsymbol{r}=0$
is set where the walker is at time
$\unicode[STIX]{x1D70F}$
. The Faraday wave field is then given by
where
$[(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})/2\unicode[STIX]{x03C0}]$
denotes the largest integer smaller than
$(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})/2\unicode[STIX]{x03C0}$
.
(a) Asymmetric wave field of a walker, calculated numerically from equation (2.57), for different values of the memory
$Me$
. From left to right,
$Me=7.7$
,
$Me=16$
,
$Me=82$
and
$Me=447$
respectively. The walker velocity is fixed to
$V_{w}=10~\text{mm}~\text{s}^{-1}$
from left to right, and other parameters are given in table 1. The field of view is
$4\times 4~\text{cm}$
. (b) Anisotropic damping length
$l(\unicode[STIX]{x1D703})$
, compared to the experimental wave field of a walker. The red curve corresponds to (2.60) with
${\mathcal{M}}=10$
and other numerical values from table 1. In the experiment, the walking speed is
$10~\text{mm}~\text{s}^{-1}$
and
${\mathcal{M}}=10$
. The scale bar is 1 cm. (c) Lateral damping length (
$\unicode[STIX]{x1D703}=\unicode[STIX]{x03C0}/2$
) for a bouncer (black curve,
$V_{w}=0$
) and a walker (red curve,
$V_{w}=11~\text{mm}~\text{s}^{-1}$
), as a function of the distance to threshold
${\mathcal{M}}$
.
This sum is evaluated in appendix D. It yields
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D701}_{w}(\boldsymbol{r},\unicode[STIX]{x1D70F}) & = & \displaystyle \frac{B_{F}^{+}k_{F}}{2\sqrt{D}}\cos (\unicode[STIX]{x1D70F}+\unicode[STIX]{x1D703}_{F}^{+})\exp \left(-\frac{vr\cos \unicode[STIX]{x1D703}}{2D}\right)\nonumber\\ \displaystyle & & \displaystyle \times \,\int _{-\unicode[STIX]{x03C0}}^{\unicode[STIX]{x03C0}}\,\text{d}\unicode[STIX]{x1D711}\frac{\exp [\text{i}k_{F}r\sin (\unicode[STIX]{x1D703}-\unicode[STIX]{x1D711})]}{\sqrt{a-\text{i}k_{F}v\sin \unicode[STIX]{x1D711}}}\exp \left(-\frac{r}{\sqrt{D}}\sqrt{a-\text{i}k_{F}v\sin \unicode[STIX]{x1D711}}\right),\end{eqnarray}$$
where
$\cos \unicode[STIX]{x1D703}=\boldsymbol{v}\boldsymbol{\cdot }\boldsymbol{r}/(vr)$
and
The corresponding wave field can be calculated by numerically integrating (2.57). Some characteristic wave fields are shown in figure 3(a), for different values of the memory
$Me$
. With increasing
$Me$
, the sum includes significant contributions from an increasing number of past impacts. Some interferences appear in the wake of the walker, similarly to what is seen experimentally (Eddi et al.
Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011).
In the limit of small walking speed and low memory (
$2\unicode[STIX]{x03C0}Mek_{F}v\ll 1$
), the integral (2.57) can be solved analytically (appendix D):
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D701}_{w}(\boldsymbol{r},\unicode[STIX]{x1D70F})=\frac{B_{F}^{+}k_{F}}{\sqrt{v^{2}+\displaystyle \frac{2D}{\unicode[STIX]{x03C0}Me}}}\cos (\unicode[STIX]{x1D70F}+\unicode[STIX]{x1D703}_{F}^{+})\text{J}_{0}\left(k_{F}\left|\boldsymbol{r}+\frac{r}{\sqrt{v^{2}+\displaystyle \frac{2D}{\unicode[STIX]{x03C0}Me}}}\boldsymbol{v}\right|\right)\exp [-r/l(\unicode[STIX]{x1D703})],\qquad & & \displaystyle\end{eqnarray}$$
where
$$\begin{eqnarray}\displaystyle l(\unicode[STIX]{x1D703})=\frac{2D}{v\cos \unicode[STIX]{x1D703}+\sqrt{v^{2}+\displaystyle \frac{2D}{\unicode[STIX]{x03C0}Me}}}. & & \displaystyle\end{eqnarray}$$
Equation (2.59) corresponds to a Doppler-shifted Bessel profile which is exponentially attenuated on an length scale
$l(\unicode[STIX]{x1D703})$
that varies with direction
$\unicode[STIX]{x1D703}$
. The resulting wavelength is smaller in front of the walker and larger in its wake. The damping distance is smaller in front of the walker than in its wake, as illustrated in figure 3(b).
For either larger walking speed of larger memory, interferences between Bessel functions are more complex and solving the integral (2.57) is not straightforward. Nevertheless, an upper bound to the wave field can be found, as derived in appendix D:
This bound indicates that the Faraday waves emitted by the walkers are always subjected to the same exponential spatial damping
$l(\unicode[STIX]{x1D703})$
defined in (2.60).
In the case of a bouncer (
$v=0$
), this damping length
$l(\unicode[STIX]{x1D703})=\sqrt{2\unicode[STIX]{x03C0}DMe}$
is independent of
$\unicode[STIX]{x1D703}$
(axisymmetric damping) and it grows indefinitely with increasing memory. At the Faraday threshold (infinite memory), a bouncer would generate waves that are not spatially damped. By contrast, waves are always spatially damped for a walker with finite speed
$v$
. Indeed, when
$Me\rightarrow \infty$
,
$l(\unicode[STIX]{x1D703})\rightarrow 2D/[v(1+\cos \unicode[STIX]{x1D703})]$
. This horizontal asymptote is illustrated in figure 3(c). Hence, equation (2.61) predicts a finite spatial damping even at infinite memory, and this for all directions except
$\unicode[STIX]{x1D703}=\unicode[STIX]{x03C0}$
.
2.5 Approximations specific to the experimental regime of walkers
The subthreshold Faraday wave theory developed in previous sections is now particularised to the regime of walkers that corresponds to most experiments reported up to now (Bush Reference Bush2015). Typical values for each parameter are summarised in table 1. The liquid is silicone oil with density
$\unicode[STIX]{x1D70C}=956~\text{kg}~\text{m}^{-3}$
, surface tension
$\unicode[STIX]{x1D70E}=20.6\times 10^{-3}~\text{N}~\text{m}^{-1}$
and viscosity
$\unicode[STIX]{x1D708}=20\times 10^{-6}~\text{m}^{2}~\text{s}^{-1}$
. The bath is vibrated sinusoidally at a frequency
$\unicode[STIX]{x1D6FA}/2\unicode[STIX]{x03C0}=80~\text{Hz}$
.
In the limit of vanishing viscosity
$\unicode[STIX]{x1D708}$
,
$\unicode[STIX]{x1D6FE}_{k}\rightarrow 0$
, and
$f_{k}(i)\rightarrow \unicode[STIX]{x1D714}_{k}^{2}-1$
. Therefore, according to (2.32), the instability threshold
$\unicode[STIX]{x1D6E4}_{F}\rightarrow 0$
in
$k=k_{0}$
such that
$\unicode[STIX]{x1D714}_{0}^{2}=\unicode[STIX]{x1D714}_{k_{0}}^{2}=1$
. For a walker,
$k_{0}=1319.5~\text{m}^{-1}$
, which corresponds to a wavelength of 4.76 mm, already close to experimental observations (Eddi et al.
Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011). The corresponding damping (at
$\unicode[STIX]{x1D708}=20~\text{cS}$
) is
$\unicode[STIX]{x1D6FE}_{0}=\unicode[STIX]{x1D6FE}_{k_{0}}=0.277$
. Although this value is not much smaller than 1, we can perform a Taylor expansion in
$\unicode[STIX]{x1D6FE}_{0}$
for each quantity of interest.
First, we prove in appendix B that the most unstable wavenumber
$k_{F}$
, defined from (2.33), can be approximated by:
where
$d_{0}=(\unicode[STIX]{x2202}^{2}\unicode[STIX]{x1D714}_{k}^{2}/\unicode[STIX]{x2202}k^{2})(k_{0})=24\unicode[STIX]{x1D70E}k_{0}/\unicode[STIX]{x1D70C}\unicode[STIX]{x1D6FA}^{2}$
and
$c_{0}=(\unicode[STIX]{x2202}\unicode[STIX]{x1D714}_{k}/\unicode[STIX]{x2202}k)(k_{0})=(2(g+3\unicode[STIX]{x1D70E}k_{0}^{2}/rho)/\unicode[STIX]{x1D6FA}^{2})$
is the group velocity of free waves around the wavenumber
$k_{0}$
. For walkers,
$c_{0}=0.97~\text{mm}$
(which corresponds to a speed
$\unicode[STIX]{x1D6FA}c_{0}/2=0.24~\text{m}~\text{s}^{-1}$
) and the approximation (2.62) gives
$k_{F}\simeq 1338.7~\text{m}^{-1}$
and
$\unicode[STIX]{x1D706}_{f}=4.693~\text{mm}$
, while a numerical computation of the exact solution of (2.33) without any approximation except the Floquet’s first-order truncation would give
$k_{F}=1334.8~\text{m}^{-1}$
and
$\unicode[STIX]{x1D706}_{f}=4.707~\text{mm}$
.
From appendix B, it follows that
$$\begin{eqnarray}\displaystyle \left.\begin{array}{@{}c@{}}\displaystyle \unicode[STIX]{x1D714}_{F}^{2}=1+\unicode[STIX]{x1D6FE}_{0}^{3/2}-\frac{2+k_{0}c_{0}}{k_{0}c_{0}}\unicode[STIX]{x1D6FE}_{0}^{2}+\frac{12+k_{0}c_{0}}{4k_{0}c_{0}}\unicode[STIX]{x1D6FE}_{0}^{5/2}+O(\unicode[STIX]{x1D6FE}_{0}^{3}),\\ \displaystyle \unicode[STIX]{x1D6FE}_{F}=\unicode[STIX]{x1D6FE}_{0}+\frac{\unicode[STIX]{x1D6FE}_{0}^{5/2}}{k_{0}c_{0}}+O(\unicode[STIX]{x1D6FE}_{0}^{3}).\end{array}\right\} & & \displaystyle\end{eqnarray}$$
Equation (2.32) for the corresponding instability threshold
$\unicode[STIX]{x1D6E4}_{F}=\unicode[STIX]{x1D6E4}_{k}(k_{F})$
is then approximated by:
The first term of
$\unicode[STIX]{x1D6E4}_{F}$
corresponds to the threshold predicted from the damped Matthieu equation (1.3) with a phenomenological factor
$\unicode[STIX]{x1D6FC}=2$
(Sampara & Gilet Reference Sampara and Gilet2016). Since for walkers
$\unicode[STIX]{x1D6FE}_{0}=0.277$
, a limitation to this first term would yield
$\unicode[STIX]{x1D6E4}_{F}=5.41$
. Considering instead the first two terms yields
$\unicode[STIX]{x1D6E4}_{F}=3.985$
. The exact solution of (2.32) obtained numerically is
$\unicode[STIX]{x1D6E4}_{F}=4.144$
, which is very close to experimental values. The correction in
$\unicode[STIX]{x1D6FE}_{0}^{3/2}$
is therefore a considerable improvement of this theory with respect to the damped Matthieu equation.
The memory
$Me$
defined in (2.40) can be approximated by:
It is therefore inversely proportional to the viscosity
$\unicode[STIX]{x1D708}$
and an inviscid fluid would yield a theoretically infinite memory. In the case of walkers,
$(1+\unicode[STIX]{x1D6FE}_{0}^{1/2}/2+\unicode[STIX]{x1D6FE}_{0}/4)/2\unicode[STIX]{x03C0}\unicode[STIX]{x1D6FE}_{0}\sim 0.76$
, which is fairly close to unity. In previous work on walkers,
$Me$
has often been mixed up with the distance to threshold
${\mathcal{M}}$
, without much consequence. Close to threshold (
$\unicode[STIX]{x1D6E4}\lesssim \unicode[STIX]{x1D6E4}_{F}$
), the damping factor of short-lived Faraday waves
$\unicode[STIX]{x1D6FF}_{F}^{-}=-2\unicode[STIX]{x1D6FE}_{0}+O(\unicode[STIX]{x1D6FE}_{0}^{3/2})$
is significantly more negative than the damping of long-lived waves
$\unicode[STIX]{x1D6FF}_{F}^{+}=-1/(2\unicode[STIX]{x03C0}Me)$
.
The diffusion coefficient
$D$
associated with long-lived Faraday waves and defined in (2.45) is approximated by
which is about
$D\simeq 4.40~\text{mm}^{2}$
for a walker.
The phase shift of long-lived and short-lived Faraday waves near the Faraday threshold is:
and consequently,
The long-lived impulse response (2.53) then becomes:
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D701}_{1}(r,\unicode[STIX]{x1D70F}) & \simeq & \displaystyle v_{F}k_{F}\sqrt{\frac{\unicode[STIX]{x03C0}}{D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})}}\text{J}_{0}(k_{F}r)\cos \left(\unicode[STIX]{x1D70F}_{i}-\frac{\unicode[STIX]{x03C0}}{4}\right)\cos \left(\unicode[STIX]{x1D70F}+\frac{\unicode[STIX]{x03C0}}{4}\right)\nonumber\\ \displaystyle & & \displaystyle \times \,\exp \left[-\frac{\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i}}{2\unicode[STIX]{x03C0}Me}-\frac{r^{2}}{4D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})}\right],\end{eqnarray}$$
which reads in dimensional form,
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D701}_{1}(r,t) & \simeq & \displaystyle v_{F}k_{F}\sqrt{\frac{2\unicode[STIX]{x03C0}}{D\unicode[STIX]{x1D6FA}(t-t_{i})}}\text{J}_{0}(k_{F}r)\cos \left(\frac{\unicode[STIX]{x1D6FA}t_{i}}{2}-\frac{\unicode[STIX]{x03C0}}{4}\right)\cos \left(\frac{\unicode[STIX]{x1D6FA}t}{2}+\frac{\unicode[STIX]{x03C0}}{4}\right)\nonumber\\ \displaystyle & & \displaystyle \times \,\exp \left[-\frac{(t-t_{i})}{MeT_{F}}-\frac{r^{2}}{2D\unicode[STIX]{x1D6FA}(t-t_{i})}\right].\end{eqnarray}$$
The upper bound (2.61) on the wave field of a walker that moves at constant normalised speed
$v$
becomes
$$\begin{eqnarray}\displaystyle \left|\frac{\unicode[STIX]{x1D701}_{w}(\boldsymbol{r},\unicode[STIX]{x1D70F})}{\cos (\unicode[STIX]{x1D70F}+\unicode[STIX]{x03C0}/4)}\right|\leqslant \frac{2\unicode[STIX]{x03C0}v_{F}k_{F}\cos (\unicode[STIX]{x1D70F}_{i}-\unicode[STIX]{x03C0}/4)}{\sqrt{v^{2}+\displaystyle \frac{c_{0}^{2}}{{\mathcal{M}}}}}\exp \left[-\frac{r}{l(\unicode[STIX]{x1D703})}\right] & & \displaystyle\end{eqnarray}$$
with
For distance to threshold
${\mathcal{M}}\gg c_{0}^{2}/v^{2}$
(which is approximately 490 for walkers),
The damping length
$l$
can therefore be as large as wanted in the wake of the walker (
$\unicode[STIX]{x1D703}=\unicode[STIX]{x03C0}$
), but it is strongly limited in front of the walker, independently of
${\mathcal{M}}$
. Nevertheless,
$l(0)$
could be increased by decreasing
$\unicode[STIX]{x1D6FE}_{0}$
, i.e. by decreasing viscosity and associated dissipation.
At sufficiently low
${\mathcal{M}}$
, given by (2.59) is approximated by:
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D701}_{w}(\boldsymbol{r},t)\simeq \frac{v_{F}k_{F}}{\sqrt{v^{2}+\displaystyle \frac{c_{0}^{2}}{{\mathcal{M}}}}}\cos \left(\unicode[STIX]{x1D70F}_{i}-\frac{\unicode[STIX]{x03C0}}{4}\right)\cos \left(\unicode[STIX]{x1D70F}+\frac{\unicode[STIX]{x03C0}}{4}\right)\text{J}_{0}\left[k_{F}\left|\boldsymbol{r}+\frac{r}{\sqrt{v^{2}+\displaystyle \frac{c_{0}^{2}}{{\mathcal{M}}}}}\boldsymbol{v}\right|\right]\text{e}^{-r/l(\unicode[STIX]{x1D703})}. & & \displaystyle \nonumber\\ \displaystyle & & \displaystyle\end{eqnarray}$$
This approximation is only valid when
$k_{F}v/a\ll 1$
(appendix D), which here reduces to
This limit is
${\mathcal{M}}\ll 4.9$
for a walker at speed
$V_{w}=11~\text{mm}~\text{s}^{-1}$
. Furthermore, since
${\mathcal{M}}\ll c_{0}^{2}/v^{2}\simeq 490$
, the Doppler shift along the walker trajectory in (2.74) can be approximated by
This expression is very similar to the one obtained numerically in Milewski et al. (Reference Milewski, Galeano-Rios, Nachbin and Bush2015).
3 Experimental observation of the surface waves emitted by a localised impact
This section aims at validating experimentally the theory of subthreshold surface waves developed in § 2. Specifically, we characterise (i) the spatial damping of the Faraday waves upon a single impact, (ii) the phase shift between these waves and the shaker position and (iii) the dependence of the wave amplitude on the phase of impact. These three points are major additions of the present theory with respect to previous models of the Faraday waves generated by walkers (Eddi et al. Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011; Moláček & Bush Reference Moláček and Bush2013).
3.1 Experimental set-up
Tilted view (a) and top view (b) of the experimental set-up. A large transparent reservoir of oil is mounted on top of a shaker. Below the reservoir a transparent sheet is printed with a synthetic schlieren. The border of the reservoir is a shallow region where surface waves are strongly damped. A glass bead is released on demand above the reservoir with a vice jaw opened with a translation stage. The acceleration of the reservoir is measured with an accelerometer placed on the border of the reservoir.

The goal of the experiment is to finely quantify the spatio-temporal evolution of the surface waves generated by the single localised impact of a bead. The experimental set-up, depicted in figure 4, comprises a transparent cylindrical reservoir of diameter 22 cm filled with approximately 5 mm of silicone oil (Sigma-Aldrich). The reservoir is delimited by a shallow region (width 3 cm, depth
${<}1~\text{mm}$
), thanks to which surface waves are strongly damped. This trick allows us to eliminate the additional waves generated by the meniscus at the boundaries of the reservoir (Bechhoefer et al.
Reference Bechhoefer, Ego, Manneville and Johnson1995; Nguyem Thu Lam & Caps Reference Nguyem Thu Lam and Caps2011). The reservoir is also large enough (
$22~\text{cm}\simeq 46\unicode[STIX]{x1D706}_{f}$
) to avoid confinement effects that may stabilise the bath when the bath diameter is less than 10
$\unicode[STIX]{x1D706}_{f}$
(Nguyem Thu Lam & Caps Reference Nguyem Thu Lam and Caps2011). The reservoir is placed on top of an electromagnetic shaker (V400LT, Data Physics). The shaker motion is accurately controlled via an accelerometer tightly fixed on the reservoir, a computer interface and a feedback loop, as detailed in Sampara & Gilet (Reference Sampara and Gilet2016). The resulting effective gravity acceleration is
$g(t)=g(1+\unicode[STIX]{x1D6E4}\cos \unicode[STIX]{x1D6FA}t)$
. The frequency is fixed to
$\unicode[STIX]{x1D6FA}=2\unicode[STIX]{x03C0}80~\text{s}^{-1}$
.
The bead is made of glass and its diameter is 1 mm. A device that can release the bead on demand is placed at 1 cm above the bath. It consists in a vice with two jaws that grip the bead. It is mechanically insulated from the shaker, so it is not sensitive to its vibrations. A translation stage allows us to open the vice and release the bead. The bead does not acquire any significant velocity from the vice, so it starts falling from rest.
Measurements must be gathered over many Faraday periods, with a time resolution well below the shaker period
$2\unicode[STIX]{x03C0}/\unicode[STIX]{x1D6FA}$
. Therefore, we here use high-speed imaging (Phantom MIRO110), with an acquisition frequency of 4000 fps (i.e. 50 frames per period of the shaker). The shaker is illuminated with 4 home-made high-power LEDs to ensure strong illumination of the waves while limiting the associated heating. Thanks to a semi-refringent mirror placed at
$45^{\circ }$
above the shaker, optical paths of both illumination and recording are vertical near the bath surface. Thanks to a macro lens Zeiss Milvus 2/100M, the spatial resolution is brought down to
$70~\unicode[STIX]{x03BC}\text{m}$
by pixel. The camera is synchronised with the shaker thanks to a DACQ (National Instruments) and a Labview interface that returns the impact phase
$\unicode[STIX]{x1D70F}_{i}$
, with respect to the Faraday period
$T_{F}$
. Details on this crucial synchronisation are provided in appendix E.
The free-surface synthetic schlieren technique developed by Moisy, Rabaud & Salsac (Reference Moisy, Rabaud and Salsac2009) is well adapted to measure wave elevations in real time. It consists in placing a transparent sheet printed with randomly spaced dots (dot diameter
$100~\unicode[STIX]{x03BC}\text{m}$
, 50 % density) over a white background at the bottom of the reservoir. The image of these dots moves owing to the light refraction at the free surface. The displacement of the imaged dots is therefore related to the local slope of the free surface. This technique was already successfully applied to measure the 3-D shape of the long-lived Faraday waves generated by walkers, with an acquisition frequency lower than
$\unicode[STIX]{x1D6FA}/(2\unicode[STIX]{x03C0})$
(Eddi et al.
Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011; Damiano et al.
Reference Damiano, Brun, Harris, Galeano-Rios and Bush2016). However, in its original form, it is not well suited to the measurements intended here. Indeed, it requires a large field of view that must not be obstructed by the device which would drop the bead. Moreover, owing to the limited amount of information (3 GB) that can be acquired by a high-speed camera in a single shot, the recording duration at full spatial resolution would be a fraction of a second, which is definitely not sufficient to capture the long-term behaviour of Faraday waves. Finally, the reconstruction algorithm of Moisy’s technique would demand large computational resources. In order to circumvent these limitations, we here use an adaptation of Moisy’s technique. Since the wave field is known to be axisymmetric, we restrict the field of view to a radial zone of
$2\times 15~\text{mm}$
from the impact point. This allows us to record much longer films (several seconds at 4000 fps), without any obstruction of the field of view by the bead-releasing device. The displacement of the imaged dots is calculated along a radius from impact point. The one-dimensional radial gradient of the elevation
$\unicode[STIX]{x1D701}(r,t)$
is then calculated with a 1-D adaptation of the reconstruction algorithm detailed in Moisy et al. (Reference Moisy, Rabaud and Salsac2009). The restriction to a quasi-1-D wave pattern severely decreases the computational time of the reconstruction. However, it can be shown that the 1-D algorithm does not benefit anymore from the error compensation mechanism present in the 2-D algorithm. In order to limit the measurement error, we first remove most of the noise with deshaking and median filters. We also avoid the error build-up inherent to the spatial integration of
$\unicode[STIX]{x2202}_{r}\unicode[STIX]{x1D701}$
to obtain
$\unicode[STIX]{x1D701}(r,t)$
. We instead directly compare the measured
$\unicode[STIX]{x2202}_{r}\unicode[STIX]{x1D701}$
to the theoretical prediction
$$\begin{eqnarray}\displaystyle \frac{\unicode[STIX]{x2202}\unicode[STIX]{x1D701}_{1}}{\unicode[STIX]{x2202}r}(r,t) & = & \displaystyle -v_{F}k_{F}\sqrt{\frac{2\unicode[STIX]{x03C0}}{D\unicode[STIX]{x1D6FA}(t-t_{i})}}\cos \left(\frac{\unicode[STIX]{x1D6FA}t_{i}}{2}-\frac{\unicode[STIX]{x03C0}}{4}\right)\cos \left(\frac{\unicode[STIX]{x1D6FA}t}{2}+\frac{\unicode[STIX]{x03C0}}{4}\right)\nonumber\\ \displaystyle & & \displaystyle \times \left[\text{J}_{1}(k_{F}r)+\frac{r\text{J}_{0}(k_{F}r)}{D\unicode[STIX]{x1D6FA}(t-t_{i})}\right]\exp \left[-\frac{(t-t_{i})}{MeT_{F}}-\frac{r^{2}}{2D\unicode[STIX]{x1D6FA}(t-t_{i})}\right].\end{eqnarray}$$
Several input parameters are systematically varied, including the impact phase of the bead relative to the shaker, the shaker amplitude and corresponding distance to threshold
${\mathcal{M}}$
, and the liquid viscosity (5 cS and 20 cS silicone oil).
Qualitative comparison of the spatio-temporal evolution of the dimensionless radial wave gradient
$\unicode[STIX]{x2202}_{r}\unicode[STIX]{x1D701}(r,t)$
, represented here with a colour code. The impact is located in
$(r,t)=(0,0)$
. (a) Experiment at
${\mathcal{M}}=30$
and
$\unicode[STIX]{x1D708}=20~\text{cS}$
. Experimental parameters are gathered in table 1. (b) Theoretical prediction (3.1).

3.2 Spatio-temporal evolution of the wave field
A typical evolution of the wave field is represented in the space–time diagram of figure 5(a). We observe that the impact first generates a large capillary wave packet that travels radially outwards with an almost constant velocity of
$16~\text{cm}~\text{s}^{-1}$
. The long-lived Faraday waves described in § 2 are observed in the wake of this wave packet. They have a pulsation
$\unicode[STIX]{x1D6FA}/2$
and a wavelength
$2\unicode[STIX]{x03C0}/k_{F}\simeq 4.7~\text{mm}$
. The surface appears still in the region that is not yet visited by the wave packet (i.e. in the upper-right corner of figure 5
a).
This experiment can be qualitatively compared to the theoretical prediction (3.1), where the amplitude of this latter is fitted on the experiments (figure 5
b). Overall, this comparison is promising. However, the capillary wave packet is not present in the prediction, which is expected since only long standing Faraday waves were considered in (3.1). This latter also predicts waves of finite amplitude everywhere on the bath at
$t>t_{i}$
, independently of the passage of the capillary wave packet, which was hard to observe experimentally. Nevertheless, their amplitude remains very small shortly after impact and far from the impact point, which may be the reason why the bath appears still this region of the experimental spatio-temporal diagram.
A more quantitative comparison is provided in figure 6(a,b), where the radial gradient of the wave is represented as a function of time (at a given position), and as a function of space (at a given time), respectively. The theoretical prediction (3.1) of long-lived Faraday waves is compared to both experiments and the previous model of Moláček & Bush (Reference Moláček and Bush2013) (given by the gradient of (1.5)). The amplitude factors (
$v_{F}$
in (3.1), and
$A\sin \unicode[STIX]{x1D6F7}$
in (1.5)) are fitted to the experimental data in the region
$(r,t)$
that corresponds to the space–time domain after the passage of the capillary wave packet.
In figure 6(a), it is seen that both models capture well the long-term evolution of the wave at a given position. Nevertheless, a zoom on a small time window (inset of figure 6
a) reveals that the factor
$\cos (\unicode[STIX]{x1D6FA}t/2+\unicode[STIX]{x03C0}/4)$
of (3.1) fits experimental data better than the factor
$\cos (\unicode[STIX]{x1D6FA}t/2)$
of (1.5). In the short-term regime (i.e. before the passage of the capillary wave packet), the experimental wave amplitude is very small. None of the models accurately capture this behaviour, although the addition of short-lived waves in (3.1) might improve the prediction. However, equation (1.5) diverges at short times, while equation (3.1) grows from zero according to an exponential–diffusive behaviour, which is much more realistic. Moreover in (1.5), the information of impact is immediately transmitted everywhere on the bath. By contrast, the present theory predicts that after an impact the bath is first quiet and destabilises progressively owing to diffusive spreading.
The radial evolution of the wave at a given time, seen in figure 6(b), is qualitatively similar for both models and experiment. Nevertheless, the spatial damping predicted by (3.1) is closer to experiments than the one of (1.5).
Radial gradient of surface waves, (a) versus time at a fixed distance
$r=1.32~\text{cm}$
from the impact, and (b) versus position at a fixed time,
$t=0.233~\text{s}$
after impact. The experimental data are a subset of those represented in figure 5(a). Black dots correspond to the gradient measured experimentally. The grey and red solid curves correspond to (1.5) and (3.1), respectively. The inset in (a) is a zoom of the main graph on a smaller time window. The present model (red curve) predicts a phase shift of
$+\unicode[STIX]{x03C0}/4$
between the Faraday waves and the forcing signal, which is in better agreement with experiments than previous models (grey curve).

The temporal damping of Faraday waves (envelope of the curve in figure 6
a) was already observed by Eddi et al. (Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011). It was then interpreted as the memory
$Me$
of the waves, which revealed to be a crucial parameter for the dynamics of walkers (Bush Reference Bush2015). However, the definition given by Eddi et al. (Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011) differs from (2.65), as the former is independent of viscosity while the latter is inversely proportional to viscosity in first approximation. In order to validate (2.40), we have measured the damping time
$MeT_{F}$
as a function of the distance to threshold
${\mathcal{M}}$
for two different silicone oils, of kinematic viscosity 5 cS and 20 cS respectively (figure 7). The comparison with the theoretical prediction (2.65) is excellent. In particular
$MeT_{F}$
is proportional to
${\mathcal{M}}$
and inversely proportional to viscosity.
Temporal damping
$MeT_{F}$
, as a function of the distance to threshold
${\mathcal{M}}$
. Red diamonds (♢) and blue circles (○) correspond to experiments with silicone oil of viscosity 5 cS and 20 cS, respectively. Red and blue solid lines correspond to (2.65) with the corresponding viscosity. The grey line represents
$MeT_{F}={\mathcal{M}}$
.

3.3 Wave amplitude versus impact phase
The amplitude of the Faraday waves generated by an impact is reported for different phases of impact
$\unicode[STIX]{x1D70F}_{i}=t_{i}\unicode[STIX]{x1D6FA}/2$
in figure 8. This amplitude is measured as the experimental envelope of
$\unicode[STIX]{x2202}_{r}\unicode[STIX]{x1D701}(r,t)$
at a particular location
$r$
and time
$t$
. It varies sinusoidally with
$\unicode[STIX]{x1D70F}_{i}$
, so there are some impact phases for which the long-term excited wave is almost non-existent. Our model (3.1) predicts a dependence in
$\cos (\unicode[STIX]{x1D70F}_{i}-\unicode[STIX]{x03C0}/4)$
which is qualitatively similar. However, a dependence in
$\cos (\unicode[STIX]{x1D70F}_{i}+\unicode[STIX]{x03C0}/5)$
would correspond much better to the measurements.
This phase shift between experiments and theory is likely due to the fact that (3.1) represents an impulse response of surface waves, while in experiments the penetration of the bead into the bath is not instantaneous. According to Duhamel’s principle, the response to the bead impact can be calculated by convolving the impulse response with the sophisticated pressure signal
$\unicode[STIX]{x1D6F1}_{k}(\unicode[STIX]{x1D70F})$
applied by the bead on the bath. This convolution is expected to approximately shift the effective impact time by half the time of penetration, which is of the order of the contact time
$5\sqrt{\unicode[STIX]{x1D70C}R_{d}^{3}/\unicode[STIX]{x1D70E}}\sim 12~\text{ms}$
for a bead of radius
$R_{d}=0.5~\text{mm}$
(Molacek & Bush Reference Molacek and Bush2013). On the other hand, the time delay corresponding to a phase shift from
$-\unicode[STIX]{x03C0}/4$
to
$\unicode[STIX]{x03C0}/5$
is 5.625 ms, which is very close to the 6 ms that such convolution would predict.
Amplitude of the Faraday waves gradient, normalised with
$A_{0}=1.9\times 10^{-4}$
(at fixed position
$r=1.32~\text{cm}$
and time
$t=0.233~\text{s}$
), as a function of impact phase
$\unicode[STIX]{x1D70F}_{i}$
. Symbols correspond to experimental data with oil viscosity
$\unicode[STIX]{x1D708}=20~\text{cS}$
,
$\unicode[STIX]{x1D6FA}=80\times 2\unicode[STIX]{x03C0}~\text{s}^{-1}$
and
${\mathcal{M}}=30$
. The red curve is the theoretical prediction
$\cos (\unicode[STIX]{x1D70F}_{i}-\unicode[STIX]{x03C0}/4)$
from the impulse response (3.1) while the black dashed line is
$\cos (\unicode[STIX]{x1D70F}_{i}+\unicode[STIX]{x03C0}/5)$
.

4 Discussion and conclusion
In § 2, we introduced a theory for the Faraday instability derived from Navier–Stokes equations and based on the only assumption that the surface waves are linear and purely subharmonic. This assumption allows us to truncate the Floquet series (2.26), and it is valid when
$\unicode[STIX]{x1D6E4}\lesssim \unicode[STIX]{x1D6E4}_{F}$
(i.e. close to the Faraday threshold) for wavenumbers
$k\simeq k_{F}$
. We have then calculated the dispersion relation of Faraday waves, the Faraday threshold
$\unicode[STIX]{x1D6E4}_{F}$
and the Faraday wavenumber,
$k_{F}$
. We have shown that for
$\unicode[STIX]{x1D6E4}\lesssim \unicode[STIX]{x1D6E4}_{F}$
, there is a range of
$k$
in the vicinity of
$k_{F}$
, called the Faraday window, in which the frequency is locked to half the driving frequency (subharmonic response). Waves within this window are standing Faraday waves, while waves outside this window are dispersive travelling waves. The Faraday waves are long lived, and their damping time is proportional to the so-called distance to threshold
${\mathcal{M}}=\unicode[STIX]{x1D6E4}_{F}/(\unicode[STIX]{x1D6E4}_{F}-\unicode[STIX]{x1D6E4})$
, as already observed in previous studies (Eddi et al.
Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011). The present theory further predicts (see (2.65)) that the damping time is approximately inversely proportional to the fluid viscosity.
The impulse response of the surface waves has been calculated in § 2.3. Similarly to previous models (e.g. (Eddi et al.
Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011; Moláček & Bush Reference Moláček and Bush2013; Milewski et al.
Reference Milewski, Galeano-Rios, Nachbin and Bush2015)), its spatial dependence involves a Bessel function
$\text{J}_{0}(k_{F}r)$
and its time evolution is a damped oscillation at frequency
$\unicode[STIX]{x1D6FA}/2$
(subharmonic response). Some of these previous models refer to a similar exponential damping in space. We show here that this damping is not strictly present in the impulse response. Instead, the wave amplitude is modulated by an exponential whose argument scales as
$-r^{2}/(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})$
. This modulation captures the diffusive propagation of the impulse excitation to the entire bath. A direct consequence is that two walkers do not instantly feel the impacts of each other. The time oscillation of the waves is phase shifted by
$\unicode[STIX]{x03C0}/4$
compared to the shaker signal. In other terms, the wave elevation
$\unicode[STIX]{x1D701}(r,t)$
is identically zero when the shaker is at its lowest position. This phase shift, confirmed experimentally, may significantly affect the vertical bouncing dynamics of the walker. Indeed, when the droplet impacts the bath, it feels an absolute acceleration that corresponds to the sum of contributions from the shaker motion and from the overlying phase-shifted Faraday waves.
The wave field of a walker can be constructed by adding impulse responses corresponding to each impact of the droplet. This calculation has here been done in the case of a single walker moving at constant speed along a straight line. Under these assumptions, the vertical bouncing motion is perfectly periodic, and all impulse responses are given the same weight. However, the amplitude of the impulse response is modulated according to the phase of impact
$\unicode[STIX]{x1D70F}_{i}$
relative to the shaker, as predicted by this theory and confirmed experimentally. We expect the vertical dynamics of the walker to be perturbed as soon as additional Faraday waves are present. These Faraday waves may originate from the presence of other walkers, from the vicinity of a boundary, or from a previous passage of the walker at the same location (in the case of very high memory). In each of these cases, the vertical dynamics is not strictly periodic. Consequently, impact phase variations must be taken into account in the construction of the resulting wave field through an appropriately weighted summation of impulse responses. These refinements of the temporal description of Faraday waves are likely to be an additional source of chaos in the dynamics of walkers.
Experimental conditions used in three previous double-slit experiments with walkers (Couder & Fort Reference Couder and Fort2006; Andersen et al.
Reference Andersen, Madsen, Reichelt, Ahl, Lautrup, Ellegaard, Levinsen and Bohr2015; Pucci et al.
Reference Pucci, Harris, Faria and Bush2018). Walking speed
$V_{w}$
, path length of the waves
$\sqrt{2}d$
, distance to threshold
${\mathcal{M}}$
, normalised damping time
$MeT_{F}V_{w}/\sqrt{2}d$
, normalised damping length
$l/\sqrt{2}d$
and normalised distance to boundaries
$R/l$
.

The wave of a regular walker is shown to be at least exponentially damped in space. The characteristic damping length
$l(\unicode[STIX]{x1D703})$
is calculated (while it was experimentally estimated in previous works, e.g. in Eddi et al. (Reference Eddi, Sultan, Moukhtar, Fort, Rossi and Couder2011)). It increases with
${\mathcal{M}}$
and reaches a finite asymptotic value for
${\mathcal{M}}\rightarrow \infty$
. It also decreases with increasing speed or viscosity. This damping length plays a crucial role for the walker dynamics, especially in the landmark double-slit experiment initially performed with walkers by Couder & Fort (Reference Couder and Fort2006) and latter contested by Andersen et al. (Reference Andersen, Madsen, Reichelt, Ahl, Lautrup, Ellegaard, Levinsen and Bohr2015). In this experiment, a walking droplet passing to one slit has to possibly interfere with the waves that it sends to the other slit. Such interference is only possible if the waves emitted by the droplet before it crosses the slits are still significantly present when they are received by the droplet beyond the slits. This requires the damping time to be larger than the time required to cross the slits, and the damping length should then be significantly larger than the path length that they have to travel. If the slits are separated by a distance
$d$
, the damping length should then be larger than
$\sqrt{2}d$
. Experimental conditions of Couder & Fort (Reference Couder and Fort2006), Andersen et al. (Reference Andersen, Madsen, Reichelt, Ahl, Lautrup, Ellegaard, Levinsen and Bohr2015), Pucci et al. (Reference Pucci, Harris, Faria and Bush2018) are reported in table 2. As a matter of fact, none of them, except the very recent one of Pucci et al. (Reference Pucci, Harris, Faria and Bush2018), satisfied the two conditions for significant interferences: the damping time was always lower than the time to cross the slit, and the damping length was not much larger than
$d$
. In order to possibly observe interference patterns with single walkers, one would have to reproduce this experiment with a significantly higher memory (i.e. closer to Faraday threshold) and with a fluid of significantly lower viscosity. Boundaries other than those that generate the expected diffraction must be placed farther than the damping length. So for example, the radius
$R$
of the reservoir must satisfy
$R\gg l$
, which was not the case in the work of Pucci et al. (Reference Pucci, Harris, Faria and Bush2018). More generally, the damping length represents the distance at which the walker can probe its surrounding, would there be a boundary or other walkers. It is therefore a crucial parameter to the investigation of many other experiments on walkers, including those involving surreal trajectories, tunnelling or interactions between several walkers.
By contrast with walkers and Faraday waves, quantum mechanics is intrinsically non-dissipative. This feature can conceptually be approached with the walkers, by considering the limit of vanishing viscosity. In appendix F, we show that the waves generated by walkers remain linear in this limit, even though the thickness of the viscous boundary layer decreases to zero. In § 2.5, we have provided developments in series in the low-viscosity limit. If viscosity vanished (i.e. here
$\unicode[STIX]{x1D6FE}_{0}\rightarrow 0$
), the Faraday threshold
$\unicode[STIX]{x1D6E4}_{F}$
would vanish too, so the amplitude of the required driving vibration could be arbitrarily small. The memory
$Me$
would diverge and excitations from the bouncing droplets would be remembered by the wave forever. The diffusion coefficient
$D$
and the damping length
$l$
would diverge too, so each droplet impact would instantly modify the whole bath surface (with the compressibility of the liquid as a limit), i.e. each walker would instantly feel the presence of others, no matter how far away they are. Therefore, in the limit of vanishing viscosity, walkers become intrinsically non-local objects composed of a local particle coupled to a non-local wave field.
Walkers currently represent the only realisation at the macroscopic scale of a pilot-wave system similar to the one proposed by de Broglie (Reference de Broglie1987) to interpret quantum mechanics (Bush Reference Bush2015). Many phenomena initially thought to be specific to the quantum realm have been observed with these walkers. The theory of Faraday waves developed in this paper brings a rigorous framework for the modelling of the wave field generated by a walker. It comprises additions to the previous Faraday wave models that are essential to the description of walkers interfering with others, with boundaries or with their past. It also provides insights on the locality of a walker, which has to be understood before experiments on the possible entanglement of two walkers could potentially be imagined. We hope that that this paper will help the building of a consistent wave theory to describe walkers similarly to Schrödinger’s equation for quantum objects. It would then better reveal the limits of the analogy between walkers and the quantum world.
Acknowledgements
The authors thank N. Sampara and B. Begasse for their help in building the experiment, as well as R. Dubertrand and W. Struyve for fruitful discussions. This work was financially supported by the ‘Actions de Recherches Concertees (ARC)’ of the Belgium Wallonia Brussels Federation under contract No. 12-17/02 and by the FNRS grant number CHAR. RECH.-1.B423.18 (Tadrist L.).
Appendix A. Truncation of the Floquet ansatz
The (2.20) of surface waves in the absence of external pressure
$\unicode[STIX]{x1D6F1}_{k}=0$
is recalled here:
It can be solved with the Floquet ansatz (2.27):
which directly implies that
Substitution of (A 2) and (A 3) in (A 1) yields
The Heaviside cover-up method is then applied to obtain a linear system of infinite dimension for the Floquet coefficients
$\unicode[STIX]{x1D701}_{k}^{(l)}$
:
or in matrix form
$$\begin{eqnarray}\displaystyle \underbrace{\left(\begin{array}{@{}cccccccc@{}}\ddots & \ddots & \ddots & 0 & 0 & 0 & 0 & 0\\ 0 & \unicode[STIX]{x1D6FC} & f(-3\text{i}) & \unicode[STIX]{x1D6FC} & 0 & 0 & 0 & 0\\ 0 & 0 & \unicode[STIX]{x1D6FC} & f(-\text{i}) & \unicode[STIX]{x1D6FC} & 0 & 0 & 0\\ 0 & 0 & 0 & \unicode[STIX]{x1D6FC} & f(\text{i}) & \unicode[STIX]{x1D6FC} & 0 & 0\\ 0 & 0 & 0 & 0 & \unicode[STIX]{x1D6FC} & f(3\text{i}) & \unicode[STIX]{x1D6FC} & 0\\ 0 & 0 & 0 & 0 & 0 & \ddots & \ddots & \ddots \end{array}\right)}_{K}\left(\begin{array}{@{}c@{}}\vdots \\ \unicode[STIX]{x1D701}^{(-1)}\\ \unicode[STIX]{x1D701}^{(0)}\\ \unicode[STIX]{x1D701}^{(1)}\\ \unicode[STIX]{x1D701}^{(2)}\\ \vdots \\ \end{array}\right)=\left(\begin{array}{@{}c@{}}\vdots \\ 0\\ 0\\ 0\\ 0\\ \vdots \\ \end{array}\right),\quad & & \displaystyle\end{eqnarray}$$
where we denoted
$\unicode[STIX]{x1D6FC}=2\unicode[STIX]{x1D6E4}gk/\unicode[STIX]{x1D6FA}^{2}$
for readability. The matrix is tridiagonal, which indicates that each Floquet mode is coupled to its neighbour modes. The condition of existence of surface waves is bound to the obtention of a non-trivial solution to this system. This can only be achieved when the determinant
$\det K=0$
, which yields the threshold
$\unicode[STIX]{x1D6E4}_{k}$
associated with each wavenumber
$k$
for the non-truncated Floquet ansatz.
In the main text, we truncated the series to the two first modes, which led to (2.29) and (2.30). The approximation of this truncation may be evaluated by considering the presence of the two next modes. With a truncation enlarged to four modes, the linear system becomes
$$\begin{eqnarray}\displaystyle \left(\begin{array}{@{}cccc@{}}f_{k}(-3\text{i}) & \unicode[STIX]{x1D6FC} & 0 & 0\\ \unicode[STIX]{x1D6FC} & f_{k}(-\text{i}) & \unicode[STIX]{x1D6FC} & 0\\ 0 & \unicode[STIX]{x1D6FC} & f_{k}(\text{i}) & \unicode[STIX]{x1D6FC}\\ 0 & 0 & \unicode[STIX]{x1D6FC} & f_{k}(3\text{i})\end{array}\right)\left(\begin{array}{@{}c@{}}\unicode[STIX]{x1D701}^{(-1)}\\ \unicode[STIX]{x1D701}^{(0)}\\ \unicode[STIX]{x1D701}^{(1)}\\ \unicode[STIX]{x1D701}^{(2)}\\ \end{array}\right)=\left(\begin{array}{@{}c@{}}0\\ 0\\ 0\\ 0\\ \end{array}\right). & & \displaystyle\end{eqnarray}$$
A non-trivial solution is found when the determinant is zero, i.e.
The curve
$\unicode[STIX]{x1D6E4}_{k}$
is given by the smallest positive solution to this equation, namely,
$$\begin{eqnarray}\displaystyle \frac{\unicode[STIX]{x1D6FC}}{|f_{k}(3\text{i})|}=\frac{1}{\sqrt{2}}\sqrt{\left(1+\frac{f_{k}(-\text{i})}{f_{k}(3\text{i})}+\frac{f_{k}(\text{i})}{f_{k}(-3\text{i})}\right)-\sqrt{\left(1+\frac{f_{k}(-\text{i})}{f_{k}(3\text{i})}+\frac{f_{k}(\text{i})}{f_{k}(-3\text{i})}\right)^{2}-4\frac{|f_{k}(\text{i})|^{2}}{|f_{k}(3\text{i})|^{2}}}}. & & \displaystyle \nonumber\\ \displaystyle & & \displaystyle\end{eqnarray}$$
Around
$k_{F}$
, in the low-viscosity limit (cf. appendix B), one can show that for
$k\sim k_{F}$
,
$|f_{k}(\text{i})|^{2}\sim 4\unicode[STIX]{x1D6FE}_{k}^{2}$
and
$|f_{k}(3\text{i})|_{F}^{2}\sim 64$
. Consequently,
$|f(\text{i})|_{F}^{2}/|f(3\text{i})^{2}|_{F}\sim \unicode[STIX]{x1D6FE}_{F}^{2}/16\ll 1$
. Equation (A 9) can then be expanded in Taylor series, which yields
The correction to
$\unicode[STIX]{x1D6E4}_{k}$
corresponding to the addition of four terms instead of two in the Floquet ansatz is of the order of
$|f(\text{i})|^{2}/2|f(3\text{i})|^{2}\sim \unicode[STIX]{x1D6FE}_{F}^{2}/32<1\,\%$
in the limit of small viscosity and for
$k\sim k_{F}$
.
Figure 9 represents the curve
$\unicode[STIX]{x1D6E4}_{k}$
, calculated with a truncation to 2 modes, to 4 modes and to 20 modes respectively. The first two modes already provide a very good estimation of the curve
$\unicode[STIX]{x1D6E4}_{k}$
, especially for
$k\simeq k_{F}$
. The curves
$\unicode[STIX]{x1D6E4}_{k}$
for four and twenty modes are almost indistinguishable. For walker parameters (cf. table 1),
$\unicode[STIX]{x1D6E4}_{F}=4.155$
with a truncation to 20 modes whereas
$\unicode[STIX]{x1D6E4}_{F}=4.144$
with a truncation to only 2 modes. Similarly, the Faraday wavelength is
$\unicode[STIX]{x1D706}_{F}=4.749~\text{mm}$
with 20 modes, instead of
$\unicode[STIX]{x1D706}_{F}=4.706$
with two modes. In both cases, the relative error from the truncation to two modes is less than 1 %.
Faraday threshold
$\unicode[STIX]{x1D6E4}_{k}$
of each wavenumber
$k$
. The dashed line represents a truncation to the two first modes. The solid line corresponds to the truncation to either four or twenty modes – which are indistinguishable from each other.

Appendix B. Low-viscosity asymptotic behaviour
B.1 Preliminary calculations
The function
$f_{k}$
and its derivatives are first recalled:
They are all self-conjugated, so e.g.
$f_{k}(s^{\ast })=f_{k}(s)^{\ast }$
.
In order to evaluate them in
$s=\pm i$
, we first calculate
Only the positive real part is considered, since the vertical fluid velocity must remain finite when
$z\longrightarrow -\infty$
in (2.19).
Therefore, without any approximation,
When the viscosity is sufficiently small,
$\unicode[STIX]{x1D6FE}_{k}\ll 1$
, and both
$f_{k}(i)$
and its derivatives can be approximated with a Taylor series in
$\unicode[STIX]{x1D6FE}_{k}$
:
The following combinations of
$f_{k}(\text{i})$
and its derivatives are regularly called in § 2:
$$\begin{eqnarray}\displaystyle |f_{k}(\text{i})|^{2} & = & \displaystyle (\unicode[STIX]{x1D714}_{k}^{2}-1)^{2}-2(\unicode[STIX]{x1D714}_{k}^{2}-1)\unicode[STIX]{x1D6FE}_{k}^{3/2}+2(\unicode[STIX]{x1D714}_{k}^{2}+1)\unicode[STIX]{x1D6FE}_{k}^{2}\nonumber\\ \displaystyle & & \displaystyle -\,\left(4+\frac{\unicode[STIX]{x1D714}_{k}^{2}-1}{2}\right)\unicode[STIX]{x1D6FE}_{k}^{5/2}+2\unicode[STIX]{x1D6FE}_{k}^{3}+O(\unicode[STIX]{x1D6FE}_{k}^{7/2}),\end{eqnarray}$$
$$\begin{eqnarray}\displaystyle & & \displaystyle \frac{f_{k}(\text{i})\ddot{f}_{k}(-\text{i})+f_{k}(-\text{i})\ddot{f}_{k}(\text{i})}{2}+{\dot{f}}_{k}(\text{i}){\dot{f}}_{k}(-\text{i})=2(\unicode[STIX]{x1D714}_{k}^{2}+1)\nonumber\\ \displaystyle & & \displaystyle \quad -\,\frac{\unicode[STIX]{x1D714}_{k}^{2}-1}{4}{\unicode[STIX]{x1D6FE}_{k}}^{3/2}+6\unicode[STIX]{x1D6FE}_{k}^{2}+\frac{3\unicode[STIX]{x1D714}_{k}^{2}-59}{16}\unicode[STIX]{x1D6FE}_{k}^{5/2}+O(\unicode[STIX]{x1D6FE}_{k}^{7/2}).\end{eqnarray}$$
Finally, the two first
$k$
-derivative of
$|f_{k}(\text{i})|^{2}$
, useful for the calculation of the Faraday threshold and the diffusion coefficient, are expanded as:
$$\begin{eqnarray}\displaystyle \frac{k}{2}\frac{\unicode[STIX]{x2202}|f_{k}(\text{i})|^{2}}{\unicode[STIX]{x2202}k} & = & \displaystyle 2k(\unicode[STIX]{x1D714}_{k}^{2}-1)\unicode[STIX]{x1D714}_{k}c_{k}-(2k\unicode[STIX]{x1D714}_{k}c_{k}+3(\unicode[STIX]{x1D714}_{k}^{2}-1))\unicode[STIX]{x1D6FE}_{k}^{3/2}+(2k\unicode[STIX]{x1D714}_{k}c_{k}+4(\unicode[STIX]{x1D714}_{k}^{2}+1))\unicode[STIX]{x1D6FE}_{k}^{2}\nonumber\\ \displaystyle & & \displaystyle -\,\left[\frac{k\unicode[STIX]{x1D714}_{k}c_{k}}{2}+\frac{5}{2}\left(4+\frac{\unicode[STIX]{x1D714}_{k}^{2}-1}{2}\right)\right]\unicode[STIX]{x1D6FE}_{k}^{5/2}+6\unicode[STIX]{x1D6FE}_{k}^{3}+O(\unicode[STIX]{x1D6FE}_{k}^{7/2}),\end{eqnarray}$$
$$\begin{eqnarray}\displaystyle \frac{\unicode[STIX]{x2202}^{2}|f_{k}(\text{i})|^{2}}{\unicode[STIX]{x2202}k^{2}} & = & \displaystyle 8\unicode[STIX]{x1D714}_{k}^{2}c_{k}^{2}+4(\unicode[STIX]{x1D714}_{k}^{2}-1)(c_{k}^{2}+\unicode[STIX]{x1D714}_{k}{\dot{c}}_{k})\nonumber\\ \displaystyle & & \displaystyle -\,\left(4(c_{k}^{2}+\unicode[STIX]{x1D714}_{k}{\dot{c}}_{k})+\frac{12}{k^{2}}(\unicode[STIX]{x1D714}_{k}^{2}-1)+\frac{24\unicode[STIX]{x1D714}_{k}c_{k}}{k}\right)\unicode[STIX]{x1D6FE}_{k}^{3/2}+O(\unicode[STIX]{x1D6FE}_{k}^{2}),\end{eqnarray}$$
where
$c_{k}=\unicode[STIX]{x2202}\unicode[STIX]{x1D714}_{k}/\unicode[STIX]{x2202}k$
is the dimensionless group velocity, and
$\dot{c_{k}}=\unicode[STIX]{x2202}c_{k}/\unicode[STIX]{x2202}k$
.
The low-viscosity approximation of the damping of Faraday waves reads
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D6FF}_{k}^{+} & = & \displaystyle -\frac{\left((\unicode[STIX]{x1D714}_{k}^{2}-1)^{2}-2(\unicode[STIX]{x1D714}_{k}^{2}-1)\unicode[STIX]{x1D6FE}_{k}^{3/2}+2(\unicode[STIX]{x1D714}_{k}^{2}+1)\unicode[STIX]{x1D6FE}_{k}^{2}-\left(4+\displaystyle \frac{\unicode[STIX]{x1D714}_{k}^{2}-1}{2}\right)\unicode[STIX]{x1D6FE}_{k}^{5/2}-(2\unicode[STIX]{x1D6E4}gk/\unicode[STIX]{x1D6FA}^{2})^{2}\right)}{4(\unicode[STIX]{x1D714}_{k}^{2}+1)\unicode[STIX]{x1D6FE}_{k}-(\unicode[STIX]{x1D714}_{k}^{2}+3)\unicode[STIX]{x1D6FE}_{k}^{3/2}-(\unicode[STIX]{x1D714}_{k}^{2}+3)\unicode[STIX]{x1D6FE}_{k}^{5/2}/4}\nonumber\\ \displaystyle & & \displaystyle +\,O(\unicode[STIX]{x1D6FE}_{k}^{3}).\end{eqnarray}$$
B.2 Determination of
$k_{F}$
,
$\unicode[STIX]{x1D714}_{F}$
and
$c_{F}$
The most unstable wavenumber
$k_{F}$
is defined as the solution of
According to the approximations (B 17) and (B 20), it satisfies
$$\begin{eqnarray}\displaystyle & & \displaystyle (\unicode[STIX]{x1D714}_{F}^{2}-1)(\unicode[STIX]{x1D714}_{F}^{2}-1-2k_{F}\unicode[STIX]{x1D714}_{F}c_{F})+(\unicode[STIX]{x1D714}_{F}^{2}-1+2k_{F}\unicode[STIX]{x1D714}_{F}c_{F})\unicode[STIX]{x1D6FE}_{F}^{3/2}\nonumber\\ \displaystyle & & \displaystyle \quad -\,(2(\unicode[STIX]{x1D714}_{F}^{2}+1)+2k_{F}\unicode[STIX]{x1D714}_{F}c_{F})\unicode[STIX]{x1D6FE}_{F}^{2}+\left(6+\frac{3}{4}(\unicode[STIX]{x1D714}_{F}^{2}-1)+\frac{k_{F}\unicode[STIX]{x1D714}_{F}c_{F}}{2}\right)\unicode[STIX]{x1D6FE}_{F}^{5/2}\nonumber\\ \displaystyle & & \displaystyle \quad -\,4\unicode[STIX]{x1D6FE}_{F}^{3}+O(\unicode[STIX]{x1D6FE}_{F}^{7/2})=0,\end{eqnarray}$$
where
$\unicode[STIX]{x1D714}_{F}=\unicode[STIX]{x1D714}_{k_{F}}$
and
$c_{F}=c_{k_{F}}$
.
When the viscosity
$\unicode[STIX]{x1D708}$
is identically zero,
$\unicode[STIX]{x1D6FE}_{F}=0$
and (B 24) is simplified
The second factor is strictly negative for
$k_{F}>0$
, so
$k_{F}$
must satisfy
$\unicode[STIX]{x1D714}_{F}^{2}=1$
. We define
for which
$\unicode[STIX]{x1D714}_{0}=\unicode[STIX]{x1D714}_{k_{0}}=1$
,
$c_{0}=c_{k_{0}}$
and
$\unicode[STIX]{x1D6FE}_{0}=\unicode[STIX]{x1D6FE}_{k_{0}}$
.
The damping factor
$\unicode[STIX]{x1D6FE}_{0}=4\unicode[STIX]{x1D708}k_{0}^{2}/\unicode[STIX]{x1D6FA}$
is a better basis for an asymptotic expansion at low viscosity, since it does not directly depend on the Faraday wavenumber
$k_{F}$
, which is still unknown. Equation (B 24) can be rewritten in terms of
$\unicode[STIX]{x1D6FE}_{0}$
since
$\unicode[STIX]{x1D6FE}_{F}=k_{F}^{2}\unicode[STIX]{x1D6FE}_{0}/k_{0}^{2}$
:
$$\begin{eqnarray}\displaystyle & & \displaystyle (\unicode[STIX]{x1D714}_{F}^{2}-1)(\unicode[STIX]{x1D714}_{F}^{2}-1-2k_{F}\unicode[STIX]{x1D714}_{F}c_{F})+(\unicode[STIX]{x1D714}_{F}^{2}-1+2k_{F}\unicode[STIX]{x1D714}_{F}c_{F})\left(\frac{k_{F}}{k_{0}}\right)^{3}\unicode[STIX]{x1D6FE}_{0}^{3/2}\nonumber\\ \displaystyle & & \displaystyle \quad -\,(2(\unicode[STIX]{x1D714}_{F}^{2}+1)+2k_{F}\unicode[STIX]{x1D714}_{F}c_{F})\left(\frac{k_{F}}{k_{0}}\right)^{4}\unicode[STIX]{x1D6FE}_{0}^{2}+\left(6+\frac{3}{4}(\unicode[STIX]{x1D714}_{F}^{2}-1)+\frac{k_{F}\unicode[STIX]{x1D714}_{F}c_{F}}{2}\right)\left(\frac{k_{F}}{k_{0}}\right)^{5}\unicode[STIX]{x1D6FE}_{0}^{5/2}\nonumber\\ \displaystyle & & \displaystyle \quad -\,4\left(\frac{k_{F}}{k_{0}}\right)^{6}\unicode[STIX]{x1D6FE}_{0}^{3}+O(\unicode[STIX]{x1D6FE}_{0}^{7/2})=0.\end{eqnarray}$$
We formulate the ansatz
where
$\unicode[STIX]{x1D6FC}>0$
, from which we directly infer that
and
Inserting this ansatz in (B 27) yields
from which we deduce
$\unicode[STIX]{x1D6FC}=3/2$
and
$\unicode[STIX]{x1D716}=1/(2c_{0})$
, so
$k_{F}=k_{0}+(1/2c_{0})\unicode[STIX]{x1D6FE}_{0}^{3/2}+o(\unicode[STIX]{x1D6FE}_{0}^{3/2})$
. We can iterate this procedure with a new ansatz
$k_{F}=k_{0}+\unicode[STIX]{x1D6FE}_{0}^{3/2}/(2c_{0})+\unicode[STIX]{x1D716}\unicode[STIX]{x1D6FE}_{0}^{\unicode[STIX]{x1D6FC}}+o(\unicode[STIX]{x1D6FE}_{0}^{\unicode[STIX]{x1D6FC}})$
in order to find higher-order corrections. This yields:
$$\begin{eqnarray}\displaystyle \!\!\left.\begin{array}{@{}c@{}}\displaystyle k_{F}=k_{0}+\frac{1}{2c_{0}}\unicode[STIX]{x1D6FE}_{0}^{3/2}-\frac{2+k_{0}c_{0}}{2k_{0}c_{0}^{2}}\unicode[STIX]{x1D6FE}_{0}^{2}+\frac{12+k_{0}c_{0}}{8k_{0}c_{0}^{2}}\unicode[STIX]{x1D6FE}_{0}^{5/2}+\left(\frac{1}{4k_{0}c_{0}^{2}}-\frac{d_{0}}{8c_{0}^{3}}\right)\unicode[STIX]{x1D6FE}_{0}^{3}+O(\unicode[STIX]{x1D6FE}_{0}^{7/2}),\\ \displaystyle \unicode[STIX]{x1D714}_{F}^{2}=1+\unicode[STIX]{x1D6FE}_{0}^{3/2}-\frac{2+k_{0}c_{0}}{k_{0}c_{0}}\unicode[STIX]{x1D6FE}_{0}^{2}+\frac{12+k_{0}c_{0}}{4k_{0}c_{0}}\unicode[STIX]{x1D6FE}_{0}^{5/2}+O(\unicode[STIX]{x1D6FE}_{0}^{3}),\\ \displaystyle \unicode[STIX]{x1D6FE}_{F}=\unicode[STIX]{x1D6FE}_{0}+\frac{\unicode[STIX]{x1D6FE}_{0}^{5/2}}{k_{0}c_{0}}+O(\unicode[STIX]{x1D6FE}_{0}^{3}),\end{array}\right\} & & \displaystyle \nonumber\\ \displaystyle & & \displaystyle\end{eqnarray}$$
where
$d_{0}=(\unicode[STIX]{x2202}^{2}\unicode[STIX]{x1D714}_{k}^{2}/\unicode[STIX]{x2202}k^{2})(k_{0})$
.
From there, we evaluate:
$$\begin{eqnarray}\displaystyle \left.\begin{array}{@{}c@{}}\displaystyle \text{Re}(f_{F}(\text{i}))=\frac{-2}{k_{0}c_{0}}\unicode[STIX]{x1D6FE}_{0}^{2}+\frac{3}{k_{0}c_{0}}\unicode[STIX]{x1D6FE}_{0}^{5/2}+O(\unicode[STIX]{x1D6FE}_{0}^{3}),\\ \displaystyle \text{Im}(f_{F}(\text{i}))=2\unicode[STIX]{x1D6FE}_{0}-\unicode[STIX]{x1D6FE}_{0}^{3/2}+\frac{8+k_{0}c_{0}}{4k_{0}c_{0}}\unicode[STIX]{x1D6FE}_{0}^{5/2}+O(\unicode[STIX]{x1D6FE}_{0}^{3}),\\ \displaystyle f_{F}(\text{i}){\dot{f}}_{F}(-\text{i})+f_{F}(-\text{i}){\dot{f}}_{F}(\text{i})=8\unicode[STIX]{x1D6FE}_{0}-4\unicode[STIX]{x1D6FE}_{0}^{3/2}+O(\unicode[STIX]{x1D6FE}_{0}^{2}),\\ \displaystyle \frac{f_{F}(\text{i})\ddot{f}_{F}(-\text{i})+f_{F}(-\text{i})\ddot{f}_{F}(\text{i})}{2}+{\dot{f}}_{F}(\text{i}){\dot{f}}_{F}(-\text{i})=2\unicode[STIX]{x1D6FE}_{0}^{3/2}+4\left(1-\frac{1}{k_{0}c_{0}}\right)\unicode[STIX]{x1D6FE}_{0}^{2}\\ \displaystyle +\,\frac{12+9k_{0}c_{0}}{2k_{0}c_{0}}\unicode[STIX]{x1D6FE}_{0}^{5/2}+O(\unicode[STIX]{x1D6FE}_{0}^{3}),\\ \displaystyle \left.\frac{\unicode[STIX]{x2202}^{2}|f_{k}(\text{i})|^{2}}{\unicode[STIX]{x2202}k^{2}}\right|_{F}=8c_{0}^{2}+O(\unicode[STIX]{x1D6FE}_{0}^{3/2}).\end{array}\right\}\quad & & \displaystyle\end{eqnarray}$$
Appendix C. Inverse Hankel transform
In this appendix, we evaluate the integral (2.52), namely
with the saddle-node method. First,
$\unicode[STIX]{x1D6FF}_{k}^{+}$
is replaced by its Taylor series around
$k_{F}$
:
Substitution in the integral yields
In the limit
$k_{F}r\ll 1$
, the Bessel function does not oscillate significantly. Then for
$\unicode[STIX]{x1D70F}\gg \unicode[STIX]{x1D70F}_{i}$
,
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D701}_{1}(r,\unicode[STIX]{x1D70F}) & \simeq & \displaystyle \text{e}^{\unicode[STIX]{x1D6FF}_{F}(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})}B_{F}^{+}\cos (\unicode[STIX]{x1D70F}+\unicode[STIX]{x1D703}_{F}^{+})\text{J}_{0}(k_{F}r)k_{F}\int _{0}^{\infty }\text{e}^{-D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})(k-k_{F})^{2}}\,\text{d}k\nonumber\\ \displaystyle & = & \displaystyle B_{F}^{+}k_{F}\text{e}^{\unicode[STIX]{x1D6FF}_{F}(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})}\cos (\unicode[STIX]{x1D70F}+\unicode[STIX]{x1D703}_{F}^{+})\sqrt{\frac{\unicode[STIX]{x03C0}}{D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})}}\text{J}_{0}(k_{F}r).\end{eqnarray}$$
For
$k_{F}r\gg 1$
, the approximation of
$\text{J}_{0}$
by
$\text{Re}\{\text{e}^{\text{i}kr+\text{i}\unicode[STIX]{x03C0}/4}/\sqrt{\unicode[STIX]{x03C0}kr/2}\}$
yields
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D701}_{1}(r,\unicode[STIX]{x1D70F}) & \simeq & \displaystyle \text{e}^{\unicode[STIX]{x1D6FF}_{F}(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})}\int _{0}^{\infty }B_{k}^{+}\cos (\unicode[STIX]{x1D70F}+\unicode[STIX]{x1D703}_{k}^{+})\text{e}^{-D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})(k-k_{F})^{2}}\text{e}^{\text{i}kr+\text{i}\unicode[STIX]{x03C0}/4}\sqrt{\frac{2}{\unicode[STIX]{x03C0}kr}}k\,\text{d}k\nonumber\\ \displaystyle & \simeq & \displaystyle \text{e}^{\unicode[STIX]{x1D6FF}_{F}(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})}B_{F}^{+}\cos (\unicode[STIX]{x1D70F}+\unicode[STIX]{x1D703}_{F}^{+})\text{J}_{0}(k_{F}r)k_{F}\int _{0}^{\infty }\text{e}^{-D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})(k-k_{F})^{2}}\text{e}^{\text{i}(k-k_{F})r}\,\text{d}k.\end{eqnarray}$$
The remaining integral is calculated as
$$\begin{eqnarray}\displaystyle \int _{0}^{\infty }\text{e}^{-D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})(k-k_{F})^{2}}\text{e}^{\text{i}(k-k_{F})r}\,\text{d}k & = & \displaystyle \exp \left[-\frac{r^{2}}{4D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})}\right]\int _{0}^{\infty }\text{e}^{-D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})[k-k_{F}-(\text{i}r)/(2D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i}))]^{2}}\,\text{d}k\nonumber\\ \displaystyle & = & \displaystyle \exp \left[-\frac{r^{2}}{4D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})}\right]\sqrt{\frac{\unicode[STIX]{x03C0}}{D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})}}.\end{eqnarray}$$
Therefore,
Coincidentally, this asymptotic behaviour at
$k_{F}r\gg 1$
is also valid for
$k_{F}r\ll 1$
. Indeed, equations (C 4) and (C 7) are identical as soon as
$\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i}\gg r^{2}/(4D)$
. This condition is necessarily satisfied as soon as
$\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i}\gg 1/(4Dk_{F}^{2})$
for
$k_{F}r\ll 1$
. For walkers, with numerical values in table 1,
$1/4Dk_{F}^{2}\sim 0.08$
so (C 7) is equivalently valid at
$k_{F}r\gg 1$
and
$k_{F}r\ll 1$
.
Appendix D. Surface waves emitted by a walker
The Faraday wave profile resulting from the previous bounces of the walker at the positions
$\boldsymbol{r}_{n}=-(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{n})\boldsymbol{v}$
and the times
$\unicode[STIX]{x1D70F}_{n}=\unicode[STIX]{x1D70F}_{i}+2\unicode[STIX]{x03C0}n$
(
$n\in \mathbb{Z}$
) reads
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D701}_{w}(\boldsymbol{r},\unicode[STIX]{x1D70F}) & = & \displaystyle \mathop{\sum }_{n=-\infty }^{[(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})/2\unicode[STIX]{x03C0}]}\unicode[STIX]{x1D701}_{1}(|\boldsymbol{r}-\boldsymbol{r}_{n}|,\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{n})\nonumber\\ \displaystyle & = & \displaystyle \mathop{\sum }_{n=-\infty }^{[(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})/2\unicode[STIX]{x03C0}]}B_{F}^{+}k_{F}\sqrt{\frac{\unicode[STIX]{x03C0}}{D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{n})}}\text{J}_{0}(k_{F}|\boldsymbol{r}-\boldsymbol{r}_{n}|)\cos (\unicode[STIX]{x1D70F}+\unicode[STIX]{x1D703}_{F}^{+})\nonumber\\ \displaystyle & & \displaystyle \times \,\exp \left(-\frac{\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{n}}{2\unicode[STIX]{x03C0}Me}-\frac{|\boldsymbol{r}-\boldsymbol{r}_{n}|^{2}}{4D(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{n})}\right),\end{eqnarray}$$
where
$[(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})/2\unicode[STIX]{x03C0}]$
denotes the largest integer smaller than
$(\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{i})/2\unicode[STIX]{x03C0}$
, and where the origin
$\boldsymbol{r}=0$
is placed at the position of the walker at time
$\unicode[STIX]{x1D70F}$
. We replace the sum in the previous equation by an integration over the delay time
$u=\unicode[STIX]{x1D70F}-\unicode[STIX]{x1D70F}_{n}$
going from
$0$
to
$\infty$
. This is justified for
$Mer^{2}\gg 2\unicode[STIX]{x03C0}D$
, since the sum can then be seen as an approximate Riemann integral where the interpolated function does not significantly vary over one bin.
The Graf’s theorem for zeroth-order Bessel function reads
with
Since Bessel functions are defined as
and since
$\sum [\text{e}^{\text{i}(\unicode[STIX]{x1D703}-\unicode[STIX]{x1D711}-\unicode[STIX]{x1D711}^{\prime })}]^{n}=\unicode[STIX]{x1D6FF}(\unicode[STIX]{x1D703}-\unicode[STIX]{x1D711}-\unicode[STIX]{x1D711}^{\prime })$
,
Substitution in (D 1) yields
$$\begin{eqnarray}\displaystyle & & \displaystyle \unicode[STIX]{x1D701}_{w}(\boldsymbol{r},\unicode[STIX]{x1D70F})=\frac{B_{F}^{+}k_{F}}{2\sqrt{\unicode[STIX]{x03C0}D}}\cos (\unicode[STIX]{x1D70F}+\unicode[STIX]{x1D703}_{F}^{+})\int _{-\unicode[STIX]{x03C0}}^{\unicode[STIX]{x03C0}}\,\text{d}\unicode[STIX]{x1D711}\exp [\text{i}k_{F}r\sin (\unicode[STIX]{x1D703}-\unicode[STIX]{x1D711})]\nonumber\\ \displaystyle & & \displaystyle \quad \times \,\int _{0}^{\infty }\frac{1}{\sqrt{u}}\exp \left(-\frac{u}{2\unicode[STIX]{x03C0}M\text{e}}-\frac{r^{2}}{4Du}-\frac{vr\cos \unicode[STIX]{x1D703}}{2D}-\frac{v^{2}u}{4D}+\text{i}k_{F}uv\sin \unicode[STIX]{x1D711}\right)\text{d}u.\end{eqnarray}$$
Using the identity
for
$\text{Re}(\unicode[STIX]{x1D6FC})>0$
and
$\unicode[STIX]{x1D6FD}>0$
, we obtain
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D701}_{w}(\boldsymbol{r},\unicode[STIX]{x1D70F}) & = & \displaystyle \frac{B_{F}^{+}k_{F}}{2\sqrt{D}}\cos (\unicode[STIX]{x1D70F}+\unicode[STIX]{x1D703}_{F}^{+})\exp \left(-\frac{vr\cos \unicode[STIX]{x1D703}}{2D}\right)\nonumber\\ \displaystyle & & \displaystyle \times \,\int _{-\unicode[STIX]{x03C0}}^{\unicode[STIX]{x03C0}}\,\text{d}\unicode[STIX]{x1D711}\frac{\exp [\text{i}k_{F}r\sin (\unicode[STIX]{x1D703}-\unicode[STIX]{x1D711})]}{\sqrt{a-\text{i}k_{F}v\sin \unicode[STIX]{x1D711}}}\exp \left(-\frac{r}{\sqrt{D}}\sqrt{a-\text{i}k_{F}v\sin \unicode[STIX]{x1D711}}\right),\end{eqnarray}$$
where
For large
$a$
such that
$k_{F}v/a\ll 1$
(for walkers, with numerical values in table 1, this corresponds to a distance to threshold
${\mathcal{M}}<4.8$
), we can justify the approximation
A reverse application of Graf’s theorem (D 5) yields
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D701}_{w}(\boldsymbol{r},\unicode[STIX]{x1D70F}) & = & \displaystyle \frac{B_{F}^{+}k_{F}}{2\sqrt{Da}}\cos (\unicode[STIX]{x1D70F}+\unicode[STIX]{x1D703}_{F}^{+})\text{J}_{0}\left(k_{F}\left|\boldsymbol{r}+\frac{r}{2\sqrt{Da}}\boldsymbol{v}\right|\right)\nonumber\\ \displaystyle & & \displaystyle \times \,\exp \left[-\left(\sqrt{\frac{a}{D}}+\frac{v\cos \unicode[STIX]{x1D703}}{2D}\right)r\right],\end{eqnarray}$$
i.e. we obtain a Doppler-shifted Bessel profile which is exponentially attenuated in space. The damping length scale, defined as
$l(\unicode[STIX]{x1D703})$
, depends on the orientation
$\unicode[STIX]{x1D703}$
with respect to the walking direction
$\boldsymbol{v}/v$
:
In the special case of a stationary bouncer with
$v=0$
, we obtain the spatial damping length
$l=\sqrt{2\unicode[STIX]{x03C0}MeD}$
which diverges in the limit
$Me\rightarrow \infty$
.
An accurate analytical evaluation of (2.57) in the more general case
$k_{F}v/a\gtrsim 1$
is more involved and beyond the scope of this work. However, we can straightforwardly provide an upper limit for the wave amplitude which holds in any case, namely through the inequality
obtained by substituting
$|\text{J}_{0}(k_{F}|\boldsymbol{r}-\boldsymbol{r}_{n}|)|\leqslant 1$
within (D 1).
Appendix E. Experimental measurement of the impact phase
Measurement of the impact time
$t_{i}$
, from the end time of the recording
$t_{f}$
and the synchronisation delay
$\unicode[STIX]{x0394}t$
. Definitions of
$t_{i}$
,
$t_{f}$
and
$\unicode[STIX]{x0394}t$
are provided in the main text.

Measuring the impact phase requires a good synchronisation of the high-speed camera with the vibration
$g(t)$
of the shaker. For this purpose, the trigger of the camera that marks the end of the recording is also sent to the DACQ. Both signals are simultaneously processed with a Labview interface that outputs the time delay
$\unicode[STIX]{x0394}t\in [0,T_{F}/2]$
from the last recorded maximum of
$g(t)$
to the end of the recording.
We then define the absolute origin of time
$t=0$
as the last maximum of
$g(t)$
before impact such that the end time of the recording is given by
in this reference time (figure 10). The impact time
$t_{i}\in [0,T_{F}]$
is then calculated as
where
$X[T_{F}]$
means
$X$
modulo
$T_{F}$
.
The impact phase
$\unicode[STIX]{x1D70F}_{i}$
is then
Practically,
$t_{f}-t_{i}$
is computed as a number of frames divided by the acquisition frequency.
The digital output of the DACQ is refreshed at 8 kHz, so the error on
$\unicode[STIX]{x0394}t$
is less than 0.125 ms. The impact is detected with an accuracy of (
$\pm 1$
frames), which corresponds to an error on
$t_{f}-t_{i}$
less than 0.25 ms. The absolute error on
$t_{i}$
(respectively
$\unicode[STIX]{x1D70F}_{i}$
) is then less than 0.375 ms (respectively 1.5 %).
Appendix F. Validity of the linear Navier–Stokes equations for walkers
The linear theory of Faraday waves for walkers developed in this manuscript is valid as long as the amplitude of the waves is significantly smaller than both the Faraday wavelength
$\unicode[STIX]{x1D706}_{F}$
and the thickness of the viscous boundary layer. This latter, here denoted
$\unicode[STIX]{x1D6FF}_{\unicode[STIX]{x1D708}}$
, is of the order of
$\sqrt{2\unicode[STIX]{x1D708}/\unicode[STIX]{x1D6FA}}$
. The amplitude of the wave field generated by classical walkers (oil viscosity 20 cSt, forcing frequency 80 Hz) is always less than
$12~\unicode[STIX]{x03BC}\text{m}$
in the vicinity of a single walker, shortly after the impact of the droplet on the bath (Damiano et al.
Reference Damiano, Brun, Harris, Galeano-Rios and Bush2016). This amplitude is therefore more than one order of magnitude smaller than both (i) the considered wavelength
$\unicode[STIX]{x1D706}_{F}\simeq 4.7~\text{mm}$
and (ii) the viscous boundary layer
$\unicode[STIX]{x1D6FF}_{\unicode[STIX]{x1D708}}\simeq 300~\unicode[STIX]{x03BC}\text{m}$
. Since walkers are commonly studied below the Faraday threshold, the amplitude of their wave decreases with time, despite the forcing, so the measured
$12~\unicode[STIX]{x03BC}\text{m}$
can be seen as an upper bound for the wave amplitude.
Data from Eddi (Reference Eddi2011). (a) Log–log scaling of the evolution of the optimal forcing frequency
$f_{0}$
of walkers with kinematic viscosity
$\unicode[STIX]{x1D708}$
. The dashed line is corresponds to
$f_{0}\unicode[STIX]{x1D708}^{1/2}=3.55~\text{m}~\text{s}^{-3/2}$
(
$f_{0}\propto \unicode[STIX]{x1D708}^{-1/2}$
). (b) Log–log scaling of the evolution of the optimal diameter
$d_{0}$
of walking droplets with kinematic viscosity
$\unicode[STIX]{x1D708}$
. The dashed line corresponds to
$d_{0}/\unicode[STIX]{x1D708}^{1/4}=3.46\times 10^{-3}~\text{m}^{1/2}~\text{s}^{1/4}$
(
$d_{0}\propto \unicode[STIX]{x1D708}^{1/4}$
).

The amplitude of the waves right below the impacting droplet might be larger than the measured
$12~\unicode[STIX]{x03BC}\text{m}$
on its sides. This region is challenging to observe by optical means. Nevertheless, a theoretical bound for the deflection of the bath right after impact can be obtained. This deflection mostly originates from the kinetic energy of the falling droplet. At impact, this energy is temporarily converted into surface energy, mostly associated with the bath deflection. A part of this stored energy is then restituted to the translation of the droplet (Molacek & Bush Reference Molacek and Bush2013), another part generates the surface wave accompanying the walker, and a last part is dissipated by viscous flows (within both the droplet and the bath) during impact. An upper bound
$H$
on the bath deflection (and on the subsequent surface waves) can be found by balancing the surface energy of the bath deflection and the kinetic energy of the incoming droplet. According to experiments (e.g. Protiere et al.
Reference Protiere, Boudaoud and Couder2006; Molacek & Bush Reference Molacek and Bush2013), the free fall of the droplet lasts approximately 2/3 of the bouncing period
$2/f$
(where
$f=\unicode[STIX]{x1D6FA}/\unicode[STIX]{x03C0}$
is the forcing frequency). Consequently, the impact speed of the droplet is
$v=2g/(3f)$
. To the leading order, the crater is assumed to be a spherical cap whose curvature matches half the curvature of the droplet (Gopinath & Koch Reference Gopinath and Koch2001). The excess of surface of this spherical cap is approximately
$2\unicode[STIX]{x03C0}H^{2}$
, independently of the droplet radius. The energy balance then yields
with a droplet diameter
$d\simeq 0.8~\text{mm}$
. This pessimistic upper bound
$H$
on the bath deflection, which did not take into account any dissipation or partial transfer of energy, is nevertheless again smaller than both the wavelength
$\unicode[STIX]{x1D706}_{F}$
and the boundary layer thickness
$\unicode[STIX]{x1D6FF}_{\unicode[STIX]{x1D708}}$
.
In the discussion, we suggested that quantum-like phenomena would be likely to appear for large damping length
$l$
, i.e. for vanishing viscosity. However, the boundary layer thickness
$\unicode[STIX]{x1D6FF}_{\unicode[STIX]{x1D708}}$
would then decrease and could possibly become smaller than the wave amplitude, which would invalidate a description of walkers with linear waves. There have been very few experimental attempts at generating walkers with viscosity different from 20 cSt or 50 cSt (Protiere et al.
Reference Protiere, Couder, Fort and Boudaoud2005; Eddi Reference Eddi2011). For each viscosity in the range
$\unicode[STIX]{x1D708}\in [10,100]$
cSt, there is a narrow window of forcing frequency
$f_{0}(\unicode[STIX]{x1D708})\propto \unicode[STIX]{x1D708}^{-1/2}$
and droplet size
$d_{0}(\unicode[STIX]{x1D708})\propto \unicode[STIX]{x1D708}^{1/4}$
in which walkers could be observed (see figure 11). Substituting
$f_{0}$
and
$d_{0}$
in (F 1) yields
$H\propto \unicode[STIX]{x1D708}^{7/8}$
. On the other hand, the boundary layer thickness
$\unicode[STIX]{x1D6FF}_{\unicode[STIX]{x1D708}}$
would scale proportionally to
$\sqrt{\unicode[STIX]{x1D708}/f_{0}}\propto \unicode[STIX]{x1D708}^{3/4}$
. Consequently, we obtain the scaling
$H/\unicode[STIX]{x1D6FF}\propto \unicode[STIX]{x1D708}^{1/8}$
, i.e. the wave amplitude
$H$
decreases faster than the boundary layer thickness
$\unicode[STIX]{x1D6FF}$
when the viscosity is decreased. Therefore, as far as walkers less viscous than
$\unicode[STIX]{x1D708}=20$
cSt are considered,
$H\ll \unicode[STIX]{x1D6FF}_{\unicode[STIX]{x1D708}}$
and the associated Faraday waves can be described with a linear theory.












