1 Introduction
Prandtl (Reference Prandtl1949) was probably the first to demonstrate experimentally a slow steady rotation of the contained liquid during the swirl-type sloshing in an upright circular tank exposed to an orbital horizontal translatory excitation (swirl-type sloshing
$=$
swirling
$=$
azimuthally propagating wave; see, chap. 9 by Faltinsen & Timokha (Reference Faltinsen and Timokha2009)). The azimuthal liquid mass transport was further observed and measured by Hutton (Reference Hutton1964) and Royon-Lebeaud, Hopfinger & Cartellier (Reference Royon-Lebeaud, Hopfinger and Cartellier2007) who suggested that its theoretical description should involve the Lagrangian-mean concept, in general, and the Stokes drift (which is an element of the concept, see, Craik Reference Craik1986; Bühler Reference Bühler2009; van den Bremer & Breivik Reference van den Bremer and Breivik2017), in particular, and that one should account for the free-surface nonlinearity and viscosity. Recent experiments by Reclari (Reference Reclari2013), Ducci & Weheliye (Reference Ducci and Weheliye2014), Reclari et al. (Reference Reclari, Dreyer, Tissot, Obreschkow, Wurm and Farhat2014) and Bouvard, Herreman & Moisy (Reference Bouvard, Herreman and Moisy2017) offered an insight into the Prandtl phenomenon, for both almost inviscid and strongly viscous liquids (Bouvard et al.
Reference Bouvard, Herreman and Moisy2017, conducted model tests with silicon oils), to show that viscosity does not qualitatively matter, at least, when the forcing frequency is close to the lowest natural sloshing frequency, but it affects both a local vortical flow in the neighbourhood of the moving contact line and the boundary-layer flow along the tank wall. However, the free-surface nonlinearity cannot be neglected at the resonance conditions.
Existing experimental works report that the mass transport (a) is co-directed with swirl-type sloshing in each inner point beneath the free surface, (b) tends to zero when approaching the upright wall, (c) reaches the maximum value at approximately the semi-radius and (d) is satisfactorily predicted by the Stokes drift only in the neighbourhood of the tank axis. The peculiarities (a)–(d) require us to account for a non-zero Eulerian-mean (vortical) velocity
$\boldsymbol{w}^{E}$
, whose appearance was extensively discussed, but not analysed in the aforementioned works by Reclari (Reference Reclari2013), Ducci & Weheliye (Reference Ducci and Weheliye2014), Reclari et al. (Reference Reclari, Dreyer, Tissot, Obreschkow, Wurm and Farhat2014), Bouvard et al. (Reference Bouvard, Herreman and Moisy2017). The governing equations for
$\boldsymbol{w}^{E}$
are the starting point of the present paper.
In § 2, we introduce the steady-state potential-flow Narimanov–Moiseev-type solution by Faltinsen, Lukovsky & Timokha (Reference Faltinsen, Lukovsky and Timokha2016) describing the steady-state resonant sloshing in an upright circular tank, which is horizontally excited with a forcing frequency close to the lowest natural sloshing frequency. The lowest-order component of this solution is determined by the four lowest-order amplitude parameters
$a,\bar{a},\bar{b}$
and
$b$
. As follows from Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016), one can introduce the non-dimensional parameter
$\unicode[STIX]{x1D6EF}=ab-\bar{b}\bar{a}$
whose sign determines the existing steady-state wave types: standing (
$\unicode[STIX]{x1D6EF}=0$
), counter- (
$\unicode[STIX]{x1D6EF}>0$
) and clockwise (
$\unicode[STIX]{x1D6EF}<0$
) swirling waves (azimuthal progressive wave). In a frequency range, the steady-state sloshing can be unstable, which causes chaotic (irregular) wave motions. Limitations of the steady-state solution are discussed.
In § 3, the Eulerian-mean (vortical) flow velocity
$\boldsymbol{w}^{E}$
is added to the second-order asymptotic component of the steady-state solution by Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016). The time averaging in the vorticity equation derives the dynamic governing equation for
$\boldsymbol{w}^{E}$
, which is the same as the inviscid Craik–Leibovich equation (Craik & Leibovich Reference Craik and Leibovich1976). In § 4, we evaluate the mean azimuthal mass flux
$M=M^{E}+M^{s}$
, where
$M^{E}$
is caused by
$\boldsymbol{w}^{E}$
, but the second summand
$M^{s}$
, which is not zero for swirling, is in the standard way interpreted as a consequence of the non-Eulerian azimuthal (Stokes-drift type) mass-transport velocity
$\boldsymbol{w}^{s}$
. Mathematical and physical reasons for the difference between
$\boldsymbol{w}^{s}$
(and
$M^{s}$
) and the formal mathematical expression of the Stokes-drift velocity
$\boldsymbol{w}^{S}$
(and the associated mass flux
$M^{S}$
) are discussed. In the forthcoming analysis, the experimentally established mass transport is associated with the sum
$\boldsymbol{w}^{P}=\boldsymbol{w}^{E}+\boldsymbol{w}^{s}$
.
Along with the inviscid Craik–Leibovich equation, the steady solenoidal
$\boldsymbol{w}^{E}$
satisfies the zero normal-velocity conditions on the wall and the mean free surface. Getting a unique
$\boldsymbol{w}^{E}$
requires tangential (azimuthal and vertical) boundary conditions on the wetted tank surface. An indicative prediction of these conditions can be obtained by using the nonlinear boundary layer (steady streaming) theory whose details are described in appendix A in the Supplementary Materials, available at https://doi.org/10.1017/jfm.2019.94. Analysis of these predictions in § 5 shows that they are mathematically contradictory. However, a much more important fact is that they do not provide the zero mass-transport velocity
$\boldsymbol{w}^{P}=\mathbf{0}$
on the wetted tank wall, which was listed above as the peculiarity (b). A physical reason is that the Eulerian steady streaming theory ignores the non-Eulerian mass transport (associated with
$\boldsymbol{w}^{s}$
) around the wall, which influences viscous stresses and may modify tangential boundary conditions.
By assuming that the non-Eulerian mass transport predominates the local tangential mean rotation of fluid particles in the neighbourhood of the wall, one can arrive, in the inviscid flow limit, at the tangential boundary condition
$\boldsymbol{w}^{P}=\mathbf{0}\Rightarrow \boldsymbol{w}^{E}=-\boldsymbol{w}^{s}$
on the wall. The latter condition makes the experimental peculiarity (b) automatically satisfied. It also means that
$\boldsymbol{w}^{E}$
counteracts
$\boldsymbol{w}^{s}$
and, thereby, implies the return flow. Adopting this boundary condition makes the inviscid Craik–Leibovich boundary value problem mathematically correct and allows for the derivation of an exact analytical solution for
$\boldsymbol{w}^{P}$
. As announced in the paper title, this purely inviscid solution neglects specific viscous streams and their influence (feedback) on the Prandtl phenomenon. The inviscid approximation of the azimuthal mass transport is validated in § 6 by using the measurements by Hutton (Reference Hutton1964) and Bouvard et al. (Reference Bouvard, Herreman and Moisy2017).
2 Steady-state asymptotic solution by Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016)
The upright circular container with an infinite liquid depth moves translatory along an horizontal elliptic orbit defined by
$\unicode[STIX]{x1D702}_{1}(t)=\unicode[STIX]{x1D702}_{1a}\cos \unicode[STIX]{x1D70E}t,\,\unicode[STIX]{x1D702}_{2}(t)=\unicode[STIX]{x1D702}_{2a}\sin \unicode[STIX]{x1D70E}t$
. Sloshing is considered in the tank-fixed coordinate system. The panel (a) introduces the original geometric notations but (b) shows the mean (hydrostatic) liquid domain
$Q_{0}$
, unperturbed free surface
$\unicode[STIX]{x1D6F4}_{0}$
and the mean wetted tank surface
$S_{0}$
as they appear in our non-dimensional analysis.

A rigid circular cylindrical container of radius
$r_{0}$
is partly filled by an inviscid incompressible liquid with a fairly deep depth (figure 1
a). Liquid sloshing is considered in the tank-fixed cylindrical coordinate system. The tank performs an orbital (longitudinal, elliptic or circular) small-magnitude horizontal translatory motion governed by
$\unicode[STIX]{x1D702}_{1}(t)=\unicode[STIX]{x1D702}_{1a}\cos \unicode[STIX]{x1D70E}t,\,\unicode[STIX]{x1D702}_{2}(t)=\unicode[STIX]{x1D702}_{2a}\sin \unicode[STIX]{x1D70E}t$
along the
$x$
and
$y$
axes, respectively. The forcing frequency
$\unicode[STIX]{x1D70E}$
is close to the lowest natural sloshing frequency
$\unicode[STIX]{x1D70E}_{1}=\unicode[STIX]{x1D70E}_{11}$
. For the infinite liquid depth, the natural sloshing frequencies are
$\unicode[STIX]{x1D70E}_{Mi}=\sqrt{gk_{Mi}/r_{0}}$
, where
$g$
is the gravitational acceleration and
$k_{Mi}$
are the roots of
$\text{J}_{M}^{\prime }(k_{Mi})=0$
(
$\text{J}_{M}$
is the Bessel function of the first kind).
Assuming irrotational flows of an incompressible perfect liquid, Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016) derived an asymptotic steady-state solution of the nonlinear resonant sloshing problem for finite deep liquid depths
$h$
(
$1.2\lesssim h^{\ast }=h/r_{0}$
). The derivations were based on the Narimanov–Moiseev asymptotic multimodal theory, which is a consequence of the Bateman–Luke variational formalism for the free-surface sloshing problem coupling the velocity potential
$\unicode[STIX]{x1D711}(r,\unicode[STIX]{x1D703},z,t)$
and function
$\unicode[STIX]{x1D701}(r,\unicode[STIX]{x1D703},t)$
describing the free-surface elevations by
$z=\unicode[STIX]{x1D701}(r,\unicode[STIX]{x1D703},t)$
. The constructed asymptotic solution by Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016) exactly satisfies the Laplace equation, boundary conditions on the wetted tank walls, but kinematic and dynamic boundary conditions on the free surface
$\unicode[STIX]{x1D6F4}(t)$
are asymptotically approximated to within the
$o(\unicode[STIX]{x1D716})$
-terms, where the lowest-order surface-wave component has then the order
$O(\unicode[STIX]{x1D716}^{1/3})$
and the highest included asymptotic terms
$O(\unicode[STIX]{x1D716})\ll 1$
are associated with the non-dimensional forcing amplitudes. Why the steady-state resonant sloshing amplitude is of the order
$O(\unicode[STIX]{x1D716}^{1/3})$
while the tank amplitude is of the order
$O(\unicode[STIX]{x1D716})$
was explained by Moiseev (Reference Moiseev1958) (see also details in chaps 8 and 9 by Faltinsen & Timokha (Reference Faltinsen and Timokha2009)).
We utilise the steady-state asymptotic solution by Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016) for
$h^{\ast }\rightarrow \infty$
and re-write it with special normalisation, in which
$1/\unicode[STIX]{x1D70E}$
and
$r_{0}/k$
are the characteristic time and length, respectively, (
$k=k_{11}=1.841183781341\ldots$
). This normalisation simplifies the forthcoming derivations and analytical expressions. The lowest-order terms in the asymptotic free-surface
$\unicode[STIX]{x1D6F4}(t)$
representation,
take then the form
where
and
$a,\bar{a},\bar{b},b=O(\unicode[STIX]{x1D716}^{1/3})$
are the only lowest-order non-dimensional amplitude parameters. The asymptotic velocity field relative to the rigid tank is defined by
where
$\hat{\boldsymbol{r}},\hat{\unicode[STIX]{x1D73D}}$
,
$\hat{\boldsymbol{z}}$
are the unit vectors in the cylindrical coordinate frame. The lowest-order potential-flow velocity component is
$$\begin{eqnarray}\displaystyle \boldsymbol{v}^{(1/3)}(r,\unicode[STIX]{x1D703},z,t) & = & \displaystyle \unicode[STIX]{x1D735}[J_{1}(r)\,\text{e}^{z}(\unicode[STIX]{x1D703}_{c}\cos t+\unicode[STIX]{x1D703}_{s}\sin t)]\nonumber\\ \displaystyle & = & \displaystyle [J_{1}^{\prime }(r)\text{e}^{z}(\unicode[STIX]{x1D703}_{c}\cos t+\unicode[STIX]{x1D703}_{s}\sin t)]\hat{\boldsymbol{r}}+[r^{-1}J_{1}(r)\text{e}^{z}(\unicode[STIX]{x1D703}_{c}^{\prime }\cos t+\unicode[STIX]{x1D703}_{s}^{\prime }\sin t)]\hat{\unicode[STIX]{x1D73D}}\nonumber\\ \displaystyle & & \displaystyle +\,[J_{1}(r)\text{e}^{z}(\unicode[STIX]{x1D703}_{c}\cos t+\unicode[STIX]{x1D703}_{s}\sin t)]\hat{\boldsymbol{z}}=v_{1}^{(1/3)}\hat{\boldsymbol{r}}+v_{2}^{(1/3)}\hat{\unicode[STIX]{x1D73D}}+v_{3}^{(1/3)}\hat{\boldsymbol{z}},\end{eqnarray}$$
where the velocity potential
$\unicode[STIX]{x1D711}^{(1/3)}(r,\unicode[STIX]{x1D703},z)=J_{1}(r)\,\text{e}^{z}(\unicode[STIX]{x1D703}_{c}\cos t+\unicode[STIX]{x1D703}_{s}\sin t)$
is taken from Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016) and
$\unicode[STIX]{x1D703}_{c}^{\prime },\,\unicode[STIX]{x1D703}_{s}^{\prime }$
are derivatives of
$\unicode[STIX]{x1D703}_{c},\,\unicode[STIX]{x1D703}_{s}$
by
$\unicode[STIX]{x1D703}$
. The relative velocity component
$\boldsymbol{v}^{(1/3)}$
is constructed in the body-fixed coordinate system and, therefore, it satisfies the impermeability condition on the wetted tank surface as well as the kinematic and dynamic free-surface boundary conditions (see, details in chaps 2 and 5 by Faltinsen & Timokha (Reference Faltinsen and Timokha2009)).
Depending on the quadratic amplitude quantity
$\unicode[STIX]{x1D6EF}=ab-\bar{a}\bar{b}$
, the free-surface representation (2.2) determines either standing (only possible for longitudinal horizontal excitations) or swirling (azimuthal progressive) wave, i.e.
We construct an inviscid vortical asymptotic solution for the swirl-type sloshing, in which the lowest-order velocity field coincides with (2.5) but a steady vorticity appears in the second-order approximation,
$O(\unicode[STIX]{x1D716}^{2/3})$
. Owing to restrictions of the Narimanov–Moiseev asymptotic theory, proceeding this way implicitly assumes:
(i) the low-viscosity contained liquid, for which the boundary layer thickness at the tank wall,
$\unicode[STIX]{x1D6FF}=1/\sqrt{Re_{s}}\ll 1$
(
$Re_{s}=(r_{0}^{2}\unicode[STIX]{x1D70E})/(\unicode[STIX]{x1D708}k)$
, is the sloshing-related Reynolds number and
$\unicode[STIX]{x1D708}$
is the kinematic viscosity) is smaller than the introduced steady vortical component, i.e. (2.7)
$$\begin{eqnarray}\unicode[STIX]{x1D6FF}={\displaystyle \frac{1}{\sqrt{Re_{s}}}}=\sqrt{{\displaystyle \frac{\unicode[STIX]{x1D708}k}{r_{0}^{2}\unicode[STIX]{x1D70E}}}}\ll O(\unicode[STIX]{x1D716}^{2/3});\end{eqnarray}$$
(ii) the fairly deep liquid depth (practically,
$2.5\lesssim h^{\ast }=h/r_{0}$
to avoid the depth effect on the natural sloshing frequencies);(iii) the forcing frequency
$\unicode[STIX]{x1D70E}$
is close to the lowest natural sloshing frequency
$\unicode[STIX]{x1D70E}_{1}=\unicode[STIX]{x1D70E}_{11}$
, so that the Moiseev detuning condition (2.8a,b )is satisfied, namely, the closeness of the forcing frequency
$$\begin{eqnarray}(\unicode[STIX]{x1D70E}_{1}^{2}/\unicode[STIX]{x1D70E}^{2}-1)=O(\unicode[STIX]{x1D716}^{2/3}),\quad k\sqrt{\unicode[STIX]{x1D702}_{1a}^{2}+\unicode[STIX]{x1D702}_{1a}^{2}}/r_{0}=O(\unicode[STIX]{x1D716})\ll 1\end{eqnarray}$$
$\unicode[STIX]{x1D70E}$
to the first natural sloshing frequency
$\unicode[STIX]{x1D70E}_{1}$
is measured on the
$O(\unicode[STIX]{x1D716}^{2/3})$
-scale, where
$O(\unicode[STIX]{x1D716})$
characterises the forcing amplitude;(iv) there are no secondary resonances, which occur, e.g. in local neighbourhoods of
$\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}=\unicode[STIX]{x1D70E}_{2k}/(2\unicode[STIX]{x1D70E}_{1}),\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}=\unicode[STIX]{x1D70E}_{0k}/(2\unicode[STIX]{x1D70E}_{1}),\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}=\unicode[STIX]{x1D70E}_{3k}/(3\unicode[STIX]{x1D70E}_{1}),\,k\geqslant 1$
and
$\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}=\unicode[STIX]{x1D70E}_{1i}/(3\unicode[STIX]{x1D70E}_{1}),\,i\geqslant 2$
(the secondary resonances were extensively discussed by Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016)).
Remark. Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016, equation (3.12)) derived a system of nonlinear algebraic secular (solvability) equations governing
$a,\bar{a},\bar{b}$
and
$b$
as functions of
$\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}$
. The secular system comes from the third-order (
$O(\unicode[STIX]{x1D716})$
) approximation and, therefore, its coefficients can modify due to the introduced second-order vortical component. However, the structure of (2.2) and (2.5) remains the same and, therefore, we can assume that
$a,\bar{a},\bar{b},b$
are the known amplitude values, but not necessarily the same as computed by Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016). How
$a,\bar{a},\bar{b},b$
change due to the non-zero vortical flow deserves a dedicated study.
3 The Eulerian-mean velocity field
The original asymptotic solution by Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016) implies the zero mean velocity field. In the present section, we assume that, after a long, and, generally, viscous-flow transient wave phase, the velocity field becomes vortical so that the steady-state solution contains the non-zero time-averaged (Eulerian-mean) solenoidal velocity component
such that
This velocity component is associated with the following framed second-order supplement to the original asymptotic solution
To get a governing (dynamic) equation for
$\boldsymbol{w}^{E}$
, we follow the asymptotic (time-averaging) procedure by Craik & Leibovich (Reference Craik and Leibovich1976) (not that in Leibovich Reference Leibovich1980). The procedure uses the vorticity equation written down in the tank-fixed (non-inertial) coordinate frame (see, Kochin, Kibel & Roze Reference Kochin, Kibel and Roze1965; Faltinsen & Timokha Reference Faltinsen and Timokha2009),
where
$\boldsymbol{v}_{O}=k(\dot{\unicode[STIX]{x1D702}}_{1}(t),\dot{\unicode[STIX]{x1D702}}_{2}(t),0)/(r_{0}\unicode[STIX]{x1D70E})=O(\unicode[STIX]{x1D716})$
is the non-dimensional translatory velocity of the tank, which appears in (3.4) because we employ the non-inertial coordinate system, and the dot means differentiation by the time. The time-periodic solution of (3.4) is asymptotically posed as
in which
$\unicode[STIX]{x1D734}^{E}$
is defined in (3.1),
$\boldsymbol{v}^{(1/3)}$
comes from (2.5),
$\boldsymbol{v}^{(2/3)}$
contains the vortical term by (3.3).
Substituting (3.5) into the vorticity equation (3.4) derives the time-periodic (3/3)-order approximation
$$\begin{eqnarray}\displaystyle \unicode[STIX]{x1D734}_{3/3}(r,\unicode[STIX]{x1D703},z,t) & = & \displaystyle \displaystyle \int \unicode[STIX]{x1D735}\times [\boldsymbol{v}^{(1/3)}\times \unicode[STIX]{x1D734}^{E}]\,\text{d}t\nonumber\\ \displaystyle & = & \displaystyle \unicode[STIX]{x1D735}\times [\unicode[STIX]{x1D735}(J_{1}\unicode[STIX]{x1D703}_{c}\text{e}^{z})\times \unicode[STIX]{x1D734}^{E}]\,\sin t-\unicode[STIX]{x1D735}\times [\unicode[STIX]{x1D735}(J_{1}\unicode[STIX]{x1D703}_{s}\text{e}^{z})\times \unicode[STIX]{x1D734}^{E}]\,\cos t.\end{eqnarray}$$
The (4/3)-order asymptotic approximation of the vorticity equation
leads to the time-periodic solution
$\unicode[STIX]{x1D734}_{4/3}$
, if and only if, the time-averaged right-hand side of (3.7) is zero. Tedious derivations (we used Maple
$^{TM}$
to simplify them) show that, if
$\unicode[STIX]{x1D734}^{E}$
is solenoidal (satisfies (3.2)), the time averaging in (3.7) yields the following equation
$$\begin{eqnarray}\unicode[STIX]{x1D735}\times [\boldsymbol{w}^{E}\times \unicode[STIX]{x1D734}^{E}]=\unicode[STIX]{x1D6EF}\,\text{e}^{2z}\Bigg({\displaystyle \frac{1}{r}}w^{S}\,\unicode[STIX]{x2202}_{\unicode[STIX]{x1D703}}\unicode[STIX]{x1D734}^{E}-\hat{\unicode[STIX]{x1D73D}}\Bigg[\underbrace{\bigg((w^{S})^{\prime }-{\displaystyle \frac{1}{r}}w^{S}\bigg)}_{3r^{-1}(-2w^{S}+w^{s})}\unicode[STIX]{x1D6FA}_{1}^{E}+2w^{S}\,\unicode[STIX]{x1D6FA}_{3}^{E}\Bigg]\Bigg),\end{eqnarray}$$
where
$w^{s}(r)=r^{-1}J_{1}^{2}(r)$
and
$w^{S}(r)$
is associated with the formally defined Stokes-drift velocity
$$\begin{eqnarray}\displaystyle \boldsymbol{w}^{S}(r,z) & \mathop{=}^{\text{def}} & \displaystyle \left\langle \displaystyle \int \boldsymbol{v}^{(1/3)}\,\text{d}t\boldsymbol{\cdot }\unicode[STIX]{x1D735}\boldsymbol{v}^{(1/3)}\right\rangle ={\displaystyle \frac{1}{2}}\unicode[STIX]{x1D735}\times \left\langle \boldsymbol{v}^{(1/3)}\times \displaystyle \int \boldsymbol{v}^{(1/3)}\,\text{d}t\right\rangle \nonumber\\ \displaystyle & \equiv & \displaystyle \unicode[STIX]{x1D6EF}\text{e}^{2z}\,\underbrace{{\displaystyle \frac{1}{2r}}\left[J_{1}^{2}(r)+\left(J_{1}^{\prime }-{\displaystyle \frac{J_{1}(r)}{r}}\right)^{2}\right]}_{w^{S}(r)}\hat{\unicode[STIX]{x1D73D}}=w_{2}^{S}(r,z)\hat{\unicode[STIX]{x1D73D}}.\end{eqnarray}$$
(We thank Dr W. Herreman who prompted the use of the second definition of the Stokes drift.)
The derived equation (3.8) is the same as the inviscid Craik–Leibovich equation (Craik & Leibovich Reference Craik and Leibovich1976) but for swirling, that is, equation (3.8), admits the well-known equivalent form
Equivalence of (3.10) and (3.8) is shown in appendix A in the Supplementary Materials.
The governing equations (3.2) and (3.8) with respect to
$\boldsymbol{w}^{E}$
require appropriate boundary conditions, which include, since we employ the tank-fixed coordinate system, the zero normal-velocity conditions on the wall and the mean free surface
as well as
However, equations (3.2) and (3.8) also need tangential boundary conditions on the wetted wall. Where the tangential boundary conditions for the inviscid velocity field
$\boldsymbol{w}^{E}$
could come from is analysed in § 5.
4 The mean mass transport through the meridional plane
Let us forget, for a while, the derivations in the previous section but assume that we know the axisymmetric time-averaged Eulerian (vortical) velocity by (3.1) in each inner point of the time-dependent liquid domain, i.e.
which has already appeared in (3.3). The lowest-order free-surface motion
$\unicode[STIX]{x1D701}^{(1/3)}$
and velocity field
$v^{(1/3)}$
are described by (2.2) and (2.5), respectively.
Simple derivations show that the mean mass fluxes through the horizontal on
$z=z_{1}<0$
and annular
$r=r_{1},\,0<r_{1}<k$
cross-sections are zero (physically, due to the liquid mass conservation). However, the integral mass transport (mass flux) through the meridional cross-section (in the
$\hat{\unicode[STIX]{x1D73D}}$
direction) is not zero and reads as
where
$v^{(2/3)}$
is defined by (3.3).
The asymptotic series in (4.2) shows that
$$\begin{eqnarray}M=\underbrace{\displaystyle \int _{-\infty }^{0}\displaystyle \int _{0}^{k}w_{2}^{E}(r,z)\,\text{d}r}_{M^{E}}+\underbrace{\displaystyle \int _{0}^{k}\langle v_{2}^{(1/3)}|_{z=0}\,\unicode[STIX]{x1D701}^{(1/3)}\rangle \,\text{d}r}_{M^{s}}+o(\unicode[STIX]{x1D716}^{2/3}),\end{eqnarray}$$
where
$M^{E}$
implies the Eulerian-mean azimuthal mass flux but the second mass-flux contribution,
$M^{s}$
, does not depend on
$\boldsymbol{w}^{E}$
, but is a consequence of a quadratic quantity by the lowest-order potential-flow terms. Following derivations by Faltinsen & Timokha (Reference Faltinsen and Timokha2009, §9.6.3) gives
which shows that
$M^{s}\not =0$
for swirling (
$\unicode[STIX]{x1D6EF}\not =0$
).
Derivations in (4.4) utilised the linear kinematic boundary condition on the free surface
$\dot{\unicode[STIX]{x1D701}}^{(1/3)}=v_{3}^{(1/3)}|_{z=0}$
. This boundary condition states that the fluid particles on the free surface move together with this surface and, in particular,
$\unicode[STIX]{x1D701}^{(1/3)}=d_{3}^{(1/3)}|_{z=0}$
, where
is the first-order Lagrangian displacement. The dual Eulerian–Lagrangian character of the wave elevation
$\unicode[STIX]{x1D701}^{(1/3)}=d_{3}^{(1/3)}$
makes it possible to interpret
$M^{s}$
in (4.3) as being caused by the Stokes-drift-type mass transport. This resolves the mathematical paradox – appearance of the extra non-zero contribution
$M^{s}$
into the total mass flux
$M$
, which is formally computed within the framework of the Eulerian specification but is not included in
$M^{E}$
.
Treatment of the mass-flux
$M^{s}$
as a product of the non-Eulerian-mean mass transport is typical for the external progressive surface-wave problems. Details and discussions can be found in books and review papers, which are exemplified by Bühler (Reference Bühler2009) and van den Bremer & Breivik (Reference van den Bremer and Breivik2017). Pursuing a self-contained presentation, we add an instructive example in appendix C of the Supplementary Materials where the link between the Stokes-drift mass flux and an integral alike
$M^{s}$
is shown for progressive waves in a rectangular channel by using the Stokes integration theorem and (3.9). (The idea of using (3.9) was proposed by Dr W. Herreman.)
By adopting definition (3.9) for the Stokes-drift velocity
$\boldsymbol{w}^{S}$
, we can examine whether the latter fact from the external surface-wave theory remains correct for the studied problem. For this purpose, we consider the integral over the azimuthal Stokes-drift velocity and apply the Stokes integration theorem:
$$\begin{eqnarray}\displaystyle M^{S} & = & \displaystyle \displaystyle \int _{-\infty }^{0}\displaystyle \int _{0}^{k}w_{2}^{S}(r,z)\,\text{d}r\,\text{d}z={\displaystyle \frac{1}{2}}\displaystyle \int _{-\infty }^{0}\displaystyle \int _{0}^{k}\left(\unicode[STIX]{x1D735}\times \left\langle \boldsymbol{v}^{(1/3)}\times \displaystyle \int \boldsymbol{v}^{(1/3)}\,\text{d}t\right\rangle \right)\boldsymbol{\cdot }\hat{\unicode[STIX]{x1D73D}}\,\text{d}r\,\text{d}z\nonumber\\ \displaystyle & = & \displaystyle {\displaystyle \frac{1}{2}}\displaystyle \int _{-\infty }^{0}\left.\left\langle v_{1}^{(1/3)}\displaystyle \int v_{2}^{(1/3)}\,\text{d}t-v_{2}^{(1/3)}\displaystyle \int v_{1}^{(1/3)}\,\text{d}t\right\rangle \right|_{r\rightarrow 0}\,\text{d}z\nonumber\\ \displaystyle & & \displaystyle +\,{\displaystyle \frac{1}{2}}\displaystyle \int _{0}^{k}\left.\left\langle v_{2}^{(1/3)}\displaystyle \int v_{3}^{(1/3)}\,\text{d}t-v_{3}^{(1/3)}\displaystyle \int v_{2}^{(1/3)}\,\text{d}t\right\rangle \right|_{z=0}\,\text{d}r\nonumber\\ \displaystyle & & \displaystyle +\,{\displaystyle \frac{1}{2}}\displaystyle \int _{0}^{-\infty }\left.\left\langle v_{1}^{(1/3)}\displaystyle \int v_{2}^{(1/3)}\,\text{d}t-v_{2}^{(1/3)}\displaystyle \int v_{1}^{(1/3)}\,\text{d}t\right\rangle \right|_{r=k}\,\text{d}z\nonumber\\ \displaystyle & & \displaystyle +\,{\displaystyle \frac{1}{2}}\displaystyle \int _{k}^{0}\left.\left\langle v_{2}^{(1/3)}\displaystyle \int v_{3}^{(1/3)}\,\text{d}t-v_{3}^{(1/3)}\displaystyle \int v_{3}^{(1/3)}\,\text{d}t\right\rangle \right|_{z\rightarrow -\infty }\,\text{d}r.\end{eqnarray}$$
Here, the third and fourth integrals are zeros but the first one implies the limit
$-(1/4)\unicode[STIX]{x1D6EF}[J_{1}^{\prime }(r)J_{1}(r)r^{-1}]\rightarrow -(1/16)\unicode[STIX]{x1D6EF}$
as
$r\rightarrow 0$
.
Because
$$\begin{eqnarray}\left.\begin{array}{@{}c@{}}\left.\left\langle v_{3}^{(1/3)}\displaystyle \int v_{2}^{(1/3)}\right\rangle \right|_{z=0}=-\left.\left\langle v_{2}^{(1/3)}\displaystyle \int v_{3}^{(1/3)}\right\rangle \right|_{z=0};\\ \left.\left\langle v_{1}^{(1/3)}\displaystyle \int v_{2}^{(1/3)}\right\rangle \right|_{r\rightarrow 0}=-\left.\left\langle v_{2}^{(1/3)}\displaystyle \int v_{1}^{(1/3)}\right\rangle \right|_{r\rightarrow 0},\end{array}\right\}\end{eqnarray}$$
the derivation line (4.6) results in
$$\begin{eqnarray}\underbrace{\displaystyle \int _{0}^{k}\langle v_{2}^{(1/3)}|_{z=0}\,\unicode[STIX]{x1D701}^{(1/3)}\rangle \,\text{d}r}_{M^{s}}=M^{S}+\underbrace{\displaystyle \int _{-\infty }^{0}\langle v_{2}^{(1/3)}d_{1}^{(1/3)}\rangle |_{r\rightarrow 0}\,\text{d}z}_{{\textstyle \frac{1}{16}}\unicode[STIX]{x1D6EF}},\end{eqnarray}$$
where
$d_{1}^{(1/3)}|_{r\rightarrow 0}=\!\int v_{1}^{(1/3)}\,\text{d}t|_{r\rightarrow 0}$
expresses the first-order radial Lagrangian displacement for fluid particles, which belong (but do not cross the tank axis) to a meridional plane
$\unicode[STIX]{x1D703}=\unicode[STIX]{x1D703}_{0}=\text{const.}$
as
$r\rightarrow 0$
.
Formula (4.8) links the actual non-Eulerian-mean mass flux
$M^{s}$
and its formal mathematic prediction based on the Stokes mass-transport concept,
$M^{S}$
. The formula and derivation line (4.6) also helps to clarify the difference between
$M^{s}$
and
$M^{S}$
. Computing
$M^{S}$
ignores the fluid particles in the meridional plane and their first-order Lagrange displacements, which cross the tank centre, namely, the fluid particles, which move between in the meridional plane
$\unicode[STIX]{x1D703}=\unicode[STIX]{x1D703}_{0}$
=const. and the meridional plane
$\unicode[STIX]{x1D703}=\unicode[STIX]{x1D703}_{0}+\unicode[STIX]{x1D70B}$
. In the contrast,
$M^{s}$
determines the azimuthal mass flux as it stands, including the effect of these cross-moving particles.
One can rebuild (extract)
which would play the same role as the Stokes-drift velocity
$\boldsymbol{w}^{S}$
for the external progressive waves. For this purpose, we replace the finite integration on
$(0,k)$
in (4.4) by a small interval
$(r,r+\unicode[STIX]{x0394}r)$
that gives, in the limit
$\unicode[STIX]{x0394}r\rightarrow 0$
, that
$w_{2}^{s}(r,z)=\unicode[STIX]{x1D6EF}F(z)r^{-1}J_{1}^{2}(r)$
. To derive
$F(z)$
, one should exclude integration by
$r$
in (4.4), fix the radial coordinate
$0<r=r_{0}<k$
and consider integration by
$z$
from
$-\infty$
to
$d_{3}^{(1/3)}|_{z=z_{0}}=\unicode[STIX]{x1D701}_{z_{0}}=e^{z_{0}}\int v_{3}^{(1/3)}(r_{0},\unicode[STIX]{x1D703},z_{0})\,\text{d}t+o(\unicode[STIX]{x1D716}^{1/3})$
in (4.4), where
$-\infty <z_{0}<0$
is a fixed vertical coordinate in the mean cross-sectional area. This gives
$\int _{-\infty }^{z_{0}}F(z)\,\text{d}z=(1/2)\text{e}^{2z_{0}}$
and, therefore,
$F(z)=\text{e}^{2z}$
. As a consequence,
where
$\boldsymbol{w}^{s}$
has already appeared for the inviscid Craik–Leibovich equation.
In summary, to describe the mass flux (4.3) in the Prandtl experiments one should consider
where
$\boldsymbol{w}^{s}$
implies, according to the generally accepted physical treatment of
$M^{s}$
, the non-Eulerian (Stokes-drift-type) velocity (4.10) but the Eulerian-mean velocity
$\boldsymbol{w}^{E}$
is governed by the boundary value problem (3.2), (3.8)–(3.12) whose unique non-trivial solution requires tangential boundary conditions on the wetted tank wall.
5 The mass-transport velocity w
$^{P}$
5.1 Where the tangential boundary conditions for
$\boldsymbol{w}^{E}$
may come from
There are no physical mechanisms causing tangential boundary conditions for
$\boldsymbol{w}^{E}$
by (3.2), (3.8)–(3.12) within the framework of the fully inviscid hydrodynamic model. Usually, the steady tangential stresses and associated boundary conditions are related to steady streaming (Riley Reference Riley2001), a non-zero Eulerian mean of fluctuating flows, which is normally caused by the action of an oscillating submerged body, or, indirectly, by the action of viscous stresses in thin boundary layers at no-slip boundaries, here, at the vertical wall. Appendix B in the Supplementary Materials derives the tangential boundary conditions (B 23a) and (B 23b) as they follow from the steady streaming theory for the swirl-type sloshing. These Eulerian derivations ignore the non-Eulerian-mean mass transport of fluid particles around (in a small neighbourhood of) the wall, which should, in contrast to the inner liquid points, influence the viscous stresses and, mathematically, modify the no-slip conditions, which must, in addition, be applied, according to experiments, to the azimuthal mass transport on the vertical wall. Physically, this means that moving fluid particles due to
$\boldsymbol{w}^{s}$
not being zero at
$r=k$
cause an extra vorticity.
We have no idea regarding how to include the non-zero
$\boldsymbol{w}^{s},\,r\rightarrow k$
into the steady streaming theory. However, we can either neglect the Stokes-drift-type mass transport at the wall or assume that
$\boldsymbol{w}^{s}$
predetermines the inviscid Eulerian mean
$\boldsymbol{w}^{E}$
so that other viscous streams lead to negligible contributions to
$\boldsymbol{w}^{P}$
.
5.1.1 Steady streaming with negligible
$\boldsymbol{w}^{s}$
at the wall
By ignoring the non-Eulerian-mean mass transport by
$\boldsymbol{w}^{s}$
on the wall, appendix B in the Supplementary Materials derives the tangential boundary conditions, which express the Eulerian-mean steady streaming. Because Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016) showed that
$\bar{a}=\bar{b}=0$
and
$a\approx b$
for undamped sloshing, at least, for the orbital tank forcing, one can focus on the framed terms in (B 23a) and (B 23b). Comparing (B 23b) with
$w_{3}^{E}=0$
in the mean free surface in (3.11) shows that using (B 23b) in our inviscid analysis is mathematically contradictory, because this causes
$\boldsymbol{w}^{E}$
to be discontinuous but
$\unicode[STIX]{x1D734}^{E}=\unicode[STIX]{x1D735}\times \boldsymbol{w}^{E}$
becomes singular (infinite) at the contact point
$r=k,\,z=0$
, and, as a consequence, asymptotic analysis in § 3 is invalid as involving the infinite function
$\unicode[STIX]{x1D734}^{E}$
in the second-order approximation. This mathematical problem is well known from viscous computational fluid dynamics (CFD). Its resolution requires us to relax the no-slip condition about the moving contact line. However, existing computational approaches are not adoptable in analytical inviscid studies.
The azimuthal tangential boundary condition (B 23a) states that
$\boldsymbol{w}^{E}$
has in each inner point the opposite sign to
$\unicode[STIX]{x1D6EF}$
and, therefore,
$\boldsymbol{w}^{E}$
implies the return flow (in opposite direction to swirling). However, (B 23a) does not provide the experimental peculiarity (b) consisting of the zero mass transport on the wall,
$w_{2}^{P}=0,\,r=k$
. Indeed, imposing
$w_{2}^{P}=0$
(
$w_{2}^{E}=-w_{2}^{s}$
) at
$r=k$
gives
which should be compared with
following from (B 23a). Calculations show that
$p_{0}^{V}/p_{0}^{s}=0.5287583024\ldots$
, and, therefore, (B 23a) underpredicts (by almost twice) the needed value.
5.1.2 The non-Eulerian mass transport
$\boldsymbol{w}^{s}$
plays a dominant role at the wall
Assuming the dominant role of
$\boldsymbol{w}^{s}$
at the wall is physically consistent with the inviscid limit,
$Re_{s}\rightarrow \infty$
, when purely viscous stresses due to no-slip conditions for oscillatory flow components and associated viscous steady streams may be neglected. Considering the inviscid limit and the time-averaging procedure of the no-slip conditions for the resulting particle motions around the wall make the latter conditions automatically satisfied for the
$O(\unicode[STIX]{x1D716}^{1/3})$
asymptotic component but the
$O(\unicode[STIX]{x1D716}^{2/3})$
approximation mathematically leads to the tangential boundary condition
This condition is exactly the same as required by the peculiarity (b) (see the Introduction). It means that liquid particles, which are involved into the mass transport
$\boldsymbol{w}^{P}$
, are unmovable (do not steadily slip) along the wall but their components,
$\boldsymbol{w}^{s}$
(non-Eulerian mean) and
$\boldsymbol{w}^{E}$
(Eulerian mean) eliminate each other on the wetted tank surface. The Eulerian mean implies the return flow. The mass transport becomes self-supported when equipped with (5.3). Decreasing
$\boldsymbol{w}^{E}$
at the wall (due to small viscous stresses) makes the mass transport around the wall (by
$\boldsymbol{w}^{s}$
) generating a tangential stresses counteracting
$\boldsymbol{w}^{s}$
until (5.3) is fulfilled.
Use of (5.3) is mathematically not contradictory but we should understand that this condition neglects purely viscous streams including the poloidal recirculation (details of the viscous phenomena are extensively discussed by Bouvard et al. (Reference Bouvard, Herreman and Moisy2017)). This condition can also be treated as an hypothetical analogy of the Kutta–Joukowski condition applied to foils in inviscid fluid with rotational flow.
5.2 The resulting mass-transport velocity
Accepting (5.3) causes an axisymmetric solution of the boundary value problem (3.2), (3.8)–(3.12). Furthermore, if the vector function
$\boldsymbol{w}^{E}(r,z)$
is continuous and finite in
$Q_{0}$
, the
$\hat{\boldsymbol{r}}$
and
$\hat{\boldsymbol{z}}$
components of the Craik–Leibovich equation (3.8) reduce to
and, therefore,
but (3.2) transforms to
Equations (5.5), (5.6) are restricted to the homogeneous boundary conditions (3.11) and (5.3b
). Because we assume a non-trivial azimuthal velocity component
$w_{2}^{E}$
, which appears in the coefficients of (5.5), (5.6), the boundary problem (5.5), (5.6), (3.11), (5.3b
) can be solved by using, for example, the method of characteristics, which leads to the trivial solution
$w_{1}^{E}=w_{3}^{E}\equiv 0$
.
The
$\hat{\unicode[STIX]{x1D73D}}$
-component of the Craik–Leibovich equation follows then from the boundary problem
Substitution of
$w_{2}^{E}(r,z)=\unicode[STIX]{x1D6EF}\text{e}^{2z}w^{E}(r)$
reduces (5.7) to the boundary value problem for the Bernoulli differential equation
which has an exact analytical solution and causes
where
$w^{S}(r)$
is associated with the Stokes-drift velocity (3.9); one should require
$c_{0}<k^{-2}$
to avoid singularity in the interval
$0\leqslant r\leqslant k$
.
When
$p_{0}=p_{0}^{s}$
, the resulting mean azimuthal mass-transport velocity reads as
where
$w^{S}$
comes from (3.9) and
$c_{0}=0.10398523061\ldots$
.
The inviscid theoretical prediction (5.10) includes three consequent multipliers,
where
$\unicode[STIX]{x1D6EF}$
is responsible for the wave amplitude (
$\unicode[STIX]{x1D6EF}=ab$
for the undamped sloshing),
$\text{e}^{2z}$
implies the exponential decay (the formula is true only for fairly deep liquid depths) and
$w^{P}(r)$
describes the radial distribution of the mean mass-transport velocity. As remarked in § 2, the wave amplitude multiplier
$\unicode[STIX]{x1D6EF}$
can have a feedback from the mean mass transport, i.e. the potential-flow values
$a,\bar{a},\bar{b},b$
by Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016) provide only an estimate for
$\unicode[STIX]{x1D6EF}$
. Dependence of
$a,\bar{a},\bar{b}$
and
$b$
on
$\boldsymbol{w}^{P}$
goes beyond these studies. Finally, the exponential decay multiplier is formally of the order
$O(1)$
, which means that (5.10) is applicable only for
$z$
far away from the mean free surface.
Remark. If we assume (even though we proved that it is not true), that the non-Eulerian mass transport is described by
$\boldsymbol{w}^{S}$
, i.e.
$\boldsymbol{w}^{P}=\boldsymbol{w}^{L}=\boldsymbol{w}^{S}+\boldsymbol{w}^{E}$
and, in addition, require the experimental peculiarity (b) implying
$\boldsymbol{w}^{E}=-\boldsymbol{w}^{S}$
at
$r=k$
, the parameter
$p_{0}$
in (5.9) becomes equal to
$w(k)$
that gives
$c_{0}=0\Rightarrow w^{E}(r)=-w^{S}(r)$
. The latter means that the return flow
$\boldsymbol{w}^{E}$
fully annihilates
$\boldsymbol{w}^{S}$
in each inner point of the liquid domain and, therefore, the total mass transport becomes zero, at least within the framework of the inviscid hydrodynamic model. This confirms that the first-order radial Lagrangian fluid particles displacements through the tank axis, which were extensively discussed in § 4 as causing the difference between
$\boldsymbol{w}^{s}$
and
$\boldsymbol{w}^{S}$
, play an important role in the description of the Prandtl phenomenon.
6 Comparison with experiments
Functions
$w^{s}(k\,r_{\ast })$
by (4.10),
$w^{E}(k\,r_{\ast })$
by (5.9) and
$w^{P}(k\,r_{\ast })$
by (5.10), which represent the radial distribution of the mean non-Eulerian, Eulerian and resulting mass-transport azimuthal velocities, respectively, by the
$r_{0}$
-scaled radial coordinate
$r_{\ast }=r/k$
. The graph
$w^{S}(k\,r_{\ast })$
illustrates the Stokes azimuthal mean velocity by (3.9).

Figure 2 presents the theoretical azimuthal mass-transport velocity and its components along the
$r_{0}$
-scaled radial axis (the non-dimensional radial coordinate
$r_{\ast }=r/k$
). These contain
$w^{s}$
(the non-Eulerian mean, (4.10)),
$w^{E}$
(the Eulerian mean, equation (5.9)) and
$w^{P}$
(summarised mass transport (5.10)). In addition, we depict the non-modified azimuthal Stokes-drift velocity
$w^{S}$
, which is governed by (3.9) and appears in the Craik–Leibovich equation.
The non-Eulerian-mean velocity is co-directed with the swirl-type propagating wave,
$w^{s}(k\,r_{\ast })>0,\,0<r_{\ast }<1$
but the Eulerian-mean velocity counteracts it, i.e.
$w^{E}(k\,r_{\ast })<0,\,0<r_{\ast }<1$
; here,
$r_{\ast }=r/k$
is the
$r_{0}$
-scaled radial coordinate. This implies, in particular, that comparing the experimental data with, independently,
$w^{s}$
and
$w^{E}$
has no meaning. The sign of
$w^{E}$
(the return flow) is opposite to the experimental discoveries, but
$w^{s}$
does not satisfy, by itself, the experimental peculiarity (b) (zero azimuthal mass transport on the vertical wall). Only the summarised mass transport distribution
$w^{P}$
(joint effect of
$w^{s}$
and
$w^{E}$
) has a physical meaning for the studied problem and can be adopted for the forthcoming experimental validation. Obviously,
$w^{P}$
satisfies the experimental peculiarities (a) and (b) (see the Introduction).
The summarised mass transport
$w^{P}$
reaches its maximum value at about the semi-radius (satisfies the experimental peculiarity (c)) so that its graph has an arc-type shape. This shape is independent of the forcing frequency and amplitude (see, discussions on that by Hutton Reference Hutton1964; Reclari Reference Reclari2013; Bouvard et al.
Reference Bouvard, Herreman and Moisy2017).
Another interesting result is that
$(w^{P})^{\prime }(0)$
is equal to
$(w^{S})^{\prime }(0)=1/8$
,
and, as a consequence,
$\boldsymbol{w}^{P}$
is satisfactorily fitted by the Stokes drift (3.9) in the neighbourhood of the tank centre (satisfies the experimental peculiarity (d)). The latter experimental fact was discussed by Bouvard et al. (Reference Bouvard, Herreman and Moisy2017) who commented that the mean mass-transport flow is nearly in solid-body rotation near the centre, rotating in the direction of the orbital shaking, for
$0<r_{\ast }<0.3$
. Comparing
$w^{P}$
with
$w^{S}$
in figure 2 confirms this interval of
$r_{\ast }$
.
In summary, the constructed solution satisfies all the experimentally established peculiarities, which were listed in the Introduction, and, therefore, it qualitatively describes the Prandtl phenomenon. The next step consists of a quantitative validation of the analytical result (5.10). Limitations of the Narimanov–Moiseev theory (i)–(iv) (see, § 2) and a lack of measurements of the Prandtl mass-transport phenomenon (including his pioneering work) for fairly deep liquid depths makes it difficult to find suitable experimental data for this validation. With known reservations, only measurements by Hutton (Reference Hutton1964) and Bouvard et al. (Reference Bouvard, Herreman and Moisy2017) can be adopted.
6.1 Experiments by Bouvard et al. (Reference Bouvard, Herreman and Moisy2017)
By using a very special stroboscopic particle image velocimetry technique, which makes it possible to detect the mass transport, Bouvard et al. (Reference Bouvard, Herreman and Moisy2017) measured the azimuthal mass transport (corresponding to
$\boldsymbol{w}^{P}$
) and mean radial (associated with
$w_{1}^{E}=0$
in our inviscid theory) velocities for swirling waves occurring due to the orbital horizontal circular forcing of the tank. In these experiments, the upright circular container was filled, consequently, by two different silicon oils so that the experimental series were characterised by Reynolds numbers equal to
$Re_{s}=r_{0}^{2}\unicode[STIX]{x1D70E}/(k^{2}\unicode[STIX]{x1D708})=1000/k^{2}$
(
$\unicode[STIX]{x1D708}=50$
mm
$^{2}$
s
$^{-1}$
) and
$100/k^{2}$
(
$\unicode[STIX]{x1D708}=500$
mm
$^{2}$
s
$^{-1}$
), respectively (
$\unicode[STIX]{x1D708}$
is the kinematic viscosity). Silicon oil with
$\unicode[STIX]{x1D708}=500$
mm
$^{2}$
s
$^{-1}$
is definitely not low viscosity. The measurements were done at the horizontal level
$z_{0}^{\ast }=z_{0}/r_{0}=-0.23$
, which is relatively far from the free surface,
$|z_{0}|=0.23\,k=O(1)$
and, therefore, formula (5.10) is applicable.
The liquid depth was sufficiently large,
$h_{\ast }=h/r_{0}=2.168$
, (
$r_{0}=51.2$
mm, the Bond number was approximately 500 so that surface tension can be neglected). The forcing amplitude was
$O(\unicode[STIX]{x1D716})=\unicode[STIX]{x1D702}_{a}^{\ast }=\unicode[STIX]{x1D702}_{1a}/r_{0}=\unicode[STIX]{x1D702}_{2a}/r_{0}=0.057$
, but the tested forcing frequencies were in the range
$0.48\lesssim \unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}\lesssim 0.78$
, which is generally speaking, away from the primary resonance zone,
$0.9\lesssim \unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}\lesssim 1.07$
, where the Moiseev detuning condition (2.8) is fulfilled. Finally, the experimental frequency range contains rather dangerous secondary resonances by the first axisymmetric (01) mode at
$\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}=0.72$
(see, an extra small amplitude response peak about this value in experiments by Reclari (Reference Reclari2013)) and for the asymmetric (21) mode (
$\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}=0.65$
).
Summarising the aforementioned input parameters shows that conditions (i), (iii) and (iv) (see, § 1) are, generally, not satisfied. However, the approximate analytical result (5.10) can, for experiments by Bouvard et al. (Reference Bouvard, Herreman and Moisy2017), be interpreted as the single
$J_{1}$
-mode approximation on the azimuthal mass transport within the framework of the linear sloshing theory.
Our first focus is on the mean angular liquid velocity
$\unicode[STIX]{x1D714}_{O}$
in the tank centre, which can, according to (5.10), be estimated as
$\unicode[STIX]{x1D714}_{O}/\unicode[STIX]{x1D70E}=\unicode[STIX]{x1D6EF}\text{e}^{kz_{0}^{\ast }}(w^{P})^{\prime }(0)=(1/8)\unicode[STIX]{x1D6EF}\text{e}^{kz_{0}^{\ast }},\,z_{0}^{\ast }=z_{0}/r_{0},$
where
$z_{0}$
is the dimensional horizontal level beneath the mean free surface. Adopting the single-mode linear potential-flow sloshing prediction of the amplitude parameter
$\unicode[STIX]{x1D6EF}$
(Faltinsen & Timokha Reference Faltinsen and Timokha2009, chap 5) gives the formula
$$\begin{eqnarray}{\displaystyle \frac{\unicode[STIX]{x1D714}_{O}}{\unicode[STIX]{x1D70E}}}={\displaystyle \frac{1}{8}}(k^{2}\unicode[STIX]{x1D702}_{a}^{\ast }\unicode[STIX]{x1D6FC})^{2}\text{e}^{2kz_{0}^{\ast }}\left(\left({\displaystyle \frac{\unicode[STIX]{x1D70E}_{1}}{\unicode[STIX]{x1D70E}}}\right)^{2}-1\right)^{-2},\quad \unicode[STIX]{x1D6FC}={\displaystyle \frac{\displaystyle \int _{0}^{k}r^{2}J_{1}(r)\,\text{d}r}{\displaystyle \int _{0}^{k}rJ_{1}^{2}(r)\,\text{d}r}},\end{eqnarray}$$
which is compared with the measured data by Bouvard et al. (Reference Bouvard, Herreman and Moisy2017) in figure 3 (the measured values for
$\unicode[STIX]{x1D708}=50\,\text{mm}^{2}\,\text{ s}^{-1}$
are marked by the empty circles but
$\unicode[STIX]{x1D708}=500\,\text{mm}^{2}\,\text{s}^{-1}$
corresponds to the filled circles). The logarithmic axis scale is kept from the original experimental work.
Theoretical prediction (6.2) of the non-dimensional angular velocity
$(\unicode[STIX]{x1D714}_{O})_{\ast }=\unicode[STIX]{x1D714}_{O}/\unicode[STIX]{x1D70E}$
in the tank centre (
$r_{\ast }=0$
and
$z_{0}^{\ast }=z_{0}/r_{0}=-0.23$
), the solid line, and the corresponding measured data by Bouvard et al. (Reference Bouvard, Herreman and Moisy2017) for two liquids with
$\unicode[STIX]{x1D708}=50~\text{mm}^{2}~\text{s}^{-1}$
(empty circles) and
$\unicode[STIX]{x1D708}=500~\text{mm}^{2}~\text{s}^{-1}$
(filled circles). The tank is exposed to the circular orbital forcing with the
$r_{0}$
-scaled amplitude
$\unicode[STIX]{x1D702}_{a}^{\ast }=0.057$
. Experiments (a) and (b) were done with
$\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}=0.78$
and 0.75, respectively. Theoretical amplitude parameter
$\unicode[STIX]{x1D6EF}$
is computed by using the linear sloshing theory with potential flows. The logarithmic scale is kept from the original publication by Bouvard et al. (Reference Bouvard, Herreman and Moisy2017).

Figure 3 demonstrates that viscosity matters for
$\unicode[STIX]{x1D708}=500~\text{mm}^{2}~\text{s}^{-1}$
so that the theory only qualitatively fits the corresponding measured values. As for silicon oil with
$\unicode[STIX]{x1D708}=50~\text{mm}^{2}~\text{s}^{-1}$
, discrepancy for
$\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}<0.75$
is theoretically clarified by the aforementioned secondary resonances at
$\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}=0.72$
and 0.65, but the theory shows a good agreement for the cases (a) (
$\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}=0.78$
) and (b) (
$\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}=0.75$
), which are away from the secondary resonances but close to the primary resonance zone. The latter closeness means an increasing contribution of the lowest natural sloshing mode.
When preparing the paper, Bouvard et al. (Reference Bouvard, Herreman and Moisy2017) measured azimuthal and radial components of the mean mass transport as functions of
$r_{\ast }$
(
$z_{0}^{\ast }=-0.23$
). Professor F. Moisy (personal communication) granted us with these
$r_{0}\unicode[STIX]{x1D70E}$
-scaled measured data. Adopting, as in (6.2), the linear potential-flow sloshing theory for computing the amplitude parameter
$\unicode[STIX]{x1D6EF}$
gives the formula
from (5.10).
Figures 4(a) and 4(b) compare (6.3) with the measured data by F. Moisy (personal communication) for the cases (a) and (b) in figure 3. Specifically, equation (6.3) states that the radial mean mass transport is zero (
$(w_{1}^{P})_{\ast }\equiv 0$
) because we neglect the poloidal recirculation and other viscous non-azimuthal mean streams. Figure 4 supports this neglect. We see that the measured radial mean mass transport is zero for
$0<r_{\ast }<0.7$
but, even though this mass transport is not zero in the interval
$0.7<r_{\ast }<1$
, it is clearly lower than the measured azimuthal mass transport. This is especially for the case (a), whose forcing frequency is relatively close to the Narimanov–Moiseev resonant zone.
The
$r_{0}\unicode[STIX]{x1D70E}$
-scaled measured (F. Moisy, personal communication) and theoretical (6.3) liquid mass-transport velocity in the azimuthal (solid circles and solid arc-type line, respectively) and radial (empty circles, the theory states zero) directions. The results with
$z_{0}^{\ast }=-0.23$
, the forcing frequencies
$\unicode[STIX]{x1D70E}/\unicode[STIX]{x1D70E}_{1}=0.78$
(a) and 0.75 (b), the forcing amplitude
$\unicode[STIX]{x1D702}_{1}^{\ast }=0.057$
(circular tank motions). The experimental tank has the non-dimensional depth
$h_{\ast }=h/r_{0}=2.168$
.

As for the azimuthal mass-transport velocity in figure 4 (solid arc and filled circles), the constructed inviscid theory by (6.3) demonstrates quantitatively good prediction. Especially, if we recall that the theory neglects viscous flows and adopts the approximate single-mode linear sloshing solution for computing the amplitude parameter
$\unicode[STIX]{x1D6EF}$
.
6.2 Experiments by Hutton (Reference Hutton1964)
The azimuthal mass-transport velocity scaled by its maximum value for the experimental case by Hutton (Reference Hutton1964), which was done with tap water (
$Re_{s}=4.5\times 10^{5}$
),
$r_{0}=22.46884$
cm,
$h_{\ast }h/r_{0}=1.4$
. Swirling wave mode was excited by the horizontal longitudinal tank forcing along the
$Ox$
axis (
$\unicode[STIX]{x1D702}_{2a}=0$
) with the forcing frequency
$\unicode[STIX]{x1D70E}=\unicode[STIX]{x1D70E}_{1}$
, for which, according to Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016), the swirling wave mode is stable. The measurements were done at the vertical level
$z_{0}=-0.43h,z_{0}^{\ast }=-0.59$
. Three different forcing amplitudes (denoted as small, middle and large) were used, which are not specified by Hutton (Reference Hutton1964) who has linked these three experimental cases to experimental crest (
$e_{max}$
)/trough (
$e_{min}$
) near the wall equal to 8.89/5.334, 11.43/6.35 and 13.97/7.62 cm, respectively. The experimental set-up employed a mechanical measurement technique with rectangular paddles of different sizes. The sizes are marked as ‘small paddle’
$=l_{r}\times l_{z}=3.937\times 2.286$
cm; ‘medium paddle’
$=4.8514\times 3.6068$
cm; ‘large paddle’
$=5.968\times 4.334$
cm. The measurements in zone F can, according to Hutton (Reference Hutton1964), be affected by the paddle feedback on the liquid flows.

Comparison of theoretical formula (6.5) with measured data by Hutton (Reference Hutton1964) with the input parameters documented in the caption of figure 5. Measurements from zone Z, where the azimuthal mass transport was measured with relatively large paddles, which influence the liquid flows, were excluded.

Another experimental series for the swirling wave mode was reported by Hutton (Reference Hutton1964). He used the longitudinal harmonic excitation (
$\unicode[STIX]{x1D702}_{1a}\not =0,\unicode[STIX]{x1D702}_{2a}=0$
) with the forcing frequency equal to
$\unicode[STIX]{x1D70E}=\unicode[STIX]{x1D70E}_{1}$
, i.e. when the swirling wave mode is stable and possesses the same features as for the orbital forcing (Faltinsen et al.
Reference Faltinsen, Lukovsky and Timokha2016). Hutton (Reference Hutton1964) utilised three different forcing amplitudes, which were, unfortunately, not specified in the original paper but, instead, he distinguished the corresponding experimental series by documenting the experimental crest (
$e_{max}$
) and through (
$e_{min}$
) near the wall in the excitation plane (see the explanation in figure 5). The geometric input was
$r_{0}=22.47$
cm,
$h=31.45$
cm (
$h_{\ast }=h/r_{0}=1.4$
),
$\unicode[STIX]{x1D70E}_{1}=8.91$
rad s
$^{-1}$
. Tap water with
$\unicode[STIX]{x1D708}=10^{-6}$
m
$^{2}~\text{s}^{-1}$
and
$Re_{s}=4.5\times 10^{5}$
was used.
To measure the azimuthal mass-transport velocity distribution, Hutton (Reference Hutton1964) used a mechanical device (‘fixture’) whose main element is a rectangular paddle lying in the meridional plane and rotating together with the contained liquid as depicted in figure 5. Three different paddles were employed, which are denoted as ‘small paddle’ (rectangle sides are
$l_{r}\times l_{z}=3.937\times 2.286$
cm), ‘medium paddle’ (4.8514
$\times$
3.6068 cm) and ‘large paddle’ (5.968
$\times$
4.334 cm). The measurements were made at the vertical levels
$z_{0}^{\ast }=-0.59\,(z_{0}^{\ast }=-0.43\,h)$
,
$z_{0}^{\ast }=-0.92\,(z_{0}^{\ast }=-0.67\,h)$
and
$z_{0}^{\ast }=-1.22\,(z_{0}^{\ast }=-0.89\,h)$
. Hutton (Reference Hutton1964) wrote that ‘the test data also indicated that the transport velocity was nearly constant over the range of depths’, which means that the azimuthal mass-transport velocity does not depend on
$z$
(as in (5.10)) due to the bottom effect for the liquid filling
$h_{\ast }=h/r_{0}=1.4$
, which is not fairy deep. The experimental input in Hutton (Reference Hutton1964) satisfies conditions (i), (iii) and (iv) in § 2 but not (ii).
The azimuthal mass-transport velocities were normalised by their maximum values and compared with our inviscid analytical solution (5.10),
$w^{P}(r)/\max |w^{P}|$
. The result is presented in figure 5. The smaller symbol implies the smaller paddle. The filling depth of the used symbols also increases with increasing forcing amplitude (Hutton Reference Hutton1964, reports the experimental crest (
$e_{max}$
)/trough (
$e_{min}$
) of azimuthal progressive wave at the wall for these three forcing amplitudes as 8.89/5.588, 11.43/6.35 and 13.97/7.62 cm, respectively). Figure 5 supports our analytical inviscid solution for measurements made with ‘small paddle’. ‘Medium’ and ‘large’ paddles give clearly larger values for one measurement probe, which is located close to the tank centre. Hutton (Reference Hutton1964) suggested that bigger paddles may yield a feedback on the liquid flows for the latter case.
As we mentioned above, because the liquid depth
$h_{\ast }$
was not large enough and the near-bottom flows matter, Hutton (Reference Hutton1964) states that there is no exponential decay
$\text{e}^{2z}$
as predicted in (5.10). Hutton (Reference Hutton1964) reports that the azimuthal mass transport is uniquely a function of
$r$
. Speculatively assuming that the near-bottom flows play an ‘averaging role’, one can take the mean value of
$\text{e}^{2z}$
over the infinite depth, which implies the replacement
$\text{e}^{2z}\rightarrow \int _{-\infty }^{0}\text{e}^{2z}\,\text{d}z=1/2$
. Furthermore, one can roughly approximate the amplitude parameters
(the meaning of coefficients
$m_{1}$
and
$m_{3}$
is explained by Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016)) and formula (5.10) transforms then to the approximate form
for the
$r_{0}\unicode[STIX]{x1D70E}$
-scaled azimuthal mass-transport velocity.
The theoretical prediction (6.5) is compared in figure 6 with the measured data by Hutton (Reference Hutton1964) except those from zone F in figure 5, where, as we stated, the larger paddle feeds back to the liquid flows. The theoretical results by (6.5) look sufficiently good if we consider how many assumptions were made to derive formula (6.5).
7 Conclusions
The mean azimuthal liquid mass transport (Prandtl’ phenomenon, 1949) generated by swirling waves in a vertical circular cylindrical tank with an infinite liquid depth is theoretically described by using the asymptotic steady-state wave solution (potential flows of incompressible liquid) by Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016). The summarised mass-transport velocity is associated with
$\boldsymbol{w}^{P}=\boldsymbol{w}^{E}+\boldsymbol{w}^{s}$
, where
$\boldsymbol{w}^{E}$
is the Eulerian-mean (vortical) velocity, which appears as a second-order supplement to the asymptotic inviscid solution by Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016), and
$\boldsymbol{w}^{s}$
is the non-Eulerian mean, which is affected by the Stokes drift and the first-order radial inflow/outflow into the meridional plane at
$r=0$
. The non-Eulerian mass-transport component
$\boldsymbol{w}^{s}$
is co-directed with azimuthally propagating (swirling) wave, but
$\boldsymbol{w}^{E}$
implies the return flow.
The Eulerian-mean velocity
$\boldsymbol{w}^{E}$
is governed by the inviscid Craik–Leibovich equation whose special analytical form is derived in the present paper by using the time averaging in the vorticity equation. Finding a unique
$\boldsymbol{w}^{E}$
requires us to know tangential boundary conditions on the wall. Derivations of these conditions from the nonlinear Eulerian boundary-layer analysis (see, appendix B in the Supplementary Materials) does not fit the existing experimental peculiarities of the summarised azimuthal mass transport, and causes mathematical conflict with the used asymptotic scheme in a neighbourhood of the contact curve between the liquid surface and the tank wall. The failure is, in our opinion, caused by the neglecting of the fluid particles’ motions by
$\boldsymbol{w}^{s}$
in the neighbourhood of the wall, which acts on viscous stresses and yields an extra vorticity. In the inviscid limit, we suggest that local mass transport
$\boldsymbol{w}^{s}$
at the wall implies the dominant mechanism to constitute the Eulerian-mean flow. Physically, this implies that typical viscous vortical streams are neglected. Mathematically, this means that one can use the inviscid Craik–Leibovich boundary problem, but adopt the time-averaged no-slip condition for the resulting (Eulerian + non-Eulerian) particle velocities on the wall. The latter causes the tangential boundary condition
$\boldsymbol{w}^{E}=-\boldsymbol{w}^{s}$
on the wall. This tangential condition can be treated as an analogy of the Kutta–Joukowski condition for steady ambient flow past a stationary foil but for the studied problem. Provided by this tangential boundary condition, we were able to derive
$\boldsymbol{w}^{P}$
in a simple analytical form. The theoretical result is compared with the measured data by Hutton (Reference Hutton1964) and Bouvard et al. (Reference Bouvard, Herreman and Moisy2017). The results are in a good quantitative agreement. The comparison confirms that viscosity and associated viscous streams may play a secondary role for the mass-transport phenomenon by Prandtl.
The present paper neglects a feedback of the mean liquid rotation on the lowest-order amplitudes
$a,b,\bar{a}$
and
$\bar{b}$
, which are taken from the steady-state asymptotic solution by Faltinsen et al. (Reference Faltinsen, Lukovsky and Timokha2016). This fact as well as the near-bottom flow effect deserve dedicated study.
Acknowledgements
The authors acknowledge the financial support of the Centre of Autonomous Marine Operations and Systems (AMOS) whose main sponsor is the Norwegian Research Council (Project number 223254–AMOS). The authors also thank Professor F. Moisy for providing unpublished experimental data, which are used for comparative analysis in figure 4.
Supplementary materials
Supplementary materials are available at https://doi.org/10.1017/jfm.2019.94.






