Exact solution for progressive gravity waves on the surface of a deep fluid

Nail S. Ussembayev∗
Computer, Electrical, and Mathematical Sciences and Engineering Divison
King Abdullah University of Science and Technology, Thuwal 23955-6900, KSA

Gerstner or trochoidal wave is the only known exact solution of the Euler equations for periodic
surface gravity waves on deep water. In this Letter we utilize Zakharov’s variational formulation
of weakly nonlinear surface waves and, without truncating the Hamiltonian in its slope expansion,
derive the equations of motion for unidirectional gravity waves propagating in a two-dimensional
flow. We obtain an exact solution of the evolution equations in terms of the Lambert W -function.
The associated flow field is irrotational. The maximum wave height occurs for a wave steepness of
0.2034 which compares to 0.3183 for the trochoidal wave and 0.1412 for the Stokes wave. Like in
the case of Gerstner’s solution, the limiting wave of a new type has a cusp of zero angle at its crest.

Background.—The water waves problem even in the tial evaluated on the free surface. The expansion coef-
arXiv:1903.11909v1 [physics.flu-dyn] 28 Mar 2019

completely idealized setting of a perfect fluid is notori- ficients simplify remarkably for two-dimensional waves
ously difficult from a mathematical point of view. This propagating in the same direction and the infinite series
is one of the reasons we have very few explicit solutions can be summed in a closed form if the wave steepness
to the fully nonlinear free-surface hydrodynamics equa- is small. From the stationary periodic solution of the
tions despite extensive research over hundreds of years. resulting nonlinear equations, an exact irrotational solu-
The first exact solution was provided by Gerstner [1] in tion to the full system of hydrodynamic equations with a
1802 for pure gravity waves of finite height propagating free surface is then recovered. The solutions are provided
on the surface of an inviscid and incompressible fluid of in Eulerian coordinates [10].
infinite depth. The practical importance of trochoidal Exact solutions play an important role not just because
waves became apparent when in the 1860s Rankine and all the properties of the motion can be conveniently ex-
Froude independently rediscovered them in connection pressed and analyzed without any approximation, but
with the ship-rolling problem [2]. The fact that these also because they can provide a key insight into the na-
waves are rotational and cannot be generated from the ture of more complex and physically more realistic flows,
fluid at rest was seen as somewhat unsatisfactory. More- say, in which waves propagate under the combined influ-
over, since the fluid particles move on a closed circular ence of gravity and surface tension.
orbit, Gerstner’s solution does not predict the observed
Let us finish this introduction with a note that two-
mass transport in the direction of wave propagation [3].
dimensional flows are not of purely theoretical interest.
In 1847 Stokes presented an approximate theory for ir- Their fundamental importance stems from the observa-
rotational waves and showed that for deep water the wave tion that in many situations the motion of water waves
profile coincides with a trochoid to the third order, but can be treated as quasi-two-dimensional. For example,
not to the fourth [4]. He also conjectured the existence waves generated in the open ocean while coming ashore
of the wave of greatest height distinguished by a sharp often appear as a wave-train of very long nearly paral-
edge and argued that the enclosing angle must be 120◦ . lel crests [11] and due to this symmetry can be more
The mere existence of such waves was rigorously verified or less accurately studied and simulated by various two-
only forty years ago [5], clearly illustrating the inher- dimensional models (see, e.g., [12]).
ently complicated structure of the underlying nonlinear Theoretical formulation.—The potential flow φ(x, y, t)
equations. The existence of extreme waves that are, as
 an ideal2 incompressible fluid in the region Ω(t) =
of
hypothesized by Stokes, strictly convex between succes- (x, y) ∈ R : − ∞ < y ≤ ξ(x, t) bounded by the free
sive crests was established very recently, but in spite of surface Γ(t) = {y = ξ(x, t)} satisfies the following set of
some numerical evidence supporting the convexity of all equations:
waves of extreme form their uniqueness remains an open
problem [6].
Apart from Crapper’s exact solution for pure capillary ξt + ξx φx − φy = 0, on Γ(t) (1a)
waves [7] found in 1957 and some explicit Gerstner-type 1 2
φx + φ2y + gξ = 0,


solutions for equatorially trapped waves discovered lately φt + on Γ(t) (1b)
2
[8], the field of exact solutions for water wave equations
φxx + φyy = 0, in Ω(t) (1c)
has not experienced significant advancement since the
last century. In this Letter, we present an exact solution φy → 0, as y → −∞ (1d)
for planar periodic waves of finite amplitude moving uni-
formly and without change of form on a fluid of unlim- where g is the constant gravitational acceleration in the
ited depth when only gravity acts as a restoring force. negative y direction and subscripts denote partial deriva-
Following Zakharov [9], we use the power of Hamilto- tives. The effects of surface tension are neglected and
nian formalism to write an asymptotic expansion of the the pressure at the free surface is taken to be zero. Once
equations of motion for a pair of canonically conjugate ξ(x, t) and φ(x, y, t) are found, the velocity field and the
variables – the surface elevation and the velocity poten- pressure within the fluid domain follow from
 2

The major difficulty in applying the Hamiltonian for-

malism for water waves lies in expressing the kinetic en-
ergy in terms of free surface variables, ξ and ψ, alone.
v(x, y, t) = ∇φ, When the wave steepness is small, |ξx |  1, one can write
|v|2
 
the expansion of the Hamiltonian in a power series in the
p(x, y, t) = −ρ gy + φt +
2 amplitudes of the waves. Let ξˆk = ξˆ−k ∗
and ψ̂k = ψ̂−k ∗

denote the Fourier components of ξ and ψ, then upon

solving the boundary-value problem for Laplace’s equa-
where ρ is the fluid density. Since φ is harmonic in Ω(t), it tion the Hamiltonian (3) reads
is uniquely determined by its trace on the free boundary,
ψ(x, t) = φ|y=ξ(x,t) , and (1d). Zakharov ingeniously real- Z
1
ized that equations (1a)-(1b) describing the wave motion H= |k|ψ̂0∗ ψ̂0 + g ξˆ0∗ ξˆ0 dk+
can be put in the form of Hamilton’s canonical equations 2
∞ Z n+1 n+1
! n+1
(n+2)
X Y X Y
+ E ψ̂0 ψ̂1 ˆ
ξi δ ki dkj .
0,1,2,...,n+1
n=1 i=2 i=0 j=0
∂ξ δH ∂ψ δH (4)
ρ = , ρ =− (2)
∂t δψ ∂t δξ
where we set ρ = 1 [13]. Computing the expansion ker-
(n+2)
with the Hamiltonian given as the sum of the potential nels E0,1,2,...,n+1 relies on a series reversion associated
energy measured with respect to the undisturbed fluid with expressing the velocity potential in terms of its value
level y = 0 and the kinetic energy [9]: at the free surface [14]. These kernels can be obtained
in principle to all orders, however, the iteration process
! quickly becomes cumbersome. A fully explicit recursion
Z ∞ Z ξ (n+2)
ρ relation for computing E0,1,2,...,n+1 without a laborious
H= dx gξ 2 + |v|2 dy . (3)
2 −∞ −∞
series reversion has been proposed recently in [15] for an
arbitrary depth and arbitrary space dimension.
The Fourier transform is a canonical transformation,
The Laplace equation (1c) and the decay condition at and hence the evolution equations (2) preserve their
infinity (1d) represent the constraints on the dynamical Hamiltonian form in the canonically conjugate variables
system whose total energy H is a constant of motion. ξˆk and ψ̂k :

∞ Z n+1 n+1
!
∂ ξˆk δH X (n+2)
Y
ˆ
X
= = |k|ψ̂k + 2 E−0,1,2,...,n+1 ψ̂1 ξi δ k − ki dk12...n+1 , (5a)
∂t δ ψ̂k∗ n=1 i=2 i=1
∞ Z n+1 n+1
!
∂ ψ̂k δH X (n+2)
Y X
=− = −g ξˆk − nE1,2,...,n+1,−0 ψ̂1 ψ̂2 ξˆi δ k − ki dk12...n+1 (5b)
∂t δ ξˆk∗ n=1 i=3 i=1

where (4) was used to compute the variational derivative.

The expansion kernels posses a remarkable property re-
vealed by the recursion relation: for any fixed n ∈ N, it ξt± ± iψx± = 0 (8a)
2
holds that (iψx± )
ψt± + gξ ± = (8b)
1 ∓ iξx±
(n+2)
E− Pn+1 k =0 (6)
i=1 i ,k1 ,k2 ,...,kn+1 where f ± = P ± f , P ± = 12 (I ± iH) are orthogonal self-
adjoint projections onto positive and negative wavenum-
if all wavenumbers have the same sign [15] and ber components, H is the Hilbert transform [16]. In sum-
mary, the small-amplitude assumption, |ξx |  1, in con-
σ n+1 Y
n+1 junction with the unidirectionality of wave propagation
(n+2)
Ek Pn+1 =− ki (7) for which relations (6)-(7) are valid in one dimension al-
1 ,k2 ,...,kn+1 ,− i=1 ki n(2π)n/2 i=1
lows us to write apparently intangible equations (5) con-
taining infinite sums in a concise form given in (8).
with σ = −1 (σ = +1) if all wavenumbers are positive A few remarks are in order. First, making a crude ap-
−1
(resp. negative). Hence, by considering waves traveling proximation (1 ∓ iξx± ) = 1+O (ξx± ) in (8b) and letting
in one direction we can drastically simplify the equations g = 0 reduces the system (8) to the equations (17)-(18)
of motion. Indeed, taking the inverse Fourier transform derived in [17] to study the formation of singularities on
of (5) and using (6)-(7) we obtain the free surface of an ideal fluid in the absence of grav-
 3

itational force. Second, differentiating (8a) with respect

to time and (8b) with respect to space yields the conser-
vation form
± 2

!
i ξ t
ξt± t ± − igξ ±


=0
1 ∓ iξx±
x

which uncouples the free surface elevation from ψ ± . Last,

using (8a) it is possible to rewrite (8b) as
2
1 1 (ξt + ξx ψx )
ψt + gξ + ψx2 − =0
2 2 1 + ξx2
and (8a) itself as a non-local equation Figure 1. Wave profiles (not to scale) for three values of the
Z ∞ dimensionless parameter ka proportional to the wave slope:
dxe−ikx+|k|ξ (iξt − sgn(k)ψx ) = 0 ka = 0.3678 (top), ka = 0.3 (middle) and ka = 0.15 (bottom).
−∞ The crest loses its differentiability when ka = e−1 . The initial
valid for every value of k ∈ R. Thus we can recover from phase is α = 0 and g = 1.
(8) the non-local formulation of Euler’s equations due to
Ablowitz, Fokas and Musslimani (c.f. [18], p. 326). The
AFM formulation is particularly useful for deriving var-
ious asymptotic approximations including the nonlinear
Schrödinger equation describing the envelopes of waves
in deep water.
General solution.—Next we attempt to find a periodic
traveling wave solution inserting the ansatz f ± (x, t) =
f ± (kx ∓ ωt) with k > 0 to (8). The resulting system of
ODEs is solved by the Lambert W -function
1  
ξ ± (x, t) = − W −kae±i(kx∓ωt+α) (9a)
k
iω  
Figure 2. Extreme wave profiles (not to scale) with the same
ψ ± (x, t) = 2 W −kae±i(kx∓ωt+α) + C (9b)
k horizontal distance between two successive maxima: Stokes
wave (top), real part of ξ ± (x, t) (middle) and Gerstner wave
√ ω verifies the deep water dispersion relation, i.e. ω =
if
(bottom). Stokes wave is plotted using a close one-term piece-
gk. The two arbitrary constants in the solution of the wise approximation suggested in [21]. Limiting Gerstner wave
equations of motion are represented by an additive con- is an upside-down cycloid.
stant C ∈ C which we can set to zero without loss of
generality and the complex amplitude A = aeiα whose
argument is the initial phase α and the absolute valuewhere λ = 2π/k is the wavelength. This value is
defines the first-order wave amplitude a (see (10) be-smaller than the maximum wave steepness for the Ger-
low). The real part of ξ + (ξ − ) describes a wave pro-
stner wave, (h/λ)max = π −1 ≈ 0.3183 [22], but larger
file moving to the right (resp. left) with speed c = ω/k
than that for the highest Stokes wave in deep water,
(see Fig. 1). One can check directly that φ(x, y, t) =(h/λ)max ≈ 0.1412 [23] as shown in Fig. 2. The free
ky ±i(kx∓ωt+α)
− iaω
k e e satisfies the system (1), is irrota-
surface velocity potential ψ ± (x, t) is zero at the crest
tional ∇ × v = 0 [19], and equals ψ ± (x, t) on the free
and trough of the wave and develops a jump disconti-
surface since for any z ∈ C the Lambert function obeysnuity at the crest as ka approaches e−1 . The horizontal
the identity z = W (z)eW (z) . Unless otherwise stated,
velocity is highest at the crest of the wave and lowest at
W (z) will denote the principle branch of the function. It
the trough whereas the vertical velocity is zero at both
has a branch cut along the negative real axis, ending at
these locations.
−e−1 [20].
Discussion.—Asserting the flow of physical interest to
The wave profile changes from sinusoidal when ka is
be free of rotation, Lamb [24] and Stokes did not endorse
small to the wave with a sharp crest of angle 0◦ when
Gerstner’s exact solution as much as naval architects and
ka = e−1 ≈ 0.3678. The position of the crest relative to
engineers. In 1847 Stokes proposed an approximate solu-
the undisturbed water level is −k −1 W (−ka) and that of
tion for irrotational motion by means of a perturbation
trough is −k −1 W (ka) so that the height of the wave is
series: he sought the free-surface elevation as an infinite
given by P∞
Fourier series n=1 an cos (n(kx − ωt)) with unknown
± ± W (ka) − W (−ka) coefficients an and without discussing the convergence
h = max ξ − min ξ = .
x∈R x∈R k properties. The convergence of Stokes’ series for small-
amplitude deep water waves was proved by Levi-Civita
The maximum wave height occurs for the wave steepness
[25] in 1925. A representation similar to the Stokes’ ex-
1 + W (e−1 )
 
h pansion can be written for the waves studied in this Let-
= ≈ 0.2034
λ max 2π ter. Indeed, for small values of the parameter ka using
 4

the Taylor series of the Lambert function we obtain Krasovskii conjectures for the wave of greatest height,
Stud. Appl. Math. 98, 311 (1997); C. J. Amick, L. E.
3 Fraenkel and J. F. Toland, On the Stokes conjecture for
Re ξ ± (x, t) = a cos β + ka2 cos(2β) + k 2 a3 cos(3β)


2 the wave of extreme form, Acta Math. 148, 193 (1982);
8 3 4 P. I. Plotnikov, A proof of the Stokes conjecture in the
+ k a cos(4β) + O (ka)5


(10) theory of surface waves, Stud. Appl. Math. 3, 217 (2002).
3
[6] P. I. Plotnikov and J. F. Toland, Convexity of Stokes
√ waves of extreme form, Arch. Rational Mech. Anal. 171,
where β = ± (kx ∓ ωt + α) with ω = gk. The radius
349 (2004).
of convergence of the series for W (z) around z = 0 is [7] G. D. Crapper, An exact solution for progressive capil-
precisely equal to 1/e, i.e. to the value of ka for the lary waves of arbitrary amplitude, J. Fluid Mech. 2, 532
wave of greatest height. One can compare expression (1957).
(10) to the corresponding fourth-order Stokes expansion [8] A. Constantin, An exact solution for equatorially trapped
in deep water waves, J. Geophys. Res.: Oceans 117, C05029 (2012); A.
Constantin, Some three-dimensional nonlinear equatorial

1 2 17 3 4
 flows, J. Phys. Oceanogr. 43, 165 (2013); D. Henry, On
ξ(x, t) = a cos β − ka + k a cos(2β)+ three-dimensional Gerstner-like equatorial water waves,
2 24 Phil. Trans. R. Soc. A 376, 20170088 (2017).
3 2 3 1 3 4 [9] V. E. Zakharov, Stability of periodic waves of finite am-
+ k a cos(3β) − k a cos(4β) + O (ka)5


8 3 plitude on the surface of a deep fluid, J. Appl. Mech.
Tech. Phys. 9, 190 (1968).
where the dispersion relation involves the amplitude [10] Recall that Gerstner waves do not have any explicit rep-
ω 2 = gk(1 + k 2 a2 + . . . ) (see, e.g., Lamb p. 419). resentation in Eulerian variables.
[11] R. S. Johnson, A Modern Introduction to the Mathemat-
Stokes theory is limited to small amplitude waves and
ical Theory of Water Waves (Cambridge Univ. Press,
due to the convergence issues cannot yield the extreme Cambridge, 1997), p. 102; D. Henry, On Gerstner’s water
wave for any value of the water depth. Nevertheless, wave, J. Nonlin. Math. Phys. 15, 87 (2008).
Stokes gave an elegant argument to show that if a cusp [12] D. V. Chalikov, Numerical Modeling of Sea Waves
is attained in an irrotational flow, then its tangents at (Springer, Cham 2016).
the apex necessarily make an angle of 120◦ . How does [13] We use a compact notation in which ξˆj = ξ(k ˆ j , t), Ql
m=n
this agree with our finding that the sharp-edged extreme for l < n is 1, dk123 = dk1 dk2 dk3 etc., and subscript 0
wave has an included angle of 0◦ at the crest? To answer corresponds to R k. The star denotes complex conjugation,
fˆ(k) = √12π R f (x)e−ikx dx is the Fourier transform of
this question let us switch to a frame moving with the R ikx
1
crest and introduce polar coordinates r, θ with the origin f (x) and the Dirac delta function is δ(k) = 2π R
e dx.
[14] V. P. Krasitskii, On reduced equations in the Hamilto-
at the vertex, i.e. let x + i(y + k −1 W (−ka)) = reiθ
nian theory of weakly nonlinear surface waves, J. Fluid
where θ is measured from one of the branches of the Mech. 272, 1 (1994); M. Stiassnie and L. Shemer, On
wave. In the vicinity of the vertex the velocity potential modifications of the Zakharov equation for surface grav-
behaves as φ ∝ rn cos(nθ) so that the tangential and ity waves, J. Fluid Mech. 143, 47 (1984); M. Glozman,
normal velocity components are v r ∝ rn−1 cos(nθ) and Y. Agnon and M. Stiassnie, High-order formulation of
v θ ∝ rn−1 sin(nθ), respectively. In steady flow, according the water-wave problem, Phys D 66, 347 (1993).
to the Bernoulli equation we have v 2 ∝ r which implies [15] N. S. Ussembayev, Non-interacting gravity waves on the
surface of a deep fluid, arXiv:1903.10854.
that n = 3/2. The normal velocity must vanish for a
[16] The Hilbert transform is defined by (Hf ) (x) =
point on the wave profile and hence 3/2θ = π, giving the R (x0 ) 0
1
p.v. R fx−x0 dx , He
ikx
= −isgn(k)eikx .
Stokes’ result. However, θ = 120◦ is not the only solution π
[17] E. A. Kuznetsov, M. D. Spector and V. E. Zakharov,
to v θ = 0 as the trivial solution θ = 0◦ also satisfies it. Formation of singularities on the free surface of an ideal
fluid, Phys. Rev. E 49, 1283 (1994).
The author is supported by the KAUST Fellowship.
[18] M. J. Ablowitz, A. S. Fokas and Z. H. Musslimani, On a
new non-local formulation of water waves, J. Fluid Mech.
562, 313 (2006).
[19] In a two-dimensional setting the vorticity vector is iden-
∗
tified with its middle component which is orthogonal to
[email protected] the plane of the flow.
[1] F. Gerstner, Theorie der Wellen, Ann. Phys. 32, 412 [20] R. M. Corless et al., On the Lambert W -function, Adv.
(1809); A. Constantin, On the deep water wave motion, Comp. Math. 5, 329 (1996).
J. Phys. A: Math. Gen. 34, 1405 (2001). [21] R. C. T. Rainey and M. S. Longuet-Higgins, A close one-
[2] W. J. M. Rankine, On the exact form of waves near the term approximation to the highest Stokes wave on deep
surface of deep water, Phil. Trans. R. Soc. Lond. 153, 127 water, Ocean Eng. 33, 2012 (2006).
(1863); W. Froude, On the rolling of ships, Trans. Inst. [22] J. V. Wehausen, E. V. Laitone, Surface Waves in Ency-
Naval Arch. 2, 180 (1861); W. J. M. Rankine, On the clopaedia of Physics (Springer-Verlag, Berlin, 1960), Vol.
rolling of ships, Trans. Inst. Naval Arch. 13, 62 (1872). 3, p. 744.
[3] R. L. Wiegel and J. W. Johnson, Elements of wave the- [23] L. W. Schwartz and J. D. Fenton, Strongly nonlinear
ory, Coast. Eng. Proc. 1, 5 (1950). waves, Ann. Rev. Fluid Mech. 14, 39 (1982).
[4] G. G. Stokes, Mathematical and Physical Papers (Cam- [24] H. Lamb, Hydrodynamics (Cambridge Univ. Press, Cam-
bridge Univ. Press, Cambridge, 1880), Vol. 1, p. 319. bridge, 1975), p. 421.
[5] J. F. Toland, On the existence of a wave of greatest [25] T. Levi-Civita, Détermination rigoureuse des ondes per-
height and Stokes’s conjecture, Proc. Roy. Soc. Lond. manentes d’ampleur finie, Math. Ann. 93, 264 (1925).
Ser. A 363, 469 (1978); J. B. McLeod, The Stokes and