Lecture notes on linear wave theory.

Lectures given at the summer school on:

WATER WAVES and OCEAN CURRENTS.
Nordfjordeid 21-29 june 2004.
Kristian B Dysthe
Department of Mathematics
University of Bergen
Norway

June 2, 2004

1 Introduction.
To give an introduction to linear wave theory for surface waves lasting for a
few hours is a nearly impossible challenge. There is no time for mathematical
details, yet the theory is mathematical in its nature. The notes are probably
going to contain more details than the lectures. Still they are rather sketchy.
Consequently I shall have to rely on my listeners’ ability to fill in the details
that are left out.
Excellent books for further reading are for example the following:

• G.B.Whitham : Linear and Nonlinear Waves. John Wiley & Sons,

1974.

• J. Lighthill : Waves in Fluids. Cambridge University Press, 1978.

• G.D. Crapper : Introduction to Water Waves. Ellis Horwood Limited

1984.

1
2 Basic equations.
We start by assuming that our fluid is of homogeneous density ρ , and also
ideal and incompressible. Consequently the continuity equation is simply

∇·v = 0 (1)

We shall also assume that vorticity has no major place in wave propaga-
tion. This, however, calls for a comment. From the theory of an ideal and
homogeneous fluid we recall that the vorticity ∇ × v is a property associ-
ated with the fluid elements. It is carried along by the fluid motion. This
implies that if a particular fluid element had zero vorticity initially, it will
always have zero vorticity. The main property of a wave is its ability to
transport information, energy and momentum over considerable distances
without transport of matter. Thus the velocity field associated with the
wave is irrotational and given by a velocity potential, ϕ , which according
to the equation (1) above satisfies the Laplace equation

∇2 ϕ = 0 (2)

The boundary conditions for (2) at the bottom is simply

∂ϕ
=0 for z = −h (3)
∂z
where we use the oceanographic convention: the z-axis pointing vertically
upwards with z = 0 at the equilibrium surface. The actual surface is located
at
z = η(t, x, y)
and the kinematic surface condition states that a ”fluid particle” at the
surface at any given time is always at the surface:
d dη ∂ϕ
(z − η(t, x, y)) = 0 ⇐⇒ = (4)
dt dt ∂z
where we use the notation
dA ∂A
≡ + ∇ϕ · ∇A
dt ∂t
There is also a dynamical condition at the free surface. Above (i.e. for
z > η) there is an atmospherical pressure pa which is taken to be constant.

2
Below (z < η) the pressure must be calculated from the Bernoulli equation,
which by our previous assumptions can be written
∂ϕ 1 p
+ (∇ϕ)2 + gz + = constant
∂t 2 ρ

The constant must be equal to pa /ρ if we allow for the fluid to be at rest

and undisturbed at some distant region of the surface (show this!). Thus at
the surface we have
∂ϕ 1
pa − p = ρ( + (∇ϕ)2 + gη) (5)
∂t 2
and this pressure difference would cause an infinite acceleration, if nonzero
and with no forces to balance it. The surface tension give a balancing force.
Using the Laplace formula we have the following condition for equilibrium
between these forces
p − pa = κT (6)
Here T is the surface tension (assumed to be constant), and κ is the mean
curvature of the surface given by
 
−∇η
κ = ∇ · n = ∇ · q  (7)
1 + (∇η)2

where n is the unit normal to the surface (pointing upwards). Thus on the
free surface we have the two conditions
 
dη ∂ϕ ∂ϕ 1 T ∇η
= , + (∇ϕ)2 +gη = ∇· q  at z=η
dt ∂z ∂t 2 ρ 2
1 + (∇η)
(8)
These equations are of course non-linear, and therefore rather difficult to
handle. In the following we shall limit ourselves to the case of small pertu-
bations from an equilibrium.

3 The linearized problem.

For waves of small steepnes (i.e. for |∇η| << 1 ) the nonlinear terms are
small. We can therefore hope that a linearization procedure can give us a
good approximation to the properties of these waves. As you will find out

3
later in the course, this is not always true: even small nonlinear terms can
(given sufficient time or fetch) produce large effects.
Nevertheless let us proceed to linearize the conditions (8) by neglecting
quadratic and higher order terms in η and ϕ . We remark that
∂ϕ
ϕ(t, x, y, η) = ϕ(t, x, y, 0) + (t, x, y, 0)η + third and higher order
∂z
This implies that the terms containing ϕ is to be evaluated at z = 0 upon
being linearized. The linearized version of (8) now becomes (fill in the
details!)
∂η ∂ϕ ∂ϕ T
= , + gη = ∇2 η at z=0 (9)
∂t ∂z ∂t ρ
The linearized problem is then to solve (2) with the boundary conditions
(3) and (9). We note that these can be formally derived as the lowest order
equations in a development in powers of a small number  which characterizes
the smallness of perturbation of the surface (e.g. some characteristic value
of |∇η| , which is a pure number). In the next section we derive a simple
class of solutions, namely those corresponding to a ”plane wave” of wave
number k and an amplitude a. A convenient small number is then  = ka
called the wave steepness.

3.1 Elementary solutions and the dispersion relation.

The most natural thing to do under these circumstances is probably to look
for solutions where the surface is varying harmonically in time and in the
horizontal spatial coordinates i.e. like a ”plane” wave

ei(k·x−ωt)

where k = (kx , ky ) is a horizontal wave vector, and x = (x, y). Thus we

make the ansatz

η = Aei(k·x−ωt) + c.c. and ϕ = B(z)ei(k·x−ωt) + c.c.

Inserting this in Laplace equation (2) B is found to satisfy the equation

∂2B
− k2 B = 0
∂z 2
with elementary solutions sinh(kz), cosh(kz). Here k = |k| is the wavenum-
ber which is related to the wavelength λ by λ = 2π/k . A solution satisfying

4
the lower boundary condition is then readily found to be

cosh [k(z + h)]

B=C
cosh(kh)

Inserted into the linearized surface boundary conditions (9) one obtains the
equations
iωA + k tanh(kh)C = 0
and
T
(g + k 2 )A − iωC = 0
ρ
The condition for existence of a plane wave solution (i.e. a nonzero solution
A and C) then becomes

T 3
ω 2 = (gk + k ) tanh(kh) (10)
ρ
This relation between the frequency ω and the wave number k is called the
dispersion relation. In real form the plane wave solution can now be written

η = a cos(k · x − Ω(k)t) (11)

and
Ω(k) cosh [k(z + h)]
ϕ= a sin(k · x − Ω(k)t) (12)
k sinh(kh)
where Ω(k) is a solution of the dispersion relation (10), and a is the
real amplitude of the wave. Before looking at more general solutions of the
linearized equations we consider how the dispersion relation (10) can be
simplified in some special parameter domains.

3.2 Waves of different wavelengths.

We rewrite (10) as
k 2
ω 2 = gk(1 + ( ) ) tanh(kh)
k0
where r
gρ
k0 =
T
is a characteristic wave number (corresponding to a wavelengh of 1.71cm for
pure water). The different parameter regimes of the two numbers kh and
k/k0 serves to distinguish between different wavetypes:

5
 • kh >> 1 deep water waves.

• kh << 1 shallow water waves.

• k/k0 << 1 gravity waves.

• k/k0 >> 1 capillary waves.

For deep water gravity waves the dispersion relation is simplified to

ω 2 = gk (13)

and for shallow water gravity waves one have

ω 2 = ghk 2 (14)

A graph of the dispersion relation for gravity waves is shown in Figure

(1). The approximations (13) and (14) are also shown.

The wave pattern (crests and troughs) moves with the phase velocity vph =
ω/k. For waves on deep water we have

2 g gk
vph = + 2
k k0
√
from which it is found that vph has a minimum of 2(T g/ρ)1/4 ('
23cm/s, pure water) for k = k0 .

3.3 Motion of the fluid particles.

The motion of a fluid particle due to the wave is found by integrating the
equation of motion
dr
= ∇ϕ (15)
dt
where r =(x(t), y(t), z(t)) gives the position of the fluid particle, and the
right hand side is evaluated at that point. This is of course a nonlinear
differential equation even though we shall use the linearized solution (12)
for ϕ. We expect the motion to consist of a periodic ocillation s(t) and
possibly a slow translatory motion of the average position, or guiding center
R(t) = (X(t), Y (t), Z(t)). We write r = R + s . To lowest significant order
in the wave steepness we evaluate ∇ϕ at the guiding center and neglect

6
the variation of R during a wave period. The equation for s now becomes
(taking the x-axis parallel to k)
ds
= Ω(A cos θ, 0, B sin θ)
dt
where θ = k · R−Ωt and
a cosh [k(Z + h)] a sinh [k(Z + h)]
A= and B=
sinh(kh) sinh(kh)
which is readily integrated to give
s = (−A sin θ, 0, B cos θ)
The trajectory of the ocillating movement is found by eliminating θ giving
x−X 2 z−Z 2
   
+ =1
A B
which is an ellipse with half axis A (horizontal) and B (vertical). It is readily
seen that B = a (the amplitude) for a surface particle and zero for a bottom
particle. For a wave on deep water we have the simplification
A = B = aekZ
implying that the trajectories are all circles with radius a at the surface,
and decreasing exponentially downwards (see Figure (2) ).
The equation for the slow motion of the guiding center is obtained by
going to the next order in wave steepness. This seems a bit strange since we
still use the expression (12) which is derived from linear wave theory. It is
not difficult, however, to show that by taking into account the next order of
approximation for ϕ one does not change the result for the guiding center.
Developing the right hand side of (15) in powers of s, and averaging over
one wave period (keeping R constant), we obtain
dR 1 cosh [2k(Z + h)]
= hs·∇∇ϕi = kΩa2
dt 2 sinh2 (kh)
This is a slow horizontal drift of the fluid particles in the wave direction
called Stokes drift. The total mass transport M due to the wave is found
by integrating the Stokes drift from the bottom to the surface
Z 0
dR ρΩa2 k
M= ρ dz = (16)
−h dt 2k tanh(kh)
For deep water we have the simplifications
dR k k
= Ωa2 e2kz and M = ρΩa2
dt k 2k

7
4 The group velocity.
As long as we consider linear wave theory, solutions can be added to produce
new solutions. Since integration is a linear operation the perturbation given
by
Z
η = Re A(k)ei(k·x−Ω(k)t) dk
Z
Ω(k) cosh [k(z + h)] i(k·x−Ω(k)t)
ϕ = Re A(k) e dk
k sinh(kh)
is still a solution provided Ω(k) satisfies the dispersion relation. Here A(k)
is an arbitrary function of k . With the integration taken over the entire k
-plane the above solution take the form of Fourier integrals, and we shall use
the phrase Fourier component about the plane wave solution ei(k·x−Ω(k)t) .
For a solution that is a sum or an integral over elementary solutions,
the question of a wave velocity comes up since vph is a function of the
wavenumber. Take the simple case with two waves of equal amplitude and
slightly different wave vectors k − ∆k and k + ∆k
h i
η = Re a(ei((k−∆k)·x−Ω(k−∆k)t) + ei(k+∆k)·x−Ω(k+∆k)t ) (17)
 
∂Ω
' 2a cos ∆k · (x− t) cos(k · x−Ω(k)t)
∂k
It is seen from the expression above that the combination of two waves of
slightly different wave vectors behaves like a single wave with a wave vector
being the average of the two and with a slowly varying amplitude given by
 
∂Ω
2a cos ∆k · (x− t)
∂k
It is seen from this expression that the amplitude is transported with the
velocity
∂Ω ∂Ω ∂Ω
vg ≡ =( , ) (18)
∂k ∂kx ∂ky
the socalled group velocity. The group velocity is generally different from
the phase velocity. The exception is gravity waves on shallow water where
there is a linear relation between ω and k (see equation (14)). For gravity
waves on deep water it is readily found that vph = 2vg . In Figure (3) we
show how the phase and group velocities vary with the wave number for
gravity waves. It is seen that they both decrease monotonically with wave
number (or increase with wave length).

8
4.1 Energy and momentum.
It is intuitively evident that the physical quantities energy and momentum
associated with the wave is transported in the same way as the amplitude
i.e. with the group velocity. We shall now derive the expressions for the
average energy and momentum in an elementary wave. For simplicity we
consentrate on gravity waves. The average (available) potential energy Ep
is given by Z η 
gρ
2 
Ep = gρzdz = η
0 2
The average kinetic energy Ek (energy per unit surface area) is given by
 Z η 
ρ 2
Ek = (∇ϕ) dz
2 −h

where hi denotes averaging over a wave period.Since we shall limit ourselves

to linear wave theory, only the quadratic part of Ek is relevant. The upper
boundary of the integral can therefore be taken to be 0, as the difference is
of third order in the perturbation. Inserting the expressions (11) and (12)
into the relations above we get

ρ(aΩ)2
Ek = (19)
4k tanh(kh)
and
gρa2
Ep =
4
Using the dispersion relation it is seen from these expressions that Ek = Ep ,
which is a general result in linear wave theory. Since the average kinetic- and
potential energies are equal for a propagating wave, the total wave energy
E can be written as

 1
E = gρ η 2 = gρa2 (20)
2
The energy flux, F, associated with the wave ((11) and (12)) is
Z 0   
2 Ωk 2kh
F= pudz = ρga 1+ = Evg
−h 4k 2 sinh(2kh)
The velocity of energy transport is therefore the group velocity.
The average momentum associated with an elementary wave is
Z η
ρΩa2

k
P= ρ∇ϕdz ' ρ hη∇ϕiz=0 = k = E (21)
−h 2k tanh(kh) Ω

9
By comparing this with the expression (16) it is seen that the average mo-
mentum is equal to the total mass transport.
These expressions for the average energy and momentum remain valid as
an approximation even if the amplitude of the wavetrain is slowly varying.
If L is a characteristic length for a significant variation of the amplitude
(corresponding to 1/ |∆k| in the example above) then (20) and (21) are
correct to the order (kL)−2 . By a slow variation it is understood that the
amplitude has a very small relative variation during a wave period.
Since both E and P are quadratic in the amplitude, the transport ve-
locity for these quantities is also the group velocity as already anticipated.

5 The initial value problem.

The linear superposition of elementary waves can be used to solve initial
value problems. In the following we consider such a problem for a gravity
wave on deep water travelling along a channel (one dimensional propaga-
tion). Let the fluid be at rest initially (i.e. ϕ = 0 at t = 0) with a perturbed
surface. We take the initial perturbation for η to be an impuls function. If
that problem can be solved (as it was by Cauchy and Poisson in 1816) the
solution for a more general initial value can be found by convolution. The
initial conditions are now
∂η
η(0, x) = δ(x) , (0, x) = 0
∂t
Consider now a general solution of the linearized equations
Z ∞ Z ∞
i(kx−Ω(k)t)
η(t, x) = R1 (k)e dk + R2 (k)ei(kx+Ω(k)t)
−∞ −∞

written as Fourier integrals. The first integral represents a wave moving to

the right, and the second a wave moving to the left (corresponding to the
two solutions ±Ω(k) of the dispersion relation). From the initial conditions
we obtain
Z ∞
∂η
(0, x) = −iΩ(k)(R1 (k) − R2 (k))eikx dk = 0
∂t −∞

and Z ∞
η(0, x) = (R1 (k) + R2 (k))eikx dk = δ(x)
−∞
implying that
1
R1 = R2 =
4π

10
Thus the solution becomes
Z ∞
1
η(t, x) = cos(kx) cos(Ω(k)t)dk (22)
π 0

which is an exact, nice and compact solution. The content of it is , however,

far from easy to see. Next we show how some important information can be
extracted from it.

5.1 Asymptotic solution of the initial value problem.

In the above initial value problem we consentrate on the waves moving to
the right. The challenge is then to find an asymptotic approximation (far
from the initial impuls) to the Fourier integral representing these waves
Z ∞ Z ∞
1 1
η(t, x) = ei(kx−Ω(k)t) dk = eitw(k) dk (23)
4π −∞ 4π −∞

where
x
− Ω(k)t
w=k
t
The leading term in an asymptotic development of this integral for large t
is known to come from a small area around the points of stationary phase
i.e. the solution of the equation
dw dΩ x
=0 ⇔ = (24)
dk dk t
The physical interpretation of this relation is that the main contribution
at time t and location x comes from the Fourier component whose group
velocity is exactly right for travelling the distance x in the timespan t. Let
K( xt ) be the relevant solution of (24) with respect to k. The leading term of
the asymptotic expansion of the integral (23) is (using the socalled stationary
phase method and observing that w00 > 0 for a gravity wave)
1 h πi
η(t, x) ' p cos tw(K) + = A cos θ (25)
2πtw00 (K) 4

where A is a slowly varying amplitude and θ is the wave phase. For deep
water waves (i.e. Ω = g |k|) an explicite solution K( xt ) of equation (24)
p

can be found as
t
K = g( )2
2x

11
giving (show this)

gt2
r
t g
θ=− and A=
4x 2 πx3

The solution (25) represent a slowly varying wavetrain that locally looks like
a plane wave with a local wave number K and frequency Ω(K) where

∂θ ∂θ
=K and = −Ω(K)
∂x ∂t
Show that K and Ω thus defined satisfies the dispersion relation. It is also
straight forward to show directly that the relative variations in the quantities
K, Ω and A over a wave period is small as long as Ωt >> 1.
Out of all this emerges the following picture: Initially a compact region
is disturbed. The Fourier spectrum of the initial disturbance is rather broad-
band. The waves corresponding to each Fourier component starts moving.
Its part of the energy is transported with the corresponding group veloc-
ity. At first all these waves add up to some rather involved pattern. After
a while, since they move with different velocities, an ordering takes place
and increases with time: The longest waves in front and the shortest in the
rear. In fact if after a long time one observes the train going by, the local
frequency is increasing linearily with time since
∂θ gt
Ω=− = (26)
∂t 2x
Although we developed these results under rather special conditions (one
dimensional propagation and an impulsive initial condition) they can readily
be generalized. For two dimensional propagation a corresponding version of
the stationary phase method can be used and the condition for stationary
phase is then
 2
∂Ω x t x
= with solutions K=g for deep water waves
∂k t 2r r

where r = |x| . The asymptotic wave phase θ now becomes

gt2 ∂θ
θ=− with K =∇θ and Ω=−
4r ∂t
An example of the effects discussed above is the common experience when
throwing a stone into a pond. The circular symmetric wavetrain resulting

12
from the splash have the long waves in front and the short waves in the rear.
As another example consider someone who is recording the frequency of the
swell arriving at a beach from a storm distant both in time and space (see
Figure (4)). If the storm was of short duration compared with the transit
time of the swell, and if all the incoming swell came from that source one
would expect the frequency to be a linear function of time. Fitting a straight
line to the measured points one can determine the distance to- and the time
of occurence of the storm. Indeed such measurements have been conducted
(Snodgrass et al., 1966, see Figure (5)).

5.2 The wave front.

The stationary phase method is an effective tool for extracting asymptotic
information from the solution of an initial value problem. In some cases
we have to modify the tools. Take for example the case of one dimensional
propagation of gravity waves in a channel discussed above. The question:
what does the front of the propagating disturbance look like? cannot be
answered by using the standard stationary phase method. The mathematical
reason for this is the fact that the stationary points ñK( xt ) merges at
the origine as xt tends to the maximum wavespeed gh. This happens
when k → 0 . Thus the secrets of the front seems to be buried in the
neighbourhood of k = 0 in the integral solution (23). Developing w around
this point to third order in k we have
√
x p ghh2 3
w ' k( − gh) + k
t 6
Inserting the expansion into the integral (23) we have
Z ∞ √
t ghh2 3
 
1 p
η' exp ik(x − ght) + i k dk
4π −∞ 6
Comparing this to the integral representation of the Airy function
Z ∞  
1 1 3
Ai(τ ) = exp i(τ p + p ) dp
2π −∞ 3
we obtain the following approximation valid around the wavefront
√
1 x − ght
η' Ai( ) (27)
2Ct1/3 Ct1/3
where  2 √ 1/3
h gh
C=
2
This is exibited in Figur (6) at some different times.

13
6 Current and refraction.
It is only possible to give a sketchy introduction to this theme. Let us start
with the case of a uniform (horizontal) current U. It can be considered as
a Gallilei transformation to a coordinate system moving with the velocity
-U with respect to the fluid. Going back to the linearized equations it is
∂ ∂
readily seen that the only difference is that ∂t is changed to ∂t + U · ∇. In
the dispersion relation this implies the transformation ω → ω − U · k . Thus
the dispersion relation (10) becomes
T 3
(ω − U · k)2 = (gk + k ) tanh(kh) (28)
ρ
The solutions of this equation we write
±Ω(k) + U · k
where Ω(k) is again a solution of the original dispersion relation (10). The
extra term, representing a frequency shift is just the Doppler shift. The
group velocity for the wave ω = Ω(k) + U · k is seen to be
∂ω ∂Ω
= +U
∂k ∂k
Let us next investigate the changes in the mean energy and momentum from
the expressions (20) and (21). We first remark that the potential energy is
unaltered. The kinetic energy is now
 Z η 
ρ 2 ρ
Ek = (∇ϕ + U) dz − U2 h
2 −h 2
ρa2 (ω − U · k)2
' + hρηU · ∇ϕi
4k tanh(kh)
ρa2 (ω 2 − (U · k)2 )
=
4k tanh(kh)
which is not equal to the potential energy. Thus we have learned that
the equipartition of potential and kinetic energy only works when these
quantities are referred to a coordinate system at rest with respect to the
fluid (the ”rest frame”).
The mean total energy Eu in the ”lab frame” (i.e. in the coordinate
system where the fluid is flowing with the uniform velocity U ) is found to
be
ρa2 ω(ω − U · k) Ω+U·k
Eu = = E
2k tanh(kh) Ω

14
where E is the mean energy density in the rest frame (see equation (20)).
This shows that the total mean energy is not invariant under a Gallilei
transformation. It is also seen that the quantity N = E/Ω called action
density is invariant since
E Eu
N= = (29)
Ω Ω+U·k
The averaged momentum density is seen from (21) to be invariant
E
P=k = kN
Ω

6.1 Refraction
In reality a current is hardly uniform in space and time, also depth h varies
with the horizontal dimensions. However, in many cases the time and length
scales associated with these quantities are much larger than those of the
wave. This can be expressed as
     
1  1   1 ∂U 
k >>  ∇h
  , k >>  ∇U 
  and ω >>   
h U U ∂t 

It is then rather natural to assume that locally (in time and space) the wave
properties are the same as that of a corresponding plane wave under uniform
conditions. Formally this can be shown to be true by an asymptotic expan-
sion in a small parameter made from the ratio of wave and current scales (or
depth scales). The lowest order result from such an exercise is the socalled
geometric optics approximation or ray theory. Since the medium through
which the wave is propagating is slowly varying in space (and possibly in
time) the amplitude, wave number and (possibly) the frequency are varying
too. Starting with a ”locally plane” wave represented by

η = Re(a(εx,εt)eiθ )

where θ is the wave phase and a is a slowly varying amplitude (the slow
variation is made explicite by a small parameter ε ). The local frequency
and wave vector are then (as earlier) defined by
∂θ
ω(εx,εt) = − and k(εx, εt)=∇θ (30)
∂t
and are also assumed to be slowly varying. Making a similar anzats for the
potential ϕ and inserting this into the linearized equations one obtain to the

15
zero order in ε just the dispersion relation (10). To the first order in ε one
obtains an equation governing the variation of the amplitude which can be
written (after a lot of work!)
∂N ∂ω
+ ∇ · ( N) = 0 (31)
∂t ∂k
where N = E/Ω is the action density and E is given in terms of the ampli-
tude by the expression (20). This has the form of a conservation equation
and tell us that action is conserved and the action density is transported
with the group velocity, like some ”wave fluid” density. Now the stream lines
of this wave fluid are the rays, which can also be thought of as trajectories
of wave packets. They are defined through the equation
dx ∂ω
= (32)
dt ∂k
To integrate (32) we need a similar equation for k . It follows from the
relations (30) that
∂k ∂θ
=∇ = −∇ω
∂t ∂t
Through the dispersion relation ω is a function of k . It may also be an
explicite function of time and space through the physical parameters entering
the dispersion relation, i.e. the depth h and the current velocity U . The
right hand side of the equation above can now be developed as
∂2θ X ∂ω ∂kj ∂ω X ∂ω ∂ki ∂ω
− = + = +
∂xi ∂t ∂kj ∂xi ∂xi ∂kj ∂xj ∂xi
j j

where the last equality follows from the fact that k is a gradient vector. We
now get the equation
dk ∂ω
=− (33)
dt ∂x
d ∂
where dt = ∂t + ∂ω
∂k · ∇ . The two equations (32) and (33) can be considered
to be dynamical relations for the motion of wave groups. In fact they have
the canonical form of the Hamiltonian equations with ω corresponding to
the Hamiltonian and k to the momentum variable. This correspondence is
not very surprising considering the wave-particle correspondence first sug-
gested by Louis de Broglie. The conservation equation for action (31) is the
corresponding evolution equation for the amplitude. Together these three
equations constitute the ”geometrical optics” relations for surface gravity
waves.
In the following we give some examples.

16
6.2 Examples.
6.2.1 Swell approaching beach.
The frequency change of swell (that we considered earlier) is a slow process.
When considering the transition of swell from deep water to shallow water we
may consider the frequency of the incoming waves to be constant. We shall
also neglect any background current. The frequency during the transition
is also constant because we have
dω ∂ω dk ∂ω dx ∂ω ∂ω
= + · + · = =0
dt ∂t dt ∂x dt ∂k ∂t
which follows from the equations (32) and (33). Consequently it is the wave
vector that must change in order to satisfy the dispersion relation with a
changing depth. Consider a straight beach parallel to the y-axis, with a
depth that depends on the x-coordinate only. From (33) we then have

dky
=0
dt
thus ky is a constant. Since we now have a steady state problem equation
(31) can be integrated to give

∂ω ωkx 2kh
E= 2
(1 + )E = constant
∂kx 2k sinh(2kh)

where the constant is equal to the component of the energy flux in the
x-direction. Using the subscript 0 to denote the wave properties of the
incoming wave at deep water we obtain from the dispersion relation and the
equations above
E kx k02 2kh
= 2
(1 + )
E0 kx0 k sinh(2kh)
and
k cos(ψ0 )
= coth(kh) =
k0 cos(ψ)
where ψ is the angle between k and the x-axis. Explicite solutions can be
found when the wave has arrived at shallow water i.e. kh << 1. In this

17
approximation we have
 2
E a cos(ψ0 )
= = (k0 h)−1/2
E0 a0 2
k cos(ψ0 )
= = (k0 h)−1/2
k0 cos(ψ)
r
ak cos(ψ0 )
= (k0 h)−3/4
a0 k0 2
It is seen that as k0 h decreases:

• the energy (and thus the amplitude) increases,

• the wavelength decreases (and thus the wavenumber increases),

• the wave direction turns towards normal incidence (i.e. ψ → π/2),

• the most dramatic increase is experienced by the wave steepnes ak.

6.2.2 Waves against current.

Consider waves on deep water moving into a region where there is a current.
We shall assume a simple geometry with the waves moving parallel to the
x-axis in the positive direction and the current −U (x) going in the opposite
direction (the incoming waves come from x = −∞ and we assume U (−∞) =
0. As in the previous example we assume that the incoming waves have a
constant frequency and denote the wave properties of the incoming waves
(far from the current region) by subscript 0. The dispersion relation is now
p p
ω = gk − kU (x) = gk0 = constant
√
which is a quadratic equation in k with the solutions
r s
k 1 p U (x) k0
= (1 ± 1 − 2p) where p= = 2U (x)
k0 p vg0 g

There is a lot to be learnt from these solutions. First the question is why
there are two of them. The incoming wave must satisfy the condition that
k → k0 when x → −∞ (i.e. when p → 0 ). The solution with the minus sign
therefore corresponds to the incoming wave. It is also seen that solutions
only exist when p < 1/2 , i.e. when U (x) < vg0 /2. At the level x = xc
where U (xc ) = vg0 /2 the wave is reflected (or blocked). The reflected wave

18
corresponds to the solution with the plus sign, and at the level of reflection
the wavenumber is seen to be the same for the two branches and equal to
4k0 . Thus the wavelength of the incoming wave decreases to λ0 /4 . After
reflection it decreases much faster. This can be seen from Figure (7).
The conclutions above are only slightly modified when the waves and
the current are not colinear. In Figure (7) the waves originate from a point
source outside the current area. Note that waves coming in at different
angles with respect to the the current direction are reflected at almost the
same level.
It is left to the reader to show that the group velocity changes sign at
the level of reflection. Note that the incoming- and reflected waves both
have phase velocities opposite to the current velocity. On the other hand:
in a steady state the incoming wave energy must be carried away by the
reflected wave. Thus we arrive at the conclution that the reflected wave must
have its phase- and group velocities pointing in opposite directions. Such a
”backward” wave looks quite strange. Imagine that we are looking at the
surface near the reflecion level. Because our visual interpretation associate
the wave speed with the motion of the wave crests i.e. with the phase
velocity, it appears that two waves of different wavelength is approaching
the reflection level and then suddenly disappears without trace.
The amplitude and steepness of the waves are also increasing as one
approaches the reflection level. In a steady state this evolution is found
using the conservation equation (21) which now implies that the energy flux
is constant i.e.
r r
1 g 1 g
vg E = ( − U (x))E = vg0 E0 = E0
2 k 2 k0
which gives r
E k
=( − p)−1
E0 k0
This shows that the average energy density (and therfore the amplitude)
increases as we approaches the level of reflection. The geometrical optics
approximation is not valid at such an ”internal reflection” (or more generally
at a caustic). Still it is possible to find uniformly valid solutions of the
linearized equations in such regions. Even if the steepness of the incoming
wave (far from the reflection level) is small this may no longer be so as
the wave approaches reflection. Consequently the linear approximation may
become invalid.
In Figure (8) and (9) it is shown how some wave parameters are evolving
according to the relations above, when the wave meet with a counter-current

19
whose variation is taken to be Gaussian (see bottom of the figures). In Figure
(8) the current maximum is higher than the critical one (i.e. vg0 /2), such
that reflection occurs at a level x = xc . In Figure (9) the current maximum
is subcritical so that the wave is able to pass through the current region.

6.2.3 Ship waves.

It is wellknown that a ship moving at a constant velocity generate a V-
shaped wave pattern that is stationary in the reference frame moving with
the ship. It is somewhat less wellknown that the half-angle of the V is 19.5◦
and that the waves at the edge of the wave pattern (caustic) is moving at
an angle of 35.3◦ to that of the ship, regardless of the ship form and speed.
This was explained by Lord Kelvin more than a hundred years ago.
The explanation starts from two observations

• In the reference frame of the ship the wave pattern is stationary i.e.
the frequency is zero.

• The ship is the wave source, therefore (still in the ship-frame) the wave
energy at a point of the pattern must have travelled along a straight
line from the ship.

In the ship-frame the water has a uniform velocity −U and the condition
of zero frequency (assuming deep water) becomes
p p
ω = gk − k · U = gk − U k cos ψ = 0 (34)

The group velocity is then (with the x-axis parallel to the ship track, see
Figure (10))
r r r
1 gk 1 g 1 g
vg = − U = (− cos ψ + U, sin ψ)
2 kk 2 k 2 k

Simple trigonometry, and using equation (34) give us

vg sin ψ sin ψ cos ψ tan ψ

tan θ = = 2 = (35)
U − vg cos ψ 1 + sin ψ 1 + 2 tan2 ψ

where the angles θ and ψ are explained in Figure (10a). Solving for tan ψ
we obtain the two real solutions
1 p
tan ψ = (1 ± 1 − 8 tan2 θ)
4 tan θ

20
corresponding to the two wave-systems in the ”Kelvin” wake (see Figure
(10b)). These different waves meet in cusps along the perifery of the wave
pattern.
√ As seen from the formula above there are no solutions for |tan θ| >
2 2 i.e. |θ| > 19.3◦ . Further, the angle that the cuspwaves at the perifery
make with the ship track, is given by tan−1 ( √12 ) ' 35.3◦ . Figure (11)
shows a picture of a ship wake pattern, where only the transversal wave
system is in evidence. Which parts of the pattern are amplified and which
are suppressed, depend on the form and speed of the ship.

21