Models For Linear Wave Propagation
Models For Linear Wave Propagation
21
Introduction
In this section we derive several different models for the propagation of waves in fluids.
In each case, we assume the amplitude is sufficiently small for the equations to be
linearised. Our first example is acoustic waves in a gas, which are governed by the wave
equation and therefore move at a well-defined constant wave-speed. In contrast, gravity
waves on the surface of a layer of fluid are dispersive, meaning that waves with different
wavelengths move at different speeds. We also derive models for internal gravity waves
in a stratified fluid and inertial waves in a rotating fluid.
2.2
+ (u) = 0,
t
u
+ (u )u = p + g,
t
(2.1b)
(2.1c)
combined with (2.1a), provide a closed system for , u and p. For the moment, we will
assume that the body force g is negligible.
The system (2.1) is nonlinear and hence very difficult to solve in general. We therefore try to simplify the equations, by considering small disturbances about an assumed
uniform initial state. Suppose the fluid starts at rest with constant density 0 and
pressure p0 = P (0 ). We perturb about these initial conditions by setting
= 0 + ,
u = 0 + u ,
p = p0 + p ,
(2.2)
University of Oxford
in which the primed variables are assumed to be small.1 If we apply Taylors Theorem
to (2.1a) and neglect products of primed variables, we obtain
p = c20 ,
(2.3)
(2.4)
u = 0.
(2.6)
t
Hence, if u = 0 initially (which is true if the fluid starts from rest) then it is zero
for all time and we can write
u = ,
(2.7)
0
where is the velocity potential. Substituting (2.7) back into (2.5), we obtain
2
+ c0 = 0
0
t
(2.8)
+ c20 = F (t)
(2.9)
t
where F is an arbitrary scalar function. Since an arbitrary function of t may be added
to without affecting (2.7), we may set F (t) 0 without loss of generality and thus
obtain
+ c20 = 0.
(2.10)
0
t
This is the linearised version of Bernoullis equation.
Linearisation of (2.1b) gives
0
+ 0 2 = 0
(2.11)
t
and, by eliminating between (2.10) and (2.11), we find that satisfies the wave
equation:
2
= c20 2 .
(2.12a)
t2
From (2.3) and (2.10), it is clear that and p also satisfy exactly the same partial
differential equation, that is
2
= c20 2 ,
t2
2 p
= c20 2 p .
t2
(2.12b)
1
Exactly what we mean by small here may be quantified by nondimensionalising the equations and
identifying an appropriate small dimensionless parameter.
B6b
23
One-dimensional waves
If p is a function only of one spatial variable x and time t, then (2.12a) reads
2
2 p
2 p
=
c
.
0
t2
x2
(2.13)
(2.14)
where F and G are arbitrary scalar functions. These represent waves, with initial shapes
given by F (x) and G(x), moving at speed c0 from left to right and from right to left
respectively. These are sound waves, and c0 is the speed of sound in the undisturbed
fluid.
Suppose, for example, that a transducer at x = 0 imposes a periodic pressure fluctuation with frequency , that is
p (0, t) = a cos(t) + b sin(t),
(2.15)
(2.16)
where A = a + ib is the complex amplitude. We can then seek a solution of (2.13) in the
form
p (x, t) = Re f (x)eit ,
(2.17)
for some (complex-valued) function f (x). In fact, because taking the real part commutes
with diffentiation, we can ignore the Re for the moment and then simply take the
real part at the end of the calculation. Note that this approach only works for linear
problems.
Substitution for p from (2.17) into (2.13) and (2.16) leads to the problem
2
d2 f
f = 0,
+
dx2
c20
f (0) = A.
(2.18)
where
+ = A.
(2.19)
Evidently one more piece of information is needed to determine the two constants and
. We impose a radiation condition, namely that the source at x = 0 can only cause
outward -travelling waves. In x > 0, we therefore want waves travelling to the right, with
p a function of x c0 t. This implies that must be zero, and an analogous argument
shows that must be zero in x < 0. We therefore obtain
(
Aeix/c0
x > 0,
f (x) =
(2.20)
ix/c0
Ae
x < 0,
University of Oxford
(2.21)
p
=
c
+
rp .
(2.22)
0
0
2
2
2
t
r
r r
r r
It follows that (rp ) satisfies the usual wave equation in r and t and, hence, that the
general solution is
F (r c0 t) G(r + c0 t)
p =
+
.
(2.23)
r
r
Notice that these radial waves are generally unbounded at the origin, with p 1/r as
r 0.
Let us again try to find the wave-field generated by a point transducer at the origin.
In the light of (2.23), we suppose that the pressure behaves like
rp (r, t) Aeit
as r 0,
(2.24)
where the real part is assumed. Seeking a separable solution of the form
p (r, t) = f (r)eit ,
(2.25)
rf (r) A as r 0.
(2.26)
To get a unique solution, we must again impose a radiation condition, namely that p
should be a function of r c0 t rather than r + c0 t. The appropriate solution of (2.26) is
f (r) =
Aeir/c0
r
(2.27)
(2.28)
B6b
-1
M = 1.0
0.5
0.5
-0.5
25
0.5
-1
-0.5
0.5
-0.5
-0.5
-1
-1
M = 1.5
1
0.5
-1
-0.5
0.5
1.5
-0.5
-1
Figure 2.1: A source moving with Mach number M = 0.5, 1.0, 1.5. In the final case, the
angle of the cone is the Mach angle .
U
.
c0
(2.29)
If M < 1 then source travels slower than the speed of sound, and is referred to as
subsonic. The sound waves propagate ahead of the source, as illustrated in figure 2.1
(when M = 0.5). The waves in front of the source are compressed while those behind
are stretched, so a fixed observer would hear the pitch of the sound decrease as the
source travels past; this is the celebrated Doppler effect.
If M = 1 (so the flow is sonic) then the source travels at exactly the same speed
as the sound that it is emitting. Thus, as illustrated in figure 2.1, it can never escape
from its own noise. Finally, if M > 1, then the source outstrips the sound waves, and is
University of Oxford
called supersonic. The noise is confined to a cone, as shown in figure 2.1 with M = 1.5.
Outside the cone is a zone of silence, and a stationary observer would hear nothing until
the cone reaches him or her. The vertex of the cone makes an angle known as the Mach
angle , and it is a simple exercise in trigonometry to show that
sin =
1
.
M
(2.30)
x + u ,
u = Ue
p = p0 + p ,
into (2.1) and again linearise with respect to the primed variables, resulting in
u
u
= p ,
p = c20 .
+U
+ 0 u = 0,
0
+U
t
x
t
x
(2.31)
(2.32)
Again, it is straightforward to show that the flow is irrotational for all time if it is so
initially, so we can introduce a velocity potential such that u = .
Now, if an arbitrary function of t is absorbed into as before, we find that and
satisfy the coupled equations
0
+ c20 = 0,
+U
+U
+ 0 2 = 0.
(2.33)
t
x
t
x
By eliminating between these, we can obtain a single partial differential equation for
either , or p , namely
2
2
2
2
+
2U
+
U
= c20 2 ,
t2
xt
x2
2
2
2
2
+
2U
+
U
= c20 2 ,
t2
xt
x2
2
2 p
2 p
2 p
+
2U
+
U
= c20 2 p .
t2
xt
x2
(2.34a)
(2.34b)
To understand the implications of these equations, let us consider some special cases.
First, for sound propagating in just one space dimension, say p = p (x, t), (2.34b)
becomes
2
2 p
2 p
2
2 p
+
2U
+
(U
c
)
= 0.
(2.35)
0
t2
xt
x2
It is readily verified that this partial differential equation is hyperbolic, with characteristics given by
dx
= U c0 .
(2.36)
dt
For subsonic flow, U < c0 and the characteristic velocities (2.36) take different signs,
indicating that waves travel in both directions. For supersonic flow, U > c0 so the
B6b
27
characteristic velocities are both positive. This means that waves can only propagate
downstream in this case, and there is a zone of silence upstream of the source.
Next, we try looking for two-dimensional steady solutions, setting p = p (x, y), so
(2.34b) becomes
2 p 2 p
= 0,
(2.37)
(1 M 2 ) 2 +
x
y 2
where M = U/c0 is the Mach number as before. For M < 1, (2.37) is elliptic, and is
qualitatively similar to Laplaces equation. For M > 1, though, (2.37) is hyperbolic, and
is analogous to the wave equation; indeed, the general solution of (2.37) in this case is
readily found to be
x
x
+g y+
,
(2.38)
p =f y
M2 1
M2 1
where f and g are two arbitrary scalar functions
The change of type of the partial differential equation (2.37) as the Mach number
passes through 1 is associated with a dramatic change in the character of the flow.
In subsonic flow, the elliptic equation causes localised effects to be felt everywhere
instantaneously. In supersonic flow, however, disturbances can only propagate along
the characteristics, given by
dy
1
= tan ,
=
dx
M2 1
(2.39)
where = sin1 (1/M ) is the Mach angle as before. This explains the differences depicted
in figure 2.1. In the subsonic flow, waves from the source penetrate the whole space while,
in supersonic flow, they are only felt in the Mach cone downstream of the source.
2.3
(2.40)
(2.41)
University of Oxford
z
z = (x, y, t)
x
z = h
111111111111111111
000000000000000000
000000000000000000
111111111111111111
Figure 2.2: Definition sketch of the geometry for Stokes waves on a fluid layer of depth
h.
Small-amplitude waves then perturb the fluid such that the free surface is displaced
to z = (x, y, t), as illustrated in figure 2.2.
On the base z = h, the normal velocity w must be zero, that is
= 0 at z = h.
z
On the free surface z = we have the kinematic boundary condition
D
(z ) = 0
Dt
which, when written out in terms of , reads
=
+
+
z
t
x x y y
at z = (x, y, t).
(2.42)
(2.43)
(2.44)
We also have the dynamic boundary condition that the pressure must be continuous
across the interface, that is
p = pa
at z = (x, y, t),
(2.45)
(2.46)
B6b
29
The problem, then, is to solve (2.40) for , subject to the boundary conditions (2.42),
(2.44) and (2.46). Notice that one more condition is needed at the free surface than at
the base, since the displacement (x, y, t) is also unknown in advance.
Although Laplaces equation is linear, the boundary conditions (2.44) and (2.46) on
the free surface are nonlinear, and the problem is therefore difficult to solve in general. If
the disturbances are small, then the boundary conditions can be simplified by linearising,
that is neglecting products of terms involving and . For example, if we neglect the
quadratic terms in (2.44), we find
=
z
t
at z = (x, y, t).
(2.47)
(x, y, , t)
(x, y, 0, t) + 2 (x, y, 0, t) + ,
z
z
z
(2.48)
in which all terms except the first are nonlinear. When linearising the boundary conditions, it is thus consistent also to evaluate the left-hand side of (2.47) at z = 0 rather
than z = . The same simplification applies when we linearise (2.46), so we end up with
the boundary conditions
=
,
z
t
+ g = 0
t
at z = 0.
(2.49)
One-dimensional waves
Now we look for solutions in which is of the form
(x, t) = Aei(kxt) ,
(2.50)
where A is a constant and the real part is assumed. (Because the problem is linear, we
can proceed with the complex solution (2.50) and then take the real part at the end.)
The parameter represents the frequency at which the surface oscillates at any fixed
position x. The wavenumber k is 2/, where is the wavelength; thus k is small for
long waves and large for short waves. The phase velocity at which the waves propagate
is related to and k by
cp = .
(2.51)
k
It is consistent with (2.50) to assume that is of the form
(x, z, t) = f (z)ei(kxt) .
(2.52)
(2.53)
University of Oxford
c2p
gh
k
Figure 2.3: The squared wave-speed c2p given by (2.57) versus wavenumber k.
Clearly f is a linear combination of ekz and ekz , and the correct combination to satisfy
the boundary condition (2.42) on the base is
f = B cosh k(z + h) ,
(2.54)
i
k sinh(kh)
A
= 0.
g i cosh(kh)
B
(2.55)
g
tanh(kh),
k
(2.57)
which depends on the wavenumber k; that is, waves with different wavenumbers move
at different speeds. Such waves are called dispersive, in contrast with acoustic waves,
which have a constant wave speed.
As depicted in figure 2.3, for positive k, the right-hand side of (2.57) is a decreasing
function, indicating that long waves travel faster than short waves. The maximum wave
speed occurs in the limit k 0, which yields
cp 6
gh.
(2.58)
B6b
211
Extensions
Here we briefly summarise a few common generalisations of Stokes waves.
Flowing fluid
We can study waves on a flowing liquid by linearising about uniform flow, setting
x + ,
u = Ue
(2.59)
where and its derivatives are again assumed to be small. It is clear that
still satisfies Laplaces equation, and the only effect is to modify the free-surface
conditions to
=
+U ,
z
t
x
+U
+ g = 0
t
x
at z = 0.
(2.60)
Two fluids
Suppose the interface z = separates two fluids with different densities, say = 1
in z < 0 and = 2 in z > 0. We denote the velocity potentials and pressures on
either side by 1 , 2 and p1 , p2 respectively. The kinematic condition (2.44) applies
to both sides of the interface, and leads to the linearised boundary conditions
1
2
=
=
t
z
z
at z = 0.
(2.61)
sin
(p1 p2 )
cos
b
cos
ds +
= 0,
sin a
(2.63)
where s is arc length and is the angle made by the interface with the horizontal.
Assuming is constant, we can rewrite this as
Z b
d cos
sin
ds = 0,
(2.64)
(p1 p2 )
+
cos
ds sin
a
University of Oxford
p2
p1
Figure 2.4: Schematic showing the surface tension acting at a fluid interface.
which holds for all a and b, so the integrand must be zero. Hence there is a pressure
jump across the interface, given by
p1 p2 = ,
where
d
=
=
ds
1+
2 )3/2
(2.65)
2
x2
(2.66)
is the curvature of the interface.2 The latter equality can be obtained from the
relations
dx
= tan ,
= cos .
(2.67)
x
ds
After linearisation, the dynamic boundary condition is thus modified to
2
1
2
1
+ g 2
+ g = 2 at z = 0
(2.68)
t
t
x
to take account of surface tension. Note that (2.62) is recovered if is set to zero.
Example
We illustrate all these effects by analysing the situation shown in figure 2.5, where a
layer of depth h2 and density 2 flows at speed U over a layer of depth h1 and density
1 . The disturbance potentials 1 , 2 and the free-surface deflection satisfy
2 1 = 0
1
=0
z
2
h1 <z < 0,
z = h1 ,
2 2 = 0
2
=0
z
0 <z < h2 ,
(2.69a)
z = h2 ,
(2.69b)
In three dimensions, (2.65) still holds, with equal to the mean curvature of the interface, that is
8
8
9
9
,s
,s
2 2 =
2 2 =
<
<
+
.
+
+
=
1+
1+
; y : y
;
x : x
x
y
x
y
B6b
213
111111111111111
000000000000000
000000000000000
111111111111111
U
h2
2 , 2 , p 2
1 , 1 , p 1
h1
111111111111111
000000000000000
Figure 2.5: Schematic of a fluid layer of thickness h2 and density 2 flowing at speed U
over a layer of thickness h1 and 1 .
1
=
,
=
+U ,
z
t
z
t
x
1
2
2
2
1
+ g 2
+U
+ g = 2
t
t
x
x
We look for travelling waves of the form
z = 0.
(2.69c)
= Aei(kxt) , 1 = Bei(kxt) cosh k(z + h1 ) , 2 = Cei(kxt) cosh k(z h2 ) ,
(2.70)
and find the dispersion relation
2 1 coth(kh1 ) + ( U k)2 2 coth(kh2 ) = (1 2 )g + k 2 k.
(2.71)
For simplicity, we take the limit where both layers are very deep, that is h1 , h2 , so
(2.71) becomes
(1 + 2 ) 2 2(2 U k) + 2 U 2 k 2 (1 2 )g + k 2 |k| = 0.
(2.72)
(2.73)
If 1 > 2 then the right-hand side of (2.73) is positive, but if 1 < 2 , it is negative for
some values of k, namely
s
(2 1 )g
|k| <
.
(2.74)
University of Oxford
U2
2(1 + 2 ) p
(1 2 )g.
1 2
(2.76)
If U > U , then there is a band of values of k for which (2.75) is satisfied and for which
is therefore complex. In other words the flow is unstable if the velocity of the upper
fluid exceeds this critical value. This KelvinHelmholtz instability is responsible for the
formation of waves by wind blowing over the sea.
2.4
Now we consider waves in a fluid that is incompressible, so the density of each fluid
element is conserved, i.e.
D
= 0 u = 0.
(2.77)
Dt
However, we no longer assume that the density is constant. The aim is to study waves
in a stratified fluid, where the density varies with depth, for example in the sea, where
depends on the salinity of the water. We therefore start with a stationary solution of
the form
u = 0,
= 0 (z),
p = p0 (z).
(2.78)
B6b
215
Stokes waves may be viewed as a special case of (2.78) in which 0 (z) is piecewise
constant. From the momentum equation (2.1c), with g = g
ez , we find that the
pressure must take the form
Z z
0 () d,
(2.79)
p0 (z) = pa g
0
= 0 (z) + ,
p = p0 (z) + p ,
(2.80)
where the primed variables are again assumed to be small. With all nonlinear terms
neglected, the equations reduce to
d0
+ w
= 0,
t
dz
0
p
u
=
,
t
x
v
p
=
,
t
y
u v w
+
+
= 0,
x
y
z
0
w
p
=
g,
t
z
(2.81a)
(2.81b)
+
=
.
(2.82)
0
0
x
y 2
t x
y
zt
Taking the time derivative of the vertical momentum equation and using the first
conservation-of-mass equation we find that
0
2 p
2 p
d0
2 w
=
g
=
+ gw
.
2
t
zt
t
zt
dz
=
.
t2 x2
y 2
z 2
0 dz
x2
y 2
g zt2
(2.83)
(2.84)
To get an idea of the behaviour of the solutions to this equation, we briefly examine a simplified version based on the following assumptions. We consider purely twodimensional disturbances so that w is a function only of x, z and t. We also suppose
that
1 d0
(2.85)
=
0 dz
is constant, which occurs if 0 is proportional to ez . Finally, we assume that gravity
dominates and we may thus neglect the final term multiplied by 1/g in (2.84), to obtain
2 2 w 2 w
2 w
+
.
(2.86)
=
g
t2 x2
z 2
x2
University of Oxford
(2.87)
where the amplitude A is arbitrary, k is the wavenumber, and the direction of propagation makes an angle with the x-axis. Substitution of (2.87) into (2.86) results in the
dispersion relation
2 = g cos2 .
(2.88a)
Hence waves travel with a positive frequency provided is negative. If is positive (so
the density decreases with depth) then is imaginary: the fluid is unstably stratified.
The most unstable waves have = 0 or , so they travel in the horizontal direction.
Note that we can check a posteriori whether the term we omitted from (2.84) actually
is negligible. With the extra term included, (2.88a) becomes
i sin
2 = g cos2 ,
(2.88b)
1
k
so the approximation is good provided k, that is, if the wavelength is much smaller
than the length-scale over which 0 varies.
2.5
Now we examine the possibility of waves in a constant-density fluid rotating with constant angular velocity . If gravity is neglected then, as shown in section 1, the governing
equations are
u = 0,
1
u
+ (u )u + 2 u + ( x) = p.
t
(2.89)
(2.90)
which may easily be proved by taking components. Hence the momentum equation may
be written in the form
1
u
+ (u )u + 2 u = P,
t
(2.91)
P = p | x|2 .
2
(2.92)
We look for small disturbances by linearising about the uniform steady state u = 0,
P = const and find that the perturbations satisfy
u = 0,
u
1
+ 2 u = P .
t
(2.93)
B6b
217
Equation (2.93) describes large-scale flows in the Earths atmosphere.3 In steady flow,
(2.93) implies that u P = 0, which explains why the wind is usually parallel to
isobars (lines of constant pressure).
If we now take the curl of (2.93b) and make use of (2.93a) when expanding the vector
triple product, we obtain
( u ) = 2 ( )u .
(2.94)
t
Applying another time derivative and a curl to (2.94), we obtain
2
(2.95)
( u ) = 2 ( ) ( u ) .
2
t
t
On the left-hand side, we can now expanding the double curl using the vector identity
( f ) ( f ) 2 f , and apply (2.93a) once more. On the right-hand side
we can substitute using (2.94). We obtain
2
2 2
+
4
u
u = 0 .
(2.96)
t2
Substituting in a plane wave proportional to ei(kxt) we obtain the dispersion relation
2 =
4 ( k)2
|k|2
(2.97)
Since is always real, these waves are always stable. Observe that the frequency
depends only on the direction of k, and not on its magnitude. The highest frequencies
are obtained when the wavevector k is aligned with the rotation axis , and indeed the
fastest wavespeeds are obtained in this direction also.
2.6
Electromagnetic waves
In free space (where there are no charges or currents) the electric field E and magnetic
field B satisfy Maxwells equations
B
E
B = 0,
B =
,
(2.98)
t
t
where and are positive constants known as the permittivity and permeability respectively. It is straightforward by cross-differentiation to find that E and B both satisfy
the wave equation
E = 0,
E =
2E
= c2 2 E,
t2
2B
= c2 2 B,
t2
(2.99)
University of Oxford
(2.100a)
p
with c = T /m. Thus the string transmits non-dispersive waves at constant speed c.
This theory is easily generalised to describe an elastic membrane stretched to a
tension T across the plane z = 0. Now small transverse displacements satisfy the twodimensional wave equation
2
2w
2w
2 w
=c
+
,
(2.100b)
t2
x2
y 2
p
where c = T / and is now the mass of the membrane per unit area. Again the
wave-speed c is constant.
2w
+ B4 w = 0,
t2
(2.103)
B6b
219
in which is the density, while and are called the Lame constants. (Roughly
speaking, measures a materials resistance to compression and its resistance to
shear.) It is straightforward to show that (2.104) supports two different kinds of waves
with different wave-speeds. First taking the divergence of (2.104), we find that ( u)
satisfies the wave equation
2 ( u)
= c2p 2 ( u),
t2
(2.105)
where c2p = ( + 2)/. Equation (2.105) describes pressure waves or P-waves moving at
speed cp . On the other hand, taking the curl of (2.104) results in an equation describing
shear waves or S-waves, namely
2 ( u)
= c2s 2 ( u),
t2
(2.106)
where c2s = /.
The existence of two different wave speeds explains why, following an underground
earthquake, two initial shocks are felt at the surface: the P-waves arrive first, followed
by the S-waves.
Quantum mechanics
A quantum-mechanical particle of mass m, moving in one-dimension under a potential
V (x), is described by the Schr
odinger equation
i~
~2 2
=
+ V (x),
t
2m x2
(2.107)
where ~ is Plancks constant. The so-called wave function is related to the probability
density function P (x, t), measuring the probability of finding the particle at position x
at time t, by
P (x, t) = |(x, t)|2 .
(2.108)
In three dimensions, the second-order x-derivative in (2.107) is replaced by 2 .
With just one t-derivative on the left-hand side, (2.107) does not at first glance
resemble a wave equation at all. However, if we set = u + iv and, for simplicity, ignore
the potential V we find that the real and imaginary parts of satisfy the beam equation
(2.101):
2v
~ 2 4u
~ 2 4v
2u
+
= 2 +
= 0.
(2.109)
t2
2m
x4
t
2m
x4
Hence (2.107) supports dispersive waves and, if V is constant, the dispersion relation is
=
~k 2 V
+ .
2m
~
(2.110)