Environmental Fluid Mechanics (2005) 5: 507–525 © Springer 2006

DOI 10.1007/s10652-005-2234-0

Kelvin–Helmholtz Instability
as a Boundary-value Problem

BENOIT CUSHMAN-ROISIN
Thayer School of Engineering, Dartmouth College, Hanover, NH 03755, USA
E-mail: [email protected]

Received 4 October 2004; accepted in revised form 16 August 2005

Abstract. The Kelvin–Helmholtz (KH) instability is traditionally viewed as an initial-value
problem, wherein wave perturbations of a two-layer shear flow grow over time into bil-
lows and eventually generate vertical mixing. Yet, the instability can also be viewed as a
boundary-value problem. In such a framework, there exists an upstream condition where
a lighter fluid flows over a denser fluid, wave perturbations grow downstream to eventu-
ally overturn some distance away from the point of origin. As the reverse of the tradi-
tional problem, this flow is periodic in time and exhibits instability in space. A natural
application is the mixing of a warmer river emptying into a colder lake or reservoir, or
the salt-wedge estuary. This study of the KH instability from the perspective of a bound-
ary-value problem is divided into two parts. Firstly, the instability theory is conducted
with a real frequency and complex horizontal wavenumber, and the main result is that
the critical wavelength at the instability threshold is longer in the boundary-value than in
the initial-value situation. Secondly, mass, momentum and energy budgets are performed
between the upstream, unmixed state on one side, and the downstream, mixed state on
the other, to determine under which condition mixing is energetically possible. Cases with
a rigid lid and free surface are treated separately. And, although the algebra is somewhat
complicated, both end results are identical to the criterion for complete mixing in the
initial-value problem.

Key words: Estuaries, Kelvin–Helmholtz instability, lakes, shear flow

1. Introduction
It is not uncommon for environmental fluids to be subject simultaneously
to the destabilizing effect of a velocity shear and the stabilizing effect
of density stratification, and, when such competition occurs, the outcome
is often the so-called Kelvin–Helmholtz (KH) instability [4]. Ever since
von Helmholtz [5] and Kelvin [7] developed the theory, this instability has
become a standard staple of fluid mechanics, and the basic theory can be
found in numerous textbooks, for example Lamb ([9], pp. 373–374), Turner
([19], pp. 93–96), Kundu ([8], pp. 373–381) and Scorer ([14], pp. 231–234)
to cite a few up to the present time. Investigations into the details of
508 BENOIT CUSHMAN-ROISIN

the instability, including secondary instabilities [13] and the mechanisms of

entrainment [2], have been carried out perhaps with far more depth than
for any other type of fluid instability.
The prototypical case is that with one layer of lighter fluid overlying
another of denser fluid, and the two moving horizontally in the same direc-
tion but with different velocities. The theory then states that a sinusoidal
perturbation of wavenumber k on the interface of velocity and density dis-
continuity is unstable if it meets the following inequality:

(ρ22 − ρ12 )g < ρ1 ρ2 k(U1 − U2 )2 , (1)

where ρ1 and ρ2 are the upper and lower density, respectively, g the gravi-
tational acceleration, and U1 and U2 the upper and lower velocity, respec-
tively. When the two densities are close to each other (e.g., warmer air
mass over colder air, or freshwater over salty seawater), then the Bous-
sinesq approximation yields

2(ρ2 − ρ1 )g < ρ0 k(U1 − U2 )2 , (2)

where ρ0 is the reference density, close to both ρ1 and ρ2 .

Outstanding laboratory simulations (Figure 1) of the KH instability
were reported by Thorpe [18], and the phenomenon has also been captured
in the midst of the action in both the atmosphere ([14], p. 245) and the sea
[21].
An alternative to studying the instability as it begins (small-amplitude
theory) and matures (laboratory and computer simulations) is to contrast
the initial, two-layer state with the final mixed state and ask whether the
transition is energetically possible. The reasoning is that, if the energy
is less in the final state than in the initial state, then the transition
can be accomplished without import of energy, and the mixing can be
spontaneous. Such bulk analysis was performed by Cushman–Roisin ([3],
pp. 152–153) and yields a criterion for the vertical extent of the eventual
mixing: If the following inequality holds true
(ρ2 − ρ1 )gH
< 1, (3)
ρ0 (U1 − U2 )2
then mixing can spontaneously occur over the entire height H of the
system.
The aforementioned studies all consider the instability process as an
initial-value problem: There exists a certain initial state, which must have
somehow arisen rather quickly and then undergoes a gradual instability
until an ultimate stable state is reached. While such situation can and does
intermittently occur in the atmosphere ([14], p. 245) and the ocean [21],
there are other situations in which the two-layer, sheared configuration is
K–H INSTABILITY 509

Figure 1. Laboratory demonstration of a Kelvin–Helmholtz instability framed as an

initial-value problem. At some initial time, a long tank filled with two resting fluids
of different densities is tipped slightly. As the denser fluid sinks toward the bottom
and the lighter fluid rises toward the top, a counterflow is generated, and the inter-
face develops unstable billows [18].
510 BENOIT CUSHMAN-ROISIN

Figure 2. Laboratory demonstration of a Kelvin–Helmholtz instability framed as a

boundary-value problem. Two layers flowing from left to right join downstream of
a thin plate (visible on the left of the top photograph). The upper and faster mov-
ing fluid is slightly less dense than the fluid below. With downstream distance (from
left to right on each photograph and from top to bottom panel), waves first turn
into billows and later degenerate into turbulence. (Photo courtesy of Gregory A.
Lawrence. For more details on the laboratory experiment, see Lawrence et al. [10]).

constantly being regenerated on the upstream side of the system, and the
instability takes the form of a boundary-value problem. Common exam-
ples are the summer discharge of warmer river water into a colder lake
and all salt-wedge estuaries, in which freshwater from a river flows encoun-
ters salty seawater. In each situation, a lighter fluid is constantly flowing
at a differential speed over a denser fluid. The Kelvin–Helmholtz instability
then develops not in time but in space, as function of downstream distance
from the point of encounter between the two fluids. A laboratory simu-
lation of this process ([10] – see Figure 2) shows that the instability still
proceeds by means of growing waves and overturning billows, except that
waves and billows co-exist at various stages of development.
K–H INSTABILITY 511

Maslowe and Kelly [11] have investigated the small-amplitude stability

of a continuously stratified and sheared flow. Huerre and Monkewitz [6]
have further shown that the condition for absolute stability depends sig-
nificantly on whether the problem is posed as an intial-value or boundary-
value problem. But, to the author’s knowledge this question has not yet
been addressed for the two-layer Kelvin–Helmholtz problem. Its answer is
the subject of the present paper.

2. Instability Theory
In this section, we consider the downstream stability of a two-layer sheared
fow. Starting at position x = 0, two fluids previously separated by a parti-
tion begin to flow together, with the lighter fluid of density ρ1 floating atop
of the denser fluid of density ρ2 . At the origin, the respective horizontal
velocities are U1 and U2 , both positives to guarantee flow toward x > 0, the
downstream direction. The vertical coordinate z increases upward from z =
0 at the level of the upstream partition. Downstream of the partition (x >
0), waves may occur on the interface, creating perturbations in the flow of
each layer, including vertical velocities.
Neglecting variations in the transverse horizontal direction, we take the
flow to be two-dimensional in the vertical plane and denote the velocity
components in layer j (j = 1, 2 for the upper and lower layers, respec-
tively) as uj (x, z, t) in the horizontal and wj (x, z, t) in the vertical. The
incompressible, inviscid, non-hydrostatic equations governing the flow in
each layer are:

∂uj ∂wj
+ =0 (4)
∂x
∂z 
∂uj ∂uj ∂uj ∂pj
ρj + uj + wj =− (5)
∂t ∂x ∂z ∂x


∂wj ∂wj ∂wj ∂pj
ρj + uj + wj =− − ρj g, (6)
∂t ∂x ∂z ∂z

which consist in a statement of volume conservation and a pair of momen-

tum equations, and where pj (x, z, t) is the pressure in layer j . The unper-
turbed state corresponds to the solution uj = Uj , wj = 0 and hydrostatic
pressure pj = P − ρj gz, where P is the pressure along the interface (z = 0).
For the boundary-value problem, both Uj values are required to be pos-
itive so that there exists a downstream direction along which the insta-
bility and mixing can proceed without affecting the upstream boundary
condition.
512 BENOIT CUSHMAN-ROISIN

With waves on the interface, the flow consists of the basic state plus a
perturbation:
uj = Uj + u
j
wj = wj

pj = P − ρj gz + pj
,
while the density ρj of each layer remains unchanged. Considering only
perturbations of weak amplitude, so that we may linearize the equations,
and dropping the primes, we obtain:
∂uj ∂wj
+ =0 (7)
∂x
∂z 
∂uj ∂uj ∂pj
ρj + Uj =− (8)
∂t ∂x ∂x


∂wj ∂wj ∂pj
ρj + Uj =− . (9)
∂t ∂x ∂z
We seek a solution of the Fourier type, with all dependent variables pro-
portional to
eiωt−ikx−aj z
where the frequency ω and wavenumber k are shared between the two lay-
ers, because matching of the appropriate expressions along the interface
will have to be satisfied for all t and x values, but the attenuation coeffi-
cient aj in the vertical need not be the same in both layers.
For the boundary-value problem under consideration here, we enforce
periodicity in time, demanding that ω take only real positive values, but
allow waves to grow or decay in the downstream direction, thus letting k
be possibly complex. With the minus sign in front of k, a positive imagi-
nary part, Im(k), would indicate growth. For the perturbation to be con-
fined vertically in the vicinity of the interface, we require Re(a1 ) > 0 and
Re(a2 ) < 0.
Substitution in the governing Equations (7)–(9) yields:
−ikuj − aj wj = 0 (10)
iρj (ω − kUj )uj = +ikpj (11)
iρj (ω − kUj )wj = +aj pj , (12)
from which we can find uj and pj in terms of wj :
aj
uj = i wj (13)
k
ρj
pj = i (ω − kUj )wj . (14)
aj
K–H INSTABILITY 513

Elimination of uj and pj leads to a single homogeneous equation for wj ,

which allows a non-trivial solution only if

ρj (ω − kUj )(aj2 − k 2 ) = 0. (15)

The solution ω = kUj is unacceptable because it would yield two differ-

ent values of k, one for each layer. Therefore, we have aj2 = k 2 , implying:

aj = ±k, (16)

with the sign to be selected such that Re(a1 ) > 0 and Re(a2 ) < 0. With
k = kR + ikI , the phase speed is c = ω/kR , and it is expected to be posi-
tive because wave propagation ought to be in the direction of the overall
flow, that is, in the positive x-direction. Thus, we expect kR > 0 and choose
a1 = +k and a2 = −k.
Taking the vertical displacement η of the interface as

η(x, t) = Aei(ωt−kx) , (17)

we impose at z = η a set of kinematic and dynamic conditions. The kine-

matic condition requires that particles of each layer at the interface move
with the interface, namely
∂η ∂η
wj = + Uj
∂t ∂x
or

wj = i(ω − kUj )A. (18)

The dynamic condition is continuity of pressure at the interface:

P − ρ1 gη + p1 = P − ρ2 gη + p2 .

or

p2 = p1 + (ρ2 − ρ1 )gA. (19)

Linearization then allows that the preceding expression be stated at z = 0

rather than z = η. Combining (14), (18) and (19) yields a homogeneous
equation for the amplitude A, which is not trivially zero only if

ρ1 (ω − kU1 )2 + ρ2 (ω − kU2 )2 = gk(ρ2 − ρ1 ). (20)

This equation is the same as for the classical Kelvin–Helmholtz stability

analysis (see for example Kundu, [8], pp. 377). The dispersion curve ω(k)
514 BENOIT CUSHMAN-ROISIN

Figure 3. Dispersion relation of the Kelvin Helmholtz wave problem (plotted in the
case of shear equal to 40% of the average velocity and small relative density varia-
tion). Temporal instability occurs when there is no real ω solution for a given real
k, and spatial instability occurs when there is no real k solution for a given real
ω. Note how the cases differ.

to which it corresponds is shown in Figure 3. Here, instead of solving the

relation for ω as done for the initial-value problem, we solve it for k:
√
2ω(ρ1 U1 + ρ2 U2 ) + gρ ± R
k= , (21)
2(ρ1 U12 + ρ2 U22 )

where ρ = ρ2 − ρ1 is the positive density difference and the expression R

under the square root is equal to:

R = g 2 ρ 2 + 4ωgρ(ρ1 U1 + ρ2 U2 ) − 4ρ1 ρ2 ω2 U 2 , (22)

with U = |U1 − U2 |, the absolute velocity difference.

The condition of instability is that the wavenumber k be complex, which
occurs when R is negative. This condition is realized only if the angular
frequency ω exceeds a threshold value ωc given by:
K–H INSTABILITY 515
  
gρ
ωc = (ρ U
1 1 + ρ U
2 2 ) + (ρ 1 + ρ2 )(ρ U
1 1
2
+ ρ U
2 2
2
) . (23)
2ρ1 ρ2 U 2
The corresponding wavenumber, obtained by substituting (23) in (21), is:
  
gρ ρ1 + ρ2
kc = (ρ1 + ρ2 ) + (ρ1 U1 + ρ2 U2 ) , (24)
2ρ1 ρ2 U 2 ρ1 U12 + ρ2 U22

which differs from the critical wavenumber of the initial-value problem [k

value extracted from (1) when the inequality is replaced by an equality].
Thus, the wavelength exhibited by the boundary-value KH is not the same
as that in initial-value KH. It can actually be shown that the critical wave-
length is shorter in the boundary-value than in the initial-value configura-
tion (Figure 3).
If the two layers are of the same fluid but with slightly different densities
(such as warmer water overflowing colder water), the Boussinesq approxi-
mation may be invoked, allowing us to replace all densities by the reference
density ρ0 except in the density difference. The preceding expressions then
reduce to
  
gρ
ωc = (U1 + U2 ) + 2(U1 + U2 )
2 2
(25)
2ρ0 U 2

gρ U1 + U2 2
kc = 1+ . (26)
ρ0 U 2 2 U12 + U22
If, furthermore, the two velocities U1 and U2 are close to each other (U1 
U2  U ), the expressions are:
2gU ρ
ωc = (27)
ρ0 U 2

2gρ
kc = , (28)
ρ0 U 2
whereas, if the lower layer is at rest (U2 = 0),
√
1 + 2 gρ gρ
ωc = = 1.207 (29)
2 ρ0 U1 ρ0 U1


1 gρ gρ
kc = 1 + √ 2
= 1.707 . (30)
2 ρ0 U1 ρ0 U12
Since random perturbations can come in all frequencies, including values
higher than ωc (which is finite as long as U is nonzero), we can conclude
516 BENOIT CUSHMAN-ROISIN

that there always exist waves that grow downstream, and the sheared flow is
always unstable.1

3. Energetics of mixing
The energetics of Kelvin–Helmholtz billows have been the subject of multi-
ple investigations. Of note are the studies by Palmer et al. [13] of the ener-
getics of the secondary instability and by Smyth et al. [15] of the work
done against buoyancy forces and its effect on the mixing efficiency. Both
of these studies, however, consider the temporal instability. Closer to the
spatial instability is the set of laboratory experiments conducted by Strang
and Fernando [16, 17], in which the sheared stratified flow has a definite
spatial origin. The data reveal relations between the various components of
the turbulent energy budget (production, work against buoyancy forces and
dissipation) and the gradient Richardson number.
Here, instead of following the developing instability and consider its inner
workings, we turn our attention to the final state, assuming that it is one of
complete mixing. For simplification we assume rigid boundaries both at top
and bottom, as depicted in Figure 4. This asssumption will be relaxed in a
subsequent section.
The question we ask is: Under the constraints of the appropriate con-
servation laws, is the downstream state of lower or higher energy than the
upstream state? If it corresponds to a level of lower energy, then mixing
liberates energy (available for turbulent dissipiation) and would likely occur
spontaneously. But if the downstream state holds higher energy than the
upstream state, then there is an energy barrier to overcome, and in the
absence of an external source of energy, mixing cannot take place, at least
not over the entire extent of the domain.

3.1. CONSERVATION LAWS

The first conservation law is that of mass, more precisely of mass flux,
requiring that the mass flux leaving the system downstream be equal to that
entering upstream. Mathematically, the vertically integrated mass flux
1
It can be shown that there is no finite frequency ω yielding a most rapidly grow-
ing mode. This is similar to the classical initial-value problem, for which there is no
wavelength of fastest growth. In other words, the Kelvin–Helmholtz model predicts an
ultra-violet catastrophe. The finite wavelengths noted in the laboratory experiments (Fig-
ures 1 and 2) arise from the presence of diffusion and viscosity, which smooth the ini-
tial/upstream density and velocity discontinuities somewhat, thus providing a length scale
to each system. Scorer ([14], p. 236) reports that the observed wavelength is about 4π
times the thickness of the diffuse interfacial layer. Models with continuous profiles do
not exhibit this behavior.
K–H INSTABILITY 517

Figure 4. Schematic representation of the system under consideration and the

attending notation.

H
ρ(z)u(z)dz
0

must be the same downstream as upstream, giving:

ρU H = ρ1 U1 H1 + ρ2 U2 H2 , (31)
where ρ and U are the downstream, mixed density and velocity, while ρi
and Ui are the density and velocity of layer i (i = 1, 2).
For a Boussinesq fluid, the three densities are very close to one
another and, since gravity does not enter this statement, the small density
differences may be neglected, and conservation of mass flux reduces to con-
servation of volumetric flux. Replacing all three densities by the reference
density ρ0 , we can solve the preceding equation for the donwstream veloc-
ity U :
H1 U1 + H2 U2
U= . (32)
H
Next, we enforce conservation of heat, salt or whatever component in
the fluid is causing the density difference between ρ1 and ρ2 . Assuming
a linear relationship between this physical property and density, we write
again the preceding mass flux but need to retain this time the small den-
sity differences. This yields (31) again, which together with (32) provides
the downstream density:
ρ1 H1 U1 + ρ2 H2 U2
ρ= . (33)
H1 U1 + H2 U2
518 BENOIT CUSHMAN-ROISIN

Figure 5. Upstream and downstream pressure forces acting on the fluid.

Finally, we consider the flux of linear momentum, which is not nec-

essarily unchanged since forces, including the pressure force, can act to
increase or decrease linear momentum, according to Newton’s second law.
The momentum outflowing at the downstream end is equal to the momen-
tum inflowing upstream plus the forward pressure force upstream minus the
retarding pressure force downstream, as shown in Figure 5. Mathematically
([20], p. 154), we have:

H H1 +H2
ρu2 |downstream dz = ρu2 |upstream dz
0 0
H1 +H2 H
+ p|upstream dz − p|downstream dz, (34)
0 0

where u is the local velocity and p the pressure, which in the absence of
vertical motion is strictly hydrostatic.
Since the pressure needs to be known only within an additive constant,
we can choose p = 0 at the top boundary on the upstream side (see Figure
5). Over depth, the downstream pressure is then:

p|upstream = ρ1 gH1 + ρ2 g(H2 − z) for 0 ≤ z ≤ H2

p|upstream = ρ1 g(H1 + H2 − z) for H2 ≤ z ≤ H1 + H2 (35)

with z the vertical coordinate increasing upward from the bottom bound-
ary. The downstream pressure may be different from zero at the top bound-
ary, and we denote its value by p0 . The downstream pressure profile is
thus:

p|downstream = p0 + ρg(H − z). (36)

K–H INSTABILITY 519

The pressure forces (per unit distance across the flow) can now be cal-
culated:
H1 +H2
p|upstream dz = 21 ρ1 gH12 + ρ1 gH1 H2 + 21 ρ2 gH22 (37)
0
H
p|downstream dz = p0 gH + 21 ρgH 2 , (38)
0

and the linear momentum budget (34) becomes:

ρU 2 H = ρ1 U12 H1 + ρ2 U22 H2
+ 21 ρ1 gH12 + ρ1 gH1 H2 + 21 ρ2 gH22
−p0 H − 21 ρgH 2 . (39)

As it is customary within the Boussinesq approximation, the actual den-

sity may be approximated by the reference density ρ0 in the inertial terms
(those with a velocity squared and no g factor):

ρ0 (H U 2 − H1 U12 − H2 U22 ) = 21 ρ1 gH12 + ρ1 gH1 H2 + 21 ρ2 gH22

−p0 H − 21 ρgH 2 . (40)

This last equation provides the value of the pressure p0 against the top
boundary. Solving (39) for p0 and using the earlier relations (32) and (33)
for U and ρ yields after some algebra:
H1 H2 ρgH1 H2 H2 U1 − (H1 + 2H2 )U2
p0 = ρ0 U 2 + . (41)
H 2 2H H1 U1 + H2 U2
Equations (32), (33) and (41) for U , ρ and p0 , respectively, specify com-
pletely the downstream flow characteristics in terms of its upstream values.

3.2. ENERGETICS

The question that needs to be asked is whether the mixed downstream state
is energetically achievable from the given upstream conditions. To answer
this, we turn to the energy flux, which is defined as ([20], p. 174):
H
E= ( 21 ρu2 + ρgz + p)udz. (42)
0

We easily recognize here the kinetic and potential energy components but
might be surprised by the last term, which involves the pressure. This term
occurs because pressure times the flowing displacement contributes work,
which affects the energy of the fluid, as in the Bernoulli principle.
520 BENOIT CUSHMAN-ROISIN

In the absence of heat and work exchanged between the fluid and
the external environment, the energy flux remains constant. But, if mix-
ing occurs as we assume, turbulent dissipation converts some mechanical
energy into heat, which is then lost to the external environment and/or used
to raise the temperature of the fluid slightly. Mixing therefore requires that
the downstream energy flux be less than the upstream energy flux:

Edownstream < Eupstream . (43)

The upstream energy flux is found to be (in Boussinesq form):

Eupstream = 21 ρ0 (H1 U13 + H2 U23 )

+ρ1 g(H12 U1 + H1 H2 U1 + H1 H2 U2 ) + ρ2 gH22 U2 , (44)

and the downstream value is (again in Boussinesq form):

Edownstream = 21 ρ0 H U 3 + p0 H U + ρgH 2 U. (45)

The difference between the upstream and downstream energy fluxes is:

ρ0 H1 H2 U 2
Eupstream − Edownstream = (H2 U1 + H1 U2 )
2H 2
gρH1 H2
− (H2 U1 + H1 U2 ). (46)
2H
Criterion (43) then tells that mixing is energetically possible as long as:

ρ0 H1 H2 U 2 gρH1 H2
2
(H2 U1 + H1 U2 ) − (H2 U1 + H1 U2 ) > 0,
2H 2H
which can be re-arranged as:

gH ρ
< 1. (47)
ρ0 U 2

This is identical to (3), the criterion of the initial-value problem and remi-
niscent of results by Miles [12] and Abarbanel et al. [1].

4. Generalization to free surface

We now generalize the analysis to a free surface. Since the downstream sur-
face may not lie at the same level as the upstream surface, we introduce the
surface displacement η (Figure 6), which is the amount by which the sur-
face rises (or drops, if η < 0) across the mixing zone.
K–H INSTABILITY 521

Figure 6. Geometry of the free-surface case.

The analysis proceeds as in the previous section. First, conservation of

volume and of density provide:

H1 U1 + H2 U2
U= (48)
H +η
ρ1 H1 U1 + ρ2 H2 U2 ρ1 H1 U1 + ρ2 H2 U2
ρ= = . (49)
(H + η)U H1 U1 + H2 U2

Then, we determine the various pressure profiles [p1 = ρ1 g(H − z), p2 =

ρ1 gH1 + ρ2 g(H2 − z) and p = ρg(H + η − z)] in order to state the momentum
budget, which yields:

ρ0 (H + η)U 2 + 21 ρg(H + η)2 = ρ0 H1 U12 + ρ0 H2 U22

+ 21 ρ1 gH12 + ρ1 gH1 H2 + 21 ρ2 gH22 , (50)

which, after use of (48) and (49) becomes:

g(ρ1 H1 U1 + ρ2 H2 U2 ) η
(H η + 21 η2 ) − (H1 U12 + H2 U22 )
ρ0 (H1 U1 + H2 U2 ) H +η
H1 H2 gρH H H
1 2 2 1 U − H U
1 2 − 2H U
2 2
= U 2 + . (51)
H +η 2ρ0 H1 U1 + H2 U2

The analysis that follows becomes quite intractable unless some simplifying
assumptions are made. One asymptotic situation that should fit most envi-
ronmental applications, including rivers discharging in lakes and estuaries,
consists in taking U12 and U22 each much smaller than gH (small external
522 BENOIT CUSHMAN-ROISIN

Froude numbers) and ρ much smaller than ρ0 (Boussinesq approxima-

tion). Then, the preceding equation can and solved for η:
H1 H2 gH ρ(H2 U1 − H1 U2 − 2H2 U2 ) + 2ρ0 (H1 U1 + H2 U2 )U 2
η= . (52)
2H 2 g(ρ1 H1 U1 + ρ2 H2 U2 )
From this solution, it can be shown that, for commensurate layer depths
(H1 and H2 each on the same order as H ), the magnitude of η is given by


η ρ U 2
=O , . (53)
H ρ0 gH
Thus, the conditions under which η is small are ρ  ρ0 (Boussinesq
aproximation) and U 2  gH (low Froude number), simultaneously. The
second condition is the more constraining of the two.
The upstream and downstream energy fluxes are:
H2
Eupstream = 1
ρ U 2 + ρ2 gz + p2
2 0 2
U2 dz
0
H
+ 1
ρ U 2 + ρ1 gz + p1
2 0 1
U1 dz
H2
H2
= 1
ρ U 2 + ρ1 gH1 + ρ2 gH2
2 0 2
U2 dz
0
H
+ 1
ρ U 2 + ρ1 gH
2 0 1
U1 dz
H2
= 21 ρ0 (H1 U13 + H2 U23 ) + ρ1 gH1 [(H1 + H2 )U1 + H2 U2 ]
+ρ2 gH22 U2 (54)
H +η
Edownstream = 1
ρ U 2 + ρgz + p
2 0
U dz
0
H +η
= 1
ρ U 2 + ρg(H
2 0
+ η) U dz
0
= 21 ρ0 U 3 (H + η) + ρgU (H + η)2
ρ0 (H1 U1 + H2 U2 )3
= + g(ρ1 H1 U1 + ρ2 H2 U2 )(H + η), (55)
2(H + η)2
and the energy difference is, after linearization with respect to the small
terms:

E = Eupstream − Edownstream
ρ0 H1 H2 U 2
= [(2H1 + H2 )U1 + (H1 + 2H2 )U2 ]
2H 2
−gρH1 H2 U2 − g(ρ1 H1 U1 + ρ2 H2 U2 )η. (56)
K–H INSTABILITY 523

Note how in this expression the kinetic-energy term (that preceded by ρ0

and with the cube of velocities) contributes positively to the energy differ-
ence, while the potential-energy terms (those with g) contribute negatively.
This confirms the expected physics, namely that, in the mixing process,
energy is supplied by the velocity shear and consumed by gravity.
Elimination of the downstream variable η by use of (52) provides the
expression of the energy difference in terms of upstream variables alone:


ρ0 gH1 H2 U 2 ρ
E = (H2 U1 + H1 U2 ) − . (57)
2H gH ρ0
Spontaneous mixing requires a drop in energy, and the criterion is thus
E > 0, which can be obtained from (57). The result is:
gH ρ
< 1, (58)
ρ0 U 2
which happens to be identical to criterion (3), once again.

5. Conclusions
The theory of the classical Kelvin–Helmholtz instability was reconsidered
in the framework of a boundary-value problem. Namely, the following
problem was considered: A lighter fluid enters a domain and flows over
a denser fluid, waves develop along the interface, these waves grow in the
downstream direction, develop into billows and eventually roll over; the
downstream end state is characterized by mixing over a finite depth. Physi-
cal examples are the discharge of a warm river in a cold lake and the salt-
wedge estuary.
The small-amplitude theory of such system provides a critical angu-
lar frequency separating unstable from stable perturbations. Since temporal
perturbations may come in all frequencies, there will always be some that
generate unstable waves, and the system is deemed to be always unstable,
as its initial-value counterpart. However, the wavelength and period of the
critical wave differ from the classical problem.
An energy analysis comparing the kinetic and potential energies of the
unmixed upstream two-layer sheared flow with those of the mixed con-
dition occurring downstream was also performed. The treatment of the
boundary-value case differs from that of the classical case in significant
ways. Among other differences, the energy flux includes a higher power of
the velocities, and pressure forces and pressure work need to be retained
in the momentum and energy budgets, respectively. Yet, the outcome is the
same stability criterion as for the initial-value configuration. In retrospect,
this might have been expected, for the energy analysis advanced by Miles
[12], which considers a two-particle exchange, leads to a unique criterion
524 BENOIT CUSHMAN-ROISIN

regardless of whether mixing proceeds in time or over downstream distance.

And, the finite-amplitude stability analysis of Abarbanel et al. [1] based on
the Liapunov method ends, too, with the same criterion, namely that the
Richardson number be greater than one for stability.
The most difficult sub-case among those explored in the present study
was that of a system with a free surface. A major simplification was
achieved by restricting the attention to incoming flows that have very small
Froude numbers in each layer. The nature of the problem with free surface
and upstream Froude numbers that are not small remains to be explored.

Acknowledgements
The problem studied in this paper was suggested by Professor Doron
Nof of Florida State University, whose many suggestions and comments
further contributed to the paper. Marcelo Dottori, Alexandre Fernandes,
J. Lopez-Dominguez and Giuseppe Colantuono of Florida State University
assisted with some of the calculations and sustained the author’s interest in
the problem. The author is also grateful for the two reviewers whose com-
ments helped bring this work in better context. Financial support was pro-
vided by the U.S. Office of Naval Research under grant N00014-02-1-0065
to Dartmouth College.

References
1. Abarbanel, H.D.I., Holm, D.D., Marsden, J.E. and Ratiu, T.S.: 1986, Nonlinear stabil-
ity analysis of stratified fluid equilibria, Phil. Trans. R. Soc. London A 318, 349–409.
2. Cortesi, A.B., Smith, B.L., Yadigaroglu, G. and Banerjee, S.: 1999, Numerical investi-
gation of the entrainment and mixing processes in neutral and stably-stratified mixing
layers, Phys. Fluids 11, 162–185.
3. Cushman-Roisin, B.: 1994, Introduction to Geophysical Fluid Dynamics, 320 pp.,
Prentice-Hall.
4. DeSilva, I., Fernando, H.J.S., Eaton, F. and Hebert, D.: 1996, Evolution of Kelvin-
Helmholtz billows in nature and laboratory, Earth Planet. Sci. Lett. 143, 217–231.
5. von Helmholtz, H.L.F.: 1868, Über discontinuirliche Flüssigkeitsbewegungen, Mona-
tsbericht Akad. Wiss. Berlin S 215–228.
6. Huerre, P. and Monkewitz, P.A.: 1985, Absolute and convective instabilities in free
shear layers, J. Fluid Mech. 159, 151–168.
7. Kelvin, Lord (William Thomson) 1871, Hydrokinetic solutions and observations, Phil
Mag. 10, 155–168.
8. Kundu, P.K.: 1990, Fluid Mechanics, 638 pp., Academic Press.
9. Lamb, H.: 1945, Hydrodynamics, 6th ed., Dover Publications, 738 pp., New York.
10. Lawrence, G.A., Browand, F.K. and Redekopp, L.G.: 1991, The stability of a sheared
density interface, Phys. Fluids A, Fluid Dyn. 3, 2360–2370.
11. Maslowe, S.A. and Kelly, R.E.: 1971, Inviscid instability of an unbounded heteroge-
neous shear layer, J. Fluid Mech. 48, 405–415.
K–H INSTABILITY 525

12. Miles, J.: 1986, Richardson’s criterion for the stability of stratified shear flow, Phys.
Fluids 29, 3470–3471.
13. Palmer, T.L., Fritts, D.C. and Andreassen, Ø.: 1996, Evolution and breakdown of
Kelvin-Helmholtz billows in stratified compressible flows, Part II: Instability structure,
evolution, and energetics, J. Atmos. Sci. 53, 3192–3212.
14. Scorer, R.S.: 1997, Dynamics of Meteorology and Climate, 686 pp., John Wiley &
Sons.
15. Smyth, W.D., Moum, J.N. and Caldwell, D.R.: 2001, The efficiency of mixing in tur-
bulent patches: Inferences from direct simulations and microstructure observations,
J. Phys. Oceanogr. 31, 1969–1992.
16. Strang, E.J. and Fernando, H.J.S.: 2001a, Entrainment and mixing in stratified shear
flows, J. Fluid Mech. 428, 349–386.
17. Strang, E.J. and Fernando, H.J.S.: 2001b, Vertical mixing and transports through a
stratified shear layer, J. Phys. Oceanogr. 31, 2026–2048.
18. Thorpe, S.A.: 1971, Experiments on the instability of stratified shear flows: miscible
fluids, J. Fluid Mech. 46, 299–319.
19. Turner, J.S.: 1973, Buoyancy Effects in Fluids, 368 pp., Cambridge University Press.
20. White, F.M.: 2003, Fluid Mechanics 5th ed., 866 pp., McGraw-Hill.
21. Woods, J.D.: 1968, Wave-induced shear instability in the summer thermocline, J. Fluid
Mech. 32, 791–800 + 5 plates.