104

19. FLUID INSTABILITIES In this case, convection will develop.

To find the conditions under which this instability
occurs, we argued that the rising blob is adiabatic, and
Fluids, magnetized or not, are subject to a wide va- also stays in pressure balance with the outside atmo-
riety of instabilities. Many so-called equlibrium con- sphere. We found, somewhat heuristically, that the sta-
figurations, that is steady-state solutions of the basic bility of the blob depends on how rapidly the density of
equations (or configurations that we might set up in the atmosphere drops compared to the adiabatic case.
the lab), are violently unstable. That is, once set up, If we didn’t have this physical argument in our head,
they rapidly evolve away from their initial state. Some- we could have done it more formally, by assuming the
times these instabilities are destructive (such as current- blob oscillates as eiN t , and exploring what N is. Check
driven instabilities of plasma pinches), and sometimes the result (6.15): the stability criterion is in this equa-
they move the system to another quasi-steady state tion. If N 2 > 0, the blob is stable; while if N 2 < 0, the
(such as convection in the atmosphere or on the sun). displacement grows exponentially and the system is un-
In this chapter we consider a few of these, both hy- stable. But the formal result, (6.15), shows that stability
drodynamic and magnetohydrodynamic. In each case, depends on the sign of dT /dz − (dT /dz)ad .
we are interested in a criterion for, and a description of, Thus: stability or instability is determined by the
instability. When – for what initial configurations – is temperature structure of the atmosphere – which is de-
a system unstable? How fast, and on what scales, does termined by external conditions (how we set the prob-
the instability grow? The answers to these questions lem up). In this case we could make a physical argu-
require some lengthy math. We will see two different ment to lead us to the stability criterion; and a formal
approaches. One is modal analysis, in which a small analysis found the same result. In more complex cases,
perturbation is studies in terms of its Fourier compo- the exact form of the physical argument is not always
nents. In this type of analysis, the system is determined evident ... but the mathematical approach, analogous to
to be unstable if the frequency of the perturbation is the eiN t method here, will lead us to the answer.
found to have imaginary components. Another method
is the energy method, in which one determines whether In the rest of this chapter I’ll treat two well-known
the potential energy of an equilibrium system is a max- fluid instabilities analytically, with and without mag-
imum (unstable), or a minimum (stable). netic fields. Hold on to your hats ...
Often the physical insight on why the instability B. The Rayleigh Taylor Instability
goes, arrives only after the analysis has been gone
through. In these notes, I will try to focus on the physics 1. THE PHYSICS
driving each instability, and present only a summary of
the mathematical analysis, rather than reproducing ev- This instability arises when a heavy fluid is supported
on top of a light one, in a gravitational field. This is
ery step (with credit to each author who presents it in
more detail). clearly unfavorable energetically; if the two fluids can
change places, the system will have a lower potential
A. Buoyancy and Thermal Convection energy. This is just what the RT instability does; any
tiny perturbation of the interface between the two flu-
To motivate things, recall one example we’ve already ids, allows “fingers” to develop – in which each fluid
seen. In chapter 6 we discussed convective stability, penetrates into the other.
and in chapter 14 we talked about the effects of a 1111111111111
0000000000000
0000000000000
1111111111111 111111111111
000000000000
000000000000
111111111111
magnetic field. Let’s revisit this briefly, aiming at the ρ 0000000000000
1111111111111 g 000000000000
111111111111
physics, but with an eye to the more formal stability 2
0000000000000
1111111111111
0000000000000
1111111111111 000000000000
111111111111
000000000000
111111111111
analysis about to come.
0000000000000
1111111111111
0000000000000
1111111111111
1111111111111
0000000000000
0000000000000
1111111111111
000000000000
111111111111
000000000000
111111111111
000000000000
111111111111
000000000000
111111111111
000000000000
111111111111
ρ 0000000000000
1111111111111 000000000000
111111111111
0000000000000
1111111111111 000000000000
111111111111
1 0000000000000
1111111111111 111111111111
000000000000
In §6.2, we considered a small vertical displacement 0000000000000
1111111111111 000000000000
111111111111
Figure 19.1. Development of the Rayleigh-Taylor
of a parcel of fluid (air, say) in a static atmosphere. We instability. The fluid on top has higher density, and gravity
expect, physically, that this parcel will be stable against points downward. Mixing due to the instability produces
the displacement if it has a higher density than its sur- fingers that eventually invert the density distribution.
roundings, once displaced; it will just bob up and down.
The frequency of that motion is the Brunt-Vaisala fre- We will work this in terms of a fluid interface; but
quency, N (as derived in §6.2). On the other hand, if any density gradient inverted compared to the local
the blob when displaced has a lower density than its gravity gives the same results. In addition, the acceler-
surroundings, it will be unstable, and continue to rise. ation does not need to come from gravity; a decleration
 105

of a light fluid into a heavy one (as in an explosion in a (??) represents a perturbation with grows exponentially
dense atmosphere) is also subject to this instability. with time: the system is unstable. Let k2 = kx2 +ky2 , and
denote z-derivatives by Df = df /dz (In this and the
2. THE MATH next section only; Chandra’s notation). These equations
Our basic approach here is modal analysis. That is: first then become
we linearize, assuming small perturbations, so that we
can drop terms which are second order in these pertur- ikx δp = −iωρu
bations. We have done this before – in deriving sound
waves and MHD waves, for instance. Our goal is to find iky δp = −iωρv
the time evolution of these perturbations: do they grow
Dδp = −iωρw − gδρ (19.3)
with time (which is an instability), or simply oscillate
(which says we have wave solutions, but not instabili-
ikx u + iky v + Dw = 0
ties). In order to do this, we do a modal analysis. We
assume that any perturbation can be treated as a sum of iωδρ = wDρ
Fourier components (which is legal if the defining equa-
tions are linear in the perturbations – right??). We can These can be used to show
then focus on the evolution of one such Fourier compo-
nent – for instance (19.2), below. k2 δp = −iωρDw (19.4)
My analysis here follows Chandrasekhar, Hydrody-
namic and Hydromagnetic Stability. He works through and
a modal analysis, and we will pick up on some of his g
results. Recall our formal derivation of sound waves, Dδp = −iωρw + wDρ (19.5)
iω
in Chapter 4: we carried through a linear analysis, ex-
panding the variables v, p, ρ in (unperturbed) + (small These two now combine – eliminating δp between them
perturbation) terms. Our basic equations are those – as
of mass conservation, incompressibility of the unper-
gk2
turbed state, and momentum conservation. Geome- D (ρDw) = wDρ + ρwk2 (19.6)
try: take g = gẑ; assume the unperturbed state had ω2
v = 0, and only has gradients on the ẑ direction. and let But now, this is an equation for w(ω, k), because Dρ
the perturbed velocity components be δv = (u, v, w). is specified in the problem. Picking w = 0 on the (dis-
The linearized equations – for the perturbations δρ, δp, tant) boundaries of the system, this becomes a Sturmian
keeping only lowest order “small” terms, are: characteristic value problem for ω 2 . One can show,
∂u ∂ then, that ω 2 > 0 if Dρ < 0 everywhere; and con-
ρ = − δp versely, ω 2 < 0 if Dρ > 0 somewhere in the fluid.
∂t ∂x That is: a density structure which decreases with ẑ is
∂v ∂
ρ = − δp stable; while a density structure which increases with ẑ
∂t ∂y is unstable.
∂w ∂ The next step is to find out the growth rates, and
ρ = − δp − gδρ (19.1)
∂t ∂z whether all wavevectors are unstable. To do this, we
∂u ∂v ∂w simplify to the case of two uniform fluids, separated by
+ + =0
∂x ∂y ∂z a horizontal interface at z = 0. Now, away from this
∂ dρ boundary, (19.6) becomes
δρ = −w
∂t dz
D2 w = k2 w (19.7)
Now, assume each perturbed quantity has the form (a
“mode”) which has simple exponential solutions. Picking the
ones which stay well-behaved at z → ±∞, we have
δf (x, y, z, t) = δf (z)ei(kx x+ky y+ωt) (19.2) away from the boundary
We will generally treat k as given, and want to find the
nature of ω. In particular: if ω is real, (19.2) simply w(z) = Ae−kz ; z>0
represents waves or oscillations – so that the system is (19.8)
kz
stable. However if ω has a negative imaginary part, then w(z) = Ae ; z<0
 106

To connect solutions through the boundary, we need (Why is it proportional to the perturbed velocity?) The
jump conditions. Define perturbed momentum equations become
∆s f = f (zs + 0) − f (zs − 0) iρωu = −ikx δp
that is, the jump in f from the below some surface zs to B
above that surface. Here zs = 0. Applied to (??), we iρωv = (ikx by − iky bx ) − iky δp
4π
find the jumps in p and ρ are related by B g
iρωw = (ikx bz − iDbx ) − iky δp + wDρ
g 4π iω
∆o δp = ws ∆o ρ (19.9)
ω (19.14)
Manipulation of the system (??) also gives
Now: using (19.11) to convert from b components to v
2
k ∆o δp = −∆o (−iωρ) (19.10) components, we can show
kx2 B 2
 
Finally, these combine to give iρω + (ikx v − iky u) = 0 (19.15)
iω 4π
k2
∆o (ρDw) = gw∆o ρ (19.11) which requires kx v = ky u. Using this result, and fur-
ω2
ther algebra, we find
But in this system, ∆o ρ = ρ2 − ρ1 (upper = lower val-
ues); and Dw = ±kw (- for upper, + for lower). The iρωDw = −k2 δp (19.16)
result (19.11) thus becomes
and
 
2 ρ2 − ρ1 B 2 kx2 g
ω = −gk (19.12) iρωw − D 2
− k 2


w = −Dδp + wDρ
ρ2 + ρ1 2
4πk iω iω
(19.17)
This is our result: ω 2 < 0 – the system is unstable – And now, we can look for the new results. First,
if ρ2 > ρ1 . This recovers what we guess from potential combine (19.16) and (19.17) to get
energy arguments, at the start. We also find that all
wavenumbers are unstable 2
√ (that is, have ω < 0), with B 2 kx2 gk2
D (ρDw)−ρwk2 − 2 2


D − k w = wDρ
a growth rate |ω| ∝ k. This last result is true for 4πk2 ω 2 ω2
small perturbations; however when the instability starts (19.18)
to grow, larger-scales dominate the system. which is the generalization of (19.4) for the non-
magnetic case.
This analysis could also have included surface ten-
sion at the interface: when included, one finds that sur- From this, first, we note that if kx = 0, which cor-
face tension stabilitizies high wavenumbers (small spa- responds to propagation perpendicular to the field, (??)
tial scales). just reduces to (19.6): cross-field perturbations have no
effect on the RT instability. The ones with kx 6= 0,
3. THE MAGNETIZED CASE however, do change the results. We can repeat the non-
magnetized analysis . . . the jump condition (19.11)
What happens if we add a magnetic field? We expect generalizes to
little effect from a vertical field, as it doesn’t hamper
vertical fluid motions. This turns out to be the case – B 2 kx2 k2
the stability conditions are not affected. It does have ∆o (ρDw) − ∆ o (Dw) = gw∆o ρ (19.19)
4πω 2 ω2
some effect on the growth rate, however (note that a
Again taking the case of two uniform fluids separated
vertical perturbed velocity must also involve horizon-
by an interface, the result (19.12) becomes
tal, cross-field flows – by the continuity equation). The
more interesting case is a horizontal magnetic field: we 2B 2 kx2
 
2 ρ2 − ρ1
expect this to affect the stability as well as the growth ω = −gk + (19.20)
ρ2 + ρ1 4π (ρ2 + ρ1 )
rates, and it does.
In this section I still follow Chandra. Consider a Thus: a horizontal magnetic field stabilizes high fre-
uniform B = Bx̂, with perturbed field quency perturbations. Instability (ω 2 < 0) requires
kx 2B 2 kx2
δB = (bx , by , bz ) = Bv (19.13) gk (ρ2 − ρ1 ) > (19.21)
ω 4π
 107

Finally, we note that this analysis assumed an in- The basic perturbed equations are
compressible fluid. That simplifies the algebra but isn’t
∂u ∂u dU ∂
required for the physics. Extending to the compressible ρ + ρU + ρw = − δp
case (as done by Shivamoggi, Theory of Hydromagnetic ∂t ∂x dz ∂x
Stability), one finds that compressibilty has a stabiliz- ∂v ∂v ∂
ρ + ρU = − δp
ing effect, in that it reduces the growth rates in both the ∂t ∂x ∂y
magnetized and unmagnetized cases, compared to their ∂w ∂w ∂
ρ + ρU = − δp (19.22)
incompressible analogs. ∂t ∂x ∂z
∂u ∂v ∂w
C. The Kelvin Helmholtz Instability + + =0
∂x ∂y ∂z
Well, that was so much fun, let’s do it again. The other ∂ ∂ dρ
δρ + U δp = −w
well-known instability of an interface is the Kelvin Hel- ∂t ∂x dz
moltz Instability. We again assume the perturbation has the form, (19.2).
The system (??) becomes
1. THE PHYSICS
ikx δp = −iνρu − ρwDU
This instability arises at a velocity shear. Consider two
fluids, or two pieces of the same fluid, in relative mo-
iky δp = −iνρv
tion. For instance, think of a horizontal interface; fluid
below is at rest, and fluid above is moving parallel to Dδp = −iνρw (19.23)
the interface. (This might describe a situation in the at-
mosphere, with high-velocity winds going past a low- ikx u + iky v + Dw = 0
velocity cloud layer). When the interface is displaced
upwards, it forces the overlying flow to deviate over iνδρ = −wDρ
the perturbation. This means the pressure there drops
(Bernoulli, remember?). The boundary thus feels a lift. where we have defined the “shifted” frequency, ν =
But of course the situation is mirror symmetric across ω + kx U . From these we can find
the interface; the downward displaced boundary feels
a downward “lift”. As the boundary follows these per- iρνDw − iρkx wDU = −k2 δp (19.24)
turbations, it is sheared by the flow into which it pene- and
trates, and rollw up into a vortex sheet.
ik2 Dδp = ρk2 νw (19.25)
(compare equations 19.4 and 19.5). These combine
v
2 (eliminate δp) to give
D [ρνDw − ρkx wDU ] − k2 ρνw = 0 (19.26)
and from this, our useful jump condition , at an inter-
v
1 face zs , is
Figure 19.2. The origin of the Kelvin-Helmholtz ∆s [ρνDw − ρkx wDU ] = 0 (19.27)
instability. Shear at an interface between two fluid layers in
relative motion generates a vortex sheet; pressure (what quantities must be held constant across an inter-
differences (think of the Bernoulli effect) lead to instability, face? how did we throw out some of the terms in ???)
and eventual mixing of the two fluids.
We again apply this to two uniform fluids. Above
and below the boundary, we again have (19.7), as all
2. THE MATH
else is constant. But now, the ratio w/(ω + kx U ) must
be continuous across the boundary (why?). Thus, (19.8)
We again follow Chandrasekhar. The methods are sim- is replaced by
ilar to those presented for the RT instability, but the ge-
ometry is slightly more complicated. We add a hori- w(z) = Aν2 e−kz ; z>0
zontal flow U = ux̂ to the unperturbed system, but we (19.28)
kz
now ignore gravity. w(z) = Aν1 e ; z<0
 108

(where the subscripts 2,1 refer to (top),(bottom) flow U; we don’t expect cross-field flows to differ from the
speeds). Our characteristic equation is simpler here. unmagnetized case.
Taking ρ1 = ρ2 = ρ, we can derive The modal equations become
(ω + kx U2 )2 + (ω + kx U1 )2 = 0 (19.29)

But this has only complex roots (if kx 6= 0). The ikx δp = −iνρu − ρwDU
quadratic solves to
B
1 i iky δp = −iνρv + (ikx by − iky bx )
ω = − kx (U1 + U2 ) ± kx (U1 − U2 ) (19.30) 4π
2 2 B
Dδp = −iνρw + (ikx bz − Dbx )
4π (19.31)
Thus, any such perturbation is unstable (ω has an
imaginary part), no matter how small the difference ikx u + iky v + Dw = 0
U1 − U2 may be.
iνδB = ikx Bv + bz DU
As Chandrasekhar quotes Helmholtz: “Every per-
fect geometrically sharp edge by which a fluid iνδρ = −wDρ
flows must tear it asunder and establish a surface
of separation, however slowly the rest of the fluid
may move”.
From these, after a fair bit of algebra, and again using
Equation (19.30) shows that higher k’s grow the the “shifted” frequency ν = ω+kx U , we find (compare
fastest when the perturbation is small. When it becomes 19.16 and 19.17)
finite, however, this analysis breaks down; experiment
shows that a characteristic and large scale will dominate
the instability.
ρνDw − ρkx wDU = ik2 δp (19.32)
3. THE MAGNETIZED CASE

Finally, we can again consider the effect of a magnetic

field. We expect the interesting case will be when B k and

B 2
 
Dw

k2 w
 2  
2 2 3B DU
ik Dδp = ρk νw + kx2 D − − kx D w (19.33)
4π ν ν 4π ν2
These two combine as
2 k2 w 2
     
B Dw 3B DU
D (ρνw − ρkx wDU ) + kx2 D − − kx D w (19.34)
4π ν ν 4π ν2

(compare 19.16). Specializing once more to two uni- teristic equation:

form fluids separated by a boundary at z = 0, our jump
condition is 2B 2
(ω + kx U2 )2 + (ω + kx U1 )2 = kx2 (19.36)
4π
2
 
B Dw The roots of this are
∆o (ρνDw) = kx2 ∆o (19.35)
4π ν 1/2
2 kx2

1 2 B 2
ω = − kx (U1 + U2 )± kx − (U1 − U2 )
2 4πρ 4
From this, going back from ν to ω, we get the charac- (19.37)
 109

And so, finally, we have our result: a B field aligned

with the flow will stabilize the KH instability at all
wavelengths if

4B 2
(U1 − U2 )2 ≤ (19.38)
4πρ

Finally, again, one can also do this for compress-

ible fluids. As with RT, one finds that compressibil-
ity is stabilizing; it reduces the growth rate, in both the
unmagnetized and magnetized cases, compared to our
incompressible calculation here.

References
As in the text, I’ve followed Chandrasekhar, Hy-
drodynamic and Hydromagnetic Stability for the hydro,
and Shivamoggi, Theory of Hydromagnetic Stability),
for the MHD.