31

6. BASICS OF COMPRESSIBLE FLOW 1. CONSTANT GRAVITY: THE EXPONENTIAL

ATMOSPHERE

Up to now we have assumed incompressible flows. We One example is hydrostatic balance in a fixed (external)
now extend to the more general case, when the den- gravitational field g. The most obvious application of
sity is allowed to vary. This is necessary to work with this is the earth’s atmosphere. The basic equation is:
(i) wave propagation; (ii) hydrostatic atmospheres, over
scales large compared with the scale height; (iii) flows ∇p = ρg (6.1)
in which the fluid speed is comparable to (or larger
Or in a vertically stratified medium, such as earth’s at-
than) the sound speed.
mosphere, this is
dp
A. Some useful thermodynamic quantities = ρg (6.2)
dz
and g is constant over any relevant scale in this applica-
Since density changes are accompanied by temperature tion. We can directly define a scale height from this. If
changes, we need to remember a bit of thermodynam- T is also constant with height, the pressure solution is
ics. I store some terms and definitions here, in no partic-
ular order (some of the definitions are circular..). When p(z) = po e−z/H (6.3)
specific expressions are needed, I work with an ideal
gas. where H = RT /g = kT /mg is the scale height.
• Equation of state. p = nkB T = ρRT ; kB is the Another example of this, perhaps less obvious,
Boltzmann constant, a fundamental constant. R is the is the interstellar medium in our galaxy. For non-
gas constant, which varies by composition. It’s related astronomers: our galaxy is a flat disk, composed of stars
to the “fundamental” gas constant R by R = R/M , if and gas rotating about a common center. The thick-
M is the molecular/atomic weight of the gas in ques- ness of the disk is controlled by the local mass density
tion. Or if you prefer microphysics, R is related to the (which produces a constant gravity, g), and the temper-
Boltzmann constant by R = kB /m, if m is the mass ature (plus random motions) of the gas or stars in the
per particle. disk. The vertical structure of the disk is controlled by
the same equation, (6.3) ... just on a much larger scale.
• Internal energy and enthalpy. Let e be the internal
energy (per mass): we define the enthalpy as h = e + 2. VARIABLE GRAVITY: THE ISOTHERMAL
p/ρ. For a simple ideal gas we have p = nkT ; e = SPHERE
kT /m (if m is the mass per particle); also ρe = p/(γ −
1); and ρh = γp/(γ − 1). A different HSEq system is a self-gravitating sphere.
• Specific heats and adiabatic index. The SH’s This might be a naive picture of a star: held together by
are defined by de/dT = cV (constant volume) and its own gravity, and supported against gravitational col-
dh/dT = cP (constant pressure). The adiabatic in- lapse by its internal energy. The basic HSEq equation,
dex can be defined as γ = cP /cV , the ratio of specific (6.1) still holds; in spherical geometry it is
heats. We also have cp − cV = R; cv = R/(γ − 1); dp GM (r)
cp = γR/(γ − 1). = −ρ (6.4)
dr r2
• Degrees of freedom and adiabatic index. Let f be
where M (r) is the mass inside r. Alternatively, this can
the number of d’s of f, so that the mean KE per particle
be written
is (f /2)kB T . This connects to the adiabatic index by
γ = (f + 2)/f . Typical values for γ are γ = 1.4 for the dp
= ρ∇Φg (6.5)
atmosphere; γ = 5/3 for a monatomic ideal gas; and dr
γ = 4/3 for a relativistic gas (such as a photon gas). where Φg is the gravitational potential, defined through
g = −∇Φg . The mass, density and Φg are connected
B. Hydrostatics: gaseous atmospheres via
Z r
M (r) = 4πr 2 ρdr ;
The effects of compressibility must be considered in 0
  (6.6)
most situations involving gases (or plasmas) in static 2 1 d 2 dΦ
∇ Φg = 2 r = 4πGρ
equilibrium in a gravitational field. r dr dr
 32

Putting this into (6.5) and differentiating with r, the ba- Numerical solutions of this equation are called poly-
sic equation becomes tropes, or Lane-Emden solutions, for polytropic index
n = 1/(γ − 1). These are physically well-behaved, in
d r 2 dρ
 
G that the density is finite at the center and falls faster than
=− 4πr 2 ρ (6.7)
dr ρ dr RT 1/r 2 at large r.
The solutions of this are less straightforward. First, we 3. REALITY: NONISOTHERMAL ATMOSPHERES
note that a basic scale length appears:
Just a note of caution here. Both of the preceding ex-
9RT 1/2
 
amples assumed a constant temperature throughout the
ao = (6.8)
4πGρo atmosphere. This naive assumption allowed us to find
(nearly) analytic solutions. More realistically, however,
if ρo is some characteristic density, say that at the ori- we expect the temperature to be a function of position,
gin. (the numerical constants appear to simplify things controlled by the thermodynamics of the system. For
later on.) We expect physical solutions of (6.7) to in- earth’s atmosphere, T (z) is determined by the com-
volve lengths scaled to ao . bined actions of solar radiation coming in at the top,
Consider an isothermal gas; these are the simplest all the chemistry and energy transfer effects within the
solutions. One solution of (6.7) is the simple power law, atmosphere, and radiative losses back into space. For a
ρ ∝ 1/r 2 . The divergeance as r → 0 keeps this from star, T (r) is determined by nuclear energy generation
being an intersting solution, however. More interesting in the star’s core, radiative energy transfer within the
physical solutions must be found numerically, starting star, and eventual radiative losses into space.
with ρ(r = 0) = ρo and working out. These solutions
do, indeed, have a turnover at r ≃ ao ; at large radii 4. ADIABATIC ATMOSPHERE
they do approach ρ ∝ 1/r 2 . (These solutions also have
Another analytic model of an atmosphere assumes the
problems, for the total mass M (r) diverges ∝ ln r as
gas is adiabatic: that is, as it compresses or expands
r → ∞. Physically, the problem is that the gas cannot
(while satisfying HSEq), it heats or cools accordingly.
be maintained isothermal everywhere.)
This is useful in analyzing the stability of an atmo-
sphere to convection (which we’ll do immediately be-
low).
To develop this idea, go back to basic HSEq, (6.1):
dp/dz = −ρg. But now, assume the gas is adiabatic,
so that
 (γ−1)/γ  1/γ
T p ρ p
= ; = (6.10)
To po ρo po
We can use these to relate dT /dz to dp/dz for an adi-
abatic atmosphere:
1 dT γ − 1 1 dp 1 dρ 1 1 dp
= ; = (6.11)
T dz γ p dz ρ dz γ p dz
Figure 6.1. Density (and projected density, Σ) solution
for a self-gravitating isothermal sphere. The dotted line is From here, assuming p = ρRT (ideal gas) and using
the asymptotic ρ ∝ 1/r2 power-law solution. From Binney cp − cv = R (refer back to §6.1) we get the condition
& Tremaine figure 6.7 for an adiabatic atmosphere:
Self-gravitating spheres can also be modelled as adi- dTad g (γ − 1) mg
abatic gases. Take the equation of state as p = Kργ . =− =− (6.12)
dz cp γ kB
The hydrostatic equation, (6.5), and the differential
form (6.7), become where cp is the isobaric specific heat, and m is the mass
per particle. This clearly gives a linear temperature
dρ dΦ drop with altitude z; g/cp ≃ 10◦ C/km for typical atmo-
Kγργ−2 = −ρ ;
dr dr spheric conditions. The vertical temperature gradient is
 (6.9)
1 d 2 dΦ often caled Γ = dT /dz; and is sometimes called the
r = 4πGρ
r 2 dr dr “lapse rate”.
 33

C. Convective Stability ings. If this is the case, we will have

 ∗ 1/γ
So ... we now have a description of hydrostatic atmo- pin
pin = pout ; ρin =ρin
∗ ∗ ∗
;
spheres. But, are they stable? One of the most impor- pin
tant instabilities, for the earth’s atmosphere or for a star,  ∗ 1/(γ−1) (6.14)
pin
is convection. We need to understand whether the at- Tin = Tin
∗
pin
mosphere is stable, or unstable, to convection – that is,
whether or not convection will develop spontaneously.
To picture the situation, here’s how Shore puts it:1 *
n no (z+dz), T o(z+dz)
in
Picture a duck sitting calmly on a pond. . . If we
say that the bird is bouyant, we mean that if we
depress him a bit by pushing from above, he will
g
bob back to the surface and, ignoring his agitation, dz
bounce up and down for awhile [this is called neu-
tral bouyancy.]. . .If we have one, on the other
hand, who is not well preened and therefore not
waterproof, and [we] push down on him, he may
sink. Now think of a blob which is hotter than nin n (z), T (z)
o o
its surroundings. It will begin to rise, since we
already know that its density will be lower than Figure 6.2. A cartoon illustrating the buoyant/convective
that of the [surrounding] medium, and it will thus instability. Imagine a blob starts in balance with its
be bouyant. If it remains underdense, it will con- surroundings. It is then displaced vertically, some dz;
during this rise it (i) remains in pressure balance with its
tinue to rise – we call this an instability. It will surroundings, and (ii) remains adiabatic. How does its
continue to rise until it reaches a level at which it density at z + dz compare to the density outside? The
is neutrally bouyant again. On the other hand, if answer to this question determines the stability of the
the blob is pushed down, and if it remains over- atmosphere.
dense, it will sink until it reaches a point at which
the density again allows for stable balance. The surroundings, however, are not necessarily adi-
abatic: they have some other dT /dz and dρ/dz val-
We would like to find a condition to tell if the sit- ues (specified by the situation – for instance the heat-
uation is stable or unstable. To get there, think about ing/cooling balance for the outer layers of a star, or the
some blob again: assume it starts at some vertical po- earth’s atmosphere). So, we can determine stability or
sition z, with density and pressure in balance with its instability by asking whether, when the blob has risen
surroundings (i.e, the “outside”, which usually refers to this ∆z, it is at a higher or lower density than the sur-
the atmosphere). Thus: it starts at ρin = ρout = ρ1 and rounding atmosphere. In the first case it will sink again
Tin = Tout = T1 . Now, raise it some distance ∆z; let (and thus the atmosphere is stable); in the second case it
the conditions in the blob at z + ∆z be labelled by “∗”. will keep rising (and thus the atmosphere is unstable).
For the blob, we assume it evolves adiabatically. Thus, Thus, our condition for instability becomes condition
it reaches a new density and temperature, on the external density gradient. If we assume the blob
remains in pressure balance with its surroundings, we
also have a condition on the atmospheric temperature
    gradient. Thus, the atmosphere is buoyantly unstable if
dρ dT
ρ∗in = ρ1 + ; Tin
∗
= T1 +        
dz dz dρ dρ dT dT
ad ad < ; >
(6.13) dz ad dz atm dz ad dz atm
If we want to evaluate the z-derivatives in (6.13), we (6.15)
need to know something about the specific situation. Because we usually consider situations with dρ/dz <
A common assumption is that the blob rises slowly 0, for instance a hydrostatic atmosphere, the condition
enough to remain in pressure balance with its surround- for instabilty is often written in terms of absolute val-
ues:
1
An Introduction to Astrophysical Hydrodynamics (Academic dρ dρ dT dT
> ; < (6.16)
Press) 1992, ch. 9. dz ad dz atm dz ad dz atm
 34

Thus: if the outside (atmospheric) temperature changes

too rapidly with altitude, the atmosphere is convectively
unstable. An underdense blob will continue to rise, and
an overdense blob will sink.
1. ADIABATIC ATMOSPHERE

Referring to our discussion of adiabatic atmospheres,

above, we recall that the real atmosphere needs to be
compared to the ideal, adiabatic case in order to de-
termine convective (in)stability. This gives a condition
for instability, written in terms of either the pressure or
temperature gradients: Figure 6.3. Vertical variation of the (a) actual and (b)
potential temperature in a typical terrestrial atmosphere.
The straight lines are the adiabatic Tad (z) structure. From
 
1 1 dp 1 dρ 1 T dp dT
> ; 1− > (6.17) Kundu figure 1.9.
γ p dz ρ dz γ p dz dz
Note that both p and T drop with z in any hydrostatic
while the ambient density has changed as
situation; so both sides of these inequalities involve
negative quantities. dρ
ρ(z + dz) = ρ(z) + dz (6.20)
2. POTENTIAL TEMPERATURE dz
(Note, I’ve now dropped the subsript “out”). Thus, the
The adiabatic gradient can be important in the convec-
bouyant force (per unit volume of the blob/duck) is
tive stability of the atmosphere. Another useful quan-
tity here is the potential temperature. Consider a par-  
dρ
 
dρ
cel of atmosphere, that starts in local balance at some Fbuoy = g − ∆z (6.21)
dz ad dz
(p, T ). Take it adiabatically to some other point, with
local pressure ps (typically sea level pressure). The Alternatively, using the definition of potential tempera-
temperature this parcel reaches at point s is called its ture, this becomes
potential temperature – called θ – which is given by
g dθ
 (γ−1)/γ
p T dθ d Fbuoy = ∆z (6.22)
T =θ ; = (Tad − T ) θ dz
ps θ dz dz
(6.18) So: the buoyant force ∝ ∆z – this is clearly simple har-
But this is useful, because the gradient of θ depends monic motion. The equation of motion (still per mass)
on the difference between the actual and adiabatic gra- can be written
dients in the atmosphere; thus θ is a handy tool when d2 ∆z 1
one’s thinking about convection. Specifically, from the 2
= Fbuoy = N 2 ∆z (6.23)
dt ρ
results above, we see that the atmosphere is convec-
tively stable if dθ/dz > 0, and unstable if dθ/dz > 0. and this last implicitly defines the Brunt-Väisälä fre-
quency, N , frequency,2
3. BRUNT- V ÄIS ÄL Ä FREQUENCY
   
2 g dρ dρ
Now, let’s return to the blob (or the duck). If the at- N = − (6.24)
mosphere is convectively stable, the blob (or duck) will ρ dz ad dz
simply bob up and down. Its easy to show this is sim- If the right hand side is positive (which is the same thing
ply harmonic motion, and to find its characteristic fre- as saying the atmosphere is convectively stable – refer
quency. back to 6.15) , the blob simply oscillates at the B-V
To proceed, picture the earth’s atmosphere. Move frequency. If the RHS is negative (if the atmosphere is
the blob vertically by some ∆z, so that it reaches a new unstable), the displacement ∆z grows exponentially.
density ρ∗in . The blob’s new density, still assuming an
adiabatic displacement, will be
2
Caveat to the student: very few authors leave N 2 in this form
1 ρ dp – they express it in terms of θ, dT /dz, specific heats, and what
ρ∗in = ρ1 + ∆ρ∗ = ρ(z) + dz (6.19)
γ p dz have you.
 35

D. Energetics of Compressible Flow find

   
We also need an equation for energy conservation. We ∂ 1 1
ρ e + v 2 +ρv · ∇ e + v 2
consider two forms of energy: kinetic energy density, ∂t 2 2 (6.28)
v 2 /2, and internal energy density, e, both defined per
unit mass. = ρg · v − ∇ · (pv) + H
The net energy in our volume V is V ρ(e+ 12 v 2 )dV .
R
This one alternate form that we will use again. We can
The net rate of change of this energy, from intrinsic isolate the rate of change of e, by subtracting v·(the
changes and from flows is momentum conservation equation) from (6.28), giving
 
∂ 1 2 ∂e
Z
ρe + ρv dV + ρ + ρv · ∇e = −p∇ · v + H (6.29)
V ∂t 2 ∂t
   (6.25)
1 2
Z
+ ∇ · v ρe + ρv dV In this expression, we can see that the rate of change of
V 2 the internal energy depends explicitly on compression
work (“pdV ” work), and on the net heating and cooling
This net energy change must be accounted for by (a) rates.
work done by the external force, f ; (b) work done by Yet another common form of the energy equation
the external pressure; (c) direct energy gains or losses,3 uses the Lagrangian derivative. Writing ∇ · v in terms
which we collect as H.4 These three energy-change of the density derivatives, and collecting the p and ρ
factors are derivatives separately, we get
Z Z Z
 
ρf · vdV − pn̂ · vdA + HdV (6.26) D p
V A V = (γ − 1)H (6.30)
Dt ργ
We can use Gauss’ law to express the pressure work which is the last of our alternate forms of the energy
term as a volume integral, and can derive one version equation. This last form allows us to consider a couple
of the differential energy conservation law: of important limits.
    
∂ 1 1 • The first is the adiabatic limit. If H = 0, so that
ρe + ρv 2 +∇ · v ρe + p + ρv 2 there is no net gain or loss of energy to the system,
∂t 2 2
(6.30) shows that
= ρf · v + H (6.27)
p
= constant (6.31)
ργ
Note that the enthalpy h = e + p/ρ appears naturally in
the second term on the LHS. which is the usual adiabatic law (the consequence
At this point, we need to look at alternative forms of there being no gain or loss of heat from a sys-
of (6.27). The forms we derived for mass conserva- tem).5
tion (1.4) and momentum conservation (1.10 or 2.2) •The second limit is the isothermal limit. A good
are pretty standard. However, there does not seem to many astrophysical calculations assume T = con-
be one standard form for the energy conservation equa- stant, which simplifies things enormously. From
tion; rather, one uses the form that works best in a given (6.22), we note that
application. Therefore, at the expense of a little algebra,
we will look at several alternate forms of (6.27). p∇ · v = H (6.32)
First, we simply separate out the ∂ρ/∂t and ∇·(ρv)
is the condition that must be satisfied if T (or e) is
terms in (6.27), using the continuity equation (1.4), and
constant.
It might be comforting to prove that we can extract
3
Examples of direct heating include resistive dissipation (of a cur- Bernoulli’s relationship from this formalism, in addi-
rent), or such things as cosmic ray heating (relevant in astro-
tion to our earlier derivation from the force (Euler’s)
physics); direct losses are most commonly by radiation
4
We will treat viscous dissipation below. We could also include
equation. We assume, again, H = 0. We use the ex-
thermal conductivity, which provides a separate means of energy pression for De/Dt, and expand out the ∇ · (pv) term,
transport out of or into the volume; it brings in second deriva- using the continuity equation. We also assume the sys-
tives, and we will not need it here. tem is in a steady state, so that ∂/∂t = 0; thus, D/Dt
 36

measures the rate of change of a quantity, due to its mo- cluding viscosity, equation (2.2); multiply it by vi and
tion through a region in which the flow field changes. interchange dummy indices, to get an expression for the
We get, time-change of kinetic energy:
   
D 1 2 D p D v2
ρ e+ v +ρ = ρg (6.33) ρ = σik,i uk + ρgk vk (6.35)
Dt 2 Dt ρ Dt 2
Now, using g = ∇Φ, Bernoulli’s relationship becomes Subtract this from the full energy equation (also written
in Cartesian), and summetrize the tensor term (in the
1 p 1
e + v 2 + + Φ = h + v 2 + Φ = constant (6.34) second step), to get two forms of the energy equation
2 ρ 2 including viscosity:
which does, indeed, recover the Bernoulli relation that
De De dvk
we derived above from momentum conservation. We ρ = σik vk,i = σik Dik ; ρ= −p +Σik Dik
recall, again, that this law holds along any one stream- Dt Dt dxk
(6.36)
line in the flow.
This brings back the (symmetrized) deformation tensor
in Cartesian:
 
1 ∂vi ∂vk
References Dij = + (6.37)
2 ∂xk ∂xi
I mostly follow Thompson for the basic energy
conservation and dissipation analysis. The isothermal and introduces what I’ll call the dissipation function:
sphere discussion follows Binney & Tremaine (Galac- D = Σij Dij . Referring back to (2.37) and (2.38),
tic Dynamics); the terrestrial atmosphere discussion it’s normal to assume νb = 2ν/3, which simplifies the
leans on Kundu and Tritton. stress and dissipation tensors. In this limit, still Carte-
sian, the energy equation becomes
 2
∂e ∂e ∂vk ∂vk
ρ + ρvi +p = ρν (6.38)
∂t ∂xi ∂xk ∂xk
E. Appendix: Viscous Dissipation
Well, that was so much fun, let’s do it again: in Carte- To close, I write out explicitly the dissipation func-
sian to be explicit. Start with our force equation, in- tions for all 3 coordinate systems.

Cartesian:
2ρν 
(D11 − D22 )2 + (D22 − D33 )2 + (D33 − D11 )2

D=
3 (6.39)
2 2 2
+ ρνb (D11 + D22 + D33 )2


+ 4ρν D12 + D13 + D23
where, as above,
 
1 ∂vi ∂vk
Dij = + (6.40)
2 ∂xk ∂xi
Cylindrical:
2 2 2 2 2 2
+ ρ(νb − 2ν/3)(∇ · v)2


D = 2ρν Drr + Dθθ + Dzz + 2Drθ + 2Dθz + 2Dzr (6.41)
where
 
∂vr 1 ∂vθ vr 1 1 ∂vr ∂ vθ
Drr = ; Dθθ = + ; Drθ = +r
∂r r ∂θ r 2 r ∂θ ∂r r
    (6.42)
∂vz 1 ∂vθ 1 ∂vz 1 ∂vr ∂vz
Dzz = ; Dθz = + ; Drz = +
∂z 2 dz r ∂θ 2 ∂z ∂r
 37

Spherical is even better (recall that θ is the polar angle and φ is the azimuthal angle):
2 2 2 2 2 2
+ ρ(νb − 2ν/3)(∇ · v)2


D = 2ρν Drr + Dθθ + Dφφ + 2Drθ + 2Dθφ + 2Dφr (6.43)

where
∂vr 1 ∂vφ vr vθ cot θ 1 ∂vθ vr
Drr = ; Dφφ = + + ; Dθθ = +
∂r r sin θ ∂φ r r r ∂θ r
   
1 1 ∂vr ∂ vφ 1 1 ∂vr ∂ vθ
Drφ = +r ; Drθ = +r (6.44)
2 r sin θ ∂φ ∂r r 2 r ∂θ ∂r r
 
1 sin θ ∂ vφ 1 ∂vθ
Dθφ = +
2 r ∂θ sin θ r sin θ ∂φ

and this is the end of this chapter.