1

1. BASIC CONCEPTS AND TOOLS The streamline is defined by

dx dy dz
= = (1.1)
vx vy vz
In this course, we will study the dynamics of fluids, If the velocity field is specified, the streamline can be
that is, hydrodynamics (HD). What, in this context, is found directly from (1.1). In addition, (1.1 is equivalent
a fluid? For the purposes of this course, there are three to ds × v = 0; no fluid crosses the streamline. We can
answers. (1) A normal, everyday liquid – such as wa- also visualize a streamtube: a tubular region within the
ter or molasses. The equation of state will depend on fluid, bounded by a set of streamlines (think of them
the substance; however, in most applications we will as pink). Because streamlines can’t intersect, the same
assume the density of the liquid is constant. (2) A dif- streamlines follow the streamtube everywhere along its
fuse gas – such as the atmosphere. The density of such length.
a gas is definitely not constant; the relevant equation of Two other concepts are sometimes used. The path
state is the ideal gas law. (3) An ionized plasma – such line is the trajectory of a fluid particle of fixed identity,
as in laboratory plasma experiments, the upper layer’s over a period of time. (Paint a single fluid element red,
of earth’s atmosphere, and nearly all astrophysical sit- say, and take a series of photos which display its lo-
uations. The ideal gas law also applies here. In ad- cation as time passes.) The streakline, through a fixed
dition, the free charges in plasmas allow current flow; spatial point, is defined as the current location of all
we must extend our subject matter to magnetohydrody- fluid particles which have passed through that point at
namics (MHD). some previous time. (Picture injecting dye at a fixed
We will generally assume a fluid can be described point in a flow, and taking one photo at some later time.
in terms of macroscopic quantites. Our three basic vari- The dye traces out the streakline). All three of these
ables will be the density ρ, the pressure p, and the ve- “lines” are the same in a steady flow; in a time-varying
locity v. We will often also use an equation of state, flow they will differ.
relating p and ρ. We will occasionally need transport:
most often the viscosity ν, also thermal conductivity κ, STREAM FUNCTION
and the electricial resistivity η. These are often deriv-
Now consider a special case, when the flow is incom-
able from the microphysics, but are usually considered
pressible. That means constant density, and turns out
as given constants for a particular problem.
to be equivalent to ∇ · v = 0 (we’ll prove this im-
A few comments are in order before we get into mediately below). Now, we know from E&M that a
specifics. vector field with zero divergeance can be written as the
• My units are cgs, not SI. This matters very little for curl of some other vector: v = ∇ × A, say. This
normal fluid mechanics, but is critical for MHD. I will isn’t necessarily useful ... unless the flow field is two-
try also to introduce SI variants of basic electrodynamic dimensional. In that case, the vector potential A has
terms and equations as they come up. only one non-zero component. To be specific, consider
• My notation is driven somewhat by the vagaries of Cartesian coordinates, and assume the flow field only
TEX. vectors are denoted by bold face roman, e.g., A; depends on (x, y). We can then write the (vector) po-
and for greek, either by bold (e.g., ω , or by over-arrows tential as A = (0, 0, ψ), for some function ψ(x, y); and
(e.g., ~ω ). Tensors are more problematic: possibilities the velocity field is
~
are direct subscripts, Tij ; T~ is possible, ab for a product v = −∇ × (ψẑ) (1.2)
of two vectors, and occasionally bold face capitals (as
This function ψ(x, y) is called the stream function.
in P).
We’ll return to this in more detail in chapter 3, where
we’ll find that streamlines in the flow are also lines of
A. Kinematics: How to Describe a Flow
constant ψ: each ψ value labels one particular stream-
line.
Visualizing a complex flow field can be important in
understanding a problem. The most common concept B. Mass Conservation: The Continuity Equa-
is the streamline. This is a line (one of an infinite set tion
of lines) that is everywhere tangent to the velocity vec-
tor in the flow. For instance, let ds = (dx, dy, dz) be This will be one of our most basic tools. Consider an
the element of arc length along a steamline; and let arbtrary volume of fluid, V , bounded by a closed sur-
v = (vx , vy , vz ) be the velocity vector at that point. face, A; let the surface have an outward normal, n̂. The
 2

R
mass within this volume is V ρdV , if ρ is the mass Now, the net rate of change of momentum in the vol-
density. The net rate of change of this mass is ume must be equal to the net force exerted on the vol-
ume. We consider external forces which act throughout
d
Z
ρdV ; the volume (“body” forces, such as gravity, electomag-
dt V netism, bouyancy, radiation pressure if the fluid is opti-
cally thin; we let g be the net force per mass), and also
if there are no sources or sinks of matter, this quantity
the force exerted on the surface by the fluid outside V .
must equal zero. Now, there are two ways this integral
The net force on the volume V is, then,
can change with time. (i) there can be intrinsic variation Z Z Z Z
of ρ, ∂ρ/∂t 6= 0; or (ii) there can be flow into or out of ρgdV − pn̂dA = ρgdV − ∇pdV (1.8)
the volume, at a rate ρv · n̂ per surface area. The sum V A V V
of (i) and (ii) must balance out to zero: where we have again used vector identities in the last
d
Z Z
∂ρ
Z step. This balance, (1.7) and (), can then be written in
ρdV = dV + ρv · n̂dA = 0 (1.3) differential form,
dt V V ∂t A
∂
R (ρv) + ∇ · (ρvv) = −∇p + ρg (1.9)
RBut the surface integral can be written as A ρv· n̂dA = ∂t
V ∇ · (ρv)dV . Since V is arbitrary, we can set the This is our basic force equation, known as Euler’s equa-
integrand to zero, and we get the differential form of tion. Note that the vv term is a tensor.1
this basic equation:
It is conventional to simplify this, by expanding the
∂ρ derivatives on the left hand side and using (1.4); we get
+ ∇ · (ρv) = 0 (1.4)
∂t ∂v
ρ + ρ(v · ∇)v = −∇p + ρg (1.10)
This is, of course, the continuity equation, applied to ∂t
mass conservation. Another version of Euler’s equation holds is useful if
there is no body force. To get to this, write the pres-
LAGRANGIAN DERIVATIVE sure in terms of the pressure tensor (for an isotropic gas
Note, the two terms in the above expression describe pressure):
the two “intrinsic” ways in which the mass in the ele-
 
p 0 0
mental volume can change. We collect them as  
P = 0 p 0 = pδij (1.11)
 
D ∂  
= +v·∇ (1.5)
Dt ∂t 0 0 p
and with this, the continuity equation becomes and note that the divergance of a tensor is found (in
Cartesian) by
Dρ
+ ρ∇ · v = 0 (1.6) ∂Pij
Dt (∇ · P)i = (1.12)
∂xj
C. Momentum Conservation: Euler’s Equation
with the Einstein summation convention assumed.
Consider again our surface A, enclosing R volume V . With this notation, the Euler equation in the case of
The momentum within this surface is V ρvdV . The no body force can be written,
net rate of change of this quantity again must reflect in- ∂
trinsic (∂/∂t 6= 0) variation and advection (flow across (ρv) + ∇ · (ρvv + P) = 0 (1.13)
the surface). Thus, we write the net rate of change of ∂t
momentum as This, and (1.4), are conservative forms of the basic
equations.
∂
Z Z
(ρv)dV + (ρv)v · n̂dA
V ∂t A
(1.7) 1
In general, a “vector product” tensor ab expands out as
∂
Z Z
= (ρv)dV + ∇ · (ρvv)dV
V ∂t
2 3
V a 1 b1 a 1 b2 a 1 b3
ab = 4a2 b1 a2 b2 a2 b3 5
In the second expression, we have used Gauss’s law for a 3 b1 a 3 b2 a 3 b3
tensors (noting that ρvv is a second-rank tensor).
 3

1. CONTROL VOLUMES But now: the right hand side of (1.15) is normal to both
the local flow field (that is normal to streamlines) and
We argued above that equating (1.7) and (1.7) will give to the local vorticity ~ω. Thus, we have one form of
an integral form of the momentum balance equation. Bernoulli’s relation: in inviscid, steady flow, the term
This is true if the volume V and surface A are fixed (not in brackets has zero gradient in the direction of the local
themselves moving). It can also be useful to consider velocity field. Thus, we have one version of Bernoulli’s
a moving volume (such as in a rocket problem). To law:
specify, let V ∗ (t) be the instantaneous control volume, 1 2
Z
dp
and A∗ (t) be its area. Let b be the local velocity of the v + + Φg = constant along streamline
boundary. Our integral equations become 2 ρ
(1.17)
d
Z Z Further, in an adiabatic gas, p ∝ ργ if γ is the adi-
ρdV + ρ(v − b) · n̂dA = 0 atabic index (the ratio of specific heats). The second
dt V ∗ (t) A∗ (t) term simplifies, so that Bernoulli’s relation for an invis-
cid adiabatic gas is
and
1 2 γ p
d
Z Z v + + Φg = constant along streamline
ρvdV + ρv(v − b) · n̂dA 2 γ−1ρ
dt V ∗ (t) A∗ (t) (1.18)
Z Z Alternatively, in an incompressible fluid, ρ is constant,
= ρgdV − pn̂dA and the second term in (1.15) becomes simply p/ρ.
V ∗ (t) A∗ (t)
Thus, for an incompressible fluid, Bernoulli’s relation
Exercise for the student: can you derive or justify these is
relations? 1 2 p
v + + Φg = constant along streamline (1.19)
2 ρ
2. BERNOULLI ’ S RELATION
D. Work in a Rotating Frame
It might be comforting to prove that we can extract
Bernoulli’s relationship from what we have so far. Euler’s equation effectively expresses force balance. It
must therefore be modified in a rotating system, where
Start with Euler’s equation, in the form (1.10). But we expect Coriolis and centrifugal forces. Recall your
now, note two useful facts. The first is that if the fluid is basic mechanics. We want to transform vectors from
barotropic – that is if p = p(ρ) only (as in an adiabatic our (intertial) frame to a frame rotating at Ω . Let r be
gas), we have the usual vector from the coordinate origin to the obser-
1
Z
dp vation point (polar coordinates), and let R be a vector
∇p = ∇ (1.14) from the rotation axis, perpendicular, to the observation
ρ ρ
point (cylindrical coordinates). The time derivative of a
(this can be verified using the chain rule; take p = general vector, P, transforms as
F (ρ), F being some function, and go from there).
   
dP dP
Thus, this term is a perfect differential. The second use- = +Ω × P
dt i dt r
ful fact is that
  Thus, velocities and accelerations transform as
1 2
v · ∇v = −v × ω + ∇ v (1.15) vi = vr + Ω × r ;
2

(this is easiest to verify by expanding out in Cartesian Ω × vr + Ω × (Ω

ai = ar + 2Ω Ω × r)
coordinates). Thus, this term is also a perfect differen-
Ω × vr − Ω2 R
= ar + 2Ω
tial. The first term on the right hand side is written in
terms of ω = ∇ × v, the local vorticity (more in chap- From this, we can write the incompressible Euler equa-
ter 4 on this). Specify the force to gravity, which can be tion in a rotating frame (now dropping the r subscripts):
expressed in terms of a potential: g = ∇Φg . If we then
Dv 1
consider steady flow, we can rewrite (1.10) as = − ∇p + g + Ω2 R − 2Ω Ω×v (1.20)
Dt ρ
 
1 2 dp
Z
The two new terms are the centrifugal and Coriolis
∇ v + + Φg = v × ω (1.16)
2 ρ forces.
 4

E. Dimensional Analysis Viscosity ν is addressed and defined in Chapter 3. This

measures the ratio of the inertial term to the viscous
Much of the art of fluid mechanics is knowing which term in the force equation. Viscosity is important at low
terms in the equations can be thrown away. We are Re; flows are laminar, i.e., we can ignore viscosity, at
guided in this by several dimensionless numbers. In this higher Re. Interestingly, the transition to turbulence oc-
section, we need “characteristic” values of the velocity curs at very high Re (even though viscosity is critically
V , the length scale L, the gas pressure p, the density important in turbulence; this is discussed later).
ρ, and the magnetic field B. In addition, transport co-
• the Mach number is
effients come in: ν is the viscosity, η the magnetic dif-
fusivity, and κ is the thermal conductivity (these quan- v
M= (1.25)
tities will be introduced in more detail later, as they are cs
needed). if cs is the sound speed (shown in §3 to be given by
As an example of why we consider these dimension- c2s = ∂p/∂ρ). An important limit is the incompress-
less numbers, return to the Euler equation, but with the ible, or Boussinesq limit, namely that the fluid density
viscous term added (just now, take my word for it, we’ll is constant (from 1.4, this is equivalent to ∇ · v = 0);
derive it in Chapter 3 and give it another name): this is justified for flows in which M ≪ 1.
Dv The most important scaling numbers for MHD are
ρ = −∇p + ρν∇2 v (1.21) • the plasma beta is
Dt
The last term, a second spatial derivative in v, is the 8πp
β= (1.26)
viscous stress term. Consider the magnitude of each of B2
the three terms: and gives the ratio of the gas pressure to the magnetic
pressure (pB = B 2 /8π).2 In low-beta
 2  2
plasmas, the
   
V V c V
O ρ ∼O ρ ; O ρ s ; O ρν 2 magnetic field dominates the dynamics; in high-beta
t L L L
plasmas the gas pressure dominates and magnetic ef-
In writing these estimates, we have introduced the fects are small.
sound speed, c2s = ∂p/∂ρ (it will appear formally in • the Magnetic Reynolds number is
Chapter 7). In addition, we have assumed we can find
a characteristic velocity V and a characteristic length LV
Rm = (1.27)
L. Now, we can form the ratio of the first (the inertial η
term) to the second (the pressure gradient):
if η is the magnetic diffusivity (addressed in Chapter
Dv/Dt
 2
V 13). Flows with high Rm have strong coupling of the
∼O (1.22) B field to the flow; flows at low Rm have only a weak
∇p c2s
coupling.
and, the ratio of the first term to the third (the viscous • the Alfven Mach number is
force): v
MA = (1.28)

VL
 vA
Dv/Dt ∼ O (1.23)
ν if vA is the Alfven speed, a characteristic signal speed
in a magnetized
√ fluid (shown in chapter 13 to be given
Thus, the relative importance of the three terms in by vA = B/ 4πρ).
the Euler equation are determined by the dimensionless In addition, several other dimensionless numbers
quantities V 2 /c2s and vL/ν. These two ratios are fun- appear here and there. Some that we may meet are
damental in hydrodynamics, and have their own names.
A similar analysis of other terms in these and/or other • the Prandtl number is
equations, as they arise, leads to several other important ν
Pr = (1.29)
dimensionless numbers. I list several of these here. κ
The two most important for hydrodynamics are: and measures the radio of viscous diffusion to thermal
• the Reynolds number is diffusion.

VL
Re = (1.24) 2
The SIs expression is β = 2µo p/B 2 .
ν
 5

• the Rayleigh number is F. Appendix: When can we use hydrodynamics?

αg∆T L3 Finally, we need to consider when we can use the hy-

Ra = (1.30) drodynamic (or magnetohydrodynamic) model. Fluid
κν
mechanics may seem the obvious choice when we are
and measures the ratio of the bouyancy force to the sta- concerned with laboratory or terrestrial liquids; they
bilizing effects of diffusion. α is the thermal expansion are “clearly” well described by macroscopic quantities
coefficient. such as density, pressure, velocity. For gases, and par-
ticularly plasmas, however, the relevance of fluid me-
• the Ekman number is
chanics is slightly subtler. In particular, some astro-
ν physical gases and plasmas are very tenuous: when can
E= (1.31) we justify a fluid approach, and when must we consider
ΩL2
the dynamics of individual particles?
for a system rotating at angular speed Ω. This measures The general answer is, HD or MHD can be
the ratio of viscous to Coriolis terms in the force equa- used when both the interparticle distance (equal to
tion. (density)−1/3 ) and the mean free path of fluid particles
• the Rossby number is (molecules, atoms, ions) is small compared to any char-
acteristic scale of the system. That is, these distances
V must be small compared to the overall scale of the sys-
Ro = (1.32) tem, and must also be small compared to the scale of
ΩL
any gradients (ρ/∇ρ, for instance, is the density gradi-
in a rotating system. It measures the ratio of inertial to ent scale).
Coriolos terms in the force equation.
• the Hartmann number is 1. HARD SPHERE COLLISIONS

Thus, we need to look at the mean free path in fluids,

BL
Ha = √ (1.33) gases and plasmas. To start, we recall the basics of
4πρνη hard-sphere collisions. If a “gas” of billiard balls, say,
has a number density n and each particle has a random
and measures the ratio of magnetic forces to diffusive velocity v and a radius a, we define the collision cross
forces in a MHD fluid. Some authors omit the 4π. section,
• the Magnetic Prandtl number is
σ = πa2 (1.35)
ν
Pm = (1.34)
η From this we find the mean free path (the average dis-
tance between collisions),
and measures the ratio of viscous diffusion to magnetic
diffusion. 1
λ≃ (1.36)
nσ
and the mean time between collisions,
References 1
Much of this discussion is “just from me” – i.e., tcoll ≃ . (1.37)
nσv
I’ve pulled the topics together “out of my head”. Good
places to go for more discussion include Kundu (who’s This last can be inverted to describe the collision rate
usually quite mathematical); Tritton (who’s much bet- per particle, t−1
coll ≃ nσv.
ter with words and concepts); or Faber (our text). For hard spheres this analysis is straightforward, of
For the plasma/Coulomb scattering discussion, course; they will not interact unless there is a direct
Spitzer’s Physics of Fully Ionized Gases is the classic “hit”, and the geometrical cross section is the relevant
reference. one to describe energy exchange. Neutral atoms and
molecules behave similarly, in that they need a very
close hit; their cross sections can be calcalated from ba-
sic physics. Typical atomic cross sections ∼ 10−14 cm2 .
 6

2. PLASMAS : THE COULOMB CROSS SECTION (clearly, we cannnot integrate from bmin = 0 to bmax =
∞, since the integral in (1.38) would diverge). bmin is
There is another type of encounter which is important in usually taken to be the distance corresponding to maxi-
plasmas: a long-range encounter between two particles mum energy transfer,
which feel a 1/r 2 Coulomb force. The particles never
directly collide with each other; but the long-range scat- e2
tering allows exchange of energy and momentum, and bmin ≃ (1.41)
2me v 2
thus makes the system act like a fluid.
• Start with a single encounter, in which particle A bmax is less straightforward. A common choice is the
(an electron, say) scatters on particle B (a proton, say; Debye shielding length (the scale over which an extra
with mp >> me , we can assume the proton stays at charge causes charge separation in a plasma):
rest. Let the incoming particle have velocity v and mass
1/2
me , and let in come in at impact parameter b. We can bmax ≃ λD = kB T /4πne2 (1.42)
solve this problem exactly, from classical mechanics, (kB is the Boltzmann constant, and T is the temper-
and find the deflection angle, θ, and the resultant veloc- ature). Thus, the best choice of ln Λ clearly depends
ity and momentum changes, ∆p = m∆v. Here, we on the exact situation one is considering. Luckily, for
will approximate this analysis. our purposes, this is only a logarithmic uncertainty, and
R • The net impulse on the electron will be ∆p = will not be critical for most of our calculations. The
F(t)dt, integrated over the collision. Now, the force choices above, with typical astrophysical parameters,
is strong only when the two particles are close. Since give ln Λ ≃ 10 − 20, in almost any diffuse-matter set-
they are close for a period of time ∆t ≃ 2b/v, we can ting.
approximate F ≃ e2 /b2 and ∆p ≃ 2F b/v. (Since we • Numerically, for a thermal plasma with 21 me v 2 ≃
know the net deflection is perpendicular to the initial kB T , the Coulomb cross section becomes,
direction of motion, we can also drop the vector nota-
tion). This gives us the net energy gain per collision, ln Λ
σc ≃ 7 × 10−13 2 cm2 (1.43)
T4
(∆p)2 2e4
∆E = ≃
2me m e b2 v 2 where T4 = T /104 K; so that

We want to extend this analysis, to find the net rate of 3/2

T4
energy exchange with the plasma. But the collision rate tcoll ≃ 4 × 104 sec (1.44)
n ln Λ
of our electron, with particles at impact parameter b is,
(collisions/second) = 2πnbv db, we find the net energy and
exchange rate by integrating over all allowed b:
T42
λ ≃ 1 × 1012 cm . (1.45)
dE
Z bmax
2e4 4πe4 n bmax
  n ln Λ
= 2 2
2πbnvdb = ln
dt bmin me b v me v bmin • A useful short way to remember the Coulomb
(1.38) cross section is as follows. Similarly to the bmin es-
• Now, we want to express this in terms of a cross timate above, we can define an effective “radius”, aef f ,
section: by equating potential and kinetic energies:
dE E e2 1
= = nvσc E (1.39) = me v 2 (1.46)
dt tcoll aef f 2
This defines the Coulomb cross section, σc : and then, estimating σc = 2πa2ef f ln Λ. This recovers
2 the form of equation (1.40), and resembles the hard-
e2

σc = 8π ln Λ (1.40) sphere cross section, (1.35), “with a factor of ln Λ
me v 2 tacked on”. In extending this to other examples, as
we will do just below, the exact numerical factor that
if ln Λ = ln (bmax /bmin ) is defined as the Coulomb scales πa2ef f ln Λ cannot be recovered by this method
logarithm. of guessing; one would have to do a more formal anal-
• The Coulomb logarithm depends on the largest ysis to get the correct order-unity numerical factor for
and smallest impact parameters that are important each cross section.
 7

3. COLLISIONLESS PLASMAS

Finally, a brief comment on a third limit. We saw

above that Coulomb collisions can transfer energy and
momentum between charges in a plasma, thus “tying
the plasma together” and justifying our treating it in
the fluid limit. In many applications, however, the
Coulomb mean free path derived using (1.40) is large
compared to the size of the system. Does this then
mean that we must use a single-particle (kinetic) ap-
proach? Not necessarily. Such systems can and do act
like fluids; one example is the solar wind, which has a
very long Coulomb mean free path, and yet is observed
to carry shocks. (We will see later that shocks are me-
diated by particle-particle collisions; their thickness is
several mean free paths). How can this be? There are
two likely causes:
• Tangled magnetic fields. A charged particle in
a magnetic field undergoes gyromotion. It is thus con-
strained to move (nearly) along the field line; gyroradii
are typically quite small. If the magnetic field is tan-
gled on a scale small compared to the system size, the
system can be “tied togeter” by this effect, justifying a
fluid approximation.
• Plasma Microinstabilites. A plasma is subject
to a wealth of microinstabilities – in which free energy
(such as that of streaming charges) is converted to wave
energy in the plasma. The plasma waves which are ex-
cited involve fluctuating electric and magnetic fields,
which in their turn scatter the plasma charges. This ef-
fect can also “tie the system together”; even a low-level
wave energy density can lead to a short mean free path
for particle-wave scattering.