85

16. MHD EFFECTS IN FLUID FLOWS Integrate this over a finite box (without losses through
the sides):
We now have the basics: we know, in principle, how
d 1 2 1
Z Z
B fields interact with fluid flows. They exert magnetic ρv dV = − j 2 dV (16.3)
tension and pressure on the fluid; the fluid, in return, dt 2 σ
modifies and/or determines the local B field, through Thus, we find that the energy lost from the flow, due
inductive effects (that is, the v × B EMF creates local to the Lorentz force, is just balanced by ohmic losses –
currents). In this chapter we visit some applications. which of course appear as heat in the flow: (Compare
the more general energy equation, at the end of Chapter
A. Magnetic damping and stirring 13 – this is a simple application of the same physics.)
Magnetic fields can have some unexpected effects on 2. M AGNETIC STIRRING .
fluid flow – if you set things up right, they can either
accelerate or decelerate the flow. Here are a couple of Magnetic fields can also induce motion in a fluid. One
brief examples; we may discuss them more in class. simple example is a magnetic field rotated (at some an-
gular frequency Ω) around a fluid that is initially sta-
1. M AGNETIC DAMPING . tionary (think about external electromagnets rotating
around the cylinder).
Magnetic fields can decelerate a flow. For a concrete
example, let’s say you try to drive a flow (maybe a jet)
across a pre-existing B field. Qualitataively, we know Ω
that the flow will initially try to “stretch” the field lines;
the resultant magnetic tension will resist – and decel-
erate – the flow. More quantitatively, the EMF caused
by the flow, v × B, will generate a transverse current j; B
this will give a “backwards” j × B force which will try
to decelerate the flow.

B Figure 16.2. Illustrating magnetic damping. A fluid sits

in a long cylinder (seen end-on). A magnetic field B
pervades the fluid, and is rotated at some Ω.
v
How does this cause motion? Think about this in a
j <−> v x B frame in which the B field is at rest; in this frame the
fluid initially has an azimuthal motion vphi = Ωr. But
this gives us an EMF,
Figure 16.1. Illustrating magnetic damping. Sent a jet,
velocity v, into a transverse B field. The resultant EMF, v ΩrB
v × B, will drive a current: which way will the Lorentz ×B → (16.4)
c c
force act? Will it accelerate or retard the flow?
This induces a current, as usual (in which direction??),
which leads to a Lorentz force
The energetics are simple. From Ohm’s law, j = j σ
σ (E + (v/c) × B). If the flow is set up with no free- × B → 2 ΩrB 2 (16.5)
c c
charge E (for instance by keeping all boundaries at the
Think through the directions: this force will accelerate
same potential), then we find that the rate of work done
the flow, in the direction of motion of the rotating B.
on the flow, by the Lorentz force, is
B. Channel flow with MHD: Hartmann flow
1 1 j2
v · (j × B) = − j · (v × B) = − (16.1) Here’s a classic example of MHD effects on a fluid
c c σ
flow; think of a low-Rm , laboratory-type channel flow.
Connect this to the force equation (which we dot with We already saw pure-HD versions of this, in Chapter 9.
v to get an energy equation): Here, we revisit this situation with MHD effects added.
We’ll find that MHD effects can either drive the flow (a
D 1 “pump”, or drive a transverse current (a “generator”),
v· = v · (j × B) − ∇ · (pv) (16.2) depending on the cionfiguration of the experiment.
Dt c
 86

1. T HE BASIC SETUP We get

The setup is: steady, viscous, incompressible, planar ∂b

Bo v + η=0
flow between plates, and across a B field. As in the ∂z
case of Poiseuille flow, which we saw in Chapter 3, the ∂ 2 v Bo ∂b ∂p
ρν 2 + − =0 (16.7)
flow here is maintained against friction by a pressure ∂z 4π∂z ∂x
difference applied across the ends of the channel. The b2

∂
presence of a magnetic field will, however, change the p+ =0
∂z 8π
nature of the solution.
The basic equations are Play algebra on these, eliminate b, and find an equation
for v:
dp d2 v Bo2
∇×E=0 − ρν 2 − v=0 (16.8)
dx dz 4πcη
∇·v =0 Because dp/dx is constant, we have a differential equa-
(16.6) tion which can be solved for v(z). If we apply no-slip
η∇ × B = E + v × B boundary conditions, the solution can be written
1
∇p − ρν∇2 v =
 
(∇ × B) × B Haz
4π v(z) = A cosh Ha − cosh (16.9)
l
The first one is from Maxwell, with ∂B/∂t = 0; the where A is an integration constant (which can be de-
third is from the steady-state induction equation (think termined from the problem), and the Hartmann number
of “uncurling” ∂B/∂t = 0. The other two should be fa- has been defined as
miliar to you. Once again, ν is the kinematic viscosity, lBo
η = c2 /4πσ is the diffusivity, and σ is the conductivity. Ha = (16.10)
(4πρνη)1/2

111111111111111111111111
000000000000000000000000 How does this compare to the non-MHD channel
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111 flows in Chapter 2? Well, the velocity at the center of
000000000000000000000000
111111111111111111111111
dp/dx
b z the flow is vo = v(z = 0) = A(cosh Ha−1). The form
(16.9) converts to the usual parabolic velocity profile,
B
v
x as B → 0 (that is as Ha → 0:
z2
 
111111111111111111111111
000000000000000000000000
000000000000000000000000
111111111111111111111111 v(z) → vo 1 − 2 (16.11)
000000000000000000000000
111111111111111111111111
000000000000000000000000
111111111111111111111111 l
Figure 16.3. The geometry for Hartmann flow: a viscous
conducting flow between parallel plates with a transverse B, (Compare the channel-flow solutions in chapter 2). Al-
and also a component b k v. Following Somov figure 15.4 ternatively, as Ha gets large, the solution flattens:
h i
v(z) → vo 1 − e−Ha(l−z)/l (16.12)
Figure 16.3 shows our setup. Let the plates lie at
z = ±l, with the flow along x̂. Things only vary with This is a much flatter velocity profile, with an exponen-
z (so the driving ∇p = dp/dx) is constant along x, as tial boundary layer close to the walls.
before). The imposed field is B = (0, 0, Bo ). We look
Note that everything else you might want for the so-
for induced fields b = (b(z), 0, ) and v = (v(z), 0, 0).
lution – current, b, etc – can be found from the basic
In general, we allow a constant transverse electric field,
equations once you have the v(z) solution.
E = Eo ŷ; but (you should be able to convince yourself
that) if we hold the boundaries at a fixed potential, we 3. AN MHD GENERATOR OR AN MHD PUMP ?
will have E = 0 inside the flow.
What determines the on-axis flow, vo ? Combine (16.9)
2. THE SOLUTION with Maxwell, to find the current density in the duct:
 2
To proceed, ignore any E field (for reasons given jy Bo ∂p Ha Haz
= − ρνA cosh (16.13)
above), and specify the system (16.6) to one dimension. c ∂x l l
 87

R
This integrates to give the total current, I = jy dz:
v
  H o
IBo ∂p Ha Haz
= 2l − 2ρνA sinh (16.14)
c ∂x l l z

B
H
(Note, this latter is an implicit solution for the velocity y

amplitude A). 0000000000000000000

1111111111111111111 x
1111111111111111111
0000000000000000000
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
Thus, the Hartmann flow speed depends on the ap- H
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
plied pressure gradient and the Lorentz force. This al- 0000000000000000000
1111111111111111111 v
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
1
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
lows two regimes of operation. (1) If the flow is main- 0000000000000000000
1111111111111111111
tained by an external pressure gradient, the duct oper-
Figure 16.4. Two plasma slabs, in relative motion. The
ates as an MHD generator, driving the transverse cur- top slab moves at vo , and is tied by B to the bottom slab.
rent I (this is useful if you somehow complete the cir- The slabs each have height H, which is also the inter-slab
cuit external to the duct..). (2) Alternatively, one can spacing. The slabs are infinite in the y direction, and have
apply an external EM driver to create the transverse cur- finite extent Lx in the x direction. Following Cravens
rent I; interaction of this current with Bo gives rise to Figure 8.19.
a Lorentz force that makes the plasma move along the
duct, that is creates an MHD pump. This may be dis- equations, for the top and bottom slabs:
cussed further in class.
Dvo dp By Bo
C. Magnetic Coupling of Two Plasma Slabs. ρo + =−
Dt dy 4πH
(16.17)
This is a quite different application, derived from space Dv1 By Bo
ρ1 = − ρ1 νcoll v1
plasmas... think of the earth’s ionosphere (which Dt 4πH
is partly ionized, partly neutral) and magnetosphere The first equation describes the “drivers”: either
(which is fully ionized). Dvo /Dt or dp/dy must be finite, in order to drive the
Picture: two plasma slabs, the top fully ionized, the system. The second equation describes the response of
bottom an ion-neutral mix, separated by some lower- the lower slab; clearly a steady state is possible there if
density plasma. A vertical magnetic field threads both. By Bo /4πH ≃ ρ1 νcoll v1 – in fact this will determine
Now let the top layer move at some vo ; if it is close to v1 , the velocity of the lower slab.
ideal, flux freezing will pull the field lines along, and But: what determines By , i.e. the curvature (and
thus exert a transverse force on the lower layer. The tension) of the field lines? If we stick to the pure-MHD
system is governed, of course, by the momentum equa- approach, we answer this using the induction equation
tion: (13.7). That is,
Dv
ρ = −∇p + j × B − ρνcoll v (16.15) ∇ × (v × B) + η∇2 B = 0
Dt
vo Bo By
We allow a ∇p term, and have added a “friction” force, +η 2 =0 (16.18)
ρνcoll v; assuming some collisional coupling term νcoll . H H
In addition we anticipate a Lorentz force, j × B. where the second equation is again from scal-
We can analyze the system in two ways. ing/dimensional analysis. We can thus estimate By ≃
Bo Hvo /η; and use this in (16.17) to solve the full sys-
1. PURE MHD tem. NOTE that we must still pick the appropriate resis-
tivity η for the bottom slab – Cravens argues for cross-
The spirit here is, ignore currents and E fields; use field Pederson conductivity, using the same ion-neutral
Maxwell to convert j to B. We thus write collision rate νcoll that provides the frictional force on
the lower slab.
Bo2 1
j × B ≃ −∇ + (B · ∇)B (16.16)
8π 4π 2. CURRENT- BASED APPROACH

But now, we ignore gradients in the magnetic pres- The above is a perfectly good solution...but Cravens
sure (they are small), and estimate the second term as notes that simple dimensional analysis may not be good
(B · ∇)B ≃ (By Bo /H)ŷ. We thus have two force enough when By is small (and presumably numerical
 88

methods may have trouble). We can also gain insight j⊥o H = j⊥1 H, or j⊥0 = j⊥1 (because we’ve cho-
by repeating the analysis in terms of the currents and sen the two slabs to have the same thickness). We can
electric fields induced by the motion. We expect two therefore combine the results above, as
cross-field currents, Pederson and Hall... The Pederson  
current is the interesting one.1 The Lorentz force in the 1 Dvo dp
v1 = − ρo + (16.21)
upper slab acts to decelerate it; that in the lower slab ρ1 νcoll Dt dy
accelerates the slab:
which allows, again, the possibility of the two drivers,
Dvo /Dt and dp/dy.
top : j × B ≃ −j⊥o Bo ŷ
To finish this, we still need to find j⊥o . Consider
bottom : j × B ≃ j⊥1 Bo ŷ (16.19) the top slab: if it is collisionless, we expect E =
−v × Bo = −vo Bo x̂ in the top slab. But also, be-
cause this is a steady-state, one-dimensional problem,
The two momentum equations become we have ∇ × E = 0, thus E must be the same in the
bottom slab (as well as in the intervening space). It fol-
Dvo dp lows that the Bo lines “map” the field from one slab
ρo + = −j⊥o Bo
Dt dy to the other: the magnetic field lines are equipotentials.
(16.20)
Dv1 We can, therefore, use Ohm’s law (for the Pederson cur-
ρ1 = j⊥1 Bo − ρ1 νcoll v1
Dt rent) to find j⊥o = j⊥1 = σ⊥ vo Bo ..thus finishing the
problem.
We can again find an equilibrium solution for the lower
slab, which has v1 = j⊥1 Bo /ρ1 νcoll . 3. ENERGETICS AND EQUIVALENT CIRCUIT

Finally, consider the energetics of this example. We can

think of j · E as a “local” source of EM energy. In this
J
v J
_
example, jo ·E > 0 while j1 ·E < 0. Thus, we can think
0
of the top slab as an MHD generator, which produces
z
J ||
z energy that is dissipated in the lower slab, the resistor.
y B Β
J
|| x
The powerRper unit length generated (top) or dissipated
000000000000000000
111111111111111111
111111111111111111
000000000000000000
000000000000000000
111111111111111111
v
1
000000000000000000
111111111111111111
111111111111111111
000000000000000000
000000000000000000
111111111111111111
(lower) is j · Edxdz ≃ Lx Hvo Bo j⊥o . This is also,
000000000000000000
111111111111111111
000000000000000000
111111111111111111 000000000000000000
111111111111111111
000000000000000000
111111111111111111
000000000000000000
111111111111111111
J_
000000000000000000
111111111111111111
000000000000000000
111111111111111111
000000000000000000
111111111111111111
J_
000000000000000000
111111111111111111
000000000000000000
111111111111111111
nicely, the same as the powerRtransferred from top to
000000000000000000
111111111111111111 000000000000000000
111111111111111111
000000000000000000
111111111111111111
000000000000000000
111111111111111111
000000000000000000
111111111111111111
000000000000000000
111111111111111111
000000000000000000
111111111111111111
000000000000000000
111111111111111111
bottom by the Poynting flux: (E × B)4π −1 dxdy ≃
000000000000000000
111111111111111111 000000000000000000
111111111111111111
vo Bp By Lx /4π (check: you can verify this is the same
Figure 16.4. Two views of the problem, emphasizing the result). We can also describe this problem in terms
induced currents. Left, same view as in the first figure; of the total circult and a global resistance...let E ≃
induced j × B Lorentz forces as shown. The top slab carries vo Bo Lx be the EMF generated (volts); the resistance
j⊥ out of the page, the bottom has j⊥ into the page. The of the lower slab is R = Lx /σ⊥ HLy (ohms); and the
induced E is into the page throughout. Right, the same
situation, but seen along the y axis. The complete induced current I = /R (amps), as it should be.
current loop can be seen, with j⊥ within the slabs (not
labelled with subscripts due to limitations of Xfig), and
field-aligned jk connecting the slabs. Following Cravens
Figure 8.19.
References
I’ve taken the magnetic damping/stirring material
Now: how do we find the current densities? First,
mostly from Davidson. My Hartmann flow discussion
consider current conservation, ∇ · j = 0 in a steady
follows Woods & Somov. The plasma-coupling discus-
state. The Pederson current flows along x̂, and the
sion follows Cravens.
slab is finite in that direction. Thus, the cross-field
currents in the plasma slabs must be connected by
field-aligned currents inbetween the slabs, as shown in
the figure. In addition, currrent conservation requires

1
the Hall current is there, in principle infinite in extent, so without
gradients, and not interesting for this problem.