16

3. POTENTIAL FLOW which becomes, explicitly,

In the first chapter we wrote down fairly general (partial ∂ψ ∂ψ
differential) equations describing fluid flow. We now vx = ; vy = − (3.5)
∂y ∂x
must look for solutions. In this chapter we consider
one of the traditional methods, potential flow, which de- (Why is this possible? From 3.4) it’s clear that ∇ · v =
scribes incompressible flows in which the velocity field 0. Thus, this works for incompressible flows.) The
has no curl. stream function has several useful properties.
• It’s related to the velocity potential; we can get
A. Setups and Assumptions one from the other:
Our first set of solutions consider flows with no vor- ∂φ ∂ψ ∂φ ∂ψ
ticity: ω = ∇ × v = 0. In addition, we will gen- vx = = ; vy = =− (3.6)
∂x ∂y ∂y ∂x
erally assume the flows are incompressible. Why are
such restrictive solutions interesting? First, we can find • ψ also satisfies Laplace’s equation; we can solve
analytic solutions: so let’s see where they lead. Sec- for ψ rather than φ if we wish.
ond, these solutions have some relation to reality . . . • Streamlines are tangent to the local velocity v.
because vorticity turns out to be a (nearly) conserved A streamline obeys
quantity, and hard to generate away from boundaries:
we will prove this later. Thus, flows which start out vy dx = vx dy (3.7)
irrotational tend to stay this way.
(compare 1.1). Thus, the differential of ψ along
1. VELOCITY POTENTIAL some path is
For irrotational flows, the fact that ∇ × v = 0 allows ∂ψ ∂ψ
the velocity to be written as the gradient of a scalar po- dψ = dx + dy (3.8)
∂x ∂y
tential:
Combining (3.7) and (3.8) shows that dψ = 0
v = ∇φ (3.1)
along a streamline. We can therefore use the func-
Explicitly, this is tion ψ to label streamlines – as in the examples.
• Finally, using (3.6), it follows directly that
∂φ ∂φ ∇φ · ∇ψ = 0, so that streamlines and equipo-
vx = ; vy = (3.2)
∂x ∂y tential lines cross each other at right angles. This
again says that v is tangential to the contours of
Further, when the flow is incompressible – which is the constant ψ – so the value of ψ provides a label for
case in all examples we will see — we have ∇ · v = 0, the streamlines in the flow.
so that
3. OTHER COORDINATES
∇2 φ = 0 (3.3)
As long as the problem is two dimensional and incom-
That is, the velocity potential satisfies Laplace’s equa- pressible, we can define φ and ψ functions.
tion. But this is great if we are traditional analytic Plane Polar Coordinates
physicists: we can use all the (glorious) solution meth- The velocity is connected to the potential by
ods worked out in electrostatics. This will be our gen-
eral approach – picking examples where we can find so- ∂φ 1 ∂φ
lutions to Laplace’s equation, based on what instinct we vr = ; vθ = (3.9)
∂r r ∂θ
have from analytic E&M. Once we have found φ(x, y),
the velocity solution comes immediately from (3.1). The incompressibility condition is
2. STREAM FUNCTION 1 ∂ 1 ∂vθ
(rvr ) + =0 (3.10)
If the flow is two-dimensional, we can also define an- r ∂r r ∂θ
other useful function. This is the stream function, ψ. From this we can define a stream function:
Define that
1 ∂ψ ∂ψ
v = −∇ × (ψẑ) (3.4) vr = ; vθ = − (3.11)
r ∂θ ∂r
 17

Cylindrical Coordinates, Axisymmetric 1. SOURCES AND SINKS , FLOW PAST HALF BODY
The velocity is connected to the potential by
We can combine these two building blocks (and work
∂φ ∂φ in cylindrical):
vr = ; vz = (3.12)
∂r ∂z
The incompressibility condition is ψ = U r sin θ

∂vz 1 ∂ C
+ (rur ) = 0 (3.13) φ = U r cos θ + ln r
∂z r ∂r 2π (3.21)
C
From this we can define a stream functionn: vr = U cos θ +
2πr
1 ∂ψ 1 ∂ψ
vz = ; vr = − (3.14) vθ = −U sin θ
r ∂r r ∂z
Spherical Polar Coordinates, Axisymmetric to get what is called flow past a half body. We see why,
The velocity is connected to the potential by by noting that the corresponding stream function is

∂φ 1 ∂φ C
vr = ; vθ = (3.15) ψ = U r sin θ + θ (3.22)
∂r r ∂θ 2π
The incompressibility condition is
1 ∂ 1 ∂
r 2 vr +


(vθ sin θ) = 0 (3.16)
r ∂r sin θ ∂θ
From this we get a stream function:
1 ∂ψ 1 ∂ψ
vr = ; vθ = − (3.17)
r 2 sin θ ∂θ r sin θ ∂r
B. Two-dimensional Planar Problems
Consider Laplace’s equation in two dimensions, in
plane polar coordinates (r, θ). Figure 3.1. Potential flow past a two-dimensional “half
body” (represented mathematically by a source function).
1 ∂2φ
 
2 1 ∂ ∂φ The boundary streamline is labelled by ψ = m/2 (that’s
∇ φ= r + 2 2 =0 C/2 in our notation); the shape of this streamline defines
r ∂r ∂r r ∂θ
the shape of the half body. S shows the stagnation point in
We know the possible general solutions: front. From Kundu figure 6.9

φ = constant ; φ ∝ r n cos(nθ)
φ ∝ ln r ; The flow clearly has a stagnation point: if it’s at po-
(3.18) sition (a, π), then v(a, π) = 0 requires a = C/2πU .
And, of course, any superposition of these is a solution. Now, the value of the stream function passing through
So, we proceed by finding useful superpositions . . . this point – its label – is
One important building block will be a simple
source or sink: C C
ψs = U a sin π + π=
2π 2
Cθ C
ψ= ; φ= ln r
2π 2π and the equation of the streamline passing through this
(3.19)
C point is found by taking ψ(r, θ) = ψs = C/2:
vr = ; vθ = 0
2πr
C C
This describes radial flow away from a source (or in- U r sin θ + θ= (3.23)
2π 2
wards to a sink if C < 0).
Another basic building block is a steady flow at U This stagnation streamline separates the flow into an
is represented in Cartesian by upstream region (v → U x̂ at ∞), and the region in-
side the streamline, a region with a smooth nose, called
φ = Ux ; v = ∇φ = U x̂ (3.20) a half body. We can, therfore, put a physical body with
 18

this shape into a flow, and keep the same external solu-
tion.
Once we have the velocity solution, we can find the
drag on the half body – the net force the fluid exerts on
the body (or vice versa). As long as we ignore friction
(i.e., viscosity), the net force on the body is just the
(vector) integral of the pressure over the surface of the
body. That is, for this simple case, we expect the drag
force to be
Z
FD = − pn̂dS (3.24)
S

if n̂ is the normal to the surface S, and the integral is

taken over the surface of the body. Bernoulli‘s relation Figure 3.2. A doublet: a source and a sink which
approach each other, keeping the product of their strength C
for an incompressible fluid (1.19) can be written as and separation ǫ constant. The streamlines become circles,
1 1 whose centers lie on the y-axis and which are tangent to the
p + ρv 2 = p∞ + ρU 2 (3.25) x-axis at the origin. From Kundu figure 6.8
2 2
where we have evaluated the constant at “infinity”, that But, now, this shows that ψ(a, θ) = 0: the streamline
is far upstream. Work this out: the (normalized) excess ψ = 0 is a circular cylinder of radius a. Thus, flows
pressure, inside this cylinder have no influence on flows outside;
Cp = 2(p − p∞ )/ρU 2 we can again replace this doublet with a physical cylin-
der, and find the same flow outside.
turns out to be positive at the front of the body, zero at
θ = 113◦ , and negative along the sides. You can show
from this, by setting up the integrals, that the net pres-
sure, integrated over the surface of the body, is zero:
there is no drag in this system.
2. FLOW PAST A CYLINDER

In another example, we start with a doublet: a paired

source and sink, of equal strength C, separated by ǫ;
let them approach each other while the product of their
separation and their strength stays finite, Cǫ/π → µ.
The potential and stream function, from Kundu, are Figure 3.3. Potential flow past a circular cylinder,
described by the ψ = 0 streamline. The inside of the
µx µy cylinder is represented mathematically by a doublet. From
φ= ; ψ=− (3.26)
x2 + y2 x2 + y2 Kundu figure 6.11
The streamlines, given by ψ = constant, are circles
Finally, we can again work out the drag. The pres-
whose centers lie on the y-axis, and which are tangent
sure difference
to the x-axis at the origin.
Now: combine this with a uniform stream flow, φ = 2(p − p∞ ) v2
Cp = = 1 − = 1 − 4 sin2 θ (3.28)
U x and ψ = U y: combining and putting in polars, with ρU 2 U2
a2 = µ/U , gives
so, again, is positive at the front and negative at the back

a2
 
a2
 . . . so that, once again, there is no net pressure drag on
φ=U r+ cos θ ; ψ = U r − sin θ this cylinder.
r r
a2 a2
   
C. Axisymmetric 3D Problems
vr = U 1 − 2 cos θ ; vθ = −U 1 + 2 sin θ
r r
Axisymmetric 3D problems are also amenable to this
(3.27) type of solution. Spherical polar coordinates are appro-
 19

priate here. The general solutions to Laplace’s equa- From this, we see that ψ = 0 for θ = 0, π at any r
tions for this geometry are (that is, the x-axis), and also for r = a = (2m/U )1/3
(any θ). Thus, again, we have a spherical surface, flows
φ =constant ; φ ∝ r n Pn (cos θ) ; inside of which have no effect on the outer stream flow.
(3.29) Again, then, we can work out the pressure/drag on
φ ∝ r −(n+1) Pn (cos θ) this system. The velocity components are
  a 3 
which are just your favorite (axisymmetric) Laplace’s vr =U 1 − cos θ ;
solutions in spherical polars. The velocity potentials r
  (3.36)
and stream functions obey 1  a 3
vθ = −U 1 + sin θ
2 r
∂φ 1 ∂ψ
vr = = 2 ; so that the pressure is
∂r r sin θ ∂θ (3.30)
1 ∂φ 1 ∂ψ 1 2

9

vθ = =− 2
p(a, θ) = p∞ + ρU 1 − sin θ (3.37)
r ∂θ r sin θ ∂r 2 4
We can run through some familiar-looking solutions. So, once again, this is symmetric in θ, leading to zero
drag on the sphere. Finally, let me refer you to Faber for
POINT SOURCE AND STREAM FLOW
a good discussion of real-world effects in this problem,
These are our two most basic. A point source has the §4.7.
potential and stream function,
A LINE SOURCE
Q Q
φ=− ; ψ=− cos θ (3.31) describes a mass source which is distributed in a line;
4πr 4π
R function k per length, so that its total
let it have a source
where Q (units: m3 /s) is the discharge rate of the source is Q = kdx.
source. A uniform stream flow has the potential and (r, θ)
stream function,
1
φ = U r cos θ ; ψ = U r 2 sin2 θ (3.32) r
2
The velocity solutions for these two basics are straight-
forward. It is worth noting that the sum of these two
gives the velocity field for an axisymmetric half body,
θ α
Q a
vr = U cos θ + ; vθ = −U sin θ (3.33) 0 ξ
4πr 2
Figure 3.4. The geometry of a line source, which extends
FLOW AROUND A SPHERE from ξ = 0 to ξ = a.

As with flow past a cylinder, this adds a uniform stream Each differential bit of the line source has potential
to a doublet which opposes the stream. and stream functions,
A Doublet, in this geometry, has the potential and kdξ kdξ
dφ = − ; dψ = − cos α (3.38)
stream function, 4πr 4π
m m if ξ is the running variable along the length of the
φ= cos θ ; ψ=− sin2 θ (3.34) source. We also let R = r sin θ be the ver-
r2 r
tical distance to the observation point, and r1 =
The velocity solutions for flow around a sphere are
1/2
r 2 + a2 − 2ra cos θ be the radial distance from
straightforward. We have, then ξ = a (the end of the source) to the observation point.
m The net stream function is then
φ = U r cos θ + cos θ ; Z a
k
Z a
k
r2 (3.35) ψ(r, θ) = dψ = − cos α(ξ)dξ = (r−r1 )
1 m 0 4π 0 4π
ψ = U r 2 sin2 θ − sin2 θ (3.39)
2 r
 20

P Q
where the last step is obtained by straightforward geo-
metrical identities and changes
R of the integration vari-
able. [Exercise: what is φ = dφ?]
1111111
0000000
0000000
1111111
0000000
1111111
0000000
1111111
D. d’Alembert’s “Paradox” 0000000
1111111
0000000
1111111 u(y)
U 0000000
1111111
o 0000000
1111111
0000000
1111111
0000000
1111111
0000000
1111111
We have now seen several examples of flow past a solid p 0000000
1111111
0000000
1111111
o 0000000
1111111
body, described by potential flow. In each case we
found that the net force exerted by the flow on the body
y
is zero. Thus, there is no drag in a potential flow around S R
a body. This result turns out to hold for any body, sym- x
metric or not. Figure 3.6. Momenteum balance of flow past an arbitrary
Does this sound wrong? We know experimentally body. The upstream flow has constant speed Uo and
that a drag force does exist when a body is immersed in pressure po . The downstream velocity may be a function of
y; in this sketch it is lower immediately behind the body (as
a fluid flow. This apparent contradiction, between the due to a wake). Note, the body need not be spherical, that
theory and experiment, became known as d’Alembert’s was just easy for me to draw... Following Kundu figure 4.7.
paradox. As you might suspect, the solution of the para-
dox is that the theory is incomplete (no, we don’t say
outflow through PS and QR is easy to calculate:
“wrong”). In particular, we have ignored two important
Z L
pieces of the physics here. One, is that these solutions PS
can have nonzero velocity at a solid surface: check Fig- Ṗ =− Uo ρUo dy = −LρUo2 ;
o
ure 3.1 or 3. This is naive: any real fluid has a finite, Z L (3.40)
if small, viscosity. But viscosity will force to a no-slip QR 2
Ṗ =ρ u(y) dy
boundary condition. This leads to a viscous boundary 0
layer, which will modify the solutions close to the sur- Now: there must be mass outflow, Ṁ , through the
face, and add tangential stress which contributes some sides, SR and PQ (this is necessary to conserve mass
drag. Two, the flow behind the body may very well be if u(y) < Uo , as shown). Mass conservation requires
turbulent – as this viscous boundary separates from the Z L
body and conects into the wake. We will see later that PQ SR
LρUo = Ṁ + Ṁ + ρ u(y)dy (3.41)
this also contributes to a drag force. Both of these ideas O
are illustrated in Figure 3.6.
But if the sides are far enough away from the rock that
the x velocity ≃ Uo there, then the rate of outflow of
x-momentum is
Z L
PQ SR
Ṗ + Ṗ = ρUo (Uo − u(y))dy (3.42)
0
Finally, we can collect all of this, to find the net drag on
the body: FD = Ṗ out = Ṗ P S + Ṗ QR + Ṗ P Q + Ṗ SR .
This gives,
Z L
out
Figure 3.5. Illustrating the nature of the flow field around FD = Ṗ =ρ u(y)(Uo − u(y))dy (3.43)
O
an arbitrarily shaped body. From Currie Figure 9.1.
And .. referring back to our simple example of potential
We can also carry out a general drag calculation, for flow past a cylinder, note from (3.27) that the down-
a finite body, that illustrates this argument. We could stream velocity field → U as r → ∞. That is, there
work out the pressure in the fluid, everywhere on the is no slow-down of the flow in this idealized case; and
surface of the body, and then integrate it over the sur- thus no drag. Drag comes from viscosity and boundary
face. Or we can try a different approach. effects which generate a wake behind the body.
Consider the region of fluid within the rectangle
SRQP; the net drag on the body within this region is just
the rate of change of momenteum flux: FD = Ṗout , i.e.
the net outflow of x-momentum from the region. The References: I’ve mostly followed Kundu here.