282 PA R T 2 Inviscid, Incompressible Flow

obtain the lift—is the essence of the circulation theory of lift in aerodynamics.
Its development at the turn of the twentieth century created a breakthrough in
aerodynamics. However, let us keep things in perspective. The circulation theory
of lift is an alternative way of thinking about the generation of lift on an aero-
dynamic body. Keep in mind that the true physical sources of aerodynamic force
on a body are the pressure and shear stress distributions exerted on the surface of
the body, as explained in Section 1.5. The Kutta-Joukowski theorem is simply an
alternative way of expressing the consequences of the surface pressure distribu-
tion; it is a mathematical expression that is consistent with the special tools we
have developed for the analysis of inviscid, incompressible flow. Indeed, recall
that Equation (3.140) was derived in Section 3.15 by integrating the pressure dis-
tribution over the surface. Therefore, it is not quite proper to say that circulation
“causes” lift. Rather, lift is “caused” by the net imbalance of the surface pres-
sure distribution, and circulation is simply a defined quantity determined from
the same pressures. The relation between the surface pressure distribution (which
produces lift L  ) and circulation is given by Equation (3.140). However, in the
theory of incompressible, potential flow, it is generally much easier to determine
the circulation around the body rather than calculate the detailed surface pressure
distribution. Therein lies the power of the circulation theory of lift.
Consequently, the theoretical analysis of lift on two-dimensional bodies in
incompressible, inviscid flow focuses on the calculation of the circulation about
the body. Once  is obtained, then the lift per unit span follows directly from
the Kutta-Joukowski theorem. As a result, in subsequent sections we constantly
address the question: How can we calculate the circulation for a given body in a
given incompressible, inviscid flow?

3.17 NONLIFTING FLOWS OVER ARBITRARY

BODIES: THE NUMERICAL SOURCE
PANEL METHOD
In this section, we return to the consideration of nonlifting flows. Recall that
we have already dealt with the nonlifting flows over a semi-infinite body and a
Rankine oval and both the nonlifting and the lifting flows over a circular cylinder.
For those cases, we added our elementary flows in certain ways and discovered
that the dividing streamlines turned out to fit the shapes of such special bodies.
However, this indirect method of starting with a given combination of elementary
flows and seeing what body shape comes out of it can hardly be used in a practical
sense for bodies of arbitrary shape. For example, consider the airfoil in Figure 3.37.
Do we know in advance the correct combination of elementary flows to synthesize
the flow over this specified body? The answer is no. Rather, what we want is a
direct method; that is, let us specify the shape of an arbitrary body and solve for the
distribution of singularities which, in combination with a uniform stream, produce
the flow over the given body. The purpose of this section is to present such a direct
method, limited for the present to nonlifting flows. We consider a numerical
method appropriate for solution on a high-speed digital computer. The technique
 CHAPTER 3 Fundamentals of Inviscid, Incompressible Flow 283

Figure 3.38 Source sheet.

is called the source panel method, which, since the late 1960s, has become a
standard aerodynamic tool in industry and most research laboratories. In fact, the
numerical solution of potential flows by both source and vortex panel techniques
has revolutionized the analysis of low-speed flows. We return to various numerical
panel techniques in Chapters 4 through 6. As a modern student of aerodynamics,
it is necessary for you to become familiar with the fundamentals of such panel
methods. The purpose of the present section is to introduce the basic ideas of the
source panel method, which is a technique for the numerical solution of nonlifting
flows over arbitrary bodies.
First, let us extend the concept of a source or sink introduced in Section 3.10.
In that section, we dealt with a single line source, as sketched in Figure 3.21.
Now imagine that we have an infinite number of such line sources side by side,
where the strength of each line source is infinitesimally small. These side-by-side
line sources form a source sheet, as shown in perspective in the upper left of
Figure 3.38. If we look along the series of line sources (looking along the z axis
in Figure 3.38), the source sheet will appear as sketched at the lower right of
Figure 3.38. Here, we are looking at an edge view of the sheet; the line sources
are all perpendicular to the page. Let s be the distance measured along the source
sheet in the edge view. Define λ = λ(s) to be the source strength per unit length
along s. [To keep things in perspective, recall from Section 3.10 that the strength
of a single line source  was defined as the volume flow rate per unit depth, that
is, per unit length in the z direction. Typical units for  are square meters per
second or square feet per second. In turn, the strength of a source sheet λ(s) is
the volume flow rate per unit depth (in the z direction) and per unit length (in
the s direction). Typical units for λ are meters per second or feet per second.]
Therefore, the strength of an infinitesimal portion ds of the sheet, as shown in
Figure 3.38, is λ ds. This small section of the source sheet can be treated as
a distinct source of strength λ ds. Now consider point P in the flow, located a
distance r from ds; the cartesian coordinates of P are (x, y). The small section
284 PA R T 2 Inviscid, Incompressible Flow

Figure 3.39 Superposition of a uniform flow and a source sheet on a body

of given shape, to produce the flow over the body.

of the source sheet of strength λ ds induces an infinitesimally small potential dφ

at point P. From Equation (3.67), dφ is given by
λ ds
dφ = ln r (3.141)
2π
The complete velocity potential at point P, induced by the entire source sheet
from a to b, is obtained by integrating Equation (3.141):
 b
λ ds
φ(x, y) = ln r (3.142)
a 2π
Note that, in general, λ(s) can change from positive to negative along the sheet;
that is, the “source” sheet is really a combination of line sources and line sinks.
Next, consider a given body of arbitrary shape in a flow with freestream
velocity V∞ , as shown in Figure 3.39. Let us cover the surface of the prescribed
body with a source sheet, where the strength λ(s) varies in such a fashion that the
combined action of the uniform flow and the source sheet makes the airfoil surface
a streamline of the flow. Our problem now becomes one of finding the appropriate
λ(s). The solution of this problem is carried out numerically, as follows.
Let us approximate the source sheet by a series of straight panels, as shown
in Figure 3.40. Moreover, let the source strength λ per unit length be constant
over a given panel, but allow it to vary from one panel to the next. That is,
if there are a total of n panels, the source panel strengths per unit length are
λ1 , λ2 , . . . , λ j . . . , λn . These panel strengths are unknown; the main thrust of the
panel technique is to solve for λ j , j = 1 to n, such that the body surface becomes
a streamline of the flow. This boundary condition is imposed numerically by
defining the midpoint of each panel to be a control point and by determining the
λ j ’s such that the normal component of the flow velocity is zero at each control
point. Let us now quantify this strategy.
Let P be a point located at (x, y) in the flow, and let r pj be the distance from
any point on the jth panel to P, as shown in Figure 3.40. The velocity potential
induced at P due to the jth panel φ j is, from Equation (3.142),

λj
φ j = ln r pj ds j (3.143)
2π j
 CHAPTER 3 Fundamentals of Inviscid, Incompressible Flow 285

Figure 3.40 Source panel distribution over the surface of a body of arbitrary shape.

In Equation (3.143), λ j is constant over the jth panel, and the integral is taken
over the jth panel only. In turn, the potential at P due to all the panels is Equa-
tion (3.143) summed over all the panels:
n n 
λj
φ(P) = φ j = ln r pj ds j (3.144)
j=1 j=1
2π j

In Equation (3.144), the distance r pj is given by


r pj = (x − x j )2 + (y − y j )2 (3.145)
where (x j , y j ) are coordinates along the surface of the jth panel. Since point P
is just an arbitrary point in the flow, let us put P at the control point of the ith
panel. Let the coordinates of this control point be given by (xi , yi ), as shown in
Figure 3.40. Then Equations (3.144) and (3.145) become
n 
λj
φ(xi , yi ) = ln ri j ds j (3.146)
j=1
2π j

and ri j = (xi − x j )2 + (yi − y j )2 (3.147)
Equation (3.146) is physically the contribution of all the panels to the potential
at the control point of the ith panel.
Recall that the boundary condition is applied at the control points; that is, the
normal component of the flow velocity is zero at the control points. To evaluate this
component, first consider the component of freestream velocity perpendicular to
the panel. Let ni be the unit vector normal to the ith panel, directed out of the body,
as shown in Figure 3.40. Also, note that the slope of the ith panel is (dy/d x)i . In
general, the freestream velocity will be at some incidence angle α to the x axis,
286 PA R T 2 Inviscid, Incompressible Flow

as shown in Figure 3.40. Therefore, inspection of the geometry of Figure 3.40

reveals that the component of V∞ normal to the ith panel is
V∞,n = V∞ · ni = V∞ cos βi (3.148)
where βi is the angle between V∞ and ni . Note that V∞,n is positive when directed
away from the body, and negative when directed toward the body.
The normal component of velocity induced at (xi , yi ) by the source panels
is, from Equation (3.146),
∂
Vn = [φ(xi , yi )] (3.149)
∂n i
where the derivative is taken in the direction of the outward unit normal vector, and
hence, again, Vn is positive when directed away from the body. When the derivative
in Equation (3.149) is carried out, ri j appears in the denominator. Consequently,
a singular point arises on the ith panel because when j = i, at the control point
itself ri j = 0. It can be shown that when j = i, the contribution to the deriva-
tive is simply λi /2. Hence, Equation (3.149) combined with Equation (3.146)
becomes

λi  λ j
n
∂
Vn = + (ln ri j ) ds j (3.150)
2 j=1
2π j ∂n i
( j=1)

In Equation (3.150), the first term λi /2 is the normal velocity induced at the ith
control point by the ith panel itself, and the summation is the normal velocity
induced at the ith control point by all the other panels.
The normal component of the flow velocity at the ith control point is the sum
of that due to the freestream [Equation (3.148)] and that due to the source panels
[Equation (3.150)]. The boundary condition states that this sum must be zero:
V∞,n + Vn = 0 (3.151)
Substituting Equations (3.148) and (3.150) into (3.151), we obtain

λi  λ j
n
∂
+ (ln ri j ) ds j + V∞ cos βi = 0 (3.152)
2 j=1
2π j ∂n i
( j=1)

Equation (3.152) is the crux of the source panel method. The values of the integrals
in Equation (3.152) depend simply on the panel geometry; they are not properties
of the flow. Let Ii, j be the value of this integral when the control point is on the
ith panel and the integral is over the jth panel. Then Equation (3.152) can be
written as
λi  λ j
n
+ Ii, j + V∞ cos βi = 0 (3.153)
2 j=1
2π
( j=1)

Equation (3.153) is a linear algebraic equation with n unknowns λ1 , λ2 , . . . , λn .

It represents the flow boundary condition evaluated at the control point of the ith
panel. Now apply the boundary condition to the control points of all the panels;
that is, in Equation (3.153), let i = 1, 2, . . . , n. The results will be a system of n
 CHAPTER 3 Fundamentals of Inviscid, Incompressible Flow 287

linear algebraic equations with n unknowns (λ1 , λ2 , . . . , λn ), which can be solved

simultaneously by conventional numerical methods.
Look what has happened! After solving the system of equations represented
by Equation (3.153) with i = 1, 2, . . . , n, we now have the distribution of source
panel strengths which, in an appropriate fashion, cause the body surface in Fig-
ure 3.40 to be a streamline of the flow. This approximation can be made more
accurate by increasing the number of panels, hence more closely representing the
source sheet of continuously varying strength λ(s) shown in Figure 3.39. Indeed,
the accuracy of the source panel method is amazingly good; a circular cylinder
can be accurately represented by as few as 8 panels, and most airfoil shapes, by
50 to 100 panels. (For an airfoil, it is desirable to cover the leading-edge region
with a number of small panels to represent accurately the rapid surface curvature
and to use larger panels over the relatively flat portions of the body. Note that, in
general, all the panels in Figure 3.40 can be different lengths.)
Once the λi ’s (i = 1, 2, . . . , n) are obtained, the velocity tangent to the sur-
face at each control point can be calculated as follows. Let s be the distance along
the body surface, measured positive from front to rear, as shown in Figure 3.40.
The component of freestream velocity tangent to the surface is
V∞,s = V∞ sin βi (3.154)
The tangential velocity Vs at the control point of the ith panel induced by all the
panels is obtained by differentiating Equation (3.146) with respect to s:

∂φ  λ j
n
∂
Vs = = (ln ri j ) ds j (3.155)
∂s j=1
2π j ∂s
[The tangential velocity on a flat source panel induced by the panel itself is zero;
hence, in Equation (3.155), the term corresponding to j = i is zero. This is
easily seen by intuition, because the panel can only emit volume flow from its
surface in a direction perpendicular to the panel itself.] The total surface velocity
at the ith control point Vi is the sum of the contribution from the freestream
[Equation (3.154)] and from the source panels [Equation (3.155)]:
n 
λj ∂
Vi = V∞,s + Vs = V∞ sin βi + (ln ri j ) ds j (3.156)
j=1
2π j ∂s
In turn, the pressure coefficient at the ith control point is obtained from Equa-
tion (3.38):
 
Vi 2
C p,i = 1 −
V∞
In this fashion, the source panel method gives the pressure distribution over the
surface of a nonlifting body of arbitrary shape.
When you carry out a source panel solution as described above, the accuracy
of your results can be tested as follows. Let S j be the length of the jth panel.
Recall that λ j is the strength of the jth panel per unit length. Hence, the strength
of the jth panel itself is λi S j . For a closed body, such as in Figure 3.40, the sum
288 PA R T 2 Inviscid, Incompressible Flow

of all the source and sink strengths must be zero, or else the body itself would be
adding or absorbing mass from the flow—an impossible situation for the case we
are considering here. Hence, the values of the λ j ’s obtained above should obey
the relation
n
λj Sj = 0 (3.157)
j=1

Equation (3.157) provides an independent check on the accuracy of the numerical

results.

EXAM PL E 3.19
Calculate the pressure coefficient distribution around a circular cylinder using the source
panel technique.
■ Solution
We choose to cover the body with eight panels of equal length, as shown in Figure 3.41.
This choice is arbitrary; however, experience has shown that, for the case of a circular
cylinder, the arrangement shown in Figure 3.41 provides sufficient accuracy. The panels
are numbered from 1 to 8, and the control points are shown by the dots in the center of
each panel.
Let us evaluate the integrals Ii, j which appear in Equation (3.153). Consider Fig-
ure 3.42, which illustrates two arbitrary chosen panels. In Figure 3.42, (xi , yi ) are the
coordinates of the control point of the ith panel and (x j , y j ) are the running coordinates
over the entire jth panel. The coordinates of the boundary points for the ith panel are
(X i , Yi ) and (X i+1 , Yi+1 ); similarly, the coordinates of the boundary points for the jth
panel are (X j , Y j ) and (X j+1 , Y j+1 ). In this problem, V∞ is in the x direction; hence,
the angles between the x axis and the unit vectors ni and n j are βi and β j , respectively.
Note that, in general, both βi and β j vary from 0 to 2π. Recall that the integral Ii, j is

Figure 3.41 Source panel distribution around a circular

cylinder.
 CHAPTER 3 Fundamentals of Inviscid, Incompressible Flow 289

Figure 3.42 Geometry required for the evaluation of Ii j .

evaluated at the ith control point and the integral is taken over the complete jth panel:

∂
Ii, j = (ln ri j ) ds j (3.158)
j ∂n i

Since ri j = (xi − x j )2 + (yi − y j )2

∂ 1 ∂ri j
then (ln ri j ) =
∂n i ri j ∂n i
1 1
= [(xi − x j )2 + (yi − y j )2 ]−1/2
ri j 2

d xi dyi
× 2(xi − x j ) + 2(yi − y j )
dn i dn i
∂ (xi − x j ) cos βi + (yi − y j ) sin βi
or (ln ri j ) = (3.159)
∂n i (xi − x j )2 + (yi − y j )2
Note in Figure 3.42 that i and j are angles measured in the counterclockwise direction
from the x axis to the bottom of each panel. From this geometry,
π
βi = i +
2
Hence, sin βi = cos i (3.160a)
cos βi = − sin i (3.160b)

Also, from the geometry of Figure 3.38, we have

x j = X j + s j cos j (3.161a)
and y j = Y j + s j sin j (3.161b)
290 PA R T 2 Inviscid, Incompressible Flow

Substituting Equations (3.159) to (3.161) into (3.158), we obtain

 Sj
Cs j + D
Ii, j = ds j (3.162)
0 s 2j + 2As j + B

where A = −(xi − X j ) cos j − (yi − Y j ) sin j

2 2
B = (xi − X j ) + (yi − Y j )
C = sin( i − j)

D = (yi − Y j ) cos i − (x i − X j ) sin i


S j = (X j+1 − X j )2 + (Y j+1 − Y j )2
√
Letting E= B − A2 = (xi − X j ) sin j − (yi − Y j ) cos j

we obtain an expression for Equation (3.162) from any standard table of integrals:

C S 2j + 2AS j + B
Ii, j = ln (3.163)
2 B
 
D − AC −1 Sj + A A
+ tan − tan−1
E E E
Equation (3.163) is a general expression for two arbitrarily oriented panels; it is not
restricted to the case of a circular cylinder.
We now apply Equation (3.163) to the circular cylinder shown in Figure 3.41. For
purposes of illustration, let us choose panel 4 as the ith panel and panel 2 as the jth panel;
that is, let us calculate I4,2 . From the geometry of Figure 3.41, assuming a unit radius for
the cylinder, we see that
X j = −0.9239 X j+1 = −0.3827 Y j = 0.3827
Y j+1 = 0.9239 i = 315◦ j = 45◦
xi = 0.6533 yi = 0.6533
Hence, substituting these numbers into the above formulas, we obtain
A = −1.3065 B = 2.5607 C = −1 D = 1.3065
S j = 0.7654 E = 0.9239
Inserting the above values into Equation (3.163), we obtain

I4,2 = 0.4018

Return to Figures 3.41 and 3.42. If we now choose panel 1 as the jth panel, keeping panel 4
as the ith panel, we obtain, by means of a similar calculation, I4,1 = 0.4074. Similarly,
I4,3 = 0.3528, I4,5 = 0.3528, I4,6 = 0.4018, I4,7 = 0.4074, and I4,8 = 0.4084.
Return to Equation (3.153), which is evaluated for the ith panel in Figures 3.40 and
3.42. Written for panel 4, Equation (3.153) becomes (after multiplying each term by 2 and
 CHAPTER 3 Fundamentals of Inviscid, Incompressible Flow 291

noting that βi = 45◦ for panel 4)

0.4074λ1 + 0.4018λ2 + 0.3528λ3 + πλ4 + 0.3528λ5

+ 0.4018λ6 + 0.4074λ7 + 0.4084λ8 = −0.7071 2π V∞ (3.164)

Equation (3.164) is a linear algebraic equation in terms of the eight unknowns, λ1 , λ2 , . . . ,

λ8 . If we now evaluate Equation (3.153) for each of the seven other panels, we obtain a
total of eight equations, including Equation (3.164), which can be solved simultaneously
for the eight unknown λ’s. The results are

λ1 /2π V∞ = 0.3765 λ2 /2π V∞ = 0.2662 λ3 /2π V∞ = 0

λ4 /2π V∞ = −0.2662 λ5 /2π V∞ = −0.3765 λ6 /2π V∞ = −0.2662
λ7 /2π V∞ = 0 λ8 /2π V∞ = 0.2662

Note the symmetrical distribution of the λ’s, which is to be expected for the nonlifting
circular cylinder. Also, as a check on the above solution, return to Equation (3.157). Since
each panel in Figure 3.41 has the same length, Equation (3.157) can be written simply as


n
λj = 0
j=1

Substituting the values for the λ’s obtained into Equation (3.157), we see that the equation
is identically satisfied.
The velocity at the control point of the ith panel can be obtained from Equa-
tion (3.156). In that equation, the integral over the jth panel is a geometric quantity
that is evaluated in a similar manner as before. The result is

∂ D − AC S 2j + 2AS j + B
(ln ri j )ds j = ln (3.165)
j ∂s 2E B
 
−1 Sj + A A
−C tan − tan−1
E E

With the integrals in Equation (3.156) evaluated by Equation (3.165), and with the values
for λ1 , λ2 , . . . , λ8 obtained above inserted into Equation (3.156), we obtain the veloci-
ties V1 , V2 , . . . , V8 . In turn, the pressure coefficients C p,1 , C p,2 , . . . , C p,8 are obtained
directly from
 2
Vi
C p,i = 1 −
V∞

Results for the pressure coefficients obtained from this calculation are compared with the
exact analytical result, Equation (3.101) in Figure 3.43. Amazingly enough, in spite of the
relatively crude paneling shown in Figure 3.41, the numerical pressure coefficient results
are excellent.
292 PA R T 2 Inviscid, Incompressible Flow

Figure 3.43 Pressure distribution over a circular cylinder; comparison of the

source panel results and theory.

3.18 APPLIED AERODYNAMICS: THE FLOW OVER

A CIRCULAR CYLINDER—THE REAL CASE
The inviscid, incompressible flow over a circular cylinder was treated in Sec-
tion 3.13. The resulting theoretical streamlines are sketched in Figure 3.26, char-
acterized by a symmetrical pattern where the flow “closes in” behind the cylinder.
As a result, the pressure distribution over the front of the cylinder is the same as
that over the rear (see Figure 3.29). This leads to the theoretical result that the
pressure drag is zero—d’Alembert’s paradox.
The real flow over a circular cylinder is quite different from that studied in
Section 3.13, the difference due to the influence of friction. Moreover, the drag
coefficient for the real flow over a cylinder is certainly not zero. For a viscous
incompressible flow, the results of dimensional analysis (Section 1.7) clearly
demonstrate that the drag coefficient is a function of the Reynolds number. The
variation of C D = f (Re) for a circular cylinder is shown in Figure 3.44, which
is based on a wealth of experimental data. Here, Re = (ρ∞ V∞ d)/μ∞ , where
d is the diameter of the cylinder. Note that C D is very large for the extremely
small values of Re < 1, but decreases monotonically until Re ≈ 300,000. At
this Reynolds number, there is a precipitous drop of C D from a value near 1 to
about 0.3, then a slight recovery to about 0.6 for Re = 107 . (Note: These results
are consistent with the comparison shown in Figure 1.54d and e, contrasting C D
for a circular cylinder at low and high Re.) What causes this precipitous drop in
C D when the Reynolds number reaches about 300,000? A detailed answer must
await our discussion of viscous flow in Chapter 4, and later in Part 4. However,
we state now that the phenomenon is caused by a sudden transition of laminar
flow within the boundary layer at the lower values of Re to a turbulent boundary