Applied Computational Aerodynamics

APPLIED
COMPUTATIONAL
AERODYNAMICS
W. H. Mason Professor of Aerospace & Ocean

Engineering Virginia Polytechnic Institute & State
University
Copyright 1998 by W. F. Mason
Preface
Objectives
These notes are intended to fill a significant gap in the literature available to students. There is a huge
disparity between the aerodynamics covered in typical aerodynamics courses and the application of
aerodynamic theory to design and analysis problems using computational methods. As an elective
course for seniors, Applied Computational Aerodynamics provides an opportunity for students to gain
insight into the methods and means by which aerodynamics is currently practiced. The specific
threefold objective is: i) physical insight into aerodynamics that can arise only with the actual
calculation and subsequent analysis of flowfields, ii) development of engineering judgment to answer
the question how do you know the answer is right? and iii) establishment of a foundation for future
study in computational aerodynamics; exposure to a variety of methods, terminology, and jargon.
Two features are unique. First, when derivations are given, all the steps in the analysis are included.
Second, virtually all the examples used to illustrate applied aerodynamics ideas were computed by the
author, and were made using the codes available to the students. The exercises are an extremely
important component of the course, where parts of the course are possibly best presented as a
workshop, rather than as a series of formal lectures. To meet the objectives, many old fashioned
methods are included. Using these methods a student can learn much more about aerodynamic design
than by performing a few large modern calculations. For example (articulated to the author by Prof.
Ilan Kroo), the vortex lattice method allows the student to develop an excellent mental picture of the
flowfield. Thus these methods provide a context within which to understand Euler or Navier-Stokes
calculations.
Audience
We presume that the reader has had standard undergraduate courses in fluid mechanics and
aerodynamics. In some cases the material is repeated to illustrate issues important to computational
aerodynamics. Access to a computer and the ability to program is assumed for the exercises.
Warnings
Computational aerodynamics is still in an evolutionary phase. Although most of the material in the
early chapters is essentially well established, the viewpoint adopted in the latter chapters is necessarily
a snapshot of the field at this time. Students that enter the field can expect to use this material as a
starting point in understanding the continuing evolution of computational aerodynamics.
These notes are not independent of other texts. At this point several of the codes used in the instruction
are based on source codes copyrighted in other sources. Use of these codes without owning the text
may be a violation of the copyright law.
The traditional printed page is inadequate and obsolete for the presentation of computational
aerodynamics information. The reader should be alert to advances in information presentation, and
take every opportunity to make use of advanced color displays, interactive flowfield visualization and
virtual environment technology.
The codes available on disk provide a significant capability for skilled users. However, as discussed in
the text, few computational aerodynamics codes are ever developed and tested to the level that they
are bug free. They are for educational use only, and are only aides for education, not commercial
programs, although they are entirely representative of codes in current use.
Acknowledgements
Many friends and colleagues have influenced the contents of these notes. Specifically, they reflect
many years developing and applying computational aerodynamics at Grumman, which had more than
its share of top flight aerodynamicists. Initially at Grumman and now at VPI, Bernard Grossman
provided access to his as yet unpublished CFD course notes. At NASA, many friends have contributed
help, insight and computer programs. Nathan Kirschbaum read the notes and made numerous
contributions to the content and clarity. Several classes of students have provided valuable feedback,
found typographical and actual errors. They have also insisted that the notes and codes be completed.
I would like to acknowledge these contributions.
W.H. Mason
Return to the main table of contents
4. Incompressible Potential Flow

Using Panel Methods
4.1 An Introduction
The incompressible potential flow model provides reliable flowfield predictions over a wide range
of conditions. For the potential flow assumption to be valid for aerodynamics calculations the
primary requirement is that viscous effects are small in the flowfield, and that the flowfield must be
subsonic everywhere. Locally supersonic velocities can occur at surprisingly low freestream Mach
numbers. For high-lift airfoils the peak velocities around the leading edge can become supersonic
at freestream Mach numbers of 0.20 ~ 0.25. If the local flow is at such a low speed everywhere
that it can be assumed incompressible (M .4, say), Laplaces Equation is essentially an exact
representation of the inviscid flow. For higher subsonic Mach numbers with small disturbances to
the freestream flow, the Prandtl-Glauert (P-G) Equation can be used. The P-G Equation can be
converted to Laplaces Equation by a simple transformation.1 This provides the basis for estimating
the initial effects of compressibility on the flowfield, i.e., linearized subsonic flow. In both
cases, the flowfield can be found by the solution of a single linear partial differential equation. Not
only is the mathematical problem much simpler than any of the other equations that can be used to
model the flowfield, but since the problem is linear, a large body of mathematical theory is
available.
The Prandtl-Glauert Equation can also be used to describe supersonic flows. In that case the
mathematical type of the equation is hyperbolic, and will be mentioned briefly in Chapter 12.
Recall the important distinction between the two cases:
subsonic flow:
supersonic flow:
elliptic PDE, each point influences every other point,

hyperbolic PDE, discontinuities exist, zone of influence
solution dependency.
In this chapter we consider incompressible flow only. One of the key features of Laplaces
Equation is the property that allows the equation governing the flowfield to be converted from a 3D
problem throughout the field to a 2D problem for finding the potential on the surface. The solution
is then found using this property by distributing singularities of unknown strength over
discretized portions of the surface: panels. Hence the flowfield solution is found by representing
2/24/98
4-1
4-2 Applied Computational Aerodynamics

the surface by a number of panels, and solving a linear set of algebraic equations to determine the
unknown strengths of the singularities.* The flexibility and relative economy of the panel methods
is so important in practice that the methods continue to be widely used despite the availability of
more exact methods (which generally arent yet capable of treating the range of geometries that the
panel method codes can handle). An entry into the panel method literature is available through two
recent reviews by Hess,23 the survey by Erickson,4 and the book by Katz and Plotkin.5
The general derivation of the integral equation for the potential solution of Laplaces equation is
given in Section 4.3. Complete details are presented for one specific approach to solving the
integral equation in Section 4.4. For clarity and simplicity of the algebra, the analysis will use the
two-dimensional case to illustrate the methods following the analysis given by Moran.6 This results
in two ironic aspects of the presentation:
The algebraic forms of the singularities are different between 2D and 3D, due to 3D
relief. You cant use the actual formulas we derive in Section 4.4 for 3D problems.
The power of panel methods arises in three-dimensional applications. Twodimensional work in computational aerodynamics is usually done in industry using
more exact mappings,** not panels.
After the general derivation, a panel method is used to examine the aerodynamics of airfoils.
Finally, an example and some distinctive aspects of the 3D problem are presented.
4.2 Some Potential Theory
Potential theory is an extremely well developed (old) and elegant mathematical theory, devoted to
the solution of Laplaces Equation:
2 = 0 .
(4.1)
There are several ways to view the solution of this equation. The one most familiar to
aerodynamicists is the notion of singularities. These are algebraic functions which satisfy
Laplaces equation, and can be combined to construct flowfields. Since the equation is linear,
superposition of solutions can be used. The most familiar singularities are the point source, doublet
and vortex. In classical examples the singularities are located inside the body. Unfortunately, an
arbitrary body shape cannot be created using singularities placed inside the body. A more
sophisticated approach has to be used to determine the potential flow over arbitrary shapes.
Mathematicians have developed this theory. We will draw on a few selected results to help
understand the development of panel methods. Initially, we are interested in the specification of the
boundary conditions. Consider the situation illustrated Fig. 4-1.
*
The singularities are distributed across the panel. They are not specified at a point. However, the boundary
conditions usually are satisfied at a specific location.
**
These will be mentioned in more detail in Chapter 9.
2/24/98
Panel Methods 4-3
Figure 4-1. Boundaries for flowfield analysis.

The flow pattern is uniquely determined by giving either:
or
on + {Dirichlet Problem: Design}
(4-2)
/n on + {Neuman Problem: Analysis}.
(4-3)
Potential flow theory states that you cannot specify both arbitrarily, but can have a mixed
boundary condition, a + b /n on + . The Neumann Problem is identified as analysis
above because it naturally corresponds to the problem where the flow through the surface is
specified (usually zero). The Dirichlet Problem is identified as design because it tends to
correspond to the aerodynamic case where a surface pressure distribution is specified and the body
shape corresponding to the pressure distribution is sought. Because of the wide range of problem
formulations available in linear theory, some analysis procedures appear to be Dirichlet problems,
but Eq. (4-3) must still be used.
Some other key properties of potential flow theory:
If either or /n is zero everywhere on + then = 0 at all interior points.
cannot have a maximum or minimum at any interior point. Its maximum value can
only occur on the surface boundary, and therefore the minimum pressure (and
maximum velocity) occurs on the surface.
2/24/98

4.3 Derivation of the Integral Equation for the Potential
We need to obtain the equation for the potential in a form suitable for use in panel method
calculations. This section follows the presentation given by Karamcheti7 on pages 344-348 and
Katz and Plotkin5 on pages 52-58. An equivalent analysis is given by Moran6 in his Section 8.1.
The objective is to obtain an expression for the potential anywhere in the flowfield in terms of
values on the surface bounding the flowfield. Starting with the Gauss Divergence Theorem, which
relates a volume integral and a surface integral,
divAdV = A n dS
R
(4-4)
we follow the classical derivation and consider the interior problem as shown in Fig. 4-2.
S0
n
z
y
R0
x
Figure 4-2. Nomenclature for integral equation derivation.
To start the derivation introduce the vector function of two scalars:
A = grad grad .
(4-5)
Substitute this function into the Gauss Divergence Theorem, Eq. (4-4), to obtain:
div(grad grad )dV = (grad grad ) n dS. .

R
(4-6)
Now use the vector identity: F= F + F to simplify the left hand side of Eq. (4-6).
Recalling that A = divA, write the integrand of the LHS of Eq. (4-6) as:
div(grad grad ) = ( ) ( )
= +
= 2 2
2/24/98
(4-7)
Panel Methods 4-5

Substituting the result of Eq. (4-7) for the integrand in the LHS of Eq. (4-6), we obtain:
( )dV = (grad grad ) n dS ,

2
(4-8)
or equivalently (recalling that grad n = / n ),
( )dV = n n dS .
2
(4-9)
Either statement is known as Greens theorem of the second form.

Now, define = 1/r and = , where is a harmonic function (a function that satisfies
Laplaces equation). The 1/r term is a source singularity in three dimensions. This makes our
analysis three-dimensional. In two dimensions the form of the source singularity is ln r, and a twodimensional analysis starts by defining = ln r. Now rewrite Eq. (4-8) using the definitions of
and given at the first of this paragraph and switch sides,
1
1 2
r r ndS = r
S0
R0
2 1
dV .
r
(4-10)
R 0 is the region enclosed by the surface S0 . Recognize that on the right hand side the first term,
2 , is equal to zero by definition so that Eq. (4-10) becomes
1
2 1
r r ndS =
S0
R0
dV .
(4-11)
1
If a point P is external to S0 , then 2 = 0 everywhere since 1/r is a source, and thus satisfies
r
Laplaces Equation. This leaves the RHS of Eq. (4-11) equal to zero, with the following result:
1
r r n dS = 0.
(4-12)
S0
However, we have included the origin in our region S0 as defined above. If P is inside S0 , then
1
2 at r = 0. Therefore, we exclude this point by defining a new region which excludes
r
the origin by drawing a sphere of radius around r = 0, and applying Eq. (4-12) to the region
between and S0 :
2/24/98

1
r r ndS r r + r2 dS = 0
(4-13)
S0
4 42443
1
444
4244 44
3 1
sphere
arbitrary region
or:
r r + r2 dS = r r ndS .
(4-14)
S0
Consider the first integral on the left hand side of Eq. (4-14). Let 0, where (as 0)
we take constant ( / r == 0 ), assuming that is well-behaved and using the mean value
theorem. Then we need to evaluate
dS
r2
over the surface of the sphere where = r. Recall that for a sphere* the elemental area is
dS = r 2 sin dd
(4-15)
where we define the angles in Fig. 4-3. Do not confuse the classical notation for the spherical
coordinate angles with the potential function. The spherical coordinate will disappear as soon as
we evaluate the integral.
z
y
x
Figure 4-3. Spherical coordinate system nomenclature.

Substituting for dS in the integral above, we get:
sin dd .
Integrating from = 0 to , and from 0 to 2, we get:
See Hildebrand, F.B., Advanced Calculus for Applications, 2nd Ed., Prentice-Hall, Englewood Cliffs, 1976 for an
excellent review of spherical coordinates and vector analysis.
2/24/98
Panel Methods 4-7

=2 =
=0 =0 sin dd = 4 .
(4-16)
The final result for the first integral in Eq. (4-14) is:
1
r r + r2 dS = 4 .
(4-17)
Replacing this integral by its value from Eq. (4-17) in Eq. (4-14), we can write the expression
for the potential at any point P as (where the origin can be placed anywhere inside S0 ):
( p) =
1
4
r r n dS
(4-18)
s0
and the value of at any point P is now known as a function of and /n on the boundary.
We used the interior region to allow the origin to be written at point P. This equation can be
extended to the solution for for the region exterior to R 0 . Apply the results to the region between
the surface SB of the body and an arbitrary surface enclosing SB and then let go to infinity. The
integrals over go to as goes to infinity. Thus potential flow theory is used to obtain the
important result that the potential at any point P' in the flowfield outside the body can be expressed
as:
( p ) =
1
4
r r n dS .
(4-19)
SB
Here the unit normal n is now considered to be pointing outward and the area can include not only
solid surfaces but also wakes. Equation 4-19 can also be written using the dot product of the
normal and the gradient as:
( p ) =
1
4
r n n r dS .
(4-20)
SB
The 1/r in Eq. (4-19) can be interpreted as a source of strength / n , and the (1/r) term in
Eq. (4-19) as a doublet of strength . Both of these functions play the role of Greens functions in
the mathematical theory. Therefore, we can find the potential as a function of a distribution of
sources and doublets over the surface. The integral in Eq. (4-20) is normally broken up into
body and wake pieces. The wake is generally considered to be infinitely thin. Therefore, only
doublets are used to represent the wakes.
2/24/98

Now consider the potential to be given by the superposition of two different known functions,
the first and second terms in the integral, Eq. (4-20). These can be taken to be the distribution of
the source and doublet strengths, and , respectively. Thus Eq (4-20) can be written in the form
usually seen in the literature,
( p ) =
1
4
r n r dS .
(4-21)
SB
The problem is to find the values of the unknown source and doublet strengths and for a
specific geometry and given freestream, .
What just happened? We replaced the requirement to find the solution over the entire flowfield
(a 3D problem) with the problem of finding the solution for the singularity distribution over a
surface (a 2D problem). In addition, we now have an integral equation to solve for the unknown
surface singularity distributions instead of a partial differential equation. The problem is linear,
allowing us to use superposition to construct solutions. We also have the freedom to pick whether
to represent the solution as a distribution of sources or doublets distributed over the surface. In
practice its been found best to use a combination of sources and doublets. The theory can be
extended to include other singularities.
At one time the change from a 3D to a 2D problem was considered significant. However, the
total information content is the same computationally. This shows up as a dense 2D matrix vs. a
sparse 3D matrix. As methods for sparse matrix solutions evolved, computationally the problems
became nearly equivalent. The advantage in using the panel methods arises because there is no
need to define a grid throughout the flowfield.
This is the theory that justifies panel methods, i.e., that we can represent the surface by panels
with distributions of singularities placed on them. Special precautions must be taken when
applying the theory described here. Care should be used to ensure that the region SB is in fact
completely closed. In addition, care must be taken to ensure that the outward normal is properly
defined.
Furthermore, in general, the interior problem cannot be ignored. Surface distributions of
sources and doublets affect the interior region as well as exterior. In some methods the interior
problem is implicitly satisfied. In other methods the interior problem requires explicit attention. The
need to consider this subtlety arose when advanced panel methods were developed. The problem is
not well posed unless the interior problem is considered, and numerical solutions failed when this
aspect of the problem was not addressed. References 4 and 5 provide further discussion.
2/24/98
Panel Methods 4-9

When the exterior and interior problems are formulated properly the boundary value problem
is properly posed. Additional discussions are available in the books by Ashley and Landahl8 and
Curle and Davis.9
We implement the ideas give above by:
a) approximating the surface by a series of line segments (2D) or panels (3D)
b) placing distributions of sources and vortices or doublets on each panel.
There are many ways to tackle the problem (and many competing codes). Possible differences
in approaches to the implementation include the use of:
- various singularities
- various distributions of the singularity strength over each panel
- panel geometry (panels dont have to be flat).
Recall that superposition allows us to construct the solution by adding separate contributions
[Watch out! You have to get all of them. Sometimes this can be a problem]. Thus we write the
potential as the sum of several contributions. Figure 4-4 provides an example of a panel
representation of an airplane. The wakes are not shown, and a more precise illustration of a panel
method representation is given in Section 4.8.
Figure 4-4. Panel model representation of an airplane.

(Joe Mazza, M.S. Thesis, Virginia Tech, 1993).
An example of the implementation of a panel method is carried out in Section 4.4 in two
dimensions. To do this, we write down the two-dimensional version of Eq. (4-21). In addition,
we use a vortex singularity in place of the doublet singularity (Ref. 4 and 5 provide details on this
change). The resulting expression for the potential is:
2/24/98
q(s)
(s)
=
{
+
lnr
ds
224
2
4
3
1
23
uniform onset flow
1
this is a vortex singularity

=V x cos +V y sin
q is the 2D
source strength
of strength (s)
(4-22)
and = tan (y/x). Although the equation above shows contributions from various components of
-1
the flowfield, the relation is still exact. No small disturbance assumption has been made.
4.4 The Classic Hess and Smith Method
A.M.O. Smith at Douglas Aircraft directed an incredibly productive aerodynamics development
group in the late 50s through the early 70s. In this section we describe the implementation of the
theory given above that originated in his group.* Our derivation follows Morans description6 of
the Hess and Smith method quite closely. The approach is to i) break up the surface into straight
line segments, ii) assume the source strength is constant over each line segment (panel) but has a
different value for each panel, and iii) the vortex strength is constant and equal over each panel.
Roughly, think of the constant vortices as adding up to the circulation to satisfy the Kutta
condition. The sources are required to satisfy flow tangency on the surface (thickness).
Figure 4-5 illustrates the representation of a smooth surface by a series of line segments. The
numbering system starts at the lower surface trailing edge and proceeds forward, around the
leading edge and aft to the upper surface trailing edge. N+1 points define N panels.
node
N -1
N
N+1
panel
Figure 4-5. Representation of a smooth airfoil with straight line segments.
The potential relation given above in Eq. (4-22) can then be evaluated by breaking the integral
up into segments along each panel:
N
= V ( xcos + ysin ) +
j=1panel j

q(s)
2 ln r 2 dS
(4-23)
In the recent AIAA book, Applied Computational Aerodynamics, A.M.O. Smith contributed the first chapter, an
account of the initial development of panel methods.
2/24/98
Panel Methods 4-11

with q(s) taken to be constant on each panel, allowing us to write q(s) = qi, i = 1, ... N. Here we
need to find N values of qi and one value of .
i+1
li
^t
i
^n
i
i+1
i
x
a) basic nomenclature
b) unit vector orientation
Figure 4-6. Nomenclature for local coordinate systems.

Use Figure 4-6 to define the nomenclature on each panel. Let the i th panel be the one between
the i th and i+1th nodes, and let the i th panels inclination to the x axis be . Under these
assumptions the sin and cos of are given by:
y y
sini = i+1 i ,
li
x x
cos i = i+1 i
li
(4-24)
and the normal and tangential unit vectors are:

ni = sin ii + cosi j
.
ti = cos ii + sini j
(4-25)
We will find the unknowns by satisfying the flow tangency condition on each panel at one
specific control point (also known as a collocation point) and requiring the solution to satisfy the
Kutta condition. The control point will be picked to be at the mid-point of each panel, as shown in
Fig. 4-7.
Y
smooth shape
control point
panel
X
Figure 4-7. Local panel nomenclature.
2/24/98

Thus the coordinates of the midpoint of the control point are given by:
x +x
xi = i i +1 ,
2
y +y
yi = i i +1
2
(4-26)
and the velocity components at the control point xi , yi are ui = u(xi , yi ), vi = v(xi , yi ).
The flow tangency boundary condition is given by V n = 0, and is written using the relations
given here as:
(ui i + vi j) ( sini i + cosi j) = 0

or
ui sini + vi cos i = 0,
for each i, i = 1, ..., N .
(4-27)
The remaining relation is found from the Kutta condition. This condition states that the flow
must leave the trailing edge smoothly. Many different numerical approaches have been adopted to
satisfy this condition. In practice this implies that at the trailing edge the pressures on the upper and
lower surface are equal. Here we satisfy the Kutta condition approximately by equating velocity
components tangential to the panels adjacent to the trailing edge on the upper and lower surface.
Because of the importance of the Kutta condition in determining the flow, the solution is extremely
sensitive to the flow details at the trailing edge. When we make the assumption that the velocities
are equal on the top and bottom panels at the trailing edge we need to understand that we must
make sure that the last panels on the top and bottom are small and of equal length. Otherwise we
have an inconsistent approximation. Accuracy will deteriorate rapidly if the panels are not the same
length. We will develop the numerical formula using the nomenclature for the trailing edge shown
in Fig. 4-8.
^t
N
N+1
^t
1
1
Figure 4-8. Trailing edge panel nomenclature.

Equating the magnitude of the tangential velocities on the upper and lower surface:
ut1 = utN .
(4-28)
and taking the difference in direction of the tangential unit vectors into account this is written as
V t 1 = V t N .
2/24/98
(4-29)
Panel Methods 4-13

Carrying out the operation we get the relation:
(u1i + v1j) ( cos1i + sin1 j) = (uN i + vN j) (cosN i + sin N j)

which is expanded to obtain the final relation:
u1 cos1 + v1 sin1 = uN cos N + vN sin N
(4-30)
The expression for the potential in terms of the singularities on each panel and the boundary
conditions derived above for the flow tangency and Kutta condition are used to construct a system
of linear algebraic equations for the strengths of the sources and the vortex. The steps required are
summarized below. Then we will carry out the details of the algebra required in each step.
Steps to determine the solution:
1. Write down the velocities, ui, v i, in terms of contributions from all the singularities. This
includes qi, from each panel and the influence coefficients which are a function of the
geometry only.
2. Find the algebraic equations defining the influence coefficients.
To generate the system of algebraic equations:
3. Write down flow tangency conditions in terms of the velocities (N eqns., N+1
unknowns).
4. Write down the Kutta condition equation to get the N+1 equation.
5. Solve the resulting linear algebraic system of equations for the qi, .
6. Given qi, , write down the equations for uti, the tangential velocity at each panel control
point.
7. Determine the pressure distribution from Bernoullis equation using the tangential
velocity on each panel.
We now carry out each step in detail. The algebra gets tedious, but theres no problem in
carrying it out. As we carry out the analysis for two dimensions, consider the additional algebra
required for the general three dimensional case.
2/24/98
Step 1. Velocities
The velocity components at any point i are given by contributions from the velocities induced
by the source and vortex distributions over each panel. The mathematical statement is:
N
j=1
j =1
j=1
j=1
ui = V cos + q jusij + uvij

vi = V sin +
(4-31)
q jvs ij + vvij
where qi and are the singularity strengths, and the usij , v sij , uvij, and v vij are the influence
coefficients. As an example, the influence coefficient usij is the x-component of velocity at x i due to
a unit source distribution over the j th panel.
Step 2. Influence coefficients
To find usij , v sij , uvij, and v vij we need to work in a local panel coordinate system x*, y* which
leads to a straightforward means of integrating source and vortex distributions along a straight line
segment. This system will be locally aligned with each panel j, and is connected to the global
coordinate system as illustrated in Fig. 4-9.
Y
Y*
lj
j+1
X*
j
j
Figure 4-9. Local panel coordinate system and nomenclature.

The influence coefficients determined in the local coordinate system aligned with a particular
panel are u* and v*, and are transformed back to the global coordinate system by:
u = u * cos j v*sin j
v = u * sin j + v * cos j
2/24/98
(4-32)
Panel Methods 4-15

We now need to find the velocities induced by the singularity distributions. We consider the source
distributions first. The velocity field induced by a source in its natural cylindrical coordinate system
is:
Q
V=
e .
(4-33)
2r r
Rewriting in Cartesian coordinates (and noting that the source described in Eq. (4-33) is
located at the origin, r = 0) we have:
u(x,y) =
Q
x
,
2
2 x + y2
v(x,y) =
Q
y
.
2
2 x + y2
(4-34)
In general, if we locate the sources along the x-axis at a point x = t, and integrate over a length l,
the velocities induced by the source distributions are obtained from:
q(t)
x t
dt
t=0 2 (x t)2 + y 2
us =
t=l
q(t)
y
vs =
dt
t=0 2 (x t)2 + y 2
t=l
(4-35)
To obtain the influence coefficients, write down this equation in the ( )* coordinate system,
with q(t) = 1 (unit source strength):
u*sij =
1 lj
xi* t
dt
2 0 (x * t) 2 + y*2
i
v*sij
1 lj
y*i
=
dt
2 0 (x * t) 2 + y*2
i
i
(4-36)
These integrals can be found (from tables) in closed form:
u*sij =
1 *
ln x t
2 i
1 t=l j
2
2
+ y*i 2
t=0
(4-37)
t=l
v*sij
j
*
1
1 yi
=
tan *
2
xi t t=0
To interpret these expressions examine Fig. 4-10. The notation adopted and illustrated in the
sketch makes it easy to translate the results back to global coordinates.
2/24/98
y*
x*, y*
i i
rij
ri,j+1
ij
0
j
j+1
lj
x*
Figure 4-10. Relations between the point x*, y* and a panel.

Note that the formulas for the integrals given in Eq. (4-37) can be interpreted as a radius and
an angle. Substituting the limits into the expressions and evaluating results in the final formulas for
the influence coefficients due to the sources:
u*sij =
1 ri, j+1
ln
2 rij
ij
(4-38)
0
v*sij = l
=
2
2
Here rij is the distance from the j th node to the point i, which is taken to be the control point
location of the i th panel. The angle ij is the angle subtended at the middle of the i th panel by the j th
panel.
The case of determining the influence coefficient for a panels influence on itself requires
some special consideration. Consider the influence of the panel source distribution on itself. The
source induces normal velocities, and no tangential velocities, Thus, u*sii = 0 and vs*ii depends on
the side from which you approach the panel control point. Approaching the panel control point
from the outside leads to ii = , while approaching from inside leads to ii = -. Since we are
working on the exterior problem,
ii = ,
(4-39)
and to keep the correct sign on ij, j i, use the FORTRAN subroutine ATAN2, which takes into
account the correct quadrant of the angle.*
*
Review a FORTRAN manual to understand how ATAN2 is used.
2/24/98
Panel Methods 4-17

Now consider the influence coefficients due to vortices. There is a simple connection between
source and vortex flows that allows us to use the previous results obtained for the source
distribution directly in the vortex singularity distribution analysis.
The velocity due to a point vortex is usually given as:
V =
e .
2r
(4-40)
Compared to the source flow, the u, v components simply trade places (with consideration of the
direction of the flow to define the proper signs). In Cartesian coordinates the velocity due to a point
vortex is:
x
u(x,y) = +
,
v(x, y) =
.
(4-41)
2
2
2
2 x + y
2 x + y2
where the origin (the location of the vortex) is x = y = 0.
Using the same analysis used for source singularities for vortex singularities the equivalent
vortex distribution results can be obtained. Summing over the panel with a vortex strength of unity
we get the formulas for the influence coefficients due to the vortex distribution:
uv*ij = +
1 lj
y*i
dt = ij
2
2 0 (x * t)2 + y*
2
i
i
1 lj
xi* t
1 ri, j+1
v*vij =
dt
=
ln
2 0 (x * t)2 + y*2
2 rij
i
i
(4-42)
where the definitions and special circumstances described for the source singularities are the same
in the current case of distributed vortices.* In this case the vortex distribution induces an axial
velocity on itself at the sheet, and no normal velocity.
Step 3. Flow tangency conditions to get N equations.
Our goal is to obtain a system of equations of the form:
N
Aij q j + Ai,N +1 = bi
i = 1,...N
(4-43)
j=1
which are solved for the unknown source and vortex strengths.
Recall the flow tangency condition was found to be:
ui sini + vi cos i = 0,
*
for each i,
i = 1,...N
Note that Morans Equation (4-88) has a sign error typo. The correct sign is used in Eq. (4-42) above.
2/24/98
(4-44)

where the velocities are given by:
N
j=1
j =1
j=1
j=1
ui = V cos + q jusij + uvij

vi = V sin +
(4-45)
q jvs ij + vvij
Substituting into Eq. (4-45), the flow tangency equations, Eq. (4-44), above:
N
N
N
N
V cos q jus uv sini + V sin + q j vs + vv cos i = 0

ij
ij
ij
ij
j =1
j=1
j=1
j =1
(4-46)
which is rewritten into:
N
[ V sini cos + V cos i sin ] sin i q j usij

j =1
+ cos i q jvsij
j=1
j =1
j=1
sin i uvij + cosi vvij = 0

or
N
V ( cosi sin sin i cos ) + cosi vsij sini usij q j

144 44
42444 44
3
144 4244 43
bi
j=1
Aij
N
N
+ cosi vvij sini uvij = 0
j=1
j=1
14444424 4444
3
(4-47)
Ai, N +1
Now get the formulas for A ij and A i,N+1 by replacing the formulas for usij , v sij ,uvij,v vij with the ( )*
values, where:
u = u * cos j v*sin j
v = u * sin j + v * cos j
and we substitute into Eq. (4-47) for the values in A ij and A i,N+1 above.
Start with:
2/24/98
(4-48)
Panel Methods 4-19

Aij = cosi vsij sini usij
= cos i u*sij sin j + v*sij cos j sini u*sij cos j v*sij sin j
(4-49)
= cosi sin j sini cos j u*sij + cos i cos j sini sin j vs*ij
and we use trigonometric identities to combine terms into a more compact form. Operating on the
first term in parenthesis:
1
1
cosi sin j = sin i + j + sin i j
2
2
(4-50)
1
1
= sin i + j sin i j
2
2
and
1
1
sini sin j = sin i + j + sin i j
(4-51)
2
2
results in:
({
})
(cos i sin j sin i cos j ) = 0 sin (i j ).
(4-52)
Moving to the second term in parentheses above:
1
1
cos i + j + cos i j
2
2
1
1
sin i sin j = cos i j cos i + j
2
2
cosi cos j =
(4-53)
and
1
1
1
1
cosi cos j +sin i sin j = cos i + j + cos i j + cos i j cos i + j
2
2
2
2
= cos i j
(4-54)
so that the expression for A ij can be written as:
Aij = sin i j u*sij + cos i j v*sij
(4-55)
and using the definitions of

Aij =
ri,j+1 1
1
sin( i j )ln
+
cos i j ij .
2
ri, j 2
(4-56)
Now look at the expression for bi identified in (4-47):

bi = V (cosi sin sini cos )
2/24/98
(4-57)

where in the same fashion used above:
1
1
cosi sin = sin (i + ) sin( i )
2
2
1
1
sini cos = sin (i + ) + sin( i )
2
2
(4-58)
cosi sin sini cos = sin (i )
(4-59)
bi = V sin(i ) .
(4-60)
and
so that we get:
Finally, work with the A i,N+1 term:
N
N
Ai,N+1 = cosi vvij sini uvij

j=1
j=1
= cos i uv*ij sin j + vv*ij cos j sini u*vij cos j vv*ij sin j
j=1
(
N
cos i sin j uv*ij
+ cos i cos j vv*ij
j=1
sini cos juv*ij
+ sin i sin j vv*ij
j=1
(4-61)
*
*
= cosi cos j + sini sin j vv ij + cos i sin j sini cos j uvij
444 43
144 44244443
j =1 14 4442
a
b
N
and a and b can be simplified to:
a = cos i j
b = sin i j
(4-62)
Substituting for a and b in the above equation:

Ai,N+1 =
{cos(i j )v*vij sin(i j )uv*ij }

N
(4-63)
j=1
and using the definition of we arrive at the final result:

Ai,N+1 =
2/24/98
ri,j+1
1 N
cos(
)ln
sin(
i
j
i
j ij .
2 j=1
ri, j
(4-64)
Panel Methods 4-21

To sum up (repeating the results found above), the equations for the A ij, A i,N+1 , and bi are
given by (4-56), (4-64), and (4-60):
Aij =
Ai, N+1 =
ri,j+1 1
1
sin(i j )ln
+
cos i j ij
2
ri, j 2
ri,j+1
1 N
cos(
)ln
sin(
i
j
i
j ij
2 j=1
ri, j
bi = V sin( i )
Step 4. Kutta Condition to get equation N+1
To complete the system of N+1 equations, we use the Kutta condition, which we previously
defined as:
u1 cos1 + v1 sin1 = uN cos N v N sin N
(4-66)
and substitute into this expression the formulas for the velocities due to the freestream and
singularities given in equation (4-31). In this case they are written as:
N
j =1
j=1
u1 = V cos + q jus1 j + uv1 j

v1 = V sin + q j vs1 j +
j=1
vv1 j
j=1
j=1
j=1
uN = V cos + q j us Nj + uvNj
vN = V sin + q j vsNj +
j=1
Substituting into the Kutta condition equation we obtain:
2/24/98
vvNj
j=1
(4-67)
N
N
V cos + q j us + uv cos1
1j
1j
j=1
j=1
N
N
+ V sin + q j vs1j + vv1 j sin1
j=1
j=1
(4-68)
N
N
+ V cos + q jus Nj + uv Nj cos N
j =1
j=1
N
N
+ V sin + q j vsNj + vv Nj sin N = 0

j=1
j =1
and our goal will be to manipulate this expression into the form:
N
AN+1,jq j + AN+1,N+1
= bN +1
(4-69)
j=1
which is the N + 1st equation which completes the system for the N + 1 unknowns.
Start by regrouping terms in the above equation to write it in the form:
cos1 + vs1 j sin1 + usNj cos N + vs Nj sin N )q j
(1us4
1j
44 4444 4424 444 4444 43
N
j=1
AN +1, j
+ uv1 j cos1 + vv1j sin1 + uvNj cos N + vv Nj sin N

j =1
1444 4444 4442444 4444 4443
(4-70)
AN +1, N +1
= (V cos cos1 + V sin sin1 + V cos cosN + V sin sin N )

144 4444 444 444
42444 444 4444 444
3
bN +1
Obtain the final expression for bN+1 first:

bN+1 = V (cos
cos
sin
sin
cos
cos
+4
sin
sin
N2
N)
144
4
412+4
44
4
3+ 1
44
44
44
43
cos( 1)
cos ( N )
(4-71)
and using the trigonometric identities to obtain the expression for bN+1 :
bN+1 = V cos (1 ) V cos( N )
2/24/98
(4-72)
Panel Methods 4-23

where we made use of cos(-A) = cos A.
Now work with A N+1,j:
AN+1, j = us1 j cos1 + vs1 j sin1 + usNj cos N + v sNj sin N
(4-73)
and replace the influence coefficients with their related ( )* values:

us1 j = us*1j cos j v *s1j sin j
vs1j = us*1 j sin j + v*s1 j cos j
(4-74)
usNj = us*Nj cos j v s*Nj sin j

vsNj = us*Nj sin j + v *sNj cos j
so that we can write:
AN+1, j = u*s1j cos j v*s1 j sin j cos1
+ u*s1 j sin j + v*s1 j cos j sin1

+
u*sNj cos j vs*Nj
(4-75)
sin j cos N
+ us*Nj sin j + vs*Nj cos j sinN

or:
AN+1, j = cos j cos1 + sin j sin1 u*s1j
)
.
+ ( cos j sin1 sin j cos1 )v*s1 j
+ ( cos j sin N sin j cos N )v*sNj
+ cos j cos N + sin j sin N us*Nj
(4-76)
Use the following trig relations to simplify the equation:
(
)
cos j cos N + sin j sin N = cos ( j N )
cos j sin1 sin j cos1 = sin ( j 1)
cos j sinN sin j cos N = sin ( j N )
cos j cos1 + sin j sin1 = cos j 1
2/24/98
(4-77)

and substitute into Eq. (4-76) to obtain:
AN+1, j = cos j 1 u*s1j + cos j N us*Nj
sin j 1 v*s1j sin j N vs*Nj
(4-78)
Use the definition of the influence coefficients:

u*s1 j =
1 r1,j+1
ln
2 r1,j
us*Nj =
1,j
v*s1 j =
v*sNj =
1 rN, j+1
ln
2 rN, j
(4-79)
N, j
2
to write the equation for A N+1,:

AN+1, j =
cos j 1 r1,j+1 cos j N rN, j+1

ln
ln
2
2
r1,j
rN, j
sin j 1
2
1, j
sin j N
2
(4-80)
N,j
Finally, use symmetry and odd/even relations to write down the final form:
sin(1 j )1,j + sin( N j ) N,j
1
r1, j+1
rN ,j+1 .
AN+1, j =
2 cos(1 j )ln
cos( N j )ln
r1,j
rN, j
(4-81)
Now work with A N+1, N+1 :

AN+1,N+1 =
(uv1 j cos1 + vv1 j sin1 + uv Nj cos N + v vNj sinN )

N
(4-82)
j=1
where we substitute in for the ( )* coordinate system, Eq. (4-32), and obtain:
*
*
*
*
uv1 j cos j vv1 j sin j cos1 + uv1 j sin j + vv1j cos j sin1
AN+1,N+1 =
(4-83)
*
*
*
*
j=1 + uv cos j vv sin j cos N + uv sin j + vv cos j sin N

Nj
Nj
Nj
Nj
or:
2/24/98
Panel Methods 4-25
cos cos + sin sin u * + cos sin sin cos v*
j
1
j
1 v1j
j
1
j
1 v1 j
144
44 244 443
144442444 43
N
cos( j 1 )
sin( j 1 )
AN+1,N+1 =
* + cos sin sin cos v*
+
cos
cos
+
sin
sin
u
j=1
j
N
j
N v
j
N
j
N v Nj
42444 44
3 Nj 14 44442444 44
3
144 44
cos( j N )
sin(j N )
(4-84)
which is:
*
*
cos( j 1 )uv1 j sin( j 1 )vv1j
AN+1,N+1 =
,
* sin( )v*
+cos(
)u
j=1
j
N
v
j
N
v
Nj
Nj
N
and using odd/even trig relations we get the form given by Moran6 :
*
*
sin(1 j )vv1 j + sin( N j )vv Nj

AN+1,N+1 =
.
* + cos( )u*
+
cos(
)u
j=1
1
j
v
N
j
v
1j
Nj
N
(4-86)
We now substitute the formulas derived above for the influence coefficients given in Eq. (442). The final equation is:
r1, j+1
rN ,j+1
1 N sin 1 j ln r + sin N j ln r
AN+1,N+1 =
i,j
N, j .
2 j =1
+ cos(1 j )1, j + cos( N j ) N,j
(4-86)
After substituting in the values of the velocities in terms of the singularity strengths, and
performing some algebraic manipulation, a form of the coefficients suitable for computations is
obtained.
The final equations associated with the Kutta condition are:
sin(1 j )1,j + sin( N j ) N,j
1
r1, j+1
rN ,j+1
AN+1, j =
2 cos(1 j )ln
cos( N j )ln
1,j
rN, j
(4-81)
r1, j+1
rN ,j+1
1 N sin 1 j ln r + sin N j ln r
AN+1,N+1 =
i,j
N, j
2 j =1
+ cos(1 j )1, j + cos( N j ) N,j
(4-86)
bN+1 = V cos(1 ) V cos( N ) .
(4-72)
2/24/98

Step 5. Solve the system for qi, .
The coefficients derived above provide the required coefficients to solve a system of linear
algebraic equations for the N+1 unknowns, qi, i = 1,...,N and given by (4-43) and (4-69):
N
Aijq j + Ai,N+1 = bi
i = 1,...N
j=1
AN+1,jq j + AN+1,N+1
(4-87)
= bN +1
j=1
This is easily done using any number of computer subroutines.

Step 6. Given qi, and , write down the equations for the tangential velocity at each
panel control point.
At each control point, (v n = 0), find ut, the tangential velocity starting with:
uti = ui cosi + vi sini
N
N
= V cos + usij q j + uvij cos i .
j=1
j=1
(4-88)
N
N
+ V sin + vsij q j + vv ij sini
j=1
j=1
Using the ( )* values of the influence coefficients,
N
N
*
*
*
*
uti = V cos + usij cos j vsij sin j q j + uv ij cos j vvij sin j cosi
j=1
j=1
N
N
+ V sin + u*sij sin j + v*sij cos j q j + uv*ij sin j + vv*ij cos j
j =1
j=1
sin i
(4-89)
or:
2/24/98
Panel Methods 4-27

uti = V cos cos i + V sin sini
+
{us*ij cos j cos i vs*ij sin j cos i + us*ij sin j sini + vs*ij cos j sini }q j .
N
j=1
{uv*ij cos j cosi v*vij sin j cos i + u*vij sin j sini + v*sij cos j sini }
N
j=1
(4-90)
Collecting terms:
uti = (cos cosi + sin sin i ) V
144 44244443
cos ( i )
+ (cos j cosi +sin j sin i )us*ij + (cos j sin i sin j cosi )vs*ij q j
14 4442444 43
14 4442444 43
j=1
cos( j i )
sin( j i )
(4-91)
+ (cos j cos i + sin j sini )u*vij + (cos j sin i sin j cos i )v*vij
144 4424 4443
14 4442444 43
j=1
cos( j i )
sin( j i )
which becomes:
N
uti = cos( i )V + cos( j i )u*sij sin( j i )v*sij q j

j=1
{cos( j i )u*vij sin( j i )vv*ij }

N
(4-92)
j=1
Using the definitions of the ( )* influence coefficients, and some trigonometric identities, we
obtain the final result:
N
r
q
uti = cos(i )V + i sin( i j )ij cos(i j )ln i, j+1
2
ri, j
j=1
ri,j+1
sin(
)ln
+
cos(
i
j
i
j ij
2 j=1
ri, j
2/24/98
(4-93)

Step 7. Finally, the surface pressure coefficient can be found from:
uti 2
CPi = 1
V
(4-94)
using ui from Eq. (4-93).

This completes our derivation of one panel method scheme in two dimensions. Imagine the
difficulty in performing the algebra required to extend this approach to three dimensions! Thats
why weve used a two-dimensional example.
4.5 Program PANEL
Program PANEL is an exact implementation of the analysis given in Section 4.4, and is
essentially the program given by Moran.6 Other panel method programs are available in the
textbooks by Houghton and Carpenter,10 and Kuethe and Chow.11 Morans program includes a
subroutine to generate the ordinates for the NACA 4-digit and 5-digit airfoils (see Appendix A for a
description of these airfoil sections). The main drawback is the requirement for a trailing edge
thickness thats exactly zero. To accommodate this restriction, the ordinates generated internally
have been altered slightly from the official ordinates. The extension of the program to handle
arbitrary airfoils is an exercise. The freestream velocity in PANEL is assumed to be unity, since
the inviscid solution in coefficient form is independent of scale.
PANELs node points are distributed employing the widely used cosine spacing function.
The equation for this spacing is given by defining the points on the thickness distribution to be
placed at:
(i 1)
xi 1
= 1 cos
c 2
( N 1)
i = 1,..., N .
(4-95)
These locations are then altered when camber is added (see Eqns. (A-1) and (A-2) in App. A).
This approach is used to provide a smoothly varying distribution of panel node points which
concentrate points around the leading and trailing edges.
An example of the accuracy of program PANEL is given in Fig. 4-11, where the results
from PANEL for the NACA 4412 airfoil are compared with results obtained from an exact
conformal mapping of the airfoil (comments on the mapping methods are given in Chapter 9 on
Geometry and Grids. Conformal transformations can also be used to generate meshes of points for
use in field methods). The agreement is nearly perfect.
Numerical studies need to be conducted to determine how many panels are required to obtain
accurate results. Both forces and moments and pressure distributions should be examined.
2/24/98
Panel Methods 4-29
-2.50
PANEL
-2.00
Exact Conformal Mapping
-1.50
-1.00
Cp
-0.50
0.00
0.50
1.00
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
x/c
Figure 4-11. Comparison of results from program PANEL with an essentially exact
mapping solution for the NACA 4412 airfoil at 6 angle-of-attack.
You can select the number of panels used to represent the surface. How many should you
use? Most computational programs provide the user with freedom to decide how detailed
(expensive - in dollars or time) the calculations should be. One of the first things the user should
do is evaluate how detailed the calculation should be to obtain the level of accuracy desired. In the
PANEL code your control is through the number of panels used.
We check the sensitivity of the solution to the number of panels by comparing force and
moment results and pressure distributions with increasing numbers of panels. This is done using
two different methods. Figures 4-12 and 4-13 present the change of drag and lift, respectively,
using the first method. For PANEL, which uses an inviscid incompressible flowfield model, the
drag should be exactly zero. The drag coefficient found by integrating the pressures over the airfoil
is an indication of the error in the numerical scheme. The drag obtained using a surface (or
nearfield) pressure integration is a numerically sensitive calculation, and is a strict test of the
method. The figures show the drag going to zero, and the lift becoming constant as the number of
2/24/98

panels increase. In this style of presentation it is hard to see exactly how quickly the solution is
converging to a fixed value.
The results given in Figures 4-12 and 4-13 indicate that 60-80 panels (30 upper, 30 lower for
example) should be enough panels. Note that the lift is presented in an extremely expanded scale.
Drag also uses an expanded scale. Because drag is typically a small number, it is frequently
described in drag counts, where 1 drag count is a CD of 0.0001.
To estimate the limit for an infinitely large number of panels the results can be plotted as a
function of the reciprocal of the number of panels. Thus the limit result occurs as 1/n goes to zero.
Figures 4-14, 4-15, and 4-16 present the results in this manner for the case given above, and with
the pitching moment included for examination in the analysis.
0.012
NACA 0012 Airfoil, = 8
0.010
CD
0.008
0.006
0.004
0.002
0.000
20
40
60
80
No. of Panels
100
120
Figure 4-12. Change of drag with number of panels.

0.980
0.975
CL
0.970
0.965
0.960
0.955
0.950
20
40
60
80
No. of Panels
100
Figure 4-13. Change of lift with number of panels.
2/24/98
120
Panel Methods 4-31
0.012
0.010
0.008
CD 0.006
0.004
0.002
0.000 0
0.01
0.02
0.03
0.04
0.05
0.06
1/n
Figure 4-14. Change of drag with the inverse of the number of panels.
0.980
0.975
CL
0.970
0.965
0.960
0.955
0.950
0.01
0.02
0.03
0.04
0.05
0.06
1/n
Figure 4-15. Change of lift with the inverse of the number of panels.
-0.240
-0.242
-0.244
Cm
-0.246
-0.248
-0.250
0.03
0.04
0.05
0.06
1/n
Figure 4-16. Change of pitching moment with the inverse of the number of panels.
2/24/98
0.01
0.02

The results given in Figures 4-14 through 4-16 show that the program PANEL produces
results that are relatively insensitive to the number of panels once fifty or sixty panels are used, and
by extrapolating to 1/n = 0 an estimate of the limiting value can be obtained.
In addition to forces and moments, the sensitivity of the pressure distributions to changes in
panel density should also be investigated. Pressure distributions are shown in Figures 4-17, 4-18,
and 4-19. The case for 20 panels is given in Figure 4-17. Although the character of the pressure
distribution is emerging, its clear that more panels are required to define the details of the pressure
distribution. The stagnation pressure region on the lower surface of the leading edge is not yet
distinct. The expansion peak and trailing edge recovery pressure are also not resolved clearly.
Figure 4-18 contains a comparison between 20 and 60 panel cases. In this case it appears that the
pressure distribution is well defined with 60 panels. This is confirmed in Figure 4-19, which
demonstrates that it is almost impossible to identify the differences between the 60 and 100 panel
cases. This type of study should (and in fact must) be conducted when using computational
aerodynamics methods.
-4.00
NACA 0012 airfoil, = 8
-3.00
20 panels
-2.00
CP
-1.00
0.00
1.00
0.0
0.2
0.4
0.6
0.8
1.0
x/c
Figure 4-17. Pressure distribution from progrm PANEL, 20 panels.
2/24/98
Panel Methods 4-33
-5.00
-4.00
20 panels
60 panels
-3.00
-2.00
CP
-1.00
0.00
1.00
0.0
0.2
0.4
x/c
0.6
0.8
1.0
Figure 4-18. Pressure distribution from progrm PANEL,

comparing results using 20 and 60 panels.
-5.00
-4.00
60 panels
100 panels
-3.00
-2.00
CP
-1.00
0.00
1.00
0.0
0.2
0.4
x/c
0.6
0.8
Figure 4-19. Pressure distribution from progrm PANEL,

comparing results using 60 and 100 panels.
2/24/98
1.0
Having examined the convergence of the mathematical solution, we investigate the agreement
with experimental data. Figure 4-20 compares the lift coefficients from the inviscid solutions
obtained from PANEL with experimental data from Abbott and von Doenhof.12 Agreement is
good at low angles of attack, where the flow is fully attached. The agreement deteriorates as the
angle of attack increases, and viscous effects start to show up as a reduction in lift with increasing
angle of attack, until, finally, the airfoil stalls. The inviscid solutions from PANEL cannot capture
this part of the physics. The difference in the airfoil behavior at stall between the cambered and
uncambered airfoil will be discussed further in Chapter 10. Essentially, the differences arise due to
different flow separation locations on the different airfoils. The cambered airfoil separates at the
trailing edge first. Stall occurs gradually as the separation point moves forward on the airfoil with
increasing incidence. The uncambered airfoil stalls due to a sudden separation at the leading edge.
An examination of the difference in pressure distributions to be discussed next can be studied to
see why this might be the case.
2.50
2.00
1.50
1.00
0.50
CL, NACA 0012 - PANEL
CL, NACA 0012 - exp. data
0.00
-0.50
-5.0
CL, NACA 4412 - PANEL

CL, NACA 4412 - exp. data
0.0
5.0
10.0
15.0
20.0
25.0
Figure 4-20. Comparison of PANEL lift predictions with experimental data, (Ref. 12).
2/24/98
Panel Methods 4-35

The pitching moment characteristics are also important. Figure 4-21 provides a comparison of
the PANEL pitching moment predictions (about the quarter chord point) with experimental data.
In this case the calculations indicate that the computed location of the aerodynamic center,
dCm / dCL = 0 , is not exactly at the quarter chord, although the experimental data is very close to
this value. The uncambered NACA 0012 data shows nearly zero pitching moment until flow
separation starts to occur. The cambered airfoil shows a significant pitching moment, and a trend
due to viscous effects that is exactly opposite the computed prediction.
0.10
0.05
-0.00
Cm
-0.05
c/4
-0.10
-0.15
Cm, NACA 0012 - PANEL
Cm, NACA 4412 - PANEL
Cm, NACA 0012 - exp. data
Cm, NACA 4412 - exp. data
-0.20
-0.25
-0.30
-5.0
0.0
5.0
10.0
15.0
20.0
25.0
Figure 4-21. Comparison of PANEL moment predictions with experimental data, (Ref. 12).
We do not compare the drag prediction from PANEL with experimental data. In twodimensional incompressible inviscid flow the drag is zero. In the actual case, drag arises from skin
friction effects, further additional form drag due to the small change of pressure on the body due to
the boundary layer (which primarily prevents full pressure recovery at the trailing edge), and drag
due to increasing viscous effects with increasing angle of attack. A well designed airfoil will have a
drag value very nearly equal to the skin friction and nearly invariant with incidence until the
maximum lift coefficient is approached.
In addition to the force and moment comparisons, we need to compare the pressure
distributions predicted with PANEL to experimental data. Figure 4-22 provides one example. The
NACA 4412 experimental pressure distribution is compared with PANEL predictions. In general
2/24/98

the agreement is very good. The primary area of disagreement is at the trailing edge. Here viscous
effects act to prevent the recovery of the experimental pressure to the levels predicted by the
inviscid solution. The disagreement on the lower surface is surprising, and suggests that the angle
of attack from the experiment is not precise.
-1.2
data from NACA R-646
-0.8
-0.4
Cp
-0.0
Predictions from PANEL
0.4
= 1.875
M = .191
Re = 720,000
transition free
0.8
NACA 4412 airfoil
1.2
0.0
0.2
0.4
x/c
0.6
0.8
1.0
1.2
Figure 4-22. Comparison of pressure distribution from PANEL with data.

Panel methods often have trouble with accuracy at the trailing edge of airfoils with cusped
trailing edges, so that the included angle at the trailing edge is zero. Figure 4-23 shows the
predictions of program PANEL compared with an exact mapping solution (FLO36 run at low
Mach number, see Chap. 11) for two cases. Figure 4-23a is for a case with a small trailing edge
angle: the NACA 651-012, while Fig. 4-23b is for the more standard 6A version of the airfoil. The
corresponding airfoil shapes are shown Fig. 4-24.
2/24/98
Panel Methods 4-37
-0.60
-0.60
PANEL
-0.40
FLO36
-0.20
Cp
0.00
FLO36
-0.40
PANEL
-0.20
Cp
0.00
0.20
0.20
NACA 651 -012
= 8.8
0.40
0.60
0.6
0.7
0.8
0.9
1.0
X/C
a. 6-series, cusped TE
0.40
1.1
0.60
0.6
NACA 651 A012

= 8.8
0.7
0.8
0.9
1.0
1.1
X/C
b. 6A-series, finite TE angle
Figure 23. PANEL Performance near the airfoil trailing edge

0.05
y/c
0.00
NACA 65(1)-012
NACA 65A012
-0.05
0.70
0.80
x/c
0.90
1.00
Figure 4-24. Comparison at the trailing edge of 6- and 6A-series airfoil geometries.
This case demonstrates a situation where this particular panel method is not accurate. Is this a
practical consideration? Yes and no. The 6-series airfoils were theoretically derived by specifying a
pressure distribution and determining the required shape. The small trailing edge angles (less than
half those of the 4-digit series), cusped shape, and the unobtainable zero thickness specified at the
trailing edge resulted in objections from the aircraft industry. These airfoils were very difficult to
use on operational aircraft. Subsequently, the 6A-series airfoils were introduced to remedy the
problem. These airfoils had larger trailing edge angles (approximately the same as the 4-digit
series), and were made up of nearly straight (or flat) surfaces over the last 20% of the airfoil. Most
applications of 6-series airfoils today actually use the modified 6A-series thickness distribution.
This is an area where the user should check the performance of a particular panel method.
2/24/98
4.6 Subsonic Airfoil Aerodynamics

Using PANEL we now have a means of easily examining the pressure distributions, and
forces and moments, for different airfoil shapes. In this section we present a discussion of airfoil
characteristics using an inviscid analysis. All the illustrative examples were computed using
program PANEL. We illustrate key areas to examine when studying airfoil pressure distributions
using the NACA 0012 airfoil at 4 angle of attack as typical in Fig. 4-25.
-2.00
Expansion/recovery around leading edge
(minimum pressure or max velocity,
first appearance of sonic flow)
-1.50
Rapidly accelerating flow,

favorable pressure gradient
-1.00
upper surface pressure recovery

(adverse pressure gradient)
CP
-0.50
lower surface
0.00
Trailing edge pressure recovery

Leading edge stagnation point
0.50

1.00
-0.1
0.1
0.3
0.5
x/c
0.7
0.9
1.1
Figure 4-25. Key areas of interest when examining airfoil pressure distributions.
Remember that we are making an incompressible, inviscid analysis when we are using
program PANEL. Thus, in this section we examine the basic characteristics of airfoils from that
point of view. We will examine viscous and compressibility effects in subsequent chapters, when
we have the tools to conduct numerical experiments. However, the best way to understand airfoil
characteristics from an engineering standpoint is to examine the inviscid properties, and then
consider changes in properties due to the effects of viscosity. Controlling the pressure distribution
through selection of the geometry, the aerodynamicist controls, or suppresses, adverse viscous
effects. The mental concept of the flow best starts as a flowfield driven by the pressure distribution
that would exist if there were no viscous effects. The airfoil characteristics then change by the
2/24/98
Panel Methods 4-39

relieving effects of viscosity, where flow separation or boundary layer thickening reduces the
degree of pressure recovery which would occur otherwise. For efficient airfoils the viscous effects
should be small at normal operating conditions.
4.6.1 Overview of Airfoil Characteristics: Good and Bad
In this section we illustrate the connection between the airfoil geometry and the airfoil pressure
distribution. We identify and discuss ways to control the inviscid pressure distribution by changing
the airfoil geometry. An aerodynamicist controls viscous effects by controlling the pressure
distribution. Further discussion and examples providing insight into aerodynamic design are
available in the excellent recent book by Jones.13 A terrific book that captures much of the
experience of the original designers of the NACA airfoils was written by aeronautical pioneer E.P.
Warner.14
Drag: We discussed the requirement that drag should be zero* for this two-dimensional
inviscid incompressible irrotational prediction method when we studied the accuracy of the method
in the previous section. At this point we infer possible drag and adverse viscous effects by
examining the effects of airfoil geometry and angle of attack on the pressure distribution.
Lift: Thin airfoil theory predicts that the lift curve slope should be 2, and thick airfoil theory
says that it should be slightly greater than 2, with 2 being the limit for zero thickness. You can
easily determine how close program PANEL comes to this value. These tests should give you
confidence that the code is operating correctly. The other key parameter is a ZL , the angle at which
the airfoil produces zero lift (a related value is CL0 , the value of CL at = 0).
Moment: Thin airfoil theory predicts that subsonic airfoils have their aerodynamic centers at
the quarter chord for attached flow. The value of Cm 0 depends on the camber. We have seen in Fig.
4-21 that the computed aerodynamic center is not precisely located at the quarter chord. However,
the slope of the moment curve in Fig. 4-21 corresponds to an aerodynamic center location of x/c =
0.2597, which is reasonably close to 0.2500.
Multi-element airfoils are also an important class of airfoils. However, their performance is so
closely connect to the effects of viscosity that the discussion of those airfoils is deferred until
Chapter 10, Viscous Flows in Aerodynamics.
Three-dimensional panel methods can estimate the induced drag.
2/24/98
-5.00
NACA 0012 airfoil
Inviscid calculation from PANEL
-4.00
= 0
-3.00
=4
CP
=8
-2.00
-1.00
0.00
1.00
-0.1
0.1
0.3
0.5
x/c
0.7
0.9
1.1
Figure 4-26. Effect of angle of attack on the pressure distribution.

The starting place for understanding airfoil characteristics is an examination of the angle of
attack effects on an uncambered airfoil. Figure 4-26 presents this effect for the NACA 0012 airfoil.
Here we see the progression from the symmetric zero angle of attack result. The = 0 case
produces a mild expansion around the leading edge followed by a monotonic recovery to the
trailing edge pressure. As the angle of attack increases the pressure begins to expand rapidly
around the leading edge, reaching a very low pressure, and resulting in an increasingly steep
pressure recovery at the leading edge.
The next effect of interest is thickness. Figure 4-27 presents airfoil shapes for NACA 4 digit
sections of 6, 12, and 18 percent thick. The associated basic pressure distributions at zero angle of
attack are shown in Fig. 4-28. Clearly the thicker airfoil produces a larger disturbance, and hence a
lower minimum pressure. However, the 18 percent thick airfoil produces a milder expansion
around the leading edge and a recompression extending further upstream than the thinner airfoils,
especially at the trailing edge.
2/24/98
Panel Methods 4-41
-0.30
NACA 0006 (max t/c = 6%)

NACA 0012 (max t/c = 12%)
NACA 0018 (max t/c = 18%)
-0.20
-0.10
y/c
0.00
0.10
0.20
0.30
-0.1
0.1
0.3
0.5
x/c
0.7
0.9
Figure 4-27. Comparison of NACA 4-digit airfoils of 6, 12, and 18% thicknesses.
-1.00
-0.50
C
0.00
0.50
1.00
-0.1
2/24/98
NACA 0006, = 0
NACA 0012, = 0
NACA 0018, = 0
0.1
0.3
0.5
0.7
0.9
1.1
x/c
Figure 4-28. Effect of airfoil thickness on the pressure distribution at zero lift.
1.1

The effect of thickness in softening the expansion and recompression around the leading edge
is even more evident at an angle of attack. Figure 4-29 shows this effect at a lift coefficient of .48.
The thinnest airfoil shows a dramatic expansion/recompression due to the location of the stagnation
point below the leading edge point, requiring a large expansion around the leading edge which has
a very small radius of curvature. The thicker airfoil results in a significantly milder expansion and
subsequent recompresion.
-3.00
-2.50
NACA 0006, = 4
NACA 0012, = 4
-2.00
NACA 0018, = 4
-1.50
C
-1.00
-0.50
0.00
0.50
1.00
-0.1
0.1
0.3
0.5
x/c
0.7
0.9
1.1
Figure 4-29. Effect of airfoil thickness on the pressure distribution at CL = 0.48.
The next effect to examine is camber. Figure 4-30 compares the shapes of the NACA 0012
and 4412 airfoils. The pressure distributions on the cambered airfoil for two different angles of
attack are shown in Figure 4-31. Note the role of camber in obtaining lift without producing a
leading edge expansion followed by a rapid recompression immediately behind the expansion. This
reduces the possibility of leading edge separation.
2/24/98
Panel Methods 4-43
0.30
0.20
0.10
y/c 0.00
-0.10
NACA 0012 (max t/c = 12%)
NACA 4412 foil (max t/c = 12%)
-0.20
-0.30
-0.1
0.1
0.3
0.5
x/c
0.7
0.9
1.1
Figure 4-30. Comparison of uncambered and cambered NACA 4-digit airfoils.
-2.00
-1.50
NACA 4412, = 0
NACA 4412, = 4
-1.00
CP
-0.50
0.00
0.50
Note: For a comparison of cambered and uncambered
presuure distributions at the same lift, see Fig. 4-32.
1.00-0.1
0.5
0.7
0.9
1.1
x/c
Figure 4-31. Effect of angle of attack on cambered airfoil pressure distributions at low lift.
2/24/98
0.1
0.3

A comparison of the NACA 0012 and NACA 4412 airfoil pressure distributions at the same
lift coefficient is presented for several values of lift in Figures 4-32, 4-33 and 4-34. As the lift
increases, the camber effects start to be dominated by the angle of attack effects, and the dramatic
effects of camber are diminished until at a lift coefficient of 1.43 the pressure distributions start to
look similar.
-2.00
-1.50
NACA 0012, = 4
NACA 4412, = 0
-1.00
-0.50
CP
0.00
0.50
1.00
-0.1
2/24/98
0.1
0.3
0.5
0.7
0.9
x/c
Figure 4-32. Camber effects on airfoil pressure distributions at CL = 0.48.
1.1
Panel Methods 4-45
-4.00
Inviscid calculations from PANEL
NACA 0012, = 8
NACA 4412, = 4
-3.00
-2.00
C
-1.00
0.00
1.00
-0.1
0.1
0.3
0.5
0.7
0.9
x/c
Figure 4-33. Camber effects airfoil pressure distributions at CL = 0.96.
1.1
-6.00
-5.00
NACA 0012, = 12
NACA 4412, = 8
-4.00
-3.00
C
-2.00
-1.00
0.00
1.00
-0.1
2/24/98
0.1
0.3
0.5
0.7
0.9
x/c
Figure 4-34. Camber effects airfoil pressure distributions at CL = 1.43.
1.1
Finally, we examine the effect of extreme aft camber, which was part of the design strategy of
Whitcomb when the so-called NASA supercritical airfoils were developed. This effect can be
simulated using the NACA 6712 airfoil, as shown in Figure 4-35. The resulting pressure
distribution is given in Figure 4-36. Note that the aft camber opens up the pressure distribution
near the trailing edge. Two adverse properties of this type of pressure distribution are the large zero
lift pitching moment and the delayed and then rapid pressure recovery on the upper surface. This
type of pressure recovery is a very poor way to try to achieve a significant pressure recovery
because the boundary layer will separate early. Whitcombs design work primarily improved the
pressure recovery curve.
0.15
y/c 0.05
-0.05
-0.1
0.1
0.3
0.5
x/c
0.7
0.9
1.1
Figure 4-35. Highly aft cambered NACA airfoil, an NACA 6712.

-2.00
-1.50
= -.6 (CL = 1.0)
-1.00
-0.50
CP
0.00
0.50
NACA 6712
1.00
-0.1
2/24/98
0.1
0.3
0.5
0.7
0.9
x/c
Figure 4-36. Example of the use of aft camber to "open up"
the pressure distribution near the trailing edge.
1.1
Panel Methods 4-47

The airfoils used to demonstrate geometry effects on pressure distributions above use
parametric geometry definition formulas developed in the 1930s. More modern airfoils are
available to the aerodynamicist. Unfortunately, to obtain improved performance, the designs were
developed without the use of simple geometric definitions, and are available only as tables of
coordinates. One modern airfoil that extends some of the previous shapes to obtain a high
performance airfoil is the GA(W)-1 airfoil.15 This 17% thick airfoil designed by NASAs Richard
Whitcomb provides better maximum lift and stall characteristics. Figure 4-37 shows the airfoil
shape, and Fig. 4-38 shows the pressure distribution.
0.15
0.10
0.05
y/c 0.00
-0.05
-0.10
0.0
0.2
0.4
x/c
0.6
0.8
1.0
Figure 4-37. GA(W)-1 airfoil, also known as NASA LS(1)-0417.

-1.00
-0.50
Cp
0.00
0.50
GA(W)-1
= 0
1.00
0.0
0.2
0.4
0.6
0.8
1.0
X/C
Figure 4-38. Pressure distribution at zero angle of attack of the GA(W)-1.
2/24/98

Notice that in this case the upper surface pressure distribution reaches a constant pressure
plateau, and then has a moderate pressure recovery. Aft camber is used to obtain lift on the lower
surface and open up the airfoil pressure distribution near the trailing edge in a manner suggested
previously in Fig. 4-36. The area of aft camber on the lower surface is know as the cove region.
If the camber is too extreme here the adverse pressure gradient will be too steep, and the flow will
separate on the lower surface before it separates on the upper surface. Also, this type of pressure
distribution has a significantly higher Cm 0 than conventional airfoil sections.
4.6.2 Geometry and Design
Effects of Shape Changes on Pressure Distributions: So far the examples have
demonstrated global effects of camber and thickness. To develop an understanding of the typical
effects of adding local modifications to the airfoil surface, Exercise 5 provides a framework for the
reader to carry out an investigation analogous to the one for which results were presented in
Section 4.6.1. It is also worthwhile to investigate the very powerful effects that small deflections
of the trailing edge can produce. This reveals the power of the Kutta condition, and alerts the
aerodynamicist to the basis for the importance of viscous effects at the trailing edge.
This approach is extremely educational when implemented in an interactive computer
program, where the aerodynamicist can make shape changes with a mouse and see the effect on the
pressure distribution immediately. An outstanding code that does this has been created by Ilan
Kroo.16 It is called PANDA, originally was for the Macintosh, but now is available for a PC.
Shape for a specified pressure distribution: There is another way that aerodynamicists view
the design problem. The local modification approach described above is useful to make minor
changes in airfoil pressure distributions. Often the aerodynamic designer wants to find the
geometric shape corresponding to a prescribed pressure distribution from scratch. This problem is
known as the inverse problem. This problem is more difficult than the analysis problem. It is
possible to prescribe a pressure distribution for which no geometry exists. Even if the geometry
exists, it may not be acceptable from a structural standpoint. For two-dimensional incompressible
flow it is possible to obtain conditions on the surface velocity distribution that ensure that a closed
airfoil shape exists. Excellent discussions of this problem have been given by Volpe17 and Sloof.18
A two-dimensional panel method has been developed by Bristow.19 Numerical optimization can
also be used to find the shape corresponding to a prescribed pressure distribution.20
2/24/98
Panel Methods 4-49
4.7 Issues in the Problem formulation for 3D flow over aircraft

The extension of panel methods to three dimensions leads to fundamental questions regarding
the proper specification of the potential flow problem for flow over an aircraft. Examples include
the proper treatment of wing tips and the treatment of the wake and fuselage aft of the wing. Hess21
provides an excellent discussion of the problems. In particular, the Kutta condition has to be
reconsidered in three-dimensional flow. There are several aspects to consider. When solving the
flow over a complete aircraft the aerodynamicist has to decide how to model the flow streaming off
the fuselage or tip tank. The Kutta condition applies to distinct edges, and the inviscid model is not
as precise. Many different approaches have been followed. Carmichael and Erickson22 also provide
good insight into the requirements for a proper panel method formulation. Similarly, references 4
and 5 provide good overviews.
Aerodynamics panel methods generally use quadrilateral panels to define the surface. Since
three points determine a plane, the quadrilateral may not necessarily define a consistent flat surface.
In practice, the methods actually divide panels into triangular elements to determine an estimate of
the outward normal. It is important that edges fit so that there is no leakage in the panel model
representation of the surface.
Other practical considerations also require fastidious attention to detail. These include making
sure that the outward surface normal is oriented in the proper direction, that all surfaces are
properly enclosed, and that wakes are properly specified. In some methods wakes are handled
automatically. In other methods the wakes must be precisely specified by the user. This provides
complete control over the simulation, but means that the user must understand precisely what the
problem statement should be. Figure 4-39 shows an example of a panel model including the details
of the wakes. For high lift cases and wakes from one surface streaming near another, wake
deflection must be computed as part of the solution. Figure 4-39 comes from a one week short
course that was given to prospective users of an advanced panel method known as PAN AIR.23
Clearly, to ensure that the problem is properly specified, and to examine the entire flowfield in
detail, a complete graphics capability is required.
There is one other significant difference. Induced drag occurs even in inviscid, irrotational
incompressible flow. However, its calculation by integration of pressures over the surface requires
extreme accuracy, as we saw above for the two-dimension examples. The use of a farfield
momentum approach is much more accurate, and is described in Chap. 5, Drag, An Introduction.
2/24/98
Impermeable Surface
Tail Wake
Nor Shown
Body Wake
Carry-Over
Wakes
Wing Wake
a) wing-body-tail configuration panel scheme with wakes
Impermeable
Surfaces
Body Wake
Tail-Body
Carry-Over
Wake
Tail Wake
Wing Wake
Wing-Body
Carry-Over
Wake
b) details of the wake model required

Figure 4-39. Example of a panel model containing wake model details.
(from a viewgraph presented at a PAN AIR users short course, Ref. 23)
2/24/98
Panel Methods 4-51

4.8 Example applications of panel methods
Many examples of panel methods have been presented. Figure 4-40 shows an example of the
use of a panel model to evaluate the effect of the space shuttle on the Boeing 747. This is a classic
example. Other uses include the simulation of wind tunnel walls, support interference, and ground
effects. Panel methods are also used in ocean engineering. Recent Americas Cup designs have
been dependent on panel methods for hull and keel design. The effects of the free surface can be
treated using panel methods.
Figure 4-40. The space shuttle mounted on a Boeing 747.

One example has been selected to present in some detail. It is an excellent illustration of how a
panel method is used in design, and provides a realistic example of the typical agreement that can
be expected between a panel method and experimental data in a demanding real application. The
work was done by Ed Tinoco and co-workers at Boeing.24 Figure 4-41 shows the modifications
required to modify a Boeing 737-200 to the 737-300 configuration.The panel method was used to
investigate the design of a new high lift system. They used PAN AIR, which is a Boeing
developed advanced panel method.25 25 Figure 4-42 shows the panel method representation of the
airplane.
2/24/98
Figure 4-41. The Boeing 737-300 relative to the model 737-200 (Ref.24).
Figure 4-42. The panel representation of the 737-300 with 15 flap deflection (Ref. 4).
2/24/98
Panel Methods 4-53
An understanding of the wing flowfield for two different takeoff flap settings was desired.
The cases are flaps 15, the normal takeoff setting, and flaps 1, the high altitude, hot day
setting. The work was conducted in concert with the flight test program to provide insight into the
flight test results by providing complete flowfield details not available from the flight test. The
computational models used 1750 panels for flaps 1 and 2900 panels for flaps 15. The modeling
used to simulate this flowfield illustrates typical idealizations employed when applying panels
methods to actual aircraft. Although typical, it is one of the most geometrically complicated
examples ever published.
Figure 4-43 shows the wing leading edge and nacelle. The inboard Krueger flap was actually
modeled as a doublet of zero thickness. The position was adjusted slightly to allow the doublet
sheet to provide a simple matching of the trailing edge of the Krueger and the leading edge of the
wing. These types of slight adjustments to keep panel schemes relatively simple are commonly
used. The outboard leading and trailing edge flap geometries were also modified for use in this
inviscid simulation. Figure 4-44 a) shows the actual and computational flaps 1 geometry. In this
case the airfoil was modeled as a single element airfoil. The flaps 15 trailing edge comparison
between the actual and computational geometry is shown in Fig. 4-44 b). The triple slotted flap
was modeled as a single element flap. At this setting the gap between the forward vane and main
flap is closed, and the gap between the main and aft flap is very small.
Figure 4-43. Inboard wing leading edge and nacelle details (Ref. 24).
2/24/98
a) Comparison of actual and computational wing geometry for the flaps 1 case (Ref. 24).
Actual Geometry
Computational Geometry
b) Actual and computational trailing edge geometry for the flaps 15 case (Ref. 4).
Figure 4-44. Examples of computational modeling for a real application.
Several three-dimensional modeling considerations also required attention. In the flaps 1 case
shown in Fig. 4-45, spanwise discontinuities included the end of the outboard leading edge slat
and trailing edge discontinuities at the back of the nacelle installation (called the thrust gate)
between the inboard and outboard flaps. At the outboard leading edge the edges of the slat and
wing were paneled to prevent leakage. A 0.1 inch gap was left between these surfaces. At the
trailing edge discontinuity a wake was included to model a continuous trailing edge from which a
trailing vortex sheet could be shed.
2/24/98
Panel Methods 4-55
Figure 4-45. Spanwise discontinuity details requiring modeling for flaps 1 case (Ref. 24).
Similar considerations are required for the flaps 15. Here, special care was taken to make sure
that the configuration was closed, and contained no holes in the surface at the ends of the flap
segments.
Another consideration is the nacelle model. This requires the specification of the inlet flow at
the engine face, a model of the strut wake, and both the outer bypass air plume and the primary
wake from the inner hot gas jet. Figure 4-46 provides the details.
Figure 4-46 Nacelle model (Ref. 24).
2/24/98

Complete details of the model are contained in Ref. 24. With the model complete, the solution
was obtained. The spanwise distribution of airfoil section lift coefficients is presented in Figure 447. The first part of the figure shows the results for the flaps 1 case, and the second part of the
figure presents the flaps 15 case. In both cases the jig shape and flight shape including aeroelastic
deformation are included. This is another consideration in making a proper aerodynamic
simulation. In both cases the shape including the deformation under load shows much better
agreement with flight and wind tunnel data. Notice the loss of lift on the wing at the nacelle station,
and the decrease in lift outboard of the trailing edge flap location.
a) flaps 1 case
b) flaps 15 case
Figure 4-47. Spanwise distribution of lift coefficient on the Boeing 737-300 (Ref.24).
2/24/98
Panel Methods 4-57

Figure 4-48 presents the change in section lift coefficient with angle of attack at several span
stations. The agreement between PAN AIR and flight test is better for the flaps 1 case. Viscous
effects are becoming important for the flaps 15 case.
a) flaps 1 case
b) flaps 15 case
Figure 4-48. Comparison of section lift coefficient change with angle of attack(Ref.24)
2/24/98
Figure 4-49 completes this example by presenting the comparison of pressure distributions for
the two cases at four spanwise stations. The flaps 1 case agreement is generally good. Calculations
are presented for both the actual angle of attack, and the angle of attack which matches the lift
coefficient. Matching lift coefficient instead of angle of attack is a common practice in
computational aerodynamics. Considering the simplifications made to the geometry and the
absence of the simulation of viscous effects the agreement is very good. The flaps 15 case starts to
show the problems that arise from these simplifications. This is a good example of the use of a
panel method. It illustrates almost all of the considerations that must be addressed in actual
applications.
a) flaps 1 case
b) flaps 15 case
Figure 4-49. Comparison of pressure distributions between flight and computations for the 737300, solid line is PAN AIR at flight lift, dashed line is PAN AIR at flight angle of attack (Ref. 24).
2/24/98
Panel Methods 4-59
4.9 Using Panel Methods

4.9.1 Common sense rules for panels
Vary the size of panels smoothy
Concentrate panels where the flowfield and/or geometry is changing rapidly
Dont spend more money and time (i.e., numbers of panels) than required
Panel placement and variation of panel size affect the quality of the solution. However,
extreme sensitivity of the solution to the panel layout is an indication of an improperly posed
problem. If this happens, the user should investigate the problem thoroughly.
Panel methods are an aid to the aerodynamicist. You must use the results as a guide to help
you develop your own judgement. (An issue: lawyers often get involved because you frequently
sign an agreement that the code developer is not liable for problems that stem from the use of the
code; the same disclaimer you see with every PC programs).
Remember that the panel method solution is an approximation of the real life problem; an
idealized representation of the flowfield. An understanding of aerodynamics that provides an
intuitive expectation of the types of results that may be obtained, and an appreciation of how to
relate your idealization to the real flow is required to get the most from the methods. This insight
requires experience and study.
4.9.2 What a Panel Method Can't Do
1. Panel methods are inviscid solutions. You will not capture viscous effects except via
user modeling by changing the geometry.
2. Solutions are invalid as soon as the flow develops local supersonic zones
[i.e., Cp < Cpcrit ]. For two-dimensional isentropic flow, the exact value of Cp for critical
flow is:
C pcrit
1 2 1
2 1 + 2 M
=
2 1 +1
M

(4-96)
4.10 Advanced panel methods: What is a Higher Order Panel Method?

So-called higher-order panel methods use singularity distributions that are not constant on
the panel, and may also use panels which are non-planar. Higher order methods were found to be
crucial in obtaining accurate solutions for the Prandtl-Glauert Equation at supersonic speeds. At
supersonic speeds, the Prandtl-Glauert equation is actually a wave equation (hyperbolic), and
2/24/98

requires much more accurate numerical solution than the subsonic case in order to avoid
pronounced errors in the solution (Magnus and Epton25 ). However, subsonic higher order panel
methods, although not as important as the supersonic flow case, have been studied in some detail.
In theory, good results can be obtained using far fewer panels with higher order methods. In
practice the need to resolve geometric details often leads to the need to use small panels anyway,
and all the advantages of higher order panelling are not necessarily obtained. Nevertheless, since a
higher order panel method may also be a new program taking advantage of many years of
experience, the higher order code may still be a good candidate for use.
4.11 Todays standard programs: a brief survey
Panel methods are widely used in the aircraft industry, and have been for a long time.
Comparisons between codes have been made, the most recent comparison being by Margason, et
al.26 In general, all the new professionally-developed codes work well. The selection of a specific
code will likely be based on non-technical considerations. In recent times, several codes have
emerged as the primary ones. The newest is known as PMARC,27 for Panel Method Ames
Research Center. These codes have received the most development effort. We provide a brief
description of the codes a new aerodynamicist will most likely encounter. Specific references are
provided in Tables 4-1 through 4-3.
PAN AIR - Boeing-developed code, funded by a variety of government agencies, and available
through COSMIC (a lease arrangement, about $7000 last time I looked, and export controlled).
This code provides total flexibility, i.e., its really an integral equation solver and not an
aerodynamicists tool per se. It uses higher order panels, and is both subsonic and supersonic. Its
relatively expensive and difficult to run (a PAN AIR user would take months to train, and it would
probably become his primary job).
To effectively use the code good pre- and post- processing systems must be available. Although
Boeing has these systems in place, they were internally developed and are not available outside the
company.
VSAERO - AMI developed (Analytical Mechanics Inc., Frank Dvorak and Brian Maskew). It uses
low order panels and is subsonic only. It also handles general geometries, and includes options to
treat viscous effects and vortex flows. The original NASA version is available through COSMIC.
However, the code has been much further developed by AMI and is for sale by this company. The
price for the current code is about $100K, and they also have a plotting package (OMNIPLOT,
about $20K) available for purchase. This code also requires considerable user training. Support
from AMI is about $10-$15K per year, and site licensing is not available (as of 1990). You pay a
2/24/98
Panel Methods 4-61

large fee for each machine on which you install VSAERO. The business of licensing codes from
developers is an important consideration in computational aerodynamics in the 90s.
The public domain version of this code was obtained by several groups that worked on the design
of the Americas Cup Yacht competitors in the mid-eighties. The code was used for hull and keel
design. One of the modifications that was made was the addition of the free surface representing
the air-water interface (recall that the free surface problem means that the surface displacement is
unknown, and the boundary condition is that a constant pressure exists at the interface).
QUADPAN - Lockheed-developed, and possibly developed at some government labs. Not widely
used by industry outside of Lockheed. This is probably because of availability.
Versions of the Hess Code - further developments of the team at Douglas now led by Hess.
Naturally, Douglas uses this code exclusively. Douglas developed numerous versions under
various government contracts, and it seems to be available mainly at Navy facilities.
Woodward: An old panel method that is sometimes encountered is the code known as the
Woodward or Woodward-Carmichael code. Woodward was a pioneer panel method
developer, and the most likely Woodward code a new aerodynamicist might encounter is a version
of USSAERO, which was developed under NASA contract. Woodwards first methods were
developed while he was at Boeing, and were supported by NASA Ames, primarily for the US SST
program (which was an important national effort in the sixties). Subsequently, Woodward went
into business and continued to develop codes. USSAERO treats both supersonic and subsonic
flow, and a version which incorporates design options Woodward 1-2 is available at VPI.
PMARC -This is the newest panel method code, and was developed at NASA Ames to provide an
extremely flexible method to simulate a wide range of very general geometries. An example is the
simulation of high lift systems and jet exhausts for VSTOL aircraft. The code is a lower order
panel method, and can simulate steady as well as unsteady flow. The wake position can be
obtained as part of the solution. It is being used for underwater applications as well as for aircraft.
This code is also available at VPI.
The history of panel methods is illustrated in the tables. Table 4-1 summarizes some of the
key early methods that were developed. W12SC3 is included because it was a valuable
combination of two early codes, providing significant design capability, particularly at supersonic
speeds. Table 4-2 reviews the extremely active era of the development of advanced methods.
Finally, Table 4-3 provides details on the current production codes likely to be used on current
aerodynamic design and analysis projects.
2/24/98
Table 4 - 1
Comparison of Some Major Panel Method Programs: Early Codes
Originator and
Method Name
Year
Hess and Smith1

1962
(Douglas)
Rubbert2
(vortex lattice)
1964
Panel
Geometry
Source
Type
Doublet
Type
Boundary
Conditions
Restrictions
flat
constant
none
specification
of normal
flow
non-lifting
wings and
bodies only
flat
none
constant
normal flow
planar wings
only
Woodward
(Woodward I)
1967
flat
constant
linear
normal flow
wings must
be planar
Rubbert and
Saaris4
(Boeing A-230)
1968
flat
constant
constant
normal flow
nearly
constant
panel density
Hess I 5
1972
flat
constant
linear
normal flow
wings and
bodies only
USSAERO 6
(Woodward II)
1973
flat
W12SC3 7
(Grumman)
1983
flat
mixed design
and
analysis
Comments
mainly
supersonic,
includes
design &
optimization
subsonic and
supersonic,
analysis only
combines
Woodward
1&2
features
Hess, J.L., and Smith, A.M.O., "Calculation of Nonlifting Potential Flow About Arbitrary
Three-Dimensional Bodies," Douglas Report ES40622, Douglas Aircraft Company, 1962.
Rubbert, P.E., "Theoretical Characteristics of Arbitrary Wings by a Nonplanar Vortex Lattice Method,"
Boeing Report D6-9244, The Boeing Company, 1964.
Woodward, F.A., Tinoco, E.N., and Larsen, J.W., "Analysis and Design of Supersonic Wing-Body
Combinations, Including Flow Properties in the Near Field," Part I - Theory and Application, NASA
CR-73106, 1967.
Rubbert, P.E., and Saaris, G.R., "A General Three-Dimensional Potential Flow Method Applied to
V/STOL Aerodynamics," SAE Paper No. 680304, 1968.
Hess, J.L., "Calculation of Potential Flow About Arbitrary 3-D Lifting Bodies," Douglas Report
MDC-J5679-01, October 1972.
Woodward, F.A., "An Improved Method for the Aerodynamic Analysis of Wing-Body-Tail Configurations
in Subsonic and Supersonic Flow," NASA CR-2228, Parts I and II, 1973.
Mason, W.H., and Rosen, B.S., "The COREL and W12SC3 Computer Programs for Supersonic Wing
Design and Analysis," NASA CR 3676, 1983 (contributions by A. Cenko and J. Malone acknowledged).
from Magnus and Epton, NASA CR 3251, April 1980 (with extensions)
2/24/98
Panel Methods 4-63
Table 4 - 2
Comparison of Some Major Panel Method Programs: Advanced Methods
Originator and
Method Name
Roberts and
Rundle1
Mercer, Weber
and Lesford 2
Year
1973 paraboloidal
1973
Morino and Kuo 3

1974
(SOUSSA)
Johnson and
Rubbert 4
Panel
Geometry
Source
Type
quadratic
Doublet
Type
quadratic
Boundary
Conditions
normal flow
normal flow
in least
planar wings
squares sense
subsonic/
supersonic,
cubic
spanwise,
quadratic
chordwise
none
smooth,
cubic/
quadratic
continuous,
hyperboloidal
constant
constant
potential
linear
quadratic
normal flow
normal flow
arbitrary in
,
Ehlers and
Rubbert 5
(Mach line
paneling)
1976
flat
linear
continuous
quadratic
Ehlers et al. 6
(PAN AIR
"pilot code")
1977
continuous
piecewise
flat
linear
continuous
quadratic
Comments
numerical
integration,
very
expensive
flat
1975 paraboloidal
Restrictions
no thin
configurations
unsteady
planar
wings,
special
paneling
supersonic
flow
subsonic and
supersonic
Roberts, A., and Rundle, K., "Computation of First Order Compressible Flow About Wing-Body
Configurations," AERO MA No. 20, British Aircraft Corporation, February, 1973.
Mercer, J.E., Weber, J.A., and Lesford, E.P., "Aerodynamic Influence Coefficient Method Using
Singularity Splines," NASA CR-2423, May 1974.
Morino, L., and Kuo, C-C, "Subsonic Potential Aerodynamics for Complex Configurations: A General
Theory," AIAA Journal, Vol. 12, No. 2, pp 191-197, February, 1974.
Johnson, F.T., and Rubbert, P.E., "Advanced Panel-Type Influence Coefficient Methods Applied to
Subsonic Flow," AIAA Paper No. 75-50, January 1975.
Ehlers, F.E., and Rubbert, P.E., "A Mach Line Panel Method for Computing the Linearized Supersonic
Flow," NASA CR-152126, 1979.
Ehlers, F.E., Epton, M.A., Johnson, F.T., Magnus, A.E., and Rubbert, P.E., "A Higher Order Panel
Method for Linearized Flow," NASA CR-3062, 1979.
2/24/98
Table 4-3
Comparison of Some Major Panel Method Programs: Production Codes
Originator and
Method Name
Year
Panel
Geometry
Source
Type
Doublet
Type
MCAIR 1
(McDonnell)
1980
flat
constant
quadratic
PAN AIR 2
(Boeing)
1980
continuous
piecewise
flat
Hess II 3
(Douglas)
1981
parabolic
linear
quadratic
normal flow
VSAERO4
(AMI)
1981
flat
constant
constant
exterior and
interior
normal flow
QUADPAN 5
(Lockheed)
1981
flat
constant
constant
PMARC 6
(NASA Ames)
1988
flat
constant
constant
continuous continuous
linear
quadratic
Boundary
Conditions
Restrictions
Comments
design
option
arbitrary in
,
subsonic and
supersonic
subsonic
unsteady,
wake rollup
Bristow, D.R., "Development of Panel Methods for Subsonic Analysis and Design," NASA CR 3234,
1980.
Magnus, A.E., and Epton, M.A., "PAN AIR - A Computer Program for Predicting Subsonic or
Supersonic Linear Potential Flows About Arbitrary Configurations Using a Higher Order Panel Method,"
Volume I - Theory Document (Version 1.0), NASA CR 3251, 1980.
Hess, J.L., and Friedman, D.M., "An Improved Higher Order Panel Method for Three Dimensional
Lifting Flow," Douglas Aircraft Co. Report No. NADC-79277-60, 1981.
Maskew, B., "Prediction of Subsonic Aerodynamic Characteristics: A Case for Lower Order Panel
Methods," AIAA Paper No. 81-0252, 1981.
Coopersmith, R.M., Youngren, H.H., and Bouchard, E.E., "Quadrilateral Element Panel Method
(QUADPAN)," Lockheed-California LR 29671, 1981.
Ashby, D.L., Dudley, M.R., and Iguchi, S.K., "Development and Validation of an Advanced Low-Order
Panel Method," NASA TM 101024, 1988 (also TM 102851, 1990).
2/24/98
Panel Methods 4-65
4.12 Exercises
1. Program PANEL.
a) Obtain a copy of program PANEL and the sample case.
b) Convert PANEL to run on your PC.
c) Run the sample case: NACA 4412, 20 pts. upper, 20 pts. lower, = 4 , and verify
against sample case.
d) Document
i) compile time required on your PC
(cite computer and compiler used)
ii) the execution time for the sample case
iii) the accuracy relative to the sample case.
iv) the exact modifications required to make the code
work on your computer
2. Start work on program PANEL
a) Save a reference copy of the working code!
b) Check convergence with panels (NLOWER+NUPPER must be less than 100
currently). How many panels do you need to get results independent of the number of
panels? What happens to the computer time as the number of panels increases?
c) Check the coordinates generated by the airfoil routine vs. exact (consider using the
NACA 0012, see App. A for geometry definition), including examination of the
coordinates at the trailing edge. This is best done by making a table of exact and
computed values at selected values of x/c. What did you find out?
d) Locate the source strengths, and sum the source strengths x panel lengths to get the
total source strength. Does it sum to zero? Should it?
e) Where is the moment reference center in this code?
Submit an assessment of your findings.
3. Modify program PANEL:
You need a version of PANEL that will allow you to compute the pressure
distribution on arbitrary airfoils. This exercise will give you this capability. Modify
the code to interpolate input airfoil points to the program defined surface points, x/c.
The resulting code should:
a) accept arbitrary airfoil input data
b) echo all the input data on the output
c) generate an output file for post processing
(both for plotting and as the input to a boundary layer code)
d) output Cm about the airfoil quarter chord point.
Hint: Dont alter the panel distribution. The paneling scheme should be independent
of the input distribution of airfoil coordinates. This produces a much more general
and accurate program. This problem is usually solved by finding both the x/c and y/c
2/24/98

values as functions of the airfoil arc length, starting at the lower surface trailing edge.
A spline fit is usually used to interpolate the values along the arc length.
Check your modified code. Run the airfoil you ran previously with internal
coordinate generation. This time use an input file with the same coordinates as
external inputs. Submit a description of your work, and assess your results.
4. Assess the accuracy of incompressible potential flow theory. Run your modified PANEL code
using the airfoil you selected in the exercise in Chap. 1. (What happens if your airfoil has a
trailing edge with finite thickness? What do you do now?)
- compare the computed pressure distribution with the experimental data
- compare the computed force and moment results with the data
(over a range of angles of attack
Turn in a CONCISE report describing the results of your work. Include a plot showing the
pressure distribution comparison, and a plot(s) showing comparison with forces and
moments. What do you conclude about the accuracy of this method?
5. Airfoil design using program PANEL
Take your reference airfoil:
a) add thickness on the bottom (mid chord)- what happens?
b) shave some thickness off the bottom (mid chord) - ?
c) add thickness on the top (mid chord)- what happens?
d) deflect the trailing edge down a couple of degrees
(how sensitive is the airfoil to changes at the TE?)
Hint: use smooth 's to the reference foil employing analytic formulas.
Turn in a CONCISE report comparing the effects on the pressure distribution due to the above
modifications.
6. How good is thin airfoil theory? Compare the thin airfoil Cp for a flat plate with program
PANEL.
Recall thin airfoil theory for an uncambered flat plate:
Cp = 4
(1 x /c )
x /c
a) pick an NACA 0012 airfoil at = 2 and 12 and run PANEL.

b) plot Cp/ as a function of x/c.
c) how many panels do you need to get a converged solution from PANEL?
d) what conclusions do you reach?
2/24/98
Panel Methods 4-67
4.13 References
1
Anderson, John P., Jr., Modern Compressible Flow, 2nd Ed., McGraw-Hill, New York,
1990, pp. 258-269.
2
Hess, J. L., Panel Methods in Computational Fluid Dynamics, Annual Review of Fluid
Mechanics, Vol. 22, 1990, pp. 255-274.
3
Hess, J.L., Linear Potential Schemes, Applied Computational Aerodynamics, P.A. Henne,
ed., AIAA, Washington, 1990. pp.21-36.
4
Erickson, L.L., Panel MethodsAn Introduction, NASA TP-2995, Dec. 1990.
5
Katz, J., and Plotkin, A., Low-Speed Aerodynamics From Wing Theory to Panel Methods,
McGraw-Hill, Inc., New York, 1991.
6
Moran, J. An Introduction to Theoretical and Computational Aerodynamics, John Wiley &

Sons, New York, 1984. pp. 103-112, 118-123, 260-287.
7
Karamcheti, K., Principles of Ideal-Fluid Aerodynamics, John Wiley & Sons, New York,
1966.
8
Ashley, H, and Landahl, M., Aerodynamics of Wings and Bodies, Addison-Wesley, Reading,
1965 (republished in paperback by Dover Publishing).
9
Curle, N., and Davis, H.J., Modern Fluid Dynamics, Volume 1: Incompressible Flow, Van
Nostrand, London, 1968.
10
Houghton, E.L., and Carpenter, P.W., Aerodynamics for Engineering Students, 4th Ed.,
Halsted Press, New York, 1993, pp. 257-265, 203-211.
11
Kuethe, A.M., and Chow, C-Y., Foundations of Aerodynamics, 4th Ed., John Wiley, New
York, 1986, pp. 128-137.
12
Abbott, I.H., and von Doenhoff, A.E., Theory of Wing Sections, Dover, New York, 1959.
13
Jones, R.T., Wing Theory, Princeton University Press, Princeton, 1990.
14
Warner, E.P., Airplane Design: Performance, McGraw-Hill, New York, 1936.

McGhee, Robert J., and Beasley, William D., Low Speed Aerodynamic Characteristics of a
17-Percent-Thick Airfoil Section Designed for General Aviation Applications, NASA TN D7428, 1973.
16
Kroo, Ilan, Aerodynamic Analyses for Design and Education, AIAA Paper 92-2664, June
1992.
17
Volpe, G., Inverse Airfoil Design: A Classical Approach Updated for Transonic Applications,
in Applied Computational Aerodynamics, ed. by P.A. Henne, AIAA Progress in Astronautics
and Aeronautics, Vol. 125, AIAA, New York, 1990, pp. 191-220.
18
Labrujere, Th. E., and Sloof, J.W., Computational Methods for the Aerodynamic Design of
Aircraft Components, Ann. Rev. of Fluid Mech., 1993, Vol. 25, pp.183-214.
19
Bristow, D.R., A New Surface Singularity Method for Multi-Element Airfoil Analysis and
Design, AIAA Paper 76-20, Jan. 1976.
20
Aidala, P.V., Davis, W.H., Jr., and Mason, W.H., Smart Aerodynamic Optimization, AIAA
Paper 83-1863, July 1983.
15
2/24/98
21
Hess, J. L. "The Problem of Three-Dimensional Lifting Potential Flow and Its Solution by
Means of Surface Singularity Distributions", Computer Methods in Applied Mechanics and
Engineering 4 (1974) pp. 283-319.
22
Carmichael, R.L., and Erickson, L.L., "PAN AIR - A Higher Order Panel Method for
Predicting Subsonic or Supersonic Linear Potential Flows About Arbitrary Configurations," AIAA
Paper No. 81-1255, June 1981.
23
PAN AIR Users Class Short Course Presentation Material, 1981.
24
Tinoco, E.N., Ball, D.N., and Rice, F.A. II, PAN AIR Analysis of a Transport High-Lift
Configuration, Journal of Aircraft, Vol. 24, No. 3, March 1987, pp. 181-188.
25
Magnus, A.E., and Epton, M.A., "PAN AIR - A Computer Program for Predicting Subsonic or
Supersonic Linear Potential Flows About Arbitrary Configurations Using a Higher Order Panel
Method," Volume I - Theory Document (Version 1.0), NASA CR 3251, April 1980.
26
Margason, R.J., Kjelgaard, S.O., Sellers, W.L., Moriis, C.E.K., Jr., Walkley, K.B., and
Shields, E.W., Subsonic Panel Methods - A Comparison of Several Production Codes, AIAA
Paper 85-0280, Jan. 1985.
27
Ashby, D.L., Dudley, M.R., Iguchi, S.K., Browne, L., and Katz, J., Potential Flow Theory
and Operation Guide for the Panel Code PMARC, NASA TM 102851, Jan. 1991.
2/24/98
6. Aerodynamics of 3D Lifting Surfaces

through Vortex Lattice Methods
6.1 An Introduction
There is a method that is similar to panel methods but very easy to use and capable of providing
remarkable insight into wing aerodynamics and component interaction. It is the vortex lattice
method (vlm), and was among the earliest methods utilizing computers to actually assist
aerodynamicists in estimating aircraft aerodynamics. Vortex lattice methods are based on
solutions to Laplaces Equation, and are subject to the same basic theoretical restrictions that
apply to panel methods.
As a comparison, vortex lattice methods are:
Similar to Panel methods:
singularities are placed on a surface
the non-penetration condition is satisfied at a number of control points
a system of linear algebraic equations is solved to determine singularity strengths
Different from Panel methods:
Oriented toward lifting effects, and classical formulations ignore thickness
Boundary conditions (BCs) are applied on a mean surface, not the actual surface
(not an exact solution of Laplaces equation over a body, but embodies some
additional approximations, i.e., together with the first item, we find Cp,
not Cpupper and Cplower)
Singularities are not distributed over the entire surface
Oriented toward combinations of thin lifting surfaces
(recall Panel methods had no limitations on thickness).
Vortex lattice methods were first formulated in the late 30s, and the method was first called
Vortex Lattice in 1943 by Faulkner. The concept is extremely simple, but because of its purely
numerical approach (i.e., no answers are available at all without finding the numerical solution of
a matrix too large for routine hand calculation) practical applications awaited sufficient
development of computersthe early 60s saw widespread adoption of the method. A workshop
was devoted to these methods at NASA in the mid 70s.1 A nearly universal standard for vortex
lattice predictions was established by a code developed at NASA Langley (the various versions
were available prior to the report dates):
Margason & Lamar2
Lamar & Gloss3
Lamar & Herbert4,5
3/11/98
1st Langley report

2nd "
"
3rd
"
"
NASA TN D-6142
NASA TN D-7921
NASA TM 83303
1971
1975
1982
6-1
6 - 2 Applied Computational Aerodynamics

Each new version had considerably more capability than the previous version. The final
development in this series is designated VLM4.997. The original codes could handle two lifting
surfaces, while VLM4.997 could handle four. Many, many other people have written vortex
lattice method codes, some possibly even better than the code described in the NASA reports.
But the NASA codes general availability, versatility, and reliability resulted in its becoming a
de-facto standard.
Some of the most noteworthy variations on the basic method have been developed by Lan6
(Quasi-Vortex Lattice Method), Hough7, DeJarnette8 and Frink9. Mook 10 and co-workers at
Virginia Tech have developed vortex lattice class methods that treat flowfields that contain
leading edge vortex type separation (see Section 6.12) and also handle general unsteady motions.
The recent book by Katz and Plotkin 11 contains another variation. At Virginia Tech, Jacob Kay
wrote a code using the method of Katz and Plotkin to estimate stability derivatives, which is
available from the department web page.12
To understand the method, a number of basic concepts must be reviewed. Then we describe
one implementation of the vlm method, and use it to obtain insights into wing and wing-canard
aerodynamics. Naturally, the method is based on the idea of a vortex singularity as the solution
of Laplaces equation. A good description of the basic theory for vortices in inviscid flow and
thin wing analysis is contained in Karamcheti,13 pp. 494-496, 499-500, and 518-534. A good
description of the vortex lattice method is given by Bertin and Smith.14 After the discussion of
wing and wing-canard aerodynamics, an example of a vortex lattice method used in a design
mode is presented, where the camber line required to produce a specified loading is found. The
chapter concludes with a few examples of the extension of vortex lattice methods to treat
situations with more complicated flowfields than the method was originally intended to treat.
6.2 Boundary conditions on the mean surface and the pressure relation
An important difference between vortex lattice methods and panel methods is the method in
which the boundary conditions are handled. Typically, the vortex lattice method uses an
approximate boundary condition treatment. This boundary condition can also be used in other
circumstances to good advantage. This is a good trick applied aerodynamicists should know
and understand. In general, this approach results in the so-called thin airfoil boundary
condition, and arises by linearizing and transferring the boundary condition from the actual
surface to a flat mean reference surface, which is typically a constant coordinate surface.
Consistent with the boundary condition simplification, a simplified relation between the pressure
and velocity is also possible. The simplification in the boundary condition and pressure-velocity
relation provides a basis for treating the problem as a superposition of the lift and thickness
contributions to the aerodynamic results. Karamcheti 13 provides an excellent discussion of this
approach.
3/11/98
Aerodynamics of 3D Lifting Surfaces 6 - 3

To understand the thin airfoil theory boundary condition treatment, we provide an example in
two dimensions. Recall (from Eqn. 2-54) that the exact surface boundary condition for steady
inviscid flow is:
V n = 0
(6-1)
on F(x, y) = 0 = y f (x) . The unit normal vector is n = F(x, y) / F(x, y) and the velocity
field is defined using the notation defined in Fig. 6-1.
Vsin
V cos
Figure 6-1. Basic coordinate system for boundary condition analysis.
Define the velocity components of V as:

V = V +
(6-2)
q(x,y)
123
a disturbance velocity
where q is a disturbance velocity with components u and v. If we assume irrotational flow, then
these components are described in terms of a scalar potential function, u = x and v = y. The
total velocity V then becomes in terms of velocity components:
uTOT = V cos + u(x,y)
vTOT = V sin + v(x, y)
(6-3)
and we can write out the boundary condition as:

F F
V n = (uTOT i + vTOT j)
i+
j = 0
x
y
or
(6-4)
[V cos + u(x, y)] x + [V sin + v(x, y)] y = 0

on F(x,y) = 0, and recalling the relationship between F and f given below Eqn. (6-1):
3/11/98
(6-5)

F
=
x
F
=
y
df (x)
{ y f (x)} =
x
dx
.
{ y f (x) } = 1
y
(6-7)
Substituting for F in Eq.(6-5) we have:
(V cos + u )
df
+ (V sin + v ) = 0
dx
(6-8)
which, solving for v, is:

df
(6-9)
V sin
dx
on y = f(x). Note that v is defined in terms of the unknown u. Thus Eq. (6-9) is a nonlinear
v = ( V cos + u )
boundary condition and further analysis is needed to obtain a useful relation.*

6.2.1 Linearized form of the boundary condition
The relation given above by Eq.(6-9) is exact. It has been derived as the starting point for
the derivation of useful relations when the body (which is assumed to be a thin surface at a small
angle of attack) induces disturbances to the freestream velocity that are small in comparison to
the freestream velocity. Thus we assume: u << V , v << V , and F/x < F/y. Note that this
introduces a bias in the coordinate system to simplify the analysis, a typical consequence of
introducing simplifying assumptions. Consistent with this assumption, the components of the
freestream velocity are:
V cos V
(6-10)
V sin V
and the expression for v in Eq.(6-9) becomes:
v = (V + u)
Dividing by V ,
df
V
dx
v
u df
= 1+
V V dx
(6-11)
(6-12)
the linearized boundary condition is obtained by neglecting u/V compared with unity
(consistent with the previous approximations). With this assumption, the linearized boundary
condition becomes:
v
df
(6-13)
=
on y = f (x) .
V dx
*
Observe that even when the flowfield model is defined by a linear partial differential equation, an
assumption which we have not yet made, the boundary condition can make the problem nonlinear.
3/11/98

This form of the boundary condition is not valid if the flow disturbance is large compared to
the freestream velocity {for aerodynamically streamlined shapes this is usually valid everywhere
except at the leading edge of the airfoil, where a stagnation point exists (u = -V ) and the slope
is infinite (df/dx = )}. In practice, a local violation of this assumption leads to a local error.
Thus, if the details of the flow at the leading edge are not important to the analysis, which
surprisingly is often the case, the linearized boundary condition can be used.
6.2.2 Transfer of the boundary condition
Although Eqn. (6-13) is linear, its hard to apply because it is not applied on a coordinate
line.* We now use a further approximation of this relation to get the useful form of the linearized
boundary condition. Using a Taylor series expansion of the v component of velocity about the
coordinate axis we obtain the v velocity on the surface:
v{x, y = f(x)} = v(x,0) + f (x)
v
+... .
y y=0
(6-14)
For the thin surfaces under consideration, f(x) is small, and because the disturbances are assumed
small, v/y is also small. For example, assume that v and v/y are the same size, equal to 0.1,
and df/dy is also about 0.1. The relation between v on the airfoil surface and the axis is:
v{x, y = f(x)} = (.1) +(.1)(.1) = .1 + .01
{ .
(6-15)
neglect
Neglecting the second term, we assume:

v{x, f (x)} v(x, 0).
(6-16)
We now apply both the upper and lower surface boundary conditions on the axis y = 0, and
distinguish between the upper and lower surface shapes by using:
f = fu
f = fl
on the upper surface

.
on the lower surface
(6-17)
Using Eq. (6-17), we write the upper and lower surface boundary conditions as:
v x,0 +
V
df
= u ,
dx
up
v x,0
V
dfl
.
dx
(6-18)
low
The simplification introduced by applying boundary conditions on constant coordinate surfaces justifies
the use of rather elaborate tranformations, which will be dicussed in more detail in Chap. 9, Geometry and
Aerodynamics.
3/11/98

These are the linearized and transferred boundary conditions. Frequently, these boundary
conditions result in a surprisingly good approximation to the flowfield, even in transonic and
supersonic flow.
6.2.3 Decomposition of boundary conditions to camber/thickness/alpha
Further simplification and insight can be gained by considering the airfoils in terms of the
combination of thickness and camber, a natural point of view. We thus write the upper and lower
surface shapes in terms of camber, fc, and thickness, ft, as:
fu = fc + ft
fl = fc ft
(6-19)
and the general problem is then divided into the sum of three parts as shown in Fig. 6-2.
General
Thickness
Camber
airfoil
at = 0
at = 0
at
Figure 6-2. Decomposition of a general shape at incidence.
Flat plate
at
The decomposition of the problem is somewhat arbitrary. Camber could also be considered
to include angle of attack effects using the boundary condition relations given above, the sign is
the same for both the upper and lower surface. The aerodynamicist must keep track of details for
a particular problem. To proceed further, we make use of the basic vortex lattice method
assumption: the flowfield is governed by a linear partial differential equation (Laplaces
equation). Superposition allows us to solve the problem in pieces and add up the contributions
from the various parts of the problem. This results in the final form of the thin airfoil theory
boundary conditions:
v x,0+
V
v
x,0
V
df c dft
+
dx dx
up
.
=
(6-20)
df c dft
dx dx
low
The problem can be solved for the various contributions and the contributions are added together
to obtain the complete solution. If thickness is neglected the boundary conditions are the same
for the upper and lower surface.
3/11/98

6.2.4 Thin airfoil theory pressure relation
Consistent with the linearization of the boundary conditions, a useful relation between the
pressure and velocity can also be obtained. For incompressible flows, the exact relation between
pressure and velocity is:
V 2
C p =1
V
(6-21)
and we expand the velocity considering disturbances to the freestream velocity using the
approximations discussed above:
V 2 = (V cos + u) + ( V sin + v)
2
V
Expanding:
V 2 = V2 + 2Vu + u2 + (V) + 2Vv + v 2
(6-22)
V2
u
u2
v v2
2
=1
+
2
+
+
+
2
+
.
V V2
V V2
V2
(6-23)
and dividing by V2 we get:
Substituting into the Cp relation, Eqn. (6-21), we get:
u
u2
v v2
2
CP = 1 1 + 2
+
+ + 2
+
V V2
V V2
= 1 1 2
u2
(6-24)
v2
u
v
2 2 2
V V
V V2
and if , u/V and v/V are << 1, then the last four terms can be neglected in comparison with
the third term. The final result is:
C p = 2
u
.
V
(6-25)
This is the linearized or thin airfoil theory pressure formula. From experience gained
comparing various computational results, Ive found that this formula is a slightly more severe
restriction on the accuracy of the solution than the linearized boundary condition. Equation (625) shows that under the small disturbance approximation, the pressure is a linear function of u,
and we can add the Cp contribution from thickness, camber, and angle of attack by
superposition. A similar derivation can be used to show that Eq. (6-25) is also valid for
compressible flow up to moderate supersonic speeds.
3/11/98

6.2.5 Delta Cp due to camber/alpha (thickness cancels)
Next, we make use of the result in Eq. (6-25) to obtain a formula for the load distribution on
the wing:
(6-26)
Cp = Cp
Cp
.
LOWER
UPPER
Using superposition, the pressures can be obtained as the contributions from wing thickness,
camber, and angle of attack effects:
C pLOWER = C pTHICKNESS + C pCAMBER + C pANGLE OF ATTACK
C pUPPER = C pTHICKNESS C pCAMBER C pANGLE OF ATTACK
so that:
Cp = CpTHICKNESS + CpCAMBER + CpANGLE OF ATTACK
C pTHICKNESS C pCAMBER C pANGLE OF ATTACK .
= 2 C pCAMBER + C pANGLE OF ATTACK
(6-27)
(6-28)
Equation (6-28) demonstrates that for cases where the linearized pressure coefficient relation is
valid, thickness does not contribute to lift to 1st order in the velocity disturbance!
The importance of this analysis is that we have shown:
1. how the lifting effects can be obtained without considering thickness, and
2. that the cambered surface boundary conditions can be applied on a flat coordinate
surface, resulting in an easy to apply boundary condition.
The principles demonstrated here for transfer and linearization of boundary conditions can be
applied in a variety of situations other than the application in vortex lattice methods. Often this
idea can be used to handle complicated geometries that cant easily be treated exactly.
The analysis here produced an entirely consistent problem formulation. This includes the
linearization of the boundary condition, the transfer of the boundary condition, and the
approximation between velocity and pressure. All approximations are consistent with each other.
Improving one of these approximations without improving them all in a consistent manner may
actually lead to worse results. Sometimes you can make agreement with data better, sometimes it
may get worse. You have to be careful when trying to improve theory on an ad hoc basis.
3/11/98
6.3 Vortex Theorems

In using vortex singularities to model lifting surfaces, we need to review some properties of
vortices. The key properties are defined by the so-called vortex theorems. These theorems are
associated with the names of Kelvin and Helmholtz, and are proven in Karamcheti.13 Three
important results are:
1. Along a vortex line (tube) the circulation, , is constant.
2. A vortex filament (or line) cannot begin or end abruptly in a fluid. The vortex line
must i) be closed, ii) extend to infinity, or iii) end at a solid boundary.
Furthermore, the circulation, , about any section is the vortex strength.
3. An initially irrotational, inviscid flow will remain irrotational.
Related to these theorems we state an important result:
A sheet of vortices can support a jump in tangential velocity [i.e. a force],
while the normal velocity is continuous. This means you can use a vortex sheet
to represent a lifting surface.
6.4 Biot-Savart Law
We know that a two-dimensional vortex singularity satisfies Laplaces equation (i.e. a point
vortex):
(6-29)
V=
e
2 r
where V is the irrotational vortex flow illustrated in Fig. 6-3.
V
1/r
V
Vortex normal
to page
a) streamlines
b) velocity distribution
Figure 6-3. The point vortex.
What is the extension of the point vortex idea to the case of a general three-dimensional
vortex filament? Consider the flowfield induced by the vortex filament shown in Fig. 6-4, which
defines the nomenclature.
3/11/98
vortex
filament
dl
q
rpq
p
Figure 6-4. General three-dimensional vortex filament.

The mathematical description of the flow induced by this filament is given by the BiotSavart law (see Karamcheti,13 pages 518-534). It states that the increment in velocity dV at a
point p due to a segment of a vortex filament dl at q is:
dl rpq
(6-30)
dVp =
.
4 r 3
pq
To obtain the velocity induced by the entire length of the filament, integrate over the length
of the vortex filament (or line) recalling that is constant. We obtain:
r r
r
dl rpq

(6-31)
Vp =
r 3 .
4
rpq
To illustrate the evaluation of this integral we give the details for several important
examples. The vector cross product definition is reviewed in the sidebar below. Reviewing the
cross product properties we see that the velocity direction dV p induced by the segment of the
vortex filament dl is perpendicular to the plane defined by dl and rpq, and its magnitude is
computed from Eq. (6-31).
Case #1: the infinitely long straight vortex.
In the first illustration of the computation of the induced velocity using the Biot-Savart Law, we
consider the case of an infinitely long straight vortex filament. The notation is given in Fig. 6-5.
3/11/98
A review: the meaning of the cross product. What does a x b mean? Consider the following
sketch:
c
b
a
Here, the vectors a and b form a plane, and the result of the cross product operation is a vector c,
where c is perpendicular to the plane defined by a and b. The value is given by:
c = a b = a b sin e
and e is perpendicular to the plane of a and b. One consequence of this is that if a and b are
parallel, then a x b = 0.
a b = area of the parallelogram
Also:
and
i
a b = Ax
Bx
j
Ay
By
k
Az = Ay Bz Az By i ( Ax Bz Az Bx ) j + Ax By Ay Bx k
Bz
dl
h
rpq
-
.
Figure 6-5. The infinitely long straight vortex.
3/11/98

Now consider the numerator in Eq. (6-30) given above using the definition of the cross
product:
V
(6-32)
dl rpq = dl rpq sin
V
so that the entire expression becomes
dVp =
=
dl rpq sin V
3
4
V
rpq
V sin dl
2
4 V rpq
(6-33)
Next, simplify the above expressions so they can be readily evaluated. Use the nomenclature
shown in Fig. 6-6.
p
r pq
h
l1
q
l
Figure 6-6. Relations for the solution of the Biot-Savart Law.

Using the notation in Fig. 6-6 to find the relations for rpq and dl. First we see that:
h = rpq sin
h
or
rpq =
sin
h
and
l1 l =
= hcot .
tan
Next look at changes with . Start by taking the differential of Eq. (6-36):
(6-34)
(6-35)
(6-36)
h
d(l1 l) = d
= d(hcot)
tan
dl = h d(cot )
which is
(6-37)
dl
d
=h
(cot) = hcosec2
d
d
and we can now write dl in terms of d :

dl = hcosec2 d
3/11/98
(6-38)

so that the Biot-Savart Law gives:
dVp =
V sin dl
2
4 V rpq
V sin hcosec2 d
.
2
4 V
h
sin
sin
=
d
(direction understood)
4 h
=
Thus we integrate:
Vp =
sin
d =
4
h
4 h
(6-39)
sin d
(6-40)
=0
where here the limits of integration would change if you were to consider a finite straight length.
Carrying out the integration:
Vp =
[cos ] =
=0
4 h
=
[(1) (1)]
4 h
and:
(6-41)
Vp =
2 h
which agrees with the two dimensional result.
Although a vortex cannot end in a fluid, we can construct expressions for infinitely long
vortex lines made up of a series of connected straight line segments by combining expressions
developed using the method illustrated here. To do this we simply change the limits of
integration. Two cases are extremely useful for construction of vortex systems, and the formulas
are given here without derivation.
Case #2: the semi-infinite vortex.
This expression is useful for modeling the vortex extending from the wing to downstream
infinity.
+
dl
0
h
p
Vp =
Figure 6-7. The semi-infinite vortex.
3/11/98
(1 + cos 0 )e
4h
(6-42)

Case #3: the finite vortex.
This expression can be used to model the vortex on a wing, and can be joined with two
semi-infinite vortices to form a vortex of infinite length, satisfying the vortex theorems.
dl
h
p
Vp =
(cos1 cos2 )e
4h
(6-43)
Figure 6-8. The finite vortex.

Systems of vortices can be built up using Eqns. (6-42) and (6-43) and the vortex theorems.
The algebra can become tedious, but there are no conceptual difficulties.
6.5 The Horseshoe Vortex
There is a specific form of vortex used in the traditional vortex lattice method. We now use
the expressions developed from the Biot-Savart Law to create a horseshoe vortex, which
extends from downstream infinity to a point in the field A, then from point A to point B,
and another vortex from point B downstream to infinity. The velocity induced by this vortex is
the sum of the three parts. The basic formulas were presented in the previous section. Here we
extend the analysis of the previous section, following the derivation and notation of Bertin and
Smith.14 In particular, the directions of the induced flow are made more precise. Our goal is to
obtain the expression for the velocity field at a general point in space (x,y,z) due to the specified
horseshoe vortex.
To create a horseshoe vortex we will use three straight line vortices: one finite length vortex
and two semi-infinite vortices. This is illustrated in Fig. 6-9. To start our analysis, we rewrite the
Biot-Savart Law in the Bertin and Smith notation,14 which is given in Fig. 6-10.
3/11/98

B
Figure 6-9. The horseshoe vortex.
dl
r2
1
r1
rp
C
ro = AB
r1 = AC
r2 = BC
Figure 6-10. Nomenclature for induced velocity calculation.

The next step is to relate the angles to the vector definitions. The definition of the dot product
is used:
(6-44)
A B = A B cos
so that:
r r
cos1 = 0 1
r0 r1
(6-45)
.
r0 r2
cos2 =
r0 r2
Next, we put these relations into the formula for the finite length vortex segment formula for the
induced velocity field given above in Eq. (6-43).
(6-46)
Vp =
(cos1 cos 2 )e
4 rp
so that, substituting using the definition in Eq. (6-44), we get
Vp =
3/11/98
r0 r0 r1 r0 r2 r1 r2
4 r1 r2 r0 r1 r0 r2 r1 r2
(6-47)
Bertin and Smith14 use the relation:

r r
r r
rp = 1 2 rp = 1 2
r0
r0
and we need to demonstrate that this is true. Consider the parallelogram ABCD shown in the
sketch:
C
D
r
0
rp
r
2
A
r
1
by definition: r1 r2 = Ap , is the area of the parallelogram.

Similarly, the area of ABC is:
1
bh
2
1
= ro rp
2
The total area of the parallelogram can be found from both formulas, and by equating these two
areas we obtain the expression we are trying to get:
AABC =
2AABC = Ap
ro rp = r1 r2
which can be rewritten to obtain the expression given by Bertin and Smith:14
r r
rp = 1 2
r0
Collecting terms and making use of the vector identity provided in the sidebar, we obtain the
Bertin and Smith statement of the Biot-Savart Law for a finite length vortex segment:
Vp =
r1 r2 r1 r2
r
.
4 r1 r2 2 0 r1 r2
(6-48)
For a single infinite length horseshoe vortex we will use three segments, each using the
formula given above. The nomenclature is given in the sketch below. The primary points are the
3/11/98

connecting points A and B. Between A and B we use a finite length vortex which is considered a
bound vortex, and from A to infinity and B to infinity we define trailing vortices that are
parallel to the x-axis.
The general expression for the velocity at a point x, y, z due to a horseshoe vortex at (x1n,
y1n, z1n), (x2n,y2n, z2n) with trailing vortices parallel to the x-axis is (from Bertin and Smith14):
V = VAB + VA + V B
(6-49)
where the total velocity is the sum of the contributions from the three separate straight line
vortex segments making up the horseshoe vortex, as shown in Fig. 6-11.
z
y
B
n
n
A
Figure 6-11. Definitions for notation used in induced velocity expressions.

The corner points of the vortex, A and B , are arbitrary, and are given by:
A = A(x1n , y1n , z1n )
.
B = B(x 2n , y2n ,z2n )
(6-50)
We now write the expression for the velocity field at a general point in space (x,y,z) due to
the horseshoe vortex system. At C(x,y,z) find the induced velocity due to each vortex segment.
Start with AB, and use Fig. 6-12.
Now we define the vectors as:
3/11/98

r0 = ( x2n x1n )i + ( y2n y1n ) j + (z2n z1n )k
r1 = ( x x1n )i + ( y y1n ) j + (z z1n )k
(6-51)
r0 = ( x x2n )i + (y y2n ) j + (z z2n )k
r
0
r
2
A
r
1
C(x,y,z)
Figure 6-12. Velocity induced at Point C due to the vortex between A and B.
and we simply substitute into:
VC =
r1 r2
r1
r2
r
.
0
0
4 r1 r2 2
r1
r2
42444
3
1
424
3 144
(6-52)
Considering the bound vortex on AB first we obtain,

VAB = n
4
(6-53)
where and are lengthy expressions. By following the vector definitions, Eqn. (6-53) can be
written in Cartesian coordinates. The vector is:
=
r1 r2
r1 r2
[( y y1n )( z z2n ) ( y y2n ) (z z1n )]i
[( x x1n )(z z 2n ) ( x x2n )( z z1n )] j
+
x
x
y
x
y
y
k
(
)
(
)
(
)
(
)
[
]
1n
2n
2n
1n
.
=
( y y )( z z ) ( y y )( z z ) 2
1n
2n
2n
1n ]
[
2
+ [( x x1n )(z z2n ) ( x x2n )(z z1n )]
+ [( x x1n )( y y2n ) ( x x2n )( y y1n )]2
3/11/98
(6-54)

The scalar portion of the expression, , is:
= ro
r1
r
ro 2
r1
r2
[( x2n x1n )( x x1n ) + ( y2n y1n )( y y1n ) + (z2n z1n )(z z1n )]
[( x2n x1n )( x x2n ) + ( y2n y1n )( y y2n ) + (z 2n z1n ) (z z 2n )]
( x x1n )
+ ( y y1n ) + ( z z1n )
2
(6-55)
( x x2n )2 + ( y y2n )2 + ( z z2n )2
We find the contributions of the trailing vortex legs using the same formula, but redefining
the points 1 and 2. Then, keeping the 1 and 2 notation, define a downstream point, 3 and let x3
go to infinity. Thus the trailing vortex legs are given by:
n (z z1n ) j + ( y1n y)k

VA =
4 (z z1n )2 + ( y1n y)2
1.0 +
(x x1n )2 + (y y1n )2 + (z z1n )2

x x1n
(6-56)
and
n ( z z2n ) j + ( y2n y)k

VB =
4 ( z z2n )2 + (y2n y )2
1.0 +
.(6-57)
( x x 2n )2 + ( y y2n )2 + (z z2n )2
x x2n
Note that n is contained linearly in each expression, so that the expression given above can
be arranged much more compactly by using (6-49) with (6-52), (6-56), and (6-57) as:
Vm = Cmnn
(6-58)
and Cmn is an influence coefficient for the nth horseshoe vortex effect at the location m,
including all three segments.
Now that we can compute the induced velocity field of a horseshoe vortex, we need to
decide where to place the horseshoe vortices to represent a lifting surface.
6.6 Selection of Control Point/Vortex Location
Since we are interested in using the horseshoe vortex defined above to represent a lifting
surface, we need to examine exactly how this might be done. In particular: where do you locate
the vortex, and where do you locate a control point to satisfy the surface boundary condition?
Tradition has been to determine their locations by comparison with known results. In particular,
we use two dimensional test cases, and then apply them directly to the three dimensional case.
3/11/98

An alternate distribution based on numerical properties of quadrature formulas has been derived
by Lan. Section 6.11 will demonstrate the use of his vortex/control point locations in the inverse
case, where the pressure is given, and the shape of the surface is sought.
a) Simplest Approach: A Flat Plate
Consider representing the flow over a flat plate airfoil by a single vortex and control point.
Comparing with the known result from thin airfoil theory we determine the spacing between the
vortex and control point which produces a lift identical with the thin airfoil theory value.
Consider the flat plate as sketched in Fig. 6-13.
control point
X
b
a
r
c
Figure 6-13. The notation for control point and vortex location analysis.
The velocity at the control point, cp, due to the point vortex is:
vcp =
.
2r
(6-59)
The flow tangency condition was given above as:

v BC dfc
=
V
dx
(6-60)
v BC
=
V
(6-61)
v BC = V .
(6-62)
= V
2r
(6-63)
and ignoring camber:
or:
Now, we equate vBC and vcp:
resulting in the expression for :
3/11/98

=
.
2rV
(6-64)
To make use of this relation, recall the Kutta-Joukowsky Theorem:

L = V
(6-65)
L = 12 V2 c{ 2
.
123 S {
C
(6-66)
and the result from thin airfoil theory:
ref
Equate the expressions for lift, Eqns. (6-65) and (6-66) and substitute for using the
expression in Eqn. (6-64) given above:
V = 12 V2c2
= 12 V2c2
1=
2rV
(6-67)
1c
2r
and finally:
c
(6-68)
.
2
This defines the relation between the vortex placement and the control point in order for the
single vortex model to reproduce the theoretical lift of an airfoil predicted by thin airfoil theory.
Since the flat plate has constant (zero) camber everywhere, this case doesnt pin down placement
(distance of the vortex from the leading edge) completely. Intuitively, the vortex should be
located at the quarter-chord point since that is the location of the aerodynamic center of a thin
flat plate airfoil. The next example is used to determine the placement of the vortex.
r=
b) Determine placement of the vortex using parabolic camber model.

Rewrite the velocity at the control point due to the point vortex in a little more detail, where
a denotes the location of the vortex, and b the location of the control point:
vcp =
2 ( a b )
(6-69)
and the boundary condition remains the same:

df
v BC = V c .
dx
3/11/98
(6-70)

Equating the above expressions (and dividing by V :):
df
= c .
2 ( a b)V dx
(6-71)
For parabolic camber,

x
fc (x) = 4 (c x )
c
(6-72)
dfc (x)
x
= 4 1 2
dx
c
(6-73)
we have the slope,
so that:
x
= 4 1 2 .
2 ( a b)V
c
(6-74)
Now use the result from thin airfoil theory:

L = 12 V2 c2 ( + 2 )
and substitute for the lift from the Kutta-Joukowsky theorem. We thus obtain an expression for
the circulation of the vortex in terms of the angle of attack and camber:
= Vc ( + 2 ) .
(6-75)
Substitute for from (6-75) into (6-74), and satisfy the boundary condition at x = b:
Vc ( + 2 )
b
= 4 1 2
c
2 ( b a )V
or:
1 c
b
( + 2 ) = 4 1 2 .

c
2 b a
(6-76)
To be true for arbitrary , , the coefficients must be equal:

1
c

= 1
2 b a
b
c
= 4 1 2
b a
(6-77)
and we solve for a and b. The first relation can be solved for (b-a):
c
(6-78)
(b a) =
2
and we obtain the same results obtained above, validating our previous analysis ( r = c/2).
3/11/98

Now, rewrite the 2nd equation:
b
c = 41 2 (b a)
c
or:
b c
c = 41 2
c 2
(6-79)
and solve for b/c:
b
1 = 21 2
c
1
b
=1 2
c
2
b
1 3
=1 + =
c
2 2
b 13 3
=
=
c 22 4,
(6-80)
and use this to solve for a/c starting with Eqn. (6-78) :
(b a) =
c
2
or:
b a 1
=
c c 2
(6-81)
and:
a b 1 3 1
= = .
c c 2 4 2
Finally:
a 1
= .
c 4
(6-82)
Thus the vortex is located at the 1/4 chord point, and the control point is located at the 3/4
chord point. Naturally, this is known as the 1/4 - 3/4 rule. Its not a theoretical law, simply a
placement that works well and has become a rule of thumb. It was discovered by Italian Pistolesi.
Mathematical derivations of more precise vortex/control point locations are available (see Lan6 ),
but the 1/4 - 3/4 rule is widely used, and has proven to be sufficiently accurate in practice.
3/11/98

To examine the use of these ideas we present a two-dimensional example. The airfoil is
divided into a number of equal size panels. Each panel has a vortex at the quarter chord point and
the non-penetration condition is satisfied at the three-quarter chord point. We use the example to
illustrate the accuracy of the classical thin airfoil theory formulation. In Fig. 6-14, we compare
the results obtained for a 5% circular arc camber.* Three solutions are presented. The linear
theory curve uses classical thin airfoil theory with results obtained satisfying the boundary
condition on the mean surface. This is compared with numerical results for the case where the
boundary condition is applied exactly on the camber line, and the result obtained applying the
boundary condition using the approximate method described above. The difference between
placing the vortex on the actual camber surface and satisfying the boundary condition on the
actual surface, and the more approximate traditional approach of locating the vortex and control
point on the mean surface is extremely small.
Cp bc's on camber line

Cp bc's on axis
Cp linear theory
0.90
0.80
0.70
Cp
0.60
0.50
0.40
5% Circular arc camber line
=0
40 panels
0.30
0.20
0.10
0.0
0.2
0.4
x/c
0.6
0.8
1.0
Figure 6-14. Comparison in 2D of the 1/4-3/4 rule for vortex-control point locations with
linear theory, and including a comparison between placing the vortex and
control point on the camber line or on the axis.
*
A relatively large camber for a practical airfoil, the NACA 4412 example we used in Chapter 4 was an extreme
case, and it has 4% camber.
3/11/98

6.7 The Classical Vortex Lattice Method
There are many different vortex lattice schemes. In this section we describe the classical
implementation. Knowing that vortices can represent lift from our airfoil analysis, and that one
approach is to place the vortex and then satisfy the boundary condition using the 1/4 - 3/4 rule,
we proceed as follows:
1. Divide the planform up into a lattice of quadrilateral panels, and put a horseshoe
vortex on each panel.
2. Place the bound vortex of the horseshoe vortex on the 1/4 chord element line of
each panel.
3. Place the control point on the 3/4 chord point of each panel at the midpoint in the
spanwise direction (sometimes the lateral panel centroid location is used) .
4. Assume a flat wake in the usual classical method.
5. Determine the strengths of each n required to satisfy the boundary conditions by
solving a system of linear equations. The implementation is shown schematically
in Fig. 6-15.
y
bound vortices
control points
X
X
X
X
X
X
X
X
x
Trailing vortices extend to infinity
Figure 6-15. The horseshoe vortex layout for the classical vortex lattice method.
Note that the lift is on the bound vortices. To understand why, consider the vector statement
of the Kutta-Joukowski Theorem, F = V . Assuming the freestream velocity is the primary
contributor to the velocity, the trailing vortices are parallel to the velocity vector and hence the
force on the trailing vortices are zero. More accurate methods find the wake deformation
required to eliminate the force in the presence of the complete induced flowfield.
3/11/98

Next, we derive the mathematical statement of the classical vortex lattice method described
above. First, recall that the velocity induced by a single horseshoe vortex is
Vm = Cm,nn .
(6-60)
This is the velocity induced at the point m due to the nth horseshoe vortex, where Cm,n is a
vector, and the components are given by Equations 6-54, 6-58 and 6-59.
The total induced velocity at m due to the 2N vortices (N on each side of the planform) is:
Vmind = umind i + vmind j+ wmind k =
2N
Cm,nn .
(6-83)
n=1
The solution requires the satisfaction of the boundary conditions for the total velocity, which
is the sum of the induced and freestream velocity. The freestream velocity is (introducing the
possibility of considering vehicles at combined angle of attack and sideslip):
V = V cos cosi V sin j + V sin cos k
(6-84)
so that the total velocity at point m is:
Vm = (V cos cos + umind ) i+ V sin + vmind j + (V sin cos + wmind )k . (6-85)

The values of the unknown circulations, n, are found by satisfying the non-penetration
boundary condition at all the control points simultaneously. For steady flow this is
V n = 0
(6-1)
F(x, y,z) = 0.
(6-86)
F
= V F = 0 .
F
(6-87)
where the surface is described by

Equation (6-1) can then be written:
V
This equation provides freedom to express the surface in a number of forms. The most general
form is obtained by substituting Eqn. (6-85) into Eqn. (6-87) using Eqn. (6-83). This can be
written as:
[(V cos cos + um
ind
)i + V sin + v mind j + (V sin cos + wmind )k

(6-88)
F
F
F
x i + y j + z k = 0
or:
3/11/98
2N
2N
2N
(V cos cos + Cm,n n )i + V sin + Cm,n n j + (V sin cos + Cm,n n )k

i
j
k
n=1
n=1
n=1
F
F
F
x i + y j + z k = 0
(6-89)
Carrying out the dot product operation and collecting terms:
2N
2N
F
F
(V cos cos + Cm,ni n )+
V sin + Cm,nj n +
x
y
n=1
n=1
2N
F
(V sin cos + Cm,nk n ) = 0 .
z
n=1
(6-90)
Recall that Eqn. (6-90) is applied to the boundary condition at point m. Next, we collect terms to
clearly identify the expression for the circulation. The resulting expression defines a system of
equations for all the panels, and is the system of linear algebraic equations that is used to solve
for the unknown values of the circulation distribution. The result is:
2N
x Cm,ni
n=1
F
F
F
F
F
Cm,nj +
Cm,nk n = V cos cos
sin
+ sin cos
y
z
x
y
z
m = 1,...2N (6-91)
This is the general equation used to solve for the values of the circulation. It is arbitrary,
containing effects of both angle of attack and sideslip (if the vehicle is at sideslip the trailing
vortex system should by yawed to align it with the freestream).
If the surface is primarily in the x-y plane and the sideslip is zero, we can write a simpler
form. In this case the natural description of the surface is
z = f(x, y)
(6-92)
F(x, y,z) = z f(x, y) = 0 .
(6-93)
F
f F
f F
= ,
= ,
= 1.
x
x y
y z
(6-94)
and
The gradient of F becomes
Substituting into the statement of the boundary condition, Eqn. (6-91), we obtain:
2N
Cm,n k x Cm,n i y Cm,n j n = V cos x sin ,
n=1
This equation provides the solution for the vortex lattice problem.
3/11/98
m = 1,...,2N .
(6-95)
Note that if an essentially vertical surface is of interest, the form of F is more naturally
F = y - g(x,z),
and this should be used to work out the boundary condition in a similar fashion.
To illustrate the usual method, consider the simple planar surface case, where there is no
dihedral. Furthermore, recalling the example in the last section and the analysis at the beginning
of the chapter, the thin airfoil theory boundary conditions can be applied on the mean surface,
and not the actual camber surface. We also use the small angle approximations. Under these
circumstances, Eq. (6-95) becomes:
wm =
2N
Cm,nk n = V dxc
df
n=1
(6-96)
.
m
Thus we have the following equation which satisfies the boundary conditions and can be used to
relate the circulation distribution and the wing camber and angle of attack:
2N
Cm,nk V n = dxc m
n=1
df
m = 1,...,2N .
(6-97)
Equation (6-97) contains two cases:
or
1. The Analysis Problem. Given camber slopes and , solve for the circulation strengths,
(/V ) [ a system of 2N simultaneous linear equations].
2. The Design Problem. Given (/V ), which corresponds to a specified surface
loading, we want to find the camber and required to generate this loading (only
requires simple algebra, no system of equations must be solved).
Notice that the way dfc/dx and are combined illustrates that the division between camber,
angle of attack and wing twist is arbitrary (twist can be considered a separate part of the camber
distribution, and is useful for wing design). However, care must be taken in keeping the
bookkeeping straight.
One reduction in the size of the problem is available in many cases. If the geometry is
symmetrical and the camber and twist are also symmetrical, then n is the same on each side of
the planform (but not the influence coefficient). Therefore, we only need to solve for N s, not
2N (this is true also if ground effects are desired, see Katz and Plotkin11). The system of
equations for this case becomes:
N
Cm,nk left + Cm,nk right V n = dxc m
n=1
3/11/98
df
m = 1,..., N .
(6-98)

This is easy. Why not just program it up ourselves? You can, but most of the work is:
A. Automatic layout of panels for arbitrary geometry. As an example, when considering
multiple lifting surfaces, the horseshoe vortices on each surface must line up. The
downstream leg of a horseshoe vortex cannot pass through the control point of
another panel.
and
B. Converting n to the aerodynamics values of interest, C L , Cm, etc., and the spanload,
is tedious for arbitrary configurations.
Nevertheless, many people (including previous students in the Applied Computational
Aerodynamics class) have written vlm codes. The method is widely used in industry and
government for aerodynamic estimates for conceptual and preliminary design predictions. The
method provides good insight into the aerodynamics of wings, including interactions between
lifting surfaces.
Typical analysis uses (in a design environment) include
Predicting the configuration neutral point during initial configuration layout, and
studying the effects of wing placement and canard and/or tail size and location.
Finding the induced drag, CDi, from the spanload in conjunction with farfield methods.
With care, estimating control and device deflection effectiveness (estimates where
viscous effects may be important require calibration. Some examples are shown in the
next section. For example, take 60% of the inviscid value to account for viscous losses,
and also realize that a deflection of f = 20 - 25 is about the maximum useful device
deflection in practice).
Investigating the aerodynamics of interacting surfaces.
Finding the lift curve slope, CL , approach angle of attack, etc.
Typical design applications include:
Initial estimates of twist to obtain a desired spanload, or root bending moment.
Starting point for finding a camber distribution in purely subsonic cases.
Before examining how well the method works, two special cases require comments. The first
case arises when a control point is in line with the projection of one of the finite length vortex
segments. This problem occurs when the projection of a swept bound vortex segment from one
side of the wing intersects a control point on the other side. This happens frequently. The
velocity induced by this vortex is zero, but the equation as usually written degenerates into a
singular form, with the denominator going to zero. Thus a special form should be used. In
practice, when this happens the contribution can be set to zero without invoking the special form.
Figure 6-16 shows how this happens. Using the Warren 12 planform and 36 vortices on each side
of the wing, we see that the projection of the line of bound vortices on the last row of the left
hand side of the planform has a projection that intersects one of the control points on the right
hand side.
3/11/98
-0.50
Warren 12 planform
0.00
0.50
example is for 36 horseshoe

vortices on each side, all
vortices not shown for
clarity
projection of bound
vortices on this row
intersect control point
on other side
of planform
horseshoe vortex
1.00
control point
1.50
2.00
2.50
trailing legs extend to infinity
3.00
-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
Figure 6-16. Example of case requiring special treatment, the intersection of the projection of
a vortex with a control point.
A model problem illustrating this can be constructed for a simple finite length vortex
segment. The velocity induced by this vortex is shown in Fig. 6-17. When the vortex is
approached directly, x/l = 0.5, the velocity is singular for h = 0. However, as soon as you
approach the axis (h = 0) off the end of the segment (x/l > 1.0) the induced velocity is zero. This
illustrates why you can set the induced velocity to zero when this happens.
This second case that needs to be discussed arises when two or more planforms are used with
this method. This is one of the most powerful applications of the vortex lattice method. However,
care must be taken to make sure that the trailing vortices from the first surface do not intersect
the control points on the second surface. In this case the induced velocity is in fact infinite, and
the method breaks down. Usually this problem is solved by using the same spanwise distribution
of horseshoe vortices on each surface. This aligns the vortex the legs, and the control points are
well removed from the trailing vortices of the forward surfaces.
3/11/98
1.00
l
0.80
x
h
x=0
p
x/l
0.5
0.60
Vp
0.40
1.5
0.20
2.0
3.0
0.00
0.0
0.5
1.0
h/l
1.5
Figure 6-17. Velocity induced by finite straight line section of a vortex.
3/11/98
2.0

6.8 Examples of the Use and Accuracy of the vlm Method
How well does the method work? In this section we describe how the method is normally
applied, and present some example results obtained using it. More examples and a discussion of
the aerodynamics of wings and multiple lifting surfaces are given in Section 6.10.
The vortex lattice layout is clear for most wings and wing-tail or wing-canard configurations.
The method can be used for wing-body cases by simply specifying the projected planform of the
entire configuration as a flat lifting surface made up of a number of straight line segments. The
exact origin of this somewhat surprising approach is unknown. The success of this approach is
illustrated in examples given below.
To get good, consistent and reliable results some simple rules for panel layout should be
followed. This requires that a few common rules of thumb be used in selecting the planform
break points: i) the number of line segments should be minimized, ii) breakpoints should line up
streamwise on front and rear portions of each planform, and should line up between planforms,
iii) streamwise tips should be used, iv) small spanwise distances should be avoided by making
edges streamwise if they are actually very highly swept, and v) trailing vortices from forward
surfaces cannot hit the control point of an aft surface. Figure 6-18 illustrates these requirements.
y
line up spanwise
breaks on a common
break, and make the
edge streamwise
first planform
Model Tips
start simple
crude representation
of the planform is all right
keep centroids of areas
the same: actual planform to
vlm model
Make edges
streamwise
second planform
x
Figure 6-18. Example of a vlm model of an aircraft configuration. Note that one side of a
symmetrical planform is shown.
3/11/98

Examples from three reports have been selected to illustrate the types of results that can be
expected from vortex lattice methods. They illustrate the wide range of uses for the vlm method.
Aircraft configurations examined by John Koegler15
As part of a study on control system design methods, John Koegler at McDonnell Aircraft
Company studied the prediction accuracy of several methods for fighter airplanes. In addition to
the vortex lattice method, he also used the PAN AIR and Woodward II panel methods (see
Chapter 4 for details of the panel methods). He compared his predictions with the three-surface
F-15, which became known as the STOL/Maneuver demonstrator, and the F-18. These
configurations are illustrated in Fig. 6-19.
Three-Surface F-15
F/A-18
Figure 6-19.
3/11/98
Configurations used by McDonnell Aircraft to study vlm method accuracy

(Reference 15).

Considering the F-15 STOL/Maneuver demonstrator first, the basic panel layout is given in
Figure 6-20. This shows how the aircraft was modeled as a flat planform, and the corner points
of the projected configuration were used to represent the shape in the vortex lattice method and
the panel methods. Note that in this case the rake of the wingtip was included in the
computational model. In this study the panel methods were also used in a purely planar surface
mode. In the vortex lattice model the configuration was divided into three separate planforms,
with divisions at the wing root leading and trailing edges. On this configuration each surface
was at a different height and, after some experimentation, the vertical distribution of surfaces
shown in Figure 6-21 was found to provide the best agreement with wind tunnel data.
Vortex Lattice
233 panels
Woodward
208 panels
Pan Air
162 panels
Figure 6-20. Panel models used for the three-surface F-15 (Ref.15).
Figure 6-21. Canard and horizontal tail height representation (Ref. 15).
3/11/98

The results from these models are compared with wind tunnel data in Table 6-1. The vortex
lattice method is seen to produce excellent agreement with the data for the neutral point location,
and lift and moment curve slopes at Mach 0.2.
Subsonic Mach number effects are simulated in vlm methods by transforming the PrandtlGlauert equation which describes the linearized subsonic flow to Laplaces Equation using the
Gthert transformation. However, this is only approximately correct and the agreement with
wind tunnel data is not as good at the transonic Mach number of 0.8. Nevertheless, the vlm
method is as good as PAN AIR used in this manner. The vlm method is not applicable at
supersonic speeds. The wind tunnel data shows the shift in the neutral point between subsonic
and supersonic flow. The Woodward method, as applied here, over predicted the shift with Mach
number. Note that the three-surface configuration is neutral to slightly unstable subsonically, and
becomes stable at supersonic speeds.
Figure 6-22 provides an example of the use of the vlm method to study the effects of
moving the canard. Here, the wind tunnel test result is used to validate the method and to provide
an anchor for the numerical study (it would have been useful to have to have an experimental
point at -15 inches). This is typical of the use of the vlm method in aircraft design. When the
canard is above the wing the neutral point is essentially independent of the canard height.
However when the canard is below the wing the neutral point varies rapidly with canard height.
Table 6-1
Three-Surface F-15 Longitudinal 1Derivatives
Neutral
Point
(% mac)
Cm
(1/deg)
CL
(1/deg)
Wind Tunnel
15.70
.00623
.0670
Vortex Lattice
15.42
.00638
.0666
Woodward
14.18
.00722
.0667
Pan Air
15.50
.00627
.0660
Wind Tunnel
17.70
.00584
.0800
Vortex Lattice
16.76
.00618
.0750
Pan Air
15.30
.00684
.0705
Wind Tunnel
40.80
-.01040
.0660
Woodward
48.39
-.01636
.0700
Data Source
M = 0.2
M = 0.8
M = 1.6
from reference 15, Appendix by John Koegler
3/11/98
16
Neutral
Point 12
percent c
Vortex lattice
Wind tunnel data
8
0.07
CL
per deg
0.06
-20
-10
10
20
Canard Height, in.

Figure 6-22. Effect of canard height variation on three-surface F-15 characteristics (Ref. 15).
Control effectiveness is also of interest in conceptual and preliminary design, and the vlm
method can be used to provide estimates. Figure 6-23 provides an apparently accurate example
of this capability for F-15 horizontal tail effectiveness. Both CLh and Cmh are presented. The
vlm estimate is within 10% accuracy at both Mach .2 and .8. However, the F-15 has an all
moving horizontal tail to provide sufficient control power under both maneuvering and
supersonic flight conditions. Thus the tail effectiveness presented here is effectively a measure of
the accuracy of the prediction of wing lift and moment change with angle of attack in a nonuniform flowfield, rather than the effectiveness of a flap-type control surface. A flapped device
such as a horizontal stabilizer and elevator combination will have significantly larger viscous
effects, and the inviscid estimate from a vortex lattice or panel method (or any inviscid method)
will overpredict the control effectiveness. This is shown next for an aileron.
The aileron effectiveness for the F-15 presented in Fig. 6-24 is more representative of
classical elevator or flap effectiveness correlation between vlm estimates and experimental data.
This figure presents the roll due to aileron deflection. In this case the device deflection is subject
to significant viscous effects, and the figure shows that only a portion of the effectiveness
predicted by the vlm method is realized in the actual data. The vlm method, or any method,
should always be calibrated with experimental data close to the cases of interest to provide an
indication of the agreement between theory and experiment. In this case the actual results are
found to be about 60% of the inviscid prediction at low speed.
3/11/98
-0.02
Cm
per deg.
-0.01
0.00
0.02
CL
H
per deg.
Wind tunnel
Vortex lattice
Woodward
0.01
0.00
0.0
0.4
0.8
1.2
Mach Number
1.6
2.0
Figure 6-23. F-15 horizontal tail effectiveness (Ref. 15).

0.0016
0.0012
0.0008
Cl
a
per deg.
0.0004
0.0000
0.0
(a = 20)
Wing
Vortex lattice: Sym chordwise
panels
6
8
10
Wind tunnel
0.2
0.4
0.6
Mach Number
0.8
Figure 6-24. F-15 aileron effectiveness (Ref. 15).
3/11/98
1.0

The F/A-18 was also considered by Koegler. In this case the contributions to the
longitudinal derivatives by the wing-tip missiles and the vertical tail were investigated (the
vertical tails are canted outward on the F/A-18). The panel scheme used to estimate the effects of
the wing-tip missile and launcher is shown in Figure 6-25. The results are given in Table 6-2.
Here the computational increments are compared with the wind tunnel increments. The vlm
method over predicts the effect of the wing-tip missiles, and under predicts the effects of the
contribution of the vertical tail to longitudinal characteristics due to the cant of the tail (recall
that on the F/A-18 the rudders are canted inward at takeoff to generate an additional nose up
pitching moment) .
Figure 6-25. F/A-18 panel scheme with wing-tip missile and launcher (Ref. 15).
3/11/98
Table 6-2
F/A-18 Increments Due To Adding Wing-Tip Missiles and
Launchers,
and Vertical Tails
Mach
Number
Wing Tip
Missiles
and
Launchers
Vertical
Tails
Neutral
Point
(% mac)
CM
(1/deg)
CL
(1/deg)
Wind
Tunnel
0.2
1.10
-0.00077
0.0020
"
0.8
1.50
-0.00141
0.0030
"
1.6
-1.60
0.00148
0.0030
Vortex
Lattice
0.2
1.48
-0.00121
0.0056
"
0.8
2.11
-0.00198
0.0082
Woodward
0.2
1.52
-0.00132
0.0053
"
0.8
1.77
-0.00180
0.0079
"
1.6
-0.17
0.00074
0.0022
Wind
Tunnel
0.2
1.50
-0.00110
0.0050
"
0.8
2.00
-0.00202
0.0080
Vortex
Lattice
0.2
1.11
-0.00080
0.0022
"
0.8
1.32
-0.00108
0.0026
from refernce 15, Appendix by John Koegler
Finally, the effects of the number of panels and the way they are distributed is presented in
Figure 6-26. In this case the vlm method is seen to take between 130 to 220 panels to produce
converged results. For the vortex lattice method it appears important to use a large number of
spanwise rows, and a relatively small number of chordwise panels (5 or 6 appear to be enough).
3/11/98

4
( ) Number of chordwise panels
Neutral
point, 2
percent c
(9)
(5)
Vortex Lattice Method
0
0
40
80
120
160
Total number of panels
200
240
a) vortex lattice sensitivity to number of panels

6
(4)
4
Neutral
point, 2
percent c
(8)
Woodward Program
0
0
40
80
120
160
200
240
b) Woodward program sensitivity to number of panels

0
Neutral
point,
percent c
-2
(3)
Pan Air
0

40
80
120
160
200
c) Pan Air program sensitivity to number of panels
Figure 6-26. F/A-18 panel convergence study (Ref. 15).
240
Although this study has been presented last in this section, a study like this should be
conducted before making a large number of configuration parametric studies. Depending on the
relative span to length ratio the paneling requirements may vary. The study showed that from
about 120 to 240 panels are required to obtain converged results. The vortex lattice methods
obtains the best results when many spanwise stations are used, together with a relatively small
number of chordwise panels. In that case about 140 panels provided converged results.
3/11/98

Slender lifting body results from Jim Pittman16
To illustrate the capability of the vortex lattice method for bodies that are more fuselage-like
than wing-like, we present the lifting body comparison of the experimental and vlm results
published by Jim Pittman of NASA Langley. Figure 6-27 shows the configuration used. Figure
6-28 provides the results of the vortex lattice method compared with the experimental data. In
this case the camber shape was modeled by specifying camber slopes on the mean surface. The
model used 138 vortex panels. The bars at several angles of attack illustrate the range of
predictions obtained with different panel arrangements. For highly swept wings, leading edge
vortex flow effects are included, as we will describe in Section 6.12. The program VLMpc
available for this course contains the option of using the leading edge suction analogy to model
these effects. Remarkably good agreement with the force and moment data is demonstrated in
Fig. 6-28. The nonlinear variation of lift and moment with angle of attack arises due to the
inclusion of the vortex lift effects. The agreement between data and computation breaks down at
higher angles of attack because the details of the distribution of vortex flow separation are not
provided by the leading edge suction analogy. The drag prediction is also very good. The
experimental drag is adjusted by removing the zero lift drag, which contains the drag due to
friction and separation. The resulting drag due to lift is compared with the vlm estimates. The
comparisons are good primarily because this planform is achieving, essentially, no leading edge
suction, and hence the drag is simply C D = CL tan.
copyrighted figure from the AIAA Journal of Aircraft
Figure 6-27. Highly swept lifting body type hypersonic concept (Ref. 16).
3/11/98
Figure 6-28. Comparison of CL , Cm, and CD predictions with data (Ref. 16).
Non-planar results from Kalman, Rodden and Giesing,17
All of the examples presented above considered essentially planar lifting surface cases. The
vortex lattice method can also be used for highly non-planar analysis, and the example cases
used at Douglas Aircraft Company in a classic paper 17 have been selected to illustrate the
capability. To avoid copyright problems, several of the cases were re-computed using the
Virginia Tech code JKayVLM, and provide an interesting comparison with the original results
from Douglas. Figure 6-29 presents an example of the prediction capability for the pressure
loading on a wing. In this case the geometry is complicated by the presence of a wing fence. The
pressures are compared with data on the inboard and outboard sides of the fence. The agreement
is very good on the inboard side. The comparison is not so good on the outboard side of the
fence. This quality of agreement is representative of the agreement that should be expected using
vortex lattice methods at low Mach numbers in cases where the flow would be expected to be
attached.
Figure 6-30 provides an example of the results obtained for an extreme non-planar case: the
ring, or annular, wing. In this case the estimates are compared with other theories, and seen to be
very good. The figure also includes the estimate of Cmq. Although not included in the present
discussion, Cmq and Clp can be computed using vlm methods, and this capability is included in
the vortex lattice method provided here, VLMpc.
3/11/98
Figure 6-29. Comparison of Cp loading on a wings with a fence (Ref. 17).
Figure 6-30. Example of aerodynamic characteristics of a ring wing (Ref. 17).
3/11/98

Figure 6-31 provides an example of the effects of the presence of the ground on the
aerodynamics of simple unswept rectangular wings. The lift and pitching moment slopes are
presented for calculations made using JKayVLM and compared with the results published by
Kalman, Rodden and Giesing, 17 and experimental data. The agreement between the data and
calculations is excellent for the lift curve slope. The AR = 1 wing shows the smallest effects of
ground proximity because of the three dimensional relief provided around the wing tips. As the
aspect ratio increases, the magnitude of the ground effects increases. The lift curve slope starts to
increase rapidlyi as the ground is approached.
8.0
Solid lines: computed using JKayVLM (Ref. 12)

Dashed lines: from Kalman, Rodden and Giesing (Ref. 17)
Symbols: Experimental data
7.0
c/4
6.0
CL
5.0
4.0
AR
4
3.0
2.0
1.0 0.0
1
0.2
0.4
0.6
0.8
1.0
1.2
1.4
h/c
0.20
0.10
CM
AR
1
2
4
0.00
-0.10
-0.20
-0.30

-0.40
0.0
0.2
0.4
0.6
0.8
1.0
1.2
h/c
1.4
Figure 6-31. Example of ground effects for a simple rectangular wing (a case from Ref. 17).
The wings also experience a significant change in the pitching moment slope
(aerodynamic center shift), and this is also shown. Note that the predictions start to differ as the
ground is approached. JKayVLM actually rotates the entire surface to obtain another solution to
use in estimating the lift curve slope. The standard procedure used by most methods is to simply
change the slope condition at the mean line, as discussed previously in this chapter. Because of
3/11/98

the proximity to the ground, this might be a case where the transfer of the boundary condition
may not be accurate. I have not asked Joe Giesing if he remebers how these calculations were
made (nearly 30 years ago!).
Figure 6-32 presents similar information for the effect of dihedral angle on a wing. In this
case the effects of anhedral, where the wing tip approach the ground, are extremely large. The
results of dihedral changes for a wing out of ground effect are shown for comparison. Both
methods agree well with each other, with differences appearing only as the wingtips approach
the ground. Here again, JKayVLM actually rotates the entire geometry, apparently resulting in
an increase in the effects as the tips nearly contact the ground. It also prevents calculations from
being obtained as close to the ground as thepublished results. In making these calculations it was
discovered that the wing panel was rotated and not sheared, so that the projected span decreases
as the dihedral increases, and this produces much more pronounced changes in the lift curve
slope due to the reduction in projected span.
12

ARref = 4, Planform rotated by .
10
CL
h/c = 0.5
6
h/c = 1
4
h/c =
2
Note: Most of the decrease in lift curve slope for the wing out of ground
effect is due to the effect of reduced span as the wing rotates from
the zero dihedral condition
0
-40
-30
-20
-10
0
10
Dihedral angle,
20
30
Figure 6-32. Example of ground effects for a wing with dihedral (a case from Ref. 17).
3/11/98
40
6.9 Program VLMpc and the Warren 12 Test Case

This vortex lattice method can be used on personal computers. The version of the Lamar
program described in NASA TN D-79213 fits easily into personal computers, and is available for
student use (students typed this code in from the listing in the TN) as VLMpc. The code and
input instructions are described in Appendix D.6.
This code is still used in advanced design work, and can be used to investigate many ideas in
wing aerodynamics. As shown above, results can be obtained and used before the large time
consuming methods of CFD are used to examine a particular idea in detail.
This section defines one reference wing case that is used to check the accuracy of vortex
lattice codes. It provides a ready check case for the evaluation of any new or modified code, as
well as a check on the panel scheme layout. This wing is known as the Warren 12 planform, and
is defined, together with the official characteristics from previous calculations, in Fig. 6-33
below.
=2 2
LE = 53.54
Swing = 2 2
AR
CL
1.50
1.91421
= 2.743 / rad
CM = 3.10 / rad
0.50
1.41421
Warren-12 Planform
Figure 6-33. Definition and reference results for the Warren-12 wing.
For the results cited above, the reference chord used in the moment calculation is the average
chord (slightly nonstandard, normally the reference chord used is the mean aerodynamic chord)
and the moment reference point is located at the wing apex (which is also nonstandard).
3/11/98

6.10 Aerodynamics of Wings
With a three-dimensional method available, we can examine the aerodynamics of wings.
Most of the results presented in this section were computed using VLMpc. One key advantage of
the vortex lattice method compared to lifting line theory is the ability to treat swept wings.
Classical Prandtl lifting line theory is essentially correct for unswept wings, but is completely
erroneous for swept wings. Aerodynamics of unswept wings are closely related to the airfoil
characteristics of the airfoil used in the wing. This relationship is less direct for swept wings.
Many of the most important wing planform-oriented characteristics of wings arise when the
planforms are swept. Even though sweep is used primarily to reduce compressibility effects, the
important aerodynamic features of swept wings can be illustrated at subsonic speeds using the
vlm method.
6.10.1 - Basic Ideas
Wings are designed to satisfy stability and handling characteristics requirements, while
achieving low drag at the design conditions (usually cruise and sustained maneuver). They must
also attain high maximum lift coefficients to meet field performance and maneuver requirements.
Although these requirements might at first appear overwhelming, a small number of key
characteristics can provide a basic physical understanding of the aerodynamics of wings.
Aerodynamic Center: The first key characteristic is the aerodynamic center of the wing,
defined as location at which dCm/dCL = 0. The neutral point of the configuration is the
aerodynamic center for the entire configuration. The vlm method was shown to provide accurate
predictions of the neutral point for many configurations in the previous sections. The location of
the neutral point is important in initial configuration layout to position the wing and any
longitudinal stability and control surfaces at the proper location on the aircraft. Subsequently this
information is fundamental in developing the control system. Wing planform shaping, as well as
positioning, is used to control the location of the configuration neutral point.
Spanload: The next key consideration is the spanload distribution, ccl/ca, where c is the local
chord, ca is the average chord, and cl is the local section lift coefficient. The spanload controls
the location of the maximum section lift coefficient, the induced drag, and the magnitude of the
wing root bending moment. The location and value of the maximum section lift coefficient
determines where the wing will stall first. * If the wing airfoil stalls in front of a control surface,
control will be poor at flight conditions where the control becomes very important. The shape of
the spanload, together with the actual value of the wingspan, determines the value of the induced
*
For a trapezoidal wing with an elliptic spanload the maximum value of the local lift coefficient occurs at = 1 - .
3/11/98
drag. For a specified span, the performance of the wing is evaluated by finding the value of the
span efficiency factor, e, as described in Chapter 5. Finally, the wing root bending moment
provides an indication of the structural loading requirements that the wing structure must be
designed to accommodate. When considering the total system, the basic aerodynamic efficiency
may be compromised to reduce structural wing weight. The shape of the spanload can be
controlled through a combination of planform selection and wing twist. Typical twist
distributions required to produce good wing characteristics are presented below.
The simplest example of planform shaping is the selection of wing aspect ratio, AR, wing
taper, , and wing sweep, . While the aerodynamicist would like to see high values the aspect
ratio, several considerations limit aspect ratio. Perhaps the most important limitation is the
increase of wing structural weight with increasing aspect ratio. In addition, the lift coefficient
required to maximize the benefits of high aspect ratio wings increases with the square root of the
aspect ratio. Hence, airfoil performance limits can restrict the usefulness of high aspects ratios,
especially for highly swept wings based on airfoil concepts. In recent years advances in both
aerodynamics and structures have allowed aircraft to be designed with higher aspect ratios and
reduced sweep. Table 6-3 provides some key characteristics of transport wings designed to
emphasize efficient cruise while meeting takeoff and landing requirements.
Taper: Several considerations are used in selecting the wing taper. For a straight untwisted,
unswept wing, the minimum induced drag corresponds to a taper ratio of about 0.4. However, a
tapered wing is more difficult and hence expensive to build than an untapered wing. Many
general aviation aircraft wings are built with no taper (all ribs are the same, reducing fabrication
cost, and the maximum section Cl occurs at the root, well away from the control surface). To
reduce structural weight the wing should be highly tapered, with < .4. However, although
highly tapered wings are desirable structurally, the section lift coefficient near the tip may
become high. This consideration limits the amount of taper employed (current jet transports use
taper ratios in the range of 0.2 to 0.3, as well as progressively increasing twist upward from the
tip). As an example, the Aero Commander 500 had an aspect ratio of 9.5 and a taper ratio of 0.5
(it also had -6.5 of twist and the quarter chord of the wing was swept forward 4).
Sweep: Sweep is used primarily to delay the effects of compressibility and increase the drag
divergence Mach number. The Mach number controlling these effects is approximately equal to
the Mach number normal to the leading edge of the wing, Meff = M cos. The treatise on swept
planforms by Kchemann is very helpful in understanding swept wing aerodynamics.18
Aerodynamic performance is based on the wingspan, b. For a fixed span, the structural span
increases with sweep, bs = b/cos, resulting in a higher wing weight. Wing sweep also leads to
aeroelastic problems. For aft swept wings flutter becomes an important consideration. If the wing
3/11/98
is swept forward, divergence is a problem. Small changes in sweep can be used to control the
aerodynamic center when it is not practical to adjust the wing position on the fuselage (the DC-3
is the most famous example of this approach).
Table 6-3
Typical Planform Characteristics of Major Transport Aircraft
1st Flight
Aircraft
W/S
AR
(c/4 )
1957
B707-120
105.6
7.04
35
0.293
1958
DC-8-10
111.9
7.32
30
0.230
1963
B707-320C
110.0
7.06
35
0.250
1970
B747-200B
149.1
6.96
37.5
0.240
1970
L-1011
124.4
8.16
35
0.200
1972
DC-10-30
153.7
7.57
35
0.230
1972
A300 B2
107.9
7.78
28
0.230
1982
A310-100
132.8
8.80
28
0.260
1986
B767-300
115.1
7.99
31.5
0.182
1988
B747-400
149.9
7.61
37.5
0.240
1990
MD-11
166.9
7.57
35
0.230
1992
A330
119.0
9.3
29.74
0.192
data courtesy of Nathan Kirschbaum
To understand the effects of sweep, the Warren 12 wing is compared with wings of the same
span and aspect ratio, but unswept and swept forward. The planforms are shown in Figure 6-34.
The wing leading edge sweep of the aft swept wing becomes the trailing edge sweep of the
forward swept wing. Figure 6-35 provides the spanload and section lift coefficient distributions
from VLMpc. The spanload, ccl / ca , is given in Fig. 6-35a, where, c is the local chord, cl is the
local lift coefficient, based on the local chord, and ca is the average chord, S/b. Using this
nomenclature, the area under the curve is the total wing lift coefficient. Note that sweeping the
wing aft increases the spanload outboard, while sweeping the wing forward reduces the spanload
outboard. This follows directly from a consideration of the vortex lattice model of the wing. In
both cases, the portion of the wing aft on the planform is operating in the induced upwash
flowfield of the wing ahead of it, resulting in an increased spanload. Figure 6-35b shows the
corresponding value of the local lift coefficient. Here the effect of sweep is more apparent. The
forward swept wing naturally results in a spanload with a nearly constant lift coefficient. This
means that a comparatively higher wing lift coefficient can be achieved before the wing stall
begins. The program LIDRAG can be used to compare the span es associated with these
spanloads (an exercise for the reader).
3/11/98
V
Forward
Swept
Wing
Unswept
Wing
Aft
Swept
Wing
Figure 6-34. Comparison of forward, unswept, and aft swept wing planforms, AR = 2.8.
3/11/98
1.60
1.40
1.20
Spanload,
1.00
c cl / ca
0.80
aft swept wing
0.60
unswept wing
0.40
forward swept wing
0.20
0.00
Warren 12 planform, sweep changed

0.0
0.2
0.4
y/(b/2)
0.6
0.8
1.0
0.8
1.0
a) comparison of spanloads
1.40
aft swept wing
1.20
1.00
Section CL
0.80
unswept wing
forward swept wing
0.60
0.40
0.20
Warren 12 planform, sweep changed
0.00
0.0
0.2
0.4
0.6
y/(b/2)
b) comparison of section lift coefficients

Figure 6-35. Effects of sweep on planform spanload and lift coefficient distributions, AR = 2.8.
3/11/98

Similar results are now presented for a series of wings with larger aspect ratios (AR = 8)
than the wings used in the study given above. Figure 6-36 shows the planforms used for
comparison, and Figure 6-37 presents the results for the spanwise distribution of lift and section
lift coefficient. These results are similar to the previous results. However, the trends observed
above are in fact exaggerated at the higher aspect ratio.
Aerodynamic problems as well as structural penalties arise when using a swept wing.
Because of the high section lift coefficient near the tip, aft swept wings tend to stall near the tip
first. Since the lift at the tip is generated well aft, the pitching moment characteristics change
when the this stall occurs. With the inboard wing continuing to lift, a large positive increase in
pitching moment occurs when the wingtip stalls. This is known as pitchup, and can be difficult to
control, resulting in unsafe flight conditions. Frequently the swept wing pitching moment
characteristics are compounded by the effects of flow separation on the outboard control surface.
Figure 6-38 provides an example of the pitching moment characteristics of an isolated aspect
ratio 10 wing using experimental data.19 The figure also includes the predictions from VLMpc.
The agreement is reasonably good at low angle of attack, but deteriorates at high angle of attack
as viscous effects become important. This is another reason that sweep is minimized.
V
Forward
Swept
Wing
Unswept
Wing
Aft
Swept
Wing
Figure 6-36. Comparison of forward, unswept, and aft swept wing planforms, AR = 8.
3/11/98
1.6
Aspect ratio 8 wings
1.4
1.2
Spanload,
cc / c 1.0
l
0.8
aft swept wing
0.6
unswept wing
0.4
forward swept wing
0.2
0.0
0.2
0.4
0.6
0.8
1.0
0.8
1.0
y/(b/2)
a) comparison of spanloads
1.50
aft swept wing
1.00
unswept wing
Section CL
forward swept wing
0.50
Aspect ratio 8 planforms

0.00
0.0
0.2
0.4
0.6
y/(b/2)
b) comparison of section lift coefficients

Figure 6-37. Effects of sweep on planform spanload and lift coefficient distributions, AR = 8.
3/11/98
0.08
data from NACA RM A50K27
0.06
Cm
AR = 10, c/4 = 35, = 0.5

Re = 10 million
0.04
x ref = c/4
0.02
0.00
-0.02
-0.04
VLMpc calculation
-0.06
-0.4
-0.2
0.0
0.2
0.4
CL
0.6
0.8
1.0
1.2
Figure 6-38. Example of isolated wing pitchup: NACA data19 compared with VLMpc.
To control the spanload, the wing can be twisted. Figure 6-39 shows typical twist
distributions for aft and forward swept wings, obtained from John Lamars program LamDes.20
(see Chapter 5 for a description). In each case the twist is used to reduce the highly loaded areas,
and increase the loading on the lightly loaded portions of the wing. For an aft swept wing this
means the incidence is increased at the wing root, known as washin, and reduced, known as
washout, at the wing tip. Just the reverse is true for the forward swept wing. The sudden drop in
required twist at the tip for the forward dwept wing case is frequently seen in typical design
methods and attribute to a weakness in the method and faired out when the aerodynamicist
gives his design to the lofting group.
Although geometric sweep is used to reduce the effective Mach number of the airfoil, the
geometric sweep is not completely effective. The flowfield resists the sweep. In particular, the
wing root and tip regions tend to effectively unsweep the wing. Aerodynamicists study lines of
constant pressure on the wing planform known as isobars to investigate this phenomenon. Figure
6-40 presents the computed isobars for a typical swept wing,21 using a transonic small
disturbance method.22 The effect is dramatic. The effective sweep may actually correspond to
the isobar line from the wing root trailing edge to the leading edge at the wing tip. To increase
the isobar sweep, in addition to geometric sweep and twist, the camber surface and thickness are
typically adjusted to move the isobars forward at the wing root and aft at the wing tip. This is a
key part of the aerodynamic wing design job, regardless of the computational, methodology used
to obtain the predicted isobar pattern.
3/11/98

6.0
5.0
4.0
, 3.0
deg. 2.0
1.0
0.0
-1.0
0.2
0.4
0.6
y/(b/2)
0.8
7.0
6.0
5.0
, 4.0
deg. 3.0
2.0
1.0
0.0
-1.0
0
0.2
0.4
0.6
y/(b/2)
0.8
a) aft swept wing

b) forward swept wing
Figure 6-39. Typical twist distribution required to improve spanload on swept wings
AR = 4
= 0.6
45
(a) L51F07 configuration
(b) Upper Surface Isobars

Figure 6-40. Example of the isobar distribution on an untwisted swept wing.22
Using the wing planform and twist, together with a constant chord loading, Fig. 6-41
provides the camber lines required to support the load near the root, the mid-span and the wing
tip. These results were also computed using LamDes.20 At each station a similar chord load is
specified. Here we clearly see the differences in the camber required. This is an explicit
illustration of the modification to an airfoil camberline required to maintain two-dimensional
airfoil-type performance when the airfoil is placed in a swept wing. These modifications
represent the explicit effects of the three dimensionality of the flowfield.
3/11/98
= 0.925
= 0.475
0.05
0.04
0.03
(z-zle)/c
0.02
= 0.075
0.01
0.00
-0.01
-0.2
0.2
0.4
x/c
0.6
0.8
1.2
Figure 6-41. Comparison of camber lines required to develop the same chord load shape at
the root, mid-span and tip region of an aft swept wing. (from LamDes 20)
Many other refinements are available to the aerodynamic designer. Insight into both the
human and technical aspects of wing design prior to the introduction of computational
aerodynamics is available in two recent books describing the evolution of the Boeing series of jet
transports.23,24 One interesting refinement of swept wings has been the addition of trailing edge
area at the wing root. Generally known as a Yehudi flap, this additional area arises for at least
two reasons. The reason cited most frequently is the need to provide structure to attach the
landing gear at the proper location. However, the additional chord lowers the section lift
coefficient at the root, where wing-fuselage interference can be a problem, and the lower
required section lift makes the design job easier. Douglas introduced this planform modification
for swept wings on the DC-8, while Boeing did not incorporate it until the -320 model of the
707. However, the retired Boeing engineer William Cook, in his book, 23 on page 83, says it was
first introduced on the B-29 to solve an interference problem between the inboard nacelle and the
fuselage. The aerodynamic benefit to the B-29 can be found in the paper by Snyder. 25 Cook says,
in a letter to me, that the device got its name because each wind tunnel part needed a name and
there was a popular radio show at the time that featured the continuing punch line Whos
Yehudi? (the Bob Hope Radio show featuring Jerry Colonna, who had the line). Thus, a Boeing
engineer decided to call it a Yehudi flap. This slight extra chord is readily apparent when
examining the B-29, but is very difficult to photograph.
3/11/98
6.10.2 The Relation Between Airfoils and Swept Wings

Chapter 4 examined the basic aerodynamics of airfoils using panel methods. This chapter has
emphasized the planform shape, and its analysis using vortex lattice methods. The connection
between the airfoil and planform is important. In most cases the integration of the airfoil concept
and the wing planform concept is crucial to the development of a successful configuration.
Simple sweep theory can be used to provide, at least approximately, the connection between the
airfoil and the planform. The typical aerodynamic design problem for an airfoil in a wing is
defined by specifying the streamwise thickness to chord ratio, t/c, the local section lift
coefficient, C Ldes, and the Mach number. This three-dimensional problem is then converted to a
corresponding two-dimensional problem. The desired two dimensional airfoils are then designed
and transformed back to the streamwise section to be used as the wing airfoil section. Examples
of the validity of this technique, together with details on other properties, including the cosine
cubed law for profile drag due to lift are available in the NACA report by Hunton. 26 The
relations between the streamwise airfoil properties and the chordwise properties (values normal
to the leading edge, as shown in Fig. 6-42) are:
Cn = Cs cos
Mn = Mcos
t/c)n = t/c)s/cos
Cn
Cs
(699)
and
CLs = CLn cos2
Figure 6-42. Swept wing definitions.

These relations demonstrate that the equivalent two-dimensional airfoil is thicker, operates
at a lower Mach number, and at a higher lift coefficient than the three dimensional wing airfoil
section. Taper effects on real wings require the selection of an effective sweep angle. Numerous
approaches have been used to determine the effective angle, where guidance has been obtained
by examining experimental data. The quarter chord sweep or shock sweep are typical choices.
3/11/98

One good example of airfoil/planform matching is the Grumman X-29. In that case wind tunnel
testing of advanced transonic maneuver airfoil sections on aft swept wing configurations led the
aerodynamicists (Glenn Spacht in particular) to conclude that the proper planform to take
advantage of the advanced airfoil section performance should be swept forward.
6.10.3 - Wing/Tail and Canard/Wing Aerodynamics
Additional lifting surfaces are used to provide control over a wide range of conditions. If
modern advanced control systems are not used, the extra surface is also designed, together with
the rest of the configuration, to produce a stable design. Considering aft tail configurations first,
the problem of pitchup described above for isolated wings must be reconsidered for aft tailled
configurations. In particular, T-tail aircraft can encounter problems when the horizontal tail
interacts with the wake of the wing at stall. Figure 6-43 provides the pitching moment
characteristics of the DC-9.27 The initial abrupt nose-down characteristic is the result of careful
design, before the large pitchup develops. Note that even though pitch-up is a viscous effect,
inviscid calculations clearly show why it happens, and can provide valuable information.
Figure 6-43 shows that a stable trim condition occurs at an angle of attack of 43. This is an
undesirable equilibrium condition, which could result in the vehicle actually trying to fly at
this angle of attack. If adequate control power is not available, it may even be difficult to
dislodge the vehicle from this condition, which is commonly known as a deep or hung stall. This
will result in a rapid loss of altitude due to the very high drag. Although for this configuration
full down elevator eliminates the possibility of getting trapped in a trimmed flight condition at
this angle of attack, the amount of pitching moment available may not be sufficient to affect a
rapid recovery from this condition. Examples of pitchup characteristics are not readily available.
Aerodynamic designers do not like to admit that their configurations might have this
characteristic. This aspect of swept wing and wing-tail aerodynamics is an important part of
aerodynamic configuration development.
Even low tail placement cannot guarantee that there will not be a problem. Figure 6-44
shows the pitching moment characteristics for an F-16 type wind tunnel model. 28 In this case a
deep stall is clearly indicated, and in fact the allowable angle of attack on the F-16 is limited to
prevent the airplane from encountering this problem. In this case the pitchup arises because of
powerful vortices generated by the strakes, which continue to provide lift as the wing stalls. This
type of flowfield is discussed in Section 6.12.
Canard configurations provide another interesting example of multiple lifting surface
interaction. The downwash from the canard wake, as it streams over the wing, reduces the
effective angle of attack locally, and hence the local lift on the wing behind the canard. Wing
3/11/98

twist is used to counteract this effect. Figure 6-45 illustrates how this interaction occurs. The
relative loading of the surfaces is an important consideration in configuration aerodynamics. The
induced drag is highly dependent on the relative wing loading, which is determined by the
selection of the configuration stability level and the requirement to trim about the center of
gravity. in determining the induced drag. Figure 6-46, also computed using LamDes,20 shows
how the trimmed drag changes with cg position. Three different canard heights are shown for a
range of cg positions, which is equivalent to varying the stability level. Figure 6-47 provides an
example of the wing twist required to account for the effect of the canard downwash. Note that
the forward swept wing twist increment due to the canard acts to reduce the twist required, which
is exactly opposite the effect for the aft swept wing.
a copyrighted figure from the AIAA Journal of Aircraft
Figure 6-43. Pitching moment characteristics of the DC-9. 27
3/11/98

0.4
Nose up
0.2
Cm
0.0
Cannot get
nose down!
-0.2
h
-0.4
Nose down
0
+25
-25
-0.6
0
20
40
, deg
60
80
Figure 6-44. Pitching moment characteristics of an F-16 type wind tunnel model.28
Upwash outboard
of canard tips
Canard wake
streams over
wing
+
w 0
A
canard wake extends to indinity

wing wake not shown
y
Downwash from canard across wing
at Section A-A
Figure 6-45. Illustration of wing-canard interaction.
3/11/98
Cp
Cp
0
x/c 1
x/c
a) Canard-wing planform and chord loads used in drag calculation.

Note: The sample case may not be a good
design, the canard is too big.
0.0400
Computations from LamDes
0.0350
M = 0.3, CL = 0.5
Canard lift must be
negative to trim
Canard height
above wing ,z/b
CDi
0.1
0.2
0.0300
Static Margin
0.3
Stable Unstable
0.0250
0.0200
-0.4 -0.3
Forward
Advantage of relaxed
static stability evident
-0.2
-0.1
0.0
0.1
x/c
Advantage of vertical
separation clearly evident
0.2
0.3
0.4
Aft
b) Minimum trimmed drag variation with trim position and canard-wing separation.
Figure 6-46. Example of relation of minimum trimmed drag to balance (stability).20
3/11/98

6.0
5.0
4.0
, 3.0 without
deg.2.0
canard
1.0
0.0
-1.0
0.0
0.2
8.0
in presence
of canard
6.0
canard tip
vortex effect
in presence
of canard
4.0
2.0
without
canard
0.0
canard tip
vortex effect
-2.0
0.0 0.2 0.4 0.6 0.8 1.0
0.4
0.6
0.8
1.0
y/(b/2)
y/(b/2)
a) aft swept wing
b) forward swept wing
Figure 6-47. Effects of canard on twist requirements. Twist required for minimum drag
using LamDes20 (Note: results depend on configuration details, balance).
6.11 Inverse Design Methods and Program DesCam

Although most of the analysis discussed above corresponds to the analysis problem, the
design problem can also be treated. In this section we provide one example: the determination of
the camber line shape required to obtain a specific chord load in the two dimensional case. We
take the opportunity to illustrate a method due to Lan6 that uses a mathematically based selection
of vortex and and control point placements instead of the 1/4 - 3/4 rule used above.
Recall that a line of vortex singularities induces a vertical velocity on the singularity line
given by (see chapter 4 and Karamcheti 13:
c
1
(x ) dx .
w(x) =
2 x x
(6-100)
For thin wing theory the vertical velocity can be related to the slope as shown above in Section
6.2. The vortex strength can be related to the streamwise velocity by = u+ - u-. This in turn can
be used to relate the vorticity to the change in pressure, Cp through:
Cp = Cpl C pu = 2u 2u+ = 2 u+ u
(6-101)
which leads to:

Cp (x)
= (x)
2
resulting in the expression for camber line slope in terms of design chord load:
(6-102)
dz
1 Cp
=
dx .
dx
4 x x
0
3/11/98
(6-103)

Here dz/dx includes the slope due to the angle of attack. Note that the integral contains a
singularity, and this singularity introduces the extra complications that require special analysis
for numerical integration. The original Lan theory was used to find Cp (in a slightly different
form), but it can also be used to obtain dz/dx from Cp. To do this, Lan derived a summation
formula to obtain the slope. Once the slope is known, it is integrated to obtain the camber line.
Lan showed that the integral in Eq. (6-100) could be very accurately found from the
summation:
dz
1 N C p
=
dx i
N k=1 4
x k (1 xk )
xi x k
(6-104)
where:
xk =
1
(2k 1)
1 cos
2N
2
k = 1,2,..., N
(6-105)
and:
xi =
1
i
1 cos
N
2
i = 0,1,2,..., N .
(6-106)
Here N + 1 is the number of stations on the camber line at which the slopes are obtained.
Given dz/dx, the camber line is then computed by integration using the trapezoidal rule
(marching forward starting at the trailing edge):
dz
x x dz
zi +1 = z i i+1 i
+
.
dx i dx i+1
2
(6-107)
The design angle of attack is then:

DES = tan1 z0
(6-108)
The camber line can then be redefined in standard nomenclature, i.e., z(x=0) = z(x=1) = 0.0:
zi = zi (1 x i )tan DES
(6-109)
How well does this work? Program DesCam implements the method described here, and the
users manual is provided in App. D.7. Here we compare the results from DesCam with the
analytic formula given in Appendix A.1 for the NACA 6 Series mean line with a = .4. The
results are shown in Figure 6-48 below. Notice that the camber scale is greatly enlarged to
3/11/98

demonstrate the excellent comparison. Even though the chord load is constructed by two straight
line segments, the resulting required camber line is highly curved over the forward portion of the
airfoil. Note also that thin airfoil theory allows only two possible values for the pressure
differential at the leading edge, zero or infinity. A close examination of the camber line shape
required to produce a finite load reveals a singularity. The slope is infinite. This feature is much
easier to study using the analytic solution, as given in Appendix A. This approach can easily be
extended to three dimensions. Notice that design problem is direct, in that it does not require the
solution of a system of equations.
(Z-Z0)/C - DesCam
Z/C - from Abbott & vonDoenhoff
0.12
2.00
Design Chord
Loading
0.10
1.50
CP
0.08
Z/C
1.00
0.06
0.50
0.04
0.00
0.02
0.00
0.0
0.2
0.4
0.6
0.8
-0.50
1.0
X/C
Figure 6-48. Example and verification of camber design using DesCam.
3/11/98
6.12 Vortex Flow Effects and the Leading Edge Suction Analogy
For highly swept wings at even moderate angles of attack, the classical attached flow/trailing
edge Kutta condition flow model weve adopted is wrong. Instead of the flow remaining attached
on the wing and leaving the trailing edge smoothly, the flow separates at the leading edge,
forming a well defined vortex. This vortex plays an important role in the design of highly swept,
or slender wing aircraft. The most notable example of this type of configuration is the
Concorde. Sharp leading edges promote this flow phenomena. The basic idea is illustrated in the
sketch from Payne and Nelson29 given here in Fig. 6-49.
Vortical flow
abova a delta wing
Primary vortex
Secondary vortex
Figure 6-49. Vortex flow development over a delta wing with sharp edges.29
An important consequence of this phenomena is the change in the characteristics of the
lift generation as the wing angle of attack increases. The vortex that forms above the wing
provides an additional low pressure force due to the strongly spiraling vortex flow. The low
pressure associated with the centrifugal force due to the vortex leads to the lower pressure on the
wing surface. As the wing increases its angle of attack the vortex gets stronger, further reducing
the pressure on the wing. The resulting increase in lift due to the vortex can be large, as shown in
Fig. 6-50, from Polhamus.30
This is an important flow feature. Slender wings have very low attached flow lift curve
slopes, and without the additional vortex lift it would be impractical to build configurations with
low aspect ratio wings. The low attached flow alone lift curve slope would prevent them from
being able to land at acceptable speeds or angle of attack. Vortex lift made the Concorde
possible. Another feature of the flow is the high angle of attack at which maximum lift occurs,
3/11/98

and typically a very mild lift loss past maximum lift. These features are a direct result of the
leading edge vortex flow structure that occurs on slender wings.
Subsonic lift
L.E. = 75
1.2
Vortex
flow
CL
Vortex lift
0.6
Attached flow
0.0
20
, deg
40
Figure 6-50. Vortex lift changes the characteristic lift development on wings. 30
Although the vortex lattice method formulation presented above does not include this effect,
vortex lattice methods are often used as the basis for extensions that do include the leading edge
vortex effects. A remarkable, reasonably accurate, flow model for leading edge vortex flows was
introduced by Polhamus31,32 at NASA Langley in 1966 after examining lots of data. This flow
model is known as the Leading Edge Suction Analogy. The concept is quite simple and was
invented for sharp edged wings. The leading edge suction that should exist according to attached
flow theory (see section 5.8) is assumed to rotate 90 and generate a vortex induced force instead
of a suction when leading edge vortex flow exists. Thus the vortex flow force is assumed to be
equal to the leading edge suction force. However, the force now acts in the direction normal to
the wing surface in the direction of lift rather than in the plane of the wing leading edge. The
concept is shown in the Fig. 6-51 from the original Polhamus NASA report.31 Further details on
the effects of vortex flow effects are also available in the reports by Kulfan.33,34
3/11/98

Potential flow (sharp edge)
CS
Potential flow (round edge)

CT /2
CS/2
CS
Separated flow
(sharp edge)
ttachment
point
CSS
Spiral
vortex
Section A-A
Figure 6-51. The Polhamus leading edge suction analogy.
Polhamus developed charts to compute the suction force for simple wing shapes. For a
delta wing with a sharp leading edge, the method is shown compared with the data of Bartlett
and Vidal35 in Figure 6-52. The agreement is quite good (my reconstruction doesnt show
agreement as good as that presented by Polhamus,31 but it is still impressive).
The figure also shows the large size of the vortex lift, and the nonlinear shape of the lift
curve when large angles are considered. This characteristic was exploited in the design of the
Concorde.
To find the vortex lift using the leading edge suction analogy, an estimate of the leading edge
suction distribution is required. However the suction analogy does not result in an actual
flowfield analysis including leading edge vortices. The Lamar vortex lattice code (VLMpc)
optionally includes a fully developed suction analogy based on Polhamus ideas, with extensions
to treat side edge suction by John Lamar3 also included.
Other approaches have been developed to compute leading edge vortex flows in more detail.
Many of these methods allow vortex filaments, simulating the leading edge vortices, to leave the
leading edge. The location of these vortices, and their effect on the wing aerodynamics as they
roll up are explicitly computed. Mook10 and co-workers are leaders in this methodology.
The area of vortex flows in configuration aerodynamics is fascinating, and an entire
conference was held at NASA Langley36 devoted to the topic. The references cited above were
selected to provide an entry to the literature of these flows. Interest in the area remains strong.
The effects of round leading edges have been investigated by Ericsson and Reding37 and
Kulfan.33 The relation between sweep, vortex lift, and vortex strength has been given recently by
Hemsch and Luckring. 38
3/11/98
2.00
AR = 1.5 ( = 69.4)
Prediction from Polhamus
Leading Edge Suction Analogy
1.50
CL
Experimental Data of
Bartlett and Vidal (ref. 35)
Vortex
Lift
1.00
0.500
Potential
Lift
0.00
0.0
10.0
20.0
30.0
40.0
Figure 6-52. Comparison of the leading edge suction analogy with data.
6.13 Alternate and Advanced VLM Methods
Many variations of the vortex lattice method have been proposed. They address both the
improvement in accuracy for the traditional case with a planar wake, and extensions to include
wake position and rollup as part of the solution. Areas requiring improvement include the ability
to predict leading edge suction, the explicit treatment of the Kutta condition, and the
improvement in convergence properties with increasing numbers of panels.8 The traditional
vortex lattice approaches assume that the wing wake remains flat and aligned with the
freestream. This assumption is acceptable for most cases. The effect of the wake on the wing that
generates it is small unless the wing is highly loaded. However, the interaction between the wake
from an upstream surface and a trailing lifting surface can be influenced by the rollup and
position.
In the basic case where the wake is assumed to be flat and at a specified location, the primary
extensions of the method have been directed toward improving the accuracy using a smaller
number of panels. Hough7 demonstrated that improvement in accuracy could be achieved by
using a lattice that was slightly smaller than the true planform area. Basically, he proposed a 1/4
panel width inset from the tips.
3/11/98

Perhaps the most important revision of the vortex lattice method was proposed by Lan6 , and
called the quasi vortex lattice method. In this method Lan used mathematical methods, rather
than the more heuristic arguments described above, to find an approximation for the thin airfoil
integral in the streamwise direction. The result was, in effect, a method where the vortex and
control point locations were established from the theory of Chebychev polynomials to obtain an
accurate estimate of the integrals with a small number of panels. The mathematically based
approach also led to an ability to compute leading edge suction very accurately.
The wake rollup and position problem has been addressed by Mook10 among others, and his
work should be consulted for details. A method similar to Mooks has been presented recently in
the book by Katz and Plotkin.11 They propose a vortex ring method, which has advantages when
vortices are placed on the true surface of a highly cambered shapes.
Unsteady flow extension
Analogous extensions have been made for unsteady flow. For the case of an assumed flat
wake the extension to harmonically oscillating surfaces was given by Albano and Rodden.39
When the vortex is augmented with an oscillating doublet, the so-called doublet-lattice method is
obtained. The doublet-lattice method is widely used for subsonic flutter calculations. Kalman,
Giesing and Rodden17 provide additional details and examples (they also included the steady
flow examples given above).
General unsteady flows calculations, including wake location as well as the incorporation of
leading edge vortices, have been carried out by Mook among others. The resulting codes have
the potential to be used to model time accurate aerodynamics of vehicles in arbitrary
maneuvering flight, including the high angle of attack cases of interest in fighter aerodynamics.
These codes are currently being used in studies where the aircraft aerodynamics is coupled with
advanced control systems. In this case the active control is incorporated and the dynamics of the
maneuver change dramatically due to the incorporation of a stability and control augmentation
system.
To sum up
The summary provided above illustrates the current state of affairs. Vortex lattice methods,
per se, are not being developed. However, they are being used in advanced methods where
several disciplines are being studied simultaneously and an affordable model of the
aerodynamics is required.
3/11/98

6.14 Exercises
1. Get a copy of VLMpc from the web site. The detailed instructions for this program are
included in Appendix D.6. Install the program on your personal computer and repeat the
sample case, checking that your output is the same as the sample output files on the web.
Study the output to familiarize yourself with the variety of information generated. Turn in a
report describing your efforts (not the output), including any mods required to make the code
run on your computer.
2. How good is thin airfoil theory? Compare the thin airfoil theory Cp for a 2D flat plate airfoil
with program VLMpc.
Flat plate thin airfoil theory:
CP = 4
(1 x/c )
x /c
i. Pick an aspect ratio 10 unswept wing at = 3 and 12 and run VLMpc.
ii. Plot (Cp)/ as a function of x/c at the wing root.
iii. How many panels do you need to get a converged solution from VLM?
iv. What conclusions do you reach?
3. Compare the validity of an aerodynamic strip theory using VLMpc. Consider an uncambered,
untwisted wing, AR = 4, = .4, le = 50, at a lift coefficient of 1. Plot the spanload, and the
Cp distribution at approximately the center section, the midspan station, and the 85%
semispan station. Compare your results with a spanload constructed assuming that the wing
flow is approximated as 2D at the angle of attack required to obtain the specified lift. Also
compare the chordloads, Cp, at the three span stations. How many panels do you need to
obtain converged results. Document your results. Do you consider this aerodynamic strip
theory valid based on this investigation? Comment.
4. Compare the wing aerodynamic center location relative to the quarter chord of the mac for the
wing in exercise 3, as well as similar wings. Consider one wing with zero sweep on the
quarter chord, and a forward swept wing with a leading edge sweep of -50. Compare the
spanloads. Document and analyze these results. What did you learn from this comparison?
5. For the wings in exercise 4, compare the section lift coefficients. Where would each one stall
first? Which wing appears to able to reach the highest lift coefficient before the section stalls.
6. For the problem in exercise 5, add twist to each wing to obtain near elliptic spanloads.
Compare the twist distributions required in each case.
7. Pick a NASA or NACA report describing wind tunnel results for a simple one or two lifting
surface configuration at subsonic speeds. Compare the lift curve slope and stability level
predicted by VLMpc with wind tunnel data. Submit a report describing your work and
assessing the results.
8. Add a canard to the aft and forward swept wings analyzed in exercise 4. Plot the sum of the
spanloads. How does the canard effect the wing spanload.
9. Consider the wings in exercise 8. How does lift change with canard deflection? Add an
equivalent tail. Compare the effect of tail or canard deflection on total lift and moment. Did
you learn anything? What?
10. Construct a design code using the 1/4 - 3/4 rule and compare with DESCAM.
11. Construct a little 2D code to study ground effects.
12. Compare wing and wing/tail(canard) results for CL with standard analytic formulas.
3/11/98
6.15 References
1
Vortex Lattice Utilization Workshop, NASA SP-405, May, 1976

2
Margason, R.J., and Lamar, J.E., Vortex-Lattice FORTRAN Program for Estimating Subsonic
Aerodynamic Characteristics of Complex Planforms, NASA TN D-6142, 1971
3
Lamar, J.E., and Gloss, B.B., Subsonic Aerodynamic Characteristics of Interacting Lifting
Surfaces With Separated Flow Around Sharp Edges Predicted by a Vortex-Lattice Method,
NASA TN D-7921,1975
4
Lamar, J.E., and Herbert, H.E., Production Version of the Extended NASA-Langley Vortex
Lattice FORTRAN Computer Program, Vol. I, Users Guide, (requires update packet, July,
1984) NASA TM 83303, 1982
5
Herbert, H.E, and Lamar, J.E., Production Version of the Extended NASA-Langley Vortex
Lattice FORTRAN Computer Program, Vol. II, Source Code, NASA TM 83304, 1982
6
Lan, C.E., A Quasi-Vortex-Lattice Method in Thin Wing Theory, Journal of Aircraft, Vol.
11, No. 9, September 1974, pp. 518-527.
7
Hough, Gary R., Remarks on Vortex-Lattice Methods, Journal of Aircraft, Vol. 10, No. 5,
May 1973, pp. 314-317.
8
DeJarnette, F.R., Arrangement of Vortex Lattices on Subsonic Wings, in Vortex Lattice
Utilization Workshop, NASA SP-405, May, 1976. pp. 301-319.
9
Frink, Neal T., Lifting-Surface Theory for Skewed and Swept Subsonic Wings, Journal of
Aircraft, Vol. 19, No. 7, July 1982, pp. 519-524.
10
Mook, D.T., and Nayfeh, A.H., Application of the Vortex-Lattice Method to High-Angle-ofAttack Subsonic Aerodynamics, SAE Paper No. 851817, October, 1985.
11
Katz, J., and Plotkin, A., Low-Speed Aerodynamics: From Wing Theory to Panel Methods,
McGraw-Hill, Inc., New York, 1991.
12
Jacob Kay, W.H. Mason, W. Durham, F. Lutze and A. Benoliel, Control Power Issues in
Conceptual Design: Critical Conditions, Estimation Methodology, Spreadsheet Assessment,
Trim and Bibliography, VPI-Aero-200, November 1993.
13
Karamcheti , K., Principles of Ideal Fluid Aerodynamics, John Wiley & Sons, New York,
1966, pp: 518
14
Bertin, J.J., and Smith, M.L., Aerodynamics for Engineers, Prentice-Hall, Inc., Englewood
Cliffs, 2nd Ed., 1989, pp: 261-282, 3rd Ed. 1998, pp: 291-311.
15
Thomas, R.W., Analysis of Aircraft Stability and Control Design Methods, AFWAL-TR-843038, Vol. II, App. B., Evaluation of Aerodynamic Panel Methods, by John Koegler, May,
1984.
16
Pittman, J.L., and Dillon, J.L., Vortex Lattice Prediction of Subsonic Aerodynamics of
Hypersonic Vehicle Concepts, Journal of Aircraft, Vol. 14, No. 10, pg 1017, October, 1977.
17
Kalman, T.P., Rodden, W.P., and Giesing, J., Application of the Doublet-Lattice Method to
Nonplanar Configurations in Subsonic Flow, Journal of Aircraft, Vol. 8, No. 6, June 1971, pp.
406-415.
18
Kchemann, D., The Aerodynamic Design of Aircraft, Pergamon Press, Oxford, 1978.
19
Tinling, B.E., and Kolk, W. R., The Effects of Mach Number and Reynolds Number on the
Aerodynamic Characteristics of Several 12-Percent-Thick Wings Having 35 of Sweepback and
Various Amounts of Camber, NACA RM A50K27, Feb. 1951.
3/11/98
20
Lamar, J.E., A Vortex Lattice Method for the Mean Camber Shapes of Trimmed NonCoplanar Planforms with Minimum Vortex Drag, NASA TN D-8090, June 1976.
21
Loving, D.L., and Estabrooks, B.B., Transonic Wing Investigation in the Langley Eight Foot
High Speed Tunnel at High Subsonic Mach Numbers and at Mach Number of 1.2, NACA RM
L51F07, 1951.
22
Mason, W.H., MacKenzie, D.A., Stern, M.A., Ballhaus, W.F, Jr., and Frick, J., Automated
Procedure for Computing the Three-Dimensional Transonic Flow Over Wing-Body
Combinations, including Viscous Effects, Vol. I, Description of Methods and Applications,
AFFDL-TR-77-122, February 1978.
23
Cook, William C., The Road to the 707, TYC Publishing, Bellevue, 1991.
24
Irving, Clive, Wide-Body: The Triumph of the 747, William Morrow, New York, 1993.
25
Snyder, George, Structural Design Problems in the B-29 Airplane, Aeronautical Engineering
Review, Feb. 1946, pp. 9-12.
26
Hunton, Lynn W., A Study of the Application of Airfoil Section Data to the Estimation of the
High-Subsonic-Speed Characteristics of Swept Wings, NACA RM A55C23, June 1955.
27
Shevell, R.S., and Schaufele, R.D., Aerodynamic Design Features of the DC-9, Journal of
Aircraft, Vol. 3, No. 6, Nov.-Dec. 1966, pp. 515-523.
28
Nguyen, L.T., Ogburn, M.E., Gilbert, W.P., Kibler, K.S., Brown, P.W., and Deal, P.L.,
Simulator Study of Stall/Post-Stall Characteristics of a Fighter Airplane With Relaxed
Longitudinal Static Stability, NASA TP 1538, Dec. 1979.
29
Payne, F.W., and Nelson, R.C., An Experimental Investigation of Vortex Breakdown on a
Delta Wing, in Vortex Flow Aerodynamics, NASA CP 2416, 1985.
30
Polhamus, E.C., Application of Slender Wing Benefits to Military Aircraft, AIAA Wright
Brothers Lecture, AIAA-83-2566, 1983..
31
Polhamus, E.C., A Concept of the Vortex Lift on Sharp Edge Delta Wings Based on a
Leading-edge Suction Analogy, NASA-TN 3767, 1966.
32
Polhamus, E.C., Prediction of Vortex Lift Characteristics by a Leading-edge Suction
Analogy,, Journal of Aircraft, Vol. 8, No. 4, 1971, pp. 193-199.
33
Kulfan, R.M., Wing Airfoil Shape Effects on the Development of Leading-Edge Vortices,
AIAA 79-1675, 1979.
34
Kulfan, R.M., Wing Geometry Effects on Leading Edge Vortices, AIAA Paper No. 79-1872,
1979.
35
Bartlett, G.E., and Vidal, R.J., Experimental Investigation of Influence of Edge Shape on the
Aerodynamic Characteristics of Low Aspect Ratio Wings at Low Speeds, Journal of the
Aeronautical Sciences, Vol. 22, No. 8, August, 1955, pp.517-533,588.
36
Vortex Flow Aerodynamics, NASA SP-2416 (volume I), and NASA SP-2417 (volume II),
October, 1985.
37
Ericsson, L.E., and Reding, J.P., Nonlinear Slender Wing Aerodynamics, AIAA Paper No.
76-19, January, 1976.
38
Hemsch, M.J., and Luckring, J.M., Connection Between Leading-Edge Sweep, Vortex Lift,
and Vortex Strength for Delta Wings, Journal of Aircraft, Vol. 27, No. 5, May 1990, pp. 473475.
39
Albano, E., and Rodden, W.P., A Doublet-Lattice Method for Calculating Lift Distributions
on Oscillating Surfaces in Subsonic Flows, AIAA J., Vol. 7, No. 2, February 1969, pp. 279-285;
errata AIAA J., Vol. 7, No. 11, November 1969, p. 2192.
3/11/98
8. Introduction to
Computational Fluid Dynamics
We have been using the idea of distributions of singularities on surfaces to study the
aerodynamics of airfoils and wings. This approach was very powerful, and provided us with
methods which could be used easily on PCs to solve real problems. Considerable insight into
aerodynamics was obtained using these methods. However, the class of effects that could be
examined was somewhat restricted. In particular, practical methods for computing fundamentally
nonlinear flow effects were excluded. This includes both inviscid transonic and boundary layer
flows.
In this chapter we examine the basic ideas behind the direct numerical solution of differential
equations. This approach leads to methods that can handle nonlinear equations. The simplest
methods to understand are developed using numerical approximations to the derivative terms in
the partial differential equation (PDE) form of the governing equations. Direct numerical
solutions of the partial differential equations of fluid mechanics constitute the field of
computational fluid dynamics (CFD). Although the field is still developing, a number of books
have been written.1,2,3,4,5,6 In particular, the book by Tannehill et al,1 which appeared in 1997 as a
revision of the original 1984 text, covers most of the aspects of CFD theory used in current codes
and reviewed here in Chapter 14. Fundamental concepts for solving partial differential equations
in general using numerical methods are presented in a number of basic texts. Smith7 and Ames 8
are good references.
The basic idea is to model the derivatives by finite differences. When this approach is used
the entire flowfield must be discretized, with the field around the vehicle defined in terms of a
mesh of grid points. We need to find the flowfield values at every mesh (or grid) point by writing
down the discretized form of the governing equation at each mesh point. Discretizing the
equations leads to a system of simultaneous algebraic equations. A large number of mesh points
is usually required to accurately obtain the details of the flowfield, and this leads to a very large
system of equations. Especially in three dimensions, this generates demanding requirements for
computational resources. To obtain the solution over a complete three dimensional aerodynamic
configuration millions of grid points are required!
3/17/98
8 -1

In contrast to the finite difference idea, approximations to the integral form of the governing
equations result in the finite volume approach. A book has been written recently devoted solely to
this approach,9 and we will cover this approach briefly here.
Thus CFD is usually associated with computers with large memories and high processing
speeds. In addition, massive data storage systems must be available to store computed results,
and ways to transmit and examine the massive amounts of data associated with a computed result
must be available. Before the computation of the solution is started, the mesh of grid points must
be established. Thus the broad area of CFD leads to many different closely related but
nevertheless specialized technology areas. These include:
grid generation
flowfield discretization algorithms
efficient solution of large systems of equations
massive data storage and transmission technology methods
computational flow visualization
Originally, CFD was only associated with the 2 nd and 3rd items listed above. Then the
problem with establishing a suitable mesh for arbitrary geometry became apparent, and the
specialization of grid generation emerged. Finally, the availability of large computers and remote
processing led to the need for work in the last two items cited. Not generally included in CFD
per se, a current limiting factor in the further improvement in CFD capability is development of
accurate turbulence models, discussed in Chapter 10.
This chapter provides an introduction to the concepts required for developing discretized
forms of the governing equations and a discussion of the solution of the resulting algebraic
equations. For the most part, we adopt the viewpoint of solving equilibrium (elliptic) problems.
This in contrast to the more frequent emphasis on solving hyperbolic systems. Although the
basic idea of CFD appears straightforward, once again we find that a successful numerical
method depends on considerable analysis to formulate an accurate, robust, and efficient solution
method. We will see that the classification of the mathematical type of the governing equations
(Sec. 2.8) plays an important role in the development of the numerical methods. Although we
adopt finite difference/finite volume methods to solve nonlinear equations, to establish the basic
ideas we consider only linear equations. Application to nonlinear equations is addressed in
Chapters 10, 11 and 12, where additional concepts are introduced and applied to the solution of
nonlinear equations. Chapter 13 describes the most advanced approaches currently in use.
8.1 Approximations to partial derivatives
There are many ways to obtain finite difference representations of derivatives. Figure 8-1
illustrates the approach intuitively. Suppose that we use the values of f at a point x0 and a point a
3/17/98
Introduction to CFD 8 - 3
distance x away. Then we can approximate the slope at x0 by taking the slope between these
points. The sketch illustrates the difference between this simple slope approximation and the
actual slope at the point x0. Clearly, accurate slope estimation dependents on the method used to
estimate the slope and the use of suitably small values of x.
True slope
at X 0
Approximate slope
at X 0
X
x0
Figure 8-1. Example of slope approximation using two values of the function.
Approximations for derivatives can be derived systematically using Taylor series
expansions. The simplest approach is to find an estimate of the derivative from a single series.
Consider the following Taylor series:
df
f (x0 + x) = f(x 0 ) + x
dx
and rewrite it to solve for
+
x0
(x )2 d 2f
2
dx2 x
0
(x )3 d 3f
6
dx3 x
0
+ .. .
(8-1)
df
:
dx x 0
f ( x0 + x ) f ( x0 )
df
1 d 2f
=
x
...
dx x 0
x
2 dx2 x
0
or:
f ( x0 + x ) f ( x0 )
df
=
+ O
(x
12
3)
dx x 0
x
(8-2)
Truncation
Error
where the last term is neglected and called the truncation error. In this case it is O(x). The term
truncation error means that the error of the approximation vanishes as x goes to zero. * The
*
This assumes that the numerical results are exactly accurate. There is a lower limit to the size of the difference step
in x due to the use of finite length arithmetic. Below that step size, roundoff error becomes important. In most
3/17/98

form of the truncation error term is frequently important in developing numerical methods. When
the order of the truncation error is O(x), the approximation is described as a first order
accurate approximation, and the error is directly proportional to x. The other characteristic of
this representation is that it uses only the information on one side of x0, and is thus known as a
one-sided difference approximation. Finally, because it uses information ahead of x0, its known
as a forward difference. Thus, Eq.(8-2) is a first order, one sided, forward difference
approximation to the derivative.
We could also write the approximation to the derivative using information prior to the point
of interest. The corresponding first order accurate one sided backward difference approximation
is obtained by expanding the Taylor series to a point prior to the point about which the expansion
is carried out. The resulting expansion is:
df
(x )2 d 2f
(x )3 d 3f
f (x0 x) = f(x0 ) x
+
+ . . ..
dx x 0
2 dx2 x
6 dx3 x
0
0
(8-3)
Solving for the first derivative in the same manner we used above, we obtain:
f ( x0 ) f ( x0 x)
df
=
+ O( x ) ,
dx x 0
x
(8-4)
the first order accurate, one sided, backward difference approximation.

Note from Fig. 8-1 above that one sided differences can lead to a fairly large truncation error.
In many cases a more accurate finite difference representation would be useful. To obtain a
specified level of accuracy, the step size x must be made small. If a formula with a truncation
term of O(x)2 is used,* the required accuracy can be obtained with significantly fewer grid
points. A second order, O(x)2, approximation can be obtained by subtracting the Taylor series
expansions, Eq.(8-3) from Eq.(8-1):
df
(x )3 d3f
f (x0 + x) f(x0 x) =+2x
+
+ . . ..
dx x 0
3 dx3 x
0
Here the O(x) terms cancel in the subtraction. When we divide by 2x and solve for the first
derivative, we get an expression with a truncation error of O(x)2. The resulting expression for
the derivative is:
cases the stepsize used for practical finite difference calculations is larger than the limit imposed by roundoff errors.
We cant afford to compute using grids so finely spaced that roundoff becomes a problem.
*
With x small, x2 is much smaller than x
3/17/98
f ( x0 + x ) f ( x0 x )
df
=
+ O( x )2 .
dx x 0
2x
(8-5)
This is a second order accurate central difference formula since information comes from both
sides of x0. Numerous other approximations can be constructed using this approach. Its also
possible to write down second order accurate forward and backward difference approximations.
We also need the finite difference approximation to the second derivative. Adding the Taylor
series expressions for the forward and backward expansions, Eq.(8-1) and Eq.(8-3), results in the
following expression, where the odd order terms cancel:
f (x0 + x) + f(x 0 x) = 2 f (x0 ) + (x )2
d 2f
+ O (x ) 4
2
dx x
0
Solving for the second derivative yields:

f ( x0 + x ) 2 f (x0 ) + f ( x0 x )
d2 f
(8-6)
=
+ O( x )2 .
2
2
dx x
(x)
0
The formulas given above are the most frequently used approximations to the derivatives
using finite difference representations. Other methods can be used to develop finite difference
approximations. In most cases we want to use no more than two or three function values to
approximate derivatives.
Forward and backward finite difference approximations for the second derivative can also be
derived. Note that formally these expressions are only first order accurate. They are:
a forward difference expression:
f ( x0 ) 2 f (x 0 +x) + f ( x0 + 2x )
d2 f
=
+ O (x )
dx2 x
(x)2
0
(8-7)
a backward difference expression:

f ( x0 ) 2 f (x 0 x) + f ( x0 2x )
d2 f
=
+ O( x ) .
2
dx x
(x)2
0
(8-8)
In addition, expressions can be derived for cases where the points are not evenly distributed.
In general the formal truncation error for unevenly spaced points is not as high as for the evenly
spaced point distribution. In practice, for reasonable variations in grid spacing, this may not be a
serious problem. We present the derivation of these expressions here. A better way of handling
non-uniform grid points is presented in the next chapter. The one sided first derivative
expressions Eq.(8-2) and Eq.(8-4) are already suitable for use in unevenly spaced situations. We
need to obtain a central difference formula for the first derivative, and an expression for the
3/17/98

second derivative. First consider the Taylor expansion as given in Eqs. (8-1) and (8-3). However,
the spacing will be different in the two directions. Use x+ and x to distinguish between the
two directions. Eqs. (8-1) and (8-3) can then be rewritten as:
( )
x +
d 2f
+
6
dx2 x
0
( )
x
d 2f
6
dx2 x
0
x
df
f (x0 x ) = f (x0 ) x
+
dx x0
2
( )
x +
+
+ df
f (x0 + x ) = f (x0 ) + x
+
dx x0
2
( )
d 3f
+ . ..
dx3 x
0
(8-9)
d 3f
+...
dx3 x
0
(8-10)
Define x+ = x . To obtain the forms suitable for derivation of the desired expressions,
replace x+ in Eq. (8-9) with x , and multiply Eq. (8-10) by . The resulting expressions
are:
x )
(
+
f (x0 + x ) = f (x0 ) + x
df
dx x 0
dx2 x
0
x )
(
+
x )
(
+
d 2f
2
( )
d 3f
+ . ..
dx3 x
0
(8-11)
x d 3f
d 2f
f (x 0 x ) = f (x0 ) x
+ . . . (8-12)
dx x 0
2
6
dx 2 x
dx3 x
0
0
To obtain the expression for the first derivative, subtract Eq ( 8-12) from Eq. (8-11).
df
f (x0 + x + ) f (x 0 x ) = f (x0 ) f(x 0 ) + 2x
df
dx x0
( )
2
2 2
x
x
d f
+
+...
2
2 dx2
x0
(8-13)
and rearrange to solve for df/dx:

df
f (x0 + x + ) + ( 1) f (x 0 ) f(x0 x )
=
+ O(x )
dx x 0
2x
(8-14)
To obtain the expression for the second derivative, add (8-11) and (8-12):
( )
2
2 2
x
x
d f
f (x0 + x + ) +f (x 0 x ) = f (x0 ) + f(x 0 ) +
+
+ O(x)3 . ..
2
2
2
dx x
0
which is then solved for d2f/dx2:
3/17/98
d 2f
f (x0 + x + ) (1 + ) f (x 0 ) +f(x0 x )
=
+ O(x )
2
2
dx x
(1+ ) x
0
2
( )
(8-15)
Note that both Eqs. (8-14) and (8-15) reduce to the forms given in Eq.(8-5) and Eq.(8-6)
when the grid spacing is uniform.
Finally, note that a slightly more sophisticated analysis (Tannehill, et al,1 pages 61-63) will
lead to a second order expression for the first derivative on unevenly spaced points:
f (x0 + x + ) + 2 1 f (x0 ) 2 f(x 0 x )

df
=
+ O(x )2
dx x 0
( +1)x
(8-16)
Tannehill, et al,1 give additional details and a collection of difference approximations using
more than three points and difference approximations for mixed partial derivatives (Tables 3-1
and 3-2 on their pages 52 and 53). Numerous other methods of obtaining approximations for the
derivatives are possible. The most natural one is the use of a polynomial fit through the points.
Polynomials are frequently used to obtain derivative expressions on non-uniformly spaced grid
points.
These formulas can also be used to represent partial derivatives. To simplify the notation, we
introduce a grid and a notation common in finite difference formulations. Figure 8-2 illustrates
this notation using x = y = const for these examples.
Assume:
i,j+1
x = y = const.
x = ix
y = jx
y or "j"
i-1,j
i,j
i+1,j
i,j-1
x or "i"
Figure 8-2. Nomenclature for use in partial differential equation difference expressions.
3/17/98

In this notation the following finite difference approximations for the first derivatives are:
f fi+1,j fi,j
=
+ O( x )
1st order forward difference
(8 -17)
x
x
f fi, j fi1,j
=
+ O( x )
1st order backward difference (8 -18)
x
x
f fi+1,j fi1, j
=
+ O( x )2
2nd order central difference
(8 -19)
x
2x
and the second derivative is:
2 f fi+1,j 2 fi,j + fi1, j
+ O( x )2
2 =
2
x
( x)
2nd order central difference
(8 - 20)
Similar expressions can be written for the y derivatives. To shorten the expressions, various
researchers have introduced different shorthand notations to replace these expressions. The
shorthand notation is then used in further operations on the difference expressions.
8.2 Finite difference representation of Partial Differential Equations (PDE's)
We can use the approximations to the derivatives obtained above to replace the individual
terms in partial differential equations. The following figure provides a schematic of the steps
required, and some of the key terms used to ensure that the results obtained are in fact the
solution of the original partial differential equation. We will define each of these new terms
below.
Steps and Requirements To Obtain a Valid Numerical Solution

Governing
Partial
Differential
Equation
Discretization
Consistency
?
Exact
Solution
System
of
Algebraic
Equations
Stability
Convergence
as x,t 0
Approximate
Solution
Figure 8-3. Overall procedure used to develop a CFD solution procedure.
3/17/98
Successful numerical methods for partial differential equations demand that the physical
features of the PDE be reflected in the numerical approach. The selection of a particular finite
difference approximation depends on the physics of the problem being studied. In large part the
type of the PDE is crucial, and thus a determination of the type, i.e. elliptic, hyperbolic, or
parabolic is extremely important. The mathematical type of the PDE must be used to construct
the numerical scheme for approximating partial derivatives. Some advanced methods obscure the
relationship, but it must exist. Consider the example given in Fig. 8-4 illustrating how
information in a grid must be used.
subsonic flow
supersonic flow
i,j+1
i,j+1
zone of
dependence
i-1,j
i,j
i,j-1
i,j depends on all

neighboring points
(elliptic system)
i+1,j
i-1,j
i,j
i+1,j
i,j-1
i,j depends only on the points

in the zone of dependence
(hyperbolic system)
Note: velocity direction relative to

the grid becomes important
Figure 8-4. Connection between grid points used in numerical method and equation type.
Any scheme that fails to represents the physics correctly will fail when you attempt to obtain
a solution. Furthermore, remember, in this case we were looking at a uniformly spaced cartesian
grid. In actual real life applications we have to consider much more complicated non-uniform
grids in non-Cartesian coordinate systems. In this section we use simple uniform Cartesian grid
systems to illustrate the ideas. The necessary extensions of the methods illustrated in this chapter
are outlined in the next chapter.
In Fig. 8-3 above, we introduced several important terms requiring definition and discussion:
discretization
consistency
stability
convergence
3/17/98

Before defining the terms, we provide an example using the heat equation:
u
2u
(8-21)
= 2 .
t
x
We discretize the equation using a forward difference in time, and a central difference in
space following the notation shown in the following sketch:
x
i+1
i
i-1
n-1
n
n+1
Figure 8-5. Grid nomenclature for discretization of heat equation.
The heat equation can now be written as:
u
2u
2 =
t424
x3
1
PDE
uin+1 uin
n
n
ui+1
2uin + ui1
+
2
t
(
x
)
1444 444 2444 444 3
FDE
2 n
4 n
2
u t + u (x ) + ... = 0
2
x 4 i 12
t i 2
1444 4442444 4443
(8-22)
Truncation Error
where we use the superscript to denote time and the subscript to denote spatial location. In Eq.
(8-18) the partial differential equation (PDE) is converted to the related finite difference equation
(FDE). The truncation error is O( t) + O ( x)2 or O t, ( x)2 . An understanding of the

truncation error for a particular scheme is important.
Using the model equation give here, we define the terms in the schematic given above:
discretization
This is the process of replacing derivatives by finite difference approximations. Replace
continuous derivatives with an approximation at a discrete set of points (the mesh). This
introduces an error due to the truncation error arising from the finite difference approximation
3/17/98
and any errors due to treatment of BCs. A reexamination of the Taylor series representation is
worthwhile in thinking about the possible error arising from the discretization process:
f f ( x 0 + x) f ( x0 x)
=
+
x
2 x
x 2 3f
3
6 2x
1
4
4
3
(8-5a)
formally valid for x0,

but when x = finite, 3 f /x 3
can be big for rapidly changing
solutions (shock wave cases)
Thus we see that the size of the truncation error will depend locally on the solution. In most
cases we expect the discretization error to be larger than round-off error.
consistency
A finite-difference representation of a PDE is consistent if the difference between the PDE
and its difference representation vanishes as the mesh is refined, i.e.,
lim
(PDE FDE ) = lim ( T.E.) = 0
mesh 0
(8-23)
mesh0
When might this be a problem? Consider a case where the truncation error is O(t/x). In
this case we must let the mesh go to zero just such that:
t
lim = 0
t,x0 x
(8-24)
Some finite difference representations have been tried that werent consistent. An example
cited by Tannehill, et al,1 is the DuFort-Frankel differencing of the wave equation.
stability
A stable numerical scheme is one for which errors from any source (round-off, truncation)
are not permitted to grow in the sequence of numerical procedures as the calculation proceeds
from one marching step, or iteration, to the next, thus:
errors grow unstable
errors decay stable
and
Stability is normally thought of as being associated with
marching problems.
Stability requirements often dictate allowable step sizes.
In many cases a stability analysis can be made to define the
stability requirements.
3/17/98

convergence
The solution of the FDEs should approach the solution of the PDE as the mesh is refined. In
the case of a linear equation there is a theorem which proves that the numerical solution to the
FDE is in fact the solution of the original partial differential equation.
Lax Equivalence Theorem1 (linear, initial value problem): For a properly posed
problem, with a consistent finite difference representation, stability is the necessary
and sufficient condition for convergence.
In practice, numerical experiments must be conducted to determine if the solution appears to
be converged with respect to mesh size. * Machine capability and computing budget (time as well
as money) dictate limits to the mesh size. Many, many results presented in the literature are not
completely converged with respect to the mesh.
So far we have represented the PDE by an FDE at the point i,n. The PDE is now a set of
algebraic equations written at each mesh point. If the grid is (in three dimensions) defined by a
grid with IMAX, JMAX and KMAX mesh points in each direction, then we have a grid with
IMAX JMAX KMAX grid points. This can be a very large number. A typical recent case
computed by one of my students was for the flow over a simple aircraft forebody. The
calculation required 198,000 grid points. Thus the ability to carry out aerodynamic analysis using
finite difference methods depends on the ability to solve large systems of algebraic equations
efficiently.
We need to obtain the solution for the values at each grid point. We now consider how this is
actually accomplished. Since the computer requirements and approach are influenced by the
mathematical type of the equation being solved, we illustrate the basic types of approaches to the
solution with two examples.
1st example - typical parabolic/hyperbolic PDEs
Explicit Scheme: Consider the finite difference representation of the heat equation given
above in Eq. (8-18). Using the notation shown in the Fig. 8-6 below, we write the finite
difference representation as:
un+1
uni
n
n
n
i
=
2 ui+1 2ui + ui1
t
(x)
(8-25)
and the solution at time step n is known. At time n+1 there is only one unknown.
This is convergence with respect to grid. Another convergence requirement is associated with the satisfaction of the
solution of a system of equations by iterative methods on a fixed grid.
3/17/98
x
i+1
x
x
i-1
n-1 n n+1
Figure 8-6. Grid points used in typical explicit calculation.
We solve for the value of u at i and the n+1 time step:
t n
n
n
un+1
= uin +
i
2 ui +1 2ui + ui1
(x )
(8-26)
and thus at each i on n+1 we can solve for un+1

algebraically, without solving a system of
i
equations. This means that we can solve for each new value explicitly in terms of known values
from the previous time step. This type of algorithm is known as an explicit scheme. It is a very
straight forward procedure. To summarize:
The algebra is simple.
The bad news for non-vector computers: stability requirements require very
small steps sizes.
The good news: this scheme is easily vectorized* and a natural for massively
parallel computation.
Implicit Scheme: Now consider an alternate finite difference representation of the heat
equation given above in Eq. (8-18). Use the notation shown in the Fig. 8-7 below to define the
location of grid points used to define the finite difference representation.
x
i+1
i
i-1
n-1 n n+1
Figure 8-7. Grid points used in typical implicit calculation.
see Chapter 3 for a brief discussion of vectorization.
3/17/98

Now we write the finite difference representation as:
un+1
uni
n+1
n+1
i
=
ui+1
2uin+1 + ui1
2
t
(x)
(8-27)
where we use the spatial derivative at time n+1. By doing this we obtain a system where at each i
on n+1, un+1
depends on all the values at n+1. Thus we need to find the values along n+1
i
simultaneously. This leads to a system of algebraic equations that must be solved. For our model
problem this system is linear. We can see this more clearly by rearranging Eq. (8-27). Defining
=
t
( x )2
(8-28)
we can re-write Eq.(8-27) after some minor algebra as:

n+1
n+1
ui1
+ (1 + 2 )uin+1 ui+1
= uin
for i = 1,..., N .
(8-29)
This can be put into a matrix form to show that it has a particularly simple form:
(1 + 2 )

(1 + 2)
O
O
O
O
O
O
0
0
0
0
O
O
0
0
0
O
(1 + 2 )
O
0
0
0
O
O
0
0 u1n+1 u1n
0
0 u2n+1 u2n
O
O M M
O
O uin+1 = uin
O
O M M
n+1 n
(1+ 2)
uN
1 u N1
(1+ 2) uNn+1 u nN
(8-30)
Equation (8-30) is a special type of matrix form known as a tridiagonal form. A particularly
easy solution technique is available to solve this form. Known as the Thomas algorithm, the
details are described in Section 8.5 and a routine called tridag is described in Appendix H-1.
Many numerical methods are tailored to be able to produce this form.
The approach that leads to the formulation of a problem requiring the simultaneous solution
of a system of equations is known as an implicit scheme. To summarize:
The solution of a system of equations is required at each step.
The good news: stability requirements allow a large step size.
The not so good news: this scheme is harder to vectorize/parallelize.
A common feature for both explicit and implicit methods for parabolic and hyperbolic
equations:
A large number of mesh points can be treated, you only need the values at a
small number of marching stations at any particular stage in the solution.
This means you can obtain the solution with a large number of grid points
using a relatively small amount of memory. Curiously, some recent codes
dont take advantage of this last fact.
3/17/98
2nd Example - elliptic PDE
We use Laplaces equation as the model problem for elliptic PDEs:
xx + yy = 0
(8-31)
and consider the grid shown below in Figure 8-8.
y
j+1
j
j-1
i-1
i+1
Figure 8-8. Grid points used in a typical representation of an elliptic equation.

Use the second order accurate central difference formulas at i,j:
xx =
i+1, j 2i,j + i1, j

+ O( x )2
2
(x )
(8-32)
yy =
i, j+1 2i,j + i,j1

+O (y)2 ,
2
(y)
(8-33)
and:
and substitute these expressions into the governing equation:

i+1, j 2i,j + i1, j
( x)
i, j+1 2i, j + i, j1
(y)
=0
(8-34)
Solve this equation for ij:

i,j =
(y)2
2 ( x)2 + (y )2
i+1, j + i 1, j +
(x)2
(i,j +1 + i,j1)
2 (x)2 + (y)2
(8-35)
where if x = y:
i,j =
3/17/98
+ i1, j + i, j+1 + i,j 1

4 i+1, j
(8-36)

This expression illustrates the essential physics of flows governed by elliptic PDEs:
ij depends on all the values around it
all values of must be found simultaneously
computer storage requirements are much greater than those required for
parabolic/hyperbolic PDEs
Because of the large number of mesh points required to resolve the flowfield details, it is
generally not practical to solve the system of equations arising from applying the above equation
at each mesh point directly. Instead, an iterative procedure is usually employed. In this procedure
an initial guess for the solution is made and then each mesh point in the flowfield is updated
repeatedly until the values satisfy the governing equation. This iterative procedure can be
thought of as having a time-like quality, which has been exploited in many solution schemes to
find the steady flowfield.
A Note on Conservation Form
Care must be taken if the flowfield has discontinuities (shocks). In that case the correct
solution of the partial differential equation will only be obtained if the conservative forms of the
governing equations are used.
8.3 Other approaches, including the finite volume technique
Finite difference methods are the most well known methods in CFD. However other methods
have also proven successful, and one method in particular, the finite volume technique, actually
forms the basis for most current successful codes. The other methods in use are categorized as
finite element and spectral. Each method eventually leads to a large set of algebraic equations,
just as with the finite difference methods. See References 1 and 3 for more details of the latter
two methods. In US aircraft aerodynamics work they dont currently have an impact.
The finite volume method is important. Instead of discretizing the PDE, select the integral
form of the equations. Recall that each conservation law had both differential and integral
statements. The integral form is more fundamental, not depending on continuous partial
derivatives.
Example of Finite Volume Approach (Fletcher,3 vol. I, pg.105-116, Tannehill, et al,1 pg 71-76)
Consider the general conservation equation (in two dimensions for our example analysis):
q F G
+
+
=0.
t x y
Pick the particular form to be conservation of mass:
3/17/98
(8-37)
q=
F = u,
G = v
(8-38)
and recall that this conservation law could also come from the integral statement:
dV = V ndS .
t
(8-39)
Introducing the notation defined above and assuming two dimensional flow, the conservation
law can be rewritten as:
where
qdV + H n dS = 0
t
(8-40)
H = (F,G) = V
(8-41)
Hx = F = u
.
Hy = G = v
(8-42)
and
y, j
n
A
B
n
x, i
Figure 8-9. Basic nomenclature for finite volume analysis.
Using the definition of n in Cartesian coordinates, and considering for illustration the
Cartesian system given in Fig. 8-9, we can write:
H ndS = H xi + H y j n dS
= (Fi + Gj) ndS
(8-43)
along AB, n = -j, dS = dx, and:

H ndS = Gdx
(8-44)
H ndS = F dy
(8-45)
along BC, n = i, dS = dy, and:
3/17/98

or in general:
H ndS = Fdy Gdx .
(8-46)
Using the general grid shown in the Fig. 8-10, our integral statement, Eq. (8-40) can be
written as:
DA
Aq j,k + ( Fy Gx) = 0.
t
AB
(8-47)
Here A is the area of the quadrilateral ABCD, and qi,j is the average value of q over ABCD.
j+1
k+1
j
C
j-1
k
D
cell centered
at j,k
B
k -1
x
Figure 8-10. Circuit in a general grid system.
Now define the quantities over each face. For illustration consider AB:
yAB = yB yA
xAB = xB x A
1
FAB = Fj,k 1 + Fj,k ,
2
1
GAB = Gj,k +1 + Gj,k
2
and so on over the other cell faces.
3/17/98
(8-48)
Assuming A is not a function of time, and combining:
q j,k 1
1
+ Fj,k 1 + Fj,k yAB Gj,k 1 + Gj,k x AB
2
2
t
1
1
+ Fj,k + Fj+1,k yBC G j,k + G j+1,k xBC
2
2
1
1
(8-49)
+ Fj,k + Fj,k+1 yCD G j,k + G j,k+1 xCD .
2
2
1
1
+ Fj1,k + Fj,k yDA G j 1,k + Gj,k x DA
2
2
=0
Supposing the grid is regular cartesian as shown in Fig. 8-11. Then A = xy, and along:
BC:
CD:
y = 0, x AB = x
x = 0, yBC = y
.
y = 0, x CD = x
DA:
x = 0, yDA = y
AB:
(8-50)
y
k +1
k -1
j
j+1
j-1
Figure 8-11. General finite volume grid applied in Cartesian coordinates.
Thus, in Eq. (8-49) we are left with:
q j,k
1
1
Gj,k 1 + Gj,k x + Fj,k + Fj+1,k y
2
2
t
.
1
1
+ Gj,k + Gj,k +1 x Fj1,k + Fj,k y = 0
2
2
xy
Collecting terms:
3/17/98
(8-51)

q j,k
t
Fj+1,k Fj1,k G j,k+1 G j,k1

+
=0
2x
2y
144
44 4
42444
4 44
3
for this reversion to Cartesian
coordinates the equation just reduces
to simple central differences of the original
partial differential equation
(8-52)
or:
q F G
+
+
=0.
t x y
(8-53)
Thus, and at first glance remarkably, the results of the finite volume approach can lead to the
exact same equations to solve as the finite difference method on a simple Cartesian mesh.
However, the interpretation is different:*
Finite difference: approximates the governing equation at a point
Finite volume:
"
"
"
"
over a volume
Finite volume is the most physical in fluid mechanics codes, and is actually used in
most codes.
Finite difference methods were developed earlier, the analysis of methods is easier
and further developed.
Both the finite difference and finite volume methods are very similar. However, there are
differences. They are subtle but important. We cite three points in favor of the finite volume
method compared to the finite difference method:
Good conservation of mass, momentum, and energy using integrals when mesh is
finite size
Easier to treat complicated domains (integral discretization [averaging] easier to
figure out, implement, and interpret)
Average integral concept much better approach when the solution has shock waves
(i.e. the partial differential equations assume continuous partial derivatives).
Finally, special considerations are needed to implement some of the boundary conditions in
this method. The references, in particular Fletcher,3 should be consulted for more details.
8.4 Boundary conditions
So far we have obtained expressions for interior points on the mesh. However, the actual
geometry of the flowfield we wish to analyze is introduced through the boundary conditions. We
use an elliptic PDE problem to illustrate the options available for handling boundary conditions.
Consider the flow over a symmetric airfoil at zero angle of attack, as shown in Fig. 8-12.
Summarized from Professor B. Grossmans unpublished CFD notes.
3/17/98
X
Figure 8-12. Example of boundary condition surface requiring consideration.
-
Here, because there is no lift, symmetry allows us to solve only the top half of the region. If
is a perturbation potential [see Chap. 2, Eq. (2-123)],
U = U + x
,
V = y
(8-54)
then far away from the surface,

u=v= 0
(8 55)
0 as x 2 + y 2 .
(8 56)
or
For a lifting airfoil, the farfield potential must take the form of a potential vortex singularity
with a circulation equal to the circulation around the airfoil.
The boundary condition on the surface of primary interest is the flow tangency condition,
where the velocity normal to the surface is specified. In most cases the velocity normal to the
surface is zero.
Consider ways to handle the farfield BC
There are several possibilities:
A.
go out far enough (?) and set = 0 for 0, as the distance from the body goes to
infinity (or v = 0, u = 0 where these are the perturbation velocities, or u = U if it is the
total velocity).
How good is this? This method is frequently used, although clearly it requires
numerical experimentation to ensure that the boundary is far enough from the body.
In lifting cases this can be on the order of 50 chord lengths in two-dimensions. In
3/17/98

addition, this approach leads to excessive use of grid points in regions where we
normally arent interested in the details of the solution.
B.
Transform the equation to another coordinate system, and satisfy the boundary
condition explicitly at infinity (details of this approach are given in Chap. 9).
Figure 8-13 demonstrates what we mean. In the system the physical distance from 0
to infinity is transformed to the range from 0 to 1. Although this approach may lead to
efficient use of grid points, the use of the resulting highly stretched grid in the physical
plane may result in numerical methods that lose accuracy, and even worse, do not
converge during an iterative solution.
points evenly spaced in plane
x = is = 1
1.20
1.00
0.80
0.60
0.40
points spaced progressively

further apart in physical plane
0.20
0.00
0.00
5.00
10.00
15.00
20.00
25.00
Figure 8-13. One example of a way to handle the farfield boundary condition.
C.
Blocks of Grids are sometimes used, a dense inner grid and a coarse outer grid. In
this approach the grid points are used efficiently in the region of interest. It is a simple
version of the adaptive grid concept, where the the grid will adjust automatically to
concentrate points in regions of large flow gradients.
D.
Match the numerical solution to an analytic approximation for the farfield boundary
condition.
This is emerging as the standard way to handle the farfield boundary conditions. It
allows the outer boundary to be placed at a reasonable distance from the body, and
properly done, it ensures that the boundary numerical solution reflects the correct
physics at the boundary. This has been found to be particularly important in the solution
3/17/98
of the Euler equations. Effort is still underway to determine the best way to implement
this approach.
To summarize this discussion on farfield boundary conditions:
BCs on the FF boundary are important, and can be especially important for Euler
codes which march in time to a steady state final solution.
How to best enforce the FF BC is still under study - research papers are still being
written describing new approaches.
Consider ways to handle the nearfield BC
There are are also several ways to approach the satisfaction of boundary conditions on the
surface. Here we discuss three.
A. Use a standard grid and allow the surface to intersect grid lines in an irregular manner.
Then, solve the equations with BCs enforced between node points. Figure 8-14
illustrates this approach. In the early days of CFD methodology development this
approach was not found to work well, and the approach discussed next was developed.
However, using the finite volume method, an approach to treat boundary conditions
imposed in this manner was successfully developed (primarily by NASA Langley and its
contractors and grantees). It has not become a popular approach, and is considered to lead
to an inefficient use of grid points. Many grid points end up inside the body.
Airfoil placed in simple

rectangular grid
Irregular Intersection of
airfoil surface and grid
Figure 8-14. Surface passing through a general grid.

B. The most popular approach to enforcing surface boundary conditions is to use a
coordinate system constructed such that the surface of the body is a coordinate surface.
An example of this approach is shown in Figure 8-15. This is currently the method of
choice and by far the most popular approach employed in CFD. It works well. However,
it complicates the problem formulation. To use this approach, grid generation became an
area of study by itself. Grid generation is discussed in Chapter 9.
3/17/98
a. Entire geometry
b. Closeup of the trailing edge
Figure 8-15. Body conforming grid for easy application of BCs on curved surfaces.10
C. Another approach is to use thin airfoil theory boundary conditions, as described in detail
in Chapter 6. This eliminates many of the problems associated with the first two
approaches. It is expedient, but at some loss in accuracy (but very likely not that much, as
shown in Chap. 6, Fig. 6-14).
Apply boundary conditions

approximately on a grid line
Figure 8-16. Approximate approach to boundary condition specification.

Finite difference representation of the BC's
After defining a coordinate system, the finite difference representation of the boundary
condition must be written down. Using Laplaces Equation as an example, consider that there are
normally two types of boundary conditions associated with the boundary: 1) the Dirichlet
problem, where is specified on the boundary, and 2) the Neumann problem, where /n is
specified. If the Dirichlet problem is being solved, the value on the boundary is simply specified
and no special difference formulas are required.
3/17/98
When the solution requires that the gradient normal to the surface be specified, a so-called
dummy row is the easiest way to implement the boundary condition. As an example, following
Moran,4 consider a case where the normal velocity, v, is set to zero at the outer boundary. The
boundary is at grid line j = NY. Assume that another row is added at j = NY + 1, as indicated in
Fig. 8-17.
j = NY + 1
j = NY
j = NY - 1
i
Figure 8-17. Boundary condition at farfield.
The required boundary condition at j = NY is:
= 0 = i,NY +1 i,NY1 + O(Y )2

n
YNY +1 YNY1
(8-57)
and to ensure that the boundary condition is satisfied, simply define:

i,NY +1 i,NY1 .
(8-58)
The equations are then solved up to Y NY, and whenever you need at NY +1, simply use the
value at NY-1.
Now, we present an example demonstrating the application of thin airfoil theory boundary
conditions at the surface. Recall that the boundary condition is:
v=
df
= U
y
dx
(8-59)
where yfoil = f(x). Assuming that in the computer code v has been nondimensionalized by U ,
the boundary condition is:
df
=
y dx
and the grid near the surface is defined following Fig. 8-18.
3/17/98
(8-60)

j=3
j = 2 (surface)
j=1
i
Figure 8-18. Boundary condition at surface.
Writing the derivative in terms of central differences at j =2,
i,3 i,1 df
=
Y3 Y1
dx
(8-61)
we solve for i,1:

i,1 = i,3 (Y3 Y1)
df
.
dx
(8-62)
Note that since j = 1 is a dummy row, you can select the grid spacing such that the spacing is
equal on both sides of j = 2, resulting in second order accuracy. Thus, as in the previous example,
anytime we need i,1 we use the value given by Eq. (8-62). Using these boundary condition
relations, the boundary conditions are identically satisfied. Note also that this approach is the
reason that in many codes the body surface corresponds to the second grid line, j = 2.
Imposition of boundary conditions is sometimes more difficult than the analysis given here
suggests. Specifically, both the surface and farfield boundary conditions for the pressure in the
Navier-Stokes and Euler equations can be tricky.
8.5 Solution of Algebraic Equations
We now know how to write down a representation of the PDE at each grid point. The next
step is to solve the resulting system of equations. Recall that we have one algebraic equation for
each grid point. The system of algebraic equations may, or may not, be linear. If they are
nonlinear, the usual approach is to form an approximate linear system, and then solve the system
iteratively to obtain the solution of the original nonlinear system. The accuracy requirement
dictates the number of the grid points required to obtain the solution. Previously, we assumed
that linear equation solution subroutines were available, as discussed in Chapter 3. However, the
development of CFD methods requires a knowledge of the types of algebraic systems of
equations.
3/17/98
Recall that linear algebraic equations can be written in the standard form:
Ax = b .
(8-63)
For an inviscid two-dimensional solution, a grid of 100 x 30 is typical. This is 3000 grid points,
and results in a matrix 3000 x 3000. In three dimensions, 250,000 300,000 grid points are
common, 500,000 points are not uncommon, and a million or more grid points are often
required. Clearly, you cant expect to use classical direct linear equation solvers for systems of
this size.
Standard classification of algebraic equations depends on the characteristics of the elements
in the matrix A. If A:
1. contains few or no zero coefficients, it is called dense,
2. contains many zero coefficients, it is called sparse,
3. contains many zero coefficients, and the non-zero coefficients are close to
the main diagonal: the A matrix is called sparse and banded.
Dense Matrix
For a dense matrix direct methods are appropriate. Gauss elimination is an example of the
standard approach to these systems. LU decomposition11 is used in program PANEL, and is an
example of a standard method for solution of a dense matrix. These methods are not good for
large matrices (> 200-400 equations). The run time becomes huge, and the results may be
susceptible to round-off error.
Sparse and Banded
Special forms of Gauss elimination are available in many cases. The most famous banded
matrix solution applies to so-called tridiagonal systems:
b1
a
2
c1
b2
O
c2
O
ai
0
O
bi
O
ci
O
a N1
O
bN1
aN
x1 d1
x2 d2
M M
xi = di
M M
c N1 x N1 dN 1
bN x N dN
(8-64)
The algorithm used to solve Eq. (8-64) is known as the Thomas algorithm. This algorithm is very
good and widely used. The Thomas algorithm is given in detail in the sidebar, and a sample
subroutine, tridag, is described in App H-1.
3/17/98

Solution of tridiagonal systems of equations
The Thomas Algorithm is a special form of Gauss elimination that can be used to solve
tridiagonal systems of equations. When the matrix is tridiagonal, the solution can be
obtained in O(n) operations, instead of O(n3/3). The form of the equation is:
ai xi1 + bi xi + ci xi+1 = di
i = 1,K,n
where a1 and cn are zero. The solution algorithm12 starts with k = 2,....,n:
m=
ak
bk 1
bk = bk mck 1 .
dk = dk mdk1
Then:
d
xn = n
bn
and finally, for k = n - 1,...1:
d c x
xk = k k k+1 .
bk
In CFD methods this algorithm is usually coded directly into the solution procedure,
unless machine optimized subroutines are employed on a specific computer.
General Sparse
These matrices are best treated with iterative methods. In this approach an initial estimate of
the solution is specified (often simply 0), and the solution is then obtained by repeatedly
updating the values of the solution vector until the equations are solved. This is also a natural
method for solving nonlinear algebraic equations, where the equations are written in the linear
equation form, and the coefficients of the A matrix are changed as the solution develops during
the iteration. Many methods are available.
There is one basic requirement for iterative solutions to converge. The elements on the
diagonal of the matrix should be large relative to the values off the diagonal. The condition can
be give mathematically as:
aii
aij
j=1
ji
and for at least one row:
3/17/98
(8-65)
aii >
aij
(8-66)
j=1
ji
A matrix that satisfies this condition is diagonally dominant, and, for an iterative method to
converge, the matrix must be diagonally dominant. One example from aerodynamics of a matrix
that arises which is not diagonally dominant is the matrix obtained in the monoplane equation
formulation for the solution of the lifting line theory problem.
One class of iterative solution methods widely used in CFD is relaxation. As an example,
consider Laplaces Equation. Start with the discretized form, Eq.(8-31). The iteration proceeds
by solving the equation at each grid point i,j at an iteration n+1 using values found at iteration n.
Thus the solution at iteration n+1 is found from:
n+1
i,j
=
1 n
n
i+1, j + in1, j + i,n j+1 + i,j
1 .
4
(8-67)
The values of are computed repeatedly until they are no longer changing. The relaxation of
the values of to final converged values is roughly analogous to determining the solution for an
unsteady flow approaching a final steady state value, where the iteration cycle is identified as a
time-like step. This is an important analogy. Finally, the idea of iterating until the values stop
changing as an indication of convergence is not good enough. Instead, we must check to see if
the finite difference representation of the partial differential equation using the current values of
actually satisfies the partial differential equation. In this case, the value of the equation should
be zero, and the actual value of the finite difference representation is know as the residual. When
the residual is zero, the solution has converged. This is the value that should be monitored during
the iterative process. Generally, as done in THINFOIL, the maximum residual and its location
in the grid, and the average residual are computed and saved during the iterative process to
examine the convergence history.
Note that this method uses all old values of to get the new value of . This approach is
known as point Jacoby iteration. You need to save all the old values of the array as well as the
new values. This procedure converges only very slowly to the final converged solution.
A more natural approach to obtaining the solution is to use new estimates of the solution as
soon as they are available. Figure 8-19 shows how this is done using a simply programmed
systematic sweep of the grid. With a conventional sweep of the grid this becomes:
n+1
i,j
=
3/17/98
1 n
n
n+1
i+1, j + in+1
1, j + i,j+1 + i, j1 .
4
(8-68)

This scheme is called the point Gauss-Seidel iteration. It also elliminates the need to store all
the old iteration values as well as all the new iteration results, which was required with the point
Jacoby method.
new values
available
old values available
X
X
value to be found here
sweep up each line, and

then move to the next,
starting at the bottom
Figure 8-19. Grid sweep approach to implement the Gauss-Seidel solution iteration scheme.
The point Gauss-Seidel iteration procedure also converges slowly. One method of speeding
up the convergence is to make the change to the value larger than the change indicated by the
normal Gauss-Seidel iteration. Since the methods that have been described are known as
relaxation methods, the idea of increasing the change is known as successive over-relaxation, or
SOR. This is implemented by defining an intermediate value:
n+1 1 n
n
n+1
i,j
= i+1, j + in+1
1, j + i,j+1 + i, j1
4
(8-69)
and then obtaining the new value as:

n+1
n
i,j
= i,j
+
(i,n+1j i,n j ) .
(8-70)
The parameter is a relaxation parameter. If it is unity, the basic Gauss-Seidel method is

recovered. How large can we make it? For most model problems, a stability analysis (presented
in the next section) indicates that < 2 is required to obtain a converging iteration. The best
value of depends on the grid and the actual equation. For most cases of practical interest the
best values of must be determined through numerical experimentation. Figure 8-20 presents an
example of the manner in which the solution evolves with iterations. The value of after 2000
iterations is approached very gradually. The figure also illustrates the time-like nature of the
iteration.
3/17/98
-0.02
value at 2000 iterations
-0.01
(16,2)
-0.01
solution of Laplace's equation

biconvex airfoil
=1.5
surface value at .2c
-0.00
0.00
100
200
300
iteration number
400
500
Figure 8-20. Typical variation of during solution iteration.

Another way to speed up the iteration is to sweep the flowfield a line at a time rather than a
point at a time. Applying over-relaxation to this process, the so-called successive line overrelaxtion, or SLOR, process is obtained. In this method a system of equations must be solved at
each line. Figure 8-21 illustrates this approach. The method is formulated so that the system of
equations is tridiagonal, and the solution is obtained very efficiently. This approach provides a
means of spreading the information from new values more quickly than the point by point sweep
of the flowfield. However, all of these approaches result in a very slow approach to the final
value during the iterations.
The effect of the value of the over relaxation parameter is shown in Figure 8-22. Here, the
convergence level is compared for various values of . Notice that as convergence requirements
are increased, the choice of becomes much more important. Unfortunately, the choice of
may not only be dependent on the particular numerical method, but also on the particular
problem being solved.
Mathematically, the convergence rate of an iterative process depends on the value of the socalled spectral radius of the matrix relating the value of the unknowns at one iteration to the
values of the unknowns at the previous iteration. The spectral radius is the absolute value of the
largest eigenvalue of the matrix. The spectral radius must be less than one for the iterative
process to converge. The smaller the value of the spectral radius, the faster the convergence.
3/17/98

values to be
found here
new values available
old values available
X
X
X
X
X
X
X
X
X
starting upstream, move downstream, solving a line at a time.
Figure 8-21. Solution approach for SLOR.
200
Convergence tolerance
= 10-5
150
Number
of
Iterations
100
= 10-4
= 10-3
50
ONERA M6 Wing
M = 0.84
= 3
0
1.60
1.65
1.70
1.75
1.80
1.85
1.90
1.95
2.00
Figure 8-22. Effect of the value of on the number of iterations required to achieve
various levels of convergence.13
Another way to spread the information rapidly is to alternately sweep in both the x and y
direction. This provides a means of obtaining the final answers even more quickly, and is known
3/17/98
as an alternating direction implicit or ADI method. Figure 8-23 illustrates the modification to the
SLOR method that is used to implement an ADI scheme. Several different methods of carrying
out the details of this iteration are available. The traditional approach for linear equations is
known as the Peaceman-Rachford method, and is described in standard textbooks, e.g., Ames8
and Isaacson and Keller.14 This approach is also known as an approximate factorization or AF
scheme. It is known as AF1 because of the particular approach to the factorization of the
operator. A discussion of ADI including a computer program is given in the first edition of the
Numerical Recipes book.15
Another approach has been found to be more robust for nonlinear partial differential
equations, including the case of mixed sub- and supersonic flow. In this case the time-like nature
of the approach to a final value is used explicitly to develop a robust and rapidly converging
iteration scheme. This scheme is known as AF2. This method was first proposed for steady flows
by Ballhaus, et al,16 and Catherall17 provided a theoretical foundation and results from numerical
experiments. A key aspect of ADI or any AF scheme is the use of a sequence of relaxation
paramters rathers a single value, as employed in the SOR and SLOR methods. Typically, the
sequence repeats each eight to eleven iterations.
Holst10 has given an excellent review and comparison of these methods. Figure 8-24, from
Holst,10 shows how the different methods use progressively better information at a point to find
the solution with the fewest possible iterations. The advantage is shown graphically in Figure 825, and is tabulated in Table 8-1 (also from Holst10). Program THINFOIL, described in Section
8.7, uses these methods and App. G-1 contains a description of the theoretical implementation of
these methods. Further details are given in Chapter 11, Transonic Aerodynamics.
first sweep along vertical lines

from upstream to downstream
next, sweep across the

grid in the other direction
Figure 8-23. ADI Scheme solution approach.
3/17/98

Current value to be found, and new information used
Old information used
Point Jacoby
Point Gauss-Seidel
Line Gauss-Seidel
SLOR
Line Jacoby
SOR
ADI
Figure 8-24. Stencil of information (Holst10)

In addition to these methods, solutions can be obtained more rapidly by using so-called
multigrid methods. These methods accelerate the convergence iterative procedures by using a
sequence of grids of different densities and have become one of the most important techniques
used to solve field problems of all types. The overall levels of the solution are established by the
solution on a crude grid, while the details of the solution are established on a series of finer grids.
Typically, one iteration is made on each successively finer grid, until the finest grid is reached.
Then, one iteration is made on each successively courser grid. This process is repeated until the
solution converges. This procedure can reduce the number of fine grid iterations from possibly
thousands, as shown above, to from 10 to 30 iterations.
This approach to the solution of partial differential equations was highly developed by
Jameson18 for the solution of computational aerodynamics problems. He used the multigrid
approach together with an alternating direction method in an extremely efficient algorithm for
the two-dimensional transonic flow over an airfoil.
The details of the multigrid method are, as they say, beyond the scope of this chapter, and the
reader should consult the standard literature for more details. This includes the original treatise
on the subject by Brandt19 (which includes an example FORTRAN program), another tutorial
which includes a FORTRAN code,20 and more recent presentations by Briggs21 and Wesseling.22
The most recent Numerical Recipes11 book also includes a brief description and sample program.
3/17/98
100000
Point Gauss Seidel
& Line Jacobi
10000
Line
Gauss Seidel
1000
Point
Jacoby
n
SOR
100
SLOR
10
ADI
1
0.001
0.010
0.100
1.000
Figure 8-25. Comparison of convergence rates of various relaxation schemes (Holst10). This is
the number of iterations estimated to be required to reduce the residual by one order of
magnitude
Table 8-1
Convergence rate estimates for various relaxation schems (Holst 10)
Algorithm
3/17/98
Number of iterations required

for a one order-of-magnitude
reduction in error
2
Point - Jacobi
2/
Point - Gauss -Seidel

SOR
1 / 2
1 / (2 )
Line - Jacobi
1 / 2
Line - Gauss- Seidel
1 / (2 2 )
SLOR
ADI
1 / (2 2 )
log( / 2) / 1.55

To carry out the solution to large systems of equations, the standard numerical procedures
require that the approach be generalized slightly from the one given above. Specificaly, we
define an operator, such that the partial differential equation is written (continuing to use
Laplaces equation as an example):
L = 0
(8-71)
2
2
+
.
x2 y2
(8-72)
where
L=
To solve this equation, we re-write the iteration scheme expressions given above in Equation.
(8-70) as:
N
Ci,n j
{
nth iteration
correction
L i,nj
1
23
= 0.
(8-63)
nth iteration
residual, = 0
when converged
solution is achieved
This form is known as the standard or delta form. C is given by

n
Ci,nj = 1,n +1
j 1, j .
(8-64)
The actual form of the N operator depends on the specific scheme chosen to solve the problem.
8.6 Stability Analysis
The analysis presented above makes this approach to solving the governing equations for
flowfields appear deceptively simple. In many cases it proved impossible to obtain solutions.
Frequently the reason was the choice of an inherently unstable numerical algorithm. In this
section we present one of the classical approaches to the determination of stability criteria for use
in CFD. These types of analysis provide insight into grid and stepsize requirements (the term
stepsize tends to denote time steps, whereas a grid size is thought of as a spatial size). In
addition, this analysis is directly applicable to a linear equation. Applications in nonlinear
problems are not as fully developed.
Fourier or Von Neumann Stability Analysis
Consider the heat equation used previously,
u
2u
= 2
t
x
(8-17)
and examine the stability of the explicit representation of this equation given by Eq. (8-21).
Assume at t = 0, that an error, possibly due to finite length arithmetic, is introduced in the form:
3/17/98
u(x,t)
123 = (t)
"error" is
introduced
jx
e{
(8-75)
actually could be a series,

take one term here
where:
a real constant
j = 1
Here we restate the explicit finite difference representation,

u(x,t + t) u(x,t)
u( x + x,t ) 2u (x,t ) + u( x x,t)
=
.
t
(x )2
(8-21)
Substitute Eq. (8-75) into this equation, and solve for (t + t). Start with
(t + t)e jx (t)e jx
(t) j (x +x )
=
2e jx + e j (x x )
2 e
t
(x )
(8-76)
and collecting terms:

(t + t)e jx = (t)e jx +
t
jx jx
e
2 + e jx
2 (t)e
(x)
14 4424 443
(8-77)
x +e jx
2+e1j
442
443
2cos x
Note that the e jx term cancels, and Eq. (8-77) can be rewritten:
t
(t + t) = (t)1 +
2 (2 + 2 cosx )
(x )
= (t) 1 2
1
cosx
424
3
(x)2 double1
angle formula
=12sin2 x
(8-78)
which reduces to:
t
2 x
(t + t) = (t)1 + 2
2 1 1+ 2sin 2
(x )
t
2 x
= (t)1 4
sin
2
( x )2
Now look at the ratio of (t + t) to (t) , which is defined as an amplification factor G,
3/17/98
(8-79)
G=
(t + t)
t
2 x
= 1 4
sin
.
(t)
2
( x )2
(8-80)
For stability the requirement is clearly:

G <1 ,
(8-81)
which means that the error introduced decays. For arbitrary , what does this condition mean?
Observe that the maximum value of the sine term is one. Thus, the condition for stability will be:
t
1 4
2 <1
(
x
)
1
424
3
(8-82)
and the limit will be:

1 4 =1 .
(8-83)
The largest that can satisfy this requirement is:

1 4 = 1
or
4 = 2 .
(8-84)
and
1
2
t
1
2 <2
( x )
Thus, the largest for |G| < 1 means
or:
t
1
2 <2.
( x )
(8-85)
(8-86)
This sets the condition on t and x for stability of the model equation. This is a real
restriction. It can be applied locally for nonlinear equations by assuming constant coefficients.
An analysis of the implicit formulation, Eq. (8-23), demonstrates that the implicit formulation is
unconditionally stable.
Is this restriction on t and x real? Rightmyer and Morton23 provided a dramatic example
demonstrating this criteria. Numerical experiments can quickly demonstrate how important this
condition is. Figure 8-26 repeats the analysis of Rightmyer and Morton,23 demonstrating the
validity of the analysis. The initial and boundary conditions used are:
3/17/98
u(x,0 = (x) (given) for0 x
u(0,t) = 0, u(,t) = 0 for t > 0
Figure 8-26a presents the development of the solution and shows the particular choice of
initial value shape, , using a value of < 1/2: 5/11. Figure 8-26b-d provide the results for a
value of > 1/2: 5/9. Theoretically, this stepsize will lead to an unstable numerical method, and
the figure demonstrates that this is, in fact, the case.
Our model problem was parabolic. Another famous example considers a hyperbolic equation.
This is the wave equation, where c is the wave speed:
2u
2u
c
= 0.
(8-87)
t 2
x 2
This equation represents one-dimensional acoustic disturbances. The two-dimensional small
disturbance equation for the potential flow can also be written in this form for supersonic flow.
Recall,
(1 M2 ) xx + yy = 0
(8-88)
(8-89)
or when the flow is supersonic:

xx
2
M
1
yy = 0
and we see here that x is the timelike variable for supersonic flow.
Performing an analysis similar to the one above, the stability requirement for Eq. (8-87) is
found to result in a specific parameter for stability:
=c
t
x
(8-90)
which is known as the Courant number. For many explicit schemes for hyperbolic equations, the
stability requirement is found to be
1.
(8-91)
This requirement is known as the CFL condition, after its discoverers: Courant, Friedrichs,
and Levy. It has a physical interpretation. The analytic domain of influence must lie within the
numerical domain of influence.
Recalling that the evolution of the solution for an elliptic system had a definite time-like
quality, a stability analysis for elliptic problems can also be carried out. For the SOR method,
that analysis leads to the requirement that the over-relaxation factor, , be less than two.
3/17/98
1.40
relative timestep
0
5
10
15
20
40
1.20
1.00
0.80
u(x,t)
0.60
initial condition
0.40
0.20
0.00
-0.5
0.0
1.5
2.0
2.5
3.0
x
a) numerical solution using a theoretically stable stepsize.
1.40
0.5
1.0
numerical solution
relative timestep
1.20
3.5
initial condition
5
1.00
stable solution
0.80
u(x,t)
0.60
0.40
0.20
0.00
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
x
t
b) numerical solution using a theoreticaly unstable stepsize,
= 5.
(x)2
3.5
Figure 8-26. Demonstration of the step size stability criteria on numerical solutions.
3/17/98
1.40
1.20
numerical solution
relative timestep
1.00
10
0.80
stable solution
u(x,t)
initial condition
0.60
0.40
0.20
0.00
-0.5
1.5
2.0
2.5
3.0
x
t
c) numerical solution using a theoreticaly unstable stepsize,
= 10.
(x)2
1.80
0.0
0.5
1.0
3.5
numerical solution
relative timestep
15
initial condition
1.20
u(x,t)
stable solution
0.60
0.00
-0.60
-0.5
0.0
0.5
1.0
1.5
x
2.0
2.5
d) numerical solution using a theoreticaly unstable stepsize,
3.0
3.5
t
= 15.
(x)2
Figure 8-26. Demonstration of the step size stability criteria on numerical solutions (concluded).
3/17/98

8.7 Program THINFOIL
An example of the solution of Laplaces Equation by finite differences is demonstrated in the
program THINFOIL. This program offers the users options of SOR, SLOR, AF1 and AF2 to
solve the system of algebraic equations for the flow over a biconvex airfoil at zero angle of
attack. An unevenly spaced grid is used to concentrate grid points near the airfoil. The program
and the theory are described in Appendix G-1. It can be used to study the effects of grid
boundary location, number of grid points, and relaxation factor, .
Figure 8-27 provides the convergence history for the case for which the comparison with the
exact solution is given below. Using SOR, this shows that hundreds of iterations are required to
reduce the maximum change between iteration approximately three orders of magnitude. This is
about the minimum level of convergence required for useful results. A check against results
converged further should be made. The reader should compare this with the other iteration
options.
103
5% Thick Biconvex Airfoil
74 x 24 grid
SOR =1.80
AF 2 factor: 1.333
10
101
Maximum
Residual
SOR
100
10-1
AF2
-2
10
10-3
10-4
100
200
300
Iteration
400
500
600
Figure 8-27. Convergence history during relaxation solution.

The convergence history presented above is actually the maximum residual of at each
iteration. The solution is assumed to have converged when the residual goes to zero. Typical
3/17/98
engineering practice is to consider the solution converged when the residual is reduced by 3 or 4
orders of magnitude. However, a check of the solution obtained at a conventional convergence
level with a solution obtained at much smaller residual (and higher cost) level should be made
before conducting an extensive analysis for a particular study.
The solution for a 5% thick biconvex airfoil obtained with THINFOIL is presented in
Figure 8-28, together with the exact solution. For this case the agreement with the exact solution
is excellent. The exact solution for a biconvex airfoil is given by Milton Van Dyke,24 who cites
Milne-Thompson25 for the derivation.
-0.30
5% Thick Biconvex Airfoil
(THINFOIL uses thin airfoil theory BC's)
-0.20
-0.10
Cp
0.00
0.10
Cp (THINFOIL)
Cp (Exact)
0.20
74 x 24 grid
0.30
0.0
0.2
0.4
0.6
0.8
1.0
X/C
Figure 8-28. Comparison of numerical solution with analytic solution for a biconvex airfoil.
The material covered in this chapter provides a very brief introduction to an area which has been
the subject of an incredible amount of research in the last thirty years. Extensions to include
ways to treat flows governed by nonlinear partial differential equations are described after some
basic problems in establishing geometry and grids are covered in the next chapter.
3/17/98

8.8 Exercises
1. How accurate are finite difference approximations? Over one cycle of a sine wave,
compare first and second order accurate finite difference approximations of the 1st
derivative and the second order accurate 2nd derivative of the shape with the exact values.
How small does the step size have to be for the numerical results to accurate to 2
significant figure? 4 figures? 6? What conclusions about step size can you make?
2. Get some experience with the solution of Laplaces Equation using finite differences.
i) Download a copy of THINFOIL from the web page
ii) Make it run on your PC.
iii) Study the program to understand the procedure.
Pick as a baseline case: Xmin=-2.2, Xmax=3.2, Ymax=2.4, and NUP=14, NDOWN=14,
NON=30, NABOVE=18
iv) Run SOR with = 1.6 and see how many iterations to convergence
v) Run with = 1.0, 1.50, 1.75, 1.90, 1.99 (400 iterations max)
vi) Plot the convergence history as a function of iteration for each . Note that it is
standard procedure to plot the log of the residual. See examples in the text.
vii) For one , increase the number of grid points and compare (watch dimensions)
- convergence rate with the same case above
- the surface pressure distribution results for the two grids
viii) Draw conclusions about SOR as a numerical method for solving PDE's.
ix) Repeat the studies using SLOR, AF1 and AF2. What do you conclude about the
relative convergence times and solution accuracy?
3. Examine the effect of the number of grid points on the solution obtained using program
THINFOIL. How many grid points are required for a grid converged solution?
4. Examine the effect of the location of the farfield boundary condition on the solution
obtained using program THINFOIL. What do you conclude?
5. Change the farfield boundary condition to set = 0, instead of /n = 0. How does this
affect the solution? the convergence rate?
6. Modify program THINFOIL to obtain the solution to the flow over an NACA 4-Digit
airfoil thickness shape. Address the following issues:
i) store the boundary condition values before the calculation begins instead of
recomputing each time the BC needs the value
ii) recognizing that the slope at the leading edge is infinite, assess two methods of
avoiding numerical problems
place the leading edge between grid points
use Riegels factor to modify the slope boundary condition, replacing df/dx by
df /d x
1 + ( df /d x)2
3/17/98
8.9 References
1
Tannehill, J.C., Anderson, D.A., and Pletcher, R.H., Computational Fluid Mechanics and Heat
Transfer, 2nd Ed., Taylor & Francis, New York, 1997. (on the first edition the order of the
authors was Anderson, D.A., Tannehill, J.C., and Pletcher, R.H)
2
Hoffman, K.A., and Chiang, S.T., Computational Fluid Dynamics for Engineers, in two
volumes, Engineering Education System, PO Box 20078, Wichita, KS, 67208-1078. 1993.
3
Fletcher, C.A.J., Computational Techniques for Fluid Dynamics, Vol. 1: Fundamental and
General Techniques, Vol. II, Specific Techniques for Different Flow Categories, SpringerVerlag, Berlin, 1988.
4
Moran, J., An Introduction to Theoretical and Computational Aerodynamics, John Wiley &
Sons, New York, 1984.
5
Hirsch, C., Numerical Computation of Internal and External Flows, Vol. 1, Fundamentals of
Numerical Discretization, 1988, and Vol. 2, Computational Methods for Inviscid and Viscous
Flows, 1990, John Wiley & Sons, New York.
6
Anderson, J.D., Jr., Computational Fluid Dynamics: The Basics with Applications, McGrawHill, Inc., New York, 1995.
7
Smith, G.D., Numerical Solution of Partial Differential Equations: Finite Difference Methods,
3rd Ed., Clarendon Press, Oxford, 1985.
8
Ames, W.F., Numerical Methods for Partial Differential Equations, 2nd Ed., Academic Press,
New York, 1977.
9
Versteeg, H.K., and Malalasekera, W., An Introduction to Computational Fluid Dynamics: The
Finite Volume Method, Addison Wesley Longman, Ltd., Harlow, England, 1995.
10
Holst, T.L., Numerical Computation of Transonic Flow Governed by the Full-Potential
Equation, VKI Lecture Series on Computational Fluid Dynamics, Rhode-St.-Genese, March,
1983.
11
Press, W.H., Flannery, B.P., Teukolsky, S.A., and Vettering, W.T., Numerical Recipes in
FORTRAN: The Art of Scientific Computing, 2nd ed., Cambridge University Press, Cambridge,
1992.
12
Conte, S.D., and deBoor, C., Elementary Numerical Analysis, McGraw-Hill, New York, 1972.
13
Mason, W.H., Mackenzie, D., Stern, M., Ballhaus, W.F., and Frick, J., An Automated
Procedure for Computing the Three-Dimensional Transonic Flow Over Wing-Body
Combinations, Including Viscous Effects, Vol. II Program Users Manual and Code
Description, AFFDL-TR-77-122, Feb. 1978.
14
Isaacson, E., and Keller, H.B., Analysis of Numerical Methods, John Wiley, New York, 1966.
15
Press, W.H., Flannery, B.P., Teukolsky, S.A., and Vettering, W.T., Numerical Recipes: The
Art of Scientific Computing (FORTRAN Version), Cambridge University Press, Cambridge,
1989.
16
Ballhaus, W.F., Jameson, A., and Albert, J., Implicit Approximate Factorisation Schemes for
the Efficient Solution of Steady Transonic Flow Problems, AIAA J., Vol. 16, No. 6, June 1978,
pp. 573-579. (also AIAA Paper 77-634, 1977).
17
Catherall, D., Optimum Approximate-Factorisation Schemes for 2D Steady Potential Flows,
AIAA Paper 81-1018, 1981.
18
Jameson, A., Acceleration of Transonic Potential Flow Calculations on Arbitrary Meshes by
the Multiple Grid Method, AIAA Paper 79-1458, Proceedings of the AIAA 4th Computational
3/17/98
Fluid Dynamics Conference, AIAA, New York, 1979, pp. 122-146.

19
Brandt, A., Multi-Level Adaptive Solutions to Boundary-Value Problems, Mathematics of
Computation, Vol. 31, No. 138, April 1977, pp. 333-390.
20
Stuben, K., and Trottenberg, U., Multigrid Methods: Fundamental Algorithms, Model
Problem Analysis and Applications, in Multgrid Methods, ed. by W. Hackbusch and U.
Trottenberg, Lecture Notes in Mathematics Vol. 960, Springer-Verlag, Berlin, 1982. pp. 1-176.
21
Briggs, W.L., A Multigrid Tutorial, SIAM, Philadelphia, 1987
22
Wesseling, P., An Introduction to Multigrid Methods, John Wiley, Chichester, 1992.
23
Richtmyer, R.D., and Morton, K.W., Difference Methods for Initial-Value Problems, 2nd Ed.,
Interscience, New York, 1967 pp. 4-9.
24
Van Dyke, M., Perturbation Methods in Fluid Mechanics, Annotated Edition, The Parabolic
Press, Stanford, 1975, problem 4.5, page 74.
25
Milne-Thompson, Theoretical Hydrodynamics, 5th ed., (1968), reprinted by Dover
Publications, Inc., New York, 1996, Section 6.51, pp. 177-179.
3/17/98

Applied Computational Aerodynamics

Uploaded by

Copyright:

Available Formats

Applied Computational Aerodynamics

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Applied Computational Aerodynamics

Uploaded by

Copyright:

Available Formats

APPLIED

W. H. Mason Professor of Aerospace & Ocean

4. Incompressible Potential Flow

elliptic PDE, each point influences every other point,

4-2 Applied Computational Aerodynamics

Panel Methods 4-3

Figure 4-1. Boundaries for flowfield analysis.

on + {Dirichlet Problem: Design}

/n on + {Neuman Problem: Analysis}.

4-4 Applied Computational Aerodynamics

div(grad grad )dV = (grad grad ) n dS. .

Panel Methods 4-5

( )dV = (grad grad ) n dS ,

or equivalently (recalling that grad n = / n ),

Either statement is known as Greens theorem of the second form.

4-6 Applied Computational Aerodynamics

Figure 4-3. Spherical coordinate system nomenclature.

Integrating from = 0 to , and from 0 to 2, we get:

Panel Methods 4-7

4-8 Applied Computational Aerodynamics

Panel Methods 4-9

Figure 4-4. Panel model representation of an airplane.

4-10 Applied Computational Aerodynamics

this is a vortex singularity

Panel Methods 4-11

b) unit vector orientation

Figure 4-6. Nomenclature for local coordinate systems.

and the normal and tangential unit vectors are:

Figure 4-7. Local panel nomenclature.

4-12 Applied Computational Aerodynamics

(ui i + vi j) ( sini i + cosi j) = 0

for each i, i = 1, ..., N .

Figure 4-8. Trailing edge panel nomenclature.

Panel Methods 4-13

(u1i + v1j) ( cos1i + sin1 j) = (uN i + vN j) (cosN i + sin N j)

4-14 Applied Computational Aerodynamics

ui = V cos + q jusij + uvij

Figure 4-9. Local panel coordinate system and nomenclature.

Panel Methods 4-15

These integrals can be found (from tables) in closed form:

4-16 Applied Computational Aerodynamics

Figure 4-10. Relations between the point x*, y* and a panel.

Review a FORTRAN manual to understand how ATAN2 is used.

Panel Methods 4-17

4-18 Applied Computational Aerodynamics

ui = V cos + q jusij + uvij

V cos q jus uv sini + V sin + q j vs + vv cos i = 0

[ V sini cos + V cos i sin ] sin i q j usij

sin i uvij + cosi vvij = 0

V ( cosi sin sin i cos ) + cosi vsij sini usij q j

+ cosi vvij sini uvij = 0

Panel Methods 4-19

(cos i sin j sin i cos j ) = 0 sin (i j ).

Moving to the second term in parentheses above:

Aij = sin i j u*sij + cos i j v*sij

and using the definitions of

Now look at the expression for bi identified in (4-47):

4-20 Applied Computational Aerodynamics

cosi sin sini cos = sin (i )

Finally, work with the A i,N+1 term:

Figure 4-10. Relations between the point x, y and a panel.

Aij = sin i j usij + cos i j vsij

{cos(i j )vvij sin(i j )uvij }

usNj = usNj cos j v sNj sin j

AN+1, j = us1j cos j vs1 j sin j cos1

+ us1 j sin j + vs1 j cos j sin1

usNj cos j vsNj

+ usNj sin j + vsNj cos j sinN

AN+1, j = cos j 1 us1j + cos j N usNj

sin j 1 vs1j sin j N vsNj

uti = cos( i )V + cos( j i )usij sin( j i )vsij q j

{cos( j i )uvij sin( j i )vvij }