Estimating Aerodynamic Properties

The aerodynamic properties of a given aircraft are critically dependent on the aerodynamic properties of
its various lifting surfaces. For example, recall that the slope of the pitch moment coefficient Cm , a term
of primary importance in determining longitudinal stability, takes the form

d
+ Cmp .
Cm = CLwb (h hnwb ) VH CLt 1
d
d
, all of which are determined by the
The value of this expression depends on CLwb , CLt , hnwb , and d
geometric properties of the wing and horizontal tail and parameters describing the flow field (such as the
Mach number and Reynolds number). In this lecture, we discuss various tools available for estimating
aerodynamic properties of lifting surfaces. These notes are adapted from [1].

fuselage
centerline

cr

y
c (y )
ct

b
2

Figure 1: Sketch of a half-wing with constant sweep and taper.

Geometric properties of lifting surfaces. Consider the half-wing shown in Figure 1. Note that the
wing has been extended to the centerline of the fuselage; the root chord cr is defined there. The mean
geometric chord of a wing is defined as
c

=
=

Z b
2 2
c(y)dy
b 0
2S
b2

or

c0 =

The wing aspect ratio is

S
.
b

AR =

b
b2
=
.
c0
S

The mean aerodynamic chord of a wing is defined as

c =

2
S

b
2

c(y)2 dy.

(Note that this definition is purely geometric. The mean aerodynamic chord length does not depend on
the wing aerodynamics.)
For the wing depicted in Figure 1, the local chord c(y) varies linearly from the root chord cr to the tip
chord ct , so we say that this wing has constant taper. We define the taper ratio

ct
.
cr

For a wing with constant taper, the mean geometric chord is

c =

1+
2

cr

and the mean aerodynamic chord is

2
c =
3

1 + + 2
1+

cr .

Note that, if = 1 (i.e., if the wing is untapered), then c0 = c = cr = ct .

The nth percent chord line of a wing is the curve connecting each point nc(y) measured aft of the local
leading edge (where 0 n 1). (For example, the 0th percent chord line is the leading edge of the wing.)
For the wing depicted, the angle of incidence of each chord line is constant, so we say that this wing has
constant sweep. For a wing with constant sweep, one defines the sweep angle n of the nth percent chord
line as shown for n = 0. Given the sweep angle m of the mth percent chord line for a wing with constant
sweep and taper, one may determine n according to the formula
4(n m)
tan n = tan m
AR

1
1+

The mean aerodynamic center (

x, y, z) is defined as that point about which the total aerodynamic moment
of the wing does not vary with angle of attack. The location of the mean aerodynamic chord (for the
half-wing) is determined relative to the mean aerodynamic center. While the mean aerodynamic chord
c is defined purely by the wing geometry, the location of the mean aerodynamic center depends on the
wing loading. Under certain assumptions about the form of the load distribution, the location of the mean
aerodynamic center may be computed explicitly. See Appendix C of [1] and, in particular, Figure C.3.
2

Figure 2: Sketch of the mean aerodynamic chord for a wing with constant sweep and taper.
Because the airplane is symmetric about the vertical plane through the centerline, the mean aerodynamic
center of the complete wing lies in this plane. For a wing with constant sweep and taper whose load
distribution is proportional to the local chord length, the distance from the leading edge of the root chord
to the leading edge of the mean aerodynamic chord is
b
m
=
6

1 + 2
1+

tan 0 .

Such a load distribution would result, for example, if the local lift coefficient of the airfoil sections constituting the wing does not vary with y. (See Appendix C of [1] for information on other load distributions
and planform shapes.)
Aerodynamic properties of airfoils (Sectional properties). The aerodynamic properties of lifting
surfaces can be estimated in terms of the aerodynamic properties of the 2-D airfoil sections which make
up the complete lifting surface. For a given lifting surface we generally need three 2-D properties
Cl
0L2D
Cm0L2D

= the 2-D lift-curve slope

= the 2-D zero-lift angle of attack, and
= the 2-D pitch moment at zero lift (i.e., the moment coefficient about the aerodynamic center).
0.09c

1
2 TE

t0.99

t0.90

Figure 3: Definition of the trailing edge angle TE .

We first consider the problem of estimating the 2D lift-curve slope Cl for a given section. The method
outlined in Section 1 of Appendix B in [1] applies to wings with constant sweep and taper and without
3

twist moving at Mach numbers less than the critical Mach number. (The critical Mach number is that
value of the free stream Mach number at which the local flow is sonic at some point on the aircraft; the
critical Mach number is always less than one.) The method proceeds as follows
1. Given the wing thickness ratio ct , estimate the theoretical 2-D lift-curve slope (Cl )theory from Figure
B.1,1(b). The curve appears to be well-approximated by the formula
t
(Cl )theory = 2 + 4.9 .
c
2. Measure or estimate the trailing edge angle TE . This angle is defined

TE = 2 arctan

1 t
2

c 90%

1
2

0.09

c 99%

where ct 90% is the wing thickness ratio at the 90% chord line and ct 99% is the wing thickness ratio
at the 99% chord line. See Figure 3.
3. Determine the factor K which corrects for trailing edge angle and Reynolds number from Figure
B.1,1(a). (Interpolate for Reynolds numbers between 106 and 108 .)
4. Correct for trailing edge angle, Reynolds number, and Mach number to obtain the 2-D lift-curve
slope

Cl =

1.05
K (Cl )theory .
1 M2

If detailed airfoil information is unavailable and the wing is thin, then an approximate value of the 2-D
lift-curve slope is

Cl =

2
.
1 M2

The parameters 0L2D and Cm0L2D must be determined by some other means, such as a wind tunnel test
or a computational fluid dynamics (CFD) model.
Aerodynamic properties of lifting surfaces. First, we determine the lift-curve slope for an untwisted
wing with constant sweep and taper moving at a subcritical Mach number. Let

1 M2
Cl .
2

Note that = 1 for a thin wing (i.e., a flat plate). Using the sectional lift-curve slope from above, and
correcting for the finite wing-span and the sweep angle, we obtain

CL =
2+

2AR

AR2 (1M 2 )
1+
2

.
tan2 1/2
(1M 2 )

+4

Next, we determine the zero-lift angle of attack 0L . We assume that the sectional zero-lift angle 0L2D is
given. Following are two cases for which 0L can be easily determined.
1. Constant sweep angle and constant airfoil section, where the cross sections are taken normal to the
nth percent chord line. (No twist.)

0L = arctan

tan 0L2D
cos n

Note that, in the special case of an unswept wing, 0L = 0L2D .

2. Geometric twist and zero sweep. (Note: A wing with geometric twist has constant airfoil sections
across the span, however the zero-lift line of the sections varies from root to tip. A wing with
aerodynamic twist achieves a similar effect by varying the airfoil section from root to tip.) Refer the
wing angle of attack to the root chord: w := root . Then
(y) = root + (y)
where (y) is the wing twist and (0) = 0. The wing zero-lift angle of attack is

0L

2
=
S

b/2
0

(0L2D (y)) c(y)dy.

We next compute the wing zero-lift pitch moment Cm0L , given Cm0L2D across the span. This term can
be easily estimated in the special case of an untwisted wing with constant sweep angle, where the airfoil
sections are taken parallel to the free stream:

Cm0L =

AR cos2 1/4
AR + 2 cos 1/4

Cm0L2D |root + Cm0L2D |tip

2

Note that, if the wing section does not vary across the span, the latter factor becomes simply Cm0L2D .
d
. The wing downwash parameter can be crudely estimated by assuming
Wing downwash parameter d
a thin, finite wing with an elliptic load distribution, which gives

d
2CL
=
.
d
AR
A more accurate method applicable to wings with constant sweep and taper is described in Appendix B.5
of [1]. The estimate takes the form
1.19

p
4.44 KA K KH cos 1/4
d

=
d
1 M2
5

where KA is a correction for the aspect ratio of the wing, K is a correction for the taper ratio of the
wing, and KH is a correction for the location of the horizontal tail. The correction factors are determined
according to the following formulas:

KA =
K =
KH =

1
1

AR 1 + AR1.7
10 3
7

1 hbH
q
2lH
b

where, in the last formula,

hH

= the (signed) orthogonal distance from the extended root chord line to the horizontal tail a.c.

lH

= the longitudinal distance from the wing a.c. to the tail a.c.

b = wingspan.

Tutorial Example
Fuselage
Reference
Line
acw

V
( - )
-i t
V
act
NACA 2412

NACA 0009

Figure 4: Sketch for the example.

Consider an aircraft with a trapezoidal wing and horizontal tail. Following are the geometric properties.
Wing properties: NACA 2412
b = 15 m,

AR = 6,

1/4 =

0L2D = 2.0 ,

rad,
6

1
= ,
4

cm0L2D = 0.047.

Tail properties: NACA 0009

bt = 6 m,

ARt = 4,

1/4t = 0 rad,
6

t = 1,

0L2Dt = 0.0 ,

cm0L2Dt = 0.0,

it = 1.0 ,

hH = 0 m,

lH = 15 m.

Compute the following:

1. The mean aerodynamic chord length and location for the wing and tail and the longitudinal location
of the wing and tail mean aerodynamic centers.
2. The lift coefficient of the wing as a function of fuselage angle of attack at Re = 106 and M = 0.5.
(This requires computing both the slope of the lift curve and the lift coefficient at = 0.)
3. The zero-lift pitch moment of the wing Cm0Lw = Cmacw .
4. The lift coefficient of the tail as a function of fuselage angle of attack at Re = 106 and M = 0.5.
d
(In addition to the lift slope, this requires computing the downwash parameter d
.)
Item #1.)

Wing m. a. chord length. The mean geometric chord is

c0 =

b
S
=
= 2.5 m.
b
AR

Because the wing is trapezoidal, we compute

cr =
The mean aerodynamic chord is
2
c =
3

2 0
c = 4 m.
1+

1 + + 2
1+

cr = 2.8 m.

Wing m. a. chord location. Assuming a uniform load distribution, the distance from the leading edge
of the root chord to the leading edge of the mean aerodynamic chord is

b 1 + 2
tan 0
m
=
6 1+
where

4(0 14 )
tan 0 = tan 1/4
AR

1
1+

= 0.68

which corresponds to a leading edge sweep angle of about 34 . We thus compute

b 1 + 2
m
=
tan 0 = 2.03 m.
6 1+
The leading edge of the mean aerodynamic chord is roughly two meters aft of the wing apex.
Wing m. a. center. We may estimate the location of the wing mean aerodynamic center from Figure
C.3 in [1]. For a taper ratio = 0 and aspect ratio AR = 6, we would find that x
w 0.32
c. For a taper
w 0.26
c. Interpolating for = 41 , we find that
ratio = 21 and aspect ratio AR = 6, we would find that x
x
w 0.29
c = 0.81 m
aft of the leading edge of the mean aerodynamic chord, which means the wing aerodynamic center is 2.84
meters aft of the wing apex.
7

Tail m. a. chord length and location. The mean geometric chord of the tail is
c0t =

St
bt
=
= 1.5 m.
bt
ARt

Because the wing is not tapered ( = 1),

ct = c0t = 1.5 m.
Tail m. a. center. From Figure C.3 in [1], we estimate that the tail mean aerodynamic center is located
at
x
t 0.20
c = 0.30 m
aft of the leading edge of the tail.
Item #2.)

Wing lift-curve slope. Based on the definition of the NACA 4-digit airfoil series, the thickness ratio for
the NACA 2412 is ct = 0.12. (See [2] for details about the definitions of the 4, 5, and 6 digit series.) From
Figure B.1,1 (b) in [1], we find that
(Cl )theory 6.87 rad1 .
Referring again to the definition of the NACA 4-digit series, one may compute
1 t
!

1 t
2 c 90% 2 c 99%
TEw = 2 arctan
= 0.26 rad
0.09
which is around 15 . From Figure B.1,1 (a), we find that
K 0.77.
The 2-D lift-curve slope, corrected for wing thickness, trailing edge angle, Reynolds number and Mach
number, is
1.05
Cl =
K (Cl )theory 6.41 rad1 .
2
1M
We next compute

1 M2
Cl = 0.88.
=
2
The formula for the lift-curve slope CLw requires the sweep angle of the mid-chord line. We compute
(

)
4( 21 14 ) 1
1/2 = arctan tan 1/4
= 0.45 rad
AR
1+
or about 26 . We thus find that
CLw =
2+
or about 0.07 per degree.

2AR

AR2 (1M 2 )
1+
2

tan2

1/2
(1M 2 )

= 4.2 rad1

+4

Wing zero-lift angle of attack. Now, the zero-lift angle of attack of the airfoil is given to be 0L2D =
2.0 . Thus, the zero-lift angle of attack of the wing is

tan 0L2D
= 0.040 rad
0Lw = arctan
cos 1/4
8

or about 2.3 . We thus find that

CLw

= CL w (w 0Lw )
= CL w ( 0Lw )

= 0.07( (2.3))
= 0.16 + 0.07,

where CL w has units of deg1 and is given in degrees.

Item #3.)

Using the formula for Cm0L presented earlier, we find that

!

AR cos2 1/4
Cm0L2D = 0.027.
Cm0Lw =
AR + 2 cos 1/4
Item #4.)

Tail lift-curve slope. The tail lift coefficient is

CLt

= CLt ( (w ) it )

d
= CLt 0 +
it
d

d
= CLt (0 + it ) + CLt 1
.
d

The critical parameters which we must compute are at , 0 , and

d
d .

Based on the definition of the NACA 4-digit airfoil series, the thickness ratio is
B.1,1 (b) in [1], we find that
(Cl )theory 6.75 rad1 .

t
c

= 0.09. From Figure

(The subscript t is omitted with the understanding that all of the following computations relate to the
horizontal tail.) Referring again to the definition of the NACA 4-digit series, one may compute
1 t
!

1 t

TEt = 2 arctan 2 c 90% 2 c 99% = 0.20 rad

0.09
which is around 11 . From Figure B.1,1 (a), we find that
K 0.79.
The 2-D lift slope, corrected for wing thickness, trailing edge angle, Reynolds number and Mach number,
is
1.05
Cl =
K (Cl )theory 6.47 rad1 .
1 M2
We next compute

1 M2
Cl = 0.89.
=
2
9

Because the horizontal tail is unswept, we compute

CLt

=
2+

2+

2AR

AR2 (1M 2 )
1+
2

tan2 1/2
(1M 2 )

2AR

AR2 (1M 2 )
2

= 3.9 rad1

+4

+4

or about 0.07 per degree. Since the horizontal tail is symmetric, the zero-lift angle of attack of the airfoil,
and of the entire tail, is zero.
d
Downwash at the tail. Next, we estimate the downwash parameter d
using the relation
1.19

p
4.44 KA K KH cos 1/4
d

.
=
d
1 M2
The correction factors are:
1
1

= 0.12
KA =
AR 1 + AR1.7
10 3
K =
= 1.32
7

1 hbH
q
= 0.79.
KH =
3

2lH
b

We thus compute

d
= 0.40.
d
To determine 0 , we observe that no downwash is generated when the wing generates no lift. The zero-lift
angle of attack of the wing is 0Lw = 2.3 . We thus compute
d
(w ) = 0 = 0 +
0Lw
d
d
= 0 +
(2.3 )
d
which tells us that 0 = 0.92 = 0.016 rad.
In the end, we obtain
CLt

= CLt (0 + it ) + CLt 1
d
= 0.14 + 0.04

where is measured in degrees. Notice that CLt < 0 when = 0. At zero angle of attack, the tail
generates a downward force which generates a nose-up moment about the center of gravity, as desired.

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.
[2] W. H. Mason.
Geometry for aerodynamicists.
http://www.aoe.vt.edu/aoe/faculty/Mason f/CAtxtAppA.pdf.

10

Notes

available

on-line

at:

Review of Linear, Time-Invariant Ordinary Differential Equations

First order, homogeneous, LTI ODEs. Following is the normal form for a homogeneous, linear,
time-invariant ordinary differential equation of first order.
x + ax = 0,

x(t0 ) = x0 .

(1)

The dot indicates differentiation with respect to time. Suppose we wish to solve (1) for x(t).
Approach #1 : Use an integrating factor. That is, multiply by a factor that will make the left hand side
an exact differential.
eat (x + ax) = 0
xe
at + axeat = 0
d at
xe
= 0
dt

Integrating both sides from t0 to t gives

Z t
Z t
d
(x( )ea ) d =
0 d

t0 d
t0

x(t)eat x(t0 )eat0 = 0

x(t) = x0 ea(tt0 )

Comments:
The integrating factor also works when there is forcing on the right hand side and when the equation
is linear time varying (i.e., when a depends on t).
As stated, the approach only works for first order linear equations. However, it can be generalized
to systems of first order linear equations. (Recall that any number of linear equations of any order
can be rewritten as a set of first order linear equations.)
Approach #2 : Motivated by divine insight, assume a solution of the form
x(t) = cet .

(2)

(Actually, this form seems quite reasonable given the solution in Approach #1.) Substituting into (1) gives
x + ax = 0

d t
ce
+ a cet
= 0
dt
( + a)cet = 0
( + a)x(t) = 0.
Now, x(t) will not be zero, in general; only in the special case that x0 = 0 will x(t) be zero. Since the
identity above must hold for any initial state, it follows that
= a.
To determine c, we evaluate x(t) given in (2) at t0 and use the initial condition:
x(t0 ) = ceat0

c = x0 eat0 .

Substituting the values of and c into (2) gives the solution

x(t) = x0 ea(tt0 ) .
1

Approach #3 : Since this is a linear, time-invariant system we may use the Laplace transform to convert the
differential equation in time into an algebraic equation in s. Having expanded the solution X(s) in a series
of partial fractions, we may then transform back to the time-domain. Assume that t0 = 0. There is no
loss of generality in doing this; because the system is time-invariant we simply change the time coordinate,
shifting the origin of time such that t0 = 0. Taking the Laplace transform of the equation gives
L {x + ax = 0}

(sX(s) x0 ) + aX(s) = 0

X(s) =

x0
.
s+a

Taking the inverse Laplace transform gives

x(t) = L

x0
s+a

= x0 eat .

To better understand the nature of the solution, define the time constant
T =

1
.
a

Dividing through by the initial state (which we assume is nonzero), we have

t
x(t)
= e T .
x0

If a > 0 (so that T > 0), then the solution decays with time. In this case, at time t = T (that is, after one
time constant has elapsed), x(t) is only 37% of its initial value. Slightly before this time, x(t) is exactly
one-half its initial value. To determine this time to half-amplitude thalf , observe that

t
1
1
x(thalf )
half
= =e T

thalf = ln
T 0.69T.
x0
2
2
Suppose that a < 0, so that T < 0. Then x(t) grows without bound. The time required for x(t) to grow
to twice its initial amplitude is obtained as follows:
tdouble
x(tdouble )
= 2 = e T
x0

tdouble = ln (2) T 0.69|T |.

Now suppose x0 = 0. Then x(t) = 0 for all time, regardless of the value of a. Thus, x(t) = 0 is an
equilibrium. Stability of this equilibrium depends on the sign of the constant a. If a > 0, then trajectories
starting near the equilibrium approach it asymptotically and we say that the equilibrium is stable. If a < 0,
then trajectories starting near the equilibrium diverge from it and we say that the equilibrium is unstable.
First order, heterogeneous, LTI ODEs. Following is the normal form for a heterogeneous, linear,
constant coefficient ordinary differential equation of first order.
x + ax = f (t),

x(t0 ) = x0 .

(3)

Suppose we wish to solve (1) for x(t). (Assume that a > 0.)
Approach #1 : Use an integrating factor.
eat (x + ax) = eat f (t)
Integrating both sides from t0 to t gives
Z t
at
at0
ea f ( )d
x(t)e x(t0 )e =
t0

d at
xe = eat f (t).
dt
a(tt0 )

x(t) = x0 e

t0

ea(t ) f ( )d

(4)

Note that x(t) is the sum of a term involving the initial condition and a term involving the force f (t). This
observation reflects the principle of superposition for solutions to linear equations. This principle is useful
in solving more general problems (i.e., heterogeneous equations of arbitrary order).
Approach #2 : The method of undetermined coefficients applies to forced LTI equations where f (t) solves
some linear differential equation. Equivalently, the method applies to forced LTI equations where a finite
set S contains f (t) and its derivatives of all orders. For example, the forcing function
f (t) = t sin(t)
is admissible because the finite set
S = {t sin(t), t cos(t), sin(t), cos(t)}
contains f (t) and its derivatives of all orders. (Or rather, f (t) and its derivatives of all orders may
be constructed by linear combinations of the elements of S.) Note that each element of S is linearly
independent; no element can be formed by a linear combination of the remaining elements. (You may
check this by computing the Wronskian determinant and making sure it is not zero for all t; see your
textbook on differential equations.)
The method of undetermined coefficients proceeds as follows:
1. Solve the associated homogeneous problem, leaving the constants arbitrary. The general solution to
the homogeneous problem is termed the complementary solution.
2. Assume a particular solution as a weighted sum of all the elements of S. Leave the coefficients of
the various terms undetermined.
3. Sum the particular and complementary solutions, substitute into the original equation, and match
the coefficients of like terms. This will determine values for some of the coefficients.
4. Impose the initial conditions on the resulting solution to determine the remaining coefficients.
Approach #3 : Use the Laplace transform.
L {x + ax = f (t)}

(sX(s) x0 ) + aX(s) = F (s)

X(s) =

x0
1
+
F (s).
s+a s+a

Taking the inverse Laplace transform, we find

x(t) = x0 eat + eat f (t)
Z t
at
= x0 e
+
ea(t ) f ( )d.
0

Compare this solution with (4).

Second order, homogeneous, LTI ODEs. We now turn our attention to the equation
x
+ a1 x + a0 x = 0,

x(t0 ) = x0 , x(t
0 ) = x 0

(5)

where a1 and a0 are scalar constants.

Approach: Again, assume that
x(t) = cet

(6)

Substituting the assumed form of x(t) into the equation gives

0 = 2 (cet ) + a1 (cet ) + a0 (cet )
= (2 + a1 + a0 )x(t).
We assume the initial condition is such that x(t) 6= 0 for all time. Then it must be true that
2 + a1 + a0 = 0.
This quadratic equation has two distinct solutions, in general,

q
1
2
1,2 =
a1 a1 4a0 .
2
Thus, x(t) generally contains two components of the form (6). By the principle of superposition, the
general solution is a sum of the two:
x(t) = c1 e1 t + c2 e2 t .
(7)
Evaluating at the initial conditions determines the parameters c1 and c2 . In the case that a21 4a0 = 0, the
two candidate solutions are linearly dependent. In that case, we obtain a secular term in the solution:
x(t) = (c1 + c2 t)e

a1
t
2

Generically, the initial condition response of a second order system can be described either as exponential
convergence or exponential divergence. The former case occurs whenever both eigenvalues have negative
real part. The latter case occurs if either eigenvalue has a positive real part. The boundary which
separates these two generic behaviors occurs when one or both of the eigenvalues lie on the imaginary axis.
In this case, and assuming both eigenvalues are not zero, the initial condition response is a sinusoid whose
frequency is the magnitude of the imaginary part.
If the initial condition response involves convergence to zero, then we say that the equilibrium at (x, x)
=
(0, 0) is stable. If the initial condition response involves divergence from zero, then we say that the
equilibrium at (x, x)
= (0, 0) is unstable. Otherwise, we say that the equilibrium at (x, x)
= (0, 0) is
neutrally stable.
We will be most concerned with the case where a0 > 0 (which is necessary for static stability). Define
the natural frequency n and the damping ratio as follows:
n =

a0

and

a1
= .
2 a0

With these definitions, the equation of motion becomes

x
+ 2n x + n2 x = 0,

x(t0 ) = x0 , x(t
0 ) = x 0

and the characteristic equation is

2 + 2n + n2 = 0.
The characteristic values are

p
1
2n 4 2 n2 4n2

2
p
=
2 1 n .

1,2 =

Consider the following two cases:

4

Im

Im

Im

!d
!n
Re

Re

-! n

increasing through 1

Re

decreasing through -1

-1 < < 1

Figure 1: Characteristic Values of a Second Order System as the damping coefficient varies.
1. 0 2 < 1: In this case, the characteristic values are a complex conjugate pair. We may write
p
1,2 = n j 1 2 n
= n jd .

The initial condition response takes the form

1
n t
(x 0 + n x0 ) sin d t
x(t) = e
x0 cos d t +
d
If > 0, this response is a damped oscillation and we say that the equilibrium at the origin is
dynamically stable, as well as statically stable. If < 0, the response diverges and we say that the
equilibrium at the origin is dynamically unstable. If = 0, the response oscillates forever and we say
that the equilibrium at the origin is neutrally stable.
Referring to the factor en t in the expression above, the real part n plays a role similar to the
reciprocal of a first order time constant. As in the case of a first order system, we may compute the
time to half-amplitude (if 0 < < 1) or the time to double amplitude (if 1 < < 0):
thalf or tdouble
Since the period of oscillation is Td =
double amplitude as
Nhalf/double

2
d ,

0.69
.
||n

we may also compute the number of oscillations to half or

thalf/double
=

Td

0.69
||n

1 2
d

= 0.11 2 .
2

2. 2 > 1: In this case, the characteristic values are real. If > 1, then both characteristic values
are negative. The response is an overdamped convergence to the equilibrium which is dominated by
the larger (less negative) characteristic value. Thus, the response is essentially a first order response
and one computes the time to half amplitude as one would for a first order system (using the larger
characteristic value). As grows larger positive, one characteristic value approaches zero from the left
while the other moves toward . As the larger characteristic value approaches zero, convergence
becomes slower and slower.
If < 1, then both characteristic values are positive. The response is a non-oscillatory divergence
which is dominated by the larger (more positive) characteristic value. One computes the time to
double amplitude as for a first order system, using the larger characteristic value. As grows larger
negative, the dominant characteristic value moves toward and the other approaches zero from the
right.
5

Second order, heterogeneous, LTI ODEs. Second order LTI ODEs with forcing can be solved using
the method of undetermined coefficients or Laplace transforms exactly as in the case of first order ODEs.1
Higher order systems. We have seen that the the response of a first order system is (generically)
exponential growth or decay. The response of a second order system is either a combination of first order
responses (in the case that both characteristic values are real) or is an oscillatory exponential growth
or decay (in the case that the characteristic values are complex conjugates). In fact, the roots of a
polynomial with real coefficients are always real or complex conjugate pairs. Thus, with the exception of
the degenerate case of repeated characteristic values, the general response of an nth order linear system
can be understood as a superposition of responses of first and second order responses.

1
Even the technique of an integrating factor extends to second order equations, provided they are re-written as a system
of first order equations; since this can always be done, the technique applies in general. The generalized integrating factor
method is related to the method of variation of parameters which you may have seen in your course on differential equations.

Evans Rules for Sketching the Root Locus

Absolute and Relative Stability. A control system is called absolutely stable if the controlled transfer
function Hd (s) from the reference signal yd (s) to the output signal y(s) has all of its poles in the open
left half plane. One technique for determining absolute stability of a control system is the Routh-Hurwitz
stability analysis technique. This very useful technique is presented in Section 5-7 of [1].
Absolute stability is an essential quality for a control system, but it says nothing about the performance
characteristics of the system, i.e., the transient response. Two absolutely stable systems can respond to
a step input in very different ways; one might exhibit a very slow, overdamped response while the other
exhibits a very fast, underdamped response.
To compare the performance of two absolutely stable systems, it is useful to consider the notion of relative
stability or degree of stability. Degree of stability can be rather narrowly defined as the horizontal
distance between the imaginary axis and the nearest pole. This distance will typically determine the speed
of response of the system, however it tells you nothing more about the nature of that response (e.g., if
it the system is overdamped, critically damped, or underdamped). More generally, one may examine the
specific locations of the closed-loop poles. Knowing these pole locations gives a good sense of the nature
of the systems transient response.
The Root Locus Method. The root locus method, also known as Evans rules in honor of W. R.
Evans, is a technique for determining how the poles of a feedback control system move in the complex plane
as a parameter is varied. Typically, the parameter is a control gain, although any parameter of interest
can be used. (For this reason, the root locus method is useful in dynamical system theory, where one is
often interested in sudden changes in a systems qualitative behavior, called bifurcations, as a parameter
varies.)

yd

Fd

Figure 1: One degree of freedom closed-loop control structure.

Consider the simple feedback control system shown in Figure 1. The closed-loop transfer function is
Hd (s) =

P (s)Fd (s)
y(s)
=
.
yd (s)
1 + P (s)Fd (s)

Closed-loop poles are values of s for which

1 + P (s)Fd (s) = 0.
Since P (s)Fd (s) is a function of a complex variable, the equation P (s)Fd (s) = 1 can be expressed in
terms of the magnitude and phase of P (s)Fd (s):
|P (s)Fd (s)| = 1

and

P (s)Fd (s) = (2k + 1) k = 0, 1, 2, . . .

In words, the magnitude of the loop gain is always one and the phase is an odd power of .
Suppose that P (s)Fd (s) can be written in the form
P (s)Fd (s) = K

b(s)
.
a(s)

This would be the case, for example, if P (s) = b(s)/a(s) and Fd = K, as for a simple proportional
controller. The control structure might be more complicated than this, however we assume that a the
multiplicative factor K appears and that this parameter may vary.
The root locus is the locus of possible roots of the closed-loop transfer function as the multiplicative
parameter K is varied. In fact, the entire root locus can be determined from the angle condition alone. The
magnitude condition is then used to determine which value of K corresponds to which set of closed-loop
poles along the locus of all possible closed-loop poles.
Rather than learn Evans rules to begin with, it is more illustrative to consider a series of increasingly
complicated examples.
Example 1. To begin, we consider the very simple example
P (s)Fd (s) = K

1
.
s(s + 2)

We will compute the closed-loop poles as explicit functions of K. In general, this is a tedious, and
uninformative exercise, but for this simple system it serves to illustrate how closed-loop poles vary as the
gain K is varied. The closed-loop transfer function is
Hd (s) =

K
s(s+2)
K
+ s(s+2)

The closed-loop poles are obtained from

K
s(s + 2)
= s2 + 2s + K.

0 = 1+

They are

1
(2 4 4K)
2

= 1 1 K.

s =

Im

Re

Figure 2: Root locus example #1.

2

When 0 < K < 1, there are two distinct poles which are located on the real axis between 0 and 2. When
K = 1, the poles coalesce at s = 1. As K continues to increase, the poles split apart and move in opposite
directions parallel to the imaginary axis.
To see that the locus of closed-loop poles shown in Figure 2 can be obtained from the angle condition

1
= (2k + 1) k = 0, 1, 2, . . . ,
s(s + 2)

we first recall some facts about complex numbers. First, a complex number can be represented in polar
form, for example z = rei where r is the radial distance from the origin to the point z and is the angle
to z measured counter-clockwise from the positive real axis. Consider the complex function
C(s) =

(s z1 ) (s zm )
.
(s p1 ) (s pn )

Each term in the numerator can be considered a vector from the zero zi to the point s. Similarly, each
term in the denominator can be considered a vector from the pole pi to the point s. Each of these vectors
has a magnitude and an angle, so we may equivalently write

rz1 eiz1 rzm eizm

C(s) =
rp1 eip1 rpn eipn

rz 1 r z m
ei(z1 ++zm p1 pn )
=
r p1 r pn
where rzi (or rpi ) is the magnitude of the vector from zi (or pi ) to s and zi (or pi ) is the angle of the
vector from zi (or pi ) to s.
Im

qp2

qp1

Re

Figure 3: Angle condition for root locus example #1.

Applying these observations to the current example, we find that

1
= s (s + 2).
s(s + 2)

(1)

Now, for any point on the real axis to the right of p1 = 0, equation (1) gives zero, which is not an odd
number times . Similarly, for any point on the real axis to the left of p2 = 2, equation (1) gives 2,
3

which is also not an odd number times . Thus, the real axis to the left of p2 and to the right of p1 is not
part of the root locus. However, for points between p2 and p1 , equation (1) gives

1
= s (s + 2) = 0,
s(s + 2)

which is an odd number times . Thus, points on the real axis between p2 and p1 are part of the root
locus.
1
Considering next the points on the vertical line s = 1, we choose a point and determine s(s+2)
. The
vectors from p1 and p2 to any such point form an isosceles triangle. The sum of the two angles is for
1
= or 3, respectively.
points above the real axis and 3 for points below the real axis, giving s(s+2)
Thus, the line s = 1 is part of the root locus.

To find the value of K which corresponds to a particular pair of closed-loop poles, we use the magnitude
condition.For example, suppose we would like to choose K so that the closed-loop system has adamping
2
ratio = 22 . Any pole lying on the radius = 3
4 in the complex plane has damping ratio = 2 . Thus,
we would like to choose K to give closed-loop poles at
s = 1 i tan

= 1 i.
4

Choosing a particular pole, say s = 1 + i, we substitute into the magnitude condition to obtain

(1 + i)((1 + i) + 2) = 1
or

K = |(1 + i)(1 + i)| = | 2| = 2.

Thus, choosing the gain K = 2 gives the closed-loop poles s = 1 i.

We have assumed that P (s)Fd (s) can be written in the form
P (s)Fd (s) = K

b(s)
.
a(s)

where b(s) has degree m, a(s) has degree n m and where K > 0 is a parameter (e.g., a control gain)
which may vary.
An important observation is that, as K 0, the closed-loop poles approach the poles of the loop gain. To
see this, write the closed-loop characteristic equation as
a(s) + Kb(s) = 0.
Clearly, as K 0 the roots of the polynomial on the left approach the roots of a(s).
One may also observe that, as K , the closed-loop poles must either diverge to or approach a zero
b(s)
must become very small so that the product
of the loop gain. To see this, recognize that as K , a(s)

b(s)
is 1. There are two ways that a(s)
can become very small. First, b(s) can go to zero (which happens
when s approaches a zero of the loop gain). Second, a(s) can go to infinity (which can only happen when
|s| goes to infinity.) In general, m branches of the root locus approach the zeros of the loop gain while the
remaining n m branches go to infinity.

Example 2. Consider the following example from [1]:

P (s)Fd (s) =

K
.
s(s + 1)(s + 2)
4

This system has poles at p1 = 0, p2 = 1, and p3 = 2. Recalling that

m
X

P (s)Fd (s) =

i=1

(s zi )

n
X
j=1

(s pj ),

we first consider which, if any, points on the real axis are part of the root locus. For any point to the
right of s = 0, P (s)Fd (s) = 0, so the positive real axis is not part of the root locus. For any point
1 < s < 0, P (s)Fd (s) = , so these points are part of the root locus. For any point 2 < s < 1,
P (s)Fd (s) = 2, so these points are not part of the root locus. Finally, for any point s < 2,
P (s)Fd (s) = 3, so these points are part of the root locus.
Next, we consider what happens to the root locus as s grows large. In the limit that s grows large, we have
K
K
= lim 3 .
|s| s(s + 1)(s + 2)
|s| s

lim P (s)Fd (s) = lim

|s|

Now, no matter how large |s| is, the angle condition must be satisfied, so we must have
lim P (s)Fd (s) =

|s|

lim P (rei )Fd (rei )

lim

K
(rei )3

= ei3
= (2n + 1)

n = 0, 1, 2, . . .

or

2n + 1
.
3
Trying n = 0 gives = 3 . Trying n = 1 gives = . Trying n = 2 gives = 5
3 . Other choices of n
give repeated angles. In the limit that |s| , the three closed-loop poles follow asymptotes that extend
radially in the directions 3 and .
=

A logical question is How large must |s| be before the root locus converges to these asymptotes? Rephrasing the question, How does the root locus look for smaller values of |s| given that it converges to these
asymptotes as |s| grows large? A partial answer can be obtained by determining where the three asymptotes are centered. A simple way to determine this is to notice that a slightly more precise approximation
for P (s)Fd (s) for large |s| is
K
.
lim P (s)Fd (s) = lim
|s|
|s| (s + 1)3
To see this, compare the polynomials
(s + 1)3 = s3 + 3s2 + 3s + 1

and

s(s + 1)(s + 2) = s3 + 3s2 + 2s.

The two polynomials agree to next-to-highest order. The root locus for the large-|s| approximation is the
three asymptotes computed previously, centered at the point s = 1. For the true system, the root locus
will behave a bit differently for small |s|, but we have at least located the origin of the asymptotes which
describe the large |s| behavior.
Two of the three asymptotes extend into the right half complex plane, while the third follows the negative
real axis. Intuitively, the closed-loop pole which starts (for small K) at s = 2 will follow the negative real
axis asymptote as K increases. Therefore, the two closed-loop poles which rest on the real axis between
s = 1 and s = 0 must coalesce and split off to follow the asymptotes at 3 .1
1
They must first coalesce because poles must be either real numbers or complex conjugate pairs and because the closed-loop
pole locations vary continuously with K.

To determine precisely where this split occurs, known as a breakaway point, we recognize that there
must be a double pole between s = 1 and s = 0 for some value of K. Recall from System Dynamics that
a double-root s = s of a polynomial Q
C(s) satisfies not only C(
s) = 0, but also C 0 (
s) = 0. You can see this
n2
2
by recognizing that C(s) = (s s)
(s

p
)
if
there
is
a
double
pole
at
s

.
i
i=1

For the feedback control system, we therefore have not only a(s) + Kb(s) = 0, when the closed-loop poles
coalesce, but also
d
(a(s) + Kb(s)) = 0.
ds
Solving for K, the value of the parameter for which the double-pole occurs, we find
K=

a0 (s)
.
b0 (s)

Substituting back into the condition for a closed-loop pole, we have

a(s)

a0 (s)
b(s) = 0
b0 (s)

or
b0 (s)a(s) a0 (s)b(s) = 0

(2)

Im

Re

Figure 4: Root locus example #2.

For our system, we have b(s) = 1 and a(s) = s(s + 1)(s + 2). Solving (2) for the multiple-pole, we obtain
(3s2 + 6s + 2) = 0.
Two solutions are

6 12
3
s=
= 1
6
3

The solution is not on the root locus, so the breakaway point must be s = 1 +

3
3

0.4.

Note: You can find the value of the gain K at which the root locus passes into the right half plane by
performing a Routh-Hurwitz stability analysis and finding conditions on K for stability.
Example 3. Next consider a system for which
P (s)Fd (s) = K
6

s2 + 3s + 2
.
s2 + 2s + 4


The loop gain has two zeros at z1 = 1 and z2 = 2 and two poles at p1,2 = 1 i 3.
First, determine if any closed-loop poles lie on the real axis. The vectors from the two poles to points on
the real axis form an isosceles triangle. The sum of their contribution to P Fd is 2. Similarly, a pair of
complex conjugate zeros would contribute 2 to the angle calculation. Thus, we see that complex conjugate
poles and zeros have no impact on whether a given segment of the real axis is part of the root locus. The
real axis poles and zeros alone determine which portion of the real axis is part of the root locus. In fact,
one can easily verify that the following statement is true:
The portion of the real axis which is part of the root locus lies to the left of an odd number of poles
and zeros.
For the present example, we observe that the only part of the real axis which contributes to the root locus
lies between the zeros at z1 and z2 . Recall that for small values of K, the closed-loop poles are close to the
poles of the loop gain P (s)Fd (s). To determine how the closed-loop poles leave p1 and p2 and approach
the real axis locus, we first compute the angle at which the locus departs from p1 . (Of course, the locus
near p2 will simply be a mirror image about the real axis.) Taking a test point s very near p1 , it is easy to
see that the contributions of z1 , z2 , and p2 to P (s)Fd (s) will remain more or less constant as we move s
in a small circle around p1 . Choosing s = p1 , a point near p1 , we must have
P (
p1 )Fd (
p1 ) = (2k + 1)

k = 0, 1, 2, . . .

= (
p1 z1 ) + (
p1 z2 ) (
p1 p1 ) (
p1 p2 )
(p1 z1 ) + (p1 z2 ) d (p1 p2 )

where d is the departure angle from p2 . Taking k = 0, we therefore have

d = + (p1 z1 ) + (p1 z2 ) (p1 p2 )
!
!
!
3
3
2 3
= + arctan
+ arctan
arctan
0
1
0

= + +
2
3
2
2
= .
3
By mirror symmetry, the departure angle from p2 must be

2
3 .

Next, we compute the breakin point exactly as we did before, i.e., by using the condition for the existence
of a double-pole. From the coalescence condition
b0 (s)a(s) a0 (s)b(s) = 0,
we require that

0 = (2s + 3) s2 + 2s + 4 (2s + 2) s2 + 3s + 2

= s2 + 4s + 8.

Two solutions
are s = 2 2 3; only the solution lies on the root locus, so the breakin point is

s = 2 2 3 1.46.
A total of m closed-loop poles approach the zeros of the loop gain P (s)Fd (s) as K . The
remaining n m closed-loop poles follow asymptotes outward to infinity.
7

Im

x
2p/3

Re

Figure 5: Root locus example #3.

Following is the general procedure for constructing a root locus plot, as adapted from [1].
Step 1. Locate the poles and zeros of the loop gain P (s)Fd (s). First, compute the zeros of the loop
gain (the roots of b(s)) and place an o at their location in the complex plane. Next, compute the poles
of the loop gain (the roots of a(s)) and place an x at their location in the complex plane.
Step 2. Determine what, if any, portion of the real axis is part of the root locus. The angle
condition requires that the real axis portion of the root locus lies to the left of an odd number of poles and
zeros. Equivalently, since complex poles and zeros must occur in conjugate pairs, the real axis portion of
the root locus lies to the left of an odd number of real poles and zeros.
Step 3. Determine the asymptotes of the root locus. Given that there are m zeros and n m
poles, m of the closed-loop poles will approach the loop gain zeros as K and the remaining n m
will converge to asymptotes which extend radially to infinity from some starting point on the real axis.
The asymptote angles are
(2k + 1)
k = 0, 1, 2, . . . ,
nm
which can be proved by approximating the loop gain with

K
snm

for large values of |s|.

The center of the asymptotes can be computed from a slightly better approximation obtained as follows.
Write
b(s)
a(s)

=
=

(s z1 ) (s zm )
(s p1 ) (s pn )
sm + (z1 zm )sm1 +
sn + (p1 pn )sn1 +

Dividing both the numerator and the denominator by the numerator gives
b(s)
1
= nm
a(s)
s
+ ((z1 + + zm ) (p1 + + pn )) snm1 +

(3)

For large values of |s|, this ratio of polynomials can be approximated by

(z1 + + zm ) (p1 + + pn ) (nm)

.
s+
nm
8

(4)

That is, (3) matches (4) to order snm1 . The root locus for this (approximate) loop gain consists of n m
rays extending radially from the point
=

(p1 + + pn ) (z1 + + zm )
.
nm

The real number is the center of the asymptotes for root locus corresponding to the true loop gain.
Step 4. Find the breakaway and breakin points. Recall that these points correspond to values of
the gain K for which the closed-loop system has multiple closed-loop poles at a particular point. For a
double-pole, the condition
b0 (s)a(s) a0 (s)b(s) = 0
must be satisfied. The roots of this algebraic equation give possible breakaway or breakin points. To
determine whether these are, in fact, breakaway or breakin points, one must check whether these points
are actually on the root locus.
Step 5. Determine the angles of departure from the loop gain poles and the angles of arrival
at the loop gain zeros. Recall that as K 0, the root locus approaches the poles of the loop gain and
as K , m branches of the root locus approach the zeros of the loop gain. The angle of departure from
the k th loop gain pole pk can be obtained from the angle condition as
X
X
d = +
(pk zi )
(pk pj ).
j6=k

That is, the departure angle is plus the sum of all the angles of vectors pointing from the loop gain zeros
to pk minus the sum of all the angles of vectors pointing from the remaining loop gain poles to pk .
Similarly, one can use the angle condition to show that the angle of arrival at the k th loop gain zero zk is
X
X
a =
(zk zi ) +
(zk pj ).
i6=k

In general, it is a good idea to also compute the value of K at which the root locus crosses into the right
half of the complex plane for the first time. This can be done using the Routh-Hurwitz procedure. The
gain value at which the root locus first crosses into the right half plane generally serves as an upper limit
on acceptable choices of the parameter K.
To determine the value of K corresponding to a particular closed-loop pole on the root locus, one must
use the magnitude condition. Recognizing that

b(s)
a(s)
K

=1
,

K=
a(s)
b(s)

we have, for a particular closed-loop pole s,

j
K= Q

|(
s pj )|

s zi )|
i |(

Example: Stabilizing an Inverted Pendulum. The nondimensional equation for an inverted pendulum
is
n2 = n2 u,
so the transfer function from torque to angle is
G(s) =

n2
.
s2 n2
9

Root Locus
10

Imaginary Axis

10
10

10

Real Axis

Figure 6: Root locus for proportional feedback.

The system has no zeros and two real-conjugate poles.
We start by applying proportional feedback Fd (s) = kp . The loop gain becomes
P Fd = kp

n2
.
s2 n2

The root locus is shown, for particular parameter values, in Figure 7. Clearly the feedback control law
only marginally stabilizes the system, provided kp is large enough.
Root Locus
20

15

10

Imaginary Axis

10

15

20
40

35

30

25

20

15

10

10

Real Axis

Figure 7: Root locus for proportional-derivative feedback.

Next, we apply proportional-derivative feedback
Fd = kd s + kp = kp

kd
s+1 .
kp

The loop gain becomes

P Fd = kp

1
2
s + 1 n
s2 n2

10

where = kpd . The compensator introduces a new loop-gain zero at . We will assume that remains
constant as kp is varied. (That is, kd varies in direct proportion to kp .) The root locus is shown in Figure 7.
Clearly, the feedback controller stabilizes the system. The independent freedom in kp and allows one to
obtain any desired closed-loop pole locations (provided the poles are both real or are complex conjugates).
Example: Longitudinal Autopilot. The longitudinal dynamics of the airplane shown in Figure 8 are
described by the following equations

V
u1
D
0
1 0

T
T

+ R ()
+ R ()R()

L
1
u
0
2
V
=

u2
q

where R() is the proper rotation matrix

R() =

cos sin
sin cos

and the aerodynamic forces satisfy

D = KD V 2 2

(5)

L = KL V (1 + ).

(6)

(Note: These equations have been normalized. All quantities are dimensionless.) The inputs are u1 ,
representing thrust, and u2 , representing the elevator deflection. The small positive scalar accounts for
a slight downward force due to positive deflections of the elevator.
u

v
V

a
q

Figure 8: Longitudinal aircraft motion.

Define the state vector x = [V, , , q]T . Assuming that KL = 1/V 2 , we linearize the dynamics about
the equilibrium x = [V , 0, 0, 0]T . A somewhat tedious series of calculations gives the linearized dynamics

0
1 1 0
1 0

22 1

0 0
d
x + 0 V u.
V
V
x =
0
0 0
0
0 1
dt
0

Suppose we take the output of interest to be flight path angle :

y = = [0, 1, 1, 0]x.
The transfer function from the elevator u2 to the flight path angle is
1

0
s
1
1 0
22 s + 1 0 0
V =
V
V
G(s) = [0, 1, 1, 0]
0
0
s 1 0
0
0
0 s
1
11

V s2

s2 s2 +

+s+
1
V

s+

1
V

2
V 2

2
s
V

=
s s2 +

1
V

s+

1
V

2
V 2

Note that this transfer function is nonminimum phase for all > 0. Recall that a stable, non-minimum
phase system initially responds to a positive step input in the negative direction. Physically, the downward
force on the elevator and tail fin due to a pitch-up command cause the airplane, initially, to accelerate
downward. This results in a negative flight path angle until the integrated effect of the tail moment
increases the aircraft pitch angle sufficiently to provide upward lift.
Root Locus
6

Imaginary Axis

6
2

10

12

Real Axis

Figure 9: Root locus for proportional feedback.

We would like to design an autopilot to regulate the flight path angle for an aircraft with flight conditions
and aircraft parameter values given by
V = 1,

and

= 0.25.

(In fact, the aircraft is already stable, but its motion is very lightly damped. We start by applying
proportional feedback Fd (s) = kp . The loop gain becomes

1
2
s

V
V

P Fd = kp
s s2 + V1 s + V22
Consider the root locus shown in Figure 9. Observe that straight proportional feedback drives this slightly
stable system unstable! The right half plane zero draws the branches of the locus into the right half plane.

Notice that the root locus in Figure 9 appears not to satisfy Evans rules. For example, the real axis locus
does not lie to the left of an odd number of poles and zeros. This is a consequence of the negative sign in
the numerator, the very term which makes the system nonminimum phase. Essentially, it is as if we have
changed the sign of the proportional gain and are now applying positive feedback. For positive feedback,
the angle condition changes to
P Fd = 2k
k = 0, 1, 2, . . . ,
and Evans rules change accordingly.
To stabilize the system, suppose we introduce an additional left half plane zero in the loop gain by applying
proportional derivative feedback. This additional zero will have the effect of drawing the branches of the
12

Root Locus
5

Imaginary Axis

5
15

10

10

Real Axis

Figure 10: Root locus for proportional-derivative feedback.

root locus into the left half plane. We choose
Fd = kd s + kp = kp

kd
s+1 .
kp

The loop gain becomes

P Fd = kp
where, once again, =

kp
kd .

+ 1 V s2 s V1

.
s s2 + V1 s + V22

1
s

Figure 10 shows the root locus for this system. The branch coming into the RHP zero from the right is
an extension of the branch moving the left. It is merely an artifact. (Root loci for nonminimum phase
systems are just weird.) Thus, we see that proportional derivative control provides stable regulation of the
flight path angle.
Example: Parametric Stability Analysis for Watts Regulator. Watts regulator is an early
example of mechanical feedback control. This apparatus can be simply modeled as a planar pendulum
which is made to rotate about its vertical axis by some device, such as a steam engine, whose speed is to
be controlled. The speed of the engine is governed by the regulator through a mechanical linkage between
the pendulum and the engine throttle. A simple dynamic model is

Figure 11: A simple model for Watts regulator.

13

+ b + sin c2 v cos = 0

,
v + av = a v k( )

(7)
(8)

where is the elevation angle between the pendulum and the vertical axis and v represents the (square of
the) angular speed of the regulator shaft. The nominal value of v is v and the nominal value of is
2
c
.
= arccos
v
The parameter a > 0 determines the response time of the engine. The constant b > 0 is a damping
parameter for the regulator and c > 0 is the undamped natural frequency of planar swinging motion
(when v 0). The parameter k is a control gain.
As manufacturing processes improved in the latter half of the nineteenth century, friction in mechanical
linkages was dramatically reduced. Coincidentally, new machines whose speed was controlled by Watt
regulators began to exhibit undesired speed oscillations due to oscillations in the elevation angle . This
behavior was referred to as hunting. Lets use a root locus plot to investigate this phenomenon. To do
so, suppose that, for a particular regulator,
a=

61
81

and

k = 80 3

and that the regulators nominal (equilibrium) state is

v)e = (,
0, v) =
(, ,

, 0, 80 .

Linearizing the dynamics about this equilibrium and computing the characteristic equation for the state
matrix gives

or

0 = 813 + (61 + 81b)2 + (4860 + 61b) + 4880

= ( + 1)(812 20 + 4880) + b(812 + 61)

0=1+b

812 + 61
.
( + 1)(812 20 + 4880)

The key observation is that the eigenvalue equation can be written in a way which resembles the characteristic equation for a feedback control system with loop gain
812 + 61
.
( + 1)(812 20 + 4880)
Evans rules apply equally well to this problem. Note that the loop gain has two zeros at = 0 and at
61
. There are also three loop gain poles: at s = 1 and at
= 81

20 400 81 4880
s =
162
0.123 i3.879.
Note that the closed-loop poles, that is the characteristic values, have positive real part for small values
of b. As b is increased, two branches of the root locus approach the two loop gain zeros and the third
follows the asymptote along the negative real axis. The portion of the root locus which lies on the real
61
and to the left of the pole at s = 1.
axis lies between the zeros at s = 0 and s = 81
We know that the three branches of the root locus leave the poles of the loop gain and that two of these
branches somehow approach the two zeros while the third diverges to infinity along the negative real axis.
14

One way this could happen is that the two branches originating at the complex conjugate poles could
coalesce between s = 0 and s = 61
81 . However, there is another possibility which becomes apparent when
we attempt to compute the break-in point. To compute the break-in point, we define
b(s) = 81s2 + 61s
a(s) = 81s3 + 61s2 + 4860s + 4880
and compute

b0 (s)a(s) a0 (s)b(s) = (162s + 61) 81s3 + 61s2 + 4860s + 4880 243s2 + 122s + 4860 81s2 + 61s
= 6561s4 9882s3 + 389939s2 + 790560s + 297680

which has roots at

s = 7.432,

s = 1.529,

s = 0.502,

and

s = 7.957.

Note that all but the last point lie on the root locus. The root locus is clearly more complicated than
we first imagined. A simple exercise in logic shows that the only feasible shape for the root locus is the
one depicted in Figure 12. Because branches of the root locus leave loop gain poles as b increases, the
point s = 1.529 must be a break-away point rather than a break-in point. The remaining two points
are obviously break-in points. Suppose the two branches which leave the complex conjugate poles were to
break in at s = 0.502 and converge to the zeros, as we originally hypothesized. Then we would have a
contradiction between the fact that the real axis to the left of s = 1 is part of the root locus and the fact
that there is a break-away and break-in point to the left of the pole at s = 1. (Note: a single pole cannot
suddenly split and become two poles!)
The angle of departure for the pole at 0.123 + i3.879 is
d = + 1.539 + 1.349 1.289

= 3.170

and, by mirror symmetry, the angle of departure for its complex conjugate is 3.170 radians.
Im
x

xo

Re

Figure 12: Root locus as b varies from 0 to . Note that as b 0, the closed loop eigenvalues move into
the right half plane.
The hunting phenomenon is a consequence of the fact that, while the damping coefficient had previously
been sufficiently large to ensure stability of the closed-loop system, improvements in machining lowered b
15

below a critical value for stability. The oscillatory behavior referred to as hunting is a consequence of
what dynamical systems theorists call a Hopf bifurcation in which a complex conjugate pair of eigenvalues
crosses over the imaginary axis as a bifurcation parameter is varied. The equilibrium about which the
dynamics are linearized becomes unstable and the nonlinear system begins to exhibit a periodic oscillation
about the equilibrium state. Note that this oscillation is not predicted by the linearized dynamics; the
right half plane poles of the linearized system would suggest that the state diverges exponentially, but this
does not happen in reality. The oscillation is a nonlinear phenomenon referred to as a limit cycle.

References
[1] K. Ogata. Modern Control Engineering, Fourth Ed. Prentice Hall, Upper Saddle River, NJ, 2002.

16

Lecture 1: Introductory Remarks

Aerospace engineering encompass a broad range of challenging topics which must be mastered in order to
design atmospheric and space flight vehicles. Topics of fundamental importance include
aerodynamics
propulsion
structures and materials, and
vehicle dynamics and control.
A fifth topic, vehicle design, envelops all of these more basic areas. It involves integrating knowledge in each
of the four subject areas in order to synthesize a complete vehicle which satisfies prescribed performance
requirements.
This course focuses on dynamics and control of atmospheric flight vehicles, particularly fixed-wing aircraft.
This topic is also referred to as flight mechanics. Flight mechanics comprises three major subtopics:
performance,
stability and control, and
aeroelasticity.
Conventionally, each of these subtopics is studied individually although the three are very much related.
In studying aircraft performance, one considers issues such as range, take-off and landing distance, and
trajectory planning for a given aircraft. This involves determining the forces necessary to achieve a given
path of motion, assuming that these desired forces can be generated. Thus, one typically models the
aircraft as a point mass subject to three control forces: lift, side force, and thrust. Performance is
concerned with the large-scale aircraft motions associated with takeoff, landing, turning, etc.
In studying stability and control, one takes a closer look at the aircraft and recognizes that lift and side force
are not true control forces. Rather, these forces are a consequence of the aircrafts orientation with respect
to the local air flow. To generate a desired lift force, for example, the vehicle must effect a particular
angle of attack. Thus, in stability and control, one is concerned with how the vehicles orientation, or
attitude, changes under the influence of moments generated by the actuators. These moments are typically
generated by a pilot through a suitably designed interface (such as a stick and pedals).
In studying performance, one assumes that the aircraft is a point mass. In studying stability and control,
one typically assumes that the aircraft is a rigid body. In studying aeroelasticity, one recognizes that
no aircraft is truly rigid and, moreover, that changes in the vehicle shape due to varying load conditions
can have dramatic effects on the vehicles motion. Aeroelastic phenomena that can arise for real aircraft
include wing or control surface flutter, roll control reversal, and other effects.
In this course, we will consider only the second sub-topic: stability and control. As the course title suggests,
there are two issues of primary importance. Stability relates to the intrinsic flying qualities of the aircraft.
Stability is a characteristic of the vehicle dynamics which is independent of the pilots actions.
Control relates to a pilots interaction with the aircraft. Of interest are the following two questions:
How effective are the various actuators at forcing the aircraft into a desired motion?
1

 How much effort is required of the pilot to generate the necessary actuator commands?
Issues related to these two questions include actuator placement and sizing, stick motion cues for aircraft
with power-assisted actuators, and the use of feedback control to enhance handling qualities and reject
disturbances.
We will begin by considering stability. To discuss stability of a steady motion, we must first introduce
some terminology to describe the motion. Suppose we fix a reference frame to some point in the aircraft,
as shown in Figure 1. We denote by xB the unit vector pointing through the nose of the aircraft. This
axis is often referred to as the longitudinal axis. We let z B represent the unit vector pointing through the
belly of the aircraft; this is often called the directional axis. Finally, we define the lateral axis in terms of
the unit vector y B = z B xB . Viewing the aircraft from behind, y B points to the right.
!

yB
xB
zB

V
yI
xI
zI

Figure 1: Inertial and body-fixed reference frames.

To describe the orientation of the aircraft, we define an inertial reference frame, which is denoted by the
fixed unit vectors xI , y I , and z I . The reason we choose to describe the aircrafts orientation with respect
to an inertial frame is that Newtons laws of motion only hold in an inertially fixed frame. In this course,
we will typically consider an earth-fixed frame to be an inertial frame. Although the resulting equations
of motion will technically be incorrect, the error due to the earths rotation, its revolution about the sun,
etc. will be small over the time periods of interest in studying stability and control.
As the aircraft is assumed to be rigid, the location of any point in the airplane is uniquely determined
by the position and orientation of the body-fixed reference frame. Therefore, we will often represent the
aircraft simply by its body-fixed reference frame. Suppose that the aircraft (i.e., the body frame) translates
at some velocity with respect to the inertial frame. We let

u
V = v
w
denote the translational velocity of the body with respect to the inertial frame, but expressed in the body
frame.1 Also, suppose that the aircraft rotates at some angular velocity with respect to the the inertial
1

Note the distinction, here! While any given vector can be expressed in any given reference frame, derivatives are always
taken with respect to a specific frame.

frame. We let

p
= q
r

denote the angular velocity of the body with respect to the inertial frame, but expressed in the body frame.
Axis
Longitudinal (xB )
Lateral (y B )
Directional (z B )

Linear
Velocity
u
v
w

Aerodynamic
Force
X
Y
Z

Angular
Displacement

Angular
Velocity
p
q
r

Aerodynamic
Moment
L
M
N

The angular displacement variables , , and do not generally represent angles about the body-fixed
axes. These angles, referred to as the Euler angles, define a series of three rotations which transform
vectors from the inertial frame to the body frame, and vice versa. We will discuss the parameterization
of vehicle attitude in more detail later in the course. Until then, we will only consider simple motions in
which, for example, the pitch angle is truly a rotation about the lateral (y B ) axis.
The aerodynamic forces and moments are conventionally denoted in terms of dimensionless coefficients.
Let V = kV k be the airspeed, let S denote a reference area, and let l denote a reference length. Then one
writes

X =
Y

Z =

L =
M

1 2
CX
V
S
2

1 2
V
S
CY
2

1 2
V
S
CZ
2

1 2
Cl
V
Sl
2

1 2
V
Sl
Cm
2

1 2
Cn
V
Sl
2

Note: The reference area is typically chosen to be the wing planform area. Reference lengths may differ
depending on the context. For the pitch moment coefficient, for example, one typically takes l = c, the
mean aerodynamic chord. For the roll and yaw moment coefficients, one takes l = b, the wing span.
The use of upper-case subscripts in the force coefficients is consistent with the notation for aerodynamic
forces. The apparently inconsistent use of lower-case subscripts in the moment coefficients avoids a potential
ambiguity between roll moment coefficient and lift coefficient.
The dimensionless coefficients CX , CY , CZ , Cl , Cm , and Cn , are primarily functions of the Mach number
M = V /a (where a is the speed of sound), the Reynolds number Re = (V l)/ (where is the fluid density
and is the dynamic viscosity), and the aerodynamic angles and . Recall that the aerodynamic angles
are defined solely in terms of the body translational velocity:
= arctan

w
u

and
3

= arcsin

v
.
V

Figure 2: Aerodynamic angles.

These angles are shown in Figure 2. In normal flight conditions, the dimensionless coefficients depend
primarily on the variables and parameters mentioned above. They also depend, to a lesser extent, on the

body angular rate and the aerodynamic angle rates and .

An often-used simplifying assumption is that an aircraft is symmetric about the xB -z B plane. Motions
which are restricted to this plane of symmetry, such as wings-level climbs and loops, are called symmetric or
longitudinal motions. Motions out of the plane of symmetry, such as banked turns, are called asymmetric
or lateral-directional motions. Accordingly, the components of velocity and aerodynamic force and moment
are often decomposed into these two groups:
Longitudinal (or symmetric) quantities: u, w, q, X, Z, M
Lateral-directional (or asymmetric) quantities: v, p, r, Y , L, N
Static longitudinal stability. When discussing flight of atmospheric vehicles, the term stability refers
to a property of a special class of motion known as steady motion. For a vehicle in steady motion, all
components of body translational velocity V and body angular velocity are constant. A special case
of steady motion is equilibrium flight, in which the vehicle acceleration is zero. Note that these two
definitions are distinct. Steady, wings-level flight at constant altitude is equilibrium flight. A horizontal
turn at constant radius and velocity is not equilibrium flight; the constant yaw rate turn requires a constant
centripetal acceleration. Equilibrium flight is a steady motion for which = 0.
Stability (or instability) is a property corresponding to a steady motion. Loosely speaking, if a vehicle
which is slightly perturbed from a steady motion returns to that steady motion, the motion is stable. If
the vehicle motion diverges in response to a small perturbation, the motion is unstable.
The flight mechanics literature distinguishes between two finer notions of stability: static and dynamic
stability. The term static stability is somewhat of a misnomer because, by definition, stability (or instability) refers to a systems motion in response to a disturbance. Static stability refers to the initial tendency
of a vehicle, if displaced from a given steady motion, to return to that motion. No information about the
vehicles subsequent motion is required, only its initial tendency. Thus, one may determine static stability
without solving the differential equations that describe the airplanes motion.
Even though a given steady motion may be statically stable, the vehicle may diverge from the given motion
with time. To characterize this latter phenomenon, one must consider dynamic stability in which the
complete vehicle motion, not just its initial motion, is important. A given steady motion is dynamically
stable if, after a small displacement, the aircraft returns to the steady motion asymptotically in time.
Dynamic stability is stronger than static stability:
Dynamic stability

Static stability

but

Static stability

6 Dynamic stability

5
Reference value for steady motion
Statically and dynamically stable
Statically stable and dynamically unstable
Statically and dynamically unstable

Nondimensional Pitch Angle

4
5
6
Nondimensional Time

10

Figure 3: Sketches depicting static and dynamic pitch stability.

For example, a vehicles state may undergo diverging oscillations about a statically stable steady motion.
See Figure 3.
We will begin by investigating the conditions under which steady wings level flight is statically stable in
pitch. Consider a rigid aircraft with a reference frame fixed in the body at its center of gravity (CG) as
shown in Figure 1. If the xB -z B plane is a plane of symmetry, then the pitch rate equation is
q =

1
(Iz Ix )pr + Ixz (r2 p2 ) + M ,
Iy

(1)

where Ii is the moment of inertia about the ith coordinate axis and Ixz is a product of inertia. This is one
of three first order ODEs for the body angular rate; there are also equations for p and r. In addition, there
are three first order ODEs for the components u, v, and w of translational velocity. These six equations
describe the aircraft dynamics. Six more first order ODEs describe the variation of position and attitude
due to changes in velocity, that is, the aircraft kinematics.
Equation (1) is a nonlinear ordinary differential equation; the dependent variables p and r appear quadratically. We consider the case of steady, wings-level flight. In this case, p = r = 0. Also, v = 0. In the
absence of asymmetric disturbances, the lateral-directional variables remain zero; the motion is purely
longitudinal. The pitch rate equation becomes simply

1
1
1 2
q = M =
V
S
c Cm .
Iy
Iy
2
The dimensionless coefficient Cm depends on a number of variables and parameters. For pure longitudinal
flight, the primary influences are angle of attack , Reynolds number Re, and Mach number M. To a
lesser extent, Cm also depends on q and .
For now, we will ignore all dependencies save and write
Cm = Cm (). Formally expanding this expression in a Taylor series about zero angle of attack gives
Cm = Cm0 + Cm + h.o.t.

(2)

If we assume that (given in radians) remains fairly small, then we may neglect the higher order terms
in (2). The term Cm0 is the pitch moment coefficient at zero angle of attack and

Cm
Cm =

=0

is the slope of the pitch moment coefficient curve. Because of its critical role in determining both static
and dynamic stability, Cm is referred to as a stability derivative.

One should keep in mind that (2) is only an approximation. It ignores the dependency of Cm on Re and
M, as well as higher order terms in . Moreover, for fast, asymmetric maneuvers, Cm also depends on
p, q, and r and possibly other variables and parameters. For now, we consider only the case of
,
, ,
symmetric (wings-level) equilibrium flight
CL

Cm

1. Cm > 0

CL

eq

2. Cm = 0

eq

eq
3. Cm < 0

Figure 4: Pitch coefficient possibilities.

From a force balance in the z I direction, we see that the lift which the aircraft generates must perfectly
balance its weight for equilibrium flight. Suppose that we measure the angle of attack from the airplanes
zero-lift line, so that
CL = CL .
Given that CL > 0 and that lift must act in the upward direction to balance the airplanes weight, a
positive angle of attack is required at equilibrium, say = eq > 0.
ASIDE: If the angle of attack (call it ,
for now) is given in reference to some other line in the body so
that
CL = CL0 + CL
,
we may shift the origin by defining
=

0L
where

0L =

CL0
.
CL

Doing so gives
CL = CL ,
as we have assumed.
We know that must be positive and constant for equilibrium flight. Moreover, equilibrium flight requires
that the angular rate be zero, i.e., that q = 0. It follows that Cm () must be zero at the precise value of
> 0 for which lift balances weight. Referring to Figure 4, there are three possibilities to consider:
6

1. Cm0 < 0 and Cm > 0

2. Cm0 = 0 and Cm = 0
3. Cm0 > 0 and Cm < 0
In each case, the pitch coefficient is zero when = eq . What distinguishes the three cases is what happens
when the equilibrium is disturbed.
Case 1. If an impulsive pitch disturbance causes the angle of attack to decrease ( < eq ), then the pitch
moment coefficient becomes negative. This results in a nose-down pitching moment which drives the angle
of attack even lower. Alternatively, if a pitch disturbance causes the angle of attack to increase ( > eq ),
then the pitch moment coefficient becomes positive. This results in a nose-up pitching moment which
drives the angle of attack even higher. Thus, if Cmcg > 0, steady wings-level flight is statically unstable.
Case 2. In this case, variations in the angle of attack have no effect on the pitch moment coefficient. Thus
no additional moment is developed which would either drive the aircraft away from the equilibrium motion
or cause equilibrium to be restored. If Cmcg = 0 (with Cmcg0 = 0), steady wings-level flight is called
neutrally stable.
Case 3. If an impulsive pitch disturbance causes the angle of attack to decrease, then the pitch moment
coefficient becomes positive, resulting in a nose-up pitching moment which drives the angle of attack back
up toward eq . Alternatively, if a pitch disturbance causes the angle of attack to become positive, then
the pitch moment coefficient becomes negative, resulting in a nose-down pitching moment which drives the
angle of attack back down toward eq . Thus, if Cmcg < 0 (with Cmcg0 > 0), steady wings-level flight is
statically stable.
Static pitch stability clearly requires that Cm () have a negative slope when it crosses the -axis at the
equilibrium angle of attack eq . Moreover:
Static longitudinal stability requires Cm < 0 and Cm0 > 0.

Lecture 2: Introduction to Static Longitudinal Stability

Transferring moments. Recall that in the previous lecture we began discussing static longitudinal
stability. We obtained requirements on the dimensionless pitch moment coefficient as a function of the
angle of attack . Specifically, we found that static longitudinal stability requires Cm < 0 and Cm0 > 0.
Before we discuss the various aircraft components and their contributions to Cm , we should review the
basic notion of equivalent representations of forces and moments.
z

z
F
xa

(M - Fxa)

Figure 1: Equivalent force and moment diagrams.

Consider the planar rigid body shown on the left in Figure 1. The body is subject to a force F acting at
the point xa and a moment M , which is a pure couple. For this system, we have
X
X
Fz = F
and
M O = M F xa .

One may easily transfer a set of forces and moments acting at a given point to any other point. For
example, one may transfer the force and moment above to the origin O, as shown at the right in Figure 1.
Lb

La

Ma

Da

Mb

Db
xb

xa
c

Figure 2: Equivalent force and moment diagram for a wing.

Now consider a rectangular wing. We assume that lift force, drag force, and aerodynamic moment are
known, as functions of angle of attack, at the point xa . (In keeping with aerodynamics convention, the
signed distance x is measured positive aft from the wing leading edge.)
Suppose we wish to transfer the forces and moment from the point xa to another point xb along the chord.
The forces are equal at either point:
Lb = La = L

and

Db = Da = D.

It remains to determine the moment Mb given Ma , La , and Da . First, compute the moment of the system
on the left about a particular point, say the leading edge:
Ml.e. = Ma L(xa cos ) D(xa sin ).
Next, compute the moment of the system on the right about the same point:
Ml.e. = Mb L(xb cos ) D(xb sin ).
1

Equating the two expressions for Ml.e. and solving for Mb gives
Mb = Ma + (L cos + D sin )(xb xa ).
Dividing through by

2 V

S
c gives
Cmb = Cma + (CL cos + CD sin )

xa
.
c

(Note: Since we are considering a rectangular wing, the mean aerodynamic chord c is simply the constant
chord length c.) Define the nondimensional distances
ha =

xa
c

and

hb =

xb
.
c

For small angles of attack1 , we have

Cmb

= Cma + (CL cos + CD sin ) (hb ha )

CD
(hb ha ) .
Cma + CL 1 +
CL

D
For a well-designed wing operating below stall , C
CL 1. And since we have already assumed that
D
is small, the product C
CL may be neglected. We therefore have the following approximate equation for
transferring an aerodynamic moment between points on a wing:

Cmb Cma + CL (hb ha ) .

(1)

Aerodynamic reference points. A common reference point for the aerodynamic forces and the pitch
moment on a wing is the aerodynamic center. The aerodynamic center is that point about which the
pitching moment does not vary with angle of attack. To find this point, note that by definition
Cmac
= 0.

Thus, if one knows CL and Cma , about some point ha , as functions of (from wind tunnel tests, for
example), one may obtain Cmac through the following procedure:
1. Let hb in equation (1) denote the aerodynamic center.
2. Set the derivative of equation (1) with respect to equal to zero:
0=

Cmac
Cma
CL
=
+
(hac ha ) .

Notice that if CL and Cm are linear in (and we will generally assume that they are), all terms in
the equation above are constants.
3. Solve for the location of the aerodynamic center:

CL 1 Cma
.
hac = ha

4. Substitute hac back into equation (1) to obtain Cmac .

1

The error in this approximation is less than 5% for ||

12

radians 15 .

(2)

Given a set of wind tunnel data, there is often a simpler approach to find Cmac . If = 0L , that is, the
angle of attack corresponding to zero lift, then CL is zero in equation (1) and we have
Cmb = Cma = Cmac .
Thus, the zero-lift pitching moment has the same value as the (constant) moment about the aerodynamic
center:
Cmac = Cm0L .
Letting xa = xac , we may re-write the moment transfer formula (1) as

Cmb

Cmac + CL (hb hac )

= Cm0L + CL (hb hac )

Aerodynamic data for wings are typically referenced to the wing aerodynamic center or some other wingrelated reference point. When writing the equations of motion for an entire aircraft, however, it is most
convenient to sum moments about the aircraft center of gravity. Thus, a typical application of the formula
above will be to transfer the wing aerodynamic moment to the aircraft center of gravity.
To this point, we have only discussed the aerodynamic center for a rectangular wing. For a more general
wing, one introduces the notion of mean aerodynamic center x
ac . Appendix C in [1] presents techniques for
determining (or approximating) this point, as well as the mean aerodynamic chord c, for wings of general
shape. In subsonic flight, the aerodynamic center is located roughly one-quarter chord aft of the wings
leading edge. In supersonic flight, the aerodynamic center shifts aftward to roughly the half-chord point.
Another reference point which is sometimes important is the point at which the moment generated by the
wing vanishes entirely. The center of pressure is the point about which the moment due to the aerodynamic
force generated by the wing (i.e., the vector sum of lift and drag) precisely balances the pure aerodynamic
couple generated by the wing. To find the center of pressure, we solve
0 = Cmcp = Cm0L + CL (hcp hac )
to obtain

Cm0L
.
CL
Note that the center of pressure varies with because CL varies with . For this reason, the center of
pressure is generally a less useful reference point in aircraft dynamic modeling.
hcp = hac

Conditions for Static Longitudinal Stability. Lets return now to the problem of static longitudinal
stability. The two requirements we obtained are that the pitch moment coefficient Cm about the center
of gravity (CG) must have a negative slope and be positive at the zero lift angle of attack 0L . The
first condition ensures that a restoring moment is generated in response to small perturbations from eq .
The second condition ensures that an equilibrium angle of attack exists for which the wing generates the
positive lift necessary to balance the airplanes weight.
Given xac and Cmac , we may write
Cmcg = Cmac + CL (hcg hac ) .
The first condition for static longitudinal stability is that Cmcg < 0, where
Cmcg =

Cmcg

Cmac
CL
+
(hcg hac )

= CL (hcg hac ) .
=

Since CL > 0 and x is measured positive aft of the leading edge, this condition says that the center of
gravity must be forward of the aerodynamic center.
Now consider the second condition for static longitudinal stability, that Cmcg |0L > 0. This condition
ensures that the pitch coefficient curve passes through zero at an angle of attack eq for which CL (eq ) is
positive. Thus, positive lift will be generated when the pitch moment is zero. Assuming that the dynamic
pressure is appropriate, the aircrafts weight will be perfectly balanced by the lift that it generates. We
can express this condition as a condition on Cmac as follows:

0 < Cmcg
0L

= [Cmac + CL (hcg hac )]0L

= Cmac .

The existence of a statically stable, balanced flight condition requires

Cmac = Cm0L > 0 and Cmcg < 0

Cmac = Cm0L > 0 and (hcg hac ) < 0.

or, equivalently,

Cm

CL

Cm , C mac
0L

CL

eq

0L

eq

eq

CL

Cm
Cm , C mac
0L

CL

eq

eq

eq

Figure 3: Generic lift and pitch moment coefficient curves. The bottom and top graphs are equivalent
except that, in the lower graphs, is measured from the zero-lift line while.
Shown in Figure 3 are representative lift and pitch moment coefficient curves, where Cm = Cmcg . At the
top, CL and Cm are plotted versus
where
is not measured from the zero-lift line. Below these plots,
CL and Cm are plotted versus , which is measured from the zero-lift line (meaning 0L = 0). Note that
the the pitch moment coefficient curve represents a statically stable wing for which the balanced angle of
attack corresponds to positive lift. That is, Cmac = Cm0L > 0 and Cmcg < 0.
Example. (Courtesy of Dr. F. Lutze) Suppose lift force and pitch moment data have been obtained for
a rectangular flying wing. Moment data are referred to the one-third chord point x = 3c where x is
measured aft from the leading edge. The data are given in the table below.
4


(deg)

CL

Cm1/3

0.5
3.0
5.5
8.0

0.2
0.4
0.6
0.8

-0.02
0.00
0.02
0.04

Clearly CL and Cm1/3 vary linearly over this range of

because each 2.5 increment in
gives an equal
increment in CL and Cm1/3 . A least squares fit (though unnecessary for these data!) gives
CL

Cm(1/3)

CL
= 0.08 deg1 = 4.6 rad1

Cm1/3
= 0.008 deg1 = 0.46 rad1

We have
CL = CL0 + CL

= 0.16 + 0.08
deg
= 0.16 + 4.6
rad .
and
Cm1/3

= Cm(1/3)0 + Cm(1/3)

= 0.024 + 0.008
deg
= 0.024 + 0.46
rad .

Let us compute the aerodynamic center of the wing from these data. Equation (2) gives
hac

CL
= ha

1 0.008
=

3
0.08
0.23

Cma

which says that the aerodynamic center is at roughly 23% chord. Recall that Cmac = Cm0L , and note from
the tabulated data that
0L = 2 . Substituting into the above equation for Cm1/3 gives
Cmac = Cm1/3 |0L = 0.04
Next, we investigate static stability. Notice from the data that the slope of the Cm1/3 curve is positive.
If the CG were located at x = 3c , then the flying wing would be statically unstable because small pitch
disturbances would drive the state away from the balanced flight condition. The critical CG location at
which the slope of the Cm curve becomes zero is the aerodynamic center, which we have computed. Recall
that a necessary condition for a flying wing of this design to be statically stable is that the CG be located
forward of the aerodynamic center, i.e.,
(hcg hac ) < 0,
for only then do we have

Cmcg
< 0.

This condition alone is not sufficient for static stability, however. The aircraft must also be capable of
generating positive lift at the balanced flight condition, in order to balance its weight. Thus, we also require
Cm0L > 0. (See Figure 3.) In our example, Cm0L = Cmac = 0.04; this flying wing is statically unstable
regardless of the CG location. If the CG is forward of the aerodynamic center, so that Cm < 0, then it
can not be balanced at a positive angle of attack (i.e., steady, wings-level flight is not and equilibrium). If
the CG is aft of the aerodynamic center, there is an equilibrium corresponding to steady, wings-level flight
however, because Cm > 0 in this case, the equilibrium is longitudinally unstable.

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.

Lecture 3: Component Effects on Static Pitch Moment

Pitch moment contribution from the wing. Suppose that the aerodynamic characteristics of a
Reference line
k
j
i

xcg

Lw

z cg
w
xac

iw

Macw
Dw

Figure 1: Longitudinal force and moment due to the wing.

particular airfoil on a particular aircraft are given with respect to the mean aerodynamic center. As in
Figure 1, we replace the wing with a line denoting its mean aerodynamic chord of length c. This line is
drawn at angle of incidence iw with respect to a fuselage reference line which, we will assume, passes
through the mean aerodynamic center. In longitudinal (wings-level) flight, the airflow about the wing
generates two components of force and one component of moment:
Lw = Lw ( sin i + cos k)
D w = Dw (cos i + sin k)
M acw

= Macw j.

Keep in mind that Macw remains constant as varies. Choose the reference point (the origin of a bodyfixed reference frame) to be some point on the reference line, as shown in Figure 1. The vector from the
aircraft center of gravity to the aerodynamic center of the wing is
xac/cg = xac xcg

= (xac i) (xcg i + zcg k)

= (xac xcg )i zcg k

Summing the moments due to the wing about the aircraft center of gravity, we obtain
M cgw

= M acw + xac/cg Lw + xac/cg D w

= [Macw + Lw ((xac xcg ) cos + zcg sin ) + Dw ((xac xcg ) sin zcg cos )] j.
c, we obtain
Normalizing by 12 V 2 S
i
h
i
h
zcg
zcg
sin + CDw (h hnw ) sin
cos ,
Cmcgw = Cmacw + CLw (h hnw ) cos +
c
c
where h = xcg /
c and hnw = xac /
c.1

1
The n in the term hnw stands for neutral point, another term for aerodynamic center. The term arises from the pivotal
role that the aerodynamic center plays in determining static longitudinal stability.

In normal operation, 1 radian. Thus, in normal operation,

zcg
zcg
CDw
Cmcgw Cmacw + CLw (h hnw ) +
+
(h hnw )
.
c
CLw
c
C

w
For a well-designed aircraft, CD
1. Thus, the contribution of drag on the wing to pitch moment is
Lw
often negligible. Typically, zcg /
c 1, as well. Thus, for most aircraft in normal operation, we may write

Cmcgw

Noting that w = + iw , we have

Cmacw + CLw (h hnw )

= Cmacw + CL0w + CLw w (h hnw )

Cmcgw = Cm0w + Cmw

where

Cm0w = Cmacw + CL0w + CLw iw (h hnw )

and

Cmw = CLw (h hnw ) .

Wing (with positive camber)

Horizontal Tail

A) Conventional arrangement

Canard

Wing (with positive camber)

B) Forward tail arrangement

Figure 2: Conventional and forward tail arrangement.

Note that Cm0w 6= Cmacw unless CL0w + CLw iw = 0, that is, unless the fuselage reference line coincides
with the zero-lift line of the wing. While this is not generally the case, let us assume that it is true for
the sake of the following discussion. (Note: When they discuss the wings contribution to pitch moment,
Etkin and Reid do assume that the fuselage reference line coincides with the zero-lift line of the wing.) As
the CG moves forward, Cmw becomes more negative but Cm0w = Cmacw remains constant. Treating the
wing as the aircraft, recall that the two conditions for static pitch stability are:
Cmw < 0

and

Cm0w > 0.

The former condition will be satisfied provided the CG is forward of the wings aerodynamic center. The
latter condition can be satisfied by an airfoil with negative camber, however wings with negative camber
are less efficient at producing lift and can introduce problems with dynamic stability. In fact, most aircraft
use airfoils with positive camber which typically results in Cm0w < 0. Thus, most conventionally used
2

airfoils can not provide balanced flight without the addition of a horizontal tail or canard. See Figure 2,
which is adapted from [2].
Pitch moment contribution from the fuselage. According to Etkin and Reid, the typical effect of
the fuselage and nacelles on longitudinal static stability is
a shift forward of the mean aerodynamic center,
an increase in the lift-curve slope, and
a negative increment in Cmac .
Although the flow around the wings, fuselage, and nacelles is generally quite complicated, there are techniques for estimating the contributions of these components to the pitch moment acting on the aircraft.
See [3] for some simple techniques. See [1] for more sophisticated techniques.
Typically, the combined effects of the body and wing are incorporated into a single wing-body contribution to the pitch moment. Let wb denote the angle of attack measured from the zero-lift line of the
combined wing and body. Then we may write
Cmcgwb

= Cmacwb + Cmwb wb
= Cmacwb + (h hnwb )CLwb wb ,

where hnwb is the neutral point of the combined wing and body, that is, the point about which the
aerodynamic moment acting on the combined wing and body does not vary with angle of attack. The
value of this constant moment is given by the coefficient Cmacwb .
Pitch moment contribution from the horizontal tail. The conventional way to address the problem
of making Cm0w > 0 for an aircraft with a positively cambered wing is to include a horizontal tail, either
aft or forward of the wing at an appropriate angle of incidence. Here, we consider the more conventional
case of an aft tail. The case of a canard can be treated similarly.
The effect of an aft tail on the aircrafts longitudinal static stability can be treated very similarly to the
effect of the wing, with at least two important differences. First, and more importantly, the downwash
behind the wing lowers the effective angle of attack seen by the horizontal tail. Second, the local dynamic
pressure at the wing may be reduced (for example, in the wake of the wing) or increased (for example, in
the flow field behind a propeller or a jet engine). We will consider the former effect explicitly. The latter
effect, which we will not discuss in depth, may be incorporated into the lift coefficient for the horizontal
tail.
Referring to Figure 3, the angle of attack at the tail is
t = wb (w ) it ,
where (w ) is the downwash angle, a function of w , and where it is the incidence angle of the tail.2
The lift and drag force generated by the tail are defined relative to the local velocity V 0 rather than the
aircraft velocity V . To combine these forces with the total aircraft lift and drag, they must be referred
to the aircraft velocity V . The angle between V 0 and V is due to the downwash aft of the main wing.
This downwash is often described as being induced by the system of bound and trailing vortices which
describe the circulation about a wing. Note that, forward of the wing, this system of vortices induces an
upwash. (See Figure 4.) Thus, while the effective angle of attack at an aft tail is smaller than the wing
angle of attack, the opposite is true for a canard.
2

Per Etkin and Reids notation, the tail incidence angle is defined to be positive with the leading edge down.

Fuselage reference line

k

Vair

Wing Downwash

xcg

z cg

V air
'

Wing/body neutral point

Lt

wb
V'
xac t

Mac t

it

Dt
z ac t

Figure 3: Longitudinal force and moment due to the tail.

Figure 4: Horseshoe vortex representing circulation due to a wing.

The combined lift due to the wing-body and tail is
L = Lwb + Lt cos Dt sin
As

Dt
Lt

and are typically small, we may approximate the total lift as

St
1 2
L = Lwb + Lt = CLw + CLt
V
S
S
2

where St is the tail area. Now, the dynamic pressure at the tail is actually somewhat different than 21 V 2
because the local velocity at the tail is different. The lift coefficient for the tail CLt is therefore determined
not only by its own geometry, but also by the effect of downwash. Some textbooks include a tail efficiency
factor to account for the change in dynamic pressure at the tail. Etkin and Reid incorporate the effect
directly into the lift coefficient CLt .
The pitch moment about the aircraft CG due to the tail may be computed in a similar way to the wings
4

contribution. First, define

Lt = Lt ( sin(wb )i + cos(wb )k)

D t = Dt (cos(wb )i + sin(wb )k)

M act

= Mact j.

The vector from the aircraft center of gravity to the aerodynamic center of the tail is
xact /cg = xact xcg

= (xact i + zact k) (xcg i + zcg k)

= (xact xcg )i + (zact zcg )k

= l t i + zt k
where
lt = xact xcg

and

zt = zact zcg

denote the signed horizontal and vertical distance from the CG to the mean aerodynamic center of the
tail.3 Summing the moments due to the tail about the aircraft center of gravity, we obtain
X
M cgt = Mcgt j
= M act + xact /cg Lt + xact /cg D t
h
= Mact + Lt (lt cos(wb ) zt sin(wb ))

+Dt (lt sin(wb ) + zt cos(wb )) j.

Now, since
Lt = CLt

1 2
V
St
2

and

Dt = CDt

1 2
V
St ,
2

we obtain
Mcgt

"
1 2
V
St Cmact ct + CLt (lt cos(wb ) zt sin(wb ))
2
#

+CDt (lt sin(wb ) + zt cos(wb )) .

Normalizing the tail moment equation, we obtain

"

ct
lt
zt
St
Cmt =
Cmact + CLt cos(wb ) sin(wb )
S
c
c
c

#
lt
zt
+CDt sin(wb ) + cos(wb )
c
c
"

St ct
St lt
zt
Cmact +
CLt
=
cos(wb ) sin(wb )
S
c
S
c
lt

#
zt
CDt
sin(wb ) + cos(wb )
+
CLt
lt
3

The definition lt is consistent with that given by Etkin and Reid. The definition of zt is not; there is a sign discrepancy
due to the authors inconsistent use of the coordinate z.

Now, in normal operation, (wb ) and CDt /CLt are small. Also, Cmact cct is typically small in comparison
to the other terms; if the tail airfoil is symmetric, for example, then Cmact = 0. Terms involving zt are
typically ignored, as well. It follows that, in estimating the pitch moment contribution of the tail, it is
often reasonable to consider only the contribution of the tails lift acting through the moment arm lt . We
therefore obtain

St lt
CLt
Cmt
S
c
The term in parentheses plays such an important role in design that we give it a special name and symbol.
The horizontal tail volume ratio
St lt

VH =
S
c
is an indicator of the horizontal tails effectiveness at generating a pitch moment. For example, a horizontal
tail with a large planform area St acting through a large moment arm lt will be very effective at generating
a restoring pitch moment in response to pitch disturbances.
Etkin and Reid note that the distance lt from the aircraft CG to the tail aerodynamic center varies with the
aircrafts mass distribution. For example, the CG location may change with different passenger or cargo
loading arrangements or as fuel is spent. Because lt varies, the horizontal tail volume ratio also varies. In
order to provide an absolute reference, define the horizontal distance lt from the wing/body aerodynamic
center to the tail aerodynamic center:
lt =: xac xac
t
wb
=
=

lt (xacwb xcg )

lt + (xcg xacwb ) .

We may now define

H :=

V
=

lt St
cS

VH +

St
(h hnwb ) .
S

In the end, we find that the contribution of the tail to the pitch moment about the CG is

Cmt

= V
H at t

St

=
VH
(h hnwb ) CLt t .
S

References
[1] USAF Stability and Control Datcom. Flight Control Division, Air Force Flight Dynamics Laboratory,
Wright Patterson Air Force Base, Fairborn, OH.
[2] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.
[3] R. C. Nelson. Flight Stability and Automatic Control. WCB McGraw-Hill, New York, NY, second
edition, 1998.

Lecture 4: Total Pitch Moment Coefficient

Recall that in the previous lecture we found
Cmt V
H CLt t
t lt
where
VH = SS
c is the horizontal tail volume ratio (with lt representing the longitudinal distance from the
center of gravity to the tail aerodynamic center). By definition,

t = wb (wb ) it ,
where the downwash angle (wb ) depends on the lift being generated by the wing and body. For reasonable
angles of attack, one may approximate (wb ) by the first two terms of its Taylor series expansion about
wb = 0:
d
(1)
= |0 +
wb
dwb 0
To simplify notation, and to remain consistent with the textbook, we will write
d
d
0 = |0
and
=
.
d
dwb 0
d
The coefficients 0 and d
may be estimated as described in Appendix B of [1]. Substituting for from (1)
in the definition of t , we find that

d
t = 1
wb 0 it .
(2)
d

The contribution to Cm from the horizontal tail may therefore be written

Cmt

= V
H CLt t
= Cm0t + Cmt wb

Cm0t
Cmt

where

d
=
VH CLt 0 +
iw + it
d

d
.
= V
H CLt 1
d

Consider the term Cm0t . Recall that we require a positive pitch moment at zero lift so that the aircraft
can be balanced at a positive angle of attack (i.e., so that the aircraft can generate appropriate lift in
equilibrium flight). Also recall that the wings contribution to the zero-lift pitch moment is negative for
airfoils with positive camber (the most commonly used type). An aft tails contribution to Cm0 will be
positive provided
d
it > 0
iw .
d
By making it large enough, one may increase the positive contribution to Cm0 . (Recall that it is positive
when the leading edge of the horizontal tail is rotated downward.) The effectiveness of the horizontal tails
contribution may be increased by increasing the moment arm lt or by increasing the tail surface area St .
d
< 1, it is clear that an aft tail will
Considering next the term Cmt , and noting that typically 0 < d
provide a stabilizing (negative) increment in Cm , as well. This increment can be increased by increasing
the horizontal tail volume ratio and also by increasing the lift curve slope CLt .

Similar analysis can be performed for a forward tail, or canard, with a few notable differences. First, the
local airflow forward of the wing experiences an upwash rather than a downwash; this tends to increase
the local angle of attack seen at the canard. Second, the signed horizontal distance lt from the CG to the
canard will likely be negative. As a result, a canards contribution to Cm reduces the airplanes static
longitudinal stability.
Propulsion effects. Formally, we may approximate the propulsion systems contribution to pitch moment
as
Cmp = Cm0p + Cmp wb ,
however the task of determining the coefficients Cm0p and Cmp can be complicated. The simplest estimate
involves taking the moment of the propulsion force about the CG, however there are other important effects
as well. For example, a propeller or a jet engine at angle of attack experiences a force component normal
to the thrust generated.
Total pitching moment. We may now write the total pitching moment coefficient for an aircraft in
wings-level equilibrium flight:
Cm = Cmwb + Cmt + Cmp .
(When the subscript cg is omitted, it should be assumed that the moment is taken about the aircraft
center of gravity.) Written more explicitly,

Cm = Cmacwb + CLwb (h hnwb )wb

VH CLt t + Cmp .

Substituting for t from (2), we may express the pitch moment coefficient as

Cm = Cm0 + Cmwb wb
where
Cm0

= Cmacwb + Cm0t + Cm0p

d
iw + it + Cm0p
= Cmacwb +
VH at 0 +
d
and

Cmwb

= Cmwb + Cmt + Cmp

= CLwb (h hnwb )
VH CLt

d
1
+ Cmp .
d

Note, once again, that Etkin and Reid assume that iw = 0. From this point on, we will make the same
assumption.
The zero-lift line for the entire aircraft is not the same, in general, as the zero-lift line for the wing and
body. This is because the horizontal tail modifies the airplanes lift characteristics. Indeed, the entire
point of the tail incidence angle it is to generate lift in order to provide a nose-up moment when the wing
and body are not creating lift. A side effect is a shift in the zero-lift line of the aircraft. Note that the
additional lift generated by the tail is downward for an aft tail and upward for a canard. An advantage of
a canard is that it not only provides a nose-up moment, but it also provides lift in the desired sense. On
the other hand, canards typically give a positive increment in Cm which tends to destabilize steady wings
level flight.

It is convenient to define a new angle of attack which is measured from the zero-lift line of the entire
aircraft such that
CL () = CL .
(3)
Now, the lift generated by the entire aircraft is simply the sum of the lift generated by the wing and body
and that generated by the tail:
L = Lwb + Lt .
Normalizing by ( 21 V 2 )S, we find that
CL = CLwb +

St
CLt .
S

Since we are assuming that wb is measured from the wing/body zero-lift line and that t is measured
from the tail zero-lift line, we have
St
CLt t
S

d
St
0 it
= CLwb wb + CLt wb 1
S
d

!
St CLt
d
St
= CLwb 1 +
1
wb CLt (0 + it ) .
S CLwb
d
S

CL = CLwb wb +

(Recall that any effect due to a modified dynamic pressure at the tail is included in the tail lift coefficient.)
Thus, we have
CL (wb ) = CL0 + CL wb
(4)
where
CL0
and

St
= CLt (0 + it )
S

CL = CLwb

St CLt
1+
S CLwb

!
d
1
.
d

Comparing equations (3) and (4), we see that

= wb

St CLt
(0 + it ) .
S CL

Note that downwash and positive tail incidence (for an aft tail) reduce the effective angle of attack. This
makes sense, of course the lift generated by the tail due to these effects is opposite the lift generated by
the wing.
We may now restate our previous results concerning the total pitch moment in terms of the new angle of
attack . First, recall that
Cm = Cmwb + Cmt + Cmp

=
Cmacwb + CLwb (h hnwb )wb
VH CLt t + Cmp

St

(h hnwb ) CLt t + Cmp

=
Cmacwb + CLwb (h hnwb )wb
VH
S

St
H CL t + Cmp
=
Cmacwb + CLwb wb + CLt t (h hnwb )
V
t
S
H CL t + Cmp .
= Cmacwb + CL (h hnwb )
V
t
3

Noting also that

t

d
=
1
wb 0 it
d

d
St CLt
=
1
+
(0 + it ) 0 it ,
d
S CL

we find that
Cm = Cmacwb

H CL
+ CL (h hnwb )
V
t

St CLt
d
+
(0 + it ) 0 it + Cmp .
1
d
S CL

As before, we may express the pitch moment coefficient as

Cm = Cm0 + Cm

Cm0

H CL
= Cmacwb +
V
t

where

St CLt
d
(0 + it ) 1
1
+ Cm0p
S CL
d
and

Cm

H CL
V
= CL (h hnwb )
t

d
1
+ Cmp .
d

H rather than the slightly varying

The alternative form above, which involves the truly constant term
V
term
VH , is stated for completeness. In the future, we will state all expressions involving the horizontal
tail volume ratio in terms of the more conventional term
VH .
Neutral Point. Because the location of the CG of an aircraft varies in flight as fuel is spent, passengers
move about, etc, it is of interest to know the absolute limit on the range of CG motion for Cm to be
negative.
Definition. The neutral point is the CG location h = hn at which Cm = 0.
Essentially, the neutral point is the aerodynamic center for the complete airplane. Because, by definition,
the slope of the Cm curve is zero when h = hn , the aircraft is neutrally stable in pitch. (Recall that h
denotes the location of the CG.) The trim condition will not be restored in response to a small deviation
from the equilibrium pitch angle. For h > xn , i.e., when the CG is aft of the neutral point, the aircraft will
diverge from the equilibrium in response to a small pitch disturbance. Thus, h = hn is a critical condition
which must be avoided.
To find hn , we set Cm as given above equal to zero and solve for h = hn . Doing so gives

CLt
d

VH 1
Cmp .
hn = hnwb +
CL
d

(5)

It is common to replace the wing/body neutral point with the aircraft neutral point by solving (5) for hnwb
and replacing that term wherever it appears. For example, substituting hnwb into Cm above gives
Cm = CL (h hn ) .
Note that we may write the pitch moment coefficient taken about the CG as
Cm = Cmn + CL (h hn )
4

(6)

where
Cmn = Cm0L = constant.
Here, Cm0L is the zero-lift pitch moment coefficient for the entire airplane, not just for the wing.
Definition The static margin for an aircraft is
Kn = hn h.
The static margin is a margin of safety for static longitudinal stability. If it is sufficiently positive, then
minor modeling discrepancies and computational errors will not affect the static longitudinal stability of
wings level equilibrium flight.
Remark #1: The term Kn is often referred to as the stick-fixed static margin, indicating that it is the
static margin with the elevator locked in place. If the elevator is able to float freely under the influence
of aerodynamic forces, it will tend to align itself with the flow and will generate no local lift force. The
horizontal tail will therefore be less effective at providing stability. In these cases, one may also define a
stick-free static margin, whose value will be somewhat smaller than the stick-fixed static margin. See the
supplemental notes on hinge moments and stick forces.
Remark #2: Occasionally, pitch moment data will be given versus lift coefficient rather than angle of
attack. In some sense, this makes the pitch stability question simpler because the zero-lift pitch moment
(which must be positive) is more obvious from the data. Differentiating (6) with respect to , we obtain
Cm = CL (h hn ) ,
as before. Thus,
hn = h

Cm
.
CL

If Cm is given in terms of CL , then we have Cm (CL ()). By the chain rule,

dCm dCL
d
Cm (CL ()) =
.
d
dCL d
Therefore,
dCm
Cm
=
dCL
CL
and we have
hn = h

dCm
.
dCL

This equation allows one to estimate the stick-fixed neutral point from Cm versus CL data for the aircraft.
(Note, however, the important discussion in Section 2.3 of [1]!)

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.

Lecture 5: Longitudinal Control

We next turn our attention to control of longitudinal motion, particularly control of the pitch attitude
. For wings-level, equilibrium flight at constant altitude, is equal to the pitch angle . Thus, we are
interested in controlling , and hence the lift generated by the aircraft.

Figure 1: Standards for stick and control surface deflections [2].

For conventional aircraft, there are three primary control surfaces which provide moments about the three
coordinate axes. Figure 1 depicts the elevator, rudder, and ailerons, as well as a pair of canards, which
are somewhat less common. The figure also depicts the sign conventions for the various stick, pedal, and
actuator deflections.
Aerodynamic control surfaces are generally small lifting surfaces which generate small local forces. These
force act through a large moment arm to generate control moments about the aircraft center of gravity.
Table 1 shows the effect of positive control surface deflections on the aerodynamic moments. For example,
a positive elevator deflection (as indicated in Figure 1) results in a negative moment about the lateral axis
because Me < 0. Because these control surfaces are simply lifting surfaces, everything we know about
wings can be applied to our study of actuator effects.
Table 1: Effect of control surface deflections on the aerodynamic moments.
Actuator Angle to be
Symbol for
Moment Sensitivity
Name
Controlled Actuator Deflection to Actuator Deflection
Aileron

a
La > 0
Elevator

e
Me < 0
Rudder

r
Nr < 0

Two primary concerns in designing a control surface, such as an elevator, are

1. control effectiveness and
2. hinge moments and stick forces.
Control effectiveness relates to the ability of a given control surface to generate the necessary control
1

moments. Hinge moments and stick forces relate to the pilots ability to deflect the control surface as
necessary.
We previously assumed that the lift of the aircraft was a function only of its angle of attack . Now, we
recognize that deflections e of the elevator will also affect the lift. Assuming these deflections are small
(e 1 radian), we may write (for the entire aircraft)
CL = CL + CLe e.
The coefficient

CL
e
is typically positive, but small relative to CL . Thus, the effect of positive (downward) elevator deflections
is a small increase in the value of CL .
CLe =

The more important effect of the elevator is on the pitch moment. The aircraft moment coefficient is
Cm = Cm0 + Cm + Cme e
where

Cm
e
is referred to as the elevator control power or simply elevator power. The elevator power determines how
effective the elevator is at generating pitch control moments.
Cme =

Cm

trim

Figure 2: Effect of elevator deflections on Cm and trim .

As shown in Figure 2, the effect of elevator deflections is to shift the Cm versus curve up or down. (Which
direction the curve moves, depends on the sign of Cme .) Equivalently, the effect is to shift the trim angle
of attack trim right or left. Intuition suggests that negative elevator displacements, which correspond to
positive (backward) deflections of the stick, result in higher trim angles of attack. We would therefore
expect that Cme < 0 so that negative (upward) elevator deflections correspond to positive (nose-up) pitch
moments.
To verify this intuition about the elevator power, we examine the effect of the elevator more carefully. The
increase in lift provided by a positive elevator deflection is

1 2
V
S.
L = CL
2
2

Note that this increase is generated entirely by the horizontal tail, so

1 2
L = Lt = CLt
V
St .
2
We therefore find that
CL =
=

St
CL
S t

St CLt
e .
S
e

The coefficient

CLt
e
is termed the elevator effectiveness and is assumed to be constant for a given horizontal tail and elevator
assembly. The elevator effectiveness determines how effective the elevator is at modifying the lift generated
by the tail.1
ae =

If the elevator is a flap at the trailing edge of the horizontal stabilizer, as is typically the case, one may
write
CLt dt
ae =
t de
or

ae =

CLt

where the constant

dt
de
depends on the ratio of the elevator surface area to the complete tail lifting surface area. If this ratio of
areas is one, i.e., if the entire tail can be deflected as an elevator, then = 1. The parameter can be
determined as described in Appendix B.2 of [1].
=

Considering once again the primary purpose of the elevator as a pitch actuator, we see that the increment
in total pitch moment due to an elevator deflection is

1 2
M lt Lt = lt CLt
V
St
2
Here, as in previous discussions, we have neglected the small contributions due to
the change in drag on
the tail and the height of the tail relative to the fuselage. Normalizing by 21 V 2 S
c gives
Cm = ae
VH e.

Since Cm = Cme e, the elevator power is

Cme = ae
VH
1
Note: Elevator effectiveness is the tail lift coefficient sensitivity to elevator deflections while elevator power is the
moment coefficient sensitivity to elevator deflections.

Note that Cme is indeed negative for an aft tail. Thus, a negative elevator deflection (trailing edge upward)
yields a positive (nose-up) pitching moment.
Longitudinal trim conditions: e 6= 0. For wings-level, equilibrium flight, we need
the thrust to balance the drag (which we assume is the case)
the lift to balance the weight:
CLtrim

1 2
V
S=W
2

where
CLtrim = CL trim + CLe etrim ,
the pitching moment to vanish
0 = Cm0 + Cm trim + Cme etrim .
For an airplane of given geometry, the aerodynamic coefficients and stability derivatives are known. The
only unknowns are the angle of attack and elevator deflection. Rearranging these two linear algebraic
equations into matrix form, we have

CL CLe
trim
CLtrim
=
.
(1)
Cm Cme
etrim
Cm0
The solution to these equations is

trim
CL CLe
CLtrim
=
etrim
Cm Cme
Cm0

1
Cme CLe
CLtrim
=
Cm0
Det Cm CL

1
Cme CLtrim + CLe Cm0
=
Det Cm CLtrim CL Cm0
where Det represents the determinant of the matrix on the left hand side of (1):
Det = CL Cme CLe Cm .
The first term CL Cme is generally negative, as we have seen, while the second term CLe Cm is positive.
Because the first term typically dominates, however, the determinant Det is typically negative.
Recall that the trim lift coefficient is given by
CLtrim =

.
1
2
2 V S

For a given aircraft in constant altitude flight, lower speed requires a larger lift coefficient. To see the effect
of variations in CLtrim on the trim angle of attack and elevator deflection, we plot

CLe Cm0
Cme
CLtrim +
.
trim (CLtrim ) =
Det
Det
and
etrim (CLtrim ) =

Cm
Det
4

CLtrim

CL Cm0
.
Det

e trim

trim
CL Cm

Det

CL
CL Cm
e

Cm
0

trim

Cm

trim

Det

Det

Det

CL

Figure 3: Trim angle of attack and elevator deflection versus lift coefficient.
The important observation to make is that, as CLtrim increases (due to decreased speed, increased weight,
or increased altitude), the trim angle of attack increases and the trim elevator deflection decreases. Recall
that elevator deflections are defined to be positive when the trailing edge moves down; a decrease in e
means a trailing-edge-up increment resulting in a nose-up increment in pitch moment.
Maximum forward CG location. The pitch moment coefficient for the aircraft is
Cm = Cm0 + Cm + Cme e.
The purpose of the elevator is to provide a counter-moment to Cm0 + Cm so that the aircraft can be
trimmed at different angles of attack. Recall that, as the CG moves forward, the Cm curve becomes
steeper. This effect enhances stability of wings-level equilibrium flight, however it simultaneously reduces
the effectiveness of the elevator because greater and greater trim deflections are needed to obtain a given
change in the trim angle of attack; see Figure 4.

Cm

Cm

e min

e min

max

a
max

Figure 4: Illustration that elevator is less effective when Cm is more negative.

For a given aircraft with given physical limits on the elevator deflection angle, we may compute the
maximum forward CG location to be that CG location at which the minimum elevator deflection (i.e.,
5

elevator trailing edge at its upper limit) corresponds to the stall angle max . That is, we require that
etrim = emin

trim = max .

If the CG were any further forward, then the minimum elevator deflection would correspond to a trim
angle trim < max and the aircraft performance envelope would be limited by the actuator limits. This
could be problematic in landing, for example, where the low airspeed dictates a large lift coefficient and
therefore a large angle of attack.
From the equation for the trim elevator deflection, we require
(CL Cme CLe Cm )emin = Cm CLmax CL Cm0

(2)

where CLmax = CL (max ) is the lift coefficient at stall. In this equation, the only terms which depend on
the CG location are Cme and Cm :
Cm
Cme

= CL (h hn )
= ae
VH

Assuming that the elevator is well aft of the aircraft CG, the horizontal tail volume remains relatively
constant, so only Cm depends significantly on the CG location. Rearranging terms in (2), we obtain
(Cm0 + Cme emin ) CL = Cm (CLmax CLe emin )
Substituting
Cm = CL (h hn )
and solving for forward-most acceptable CG location gives

hmin = hn

Cm0 + Cme emin

.
CLmax CLe emin

By definition, hmin is the (nondimensional) CG location at which the trim elevator deflection emin yields
the maximum lift coefficient CLmax . If the CG lies aft of hmin , then the full range of lift possible lift
coefficients will be obtainable by the pilot.

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.
[2] R. C. Nelson. Flight Stability and Automatic Control. WCB McGraw-Hill, New York, NY, second
edition, 1998.

Lecture 6: Longitudinal Maneuvering Flight

Symmetric pull-up. Consider an aircraft in wings-level flight which is executing a steady pull-up, that
is, a pitch-up maneuver at constant pitch rate. As a simple, but representative case, we will consider
the situation where the airplane just passes through horizontal flight. The centripetal acceleration at the
instant shown in Figure 1 is a = az k where
az =

V2
Rpullup

Now, if the airplane continued around the curved path and performed a circular loop at constant speed, it
would have performed one full pitch rotation. The rate of pitch rotation is
q=

V
.
Rpullup

Therefore,
az = qV.
i

R pull-up = V/q

L
V

D
T
W

Figure 1: A wings-level pull-up.

Applying Newtons second law of motion gives
Fx = 0
Fz = maz

T D =0

W L = mqV

Because we are considering steady motion (i.e., flight at constant v and ), the aerodynamic moment must
be zero.
Define the load factor
L
.
W
In wings-level, equilibrium flight the load factor is one. Although it is somewhat of a misnomer, we say
that the pilot experiences one g of upward acceleration. In reality, the pilot is not accelerating at all.
Rather, he senses the seat below him preventing him from accelerating downward at one g. As far as the
pilots senses are concerned, this condition is equivalent to accelerating upward at one g in a zero-gravity
environment. For a pull-up maneuver, the load factor increases. If n = 2, for example, the lift being
generated is twice the weight of the aircraft which causes it to accelerate upward at 9.8 m/s2 ; the pilot
thus experiences two gs.
n :=

Dividing through by W = mg in the z-direction force balance equation gives

1
1 n = qV.
g
Solving for pitch rate in terms of velocity and load factor, we find for a symmetric pull-up:

q = (n 1)

g
.
V

Horizontal turn at constant radius. Now consider an aircraft undergoing a steady turn at constant
radius and constant altitude. Strictly speaking, this is not a purely longitudinal flight condition like the
symmetric pull-up. In fact, the process of switching from wings-level equilibrium flight to a horizontal turn
is fairly complicated, involving motion in all six degrees of freedom. However, the problem of maintaining
a steady horizontal turn once it has been achieved is largely a question of longitudinal control.

L
R turn = V/
i

Figure 2: A steady turn at constant rate and altitude.

As shown in Figure 2, the lift vector in a steady turn is deflected through the roll angle . The centripetal
acceleration at the instant shown in Figure 2 is a = ay j where

V2
V

ay =
=
V = V.
Rturn
Rturn
Recalling that the vertical component of lift must balance weight for constant altitude flight, Newtons
second law of motion gives

Fx = 0
Fy = may
Fz = 0

T D =0

L sin = mV

W L cos = 0

The general expression for pitch rate is

q = cos + cos sin
For a steady turn at constant
For a symmetric motions, such as a pull-up, = 0 and we find that q = .
radius, the pitch angle is constant so that = 0. Moreover, if the pitch angle remains relatively small,
we have
q sin .
2

Replacing in the y-direction force balance equation gives

L sin =

mV q
.
sin

Dividing by W = mg, substituting the load factor n =

n sin2 =
Solving for pitch rate then gives
q=

L
W

and rearranging gives

1
V q.
g

g
n sin2 .
V

Now, from the z-direction force balance equation,

cos =

1
n

sin2 = 1 cos2 = 1

1
.
n2

Thus, for a horizontal turn at constant altitude and radius, we have

1 g
q = n
n V
Effect of pitch rate on lift force and pitch moment. We would like to know how the two maneuvering
flight conditions discussed above relate to the elevator control authority. In particular, we would like to
have some measure of the elevator angle necessary to execute a given maneuver, that is, the elevator angle
per g for these two maneuvering flight conditions. To do this, we must first account for the effect of pitch
rate on the lift force and pitch moment.
Previously, we considered only equilibrium flight. We therefore assumed that lift and pitch moment depended only on and e. More generally, in longitudinal maneuvering flight, lift and pitch moment also
vary with pitch rate. We assume that the dependence is linear:
Cm = Cm0L + Cm + Cme e +
CL = CL + CLe e +

Cm
q
q

CL
q.
q

Notice that every stability derivative introduced thus far has been dimensionless. (Consider, for example,
Cm or Cme .) This is because angles, measured in radians, are dimensionless. On the other hand, angular
rate has units of radians per second. To keep things nondimensional, we define a dimensionless pitch rate
q :=
(Note the enigmatic factor of
the stability derivatives

1
2

q
V
c/2

c
q.
2V

which is an artifact of early literature on aeroelasticity.) We now define

CL
2V CL
=
q
c q
Cm
2V Cm
=
=
.
q
c q

CLq =
Cmq

To determine the value of Cmq we consider the incremental contribution of pitch rate to pitch moment due
to an increase in lift generated by the tail. This increase in lift is the effect of an increased angle of attack
at the tail due to pitch rotation about the center of gravity. Of course, the wing experiences a similar
change in angle of attack, but the moment arm from the wing aerodynamic center to the aircraft CG is
small compared with lt and the contribution to the total pitch moment is generally small. Typically, one
accounts for the small contribution Cmq from the wing and fuselage by including a small correction factor
in the contribution due to the tail.

q lt
V
lt

Figure 3: Effect of pitch rate on tail angle of attack.

Consider the sketch in Figure 3. The increment in tail angle of attack is
t =

qlt
V

The increment in the tail lift coefficient due to pitch rate is

qlt
CLt = CLt t = CLt
V
The increment in total lift coefficient is
1
Lt
2
2 V S

1
1 2
= 1 2 CLt
V St
2
2 V S
qlt St
= CLt
.
V S

CL =

Differentiating with respect to q and nondimensionalizing gives

2V CLt lt St
CLq =
c
V
S

St lt
= 2CLt
S
c
or

CLq = 2CLt
VH .
Turning now to the pitch moment coefficient, we note that the pitch moment due to the increment in tail
lift is

1 2
Mt = lt CLt
V St .
2
4

Nondimensionalizing gives
Cmt

lt St
CLt
cS

= V
H

qlt
CLt
V

Differentiating with respect to q and nondimensionalizing once again gives

2V CLt lt

VH
Cmtq =
c
V
lt
= 2CLt
VH
c
A factor k 1.1 is generally applied to account for wing-fuselage effects. Thus, we have
lt
Cmq = 2kCLt
VH .
c
Elevator increment and elevator angle per g for longitudinal maneuvers. Recall that, for equilibrium flight, the trim conditions were
CLtrim = CW = CL trim + CLe etrim

(1)

Cmtrim = 0

(2)

= Cm0 + Cm trim + Cme etrim .

Solving these two linear algebraic equations in two unknowns gives the angle of attack and elevator deflection for equilibrium flight. Now, however, we are considering the case of maneuvering flight. For both of
the cases considered (a pull-up and a horizontal turn), the angular rate is constant. The equations above
become
CLmaneuver = nCW = CL maneuver + CLe emaneuver + CLq qmaneuver

(3)

Cmmaneuver = 0

(4)

= Cm0 + Cm maneuver + Cme emaneuver + Cmq qmaneuver .

Subtracting equations (1) and (2) for wings-level, equilibrium flight from equations (3) and (4) for maneuvering flight leaves
(n 1)CW

= CL + CLe e + CLq q

0 = Cm + Cme e + Cmq q

where
= maneuver trim

and

e = emaneuver etrim .

Rearranging the equations above, we have

CL CLe

(n 1)CW CLq q
=
.
Cm Cme
e
Cmq q
The solution to these equations is

1
Cme

=
e
Det Cm

CLe
CL

(n 1)CW CLq q
Cmq q

where
Det = CL Cme CLe Cm ,
5

as before. (Recall that Det is normally negative.) The increment in elevator deflection for a given maneuver
is

1
e =
Cm (n 1)CW CLq q + CL Cmq q
Det
or

1
e =
Cm (n 1)CW + Cm CLq CL Cmq q
Det
Notice that e vanishes when n = 1 and q = 0, as it must.

Symmetric pull-ups. Recall that, for a pull-up maneuver,

q = (n 1)

g
.
V

Thus, we have
q =

q
c
2V

g
c
2V 2
g
c mS
= (n 1) 2
2V
mS

!
mg

cS
= (n 1) 1 2
4m
2 V S
= (n 1)

= (n 1)CW

cS
.
4m

Substituting into the equation for e gives the elevator increment for a pull-up maneuver

(n 1)CW
cS
e =
Cm + Cm CLq CL Cmq
.
Det
4m
Differentiating with respect to n gives the elevator angle per g for a pull-up maneuver :

e
CW
cS
=
Cm + Cm CLq CL Cmq
.
n
Det
4m
Using the expression above for elevator deflection per g, one may easily compute the stick force per g, a
critical parameter in sizing the pilot interface.
Horizontal turns. For a horizontal turn,
q=

1 g
n
.
n V

Thus, we have
q
c
q =
2V

1
g
c
=
n
n 2V 2

cS
=
n
CW
.
n
4m

Substituting into the equation for e gives the elevator angle required for a horizontal turn:
6

n + 1
cS
CW
Cm + Cm CLq CL Cmq
.
e = (n 1)
Det
n
4m
As above, one could differentiate with respect to n to determine the elevator angle per g for a horizontal
turn:

n + 1
CW
cS
e =
Cm + Cm CLq CL Cmq
.
Det
n
4m
Alternatively, given the necessary aircraft parameters and actuator limits, one could use the prior equation
to determine the turn of smallest radius (or largest load factor) which a given aircraft can execute.
In [1], the authors define the stick-fixed maneuver point hm as that CG location at which the elevator angle
per g (for a pull-up maneuver) is zero. Computing this number, one finds that the stick-fixed maneuver
point is aft of the stick-fixed neutral point hn , meaning that, for a statically stable aircraft, some nonzero
elevator angle will be required to execute a pull-up maneuver. In general, the elevator angle per g will be
proportional to hm h, which is referred to as the stick-fixed maneuver margin. The larger the maneuver
margin, the larger the elevator deflection that is required to pull up at a given acceleration.

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.

Lecture 7: Directional Stability & Control

With the exception of steady turning flight at constant radius, we have only considered wings level flight. In
this setting, the three lateral-directional velocities (v, p, and r) are initially zero and they remain zero under
the assumption that there are no lateral-directional disturbance forces or torques to perturb their values. In
reality, there are always disturbances. A major design objective, therefore, is to provide lateral-directional
stability through careful choice of available parameters (vertical tail size and location, wing dihedral and
sweep angle, etc.). Following the textbook, we will divide our discussion of lateral-directional stability into
two topics: directional (or weathercock) stability and roll stability. We first discuss directional stability.
V

u
v

Figure 1: Top view of an aircraft in sideslip.

Directional stability. Note that the assumption that v is zero implies that the sideslip angle = sin1 Vv
is zero. If a yaw disturbance occurs which changes the value of , then we would prefer that the airplane
naturally reject this disturbance by driving back toward zero. This is the purpose of the vertical
stabilizer.
A conventional vertical stabilizer is a vertical lifting surface which produces a lateral force in response to
its angle of attack, which is a function of the sideslip angle. This force acts through a moment arm to
produce a yaw moment about the airplanes center of gravity in response to sideslip. Let lvt denote the
longitudinal distance from the center of gravity to the aerodynamic center of the vertical tail. The yaw
moment due to sideslip is
N = Nwb Lvt lvt cos vt Dvt lvt sin vt
where vt is the angle from the longitudinal axis to the relative air velocity at the vertical tail. In [1], tail
lift is defined to be positive when the resulting side force on the aircraft is positive. Thus, the sign of vt
is opposite that of . The contribution Nwb of the wing-body to the yaw moment vanishes at zero sideslip
angle and typically makes a destabilizing contribution when 6= 0.

Making the usual assumptions that vt is small and that Dvt Lvt , and normalizing by 21 V 2 Sb, we
obtain
Cn = Cnwb + Cnvt
Cnwb
= Cnwb

(1)

1 2
1
CLvt
1 2
V
Svt lvt
2 vt
Sb
2 V

Vvt 2
CLvt

VV
V

where

VV =

Svt lvt
.
Sb
1

(2)

V
-

Nwb

Vvt
vt

lvt

Lvt
Dvt
Figure 2: Tail forces and yaw moment on an aircraft in sideslip.
The parameter
VV is referred to as the vertical tail volume ratio. In analogy with the horizontal tail volume
ratio VH , the vertical tail volume ratio governs the effectiveness of the vertical tail at providing directional
stability.
A symmetric vertical tail generates no lift when vt is zero, so
CLvt = CLvt vt .
The angle of attack vt is
vt = + ,

where the sidewash angle results from the asymmetry in the distribution of vorticity generated by the
wing. In [1], the sidewash angle is defined to be positive when the induced flow is in the positive y direction.
In simple terms, one may observe that, of the two trailing vortices for an airplane in negative sideslip (as
shown in Figure 2), the trailing vortex from the left wing tip will dominate the local flow at the tail. If
the vertical tail is mounted level with or above the wings, the induced flow at the vertical tail will have
a substantial component in the positive y direction. Thus, a negative sideslip angle will cause a positive
sidewash angle. Conversely, a positive sideslip condition ( > 0) will cause a negative sidewash angle. In
both cases, the effect of sidewash on a top-mounted vertical tail is to increase the magnitude of the angle
of attack vt , as indicated by the definition above. Note that the sidewash angle is a function of the
sideslip angle just as the downwash angle was a function of the wing angle of attack.
For an airplane for which the xz-plane is a plane of symmetry, Cn = 0 when = 0. Thus, for small ,
Cn = Cn
where
Cn =

Cn
.

Referring to (2), we have

Cn = Cnwb + Cnvt

Vvt 2
CLvt ( + ) .
= Cnwb
VV
V
2

Thus, we find
Cn = Cnwb + Cnvt
where
Cnvt =
VV

Vvt
V

CLvt

d
1
.
d

V vt
vt

Figure 3: Sketch of a positively deflected rudder.

Directional control. Now, suppose that a rudder flap is included on the vertical tail. Rudder deflections
r are defined to be positive when the trailing edge of the flap moves to the left. Recalling that tail lift
is defined to be positive when the resulting side force on the aircraft is positive, it is easy to see that a
positive rudder deflection will cause a positive increment in tail lift. Thus CLrvt > 0.
The total yaw moment is
Cn = Cn + Cnr r,
where Cnr is the rudder power. To determine the rudder power, we refer once again to (2)
Cn = Cnwb +
VV

Vvt
V

CLvt .

With the rudder included, we now have

CLvt = CLvt vt + CLrvt r
It follows by the previous argument that
Cn = Cnwb
VV

Vvt
V

CLvt vt + CLrvt r ,

from which we can easily see that

Cnr = V
V

Vvt
V

CLrvt .

Note that the rudder power is negative which is consistent with intuition given the sign convention for
rudder deflections. (By convention, a positive rudder deflection results in a negative increment in yaw
moment, causing the airplane to nose left.)
To summarize, the total yaw moment due to steady sideslip and rudder deflections is

Cn = Cn + Cnr r,

Cn

where

Vvt 2
d
= Cnwb +V
V
CLvt 1
V
d

Cnr

and

Vvt 2
CLrvt .
= V
V
V

Rudder considerations. The vertical tail alone tends to direct the nose of the aircraft into the wind,
thus driving the sideslip angle to zero. So why does one need a rudder and how does one decide how large
to make it? There are several considerations that require the use of a rudder and that dictate its size.
1. Adverse yaw. The outer (upper) wing of an airplane which is performing a banked turn generates
more lift and therefore more drag than the inner wing. This results in a yaw moment which is counter
to the desired turn direction. The rudder must counter this adverse yaw moment in order to execute
a coordinated turn.
2. Asymmetric power effects.
(a) Propeller effects.
For an airplane with one or more propellers rotating in one direction, there will be a steady
yaw torque generated which must be countered by the rudder. In particular, for an airplane
flying at low speed (and thus at high angle of attack), each propeller will generate a yaw
moment due to the differential angle of attack seen by the downward-moving and upwardmoving propeller blades. The downward moving blade sees a higher angle of attack and
therefore generates more thrust which creates a yaw moment.
The reaction torque from the propeller shaft onto the airplane causes a negative roll moment which must be countered by the ailerons (still to be discussed). A side effect of the
compensatory roll torque is a yaw moment due to the drag differential between the wings;
the phenomenon is essentially identical to the case of adverse yaw described above.
The swirling flow from the propeller causes additional sidewash at the tail which generates
a yaw moment.
(b) Engine out. A multi-engine aircraft flying at low speed must be able to maintain equilibrium
flight with a single engine failure. The rudder must be sizable enough to counter the yaw moment
resulting from this asymmetric flight condition.
3. Cross wind landings. In order to align an aircrafts longitudinal axis with the runway on final approach
when landing in a cross wind, it is necessary to maintain a steady sideslip angle for some period of
time. A nonzero rudder deflection is necessary to balance the resulting yaw moment due to sideslip.
Crude Rudder Sizing for Asymmetric Thrust. Suppose a two engine airplane has lost its left engine.
For the airplane to maintain equilibrium flight, it is necessary for the rudder to counter the yaw moment
generated by the thrust of the good engine. Assuming that thrust is parallel to longitudinal axis, the
moment about the airplane center of gravity is
NT = T yp ,
4

where yp is the lateral distance to the line of action of thrust. Thrust is often given in terms of a dimensionless coefficient CT :

1 2
T = CT
V
Sp ,
2
where Sp is an appropriate reference area for the engine (as specified by the manufacturer). The contribution of thrust to the yaw moment coefficient is
CnT

(T yp )
2 Sb
2 V
Sp yp
= CT
.
Sb
=

The total yaw moment coefficient is then

Cn = Cn + Cnr r + CnT .
For equilibrium flight, we require that Cn = 0. Moreover, to minimize drag the airplane should fly at
zero sideslip angle: = 0. Suppose that the rudder deflection satisfies (r )max r (r )max for some
maximum rudder deflection (r )max . Then, designing for the critical case of an engine failure, the rudder
should be sized such that
CnT
Cnr
.
(r )max
This is a relatively crude analysis. We will refine the condition once we have discussed roll stability and
control.

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.

Lecture 8: Roll Stability & Control

For both pitching motion and yawing motion, the primary effect of the empennage (horizontal and vertical
tail assembly) is to provide stabilizing moments which tend to keep the nose of the airplane pointing into
the wind. Thus, the notion of static stability is well-defined for these motions: the initial tendency of
the aircraft, when perturbed from wings level, equilibrium flight, is to return to that equilibrium. More
specifically, the horizontal tail provides a counter-pitching moment and the vertical tail provides a counteryawing moment. Because these restoring moments are proportional to the deflections in the aerodynamic
angles and , the effects are sometimes referred to as pitch stiffness and yaw stiffness. The more
effective the tail surfaces are at generating these moments, the more stiff the system is in the sense of a
mass-spring mechanism.
In the case of rolling motion, there is no feature of an airplane which provides static roll stability per se.
Etkin and Reid give the example of an airplane model pinned about its longitudinal axis in a wind tunnel
[1]. If the airplane is perturbed from wings level, no roll moment will develop to return it to that state.
While an aircraft has no first order roll stiffness, some roll stability can be provided by thoughtful design.
To see how, first consider an airplane which has suffered a positive roll disturbance so that > 0. Because
the lift vector does not remain vertical, it no longer balances the weight, which now has a component in
the direction of the right wing. The airplane begins to slide in the direction of the right wing so that
v > 0 which means that > 0. We will say that wings level, equilibrium flight is statically stable in roll
if the result of such a perturbation is a negative roll moment, i.e., one which tends to return the airplane
to the wings level condition.
Defining the roll moment coefficient
Cl =

L
,
QSb

we require that
Cl :=

Cl
<0

for roll stability. There are four primary factors which influence the value of Cl . The most important
of these is the wing dihedral angle. Other contributors are wing-body interaction, wing sweep, and the
vertical tail.
Dihedral effect. The principal means of satisfying the condition Cl < 0 is to angle the wings upward so
that the tip chord is higher than the root chord. The angle between the reference plane of the wing and
the aircraft xy-plane is called the dihedral angle and the stabilizing effect of this configuration is known as
the dihedral effect.

Vn

w
u

Figure 1: Illustration of wing dihedral.

Let Vn denote the component of velocity normal to the reference plane of the right wing, as shown in
1

Figure 1, which is adapted from [1]. A bit of thought will verify that
Vn = w cos + v sin w + v.
For

v
V

1 and

w
V

1, the local angle of attack of the right wing is

w + v
right arctan
+ .
u

Conversely, the component of velocity normal to the left wing is approximately w v and the local angle
of attack is

w v
.
left arctan
u

Thus, if > 0, then positive sideslip results in an increased angle of attack, and therefore increased lift, on
the right wing. Conversely, positive sideslip results in a decreased angle of attack, and therefore decreased
lift, on the left wing. The net result is a negative roll moment which tends to restore wings level flight.

Figure 2: Illustration of fuselage contribution to dihedral effect. (Viewed from the front)
Wing-body interaction. Mounting the wing above the fuselage enhances the dihedral effect while mounting the wing below the fuselage decreases this effect. To understand this, consider Figure 2, which depicts
the lateral airflow about a fuselage when the airplane slips sideways. For the high wing configuration, the
locally increased angle of attack on the right wing, and the decreased angle of attack on the left wing,
augment the dihedral effect. For the low wing configuration, the converse is true. Because the fuselage
contribution to roll stability is generally detrimental for a low-wing airplane, the dihedral angle must
generally be larger.
Wing sweep. The wing sweep angle affects roll stability, as well. Consider the case of positive sideslip
depicted in Figure 3, which might result from a positive roll disturbance. Any given chord line of the right
wing (e.g., the leading edge) experiences a relative increase in normal velocity, and thus an increase in the
local angle of attack. Conversely, any given chord line of the left wing experiences a relative decrease in
normal velocity and a decrease in angle of attack. Consequently, the right wing generates more lift than
the left and a negative (restoring) roll moment results.
The contribution Clwb to Cl due to the wing and body can be estimated as described in Section B.9
of [1]. The estimate accounts for dihedral angle, sweep angle, taper ratio, and twist, as well as the finite
aspect ratio and Mach number effects.
Vertical tail. The contribution to Cl due to the vertical tail is straight forward to estimate. The key
physical observation is that the lift force generated by a vertical tail in sideslip (the same force which exerts
the weathercock yaw moment that tends to decrease ) acts, in general, at some distance above or below
the longitudinal axis of the aircraft. Thus, a roll moment results as well.
2

Vt

Vt

V
Vn
V

Vn

Figure 3: Illustration of wing sweep effect on roll stability.

V

z
y
acvt

l vt

Lvt

z ac vt

Figure 4: Roll moment due to vertical tail.

Let zacvt denote the vertical distance (measured positive up) from the longitudinal axis to the vertical tail
aerodynamic center. For positive sideslip angles and for zacvt > 0 (i.e., a topside vertical tail), the roll
moment is negative, which means that the vertical tail increases roll stability. In fact, it is a simple matter
to show that, for an aft, topside vertical tail,
zac
Clvt = Cnvt vt < 0,
lvt
where, from the previous lecture, lvt > 0 and

Vvt 2
d
Cnvt =
VV
CLvt 1
> 0.
V
d
In the end, we have
Cl = Clwb + Clvt .
Roll Control. As with the other two control moment devices we have studied (namely the elevator and
the rudder), the roll control device acts by exerting a small force over a large moment arm. The ailerons
are a pair of flaps located symmetrically about the xz-plane at the trailing edge of the left and right wings.
These flaps are slaved to move in opposite directions, so only one parameter a is necessary to specify the
aileron displacement. For a > 0, the left aileron is deflected downward to increase the lift generated by
the left wing. The right aileron is simultaneously deflected upward to decrease the lift generated by the
right wing. The primary result is a positive roll moment.1
1

The convention stated here is opposite to that used by Etkin and Reid; see Figure 3.20 in the text. In fact, their convention

fuselage
centerline
b
2

y
c(y)
y1
y2

Figure 5: Sketch of the right wing with aileron.

To determine the aileron power, first note that the roll moment increment due to an aileron deflection is
Z b/2
L = 2
yLift0 (y)dy,
0

where y is the spanwise station along the right wing and where Lift0 (y) is the additional lift per unit
spanwise length at y due to a positive aileron deflection. The factor of two accounts for the effect of the left
aileron. The minus sign accounts for the fact that a positive aileron deflection decreases the lift generated
by the right wing. (By convention, Lift0 (y) for the right wing is negative for positive aileron deflections.
However, the resulting roll moment L is positive.)
If we assume that the flow over the wing is only altered at the aileron, then we may write
( 0

(CLa a) 12 V 2 c(y)
Lift0 (y) =
0

0 y < y1
y1 y y 2
y2 < y < 2b

(1)

where y1 is the spanwise location of the ailerons inner edge, y2 is the location of its outer edge, and c(y)
is the wing chord length at y. The aileron effectiveness is
CLa = CLw ,
where the parameter can be estimated as in Section B.2 of [1], exactly as in the case of an elevator or a
rudder. Also see Section B.9. The minus sign in (1) accounts for the fact that a positive aileron deflection
decreases the lift generated by the right wing.
is quite inexplicable. In the case of an elevator, one may argue that a positive deflection e should correspond to an increase
in lift. The argument is unconvincing, in my opinion, since the result is a negative (nose down) pitch control moment. Since
exerting pitch control moments is the entire purpose of the elevator, it would seem to me that a positive deflection should
yield a positive moment. At least there is a somewhat defensible reason for the convention, though, and we have adopted
Etkin and Reids convention for elevator deflections. In my view, however, there is simply no logical explanation for defining
a the way that Etkin and Reid do. We will not adopt their convention for a.

Substituting Lift0 (y) into our expression for L, and normalizing by QSb, we obtain the increment in
roll moment coefficient due to an aileron deflection:
Z y2

1
Claileron =
2
y CLw a Qc(y) dy
QSb
y1
Z
2CLw a y2
=
c(y)ydy.
Sb
y1
Therefore, the aileron power is
Cla

2CLw
Cl
=
=
a
Sb

y2

c(y)ydy.

y1

The aileron power can be increased

by making the flap chord larger (which increases ), by
R yplacing it farther
Ry
outboard (which increases y12 c(y)ydy), or by increasing its span (which also increases y12 c(y)ydy).
At this point, we have described the primary effects of the lateral directional control surfaces. Namely, we
have described how the rudder exerts a yaw moment and how the ailerons exert a roll moment. In addition,
these actuators have secondary effects which couple them together. For example, in discussing the necessity
of a rudder, we mentioned the adverse yaw that results from aileron deflections. The contribution of the
ailerons to yaw moment is captured by the term Cna a. Conversely, a rudder deflection causes a small
roll moment Clr r which must be countered by the ailerons.

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.

Lecture 9: Lateral-Directional Steady Flight

A complete discussion of lateral-directional equilibrium flight requires some advanced material which we
will cover in coming lectures. For this reason, the authors of [1] postpone the topic until Section 7.8.
There is no great difficulty in studying lateral-directional equilibrium flight, however, if one is willing to
temporarily accept a few observations without proof.
As we will see shortly, the attitude kinematic equations for a rigid aircraft are

1 sin tan cos tan

p
= 0

cos
sin
q .
0 sin sec cos sec
r

(1)

In considering lateral-directional steady flight, we will assume that and remain constant. For most
flight conditions, the 3 3 matrix above is invertible. Moreover, if and remain constant,

p
sin

q = cos sin .
(2)
r
cos cos
If the aircraft is not turning, then = 0 and = 0. Otherwise, is a constant vector which is parallel to
the inertial vertical axis.
As we will soon derive, the dynamic equations are

X(v, , u)
mv = mv + Y (v, , u)
Z(v, , u)

L(v, , u)
I = I + M (v, , u)
N (v, , u)

sin
+ W cos sin
cos cos

(3)

(4)

For steady flight, v = 0 and = 0. Moreover, in normal flight, all components of v and are small, with
the exception of u V . Arguing (informally, for now) that products of small terms are negligible we may
write

0
mv mV r
and
I 0.
mV q
Thus, for steady flight, we require

X
0
sin
Y = mV r + W cos sin
Z
mV q
cos cos

L
M = 0.
N

Non-turning flight with steady sideslip. For steady, non-turning flight, = 0 which implies that
q = r = 0. In this case, the lateral-directional force and moment equations are
Y

W sin

L = 0
N

= 0.
1

For small roll angles, we may write

CY + CYa a + CYr r CW
Cl + Cla a + Clr r = 0
Cn + Cna a + Cnr r = 0
W
is the wing loading. These are three linear algebraic equations in the four unknowns
( 12 V 2 )S
, a, r and . The system is under-determined. To solve the equations, we fix one value to obtain three
equations in three unknowns.

where CW =

Landing in a cross-wind. For example, suppose one wishes to land in a cross-wind. To maintain align
the aircraft with the runway requires flying at a steady sideslip angle

1 Crosswind velocity
.
= sin
Total airspeed
Then one would solve the equations

CYa
Cla
Cna

CY
CW
a
0 r = Cl
0

Cn

CYr
Clr
Cnr

to obtain the necessary control commands and the corresponding roll angle.
Example. As an example, lets consider the Navion, whose mass and geometric parameters are
W = 2750 lbs,

S = 184 ft2 ,

b = 33.4 ft.

For sea level flight at M = 0.158, we have the following stability derivatives (all in units of per radian):

a
r

CY()

Cl()

Cn()

-0.564
0
0.157

-0.074
0.134
0.107

0.071
-0.0035
-0.072

The density is = 2.3769 10 3 slugs/ft3 and the airspeed is V = 176 ft/s. The weight coefficient is
therefore
2750

CW = 1
= 0.406.
2
2 (2.3769E 3)(176) (184)

Suppose one wishes to land in a 40 ft/s cross-wind; in this case,

40
= arcsin
= 0.223 rad 13.1 .
176
Solving

0
0.157 0.406
a
0.564
0.134
0.107
0 r = 0.074 (0.223)
0.0035 0.072
0

0.071
gives

a
0.056
3.2
r = 0.229 rad 13.1

0.230
13.2

Maximum allowable cross-wind at landing. Alternatively, suppose one wishes to compute the maximum crosswind (equivalently, the maximum sideslip angle ) in which an airplane can land. Assuming
that maximum rudder is the limiting condition, one would solve

CY CYa CW

CYr
Cl Cla
0 a = Clr (r)max .

Cnr
Cn Cna
0
If the resulting aileron deflection a is larger than the maximum allowable deflection, then the aileron is
the limiting factor and one must solve for the corresponding sideslip angle, rudder angle, and roll angle.
Control conditions for asymmetric thrust. One may also use the given equations to determine the
complete control conditions for asymmetric thrust. In this case, the original equations become
CY + CYa a + CYr r = CW
Cl + Cla a + Clr r = 0
Cn + Cna a + Cnr r = CnT
where, as was mentioned in a previous lecture,
Sp yp
T yp

= CT
.
2
Sb
Sb
2 V

CnT = 1

Requiring that the sideslip angle be zero, we obtain

CYa CYr CW
a
0
Cla Clr
0 r = 0 .
Cna Cnr
0

CnT
If either the resulting rudder angle or the resulting aileron angle is beyond the capability of the actuator,
then the actuators must be modified to provide more control power.
Steady turning flight revisited. Recall that we have already discussed the longitudinal conditions for
a steady turn at constant radius. The essential difference between this condition and wings level (nonslipping) flight is that the lift vector is deflected inward. The vertical component of lift must balance
the airplanes weight while the horizontal component provides the centripetal acceleration necessary to
maintain the turn.
Per the discussion in Section 7.8 of [1], we define a truly banked turn (or a coordinated turn) as one for
which (1) the angular velocity vector is constant and vertical (in the inertial frame) and (2) the resultant
of gravity and centrifugal force lies in the plane of symmetry.1 Equivalently, it is a turn for which the
lateral aerodynamic force Y is identically zero. Thus, in a truly banked turn, the pilot and passengers will
feel the combination of their own weight and their inward acceleration through the seat of their pants.
Because Y = 0 for a truly banked turn, we have
W sin = mV r

= mV cos cos
from which we obtain in terms of turn rate and climb angle :2
tan =

V
cos .
g

Recall that centrifugal force is a fictitious force which explains the feeling of being flung outward when experiencing a
turn. It is the negative of the (very real) force which is necessary to maintain the turn.
2
The term climb angle presumes that the equations have been expressed in stability axes. In this alternate body-fixed
reference frame, which we will discuss shortly, the pitch angle is re-defined so that it is zero in horizontal flight.

For a steady turn at constant altitude, the climb angle is zero and
tan =

V2
.
gRturn

In this case, may be specified by the desired speed and turn radius.
The non-dimensional lateral-directional force and moment equations for steady turning flight are
CY + CYp p + CYr r + CYa a + CYr r =

(W sin + mV r)
2 Sb
2 V

Cl + Clp p + Clr r + Cla a + Clr r = 0

Cn + Cnp p + Cnr r + Cna a + Cnr r = 0

where p and r represent nondimensional roll and pitch rate, respectively:

p =

b
p
2V

and

r =

b
r.
2V

Recalling that W sin = mV r for a truly banked turn, the right hand side of the Y coefficient equation
vanishes. Now recall equation (2):

p
sin

q = cos sin .
r
cos cos
Substituting into the definitions p and r and rearranging the steady flight equations above gives

CY
Cl
Cn

CYa
Cla
Cna

CYr

CYp
Clr a = Clp
r
Cnp
Cnr

CYr
b
sin
Clr
cos cos 2V
Cnr

where

V
= arctan
cos .
g
and climb angle , one may compute the sideslip angle and the
Thus, for a given speed V , turn rate ,
aileron and rudder deflections necessary to maintain a banked turn.

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.

Lecture 10: Review of Particle Dynamics

Static stability versus dynamic stability. At this point, we have discussed the notion of static stability
for longitudinal and lateral-directional motions. Static stability relates to an airplanes initial tendency
in response to a perturbation from equilibrium flight if this initial tendency is to return to equilibrium
flight, then we say the motion is statically stable. Dynamic stability concerns not only the short term, but
the long term as well. Dynamic stability requires that the system truly returns to the equilibrium after
a small perturbation. The advantage of considering static stability over dynamic stability is that static
stability can be assessed without solving, or even deriving, the complete equations of motion.

m
k
Figure 1: A mass-spring system.
As an example, consider a simple mass-spring
m
x = kx,
where x is measured from the equilibrium position. We can solve this simple linear, time-invariant ODE
explicitly, but that is unnecessary to determine static stability of the equilibrium. Simply recognize that,
if the system is perturbed to a new state (x, x)
= (x0 , 0) where x0 6= 0, then the system will initially tend
to return to the equilibrium provided k > 0:
(
k
<0
x0 > 0
x
(0) = x(0)
>0
x0 < 0
m
If the mass is pulled to some initial point x0 , held at rest, and then released, it will initially accelerate back
toward the equilibrium. If k < 0, on the other hand, the perturbed mass will initially tend to move away
from the equilibrium. Thus, positive spring stiffness is necessary and sufficient for static stability.

m
b
Figure 2: A mass-spring-damper system.
Next consider a mass-spring-damper
m
x = bx kx,

(1)

where we assume that m > 0 and k > 0, but we allow the possibility that b < 0. Once again, suppose that
the system is perturbed to a new state (x, x)
= (x0 , 0) where x0 6= 0. Noting that
k
b
x(0)

x(0)
m
m
k
= x(0),
m

x
(0) =

as before, we see that the system is, once again, statically stable. The perturbed mass will initially move
in the direction of the equilibrium regardless of the value of b.
1

Assuming that 1 1, the general solution to equation (1), with the given perturbed initial condition,
is
p
x(t) = x0 en t cos d t
where
d = 1 2 n

where

p
k
b
,
=
,
and
n = n 1 2 .
m
2 km
If b < 0, then 1 < < 0 and the exponential envelope, and the sinusoid which it bounds, diverges as
t . The equilibrium is (dynamically) unstable! Clearly, some additional condition(s) besides k > 0
must be imposed to ensure dynamic stability. While the requirement b > 0 is certainly necessary for
dynamic stability, it turns out not to be sufficient in general.
n =

As a first step toward understanding dynamic stability of equilibrium flight, we must develop the dynamic
equations for an airplane. We will start at the very beginning.

Fi

Xi

X cm

yI
xI

zI

Figure 3: System of particles.

Particle dynamics: Translational Motion. Consider a collection of N point masses, as shown in
Figure 3. Suppose that the ith point mass has mass mi and that its location in inertial space is given by
the vector X i . (We will use capital letters to denote vectors expressed in the inertial reference frame.)
Suppose that a net external force F i acts on the ith point mass. Also, suppose there is a force of interaction
F ij exerted by the j th particle on the ith . Newtons second law of motion applied to the ith particle gives
i = Fi +
mi X

N
X

F ij .

j=1

Summing over the N particles gives

d2
dt2

N
X

mi X i =

i=1

N
X
i=1

F i +

N
X
j=1

F ij .

Now, assume that the interaction forces are equal and opposite: F ij = F ji . (Note that this implies
Fii = 0; the ith point mass exerts no force on itself.) All of the forces of interaction cancel in the
summation, leaving
N
N
X
d2 X
m
X
=
F i.
(2)
i
i
dt2
i=1

i=1

The center of mass of the collection of particles is defined as

N
!
1 X
X cm =
mi X i
where
m

m=

i=1

Referring to (2), we see that

N
X

mi .

i=1

X
d2
(mX
)
=
F i.
cm
dt2
i=1

Recall that two force systems are equivalent provided the net magnitude and direction of the forces is
equal. (This is not true for moment systems, where the point of action of each force is also important.)
We may therefore define an equivalent force
F =

n
X

Fi

i=1

which acts at the point X cm . Thus, we may write

cm = F .
mX

(3)

The effect of the N point forces F i acting on the each of the N particles is equivalent to the effect of a
single force F on a single point mass m located at X cm . If one were to solve the dynamic equations for
each of the point masses as well as equation (3), one would find that the point X cm (t) remains the center
of mass even as the various particles move toward each other, collide elastically or plastically, move away
from each other, and so on. This is an important, fundamental observation about the motion of particle
systems.

Xi

Fi

Xi

X cm

yI
xI

zI

Figure 4: System of particles.

Particle Dynamics: Rotational Motion. To determine the rotational motion of the particle system,
i denote the position of
we must go back and consider the moment generated by each point force F i . Let X
th
the i point mass with respect to the center of mass, expressed in the inertial frame. With this definition,
we may take the moment of translational momentum (the angular momentum) of each particle about the
inertial frame origin:
i.
H i = X i mi X
3

Summing the contributions over the N particles gives the total system momentum about the inertial frame
origin
H =

N
X

i
X i mi X

i=1
N n
X
i=1

N
X

i mi X
cm + X
i
X cm + X

mi X i + mi X i X cm + X i mi X i
=
X cm mi X cm + X cm
dt
i=1
N
!
N
! N
!
N

X
X
X
X
d

i mi X
i
= X cm
mi X cm + X cm
mi X i +
mi X i X cm +
X
dt
i=1

i=1

i=1

i=1

The middle two terms vanish by definition of the center of mass. Thus, the total angular momentum of
the system about the inertial coordinate origin is the sum of the moment of translational momentum (as
if concentrated at the center of mass) and the angular momentum of the system about its center of mass:
cm +
H = X cm mX

N
X
i=1

i mi X
i .
X

Differentiating, we obtain
=X
cm mX
cm + X cm mX
cm +
H

N
X
i=1

i mi X
i +
X

N
X
i=1

.
i mi X
X
i

Because the cross-product of a vector with itself is zero, the first and third terms vanish. Thus,
= X cm mX
cm +
H

N
X
i=1

.
i mi X
X
i

By Newtons second law,

= X cm F +
H

N
X
i=1

i F i +
X

N
X
j=1

F ij .

In words, the rate of change of total angular momentum about the origin of the inertial reference frame is
equal to the sum of
1. the moment of all external forces about the origin of the inertial reference frame and
2. the moment of all (internal and external) forces about the center of mass.
We define the total moment M cm acting about the center of mass as

N
N
X
X
i F i +
M cm =
X
F ij
i=1

j=1

Xj
Fji
( Xi - Xj)
Fij

Xi
Figure 5: Forces of interaction between the ith and j th particles.
Recall the assumption that F ij = F ji . We may therefore rewrite the contributions to M cm due to the
internal forces as follows
N X
N
N X
i
X
X

i F ij + X
j F ji
i F ij =
X
X
i=1 j=1

i=1 j=1

i
N X
X

i=1 j=1

N X
i
X

i=1 j=1

i F ij + X
j (F ij )
X

iX
j F ij .
X

Assuming that the forces of interaction between particles act along the line joining the particles, these
terms vanish leaving
N
X

i Fi .
M cm =
X
i=1

We therefore have

= X cm F + M cm .
H

(4)

Everything we have said thus far applies to a system of particles in unconstrained motion. The only
assumption we have made is that the forces of interaction between particles are equal and opposite and
that the line of action is along the line joining the points. The conclusions are quite general.
In the next lecture, we will consider a collection of particles which are rigidly constrained to one another.
Starting from equation (4), we will then obtain a rotational version of equation (3) for a rigid body. Along
the way, we will define an important geometric object known as the inertia tensor. Ultimately, we will
obtain six differential equations that completely describe the dynamics of a rigid body moving under the
influence of external forces and moments.

Lecture 11: Rigid Body Dynamics

Xi
Fi

Xi

Xcm

yI
xI

zI

Figure 1: N rigidly constrained particles.

Recall that we defined H to be the angular momentum of a particle system about the origin of an inertial
reference frame. We showed that
H =

N
X
i=1

i
X i mi X

cm +
= X cm mX

N
X
i=1

i mi X
i .
X

Assuming that the collection of particles forms a rigid body (i.e., that no particle moves relative to any
other particle), we find that
cm +
H = X cm mX

N
X
i=1

i mi ( I X
i ),
X

where I is the angular velocity of the body with respect to inertial space, written in the inertial frame.
The latter term is simply the angular momentum of the system of particles about the center of mass
(expressed in the inertial frame), which we will denote H cm . Thus
cm + H cm .
H = X cm mX

(1)

Suppose that each particle has the same infinitesimal mass mi = dm. In the limit N , we may write
Z
(X
I )dm.
H cm =
X
V

= XI xI + YI y I + ZI z I denotes a point in the rigid body. As an exercise, you

where the inertial vector X
may check that
2

Z
Z
YI + ZI2 XI YI
XI ZI
(X
I )dm = I I I
XI YI X 2 + Z 2
YI ZI dm.

X
where
II =
I
I
V
V
XI ZI
YI ZI XI2 + YI2
1

Differentiating (1) gives

= X cm mX
cm + H
cm
H
cm + H
cm
= X cm mX

(2)

Next, recall from the previous lecture that

Comparing (2) and (3) gives

= X cm F + M cm .
H

(3)

cm = M cm
H

(4)

where H cm = I I I .
The problem with equation (4) is that the inertia tensor I I that defines H cm varies with time. Because the
body rotates with respect to inertial space, the positions of the infinitesimal mass elements appearing in the
definition of the inertia tensor change with time. It turns out to be much easier to express the equations of
motion in a coordinate frame that is fixed in the body because, in this case, the inertia tensor is constant.
Another advantage to using a body-fixed reference frame is that aerodynamic forces and moments are more
naturally expressed with respect to the body than with respect to inertial space.

yB
Xi

Fi

xB
zB

Xi

X cm

yI
xI

zI

Figure 2: Inertial and body-fixed coordinate frames.

Let us re-express the preceding developments in terms of a coordinate frame (xB , y B , z B ) which is fixed to
the rigid body at its center of mass. To start with, we define the angular momentum hcm about the center
of mass and expressed in the body frame, as
Z
(
hcm =
x
x B )dm.
V

= xB xB + yB y B + zB z B denotes a point in the rigid body and where

where the body frame vector x
T
B = [p, q, r] is the angular velocity of the body with respect to inertial space, but expressed in the
body frame. We thus have
2

2
Z
Z
yB + z B
xB yB xB zB
xB yB x2 + z 2 yB zB dm.
(

x
x B )dm = I B B
where
IB =
(5)
B
B
2
2
V
V
xB zB yB zB xB + yB
2

Because the reference frame in which these integrals are expressed moves along with the body, every
component of I B is constant. We define the moments and products of inertia according to the relationship

2
Z
Ix
Ixy Ixz
yB + z B
xB yB xB zB
Ixy
xB yB x2 + z 2 yB zB dm
Iy
Iyz :=
B
B
2
V
Ixz Iyz
Iz
xB zB yB zB x2B + yB
Comment: Suppose that the proper rotation matrix RIB (t) transforms free vectors expressed in the body
reference frame into the inertial reference frame. Then I = RIB B and H cm = RIB hcm , so that
I I I = RIB (I B B )
= RIB I B R1
IB I .
Since this equation holds for any angular velocity vector I , we conclude that
I I = RIB I B R1
IB .
The rotation matrix RIB is a function of time because the body is rotating. Thus, while the matrix I B is
a matrix of constants (see definition in (5)), the matrix I I is time-varying; at any instant, it depends on
the orientation of the body-fixed frame with respect to inertial space.
To summarize, we have hcm = I B B , where I B is a constant matrix. If we define mcm = RBI M cm to
be the total moment about the center of mass, expressed in the body reference frame, then we have only
to relate the rate of change of hcm to mcm . This is not quite as simple as it might appear: h cm 6= mcm !
While having a constant inertia matrix does simplify the equations of motion, a new complication arises
due to the rotating body-fixed reference frame.
Rotating reference frames. The time rate of change of a vector written with respect to an inertially
fixed frame is simply the derivative of the vectors components. The time rate of change of a vector written
with respect to a rotating coordinate frame, however, depends also on the rate of rotation of the coordinate
frame. Let a(t) be a vector expressed with respect to a rotating reference frame. Let A(t) = RIB (t)a(t)
represent the same vector, but expressed in an inertial frame. Then
d
d
(a) =
(a) + B a

dt Inertial
dt Body
= a + B a
(6)
where

(a) = RBI A.
(7)

dt Inertial
(See any undergraduate dynamics textbook for a proof.) On the left hand side of equation (6), the time
derivative is taken with respect to inertial space; on the right hand side, the time derivative is taken with
respect to the rotating coordinate frame. On both sides of the equation, vectors are expressed in the
body frame. Notice the important distinction between the phrases with respect to and expressed in.
Although they are taken with respect to different reference frames, the time derivatives on either side of (7)
are expressed in the same frame. If one desired, one could re-express the vectors in the inertial frame by
premultiplying both sides by RIB = R1
BI . In that case, we would have
= RIB (a + B a)
A

= RIB a + (RIB B ) (RIB a)

= RIB a + I A.

(Notice that left multiplication by a rotation matrix distributes over the cross-product; this is not true for
general, non-rotation matrices.)
3

zI , zB
y(t)

yI

yB

xB
y(t)

xI

Figure 3: An ant on a turntable.

Example: An ant on a turntable. Think of an ant walking on top of a spinning turntable. The angle
is the counterclockwise angle from the inertial xI axis to the body-fixed xB axis, as shown in Figure 3.
The transformation from the body frame to the inertial frame is given by the rotation matrix

cos (t) sin (t) 0

RIB (t) = sin (t) cos (t) 0 .
0
0
1
Let us check this assertion by mapping the unit vector xB into the inertial frame:

cos (t)
1
cos (t) sin (t) 0
RIB (t)xB = sin (t) cos (t) 0 0 = sin (t)
0
0
0
1
0 B
I
As a simple check, refer to Figure 3 and notice that, for small, positive angles , the vector xB , re-expressed
in the inertial frame, should have a relatively large positive component (cos ) in the xI direction and a
small positive component (sin ) in the y I direction.
The inverse transformation, i.e., the map from the inertial frame to the body frame, is

cos (t) sin (t) 0

RBI (t) = RIB (t)1 = RIB (t)T = sin (t) cos (t) 0 .
0
0
1
The turntable spins with inertial angular velocity

0
d
I= 0 .
((t)z I ) = z
I =
dt

(t)
I
We may convert the angular velocity from the inertial

cos (t) sin (t)

sin (t) cos (t)

B = RBI (t) I =
0
0

frame to the body frame by writing

0
0
0

0 0 = 0 = (t)z
B.

1
I
(t) B

For this simple example of planar rotation, the two vectors I and B are numerically equal. This is not
true for general, three-dimensional rotations.
Suppose that, at a given instant t, the ant is located by the body frame vector

x(t)
xant = y(t) .
0
B
4

In the inertial frame, we have

X(t)
cos (t) sin (t) 0
x(t)
X ant = Y (t) = RIB xant = sin (t) cos (t) 0 y(t)
Z(t) I
0
0
1
0
B

x(t) cos (t) y(t) sin (t)

= x(t) sin (t) + y(t) cos (t) .
0
I

(8)

The velocity of the ant with respect to inertial space, but written in the body frame, can be obtained
from (6) as
d
(xant ) = x ant + B xant

dt Inertial
or

0
x(t)

x(t)
x(t)
y(t)
ant = y(t)
+ 0 y(t) = y(t)

RBI X
+ x(t) .

0
0
B
0
B
B
B
As an exercise, you might verify this result by computing the total time-derivative of X ant directly from (8)
and then premultiplying by RBI .
If we smash the ant, so that it becomes part of the rigid body, then the relative velocity x ant will be zero
and the inertial velocity is due entirely to the angular velocity of the turntable.
cm represent the inertial velocity
Rigid body dynamic equations in body coordinates. Let V cm = X
of a given rigid bodys center of mass and define the inertial translational momentum of the center of mass
Gcm = mV cm .
We have already defined the inertial angular momentum H cm = I I I , where I I is the inertia matrix
computed with respect to the inertial frame and where I is the angular velocity of the body with respect
to inertial space, written in the inertial frame. The complete dynamic equations for a rigid body, written
in the inertial frame, are
cm = F
G
cm = M cm
H

(9)

These two vector equations seem harmless enough, but recall that the inertia I I is time-dependent because
the body rotates. Because of this, and because aerodynamic forces on an airplane are more easily expressed
in a body-fixed coordinate frame, it is more convenient to express the dynamic equations in the body frame.
To do so, we use the identity (7). Let

u
v cm = RBI V cm = v
w B
represent the translational velocity of the mass center with respect to the inertial frame, but written in the
body frame. With this definition, we may write the translational momentum and the angular momentum
about the mass center in the body frame as
g cm = RBI Gcm

= RBI (mV cm )

hcm = RBI H cm = RBI (I I I )

= mv cm
= I BB.

(10)

1
(In the latter equation, we have substituted B = R1
IB I and used the observation that I B = RBI I I RBI .)
Applying the identity (6) to compute the derivatives of g cm and hcm with respect to inertial space gives
d
cm = g cm + B g cm
(g ) = RBI G

dt Inertial cm
d
cm = h cm + B hcm .
(hcm ) = RBI H

dt Inertial
cm = F and H
cm = M cm from (9), as well as the definitions of g cm and hcm from (10),
Substituting G
gives

RBI F

= mv cm + B mv cm

RBI M cm = I B B + B I B B .
Defining the body frame expressions for force and moment, f = RBI F and mcm = RBI M cm , we finally
obtain
mv cm + B mv cm = f

I B B + B I B B = mcm .
As we will almost always express equations in a body frame fixed at the center of mass, we will typically
omit the related subscripts:
mv + mv = f

I + I = m.

Because the xz-plane of an aircraft is typically a plane of symmetry, it is possible to choose the body
reference frame such that Ixy = Iyz = 0.1 Writing the equations of motion explicitly, assuming such a
choice of body reference frame, we have
m(u + qw rv) = X + Wx

m(v + ru pw) = Y + Wy

m(w + pv qu) = Z + Wz

Ix p Ixz r + qr(Iz Iy ) Ixz pq = L

Iy q + pr(Ix Iz ) Ixz (p2 r2 ) = M

Iz r Ixz p + pq(Iy Ix ) + Ixz qr = N.

These six first order, nonlinear, ordinary differential equations (ODEs) describe the motion of a rigid body
whose body xz-plane is a plane of symmetry. The force terms Wx , Wy , and Wz are the components of
the body weight in the body reference frame. The remaining force and moment components on the right
arise due to aerodynamic effects. These equations are impossible to solve analytically, except in very
special cases. One may certainly solve the equations numerically, for given forcing and initial conditions,
but pencil-and-paper analysis is challenging. We therefore must develop a simpler set of approximate
equations. This will be a topic for a later class.
The equations above are the dynamic equations, which describe how external forces and moments affect
the translational and rotational velocity of the rigid body. In the next lecture, we will derive six more first
order ODEs which relate translational and rotational velocity to position and attitude. These equations
are known as the rigid body kinematic equations.
1
In fact, one could choose a reference frame whose axes are aligned with the principal axes of inertia. In this case, all
cross products of inertia would vanish. Such a choice, however, makes the expression of aerodynamic forces and moments less
convenient.

Lecture 12: Rigid Body Dynamics & Kinematics

yB
yI

xI
xB

zI zB

Xcm

yI
xI

zI

Figure 1: Rigid body coordinate frames.

Recall that we are considering the dynamics of a rigid body. The rigid body dynamic equations, expressed
in a reference frame fixed in the body at the center of gravity, are
mv + mv = f
I + I = m.
Because the xz-plane of an aircraft is typically a plane of symmetry, it is possible to choose the body
coordinate axes in such a way that Ixy = Iyz = 0. Writing the equations of motion explicitly, assuming
such a choice of body coordinates, we have
m(u + qw rv) = X + Wx
m(v + ru pw) = Y + Wy
m(w + pv qu) = Z + Wz
Ix p Ixz r + qr(Iz Iy ) Ixz pq = L
Iy q + pr(Ix Iz ) Ixz (p2 r2 ) = M
Iz r Ixz p + pq(Iy Ix ) + Ixz qr = N,
where X, Y, and Z are the components of the total aerodynamic force (including propulsive forces), Wx , Wy ,
and Wz are the components of weight, and L, M, and N are the components of the total aerodynamic
moment. These are the rigid body dynamic equations, which describe how the forces and moments affect
the translational and rotational velocity of the rigid body. We must also develop the kinematic equations,
which relate translational and rotational velocity to position and attitude.
Suppose that the body merely rotates about its center of mass, without translating, and that the center of
mass is located at the origin of the inertial reference frame. (Alternatively, suppose that a reference frame
which is aligned with the inertial reference frame is pinned at the center of mass and translates along with
the body.) Then the two frames are related by a matrix RBI satisfying
xB = RBI xI ,

y B = RBI y I ,
1

z B = RBI z I ,

where RBI is a proper rotation matrix, i.e., it is a 3 3 rotation matrix which preserves cross-products.
Mathematically, we say that

RBI A R33 | A1 = AT , det(A) = +1 ,

where the set on the right is the set of all real-valued, proper, 3 3 rotation matrices. While a 3 3
matrix contains as many as nine independent entries, the condition A1 = AT can be shown to impose
six constraints on the matrix RBI . Thus, at most three independent numbers are required to specify the
matrix RBI . (The additional condition that det(A) = +1 limits RBI to be a proper rotation, i.e., one
which preserves right-handed coordinate frames.) There are a great many attitude parameterizations
available to express the matrix RBI , each with advantages and disadvantages. The one most commonly
used in aircraft dynamics is the XY Z Euler angles.
x2

yI
y1

x1

x2 , x B

f
y2

y1 , y2

x1

xI

yB

q
y

z2

z I , z1

y1

yI

x2

zB
z2

z1

x1

zB

x1

z2

y
z I , z1

yB
f

q
xI

z2

z1

y 1 , y2

x 2 , xB

y2

Figure 2: XY Z Euler angle rotations.

Suppose that one wishes to transform a vector given in inertial coordinates (xI , y I , z I ) to the rotating body
coordinate frame (xB , y B , z B ). To do so using Euler angles, one first expresses the vector in an intermediate
coordinate frame (x1 , y 1 , z 1 ). This intermediate frame is obtained by rotating the inertial frame in the
positive direction about the z I axis through the yaw angle . The transformation is given by the rotation
matrix

cos sin 0
R1I = sin cos 0 .
0
0
1
For example, if one wishes to transform the vector xI = [1, 0, 0]TI from the inertial frame to the 1-frame,
one computes

1
cos
R1I 0 = sin = cos x1 sin y 1 .
0 I
0
1
Next, one rotates the intermediate 1-frame in the positive direction about the y 1 axis through the pitch
angle . This transformation defines a new coordinate frame, call it the 2-frame, and is given by the

rotation matrix
R21

cos 0 sin
.
1
0
= 0
sin 0 cos

Finally, one rotates the intermediate 2-frame in the positive direction about the x2 axis through the roll
angle . This transformation is given by the rotation matrix

1
0
0
RB2 = 0 cos sin .
0 sin cos
The complete transformation from inertial to body coordinates is given by the composition of the three
transformations:

cos cos
cos sin
sin
RBI = RB2 R21 R1I = cos sin sin cos sin cos cos + sin sin sin cos sin . (1)
cos sin cos + sin sin sin cos + sin cos sin cos cos
Example. Suppose we wish to determine the pitch and roll angle in equilibrium flight from a bodyfixed accelerometer measurement. Because an accelerometer measures actual acceleration plus a sensed
acceleration due to gravity, the measurement in equilibrium flight will be gz I . Computing

0
sin
ameasured = RBI (gz I ) = RBI 0 = g cos sin ,
g I
cos cos B
we see that

= arcsin

ameasured1
g

and

ameasured2
= arcsin
,
g cos

provided cos 6= 0 (i.e., provided 6= 2 ). Notice that we can not obtain the yaw angle from a
measurement of the direction of gravity. This makes sense because is measured about the direction of
gravity. The accelerometer could be rotated through any yaw angle without affecting the measurement.
Notice also that when = 2 , the axes z I and xB are collinear. In this condition, the yaw angle is
indistinguishable from the roll angle . The XYZ Euler angle representation is singular when = 2 .
Because RBI is a rotation matrix, its inverse is its transpose. Thus, the inverse transformation from the
body to inertial frame is
T
RIB = R1
(2)
BI = RBI .
From here, it is straightforward to express the translational kinematics. If the position of the bodys center
of mass is given by the inertial coordinate vector [x, y, z]TI , then we have

x
u
y = RIB v .
z I
w B
Thus, given a translational velocity history in the body frame and the attitude history RIB (t), one may
integrate from the initial position in order to determine the bodys position in inertial space.
To determine the attitude kinematics (i.e., the relationship between and the rate of change of RIB ), we
I , y
, and x
2 in a
let the yaw, pitch, and roll angles vary with time and sum their time derivatives z
1
3

compatible reference frame, say the body reference frame. To this end, we note that

0
sin

I = RBI 0 = cos sin ,

RBI z
cos cos B
I

and that

RB1 y
1 = (RB2 R21 ) y 1 = RB2 y 2
and finally that
2
RB2 x

1
0
0
= 0 cos sin
0 sin cos

0
0
= cos ,
sin B
0 2

1
B = 0 .
= x
0 B

The angular velocity of the body with respect to inertial space, but written in the body frame, is

sin
0
p
1
= 0 + cos + cos sin
= q
cos cos B
sin B
0 B
r B

1
0
sin

0 cos cos sin

=
.
0 sin cos cos

q 6= ,
and r 6= ,
in general!
Note that p 6= ,

We define the matrix mapping Euler angle rates to body angular velocity as

1
0
sin
LBI = 0 cos cos sin .
0 sin cos cos

As with the rotation matrix RBI , the subscript BI here connotes a map from the inertial frame to the
h
iT
,
is not really a free vector
body frame. This is a slight abuse of notation, however, as the vector ,
in the same sense as angular rate and angular velocity. The column vector of Euler angle rates is not
affiliated with any particular reference frame; it is merely a column of numbers.
The determinant of LBI is
detLBI = cos .
Thus, the relationship between body angular velocity and the Euler angle rates may be inverted provided
that 6= 2 . Assuming this is the case, we have

= LIB

where

LIB = L1
BI

1 sin tan cos tan

cos
sin .
= 0
0 sin sec cos sec

(3)

Unlike RBI , the matrix LBI is not a rotation matrix. So, for example, its inverse is not its transpose!
It should be pointed out that the singularity at = 2 is purely a consequence of the coordinate choice.
Certainly there is nothing physical which limits an airplane from obtaining such a pitch angle. The problem
can be understood by recalling that, when = 2 , the axes z I and xB are collinear. In this condition,
the yaw angle is indistinguishable from the roll angle .
4

To summarize, the translational and attitude kinematics, with rotations parameterized by Euler angles,
are

x
y = RIB v
(4)
z

= LIB
(5)

where RIB is given by (2) and (1) and where LIB is given by (3). These equations hold provided that
6= 2 .
For convenience, we will lump the primary control parameters into one control vector

a
e

u=
r
T

where T represents a thrust command. Noting that the force of gravity expressed in body coordinates is

0
sin
RBI 0 = mg cos sin ,
mg I
cos cos B
the rigid body dynamic equations are

X(v, , u)
sin
mv + mv = Y (v, , u) + mg cos sin
Z(v, , u)
cos cos

L(v, , u)

M (v, , u) .
I + I =
N (v, , u)

(6)

(7)

Equations (4) through (7) completely describe the motion of a rigid airplane. They are twelve first order,
nonlinear, time-invariant, ordinary differential equations. In general, these equations are impossible to
solve analytically. Often, however, one may learn quite a lot about the behavior of solutions to these
equations by studying a simpler set of approximate equations which are linear, time-invariant. We will
next discuss the process of linearization.

Lecture 13: Linearization

Consider a set of nonlinear, time-invariant ODEs. Recall that any nth order ODE can be re-written as n
first order ODEs. We may therefore assume, without loss of generality, that the system takes the form
x = f (x, u),

(1)

where f is a vector-valued function of the state x and input u. While we may well know the form of the
input u, we do not necessarily know the resulting form of x(t), for a given initial state x(0). A solution or
(t)) satisfying
trajectory of (1) is a pair (
x(t), u
= f (
).
x
x, u
Some solutions are special in the sense that they do not depend on time. An equilibrium is a solution
(t)) = (xe , ue ) for which xe and ue are constant, that is
(
x(t), u
x e = 0 = f (xe , ue ).
) equal to zero and solves the resulting nonlinear
To determine equilibria for a given system, one sets f (
x, u
algebraic equations for x = xe and u = ue .
Linearization: View #1. Given a solution, and assuming that the components of f are suitably smooth,
one may expand the right-hand side of equation (1) in a multivariable Taylor series. Doing so, one obtains


f
f
) +
) + h.o.t.,
) +
(x x
(u u
(2)
x = f (
x, u
x
u
(t)). So far, we have
where the asterisk denotes that the quantity is evaluated along the solution (
x(t), u
said nothing new. Equation (2) is an equivalent representation of equation (1). Rearranging (2), we obtain


f
f

) +
) + h.o.t.,
=
(x x
(u u
x x
x
u
(t)) is a solution in order to replace f (
) with x
. The
where we have used the fact that (
x(t), u
x, u
)k and k(u u
)k remain small, so that we may neglect
linearization occurs when we assume that k(x x
the higher order terms in the Taylor series expansion. We make the following definitions:

x = x x

u = u u

f
A(t) =
x

f
B(t) =
u
The linearized state equation is
d
x = A(t)x + B(t)u
dt
So long as x and u remain small enough, these equations will provide a good approximation to
equations (1). Note that the state itself need not remain small. Only the difference between its actual and
nominal values must remain small.
In general, the state matrix A and the input matrix will depend on time explicitly. Even though the
(t)) on time will carry
original nonlinear system is time-invariant, the dependence of the solution (
x(t), u
1

b
l

mg

Figure 1: A planar pendulum with linear damping and an input torque.

through to the components of A and B. Linearizing about an equilibrium (xe , ue ), however, gives n first
order, linear time-invariant ODEs
d
x = Ax + Bu.
dt
Example. Consider the planar pendulum shown in Figure 1. Summing moments about the pivot gives
the following equation of motion
ml2 = mgl sin b + T.
Letting
x=

and

u = T,

we may rewrite this second order, nonlinear, time-invariant ODE as two first order equations

x2
x 1
.
=
x 2
gl sin x1 mlb 2 x2 + ml1 2 u

(3)

More compactly, we say

x = f (x, u)
where
f (x, u) =

gl sin x1

x2
b
x
ml2 2

1
u
ml2

One equilibrium of the system (3) is

(xe , ue ) = (0, 0) .

(4)

The equilibrium (4) corresponds to the case where the pendulum hangs vertically downward at rest. (There
is one other equilibrium for which ue = 0, corresponding to the pendulum standing vertically on end.)
Linearizing equations (3) about the equilibrium (4), as described above, we first obtain

f1

f1
f1
f
f
0
1
0
x
x
u
1
2
=
=
and
=
=
.
1
f2
f2
f2
gl cos x1 mlb 2
x
u
ml2
x1
x2
u
Evaluating these matrices at the equilibrium (4) gives

0
1
A=
and
gl mlb 2
2

B=

0
1
ml2

Thus, the dynamics linearized about the equilibrium (4) are

d
x = Ax + Bu.
dt
Written out explicitly, these equations are

d
x2
x1
=
g
b
x

x2 +
x

dt
2
1
l
ml2

1
u
ml2

These are precisely the equations one would obtain by using the small angle approximation sin .
Indeed, that approximation is simply the linearization of sin about the angle = 0.
Although the nonlinear equations (3) are difficult to solve analytically, one may use Matlab to compute a
numerical solution for the pendulum response. Suppose the physical parameters are
g = 10 m/s2 ,

m = 0.1, kg,

l = 1 m,

and

b = 0 Nm/(rad/s).

Assume that no control torque is applied (T = 0), so that the response is due entirely to the initial
condition, say:

(0) = 0
(0)
= 0.
Consider the following three cases.
p
g/l rad/s.
p
2. 0 = 179.9 : The linear system oscillates at g/l rad/s but the nonlinear system takes a full 10
seconds to swing to the other side and back again.
p
3. 0 = 180 : The linear system oscillates at g/l rad/s but the nonlinear system remains at the inverted
equilibrium! Multiple isolated equilibria are a phenomenon exhibited only in nonlinear systems.

1. 0 = 20 : Both systems oscillate at around

Except in the last example, damping would gradually bring to zero. However, the transient response for
the two systems would remain drastically different for large . This example should underscore the fact
that linearization only provides a valid approximation when the nonlinear systems motion stays close
to the nominal motion. For the pendulum, this means that should remain small.
Angle versus Time for a Simple Pendulum
200
True System Response
Linear Approximation
150

100

(deg)

50

50

100

150

200

5
Time (s)

10

Figure 2: Initial Condition Response: 0 = 20 .

Angle versus Time for a Simple Pendulum

200
True System Response
Linear Approximation
150

100

(deg)

50

50

100

150

200

5
Time (s)

10

Figure 3: Initial Condition Response: 0 = 179.9 .

Angle versus Time for a Simple Pendulum
200

150

100

(deg)

50

True System Response

Linear Approximation

50

100

150

200

5
Time (s)

10

Figure 4: Initial Condition Response: 0 = 180 .

Linearization: View #2. Another way to think about linearization is to suppose that a system is slightly
disturbed from a nominal state of motion. In this view, the actual state x is the sum of the nominal state,
say x0 , and the state disturbance, say x. Similarly, the actual input u is the sum of the nominal input,
say u0 , and the input disturbance, say u.
For the pendulum equations

x 1
x 2

gl sin x1

x2
b
x
ml2 2

1
u
ml2

let x = x0 + x and u = u0 + u, where x0 is a nominal value of the state (e.g., an equilibrium value)
and u0 is a nominal value of the input. Substituting above gives

d
x20 + x2
x1
x 10 + dt
=
d
gl sin(x10 + x1 ) mlb 2 (x20 + x2 ) + ml1 2 (u0 + u)
x 20 + dt
x2

x20 + x2
=
.
gl (cos x10 sin x1 + sin x10 cos x1 ) mlb 2 (x20 + x2 ) + ml1 2 (u0 + u)
The equations above are equivalent to the original equations. The linearization takes place when we
4

assume that the disturbance is small. In that case, cos x1 1 and sin x1 x1 and the equations
above can be re-written as

d
x20
x2
x 10
x1
=
+
.
+
x 20
gl sin x10 mlb 2 x20 + ml1 2 u0
gl cos x10 x1 mlb 2 x2 + ml1 2 u
dt x2
Now, the nominal state x0 and input u0 about which we have linearized are a assumed to be a solution of
the dynamic equations:

x20
x 10
.
=
x 20
gl sin x10 mlb 2 x20 + ml1 2 u0
Subtracting this equation from the one above leaves the small disturbance or perturbation equation

d
x2
x1
=
.
gl cos x10 x1 mlb 2 x2 + ml1 2 u
dt x2

If we consider the nominal state (x0 , u0 ) = (0, 0), as before, then these equations are identical to the
linearized equations obtained previously.
Nominal flight condition for a rigid airplane. Consider a rigid airplane with a coordinate frame fixed
at the center of gravity such that the xz-plane is a plane of symmetry. Written in these coordinates, the
kinematic equations are

x
y = RIB (, , )v
z

= LIB (, )

and the dynamic equations are

X(v, , u)
sin
mv + mv = Y (v, , u) + mg cos sin
Z(v, , u)
cos cos

L(v, , u)

M (v, , u) .
I + I =
N (v, , u)

yB
V0

xB
zB

Figure 5: Stability Axes.

There are several choices of body-fixed axes which may be more or less convenient in studying stability and
control of aircraft motion. Three common choices, all of which have the xz-plane as a plane of symmetry,
are
5

1. Body axes. This is the most general body-fixed coordinate frame for which the xz-plane is still a
plane of symmetry. The coordinate axes might be chosen, for example, such that the x-axis is parallel
with the airplanes zero-lift line.
2. Principal axes. In this coordinate frame, all three cross-products of inertia are zero.
3. Stability axes. This coordinate frame is defined relative to a particular wings-level equilibrium flight
condition. The stability axes are defined such that the x-axis is aligned with the velocity vector in
this nominal flight condition; v and w are nominally zero in these coordinates.
For the purpose of defining the small disturbance equations of motion, we will assume that a nominal
wings-level, equilibrium flight condition has been identified and that the stability axes serve as the bodyfixed coordinate frame. We will substitute the following values into the equations of motion
x

u
p
X
L

=
=
=
=
=
=

x0 (t) + x
0 +
u0 + u
p0 + p
X0 + X
L0 + L

v
q
Y
M

=
=
=
=
=
=

y0 (t) + y
0 +
v0 + v
q0 + q
Y0 + Y
M0 + M

w
r
Z
N

=
=
=
=
=
=

z0 (t) + z
0 +
w0 + w
r0 + r
Z0 + Z
N0 + N.

Note that, because we are considering an equilibrium motion, the nominal position (given by x, y, and z)
will change with time; we denote this dependence explicitly above. In stability axes, we have
y0 (t) 0, 0 0, 0 0, v0 0, w0 0, p0 0, q0 0, r0 0.
The nominal values of the remaining variables (for example, u and ) are generically nonzero. (Note that
the assumption 0 = 0 is entirely arbitrary because does not appear anywhere in the equations of motion.
This choice corresponds, for example, to due north flight.) In the next lecture, we will proceed with the
linearization procedure to obtain a set of first order, linear time-invariant equations which describe the
motion of a rigid airplane operating near wings-level equilibrium flight.

Lecture 14: Small Disturbance Equations of Motion

We are considering a rigid airplane with a coordinate frame fixed at the center of gravity such that the
xz-plane is a plane of symmetry. Written in these coordinates, the kinematic equations are

x
cos cos cos sin sin cos sin cos sin cos + sin sin
u
y = RIB v = cos sin cos cos + sin sin sin sin cos + sin cos sin v
z
sin
cos sin
cos cos
w

p
1 sin tan cos tan
= LIB = 0
cos
sin q
0 sin sec cos sec
r

and the dynamic equations are

X(v, , u)
sin
mv = mv + Y (v, , u) + mg cos sin
Z(v, , u)
cos cos

L(v, , u)
I = I + M (v, , u) .
N (v, , u)

yB
V0

xB
zB

Figure 1: Stability Axes.

For the purpose of defining the small disturbance equations of motion, we will assume that a nominal
wings-level, equilibrium flight condition has been identified and that the stability axes serve as the bodyfixed reference frame. We will substitute the following values into the equations of motion
x

u
p
X
L

=
=
=
=
=
=

x0 (t) + x
0 +
u0 + u
p0 + p
X0 + X
L0 + L

v
q
Y
M

=
=
=
=
=
=

y0 (t) + y
0 +
v0 + v
q0 + q
Y0 + Y
M0 + M

w
r
Z
N

=
=
=
=
=
=

z0 (t) + z
0 +
w0 + w
r0 + r
Z0 + Z
N0 + N.

In stability axes, we have

0 0, 0 0, v0 0, w0 0, p0 0, q0 0, r0 0.
The nominal values of the remaining variables (for example, u and ) are generically nonzero. (The
assumption 0 = 0 is entirely arbitrary because does not appear anywhere in the equations of motion.
This choice corresponds, for example, to due north flight.)
1

Linearized Kinematics. We will assume small perturbations so that the linearized equations accurately
approximate the nonlinear ones. To start, take the first view of linearization, as discussed in the previous
lecture, and consider only the kinematic equation for position x. Define f1 such that
x = f1 (x, y, z, , , , u, v, w, p, q, r, T, a, e, r)
= (cos cos ) u + (cos sin sin cos sin ) v + (cos sin cos + sin sin ) w

(1)

Note that f1 is the first component of the twelve-dimensional vector field f (x, u) that defines the airplane
equations of motion. To linearize, compute
f1
x
x

f1
f1
f1
f1
f1
f1
+
+
+
u +
v +
w
=
0
0
0
u 0
v 0
w 0
= u0 sin 0 + cos 0 u + sin 0 w.

x =

Alternatively, using the second view of linearization, we may use small angle approximations and neglect
products of perturbation variables. We may write
cos = cos() 1
sin = sin()
cos = cos(0 + ) = cos 0 cos sin 0 sin
cos 0 sin 0
sin = sin(0 + ) = sin 0 cos + cos 0 sin
sin 0 + cos 0
cos = cos() 1
sin = sin() .
Substituting the above approximations, along with the identities u = u0 + u, v = v, and w = w,
into (1) and neglecting perturbation terms higher than first order gives
x 0 + x (u0 + u) cos 0 u0 sin 0 + sin 0 w.
Following this approach for each of the kinematic equations, we find that the first-order approximation to
the complete set of kinematic equations is
x 0 + x = (u0 + u) cos 0 (u0 sin 0 ) + sin 0 w
y 0 + y = (u0 cos 0 ) + v
z0 + z = (u0 + u) sin 0 (u0 cos 0 ) + cos 0 w
0 + = p + tan 0 r
0 + = q
0 + = sec 0 r.
Setting the perturbation values to zero leaves the nominal equations. Subtracting these from the complete
equations then gives the perturbed equations. The resulting nominal and perturbation equations for the
vehicle kinematics are:

x 0 = u0 cos 0

x = cos 0 u (u0 sin 0 ) + sin 0 w

y 0 = 0

y = (u0 cos 0 ) + v

z0 = u0 sin 0
0 = 0
0 = 0

z = sin 0 u (u0 cos 0 ) + cos 0 w

= p + tan 0 r
= q

and

0 = 0

= sec 0 r

Linearized Dynamics. First, consider the translational dynamics. Written explicitly, we have

u
p
X(v, , u)
sin
u
m v = m v q + Y (v, , u) + mg cos sin .
Z(v, , u)
cos cos
w
w
r

Substituting nominal-plus-perturbed values for each component of the system state gives

u
u0
u
p
X0 + X
sin(0 + )
m v = m 0 + v q + Y0 + Y + mg cos(0 + ) sin .
w
0
w
r
Z0 + Z
cos(0 + ) cos
Ignoring higher order perturbation terms leaves the following first order approximate equations:
mu = X0 + X mg(sin 0 + cos 0 )
mv = mu0 r + Y0 + Y + mg cos 0
mw = mu0 q + Z0 + Z + mg(cos 0 sin 0 )
Turning next

Ix
0
Ixz

to the rotational dynamics, we have

p
p
L(v, , u)
p
Ix
0 Ixz
0 Ixz
Iy
0 q q + M (v, , u) .
Iy
0 q = 0
r
r
N (v, , u)
r
Ixz 0
Iz
0
Iz

Substituting nominal-plus-perturbed values for

p
Ix
Ix
0 Ixz

0
q
0
Iy
0
=
r
Ixz
Ixz 0
Iz

each component of the system state gives

p
p
L0 + L
0 Ixz
Iy
0 q q + M0 + M .
r
r
N0 + N
0
Iz

Ignoring higher order perturbation terms leaves the following first order approximate equations:
Ix p Ixz r = L0 + L
Iy q = M0 + M
Iz r Ixz p = N0 + N.

Setting the perturbation values to zero leaves the nominal equations. Subtracting these from the complete
equations then gives the perturbed equations. The resulting nominal and perturbation equations for the
vehicle dynamics are:
0 = X0 mg sin 0

mu = X mg cos 0

0 = Y0

mv = mu0 r + Y + mg cos 0

0 = Z0 + mg cos 0

mw = mu0 q + Z mg sin 0

and

0 = L0

Ix p Ixz r = L

0 = M0

Iy q = M

0 = N0 .

Iz r Ixz p = N.
3

The challenge at this point is to express the aerodynamic forces and moments which result from perturbations from the nominal flight condition. The aerodynamic forces and moments developed over a moving
rigid body are functions of the bodys geometry, the local density of air, the control surface deflections, the
air-relative velocity, and, in general, the entire time history of the bodys motion. In practice, however,
the following assumptions are well-justified for most flight conditions:
The asymmetric force Y and moments L, and N are, to first order, independent of the symmetric
state and control variables. That is, the derivatives of Y , L, and N with respect to , u, w, q, T ,
and e are identically zero.
The symmetric forces X and Z and moment M are, to first order, independent of the asymmetric
state and control variables. That is, the derivatives of X, Z, and M with respect to , , v, p, r, a
and r are identically zero.
The only dependence of the aerodynamic forces and moments on acceleration is the dependence of
Z and M on w.
(Because w V0 = u0 ,
these terms essentially account for the dependence
of lift and moment on the rate of change of angle of attack.)
The force X is independent of pitch rate q.
Density remains constant over the range of perturbations.
Define the dimensional derivatives

X
,
X() =
() 0

Also, define

M
L
,
M() =
,
=
() 0
() 0

(For example, let Xu =

L()

X
u .)

Y()

Y
=
,
() 0

and

and

Z()

N()

Z
=
.
() 0

N
=
.
() 0

With the assumptions and definitions above, we may rewrite the linearized dynamic equations as
mu = (Xu u + Xw w + Xe e + XT T ) mg cos 0
mv = mu0 r + (Yv v + Yp p + Yr r + Ya a + Yr r) + mg cos 0
mw = mu0 q + (Zu u + Zw w + Zw w + Zq q + Ze e + ZT T ) mg sin 0
Ix p Ixz r = (Lv v + Lp p + Lr r + La a + Lr r)
Iy q = (Mu u + Mw w + Mw w + Mq q + Me e + MT T )
Iz r Ixz p = (Nv v + Np p + Nr r + Na a + Nr r) .
Per our assumptions, the asymmetric state and control variables do not appear in the equations for u,

w,
and q.
Conversely, the symmetric state and control variables do not appear in the equations for
v,
p,
and r.

Lecture 15: Longitudinal and Lateral-Directional Dynamics

Recall, from the last lecture, that the perturbation equations for a rigid aircraft linearized about wings-level
equilibrium flight are
x = cos 0 u (u0 sin 0 ) + sin 0 w
y = (u0 cos 0 ) + v

z = sin 0 u (u0 cos 0 ) + cos 0 w

= p + tan 0 r
= q
= sec 0 r
mu = (Xu u + Xw w + Xe e + XT T ) mg cos 0

mv = mu0 r + (Yv v + Yp p + Yr r + Ya a + Yr r) + mg cos 0

mw = mu0 q + (Zu u + Zw w + Zw w + Zq q + Ze e + ZT T ) mg sin 0

Ix p Ixz r = (Lv v + Lp p + Lr r + La a + Lr r)

Iy q = (Mu u + Mw w + Mw w + Mq q + Me e + MT T )

Iz r Ixz p = (Nv v + Np p + Nr r + Na a + Nr r) .
Per our assumptions, the asymmetric state and control variables do not appear in the equations for u,

w,
and q.
Conversely, the symmetric state and control variables do not appear in the equations for
v,
p,
and r.

Notice that the forcing on the right-hand side of the equations for mw and Iy q include an aerodynamic
contribution due to w.
Thus, to obtain a simple expression for w,
one must bring the right-hand side
term involving w across to the left. To obtain a decoupled equation for q,
one must substitute the
resulting expression for w on the right-hand side. Also notice the inertial coupling between the roll and
yaw dynamic equations. One must decouple these two equations for p and r in order to express the
linearized dynamics in their simplest form.
Linearized Longitudinal Equations. We may partition the state and control vector into longitudinal
(or symmetric) state and control vectors and lateral-directional (or asymmetric) state and control vectors.
The longitudinal state and control vectors are:

x
z

u
e

.
xL =
and
uL =

T
w
q

The linearized longitudinal equations take the form

x L = AL xL + B L uL

where

and

AL =
0

0
0
0

cos 0
sin 0
1
m Xu

sin 0
cos 0
1
m Xw

0
0
0

Zu
mZw

Zw
mZw

(Zq +mu0 )
mZw

1
Iy

Mu +

Mw Zu
mZw

1
Iy

Mw Zw
mZw

1
Iy

0
0
1
m Xe

BL =

Mw +

Me +

Mw (Zq +mu0 )
mZw

Mq +

1
0
0
1
m XT

Ze
mZw
1
Iy

u0 sin 0
u0 cos 0
g cos 0

Mw Ze
mZw

1
Iy

1
Iy

mg sin 0

mZw

Mw (mg sin 0 )

mZw
0

ZT

mZw

w
ZT

MT + M
mZ
w

The matrices are partitioned to emphasize that the state variables u, w, q, and evolve independently of the state variables x and z. The x and z position play no role in the dynamic equations (aside
from a second order dependence of air density on altitude). In studying aircraft stability and control, one
typically ignores horizontal and vertical position.
The stick-fixed longitudinal response (i.e., the response with e = 0 and T = 0) can be understood
as the superposition of two oscillatory modes of motion. One mode, referred to as the phugoid mode,
corresponds to a severely underdamped oscillation with a fairly large damped natural period. The other
mode, referred to as the short period mode, corresponds to a well-damped oscillation with a relatively small
damped natural period. Stability of these modes of motion determines longitudinal dynamic stability.

u(t) (ft/s)

220
215
210
205

10

10

10

10

w(t) (ft/s)

30
20
10
0

q(t) (deg/s)

40
20
0
20

(t) (deg)

6
4
2
0

t (s)

Figure 1: Short period contribution to the longitudinal response.

2

u(t) (ft/s)

220

210

200

20

40

60

80

100

120

140

160

180

20

40

60

80

100

120

140

160

180

20

40

60

80

100

120

140

160

180

20

40

60

80

100

120

140

160

180

w(t) (ft/s)

40
20
0
20

q(t) (deg/s)

40
20
0
20

(t) (deg)

10
5
0
5

t (s)

Figure 2: Long period (phugoid) contribution to the longitudinal response.

Figure 1 shows a longitudinal time history for a particular short takeoff and landing (STOL) airplane in
response to an impulsive pitch disturbance (modeled as a non-zero initial pitch rate). Notice that w, which
is proportional to the angle of attack for small perturbations, converges to near zero in only a few seconds.
This quick convergence of the angle of attack is characteristic of a stable short period mode. Figure 2
shows a longer record of the time history for the same perturbation. Notice the much slower convergence
of speed and pitch angle. This long period, lightly damped oscillation is characteristic of a stable long
period, or phugoid, mode.
Aside: Approximating the Phugoid Mode. In W. F. Lanchesters original 1908 investigation of the
phugoid mode, he considered the nonlinear dynamic equations for an airplane in longitudinal flight. He
made three important assumptions:
The angle of attack remains identically zero,
thrust exactly balances drag throughout the motion, and
the lift coefficient is constant and equal to nominal weight coefficient.
Let 0 = 0. Moreover, assume that Zq and Zw are negligibly small. In the case of small perturbations, the
assumption that 0 means that
w 0

w 0 =

1
Zu u + u0 q
m

q =

Zu
u.
mu0

The linearized equation for w is trivial, under the given assumptions. The equation for q is simply a
scaling of the equation for u,
under the given assumptions. Both equations may therefore be ignored.
The remaining two first order equations which approximate the phugoid mode are therefore
u =

1
Xu u g
m
3

 = q
Zu
=
u.
mu0
Solving the latter equation for u and substituting into the former equation gives a single second order
ODE:
Xu gZu


= 0.
m
mu0
This equation represents a damped linear oscillator (e.g., a mass-spring-damper system). Assuming that
the discriminant

gZu
Xu 2
4

m
mu0

is negative, the system is underdamped and we may compute the approximate phugoid natural frequency
and damping ratio:
r
Zu g
Xu
nP
and
P
.
mu0
2mnP
Although we will not discuss methods for computing dimensional stability derivatives, such as Xu and Zu ,
until the next lecture, one may show that in the case of a jet, where thrust is independent of speed,
g
1 1
nP 2
and
P
.
u0
2 L/D
A higher nominal speed thus corresponds to a lower phugoid frequency (that is, a longer period of oscillation). A higher lift-to-drag ratio corresponds to lower damping. A derivation of this incompressible
phugoid mode approximation, and a very nice discussion of Lanchesters original phugoid theory, are given
in [1].
We will revisit this approximation for the phugoid mode, and develop another for the short period mode,
after we have discussed methods for computing dimensional stability derivatives.
Linearized Lateral-Directional Equations. The lateral-directional state and control vectors are:

v
a

xLD =
and
uLD =
.

r
p
r

Let

2
= Ix Iz Ixz
.

The linearized lateral-directional equations take the form

x LD = ALD xLD + B LD uLD

where

ALD

0 u0 cos 0
0
0

0
0

= 0
0

0
0

0
0

g cos 0

1
1
1

(I
L
+
I
N
)
(I
L
+
I
N
)
(I
L
+
I
N
)
0
z
v
xz
v
z
p
xz
p
z
r
xz
r

1
1
1
(I
L
+
I
N
)
(I
L
+
I
N
)
(I
L
+
I
N
)
0
xz
v
x
v
xz
p
x
p
xz
r
x
r

1
0
1
m Yv

0
0
1
m Yp

0
sec 0
1
m Yr u0

tan 0

and

B LD

0
0
1
m Ya

=
1 (Iz La + Ixz Na )

1
(Ixz La + Ix Na )
0

0
0
1
m Yr

.
1
(I
L
+
I
N
)
z r
xz r

1
(I
L
+
I
N
)
xz r
x r

The matrices are partitioned to emphasize that the state variables v, p, r, and evolve independently of the state variable y and . That is, y position and heading play no role in the dynamic
equations. In studying aircraft stability and control, one typically ignores lateral position and heading.
If the stability axes happen to coincide with the principal axes of inertia, then Ixz = 0. In this case, the
linearized dynamic equations (less the equations for y and ) simplify to

v

d
p =

r
dt

1
m Yv
1
Ix Lv
1
Iz Nv

1
m Yp
1
Ix Lp
1
Iz Np

1
m Yr u0
1
Ix Lr
1
Iz Nr

tan 0

g cos 0
v
p
0

r
0

1
m Ya
1
Ix La
1
Iz Na

1
m Yr
1
Ix Lr
1
Iz Nr

a
r

Notice that, while the inertial coupling between roll rate and yaw rate vanishes in this case, the aerodynamic
coupling does not; terms like Lr , Np , Lr and Na are not necessarily zero.
We will see shortly that the stick-fixed lateral-directional response (i.e., the response with a = 0 and
r = 0) can be understood as the superposition of two non-oscillatory modes of motion (the spiral and
roll modes) and an oscillatory mode (the Dutch roll mode). Stability of these modes of motion determines
lateral-directional dynamic stability.
Recap. At this point, we have obtained twelve first order, linear time-invariant ordinary differential equations describing the motion of an airplane in response to small perturbations from wings-level equilibrium
flight. Under certain assumptions, these twelve equations decouple into two sets of six ODEs. Moreover,
in considering dynamic stability, we may ignore the variables x, y, z and and study only two
sets of four ODEs: the (reduced) longitudinal equations and the (reduced) lateral-directional equations.
First, though, we must understand how to relate estimates of aerodynamic properties such as lift, drag
and pitching moment to dimensional stability derivatives such as Xu , Zw and Mq .

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.

Lecture 16: Longitudinal Stability Derivatives

The term stability derivative arises from the linearization of the aerodynamic terms in the nonlinear
dynamic equations. The connection to stability comes from the role these terms play in the stability
of wings-level equilibrium flight. As an example, recall that the linearized aerodynamic force in the Xdirection is
X = X0 + X = X0 + (Xu u + Xw w + Xe e + XT T )
where

X
Xu =
u 0

and so on. The subscript 0 indicates that the partial derivative is evaluated at the nominal, equilibrium
flight condition. The term Xu , and similar terms, are sometimes referred to as the dimensional stability
derivatives because these terms have physical dimensions. The shorter phrase stability derivative is
typically reserved for the nondimensional form of these terms. We now turn to the problem of relating
stability derivatives to the physical properties of a given aircraft.
The u-derivatives: Xu , Zu , and Mu . We first compute the longitudinal stability derivatives with respect
to u. Aerodynamic forces, such as X, are conventionally represented in coefficient form:

1 2
V
SCX
X=
2
where V 2 = u2 + v 2 + w2 . Thus, an aerodynamic force depends on u in two ways: through the dynamic
pressure and through the (not necessarily constant) aerodynamic coefficient. In the case of the force X,
we have

CX
1 2
V
S
Xu = uSCX +
2
u 0

1 2
CX
= u0 SCX0 +
u S
2 0
u 0

1 2
2CX0
CX
=
u S
+
.
2 0
u0
u
0

Define the nondimensional term

CXu
With this definition, we may write
Xu

CX
CX
=
= u0
.
(u/u0 ) 0
u 0

1 2
2CX0
CXu
=
u S
+
2 0
u0
u0

1
=
u0 S (2CX0 + CXu ) .
2

(1)

Similarly, we have
Zu =

1
u0 S (2CZ0 + CZu )
2

where

Considering the aerodynamic pitching moment

M=

1 2
V
S
cCm ,
2
1

CZu

CZ
=
.
(u/u0 ) 0

(2)

Nominal Flight

Longitudinally Perturbed Flight

V
V0

xB

xB

xI

xI

zB

zB

zI

zI

Figure 1: Stability Axes.

we also have
Mu =

1
u0 S
c (2Cm0 + Cmu )
2

where

Cmu

Cm
=
.
(u/u0 ) 0

(3)

We must next relate the body frame forces X and Z to the more familiar aerodynamic forces lift L, drag
D, and thrust T . Recall that we are working in stability axes, a body-fixed coordinate frame defined such
that the longitudinal (xB ) axis is aligned with the velocity vector when the airplane is in its nominal,
wings-level equilibrium condition. Of course, this frame moves with the aircraft when the flight condition
is perturbed.
Consider first the case of nominal flight. For simplicity, we also assume that thrust and drag are aligned
with the xB axis and that lift acts in the negative z B direction.1 In nominal flight, we have
X0 = T0 D0 = mg sin 0
Z0 = L0

= mg cos 0

M0 = 0
Nondimensionalizing gives
CX0 = CT0 CD0 = CW0 sin 0
CZ0 = CL0

= CW0 cos 0

Cm0 = 0
(For convenience, we have assumed that the reference area for the thrust coefficient is the wing planform
area Sp = S.)
Now suppose we allow small perturbations in speed and angle of attack; see Figure 1. In the general case,
we have
X = D cos + L sin + T D + L + T
Z = D sin L cos

D L.

Dividing through by dynamic pressure and area gives

CX

= CD + CL + CT

CZ

= CD CL .

1
This is not an especially good assumption, but it simplifies the discussion. The more general case simply involves some
geometry.

The (nondimensional) stability derivative CXu is obtained by computing

(CD + CL + CT )
CXu = u0
u
0

CD
CL

CT
= u0
+ u0
+ u0 CL
+ u0
.
u
u
u
u 0
Recall that
tan =
so
sec2
Solving for

w
,
u

w
1
= 2 = tan .
u
u
u

gives
1
1

= sin cos = sin 2.

u
u
2u

We therefore have
CXu

CD
CL
CT
0
= u0
+ u0
+ CL sin 2 + u0
u
u
2u
u

CD
CT
= u0
+ u0
.
u 0
u 0

Similarly, we find that

CZu

(CD CL )
u
0

CL
CD
0
u0 CD sin 2 u0
= u0
u
2u
u 0

CL
.
= u0
u 0
=

u0

Defining
CDu
CLu
CTu
we have

CD
CD
=
= u0
(u/u0 ) 0
u 0

CL
CL
=
= u0
(u/u0 ) 0
u 0

CT
CT
=
= u0
,
(u/u0 ) 0
u 0
CXu

= CDu + CTu

CZu

= CLu .

Summarizing to this point:


1
u0 S [2 (CD0 + CT0 ) + (CDu + CTu )]
Xu =
2

1
=
u0 S [2CW0 sin 0 + (CDu + CTu )]
2

1
Zu =
u0 S (2CL0 CLu )
2

1
u0 S (2CW0 cos 0 CLu )
=
2

1
Mu =
u0 S
c (Cmu )
2

It still remains for us to determine how CD , CT , and CL depend on u. In [1], the authors claim that the
dependence arises primarily through three mechanisms: the type of propulsion, compressibility effects, and
aeroelastic effects. We will consider only the first two.
First, consider the thrust coefficient
T
.
2 S
V
2

CT = 1

We compute
CTu

"

1
T
2T
V
= u0 1 2
1 3
S u
S u
2 V
2 V

1
T

= 1
2CT0 .
u 0
u
S
0
2

The variation of thrust with u depends on the type of engine and/or mode of flight. For unpowered flight
(e.g., for a sailplane), thrust is identically zero so that CTu = 0. For constant thrust propulsion (e.g., for a
jet aircraft in cruising flight), thrust does not vary with speed so that CTu = 2CT0 . For constant power
propulsion (e.g., for propeller driven aircraft), the propulsive power T u does not vary with speed. In this
case, we compute

T
T
1

(T u) = 0 =
u+T

= T0
u
u
u 0
u0

so that CTu = 3CT0 .

Propulsion Type
Zero Thrust
Constant Thrust
Constant Power

CTu
0
2CT0
3CT0

The variation of CD , CL , and Cm with u is primarily due to compressibility effects which are characterized
by the Mach number Ma. Consider a general nondimensional coefficient CA which depends on u through
its dependence on Mach number. Since
V
Ma = ,
a
where a is the speed of sound (assumed to remain constant), we may compute

CA Ma
CA 1 1 2u
CA
1
CA
=
=
=
.

u 0
Ma u 0
Ma a 2 V
Ma 0 a
0
4

Multiplying through by u0 to nondimensionalize, we obtain

CA
CA
CA
CAu = u0
=
=
Ma .
u 0
(u/u0 ) 0
Ma
0

In the case of the longitudinal stability derivatives, and omitting the subscript 0, we have

CDu =

CD
Ma
Ma

CLu =

CL
Ma
Ma

and

Cmu =

Cm
Ma.
Ma

Explicit formulas for the partial derivatives on the right can be obtained from inviscid flow theory; see [1],
for example.
It should be pointed out, once again, that aeroelasticity can have an important effect on stability derivatives.
The deformations in the fuselage and lifting surfaces which occur as a consequence of changes in speed,
can dramatically affect the aircrafts dynamics and stability. These considerations are, however, beyond
the scope of this course.
The w-derivatives. Equivalently, the -derivatives: X , Z , and M . Recall that the state
variables of interest in studying longitudinal dynamics are u, w, q, and . (We ignore x and z, which are
irrelevant to the question of dynamic stability.) Logically, it would seem that we should next study the
dependence of X, Z, and M on the plunge rate w. Equivalently, we may study the dependence of X, Z,
and M on the angle of attack . Recall once again that
tan =
Differentiating with respect to w gives
sec2

w
.
u

=
w
u

or

1
= cos2 .
w
u
For small perturbations from nominal flight, we have

.
w
u0

Thus, for the linearized equations at least, we may write

= u0
.

w
Stability derivatives may appear either with respect to or with respect to w; for small perturbations, the
two are directly proportional. In particular, we have
X = u 0 Xw ,
Define the nondimensional coefficients

CX
,
CX =
0

Z = u0 Zw ,

CZ

and recall the force coefficient approximations

CZ
=
,
0

and

and

CX

= CD + CL + CT

CZ

= CD CL .
5

M = u 0 Mw .

Cm

Cm
=
0

One may compute

CX

[CD + CL + CT ]0

CT
= CD + CL + CL +

= CD + CL0 ,

where we assume that thrust does not vary with angle of attack.
Turning to the Z force coefficient, we find that

CD
+ CL
CZ = CD +

0
= (CD0 + CL ) .
Summarizing the force coefficients:

CX

= CD + CL0

CZ

= CD0 CL

Of course, we have already spent a great deal of time discussing the term Cm .

Cm

= Cmw + Cmt + Cmf + Cmp

d
+ Cmf + Cmp
= CLw (h hacw ) VH CLt 1
d
= CL (h hn ) .

The q-derivatives: Zq and Mq . We have already considered stability derivatives related to pitch rate in
a previous lecture. Those results are restated here. Define the nondimensional terms
CL
2u0 CL
=
q
c q
Cm
2u0 Cm
=
=
.
q
c q

CLq =
Cmq

Considering the increment in lift generated by the tail due to a nonzero pitch rate, one finds that

CLq

= 2CLt
VH

Cmq

lt
= 2kCLt
VH ,
c

where k (usually chosen as 1.1) corrects for the contribution to pitch damping from the wing. Referring to the force coefficient relations at the beginning of this lecture, one sees that CZq = CLq . Redimensionalizing, we have

Zq =
Mq =

1 2
c
CZ
u S
2u0 q 2 0

c
1 2
Cmq
u S
c.
2u0
2 0

The w-derivatives.

Equivalently, the -derivatives:

Z and M . Recall that, for small angles

of attack, we may change variables by replacing w everywhere with u0 . Similarly, we may replace w
everywhere with u0 .
Stability derivatives with respect to w appeared only in the Z-force and M -moment
equations. These terms account for the fact that there is a time delay before any change in downwash
generated by the wing is felt at the tail. That delay is approximately t = ult0 . The downwash at the
tail at an instant t corresponds to the wing angle of attack at a previous instant t t. Therefore the
deviation of the downwash from its nominal value at an instant t corresponds to the deviation of the wing
angle of attack from its nominal value at t t. Treating as an explicit function of time, we have
= (t) (t t).
Of course, is actually a function of (which is, itself, a function of time) so that = ((t)). We therefore
have
d
d
=
((t) (t t)).
=
d
d
t
Multiplying the right-hand side by t
and assuming that ult0 is small, we may write
d

d
d
t

d
d lt
.
d u0

=
=
=

This is precisely the change in the tail angle of attack due to :

t = =

d lt
.
d u0

Following the same procedure as we did earlier in the course, when we computed CZq and Cmq , we find
that
d
CZ = 2CLt VH
d
and
lt d
Cm = 2CLt VH
.
c d
Re-dimensionalizing, we have:

Z =
=
M =
=

c
1 2
CZ
V
S
2u0 2

d
1 2
c
2CLt VH
V
S
2u0
d
2

1 2
c
Cm
V
S
c
2u0
2

c
lt d
1 2
2CLt VH
V
S
c
2u0
c d
2
7

The force and moment derivatives Zw and Mw can be computed using the approximate relation w = u0 :

=
= u0
.

(w/u
0)
w
Thus,
Zw =

1
Z
u0

and

Mw =

1
M .
u0

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.

Lecture 17: Lateral-Directional Stability Derivatives

The v-derivatives. Equivalently, the -derivatives: Y , L , and N . We have assumed that the
asymmetric force Y and moments L and N depend only on the asymmetric state and control variables v,
p, r, a, and r. We first consider the dependence of Y , L, and N on v. Equivalently, we may consider
their dependence on . Recall that
v
sin = .
V
Differentiating with respect to v gives
cos

1
v2
1
1

=
3 =
1 sin2 = cos2 .
v
V
V
V
V

For small perturbations from nominal flight, we have

v
u0
Thus, and v are directly proportional when operating near the nominal flight condition.
Consider first the term Cy . The primary contributor to this term is the vertical tail, for which we may
write

1 2
V
Svt .
Yvt = CLvt ( + ())
2
(Recall that any change in tail efficiency due to differences in dynamic pressure at the tail
and in the
free stream are assumed to be represented in the lift curve slope CLvt .) Normalizing by 21 V 2 S and
differentiating with respect to gives

d Svt
.
CY = CLvt 1 +
d
S
The term Cn was discussed in a previous lecture. It depends on the wing-body and the vertical tail:

d
Cn = Cnwb +
VV CLvt 1
.
d
The term
VV above is the vertical tail volume ratio.
As discussed in Lecture 10, the term Cl owes primarily to the dihedral effect. Having computed the
nondimensional stability coefficients CY , Cl , and Cn , the dimensional stability derivatives are

1 2
1 2
1 2
Y = CY
V
S,
L = Cl
V
Sb,
and
N = Cn
V
Sb.
2
2
2
The p-derivatives: Yp , Lp , and Np . When an airplane in equilibrium flight experiences a perturbation
which results in a nonzero roll rate p, the result is a change in lift generated by each of the lifting
surfaces. The wing, for example, experiences a change in lift profile such that the downward moving wing
experiences an increase in lift and the upward moving wing experiences a decrease in lift. (The increment
in lift varies linearly from wingtip to wingtip.) The same is true for the horizontal tail. Also, an increment
in angle of attack is induced at the vertical tail. All of these effects contribute to the three roll rate stability
derivatives
2u0 Cl
2u0 Cn
2u0 CY
,
Clp =
,
and
Cnp =
.
CYp =
b p
b p
b p
The only effect which can be easily estimated, however, is that due to the vertical tail.
1

(pb/2)
u0

u0

yB

ac vt

xB

zB
(p|z acvt|)
u0
(pb/2)

Figure 1: Effect of roll rate.

Etkin and Reid [1] suggest that the side force due to roll rate can often be neglected. Primary contributors
to CYp are the vertical tail and the wing. The vertical tail contribution can be estimated by recognizing
that the non-zero roll rate induces a sideslip angle
vt =

pzacvt
u0

at the vertical tail. (While the sideslip angle actually varies over the span of the vertical tail, we assume
that the variations cancel out so that the value above is a good approximation.) For positive roll rate p > 0
and an above-board vertical tail, the sideslip angle vt is positive. Consequently, the vertical tail exerts
a side force in the negative y B direction,
a negative roll moment, which opposes the roll rate, and
a positive yaw moment.
While the wing contribution is not easily estimated, it is fairly easy to understand physically. Because
the downward moving wing generates more lift and the upward moving wing generates less, the primary
effect is a roll moment which opposes the roll rate. Thus Clp is often called the roll damping derivative.
Also, because drag is proportional to the square of lift, the downward moving wing experiences greater
drag resulting in a yaw moment which is positive for positive roll rate.
Techniques for estimating Clp and Cnp are given in Appendix B.10 of [1]. Having computed the nondimensional stability coefficients CYp , Clp , and Cnp , the dimensional stability derivatives are

1 2
b
1 2
b
1 2
b
V
S,
Lp = Clp
V
Sb,
and
Np = Cnp
V
Sb.
Yp = CYp
2u0
2
2u0
2
2u0
2
The r-derivatives: Yr , Lr , and Nr . The asymmetric force and moments vary with yaw rate through
two primary effects. First, and more obviously, the vertical tail experiences a change in angle of attack due
to the yaw rate; a positive yaw rate results in a negative sideslip angle, as shown in Figure 2. The effect
of the resulting side force generated by the tail is easily estimated. Second, the lift generated by the wings
varies due to the relative change in forward speed at different points along the span. This contribution is
not as easy to estimate, though it is fairly easy to understand. For a positive yaw rate, the airspeed of
2

the left wing increases and the airspeed of the right wing decreases. The result is an increase in lift and
drag on the left wing and a decrease in lift and drag on the right. These changes result in a negative yaw
moment, which opposes the positive yaw rate, and a positive roll moment. All of these effects contribute
to the three yaw rate stability derivatives
CYr =

2u0 CY
,
b r

Clr =

2u0 Cl
,
b r

Cnr =

2u0 Cn
.
b r

The only effect which can be easily estimated, however, is that of the vertical tail.
To estimate the effect of the vertical tail on the stability derivatives CYr , Clr , and Cnr , note that a side
force Y results:

1 2
V
Svt ,
Y = CLvt
2
where
=
Normalizing by

2 V

rlvt
.
u0

S gives
CY = CLvt

rlvt Svt

u0
S

Differentiating with respect to r and re-normalizing gives

2u0
lvt Svt
CYr =
CLvt
b
u0
S
= 2CLvt
Vv
u 0 - (rb/2)

yB
xB

acvt

u0
zB

r
u 0 + (rb/2)

Figure 2: Effect of roll rate.

The side force Y resulting from a non-zero yaw rate generates a yaw moment
N

= Y lvt

1 2
V
Svt lvt
= CLvt
2

1 2
rlvt
V
Svt lvt .
= CLvt
u0
2

rlvt

Note that the tail contribution to yaw moment

for a positive yaw rate. The yaw moment due
is negative

to the tail opposes yaw rate. Normalizing by 21 V 2 Sb gives

rlvt

Vv
Cn = CLvt
u0
Differentiating with respect to r and re-normalizing gives

lvt
2u0
CLvt

Vv
Cnr =
b
u0
lvt
= 2CLvt
Vv .
b
The vertical tails contribution to roll moment arises because the tail aerodynamic center is a signed
distance zacvt below the xB axis. (The distance zacvt is negative for a top-mounted vertical tail.) The
tail contribution to the two moment coefficients Cnr and Clr are directly proportional; neglecting other
contributions, we have
Clr

zacvt
Cnr
lvt

zacvt
lvt
=
2CLvt
Vv
.
lvt
b
=

Note that Clr is positive for a top-mounted vertical tail. This means that the tail will provide a positive
roll moment in response to a positive yaw rate. The roll moment contribution due to the vertical tail acts
in the same direction as that of the wing.
Having computed the nondimensional stability coefficients CYr , Clr , and Cnr , the dimensional stability
derivatives are

1 2
b
1 2
b
1 2
b
V
S,
Lr = Clr
V
Sb,
and
Nr = Cnr
V
Sb.
Yr = CYr
2u0
2
2u0
2
2u0
2

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.

Lecture 18: LTI Systems in State-Space Form

Consider the homogeneous LTI system
x = Ax,

x(t0 ) = x0 .

(1)

The dynamics of an LTI system are invariant under shifts in the initial time. Without loss of generality, we
may assume that t0 = 0 because if t0 were other than zero, we could simply shift the computed response
by that amount of time.
Using Laplace transforms, one may easily compute
X(s) = (sI A)1 x0
1
Adj (sI A) x0 ,
=
a(s)
where
a(s) = det (sI A)

and where Adj (sI A) is the adjugate matrix, which is the transpose of the matrix signed cofactors of
(sI A). The nth order polynomial a(s) is the characteristic polynomial for the matrix A; its roots are
eigenvalues of A. Pursuing the solution further by taking the inverse Laplace transform of each component
of X(s), one would find that x(t) is a sum of exponentials whose arguments are the roots of a(s), i.e., the
eigenvalues of A.
To further explore the linear algebraic properties of this LTI system, suppose that there exist n linearly
independent eigenvectors v i corresponding to n eigenvalues i , that is, the n roots of a(s). (If there are
repeated eigenvalues, then there may not exist n linearly independent eigenvectors. Ignore this case, for
now.)
Proposition: The solution to the homogeneous system (1) with x0 = v i is
v i ei t .
Proof:

d i t
vie
= i v i ei t = A v i ei t .
dt

Suppose that 1 and its corresponding eigenvector v 1 are real. Then, in the n-dimensional state space, the
trajectory either
follows the direction of the eigenvector v 1 in to the origin, if 1 < 0,
follows the direction of the eigenvector v 1 out to infinity, if 1 > 0, or
remains at v 1 , if 1 = 0.
In any case, we say that only the mode (or eigenmode) corresponding to 1 is excited by the initial
state.
By assumption, the eigenvectors v i of (1) are independent. Therefore, one may express any initial condition
x0 in terms of these vectors:
x0 =

n
X

ci v i

i=1

= [v 1 , , v n ] c
1

where the coefficients ci are constant and where c = [c1 , , cn ]T . Moreover, the set
n
o
v 1 e1 t , , v n en t

forms a basis of homogeneous solutions from which any solution x(t) to (1) can be formed. Define the
matrix
h
i
V (t) = v 1 e1 t , , v n en t .
Then

x(t) = c1 v 1 e1 t + + cn v n en t
= V (t)c

= V (t)V (0)1 x0 .
Aside: Define the nonsingular modal matrix T = [v 1 , , v n ] and the diagonal matrix = diag(1 , , n ).
Notice, from the definition of eigenvalues and eigenvectors, that
AT = T .
Using T , one may transform the state equations (1) into modal coordinates
z = T 1 x.
The transformed state equations are
z = z,

z(t0 ) = T 1 x0 .

Notice that these are simply n decoupled, first order, LTI equations.
So far, we have ignored the question of what happens when i is complex. Suppose that 1 = + i
1 = i. It is easily proved that v 1 must be complex, say v 1 = a + ib, and that
and that 2 =
1 = a ib is its complex conjugate.
v2 = v
We previously considered what happens when the initial state is an eigenvector. Obviously we can not
have a complex initial state, so here we consider what happens when the initial state is either a or b, both
of which are real vectors. First, suppose the initial state is

1/2
1/2

x0 = a = (v 1 + v 2 ) = V (0) 0 .
..
2
.
0

Then, from our previous discussion we have

1/2
1/2

x(t) = V (t)V (0)1 x0 = V (t) 0

..
.
0

1
(a + ib)e(+i)t + (a ib)e(i)t
2

1 t
e
(a + ib)eit + (a ib)eit
2

1 it
1 it
t
it
it
= e
a e +e
b
e e
2
2i
= et (a cos t b sin t) ,
=

which is real-valued, as x(t) must be.

Similarly, one may show that the zero-input response to the initial condition x0 = b is
x(t) = et (a sin t + b cos t) .
An interesting observation is that
A[a, b] = [a, b]

While it is also true that

A[v 1 , v 2 ] = [v 1 , v 2 ]

1 0
0 2

the former expression only involves real-valued elements and may therefore be more convenient to work
T 1 where the columns
with. Suppose we use this observation to define a different decomposition A = T
of T are either
the real eigenvectors v i , corresponding to real eigenvalues i , or
the real and imaginary component vectors (a and b) of complex conjugate pairs of eigenvectors
corresponding to complex conjugate pairs of eigenvalues.
For example, suppose that the 1 and 2 are complex conjugates while the remaining n 2 eigenvalues are
real, with corresponding real, independent eigenvectors v i . Then defining the real-valued transformation
matrix
T = [a, b, v 3 , . . . , v n ]
gives the the state equations

z1

z 2

z 3

..

.
zn

= T
written in the quasi-modal coordinates z

3 0
=

.. . .
..
0

.
. .
0 n

x:

z1
z2
z3
..
.
zn

The system response is

cos t sin t
z1
t
0

e
sin t cos t
z2

3
z3
e

=
z0 .

..
..
..
..
0

.
.
.
.

t
n
zn
0 e

When there are repeated real or complex conjugate eigenvalues, things can become slightly more complicated. In this case, there may not be n linearly independent eigenvalues, meaning one cannot construct
the nonsingular modal matrix T . However, the ideas discussed here can be extended using a device known
as Jordan form.
To better understand the role of eigenvalues and eigenvectors in the initial condition response of a linear,
time-invariant system of equations, it is helpful to consider the special case of planar systems. Each of the
following examples involves the LTI system
x = Ax
3

where x has dimension two.

Example 1.
A=

1 3
3 1

By definition, an eigenvalue of a matrix A is a (possibly complex) scalar for which there exists a nonzero
vector v, called an eigenvector, satisfying Av = v. Since (I A)v = 0 and v is nonzero, the matrix
I A has a nontrivial null space which contains the vector v. It follows that I A is not full rank and
its determinant is zero. The eigenvalues of A are therefore the scalar values for which det(I A) = 0;
that is, the eigenvalues are the roots of

1
3
det(I A) = det
= ( 1)2 9
3
1
= 2 2 8

= ( 4)( + 2).

The eigenvalues are thus 2 and 4. To compute the eigenvectors, we must determine the null space of
I A for each of the two eigenvalues. First, we must find v 1 such that

3 3
0 = (2I A)v 1 =
v1.
3 3
Of course, the equations are redundant because the matrix is rank deficient. That is precisely how an
eigenvalue is defined! One choice of eigenvector (there are infinitely many choices) is obtained by setting
the first component v 11 = 1 and computing the second, which gives v 12 = 1 or

1
v1 =
.
1
Next, we must find v 2 such that
0 = (4I A)v 2 =

3 3
3 3

v2.

Proceeding as before, we find

v2 =

1
1

Forming the transformation matrix

T = [v 1 , v 2 ] =

1 1
1 1

and defining
z = T 1 x,
we obtain the equations of motion in modal coordinates

2 0
z = z =
z.
0 4
If z 0 = [1, 0]T , which corresponds to x0 = v 1 , then only the first mode is excited and the response is
2t

e
1
2t
z(t) =
or x(t) = T z(t) = e T
= v 1 e1 t .
0
0
4

10

z2

x2

10

10
10

0
x1

10
10

10

0
z1

10

Figure 1: Phase portrait for Example #1. Original coordinates (left) and quasi-modal coordinates (right).
If z 0 = [0, 1]T , which corresponds to x0 = v 2 , then only the second mode is excited and the response is

0
0
4t
z(t) =
or x(t) = T z(t) = e T
= v 2 e2 t .
e4t
1
More generally, the response will be a linear combination of these two responses as depicted in the subplot
to the right in Figure 1. The plot shows the system response, in modal coordinates, to several initial
conditions (marked by small circles). These initial condition responses are superimposed on a plot of

the vector field z. (A vector field defines, at any point z, a velocity vector z.)
We may convert
the modal response back into x coordinates (shown in the subplot to the left) by multiplying z(t) by
the transformation matrix T . Notice that the line z2 = 0, corresponding to a pure Mode 1 response
transforms to the line x2 = x1 , which is defined by v 1 for all R. Similarly, the line z1 = 0 transforms
to the line x2 = x1 , which corresponds to the eigenvector v 2 .
ASIDE: For a planar system with two real eigenvalues, the equilibrium at the origin is called a stable
node or a sink if both eigenvalues are real and negative. If the eigenvalues are both real and positive, the
equilibrium is called an unstable node or a source. If one eigenvalue is positive and the other is negative,
as in the example above, then the equilibrium is called a saddle node.
Example 2.
A=

1 21
2 1

The eigenvalues are

1,2 = 1 i
and the eigenvectors are
v 1,2 =
Define
a=

1
0

1
0

and

0
2

b=

0
2

Using the transformation matrix T = [a, b], define

1
z = T x.

The equations of motion in quasi-modal coordinates are

z = 1 1
z =
z.
1 1
The general response is
z(t) = e

cos(t) sin(t)
sin(t) cos(t)

z0 .

If z0 = [1, 0]T , which corresponds to x0 = a, then the response is

cos t
z(t) = et
.
sin t

10

10

z2

x2

If z0 = [0, 1]T , which corresponds to x0 = b, then the response is

sin t
t
z(t) = e
.
cos t

10
10

0
x1

10
10

10

0
z1

10

Figure 2: Phase portrait for Example #2. Original coordinates (left) and quasi-modal coordinates (right).
More generally, the response will be a linear combination of these two responses. The subplot to the right
in Figure 2 shows the system response to several initial conditions superimposed on a plot of the vector
z . We may convert this response back into x coordinates (shown to the left) by multiplying z(t) by
field
the transformation matrix T.
ASIDE: For a planar system with complex conjugate eigenvalues, as above, the equilibrium at the origin
is called a stable focus if < 0. If > 0, then the equilibrium is called an unstable focus. If = 0, then
the equilibrium is called a center.

Lecture 19: Stick-Fixed Longitudinal Dynamics

The small perturbation equations describing the stick-fixed longitudinal motion of an airplane are
Lx
L = A
L
x
TL = [u, w, q, ] and
where x

1
m Xu
Zu

mZw

L =
A
1
Mw Zu
I Mu + mZ
y
w

1
m Xw
Zw
mZw

Mw Zw
1
M
+
w
Iy
mZw

1
Iy

g
(Zq +mu0 )
0

mZw

.
M (Zq +mu0 )
Mq + w mZ
0

(1)

Figure 1: A short takeoff and landing airplane [1].

Longitudinal Dynamics of a STOL Airplane. Consider the STOL aircract shown in Figure 1.
W = 40, 000 lb (m = 1242.2 slug), Iy = 215, 000 slug ft2 , S = 945 ft2 , c = 10.1 ft.
For constant altitude, wings level flight at speed 400 ft/s and altitude 10,000 ft, the relevant stability
derivatives are

u
w
q
w

X()
35.7 lb/(ft/s)
121.2 lb/(ft/s)
0 lb/(rad/s)
0 lb/(ft/s2 )

Z()
200.7 lb/(ft/s)
1744.7 lb/(ft/s)
13.1 103 lb/(rad/s)
5.6 lb/(ft/s2 )

M()
0 (ft lb)/(ft/s)
2.6 103 (ft lb)/(ft/s)
600.4 (ft lb)/(rad/s)
256.0 (ft lb)/(ft/s2 )

L
We find that the eigenvalues and corresponding eigenvectors for A

0.0456
0.9986
1,2 = 2.3297 1.7818j with v 1,2 =
0.0024
0.0016
and

3,4 = 0.0102 0.0848j with v 3,4

are

0.0248

0
j

0.0046
0.0008

0.9986
0
0.0523
0.0008

=
0.0002 j 0.0000
0.0004
0.0026

The real part of the eigenvalues 1 and 2 (the n component) is comparatively large, meaning that the
contribution of these terms to the initial condition response varies rapidly. Also, the ratio of the imaginary
part to the real part is smaller meaning that these terms contribute fairly well-damped oscillations. The
eigenvalues 1 and 2 correspond to the short-period mode. This mode is governed largely by the size
and location of the horizontal tail. It is the more critical mode in terms of longitudinal stability. The short
period natural frequency and damping ratio are
nsp = 2.93 rad/s

and

sp = 0.79.

The time and number of cycles to half amplitude for this well-damped, stable mode are
thalf sp = 0.30 s

and

Nhalf sp = 0.08.

The real part of the eigenvalues 3 and 4 (the n component) is relatively small in magnitude,
meaning that the contribution of these terms to the initial condition response varies slowly. Also, the ratio
of the imaginary part to the real part is fairly large meaning that these terms contribute underdamped
oscillations. The eigenvalues 3 and 4 correspond to the long-period or phugoid mode. This mode is
characterized by a relatively slow trade-off between kinetic energy (speed) and potential energy (altitude).
The phugoid natural frequency and damping ratio are
np = 0.085 rad/s

and

p = 0.12.

The time and number of cycles to half amplitude for this lightly damped, stable mode are
thalf p = 68 s

and

Nhalf p = 0.91.

Phugoid Mode Approximation. While it is possible to formally compute eigenvalues and eigenvectors
analytically, without substituting values for parameters, it is much more convenient and insightful to obtain
simple approximations which capture the essential physics. Suppose, then, that we wish to approximate
the phugoid mode. To do so, we must isolate this lightly damped, long period mode from the general
transient response. Note that one characteristic of the short period response is that converges quickly
to zero, or very nearly so. Using this observation, in a previous lecture, we approximated the phugoid
mode by assuming that w 0. We also neglected Zq and Zw . For the STOL aircraft in Figure 1, the
L in (1), it is
magnitude of Zq is only 3% of mu0 while the magnitude of Zw is 0.5% of m. Referring to A
clear that we can neglect these terms.
The resulting approximation was

np

Zu g
mu0

and

Xu
.
2mnP

For the STOL example above, this approximation gives

np 0.11 rad/s

and

p 0.12.

Comparing this approximation with the actual values computed earlier, we see that the damping ratio
approximation is quite accurate, although the natural frequency is slightly over-estimated.
Short Period Mode Approximation. As we have seen, a good approximation of the Phugoid mode
may be obtained by assuming that 0. This assumption allows us to express the mode in terms of the
variables u and . It stands to reason, then, that the short period mode should involve the remaining
2

two variables w and q. Specifically, if we assume that u = 0 and ignore the pitch angle , we obtain
the equations

1
u0
w
w
m Zw

=
.
1
1
1
q
q
Iy Mw + m Mw Zw
Iy (Mq + Mw u0 )

The characteristic polynomial for the system above is

1
1
u0
1
2
Zw +
(Mq + u0 Mw ) +
Zw Mq Mw .

m
Iy
mIy
Iy

The coefficient of 1 is positive because Zw , Mq and Mw are all negative. Because Mw is negative, as well,
the coefficient of o is also positive. Thus, we may write

1
1
Zw +
(Mq + u0 Mw )
2sp nsp
m
Iy
and
n2 sp

1
u0
Zw Mq Mw .
mIy
Iy

The approximate short period natural frequency and damping ratio are

nsp

1
u0
Zw Mq Mw
mIy
Iy

and

sp

2nsp

1
1
Zw +
(Mq + u0 Mw ) .
m
Iy

For the STOL example given previously, the approximation above gives
nsp 2.96 rad/s

and

sp 0.79.

These values agree quite well with the true values computed from the complete state matrix.
Im

Im
Zu
u0 increasing.

-Mw increasing.

!n p

!n sp
-X u increasing.

-Mq increasing.

- sp! nsp

Re

- p!np

Re

Phugoid Mode

Short Period Mode

Figure 2: Effect of key parameters on longitudinal modes.

Figure 2 illustrates the effect of various longitudinal parameters on modal frequencies and damping ratios.
The key parameters affecting the short period mode are the pitch stiffness and the pitch damping. Both
of these values are determined, in large part, by the horizontal tail volume ratio. Key parameters affecting
the phugoid mode include speed and lift-to-drag ratio. Note that, while the tail volume may be considered
a free design parameter, the cruise speed and lift-to-drag ratio are typically dictated by basic customer
requirements.
3

Performance Specifications: Flying Qualities. The term flying qualities is a seemingly vague
reference to the overall feel of an airplane. In fact, a considerable amount of effort has gone into
developing a firm and quantifiable definition of this term. The Cooper-Harper Pilot Opinion Rating Scale
is a questionnaire developed for test pilots in order to quantify their opinions about the aircraft that they
fly. Essentially, a pilot rates an airplanes performance on a scale of one to ten, with one representing
excellent, highly desirable performance characteristics and with ten representing major deficiencies in
performance. This rating can be related to the following three levels of flying qualities:
Level 1: Flying qualities clearly adequate for the mission flight phase.
Level 2: Flying qualities adequate to accomplish the mission flight phase but some increase in pilot
workload or degradation in mission effectiveness exists.
Level 3: Flying qualities such that the airplane can be controlled safely, but pilot workload is
excessive or mission effectiveness is inadequate or both.
Using the Cooper-Harper scale, these three levels of flying qualities have in turn been related to formal
numerical bounds on the parameters that characterize an airplanes dynamics, such as short period and
phugoid natural frequency and damping ratio. As a basic requirement, the short period mode must be
dynamically stable. Table 1, adapted from [1], shows bounds on the short period damping ratio for two
representative flight phases and the three flying quality levels.
Table 1: Bounds
Take-off
& Landing
Level min max
1
0.35 1.30
2
0.25 2.00
3
0.15
-

on sp .
Cruising
Flight
min max
0.30 2.00
0.20 2.00
0.15
-

Better still, Figure 3 shows a thumbprint plot of short period natural frequencies and damping ratios.
The center of the contour marked
Good corresponds roughly to a natural frequency of radians per

2
second and a damping ratio of 2 .
The phugoid mode plays a much less important role in an airplanes dynamics. Table 2, also adapted
from [1], shows bounds on the phugoid damping ratio for the three flying quality levels. Notice that, at
least for Level 3 flying qualities, an unstable phugoid mode is acceptable.
Table 2: Bounds on p .
Level
Condition
1
p > 0.04
2
p > 0
3
tdoublep > 55 s

References
[1] R. C. Nelson. Flight Stability and Automatic Control. WCB McGraw-Hill, New York, NY, second
edition, 1998.

Figure 3: Short period natural frequency versus damping ratio [1].

Lecture 20: Stick-Fixed Lateral-Directional Dynamics

Define the reduced lateral-directional state vector

T
LD = v p r
x
.

(In studying aircraft stability, one may ignore lateral position and heading.) The reduced, linearized
lateral-directional equations (with controls fixed at zero) take the form
LD x
LD = A
LD
x
where

LD =
A

1
m Yv
1
(Iz Lv + Ixz Nv )
1
(Ixz Lv + Ix Nv )

1
m Yp
1
(Iz Lp + Ixz Np )
1
(Ixz Lp + Ix Np )

1
m Yr u0
1
(Iz Lr + Ixz Nr )
1
(Ixz Lr + Ix Nr )

tan 0

g cos 0

0
0

(1)

where
2
= Ix Iz Ixz

In this case, we have

v

d
p =

r
dt

1
m Yv
1
Ix Lv
1
Iz Nv

1
m Yp
1
Ix Lp
1
Iz Np

1
m Yr u0
1
Ix Lr
1
Iz Nr

tan 0

g cos 0
v
p
0

r
0

where the terms Lv , Lp , and so on may be obtained by direct comparison with (1). Thus, we find that
L() =
N() =

Ix
Iz L() + Ixz N()

Iz
Ixz L() + Ix N()

where the dot represents a dependent variable (v, p, or r). If the stability axes happen to coincide with
the principal axes of inertia, so that Ixz = 0, we simply have L() = L() and N() = N() .
As in the longitudinal case, we will find that the general initial condition response can be divided into
characteristic modes, each of which corresponds to a single real eigenvalue or a pair of complex conjugate
eigenvalues. Stick-fixed dynamic stability is determined by the real part of these eigenvalues:
If every eigenvalue has negative real part, then the lateral directional dynamics are stable.
If any eigenvalue has positive real part, then the lateral directional dynamics are unstable.
Lateral-Directional Dynamics of a STOL Airplane. We consider, once again, the STOL aircraft
shown in Figure 1. The relevant geometric parameters are
W = 40, 000 lb (m = 1242.2 slug), Ix = 273, 000 slug ft2 , Iz = 447, 000 slug ft2 , Ixz = 0 slug ft2
and
S = 945 ft2 , b = 96 ft.
For constant altitude flight at speed 400 ft/s and altitude 10,000 ft, the relevant stability derivatives are
1

Figure 1: A short takeoff and landing airplane [2].

v
p
r

Y()
119.7 lb/(ft/s)
0 lb/(rad/s)
0 lb/(rad/s)

L()
3968 (ft lb)/(ft/s)
807.6 103 (ft lb)/(rad/s)
172.2 103 (ft lb)/(rad/s)

N()
3206 (ft lb)/(ft/s)
56.4 103 (ft lb)/(rad/s)
260.6 103 (ft lb)/(rad/s)

LD , we find that the eigenvalues are:

Having determined the various elements of the matrix A
1 = 0.0132,

2 = 3.05,

and

3,4 = 0.288 j1.757.

The first eigenvalue corresponds to a stable mode with a very large time constant. Analysis of the associated
eigenvector indicates that this mode corresponds, roughly, to a coordinated turn, that is, a banked turn
with nearly zero sideslip angle. This mode is referred to as the spiral mode, and we denote s = 1 . For
the case shown here, the spiral mode is stable, meaning that the airplanes response to a perturbation in
the direction of this eigenmode would converge (slowly) back to the equilibrium flight condition. Like the
phugoid mode, the spiral mode can sometimes be unstable.
The second eigenvalue corresponds to a stable mode with a very small time constant. Analysis of the
associated eigenvector indicates that this mode corresponds roll rate. Thus, a perturbation in the roll rate
will be quickly damped. This mode of response is called the roll mode or the rolling convergence mode,
which emphasizes the stability of this mode. We denote this mode r = 2 .
The third and fourth eigenvalues correspond to a stable, second order mode with natural frequency 1.781
radians per second and damping ratio 0.162. Analysis of the associated eigenvector indicates that this mode
is a relatively complex oscillatory motion in which the airplane yaws, then rolls, then yaws back, and rolls
back again. The motion has been described as reminiscent of a Dutch ice skater and was therefore dubbed
the Dutch roll mode. We let ndr and dr be the natural frequency and damping ratio corresponding to the
complex conjugate eigenvalues 3 and 4 .
Spiral Mode Approximation. As in the longitudinal case, it is often more convenient to estimate the
characteristics of the various lateral-directional modes. Simple, approximate expressions for properties
such as time constants, natural frequencies, and damping ratios can speed analysis and, more importantly,
can help one to develop an intuition for the effect of aircraft geometry on motion characteristics.
The characteristic polynomial for the fourth order lateral-directional system is of the form
4 + a3 3 + a2 2 + a1 + a0 .
Because the characteristic value corresponding to the spiral mode is typically quite small in comparison
to the other characteristic values, we may assume that higher powers of this eigenvalue will be extremely
2

small. Thus, the spiral mode can be approximated by solving

s + a0 = 0
a1

s = a0 .

a1

In terms of the stability derivatives, according to [1], this expression becomes

s =

(Lr Nv Lv Nr )
.
Iz Lv + ug0 (Lv Np Lp Nv )

Substituting from the table of values above, we find that

s = 0.0128,

which is clearly a good approximation to s = 0.0132.

Roll Mode Approximation. A crude approximation of the roll mode arises from simply ignoring all
coupling between roll and the other state variables:
p =

1
Lp p.
Ix

The single characteristic value associated with this first order equation is
r = 1 Lp .

Ix
This approximation provides a reasonable estimate of the roll mode, although not an extremely accurate
r = 4.844 with r = 3.049.)
one. (Compare
Note that the general solution to the equation above is
p(t) = p(0)e(Lp /Ix )t ,
which converges rapidly to zero, since Lp is large negative.
Dutch Roll Mode Approximation. The Dutch roll motion involves both banking and yawing. A
reasonable approximation arises if one simply ignores the roll rate equation and the roll angle equation.
The result is the second order system

1
d
Yv m
Yr u0
v
v
m
=
1
1
r
dt r
Iz Nv
Iz Nr
The characteristic polynomial is

1
1
1
u
0
2

dr +

Yv + Nr
(Yv Nr Nv Yr ) + Nv .
dr
m
Iz
mIz
Iz
The mode is typically oscillatory, which can be verified in a given case by checking that the discriminant
is negative. Assuming this is the case, we may write

ndr =

1
u0
(Yv Nr Nv Yr ) + Nv
mIz
Iz

and

dr =

1
2ndr

1
1
Yv + Nr .
m
Iz

Substituting from the table of values above, we find, for the STOL example, that

ndr = 1.327 rad/s

and

dr = 0.170.

These values are fairly close to the true values ndr = 1.781 radians per second and dr = 0.162.
Performance Specifications: Flying Qualities. As there were for the two longitudinal modes, there
are standard requirements for the three lateral-directional modes. For the spiral mode, the requirement
comes in the form of a minimum time to double amplitude (12 to 20 seconds for Level 1 flying qualities and
4 seconds for Level 3). Thus, the spiral mode can actually be unstable, so long as the rate of divergence is
not too high.
The roll mode, on the other hand, must be stable. The flying qualities specifications on this mode come
in the form of a maximum allowable time to half amplitude. This time varies with the type of aircraft and
flight phase but is generally less than one second for Level 1 flying qualities and less than ten seconds for
Level 3.
Specifications on the Dutch roll mode call for a minimum damping ratio and a minimum natural frequency.
Again, the specifications vary with aircraft type and flight phase. For specific values, see [2].

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.
[2] R. C. Nelson. Flight Stability and Automatic Control. WCB McGraw-Hill, New York, NY, second
edition, 1998.

Lecture 21: Introduction to Aircraft Control

Longitudinal Control. Recall the following small perturbation equations for the longitudinal dynamics:
Lx
Lu
L = A
L + B
L
x
where the reduced longitudinal state vector, input vector, and input matrix are

m Xe
u

Ze

w
e
mZw

L =
L =
x
,
u
,
and
BL =

q
T
1
Mw Ze
M
+
Iy
e
mZw

1
m XT
ZT
mZw
1
Iy

MT

w
ZT
+M

mZw
0

L was given in Lecture 19.)

(The reduced longitudinal state matrix A
When we considered stick-fixed longitudinal dynamic stability in previous lectures, we assumed that the
longitudinal inputs e and T were both zero. Here, we consider what happens when e or T is
varied. As a representative case, we will consider step inputs in elevator angle and thrust. First, though,
we must develop the expressions appearing in B L . The e derivatives are depend on elevator effectiveness and elevator power, terms which can be determined using semi-empirical techniques described in
Appendix B of [1]. Table 1 shows the dimensional e derivatives, along with the remaining longitudinal
stability derivatives.

u
w

q
w

Table 1: Longitudinal Dimensional Stability Derivatives

X()
Z()
1
1
2 u0 S [2 (CD0 + CT0 ) + (CDu + CTu )] 2 u0 S (2CL0 + CLu )
1
12 u0 S (CD0 + CL )
2 u0 S (CD + CL0 )
u0 X w
u0 Zw
0
41 u0 S
cCLq
1
cCL
0
4 S
0
u0 Zw
1
2
0
2 u0 S (CLe )

M()
1
cCmu
2 u0 S
1
cCm
2 u0 S
u 0 Mw
1
u
c2 Cmq
0 S
4
1
c2 Cm
4 S
u0 Mw
1
2 cC
me
2 u0 S

The T derivatives are really functions of the propulsion system and must be determined case by case.
For simplicity, we will typically assume that ZT and MT are zero. (In fact, engines are often installed
at a slight incidence to provide additional propulsive lift, which would contribute to both ZT and MT ,
but we ignore this effect here.) The term XT is somewhat ambiguous. For starters, it is not entirely
clear what a unit step input in throttle corresponds to. If the unit of throttle displacement is small, one
would expect a correspondingly small change in thrust, relative to the aircrafts weight. In examples, we
will arbitrarily assume that XT = 0.1W ; please note, however, that this assumption is completely without
physical motivation.
Matlab contains a number of tools for dealing with LTI systems and for transferring between various
representations. For example, the Matlab command ss takes the matrices A, B, C, and D as arguments
and generates an object called a system which is simply the state-space representation. This object can
be operated on using commands such as step (to obtain the systems step response) or initial (to obtain
the initial condition response). The command lsim provides the system response to initial conditions and
a user-defined input signal. Having defined a system in state-space form using ss, one may determine the
systems transfer function matrix using the command ss2tf.
1

Response to an elevator step input (1 deg)

u (ft/s)

0
50
100
150

10

12

14

16

18

20

10

12

14

16

18

20

10

12

14

16

18

20

10

12

14

16

18

20

(deg)

(deg)

40
20
0
20

(deg)

40
20
0
20

Figure 1: Elevator step response for a STOL airplane [2].

Figure 1 shows the short-term elevator step response for the short takeoff and landing aircraft we have
considered in previous lectures. The figure shows four plots representing change in speed u, change in
angle of attack , change in pitch angle , and change in flight path angle = . The
system begins at equilibrium but experiences a 1 elevator deflection at t = 5 seconds. Recall that, by
convention, a negative elevator deflection gives a nose-up moment. Shortly after the command is executed,
the angle of attack converges to a small positive value. The time required for this convergence is related
to the time constant for the short period mode. Thus, we see that the stick-fixed dynamics impact control
characteristics. Also notice that the airplanes speed begins to drop, and the pitch (and flight path) angle
begins to rise.
Over a longer period, as shown in Figure 2, the system undergoes a phugoid-like convergence to a new
horizontal equilibrium flight condition in which the airplane travels slower and at a correspondingly higher
angle of attack.
Next, consider the effect of a unit increase in thrust. As shown in Figure 3, the long-term effect is not an
increase in speed u or the angle of attack , but rather an increase in pitch (and thus flight path) angle.
That is, the long-term effect of an increase in throttle is for the aircraft to climb at its trim speed and lift
coefficient.
Pitch Stabilization Example. Consider a wind-tunnel model which is pinned to allow pitch rotation
about the center of gravity. Rather than a conventional horizontal stabilizer with elevators, the model uses
servo-actuated canards to provide longitudinal stability and control. For small angles of attack, the model
dynamics are well-described by the second order ODE
Mq M = Mc c

(1)

where c represents the canard deflection. (Since the model is mounted in a wind tunnel, the angle of
attack is identical to the aircraft pitch angle .) Pitch damping due to a canard acts to oppose pitch
2

Response to an elevator step input (1 deg)

u (ft/s)

0
100
200
300

50

100

150

200

250

300

50

100

150

200

250

300

50

100

150

200

250

300

50

100

150

200

250

300

(deg)

15
10
5
0

(deg)

40
20
0
20

(deg)

40
20
0
20

Figure 2: Long-term elevator step response.

rate just as it does for an aft tail. Thus, one finds that Mq < 0. The pitch stiffness, however, is diminished
by a forward tail and, for the case shown in Figure 4, one finds that M > 0. The airplane is not statically
stable.
Given a desired pitch angle d , define the error
e = d .
For simplicity, suppose that d = 0 and define the following proportional-derivative (PD) feedback
control law
c = kp e + kd e

= kp kd .
That is, apply a control deflection in which is the sum of terms directly proportional to the error and the
rate of increase of the error. Substituting into the dynamic equation gives
+ (Mc kd Mq ) + (Mc kp M ) = 0.
For stability, we require that
Mc kd > Mq

and

Mc kp > M .

(2)

Since Mq < 0 and Mc > 0, we could actually just choose kd = 0, which would give a simple proportional
feedback control structure. On the other hand, because M > 0, we must choose kp large enough to
dominate that destabilizing term. Moreover, in order to obtain an arbitrary closed-loop natural frequency
and damping ratio, we should retain the derivative term and choose
kp =

1
M + n2
Mc

and
3

kd =

1
(Mq + 2n ) .
Mc

Response to a thrust step input (0.1 g)

u (ft/s)

50

50

50

100

150

200

250

300

50

100

150

200

250

300

50

100

150

200

250

300

50

100

150

200

250

300

(deg)

0.5

0.5

(deg)

10

(deg)

10

Figure 3: Thrust step response.

dc

Figure 4: An airplane with canards.

Now suppose that d is some nonzero constant. Then we have
+ (Mc kd Mq ) + (Mc kp M ) = Mc kp d .
Assuming conditions (2) hold, one may use the final value theorem (FVT) to show that
lim =

Mc kp
d
Mc kp M

If kp is chosen large enough, then will approach d in time, however it will never really converge to the
desired value. While letting kp would make the error arbitrarily small, there are practical concerns
associated with such high gain feedback, including actuator limits and destabilization of unmodeled
dynamics. A better approach to eliminating the steady-state error is to incorporate an integral term in
the controller. At this point, it may be easier to proceed in the s-domain rather than the t-domain
Re-expressing the pitch dynamics (1) in the s-domain, we find that the plant transfer function is
P (s) =

Mc
(s)
= 2
.
c(s)
s Mq s M

Given the desired pitch angle history d (t), the error signal is
e(s) = d (s) (s).
4

In order to stabilize the system, so that the angle of attack may be prescribed as desired, we will implement
the PID compensator
c(s)
e(s)

F (s) =

1
= kp + ki + kd s
s
kp s + ki + kd s2
.
=
s

(The fact that the degree of the numerator polynomial is higher than that of the denominator is a bit
problematic, because it suggests that the compensator is acausal, i.e., that current outputs c depend on
future inputs e. In practice, there are simple ways around this problem.)
e

Figure 5: The closed-loop control system.

The feedback control structure is shown in Figure 5. To solve for the closed-loop transfer function (from
d to ), we note that
(s) = P (s)c(s)
= P (s)(F (s)e(s))
= P (s)F (s)(d (s) (s)).
Solving for (and omitting the independent variable s) gives
(1 + P F ) = P F d
so that the closed-loop transfer function is
H(s) =

PF
(s)
=
.
d (s)
1 + PF

Substituting the definitions of P and F and manipulating a bit gives

2
kp s+ki +kd s
s

H(s) =

1+
=
=

kp s+ki +kd s2
s

Mc
s2 Mq sM

Mc
s2 Mq sM

kp s + ki + kd s2 (Mc )
s (s2 Mq s M ) + Mc (kp s + ki + kd s2 )

Mc kp s + ki + kd s2
.
s3 + (Mc kd Mq ) s2 + (Mc kp M ) s + (Mc ki )

Lets revisit the PD feedback control problem for a moment by assuming that ki = 0. Then the closed-loop
transfer function becomes

Mc kp s + kd s2
H(s) = 3
s + (Mc kd Mq ) s2 + (Mc kp M ) s
Mc (kp + kd s)
(3)
= 2
s + (Mc kd Mq ) s + (Mc kp M )

Using the FVT, the steady state response to a step input, say d 1s , is

Mc kp
1

=
d .
lim = lim sH(s) d
t
s0
s
Mc kp M
Except in the limit kp , the steady-state response will not converge to precisely the desired value.
Now consider the PID-controlled system. Stability requires that the roots of the denominator polynomial
in (3) have negative real part. (This polynomial plays precisely the same role as the characteristic polynomial for the state matrix in a state-space LTI system.) A necessary condition is that each coefficient in
the polynomial be positive, which implies that ki must be positive (in addition to the previous stability
requirements). Suppose stabilizing values of the control gains (kp , ki , and kd ) are chosen and consider once
again the step response problem:

Mc ki
1

=
d = d .
lim = lim sH(s) d
t
s0
s
Mc ki
Thus, integral control eliminates the steady-state error in response to a step input. More generally, integral
control is useful for rejecting unmodeled constant biases. So, for example, if the pitch dynamics (1) included
a constant term due to some peculiar flow phenomenon, PID control could eliminate that bias, whereas
PD control could not.
There are experimentally-based algorithmic techniques for tuning the gains of a PID feedback controller,
the most common being the Ziegler-Nichols tuning rules. For more information see [?].
Lateral-Directional Control. Next, recall that the reduced lateral-directional equations take the form
LD x
LD u
LD = A
LD + B
LD
x
where the reduced, lateral-directional state vector, input vector, and input matrix are

1
Ya m
Yr
v
m

p
a
La Lr
LD =

LD =

x
,
u
=
,
and
B
.

LD
r
r
N
Nr
a

0
0
LD given above,
In the input matrix B
L() =
N() =

Ix
2
Ix Iz Ixz
Iz
2
Ix Iz Ixz

Iz L() + Ixz N()

Ixz L() + Ix N() .

LD was given in Lecture 20.)

(The reduced lateral-directional state matrix A
The dimensional stability and control derivatives are given, in terms of the nondimensional derivatives, in
Table 2.
Figure 6 shows the aileron step response for the STOL aircraft. The four plots represent change in sideward
speed v, change in roll rate p, change in yaw rate r, and change in roll angle . The system begins
at equilibrium but experiences a 1 aileron deflection at t = 5 seconds. The result is a relatively quick,
underdamped convergence to a positive roll rate. As the aircraft begins to slip sideways, v grows and,
as the vertical tail tries to align the nose with the oncoming wind, the plane begins to yaw to the right.
The roll rate tapers off somewhat as this spiral-like behavior develops. Note that the validity of the model
becomes questionable as the roll angle grows very large.
6

Table 2: Lateral-Directional Dimensional Stability Derivatives

Y()
L()
N()
1
1
1
v
2 u0 SCY
2 u0 bSCl
2 u0 bSCn
p 14 u0 bSCYp 14 u0 b2 SClp 41 u0 b2 SCnp
1
1
1
2
2
r
4 u0 bSCYr
4 u0 b SClr
4 u0 b SCnr
1
1
1
2
2
2
a 2 u0 SCYa 2 u0 bSCla 2 u0 bSCna
1
2
r 12 u20 SCYr 12 u20 bSClr
2 u0 bSCnr

Response to an aileron step input (1 deg)

v (ft/s)

20

10

20

40

60

80

100

120

20

40

60

80

100

120

20

40

60

80

100

120

20

40

60

80

100

120

p (deg/s)

3
2
1
0

r (deg/s)

20
10
0
10

(deg)

200

100

Figure 6: Aileron step response for a STOL airplane [2].

Figure 7 shows the rudder step response for the STOL aircraft. The system begins at equilibrium but
experiences a 1 rudder deflection at t = 5 seconds. The result is a relatively quick, underdamped convergence to a positive sideslip angle (nose left) which results in a slight negative roll rate. As the aircraft
begins to slip to the left, v eventually grows negative and, as the vertical tail tries to align the nose
with the oncoming wind, the plane begins to yaw nose-left. Once again, a spiral-like behavior develops,
but in the opposite direction to before. Also, once again, we note that the validity of the model becomes
questionable as the roll angle grows large.

References
[1] B. Etkin and L. D. Reid. Dynamics of Flight: Stability and Control. John Wiley and Sons, New York,
NY, third edition, 1996.
[2] R. C. Nelson. Flight Stability and Automatic Control. WCB McGraw-Hill, New York, NY, second
edition, 1998.
7

Response to a rudder step input (1 deg)

v (ft/s)

10

10

20

40

60

80

100

120

20

40

60

80

100

120

20

40

60

80

100

120

20

40

60

80

100

120

p (deg/s)

r (deg/s)

10
0
10
20

(deg)

0
50
100
150

Figure 7: Rudder step response for a STOL airplane [2].

Lecture 22: A Brief Introduction to Linear Control

Figure 1, which is taken from [1], represents the general form of a control system. The fundamental
component is the plant itself. The Plant Dynamics block in Figure 1 represents some system whose
behavior is influenced by the application of a control input u. While the complete behavior of the plant is
given by its state history, there may be some subset of state variables which are of particular interest to
the control designer; these define the output y. While we will consider only linear, time-invariant systems,
this control system structure is valid for a much more general class of systems. For example, the dynamics
in any of the blocks might be described by nonlinear, time-varying ODEs or even PDEs. Of course, the
problem of control design and analysis is much more difficult in those cases.
w

wm
yd

Disturbance
Sensor

Controller

Plant
Dynamics

Output
Sensor

ym

Figure 1: Functional diagram of a general control system.

Besides the input u, the plants behavior
controller, it is necessary to consider the
Sometimes, in order to meet performance
explicitly and pre-compensate for them.
dynamics and disturbance sensor noise.

may also be influenced by a disturbance w. In designing a

nature of the disturbances which will influence the system.
objectives, it is even necessary to measure the disturbances
In this case, one might also consider the disturbance sensor

Referring again to Figure 1, notice that the output y is not the output that someone who observes the
system would see. Rather, she or he would see a measurement of the output, say y m , which is generated by
a physical sensor. This sensor can itself be thought of as a plant which takes y as its input and produces
y m as its output. It can be assumed that the sensor dynamics are stable, in the sense that the output
responds in proportion to the input; otherwise the sensor would not be of much use. Because the sensor is
a dynamical system, one may expect that it will take some amount of time for it to respond to the signal
it is measuring. However, one would hope that the sensor dynamics are much faster than the dynamics of
the process being measured; otherwise, the sensor will not be able to keep up with y and will give an
erroneous measurement. Another possible source of discrepancy between y and y m is a noise signal v. If
the measured output is to be compared with the desired output y d in a feedback loop, then it is important
that the controller compensate for any discrepancy between y and y m .
The control design problem is to design a controller (that is, a device or algorithm which generates the
control input u) in order to make the system, or at least its output, behave in a desired way. Depending
on the situation, one might choose the controller to make use of any or all of the following information
the desired output y d ,
the disturbance measurement wm , and
the output measurement y m .
Now, assume that every element of the system (the plant, the sensors, and the controller) is linear, timeinvariant. In this case, we may exploit the superposition property to express all of the signals as sums of
1

Psw
wm
Fw
yd

Fd

+
+

Pw
u

Ps

+
+

ym

Fm

Figure 2: Functional diagram of an LTI control system.

linear operations on other signals. For example, in the Laplace domain, we may write
u(s) = F d (s)y d (s) + F m (s)y m (s) + F w (s)wm (s)
= F d (s)y d (s) + F m (s)y m (s) + F w (s)P sw w(s)
y(s) = P (s)u(s) + P w (s)w(s)
y m (s) = P s (s)y(s) + v(s),
where upper case letters represent transfer function matrices and lower case letters represent signal vectors.
Figure 2 shows the system diagram for this more specific case of an LTI system.
Aside: Transfer Functions from LTI State-Space Models. In general, a system of first order linear,
time-invariant ODEs may be represented in the state-space form
x = Ax + Bu
y = Cx + Du,
where u is a vector of input signals, y is a vector of output signals, and x is the system state. The set
of equations may be transformed into transfer function form by taking the Laplace transform of all of the
signals and determining the map from U (s) to Y (s). Doing so, we find that
sX(s) x0 = AX(s) + BU (s)
Y (s) = CX(s) + DU (s).
Rearranging the first equation, we obtain
(sI A) X(s) = x0 + BU (s)
or
X(s) = (sI A)1 (x0 + BU (s)) ,
where I is the identity matrix. Substituting into the output equation gives
Y (s) = C (sI A)1 (x0 + BU (s)) + DU (s)
or

Y (s) = C (sI A)1 x0 + C (sI A)1 B + D U (s).

2

Setting the initial state x0 to zero (as is required to determine a transfer function), we see that the matrix
P (s) = C (sI A)1 B + D
maps the vector U (s) of input signals to the vector Y (s) of output signals. If the system is single-input,
single output (SISO), then the matrix transfer function becomes a scalar, say P (s).
The system transfer function represents the complete system dynamics in the form of a map from input
signals to output signals. The control problem is to find an input signal that makes the output signal behave
in some desired way. If the system being considered is an airplane, then u(t) might represent thrust and
elevator inputs and y(t) might represent speed and flight path angle. In designing a longitudinal controller,
one would seek thrust and elevator histories which would force the speed and flight path angle to follow
some desired time history.
Controller Structures. While Figures 1 and 2 represent a general control system, it is not always
necessary or possible to use this complete structure. Following are some simpler control structures which
are commonly used in practice:
Open-loop: Neither output nor disturbance measurements are used to compute u; the input only
depends on the reference signal y d . (In this case, we set F m = F w = 0 and remove the unnecessary
output and disturbance sensors.)
Closed-loop (1 Degree of Freedom): The desired output y d is compared directly with the measured
output y m to generate an error signal which determines u. No disturbance measurement is used. (In
this case, we set F m = F d and F w = 0 and remove the unnecessary disturbance sensor.)
Feedforward control: The disturbance is measured and compensated for by the input u. Typically,
the problem is one of regulating a set point so that one may take y d = 0 without loss of generality.
(In this case, we set F m = F d = 0 and remove the unnecessary output sensor.)
There is also a slightly more general form of feedback, called 2 degree of freedom feedback, in which
both y d and y m are used to determine u, however F m 6= F d . Of course, there are other variations, as
well. For example, one may combine feedforward and feedback control.
In every case, the goal of the control designer is to make the true output y follow as closely as possible the
desired output y d . In other words, the goal is to maintain the error e = y d y as small as possible.
Following are some general observations comparing open- and closed-loop control.
Open-loop control . . .
1. . . . can not stabilize an unstable plant.
2. . . . does not attenuate disturbances.
3. . . . does not mitigate sensitivity to plant parameter variations.
4. . . . requires little or no special equipment (such as expensive sensors)
On the other hand, closed-loop control . . .
1. . . . can stabilize an unstable plant. (It can also destabilize a stable plant!)
2. . . . does attenuate disturbances.
3

3. . . . does mitigate sensitivity to plant parameter variations.

4. . . . does require sensors and hardware for processing the sensor signals.
Basic Classical Control Elements. Perhaps the most well-known acronym in the control community is
PID. This stands for proportional-integral-derivative and it refers to a particular one degree of freedom
control structure which is used in a wide variety of applications. As you know, the controller in a one
degree of freedom feedback structure generates a control input u to the plant in response to an error signal
e = yd y. The PID controller generates this input as the sum of three signals:
u(t) = up + ui + ud .
Proportional action. The first component of the PID control signal is proportional to the error. That is,
up (t) = kp e(t),
where kp is the proportional gain. Large error signals generate large control signals and, as the error
decreases, so does the control effort. Of course, the trick is to ensure that large control signals tend to
drive the error toward zero instead of making it even bigger. That is, one must check that the proportional
control law actually stabilizes the system.
Integral action. The second component of the PID control signal is proportional to the time integral of the
error,
Z
t

e( )d,

ui (t) = ki

where ki is the integral gain. This term has little effect until the error accumulates sufficiently. Of
course, error is a signed quantity, so negative error can cancel with positive error. Consequently, integral
action can increase oscillation in the closed-loop system transient response. Again, one must be careful
that the integral term does not destabilize the system.
Derivative action. The third component of the PID control signal is proportional to the time derivative of
the error,
de
ud (t) = kd ,
dt
where kd is the derivative gain. This term responds not to the actual error, but to the rate of increase
of error. The derivative term is often described as anticipatory because it takes effect before the error
ever has a chance to grow.
Summing the various terms gives the classical PID controller:
Z t
de
u = kp e(t) + ki
e( )d + kd .
dt
0
In the Laplace domain, we have
F (s) =

U (s)
1
= kp + ki + kd s.
E(s)
s

Of course, individual components can be removed by setting their respective gain to zero. For example,
setting ki = 0 reduces the controller above to a proportional-derivative or PD controller.
Example: Flight path regulator. For the STOL aircraft we have considered in previous lectures, the
transfer function from elevator to flight path angle is
P (s) =

7s3 + 0.1s2 1000s 8

(s)
4
.
e(s)
s + 4.7s3 + 8.7s2 + 0.2s + 0.06
4

Figure 3: The closed-loop control system for Problem 1.

Suppose we apply proportional feedback
F (s) =

e(s)
= kp
e(s)

in order to regulate the flight path angle to some desired value d . The closed-loop transfer function is
H(s) =

(s)
d (s)

=
=
=

P (s)F (s)
1 + P (s)F (s)

kp 7s3 + 0.1s2 1000s 8

(s4 + 4.7s3 + 8.7s2 + 0.2s + 0.06) + kp (7s3 + 0.1s2 1000s 8)
7s3 + 0.1s2 1000s 8
4
3
s + (4.7 + 7kp ) s + (8.7 + 0.1kp ) s2 + (0.2 1000kp ) s + (0.06 8kp )

The characteristic values for the closed-loop system are the poles of H(s), i.e., the roots of the denominator
polynomial. A necessary condition for these roots to have negative real part is that all coefficients have the
same sign. Since the coefficient of s4 is positive one, we require that all coefficients of lower order terms
also be positive. We obtain the following conditions on the proportional gain kp .
kp >

4.7
,
7

kp >

8.7
,
0.1

kp <

0.2
,
1000

and

kp <

0.06
.
8

Thus, it is necessary that

0.2
4.7
< kp <
.
7
1000
These conditions are only necessary. To obtain necessary and sufficient conditions, one must apply the
Routh-Hurwitz stability criterion to be discussed in the next lecture.

References
[1] P. R. Belanger. Control Engineering: A Modern Approach. Saunders College Publishing, Philadelphia,
PA, 1995.

Lecture 23: Routh-Hurwitz Stability Analysis

The following discussion follows that of [2]. For more detailed information about Routh-Hurwitz stability
analysis, see [1].
Suppose we have a transfer function
Q(s) =

B(s)
.
A(s)

The transfer function Q(s) might represent the dynamics of an uncontrolled plant, an open-loop controlled
system, or a closed-loop (feedback controlled) system. In any case, Q(s) describes the relationship between
an input, say U (s), and an output, say X(s). For example, the system response to a unit step input can
be obtained as

1
1
1
x(t) = L [X(s)] = L
Q(s) .
s
If any root of A(s) has positive real part, then the system (or, equivalently, the transfer function) is said to
be unstable. As a result x(t) may grow large with time. This is clearly unacceptable if one is attempting
to drive x to some desired value xd . Absolute stability (i.e., having all characteristic values in the left half
complex plane) is a fundamental requirement for a control system.
The problem is that A(s) may be very high order, so finding its roots explicitly (in order to check that
they have negative real part) could be difficult. The quadratic formula works for second order polynomials.
Formulae exist up to order five, but they are very messy and numerical methods are more often used for
third and higher degree polynomials. Rouths stability criterion provides a simple algorithm for determining
whether a system is unstable without solving for the roots explicitly.
Rouths Stability Algorithm.
Step 1. Write the polynomial in the form
A(s) = a0 sn + a1 sn1 + + an1 s + an .
One may assume that an is nonzero. Otherwise, there is a root at s = 0 and one may divide A(s) by s to
check the remaining roots. A necessary (but not sufficient) condition for A(s) to be Hurwitz, i.e., for all of
its roots to have strictly negative real part, is that all coefficients be positive. If any coefficient is zero or
negative, then there is at least one root in the closed right-half plane.
Step 2. If all coefficients are positive, form the Routh Array as follows
sn
sn1
sn2
sn3
..
.
s2
s1
s0

a0 a2 a4
a1 a3 a5
b1 b2 b3
c1 c2 c3
..
..
..
.
.
.
d1 d2
e1
f1

The powers of s to the left serve as place-keepers to remind you how many rows of the array remain to
be computed. (More in a moment.)
The third row of the array is computed according to the formula
bi =

a1 a2i a0 a2i+1
.
a1
1

Start with i = 1 and iterate until bi = 0 for all higher values of i.

The fourth row of the array is computed according to the formula
ci =

b1 a2i+1 a1 bi+1
.
b1

Start with i = 1 and iterate until ci = 0 for all higher values of i.

Continue until f1 has been computed.
Step 3. Apply Rouths Stability Criterion to determine the number of roots with positive real part:
Rouths Stability Criterion: The number of roots with positive real part is equal to the number of sign
changes in the left-most column of the Routh array.
Example: Determine how many roots of the polynomial
3s6 + s5 + 2s3 + s2 + 5s + 1
have positive real part. We may rewrite this polynomial as
3s6 + 1s5 + 0s4 + 2s3 + 1s2 + 5s + 1
The Routh array may be computed as
s6
s5
s4

3
1

0
2

1
5

1032
1

1135
1

1130
1

14

s3

621(14)
6
13

6511
6
31
6

s2

(2)(14)(6)31
2

(2)1(6)0
2

s1

(107)31(2)1
107
3315
107

s0

31

107

Because there are two sign changes in the left-most column, there are two roots with positive real part.
Remark: If the sign of all coefficients in the polynomial A(s) is not the same, then at least one root of A(s)
has positive real part. If determining stability is the only objective, one may stop here! (However, if one
wants to know the precise number of unstable roots, one must proceed with the algorithm as described.)
Remark: Rouths stability criterion describes how many roots have positive real part. The remaining
roots may have negative or zero real part. A necessary and sufficient condition for all roots to have strictly
negative real part is that all coefficients in the polynomial A(s) are strictly positive (no zero coefficients!)
and no sign changes occur in the left-most column of the Routh array.
d

Figure 1: The closed-loop control system for Problem 1.

2

Example: Flight path regulator. Recall the STOL aircraft example we considered in the previous
lecture. The transfer function from elevator to flight path angle is
P (s) =

(s)
7s3 + 0.1s2 1000s 8
4
.
e(s)
s + 5s3 + 9s2 + 0.2s + 0.06

(For simplicity, we have only retained a single significant digit in the polynomial coefficients.) With
proportional error feedback
e(s)
= kp ,
F (s) =
e(s)
where e = d , the closed-loop transfer function is
H(s) =

(s)
d (s)

P (s)F (s)
1 + P (s)F (s)

kp 7s3 + 0.1s2 1000s 8

s4 + (5 + 7kp ) s3 + (9 + 0.1kp ) s2 + (0.2 1000kp ) s + (0.06 8kp )

We previously obtained the following necessary conditions on the proportional gain kp :

4.7
0.2
< kp <
.
7
1000

To obtain necessary and sufficient conditions, we apply the Routh-Hurwitz stability analysis technique.
s4
s3
s2
s1
s0

1
5 + 7kp

9 + 0.1kp
0.2 1000kp

(5+7kp )(9+0.1kp )(1)(0.21000kp )

(5+7kp )
0.7kp2 +1000kp +50
5+7kp
7E3kp3 1E6kp2 5E4kp +10
0.7kp2 +1000kp +50

0.06 8kp

(5+7kp )(0.068kp )(1)(0)

(5+7kp )

0.06 8kp

0.06 8kp

Stability requires not only the previous linear inequalities, but also that
0.7kp2 + 1000kp + 50 > 0

and

7E3kp3 1E6kp2 5E4kp + 10 > 0.

Thus, the conditions on kp are considerably more complicated than simple linear inequalities. The end
result is that only a very tiny range of proportional gains kp > 0 will ensure closed-loop stability.
It is evident that controlling flight path angle using elevator alone is a challenging control design problem.
Of course, there are more controls available than just the elevator and control design should logically take
this into consideration. The traditional approach, however, has been to break the system into a set of
critical single-input, single-output (SISO) channels and to consider them one at a time. At the end,
the control designer verifies that the complete, closed-loop multi-input, multi-output (MIMO) system is
well-behaved. An alternative to this classical approach is to use modern control theory to design a
regulator for the complete MIMO system. To learn more about classical and modern control techniques,
you may be interested in taking AOE 4004 the next time it is offered.
The Routh-Hurwitz approach is a method of assessing absolute stability. It provides necessary and sufficient
conditions for all of the characteristic values to have strictly negative real part. It does not, however,
indicate anything else about the closed-loop characteristic values. Recall that aircraft handling quality
specifications give desired ranges for damping ratio and natural frequency for the characteristic of modes
of motion. These specifications can be translated into specific desired pole locations for the feedbackcontrolled system. To study this problem of relative stability, one must use a tool such as the root locus
3

method, which shows how closed-loop characteristic values move in the complex plane as a control gain
is varied. Again, to learn more about classical and modern control techniques, consider taking AOE 4004
the next time it is offered.
Appendix: Special Cases in Routh-Hurwitz Analysis
A zero in the left-most column. If there is a zero in the left-most column, the procedure cannot be
applied as stated because one must divide by zero. To avoid this problem, replace the zero with a small,
positive constant and proceed as before.
If the sign changes from the row before to the row after , then this indicates one root with positive
real part.
If the sign does not change from the row before to the row after , then there is a complex pair of
roots.
Example: A root with positive real part exists if,
si+1
si
i1
s

5

2

si+1
si
i1
s

or

1

1

A pure imaginary pair of roots exists if,

si+1
si
i1
s

5

2

si+1
si
i1
s

or

1

1

Example: Determine how many roots of the polynomial

s3 + s + 10
have positive real part. We may rewrite this polynomial as
1s3 + 0s2 + 1s + 10
The Routh array may be computed as
s3
s2

1
0

s1

1110

1 10

s0

1
10
10

Because 0 < 1, the third element in the left-most column is negative. Thus, there are two sign changes
and therefore two roots with positive real part. One may verify that the roots are
s = 2, 1 2j.
An entire row of zeros. An entire row of zeros indicates a symmetry in the distribution of roots. To
continue with the procedure, one replaces this row of the Routh array using coefficients from the derivative
of an auxiliary polynomial.
4

Example: Determine how many roots of the polynomial

s5 + 2s4 + s + 2.
have positive real part. We may rewrite this polynomial as
1s5 + 2s4 + 0s3 + 0s2 + 1s + 2.
The Routh array terminates too early with a zero row
s5
s4
s3

1
2
0
0

0
0

1
2

2112
2

s2
s1
s0
The auxiliary polynomial P (s) is formed from the row preceding the zero row:
P (s) = 2s4 + 0s2 + 2
= 2(s4 + 1).
As an aside, note that four of the roots of the original polynomial may be obtained from the roots of P (s).
To continue with the Routh procedure, take the derivative of P (s):
d
P (s) = 8s3 .
ds
We may continue the Routh array by substituting the coefficients of
s5
s4
s3
s2

1
2
8
0

s1

082

16

s0

0
0
0

dP
ds .

1
2

8220
8

There are two sign changes so there are two roots with positive real part. One may check that the roots
are
s = 2, 1 j

References
[1] L. Cesari. Asymptotic Behavior and Stability Problems in Ordinary Differential Equations. SpringerVerlag, New York, 3rd edition, 1971.
[2] K. Ogata. Modern Control Engineering. Prentice Hall, Englewood Cliffs, NJ, 4th edition.

Lecture 24: Root Locus Diagrams

Absolute and Relative Stability. A LTI control system is called absolutely stable if the controlled
transfer function Hd (s) from the reference signal yd (s) to the output signal y(s) has all of its poles in the
open left half plane. One technique for determining absolute stability of a control system is the RouthHurwitz stability analysis technique. This very useful technique is presented in Section 5-7 of [1].
Absolute stability is an essential quality for a control system, but it says nothing about the performance
characteristics of the system, i.e., the transient response. Two absolutely stable systems can respond to
a step input in very different ways; one might exhibit a very slow, overdamped response while the other
exhibits a very fast, underdamped response.
To compare the performance of two absolutely stable systems, it is useful to consider the notion of relative
stability or degree of stability. Degree of stability can be rather narrowly defined as the horizontal
distance between the imaginary axis and the nearest pole. This distance will typically determine the speed
of response of the system, however it tells you nothing more about the nature of that response (e.g., if
it the system is overdamped, critically damped, or underdamped). More generally, one may examine the
specific locations of the closed-loop poles. Knowing these pole locations gives a good sense of the nature
of the systems transient response.
The Root Locus Method. The root locus method, also known as Evans rules in honor of W. R.
Evans, is a technique for determining how the poles of a feedback control system move in the complex plane
as a parameter is varied. Typically, the parameter is a control gain, although any parameter of interest
can be used. (For this reason, the root locus method is useful in dynamical system theory, where one is
often interested in sudden changes in a systems qualitative behavior, called bifurcations, as a parameter
varies.)

yd

Fd

Figure 1: One degree of freedom closed-loop control structure.

Consider the simple feedback control system shown in Figure 1. The closed-loop transfer function is
Hd (s) =

y(s)
P (s)Fd (s)
=
.
yd (s)
1 + P (s)Fd (s)

Closed-loop poles are values of s for which

1 + P (s)Fd (s) = 0.
Since P (s)Fd (s) is a function of a complex variable, the equation P (s)Fd (s) = 1 can be expressed in
terms of the magnitude and phase of P (s)Fd (s):
|P (s)Fd (s)| = 1

and

P (s)Fd (s) = (2k + 1) k = 0, 1, 2, . . .

In words, the magnitude of the loop gain is always one and the phase is an odd power of .
Suppose that P (s)Fd (s) can be written in the form
P (s)Fd (s) = K

b(s)
.
a(s)

This would be the case, for example, if P (s) = b(s)/a(s) and Fd = K, as for a simple proportional
controller. The control structure might be more complicated than this, however we assume that a the
multiplicative factor K appears and that this parameter may vary.
The root locus is the locus of possible roots of the closed-loop transfer function as the multiplicative
parameter K is varied. In fact, the entire root locus can be determined from the angle condition alone. The
magnitude condition is then used to determine which value of K corresponds to which set of closed-loop
poles along the locus of all possible closed-loop poles.
Rather than learn Evans rules to begin with, it is more illustrative to consider a series of increasingly
complicated examples.
Example 1. To begin, we consider the very simple example
P (s)Fd (s) = K

1
.
s(s + 2)

We will compute the closed-loop poles as explicit functions of K. In general, this is a tedious, and
uninformative exercise, but for this simple system it serves to illustrate how closed-loop poles vary as the
gain K is varied. The closed-loop transfer function is
Hd (s) =

K
s(s+2)
K
+ s(s+2)

The closed-loop poles are obtained from

K
s(s + 2)
= s2 + 2s + K.

0 = 1+

They are

1
(2 4 4K)
2

= 1 1 K.

s =

Im

Re

Figure 2: Root locus example #1.

2

When 0 < K < 1, there are two distinct poles which are located on the real axis between 0 and 2. When
K = 1, the poles coalesce at s = 1. As K continues to increase, the poles split apart and move in opposite
directions parallel to the imaginary axis.
To see that the locus of closed-loop poles shown in Figure 2 can be obtained from the angle condition

1
= (2k + 1) k = 0, 1, 2, . . . ,
s(s + 2)

we first recall some facts about complex numbers. First, a complex number can be represented in polar
form, for example z = rei where r is the radial distance from the origin to the point z and is the angle
to z measured counter-clockwise from the positive real axis. Consider the complex function
C(s) =

(s z1 ) (s zm )
.
(s p1 ) (s pn )

Each term in the numerator can be considered a vector from the zero zi to the point s. Similarly, each
term in the denominator can be considered a vector from the pole pi to the point s. Each of these vectors
has a magnitude and an angle, so we may equivalently write

rz1 eiz1 rzm eizm

C(s) =
rp1 eip1 rpn eipn

rz 1 r z m
=
ei(z1 ++zm p1 pn )
r p1 r pn
where rzi (or rpi ) is the magnitude of the vector from zi (or pi ) to s and zi (or pi ) is the angle of the
vector from zi (or pi ) to s.
Im

qp2

qp1

Re

Figure 3: Angle condition for root locus example #1.

Applying these observations to the current example, we find that

1
= s (s + 2).
s(s + 2)

(1)

Now, for any point on the real axis to the right of p1 = 0, equation (1) gives zero, which is not an odd
number times . Similarly, for any point on the real axis to the left of p2 = 2, equation (1) gives 2,
3

which is also not an odd number times . Thus, the real axis to the left of p2 and to the right of p1 is not
part of the root locus. However, for points between p2 and p1 , equation (1) gives

1
= s (s + 2) = 0,
s(s + 2)

which is an odd number times . Thus, points on the real axis between p2 and p1 are part of the root
locus.
1
. The
Considering next the points on the vertical line s = 1, we choose a point and determine s(s+2)
vectors from p1 and p2 to any such point form an isosceles triangle. The sum of the two angles is for
1
points above the real axis and 3 for points below the real axis, giving s(s+2)
= or 3, respectively.
Thus, the line s = 1 is part of the root locus.

To find the value of K which corresponds to a particular pair of closed-loop poles, we use the magnitude
condition.For example, suppose we would like to choose K so that the closed-loop system has adamping
2
ratio = 22 . Any pole lying on the radius = 3
4 in the complex plane has damping ratio = 2 . Thus,
we would like to choose K to give closed-loop poles at
s = 1 i tan

= 1 i.
4

Choosing a particular pole, say s = 1 + i, we substitute into the magnitude condition to obtain

(1 + i)((1 + i) + 2) = 1
or

K = |(1 + i)(1 + i)| = | 2| = 2.

Thus, choosing the gain K = 2 gives the closed-loop poles s = 1 i.

We have assumed that P (s)Fd (s) can be written in the form
P (s)Fd (s) = K

b(s)
.
a(s)

where b(s) has degree m, a(s) has degree n m and where K > 0 is a parameter (e.g., a control gain)
which may vary.
An important observation is that, as K 0, the closed-loop poles approach the poles of the loop gain. To
see this, write the closed-loop characteristic equation as
a(s) + Kb(s) = 0.
Clearly, as K 0 the roots of the polynomial on the left approach the roots of a(s).
One may also observe that, as K , the closed-loop poles must either diverge to or approach a zero
b(s)
of the loop gain. To see this, recognize that as K , a(s)
must become very small so that the product

b(s)
is 1. There are two ways that a(s)
can become very small. First, b(s) can go to zero (which happens
when s approaches a zero of the loop gain). Second, a(s) can go to infinity (which can only happen when
|s| goes to infinity.) In general, m branches of the root locus approach the zeros of the loop gain while the
remaining n m branches go to infinity.

Example 2. Consider the following example from [1]:

P (s)Fd (s) =

K
.
s(s + 1)(s + 2)
4

This system has poles at p1 = 0, p2 = 1, and p3 = 2. Recalling that

P (s)Fd (s) =

m
X
i=1

(s zi )

n
X
j=1

(s pj ),

we first consider which, if any, points on the real axis are part of the root locus. For any point to the
right of s = 0, P (s)Fd (s) = 0, so the positive real axis is not part of the root locus. For any point
1 < s < 0, P (s)Fd (s) = , so these points are part of the root locus. For any point 2 < s < 1,
P (s)Fd (s) = 2, so these points are not part of the root locus. Finally, for any point s < 2,
P (s)Fd (s) = 3, so these points are part of the root locus.
Next, we consider what happens to the root locus as s grows large. In the limit that s grows large, we have
K
K
= lim
.
s(s + 1)(s + 2) |s| s3

lim P (s)Fd (s) = lim

|s|

|s|

Now, no matter how large |s| is, the angle condition must be satisfied, so we must have
lim P (s)Fd (s) =
|s|

lim P (rei )Fd (rei )

lim

K
(rei )3

= ei3
= (2n + 1)

n = 0, 1, 2, . . .

or

2n + 1
.
3
Trying n = 0 gives = 3 . Trying n = 1 gives = . Trying n = 2 gives = 5
3 . Other choices of n
give repeated angles. In the limit that |s| , the three closed-loop poles follow asymptotes that extend
radially in the directions 3 and .
=

Im

Re

Figure 4: Root locus example #2.

Two of the three asymptotes extend into the right half complex plane, while the third follows the negative
real axis. Intuitively, the closed-loop pole which starts (for small K) at s = 2 will follow the negative real
axis asymptote as K increases. Therefore, the two closed-loop poles which rest on the real axis between
s = 1 and s = 0 must coalesce and split off to follow the asymptotes at 3 .1
1
They must first coalesce because poles must be either real numbers or complex conjugate pairs and because the closed-loop
pole locations vary continuously with K.

Note: You can find the value of the gain K at which the root locus passes into the right half plane by
performing a Routh-Hurwitz stability analysis and finding conditions on K for stability.

References
[1] K. Ogata. Modern Control Engineering, Fourth Ed. Prentice Hall, Upper Saddle River, NJ, 2002.