L1 A3-401 - Newtons Law of Gravitation

The N-body problem

We are interested in the motion of n particles or masses interacting with each other via
mutual gravitation. We will denote the mass of each particle as m
i
and the position and
velocity of each particle as r
i
and v
i
respectively
1
.
We can denote the distance between two bodies i and j as
r
ij
:= r
i
r
j
=
_
(r
i
r
j
)

(r
i
r
j
).
Newtons law of universal gravitation gives us the total force of attraction experienced by
each of the bodies due to the others. The force experienced by body i due to the attraction
of body j is Gm
i
m
j
/r
2
ij
. The constant G is the universal gravitational constant. In SI
units, G 6.67 10
11
N m
2
kg
2
. The force on body i is directed radially towards
particle j, i.e. in the direction of the unit vector (r
j
r
i
)/r
ij
.
_
In this course we will ignore any additional forces besides gravity that may act on
the particles unless otherwise specied. Such forces can include atmospheric drag
forces, solar pressure, use of thrusters on spacecraft, etc.
_
If we have a system of N particles and denote the total force acting on body i as f
i
, then
the above gives us the total force applied to body i due to the other N 1 bodies:
Newtons Law of Universal Gravitation for N bodies For a system of N particles
each with mass m
i
and position r
i
, the total force experienced by particle i is:
f
i
=
N

j=1, i=j
G
m
i
m
j
(r
j
r
i
)
r
3
ij
Newtons second law of motion tells us that the acceleration of each particle is
f
i
= m
i
dv
i
dt
= m
i
d
2
r
i
dt
2
.
Combining this with the law of gravitation of above, we get the following:
1
In general these notes will follow the notation in the text by Battin (see suggested reading list), with
vector quantities in bold, their associated magnitude in plain font, and unit vectors denoted as i with an
appropriate subscript, i.e. for vector r,
r := r , i
r
:=
r
r
.
L1 A3-401 - Newtons Law of Gravitation
Equations of motion for N bodies For an N-body system, the equations of motion are:
d
2
r
i
dt
2
=
N

j=1, i=j
G
m
j
(r
j
r
i
)
r
3
ij
Important: The only physics used in this course are Newtons laws of motion and New-
tons law of universal gravitation. Everything else in the course is based on mathemati-
cal manipulation of the above. If you are extremely good at maths, you can stop reading
now!
The above describes a set of second order, coupled, nonlinear differential equations for a
set of N bodies. Except in certain special cases, there is no closed-form solution for these
equations. However, if the problem is restricted to two bodies, we can solve the equations
of motion explicitly.
The 2-body problem
For a system of two bodies, our equations of motion are:
m
1
d
2
r
1
dt
2
= G
m
1
m
2
(r
2
r
1
)
r
3
12
m
2
d
2
r
2
dt
2
= G
m
1
m
2
(r
1
r
2
)
r
3
21
.
However, we generally are more interested in the relative motion of the two bodies. We
can therefore dene
r := r
2
r
1
v := v
2
v
1
r := r = r
2
r
1

:= G(m
1
+ m
2
)
We can then take the difference between the two equations above, to arrive at the funda-
mental differential equation of orbital mechanics for the two-body problem:
d
2
r
dt
2
+
r
r
3
= 0.
Note that we do not assume that either mass is xed in space.
L1 A3-401 - Newtons Law of Gravitation
Integrals of Two-Body Motion
By taking various integrals of the two-body equations of relative motion, we can nd sev-
eral constants of motion for the system. The constants of motion will help us to characterize
completely the solution of these equations.
The Angular Momentum Vector
We can take the cross product of relative position and acceleration and integrate to arrive
at our rst constant of orbital motion. First take the cross product
d
dt
(r v) = r
dv
dt
= r
_

r
3
r
_
= 0,
then integrate once to get
r v = h = constant.
The vector h is the (massless) angular momentum vector. The motion of the two bodies
takes place in the plane normal to h.
The Eccentricity Vector
We next take the cross product of h with dv/dt and apply various vector identities along
with the fundamental equation of dynamics above, to obtain
d
dt
_
v h
r
r
_
= 0.
Verication of the above is left as a tutorial exercise. We then integrate once to obtain
another constant, this time the Laplace vector e:
v h
r
r
= e = constant.
The vector e is known more commonly as the eccentricity vector, and its magnitude e as
the eccentricity of the orbit. The eccentricity vector points in the direction of periapsis, i.e.
the direction in which the two bodies make their closest approach. (An apse is a position
in an orbit where the position and velocity vectors are orthogonal to each other).
L2 A3-401 - Elliptic Orbits and Keplers Equation
So far we have identied 6 constants of integration for the 2nd order differential equation
describing the motion of 2-bodies, i.e. we have found the constant vectors e and h. How-
ever, these two vectors are not independent (i.e. h e = 0), so there is still one constant
missing.
Time of pericenter passage
The missing constant is the one relating time to orbital position. Using the equation of orbit
from the previous lecture, we can write
_

p
3
dt =
df
(1 + e cos f)
2
.
If we integrate this equation once (not easy!), we get a relationship between the true
anomaly f and the time t, plus our nal constant of integration. The usual choice for
this constant is the time of pericenter passage, denoted , which gives the time of closest
approach of the two bodies.
The eccentric anomaly
In order to nd a solution to the above differential equation, we will rst require a few new
denitions relating to the geometry of our orbit. In particular, we will dene a new angle
E, known as the eccentric anomaly. The relation of this angle to the true anomaly f is
shown in the gure below:
b
a
P
Q
f E
r
ae
S
V
F O
x
y
p
T
U
L2 A3-401 - Elliptic Orbits and Keplers Equation
In this gure, we can see the following features of the orbit:
a length of the semimajor axis (same as the distance OV )
b length of the semiminor axis
p the parameter
F the focus of the orbit
T the vacant focus
O the center of the orbit
U apoapsis
V periapsis
f the true anomaly
E the eccentric anomaly
The outer circle in the gure (whose radius matches the semimajor axis of the orbital el-
lipse), is known as the auxiliary circle, and is a useful device for determining orbital posi-
tion as a function of time.
We rst nd a relationship between the true and eccentric anomalies. From the gure,
cos E =
ae + r cos f
a
.
From the equation of orbit, we also know that
r =
a(1 e
2
)
1 + e cos f
.
Combining the two, we nd the relationship between E and f:
cos E =
e + cos f
1 + e cos f
, cos f =
cos E e
1 e cos E
Keplers Equation
The reason for introducing the eccentric anomaly is that we can relate it to an explicit
function of time. First note that the ratio of y positions of points on the ellipse to points on
the auxiliary circle (e.g. the ratio of the lengths SP/SQ) is always b/a.
Recall from Keplers second law that equal areas of the ellipse are swept out in equal times.
We would like to nd the area swept out by the true anomaly f, i.e. the area of the sector
of the ellipsoid bounded by the points FPV , which we will denote .
If we denote the area of the triangle SPF as , then simple geometric arguments show that
=
ab
2
(e sin E cos E sin E)
L2 A3-401 - Elliptic Orbits and Keplers Equation
We can then express the sum + as a fraction of the area of the auxiliary circle swept
out by E minus the area of triangle OSQ.
b
a

_
E
2
a
2

a
2
2
sin E cos E
_
= + ,
so that
=
ba
2
(E e sin E).
From Keplers second law (equal areas are swept out in equal times), we can conclude that

ab
=
(t )
P
,
where P = 2
_
a
3
/ is the orbital period and is the time at periapsis. The resulting
expression is Keplers equation:
M =
_

a
3
(t ) = E e sin E.
The term M is the mean anomaly, and can be thought of as the position of a point that
travels the auxiliary circle at constant speed and with period P.
Keplers Equation relates the mean anomaly M to the eccentric anomaly E:
M = E e sin E.
Solving Keplers Equation
The advantage of the above equation is that it is very easy to nd a value for M given
the current time and the time at periapsis. If we can solve Keplers equation for E (and
consequently f) given M, then we can determine the position of an orbiting body as a
function of time. Unfortunately, solving Keplers equation is not straightforward.
There are a huge number of methods for solving this equation (see, for example, all of
chapter 5 of the book by Battin). The most straightforward of these (but certainly not the
best) is the method of successive substitution. Starting from E
0
= 0, assign:
E
k
M + e sin E
k1
,
which is guaranteed to converge (cf. Battin 5.1).
L3 A3-401 - Coordinate Systems and Orbital Elements
Inertial Reference Frames
In order to characterize the position and velocity of a satellite relative to the earth (or any
other central body), we require an inertial frame of reference. Dening such a frame of
reference (e.g. for satellites orbiting Earth) is not straightforward, because the orbit plane
and spin axis of the earth are not xed. Two of the most important sources of difculty are:
Precession of the Equinoxes: The motion of the Earths spin axis. The spin axis is
inclined relative to the Earths orbit around the sun at an angle of 23.5

(which is
time-varying), and this vector traces a cone in space with a period of 26, 000 years.
Precession of the Ecliptic: The ecliptic plane is the plane of the Earths rotation
about the sun (and is the plane in which all of constellations of the zodiac are found).
The ecliptic plane is not xed in space, because the angular momentum vector of the
Earths orbit drifts with a period of 100, 000 years.
The vernal equinox is the point in the Earths orbit where the plane of the equator intersects
the ecliptic (i.e. where the Earth is not tilted towards or away from the sun). The vernal
equinox is time varying due to the precession of the equinoxes. It is also called the rst
point of Aries, although this point no longer corresponds exactly to the vernal equinox.
The invariable plane is the plane dened by the total angular momentum of the solar sys-
tem, and is dominated by the angular momentum of the planet Jupiter. The earths orbit is
inclined to the invariable plane by 1

.
Standard inertial reference frames
Given the above, it is important to be very careful when dening coordinate systems for
use in orbital navigation. The one in most common use is the J2000 system, which uses:
i
x
vector in the direction of the mean vernal equinox on
1 January 2000 at 12:00:00.00. (Sometimes denoted
as in reference to the constellation Aries).
i
z
vector in the direction of the Earths mean rotation axis on
1 January 2000 at 12:00:00.00.
i
y
completes the set, i.e. i
y
= i
z
i
z
.
The J in J2000 stands for Julian Date, which is the number of days elapsed since January 1,
4713 BC at noon, so that the J2000 system is relative to Julian Date 2451545.0. It is
customary for inertial reference systems to be updated every 50 years starting in 2025
most calculations will be made with respect to the J2050 coordinate system.
L3 A3-401 - Coordinate Systems and Orbital Elements
Orbital Elements
In order to describe the location of a satellites orbit relative to the earth (or any other central
body), we require 6 parameters, corresponding to the 6 constants of integration from the
equation of motion in the two body problem, plus a reference time.
It is of course possible to just specify a position and velocity pair (r, v) and a reference,
but this does not give a clear picture of the shape of the orbit. It is therefore more common
to use the following set of orbital elements:
a the semimajor axis
e the eccentricity
f the true anomaly
i the orbital inclination
the argument of periapsis
the right ascension of the ascending node (RAAN)
The rst three values above describe the shape of the orbit (a, e) and the location of the
orbiting body within this shape (f). The other three values describe the orientation of the
orbit plane relative to the inertial frame of reference (see gure below). The ascending
node (denoted ) is in the direction of the intersection of the orbit plane with the reference
plane, on the ascending side of the orbit.
i
h
f

i
x
i
z i
e
r
i

Figure : The classical orbit elements

L3 A3-401 - Coordinate Systems and Orbital Elements
In the previous gure, the position of a body in the orbital frame of reference is given by:
r
orb
=
a(1 e
2
)
1 + e cos f

_
cos f i
e
+ sin f i
p
.
_
This can then be converted into the inertial reference frame using a standard coordinate
transformation:
r
inertial
= Ar
orb
where
A =
_
_
i
x
i
e
i
x
i
e
i
x
i
e
i
y
i
p
i
y
i
p
i
y
i
p
i
z
i
h
i
z
i
h
i
z
i
h
_
_
=
_
_
cos cos sin sin cos i cos sin sin cos cos i sin sin i
sin cos + cos sin cos i sin sin + cos cos cos i cos sin i
sin sin i cos sin i cos i
_
_
.
L4 A3-401 - Preliminary Orbit Determination
Gibbs Method
In this section, we will describe a method for determining an orbit based on three position
measurements.
Assume that we have three position measurements (e.g. radar measurements of a satellite
from a ground station) we will label these measurements (r
1
, r
2
, r
3
) respectively, and
assume that they are taken relatively close together (i.e. within a space of time less than
1
2
the orbital period). We also assume that the vectors are coplanar, i.e.
(r
1
r
2
) r
3
= 0.
Gibbs Orbit Determination Algorithm
1. Calculate a vector normal to the orbit plane (and therefore in the same direction
as the angular momentum h):
n = r
1
r
3
.
2. Write the middle vector as a linear combination of the other two:
r
2
= r
1
+ r
3
,
where
= (r
2
r
3
)
n
n
2
, = (r
1
r
2
)
n
n
2
.
3. Solve for the parameter p of the orbit using
p =
r
2
r
1
r
3
1 .
4. Find the eccentricity vector e of the orbit using
e =
1
n
2
[(r
3
n)(p r
1
) (r
1
n)(p r
3
)] .
5. Find the true anomaly for the orbit, at a time corresponding to measurement
of r
3
, using
cos f = i
e
i
r
3
L5 A3-401 - The Initial Value Problem
The Velocity Vector
In previous lectures we have shown how to calculate the orbital position and orbital el-
ements by taking various dot and cross products of the orbital equation of motion and
quantities derived from it. We now do this one more time to obtain an expression for the
velocity vector v as a function of orbital constants and the position vector r.
First recall that the Laplace vector was dened as
e = v h
r
r
.
We can take the cross product of this vector with the angular momentumh to obtain
h e = h (v h) h
_

r
r
_
.
Then apply the vector triple product identity
a (b c) = b(a c) c(a b),
so that the above simplies to
h e = vh
2

r
(h r).
We can then rearrange this to obtain the velocity in isolation:
v =

h
2
h
_
e +
r
r
_
.
Note that in the above, the velocity is actually a function of the orbital constants (h, e, )
and the direction vector r/r = i
r
only, and not the actual position.
The Initial Value Problem
We now have all of the pieces required to propogate the position and velocity (r(t), v(t))
of a satellite from an initial position and velocity pair (r
0
, v
0
).
We rst decompose the position and velocity vectors into components in the orbit plane,
so that
r = r cos f i
e
+ r sin f i
p
and
v =

h
i
h
(e i
e
+ cos f i
e
+ sin f i
p
)
v =

h
sin f i
e
+

h
(e + cos f) i
p
.
L5 A3-401 - The Initial Value Problem
r
i
p
i
e
i
h
f
These can then be written in matrix form as
_
r
v
_
=
_
r cos f r sin f

h
sin f

h
(e + cos f)
_ _
i
e
i
p
_
.
The important thing to note about the equation above is that (r, v) varies only with the true
anomaly f, since the vectors (i
e
, i
p
) are constant. Writing the matrix term above as (f),
we have
_
r
v
_
= (f)
_
i
e
i
p
_
.
We now ask whether we can use the above relationship to nd a mapping (r
0
, v
0
) (r, v).
If we denote the true anomaly at time 0 as f
0
(i.e. the true anomaly at which the position /
velocity vectors are (r
0
, v
0
)), then we can invert the matrix (f
0
) to get
_
i
e
i
p
_
=
1
(f
0
)
_
r
0
v
0
_
.
In order to obtain an explicit expression for
1
(f
0
), we note that (f
0
) is a 2 2 matrix,
so its inverse can be written as

1
(f
0
) =
1
det (f
0
)
_

h
(e + cos f
0
) r sin f
0

h
sin f
0
r cos f
0
_
,
where the determinant simplies to
det (f
0
) =
r
0

h
_
cos
2
f
0
+ e cos f
0
+ sin
2
f
0
_
=
r
0

h
(1 + e cos f
0
)
=
r
0

h
p
r
0
=
r
0

h
h
2

= h
L5 A3-401 - The Initial Value Problem
We can therefore write the constant vectors (i
e
, i
p
) as
_
i
e
i
p
_
=
1
(f
0
)
_
r
0
v
0
_
=
_

h
2
(e + cos f
0
)
r
h
sin f
0

h
2
sin f
0
r
h
cos f
0
_ _
r
0
v
0
_
Finally, we can combine all of the above to obtain the mapping (r
0
, v
0
) (r, v):
_
r
v
_
= (f)
1
(f
0
)
_
r
0
v
0
_
Calculating (r, v) as a function of time
The expression (f)
1
(f
0
) is a 2 2 matrix mapping initial and nal conditions as a
function of the initial and nal true anomalies f
0
and f. The form of this expression above
is generic; it can be used for any elliptical, parabolic or hyperbolic orbit. However, if one
wishes to determine (r, v) at some specic future time, then it is still necessary to obtain a
transformation from t f. For elliptical orbits, this can be done via Keplers equation. In
the parabolic or hyperbolic case, the transformation t f requires solving either Barkers
equation (for a parabolic orbit), or the Gudermannian equation (for a hyperbolic orbit).
Lagrange Coefcients
If we form the matrix product (f)
1
(f
0
), we get the Lagrange coefcients F, G, F
t
, G
t
:
(f)
1
(f
0
) =
_
F G
F
t
G
t
_
,
so that the position and velocity can be written as
r = Fr
0
+ Gv
0
v = F
t
r
0
+ G
t
v
0
L6 A3-401 - Enckes Method
Non-Keplerian Orbits
Thus far we have focussed on the classical two-body problem, where the dynamics were
described by
d
2
r
dt
2
+
r
r
3
= 0.
Although it is not possible to nd a closed-form solution to this problem, we were able to
nd a number of constants of integration, and to thereby showthat solutions to this equation
result in the classical conic orbits described (in part) by Kepler.
We now address the problem of nding solutions to problems of orbital motion that contain
a disturbing acceleration a
d
, so that the above dynamic equation becomes
d
2
r
dt
2
+
r
r
3
= a
d
.
In the above, the disturbance acceleration term can represent result from gravitational
forces from third bodies (e.g. the sun, moon or other planets for an Earth-orbiting satel-
lite), disturbances due to solar pressure, tides, a non-spherical central body, atmospheric
drag etc. The term a
d
can also be treated as an engine thrust term in problems of orbit
targeting or spacecraft station-keeping.
In any of the above cases, the problem of nding a solution to the above ODE becomes
much more complicated, and can only be solved numerically in most cases. However,
we wish to make as much use as possible of the solution to the non-disturbed Keplerian
solution to aid in this numerical solution. The procedure we use to do this is known as
Enckes method.
Enckes Method
Partition the position r and velocity v into reference terms (r
o
,v
o
) and perturbation terms
(, ):
r = r
o
+
v = v
o
+
We assume that the reference trajectory r
o
(t) represents the solution to the unperturbed
problem
d
2
r
o
dt
2
+
r
o
r
3
o
= 0.
If we subtract the dynamics of the reference trajectory from the dynamics of the true tra-
jectory r(t) and collect all terms in on the left-hand side, we are left with
d
2

dt
2
+

r
3
o
= a
d
+

r
3
o
_
1
r
3
o
r
3
_
r.
L6 A3-401 - Enckes Method
The rst term on the right hand side represents an acceleration due to external forcing. The
second term accounts for incremental changes to the gravitational force when r = r
o
. If
we integrate this ODE, we get an exact value for the difference between the nominal
trajectory r
o
and the true trajectory r. Enckes method for orbit propagation uses this idea
in the following algorithm:
Enckes Method
1. Calculate a reference trajectory r
0
(t) over some interval t [0, T] using
Keplers equation / Lagrange coefcients.
2. Integrate (t) over the interval t [0, T] with initial condition (0) = 0.
3. Rectify the two orbits by setting r
0
(T) r
o
(T) +(T)
4. use r
o
(T) as a new initial condition and repeat.
Enckes method creates a sequence of diverging and re-converging trajectories as shown in
the gure below. These trajectories are referred to as osculating orbits (from the Latin verb
osculari to kiss).
r
r
o

Figure : Osculating Orbits

The above procedure is an exact method and will work no matter how large is allowed
to get. However, in practice the idea is to employ this method only when a
d
(and therefore
) is relatively small. The main advantage of this method is that one can use a much larger
step size in solving for than would be possible if the original ODE in (r, a
d
) were to be
solved directly.
Note that in the above algorithm, it is implied that the rectication step happens at xed
time intervals of size T. In practice one might instead choose to implement the rectication
step when reaches some threshold value relative to r.
L6 A3-401 - Enckes Method
Improving Numerical Accuracy in Enckes Method
Although the method above is exact, the term
_
1
r
3
o
r
3
_
can cause difculties due to numerical accuracy problems when r
o
r. We can get around
this difculty by rewriting the awkward term above as
r
3
o
r
3
=
_
(r
o
+) (r
o
+)
r
2
o
_

3
2
=
_
1 +
(2r
o
+)
r
2
o
_

3
2
.
If we dene a new variable q = q(, r
o
) as
q :=
(2r
o
+)
r
2
o
,
and dene a new function f such that
f(q) = 1 (1 +q)

3
2
=
_
1
r
3
o
r
3
_
,
then we can rewrite our ODE for the perturbation term as
d
2

dt
2
+

r
3
o
= a
d
+

r
3
o
f(q)(r
o
+),
with initial condition (0) = 0. The numerical conditioning of the above ODE is generally
better than the original form, although it is still a nonlinear ODE that must be integrated
numerically.
L7 A3-401 - The Gravity Gradient
In the previous lecture we used Enckes method to integrate exactly the ODE
d
2
r
dt
2
+
r
r
3
= a
d
.
This was done by partitioning the position and velocity vectors as
r = r
o
+
v = v
o
+
and integrating the disturbance terms separately from the nominal components (r
o
, v
o
) that
would be found in Keplerian motion.
All of the difculties that arise in integrating this equation result from the fact that the
gravity term
g(r) =
r
r
3
is nonlinear. In this lecture we develop an approximate method to calculate perturbations
to the position r and velocity v when a disturbing acceleration is present. Our approach
will be based on linearization of the gravity term.
The Gravity Gradient
We can approximate the gravitational acceleration at a point r = r
o
+ as
g(r) g(r
o
) + G(r
o
) +O(
2
),
where the term
G(r) = g(r) =
g(r)
r
is the gravity gradient, and the term O(
2
) represents all of the higher order terms. If is
small, then the dynamics of our system can be written as
d
2
r
o
dt
2
+
d
2

dt
2
= g(r
o
) + G(r
o
) +a
d
.
If we assume that the reference trajectory r
o
is the solution to
d
2
r
o
dt
2
= g(r
0
),
then these terms can be cancelled, leaving
d
2

dt
2
= G(r
o
) +a
d
.
The critical difference between this ODE from and the one used in Enckes method is that
the above, although approximate, is linear, and therefore much easier to solve.
L7 A3-401 - The Gravity Gradient
State-Space Form of the Perturbation Errors
An important application of the above linearization is for spacecraft control. If we rewrite
the dynamics of the position error in terms of the velocity error term , we get
d
dt
=
d
dt
= G(r
o
) +a
d
.
The above can be rewritten as
d
dt
_

_
=
_
0 I
G(r
o
) 0
_ _

_
+
_
0
I
_
a
d
,
which is in standard state-space form. In the above the disturbing acceleration a
d
is best
viewed as a thrust that can be used to force the position error to zero for either trajectory
following (e.g. a moon landing problem) or stationkeeping (e.g. to keep a geostationary
satellite at its desired location).
Properties of the Gravity Gradient
We now work out in detail some of the properties of the gravity gradient term G(r
0
):
G(r) =
g(r)
r
=

r
_

r
r
3
_
=
1
r
3

_
3rr

r
2
I
_
,
where the term rr

is the outer product of the vector r with itself, i.e. a 33 matrix formed
by the multiplication of column by a row.
Note that if the nominal position vector r
o
is written as
r
o
=
_
_
r
x
r
y
r
z
_
_
,
then the above is equivalent to
G(r
o
) =

r
5
o
_
_
3r
2
x
r
2
3r
x
r
y
3r
x
r
z
3r
x
r
y
3r
2
y
r
2
3r
y
r
z
3r
x
r
z
3r
y
r
z
3r
2
z
r
2
_
_
.
L7 A3-401 - The Gravity Gradient
Eigenvectors of the Gravity Gradient
We next determine the eigenvectors of the gravity gradient matrix G(r
o
) in order to under-
stand its effect on the perturbed position . First note that the product of G(r
0
) with r
o
is
G(r
o
)r
o
=
2
r
3
o
r
o
.
We therefore conclude that r
o
is an eigenvector with corresponding eigenvalue 2/r
3
o
> 0.
The other two eigenvectors are both orthogonal to r
o
denote either of them as r

, so that
G(r
o
)r

=

r
3
o
.
In this case we conclude that any vector orthogonal to r
o
is also an eigenvector, with corre-
sponding eigenvalue /r
3
o
< 0.
We therefore conclude that any perturbation that is in the same direction as the refer-
ence position r
o
results in a further radial acceleration away from the reference (e.g. a
perturbation upwards results in further acceleration up and away from the reference).
Any perturbation in an orthogonal direction results in an acceleration back towards to the
reference position.
L8 A3-401 - The Clohessy-Wiltshire Equations
In this lecture we consider the relative motion of two objects (e.g. two satellites) orbiting a
central body. We assume that the mass of both objects is negligible relative to the mass of
the central body.
We designate one of the two satellites as the reference object, and denote its position as r
o
.
We assume that the other object is in a nearby orbit with position r = + r
o
, and assume
that r
o
.
Recall from the previous lecture that, using a linear approximation, the dynamics of the
relative position vector can be written as
d
2

dt
2
= G(r
o
).
The matrix term G(r
o
) is the gravity gradient evaluated at the reference position r
o
and can
be written as
G(r
o
) =

r
3
o
_
3r
o
r

o
r
2
o
I
_
.
Relative Position in a Rotation Reference Frame
In order to describe the relative position of the two bodies, we will nd it convenient to
work in frame of reference that is xed with respect to the reference object, as shown in the
gure below:
i
y
i
x
direction of motion
i
z
r
o

The vector i
x
represents the local vertical direction. The vectors (i
y
, i
z
) are locally hori-
zontal, with i
y
facing in the direction of orbital motion and i
z
facing in the direction of the
angular momentum vector of the reference orbit.
L8 A3-401 - The Clohessy-Wiltshire Equations
We assume that the reference orbit is circular, so that r
o
is constant and the period of the
reference orbit is
P = 2

r
3
o

.
The angular velocity of the rotating reference frame therefore satises
=
_

r
3
o
.
We note that the acceleration of the relative position term due to the gravity gradient has
thus far always been in an inertial frame. To emphasise this point, we write it henceforward
as
I
.
We denote the relative position in the rotating frame xed to the reference orbit as
B
. A
standard result is

I
..
gravity
=

B
..
observed
+2(

B
)
. .
Coriolis
+(
B
)
. .
Euler
+ (
B
)
. .
centripetal
.
Note that the Euler term above is zero in this case since = 0.
If we write
= i
z
, r
o
= r
o
i
x
,
B
= xi
x
+ yi
y
+ zi
z
,
then the above can be simplied to
2
2
xi
x

2
yi
y

2
zi
z
. .
gravity
= xi
x
+ yi
y
+ zi
z
. .
observed
+2 xi
y
2 yi
x
. .
Coriolis

2
xi
x

2
yi
y
. .
centripetal
.
The above vector differential equation can be separated into a set of three coupled linear
constant-coefcient differential equations, known as the Clohessy-Wiltshire equations.
Clohessy-Wiltshire Equations
The position of a satellite relative to a circular reference orbit is modelled approximately
by
x 3
2
x 2 y = 0
y + 2 x = 0
z +
2
z = 0
An important point to notice is that equation in (x, y) are coupled to each other, but not
to z. This indicates that local vertical and horizontal motions are coupled to each other and
independent of motion normal to the reference plane.
L8 A3-401 - The Clohessy-Wiltshire Equations
Note that if we wish to add disturbing acceleration terms (e.g. due to an engine on board
the satellite), then we simply add these to the right-hand side of the C-W equations, i.e.
x 3
2
x 2 y = a
x
y + 2 x = a
y
z +
2
z = a
z
.
In this case a
x
represents an acceleration in the local vertical direction, a
y
in the local
horizontal direction, and a
z
out of the reference plane.
Solution to the Clohessy-Wiltshire Equation
The C-W equations are linear constant-coefcient differential equations, and can be solved
exactly. If initial values are available for the relative position (x
0
, y
0
, z
0
) and relative ve-
locity ( x
0
, y
0
, z
0
) , then the solution to the C-W equations is
_

_
x(t)
y(t)
x(t)
y(t)
_

_
=
_

_
4 3 cos t 0
1

sin t
2

(1 cos t)
6(sin t t) 1
2

(cos t 1)
1

(4 sint 3t)
3 sin t 0 cos t 2 sint
6(cos t 1) 0 2 sin t 4 cos t 3
_

_
_

_
x
0
y
0
x
0
y
0
_

_
and
_
z(t)
z(t)
_
=
_
cos t
1

sin t
sin t cos t
_ _
z
0
z
0
_