Thurman Worfolk Geometry Halo Orbits

Download as pdf or txt
Download as pdf or txt
You are on page 1of 46

The geometry of halo orbits in

the circular restricted three-body problem


Robert Thurman and Patrick A. Worfolk
The Geometry Center
University of Minnesota
1300 S. Second St., Suite 500
Minneapolis, MN 55454
October 25, 1996

Abstract
In the summer of 1996, we supervised two undergraduate students during a nine-week
summer program at the Geometry Center. They worked on a project using dynamical
systems techniques to compute and visualize orbits in the circular restricted three-body
problem. This project was motivated by recent interest in the space science community
to send missions near to the Sun-Earth libration points. A fuller understanding of the
geometry of the phase space of the circular restricted three-body problem could provide
new possibilities for baseline trajectory design. To this end, the goal of this project
was to develop computational and visualization tools to aid in trajectory design. In
particular, we wanted to be able to easily and interactively explore the geometry of the
halo orbits and their stable and unstable manifolds. This report provides a summary of
the mathematics underlying the project and a brief discussion of the results.

Keywords: restricted three-body problem, halo orbits, (un)stable manifolds.


Contents
1 Introduction 1
2 The central force problem 2
3 The two-body problem 3
4 The circular restricted three-body problem 4
5 Libration points 8
6 The Jacobian integral 8
7 Symmetries 9
8 The Lagrangian formulation 9
9 The Hamiltonian formulation 10
10 Richardson's approximations for halo orbits 11
11 Numerical computation of halo orbits 20
12 Stability of the periodic orbits 23
13 Stable and unstable manifolds of the halo orbits 25
14 Results 26
A Astronomical constants 31
B Solving the inhomogeneous linear equation 31
C Coe cients and quantities from Section 10 33
D Solutions to exercises 36

ii
1 Introduction
For six consecutive years, the Geometry Center has sponsored a Summer Institute for undergraduate
students, who work individually or in small groups under the direction of Geometry Center sta

members. In the summer of 1996, we supervised two students for the nine-week period of the
program on a dynamical systems project. Essentially, the focus of the project was to use dynamical
systems techniques to compute and visualize new spacecraft trajectories and was motivated by our
interaction with mathematicians at NASA's Jet Propulsion Laboratory. This report provides the
mathematical background that we provided our students as well as a brief discussion of their results.
An online HTML document 9] created by our students further illustrates these results, in addition
to describing the software tools they developed.
Recently, the space science community has shown considerable interest in missions which take
place in the vicinity of the libration points of the Sun-Earth system. Designing trajectories for
these missions is challenging because conic approximations (solutions of the two-body problem) are
inadequate and, in the past, manual numerical searches have been the only recourse. Recent work of
Barden, Howell, and Lo 3] has shown that a greater understanding of the dynamics of the restricted
three-body problem could lead to innovative baseline mission concepts. Dynamical systems theory
could provide insights into the qualitative nonlinear behavior. Knowledge of the geometry of phase
space and the existence of stable and unstable manifolds which both separate regions of space and
provide natural transfer mechanisms can all aid in trajectory design. Once a preliminary trajectory
design has been accomplished within the framework of the restricted three-body problem, the nal
solution is computed using a model that incorporates ephemeris data and solar radiation pressure.
The primary goal of our project was to develop computational and visualization capabilities for
the study of trajectories near the libration points in the restricted three-body problem. This goal was
to be accomplished by extending the capabilities of the dynamical systems software package DsTool
2] and using the Geometry Center software Geomview 12] for the visualization. We wanted to
create a software environment where it was easy to interactively explore and visualize the dynamics
of the restricted three-body problem. Thus for the summer, our rst goals were to provide such an
environment for the computation of the libration points and the symmetric halo (periodic) orbits
surrounding the collinear libration points. Our \championship goals" were to compute and visualize
the stable and unstable manifolds of these halo orbits with an emphasis on being able to discover
new trajectories which would transfer a spacecraft naturally from a parking orbit about the Earth
to the vicinity of one of the libration points.
This paper contains the basic mathematics needed to accomplish the goals outlined above. Most
of the mathematics is generally well-known, though it is sprinkled throughout the literature. We
have thus tried to bring together the basic ideas, in a manner which is accessible to undergraduates.
The details are sketchy in many parts, but all the ideas we found necessary to formally convey to
our students have been included. We have ignored the long history of the study of the restricted
three-body problem which is covered quite completely by Szebehely 17]. We have also failed to give
a complete description of Lagrangian and Hamiltonian mechanics and any formal introduction to
dynamical systems.
An overview of the contents is as follows. In Sections 2 and 3, we discuss the central force

1
problem and the two-body problem. From here, we derive the equations of motion for the circular
restricted three-body problem in Section 4. In the next section we calculate the libration points and
then we present the Jacobian integral. Following this we discuss the symmetries of the equations
of motion, since we will take advantage of these in our numerical calculations. In Section 8, we
present the Lagrangian formulation of the equations of motion, and in the following section, the
Hamiltonian formulation. We present Richardson's technique for analytically approximating halo
orbits in Section 10, including many of the gory details not included in his published work. In
working through this, we discovered an error in his calculations and present the corrected version
here. Following this is a section on the numerical computation of periodic orbits. In particular,
an algorithm is given for the computation of the symmetric halo orbits in which we are interested.
In Section 12 we present an ecient method for computing the monodromy matrix of the periodic
halo orbits. Following this we give a naive, yet e
ective, algorithm for approximating the stable and
unstable manifolds of the halo orbits. Finally, we discuss some of the results of our students' work
and point out a variety of interesting questions which we were unable to resolve in the nine-week
period.
We found it useful to provide our students (and ourselves) with a number of exercises as a means
of \getting our hands dirty" with the background material. We have left these exercises embedded
in the document. Solutions are given in Appendix D.
Acknowledgments. Motivation for this project originated from discussions we had with sci-
entists and mathematicians at the Jet Propulsion Laboratory. Martin Lo, in particular, has been
instrumental in pushing for a dynamical systems approach to spacecraft trajectory design, and our
thanks go to him for both inspiring the project, and providing us with the known techniques for
computing libration point halo orbits.
The great success of this project was due primarily to the work of our students, Molly Megraw
and Christopher Sinclair. It was a joy to stand back and watch them take o
with the project
after we provided them with the basic background material. The striking pictures of manifolds that
appear in this document were created with their software.

2 The central force problem


Newton was the rst to discover that the gravitational force due to a massive central object is
one example of a force vector eld whose magnitude is inversely proportional to the square of the
distance to the object, and directed towards it: F (X ) = ;kX=jX j3, where X = (x y z ), and the
object is located at the origin. In such a case, F has a potential function V : R3 ! R given by
V (X ) = ;k=jX j, so that F = ;rV .
Exercise 2.1. Verify this.
The equations of motion for a particle of unit mass under the inuence of F are given by Newton's
second law, which says that F = ma. With m = 1 and a = X (t), we have
X (t) = ;kX=jX j3:

2
This is a 6 dimensional system a solution is uniquely determined by six initial conditions { 3 for
position and 3 for velocity.
Exercise 2.2. Write down the equations of motion as a system of 6 rst-order equations.
The motion is planar
It shouldn't be too surprising that the motion of an object in a central force eld stays in a plane,
that spanned by the object's initial position and velocity vectors.
Exercise 2.3. Show this. To do so, compute
d _
dt (X (t)  X (t))
for a solution X (t). Recall that the normal product rule holds for the derivative of cross-products.
Now think geometrically! The vector mX  X_ for a mass m is the angular momentum vector.
Explain the phrase \angular momentum is conserved" for this motion.

Circular solutions
It can be shown that the path of the object is not only planar but in fact traces out a conic section
(see, for example, 5, Chapter 1, Section 1.5]). The proof is not dicult, but involves changing to
polar coordinates along with a little trickery. For our purposes, we will just verify that a circular
trajectory is indeed one possible solution.
Exercise 2.4. Verify that (a cos !t a sin !t 0) is a solution. How does the angular velocity ! depend
on a and k?
An important aside: The discovery of conics as solutions involves changing to polar coordinates
(r ) in the plane of motion, and then transforming the equations of motion and eliminating time
as the dependent variable. The resulting equation can be solved explicitly for r in terms of . Thus
it is possible to deduce from initial conditions precisely which conic section path an object under
the inuence of a central force will follow. It is not in general possible, however, to nd an explicit
parameterization, using elementary functions, of the path with time as the dependent variable. To
do so involves inverting a transcendental function. We have to resort to numerical approximations
to nd this parameterization. Clearly, the family of circular orbits is an exception.

3 The two-body problem


The two-body problem is to describe the motion of two bodies, of mass m1 and m2 , under the
inuence of the gravitational eld induced by each. Let X1 be the vector position of m1 and X2 the
position of m2 . Newton's law of gravitation says that the force on m1 due to m2 is
F12 = Gmjr1j3m2 r

3
where r = X2 ; X1 is the vector from m1 to m2 , and G is the \universal gravitational constant" (see
Appendix A for its value). The force on m2 due to m1 is F21 = ;F12 . Newton's second law allows
us to write the equations of motion for the two objects. After dividing out the common masses from
both sides of the equations, we get
X1 = Gm
jrj3 r 
2

X2 = ; Gm
jrj3 r :
1

The following series of exercises illustrate how the two-body problem may be reduced to the
central force problem.
Exercise 3.1. Write down the equations of motion for the relative position vector r and show that
the problem of two bodies reduces to a central force problem. This means that the motion of one
body relative to another traces out a conic.
Exercise 3.2. Assume the Earth's orbit around the sun is circular, and use what you've learned so
far to determine the angular rotation rate and period of the Earth's orbit. How does this compare
with the \experimental" value? (Assume the mass of the Earth is negligible. The radius of Earth's
orbit is 1 A.U. (astronomical unit). See Appendix A).
Exercise 3.3. Towards further analysis, compute the location r0 of the center of mass of m1 and
m2 , in terms of X1 and X2 . The center of mass is the location on the line connecting the masses
at which point the moments m1 jr0 ; X1 j and m2 jr0 ; X2 j are equal. (Answer: r0 = X1 m 1 +X2 m2
m1 +m2 )
Then compute the equations of motion for r0 and show that the center of mass always moves along
the linear path r0 (t) = at + b, for arbitrary constants a and b. Thus we can assume that the center
of mass remains stationary at the origin.
Exercise 3.4. Write X1 and X2 in terms of r, assuming r0 = 0.

4 The circular restricted three-body problem


The restricted three-body problem (RTBP) introduces a third body whose motion is a
ected by
but does not aect the original two bodies, called the primaries. The primaries move along a
conic section, as described above, and the potential well (which evolves with time) generated by
the primaries controls the motion of the third body. In the circular restricted three-body problem
(CRTBP) we assume a circular orbit for the primaries. This problem, especially restricted to motion
of the third body in the plane of motion of the primaries, has been well studied 17].

The equations of motion


Let the center of mass of the primaries constitute the origin of our (x y z ) coordinate system, and
assume that the primaries orbit in the (x y)-plane. Let X (t) denote the position of the third body
with mass m and let X1 (t) and X2 (t) denote the positions of the primaries.

4
Exercise 4.1. For the moment, assume the positions X1 and X2 are xed, and use the fact that
the potential (and force) due to two masses is just the sum of the individual potentials (and forces)
to write down the two-body potential (and force) in terms of X1 and X2 .
For the CRTBP we assume the primaries move in circles (of radius a for m1 and b for m2 , say)
about the origin, with common angular velocity !.
Exercise 4.2. Use this to write down the time-dependent equations of motion for X . Assume at
t = 0 that X1 is on the positive x-axis, and X2 on the negative x-axis.
Dimensionless coordinates
Through a sequence of coordinate transformations, we can reduce the number of free parameters in
the equations of motion to one. The nal coordinates are called dimensionless coordinates.
We can write the equations of motion in terms of the potential as
X = ;rX V
where  m m

V (X t) = ;G jX ; X j + jX ; X j 
1 2
1 2
and
X1 (t) = (a cos !t a sin !t 0)
X2 (t) = (;b cos !t ;b sin !t 0):
Normalize length. First, we normalize so that the distance l = a + b separating the primaries
is one. Thus we want to substitute  = X=l. This means  = X=l , so that the new equations of
motion become
 = ; 1l rX V (l ):
But now let ~( t) = l12 V (l t). Then
r ~ = l12 rX V (l ) dX
d
1
= l rX V (l )
which allows us the compact notation  = ;r~.
Normalize time. Next we normalize the time so that the angular velocity of the primaries is
one, by substituting t~ = !t. We have
d2  = 1 d2 
dt~ !2 dt2  
~
= ; 12 r~ t! !
= ;r

5
where  !
( t~) = !12 ~( t~=!) = ; !G2l3 m1 m2
j ; l X1 (t)j j ; 1l X2 (t~)j :
1 ~ +
As discovered in Exercise 2.4, the angular velocity ! satises !2 = GM
l3 , where M = m1 + m2 . Thus
 !
( t~) = ; M1 m1 m2
j ; l X1 (t)j j ; 1l X2 (t~)j :
1 ~ +

Normalize mass. Finally we normalize so that the total mass of the primaries is one. Substi-
tuting  = m1 =M and  = m2 =M = 1 ; , we get
0

 !
( t~) = ; 
~ + 1 1;  ~ :
j ; l X1 (t)j j ; l X2 (t)j
1

And since none of the dependent variables depends on the masses, the equations of motion remain
 = ;r.
Furthermore, we have the relations bl = mM1 and al = mM2 .
Exercise 4.3. Show this.
Thus
1 X (t~) = ;(1 ; ) cos t~ (1 ; ) sin t~ 0
l 1
1 X (t~) = ;; cos t~ ; sin t~ 0 :
l 2

Set X = (x y z ) =  and t~ = t, and dene


1 = jX ; X1 j
2 = jX ; X2 j 
where we redene X1 and X2 to be the normalized primary vectors (1 ; )(cos t~ sin t~ 0), and
;(cos t~ sin t~ 0), respectively. Then
  1 ; 
(x y z t) = ; + 1 2
and the nal equations of motion are
x(t) = ;x =  x ; (1 ;3) cos t + (1 ; ) x + 3cos t
1 2

y(t) = ;y =  y ; (1 ;  ) sin t + (1 ;  ) y +  sin t


1
3 2
3
z
z(t) = ;z =  3 + (1 ; ) 3 z
1 2
The only free parameter is , which is usually taken to be less than or equal to 1/2 to represent the
smaller of the two primary masses.

6
Rotating coordinates
A disadvantage of the above representation is that the independent variable t appears explicitly in
the potential function . We will now introduce another change of coordinates which takes out the
rotation of the primaries, and removes time explicitly from the di
erential equations.
Let R denote the matrix of rotation (clockwise) by angle about the z -axis:
0 cos sin 0
1
R = @ ; sin cos 0 A
0 0 1
Since at time t the primaries lie on a line at angle t measured counterclockwise from the x-axis, we
want to make the change of coordinates W = Rt X .
Exercise 4.4. Show that under this change of coordinates, the equations of motion become
W ; 2K W_ + K 2 W = ;rW U (W )
where 0 0 1 0
1 0 ;1 0 0
1
K = @ ;1 0 0 A  K 2 = @ 0 ;1 0 A
0 0 0 0 0 0
and
 
U (W ) = ; jR W ; X (t)j + jR W1 ;;X (t)j
 t  1
;

1
t
; 
;
2
= ; jW ; (1 ; )e j + jW + e j 
1 1

and where e1 = (1 0 0). U is the transformed potential.


Re-introduce the variable X = (x y z ) for W , and dene


$(X ) = ; 21 K 2 X X ; U (X ):
Then the equations of motion become
X ; 2K X_ = rX $
or

x ; 2y_ = $x (1)
y + 2x_ = $y (2)
z = $z : (3)

7
5 Libration points
The CRTBP, when expressed by (1)-(3) in rotating coordinates, has ve equilibrium points, called
Lagrangian, or libration points.
Exercise 5.1. Find them! Show that equilibrium solutions occur when $x = $y = $z = 0. Argue
that all solutions lie in the x y-plane, and show that two of the solutions (called L4 and L5 ) form
equilateral triangles with the primaries. The other three libration points (L1 {L3 ) lie on the x-axis,
bracketing the masses. Show that the x-coordinates of these points can be found by solving quintic
polynomials.

6 The Jacobian integral


An integral of a second order system is a non-constant function G(x1  : : :  xn  x_ 1  : : :  x_ n ) that is
constant on solutions to the system. Thus, every solution lies on some level set of G. This is
expressed mathematically by the equation
d
dt G(x1 (t) : : :  xn (t) x_ 1 (t) : : :  x_ n (t)) = 0
for any solution (x1 (t) : : :  xn (t)).
Exercise 6.1. Verify that
J  2$(x y z ) ; (x_ 2 + y_ 2 + z_ 2 ) (4)
is an integral for the CRTBP. It is called the Jacobian integral, and the value it takes on a solution
is called the Jacobian constant.
The Jacobian integral allows one in principle to reduce the equations of motion in the CRTBP
from a 6th order system to a 5th order system. Another immediate application of the Jacobian
integral is the following. A given set of initial conditions denes a particular value for the Jacobian
constant, say C . The simple observation that the square of the velocity x_ 2 + y_ 2 + z_ 2 must be positive
implies with (4) that the range of motion in x y z -space for that particular solution must lie on one
side of the surface dened by $(x y z ) = C=2. This surface is called the surface of zero velocity for
the Jacobian constant C . (See 17, Chapter 4, Section 10.2] for properties of $, and a discussion of
curves and surfaces of zero velocity.)
Note that despite the name, if a particle is on a surface of zero velocity, it does not necessarily
have zero velocity. The surface represents a barrier through which solutions for a particular value
of the Jacobian constant cannot pass. Solutions with di
erent Jacobian constants will have di
erent
surfaces of zero velocity.
Exercise 6.2. Use Maple or Mathematica to graph some surfaces of zero velocity for various Jaco-
bian constants, and identify the regions of possible motion.

8
7 Symmetries
Symmetries in the equations of motion allow us to nd new solutions when some solutions are given.
We present two symmetries here. The rst symmetry is a reection across the (x y)-plane and the
second is a reection across the (x z )-plane with a time-reversal. Assume that (x(t) y(t) z (t)) is a
solution to the CRTBP, then the following are also solutions:
(x(t) y(t) ;z (t))
(x(;t) ;y(;t) z (;t)) :
There is also a symmetry involving parameters in the equation. This is given by:
(x(t) y(t) z (t) ) 7! (;x(t) ;y(t) z (t) 1 ; ):
This means that in fact we only need to study the equations in the parameter range 0 <  < 1=2.
We will take advantage of the symmetry properties of the CRTBP equations in the later analysis.
Exercise 7.1. Verify these three symmetries.

8 The Lagrangian formulation


We digress slightly to present two reformulations of the equations of motion which will be of use to
us later. The Lagrangian formulation will be used to develop Richardson's analytic approximation
to the ow in the neighborhood of a libration point. The Hamiltonian formulation will be of use in
our numerical scheme to nd actual halo orbits.
The motion of a point mass in a Newtonian potential system can be formulated in terms of a
Lagrangian function. The Lagrangian function is dened by
L(x x_ t) = 12 x_ 2 ; U (x t) 
where x_ 2 =2 is the kinetic energy and U (x t) is the potential energy. The equations of motion are
given by the Euler-Lagrange equations:
d @L @L
dt @ x_ ; @x = 0 :
Exercise 8.1. Verify that the Lagrangian equations of motion correspond to Newton's equations
of motion x = ;rU .
The beauty of this formulation arises from the fact that it is independent of the coordinates.
The Lagrangian can be written as the di
erence of the kinetic and potential energies in generalized
coordinates and the equations of motion may be easily derived using the Euler-Lagrange equations.
This fact follows from Hamilton's principle and formulating variational equations for the action of
the system. See any standard mechanics book for a detailed discussion (e.g., 1, 4, 7]).

9
The CRTBP in dimensionless non-rotating coordinates has a Lagrangian which is given by
_ t) = 1 X_ 2 ; U (X t)
L(X X 2
where
U (X t) = ; jX ; X (t)j ; jX 1;;X(t)j :
1 2
We can write the Lagrangian in rotating coordinates W = Rt X . First calculate X = R t W and
;

X_ = R_ t W + R t W_ and substitute into the equation for L to get


; ;

L = 21 (W_ ; KW )2 ; U (W ) 
where
U (W ) = ; jW ; T X (t)j ; jW ;1 R; X 
t 1 t 2 (t)j
= ; jW ; (1; )e j ; jW1 +; e  :
j
1 1

Exercise 8.2. Verify that the equations of motion in rotating coordinates derived from using this
Lagrangian and the Euler-Lagrange equations are the same as the ones we derived earlier.

9 The Hamiltonian formulation


The CRTBP is a Hamiltonian dynamical system. This means that there exist coordinates p q and
a function H (p q t) such that the equations of motion are given by
q_ = @H=@p (5)
p_ = ;@H=@q :
This is often written in the compact form
z_ = J rz H (z ) 
where z = (q p) and J is the matrix of four square blocks given by
 0 1
J = ;1 0 :
For a discussion of this extensive theory, see, for example, 1, 4, 7, 10].
We will write down the Hamiltonian and corresponding canonical coordinates in rotating variables
since this will be most useful for us. Typically this is done by starting from the Lagrangian written
using the traditional coordinates q q_. Thus, we start with
L(q q_ t) = 12 (q_ ; Kq)2 ; U (q) :

10
The generalized momenta p are given by p = @L=@ q_, so
p = q_ ; Kq :
The Hamiltonian is dened by H (p q t) =< p q_ > ;L, so we compute
H = 12 p2 + < p Kq > +U (q) :
The equations of motion are given by applying Equation 5 to this Hamiltonian.
Exercise 9.1. Verify that these equations of motion coincide with those from the Lagrangian for-
mulation.
Since the Hamiltonian for this problem is independent of time, it is a constant of the motion.
This can be veried by computing dH=dt = @H @H
@q q_ + @p p_ = 0.
Exercise 9.2. Verify that H = ;J =2, where J is the Jacobian integral introduced in Section 6.

10 Richardson's approximations for halo orbits


In this section we discuss a third-order analytic approximation to the equations of motion which
produces periodic solutions about the collinear libration points. Our discussion follows the work of
D. L. Richardson 14, 15, 16], who used this technique successfully in designing orbits for the ISEE3
mission to L1 in the late 1970s.

Yet another change of coordinates


We will discuss ow in a neighborhood of L1 and L2 , recalling that ow near L3 is the same as that
near L2 but for a di
erent parameter value. In order to study the motion near a libration point it is
easiest to choose a coordinate system that is centered at the libration point. We will also normalize
distances so that the distance between the libration point and the mass M1 is one unit.
In the rotating coordinate system, let the locations of the libration points Li be given by
Li = i e1 
where the i are roots of the appropriate quintic as discussed in an earlier section. Let the vectors
from Li to the masses M1 and M2 be given by
r1 = (1 ; )e1 ; Li = (1 ;  ; i )e1 
r2 = ;e1 ; Li = ;( + i )e1 :
New rotating coordinates centered at the libration point Li and rescaled by jr1 j are given by
 = (W ; Li )=jr1 j :

11
We write the Lagrangian in these new coordinates as
L( _ ) = jr21 j (_ ; K)2 + i jr1 j <  e1 > ;U ()
2

+ i jr1 j < _ e2 > + 12 2i 


with
U () = ;j= jr1 j + ;(1 ; )=jr1 j 
 e1 j j + rr21 e1 j j j
j j

where from now onwards the top sign is taken for L1 and the bottom for L2. We want to develop
this as a power series in jj and thus want jr2 j  jr1 j. Consequently, this analysis will give results
for L1 when 0 <   1=2 and for L2 when 0 <  < 1. From the symmetry discussed earlier we can
produce orbits for L1 when 1=2 <  < 1 by applying the analysis for L1 using 1 ; . Similarly, we
can study orbits around L3 by using the symmetry applied to orbits about L2 .
Note that rescaling of the Lagrangian by a constant does not a
ect the equations of motion so
we shall divide by jr1 j2 . Also, constants and linear terms which contain only _ play no role in the
equations of motion, so the nal two terms may be discarded. We now write the Lagrangian as
L( _) = 12 (_ ; K)2 + i jrxj ; U () 
1

with
U () = j;  1 ;(1 ; ) jr2 j=jr1 j
r1 j3 j  e1 j + jr1 j2 jr2 j j + rr12 e1 j :
j j
j j

The next step is to write the potential function U using a power series of Legendre polynomials.
To introduce Legendre polynomials we recall a couple of identities. First an application of the
law of cosines gives
jrj = 1 ; 2 jj cos + jj2 1=2  ;

j ; r j jrj jrj2
where is the angle between  and r. Then we recognize that (1 ; 2zx + z 2 ) 1=2 is the generating ;

function for the Legendre polynomials. In other words,


1=2
X
1

(1 ; 2z cos + z )
2 ;
= Pn (cos )z n 
n=0
and
jrj = X P (cos ) jjn :
1

j ; rj n=0 n jr jn
The Legendre polynomials can also be developed from a recursion relation given by
P0 (cos ) = 1
P1 (cos ) = cos
 2n ; 1  n ; 1
P (cos ) =
n cos P ;
n n;P :
1
n n 2
;

12
Note that Pi is even or odd in cos exactly when i is even or odd. Also recall the formula for the
derivative of a Legendre polynomial,
dPn (C ) = X
dC (2n ; 4k ; 1)Pn 2k 1 (C ) : ; ;

0 k (n 1)=2k Z
  ; 2

Applying this to terms in our potential function we see that


1 X
1
n
j  e1 j = n=0 Pn (
x =jj)jj 
jr2 j=jr1 j X
Pn (;x =jj) jjrr1 jjn jjn :
1 n
r2 =
j + j
j r1
j
j
e1 j n=0 2

When introducing the expansions into the Lagrangian we can once again neglect the constant terms
and we will collect the linear terms. This results in a Lagrangrian of
X
L( _) = 21 (_ ; K)2 + jrj3 Pn (
x=jj)jjn
1

1 n=2

+ j(1 ; ) X P (; =jj) jr1 jn jjn


1

r1 j2 jr2 j n=2 n x jr2 jn



  1 ; 

x
+ jr j i
jr j2 ; jr j2 :
1 1 2

Now recall that


i
jrj2 ; 1jr;j2 = i
( ; (1 ; ))2 ; ( 1 ;  =0
1 2 i i + )2
since this is the equation which denes the libration point! Of course we know this quantity will be
zero simply because we have placed the origin of the coordinate system at an equilibrium point thus
guaranteeing that linear terms will not exist.
Combining the sums in the Lagrangian, we write
X
L( _) = 12 (_ ; K)2 +
1

cn Pn (x =jj)jjn 
n=2
where
cn = (
1)n jrj3 + (;1)n (1 ;jr)jnjr+11 j
n 2 ;

1 2

= (
1)n j ; (1 ; )j3 + (;1)n (1 ; ) j i j;
 (1 ; )jn 2 : ;

i i + jn+1
If you insist on nding the equations of motion in a neighborhood of L3, then in the coordinates
 = (L3 ; W )=jr2 j the above Lagrangian holds and the cn are the same as those for L2 except they

13
must be computed for parameter value 1 ; . Performing this computation, we nd that the cn for
the Lagrangian at L3 are given by
c = (;1)n 1 ;  + (;1)n jr2 j :
n 2 ;

n jr2 j3 jr1 jn+1


The equations of motion are derived from the Lagrangian using the Euler-Lagrange equations.
Consequently, the equations of motion are
X
1

 ; 2K _ + K 2  = ncn Pn (x =jj)jjn 2 


;

n=2
X
cn Pn (x =jj)jjn 2 (jje1 ; jxj ) :
1

+ 0 ;

n=2
We can now approximate the dynamical system near the libration points by using perturbation
methods. Essentially, we will expand all quantities in a power series of a small parameter (e.g., the
amplitude of the solution) and consider only a nite number of terms in the innite sum. We can
then attempt to nd explicit solutions for these dynamical systems and hope that they approximate
solutions of the full problem.

The linear approximation


First we will study a linear approximation to the equations of motion just developed. We show
that solutions to such an approximation are inadequate in the sense that they are not, in general,
periodic. We will require these solutions, however, when we move to higher-order approximations.
The linear approximation involves taking only the quadratic terms in the Lagrangian. This gives
us
L2( _) = 12 (_ ; K)2 + c2 (32x ; 2 )=2
with
c2 = j ; (1 ; )j3 + j 1 ;  :
i i j3
+
Using the Euler-Lagrange equations we write down the equations of motion to get
0 ;(2c2 + 1) 0 0 1
 ; 2K _ + @ 0 c2 ; 1 0 A  = 0 :
0 0 c2
This is a linear degree two system of ordinary di
erential equations which we can solve by spectral
methods. Note that the z -component is decoupled from the other two. We compute the characteristic
polynomial to be
(%2 + c2 )(%4 + %2 (2 ; c2 ) + 1 + c2 ; 2c22) = 0
and solving for % nd s
p
p 2
c 2;2
9c2 ; 8c2
% =
j c2  % =
2
14
where j represents the complex involute. If c2 > 1 then 9c22 ; 8c2 > (c2 ; 2)2 and there will be two
pairs of purely imaginary eigenvalues and one pair of real eigenvalues.
Exercise 10.1. Show that c2 > 1.
We can thus write the solution to the linear equation in the form:
x(t) = A1 et + A2 e t + A3 cos t + A4 sin t
;

y(t) = ;k1 A1 et + k1 A2 e t ; k2 A3 sin t + k2 A4 cos t


;

z (t) = A5 cos pc2 t + A6 sin pc2 t 


where
s p
= c2 ; 2 + 9c22 ; 8c2
2
s p
; c2 ; 2 ; 9c2 ; 8c2
2
 = 2
k1 = (2c2 + 1 ; )=2
2

k2 = (2c2 + 1 + 2 )=2 
and A1  : : :  A6 are arbitrary constants determined by the initial conditions.
Since we are interested in periodic and quasiperiodic solutions we take A1 = A2 = 0. Then the
solution to the linear problem can be written in terms of amplitude and phase as
x(t) = ;Ax cos(t + )
y(t) = kAx sin(t + )
z (t) = Az sin(pc2 t + ) 
with k = k2 . For out of plane solutions (Az 6= 0), we do not expect  and c2 to be rationally related
and thus expect quasiperiodic solutions. We must therefore include nonlinearities if we hope to nd
periodic solutions.

The method of Lindstedt-Poincare


To nd better approximations to the nonlinear problem in a neighborhood of the equilibrium point
solutions we will use the perturbation techniques of Lindstedt-Poincar&e 11].
The rst thing we do is allow for a frequency correction by setting  = !t and letting ' ' denote
0

d=d . We will then truncate the equations of motion at degree 3. Doing this we get
!2 x ; 2!y ; (1 + 2c2)x = 32 c3 (2x2 ; y2 ; z 2)
00 0

+2c4(2x2 ; 3y2 ; 3z 2)x


!2 y + 2!x + (c2 ; 1)y = ;3c3xy ; 32 c4 (4x2 ; y2 ; z 2)y
00 0

!2 z + 2 z = ;3c3xz ; 23 c4 (4x2 ; y2 ; z 2 )z + 'z 


00

15
where
' = 2 ; c2 :
We continue the perturbation analysis by assuming solutions of the form:
x( ) = x1 ( ) + 2x2 ( ) + 3 x3 ( ) + : : :
y( ) = y1 ( ) + 2 y2 ( ) + 3 y3 ( ) + : : :
z ( ) = z1 ( ) + 2 z2 ( ) + 3 z3 ( ) + : : :
and letting
! = 1 + !1 + 2 !2 + : : : :
We substitute these quantities into the equations of motion and equate components of the same
order. Here we make the assumption that ' = O(2 ) and set  = .

The rst order equations


The O() equations are the linearization of the vector eld which we solved in the previous section,
with a modied frequency for the out of plane oscillations:
x1 ; 2y1 ; (1 + 2c2 )x1 = 0
00 0

y1 + 2x1 + (c2 ; 1)y1 = 0


00 0

z1 + 2 z1 = 0 :
00

Since we are only interested in bounded solutions we write the solutions


x1 ( ) = ;Ax cos( + )
y1 ( ) = kAx sin( + )
z1 ( ) = Az sin( + ) :
Later, in order to avoid secular solutions we will need to put constraints on the constants Ax  Az   ,
but for now they are arbitrary.

The second order equations


The O(2 ) equations depend on the above solutions and are given by

x2 ; 2y2 ; (1 + 2c2 )x2 = ;2!1(x1 ; y1 ) + 32 c3 (2x21 ; y12 ; z12)


00 0 00 0

y2 + 2x2 + (c2 ; 1)y2 = ;2!1(y1 + x1 ) ; 3c3x1 y1


00 0 00 0

z2 + 2 z2 = ;2!1z1 ; 3c3x1 z1 :
00 00

16
Substituting in the solutions for x1  y1  z1 , we get
x2 ; 2y2 ; (1 + 2c2)x2 = +2!1Ax (k ; ) cos 1
00 0

+ 1 + 1 cos 21 + 2 cos 22


y2 + 2x2 + (c2 ; 1)y2 = 2!1Ax (k ; 1) sin 1
00 0

+1 sin 21


z2 +  z2 = 2!1Az 2 sin 2
2 00

+1 sin(1 + 2 ) + 1 sin(2 ; 1 ) 


where
1 =  + 
2 =  +  
and the expressions for the other coecients are given in Appendix C. These are a set of inhomo-
geneous linear di
erential equations whose solutions are summarized in Appendix B. We know the
bounded homogeneous solution which is incorporated into the solution to the rst order equations
and simply need to nd a particular solution. The secular terms (those responsible for producing un-
bounded solutions) are the sin 1 , cos 1 and sin 2 terms which are all eliminated by setting !1 = 0.
Thus we nd the solution
x2 = 20 + 21 cos 21 + 22 cos 22
y2 = 21 sin 21 + 22 sin 22
z2 = 21 sin(1 + 2 ) + 22 sin(2 ; 1 ) 
where the expressions for the coecients are given in Appendix C.
The third order equations
The O(3 ) equations are (after setting !1 = 0)
x3 ; 2y3 ; (1 + 2c2 )x3 = ;2!2(x1 ; y1 )
00 0 00 0

+3c3 (2x1 x2 ; y1 y2 ; z1z2 )


+2c4 x1 (2x21 ; 3y12 ; 3z12)
y3 + 2x3 + (c2 ; 1)y3 = ;2w2 (x1 + y1 )
00 0 0 00

;3c3 (x1 y2 + x2 y1 )
; 32 c4 y1 (4x21 ; y12 ; z12)
z3 + 2 z3 = ;2!2z1 + '2 z1
00 00

;3c3 (x2 z1 + x1 z2)


; 23 c4 z1 (4x21 ; y12 ; z12) :

17
Substituting in the solutions for x1  y1  z1  x2  y2  z2 , we get
x3 ; 2y3 ; (1 + 2c2 )x3 = 1 + 2!2Ax (k ; )] cos 1
00 0

+3 cos 31 + 4 cos(1 + 22 )


+5 cos(22 ; 1 )
y3 + 2x3 + (c2 ; 1)y3 = 2 + 2!2Ax (k ; 1)] sin 1
00 0

+3 sin 31 + 4 sin(1 + 22 )


+5 sin(22 ; 1 )
z3 +  z3 = 3 + Az (2!22 + '=2 )] sin 2
2 00

+3 sin 32 + 4 sin(21 + 2 )


+5 sin(21 ; 2 ) 
where the expressions for the coecients are given in Appendix C. There are secular terms in the
x3 ; y3 equations and in the z3 equations which can no longer be removed by simply setting a value
for the frequency correction !2 . We start by examining the secular terms in the z3 equations more
closely.
To remove the 5 sin(21 ; 2 ) term we would need 5 = 0. Since we cannot freely adjust this
parameter, we can attempt to remove the secular term by adjusting the phases of 1 and 2 so that
sin(21 ; 2 ) sin 2 . To do this we need
 =  + n 2 
where n = 0 1 2 3. Then the z3 solution will be bounded if
3 + Az (2!2 2 + '=2 ) + 5 = 0 
where  = (;1)n . This phase constraint a
ects the x3 ; y3 equations, which now each contain one
secular term. Instead of forcing the removal of these terms with two additional constraint equations,
we see from Appendix C that we can simultaneously eliminate the unbounded terms from both
particular solutions with a single constraint equation. The requirement is
(1 + 2!2Ax (k ; ) + 5 ) ; k(2 + 2!2Ax (k ; 1) + 5 ) = 0 :
We satisfy this last equation by setting

; k2 +  (5 ; k5 )


!2 = 21A
x (1 + k 2 ) ; 2k ]
= s1 Ax + s2 A2z 
2

where the expressions for s1  s2 are given in Appendix C. Substituting this into the rst constraint
we get
l1 A2x + l2 A2z + '=2 = 0 

18
with l1  l2 given in Appendix C, and thus can satisfy this constraint by letting one amplitude be
determined by the other.
Assuming these constraints the third order equations reduce to

x3 ; 2y3 ; (1 + 2c2)x3 = k6 cos 1 + (3 + 4 ) cos 31


00 0

y3 + 2x3 + (c2 ; 1)y3 = 6 sin 1 + (3 + 4 ) sin 31


 (;1)n=2( +  ) sin 3
00 0

z3 +  z3 =
2 00 3 4 1 n = 0 2
(;1)(n 1)=2 (4 ; 3 ) cos 31 n = 1 3 
;

where 6 = 2 + 2!2Ax (k ; 1) + 5 . The solution to this is given by

x3 = 31 cos 31


y3 = 31 sin 31 + 32 sin 1
 (;1)n=2 sin 3 n = 0 2
z3 = 31 1
(;1)(n 1)=232 cos 31 n = 1 3 
;

where the expressions for the coecients are given in Appendix C. Richardson's solution 14, 15, 16]
neglected the 6 terms and thus did not include the 32 correction { a cubic correction to the
amplitude of the primary frequency.

The nal approximation

Applying the mapping Ax 7! Ax = and Az 7! Az = will remove  from all the equations. Then we
can use Ax or Az as a small parameter. Combining the solutions we have computed to third order,
we get

x( ) = 20 ; Ax cos 1 + (21 + 22 ) cos 21 + 31 cos 31
y( ) = (kAx + 32 ) sin 1 + (21 + 22 ) sin 21 + 31 sin 31
 (;1)n=2A sin  n = 0 2
z ( ) = z 1
(;1)(n 1)=2 A cos  n = 1 3
 (;1)n=2( z sin 21 +  sin 3 )
;

+ 21 1 31 1 n = 0 2
(;1)(n 1)=2 (21 cos 21 + 22 + 32 cos 31 ) n = 1 3 :
;

19
These equations may be rewritten to more clearly see the dependence on the small parameters Ax
and Az . This is
x( ) = ;Ax cos 1 + a21 A2x + a22 A2z + (a23 A2x + a24 A2z ) cos 21 (6)
+(a31 A3x + a32 Ax A2z ) cos 31
y( ) = kAx sin 1 + (b21 A2x + b22 A2z ) sin 21
+(b31 A3x + b32 Ax A2z ) sin 31 (7)
+(b33 A3x + b34 Ax A2z + b35 Ax A2z ) sin 1
 (;1)n=2A sin  n = 0 2
z ( ) = z 1
(;1)(n 1)=2 Az cos 1 n = 1 3
;

 (;1)n=2d A A sin 2 n = 0 2
+ (;1)(n 1)21=2 dx Az A (cos 1
2  ; 3) n = 1 3 (8)
 (;1)n=2(d A213 +xd z A2 A )1sin 3
;

+ (;1)(n 1)=312 (dz A A322 ;x dz A3 ) cos1 3 nn = 0 2


= 1 3 :
32 z x 31 z
;
1

The expressions for the coecients are given in Appendix C.

11 Numerical computation of halo orbits


In this section we discuss some strategies for the numerical computation of periodic orbits. The
method we will discuss is basically a Newton's method scheme, with the idea that Richardson's
analytic approximation will provide convergent initial seeds for the method. We will rst examine
the problem of computing periodic orbits in the 2-degrees of freedom CRTBP { that is, orbits which
lie in the (x y)-coordinate plane. Then we will look at out-of-plane orbits. In both cases we will
restrict the discussion to nding orbits which have a symmetry across the x z -plane.

Symmetric periodic orbits for the 2-degrees of freedom CRTBP


We would like to nd a periodic orbit which lies in the x y-coordinate plane and is symmetric about
the x-axis. There are families of such orbits surrounding the collinear libration points. We shall
dene one of these orbits by the location of its intersection with the x-axis and its velocity. Thus
if a point in phase space is given by X = (x y vx  vy ), then a periodic orbit can be specied by
(x 0 0 vy ). We have set y = 0 to specify the intersection of the orbit with the x-axis, and we set
vx = 0 in order to satisfy the symmetry condition.
Suppose that the orbit only crosses the x-axis in two points. Then the second point will be
halfway around. We can then use the symmetry of the problem to generate the entire orbit from
the half of the orbit lying to one side of the x-axis.
Let X_ = f (X ) be the equations of motion. We are of course here only interested in the CRTBP,
but these techniques will work in a great variety of problems. Let ((X t) be the solution to the
di
erential equation, which means that ((X 0) = X and @ ((X t)=@t = f (((X t)).

20
Let X = (x 0 0 vy ) be an initial guess for a point on a symmetric planar periodic orbit of the
CRTBP. Flow the point under the vector eld until y = 0 again and let T1=2 be the time of the ow.
Thus ((X T1=2 ) = (~x 0 v~x  v~y ). If v~x = 0 then the initial guess is part of the periodic orbit and we
are done if not then we need to improve the initial guess in order to attempt to drive v~x to zero.
Thus we compute
@ ((X T1=2)
((X + 'X T1=2 + 't) = ((X T1=2 ) + @X 'X
@ ((X T1=2)
+ @t 't + h.o.t. :
Since we wish to restrict our choice of initial conditions, we let 'X = ('x 0 0 'vy ) and solve the
following equations which come from the approximation found by dropping the higher order terms
in the above equation:
0 1 0 1
@ ( B '0x C
B@ CA + f (((X T1=2)) 't = BB@ 0 CC :
A
@X 0 ;v~x
'vy
We use ' ' in the above equation to indicate quantities whose values do not concern us. So we end
up with exactly two equations and three unknowns. This is expected since we will nd a family
of periodic solutions which can be parameterized by, for example, the intersection with the x-axis.
Fixing this quantity, we must take 'x = 0 and then can solve for 'vy and 't. This will give us a
new guess for the initial velocity vy + 'vy and an estimate of T1=2 + 't for the return time. The
whole process may be repeated.  @ . Since we do not have an explicit formula for
We have not yet discussed the quantity @X
((X t), we cannot compute this matrix of functions explicitly. In the next section we discuss this
quantity, which is known as the fundamental solution matrix.

The fundamental solution matrix


Let ((X t) solve the n-dimensional di
erential equation X_ = f (X ), so ((X 0) = X and @ ((X t)=@t =
f (((X t)). If we wish to see how a small change in initial condition
 X will a
ect the solution after

@
time t, then to approximate this to rst order we must compute @X . This matrix of functions is a

particular fundamental solution matrix. The role of this matrix should be clear from the following
calculation:

((X + 'X t + 't) = ((X t) + @ ((X t) 'X + @ ((X t) 't
@X @t
+h.o.t.
Since ((X 0) = X , we compute @ ((X 0)
@X =1:

21
Thus for t = 0 the matrix is the identity matrix. We can then see how the matrix evolves in time
by computing

@ @ ( = @ @ ( = @ f (() = @f j
@(
@t @X @X @t @X :
@X X = @X
Thus we have a linear, nonautonomous di
erential equation in n2 dimensions which describes the evo-
lution of a fundamental solution matrix. These are sometimes called the rst variational equations.
We canset up a system to simultaneously compute the solution ( and the fundamental solution
matrix @X @ , by augmenting the initial n dimensional vector eld by n2 dimensions. Letting M
denote the fundamental solution matrix, the resulting n2 + n dimensional initial value problem that
we solve is:
X_ = f (X )
M_ = @X @f M 

with initial conditions


X (0) = X0
M (0) = 1 :
When the solution is a periodic orbit and this fundamental solution matrix is computed at a time
equal to one period, then the resulting matrix is called the monodromy matrix. The eigenvalues of
this matrix determine the stability of the periodic orbit.

Halo orbits - symmetric periodic orbits for the 3-degrees of freedom CRTBP
In this section we extend the previous discussion regarding the computation of periodic orbits to
the full 3 degrees of freedom problem. This is nearly identical to the previous discussion, except
that the dimensions are increased. These orbits surrounding the libration points L1 , L2 , and L3 are
called halo orbits.
Let X = (x y z vx  vy  vz ). An initial guess is now located on the (x z )-plane with a component
of velocity only in the y direction. Let X = (x 0 z 0 vy  0) be an initial guess. If T1=2 is the time
of rst return to the (x z )-plane, then we compute ((X T1=2 ) = (~x 0 z~ v~x  v~y  v~z ): We are looking
for initial conditions that give us v~x = v~z = 0 in order to give us the rst half of a periodic orbit.
Adjusting the initial condition and ow time we get:
@ ((X T1=2)
((X + 'X T1=2 + 't) = ((X T1=2 ) + @X 'X
@ ((X T1=2)
+ @T 't + h.o.t. :

22
Restricting 'X to 'X = ('x 0 'z 0 'vy  0), we nd a new guess by solving:
0 'x 1 0 1
B CC BB 0 CC
@ ( BBB '0z CC B CC
@X B BB 0 CC + f (((X T1=2) 't = BBB ; v~x CC :
@ 'vy CA B@ CA
0 ;v~z
The ' ' represent quantities we do not care about, so we end up with three equations and four
unknowns. Thus we can parameterize the family of solutions by one quantity, say the out of plane
amplitude z , in which case we require that 'z = 0. This allows us to solve for a new initial x-position
x + 'x and new initial y-velocity vy + 'vy . This process is repeated in the expectation of having
found a better approximation to a periodic orbit. In Section 14 we present the results of applying
this algorithm.
Another type of symmetric periodic orbit is suggested by cases n = 0 2 in the Richardson
approximation (6) { (8). These orbits cross the x-axis perpendicularly at 1 = 0 . Any solution to
the CRTBP with this characteristic would necessarily be symmetric across the x-axis. Nevertheless,
we have been unable to apply techniques similar to those above to locate solutions with this symmetry
close to the n = 0 2 Richardson approximation.

12 Stability of the periodic orbits


In order to compute the stability of a periodic orbit, one must compute the eigenvalues of the
monodromy matrix. Recall that the monodromy matrix is a particular fundamental solution matrix
for the variational equations along a periodic orbit evaluated after one period of the orbit. Although
the monodromy matrix is dened in terms of a specic point on the periodic orbit, all the monodromy
matrices (for di
erent points on the orbit) are related by similarity transformations. Thus the
eigenvalues of the monodromy matrix are an invariant quantity. The monodromy matrix M from
a volume preserving Hamiltonian system is a symplectic mapping | that is, it satises M 1 = ;

;JM J , where denotes matrix transpose and J is the matrix introduced in Section 9 (see 10],
 

for example). Symplectic matrices have the following properties:


1. det(M ) = 1 .
2. If  is an eigenvalue of M , then so is 1=.
In practice the monodromy matrix is computed by numerical integration. When the periodic orbit
is symmetric, as are those we are interested in computing, then one can take advantage of the
symmetry to more eciently compute the matrix.
Let S be the matrix that represents the symmetry of the periodic orbit we are considering. If
X = (x y z vx  vy  vz ), then SX = (x ;y z ;vx vy  ;vz ) denes the time-reversal symmetry for
the halo orbits we are interested in. It is a simple check to prove

23
Lemma 1. If X (t) is a solution to X_ = f (X ), and ;Sf = f  S , then SX (;t) is also a solution.
We veried earlier that the CRTBP in rotating coordinates has this symmetry. We now show
that the variational equations also have this symmetry.
Lemma 2. Assume that ;Sf = f  S . Let F (X ) = @f=@X and let M_ = F (X )M , then ;SF (X ) =
F (SX )S .
Thus, we have
Lemma 3. If (X (t) M (t)) is a solution to
X_ = f (X )
M_ = F (X )M 
then so is (SX (;t) SM (;t)).
Now suppose that X (0) is part of a periodic orbit with period 2T that is symmetric, i.e., X (t)
and SX (;t) are the same orbits. Further, assume that X (0) = SX (0), a point of symmetry. Now
let M (t X (t0 ) M (t0 ) t0 ) denote the solution to the variational equations for the periodic orbit X (t).
The monodromy matrix is given by M (2T  X (0) 1 0). We wish to use the symmetry of the periodic
orbit in order to write the monodromy matrix in terms of the solution for only halfway around the
orbit, M (T  X (0) 1 0). We will prove the following
Theorem 4. 6]
 1 
0
 2K ;1 
M (2T  X (0) 1 0) = S ;1 2K M (T  X (0) 1 0) 1 0 SM (T  X (0) 1 0) 


where the represents matrix transpose.




Proof. We start by noting that


M (2T  X (0) 1 0) = M (2T  X (t) 1 T )M (T  X (0)1 0) :
Next we use the symmetry to write
M (2T  X (T ) 1 T ) = M (0 X (;T ) 1 ;T ) :
Since (X (t X (T ) T ) M (t X (T ) 1 T ) is a solution, then we can apply our symmetry to get a
solution (SX (;t X (;T ) ;T ) SM (;t X (;T ) 1 ;T )). At time T , this new solution takes on the
values (SX (;T ) S ). Thus by uniqueness of solutions we have the equalities:
SX (;t X (;T ) ;T ) = X (t SX (;T ) T )
SM (;t X (;T ) 1 ;T ) = M (t SX (;T ) S T ) :
Evaluating at t = 0 and using the periodicity and symmetry of X (t) we get
SM (0 X (;T ) 1 ;T ) = M (0 X (T ) 1 T )S :

24
So now we have
M (2T  X (0) 1 0) = SM (0 X (T ) 1 T )SM (T  X (0)1 0) :
We are in principle done with expressing the monodromy matrix in terms of the fundamental solution
matrix evaluated at time T since M (0 X (T ) 1 T ) = M (T  X (0)1 0) 1. But we can do a little
;

better than this, since M (T  X (0)1 0) 1 may be written in terms of M (T  X (0)1 0) . This is done
; 

by recalling that the fundamental solution matrix for a Hamiltonian system is symplectic 10]. Thus
we can perform a change of coordinates and use the symplectic property to write the inverse in terms
of a transpose.
Writing M in symplectic coordinates we get
 1 0
  1 0

M~ = ;K 1 M K 1 :
 
0 1 . Thus we compute that
For symplectic M~ , we have M~ 1
;
= ;J M~ J , where J =


;1 0
 1  M  2K ;1  :
0
M 1=;

;1 2K 1 0


Putting this all together we get the conclusion of the theorem.

13 Stable and unstable manifolds of the halo orbits


In this section we give a rough method to approximate the stable and unstable manifolds of the
halo orbits. Very briey, the stable manifold of a periodic orbit is the set of points which converge
to the periodic orbit in forward time. Conversely, the unstable manifold consists of those points
which converge to the orbit in backwards time. In general, these are manifolds and their smoothness
depends on the smoothness of the dynamical system. The idea of our algorithm is to nd a linear
approximation in the neighborhood of the periodic orbit and then to globalize this approximation
by owing points on it under the vector eld.
The stable manifold of a periodic orbit is tangent to the space of eigenvectors of the mon-
odromy matrix whose eigenvalues have magnitude less than one. It is also tangent to the periodic
orbit. For the nearly planar halo orbits in the Sun-Earth system, there is only one eigenvalue with
magnitude less than one so the stable manifold is two-dimensional. The unstable manifold is also
two-dimensional. Eigenvectors of the monodromy matrix are used to construct the local linear ap-
proximations. At each point along the trajectory, the eigenvectors of the monodromy matrix may be
calculated. A small step in the direction of the appropriate eigenvector will provide an approximation
of a point on the stable or unstable manifold.
We note that this method cannot in general be expected to work well. In particular, computa-
tional diculties in computing a good approximation of the stable manifold include
 disparate growth scales,
25
 poor initial guesses from the linear approximation, and
 an exponential growth of errors.
In general there are diculties with trying to evolve even a smooth curve in a smooth vector eld in
that the resulting surface need not be smooth. Some of these issues are discussed in Guckenheimer
and Worfolk 8]. For the Sun-Earth system, the stretching away from the periodic orbit is strong
enough that the manifold appears to grow nicely. There are complications in the computation of
the manifold when it comes close to Earth due to the singularity in the vector eld. In the following
section we will present some manifolds that were computed using this algorithm.

14 Results
In this section we summarize some of the discoveries our students made in the implementation of the
schemes discussed in the earlier sections. For more details on these results, we urge you to examine
the online documents of our students 9]. All these calculations were done for the Sun-Earth system
which we take to be given by  = 3:03591  10 6.
;

The analytic approximation for halo orbits should work well in a neighborhood of the libration
points. Since the (out of plane) halo orbits have a minimum amplitude, it is not clear whether the
approximation is valid. However for the Sun-Earth system and for halo orbits where n = 1 3, we nd
that the approximation works very well and provides a convergent seed for our Newton's method for
orbits with an out of plane amplitude of up to approximately 0:008 AU  1:2 million kilometers. In
Figure 1 we give one family of halo orbits. The analytic orbits and the numerically converged ones
are, on the scale of this gure, indistinguishable. The family of orbits can be numerically continued
beyond the size where good analytic approximations are provided. Doing this we nd that the orbits
become stable, but we have not yet explored further details of the extension of this family.
In the analytic approximation to the CRTBP, there are also halo orbits with a second symmetry
given by taking n = 0 2 in Equations 6. Recall that n=2 is the phase di
erence between the
oscillation in the x direction and the z direction. For the Sun-Earth system, these orbits are much
larger than the others and we have not been able to numerically nd any periodic orbits which
are similar to these. Thus, they are either an artifact of the truncated analytic model or not well
approximated by the analytic model. Further exploration may be able to resolve this question.
Stable and unstable manifolds for class n = 1 3 halo orbits were successfully computed by our
students using the method described in Section 13. Visualization of the projection of the manifolds
into 3-dimensional position space was provided using Geomview 12]. A few coloring schemes were
implemented to indicate stability of the manifolds, and to provide velocity information at each point
on the manifold. See the online HTML document 9] for details and pictures. Some examples are
given in Figures 2, 3 and 4.
In Figure 2 the halo orbit bounds the open end of the half of the manifold trailing away from
the Earth, which is located at the origin of the coordinate axes. L1 appears as the small sphere
at the center of the orbit. A coloring scheme has been chosen so that each dark band on the
manifold represents a trajectory asymptotically approaching the halo orbit. Figures 3 and 4 depict

26
Figure 1: The n = 1 family of halo orbits around L1 in the Sun-Earth system. L1, the Earth, and
L2 are represented by small spheres.

the Earth-side branch of the stable manifold of the same halo orbit. The manifold passes close to
the Earth before doubling back and intersecting itself in physical space (but not phase space!). A
single trajectory on the manifold has been superimposed in Figure 4.
The potential for further investigations is great. Future projects of interest include
 a study of the geometry of the surfaces of halo orbits
 a study of how the stable and unstable manifolds change along the family of halo orbits
 an exploration of the structure of the manifolds in the proximity of the Earth in a search for
low energy transfers from Earth parking orbits to the stable manifold of a halo orbit
 similar explorations for transfers between halo orbits about L1 and L2 
 an exploration of the geometry of the phase space for other mass ratio parameters, say for the
Sun-Jupiter system and
 the computation and visualization of invariant tori in the vicinity of the libration points which
give rise to \Lissajous" orbits.

27
Figure 2: One branch of an L1 halo orbit's stable manifold, extending away from Earth.

28
Figure 3: The other branch of the halo orbit's stable manifold, extending towards Earth.

Figure 4: A di
erent view of the same stable manifold, with a trajectory emphasized.

29
References
1] V. I. Arnold. Mathematical Methods of Classical Mechanics. Springer-Verlag, New York, 1978.
2] A. Back, J. Guckenheimer, M. Myers, F. Wicklin, and P. Worfolk. dstool: Computer assisted
exploration of dynamical systems. Notices of the Amer. Math. Soc., 39(4):303{309, 1992. Soft-
ware available via anonymous ftp at cam.cornell.edu.
3] B. T. Barden, K. C. Howell, and M. W. Lo. Application of dynamical systems theory to
trajectory design for a libration point mission. preprint, 1996.
4] V. D. Barger and M. G. Olsson. Classical Mechanics: A Modern Perspective. McGraw-Hill,
New York, second edition, 1995.
5] R. R. Bate, D. D. Mueller, and J. E. White. Fundamentals of astrodynamics. Dover, New York,
1971.
6] J. V. Breakwell and J. V. Brown. The `halo' family of 3-dimensional periodic orbits in the
Earth-Moon restricted 3-body problem. Celestial Mechanics, 20:389{404, 1979.
7] H. Goldstein. Classical Mechanics. Addison-Wesley, Reading, 2nd edition, 1980.
8] J. Guckenheimer and P. Worfolk. Dynamical systems: Some computational problems. In
D. Schlomiuk, editor, Bifurcations and Periodic Orbits of Vector Fields, pages 241{277. NATO
ASI Series. Series C: Mathematical and Physical Sciences, Kluwer, 1993.
9] M. Megraw and C. Sinclair. Orbit visualization project home page. Online HTML document,
http://www.geom.umn.edu/ megraw/home.html, 1996.
10] K. R. Meyer and G. R. Hall. Introduction to Hamiltonian Dynamical Systems and the N-body
Problem. Springer-Verlag, New York, 1992.
11] A. H. Nayfeh. Perturbation Methods. Pure and applied math. John Wiley & Sons, 1973.
12] M. Phillips. Geomview manual. Technical Report GCG62, The Geometry Center, University
of Minnesota, 1993.
13] The Royal Astronomical Society of Canada R. L. Bishop, editor. Observer's Handbook 1996.
University of Toronto Press, 1995.
14] D. L. Richardson. Periodic orbits about the L1 and L2 collinear points in the circular-restricted
problem. Technical Report CSC/TR-78/6002, Computer Sciences Corporation, 1978.
15] D. L. Richardson. Analytic construction of periodic orbits about the collinear points. Celestial
Mechanics, 22:241{253, 1980.
16] D. L. Richardson. Halo orbit formulation for the ISEE-3 mission. J. Guidance and Control,
3(6):543{548, 1980.
17] V. Szebehely. Theory of Orbits. Academic Press, New York, 1967.

30
A Astronomical constants
Taken from 13], except where noted.
G = 6:672  10 11 m3 kg 1s 2 ; ; ;

MS (mass of the Sun) = 1:9891  1030kg


ME (mass of the Earth) = 5:974  1024 kg
 (Earth/Sun) = 3:0035  10 6 ;

 (Earth-Moon/Sun) = 3:0404  10 6 ;

 (Earth-Moon/Sun) = 3:0359  10 6 (from 17]) ;

 (Jupiter/Sun) = 3:1463  10 3 ;

1 A:U: = 1:4960  1011m

B Solving the inhomogeneous linear equation


In this section we discuss how to solve the inhomogeneous linear equations which arise in the
Poincar&e-Linstedt method for this problem. First note that the x ; y equations and the z equations
are completely decoupled, so we can consider them independently.

The x ; y equations
Consider the following linear inhomogeneous di
erential equations with c2 > 1 :
x ; 2y ; (1 + 2c2 )x = A cos(qt + p) 
00 0

y + 2x + (c2 ; 1)y = B sin(qt + p) :


00 0

The homogeneous solution is given by


xh (t) = ;C1 cos(t + ) + C2 et + C3 e t  ;

yh (t) = k2 C1 sin(t + ) ; k1 C2 et + k1 C3 e t  ;

where
s p
= c2 ; 2 + 9c22 ; 8c2 
2
s p
; c2 ; 2 ; 2 9c2 ; 8c2 
2
 =
k1 = (2c2 + 1 ; 2 )=2 
k2 = (2c2 + 1 + 2 )=2 :

31
There are two cases for particular solutions, depending on whether on the forcing is at the natural
frequency  or not. Thus, we have

xp (t) = q4 +2qqB (q +1 c2 )A
2
2 (c2 2) 2c2 +c2 +1 cos(qt + p)
; ;
;
2 ;
, when q 6=  
yp (t) = q4 +2qA ;(q 2 +1+2c2 )B
q2 (c2 2) 2c22 +c2 +1 sin(qt + p)
; ;

and

xp (t) = 2(kA2 +kB


;
 2k) t sin(qt + p)
;
k(A(1 k) B ( k)) , when q =  :
yp (t) = k 2(k2 +kB
A ;
 2k) t cos(qt + p) + 2(k2 + 2k) sin(qt + p)
;
; ;
;
;

The z equation
For the z equation, we simply have

z + 2 z = A sin(qt + p)
00

The homogeneous solution is given by

zh (t) = C sin(t + ) :

Once again there are two cases for the particular solution given by

zp (t) = 2 A; q2 sin(qt + p)  q 6=  

and

zp (t) = ; 2A t cos(qt + p)  q =  :

32
C Coe cients and quantities from Section 10
s q 
 = 2 ; c2 + 9c ; 8c2 =2
2
2

k = 2c2 +21 +  = 2 +21; c


2

1 =  + 
2 =  + 

1 = ; 34 c3 A2x (k2 ; 2) + A2z ]

1 = 32 c3 kA2x
3 = 32 c3 Ax (21 ; k21 ) ; 38 kc4 A3x (4 + k2 )
4 = 23 c3 Ax (22 ; k22 ) ; 38 c4 kAx A2z
5 = 23 c3 Ax (22 + k22 ) + 38 c4 kAx A2z
6 = 2 + 2!2Ax (k ; 1) + 5

1 = 43 c3 A2x (2 + k2 )
2 = 34 c3 A2z
3 = ; 32 c3 Ax (221 ; k21 ) ; 12 c4 A3x (2 + k2 )
4 = 23 c3 (kAx 22 + Az 21 ; 2Ax 22 ) ; 32 c4 Ax A2z
5 = ; 32 c3 (2Ax 22 + kAx 22 ; Az 22 ) ; 32 c4 Ax A2z

1 = 23 c3 Ax Az
3 = ; 32 c3 22 Az ; 83 c4 A3z
4 = 32 c3 (Ax 21 ; Az 21 ) ; 38 c4 Az A2x (4 + k2 )
5 = ; 23 c3 (Ax 22 ; Az 21 ) + 38 c4 A2x Az (4 + k2 )

33
21 = ;1 =32
22 = 1 =2
31 = ; 38+24
32 = 38;24

1 = ;3c3(Ax (220 + 21 + k2 21 ) + 12 Az (21 + 22 ))


+ 23 c4 Ax (A2x (k2 ; 2) + 2A2z )
2 = 32 c3 Ax (21 ; 2k20 + k21 ) + 32 c4 kAx A2x ( 43 k2 ; 1) + 12 A2z ]
3 = 23 c3 (Ax (21 + 22 ) + Az (22 ; 220))
+ 23 c4 Az ( 34 A2z + ( 12 k2 ; 2)A2x )

20 = ; 1 =(1 + 2c2 )


21 = 164 4+412;(c1;(42) ;+ 21c;2 +c2c) + 1
2

2 2 2

22 = 164 + 4;22((4  + 1 ; c2 )


2

c2 ; 2) ; 2c22 + c2 + 1
31 = 6(81 3 + 4 ) ; (92 + 1 ; c2 )(3 + 4 )
4 + 92 (c2 ; 2) ; 2c22 + c2 + 1
21 = 1644+41;2 (c 1 (4 2 + 1 + 2c2)
2 ; 2) ; 2c2 + c2 + 1
2

22 = 164 + 42 (c 4;2)2 ; 2c2 + c + 1


2 2 2

31 = 6  ( 3 +  4 ) ; (9  2 + 1 + 2c )( +  )
2 3 4
814 + 92 (c2 ; 2) ; 2c22 + c2 + 1
32 = ; 2k 6
a1 = ; 32 c3 (2a21 ; a23 + d21 (2 ; 3 )) ; 38 c4 (8 ; 4 ; k2 (2 +  ))
a2 = 23 c3 (a24 ; 2a22 ) + 89 c4

34
c3 (k2 ; 2)
a21 = 34(1 + 2c2)
a22 = 4(1 +c32c )
3
2
3 c
a23 = ; 4kD (3k3  ; 6k(k ; ) + 4)
3
1
3 c
a24 = ; 4kD (2 + 3k)
3
1

a31 = ; D (c3 (ka23 ; b21 ) + kc4 (1 + 14 k2 )) + 9 +


9  2 1 ; c2 (3c (2a ; kb ) + c (2 + 3k2))
2 2D2 3 23 21 4

a32 = ; 49D (4c3 (ka24 ; b22 ) + kc4 )


2

;  2+D1 ; c2 ) (c3 (kb22 + d21 ; 2a24) ; c4 )


3(9 2

b21 = ; 32cD3 (3k ; 4)


1
3
b22 = Dc 3
1

b31 = D1 3(3c3 (kb21 ; 2a23) ; c4 (2 + 3k2 ))


2
+(92 + 1 + 2c2)(12c3 (ka23 ; b21 ) + 3kc4(4 + k2 ))=8]
b32 = D1 3(3c3 (kb22 + d21 ; 2a24) ; 3c4 ) + (92 + 1 + 2c2 )(12c3 (ka24 ; b22 ) + 3c4 k)=8]
2

b33 = ; 16k (12c3 (b21 ; 2ka21 + ka23 ) + 3c4 k(3k2 ; 4) + 16s1(k ; 1))
b34 = ; 8k (;12c3ka22 + 3c4 k + 8s2 (k ; 1))
b35 = ; 16k (12c3 (b22 + ka24 ) + 3c4 k)

d21 = ; 2c32
d31 = 6432 (4c3 a24 + c4 )
d32 = 6432 (4c3 (a23 ; d21 ) + c4 (4 + k2 ))

D1 = 164 + 42 (c2 ; 2) ; 2c22 + c2 + 1


D2 = 814 + 92 (c2 ; 2) ; 2c22 + c2 + 1
D3 = 2((1 + k2 ) ; 2k)

35
l1 = 2s1 2 + a1
l2 = 2s2 2 + a2
3 
s1 1
= D 2 c3 2a21 (k ; 2) ; a23 (k + 2) ; 2kb21] ; 83 c4 3k4 ; 8k2 + 8]
2 2

1
3
3
s2 = D 2 c3 2a22 (k2 ; 2) ; a24 (k2 + 2) ;  2kb22 + d21 (2 ; 3 )]
3
3

+ 8 c4 (8 ; 4 ) ; k (2 +  )]
2

D Solutions to exercises
Solution 2.1
If
;k 
U (x) = j;Xkj = (P X 2 )1=2
i
then
;k 2X = kXi :
@X=@Xi = ; 12 (P X 2 )3=2 i jX j3
i
Thus
;rU (X ) = ; jkX
X j3 :

Solution 2.2
Let u = X and v = X_ , then
u_ = v
v_ = ;ku=juj3 :
Written component-wise, this is a system of 6 rst-order ordinary di
erential equations.
Solution 2.3

d _ _ _ 
dt (X (t)  X (t)) = X (t)  X (t) + X (t)  X (t)
= 0+0:
This implies that X  X_ is a constant of the motion and thus the motion is in a plane which has
normal X  X_ .

36
Solution 2.4
Substituting X (t) = (a cos !t a sin !t 0) into the equation of motion we compute:
0 cos !t 1 0 cos !t 1
ka
;a!2 @ sin !t A = ; a3 @ sin !t A :
0 0
This equation is satised when
!2 = ak3 :
Solution 3.1
Since r = X2 ; X1 ,
r = X2 ; X1
= ;(Gm1 + Gm2 )r=jrj3 :
This is the central force motion equation with k = G(m1 + m2 ).
Solution 3.2
p
In Exercise 2.4, we found that ! = k=a3. Using k = G(m1 + m2)  6:672  10 ;11 1:9891 
10 m3 s 2 and a = 1:496  1011m, we nd that
30 ;

!  1:991  10 7s 1
; ;

and the period T = 2=! is


T  3:156  107sec  365:26 days .

Solution 3.3
The center of mass r0 is given by solving the following equation:
(m1 + m2 )r0 = m1 X1 + m2 X2 :
Thus,
r0 = (m1 X1 + m2 X2 )=(m1 + m2) :
Substituting in the equations of motion for X1 and X2 we nd that r0 = 0. Thus we can integrate
and solve explicitly to get r0 (t) = at + b, where a and b and the initial velocity and position,
respectively.
Solution 3.4
Since r0 = 0, we have m1 X1 + m2 X2 = 0. By denition, r = X2 ; X1 . We rst eliminate X2 to
nd
X1 = ; mMr
2

37
and then nd
X2 = m
Mr:
2

Solution 4.1
The potentials U1 and U2 from the masses m1 and m2 located at X1 and X2 , respectively, acting
on a mass m located at X are given by
U1 = ;Gm1 m=jX ; X1 j  U2 = ;Gm2 m=jX ; X2 j :
The full potential U is given by U = U1 + U2 . The force on the mass m is given by F = ;rU =
;rU1 ; rU2 .
Solution 4.2
Let X1 (t) = a(cos !t sin !t 0) and X2 (t) = ;b(cos !t sin !t 0). Then
X = ; jX ;Gm Gm2
X (t)j3 (X ; X1 (t)) ; jX ; X (t)j3 (X ; X2 (t)) :
1
1 2

Solution 4.3
Recall that l = a + b and M = m1 + m2 . Because the center of mass is at the origin, we have
the relation am1 = bm2 . Then
b = m1 a=m2 = m1 a = m1
l a+b m2 a + m1 a M
and
a = m2 b=m1 = m2 b = m2 :
l a+b m2 b + m1 b M
Solution 4.4
Recall that Rt 1 = RtT = R t . Let W = Rt X , then
;
;

W_ = R_ t X + Rt X_
= R_ t R t W + Rt X_ ;

and solving for X_ we nd


X_ = R t W_ ; R t R_ t R t W :
; ; ;

Next,
W = Rt X + R_ t X_ + R_ t X_ + Rt X
= Rt R t W + 2(R_ t R t W_ ; R_ t R t R_ t R t W ) + Rt (;rX (X t))
; ; ; ;

= (Rt R t ; 2R_ t R t R_ t R t )W + 2R_ t R t W_


; ; ; ;

;Rt(RT t ) 1 rW (R t W ) :
;
;
;

38
Now let 0 0 1 0
1
K = R_ t R t = @ ;1
; 0 0 A
0 0 0
and notice that 0 ;1 0 0
1
;(Rt R t ; 2R_ t R t R_ t R t ) = @ 0 ;1
; ; ; 0 A = K2 :
0 0 0
Thus, letting U (W ) = (R t W t) we nd that
;

W ; 2K W_ + K 2 W = ;rU (W ) :

Solution 5.1
The libration points are stationary points and are solutions to r$ = 0. This is equivalent to
x ; Ux = 0
y ; Uy = 0
Uz = 0 :
Let 1 = jX ; (1 ; )e1 j and 2 = jX + e1 j be the distances from the primary masses to X . Then
the above equations for the equilibrium points are
x ; (x ; (13 ; )) ; (1 ; )(3x + ) = 0
1 2

y ; y ; (1 ; )y = 0
1
3 32
; z ; (1 ; )z = 0 :
1
3 32
From the third equation, we discover that z = 0. From the second, we nd there are two types of
solutions, one type for y 6= 0 and another for y = 0. We will rst solve for the solutions with y 6= 0.
For y 6= 0,
1 ; =31 ; (1 ; )=32 = 0 :
We use this equation to eliminate 2 from the rst equation, thus discovering that 1 = 1. Using
this, we nd that also 2 = 1. Thus the two points which are unit distance from both masses are
libration points. Since the masses are unit distance apart, these libration points lie on the vertices
of an equilateral triangle whose other p two vertices are the locations of the masses. Solving explicitly,
we nd they are given by (1=2 ; 
3=2 0) :
When y = 0, then we are looking for stationary points which lie on the line connecting the two
masses | the collinear libration points. We are reduced to solving for x in the following equation:
x ; jx(x;;(1(1;;)j))3 ; (1 ;jx+)(xj+3 ) = 0 :

39
By considering the three di
erent regions separated by the location of the primary masses
; < x < 1 ; 
1;<x
x < ; 
this equations may be written as
x ; s1 (x ; (1; ))2 ; s2 (x1+;)2 = 0 
where the si have the following values in the three regions, respectively,
s1 = ;1 s2 = 1
s1 = 1 s2 = 1
s1 = ;1 s2 = ;1 :
This in turn leads to solving a fth-order algebraic equation (for each region).
Solution 6.1
Let
J = 2$(x y z ) ; (x_ 2 + y_ 2 + z_ 2) :
We compute
dJ =dt = 2 < rX $ X_ > ;2 < X  X_ >
= 2 < rX $ ; X X_ >
= 2 < ;(2y_ ;2x_ 0) (x_ y_ z_ ) >
= 0:
So J is an integral.
Solution 6.2
The following Maple code does the job.
with(plots):
mu := 0.25
rho1 := sqrt((x-(1-mu))\^2+y\^2+z\^2)
rho2 := sqrt((x+mu)\^2+y\^2+z\^2)
U := -(mu/rho1+(1-mu)/rho2)
Omega := 0.5*(x\^2+y\^2)-U
C := 2.0
implicitplot3d(Omega=C/2, x=-2..2, y=-2..2, z=-2..2,
title=cat(`Mu=`,convert(mu,string),`, C=`,convert(C,string)),
axes=boxed)

40
Solution 7.1
Assuming that (x(t) y(t) z (t)) satises the equations of motion, we must show that the sym-
metric solutions do also. Doing this for (x(t) y(t) ;z (t)) we get
x(t) ; 2y_ (t) = $x (x(t) y(t) ;z (t))
y(t) + 2x_ (t) = $y (x(t) y(t) ;z (t))
;z(t) = $z (x(t) y(t) ;z (t)) :
Since
$x(x(t) y(t) ;z (t)) = $x (x(t) y(t) z (t))
$y (x(t) y(t) ;z (t)) = $y (x(t) y(t) z (t))
$z (x(t) y(t) ;z (t)) = ;$z (x(t) y(t) z (t)) 
the above equations hold, and (x(t) y(t) ;z (t)) is indeed a solution.
Similarly substituting (x(;t) ;y(;t) z (;t)) into the equations of motion, we get
x(;t) ; 2y_ (;t) = $x (x(;t) ;y(;t) z (;t))
;y(;t) ; 2x_ (;t) = $x (x(;t) ;y(;t) z (;t))
z(;t) = $z (x(;t) ;y(;t) z (;t)) :
Now,
$x (x(;t) ;y(;t) z (;t)) = $x (x(;t) y(;t) z (;t))
$x (x(;t) ;y(;t) z (;t)) = ;$x(x(;t) ;y(;t) z (;t))
$z (x(;t) ;y(;t) z (;t)) = $z (x(;t) y(;t) z (;t)) 
so the equations hold (at time -t), and (x(;t) ;y(;t) z (;t)) is a solution.
We verify the third symmetry in the same fashion.
Solution 8.1
For the Lagrangian
L = 12 x_ 2 ; U (x)
we have, from the Euler-Lagrange equations,
d @L ; @L
0 = dt @ x_ @x
d
= dt x_ ; rU (x)
= x ; rU (x)
which is the same as the Newton formulation.

41
Solution 8.2
The Lagrangian in rotating coordinates is
L = 12 (W_ ; KW )2 ; U (W ):
The equations of motion are given by the Euler-Lagrange equations
d @L ; @L = 0:
dt @ W_ @W
We have
@L = W_ ; KW
@ W_
d @L = W ; K W_
dt @ W_
@L _
@W = (W ; KW )(;K ) ; rU (W ):
Now use the fact that K T = ;K to see
0 = dt d @L ; @L
@ W_ @W
= W ; K W_ + WK
 _ ; KWK + rU
= W ; 2K W_ + K 2W + rU
which are the equations of motion we know and love.
Solution 9.1
Let p = @L=@ q_, where L(q q_) is the Lagrangian. Then set H (p q) =< p q_ > ;L and compute
@H=@p = q:_
So the rst Hamiltonian equation of motion q_ = @H=@p is trivially satised.
We then compute
@H=@q = ;@L=@q
and see that the second Hamiltonian equation of motion p_ = ;@H=@q becomes
d @L @L
dt @ q_ = @q 
which is the Lagrangian form of the equations of motion.
Solution 9.2
The Jacobian integral is
J (q q_) = 2$(q) ; q_2 = ; < K 2q q > ;2U (q) ; q_2 :
42
The Hamiltonian is
H (q q_) = 21 (q_ ; Kq)2 + < q_ ; Kq Kq > +U (q)
= 12 q_2 ; < q_ Kq > + 12 < Kq Kq > + < q_ Kq > ; < Kq Kq > +U (q)
= 12 q_2 ; 12 < Kq Kq > +U (q)
= 21 q_2 + 12 < K 2q q > +U (q) 
and we see H = ;J =2.
Solution 10.1 (Solution due to Molly Megraw.)
Let x = i be the x-coordinate of the collinear libration point Li . Then x solves
x ; jx ; j3 (x ;  ) ; jx + j3 (x + ) = 0
0
0
0

where  = 1 ; . Set
0

S1 = x  3  S2 = x+ 3 :
0

j ; 0j j j

Then
S1 = x ; xS2;(x+ )  0

and we are trying to show S1 + S2 > 1. But

 S1 + S2 > 1 ()
S2 1 ; x ;+  > 1 ; x ;x  ()
x
0 0

S2 ;x;;  > x;; ()


0 0

0 0

S2 x ;;1 > x;; :


0

0 0

Now if x is at L2 , then x ;  > 0, and the above holds if and only if


0

S2 <  () 0

1
jx + j3 < 1 ()
jx + j > 1:
But this holds, since the distance between the primaries is 1, and the mass at 1 ;  separates L2
from the mass at ;. The inequalities above are reversed for x at L1, leading to the inequality
jx + j < 1. This inequality holds trivially, since L1 is between the primaries.

43
Corrections list for Richardson's work
In this section we list the di
erences between the formulae presented here and those in the original
work of Richardson 14].

Typographical errors
These rst corrections to 14] appear to simply be typos. We shall present them here for complete-
ness.
1. The rst term in the formula for 3 is missing a factor of 1=2. This also shows up in the
parameter d31 . The formula for d31 is corrected in the journal paper 15].
2. The expressions for a32 and b32 have a factor of  which should not be there. The journal
paper only uses  = ;1 and thus does not have this factor the expressions for a32 and b32 are
given correctly.
3. In equations 2-65, the rst occurrence of 21 should be 20 .
4. In equations 2-68, the term cos 31 with coecient d21 Ax Ax should be cos 21 .
5. In equations 2-65 and 2-68, the term sin 1 in the z equation should switch between sin and
cos with various signs depending on the value of n.

Removal of the third order secular terms


The next corrections are due to the fact that the secular terms in the third-order equations were
removed incorrectly. This a
ects both the third order equations and the third order solutions. The
reader can easily refer back to the text to see the di
erences between the expression in this paper
and the original 14]. Essentially, this results in an additional term with coecient 32 being added
to the expression for y3 , which in turn results in third order corrections to the amplitude of sin 1
in the nal expression for y( ) (with new coecients b33  b34  b35 ).

44

You might also like