Computation of Periodic Solutions of Conservative Systems With Application To The 3-Body Problem

Tutorials and Reviews
International Journal of Bifurcation and Chaos, Vol. 13, No. 6 (2003) 1353–1381
c World Scientific Publishing Company
COMPUTATION OF PERIODIC SOLUTIONS OF

CONSERVATIVE SYSTEMS WITH
APPLICATION TO THE 3-BODY PROBLEM
E. J. DOEDEL∗ , R. C. PAFFENROTH and H. B. KELLER

ACM 217-50, California Institute of Technology, Pasadena CA 91125, USA
[email protected]
D. J. DICHMANN
Consultant, Torrance, CA, USA
J. GALÁN-VIOQUE
Matemática Aplicada II, Escuela Superior de Ingenieros,
Universidad de Sevilla, Camino de los Descubrimientos s/n,
Sevilla 41092, Spain
A. VANDERBAUWHEDE
Vakgroep Zuivere Wiskunde en Computeralgebra,
Universiteit Gent, Krijgslaan 281, B-9000 Gent, Belgium
Received May 24, 2002; Revised August 22, 2002
We show how to compute families of periodic solutions of conservative systems with two-point
boundary value problem continuation software. The computations include detection of bifurca-
tions and corresponding branch switching. A simple example is used to illustrate the main idea.
Thereafter we compute families of periodic solutions of the circular restricted 3-body problem.
We also continue the figure-8 orbit recently discovered by Chenciner and Montgomery, and nu-
merically computed by Simó, as the mass of one of the bodies is allowed to vary. In particular,
we show how the invariances (phase-shift, scaling law, and x, y, z translations and rotations) can
be dealt with. Our numerical results show, among other things, that there exists a continuous
path of periodic solutions from the figure-8 orbit to a periodic solution of the restricted 3-body
problem.
Keywords: Numerical bifurcation; numerical continuation; conservative systems; periodic solu-

tions; boundary value problems; three-body problem; restricted three-body problem; figure-8
orbit.
1. Introduction single planar figure-8 orbit (see Fig. 15). The exis-
tence of this solution was proved by Chenciner and
There is much interest recently in new periodic so- Montgomery [2000], and it was computed numeri-
lutions of the classical n-body problem of celes- cally by Simó. Simó has also discovered many other
tial mechanics. A particularly surprising solution is planar single-curve periodic solutions for n ≥ 3
the one where 3 bodies follow each other along a [Simó, 2000, 2002].
∗
Author for correspondence. (On leave from Concordia University, Montreal).
1353
1354 E. J. Doedel et al.
In this paper we show how two-point bound- multiple invariances (phase-shift, scaling law, and x,
ary value problem continuation software like AUTO y, z translations and rotations) can be dealt with.
[Doedel, 1981; Doedel et al., 1997; Doedel et al., The local continuation of the figure-8 orbit, as the
2000] can be used to compute families of periodic mass of one of the bodies is allowed to vary on a
solutions of conservative systems, i.e. systems hav- very small scale, is presented in Sec. 4.2. In Sec. 4.3
ing a first integral. The method is applicable to we discuss the global continuation of the figure-8 or-
the computation of periodic solutions of the n-body bit, as the mass of one of the bodies varies between
problem and, in particular, to the numerical contin- 1 and 0. Our numerical results show, among other
uation of the figure-8 orbit, as the mass of one of things, that there exists a continuous path of peri-
the bodies is allowed to vary. The basic idea of the odic solutions from the figure-8 orbit to a periodic
method is very simple, and for the case of one con- solution of the restricted 3-body problem.
served quantity it was already used, without spe-
cific mention, in [Doedel, 1981], and subsequently in
2. Continuation of Solutions
many other applications, for example [Zufirı́a, 1987]
and [Doedel et al., 1991a]. Numerical continuation enables the computa-
First, in Secs. 2.1 and 2.2, we recall some ba- tion of solution manifolds (“solution families”).
sic notions of numerical continuation. In Sec. 2.3, we Most existing algorithms are for the computa-
give a simple example to illustrate the principal idea tion of one-dimensional solution manifolds (“solu-
used in the computation of families of periodic so- tion branches”), see, e.g. [Rheinboldt, 1986; Doedel
lutions of conservative systems. In Sec. 2.4 we state et al., 1991b; Doedel et al., 1991c; Seydel, 1995;
a mathematical result that illustrates justification Allgower & Georg, 1996; Beyn et al., 2002], and
of the general computational procedure under ap- [Kuznetsov, 1998, Chap. 10], but continuation al-
propriate assumptions. A much more detailed theo- gorithms have also been developed for higher-
retical treatment is given in [Muñoz-Almaraz et al., dimensional manifolds; see, in particular, the recent
2003]. paper by Henderson [2002], and references therein.
In Sec. 3 we apply the continuation method to In this paper we only consider the computation of
the circular restricted 3-body problem (CR3BP). solution branches.
The mass-ratio parameter used in our calculations
is approximately that of the Earth–Moon system. 2.1. Finite-dimensional systems
First we compute a family of planar periodic solu-
tions of the CR3BP, namely, the Lyapunov orbits To recall the basic notions of continuation, first con-
that originate from the libration point L1. Our cal- sider the finite-dimensional equation
culation illustrates how easily these can be com- F (X) = 0 , F : Rn+1 → Rn , (1)
puted by numerical continuation until very close to
a collision orbit. Then, in Sec. 3.2, we give graphical where F is assumed to be sufficiently smooth. This
results that illustrate the rich solution structure of equation has one more variable than it has equa-
3D periodic orbits in the CR3BP. Bifurcations are tions. Given a solution X0 , one has, generically,
easily detected in our boundary value continuation a locally unique branch of solutions that passes
approach, and lead to previously known and possi- through X0 . To compute a next point, say, X1 , on
bly new periodic orbits. The known orbits include this branch, one can use Newton’s method to solve
the Halo orbits, whose behavior and stability prop- the extended system
erties have been computed before; see, for exam- (a) F (X1 ) = 0 ,
ple [Howell, 1984] for a detailed numerical study. In (2)
(b) (X1 − X0 )∗ Ẋ0 = ∆s .
Sec. 3.3 we discuss the use of libration point orbits
in space mission design and we give some related Here Ẋ0 is the unit tangent to the path of solutions
references. at X0 , the symbol ∗ denotes transpose, and ∆s is a
Our most elaborate application of the ba- step size in the continuation procedure. The vector
sic method for computing periodic orbits in con- Ẋ0 is, of course, also a null vector of the Jacobian
servative systems is presented in Sec. 4, where matrix FX (X0 ), and can be computed at little cost
we continue the figure-8 orbit of Chenciner and [Doedel et al., 1991b]. The geometrical interpreta-
Montgomery, as the mass of one of the bodies tion of this well-known method, known as Keller’s
is allowed to vary. In Sec. 4.1 we show how the pseudo-arclength method [Keller, 1977], is shown in
Computation of Periodic Solutions of Conservative Systems 1355
This equation must be solved for X1 = (x1 (·), T1 ,

λ1 ), given a solution X0 = (x0 (·), T0 , λ0 ) and the
path tangent Ẋ0 = (ẋ0 (·), Ṫ0 , λ̇0 ). Here T1 ∈ R is
X1
the unknown period. Equation (5a2 ) imposes unit
X0 periodicity, after rescaling of the independent vari-
able t. Equation (5a3 ) is a phase condition, which
X fixes the phase of the new orbit x1 (·) relative to the
0
∆s given orbit x0 (·). It may be replaced by the classical
Poincaré phase condition
Fig. 1. Pseudo-arclength continuation.
(a03 ) (x1 (0) − x0 (0))∗ x00 (0) = 0 .
However, the integral phase condition (5a 3 ) has the
Fig. 1. The size of the pseudo-arclength step size ∆s desirable property of minimizing phase drift rela-
is normally adapted along the branch, depending, tive to x0 (·), which often allows bigger continuation
for example, on the convergence history of Newton’s steps to be taken. Equation (5b) is the functional
method. form of the pseudo-arclength constraint (2b). More
It can be shown that the above continuation details on this boundary value approach for com-
method works near a regular solution X 0 , i.e. if the puting periodic solutions can be found in [Doedel
null space of FX (X0 ) is one-dimensional. In fact, et al., 1991c]; further references include [Doedel,
in this case the Jacobian of (2) evaluated at X 0 , 1981; Jepson, 1981; Doedel et al., 1984].
i.e. the n + 1 by n + 1 matrix In each continuation step Eq. (5) is solved
! by numerical boundary value algorithms. In par-
FX (X0 ) ticular, AUTO uses piecewise polynomial colloca-
, (3)
Ẋ0∗ tion with Gauss–Legendre collocation points (“or-
thogonal collocation”) [de Boor & Swartz, 1973;
is easily seen to be nonsingular. By the Implicit
Ascher et al., 1995; Ascher & Petzold, 1998], simi-
Function Theorem, this guarantees the existence
lar to COLSYS [Ascher et al., 1981], and COLDAE
of a locally unique solution branch through X 0 .
[Ascher & Spiteri, 1995], with adaptive mesh se-
This branch can be parametrized locally by ∆s.
lection [Russell & Christiansen, 1978]. Combined
Moreover, for ∆s sufficiently small, and for suffi-
(0) with continuation, this allows the numerical so-
ciently accurate initial approximation (e.g. X 1 =
lution of “difficult” problems, as illustrated by
X0 + ∆sẊ0 ), Newton’s method for solving (2) can the near-homoclinic and near-collision orbits in
be shown to converge. this paper. (For other challenging applications see,
for example [Doedel, 1997].) Orthogonal colloca-
2.2. Periodic solutions
tion also has the desirable property of preserving
Here we recall how pseudo-arclength continuation the symplectic structure of Hamiltonian systems.
can be used to compute a branch of periodic solu- AUTO determines the characteristic multipliers (or
tions of a dynamical system Floquet multipliers), that reflect asymptotic stabil-
x0 (t) = f (x(t), λ) , f : R n × R → Rn , (4) ity and bifurcation properties, as a by-product of
the decomposition of the Jacobian of the bound-
where λ ∈ R is a physical parameter. In this case the ary value collocation system [Doedel et al., 1991c;
continuation step corresponding to Eq. (2) takes the Fairgrieve & Jepson, 1991]; see also [Lust, 2001].
form of the following constrained periodic boundary
value problem:
(a1 ) x01 (t) = T1 f (x1 (t), λ1 ) , 2.3. A model conservative system
(a2 ) x1 (0) = x1 (1) , As a simple introductory example, consider the
Z 1
system
(a3 ) x1 (τ )∗ x00 (τ )dτ = 0 ,
0 (5) x01 = x2 ,
(6)
Z 1
∗
x02 = −x1 (1 − x1 ) ,
(b) (x1 (τ ) − x0 (τ )) ẋ0 (τ )dτ
0 which has one first integral, namely the Hamilto-
+ (T1 − T0 )Ṫ0 + (λ1 − λ0 )λ̇0 = ∆s . nian H(x1 , x2 ) = (1/2)x21 + (1/2)x22 − (1/3)x31 . This
equation has a family of periodic solutions, of which sumed that there is an explicitly available problem
representative orbits can be seen in Fig. 2(b). The parameter, namely λ, which is part of the set of
family originates at (x1 , x2 ) = (0, 0), and termi- unknowns in each continuation step.
nates in a homoclinic orbit containing the saddle A straightforward approach to deal with this
point equilibrium (x1 , x2 ) = (1, 0). discrepancy is to use the conserved quantity (the
Note that Eq. (6) does not contain any pa- Hamiltonian, in the example above), to eliminate
rameters; yet it allows an entire family of periodic one of the variables; then choose a suitable Poincaré
solutions. This nongeneric behavior is typical for section for the flow, and look for periodic or-
conservative systems. However, in Eq. (5) it is as- bits in the reduced problem. This scheme can be
0.8
1.2
1.0
0.8
0.2
L2 norm
0.6
0.4 x2 −0.2
0.2
0.0
−0.2 −0.8
−1.0 −0.5 0.0 0.5 1.0 −0.5 0.0 0.5 1.0
λ x1
(a) (b)
1.0
0.5
x1
0.0
−0.5
0.0 0.2 0.4 0.6 0.8 1.0
scaled time
(c)
Fig. 2. (a) Bifurcation diagram for Eq. (7), computed with AUTO. The vertical axis is the L2 norm. For λ 6= 0 there are no
nontrivial periodic solutions, while for λ = 0 there is a vertical branch of periodic orbits. The computed branch terminates
near a homoclinic orbit where kxk = 1. (b) Phase plane representation of some representative periodic orbits. (c) As a function
of scaled time t, the solution component x1 tends to a “downward delta function”, which stays at practically the same location
as the period T tends to infinity, due to the phase condition (5a3 ).
extended to the case of several constants of mo- ponent x1 at practically the same location when the
tion and has been extensively used in the literature, period becomes large, as can be seen in Fig. 2(c).
see, for example [Simó, 1996]. It requires numerical This allows much bigger continuation steps to be
integration of the differential equations, which can taken compared to phase conditions that allow this
lead to instability in case of very stiff equations. sharp dip to move.
Numerical integration is also difficult to apply to The unfolding term in Eq. (7) was chosen on
very unstable orbits. It also makes it difficult to the basis of a physical argument, namely the ef-
use integral constraints, such as the integral phase fect of positive and negative damping. It is more
condition (5a3 ) and the continuation equation (5b), satisfactory, however, to have a precise procedure
which have often significant computational advan- for regularizing periodic orbit continuation in con-
tages over pointwise constraints. servative systems. The general results in [Muñoz-
Another approach, adopted in this paper, is to Almaraz et al., 2003], (see Sec. 2.4 below) imply
reformulate the problem so that boundary value that the unfolding parameter times the gradient of
continuation methods can be applied. In our model the conserved quantity is an appropriate choice for
problem this can be done by simply introducing an the unfolding term. In the case of our test prob-
“unfolding” term, with unfolding parameter λ, for lem, Eq. (6), this gives the following alternate for-
example, mulation that can be used to continue the periodic
x01 = x2 , orbits:
(7) x01 = x2 + λx1 (1 − x1 ) ,
x02 = −x1 (1 − x1 ) + λx2 . (8)
x02 = −x1 (1 − x1 ) + λx2 .
The unfolding term chosen here, corresponds to
damping. When the damping parameter λ is zero
then Eqs. (6) and (7) are identical. When λ is 2.4. A model result for
nonzero then all nontrivial periodic solutions dis- conservative systems
appear. Starting the computation from, for exam-
Here we state a simple continuation result for the
ple, λ = −1, with (x1 , x2 ) = (0, 0), AUTO locates
case of a conservative system with one conserved
(x1 , x2 , λ) = (0, 0, 0) as a Hopf bifurcation from
quantity. More general results are given in [Muñoz-
the trivial solution, and then computes the ema-
Almaraz et al., 2003]. Consider the dynamical
nating branch of periodic solutions at λ = 0 using
system
the system (5). Along the periodic solution branch
the value of λ, computed as part of the solution in x0 (t) = f (x(t)) , f : R n → Rn , (9)
each continuation step, will be zero (up to numeri- which has C : Rn → R as a conserved quantity,
cal precision). Figure 2(a) shows the computed bi- i.e. C(x(t)) is constant along any orbit x(t). Let
furcation diagram, in which the vertical axis is cho- x0 (t) be a nontrivial periodic solution of Eq. (9)
sen to be the L2 -norm of the solution. Each point with period T0 . Finding a periodic solution with pe-
on the “vertical branch” in the bifurcation diagram riod T of Eq. (9) is equivalent to solving the bound-
represents a periodic solution. Note that there is no ary value problem
“distinguished” parameter in pseudo-arclength con-
tinuation; λ is just one of the quantities solved for x0 (t) = Tf (x(t)) + λ∇C(x(t)) ,
(10)
in each continuation step. This allows the compu- x(1) = x(0) ,
tation past folds and, as the current example illus-
provided λ vanishes. For theoretical purposes, the
trates, continuation of vertical solution branches.
continuation problem can be formulated as
As mentioned in Sec. 2.2, orthogonal colloca-
tion with adaptive mesh selection is used to solve p = x(1; p, T, λ) , (11)
the boundary value problem (5) at each continua-
where x(t; p, T, λ) is the solution of Eq. (10) with
tion step. This allows the solution to be computed
x(0) = p as initial condition. By hypothesis, x 0 (t)
up to large periods; i.e. until very close to the homo-
is a solution of Eq. (10) such that x0 (1) = x0 (0),
clinic orbit that terminates the branch. Periods as
T = T0 and λ = 0. Let V (t) be the fundamental so-
large as T = 109 are easily attained with a relatively
lution matrix of the variational equation, i.e. V (·)
small number of mesh intervals.
satisfies
Note that the integral phase condition (5a 3 ) has
the property of keeping the “dip” in solution com- V 0 (t) = T0 Df (x0 (t))V (t) , V (0) = I . (12)
Using, for simplicity of presentation, the classical the orthogonal coordinate system. The parameter µ
Poincaré phase condition (5a03 ), rather than the in- represents the ratio of the mass of the smaller pri-
tegral phase condition (5a3 ), we can formulate the mary to the total mass. In this section we consider
continuation problem as finding the zeros of the Earth–Moon system, for which µ = 0.01215.

x(1; p, T, λ) − p
The units are chosen so that the distance between
F (p, T, λ) = . (13) the primaries, the sum of the masses of the pri-
(p − x0 (0))∗ x00 (0)
maries, and the angular velocity of the primaries are
The Implicit Function Theorem can now be used to all equal to one. The larger and smaller primaries
prove the following: are then located at (−µ, 0, 0) and (1 − µ, 0, 0),
respectively, and the equations of motion are
Theorem 2.1. Let x0 be a solution of Eq. (10) given by
whose monodromy matrix V (1) has 1 as eigenvalue
with geometric multiplicity 1. Then there exists a x00 = 2y 0 + x − (1 − µ)(x + µ)r1−3
unique branch of solutions of F (p, T, λ) = 0 pass- − µ(x − 1 + µ)r2−3 ,
ing through (x0 (0), T0 , 0). Moreover the parameter (14)
y 00 = −2x0 + y − (1 − µ)yr1−3 − µyr2−3 ,
λ vanishes along this branch.
z 00 = −(1 − µ)zr1−3 − µzr2−3 ,
The eigenvalues of V (1) are the Floquet mul- where
tipliers (or “characteristic multipliers”) of the peri- p
odic orbit x0 [Meyer & Hall, 1999]. We note that the r1 = (x + µ)2 + y 2 + z 2 ,
algebraic multiplicity of the multiplier 1 is necessar-
p
r2 = (x − 1 + µ)2 + y 2 + z 2 .
ily greater than one, for if it were equal to one then
x0 would be a locally unique periodic solution at This dynamical system has one integral of motion,
λ = 0. The above theorem is essentially a computa- namely the energy (or Jacobi constant)
tionally appropriate restatement of the cylinder the- E = (vx2 + vy2 + vz2 )/2
orem in conservative systems [Meyer, 1999]. It forms
the basis for the justification of the simplest cases − U (x, y, z) − µ(1 − µ)/2 ,
presented in this paper, namely, our model prob- where vx = x0 , vy = y 0 , vz = z 0 , and
lem, and the continuation of periodic orbits in the 1 1−µ µ
restricted 3-body problem. Our computational re- U = (x2 + y 2 ) + + .
2 r1 r2
sults for the general 3-body problem, starting from In accordance with Eq. (5) we rewrite Eq. (14)
the figure-8 orbit of Chenciner and Montgomery, re- as a first order system, we scale time, so that the
quires an extension of the theorem to the case of a period T appears explicitly in the equations, and we
Hamiltonian system with several independent con- add periodic boundary conditions and the integral
served quantities. General results of this nature are phase constraint (5a3 ). Following the discussion in
given in [Muñoz-Almaraz et al., 2003]. the preceding section, we also introduce the unfold-
ing term λ∇E. The resulting system of differential
equations is
3. The Restricted 3-Body Problem
x0 = T vx + λ ∂E/∂x ,
The circular restricted 3-body problem (CR3BP)
describes the dynamics of a body with negligible y 0 = T vy + λ ∂E/∂y ,
mass under the gravitational influence of two mas- z 0 = T vz + λ ∂E/∂z ,
sive bodies, called the primaries, where the pri-
maries move in circular orbits about their barycen- vx0 = T [2vy + x − (1 − µ)(x + µ)r1−3
(15)
ter [Danby, 1992]. Let (x, y, z) denote the position − µ(x − 1 + µ)r2−3 ] + λ ∂E/∂vx ,
of the negligible-mass body in a rotating barycentric
vy0 = T [−2vx + y − (1 − µ)yr1−3 − µyr2−3 ]
coordinate system, where the x-axis points from the
larger to the smaller primary; the z-axis is orthog- + λ ∂E/∂vy ,
onal to the orbital plane, and the y-axis completes vz0 = T [−(1 − µ)zr1−3 − µzr2−3 ] + λ ∂E/∂vz .
1.5 20.0
10.0
0.5
0.0
y
vx
−0.5
−10.0
−1.5 −20.0
−1.0 −0.5 0.0 0.5 1.0 1.5 0.0 0.2 0.4 0.6 0.8 1.0
x scaled time
(a) (b)
Fig. 3. (a) x–y representation of a selection of Lyapunov orbits. The smallest orbits are close to L1, while the largest orbits
are near collision. The Earth and the Moon are located at (−µ, 0) and (1 − µ, 0), respectively, where µ = 0.01215. (b) Velocity
component vx (t) as a function of scaled time t for a near-collision orbit. Note the rapid changes of vx (t) near the Earth and
near the Moon. (The highest velocity along the orbit occurs near the Earth.)
3.1. Periodic orbits near the primaries. All five libration points can be seen in
libration points the schematic bifurcation diagram in Fig. 4, which
will be discussed in detail in Sec. 3.2 below.
The phase space of Eq. (14) has an invariant sub- The libration points L1, L2, L3, each give rise
space of planar orbits, for which z = z 0 = 0. The to a family of planar periodic orbits, the Lyapunov
corresponding reduced system is given by orbits. Computational results for the Lyapunov or-
bits that emanate from L1 are shown in Fig. 3.
x00 = 2y 0 + x − (1 − µ)(x + µ)r1−3
As solutions to [Eq. (16)], i.e. as orbits in four-
− µ(x − 1 + µ)r2−3 , (16) dimensional phase space, these orbits are unstable,
y 00 = −2x0 + y − (1 − µ)yr1−3 − µyr2−3 , with two Floquet multipliers equal to 1, one real
multiplier greater than one, and one real multiplier
where less than one.
p
r1 = (x + µ)2 + y 2 , r2 =
p
(x − 1 + µ)2 + y 2 . However, as orbits in six-dimensional phase
space, i.e. as solutions of Eq. (14), there are various
It is well known that for each value of µ Eq. (16) bifurcations from the Lyapunov orbits. On portions
[and Eq. (14)] has five equilibria in the orbit plane of the solution branch the orbits have complex mul-
of the primaries, called Lagrange points or libra- tipliers that move around the unit circle in the com-
tion points [Szebehely, 1967; Gómez et al., 2001b; plex plane, as can be determined numerically using
AUTO, giving rise to invariant tori, subharmonic bi-
Gómez et al., 2001c]. Three of the libration points,
furcations and period-doubling bifurcations. There
denoted L1, L2 and L3, are collinear with the pri- are also bifurcation points,1 where distinct solution
mary bodies; one of them, L1, lies between the branches intersect, with identical orbit and period
primaries. In the case of the Earth–Moon system, at the point of intersection. (Recall that, in our con-
for which µ = 0.01215, L2 is the libration point tinuation setting, a periodic solution X has compo-
beyond the moon, while L3 is the libration point nents (x(·), T, λ), where x denotes the orbit, T is
beyond the earth. Each of the other two points, the period, and where the unfolding parameter λ is
L4 and L5, forms an equilateral triangle with the zero along the solution branch.)
1
In this paper we reserve the term bifurcation point (or bifurcation orbit), when not further qualified, for transcritical and
pitchfork bifurcations, excluding period-doubling, torus and subharmonic bifurcations.
V
Y
H
L
C
M
Fig. 4. Schematic bifurcation diagram of periodic orbits emanating from L1 and from subsequent bifurcation points. The
five libration points are shown as gray cubes. The red line (labeled L) represents the Lyapunov orbits and the green curve (V)
represents the Vertical orbits (see Figs. 6–8). Any solution branch that touches the gray plane has a planar periodic solution
at that point. The red line of Lyapunov orbits shows two bifurcation points, the first gives the blue family (H) of Halo orbits
(see Figs. 9 and 10) and the second gives the yellow (Y) family of orbits that connects the Lyapunov and Vertical families
(see Fig. 11). The two symmetry related bifurcation points on the blue curve of Halo orbits give rise to the cyan (C) family
of orbits (see Fig. 12). The cyan family of orbits has two symmetry related bifurcation points which give rise to the magenta
(M) family of orbits, which connects the “Vertical” orbits emanating from L4 and L5 (see Figs. 13 and 14).
Moreover, in the full six-dimensional model, the be viewed as a “skeleton” [Deprit & Henrard, 1969],
Jacobian at the libration points L1, L2 and L3 has from which many other solutions may be reached.
two pairs of purely imaginary eigenvalues, giving Also, to keep the presentation relatively simple, we
rise to a second branch of well-known periodic solu- do not present results for the planar and Vertical
tions, namely the so-called Vertical orbits [Howell, families that originate from L2 and L3. Some of
2001]. Furthermore, for µ less than a critical value, the branches that we present lead to L4 and L5;
however, we also omit results for the remaining two
p
µ1 ≡ 1/2 − 23/108 ≈ 0.0385, each of the libra-
tion points L4 and L5 gives rise to three families of families that emanate from each of L4 and L5.
periodic solutions. In the numerical results below
we restrict our presentation to the Lyapunov and 3.2. A tour of the bifurcation
Vertical families from L1, and some of the branches diagram
that bifurcate from them. We do not present results Here we explain the bifurcation diagram in Fig. 4.
here on branches of periodic solutions that bifur- The solution branches shown are those obtained
cate from subharmonic- and period-doubling bifur- by following the two branches of periodic orbits
cations. The solution structure that we present can that emanate from L1, and subsequent branches
Fig. 5. Some well-known periodic orbits: The red plane consists of Lyapunov orbits, with selected orbits shown as curves.
Similarly, the green surface consists of Vertical orbits, and the blue surface of “northern” Halo orbits. The coloring scheme is
the same as used in Fig. 4. The thick blue orbit is the Lyapunov orbit from which the Halos bifurcate. The symmetry related
family of “southern” Halo orbits is not shown.
that originate at bifurcation points along computed branches themselves are only schematic. The rel-
branches. ative positions of the various bifurcating branches
In Fig. 4 we use colored curves to represent should not be interpreted as signifying any physical
various families of periodic solutions. For reference, property of the solutions, other than those discussed
Fig. 4 and later figures in this section show the above.
Earth and Moon as two textured spheres. The five As mentioned, the red line in Fig. 4 emanat-
libration points are shown as gray cubes, and fam- ing from L1 represents the Lyapunov orbits. Two
ilies of periodic orbits are shown emanating from bifurcation points are shown along the path of Lya-
the appropriate cube. For example, the red line of punov orbits, the first giving rise to the blue family
Lyapunov orbits emanates from the cube represent- of Halo orbits (see Figs. 9 and 10). On the green
ing L1. All bifurcation points are marked as small curve of Vertical orbits (see Figs. 6 and 7) there is
white spheres except for four brown colored bifur- a bifurcation point which gives rise to a yellow fam-
cation points at which we do not display the bi- ily of orbits that connects to the second bifurcation
furcating branches. In addition, we include a gray point on the Lyapunov family (see Fig. 11). On the
rectangle in the x–y plane. Any solution branch blue curve of Halo orbits there are two symmetry re-
that touches this plane has a planar solution at lated bifurcation points giving rise to the cyan fam-
that point. For example, since the entire branch of ily of orbits (see Fig. 12). The cyan family of orbits
Lyapunov orbits is planar, the entire red line lies in has two symmetry related bifurcation points which
the gray plane. We emphasize that even though the give rise to the magenta family of orbits, which con-
various visualization aids are in the proper physical nects the “Vertical” orbits emanating from L4 and
position with respect to each other; the bifurcating L5 (see Figs. 13 and 14). There is a brown colored
V1
Fig. 6. The Vertical orbits up to the first bifurcation point: The green surface contains all Vertical orbits up to the first
bifurcation point, which is shown as a thick green orbit. The part of the bifurcation diagram where these orbits are found is
indicated on the right by a white arrow.
bifurcation point on this branch through which point. In this and all following figures in this section
one can reach L3 as well, although this connecting the bifurcation diagram from Fig. 4 is shown on the
branch is not shown here. right, with the appropriate branch portion indicated
We begin our tour of the periodic orbits by by a white arrow. As before, the bifurcation orbit
discussing several well-known families which em- is shown as a thickened tube. Similarly, Figs. 7 and
anate from L1, as shown in Fig. 5. Previous work 8 show the family of Vertical orbits starting at the
has mapped portions of the families of periodic first bifurcation point. As can be seen, these orbits
orbits for various values of µ; cf. [Howell, 2001] grow to encompass the Earth–Moon system, and
and references therein. Several authors have in- end in a period-doubling bifurcation from a family
vestigated bifurcations of these families, including of planar solutions. Accordingly, the green curve in
[Hadjidemetriou, 1975; Ichtiaroglou & Michalodim- Fig. 4 touches the gray plane in two places; the first
itrakis, 1980; Markellos, 1981] and [Hénon, 1997; corresponding to the creation of the family at L1
Hénon, 2001; Howell & Campbell, 1999]. The red and the second at the period-doubling bifurcation
plane in Fig. 5 contains a subset of small-amplitude from a planar solution (B2 in Fig. 7).
Lyapunov orbits from Fig. 3 near L1, with selected There are two bifurcation points on the green
individual orbits shown as curves. Similarly, the branch of Vertical solutions away from planar solu-
green surface contains Vertical orbits and the blue tions. As can be seen in Fig. 4, the first bifurcating
surface contains Halo orbits. The coloring scheme branch connects the red Lyapunov and green Ver-
is the same as used in Fig. 4. As a visualization tical families, and the orbits themselves are shown
aid we show orbits from which other families of pe- in yellow in Fig. 11. The red plane is a collection of
riodic orbits bifurcate as thickened tubes. Accord- Lyapunov orbits and the thick planar orbit is the
ingly, the thick blue orbit which lies in the plane second bifurcation orbit on the Lyapunov family.
of the Lyapunov orbits is the orbit from which the The thick green orbit is the first bifurcation orbit on
Halos bifurcate. To reduce clutter in the diagram we the Vertical branch, as shown in Fig. 6. The yellow
have only plotted the “northern” Halo orbits and orbits are a representative collection of orbits which
not the symmetry related “southern” Halo orbits connect these two bifurcation orbits. Some of the
which are a mirror image of the “northern” Halo orbits on the “Y” branch were plotted in Zagouras
orbits reflected across the x–y plane. and Kazantzis for the Sun–Jupiter case [Zagouras
We now turn our attention to the green Ver- & Kazantzis, 1979]. There is a second symmetry re-
tical orbits. In Fig. 6 we show the Vertical orbits lated branch not shown here, and the whole family
from their origin at L1 up to their first bifurcation of orbits forms a loop as shown in Fig. 4.
Fig. 7. An overall view of the Vertical orbits on the indicated branch portion (V2), between the white and the brown (B1)
bifurcation points. The orbits at the bifurcation points are shown as a thick green orbits. The branch of Vertical orbits connects
via a “reverse period-doubling” (B2) to a family of circular planar orbits (not shown) just beyond the second bifurcation point
(B1). The branch bifurcating from B1 is not shown.
Fig. 8. A closeup view of the Vertical orbits near the first (white) bifurcation point in Fig. 7.
The first bifurcation point on the red Lyapunov 1993; Howell, 2001]. Figure 9 shows a selection of
branch gives rise to the well-known Halo orbits; “northern” Halo orbits between the bifurcation
see [Farquhar & Kamel, 1973; Breakwell & Brown, point on the Lyapunov branch from which they arise
1979], and [Howell, 1984; Howell & Farquhar, up to the first bifurcation point on the Halo family
H1
Fig. 9. The Halo orbits (H1) up to the first bifurcation point: The blue curves are a representative collection of orbits from
the “northern” families of Halo orbits. The thick blue orbit between the Earth and Moon is the Lyapunov orbit from which
the Halo orbits bifurcate. The thick blue orbit near the Moon is the first bifurcation point on the branch of Halo orbits.
H2
Fig. 10. The “northern” Halo orbits (H2) after the first bifurcation point. For reference, a set of red Lyapunov orbits is also
shown. The thick blue orbit near the Moon is the first bifurcation point on the branch of “northern” Halo orbits. The thick
blue orbit which encompasses the Earth is a bifurcation point to a family of planar orbits which is not shown here. Also not
shown are the symmetry related “southern” Halo orbits, although the corresponding branch does appear in the bifurcation
diagram.
itself. We have removed the surface of orbits from Accordingly, the blue family of Halo orbits in Fig. 4
this figure in order to make the two bifurcation or- is a loop.
bits easier to distinguish. Figure 10 begins where As is evident, the connectivity of the bifurca-
Fig. 9 ends and shows the “northern” Halo orbits tion diagram is quite complex. On the branch of
from the first bifurcation point on the Halo branch Halo orbits there are two symmetry related bifurca-
up to the second bifurcation point which gives rise tion points. They are connected by the cyan loop of
to a planar family of solutions, not shown here. As orbits shown in Fig. 4. Two additional symmetry re-
can be seen in Fig. 4, the Halo branch can be con- lated bifurcation points occur on the cyan loop, and
tinued past the planar bifurcation point to obtain they give rise to the magenta branch which twice
the symmetry related branch of “southern” Halos. intersects the cyan loop transcritically.
Fig. 11. A family of orbits (Y) connecting the Lyapunov and Vertical orbits. The red plane is a collection of Lyapunov orbits
and the thick red orbit is the second bifurcation point in the Lyapunov family. The thick green orbit is the first bifurcation
point on the Vertical branch, as also shown in Fig. 6. The yellow orbits are a representative collection of orbits which connect
these two bifurcation points. There is a second symmetry related branch not shown here which consists of the above orbits
reflected across the x–z plane. Accordingly, the whole family of orbits forms a loop as shown in Fig. 4.
Fig. 12. The cyan curves are a representative collection of orbits which emanate from the first bifurcation point on the Halo
family. The elliptical bifurcation orbit close to the Moon from Fig. 9 corresponds to the small thick cyan orbit. The thick
cyan figure-8 orbit in the middle of the figure is a bifurcation orbit that connects with a family of “Vertical”-like orbits which
emanates from L4.
On the cyan loop there are four symmetric The thick cyan figure-8 orbit in the middle of Fig. 12
segments separated by the four bifurcation points. is a bifurcation orbit and connects with the magenta
Figure 12 shows a representative collection of orbits family.
on the cyan loop from one of these segments. The The magenta family of orbits is of particular
other segments have the same structure as Fig. 12 interest in that it connects the “Vertical”-like or-
except that they are reflected across the x–y plane bits from L4 with the “Vertical”-like orbits from L5.
and/or the x–z plane. The small thick cyan orbit In fact, the “Vertical” orbits of L4 and L5 are the
corresponds to the larger bifurcation orbit in Fig. 9. same family of orbits. The magenta family of orbits
L5
L5V M1
L4V
L4
Fig. 13. The L4 “Vertical” orbits: The magenta curves are a representative collection of orbits emanating from L4 which are
very similar to the Vertical orbits that emanate from L1. There are two families of such orbits and the ones shown here (M1)
can be found in the bifurcation diagram between the white arrows marked “L4” and “L4V”. The other family is found in the
bifurcation diagram between the white arrows marked “L5” and “L5V”. The thick magenta orbit is the same bifurcation orbit
shown in Fig. 12.
L5V
L4V
M2
L45
Fig. 14. The magenta curves are a representative collection of orbits between the bifurcation points marked “L4V” and
“L45”. The thick magenta orbit, which encompasses the Earth, is the bifurcation point marked “L45”. This orbit connects to
the “Vertical” orbits of L3 which are not shown here. Continuing on the branch past “L45” one reaches the symmetric image
of “L4V”, namely “L5V”, which connects to the “Vertical” orbits for L5.
transcritically intersects the cyan family of orbits in shown in Fig. 12, and is the first bifurcation point
two places, and there is a third bifurcation point in one encounters on the magenta branch as one moves
between these two (shown as the brown bifurcation away from L4.
point in Fig. 4). Figure 14 shows a representative collection of
Figure 13 shows a representative collection of orbits between the figure-8 bifurcation orbit shown
orbits emanating from L4 which are very similar in Figs. 12 and 13 and the brown bifurcation point
to the Vertical orbits which emanate from L1. The on the magenta family. The thick magenta or-
thick magenta orbit is the same bifurcation orbit as bit, which encompasses the Earth, is a bifurcation
orbit which connects to the “Vertical” orbits of L3, grams similar to that in Fig. 4 for any given value
though this branch of orbits is not shown here. This of µ. Moreover, these automated calculations can
bifurcation point is shown in brown in Fig. 4. Con- be made to follow all branches that emanate from
tinuation of the branch past the brown bifurcation each of the five libration points. Furthermore, one
point gives a symmetry related family of “Vertical” can use numerical continuation of singular points to
orbits for L5. This family consists of the symmetric follow the bifurcation points as µ is allowed to vary.
version of Figs. 13 and 14 reflected through the x–z We do not report these more extensive results in
plane. this paper; however, below we do include a discus-
The family of magenta orbits shown in Fig. 13 sion of the relevance of libration points and their
is the Vertical family emanating from L4; an anal- associated periodic orbits in actual space mission
ogous Vertical family emanates from L5. Zagouras design.
[1985] used Poincaré–Lindstedt series to study the
continuation and bifurcation of the Vertical fami- 3.3. Libration point orbits in the
lies emanating from the equilateral libration points solar system
for the Sun–Jupiter case. Zagouras traced the Ma-
genta family of orbits from L4 to the Vertical family As noted above, the periodic orbits around the equi-
emanating from L5, and identified the bifurcation lateral libration points are significant in the so-
points that we call L4V and L45 shown in Fig. 14. lar system dynamics. In addition, some of the li-
Zagouras identified L45 as a branch point to the bration point orbits discussed above have been or
Vertical family emanating from L3, and numerically could be exploited in space mission design. Due
continued the family branching from L4V which we to the fixed positions of L1 and L2 along the line
between the primaries in the rotating frame, the
call the cyan family. The work in [Zagouras, 1985]
regions around L1 and L2 provide excellent loca-
was extended in [Papadakis & Zagouras, 1992] to
tions for scientific observation spacecraft and for
consider the families of three-dimensional periodic
communication relays [Farquhar, 1968; Farquhar &
orbits for other values of µ.
Dunham, 1990; Farquhar, 2001] analogous to the
For µ less than the critical value µ1 ≈ 0.0385
geostationary communications relays already pro-
there are planar periodic orbits emanating from L4
posed in [Clarke, 1945, 1947].
and L5 [Danby, 1992; Szebehely, 1967]. Also, L4 and
In 1978, the NASA International Sun–Earth
L5 are linearly stable in the CR3BP for µ < µ 1 .
Explorer 3 (ISEE-3) became the first spacecraft
Astronomers have therefore looked to see whether
to orbit the Sun–Earth L1 point, where it traced
natural satellites exist near the equilateral points a Halo orbit [Farquhar, 2001]. The Halo or-
in the solar system. The so-called Trojan asteroids bits for the ISEE-3 mission were computed us-
have been observed near the equilateral points in ing the Lindstedt–Poincaré method, combined with
the Sun–Jupiter and Sun–Mars systems. In addi- Newton’s method, to provide an accurate first ap-
tion, the equilateral points play a role in the in- proximation [Farquhar & Kamel, 1973; Breakwell &
tricate dynamics of Saturn’s moon and ring sys- Brown, 1979; Richardson, 1980].
tem [Murray & Dermott, 1999]. The families of Since that time, the European Space Agency
planar periodic orbits and their bifurcations ema- (ESA)/NASA SOHO mission and the NASA ACE
nating from L4 and L5 have been studied exten- mission have used Halo orbits about the Sun–Earth
sively for subcritical values of µ. See in particular L1 point for solar observations [Rodriguez-Canabal
[Deprit & Henrard, 1968, 1970; Gómez & Noguera, & Hechler, 1989; Farquhar, 2001]. It is known that
1985; Gómez et al., 2001b]. Ragos et al. [1997] ex- the libration points L1 and L2 are linearly un-
amined the planar families around the equilateral li- stable in the CR3BP for all values of µ [Danby,
bration points for values of µ both below and above 1992], and the Halo orbits used for these missions
the critical value µ1 . are themselves unstable. Nevertheless, a spacecraft
Although our numerical results in this section can remain in orbit around the Sun–Earth L1 or
are limited to the Earth–Moon system, we have L2 point for years by performing small maneuvers
also applied the techniques to other values of the every few months [Dunham & Roberts, 2001]. Re-
mass-ratio parameter µ. In fact, these calculations cent libration point mission design has exploited dy-
can be automated by means of scripts [Doedel & namical systems theory to compute stable, unsta-
Paffenroth, 2001] that generate bifurcation dia- ble and center manifolds for libration point orbits
[Howell et al., 1997; Gómez et al., 2001a]. In 2001 as a system of eighteen first order differential equa-
the MAP spacecraft was placed in orbit about the tions. It has seven independent conserved quanti-
Sun–Earth L2 point [Folta & Richon, 1998]. Due to tites, namely, the Hamiltonian, thePthree compo-
mission constraints the MAP mission does not use nents of the linear momentum P = 3i=1 mi x0i , and
an orbit from the Halo family, because the mini- the three components of the angular momentum
L = 3i=1 mi xi ∧ x0i .
P
mum amplitude of the Halo family in the x–y plane
is too large. Instead MAP uses a smaller, quasiperi- Despite its apparent simplicity, it has been no-
odic orbit. The Genesis mission, launched in 2001, toriously difficult to obtain detailed information on
calls for the spacecraft to first orbit the Sun–Earth the global solution structure of Eq. (17). In particu-
L1 point, then follow a heteroclinic connection to an lar, much effort has been devoted to the study of pe-
orbit around L2 before returning to Earth [Koon riodic solutions; see, for example [Hadjidemetriou,
et al., 1999; Koon et al., 2000; Lo et al., 2001]. 1975; Arenstorf, 1976; Meyer, 1981a; Davoust &
Some other scientific missions that are planned to Broucke, 1982; Meyer, 1999; Bruno, 2001]. One well-
orbit the Sun–Earth L2 point are NASA’s Next known periodic solution is the Lagrange equilateral
Generation Space Telescope (NGST) [Folta et al., triangle solution, in which three equal-mass bodies
2001a] and ESA’s FIRST and Planck missions are located at the vertices of an equilateral trian-
[Felici et al., 2001]. In a more elaborate design, the gle that rotates with constant angular velocity; the
Terrestrial Planet Finder (TPF) mission will exploit three bodies thereby describing a circle. Another
the center manifold of a Halo orbit to fly two space- well-known periodic solution is Euler’s collinear so-
crafts in formation along Lissajous orbits about the lution, where two bodies rotate at constant angular
Sun–Earth L2 point [Gómez et al., 1997; Barden & velocity around a circle, while the third body is at
Howell, 1998; Gómez et al., 2001d]. The large Verti- rest at the center.
cal orbits shown in Fig. 7 may also be useful to ob- In a spectacular discovery, Chenciner and
serve the Earth’s polar regions [Folta et al., 2001b]. Montgomery [2000] recently proved the existence of
During NASA’s Apollo program, it was pro- the planar figure-8 periodic solution of the equal-
posed that an orbit about the Earth–Moon L2 point mass 3-body problem shown in Fig. 15. Such a so-
be used for a relay satellite that would allow a lan- lution had earlier been predicted by Moore [1993].
der on the far side of the Moon to communicate The proof uses variational arguments; after certain
with Earth [Farquhar, 1968, 2001]. This innovative
concept was never implemented. Recently a related
study has considered placing spacecraft in orbits
about the Sun–Mars L1 and L2 points to support
communications between Mars landers and Earth
[Strizzi et al., 2001].
4. The General 3-Body Problem

Since the formulation of the law of gravitation by
Newton, the 3-body problem of celestial mechan-
ics has been one of the most studied problems in
physics and mathematics. It is easy enough to state;
in fact, the equations are given by
x1 − x 2 x1 − x 3
x001 = −m2 − m3 ,
|x1 − x2 |3 |x1 − x3 |3
x2 − x 1 x2 − x 3
x002 = −m1 3
− m3 , (17)
|x1 − x2 | |x2 − x3 |3
x3 − x 2 x3 − x 1
x003 = −m2 3
− m1 ,
|x3 − x2 | |x1 − x3 |3
where xi ≡ (xi , yi , zi )∗ , and where mi denotes the Fig. 15. The planar figure-8 orbit [Chenciner & Mont-
mass of the ith body. This equation can be written gomery, 2000; Simó, 2000].
reductions the action integral is minimized over a can be extended, under certain nondegeneracy con-
restricted set of symmetric arcs. The variational ar- ditions, to the case of k linearly independent con-
gument does not provide the stability properties of served quantities (k ≥ 1). If H is the Hamiltonian,
the figure-8 solution. However, Simó [2000] numeri- and if P ≡ (Px , Py , Pz ) (the linear momentum vec-
cally computed this remarkable solution with great tor) and L ≡ (Lx , Ly , Lz ) (the angular momentum
accuracy, and determined elliptic stability; i.e. the vector) are the additional six conserved quantities,
Floquet multipliers of the periodic orbit are on the then the resulting system has the form
unit circle in the complex plane. As will be seen in
u0 = T J∇H + α∇H + β1 ∇Px + β2 ∇Py
Sec. 4.2, elliptic stability is preserved for very small
changes (of the order 10−5 ) in one of the masses. + β3 ∇Pz + γ1 ∇Lx + γ2 ∇Ly + γ3 ∇Lz , (18)
A periodic solution of Eq. (17) has eighteen Flo-
where u(t) ≡ (x, v) ∈ R18 is the state vector, con-
quet multipliers and, as the system is Hamiltonian,
taining nine spatial coordinates and nine velocities,
they appear in pairs (ψ, 1/ψ). Using Proposition 12
and where
of [Muñoz-Almaraz et al., 2002] one can show that
the presence of the seven independent conserved O I
J≡
quantities implies that at least 14 of these multipli- −I 0
ers are located at 1 on the unit circle if the angular is the standard symplectic matrix, with I denoting
momentum L is zero, and at least 12 if L 6= 0. (The the nine-dimensional identity matrix. The unfold-
linear momentum P is always zero along a periodic ing parameters α, and βi , γi , (i = 1, 2, 3), which
orbit.) To study the stability and bifurcations of are treated as unknowns in the solution procedure,
periodic solutions of Eq. (17), one has to monitor are zero at periodic solutions.
the behavior of the remaining “nontrivial” multipli- In full detail, the system (18) can be written as
ers. For the figure-8 orbit we have L = 0, so that
there are four nontrivial multipliers, whose values x01 = T v1 + m1 G(v1 , λ) ,
are given by ψ = exp(±2πiνj ), with ν1 = 0.008422, x02 = T v2 + m2 G(v2 , λ) ,
ν2 = 0.298092. The smallness of ν1 indicates that x03 = T v3 + m3 G(v3 , λ) ,
the corresponding multipliers are very close to a
double +1 situation. Indeed, the numerical results 0 x1 − x 2 x1 − x 3
v1 = −T m2 + m3
in Sec. 4.2 show that there is a bifurcation point as |x1 − x2 |3 |x1 − x3 |3
well as a fold near the figure-8 orbit. + m1 N (x1 , v1 , λ) ,
Simó discovered hundreds of other single-curve (19)
planar periodic solutions for the 3-body problem 0 x2 − x 1 x2 − x 3
v2 = −T m1 + m3
[Simó, 2002], and many more for the case of n |x1 − x2 |3 |x2 − x3 |3
bodies, with n as high as 799! (see [Simó, 2000]). + m2 N (x2 , v2 , λ) ,
Simó called these solutions “choreographies”; the
defining property being that all bodies follow a sin- 0 x3 − x 2 x3 − x 1
v3 = −T m2 + m1
gle closed curve in phase space with a fixed de- |x3 − x2 |3 |x1 − x3 |3
lay. From a historical point of view, the solution + m3 N (x3 , v3 , λ) ,
of Lagrange in 1772 may be considered as the first
“choreography”, and it has taken more than two where the unfolding terms G and N are given by
hundred years for other choreographies to be found! 
0
 
−vz
 
vy

G(v, λ) = γ1  vz  + γ2  0  + γ3  −vx ,
4.1. Computational formulation −vy vx 0
As in Sec. 2.2, we rewrite Eq. (17) as a first or-
β1 0
  
der system with scaled time variable t, so that the
unknown period T appears explicitly. We also sup- N (x, v, λ) = αv+  β2  + γ1  −z  (20)
plement the equations with appropriate unfolding β3 y
terms, applying the procedure outlined in Sec. 2.4,
z −y
   
generalized to the current situation of seven con-
served quantities. In [Muñoz-Almaraz et al., 2003] + γ 2  0  + γ 3  x ,
it is shown that the continuation theorem of Sec. 2.1 −x 0
with (u(0) − û(0))∗ J∇Lz (û(0))
λ ≡ (α, β1 , β2 , β3 , γ1 , γ2 , γ3 ) , λ ∈ R7 . 3
mi [x̂i yi + x̂˙ i ẏi − ŷi xi − ŷ˙ i ẋi ] = 0 .
X
≡
Equation (18) is supplemented with the periodicity i=1
boundary conditions
x1 (0) − x1 (1) = 0 , x2 (0) − x2 (1) = 0 , x3 (0) − x3 (1) = 0 , The computational formulation now consists of
v1 (0) − v1 (1) = 0 , v2 (0) − v2 (1) = 0 , v3 (0) − v3 (1) = 0 . Eqs. (19), (21) and (22); that is, 18 first order
ODEs, subject to 18 periodicity conditions and 7
(21) additional boundary constraints. The unknowns in
Even with fixed masses mi , (i = 1, 2, 3), a solu- each continuation step are the orbit (x(·), v(·)), the
tion of Eq. (19) is not unique, due to the freedom of unfolding parameters λ, and one physical parame-
phase-shift, translation and rotation, and the scal- ter. It follows from the theory in [Muñoz-Almaraz
1
ing invariance x → cx, v → c− 2 v, T → c 2 T .
3 et al., 2003] that the unfolding parameters λ van-
As shown in [Muñoz-Almaraz et al., 2003] (see, in ish along the solution branch, as is also observed in
particular, Theorem 13 and the discussion follow- actual numerical computations.
ing that theorem), the scaling invariance can be re- In our calculations, as described in Secs. 4.2 and
moved by simply fixing the period. In our calcula- 4.3, we use the mass m1 of Body 1, as the free phys-
tions we have fixed the period at 2π. Furthermore, ical parameter; the remaining two masses have the
following the theory in [Muñoz-Almaraz et al., fixed value 1. Numerical continuation, starting from
2003], the remaining seven invariances can be re- the figure-8 orbit of Chenciner and Montgomery,
moved by adding appropriate additional boundary then gives rise to a solution branch along which the
constraints. To be precise, let û(t) ≡ (x̂(t), v̂(t)) computed values of the unfolding vector λ are zero,
denote a reference periodic solution; as in Sec. 2 up to numerical precision. We note that one can
the reference solution is typically the preceding so- also keep all masses fixed at value 1, while allowing
lution in the continuation process. The additional the period T to be the free physical variable. This
boundary conditions are then given by calculation gives the family of figure-8 orbits that
arise from the scaling law; each solution is just a
(u(0) − û(0))∗ J∇H(û(0)) scaled version of the previous one with exactly the
same stability properties.
≡ (u(0) − û(0))∗ û0 (0) = 0 ,
Calculations have been done using the unfold-
(u(0) − û(0))∗ J∇Px (û(0)) ing term (20) and the extra boundary constraints
3
X (22). However, the choice of unfolding terms and
≡ mi (xi − x̂i ) = 0 , corresponding constraints is not unique. One alter-
i=1 native is to supplement the equations with addi-
(u(0) − û(0))∗ J∇Py (û(0)) tional integral constraints, rather than boundary
3
conditions. Although the theory in [Muñoz-Almaraz
X et al., 2003] is based on additional boundary
≡ mi (yi − ŷi ) = 0 ,
constraints (and unfolding parameters), it is pos-
i=1
sible to prove similar results based on integral con-
(u(0) − û(0))∗ J∇Pz (û(0)) (22)
straints (and unfolding parameters). Specifically, we
3
X have done extensive calculations with the integral
≡ mi (zi − ẑi ) = 0 , constraints described below, which basically corre-
i=1 spond to integrated versions of the boundary con-
(u(0) − û(0))∗ J∇Lx (û(0)) straints in Eq. (22).
3 The phase can be fixed by the integral con-
mi [ŷi zi + ŷ˙ i żi − ẑi yi − ẑ˙ i ẏi ] = 0 , straint (5a3 ), which, leaving out the v-component,
X
≡
i=1 can here be written as
(u(0) − û(0))∗ J∇Ly (û(0))
3 3 Z 1
mi [ẑi xi + ẑ˙ i ẋi − x̂i zi − x̂˙ i żi ] = 0 ,
X X
≡ hxi (τ ), x̂0i (τ )idτ = 0 . (23)
i=1 i=1 0
Note that Eq. (23) is a necessary condition for the The integral phase constraint (23) regularizes
relative phase-drift measure the continuation whenever the Poincaré pointwise
3 Z 1 phase condition does [Doedel et al., 1991c]. Similar
results hold for the integral versus boundary con-
X
kxi (τ + σ) − x̂i (τ )k2 dτ ,
i=1 0 ditions for fixing the translation and rotation in-
variances. The integral phase condition often dras-
to be minimized over σ ∈ R [Doedel et al., 1991c].
tically reduces the number of steps required to trace
As before, the reference solution (x̂(·), v̂(·)) is nor-
out a solution branch, especially near homoclinic
mally the preceding solution computed in the con-
orbits and near simple collisions. Properly chosen
tinuation process. Thus, as already mentioned in
integral constraints for fixing translation and ro-
Sec. 2.2, the integral phase condition (23) has the
tation have similar advantages over corresponding
effect of aligning the current solution as much as
boundary constraints.
possible with the previous solution. Often this al-
For periodic orbits near heteroclinic orbits, or
lows much bigger continuation steps to be taken
near multiple collisions, the advantage of integral
along a solution branch.
constraints over pointwise constraints is less clear.
The translation freedom can be removed by
In this case there are multiple small regions (in
requiring
scaled time) where solutions vary rapidly, and these
X3 Z 1 regions may necessarily move relative to each other
mi [xi (τ ) − x̂i (τ )]dτ = 0 . (24) during continuation. This renders impossible the
i=1 0
desirable property of having all mesh point accumu-
Note that (24) consists of three scalar integral lations in fixed locations, as was the case, for exam-
constraints, which correspond to the necessary ple, in Fig. 2(c). Indeed, such calculations typically
conditions for minimizing the relative translation require very small continuation steps.
measure
X3 Z 1
mi kxi (τ ) + s − x̂i (τ )k2 dτ , 4.2. Local continuation of the
i=1 0
figure-8 orbit
over s ∈ R3 . The local continuation of the Chenciner–
The rotational freedom can be eliminated by Montgomery figure-8 solution, as the value of one
adding the integral constraints of the masses is allowed to vary on a very small
X3 Z 1 scale, is represented in Fig. 16(c), where the L 2
mi xi (τ ) × x̂i (τ )dτ = 0 , (25) norm of the solution is plotted versus the mass m 1 .
0
i=1 Solution 1 is the figure-8 orbit, and the solid curve
where we have left out the v-component. Equa- segment containing Solution 1 corresponds to the
tion (25) expresses the fact that the x i are “on the portion of the solution branch where solutions are
average” parallel to x̂i . Written individually, these elliptically stable. The nonunity multipliers of the
constraints are given by the equations figure-8 solution are given by ψj = exp(±2πiνj ),
3 Z 1 with ν1 = 0.008422 and ν2 = 0.298092, which is in
agreement with Simó’s results. All solutions repre-
X
(a) mi {ẑi (τ )yi (τ ) − ŷi (τ )zi (τ )}dτ = 0 ,
i=1 0 sented in Fig. 16(c) have angular momentum L = 0,
and, as discussed earlier, they therefore have four
3
X Z 1 nontrivial multipliers.
(b) mi {x̂i (τ )zi (τ ) − ẑi (τ )xi (τ )}dτ = 0 , For increasing m1 the family reaches a fold bi-
i=1 0 furcation, or limit point (LP), where the branch
turns back towards lower values of m 1 . At the fold
3 Z 1
X the solution loses stability and becomes hyperbolic.
(c) mi {ŷi (τ )xi (τ ) − x̂i (τ )yi (τ )}dτ = 0 ,
0 Continuing past the fold, one arrives at another pe-
i=1
riodic orbit for which all masses are equal to 1,
(26)
namely, Solution 1a. (The “straight line” solution
which fix rotation about the x, y and z axes, re- branch that appears to pass through Solution 1a
spectively, relative to the reference orbit x̂(·). will be described at the end of this section.) The
3
2.5
L2 norm
2.3
2
1
E
2.1 5 4
L
7 6
1.9
0.0 0.2 0.4 0.6 0.8 1.0
m1
(a) (b)
2.1364
1a
L2 norm
2.1363 BP LP
1
2.1362
0.99980 0.99990 1.00000
m1
(c)
Fig. 16. Continuation of the figure-8 orbit. The solution L2 norm is plotted versus the mass of Body 1. The lower figures
are successive blow-ups of the top figure. Solutions at labeled points appear in Figs. 15, 17 and 19. (a) The primary branch
of planar solutions obtained by continuing the figure-8 orbit (Solution 1), and a branch of nonplanar solutions that bifurcates
at Solution 2. The end points of the primary branch, Solutions 3 and 4, are near collision orbits. Solution 5 is a bifurcation
point on the nonplanar solution branch; the bifurcating branch is not shown. Solutions 6 and 7 are further nonplanar solu-
tions. Solution 7 is also a solution of the restricted 3-body problem. (b) A blow-up of the primary solution branch near the
figure-8 orbit. Solution 4 is near a collision orbit. (c) A further blow-up of the primary solution branch near the figure-8 orbit.
Solution 1a is another solution at m1 = 1 on the primary branch; however, it is not a choreography (see Fig. 17). LP and BP
denote a fold and a pitchfork bifurcation; at these points the primary solutions lose stability. The bifurcating pitchfork branch
appears to reconnect to the primary branch at Solution 1a; however, this is not an actual bifurcation point.
unstable Solution 1a is homotopic to the figure-8 Remarkably, within the precision of our calcu-
solution, as one can continuously deform one orbit lations, the action integral of Solution 1a is identical
into the other. However, Solution 1a is hyperbolic to that of the figure-8 orbit. Recall that if x(t) is a
and not a choreography; the three bodies follow trajectory that goes from x1 = x(t1 ) to x2 = x(t2 ),
slightly different figure-8 paths. In fact, Solution 1 then the action, usually denoted by S, is the integral
is the only choreography on the branch. of the Lagrangian over time from t1 to t2 , where the
Lagrangian L is the difference between the kinetic Minimizing orbits are unstable in Hamiltonian sys-
energy K and the potential energy V : tems with two degrees of freedom [Birkhoff, 1927].
Z t2 Z t2 However, for higher-dimensional systems there are
S= L(t)dt = K(x0 (t)) − V (x(t))dt . counter-examples to this statement. Thus it is pos-
t1 t1 sible that there exists a less symmetric solution that
The action depends therefore on the initial and final is the actual minimizer of the action for the equal-
time and on the trajectory. The Principle of Least mass 3-body problem. From the variational view
Action states that the true trajectory, i.e. the one point, if one enlarges the space of arcs over which
that satisfies Newton’s laws, is the minimizer of the the action is minimized, then obviously the action
action. The problem is to find the minimizer among will not increase. The actual minimizing orbit would
all possible trajectories, i.e. minimizing the action be (i) in the same homotopy class as the figure-8
is an alternative to applying Newton’s Laws. (For orbit, and (ii) no longer a choreography. Note that
an excellent exposition see Lecture 19 in Volume II Solution 1a in Fig. 16(c) satisfies these two prop-
of the Feynman Lectures on Physics [Feynman erties. However, we cannot explain why the action
et al., 1964].) Note that our solution algorithm, integral of Solution 1a is, to numerical precision,
which uses a boundary value approach coupled to a identical to that of the figure-8 orbit. Two other
continuation algorithm, does not use the minimiz- well-known periodic solutions of the equal-mass 3-
ing property of the action integral for orbits. We body problem, with normalized period 2π, have
can, of course, easily compute the action integral lower values of the action; namely, the Lagrange so-
as a byproduct of our numerical scheme. For the lution SL = 3π32/3 ≈ 19.60436, which is stable, and
figure-8 orbit its value is found to be S = 24.37197. the Euler solution SE = 3π52/3 /21/3 ≈ 21.87299,
There has been some question about the sta- which is unstable.
bility of the actual minimizer of the action. Appar- Figure 17 shows Solutions 1 and 1a in pro-
ently it is not clear whether the minimizer is elliptic. jected phase space. The behavior of the complex
0.4
0.2
0.4
0.0
0.2 1
y
−0.2
0.0 1
y
−0.4
−0.2
0.4 (a)
−0.4
0.2
0.4
0.0
0.2 1a
y
−0.2
0.0
1a
y
−0.4
−0.2
−1.2 −0.2 0.8
−0.4 x
−1.2 −0.2 0.8
x
(b)
Fig. 17. Projected phase-space representation of Solutions 1 and 1a from Fig. 16.
0.20
0.20
0.10
Log|µι|
0.10
0.00
Log|µι
|
0.00
−0.10
−0.10
−0.20
0.9998 0.9999 1.0000
0.10
−0.20 (a)
0.9998 0.9999 1.0000
0.05
0.10
Arg(µ ι)
0.00
0.05
Arg(µ ι)
−0.05
0.00
−0.10
−0.05
0.9998 0.9999 1.0000
−0.10 m1
0.9998 0.9999 1.0000
m1
(b)
Fig. 18. The behavior of the complex conjugate pair of Floquet multipliers that is closest to 1, for the family of orbits that
passes through the figure-8 orbit. (a) The logarithm of the modulus of the Floquet multipliers. (b) The principal argument of
the Floquet multipliers.
conjugate pair of Floquet multipliers closest to 1, solutions at m1 = 1 on the two legs of the pitch-
as a function of the mass m1 , is shown in Fig. 18. fork branch. In the bifurcation diagram the pitch-
The upper panel is the logarithm of the modulus of fork branch appears to reconnect to the primary
this multiplier and the lower panel is its principal branch at Solution 1a. However, this is an artifact
argument. (The second complex conjugate pair, of our choice of vertical axis and not an actual bifur-
which is not shown, remains complex in the param- cation point. Further continuation of the two legs of
eter range under consideration, it varies very little, the pitchfork branch for m1 > 1 leads to collision
and its argument is outside the scale of Fig. 18(b).) orbits that we do not discuss here.
The elliptic region is restricted to the m 1 -interval
where the multipliers have nonzero argument. Ex-
amining the nonzero values of the logarithm of the 4.3. Global continuation of the
modulus, one can identify two bifurcations in this figure-8 orbit
diagram, namely, the fold bifurcation mentioned We now describe some computational results for the
above, and a pitchfork bifurcation at a value of m 1 continuation of the figure-8 orbit as the mass of
just below 1. Body 1 is allowed to vary on a larger scale, namely,
The pitchfork bifurcation can be seen in between 1 and 0. As discussed above, Fig. 16(c)
Fig. 16(c), where its two legs coincide due to our shows the local continuation of the figure-8 orbit,
choice of vertical axis. There are four distinct solu- for values of m1 very close to 1. Solutions 1 and 1a
tions having m1 = 1 in this diagram, namely, So- lie on the same branch; they are connected via a
lution 1 (the figure-8 orbit), Solution 1a, and the fold. It is of interest to continue these two solutions
(a) (b)
(c) (d)
(e) (f)
Fig. 19. Selected orbits from Fig. 16: (a) Solution 2 (at the bifurcation point to nonplanar solutions). (b) Solution 3 (close to
small body–large bodies collisions). (c) Solution 4 (near collisions of the large bodies). (d) Solution 5 (a nonplanar solution at
a bifurcation point). (e) Solution 6 (another nonplanar solution). (f) Solution 7 (a solution of the restricted 3-body problem).
to smaller values of m1 . The result of this continua- of the restricted 3-body problem, where the orbits
tion is shown in Fig. 16(a). The two lower diagrams remain planar along the entire homotopy path.
correspond to consecutive blow-ups of the top dia- As mentioned before, the figure-8 orbit is el-
gram near m1 = 1. liptically stable, and this stability is preserved in
Solutions at labeled points in Fig. 16 are shown a very small neighborhood of m1 -values. However,
in separate figures: Solution 1 is the figure-8 or- most solutions determined in the continuation pro-
bit of Chenciner and Montgomery, which appears cess are unstable (hyperbolic). Our computational
in Figs. 15 and 17, while Solutions 2–7 are shown scheme does not distinguish between stable and
in Fig. 19. unstable periodic solutions; it can compute both
The branch containing Solutions 1–4 consists solution types equally well. In real astronomical ob-
entirely of planar periodic orbits with zero angular servations, on the other hand, generally only stable
momentum. Solution 2 marks a bifurcation point, periodic orbits are observed. By carefully monitor-
where a solution branch consisting of nonplanar pe- ing the Floquet multipliers along solution branches
riodic orbits with nonzero angular momentum bi- and by following more bifurcating branches, we have
furcates. We have located several more bifurcation located other regions along solution branches where
points, for example, Solution 5, but to keep the pre- elliptically stable periodic orbits exist. However,
sentation simple we only show the branch that bi- some of these regions are very small, so that the
furcates from Solution 2. likelihood of encountering such orbits in physical
Continuation of the figure-8 orbit, i.e. Solu- observations is exceedingly small.
tion 1, in the direction of decreasing m 1 leads to
a collision orbit near Solution 3, where the smaller
mass, i.e. Body 1, collides with the larger bodies. 5. Conclusion
There are four such near-collisions along Solution 3.
We have shown how standard numerical continua-
Continuation of the figure-8 orbit in opposite direc-
tion and bifurcation methods can be used to com-
tion, via the fold and via Solution 1a in Fig. 16(c)
pute families of periodic orbits in conservative sys-
leads to another type of collision orbit near Solu-
tems and determine their bifurcations. We have
tion 4, where the two larger bodies collide with each
applied the computational techniques to the re-
other. There are two such near-collisions along So-
stricted 3-body problem and to the general 3-body
lution 4, while the smaller body completes a near-
problem, and we have presented a limited collec-
horizontal motion.
tion of numerical results that illustrate the power
Solution 5 marks a bifurcation point on the and ease of application of the methods. Systematic
branch of nonplanar solutions; the branch that bi- application of the techniques can lead to new discov-
furcates at this point is not shown. Solutions 6 eries that increase our understanding of the global
and 7 are also on the branch of nonplanar solu- periodic solution structure of these problems.
tions. Note that m1 = 0 at Solution 7; therefore Future work on the restricted 3-body problem
it is also a solution of the restricted 3-body prob- includes a complete classification of the periodic so-
lem. (Note, however, that the mass-ratio parame- lution branches that emanate from all five libra-
ter µ is equal to 0.5 here, while in Sec. 3 we had tion points, and their subsequent bifurcations, for
µ = 0.01215, corresponding to the Earth–Moon sys- all values of the mass-ratio parameter µ. Future
tem.) Our computations therefore establish a ho- work on the general n-body problem includes the
motopy from the figure-8 orbit of Chenciner and extension of the methods in order to deal with
Montgomery to a periodic solution of the restricted the continuation of periodic solutions from the
3-body problem. The complete homotopy path re- Euler solution and the Lagrange solution. An in-
quires branch switching at the bifurcation point de- teresting conjecture is the existence of a homo-
noted as Solution 2. More extensive computations, topy from the figure-8 solution of Chenciner and
that we do not report here, establish the existence Montgomery to the 3-body Lagrange solution. In
of further homotopy paths from the figure-8 orbit fact, it was this conjecture of Professor J. B. Keller
to other solutions of the restricted 3-body problem. that motivated us to design the current compu-
All these paths pass through Solution 2 and switch tational scheme for the continuation of periodic
to the branch of 3D orbits there. We have not found solutions of the 3-body problem. The conjecture
a homotopy from the figure-8 solution to a solution remains open, however.
Supplementary Bibliography Acknowledgments

A more general and detailed theoretical treatment E. J. Doedel, R. C. Paffenroth and H. B. Keller
of our computational methods for conservative sys- thank Professor J. B. Keller (Stanford University)
tems can be found in [Muñoz-Almaraz et al., 2003]. for asking us to continue the figure-8 orbit of
For related theory see [Vanderbauwhede, 1982; Chenciner and Montgomery. E. J. Doedel is in-
Meyer, 1973; Schmidt, 1978; Kirchgraber, 1980] debted to Dr. Neal Hulkower of RAF (Redmond
and, more recently [Sepulchre & MacKay, 1997]. WA) for enjoyable informative conversations and
For recent papers with alternate numerical ap- for access to documents stored in his garage. The
proaches for conservative systems see [Jorba, 1999] work of E. J. Doedel and R. C. Paffenroth has
and [Viswanath, 2001]. been partially supported by NSF grant KDI/NCC
We can only give some pointers to the vast lit- SBR-9873173. E. J. Doedel is also supported by
erature on the general n-body problem, the CR3BP, NSERC Canada, Research Grant A4274. D. J.
and related topics. Les Méthodes Nouvelles de Dichmann thanks R. W. Farquhar and K. C.
la Mécanique Céleste of Poincaré [1899] (English Howell for introducing him to the fascinating sub-
translation [1993]) is a seminal work on the CR3BP ject of libration point mission design. J. Galán-
which developed fundamental concepts in dynami- Vioque is indebted to E. Freire, F. J. Muñoz-
cal systems. Orbital mechanics texts that examine Almaraz, J. Llibre, R. Montgomery and A. Zevin
the n-body problem and the CR3BP in particular, for helpful discussions. J. Galán-Vioque acknowl-
include [Roy, 1988] and [Danby, 1992]. Meyer and edges funding from NATO (CRG-972260), DGES
Hall [1999] considered the n-body problem and the PB-1152 and PB-0142. The authors thank David
CR3BP from the point of view of Hamiltonian sys- Goldstein at the Aerospace Corporation for valu-
tems theory, while Meyer [1999] focused on peri- able comments. They also gratefully acknowledge
odic solutions. For books on the 3-body problem the artistic assistance of Christine Thorn in the
see [Szebehely, 1967] and [Bruno, 1994], both of preparation of the figures.
which focus on the planar problem, and [Marchal,
1990].
The use of libration points for space mis- References
sions was already mentioned by Clarke [1947]. The Allgower, E. L. & Georg, K. [1996] “Numerical path fol-
January–March 2001 issue of the Journal of the lowing,” in Handbook of Numerical Analysis, Vol. 5,
Astronautical Sciences contains many articles on eds. Ciarlet, P. G. & Lions, J. L. (North Holland
libration point missions, including several already Publishing).
cited here. Other references on orbits near libra- Arenstorf, R. F. [1976] “Some families of periodic solu-
tion points can be found in [Bray & Goudas, 1967a, tions in the planar n-body problem,” Celest. Mech.
1967b; Meer, 1985], and [Gabern & Jorba, 2001]. 14, 5–9.
Corresponding mission design aspects are consid- Ascher, U. M., Christiansen, J. & Russell, R. D. [1981]
“Collocation software for boundary value ODEs,”
ered in [Gómez et al., 1991] and [Gómez et al., 1993].
ACM Trans. Math. Software 7, 209–222.
For additional literature on 3D orbits near the Ascher, U. M., Mattheij, R. M. M. & Russell, R. D. [1995]
libration points see [Goudas, 1963; Kazantzis, 1978, Numerical Solution of Boundary Value Problems for
1979a, 1979b; Robin & Markellos, 1980], and [Jorba Ordinary Differential Equations (SIAM).
& Masdemont, 1999]. Recent results on verti- Ascher, U. M. & Spiteri, R. J. [1995] “Collocation soft-
cal orbits can be found in [Jorba & Villanueva, ware for boundary value differential-algebraic equa-
1998], and for collisions see [Simó, 2001a]. Some tions,” SIAM J. Sci. Comput. 15, 938–952.
other relevant papers on the CR3BP include Ascher, U. M. & Petzold, L. R. [1998] Computer Methods
[Michalodimitrakis, 1978; Kwok, 1980; Meyer, for ODEs and DAEs (SIAM).
1981a], and [Meyer & Schmidt, 2000]. Barden, B. T. & Howell, K. C. [1998] “Fundamental mo-
For more papers on the figure-8 orbit see tions near collinear libration points and their transi-
tions,” J. Astronaut. Sci. 46, 361–378.
[Chenciner et al., 2001] and [Simó, 2001b]. Fur-
Beyn, W.-J., Champneys, A., Doedel, E. J., Govaerts,
ther pointers to the general n-body literature in- W., Sandstede, B. & Kuznetov, Y. A. [2002] “Numer-
clude [Broucke, 1978; Hulkower, 1980; Gómez & ical continuation and computation of normal forms,”
Llibre, 1981; Montgomery, 1997], and [Stuchi et al., in Handbook of Dynamical Systems, Vol. 2, ed. Fiedler,
2000]. B. (Elsevier Science), pp. 149–219.
Birkhoff, G. D. [1927] Dynamical Systems (Amer. Math. Doedel, E. J., Jepson, A. D. & Keller, H. B. [1984]
Soc.) “Numerical methods for Hopf bifurcation and con-
Bray, T. A. & Goudas, C. L. [1967a] “Doubly symmet- tinuation of periodic solution paths,” in Computing
ric orbits about the collinear Lagrangian points,” The Methods in Applied Sciences and Engineering VI,
Astron. J. 72, 202–213. eds. Glowinski, R. & Lions, J. L. (North Holland,
Bray, T. A. & Goudas, C. L. [1967b] “Three-dimensional Amsterdam), pp. 127–136.
periodic oscillations about L1, L2 and L3,” in Ad- Doedel, E. J., Aronson, D. G. & Othmer, H. G. [1991a]
vances in Astronomy and Astrophysics, Vol. 5, ed. “The dynamics of coupled current-biased Joseph-
Kopal, Z. (Academic Press), pp. 71–130. son junctions II,” Int. J. Bifurcation and Chaos 1,
Breakwell, J. V. & Brown, J. V. [1979] “The Halo fam- 51–66.
ily of 3-dimensional periodic orbits in the Earth– Doedel, E. J., Keller, H. B. & Kernévez, J. P. [1991b]
Moon restricted 3-body problem,” Celest. Mech. 20, “Numerical analysis and control of bifurcation prob-
389–404. lems: I,” Int. J. Bifurcation and Chaos 1, 493–520.
Broucke, R. [1978] “On the isosceles triangle configura- Doedel, E. J., Keller, H. B. & Kernévez, J. P. [1991c]
tion in the planar general three-body problem,” As- “Numerical analysis and control of bifurcation prob-
tron. Astrophys. 73, 303–313. lems: II,” Int. J. Bifurcation and Chaos 1, 745–772.
Bruno, A. D. [1994] The Restricted 3-Body Problem: Doedel, E. J. [1997] “Nonlinear numerics,” Int. J. Bifur-
Plane Periodic Orbits (De Gruyter, Berlin, NY). cation and Chaos 7, 2127–2143.
Bruno, A. D. [2001] “Families of periodic solutions of the Doedel, E. J., Champneys,
Beletskii equation,” Technical Report, Keldysh Insti- A. R., Fairgrieve, T. F., Kuznetsov, Y. A., Sandst-
tute of Applied Mathematics of RAS, Moscow. ede, B. & Wang, X. J. [1997] “Continuation and bi-
Chenciner, A. & Montgomery, R. [2000] “A remarkable furcation software for ordinary differential equations,”
periodic solution of the three body problem in the Available via http://cmvl.cs.concordia.ca.
case of equal masses,” Ann. Math. 152, 881–901. Doedel, E. J., Paffenroth, R. C., Champneys, A. R.,
Chenciner, A., Gerver, J., Montgomery, R. & Simó, C. Fairgrieve, T. F., Kuznetsov, Y. A., Oldeman,
[2001] “Simple choreographic motions of n bodies: A B. E., Sandstede, B. & Wang, X. J. [2000]
preliminary study,” Technical Report, Departament “AUTO2000: Continuation and bifurcation software
de Matemàtica i Anàlisi, Universitat de Barcelona. for ordinary differential equations,” Available via
Clarke, A. C. [1945] “Extra-terrestial relays,” Wire- http://cmvl.cs.concordia.ca.
less World, pp. 305–308; Reprinted [1984] Ascent Doedel, E. J. & Paffenroth, R. C. [2001] “The
to Orbit: A Scientific Autobiography, Clarke, A. C. AUTO2000 command line user interface,” Proc. 9th
(Wiley). Int. Python Conf., Long Beach, pp. 233–241.
Clarke, A. C. [1947] “Stationary orbits,” J. British As- Dunham, D. W. & Roberts, C. E. [2001] “Stationkeep-
tron. Assoc., pp. 232–237; Reprinted [1984] Ascent ing techniques for libration-point satellites,” J. Astro-
to Orbit: A Scientific Autobiography, Clarke, A. C. naut. Sci. 49, 127–144.
(Wiley). Fairgrieve, T. F. & Jepson, A. D. [1991] “O. K.
Danby, J. M. A. [1992] Fundamentals of Celestial Me- Floquet multipliers,” SIAM J. Numer. Anal. 28,
chanics (Willmann-Bell). 1446–1462.
Davoust, E. & Broucke, R. [1982] “On the manifold of pe- Farquhar, R. [1968] The Control and Use of Libration-
riodic orbits in the planar general three-body problem Point Satellites, Ph.D. thesis, Department of Aero-
with equal masses,” Astron. Astrophys. 112, 305–320. nautics and Astronautics, Stanford University.
de Boor, C. & Swartz, B. [1973] “Collocation at Gaussian Farquhar, R. W. & Kamel, A. K. [1973] “Quasi-periodic
points,” SIAM J. Numer. Anal. 10, 582–606. orbits about the translunar libration point,” Celest.
Deprit, A. & Henrard, J. [1968] “A manifold of periodic Mech. 7, 458–473.
orbits,” in Advances in Astronomy and Astrophysics, Farquhar, R. W. & Dunham, D. W. [1990] “Use of libra-
ed. Kopal, Z. (Academic Press), pp. 1–124. tion points for space observatories,” in Observatories
Deprit, A. & Henrard, J. [1969] “Construction of or- in Earth Orbit and Beyond (Kluwer Academic Pub-
bits asymptotic to a periodic orbit,” Astron. J. 74, lishers), pp. 391–395.
308–316. Farquhar, R. W. [2001] “The flight of ISEE-3/ICE: Ori-
Deprit, A. & Henrard, J. [1970] The “Trojan manifold — gins, mission history and a legacy,” J. Astronaut. Sci.
Survey and conjectures,” in Periodic Orbits, Stability 49, 23–73.
and Resonances, ed. Giacaglia, G. E. O. (D. Reidel Felici, F., Hechler, M. & Vanderbussche, F. [2001] “The
Publishing Co., Dordrecht-Holland), pp. 1–18. ESA astronomy missions at L2: FIRST and Planck,”
Doedel, E. J. [1981] “AUTO, a program for the auto- J. Astronaut. Sci. 49, 185–196.
matic bifurcation analysis of autonomous systems,” Feynman, R. P., Leighton, R. B. & Sands, M. [1964] The
Congr. Numer. 30, 265–384. Feynman Lectures on Physics (Addison-Wesley).
Folta, D. & Richon, K. [1998] “Libration orbit mission Henderson, M. [2002] “Multiple parameter continuation:
design at L2: A MAP and NGST perspective,” in Computing implicitly defined k-manifolds,” Int. J. Bi-
AIAA/AAS Astrodynamics Specialist Conf., AIAA furcation and Chaos 12, 451–476.
98-4469. Hénon, M. [1997] Generating Families in the Restricted
Folta, D., Cooley, S. & Howell, K. [2001a] “Trajectory Problem (Springer-Verlag).
design strategies for the NGST L2 libration point mis- Hénon, M. [2001] Generating Families in the Re-
sion,” in AAS/AIAA Space Flight Mechanics Meeting, stricted Problem, II, Quantitative Study of Bifurca-
AAS 01-205. tions (Springer-Verlag).
Folta, D., Young, C. & Ross, A. [2001b] “Unique non- Howell, K. C. [1984] “Three-dimensional, periodic, Halo
Keplerian orbit vantage locations for Sun–Earth con- orbits,” Celest. Mech. 32, 53–71.
nections and Earth Science Vision roadmaps,” in Howell, K. C. & Farquhar, R. W. [1993] “John Break-
NASA Goddard Flight Dynamics Symp. well, the restricted problem, and Halo orbits,” Acta
Gabern, F. & Jorba, A. [2001] “A restricted four-body Astronautica 29, 485–488.
model for the dynamics near the Lagrangian points of Howell, K. C., Barden, B. T. & Lo, M. W. [1997] “Ap-
the Sun–Jupiter system,” Discr. Contin. Dyn. Syst.- plications of dynamical systems theory to trajectory
Series B 1, 143–182. design for a libration point mission,” J. Astronaut.
Gómez, G. & Llibre, J. [1981] “A note on a conjecture Sci. 45, 161–178.
by Poincaré,” Celest. Mech. 24, 335–343. Howell, K. C. & Campbell, E. T. [1999] “Three-
Gómez, G. & Noguera, M. [1985] “Some manifolds of dimensional periodic solutions that bifurcate from
periodic orbits in the restricted three-body problem,” Halo families in the circular restricted three-body
Celest. Mech. 35, 235–255. problem,” in Spaceflight Mechanics, AAS 99-161.
Gómez, G., Jorba, A., Masdemont, J. & Simó, C. [1991] Howell, K. C. [2001] “Families of orbits in the vicinity of
“Moon’s influence on the transfer from the Earth to the collinear libration points,” J. Astronaut. Sci. 49,
a Halo orbit around L1,” in Predictability, Stability, 107–125.
and Chaos in N-Body Dynamical Systems, ed. Roy, Hulkower, N. D. [1980] “Central configurations and
A. E. (Plenum Press), pp. 283–290. hyperbolic-elliptic motion in the three-body prob-
Gómez, G., Jorba, A., Masdemont, J. & Simó, C. [1993] lem,” Celest. Mech. 21, 37–41.
“Study of the transfer from the Earth to a Halo orbit Ichtiaroglou, S. & Michalodimitrakis, M. [1980] “Three-
around the equilibrium point L1,” Celest. Mech. Dyn. body problem: The existence of families of three-
Astron. 56, 541–562. dimensional periodic orbits,” Astron. Astrophys. 81,
Gómez, G., Masdemont, J. & Simó, C. [1997] 30–32.
“Lissajous orbits around Halo orbits,” in AAS/AIAA Jepson, A. D. [1981] Numerical Hopf Bifurcation, Ph.D.
Space Flight Mechanics Meeting, AAS 97-106. thesis, Applied Mathematics, California Institute of
Gómez, G., Koon, K. S., Lo, M. W., Marsden, J. E., Technology.
Masdemont, J. & Ross, S. D. [2001a] “Invariant man- Jorba, A. & Villanueva, J. [1998] “Stability and asymp-
ifolds, the spatial three-body problem and space mis- totic behavior of the vertical family of periodic or-
sion design,” in AAS/AIAA Astrodynamics Specialist bits around L5 of the restricted three body problem,”
Conf., AAS 01-301. in Actas del XV CEDYA/V CMA (Universidad de
Gómez, G., Llibre, J., Martı́nez, R. & Simó, C. [2001b] Vigo), pp. 303–308.
Dynamics and Mission Design Near Libration Points, Jorba, A. [1999] “A methodology for the numerical com-
Vol. II: Fundamentals: The Case of Triangular Libra- putation of normal forms, centre manifolds, and first
tion Points, Vol. 3 (World Scientific, Singapore). integrals of Hamiltonian systems,” Experim. Math. 8,
Gómez, G., Llibre, J., Martı́nez, R. & Simó, C. 155–195.
[2001c] Dynamics and Mission Design Near Libra- Jorba, A. & Masdemont, J. [1999] “Dynamics in the cen-
tion Points, Volume I: Fundamentals: The Case of ter manifold of the collinear points of the restricted
Collinear Libration Points, Vol. 2 (World Scientific, three body problem,” Physica D132, 189–213.
Singapore). Kazantzis, P. G. [1978] “The structure of peri-
Gómez, G., Lo, M. W., Masdemont, J. & Museth, K. odic solutions in the restricted problem of three-
[2001d] “Simulation of formation flight near Lagrange body: Sun–Jupiter case,” Astrophys. Space Sci. 59,
points for the TPF mission,” in AAS/AIAA Astrody- 355–371.
namics Specialist Conf., AAS 01-305. Kazantzis, P. G. [1979a] “Families of three-dimensional
Goudas, C. L. [1963] “Three-dimensional periodic orbits axisymmetric periodic orbits in the restricted three-
and their stability,” Icarus 2, 1–18. body problem: Sun–Jupiter case,” Astrophys. Space
Hadjidemetriou, J. [1975] “The continuation of periodic Sci. 61, 477–486.
orbits from the restricted to the general three-body Kazantzis, P. G. [1979b] “Families of three-dimensional
problem,” Celest. Mech. 12, 155–174. plane-symmetric periodic orbits in the restricted
three-body problem: Sun–Jupiter case,” Astrophys. Muñoz-Almaraz, F. J., Freire, E., Galán, J., Doedel,
Space Sci. 61, 357–367. E. J. & Vanderbauwhede, A. [2003] “Continuation of
Keller, H. B. [1977] “Numerical solution of bifurca- periodic orbits in conservative and Hamiltonian
tion and nonlinear eigenvalue problems,” in Appli- systems,” Physica D, to appear.
cations of Bifurcation Theory, ed. Rabinowitz, P. H. Murray, C. D. & Dermott, S. F. [1999] Solar System
(Academic Press), pp. 359–384. Dynamics (Cambridge University Press).
Kirchgraber, U. [1980] “A note on Liapunov’s center Papadakis, K. E. & Zagouras, C. G. [1992] “Bifurcation
theorem,” J. Math. Anal. Appl. 73, 568–570. points and intersection of families of periodic orbits
Koon, W. S., Lo, M. W., Marsden, J. E. & Ross, S. D. in the three-dimensional restricted three-body prob-
[1999] “The Genesis trajectory and heteroclinic con- lem,” Astrophys. Space Sci. 199, 241–256.
nections,” in AAS/AIAA Astrodynamics Specialist Poincaré, H. [1899] Les Méthodes Nouvelles de la
Conf., AAS 99-451. Mécanique Céleste (Gauthier-Villars).
Koon, W. S., Lo, M. W., Marsden, J. E. & Ross, S. D. Poincaré, H. [1993] New Methods of Celestial Mechanics,
[2000] “Heteroclinic connections between periodic or- Vols. 1–3 (American Institute of Physics) ed. Goroff,
bits and resonance transitions in celestial mechanics,” D. L.
Chaos 10, 427–469. Ragos, O., Papadakis, K. E. & Zagouras, C. G. [1997]
Kuznetsov, Y. A. [1998] Elements of Applied Bifurcation “Stability regions and quasi-periodic motion in the
Theory (Springer Verlag), 2nd edition. vicinity of triangular equilibrium points,” Celest.
Kwok, J. H. [1980] “Periodic orbits of the asteroidal Mech. Dyn. Astron. 67, 251–274.
type in the circular restricted three-body problem,” in Rheinboldt, W. C. [1986] Numerical Analysis of
AIAA/AAS Astrodynamics Specialist Conf., AIAA- Parametrized Nonlinear Equations, University of
80-1654. Arkansas, Lecture Notes in the Mathematical Sciences
Lo, M. W., Williams, B., Bollman, W., Han, D., Hahn, (Wiley-Interscience).
Y., Bell, J., Hirst, E., Corwin, R., Hong, P., Howell, Richardson, D. L. [1980] “Halo orbit formulation for the
K., Barden, B. & Wilson, R. [2001] “Genesis mission ISEE-3 mission,” J. Guid. Contr. 3, 543–548.
design,” J. Astronaut. Sci. 49, 169–184. Robin, I. A. & Markellos, V. V. [1980] “Numerical de-
Lust, K. [2001] “Improved numerical Floquet multipli- termination of the three-dimensional periodic orbits
ers,” Int. J. Bifurcation and Chaos 11, 2389–2410. generated from vertical self-resonant satellite orbits,”
Marchal, C. [1990] The Three Body Problem (Elsevier). Celest. Mech. 21, 395–434.
Markellos, V. [1981] “Bifurcations of planar to three- Rodriguez-Canabal, J. & Hechler, M. [1989] “Or-
dimensional periodic orbits in the general three-body bital aspects of the SOHO mission design,” in Or-
problem,” Celest. Mech. 25, 3–31. bital Mechanics and Mission Design, AAS 89-171,
Meer, J.-C. v. d. [1985] The Hamiltonian Hopf Bifurca- pp. 347–356.
tion (Springer-Verlag). Roy, A. E. [1988] Orbital Motion (Adam Hilger).
Meyer, K. R. [1973] “Symmetries and integrals in me- Russell, R. D. & Christiansen, J. [1978] “Adaptive mesh
chanics,” in Dynamical Systems, ed. Peixoto, M. selection strategies for solving boundary value prob-
(Academic Press), pp. 259–272. lems,” SIAM J. Numer. Anal. 15, 59–80.
Meyer, K. R. [1981a] “Periodic orbits near infinity in the Schmidt, D. S. [1978] “Hopf’s bifurcation theorem and
restricted n-body problem,” Celest. Mech. 23, 69–81. the center theorem of Liapunov with resonance cases,”
Meyer, K. R. [1981b] “Periodic solutions of the n-body J. Math. Anal. Appl. 63, 354–370.
problem,” J. Diff. Eqs. 39, 2–38. Sepulchre, J. A. & MacKay, R. S. [1997] “Localized os-
Meyer, K. R. [1999] Periodic Solutions of the N-Body cillations in conservative or dissipative networks of
Problem (Springer-Verlag). weakly coupled autonomous oscillators,” Nonlinear-
Meyer, K. R. & Hall, G. R. [1999] Introduction to Hamil- ity 10, 679–713.
tonian Dynamical Systems and the N-body problem Seydel, R. [1995] From Equilibrium to Chaos. Practi-
(Springer-Verlag). cal Bifurcation and Stability Analysis, 2nd edition
Meyer, K. R. & Schmidt, D. S. [2000] “From the re- (Springer-Verlag, NY).
stricted to the full three-body problem,” Trans. AMS Simó, C. [1996] “Effective computations in Hamilto-
352, 2283–2299. nian dynamics,” in Cent ans après les Méthodes Nou-
Michalodimitrakis, M. [1978] “A new type of connection velles H. Poincaré (Société Mathématique de France),
between the families of periodic orbits of the restricted pp. 1–23.
problem,” Astron. Astrophys. 64, 83–86. Simó, C. [2000] “New families of solutions in n-
Montgomery, R. [1997] The 3-Body Problem, Lecture body problems,” Proc. ECM 2000, Barcelona
Notes, CIMAT, Guanajuato, Mexico. http://mat.uab.es/art3ecm/simo.pdf, Vol. I, Progress
Moore, C. [1993] “Braids in classical gravity,” Phys. Rev. in Mathematics, Vol. 201 (Birkhäuser Basel),
Lett. 70, 3675–3679. pp. 101–115.
Simó, C. [2001a] “Consecutive quasi-collisions in the pla- Vanderbauwhede, A. [1982] “Families of periodic orbits
nar circular RTBP,” Technical Report, Departament for autonomous systems,” in Dynamical System II,
de Matemàtica i Anàlisi, Universitat de Barcelona. eds. Bednarek, A. & Cesari, L. (Academic Press),
Simó, C. [2001b] “Periodic orbits of the planar n- pp. 427–446.
body problem with equal masses and all bodies on Viswanath, D. [2001] “The Lindstedt–Poincaré tech-
the same path,” Technical Report, Departament de nique as an algorithm for computing periodic orbits,”
Matemàtica i Anàlisi, Universitat de Barcelona. SIAM Rev. 43, 478–495.
Simó, C. [2002] “Dynamical properties of the figure eight Zagouras, C. G. & Kazantzis, P. G. [1979] “Three-
solution of the three-body problem,” Contemp. Math. dimensional periodic oscillations generating from
292, 209–228. plane periodic ones around the collinear Lagrangian
Strizzi, J. D., Kutrieb, J. M., Damphousse, P. E. & points,” Astrophys. Space Sci. 61, 389–409.
Carrico, J. P. [2001] “Sun–Mars libration points Zagouras, C. G. [1985] “Three-dimensional periodic or-
and Mars mission simulations,” in AAS/AIAA Space bits about the triangular equilibrium points of the
Flight Mechanics Meeting, AAS 01-159. restricted problem of three bodies,” Celest. Mech.
Stuchi, T. J., Antunes, A. C. B. & Andreu, M. A. [2000] Dyn. Astron. 37, 27–46.
“Muonic helium atom as a classical three-body prob- Zufirı́a, J. A. [1987] Ph.D. thesis, Applied Mathematics,
lem,” Phys. Rev. E62, 7831–7841. California Institute of Technology.
Szebehely, V. [1967] Theory of Orbits: The Restricted
Problem of Three Bodies (Academic Press).

Computation of Periodic Solutions of Conservative Systems With Application To The 3-Body Problem

Uploaded by

Copyright:

Available Formats

Computation of Periodic Solutions of Conservative Systems With Application To The 3-Body Problem

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Computation of Periodic Solutions of Conservative Systems With Application To The 3-Body Problem

Uploaded by

Copyright:

Available Formats

Tutorials and Reviews

COMPUTATION OF PERIODIC SOLUTIONS OF

E. J. DOEDEL∗ , R. C. PAFFENROTH and H. B. KELLER

Received May 24, 2002; Revised August 22, 2002

Keywords: Numerical bifurcation; numerical continuation; conservative systems; periodic solu-

This equation must be solved for X1 = (x1 (·), T1 ,

4. The General 3-Body Problem

with (u(0) − û(0))∗ J∇Lz (û(0))

Supplementary Bibliography Acknowledgments

You might also like