Perturbation Theory in Celestial Mechanics

Alessandra Celletti
Dipartimento di Matematica
Università di Roma Tor Vergata
Via della Ricerca Scientifica 1, I-00133 Roma (Italy)
(celletti@[Link])

December 8, 2007

Contents
1 Glossary 2

2 Definition 2

3 Introduction 2

4 Classical perturbation theory 4

4.1 The classical theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4.2 The precession of the perihelion of Mercury . . . . . . . . . . . . . . . . . . . . 6
4.2.1 Delaunay action–angle variables . . . . . . . . . . . . . . . . . . . . . . 6
4.2.2 The restricted, planar, circular, three–body problem . . . . . . . . . . . 7
4.2.3 Expansion of the perturbing function . . . . . . . . . . . . . . . . . . . . 7
4.2.4 Computation of the precession of the perihelion . . . . . . . . . . . . . . 8

5 Resonant perturbation theory 9

5.1 The resonant theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
5.2 Three–body resonance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
5.3 Degenerate perturbation theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5.4 The precession of the equinoxes . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

6 Invariant tori 14
6.1 Invariant KAM surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
6.2 Rotational tori for the spin–orbit problem . . . . . . . . . . . . . . . . . . . . . 15
6.3 Librational tori for the spin–orbit problem . . . . . . . . . . . . . . . . . . . . . 16
6.4 Rotational tori for the restricted three–body problem . . . . . . . . . . . . . . . 17
6.5 Planetary problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

7 Periodic orbits 18
7.1 Construction of periodic orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7.2 The libration in longitude of the Moon . . . . . . . . . . . . . . . . . . . . . . . 20

1
8 Future directions 20

9 Bibliography 21
9.1 Books and Reviews . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
9.2 Primary Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

1 Glossary

KAM theory: it provides the persistence of quasi–periodic motions under a small perturbation
of an integrable system. KAM theory can be applied under quite general assumptions, i.e. a non–
degeneracy of the integrable system and a diophantine condition of the frequency of motion.
It yields a constructive algorithm to evaluate the strength of the perturbation ensuring the
existence of invariant tori.
Perturbation theory: it provides an approximate solution of the equations of motion of a
nearly–integrable system.
Spin–orbit problem: a model composed by a rigid satellite rotating about an internal axis
and orbiting around a central point–mass planet; a spin–orbit resonance means that the ratio
between the revolutional and rotational periods is rational.
Three–body problem: a system composed by three celestial bodies (e.g. Sun–planet–satellite)
assumed as point–masses subject to the mutual gravitational attraction. The restricted three–
body problem assumes that the mass of one of the bodies is so small that it can be neglected.

2 Definition
Perturbation theory aims to find an approximate solution of nearly–integrable systems, namely
systems which are composed by an integrable part and by a small perturbation. The key point
of perturbation theory is the construction of a suitable canonical transformation which removes
the perturbation to higher orders. A typical example of a nearly–integrable system is provided
by a two–body model perturbed by the gravitational influence of a third body whose mass is
much smaller than the mass of the central body. Indeed, the solution of the three–body problem
greatly stimulated the development of perturbation theories. The solar system dynamics has
always been a testing ground for such theories, whose applications range from the computation
of the ephemerides of natural bodies to the development of the trajectories of artificial satellites.

3 Introduction
The two–body problem can be solved by means of Kepler’s laws, according to which for negative
energies the point–mass planets move on ellipses with the Sun located in one of the two foci. The
dynamics becomes extremely complicated when adding the gravitational influence of another
body. Indeed Poincaré showed ([12]) that the three–body problem does not admit a sufficient
number of prime integrals which allow to integrate the problem. Nevertheless a special attention
deserves the so–called restricted three–body problem, namely when the mass of one of the three
bodies is so small that its influence on the others can be neglected. In this case one can assume

2
that the primaries move on Keplerian ellipses around their common barycenter; if the mass
of one of the primaries is much larger than the other (as it is the case in any Sun–planet
sample), the motion of the minor body is governed by nearly–integrable equations, where the
integrable part represents the interaction with the major body, while the perturbation is due
to the influence of the other primary. A typical example is provided by the motion of an
asteroid under the gravitational attraction of the Sun and Jupiter. The small body may be
taken not to influence the motion of the primaries, which are assumed to move on elliptic
trajectories. The dynamics of the asteroid is essentially driven by the Sun and perturbed by
Jupiter, since the Jupiter–Sun mass–ratio amounts to about 10−3 . The solution of this kind
of problem stimulated the work of the scientists, especially in the XVIII and XIX centuries.
Indeed, Lagrange, Laplace, Leverrier, Delaunay, Tisserand and Poincaré developed perturbation
theories which are at the basis of the study of the dynamics of celestial bodies, from the
computation of the ephemerides to the recent advances in flight dynamics. For example, on
the basis of perturbation theory Delaunay ([8]) developed a theory of the Moon, providing
very refined ephemerides. Celestial Mechanics greatly motivated the advances of perturbation
theories as witnessed by the discovery of Neptune: its position was theoretically predicted by
John Adams and by Jean Urbain Leverrier on the basis of perturbative computations; following
the suggestion provided by the theoretical investigations, Neptune was finally discovered on 23
September 1846 by the astronomer Johann Gottfried Galle.
The aim of perturbation theory is to implement a canonical transformation which allows to find
the solution of a nearly–integrable system within a better degree of approximation (see section 4
and references [2], [4], [9], [10], [11], [13], [14]). Let us denote the frequency vector of the system
by ω (see [Hamiltonian Normal Forms?], [KAM theory?]), which we assume to belong to Rn ,
where n is the number of degrees of freedom of the system. Classical perturbation theory can
be implemented provided that the frequency vector satisfies a non–resonant relation, which
means that there do not exist a vector m ∈ Zn such that ω · m ≡ nj=1 ωj mj = 0. In case there
P

exists such commensurability condition, a resonant perturbation theory can be developed as

outlined in section 5. In general, the three–body problem (and, more extensively, the N –body
problem) is described by a degenerate Hamiltonian system, which means that the integrable
part (i.e., the Keplerian approximation) depends on a subset of the action variables. In this
case a degenerate perturbation theory must be implemented as explained in section 5.3. For all
the above perturbation theories (classical, resonant and degenerate) an application to Celestial
Mechanics is given: the precession of the perihelion of Mercury, orbital resonances within a
three–body framework, the precession of the equinoxes.
Even if the non–resonance condition is satisfied, the quantity ω·m can become arbitrarily small,
giving rise to the so–called small divisor problem; indeed, these terms appear at the denominator
of the series defining the canonical transformations necessary to implement perturbation theory
and therefore they might prevent the convergence of the series. In order to overcome the small
divisor problem, a breakthrough came with the work of Kolmogorov ([31]), later extended to
different mathematical settings by Arnold ([16]) and Moser ([37]). The overall theory is known
with the acronym of KAM theory. As far as concrete estimates on the allowed size of the
perturbation are concerned, the original versions of the theory gave discouraging results, which
were extremely far from the physical measurements of the parameters involved in the proof.
Nevertheless the implementation of computer–assisted KAM proofs allowed to obtain results
which are in good agreement with the reality. Concrete estimates with applications to Celestial

3
Mechanics are reported in section 6.
In the framework of nearly–integrable systems a very important role is provided by periodic
orbits, which might be used to approximate the dynamics of quasi–periodic trajectories; for
example, a truncation of the continued fraction expansion of an irrational frequency provides a
sequence of rational numbers, which are associated to periodic orbits eventually approximating
a quasi–periodic torus. A classical computation of periodic orbits using a perturbative approach
is provided in section 7, where an application to the determination of the libration in longitude
of the Moon is reported.

4 Classical perturbation theory

4.1 The classical theory
Consider a nearly–integrable Hamiltonian function of the form

H(I, ϕ) = h(I) + εf (I, ϕ) , (1)

where h and f are analytic functions of I ∈ V (V open set of Rn ) and ϕ ∈ Tn (Tn is the
standard n–dimensional torus), while ε > 0 is a small parameter which measures the strength
of the perturbation. The aim of perturbation theory is to construct a canonical transformation,
which allows to remove the perturbation to higher orders in the perturbing parameter. To this
end, let us look for a canonical change of variables (i.e., with symplectic Jacobian matrix)
C : (I, ϕ) → (I 0 , ϕ0 ), such that the Hamiltonian (1) takes the form

H 0 (I 0 , ϕ0 ) = H ◦ C(I, ϕ) ≡ h0 (I 0 ) + ε2 f 0 (I 0 , ϕ0 ) , (2)

where h0 and f 0 denote the new unperturbed Hamiltonian and the new perturbing function.
To achieve such result we need to proceed along the following steps: build a suitable canon-
ical transformation close to the identity, perform a Taylor series expansion in the perturbing
parameter, require that the unknown transformation removes the dependence on the angle
variables up to second order terms, expand in Fourier series in order to get an explicit form of
the canonical transformation.
The change of variables is defined by the equations
∂Φ(I 0 , ϕ)
I = I0 + ε
∂ϕ
0 ∂Φ(I 0 , ϕ)
ϕ = ϕ+ε , (3)
∂I 0
where Φ(I 0 , ϕ) is an unknown generating function, which is determined so that (1) takes the
form (2). Decompose the perturbing function as

f (I, ϕ) = f0 (I) + f˜(I, ϕ) ,

where f0 is the average over the angle variables and f˜ is the remainder function defined through
f˜(I, ϕ) ≡ f (I, ϕ) − f0 (I). Define the frequency vector ω = ω(I) as
∂h(I)
ω(I) ≡ .
∂I

4
Inserting the transformation (3) in (1) and expanding in Taylor series around ε = 0 up to the
second order, one gets
∂Φ(I 0 , ϕ) ∂Φ(I 0 , ϕ)
h(I 0 + ε ) + εf (I 0 + ε , ϕ)
∂ϕ ∂ϕ
∂Φ(I 0 , ϕ)
= h(I 0 ) + ω(I 0 ) · ε + εf0 (I 0 ) + εf˜(I 0 , ϕ) + O(ε2 ) .
∂ϕ

The new Hamiltonian is integrable up to O(ε2 ) provided that the function Φ satisfies:
∂Φ(I 0 , ϕ)
ω(I 0 ) · + f˜(I 0 , ϕ) = 0 . (4)
∂ϕ
In such case the new integrable part becomes

h0 (I 0 ) = h(I 0 ) + εf0 (I 0 ) ,

which provides a better integrable approximation with respect to (1). The solution of (4) yields
the explicit expression of the generating function. In fact, let us expand Φ and f˜ in Fourier
series as

Φ(I 0 , ϕ) = Φ̂m (I 0 ) eim·ϕ ,

X

m∈Zn \{0}

f˜(I 0 , ϕ) = fˆm (I 0 ) eim·ϕ ,

X
(5)
m∈I

where I denotes the set of integer vectors corresponding to the non–vanishing Fourier coeffi-
cients of f˜. Inserting the above expansions in (4) one obtains

ω(I 0 ) · m Φ̂m (I 0 ) eim·ϕ = − fˆm (I 0 ) eim·ϕ ,

X X
i
m∈Zn \{0} m∈I

which provides
fˆm (I 0 )
Φ̂m (I 0 ) = − .
i ω(I 0 ) · m
Casting together the above formulae, the generating function is given by

fˆm (I 0 )
Φ(I 0 , ϕ) = i eim·ϕ .
X
(6)
m∈I
ω(I 0 ) · m

We stress that this algorithm is constructive in the sense that it provides an explicit expression
for the generating function and for the transformed Hamiltonian. We remark that (6) is well
defined unless there exists an integer vector m ∈ I such that

ω(I 0 ) · m = 0 .

On the contrary, if ω is rationally independent, there are no zero divisors in (6), though these
terms can become arbitrarily small with a proper choice of the vector m. This problem is known
as the small divisor problem, which can prevent the implementation of perturbation theory (see
[Hamiltonian Normal Forms?], [KAM theory]).

5
4.2 The precession of the perihelion of Mercury
As an example of the implementation of classical perturbation theory we consider the compu-
tation of the precession of the perihelion in a (restricted, planar, circular) three–body model,
taking as sample the planet Mercury. The computation requires the introduction of Delau-
nay action–angle variables, the definition of the three–body Hamiltonian, the expansion of the
perturbing function and the implementation of classical perturbation theory (see [7], [15]).

4.2.1 Delaunay action–angle variables

We consider two bodies, say P0 and P1 with masses, respectively, m0 , m1 ; let M ≡ m0 + m1
and let µ > 0 be a positive parameter. Let r be the orbital radius and ϕ be the longitude of P1
with respect to P0 ; let (Ir , Iϕ ) be the momenta conjugated to (r, ϕ). In these coordinates the
two–body problem Hamiltonian takes the form

1 Iϕ2 µM
H2b (Ir , Iϕ , r, ϕ) = (Ir2 + 2 ) − . (7)
2µ r r

On the orbital plane we introduce the planar Delaunay action–angle variables (Λ, Γ, λ, γ) as
follows ([21]). Let E denote the total mechanical energy; then:
s
2µ2 M Iϕ2
Ir = 2µE + − 2 .
r r
q
3 2
Since (7) does not depend on ϕ, setting Γ = Iϕ and Λ = − µ 2E
M
, we introduce a generating
function of the form
s
µ4 M 2 2µ2 M Γ2
Z
F (Λ, Γ, r, ϕ) = − + − dr + Γϕ .
Λ2 r r2

From the definition of Λ the new Hamiltonian H2D becomes

µ3 M 2
H2D (Λ, Γ, λ, γ) = − ,
2Λ2
where (Λ, Γ) are the Delaunay action variables; by Kepler’s laws one finds that (Λ, Γ) are related
to the semimajor axis a and to the eccentricity e of the Keplerian orbit of P1 around P0 by the
formulae: √ p
Λ = µ Ma , Γ = Λ 1 − e2 .
Concerning the conjugated angle variables, we start by introducing the eccentric anomaly u as
follows: build the auxiliary circle of the ellipse, draw the line through P1 perpendicular to the
semimajor axis whose intersection with the auxiliary circle forms at the origin an angle u with
the semimajor axis. By definition of the generating function, one finds

∂F µ4 M 2
Z
λ= = q dr = u − e sin u ,
∂Λ Λ3 − µ ΛM2 +
4 2 2µ2 M
− Γ2
r r2

which defines the mean anomaly λ in terms of the eccentric anomaly u.

6
In a similar way, if f denotes the true anomaly related to the eccentric anomaly by tan f2 =
q
1+e
1−e tan u2 , then one has:

∂F Γ
Z
γ= =ϕ− q dr = ϕ − f ,
∂Γ r2 − µ4 M 2
+ 2µ2 M
− Γ2
Λ2 r r2

which represents the argument of perihelion of P1 , i.e. the angle between the perihelion line
and a fixed reference line.

4.2.2 The restricted, planar, circular, three–body problem

Let P0 , P1 , P2 be three bodies of masses, respectively, m0 , m1 , m2 . We assume that m1 is
much smaller than m0 and m2 (restricted problem) and that the motion of P2 around P0 is
circular. We also assume that the three bodies always move on the same plane. We choose
the free parameter µ as µ ≡ 12/3 , so that the two–body Hamiltonian becomes H2D = − 2Λ1 2 ,
m0
m2
while we introduce the perturbing parameter as ε ≡ 2/3 ([21]). Set the units of measure so
m0
that the distance between P0 and P2 is one and so that m0 + m2 = 1. Taking into account the
interaction of P2 on P1 , the Hamiltonian function governing the three–body problem becomes
1  1 
H3b (Λ, Γ, λ, γ, t) = − + ε r 1 cos(ϕ − t) − ,
2Λ2
q
1 + r12 − 2r1 cos(ϕ − t)

where r1 is the distance between P0 and P1 . The first term of the perturbation comes out from
the choice of the reference frame, while the second term is due to the interaction with the
external body. Since ϕ − t = f + γ − t, we perform the canonical change of variables

L = Λ `=λ
G = Γ g =γ−t ,

which provides the following two degrees–of–freedom Hamiltonian

1
H3D (L, G, `, g) = − − G + εR(L, G, `, g) , (8)
2L2
where
1
R(L, G, `, g) ≡ r1 cos(ϕ − t) − q (9)
1 + r12 − 2r1 cos(ϕ − t)
with r1 and ϕ − t must be expressed in terms of the Delaunay variables (L, G, `, g). Notice
that when ε = 0 one obtains the integrable Hamiltonian function h(L, G) ≡ − 2L1 2 − G with
∂h ∂h
associated frequency vector ω = ( ∂L , ∂G ) = ( L13 , −1).

4.2.3 Expansion of the perturbing function

We expand the perturbing function (9) in terms of the Legendre polynomials Pj obtaining
∞
1 X 1
R = − Pj (cos(ϕ − t)) j .
r1 j=2 r1

7
The explicit expressions of the first few Legendre polynomials are:
1 3
P2 (cos(ϕ − t)) = + cos 2(ϕ − t)
4 4
3 5
P3 (cos(ϕ − t)) = cos(ϕ − t) + cos 3(ϕ − t)
8 8
9 5 35
P4 (cos(ϕ − t)) = + cos 2(ϕ − t) + cos 4(ϕ − t)
64 16 64
15 35 63
P5 (cos(ϕ − t)) = cos(ϕ − t) + cos 3(ϕ − t) + cos 5(ϕ − t) .
64 128 128
We invert Kepler’s equation ` = u − e sin u to the second order in the eccentricity as

e2
u = ` + e sin ` + sin(2`) + O(e3 ) ,
2
from which one gets
5
ϕ − t = g + ` + 2e sin ` + e2 sin 2` + O(e3 )
4
 1 2 1 
r1 = a 1 + e − e cos ` − e2 cos 2` + O(e3 ) .
2 2
Then, up to inessential constants the perturbing function can be expanded as

R = R00 (L, G) + R10 (L, G) cos ` + R11 (L, G) cos(` + g)

+ R12 (L, G) cos(` + 2g) + R22 (L, G) cos(2` + 2g)
+ R32 (L, G) cos(3` + 2g) + R33 (L, G) cos(3` + 3g)
+ R44 (L, G) cos(4` + 4g) + R55 (L, G) cos(5` + 5g) + ... , (10)
q
G2
where the coefficients Rij are given by the following expressions (recall that e = 1− L2
):

L4 9 3 L4 e 9
R00 = − (1 + L4 + e2 ) + ... , R10 = (1 + L4 ) + ...
4 16 2 2 8
3 6 5 4 L4 e
R11 = − L (1 + L ) + ... , R12 = (9 + 5L4 ) + ...
8 8 4
L4 5 4 3
R22 = − (3 + L ) + ... , R32 = − L4 e + ...
4 4 4
5 6 7 35
R33 = − L (1 + L4 ) + ... , R44 = − L8 + ...
8 16 64
63 10
R55 = − L + ... (11)
128

4.2.4 Computation of the precession of the perihelion

We identify the three bodies P0 , P1 , P2 with the Sun, Mercury and Jupiter. Taking ε as
perturbing parameter, we implement a first order perturbation theory, which provides a new
integrable Hamiltonian function of the form
1
h0 (L0 , G0 ) = − − G0 + ε R00 (L0 , G0 ) .
2L02

8
From Hamilton’s equations one obtains
∂h0 (L0 , G0 ) ∂R00 (L0 , G0 )
ġ = = −1 + ε ;
∂G0 ∂G0
neglecting O(e3 ) in R00 and recalling that g = γ − t, one has
∂R00 (L0 , G0 ) 3
γ̇ = ε 0
= εL02 G0 .
∂G 4
Notice that to the first order in ε one has L0 = L, G0 = G. The astronomical data are m0 =
2·1030 kg, m2 = 1.9·1027 kg, which give ε = 9.49·10−4 ; setting to one the Jupiter–Sun distance
one has a = 0.0744, while e = 0.2056. Taking into account that the orbital period of Jupiter
amounts to about 11.86 years, one obtains
arcsecond
γ̇ = 154.65 ,
century
which represents the contribution due to Jupiter to the precession of perihelion of Mercury.
The value found by Leverrier on the basis of the data available in the year 1856 was of 152.59
arcsecond/century ([6]).

5 Resonant perturbation theory

5.1 The resonant theory
Let us consider an Hamiltonian system with n degrees of freedom of the form
H(I, ϕ) = h(I) + εf (I, ϕ)

and let ωj (I) = ∂h(I)

∂Ij (j = 1, ..., n) be the frequencies of the motion, which we assume to satisfy
`, ` < n, resonance relations of the form
ω · mk = 0 for k = 1, ..., ` ,
for suitable rational independent integer vectors m1 , ..., m` . A resonant perturbation theory can
be implemented to eliminate the non–resonant terms. More precisely, the aim is to construct
a canonical transformation C : (I, ϕ) → (J 0 , ϑ0 ) such that the transformed Hamiltonian takes
the form
H 0 (J 0 , ϑ0 ) = h0 (J 0 , ϑ01 , ..., ϑ0` ) + ε2 f 0 (J 0 , ϑ0 ) , (12)
where h0 depends only on the resonant angles ϑ01 , ..., ϑ0` . To this end, let us first introduce the
angles ϑ ∈ Tn as
ϑj = mj · ϕ j = 1, ..., `
ϑ k = mk · ϕ k = ` + 1, ..., n ,
where the first ` angle variables are the resonant angles, while the latter n − ` angle variables
are defined as suitable linear combinations so to make the transformation canonical together
with the following change of coordinates on the actions J ∈ Rn :
Ij = mj · J j = 1, ..., `
Ik = mk · J k = ` + 1, ..., n .

9
The aim is to construct a canonical transformation which removes (to higher order) the de-
pendence on the short–period angles (ϑ`+1 , ..., ϑn ), while the lowest order Hamiltonian will
necessarily depend upon the resonant angles. Let us decompose the perturbation as

f (J, ϑ) = f (J) + fr (J, ϑ1 , ..., ϑ` ) + fn (J, ϑ) , (13)

where f is the average of the perturbation over the angles, fr is the part depending on the
resonant angles and fn is the non–resonant part. In analogy to the classical perturbation theory,
we implement a canonical transformation of the form
∂Φ 0
J = J0 + ε (J , ϑ)
∂ϑ
∂Φ
ϑ0 = ϑ + ε 0 (J 0 , ϑ) ,
∂J
such that the new Hamiltonian takes the form (12). Taking into account (13) and developing
up to the second order in the perturbing parameter, one obtains:
∂Φ
h(J 0 + ε ) + εf (J 0 , ϑ) + O(ε2 )
∂ϑ
n
∂h ∂Φ
= h(J 0 ) + ε + εf (J 0 ) + εfr (J 0 , ϑ1 , ..., ϑ` ) + εfn (J 0 , ϑ) + O(ε2 ) .
X

k=1
∂J k ∂ϑ k

Equating same orders of ε one gets that

h0 (J 0 , ϑ01 , ..., ϑ0` ) = h(J 0 ) + εf (J 0 ) + εfr (J 0 , ϑ01 , ..., ϑ0` ) , (14)

provided that
n
∂Φ
ωk0 = −fn (J 0 , ϑ) ,
X
(15)
k=1
∂ϑk
0
where ωk0 = ωk0 (J 0 ) ≡ ∂h(J )
∂Jk0 . The solution of (15) gives the generating function, which allows
to reduce the Hamiltonian to the required form (12); as a consequence the conjugated action
variables, say J`+10 , ..., J 0 , are constants of the motion up to the second order in ε. We conclude
n
by mentioning that using the new frequencies ωk0 , the resonant relations take the form ωk0 = 0
for k = 1, ..., `.

5.2 Three–body resonance

We consider the three–body Hamiltonian (8) with perturbing function (10)–(11) and let ω ≡
(ω` , ωg ) be the frequency of motion. We assume that the frequency vector satisfies the resonance
relation
ω` + 2ωg = 0 .
According to the theory described in the previous section we perform the canonical change of
variables
1
ϑ1 = ` + 2g J1 = G
2
1 1
ϑ2 = 2` J2 = L − G .
2 4

10
In the new coordinates the unperturbed Hamiltonian becomes
1
h0 (J) ≡ − + 2J1 ,
2(J1 + 2J2 )2
∂h0 (J)
with frequency vector ω 0 ≡ ∂J , while the perturbation takes the form

1 1 1
R(J1 , J2 , ϑ1 , ϑ2 ) ≡ R00 (J) + R10 (J) cos( ϑ2 ) + R11 (J) cos( ϑ1 + ϑ2 )
2 2 4
1
+ R12 (J) cos(ϑ1 ) + R22 (J) cos(ϑ1 + ϑ2 )
2
3 3
+ R32 (J) cos(ϑ1 + ϑ2 ) + R33 (J) cos( ϑ1 + ϑ2 )
2 4
5 5
+ R44 (J) cos(2ϑ1 + ϑ2 ) + R55 (J) cos( ϑ1 + ϑ2 ) + ...
2 4
with the coefficients Rij as in (11). Let us decompose the perturbation as R = R(J)+Rr (J, ϑ1 )+
Rn (J, ϑ), where R(J) is the average over the angles, Rr (J, ϑ1 ) = R12 (J) cos(ϑ1 ) is the reso-
nant part, while Rn contains all the remaining non–resonant terms. We look for a canonical
transformation close to the identity with generating function Φ = Φ(J 0 , ϑ) such that

∂Φ(J 0 , ϑ)
ω 0 (J 0 ) · = −Rn (J 0 , ϑ) ,
∂ϑ

which is well defined since ω 0 is non–resonant for the Fourier components appearing in Rn .
Finally, according to (14) the new unperturbed Hamiltonian is given by

h0 (J 0 , ϑ01 ) ≡ h(J 0 ) + εR00 (J 0 ) + εR12 (J 0 ) cos ϑ01 .

5.3 Degenerate perturbation theory

A special case of resonant perturbation theory is obtained when considering a degenerate
Hamiltonian function with n degrees of freedom of the form

H(I, ϕ) = h(I1 , ..., Id ) + εf (I, ϕ) , d<n; (16)

notice that the integrable part depends on a subset of the action variables, being degenerate
in Id+1 , ...,In . In this case we look for a canonical transformation C : (I, ϕ) → (I 0 , ϕ0 ) such that
the transformed Hamiltonian becomes

H 0 (I 0 , ϕ0 ) = h0 (I 0 ) + εh01 (I, ϕ0d+1 , ..., ϕ0n ) + ε2 f 0 (I 0 , ϕ0 ) , (17)

where the part h0 + εh01 admits d integrals of motion. Let us decompose the perturbing function
in (16) as
f (I, ϕ) = f (I) + fd (I, ϕd+1 , .., ϕn ) + f˜(I, ϕ) , (18)
where f is the average over the angle variables, fd is independent on ϕ1 , ..., ϕd and f˜ is
the remainder. As in the previous sections we want to determine a near–to–identity canonical

11
transformation Φ = Φ(I 0 , ϕ) of the form (3), such that in view of (18) the Hamiltonian (16)
takes the form (17). One obtains
d
∂h ∂Φ
h(I10 , ..., Id0 ) + ε + εf (I 0 ) + εfd (I 0 , ϕd+1 , ..., ϕn ) + εf˜(I 0 , ϕ) + O(ε2 )
X

k=1
∂Ik ∂ϕk
= h0 (I 0 ) + εh01 (I 0 , ϕd+1 , ..., ϕn ) + O(ε2 ) ,

where

h0 (I 0 ) = h(I10 , ..., Id0 ) + εf (I 0 )

h01 (I 0 , ϕd+1 , ..., ϕn ) = fd (I 0 , ϕd+1 , ..., ϕn ) ,

while Φ is determined solving the equation

d
∂h ∂Φ
+ f˜(I 0 , ϕ) = 0 .
X

k=1
∂Ik ∂ϕk

Expanding Φ and f˜ in Fourier series as in (5) one obtains that Φ is given by (6) where ω · m =
Pd
k=1 mk ωk , being ωk = 0 for k = d + 1, ..., n. The generating function is well defined provided
that ω · m 6= 0 for any m ∈ I, which is equivalent to require that
d
X
mk ωk 6= 0 for m ∈ I .
k=1

5.4 The precession of the equinoxes

An example of the application of the degenerate perturbation theory in Celestial Mechanics is
provided by the computation of the precession of the equinoxes.
We consider a triaxial rigid body moving in the gravitational field of a primary body. We
introduce the following reference frames with common origin in the barycenter of the rigid
(i) (i) (i) (b) (b) (b)
body: (O, i1 , i2 , i3 ) is an inertial reference frame, (O, i1 , i2 , i3 ) is a body frame oriented
(s) (s) (s)
along the direction of the principal axes of the ellipsoid, (O, i1 , i2 , i3 ) is the spin reference
frame with the vertical axis along the direction of the angular momentum. Let (J, g, `) be the
Euler angles formed by the body and spin frames, and let (K, h, 0) be the Euler angles formed
by the spin and inertial frames. The angle K is the obliquity (representing the angle between
the spin and inertial vertical axes), while J is the non–principal rotation angle (representing
the angle between the spin and body vertical axes).
This problem is conveniently described in terms of the following set of action–angle variables
introduced by Andoyer in [1] (see also [24]). Let M 0 be the angular momentum and let M0 ≡
|M 0 |; the action variables are defined as
(s)
G ≡ M 0 · i3 = M0
(b)
L ≡ M 0 · i3 = G cos J
(i)
H ≡ M 0 · i3 = G cos K ,

while the corresponding angle variables are the quantities (g, `, h) introduced before.

12
We limit ourselves to consider the gyroscopic case in which I1 = I2 < I3 are the principal
moments of inertia of the rigid body E around the primary S; let mE and mS be their masses
and let |E| be the volume of E. We assume that E orbits on a Keplerian ellipse around S with
semimajor axis a and eccentricity e, while λE and rE denote the longitude and instantaneous
orbital radius (due to the assumption of Keplerian motion λE and rE are known functions of
the time). The Hamiltonian describing the motion of E around S is given by ([23])

G2 I1 − I3 2
H(L, G, H, `, g, h, t) = + L + V (L, G, H, `, g, h, t) ,
2I1 2I1 I3
where the perturbation is implicitly defined by

G̃mS mE dx
Z
V ≡− ,
E |rE + x| |E|

G̃ being the gravitational constant. Setting rE = |rE | and x = |x|, we can expand V using the
Legendre polynomials as

G̃mS mE dx h x·r 1 (x · rE )2 x
Z i
V =− 1 − 2 E + 2 (3 2 − x2
) + O(( )3 ) .
rE E |E| rE 2rE rE rE

We further assume that J = 0 (i.e. G = L) so that E rotates around a principal axis. Let
G0 and H0 be the initial values of G and H; if α denotes the angle bewteen rE and k, the
perturbing function can be written as

3 G2 (1 − e cos λE )3
V = ηω 0 cos2 α
2 H0 (1 − e2 )3

I3 −I1 G̃mS H0
with η = I3 and ω = a3
I3 G20
. Elementary computations show that
s
H2
cos α = sin(λE − h) 1− .
G2
3
Neglecting first order terms in the eccentricity, we approximate (1−e cos λE )
(1−e2 )3
with one. A first
order degenerate perturbation theory provides that the new unperturbed Hamiltonian is given
by
G2 3 G20 G2 − H 2
K(G, H) = + ηω .
2I3 2 H0 2G2
Therefore the average angular velocity of precession is given by

∂K(G, H) 3 G2 H
ḣ = = − ηω 0 2 .
∂H 2 H0 G
At t = 0 it is
3 3
ḣ = − ηω = − ηωy2 ωd−1 cos K , (19)
2 2
where we used ω = ωy2 ωd−1 cos K with ωy being the frequency of revolution and ωd the frequency
of rotation.

13
 1
In the case of the Earth, the astronomical measurements show that η = 298.25 , K = 23.45o .
The contribution due to the Sun is thus obtained inserting ωy = 1 year, ωd = 1 day in (19),
which yields ḣ(S) = −2.51857 · 10−12 rad/sec, corresponding to a retrograde precessional period
of 79107.9 years. A similar computation shows that the contribution of the Moon amounts
to ḣ(L) = −5.49028 · 10−12 rad/sec, corresponding to a precessional period of 36289.3 years.
The total amount is obtained as the sum of ḣ(S) and ḣ(L) , providing an overall retrograde
precessional period of 24877.3 years.

6 Invariant tori
6.1 Invariant KAM surfaces
We consider an n–dimensional nearly–integrable Hamiltonian function

H(I, ϕ) = h(I) + εf (I, ϕ) ,

defined in a 2n–dimensional phase space M ≡ V × Tn , where V is an open bounded region

of Rn . A KAM torus associated to H is an n–dimensional invariant surface on which the flow
is described parametrically by a coordinate θ ∈ Tn such that the conjugated flow is linear,
namely θ ∈ Tn → θ + ωt where ω ∈ Rn is a Diophantine vector, i.e. there exist γ > 0 and
τ > 0 such that
γ
|ω · m| ≥ , ∀ m ∈ Zn \{0} .
|m|τ
Kolmogorov’s theorem ([31], see also [KAM theory]) ensures the persistence of invariant tori
with diophantine frequency, provided ε is sufficiently small and provided the unperturbed
Hamiltonian is non–degenerate, i.e. for a given torus {I 0 } × Tn ⊂ M
!
00 ∂2h
det h (I 0 ) ≡ det (I ) 6= 0 . (20)
∂Ii ∂Ij 0 i,j=1,...,n

The condition (20) can be replaced by the isoenergetic non–degeneracy condition introduced
by Arnold ([16])  00
h (I 0 ) h0 (I 0 )

det 6= 0 , (21)
h0 (I 0 ) 0
which ensures the existence of KAM tori on the energy level corresponding to the unperturbed
energy h(I 0 ), say M0 ≡ {(I, ϕ) ∈ M : H(I, ϕ) = h(I 0 )}. In the context of the n–body problem
Arnold ([16]) addressed the question of the existence of a set of initial conditions with positive
measure such that, if the initial position and velocities of the bodies belong to this set, then the
mutual distances remain perpetually bounded. A positive answer is provided by Kolmogorov’s
theorem in the framework of the planar, circular, restricted three–body problem, since the
integrable part of the Hamiltonian (8) satisfies the isoenergetic non–degeneracy condition (21);
denoting by (L0 , G0 ) the initial values of the Delaunay’s action variables, if ε is sufficiently small,
there exist KAM tori for (8) on the energy level M0 ≡ {H3D = − 2L1 2 − G0 }. In particular,
0
the motion of the perturbed body remains forever bounded from the orbits of the primaries.
Indeed, a stronger statement is also valid: due to the fact that the two–dimensional KAM
surfaces separate the three dimensional energy levels, any trajectory starting between two
KAM tori remains forever trapped in the region between such tori.

14
In the framework of the three–body problem, Arnold ([16]) stated the following result: “If the
masses, eccentricities and inclinations of the planets are sufficiently small, then for the majority
of initial conditions the true motion is conditionally periodic and differs little from Lagrangian
motion with suitable initial conditions throughout an infinite interval of time −∞ < t < ∞”.
Arnold provided a complete proof for the case of three coplanar bodies, while the spatial three–
body problem was investigated by Laskar and Robutel in [32], [38] using Poincaré variables,
the Jacobi’s “reduction of the nodes” (see, e.g., [5]) and Birkhoff’s normal form ([14], [2], [3]).
The full proof of Arnold’s theorem was provided in [26], based on Herman’s results on the
planetary problem; it makes use of Poincaré variables restricted to the symplectic manifold of
vertical total angular momentum.
Explicit estimates on the perturbing parameter ensuring the existence of KAM tori were given
by M. Hénon ([30]); he showed that direct applications of KAM theory to the three–body
problem lead to analytical results which are much smaller than the astronomical observations.
For example, the application of Arnold’s theorem to the restricted three–body problem is valid
provided the mass–ratio of the primaries is less than 10−333 . This result can be improved up to
10−48 by applying Moser’s theorem, but it is still very far from the actual Jupiter–Sun mass–
ratio which amounts to about 10−3 . In the context of concrete estimates, a big improvement
comes from the synergy between KAM theory and computer–assisted proofs, based on the
application of interval arithmetic which allows to keep rigorously track of the rounding–off
and propagation errors introduced by the machine. Computer–assisted KAM estimates were
implemented in a number of cases in Celestial Mechanics, like the three–body problem and the
spin–orbit model as briefly recalled in the following subsections.
Another interesting example of the interaction between the analytical theory and the computer
implementation is provided by the analysis of the stability of the triangular Lagrangian points;
in particular, the stability for exponentially long times is obtained using Nekhoroshev theory
combined with computer–assisted implementations of Birkhoff normal form (see, e.g., [17], [22],
[25], [27], [28], [29], [36], [39]).

6.2 Rotational tori for the spin–orbit problem

We study the motion of a rigid triaxial satellite around a central planet under the following
assumptions ([18]):
i) the orbit of the satellite is Keplerian,
ii) the spin–axis is perpendicular to the orbital plane,
iii) the spin–axis coincides with the smallest physical axis,
iv) external perturbations as well as dissipative forces are neglected.
Let I1 < I2 < I3 be the principal moments of inertia; let a, e be the semimajor axis and
eccentricity of the Keplerian ellipse; let r and f be the instantaneous orbital radius and the
true anomaly of the satellite; let x be the angle between the longest axis of the triaxial satellite
and the periapsis line. The equation of motion governing the spin–orbit model is given by:
3 I2 − I1 a 3
ẍ + ( ) sin(2x − 2f ) = 0 . (22)
2 I3 r
Due to assumption i), the quantities r and f are known functions of the time. Expanding the
second term of (22) in Fourier–Taylor series and neglecting terms of order 6 in the eccentricity,

15
setting y ≡ ẋ one obtains that the equation of motion corresponds to Hamilton’s equations
associated to the Hamiltonian
y2 e e3 5 5
H(y, x, t) ≡ − ε [(− + − e ) cos(2x − t) +
2 4 32 768
1 5 13 7 123 3 489 5
+ ( − e2 + e4 ) cos(2x − 2t) + ( e − e + e ) cos(2x − 3t) +
2 4 32 4 32 256
17 115 4 845 3 32525 5
+ ( e2 − e ) cos(2x − 4t) + ( e − e ) cos(2x − 5t) +
4 12 96 1536
533 4 228347 5
+ e cos(2x − 6t) + e cos(2x − 7t) ] , (23)
32 7680
where ε ≡ 23 I2I−I
3
1
and we have chosen the units so that a = 1, T2π rev
= 1, where Trev is the
period of revolution. Let p, q be integers with q 6= 0; a p : q resonance occurs whenever hẋi = pq ,
meaning that during q orbital revolutions, the satellite makes on average p rotations. Being the
phase–space three–dimensional, the two–dimensional KAM tori separate the phase–space into
invariant regions, thus providing the stability of the trapped orbits. In particular, let P( pq ) be
the periodic orbit associated to the p : q resonance; its stability is guaranteed by the existence
of trapping rotational tori with frequencies T (ω1 ) and T (ω2 ) with ω1 < pq < ω2 . For example,
one can consider the sequences of irrational rotation numbers
(p/q) p 1 (p/q) p 1
Γk ≡ − , ∆k ≡ + , k ∈ Z, k ≥ 2
q k+α q k+α
√
with α ≡ 5−1
2 . In fact, the continued fraction expansion of 1
k+α is given by 1
k+α = [0, k, 1∞ ].
(p/q) (p/q)
Therefore, both Γk and ∆k are noble numbers (i.e. with continued fraction expansion
definitely equal to one); by number theory they satisfy the diophantine condition and bound
p
q from below and above.
As a concrete sample we consider the synchronous spin–orbit resonance (p = q = 1) of the
Moon, whose physical values of the parameters are ε ≡ 3.45 · 10−4 and e = 0.0549. The
(1) (1)
stability of the motion is guaranteed by the existence of the surfaces T (Γ40 ) and T (∆40 ),
which is obtained implementing a computer–assisted KAM theory for the realistic values of the
parameters. The result provides the confinement of the synchronous periodic orbit in a limited
region of the phase space.

6.3 Librational tori for the spin–orbit problem

The existence of invariant librational tori around a spin–orbit resonance can be obtained as fol-
lows ([19]). Let us consider the 1:1 resonance corresponding to Hamilton’s equations associated
to (23). First one implements a canonical transformation to center around the synchronous
periodic orbit; after expanding in Taylor series, one diagonalizes the quadratic terms, thus
obtaining a harmonic oscillator plus higher degree (time–dependent) terms. Finally, it is con-
venient to transform the Hamiltonian using the action–angle variables (I, ϕ) of the harmonic
oscillator. After these symplectic changes of variables one is led to a Hamiltonian of the form

H(I, ϕ, t) ≡ ωI + εh(I) + εR(I, ϕ, t) , I ∈ R, (ϕ, t) ∈ T 2 ,

where ω ≡ ω(ε) is the frequency of the harmonic oscillator, while h(I) and R(I, ϕ, t) are suitable
functions, precisely polynomials in the action (or the in the square of the action). Then apply

16
a Birkhoff normal form (see [Hamiltonian Normal Forms?], [Normal Forms?]) up to the order
k (k = 5 in [19]) to obtain the following Hamiltonian:
Hk (I 0 , ϕ0 , t) = ωI 0 + ε hk (I 0 ; ε) + εk+1 Rk (I 0 , ϕ0 , t) .
Finally, implementing a computer–assisted KAM theorem one gets the following result: consider
the Moon–Earth case with εobs = 3.45 · 10−4 and e = 0.0549; there exists an invariant torus
around the synchronous resonance corresponding to a libration of 8o .79 for any ε ≤ εobs /5.26.
The same strategy applied to different samples, e.g. the Rhea–Saturn pair, allows to prove
the existence of librational invariant tori around the synchronous resonance for values of the
parameters in full agreement with the observational measurements ([19]).

6.4 Rotational tori for the restricted three–body problem

The planar, circular, restricted three–body problem has been considered in [21], where the
stability of the asteroid 12 Victoria has been investigated under the gravitational influence
of the Sun and Jupiter. On a fixed energy level invariant KAM tori trapping the motion of
Victoria have been established for the astronomical value of the Jupiter–Sun mass–ratio (about
10−3 ). After an expansion of the perturbing function and a truncation to a suitable order (see
[21]), the Hamiltonian function describing the motion of the asteroid is given in Delaunay’s
variables by
1
H(L, G, `, g) ≡ − 2 − G − εf (L, G, `, g) ,
2L
q
G2
where setting a ≡ L2 , e = 1− L2
, the perturbation is given by

a2 9 4 3 2 2 1 9 
f (L, G, `, g) = 1 + + a + a e − + a2 a2 e cos `
4 64 8 2 16
3 15 5  9 5 
3
+ a + a cos(` + g) − + a2 a2 e cos(` + 2g)
8 64 4 4
3 5  3
+ a2 + a4 cos(2 ` + 2 g) + a2 e cos(3 ` + 2 g)
4 16 4
5 35 5  35 4
3
+ a + a cos(3 ` + 3 g) + a cos(4 ` + 4 g)
8 128 64
63 5
+ a cos(5` + 5g) .
128
For the asteroid Victoria the orbital elements are aV ' 2.334 AU, eV ' 0.220, which give the
observed values of the Delaunay’s action variables as LV = 0.670, GV = 0.654. The energy
level is taken as
(0) 1 (1) (0) (1)
EV ≡ − 2 −GV ' −1.768 , EV ≡ −hf (LV , GV , `, g)i ' −1.060 , EV (ε) ≡ EV +εEV .
2LV
The osculating energy level of the Sun–Jupiter–Victoria model is defined as
∗ (0) (1)
EV ≡ EV (εJ ) = EV + εJ EV ' −1.769 .
We now look for two invariant tori bounding the observed values of LV and GV . To this end,
let L̃± = LV ± 0.001 and let
 1 
ω̃ ± = , −1 ≡ (α̃± , −1) .
L̃3±

17
To obtain diophantine frequencies, the continued fraction expansion of α̃± is modified adding a
tail of one’s after the order 5; this procedure gives the diophantine numbers α± which define the
bounding frequencies as ω ± = (α± , −1). By a computer–assisted KAM theorem, the stability
of the asteroid Victoria is a consequence of the following result ([21]): for |ε| ≤ 10−3 the
unperturbed tori can be analytically continued into invariant KAM tori for the perturbed
system on the energy level H −1 (EV (ε)), keeping fixed the ratio of the frequencies. Therefore
the orbital elements corresponding to the semimajor axis and to the eccentricity of the asteroid
Victoria stay forever ε–close to their unperturbed values.

6.5 Planetary problem

The dynamics of the planetary problem composed by the Sun, Jupiter and Saturn is investigated
in [33], [34] and [35]. In [33] the secular dynamics of the following model is studied: after the
Jacobi’s reduction of the nodes, the 4–dimensional Hamiltonian is averaged over the fast angles
and its series expansion is considered up to the second order in the masses. This procedure
provides a Hamiltonian function with two degrees of freedom, describing the slow motion of
the parameters characterizing the Keplerian approximation (i.e., the eccentricities and the
arguments of perihelion). Afterwards, action–angle coordinates are introduced and a partial
Birkhoff normalization is performed. Finally, a computer–assisted implementation of a KAM
theorem yields the existence of two invariant tori bounding the secular motions of Jupiter and
Saturn for the observed values of the parameters.
The approach sketched above is extended in [35] so to include the description of the fast vari-
ables, like the semi–major axes and the mean longitudes of the planets. Indeed, the preliminary
average on the fast angles is now performed without eliminating the terms with degree greater
or equal than two with respect to the fast actions. The canonical transformations involving the
secular coordinates can be adapted to produce a good initial approximation of an invariant
torus for the reduced Hamiltonian of the three–body planetary problem. This is the starting
point of the procedure for constructing the Kolmogorov’s normal form which is numerically
shown to be convergent. In [34] the same result of [35] has been obtained for a fictitious plan-
etary solar system composed by two planets with masses equal to 1/10 of those of Jupiter and
Saturn.

7 Periodic orbits
7.1 Construction of periodic orbits
One of the most intriguing conjectures of Poincaré concerns the pivotal role of the periodic
orbits in the study of the dynamics; more precisely, he states that given a particular solution of
Hamilton’s equations one can always find a periodic solution (possibly with very long period)
such that the difference between the two solutions is small for an arbitrary long time. The
literature on periodic orbits is extremely wide (see, e.g., [3], [7], [10], [14], [15] and references
therein); here we present the construction of periodic orbits implementing a perturbative ap-
proach (see [20]) as shown by Poincaré in [12]. We describe such method taking as example the
2
spin–orbit Hamiltonian (23) that we write in a compact form as H(y, x, t) ≡ y2 − εf (x, t) for
a suitable function f = f (x, t); the corresponding Hamilton’s equations are
ẋ = y

18
 ẏ = εfx (x, t) . (24)
A spin–orbit resonance of order p : q is a periodic solution of period T = 2πq (q ∈ Z\{0}), such
that
x(t + 2πq) = x(t) + 2πp
y(t + 2πq) = y(t) . (25)
From (24) the solution can be written in integral form as
Z t
y(t) = y(0) + ε fx (x(s), s) ds
0
Z tZ τ Z t
x(t) = x(0) + y(0)t + ε fx (x(s), s) ds dτ = x(0) + y(s) ds ;
0 0 0
combining the above equations with (25) one obtains
Z 2πq
fx (x(s), s)ds = 0
0
Z 2πq
y(s)ds − 2πp = 0 . (26)
0
Let us write the solution as the series
x(t) ≡ x + yt + εx1 (t) + ...
y(t) ≡ y + εy1 (t) + ... , (27)
where x(0) = x and y(0) = y are the initial conditions, while x1 (t), y1 (t) are the first order
terms in ε. Expanding the initial conditions in power series of ε, one gets:
x = x0 + εx1 + ε2 x2 + ...
y = y 0 + εy 1 + ε2 y 2 + ... (28)
Inserting (27) and (28) in (24), equating same orders in ε and taking into account the periodicity
condition (26), one can find the following explicit expressions for x1 (t), y1 (t), y 0 , y 1 :
Z t
y1 (t) = y1 (t; y, x) = fx (x0 + y 0 s, s) ds
0
Z t
x1 (t) = x1 (t; y, x) = y1 (s) ds
0
p
y0 =
q
1
Z 2πq Z t
y1 = − fx (x0 + y 0 s, s) ds dt .
2πq 0 0
Furthermore, x0 is determined as a solution of
Z 2πq
fx (x0 + y 0 s, s) ds = 0 ,
0
while x1 is given by
Z 2πq Z 2πq
1 h
0 0
i
x1 = − R 2πq y1 tfxx dt + fxx x1 (t) dt ,
0 dt
fxx 0 0
0
0 = f (x + y t, t).
where, for shortness, we have written fxx xx 0 0

19
7.2 The libration in longitude of the Moon
The previous computation of the p : q periodic solution can be used to evaluate the libration
in longitude of the Moon. More precisely, setting p = q = 1 one obtains

x0 = 0
y0 = 1
x1 (t) = 0.232086t − 0.218318 sin(t) − 6.36124 · 10−3 sin(2t)
− 3.21314 · 10−4 sin(3t) − 1.89137 · 10−5 sin(4t)
− 1.18628 · 10−6 sin(5t)
y1 (t) = 0.232086 − 0.218318 cos(t) − 0.0127225 cos(2t)
− 9.63942 · 10−4 cos(3t) − 7.56548 · 10−5 cos(4t)
− 5.93138 · 10−6 cos(5t)
x1 = 0
y 1 = −0.232086 ,

where we used e = 0.05494, ε = 3.45 · 10−4 . Therefore the synchronous periodic solution,
computed up to the first order in ε, is given by

x(t) = x0 + y 0 t + εx1 (t) = t − 7.53196 · 10−5 sin(t) − 2.19463 · 10−6 sin(2t)

− 1.10853 · 10−7 sin(3t) − 6.52523 · 10−9 sin(4t)
− 4.09265 · 10−10 sin(5t)
y(t) = y 0 t + εy1 (t) = 1 − 7.53196 · 10−5 cos(t) − 4.38926 · 10−6 cos(2t)
− 3.3256 · 10−7 cos(3t) − 2.61009 · 10−8 cos(4t)
− 2.04633 · 10−9 cos(5t) .

It turns out that the libration in longitude of the Moon, provided by the quantity x(t) − t, is
of the order of 7 · 10−5 in agreement with the observational data.

8 Future directions
The end of the XX century has been greatly marked by astronomical discoveries, which changed
the shape of the solar system as well as of the entourage of other stars. In particular, the
detection of many small bodies beyond the orbit of Neptune has moved forward the edge of
the solar system and it has increased the number of its population. Hundreds objects have
been observed to move in a ring beyond Neptune, thus forming the so–called Kuiper’s belt.
Its components show a great variety of behaviors, like resonance clusterings, regular orbits,
scattered trajectories. Furthermore, far outside the solar system, the astronomical observations
of extrasolar planetary systems have opened new scenarios with a great variety of dynamical
behaviors. In these contexts classical and resonant perturbation theories will deeply contribute
to provide a fundamental insight of the dynamics and will play a prominent role in explaining
the different configurations observed within the Kuiper’s belt as well as within extrasolar
planetary systems.

20
9 Bibliography

References

9.1 Books and Reviews

[1] H. Andoyer, Mécanique Céleste, Gauthier-Villars, Paris (1926)

[2] V.I. Arnold, Mathematical methods of classical mechanics, Springer–Verlag, Berlin, Hei-
delberg, New York (1978)

[3] V.I. Arnold (editor), Encyclopedia of Mathematical Sciences, Dynamical Systems III,
Springer–Verlag 3 (1988)

[4] D. Boccaletti, G. Pucacco, Theory of orbits, Springer–Verlag, Berlin, Heidelberg, New

York (2001)

[5] A. Celletti, L. Chierchia, KAM tori for N-body problems: a brief history, Celestial Mechan-
ics and Dynamical Astronomy 95, 1, 117–139 (2006)

[6] A.G. Chebotarev, Analytical and Numerical Methods of Celestial Mechanics, American
Elsevier Publishing Co., New York (1967)

[7] D. Brouwer, G. Clemence, Methods of Celestial Mechanics, Academic Press, New York
(1961)

[8] C. Delaunay, Mémoire sur la théorie de la Lune, Mém. de l’Acad. des Sciences 28 (1860)
and 29 (1867)

[9] S. Ferraz–Mello, Canonical Perturbation Theories, Springer–Verlag, Berlin, Heidelberg,

New York (2007)

[10] Y. Hagihara, Celestial Mechanics, MIT Press, Cambridge (1970)

[11] K.R. Meyer, G.R. Hall, Introduction to Hamiltonian dynamical systems and the N –body
problem, Springer–Verlag, Berlin, Heidelberg, New York (1991)

[12] H. Poincarè, Les Méthodes Nouvelles de la Méchanique Céleste, Gauthier Villars, Paris
(1892)

[13] J.A. Sanders, F. Verhulst, Averaging methods in nonlinear dynamical systems, Springer–
Verlag, Berlin, Heidelberg, New York (1985)

[14] C.L. Siegel, J.K. Moser, Lectures on Celestial Mechanics, Springer–Verlag, Berlin, Heidel-
berg, Berlin (1971)

[15] V. Szebehely, Theory of orbits, Academic Press, New York and London (1967)

21
 9.2 Primary Literature

[16] V.I. Arnold, Small denominators and problems of stability of motion in classical and ce-
lestial mechanics, Uspehi Mat. Nauk 18, no. 6 (114), 91–192 (1963)

[17] G. Benettin, F. Fasso’, M. Guzzo, Nekhoroshev–stability of L4 and L5 in the spatial re-

stricted three–body problem, Regular and Chaotic Dynamics 3, n. 3, 56–71 (1998)

[18] A. Celletti, Analysis of resonances in the spin-orbit problem in Celestial Mechanics: The
synchronous resonance (Part I), Journal of Applied Mathematics and Physics (ZAMP)
41, 174–204 (1990)

[19] A. Celletti, Construction of librational invariant tori in the spin-orbit problem, Journal of
Applied Mathematics and Physics (ZAMP) 45, 61–80 (1993)

[20] A. Celletti, L. Chierchia, Construction of stable periodic orbits for the spin–orbit problem of
Celestial Mechanics, Regular and Chaotic Dynamics, (Editorial URSS) 3, 107–121 (1998)

[21] A. Celletti, L. Chierchia, KAM Stability and Celestial Mechanics, Memoirs of the American
Mathematical Society 187, 878 (2007)

[22] A. Celletti, A. Giorgilli, On the stability of the Lagrangian points in the spatial restricted
problem of three bodies, Celestial Mechanics and Dynamical Astronomy 50, 31–58 (1991)

[23] L. Chierchia, G. Gallavotti, Drift and diffusion in phase space, Ann. de l’Inst. H. Poincaré
60, 1–144 (1994)

[24] A. Deprit, Free rotation of a rigid body studied in the phase space, Am. J. Phys. 35, 424-428
(1967)

[25] C. Efthymiopoulos, Z. Sandor, Optimized Nekhoroshev stability estimates for the Trojan
asteroids with a symplectic mapping model of co-orbital motion, MNRAS 364, n. 6, 253-271
(2005)

[26] J. Féjoz, Démonstration du “théorème d’Arnold” sur la stabilité du système planétaire

(d’après Michael Herman), Ergod. Th. & Dynam. Sys. 24, 1–62 (2004)

[27] F. Gabern, A. Jorba, U. Locatelli, On the construction of the Kolmogorov normal form
for the Trojan asteroids, Nonlinearity 18 1705–1734 (2005)

[28] A. Giorgilli, C. Skokos, On the stability of the trojan asteroids, Astron. Astroph. 317
254–261 (1997)

[29] A. Giorgilli, A. Delshams, E. Fontich, L. Galgani, C. Simó, Effective stability for a Hamil-
tonian system near an elliptic equilibrium point, with an application to the restricted three–
body problem, J. Diff. Eq. 77 167–198 (1989)

[30] M. Hénon, Explorationes numérique du problème restreint IV: Masses egales, orbites non
periodique, Bullettin Astronomique 3, 1, fasc. 2, 49–66 (1966)

[31] A. N. Kolmogorov, On the conservation of conditionally periodic motions under small

perturbation of the Hamiltonian, Dokl. Akad. Nauk. SSR 98, 527–530 (1954)

22
[32] J. Laskar, P. Robutel, Stability of the planetary three–body problem. I. Expansion of the
planetary Hamiltonian, Celestial Mechanics and Dynamical Astronomy 62, no. 3, 193–217
(1995)

[33] U. Locatelli, A. Giorgilli, Invariant tori in the secular motions of the three-body planetary
systems, Celestial Mechanics and Dynamical Astronomy 78, 47–74 (2000)

[34] U. Locatelli, A. Giorgilli, Construction of the Kolmogorov’s normal form for a planetary
system, Regular and Chaotic Dynamics, 10, 153–171 (2005)

[35] U. Locatelli, A. Giorgilli, Invariant tori in the Sun–Jupiter–Saturn system, Discrete and
Continuous Dynamical Systems – Series B, 7, 377–398 (2007)

[36] Ch. Lhotka, C. Efthymiopoulos, R. Dvorak, Nekhoroshev stability at L4 or L5 in the elliptic

restricted three body problem – application to Trojan asteroids, MNRAS in press (2007)

[37] J. Moser, On invariant curves of area-preserving mappings of an annulus, Nach. Akad.

Wiss. Göttingen, Math. Phys. Kl. II 1, 1 (1962)

[38] P. Robutel, Stability of the planetary three–body problem. II. KAM theory and existence of
quasi–periodic motions, Celestial Mechanics and Dynamical Astronomy 62, no. 3, 219–261
(1995)

[39] P. Robutel, F. Gabern, The resonant structure of Jupiter’s Trojan asteroids - I. Long–term
stability and diffusion, MNRAS 372, n. 4, 1463–1482 (2006)

23