See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/307636645

Basic Celestial Mechanics

Article · September 2016

CITATIONS READS
0 1,003

1 author:

Sergei A Klioner
Technische Universität Dresden
190 PUBLICATIONS   3,434 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Gaia mission View project

Gaia data processing View project

All content following this page was uploaded by Sergei A Klioner on 31 October 2016.

The user has requested enhancement of the downloaded file.

 Lecture Notes on
Basic Celestial Mechanics
arXiv:1609.00915v1 [astro-ph.IM] 4 Sep 2016

Sergei A. Klioner

2011
Contents

1 Introduction 5

2 Two-body Problem 6
2.1 Equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Integrals of angular momentum and energy . . . . . . . . . . . . . . . . . . . 9
2.3 Possible Orbits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Orbit in Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.5 Kepler Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.6 Solving the Kepler Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.7 Hyperbolic and Parabolic Motion . . . . . . . . . . . . . . . . . . . . . . . . 20
2.8 Relation between Position, Velocity and the Kepler Elements . . . . . . . . . 23
2.9 Series Expansions in Two-Body Problem . . . . . . . . . . . . . . . . . . . . 25
2.9.1 Taylor expansions in powers of time . . . . . . . . . . . . . . . . . . . 25
2.9.2 Fourier expansions in multiples of the mean anomaly . . . . . . . . . 27
2.9.3 Taylor expansions in powers of the eccentricity . . . . . . . . . . . . . 28

3 The N-body problem 29

3.1 Equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Classical integrals of the N-body motion . . . . . . . . . . . . . . . . . . . . 30
3.3 The disturbing function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4 Overview of the three-body problem . . . . . . . . . . . . . . . . . . . . . . 33
3.5 Planetary ephemerides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4 Elements of the Perturbation Theory 39

4.1 The method of the variation of constants . . . . . . . . . . . . . . . . . . . . 39
4.2 Gaussian perturbation equations . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2.1 Derivation of differential equations for osculating elements . . . . . . 41
4.2.2 Discussion of the derived equations . . . . . . . . . . . . . . . . . . . 45
4.3 Lagrange equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

5 Three-body problem 50
5.1 The Lagrange solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.2 The restricted three-body problem . . . . . . . . . . . . . . . . . . . . . . . 50
5.3 Motion near the Lagrange equilibrium points . . . . . . . . . . . . . . . . . . 50

6 Gravitational Potential of an Extended Body 51

6.1 Definition and expansion of the potential . . . . . . . . . . . . . . . . . . . . 51

2
CONTENTS 3

6.1.1 Definition of the potential of an extended body . . . . . . . . . . . . 51

6.1.2 Legendre polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
6.1.3 Expansion of the potential . . . . . . . . . . . . . . . . . . . . . . . . 56
6.2 First terms of the expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.2.1 The term for n = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.2.2 The terms for n = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.2.3 The terms for n = 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.3 Symmetric bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.3.1 Axial symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.3.2 Axial symmetry and the symmetry about xy-plane . . . . . . . . . . 62
6.3.3 Spherical symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.3.4 Symmetry with respect to three coordinate planes . . . . . . . . . . . 63
6.4 Spherical functions and the classification of the coefficients . . . . . . . . . . 64

7 Satellite Motion 67
7.1 Typical perturbations in satellite motion . . . . . . . . . . . . . . . . . . . . 67
7.2 Motion in the quadrupole field . . . . . . . . . . . . . . . . . . . . . . . . . . 68
7.2.1 Disturbing potential due to J2 . . . . . . . . . . . . . . . . . . . . . . 68
7.2.2 Exact consequence of the axially symmetric perturbation . . . . . . . 69
7.2.3 Secular part of the disturbing potential . . . . . . . . . . . . . . . . . 70
7.2.4 Secular perturbations of osculating elements . . . . . . . . . . . . . . 72
7.2.5 Analysis of the secular perturbations . . . . . . . . . . . . . . . . . . 73
7.2.6 Additional remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
7.3 Atmospheric drag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
7.3.1 Model for the drag force . . . . . . . . . . . . . . . . . . . . . . . . . 75
7.3.2 Model for the air density . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.3.3 Gaussian perturbation equations in the axes aligned with the velocity
vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
7.3.4 Osculating elements for the air drag . . . . . . . . . . . . . . . . . . . 79
7.3.5 Averaged equations for the osculating elements . . . . . . . . . . . . 80
7.3.6 Averaged osculating elements for small eccentricities . . . . . . . . . 81
7.3.7 Discussion of the solution for osculating elements . . . . . . . . . . . 82

8 Numerical integration 84
8.1 Basic notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
8.2 Methods of numerical integration . . . . . . . . . . . . . . . . . . . . . . . . 84
8.3 Reliability of numerical integration . . . . . . . . . . . . . . . . . . . . . . . 84

Index 84

Bibliography 85
List of Figures

2.1 One body problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.2 General two body problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Orbit in space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.1 Lagrange points in the three-body problem . . . . . . . . . . . . . . . . . . . 35

6.1 N body system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

6.2 Extended body as an N body system . . . . . . . . . . . . . . . . . . . . . . 52
6.3 Definition of the angle H . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
6.4 Legendre polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.5 Spherical coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.6 Spherical functions on a sphere . . . . . . . . . . . . . . . . . . . . . . . . . 66

7.1 Precession of the pericenter . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

7.2 Precession of the longitude of the node . . . . . . . . . . . . . . . . . . . . . 75
7.3 Drag coefficients for some bodies . . . . . . . . . . . . . . . . . . . . . . . . 76
7.4 Air density as function of altitude . . . . . . . . . . . . . . . . . . . . . . . . 77
7.5 Relation between (S, T ) and (Fτ , Fn ) . . . . . . . . . . . . . . . . . . . . . . 79
7.6 Modified Bessel functions of the first kind . . . . . . . . . . . . . . . . . . . 82
7.7 The effect of the atmospheric drag on a satellite orbit . . . . . . . . . . . . . 83

4
Chapter 1

Introduction

Summary: Research field of celestial mechanics. Historical overview: apparent motion of planets,
and solar and lunar eclipse as impetus for celestial mechanics. Ancient celestial mechanics. Ap-
polonius and the idea of epicyclic motion. Ptolemy and the geocentric system. Copernicus and the
heliocentric system. Kepler and the three Kepler laws. Galileo: satellites of Jupiter as a model
for the Solar system, the begin of mechanics. Newton: mathematical formulation of mechanics,
gravitational force. Einstein: the problem of perihelion advance of Mercury and the general theory
of relativity.
Three aspects of celestial mechanics: physics of motion, mathematics of motion and (numerical)
calculation of motion. The astronomical objects and specific goals and problems of the modelling of
their motion: artificial satellites, the Moon, major planets, asteroids, comets, Kuiper belt objects,
satellites of the major planets, rings, interplanetary dust, stars in binary and multiple systems,
stars in star clusters and galaxies.

5
Chapter 2

Two-body Problem

2.1 Equations of motion

Summary: Equations of motion of one test body around a motionless massive body. Equations
of general two-body problem. Center of mass. Relative motion of two bodies. Motion relative to
the center of mass.

Let us first consider the simplest case: the motion of a particle having negligibly small
mass m in the gravitational field of a body with mass M (m ≪ M). Here we neglect the
influence of the smaller mass on the larger one and assume that the larger mass is at rest at
the origin of our coordinate system. Let r be the position of the mass m (Fig. 2.1). Then
according to the Newtonian law of gravity the force acting on the smaller mass reads
Mm r Mm
F =G 2
· = −G 3 r.
r r r
Here and below the absolute value of a vector is designated by the same symbol as the
vector itself, but not in boldface (e.g., r = |r|). In Newtonian mechanics force is equal to
the product of the mass and acceleration of the particle. Therefore, one has
d2 r
F =m = m r̈
dt2

Figure 2.1: One body problem. Body P with mass M is assumed to be at rest at the origin O
of the coordinate system. The motion of a test particle, that is, of a body with negligible mass
m ≪ M is then considered assuming that m is so small that its influence on the body with mass
M can be totally neglected. Position of that test particle is denoted by r.

6
CHAPTER 2. TWO-BODY PROBLEM 7

P1 r
P2

!1
!2

Figure 2.2: General two body problem. Bodies P1 and P2 with masses m1 and m2 have positions
ρ1 and ρ2 , respectively. The position of body P2 relative to body P1 is r = ρ2 − ρ1 . Point O is
the origin of the chosen coordinate system.

(a dot over a symbol denote the time derivative of the corresponding quantity and a double
dot the second time derivative), and finally the equations of motion of the mass m read
r
r̈ + GM = 0. (2.1)
r3
Let us now consider the general case of two bodies experiencing mutual gravitational
attraction. Let vectors ρ1 and ρ2 are the positions of bodies P1 and P2 with masses m1 and
m2 , respectively, in some coordinate system, and r = ρ2 − ρ1 is the position of body P2 with
respect to body P1 (Fig. 2.2). Then, the equations of motions of the two bodies read
r
m1 ρ̈1 = Gm1 m2 ,
r3
r
m2 ρ̈2 = − Gm1 m2 3 . (2.2)
r
This is a system of differential equations of order 12 (we have a differential equation of
order 2 for each of the 3 components of the two vectors ρ1 and ρ2 , the equations being
coupled to each other). Now, summing these two equations one gets that the following
linear combination of the position vectors remains zero at any moment of time:

m1 ρ̈1 + m2 ρ̈2 = 0. (2.3)

Since the masses are considered to be constant in our consideration, this equation can be
integrated twice:

m1 ρ̇1 + m2 ρ̇2 = A, (2.4)

m1 ρ1 + m2 ρ2 = A t + B, (2.5)

where vectors A and B are some arbitrary integration constants. Clearly these equations
express that the barycenter (center of mass) of the system of two bodies moves uniformly
CHAPTER 2. TWO-BODY PROBLEM 8

and rectilinear (that is, with a constant velocity proportional to A). The position of the
barycenter is
m1 ρ1 + m2 ρ2
R= ,
m1 + m2
so that one has
A
Ṙ = = const,
m1 + m2
A B
R= t+ .
m1 + m2 m1 + m2

Quantities which remain constant during the motion are called integrals of motion. Here
we have found 6 integrals of motion (3 components of A and 3 components of B). One often
can use the integrals of motion to reduce the order of the system of differential equations
describing the motion. Let us demonstrate that using (2.5) one can reduce the order of (2.2)
by 6. Two ways of thinking are possible here. First, let us consider the motion of body P2
relative body P1 . In this case we need an equation for r. Cancelling factors m1 and m2 in
the first and second equation of (2.2), respectively,
r
ρ̈1 = Gm2 ,
r3
r
ρ̈2 = − Gm1 3
r
and subtracting the resulting equations one gets
r
r̈ + G(m1 + m2 ) = 0. (2.6)
r3
This is a system of differential equations of order 6 ( components of r are defined by a system
of 3 coupled equations of order 2). Having a solution of this equation (that is, assuming that
r as a function of time is known) one has two linear equations for vectors ρ1 and ρ2 :

m1 ρ1 + m2 ρ2 = A t + B,
ρ2 − ρ1 = r.

The 6 constants A and B can be chosen arbitrarily (for example, computed from the initial
values for the positions ρ1 and ρ2 , and velocities ρ̇1 and ρ̇2 ).
Another possible way to use the integrals of motion (2.5) to reduce the order of (2.2) is
to consider the motion of each body relative to the barycenter. This corresponds to choosing
the coordinate system with the origin at the barycenter and setting A = 0 and B = 0. This
is always possible due to the Galilean relativity principle stating that coordinate systems
moving with a constant velocity relative to each other are equivalent and can be equally
used to describe the motion. From

m1 ρ1 + m2 ρ2 = 0,
ρ2 − ρ1 = r
CHAPTER 2. TWO-BODY PROBLEM 9

one has
m2
ρ2 = − r,
m1 + m2
m1
ρ1 = r.
m1 + m2
Substituting these two equations into (2.6) one gets two uncoupled equations for ρ1 and ρ2 :
m32 ρ1
ρ̈1 + G = 0,
(m1 + m2 )2 ρ31
m31 ρ2
ρ̈2 + G = 0. (2.7)
(m1 + m2 )2 ρ32
Note that the second equation in (2.7) can be derived from the first one by interchanging
the indices 1 ↔ 2.
Now we notice that in all cases considered above the equations of motion (2.1), (2.6) and
(2.7) have the form
r
r̈ + κ2 = 0, (2.8)
r3
where κ > 0 is a constant depending on the masses of the bodies. In the following we
consider the equations of motion in their generic form (2.8).

Remark. Note that the same equation (2.8) describes also the position of any body of a system
of N bodies with N > 2 when some special configuration (special set of positions and velocities) of
the N bodies is considered. Such a configuration must possess certain symmetry so that the sum
of all gravitational forces acting on each of the N bodies is always directed toward the center of
mass of the N -body system. Such configurations are called central configurations.

Exercise. Find some examples of the central configuration. Hint: consider the bodies at the
vertices of equilateral polygons.

2.2 Integrals of angular momentum and energy

Summary: Integral of angular momentum (the law of areas). The second Kepler’s law. Integral
of energy. Integrals of angular momentum and energy in polar coordinates.

The equations (2.8) can be further simplified by using the so-called integrals of area (or
angular momentum) and energy. Let us first compute the cross product of (2.8) by r:
 r
r × r̈ + κ2 3 = 0
r
which implies
r × r̈ = 0.
The latter equation can be integrated to give
r × ṙ = c, (2.9)
where c = const. Eq. (2.9) has two consequences:
CHAPTER 2. TWO-BODY PROBLEM 10

1. The motion is planar. Indeed, the constant vector c is orthogonal to both position
vector r and velocity vector ṙ at any moment of time. The last two vectors define the
momentary plane of motion and since c = const this plane does not change.

2. The area swept out by position vector r within an infinitely small interval of time
dt remains constant. Indeed, if at some moment of time t the position vector is r
then at time t + dt the position vector is r + ṙ dt, where ṙ is the velocity vector
at time t. The area encompassed by vectors r and r + ṙ dt can be calculated as
dA = 12 |r × (r + ṙ dt)| = 12 |r × ṙ| dt = 21 c dt. Therefore, Ȧ = 12 c = const. This is
nothing else than the second Kepler law in differential form (its usual integral form
∆A = 12 c ∆t immediately follows from the differential form Ȧ = 21 c). Let us remind
that the standard formulation of the second Kepler’s law states that “a line joining a
planet and the Sun sweeps out equal areas during equal intervals of time”.

Remark. We have used only one property of (2.8) in order to derive (2.9): the property that
the force is proportional to r. The coefficient of proportionality plays no role here and can be any
function of time t, position r, velocity ṙ. Such forces are called central forces. Motion with any
central forces is, therefore, planar and satisfies the second Kepler’s law.

Denoting the components of vectors as r = (x, y, z), ṙ = (ẋ, ẏ, ż), c = (cx , cy , cz ) one can
rewrite (2.9) in the form

y ż − ẏ z = cx ,
z ẋ − ż x = cy , (2.9 ′)
x ẏ − ẋ y = cz ,

Let us now use the fact that the motion is planar and re-orient our coordinates in such
a way that one of the body’s coordinate remain identically zero. This means that the body
remains in the plane z = 0 of the coordinate system. Let us then denote the two other
coordinates as ξ and η. In these new coordinates Eq. (2.9) reads

ξ η̇ − ξ˙ η = c, (2.10)

and the equations of motion can be rewritten as

ξ
ξ¨ + κ2 3 = 0,
r
2 η
η̈ + κ 3 = 0 (2.11)
r
p
where r = ξ 2 + η 2 .
Let us now multiply the first equation of (2.11) by 2 ξ̇ and the second one by 2 η̇, and
then add the two resulting equations to get

2ξ˙ ξ + 2η̇ η
2ξ˙ ξ¨ + 2η̇ η̈ = −κ2 .
r3
CHAPTER 2. TWO-BODY PROBLEM 11

Both sides of the latter equation are total time derivatives. Integrating this equation one
gets
1
ξ˙2 + η̇ 2 = 2κ2 + h, (2.12)
r
where h = const is a constant of integration. The validity of (2.12) can be checked by
calculating its derivative with respect to time t and comparing it with the previous equations.
Quantity h represents one more integral of motion which is called energy constant. Indeed,
the left-hand side of (2.12) is doubled kinetic energy of the body per unit of mass and the
right-hand side is minus doubled potential energy per unit of mass plus constant h.
Let us now introduce polar coordinates r and u instead of Cartesian coordinates ξ and
η. Using standard relations ξ = r cos u and η = r sin u which imply, for example,

ξ˙ = ṙ cos u − r sin u u̇,

η̇ = ṙ sin u + r cos u u̇

one gets the integrals of area (2.10) and of energy (2.12) in polar coordinates

r 2 u̇ = c, (2.13)

1
ṙ 2 + r 2 u̇2 = 2κ2 + h. (2.14)
r
The equations of motion (2.11) can be also expressed in polar coordinates r and u. One can
show that the only non-trivial equation reads
κ2
r̈ − r u̇2 + = 0. (2.15)
r2

Exercise. Rewrite the equations of motion (2.11) in polar coordinates r and u explicitly. Show
that that they can be expressed as a sum of derivatives of the integrals (2.13) and (2.14).

2.3 Possible Orbits

Summary: Conic sections as possible orbits in the two-body problem. The first Kepler’s law.
Definitions of the semi-latus rectum, eccentricity and the argument of pericenter. Apocenter and
pericenter. Apsidal line. Elliptical, parabolic, hyperbolic and rectilinear motions.

Our aim now is investigate the form of the orbits implies by (2.13) and (2.14). Since we
are interested in the form of the orbits only, we can eliminate the time variable from the
two equations. Eq. (2.13) implied that u̇ = c r −2. Therefore, ṙ = dtd r = du
dr du
dt
dr
= du c r −2 .
Substituting this into (2.14) one gets
 2 2 2
dr c 2 c 1
4
+ r 4
= 2κ2 + h
du r r r
CHAPTER 2. TWO-BODY PROBLEM 12

and, therefore,
 c 2  dr 2 1 c2
− 2 = 2κ2 + h − 2 .
r du r r
Now let us consider first the case c 6= 0. The previous equations can be rewritten in the
form
 2
dσ κ4
= h + 2 − σ2 ,
du c
where
c κ2
σ= − .
r c
Since the right-hand side of this equation is non-negative (as a square of a real number
4 4
dσ/du) one has also h + κc2 − σ 2 ≥ 0 or h + κc2 ≥ σ 2 . Considering that σ 2 ≥ 0 one gets
κ4
h+ ≥ 0.
c2
 1/2
κ4
Therefore, one can designate h + c2
= A, A ≥ 0.

Exercise. Prove that for any position r and velocity ṙ the integrals h and c take such values
that h + κ4 c−2 ≥ 0. Hint: use the definitions of h and c as functions of r and ṙ.

Therefore, one get the differential equation for the orbit

 2
dσ
= A2 − σ 2 . (2.16)
du
Let us first consider the case A 6= 0. One can consider that dσ/du > 0. The second case of
dσ/du < 0 can be derived from the first one by setting u = −u, that is by mirroring the first
case. It is clear, however, that the orbit in both cases remains the same and it is only the
direction in which the body moves along the orbit which changes. The direction of motion
is not interesting for us for the moment. Therefore, the solution can be written as
σ
u = arccos + ω,
A
or
σ = A cos(u − ω),
ω = const being an arbitrary constant. Taking into account the definitions of σ and A one
has
p
r= , (2.17)
1 + e cos(u − ω)
where p > 0 and e ≥ 0 represent two parameters of the orbit defined through the integrals
of motion h and c and parameter κ:
c2
p= , (2.18)
κ2
CHAPTER 2. TWO-BODY PROBLEM 13
r
c2
e= 1+h . (2.19)
κ4

Remark. One can see that formally Eq. (2.16) has one more solution: σ = ±A which means
that r = p/(1 ± e) is constant. Using the equations of motion in polar coordinates (2.15) one can
see that this is valid only when A = 0 (this case is treated below separately). Indeed, a solution
of equations of motion (e.g. a solution of (2.11)) must satisfy also the corresponding integrals of
motion (e.g. (2.13)–(2.14)), but not any solution satisfying the integrals of motion also satisfy the
equations of motion. That is, the integrals of motion are necessary, but not sufficient conditions
for a function to be a solution of the equations of motion. Whether a function satisfying of the
integrals of motion also satisfies the equations of motion should be checked explicitly. One can
easily see that (2.17) is really a solution of (2.15), but σ = ±A 6= 0 is not.

Eq. (2.17) shows that the orbit in this case (we assumed c 6= 0 and A 6= 0) is a conic section.
The parameter p > 0 is called semi-latus rectum and e represents the eccentricity of the
conic section. From (2.17) one sees that for e < 1 (this corresponds to h < 0) the orbit is
an ellipse, for e = 1 (h = 0) a parabola and for e > 1 (h > 0) a hyperbola. This proves the
first Kepler’s law: the orbit of every planet is an ellipse with the Sun at one of the two foci.
For any e the polar angle u can take the value u = ω. In this case the denominator of
(2.17) takes its maximal value 1 + e. Therefore, the radial distance r is minimal at this point
p
r= = rmin.
1+e
The point of the orbit where the distance r takes its minimal value is called pericenter
or periapsis (or perihelion when motion relative to the Sun is considered, or perigee when
motion relative to the Earth is considered, or periastron when motion of a binary star is
considered). The constant ω is called argument of pericenter.
For e < 1, polar angle u can also take the value u = ω + π (π = 3.14 . . .). Here the
distance r takes its maximal value
p
r= = rmax .
1−e
The point of the elliptic orbit where the distance r takes its maximal value is called apocenter
or apoapsis (or aphel when motion relative to the Sun is considered, or apogee when motion
relative to the Earth is considered, or apoastron when motion of a binary star is considered).
Pericenter and apocenter are called apsides. A line connecting pericenter and apocenter is
called line of apsides or apse line.
The mean distance of the body calculated as arithmetic mean of the maximal and minimal
values of r is called semi-major axis of the orbit:
1 p
a= (rmin + rmax ) =
2 1 − e2
or

p = a (1 − e2 ). (2.20)
CHAPTER 2. TWO-BODY PROBLEM 14

Substituting (2.18) and (2.19) into (2.20) one gets the relation between a and the integrals
of motion:
κ2
a=− . (2.21)
h
One can see that a depends only on κ and the energy constant h, and not on c. Eqs. (2.20)
and (2.21) represent definition of a for any non-negative value of e (or for any h and c).
From (2.21) it follows that a is infinite for parabolic motion (e = 1, h = 0) and negative for
hyperbolic one (e > 1, h > 0).
Let us consider now the two
remained cases. First, for c 6= 0 and A = 0 the differential
2
equation for the orbit reads dσdu
= −σ 2 , which means that both σ and dσ du
should be zero
and therefore σ ≡ 0. This means that
c κ2
− =0
r c
and
r = p. (2.22)
This solution coincides with (2.17) for e = 0 (this agrees also with the definition of e: if
A = 0, one has h = −κ4 /c2 and from (2.19) it follows that e = 0).
Finally, if c = 0 from (2.13) one gets u̇ = 0 and therefore u = const, which means that
the motion is rectilinear. Substituting this into the energy integral (2.14) one gets
1
ṙ 2 = 2κ2 + h.
r
For h < 0 the motion is bounded since both sides of the latter equations must be non-
2
negative. For negative h this means that r ≤ − 2κh > 0. This is rectilinear motion of
elliptical kind (the elliptical motion with (2.17) with h < 0 and e < 1 is also bounded in
space). For non-negative h one √ can calculate the velocity of the body for infinite distance
r → ∞: v∞ = limr→∞ ṙ = h. For h = 0 velocity goes to zero: v∞ = 0. For h > 0 the
velocity always remains positive: v∞ > 0. These are rectilinear motions of parabolic and
hyperbolic kinds, respectively.

Exercise. For parabolic case h = 0, one has ṙ 2 = 2κ2 1r . This equation has a simple analytical
solution. Find this solution in its most general form.

2.4 Orbit in Space

Summary: Three Euler angles defining the orientation of the orbit in space: longitude of the
ascending node, inclination and the argument of pericenter. The rotational matrix between inertial
coordinates in space and the coordinates in the orbital plane.

Let us now consider the orientation of the orbit in space. Above we have seen that the
orbit lies in a plane perpendicular to vector c. Let us consider two orthogonal Cartesian
coordinate systems: (1) (x, y, z) is some arbitrary inertial reference system where the equa-
tions of motion (2.8) are initially formulated and (2) (X, Y, Z) is oriented in such a way that
CHAPTER 2. TWO-BODY PROBLEM 15

XY -plane contains the orbit (that is, axis Z is parallel to vector c), and the pericenter of the
orbit lies on axis X. The origins of both reference systems coincide and the transformation
between these coordinates is a pure three-dimensional (time-independent) rotation. The ro-
tation can be parametrized in a multitude of ways. Historically, it was parametrized by three
Euler-type angles. Let us first consider the points where the orbit intersects the xy-plane.
These points are called nodes. The node at which the body, in course of its motion, proceeds
from the area of negative z to that of positive z is called ascending node. Let us introduce
an intermediate coordinate system (x1 , y1, z1 ) that is obtained from (x, y, z) by a rotation
around axis z = z1 so that axis x1 contains the ascending node of the orbit:
   
x1 x
 y1  = Az (Ω)  y  , (2.23)
z1 z
where Az is the rotational matrix around z-axis
 
cos α sin α 0
Az (α) =  − sin α cos α 0  . (2.24)
0 0 1
Angle Ω is the called the longitude of the ascending node or simply longitude of the node.
One more intermediate system (x2 , y2 , z2 ) is obtained from (x1 , y1, z1 ) by a rotation around
axis x1 = x2 so that the direction of axis z2 coincides with vector c:
     
x2 x1 x
 y2  = Ax (i)  y1  = Ax (i) Az (Ω)  y  , (2.25)
z2 z1 z
where Ax is the rotational matrix around x-axis
 
1 0 0
Ax (α) =  0 cos α sin α  . (2.26)
0 − sin α cos α
Angle i is called inclination. Two angles – longitude of the node Ω and inclination i –
fully define the orientation of the orbital plane in space. The orbit lies in the x2 y2 -plane.
Coordinates (x2 , y2) coincides with coordinates (ξ, η) used above. The last step is to define
the orientation of the orbit in the orbital plane. This is done by using argument of pericenter
ω. The final coordinate system (X, Y, Z) can be obtained from (x2 , y2, z2 ) by a rotation
around axis z2 :
     
X x2 x
 Y  = Az (ω)  y2  = Az (ω) Ax (i) Az (Ω)  y  . (2.27)
Z z2 z
In the following the inverse transformation plays an important role:
   
x X
y  = P Y  , (2.28)
z Z
CHAPTER 2. TWO-BODY PROBLEM 16

P
c

v
P

y
i
W

x N

Figure 2.3: Orbit in space. The picture shows the definition of three Euler-like angles Ω, i and ω
that fully describe the orientation of the orbit of a two-body problem in space. N is the ascending
node, Π is the pericenter, P is the position of the body. One more angle (true anomaly v) defines
the position of the body on the orbit.

P = ATz (Ω) ATx (i) ATz (ω) . (2.29)

Here superscript T denotes the transpose of the corresponding matrix. Note that for any
rotational matrix RT = R−1 . Explicitly one has:
 
cos Ω cos ω − sin Ω cos i sin ω − cos Ω sin ω − sin Ω cos i cos ω sin Ω sin i
P =  sin Ω cos ω + cos Ω cos i sin ω − sin Ω sin ω + cos Ω cos i cos ω − cos Ω sin i  .
sin i sin ω sin i cos ω cos i
(2.30)

2.5 Kepler Equation

Summary: True Anomaly. Kepler equation in true anomaly. Eccentric anomaly. Various relations
between the true and eccentric anomaly. Kepler equation in eccentric anomaly. Mean anomaly.
The period of motion and the third Kepler law.

The true anomaly v is defined as v = u − ω. Since ω = const one has v̇ = u̇. Therefore,
integral of areas (2.13) can be written as
√ 1
v̇ = c = κ p 2.
r
Using (2.17) one gets
dv
2
= κ p−3/2 dt , (2.31)
(1 + e cos v)
CHAPTER 2. TWO-BODY PROBLEM 17

which after integration gives the Kepler equation in true anomaly

Z
−3/2 dv
κp t= + const .
(1 + e cos v)2
In this Section from now on we consider only the case of elliptical motion with e < 1.
The integral above cannot be computed analytically. In order to simplify the computations
one introduces the so-called eccentric anomaly. The eccentric anomaly E is defined by its
relation to the true anomaly v:
cos E − e
cos v = , (2.32)
1 − e cos E
It is easy to calculate sin2 v from this equation. Choosing the sign so that E = v for e = 0
one gets
√
1 − e2 sin E
sin v = . (2.33)
1 − e cos E
Solving (2.32) for cos E one gets
cos v + e
cos E = , (2.34)
1 + e cos v
so that
1 − e2
1 − e cos E = , (2.35)
1 + e cos v
which together with (2.33) gives
√
1 − e2 sin v
sin E = . (2.36)
1 + e cos v
It is easy to see that (2.34) and (2.36) can be obtained from (2.32) and (2.33) by making
the substitution (E, v, e) −→ (v, E, −e). From (2.32) and (2.33) one can derive a relation
between v and E which is convenient for numerical calculations
r
v 1+e E
tan = tan . (2.37)
2 1−e 2
From this equation it is easy to see that E and v are equal at apsides.
Now let us express the Kepler equation (2.31) in terms of the eccentric anomaly. Taking
a derivative of (2.35) one gets
(1 − e2 ) sin v
sin E dE = dv,
(1 + e cos v)2
which together with (2.33) gives
dv
= (1 − e2 )−3/2 (1 − e cos E) dE.
(1 + e cos v)2
CHAPTER 2. TWO-BODY PROBLEM 18

Substituting this into (2.31) one gets

(1 − e cos E) dE = κ a−3/2 dt (2.38)

or after integrating

E − e sin E = M, (2.39)

where M = n(t − T ) is the mean anomaly, n = κ a−3/2 is the mean motion, and T is an
integration constant representing the moment time at which M = 0. It is easy to see from
(2.39) and (2.37) that for t = T one has M = E = v = 0 and the body is situated in
its pericenter. However, for e = 0 the orbit is circular so that any point can be declared
to be pericenter. That is why, the definition of the mean anomaly is often taken to be
M = M0 + n(t − t0 ), where M0 is the value of the mean anomaly at some moment t = t0 .
The period of motion P an be defined as a time interval between two successive pericenters
(M = 0 and M = 2π). Then it is clear that P = 2π n
, which can be re-written as
P2 4π
3
= 2 = const.
a κ
This is the third Kepler’s law: the square of the orbital period of a planet is directly pro-
portional to the cube of the semi-major axis of its orbit. However, the constant κ depends
on the masses of both bodies of the two-body problem. Considering the motion of two
planets relative to the Sun in the framework of the two-body problem we have two differ-
ent constants κi entering the corresponding equations of motion of each of the two planets
κ2i = G (mSun + mi ). Hence, one gets the third Kepler’s law in its correct form
 2  3
P1 a1 mSun + m1
=
P2 a2 mSun + m2
In the Solar system mi /mSun < 10−3 and the last factor is almost unity.

Using the eccentric anomaly E it is easy to calculate the position and velocity of the
object in two-body motion. For the position one has (here we use the known expressions of
the true anomaly v and relation (2.32)–(2.33)):
p
r= = a (1 − e cos E), (2.40)
1 + e cos v
X = r cos v = a (cos E − e), (2.41)
√
Y = r sin v = a 1 − e2 sin E. (2.42)
Differentiating (2.41)–(2.42) with respect to time t and using that
n
Ė = (2.43)
1 − e cos E
(the latter equation can be derived by differentiating the Kepler equation (2.39)) one gets
a n sin E
Ẋ = − , (2.44)
1
√ − e cos E
a 1 − e2 n cos E
Ẏ = . (2.45)
1 − e cos E
CHAPTER 2. TWO-BODY PROBLEM 19

We have now all formulas that are necessary to compute the position and velocity of a body
in two-body motion. These calculations can be performed if eccentric anomaly E is known
as a function of time t. The relation between E and t is given by the Kepler equation (2.39).
However, the latter equation is transcendent and cannot be solved algebraically. Let us turn
to the analysis of the Kepler equation.

2.6 Solving the Kepler Equation

Summary: Existence and uniqueness of the solution. Iterative solution. Newtonian solution.

Let us confine ourselves by the case of elliptic motion with 0 ≤ e < 1. The Kepler
equation E − e sin E = M can be considered as an implicit function E(M) or as explicit
function M(E) (e being a parameter in both cases).

1) E(M) is a continuous function as an inverse function to a continuous function M(E).

2) Since |E − M| = e | sin E| < 1 one has

lim M(E) = +∞,

E→+∞
lim M(E) = −∞.
E→−∞

From properties 1 and 2 follows that for any M there exists at least one E such that the
Kepler equation is satisfied.
dM
3) Since = 1−e cos E > 0, M(E) is monotone, and therefore, E(M) is also monotone
dE
(as an inverse function of M(E)).

From these three properties it follows that for any M there exists only one E such that
E − e sin E = M.
Let us now consider how one can solve the Kepler equation. Generally we have a tran-
scendent equation

f (x) = 0, (2.46)

that should be solved numerically. Moreover, we just have proved that in case of the Kepler
equation one has only one solution for any 0 ≤ e < 1 and any M. Many numerical methods
to find the solution are known. Let us consider two simplest methods which, however, are
sufficient in many cases.

I. Iterations
The method consists in starting with some initial value for x (say x0 ) and iterating
the formula xi+1 = xi − f (xi ) until the subsequent values of x (xi and xi+1 ) are close
enough to each other: if |xi+1 − xi | < ε, then xi+1 is a solution of f (x) = 0 such that
|f (xi )| < ε. Convergence of the iterative sequence is guaranteed if the derivative f ′ (x)
of f (x) satisfies the inequality |f ′ (x) − 1| < 1 (Schwarz, 1993). One can easily show
CHAPTER 2. TWO-BODY PROBLEM 20

that this inequality holds in case of the Kepler equation with e < 1. In general, the
convergence also depends on the starting point x0 .
For the Kepler equation one gets f (E) = E − e sin E − M and the algorithm can be
written as

E0 = M,
Ei+1 = M + e sin Ei , i = 0, 1, . . .

One can prove that the initial condition E0 = M guarantees that the iterations converge
for any e and M.

II. Newton’s method

Another well-known method is the Newton’s (or Newton-Raphson) one. Again starting
from some initial value for the root’s estimate x0 one iterates xi+1 = xi − f (xi )/f ′ (xi ),
where f ′ (x) is the derivative of f (x). Again if |xi+1 − xi | < ε, then xi is a solution of
f (x) = 0 with the corresponding accuracy. Convergence of the iterative sequence is
guaranteed if the initial guess x0 lies close enough to the root (Section 9.4 of Press et
al. 1992).
For the Kepler equation this implies the scheme

E0 = M,
M + e (sin Ei − Ei cos Ei )
Ei+1 = , i = 0, 1, . . .
1 − e cos Ei

Again one can prove that the initial condition E0 = M guarantees that the iterations
converge for any e and M. It is well known that if the initial guess of the rule is good
enough, the Newton’s method converges much faster than the iteration method.

2.7 Hyperbolic and Parabolic Motion

Summary: The eccentric anomaly and the Kepler equation for the hyperbolic motion. Explicit
solution for the parabolic motion.

When we introduced the eccentric anomaly E and discussed the Kepler equation above
we have concentrated on elliptic motion with eccentricity 0 ≤ e < 1. Let us now consider
two other cases: hyperbolic motion with e > 1 and parabolic one with e = 1. The formula
for the form of the orbit
p
r=
1 + e cos v
remains valid for any e. Let us first consider the case of hyperbolic motion with e > 1.
2
The semi-latus rectum is non-negative p = a(1 − e2 ) = κc 2 > 0 (the case c = 0 leads to
the rectilinear
q motion, has been considered above and will not be considered here). But
2
e= 1 + h κc 4 and, therefore, e > 1 implies h > 0, i.e. the total energy of the two-body
CHAPTER 2. TWO-BODY PROBLEM 21

2
system is positive. Since a = − κh , h > 0 implies in turn that the semi-major axis is negative
a < 0.
If one takes the transformations (2.34) and (2.36) relating sine and cosine of the eccentric
one E to the sine and cosine of the true anomaly v, it is easy to see that sin E is imaginary
and cos E is real, but can exceed unity. Therefore, for hyperbolic motion E is imaginary. One
can continue in this way and work directly with complex numbers, but it is not convenient.
We, therefore, try to make all our equations real again (as it is in the case of elliptical
motion). To this end we define a new anomaly instead of E. Since
◦
eı x = cos x+ ı sin x,
◦

◦
ı being imaginary unit, one has
 ◦   ◦ 
1 ı x
◦
−ıx 1 ı x
◦
−ı x
cos x = e +e , sin x = ◦ e − e ,
2 2ı
◦
and for x = ı y one gets
◦
cos ı y = cosh y,
◦ ◦
sin ı y = ı sinh y.
◦
Using these formulas one can introduce a new anomaly H defined as H = ı E, so that

cos E = cosh H,
◦
sin E = − ı sinh H.

This allows one to re-write Eq. (2.40)–(2.42) as

p
r= = a (1 − e cos E) = |a| (e cosh H − 1), (2.47)
1 + e cos v
X = r cos v = a (cos E − e) = |a| (e − cosh H), (2.48)
√ √
Y = r sin v = a 1 − e2 sin E = |a| e2 − 1 sinh H, (2.49)

and the Kepler equation (2.39) as

e sinh H − H = Mhyp , (2.50)

Mhyp = κ |a|−3/2 (t − T ). (2.51)

The signs in (2.49) and (2.50) are chosen so that the body moves in the positive direction of
the Y -axis for t = T . Eq. (2.50) is the Kepler equation for hyperbolic motion. It has only
one solution for any value of mean anomaly Mhyp .
Let us now consider the simple case of parabolic motion. Parabolic motion corresponds
to e = 1 and this implies that the total energy of the system is zero: h = 0. The equation
for the form of the orbit can be simplified to
p
r= = q (1 + σ 2 ),
1 + cos v
CHAPTER 2. TWO-BODY PROBLEM 22

where σ = tan v2 and q = p/2 is the perihelion distance of the parabolic orbit (since in the
perihelion one has v = 0 and r = p/2). From the definition of σ one has
2σ 1 − σ2
sin v = , cos v = .
1 + σ2 1 + σ2
Therefore, the distance and coordinates of the body on a parabolic orbit can be written as

r = q (1 + σ 2 ), (2.52)
X = r cos v = q (1 − σ 2 ), (2.53)
Y = r sin v = 2 q σ. (2.54)

Now, using the integral of angular momentum one gets

√ p
r 2 v̇ = c = κ p = κ 2 q

and, therefore,
√
v̇ = κ 2 q −3/2 (1 + σ 2 )−2 . (2.55)

Since σ = tan v2 we have

dσ 1
= (1 + σ 2 ).
dv 2
Integrating this equation and using (2.55) one gets
1 3
σ + σ = Mpar , (2.56)
3
κ
Mpar = √ q −3/2 (t − T ). (2.57)
2

Eq. (2.56) is the Kepler equation for parabolic motion. This equation can be solved analyt-
ically using, e.g., the well-known Kardan formulas. Indeed, the equation has only one real
solution for any value of Mpar
1 1/3
σ= Q − 2 Q−1/3 ,
2 q
Q = 12 Mpar + 4 2 .
4 + 9 Mpar (2.58)

Note that for any Mpar value of Q remains positive. This means that the parabolic motion
can be represented by an explicit analytical formula (this can be also done for circular motion
with e = 0 and for rectilinear motion of parabolic type with c = 0 and h = 0).

Exercise. Write the explicit formulas for the coordinates for the case e = 0.
CHAPTER 2. TWO-BODY PROBLEM 23

2.8 Relation between Position, Velocity and the Ke-

pler Elements
Summary: Calculation of position and velocity from the Kepler elements. Calculation of the
Kepler elements from the position and velocity. Orbit determination (an overview).

We have seen above that there are two equivalent ways to represent a particular two body
motion: (1) to specify the initial conditions for the equation of motion, i.e. the position and
velocity vectors r = (x, y, z) and ṙ = (ẋ, ẏ, ż) together with the corresponding moment of
time t0 and the parameter κ, and (2) to fix the whole set of the six Kepler elements a, e, i,
ω, Ω, M0 = M(t0 ) again together with the moment of time t0 for which the mean anomaly
M0 is supposed to be known and the parameter κ. Very often in the practical calculation
one wants to switch between these two representations, that is to transform the position
and velocity into the corresponding Kepler elements or vice verse. Here we give the set of
formulas enabling one to perform these transformations for the case of elliptic motion.
The transformation from the Kepler elements to the position and velocity vectors can be
done in the following way:

1. calculate mean motion as n = κ a−3/2 and mean anomaly as M = n (t − t0 ) + M0 (here

the position and velocity vectors can be calculated for any arbitrary moments of time
t, not necessarily for the moment t0 for which the mean anomaly M0 is specified),

2. calculate eccentric anomaly E from E − e sin E = M,

3. calculate position and velocity in the orbital plane:

X = a (cos E − e),
√
Y = a 1 − e2 sin E,
a n sin E
Ẋ = − ,
1 − e cos E
√
a n 1 − e2 cos E
Ẏ = ,
1 − e cos E

4. calculate the position and velocity vectors in space as

   
x X
y =P  Y ,
z 0

   
ẋ Ẋ
 ẏ  = P  Ẏ  ,
ż 0

where P = ATz (Ω) ATx (i) ATz (ω) defined by (2.30).

CHAPTER 2. TWO-BODY PROBLEM 24

The transformation from the position and velocity vectors to the Kepler elements is a bit
more complicated and can be done as follows:

1. from the integrals of the areas r × ṙ = c one gets c = (cx , cy , cz )

cx = y ż − ẏ z,
cy = z ẋ − ż x,
cz = x ẏ − ẋ y,
q
c = |c| = c2x + c2y + c2z .

Then the semi-latus rectum can be calculated as

c2
p = 2, (2.59)
κ
and from
     
cx 0 c sin i sin Ω
 cy  = ATz (Ω) ATx (i)  0  =  −c sin i cos Ω 
cz c c cos i

which gives us three equations. The third equation can be written as

cz
cos i = (2.60)
c
and since 0 ≤ i ≤ π this one equation is sufficient to calculate the inclination i. The
two other equations read
cx
sin Ω = p 2 ,
cx + c2y
cy
cos Ω = − p 2 (2.61)
cx + c2y

and allow one to calculate Ω. Note that if c2x + c2y = 0, the inclination i = 0 and Ω is
not defined.
2. From Eq. (2.17) and v = u − ω one gets two equations
p
e cos v = − 1,
r√
p r · ṙ
e sin v = (2.62)
κ r
which can be used to calculate both the eccentricity e and the true anomaly v. Then
using (2.37) one can calculate the eccentric anomaly E, and from (2.39) the mean
anomaly M. All these values of anomalies v, E and M correspond to time t0 for which
the position and velocity of the body is specified. Finally, from p and e it is easy to
−1
calculate the semi-major axis as a = p (1 − e2 ) ;
CHAPTER 2. TWO-BODY PROBLEM 25

3. From
   
x r cos(v + ω)
 y  = ATz (Ω) ATx (i)  r sin(v + ω) 
z 0

one gets
x y
cos(v + ω) = cos Ω + sin Ω,
r x r
y  z
sin(v + ω) = − sin Ω + cos Ω cos i + sin i. (2.63)
r r r
From these two equations one calculates the angle v + ω and since v is known, the
argument of perihelion ω.

2.9 Series Expansions in Two-Body Problem

Summary: Series in powers of time. Fourier series in multiples of the mean anomaly. Series in
powers of the eccentricity.

As we have seen above a fully analytical solution of the two-body problem is impossible:
one has a transcendent Kepler equation cannot be solved analytically. The only possibility
to get an analytical solution is to use some kind of expansions. Below we consider three
types of expansions which are widely used in celestial mechanics.

2.9.1 Taylor expansions in powers of time

The first kind of expansion is the Taylor expansion in powers of time. Let us consider the
positional vector r and expand it into Taylor series
X∞
1 (k)
r(t + τ ) = r (t) τ k , (2.64)
k=0
k!

where r (k) are the derivatives of r of order k. For k = 0 and k = 1 they represent the initial
conditions for the motion r (0) (t) = r(t) and r (1) (t) = ṙ(t). Using the equation of motion
r
r̈ = −κ2
r3
it is clear that the higher derivatives for k ≥ 2 can be calculated in terms of r and ṙ. E.g.,
r (2) = r̈ = −κ2 S −3/2 r,
3 2 −5/2
r (3) = κ S Ṡ r − κ2 S −3/2 ṙ, (2.65)
2
and so on. Here S = r 2 , Ṡ = 2 r · ṙ, and the second- and higher-order derivatives of S
appearing in r (k) for k ≥ 4 can be calculated using
S̈ = 2 h + 2κ2 S −1/2 ,
CHAPTER 2. TWO-BODY PROBLEM 26

2
where h = ṙ · ṙ − 2κr is the energy integral. Therefore, it is clear that r(t + τ ) can be
represented as a linear combination of r(t) and ṙ(t)

r(t + τ ) = F r(t) + G ṙ(t), (2.66)

while the functions F and G can be expanded in their Taylor series in powers of τ :
X∞
1
F = Fk τ k ,
k=0
k!

X∞
1
G= Gk τ k . (2.67)
k=0
k!

The coefficients Fk and Gk are functions of κ, h, S and Ṡ only, and, therefore, can be
calculated from the initial conditions r(t) and ṙ(t). Comparing (2.66)–(2.67) to (2.64) one
gets for any k

r (k) = Fk r + Gk ṙ. (2.68)

Taking the derivative of (2.68) and comparing again with (2.68) written for k + 1

r (k+1) = Fk+1 r + Gk+1 ṙ

one gets the recursive formulas for Fk and Gk :

Fk+1 = Ḟk − κ2 S −3/2 Gk , (2.69)

Gk+1 = Fk + Ġk . (2.70)

The initial conditions for (2.69)–(2.70) can be derived by considering the zero-order expansion
r(t + τ ) = r(t) + O(τ ):

F0 = 1, (2.71)
G0 = 0. (2.72)

Using (2.69)–(2.70) with (2.71)–(2.72) one gets, for example,

F1 = 0,
G1 = 1,
F2 = −κ2 S −3/2 ,
G2 = 0,
... (2.73)

A detailed analysis of the two-body function by means of the complex analysis shows that the
convergence of the derived series is guaranteed only for |τ | smaller than some limit depending
on parameters of motion:
1 3/2
|τ | < q α(e), (2.74)
κ
CHAPTER 2. TWO-BODY PROBLEM 27

where q is the perihelion distance, e is the eccentricity and

  √ 
−3/2 1 + 1 − e2 √
 (1 − e) log − 1 − e2 , e ≤ 1
α(e) =  e (2.75)
√ √

(e − 1)−3/2 e2 − 1 − arctan e2 − 1 , e > 1

Note that α(e) is a continuous monotone function for e ≥ 0 and

lim α(e) = ∞,
e→0
√
2 2
lim α(e) = ,
e→1 3
lim α(e) = 0.
e→∞

This means that the convergence is guaranteed for any τ if and only if the eccentricity of
the orbit is zero and that the higher the eccentricity is the lower is the maximal τ for which
the series in powers of time converge.

2.9.2 Fourier expansions in multiples of the mean anomaly

It is well known that any continuous complex function f (x) of a real argument x with a
period of 2π (i.e. f (x + 2π) = f (x) for any x) can be expanded into Fourier series
+∞
X ◦
f (x) = fk eı k x ,
k=−∞
Z 2π
1 ◦
fk = f (x) e−ı k x dx, (2.76)
2π 0

which converges for any x. This kind of expansions can also be applied to the two-body
problem. If f (x) has additional properties (i.e., real or odd) formula (2.76) can be simplified.
A nice overview of all special cases can be found in Chapter 12 of Press et al. (1992).
Let us consider function f (E) = e sin E. This function is obviously real and odd
(f (−E) = −f (E)). Therefore, the Fourier expansion can be simplified to be
∞
X
E = M + e sin E = M + ak sin kM, (2.77)
k=1

Z π
2 2
ak = e sin E sin kM dM = Jk (k e), (2.78)
π 0 k
where Jn (x) are the Bessel functions of the first kind defined as
Z
1 π
Jn (x) = cos (n θ − x sin θ) dθ. (2.79)
π 0
Many properties of Jn (x) can be found e.g. in Chapter 9 of Abramowitz & Stegun (1965).

Exercise. Prove the second equality in (2.78) by taking the integral by parts.
CHAPTER 2. TWO-BODY PROBLEM 28

Therefore,
X∞
2
E =M + Jk (k e) sin kM. (2.80)
k=1
k

To give one more example let us note that

a 1
= Ė .
r n
Therefore, taking a derivative of (2.80) one gets
X ∞
a
=1+ 2 Jk (k e) cos kM. (2.81)
r k=1

In general one has

 r n ◦
∞
X ◦
eı m v = Xkn,m(e) eı k M , (2.82)
a k=−∞

where Xkn,m (e) are a three-parametric family of functions called Hansen coefficients.

2.9.3 Taylor expansions in powers of the eccentricity

The third kind of expansions are series in powers of eccentricity e. Let us consider these
series for the example of the eccentric anomaly. Re-writing the Kepler equation in the form
E = M + e sin E, (2.83)
one has iteratively
E = M + O(e),
E = M + e sin(M + O(e)) = M + e sin M + O(e2 ),
1
E = M + e sin(M + e sin M + O(e2 )) = M + e sin M + e2 sin 2M + O(e3 ),
2
...
Here we used the expansion sin(M + e sin M) = sin M + 21 e sin 2M + O(e2 ). Note that at
each step of the iteration the expansion for E derived on the previous step in substituted
under sinus in the right-hand side of (2.83) and the sinus is expanded in powers of e to the
corresponding order. In general one can write
∞
X
E=M+ ak (M) ek , (2.84)
k=1

and the coefficients a1 and a2 have been explicitly calculated above. Further coefficients can
be calculated by the same iterative scheme. The series in powers of e converge for all e lower
than the so-called Laplace limit:
0 ≤ e < e∗ = 0.6627434193492 . . . (2.85)
Chapter 3

The N-body problem

3.1 Equations of motion

Summary: Equations of motion of the N-body problem. Gravitational potential.

Let us consider N bodies having positions ρi , i = 1, . . . , N in an inertial reference system

and characterized by their masses mi . Here index i enumerates the bodies. Introducing the
position of body j with respect to body i as ρij = ρj − ρi one gets the equations of motion
of such a system
N
X mi mj
mi ρ̈i = G ρij (3.1)
j=1, j6=i
ρ3ij
or
N
X mj
ρ̈i = G ρ . (3.2)
j=1, j6=i
ρ3ij ij
These equations can also be be written in another form:
mi ρ̈i = gradi U (3.3)
where gradi U is the vector of partial derivatives of U with respect to the components of ρi .
Denoting ρi = (xi , yi , zi ) for any function f one defines gradi f as a vector with the following
components
 
∂ ∂ ∂
gradi f = f, f, f . (3.4)
∂xi ∂yi ∂zi
The potential U of N gravitating bodies reads
N N N Xi−1
1 X X G mi mj X G mi mj
U= = . (3.5)
2 i=1 j=1, j6=i ρij i=1 j=1
ρij
Since gradient of U can be written as
N
X XN
1 G mk mj
gradk U = G mk mj gradk = 3
ρkj
j=1, j6=k
ρkj j=1, j6=k
ρkj

1 ρkj
(the last equality uses that gradk = 3 ), it can be seen that (3.3) really holds.
ρkj ρkj

29
CHAPTER 3. THE N-BODY PROBLEM 30

3.2 Classical integrals of the N -body motion

Summary: Center of mass integral in the N -body. Integral of angular momentum in the N -body
problem. Integral of energy in the N -body problem.

The equations of motion of the N-body problem possess similar 10 integrals of motion
that we already discussed for the two-body problem. Summing up the equations (3.1) one
sees that
XN
mi ρ̈i = 0
i=1

Since the masses mi are constant this leads to

N
X
mi ρ̇i = A = const (3.6)
i=1

and
N
X
mi ρi = At + B, B = const . (3.7)
i=1

Components of A and B are six center of mass integrals in the N-body problem. These
are fully analogous to (2.3)–(2.5). The position of the center of mass of the N-body system
obviously read PN
i=1 mi ρi
R= P N
i=1 mi

Eq. (3.1) also implies

N
X
mi ρi × ρ̈i = 0.
i=1

Integrating one gets three more integrals – integrals of angular momentum:

N
X
mi ρi × ρ̇i = C = const. (3.8)
i=1

The plane perpendicular to vector C remains time-independent (since C is constant). This

plane is called invariant plane of the N-body system or Laplace plane. In the Solar system
the invariant plane lies close to the orbital plane of Jupiter.
Finally, summing up scalar products of each of the equations (3.3) with ρ̇i one gets
N
X N
X dU
mi ρ̇i · ρ̈i = gradi U · ρ̇i = (3.9)
i=1 i=1
dt

On the other hand, the left-hand side of this equation can be written as
N
X dT
mi ρ̇i · ρ̈i = ,
i=1
dt
CHAPTER 3. THE N-BODY PROBLEM 31

where
N
1 X
T = mi ρ̇i · ρ̇i .
2 i=1
Since both sides of (3.9) represent full derivatives one can integrate the equation to get the
integral of energy in the N-body problem
N N N
1 X 2 1 X X G mi mj
mi ρ̇i − = H = const , (3.10)
2 i=1 2 i=1 ρij
j=1, j6=i

where H is the total (mechanical) energy of the system of N gravitating bodies.

These 10 integrals can be used to decrease the order of the system (3.2) or to check the
accuracy of numerical integrations. One often uses barycentric coordinates of the N-body
system in which A = 0 and B = 0. In this case (3.6)–(3.7) can be used to compute the
position and velocity of one arbitrary body if the positions and velocities of other N − 1
bodies are known. This procedure can be used to compute initial conditions satisfying
(3.6)–(3.7) with A = 0 and B = 0. Alternatively, one body can be completely eliminated
from the integration, so that at each moment of time position and velocity for that body
are calculated using (3.6)–(3.7) with A = 0 and B = 0, and the corresponding equation is
excluded from (3.2) or (3.1). Four remaining integrals are usually used to check the accuracy
of the integration, the integral of energy is especially sensitive to numerical errors of usual
(non-symplectic) integrators.

3.3 The disturbing function

Summary: Planetary motion as perturbed two-body motion. The planetary disturbing function.

If the mass of one body is much larger than other masses in the N-body system it is
sometimes advantageous to write the equations of motion in a non-inertial reference system
centered on that dominating body. In solar system the Sun is obviously dominating, having
the mass about 1000 times larger than the planets.
Let us consider a system of N + 1 body numbered from 0 to N. Suppose that the mass
of body 0 is much larger than the masses of all other bodies mi ≪ m0 for i = 1, . . . , N. The
equations of motion (3.2) for bodies 0 and i can be written as
N
X N
X
G mj ρ0j G mi ρ0i G mj ρ0j
ρ̈0 = = + ,
j=1
ρ30j 3
ρ0i ρ30j
j=1, j6=i

XN XN
G mj ρij G m0 ρi0 G mj ρij
ρ̈i = 3
= 3
+ 3
. (3.11)
j=0, j6=i
ρij ρi0 j=1, j6=i
ρij

Let us designate the position of body i relative to body 0 as r i ≡ ρ0i = ρi − ρ0 . Then

subtracting two equations in (3.11) one has

XN  
G m0 r i G mi r i G mj r ij G mj r j
r̈ i = ρ̈i − ρ̈0 = − − + − , (3.12)
ri3 ri3 j=1, j6=i
rij3 rj3
CHAPTER 3. THE N-BODY PROBLEM 32

where r ij ≡ r j − r i = ρij . Finally, the equations of motion of body i with respect to body
0 can be written as
XN  
G (m0 + mi ) r i r ij rj
r̈ i + = G mj − 3 . (3.13)
ri3 j=1, j6=i
rij3 rj

This equation coincide with the equations of motion (2.8) of two-body problem if the right-
hand side is zero (e.g., for mj = 0, 1 ≤ j ≤ N, j 6= i). The right-hand side can be considered
as a perturbation of two-body motion. The same equations of motion can be rewritten in
the form
G (m0 + mi ) ri
r̈i + = gradi R , (3.14)
ri3
XN  
1 rj · ri
R= G mj − . (3.15)
rij rj3
j=1, j6=i

These equations can be directly integrated numerically or analyzed analytically to obtain

the motion of planets and minor bodies with respect to the Sun. It is also clear that if we
have only two bodies, R vanishes and the remaining equations of motion describe two-body
problem. Therefore, the forces coming from R can be considered as perturbations of two-
body problem. These perturbations are small in the case when m0 + mi ≫ mj and the
heliocentric motion of body is close to the solution of two-body problem. For this reason, R
is called planetary disturbing function. The idea to treat any motion of a dynamical system
as a perturbation of some known motion of a simplified dynamical system is natural and
widely used in my areas of physics and astronomy. In case of dynamics of celestial bodies
a suitable simplification is the two-body motion which is simple and given by analytical
formulas. As perturbations one can consider not only N-body forces as given above, but
also non-gravitational forces, relativistic forces, etc.
The problem of motion of N-bodies is a very complicated problem. Since its formulation
the N-body problem has led to many new branches in mathematics. Here let us only
mention the Kolmogorov-Arnold-Moser (KAM) theory that proves the existence of stable
quasi-periodic motions in the N-body problem. A review of mathematical results known
in the area of the N-body problem is given in the encyclopedic book of Arnold, Kozlov &
Neishtadt (1997).
The main practical tool to solve the equations of the N-body problem is numerical
integration. One can distinguish three different modes of these numerical integrations. First
mode is integrations for relatively short time span and with highest possible accuracy. This
sort of solutions is used for the solar system ephemerides and space navigation. Some
aspects of these high-accuracy integrations are discussed in Section 3.5 below. Second sort
of integrations are integrations of a few bodies over very long periods of time with the goal
to investigate long-term dynamics of the motion of the major and minor bodies of the solar
system or exoplanetary systems. Usually one considers a subset of the major planets and the
Sun as gravitating bodies and investigates long-term motion of this system or the long-term
dynamics of massless asteroids. For this sort of solution, it is important to have correct
phase portrait of the motion and not necessarily high accuracy of individual orbits. Besides
that, usually the initial conditions of the problem are such that no close encounters between
massive bodies should be treated. Symplectic integrators are often used for these integrations
CHAPTER 3. THE N-BODY PROBLEM 33

because of their nice geometrical properties (e.g., the symplectic integrators do not change
the integral of energy). Resonances of various nature play crucial role for such studies and
are responsible for the existence of chaotic motions. A good account of recent efforts in
this area can be found in Murray & Dermott (1999) and Morbidelli (2002). Third kind of
numerical integrations is integrations with arbitrary initial conditions that do not exclude
close encounters between gravitating bodies. Even for small N numerical integrations of
(3.2) in this general case is not easy, e.g. because of possible close encounters of the bodies
which make the result of integration extremely sensitive to small numerical errors. During
last half of a century significant efforts have been made to improve the stability and reliability
of such numerical simulations. This includes both analytical change of variables known as
“regularisation” and clever tricks in the numerical codes. An exhaustive review of these
efforts can be found in Aarseth (2003). To increase the performance and make it possible
to integrate the N-body problem for large N special-purpose hardware GRAPE has been
created on which special parallel N-body code can be run. Nowadays, direct integrations
of the N-body problem are possible with N up to several millions. This makes it possible
to use these N-body simulations to investigate the dynamics of stellar clusters and galaxies
(Aarseth, Tout & Mardling, 2008).

3.4 Overview of the three-body problem

Special cases of the N-body problem are the two- and three-body problems. The two-body
problem is the basis of all practical computations of the motion of celestial bodies and has
been considered above. The three-body problem also has important practical applications.
The real motion of the Moon is much better described by the three-body system Sun-Earth-
Moon than by the two-body problem Earth-Moon. The motion of asteroids and comets can
often be approximated by the system Sun-Jupiter-asteroid. The two-body problem can be
solved completely. Already the three-body problem is so much complicated that, in general
case, it cannot be solved in analytically closed form. The motion of three attracting bodies
already contains most of the difficulties of the general N-body problem. However, many
theoretical results describing solutions of the three-body problem have been found. For
example, all possible final motions (motions at t → ±∞) are known. Also many classes
of periodic orbits were found. The three-body problem has five important special solutions
called Lagrange solutions. These are points of dynamical equilibrium: all three masses
remain in one plane and have in that plane a Keplerian orbit (being a conic section) with
the same focus and with the same eccentricity. Therefore, in this case the motion of each
body is effectively described by the equations of the two-body problem. The geometrical
form of the three-body configuration (i.e. the ratio of mutual distances between the bodies)
remains constant, but the scale can change and the figure can rotate. In a reference system
where positions of arbitrary two bodies are fixed there are five points where the third body
can be placed (see Figure 3.1). The three bodies either are always situated on a straight
line (three rectilinear Lagrange solutions L1 , L2 , and L3 ) or remain at the vertices of an
equilateral triangle (two triangle Lagrange solutions L4 and L5 ). Three rectilinear solutions
were first discovered by Leonhard Euler (1707 – 1783) and are sometimes called Euler’s
solutions.
A simplified version of the three-body problem – the so-called restricted three-body prob-
CHAPTER 3. THE N-BODY PROBLEM 34

lem – is often considered. In the restricted three-body problem the mass of one of the bodies
is considered to be negligibly small, so that the other two bodies can be described by the
two-body problem and the body with negligible mass moves in the given field of two bodies
with given Keplerian motion. Clearly, in many practical situations the mass of the third
body can indeed be neglected (e.g., for the motion of minor bodies or spacecrafts in the
field of the Sun and one of the planets). Sometimes, it is further assumed that the motion
of all three bodies is co-planar (“planar restricted three-body problem”) and/or that the
orbit of the two massive bodies is circular (“circular restricted three-body problem”). The
five Lagrange solutions do exist also in these restricted versions of the three-body problem.
In the circular restricted three-body problem the five configurations remain constant in the
reference system co-rotating with the two massive bodies. These points are called libration
or equilibrium points. Oscillatory (librational) motion around these points has been inves-
tigated in detail. In the linear approximation, such librational orbits around L4 and L5 in
the circular restricted three-body problempare stable provided that the ratio of the masses
of two massive bodies is less than 1/2 − 23/108 ≈ 0.03852. The orbits around L1 , L2 ,
and L3 are unstable. Interestingly, librational motions around L4 and L5 are realized in
the Solar system. E.g. the asteroid family called Trojans has orbits around L4 and L5 of
the system Sun-Jupiter-asteroid. These librational orbits are stable since the ratio of the
masses of Jupiter and the Sun is about 10−3 which is much smaller than the limit given
above. The rectilinear Lagrange points have also practical applications. Librational orbits
around these points – the so-called Lissajous orbits – are very attractive for scientific space
missions. Lissajous orbits around L1 and L2 of the system Sun-Earth-spacecraft are used
for such space missions as WMAP, Planck, Herschel, SOHO, Gaia and James Webb Space
Telescope. Points L1 and L2 of the system Sun-Earth-spacecraft are situated on the line
Sun-Earth at the distance of about 1.5 million kilometers from the Earth (see Figure 3.1).
Although the Lissajous orbits are unstable, the maneuvers needed to maintain these orbits
are simple and require very limited amount of fuel. On the other side, placing a spacecraft
on an orbit around L1 or L2 guarantees almost uninterrupted observations of celestial ob-
jects, good thermal stability of the instruments, and optimal distance from the Earth (too
far for the disturbing influence of the Earth’s figure and atmosphere, and close enough for
high-speed communications).
One more important result in the circular restricted three-body problem is the existence
of an additional integral of motion called the Jacobi integral. This integral can be used to
recognize, e, g. comets even after close encounters with planets. This is the so-called Tis-
serand criterion: the Jacobi integral should remain the same before and after the encounter
even if the heliocentric orbital elements of the comet have substantially changed. The value
of the Jacobi integral also defines (via the so-called Hill’s surfaces of zero velocity) spatial
region in which the massless body must be found. The details on the Jacobi integral can be
found, e.g. in the book of Roy (2005).
Finally, let us note that although the N-body problem in general and three-body problem
in particular are one of the oldest problems in astronomy, new results in this area continue to
appear. Good example here is a remarkable figure-eight periodic solution of the three-body
problem discovered by Chenciner & Montgomery (2000).
CHAPTER 3. THE N-BODY PROBLEM 35

Figure 3.1: Lagrange points L1 , L2 , . . ., L5 in the three-body problem. Two masses labeled with
m1 and m2 are at the indicated positions. The third body has mass m3 . The positions of the
Lagrange points depend only on the mass ratios between m1 , m2 , and m3 . For this plot masses
with m2 /m1 = m3 /m1 = 0.1 were used.

3.5 Planetary ephemerides

Summary: Modern ephemerides: dynamical models, observations, representations.

Modern ephemerides of the solar system bodies are numerical solutions obtained by nu-
merical integration of the differential equations of motion and by fitting the initial conditions
of these integrations and other parameters of the force model to observational data.
The equations of motion used here are the N-body equations discussed above augmented
by a number of smaller forces. These forces include relativistic N-body forces (the so-called
Einstein-Infeld-Hoffmann equations), Newtonian forces due to asteroids, the effects of the
figures (non-sphericity) of the Earth, Moon and the Sun as well as some non-gravitational
forces. For the Sun it is sufficient to consider the effect of the second zonal harmonics J2⊙ .
The zonal harmonics Jn of the Earth and the Moon are usually used up to n ≤ 4. Mostly
one needs forces coming from the interaction of these zonal harmonics with other bodies
modeled as point masses. The dynamics of the Earth-Moon system requires even more
detailed modeling since the translational motion of the Earth and the Moon are coupled
with their rotational motions and deformations in a tricky way. For the Moon even more
subtle effects due to tesseral harmonics Cnk and Snk again with n ≤ 4 should be taken into
account. Tidal deformations of the Earth’s gravitational field influence the translational
motion of the Moon and should be taken into account. Rotational motion of the Earth is
well known and obtained from dedicated observations by the International Earth Rotation
and Reference Systems Service (IERS). These results are good enough and usually taken for
granted for the solar system ephemerides. Rotational motion of the Moon is often called
physical libration and is an important part of the process of construction of solar system
CHAPTER 3. THE N-BODY PROBLEM 36

ephemerides. Physical libration is modeled as rotational motion of a solid body with tidal
and rotational distortions, including both elastic and dissipational effects. A discussion of
all these forces coming from the non-point-like structure of gravitating bodies can be found
in Standish & Williams (2010).
Asteroids play an important role for high-accuracy modeling of the motion of the inner
solar system, the motion of Mars being especially sensitive to the quality of the model of
asteroids. Since masses of asteroids are poorly known for most of them, the modeling is not
trivial. Usually, asteroids are treated in three different ways. A number of “big” asteroids
are integrated together with the major planets, the Moon and the Sun. For these “big”
asteroids the masses are estimated from the same observational data that are used to fit the
ephemeris. Among these “big” asteroids are always “the big three” – Ceres, Pallas and Vesta
– and, sometimes, up to several ten asteroids which influence the motion of Mars more than
other asteroids. For some hundred asteroids their masses are estimated using their taxonomic
(spectroscopic) classes and their estimated radii that are determined by photometry, radar
data or observations of stellar occultations by asteroids. For each of the three taxonomic
classes C (carbonaceous chondrite), S (stony) and M (iron) the mean density is determined
as a part of the ephemeris construction. The cumulative effect of other asteroids is sometimes
empirically modeled by a homogeneous massive ring in the plane of ecliptic. The mass of the
ring and its radius are again estimated from the same data that are used for the ephemeris
(Pitjeva, 2007; Kuchynka et al. 2010).
Since the equations of motion are ordinary differential equations any method for numer-
ical integration of ordinary differential equations can be used to solve them. A very good
practical overview of numerical integration methods is given in Chapter 4 of Montenbruck
& Gill (2000). Even more details can be found in Chapter 7 of Beutler (2005, Part I). In
practice, for planetary motion, one uses either multistep Adams (predictor-corrector) meth-
ods (Standish & Williams, 2008; Fienga et al., 2008) or the Everhart integrator (Everhart,
1985; Pitjeva, 2005). The latter is a special sort of implicit Runge-Kutta integrators. Nu-
merical round-off errors are an important issue for the integrations of planetary ephemerides.
Usual double precision (64 bit) arithmetic is not sufficient to achieve the goal accuracy and
one often uses quadruple precision (128 bit) arithmetic. Since the beginning of the 1970s
the JPL ephemeris team uses the variable stepsize, variable order multistep Adams integra-
tors called DIVA/QIVA (Krogh, 2004). Fienga at al. (2008) have shown that only a few
arithmetical operations in the classical Adams integrator of order 12 must be performed with
quadruple precision to achieve an acceptable accuracy over longer integration intervals. This
substantially increases the performance of numerical integrations.
Observational data used for planetary ephemerides include radar observations of earth-
like planets, radar and Doppler observations of spacecrafts (especially planetary orbiters),
VLBI observations of spacecrafts relative to some reference quasars, Lunar Laser Ranging
data, and, finally, optical positional observations of major planets and their satellites (espe-
cially important for outer planets with very few radiometric observations).
A total of 250 parameters are routinely fitted for the construction of planetary
ephemerides. These parameters include initial positions and velocities of the planets and
some of their satellites, the orientation of the frame with respect to the ICRF, the value of
Astronomical Unit in meters (or the mass parameter GM⊙ of the Sun), the parameters of the
model for asteroids (see above), various parameters describing rotational and translational
motion of the Earth-Moon system, various parameters used in the reduction of observational
CHAPTER 3. THE N-BODY PROBLEM 37

data (phase corrections for planetary disk observations, corrections to precession and equinox
drift, locations of various relevant sites on the Earth and other bodies, parameters of the
solar corona, parameters describing the geometrical figures of Mercury, Venus and Mars,
etc.). Useful discussion of various models used for data modeling is given by Moyer (2003)
and Standish & Williams (2010). The masses of the major planets can be also fitted from
the same data, but, when available, they are taken from the special solutions for the data of
planetary orbiters. However, the masses of the Earth and the Moon are often determined in
the process of construction of planetary ephemerides.
Modern ephemerides are represented in the form of Chebyshev polynomials. The details
of the representation can vary from one ephemeris to another, but the principles are the
same: each scalar quantity is represented by a set of polynomials pi of the form
Ni
X (i) 2t − ti − ti+1
pi (t) = ak Tk (x), x= , (3.16)
k=0
ti+1 − ti

where Tk (x) are the Chebyshev polynomials of the first kind (given by the recurrent rela-
(i)
tions T0 = 1, T1 = x, and Tk+1 = 2xTk − Tk−1 ), and coefficients ak are real numbers. Each
polynomial pi (t) is valid for some interval of time ti ≤ t ≤ ti+1 (so that −1 ≤ x ≤ 1).
The representation (3.16) is close to the optimal uniform approximation of a function by
polynomials of given order (Press et al., 2007, Section 5.8), and, thus, gives nearly optimal
representation of a function using given number of free parameters. The orders of polyno-
mials Ni are usually the same for all time intervals, but do depend on the quantity to be
represented. Sometimes (e.g. for the JPL ephemerides) one polynomial represents both the
position and the velocity of a body. The velocity can be then calculated as a derivative of
(3.16):

XNi
d 2 (i)
pi (t) = kak Uk (x), (3.17)
dt ti+1 − ti k=1

where Uk are the Chebyshev polynomials of the second kind (given by the recurrent relations
U0 = 1, U1 = 2x, and Uk+1 = 2xUk − Uk−1 ). At the boundaries ti of the time intervals,
the polynomials pi must satisfy conditions like pi−1 (ti ) = pi (ti ), so that the approximat-
ing function is continuous. One can also imply additional constraints dtd pi−1 (ti ) = dtd pi (ti )
for the derivatives to make the approximating function continuously differentiable. An ef-
(i)
ficient technique to compute the coefficients ak starting from values of the quantity to be
represented is described by Newhall (1989).
There are three sources of modern planetary ephemerides: Jet Propulsion Laboratory
(JPL, Pasadena, USA; DE ephemerides), Institut de Méchanique Céleste et de Calcul des
Éphémérides (IMCCE, Paris Observatory, France; INPOP ephemerides) and Institute of Ap-
plied Astronomy (IAA, St.Petersburg, Russia; EPM ephemerides). All of them are available
from the Internet:

– http://ssd.jpl.nasa.gov/?planet_eph_export for the DE ephemerides,

– http://www.imcce.fr/fr/presentation/equipes/ASD/inpop/ for the INPOP, and

– ftp://quasar.ipa.nw.ru/incoming/EPM2004/ for the EPM ephemerides.
CHAPTER 3. THE N-BODY PROBLEM 38

Different versions of the ephemerides have different intervals of validity, but typically
these are several hundred years around the year 2000. Longest readily available ephemerides
are valid for a time span of 6000 years. Further details on these ephemerides can be found
in Standish & Williams (2010), Folkner (2010), Fienga et al. (2008), and Pitjeva (2005),
respectively.
Let us also mention that for the lower-accuracy applications, semi-analytical theories of
planetary motion called VSOP are available (Bretagnon, Francou, 1988; Moisson, P Bretagnon,
nk
2001). The semi-analytical theories are given in the form of Poisson series k ck t cos(ak t +
bk ), where ak , bk , and ck are real numbers, and nk is the integer power of time t. Formally any
value of time can be substituted into such series, but the theory is meaningful only for several
thousand years around the year 2000. The VSOP ephemeris contains only major planets
and the Earth-Moon barycenter. The best semi-analytical theory of motion of the Moon
with respect to the Earth is called ELP82. This theory can also be used in low-accuracy
applications.
Chapter 4

Elements of the Perturbation Theory

4.1 The method of the variation of constants

Summary: The variation of constants as a method to solve differential equations. Instantaneous
elements. Osculating elements.

The equations of motion of two-body problem considered above in great detail read
r
r̈ + κ2
= 0.
r3
The simplicity of the two-body motion and the fact that many practical problems of celestial
mechanics are sufficiently close to two-body motion make it practical to use two-body motion
as zero approximation to the motion in more realistic cases and treat the difference by
the usual perturbative approach. Special technique for the motion of celestial bodies is
called osculating elements. The solution of the two-body problem discussed above can be
symbolically written as

r = f (t, e1 , e2 , e3 , e4 , e5 , e6 ), ei = const, i = 1 . . . 6, (4.1)

where ei are six Keplerian elements: semi-major axis a, eccentricity e, inclination i, argument
of pericenter ω and longitude of the node Ω. In general case of arbitrary additional forces it
is always possible to write the equations of motion of a body as
r
r̈ + κ2 = F, (4.2)
r3
where F = F (t, r, ṙ, . . .) is arbitrary force depending in general on the position and velocity
of the body under study, time and any other parameters. One example of such a disturbing
force is given by (3.13) for the N-body problem. The general idea is to use the same functional
form for the solution of (4.2) as we had for the two-body motion, but with constants (former
Kepler elements) being time-dependent:

r = f (t, e1 , e2 , e3 , e4 , e5 , e6 ), ei = ei (t), i = 1 . . . 6. (4.3)

This is always possible since F has three degrees of freedom (three arbitrary components)
and representation (4.3) involves six arbitrary functions of time. Let us stress the following.

39
CHAPTER 4. ELEMENTS OF THE PERTURBATION THEORY 40

Eq. (4.3) means that if elements ei (t) are given for some t, the position r of the body under
study for that t can be computed using usual formulas of two-body problem as summarized
in Section 2.8. The idea of (4.3) is closely related to the idea of the method of variation of
constants, also known as variation of parameters, developed by Joseph Louis Lagrange. This
method is a general method to solve inhomogeneous linear ordinary differential equations.
As mentioned above the representation in (4.3) has three “redundant” degrees of freedom.
These three degrees of freedom can be used to make it possible to compute not only position
r, but also velocity ṙ from the solution of (4.2) using standard formulas of the two-body
problem summarized in Section 2.8. This can be done if the elements ei (t) satisfy the
following condition
6
X ∂f
ėi = 0 . (4.4)
i=1
∂ei

Indeed, in general case, time derivative of r given by (4.3) reads

6
∂f X ∂f
ṙ = + ėi . (4.5)
∂t i=1
∂ei

Therefore, condition (4.4) guarantees that the time derivative of (4.3) is given by the partial
derivative of f with respect to time

∂
ṙ = f (t, e1 , e2 , e3 , e4 , e5 , e6 ) . (4.6)
∂t
This means that velocity ṙ can be calculated by the standard formulas of two-body problem
(indeed, (4.6) coincided with the derivative of (4.1 with constant ei that represent the usual
solution of the two-body problem). The elements having these properties are called osculat-
ing elements. The osculating elements are in general functions of time. To compute position
and velocity at any given moment one first has to calculate the values of the six osculating
elements for this moment of time and then use the standard equations summarized in Section
2.8.
Let us stress that, with osculating elements, not only vectors of position r and velocity
ṙ can be computed using formulas of the two-body problem, but also any functions of these
two vectors. Let us give an explicit example here. For a given moment of time t the absolute
value of r is given as r(t) = a(t) (1 − e(t) cos E(t)). Here a(t) and e(t) are osculating
semi-major axes and eccentricity. Osculating eccentric anomaly E(t) can be computed from
Kepler equation E − e(t) sin E = M, where again osculating eccentricity e(t) should be used.
κ a−3/2
The derivative of r can be computed ṙ = Ė a e cos E = a e cos E 1−e cos E
= √ κ e 2 sin v
a (1−e )
and all elements are again functions of time: a = a(t), e = e(t), etc. Also the anomalies
– eccentric E, true v and mean M – are related to each other in the same way as in the
two-body problem.
CHAPTER 4. ELEMENTS OF THE PERTURBATION THEORY 41

4.2 Gaussian perturbation equations

Summary: The radial, tangential and transverse components of the disturbing force. The Gaussian
perturbation equations: the differential equations for the osculating elements. Other variants of
the Gaussian perturbation equations.

Let us derive the equations for osculating elements for a general disturbing force F . First,
we introduce a new Cartesian coordinate system. The origin of the new system (S, T, W )
is the same as usual, but the orientation is different and depends at each moment of time
on the position r and velocity ṙ of the considered body. Axis S is directed radially, that is
parallel to r. Axis T lies in the momentary orbital plane (the plane containing both r and
ṙ), perpendicular to S (and, therefore, to r) and the angle between T and ṙ does not exceed
90◦ . Axis W is perpendicular to both S and T (that is, perpendicular to both r and ṙ)
and completes S and T to a right-hand coordinate system (S, T, W ). Coordinates (S, T, W )
rotate as the body moves along its orbit.
The components of disturbing force F in axes (S, T, W ) are also denoted by (S, T, W )
and can be computed as
r
S = · F, (4.7)
r
(r × ṙ) × r
T= · F, (4.8)
|r × ṙ| r
r × ṙ
W= · F, (4.9)
|r × ṙ|
where ’·’ and ’×’ denote the scalar and cross products of two vectors. Obviously, the relation
of (S, T, W ) and our usual coordinates (x, y, z) reads
   
ax aA
 ay  = ATz (Ω) ATx (i) ATz (u)  aT  , (4.10)
az aW

where angle u = v+ω is called argument of latitude. The matrix ATz (Ω) ATx (i) ATz (u) is given
by (2.30) with u substituted for ω. Here for an arbitrary vector a its components in coordi-
nates (x, y, z) are denoted as (ax , ay , az ) and the corresponding components in coordinates
(S, T, W ) are (aS , aT , aW ).

4.2.1 Derivation of differential equations for osculating elements

Now, let us derive the required equations one by one. First, let us consider the integral of
areas c = r × ṙ. This leads to c2 = (r × ṙ) · (r × ṙ), where c = |c|. A time derivative of c2
then reads
dc2
= 2 (r × ṙ) · (r × r̈) = 2 (r × ṙ) · (r × F ) = 2 ((r × ṙ) × r) · F = 2 r 2 ṙ · F − 2 r ṙ r · F
dt
(ṙ · r = ṙ r is used here). Since F = (S, T, W ) and r = (r, 0, 0) in STW-coordinates one gets
r · F = r S. Let us now consider ṙ · F . We need vectors ṙ and F in STW-coordinates. Since
the instantaneous plane of the orbit is defined by the instantaneous position and velocity
CHAPTER 4. ELEMENTS OF THE PERTURBATION THEORY 42

vectors of the body, the W component of the velocity ṙ vanish by definition. Obviously, the
S component is ṙ and the T component is√r v̇. For the latter from the integral of areas in√polar
√ κ p κ p
coordinates r 2 v̇ = κ p, one
√
has r v̇ = r . Therefore, in STW coordinates ṙ = (ṙ, r , 0).
κ p 2
Therefore, ṙ · F = ṙ S + r T . Substituting r · F and ṙ · F into the equation for dcdt derived
above and taking into account that the semi-latus rectum p = c2 /κ2 we get
 
T
ṗ = 2 p r √ . (4.11)
κ p
Now, let us consider the time derivative of the integral of area itself. One has c = r × ṙ
and, therefore, ċ = dtd (r × ṙ) = r × r̈ = r × F . It is clear that c is parallel to axis W
of the STW system (since c is defined as c = r × ṙ it is perpendicular to both r and
√
ṙ). For this reasons and considering that c = κ p, the components of c in STW axes
√
read c = (0, 0, κ p). Using transformation (4.10) to convert the ST W -components into
xyz-ones, one gets
   √ 
cx sin i sin Ω κ p
 cy  =  − sin i cos Ω κ √p  . (4.12)
√
cz cos i κ p
In the STW system one has r = (r, 0, 0) and F = (S, T, W ). Therefore, in STW components
one gets r × F = (0, −r W, r T ). Again using transformation (4.10) it is easy to calculate
that the z component of r × F reads (r × F )z = − sin i cos u r W + cos i r T . Considering
the z component of ċ = r × F one gets
d
cz = − sin i cos u r W + cos i r T .
dt
On the other hand, from (4.12) one gets
d d √ √ d d √
cz = (cos i κ p) = −κ p sin i i + cos i (κ p) .
dt dt dt dt
From (4.11) it is easy to see that
d √ 1
(κ p) = κ √ ṗ = r T ,
dt 2 p
and we finally get the following equation for the derivative of inclination i:
 
d W
i = r cos u √ . (4.13)
dt κ p
√
Analogously, considering the time derivative of cx = sin i sin Ω κ p and computing the
x component of r × F from (4.10) one gets
d d √
cx = (sin i sin Ω κ p) = (− cos Ω sin u − cos i sin Ω cos u) (−r W ) + sin i sin Ω r T .
dt dt
d
Using here equations (4.11) and (4.13) for ṗ and dt
i one gets the equation for the time
derivative of Ω:
 
d r sin(v + ω) W
Ω= √ . (4.14)
dt sin i κ p
CHAPTER 4. ELEMENTS OF THE PERTURBATION THEORY 43

Clearly, the y component of ċ = (r × F ) gives no new information since cy depends on the

same elements as cx . We have therefore got all possible equations from the integral of areas
of the two-body problem. Now let us turn to the integral of energy (2.12). This integral can
be rewritten in the form
1 κ2 κ2
ṙ · ṙ − =− . (4.15)
2 r 2a
The derivative of this equation
κ2 κ2
ṙ · r̈ + ṙ = ȧ
r2 2a2
can be simplified using
 2 
κ κ2
ṙ · r̈ = ṙ · − 3 r + F = − 3 ṙ · r + ṙ · F ,
r r

κ2 κ2 κ2
ṙ · r̈ + ṙ = − ṙ · r + ṙ · F + ṙ = ṙ · F
r2 r3 r2
(here we used that ṙ r = ṙ · r) and computing ṙ · F from the STW components of ṙ and F
already given above √
κ p
ṙ · F = ṙ S + T.
r
This allows one to get the equation for ȧ:
   
d 2 S 2 p T
a = 2 a e sin v √ +2a √ . (4.16)
dt κ p r κ p
Here we used that ṙ = √κep sin v. Having equations (4.11) and (4.16) for ṗ and ȧ it is easy to
derive the equation for ė. Indeed, from p = a(1 − e2 ) one gets ṗ = ȧ(1 − e2 ) − 2e a ė. Solving
for ė and substituting (4.11) and (4.16) one gets
   
d S T
e = p sin v √ + p (cos v + cos E) √ . (4.17)
dt κ p κ p
Now, let us turn to the derivation of ω̇. The derivation consists of several steps. First,
from
p
r=
1 + e cos v
one gets
p
1 + e cos v = .
r
Computing time derivative of the latter equations
ṗ p
−e sin v v̇ + ė cos v = − 2 ṙ
r r
and using equations for ṙ, ṗ and ė derived above one gets the derivative of true anomaly v
as
√    
κ p p cos v S p+r T
v̇ = 2 + √ − sin v √ . (4.18)
r e κ p e κ p
CHAPTER 4. ELEMENTS OF THE PERTURBATION THEORY 44

As the second step in the derivation of ω̇, let us consider the formula for the z component
of vector r: z = r sin i sin u, where u = v + ω. This formula can be derived e.g., from the
equation immediately before (2.63). On the one hand, ż is the component of the velocity
vector ṙ and can be calculated considering all osculating elements
√
as constants (from the
2 √ κ p
integral of area in polar coordinates r v̇ = κ p one has v̇ = r2 ):
√
κ p
ż = ṙ sin i sin u + r v̇ sin i cos u = ṙ sin i sin u + r 2 sin i cos u
r
However, the same ż can be computed not assuming that the osculating elements are con-
stants (that is, explicitly considering the time derivatives of the osculating elements). This
should give the same results according to the idea of osculating elements described in Section
4.1. Therefore, one has
di du
ż = ṙ sin i sin u + r cos i sin u + r sin i cos u .
dt dt
Equating the last two expressions for ż one gets
√
κ p di
u̇ = 2 − cot i tan u . (4.19)
r dt
Finally, since u = v + ω one has ω̇ = u̇ − v̇ and using equations (4.19) and (4.18) for u̇ and
di
v̇ and (4.13) for dt one gets
     
d p cos v S r+p T W
ω=− √ + sin v √ − r sin(v + ω) cot i √ . (4.20)
dt e κ p e κ p κ p
The only equation still to be derived is that for the mean anomaly of an epoch. Let us
first consider the definition of the mean anomaly

M = M0 + n(t − t0 ) , (4.21)

where t0 is a given fixed epoch for which M = M0 . In the framework of the two-body
problem n = κ a−3/2 is constant. In case of osculating elements n is time-dependent (since a
is time-dependent) and the derivative of M reads
d d dn
M= M0 + n + (t − t0 ). (4.22)
dt dt dt
This formula contains time t explicitly. This is not convenient for many applications. This
can be avoided if instead of (4.21) one defines the mean anomaly as
Z t
M = M0 + n dt . (4.23)
t0

In the framework of the two-body problem (4.23) is fully equivalent to (4.21). However, for
osculating elements derivative of M from (4.23) reads
d d
M= M0 + n . (4.24)
dt dt
CHAPTER 4. ELEMENTS OF THE PERTURBATION THEORY 45

Clearly, dtd M 0 = dtd M0 + ṅ (t − t0 ), where ṅ = dn

dt
= − 23 κ a−5/2 ȧ. Definition (4.23) is used
below.
In order to derive the equation for dtd M 0 let us consider two equations

E − e sin E = M ,
r = a (1 − e cos E) .

Differentiating these equations one gets

Ė (1 − e cos E) − ė sin E = Ṁ ,
ṙ = ȧ (1 − e cos E) − a ė cos E + a e sin E Ė.

Considering that ṙ = √κ e sin v one gets

p
√
1 − e2  r 
Ṁ = n + ė cot v − ȧ 2 . (4.25)
e a sin v
Comparing to (4.24) and substituting (4.17) and (4.16) for ė and ȧ one finally gets
√     
d 1 − e2 S T
M0 = (p cos v − 2 e r) √ − (r + p) sin v √ . (4.26)
dt e κ p κ p

4.2.2 Discussion of the derived equations

Gathering all equations for the derivatives of osculating elements derived above one gets the
full set of equations:
   
d 2 S 2 p T
a = 2 a e sin v √ +2a √ , (4.27)
dt κ p r κ p
   
d S T
e = p sin v √ + p (cos v + cos E) √ , (4.28)
dt κ p κ p
 
d W
i = r cos(v + ω) √ , (4.29)
dt κ p
     
d p cos v S r+p T W
ω=− √ + sin v √ − r sin(v + ω) cot i √ , (4.30)
dt e κ p e κ p κ p
 
d r sin(v + ω) W
Ω= √ , (4.31)
dt sin i κ p
√     
d 1 − e2 S T
M0 = (p cos v − 2 e r) √ − (r + p) sin v √ . (4.32)
dt e κ p κ p

These equations were derived by Johann Carl Friedrich Gauß (1777 – 1855) and called Gaus-
sian perturbation equations. The equations were also independently derived by Leonhard
Euler some time before Gauss and are sometimes called Euler equations. We will use these
equations below when discussing the influence of atmospheric drag on the motion of Earth’s
satellites.
CHAPTER 4. ELEMENTS OF THE PERTURBATION THEORY 46

Numerous alternative forms of the Gaussian perturbation equations are given in Beutler
(2005). In particular, one can use any other components of the disturbing force F instead
of (S, T, W ). Eqs. (4.27)–(4.32) are valid for elliptic motion (e < 1). Similar equations for
hyperbolic motion with e > 1 can be derived, but are rarely used in practice. Let us note
that Eqs. (4.27)–(4.32) have singularities for e = 0 or very small e (equations for ω and
M 0 ), i = 0 and i = π (equations for ω and Ω), and e close to 1 (e.g. since p is small for
this case). Indeed, the right-hand sides of the equations are not well defined in these cases.
These singularities are related to the fact that some of the Keplerian elements are not well
defined for e = 0, i = 0 or i = π. If the application requires these cases to be included one
introduces

h = e sin ω,
k = e cos ω, (4.33)

instead of e and ω and

p = tan i sin Ω,
q = tan i cos Ω. (4.34)

instead of i and Ω. The equations for the derivatives h, k, p and q can be easily derived
from (4.27)–(4.32). The equations for ḣ, k̇, ṗ and q̇ and do not contain singularities.

Exercise. Derive equations for the derivatives of osculating h, k, p and q using corresponding
equations for ω̇, ė, i̇ and Ω̇ in (4.27)–(4.32).

4.3 Lagrange equations

Summary: The potential disturbing force. Lagrange equations for the osculating elements. Prop-
erties of the Lagrange equations.

The Gaussian perturbation equations are valid for any disturbing force F . For the special
case when the perturbing force has a potential R one has
∂
Fi = R.
∂r i
In this case one can modify the Gaussian perturbation equations so that the disturbing
potential R appear in the equations instead of the components of perturbing force F . This
case of potential perturbations is very important for practical applications. For example,
above we have seen that the N-body problem can be considered as perturbed two-body
problem with disturbing force having potential (3.15).
In general both F and R can depend on time t, position r and velocity ṙ of the body
under study: R = R(t, r, ṙ). Here we consider the simpler situation when both F and R
do not depend of velocity ṙ. Therefore, we assume that R = R(t, r). This covers the most
important applications of potential perturbations.
First, we consider potential R as function of osculating elements R = R(t, e1 , e2 , e3 , e4 , e5 , e6 ).
This parametrization can be directly derived by substituting (4.3) into R(t, r). Our goals
CHAPTER 4. ELEMENTS OF THE PERTURBATION THEORY 47

is to replace the components S, T and W of the disturbing force in (4.27)–(4.32) by partial

∂R
derivatives of R. First, we should compute ∂e i
as functions of S, T and W . One has

3 3
∂R X ∂R ∂r j X ∂r j
= j ∂e
= Fj .
∂ei i=1
∂r i i=1
∂ei

Here r j is the j-th component of vector r and F j is the j-th component of F . Therefore one
j
should only know ∂r ∂ei
to compute ∂e∂R
i
.
Let us consider the example of semi-major axis a. According to the formulas of the
two-body problem summarized in Section 2.8, radius-vector r is linearly proportional to a.
j j
Therefore, one has ∂r ∂a
= ra and, finally,
3 3
∂R X j ∂r j 1 X j j 1 r
= F = F r = F · r = S. (4.35)
∂a i=1
∂a a i=1 a a

Analogously, one gets

   
∂R r
= a −S cos v + T 1 + sin v , (4.36)
∂e p
∂R
= r W sin (v + ω) , (4.37)
∂i
∂R
=rT , (4.38)
∂ω
∂R
= r T cos i − r W sin i cos (v + ω) , (4.39)
∂Ω
∂R a  p 
=√ e S sin v + T . (4.40)
∂M 0 1 − e2 r
Now, one should invert (4.35)–(4.40) to get three components S, T and W as functions of six
∂R
partial derivatives ∂e i
. Clearly, this inversion is not unique. In principle, any possible inverse
of (4.35)–(4.40) can be inserted to the Gaussian perturbation equations (4.27)–(4.32) and
lead to a correct set of equations for osculating elements. However, the resulting equations
∂R
can be made especially simple if one requires that the coefficients of ∂e i
in the final equations
do not depend on time explicitly (e.g., do not contain radius-vector r and true anomaly v).
With this requirement the inversion of (4.35)–(4.40) is also unique. Finally, the equations
read
d 2 ∂R
a= , (4.41)
dt n a ∂M 0
√
d 1 − e2 ∂R 1 − e2 ∂R
e= − , (4.42)
dt e n a2 ∂M 0 e n a2 ∂ω
d cot i ∂R cosec i ∂R
i= √ − √ , (4.43)
dt n a2 1 − e2 ∂ω n a2 1 − e2 ∂Ω
√
d 1 − e2 ∂R cot i ∂R
ω= 2
− √ , (4.44)
dt e n a ∂e n a 1 − e ∂i
2 2
CHAPTER 4. ELEMENTS OF THE PERTURBATION THEORY 48

d cosec i ∂R
Ω= √ , (4.45)
dt n a2 1 − e2 ∂i
d 2 ∂R 1 − e2 ∂R
M0 = − − , (4.46)
dt n a ∂a e n a2 ∂e
where the potential R is considered as function of osculating elements and possibly time
t: R = R(t, a, e, i, ω, Ω, M 0 ). These equation were first derived by Joseph Louis Lagrange
(1736 – 1813) and are called Lagrange equations. Clearly, the Lagrange equations are fully
equivalent to the Gaussian perturbation equations.
The structure of the Lagrange equations is very interesting. As noted above time appear
in (4.41)–(4.46) (if at all) only in R. The coefficients the partial derivatives depend only on
osculating elements that are constants in the two-body problem. Furthermore, the elements
can be divided into two groups: αk = (a, e, i) and βk = (ω, Ω, M 0 ), k = 1, 2, 3. Then the
Lagrange equations can be symbolically written as
3
X ∂R
α̇k = Akl ,
l=1
∂βl
3
X ∂R
β̇k = − Alk .
l=1
∂αl

This means that

(a) the derivatives of the elements of one group depend only on the partial derivatives of
R with respect to the elements of the other group and
∂R
(b) if the coefficient of in α̇k is Akl then the same coefficient (with minus) appears at
∂βl
∂R
in β̇l .
∂αk
∂R d 1 − e2
The last property should be illustrated. For example, the coefficient of in dt
e is
∂M 0 e n a2
1 − e2 ∂R
(see Eq. (4.42)). The same coefficient 2
(with minus) appears in front of in dtd M 0
ena ∂e
(see Eq. (4.46)). The same symmetry holds for all coefficients. Finally, among nine possible
Akl four vanish and one has only five different coefficients in (4.41)–(4.46). This structure of
the Lagrange equations is used to introduce the so-called canonical elements which will not
be considered here.
Osculating elements are very convenient for analytical assessments of the effects of par-
ticular perturbations. They are also widely used for practical representations of orbits of
asteroids and artificial satellites. Such a representation is especially efficient when the oscu-
lating elements can be represented using simple functions of time requiring a limited number
of numerical parameters (i.e., when a relatively “lower-accuracy” representation over rela-
tively “short” period of time is required, the exact meaning of “lower-accuracy” and “short”
depending on the problem). For example, the predicted ephemerides (orbits) of GPS satel-
lites are represented in form of osculating elements using simple model for the osculating
elements (linear drift plus some selected periodic terms). The numerical parameters are
CHAPTER 4. ELEMENTS OF THE PERTURBATION THEORY 49

broadcasted in the GPS signals from each satellite. Osculating elements are also used in
many cases by the Minor Planet Center of the IAU to represent orbits of asteroids and
comets.
Chapter 5

Three-body problem

5.1 The Lagrange solutions

Summary: The case when the three-body motion can be described by the equations of motion of
the two-body problem: the five Lagrange solutions. Examples of the Lagrange motion in the Solar
system.

5.2 The restricted three-body problem

Summary: The equations of motion. Corotating coordinates. The equations of motion in the
corotating coordinates. The Jacobi integral. The Hill’s surfaces of zero velocity.

5.3 Motion near the Lagrange equilibrium points

Summary: The Lagrange points as equilibrium points. Libration. The equations of motion for
the motion near the Lagrange points. The stability of the motion near the equilibrium points

50
Chapter 6

Gravitational Potential of an
Extended Body

6.1 Definition and expansion of the potential

Summary: Gravitational potential of an extended body as an integral. The Laplace equation.
Legendre polynomials. Associated Legendre polynomials. Expansion of the potential in terms of
the associated Legendre polynomials.

6.1.1 Definition of the potential of an extended body

Let us again consider the equations of motion of a body with mass m0 under the influence
of N bodies with masses mi , i = 1, . . . , N as we did in Section 3.3 (see Fig. 6.1). The first
equation in (3.11) can be written as
N
X N
X
G mi ρ i0 G mi
ρ̈0 = − = grad0 , (6.1)
i=1
ρ3i0 i=1
ρi0

where grad0 defined by (3.4) is the vector of partial derivatives with respect to the compo-
nents of the position ρ0 of body 0. Of course, this equation does not depend on mass m0 .
Therefore, m0 can also be considered as zero. In that case we can think of the influence of
a system of N massive bodies on a test (massless) particle situated at ρ0 . In this way, for
any arbitrary position x at each moment of time t the gravitational potential of the system
of N bodies reads
N
X G mi
U(t, x) = , ρix = |ρi (t) − x|, (6.2)
i=1
ρix

where ρi (t) is the position of body i as function of time t. This potential, through Eq. (6.1),
gives the equations of motion of a test particle in the gravitational field of N massive bodies.
Let us now consider an extended body with some continuous mass distribution. The task
is to calculate its gravitational potential at a point lying outside of the body. We can split
the whole body into N parts or “cells” with some arbitrary N ≥ 1 (see Fig. 6.2). Now, if as
an approximation, we replace each cell by a point-like body situated at the center of mass

51
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 52

m1
m2
m3
m4

mN m0

Figure 6.1: A system of N massive bodies acting on particle m0 .

m0

Figure 6.2: An extended body is split into a number of N parts or cells. Each cell is then
approximated by a point-like body situated at the center of mass of the cell and having mass mi
equal to the mass of the cell. In this way the extended body is approximated by a system of N
point-like bodies.
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 53

of the cell and having mass mi equal to the mass of the cell, we get a system of N point-like
bodies instead of the extended body. The gravitational potential of such a system is again
given by (6.2). Clearly, the larger is the number of cells N, the better is the approximation.
In order to get the potential of an extended body we can simply consider the limit N → P∞.
It is clear that such a limit means mathematically
R that we proceed from a finite sum N i=1
to an integral over the volume of the body V . Masses mi of the cells should be replaced by
mass elements dm = σd3 x, where σ is the mass density of the body (in general as function
of time and position within the body) and d3 x is the volume element. In this way one gets
Z Z
G dm G σ(t, x′ ) d3 x′
U(t, x) = = , (6.3)
V ρ V |x − x′ |
where the integration goes over all points x′ inside the body (that is, where σ > 0). This
equation is valid for any point x at which the potential is evaluated irrespective if x is
situated inside or outside the body. However, in the following we consider only points x
lying outside of the body. Mathematically this can be written as follows. For any body
there exists such a radius R so that for each points x such that |x| > R the density σ of the
body vanishes:
∃ R : ∀ x : |x| > R, σ(t, x) = 0 . (6.4)
Radius R can be called maximal radius of the body in the selected reference system. In the
following we consider U(t, x) only for such x that |x| > R. It is easy to see that in this case
function U(t, x) defined by (6.3) satisfies the Laplace equation
∆ U(t, x) = 0, (6.5)
where ∆ is the Laplace operator defined for any function f as
∂2 ∂2 ∂2
∆f = f + f + f. (6.6)
∂x2 ∂y 2 ∂z 2
One can check directly that U(t, x) defined by (6.3) satisfies (6.5). Indeed, considering that
vectors x and x′ have components (x, y, z) and (x′ , y ′, z ′ ), respectively, one gets
Z
∂U x − x′ 2 ′
=− G σ(t, x′ ) dx (6.7)
∂x V |x − x′ |
and
Z
∂2U 3 (x − x′ )2 − |x − x′ |2 2 ′
= G σ(t, x′ ) dx, (6.8)
∂x2 V |x − x′ |5
2 2 2 2 2
and analogous for ∂∂yU2 and ∂∂zU2 . Summing up ∂∂xU2 , ∂∂yU2 and ∂∂zU2 one sees that (6.5) is satisfied.
Functions satisfying Laplace equation are called harmonic functions. Therefore, one can say
that the gravitational potential of an extended body is harmonic function outside of the
body.
Now let us turn to the calculation of U(t, x) for a given body. If density σ is given, for any
given t and x the potential U can in principle be calculated by numerical integration of (6.3).
The involved integral over the volume is a three-dimensional integral and its calculation is
rather complicated. For each x the integral should be computed anew and this is extremely
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 54

x-x’ x
x’

Figure 6.3: Point x′ is situated within the body, while point x outside of it. The distance between
two points is given by |x−x′ | and the angles between vectors x and x′ is denoted by H (0 ≤ H ≤ π).

inconvenient. It turns out, however, that this integral can be represented in such a way that
the whole dependence on x is explicit. Let us write
1 1
′
=√ , (6.9)
|x − x | 2
r + r − 2 r r ′ cos H
′2

where r = |x|, r ′ = |x′ | and 0 ≤ H ≤ π is the angle between vectors x and x′ (see Fig. 6.3).
r′
Denoting = z and cos H = x (−1 ≤ x ≤ 1) one can write
r
1 1 1 1
′
=√ = √ . (6.10)
|x − x | r 2 + r ′2 − 2 r r ′ cos H r 1 + z2 − 2 x z
Note that notations x and z have now nothing to do with the components of vector x. Since
we only consider points with r = |x| > R and since for any point of the body r ′ = |x′ | ≤ R,
one has z < 1.

6.1.2 Legendre polynomials

Mathematically, the second factor on the right-hand side of (6.10) can be expanded as
∞
X
1
√ = Pn (x) z n , (6.11)
1 + z2 − 2 x z n=0

where Pn (x) are the so-called Legendre polynomials. The left-hand side of (6.11) is called
generating function for the Legendre polynomials. Direct calculations show that
1 3 x2 − 1 2
√ = 1+ xz + z + O(z 3 ) . (6.12)
1 + z2 − 2 x z 2
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 55

This allows one derive explicitly lower-order polynomials Pn :

P0 (x) = 1 ,
P1 (x) = x ,
3 1
P2 (x) = x2 − . (6.13)
2 2
In principle, computing higher-order terms in (6.12) directly one can derive higher-order
Legendre polynomials. However, the amount of computational work grows very quickly and
it is much better to proceed differently. Let us derive recurrent relations for Pn (x) starting
from their definition (6.11). To this end let us compute the derivative of (6.11) with respect
to z:
∞
d 1 d X
√ = Pn (x) z n . (6.14)
dz 1 + z − 2 x z
2 dz n=0

This gives
∞
X
x−z
√
3 = n Pn (x) z n−1 (6.15)
1+ z2 − 2xz n=0

or multiplying both sides by 1 + z 2 − 2 x z

X∞
x−z 2
√ = (1 + z − 2 x z) n Pn (x) z n−1 . (6.16)
1 + z2 − 2 x z n=0

Using again (6.11) in the left-hand side one finally gets

∞
X ∞
X
n 2
(x − z) Pn (x) z = (1 + z − 2 x z) n Pn (x) z n−1 . (6.17)
n=0 n=0

Now, since this equations must be satisfied identically (that is, for any z), the idea is to
equate the coefficients at equal powers of z on the left-hand and right-hand sides of (6.17).
Expanding (6.17) one gets
∞
X ∞
X
n
x Pn (x) z − Pn (x) z n+1
n=0 n=0
X∞ ∞
X ∞
X
= n Pn (x) z n−1 + n Pn (x) z n+1 − 2 n x Pn (x) z n . (6.18)
n=0 n=0 n=0

Using that
∞
X ∞
X
Pn (x) z n+1 = Pn−1 (x) z n , (6.19)
n=0 n=1
∞
X ∞
X
n Pn (x) z n−1 = (n + 1) Pn+1(x) z n , (6.20)
n=0 n=0
X∞ X∞
n Pn (x) z n+1 = (n − 1) Pn−1 (x) z n (6.21)
n=0 n=1
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 56

one gets
∞
X ∞
X
x Pn (x) z n − Pn−1 (x) z n
n=0 n=1
X∞ ∞
X ∞
X
n n
= (n + 1) Pn+1 (x) z + (n − 1) Pn−1 (x) z − 2 n x Pn (x) z n . (6.22)
n=0 n=1 n=0

Now, equating the coefficients at equal powers of z one gets

x P0 (x) = P1 (x), for n = 0 , (6.23)
x Pn (x) − Pn−1 (x) = (n + 1)Pn+1 (x) + (n − 1)Pn−1 (x) − 2nxPn (x), for n > 0 . (6.24)
From (6.13) one can see that the first equation is satisfied by P0 (x) and P1 (x) and, therefore,
gives no additional information. The second equation can be written as
(n + 1) Pn+1(x) = (2 n + 1) x Pn (x) − n Pn−1 (x), n ≥ 1. (6.25)
This equation allows one to compute Pn (x) for a given x if Pn−1 (x) and Pn−2 (x) are given.
This means that, starting from P0 (x) = 1 and P1 (x) = x as given by (6.13), Pn (x) can be
computed for any x and n ≥ 2. Below we will need also the following properties of Legendre
polynomials:
Pn (−x) = (−1)n Pn (x) , (6.26)
Z 1 h
0, n 6= m
Pn (x)Pm (x)dx = 2 . (6.27)
−1 2n+1
, n=m

Eq. (6.26) means that the Legendre polynomials Pn are even functions of x for even n and
odd functions for odd n. For n = 0 and n = 1 this relation can be seen directly. For n > 1
it is easy to prove (6.26) from the recurrent formula (6.25).

Exercise. Prove (6.26) from (6.25) for n > 1 using that (6.26) is correct for n = 0 and n = 1.

The proof of (6.27) and further properties of Legendre polynomials can be found, e.g., in
Chapter 8 of Abramowitz & Stegun (1965). Fig. 6.4 shows several first Legendre polynomials.
One can prove that −1 ≤ Pn (x) ≤ 1 for any n.

6.1.3 Expansion of the potential

Using the expansion (6.11) in (6.10) the potential (6.3) can be written as
X∞ Z
1 n
U= n+1
r ′ Pn (cos H) G σ(t, x′ ) d3x′ . (6.28)
n=0
r V

Here we have reached our goal – to get an expression for U(t, x) in which the dependence on
x is explicit – only partially. Indeed, the dependence on r = |x| is explicitly written since
all quantities under the integral are independent of r. On the other hand cos H under the
integral does depend on the orientation of x, that is, on the unit vector x/r. Such unit vector
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 57

Figure 6.4: Legendre polynomials Pn (x) are shown here for −1 ≤ x ≤ 1 and for n = 0, 1, . . . , 5
(red, green, brown, dark blue, magenta and light blue, respectively).

z
(r,l,q)

r
q

y
ll

Figure 6.5: Definition of the spherical coordinates: longitude λ (0 ≤ λ ≤ 2π) and co-latitude θ
(0 ≤ θ ≤ π).
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 58

can be characterized for example by the corresponding spherical coordinates. Introducing

longitude λ (0 ≤ λ ≤ 2π) and co-latitude θ (0 ≤ θ ≤ π) (see Fig. 6.5) – one has
cos H = cos θ cos θ′ + sin θ sin θ′ cos(λ − λ′ ), (6.29)
where λ and θ correspond to x while λ′ and θ′ correspond to x′ . This formula can be easily
derived using
 
r cos λ sin θ
x =  r sin λ sin θ  , (6.30)
r cos θ
x · x′
similar formula for x′ , and noting that cos H = , where ’·’ denotes scalar product of
r r′
two vectors. Furthermore, one has
Xn
(n − k)!
Pn (cos H) = (2 − δk0 ) Pnk (cos θ) Pnk (cos θ′ ) cos k(λ − λ′ ), (6.31)
(n + k)!
k=0

where
h
1, k=0
δk0 = (6.32)
0, k>0
and Pnk (x) are the associated Legendre polynomials defined as

k/2 dk
Pnk (x) = 1 − x2 Pn (x), 0 ≤ k ≤ n. (6.33)
dxk
Clearly, Pn0 (x) = Pn (x). Mathematical proof of (6.31) and further properties of the associ-
ated Legendre polynomials can be found, e.g., in Chapter 8 of Abramowitz & Stegun (1965).
Note that in the literature one can find different sign conventions related to the associated
Legendre polynomials, in particular regarding an additional factor (−1)k .
Substituting (6.31) into (6.28) one finally gets the expansion of the gravitational potential
of an extended body in the form
X∞ X n
Pnk (cos θ)
U= G n+1
(Cnk cos kλ + Snk sin kλ) . (6.34)
n=0 k=0
r
Coefficients Cnk and Snk are real numbers that fully characterize the gravitational potential
of the body and defined as
Z
(n − k)! n
Cnk = (2 − δk0 ) r ′ Pnk (cos θ′ ) cos kλ′ σ(t, x′ ) d3x′ , (6.35)
(n + k)! V
Z
(n − k)! n
Snk = (2 − δk0 ) r ′ Pnk (cos θ′ ) sin kλ′ σ(t, x′ ) d3 x′ . (6.36)
(n + k)! V
It is clear that Cnk and Snk do not depend on x, but only on the mass distribution inside
the body. Coefficient Sn0 = 0 for any n because of the factor sin kλ′ under the integral in its
definition (the latter factor vanishes for k = 0). One has, therefore, 2n + 1 coefficients Cnk
and Snk characterizing the gravitational field of an extended body for each n. Note that Cnk
and Snk are not dimensionless. Their dimensionality is kg · mn .
The expansion (6.34) for U(t, x) gives explicitly the dependence of U on the point x at
which U should be evaluated. This is done using spherical coordinates (r, λ, θ) of x. We
have thus achieved our goal.
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 59

6.2 First terms of the expansion

Summary: The mass of the body. The relations between the potential and the choice of the origin
and the orientation of the coordinate system.

Up to now the reference system where we described the body and its gravitational poten-
tial U was arbitrary. Here we show that the lower-order terms in (6.34) are closely related
to the choice of the origin and the orientation of the reference system. In the following we
X∞
consider that U = Un , where Un are all terms in (6.34) or (6.28) corresponding to a given
n=0
n, that is
Z
1 n
Un = G r ′ Pn (cos H) σ(t, x′ ) d3 x′
r n+1 V
n
X Pnk (cos θ)
= G (Cnk cos kλ + Snk sin kλ) . (6.37)
k=0
r n+1

1
Note that Un is the part of potential U falling off as for r going to infinity.
r n+1

6.2.1 The term for n = 0

For n = 0 the first line of (6.37) gives (note that P0 (cos H) = 1)
Z
1 GM
U0 = G σ(t, x′ ) d3 x′ = , (6.38)
r V r
R
where M = V σ(t, x′ ) d3x′ is the total mass of the body. From the second line of (6.37) we
see that for n = 0 there is only one coefficient C00 . Definition (6.35) again shows that

C00 = M . (6.39)

Thus, we get an important result: the main r −1 term of the gravitational potential of an
arbitrary extended body is the same as if the body would be a mass point with mass M
equal to the total mass of the extended body. This term cannot be affected by any change
of the reference system.

6.2.2 The terms for n = 1

x · x′
For n = 1 the first line of (6.37) gives (note that P1 (cos H) = cos H and r ′ cos H = )
r
Z
1 x · x′ GM
U1 = 2 G σ(t, x′ ) d3x′ = 3 x · xc , (6.40)
r V r r
Z
1
where xc = σ(t, x′ ) x′ d3 x′ are the coordinates of the center of mass of the body in
M V
the chosen reference system. On the other hand, considering the second line of (6.37) one
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 60

sees that U1 is characterized by C10 , C11 , and S11 (recall that Sn0 = 0 for any n). These
three coefficients are equivalent to the three components of xc . Indeed, it is easy to see that
C10 = M zc ,
C11 = M xc ,
S11 = M yc , (6.41)
where (xc , yc , zc ) are the components of vector xc . It is clear that choosing the reference
system in such a way that its origin coincides with the center of mass of the body under
study, one gets xc = 0. In this case, we have U1 = 0 and, therefore, C10 = C11 = S11 = 0.
Usually, this possibility is indeed used and one puts C10 , C11 , S11 to zero.

6.2.3 The terms for n = 2

For n = 2 the first line of (6.37) gives (note that P2 (cos H) = 32 cos2 H − 21 )
Z  
1 ′2 3 2 1
U2 = 3 G r cos H − σ(t, x′ ) d3 x′
r V 2 2
Z
1 3 (x · x′ )2 − r 2 r ′ 2
= 3G 2
σ(t, x′ ) d3 x′
r V 2 r
3 3
3 X X ˆij xi xj
=− G I 5
, (6.42)
2 i=1 j=1
r

where Iˆij is the trace-free part of the tensor of inertia I ij of the body. Namely, Iˆij =
I ij − 13 δ ij I kk , where I kk = I 11 + I 22 + I 33 is the trace of I ij and δ ij are the components of
identity matrix (δ ij = 1 for i = j and δ ij = 0 for i 6= j). The tensor of inertia I ij has its
usual definition
Z
ij


I = δ ij |x|2 − xi xj σ(t, x) d3 x . (6.43)
V

For n = 2 there are five coefficients Cnk and Snk : C20 , C21 , C22 , S21 , and S22 . On the
other hand the symmetric trace-free matrix Iˆij can be written as
 
a b c
Iˆij = −M R2  b d e  (6.44)
c e −a − d
and also has five independent components. Dimensionless numbers a, b, c, d and e fully
characterize the gravitational potential U2 and are directly related to the five coefficients
C20 , C21 , C22 , S21 , and S22 . One can demonstrate that
3
C20 = − (a + d) M R2 ,
2
C21 = c M R2 ,
1
C22 = (a − d) M R2 ,
4
S21 = e M R2 ,
1
S22 = b M R2 (6.45)
2
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 61

or inverting
1
M R2 a = − C20 + 2 C22 ,
3
2
M R b = 2 S22 ,
M R2 c = C21 ,
1
M R2 d = − C20 − 2 C22 ,
3
M R2 e = S21 . (6.46)
Since we consider rectangular Cartesian right-handed system there are only two sorts of
freedom to define it. First, the choice of the origin used for n = 1 above. Second, the choice
of spatial orientation of the axes. Clearly, the components of Iˆij depend on the orientation of
the coordinate system. Generally, a rotation in three-dimensional space is defined by three
parameters. Therefore, three coefficients among C20 , C21 , C22 , S21 , and S22 can be made
zero by choosing some special orientation of the coordinate system. For example, the matrix
Iˆij can be diagonalized by a suitable rotation to get C21 , S21 and S22 to zero (see (6.45) and
consider that for a diagonal matrix b = c = e = 0 in (6.44)). More detailed analysis shows
that the orientation can be chosen in such a way that C21 , S21 and either C22 or S22 vanish.
This possibility is usually not used since the orientation of the reference system is fixed from
the consideration of continuity and convenience.
Let us note that it makes no sense to consider separately terms Un for n > 2 since the
freedom in the definition of the coordinate system is already exhausted.

6.3 Symmetric bodies

Summary: Axial symmetry. Axial symmetry and the symmetry between the north and the south.
Spherical symmetry. Symmetry with respect to three coordinate planes.

In this Section we will simplify the general expansion (6.34) for the gravitational potential
U for the case of extended bodies having several sorts of symmetries. This consideration
allows one to understand which coefficients Cnk and Snk describe which properties of the
mass distribution within the body.

6.3.1 Axial symmetry

Let us first consider a body that is symmetric with respect to some axis and let us choose
that symmetry axis as z-axis of our coordinate system. Then the symmetry means that the
density σ does not depend on the spherical coordinate λ (see Fig. 6.5): σ 6= σ(λ). Therefore,
σ = σ(t, r, θ) and the integrals (6.35)–(6.36) can be written as
(n − k)!
Cnk = (2 − δk0 )
(n + k)!
Z R Z π  Z 2π 
′ ′ n+2 ′ ′ ′ ′ ′ ′ ′
× dr r dθ sin θ Pnk (cos θ ) σ(t, r , θ ) × dλ cos kλ , (6.47)
0 0 0
(n − k)!
Snk = (2 − δk0 )
(n + k)!
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 62
Z R Z π  Z 2π 
′ ′ n+2 ′ ′ ′ ′ ′ ′ ′
× dr r dθ sin θ Pnk (cos θ ) σ(t, r , θ ) × dλ sin kλ . (6.48)
0 0 0

Here we used the expression for the volume element in spherical coordinates: d3 x′ =
r ′ 2 sin θ′ dr ′ dθ′ dλ′. The last integral in (6.47)–(6.48) over dλ′ is zero for any k in Snk and
for k > 0 in Cnk . Therefore, for an axially symmetric body Cnk = 0 for k > 0 and Snk = 0
for any k. This holds for any n. It means that only coefficients Cn0 are not zero and fully
characterize the gravitational field of the body. The expansion of U in this case takes the
form
∞
X Pn (cos θ)
U= G Cn0 . (6.49)
n=0
r n+1

Therefore, we conclude that Cnk for any n and k > 0 and Snk for any n and k characterize
the deviation of the body from axial symmetry.

6.3.2 Axial symmetry and the symmetry about xy-plane

Let us now consider the case when, in addition to the axial symmetry (about z-axis), the
body is symmetric about the xy-plane (one can speak also of “symmetry between the north
and the south”). In this case density has the property that σ(t, r, θ) = σ(t, r, π − θ). Using
(6.47) for k = 0 one gets
Z R Z π
′ ′ n+2
Cn0 = 2π dr r dθ′ sin θ′ Pn (cos θ′ ) σ(t, r ′, θ′ ) , (6.50)
0 0

The integral over dθ′ can be split into two parts:

Z π/2 Z π
′ ′ ′ ′ ′
dθ sin θ Pn (cos θ ) σ(t, r , θ ) + dθ′ sin θ′ Pn (cos θ′ ) σ(t, r ′ , θ′ ) . (6.51)
0 π/2

Now, using the symmetry of the body one can show that these two integrals are related to
each other. Indeed, let us introduce θe′ = π − θ′ and replace θ′ through θe′ in the integral from
π/2 to π:
Z π
dθ′ sin θ′ Pn (cos θ′ ) σ(t, r ′, θ′ )
π/2
Z 0
=− dθe′ sin θe′ (−1)n Pn (cos θe′ ) σ(t, r ′ , θe′ )
π/2
Z π/2
= (−1) n
dθe′ sin θe′ Pn (cos θe′ ) σ(t, r ′, θe′ )
0
Z π/2
n
= (−1) dθ′ sin θ′ Pn (cos θ′ ) σ(t, r ′, θ′ ).
0

Here we used that dθ′ = −dθe′ , σ(t, r ′ , θ′ ) = σ(t, r ′ , π − θe′ ) = σ(t, r ′ , θe′ ), sin θ′ = sin(π −
θe′ ) = sin θe′ , cos θ′ = cos(π − θe′ ) = − cos θe′ , and that, according to (6.26), Pn (− cos θe′ ) =
(−1)n Pn (cos θe′ ). Finally, for the boundaries of integration we get that θ′ = π/2 and θ′ = π
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 63

correspond to θe′ = π/2 and θe′ = 0, respectively. In the last equality we simply replaced θe′
by θ′ as a change of notation.
It is, therefore, clear that with this symmetry Cn0 = 0 for odd n. The expansion of U
thus reads
X ∞
P2n (cos θ)
U= G C2n,0 . (6.52)
n=0
r 2n+1
We conclude that C2n+1,0 characterize the deviation of the (axially symmetric) body from
the symmetry about the xy-plane.

6.3.3 Spherical symmetry

We assume now an even stronger symmetry. Namely the spherical one. In this case the
density only depends on the radial coordinate and dies not depend on λ and θ: σ = σ(t, r).
In this case from (6.50) one has
Z R  Z π 
′ ′ n+2 ′ ′ ′ ′
Cn0 = 2π dr r σ(t, r ) × dθ sin θ Pn (cos θ ) , (6.53)
0 0
′
The integral over dθ can be written as
Z π Z π Z 1
′ ′ ′ ′ ′
dθ sin θ Pn (cos θ ) = − Pn (cos θ ) d cos θ = − Pn (s) ds, (6.54)
0 0 −1

where s = cos θ′ . According to (6.27) where we can put m = 0 and Pm (x) = 1 one gets
Z 1 
2, n = 0
Pn (s) ds = . (6.55)
−1 0, n ≥ 1
Therefore, only C00 = M (see Section 6.2.1) does not vanish for a spherical symmetric body
and the gravitational field of a spherically symmetric body reads
G C00 GM
U= = . (6.56)
r r
Note that we have now proved this formula for any spherically symmetric distribution of the
density σ = σ(t, r). This is a substantial generalization with respect to the case of point-like
bodies. We conclude that Cnk for n > 0 and Snk for any n and k characterize the deviation
of the body from the spherical symmetry.

6.3.4 Symmetry with respect to three coordinate planes

One more interesting case is a body symmetric with respect to all three coordinate planes:
xy-plane, xz-plane and yz-plane. For example, a triaxial ellipsoid possesses such a symmetry.
One can demonstrate that in this case the expansion of U reads
∞
X n
X
1
U =G C2n,2k P2n,2k (cos θ) cos 2kλ . (6.57)
n=0
r 2n+1 k=0

We conclude that Cnk with odd n and/or k and all Snk characterize the deviation of the body
from the symmetry about all three coordinate planes. The proof of (6.57) can be found, e.g.,
in Chandrasekhar (1987).
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 64

6.4 Spherical functions and the classification of the co-

efficients
Summary: Definition of spherical functions. The expansion of the gravitational potential in terms
of spherical functions. The principal part of the potential: spherically-symmetric gravitational
field. Zonal coefficients. Sectorial and tesseral coefficients.

Expansion (6.34) is a special case of general expansion of an arbitrary complex function of

spherical coordinates in terms of spherical functions. Namely, the factors Pnk (cos θ), cos kλ
and sin kλ can be combined into a single complex function:
s
2n + 1 (n − k)! ◦
Ynk (λ, θ) = (−1)k Pnk (cos θ) eı kλ . (6.58)
4π (n + k)!
◦ √
Note that eı kλ = cos kλ+ ı sin kλ, ı being the imaginary unit ı= −1. Functions Ynk
◦ ◦ ◦

depend only on λ and θ and are, therefore, defined on a sphere (the radius of which plays no
role). The numerical factor under the square root in (6.58) is only for a specific normalization
Z π Z 2π
that the integral dθ sin θ dλ Ynk (λ, θ) Yn∗′k′ (λ, θ) is equal to 1 if n = n′ and k = k ′
0 0
and vanishes otherwise (“∗” meaning complex conjugate). This factor does not play any role
in the following and will not be further discussed. Functions Ynk constitute full functional
basis on a sphere. It means that any complex function g(λ, θ) defined on a sphere can be
represented as
l
∞ X
X
g(λ, θ) = Alm Ylm (λ, θ), (6.59)
l=0 m=−l

where Alm are some complex numbers. If function g(λ, θ) is real, coefficients Alm have some
symmetry properties so that the whole
sum on the right-hand side of (6.59) remains real.
π
Let us also note that since cos k λ − 2k = sin kλ the real and imaginary
parts of Ynk are
π
related to each other in a simple way: Im Ynk (λ, θ) = Re Ynk λ − 2k ,θ .
Spherical functions are convenient to discuss the character of the coefficients Cnk and
Snk . The expansion (6.34) is often written in the form
( ∞  n !
GM X R
U= 1− Jn Pn (cos θ)
r n=2
r
∞ X n  n
)
X R

+ Pnk (cos θ) C nk cos kλ + S nk sin kλ , (6.60)
n=2 k=1
r

where R, as before, is the radius of a sphere encompassing the body and M = C00 is the
total mass of the body. Here one uses the coordinate system the origin of which coincides
with the center of mass of the body, so that all terms in (6.34) with n = 1 vanish and do
not appear in (6.60). Coefficients Jn , C nk and S nk are dimensionless real numbers that are
trivially related to Cnk and Snk :
1
Jn = − Cn0 , (6.61)
M Rn
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 65

1
C nk = Cnk , (6.62)
M Rn
1
S nk = Snk . (6.63)
M Rn
Coefficients Cnk and Snk (or Jn , C nk and S nk ) can be divided into four parts:
GM
1. The main part of the potential corresponds to C00 = M and to Y00 = √14π = const.
r
q
2n+1
2. The zonal harmonics Jn or Cn0 correspond to Yn0 = 4π
Pn (cos θ) that do not
depend on λ (see Fig. 6.6). For odd n the coefficients Jn characterize the oblateness
of the body. For even n they characterize the asymmetry of the body with respect to
the xy-plane (that is, asymmetry between the north and the south). This has been
discussed in Section 6.3.

3. The sectorial harmonics C nn and S nn correspond to Ynn and describe the effects that
depend only on longitude λ (see Fig. 6.6).

4. The tesseral harmonics C nk and S nk for n ≥ 2 and 1 ≤ k < n correspond to Ynk

with the same indices and describe the effects that depend both on longitude λ and
co-latitude θ. The word ’tessera’ means ’four’ or ’rectangle’. The surface of a sphere
with Ynk plotted on it appears to be divided into various “rectangles” (see Fig. 6.6).

In general, the coefficients with larger n and k describes finer details of the potential U.
CHAPTER 6. GRAVITATIONAL POTENTIAL OF AN EXTENDED BODY 66

n=2k=0 n=3k=0

n=3k=3 n=5k=5

n=3k=2 n=5k=3

Figure 6.6: Real parts of spherical functions Ynk for several sets of n and k plotted on a sphere.
The character of the functions is shown using the color code: dark gray corresponds to the areas
where Ynk is positive, light gray is used where Ynk is negative.
Chapter 7

Satellite Motion

7.1 Typical perturbations in satellite motion

Summary: Non-sphericity of the gravitational field of the Earth. Gravitational forces of the Moon
and the Sun. Planetary perturbations. Atmospheric drag. Light pressure. Magnetic field of the
Earth. Neutral and charged particles. Relativistic perturbations.

It is obvious that nowadays the task to model the motion of Earth satellites plays an
important role and even influences our every-day life. It is sufficient to mention communica-
tion satellites as well as the GPS, GLONASS and Galileo satellites. Also for scientific work
Earth satellites play important role. It is well known, for example, that the analysis of the
motion of specially designed satellites (LAGEOS, CHAMP, GRACE, GOCE, etc.) is the
best way to measure the gravitational field of the Earth.
If the Earth were spherically symmetric, had no atmosphere and were the only massive
body in the universe, the orbit of a satellite in the framework of Newtonian physics would
be one of the solutions of the two-body problem discussed in Chapter 2. None of these
conditions is met in reality. An orbit that would be an ellipse in the ideal case, is still close
to an ellipse in real world. However, several sorts of perturbations lead to time-dependent
osculating elements of the orbit. Which perturbations have to be taken into account to
describe the motion of a satellite depends both on the parameters of the orbit and on
the required accuracy. The main perturbations that must be taken into account for most
satellites are:

- the effects of the deviation of the Earth’s gravity from spherical symmetry as described
by coefficients Jn , C nk and S nk in (6.60);

- air drag caused by the motion of the satellite through the rarefied upper atmosphere;

- N-body perturbations due to the Moon and the Sun.

The first perturbation in this list is the largest one for asteroids flying at the altitudes
between approximately 300 and 30000 km above the surface of the Earth. Air drag is the
largest perturbation for lower satellites (whose with the altitudes lower than about 300 km).
For higher flying satellites (altitude over about 30000 km) the N-body perturbations from
the Moon and the Sun are the most important ones.

67
CHAPTER 7. SATELLITE MOTION 68

For high accuracy modelling (mostly important for scientific satellites of various kinds)
more subtle effects should be taken into account:

- the effects of the tidal deformations of the gravitational field (tidally induced temporal
variations of Jn , C nk and S nk ); both solid tides and ocean tides should be taken into
account here;

- N-body perturbations due to other bodies of the Solar system (mostly due to Jupiter
and Venus)

- light pressure from the light from the Sun and reflected light from the Earth; the effects
of umbra (shadow) and penumbra (the part of a shadow where the light source is only
partially blocked) should be taken into account; the “light” should be considered not
only in the “visual band”, but also in other wavebands (especially, infrared);

- effects of the general theory of relativity;

- effects of the magnetic field of the Earth;

- effects of cosmic particles (both neutral and charged ones).

Several perturbations mentioned above (e.g., light pressure and magnetic field) require to
model simultaneously also the dynamics of the attitude (spatial orientation) of the satel-
lite. In general, high-accuracy modelling of satellite motion is a truly complicated task.
In this Chapter we will discuss only two sorts of perturbations in their simplest form: the
perturbations due to the oblateness of the Earth and those due to the atmospheric drag.

7.2 Motion in the quadrupole field

Summary: The disturbing function for the oblate Earth. Solution in osculating elements. Secular
perturbations. Numerical example. Periodical perturbations.

For the Earth (and all major planets of the solar system) the second zonal harmonic J2
is significantly larger than all other coefficients in the expansion (6.60) of the gravitational
potential. Indeed, for the Earth J2 ≈ 1.083 × 10−3 while other coefficients are of the order
of 10−6 . For Earth satellites with the altitudes between about 300 and 30000 km the effects
of J2 is the largest perturbation effect. Let us consider it in more detail.

7.2.1 Disturbing potential due to J2

Let us neglect all other terms in (6.60) and write the gravitational potential of the Earth as
 
GM R2
U= 1 − J2 2 P2 (sin ϕ) , (7.1)
r r
π
where ϕ = − θ is the geographic latitude, R is now the equatorial radius of the Earth
2
(R ≈ 6378 km), and M is the mass of the Earth (GM ≈ 3.986004 × 1014 m3 /s2 ). The
gravitational potential does not explicitly depend on time (we assume J2 = const) and
CHAPTER 7. SATELLITE MOTION 69

does not depend on the longitude λ. The latter circumstance is related to the fact that if
we only consider J2 the gravitational field is axially symmetric (see Section 6.3.1 above).
The gravitational potential U is considered now as a function of coordinates r and ϕ. The
potential (7.1) can be written as
GM
U= + R, (7.2)
r
GM R2
R=− J2 2 P2 (sin ϕ), (7.3)
r r
where R can be considered as a perturbation of the two-body motion with potential GM r
.
Then the equations of motion of the satellite are given by (3.13) with m0 = 0 (the influence
of the satellite on the motion of the Earth is neglected) and with gradient of R on the
right-hand side.
The analysis of the motion can be done using the Lagrange equations discussed in Section
4.3. In order to use those equations we have to express R as function of orbital elements.
Position r = (x, y, z) of the satellite can be described using its geocentric distance r and
geographical longitude λ and latitude ϕ as
   
x r cos λ cos ϕ
 y  =  r sin λ cos ϕ  .
z r sin ϕ
Using (2.28), (2.30) and (2.41)–(2.42) one gets
z
sin ϕ = = sin i sin(v + ω), (7.4)
r
where i is the inclination of the orbit, ω is the argument of geocenter and v is the true
anomaly. Therefore, one gets
3 1 3 sin2 i − 2 3
P2 (sin ϕ) = sin2 i sin2 (v + ω) − = − sin2 i cos 2(v + ω) . (7.5)
2 2 4 4
The disturbing potential can than be written as
 
µ a3 3 sin2 i − 2 3 2
R= 3 3 − sin i cos 2(v + ω) , (7.6)
a r 4 4
2
µ = J2 G M R , (7.7)
where a is the semi-major axis or the orbit.

7.2.2 Exact consequence of the axially symmetric perturbation

Since
a 1 + e cos v
= ,
r 1 − e2
e being the eccentricity, and the true anomaly is a function of the mean anomaly M and
the eccentricity e only, one can see that the disturbing potential R depends on a, e, i, ω
and M, but does not depend on the longitude of the node Ω. This is again the consequence
CHAPTER 7. SATELLITE MOTION 70

of the fact that the assumed model for the Earth gravity is axially symmetric. Indeed, the
only effect of a change of Ω is a rotation of the orbit as a whole with respect to the z-axis
(see Fig. 2.3), but, in our model, the gravitational force does not depend on such a rotation.
One can demonstrate that if R = 6 R(Ω) the following combination of osculating elements
remains constant:
p
a (1 − e2 ) cos i = const . (7.8)

Exercise. Demonstrate that

d p  1 ∂R
a (1 − e2 ) cos i = , (7.9)
dt κ ∂Ω
where κ2 = n2 a3 , n being the mean motion. Hint: use the Lagrange equations (4.41)–(4.43).

p This means that as soon as the disturbing potential is axially symmetric, the combination
a (1 − e2 ) cos i remains constant.

7.2.3 Secular part of the disturbing potential

Now, let us confine our discussion to secular changes of the osculating elements, that is to the
changes that polynomially depend on time. To this end let us consider the Fourier expansion
of R in multiples of the mean anomaly M:
∞ 
X 
R = R0 + e k sin kM ,
Rk cos kM + R (7.10)
k=1

where Rk = Rk (a, e, i, ω) for k ≥ 0. We are only interested in the effect of R0 .

a3
Let us calculate the Fourier expansion of (7.6). Two functions should be expanded:
r3
a3
and cos 2(v + ω). Eq. (2.82) allows one to write
r3
∞
X ∞
X
a3 −3,0 −3,0
3
= Xk (e) cos kM = X0 (e) + 2 Xk−3,0(e) cos kM. (7.11)
r
k=−∞ k=1

The first equality uses the fact that the Hansen coefficients Xkn,m are real and, therefore, the
real parts and imaginary parts of (2.82) also hold:
 r n ∞
X
cos m v = Xkn,m(e) cos k M , (7.12)
a k=−∞
 r n X∞
sin m v = Xkn,m(e) sin k M . (7.13)
a k=−∞

For the second equality in (7.11) we use the well-known general symmetry property of the
Hansen coefficients

Xkn,m = X−k
n,−m
. (7.14)
CHAPTER 7. SATELLITE MOTION 71

◦ ◦
Indeed, the substitution m → −m, k → −k and ı→ − ı in (2.82) changes nothing and
n,0
demonstrate (7.14). For m = 0 as in (7.11) one has X−k = Xkn,0 so that each term with
k < 0 can be written together with the corresponding term with k > 0. This leads to (7.11).
◦
On the other hand, using (2.82) for n = −3 and m = 2 ( ı being the imaginary unit,
◦2
ı = −1)
 a 3 ◦
∞
X ◦
eı 2 v = Xk−3,2 (e) eı k M , (7.15)
r k=−∞

one can write

 a 3 ◦ ◦ X∞ ◦ ◦
ı
e e2v ı2ω
= Xk−3,2 (e) eı k M eı 2 ω , (7.16)
r k=−∞
or
 a 3 ◦
∞
X ◦
eı 2 (v+ω) = Xk−3,2 (e) eı (kM +2ω) (7.17)
r k=−∞

and, finally,
 a 3 ∞
X
cos 2(v + ω) = Xk−3,2 (e) cos(kM + 2ω)
r k=−∞
X
= X0−3,2 (e) cos 2ω + Xk−3,2 (e) cos(kM + 2ω) . (7.18)
k6=0

This means that

 
µ 3 sin2 i − 2 −3,0 3 2 −3,2
R0 = 3 X0 − sin i X0 cos 2ω . (7.19)
a 4 4
The Hansen coefficients Xkn,m (e) can be computed in a variety of ways (see, e.g. Giacaglia,
1976). One possible (but, generally speaking, inefficient) way is to compute the Hansen
coefficients from their definition as a Fourier coefficient:
Z 2π  n
n,m 1 1 − e2
Xk = cos(m v − k M) dM . (7.20)
2π 0 1 + e cos v
For k = 0 it is sufficient to use (2.31) to replace dM by dv as
3/2
(1 − e2 )
dM = dv (7.21)
(1 + e cos v)2
to get
Z 2π
1
n+3/2 cos m v
X0n,m = 1 − e2 dv . (7.22)
2π 0 (1 + e cos v)n+2
Computing this integral for n = −3, and m = 0 and m = 2 is trivial and one gets
X0−3,0 = (1 − e2 )−3/2 , (7.23)
X0−3,2 = 0 . (7.24)
With these expressions for the required Hansen coefficients one, finally, gets
µ 3 sin2 i − 2
R0 = (1 − e2 )−3/2 . (7.25)
a3 4
CHAPTER 7. SATELLITE MOTION 72

7.2.4 Secular perturbations of osculating elements

Since R0 depends only on a, e and i, one has
∂R0 ∂R0 ∂R0
= = = 0.
∂M 0 ∂ω ∂Ω
The Lagrange equations (4.41)–(4.43) give in this case

a(0) = const , (7.26)

e(0) = const , (7.27)
i(0) = const . (7.28)

This means that the semi-major axis a, eccentricity e and inclination i remain constant.
Index ’(0)’ in (7.26)–(7.28) and in equations for ω, Ω and M 0 below stresses that the equations
are valid only in the approximation R = R0 .
The partial derivatives of (7.25) read
∂R0 3 (2 − 3 sin2 i)

2 −3/2
= −µ 1 − e , (7.29)
∂a 4 a4
∂R0 3 e (2 − 3 sin2 i)

2 −5/2
= µ 1 − e , (7.30)
∂e 4 a3
∂R0 3 sin i cos i

2 −3/2
= −µ 1 − e . (7.31)
∂i 2 a3
Substituting these partial derivatives in the Lagrange equations (4.44)–(4.46) and considering
that semi-major axis a, eccentricity e and inclination i remain constant according to (7.26)–
(7.28) one get the following simple solution for other three osculating elements:

ω (0) (t) = nω (t − t0 ) + ω (0) (t0 ) , (7.32)

Ω(0) (t) = nΩ (t − t0 ) + Ω(0) (t0 ) , (7.33)
(0) (0)
M 0 (t) = nM 0 (t − t0 ) + M 0 (t0 ) , (7.34)

where the drift rates read

3µ (4 − 5 sin2 i)
nω = , (7.35)
4 κ a7/2 (1 − e2 )2
3µ cos i
nΩ = − , (7.36)
2 κ a7/2 (1 − e2 )2
3µ (2 − 3 sin2 i)
nM 0 = . (7.37)
4 κ a7/2 (1 − e2 )3/2
Thus, in the general case the averaged elliptical orbit of a satellite in the field of oblate Earth
characterized by J2 is an ellipse that (a) linearly precesses around the z-axis (linear change
of Ω), (b) linearly precesses in the orbital plane (linear change of ω), and (c) has the period
different by a constant from P = 2π n
, n = κ a−3/2 (linear change in M 0 ).
CHAPTER 7. SATELLITE MOTION 73

7.2.5 Analysis of the secular perturbations

√
Let us give a numerical value for the drifts for a particular satellite (we recall that κ = GM
and µ is defined by (7.7)):
 
 a = 12000 km  nω = 1.9◦ /day
i = 20◦ =⇒ nΩ = −1.4◦ /day . (7.38)
  ◦
e = 0.1 nM 0 = 0.9 /day

This demonstrates that the effects are significant even for relatively high satellites (a =
12000 km, e = 0.1 corresponds to the altitude of about 4500 km in the perigee).
From (7.35)–(7.37) one can see that choosing some specific inclinations i each of the drifts
(7.35)–(7.37) can be made zero. Indeed,

- For i = 90◦ one has no rotation of the orbital plane nΩ = 0. The orbits with i = 90◦ go
straightly over the poles of the Earth and are called polar orbits. These orbits are used
e.g., for special scientific satellites that have to observe some objects fixed in space.
For example, the mission GP-B (Gravity Probe B) aimed at high-accuracy testing
of general relativity, the future astrometric mission J-MAPS (Joint Milli-Arcsecond
Pathfinder Survey) of the US Naval Observatory and the mission GRACE to monitor
the gravity field of the Earth all use such polar orbits. The argument of perigee still
changes for such orbits. Fig. 7.1 illustrates the form of the orbit in the orbital plane.
2 2
- For i = arcsin √ ≈ 63◦ 26′ 5.82′′ and i = π − arcsin √ ≈ 116◦33′ 54.18′′ the perigee
5 5
ω does not precess since nω = 0. The orbits with such inclinations are used e.g. for
special communication satellites especially useful for polar regions.
Indeed, usual communication or broadcasting satellites are placed on the so-called geo-
stationary orbit. Geostationary orbit is a circular orbit (e = 0) with inclination i = 0
(orbital plane of the satellite coincides with the equatorial plane of the Earth) and
semi-major axis a = 42164 km chosen in such a way that the orbital period of the
satellite exactly coincides with the rotational period of the Earth. From the point of
view of an observer on the surface of the Earth a geostationary satellite is “seen” at a
fixed position on the sky. This allows the observer to orient his communication anten-
nas only once. The altitude of a geostationary satellite over the horizon is 90◦ − |ϕ|,
where ϕ is the geographical latitude of the observer. If the observer is situated in polar
regions with |ϕ| > 70◦ the geostationary satellites are “seen” too low over the horizon.
This would require significant power increase to guarantee reliable communications.
Already since the middle of 1960s a series of Soviet/Russian Molniya (Russian: “Light-
ning”) communications satellites for polar regions used a different principle. Molniya
satellites have orbits with inclination i ≈ 63◦ 26′ , high eccentricity e ≈ 0.722 and peri-
ods of 12 hours. Such an orbit is often called Molniya orbit. Because of high eccentricity
the satellite spend most of time far away from the Earth moving relatively slowly with
respect to the Earth surface. As a result one such satellite provides communication
and broadcasting services for about 8 hours per day. Three such satellites are sufficient
to make the service available at each moment of time.
Fig. 7.2 illustrates the form of the orbit with precessing Ω.
CHAPTER 7. SATELLITE MOTION 74

Figure 7.1: An orbit with precessing argument of pericenter ω is shown for 3, 10 and 36 periods
of motion. The change of ω is taken to be 10◦ per period.

r
2 2
- Finally, for i = arcsin ≈ 54◦ 44′ 8.20′′ and i = π − arcsin √ ≈ 125◦ 15′ 51.80′′ the
3 5
orbital period of the satellite is given by the unperturbed two-body relations P = 2π n
,
−3/2
n = κa since nM 0 = 0. No practical applications of such orbits are known.

7.2.6 Additional remarks

The problem of motion in the gravitational field of J2 can be solved exactly without averag-
ing. In addition to secular effects that were considered above one can consider all periodic
terms in (7.10). In a more elegant way this can be done by considering the whole perturba-
tion given by (7.6) and using simultaneously both true and mean anomalies in the resulting
formulas. This allows one to derive the formulas for the first-order variations of osculating
elements in closed form (see Roy, 2005, pp. 317–318). The first-order variations mean here
that terms that are at least quadratic in J2 are neglected.
The effects of other coefficients Jn , n ≥ 3 as well as C nk and S nk in (6.60) can be analyzed
in a similar way. One can show, for example, that the coefficients Jn with odd n lead only to
periodic effects. For Lageos all coefficients with n up to n = 50 should be taken into account.
The motion of missions like CHAMP, GRACE, GOCE, etc. is sensitive to the coefficients
with much higher values of n. Thus, the data of CHAMP is sensitive to the coefficients with
n up to n = 140, that of CRACE up to n = 180 and GOCE up to n = 250.

7.3 Atmospheric drag

Summary: The model of the perturbing force. Models for the atmospheric pressure. Gaussian
perturbation equations for the atmospheric drag. Averaging of the equations. The simplified
Gaussian perturbation equations for small eccentricities. Solution and its properties.

Let us now turn to the perturbations due to the atmospheric drag. Although the air in
the upper atmosphere has very low density, it influences significantly the motion of satellites
(recall that the velocity of satellites amounts to several km/s). For low satellites with
altitude 300 km and less the atmospheric drag is the largest perturbation. Air drag can be
CHAPTER 7. SATELLITE MOTION 75

Figure 7.2: An orbit with precessing longitude of the node Ω is shown for 3, 10 and 36 periods of
motion. The change of Ω is taken to be 10◦ per period.

understood as a result of friction between the air and the body of the satellite. As with all
other friction forces the mechanical energy of the system does not remain constant, but is
partially transformed into other sorts of energy (thermal one, etc.). It means that the air drag
force does not have a potential and should be considered using the Gaussian perturbation
equations (4.27)–(4.32). In general modelling of atmospheric drag is a very complex problem.
Here we consider the simplest case. More detailed discussion can be found e.g. in King-Hele
(1987).

7.3.1 Model for the drag force

We consider here the problem of motion of a body through a medium (fluid or gas). The
relative velocity of this motion is denoted v. It is well known that for the case of small
velocities v = |v| the flow of the media is laminar and the friction force is linearly proportional
to v: FS ∝ v. This is called Stokes’ friction model (George Gabriel Stokes 1819–1903).
For larger velocities v the flow becomes turbulent and the drag force is described by the
Newtonian friction model:
1
FN = ρ Cd S v 2 , (7.39)
2
where ρ is the air density, S is the area of the cross section of the body perpendicular to
the direction of motion, Cd is the numerical drag coefficient that depends on the geometry
of the body (see Fig. 7.3 for a few examples).
In reality the situation is more difficult. The effect of atmosphere results in several
effects: (1) air drag force with a magnitude given by (7.39) and directed along −v, (2) lift
force directed perpendicular to v (similar to the lift force for airplanes), (3) angular moment
leading to a rotation of the body. We completely ignore the lift forces since they are important
for higher densities of air when the motion in lower atmosphere is considered. As for the
angular moment, we simply average the force (7.39) over several rotations of the satellite.
Finally we get the disturbing force due to atmospheric drag in the form:
v
F = −C ρ v 2 , (7.40)
v
CHAPTER 7. SATELLITE MOTION 76

Cd = 0.4 Cd = 1.3 Cd = 0.3

Figure 7.3: Drag coefficients for some bodies: a sphere and a half-sphere moving along its symmetry
axis in two directions.

where C = 12 C d S, and an overline means averaging over the rotation of the satellite. We
consider parameter C to be known. Since velocity v can be computed from the osculating
elements of the orbit, the only unknown in (7.40) is the air density ρ.

7.3.2 Model for the air density

The air density can be taken from available models for the Earth’s atmosphere. These
models are based both of theoretical modelling and on the results of measurements of various
atmospheric parameters. The models are thus semi-empirical. Fig. 7.4 shows the air density
as a function of the height for different conditions. We see that the air density ρ at a given
altitude h depends on the time (it is higher during daytime and lower at night): at a given
altitude the air density can be different by a factor 2–10 depending on the time of the day.
It also depends on the level of solar activity (the higher is the solar activity the higher is
the density): ρ can be different by a factor 2–100 depending on the solar activity. There are
also a number of smaller effects: (1) the atmosphere rotates, (2) ρ depends not only on the
height h, but, to a smaller degree, also on the geographical longitude λ and the latitude φ.
Here we ignore all these complications and consider a simple model of exponential decay
of the density:
 
h
ρ = A exp − , (7.41)
H
where exp x ≡ ex , h is the height over the Earth’s surface, A = ρ(0) is the density at h = 0,
and H is height scale (that is, the height difference for which the density decreases by a
factor of e = 2.71 . . .). At the surface of the Earth A ≈ 1.3 kg/m3 and H ≈ 8 km. Such a
model describes the air density quite well if the considered region of heights is sufficiently
small. Density ρ is shown on Fig. 7.4 in logarithmic scale. Therefore, the model (7.41)
looks on Fig. 7.4 as a straight line. Clearly, model (7.41) can be used for a sequence of
layers h ∈ [hi , hi+1 ], i = 1, . . . , K with some boundaries hi and with constants A = Ai and
H = Hi depending on the layer. Such a layered model means that the curves on Fig. 7.4
are approximated by a piecewise linear curve. This is always possible, and the thinner are
the layers the better is the approximation. Thus, the height scale H can be considered as
function h. Both theoretical considerations and Fig. 7.4 demonstrate that dH/dh > 0 that
is, the density decreases slower than exponentially (slower than prescribed by (7.41)). In the
following we simplify the model further and consider that the orbit lies within one layer of
CHAPTER 7. SATELLITE MOTION 77

Figure 7.4: The density of air is shown as function of the altitude between 150 km and 1000 km
for different conditions (day and night, low, high and exceptionally high solar activity). The plot
is based on COSPAR International Reference Atmosphere 1972 (CIRA 1972).

h and the coefficients A and H in (7.41) are some given constants. Then we should express
ρ as function of osculating elements of the orbit. Considering that the radial distance of the
satellite r = R + h and, on the other side, r = a(1 − e cos E), cf. Eq.(2.40), we can write
       r
h r−R R
ρ = A exp − = A exp − = A exp exp −
H H H H
       
R a − a e cos E a−R a e cos E
= A exp exp − = A exp − exp . (7.42)
H H H H
Introducing
 
a−R
B = A exp − (7.43)
H
one finally gets

ρ = B exp (µ cos E) , (7.44)

ae
µ= . (7.45)
H
Note that B has a simple meaning: B is the air density ρ at the height equal to a − R,
i.e. at the mean altitude of the satellite. From the discussion above it is clear that this
model works better for orbit with smaller eccentricities e. Below we will see that the air
drag reduces the eccentricity e of the orbit. Therefore, the model (7.44) for ρ works better
when one considers later stages of the orbital evolution.
CHAPTER 7. SATELLITE MOTION 78

7.3.3 Gaussian perturbation equations in the axes aligned with

the velocity vector
We have seen above that the direction of the disturbing force is related to the direction of
the orbital velocity v. On the other hand, the Gaussian perturbation equations (4.27)–(4.32)
are expressed through the components S, T , and W of the disturbing force F as defined
by (4.7)–(4.9). The components (S, T, W ) are immediately related to the vector r of orbital
position, S being parallel to r. Let us introduce another coordinate system (τ, n, W ) instead
of (S, T, W ). Namely, let the axis τ be directly along the orbital velocity v and the axis n
perpendicular to τ in the instantaneous orbital plane given by the vectors of orbital position
r and velocity ṙ = v. Let Fτ and Fn be components of F in the axes τ and n, respectively.
One can write:
ṙ
Fτ = · F, (7.46)
|ṙ|
(r × ṙ) × ṙ
Fn = · F. (7.47)
|r × ṙ| |ṙ|
Our goal is now to express the components S and T of F as given by (4.7)–(4.8) in terms of
Fτ and Fn . The relation between the axes is shown on Fig. 7.5. It is clear from the Figure
that the components (S, T ) can be derived from (Fτ , Fn ) by a rotation in the orbital plane
by the angle −α, α being the angle between vectors r and ṙ. It means
S = Fτ cos α − Fn sin α ,
T = Fτ sin α + Fn cos α . (7.48)
where cos α and sin α are defined as
ṙ · r e sin E
cos α = =√ ,
|ṙ| |r| 1 − e2 cos2 E
√
ṙ × r 1 − e2
sin α = =√ . (7.49)
|ṙ| |r| 1 − e2 cos2 E
Here we used (2.41)–(2.42) and (2.44)–(2.45) for the components of r and ṙ, respectively.
Substituting this transformation into the Gaussian perturbation equations (4.27)–(4.32) one
get a version of the latter with components Fτ and Fn :
r
d 2 1 + e cos E
a= Fτ , (7.50)
dt n 1 − e cos E
√ r
d 2(1 − e2 ) cos E 1 − e2 sin E 1 − e cos E
e= √ Fτ − Fn , (7.51)
dt a n 1 − e2 cos2 E an 1 + e cos E
√ r
d 2 1 − e2 sin E cos E + e 1 − e cos E
ω= √ Fτ + Fn
dt na e 1 − e2 cos2 E ane 1 + e cos E
 
W
−r sin(v + ω) cot i √ , (7.52)
κ p
d 2 (1 − e3 cos E) sin E
M0 = − √ Fτ
dt ane 1 − e2 cos2 E
CHAPTER 7. SATELLITE MOTION 79

T
S
F!
!

Fn
O

Figure 7.5: The plot shows the relation between the components of the perturbing force in two
coordinate systems (S, T ) and (Fτ , Fn ). Both systems are rectangular Cartesian ones so that the
components are mutually orthogonal. Axis S is directed parallel to the positional vector r (from
the origin O to the current position P of the body). Axis τ is directed along the instantaneous
velocity of the body. The trajectory of the body is shown by a dotted curve and the arrow on that
curve shows the direction of motion. Axis τ is tangential to the trajectory at a given moment of
time. Finally, the angle between axes S and τ is denoted by α. Clearly, the relation between (S, T )
and (Fτ , Fn ) is a simple rotation given by (7.48).

√ r
1 − e2 (cos E − e) 1 − e cos E
− Fn . (7.53)
ane 1 + e cos E
The equations (4.29) and (4.31) for i and Ω remain unchanged.

7.3.4 Osculating elements for the air drag

The disturbing force from the air drag is given by (7.40) and (7.44)–(7.45). The components
of the disturbing force F can be written as:

Fτ = −C ρ v 2 , (7.54)
Fn = 0 , (7.55)
W =0. (7.56)

Therefore, the air drag does not change i and Ω, i.e. the orbital plane remains unchanged.
In (7.50)–(7.53) only the terms proportional to Fτ should be considered. From (2.44)–(2.45)
one gets
1 + e cos E
v 2 = n2 a2 . (7.57)
1 − e cos E
Let us also change the left-hand side of (7.50)–(7.53) from the derivatives with respect to
time t to the corresponding derivatives with respect to the eccentric anomaly E. Using (2.43)
CHAPTER 7. SATELLITE MOTION 80

for any element ǫ one has

 −1
dǫ dǫ dE 1 − e cos E dǫ
= = . (7.58)
dE dt dt n dt
In the following we are only interested in the form of the trajectory. Therefore, the element
M 0 plays no role and will be ignored below. Substituting (7.54) and (7.57) into (7.50)–(7.52)
and using (7.58) for all elements one gets:
r
da (1 + e cos E)3
= −2a2 C ρ , (7.59)
dE 1 − e cos E
r
de 1 + e cos E
= −2a C ρ (1 − e2 ) cos E , (7.60)
dE 1 − e cos E
r
dω −1


2 1/2 1 + e cos E
= −2a C ρ e 1−e sin E . (7.61)
dE 1 − e cos E
Now we can substitute the density model (7.44)–(7.45) and get
r
da (1 + e cos E)3
= −2a2 C B exp (µ cos E) , (7.62)
dE 1 − e cos E
r
de 2 1 + e cos E
= −2a C B (1 − e ) exp (µ cos E) cos E , (7.63)
dE 1 − e cos E
r
dω −1


2 1/2 1 + e cos E
= −2a C B e 1−e sin E exp (µ cos E) . (7.64)
dE 1 − e cos E

7.3.5 Averaged equations for the osculating elements

Let us investigate now only secular perturbations of a, e and ω. First, let us calculate the
mean value of the derivatives of these three elements over a period of motion as
  Z 2π Z π
dǫ 1 dǫ 1 dǫ
= dE = dE, (7.65)
dE 2 π 0 dE 2 π −π dE
where again ǫ is any of the elements. Since dω/dE given by (7.64) is an odd function of E
one has
 
dω
= 0. (7.66)
dE
This means that the osculating argument of perigee does not have secular variations, but
only periodic ones. For the other two elements one gets
  Z r
da a2 C B 2π (1 + e cos E)3
=− exp (µ cos E) dE , (7.67)
dE π 0 1 − e cos E
  Z 2π r
de aC B 1 + e cos E
=− (1 − e2 ) exp (µ cos E) cos E dE . (7.68)
dE π 0 1 − e cos E
CHAPTER 7. SATELLITE MOTION 81

7.3.6 Averaged osculating elements for small eccentricities

At this point we need one more sort of special functions. Namely, Im (x) defined as
Z 2π
1
Im (x) = exp(µ cos E) cos mE dE (7.69)
2π 0
are called modified Bessel functions of the first kind (sometimes hyperbolic Bessel functions
of the first kind). Many properties of these functions are known (see Abramowitz & Stegun
(1965), Chapter 9). They are related to the Bessel functions of the first kind given by (2.79)
as
◦−m ◦
Im (x) = ı Jm ( ı x), (7.70)

and can be calculated through the following power series

X∞  m+2l
1 1
Im (x) = x . (7.71)
l=0
l!(l + m)! 2

Functions Im (x) are shown on Fig. 7.6 for a few values of m. Below it will be important
that

1. Im (0) 6= 0 only for m = 0 (one has I0 (0) = 1);

2. Im (x) ≥ 0 for any m and x ≥ 0;

3. dIm (x)/dx ≥ 0 for any m and x ≥ 0.

These properties can be seen directly from (7.70).

Using Im (x) and considering the Taylor series
r
(1 + e cos E)3 3
= 1 + 2 cos E e + (1 + cos 2E) e2 + O(e3 ) , (7.72)
1 − e cos E 4
r
1 + e cos E 1 1
cos E = cos E + (1 + cos 2E) e + (3 cos E + cos 3E) e2 + O(e3 ) (7.73)
1 − e cos E 2 8
one gets
   
da 2 3 2 3
= −2a C B I0 (µ) + 2e I1 (µ) + e (I0 (µ) + I2 (µ)) + O(e ) , (7.74)
dE 4
  
de 1
= −2a C B (1 − e2 ) I1 (µ) + e (I0 (µ) + I2 (µ))
dE 2

1 2 3
+ e (3I1 (µ) + I3 (µ)) + O(e ) . (7.75)
8
Here we expanded the equations in power of eccentricity e and neglected terms of the order of
e3 and higher. It means that (7.74)-(7.75) are valid only for sufficiently small eccentricities.
In principle, higher-order terms in e can be calculated. However, for large e it is better to
solve (e.g., numerically) directly (7.67)-(7.68) or, directly, (7.62)-(7.64).
CHAPTER 7. SATELLITE MOTION 82

Figure 7.6: Modified Bessel functions of the first kind In (x) are shown for 0 ≤ x ≤ 3 and k = 0,
1, 2 and 3 (red, green, yellow, blue, magenta).

7.3.7 Discussion of the solution for osculating elements

Equations (7.74)-(7.75) are two differential equations for two unknown functions a(E) and
e(E). These equations are coupled. Note also that µ depends on both a and e as given by
(7.45). For a numerical example, the solution of (7.74)-(7.75) is shown on Fig. 7.7. Let us
make several remarks.

1. Since Im (µ) ≥ 0 the right-hand sides of (7.74)-(7.75) are non-positive. It means that
the averaged values of a and e are non-increasing function of E (and, therefore, of time
t). That is, both a and e becomes smaller during the evolution of the orbit.

2. For circular orbits (e = 0) the equations can be drastically simplified:

 
da
= −2a2 C B , (7.76)
dE
 
de
=0. (7.77)
dE

It means that the orbit remains circular with e = 0 and that the semi-major axis
decreases hyperbolically:

1
a(E) = 1 , (7.78)
a(E0 )
+ 2C B (E − E0 )

a(E0 ), being an integration constant, is the value of a for some initial moment E = E0 .
CHAPTER 7. SATELLITE MOTION 83

Figure 7.7: The left pane shows the variation of the osculating semi-major axes a (upper green
curve) and eccentricity e (lower red curve) over several orbital periods. On the right pane the orbit
of a satellite affected by the atmospheric drag is shown over the same interval of time. One can see
that both the semi-major axis and the eccentricity decrease with time. The effect is exaggerated
to make it better visible. Initial eccentricity is e = 0.3.

3. The smaller becomes the eccentricity e, the smaller is the absolute value of the right-
hand sides of (7.74)-(7.75). Here we used that dImdx(x) ≥ 0 for any n and x ≥ 0. This
means that the rate of change of both a and e decreases with time.

4. Finally, let us note that because of the air drag, satellites become faster and not slower
as one could expect from a friction force. This is of course related with the gravitational
character of motion: if semi-major axes a decreases, the velocity increases as a−1/2 .
Indeed, Eq. (7.57) can be written as
r r
GM 1 + e cos E
v= . (7.79)
a 1 − e cos E

Although the air drag as any friction force decreases the total mechanical energy of
the system Earth-satellite, the potential energy −GM/r of the system is transformed
into the kinetic energy v 2 /2 and the latter increases.
Chapter 8

Numerical integration of ordinary

differential equations

8.1 Basic notions

Summary: Euler step for the differential equations of the first order. Discretization. Three kinds
of errors: the local truncation error, the global error and the roundoff error.

8.2 Methods of numerical integration

Summary: The method of Taylor expansion. The Runge-Kutta method. Stepsize control for
the Runge-Kutta methods (Fehlberg method). The Runge-Kutta-Nyström method. Multistep
methods. Explicit and implicit methods. Predictor-corrector methods. Adams-Bashforth and
Adams-Moulton methods. Extrapolation methods.

8.3 Reliability of numerical integration

Summary: Close encounters. Regularization. Accuracy control.

84
Bibliography

Aarseth, S.J. 2003: Gravitational N-body Simulations: Tools and Algorithms (Cambridge:
Cambridge University Press)
Aarseth, S.J., Tout, Chr.A., Mardling, R.A. 2008: The Cambridge N-Body Lectures (Berlin:
Springer)
Abramowitz, M., Stegun, I.A. 1965: Handbook of Mathematical Functions with Formulas,
Graphs, and Mathematical Tables (New York: Dover)
Arnold, V.I., Kozlov, V.V., Neishtadt, A.I. 1997: Mathematical Aspects of Classical and
Celestial Mechanics (Berlin: Springer)
Beutler, G. 2005: Methods of Celestial Mechanics (Berlin: Springer)
Bottke, Jr., W.F., Vokrouhlický, D., Rubincam, D.P., Nesvorný, D. 2006: Annu. Rev. Earth
Planet. Sci. 34, 157
Bretagnon, P., Francou, G. 1988: Astron.Astrophys., 202, 309
Brouwer, D., Clemence, G.M. 1985: Methods of Celestial Mechanics, 2nd impr. (Orlando:
Academic Press)
Chandrasekhar, S. 1987: Ellipsoidal Figures of Equilibrium (New York: Dover Publications)
Chenciner, A., Montgomery, R. 2000: Ann. Of Math. 152, 881
Everhart, E. 1985: in Dynamics of Comets: Their Origin and Evolution, ed. A.Carusi
& G.B.Valsecci, Astrophysics and Space Science Library, 115, 185 (Dordrecht: Reidel)
Fienga, A., Manche, H., Laskar, J., Gastineau, M. 2008: Astron.Astrophys., 477, 315 (see
also arXiv:0906.2860)
Folkner, W.M. 2010: in ”Relativity in Fundamental Astronomy”, S.A. Klioner, P.K. Seidel-
mann & M.H. Soffel, eds. (Cambridge: Cambridge University Press)
Fukushima, T. 1996: Astron.J., 112, 2858
Giacaglia, G.E.O. 1976: Cel.Mech., 14, 515
Guthmann, A. (2000): Einführung in die Himmelsmechanik und Ephemeridenrechnung (Hei-
delberg: Spektrum)
King-Hele, D.G. 1987: Satellite Orbits in an Atmosphere: Theory and Application (London:
Blackie)
Krogh, F.T. 1994: Annals of Numerical Mathematics, 1, 423 (software DIVA/QIVA is avail-
able from http://mathalacarte.com)
Kuchynka, P., Laskar, J., Fienga, A., Manche, H. 2010: Astron.Atrophys., 514, A96
Marsden, B.G., Sekanina, Z., Yeomans, D.K. 1973: Astron.J., 78, 211
Montenbruck, O., Gill, E. 2000: Satellite Orbits: Models, Methods and Applications (Berlin:
Springer)

85
 BIBLIOGRAPHY 86

Moisson, X., Bretagnon, P. 2001: Cel.Mech.Dyn.Astron., 80, 205

Montenbruck, O., Gill, E. (2000): Satellite Orbits Springer, Berlin
Morbidelli, A. 2002: Modern Celestial Mechanics: Aspects of Solar System Dynamics (Lon-
don: Tyalor & Francis)
Moyer, T.D. 2003: Formulation for Observed and Computed Values of Deep Space Network
Data Types for Navigation (Hoboken: Wiley-Interscience)
Murray, C., D., Dermott, S.F. 1999: Solar System Dynamics (Cambridge: Cambridge Uni-
versity Press)
Newhall, X.X. 1989: Cel.Mech., 45, 305
Pitjeva, E.V. 2005: Solar System Research, 39, 176
Press, W.H., Teukolsky, S.A., Vetterling, W.T., Flannery, B.P. 2007: Numerical Recipes:
The Art of Scientific Computing, 3rd ed. (Cambridge: Cambridge University Press)
Schneider M. (1984): Himmelsmechanik (Mannheim: B.I.-Wissenschaftsverlag)
Standish, E.M., Williams, J.G. 2010: in Explanatory Supplement to the Astronomical Al-
manac, P.K. Seidelmann (ed.), 2nd edn., in press
Schwarz, H.R., Waldvogel, J. 1993: Numerische Mathematik (Stuttgart: B.G. Teubner)
Roy, A.E. (2005): Orbital Motion, 4th edn. (Bristol: Institute of Physics Publishing)

View publication stats