Random classical mechanics

Teo Banica
Department of Mathematics, University of Cergy-Pontoise, F-95000
Cergy-Pontoise, France. teo.banica@gmail.com
 2010 Mathematics Subject Classification. 60B20
Key words and phrases. Classical mechanics, Celestial mechanics

Abstract. This is an introduction to classical mechanics, and more specifically to the

gravitational N -body problem, viewed from a probabilistic viewpoint, in the N >> 0
regime. Our aim is to describe the various available techniques, and to distinguish,
probabilistically speaking, between order and chaos. At the level of applications, we insist
on various configurations met in the context of celestial mechanics, from thin interstellar
dust, up to macroscopic phenomena such as galaxies, and clusters of galaxies.
 Preface

What is the difference between order and chaos? Not very clear, and this is rather
a question for God who created this world. As humans, we can however make a few
observations, as to have a bit of an idea of what that divine “order” really means.

The simplest such observations, that we all made since childhood, concerm the life
surrounding us. Cat eats mouse, mouse eats grains, grains eat minerals from the soil.
Obviously, matter tends to organize into an hierarchic way, and this is where the beauty
and order of this world come from. Technically, all this can be fully understood too, via
organic chemistry, and then a lot of biology, culminating with Darwin.

The same phenomenon, matter organizing into an orderly and hierarchic way, can
be observed as well at smaller scales. Quarks manage to team up and form elementary
particles, then elementary particles team up and form atoms, and then atoms combine
and form molecules. And afterwards, the biggest and most advanced molecules, which
are the organic ones, team up and form cells, which cells organize into life.

Things are not over here, with the same phenomenon, matter organizing into an
orderly and hierarchic way, being observable as well at bigger scales. Here the unit is
typically some sort of big chunk of matter, and the organization which results from this,
via the laws of physics, is into Solar systems, galaxies, clusters of galaxies, and so on.

In this book we will be interested in these latter phenomena. The point indeed is that,
while seemingly being a bit far away from us, and from our potential scientific knowledge
and tools, these latter phenomena are in fact the simplest. To be more precise, passed
the formation and functioning of stars, which is certainly a difficult topic, involving all
sorts of physics, things up in the skies are basically governed by gravity. Thus, what we
need in order to understand these phenomena is just good old classical mechanics.

With the remark, however, that what we need is a probabilistic-flavored version of

classical mechanics. Indeed, in order to understand how matter organizes in a hierarchic
way, under the effect of gravity, we certainly have to deal with the gravitational N -body
problem, but viewed from a probabilistic viewpoint, in the N >> 0 regime.
3
4 PREFACE

So, this will be the plan for the book, first reviewing all sorts of basic questions and
methods from classical mechanics and dynamical systems, and fluid mechanics too, and
with a touch of statistical mechanics as well, with emphasis on the difference between
mechanical order and chaos, probabilistically speaking, and then moving ahead with an
exploration of various questions from celestial mechanics, armed with these tools, starting
with basic phenomena which are relentlessly at work, concerning interstellar dust, up to
truly macroscopic phenomena such as galaxies, and clusters of galaxies.
The book contains for the most very basic material, based of course on standard
classical mechanics, but we will recall the classical mechanics basics. Mathematically, we
will assume some familiarity with basic calculus, and probability theory, but again, we
will recall the basics here, whenever needed. Finally, in what regards the main problems
to be solved, these will be a bit more advanced, mostly at the level of classical mechanics
from the mid 20th century, but again, we will explain all these techniques.
I would like to thank my colleagues in Cergy, with everyone doing some sort of quantum
there, the discussions in the coffee room usually revolve around classical and celestial
mechanics, and other peaceful themes, and I learned many things in this way. Many
thanks to Vladimir Arnold for his wonderful books, and for his talks too. I had the
pleasure and privilege of attending one of them in the early 90s, as an undergraduate
student, and boy, that was some talk, that I will always have in mind.
Finally, many thanks go to my cats. We will be talking about order and chaos in
this book, with what happens in the skies being our main source of inspiration. But, no
really need to go up there for observing things and drawing conclusions: extreme order
and chaos can be an everyday matter, happening at home, right in front of your eyes.
 Contents

Preface 3

Part I. Classical mechanics 9

Chapter 1. Pendulum, clocks 11

1a. The pendulum 11
1b. Harmonic oscillators 15
1c. Friction, resonance 19
1d. Engineering, clocks 23
1e. Exercises 24

Chapter 2. Kepler and Newton 25

2a. Ellipses, conics 25
2b. Planets and comets 29
2c. Free falls, energy 37
2d. Rotating objects 43
2e. Exercises 48

Chapter 3. Lagrange, Hamilton 49

3a. Conservative forces 49
3b. Lagrange equations 58
3c. Mechanics, revised 61
3d. Hamilton equations 61
3e. Exercises 64

Chapter 4. The N body problem 65

4a. Center of mass 65
4b. The N body problem 76
4c. Lagrange points 77
4d. Friction and drag 78
4e. Exercises 84
5
6 CONTENTS

Part II. Order and chaos 85

Chapter 5. Noether theorem 87

5a. 87
5b. 87
5c. 87
5d. 87
5e. Exercises 87

Chapter 6. Bifurcation theory 89

6a. 89
6b. 89
6c. 89
6d. 89
6e. Exercises 89

Chapter 7. Singularities, ADE 91

7a. 91
7b. 91
7c. 91
7d. 91
7e. Exercises 91

Chapter 8. Fluids, granularity 93

8a. 93
8b. 93
8c. 93
8d. 93
8e. Exercises 93

Part III. Galaxies, clustering 95

Chapter 9. Celestial mechanics 97

9a. 97
9b. 97
9c. 97
9d. 97
9e. Exercises 97
 CONTENTS 7

Chapter 10. KAM theory 99

10a. 99
10b. 99
10c. 99
10d. 99
10e. Exercises 99

Chapter 11. Galaxies, clustering 101

11a. 101
11b. 101
11c. 101
11d. 101
11e. Exercises 101

Chapter 12. Numeric aspects 103

12a. 103
12b. 103
12c. 103
12d. 103
12e. Exercises 103

Part IV. Interstellar dust 105

Chapter 13. Star collapse 107

13a. 107
13b. 107
13c. 107
13d. 107
13e. Exercises 107

Chapter 14. Interstellar dust 109

14a. 109
14b. 109
14c. 109
14d. 109
14e. Exercises 109

Chapter 15. Charges, demons 111

8 CONTENTS

15a. 111
15b. 111
15c. 111
15d. 111
15e. Exercises 111
Chapter 16. Back to stars 113
16a. 113
16b. 113
16c. 113
16d. 113
16e. Exercises 113
Bibliography 115
Index 119
 Part I

Classical mechanics
 Video killed the radio star
Video killed the radio star
In my mind and in my car
We can’t rewind we’ve gone too far
 CHAPTER 1

Pendulum, clocks

1a. The pendulum

I bet that, perhaps a bit scared by the complexity of mechanical watches, you don’t
have a Rolex. For basic timekeeping and fashion matters, an average Casio will do.
And for a few dozen bucks more, they even have fancy models, with several dials and
everything, a bit like the Daytona. Add to this some cheap tools for replacing the battery,
which cannot be properly done with bare hands, and you’re ready to go.

This being said, in this book we will be mainly interested in mechanical order and
chaos, and the best starting point for all this are mechanical watches. What can be more
basic and orderly than a swinging pendulum keeping the time, of course skilfully adjusted
as to really work, and with this being the secret of mechanical watches.

Let us start our discussion with something very basic, namely:

Definition 1.1. A simple pendulum is a device of type
×

•m
consisting of a bob of mass m, attached to a rigid rod of length l.
In order to study the physics of the pendulum, which can easily lead to a lot of
complicated computations, when approached with bare hands, the most convenient is to
use the notion of energy. For a particle moving under the influence of a force F , the
position x, speed v and acceleration a are related by the following formulae:
v = ẋ , a = v̇ = ẍ , F = ma
The kinetic energy of our particle is then given by the following formula:
mv 2
T =
2
11
12 1. PENDULUM, CLOCKS

By differentiating with respect to time t, we obtain the following formula:

Ṫ = mv v̇ = mva = F v
Now by integrating, also with respect to t, this gives the following formula:
Z Z Z
T = F vdt = F ẋdt = F dx

But this suggests to define the potential energy V by the following formula, up to a
constant, with the derivative being with respect to the space variable x:
V ′ = −F
Indeed, we know from the above that we have T ′ = F , so if we define the total energy
to be E = T + V , then this total energy is constant, as shown by:
E′ = T ′ + V ′ = 0
Very nice all this, and by getting back now to the pendulum from Definition 1.1, we
can have this understood with not many computations involved, as follows:
Theorem 1.2. For a pendulum starting with speed v from the equilibrium position,
×

•m /
v

the motion will be confined if v 2 < 4gl, and circular if v 2 > 4gl.
Proof. There are many ways of proving this result, along with working out several
other useful related formulae, for which we will refer to the proof below, and with a quite
elegant approach to this, using no computations or almost, being as follows:
(1) Let us first examine what happens when the bob has traveled an angular distance
θ > 0, with respect to the vertical. The picture here is as follows:
×

l
l

•
; m
•m
 1A. THE PENDULUM 13

The distance traveled is then x = lθ. As for the force acting, this is Ftotal = mg
oriented downwards, with the component alongside x being given by:

F = −||Ftotal || sin θ
= −mg sin θ
x
= −mg sin
l
(2) But with this, we can compute the potential energy. With the convention that
this vanishes at the equilibrium position, V (0) = 0, we obtain the following formula:
x
′ ′
V = −F =⇒ V = mg sin
l  
 x
=⇒ V = mgl 1 − cos
l
=⇒ V = mgl(1 − cos θ)

(3) Alternatively, in case this sounds too wizarding, we can compute the potential
energy in the old fashion, by letting the bob fall, the picture being as follows:

l
l

•
h
•
The height of the fall is then h = l − l cos θ, and since for this fall the force is constant,
F = −mg, we obtain the following formula for the potential energy:

V ′ = −F =⇒ V ′ = mg
=⇒ V = mgh
=⇒ V = mgl(1 − cos θ)

Summarizing, one way or another we have our formula for the potential energy V .

(4) Now comes the discussion. The motion will be confined when the initial kinetic
energy, namely E = mv 2 /2, satisfies the following condition:

mv 2
E < sup V = 2mgl ⇐⇒ < 2mgl
θ 2
⇐⇒ v 2 < 4gl
14 1. PENDULUM, CLOCKS

In this case, the motion will be confined between two angles −θ, θ, as follows:

l
l
l

•m c •
; m
•m

To be more precise here, the two extreme angles −θ, θ ∈ (−π, π) can be explicitly
computed, as being solutions of the following equation:

mv 2
V =E ⇐⇒ mgl(1 − cos θ) =
2
v2
⇐⇒ 1 − cos θ =
2gl

(5) Regarding now the case v 2 > 4gl, here the bob will certainly reach the upwards
position, with the speed w > 0 there being given by the following formula:

mw2 mw2 mv 2
= E − 2mgl =⇒ = − 2mgl
2 2 2
=⇒ w2 = v 2 − 4gl
p
=⇒ w = v 2 − 4gl

Thus, with the convention in the statement for v, that is, going to the right, the
motion of the pendulum will be counterclockwise circular, and perpetual:

o w
•n
_
l

l
?
- •m /
v
 1B. HARMONIC OSCILLATORS 15

(6) Finally, in the case v 2 = 4gl, the bob will also reach the upwards position, but
with speed w = 0 there, and then, at least theoretically, will remain there:

•? m

l
=

•m /
v

(7) Actually, it is quite interesting in this latter situation, v 2 = 4gl, to further speculate
on what can happen, when making our problem more realistic. For instance, we can add
to our setting the assumption that when the bob is stuck on top, with speed 0, there is a
33% chance for it to keep going, to the left, a 33% chance for it to come back, to the right,
and a 33% chance for it to remain stuck. In this case there are infinitely many possible
trajectories, which are best investigated by using probability. Welcome to chaos.
(8) As a final comment, yes I know that the figures in (7) don’t add up to 100%.
This is because there is as well a remaining 1% possibility, where a relativistic black
cat appears, with a continuous effect on the bob, via a paw slap, when on top, with
speed w′ ∈ (0.3c, 0.7c), with c being the speed of light. In this case, the set of possible
trajectories becomes uncountable, and is again best investigated by using probability. □
And good news, done with the pendulum. Never ever will we be scared by it, all the
above was very nice, and the continuation of the book will be the same, nice too.

1b. Harmonic oscillators

Let us discuss now the motion of a particle near an equilibrium point. We have two
basic examples of such points provided by the pendulum, namely the downwards one,
which is stable, and the upwards one, which is unstable. But our discussion here will be
valid for any other types of particles moving, under the influence of forces.

As a first observation, our generalities about motion and energy provide us with:
Theorem 1.3. For a particle moving near an equilibrium point x = 0, the following
equivalent conditions must be satisfied, infinitesimally:
(1) The potential energy is V = kx2 /2, when assuming V (0) = 0.
(2) The force acting on our particle is F = −kx.
(3) The equation of motion is mẍ + kx = 0, with m being the mass.
16 1. PENDULUM, CLOCKS

Proof. This is something very standard, the idea being as follows:

(1) Let us start with some generalities regarding the potential energy V . Around any
given point, that we can choose by translation to be x = 0, we can write:
V ′′ (0)x2 V ′′′ (0)x3
V (x) = V (0) + V ′ (0)x + + + ...
2 6
By definition of V , we can assume V (0) = 0. Regarding now the second term, this
vanishes too, because our condition of equilibrium reads:
V ′ (0) = −F (0) = 0
Thus, with the above normalizations x = 0 and V (0) = 0 made, our general formula
above for V takes at equilibrium the following form, with k = V ′′ (0):
kx2
V (x) = + ...
2
Thus, we are led to the conclusion in the statement, provided that we are indeed in
the non-degenerate case, where k ̸= 0, which amounts in saying that F ′ (0) ̸= 0.
(2) This follows indeed from (1), and from V ′ = −F .
(3) This follows indeed from (2), and from F = ma = mẍ. □
The above result suggests formulating the following definition:
Definition 1.4. A harmonic oscillator is a particle moving as above, following
mẍ + kx = 0
with k ̸= 0. In the case k > 0, we say that we have a simple harmonic oscillator.
There the last convention comes from the fact that our oscillator is unstable when
k < 0, and stable k > 0, and it is in this latter case that we are mostly interested in.
And with this, stability depending on the sign of k, coming either from some abstract
reasoning along the lines of Theorem 1.3, or from the explicit formulae below.

Very nice, so let us solve now the equation of motion. We have here:
Theorem 1.5. The solutions of the equation of motion mẍ + kx = 0 for the harmonic
oscillators are as follows:
p
(1) x = aept + be−pt with p = −k/m, p when k < 0.
(2) x = c cos wt + d sin wt with w = k/m, when k > 0.
Proof. This is standard mathematics, as follows:
p
(1) Assume first that we are in the case k < 0. Here, with p = −k/m as in the
statement, the equation of motion takes the following form:
ẍ = p2 x
 1B. HARMONIC OSCILLATORS 17

But the functions ept , e−pt being solutions of this equation, by linearity we obtain that
the solutions are exactly the linear combinations of these two functions, as claimed.
p
(2) Assume now that we are in the case k > 0. Here, with w = k/m as in the
statement, the equation of motion takes the following form:
ẍ = −w2 x
But the functions cos wt, sin wt being solutions, by linearity we obtain that the solu-
tions are exactly the linear combinations of these two functions, as claimed. □
Observe that, as already mentioned above, the formulae that we obtained make it
clear that our oscillator is unstable when k < 0, and stable when k > 0. In fact, we have
the following simple consequences of the general formulae obtained above:
Proposition 1.6. The short and long time behavior of a harmonic oscillator, moving
according to mẍ + kx = 0, are as follows:
(1) In the case k < 0, with x = aept + be−pt as above, we have x ≃ (a + b) + p(a − b)t
for t > 0 small, and x ≃ aept for t >> 0.
(2) In the case k > 0, with x = c cos wt + d sin wt as above, we have x ≃ c + dwt for
t > 0 small, and there is no asymptotics for t >> 0.
Proof. This is indeed standard mathematics based on Theorem 1.5, as follows:
(1) In the case k < 0, with x = aept + be−pt as in Theorem 1.5, in the t > 0 small
regime we have indeed the following estimate, coming from ez ≃ 1 + z:
x = aept + be−pt
≃ a(1 + pt) + b(1 − pt)
= (a + b) + p(a − b)t
As for the other estimate, namely x ≃ aept for t >> 0, this is clear.
(2) In the case k > 0, with x = c cos wt + d sin wt as in Theorem 1.5, in the t > 0
small regime we have indeed the following estimate, coming from standard calculus:
x = c cos wt + d sin wt
≃ c(1 + o(t)) + dwt
≃ c + dwt
As for the last assertion, regarding the lack of asymptotics at k > 0 in the t >> 0
regime, this is clear, because neither cos, nor sin have such asymptotics, and the same
happens for any linear combination of them, with non-trivial coefficients. Of course,
interesting exercice for you to figure out all this, abstractly, this being not hard. □
As a last piece of mathematics, using this time complex numbers, we have:
18 1. PENDULUM, CLOCKS

Theorem 1.7. The solutions of the equation mẍ + kx = 0 are as follows, regardless
of the sign of k, and with a, b, c, d ∈ C chosen as to have x ∈ R:
p
(1) x = aept + be−pt , with p = −k/m. p
(2) x = c cos wt + d sin wt, with w = k/m.

Proof. This is standard complex number business, the idea being as follows:

(1) As before in the proof of Theorem 1.5 (1), but by using this time complex numbers,
we are led to the conclusion in the statement. With two remarks, namely:

– In the case k < 0 we have p ∈ R, and so a, b ∈ R, and we recover in this way

Theorem 1.5 (1) itself.
p
– As for the case k > 0, here we can write p = iw with w = k/m ∈ R, and the
formula that we get, according to the above, is as follows:

x = aeiwt + be−iwt

Now in order to have x ∈ R, which is the same as saying that x = x̄, we need:

a = b̄

Thus we can write a = c − id, b = c + id with c, d ∈ R, and with these substitutions

made, the solution found above takes the following form:

x = aeiwt + be−iwt
= (c − id)(cos wt + i sin wt) + (c + id)(cos wt − i sin wt)
= 2(c cos wt + d sin wt)

Thus at k > 0, up to a 2 factor, we obtain the formula from Theorem 1.5 (2).

(2) Things are similar here. Indeed, as before in the proof of Theorem 1.5 (2), we are
led to the conclusion in the statement, and with two remarks to be made, namely:

– In the case k > 0 we have w ∈ R, and so c, d ∈ R, and we recover in this way

Theorem 1.5 (2) itself.
 1C. FRICTION, RESONANCE 19
p
– As for the case k < 0, here we can write w = −ip with p = −k/m ∈ R, and the
formula that we get, according to the above, is as follows:
x = c cos wt + d sin wt
= c cos(−ipt) + d sin(−ipt)
= c cos(ipt) − d sin(ipt)
ei(ipt) + e−i(ipt) ei(ipt) − e−i(ipt)
= c· −d·
2 2i
−pt pt −pt pt
e +e e −e
= c· −d·
 2   2i  
1 d pt d −pt
= c+ e + c− e
2 i i
Now observe that in order to have x ∈ R, we must have c ± d/i ∈ R. Thus c ∈ R, and
d = if with f ∈ R, and with this latter substitution made, and then aftwerwards with
the notations a = (c + f )/2 and b = (c − f )/2, we obtain:
1
(c + f )ept + (c − f )e−pt


x =
2
= aept + be−pt
Thus at k < 0, we obtain the formula from Theorem 1.5 (1). □
Many other things can be said about the harmonic oscillators, in complement to what
was said above, and we will be back to this, on a regular basis, in what follows.

1c. Friction, resonance

Let us study now the damped oscillators, obtained by adding to the picture friction,
of some other extra force, which will bring us closer to engineering, and clocks. We recall
from Theorem 1.3 that the equation of the harmonic oscillator is as follows:
F = −kx
This equation can be modified as follows, with λ ̸= 0 being a constant:
F = −kx − λẋ
In terms of the equation of motion, the result is as follows:
Theorem 1.8. For a damped oscillator, which is subject by definition to a force of
type F = −kx − λẋ, the equation of motion is
mẍ + λẋ + kx = 0
with m being as before the mass.
20 1. PENDULUM, CLOCKS

Proof. This is clear indeed from F = ma = mẍ, which gives:

F = −kx − λẋ ⇐⇒ mẍ = −kx − λẋ
⇐⇒ mẍ + λẋ + kx = 0
Thus, we are led to the conclusion in the statement. □
Now let us try to solve the equation of motion. When looking for solutions of type
x = ept , with p ∈ C constant, the equation of motion takes the following form:
mp2 + λp + k = 0
But this is a degree 2 equation, that we can solve right away, and we get:
√
−λ ± λ2 − 4mk
p=
2m
It is convenient to write these solutions that we found, and the overall final result
about damping, in the following more convenient way:
Theorem 1.9. The generic solutions of mẍ + λẋ + kx = 0 are the real linear combi-
nations of the functions ept , with the parameter p ∈ C being given by
r
p
2 2
λ k
p = −γ ± γ − w , γ = , w=
2m m
with on the right w being the frequency of the usual, undamped oscillator.
Proof. This follows indeed from the formula of p found above, by dividing everything
by 2m. Observe that w is indeed the frequency of the usual, undamped oscillator. □
Now assume that we are in the case λ > 0, which is the most usual one, in practice,
meaning that our oscillator loses energy. We have then three cases, as follows:
Proposition 1.10. The oscillator damping with λ > 0, with this assumption meaning
that our oscillator loses energy over the time, can be of three types:
(1) Large damping,
p with λ > 0 being such that γ > w. Here the roots found above
p = −γ ± γ − w2 are both real, negative, and distinct.
2

(2) Small damping, with

p λ > 0 being such that γ < w. In this case the roots that we
found p = −γ ± γ 2 − w2 are complex and conjugate.
(3) Critical damping, with λ > 0 being such that γ = w. Here we have a double root,
which is real and negative, namely p = −γ.
Proof. All this is clear indeed from the formula found in Theorem 1.9. □
Now let us study more in detail the above three types of damping. In what regards
the large damping, things here are very simple and intuitive, as follows:
 1C. FRICTION, RESONANCE 21

Proposition 1.11. For large oscillator damping, the trajectory is given by

x = ae−γ+ t + be−γ− t
with the parameters γ+ > γ− > 0 being given by the formulae
p
γ± = γ ± γ 2 − w2
and with a, b ∈ R. With t >> 0 we have x ≃ be−γ− t → 0.
Proof. All this is indeed self-explanatory, and clear from the formula that we found
in Theorem 1.9, under the large damping assumption from Proposition 1.10 (1). □
In what regards the small damping, things here are again simple and intuitive, involv-
ing this time complex numbers and trigonometric functions, as follows:
Proposition 1.12. For small oscillator damping, the trajectory is given by
x = 2re−γt cos(ρt − θ)
with γ = λ/2m as before, with the parameter ρ > 0 being given by the formula
p
ρ = w2 − γ 2
and with r > 0 and θ ∈ R. With t >> 0 we have x → 0, exponentially.
Proof. Assume indeed that we are in the small damping regime, where λ > 0 is such
that γ < w. The roots that we found are then complex conjugate, as follows:
p
p = −γ ± iρ , ρ = w2 − γ 2
As for the solution itself, this is given by the following formula, with c, d ∈ C:
x = cep+ t + dep− t
= ce−γt+iρt + de−γt−iρt
= e−γt (ceiρt + de−iρt )
¯
Now in order to have x ∈ R, which is the same as saying x = x̄, we must have c = d.
Thus we can write c = re−iθ and d = reiθ with r > 0 and θ ∈ R, and we obtain:
x = e−γt (ceiρt + c̄e−iρt )
= 2e−γt Re(ceiρt )
= 2re−γt Re(ei(ρt−θ) )
= 2re−γt cos(ρt − θ)
Finally, the fact that we have indeed x → 0, exponentially, is clear. □
As for the critical damping case, the result here is as follows:
22 1. PENDULUM, CLOCKS

Theorem 1.13. For critical oscillator damping, the trajectory is given by

x = (a + bt)e−γt
with the parameter γ > 0 being given as usual by the formula
λ
γ=
2m
and with a, b ∈ R. With t >> 0 we have x ≃ bte−γt → 0.
Proof. Assume indeed that we are in the critical damping regime, where λ > 0 is
such that γ = w. In this case the roots that we found in Theorem 1.9 are both given by
p = −γ, and so that result provides us with solutions as follows, with a ∈ R:
x = ae−γt
Thus, we must find in this case some further solutions of the equation mẍ+λẋ+kx = 0,
and our claim is that the following functions are solutions too:
x = bte−γt
Indeed, let us verify this, under the above critical damping assumption. By linearity
it is enough to do this for b = 1, and here the derivatives are computed as follows:
x = te−γt
=⇒ ẋ = e−γt − γte−γt = (1 − γt)e−γt
=⇒ ẍ = −γe−γt − γ(1 − γt)e−γt = −γ(2 − γt)e−γt
We can verify now that the equation is indeed satisfied, as follows:
mẍ + λẋ + kx
= −mγ(2 − γt)e−γt + λ(1 − γt)e−γt + kte−γt
= (−2mγ + mγ 2 t + λ − λγt + kt)e−γt
= (mγ 2 t − λγt + kt)e−γt
= (mγ 2 − λγ + k)te−γt
= 0
Here we have used at the end the fact, that we know from Theorem 1.9 andpits proof,
that the solutions of the equation mp2 + λp + k = 0 are given by p = −γ ± γ 2 − w2 .
Indeed, in the present critical damping regime these solutions are both given by p = −γ,
and so substituting this particular value in the equation gives zero, as needed:
mγ 2 − λγ + k = 0
Thus, we are led to the conclusions in the statement. □
 1D. ENGINEERING, CLOCKS 23

As a conclusion to all this, the precise mathematics of the oscillator damping can be
explicitly worked out, in each of the three cases that can appear, large, small or critical,
and of particular interest is the mathematics and physics of the critical damping.

Moving forward, the next question is that of understanding what happens when a
periodic force is applied, and also to understand the related concept of resonance.

Let us start with formulating the phenomenon that we would like to understand. In
view of the applications that we have in mind, such as clocks, this is as follows:
Fact 1.14. In the case of an oscillator system subject to an external force F , which
depends only on time, the equation of motion is as follows:
mẍ + λẋ + kx = F
We are particularly interested, mathematically speaking, in the case where the external
force F is periodic, given by a formula of the following type:
F = φ cos νt
Also, physically speaking, we are particularly interested in the case where the external
force F is again periodic, but coming this time in the form of periodic impulses.
So, this will be our problematics, mainly motivated, as already said above, by devices
such as clocks. In practice now, there are three things to be done, as follows:

(1) Mathematical study of the first situation, which is the simplest one to examine,
where F = φ cos νt, with φ, ν ∈ R constants. The study here is quite routine.

(2) Next, study of the general situation where we have a periodic force F acting. Here
less things can be said, this framework being a bit too abstract and general.

(3) And finally, study of the case that we are mainly interested in, with the external
force F being again periodic, but coming this time in the form of periodic impulses.

So, this will be our plan, which does not look hard, mathematically speaking, and
with the extra requirement of understanding well the related concept of resonance.

1d. Engineering, clocks

Engineering, clocks.
24 1. PENDULUM, CLOCKS

1e. Exercises
Exercises:
Exercise 1.15.
Exercise 1.16.
Exercise 1.17.
Exercise 1.18.
Exercise 1.19.
Exercise 1.20.
Exercise 1.21.
Exercise 1.22.
Bonus exercise.
 CHAPTER 2

Kepler and Newton

2a. Ellipses, conics

Looking up in the sky, the first thing that you see is the Sun, seemingly moving
around the Earth on a circle. But, a more careful study reveals that this circle is rather
a deformed circle, called ellipsis. As for the other stars and planets, these have all sorts
of weird trajectories, but a more careful study reveals that, with due attention to what
the best “center” is, replacing our Earth, the trajectories are often ellipses too.

Summarizing, in order to understand what happens in the sky, we are basically led to
the ellipses, and their mathematics. And good news, a full theory of ellipses is available,
and this since the ancient Greeks, whose main findings were as follows:

Theorem 2.1. The ellipses, taken centered at the origin 0, and squarely oriented with
respect to Oxy, can be defined in 4 possible ways, as follows:
(1) As the curves given by an equation as follows, with a, b > 0:
 x 2  y 2
+ =1
a b
(2) Or given by an equation as follows, with q > 0, p = −q, and l ∈ (0, 2q):

d(z, p) + d(z, q) = l

(3) As the curves appearing when drawing a circle, from various perspectives:

⃝ → ?

(4) As the closed non-degenerate curves appearing by cutting a cone with a plane.

Proof. This might look a bit confusing, and you might say, what exactly is to be
proved here. Good point, and in answer, what is to be proved is that the above construc-
tions (1-4) give rise to the same class of curves. And this can be done as follows:

(1) To start with, let us draw a picture from what comes out of (1), which will be our
main definition for the ellipses, in what follows. Here that is, making it clear what the
25
26 2. KEPLER AND NEWTON

parameters a, b > 0 stand for, with 2a × 2b being the gift box size for our ellipsis:

•b

•−a •a

•−b

(2) Let us prove now that such an ellipsis has two focal points, as stated in (2). We
must look for a number r > 0, and a number l > 0, such that our ellipsis appears as
d(z, p) + d(z, q) = l, with p = (0, −r) and q = (0, r), according to the following picture:

•b

•−a •−r •r •a

•−b

(3) Let us first compute these numbers r, l > 0. Assuming that our result holds indeed
as stated, by taking z = (0, a), we see that the length l is:

l = (a − r) + (a + r) = 2a

As for the parameter r, by taking z = (b, 0), we conclude that we must have:

√ √
2 b2 + r2 = 2a =⇒ r = a2 − b2
 2A. ELLIPSES, CONICS 27

(4) With these observations made, let us prove now the result. Given l, r > 0, and
setting p = (0, −r) and q = (0, r), we have the following computation, with z = (x, y):
d(z, p) + d(z, q) = l
p p
⇐⇒ (x + r)2 + y 2 + (x − r)2 + y 2 = l
p p
⇐⇒ (x + r)2 + y 2 = l − (x − r)2 + y 2
p
⇐⇒ (x + r)2 + y 2 = (x − r)2 + y 2 + l2 − 2l (x − r)2 + y 2
p
⇐⇒ 2l (x − r)2 + y 2 = l2 − 4xr
⇐⇒ 4l2 (x2 + r2 − 2xr + y 2 ) = l4 + 16x2 r2 − 8l2 xr
⇐⇒ 4l2 x2 + 4l2 r2 + 4l2 y 2 = l4 + 16x2 r2
⇐⇒ (4x2 − l2 )(4r2 − l2 ) = 4l2 y 2
(5) Now observe that we can further process the equation that we found as follows:
4x2 − l2 4y 2
(4x2 − l2 )(4r2 − l2 ) = 4l2 y 2 ⇐⇒ =
l2 4r2 − l2
4x2 − l2 y2
⇐⇒ =
l2 r2 − l2 /4
!2
 x 2 y
⇐⇒ −1= p
2l r2 − l2 /4
!2
 x 2 y
⇐⇒ + p =1
2l r2 − l2 /4
(6) Thus, our result holds indeed, and with√the numbers l, r > 0 appearing, and no
surprise here, via the formulae l = 2a and r = a2 − b2 , found in (3) above.

(7) Getting back now to our theorem, we have two other assertions there at the end,
labelled (3,4). But, thinking a bit, these assertions are in fact equivalent, and in what
concerns us, we will rather focus on (4), which looks more mathematical. And in what
regards this assertion (4), this can be established indeed, by doing some 3D computations,
that we will leave here as an instructive exercise, for you. And with the promise that we
will come back to this in a moment, with a full proof, in a more general setting. □
All this is very nice, but before getting into physics, with some explanations for the fact
that planets travel indeed on ellipses, which is something that we must surely understand,
before going with some further math, let us settle as well the question of wandering
asteroids. Observations show that these can travel on parabolas and hyperbolas, so what
we need as mathematics is a unified theory of ellipses, parabolas and hyperbolas. And
fortunately, this theory exists, also since the ancient Greeks, summarized as follows:
28 2. KEPLER AND NEWTON

Theorem 2.2. The conics, which are the algebraic curves of degree 2 in the plane,
n o
2
C = (x, y) ∈ R P (x, y) = 0
with deg P ≤ 2, appear modulo degeneration by cutting a 2-sided cone with a plane, and
can be classified into ellipses, parabolas and hyperbolas.
Proof. This follows by further building on Theorem 2.1, as follows:
(1) Let us first classify the conics up to non-degenerate linear transformations of the
plane, which are by definition transformations as follows, with det A ̸= 0:
   
x x
→A
y y
Our claim is that as solutions we have the circles, parabolas, hyperbolas, along with
some degenerate solutions, namely ∅, points, lines, pairs of lines, R2 .
(2) As a first remark, it looks like we forgot precisely the ellipses, but via linear
transformations these become circles, so things fine. As a second remark, all our claimed
solutions can appear. Indeed, the circles, parabolas, hyperbolas can appear as follows:
x2 + y 2 = 1 , x2 = y , xy = 1
As for ∅, points, lines, pairs of lines, R2 , these can appear too, as follows, and with
our polynomial P chosen, whenever possible, to be of degree exactly 2:
x2 = −1 , x2 + y 2 = 0 , x2 = 0 , xy = 0 , 0=0
Observe here that, when dealing with these degenerate cases, assuming deg P = 2
instead of deg P ≤ 2 would only rule out R2 itself, which is not worth it.
(3) Getting now to the proof of our claim in (1), classification up to linear transfor-
mations, consider an arbitrary conic, written as follows, with a, b, c, d, e, f ∈ R:
ax2 + by 2 + cxy + dx + ey + f = 0
Assume first a ̸= 0. By making a square out of ax2 , up to a linear transformation in
(x, y), we can get rid of the term cxy, and we are left with:
ax2 + by 2 + dx + ey + f = 0
In the case b ̸= 0 we can make two obvious squares, and again up to a linear transfor-
mation in (x, y), we are left with an equation as follows:
x2 ± y 2 = k
In the case of positive sign, x2 + y 2 = k, the solutions are the circle, when k ≥ 0, the
point, when k = 0, and ∅, when k < 0. As for the case of negative sign, x2 − y 2 = k,
which reads (x − y)(x + y) = k, here once again by linearity our equation becomes xy = l,
which is a hyperbola when l ̸= 0, and two lines when l = 0.
 2B. PLANETS AND COMETS 29

(4) In the case b ̸= 0 the study is similar, with the same solutions, so we are left with
the case a = b = 0. Here our conic is as follows, with c, d, e, f ∈ R:
cxy + dx + ey + f = 0
If c ̸= 0, by linearity our equation becomes xy = l, which produces a hyperbola or two
lines, as explained before. As for the remaining case, c = 0, here our equation is:
dx + ey + f = 0
But this is generically the equation of a line, unless we are in the case d = e = 0,
where our equation is f = 0, having as solutions ∅ when f ̸= 0, and R2 when f = 0.
(5) Thus, done with the classification, up to linear transformations as in (1). But this
classification leads to the classification in general too, by applying now linear transforma-
tions to the solutions that we found. So, done with this, and very good.
(6) It remains to discuss the cone cutting. By suitably choosing our coordinate axes
(x, y, z), we can assume that our cone is given by an equation as follows, with k > 0:
x2 + y 2 = kz 2
In order to prove the result, we must in principle intersect this cone with an arbitrary
plane, which has an equation as follows, with (a, b, c) ̸= (0, 0, 0):
ax + by + cz = d
(7) However, before getting into computations, observe that what we want to find is a
certain degree 2 equation in the above plane, for the intersection. Thus, it is convenient
to change the coordinates, as for our plane to be given by the following equation:
z=0
(8) But with this done, what we have to do is to see how the cone equation x2 +y 2 = kz 2
changes, under this change of coordinates, and then set z = 0, as to get the (x, y) equation
of the intersection. But this leads, via some thinking or computations, to the conclusion
that the cone equation x2 + y 2 = kz 2 becomes in this way a degree 2 equation in (x, y),
which can be arbitrary, and so to the final conclusion in the statement. □

Many other things can be said about conics. We will be back to this.

2b. Planets and comets

You surely know a bit about gravity, including the findings of Kepler and Newton.
The result here, and its proof, which is the pride of mathematics, physics, and human
knowledge in general, is the following famous theorem:
30 2. KEPLER AND NEWTON

Theorem 2.3 (Kepler, Newton). Planets and other celestial bodies move around the
Sun on conics, that is, on curves of type
n o
C = (x, y) ∈ R2 P (x, y) = 0
with P ∈ R[x, y] being of degree 2. The same is true for any body moving around another
body, provided that we are not in the situation of a free fall.
Proof. This is something very standard, the idea being as follows:
(1) According to observations and calculations performed over the centuries, since the
ancient times, and first formalized by Newton, following some groundbreaking work of
Kepler, the force of attraction between two bodies of masses M, m is given by:
Mm
||F || = G · 2
d
Here d is the distance between the two bodies, and G ≃ 6.674 × 10−11 is a constant.
Now assuming that M is fixed at 0 ∈ R3 , the force exterted on m positioned at x ∈ R3 ,
regarded as a vector F ∈ R3 , is given by the following formula:
x GM m x GM mx
F = −||F || · =− · = −
||x|| ||x||2 ||x|| ||x||3
But F = ma = mẍ, with a = ẍ being the acceleration, second derivative of the
position, so the equation of motion of m, assuming that M is fixed at 0, is:
GM x
ẍ = −
||x||3
(2) Obviously, the problem happens in 2 dimensions, and you can even find, as an
exercise, a formal proof of that, based on the above equation. Now here the most conve-
nient is to use standard x, y coordinates, and denote our point as z = (x, y). With this
change made, and by setting K = GM , the equation of motion becomes:
Kz
z̈ = −
||z||3
In other words, in terms of the coordinates x, y, the equations are:
Kx Ky
ẍ = − 2 2 3/2
, ÿ = − 2
(x + y ) (x + y 2 )3/2
(3) Let us begin with a simple particular case, that of the circular solutions. To be
more precise, we are interested in solutions of the following type:
x = r cos αt , y = r sin αt
In this case we have ||z|| = r, so our equation of motion becomes:
Kz
z̈ = − 3
r
 2B. PLANETS AND COMETS 31

On the other hand, differentiating x, y leads to the following formula:

z̈ = (ẍ, ÿ) = −α2 (x, y) = −α2 z
Thus, we have a circular solution when the parameters r, α satisfy:
r 3 α2 = K
(4) In the general case now, the problem can be solved via some calculus. Let us write
indeed our vector z = (x, y) in polar coordinates, as follows:
x = r cos θ , y = r sin θ
We have then ||z|| = r, and our equation of motion becomes, as in (3):
Kz
z̈ = − 3
r
Let us differentiate now x, y. By using the standard calculus rules, we have:
ẋ = ṙ cos θ − r sin θ · θ̇
ẏ = ṙ sin θ + r cos θ · θ̇
Differentiating one more time gives the following formulae:
ẍ = r̈ cos θ − 2ṙ sin θ · θ̇ − r cos θ · θ̇2 − r sin θ · θ̈
ÿ = r̈ sin θ + 2ṙ cos θ · θ̇ − r sin θ · θ̇2 + r cos θ · θ̈
Consider now the following two quantities, appearing as coefficients in the above:
a = r̈ − rθ̇2 , b = 2ṙθ̇ + rθ̈
In terms of these quantities, our second derivative formulae read:
ẍ = a cos θ − b sin θ
ÿ = a sin θ + b cos θ
(5) We can now solve the equation of motion from (4). Indeed, with the formulae that
we found for ẍ, ÿ, our equation of motion takes the following form:
K
a cos θ − b sin θ = − 2 cos θ
r
K
a sin θ + b cos θ = − 2 sin θ
r
But these two formulae can be written in the following way:
 
K
a + 2 cos θ = b sin θ
r
 
K
a + 2 sin θ = −b cos θ
r
32 2. KEPLER AND NEWTON

By making now the product, and assuming that we are in a non-degenerate case,
where the angle θ varies indeed, we obtain by positivity that we must have:
K
a+ =b=0
r2
(6) We are almost there. Let us first examine the second equation, b = 0. Remember-
ing who b is, from (4), this equation can be solved as follows:
b=0 ⇐⇒ 2ṙθ̇ + rθ̈ = 0
θ̈ ṙ
⇐⇒ = −2
θ̇ r
⇐⇒ (log θ̇) = (−2 log r)′
′

⇐⇒ log θ̇ = −2 log r + c
λ
⇐⇒ θ̇ = 2
r
As for the first equation the we found, namely a + K/r2 = 0, remembering from (4)
that a was by definition given by a = r̈ − rθ̇2 , this equation now becomes:
λ2 K
r̈ −
+ 2 =0
r3 r
(7) As a conclusion to all this, in polar coordinates, x = r cos θ, y = r sin θ, our
equations of motion are as follows, with λ being a constant, not depending on t:
λ2 K λ
r̈ = − , θ̇ =
r3 r2 r2
Even better now, by writing K = λ2 /c, these equations read:
λ2 1 1
 
λ
r̈ = 2 − , θ̇ = 2
r r c r
(8) As an illustration, let us quickly work out the case of a circular motion, where r
is constant. Here r̈ = 0, so the first equation gives c = r. Also we have θ̇ = α, with:
λ
α=
r2
Assuming θ = 0 at t = 0, from θ̇ = α we obtain θ = αt, and so, as in (3) above:
x = r cos αt , y = r sin αt
Observe also that the condition found in (3) is indeed satisfied:
λ2 λ2
r 3 α2 = = =K
r c
 2B. PLANETS AND COMETS 33

(9) Back to the general case now, our claim is that we have the following formula, for
the distance r = r(t) as function of the angle θ = θ(t), for some ε, δ ∈ R:
c
r=
1 + ε cos θ + δ sin θ
Let us first check that this formula works indeed. With r being as above, and by using
our second equation found before, θ̇ = λ/r2 , we have the following computation:
c(ε sin θ − δ cos θ)θ̇
ṙ =
(1 + ε cos θ + δ sin θ)2
λc(ε sin θ − δ cos θ)
= 2
r (1 + ε cos θ + δ sin θ)2
λ(ε sin θ − δ cos θ)
=
c
Thus, the second derivative of the above function r is given, as desired, by:
λ(ε cos θ + δ sin θ)θ̇
r̈ =
c
2
λ (ε cos θ + δ sin θ)
= 2
2
 r c
λ 1 1
= 2 −
r r c
(10) The above check was something quite informal, and now we must prove that our
formula is indeed the correct one. For this purpose, we use a trick. Let us write:
1
r(t) =
f (θ(t))
Abbreviated, and by always reminding that f takes θ = θ(t) as variable, this reads:
1
r=
f
With the convention that dots mean as usual derivatives with respect to t, and that
the primes will denote derivatives with respect to θ = θ(t), we have:
f ′ θ̇ f′ λ
ṙ = − = − · = −λf ′
f2 f 2 r2
By differentiating one more time with respect to t, we obtain:
λ λ2 ′′
r̈ = −λf ′′ θ̇ = −λf ′′ · = − f
r2 r2
34 2. KEPLER AND NEWTON

On the other hand, our equation for r̈ found in (7) reads:

λ2 1 1 λ2
   
1
r̈ = 2 − = 2 f−
r r c r c
Thus, in terms of f = 1/r as above, our equation for r̈ simply reads:
1
f ′′ + f =
c
But this latter equation is elementary to solve. Indeed, both functions cos t, sin t satisfy
g” + g = 0, so any linear combination of them satisfies as well this equation. But the
solutions of f ′′ + f = 1/c being those of g ′′ + g = 0 shifted by 1/c, we obtain:
1 + ε cos θ + δ sin θ
f=
c
Now by inverting, we obtain the formula announced in (9), namely:
c
r=
1 + ε cos θ + δ sin θ
(11) But this leads to the conclusion that the trajectory is a conic. Indeed, in terms
of the parameter θ, the formulae of the coordinates are:
c cos θ
x=
1 + ε cos θ + δ sin θ
c sin θ
y=
1 + ε cos θ + δ sin θ
But these are precisely the equations of conics in polar coordinates.
(12) To be more precise, in order to find the precise equation of the conic, observe
that the two functions x, y that we found above satisfy the following formula:
c2 (cos2 θ + sin2 θ)
x2 + y 2 =
(1 + ε cos θ + δ sin θ)2
c2
=
(1 + ε cos θ + δ sin θ)2
On the other hand, these two functions satisfy as well the following formula:
2

2
c ε cos θ + δ sin θ − (1 + ε cos θ + δ sin θ)
(εx + δy − c)2 =
(1 + ε cos θ + δ sin θ)2
c2
=
(1 + ε cos θ + δ sin θ)2
We conclude that our coordinates x, y satisfy the following equation:
x2 + y 2 = (εx + δy − c)2
But what we have here is an equation of a conic, as claimed. □
 2B. PLANETS AND COMETS 35

Still with me, I hope, after all these computations. For further applications, here is a
sort of “best of” the formulae found in the proof of Theorem 2.3:
Theorem 2.4 (Kepler, Newton). In the context of a 2-body problem, with M fixed at
0, and m starting its movement from Ox, the equation of motion of m, namely
Kz
z̈ = −
||z||3
with K = GM , and z = (x, y), becomes in polar coordinates, x = r cos θ, y = r sin θ,
λ2 1 1
 
λ
r̈ = 2 − , θ̇ = 2
r r c r
for some λ, c ∈ R, related by λ2 = Kc. The value of r in terms of θ is given by
c
r=
1 + ε cos θ + δ sin θ
for some ε, δ ∈ R. At the level of the affine coordinates x, y, this means
c cos θ c sin θ
x= , y=
1 + ε cos θ + δ sin θ 1 + ε cos θ + δ sin θ
with θ = θ(t) being subject to θ̇ = λ2 /r, as above. Finally, we have
x2 + y 2 = (εx + δy − c)2
which is a degree 2 equation, and so the resulting trajectory is a conic.
Proof. As already mentioned, this is a sort of “best of” the formulae found in the
proof of Theorem 2.3. And in the hope of course that we have not forgotten anything.
Finally, let us mention that the simplest illustration for this is the circular motion, and
for details on this, not included in the above, we refer to the proof of Theorem 2.3. □
As a first question, we would like to understand how the various parameters appearing
above, namely λ, c, ε, δ, which via some basic math can only tell us more about the shape
of the orbit, appear from the initial data. The formulae here are as follows:
Theorem 2.5. In the context of Theorem 2.4, and in polar coordinates, x = r cos θ,
y = r sin θ, the initial data is as follows, with R = r0 :
c
r0 = , θ0 = 0
1+ε
√ √
δ K Kc
ṙ0 = − √ , θ̇0 =
c R2
εK 4δK
r̈0 = 2 , θ̈0 = 2
R R
The corresponding formulae for the affine coordinates
√ x, y can be deduced from this. Also,
the various motion parameters c, ε, δ and λ = Kc can be recovered from this data.
36 2. KEPLER AND NEWTON

Proof. We have several assertions here, the idea being as follows:

(1) As mentioned in Theorem 2.4, the object m begins its movement on Ox. Thus we
have θ0 = 0, and from this we get the formula of r0 in the statement.
(2) Regarding the initial speed now, the formula of θ̇0 follows from:
√
λ Kc
θ̇ = 2 = 2
r r
Also, in what concerns the radial speed, the formula of ṙ0 follows from:
c(ε sin θ − δ cos θ)θ̇
ṙ =
(1 + ε cos θ + δ sin θ)2
√
c(ε sin θ − δ cos θ) Kc
= 2 2
· 2
c /r r
√
K(ε sin θ − δ cos θ)
= √
c
√
(3) Regarding now the initial acceleration, by using θ̇ = Kc/r2 we find:
√
√ 2rṙ 4 Kc · ṙ
θ̈ = −2 Kc · 3 = −
r r2
In particular at t = 0 we obtain the formula in the statement, namely:
√ √ √
4 Kc · ṙ0 4 Kc δ K 4δK
θ̈0 = − 2
= 2
· √ = 2
R R c R
√
(4) Also regarding acceleration, with λ = Kc our main motion formula reads:
 
Kc 1 1
r̈ = 2 −
r r c
In particular at t = 0 we obtain the formula in the statement, namely:
 
Kc 1 1 Kc ε εK
r̈0 = 2 − = 2 · = 2
R R c R c R
(5) Finally, the last assertion is clear, and since the formulae look better anyway in
polar coordinates than in affine coordinates, we will not get into details here. □
With the above formulae in hand, which are a precious complement to Theorem 2.4,
we can do some reverse engineering at the level of parameters, and work out how various
inital speeds and accelerations lead to various types of conics. There are many things
that can be said here, of course, and we will discuss some of them later.
 2C. FREE FALLS, ENERGY 37

2c. Free falls, energy

As a complement to the above results, solving the generic problem, in 2D, let us discuss
now the 1D case. We will be interested here in free falls, represented vertically:
◦m
↓

•M
The result here, which is something quite familiar, and that we can establish right
from the Newton principles, with just a pinch of basic calculus, is as follows:
Theorem 2.6. In the context of a free fall from distance x0 = R >> 0, with initial
velocity v0 = 0, the equation of the trajectory is
gt2
x≃R−
2
2
with the constant being g = GM/R , called gravity of M , at distance R from it.
Proof. As before, the equation of motion of our object m is as follows:
Kx
ẍ = −
||x||3
In one dimension now, things get simpler, and the equation of motion reads:
K
ẍ = − 2
x
Since we assumed R >> 0, we must look for a solution of type x ≃ R + ct2 , with the
lack of the t term coming from v0 = 0. But with x ≃ R + ct2 , our equation reads:
K
2c ≃ − 2
R
2
Now by multiplying by t /2, and adding R, we obtain as solution:
Kt2
x≃R−
2R2
Thus, we have indeed x ≃ R − gt2 /2, with g being the following number:
K GM
g= 2 = 2
R R
We are therefore led to the conclusion in the statement. □
As an illustration for the above 1D computation, let us do a numeric check. The
gravitational constant, the mass of the Earth, and the average radius of the Earth are as
follows, expressed as usual in SI units, namely meters, kilograms and seconds:
G = 6.674 × 10−11 , M = 5.972 × 1024 , R = 6.371 × 106
38 2. KEPLER AND NEWTON

We obtain the following value for the number g computed above:

6.674 × 5.972
g= × 10 = 9.819
6.371 × 6.371
And this is quite decent, compared to the observed value g = 9.806.

As a second result now, related to the above computations, which is more advanced,
lying somewhere between 1D and 2D, let us add an arbitrary initial speed v0 = v to the
above situation, which in addition is allowed to be a vector in R2 , as follows:
◦m
↙v

•M
We obtain in this way the following generalization of Theorem 2.6:
Theorem 2.7. In the context of a free fall from distance x0 = R >> 0, with initial
plane velocity vector v0 = v, the equation of the trajectory is
gt2
x ≃ R + vt −
2
2
where g = GM/R as usual, and with the quantities R, g in the above being regarded now
as vectors, pointing upwards. The approximate trajectory is a parabola.
Proof. We have several assertions here, the idea being as follows:
(1) Let us first discuss the simpler case where we are still in 1D, as in Theorem 2.6,
but with an initial velocity v0 = v added. In order to find the equation of motion, we can
just redo the computations from the proof of Theorem 2.6, with now looking for a general
solution of type x ≃ R + vt + ct2 , and we get, as stated above:
gt2
x ≃ R + vt −
2
Alternatively, we can simply argue that, by linearity, what we have to do is to take
the solution x ≃ R − gt2 /2 found in Theorem 2.6, and add an extra vt term to it.
(2) In the general 2D case now, where the initial velocity v0 = v is a vector in R2 , the
same arguments apply, either by redoing the computations from the proof of Theorem
2.6, or simply by arguing that by linearity we can just take the solution x ≃ R − gt2 /2
found there, and add an extra vt term to it. Thus, we have our solution.
(3) Let us study now the solution that we found. In standard (x, y) coordinates, with
v = (p, q), and with R, g being now back scalars, our solution looks as follows:
gt2
x = pt , y ≃ R + qt −
2
 2C. FREE FALLS, ENERGY 39

From the first equation we get t = x/p, and by substituting into the second:
qx gx2
y ≃R+ − 2
p 2p
We recognize here the approximate equation of a parabola, and we are done. □
Let us discuss now an important topic, namely the conservation of energy, in the
gravitational context. This is something that you are very familiar with, because when
you throw a rock up in the sky, the more energy you put into your throw, the higher the
rock will get. And the higher the rock will get, the faster it will come back on your head.
In other words, the kinetic energy T gets converted into height, and vice versa, so if we
find a way of calling that height “potential” energy V , then our conservation principle
will simply state that the total energy E = T + V is constant.

The simplest situation is that of a free fall with initial velocity v0 = 0, and our
conservation principle here is as follows:
Proposition 2.8. In the context of a free fall from distance x0 = R >> 0, with initial
velocity v0 = 0, if we define the potential energy to be
V = mgx
then the total energy E = T + V , with T = mv 2 /2 as usual, is constant, E ≃ mgR.
Proof. We know that the equation of motion is as follows, with g = GM/R2 :
gt2
x≃R−
2
The kinetic energy, from now on to be denoted T , is then given by:
mv 2 mg 2 t2
T ≃ =
2 2
Thus with V = mgx as in the statement, and then with E = T + V , we have:
E = T + V ≃ mgR
But this is a constant, and so we have our conservation principle, as desired. □
At this point, we are rather inside math, and many questions arise. We know that
E ≃ mgR, but by some kind of miracle, do we actually have E = mgR? Also, what is the
meaning of V ? What about the meaning of E? What about adding a suitable constant
to V , and so to E too, will that make these quantities easier to understand?

These questions will be answered in due time. For the moment, let us keep doing the
math, that is always relaxing, and one solid lead. As a next result, we have:
40 2. KEPLER AND NEWTON

Theorem 2.9. In the context of a free fall from distance x0 = R >> 0, with initial
velocity vector v0 ∈ R2 , if we define the potential energy to be
V = m < g, x >
with g = GM/R2 being regarded as usual as a vector pointing upwards, then
E =T +V
with T = m||v||2 /2 as usual, is constant, E ≃ T0 + mgR, with g now back scalar.
Proof. We can do this in two steps, first by adding an extra parameter to the com-
putation in Proposition 2.8, and then by adding another extra parameter:
(1) Let us first examine the 1D case, where v0 = s is a vector aligned to x, and so a
number. Here the equation of motion is as follows, with g = GM/R2 as usual:
gt2
x ≃ R + st −
2
The speed being v ≃ s − gt, with V = mgx and E = T + V as above, we have:
E = T +V
m(s − gt)2 gt2
 
≃ + mg R + st −
2 2
2
ms
= + mgR
2
= T0 + mgR
(2) In the general case now, with v0 = s, the equation of motion is as before, with
R, g being now vectors pointing upwards, and if we write s = (a, b), then we have:
m||s − gt||2
T ≃
2
m((a − gt)2 + b2 )
=
2
m(a + b2 )
2
mg 2 t2
= − magt +
2  2
gt2

= T0 − mg at −
2
With g vector pointing upwards, the last quantity is m < g, x − R >, so if we add
V = m < g, x >, we obtain T0 + mgR, with g, R being back scalars, as desired. □
With the above done, let us get back to the real thing, 3D gravity. We are interested in
the general 2-body problem, where M is fixed at 0, and m moves under the gravitational
 2C. FREE FALLS, ENERGY 41

force of M . The above computations, coming from our “kinetic energy gets converted
into height, and vice versa” principle, suggest defining the potential energy as:

V ∼ ||x||

However, this is wrong, because in our formula V = mgx the quantity g = GM/R2
depends on the average height, which is the parameter R, no longer assumed to satisfy
R >> 0. In view of this, the correct formula for the potential energy should be:
1
V ∼
||x||
In order now to find the constant, it is enough to rewrite V = mgx by getting rid of
the parameter g = GM/R2 . We obtain in this way, with K = GM as usual:
mGM x mGM Km
V = mgx = 2
≃ =
R ||x|| ||x||
Thus, we have our formula for V , and the question now is if E = T +V is constant. And
the answer here is unfortunately no, due to some bizarre reasons, with rather E = T − V
appearing to be constant, or at least that’s what computations tend to suggest.

So, let us simply change the sign of V , and see what we get. We are led in this way
to the following remarkable result, which not only says that E is approximately constant,
as in our previous computations, but is actually a plain constant:

Theorem 2.10. In the context of the 2-body problem, with M fixed at 0 and with m
moving, if we define the kinetic and potential energy of m to be
m||v||2 Km
T = , V =−
2 ||x||
with K = GM as usual, then the total energy E = T + V is constant.

Proof. The idea will be that of proving Ė = 0. We can do this as follows:

(1) In what regards the derivative of T , the computation here is something very simple,
coming straight from the formula ||v||2 =< v, v >, as follows:

m(< v, v̇ > + < v̇, v >)

Ṫ =
2
= m < v, v̇ >
= m < v, a >
42 2. KEPLER AND NEWTON

(2) In order to compute now the derivative of V , let us first compute the derivative of
the function f (x) = 1/||x||. Again by using ||x||2 =< x, x >, we obtain:
1 < x, ẋ > + < ẋ, x >
f˙ = − ·
2 < x, x >3/2
< x, ẋ >
= −
< x, x >3/2
< x, v >
= −
||x||3
(3) Thus, getting now to the potential energy V , we have the following formula:
Km < x, v >
V̇ =
||x||3
In order to further process this, remember the equation of motion of m, namely:
Kx
a=−
||x||3
We will of course jump on this, as to get rid of ||x||3 , and we finally obtain:
V̇ = −m < a, v >
(4) We are ready now to prove our result. Indeed, we have:
Ė = Ṫ + V̇ = m < v, a > −m < a, v >= 0
Now since the derivative vanishes, E is constant, as claimed. □
Nice all this, but we still have to understand the relation with Proposition 2.8 and
Theorem 2.9, with that sign of V mysteriously switching. And we have here the following
result, upgrading Proposition 2.8 and Theorem 2.9, and clarifying the whole thing:
Theorem 2.11. In the context of a free fall from distance x0 = R >> 0, with initial
velocity v0 = 0, if we define the kinetic and potential energy of m to be
mv 2 Km
T = , V =−
2 x
with K = GM as usual, then the total energy E = T + V is constant. Moreover,
V ≃ mgx − 2mgR
with g = GM/R , and so E = T + mgx is appoximately constant, E ′ ≃ mgR. The same
2 ′

happens for a free fall from x0 = R >> 0, with initial velocity vector v0 ∈ R2 .
Proof. The first assertion is something that we know, coming from Theorem 2.10.
In order to clarify now the relation with Proposition 2.8, we first have:
Km GM m mgR2
V =− =− =−
x x x
 2D. ROTATING OBJECTS 43

Now by writing x = R(1 − ε), we obtain the estimate in the statement, namely:
mgR
V = −
1−ε
≃ −mgR(1 + ε)
= mgR[(1 − ε) − 2]
= mgx − 2mgR
Thus with V ′ = mgx we have V ≃ V ′ − 2mgR, and so E ′ = T + V ′ satisfies:
E′ ≃ E + 2mgR
= E0 + 2mgR
= V0 + 2mgR
= mgR
Finally, the last assertion, which is a bit more general, follows in the same way. □
Observe how impressively Theorem 2.11 clarifies our previous computations, and gen-
erally speaking, answers all our previous questions, raised above. We will be back to this
in the next chapter, with more explanations, following Lagrange and Hamilton.

2d. Rotating objects

Let us discuss now angular momentum, and rotating objects. For this purpose, we
will use vector products in R3 , whose definition, which is quite tricky, is as follows:
Definition 2.12. The vector product of two vectors in R3 is given by
x × y = ||x|| · ||y|| · sin θ · n
where n ∈ R3 with n ⊥ x, y and ||n|| = 1 is constructed using the right-hand rule:
↑x×y
←x
↙y
Alternatively, in usual vertical linear algebra notation for all vectors,
     
x1 y1 x2 y3 − x3 y2
x2  × y2  = x3 y1 − x1 y3 
x3 y3 x1 y2 − x2 y1
the rule being that of computing 2 × 2 determinants, and adding a middle sign.
Obviously, this definition is something quite subtle, and also something very annoying,
because you always need this, and always forget the formula.
44 2. KEPLER AND NEWTON

Here are my personal methods, for remembering all this, because as calculus teacher,
I have to be good at it. With the first definition, what I always remember is that:

||x × y|| ∼ ||x||, ||y||

x×x=0

e1 × e2 = e3
So, here’s how it works. We’re looking for a vector x × y whose length is proportional
to those of x, y. But now the second formula tells us that the angle θ between x, y must be
involved via 0 → 0, and so the factor can only be sin θ. And with this we’re almost there,
it’s just a matter of choosing the orientation, and this comes either from the right-hand
rule (or perhaps left-hand rule, do I remember right?) or from e1 × e2 = e3 .

As with the second definition, that I like the most, what I remember here is simply:

1 x1 y 1
1 x2 y2 =?
1 x3 y 3

Indeed, when trying to compute this determinant, by developing over the first column,
what you get as coefficients are the entries of x × y. And with the good middle sign.

It is also good to know that x × y exists only in 3 dimensions, with our only tool in
N ̸= 3 dimensions being the usual < x, y >. This is actually quite interesting for us, in
relation with our conservation principles for gravity, but more on this later.

We are now ready to state our first conservation principle, the momentum one. This
is something quite technical, at least at the first glance, as follows:

Theorem 2.13. In the gravitational 2-body problem, the angular momentum

J =x×p

with p = mv being the usual momentum, is conserved.

Proof. There are several things to be said here, the idea being as follows:

(1) The usual momentum, p = mv, is not conserved, for instance because in the context
of the circular motion, the moment gets turned around. But this suggests precisely that,
in order to fix the lack of conservation, we have to consider J = x × p, as above.
 2D. ROTATING OBJECTS 45

(2) Regarding now the proof, consider indeed a particle m moving under the gravita-
tional force of a particle M , assumed, as usual, to be fixed at 0. We have:
J˙ = ẋ × p + x × ṗ
= v × mv + x × ma
= m(v × v + x × a)
= m(0 + 0)
= 0
Now since the derivative of J vanishes, this quantity is constant, as stated. □
The above principle has several notable consequences, as follows:
Theorem 2.14. In the context of a 2-body problem, the following happen:
(1) The fact that the direction of J is fixed tells us that the trajectory of one body
with respect to the other lies in a plane.
(2) The fact that the magnitude of J is fixed tells us that the Kepler 2 law holds,
namely that we have same areas sweeped by Ox over the same times.
Proof. This follows indeed from Theorem 2.13, as follows:
(1) We have by definition J = m(x × v), and since a vector product is orthogonal on
both the vectors it comes from, we deduce from this that we have:
J ⊥ x, v
But this can be written as follows, with J ⊥ standing for the plane orthogonal to J:
x, v ∈ J ⊥
Now since J is fixed by Theorem 2.13, we conclude that both x, v, and in particular
the position x, and so the whole trajectory, lie in this fixed plane J ⊥ , as claimed.
(2) Conversely now, forget about Theorem 2.13, and assume that the trajectory lies
in a certain plane E. Thus x ∈ E, and by differentiating we have v ∈ E too, and so
x, v ∈ E. Thus E = J ⊥ , and so J = E ⊥ , so the direction of J is fixed, as claimed.
(3) Regarding now the last assertion, we know that Kepler 2 is more or less equivalent
to θ̇ = λ/r2 . However, the derivation of θ̇ = λ/r2 was something tricky, and what we
want to prove now is that this appears as a simple consequence of ||J|| = constant.
(4) In order to to so, let us compute J, according to its definition J = x × p, but in
polar coordinates, which will change everything. Since p = mẋ, we have:
   
cos θ ṙ cos θ − r sin θ · θ̇
J = r  sin θ  × m ṙ sin θ + r cos θ · θ̇ 
0 0
46 2. KEPLER AND NEWTON

Now recall from the definition of the vector product that we have:
     
a c 0
 b  × d =  0 
0 0 ad − bc

Thus J is a vector of the above form, with its last component being:

cos θ ṙ cos θ − r sin θ · θ̇

Jz = rm
sin θ ṙ sin θ + r cos θ · θ̇
= rm · r(cos2 θ + sin2 θ)θ̇
= r2 m · θ̇

(5) Now with the above formula in hand, our claim is that the magnitude ||J|| is
constant precisely when θ̇ = λ/r2 , for some λ ∈ R. Indeed, up to the obvious fact that
the orientation of J is a binary parameter, who cannot just switch like that, let us just
agree on this, knowing J is the same as knowing Jz , and is also the same as knowing ||J||.
Thus, our claim is proved, and this leads to the conclusion in the statement. □

Finally, let us discuss the rotating frames. We first have here:

Theorem 2.15. Assume that a 3D body rotates along an axis, with angular speed w.
For a fixed point of the body, with position vector x, the usual 3D speed is

v =ω×x

where ω = wn, with n unit vector pointing North. When the point moves on the body

V = ẋ + ω × x

is its speed computed by an inertial observer O on the rotation axis.

Proof. We have two assertions here, both requiring some 3D thinking, as follows:

(1) Assuming that the point is fixed, the magnitude of ω × x is the good one, due to
the following computation, with r being the distance from the point to the axis:

||ω × x|| = w||x|| sin t = wr = ||v||

As for the orientation of ω × x, this is the good one as well, because the North pole
rule used above amounts in applying the right-hand rule for finding n, and so ω, and this
right-hand rule was precisely the one used in defining the vector products ×.
 2D. ROTATING OBJECTS 47

(2) Next, when the point moves on the body, the inertial observer O can compute its
speed by using a frame (u1 , u2 , u3 ) which rotates with the body, as follows:
V = ẋ1 u1 + ẋ2 u2 + ẋ3 u3 + x1 u̇1 + x2 u̇2 + x3 u̇3
= ẋ + (x1 · ω × u1 + x2 · ω × u2 + x3 · ω × u3 )
= ẋ + w × (x1 u1 + x2 u2 + x3 u3 )
= ẋ + ω × x
Thus, we are led to the conclusions in the statement. □
In what regards now the acceleration, the result, which is famous, is as follows:
Theorem 2.16. Assuming as before that a 3D body rotates along an axis, the accel-
eration of a moving point on the body, computed by O as before, is given by
A = a + 2ω × v + ω × (ω × x)
with ω = wn being as before. In this formula the second term is called Coriolis accelera-
tion, and the third term is called centripetal acceleration.
Proof. This comes by using twice the formulae in Theorem 2.15, as follows:
A = V̇ + ω × V
= (ẍ + ω̇ × x + ω × ẋ) + (ω × ẋ + ω × (ω × x))
= ẍ + ω × ẋ + ω × ẋ + ω × (ω × x)
= a + 2ω × v + ω × (ω × x)
Thus, we are led to the conclusion in the statement. □
The truly famous result is actually the one regarding forces, obtained by multiplying
everything by a mass m, and writing things the other way around, as follows:
ma = mA − 2mω × v − mω × (ω × x)
Here the second term is called Coriolis force, and the third term is called centrifugal
force. These forces are both called apparent, or fictious, because they do not exist in the
inertial frame, but they exist however in the non-inertial frame of reference, as explained
above. And with of course the terms centrifugal and centripetal not to be messed up.

In fact, even more famous is the terrestrial application of all this, as follows:
Theorem 2.17. The acceleration of an object m subject to a force F is given by
ma = F − mg − 2mω × v − mω × (ω × x)
with g pointing upwards, and with the last terms being the Coriolis and centrifugal forces.
Proof. This follows indeed from the above discussion, by assuming that the acceler-
ation A there comes from the combined effect of a force F , and of the usual g. □
48 2. KEPLER AND NEWTON

We refer to any standard undergraduate mechanics book, such as Feynman [36], Kibble
[61] or Taylor [88] for more on the above, including various numerics on what happens
here on Earth, the Foucault pendulum, history of all this, and many other things. Let
us just mention here, as a basic illustration for all this, that a rock dropped from 100m
deviates about 1cm from its intended target, due to the formula in Theorem 2.17.
2e. Exercises
Exercises:
Exercise 2.18.
Exercise 2.19.
Exercise 2.20.
Exercise 2.21.
Exercise 2.22.
Exercise 2.23.
Exercise 2.24.
Exercise 2.25.
Bonus exercise.
 CHAPTER 3

Lagrange, Hamilton

3a. Conservative forces

We have seen that the momentum and energy conservation principles from the very
basic, acceleration-free mechanics, have gravitational analogues, namely the conservation
of the angular momentum J, and the conservation of the total energy E = T + V . Things
have been quite tricky, both with J and E, but somehow trickier with E, and our purpose
in what follows will be to develop some further general theory, in order to understand
what the potential energy V , and then what the total energy E = T + V , truly are.

We will need some multivariable calculus. Let us start with something that you surely
know well, but which is perhaps expressed here in a more conceptual way, namely:
Definition 3.1. The derivative of a function f : RN → RM at a point x ∈ RN is the
linear map f ′ (x) : RN → RM making the approximation formula
f (x + t) ≃ f (x) + f ′ (x)t
work, around that given point x ∈ RN .
In the case N = M = 1 the derivative is easy to compute, because the linear map
′
f (x) : R → R must be the multiplication by a constant, and that constant, that we will
agree to denote by f ′ (x) as well, must be given by the following formula:
f (x + t) − f (x)
f ′ (x) = lim
t→0 t
′
In general, by linear algebra, the derivative f (x) : RN → RM is no longer a number,
but rather a rectangular matrix, that we will agree to denote by f ′ (x) as well:
f ′ (x) ∈ MM ×N (R)
And the question is now, with this picture of the derivative in mind, how to compute
this matrix. And here, you certainly know the answer, which is as follows:
Theorem 3.2. The derivative of f : RN → RM at a point x ∈ RN is given by:
 
′ dfi
f (x) =
dxj ij
That is, the derivative is the matrix formed by the partial derivatives.
49
50 3. LAGRANGE, HAMILTON

Proof. This is something which certainly holds at N = M = 1, where the equality

in question reads f ′ (x) = f ′ (x). More generally, the result holds at N = 1, because in
this case, with the notation f = (f1 , . . . , fM ), the following column vector, or rather the
linear map associated to it, makes work the approximation formula in Definition 3.1:
 ′
f1 (x)


f ′ (x) =  ... 
′
fM (x)
Thus, done with N = 1, and this study also teaches us that, in order to deal with
the general case, we can restrict the attention to the case M = 1, because afterwards
things will extend without problems to the case where M ∈ N is arbitrary. So, consider a
function f : RN → R. We must prove that the following row vector, or rather the linear
map associated to it, makes work the approximation formula in Definition 3.1:
 
df df
f ′ (x) = dx 1
. . . dxN

But this is something standard, say by proceeding by recurrence on N ∈ N, starting

from the case N = 1 which is clear, as mentioned above, and so we get the result. □

Many things can be said about derivatives, and of particular interest is:
Theorem 3.3. We have the chain derivative formula
(f ◦ g)′ (x) = f ′ (g(x)) · g ′ (x)
as an equality of matrices.
Proof. This is something standard in one variable, and in several variables the proof
is similar. To be more precise, consider a composition of functions, as follows:
f : RN → RM , g : RK → RN , f ◦ g : RK → RM
According to Definition 3.1, the derivatives of these functions are certain linear maps,
corresponding to certain rectangular matrices, as follows:
f ′ (g(x)) ∈ MM ×N (R) , g ′ (x) ∈ MN ×K (R) (f ◦ g)′ (x) ∈ MM ×K (R)
Thus, our formula makes sense indeed. As for proof, this comes from:
(f ◦ g)(x + t) = f (g(x + t))
≃ f (g(x) + g ′ (x)t)
≃ f (g(x)) + f ′ (g(x))g ′ (x)t
Thus, we are led to the conclusion in the statement. □
 3A. CONSERVATIVE FORCES 51

Observe that the above proof does not use at all Theorem 3.2. In fact, Theorem 3.3
is more of a triviality, based on our notion of derivative from Definition 3.1.

Now back to physics, and to our energy results from chapter 2, everything there was
based on two main formulae, both involving V , which are as follows, with the first one
being our definition of V , and the second one coming from the equation of motion:
Km
V =− , V̇ = −m < a, v >
||x||
The point now, which is something tricky, is that in order to fully understand the
meaning of these formulae, and of V itself, we need a third formula, as follows:
Proposition 3.4. In the context of the 2-body problem, the force applied to m is
F = −∇V
where V = −Km/||x|| is as usual the potential energy of m.
Proof. According to the Newton principles, and with K = GM as usual, the force
applied to m by the object M positioned at 0 is given by:
x
F = −||F || ·
||x||
mM x
= −G · 2
·
||x|| ||x||
Kmx
= −
||x||3
Now let us compute the gradient ∇V , pronounced “del V ” or “nabla V ” which is by
definition the vector ∇V ∈ R3 formed by the 3 spatial derivatives of V . We have:
p
dV dKm/ x21 + x22 + x23
= −
dxi dxi
d(x21 + x22 + x23 )−1/2
= −Km ·
dxi
 
1 2xi
= −Km · − · 2
2 (x1 + x22 + x23 )3/2
Kmxi
=
||x||3
Thus we have ∇V = Kmx/||x||3 , which by the above equals −F , as desired. □
The above result is interesting in connection with energy conservation problems, due
to the following general fact, which is something going beyond gravitation:
52 3. LAGRANGE, HAMILTON

Theorem 3.5. Assuming that an object of mass m moves under the influence of a
force F which is conservative, in the sense that
F = −∇V
for a certain function V , then if we call this function V potential energy of m, and set
E =T +V
with T = m||v||2 /2 being as usual the kinetic energy of m, then E is constant.
Proof. This statement looks a bit head-scratching, probably reminding some terribly
abstract things that you had to endure, coming from your math professors. But please
stay with us, we’re doing physics here, not mathematics. We have:
m(< v̇, v > + < v, v̇ >)
Ṫ =
2
= m < v̇, v >
= m < a, v >
= < F, v >
= − < ∇V, v >
X
= − (∇V )i vi
i
X dV dxi
= − ·
i
dxi dt
dV
= −
dt
= −V̇
Thus we have Ė = Ṫ + V̇ = 0, and so the energy E is constant, as claimed. □

Here we have used the chain rule from Theorem 3.3, suitably adapted to the present
context. To be more precise, in our context the function V = V (x(t)) appears by com-
posing the function V : R3 → R, having as derivative the transpose of the gradient
vector (∇V )t ∈ M1×3 (R), with the function x : R → R3 , having as derivative the vector
ẋ ∈ M3×1 (R). Thus, the chain rule gives in this case the following number:

V̇ = (∇V )t ẋ =< ∇V, ẋ >

All this might seem a bit complicated, but several variables means linear algebra, so
it is. In practice, all this is best remembered as in the
P proof of Theorem 3.5, with that
dxi factors miraculously simplifying, and taking the i symbol with them.
 3A. CONSERVATIVE FORCES 53

Back to physics now, what we have in Theorem 3.5 is quite conceptual. It is possible
to say more about conservative forces, such as gravitation, but before that, let us restate
and reprove our main energy result from chapter 2, more conceptually, as follows:
Theorem 3.6. The gravitation force F is conservative, F = −∇V , coming from
Km
V =−
||x||
and therefore the total energy E = T + V , with T = m||v||2 /2 as usual, is constant.
Proof. Here the first assertion comes from Proposition 3.4, and the second assertion,
that we already know from chapter 2, comes now from Theorem 3.5. □
In order to further clarify our concept of “conservative force”, which generalizes in an
efficient and elegant way the usual gravity, the best is to start by talking about work.
This is something familiar, I hope, but in order to do our math, we will need:
Proposition 3.7. Given a path γ ⊂ R3 , we can talk about integrals of type
Z
I = f (x)dx1 + g(x)dx2 + h(x)dx3
γ

with f, g, h : R3 → R, which are independent on the chosen parametrization of the path.

Proof. This is something quite straightforward, the idea being as follows:
(1) Regarding the statement itself, assume indeed that we have a path in R3 , which
can be best thought of as corresponding to a function as follows:
γ : [a, b] → R3
Observe that this function γ is not exactly the path itself, for instance because the
following functions produce the same path, parametrized differently:
δ : [0, b − a] → R3 , δ(t) = γ(t + a)
ε : [0, 1] → R3 , ε(t) = δ((b − a)t)
φ : [0, 1] → R3 , φ(t) = ε(t2 )
ψ : [0, 1] → R3 , ψ(t) = ε(1 − t)
..
.
Our claim, however, is that we can talk about integrals as follows, with f, g, h : R3 →
R, which are independent on the chosen parametrization of our path:
Z
I = f (x)dx1 + g(x)dx2 + h(x)dx3
γ
54 3. LAGRANGE, HAMILTON

(2) In order to prove this, let us choose a parametrization γ : [a, b] → R3 as above.

This parametrization has as components three functions γ1 , γ2 , γ3 , given by:

γ = (γ1 , γ2 , γ3 ) : [a, b] → R3

In order to construct the integral I, it is quite clear, by suitably cutting our path
into pieces, that we can restrict the attention to the case where all three components
γ1 , γ2 , γ3 : [a, b] → R are increasing, or decreasing. Thus, we can assume that these three
components are as follows, increasing or decreasing, and bijective on their images:

γi : [a, b] → [ai , bi ]

(3) Moreover, by using the obvious symmetry between the coordinates x1 , x2 , x3 , in

order to construct I, we just need to construct integrals of the following type:
Z
I1 = f (x)dx1
γ

(4) So, let us construct this latter integral I1 , under the assumptions in (2). The
simplest case is when the first path, γ1 : [a, b] → [a1 , b1 ], is the identity:

γ1 : [a, b] → [a, b] , γ1 (x) = x

In other words, the simplest case is when our path is of the following form, with
γ2 , γ3 : [a, b] → R being certain functions, that should be increasing or decreasing, as per
our conventions (2) above, but in what follows we will not need this assumption:

γ(x1 ) = (x1 , γ2 (x1 ), γ3 (x1 ))

But now, with this convention made, we can define our contour integral, or rather its
first component, as explained above, as a usual one-variable integral, as follows:
Z b
I1 = f (x1 , γ2 (x1 ), γ3 (x1 ))dx1
a

(5) With this understood, let us examine now the general case, where the first path,
γ1 : [a, b] → [a1 , b1 ], is arbitrary, increasing or decreasing, and bijective on its image. In
this case we can reparametrize our curve, as to have it as in (4) above, as follows:

γ̃ = (id, γ2 γ1−1 , γ3 γ1−1 ) : [a1 , b1 ] → R3

 3A. CONSERVATIVE FORCES 55
R
Now since we want our integral I1 = γ f (x)dx1 to be independent of the parametriza-
tion, we are led to the following formula for it, coming from the formula in (4):
Z
I1 = f (x)dx1
γ̃
Z b1
= f (x1 , γ2 γ1−1 (x1 ), γ3 γ1−1 (x1 ))dx1
a1
Z b
= f (γ1 (y1 ), γ2 (y1 ), γ3 (y1 ))γ1′ (y1 )dy1
a

Here we have used at the end the change of variable formula, with x1 = γ1 (y1 ).
(6) Thus, job done, we have our definition for the contour integrals, with the formula
being as follows, obtained by using (5) for all three coordinates x1 , x2 , x3 :
Z b
I = f (γ1 (y1 ), γ2 (y1 ), γ3 (y1 ))γ1′ (y1 )dy1
a
Z b
+ g(γ1 (y2 ), γ2 (y2 ), γ3 (y2 ))γ2′ (y2 )dy2
a
Z b
+ h(γ1 (y3 ), γ2 (y3 ), γ3 (y3 ))γ3′ (y3 )dy3
a

And with this, we are led to the conclusion in the statement. □

With the above understood, here is now the exact, physical definition of work:
Definition 3.8. The work done by a force F = F (x) for moving a particle from point
p ∈ R3 to point q ∈ R3 via a given path γ : p → q is the following quantity:
Z
W (γ) = < F (x), dx >
γ

We say that F is conservative if this work quantity W (γ) does not depend on the chosen
path γ : p → q, and in this case we denote this quantity by W (p, q).
We will see in a moment that this definition is compatible with our previous definition
for the conservative forces. As a first comment now, assume that we have two paths
γ : p → q and δ : p → q. We can then consider the path ◦ : p → p obtained by going
along γ : p → q, and then along δ reversed, δ −1 : q → p, and we have:
W (◦) = W (γ) − W (δ)
Thus F is conservative precisely when, for any loop ◦ : p → p, we have:
W (◦) = 0
56 3. LAGRANGE, HAMILTON

Intuitively, this means that F is some sort of “clean”, ideal force, with no dirty things
like friction involved. As we will soon see, gravity is such a clean force, with a simple
example coming from throwing a rock up in the sky. That rock will travel on a loop
p → q → p, and will come back here to p unchanged, save for the fact that its speed
vector is reversed. Thus, and assuming now that work has something to do with energy,
which is intuitive, there has been no overall work of gravity on this loop, W (◦) = 0.

An even better example, avoiding any reference to energy, is the movement of the
Earth around the Sun. Every year that passes the Earth makes a loop, and wih the Sun
obviously not even noticing that, so the yearly work done by the Sun is W (◦) = 0.

As a counterexample now, friction is not conservative. I would definitely prefer to

make loops with my lawn mower in my garden, and say to myself, for motivation, that
I’m doing −W (◦) = 0, rather than taking that thing up to the North Pole, and back.

As a first result now, regarding the conservative forces, we have:

Theorem 3.9. The work done by a conservative force F on a mass m object is
W (p, q) = T (q) − T (p)
with T = m||v||2 /2 standing as usual for the kinetic energy of the object.
Proof. Assuming that F is conservative, and acts via the usual formula F = ma on
our object of mass m, we have the following computation, as desired:
Z q
W (p, q) = < F (x), dx >
p
Z q
= m < a(x), dx >
p
Z q 
dv(x)
= m , v(x)dt
p dt
m q d < v(x), v(x) >
Z
= dt
2 p dt
m q d||v(x)||2
Z
= dt
2 p dt
m
||v(q)||2 − ||v(p)||2


=
2
= T (q) − T (p)
Here we have used in the middle the fact that the time derivative of a scalar product
of functions < v, w > consists of two terms, which are equal when v = w. □
Next, we have the following result, which uses some more advanced mathematics:
 3A. CONSERVATIVE FORCES 57

Theorem 3.10. A force F is conservative precisely when it is of the form

F = −∇V
for a certain function V , and in this case the work done by it is given by:
W (p, q) = V (p) − V (q)
Also, the gravitation force is conservative, coming from V = −Km/||x||.
Proof. This is something quite tricky, the idea being as follows:
(1) In one sense, assume that F is conservative. Since the work W (p, q) = W (γ) is
independent of the chosen path γ : p → q, we can find a function V such that:
W (p, q) = V (p) − V (q)
Observe that this function V is well-defined up to an additive constant. Now with
this formula in hand, we further obtain, as desired:
W (p, q) = V (p) − V (q) =⇒ < F, dx >= −dV
dV
=⇒ < F, xi >= −
dxi
=⇒ F = −∇V

(2) In the other sense now, assuming F = −∇V , we have the following computation,
valid for any loop ◦ : p → p, which shows that F is indeed conservative:
Z
W (◦) = − ∇V = 0
◦
More generally, regarding the work done by such a force F = −∇V , along a path
γ : p → q, which is independent on this path γ, this is given by:
Z q
W (p, q) = − ∇V = V (p) − V (q)
p

(3) Finally, the last assertion, regarding the gravitation, this is something that we
know from Theorem 3.5, coming via a quick gradient computation. □
We can put now everything together, and we have the following result:
Theorem 3.11. Given a conservative force F , appearing as follows, with V being
uniquely determined up to an additive constant,
F = −∇V
the movements of a particle under F preserve the total energy, given by
E =T +V
with T = m||v||2 /2 being the kinetic energy, and with V being called potential energy.
58 3. LAGRANGE, HAMILTON

Proof. This is something that we already know from before, established there by
using a computation using the chain rule for derivatives, the idea being as follows:
Ṫ = < F, v >
= − < ∇V, v >
= −V̇
However, we can now provide an alternative proof for this fact, based on the theory
developed above, and more specifically on Theorems 3.9 and 3.10, which give:
W (p, q) = T (q) − T (p)

W (p, q) = V (p) − V (q)

Indeed, we obtain from these equalities the following formula:
T (p) + V (p) = T (q) + V (q)
Thus, the total energy E = T + V is conserved, as claimed. □

As a side remark here, observe that Theorem 3.11 completely closes the discussion
about conservation of energy, at least linguistically, our conclusion being:
Conclusion 3.12. Conservative forces conserve energy.
So, this will be the general principle to remember. This being said, don’t leave a fish
out in the sun, it will not be conserved by gravity. Instead, use a refrigerator.

Many other things can be said about conservative forces. We will be back to this.

3b. Lagrange equations

Back now to gravity, the point is that by using the above energy theory, we can
reformulate the whole classical mechanics formalism and equations, in a far more efficient
way. Discussing this, following Lagrange and Hamilton, will be our next task.

Things are quite tricky here, involving some sort of unexpected discovery, of the type
that you can stumble upon when looking at various formulae coming from physics, with
a good mathematical background. So, let us begin with mathematics.

And here, the “good mathematical background” that we will need, and that Lagrange
needed too, of course, for his discovery, is the following theorem, which is due to guess
who, Euler and Lagrange himself:
 3B. LAGRANGE EQUATIONS 59

Theorem 3.13. Given a function f : RN × RN → R, the integral

Z x1
I= f (u, u̇)dx
x0

is stationary, in the sense that it is left unchanged by small variations of u = u(x), which
vanish at the endpoints x0 , x1 , precisely when u = u(x) satisfies the equations
 
df d df
=
dui dx du̇i
called Euler-Lagrange equations.
Proof. Let us just work out the case N = 1, the general case being similar. Consider
a small variation ∆u(x), which vanishes at the endpoints x0 , x1 , as required above:
∆u(x0 ) = ∆u(x1 ) = 0
The corresponding variation of f (u, u̇), at first order, is then given by:
df df
∆f = ∆u + ∆u̇
du du̇
Thus the corresponding variation of the integral in the statement is given by:
Z x1 Z x1
df df
∆I = ∆u dx + ∆u̇ dx
x0 du x0 du̇
Z x1 Z x1
df df d(∆u)
= ∆u dx + · dx
x0 du x0 du̇ dx
Z x1  x1 Z x1  
df df d df
= ∆u dx + ∆u − ∆u dx
x0 du du̇ x0 x0 dx du̇
Z x1 Z x1  
df d df
= ∆u dx − ∆u dx
x0 du x0 dx du̇
Z x1   
df d df
= − ∆u dx
x0 du dx du̇
We conclude that I is stationary precisely when the following equation is satisfied:
 
df d df
=
du dx du̇
But this is the Euler-Lagrange equation in the statement, as desired. □
The point now with the above is that, when looking at the usual motion equations of
mechanics, but written in a somewhat bizarre way, we will get precisely into the Euler-
Lagrange equations. So, remember our struggle from chapter 2 with the gravitational
potential, and more specifically with E = T + V vs E = T − V . We have seen there that
60 3. LAGRANGE, HAMILTON

E = T + V is the good formula, but that the quantity E = T − V looks like something
quite interesting too. So, based on this, let us formulate:
Definition 3.14. In the context of a conservative force F acting, the quantity
L=T −V
is called Lagrangian of the system.
This is something quite tricky, but we will get more familiar with it when working out
explicit examples, and with the reminder of course that we have already played a bit with
T − V , in the simplest case, that of a 1D free fall, in chapter 2.

In relation now with the mathematics in Theorem 3.13, the connection is very simple,
called Hamilton principle, as follows:
Theorem 3.15. The following integral, called action integral, is stationary,
Z t1
I= L dt
t0
and the corresponding Euler-Lagrange equations are precisely the equations of motion.
Proof. According to Definition 3.14, the Lagrangian is given by the following for-
mula, with V being the potential associated to our conservative force, via F = −∇V :
m
L = (ẋ2 + ẏ 2 + ż 2 ) − V (x, y, z)
2
Thus, we are in the general framework of Theorem 3.13, with the function u there being
played by the coordinates x, y, z. Now let us pick one of these coordinates, s = x, y, z,
and compute the derivatives of L with respect to s, ṡ. By using F = −∇V we have:
dL dV
=− = Fs
ds ds
Also since the potential V is time-independent, we have:
dL
= mṡ = mas
dṡ
Now consider the equation of motion, under the influence of the force F :
F = ma
This is a vector equation, with 3 components, and according to the above formulae its
3 components can be written as follows, in terms of the Lagrangian L:
 
dL d dL
=
ds dt dṡ
But theseRare precisely the Euler-Lagrange equations for the stationarity of the action
integral I = L, and we are therefore led to the conclusions in the statement. □
 3D. HAMILTON EQUATIONS 61

The point now with the above result is that it leads right away into another result,
which this time is something fundamental and powerful, as we will soon discover:
Theorem 3.16. The Euler-Lagrange equations for the action integral
Z t1
I= L dt
t0

hold in any system of coordinates (q1 , q2 , q3 ), and are as follows:

 
dL d dL
=
dq dt dq̇
These latter equations are called the Lagrange equations of motion.
Proof. We know from Theorem 3.15 that the action integral is stationary, with
respect to the standard coordinates (x, y, z). But this shows that the action integral is
stationary with respect to any system of coordinates (q1 , q2 , q3 ), and so the corresponding
Euler-Lagrange equations, which are the equations in the statement, hold indeed. □

We will see illustrations for all this, in a moment.

3c. Mechanics, revised

All the above might seem a bit complicated, but we will see examples in what follows.
The idea every time will be the same, namely thinking a bit, than picking up a suitable
system of coordinates (q1 , q2 , q3 ) for our problem, and then instead of doing all sorts of
computations for reformulating the equations of motion in terms of these coordinates
(q1 , q2 , q3 ), simply writing the Lagrange equations, which are there for that.

As a basic illustration here, the Newton solution to the Kepler 2-body problem in
classical mechanics, that we discussed in chapter 2, was using polar coordinates, or rather
cylindrical coordinates with z not mattering, to be fully correct.

But the computations there can be heavily simplified by starting directly with the
Lagrange equations in cylindrical coordinates.

There are of course many other applications, along the same lines.

3d. Hamilton equations

Moving ahead now, let us discuss some further interesting manipulations on the La-
grangian and the Lagrange equations, due to Hamilton. Let us start with:
62 3. LAGRANGE, HAMILTON

Definition 3.17. Given a system of coordinates q1 , . . . , qn , the quantities

dL
pi =
dq̇i
are called generalized momenta. In terms of them, the Lagrange equations read
dL
= ṗi
dqi
analogously to the usual motion equations F = ṗ.
What are these new variables good for? Let us recall from the above that the La-
grangian was by definition a function as follows:
L = L(q1 , . . . , qn , q̇1 , . . . , q̇n )
The point now, which is something quite subtle, and useful for all sorts of purposes,
in classical mechanics, and especially in its versions and generalizations, is that we can
get rid of the derivatives q̇1 , . . . , q̇n , which are variables in the Lagrange formulation of
mechanics, by replacing them by the generalized momenta p1 , . . . , pn constructed above.
And, explaining this will be our goal, in what follows.

In order to do so we need a clever new quantity H, replacing the Lagrangian L, and

we have here the following definition:
Definition 3.18. With q1 , . . . , qn and p1 , . . . , pn being as above, the quantity
X
H(q, p) = pi q̇i − L
i

is called Hamiltonian of the system.

As we will soon see, for many simple systems H is in fact the total energy. Before
that, however, let us explain how H replaces L, as a quantity which encapsulates as well
what is going on, namely the equations of motion. The result here is as follows:
Theorem 3.19. The Hamiltonian H = H(q, p) is subject to the equations
dH dH
= q̇i , = −ṗi
dpi dqi
called Hamilton equations, which are equivalent to the usual equations of motion.
Proof. As a first observation, this statement reminds right away the Lagrange for-
mulation of mechanics, from before, who was claiming the same type of thing. However,
there are some differences. On one hand, the new variables q1 , . . . , qn and p1 , . . . , pn are
certainly a bit more abstract than the old ones q1 , . . . , qn and q̇1 , . . . , q̇n . On the other
 3D. HAMILTON EQUATIONS 63

hand, the new equations look great. Regarding now the proof, everything follows from
the definition of the variables pi and from the Lagrange equations, namely:
dL dL
pi = , = ṗi
dq̇i dqi
(1) By using the definition of the variables pi , we obtain the following formula:
P 
dH d j p j q̇ j − L
=
dpi dpi
X dq̇j X dL dq̇j
= q̇i + pj − ·
j
dpi j
dq̇j dpi
X dq̇j X dq̇j
= q̇i + pj − pj
j
dp i j
dpi
= q̇i

(2) By using the Lagrange equations, we obtain the following formula:

P 
dH d j p j q̇ j − L
=
dqi dqi
dL X dq̇j X dL dq̇j
= − + pj − ·
dqi j
dqi j
dq̇j dqi
X dq̇j X dq̇j
= −ṗi + pj − pj
j
dq i j
dqi
= −ṗi

Thus, we are led to the conclusions in the statement. □

As before with the Lagrange equations, the Hamilton equations can be used in order
to considerably simplify a number of standard classical mechanics computations.

All this is quite instructive to work out, and the comparison between the methods of
Lagrange and Hamilton is something instructive to work out too, in each case.

Finally, there are many other theoretical things that can be said about the Lagrangian
L and the Hamiltonian H, and we will see more examples and general theory, in what
follows. In fact, these will be our main mathematical tools, in what follows.
64 3. LAGRANGE, HAMILTON

3e. Exercises
Exercises:
Exercise 3.20.
Exercise 3.21.
Exercise 3.22.
Exercise 3.23.
Exercise 3.24.
Exercise 3.25.
Exercise 3.26.
Exercise 3.27.
Bonus exercise.
 CHAPTER 4

The N body problem

4a. Center of mass

We have now in our bag all the standard tools of the classical mechanician. There
are all sorts of problems that we can solve with them, and as usual here we refer to our
standard books [2], [36], [39], [61], [64], [88]. In what follows we will mainly focus on
the generalizations of the Kepler 2-body problem, which is more or less the only serious
problem that we have solved, so far. These generalizations follow into several classes:

(1) 2-body problem with round bodies.

(2) 2-body problem with atmospheric drag.

(3) 2-body problem with one of the bodies rotating.

(4) 3-body problem with 2 bodies being fixed.

(5) 3-body problem with 1 big, distant body.

(6) 3-body problem with 1 body being tiny.

(7) Combinations of the above, and more.

All this screams for help, looks like we are in a complete jungle. And the problem is
that (1-7) above are all serious questions, related to problems in the real life. Just throw a
rock in front of you, and you’re instantly into (1,2,3), taken altogether. Throw something
more complicated, like a rocket with a satellite, and you’ll have a taste of (4,5,6) too.

Can mathematics save us? We know so far how to solve the Kepler 2-body problem,
and looking at the above list (1-7), and thinking underlying math, suggests that question
(4) might be the easiest. Indeed, we should normally be able to replace our 2 fixed bodies
with a single one, then solve the problem, and have our first 3-body theorem.

In order to discuss this, let us start with the following general notion:
65
66 4. THE N BODY PROBLEM

Definition 4.1. Associated to a system of bodies M1 , . . . , Mk , located at positions

c1 , . . . , ck ∈ R3 is their center of mass, located at the following position:
P
ci Mi
c = Pi
i Mi
P
A single body of mass i Mi located there, at the center of mass, and with M1 , . . . , Mk
being erased, will be called average of the system formed by M1 , . . . , Mk .
You are certainly a bit familiar with this, and we will work out the math in a moment.
However, before doing that, let us see if this can help in connection with problem (4)
above. When looking in real life for two fixed bodies M1 , M2 the first thing which comes
to mind is a dumbell. Which is however something quite small, so let us formulate:
Definition 4.2. The Devil’s dumbell is a system of two fixed objects
•M1 −−−d −− •M2
which can have arbitrary characteristics M1 > 0, M2 > 0, d > 0.
As already mentioned, the simplest example is a usual dumbell, which however at the
galactic level is subject to Mi ≃ 0 and d ≃ 0. Bigger examples exist in distant galaxies,
with the giants there training with specially designed dumbells, Mi being roughly the mass
of an average planet, and d being accordingly large. As for the fully versatile dumbell in
Definition 4.2, that is a creation of the Devil, for annoying us physicists.

Getting back now to question (4), we would like to understand the motion of an object
of mass m around a dumbell of parameters M1 , M2 , d, as in Definition 4.2, by using the
notion of center of mass of that dumbell, constructed according to Definition 4.1:
◦m
•M1 −−− ⋆c −−•M2
And the preliminary observations here are quite grim, as follows:
Observations 4.3. In the context of a body of mass m moving around a dumbell of
parameters M1 , M2 , d, the following happen:
(1) When m is at distance x >> 0 from the dumbell, the gravitation force acting on
it is F1 + F2 ≃ Fc , so m should travel on some sort of approximate conic.
(2) However, when m is at distance x ≃ 0 from the dumbell, the physics and trajectory
have nothing to do with the center of mass, F1 + F2 ̸≃ Fc .
Here in what regards forces, (1) is something quite obvious and intuitive, and we will
do the math in a moment, with a proof of F1 + F2 ≃ Fc , and a study of the correction
term as well. As for (2), this is something obvious and intuitive too, because if you place
m on the line joining M1 , M2 , bad things will happen, and the only possibility where c
 4A. CENTER OF MASS 67

can be of help is when m was initially placed precisely on c, and with this, point on a line
being at a prescribed location, being an event happening with probability 0.

As for the trajectory claim in (1), that is something sort of intuitive, but not really,
because as you might know from observations with pendulums, balls rolling on various
surfaces, and so on, involving equilibrium and non-equilibrium, there is no guarantee that
the solution of a perturbed problem is a perturbation of the initial solution.

Getting back now to our list (1-7), it looks like the simplest question there, (4), while
suggesting a quick solution using the center of mass, is something quite difficult.

So let us start with the beginning, basic mathematics of the center of mass, as con-
structed in Definition 4.1. To be kept in mind first is:
Proposition 4.4. The center of mass is not a center of gravity, in the sense that the
gravity there is not necessarily 0. For instance the center of mass of a dumbell is
•M1 −
| −{z
−−−−
} ⋆cm −
| −−−{z
−−−−−
} ◦M2
M2 d M1 d
M1 +M2 M1 +M2

while the center of gravity, which is the unique point where the gravity is 0, is:
•M1 −
| −−√−{z } ⋆cg −
−−−−− |−−−−−
{z } ◦M2
√
M1 d M2 d
√ √ √ √
M1 + M2 M1 + M2

The systems of k ≥ 3 bodies might have several centers of gravity, usually uncomputable.
Proof. There are several assertions here, the idea being as follows:
(1) Regarding the dumbell, pictured above with M1 > M2 , the formula for the center
of mass is clear from definitions. Regarding now the center of gravity, the formula there
can be found by doing the math, and it works, because the acceleration there is:
GM1 GM2
a = − √ 2 +  √ 2 = 0
√ M√1d √ M√2d
M1 + M2 M1 + M2

(2) Getting now to systems M1 , . . . , Mk with k ≥ 3, things here are quite complicated.
With ci being the position of Mi , a center of gravity x ∈ R3 must satisfy:
X Mi (x − ci )
=0
i
||x − ci ||3
Equivalently, we are looking for solutions of the equation ∇V = 0, where:
X GmMi
V =−
i
||x − ci ||
68 4. THE N BODY PROBLEM

(3) Let us first examine the simplest case, that of 3 bodies on a line, at distinct
positions. Here, by obvious reasons, we have 2 centers of gravity, as follows:
•M1 −− ⋆x1 −− •M2 −− ⋆x2 −− •M3
More generally, again by obvious reasons, a system of aligned bodies M1 , . . . , Mk has
k − 1 centers of gravity, one in between each pair of consecutive bodies:
•M1 −− ⋆x1 −− •M2 −− ⋆x2 −− . . . . . . −− ⋆xk−1 −− •Mk
(4) In the 2D case now, and with k ≥ 3, we are looking the the center of gravity of a
triangle, with vertices weighted by masses M1 , M2 , M3 . The simplest possible situation is
that of an equilateral triangle, with equal masses M, M, M at its vertices, and it is quite
clear here that we will have 3 solutions, 1 lying on each of the 3 symmetry axes.
(5) But this is of course quite bad news, because we have now 3 solutions, instead of
the 2 ones found in one dimension, in (3). And for the disaster to be complete, let us
attempt now to compute these solutions, and see how they look like.
(6) For this purpose, the best is to use here complex numbers, with our triangle being
1, w, w2 in the complex plane, with w being the standard third root of unity, namely:
w = e2πi/3
We will only look for the solution on the Ox axis, say r ∈ R, with the other solutions
being wr, w2 r, by symmetry reasons. The equation of our solution is as follows:
r−1 r−w r − w2
+ + =0
|r − 1|3 |r − w|3 |r − w2 |3
By simplifying at left and using 1 + w + w2 = 0 for the right terms, this reads:
2r + 1 1
3
=
|r − w| (1 − r)2
(7) And that is pretty much it, for computing r we must raise this to the square, which
leads us into a degree 6 equation, and we will not do this. As a conclusion, things are
hard for the equilateral triangle equally weighted, and can only be harder in general.
(8) As a final comment, however, and forgetting perhaps about exact numerics, this is
a geometry problem. Indeed, the equations in (2) suggest looking at R3 with a “hole” at
each Mi , and the problem is that of understanding which points are in equilibrium. This
is some sort of an Einstein idea, and we will be back to this later in this book. □

Moving ahead, and looking for an easier question, let us still examine the gravity of a
rigid object, formed by fixed bodies M1 , . . . , Mk , but at a distance. We have here:
 4A. CENTER OF MASS 69

Theorem 4.5. Consider a rigid object, consisting of fixed bodies M1 , . . . , Mk , located

at positions c1 , . . . , ck ∈ R3 . The corresponding gravitation force, F = −∇V with
X GmMi
V =−
i
||x − ci ||
can be approximated by the force coming from the center of mass, Fc = −∇V with
P
Gm i Mi
Vc = −
||x − c||
at order zero, when x >> ci . The correction term can be computed as well.
Proof. We have several assertions here, the idea being as follows:
(1) The first assertion, F ≃ Fc when x >> ci , is something clear, and with this not
even needing c to be the center of mass. Indeed, with V, Vc as above, we have:
X GmMi
V = −
i
||x − ci ||
X GmMi
≃ −
i
||x − c||
= Vc
(2) Regarding now the correction term, the error to be estimated is:
P
X GmMi Gm i Mi
V − Vc = − +
i
||x − ci || ||x − c||
 
X 1 1
= GmMi −
i
||x − c|| ||x − ci ||
(3) By translation we may assume c = 0. In order to evaluate the difference of inverses
on the right, we use the following trick, valid for any two vectors x >> d:
1 1 ||x − d|| − ||x||
− =
||x|| ||x − d|| ||x|| · ||x − d||
||x − d||2 − ||x||2
=
||x|| · ||x − d|| · (||x|| + ||x − d||)
||d||2 − 2 < x, d >
=
||x|| · ||x − d|| · (||x|| + ||x − d||)
< x, d >
≃ −
||x||3
To be more precise, in the last step we have neglected upstairs the order 0 term ||d||2 ,
and downstairs we have approximated all norms appearing there by ||x||.
70 4. THE N BODY PROBLEM

(4) Getting back now to the formula in (2), assuming c = 0 by translation, as already
mentioned above, and by using the trick in (3), we obtain:
 
X 1 1
V − Vc = GmMi −
i
||x|| ||x − ci ||
X < x, ci >
≃ − GmMi
i
||x||3
* +
Gm X
= − x, Mi ci
||x||3 i

(5) Before doing some more mathematics, let us think a bit at the meaning of the
above formula. Normally this is the formula of the first order correction, but when c = 0
is the center of mass, this means precisely that we must have:
X
Mi ci = 0
i

Thus, this correction that we computed vanishes. So, in short, on one hand good
news, we are on the good way, with the center of mass c being the ideal location for the
origin of the approximating potential Vc , but on the other hand, bad news, precisely due
to this fact, we must fine-tune our tricks from (3) above, get a better inequality there,
that we can use in order to compute the nonzero correction at c.
(6) So here we go again with estimates. Getting back to the end of the computation
in (3), we need there an estimate for ||x − d||, and this can be found as follows:
p
||x − d|| = ||x||2 + ||d||2 − 2 < x, d >
p
≃ ||x||2 − 2 < x, d >
≃ ||x||− < x, d >
With this in hand, we can improve our master estimate in (3), as follows:
1 1 ||d||2 − 2 < x, d >
− =
||x|| ||x − d|| ||x|| · ||x − d|| · (||x|| + ||x − d||)
||d||2 − 2 < x, d >
≃
||x|| · (||x||− < x, d >) · (2||x||− < x, d >)
||d||2 − 2 < x, d >
≃
||x|| · (2||x||2 − 3||x|| < x, d >)
||d||2 − 2 < x, d >
=
||x||2 · (2||x|| − 3 < x, d >)
 4A. CENTER OF MASS 71

Now if we denote by α the angle between x and d, this formula becomes:

1 1 ||d||2 − 2||x|| · ||d|| · cos α
− ≃
||x|| ||x − d|| ||x||2 · (2||x|| − 3||x|| · ||d|| · cos α)
||d|| ||d|| − 2||x|| · cos α
= ·
||x||3 2 − 3||d|| · cos α
(7) Thus, we can improve the estimate found in (4), with the conclusion that at the
center of mass, taken to be at the origin, c = 0, the error is as follows, with αi being the
angles between our body m, and the components M1 , . . . , Mk of the rigid body:
 
X 1 1
V − Vc = GmMi −
i
||x|| ||x − ci ||
X ||ci || ||ci || − 2||x|| · cos αi
≃ ·
i
||x||3 2 − 3||ci || · cos αi
(8) Assuming in addition that we are in a generic position, where αi ̸= π/2 for any i,
the upper terms can be further estimated, and we obtain in this way:
X ||ci || 2||x|| · cos αi
V − Vc ≃ − ·
i
||x||3 2 − 3||ci || · cos αi
X ||ci || 2 cos αi
= − 2
·
i
||x|| 2 − 3||ci || · cos αi
Summarizing, we have proved our result, and with a few bonus conclusions, namely
that the center of mass c is indeed the ideal location for the approximate potential Vc ,
and that the computation of the error term there ultimately involves the angles α1 , . . . , αk
between our body m, and the components M1 , . . . , Mk of our rigid body. □
Before going ahead and leaving this subject, let us mention that an interesting gen-
eralization of the above comes when considering a “true” rigid body, made of matter
arranged according to a certain density function ρ inside it. We will not go into details
here, and instead let us just formulate a basic statement, as follows:
Theorem 4.6. Consider a rigid body, made of matter arranged according to a certain
density function ρ inside it. Its gravitational force is then F = −∇V with
Z
Gmρ(z)
V =− dz
||x − z||
and can be approximated by the force coming from the center of mass, Fc = −∇V with
R
Gm ρ(z)dz
Vc = − R dz
x − uρ(u)du
at order zero, when m is far away. The correction term can be computed as well.
72 4. THE N BODY PROBLEM

Proof. Here the formulae in the statement, which are perfectly similar to those
in Theorem 4.5, can be obtained via the usual philosophy “replace sums by integrals”.
Observe in particular the formula of the center of mass, producing Vc , namely:
Z
c = uρ(u)du

As for the last assertion, this can only hold too, by proceeding as in the proof of
Theorem 4.5, and replacing everywhere at the end the sums by integrals. □

The above results, Theorem 4.5 and Theorem 4.6, are both quite interesting, and
suggest a whole string of further questions, and potential generalizations. What happens
in the context of Theorem 4.5 when the constituents M1 , . . . , Mk are allowed to move a bit,
say by being confined by an external force? Then, what happens when the constituents
M1 , . . . , Mk are allowed to freely move? Also, what about Theorem 4.6, if we allow there
some kind of fluid movement inside the body? And so on. These are obviously all difficult
questions, of general N -body problem type, so perhaps time to stop here.

Moving ahead now, let us examine now various conservation questions. The simplest
problematics here is most likely that of the angular momentum.

As a first question, we know from chapter 2 that when M2 moves around M1 , positioned
at 0, its angular momentum J2 is constant. By symmetry, if we regard M1 moving around
M2 , fixed at 0, its angular momentum J1 will be constant too. Thus in both cases
J = J1 + J2 is constant, which raises the question whether J is constant or not when
computed at other points of R3 . And here, we first have the following result:
Proposition 4.7. In the context of the 2-body problem, the following happen:
(1) J is conserved when assuming that M1 or M2 is fixed at 0.
(2) More generally, J is conserved at any λ1 M1 + λ2 M2 , with λ1 + λ2 = 1.
(3) In particular, J is conserved when computed at the center of mass.
(4) However, J is not conserved when assuming that M1 or M2 is fixed at d ̸= 0.
Proof. We have several assertions here, the idea being as follows:
(1) This is something that we know well, from chapter 2, as explained above.
(2) Assume first, as in (1), that M1 is fixed at 0, and that M2 moves around it, with
position vector x ∈ R3 . Given parameters λ1 , λ2 satisfying λ1 + λ2 = 1, let us set:
y = λ1 · 0 + λ2 · x = λ2 x
We make now the convention that at t = 0 this point was the origin, with coordinate
axes parallel to our original coordinate axes, and that at any t > 0 this is still the origin,
 4A. CENTER OF MASS 73

with the directions of the coordinate axes being unchanged. Thus, we have a new frame,
and the coordinates of our objects M1 , M2 with respect to this new frame are:
z1 = −λ1 x , z2 = λ2 x
But in this new frame the momenta of M1 , M2 are both proportional, by factors
λ21 M2 /M1 and λ22 , to the original momentum of M2 , computed in (1), which was constant.
Thus both these momenta J1 , J2 are constant, and so is their sum J = J1 + J2 .
(3) This follows from (2), with λ1 = M2 /(M1 + M2 ) and λ2 = M1 /(M1 + M2 ).
(4) Assuming that M1 is fixed at a given point d ̸= 0, and that M2 travels around it,
with position vector d + x, with x being as in (1), we still have J1 = 0, and so:
J = J2
= (d + x) × p
= d×p+x×p
= d × p + constant
̸= constant
Thus, we are led to the conclusions in the statement. □

As a conclusion, the conservation of angular momentum depends on the “quality” of

the frame that we are using. In a good frame, as in (1,2,3) above, the momentum will be
conserved, while in a bad frame, as in (4), the momentum will be not conserved.

Generally speaking, we will be talking frames later, when discussing relativity. In

connection with our questions here we will be quite brief, and in order to keep moving,
let us formulate the following informal definition, that will do:
Definition 4.8. An inertial frame is a frame where all basic formulae, namely
Gm1 m2
||F || = , F = ma , a = v̇ , v = ẋ , F12 = −F21
||x1 − x2 ||2
hold, with the last formula standing for Newton’s action-reaction principle.
To be more precise here, the first 4 formulae are something that we have been heavily
using, so far in this book. As for the last formula, also called Newton’s third law, this
expresses the fact that when an object 1 acts on an object 2, say via gravity, with force
F12 , then object 2 acts as well on object 1, with force F21 = −F12 .

As already mentioned, we will discuss all this later, more in detail. In relation with
our present considerations, we have the following basic examples:
74 4. THE N BODY PROBLEM

Proposition 4.9. In the context of the 2-body problem, the frames of type

λ1 M1 + λ2 M2

constructed above are all non-inertial, including the center of mass frame.

Proof. Since our definition of an inertial frame was something quite informal, so will
be this proof. We want to check whether the forces between M1 , M2 satisfy:

GM1 M2 (x1 − x2 )
F12 = −F21 =
||x1 − x2 ||3

(1) In the case of the frame centered at M1 , the formula F12 = −F21 certainly does
not hold, because the acceleration of M1 is in this case 0̈ = 0, and so no force acting upon
it, at least from our calculus viewpoint. The same holds for the frame centered at M2 .

(2) In general now, where we have parameters λ1 , λ2 satisfying λ1 + λ2 = 1, as in

Proposition 4.7, as explained there, the positions of M1 , M2 are:

z1 = −λ1 x , z2 = λ2 x

Thus the forces acting upon M1 , M2 , computed according to calculus, are:

F21 = −M1 λ1 ẍ , F12 = −M2 λ2 ẍ

Thus, in order to have F12 = −F21 , the parameters λ1 , λ2 satisfy M1 λ1 = M2 λ2 . But

these are exactly the parameters of the center of mass.

(3) But the center of mass frame is not inertial either, because due to the fact that
we performed a dilation, the magnitude of F12 = −F21 is not the correct one. □

Now back to momentum, we have the following extension of Proposition 4.7 (3):

Theorem 4.10. In an inertial frame, the total angular momentum

X
J= xi × p i
i

of a system of bodies M1 , . . . , Mk is conserved.

 4A. CENTER OF MASS 75

Proof. Our inertial frame assumption tells us that we can use at will all formulae in
Definition 4.8, and by using them, and notably F12 = −F21 , we obtain:
X X
J˙ = xi × Fji
i j̸=i
X
= xi × Fji + xj × Fij
i<j
X
= xi × Fji − xj × Fji
i<j
X
= (xi − xj ) × Fji
i<j
= 0
Now since we have J˙ = 0, the angular momentum J is conserved, as claimed. □
Moving ahead now, our next problem will concern the conservation of energy. Here
things are a bit similar with angular momentum, but more can be said, as follows:
Theorem 4.11. With a suitable potential formalism, the total energy
X
E= Ti + Vi
i
of a system of bodies M1 , . . . , Mk is conserved. Also, the individual energy
X
E′ = T ′ + Vi′
i
of an extra body m added is conserved as well, again with a suitable formalism.
Proof. There are several questions here, the idea being as follows:
P
(1) In what regards T = i Ti we have, exactly as in the 2-body problem:
X
Ṫ = < vi , Fji >
i̸=j
X
= < vi , −∇Vji >
i̸=j
X
= − V̇ji
i̸=j

(2) With this in hand, we can group pairs of terms, as in the proof of Theorem 4.10,
and we are led to the conclusion in the statement, and with the remark however that all
the potentials appearing there are now time-dependent.
(3) In what regards now the second assertion, this is not exactly something of the same
nature as the first assertion, becase assuming that by some kind of miracle we would have
76 4. THE N BODY PROBLEM

a theory where all the bodies conserve their energy, the total energy of the system would
be trivially conserved too, just by summing, and this does not look normal. So, getting
now to the second assertion as formulated, we have, by computing as in (1) above:
X
Ṫ ′ = − V̇i′
i

(4) Thus, we are led to the conclusion in the statement, with the problem however
that all the potentials appearing there are now time-dependent. □
The above results raise several interesting theoretical questions. We will be back to
all this later in this book, when discussing general relativity.

4b. The N body problem

With the above questions discussed, let us go back now to our to-do list (1-7), from
the beginning of this section. We have all sorts of difficult questions there, and we will
focus first on (6), namely 3-body problem with one of the bodies being tiny.

Things here are quite complicated, technically speaking. Let us start with:
Definition 4.12. The 3-body problem is the general gravitational problem for three
bodies M1 , M2 , M3 , with the corresponding equations of motion being as follows:
GM2 (x1 − x2 ) GM3 (x1 − x3 )
ẍ1 = − −
||x1 − x2 ||3 ||x1 − x3 ||3
GM3 (x2 − x3 ) GM1 (x2 − x1 )
ẍ2 = − −
||x2 − x3 ||3 ||x2 − x1 ||3
GM1 (x3 − x1 ) GM2 (x3 − x2 )
ẍ3 = − −
||x3 − x1 ||3 ||x3 − x2 ||3
The planar 3-body problem is this problem, in a plane. The restricted 3-body problem is
also this problem, in the situation M1 , M2 >> M3 , and usually considered in a plane.
The general 3-body problem has all sorts of weird solutions, and cannot be solved via
exact mathematics. You would probably say no problem, just give it to a computer, but
guess what, the computer cannot solve that either. The problem with computers is that,
no matter how big and powerful they are, they can still only operate with a finite amount
of data, and at a certain speed, and what happens, somehow, is that the 3 equations in
Definition 4.12 can produce after some time t > 0 all sorts of bizarre phenomena, going
in all senses, and managing all this is impossible, even for a powerful computer.

The best here is to go on internet, and look up some animations. Just by looking at
them, and how bizarre they can be, you will quickly realize that such things are beyond
what you, what your professors, and even what a powerful computer, can do.
 4C. LAGRANGE POINTS 77

4c. Lagrange points

This being said, and in connection now with our satellite problem that we have in
mind, question (6) on our to-do list, some mathematics is however possible, in certain
simple cases, and we have the following remarkable result, due to Lagrange:
Theorem 4.13. The restricted 3-body problem has 5 distinguished solutions,
·L4

·L3 •M ·L1 •m ·L2

·L5
called Lagrange points L1-L5, whose positions with respect to M, m are as above.
Proof. This is certainly something quite complicated, using all sorts of advanced
mathematics, further building on what has been developed in the above, and we won’t
get into details here. Instead, let us describe at least how each L1-L5 functions:
(1) L1 is the most intuitive solution, placed in between M and m, and closer to m,
at that precise point where the gravitation of M equals in magnitude the gravitation of
m. The math here for finding the distances is very simple, but recall however that m is
supposed to move around M , on an ellipsis. Thus, the picture is that an object ε placed
at L1 moves as well around M , by staying aligned with m, on a smaller ellipsis.
(2) One problem with L1 comes from the fact that it is unstable, in the sense that an
object ε placed around there, but not exactly there, will not stay there. Indeed, was ε to
be placed a tiny little bit towards M , it will start slowly moving towards M , and gone
it will be. And the same can happen in the other sense, with m being able to capture
it too, since L1 is some sort of no man’s land between the gravities of M, m, which look
equal from there. Thus, L1 looks like some sort of fake solution to the problem.
(3) However, all this is useful in practice, because with just a little bit of homemade
acceleration, from time to time, in order to correct the trajectory and keep it on L1, a
satellite can be placed there at L1, and will stay there. In fact, most of the scientific
satellites are placed there at L1, first because its proximity to Earth, but also because
there is no dirt like asteroids trapped there, due to the fact that L1 is unstable.
(4) Getting to L2 now, the functioning mechanism here is different, crucially relying
on the fact that m, and so L2 too, does move around M , on an elliptic orbit. Indeed, what
looks impossible in 1D, namely an object ε placed behind m not to be attracted by both
M, m, and start going towards them, is now possible in the context of the 2D elliptical
movement, with the normal movement of ε with respect to M being slightly altered by
the presence of m, which in practice tends to pull ε away from M, m, and with the precise
78 4. THE N BODY PROBLEM

distance being that where equilibrium is achieved, in all this. As for L1, this point L2 is
not far from Earth, and unstable, making it usable for satellites.
(5) In what regards now L3, yet another functioning mechanism going on here, which
is this time something very simple, namely an object ε placed there at L3 will simply
travel around M , in a standard elliptic way, basically on the same orbit as m, slightly
adjusted as to take into account the gravity of m too. Again, this is an unstable point,
suitable for satellites, but who would go up there to install one.
(6) Finally, regarding L4 and L5, these are two extra solutions, discovered later by
Lagrange, located at the positions where they form, along with M, m, equilateral triangles.
An object ε say placed at L4 will stay there, or rather travel on an elliptical orbit around
M passing through L4, keeping its L4 relative position with respect to M, m, due to the
fact that, due to the geometry of the equilateral triangle ε − m − M , the extra pull from
m keeps it in tune with m, on that smaller orbit. And the same goes for L5.
(7) These two last points L4, L5 are stable, provided that the masses M, m of the two
big objects satisfy M/m > 24.96, which is the case for instance for the Sun-Earth system,
and for the Earth-Moon system too. However, the stability makes them unsuitable for
satellite use, due to the tons of space garbage accumulated there, over the years. □
So long for the N -body problem, and for the Lagrange points. Obviously there is some
very interesting mathematics and physics going on here, which is relevant for anything
in relation with the Solar System, be that natural or human-made. In fact, speaking
astronomy, the Sun-Earth-Moon system is already something that you can spend your
whole life of scientist on, because due to pure 3-body gravitation, and then also to various
imperfections in the shape, density, and many other parameters of these objects, things
in this system are continuously evolving, over the passing years, and with the very long
term predictions on what will really happen being quite complicated to make.

4d. Friction and drag

Getting back now to our to-do list from the beginning of this chapter, let us focus
now on question (2) there, namely the 2-body problem with atmospheric drag. This is
something not discussed yet, and of crucial real-life importance, and for everything in
relation with engineering, and which can bring us far, deep into fluid mechanics.

Mathematically, we have to go back to the Kepler 2-body problem, with the aim of
doing some more study here, in the parabolic trajectory case, which is traditionally the
field of ballistics. A well regulated militia being necessary to the security of a free state,
we have the following result, that you might find of interest:
Proposition 4.14. Ballistics.
 4D. FRICTION AND DRAG 79

Proof. This follows indeed by doing some computations. In what regards the notion
of escape velocity, the conclusions here might seem quite surprising. □
Speaking arms, war and related topics, we have as well the following result:
Proposition 4.15. Rockets, again.
Proof. We already know about rockets from chapter 2. In order to beat now grav-
ity, we can just write down the rocket equations, coming from the conservation of the
momentum principle, and do some computations in relation with gravitation. □
In the elliptic trajectory case, which is the most interesting, mathematically speaking,
and for various peaceful applications too, such as scientific satellites, we have:
Proposition 4.16. Satellites.
Proof. This follows again by doing some computations. □
Still in the elliptic trajectory case, at a more advanced level now, allowing the launch
of satellites far away from the dust, radiation and other mess on Earth, we have:
Proposition 4.17. Outer space satellites.
Proof. This is something that we already know, from our study of Lagrange points,
performed in the above. □
Moving ahead now towards a different problematics, and in practice by getting back
to, guess what, the good old Kepler 2-body problem, most of the questions studied above,
regarding ballistics and rockets, are subject in the real life to atmospheric drag. Thus, we
must discuss now the needed corrections. We first have here the following result:
Proposition 4.18. Ballistics, with drag.
Proof. This follows indeed by doing some computations. In what regards the notion
of escape velocity, the conclusions here might seem quite surprising. □
Regarding now rockets, we have here the following result:
Proposition 4.19. Rockets, with drag.
Proof. This follows again by doing some computations. □
There are of course many other things that can be said on all the above, and about
versions of the various systems considered above, with all this being very useful for engi-
neering, or any other concrete application of mechanics, and we refer here as usual to our
go-to mechanics books, namely [2], [36], [39], [61], [64], [88].

Then, what about forgetting about objects moving through fluids, and investigating
the fluids themselves? There are many interesting things that can be said here.
80 4. THE N BODY PROBLEM

Let us start with the following basic fact, which was the beginning of everything, going
back to work of Boyle, Charles, Avogardo, Gay-Lussac, Clapeyron and others from the
17th, 18th and 19th centuries, and with final touches from Maxwell, Boltzmann, Gibbs
and others, in the late 19th and early 20th centuries:
Fact 4.20. The ideal gases satisfy the equation P V = kT , where:
(1) V is the volume of the gas, independently of the shape of the container used.
(2) P is the pressure of the gas, measured with a manometer.
(3) T is the temperature of the has, measured with a thermometer.
(4) k is a constant, depending on the gas.
That is, P V = kT basically tells us that “pressure and temperature are the same thing”.
At the first glance, for instance if you are a mathematician not used to this, this looks
more like a joke. Why not defining then P = T or vice-versa, you would say, and what is
the point with that long list of distinguished gentlemen having worked hard on this.
Error. The point indeed comes from the following:
Explanation 4.21. In the equation of state P V = kT , as formulated above, the
pressure P and the temperature T appear more precisely as follows,
(1) The manometer read comes from the gas molecules pushing a piston, so P is a
statistical quantity, coming from the statistics of the molecular speeds,
(2) The thermometer read is something even more complicated, and T is as well a
statistical quantity, coming from the statistics of the molecular speeds,
so P V = kT is something non-trivial, telling us that the mathematical machinery produc-
ing P, T , via manometer and thermometer, out of the molecular speeds, is the same.
Hope you got my point, and getting back now to historical details, Boyle, Charles,
Avogardo, Gay-Lussac, Clapeyron, joined by Clausius, Carnot, Joule, Lord Kelvin and
others, first observed P V = kT , and then reached to a good understanding of what
this means, via an axiomatization of P and T . Later Maxwell started to look into the
molecular speeds and their statistics, then Boltzmann came with a tough mathematical
computation, proving P V = kT , and then, even later, Gibbs and others further built on
all this, by formalizing modern thermodynamics, in the form that is still used today.
More generally now, still regarding the gases, we have the following key result:
Theorem 4.22. Beyond the ideal gas setting, stating that we should have
P V = kT
the gases are subject to the Van der Waals equation
 α
P + 2 (V − β) = kT
V
depending on two parameters α, β > 0.
 4D. FRICTION AND DRAG 81

Proof. This is something quite tricky, with the correction parameters α, β > 0 ap-
pearing from a detailed study of the gas, from a kinetic viewpoint. For an introduction
to all this, we refer for instance to Huang [55], or Schroeder [80]. □

The above result is of key importance, and takes us into rethinking everything that
what we know about the ideal gases, which must be replaced with Van der Waals gases,
at the advanced level. Among the main consequences of this replacement, the isobars,
isochores, isothermals and adiabatics of the ideal gases, given by simple formulae, must
be replaced by isobars, isochores, isothermals and adiabatics for the Van der Waals gases,
which are no longer something trivial, with some interesting math being now involved.

Among others, the Van der Waals gas study makes appear some interesting points on
the isothermals, called triple and critical points of the gas.

Our study of the gases so far, using various techniques from thermodynamics and
statistical mechanics, which makes appear various critical points, makes the connection
with the general theory of matter, which can be summarized as follows:
Fact 4.23. Ordinary matter appears in 3 forms, namely solid, liquid and gaseous,
roughly appearing according to the following generic diagram
PO

solid cp

liquid

tp gas

0 / T
with tp, cp standing for the triple and critical points. Also, at low or high temperatures we
have interesting phenonema like Bose-Einstein condensation, and plasma.
Generally speaking, the fluids are by definition the non-solids, and as such, they fall
into liquids, gases and plasma. Thus, a fluid is something quite complicated, whose
understanding requires advanced thermodynamics, and states of matter theory. Also, we
can also see that the modelling of a fluid can only be something quite complicated too,
again requiring advanced thermodynamics, and states of matter theory.
82 4. THE N BODY PROBLEM

Getting started now, fluid mechanics is a complex science, which is organized by taking
into account the two main properties that a fluid can have or not, which are:

(1) Compressibility.

(2) Viscosity.

Regarding compressibility, this is certainly a property of the gases, but not of most of
the liquids, at least when idealized. We will assume here that our fluid is incompressible,
which in practice means more or less that we are dealing with liquids, of rather “regular”
type. However, this will be not the general rule, for instance because materials like sand,
or snow, that we are particularly interested in, in view of their obvious strong link with
classical mechanics, and with our modelling abilities so far, are incompressible too.

Regarding viscosity, this is something far more tricky. Intuitively, this comes from
the mutual “friction” of the constituent molecules, when the fluid is moving, and with
this being something quite difficult to model and understand, via precise mathematical
equations. Without getting into details, for the moment, let us mention that, from this
point of view, the fluids, or rather liquids, fall into 3 main classes, namely:

– Inviscid. This is intuitively the case of regular water, and other familiar liquids.
However, this remains an idealization, with the true inviscid fluids, in the real life, being
basically only the superfluids, met at very low temperatures.

– Newtonian. This is intuitively the case for most of the familiar visquous fluids,
from the real life, whose viscosity is proportional to the applied stress, and with this
proportionality being known as Newton’s law on viscosity.

– Non-Newtonian. These are visquous fluids which do not obey to Newton’s law on
viscosity, and there are plenty of them, all very interesting, such as paint, toothpaste,
ketchup and many more, not to forget basic things like snow or sand.

As a conclusion, we can see that, even when looking at the simplest 2 possible things
that can be said about a fluid, namely compressibility and viscosity, we end up with a
whole menagerie of fluids, with each of them corresponding to its own branch of fluid
mechanics. In what concerns us, in relation with our goals here, we will be mostly
interested in the incompressible fluids, which can be either inviscid, or non-Newtonian.

As a first observation, which is of key importance in fluid dynamics, we have:

Observation 4.24. An incompressible fluid is right away something mathematical,
whose dynamics is described by a diffeomorphism, evolving in time t > 0.
 4D. FRICTION AND DRAG 83

Obviously, this is something very basic, coming from the very nature of the incom-
pressibility property. To be more precise, assuming for instance that we have colored our
fluid initially, say into tiny little cubes of red, yellow, green and so on, after some time
t > 0 we will obviously still have a mixture of red, yellow, green and so on, appearing in
equal parts, and the function f : R3 → R3 mapping red to red, yellow to yellow, green to
green, and so on, is what we call in mathematics a diffeomorphism.

In practice now, in order to reach from this to a PDE, let us assume that we are in
the simplest case, where our fluid is inviscid, and also adiabatic, or with zero thermal
conductivity. The dynamics of the diffeomorphisms f : R3 → R3 will be then, intu-
itively speaking, basically governed by the mechanics of the red, yellow, green and so on
components, and we are led in this way to the Euler equations, which are as follows:
u̇+ < u, ∇ > u = −∇w
< ∇, u >= 0
Here u is the vector velocity field, and w is the thermodynamic work, with the quantity
on the right from the first equation being as follows, p being the pressure:
∇p
∇w =
ρ
With this replacement made, the Euler equations become:
∇p
u̇+ < u, ∇ > u = −
ρ
< ∇, u >= 0
Here the first equation, which is the important one, is called the Euler momentum
equation. This equation can be further complicated by adding an acceleration term +g
on the right, accounting for exterior forces, gravitational, or magnetic or of some other
kind. As for the second equation, this is the incompressibility constraint.

Although we will not really need this here, at least at this stage of our discussion,
let us record as well what happens in the visquous case, by staying as before in the
incompressible setting. Here the Euler momentum equation gets replaced by the Navier-
Stokes equation, which is as follows, with v being the kinematic viscosity:
∇p
u̇+ < u, ∇ > u = − + v∆u
ρ
Summarizing, in relation with our modelling questions for the incompressible fluids,
we will be mostly interested in the Euler equation, and its versions.

Finally, we have seen the basic physics of the solid-fluid and fluid-fluid interactions.
For our discussion to be complete, we should talk as well about solid-solid interactions.
84 4. THE N BODY PROBLEM

4e. Exercises
Exercises:
Exercise 4.25.
Exercise 4.26.
Exercise 4.27.
Exercise 4.28.
Exercise 4.29.
Exercise 4.30.
Exercise 4.31.
Exercise 4.32.
Bonus exercise.
 Part II

Order and chaos

 Gotta make a move to a town
That’s right for me
Town to keep me moving
Keep me grooving with some energy
 CHAPTER 5

Noether theorem

5a.
5b.
5c.
5d.
5e. Exercises
Exercises:
Exercise 5.1.
Exercise 5.2.
Exercise 5.3.
Exercise 5.4.
Exercise 5.5.
Exercise 5.6.
Exercise 5.7.
Exercise 5.8.
Bonus exercise.

87
 CHAPTER 6

Bifurcation theory

6a.
6b.
6c.
6d.
6e. Exercises
Exercises:
Exercise 6.1.
Exercise 6.2.
Exercise 6.3.
Exercise 6.4.
Exercise 6.5.
Exercise 6.6.
Exercise 6.7.
Exercise 6.8.
Bonus exercise.

89
 CHAPTER 7

Singularities, ADE

7a.
7b.
7c.
7d.
7e. Exercises
Exercises:
Exercise 7.1.
Exercise 7.2.
Exercise 7.3.
Exercise 7.4.
Exercise 7.5.
Exercise 7.6.
Exercise 7.7.
Exercise 7.8.
Bonus exercise.

91
 CHAPTER 8

Fluids, granularity

8a.
8b.
8c.
8d.
8e. Exercises
Exercises:
Exercise 8.1.
Exercise 8.2.
Exercise 8.3.
Exercise 8.4.
Exercise 8.5.
Exercise 8.6.
Exercise 8.7.
Exercise 8.8.
Bonus exercise.

93
 Part III

Galaxies, clustering
Join me for a ride
Speed up the music
Join me for a ride
Maximum overdrive
 CHAPTER 9

Celestial mechanics

9a.
9b.
9c.
9d.
9e. Exercises
Exercises:
Exercise 9.1.
Exercise 9.2.
Exercise 9.3.
Exercise 9.4.
Exercise 9.5.
Exercise 9.6.
Exercise 9.7.
Exercise 9.8.
Bonus exercise.

97
 CHAPTER 10

KAM theory

10a.
10b.
10c.
10d.
10e. Exercises
Exercises:
Exercise 10.1.
Exercise 10.2.
Exercise 10.3.
Exercise 10.4.
Exercise 10.5.
Exercise 10.6.
Exercise 10.7.
Exercise 10.8.
Bonus exercise.

99
 CHAPTER 11

Galaxies, clustering

11a.
11b.
11c.
11d.
11e. Exercises
Exercises:
Exercise 11.1.
Exercise 11.2.
Exercise 11.3.
Exercise 11.4.
Exercise 11.5.
Exercise 11.6.
Exercise 11.7.
Exercise 11.8.
Bonus exercise.

101
 CHAPTER 12

Numeric aspects

12a.
12b.
12c.
12d.
12e. Exercises
Exercises:
Exercise 12.1.
Exercise 12.2.
Exercise 12.3.
Exercise 12.4.
Exercise 12.5.
Exercise 12.6.
Exercise 12.7.
Exercise 12.8.
Bonus exercise.

103
 Part IV

Interstellar dust
 I got the poison
I got the remedy
I got the pulsating
Rhythmical remedy
 CHAPTER 13

Star collapse

13a.
13b.
13c.
13d.
13e. Exercises
Exercises:
Exercise 13.1.
Exercise 13.2.
Exercise 13.3.
Exercise 13.4.
Exercise 13.5.
Exercise 13.6.
Exercise 13.7.
Exercise 13.8.
Bonus exercise.

107
 CHAPTER 14

Interstellar dust

14a.
14b.
14c.
14d.
14e. Exercises
Exercises:
Exercise 14.1.
Exercise 14.2.
Exercise 14.3.
Exercise 14.4.
Exercise 14.5.
Exercise 14.6.
Exercise 14.7.
Exercise 14.8.
Bonus exercise.

109
 CHAPTER 15

Charges, demons

15a.
15b.
15c.
15d.
15e. Exercises
Exercises:
Exercise 15.1.
Exercise 15.2.
Exercise 15.3.
Exercise 15.4.
Exercise 15.5.
Exercise 15.6.
Exercise 15.7.
Exercise 15.8.
Bonus exercise.

111
 CHAPTER 16

Back to stars

16a.
16b.
16c.
16d.
16e. Exercises
Congratulations for having read this book, and no exercises for this final chapter.

113
 Bibliography

[1] V.I. Arnold, Ordinary differential equations, Springer (1973).

[2] V.I. Arnold, Mathematical methods of classical mechanics, Springer (1974).
[3] V.I. Arnold, Catastrophe theory, Springer (1974).
[4] V.I. Arnold, Lectures on partial differential equations, Springer (1997).
[5] V.I. Arnold and B.A. Khesin, Topological methods in hydrodynamics, Springer (1998).
[6] V.I. Arnold, V.V. Kozlov and A.I. Neishtadt, Mathematical aspects of classical and celestial me-
chanics, Springer (2006).
[7] N.W. Ashcroft and N.D. Mermin, Solid state physics, Saunders College Publ. (1976).
[8] M.F. Atiyah, K-theory, CRC Press (1964).
[9] M.F. Atiyah, The geometry and physics of knots, Cambridge Univ. Press (1990).
[10] M.F. Atiyah and I.G. MacDonald, Introduction to commutative algebra, Addison-Wesley (1969).
[11] T. Banica, Introduction to modern physics (2024).
[12] T. Banica, Principles of thermodynamics (2024).
[13] T. Banica, The joys of relativity theory (2024).
[14] G.K. Batchelor, An introduction to fluid dynamics, Cambridge Univ. Press. (1967).
[15] R.J. Baxter, Exactly solved models in statistical mechanics, Academic Press (1982).
[16] N. Berline, E. Getzler and M. Vergne, Heat kernels and Dirac operators, Springer (2004).
[17] S.J. Blundell and K.M. Blundell, Concepts in thermal physics, Oxford Univ. Press (2006).
[18] S.M. Carroll, Spacetime and geometry, Cambridge Univ. Press (2004).
[19] P.M. Chaikin and T.C. Lubensky, Principles of condensed matter physics, Cambridge Univ. Press
(1995).
[20] A.R. Choudhuri, Astrophysics for physicists, Cambridge Univ. Press (2012).
[21] D.D. Clayton, Principles of stellar evolution and nucleosynthesis, Univ. of Chicago Press (1968).
[22] E.A. Coddington, An introduction to ordinary differential equations, Dover (1961).

115
116 BIBLIOGRAPHY

[23] A. Connes, Noncommutative geometry, Academic Press (1994).

[24] W.N. Cottingham and D.A. Greenwood, An introduction to the standard model of particle physics,
Cambridge Univ. Press (2012).
[25] P.A. Davidson, Introduction to magnetohydrodynamics, Cambridge Univ. Press (2001).
[26] P. Di Francesco, Meander determinants, Comm. Math. Phys. 191 (1998), 543–583.
[27] P.A.M. Dirac, Principles of quantum mechanics, Oxford Univ. Press (1930).
[28] M.P. do Carmo, Differential geometry of curves and surfaces, Dover (1976).
[29] M.P. do Carmo, Riemannian geometry, Birkhäuser (1992).
[30] S. Dodelson, Modern cosmology, Academic Press (2003).
[31] R. Durrett, Probability: theory and examples, Cambridge Univ. Press (1990).
[32] A. Einstein, Relativity: the special and the general theory, Dover (1916).
[33] L.C. Evans, Partial differential equations, AMS (1998).
[34] W. Feller, An introduction to probability theory and its applications, Wiley (1950).
[35] E. Fermi, Thermodynamics, Dover (1937).
[36] R.P. Feynman, R.B. Leighton and M. Sands, The Feynman lectures on physics, Caltech (1963).
[37] A.P. French, Special relativity, Taylor and Francis (1968).
[38] S. Galtier, Introduction to modern magnetohydrodynamics, Cambridge Univ. Press (2016).
[39] H. Goldstein, C. Safko and J. Poole, Classical mechanics, Addison-Wesley (1980).
[40] D.L. Goodstein, States of matter, Dover (1975).
[41] M.B. Green, J.H. Schwarz and E. Witten, Superstring theory, Cambridge Univ. Press (2012).
[42] W. Greiner, L. Neise and H. Stöcker, Thermodynamics and statistical mechanics, Springer (2012).
[43] D.J. Griffiths, Introduction to electrodynamics, Cambridge Univ. Press (2017).
[44] D.J. Griffiths and D.F. Schroeter, Introduction to quantum mechanics, Cambridge Univ. Press
(2018).
[45] D.J. Griffiths, Introduction to elementary particles, Wiley (2020).
[46] D.J. Griffiths, Revolutions in twentieth-century physics, Cambridge Univ. Press (2012).
[47] G.H. Hardy and E.M. Wright, An introduction to the theory of numbers, Oxford Univ. Press (1938).
[48] J. Harris, Algebraic geometry, Springer (1992).
[49] W.A. Harrison, Solid state theory, Dover (1970).
[50] A. Hatcher, Algebraic topology, Cambridge Univ. Press (2002).
 BIBLIOGRAPHY 117

[51] M.P. Hobson, G.P. Efstathiou and A.N. Lasenby, General relativity, Cambridge Univ. Press (2006).
[52] H. Hofer and E. Zehnder, Symplectic invariants and Hamiltonian dynamics, Birkhäuser (1994).
[53] L. Hörmander, The analysis of linear partial differential operators, Springer (1983).
[54] R.A. Horn and C.R. Johnson, Matrix analysis, Cambridge Univ. Press (1985).
[55] K. Huang, Introduction to statistical physics, CRC Press (2001).
[56] K. Huang, Quarks, leptons and gauge fields, World Scientific (1982).
[57] K. Huang, Fundamental forces of nature, World Scientific (2007).
[58] K. Huang, A superfluid universe, World Scientific (2016).
[59] V.F.R. Jones, Subfactors and knots, AMS (1991).
[60] L.P. Kadanoff, Statistical physics: statics, dynamics and renormalization, World Scientific (2000).
[61] T. Kibble and F.H. Berkshire, Classical mechanics, Imperial College Press (1966).
[62] M. Kumar, Quantum: Einstein, Bohr, and the great debate about the nature of reality, Norton
(2009).
[63] T. Lancaster and K.M. Blundell, Quantum field theory for the gifted amateur, Oxford Univ. Press
(2014).
[64] L.D. Landau and E.M. Lifshitz, Course of theoretical physics, Pergamon Press (1960).
[65] P. Lax, Linear algebra and its applications, Wiley (2007).
[66] P. Lax, Functional analysis, Wiley (2002).
[67] J.M. Lee, Introduction to topological manifolds, Springer (2011).
[68] J.M. Lee, Introduction to smooth manifolds, Springer (2012).
[69] J.M. Lee, Introduction to Riemannian manifolds, Springer (2019).
[70] M.L. Mehta, Random matrices, Elsevier (2004).
[71] D. McDuff and D. Salamon, Introduction to symplectic topology, Oxford Univ. Press (2017).
[72] R.K. Pathria and and P.D. Beale, Statistical mechanics, Elsevier (1972).
[73] P. Petersen, Linear algebra, Springer (2012).
[74] P. Petersen, Riemannian geometry, Springer (2006).
[75] B.M. Peterson and B. Ryden, Foundations of astrophysics, Cambridge Univ. Press (2010).
[76] W. Rudin, Principles of mathematical analysis, McGraw-Hill (1964).
[77] W. Rudin, Real and complex analysis, McGraw-Hill (1966).
[78] B. Ryden, Introduction to cosmology, Cambridge Univ. Press (2002).
118 BIBLIOGRAPHY

[79] B. Ryden and R.W. Pogge, Interstellar and intergalactic medium, Cambridge Univ. Press (2021).
[80] D.V. Schroeder, An introduction to thermal physics, Oxford Univ. Press (1999).
[81] B. Schutz, A first course in general relativity, Cambridge Univ. Press (2009).
[82] I.R. Shafarevich, Basic algebraic geometry, Springer (1974).
[83] R. Shankar, Principles of quantum mechanics, Springer (1980).
[84] A.M. Steane, Thermodynamics, Oxford Univ. Press (2016).
[85] A.M. Steane, Relativity made relatively easy, Oxford Univ. Press (2012).
[86] S. Sternberg, Dynamical systems, Dover (2010).
[87] C.H. Taubes, Differential geometry, Oxford Univ. Press (2011).
[88] J.R. Taylor, Classical mechanics, Univ. Science Books (2003).
[89] M. Tenenbaum and H. Pollard, Ordinary differential equations, Dover (1963).
[90] J. von Neumann, Mathematical foundations of quantum mechanics, Princeton Univ. Press (1955).
[91] J. von Neumann and O. Morgenstern, Theory of games and economic behavior, Princeton Univ.
Press (1944).
[92] S. Weinberg, Foundations of modern physics, Cambridge Univ. Press (2011).
[93] S. Weinberg, Lectures on quantum mechanics, Cambridge Univ. Press (2012).
[94] S. Weinberg, Lectures on astrophysics, Cambridge Univ. Press (2019).
[95] S. Weinberg, Cosmology, Oxford Univ. Press (2008).
[96] H. Weyl, The theory of groups and quantum mechanics, Princeton Univ. Press (1931).
[97] H. Weyl, The classical groups: their invariants and representations, Princeton Univ. Press (1939).
[98] H. Weyl, Space, time, matter, Princeton Univ. Press (1918).
[99] E. Witten, Quantum field theory and the Jones polynomial, Comm. Math. Phys. 121 (1989), 351–
399.
[100] B. Zwiebach, A first course in string theory, Cambridge Univ. Press (2004).
 Index

action integral, 60 Hamilton principle, 60

algebraic curve, 25, 27 Hamiltonian, 62
amplitude, 12 harmonic oscillator, 16
approximate conic, 66 hyperbola, 27

Bose-Einstein condensate, 81 ideal gas, 80

inertial frame, 73
center of gravity, 67
center of mass, 65, 67 kinetic energy, 11, 41
center of mass frame, 73
chain rule, 50, 51 Lagrange equations, 61
circular motion, 12 Lagrange points, 77
confined motion, 12 Lagrangian, 60
conic, 25, 27 linear motion, 11
conservation of energy, 11 linear transformation, 27
conservative force, 51, 55, 60
critical point, 81 momentum conservation, 72
curve, 25, 27
N-body problem, 65
cutting cone, 25, 27
non-inertial frame, 73
degree 2, 27
orbit, 35
Devil’s dumbell, 66
oscillation, 12
ellipsis, 25, 27 oscillator, 16
energy, 39
parabola, 27, 38
equilibrium point, 15
pendulum, 11
equilibrium position, 12
perspective, 25
Euler-Lagrange equations, 58
plasma, 81
focal point, 25 polar coordinates, 35
frame, 72 potential, 57
free fall, 37–39 potential energy, 11, 39, 41
friction, 55
restricted 3-body problem, 77
generalized momenta, 61 right-hand rule, 43
rigid body, 71
Hamilton equations, 62 rigid object, 68
119
120 INDEX

satellites, 77
simple harmonic oscillator, 16
simple oscillator, 16
simple pendulum, 11
states of matter, 81
stationary integral, 58

total angular momentum, 74

total energy, 11, 39, 41
triple point, 81

Van der Waals equation, 80

vector product, 43

work, 55