Quasi periodi

motions from Hippar hus to Kolmogorov.

Giovanni Gallavotti

Fisi a, Universita di Roma \La Sapienza"

Institut des Hautes Etudes S ientiques
(11 sept 1999)

We shall say, then, that the motion is omposed by

n uniform ir ular motions if it is quasi periodi in the

The evolution of the on eption of motion as omposed by

ir ular uniform motions is analyzed, stressing its ontinuity
from antiquity to our days.1

sense of (1).
In reality in Greek Astronomy it is always lear that
the motion of the solar system, on eived as quasi periodi , is only a possible one within a wider family of
motions that have the form
x(t) = f ('1 + !1 t; : : : ; 'n + !n t):
(2)
Hen e it is in a stronger sense that motions are thought
of as omposed by elementary ir ular ones. Indeed all
the n-ples ('1 ; : : : ; 'n ) of phases are onsidered as des ribing possible states of the system. This means that
one thinks that the phases ('1 ; : : : ; 'n ) provide a system
of oordinates for the possible states of the system. The
observed motion is one that orresponds, onventionally,
to the initial state with phases '1 = '2 = : : : = 0; but
also the other states with arbitrary phases are possible
and are realized in orresponden e of dierent given initial onditions and, furthermore, we an get lose to them
by waiting long enough.
Summarizing: to say that the motions of a system are
omposed by n ir ular uniform motions, of angular velo ities !1; : : : ; !n is equivalent to say that it is possible
to nd a system of oordinates ompletely des ribing the
states of the system (relevant for the dynami al problem
under study) whi h an be hosen to be n angles so that
, furthermore, the motion is simply a uniform ir ular
motion of every angle, with suitable angular velo ities
!1 ; : : : ; !n . This is indeed manifestly equivalent to saying that an arbitrary observable of the system, evolving
in time, admits a representation of the type (2).
In Greek physi s no methods were available (that we
know of) for determining the angle oordinates in terms
of whi h the motion would appear ir ular uniform, i.e.
no methods were available for the omputation of the
oordinates 'i and of the fun tions f , in terms of oordinates with dire t physi al meaning (e.g. polar or Cartesian oordinates of the several physi al point masses of
the system). Hen e Greek astronomy did onsist in the
hypothesis that all the motions ould have the form (2)
and in deriving, then, by experimental observations the
fun tions f and the velo ities !i well suited to the des ription of the planets and stars motions, with a pre ision that, even to our eyes (used to the s reens of digital
omputers), appears marvelous and almost in redible.
After Newton and the development of innitesimal al ulus it has be ome natural and ustomary to imagine
dynami al problems as developing starting from initial

ar hived: mp ar #99-244, hao-dyn #9907004

I. HIPPARCHUS AND PTOLEMY

Contemporary resear h on the problem of haoti motions in dynami al systems nds its roots in the Aristotelian idea, often presented as kind of funny in high
s hool, that motions an always be onsidered as omposed by ir ular uniform motions,3;4.
The reason of this on eption is the perfe tion and simpli ity of su h motions (of whi h the uniform re tilinear
motion ase must be thought as a limit ase).
The idea is far more an ient than Hippar hus (from
Ni ea, 194-120 a.C.) from whom, for simpli ity of exposition it is onvenient to start. The rst step is to understand learly what the Greeks really meant for motion
omposed by ir ular uniform motions. This indeed is by
no means a vague and qualitative notion, and in Greek
s ien e it a quired a very pre ise and quantitative meaning that was summarized in all its surprising rigor and
power in the Almagest of Ptolemy (100-175 d.C.).5{7
We thus dene the motion omposed by n uniform ir ular motions with angular velo ities !1 ; : : : ; !n that is,
impli itly, in use in the Almagest, but following the terminology of ontemporary mathemati s.
A motion is said quasi periodi if every oordinate of
any point of the system, observed as time t varies, an
be represented as:
x(t) = f (!1 t; : : : ; !nt)
(1)
where f ('1 ; : : : ; 'n ) is a multiperiodi fun tion of n angles, with periods 2 and !1 ; : : : ; !n are n angular velo ities that are \rationally independent";8 they were alled
the [velo ities of the \motors" of the Heavens.
We must think of su h fun tion f as a fun tion of
the positions '1 ; : : : ; 'n , (\phases" or \anomalies"), of
n points on n ir les of radius 1 and, hen e, that the
state of the system is determined by the values of the
n angles. Therefore to say that an observable x evolves
as in (1) is equivalent to say that the motion of the system simply orresponds to uniform ir ular motions of
the points that, varying on n ir les, represent the state
of the system.
1

that is, in a way, too \simple" to allow us to appre iate

the dieren es between the Ptolemai and Coperni an
theories) is provided by the theory of the Moon.
As rst example I onsider Hippar hus' theory of the
Moon.
In general motions of heavenly bodies appear, in a rst
approximation, as uniform ir ular motions around the
enter of the Earth. Or in average the position of the
heavenly body an be dedu ed by imagining it in uniform
motion on a ir le, (\the oblique ir le that arries the
planets along", Dante, Par., X), with enter on the Earth
and rigidly atta hed to the sphere (\Sky") of the Fixed
Stars, that in turn rotates uniformly around the Earth.
This average motion was alled deferent motion: but
the heavenly body almost never o upies the average position, rather it is slightly away from it sometimes overtaking it and sometimes lagging behind.

onditions that an be quite dierent from those of immediate interest in every parti ular problem. For example it
is ommon to imagine solar systems in whi h the radius of
the orbits of Jupiter is double of what it a tually is, or in
whi h the Moon is at a distan e from the Earth dierent
from the observed one, et .. Situations of this kind an be
in luded in the Greek s heme simply by imagining that
the oordinates '1 ; : : : ; 'n are not a omplete system of
oordinates, and other oordinates are needed to des ribe
the motions of the same planets if they are supposed to
have begun their motion in situations radi ally dierent
from those whi h, soon or later, they would rea h when
starting at the given present states (and whi h \just"
orrespond to states with arbitrary values of the phase
oordinates '1 ; : : : ; 'n ).
To get a omplete des ription, of su h \other possible
motions" of the system, other oordinates A1 ; : : : ; Am are
ne essary: they are, however, onstant in time on every
motion and hen e they only serve to spe ify to whi h
family of motions the onsidered one belongs. Obviously
we shall have to think that the !1 ; : : : ; !n themselves are
fun tions of the Ai and, in fa t, it would be onvenient
to take the !i themselves as part of the oordinates Ai ,
parti ularly when one an show that m = n and that the
!i an be independent oordinates.
Let us imagine, therefore, that the more general motion has the form:
x(t) = f (A1 ; : : : ; Am ; !1 t + '1 ; : : : ; !n t + 'n ) (3)
where !1 ; : : : ; !n are fun tions of A1 ; : : : ; Am and the oordinates A1 ; : : : ; Am ; '1 ; : : : ; 'n are a omplete system
of oordinates.
In Greek astronomy there is no mention of a relation between m and n: probably only be ause no
mention is made of the oordinates A1 ; : : : ; Am sin e
the Greeks depended ex lusively on a tual observations
hen e they ould not on eive studying motions in whi h
the A1 ; : : : ; Am (e.g. the radii of the orbits of the planets, the in linations of the orbits, et .) were dierent
from the observed values.
In this respe t it is important to remark that Newtonian me hani s shows that it must be m = n = 3N =
fnumber of degrees of freedom of the systemg, if N is
the number of bodies, even though in general it an happen that the !1 ; : : : ; !n annot be taken as oordinates
in pla e of the Ai 's be ause they are not always independent of ea h other (for instan e the Newtonian theory
of the two body problem gives that the three !i are all
rational multiples of one of them, as otherwise the motion would not be periodi ). Nevertheless this identity
between m and n has to be onsidered one among the
great su esses of Newtonian me hani s.
Returning to Greek astronomy it is useful to give some
example of how one on retely pro eeded to the determination of '1 ; : : : ; 'n , of !1 ; : : : ; !n and of f .
A good example (other than the motion of the Fixed
Stars, that is too \trivial", and the motion of the Sun

(h1)



C0

Fig. (h1): Hippar hus' Moon theory: deferent and epi y le.

In the ase of the motion in longitude of the Moon (i.e.

of the proje tion of the lunar motion on the plane of the
e lipti ) the simplest representation of these os illations
with respe t to the average motion is by means of two
ir ular motions, one on a ir le of radius CO = R, deferent, with velo ity !0 and another on a small ir le of
radius CD = r, epi y le, with angular velo ity !1 .
Motion is observed in a rotating frame so that the \average Sun" appears xed (this is a Sun whi h moves exa tly on a ir le with uniform motion and period one
solar day; we ould use instead a onjun tion between
the Moon and a xed Star, with obvious hanges).
We re kon the angles from a onjun tion between the
Moon and the average Sun, i.e. when OC0 proje ted on
the e lipti plane points at the position of the average
Sun.
The enter of the epi y le rotates on the deferent with
angular velo ity !0 and the Moon L rotates on the epi y le with velo ity !1 ,4
By using the omplex numbers notation to denote a
ve tor in the e lipti plane and beginning to ount angles
from the position of apogee on the epi y le we see that
the ve tor z that indi ates the longitudinal position of
the Moon is
n
 = !0 t
i!0 t
i(!1 !0 )t
(4)
= !1t ) z = Re + re
where  = !0 t is the angle C0 OC , = !1 t is the angle

referen e x{axis on a omplex z {plane) is still  = !0 t,

but the angle = !1 t that determines the position L
of the Moon on the epi y le is now re koned from D1 .
In formulae, if R0 = C1 T; R~ = C1 F1 = R s and # is
the angle between D1 F1 and D0 F0 , one nds:
 = !0 t;
~ i# = R0 ei!0 t se i!0 t
Re
R~ = jR0 ei!0 t se i!0 t j;
R0  R0 (!0 t; R; s) =q
= s os 2!0t + R~ (1 (s=R~ )2 sin2 2!0 t)
z = R0 (!0 t; R; s)ei!0 t + re i(!1 #)t
(5)
whi h redu es to Hippar hus' moon theory if s = 0. It
also gives the same result in onjun tion and in opposition (i.e. when  = 0; ); it gives a loser Moon at
quadratures (i.e. when  = 12 ; 32 ).
This representation reveals itself su ient for the omputation of the ephemerides also in quadrature positions,
but it is insu ient (although o by little) for the omputation of the ephemerides in o tagonal positions (i.e.
at 45o from the axes).
Note that, rightly so, no new periods are introdu ed:
the motion has still two basi frequen ies and (5) only
has more Fourier harmoni s with respe t to (4).
The theory was therefore further rened by Ptolemy
himself,4;5 , who supposed that, in the pre eding representation, the omputation of the angle = !1t on the
epi y le should not be performed by starting from the
axis C1 F1 , as in the previous ase, but rather from the
axis F "H of the gure:

DCL and 2=!1 is the time T that elapses between two

su essive returns of the Moon in apogee position on her
epi y le (sin e !0  !1 the new apogee will happen at a
time T for whi h there exists a small angle su h that:
!0 T = 2 + and (!0 !1 )T = ) !1T 2 = 0
) T = 2=!1 = anomaly month).
In modern language we say that the Moon, having
three degrees of freedom, shall have a motion with respe t to the Earth (assumed on a ir ular orbit) endowed
with 3 periods: the month of anomaly (i.e. return to
the apogee and \true period of revolution") of approximately 27d, the period of rotation of the apogee of approximately 9y (in a dire tion on ording with that of
the revolution) and the period of pre ession (retrograde)
of the node between the lunar world and the e lipti of
approximately 18:7y . This last period, obviously, does
not on ern the motion in longitude whi h, therefore, is
hara terized pre isely by two fundamental periods: for
instan e the month of anomaly and the sidereal month
(return to the same xed star: note that the dieren e
between the the two angular velo ities !0 !1 is obviously the velo ity of pre ession of the apogee).
Hippar hus' theory of the motion in longitude of the
Moon yields, as we see, a quasi periodi motion with one
deferent, one epi y le and two frequen ies (or \motors").
It reveals itself su ient (if ombined with the theory
of the motion in latitude, that we do not dis uss here)
for the theory of the e lipses, but it provides us with
ephemerides (somewhat) in orre t when the Moon is in
position of quadrature.
Ptolemy develops a more rened theory of this motion
in longitude,4;5 , see Fig. (t1) below:

S0
D0

S0
D0

C0

C0
D1

D1

(t1)

C1

R0

F0

F0
 

(t2)

R s

R s

F1 =se i

 

F1
s

Fig. (t1): Ptolemy's orre tion to Hippar hus' Moon theory.

F 00

Fig. (t2): The more rened theory of Ptolemy

In formulae, with R0 as in (5)

Again assume that the angles in longitude are re koned

from the mean Sun S0 starting at a onjun tion, as in the
previous theory. In a rst version he imagines that the
enter of the epi y le moves at the extremity of a segment
of length R s that, however, does not have origin on
the Earth T but in a point F1 that moves with (angular)
velo ity !0 on a small ir le of radius s entered on
T ; we suppose that the enter of the epi y le is C1 so
that the angle between C1 T and the axis of apogee (our

R0 e
z = R0 ei!0 t + r 0
jR e

+ sei!0 t e i!1 t
(6)
i!0 t + sei!0 t j
It is lear that with orre tions of this type it is possible to obtain very general quasi periodi fun tions. Note
that the above theory oin ides with the pre eding one
3

i!0 t

at onjun tion, opposition and quadratures and it is otherwise somewhat dierent (in parti ular at the o tagonal
positions).
The values that Ptolemy nds for R; r; s, so that the
theoreti al ephemerides onform with the experimental
ones, are however su h that the possible variations of the
Earth{Moon distan e (between R r s and R + r + s)
are very important and in ompatible with a not observed
orresponding variation of the apparent diameter of the
heavenly body :t is not known why the apparent diameter of the Moon did not seem to worry Ptolemy. Astronomi al distan es (as opposed to elestial longitudes
and latitudes of planets) were not, however, really measured in Greek times (due to the di ulty of parallax
measurements): but we shall see that in Kepler's theory
the measurability of their value payed a major role.

of Ptolemy's lunar theories are still interpretable as motions of deferents and epi y les. Whi h is not ompletely
obvious sin e some of the axes of referen e of Ptolemy
do not move of uniform ir ular motion, to an extent
that by several a ounts, still today, Ptolemy is \a used" of having abandoned the purity of the ir ular
uniform motions with the utilitarian s ope of obtaining
agreement between the experimental data and their theoreti al representations10.
I just quote here Coperni us' Commentariolus, few
lines before the statement of his famous se ond postulate
setting the Earth away from the enter of the World:
\Nevertheless, what Ptolemy and several others legated
to us about su h questions, although mathemati ally a eptable, did not seem not to give rise to doubts and dif ulties" ... \So that su h an explanation did not seem
su iently omplete nor su iently onform to a rational riterion" ... \Having realized this, I often meditated
whether, by han e, it would be possible to nd a more rational system of ir les with whi h it would be possible to
explain every apparent diversity; ir les, of ourse, moved
on themselves with a uniform motion", see11 p.108.

II. COPERNICUS

The skies are painted with unnumber'd

sparks,
They are all re, and every one doth shine;
But there's but one in all doth hold his
pla e,38

Therefore let us he k what was, in some form, probably so obvious to Ptolemy that he did not seem to
feel the ne essity of justifying his alleged deviation from
the \dogma" of de omposability into uniform motions.
Namely we he k that also the motions of the Ptolemai
lunar theories, as a tually all quasi periodi motions, an
be interpreted in terms of epi y les.
Consider for simpli ity the ase of quasi periodi motions with two frequen ies !1; !2 . Then the position will
be

Coperni us (1473-1543)9 (who was, indeed, very worried by the latter problem) tried to nd a remedy by
introdu ing a se ondary epi y le: his model goes ba k
to that of Hippar hus, \improved" by imagining that the
point of the epi y le on whi h Hippar hus set the Moon
was instead the enter of a smaller se ondary epi y le,
of radius s, on whi h, the Moon journeyed with angular
velo ity 2!0
F


S0

z (t) =
(

 = !0 t
 = !1t; CF = r ( 1)
= 2!0t; F L = s

1 ;2

1 2 ei(!1 1 +!2 2 )t 

X
j

j ei
j t

(8)

by the Fourier series theorem, where i are arbitrary integers and j , in the se ond sum, denotes a pair 1 ; 2 and

j  !1 1 + !22 . Imagine, for simpli ity, also that the

enumeration with the label j of the pairs 1 ; 2 ould be
made, and is made, so that 1 >> 2  3 > : : :.
Then r(t) an be rewritten as

1;1
X



r(t) = 1 ei
1 t 1 + 2 ei(
2
1 )t

1



1 + 3 ei(
3
2 )t (1 + : : :)
2

Fig. ( 1): Coperni us' Moon theory with two epi y les

and in formulae:
rL = Rei!0 t + e (!1 !0 )t (r + se2i!0 t )
(7)
This gives a theory of the longitudes of the Moon essentially as pre ise as that of Ptolemy. Note that, again, the
same two independent angular velo ities are su ient.
Before attempting a omparison between the method
of Ptolemy and that of Coperni us it is good to larify the modern interpretation of the notions of deferent and epi y le and to larify, also, that the motions

(9)

whi h, negle ting 2 ; 3 : : :, is the uniform ir ular motion on the deferent of radius j1 j with angular velo ity

1 ; negle ting only 3 ; 4 ; : : : it is a motion with a deferent of radius j1 j rotating at velo ity
1 on whi h rests
an epi y le of radius j2 j on whi h the planet rotates at
velo ity
2
1 ; negle ting only j ; j  4 one obtains a
motion with one deferent and two epi y les, as that used
by Coperni us in the above lunar model.
4

If j1 j is not substantially larger than the other radii

(and pre isely if j1 j not is mu h larger than the sum of
the other jj j, a situation that is not met in an ient astronomy), what said remains true ex ept that the notion
of deferent is no longer meaningful. Or, in other words,
the distin tion between main ir ular motion and epi y les is no longer so lear from a physi al and geometri al
viewpoint. The epi y le with radius mu h larger than
the sum of the radii of the other epi y les, if existent,
essentially determines the average motion and is given
the privileged name of \deferent". In the other ases,
although the average motion still makes sense,12 , it is
no longer asso iated with a parti ular epi y le, but all
of them on ur to dene it, for an example see [AA68,
p.138.
We see, therefore, the omplete equivalen e between
the representation of the quasi periodi motions by means
of a Fourier transform and that in terms of epi y les,10 .
Greek astronomy, thus, onsisted in the sear h of the
Fourier oe ients of the quasi periodi motions of the
heavenly bodies, geometri ally represented by means of
uniform motions.
But Ptolemy's method is in a ertain sense not systemati (see, however, below and17 ): the intri ate interplay
of rotating sti ks that explains, or better parameterizes,
the motion of the Moon is very lever and pre ise but it
seems quite learly not apt for obvious extensions to the
ases of other planets and heavenly bodies.
Coperni us' idea, instead, of introdu ing epi y les of
epi y les, as many as needed to an a urate representation of the motion, is systemati and, as seen above,
oin ides with the omputation of the Fourier transform
of the oordinates with oe ients ordered by de reasing
absolute value. Coperni us' work (with the only ex eption, and su h only in a rather restri ted sense that it is
not possible to dis uss here, of some details of the motion
of Mer ury) is stri tly oherent with this prin iple. set
in his early proje t quoted above.
This is perhaps16 the great innovation of Coperni us
and not, ertainly, the one he is always redited for, i.e.
having referred the motions to the (average) Sun rather
than to the Earth: that is a trivial hange of oordinates,
known as possible and already studied in antiquity,14;15 ,
by Aristar hus (of Samos, 310-235 a.C.), Ptolemy et .,
but set aside by Ptolemy for obvious reasons of onvenien e, be ause in the end it is from Earth that we observe the heavens (so that still today many ephemerides
are referred to the Earth and not to an improbable observer on the Sun), and also be ause he seemed to la k an
understanding of the prin iple of inertia (as we would say
in modern language). See the Almagest, p.45 where allegedly Ptolemy says: \...although there is perhaps noth-

re tly series whi h ontain innitely many Fourier oe ients (see R0 (!0 t) in (5) where this happens be ause of
the square root), i.e. innitely many epi y les, most of
whi h are obviously very small and hen e irrelevant.
We an therefore obtain the same results with several
arrangements of sti ks, provided that the motion that
results has Fourier oe ients, I mean those whi h are
not negligible, equal or lose to those of the motion that
one wants to represent: it is this absen e of uniqueness
that makes the Ptolemai method appear unsystemati .
It has, however, the advantage that, if applied by an
astronomer like Ptolemy, it apparently requires, at equal
approximation, less elementary uniform ir ular motions
(whi h, on the ontrary, very often in the Ptolemai onstru tions, are innitely many as we see in the ase of
(5),(6)): this has been erroneously interpreted as meaning less epi y les than usually ne essary with the methods of Coperni us: a fa t that was and still is onsidered
a grave defe t of the Coperni an theory ompared to the
Ptolemai . Ptolemy identies 43 fundamental uniform
ir ular motions (that ombine to give rise to quasi periodi fun tions endowed with innitely many harmoni s
formed with the 43 fundamental frequen ies) to explain
the whole system of the World: young Coperni us hopes
initially (in the Commentariolus) to be able to explain
everything with 34 harmoni s, only to nd out in the De
Revolutionibus, at the end of a lifetime work, that he is
for ed to introdu e several more. See Neugebauer in5 ,
vol. 2, p.925- 926.18
One should not, however, miss stressing also that
Coperni us helio entri assumption made possible a simple and unambiguous omputation of the planetary
distan es.16 If the Sun is assumed as the enter, and the
orbits are supposed ir ular (to make this remark simplest) then the radii of the epi y les of the external planets (for instan e) are automati ally xed to be all equal
to the distan e Earth{Sun. Then, knowing the periods of
revolution and observing one opposition (to the Sun) of
a planet and one position o onjun tion at a later time,
one easily dedu es the distan e of the planet to the Earth
and to the Sun, in units of the Earth{Sun distan e. In
a geo entri system the radii of the epi y les are simply
related to their deferents sizes and the latter are a priori
unrelated to the Sun{Earth distan e: also for this reason (although mainly be ause of the di ulty of parallax
measurements) in an ient astronomy the size of the planetary distan es was a big open problem. One an \save
the phenomena" by arbitrarily s aling deferent and epi y les radii independently for ea h planet! The possibility
of reliably measuring distan es, applied by Coperni us
and then by Ty ho and Kepler, was essential to establishing the helio entri system and to Kepler, who ould
thus see that the saving of the phenomena in longitudinal
observation was not the same as saving them in the radial observations, a more di ult but very illuminating
task, see23 .

ing in the elestial phenomena whi h would ount against

that hypothesis [that the Sun is the enter of the World...
one an see that su h a notion is quite ridi ulous.17

Ptolemy, with lever and auda ious geometri onstru tions does not ompute oe ient after oe ient
the rst few terms of a Fourier transform. He sees di5

In reality also the ritique of la k of a systemati

method in Ptolemy, the starting point of the Coperni an
theory, should be re onsidered and subje t to s rutiny:
indeed we do not know the theoreti al foundations on
whi h Ptolemy based the Almagest nor through whi h
dedu tions he arrived at the idea of the equant and to
other marvelous devi es. One an even dare the hypothesis that the Almagest was just a volume of ommented
tables based on prin iples so well known to not even deserve being mentioned. It is di ult to imagine that
Ptolemy had pro eeded in an absolutely empiri al manner in the invention of anomalous obje ts like \equant
points" and strange epi y les (like those he uses in the
theory of the Moon, see above) and yet he did not feel
that he was departing from the main stream based on
the axiom that all motions were de omposable into uniform ir ular motions: it is attra tive, instead, to think
that he did not feel, by any means, to have violated the
aristotelian law of the omposition of motions by ir ular
uniform ones.
One should note that if a s ientist of the stature of
Coperni us in a 1000 years from now, after mankind re overed from some great disaster, found a opy the Ameri an Astronomi al Almana 19 (possibly translated from
translations into some new languages) he would be astonished by the amount of details, and by the data orre tness, des ribed there and he would be left wandering
how all that had been ompiled: be ause it is very dif ult, if not impossible to derive even the Kepler's laws,
dire tly from it (not to mention the present knowledge on
the three body problem). And he would say \surely there
must be a simpler way to represent the motions of the
planets, stars and galaxies", and the whole pro ess might
start anew, only to end his life (as Coperni us in \De
revolutionibus")11 with new tables that oin ided with
an appropriately updated version of the ones he found
in his youth. The Ameri an Astronomi al Almana an
be perhaps better ompared to Ptolemy's Planetary Hypotheses if the latter is really due to him, as universally
a epted, while the Almagest is an earlier but more detailed version of it.
After the dis overy of the Kepler laws the theory of
gravitation of Newton (1642-1727) was soon rea hed,24.
Contrary to what at times is said, far from marking the
end of the grandiose Greek on eption of motion as omposed by ir ular uniform motions, Newtonian me hani s
has been, instead, its most brilliant onrmation.
For example, if # denotes the angle between the major semiaxis and the a tual position on the orbit (\true
anomaly"), ` denotes the average anomaly, a is the major
semiaxis of the ellipse and e is its e entri ity, the Keplerian motion of the Mars around the Sun is des ribed by
the equations:
z = pei# (1 e os #) 1 ; p = a(1 e2 )
5
# = ` 2e sin ` + e2 sin 2` + O(e3 ); ` = !t
(10)
4

III. KEPLER

Yet in the number I do know but one

That unassailable holds on his rank,
Unshak'd of motion,38

Today we would say that Ptolemy's theory was nonperturbative be ause it immediately represented the motions

as quasi periodi fun tions (with innitely many Fourier

oe ients). Coperni us' is, instead, perturbative and it
systemati ally generates representations of the motions
by means of developments with a nite number of harmoni s onstru ted by adding new pure harmoni s, one
after the other, with the purpose of improving the agreement with experien e. The larger number of harmoni s
in Coperni us is simply explained be ause, from his point
of view, harmoni s multiple of others ount as dierent
epi y les, while in Ptolemy the geometri onstru tions
asso iated with an epi y le sometimes introdu e also harmoni s that are multiples, or ombinations with integer
oe ients, of others already existent and produ e an
\apparent" saving of epi y les.
But the systemati nature of the Coperni an method
permitted to his su essors to organize the large amount
of new data of the astronomers of the Renaissan e and
of the Reform time. Eventually it allowed Kepler (15711630) to re ognize that what was being painfully onstru ted, oe ient after oe ient, was in the simplest
ases just the Fourier series of a motion that developed
on an ellipse with the Sun, or the Earth in the ase of
the Moon, in the fo us and with onstant area velo ity.
For reasons that es ape me History of S ien e often
redits Kepler for making possible the reje tion of the
s heme of representation of the Heavens in terms of deferents and epi y les, in favor of motions on ellipses.
But it is instead lear that Keplerian motions are still
interpretable in terms of epi y les whose amplitudes and
positions are omputed with the Coperni an or Ptolemai methods (that he regarded as equivalent in a sense
that reminds us of the modern theories of \equivalent
ensembles" in statisti al me hani s, see20 Ch. 1-4) or,
equivalently, via the modern Fourier transform. Nor it
should appear as making a dieren e that the epi y les
are, stri tly speaking, innitely many, (even though all
ex ept a small number have amplitudes, i.e. radii, whi h
are ompletely negligible): already the Ptolemai motions, with the their auda ious onstru tions based on rotating sti ks did require, to be representable by epi y les
(i.e. by Fourier series), innitely many oe ients (or
harmoni s), see (5),(6) above, in whi h the r.h.s. manifestly have innitely many nonvanishing ones.
Only Coperni an astronomy was built to have a nite
number of epi y les: but their number had to be ever
in reasing with the in rease of the pre ision of the approximations. Ptolemy seemed to be looking and Kepler
ertainly was looking for exa t theories, Coperni us appears to our eyes doomed to look for better and better
approximations.

hen e

z = p(1 e2 )1=2 ei# (1 + 2

1
X

n=1
1
2
1
=
2
1
e ) )e = e + O(e3 )

(e)  (1 (1

with respe t to the Fixed Stars Sky and it is the theory

that one nds in the Almagest, after onverting it to the
inertial frame of referen e xed with the Sun.
The motion of the Earth around the Sun (or vi eversa
if one prefers) is similar ex ept that the enter of the deferent ir le is dire tly the equant point, see4 p.192, see
also20 Ch.2-4: this is usually quoted by saying the \for
the Earth Ptolemy (Coperni us and Ty ho) did not bise t
the e entri ity", meaning that the enter and the equant
were identi al and both 2 e a away from the Sun: from23
we dedu e that this did not matter for the Earth whi h
has a mu h smaller e entri ity (than Mars). Before dis overing the ellipse Kepler had to redress this \anomaly"
and he indeed bise ted also the Earth e entri ity, see23 ,
making the Coperni an Earth lose one more distinguishing feature with respe t to the other planets.22
The above, however, is not the path followed by Kepler, see23 where the latter is dis ussed in some detail.
Thus bringing the development in e to rst order one
rea hes a level of approximation qualitatively satisfa tory
for the observations to whi h Kepler's and Ty ho's prede essors had a ess, not only for the Sun but also for the
more anomalous planets like Mer ury, Moon and Mars:
to se ond order however the equant be omes insu ient
and Kepler realized that the ellipse had to be des ribed
at onstant area velo ity with respe t to the fo us.
We an say that the experimental data agree within a

(e)n os n#)

(11)

and to rst order in e:

z = aei!t (1 2e sin !t)(1 + 2e os t) + O(e2 ) =
= aei!t (1 + e(1 + i)ei!t + e(1 i)e i!t + O(e2 )) (12)
whi h an be des ribed to lowest order in e, as omposed
by a deferent and two epi y les. Two more would be
ne essary to obtain an error of O(e3 ).
In this respe t it is interesting to observe how one an
arrive to an ellipse with fo us on the Sun, by onsidering epi y li al motions. Indeed the simplest epi y li al
motion is perhaps that in whi h one onsiders innitely
many pairs of epi y les, run with respe tive angular velo ity n!, with n = 1; 2; : : :, and with radii de reasing
in geometri progression, i.e. :
1
X
0
i#
z (#) = p e
0n os n#
n=0

(13)

for some p0 ; 0 , that leads to the ellipse in the rst of the

(11).
What is less natural in the Kepler laws, is that the
time law whi h gives the motion on the ellipse, instead
of # ! !t, is rather # ! !t 2e sin !t + : : :. Su h
motion is however an old \Ptolemai knowledge" being,
at least at lowest order in e, a uniform angular motion
around a point Sequant of abs issa 2ea from the point S
with respe t to whi h the anomaly # is evaluated and of
abs issa ea with respe t to the enter C of the ir le on
whi h the (manifestly nonuniform) motion takes pla e

third order error in the e entri ity with the hypothesis of

an ellipti al motion and with a time law based on the area
law: this, within a se ond order error in the e entri ity,
oin ides with the Ptolemai law of the equant.
IV. MODERN TIMES

To realize better the originality of the Newtonian theory we must observe that in the approximations in whi h
Kepler worked it was evident that the laws of Kepler were
not absolutely valid: the pre ession of the lunar node, of
the lunar perigee and of the Earth itself did require, to
be explained, new epi y les: in a ertain sense the Keplerian ellipses be ame \deferent" motions that, if run with
the law of the areas, did permit us to avoid the use of
equants and of other Ptolemai \tri ks".
A stri t interpretation of Kepler's laws would be manifestly in ontrast with ertain elementary astronomi al
observations unless ombined with suitable onstru tions
of epi y les as Kepler himself realized and applied to the
theory of the Moon,21 . The theory of Newtonian gravitation follows after the abstra tion made by Newton that
Kepler's laws would be rigorously exa t in the situation in
whi h we ould negle t the perturbations due to the other
planets, i.e. if we onsider the \two body problem" and
we reinterpreted in a novel way the Keplerian on eption
that the motion of a planet was mostly due to a for e
exer ized by the Sun and partly to a for e due to itself.
The theory of gravitation not only predi ts that the

M
CM =a; SC =ea

#
S

(e)
ea

`
Sequant

Fig. (e): The equant onstru tion of Ptolemy adapted to a

helio entri theory of Mars; S is the Sun, M is Mars,
enter of the orbit and the equant point is Sequant .

the

this means that the angle ` in the drawing rotates uniformly and
# = ` 2e sin ` + e2 sin 2` + O(e3 )
(14)
Trun ating the series in (13) and (14) to rst order in
the e entri ity we obtain (12) and hen e a des ription
in terms of one deferent, two epi y les and an equant:
it is a des ription quite a urate of the motion of Mars
7

 ians who, with Birkho, Hopf, Siegel in parti ular,

started from Poin are to begin the onstru tion of the
orpus that is today alled the theory of haos.
But only around the middle of the 1950's it has been
possible to understand the paradox onsisting in the di hotomy generated by Poin are:
(1) on the one hand the su esses of lassi al astronomy based on Newtonian me hani s and on the perturbation theory of Lapla e, Lagrange, et . seemed to onrm the validity of the quasi periodi on eption of motions (re all for instan e Lapla e's theory of the World,
or Gauss' \redis overy" of Ceres,35 and the dis overy of
Neptune,32 ).
(2) on the other hand the theorem of Poin are ex luded the onvergen e of the series used in (1).
The fundamental new ontribution ame from Kolmogorov,33;34: he stressed the existen e of two ways of performing perturbation theory,35 . In the rst way, the lassi al one, one xes the initial data and lets them evolve
with the equations of motion. Su h equations, in all appli ations, depend by several small parameters (ratios of
masses, et .) denoted above generi ally by ". And for
" = 0 the equations an be solved exa tly and expli itly, be ause they redu e to a Newtonian problem of two
bodies or, in not heavenly problems, to other integrable
systems. One then tries to show that the perturbed motion, with " 6= 0, is still quasi periodi , simply by trying to ompute the periodi fun tions f that should represent the motion with the given initial data (and the
orresponding phases 'i , angular velo ities !i , and the
onstants of motion Ai ) by means of power series in ".
Su h series, however, do not onverge or sometimes even
ontain divergent terms, deprived of meaning, see34 Se .
5.10.
A se ond approa h onsists in xing, instead of given
initial data (note that it is in any ase illusory to imagine knowing them exa tly), the angular velo ities (or frequen ies) !1 ; : : : ; !n of the quasi periodi motions that
one wants to nd. Then it is often possible to onstru t
by means of power series in " the fun tions f and the
variables A; ', in terms of whi h one an represent quasi
periodi motions, with the prexed frequen ies.
In other words, and making an example, we ask the
possible question: given the system Sun, Earth, Jupiter
and imagining for simpli ity the Sun xed and Jupiter on
a Keplerian orbit around it, is it or not possible that in
eternity (or also only up to just a few billion years) the
Earth evolves with a period of rotation around to Sun
of about 1 year, of revolution around its axis of about
1 day, of pre ession around the heavenly poles of about
25:500 years, et .?
One shall remark that this se ond type of question
is mu h more similar to the ones that the Greek astronomers asked themselves when trying to dedu e from
the periods of the several motions that animated a heavenly body the equations of the orresponding quasi peri-

motions of the heavenly bodies are quasi periodi , apparently even in the approximation in whi h one does not
negle t the re ipro al intera tions between the planets,
but it gives us the algorithms for omputing the fun tions
f ('1 ; : : : ; 'n ).
The summa of Lapla e (1749-1827) on the Me anique
eleste of 1799,25, makes us see how the des ription of the
solar system motions, even after taking into a ount the
intera tions between the planets, ould be made in terms
quasi periodi fun tions. Newtonian me hani s allows
us to ompute approximately the 3N oordinates A =
(A1 ; : : : ; A3N ) and the 3N angles '1 ; : : : ; '3N and the 3N
angular velo ities !1 (A); : : : ; !3N (A) in terms of whi h
the motion simply onsists of 3N uniform rotations of
the 3N angles while the A remain onstant.
Lapla e makes us see that there is an algorithm that
allows us to ompute the Ai ; !i ; 'i by su essive approximations in a series of powers in several parameters (ratios of masses of heavenly bodies, e entri ities, ratios of
the planets radii to their orbits radii et .), that will be
denoted here with the only symbol ", for simpli ity.
After Lapla e approximately 80 years elapse during
whi h the te hnique and the algorithms for the onstru tion of the heavenly series are developed and rened leading to the onstru tion of the formal stru ture of analyti
me hani s. Then Poin are was able to see learly the new
phenomenon that marks the rst true and denitive blow
to the Greek on eption of motion: with a simple proof,
elebrated but somehow little known, he showed that the
algorithms that had obtained so many su esses in the
astronomy of the 1800's were in general non onvergent
algorithms,26.
Few did realize the depth and the revolutionary hara ter of Poin are's dis overy: among them Fermi who
tried of dedu e with the methods of Poin are a proof of
the ergodi ity of the motions of Hamiltonian dynami al
systems that are \not too spe ial". The proof of Fermi,
very instru tive and witty although stri tly speaking not
on lusive from a physi al viewpoint, remained one of the
few attempts made in the rst sixty years of the 1900's
by theoreti al physi ists, to understand the importan e
of of Poin are's theorem.
Fermi himself, at the end of his history, ame ba k on
the subje t to rea h on lusions very dierent from the
ones of his youth (with the elebrated numeri al experiment of Fermi-Pasta-Ulam).28
And the Greek on eption of motion nds one of its last
(quite improbable, a priori) \advo ates" in Landau that,
still in the 1950's, proposes it as a base of his theory of
turbulen e in uids.29 His on eption has been riti ized
by Ruelle and Takens (and apparently by others)30;31 on
the basis of the ideas that, at the root, went ba k to
Poin are.
The alternative proposed by them began modern resear h on the theory of the development of turbulen e
and the renewed attempts at establishing a theory of developed turbulen e.
The attitude was quite dierent among mathemati8

thinks that a sudden transition, as the intensity " of the

perturbation in reases, is possible from a regime in whi h
diusion times are super astronomi al in orresponden e
of the interesting values of the parameters (i.e. times of
several orders of magnitude larger than the age of the
Universe) to a regime in whi h su h times be ome so
short to be observable on human s ales. This is one of
the entral themes of the present day resear h on the
subje t,36 .

odi motion.
The answer of Kolmogorov is that if !1 ; : : : ; !n are the
n angular velo ities of the motion of whi h we investigate
the existen e it will happen that for the most part of the
hoi es of the !i there a tually exists a quasi periodi
motion with su h frequen ies and its equations an be
onstru ted by means of a power series in ", onvergent
for " small.33
The set of the initial data that generate quasi periodi
motions has a omplement of measure that tends to zero
as " ! 0, in every bounded part of the phase spa e ontained in a region in whi h the unperturbed motions are
already quasi periodi .
One annot say, therefore, whether a preassigned initial datum a tually undergoes a quasi periodi motion,
but one an say that near it there are initial data that
generate quasi periodi motions. And the loser the
smaller " is.
By the theorem of ontinuity of solutions of equations
of motion with respe t to variations of initial data it follows that every motion an be simulated, for a long time
as " ! 0, by a quasi periodi motion.
But obviously there remains the problem:
(1) are there, really, initial data whi h follow motions
that, in the long run, reveal themselves to be not quasi
periodi ?
(2) if yes, is it possible that in the long run the motion
of a system diers substantially from that of the (abundant) quasi periodi motions that develop starting with
initial data near it?
The answer to these questions is armative: in many
systems motions that are not quasi periodi do exist and
be ome easily visible as " in reases. Sin e ele troni omputers be ame easily a essible it is easy for everybody to
observe personally on omputer s reens the very omplex
drawings generated by su h motions (as seen by Poin are,
Birkho, Hopf et ., without using a omputer).
Furthermore quasi periodi motions although being, at
least for " small, very ommon and almost dense in phase
spa e probably do not onstitute an obsta le to the fa t
that the not quasi periodi motions evolve very far, in the
long run, from the points visited by the quasi periodi
motions to whi h the initial data were lose. This is the
phenomenon of Arnold's diusion of whi h there exist
quite a few examples: it is a phenomenon of wide interest.
For example if diusion was possible in the solar system,
then the o urren e of important variations of quantities
su h as the radii of the orbits of the planets would be
on eivable, with obvious (dramati ) onsequen es on the
stability of the solar system.
In this last question the true problem is the evaluation of the time s ale on whi h the diusion in phase
spa e ould be observable. In systems simpler than the
solar system (to whi h, stri tly speaking, Kolmogorov's
theorem does not dire tly apply, for some reasons that
we shall not attempt to analyze here,34 Se . 5.10) one

This is, in part, a translation of the text of a onferen e

at the University of Roma given around 1989, ir ulated in
the form of a preprint sin e. The Italian text, not intended
for publi ation, has ir ulated widely and I still re eive requests of opies. I de ided to translate it into English and
make it available more widely also be ause I nally went
into more detail in the part about Kepler, that I onsidered
quite super ial as presented in the original text. Therefore this preprint diers from the previous, see2 , mainly
(but not only) for the long new part (in footnote23 ) about
Astronomia nova, whi h might be of independent interest.
2
The original Italian text an be found and freely downloaded from http://ippar o.roma1.infn.it, at the page
\ 1994".
3
A general history of astronomy is in: J. Dreyer,: A history
of astronomy, Dover, 1953.
4
For a simple introdu tion to the Ptolemai system see:
Neugebauer, O.: The exa t s ien es in antiquity, Dover,
1969. The gures of the text are taken from this volume:
see p. 193{197.
5
A riti al and ommented version of Ptolemy's theory, both
of the Almagest and of the Planetary Hypothesis, is in:
Neugebauer, O.: A history of an ient mathemati al astronomy, part 2, Springer{Verlag, 1975.
6
A re ent edition of the Almagest, with omment, is:
Ptolemy's Almagest, edited by G. Toomer, Springer Verlag, 1984.
7
Some ritiques to Ptolemy are in: R. Newton: The rime of
Claudius Ptolemy, John's Hopkins Univ. Press, 1979. The
\book"37 also provides us with a very onvin ing dis ussion of the fraud (or \ rime") allegations brought against
Ptolemy.
8
i.e. no linear ombination of them with rational oe ients
an vanish unless all oe ients vanish.
9
Re ent is the volume on Coperni us: O. Neugebauer, N.
Swerdlow: Mathemati al astronomy in Coperni us' de revolutionibus, vol. 1,2, Springer Verlag, 1984.
10
The interpretation that the Fourier transform (and (9))
has in terms of deferent and epi y li al motions has been
noted by many; the more an ient that I ould retrieve is in
a memory of 1874 of G. S hiaparelli, reprinted in G. S hiaparelli, S ritti sulla storia dell' astronomia anti a, part
I, tomo II, Le sfere omo entri he di Eudosso, di Callippo
e di Aristotele, p. 11, Zani helli, Bologna, 1926. It should

time):16 a problem in whi h he had a strong interest as

he dedi ated a book14;15 to his determination of the distan e of the Moon to the Earth, and this should have been
reported by Ptolemy. From the extant information about
Aristar hus' theory there is, strangely, no tra e of an appli ation of the helio entri system to planets other than the
Earth and the Sun, see14 p. 299 and following, although
it would be surprising that there was none. For a riti al
a ount of the Planetary Hypotheses see5 .
18
Note that, from Newtonian me hani s and from the dis ussion below, the motions of the 8 lassi al planets (the Fixed
Stars Sky, Sun, Mer ury, Venus, Moon, Mars, Jupiter,
Saturn not ounting the Earth (whose rigid motions are
des ribed by those of the Fixed Stars), or alternatively
not ounting the Sun and the Fixed Stars but regarding the Earth as having 6 degrees of freedom, requires a
maximum of 24 = 3  8 independent \fundamental" frequen ies namely three for ea h planet: so that both in
Ptolemy and in Coperni us there must be epi y les rotating at speeds multiple of the fundamental frequen ies or
at least at speeds whi h are linear ombinations with integer oe ients of the fundamental frequen ies: hen e the
43 frequen ies annot be rationally independent of ea h
other; see for instan e the gure on Coperni us' Moon and
note that the \number of epi y les" in Ptolemy's theory,
see Fig. (t1), ould be ounted dierently.
19
The astronomi al almana for the year 1989, issued by the
National Almana O e, US government printing o e,
1989.
20
Kepler, J.: Astronomia nova, reprint of the fren h translation by J. Peyroux, Blan hard, Paris, 1979.
21
Stephenson, B.: Kepler's physi al astronomy, Prin eton
University Press, 1994.
22
It is interesting to ompare in detail the theory of Mars of
Coperni us and that of Ptolemy (redu ed to a helio entri
one). The rst has a deferent of radius a on whi h a rst
epi y le of radius 32 ea ounter{rotates at equal speed and,
on it, a se ond epi y le rotates at twi e the speed; the
starting onguration being the rst epi y le at aphelion
and the se ond opposite to the aphelion of the rst. In
other words the position zC from the aphelion is, at average
anomaly ` given by zC

be stressed that the above \redu tion" of a quasi periodi

motion to an epi y li series is not unique and other paths
an be followed: this will be very lear by the examples
below23 .
11
The work of Coperni us and Newton an be easily found
in English as there are plenty of reprints; in Italian I quote
the olle tion printed by UTET dire ted by L. Geymonat:
N. Coperni us, Opere, ed. F. Barone, UTET, Torino, 1979.
We nd here in parti ular the so alled Commentariolus
that presents the plan of the Coperni an work, as optimisti ally viewed by the young (and perhaps still naive)
Coperni us himself before he really onfronted himself with
a work of the dimensions of the Almagest. Great were the
di ulties that he then met, while dedi ating the rest of
his life to a omplete realization of the program sket hed
in the Commentariolus (1530), so that the De revolutionibus presents solutions quite more elaborate than those
programmed in the quoted work. Nevertheless the operni an revolution appears already learly from this brief and
illuminating work: here one nds the passage quoted in the
text (p.108 of the Italian edition).
12
And for a detailed treatment, also of the notion of average
motion, see in parti ular (D. Bo aletti pointed out to me
this bibliographi note, together with the pre eding one10 ):
S. Sternberg, Celestial me hani s, vol. 1, Benjamin, 1969.
13
Arnold, V.I., Avez, A.: Ergodi problems of lassi al me hani s, Benjamin, New York, 1968.
14
It is interesting, not only for the personality of Aristar hus
of Samos, to read the volume: T. Heath: Aristar hus of
Samos. The an ient Coperni us, Dover, 1981.
15
Heath, T.L.: Greek Astronomy, Dover, 1991. This ontains
a very illuminating olle tion of translations of fragments
from greek originals.
16
Neugebauer assesses very lu idly Coperni us' ontribution:
see4 p. 205.
17
See Ptolemy, C.: The Almagest, ed. G.J. Toomer, Springer
Verlag, New York, 1984. See also Theon, Commentaires
de Pappus et de Theon d'Alexandrie sur l'Almageste, T.
II, ed. annotee par A. Rome, Bibliote a Apostoli a Vati ana, Citta del Vati ano, 1936. I often wonder whether it is
possible that this passage has been ontaminated by later
ommentators. Although ertainly not mu h later, be ause
it is already ommented by Theon of Alexandria in the
se ond half of the fourth entury: however two enturies
is a very long time for S ien e (if one thinks to what happened sin e Lapla e). In a way this and the argument that
follows it is mu h too rough ompared to the level of the
rest of the Almagest. Nevertheless if one attributes, as it
seems right to do,5 p. 900, the Planetary Hypotheses books
to Ptolemy, then one is led to think that the passage is indeed original. This is perhaps also proved by the fa t that
Ptolemy does not seem to realize that the helio entri hypothesis would have allowed a lear determination of the
average radii of the orbits, missing in his work. In turn
this makes us wonder whi h exa tly was the famous helio entri hypothesis of Aristar hus and if it went beyond
a mere qualitative hange of oordinates. Had it been the
same as Coperni us' he ould have determined the sizes
of the orbits (at least in prin iple as the ne essary parallax measurements were at the border of feasibility at the

3
1 2i`
ae
eae
=
2
2
= ea + aei` (1 ie sin `)
Ptolemy has the planet on an e entri ir le, entered ea
away from the Sun, whose enter rotates at onstant speed
around the equant point whi h, in turn, is ea further away
from the enter of the orbit. Hen e if  is the e entri
anomaly (i.e. the longitudinal position of the planet on the
orbit as seen from the enter) it is
zC

= aei` +

= ea + aei
and from Fig. (e) we see that the relation between the average anomaly ` (i.e. the longitudinal position of the planet
as seen from the equant point) is related to  by
zT

sin(`
so that

10

)

= e sin `

=`

e sin ` + O(e

faithfully until the end of his work. The main one, for our
purposes, was the (vituperated or, more mildly, simply riti ized) law that the \speed of the motion due to the Sun
is inversely proportional to the distan e to the Sun", see
below.
After as ertaining that the Earth and Mars orbits lie on
planes through the Sun (rather than through the mean
Sun, as in Coperni us and Ty ho) he tried to \imitate the
an ients" by assigning to Mars an orbit, on an e entri
ir le that I will all the deferent, and an equant: he noted
that a very good approximation of the longitudes followed
if one abandoned Ptolemy's theory of the enter C of the
orbit being half way between the Sun S and the equant
E . His equant was set, to save the phenomena, a little
loser to the enter (with respe t to the Ptolemai equant
point) at distan e e0 a from the enter C of the deferent
rather than at distan e ea. If z denotes the position with
respe t to the enter C in the plane of the orbit with x{axis
along the apsidal line of Mars and  denotes the position
with respe t to the Sun S (e entri by ea away from C )
and if  denotes the position on the deferent of the planet,
alled the e entri anomaly, and if ! is the angular velo ity
with respe t to the equant point E this means (using the
omplex numbers notation of Se . 1, Eq. 4 with x{axis
along the apsides line perihelion{aphelion)

1 2 2
3
e sin `) + O(e )
2
and we see that the longitudinal dieren e (i.e. the differen e of the true anomalies or longitudes from the Sun
position S ) is arg zzCT or
zT

= ea + aei` (1

ie sin `


1
e2 sin2 `
= O(e3 sin3 `)
2 e e i` + 1 ie sin `
However the dieren e in distan e is jzC j jzT j of the order
O( 12 e2 sin2 `) so that Coperni us' epi y les are equivalent
to Ptolemy's equant within O(e3 ), or about 40 , in longitude
measurements and within O(e2 ), or about 400 in distan e
measurements (i.e. to mat h distan es at quadrature, say,
one should alter by about 400 the average anomaly, whi h
means to delay the observations by about one day sin e
Mars period is about 2 years (provided the distan es ould
be measured a urately enough, whi h was not the ase at
Ptolemy's time).
Before Ty ho (relative) dieren es of O(e2 ) ould not be
appre iated experimentally: but their existen e was derived from the theories, and used, by Kepler who dis ussed
them at the beginning of his book in Ch.4, where the above
al ulation is performed for ` = 2 , where the dis repan y
is maximal, see20 p. 23 (or p. 16 in the original edition).
Kepler's theory diers from both to order O(e2 ). He rst
derived a better theory for the longitudinal observations
(whi h turned out eventually to agree with the omplete
theory already to O(e3 )) and used it to nd the \ orre t"
theory agreeing within O(e3 ) with the data for the distan e measurements (that had be ome possible, see16 , after
Coperni us).
23
The following a ount of the work on the se ond Kepler's
law attempts at providing a self ontained exposition whi h
annot be regarded as a substitute for the series of papers in37 whi h impressively analyses various aspe ts of
Astronomia Nova,20 , in luding a lear te hni al analysis
of the \vi arious hypothesis" (see37 , p. 311) and an interesting omputer analysis of some of Kepler's al ulations
(see37 , p. 367). The only point on whi h something not already ontained in37 or in21 (where omplementary areful
and detailed analysis of Kepler's dis overies are presented
and from where I derive most of what follows) is perhaps
the analysis of Kepler's 1=r for e law.
A key point to keep in mind in this footnote is that the
resolution of the observations available to Ptolemy (and
Coperni us) was of the order of 100 so that errors were in
the order of tens of primes: this meant that one ould observe rst order orre tions in the e entri ity of Mars but
the se ond order orre tions (of order e2 ' 10 2 or about
300 ) were barely non observable (the ensuing di ulties in
interpreting the data earned Mars the name of inobservabile sidus, unobservable star, after Plinius). However the
observations of Ty ho were of the order of a few primes so
that se ond order orre tions were learly observable, see3
p. 385, be ause the third order amounts to about 30 .
Another major point to keep in mind is, as learly stressed
in21 , that Kepler was the rst to have (perhaps sin e Greek
times) a physi al theory to he k: his language is not the
one we have be ome a ustomed to after Newton, but he
had very pre ise laws in mind whi h he kept following very

arg 1

= aei ;

=`

ar sin e0 sin ` = `

0 sin ` + O(e03 );

` = !t
1
= ea + aei ; j j = a ((" + os  )2 + sin2  ) 2 =
1
= a (1 + e2 + 2e os  ) 2
whi h was alled the vi arious hypothesis, illustrated in Fig.
(k1):


P0

#
Pph


C

(k1)

Pap

Fig. (k1) The vi arious hypothesis: here the e entri ities

are e = 45 0:4 and e0 = 34 0:4,  4 times larger than real to
make a learer pi ture. The dashed ir les are the deferent
( entered at S ) and the epi y le ( entered at P0 ) while the
planet is in P . The ontinuous ir le is the a tual orbit.
The segment SP0 is parallel to CP (partially drawn) so that
P CPap = P0 SPap is the e entri anomaly while P SPap
is the true anomaly # and P EPap is the average anomaly
`, that rotates uniformly around the equant E .

11

the deferent and the epi y le are drawn. The a tual orbit is
a ir le with the horizontal segment as a diameter

The hypothesis illustrated above an be ompared with the

a tual (later) Kepler law
p

Then Kepler was assailed by the suspi ion that the dis repan ies in the distan es that he was nding were rather
due to a defe tive theory of the Earth motion (whi h was
needed to onvert terrestrial observation into solar ones):
he was thus led to realize that the Earth too had an equant
and he ould he k that the dis repan y between distan es
theory and observations was not due to an erroneous theory
of the Earth motion: introdu ing an equant for the Earth
did not ae t sensibly the data for the Sun and Mars (this
meant a large amount of he king, a year or so).
Returning to Mars he tried to he k (again) his basi hypothesis that the velo ity was inversely proportional to the
distan e from the Sun; he assumed that what pro eeded
at velo ity inversely proportional to the distan e from the
Sun was the enter of an epi y le of radius ea; the epi y le
enter had anomaly  on a ir le of radius a around S and,
on it, the planet was rotating at rate _. So that the planet
was a tually moving on a ir le of radius a around the e entri enter C with angular velo ity _ too. The value of
_ was determined by the Kepler original law  _ = onst,
see21 p. 101. The planet equations would be:

= a ( os  + i 1 e2 sin  ); ` =  + e sin ;
3
 = ` e sin ` + O(e ); ` = !t
so we see that the vi arious hypothesis gave an in orre t
distan e Mars{Sun be ause of the dieren e of order O(e
e0 ) + O(e3 ) = O(e2 ), as e e0 is numeri ally ' 5e2 (from
the data for e0 ): for instan e the y { oordinate was about
21
1 2
p. 46, at  = 2 .
2 e higher than it should have, see
Nevertheless the vi arious hypothesis gave very a urate
longitudes. Therefore the hypothesis was used by Kepler
as a qui k means to ompute the longitudes until the very
end of his resear h. The hypothesis was dis arded be ause,
after Coperni us and Ty ho, it had be ome possible (see
Se . II above) to measure distan es from the Sun and they
were in orre tly predi ted by the hypothesis be ause the
se ond order orre tions in the e entri ity were already
visible (in the ase of Mars).
It is interesting to remark that a posteriori it is lear why
the vi arious hypothesis worked so well. The relation between the longitude # orresponding to a theory in whi h
the Sun is e entri by ea from the enter of the orbit and
the equant is e0 a further away gives a longitude # as
z

= j j = a (1 + 2e os  + e2 ) 2
plus the dynami al law _ = onst: this gives both
anomaly and distan e diering from the vi arious hypothesis by O(e2 ): hen e still in ompatible with the observations
(a tually worse than the vi arious hypothesis whi h at least
gave orre tly the longitudes).

1
(e + e0 ) sin ` + (ee0 + e2 ) sin 2` + O(e3 )
2
while the nal theory gives # in terms of the ellipse e entri ity e as
#

=`

5
2e sin ` + e2 sin 2` + O(e3 )
4
therefore a \non bise ted" e entri ity (i.e. e 6= e0 with
0
1
 and e = 45 e; e0 = 34 e) gives an agreement
2 (e + e ) = e
3
to order e ' 10 3 : this is a few primes, well out of observability. As Kepler noted Ty ho's and his observations
gave e + e0 = 2e = 0:18564 and e = 0:11332 very lose
to 45 e = 0; 11602, see21 p. 44. This provides an explanation of why the vi arious hyptothesis is so a urate for the
longitudes.
A realisti drawing of the vi arious hypothesis would be
illustrated by Fig. (k2):
#=`

= aei ;  = ea + aei ;

P0
a

k3


Pph

Pap
S

Fig. (k3): The rst attempt to establish the law _ = onst.
E entri ity (e = 0:4) is mu h larger than the real value,
for illustration purposes. The deferent is a ir le of radius
a around S and the epi y le of radius e a is entered at P0 ,
with anomaly  with respe t to S , the planet is in P

(k2)

The su essful (among others) attempt was driven by

the remark that one needed to lower the y { oordinate at
quadrature by 12 e2 a (i.e. to eliminate a \lunula" between
the vi arious hypothesis orbit and the observed orbit). In
fa t Kepler dis overs, by han e as he reports, that the
observed distan e of Mars from the enter of the deferent
is b, shorter than apand pre isely su h that ab = se # if
sin # = e, or: b = a 1 e2 .
So one sees that the distan e from S is at aphelion or perihelion a (1  e) from S while at quadrature on the def-

Fig. (k2): Same as Fig. (k1) with e entri ity e = 0:1; only

12

erent (i.e. at e entri anomaly  = 2 ; 32 ) it is just a,

or thepdistan e to the enter C is apin the rst ases and
1 e2 , in the se ond
b = a 1 e2 = a  a with  = 1
ases.
\Therefore as if awakening from sleep" it follows that at a
position with e entri anomaly  the planet y oordinate
is lower (in the dire tion orthogonal to the apsides line) by
a  sin  while the x{ oordinate is still a os  , see21pp. 125:
hen e the orbit is an ellipse (with axes a and b = a 1 e2
and the distan e to the Sun is then easily omputed to be
 = a (1 + e os  )). Indeed the oordinates of the planet
with
p respe t to the Sun be ome x = e a + a os ; y =
a 1 e2 p
sin  (rather than the previous y = a sin  ) so
that  = x2 + y 2 = a (1 + e os  ).
Combining this with the basi dynami al law  _ = onst
we dedu e that the motion is over an ellipse run at onstant
area velo ity around S (not C ), as one readily he ks, (see
below).
Kepler's interpretation of the above relation was in the ontext of his attempt at a des ription of the motion \imitating the an ients". The e entri anomaly  denes the
enter P 0 of an epi y le on a deferent ir le entered at
S (not at C ) and with radius ea on whi h the planet
should have traveled an angle  away from the P 0 S axis;
but in fa t the a tual position P was really loser to the
apsides line by the p(very small, yet observable after Ty ho) amount a (1
1 e2 ) os  so that the distan e P S
was  = a (1 + e os  ) (as one readily he ks). The law
 _ = onst was quite natural as he attributed the motion
around the Sun as partly due to the Sun and partly to the
planet: the latter (somewhat obs urely, perhaps) was responsible for the epi y li ex ursion so that  _ would be
the orre t variation of the e entri anomaly whi h had,
therefore, a physi al meaning (this is not the ommon interpretation of Kepler3;21 ).

1
2


1
P

(k4)


#
Pph

Pap

The deferent is a ir le entered at the enter of the orbit C

with radius 12 (a + b): the planet is not on the deferent but
on a (very small) epi y le of diameter a b. If the enter
of the epi y le is at e entri anomaly  the planet on the
epi y le has traveled retrograde by 2 with respe t to the
radius from C to the epi y le enter.
Note that in Fabri ius' onstru tion the distan e of the ellipse to the deferent ir le is very small as the epi y le has
radius O(e2 )a; unlike what happens in Kepler's onstru tion in whi h the deferent is entered at S and therefore the
epi y le is mu h larger having a radius ea. The Fabri ius'
onstru tion, if done with e = 0:1, approximately the Mars
e entri ity, makes us appre iate how subtle and rened
had to be the analysis of Kepler to dete t and understand
the ellipse and the areal law. If one draws on the s ale
of this page the Fabri ius' pi ture and Kepler pi ture at
quadrature, for insan e, one get (of ourse) the same ellipse
but one an see Kepler's epi y le while the Fabri ius' one
is not visible on this s ale. The following gure Fig. (k6)
(representing at quadrature Fabri ius', left, and Kepler's
(right) ellipses, ir ums ribed ir le (drawn but not distinguishable on this s ale) and epi y les (both drawn but only
Kepler's being visible) is a quite eloquent illustration.

ea

Fig. (k5): Fabri ius' interpretation, see3 p. 402-403, of Kepler's ellipse. The ellipse has very high e entri ity (e =
0:8!) to make visible the small epi y le. The planet is in
P and C 3 = b, C 1 = a and the enter of the epi yle is 2.
E entri anomaly re koned at C and the angle 1 2 P is 2

ea os

3


#
Pph

(k5)

Pap

(k6)

Fig. (k4): Kepler's onstru tion of the ellipse. The e entri ity is e = 0:4 to make a learer drawing. The planet is
in P . The segment 23 is the proje tion on S 3 of the segment
31 and the segment SP is equal to S 3. The epi y le (dashed
ir le) is no longer holding the planet but it determines its
position. The anomaly  is the anomaly of the enter of the
epi y le. Motion veries the law  _ = onst.

Fig. (k6): Fabri ius' (left) and Kepler's (right) deferents,

epi y les and ellipses with e entri ity e = 0:1. On the
drawing s ale one neither appre iates the dieren e between
ellipse and the deferent ir les nor the epi y le in the rst
drawing. The solid line and dashed lines (indistinguishable
in the pi ture) are respe tively the ellipse and the deferent

An equivalent formulation is des ribed in Fig. (k5):

13

nova, a hasty on lusion to say the least. It is not even

lear that Kepler himself thought so.
24
Newton, I.: Prin ipia mathemati a, UTET, Torino, 1962.
25
Re ent is the reprint (e onomi ) of the lassi al English
translation of E. Bowdit h: P. de la Pla e: Celestial Me hani s, ed. E. Bowdit h, Chelsea, 1966.
26
Poin are, H.: Les methodes nouvelles de the me anique
eleste, re enly reprinted by Blan hard, Paris, 1987.
27
Fermi, E.: Beweis dass ein me hanis hes normalsysteme im algemeinen quasi{ergodis h ist, Physikalis he Zeits hrift, 24, 261{265, 1923, reprinted in E., Fermi, Colle ted
papers, vol I, p. 79{84, A ademia dei Lin ei and Chi ago
University Press 1962.
28
Fermi, E., Pasta, J., Ulam, S.: Studies of nonlinear problems, Los Alamos report LA-1940, 1955, reprinted in E.,
Fermi, Colle ted papers, vol. II, A ademia dei Lin ei and
Chi ago University Press, 1965.
29
Landau, L., Lif hitz, E.: Me anique des uides, MIR,
Mos a, 1971.
30
Ruelle, D., Takens, F.: On the nature of turbulen e, Communi ations in mathemati al Physi s, 20, 167, 1971.
31
Ruelle, D., Takens, F.: Note on erning our paper "On the
nature of turbulen e", Communi ations in Mathemati al
Physi s, 23, 343{344, 1971.
32
A history of some important appli ations of the perturbation theory to the motion of the main planets, su essive
to the work of Lapla e, is: M. Grosser, The dis overy of
Neptune, Dover 1979.
33
Kolmogorov, A.N.: Preservation of onditionally periodi
movements with small hange in the Hamilton fun tion;
reprinted in: Casati, G., Ford, J.: Sto hasti behavior in
lassi al and quantum systems, Le ture Notes in Physi s,
vol. 93, Springer{Verlag, 1979.
34
For Kolmogorov's theorem see: G. Gallavotti: Elementary
me hani s , Springer Verlag, 1984.
35
A lassi al work on the theory of unperturbed Keplerian
motions is the book of Gauss, useful to whoever wishes
to realize the size of even the simplest astronomi al omputation and thus desires to appre iate the greatness of
the Greeks astronomers' work: K. Gauss, Theory of the
motion of heavenly bodies moving about the Sun in oni
se tions, Dover 1971. See also the appendix Q in the se ond Italian edition of34 , Me ani a elementare, Boringhieri,
Torino, 1985.
36
A olle tion of modern but lassi al papers on the theory
of quasi periodi and haoti motions an be found in: R.
M Kay, J. Meiss, Hamiltonian dynami al systems, Hilger,
1987.
37
Gingeri h, O. The eye of the Heaven, Ptolemy, Coperni us,
Kepler, Ameri an Institute of Physi s, 1993.
38
Among the last few words of C.J. Caesar as reported by
Shakespeare.

In other words the epi y le enter moves on the deferent at

speed _ while the planet moves at speed 2_ on the epi y le. This very Ptolemai interpretation of Kepler's ellipse,
formally pointed out to Kepler by Fabri ius, see3 p. 402403 (but it is simply impossible that Kepler had not known
it immediately), is \ruined" by the very original Kepler law
 _ = onst whi h implies that _ is not onstant.
No matter how auda ious the last on eptual jump, i.e.
= onst, may look it is absolutely right and it identied
the ellipse and the area law, as Kepler proved immediately,
regarding visibly this fa t as a nal proof of his hypothesis
that the area law and the inverse proportionality of the
speed to the distan e were absolutely orre t and in fa t
identi al.
The law  _ = onst, i.e. the area law, is ompletely out
of the Coperni an views and it re alls to mind the mysterious Ptolemai lunar onstru tions for whi h we have
apparently no lue on how they were derived.
That  _ is the area law is worth noting expli itly as it
is little remarked in the elementary dis ussions of the two
body problem. If # denotes the true anomaly, ` the average
anomaly and  the e entri anomaly then the equation of
an ellipse in polar oordinates (; #) with enter at the
attra tion enter S an be written  = p=(1 e os #) =
a (1+e os  ) with a the major semiaxis and p = a (1 e2 ). It
was possibly well known, sin e Apollonius (?), that in su h
oordinates the area spanned by the radius  per unit time
is 21 p  _ or, as Kepler infers from his \physi al on eption"
that \velo ity [on the deferent is inversely proportional to
the distan e from the Sun". This statement that is, often,
interpreted as an error made by Kepler, see3 p.388, possibly
onfusing the speed on the deferent with the speed around
the enter whi h is  #_ : the elementary relation 2 #_ = p  _
may explain why Kepler did not see the two laws in on i t.
Although some omments on this would have helped a lot
the readers, it seems unlikely that he did not noti e that  _
is not proportional to the area velo ity unless the motion
is on an ellipse with e entri anomaly  , parti ularly after
Fabri ius' omments on the Ptolemai version of the ellipse.
And an error on the part of Kepler in measuring the areas
is obviously ex luded from what he writes: see20 p. 248{251
(or p. 193{196 of the original edition).
Other interpretations of  are in ompatible with the area
law to rst order in the e entri ity. Sin e the e entri ity
of Mars is \large" and the measurements of Ty ho{Brahe
allowed us even to see orre tions to the distan es of se ond
order in the e entri ity it was possible to realize that _ was
indeed inversely proportional to  so that ` =  + e sin 
followed. And the natural assumption that all planets (but
the too lose Moon and perhaps the e entri Mer ury)
veried the same laws was easily he ked (by Kepler) to be
fully onsistent with the data known at the time for the
major planets.
We see that although the above Kepler approa h is very
original ompared to Ptolemy's and Coperni us' the on lusion in3 p.393 that \the dis overy of the ellipti orbit
of Mars was an absolutely new departure, as the prin iple
of ir ular motion had been abandoned..." seems, after examining the methods and the ideas followed in Astronomia
 _

I know that I am mortal and the reature of a day; but when I sear h out the
massed wheeling ir les of the stars, my
feet no longer tou h the Earth, but, side
by side with Zeus himself, I take my ll
of ambrosia, the food of Gods: (Ptolemy:

see37 ).

14

Giovanni Gallavotti
Dipartimento di Fisi a
Universita di Roma \La Sapienza"
P.le Moro 2, 00185 Roma, Italy

email:
gallavottiroma1.infn.it
web: http://ippar o.roma1.infn.it
A knowledgments: I am grateful to I. Vardi for pointing out
the referen e37 . Partial nan ial support from IHES is gratefully a knowledged.

REVTEX

15