Integrability of Dynamical Systems through

Di↵erential Galois theory: a practical guide

Juan J. Morales-Ruiz
and
Jean-Pierre Ramis
July 8, 2009

⇤
The research of the first author has been partially supported by grant MCyT-FEDER
MTM2006-00478 of Spanish goverment. Key words: Hamiltonian Systems, Integrability, Vari-
ational Equations, Di↵erential Galois Theory. Mathematics Subject Classification: 70H06, 70H07,
70H33, 70F07, 70F10, 34A05, 34A30, 34C14, 34M15, 34M35.

ii
 iii

... Quan arribis, si et sento,

hauré de saludar-te amb un gran
crit. Car moro sense cap saviesa,
però molt ric de passos de perdut
vianant.
(Salvador Espriu, Thànatos)

IN MEMORIAM OF OUR FRIEND JOSEP-MARIA PERIS

(1951–2006)
iv CONTENTS

Abstract
We survey recent advances in the non-integrability criteria for Hamiltonian
Systems which involve the di↵erential Galois group of variational equations
along particular solutions. The emphasis is on algorithms and applications,
not theory. For most applications one does not need a deep understanding
of the di↵erential Galois theory. All essentials are presented here, along with
numerous concrete examples.

Contents
1 Introduction 2

2 General Non-integrability Theorems 5

2.1 Algebraic groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Picard-Vessiot Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Non-integrability by First Order Variational Equation . . . . . . . . 9
2.4 Extension to higher order variational equations . . . . . . . . . . . . 12

3 Homogeneous Potentials and Related Problems 18

3.1 Non-integrability of Homogeneous Potentials . . . . . . . . . . . . . 18
3.2 Homogenous Polynomial Potentials . . . . . . . . . . . . . . . . . . . 22
3.3 Some Rational Potentials . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3.1 The determination of the potentials . . . . . . . . . . . . . . . 24
3.3.2 Non-integrability . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4 Celestial Mechanical Problems . . . . . . . . . . . . . . . . . . . . . 32
3.4.1 Some N -Body Problems . . . . . . . . . . . . . . . . . . . . . 32
3.4.2 Hill’s problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.5 Other N-bodiy Problems, Mechanical and Physical Problems . . . . . 52

4 Hamiltonian Rigid Body Problem 55

4.1 The equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2 Non–integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

5 Cosmological Models 58
5.1 Bianchi’s Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.2 Friedman-Robertson-Walker’s Models . . . . . . . . . . . . . . . . . 60

6 An Application to Painlevé’s Transcendents 62

6.1 Painlevé II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.2 Painlevé VI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.3 More about the non integrability of Painlevé I and Painlevé VI . . . 66

A Algorithmic Considerations 69
A.1 Kovacic Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
A.2 Algebrization Procedure . . . . . . . . . . . . . . . . . . . . . . . . . 73
A.3 The importance of logarithmic terms . . . . . . . . . . . . . . . . . . 74
CONTENTS 1

B Hypergeometric Equation 80

C Lamé Equation 82
2 1 INTRODUCTION

1 Introduction
It has been several years since our papers [97, 96, 98] and the book [94] were pub-
lished; in the meantime, new lines of research have been opened, new results have
been obtained by several authors, and old results are included in a natural way in
the framework. Some of them are the following:
a) A proof of the conjecture about the higher order variational equations.

b) The obtention of new results of a global nature as oriented to the classification

of integrable cases of homogeneous polynomial potentials.

c) New non-integrability results for several N -body problems.

d) New non-integrability results for several cosmological models.

e) Non-integrability results for other specific families of systems such as Painlevé’s

transcendents, including new simple proofs of old results, e.g. for the rigid
body.

f) Obstructions to the existence of real analytical first integrals.

g) New contributions about connections of our approach with chaotic dynamics

– more specifically, splitting of separatrices.

h) The proposal of some extensions to non-holonomic mechanical systems, control

theory, and other not necessarily Hamiltonian systems.

Our present aim is to survey these new results. Due to limitations of space
it was impossible to give a complete account of all the above items. The choice
of the topics was subjective, a fact for which we apologize. In particular, topics
f), g) and h) will be not mentioned within the text. Readers interested in f) can
read the papers by Ziglin ([149]) and Audin ([10]). Topic g) started over eight
years ago in a joint publication by Josep-Maria Peris and one of the authors ([95])
and was followed by Yagasaki and others ([142, 143, 144]). Topic h) is now being
a very active research field under Tsygvintsev, Dullin, Maciejewski, Przybylska,
Respondek, Weil, and others; as some of the fundamentals of this area are still not
completely finished, although some important results are already obtained, we only
give a few references. In [31], the obstructions to the existence of analytical first
integrals to the Rattleback problem, a difficult non-Hamiltonian and non-holonomic
rigid mechanical problem, are studied. Reference [70] features the study of the non-
integrability of a non-holonomic Hamiltonian problem: the Suslov problem. Paper
[80] is devoted to proving the non-integrability of a sub-Riemannian problem which
is important in control theory.
The emphasis will be put on the applications and our approach will essentially be
about methodology and algorithms. We also made a considerable e↵ort to remain
at a relatively elementary level and to write a self-contained text. This contribution
is also the answer to some colleagues concerning the need of an introductory text in
 3

our field. Thus, for an important part of the text we assume virtually no di↵erential
Galois prerequisite from the reader. Whoever is interested can find some of the main
results of this theoretical framework in the book [94] or an elementary introduction
with proofs in the monograph of Audin [8], or else in our original articles [96, 97,
98, 101].
As the reader may check, applications are possible through a unified and sys-
tematic approach:

1. Selecting a particular integral curve.

2. Computing the VE.

3. Checking whether the identity component of the di↵erential Galois group of

the VE is commutative.

Step 2 is easy, as we will see. Step 3 is generally quite involved. Fortunately, for
particular cases occurring in many applications, some efficient algebraic algorithms
do exist. The prototype is Kovacic’s algorithm for second-order equations. In nearly
all applications known to the authors, Step 1, common to all classical proofs of non-
integrability, is achieved due to the existence of a completely integrable subsystem,
typically due to the presence of an invariant plane.
In a joint work of the authors with Simó, the above method has been general-
ized to the higher order variational equations, V Ek , where their solutions are the
quadratic, cubic, ... contributions to the Taylor series of the flow along the partic-
ular integral curve. Hence, in the above Steps 2 and 3 we can replace V E1 := V E
by V Ek .
Finally, and although the numerous results and methods in this contribution are
nowadays collected under the umbrella name “Morales-Ramis theory”, it is the au-
thors’ contention that a more proper denomination should be “Ziglin-Morales-Ramis
theory”, since it was Ziglin who first introduced the monodromy group approach to
the variational equations as a fundamental tool for obtaining, and masterfully apply-
ing, necessary conditions to Hamiltonian integrability. Thus, in 1982 Ziglin stated
his fundamental theorem about the monodromy group of the variational equations
for Hamiltonian systems in presence of meromorphic first integrals.

Theorem 1 ([147]). Assume the Hamiltonian system XH admits n m additional

analytical first integrals, independent over a neighborhood of (but not necessarily
on itself ). Assume moreover that the monodromy group of the normal variational
equation ( NVE) contains a non-resonant transformation g. Then any other element
of the monodromy group of the NVE sends eigendirections of g into eigendirections
of g.

(see Section 2.3 for the necessary definitions and notations)

Bearing in mind that the monodromy group is contained in the symplectic group,
we recall that a linear transformation g 2 Sp(2n, C) is resonant if there exist integers
r1 , ..., rn , not all of which equal zero, such that r11 · · · rnn = 1, where we denoted
the eigenvalues of g as i , i 1 .
4 1 INTRODUCTION

This said, in 1983 Ziglin applied Theorem 1 to the non-integrability of several

Hamiltonian systems, in particular solving the integrability problem of the rigid
body with a fixed point completely ([148]).
The connection of Theorem 1 with our theorems in Section 2.3 comes from the
fact that, given a linear di↵erential system, the Galois group contains the mon-
odromy group, and is actually its Zariski closure if the system is Fuchsian – see
Section 2.2. We also remark that for n m 2 Ziglin’s in Theorem 1 does not
assume the complete integrability of the Hamiltonian system, since the latter trait
assumes the involutivity of the first integrals – see Section 2.3.
We would like to thank Sergi Simon for remarks concerning Section 3.4 as well as
for linguistic help. Special thanks are also due to David Blázquez for remarks about
Section 3.3. Also thanks to the anonymous referee for their constructive remarks.
 5

2 General Non-integrability Theorems

2.1 Algebraic groups
The minimum necessary results of linear algebraic groups are presented. An intro-
duction to linear algebraic groups is given in [21]. For more information see the
monographs [46, 20].
A linear algebraic group G (over C) is a subgroup of GL(m, C) whose matrix
coefficients satisfy polynomial equations over C. It has both the structure of a non-
singular algebraic variety and that of a group, both structures being compatible in
that the group operation and inversion are morphisms of algebraic varieties. We
note that in a linear algebraic group there are two di↵erent topologies: the Zariski
topology, where the closed sets are the algebraic sets, and the usual Hausdor↵ topol-
ogy. In particular, an algebraic group is a complex analytical Lie group and we can
consider its Lie algebra. Therefore the dimension of G is the dimension of the Lie
algebra of G. Given a linear algebraic group, the maximal connected subgroup G0
which contains the identity is an algebraic group called the identity component of
G. For those familiar to Lie theory, and since G is a complex analytical Lie group
due to satisfying algebraic equations, the underlying group of G0 coincides with
the identity component of G considered as a complex analytical Lie group. We
remark that an algebraic linear (or affine) group G is usually defined as an affine
algebraic variety with a group structure, with the above compatibility condition as
to group multiplication and taking of inverses. Then, there is a rational faithful
representation of G as a closed subgroup of GL(m, C), for some m, and we obtain
the equivalence with our definition.
It is clear that the classical linear complex groups are linear algebraic groups.
For instance SL(n, C), SO(n, C) (rotation group) and Sp(n, C) ⇢ Gl(2n, C) (sym-
plectic group) are linear algebraic groups, since they are defined by polynomial
identities.

Proposition 1. The identity component G0 of a linear algebraic group G is a closed

(with respect to the above two topologies) normal subgroup of G of finite index and
is connected with respect to the above two topologies. Furthermore G/G0 is a finite
group given by the classes of the irreducible connected components of G.

We note that by the above proposition G0 is also a linear algebraic group and
the Lie algebra of G, Lie(G) = G coincides with the Lie algebra of G0 , Lie(G0 ) = G.
As for every Lie group, G0 is solvable or commutative if, and only if, G is solvable,
respectively commutative. Furthermore, G is connected if, and only if, G = G0 .
The characterization of the connected solvable linear algebraic groups is given
by the Lie-Kolchin theorem.

Theorem 2 (Lie-Kolchin Theorem). A connected linear algebraic group is solvable

if, and only if, it is conjugate to a triangular group.

Given a subset S ⇢ GL(n, C), let M be the group generated by S and let G be
the Zariski closure of the group M . By definition the group G is a linear algebraic
6 2 GENERAL NON-INTEGRABILITY THEOREMS

group; we say G is topologically generated by M . All through the text, M will be

typically the monodromy group of a Fuchsian linear di↵erential equation and G will
be its Galois group.
Since most of the examples of irreducible equations appearing later on are second-
order and symplectic, we end this section with a classification of the algebraic sub-
groups of SL(2, C) = Sp(1, C). We shall need two lemmas.

Lemma 1 ([54]). Let G be an algebraic group contained in SL(2, C). Assume that
the identity component G0 of G is solvable. Then G is conjugate to one of the
following types:
(1) G is finite,
⇢✓ ◆ ✓ 1
◆
0 0
(2) G = 1 , , 2 C⇤ ,
0 0
(3) G is triangular.

Lemma 2. Let G be an algebraic subgroup of SL(2, C) such that the identity com-
ponent G0 is not solvable. Then G = SL(2, C).

The last lemma is well–known and follows easily from consideration of the Lie
algebra of G ⇢ SL(2, C). Indeed, if G0 is not solvable then the dimension of G must
be equal to 3, because all 2-dimensional Lie algebras are solvable.

Proposition 2 ([92]). Any algebraic subgroup G of SL(2, C) is conjugate to one of

the following types:
✓ ◆
0 1 0
1. Finite, G = {1}, where 1 = .
0 1
⇢✓ ◆
0 1 0
2. G = G = ,µ2C .
µ 1
⇢✓ ◆
0
3. Gk = 1 , is a k-root of unity, µ 2 C ,
µ
⇢✓ ◆
0 1 0
G = ,µ2C .
µ 1
⇢✓ ◆
0
4. G = G =0
1 , 2 C⇤ .
0
⇢✓ ◆ ✓ 1
◆
0 0
5. G = 1 , , 2 C⇤ ,
0 0
⇢✓ ◆
0
0
G = 1 , 2 C⇤ .
0
⇢✓ ◆
0
6. G = G =0
1 , 2 C⇤ , µ 2 C .
µ
2.2 Picard-Vessiot Theory 7

7. G = G0 = SL(2, C).

The above proposition is analogous to that found in [58], page 7. However, we not
only need to know when the identity component of the Galois group is solvable, but
when it is commutative. We remark that the identity component G0 is commutative
in cases (1)–(5) and solvable in cases (1)–(6).

2.2 Picard-Vessiot Theory

At the end of the nineteenth century, Picard ([110, 111], [112, Chapitre XVII]) and,
in a clearer way, Vessiot in his PhD Thesis ([139]), created and developed a Galois
theory for linear di↵erential equations. This field of study, henceforth called Picard-
Vessiot theory, was continued from the forties to the sixties of the twentieth-century
by Kolchin , through the introduction of the modern algebraic abstract terminology
and the obtention of new important results, see [57] and references therein. Today
the standard reference of this theory is the monograph [138].
In the last years, a new revival of interest in the di↵erential Galois theory
is being observed. This is partially due to the connections and applications to
other areas of mathematics: number theory [14, 55], asymptotic theory [85], non-
integrability of dynamical systems, etc. Here we are interested in the applications
to non-integrability. As we shall see, within di↵erential Galois theory there is a
very nice concept of “integrability”, i.e., solutions in closed form. Furthermore, all
information about the integrability of the equation is coded in the identity compo-
nent of the Galois group: a linear equation is integrable if, and only if, the identity
component of its Galois group is solvable. We only review the necessary definitions
and results of the Picard-Vessiot theory in order to understand the applications to
non-integrability. For more information see [138].
A di↵erential field K is a field with a derivative (or derivation) = 0 , i.e., an
additive mapping satisfying the Leibniz rule. The only case we are interested in is
K = M( ), the meromorphic functions over a connected Riemann surface . The
reason for this notation will be clear below: will be the set of singular points
of the linear di↵erential equation, i.e., poles of the coefficients with dtd as derivation,
t being a local coordinate over the Riemann surface . A particular classical case
is when K = C(t) = M(P1 ) is the field of rational functions, i.e., the field of
meromorphic functions over the Riemann sphere P1 . Another interesting example
for the applications is a field of elliptic functions.
We can define di↵erential subfields and di↵erential extensions in a direct way by
requiring that inclusions commute with the derivation. Analogously, a di↵erential
automorphism in K is an automorphism commuting with the derivative. The field
of constants of K is the kernel of the derivative. In the above examples C is such a
kernel. From now on we will assume this is the case.
Let

⇠ 0 = A⇠, A 2 M at(m, K) (1)

8 2 GENERAL NON-INTEGRABILITY THEOREMS

be a linear di↵erential equation. We now proceed to associate to (1) the so-called

Picard-Vessiot extension of K. The Picard-Vessiot extension L of (1) is an extension
of K, such that if u1 , ..., um is a “fundamental” system of solutions of the equation (1)
(i.e., linearly independent over C), then L = K(uij ) (rational functions in K in the
coefficients of the “fundamental” matrix U = (u1 · · · um ) ). This is the extension
of K generated by K together with uij . We observe that L is a di↵erential field
(by (1)). The existence and unicity of the Picard-Vessiot extensions was proven by
Kolchin. In the analytical case K = M( ), 0 = d/dt, this result is essentially the
existence and uniqueness theorem for linear di↵erential equations and it is already
in Vessiot [139].
As in classical Galois theory of algebraic equations, we define the Galois group of
(1), G := GalK (L) = Gal(L/K), as the group of all the (di↵erential) automorphisms
of L leaving the elements of K fixed. Then one of the main results of the theory
is that the Galois group of (1) is faithfully represented as an algebraic linear group
over C, the representation is given by the action 2 G,

(U ) = U B ,

B 2 GL(m, C). The other fundamental result is the Galois correspondence between
algebraic subgroups and intermediate extensions.
Theorem 3. Let L/K be the Picard-Vessiot extension associated to a linear di↵er-
ential equation. Given any subgroup H ⇢ G := GalK (L), let KH denote the subfield
of L consisting of those elements fixed by H. Then the mapping H 7! M := KH
restricts to a bijection between between the algebraic subgroups of G and the inter-
mediary di↵erential fields K ⇢ M ⇢ L. Furthermore, we have
(i) To the algebraic subgroups H ⇢ G := GalK (L) correspond the Picard-Vessiot
extensions L/KH .
(ii) The group H is a normal algebraic subgroup of G if, and only if, the extension
KH /K is a Picard-Vessiot extension. Then the group G/H is a linear algebraic group
and G/H = GalK (KH ).
(iii) For an arbitrary subgroup H ⇢ G the group GalKH (L) is the Zariski closure
(over the complex field C) of H.
As a corollary, when we consider the (relative) algebraic closure K of K in L,
we obtain GalK (K) = G/G0 , where G0 = GalK (L) is the identity component of
the Galois group G corresponding to the transcendental part of the Picard-Vessiot
extension, i.e., by definition, the extension L/K is the maximal transcendental ex-
tension among those L/L1 , L1 being an extension of K. If K = K (i.e., if G = G0 ),
we say L/K is a purely transcendental extension.
We call a linear di↵erential equation integrable if we can obtain its Picard-Vessiot
extension K ⇢ L and, hence, its general solution, by adjunction to K of integrals,
exponentials of integrals or algebraic functions of elements of K. In other words,
there exists a chain of di↵erential extensions K1 := K ⇢ K2 ⇢ · · · ⇢ Kr := L,
where each extension is given by the adjunction of one element a, Ki ⇢ Ki+1 =
Ki (a, a0 , a00 , ...), such that a satisfies one of the following conditions:
2.3 Non-integrability by First Order Variational Equation 9

(i) a0 2 Ki ,
(ii) a0 = ba, b 2 Ki ,
(iii) a is algebraic over Ki .
Then, it can be proven that a linear di↵erential equation is integrable if, and only
if, the identity component G0 of the Galois group is a solvable group. In particular,
if G0 is commutative, the equation is integrable.
The usual terminology for integrable linear equations is that the associated
Picard-Vessiot extension is a Liouville extension [54]. We prefer to use a termi-
nology in agreement with our dynamical approach and with the creators of the
theory ([139]).
Furthermore, by a classical theorem credited to Schlesinger, the relation between
the monodromy and the Galois group is as follows. Let be the set of singular
points of the equation i.e., the poles of the coefficients on . We recall that the
monodromy group of the equation is the subgroup of the linear group defined as
the image of a representation of the fundamental group ⇡1 ( ) into the linear group
GL(m, C). This representation is obtained by analytical continuation of the solu-
tions along the elements of ⇡1 ( ). The monodromy group M is contained in the
Galois group G and if the equation is Fuchsian (i.e., it has regular singular points
only), then M is Zariski dense in G, see for instance [138]. In particular, this implies
that for Fuchsian di↵erential equations the Galois group is solvable or commutative,
if, an only if, the monodromy group is solvable or commutative, respectively. In
the general case, the second author found a generalization of the above and, for
example, he showed that the Stokes matrices associated to an irregular singularity
belong to the Galois group, see [85].
We would like to point out that in the last few years a new non-linear di↵erential
Galois theory has come into being ([82, 83, 22, 137]. The authors are convinced that
this theory will play an important role in the context of the integrability of dynamical
systems.

2.3 Non-integrability by First Order Variational Equation

Given a dynamical system,

ż = X(z), (2)
with a particular integral curve z = (t), at the end of the nineteenth century
Poincaré introduced the variational equation (VE ) along z = (t),

⇠˙ = X 0 ( (t))⇠, (3)
as the fundamental tool to study the behavior of (2) in a neighborhood (t) [108].
Equation (3) describes the linear part of the flow of (2) along z = (t).
We have the following General Principle:
General Principle: If we assume that the dynamical system (2) is “integrable” in
any reasonable sense, then it is natural to conjecture that the linearized di↵erential
equation (3) must be also “integrable”.
10 2 GENERAL NON-INTEGRABILITY THEOREMS

It seems clear that in order to convert this principle in a true conjecture it is nec-
essary to clarify what kind of “integrability ” is considered for equations (2) and (3).

As (3) is a linear di↵erential equation, it is natural to consider the integrability

of this equation in the context of the Galois theory of linear di↵erential equations.
In order to do that, as was explained in the previous section, we need to assume the
field of constants is the complex field. Therefore, we have to extend ourselves to the
complex analytical category, i.e., all equations are complex analytical and defined
over complex analytical spaces.
For complex analytical Hamiltonian systems the General Principle works well
and we obtained the following result, which in some sense may be considered as a
generalization of a result by Ziglin in 1982 [147]. The essential idea is to consider in
the General Principle not only integrability of the variational equations (character-
ized by the solvability of the identity component of its Galois group) but commu-
tativity of the identity component of the Galois group of the variational equations.
This is natural because, for integrable Hamiltonian systems, we have an abelian
Poisson Lie algebra of first integrals of maximal dimension.
Let H be a complex analytical Hamiltonian function defined on a symplectic
manifold M of (complex) dimension 2n and let XH be the Hamiltonian system de-
fined by H. In canonical coordinates, z = (x1 , ..., xn , y1 , ..., yn ), it is given classically
by

@H
ẋi = ,
@yi
@H
ẏi = ,
@xi
i = 1, ..., n.
We recall here the definition of integrability for Hamiltonian systems. One says
that XH = (@H/@yi , @H/@xi ) i = 1, ..., n, is completely integrable or Liouville
integrable if there are n functions f1 = H, f2 ,..., fn , such that
(1) they are functionally independent i.e., the 1-forms dfi i = 1, 2, ..., n, are
linearly independent over a dense open set U ⇢ M , Ū = M ;
(2) they form an involutive set, {fi , fj } = 0, i, j = 1, 2, ..., n.
We recall that in canonical coordinates the Poisson bracket has the classical
expression

Xn
@f @g @f @g
{f, g} = .
i=1
@yi @xi @xi @yi

We remark that in virtue of item (2) above the functions fi , i = 1, ..., n are first
integrals of XH . It is very important to be precise regarding the degree of regu-
larity of these first integrals. In our contribution we assume that the first integrals
are meromorphic. Unless otherwise stated, this is the only type of integrability of
2.3 Non-integrability by First Order Variational Equation 11

Hamiltonian systems that we consider in the next pages. Sometimes, to recall this
fact we shall talk about meromorphic (complete) integrability.
Now we can write the variational equations along a particular integral curve
z = (t) of the vector field XH

⇠˙ = XH
0
( (t))⇠. (4)
Using the linear first integral dH(z(t)) of the variational equation it is possible
to reduce this variational equation and to obtain the so-called normal variational
equation which, in suitable coordinates, can be written as a linear Hamiltonian
system

⌘˙ = JS(t)⌘,
where, as usual, ✓ ◆
0 I
J=
I 0
is the standard matrix of the symplectic form of dimension 2(n 1).
More generally, if, including the Hamiltonian, there are m meromorphic first in-
tegrals independent over and in involution, we can reduce the number of degrees of
freedom of the variational equation (4) by m and obtain the normal variational equa-
tion (NVE) which, in suitable coordinates, can be written as a 2(n m)-dimensional
linear system

⌘˙ = JS(t)⌘, (5)
where now J is the matrix of the symplectic form of dimension 2(n m). For more
details about the reduction to the NVE, see [96] (or [94]).

Theorem 4. [[96], see also [94]] Assume a complex analytic Hamiltonian system is
meromorphically completely integrable in a neighborhood of the integral curve z =
(t) . Then the identity components of the Galois groups of the variational equations
(4) and of the normal variational equations (5) are commutative groups.

We remark that it is a typical version of several possible theorems. In some cases

it is interesting to add to the manifold M some points at infinity; thus we suppose
that we are in the following situation: M is an open subset of a complex manifold
M , M \ M is an hypersurface (which is by definition the hypersurface at infinity),
the two-form ! on M defining the symplectic structure extends meromorphically on
M and the vector field XH extends meromorphically on M . In such a case, when (4)
has irregular singular points at infinity, we only obtain obstructions to the existence
of first integrals which are meromorphic along , ie, also at the points at infinity of
; for example, for rational first integrals when M is a projective manifold. From a
dynamical point of view, the singular points of the variational equation (4), ,
correspond to equilibrium points, meromorphic singularities of the Hamiltonian field
or points at infinity.
12 2 GENERAL NON-INTEGRABILITY THEOREMS

Theorem 5. ([96], see also [94]) Consider a complex symplectic manifold (M, !),
which is an open subset of a complex manifold M , M \ M being an hypersurface
and ! admitting a meromorphic extension on M . Let XH be a meromorphic vector
field on M which is analytic and Hamiltonian on M . If the system ż = XH (z) is
meromorphically integrable in a neighborhood in M of some integral curve with
first integrals which extend into meromorphic functions on a neighborhood of , then
the identity component of the Galois groups of (4) and (5) (interpreted as di↵erential
equations on ) are commutative groups. In particular, let M be an open domain
of a symplectic complex space and assume the points at infinity of (4) (or (5)) are
irregular singular points and the identity component of the Galois group of (4) (or
(5)) is not commutative, then the Hamiltonian system is not integrable by rational
first integrals.
One of the essential points in the proof of the above theorems is the following
lemma:
Key Lemma: ([96], see also [94]) Let f be a meromorphic first integral of the
dynamical system (2). Then the Galois group of (3) has a non-trivial rational in-
variant.
We remark that this Lemma is valid for general dynamical systems, not only
for Hamiltonian ones. Moreover, it is possible to generalize this lemma to tensor
invariants; for instance, to symplectic forms in the case of Hamiltonian systems or
to invariant volume forms. We shall not discuss these ideas here.

2.4 Extension to higher order variational equations

Theorem 4 (and 5) has been generalized to higher order variational equations VE k
along , with k > 1 (the solutions of these equations are the quadratic, cubic, etc.
contributions to the flow of the Hamiltonian system along the particular solution
z = (t) = (z0 , t)), V E1 being the equation (3)[101].
The “fundamental” solution of VE k of a dynamical system (2) is given by
(1) (2) (k)
( (t), (t), . . . , (t)),

where
(1) 1 (k)
(z, t) = (z0 , t) + (t)(z z0 ) + . . . + (t)(z z0 )k + . . .
k!
the Taylor series up to order k of the flow (z, t) with respect to the variable z at the
@k
point (z0 , t). That is, (k) (t) = k (z0 , t). The initial conditions are (1) (0) = Idm
@z
and (j) (0) = 0 for all j > 1. We stress that, in contrast to some definitions, we
do not consider the di↵erential equations for (k) , but for ( (1) , (2) , . . . , (k) ), as
variational equation of order k.
The variational equation VE k is not linear, but it is in fact equivalent to a
linear di↵erential equation: there exists a linear di↵erential equation LVE k with
coefficients in the field of meromorphic functions over such that the di↵erential
2.4 Extension to higher order variational equations 13

extensions generated by the solutions of VE k coincide with the Picard-Vessiot exten-

sions of LVE k . Then we can consider the Galois group Gk of VE k , Gk = Gal(VE k ),
i.e., of the LVE k . For simplicity we shall denote both VE k and LVE k by VE k .
Furthermore, the singular points of the equations VE k are the same as the first
order variational equation VE 1 and a singular point of VE k (k 1) is irregular if
and only if it is irregular for VE 1 .
Although in order to obtain the main theoretical results it is convenient to work
with the jet formalism (see [101]), from a computational practical point of view, the
higher order variational equations can be obtained using the small parameter method
of the masters (Poincaré, Liapunov,...). Hence, we expand the general solution of
the non-linear equation (2) along the particular solution (t)

⇠ (2) 2 ⇠ (k) k
z(t) (t) = ⇠ (1) " + " + ··· + " + ··· ,
2! k!
being " a small parameter. Introducing the above in equation (2) and equating the
same powers of ", using the fact that

1 (2)
X(z) = X( (t)) + X 0 ( (t))(z(t) (t)) + X ( (t))(z(t) (t))2 + · · · +
2!

1 (k)
X ( (t))(z(t) (t))k + · · · ,
k!
we obtain the variational equation of order k, VE k ,

⇠˙(j) = X 0 ( (t))⇠ (j) + Pj (⇠ (1) , . . . , ⇠ (j 1)

), j = 1, . . . , k, (6)
where P1 ⌘ 0 (i.e., VE 1 becomes (3)) and Pj is polynomial of degree j in ⇠ (1) , . . . , ⇠ (j 1) ,
with meromorphic coefficients over (i.e., in the same di↵erential field of coef-
ficients of the first order variational equation). The initial conditions of (6) are
(1) (2) (k)
(⇠0 , ⇠0 , ..., ⇠0 ) = (Idm , 0, ..., 0). We observe that the solution of (6) is obtained
from the solution of the first order variational equation by successive applications
of the constants variation method of Lagrange: we substitute the solutions of VE 1
in P2 (⇠1 ), we solve the second of the equations of VE 2 , i.e., the equation in ⇠ (2)
(2)
(with initial conditions ⇠0 = 0) and we substitute all the above solutions of VE 2 in
P3 (⇠1 , ⇠2 ), and so on. In particular, (6) is integrable if, and only if, VE 1 is integrable.
For instance, the third order variational equation, VE 3 , is explicitly given by the
system

d
P
⇠
dt j,k
= Pi @i Xj ⇠i,k , P
d
⇠
dt j,k1 k2
= @i Xj ⇠i,k1 k2 + i1 ,i2 @i21 ,i2 Xj ⇠i1 ,k1 ⇠i2 ,k2 ,
d
Pi P
⇠
dt j,k1 k2 k3
= @X⇠ + i1 ,i2 @i21 ,i2 Xj ⇠i1 ,k1 k2 ⇠i2 ,k3 + (7)
Pi i j2 i,k1 k2 k3 P 2
@ X j ⇠ i ,k k ⇠i ,k + i1 ,i2 @i1 ,i2 Xj ⇠i1 ,k1 ⇠i2 ,k2 k3 +
P1 2
i ,i i1 ,i
3
2 1 1 3 2 2

i1 ,i2 ,i3 @i1 ,i2 ,i3 Xj ⇠i1 ,k1 ⇠i2 ,k2 ⇠i3 ,k3 ,
14 2 GENERAL NON-INTEGRABILITY THEOREMS

where ⇠ (1) = (⇠j,k ), ⇠ (2) = (⇠j,k1 k2 ), ⇠ (3) = (⇠j,k1 k2 k3 ), ⇠j,k := @z@k i (z0 , t), ⇠j,k1 k2 :=
@2 3
(z , t), ⇠j,k1 k2 k3 := @zk @z@k @zk i (z0 , t), j = k = k1 = k2 = k3 = 1, ..., m. We
@zk1 @zk2 i 0 1 2 3
remark that the matrix (⇠j,k ) is in fact the fundamental matrix of the equation (4).
Now we are going to describe the practical method of linearization of the equa-
tions VE k . The problem is to find a system of linear equations for
(⇠ (1) (t), ⇠ (2) (t), ..., ⇠ (k) (t))
equivalent to VE k . It is enough to write the equations satisfied by the monomials
appearing in Pj . This is the content of the next lemma.
Lemma 3. Let z 2 Cq . Assume the componentsP (z1 , . . . , zq ) of z satisfy linear
homogeneous di↵erential equations żi = qj=1 aij (t)zj . Then the monomials z k :=
Qq ki
i=1 zi of order |k| = k1 + · · · + kq satisfy also a system of linear homogeneous
di↵erential equations.
Proof. Let k = (k1 , . . . , kq ) a multi-index of non-negative integers. Then
q q q
!
d k X kj 1
X Y
ki
z = kj zj ajr zr zi , (8)
dt j=1 r=1 i=1,i6=j

the right hand side being also homogeneous of degree |k| in z. 2

We observe that the above Lemma is nothing other than the pull–back to the
symmetric fibrePbundle, S k (Cq ), of the connection associated to the linear di↵erential
q
equation żi = j=1 aij (t)zj , connected to the Tannakian formalism (see [94], Section
2.3).
Then, in order to linearize VE k , after the last equation corresponding to VE k
we can supplement the system of linear di↵erential equations with the equations
for the components of (⇠ (i1 ) )m1 , (⇠ (i2 ) )m2 , . . . , (⇠ (is ) )ms . For more details about this
linearization see our original paper, [101].
Since we can now consider the equations VE k as linear di↵erential equations,
we can talk about their Picard-Vessiot extensions and about their Galois groups
Gk . Then it was conjectured in [94] (Section 8.3) that a necessary condition for
meromorphic integrability is that the identity component of Gk must be commuta-
tive, for any k. This can be interpreted as some generalized version of our General
Principle to this context. So, using also a generalized version of the Key Lemma to
Hamiltonian dynamical systems, in a joint work of the authors with Simó we proved
the following theorem.
Theorem 6 ([101] ). Assume that a complex analytical Hamiltonian system is inte-
grable by meromorphic first integrals in a neighborhood of the integral curve z = (t).
Then the identity components (Gk )0 , k 1, of the Galois groups of the variational
equations along are commutative.
Example: a Hénon-Heiles system. Let us to consider the two degrees of freedom
Hamiltonian system defined by the cubic Hamiltonian
1 1 1 1 1
H = (y12 + y22 ) + x21 + x22 + x31 + x1 x22 . (9)
2 2 2 3 2
2.4 Extension to higher order variational equations 15

In [101] it was proven non-integrable, we follow this reference. In order to prove this
we use the method in Appendix A, A.3, a).
This system is one of the Hénon-Heiles family of one-parameter Hamiltonians
considered by Ito [49]. By means of the first order variational equation it was proven
that for all but four of the values of the parameter, the systems in this family are non-
integrable, see [49, 93, 94]. Three of these remaining cases are trivially integrable.
The fourth case is (9) and its non-integrability was conjectured from numerical
experiments. We can prove the non-integrability of this last case using Theorem
6. As we will see, we obtain an obstruction given by a non trivial residue at the
integrand of a quadrature appearing in the solution of the third variational equation.
For the Hamiltonian (9), the plane x2 = y2 = 0 is invariant, foliated by integral
curves h which are generically elliptic curves

1 2 1 2 1 3
y + x + x = h,
2 1 2 1 3 1
parametrized in time by x1 = 6}(t) 12 , y1 = ẋ1 = 6}(t), ˙ x2 = y2 = 0, where }
1 1 h
is the Weiertrass elliptic function with invariants g2 = 12 and g3 = 108 18
. In the
above computations we used the di↵erential equation satisfied by the function }(t):
}˙ 2 = 4}3 g2 } g3 , see Appendix C.
For h⇤ = 1/6, the elliptic curve degenerates to a rational one, h⇤ := : the real
period of x1 (t) and y1 (t) goes to infinity, and is parametrized in time by

3/2 (3/2) sinh(t/2)

x1 (t) = 1, y1 (t) = , x2 = y2 = 0. (10)
cosh2 (t/2) cosh3 (t/2)

We observe that in any case for h , h arbitrary, x1 (t) and y1 (t), have only one pole
in a fundamental domain in the complex plane. This pole will be a singular point
of the corresponding variational equations.
Then the first variational equation VE 1 along h 6= is given by

⇠¨1 = 12 }(t)⇠1 ,
(11)
⇠¨2 = 6 }(t) 12 ⇠2 .
(in fact, we are interested in the fundamental matrix (⇠i,j ) i, j = 1, .., 4 of the above
system). The first of the equations in (11) is the tangential variational equation
and the second one is the normal variational equation, with coefficients 1 2x1 (t)
and 1 x1 (t), respectively. We know that the tangential variational equation
has a solution belonging to the field of elliptic functions K = C(}(t), }(t)),˙ since a
particular solution of the tangential variational equation is obtained by derivation of
the particular solution z = z(t) along which we compute the variational equation and
thus, in this case, ⇠1 = ẋ1 = y1 = 6}(t).
˙ In fact a direct computation verifies that
}(t)
˙ is a particular solution, where again we use the di↵erential equation satisfied by
}(t). Thus, the first of the equations in (11) falls in the Lamé case, see Appendix C.
In an analogous way, the normal variational equation has also a particular solution
1
⇠2 = }(t) + 12 2 K. Hence, both equations in (11) fall in the Lamé case and the
Galois group of both of them is not finite (Appendix C).
16 2 GENERAL NON-INTEGRABILITY THEOREMS

From the results in Appendix A, A.3, a), Proposition 7, we know that if a

particular solution of VE k , with k > 1, has a solution with a local logarithm, then
the system is not integrable. We will prove this is the case for VE 3 , i.e., that some
of the final elements ⇠j,k1 k2 k3 are not meromorphic. We select the ⇠2,222 . Instead of
working with the family of curves h , h 6= h⇤ , with VE 1 given by (11), we shall
work with the integral curve . Then it is clear that, from the analytic dependence
on the parameter h, if we have a non-trivial monodromy (i.e., a local logarithm)
around the singular point in the variational equation VE 3 over the integral curve ,
then the variational equation over h also has a non-trivial monodromy for h 6= h⇤
and small |h h⇤ |.
Let a31 = 1 2x1 , a42 = 1 x1 the only non-zero and non-trivial elements
in the matrix of coefficients of V E1 along . To obtain ⇠2,222 from the equation (7),
we only need to integrate the following systems
✓ ◆ ✓ ◆✓ ◆ ✓ ◆ ✓ ◆✓ ◆ ✓ ◆
⇠˙2,2 0 1 ⇠2,2 ⇠˙1,22 0 1 ⇠1,22 0
= , = + (12)
⇠˙4,2 a42 0 ⇠4,2 ⇠˙3,22 a31 0 ⇠3,22 2
⇠2,2

and ✓ ◆ ✓ ◆✓ ◆ ✓ ◆
⇠˙2,222 0 1 ⇠2,222 0
= + . (13)
⇠˙4,222 a42 0 ⇠4,222 3⇠2,2 ⇠1,22
It is clear that to integrate (12) and (13) we need to solve the first order variational
equations, both tangential and normal which, as above for h (equation(11)), are
uncoupled. The solutions can be written explicitly. To shorten the notation we
introduce c := cosh(t/2) and s := sinh(t/2). Then

15ts 15 5 c2 4
⇠1,1 = + , ⇠1,3 = y1 ,
16c3 8c22 8 4 3
15t(3 2c ) 45s sc 4
⇠3,1 = 4 3
, ⇠3,3 = (x1 + x21 ),
32c 16c 4 3 (14)
tx1 3s
⇠2,2 = 2x1 , ⇠2,4 = + ,
2 2c
x1 ty1 3
⇠4,2 = 2y1 , ⇠4,4 = + + 2.
2 2 4c
Furthermore ✓ ◆
2 16 3
⇠1,22 = ⇠1,1 x + ⇠1,3 K(t), (15)
8 9 1
where ✓ ◆ ✓ ◆
45 45 15 45 15 3
K(t) = t 6
+ 4 +s + 3 +c .
16c 8c 4c2 8c 5 2c c
We remark that one of the columns of the fundamental matrix of the normal vari-
ational equations coincides (except by a factor of 2) with (10). This is true for any
h because (x1 , y1 ) are solutions of the first equation in (12).
Having (14) and (15) we are ready to solve (13). As the homogeneous part
coincides with the first order normal variational equation, the solution, after closing
the loop, is given by the method of variation of constants
2.4 Extension to higher order variational equations 17

✓ ◆ ✓ ◆Z ✓ ◆
⇠2,222 ⇠2,2 ⇠2,4 ⇠2,4 R dt
= , where R(t) = 3⇠2,2 ⇠1,22 . (16)
⇠4,222 ⇠4,2 x4,4 ⇠2,2 R dt

It is readily checked that the residues inside the integral are 72/5 and 0, respectively.
72
Hence, the final value of ⇠2,222 after the loop is 2⇡ i, due to the existence of a local
5
logarithmic term. Hence, from Proposition 7 of Appendix A, we have proven the
following result.

Proposition 3 ([101] ). The system (9) is not integrable by means of meromorphic

first integrals.

We observe that in order to prove there is non–trivial local monodromy in ⇠2,222 ,

it is also possible to use another method: instead of using the analytic dependence
with respect to h, we can work directly with the variational equations along h with
some h 6= 0 and in an analogous way to check the existence a non-trivial residue in
the integrand ⇠2,4 R. This is the standard method followed by several authors.
We remark that since the Galois group G3 is connected (i.e., G3 = (G3 )0 ), it is
possible to use the weaker result given by Lemma 11 of Appendix A rather than
Proposition 7.
Recently a general methodology to deal with non-integrability criteria using
higher variational equations was developed by R. Martı́nez and C. Simó [86, 87, 88].
Using this methodology, they prove in particular the non-integrability of a non-
linear spring pendulum problem and the non-integrability of the Swinging Atwood
Machine for the values of the parameter that, in each case, cannot be decided using
first order variational equations ([77]). For the Swinging Atwood Machine with pul-
leys it is possible to prove non-integrability in all cases using only first variational
equations [114].
As a conclusion to this chapter, we can say that all of our approach is based upon
two simple facts:
(i) A heuristic guiding General Principle.
(ii) The Key Lemma.
18 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

3 Homogeneous Potentials and Related Problems

In this chapter we survey how the preceding methods have been applied to classical
mechanics systems involving homogenous potentials. Some other related problems,
like Celestial Mechanical ones, are included.

3.1 Non-integrability of Homogeneous Potentials

Here we recall the main general non-integrability result about Hamiltonians with
homogeneous potentials obtained by the authors some years ago [97].
Consider an n-degrees-of-freedom Hamiltonian system with Hamiltonian
1
H(x, y) = T + V = (y12 + ... + yn2 ) + V (x1 , ..., xn ), (17)
2
V being a complex homogeneous function of integer degree k and 2  n.
From the homogeneity of V , it is possible to obtain an invariant plane

x = z(t)c,
y = ż(t)c,
where z = z(t) is a solution of the (scalar) hyperelliptic di↵erential equation
2
ż 2 =
(1 z k )
k
6 0), and c = (c1 , c2 , ...cn ) is a solution of the equation
(where we assume case k =

c = V 0 (c). (18)
This is our particular solution along which we compute the variational equation
VE and the normal variational equation NVE. We shall call these the homothetical
solutions of the Hamiltonian system (17) and denote solutions of (18) as homoth-
etical points. In most of the references about the integrability of the homogeneous
potentials the solutions of (18) are called Darboux points (see [74], for instance); we
use the standard terminology in the Celestial Mechanical case (see later).
The VE along is given in the temporal parametrization by

⌘¨ = z(t)k 2 V 00 (c)⌘.
Assume V 00 (c) is diagonalizable. Due to the symmetry of the Hessian matrix V 00 (c),
it is possible to express the VE as a direct sum of second order equations

⌘¨i = z(t)k 2
i ⌘i , i = 1, 2, ..., n,
where we keep ⌘ for the new variable, i being the eigenvalues of the matrix V 00 (c).
We call these eigenvalues Yoshida coefficients. One of the above second order
equations is the tangential variational equation, say, the equation corresponding
to n = k 1. This equation is trivially solvable, whereas the NVE is an equation
in the variables ⇠ := (⌘1 , ..., ⌘n 1 ) := (⇠1 , ..., ⇠n 1 ), i.e.,
3.1 Non-integrability of Homogeneous Potentials 19

⇠¨ = z(t)k 2 diag( 1 , ..., n 1 )⇠.

Now, following Yoshida [145], we consider the change of variable (which happens
to be a finite branched covering map),

! P1 ,

given by t 7! x, where x =: z(t)k (here is the compact hyperelliptic Riemann

surface of the hyperelliptic curve w2 = k2 (1 z k ), see [97] or [94] for the notation and
technical details). Thanks to the symmetries of this problem, we obtain as NVE a
system of independent hypergeometric di↵erential equations in the new independent
variable x

d2 ⇠ k 1 3k 2 d⇠ i
x(1 x) 2
+( x) + ⇠ = 0, i = 1, 2, ..., n 1. (ANVEi )
dx k 2k dx 2k

Each of these equations (ANVEi ), corresponding to the Yoshida coefficient i , is

part of the system called the algebraic normal variational equation ANVE (see Ap-
pendix A, Section A.2). In fact, the ANVE splits into a system of n 1 independent
equations (ANVEi ), i = 1, ..., n 1. Then it is clear that the ANVE is integrable
if, and only if, each of the (ANVEi ) is also integrable. That this, the identity com-
ponent of the Galois Group of the ANVE is solvable if, and only if, each one of
the identity components of the Galois Group of the (ANVEi ) i = 1, 2, . . . , n 1, is
solvable.
As was observed by Yoshida, each one of the above (ANVEi ) is an hypergeometric
equation (equation (129) of Appendix B) with three regular singular points at x = 0,
x = 1 and x = 1. By Theorem 24 of Appendix A the identity component of the
Galois Group of the NVE is the same as the identity component of the Galois
Group of the ANVE. By adapting Kimura’s table (Theorem 26 of Appendix B)
of integrable hypergeometric equations to the new hypothesis, namely assuming
that the Galois di↵erential group of each of the variational equations must have a
commutative identity component, we obtain the following result (in fact, for this
particular family of hypergeometric equations all the cases of integrability have a
commutative identity component):

Theorem 7 ([97], see also [94]). Let XH be a Hamiltonian system given by (17) and
c an homothetical point such that V 00 (c) is diagonalizable. If XH is meromorphically
completely integrable, then each pair (k, i ) matches one of the following items (p
20 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

being an arbitrary integer):

k k
2
1 k p + p (p 1) k2 10 3 25
24
1
24
12
5
+ 6p
2 2 arbitrary z 2 C 11 3 1
24
+ 1
24
(2 + 6p)2
2
3 2 arbitrary z 2 C 12 3 1
24
+ 1
24
3
2
+ 6p
49 1 10 2 1 1 6 2
4 5 40 40 3
+ 10p 13 3 24
+ 24 5
+ 6p
(19)
2
5 5 49
40
1
40
(4 + 10p)2 14 3 1
24
+ 1
24
12
5
+ 6p
9 1 4 2 1 1 4 2
6 4 8 8 3
+ 4p 15 4 8
+ 8 3
+ 4p
2
7 3 25
24
1
24
(2 + 6p)2 16 5 9
40
1
+ 40 10
3
+ 10p
2
8 3 25
24
1
24
3
2
+ 6p 17 5 9
40
1
+ 40 (4 + 10p)2
25 1 6 2 1 k 1
9 3 24 24 5
+ 6p 18 k 2 k
+ p (p + 1) k

This theorem is a generalization of a necessary condition of integrability obtained

by Yoshida using Ziglin’s approach [145].
Hence, in order to prove the non-integrability of a given Hamiltonian system
with a homogeneous potential:

(i) we find the homothetical points, solutions ci of the equation

c = V 0 (c)

(ii) we prove that for some of the ci in (i), at least one of the eigenvalues of of
V 00 (ci ) is not inside the table (7).

Furthermore, we can stretch the above result a bit further thanks to the results
in [102]. If XH has p first integrals f1 = H, . . . , fp in involution and independent on
, including the Hamiltonian we have a set of m eigenvalues (k 1 among them,
corresponding to H) that belong to Table (19) and the normal variational equations,
NVE, are now n p of the initial variational equations. Reordering indexes if needed,
let us write them as VEp+1 , . . . , VEn with corresponding di↵erential Galois groups
Gp+1 , . . . , Gn and let us write the eigenvalues corresponding to f1 , . . . , fp as 1 =
k 1, . . . , p in Table (19). Then, if there is an additional first integral independent
of the set {f1 , . . . , fp }, the Galois group of the normal variational equations must
necessarily possess a rational invariant. Recently A. J. Maciejewski, M. Przybylska
and H. Yoshida proved the following:

Theorem 8 ([78]). Let XH be a Hamiltonian field given by (17). If there is at least

an additional single first integral f independent with {f1 , . . . , fp } on a neighborhood
of (but may be dependent on ), then we have one of the following two situations:

1. At least one of the eigenvalues 1, . . . , n p belongs to Table (19).

3.1 Non-integrability of Homogeneous Potentials 21

2. There are 1  i < j  n p such that

p q
(k 2)2 + 8k i (k 2)2 + 8k j 2 2kZ.

We will actually perform a step further and, as a by-product, obtain an alterna-

tive proof for Theorem 8.

Theorem 9 ([102]). Let XH be a Hamiltonian field given by (17). If there is at least

an additional single first integral f independent with {f1 , . . . , fp } on a neighborhood
of (but may be dependent on ), then we have one of the following two situations:

1. At least one of the eigenvalues 1, . . . , n p belongs to Table (19).

2. There exist 1  i < j  n p such that

p q
(k 2)2 + 8k i (k 2)2 + 8k j 2 2kZ. (20)

Moreover if we divide the set of eigenvalues { 1 , . . . , n p } in equivalence

classes, ⇤1 = { 1,1 , . . . , 1,k1 }, . . . ⇤r = { r,1 , . . . , r,kr }, { K+1 }, . . . , { n p },
with respect to the relation defined by (20) with k1 , k2 , . . . , kr all Pgreater than
1 (by reordering the eigenvalues we can assume this) and K := ri=1 ki . Then
XH can have at most 2K 3r additional meromorphic first integrals.

There is another extension of Theorem 7 to non-homogeneous potentials. Given

a potential which is not homogenous, but is expressed as a finite sum of homogeneous
potentials

V = Vk1 + . . . Vkm , (21)

being km the degree of homogeneity of the corresponding term, ki < ki+1 , it is

possible to apply Theorem 7 by using a transformation that traces back to the Levi-
Civita regularization. Therefore, the transformed system depends on a parameter
✏, which is essentially the inverse of the energy level. For the limit cases ✏ = 0, 1
we obtain systems with homogeneous potential Vk1 , Vkm . Then Mondéjar, using a
previous result about parametric Hamiltonian systems ([90]), was able to prove the
following result.

Theorem 10 ([91]). If the Hamiltonian system with potential (21) is completely

integrable with meromorphic first integrals, then both potentials Vk1 and Vkm must
satisfy the conditions (1)-(18) of Theorem 7.

It is worth noting that, in some sense, the above theorem generalizes previous
results for two–degrees–of–freedom systems obtained by Hietarinta and Yoshida [43,
146].
22 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

3.2 Homogenous Polynomial Potentials

A natural problem is to apply the table (19) to the study of the integrability of
Hamiltonian systems associated to polynomial potentials. So, in this section we
assume that the potential in (17) is polynomial of degree k. Although this problem
has already been studied by several authors such as Maciejewski, Nakagawa, Yoshida
and Przybylska, it remains a very active area of research. The final ambitious goal
is to obtain a complete classification of the integrable potentials. The strategy is as
follows:

(i) applying the results of the above section; this gives us some candidates to
integrability,

(ii) trying a direct integrability proof for the above candidates for integrability.

The classification is completed for n = 2 (two–degrees of freedom) and k = 3; it

is almost completed for n = 2 and k = 4 (we remark that the cases k = 1, 2 are
trivially integrable). Now we restrict to two-degrees of freedom.
In 2001–2002 Nakagawa and Yoshida obtained the list of all two– degrees of
freedom integrable homogeneous polynomial potentials such that the additional first
integral is polynomial of degree four in the momenta [104, 105]. In the proof they
used Ziglin’s lemma ([147]) around the homothetical solutions (i.e., something very
related to the Key Lemma in Section 2.1) and a previous study by Hietarinta by
means of the direct method, (ii) above [42, 43]. Then Maciejewski and Przybylska
studied the meromorphic non-integrability for k = 3, 4, [71, 74]. We follow [74].
If we restrict table (19) to natural numbers we obtain the following necessary
conditions for integrability.

k
1 k p + p (p 1) k2
2 3 1
24
1
+ 24 (2 + 6p)2
1 1 3 2
3 3 24
+ 24 2
+ 6p
1 1 6 2
4 3 24
+ 24 5
+ 6p
2
(22)
1 1 12
5 3 24
+ 24 5
+ 6p
1 2
6 4 8
+ 18 43 + 4p
9 1 10 2
7 5 40
+ 40 3
+ 10p
8 5 9
40
1
+ 40 (4 + 10p)2
1 k 1
9 k 2 k
+ p (p + 1) k
where we do not consider the quadratic potentials, since all of them are integrable,
i.e., from now on in this section k > 2.
For n = 2, we have only two eigenvalues 1 = k 1, 2 of V 00 (ci ), at a homoth-
etical point c = (c1 , c2 ). Denoting the non-trivial eigenvalue 2 := , by means of a
detailed algebraic analysis and using the residue theorem over the Riemann sphere
3.2 Homogenous Polynomial Potentials 23

along a suitable di↵erential form, the following remarkable global universal relation
between the several ’s for a given degree of the potential is proven.

Theorem 11 ([74]). Assume the polynomial homogeneous potential V (x1 , x2 ) has

k di↵erent homothetical points c1 , ..., ck with corresponding non-trivial eigenvalues
1 ,..., k . Then,
X k
1
= 1. (23)
i=1 i
1

For example, if we apply the above theorem, as well as table (22), to potentials
of degree k = 4 with four homothetical points, it is shown in [74] that the only
possible cases for { 1 , 2 , 3 , 4 } are as in table (24).

1 {0, 0, 3, 3}
2 { 38 , 6, 6, 6}
3 { 38 , 3, 21, 21} (24)
4 { 38 , 35
8 8
, 35 , 136}
5 { 38 , 3, 15, 36}
It is worth pointing out that, for generic homogeneous polynomial potentials,
the assumptions of Theorem 11 are satisfied, since for two–degrees of freedom, the
potential is defined in a natural way over the Riemann sphere P1 , and the exis-
tence of homothetical points is reduced to the search of solutions of some suitable
polynomials in one single variable ([74]).
Then, using our table (22) and Theorem 11, the authors obtained that under
the above assumptions the number of integrable potentials of a given degree must
be finite:

Theorem 12 ([74]). For a given degree k, the family of inequivalent homogenous

integrable polynomials which satisfy the assumption of Theorem 11 is finite.

For non-generic potentials of degree k the number of homothetical points is less

than k, but for k = 3 and k = 4 it is possible to (nearly) finish the classification and
reconstruct the possible integrable potentials; in particular, for some non-generic
families it is possible to generalize Theorem 11 in a suitable way. So, using some
constructive methods of algebraic geometry and an analysis of case by case it is
shown in [71] that for k = 3 there are no other integrable cases that those seen
already in references [42, 43]. For k = 4 the problem is more difficult, and the
classification of the integrable cases is not complete: integrability remains open for
the discrete infinite family of potentials
1 ↵ 1
V (x1 , x2 ) = x21 (x1 + ix2 )2 + (x21 + x22 )2 , (25)
2 4
where the parameter ↵ takes values in some discrete set (obtained by means of
the table (22)). It is worth remarking that in order to study the integrability of
24 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

the generic family given by table (24) (and a similar, simpler table for k = 3), an
essential tool was the use of the higher variational equations, i.e., Proposition 7 of
Appendix A.
For more than two degrees of freedom it is possible to generalize some of the
above results. In [115], the author was able to prove, for arbitrary n > 2, the
finiteness of the number of integrable homogeneous potentials of a fixed degree k
with a maximal number of homothetical points. A new method enters in the proof:
the Kovalewskaya exponents of an auxiliary gradient system in the configuration
space with the field given by the gradient of the potential. The equilibrium points
of the above gradient system are the homothetical points of the potential. It is
easy to prove that the Kowaleskaya exponents at the homothetical point ⇤1 , ..., ⇤n
coincide with shifted eigenvalues of the Hessian of the potential at this point c,
i = ⇤i + 1. Then a universal global relation which generalizes the equation (23)
is studied for potentials with a maximal number of homothetical points. From this
relation the author obtains the finiteness of the integrable potentials. See [115] for
details.

3.3 Some Rational Potentials

In a joint work of Simó with the first author the integrability of families of two-
degrees of freedom potentials with an invariant plane and normal variational equa-
tions of Lamé type ([93] it was studied, see also [94]). Under suitable assumptions
of regularity, it is easy to see that if the invariant plane is given by x2 = y2 = 0, the
potential should be of the form

x22
V (x1 , x2 ) = (x1 ) ↵(x1 )+ (x1 , x2 )x32 . (26)
2
The NVE associated to any integral curve lying on the invariant plane is

⇠¨ = ↵(x1 (t))⇠. (27)

It is clear that the first problem is to find the families of potentials with a given
(27). In the recent paper [2], assuming that (27) can be expressed with polyno-
mial coefficients, the authors completely solved this problem as stated in [93] in
an algorithmic way and applied it to the integrability of several families of rational
potentials. Here we review these results in [2].
From now on, we will write a(t) = ↵(x1 (t)), for a generic curve z = z(t) =
(x1 (t), y1 (t)) lying on the invariant plane and parameterized by t. Then, the NVE
is written
⇠¨ = a(t)⇠. (28)

3.3.1 The determination of the potentials

Problem. Assume that a(t) is a root of a given di↵erential polynomial Q(a, ȧ, ä, . . .) 2
C[a, ȧ, ä, . . .]. We want to compute all potentials in (26) satifiying such a condition.
3.3 Some Rational Potentials 25

So, we shall give a method to compute, for any given Q(a, ȧ, . . .), the family of
potentials with invariant plane x2 = y2 = 0 such that, for any integral curve lying
on this invariant plane, the coefficient a(t) of the NVE satisfies,

Q(a, ȧ, ä, . . .) = 0, (29)

by solving certain di↵erential equations.

We should notice that, for a generic integral curve z(t) = (x1 (t), y1 (t)), y1 (t) =
ẋ1 (t), lying on x2 = y2 = 0, equation (28) depends only on the values of functions ↵
and . It depends on ↵(x1 ), since a(t) = ↵(x1 (t)). We observe that the curve z(t)
is a solution of the restricted Hamiltonian,

y12
h= + (x1 ) (30)
2
whose associated Hamiltonian vector field is,

@ d @
Xh = y1 , (31)
@x1 dx1 @y1
d
thus x1 (t) is a solution of the di↵erential equation, ẍ1 = dx1
, and then, the relation
of x1 (t) is given by .
Since z(t) is an integral curve of Xh , for any function f (x1 , y1 ) defined on the
invariant plane x2 = y2 = 0 we have

d ⇤
z (f ) = z ⇤ (Xh f ),
dt
where z ⇤ denotes the usual pull–back of functions. Then, using a(t) = z ⇤ (↵), we
have for each k 0,
dk a
= z ⇤ (Xhk ↵), (32)
dtk
so that,
Q(a, ȧ, ä, . . .) = Q(z ⇤ (↵), z ⇤ (Xh ↵),⇤ (Xh2 ↵), . . .).
There is an integral curve of the Hamiltonian through each point of x2 = y2 = 0,
and thus we have the following.

Proposition 4. Let H be a Hamiltonian of the family (26), and Q(a, ȧ, ä, . . .) a
di↵erential polynomial with constant coefficients. Then, for each integral curve lying
on x2 = y2 = 0, the coefficient a(t) of the NVE (28) satisfies Q(a, ȧ, ä, . . . , ) = 0, if
and only if the function

Q̂(x1 , y1 ) = Q(↵, Xh ↵, Xh2 ↵, . . .),

vanishes on x2 = y2 = 0.
26 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

Now we see that Q̂(x1 , y1 ) is a polynomial in y1 and its coefficients are di↵er-
ential polynomials in ↵, . So, if we write down the expressions for successive Lie
derivatives of ↵, we obtain
d↵
Xh ↵ = y1 , (33)
dx1

d2 ↵ d d↵
Xh2 ↵ = y12 2
, (34)
dx1 dx1 dxi
✓ ✓ ◆ ◆
d3 ↵ d d d↵ d d2 ↵
Xh3 ↵ = y13 y1 +2 , (35)
dx31 dx1 dx1 dx1 dx1 dx21

✓ ✓ ✓ ◆ ◆ ◆
d4 ↵ d d d d↵ d d2 ↵ d3 ↵ d
Xh4 ↵ = y14 4 y12 +2 +3 3 +
dx1 dx1 dx1 dx1 dx1 dx1 dx21 dx1 dx1
✓ ✓ ◆ ◆
d d d↵ d d2 ↵ d
+ +2 . (36)
dx1 dx1 dx1 dx1 dx21 dx1
In general form we have,

@Xhn ↵ d @Xhn
Xhn+1 ↵ = y1 , (37)
dx1 dx1 @y1

and it inductively follows that they all are polynomial in y1 , its coefficients being
di↵erential polynomials in ↵, . If we write it down explicitly,
X
Xhn ↵ = En,k (↵, )y1k , (38)
n k 0

h r s
i
we can see that the coefficients En,k (↵, ) 2 C ↵, , ddx↵r , ddxs , satisfy the following
1 1
recurrence law,

d d
En+1,k (↵, ) = En,k 1 (↵, ) (k + 1)En,k+1 (↵, ) (39)
dx1 dx1

with initial conditions,

d↵
E1,1 (↵, ) = , E1,k (↵, ) = 0 8k 6= 1. (40)
dx1

We observe that the recurrence law (39) and initial conditions (40) determine
the coefficients En,k (↵, ). We can compute the value of some of them easily:
dn ↵
• En,n (↵, ) = dxn
for all n 1.
1

• En,k (↵, ) = 0 if n k is odd, or k < 0, or k > n.

3.3 Some Rational Potentials 27

As an illustration of the above method we now compute families of potentials

(26) associated to a specific well–known NVE. Although, in order to perform these
computations, we need to solve polynomial di↵erential equations, we will see that
we can deal with this in a series of cases. Particularly, when Q is a di↵erential
linear operator, we will obtain equations involving products of few linear di↵erential
operators.
Example 1: NVE of harmonic oscillator type. Harmonic oscillator equation
is
⇠¨ = c0 ⇠, (41)
with c0 constant. Then, a Hamiltonian of type (26) gives such NVE if ȧ = 0.
d↵
Looking at formula (33), it follows that dx 1
= 0, so that ↵ is a constant. We
conclude that the general form of a Hamiltonian (26) which gives rise to NVE of
the type (41) is,
y12 + y22 2
H= + (x1 ) + 0 x2 + (x1 , x2 )x32 ,
2
0 being a constant, and , arbitrary analytical functions.
Example 2: NVE of Airy type. In [8], Audin notices that the Hamiltonian,
y12 + y22
+ x1 x22
2
gives an example of a simple non-integrable classical Hamiltonian, since its NVE
along any integral curve in the invariant plane x2 = y2 = 0 is an Airy equation.
Here we compute the family of classical Hamiltonians that have NVE of type Airy
for integral curves lying on the above invariant plane. General form of Airy equation
is
⇠¨ = (c0 + c1 t)⇠ (42)
with c0 , c1 6= 0 two constants. If follows that a Hamiltonian gives rise to NVE of
this type if ä = 0, and ȧ 6= 0. The equation ä = 0 gives, by Proposition 4 as we see
in formula (34), the following system:
d2 ↵ d d↵
= 0, = 0. (43)
dx21 dx1 dx1
It splits into two independent systems,
(
d2 ↵
d↵ dx21
=0
= 0, d (44)
dx1 dx1
=0

Solutions of the first one fall into the previous case of harmonic oscillator. Then,
taking the general solution of the second system, we conclude that the general form
of a classical Hamiltonian of type (26) with Airy NVE is:
y12 + y22 2 2
H= + 0 + 1 x2 + 2 x1 x2 + (x1 , x2 )x32 , (45)
2
28 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

with 2 6= 0.
Example 3: NVE of quantum harmonic oscillator type. Let us now consider
3 2
now equations with ddt3a = 0, and ddt2a 6= 0, it is

⇠¨ = (c0 + c1 t + c2 t2 )⇠ (46)

with c2 6= 0. Those equations can be reduced to a quantum harmonic oscillator

equation by an affine change of t. Using Proposition 4 and formula (35), we obtain
the following system of di↵erential equations for ↵ and :

d3 ↵ d↵ d2 d2 ↵ d
= 0, + 3 = 0.
dx31 dx1 dx21 dx21 dx1
The general solution of the first equation is

1 2 3
↵= + x1 + x21 ,
2 2 2
and substituting it into the second equation we obtain a linear di↵erential equation
for ,
d2 2 3 d
2
+3 = 0,
dx1 2 + 2 3 x1 dx1

this equation is integrated by two quadratures, and its general solution is

4
= + 0.
( 2 + 2 3 x1 )2

We conclude that the general formula for Hamiltonians of type (26) with NVE (46)
for any integral curve lying on x2 = y2 = 0 is

y12 + y22
H= +
2
4 2 2 2 2
2
+ 0 1 x2 2 x1 x2 3 x1 x2 + (x1 , x2 )x32 , (47)
( 2 + 2 3 x1 )

with 3 6= 0.
We observe that formula (47) yields non-linear dynamics in the invariant plane
x2 = y2 = 0. Notice that these dynamics are continuously deformed to linear
dynamics when 4 tends to zero. In the general case, for a fixed energy h, we have
the general integral of the equation:

8 23 h2 (t t0 )2 = h( 2 + 2 3 x1 )2 4.

Example 4: NVE with polynomial coefficient a(t) of odd degree. Here we

generalize Example 2. Let us consider for n > 0 the following di↵erential polynomial,
dm a
Qm (a, ȧ, . . .) = .
dtm
3.3 Some Rational Potentials 29

It is obvious that a(t) is polynomial of degree n if and only if Qn (a, ȧ, . . .) 6= 0 and
Qn+1 (a, ȧ, . . .) = 0.
Looking a Proposition 4, we see that a Hamiltonian (26) has NVE along a generic
integral curve lying on x2 = y2 = 0,

⇠¨ = Pn (t)⇠, (48)

where Pn (t) polynomial of degree n, if and only if Xhn ↵ 6= 0 and Xhn+1 ↵ vanishes on
x2 = y2 = 0. Let us remind expression (38), Xhn+1 ↵ vanish in x2 = y2 = 0 if and
only if (↵, ) is a solution of the di↵erential system

Rn+1 = {En+1,0 (↵, ) = 0, . . . , En+1,n+1 (↵, ) = 0}.

A particular solution of Rn+1 which does not satisfy Rn , is given by = 0,

↵(x1 ) = Qn (x1 ), polynomial of degree n. Then the Hamiltonians,

y12 + y22
H= + 0 + Qn (x1 )x22 + (x1 , x2 )x32 , (49)
2
have NVE, along a generic integral curve lying on x2 = y2 = 0, of the form (48).
If n is an even number, there are more solutions of the di↵erential system Rn+1
not verifying Rn , being a particular case the potentials with generic quantum har-
monic oscillators, computed above. We will prove, using the recurrence law (39),
that for odd n, the above family is the only solution of Rn+1 not verifying Rn .
d
Lemma 4. Let (↵, ) be a solution of R2m . Then, if dx1
6= 0, then (↵, ) is a
solution of R2m 1 .

Proof. As 2m 1 is odd, E2m 1,2k (↵, ) = 0 for all m 1 k 0. Then let

us prove that E2m 1,2k+1 (↵, ) = 0 for all m 2 k 0.
In the first step of the recurrence law defining R2m ,
dE2m 1,1 d
0 = E2m,0 (↵, ) = (↵, ) E2m 1,1 (↵, ),
dx1 dx1
d
we use dx1
6= 0, and E2m 1, 1 (↵, ) = 0 to obtain

E2m 1,1 (↵, ) = 0.

If we assume E2m 1,2k+1 ( , ↵) = 0, substituting it in the recurrence law

dE2m 1,2k d
E2m,2k+1 (↵, ) = (↵, ) 2(k + 1) E2m 1,2(k+1) (↵, ),
dx1 dx1
we obtain that
E2m 1,2(k+1) (↵, ) = 0,
and we conclude by finite induction. 2
30 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

Corollary 1. Let H be a classical Hamiltonian of type (26). Then the following

statements are equivalent,

1. The NVE for generic integral curve (28) lying on x2 = y2 = 0 has polynomial
coefficient a(t) of degree 2m 1.

2. H can be written
y12 + y22
H= + 0 P2m 1 (x1 )x22 + (x1 , x2 )x32 , (50)
2
for 0 constant and P2m 1 (x1 ) polynomial of degree 2m 1.

Proof. It is clear that condition 1 is satisfied if and only if (↵, ) is a solution

of R2m and it is not a solution of R2m 1 . By the previous Lemma, this implies
d 2m
dx1
= 0, and the system R2m is thereby reduced to ddx2m↵ = 0 and then, is therefore
1
a constant and ↵ must be a polynomial of degree at most 2m 1. 2

Example 5: NVE of Mathieu type. The standard Mathieu equation is

⇠¨ = (c0 + c1 cos(!t))⇠, ! 6= 0. (51)

We cannot apply our method to compute the family of Hamiltonians correspond-

ing to this equation, because {c0 + c1 cos(!t)} is not the general solution of any
di↵erential polynomial with constant coefficients. But let us consider

d3 a da
Q(a) = 3
+ !2 , (52)
dt dt
the general solution of {Q(a) = 0} is

a(t) = c0 + c1 cos(!t) + c2 sin(!t).

Just notice that

q ✓ ◆
2 2 c2
c1 cos(!t) + c2 sin(!t) = c1 + c2 cos !t + arctan ,
c1

thus NVE (28), when a is a solution of (52), is reducible to Mathieu equation (51)
by a translation of time.
Using Proposition 4, we find the system of di↵erential equations that determine
the family of Hamiltonians,

d3 ↵ d↵ d2 d2 ↵ d d↵
= 0, +3 !2 = 0.
dx31 2
dx1 dx1 dx1 dx1 dx1
The general solution of the first equation is
2
↵= 0 + 1 x1 + 2 x1 .
3.3 Some Rational Potentials 31

d
substituting it in the second equation, and writing y = dx1
, we obtain a non–
homogeneous linear di↵erential equation for y,
dy 6 2y
+ = !2. (53)
dx1 1 + 2 2 x1

We must distinguish two cases depending on the parameter. If 2 = 0, then we just

integrate the equation by trivial quadratures, obtaining
! 2 x21
= µ0 + µ1 x 1 +
2
and then,
y12 + y22 ! 2 x21 2 2
H= + µ0 + µ1 x 1 + 0 x2 1 x1 x2 + (x1 , x2 )x32 , (54)
2 2
If 2 6= 0, we can then reduce the equation to separable using
6 2y
u= ,
1 + 2 2 x1

obtaining
3du 6 2 dx 3! 2 3µ1
= , u= + ,
3! 2 4u 1 + 2 2 x1 4 4( 1 + 2 2 x1 )4
and then ✓ ◆
1 2 2 µ1
y= ! 1 + 2! 2 x1 + ,
8 2 ( 1 + 2 2 x1 )3
and finally we integrate it to obtain ,
Z
µ1 1 ! 2 1 x1 ! 2 x21
= ydx1 = µ0 + + ,
32 22 ( 1 + 2 2 x1 )
2 8 2 8
scaling the parameters adequately we write down the general formula for the Hamil-
tonian,
y 2 + y22 µ1 2
1 ! x1 ! 2 x21
H= 1 + µ0 + + + +
2 ( 1 + 2 2 x1 )2 8 2 8
2 2 2 2
0 x2 1 x1 x2 2 x1 x2 + (x1 , x2 )x32 . (55)

3.3.2 Non-integrability
One of the main results in [2] is the following.
Theorem 13 ([2]). The Galois group of the equation,

⇠¨ = Q(x)⇠,

with Q(x) a non-constant polynomial of degree k, with coefficient field K = C(x),

is a connected non-commutative group. In fact, the Galois group falls in either one
of the following cases:
32 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

1. Case 4 of Proposition 2 for k = 2n even and provided some concrete algebraic

relations in the coefficients of Q are satisfied.

2. SL(2, C) (case 7 of Proposition 2).

The proof of this theorem is given by means of a method of completion of squares

for the polynomial Q and by using the Kovacic algorithm (Appendix A, Section A.1).
From Theorem 13 and Theorem 5 it one obtains the following non-integrability
result.

Theorem 14 ([2]). The Hamiltonian systems (47) with 3 6= 0, (49) with n 1,

(54) with 1 6= 0 and (55) with ( 1 , 2 ) 6= (0, 0) are not integrable by means of
rational first integrals.

For further details see the original paper [2].

3.4 Celestial Mechanical Problems

Here we survey some recent results about the non integrability of some Celestial
Mechanical problems.

3.4.1 Some N -Body Problems

Using the Ziglin approach along the triangular parabolic solution of Lagrange,
A. V. Tsygvintsev ([130, 131, 132, 133, 134, 135]) proved the meromorphic non-
integrability of the Three-Body Problem and ultimately settled the non-existence of
a single meromorphic first integral; he established both things for all except three spe-
cial cases. On the other hand, using Theorem 4 also along the triangular parabolic
solution of Lagrange and by means of a result about the non-commutativity of
the identity component of the Galois group in presence of logarithmic terms for
completely reducible variational equations (Appendix A, Section A.3), Boucher and
Weil ([15, 16, 18]) also proved the meromorphic non-integrability of the Three-Body
Problem. It is finally worth noting that Ziglin ([150]) managed to settle strong con-
ditions on the integrability of the Three-Body Problem and the equal-mass N -Body
Problem.
In the already cited joint work of the first author with Simon ([102]) we are
reobtaining in simpler ways, strengthening and generalizing the results mentioned
in the previous paragraph. The proof is based on Theorems 9 and 7, i.e., we use
homothetical solutions. We follow [102].
The Hamiltonian of the general N -Body Problem in dimension d is

1
HN,d (x, y) := y T M 1
y + UN,d (x) , (56)
2
defining
M = diag (m1 , . . . , m1 , · · · , mN , . . . , mN ) 2 M at (N d, R) ,
3.4 Celestial Mechanical Problems 33

and assembling the coordinates of our phase space among the N d-dimensional vec-
tors

x (t) = (xi (t))i=1,...,N , y (t) = (yi (t))i=1,...,N := (mi ẋi (t))i=1,...,N

of positions and momenta, respectively. The Newtonian gravitational potential is

X mi mk
UN,d (x) := .
1i<kN
kxi xk k

The solutions of the equation

UN0 d (x) = M x, (57)

where > 0, are called central configurations. If the bodies are released with zero
initial velocity, with initial conditions at a (real) solution x of (57), this defines a
homothetical solution of the N -Body Problem.
0
We remark that we can normalize to one. Indeed, the 2 -homogeneity of UN,d
0 2↵ 0 0
assures us UN,d ( ↵ x) = UN,d (x); thus, assuming UN,d (x) = M x, defining
0
x̃ = x and asking for UN,d (x̃) = M x̃ to hold, we obtain ↵ = 1.
The problem of computing central configurations (i.e., solving the system of
algebraic equations (57), where we can assume = 1) is an old difficult classical
problem in Celestial Mechanics. In fact, it is only solved in complete generality for
N = 3, thanks to Euler (collinear central configurations: the masses are on a line)
and Lagrange (triangular central configurations: the masses are at the vertexes of an
equilateral triangle). For more information on central configurations see the article
[89].
From symmetry considerations, it is clear that whenever the masses are equal,
regular N -polygons with the masses at the vertexes give rise to homothetical so-
lutions, i.e., if the masses start with zero velocity from a such configuration, they
remain at a regular polygon. These are central configurations ([109]).
The connection of our work with homothetical solutions, and hence with central
configurations, is due to the following. For the N -Body Problem the real homo-
thetical points in Section 3.1 are central configurations and the particular integral
curves considered there are homothetical solutions. Indeed, a symplectic change
x = M 1/2 q, y = M 1/2 p renders HN,d a classical Hamiltonian HN,d = 12 p2 + VN,d (q)
with a potential which is homogeneous of degree 1, VN,d = UN,d (M 1/2 q). Since
0 0 0
M 1/2 VN,d (q) = UN,d M 1/2 q and thus UN,d (x) = M x (for x = M 1/2 q) is equiv-
alent to
0
VN,d (q) = M 1/2 M M 1/2 q = q.
Thus, we can consider the homothetical points in an N -Body Problem in Celestial
Mechanics as complex central configurations and the associated particular solution
considered in Section 3.1 as an homothetical solution; this justify our terminology.
In virtue of Theorem 7, performing the following two steps would prove HN,d
not meromorphically integrable:
 34 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

Step I either explicitly finding or proving the existence of an adequate constant vector
c 2 C2N such that
0
VN,d (c) = c; (58)
00
Assume VN,d (c) is diagonalizable.
00
Step II proving that at least one of the eigenvalues of VN,d (c) does not belong to
the set given by items 1 and 18 in Table (19), which happens to be a set of
integers:
⇢ ⇢
p (p 3) (p + 2) (p 1)
S := :p2Z = : p 2 Z ⇢ Z, (59)
2 2
whose symmetry allows for the assumption p > 1; the size of the consecutive
gaps in this discrete set is strictly increasing, as is seen in its first elements:
{1, 0, 2, 5, 9, 14, 20, 27, 35, . . .}.
In virtue of Theorem 9, isolating an adequate set of eigenvalues and performing
the following third step would be enough to set a very precise upper bound on the
amount of additional meromorphic integrals:
Step III proving that, except for a set S̃ of notable eigenvalues corresponding to the
00
set of classical first integrals, there is no other eigenvalue of VN,d (c) in S.
And in virtue of either Theorem 8 or Theorem 9, the following fourth step would
be enough to discard the existence of even a single additional meromorphic integral:
00
Step IV performing Step III and proving that, except for said notable set S̃, Spec VN,d (c) \
S̃ consists exclusively of eigenvalues not satisfying relation (20) pairwise.
We are performing steps I–IV for the Three Body Problem with arbitrary masses,
steps I–III for the N-Body Problem with equal masses, N = 4, 5, 6, as well as steps
I and II for the N-Body Problem with equal masses with N 3. In all cases we
consider the planar case d = 2, although the proof for N = 3 can be very easily
established regardless of the dimension d 2 (see [102] for details).
a) Three Body Problem. Step I is computing a solution c of (58) for N = 3. Let
us define m = m1 + m2 + m3 (which may be always set to 1 by the reader if even
simpler calculations are sought all through this section) and D = m1 m2 + m2 m3 +
m3 m1 , and consider vectors of the form c = m 2/3 M 1/2 ĉ, where M = (mi Idd )i=1,...,N
and 0 1
a2 m2 + a3 m3
B b2 m 2 + b3 m 3 C
B C
B a3 m3 a2 (m1 + m3 ) C
ĉ = B
B b3 m3 b2 (m1 + m3 ) C
C (60)
B C
@ a2 m2 a3 (m1 + m2 ) A
b2 m2 b3 (m1 + m2 )
and a2 , a3 , b2 , b3 are solutions to
3/2 3/2 ⇥ ⇤3/2
a22 + b22 = a23 + b23 = (a2 a3 )2 + (b2 b3 )2 = 1.
3.4 Celestial Mechanical Problems 35

An example of such a vector ĉ is

0 1
(m2 + 2m3 ) ↵
B m2 C
B C
B (m1 m3 ) ↵ C
ĉ = B
B
C,
C (61)
B (m1 + m3 ) C
@ (2m1 + m2 ) ↵ A
m2

where ↵2 + 2 = 1 and ↵3 = 1/8. The possible choices of ↵ and add up to two such
vectors as (61), and thus two solutions c = m 2/3 M 1/2 p
ĉ and c⇤ = m 2/3 M 1/2 ĉ⇤ for
⇤ 1+i 3
(58): those corresponding to ↵ = 1/2 and ↵ = 4
, respectively; where square
roots are taken in their principal determination. A simple, if tedious computation
proves c and c⇤ solutions to (58), indeed. In fact, c yields an explicit parametrization
for the (homothetical) Lagrange triangular solution where the three masses start at
rest on the vertex of an equilateral triangle.
The rest of the proof is based on performing both Steps II and III at a time.
The eigenvalues of V300 (c) are { 2, 0, 0, 1, + , }, where
p
1 3 m21 + m22 + m23 m1 m2 m1 m3 m2 m3
± := ± .
2 2 (m1 + m2 + m3 )

It is not difficult to prove that the eigenvalues 2, 0, 0, 1 correspond to the classical

first integrals: the energy, the two components of the linear momentum and the
angular momentum (see [102]). Then as said in Theorem 9, one of the necessary
conditions for the existence of a single additional meromorphic integral for XH3
1
implies either ⇤+ 2 S or ⇤ 2 S, where S = p (p 3) : p > 1 , which means
p 2
2
(defining R := m 2 3D) that ±3R 2 {(p 3p 1) m : p > 1} and therefore

27 (m1 m2 + m1 m3 + m2 m3 ) 2 m2 (p 1) (p 2) (p 4) (p + 1) : p > 1 . (62)

This is impossible if p 2 {2, 4} or p > 4, since it would have a strictly negative

number equaling a non-negative one. For p = 3 (62) becomes 8m2 = 27D, that is,

m1 m2 + m1 m3 + m2 m3 8
2 = . (63)
(m1 + m2 + m3 ) 27
p
1
The eigenvalues of V300 (c⇤ ) are 2, 0, 0, 1, ⇤
+,
⇤
, where ⇤
± = 2
± 23p2m
A
, and
p
A = 2m21 + 2m22 + 2m23 5m1 m2 5m2 m3 + 7m1 m3 i 3(m1 m2 + m2 m3 5m1 m3 ).

Again, the thesis

p amounts to either ⇤+ 2 S or ⇤ 2 S, which here
in Theorem 9 p
becomes ±3 A = (p2 3p 1) 2m, and thus
⇢
2
A 2m2 2 (p 1) (p 2) (p 4) (p + 1) m2 : p > 1 ;
9
36 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

a necessary condition for this to hold with real masses is the vanishing of the imag-
inary term in A, p
i 3 (m1 m2 + m2 m3 5m1 m3 ) = 0, (64)
implying m1 m2 + m2 m3 = 5m1 m3 . Thus,

378m1 m3 = 2 (p 1) (p 2) (p 4) (p + 1) m2 , (65)

for some p > 1. We discard p = 2, 4 in (65) assuming the strict positiveness of

m1 and m3 . The only integer p > 1 for which the right side can be negative is 3,
implying 378m1 m3 = 16 (m1 + m2 + m3 )2 . These two constraints arising from
(64) and (65),

189
5m1 m3 = m1 m2 + m2 m3 , m1 m3 = (m1 + m2 + m3 )2 , (66)
8
cannot hold at the same time as condition (63). Indeed, the former two substituted
into the latter would yield (5m189
1 m3 +m1 m3 )
m1 m3
8
= 27 , i.e. 16
63
8
= 27 which is obviously absurd.
8
Thus, either (63) holds or both equations in (66) hold.
Let us now prove that V3 does not psatisfy the p remaining thesis in said Theorem.
The di↵erence in (20), E ( i , j ) = 9 8 j 9 8 i /2, will be studied both
for Spec (V300 (c⇤ )) and Spec (V300 (c)). Let
1/2 1
a := m21 + m22 + m23 m1 m2 m1 m3 m2 m3 (m1 + m2 + m3 ) 0.

The only case worth considering for the real eigenvalues is

p p
13 + 12a 13 12a
E ( +, ) = ,
2
⇥ ⇤
which is real only if a 2 0, 1312
. In this interval, moreover,
p the only possible integer
2
values of E ( + , ) are 0, 1, 2. Note that a = 1 3Q, where p Q = D/m =
2
(m1 m2 + m1 m3 + m2 m3 ) (m1 + m2 + m3 ) . The solutions to 1 3Q = n for
n = 0, 1, 2 are, respectively, Q = 1/3, 0, 1, among which the only possible value for
Q is 1/3. Hence, E ( + , ) can only be real if a = 0, i.e. p
Q = 1/3.
1 3 a⇤
Now consider the complex eigenvalues ± = 2 ± 2 of V300 (c⇤ ). Since
⇤

0q q 1
12 ⇤ 12 ⇤
p 1+ 13
a 1 13
a
E ⇤
+,
⇤
= 13 @ A,
2

it is enough to prove that (a⇤ )2 is always never real when Q = 1/3. Indeed, if z = z1 +
p p p p 2
z2 i with z1 z2 6= 0, then 1 + z 1 z is always complex: 1+z 1 z =
p p
2 2 1 z 2 and since z 2 is non-real, so is 2 2 1 z 2 .
In order to prove a⇤ , (a⇤ )2 2 R \ C, we will see that the imaginary term inside
the square root, 5m1 m3 + m2 m1 + m2 m3 , is always nonzero if Q = 13 . Indeed,
3.4 Celestial Mechanical Problems 37

5m1 m3 +m1 m3
otherwise (m1 +m2 +m3 )2
= 13 , i.e. 16m1 m3 m21 2m2 m1 m22 2m2 m3 m23 = 0;
5m1 m3
from 5m1 m3 = m2 m1 + m2 m3 , we also deduce m2 = m1 +m3
and therefore

4m31 m3 15m21 m23 + 4m1 m33 m41 m43

16m1 m3 m21 2m2 m1 m22 2m2 m3 m23 = = 0,
(m1 + m3 )2

and the only values of m3 allowing this are

p p
(2 + 3i) ± (1 + 2i) 3 (2 3i) ± (1 2i) 3
m1 m1 ,
2 2
which are obviously not positive real numbers. The lack of an additional meromor-
phic first integral for arbitrary m1 , m2 , m3 > 0 is thus proven in the planar case.
Hence, we have proven.

Theorem 15 ([102]). For the planar Three Body Problem, there is no additional
meromorphic first integral with arbitrary positive masses which is independent with
the classical first integrals.

It is worth noting that the two cases forcing us to resort to a second solution to
(58) are precisely two of the three cases exceptional to Tsygvintsev’s proof ([135]):
⇢
D 1 23 2
2
2 , , . (67)
m 3 33 32

b) N -Body Problem with equal masses. Here we consider the integrability

problem of N equal masses in a plane. Defining q = (q1 , . . . , qN ) (qi = (q2i 1 , q2i )),
we have
X n
@VN,2 p
= mk (mi mk )3/2 Di,k3 Di,k , i = 1, . . . , N, (68)
@qi k=1,k6=i

T p p
where Di,j = (d2i 1,2j 1 , d2i,2j ) := mj qi
mi qj for each i, j = 1, . . . , N , and
⇣ ⌘
00
we obtain the block expression for the Hessian matrix: VN,2 (q) = Ũi,j ,
i,j=1,...,N
defining ⇢ p
P mi mj Ui,j , i 6= j,
Ũi,j := (69)
k6=i mk Ui,k , i=j
where (
02⇥2 , i = j,
Ui,j = Uj,i = 5/2 (70)
(mi mj )3/2 d22i 1,2j 1 + d22i,2j Si,j , i < j,
being 02⇥2 the zero square matrix of dimension two and
✓ 2 ◆
d2i,2j 2d22i 1,2j 1 3d2i 1,2j 1 d2i,2j
Si,j = Sj,i := , i 6= j. (71)
3d2i 1,2j 1 d2i,2j d22i 1,2j 1 2d22i,2j

For simplicity, we will denote VN := VN,2 from now on.

38 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

Since every choice of mass units amounts to a symplectic change in the extended
phase space, we may set m1 = · · · = mN = 1. Expressions (68) and (69) may be
found explicitly in terms of trigonometric functions if we choose the regular central
polygonal configuration, where each of the masses are at the vertex of a regular
polygon, as a solution to (58). Define

⇡k ⇡k
sk := sin , ck := cos , k 2 N,
N N
2⇡i
and ⇣ = e N = c2 + is2 .
1/3
Lemma 5. Vector cP = (c1 , . . . , cN ) defined by cj = N (c2j , s2j ), where N =
1
PN 1 ⇡k 0
4 k=1 csc N , is a solution for VN (q) = q.

Proof. Indeed, assume cj = A cos 2⇡j

N
, sin 2⇡j
N
for some A > 0. We have
0 P 2⇡j
1
N 1 cos N
@VN 1 @ k=1 sin N k A
⇡
(cP ) = PN 1 sin 2⇡j ,
@qj 4A2 ⇡
N
k=1 sin N
k

due to the fact that

N
X N 1X 1 (c2k + is2k )
⇣j ⇣k j
= ⇣ ,
k=1,k6=j
|⇣ j ⇣ k |3 k=1
|1 ⇣ k |3

and, since the imaginary part of this sum satisfies:

N
X1 N
X1 2sk ck N 1
s2k 1 X ck
= = = 0,
k=1
|1 ⇣ k |3 k=1
8c3k 4 k=1 s2k
P 1 1 (c2k +is2k ) P 1 1
we finally obtain ⇣ j N
k=1 3 = 14 ⇣ j N 0
k=1 sk . Now V (cP ) = cP if and only
|1 ⇣ |
k
P 1 1 1/3
if N k=1 4A2 sk = A. The latter holds for A = N .

Let us see how this specific vector simplifies VN00 . Keeping expression (69) in
1/3
consideration we have d2i 1,2j 1 + id2i,2j = N (⇣ i ⇣ j ) which implies
⇣ ⌘2 ✓ 3c 1 3s2(i+j)
◆
1/3 2(i+j)
Si,j = 2 N si j ,
3s2(i+j) 3c2(i+j) 1

for each 1  i, j  N , and thus

Ui,i = 02⇥2 , i = 1, . . . , N,
⇣ ⌘ 5
1/3
Ui,j = Uj,i = 2 N si j Si,j
✓ ◆
|si j | 3 3c2(i+j) 1 3s2(i+j)
= , i 6= j,
16 N 3s2(i+j) 3c2(i+j) 1
3.4 Celestial Mechanical Problems 39

from which defining

X |si j | 3 ✓ 3c2(i+j) 1 3s2(i+j)
◆
Ũi,i = ,
16 N 3s2(i+j) 3c2(i+j) 1
j6=i
✓ ◆
|si j | 3 1 3c2(i+j) 3s2(i+j)
Ũi,j = , i=6 j,
16 N 3s2(i+j) 3c2(i+j) + 1
⇣ ⌘
we have VN00 (cP ) = Ũi,j .
i,j=1,...,N
PN 1 ⇡k
Lemma 6. The trace for VN00 (cP ) is equal to (N/8) (↵N / N ), where ↵N = k=1 csc3 N
and N is defined as in Lemma 5.
Proof. In virtue of the above simplifications for (69), tr (VN00 (cP )) is equal to
2 X 3
µN := ⇣ 2k1 ⇣ 2k2 .
N
1k1 <k2 N
PN P 3
We have µ4N k=11 csc ⇡k N
= 1k1 <k2 N 2 ⇣ 2k1 ⇣ 2k2 ; on the other hand, the
symmetry of a regular polygon assures
X N
X1
3 3
2 |2sk2 k1 | =N (2sk ) ;
1k1 <k2 N k=1
PN 1 ⇡k
PN 1 ⇡k
thus, 2µN k=1 csc N
= N k=1 csc3 N
.

Case 1: N = 3, 4, 5, 6.
We can a↵ord obtain a result stronger than non-integrability for these values
without using Lemma 6. We just have to prove the following
Lemma 7. VN00 (cP ), N = 3, 4, 5, 6, has only four eigenvalues in S: 1 = 2, 2 =
3 = 0, 4 = 1.Furthermore, the sets of equivalence classes given by relation E ( i , j ) 2
Z in (20) with cardinality greater than one are (assuming j > 4):
1. a double eigenvalue for N = 3, 4;
2. three double eigenvalues for N = 5, 6.
Proof. The eigenvalues of V300 (cP ) are p 1 , 2 , 3 , 4 pand 5,6 = 1/2. Those of
2(5 3 2) 2( 2 4) p
6 2 17
V400 (cP ) are 1 , 2 , 3 , 4 and 5 = 7
, 6,7 = 7
, 8 = 7
. The corre-
sponding relations are
p p ! q q
2 5 3 2 2 2 4 1 p 1 p
E , = 119 + 336 2 + 889 112 2,
7 7 14 14
p p ! q q
6 2 17 2 2 4 1 p 1 p
E , = 1393 336 2 + 889 112 2,
7 7 14 14
p p ! q q
6 2 17 2 5 3 2 1 p 1 p
E , = 1393 336 2 + 119 + 336 2.
7 7 14 14
40 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

V500 (cP ) has six di↵erent non-trivial double eigenvalues:

p p p p
5 5 ± 518 222 5 5 4
5,6,7,8 = , 9,10 = .
4 2
Relations are
p p p p p p p p p p
19 2 5 + 6 37 2 37 5 19 2 5 6 37 + 2 37 5
E( 5,6 , 7,8 ) = ,
p p 2
p p p p
25 4 5 19 2 5 6 37 + 2 185
E( 5,6 , 9,10 ) = ,
p p 2
p p p p
25 4 5 19 2 5 + 6 37 2 185
E( 7,8 , 9,10 ) = .
2

The eight non-trivial eigenvalues for V600 (cP ) are

p p p p
4 29 3 94 34 3 133465 59584 3 157
5 = , 6,7 = ,
59 p 118
p p p
2 7 3 41 34 3 + 133465 59584 3 157
8,9 = , 10,11 = ,
59 p 118
4 53 22 3
12 = .
59
The relations are
p
s1 + 236s2 s3 s4 s3
E ( 5, 6,7 ) = , E ( 5 , 8,9 ) = ,
p 118 118
s1 236s2 s3 s5 s3
E ( 5 , 10,11 ) = , E ( 5 , 12 ) = ,
p 118 118
s1 236s2 s4 s5 s4
E ( 8,9 , 10,11 ) = , E ( 8.9 , 12 ) = ,
p118 118p
s4 s1 + 236s2 s5 s1 + 236s2
E ( 6,7 , 8,9 ) = , E ( 6,7 , 12 ) = ,
p 118 118
s5 s1 236s2
E ( 10,11 , 12 ) = ,
p 118 p
s1 236s2 s1 + 236s2
E ( 6,7 , 10,11 ) = ,
118
p p p p p
with p
s1 = 68381 8024 3, s2p = 133465 59584 3, s3 = 208801 54752 3,
p p
s4 = 70033 6608 3, s5 = 68735 + 41536 3.
Let us now determine an upper bound for the amount of meromorphic first
integrals for the equal–mass Problem.
1. For N = 3, we have ⇤1 = { 5,6 } = {1/2}. Then K = 2, r = 1, which by
Theorem 9 in an absence of further information would allow the existence of
an additional meromorphic first integral. That possibility, however, is ruled
out by Theorem 15.
3.4 Celestial Mechanical Problems 41

2. For N = 4, we have ⇤1 = { 6,7 } and two simple eigenvalues: { 5 } and { 8 }:

Thus K = 2, r = 1 and there can be at most one additional meromorphic first
integral.

3. For N = 5, we have ⇤1 = { 5,6 }, ⇤2 = { 7,8 }, ⇤3 = { 9,10 }. Then K = 6 and

r = 3. Hence, there may be at most three additional meromorphic integrals.

4. For N = 6, we have ⇤1 = { 6,7 }, ⇤2 = { 8,9 }, ⇤3 = { 10,11 }, and two simple

eigenvalues { 5 } and { 12 }. As for N = 5, we have at most three additional
meromorphic integrals.

Case 2: N = 7, 8, 9.
Proceeding from Lemma 6, it is straightforward to see the traces for VN00 (c) for
these three values of N are non-integers since
q p p
413 + 56 7 cos 13 arctan 3 3
µ7 = ⇣ p ⌘ 2 ( 12, 11) ,
2 cos 16 arctan 3133
⇣ p p p ⌘
4 2633 + 766 2 + 4 118010 68287 2
µ8 = 2 ( 17, 16) ,
p
241
9 8 9 3 + csc3 ⇡9 + csc3 2⇡
9
+ csc3 4⇡
9
µ9 = p 2 ( 22, 21) .
2 2 3 + csc ⇡ + csc 2⇡ + csc 4⇡
3 9 9 9

Case 3: N 10
We will prove VN00 (cP ) has at least an eigenvalue greater than 1. We know the
following holds ([1]),
k 1
1 1 X ( 1) 2 22k 1 1 B2k x2k 1
csc x = + f (x) := + , (72)
x x k 1 (2k)!

f being analytical for |x| < ⇡ (which obviously holds if x = ⇡j

N
, j = 1, . . . , N 1)
and Bk , k 1, being the Bernoulli numbers ([1, Chapter 23]).
P 1
Lemma 8. For each N 10, SN := 2 N 2 j⇡
j=1 csc N 5 csc j⇡
N
> 0.

Proof. Recall the Euler-MacLaurin summation formula ([127, §3.3]): for any f 2
C 2s+2 ([a, b]) and n 2 N, and defining h = b na , the following holds,
n Rb s
X f f (a) + f (b) X 2r 1 f (2r 1)
(b) f (2r 1)
(a)
a
f (a + jh) = + + h B2r + Rs ,
j=0
h 2 r=1
(2r)!

B2s+2 (2s+2)
where Rs = nh2s+2 (2s+2)! f (↵) for some ↵ 2 (a, a + nh). Substituting in a =
⇡(N 1)
h = ⇡/N , n = N 2, b = a + hn = N
, f (x) = 2 (csc2 x 5) csc x and s = 2,
42 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

we obtain
Rb
a
f (x) dx 2N ⇣ ⇡ ⇡ ⇣ ⇡ ⌘⌘
= cot csc + 9 ln tan ,
h ⇡ N N 2N
f (a) + f (b) ⇣ ⇡ ⌘ ⇡
= 2 csc2 5 csc ,
2 N N
f (b) f 0 (a)
0 ⇡ cot N⇡ csc N⇡ 3 csc2 N⇡ 5
hB2 = ,
2 3N
f 000
(b) f 000 (a) ⇡ 3 csc6 N⇡ 742 cos N⇡ + 213 cos 3⇡ + 5 cos 5⇡
h3 B4 = N N
4! 2880N 3
⇡ 3 (742 + 213 + 5) csc6 N⇡ ⇡ 3 csc6 N⇡
> = ,
2880N 3 3N 3
and
csc9 (↵) (N 2) ⇡ 6 P (↵)
R2 (↵) = ,
1935360N 6
where P (x) := 1110231 + 1256972 cos 2x + 206756 cos 4x + 6516 cos 6x + 5 cos 8x;.
In previous formulae, we have used B2 = 1/6, B4 = 1/30, B6 = 1/42 and several
trigonometric identities in order to express the di↵erent terms in a suitable way for
what follows.
Introducing variable w = cos 2x, we may write the function defined by the first
three terms in P (x) as

Pb (w) := 903475 + 1256972w + 413512w2 .

Then, for each w 2 [ 1, 1], one has Pb0 (w) > 0; hence, for x 2 (0, ⇡) we obtain
P (x) Pb ( 1) 6516 5 > 0 and therefore R2 (↵) > 0, which leads to the
following:
Rb 2
f f (a) + f (b) X 2r 1 f (2r 1) (b) f (2r 1) (a)
SN = a + + h B2r + R2 (↵)
h 2 r=1
(2r)!
Rb 2
f (x) dx f (a) + f (b) X 2r 1 f (2r 1) (b) f (2r 1) (a)
> a + + h B2r
h 2 r=1
(2r)!
2N cot N⇡ csc N⇡ + 9 ln tan 2N ⇡ ⇣ ⇡ ⌘ ⇡
> + 2 csc2 5 csc
⇡ N N
⇡ cot N⇡ csc N⇡ 3 csc2 N⇡ 5 ⇡ 3 csc6 N⇡
+ .
3N 3N 3
There is a number of possible ways of proving this latter lower bound strictly positive.
For instance, since, for N 10, cot N⇡ > 3, we have
2N ⇣ ⇡ ⇡ ⇣ ⇡ ⌘⌘ ⇣ ⇡ ⌘ ⇡
SN > cot csc + 9 ln tan + 2 csc2 5 csc
⇡ N N 2N N N
⇡ ⇡ ⇣ ⇡ ⌘ ⇡ 3 csc6 ⇡
+ csc 3 csc2 5 N
N N N 3N 3
=: N .
3.4 Celestial Mechanical Problems 43

2N ⇡
The first term in that sum is exactly ⇡
F tan 2N , where
2
z z2
F : (0, 1) ! R, F (z) := + 9 ln z,
4
p ⇡
p
is strictly decreasing in 0, 5 2 . Since tan 2N < 5 2 for all N 10, we have
⇣ ⇡ ⌘ ⇣ ⇡⌘ 20
F tan F tan > ,
2N 20 3
and thus,
✓ ◆ ⇣ ⌘
2N 20 ⇡ ⇡ ⇡ ⇡ ⇣ ⇡ ⌘ ⇡ 3 csc6 N⇡
N > + 2 csc2 5 csc + csc 3 csc2 5
⇡ 3 N N N N N 3N 3
⇡
csc N ⇣ ⇡ ⌘
> GN csc ,
3N 3 N
where GN (x) := ⇡ 3 x5 + 3N 2 (2N + 3⇡) x2 N 2 (55N + 15⇡) and we have used
csc (x) > x1 for all x 2 (0, ⇡) (see (72)) and thus 40N
3⇡
> 403
csc N⇡ for all N 2.
0
It is immediate that GN (x) > 0 if
✓ ◆1/3 ! ✓ ◆
N 12 + 18 N⇡2 N4
x 2 0, 0, .
⇡ 5 ⇡3
⇥ ⇤
For all N 3, the latter interval contains N⇡ , csc N⇡ , thus allowing us to lower-
bound GN csc N⇡ by
✓ ◆ ✓ ◆
N N5 9⇡ 55⇡ 2 15⇡ 3
GN = 2 1+6+ > 0, N 10.
⇡ ⇡ N N2 N4

csc( N
⇡
)
In this way we obtain SN > N > 3N 3
G csc N⇡ > 0, N 10.

Lemma 9. For N 10, VN00 (cP ) has at least one eigenvalue greater than 1.

Proof. Indeed, let A = (ai,j )i,j=1,...,2N = VN00 (cP ). The Rayleigh quotient for vector
v = e2N,2N 1 = (0, 0, · · · , 0, 1, 0)T is
PN 1
v T Av T
vN ŨN,N vN j=1 csc3 j N⇡ 3 cos 2j N⇡ 1
= T
= a2N 1,2N 1 = P 1 ,
T
v v vN vN 4 N ⇡
j=1 csc j N

and it will be strictly greater than 1 if and only if

X1 ✓
N
2j⇡
◆
j⇡
N
X1 j⇡ X1 ✓
N
j⇡
◆
j⇡
3 cos 1 csc3 4 csc = 2 csc2 5 csc > 0,
j=1
N N j=1
N j=1
N N

which we already know holds for N 10 by Lemma 8. Elementary Linear Algebra

then yields the existence of at least one eigenvalue ˜ > 1 for VN00 (cP ).
44 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

Since max S = 1 < ˜ , ˜ 2/ S, we have proved that the planar equal masses
N -Body Problem with N 3 is not integrable by meromorphic first integrals.
Summarizing, we have proven:

Theorem 16 ([102]). For the planar equal masses N - Body Problem the following
assertions hold:

1. The number of additional meromorphic first integrals is no greater than:

a) one if N = 4;
b) three if N = 5, 6.

In particular, the Problem is not meromorphically integrable in the sense of

Liouville for all three values of N .

2. For N 3 problem is not meromorphically integrable in the sense of Liouville.

For N = 3, 4, 5, 6, there is no additional meromorphic first integral which is
independent with the classical first integrals.

For more details, see the original paper [102].

We must observe that Hamiltonian HN,d is not meromorphic. However, any first
integral of XHN,d (e.g. HN,d itself), when restricted to a domain of each determi-
nation of HN,d , is meromorphic and thus amenable to the whole theory explained
so far; see, for instance, [72, pp. 156-157] for more details as applied to a di↵erent
homogeneous potential.

3.4.2 Hill’s problem

In a joint work with Simó and Simon the first author proved the non–integrability
of Hill’s problem [99]. We follow this reference henceforth. As the proof is technical
and strongly based on the Galoisian correspondence, normality of the Picard-Vessiot
extensions (see Section 2.2) and algebraic groups, we only review in detail the compu-
tational steps in the proof, in agreement with our approach, where we are interested
mainly on algorithms and methodology.
Hill’s problem, usually called lunar as an homage to its earliest motivation, or
planar in order to distinguish it from its own extension to R3 , is a model originally
based on the Moon’s motion under the joint influence of Earth and Sun ([44]).
We can think of the Hill’s problem as a limiting case of the Restricted Three Body
Problem for a negligible mass of the Earth and when the distance of the Sun tends to
infinity. After some manipulations using the Levi-Civita regularization it is possible
to write the Hamiltonian of this problem as a polynomial of degree six ([123]):

H(Q, P ) = H2 + H4 + H6, (73)

a sum of homogeneous polynomials of degrees 2, 4 and 6, respectively:
1 1
H2 = (P12 + P22 ) + (Q21 + Q22 ), H4 = 2(Q21 + Q22 )(P2 Q1 P1 Q2 ),
2 2
3.4 Celestial Mechanical Problems 45

H6 = 4Q2 (Q41 4Q21 Q22 + Q42 ),

where, as usual, Q = (Q1 , Q2 ) are the positions and P = (P1 , P2 ) are the momenta;
here we prefer to use this notation in agreement with ref. [123], rather than x and
y as in other parts of the text.
Now we recall that our method (up to first order, i.e., we only need here the first
order obstruction to integrability given by V E := V E1 ) is given by the following
steps:
(1). Find a particular integral curve.
(2). Write the V E.
(3). Check if the identity component of the di↵erential Galois group of the V E
is commutative.
(1) Particular✓integral
◆ curve. We find a particular integral curve as follows. The
1 1 i
matrix A = p2 provides for a symplectic change of variables,
i 1
✓ ◆ ✓ ◆ ✓ ◆ ✓ ◆
Q1 Q̄1 P1 1 T P̄1
=A , = A ,
Q2 Q̄2 P2 P̄2

which in turn transforms Hamiltonian (73) into

H̄ = i(Q̄1 Q̄2 P̄1 P̄2 ) 4i(3Q̄41 2Q̄21 Q̄22 + 3Q̄42 )Q̄1 Q̄2 4Q̄1 Q̄2 (Q̄1 P̄1 Q̄2 P̄2 ).

The corresponding di↵erential system z̄ 0 = XH̄ (z̄) now displays two invariant planes

⇡1 : Q̄2 = P̄1 = 0 , ⇡2 : Q̄1 = P̄2 = 0 ,

in any of which all nontrivial information of that system reduces to a hyperelliptic

equation,
00
= + 12 5 , (74)
0
which through multiplication by and subsequent integration becomes
2
( 0) = 2
+4 6
+ 2h. (75)
2 0
Defining w = , z=2 , we arrive to the system

w0 = z, z0 = 4 w + 8w3 + h , (76)

whose Hamiltonian (at level zero energy) is K(w, z) = 12 z 2 + 2w2 8w4 4hw.
The solution to system (76), or equivalently to equation (w0 )2 = 4w2 + 16w4 +
8hw, is the inverse of an elliptic integral:
Z w(t)
t=± ( 4y 2 + 16y 4 + 8hy) 1/2
dy + C1 , C1 2 C,
0
46 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

translation t 7! t C1 being the next obvious step. It is a known fact (see, for
instance, [140]) that given a polynomial of degree four without repeated factors,
p4 (x) = a4 x4 + 4a3 x3 + 6a2 x2 + 4a1 x + a0 , and defining constants (called invariants)

g2 = a4 a0 4a3 a1 + 3a22 , g3 = a0 a2 a4 + 2a1 a2 a3 a32 a4 a21 a23 a0 ,

R w(t) 1/2
the solution for t = a
(p4 (x)) dx is the following:
p ⇥ ⇤
p4 (a)}0 (t; g2 , g3 ) + 12 p04 (a) }(t; g2 , g3 ) 1 00
p (a) + 24
24 4
1
p4 (a)p000
4 (a)
w(t) = a + ⇥ ⇤2 (4)
,
1 00 1
2 }(t; g2 , g3 ) 24 p4 (a) 48
p 4 (a)p 4 (a)

where }(t; g2 , g3 ) is the Weierstrass elliptic function. In our specific case, this be-
comes
w(t) = 6h/F (t), z(t) = 18h}0 (t; g2 , g3 )/F 2 (t),
where F (t) := 3}(t; g2 , g3 ) + 1. In particular,
p
1 (t) = 6h/F (t), 2 (t) = 1 (t),

are solutions
p to original equation (74). Furthermore, a simple calculation proves
h⇤ = 1/(6 3) to be a separatrix value in which 21 (t) = 22 (t) degenerates into
combinations of hyperbolic functions. We assume 0 < h < h⇤ .
We are now proving that, for the above range of h, w(t) has two simple poles in
each period parallelogram, the sides of which will be denoted as 2!1 , 2!2 , as usual.
In virtue of [34, p. 96], expression 1/(}(t) }(t⇤ )) (in our case, }(t⇤ ) = 1/3) has
exactly two simple poles in t⇤ , t⇤ (mod 2!1 , 2!2 ), with respective residues 1/}0 (t⇤ )
and 1/}0 (t⇤ ). Therefore, all double poles, if any, of 1/(}(t) }(t⇤ )), expanding
around t = t⇤ , are precisely those t⇤ such that }0 (t⇤ ) = 0. We have
4 8
(}0 (t; g2 , g3 ))2 = 4(}(t; g2 , g3 ))3 g2 }(t; g2 , g3 ) g3 = 4}3 } + 64h2 ,
3 27
and every pole (whether double or not) must satisfy }(t⇤ ) = 1/3; X = 1/3 is
obviously not a root of 4X 3 4X/3 8/27 + 64h2 unless h = 0.
So, just we proved that the Hamiltonian system defined by (73) has a particular
integral curve (depending on the energy level h) of the form
1
(Q1 (t), Q2 (t), P1 (t), P2 (t)) = p ( (t), i (t), 0 (t), i 0 (t)) . (77)
2
p
For all 0 < h < 1/ 6 3 , 2 (t) is an elliptic function with two simple poles in each
parallelogram period.
(2) Variational equation. We compute the variational equation along the above
integral curve. Reordering the vector of dependent canonical variables as (Q̄1 , P̄2 , Q̄2 , P̄1 )T
and restricting ourselves to the particular solution,

Q̄1 = , Q̄2 = 0, P̄1 = 0, P̄2 = i 0 ,

3.4 Celestial Mechanical Problems 47

the variational equations (VE ) along that solution are written as

0 0 1 0 10 1 0 1
⇠¯ 0 i 4w 0 ⇠¯ ✓ ◆ ⇠¯
B ⌘¯0 C B i(60w2 1) 0 4iz 4w C B C B C
B 0 C=B C B ⌘¯ C =: A1 B1 B ⌘¯ C ,
@ ⇠ A @ 0 0 0 i A @ ⇠ A 0 A1 @ ⇠ A
0 2
⌘ 0 0 i(60w 1) 0 ⌘ ⌘
(78)
and their lower right block, the normal variational equations (NVE )
✓ 0 ◆ ✓ ◆✓ ◆
⇠ 0 i ⇠
= , (79)
⌘0 i(60w2 1) 0 ⌘
that is,
⇠ 00 (t) = (60w2 (t) 1)⇠(t). (80)
Next step is to obtain a fundamental matrix for (79). An obvious shortcut is to take
w as new independent variable and to define ⌅(w), H(w) such that ⇠ = ⌅ w and
⌘ = H w. We have
✓ ◆
d2 ⌅ w 8w3 h d⌅ 60w2 1
=4 + ⌅, (81)
dw2 wf (w, h) dw wf (w, h)
also expressible in matrix form
✓ d ◆ ✓ ◆✓ ◆
⌅ 1 0 i ⌅
dw
d =p 2 , (82)
dw
H wf (w, h) i(60w 1) 0 H

where f = f (w, h) = 4(4w3 w + 2h).

The VE, equation (78), is integrable; of course, we know from Chapter 2 that this
is due to the solvability of the identity component of the Galois group, see below.
Let us start from the block notation
✓ ◆
P Q
= , (83)
R S
P, Q, R, S being 2 ⇥ 2 matrices with their entries in some di↵erential field to be
described below. We can assume (0) = Id4 , which, along with the triangular
form of (78), assures R ⌘ 0. In particular, the matrix form of the NVE (79) can
be written as S 0 = A1 S. Let us now proceed to integrate these normal equations.
More precisely, let us collect all necessary information about the fundamental matrix
N (t) of (79) with initial condition N (0) = Id2 .
0
p Using well-known properties of } and }00 , it is easy to prove that ⌅1 (w) =
f (w, h) is a solution of (81), and therefore
3/2
⇠1 (t) = ⌅1 (w(t)) = }0 (t; g2 , g3 ) (3}(t; g2 , g3 ) + 1)

is a solution of (80). A first solution of (79) is then

✓ ◆ ✓ p ◆
⇠1 (t) 16w3 (t) 4w(t)p+ 8h
= C1 , C1 2 C.
⌘1 (t) 2i (12w2 (t) 1) w(t)
48 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

We now recall d’Alembert’s method ([47, p. 122]) in order to obtain a second

solution of (80) independent of ⇠1 . This solution is
Z t
⇠2 (t) = ⇠1 (t) {⇠1 (⌧ )} 2 d⌧. (84)
0

After recovering our former independent variable t through composition, we have a

fundamental matrix for the NVE, that is, the block S in (83),
✓ ◆ ✓ ◆
⇠1 ⇠2 ⇠1 ⇠2
N (t) = = .
⌘1 ⌘ 2 i⇠10 i⇠20

In particular, P (t) ⌘ S(t) since they are both fundamental matrices for the same
initial value problem. We now compute the block Q in (83); the standing equations
(in vector form) are
✓ 0 ◆ ✓ ◆✓ ◆ ✓ ◆✓ ◆
⇠¯ 0 i ⇠¯ 4w 0 ⇠
0 = 2 + , (85)
⌘¯ i(60w 1) 0 ⌘¯ 4iz 4w ⌘

where (⇠, ⌘)T are the solutions to the NVE. Applying variation of constants to (85)
we obtain Z t
Q(t) = N (t) V (⌧ )d⌧, (86)
0
where ✓ ◆
4w(t) 0 1
C(t) = , V (t) = N (t)C(t) N (t).
4iz(t) 4w(t)
In other words, the fundamental matrix of (78) has the form
✓ Rt ◆
N (t) N (t) 0 V (⌧ )d⌧
(t) = . (87)
0 N (t)

In view of (86), computing explicitly would now only take the computation of
four integrals. The path we are taking, however, is a di↵erent one, although we are
keeping in mind all of this notation and the final expression (87).
Our next aim is to prove only two specific properties of the fundamental ma-
trix of (78), namely the existence of first and second class elliptic integrals and
logarithmic terms in its coefficients.
Let K be the field of all elliptic functions of the complex plane (with some fixed
periods). We know a solution of (80),
p
⇠1 (t) = 4w3 (t) w(t) + 2h,

and can obtain a second one using (84) and the chain rule. Let us define ↵1 , ↵2 , ↵3
as the values of w for which f (w, h) = 0, the functions
s ! s
w(↵3 ↵1 ) ↵3 (↵1 ↵2 )
(w, h) := arcsin , k(h) := ,
↵3 (w ↵1 ) ↵2 (↵1 ↵3 )
3.4 Celestial Mechanical Problems 49

(both attaining complex, nonzero values if h 2 (0, h⇤ ) and therefore w(t) 6= 0) and
let
Z Z
1 1
2 2
E( |k) := (1 k sin ✓) d✓,2 F ( |k) := (1 k 2 sin2 ✓) 2 d✓.
0 0

be the elliptic integrals of first and second class, respectively (see [34], [140]). We
then obtain a fundamental matrix for the NVE (82),
!
⌅1 (w) ⌅2 (w)
N (w) = =
H1 (w) H2 (w)
p !
f (w, h) g1 {f1 E( |k) + f2 F ( |k) + g2 }
p d
,
2i w( 1 + 12w2 ) i dw (g1 {f1 E( |k) + f2 F ( |k) + g2 })
for some f1 = f1 (h), f2 = f2 (h), g1 = g1 (w, h), g2 = g2 (w, h), the first three non-
vanishing if h 2 (0, h⇤ ), and the last two linked to w by algebraic equations. In
particular, this yields our fundamental matrix N (t) = N (w(t)) for (79).
The fundamental trait of E( |k) and F ( |k) is that they are transcendental over
K. Indeed, nontrivial elliptic integrals of the first and second classes are not elliptic
functions (see [34, Theorem 6.5 and its proof]) and they stem from quadratures;
thus, as we said before, E( |k) and F ( |k) cannot be expressed in terms of elliptic
functions under any relation of algebraic dependence.
Let us prove the existence of terms with nonzero residue in the diagonal of matrix
V (t). Since
✓ ◆ ✓ ◆
⇠1 ⇠2 ⇠1 ⇠2
N (t) = =
⌘1 ⌘ 2 i⇠10 i⇠20
is the fundamental matrix of a Hamiltonian linear system, it is symplectic. The
integrand in (86) becomes
!
w(⇠2 ⇠10 + ⇠1 ⇠20 ) + w0 ⇠1 ⇠2 ⇠2 (2⇠20 w ⇠2 w0 )
V (t) = 4 i =:
(2w⇠10 w0 ⇠1 )⇠1 w(⇠1 ⇠20 +⇠10 ⇠2 ) w0 ⇠1 ⇠2
!
u(t) v1 (t)
4i .
v2 (t) u(t)
For every h 2 (0, h⇤ ), we expand these four entries around a simple pole t⇤ of w(t);
expressing only the first term in each power series, we have

w(t) = C0 (t t⇤ ) 1
+ O(1),
3/2
⇠1 (t) = 2C0 (t t⇤ ) 3/2
+ O (t t⇤ ) 1/2
,
3/2
C0
⇠2 (t) = (t t⇤ )5/2 + O (t t⇤ )7/2 ,
8
50 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

for some C0 = C0 (h) 2 C; therefore,

C0
u(t) = (t t⇤ ) 1 + O(1),
2
3
v1 (t) = (t t⇤ )3 + O (t t⇤ )4 ,
32C02
v2 (t) = 8C04 (t t⇤ ) 5 + O (t t⇤ ) 4
.
Hence, and save for the only value of h forcing C0 = 0 (i.e. h = 0), we have a
nonzero residue in u (t), which results in the aforementioned logarithmic terms in
the diagonal of
Z t Rt Rt !
0
u(⌧ )d⌧ v (⌧ )d⌧
0 1
V (⌧ )d⌧ = R t Rt .
0
0 2
v (⌧ )d⌧ 0
u(⌧ )d⌧
We have now obtained a class of functions which cannot be linked algebraically to
the former. Indeed, logarithms are special cases of elliptic integrals of the third class,
which are neither elliptic functions nor elliptic integrals of first or second class (see
[34, Theorem 6.5 and its proof] once more), and in this case the logarithms have
been obtained through a quadrature.
Summarizing, we have obtained the following information about the fundamental
matrix of the variational equation. The variational equation along solution (77) have
a fundamental matrix of the form
✓ Rt ◆
N (t) N (t) 0 V (⌧ )d⌧
(t) = ,
0 N (t)

where ✓ ◆
⇠1 (t) ⇠2 (t)
N (t) =
i⇠10 (t) i⇠20 (t)
is a fundamental matrix of the normal variational equation; furthermore, ⇠2 is a
linear combination Rof elliptic functions and nontrivial elliptic integrals of first and
t
second classes, and 0 V (⌧ )d⌧ is a 2⇥2 matrix function containing logarithmic terms
in its diagonal.
Let us interpret our results in terms of field extensions. First of all, we note that
using coordinates (x, y) = ( , 0 ) all solutions of the equation (75) are included in
the hyperelliptic curve

h := (x, y) 2 C2 : y 2 = x2 + 4x6 + 2h .
The previous transformation w = x2 , z = 2xy induces a finite branched covering

h ! ⇤h ,
where ⇤h is the elliptic curve defined by
⇤h := {z 2 /2 + 2w2 8w4 4hw = 0}.
Keeping K (= M(⇤h )) as the field of all elliptic functions, let us describe the
Picard-Vessiot extension over K for VE h in detail
3.4 Celestial Mechanical Problems 51

1. First of all, let us define the extension

K ⇢ K1 := K(⇠1 , ⇠10 ),

based on the adjunction of the first solution ⇠1 of (79) and its derivative, which
is an algebraic (in fact, quadratic) one.

2. Second of all, adjoining the solution ⇠2 from (84) to this new field, we obtain
the extension
K1 ⇢ L1 := K1 (⇠2 , ⇠20 ) = K(⇠1 , ⇠10 , ⇠2 , ⇠20 ),
which is transcendental, since it is nontrivial and defined exclusively by an
adjunction of quadratures.

3. Third of all, adjoining the matrix integral from (86) to L1 , we have

✓Z t Z t Z t ◆
L1 ⇢ L2 := L1 u, v1 , v2 ,
0 0 0

also given by quadratures, nontrivial, and thus transcendental.

So far, the Picard-Vessiot extension L2 | K of the (78) splits as follows

K ⇢ K1 ⇢ L 1 ⇢ L 2 .

(3 ) The identity component of the Galois group G of (78) is not commu-

tative. This step is a bit more technical and we do not show the details here; we
only comment that it is based on the following facts:
(a). By definition, G is the Galois group of the Picard-Vessiot extension Gal(L2 /K).
(b). By the Galoisian correspondence G0 = Gal(L2 /K2 ), since K2 ⇢ L2 is transcen-
dental; see Section 2.2.
(c). A detailed analysis of the extension K2 ⇢ L2 , again using the Galoisian corre-
spondence, proves that in the representation of G0 given by its action on the funda-
mental matrix of (78) this group is a non-commutative subgroup of the unipotent
group

80 1 9
>
> 1 µ +µ + µ >
>
<B C =
b0 = B 0 1
G
 C : µ 2 C,  2 S1 , 2 S2 , 2 S3 , (88)
>@ 0 0 1 µ A >
>
: >
;
0 0 0 1

where a fundamental role is played by the fact that the extension K2 ⇢ L2 splits
in K2 ⇢ L1 and L1 ⇢ L2 , the former given by non-trivial elliptic integrals and the
latter with a logarithmic term. See the original paper [99] for the details. As a
conclusion the following is proven.
Theorem 17 ([99]). The Hamiltonian system defined by Hamilton function (73) is
not integrable with meromorphic first integrals.
52 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

This is the second time we meet the relevance of the logarithmic terms in our
approach; it appeared before in the proof of the non-integrability of the Hénon-Heiles
system using the higher order variational equations in Section 2.4.
Another remark is that this is the only case we know in which the obstruction
to integrability is obtained by an integrable first order variational equation. From
Picard-Vessiot theory (Section 2.2), we knew that the reason for this behavior is
that the identity component of the Galois group is solvable but not commutative.
In particular, it is easy to see that Ziglin’s method does not work in this problem,
essentially because the identity component of the Galois group is unipotent.
For more details about the non-integrability of Hill’s problem see the original
reference [99].
In paper [117] a proof is given of the non-integrability of Hill’s problem by means
of algebraic first integrals.

3.5 Other N-bodiy Problems, Mechanical and Physical Prob-

lems
We briefly survey some works (with references) by several authors about other Ce-
lestial Mechanics problems, (non-Celestial) N- body problems and some physical
problems using our main Theorems of Section 2.3 and Section 3.2. Although some
of them are not related to the main topics in this Chapter, we decided to include
them here.
a) Satellite’s and asteroid’s dynamics. In some Celestial Mechanics problems,
such as satellite’s dynamics, it is important to take into account the shape of the
bodies. If the potential is expanded in harmonics and we truncate it, we may ask
about the non-integrability of these truncated potentials.
In the same way, a system arising from truncation of the harmonics was consid-
ered by Arribas and Elipe in [6]. These authors consider the motion of a particle
around a massive straight segment; for instance, this system can approximatively
represent the dynamics around some asteroids. The potential is of logarithmic type
and homogeneous of degree k = 0. Thus, we cannot apply Theorem 7. Then the po-
tential is expanded in harmonics and, as above, by truncation it is possible to apply
the Mondéjar result, Theorem 10, and the truncated system is non-integrable.
We remark that, without any further analysis, from the non-integrability of the
truncated potential does not imply the non-integrability of the original potential.
For instance, non-integrable systems are obtained by truncation of an expansion of
an integrable Toda lattice [146].
Another sort of problems connected to satellite dynamics comes from considering
the satellite as a finite rigid body rather than as a point particle. It is well-known
that when the three principal inertial moments of the satellite are the same the
system is integrable. Under the assumption that the satellite has axial symme-
try, i.e., that two of the inertia moments are equal, Audin and Boucher studied
the non-integrability of this system using a previous result by Boucher and Weil
about the non-commutativity of the identity component of the Galois group of the
3.5 Other N-bodiy Problems, Mechanical and Physical Problems 53

normal variational equations in presence of formal logarithmic branching points in

irreducible factors (see the Appendix A, Section A.3) ([17, 9]); in the proof they
used the computation of the normal variational equation along a suitable integral
curve previously obtained by Maciejewski (unpublished). Maciejewski and Przybyl-
ska studied the non-integrability of this system by considering also the action of the
Earth’s magnetic field ([69]).
b) Other N-bodies, Mechanical and Physical problems. Several perturba-
tions of the integrable hydrogen atom are considered by Ferrer and Mondéjar. The
Zeeman Hamiltonian is given by a perturbation of the atom with a magnetic field.
Their non-integrability was shown by Kummer and Sáenz in [61] by using Ziglin’s
theorem. The Stark-Zeeman Hamiltonian is a generalization of the Zeeman Hamil-
tonian obtained through the additional consideration of action of an electric field
parallel to the magnetic field. The non-integrability of the Stark-Zeeman model by
rational first integrals was proven in [37] using Theorem 5. The non-integrability of
the hydrogen atom under a combination of magnetic field and circularly polarized
microwaves was studied in [38], and a proof of the non-integrability of the generalized
van der Waals Hamiltonian was given in [39].
The non-integrability of several Three-body Problems in atomic physics is stud-
ied by Almeida, Stuchi and López-Castillo ([3, 121]. In the non–integrability proof
they used Theorem 7, but we believe that it would be an interesting exercise for the
interested reader to go further and obtain the non-existence of an additional mero-
morphic first integral using Theorem 9 as we did in Subsection 3.4 for the Celestial
Mechanics case. In fact, in [121] three of the eigenvalues of the Hessian matrix along
a suitable homothetical point are 2, 1, 1; they come from the classical first inte-
grals in involution: the energy and the two components of the angular momentum,
respectively. The other three eigenvalues depend on the parameters (masses and
atomic number) of the system.
The Gross-Neveu systems are classical Hamiltonian systems with n-degrees of
freedom given by a potential linked to the root system of a simple Lie algebra
X
V (x1 , . . . , xn ) = exp(< ↵, x >),
↵

where x = (x1 , . . . , xn ) and ↵ = (↵1 , . . . , ↵n ) moves along the root system. These
systems can be considerered as certain generalizations of the Toda lattice. Maciejew-
ski, Przybylska and Stachowiak proved in [75] the non–integrability of the Gross-
Neveu systems when the associated Lie algebras are so(2n), so(2n + 1), sl(n + 1)
and sp(2n) (n 2).
Maciejewski and Przybyska completely solved the integrability problem of a gen-
eralized two–fixed–centres–problem whose interaction potential is V = ar 2n . This
is a generalization of the classical two fixed centres problem of Celestial Mechanics for
the Newtonian potential. The system is integrable if and only if n 2 { 2, 1, 0, 1/2}
([72]).
The integrability of the spring–pendulum system was studied by Churchill, Del-
gado and Rod in [25]. It was also studied by the authors in [98], where we completely
54 3 HOMOGENEOUS POTENTIALS AND RELATED PROBLEMS

solved the problem for physical values of the mass parameter . A generalization was
considered in [77].
The integrability of the dynamical problem of the geodesic on an ellipsoid was
proven by Jacobi. Bardin, Maciejewski and Przybylska completely solved the inte-
grability of a generalization of this problem, when the particle moves on the ellip-
soid a1 x2i + a2 x22 + a3 x23 = 1 under the action of a quadratic homogeneous potential
V = 1/2(b1 x2i + b2 x22 + b3 x23 ) ([13]).
The Stormer problem is a Hamiltonian system representing the dynamics of a
charged particle in a magnetic dipole. It is a model of the dynamics of charged
particles under the action of Earth’s magnetic field. The non-integrability of the
Stormer problem was proven by Kummer and Sáenz by means of Ziglin’s Theorem
in [62]. The non-integrability of an anisotropic Stormer problem was studied by
Almeida and Stuchi in [3].Furthermore, Sáenz in [116] proved the non-integrability
of another problem of magnetic confinement. The points at infinity are irregular
singularities; in fact, although it was not explicitly stated in [116], by means of
Theorem 5, Sáenz proved the obstruction to the existence of rational first integrals.
Other anisotropic families of potentials generalizing the anisotropic Kepler prob-
lem, such as the anisotropic Mane↵ problem, are studied in [7] by Arribas, Elipe
and Riaguas.
 55

4 Hamiltonian Rigid Body Problem

In 1983 using Theorem 1 Ziglin completely solved the integrability problem of the
heavy rigid body problem by means of complex meromorphic first integrals ([148]),
and in 1997 the integrability by means of real meromorphic first integrals. In 2005
Maciejewski and Przybylska revisited the Ziglin results at the light of Theorem 4
and recover Ziglin’s results in a simpler way in the Ziglin original papers ([73]).
We give here a necessarily brief account of this nice work. For space limitations
we only survey the first part of the work, i.e., integrability by means of complex
meromorphic first integrals. It is apparent that in order to recover Ziglin’s result
the authors used several of the methods described in our text. This gives an idea of
the need of attacking a non-trivial problem from several points of view.

4.1 The equations of motion

The Euler-Poisson equations governing the dynamics of a rigid body with a fixed
point within a constant gravitational field, in the moving frame of coordinates at-
tached to the principal axes of inertia, are given by

Ṁ + ⌦ ⇥ M = µk ⇥ l, k̇ = k ⇥ ⌦, (89)
being M = (p, q, r) the angular momentum, ⌦ = (p/A, q/B, r/C) the angular ve-
locity, (A, B, C) the principal momenta of inertia, k = (↵, , ) the unitary vertical
vector, l = (x0 , y0 , z0 ) the unitary vector with origin at the fixed point and pointed
towards the center of gravity and µ the weight of the body multiplied by the distance
from the fixed point to the center of masses.
The system (89) is a dynamical system defined over C6 . It has five free real
parameters (l, A, B, C), l · l = 1 and three classical first integrals:
i) The energy integral:
1
H = M · ⌦ + µk · l.
2
ii) The geometrical integral:

f1 = k · k = 1.

iii) The vertical angular momentum:

f2 = M · k.

Using these integrals it is possible to reduce the system (89) to a two degrees of
freedom Hamiltonian system, see [60], pp. 31-32. From a modern point of view this
reduction is studied in the context of the Poisson actions of Lie groups on symplectic
manifolds and the associated momentum map, see for instance [5], Appendix 5. In
fact, the Euler-Poisson equations (89) can be considered as a Hamiltonian system
with three degrees of freedom, whose configuration space is the three dimensional
56 4 HAMILTONIAN RIGID BODY PROBLEM

rotation group and with the additional first integral f2 ; in this case, the geometric
integral f1 is automatically equal to one. Let us denote

Mm = {(M, k) 2 C6 : f2 (M, k) = m, f1 (M, k) = 1}.

On each level manifold
Mm ,
H|Mm defines a two–degrees of freedom Hamiltonian system. We say that the Euler
equation is (globally) integrable if there exits an additional independent (of H and
f2 ) first integral f3 globally defined in the five–dimensional manifold

M = {(M, k) 2 C6 : f1 (M, k) = 1};

in this case the Hamiltonian systems defined by H|Mm are integrable in Liouville’s
sense and we can apply our methods.
We remark that from our variational equations point of view it is not necessary
to perform the reduction in this non-linear way, since, at the level of the variational
equations, the reduction is reflected in the obtention of the normal variational equa-
tions N V E.
The known cases of integrability of the equations (89) are:
1. (Euler 1758). In the Euler case either there is no gravity (µ = 0 ) or the fixed
point is at the center of masses (l = 0).
2. (Lagrange 1788). The body is axially symmetric, A = B, and the fixed point
belongs to the symmetry axis, x0 = y0 = 0.
3. ( Kovalevskaya 1889). For over a century the only cases of integrability known for
the Euler-Poisson equations were those of Euler and Lagrange, but in a celebrated
paper Sophie Kovalevskaya obtained a new and highly non-trivial case of integra-
bility for A = B = 2C and z0 = 0 ([59]). This paper is the seminal paper that
motivated the actual theory of algebraically completely integrable systems; for the
historical transcendence of this paper see the interesting article of Michèle Audin
[11].
4. (Goryachev-Chaplygin 1910). Under some conditions, a new case of (partial)
integrability was obtained only for m = 0, i.e., if z0 = 0 and A = B = 4C, the
Hamiltonian system defined by H|M0 is integrable in the sense of Liouville ([48]).

4.2 Non–integrability
Using Theorem 1, Ziglin proved the following non–integrability result which solved
the integrability problem of the heavy top by means of meromorphic first integrals.

Theorem 18 ([148]). The two–degrees of freedom Hamiltonian system defined by

H|M0 on the zero level symplectic manifold of the vertical angular momentum f2 ,
M0 , is integrable by meromorphic first integrals only in the classical integrable cases
(i)–(iv).
4.2 Non–integrability 57

As said before, Maciejewski and Przybylska obtained a new simpler proof of

Theorem 18 using Theorem 4. Their proof follows the following lines (parallel to
the Ziglin’s original proof):

1. There exists a family of particular pendulum–like solutions k of the Hamil-

tonian H|M0 .

2. It is possible to normalize the five parameters of the system in such a way that
y0 = 0 and B = 1.

3. Using a change of variables (i.e., algebrization procedure, Section A.2 ), the

normal variational equations, NVE, along the family of solutions k are re-
duced to a second order equation with rational coefficients with four finite
singularities and being the point at infinity an apparent singularity, provided
the parameter

1 1 1
d= ( )x0 z0
2 C A
is di↵erent from zero. In this case using Kovacic’s algorithm and the invariance
of the identity component by a finite ramified covering (Appendix A, Theorem
24), the authors proved that the identity component of the Galois group of
the NVE is not commutative.

4. When d = 0 and we do not consider the Lagrange integrability case (ii), z0 = 0

and the above N V E degenerate trough confluence into an hypergeometric
equation and applying Kimura’s theorem (Appendix B, Theorem 26) only five
possible discrete families for the values of the parameter C are compatible
with the commutativity of the identity component of the Galois group of this
hypergeometric equation.

5. For z0 = 0 there is another family of particular elliptic solution curves 1k ,

such that the normal variational equation along them, N V E1 , are reduced to
a family of Lamé type equations (see Appendix C):

d2 ⇠ e
2
((2C(2C + 1)}(t) + C(1 4C))⇠ = 0, (90)
dt 3
where e = 2k 2 1, (k 2 (0, 1) is the moduli). Using Corollary 3 of Appendix C
and Dwork’s result (Proposition 9 of Appendix C), it is possible to prove that
the only values of C for which there is a commutative identity component in
the Galois group (of N V E1 ) for the previously obtained five discrete families
are C = 1, C = 1/2 and C = 1/4, which correspond to Euler, Kovalevskaya
and Goryachev-Chaplygin integrable cases, respectively.
58 5 COSMOLOGICAL MODELS

5 Cosmological Models
Some years ago we proved the non-integrability of the Bianchi IX model by means
of rational first integrals. Along these years other non-integrability results of some
cosmological models were obtained by several authors. We survey these results.

5.1 Bianchi’s Models

The Bianchi cosmological models are a family of relativistic homogeneous anisotropic
models (in the spatial variables) reducing to finite dimensional Hamiltonian systems.
In this section we sketch the non-integrability proof of two members of this family:
Bianchi IX and Bianchi VIII. For the obtention of the Bianchi family of models, see
[64].
The non-integrability proof of the Bianchi IX model was obtained in [98] (see also
[94]); we follow this reference. The Bianchi IX Cosmological model is a dynamical
system given by the equations in “logarithmic” time ([64]),

d2 log x1
= (x2 x3 )2 x21 ,
dt2
d2 log x2
= (x3 x1 )2 x22 , (91)
dt2
d2 log x3
= (x1 x2 )2 x23 ,
dt2
with the energy constraint (from physical considerations)

ẋ1 ẋ2 ẋ2 ẋ3 ẋ3 ẋ1

E= ( + + ) + x21 + x22 + x23 2(x1 x2 + x2 x3 + x3 x1 ) = 0.
x1 x2 x2 x3 x3 x1
Thus, we get a dynamical system of dimension five on the zero level energy
manifold M0 .
In fact this is a Hamiltonian system with position variables x1 , x2 , x3 and conju-
gate moments given by

1 d
y1 = log(x2 x3 ),
x1 dt
1 d
y2 = log(x1 x3 ),
x2 dt
1 d
y3 = log(x1 x2 ).
x3 dt
Hence, the energy becomes the two degrees of freedom Hamiltonian
1
HIX = (x21 y12 + x22 y22 + x23 y32 2x1 x2 y1 y2 2x2 x3 y2 y3 2x1 x3 y1 y3 )
4
5.1 Bianchi’s Models 59

+x21 + x22 + x23 2x1 x2 2x2 x3 2x1 x3 = 0.

Our proof of the non-integrability relies on the study of the variational equations
along particular solutions of the Taub family.
As Taub noticed, the subspace x2 = x3 (ẋ2 = ẋ3 ) is invariant by the flow
of the system, and the reduction to this invariant four dimensional space (three
dimensional if we consider the restriction of the system on the restricted manifold
M0 ) is completely integrable (i.e. it is an integrable subsystem T ) and the solutions
can be calculated explicitly ([129], p. 481). From the Taub family of solutions we
select the particular ones

2k k cosh(2kt)
x1 = , x2 = x 3 = ,
cosh(2kt) 2 cosh2 (kt)

being k a parameter. This particular integral curve (for a fixed value of k) is our
integral curve and is contained in M0 .
Along the above particular integral curve (for a fixed value of k) we compute the
variational equation and the normal variational equation, NVE. The NVE can be
reduced to a second order di↵erential equation with rational coefficients

1 1 5 1 3 1
⌘¨ + ( + 2
+ )⌘ = 0. (92)
4 x 1 4 (x 1) 16 x2
This equation has x = 0, 1 as regular singularities, being x = 1 an irregular singu-
larity; in fact, it is a confluent Heun’s equation, i.e., an equation obtained from a
confluence of two singular points into an equation with four regular singularities on
the Riemann sphere. For the physical meaning of these singular points, see [98] (or
[94]).
Then by means of our Theorem 5 and using Kovacic’s algorithm (Appendix
A, Section A.1), we proved in [98] the non-integrability of this system by means of
rational first integrals. We point out that we will use Theorem 5, instead of Theorem
4, because the points at 1 of the particular integral curve in phase space correspond
to an irregular singularity of the NVE (see the original reference [98], or the book
[94]), for details).
In an analogous way, Maciejewski, Strelcyn and Szydlowski proved in 2001 the
non-integrability of the Bianchi VIII cosmological model ([79]). The Hamiltonian is
now

HV III = x22 y12 x21 (1 + y22 ) 2x1 y1 (x2 y2 x3 y3 )

1 2
+x2 x3 (2y2 y3 x (1 + 4y32 ).
1)
4 3
Inasmuch as for Bianchi IX, the Hamiltonian HV III restricted on the Taub man-
ifold X1 = 0, x1 = 0 is an integrable subsystem, which on the five dimensional zero
1
energy level manifold, M0 = HV III (0), gives rise to a three dimensional integrable
60 5 COSMOLOGICAL MODELS

(Taub) subsystem T . Then the authors in [79] integrate the subsystem T and com-
pute the NVE along this three-parametric family of particular solutions.As in HIX
it is possible to reduce these variational equations to second order variational equa-
tions with rational coefficients and with our Theorem 4 and Kovacic’s algorithm they
proved the non-integrability of the Hamiltonian system of HV III by means of mero-
morphic first integrals. Independently of the analogies, there two main di↵erences
here with respect to our previous result on HIX :

1. In [79] is computed the NVE along the complete three parameter family defined
by the Taub subsystem on M0 , not only through a one parameter family as was
the case for [98]. This implies that the authors obtain the non-integrability
of the Hamiltonian system defined by HV III in a neighborhood of the Taub
family of solutions restricted to M0 .

2. The NVE in [79] have regular singular points at 1, for this reason it is possible
to apply Theorem 4 and to find obstruction to the existence of meromorphic
first integrals instead of rational ones.

5.2 Friedman-Robertson-Walker’s Models

The Friedman-Robertson-Walker (FRW) cosmological models are the classical rela-
tivistic homogeneous isotropic models that also can be reduced to finite dimensional
Hamiltonian systems (see [66]).
One important FRW family of models with a conformally coupled self-interacting
scalar field is defined by the two degrees of freedom Hamiltonian

1 1 ⇤
HF RW = ( y12 + y22 ) + k(x22 x21 ) + m2 x21 x22 + x41 + x41 ,
2 2 2 2
where k 2 { 1, 0, 1} is the spatial curvature and m, ⇤, are real parameters which
represent the mass of the scalar field, the cosmological constant and the self-coupling
constant, respectively.
The first application of the methods in this text to the non-integrability of a
FRW model was obtained by Maciejewski and Szydlowski in 2000 as follows ([66]).
The flat subfamily with k = 0 in the family HF RW gives rise through the complex
canonical change (x1 , y1 ) ! ( ix1 , iy1 ) to a classical Hamiltonian with an homoge-
neous potential of degree four. Thus, it is possible to apply Theorem 7 and to obtain
necessary conditions for integrability: if the parameters m, ⇤, and do not satisfy
some concrete discrete families of algebraic relations, the system is not integrable
with meromorphic first integrals.
Coelho, Skea and Stuchi also studied in [27] the non-integrability of the family
HF RW but now with k 6= 0. As x1 = y1 = 0 is an invariant plane, it gives rise to
a family of particular solutions parametrized by the energy h, like in Section 3.3.
The NVE along this family is a Lamé type equation (Appendix C) that the authors
write in algebraic form and by Theorem 4 using Kovacic’s algorithm (Appendix A)
the authors proved that a necessary condition for meromorphic integrability is that
5.2 Friedman-Robertson-Walker’s Models 61

2m2
{ , ⇤} ⇢ { : p 2 N} (93)
(p + 1)(p + 2)
Very recently Boucher and Weil continued the work of [27] and obtained stronger
restrictions on the parameters compatible with the integrability of HF RW ([19]).
These authors systematically used the obstruction to the integrability given by the
existence of logarithmic terms either in the first variational equation or in higher
order variational equations. In particular, they recovered the conditions (93) for
integrability applying their own criterium to the first order NVE (see Appendix A,
Section A.3) and, as the first order variational equation is given by a two uncoupled
Lamé-Hermite equations (122), by means of Proposition 7 of Appendix A these
authors obtain other restrictions for meromorphic integrability of the Hamiltonian
system defined by HF RW . Moreover, Boucher and Weil conjectured that this system
is integrable if, and only if, either = ⇤ = m2 or = ⇤ = m2 /3.
In another recent paper, Maciejewski, Przybylska, Stachowiak and Szydlowski
studied the integrability of the FRW cosmological model defined by the Hamiltonian

1 y22
H= y12 + kx21 + Lx41 + m2 x41 x22 . (94)
2 x21
This Hamiltonian represents a FRW cosmological model with a complex scalar field.
As with the Hamiltonian HF RW , k and m are the curvature and the mass of the field,
respectively, being L essentially (i.e., modulo a constant factor) the cosmological
constant. For L = 0 the authors obtain an algebraic form of a NVE of Whittaker
type, i.e., a confluent hypergeometric equation with a regular singular point at the
origin and an irregular one at the infinity and, using a result of the second author
about the Galois group of this kind of equations (see [85]), by Theorem 5 a necessary
condition is obtained for integrability with rational first integrals: namely, that the
curvature k and energy h must be zero. For L 6= 0 and h = 0 then either k = 0
or 9 4m2 /L = (2p + 1)2 , for some p 2 Z, using Kimura’s theorem (Theorem 26
of Appendix B) and Theorem 4, because the NVE is reduced to an hypergeometric
equation, equation (129) of Appendix B. For L 6= 0 and h 6= 0 (i.e., we are on the
physical manifold M0 ) then either k = 0 or 9 4m2 /L = 4p)2 , for some p 2 Z, using
again Proposition 7 of Appendix A.
Although it is not directly connected with the main body of this text, we mention
here the recent work in [126], where the di↵erential Galois theory is applied directly
(i.e., without use of the variational equations) to the integrability in closed form of
the di↵erential equations obtained as linear density perturbations of FRW model.
The paper is a nice and relatively simple example of another application of the
di↵erential Galois theory to integrability problems. The main technical tool used by
the authors was the Kovacic algorithm together with the algebrization mechanism
(Section A.2 of Appendix A).
62 6 AN APPLICATION TO PAINLEVÉ’S TRANSCENDENTS

6 An Application to Painlevé’s Transcendents

The Painlevé transcendents are the solutions of the six Painlevé’s families of equa-
tions (P1 , P2 , P3 , P4 , P5 , P6 ). Since the work of Malmquist we know that the
Painlevé transcendents can be expressed as Hamiltonian systems of 1 + 1/2 degrees
of freedom.
We conjecture that all Painlevé equations are non-integrable by means of rational
first integrals. Using the fundamental theorems of subsections 2.3 and 2.4 it is
possible to prove this result for some particular cases (for some particular equations
and some particular values of the parameters); in the state of the art it works for
some special values of the parameters in the cases of P1 and P6 , as we shall explain
below. Moreover, using an extension of our main theorems, replacing di↵erential
Galois groups by di↵erential Galois groupoids, thus linear di↵erential Galois theory
by non-linear di↵erential Galois theory, it is possible to prove that P1 and P6 (for all
the values of the parameters) are non-integrable by means of rational first integrals.
In the cases of P1 , P2 , P3 , P4 , P5 , as the variational equation along each particular
solution will have an irregular singular point at infinity, then, using theorem 5 (or
its extension to higher order variational equations), we can only hope to obtain
obstruction to the integrability by means of rational first integrals. In the case of P6
it is possible to obtain an obstruction to the integrability by means of meromorphic
first integrals for some special values of the parameters.
Here we illustrate our approach with Painlevé II and Painlevé VI, but, of course,
we believe that similar studies can be done for other Painlevé’s families with rational
particular solutions.

6.1 Painlevé II
We shall obtain a non-integrability result for a discrete subfamily of Painlevé II
equations (cf. also [128]). We follow [100], to which we address the reader for more
details and remarks.
The second Painlevé transcendent is given by the solutions of the Painlevé II
equation

ẍ = 2x3 + tx + ↵, (95)
↵ being a complex parameter.
For Painlevé II the Hamiltonian is
1 1 1
H0 (y, x, t) = y 2 (x2 + t)y (↵ + )x,
2 2 2
and the di↵erential equation (95) is equivalent to the Hamiltonian system

@H0 1 @H0 1
ẋ = =y x2 t, ẏ = = 2xy + ↵ +
@y 2 @x 2
([84, 106]).
6.1 Painlevé II 63

Now, by a standard procedure in Hamiltonian dynamics, from the above non-

autonomous Hamiltonian system we can obtain a two degrees of freedom autonomous
Hamiltonian system such that the non-autonomous system is included as a subsys-
tem. For the Hamiltonian H0 , it is given by

H(y, x, z, e) = H0 (x, y, z) + e.
Thus, the associated Hamiltonian system is

ẋ = y x2 12 z,
ẏ = 2xy + ↵ + 12 ,
(96)
ż = 1,
ė = 12 y.
It seems clear that the dynamical system (96) is equivalent to the Painlevé II
equation (95), in the sense that from the solutions of one we can immediately obtain
the solutions of the other. In particular, for any reasonable meaning of the word
“integrable”, the integrability of one of
R them implies the integrability of the other.
We remark that the function e(t) = 12 y(t)dt is closely related to the ⌧ function of
the Painlevé equation (95) ([107]).
The variational equation along : x = x(t), y = y(t), z = z(t), e = e(t) is
0 1 0 1
10 1
⇠1 2x(t) 1 2
0 ⇠1
d B C B
B ⇠2 C = B 2y(t) 2x(t) 0 0 C B ⇠2 C .
CB C
@ A @ A @ (97)
dt ⇠3 0 0 0 0 ⇠3 A
1
⇠4 0 2
0 0 ⇠4
The normal variational equation is given by
✓ ◆ ✓ ◆✓ ◆
d ⇠1 2x(t) 1 ⇠1
= . (98)
dt ⇠2 2y(t) 2x(t) ⇠2
Given a di↵erential system
✓ ◆ ✓ ◆✓ ◆
d ⇠1 a(t) b(t) ⇠1
= , (99)
dt ⇠2 c(t)) d(t)) ⇠2
with coefficients in a di↵erential field K, by an elimination process it is equivalent
to the second order equation

ḃ(t) ˙ a(t)ḃ(t)
⇠¨ (a(t) + d(t) + )⇠ (ȧ(t) + b(t)c(t) a(t)d(t) )⇠ = 0, (100)
b(t) b(t)

where ⇠ := ⇠1 . We remark that the equations (99) and (100) are equivalent in the
sense that they represent the same D-module (see [138]). In particular, the Galois
groups of both equations are the same.
Hence the normal variational equation (98) is equivalent to the second order
equation
64 6 AN APPLICATION TO PAINLEVÉ’S TRANSCENDENTS

⇠¨ (2y(t) 2ẋ(t) + 4x2 (t))⇠ = 0. (101)

Now by using the Hamilton equations (96) and taking z(t) = t, we obtain

⇠¨ (6x2 (t) + t)⇠ = 0. (102)

Now we fix ↵ = 1. Then it is well-known that the equation (95) has the particular
solution (see, for instance, [40])

1
x= (103)
t
and the associated Hamiltonian system (96) has the particular rational solution

1 2 t 1 t2
: x(t) = , y(t) = 2 + , z(t) = t, e(t) = + . (104)
t t 2 t 8
For this particular solution, (102) is given by

6
⇠¨ ( + t)⇠ = 0. (105)
t2
By means of the change of variable ⇠(t) = t1/2 ⌘(x), x = i 23 t3/2 , it is transformed into
Bessel’s equation

d2 ⌘ d⌘
x2 2
+ x + (x2 n2 )⌘ = 0, (106)
dx dx
with n = 5/3.
Now it is well-known that when n 2 / Z + 1/2 the identity component of Galois
group of Bessel’s equation is non-commutative, indeed, for these values the Galois
group is SL(2, C) (see, [94], Subsection 2.8.2, for a simple proof using Stokes ma-
trices). As the point at z = t = 1 is an irregular singular point of the variational
equation, by Theorem 5, we have proven the following proposition:

Proposition 5. For ↵ = 1, the Hamiltonian system (96) associated to the Painlevé

II equation is not integrable by means of rational first integrals.

Furthermore, it a classical fact that not only for ↵ = 1, but for any integer ↵
the Painlevé II equation has rational particular solutions (such a solution is (103)
for ↵ = 1) and there are rational changes of variables in the phase variables called
Bäcklund (or canonical) transformations between the members of this discrete family
of Hamiltonian systems ([40, 107]). Hence if one of them is non-integrable by rational
first integrals, any member of this family satisfies the same property. We have proven
the following:

Corollary 2. For ↵ 2 Z, the Hamiltonian system (96) associated to the Painlevé

II equation is not integrable by means of rational first integrals.
6.2 Painlevé VI 65

6.2 Painlevé VI
The Painlevé VI transcendent is given by the solutions of the Painlevé VI equation

ẍ = 12 x1 + x 1 1 + x1 t ẏ 2 1
t
+ t 1 1 + x1 t ẏ
(107)
+ x(xt2 (t1)(x
1)2
t)
↵ + xt2 + (t 1)
(y 1)2
+ (yt(t 1)
t)2
↵, , , being complex parameters.
There are other ways to write the parameters, more natural in the interpretation
2 ✓2
of Painlevé VI as related to an isomonodromic deformation: ↵ = (✓4 2 1) , = 21 ,
✓2 1 ✓2
= 22 , = 2 3 .
The case ↵ = = = 0, = 12 (✓1 = ✓2 = ✓3 = 0, ✓4 = 1) was studied by E.
Picard before Painlevé discovery of Painlevé equations, it is called Picard-Painlevé
case: P P6 .
In [45] the authors proved that Painlevé VI with parameters ↵ = = =
= 0 (and more generally with parameters related by Backlünd transformations)
is non integrable by means of meromorphic first integrals. We will only sketch
their proof and we address the reader to [45] for more details and remarks. In
this case it is necessary to use the second variational equation and the authors
obtain an obstruction using a special function, the dilogarithm. More generally
we can think that in the case of an “elementary” first variational equation with
fuchsian logarithmic singularities only at 0, 1, 1 2 P1 and an abelian monodromy,
the polylogarithms could be an efficient tool to obtain an obstruction to integrability
by means of higher variational equations.
We suppose ↵ = = = = 0. Then (107) is equivalent to the Hamiltonian
system

2x(x 1)(x t)
ẋ = y
t(t 1)
y2
ẏ = (x 1)(x t) + x(x 1) + x(x t)
t(t 1) (108)
ṫ = 1
1 2t 1
ė = x(x 1)y 2 + 2 x(x 1)(x t)y 2 .
t(t 1) t (t 1)2
This system possesses a simple family of solutions x = c, y = 0, e = E, where
c, E are constants. The first normal variational equation is

2c(1 c)(c t)
⇠˙1 = ⇠2
t(1 t) (109)
⇠˙2 = 0
Choosing ⇠2 = 1, and setting C = c(1 c), we get ⇠1 = 2C c ln(t) + (1 c) ln(1
t) . The corresponding di↵erential Galois group is commutative; thus, there is no
obstruction to integrability at the first order. We remark that this Galois group is
connected, therefore the Galois groups of the higher variational equations are also
connected.
66 6 AN APPLICATION TO PAINLEVÉ’S TRANSCENDENTS

The second order component of the second variational equation (in its non linear
form) is

c 1 c ˙(2) c(3c 2) (1 c)(3c 1) ˙ ˙

⇠˙1 = 2C
(2)
⇠ +2 + ⇠1 ⇠2
t 1 t 2 t 1 t (110)
c(3c 2) (1 c)(3c 1) ˙2
⇠˙2 =
(2)
⇠2
t 1 t
We set A0 = c(3c 2), A1 = (1 c)(1 3c), then the linear version of the sec-
ond variational equation is given by the first variational equation and the following
equation:

0 (2) 1 0 c 1 c A0 A1 10 (2) 1
⇠1 0 2C t 1 t
2 t
+ 1 t
0 ⇠1
B (2)
d B⇠1 C B0C B 0 0 A0
+ A1 C B⇠ (2) C
= t 1 t CB 1 C (111)
dt @ u A @0 0 0 2C c
t
1 c
1 t
A@ u A
v 0 0 0 0 v

where u := ⇠1 ⇠2 , v := ⇠22 .
R t ln(1 s)
We introduce the dilogarithm Li2 (t) = 0 s
ds. Using the monodromies of
Li2 (t) and Li2 (1 t) it is possible to compute two generators of the monodromy of
(111) (in a convenient basis), we obtain
0 1
1 4⇡iCc 4⇡iA0 4⇡ 2 A0 Cc
B0 1 0 2⇡iA0 C
M0 := B@0
C (112)
0 1 4⇡iCc A
0 0 0 1
corresponding to a loop around 0 and
0 1
1 4⇡iC(1 c) 4⇡iA1 4⇡ 2 A1 C(1 c)
B0 1 0 2⇡iA1 C
M1 := B@0
C (113)
0 1 4⇡iCc(1 c) A
0 0 0 1
corresponding to a loop around 1.
The matrices M0 and M1 do not commute for a generic value of c. By theorem 6,
this gives us an obstruction to integrability by means of meromorphic first integrals,
because the Galois group of the second variational equation is connected.

6.3 More about the non integrability of Painlevé I and

Painlevé VI
There is a non-linear version of di↵erential Galois theory due independently to B.
Malgrange [82] and H. Umemura [137]. J. Drach tried to built such a theory in
his thesis but there are some gaps [30]. We will use Malgrange version, because its
6.3 More about the non integrability of Painlevé I and Painlevé VI 67

geometric approach is more adapted to our purposes. Since the non-linear di↵eren-
tial Galois theory is quite technical we will only sketch some ideas without precise
definitions, addressing the reader to the original papers for more details.
We recall that a groupoid is a small category whose all the morphisms are iso-
morphisms. Malgrange introduced Lie D-groupoids (we will say Lie groupoids for
simplicity); roughly speaking they are the subgroupoids of the groupoid of germs of
analytic di↵eomorphisms of an analytic complex manifold M defined by systems of
analytic PDE. A Lie groupoid has a Lie algebra. There is also an algebraic version
when M is an algebraic manifold.
By definition, the Galois groupoid of an analytic singular foliation is the smallest
Lie groupoid among the Lie groupoid whose Lie algebra “contains the tangent pseu-
dogroup of the foliation”. Given a system, its Galois groupoid is the Galois groupoid
of the corresponding foliation; it is defined on the phase space in the autonomous
case and on the extended phase space in the general case.
There is a non linear version of theorem 6 when one replaces the usual di↵erential
Galois theory by the non-linear Galois theory (it is due to the second author, cf.
[101], part 5, p. 27): if a Hamiltonian system is integrable, then the Lie algebra of
its non linear Galois groupoid is abelian.
For Painlevé’s equations the Galois groupoid is defined on the extended phase
space C3 and it always preserves a non trivial closed rational 2-form ! = iX dt ^ dx ^
dy, where X is the vector field on C3 associated to the equation (which is divergence
free).
Painlevé I is the di↵erential equation:

ẍ = 6x2 + t. (114)
3 @ @ 2 @
The associated vector field on the phase space C is X = @t
+ y @x
+ (6x + t) @y .
In this case the preserved form is ! = iX dt ^ dx ^ dy. We have the following result
due to G. Casale [23].
Theorem 19. The Galois groupoid of Painlevé I is the algebraic Lie groupoid on
the phase space preserving the form ! = iX dt ^ dx ^ dy. Its solutions are the germs
of transformations f of C3 such that f ⇤ ! = !.
P. Painlevé and J. Drach “proved” a similar result but there were gaps and errors
in their proofs.
Therefore the Lie algebra of the Galois groupoid of Painlevé I is non abelian and,
applying the non linear version of our theorem, we obtain the following result.
Theorem 20. Painlevé I is not integrable by means of rational functions.
The following result is due to S. Cantat and F. Loray [65] (the proof uses deep
results of G. Casale on non-linear Galois theory and of the japanese school on the
dynamics of Painlevé VI [50]).
Theorem 21. The Galois groupoid of Painlevé VI is the algebraic Lie groupoid on
the phase space preserving the form ! = iX dt ^ dx ^ dy, except in each one of the
cases:
68 6 AN APPLICATION TO PAINLEVÉ’S TRANSCENDENTS

1
• ✓j 2 2
+ Z, j = 1, 2, 3, 4;

• ✓j 2 Z, j = 1, 2, 3, 4 and ✓1 + ✓2 + ✓3 + ✓4 is odd.

All these cases are equivalent modulo Okamoto symmetries to the case of P P 6:

✓1 = ✓2 = ✓3 = 0, ✓4 = 1.

The following result is due to G. Casale [23].

Theorem 22. The Galois groupoid of Painlevé-Picard VI is transversally affine.

Using Okamoto symmetries we can get similar results in each one of the excep-
tional cases of the preceding theorem. Then, in all the cases, exceptional or not, the
Lie algebra of the Galois groupoid of Painlevé VI is non abelian, and, applying the
non linear version of our results, we obtain the following result.

Theorem 23. For all values of the parameters Painlevé VI is not integrable by mean
of rational functions.

There is some hope to extend the approach of Cantat and Loray [65], using cubic
surfaces, to P2 , P3 , P4 and P5 , replacing the usual dynamics (linear and non-linear
monodromy) by some “Stokes dynamics” (linear and non-linear Stokes phenomena).
Using the non linear version of our theorem we can avoid the choice of a particular
solution, which is very good, but unfortunately there is a price to pay: the proofs
and computations are more difficult.
 69

A Algorithmic Considerations
A.1 Kovacic Algorithm
For the sake of completeness we include here the standard Kovacic algorithm, be-
cause, as it was shown in the previous pages, it is still very useful in the applications
to non-integrability.
The Kovacic algorithm gives us a procedure in order to compute the Picard-
Vessiot extension (i.e., a fundamental system of solutions) of a second order di↵er-
ential equation, provided the di↵erential equation is integrable. Reciprocally, if the
di↵erential equation is non-integrable, the algorithm does not work (see[58]). In this
(necessarily brief) description of the algorithm we essentially follow the version of
the algorithm given in [32, 33].
Given a second order linear di↵erential equation with coefficients in C(x), it is
a classical fact that it can be transformed into the so-called reduced invariant form

⇠ 00 + g⇠ = 0, (115)
with g = g(x) 2 C(x).
We remark that in this change we introduce the exponentiation of a quadrature
and the integrability of the original equation is equivalent to the integrability of the
above equation although, in general, the Galois groups are not the same.
The algorithm is based on the following two general facts:
(A) The classification of the algebraic subgroups of SL(2, C) given in Proposition 2
of Section 2.1 (the Galois group of the equation (115) is contained in SL(2, C)).
(B) The well-known transformation to a Riccati equation, by the change v = ⇠ 0 /⇠,

v0 = g + v2. (116)
Then (see Section 2.2) the di↵erential equation (115) is integrable, if and only if,
the equation (116) has an algebraic solution. The key point now is that the degree
n of the associated minimal polynomial Q(v) (with coefficients in C(x)) belongs to
the set

Lmax = {1, 2, 4, 6, 12}.

The determination of the set L of possible values for n, is the First Step of
the algorithm. We remark that for n = 4, n = 6 and n = 12, the Galois group of
(115) is finite (hence these values are related to the crystalographic groups). The
two other steps of the algorithm (Second Step and Third Step) are devoted to
computation of the polynomial Q(v) (if it exists). If the algorithm does not work
(i.e., if the equation (116) has no algebraic solution) then equation (115) is non-
integrable and its Galois group is SL(2, C).
Now we will describe the algorithm.
Let
s(x)
g = g(x) = ,
t(x)
70 A ALGORITHMIC CONSIDERATIONS

with s(x), t(x) relatively prime polynomials, and t(x) monic. We define the following
function h on the set Lmax = {1, 2, 4, 6, 12}, h(1) = 1, h(2) = 4, h(4) = h(6) =
h(12) = 12.

First Step
If t(x) = 1 we put m = 0, else we factorize t(x) in monic relatively prime
polynomials. Then
1.1. Let 0 be the set of roots of t(x) (i.e., the singular points at the finite complex
plane) and let = 0 [ 1 be the set of singular points. Then the order at a singular
point c 2 0 is, as usual, o(c) = i if c is a root of multiplicity i of t(x). The order at
infinity is defined by o(1) = max(0, 4 + deg(s) deg(t)). We call m+ the maximum
value of the order that appears at the singular points in , and i is the set of
singular points of order i  m+ .
1.2. If m+ 2 then we write 2 = card( 2 ), else 2 = 0. Then we compute
[
= 2 + card( k ).
k odd
3km+

1.3. For the singular points of order one or two, c 2 2 [ 1, we compute the
principal parts of g:

2
g = ↵c (x c) + c (x c) + O((x c)2 ),
0
if c 2 , and

2 3
g = ↵1 x + 1x + O(x 4 ),
for the point at infinity.
1.4. We define the subset L0 (of possible values for the degree of the minimal
polynomial Q(v)) as {1} ⇢ L0 if = 2 , {2} ⇢ L0 if 2 and {4, 6, 12} ⇢ L0 if
+
m  2.
1.5. We have the three following mutually exclusive cases:
1.5.1. If m+ > 2, then L = L0 .
1.5.2. If m+  2 and the two following conditions are satisfied:
p P
1.5.2.1. For any c 2 , 1 + 4↵c 2 Q, and c2 0 c = 0,
p
1.5.2.2. For any c 2 such that 1 + 4↵c 2 Z, logarithmic terms do not appair
in the local solutions in a neigbourhood of c,
then L = L0 .
1.5.3. If cases 1.5.1 and 1.5.2 do not hold then L = L0 {4, 6, 12}.
1.6. If L = ;, then equation (115) is non-integrable with Galois group SL(2, C),otherwise
one writes n for the minimum value in L.
A.1 Kovacic Algorithm 71

We remark that condition 1.5.2.2 is not stated in the original Kovacic’s paper.
As the reader can check, it follows trivially from the fact that the existence of a
logarithm in a local solution is an obstruction to a finite monodromy and Galois
group. We decided to include this condition here because it has successfully been
applied to some important Hamiltonian systems ([52, 51]).

For the Second Step and the Third Step of the algorithm we consider the
value of n fixed.
Second Step
2.1. If 1 has order 0 we write the set
h(n) h(n) h(n) h(n)
E1 = {0, ,2 ,3 , ..., n }.
n n n n

2.2. If c has order 1, then Ec = {h(n)}.

2.3. If n = 1, for each c of order 2 we define

1 p 1 p
Ec = { (1 + 1 + 4↵c , (1 1 + 4↵c }.
2 2

2.4. If n 2, for each c of order 2, we define

h(n) p h(n) p
Ec = Z \ { (1 1 + 4↵c ) + k 1 + 4↵c : k = 0, 1, ..., n}.
2 n

2.5. If n = 1, for each singular point of even order 2⌫, with ⌫ > 1, we compute the
numbers ↵c and c defined (up to a sign) by the following conditions:
0
2.5.1. If c 2 ,

⌫ 1
X
⌫
g = {↵c (x c) + µi,c (x c) i }2 + c (x c) ⌫ 1
+ O((x c) ⌫ ),
i=2

and we write
⌫ 1
X
p ⌫
g c := ↵c (x c) + µi,c (x c) i .
i=2

2.5.2. If c = 1,
⌫ 3
X
g = {↵1 x⌫ 2
+ µi,1 xi }2 1x
⌫ 3
+ O(x⌫ 4 ),
i=0

and we write
72 A ALGORITHMIC CONSIDERATIONS

⌫ 3
X
p
g 1 := ↵1 x⌫ 2
+ µi,1 xi .
i=0

Then for each c as above, we compute

1 c
Ec = { (⌫ + ✏ ) : ✏ = ±1},
2 ↵c
and the sign function on Ec is defined by

1 c
sign( (⌫ + ✏ )) = ✏,
2 ↵c
being +1 if c = 0.
2.6. If n = 2, for each c of order ⌫, with ⌫ 3, we write Ec = {⌫}.

Third Step
3.1.
Q For n fixed, we try to obtain elements e = (ec )c2 in the cartesian product
c2 Ec , such that:
n
P
(i) d(e) := n h(n) c2 ec is a non-negative integer,
(ii) If n = 2 then there is at least one odd number in e.
If no element e is obtained, we select the next value in L and go to the Second
Step, else n is the maximum value in L and the Galois group is SL(2, C) (i.e., the
equation (115) is non-integrable).
3.2. For each family e as above, we try to obtain a rational function Q and a
polynomial P , such that
(i)
n X ec X p
Q= + n1 sign(ec ) g c ,
h(n) c2 0 x c c2[⌫>1 2⌫

where n1 is the Kronecker delta.

(ii) P is a polynomial of degree d(e) and its coefficients are found as a solution of
the (in general, overdetermined) system of equations

P 1 = 0,
Pi 1 = (Pi )0 QPi (n i)(i + 1)gPi+1 , n i 0,
Pn = P.
If a pair (P, Q) as above is found, then equation (115) is integrable and the Riccati
equation (116) has an algebraic solution v given by any root v of the equation
n
X Pi
v i = 0.
i=0
(n 1)!
A.2 Algebrization Procedure 73

If no pair as above is found we take the next value in L and we go to the Second
Step. If n is the greatest value in L then equation (115) is non-integrable and the
Galois group is SL(2, C).

Under some assumptions, in the literature there are other algorithms to compute
the Galois group of a linear di↵erential equation. We notice that a remarkable
simplification of the above algorithm was obtained in [136] for irreducible di↵erential
equations. An algorithm for third order di↵erential equations is given in [124, 125].
For completely reducible equations, i.e. when the Galois group is reductive, an
algorithm is presented in [29].

A.2 Algebrization Procedure

In concrete di↵erential equations it is useful, if possible, to replace the original dif-
ferential equation over a compact Riemann surface, by a new di↵erential equation
over the Riemann sphere P1 (i.e., with rational coefficients) by a change of the inde-
pendent variable. This equation on P1 is called the algebraic form of the equation.
In a more general way we will consider the e↵ect of a finite ramified covering on the
Galois group of the original di↵erential equation.
Theorem 24. ([96], see also [94]) Let be a Riemann surface,
d
⇠ = A(x)⇠, A 2 M at(m, C(x)) (117)
dx
a linear di↵erential equation on P1 and x : ! P1 , x = x(t) a finite ramified
1
covering of P (t a local parameter in ). Let
d
⇠ = x⇤ (A)(t)⇠, x⇤ (A) 2 M at(m, M( )) (118)
dt
be the pull-back of equation (117) by x (i.e., the equation obtained by the change
of variables x = x(t)). Then the identity components of the Galois groups of the
equations (117) and (118) are the same.
We say that a linear di↵erential equation
d
⇠ = A0 (t)⇠, A0 2 M at(m, M( )) (119)
dt
is algebrizable if it is the pull-back of a linear di↵erential equation (117). In order
to apply Kovacic’s algorithm it is important to know whether a given second order
linear di↵erential equation is algebrizable. An algorithm to algebrize equations of
the type
ÿ = r(t)y
is proposed recently in [2].
We say that a change of variable x = x(t) is Hamiltonian if and only if (x(t), ẋ(t))
is a solution curve of the autonomous 1-degree of freedom Hamiltonian system
ẋ2
H = H(x, ẋ) = + V (x).
2
74 A ALGORITHMIC CONSIDERATIONS

Proposition 6 (Algebrization algorithm). ([2]) The di↵erential equation

ÿ = r(t)y (120)

is algebrizable through a Hamiltonian change of variable x = x(t) if, and only if,
there exist f, ↵ such that
↵0 f
, 2 C(x), where f (x(t)) = r(t), ↵(x) = 2(h V (x)) = ẋ2 .
↵ ↵
Furthermore, the algebraic form of the equation ÿ = r(t)y is

1 ↵0 0 f
y 00 + y y = 0, (121)
2↵ ↵
where 0 = d/dx.
From the above we know that when r(t) belong to the field of meromorphic
functions over , M( ), then the identity component of the Galois group is preserved
by the above change of variables x = x(t).

A.3 The importance of logarithmic terms

We observe that in the Kovacic algorithm we include a logarithmic condition. This is
not an isolated behavior: the appearance of logarithmic terms in the solution of the
variational equations is very often an obstruction to the integrability. In our opinion
this fact gives some insight about the success of the so-called Kovalewskaya-Painlevé
heuristic analysis, although a complete clarification remains open.
In Section 3.4 we already found that in the Hill Problem the existence of a
logarithmic term was essential for proving its non-integrability. There are other
possible situations where the logarithmic terms are essential.
a) Higher order variational equations. Assume that the first order variational
equation VE 1 of a Hamiltonian system around a particular solution has a commu-
tative identity component of the Galois group G01 . At first order we cannot obtain
obstruction to the integrability. For the moment the only family of systems where
Theorem 6 was applied with success is the following; we follow [94], chapter 8.
Consider a two degrees of freedom Hamiltonian system with a first order vari-
ational equation, VE 1 , such that is given by the direct sum of two Lamé type
equations (see Appendix C):

⇠¨1 = n1 (n1 + 1) }(t) + B1 ⇠1 ,

(122)
⇠¨2 = n2 (n2 + 1) }(t) + B2 ⇠2 ,
where the field of coefficients of (122), as well as the field of coefficients of the VE k
for k > 1, is a field of elliptic functions K = C(}(t), }(t)),
˙ isomorphic to the field of
meromorphic functions over . For simplicity we consider the case of a first order
variational equation given by only two Lamé type equations but in an obvious way
it can be generalized to an n–degrees–of–freedom Hamiltonian system with a first
A.3 The importance of logarithmic terms 75

order variational equation given by a direct sum of n Lamé type equations. One
of the equations (122) is the first order normal variational equation, say the second
one.
Let K ⇢ L1 ⇢ L2 ⇢ · · · Lk the Picard-Vessiot extension when we solve VE k .
From Section 2.3 we know that once the solutions of VE 1 , K ⇢ L1 , are obtained
the solutions of the second order, third order, etc., L1 ⇢ L2 ⇢ L3 · · · are obtained
by the method of variation of constants. So, to get the extension Lk /L1 we only
use quadratures, this extension is a purely transcendental one and by Picard-Vessiot
general theory (Section 2.1) the Galois group Gal(Lk /L1 ) is connected. If the Galois
group G1 is also connected then the extension K ⇢ L1 is also transcendental, the
total Picard-Vessiot extension K ⇢ Lk of VE k is transcendental and Gk = (Gk )0 .
We have proven the following:

Lemma 10 ([94]). Under the above assumptions, Gk = (Gk )0 if and only if G1 =

(G1 )0 .

The variational equation VE 1 is Fuchsian. Therefore all higher order variational

equations VE k (more precisely, their linear counterparts) are also Fuchsian (for
k 1, see [101] for the details) and Gk is the Zariski closure of their monodromy
groups.
We have the following lemma.

Lemma 11 ([94]). Assume that the first order variational equation VE 1 splits into
a direct sum of Lamé–type equations with n1 , n2 integers (122). Then Gk is com-
mutative if and only if the solutions of VE k are meromorphic functions with respect
to variable t.

Proof. The proof is easy. The monodromy group of each of the VE k is a linear rep-
resentation of the fundamental group of = {1} (the point 1 is represented in
the Weierstrass parametrisation by the origin modulo periods) and this fundamental
group is free, non-commutative and generated by the translations along the periods.
The commutator of these two generators is represented by a simple loop around the
singular point 1. Hence, a monodromy group is commutative if and only if the
monodromy associated to this simple loop is trivial. By Zariski closure, a di↵eren-
tial Galois group Gk is commutative if and only if the corresponding monodromy
subgroup is commutative. 2
Therefore we can check the commutativity of Gk locally at 1. Recursively, by
local power series expansions of the solutions of VE k 1 and quadratures, it is easy
to check whether VE k has branched solutions around 0. One only needs to check
for the existence of a residue di↵erent from zero, which will give rise by integration,
when we apply the method of variations of constants, to a local logarithm.
Assume both equations in (122) fall in the Lamé case with a particular solution
in the coefficient field of elliptic functions (see Appendix C); then the Galois group
of (122) is given by unipotent matrices of the type
76 A ALGORITHMIC CONSIDERATIONS

0 1
1 0 0 0
B ↵ 1 0 0 C
B C, (123)
@ 0 0 1 0 A
0 0 1
and G1 ⇢ (C2 , +). Necessarily G1 is trivial or either (C, +) or (C2 , +). In any case
G1 = (G1 )0 is commutative. Hence, by Lemma 10 Gk = (Gk )0 and by Lemma 11
(Gk )0 is commutative if, and only if, the solutions of VE k are meromorphic functions
in the variable t.
In fact we can go further in our analysis. The following result is new.

Lemma 12. Assume that the first order variational equation VE 1 splits into a direct
sum of Lamé–type equations with n1 , n2 integers (122) and that the Galois group of
the first order normal variational equation is not finite. Then Gk is commutative if
and only if (Gk )0 is commutative.

Proof. We first study the Galois group G1 of the first order variational equations.
The tangential variational equation has a solution in the field of meromorphic func-
tions over the elliptic integral curve , K = M( ), it falls into the Lamé case and its
Galois group is connected (Appendix C). Hence, we reduce the problem to studying
the normal first order variational equation NVE, for instance the first equation in
(122),
⇠¨1 = n1 (n1 + 1) }(t) + B1 ⇠1 .
This equation falls in either
a) the Lamé or
b) Hermite case.
In the first case, one of the particular solutions is a Lamé function ⇠1 either
belonging to K or to a quadratic extension K of K, the other independent solution
being transcendent, see Appendix C. If ⇠1 2 K, then G1 = G01 , therefore Gk = G0k
and the result is trivial. Therefore we can assume ⇠1 62 K. We only have to prove that
if (Gk )0 is commutative then Gk is also commutative, the converse being evident.
We assume the contrary: there exists k 2 N, k 2, such that G0k is commutative
and such that Gk is not commutative – we can assume k minimal: Gk0 is commutative
for all 1  k 0 < k. Then, by Lemma 11, for every 1  k 0 < k, the solutions of V Ek0
are meromorphic functions in the variable t.
K is the field of meromorphic functions of the elliptic curve E = C/(2Z!1
2Z!3 ). We consider, see Appendix C, the field K1 of meromorphic functions of the
elliptic curve E1 = C/(4Z!1 4Z!3 ). The identity of C induces a map ⇡ : E1 ! E
and an inclusion of fields K ⇢ K1 , ⇡ is a covering of order 4 and Gal(K1 /K) is
a group of order 4 isomorphic to Z2 Z2 . We have di↵erential field inclusions
K ⇢ K ⇢ K1 and Gal(Lk /K) = G0k , therefore the action of the monodromy of E1
on the solutions of V Ek is abelian.
We consider the parallelogram P ⇢ C defined by the four points !1 !3 , 3!1
!1 , 3!1 + 3!3 , !1 + 3!3 . It is a fundamental domain for the elliptic curve E1 =
A.3 The importance of logarithmic terms 77

C/(4Z!1 4Z!3 ) and the oriented boundary @P of P corresponds to the commu-

tator of two fundamental loops of E1 .
If Gk is not commutative, then there exists a solution of V Ek with a ramification
at zero (mod (!1 , !3 )), which is the unique singularity of V Ek (mod. (2!1 , 2!3 ));
therefore, if we compute the solutions of V Ek using expressions ⌃ in the solutions
of the V Ek0 , for 1  k 0 < k, and a quadrature, we see that one of the scalar
components ⇣ of one of the expressions ⌃ must be a meromorphic function in the
variable t, with a pole at 0 such that the corresponding residue a is not trivial (if
it is not the case, then all the solutions of V Ek in the variable t are meromorphic
and Gk is commutative). If we interpret ⇣ as a (perhaps ramified) function on E1 ,
i.e. mod. (4!1 , 4!3 ), we get exactly four poles, at 0, 2!1 , 2!2 , 2!1 + 2!2 (using
the variable t, there are 4 poles in the fundamental domain P), the corresponding
residues of ⇣ being respectively a, a, a, a (the respective
R singularities
R correspond by
translations). Then the action of the loop @P on ⇣(t)dt is ⇣(t)dt + 8ai⇡ and is
not trivial. Therefore the action of the commutator of two fundamental loops of E1
on some particular solution of V Ek is not trivial. This is a contradiction.
Assume we are in case b). The proof is now easier, because the Galois group
of the NVE must then be the multiplicative group, since we assume that its Galois
group is not finite. Hence, the Galois group G1 is connected, the higher order Galois
group Gk being also connected: Gk = (Gk )0 . 2
We observe that in the proof we used the fact that the tangential variational
equation always falls in the Lamé case, one of its solutions being in the field K of
elliptic functions: one particular solution is given by the temporal derivative of the
function which defines the integral curve . On the other hand, we know that in
the Lamé case the Galois group is not finite (see Appendix C). Then, as a corollary
of the above Lemmas and of Theorem 6, we obtained the following logarithmic
non-integrability criterium.

Proposition 7. Assume that:

1. The first order variational equation VE 1 splits in a direct sum of Lamé type
equations, (122), with n1 , n2 integers (the tangential and the normal varia-
tional equations, NVE),

2. the Galois group of the NVE is not finite.

Then a sufficient condition for meromorphic non-integrability is the existence of a

local logarithm at t = 0 in a particular solution of a higher order variational equation
VE k , k > 1.

We remark that in all applications of this proposition considered in this con-

tribution (for example, in Section 2.4) we have a two-degree-of-freedom Hamilto-
nian system with a potential V = V (x1 , x2 ) and an invariant plane (for instance,
x2 = y2 = 0). Then the integral curve is defined by x2 = y2 = 0, x1 = x1 (t) 2 K,
y1 = y1 (t) = ẋ1 (t) 2 K, and y1 (t) is a particular solution of the tangential varia-
tional equation. Furthermore, the above proposition is currently the only practical
78 A ALGORITHMIC CONSIDERATIONS

criterion being used in connection with our general higher order non-integrability
theorem (Theorem 6) and specifically exploiting V Ek for k > 1.
b) Boucher–Weil criterion. In their studies on the integrability of the Three-
Body Problem, Boucher and Weil introduced the following criterion for the non–
commutativity of the Galois group. If the NVE is a system of dimension 2(n m), by
the cyclic vector method, it is possible to obtain a scalar linear di↵erential equation
of order 2(n m),

L(⇠) = 0, (124)
being L a linear di↵erential operator of order 2(n m) ([138]). In fact, the elimination
process for obtaining equation (100) from (99) in Chapter 6 is a particular case of this
method. Then using Theorem 4, and their own result about the non–commutativity
of the identity component of the Galois group for equation (124) in presence of
logarithmic terms, they obtained the following theorem.

Theorem 25 ([15, 16, 18]). If equation (100) has a completely reducible factor whose
local solutions at a singular point contain logarithmic terms, then the Hamiltonian
system XH is not integrable with meromorphic first integrals.

A particular case of the above theorem is for a NVE with only one irreducible
factor. Thus, if equation (100) is irreducible and has local solutions at a singular
point with logarithmic terms, then the Hamiltonian system is not integrable.
We illustrate this criterion with an example taken from the study of a FRW
cosmological model in [19]. The Hamiltonian is given by

1 1 ⇤ 4
H = (y12 + y22 ) + x22 + x21 m2 x21 x22 + x + x4 , (125)
2 2 2 1 2 1
obtained from the Hamiltonian HF RW of Chapter 5 for k = 1 and with the usual
change (x1 , y1 ) ! ( ix1 , iy1 ).
As was said in Section 5.2, this Hamiltonian has the invariant plane x1 = y1 = 0
which, on the energy level h = 0, defines a particular solution with NVE ; further-
more, by means of the algebrization procedure of Section A.2, an algebraic form is
obtained for the NVE,

2 2 2m2 2m2
(3x 1)(3x + 1)2 ⇠ 00 + (3x + 1)(3x 1)⇠ 0 + x+1+ ⇠=0 (126)
3 3 3
This equation is Fuchsian with three (regular) singular points: x = 1/3, 2/3
and 1 (it can be reduced to an hypergeometric equation, but we do not use this
fact here).
The exponents at the singular point x = 1/3 are 1/2 and 1/2 (roots of the
indicial equation at that point). Then the di↵erence of exponents is an integer, and
if one of the solutions is ⇠1 = (3x+1)1/2 f (x), f (x) without singularities at x = 1/3,
the other solution has a logarithmic term provided m is di↵erent from zero.
A.3 The importance of logarithmic terms 79

In order to apply Theorem 25, we must study the necessary conditions for re-
ducibility. At the singular point x = 2/3 the exponents are ↵1 = 0 and ↵2 = 1/2
and at x = 1 the indicial equation is

2 ⇢2 ⇢ + m2 = 0.
Using the reference [120] the authors obtained the following. If the equation (126)
is reducible it must have an exponential solution of the type

⇠ = (3x + 1)1/2 (3x 2)↵i P (x), (127)

where ↵i is one of the exponents at x = 2/3 (i.e., either 0 or 1/2), P (x) is a
polynomial of degree d and
1
d+ + ↵i + ⇢i = 0,
2
⇢i being one of the roots of the indicial equation at 1.
Then a necessary condition for the existence of the exponential solution (127) is
obtained, namely that ⇢i be equal to either d 1 or d 1/2. And this implies

2m2
= ,
(p + 1)(p + 2)

with p 2 N. By an obvious symmetry argument, an analogous condition is obtained

for the other parameter ⇤. In this manner, Boucher and Weil recovered the necessary
integrability conditions obtained in Coelho, Skea and Stuchi in [27] (see Section 5.2).
In a similar way it is possible to study the commutativity of the identity com-
ponent of the Galois group in presence of irregular singular points; see the example
in [16], pag. 98 , where this method is applied to a Hénon-Heiles type Hamiltonian
system with a NVE of Bessel type, studied in [96, 94] by means of other methods.
80 B HYPERGEOMETRIC EQUATION

B Hypergeometric Equation
The hypergeometric (or Riemann) equation is the most general second order linear
di↵erential equation over the Riemann sphere with three regular singular singulari-
ties. If we place the singularities at x = 0, 1, 1 it is given by

d2 ⇠ 1 ↵ ↵0 1 0
d⇠
+ ( + )
dx2 x x 1 dx
↵↵0 0 0
↵↵0 0
+ ( 2 + + )⇠ = 0, (128)
x (x 1)2 x(x 1)
where (↵, ↵0 ), ( , 0 ),( , 0 ) are the exponents at the singular points and must satisfy
the Fuchs relation ↵ + ↵0 + + 0 + + 0 = 1. We denote the exponent di↵erences
by ˆ = ↵ ↵0 , ⌫ˆ = 0
and µ̂ = 0
.
We also use one of its reduced forms
d2 ⇠ c (a + b + 1)x d⇠ ab
+ ⇠ = 0, (129)
dx2 x(x 1) dx x(x 1)
where a, b, c are parameters, with the exponent di↵erences ˆ = 1 c, ⌫ˆ = c a b
and µ̂ = b a, respectively.
Now, we recall a theorem of Kimura giving necessary and sufficient conditions
for the integrability of the hypergeometric equation.
Theorem 26 ([56]). The identity component of the Galois group of the hypergeo-
metric equation (128) is solvable if, and only if, either
(i) at least one of the four numbers ˆ + µ̂ + ⌫ˆ, ˆ + µ̂ + ⌫ˆ, ˆ µ̂ + ⌫ˆ, ˆ + µ̂ ⌫ˆ is
an odd integer, or
(ii) the numbers ˆ or ˆ , µ̂ or µ̂ and ⌫ˆ or ⌫ˆ belong (in an arbitrary order) to
one or more of the following fifteen families
1 1/2 + l 1/2 + m arbitrary complex number
2 1/2 + l 1/3 + m 1/3 + q
3 2/3 + l 1/3 + m 1/3 + q l + m + q even
4 1/2 + l 1/3 + m 1/4 + q
5 2/3 + l 1/4 + m 1/4 + q l + m + q even
6 1/2 + l 1/3 + m 1/5 + q
7 2/5 + l 1/3 + m 1/3 + q l + m + q even
8 2/3 + l 1/5 + m 1/5 + q l + m + q even
9 1/2 + l 2/5 + m 1/5 + q l + m + q even
10 3/5 + l 1/3 + m 1/5 + q l + m + q even
11 2/5 + l 2/5 + m 2/5 + q l + m + q even
12 2/3 + l 1/3 + m 1/5 + q l + m + q even
13 4/5 + l 1/5 + m 1/5 + q l + m + q even
14 1/2 + l 2/5 + m 1/3 + q l + m + q even
15 3/5 + l 2/5 + m 1/3 + q l + m + q even
Here l, m and q are integers.
 81

We recall that Schwarz’s table provides us with the cases for which the Galois
(and monodromy) groups are finite (i.e., the identity component of the Galois group
is reduced to the identity element) and is given by fifteen families. These are given
by families 2–15 of the table above and by the family (1/2 + Z) ⇥ (1/2 + Z) ⇥ Q
(see, for instance, [113]). Since the latter family is already contained in family 1 in
the above table, so are, of course, all families due to Schwartz.
82 C LAMÉ EQUATION

C Lamé Equation
The algebraic form of the Lamé Equation is [113, 140]
d2 ⇠ f 0 (x) d⌘ Ax + B
+ ⇠ = 0, (130)
dx2 2f (x) dx f (x)
where f (x) = 4x3 g2 x g3 , with A, B, g2 and g3 parameters such that the
discriminant of f , 27g32 g23 is non-zero. This equation is a Fuchsian di↵erential
equation with four singular points over the Riemann sphere.
With the well–known change x = }(t), we get the Weierstrass form of the Lamé
equation

d2 ⇠
(A}(t) + B)⇠ = 0, (131)
dt2
where } is the elliptic Weierstrass function with invariants g2 , g3 (we recall that }(z)
is a solution of the di↵erential equation ( dx
dt
)2 = f (x)). It is a 4-parametric family
of equations in the parameters A, B, g2 and g3 . Classically the equation is written
with the parameter n instead of A, with A = n(n + 1). This equation is defined
on a torus ⇧ (a genus one Riemann surface or elliptic curve y 2 = f (x)) with only
one singular point at the origin. It is also a Fuchsian linear di↵erential equation.
Let 2!1 , 2!3 be the two periods of the Weierstrass function } and g1 , g2 their
corresponding monodromies along these periods. If g⇤ represents the monodromy
around the singular point, then g⇤ = [g1 , g2 ] ([140, 113]). The Lamé equation in the
form (131) was intensively studied by Halphen [41].
By Theorem 6 of Appendix A we know that the identity component of the Galois
group is preserved by the change of variables ⇧ ! P1 , t 7! x. The relation between
the monodromy groups of equations (130) and (131) is discussed in [113], Chapter
IX and, from a modern point of view, in [26].
Now the known mutually exclusive cases of closed form solutions of the Lamé
equation (131) are as follows:
(i) The Lamé–Hermite case [35, 41, 113, 140]. In this case n 2 Z the three other
parameters are arbitrary.
(ii) The Brioschi-Halphen-Crawford solutions [12, 35, 41, 113]. Now m := n+ 12 2
N and the parameters B, g2 and g3 must satify an algebraic equation

0 = Qm (g2 /4, g3 /4, B) 2 Z[g2 /4, g3 /4, B],

where Qm has degree m in B. This polynomial is known as the Brioschi determinant.
(iii) The Baldassarri solutions [12]. The condition on n is n+ 12 2 13 Z[ 14 Z[ 15 Z Z,
with additional (involved) algebraic restrictions on the other parameters.
It is possible to prove that the only integrable cases of integrability for the Lamé
equation are the cases (i)–(iii) above. Integrability here means in the Galois sense,
where the coefficient field is the field of elliptic functions C(}(t), }0 (t)), isomorphic
to the field of meromorphic functions on the torus ⇧.
 83

Proposition 8 ([92],[94]). Equation (131) is integrable only in the cases (i), (ii)
and (iii) above.
Corollary 3. A necessary condition for the commutativity of the identity component
of the Galois group of equation (131) is that the latter belong to one of the cases (i),
(ii) or (iii) above.
We recall that the moduli of the elliptic curve y 2 = 4x3 g2 x g3 (we write the
elliptic curve in the canonical form, where as above g2 and g3 are the invariants) is
characterized by the value of the modular function j,

g23
j = j(g2 , g3 ) = . (132)
g23 27g32
We recall that two elliptic curves are birationally equivalent if, and only if, they
have the same value of the modular function (see, for instance [119]).
Although the conditions on g2 , g3 and B for a finite Galois group (case (iii)) are
difficult to systematize, there is, in this case, a general result by Dwork answering a
question posed by Baldassarri in [12].
Proposition 9 ([36]). Assume that the Galois group of equation (131) is finite.
Then for a fixed value of n, the number of possible couples (j, B) is finite.
We note that the proof by Dwork was stated for the algebraic form of the Lamé
equation (equation (130)). But since the identity component of the Galois group
is preserved by a finite covering (Theorem 24 of Appendix A), then the finiteness
of the Galois group of equation (130) is equivalent to the finiteness of the Galois
group of equation (131) (a linear algebraic group is finite if, and only if, its identity
component is trivial) and the result is valid also for equation (131).
The first author is indebted to B. Dwork for sending the above result.
One more reference about the case (iii) of Lamé equation is [81]. This reference
corrected a mistake in the paper [12].
Now we center our attention on the classical Lamé–Hermite case (i). It is easy
to see that a necessary and sufficient condition for the total Galois group of (131)
to be commutative is that n 2 Z. We sketch the steps of the proof. Indeed, this is
a classical well-known necessary and sufficient condition for the monodromy group
M of the equation (131) to be commutative (it is clear that, as M is generated by
g1 and g2 , an equivalent condition for the commutativity of M is g⇤ = 1 (identity),
and the indicial equation at the singularity is ⇢2 ⇢ n(n + 1) = 0, and there is
no logarithmic term for integer n (see [113]). Therefore, since G is topologically
generated by M , it must also be commutative.
There are two excluding cases for (i):
1. (Lamé) There is one solution which is a Lamé function ⇠1 either belonging to
the coefficient field K = C(}(t), }0 (t)) or such that ⇠12 belongs to K ([113]).
Hence one solution belongs to a quadratic extension of the coefficient field.
The other independent solution ⇠2 is transcendent over the field K ([12]). For
a fixed n 2 Z, the parameters B, g2 and g3 must satisfy also an algebraic
equation, 0 = P2n+1 (g2 , g3 , B) 2 Q[g2 , g3 , B], of degree 2n + 1 in B ([41]).
84 C LAMÉ EQUATION

2. (Hermite) There are two particular C-independent solutions ⇠1 , ⇠2 , such that

the product ⇠1 ⇠2 belong to the field K. No other conditions are satisfied for the
parameters, except that n 2 Z and they do not satisfy the algebraic conditions
of the Lamé case ([41, 113, 140].

In the case of Lamé when ⇠1 2 K, the Galois group G of equation (131) is

connected of the type 2 of Proposition 2 in Section 2.1:
⇢✓ ◆
0 1 0
G=G = ,µ2C .
µ 1
When ⇠12 2 K, ⇠1 2 / K, we have 2⇠10 ⇠1 2 K, therefore ⇠10 2 K. We set K =
K(⇠1 ), then Gal(K/K) is a cyclic group of order 2 and K = K(⇠1 , ⇠10 ). Setting
L = K(⇠1 , ⇠2 , ⇠10 , ⇠20 ) = K(⇠2 , ⇠20 ), then Gal(L/K) is connected and isomorphic to the
additive group C.
By the Galoisian correspondence (see Section 2.2), the algebraic closure of K,
in the Picard-Vessiot extension L = K(⇠1 , ⇠2 , ⇠10 , ⇠20 ) is K and we have the chain of
di↵erential fields

K ⇢ K ⇢ L,
0 0
with G/G = Gal(K/K) and G = Gal(L/K).
In this case, the Galois group G of equation (131) is of the type 3 of Proposition
2 of Section 2.1:
⇢✓ ◆
0
G = G2 = 1 , is a 2-root of unity, µ 2 C ,
µ
⇢✓ ◆
1 0
G0 = ,µ2C .
µ 1
The extension K ⇢ L is a purely transcendent Picard-Vessiot extension with an
associated linear di↵erential equation which can be made explicit by means of the
Halphen transformation [41, 113].
The field K is the field of meromorphic functions of the elliptic curve E =
C/(2Z!1 2Z!3 ). We consider the field K1 of meromorphic functions of the elliptic
curve E1 = C/(4Z!1 4Z!3 ). The identity of C induces a map ⇡ : E1 ! E and an
inclusion of fields K ⇢ K1 , ⇡ is a covering of order 4 and Gal(K1 /K) isomorphic to
Z2 Z2 . We can interpret K as a subfield of K1 (see [113]). First we perform the
change of independent variable t = 2⌧ (which induces an isomorphism between the
elliptic curves E1 and E) and use the addition theorem for } (see [113]) we obtain
 ⇣ 1 ⇣ }00 (⌧ ) ⌘2 ⌘
d2 ⇠
4 n(n + 1) 2}(⌧ ) + B ⇠ = 0. (133)
d ⌧2 4 }0 (⌧ )
Now, in order to complete the Halphen transformation, we perform the change
n
⇠ = }0 (⌧ ) ⌘, obtaining

d2 ⌘ }00 (⌧ ) d⌘
2n + 4 n(2n 1) }(⌧ ) B ⌘ = 0, (134)
d ⌧2 }0 (⌧ ) d⌧
 85

with singularities at ⌧ = 0 (modulo the periods (2!1 , 2!3 )). In other words, equation
(134) corresponds to the Picard-Vessiot extension K ⇢ L with a connected Galois
group G0 and, just like Lamé’s equation, it is also a Fuchsian linear di↵erential
equation defined over an elliptic curve (a copy of the initial one) with only one
singular point.
For some other applications of the Lamé equation, di↵erent from those considered
in this contribution, see [94], Chapters 6 and 7.
86 REFERENCES

References
[1] M. Abramowitz, I.A. Stegun Editors, I. A. Handbook of mathematical functions
with formulas, graphs, and mathematical tables, John Wiley & Sons Inc., New
York, 1984.

[2] P. Acosta-Humánez, D. Blázquez-Sanz, Non-integrability of some Hamiltonians

with rational potentials, Discrete Contin. Dyn. Syst. 10 (2008), 265–293.

[3] A. Almeida, A. López-Castillo, T. Stuchi, Non-integrability proof of the frozen

planetary atom configuration, J. Phys. A: Math. Gen. 36 (2003) 4805–4814.

[4] A. Almeida, T.J. Stuchi, The integrability of the anisotropic Stormer problem
with angular momentum, Physica D 189 (2004) 219–233.

[5] V.I. Arnold, Mathematical methods in classical mechanics. Springer-Verlag,

Berlin, 1978.

[6] M. Arribas, A. Elipe, Non-integrability of the motion of a particle around a

massive straight segment, Physics Letters A281 (2001) 142–148.

[7] M. Arribas, A. Elipe, A. Riaguas, Non-integrability of anisotropic quasihomo-

geneous Hamiltonian systems, Mech. Res. Comm. 30 (2003) 209–216.

[8] M. Audin, “Les systèmes Hamiltoniens et leur intégrabilité”, Cours Spécialisés,

Collection SMF 8 Société Mathematique de France, Marseille 2001.

[9] M. Audin, La réduction symplectique appliquée à la non-intégrabilité du

problème du satellite, Ann. Fac. Sci. Toulouse Math. 12 (2003) 25–46.

[10] M. Audin Exemples de hamiltoniens non intégrables en mécanique analytique

réelle, Ann. Fac. Sci. Toulouse Math. 12 (2003), 1-23

[11] M. Audin, Mon choix de Sophie, preprint 2006 (http://www-irma.u-

strasbg.fr/ maudin/choix-sophie.pdf).

[12] F. Baldassarri, On algebraic solutions of Lamé’s di↵erential equation, J. of Di↵.

Eq. 41, (1981), 44-58.

[13] B. S. Bardin, A. J. Maciejewski, A. J., M. Przybylska, Integrability of general-

ized Jacobi problem, Regul. Chaotic Dyn. 10 (2005), 437–461.

[14] F. Beukers, Di↵erential Galois Theory, From Number Theory to Physics,

W.Waldschmidt, P.Moussa, J.-M.Luck, C.Itzykson Ed., Springer-Verlag, Berlin
1995, 413–439.

[15] D. Boucher, Sur la non-intégrabilité du problème plan des trois corps de masses
égales, C.R. Acad. Sci. Paris SérieI 331 (2000) 391–394.
REFERENCES 87

[16] D. Boucher, Sur les équations di↵érentielles paramétrées, une application aux
systèmes hamiltoniens, Thèse Faculté des Sciencies de Limoges 2000.

[17] D. Boucher, Non complete integrability of a satellite in circular orbit. Port.

Math. (N.S.) 63 (2006) 69–89.

[18] D. Boucher, J.-A. Weil, Application of J.-J. Morales and J.-P. Ramis’ theorem
to test the non-complete integrability of the planar three-body problem. From
combinatorics to dynamical systems, 163–177, IRMA Lect. Math. Theor. Phys.,
3, de Gruyter, Berlin, 2003.

[19] D. Boucher, J.-A. Weil, On the non-integrability in the Friedmann-Robertson-

Walker Cosmological Model Brazilian J. Phys. 37 (2007) 398–405.

[20] A. Borel, Linear algebraic groups, Springer-Verlag, New-York, 1991.

[21] R. Carter, G. Seagal, I. Macdonald, Lectures on Lie Groups and Lie Algebras,
Cambridge University Press, Cambridge, UK, 1995.

[22] G. Casale, Sur le groupoı̈de de Galois d’un feuilletage. Thèse, Toulouse 2004.

[23] G. Casale, Le groupe de Galois de P1 et son irreductibilité, Comment. Math.

Helv. 83 (2008) 471–519.

[24] G. Casale, The Galois groupoid of Picard Painlevé VI equation, in Algebraic,

analytic and geometric aspects of complex di↵erential equations and their de-
formations, Painlevé hierarchies RIMS Kôkyûroku Bessatsu, B2 (2007), 15–20.

[25] R.C. Churchill, J. Delgado, D.L. Rod, M. Alvarez, J. Delgado, The spring-
pendulum system and the Riemann equation, New trends for Hamiltonian sys-
tems and celestial mechanics (Cocoyoc, 1994), 97–103, Adv. Ser. Nonlinear
Dynam. 8, World Sci. Publ., River Edge, NJ, 1996

[26] R.C. Churchill, Two Generator Subgroups of SL(2, C) and the Hypergeometric,
Riemann and Lamé Equations, J. Symbolic Computation 28 (1999) 521–545.

[27] L.A.A. Coelho, J.E.F. Skea, T.J. Stuchi, On the Non-integrability of a Class of
Hamiltonian Cosmological Models, Brazilian J. Phys. 35 (2005) 1048–1049.

[28] L.A.A. Coelho, J.E.F. Skea, T.J. Stuchi, Friedmann Robertson Walker models
with Conformally Coupled Massive Scalar Fields are Non-integrable, preprint
2006.

[29] E. Compoint, M.F. Singer, Computing Galois groups of Completely Reducible

Di↵erential Equations, J. Symbolic Computation 11 (1998) 1–22.

[30] J. Drach, Essai sur une théorie générale de l’intégration et sur la classification
des transcendantes Ann. Sci. École Normale Sup. 15 (1898) 243–324.
88 REFERENCES

[31] H. Dullin, A. Tsygvintsev, On the analytic non-integrability of the Rattleback

problem, Annales de la faculté des sciences de Toulouse, 17 (2008)495-517.

[32] A. Duval, M. Loday-Richaud, Kovacic’s algorithm and its application to some

families of special functions, Applicable Algebra in Engineering, Communication
and Computing 3 (1992), 211-246.

[33] A. Duval, The Kovacic Algorithm with applications to special functions. Dif-
ferential Equations and Computer Algebra, M. Singer, Ed., Academic Press,
London, 1991, 113-130.

[34] P. Du Val, Elliptic functions and elliptic curves, Cambridge University Press,
1973.

[35] B. Dwork, Di↵erential operators with nilpotent p-curvature, Am. J. Math. 112
(1990), 749-786.

[36] B. Dwork, private communication.

[37] S. Ferrer, F. Mondéjar, On the non-integrability of the Stark–Zeemann Hamil-

tonian system, Comm. in Math. Phys. 208(1999) 55–63.

[38] S. Ferrer, F. Mondéjar, On the non-integrability of the 3-D hydrogen atom

under motional Stark e↵ect or circularly polarized microwave combined with
magnetic fields, Phys. Lett. A 264 (1999) 74–83.

[39] S. Ferrer, F. Mondéjar, On the non-integrability of the Generalized van der

Waals Hamiltonian, J. of Math. Phys.41 (2000) 5445–5452.

[40] V.I. Gromak, I. Laine, S. Shimomura, Painlevé Di↵erential Equations in the

Complex Plane, Walter de Gruyter, Berlin 2002.

[41] G. H. Halphen, Traité des fonctions elliptiques Vol. I, II. Gauthier-Villars, Paris,
1888.

[42] J. Hietarinta, A search for integrable two-dimensional Hamiltonian systems

with polynomial potential, Physics Letters A 96 (1983) 273–278.

[43] J. Hietarinta, Direct methods for the search of the second invariant Phys. Rep.
147 (1987) 87–154.

[44] G. W. Hill, Researches in the Lunar theory, American Journal of Mathematics

1 (1878), p. 5-6, 129-147, 245-260.

[45] E. Horozov,T. Stoyanova, Non-Integrability of Some Painlevé VI-Equations and

Dilogarithms Regular and chaotic Dynamics, 12 (2007) 622-629.

[46] J.E. Humphreys, Linear Algebraic Groups, Springer-Verlag, New York, 1981.

[47] E. L. Ince, Ordinary di↵erential equations, Dover, Nova York, 1956.

REFERENCES 89

[48] D.I. Goryachev, New integrable cases of integrability of the Euler dynamical
equations, Warsaw Univ. Izv. 1910 (in russian).

[49] H. Ito, Non-integrability of the Hénon-Heiles system and a theorem of Ziglin,

Koday Math. J. 8 (1985) 129-138.

[50] K. Iwasaki, Some dynamical aspaects of Painlevé VI Algebraic Analysis of Dif-

ferential Equations in honor of Prof. Takahiro Kawai on the occasion of his
sixtieth birthday, T. Aoki, Y. Takei, N. Tose, H. Majima eds.(2007) 143-156.

[51] E. Juillard Tosel, Un rèsultat de non-integrabilité pour le potentiel en 1/r2 , C.

R. Acad. Sci. Paris, t. 327, Série I (1998) 387–329.

[52] E. Juillard,Non-integrabilité algebrique et méromorphe de problèmes de N corps,

Thése Univ. Paris VI, Janvier 1999.

[53] E. Juillard Tosel, Un nouveau critère de non-integrabilité méromorphe d’un

Hamiltonian, C. R. Acad. Sci. Paris, t. 330, Série I (2000) 1097–1102.

[54] I. Kaplansky, An Introduction to Di↵erential Algebra. Hermann, Paris 1976.

[55] N.M. Katz, A conjecture in the arithmetic theory of di↵erential equations. Bull.
Soc. Math. France 110 (1982), 203-239.

[56] T. Kimura, On Riemann’s Equations wich are Solvable by Quadratures, Funk-

cialsj Ekvacioj 12 (1969) 269–281.

[57] E. Kolchin, Di↵erential algebra and algebraic groups. Academic Press, New
York, 1973.

[58] J.J. Kovacic, An Algorithm for Solving Second Order Linear Homogeneous
Di↵erential Equations. J. Symbolic Computation 2 (1986), 3-43.

[59] S. Kowalevski, Sur le probleme de la rotation d’un corps solide autour d’un
point fixe, Acta Math. 12 (1889), 177-232.

[60] V.V. Kozlov, Non-existence of univalued integrals and branching of solutions

in rigid body dynamics Pikl. Mat. Mekh. 42, no. 3 (1978), 400-406.

[61] M. Kummer, A.W. Saenz, Nonintegrability of the Zeeman Hamiltonian, Com-

mun. in Math. Phys. 162 (1994) 447–465.

[62] M. Kummer, A.W. Saenz, Nonintegrability of the Stormer problem, Physica D

86 (1995) 363–372.

[63] J.L. Lagrange, Ouvres, Vol. 6, Paris 1873.

[64] L. Landau, E. Lifchitz, Théorie des champs. Mir, Moscou 1970.

[65] F. Loray, S. Cantat, Holomorphic Dynamics, Painlevé VI equation and charac-

ter varieties, preprint, Université de Rennes, 2007.
90 REFERENCES

[66] A.J. Maciejewski, M. Szydlowski, Towards a description of complexity of the

simplest cosmological systems, J. Phys.A: Math. Gen. 33 (2000) 9241–9254.

[67] A.J. Maciejewski, M. Szydlowski, Integrability and Non-integrability of Planar

Hamiltonian Systems of Cosmological Origin, J. of Nonlinear Math. Phys. 8
(2001) 200–206.

[68] A.J. Maciejewski, M. Przybylska. Non-integrability of restricted two-body prob-

lems in constant curvature spaces. Regul. Chaotic Dyn., 8 (4) (2003) 413–430.

[69] A.J. Maciejewski, M. Przybylska, Non-integrability of the problem of a rigid

satellite in gravitational and magnetic fields, Celestial Mech. Dynam. Astronom.
87 (2003) 317–351.

[70] A.J. Maciejewski, M. Przybylska, Non-integrability of the Suslov problem, J.

Math. Phys. 45 (2004) 1065–1078.

[71] A.J. Maciejewski, M. Przybylska, All Meromorphically Integrable 2D Hamil-

tonian Systems with Homogeneous Potential of Degree 3, Phys. Lett. A 327
(2004) 461–473.

[72] A.J. Maciejewski, M. Przybylska, Non-integrability of the generalized two fixed

centres problem, Celestial Mech. Dynam. Astronom., 89 (2004), 145–164.

[73] A.J. Maciejewski, M. Przybylska, Di↵erential Galois Approach to the Non-

integrability of the Heavy Top Problem, Annales de la Faculté des Sciences de
Toulouse Sér. 6, 14 (2005) 123–160.

[74] A.J. Maciejewski, M. Przybylska, Darboux points and integrability of Hamilto-

nian systems with homogeneous polynomial potential, J. Math. Phys. 46 (2005)
062901.1–062901.33.

[75] A.J. Maciejewski, M. Przybylska, T. Stachowiak, Non integrability of Gross-

Neveu systems, Physica D 201 (2005) 249–267.

[76] A.J. Maciejewski, M. Przybylska, T. Stachowiak, M. Szydlowski, Global Dy-

namics of cosmological scalar fields-Part I, preprint 2007.

[77] A.J. Maciejewski, M. Przybylska, J.-A. Weil, Non-integrability of the general-

ized spring-pendulum problem, J. Phys. A: Math. Gen. 37 (2004) 2579–2597.

[78] A.J. Maciejewski, M. Przybylska, H. Yoshida, Necessary conditions for partial

and superintegrability of Hamiltonian systems with homogeneous potential,
preprint 2007.

[79] A. J. Maciejewski, J. M. Strelcyn, M. Szydlowski, Non-integrability of Bianchi

VIII Hamiltonian System, J. Math. Phys.42 (2001), 1728–1743.
REFERENCES 91

[80] A.J. Maciejewski, W. Respondek, The nilpotent tangent 3-dimensional sub-

Riemannian problem is nonintegrable, Decision and Control 2004. CDC. 43rd
IEEE Conference on, 438- 443 Vol.1.

[81] R. Maier, Algebraic Solutions of the Lamé Equation, Revisited, J. of Di↵.

Equations198 (2004), 16-34.

[82] B. Malgrange, Le groupoı̈de de Galois d’un feuilletage, L’enseignement

mathématique 38, vol 2 (2001) 465–501.

[83] B. Malgrange, On the non linear Galois theory, Chinese Ann. Math. Ser. B 23,
2 (2002) 219–226.

[84] J. Malmquist, Sur les équations di↵érentielles du second ordre dont l’intégrale
générale a ses points critiques fixes, Arkiv. Mat., Astron. Fys. 18(8)(1922) 1-89.

[85] J. Martinet, J.P. Ramis, Théorie de Galois di↵érentielle et resomma-

tion.Computer Algebra and Di↵erential Equations, E. Tournier ed. Academic
Press, London, 1989, 117–214.

[86] R. Martı́nez, C. Simó, Non-Integrability of Hamiltonian Systems Through High

Order Variational Equations: Summary of Results and Examples, Regular and
Chaotic Dynamics (2009) 14 (3) 323-348.

[87] R. Martı́nez, C. Simó, Non-Integrability of the degenerate cases of the Swinging

Atwood’s Machine using higher order variational equations, preprint (2009).

[88] R. Martı́nez, C. Simó, Efficient numerical implementation of integrability cri-

teria based on high order variational equations, preprint (2009).

[89] Richard Moeckel, On central configuration, Math. Z. 205 (1990), 499-517.

[90] F. Mondéjar, Non-integrability of parametric Hamiltonian systems by di↵eren-

tial Galois theory, Monografı́as de la Acad. de Ciencias. Zaragoza 14 (1999)
60-66.

[91] F. Mondéjar, On the Non-integrability of Hamiltonian Systems with sum of

Homogeneous Potentials, unpublished.

[92] J.J. Morales-Ruiz, C. Simó, Picard-Vessiot Theory and Ziglin’s Theorem. J. of

Di↵. Equations107(1994), 140-162.

[93] J.J. Morales-Ruiz, C. Simó, Non-integrability criteria for Hamiltonians in the

case of Lamé Normal Variational Equations. J. of Di↵. Equations129 (1996)
111–135.

[94] J. J. Morales-Ruiz, Di↵erential Galois Theory and Non-Integrability of Hamil-

tonian Systems. Birkhäuser, Basel 1999.
92 REFERENCES

[95] J.J. Morales-Ruiz, J.M. Peris, On a Galoisian Approach to the Splitting of

Separatrices, Ann. Faculté Sciencies de Toulouse VIII (1999) 125–141.

[96] J.J. Morales-Ruiz, J.P. Ramis, Galoisian obstructions to integrability of Hamil-

tonian systems, Methods and Applications of Analysis 8 (2001) 33–96.

[97] J.J. Morales-Ruiz, J.P. Ramis, A Note on the Non-Integrability of some Hamil-
tonian Systems with a Homogeneous Potential, Methods and Applications of
Analysis 8 (2001) 113–120.

[98] J.J. Morales-Ruiz, J.P. Ramis, Galoisian Obstructions to integrability of Hamil-

tonian Systems II, Methods and Applications of Analysis 8 (2001) 97–102.

[99] J.J. Morales-Ruiz, C. Simó, S. Simon, Algebraic proof of the non-integrability

pf Hill’s problem, Ergod. Th. & Dynam. Sys. 25 (2005), 1237–1256.

[100] J. J. Morales-Ruiz, A Remark about the Painlevé Transcendents, in “Théories

asymptotiques et équations de Painlevé”, S.M.F., Séminaires et Congrès 14
(2005), 229–235.

[101] J.J. Morales-Ruiz, J.P. Ramis, C. Simó, Integrability of Hamiltonian Systems

and Di↵erential Galois Groups of Higher Variational Equations, to appear in
Ann. Sc. École Norm. Sup. 40 (2007) 845-884.

[102] J. J. Morales-Ruiz, S. Simon, On the meromorphic non-integrability of some

N -body problems, Discrete and Continuous Dynamical Systems 24 (2009)
1225-1273.

[103] J. Moser, Three integrable Hamiltonian systems , Advances in Math. 16 (1975)

197-220.

[104] K. Nakagawa, Direct construction of polynomial first integrals for Hamilto-

nian systems with a two-dimensional homogeneous polynomial potential, Dep.
of Astronomical Science, The Graduate University for Advanced Study and the
National Astronomical Observatory of Japan, Ph. D. Thesis 2002.

[105] K. Nakagawa, H. Yoshida, A necessary condition for the integrability of homo-

geneous Hamiltonian systems with two degrees of freedom, J. Phys. A: Math.
Gen. 34 (2001) 2137–2148.

[106] K. Okamoto, Polynomial Hamiltonians associated to Painlevé Equations I,

IIProc. Japan Acad. Ser. A 56 (1980) 264-268, 367-371.

[107] K. Okamoto, Studies of the Painlevé Equations, III. Second and Fourth
Painlevé Equations, PII and PIV , Math. Ann. 275 (1986) 221-255.

[108] H. Poincaré, Les Méthodes Nouvelles de la Mécanique Céleste, Vol. I.

Gauthiers-Villars, Paris, 1892.
REFERENCES 93

[109] L. M. Perko, E. L. Walter, Regular polygon solutions of the N -body problem,

Proceedings of the A.M.S. 94 (1985), 301-309.
[110] E.Picard, Sur les groupes de transformation des équations di↵érentielles
linéaires, C.R. Acad. Sci. Paris 96(1883), 1131–1134.
[111] E.Picard, Sur équations di↵érentielles et les groupes algébriques des transfor-
mation, Ann. Fac. Sci. Univ. de Toulouse (1) 1(1887), A1–A15.
[112] E.Picard, Traité d’Analyse, Tome III, Gauthiers-Villars, Paris, 1928.
[113] E.G.C. Poole,Introduction to the theory of Linear Di↵erential Equations. Ox-
ford Univ. Press, London, 1936.
[114] O. Pujol, J.-P. Perez, S. Simon, J.-P. Ramis, J.-A. Weil, C. Simó, Swinging
Atwood’s machine: experimental and theoretical study, preprint (2009).
[115] M. Przybilska, Finitesness of integrable n-dimensional homogeneous polyno-
mial potentials, Physics Letters A 369 (2007)180-187.
[116] A.W. Sáenz, Nonintegrability of the Dragt-Finn model of magnetic confina-
ment: a Galoisian-group approach Physica D144, (2000) 37–43.
[117] S. T. Sadetov, On algebraic integrals of Hill Problem and Restricted Circle
Plane Three-Body Problem on a level of energy, Regul. Chaotic Dyn.10 323-332
.
[118] A. V. Shchepetilov, Nonintegrability of the two-body problem in constant
curvature spaces,J. Phys. A 39 (2006), 5787–5806.
[119] C.L. Siegel, Topics in complex function theory. Wiley, New York, 1969.
[120] M. Singer, F. Ulmer, Necessary conditions for Liouvillian solutions of (third
order) linear di↵erential equations, Appl. Algebra Engrg. Comm. Comput. 6,
(1995).
[121] T. Stuchi, A. López-Castillo,M.A. Almeida, Nonintegrability of the three-body
problems for the classical helium atom, J. Math. Phys. 47 (2006), 093506–
093513.
[122] T. J. Stuchi, A. López-Castillo,M. A. Almeida, Nonintegrability of the three-
body problems for the classical helium atom, J. Math. Phys.47 (2006) 093506.
[123] C. Simó, T. J. Stuchi, Central stable/unstable manifolds and the destruction
of KAM tori in the Planar Hill Problem, Physica D140 (2000), 1-32.
[124] M.F. Singer, F. Ulmer, Galois Groups of second and third order linear di↵er-
ential equations, J. Symbolic Computation 16 (1993), 1-36.
[125] M.F. Singer, F. Ulmer, Liouvillian and algebraic solutions of second and third
order linear di↵erential equations, J. Symbolic Computation 16 (1993), 37-73.
94 REFERENCES

[126] T. Stachowiak, M. Szydlowski, A.J. Maciejewski, Nonintegrability of density

perturbations in the Friedmann-Robertson-Walker universe, J. Math. Phys. 47
(2006), 032502-032513.

[127] J. Stoer, J., R. Bulirsch, Introduction to numerical analysis, Springer-Verlag,

New York, 2002.

[128] T. Stoyanova, O. Christov, Non-Integrability of the Second Painlevé Equation

as a Hamiltonian system, Compte Rendu de l’académie bulgare des sciences, 60
(2007) 1.

[129] A.H.Taub, Empty Space-Times admitting a Three Parameter Group of mo-

tions. Annals of Mathematics 53 (1951), 472–490.

[130] A. Tsygvintsev, La non-integrabilité méromorphe du probléme plan des trois

corps, C.R. Acad. Sci. Paris SérieI 331 (2000) 241-244.

[131] A. Tsygvintsev, Sur l’absence d’une intégral première méromorphe dans le

probléme plan des trois corps, C.R. Acad. Sci. Paris Série I 333 (2001) 125–
128.

[132] A. Tsygvintsev, Sur l’absence d’une intégrale première méromorphe

supplémentaire dans le problème plan des trois corps, C.R. Acad. Sci. Paris, t.
333, Série I (2001) 241–244.

[133] A. Tsygvintsev, The meromorphic non-integrability of the three-body prob-

lem, J. Reine Angew. Math. 537 (2001), 127–149.

[134] A. Tsygvintsev, V. Non-existence of new meromorphic first integrals in the

planar three-body problem, Celestial Mech. Dynam. Astronom. 86 (2003),
237–247.

[135] A. Tsygvintsev, On some exceptional cases in the integrability of the three-

body problem, Celestial Mechanics and Dynamical Astronomy 99 (2007) 23-29.

[136] F. Ulmer, J.A. Weil, Note on Kovacic’s Algorithm, J. Symbolic Computation

22 (1996), 179-200.

[137] H. Umemura, Di↵erential Galois Theory of infinite dimension, Nagoya Math-

ematical Journal 144 (1996) 59–135.

[138] M. van der Put and M. Singer, Galois Theory of Linear Di↵erential Equations,
Springer, Berlin 2003.

[139] M.E. Vessiot, Sur l’intégration des équations di↵érentielles linéaires, Ann. Sci.
de l’École Norm. Sup. (3) 9 (1892), 197–280.

[140] E.T. Whittaker, E.T. Watson, A Course of Modern Analysis. Cambridge Univ.
Press, Cambridge, UK, 1969.
 REFERENCES 95

[141] A. Wintner, The analytical foundations of celestial mechanics, Princeton Uni-

versity Press, Princeton, N. J., 1947 (1941).

[142] K. Yagasaki, Galoisian obstructions to integrability and Melnikov criteria for

chaos in two-degree-of-freedom Hamiltonian systems with saddle centres Non-
linearity 16 (2003), 2003-2012.

[143] K. Yagasaki, Nonintegrability of an infinite-degree-of-freedom model for un-

forced and undamped, straight beams, J. of Applied Mechanics 70 (2003), 732-
738.

[144] K. Yagasaki,T. Wagenknecht, Detection of symmetric homoclinic orbits to

saddle-centres in reversible systems, Physica D 214 (2006), 169-181.

[145] H. Yoshida, A criterion for the non-existence of an additional integral in Hamil-

tonian systems with a homogeneous potential, Physica D29 (1987) 128–142.

[146] H. Yoshida, Non integrability of the truncated Toda lattice at any order,
Commun. in Math. Phys. 116 (1988), 529–538.

[147] S.L. Ziglin, Branching of solutions and non-existence of first integrals in Hamil-
tonian mechanics I, Funct. Anal. Appl. 16 (1982), 181–189.

[148] S.L. Ziglin, Branching of solutions and non-existence of first integrals in Hamil-
tonian mechanics II. Funct. Anal. Appl. 17 (1983), 6–17.

[149] S.L. Ziglin, On the absence of a real-analytic first integral in some problems
of dynamics. Funct. Anal. Appl. 31 (1997), 3–9.

[150] S.L. Ziglin, On involutive integrals of groups of linear symplectic transforma-

tions and natural mechanical systems with homogeneous potential. Funktsional.
Anal. i Prilozhen. 34 (2000), 26–36.

View publication stats