Developments in Mathematics

Xiang Zhang

Integrability
of Dynamical
Systems: Algebra
and Analysis
Developments in Mathematics

Volume 47

Series editors
Krishnaswami Alladi, Gainesville, USA
Hershel M. Farkas, Jerusalem, Israel
More information about this series at http://www.springer.com/series/5834
Xiang Zhang

Integrability of Dynamical
Systems: Algebra
and Analysis

123
Xiang Zhang
School of Mathematical Sciences
Shanghai Jiao Tong University
Shanghai
China

ISSN 1389-2177 ISSN 2197-795X (electronic)

Developments in Mathematics
ISBN 978-981-10-4225-6 ISBN 978-981-10-4226-3 (eBook)
DOI 10.1007/978-981-10-4226-3
Library of Congress Control Number: 2017934316

Mathematics Subject Classification: 34A34, 34Cxx, 37C10, 37C27, 37C15

© Springer Nature Singapore Pte Ltd. 2017

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.

Printed on acid-free paper

This Springer imprint is published by Springer Nature

The registered company is Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore
Preface

The theory of integrability plays an important role in the study of the dynamics of
differential systems. This theory is related to several branches of mathematics, such
as algebraic geometry, algebraic topology, the theory of Lie groups and Lie alge-
bras, real and complex analysis, differential geometry, and Riemannian geometry.
Here we are mainly concerned with the algebraic and analytic aspects of the
integrability of ordinary differential equations and of dynamical systems.
This book summarizes the last two decades of research on the integrability of
dynamical systems and related topics obtained by the author and his coauthors,
together with relevant results of other specialists. The main emphasis is on ordinary
differential equations and singular holomorphic foliations given by the orbits of a
projective one-form in the projective real or complex plane.
The theory of integrability was born with Newton’s mechanics and developed
rapidly. However, until Liouville’s work on Riccati equations in 1841, this research
generally reduced to the search for different methods of integrating distinct dif-
ferential equations. The fact that the simple Riccati equations cannot be integrated
by quadratures forced researchers to explore new theories and methods to determine
if a dynamical system has first integrals, or is integrable, to study the regularity of
first integrals (if they exist), and to determine the topology and geometry of their
level surfaces.
For higher-dimensional differential systems it is in general difficult to study their
dynamics. If a system has a number, say k, of functionally independent first inte-
grals, the study of its dynamics can be reduced to ðn  kÞ–dimensional systems.
Then it becomes easier to study the dynamics of the reduced lower-dimensional
system. Since many mechanical systems have first integrals or invariants, the
problems of how to prove their existence and determine their expressions are
important and have attracted the interest of many physicists and mathematicians.
This book consists of seven chapters. Chapter 1 introduces some preliminary
notions and methods on the theory of integrability of ordinary differential equations,
which include the existence and number of functionally independent first integrals
surrounding a regular point, and their applications in solving linear and quasilinear
partial differential equations. We also characterize the global existence and

v
vi Preface

regularity of the first integrals of planar differential systems in their canonical

regions. Section 1.4 contains some fundamental results on the Lax pairs of ordinary
differential equations and their application to the search for first integrals. The main
tools used in this chapter are the flow box theorem, the C k (orbital) equivalence of
flows, the limit sets of two-dimensional flows, the local topological structure of
analytic vector fields around singularities, the finiteness of limit cycles and the
Bendixson compactification.
Chapter 2 concentrates on the Jacobian and the inverse Jacobian multipliers,
which are fundamental tools in the study of integrability. In Sect. 2.1 we study the
existence of Jacobian multipliers and their relation to volume preserving flows.
Section 2.2 concerns inverse Jacobian multipliers and their application to the study
of the dynamics of differential systems, for instance, the existence and multiplicity
of limit cycles and so on. In this subsection we also summarize the results on the
existence and regularity of inverse integrating factors in a neighborhood of a sin-
gularity or of a limit cycle. In Sect. 2.3 we mainly consider the existence of C1 or
C x local inverse Jacobian multipliers at a singularity of an analytic differential
system having a pair of pure imaginary eigenvalues and a two-dimensional center
manifold. We also study the Hopf bifurcation using the vanishing multiplicity of
inverse Jacobian multipliers. In the last section we introduce the notion of Lie
symmetry for a vector field and its application to the construction of inverse
Jacobian multipliers. The key tools in this chapter are the theory of normal forms,
the symmetry of Lie groups, and the theory of differential forms.
Chapter 3 is on the Darboux theory of integrability and its applications. The
Darboux theory of integrability is an important part of the theory of integrability
and plays an important role in the study of the dynamics of polynomial differential
systems. Section 3.1 focuses on the classical Darboux theory of integrability and on
its proof. Section 3.2 presents new developments in the Darboux theory of inte-
grability. In Sect. 3.3 we introduce the Liouvillian and the elementary integrability
of polynomial differential systems, and show their relations with the Darboux
theory of integrability. Section 3.4 provides a relation between the Liouvillian first
integrals and the Darboux polynomials, and its application to Liénard differential
systems. The tools used in this chapter are the expression of functionally inde-
pendent first integrals, the algebraic multiplicity of invariant algebraic hypersur-
faces and differential field extension.
A key role in the Darboux theory of integrability is played by the Darboux
polynomials, which were first used by Darboux in 1878 to construct first integrals
or integrating factors. Chapter 4 is devoted to the study of the degree and the
existence of Darboux polynomials of polynomial differential systems. Section 4.1
studies the degree of the Darboux polynomials of planar singular holomorphic
foliations. This is one of the classical Poincaré problems. Section 4.2 illustrates
methods for classifying the Darboux polynomials of some concrete differential
systems from physics and mechanics. This study involves algebraic geometry,
weight homogeneous polynomials, and the method of characteristic curves for
solving linear partial differential equations.
Preface vii

Chapter 5 studies the equivalence between algebraic integrability and rational

integrability. Using this result, we further characterize the polynomial, rational and
analytic integrability of some concrete physical models. Section 5.1 focuses on the
algebraic integrability of some specific models via the Darboux polynomials.
Section 5.2 presents some results on the polynomial and rational integrability of
natural Hamiltonian systems. Section 5.3 summarizes the results on the mero-
morphic integrability of Hamiltonian systems along a given orbit via the differential
Galois group. Section 5.4 introduces algorithms to compute the Darboux polyno-
mials and the rational first integrals. In the study of these results we need the
Frobenius integrability theorem, the differential Galois group, and so on.
Chapter 6 presents some applications of the Darboux theory of integrability to
the center-focus problem, to a weakened version of Hilbert’s 16th problem on
algebraic limit cycles, and to many other specific models.
Chapter 7 is devoted to the local theory of integrability. The materials are on the
existence of local analytic, formal, rational, meromorphic and local Darboux first
integrals of analytic (or formal) differential systems in a neighborhood of a
singularity. The emphasis is on the equivalent characterization of the analytical
dynamical systems which are locally integrable at a singularity, showing that
analytic integrability implies the existence of analytic normalization. Furthermore,
we consider the existence of embedding flows for integrable diffeomorphisms on a
manifold. Finally, we characterize the varieties and the asymptotic expressions
of the first integrals of the integrable or partially integrable differential systems.
A preliminary version of this book appeared as a series of lectures for graduate
students at Shanghai Jiao Tong University, China. My colleagues, Profs. Jiang Yu
and Yilei Tang, attended the lectures and pointed out lots of improvements in the
mathematics and its presentation. I really appreciate our interesting discussions
and their suggestions. I would also like to thank those students enrolled in my
lectures, who corrected many misprints in the mathematics. I must appreciate
Prof. Jaume Llibre for his careful reading of the English presentation and his
corrections. I am sincerely indebted to several anonymous referees who offered
valuable comments and suggestions for improving the structure and presentation of
this book. I thank Ramon Peng, the editor of Springer, who has worked a lot during
the review process. Finally, I should thank my wife, Niannian Su, for giving me the
gift of time during the period in which this book was written.

Shanghai, China Xiang Zhang

November 2016
Contents

1 The Fundamentals of the Theory of Integrability

of Differential Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Existence and Properties of the First Integrals . . . . . . . . . . . . . . . . 1
1.1.1 Characterization and Properties of the First Integrals . . . . . 3
1.1.2 Existence of First Integrals Near a Regular Point . . . . . . . . 7
1.2 First Integrals of Differential Systems in Canonical Regions . . . . . 10
1.3 Applications of Integrability Theory to Partial Differential
Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 20
1.3.1 First-Order Linear Homogeneous Partial Differential
Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 21
1.3.2 First-Order Quasilinear Partial Differential Equations . .... 23
1.4 Lax Pairs and Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 26
2 Jacobian and Inverse Jacobian Multipliers . . . . . . . . . . . . . . . . . . . . . 35
2.1 Jacobian Multipliers, First Integrals and Integrability . . . . . . . . . . . 35
2.2 Inverse Jacobian Multipliers and Their Vanishing Sets . . . . . . . . . . 45
2.3 Inverse Jacobian Multipliers and the Center-Focus Problem . . . . . . 59
2.3.1 The Center-Focus Problem via Inverse Integrating
Factors or Inverse Jacobian Multipliers . . . . . . . . . . . . .... 61
2.3.2 Hopf Bifurcation via Inverse Jacobian Multipliers . . . .... 73
2.4 Inverse Jacobian Multipliers via Lie Groups . . . . . . . . . . . . . . .... 79
3 Darboux and Liouvillian Integrability . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.1 The Classical Darboux Theory of Integrability . . . . . . . . . . . . . . . . 89
3.1.1 The Existence of Darboux First Integrals . . . . . . . . . . . . . . 90
3.1.2 The Darboux–Jouanolou Integrability Theorem . . . . . . . . . . 94
3.2 Generalization of the Classical Darboux Theory
of Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ......... 98
3.2.1 Taking into Account Independent Singularities . ......... 98
3.2.2 Taking into Account Algebraic Multiplicities . . ......... 100

ix
x Contents

3.2.3 Taking into Account the Multiplicity of the Hyperplane

at Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.2.4 On Nonautonomous Differential Systems
via the Wronskian Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . 115
3.2.5 Differential Systems in the Sparse Case . . . . . . . . . . . . . . . 122
3.2.6 Other Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.3 Liouville and Elementary First Integrals . . . . . . . . . . . . . . . . . . . . . 126
3.3.1 Background on Differential Field Extensions . . . . . . . . . . . 126
3.3.2 The Prelle and Singer Integrability Theorems . . . . . . . . . . . 131
3.4 Liouvillian Integrability Versus Darboux Polynomials . . . . . . . . . . 145
4 Existence and Degree of Darboux Polynomials . . . . . . . . . . . . . . .... 149
4.1 The Degree of Invariant Algebraic Curves . . . . . . . . . . . . . . . .... 149
4.1.1 Examples of Invariant Algebraic Curves
of Arbitrary Degree . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 150
4.1.2 Invariant Algebraic Curves in the Projective Plane . . . .... 152
4.1.3 The Degree of Invariant Algebraic Curves
in the Nodal and Nondicritical Cases . . . . . . . . . . . . . .... 156
4.2 Existence of Darboux Polynomials . . . . . . . . . . . . . . . . . . . . . .... 163
4.2.1 Polynomial Vector Fields Without Darboux
Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 164
4.2.2 Liénard Differential Systems: Invariant Algebraic
Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 167
4.2.3 Lorenz Systems: Invariant Algebraic Surfaces. . . . . . . .... 177
4.3 Other Results on Darboux Polynomials . . . . . . . . . . . . . . . . . .... 192
5 Algebraic, Analytic and Meromorphic Integrability . . . . . . . . . . . . . . 197
5.1 Algebraic First Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
5.1.1 Algebraic and Rational Integrability: Their
Equivalence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
5.1.2 Kirchoff Equations: Polynomial and Rational First
Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
5.1.3 Euler Equations on the Lie Algebra soð4Þ: Polynomial
First Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
5.1.4 The 5-Dimensional Lorenz Systems: Darboux and
Analytic Integrability. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
5.2 Natural Hamiltonian Systems: Polynomial and Rational
Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
5.2.1 Hamiltonian Systems in the Canonical Form . . . . . . . . . . . 221
5.2.2 Hamiltonian Systems in a Generalized Form. . . . . . . . . . . . 225
5.3 Hamiltonian Systems: Integrability via the Differential Galois
Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
5.3.1 Hamiltonian Systems Having Homogeneous Potentials . . . . 228
5.3.2 Hamiltonian Systems: Integrability via Other
Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
Contents xi

5.3.3 Extension of the Morales–Ramis Theory to General

Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
5.4 Calculations of Rational First Integrals and Darboux
Polynomials of Planar Polynomial Vector Fields . . . . . . . . . . . . . . 240
6 Applications of the Darboux Theory of Integrability . . . . . . . . . .... 253
6.1 The Center Problem via the Darboux Theory of Integrability . .... 253
6.2 Algebraic Limit Cycles: Existence and Uniqueness . . . . . . . . .... 258
6.2.1 The Existence of Rational First Integrals
via Intersection Numbers . . . . . . . . . . . . . . . . . . . . . . . .... 259
6.2.2 Quadratic Differential Systems: Algebraic
Limit Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .... 265
6.3 Hilbert’s 16th Problem: A Weak Version on Algebraic
Limit Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
6.4 Applications to Concrete Models . . . . . . . . . . . . . . . . . . . . . . . . . . 276
6.4.1 Concrete Two-Dimensional Models in Applications . . . . . . 276
6.4.2 Concrete Three-Dimensional Models in Applications . . . . . 277
6.4.3 Concrete Higher-Dimensional Models in Applications . . . . 282
6.4.4 Abel Equations and Foliations . . . . . . . . . . . . . . . . . . . . . . 283
7 Local Integrability of Differential Systems. . . . . . . . . . . . . . . . . . . . . . 287
7.1 The Foundations of Poincaré Normal Form Theory . . . . . . . . . . . . 289
7.2 Local Analytic and Formal First Integrals. . . . . . . . . . . . . . . . . . . . 295
7.2.1 Local Integrability via Partial Nonresonances . . . . . . . . . . . 296
7.2.2 Local Integrability via Resonances . . . . . . . . . . . . . . . . . . . 301
7.3 Local (Formal) Meromorphic First Integrals . . . . . . . . . . . . . . . . . . 307
7.3.1 The Equivalence Between Algebraic and Functional
Independence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
7.3.2 The Lowest Order Parts of Functionally Independent
First Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
7.3.3 Proof of the Results on Functionally Independent
First Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
7.4 The Local Theory of Darboux Integrability . . . . . . . . . . . . . . . . . . 314
7.4.1 Local Darboux First Integrals . . . . . . . . . . . . . . . . . . . . . . . 315
7.4.2 Applications of the Local Theory of Darboux
Integrability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318
7.5 Analytic Normalization of Analytic Integrable Systems . . . . . . . . . 320
7.5.1 Equivalent Characterization of Analytic Integrable
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
7.5.2 Integrable Discrete Dynamical Systems
and Embedding Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
7.6 Varieties and Normalization of Partially Integrable Systems . . . . . . 334
7.6.1 Varieties of Partially Integrable Systems . . . . . . . . . . . . . . . 334
7.6.2 Analytic Normalization of Partially Integrable Systems. . . . 341
xii Contents

7.6.3 Generic Divergence of Normalizations of Partially

Integrable Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
7.7 Other Results on Local Integrability . . . . . . . . . . . . . . . . . . . . . . . . 347
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
Symbols

N Set of positive integers

Z Set of integers (or group of integers)
Zþ Set of nonnegative integers
Q Set of rational numbers
R Set of real numbers (or field of real numbers)
Rþ Set of positive real numbers
C Set of complex numbers (or field of complex numbers)
Cn Complex n-dimensional linear space
RPn Real projective space of dimension n
CP2 Complex projective plane
CPn Complex projective space of dimension n
; Empty set
X A region in Rn or Cn
½a; b An interval of R
S1 Unit circle
ðRn ; 0Þ A neighborhood of the origin in Rn
ðCn ; 0Þ A neighborhood of the origin in Cn
A=B Quotient space of A modulo B
V 1 ð0Þ Set of zeros of function V
Vðf Þ Set of zeros of function f
n Set-theoretic difference
 Set-theoretic inclusion
\ Set-theoretic intersection
[ Set-theoretic union
2 Element of a set
9 Existence
n! Product of 1 to n, i.e. nðn  1Þ  2  1
cl Closure of a set
det Determinant of a matrix
dim Dimension of a linear space

xiii
xiv Symbols

div Divergence of a function

gcd Greatest common divisor
Im Imaginary part of a complex number
rank Rank of a matrix
Re Real part of a complex number
tr Trace of a matrix
s Transpose of a matrix
CðXÞ, C 0 ðXÞ Sets of continuous functions on X
C k ðXÞ Set of functions on X whose derivatives until kth order all
continuous
C 1 ðXÞ Set of infinitely differentiable functions on X
C x ðXÞ Set of analytic functions on X
C½a; b Set of continuous functions on ½a; b
C0 Set of continuous functions
Ck Set of functions whose derivatives until kth order all continuous
C1 Set of infinitely differentiable functions
Cx Set of analytic functions
R½x Ring of polynomials with coefficients in R
C½x Ring of polynomials with coefficients in C
CðxÞ Field of rational functions with coefficients in C
Cm ½x Set of polynomials of degree at most m
C½x½e Ring of polynomials in e with coefficients in C½x
hek i Ideal of C½x½e generated by ek
CðxÞfeg Ring of series in e with coefficients in CðxÞ
C½½x Ring of formal series in x
C½½xc Ring of convergent power series in x
CððxÞÞ Field of formal meromorphic functions in x
CððxÞÞc Field of meromorphic functions in x
@0 Degree of an objective
vðMÞ Euler characteristic of a surface M
½ Integer part function
#J Number of elements of a set J
 Composition of two maps
G1 Inverse of the transformation G
 Cartesian product of sets
 Cross product of vectors
h; i Inner product of vectors
½;  Lie bracket of vector fields or linear operators
^ Wedge product of differential forms
 Summation of elements in a group
Multiplication of elements in a group
SO(2) Rotation group on the plane
S1  S1 Cross product of two circles
 Uniformly convergent
Symbols xv

A,B Proposition A is equivalent to proposition B

ðM; uÞ Manifold M with flow u
aðci Þ Negative limit set of orbit ci
xðci Þ Positive limit set of orbit ci
@x H, rH Gradient of a function HðxÞ
DF Jacobian of vector functions FðxÞ
@f
fy , @y f ,@y Jacobian matrix of vector functions f ðx; yÞ with respect to y
Df
Dy f ; Dy Jacobian of vector functions f ðx; yÞ with respect to y
Mðn; KÞ Set of matrices of order n with entries in the field K
Chapter 1
The Fundamentals of the Theory
of Integrability of Differential Systems

This chapter contains some definitions and properties on the first integrals of ordinary
differential equations. We will prove the local integrability of differential systems
around a regular point, and the global existence and regularity of the first integrals
of planar smooth differential systems in the canonical regions. As a preliminary
application of the integrability theory, we use it to solve linear and quasilinear partial
differential equations. Lastly, we introduce a few fundamental results on Lax pairs
for finding the first integrals of ordinary differential equations.

1.1 Existence and Properties of the First Integrals

Consider an autonomous differential system in an n-dimensional space

ẋ = f (x), x ∈ Ω ⊂ Rn , (1.1)

where Ω ⊂ Rn is an open subset, and f (x) = ( f 1 (x), . . . , f n (x)) is an

n-dimensional vector-valued function defined on Ω. The differential system (1.1) is
smooth if f ∈ C 1 (Ω); or C k smooth, k ∈ N ∪ {∞}, if f ∈ C k (Ω); or analytic if
f ∈ C ω (Ω). Hereafter, N and R are, respectively, the sets of positive integers and
real numbers; C k (Ω), k ∈ N, denotes the set of functions whose derivatives until kth
order are continuous on Ω; C ∞ (Ω) is the set of infinitely differentiable functions on
Ω; and C ω (Ω) is the set of analytic functions on Ω. A function is analytic on Ω if
it can be expanded as a convergent Taylor series in some neighborhood of any point
in Ω.
Recall from [11, 89] that for any x0 ∈ Ω and any t0 ∈ R, the smooth differential
system (1.1) has a unique solution x = ϕ(t), t ∈ J , satisfying ϕ(t0 ) = x0 and
ϕ ∈ C 1 (J ), where t0 ∈ J and J is the maximal interval where the solution is

© Springer Nature Singapore Pte Ltd. 2017 1

X. Zhang, Integrability of Dynamical Systems: Algebra and Analysis,
Developments in Mathematics 47, DOI 10.1007/978-981-10-4226-3_1
2 1 The Fundamentals of the Theory …

defined. The set Γ := {ϕ(t)| t ∈ J } ⊂ Ω is called an orbit or trajectory of system

(1.1) in Ω, and x0 is the initial point of the orbit Γ . The space Rn is called the phase
space if n > 2 or the phase plane if n = 2. The space Rn × R, taking into account
time, is called the extended phase space (resp. the extended phase plane) if n > 2
(resp. n = 2). The set {(ϕ(t), t)| t ∈ J } in the extended phase space is called an
integral curve of system (1.1).
Denote by X the vector field associated to system (1.1), i.e.

∂ ∂
X = f 1 (x) + · · · + f n (x) .
∂ x1 ∂ xn

A function H (x) is a first integral of system (1.1) in Ω if

• H (x) is defined on a full Lebesgue measure subset D of Ω;
• H (x) is not a constant on any positive Lebesgue measure subset of D (or equiva-
lently, H (x) is not locally constant);
• H (x) ∈ C(D), and it is a constant along each orbit of system (1.1) located in D.
Recall that C(D) represents the set of continuous functions defined on D.

Example The planar linear differential system

ẋ = −y, ẏ = x,

has a general solution

x = c1 cos t + c2 sin t, y = c1 sin t − c2 cos t

where c1 and c2 are integral constants, which contains all solutions of the system.
We can check easily that H = x 2 + y 2 is one of the first integrals of the system. Note
that this first integral is defined in the full plane.

Remark Even though the differential system (1.1) is defined on Ω, its first integrals
may not be defined on the full Ω. For instance, the differential system

ẋ = x, ẏ = y

is defined on R2 and it is analytic, but its first integral H (x, y) = y/x can only be
defined on the full Lebesgue measure subset R2 \ {x = 0}.

Assume that H is a first integral of system (1.1). Then for c ∈ R, roughly speaking
{x ∈ Ω| H (x) = c} is an (n − 1)-dimensional hypersurface of the n-dimensional
space, we call it a level hypersurface, which is also denoted by H = c for brevity. Thus
the n-dimensional space is foliated by the (n − 1)-dimensional level hypersurfaces.
1.1 Existence and Properties of the First Integrals 3

In particular, if n = 3, we call H = c a level surface. If n = 2, we call H = c a level

curve.
Any orbit starting from a level hypersurface will always be located in this hyper-
surface. So each level hypersurface is an invariant set of system (1.1). Recall that
a set Λ ⊂ Ω is an invariant set of system (1.1) if all orbits of system (1.1) with
their initial points in Λ will always be located in Λ. The set Λ ⊂ Ω is a positively
(resp. negatively) invariant set of system (1.1) if all positive (resp. negative) orbits
of system (1.1) with their initial points in Λ will always be located in Λ.

1.1.1 Characterization and Properties of the First Integrals

In this subsection we provide an equivalent characterization of the first integrals and

give an upper bound on the number of functionally independent first integrals. The
first result presents an equivalent characterization of continuously differentiable first
integrals.
Proposition 1.1 Assume that H (x) is a continuously differentiable function defined
on a full Lebesgue measure subset D of Ω, and it is not locally constant in D. Then
H is a first integral of system (1.1) in Ω if and only if X (H ) ≡ 0, x ∈ D, i.e.

 f (x), ∇x H (x) ≡ 0, x ∈ D, (1.2)

where ·, · denotes the inner product of two vectors, ∇x H (x) = (∂x1 H (x), . . . ,
∂xn H (x)) is the gradient of H and ∂xi H (x) = ∂ H∂ x(x)
i
, i = 1, . . . , n, is the partial
derivative of H (x) with respect to xi .
Proof The proof is easy. We leave it to the reader as an exercise. 

Remark In this book we are mainly concerned with the orbits of the autonomous
differential system (1.1) in the phase space, and so in the definition of first integrals
we do not take into account the time t. The first integral depending on the time t, say
H (t, x), can be defined similarly, and it is also called an invariant by some authors.
Hence a continuously differentiable function H (t, x), (t, x) ∈ J × D ⊂ R × Ω, is
an invariant of system (1.1) if and only if

∂t H (t, x) +  f (x), ∇x H (t, x) ≡ 0, (t, x) ∈ R × D.

Similarly, for a nonautonomous differential system

ẋ = f (t, x), (t, x) ∈ J × Ω,

roughly speaking, a continuous function H (t, x) is a first integral of the system if

H (t, x) is a constant along each integral curve {(t, ψ(t))| t ∈ J }, where ψ(t) is any
solution of this last differential system.
4 1 The Fundamentals of the Theory …

In the rest of this book, if not specified, a first integral is always independent of
time. Let H1 (x), . . . , Hk (x) be smooth first integrals of system (1.1).
• If their gradients, i.e. the k vector-valued functions

∇x H1 , . . . , ∇x Hk ,

are linearly independent in a full Lebesgue measure subset of Ω, we call them

functionally independent in Ω.
• The vector constants (c1 , . . . , ck ) ∈ Rk are regular values of H1 (x), . . . , Hk (x) if
the k vector functions ∇x H1 , . . . , ∇x Hk have full rank, i.e.rank k, at all points of
 k k
i=1 {x ∈ Ω| Hi (x) = ci }. If this is the case, the points in i=1 {x ∈ Ω| Hi (x) =
ci } are called regular points of H1 (x), . . . , Hk (x).
The next result, due to Olver [351, Theorem 2.16], reveals an essential property
of functional independence.

Theorem 1.1 Assume that M ⊂ Rn is a smooth manifold, and g1 , . . . , gk are real

smooth functions defined on M. Then g1 , . . . , gk are functionally dependent on M if
and only if for each x ∈ M, there exists a neighborhood U of x and a smooth real
function F(z 1 , . . . , z k ) of k variables such that

F(g1 (x), . . . , gk (x)) ≡ 0, x ∈ U.

Roughly speaking, an n-dimensional smooth manifold locally resembles an n-

dimensional Euclidean space. More precisely, an n-dimensional smooth manifold M
is a space such that for each p ∈ M, there exists a neighborhood U p ⊂ M of p and
a diffeomorphism χ p : U p → Rn such that if U p ∩ Uq = ∅ for some p, q ∈ M,
then χq ◦ χ p−1 : χ p (U p ∩ Uq ) → χq (U p ∩ Uq ) is a diffeomorphism. Hereafter,
the symbol ◦ denotes the composition of two maps. Each pair (χ p , U p ) is a coordi-
nate chart (or simply a chart) of the manifold M. A collection of charts {(χ p , U p )}
with the U p ’s covering M is an atlas. A manifold is C k smooth, k ∈ N ∪ {∞, ω},
if the transformation χ p is a C k diffeomorphism for all p ∈ M. By convention,
a diffeomorphism means a C 1 diffeomorphism. Let M and N be two C k mani-
folds with atlases {(χ p , U p )} and {(ςq , Vq )}, respectively. A map F : M → N
is C k smooth if for each p and q, ςq ◦ F ◦ χ p−1 : χ p (U p ) → ςq (Vq ) is a C k
smooth function. A manifold with boundary is a manifold limited by edges or faces.
A differentiable manifold is orientable if it has a coordinate atlas {(χ p , U p )} such
that all χ p ’s have positive Jacobian determinants. For more details, see for instance
[1, 207, 213]. Recall that a C k diffeomorphism, k ∈ N ∪ {∞, ω}, is a C k smooth
bijection between two C r manifolds, r ≥ k, with a C k smooth inverse. A homeomor-
phism is a continuous bijection between two topological spaces with a continuous
inverse.
The next result describes the relation between the first integrals of differential
systems through invertible transformations, which will be used later on.
1.1 Existence and Properties of the First Integrals 5

Proposition 1.2 Assume that y = G(x) is a continuously differentiable and invert-

ible map from Ω to G(Ω) := {y| y = G(x), x ∈ Ω} which transforms system (1.1)
into the system
ẏ = g(y), y ∈ G(Ω). (1.3)

Then the following statements hold.

(a) If a continuously differentiable function H (x) is a first integral of system (1.1)
in Ω, then system (1.3) has the first integral H ◦ G −1 (y) in G(Ω), where G −1
is the inverse of the transformation G.
(b) If the continuously differentiable functions H1 (x), . . . , Hk (x) are functionally
independent first integrals of system (1.1) in Ω, then system (1.3) has the func-
tionally independent first integrals H1 ◦ G −1 (y), . . . , Hk ◦ G −1 (y) in G(Ω).
Proof The result follows by direct calculation. 

Now we study the maximal number of functionally independent first integrals that
system (1.1) can have, and the relationship among the first integrals of this system.
Proposition 1.3 Assume that the n-dimensional vector-valued function f (x) in (1.1)
is continuously differentiable in Ω and does not vanish identically. The following
statements hold.
(a) System (1.1) has at most n − 1 functionally independent and continuously
differentiable first integrals in Ω.
(b) If H1 (x), . . . , Hn−1 (x) are functionally independent and continuously differ-
entiable first integrals of system (1.1) in Ω, then any continuously differentiable
first integral I (x) of system (1.1) can be locally represented as a continuously
differentiable function in terms of H1 , . . . , Hn−1 .
Proof (a) On the contrary, we assume that system (1.1) has n functionally indepen-
dent and continuously differentiable first integrals H1 (x), . . . , Hn (x) in Ω. From
Proposition 1.1 we get

 f (x), ∇x Hi (x) ≡ 0, i = 1, . . . , n, x ∈ Λ ⊂ Ω, (1.4)

where Λ is a full Lebesgue measure subset of Ω. Since H1 , . . . , Hn are functionally

independent, and f (x) is continuously differentiable in Ω, the system of linear
algebraic equations (1.4) in f (x) has only the zero solution in Ω, i.e. f (x) ≡ 0,
x ∈ Ω. This is in contradiction with the assumption.
(b) Let V ⊂ Ω be an open subset such that ∇x H1 (x), . . . , ∇x Hn−1 (x) are linearly
independent on V . Choosing a continuously differentiable function J (x) such that

y = G(x) := (H1 (x), . . . , Hn−1 (x), J (x))

is an invertible transformation on V except perhaps on a zero Lebesgue measure

subset. Then under this change of coordinates system (1.1) can be transformed into
6 1 The Fundamentals of the Theory …

ẏi = 0, i = 1, . . . , n − 1, ẏn = X (J ) ◦ G −1 (y). (1.5)

By statement (a) and the choice of J (x), we know that X (J ) ◦ G −1 (y) does not
vanish identically on G(V ).
From Proposition 1.2 we know that I ◦ G −1 (y) is a first integral of system (1.5).
So we must have
 
∂ yn I ◦ G −1 (y) X (J ) ◦ G −1 (y) ≡ 0, y ∈ G(V ).

This implies that I ◦ G −1 (y) is a function of y1 , . . . , yn−1 . Consequently the first

integral I (x) of system (1.1) is a function of

H1 (x), . . . , Hn−1 (x).

This proves the proposition. 

Remark An n-dimensional differential system (1.1) can have n functionally inde-

pendent first integrals depending on the time t, see e.g. [126, 473].

According to Proposition 1.3 we have the following definitions of the complete

integrability of system (1.1).

Definition System (1.1) is

• completely integrable (or simply integrable) in Ω if it has n − 1 functionally
independent first integrals in Ω;
• C k completely integrable (or simply C k integrable) in Ω, k ∈ Z+ ∪ {∞, ω}, if it
has n − 1 functionally independent C k first integrals in Ω, where Z+ is the set of
non-negative integers;
• locally C k completely integrable (or simply locally C k integrable) at x0 ∈ Ω, if it
has n − 1 functionally independent C k first integrals in a neighborhood of x0 .

Remark The level hypersurfaces of k < n functionally independent first integrals

intersect transversally at regular points. The normal vectors of the hypersurfaces at
each of the intersection points span a linear space of dimension k. So the set of
intersection points of the level hypersurfaces of k < n functionally independent first
integrals is an (n − k)-dimensional invariant set in a full Lebesgue measure subset
of Ω. If k = n − 1, the intersection of the level hypersurfaces of these first integrals
is in general a one-dimensional invariant set, and so it is formed by orbits of system
(1.1). Hence if an n-dimensional differential system (1.1) has n − 1 functionally
independent first integrals, then all orbits of the system will be determined by the
first integrals except perhaps in a zero Lebesgue measure subset of Ω, where the first
integrals are not well defined, or not functionally independent. This is an essential
property of complete integrability.
1.1 Existence and Properties of the First Integrals 7

Proposition 1.3 provides an upper bound on the number of functionally inde-

pendent first integrals of autonomous differential systems. The question is: can this
upper bound be achieved?
Note that an n-dimensional differential system could have n functionally inde-
pendent invariants.
In the next subsection we will answer this question for the differential system
(1.1) in a neighborhood of a regular point.

1.1.2 Existence of First Integrals Near a Regular Point

A point x0 ∈ Ω is called a regular point of system (1.1) if f (x0 ) = 0. Otherwise

x0 is called a singular point of the system. In what follows, a singular point is also
called a singularity, or a critical point, or an equilibrium.

Theorem 1.2 Assume that f (x) ∈ C k (Ω), k ∈ N ∪ {∞, ω}, and x0 is a regular
point of system (1.1). Then there exists a neighborhood V ⊂ Ω of x0 such that system
(1.1) has n − 1 C k functionally independent first integrals in V .

Theorem 1.2 will be proved using the flow box theorem. To introduce this theorem
we need the notion of the equivalence of two differential systems.
Let ϕt (x), t ∈ Jx , be the solution of the autonomous differential system (1.1) in
Ω satisfying the initial condition ϕ0 (x) = x ∈ Ω, where Jx is the maximal interval
containing 0 where the solution is defined. Then ϕt (x) is a flow, i.e., by definition
it satisfies ϕ0 (x) = x for all x ∈ Ω and ϕt+s (x) = ϕt (ϕs (x)) for all x ∈ Ω and all
t, s ∈ Jx with t + s ∈ Jx .
The differential systems (1.1) and

ẏ = g(y) y ∈ Θ ⊂ Rn an open subset, (1.6)

are C k equivalent (or topologically equivalent), k ∈ N ∪ {∞, ω}, if there exists a

C k diffeomorphism (or homeomorphism) h : Ω −→ Θ such that h conjugates the
flows of the two differential systems, i.e.

h ◦ ϕt = ψt ◦ h,

where ϕt and ψt are respectively the flows of systems (1.1) and (1.6). We call h a
conjugacy between the flows of the two differential systems or simply between the
two differential systems.
The differential systems (1.1) and (1.6) are (C k ) orbitally equivalent if there
exists a (C k diffeomorphism) homeomorphism which sends the orbits of one system
to orbits of the other one.
8 1 The Fundamentals of the Theory …

Remark
• The conjugacy h preserves not only the orientation of all orbits but also the times
that the two flows run. However, orbital equivalence is in general not necessary to
preserve flow running times.
• The smooth equivalence of systems (1.1) and (1.6) can also be equivalently defined
as
h ∗ (x) f (x) = g ◦ h(x),

where h ∗ is the tangent map of h. Since we are in Euclidean space, here h ∗ is in

fact the Jacobian matrix of h with respect to x.
• The conjugacy and equivalence mentioned above are globally defined on a mani-
fold. In applications we usually consider local conjugacy or local equivalence, that
is, we consider the conjugacy or (orbital) equivalence of two differential systems
in a neighborhood of a regular point or a singularity. If the C k equivalence or C k
orbital equivalence is only defined in a neighborhood of a point, then it will be
called local C k equivalence or local C k orbital equivalence.
Theorem 1.3 (Flow box theorem) Assume that f (x) ∈ C k (Ω), k ∈ N ∪ {∞, ω},
and x0 is a regular point of system (1.1). Then there exists a neighborhood V ⊂ Ω
of x0 in which system (1.1) is C k equivalent to the differential system

ẏ1 = 1, ż = 0, where z = (y2 , . . . , yn ).

Proof A proof can be found, for instance, in [213]. Here we provide a proof for
completeness.
Since x0 is a regular point, we assume without loss of generality that f 1 (x0 ) = 0.
So there exists a neighborhood U ⊂ Ω of x0 , in which f 1 (x) = 0, x ∈ U . Thus in
U system (1.1) can be written as

dw
w := = p(x), x ∈ U, (1.7)
d x1

where x = (x1 , w), w = (x2 , . . . , xn ), p(x) = ( f 2 (x)/ f 1 (x), . . . , f n (x)/ f 1 (x)).

Clearly system (1.7) is C k in U , and the integral curves of system (1.7) are the orbits
of system (1.1) in U .
Set x0 = (x10 , . . . , xn0 ). For arbitrary x ∗ = (x10 , z) ∈ U , denote by Φx1 (x ∗ ) the
orbit of system (1.7) passing through x ∗ . Define a map along the integral curves of
system (1.7) as follows:

h : (x1 , z) −→ (x1 , Φx1 (x10 , z)).

Clearly h maps each line z = constant parallel to the x1 -axis in a neighborhood Vx0
of x0 to an integral curve of system (1.7), i.e. it maps the integral curves of the system

z = 0
1.1 Existence and Properties of the First Integrals 9

w w

z x∗ h x∗

x0 x0

O x10 x1 O x10 x1

Fig. 1.1 Mapping the straight lines parallel to the x1 -axis to the integral curves of system (1.7)

to the integral curves of system (1.7), see Fig. 1.1. By the continuous differentiability
of solutions of differential equations with respect to initial values, it follows that h
is a C k function on Vx0 . Next we will prove that h is a C k diffeomorphism from Vx0
to h(Vx0 ). With this aim in mind, we only need to prove the invertibility of h.
Since Φx1 (x10 , z) is the solution of Eq. (1.7) passing through (x10 , z), and

dΦx1 (x10 , z) 
 = p(x10 , z),
d x1 (x10 ,z)

we have
Φx1 (x10 , z) = z + p(x10 , z)(x1 − x10 ) + o(|x1 − x10 |).

This verifies that   

∂h  1 0
= .
∂(x1 , z) (x1 ,z)=x0 ∗E

Hence there exists a neighborhood V ⊂ U ∩ Vx0 ⊂ Ω of x0 on which h is invertible.

By the inverse function theorem [10, p. 372], the map h is a C k diffeomorphism
between V and h(V ). Note that the integral curves of system z  = 0 are the phase
orbits of system ẋ1 = 1, ż = 0, and the integral curves of system (1.7) are the phase
orbits of system (1.1). This proves the theorem. 

Proof of Theorem 1.2. From the proof of Theorem 1.3, we know that there
exists a neighborhood V ⊂ Ω of x0 and a C k diffeomorphism y = h(x) =
(h 1 (x), h 2 (x), . . . , h n (x)) defined on V which sends the orbits of system (1.1) to
those of the system
ẏ1 = 1, ẏi = 0, i = 2, . . . , n.

Obviously the latter has n − 1 functionally independent first integrals y2 , . . . , yn . So

we get from Proposition 1.2 that h 2 (x), . . . , h n (x) are functionally independent C k
first integrals of system (1.1). This completes the proof of the theorem. 

10 1 The Fundamentals of the Theory …

Theorem 1.2 verifies the local integrability of system (1.1) in a neighborhood of

a regular point. But the problem of the integrability of a differential system near a
singularity is much more complicated. It will be discussed in Chaps. 2 and 7.
In the proof of Theorem 1.2 we used the flow box theorem, which ensures that a
smooth differential system in a neighborhood of a regular point can be transformed
into a very simple form. This is a part of the theory of normal forms. Roughly
speaking, the theory of normal forms studies how to transform a differential system
in a neighborhood of a specified orbit into a simple form through an invertible change
of variables, where the latter can be easily studied from the dynamical point of view.
The theory of normal forms has been extensively studied and widely applied to study
the center-focus problem, center manifolds, limit cycles, chaos, different kinds of
bifurcations, and so on. For more details, see for instance [24–26, 30, 33, 42, 82, 83,
91, 92, 97, 132, 134, 138, 192, 195, 197, 198, 213, 230, 232, 246, 339, 344, 367,
390, 400, 401, 408, 409, 412, 413, 420, 444, 466, 487] and the references therein.
In Chaps. 2 and 7 we will introduce the basics of the theory of normal forms and its
applications.

1.2 First Integrals of Differential Systems in Canonical

Regions

Theorem 1.2 verifies not only the existence of local first integrals of the differential
system (1.1) in a neighborhood of a regular point, but also the regularity of the first
integrals. The question is: what about the existence and regularity of first integrals
of system (1.1) in the region formed by regular points? This problem was solved by
Li et al. [231] in 2002.
Let M be a two-dimensional smooth manifold. In this section all smooth manifolds
are either C ∞ or C ω and are connected without boundary (not necessary compact,
oriented). For precise definitions, see for example Sharpe [392, Sect. 1]. We note that
a compact set in a Euclidean space is bounded and closed. Let

X = P∂x + Q∂ y

be a C r vector field defined on M, r ∈ N ∪ {∞, ω}, where ∂x and ∂ y denote the

partial derivatives with respect to x and y, respectively. Of course, if the vector field
is C ω (i.e. P and Q are C ω ), it is necessary that the manifold must be C ω .
For arbitrary p ∈ M, denote by φ p (t) the solution of the vector field X passing
through p, and by γ p the orbit in M associated to φ p (t). Let I p be the maximal open
interval in R for which the solution φ p (t) is defined. That is

dφ p (t)
= X (φ p (t)), t ∈ Ip.
dt
Then the map
φ : D := {(t, p)| p ∈ M, t ∈ I p } −→ M
1.2 First Integrals of Differential Systems in Canonical Regions 11

is a C r flow on M, denoted by (M, φ).

The flow (M, φ) is called C k parallel, 1 ≤ k ≤ r , if it is C k equivalent to one of
the following flows:
• strip flow, i.e. (R2 , ψ), the flow ψ in cartesian coordinates is defined by ẋ =
1, ẏ = 0.
• annular flow, i.e. (R2 \ {0}, ψ), the flow ψ in polar coordinates is defined by
ṙ = 0, θ̇ = 1.
• spiral flow or radial flow, i.e. (R2 \ {0}, ψ), the flow ψ in polar coordinates is
defined by ṙ = r, θ̇ = 1 or ṙ = r, θ̇ = 0. These last two flows are topologically
equivalent. Here R2 \ {0} denotes the real plane without the origin.
• toral flow, i.e. (S1 × S1 , ψ), the flow ψ is a rational flow on the torus. On the
covering space R2 of the torus, the flow is defined by ẋ = α, ẏ = β, where α/β
or β/α is a rational number.
We remark that C k parallel flow was introduced by Neumann [348] in 1975,
whereas C 0 parallel flow was defined by Markus [330] in 1954.
A limit cycle γ of a smooth differential system defined on a two-dimensional
smooth manifold is a closed orbit (i.e. a periodic orbit) for which there exists a
neighborhood where the unique closed orbit is γ .

Remark
• If a two-dimensional differential system is analytic, then any periodic orbit Γ of
the system has a neighborhood which is either filled up with periodic orbits or
does not contain any periodic orbits other than Γ . This follows from the properties
of zeros of analytic functions. Hence a limit cycle of a two-dimensional analytic
differential system is also defined as an isolated periodic orbit.
• For non-analytic planar differential systems, there are examples showing that a
planar C ∞ differential systems may have a periodic orbit whose arbitrary small
neighborhood contains both periodic and non-periodic orbits, see e.g. Ye [451,
p. 47, Example 2].
• For planar polynomial differential systems, there stands Hilbert’s 16th problem,
which is one of the 23 problems posed by Hilbert at the second international
congress of mathematicians, Paris, 1900. This problem asks to give the maximal
number of limit cycles of planar polynomial differential systems of degree n and
their distribution.
Recall that the degree of a polynomial differential system ẋi = f i (x), i = 1, . . . , n,
x = (x1 , . . . , xn ), by definition is equal to max{deg f 1 (x), . . . , deg f n (x)}, where
deg f i denotes the degree of the polynomial f i , i = 1, . . . , n. The degree of a
vector field associated to this last polynomial differential system is the degree of
the system.

For p ∈ M and the orbit γ p passing through p, we denote by γ p+ and γ p− respec-

tively the positive and negative semi-orbits of γ p . The α limit set of γ p− and the ω
limit set of γ p+ are defined respectively as
12 1 The Fundamentals of the Theory …

α( p) = cl(γ p− ) \ γ p− and ω( p) = cl(γ p+ ) \ γ p+ ,

where cl denotes the closure of a set. We remark that the definitions of α and ω limit
sets given here differ from the classical ones (see Hirsch and Smale [207], or Chicone
[89]). However, this definition will be useful when defining a separatrix later on. We
note that in this definition, the limit sets of singularities and periodic orbits are empty,
whereas in the classical definition, the limit sets of singularities and periodic orbits
are themselves.
An open neighborhood U of an orbit γ p of a C r flow (M, φ) is C k parallel,
1 ≤ k ≤ r , if (U, φ) is C k equivalent to a parallel flow. For example, any periodic
orbit in a period annulus has a parallel neighborhood, whereas any neighborhood of
a limit cycle cannot be a parallel neighborhood. Recall that a period annulus is a
region which is filled up with periodic orbits.
The following definition of a separatrix was given by Neumann [348] and Markus
[330]. An orbit γ p of a flow (M, φ) is a separatrix of the flow if it cannot be contained
in a parallel neighborhood (U, φ) which satisfies the conditions
(S1 ) for any q ∈ U , α(q) = α( p), ω(q) = ω( p);
(S2 ) cl(U ) \ U consists of α( p), ω( p) and two orbits γa and γb satisfying α(a) =
α( p) = α(b), ω(a) = ω( p) = ω(b).

To further understand separatrices, we consider the orbits which are contained in

a parallel neighborhood satisfying (S1 ) and (S2 ). A regular orbit (i.e. not a singular
point) must be of one of the following three types: an orbit homeomorphic to R and
having different α and ω limit sets; an orbit homeomorphic to R and having the
same α and ω limit set; or a closed orbit. A parallel neighborhood in the first case is
shown in Fig. 1.2a, where α( p) and ω( p) can be a singular point, or a limit cycle,
or a polycycle. A parallel neighborhood in the second case is shown in Fig. 1.2b. In
addition any parallel neighborhood of a periodic orbit can only be an annulus filled
up with periodic orbits in the plane, or a region on a torus filled up with orbits of a
rational flow.
From the definition of a separatrix and the above remarks, we know that a singu-
larity, a limit cycle and the boundaries of a hyperbolic sector of a singularity are all

ω (p) α (p) = ω (p)

γp

γb
γa γb
γp
γb
α (p) γp
γa γa

(a) (b) (c)

Fig. 1.2 The parallel neighborhoods satisfying (S1 ) and (S2 )

1.2 First Integrals of Differential Systems in Canonical Regions 13

separatrices. The reason is that a neighborhood of a singularity cannot be parallel;

and a neighborhood of a limit cycle, or of a boundary of a hyperbolic sector of a
singularity either cannot be parallel, or is parallel but does not satisfy at least one of
the conditions (S1 ) and (S2 ).

Remark A rational flow on a torus has no separatrices, whereas each orbit of any
irrational flow on a torus is a separatrix.
 
Denote by the set of all separatrices of the flow (M, φ). Then is a closed

subset of M and is invariant under the flow φ. Each connected component of M \
is called a canonical region of the flow (M, φ) or simply φ. For example, the system
ẋ = x, ẏ =y has = {(0, 0)} as the set of separatrices and a unique canonical
region R2 \ . Whereas the system ẋ = −x, ẏ = y has = {(x, 0)| x ∈ R} ∪
{(0, y)| y ∈ R} as the set of separatrices and four canonical regions, which are the
four quadrants of the (x, y) plane.

A C k function H : M \ → R is a weak first integral of the vector field X if
H is a constant along each orbit located in each canonical region of the flow (M, φ)
of X , and it is not locally constant in M \ . We remark that the difference between
this definition and that of first integrals is that a weak first integral could not be
defined on the set formed by the separatrices.
The next result, due to Neumann [348], verifies the existence of continuous first
integrals.

Theorem 1.4 Any smooth flow φ on a two-dimensional smooth manifold M has a

continuous first integral in each canonical region of the flow.

For example, the system ẋ = x, ẏ = y has a continuous first integral in the

unique canonical region R2 \ {(0, 0)}:
⎧
⎨ y/x, for (x, y) between y = x and y = −x with x > 0 or x < 0,
H (x, y) = x/y, for (x, y) between y = x and y = −x with y > 0 or y < 0,
⎩
1, for y = x or y = −x.

The main tool for proving Theorem 1.4 is the following result obtained by Neu-
mann [348] in 1975.

Theorem 1.5 (Neumann’s Theorem) Let φ be a continuous flow defined on the

two-dimensional manifold M, then each canonical region of (M, φ) is C 0 parallel.

Proof The main part of the proof follows from Neumann [348] with a little modifi-
cation, see Chavarriga et al. [68].
Let R be a canonical region of φ, and ϕ = φ| R be the restriction of φ on R. Then
R is an open set formed either by orbits all homeomorphic to S1 , or by orbits all
homeomorphic to R. Otherwise, there exists at least one separatrix in R, but this
contradicts the fact that R is a canonical region.
14 1 The Fundamentals of the Theory …

In addition, any two different orbits of ϕ can be separated by two parallel neigh-
borhoods without intersection. In fact, for any two different orbits γ p , γq ∈ R, take
two sufficiently small open sections of the flow passing through p and q such that the
orbits with the initial points on the two sections form two parallel neighborhoods,
which do not intersect. Collapsing each orbit in R to a point, we obtain a quotient
space R/ϕ, which is a (Hausdorff) one-dimensional manifold. So the natural projec-
tion π : R → R/ϕ is a local trivial fiber of R over R or S1 . Since the flow provides
a natural orientation on the fibers, it follows that if the one-dimensional manifold
R/ϕ is compact, we have a rational flow on a torus, or a spiral flow or a radial flow,
and if R/ϕ is not compact, we have a strip flow, or an annulus flow.
By continuity of the projection map and the fact that a one-dimensional manifold
is homeomorphic to either S1 or R, it follows that the canonical region R is C 0
parallel. This completes the proof of the theorem. 

Proof of Theorem 1.4. Since each parallel flow has an analytic first integral, we get
from Neumann’s theorem that the flow (M, φ) has a continuous first integral in each
canonical region. 

In 1999 Chavarriga et al. [68] posed an open problem: What is the maximal order of
differentiability of the first integrals on a canonical region of a given two-dimensional
flow ϕ as a function of the order of differentiability of the flow?
Li et al. [231] in 2002 solved this problem, and obtained the following result.

Theorem 1.6 Let φ be a C r flow induced by a vector field X on a two-dimensional

C ∞ or C ω manifold M, r ∈ N ∪ {∞, ω}. The following statements hold.
(a) Each canonical region U of the flow φ is C r parallel.
(b) The flow φ restricted to each canonical region has a C r first integral.

Proof Statement (a) was proved by Neumann [348] in the C 0 case, see
Theorem 1.4. This is the key point in the proof of this theorem. Let M be a two-
dimensional C ∞ or C ω manifold where the C r flow φ is defined. Next we give a
sketch of the proof. For details, see [231].
If the manifold M is compact, the flow φ is complete; i.e. by definition, for every
q ∈ M, the orbit φq (t) = φ(t; q) is defined for all t ∈ R. See e.g. Sharpe [392].
Generally the manifold M is not compact and so the flow is not necessary com-
plete. We claim that the flow restricted to any non-compact canonical region U can be
rescaled so that it becomes complete. The proof of this claim will use the existence
of a Riemannian metric on U which is geodesically complete and is of the same
class of differentiability (C ∞ or C ω ) as the manifold M. We now prove the claim,
distinguishing C ∞ and C ω .
Case 1. The manifold M is of class C ∞ . Let g be an arbitrary Riemannian metric on
U of class C ∞ (see e.g. Abraham and Marsden [1] for a proof of the existence of the
metric). It is well known (see e.g. Warner [438]) that there is an increasing sequence
of compact subsets {Vm } such that Vm ⊂ U , Vm is contained in the interior of Vm+1
1.2 First Integrals of Differential Systems in Canonical Regions 15


and U = Vm . Then one can construct a C ∞ function σ on U for which the new
Riemannian metric g ∗ = eσ g has the property that the distance (with respect to g ∗ )
between Vm and cl(U − Vm+1 ) is bigger than one. As before, cl is the closure of a
set. Then it follows from the Hopf–Rinow theorem (see e.g. do Carmo [128]) that
g ∗ is geodesically complete. The Hopf–Rinow theorem states that for a connected
Riemannian manifold (M, g), the following statements are equivalent:
• All closed and bounded subsets of M are compact;
• M is a complete metric space, that is, any Cauchy sequence in M converges;
• M is geodesically complete; that is, for every p ∈ M, the exponential map Exp p
is defined on the entire tangent space T p M.

Case 2. The manifold M is of class C ω . This case is very involved. Using Whitney’s
embedding theorem [441] (which states that for a given n-dimensional differentiable
manifold N , there is a smooth proper embedding f : N −→ R2n+1 such that
f (N ) is an analytic submanifold of R2n+1 ) and the Hopf–Rinow theorem we can
get an analytic and geodesically complete metric g.
Define the vector field Z = X /X  on U , where  ·  is the norm associated to
the metric constructed above and let ψ be the flow of Z on U. Then ψ is complete
and C r .
Let (U, ψ) be a canonical region of M where ψ is either the complete C r flow
constructedabove if U is noncompact, or the original flow φ if U is compact. For
p ∈ U , let p be a local transversal section
 of the flow ψ with p ∈ p , and let U p
be the set {ψ(t, q) ∈ U | t ∈ R and q ∈ p }. Then if the canonical region (U, ψ) is a
strip or spiral, respectively annular or toral, the flow ψ defines a C r diffeomorphism:

ρ −1
p =ψ : p ×R −→ U p ,
(q, t) −→ ψ(t, q),

respectively 
ρ −1
p =ψ : p×S1 −→ U p ,
(q, t) −→ ψ(t T (q), q),

where T (q) is the minimal period for which the flow ψ with the initial point q moves
along S1 . Moreover, in both cases the inverse map ρ p is also a C r diffeomorphism.
In addition, the quotient space U/ψ has a natural structure of a C r manifold of
dimension 1, which is C r equivalent
 to either R or S1 . This follows from the fact that
for p, q ∈ U such that U p Uq = ∅, the map ρq ◦ ρ −1 r
p is a C diffeomorphism.
The above proof shows that the projection U → U/ψ is a locally trivial fibre
bundle of class C r with the fibre R or S1 (for the notions of fibre and fibre bundle,
see e.g. Sharpe [392]). Neumann’s theorem implies that this fibre bundle is trivial.
According to the different possibilities, the flow (U, ψ) is C r equivalent to either a
strip flow, an annular flow, a spiral flow, or a toral flow. This proves that the canonical
region U of M is C r parallel, which ends the proof of statement (a).
16 1 The Fundamentals of the Theory …

(b) From statement (a) it follows that for the C r flow ψ associated with the vector
field Z , in every canonical region U of M there exists a C r diffeomorphism h from
(U, ψ) onto (V, ξ ) which takes orbits of ψ onto orbits of ξ preserving or reversing
simultaneously the sense of all orbits, where (V, ξ ) is one of the four parallel flows.
For the strip and toral flows the first integral is H (x, y) = y, for the annular flow
the first integral is H (r, θ ) = r , and for the spiral or radial flow the first integral is
H (r, θ ) = r e−θ or H (r, θ ) = θ . These verify that h −1 ◦ H is a C r first integral of
the C r flow ψ on the canonical region U . This proves the statement.
This completes the proof of the theorem. 

The above provides a geometric proof of Theorem 1.6. It would also be interesting
to present an analytic one, but such a proof has not been published.

Remark A C r flow on a two-dimensional C ω manifold may not have C r first integrals

in the closure of a canonical region, r ∈ N ∪ {∞, ω}. For example, the planar linear
differential system
ẋ = −μx, ẏ = νy

is analytic in the full plane, where μ, ν > 0, and so its flow is analytic. Clearly the
origin is a saddle because this linear differential system has a positive and a negative
eigenvalues, and the system has the analytic first integral H (x, y) = |x|ν |y|μ in each
of the four canonical regions (i.e. the four quadrants). Obviously this first integral is
defined in the full plane, but it is in general not analytic, and even not smooth on the
separatrices, i.e. the x and y-axes.

For another example, the planar analytic differential system

ẋ = −y − x(x 2 + y 2 − 1), ẏ = x − y(x 2 + y 2 − 1)


has the set of separatrices := {x 2 + y 2 = 1} ∪ {0}, where the unit circle is a limit
cycle of the system, and the origin is a focus, because the linear part of the system has
a pair of complex (but not real) eigenvalues. Clearly the system has two canonical
regions. In polar coordinates the system has the first integral

H (r, θ ) = (r 2 − 1)r −2 e2θ ,

which is not defined at r = 0, and is analytic except at r = 0. Of course

(r, θ ) = (r 2 − 1)−1r 2 e−2θ

H

is also a first integral of the system, which is not defined on the limit cycle r = 1,
and is analytic outside r = 1.
Using Theorem 1.6, we can get a stronger result for planar polynomial differential
systems, see Li et al. [231, Theorem 3 and Corollary 4].
1.2 First Integrals of Differential Systems in Canonical Regions 17

Theorem 1.7 For planar polynomial vector fields

∂ ∂
X = P(x, y) + Q(x, y) , (x, y) ∈ R2 ,
∂x ∂y

with P and Q real polynomials and relatively prime (denoted by gcd(P, Q) = 1),
the following statements hold.
(a) There exist finitely many invariant
l curves and singularities of X in R , denoted
2

by γi, i = 1,
 . . . , l, such that i=1 γi consists of all separatrices of X , and
R2 \ ∪li=1 γi is composed of finitely many connected open sets, each of which is
a canonical region of the vector field X .
(b) The vector field X has an analytic first integral in each canonical region.

For simplicity, in what follows we also write the planar vector field X as X =
(P, Q).
The proof of this last theorem needs the following results. The first one charac-
terizes the limit sets of an orbit of a two-dimensional vector field, see e.g. Anosov
and Arnold [9], or Ye [451]. Recall that a polycycle consists of finitely many sin-
gularities p1 , . . . , pm and finitely many regular orbits γ1 , . . . , γm which satisfy
α(γi ) = pi , i = 1, . . . , m, ω(γi ) = pi+1 , i = 1, . . . , m − 1, ω(γm ) = p1 , see
Fig. 1.3.

Theorem 1.8 (Poincaré–Bendixson theorem) Let Y be a vector field on a compact

surface S with genus zero. If Y has only isolated singularities, then the ω limit set
of any phase orbit of Y can only be a singularity, a periodic orbit, or a polycycle.
Since we will not use the notion of genus, its definition is omitted here. We refer
the reader to do Carmo [129, p. 273]. Note that a sphere in R3 is a compact surface
of genus 0. This fact will be used later on.
The second result characterizes the local structure of isolated singularities of
planar analytic vector fields, see e.g. Ilyashenko [212], Lefschetz [227], or Ye [451].

Fig. 1.3 Illustration of a p2

polycycle

γ1 γ2

p1 p3

γm
γ3

pm p4
18 1 The Fundamentals of the Theory …

γ
γ γ γ

(a) (b) (c)

Fig. 1.4 Hyperbolic, elliptic and parabolic sectors, respectively

With this aim we need the following definitions. A hyperbolic sector of a planar
differential system at a singularity is a local region limited by two orbits inside
which all orbits resemble hyperbolae, see Fig. 1.4a. An elliptic sector at a singularity
is a local region limited by an orbit inside which all orbits have the singularity as
both α and ω limit sets, see Fig. 1.4b. A parabolic sector at a singularity is a local
region limited by two orbits inside which all orbits positively (or all negatively) flow
into the singularity, see Fig. 1.4c.

Theorem 1.9 An isolated singularity of a two-dimensional analytic vector field can

only be a center, a focus, or be formed by finitely many hyperbolic, elliptic and
parabolic sectors.

The Poincaré–Bendixson theorem is stated for vector fields on compact surfaces.

For a planar polynomial vector field X we need to compactify it in such a way that
we can apply the Poincaré–Bendixson theorem on S2 . Usually there are two ways
to compactify a planar polynomial vector field: Poincaré compactification and Ben-
dixson compactification. Here we choose the latter. By Bendixson compactification
we can get an equivalent vector field b(X ) on the sphere S2 , called the Bendixson
sphere. The Bendixson compactification is obtained by the stereographic projection
from N = (0, 0, 1) of R3 to a plane. More precisely, taking a straight line passing
through N , which intersects the unit ball centered at the origin and the plane (x, y, 0),
we get a projection from the ball to the plane (x, y, 0), which is a one to one mapping
from the ball minus the point N to the (x, y) coordinate plane, see Fig. 1.5. Treating
the plane R2 as the coordinate plane (x, y, 0) in R3 , then the ‘line at infinity’ of the
plane R2 corresponds to the north pole of the ball S2 . The flow of the compactified
vector field b(X ) in S2 \ {N } is topologically equivalent to that of X in R2 . The
stereographic projection can be expressed as follows. Let (x, y, 0) be the coordinates
of the coordinate plane and (u, v, w) be the coordinates in the ball S2 , then we have
 
2x 2y −1 + x 2 + y 2
(u, v, w) = , , .
1 + x 2 + y2 1 + x 2 + y2 1 + x 2 + y2
1.2 First Integrals of Differential Systems in Canonical Regions 19

Fig. 1.5 The stereographic z

projection
N

(u, v, w)

(x, y, 0)

Clearly this map is analytic.

Theorem 1.10 For a planar polynomial vector field X = (P, Q) with

gcd(P, Q) = 1, its Bendixson compactified vector field b(X ) is analytic in S2 ,
and has finitely many separatrices and finitely many canonical regions.

Proof Since b(X ) is obtained from X by stereographic projection, from the expres-
sion of the stereographic projection and some direct calculations we get that b(X )
is analytic.
Assume that the degree of the vector field X is m. Since P and Q are relatively
prime, Bézout’s theorem (see e.g. Fulton [157, p.112], or the following Theorem 1.11)
indicates that they have at most m 2 intersection points in the affine plane. Obviously
each intersection point is a singularity of the vector field X . Hence b(X ) has at most
m 2 + 1 singularities in S2 , including the north pole. Since b(X ) is an analytic vector
field, Theorem 1.9 shows that these singularities can only be a center, a focus, or one
whose neighborhood is formed by finitely many hyperbolic, elliptic and parabolic
sectors. The last one has only finitely many local separatrices.
By Theorem 1.8, on the Bendixson sphere S2 the α and ω limit sets of each orbit
of the vector field b(X ) can only be a singularity, a periodic orbit, or a polycycle.
We claim that the separatrices of b(X ) can only be singularities, limit cycles or the
boundaries of hyperbolic sectors of singularities. Indeed, by definition it follows that
singularities and limit cycles are separatrices. The boundaries of hyperbolic sectors
are separatrices because none of its neighborhoods can have a parallel neighbor-
hood satisfying the conditions (S1 ) and (S2 ) appearing in the definition of a parallel
neighborhood. In addition, it follows from Theorem 1.8 (i.e. the Poincaré–Bendixson
theorem) that the separatrices of b(X ) can only be singularities, limit cycles, or the
boundaries of hyperbolic sectors of singularities. Here we have used the fact that the
20 1 The Fundamentals of the Theory …

spiral orbits from a singularity to a limit cycle or a polycycle are not separatrices.
This proves the claim.
By the result of Ilyashenko [212] or of Écalle [138] which states that any given
planar polynomial vector field has at most finitely many limit cycles, we know that
the analytic vector field b(X ) has finitely many limit cycles on S2 . The above
proof shows that b(X ) has finitely many separatrices on the Bendixson sphere. In
addition, the boundaries of each canonical region consist of separatrices. All these
facts show that b(X ) has finitely many canonical regions on the Bendixson sphere.
This completes the proof of the theorem. 

Theorem 1.11 (Bézout’s theorem) Let F and G be projective planar curves of

degree m and n. If F and G have no common factors, then the number of their
intersection points in the projective plane is mn, taking into account multiplicity.

Remark In Chap. 4 we will introduce the background and some fundamental results
on the projective plane. Roughly speaking, a real projective plane is the quotient
space R3 / ∼, where ∼ represents the set of straight lines passing through the origin
of R3 but without the origin. The coordinates of a projective plane can be written
as [X : Y : Z ] with X, Y, Z not all zero. A projective planar curve of degree m is
defined by the set of zeros of a homogenous polynomial F(X, Y, Z ) of degree m in
the projective plane.

Proof of Theorem 1.7. By Theorem 1.10, the vector field b(X ) on the ball S2 has
finitely many separatrices and canonical regions. The analyticity of b(X ) together
with Theorem 1.6 show that b(X ) has an analytic first integral, say H , in each
canonical region of S2 .
Each of the separatrices of b(X ) in S2 except the north pole is an orbit of X in
R , i.e. a finite singularity or regular orbit. Clearly the number of these separatrices
2

is finite, and consequently X has finitely many separatrices and canonical regions
in R2 . Moreover, the map f from R2 to S2 defining the Bendixson compactification
is analytic. Hence H ◦ f is an analytic first integral of X in each canonical region
of R2 .
This completes the proof of the theorem. 

1.3 Applications of Integrability Theory to Partial

Differential Equations

In this section we apply the integrability theory of ordinary differential equations to

study the solutions of first-order linear homogeneous partial differential equations
and of first-order quasilinear partial differential equations.
1.3 Applications of Integrability Theory to Partial Differential Equations 21

1.3.1 First-Order Linear Homogeneous Partial Differential

Equations

For the first-order linear homogeneous partial differential equation


n
∂u
ai (x) = 0, x = (x1 , . . . , xn ) ∈ D ⊂ Rn an open region, (1.8)
i=1
∂ xi

we will use integrability theory to find its general solution and the solutions satisfying
initial conditions. Comparing with the equation that the first integrals of the system
(1.1) satisfy, we know that solving the partial differential equation (1.8) is equivalent
to finding all first integrals of the system

ẋ1 = a1 (x), . . . , ẋn = an (x). (1.9)

The differential equations (1.9) or their equivalent symmetric forms

d x1 d xn
= ··· = (1.10)
a1 an

are called the characteristic equations of the partial differential equation (1.8).
To ensure the existence of a solution of the system (1.9) or of (1.10), we assume
that

a1 , . . . , an ∈ C 1 (D) and |a1 (x)| + · · · + |an (x)| > 0, x ∈ D. (1.11)

Then systems (1.9) or (1.10) are the first-order n-dimensional autonomous ordinary
differential equations, and have all x ∈ D as regular points. So they have n − 1
functionally independent local first integrals in some neighborhood of any point in
D.

Theorem 1.12 Assume that system (1.9) satisfies conditions (1.11), and has n − 1
functionally independent and continuously differentiable first integrals in D

V1 (x), . . . , Vn−1 (x).

Then the general solution of the first-order linear partial differential equation (1.8)
is
u = F(V1 (x), . . . , Vn−1 (x)),

where F(·, . . . , ·) is an arbitrary continuously differentiable function of n − 1 vari-

ables.
22 1 The Fundamentals of the Theory …

Proof If V (x) is a first integral of (1.9), we get from Proposition 1.3 that there exists
a continuously differentiable function F of n − 1 variables such that

V (x) = F(φ1 (x), . . . , φn−1 (x)).

Furthermore, the solutions of the partial differential equation (1.8) are exactly the first
integrals of the characteristic equations (1.9). This proves that the general solution
of (1.8) is an arbitrary continuously differentiable function of V1 (x), . . . , Vn−1 (x).
The theorem follows. 

Remark The expression of the general solution in Theorem 1.12 is only local, because
it was obtained by using the Implicit Function Theorem.

Example Find the solution of the partial differential equation

∂u ∂u
y + z2 =0 (1.12)
∂x ∂z

passing through the surface x = 1, u = 1

z
+ 1y .

Solution The characteristic equations of (1.12) are

ẋ = y, ẏ = 0, ż = z 2 .

Clearly V1 (x, y, z) = y is a first integral of the characteristic equations. The first

integral V1 is defined in the full space and any integral curve of the characteristic
equations is located on some level surface of the first integral. So solving the equations

ẋ = y, ż = z 2

on the level surface y = c = 0 gives

x 1
= − + c1 .
c z

This shows that the characteristic equations have the first integral

x 1
V2 (x, y, z) = + ,
y z

which is functionally independent of V1 . By Theorem 1.13 we get the general solution

 
x 1
u = F y, +
y z
1.3 Applications of Integrability Theory to Partial Differential Equations 23

of the partial differential equation (1.12), where F is an arbitrary continuously dif-

ferentiable function.
Next we determine the function F satisfying the initial condition. Putting the
initial condition into the general solution yields
 
1 1 1 1
F y, + = + . (1.13)
y z z y

Setting ξ = y, η = 1y + 1z , we have y = ξ, z = η−ξ1 −1 . Then the functional equation

(1.13) has the expression
F(ξ, η) = η.

Consequently the partial differential equation (1.12) with the prescribed conditions
has the unique solution
x 1
u=η= + .
y z

1.3.2 First-Order Quasilinear Partial Differential Equations

In this subsection we study solutions of first-order quasilinear partial differential

equations


n
∂u
ai (x, u) = b(x, u), (x, u) ∈ G ⊂ Rn+1 an open region. (1.14)
i=1
∂ xi

To do so, we assume that

a1 , . . . , an , b ∈ C 1 (G) and |a1 (x, u)| + · · · + |an (x, u)| > 0, (x, u) ∈ G.
(1.15)
The system of ordinary differential equations

ẋ1 = a1 (x, u), . . . , ẋn = an (x, u), u̇ = b(x, u), (1.16)

or its equivalent symmetric forms

d x1 d xn du
= ··· = = , (1.17)
a1 (x, u) an (x, u) b(x, u)

are both called the characteristic equations of the partial differential equation (1.14).
Under the conditions (1.15), all points in G are regular points of systems (1.16) and
(1.17).
24 1 The Fundamentals of the Theory …

Theorem 1.13 Assume that system (1.16) satisfies conditions (1.15), and it has n
functionally independent and continuously differentiable first integrals in G

V1 (x, u) = c1 , . . . , Vn (x, u) = cn .

Then the general solution of the partial differential equation (1.14) is

F(V1 (x, u), . . . , Vn (x, u)) = 0, (1.18)

where F(·, . . . , ·) is an arbitrary continuously differentiable function of n variables.

Proof By Proposition 1.3 all first integrals of the characteristic equations (1.16) are
of the form
V (x, u) := F(V1 (x, u)), . . . , Vn (x, u)),

where F is an arbitrary continuously differentiable function of n variables. Theorem

1.12 shows that V (x, u) is a general solution of the partial differential equation

∂V ∂V ∂V
a1 (x, u) + · · · + an (x, u) + b(x, u) = 0. (1.19)
∂ x1 ∂ xn ∂u

First we claim that the solution u = ϕ(x) of the functional equation (1.18) is
a solution of the partial differential equation (1.14) (here we ask that ∂u V = 0).
Indeed, by the Implicit Function Theorem we have

∂ϕ(x) ∂V ∂V
=− / , i = 1, . . . , n.
∂ xi ∂ xi ∂u

Substituting this last expression into (1.19) gives

∂ϕ ∂ϕ
a1 (x, ϕ(x)) + · · · + an (x, ϕ(x)) = b(x, ϕ(x)).
∂ x1 ∂ xn

This verifies the claim.

Next we prove that any solution of the partial differential equation (1.14) can be
expressed in the form (1.18). Let u = ψ(x) be an arbitrary solution of (1.14). Set

Wi (x) = Vi (x, ψ(x)), i = 1, . . . , n.

Then for i = 1, . . . , n by (1.19) we have

∂ Wi ∂ Wi ∂ Vi ∂ Vi ∂ Vi
a1 + · · · + an = a1 + · · · + an +b = 0,
∂ x1 ∂ xn ∂ x1 ∂ xn ∂u
1.3 Applications of Integrability Theory to Partial Differential Equations 25

where we have used the facts that ψ(x) is a solution of (1.14) and Vi (x, u) is a first
integral of the characteristic equations (1.16).
The above proof shows that W1 (x), . . . , Wn (x) are the n first integrals of the
first-order n-dimensional system of the ordinary differential equations

ẋ1 = a1 (x, ψ(x)), . . . , ẋn = an (x, ψ(x)),

and so they must be functionally dependent. Hence there exists a continuously dif-
ferentiable function F of n variables such that

F(W1 (x), . . . , Wn (x)) = 0.

This shows that u = ψ(x) is a solution of the functional equation

F(V1 (x, u), . . . , Vn (x, u)) = 0.

The theorem is proved. 

Example Find the solution of the first-order partial differential equation

∂u ∂u ∂u x y2
x +y +z =u+ 2 .
∂x ∂y ∂z z

Solution The characteristic equations of the partial differential equation are

dx dy dz du
= = = .
x y z u + x y 2 z −2

Clearly they have the functionally independent first integrals

V1 (x, y, z, u) = yz −1 , V2 (x, y, z, u) = zx −1 .
y
On the level surface z
= c1 , the system of equations

dx du du 1
= , i.e. = u + c12 ,
x u + x y 2 z −2 dx x

has the general solution

u = x(c12 ln |x| + c),

which provides a first integral

V3 (x, y, z, u) = x −1 u − y 2 z −2 ln |x|
26 1 The Fundamentals of the Theory …

of the characteristic equations. It is easy to check that V1 , V2 , V3 are functionally

independent. Theorem 1.13 shows that the general solution of the original partial
differential equation is
 
F yz −1 , x −1 z, x −1 u − y 2 z −2 ln |x| = 0,

where F(ξ, η, ζ ) is an arbitrary continuously differentiable function with Fζ ≡ 0.

We can write the general solution of the original partial differential equation in the
explicit form  
u = xg yz −1 , x −1 z + x y 2 z −2 ln |x|,

where g is an arbitrary continuously differentiable function.

1.4 Lax Pairs and Integrability

Lax pairs were introduced by Peter Lax in 1968 in order to study the solutions of the
Korteweg–de Vries (KdV) equation

u t = 6u∂x u − ∂x3 u,

where ∂xk , k ∈ N, denotes the kth order partial derivative with respect to x. He
introduced a pair of linear operators

L = −∂x2 + u, (Sturm−Liouville operator)

P = −4∂x + 3(u∂x + ∂x u),
3

such that the KdV equation can be written as the Lax pair equation

∂t L = [P, L], (1.20)

where [P, L] = P L − L P is the commutator of the linear operators L and P. We

note that ∂t L = ∂t u. In fact, for an arbitrary continuously differentiable function
v(x, t), differentiating the equality

Lv = −∂x2 v + uv

with respect to t yields

∂t L v + L ∂t v = −∂t (∂x2 v) + u ∂t v + v ∂t u = L ∂t v + v ∂t u.

This reduces to ∂t L v = v ∂t u. Hence we have ∂t L = ∂t u.

1.4 Lax Pairs and Integrability 27

Generally, a Lax pair is a pair of matrices or linear operators (L , P), depending

on time, which act on a given Hilbert space and satisfy the Lax equation (1.20). Lax
pairs play an important role in the study of the integrability of differential systems.
In this section we mainly discuss the integrability of the system (1.1) via a Lax pair
(L , P), where L and P are two nth order matrices related to the solution x = x(t)
of system (1.1). In the Lax pair equation, ∂tL denotes
 the derivative of L along the
orbits of system (1.1). For example, if L = ai j (x) 1≤i, j≤n , then
   
∂t L = ∂t ai j (x(t)) 1≤i, j≤n = X (ai j )(x) 1≤i, j≤n ,

where x = x(t) is a solution of system (1.1).

Example The three-dimensional autonomous homogeneous differential system

ẋ = x(y − z), ẏ = y(z − x), ż = z(x − y) (1.21)

has the Lax pair

⎛ ⎞ ⎛ ⎞
0 1 x x+y 0 1
L = ⎝y 0 1⎠, P = ⎝ 1 y + z 0 ⎠.
1 z 0 0 1 x+z

This can be checked easily.

Remark In this last example, the Eq. (1.21) can be obtained from the Lax pair equa-
tion ∂t L = [P, L], because the Lax pair equation can be written as
⎛ ⎞ ⎛ ⎞
0 0 ẋ 0 0 x(y − z)
⎝ ẏ 0 0 ⎠ = ⎝ y(z − x) 0 0 ⎠.
0 ż 0 0 z(x − y) 0

Generally, if (L , P) is a Lax pair of system (1.1), and this system can be obtained
from the Lax pair equation ∂t L = [P, L], we call the Lax pair (L , P) regular.
Otherwise it is called nonregular. For example,
⎛ ⎞
a(x) 1  
0 0
L=⎝ 2 k ⎠, P=
−a (x) − ν j F j −a(x) −X (a) 0
j=1

is a nonregular Lax pair of system (1.1), where a(x) is an arbitrary continuously

differentiable function, F1 , . . . , Fk are first integrals of system (1.1), ν1 , . . . , νk are
arbitrary constants, and X is the vector field associated to system (1.1). But we
cannot get any useful information about system (1.1) from this Lax pair.
28 1 The Fundamentals of the Theory …

We now consider how to use Lax pairs to obtain first integrals of a differential
system.

Example The planar differential system

ẋ = a(t)y, ẏ = −a(t)x (1.22)

is a generalized Hamiltonian system, where a(t) is a continuously differentiable

function on R. It has the Lax pair
   
x y 0 a/2
L= , P= .
y −x −a/2 0

Some easy calculations show that L has the eigenvalues λ1,2 = ± x 2 + y 2 . Clearly
λ1,2 are first integrals of the system (1.22).

This simple example stimulates us to ask if a differential system has a Lax pair
(L , P), are the eigenvalues of L first integrals of the system?

Theorem 1.14 Assume that (L , P) is a Lax pair of system (1.1). The following
statements hold.
(a) If λ is an eigenvalue of L, then λ is a first integral of system (1.1).
(b) The determinant det L of L is also a first integral of system (1.1).

Proof (a) Assume that x = x(t) is an arbitrary nontrivial solution of system (1.1)
(i.e. it is not a singularity). Set L(t) := L(x(t)). Let U (t, s) be the solution matrix
of the initial value problem of the linear differential equation

dU
= P(x(t))U, U (s, s) = E, E is the unit matrix.
dt
Then one has
L(t) = U (t, s)L(s)U (t, s)−1 .

Indeed, direct calculations show that

dU (t, s)L(s)U (t, s)−1  

= P(x(t)), U (t, s)L(s)U −1 (t, s) .
dt
By the uniqueness of solutions of the initial value problem of the system of differential
equations, the claim follows.
This indicates that the matrices L(t) and L(s) are similar for all t, s ∈ R. Hence
the eigenvalues of L(t) are constants, i.e. each eigenvalue of L(x) is a constant along
each solution of system (1.1), and consequently it is a first integral of the system.
(b) This follows from (a), because det L is product of the eigenvalues of L(x).

1.4 Lax Pairs and Integrability 29

Theorem 1.14 has the following consequence.

Corollary 1.1 Assume that a matrix L(x) has n linearly independent eigenvectors
Vi (x), whose corresponding eigenvalues are λi (x), i = 1, . . . , n. Let X be the
vector field associated to system (1.1). Set

P(x) = (X (V1 ), . . . , X (Vn ))(V1 (x), . . . , Vn (x))−1 .

Then (L , P) is a Lax pair of system (1.1) if and only if each λi is a first integral of
system (1.1).

Proof By the assumption we have

L(x)Vi (x) = λi (x)Vi (x), X (Vi (x)) = P(x)Vi (x).

Let x(t) be a solution of the vector field X . Substituting this solution into the last
equation, and differentiating the first equation along the solution x(t) with respect
to t, we get

L t Vi + L(Vi )t = (λi )t Vi + λi (Vi )t = (λi )t Vi + λi P Vi ,

where L t denotes the derivative with respect to t. This last equation can be written
as
(L t − (P L − L P))Vi = (λi )t Vi ,

where we have used the fact that (Vi )t = X (Vi ) = P Vi . Hence, L t = [P, L] if and
only if (λi )t ≡ 0, i = 1, . . . , n. This implies that (L , P) is a Lax pair if and only if
the eigenvalues λi are all first integrals of system (1.1). 

Remark If the matrix L is semisimple, then L has n linearly independent eigenvec-

tors. Recall that a matrix is semisimple if it is diagonalizable.
The next result provides another application of Lax pairs in finding first integrals.
Hereafter tr(L) denotes the trace of the matrix L.
Theorem 1.15 If (L , P) is a Lax pair of system (1.1), then tr(L k ) is a first integral
of system (1.1) for arbitrary k ∈ N.

Proof Since

d Lk d L k−1 d L k−2 dL dL
= L +L L + · · · + L k−2 L + L k−1
dt dt dt dt dt
= (P L − L P)L k−1 + L(P L − L P)L k−2 + · · · + L k−2 (P L − L P)L
+ L k−1 (P L − L P)
= P L k − L k P,
30 1 The Fundamentals of the Theory …

one has  k
dtr(L k ) dL  
= tr = tr P L k − L k P ≡ 0,
dt dt

where in the last equality we have used the fact: tr(M N ) = tr(N M) for arbitrary
nth order matrices M and N . This proves that tr(L k ) is a constant along each orbit
of system (1.1). Consequently tr(L k ) is a first integral of system (1.1). The theorem
follows. 

Example For the Lax pair (L , P) of system (1.21), some calculations show that

tr(L) = 0,
tr(L 2 ) = 2(x + y + z),
tr(L 3 ) = 3(1 + x yz).

By Theorem 1.15 the three-dimensional system (1.21) has the two functionally inde-
pendent first integrals
H1 = x + y + z, H2 = x yz

and so it is completely integrable.

The next example shows that the first integrals of a differential system having a
Lax pair cannot be completely determined by Theorem 1.15.

Example The three-dimensional autonomous homogeneous differential system

ẋ = (2 − 3 )yz, ẏ = (3 − 1 )x z, ż = (1 − 2 )x y, (1.23)

with 1 , 2 , 3 real constants, has the Lax pair

⎛ ⎞ ⎛ ⎞
0 −z y 0 −3 z 2 y
L=⎝ z 0 −x ⎠ , P = ⎝ 3 z 0 −1 x ⎠ .
−y x 0 −2 y 1 x 0

Direct calculations show that

⎛ 2 ⎞
−y − z 2 xy xz
L =⎝
2
xy −x 2 − z 2 yz ⎠, L 3 = −(x 2 + y 2 + z 2 )L .
xz yz −x 2 − y 2

Furthermore we can prove by induction that for k ∈ N

L 2k+1 = (−1)k (x 2 + y 2 + z 2 )k L , L 2k = (−1)k−1 (x 2 + y 2 + z 2 )k−1 L 2 .

1.4 Lax Pairs and Integrability 31

Hence    
tr L 2k+1 = 0, tr L 2k = 2(−1)k (x 2 + y 2 + z 2 )k .

Thus by Theorem 1.15 we can get only the unique functionally independent first
integral H1 = x 2 + y 2 + z 2 of the system from the Lax pair.
But we can easily check that if 1 , 2 , 3 are pairwise distinct, system (1.23) has
the first integral H2 = 1 x 2 + 2 y 2 + 3 z 2 , which is functionally independent of H1 .
Even though the first integral H2 of system (1.23) cannot be obtained from the
trace of L k , sometimes by modifying the Lax pair we can find the second first integral.
Proposition 1.4 Assume that (L , P) is a Lax pair of system (1.1). Then the pair of
matrices with parameters
⎛ ⎞

k 
k
( := ⎝ L +
L, P) γj Mj, P + γ j N j ⎠ , γ1 , . . . , γk arbitrary constants,
j=1 j=1

is a Lax pair of system (1.1) if and only if

Ṁ j = [N j , L] + [P, M j ], j = 1, . . . , k,

(1.24)
M j , Ni + [Mi , N j ] = 0, 1 ≤ i, j ≤ k.

Proof This can be verified directly from the definition. 

Remark In Proposition 1.4, the equations in M j and N j are in general difficult to

solve. For an application, we consider a special case, taking k = 1, and M1 and N1
both diagonal.
For the Lax pair (L , P) of system (1.23), set k = 1, M1 = diag(m 1 , m 2 , m 3 ) and
N2 = diag(n 1 , n 2 , n 3 ). The system of equations (1.24) has a pair of solutions
 
−1 + 2 + 3 1 − 2 + 3 1 + 2 − 3
M = diag , , ,
2 2 2
 
(1 − 2 − 3 )2 (1 − 2 + 3 )2 (1 + 2 − 3 )2
N = diag , , .
4 4 4

Hence system (1.23) has the Lax pair (L + γ M, P + γ N ), where γ is an arbitrary

constant. Some calculations show that
(1 + 2 + 3 )γ
tr(L + γ M) = ,
2
2 2 2
(31 + 32 + 33 − 21 2 − 22 3 − 23 1 )γ 2
tr(L + γ M)2 = − 2(x 2 + y 2 + z 2 ),
4
tr(L + γ M)3 = K − 3γ (1 x 2 + 2 y 2 + 3 z 2 ),
32 1 The Fundamentals of the Theory …

where  

3 
i3 +3 i2  j − 181 2 3 γ 3
i=1 1≤i= j≤3
K = .
8
By Theorem 1.15 we get that system (1.23) has the first integrals

H1 = x 2 + y 2 + z 2 and H2 = 1 x 2 + 2 y 2 + 3 z 2 .

Of course, if 1 , 2 , 3 are pairwise distinct, then H1 and H2 are functionally inde-

pendent.
The last process provides a method to construct new Lax pairs from a given one.
The next result presents another method.

Theorem 1.16 Assume that (L , P) is a Lax pair of system (1.1). The following
statements hold.
(a) If A is an analytic function, then (A(L), P) is also a Lax pair of system (1.1).
(b) (L , Q) is a Lax pair of system (1.1) if and only if [L , P − Q] = 0.

Proof (a) Expanding A(z) as a Taylor series

∞

A(z) = ak z k ,
k=0

we have
∞

A(L) = ak L k .
k=0

Then the proof follows from the facts that for each c ∈ C and each k ∈ N, (cL k , P)
is a Lax pair of system (1.1), and that (L + K , P) is a Lax pair of system (1.1)
provided that (L , P) and (K , P) are both Lax pairs of system (1.1).
We now prove these two facts. From the proof of Theorem 1.15 we have

cL kt = c(P L k − L k P) = [P, cL k ].

On the other hand, we have

(L + K )t = L t + K t = (P L − L P) + (P K − K P)
= P(L + K ) − (L + K )P = [P, L + K ].
1.4 Lax Pairs and Integrability 33

This proves the two facts, and so the statement follows.

(b) By the assumption we have

L t = P L − L P, L t = Q L − L Q.

These are equivalent to

(P − Q)L − L(P − Q) = 0, i.e. [P − Q, L] = 0.

Hence the theorem is proved. 

In this section we have introduced some fundamental results on Lax pairs of

differential systems and their applications in finding first integrals. It is worth further
exploring the methods to construct suitable Lax pairs for a given differential system.
Chapter 2
Jacobian and Inverse Jacobian Multipliers

This chapter is mainly concerned with the existence and regularity of the Jacobian and
the inverse Jacobian multipliers of differential systems near a singularity, a periodic
orbit, or a polycycle. We will also use the vanishing multiplicity of inverse Jacobian
multipliers to study the multiplicity of a limit cycle, or of a homoclinic loop, and the
cyclicity of a singularity.

2.1 Jacobian Multipliers, First Integrals and Integrability

For n = 2, system (1.1) is called a planar differential system. It can be written in the
Pfaffian form or as a differential one-form

ω := f2 (x)dx1 − f1 (x)dx2 = 0. (2.1)

If there exists a C 1 function H such that

dH = ω,

we say the differential form ω is exact, and H is called a first integral of the differential
form. In this case system (1.1) is Hamiltonian, i.e.

dx1 dx2
= ∂x2 H, = −∂x1 H.
dt dt
Clearly H is a first integral of the Hamiltonian system.

Remark
• In R2m , let (x, y) ∈ Rm × Rm be the coordinate systems. A canonical Hamiltonian
system can be written as

© Springer Nature Singapore Pte Ltd. 2017 35

X. Zhang, Integrability of Dynamical Systems: Algebra and Analysis,
Developments in Mathematics 47, DOI 10.1007/978-981-10-4226-3_2
36 2 Jacobian and Inverse Jacobian Multipliers
⎞ ⎛
∂y1 H
⎜ .. ⎟
⎜ .⎟
    ⎜ ⎟
ẋ 0 E ⎜ ∂ym H ⎟
= ⎜
∇(x,y) H = ⎜ ⎟,
−E 0 ⎟
ẏ ⎜ −∂x1 H ⎟
⎜ .. ⎟
⎝ .⎠
−∂xm H

where E is the mth order unit matrix, and H is called a Hamiltonian function.
Obviously the Hamiltonian function is a first integral of the Hamiltonian system.
Recall that ∇(x,y) H is the gradient of H.
• If the differential form ω in (2.1) is exact, the divergence div(f ) = ∂x1 f1 + ∂x2 f2 of
f identically vanishes, where f = (f1 , f2 ) is the vector field associated to system
(1.1) with n = 2. So we have dω = 0. In this case, by definition the differential
form ω is closed. Generally, if a differential form in an n-dimensional space is
exact, it must be closed; but the inverse is not necessarily correct. We refer to Bott
and Tu [41] and Olver [351].
For the planar differential system (1.1), the following results are well known.
Proposition 2.1 Assume that Ω is a simply connected region of R2 , and that a
function M(x) is continuously differentiable on Ω, and M(x) = 0, x ∈ Ω. If the
divergence of the vector field Mf vanishes identically on Ω, i.e.

div(Mf ) := ∂x1 (Mf1 ) + ∂x2 (Mf2 ) ≡ 0, x ∈ Ω,

then system (1.1) has a continuously differentiable first integral in Ω. Moreover,

x
H(x) = Mf2 dx1 − Mf1 dx2 , x ∈ Ω,
x0

is one of the first integrals of system (1.1), where x0 ∈ Ω is an arbitrary fixed point,
and the integral is taken over any simple curve connecting x0 and x which is located
in Ω.
Remark
• The condition that Ω is simply connected ensures that the integral in Proposi-
tion 2.1 is uniquely determined by x. In other words, the integral is independent of
the choice of the passage from x0 to x.
• The assumption M(x) = 0, x ∈ Ω implies that system (1.1) and the system
ẋ = M(x)f (x) have the same first integrals and the same orbits. Their dynamics
can be determined by the first integral H.
• The function M(x) in Proposition 2.1 is called the integrating factor of system
(1.1).
• In some applications the integrating factor M(x) may vanish or may not be well
defined on a zero Lebesgue measure subset, say Ω ∗ , of Ω. If this is the case, the
2.1 Jacobian Multipliers, First Integrals and Integrability 37

dynamics of system (1.1) in Ω ∗ cannot be studied using the first integral, and we
need some other tools to investigate it. For instance, the differential equation

ydx − xdy = 0

has the integrating factor 1/y2 , which is defined on R2 \{y = 0}.

In higher-dimensional spaces, the corresponding notion of integrating factor is the

Jacobian last multiplier or simply Jacobian multiplier, which by definition is a non-
vanishing and continuously differentiable function M(x) defined on Ω, satisfying

div(Mf ) := ∂x1 (Mf1 ) + · · · + ∂xn (Mfn ) ≡ 0, x ∈ Ω,

where f (x) = (f1 (x), . . . , fn (x)) is the vector-valued function appearing in the right-
hand side of system (1.1).

Remark Similar to an integrating factor, a Jacobian multiplier may vanish or may

not be well defined on a zero Lebesgue measure subset of Ω.

The following proposition is one of Jacobi’s classical results, which characterizes

the existence of first integrals of system (1.1) via a Jacobian multiplier.

Proposition 2.2 Assume that M1 and M2 are two Jacobian multipliers of system
(1.1), and at least one of them is not zero on a full Lebesgue measure subset of Ω,
say M1 . If H(x) = M2 (x)/M1 (x) is not a constant, it is a first integral of system (1.1)
in Ω.

The proof of Proposition 2.2 follows easily from the definitions of first integral
and Jacobian multiplier. The details are left to the reader as an exercise.
Note that if the Jacobian multiplier M ≡ 1, the divergence of system (1.1) van-
ishes, i.e. divf ≡ 0. One of Liouville’s classical results shows that any volume in the
phase space is invariant under the flow of system (1.1) when its divergence vanishes,
see e.g. Arnold [13, Sect. 16, Theorem 1].

Theorem 2.1 (Liouville’s theorem) Let ϕt (x) be the flow of system (1.1) satisfying
ϕ0 (x) = x ∈ Ω and let D ⊂ Ω be an arbitrary bounded region. Set ϕt (D) =
{ϕt (x)| x ∈ D}. Assume that the divergence of system (1.1) is identically zero. If
ϕt (D) ⊂ Ω then
the volume of ϕt (D) = the volume of D.

In the following we continue to use the notations given in Liouville’s theorem.

For any continuously differentiable function M(x) defined on Ω, if the integral

M(x)dx (2.2)
ϕt (D)

is independent of the time t, it is called an invariant integral of system (1.1).

38 2 Jacobian and Inverse Jacobian Multipliers

Liouville’s theorem was extended by Poincaré, taking into account not only the
Jacobian multiplier and but also its integral along the phase flow of system (1.1).

Theorem 2.2 (Poincaré’s theorem) Assume that ϕt (x) is the flow of system (1.1)
satisfying ϕ0 (x) = x ∈ Ω, D ⊂ Ω is an arbitrary bounded region, and M(x) ∈
C 1 (Ω). Then (2.2) is an invariant integral of system (1.1) if and only if M(x) is a
Jacobian multiplier of system (1.1).

Clearly, Liouville’s theorem is a special case of Poincaré’s theorem.

Proof of Theorem 2.2. To prove that (2.2) is an invariant integral of system (1.1) is
equivalent to proving

d 
M(x)dx  ≡ 0, (2.3)
dt ϕt (D) t=t0

where t0 is an arbitrary point in the interval where the flow ϕt of system (1.1) is
defined. Since D ⊂ Ω is arbitrary, we only need to prove that (2.3) holds at t = 0.
Since ϕt : x → ϕt (x) is a diffeomorphism, and

ϕt (x) = x + f (x)t + O(t 2 ), |t| 1,

∂x ϕt (x) = E + ∂x f (x)t + O(t ),
2

where ∂x ϕt (x) is the Jacobian matrix of ϕt (x) with respect to x, we get from linear
algebra that

det(∂x ϕt (x)) = 1 + tr(∂x f (x))t + O(t 2 ), |t| 1,

where tr denotes the trace of a matrix, that is, the sum of the entries on the diagonal.
By calculus one has

M(y)dy = M(ϕt (x)) det(∂x ϕt (x))dx.

ϕt (D) D

Hence


d 
M(y)dy
dt ϕt (D)

 t=0
 
dϕt (x) d det(∂x ϕt (x)) 
= ∂y M(ϕt (x)) det(∂x ϕt (x)) + M(φt (x))  dx
dt dt

D
t=0

= (∂x M(x)f (x) + M(x)tr(∂x f (x))) dx = div(M(x)f (x))dx.

D D

Since M(x) and f (x) are both continuously differentiable, and D is arbitrary, the
equality (2.3) holds if and only if

div(M(x)f (x)) ≡ 0, x ∈ Ω.
2.1 Jacobian Multipliers, First Integrals and Integrability 39

This proves the theorem. 

In Chap. 1 we introduced the relation between first integrals of equivalent dif-
ferential systems. Next we discuss the relation between the Jacobian multipliers of
equivalent differential systems.
Proposition 2.3 Assume that y = G(x) is a continuously differentiable and invert-
ible transformation defined on Ω, and that a continuously differentiable function
M(x) is a Jacobian multiplier of system (1.1) in Ω. Then

N(y) := M(G −1 (y))DG −1 (y)

is a Jacobian multiplier of the differentiable system

ẏ = ∂x G(G −1 (y))f (G −1 (y)), (2.4)

where DG −1 (y) is the Jacobian of G −1 (i.e. the determinant of the Jacobian matrix
of G −1 ), and ∂x G(x) is the Jacobian matrix of G.
Proof We could prove this result from the definition, but the calculations are very
complicated. Instead, we follow an idea of Berrone and Giacomini [32] and prove it
by applying invariant integrals.
Let ϕt and ψt respectively be the flows of systems (1.1) and (2.4), which satisfy
ϕ0 (x) = x and ψ0 (y) = y, respectively, with y = G(x). Since y = G(x) is an
invertible transformation sending system (1.1) to system (2.4), it follows that the
flows of the systems are conjugate, i.e.

G −1 ◦ ψt = ϕt ◦ G −1 .

Assume that V ⊂ G(Ω) is an arbitrary bounded region. Then we have

N(y)dy = M(G −1 (y))DG −1 (y)dy

ψt (V ) ψt (V )

= M(x)dx = M(x)dx.
G −1 ◦ψt (V ) ϕt ◦G −1 (V )

Since M(x) is a Jacobian multiplier of system (1.1), it follows from Theorem 2.2
that the integral in the left-hand side of these last equalities is an invariant integral
of system (2.4). Hence N(y) is a Jacobian multiplier of system (2.4). 
We now discuss how to use Jacobian multipliers to construct first integrals. For k
(k ≤ n) functions p1 (x), . . . , pk (x) defined on a domain Ω of an n-dimensional space,
we recall that (c1 , . . . , ck ) ∈ Cn are regular values of these functions if p1 , . . . , pk are
functionally independent at all points x ∈ Ω such that p1 (x) = c1 , . . . , pk (x) = ck .
This kind of point x is a regular point of the functions p1 (x), . . . , pk (x). If this is not
the case, we call (c1 , . . . , ck ) critical values of p1 , . . . , pk , and correspondingly x is
a critical point.
40 2 Jacobian and Inverse Jacobian Multipliers

Sard’s theorem states that critical values form a zero Lebesgue measure subset,
see Sard [382], or [410, Theorem II.3.1] and [354].

Theorem 2.3 (Sard’s theorem) Let

f : Rn −→ Rm

be a C k function, where k ≥ max{n − m + 1, 1}. If V is a critical set of f , i.e. a set

formed by critical points of f , then f (V ) ⊂ Rm is a zero Lebesgue measure subset.

Theorem 2.4 (Jacobi’s theorem) Assume that the differential system (1.1) has n −2
functionally independent and continuously differentiable first integrals in Ω

H1 (x), . . . , Hn−2 (x)

and it has a continuously differentiable Jacobian multiplier M(x). Then system

(1.1) has a first integral defined in some neighborhood of each regular point of
(H1 , . . . , Hn−2 ) which is functionally independent of H1 (x), . . . , Hn−2 (x).

Proof Assume that V ⊂ Ω is an open domain, and consists of regular points of

(H1 , . . . , Hn−2 ). Without loss of generality we assume that

yi = Hi (x), i = 1, . . . , n − 2,
yn−1 = xn−1 , (2.5)
yn = xn

is invertible in V . Denote by y = G(x) the transformation defined in (2.5), and by

Δ the Jacobian of this transformation, i.e. Δ = det (∂x G(x)). Then system (1.1) is
equivalent to the system

ẏi = 0, i = 1, . . . , n − 2,
−1
ẏn−1 = fn−1 ◦ G (y), (2.6)
−1
ẏn = fn ◦ G (y),

via the transformation (2.5). Clearly system (2.6) has the functionally independent
first integrals Ii (y) = yi , i = 1, . . . , n − 2. By Proposition 2.3, system (2.6) has the
Jacobian multiplier

N(y) := M ◦ G −1 (y)DG −1 (y), y ∈ G(V ).

This implies that the two-dimensional system

ẏn−1 = fn−1 ◦ G −1 (I1 , . . . , In−2 , yn−1 , yn ) =: gn−1 (yn−1 , yn ),

ẏn = fn ◦ G −1 (I1 , . . . , In−2 , yn−1 , yn ) =: gn (yn−1 , yn )
2.1 Jacobian Multipliers, First Integrals and Integrability 41

has the integrating factor

L(yn−1 , yn ) = M ◦ G −1 (y)DG −1 (I1 , . . . , In−2 , yn−1 , yn ).

Hence it follows from Proposition 2.1 that this last two-dimensional system has the
first integral

In−1 (yn−1 , yn ) = Lgn dyn−1 − Lgn−1 dyn .

Obviously In−1 is functionally independent of I1 , . . . , In−2 , because the latter are

independent of yn−1 and yn .
By Proposition 1.2, Hn−1 := In−1 ◦ G(x) is a first integral of system (1.1).
From the functional independence of I1 , . . . , In−2 , In−1 , we can easily prove that
H1 (x), . . . , Hn−2 (x), Hn−1 (x) are functionally independent.
This completes the proof of the theorem. 
Next we discuss the relation between Jacobian multipliers and first integrals of
completely integrable autonomous differential systems. In this book the superscript
τ will denote the transpose of a matrix, or of a vector.
Proposition 2.4 Assume that system (1.1) has n − 1 functionally independent and
twice continuously differentiable first integrals H1 , . . . , Hn−1 in Ω. Set

Mi = ∂1 H, . . . , ∂i−1 H,
∂i H, ∂i+1 H, . . . , ∂n H ,

for i = 1, . . . , n, where the hat denotes the absence of that element in the matrix,
∂j H = (∂j H1 , . . . , ∂j Hn−1 )τ and ∂j is the partial derivative with respect to xj , j =
1, . . . , n. The following statements hold.
(a) For any smooth function I(x) we have

X (I) = a(x)Dx (I, H1 , . . . , Hn−1 ),

where X is the vector field associated with system (1.1), Dx (I, H1 , . . . , Hn−1 )
is the Jacobian of (I, H1 , . . . , Hn−1 ) with respect to x, and

a(x) = fi (x)Wi (x)−1 ,

for some i ∈ {1, . . . , n} such that Wi (x) := (−1)1+i Δi = 0 with Δi = det Mi .

(b) a(x) is an inverse Jacobian multiplier, i.e. 1/a(x) is a Jacobian multiplier.
Proof First we claim that if Wi (x) = 0, then

fj (x) = Wj (x)Wi (x)−1 fi (x), j = 1, . . . , n.

We now prove this claim. Since H1 (x), . . . , Hn−1 (x) are functionally independent
and smooth first integrals of X , we have
42 2 Jacobian and Inverse Jacobian Multipliers

f1 (x)∂1 H(x) + · · · + fn (x)∂n H(x) = 0.

This equation can be equivalently written as

⎛ ⎞
f1
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
⎜ fi−1 ⎟
Mi ⎜ ⎟
⎜ fi+1 ⎟ = −fi ∂i H.
⎜ ⎟
⎜ . ⎟
⎝ .. ⎠
fn

Solving this linear algebraic equation via Cramer’s rule gives

fj = −Aj Δi −1 fi , j = 1, . . . , i − 1, i + 1, . . . , n,

where Aj is the determinant of the matrix obtained from Mi replacing its jth column
by ∂i H. Direct calculations show that

Aj = (−1)i−j−1 Δj = (−1)i−1 Wj .

This proves the claim.

Next we claim that

n
∂j Wj ≡ 0.
j=1

In fact, we get from

 
Δi = det ∂1 H, . . . , ∂i−1 H, ∂
i H, ∂i+1 H, . . . , ∂n H

that

n
 
∂i Wi = (−1)1+i det ∂1 H, . . . , ∂i ∂s H, . . . , ∂
i H, . . . , ∂n H .
s=1,s=i

Note that any element in the summation of ∂i Wi , for example

 
(−1)1+i det ∂1 H, . . . , ∂i ∂j H, . . . , ∂
i H, . . . , ∂n H , (2.7)

with i > j, has a counterpart in ∂j Wj , i.e.

 
(−1)1+j det ∂1 H, . . . , ∂
j H, . . . , ∂i ∂j H, . . . , ∂n H , (2.8)

such that their sum identically vanishes. This proves the claim.
Now we can prove the theorem.
2.1 Jacobian Multipliers, First Integrals and Integrability 43

(a) The first claim shows that for any smooth function I


n 
n 
n
X (I) = f j ∂j I = Wj Wi −1 fi ∂j I = Wi −1 fi Wj ∂j I. (2.9)
j=1 j=1 j=1

By the properties of the determinant one can check that


n
Dx (I, H1 , . . . , Hn−1 ) = Wj ∂j I. (2.10)
j=1

Then the proof of statement (a) follows from (2.9) and (2.10).
(b) This statement follows from
 
X Wi −1 fi = Wi −1 fi div(X ). (2.11)

We now prove this fact. Replacing I in (2.9) by fi /Wi yields

  
n
 
X Wi −1 fi = Wi −1 fi Wj ∂j Wi −1 fi . (2.12)
j=1

Since fj Wi = fi Wj , differentiating this equation with respect to xj gives

∂j fj Wi + fj ∂j Wi = ∂j fi Wj + fi ∂j Wj . (2.13)

Hence we have
 
n
 
X Wi −1 fi = Wi −3 fi ∂j fi Wi − fi ∂j Wi Wj
j=1
n
  
n
= Wi −2 fi ∂j fj Wi + fj ∂j Wi − fi ∂j Wj − Wi −3 fi fi ∂j Wi Wj
j=1 j=1

n
 
= Wi −1 fi div(X ) + Wi −3 fi Wi fj ∂j Wi − fi Wj ∂j Wi (2.14)
j=1

= Wi −1 fi div(X ),

where in the second equality we have used (2.13), and in the third and fourth equalities
we have used the second and first claims, respectively. This proves statement (b) and
consequently the theorem. 

Proposition 2.4 has been extended to systems of vector fields for matrix Jacobian
multipliers, see Weng and Zhang [440, Theorem 1.3].
44 2 Jacobian and Inverse Jacobian Multipliers

The last proposition verifies the existence of Jacobian multipliers of a completely

integrable differential system. The next result further characterizes the essential
property of completely integrable differential systems, and their relation with their
Jacobian multipliers.

Theorem 2.5 Consider the C k autonomous differential system (1.1) and its associ-
ated vector field X with k ∈ (N\{1}) ∪ {∞, ω} defined in Ω. Assume that system
(1.1) is C r completely integrable in Ω with 2 ≤ r ≤ k, div X ≡ 0, and that the
Lebesgue measure of the set of its singularities is zero. Let H1 (x), . . . , Hn−1 (x) be
n − 1 functionally independent C r first integrals. Then the following statements hold.

(a) If J(x) is a smooth Jacobian multiplier of system (1.1), then J(x) is functionally
independent of H1 (x), . . . , Hn−1 (x).
(b) There exists a full Lebesgue measure subset Ω0 ⊂ Ω in which system (1.1) is
C r−1 orbitally equivalent to the linear differential system

ẏ = y. (2.15)

We remark that statement (a) was first proved by Llibre et al. in [292]. Statement
(b) was proved for two-dimensional differential systems by Giné and Llibre in [183],
and for any finite-dimensional differential systems by Llibre et al. in [292]. We note
that Theorem 2.5 is similar to the flow box theorem in some sense.
Proof of Theorem 2.5. Since H1 (x), . . . , Hn−1 (x) are functionally independent C r
first integrals, we can assume without loss of generality that

W (x) := det (∂1 H, . . . , ∂n−1 H) = 0, x ∈ Ω0 ⊂ Ω,

where Ω0 is a full Lebesgue measure subset of Ω, and H = (H1 , . . . , Hn−1 )τ . For

i = 1, . . . , n − 1, set

Wi (x) := det (∂1 H, . . . , ∂i−1 H, ∂n H, ∂i+1 H, . . . , ∂n−1 H) .

Statement (b) of Proposition 2.4 shows that

Q(x) = W (x)fn (x)−1

is a Jacobian multiplier of system (1.1) in Ω0 except perhaps in a zero Lebesgue

measure subset. Note that fn (x) can vanish only on a zero Lebesgue measure subset
of Ω0 , otherwise f (x) = (f1 (x), . . . , fn (x)) identically vanishes, because fi (x) =
−Wi (x)W (x)−1 fn (x), i = 1, . . . , n − 1, which could be obtained from the proof of
Proposition 2.4.
(a) First we prove that Q(x) is functionally independent of H1 (x), . . . , Hn−1 (x). Set
 
∂1 Q · · · ∂n−1 Q ∂n Q
A := det .
∂1 H · · · ∂n−1 H ∂n H
2.1 Jacobian Multipliers, First Integrals and Integrability 45

Expanding the determinant A with respect to the first row yields


n
A= ∂i Q Qi∗ , (2.16)
i=1

where
Qi∗ = (−1)1+i Δi , (2.17)

with Δi the determinant of the (n − 1) × (n − 1)-matrix which is obtained from the

matrix A by removing the first row and the ith column. Then

W (x) = Δn , Wi (x) = (−1)n−1−i Δi , i = 1, . . . , n − 1. (2.18)

From (2.16)–(2.18) together with Wi (x) = −Qfi one gets

A = (−1)n+1 QX (Q) = (−1)n+1 Q2 divX .

This shows that A can vanish only in a zero Lebesgue measure subset of Ω. Hence,
Q, H1 , . . . , Hn−1 are functionally independent in Ω.
If J/Q ≡ constant, it is clear that J is functionally independent of H1 , . . . , Hn−1 . If
J/Q ≡ constant, then J/Q is a first integral of system (1.1). By Theorem 1.1 J/Q and
H1 , . . . , Hn−1 are functionally dependent. It follows that J is functionally independent
of H1 , . . . , Hn−1 , because Q is functionally independent of H1 , . . . , Hn−1 , which ends
the proof of the statement.
(b) By Proposition 2.4 and statement (a) it follows that ∇J, ∇H1 , . . . , ∇Hn−1 are
linearly independent at all points of a full Lebesgue measure subset Ω 0 ⊂ Ω. Taking
the invertible change of coordinates

y1 = J(x)H1 , . . . , yn−1 = J(x)Hn−1 , yn = J(x), 0 ,

x∈Ω

one has

ẏi = −yi divf , ẏn = −yn divf .

This proves the statement, and consequently the theorem. 

2.2 Inverse Jacobian Multipliers and Their Vanishing Sets

Consider the smooth differential system (1.1) defined in Ω or its associated vector
field X , let Ω0 be an open subset of Ω. A function V ∈ C 1 (Ω0 ) is an inverse
Jacobian multiplier of system (1.1) if
46 2 Jacobian and Inverse Jacobian Multipliers

X (V ) = V divX .

For n = 2, the function V is also called an inverse integrating factor. Here ‘inverse’
comes from the fact that if V is an inverse Jacobian multiplier of (1.1), then M = V −1
is a Jacobian multiplier of (1.1) in Ω0 \V −1 (0), where V −1 (0) = {x ∈ Ω0 | V (x) = 0}.

Proposition 2.5 If V is an inverse Jacobian multiplier of the vector field X , then

V −1 (0) is an invariant set of X .

Proof By definition of the inverse Jacobian multiplier, it follows that the vector field
X is tangent to V −1 (0). So V −1 (0) is formed by the orbits of the vector field X .
This means that V −1 (0) is invariant under the flow of the vector field. 

The above provides a geometric proof of Proposition 2.5. It can also be proved
using 
t 
V (ϕt (x)) = V (x) exp divX ◦ ϕs (x)ds , (2.19)
0

which follows from

dV (ϕt (x))
= X (V ) ◦ ϕt (x) = V divX ◦ ϕt (x),
dt

where ϕt (x) is the solution of system (1.1) satisfying the initial condition ϕ0 (x) = x.
The following result characterizes the location of limit cycles of a planar smooth
differential system via the inverse integrating factors.

Theorem 2.6 Assume that the planar analytic vector field X defined in Ω has an
inverse integrating factor V . If Γ is a limit cycle of X , then Γ ⊂ V −1 (0).

This result was obtained by Giacomini et al. [172, Theorem 9] in 1996, and plays
an important role in the study of integrability, the center-focus problem, the limit
cycle bifurcation and its relation with Lie symmetry, and so on. In this direction
there are plenty of results, see e.g. [31, 64, 66, 68, 70, 72, 99, 115, 116, 139, 149,
150, 160, 161, 163, 164, 167, 168, 171, 178, 185, 186, 209, 226, 257] and the
references therein.
In the following we provide a proof of Theorem 2.6, which is different from
the original one in [172], where Giacomini et al. applies the first-order de Rham
cohomology.
Proof of Theorem 2.6. By Proposition 2.5 it follows that V −1 (0) is an invariant set of
system (1.1). So if V −1 (0) ∩ Γ = ∅, then
 Γ ⊂ V −1 (0).
−1
On the contrary we assume that Γ V (0) = ∅. Then there exists a neighbor-
hood Ω0 of Γ in which V does not vanish. So M = 1/V is an integrating factor of
system (1.1) in Ω0 . Obviously X and MX have the same orbits in Ω0 . Hence Γ is
also a limit cycle of MX . In addition, it follows from Theorem 2.1 (i.e. Liouville’s
theorem) that the flow of the vector field MX is area preserving. But this is not
possible, because for a domain ω0 ⊂ Ω0 located in one side of Γ the limit set of
2.2 Inverse Jacobian Multipliers and Their Vanishing Sets 47

ω0 under either positive or negative flow of MX is the limit cycle, and so its area
hasthe limit zero as either t → ∞ or t → −∞. This contradiction implies that
Γ V −1 (0) = ∅ and so Γ ⊂ V −1 (0). 
Theorem 2.6 was extended to smooth vector fields on orientable, connected and
smooth Riemannian surfaces by Athanassopoulos [16] in 2007.

Remark Theorem 2.6 shows that inverse integrating factors have played a more
important role than integrating factors in the study of the dynamics of planar differ-
ential systems. We refer the reader to the survey paper [164] by García and Grau.

We now study the multiplicity of limit cycles via inverse integrating factors.
Assume that Γ is a limit cycle of system (1.1). For any q ∈ Γ and a suffi-
ciently small neighborhood Ω0 of Γ , we define a Poincaré section  := {ψ(s)| s ∈
(−ε, ε)} ⊂ Ω0 , which is transversal to the flow of system (1.1), where ψ is an ana-
lytic function satisfying ψ(0) = q. Let φt (x) be the flow of system (1.1). It induces
a map on the Poincaré section , called the Poincaré map, and denoted by P. It is
well known that P has the same regularity as that of the differential system, see e.g.
Ilyashenko and Yakovenko [213], or Zhang [473, Sect. 6.2].
Set S1 := R/Z and J = (−ε, ε). Then there exists a diffeomorphism Ψ : J ×
S −→ Ω0 , a sufficiently small neighborhood of Γ , which provides a local coordinate
1

system (s, θ ) in Ω0 with s ∈ J and θ ∈ S1 , satisfying Ψ −1 ◦ ψ(s) = (s, 0), see e.g.
Fig. 2.1. Clearly the last condition means that  = {Ψ (s, 0), s ∈ J}. Ye [451, Sect.
2] presented a concrete such diffeomorphism, i.e.

x(s, θ ) = ϕ(θ ) − sψ  (θ ), y(s, θ ) = ψ(θ ) + sϕ  (θ ),

where (ϕ(θ ), ψ(θ )) is a parameterization of the limit cycle Γ .

(p)
Ψ (s, 0)
p
q q
Ψ Ψ (s, θ )
ε Σ Σ

(s, θ )
0 1

−ε Γ Γ

(a) Local coordinates (b) Poincaré map

Fig. 2.1 A Poincaré section and Poincaré map

48 2 Jacobian and Inverse Jacobian Multipliers

Under the local coordinates (s, θ ), system (1.1) can be written as

ṡ = g1 (s, θ ), θ̇ = g2 (s, θ ).

Obviously g1 (0, θ ) ≡ 0, and g2 = 0 on J × S1 for |ε| sufficiently small. So this last

system of equations can be written as a differential equation

ds
= F(s, θ ). (2.20)
dθ
Denote by

X = ∂θ + F(s, θ )∂s

the vector field associated to the Eq. (2.20). Note that if system (1.1) is analytic then

X is analytic, because we can choose the diffeomorphism Ψ (s, θ ) to be analytic. Let
ηθ (s) be the solution of this last equation with η0 (s) = s. In the coordinates (s, θ ),
the Poincaré map P can be expressed as P(s) = η1 (s), because this last equation is
periodic of period 1 in θ . It is well known that if system (1.1) is analytic, then the
Poincaré map P is analytic.
By construction of the Poincaré map P, it follows that P(0) = 0 because s = 0
corresponds to the limit cycle Γ . We say the limit cycle Γ is
• hyperbolic, if P (0) = 1;
• of multiplicity m (m ≥ 2), if P(s) = s + βm sm + O(sm+1 ), βm = 0.
Note that if P(s) ≡ s then Γ belongs a period annulus, and it is not a limit cycle.
Assume that the planar differential system (1.1) is analytic, and it has the limit
cycle Γ . Let V be an inverse integrating factor of system (1.1) in a neighborhood of
Γ . By Proposition 2.3 the inverse integrating factor V of system (1.1) is transformed
 of Eq. (2.20) with
by (x, y) = Ψ (s, θ ) to the inverse integrating factor V

(s, θ ) = V (Ψ (s, θ ))
V .
g2 (s, θ )DΨ (s, θ )

Recall that DΨ (s, θ ) is the Jacobian of Ψ (s, θ ). The inverse integrating factor V has
a zero of multiplicity m over Γ , or vanishing multiplicity m over Γ , if

(s, θ ) = g(θ )sm + O(sm+1 ),

V

with g(θ ) ≡ 0 on Γ . García et al. [160, Lemma 7] proved that g(θ ) = 0 for θ ∈ S1 ,
which can be obtained in the following way. By (2.19) we have

θ 
V (η0 (s0 ), 0) exp
(ηθ (s0 ), θ ) = V ∂s F(ηρ (s0 ), ρ)dρ , (2.21)
0
2.2 Inverse Jacobian Multipliers and Their Vanishing Sets 49

where ηθ (s0 ) is the solution of equation (2.20) satisfying the initial condition s(0) =
s0 . Expanding the functions on the two sides of (2.21) as Taylor series of s0 , and
comparing the coefficients of s0m , one has

θ 
g(θ ) = g(0) exp ∂s F(0, ρ)dρ .
0

Since g(θ ) ≡ 0 on S1 , it follows that g(θ ) = 0 on S1 . Otherwise g(θ ) ≡ 0 on S1 .

The next result, due to García et al. [160, Theorem 4], reveals the relation between
the inverse integrating factors and the multiplicity of a limit cycle. Recall that a period
annulus is a region filled up with periodic orbits.
Theorem 2.7 Let Γ be a limit cycle of a planar analytic differential system (1.1),
and let V be an analytic inverse integrating factor of system (1.1) in a neighborhood
of Γ . The following statements hold.
(a) If Γ is a limit cycle of multiplicity m, then V has a zero of multiplicity m over Γ .
(b) If V has a zero of multiplicity m over Γ , then Γ is either a limit cycle of multi-
plicity m or belongs to a period annulus.
 of Eq. (2.20) and the
Proof We first claim that the inverse integrating factor V
Poincaré map P(s) satisfy

(P(s), 1) = V
V (s, 0)P (s). (2.22)

Indeed, it follows from (2.21) that


1 
(P(s), 1) = V
V (s, 0) exp ∂s F(ηθ (s), θ )dθ . (2.23)
0

Note that ∂s ηθ (s) satisfies the variational equation

∂w ∂F
(s, θ ) = (ηθ (s), θ )w(s, θ ), w(s, 0) = 1.
∂θ ∂s
Integrating this equation with respect to θ from 0 to 1 gives

1 
w(s, 1) = exp ∂s F(ηθ (s), θ )dθ .
0

By definition, one has P (s) = ∂s η1 (s) = w(s, 1). This proves the claim.
The remaining proof can be completed by using (2.22) and the Taylor expansions
(s, 1) and P(s) with respect to s. The details are omitted.
of V
We complete the proof of the theorem. 
Remark Giacomini et al. [173] studied the semistable and convex limit cycles by
using the multiplicity of the inverse Jacobian multipliers on the limit cycles.
50 2 Jacobian and Inverse Jacobian Multipliers

Theorem 2.7 establishes a relation between the multiplicities of a limit cycle Γ and
of the inverse integrating factors over Γ . What about the existence and the regularity
of inverse integrating factors of planar differential systems in a neighborhood of a
limit cycle?
Ensiso and Peralta-Salas [139] in 2009 conducted pioneering work in this direc-
tion, and obtained a series of fundamental results in some neighborhood of limit
cycles and of singularities.

Theorem 2.8 Assume that Γ is a limit cycle of a planar analytic differential system
(1.1). The following statements hold.
(a) System (1.1) has a C ∞ inverse integrating factor V in a neighborhood Ω0 of
Γ , and V vanishes only on Γ .
(b) The Taylor expansion of V at any point of Γ is not identically zero.
(c) V is analytic in Ω0 if and only if the Poincaré map defined in a neighborhood
of Γ can be embedded in a flow of a one-dimensional analytic vector field.
(d) If the limit cycle Γ is hyperbolic, then the inverse integrating factor V is analytic
in Ω0 .

The proof of Theorem 2.8 requires knowledge of the existence of embedding

flows of diffeomorphisms and of normal forms of one-dimensional vector fields.
The definition of a normal form and the related background will be introduced in the
following sections and chapters.
A map y = F(x), x ∈ Ω0 ⊂ Rn , can be embedded in a flow ϕt (x) (or embedded
in a vector field X , its solutions are also denoted by ϕt (x)), if there exists a fixed
T > 0 such that ϕT (x) = F(x). We also say that ϕt (x) is an embedding flow of the
map and X is an embedding vector field of the map. Hereafter we call ϕT (x) the
time T map of the flow ϕt (x).
Proof of Theorem 2.8. The main ideas of the proof follow from Ensiso and
Peralta-Salas [139].
(a) Since the planar differential system (1.1) is analytic, the associated Poincaré map
P of the limit cycle Γ is analytic, and it has the Taylor expansion

P(s) = s + β0 sm + o(sm ),

with β0 = 0 and m ≥ 1. Note that if m = 1 then β0 > −1 and the limit cycle Γ is
hyperbolic; and that if m > 1 then the limit cycle Γ is of multiplicity m.
By Takens [420] and Yakovenko [450] the Poincaré map is C ∞ conjugate to the
time one map of the flow of the one-dimensional vector field

(sm + β1 s2m−1 )∂s .

2.2 Inverse Jacobian Multipliers and Their Vanishing Sets 51

Moreover, it follows from Yakovenko [450] that there exists a C ∞ diffeomorphism

Φ from Ω0 to Θ0 = (R, 0) × S1 under which the vector field X associated to system
(1.1) is transformed to the normal form vector field
 
X = Φ∗ X = Pm−1 (s) T −1 ∂θ + (sm + β1 s2m−1 )∂s ,

where T is the minimal positive period of Γ , Ω0 is the neighborhood of Γ defined

as before, and Pm−1 is a polynomial of degree m − 1 satisfying Pm−1 (0) = 1.
Clearly the vector field X has the inverse integrating factor
 
V := Pm−1 (s) sm + β1 s2m−1 .

Then
V := V DΦ −1 ◦ Φ

is an inverse integrating factor of X in a neighborhood of the limit cycle Γ , where

DΦ −1 represents the Jacobian of Φ −1 . This proves the statement.
(b) The proof follows from the expressions of V and V given in the proof of
statement (a).
(c) Necessity. Here we will use the notations given in the proof of Theorem 2.7.
Note that V(s, 0) = V (s, 1); we denote them by W  (s). By the assumption that V
  (s)
is analytic, it follows that V (s, θ ) is analytic for |s| sufficiently small. Hence W
is analytic and consequently it has s = 0 as a unique zero point in a suitably small
neighborhood of s = 0, say (−ε0 , ε0 ).
By (2.22) one has

P (s) 1
= , 0 < |s| < ε0 .
 (P(s))
W  (s)
W

Choosing any fixed but sufficiently small s0 = 0, and integrating this last equation
from P−1 (s0 ) to s0 gives

P(s0 )

dμ s0
P (s) s0
ds
= ds = =: σ0 . (2.24)
s0  (μ)
W P−1 (s0 )  (P(s))
W P−1 (s0 )  (s)
W

Clearly σ0 is a finite number and it depends only on s0 .

Consider the one-dimensional analytic differential equation

dρ  (ρ).
= σ0 W (2.25)
dt

Let ρt (s) be the solution of equation (2.25) satisfying the initial condition ρ0 (s) = s.
Then ρt (0) ≡ 0 for all t ∈ R, and ρt (s) = 0 for s = 0 small. Integrating equation
(2.25) along the solution ρt (s0 ) from 0 to t yields
52 2 Jacobian and Inverse Jacobian Multipliers

ρt (s0 )
dρ
= σ0 t.
s0 
W (ρ)

Comparing this last equality at t = 1 with (2.24) we readily obtain ρ1 (s0 ) = P(s0 ).
In addition, it is clear that ρ1 (0) = P(0). This proves that the Poincaré map P is the
time one map of the flow of the one-dimensional analytic equation (2.25).
Sufficiency. Assume that h(ζ )∂ζ is the one-dimensional analytic vector field and
ψt (ζ ) is its associated flow satisfying P(ζ ) = ψ1 (ζ ). Consider the vector field

Y := h(ζ )∂ζ + ∂τ , (ζ, τ ) ∈ (R, 0) × S1 .

Since the Poincaré maps of Y and X associated to (2.20) are equal, by [233, Lemma
8], which states that two real C (1 ≤ k ≤ ω) periodic differential systems of period
k

1 are C k equivalent if and only if their Poincaré maps are C k conjugate, it follows
that the vector fields Y and X are analytically equivalent. Note that the vector field
Y has the analytic inverse integrating factor V ∗ (ζ, τ ) = h(ζ ). Then the vector field
X has the analytic inverse integrating factor V(s, θ ) := V ∗ DH −1 ◦ H(s, θ ), where
(ζ, τ ) = H(s, θ ) is the analytic invertible change of coordinates which transforms
Y to X . Furthermore, it follows from the analytic equivalence of the vector fields
X and X that the vector field X has an analytic inverse integrating factor defined
in a neighborhood of the limit cycle Γ .
This proves the sufficiency and consequently the statement.
(d) Since the limit cycle Γ is hyperbolic, it follows that the Poincaré map is hyper-
bolic, i.e.
P(s) = v0 s + o(s),

with v0 = 1. Because of the analyticity of P(s) we get from Belitskii and Tkachenko
[30, Theorem 2.5] that P(s) is C ω conjugate to its linear part. Obviously, the linear
map s → v0 s is the time one map of the flow of the linear vector field ln v0 s∂s + ∂τ .
Hence P(s) can be embedded in a one-dimensional analytic vector field. Then the
proof follows from statement (c).
This completes the proof of the theorem. 

Remark If the limit cycle of an analytic differential system is not hyperbolic, there
may not exist an analytic inverse integrating factor in a neighborhood of the limit
cycle. Ensiso and Peralta-Salas [139, Example 2.8] constructed a planar analytic
differential system which has a semistable limit cycle, and it has no analytic inverse
integrating factor in any neighborhood of the limit cycle.

We now turn to the study of the existence and regularity of the inverse integrating
factors of planar differential systems in a neighborhood of elementary singularities.
Assume that q ∈ Ω is a singularity of the planar differential system (1.1), and λ1,2
are the two eigenvalues of the Jacobian matrix of the system at the singularity q.
• The singularity q is elementary if at least one of λ1,2 is not zero.
2.2 Inverse Jacobian Multipliers and Their Vanishing Sets 53

• The singularity q is nondegenerate if λ1,2 are both not zero.

• The singularity q is hyperbolic if the real parts of λ1,2 do not vanish.
• The singularity q is semihyperbolic if one of the real parts of λ1,2 is not zero and
the other is zero.
• The singularity q is a hyperbolic saddle if λ1,2 have different signs.
• The singularity q is a weak hyperbolic saddle if it is a hyperbolic saddle and
λ1 + λ2 = 0.
• The singularity q is a strong hyperbolic saddle if it is a hyperbolic saddle and
λ1 + λ2 = 0.
• The singularity q is a hyperbolic node if λ1,2 have the same sign.
• The singularity q is a center if there is a neighborhood of q which is filled up with
periodic orbits.
• The singularity q is a nondegenerate center if q is a center, and λ1,2 are a pair of
pure imaginary eigenvalues.
• The singularity q is a focus if all orbits in some neighborhood of q spirally approach
this singularity in either the positive or negative sense.
• The singularity q is a strong focus if q is a focus, and the real parts of λ1,2 are not
zeros.
• The singularity q is a weak focus if q is a focus, and λ1,2 are a pair of pure imaginary
eigenvalues.
• The singularity q is nilpotent if the Jacobian matrix of system (1.1) at q is not zero,
but its eigenvalues λ1,2 are both zero.
We remark that the topological structures of nilpotent singularities have been
characterized by Andreev [8], see also Zhang et al. [480].

Theorem 2.9 If q is an elementary singularity of the planar analytic differential

system (1.1), then the system has an inverse integrating factor, say V , in some neigh-
borhood of q. Moreover, the regularity of V can be characterized as follows:
(a) If q is a hyperbolic saddle or a semihyperbolic singularity, then V is C ∞ .
(b) If q is a hyperbolic node, or a nondegenerate center, or a strong focus, then V
is C ω .
(c) If q is a weak focus, then V is C ∞ . Furthermore, V is C ω if and only if the Poincaré
map of system (1.1) can be embedded in a flow of an analytic differential system
in a neighborhood of q.
(d) In all the above cases, the multiplicity of q as a zero of V is finite.

Recall that the inverse integrating factor V at the singularity q has

• multiplicity m (m ≥ 1) if the lowest order term in the Taylor expansion of V at q
is of degree m.
• infinite multiplicity if V at q is flat, in other words, the Taylor expansion of V at q
is identically zero.

Remark Theorem 2.9 and its proof were given in [139, Theorem 1.3] by Ensiso
and Peralta-Salas, in which the main tools are the normal forms of system (1.1)
54 2 Jacobian and Inverse Jacobian Multipliers

at singularities, see Ilyashenko and Yakovenko [213]. By computing the inverse

integrating factors of the normal form systems, the authors obtained the inverse
integrating factors of system (1.1) at the singularity through the relation between the
inverse integrating factors of two equivalent differential systems. We will not prove
Theorem 2.9, instead referring the reader to the original paper.

The next result, due to Giné and Peralta-Salas [186], characterizes the existence
of a C ∞ inverse integrating factor in a neighborhood of the center.

Theorem 2.10 The planar analytic differential system (1.1) with a (possibly degen-
erate) center at the origin O has a C ∞ inverse integrating factor V defined in a
neighborhood B of the origin, which is positive in B\{O} and is flat at O (i.e. V and
all its derivatives at O vanish).

García and Maza [167] studied the existence and regularity of inverse integrating
factors of planar analytic differential systems in a neighborhood of a simple mon-
odromy singularity. They applied a generalized polar blow up to system (1.1) at the
origin, then system (1.1) in a neighborhood of the origin is transformed to a system
defined on a cylinder {(r, θ ) ∈ (R, 0) × S1 }. Their results were stated in terms of
generalized polar coordinates.
It is still an open problem to characterize the existence and regularity of planar
analytic differential systems in a neighborhood of non-elementary singularities.
The above results have been extended to polycycles. Recall that a polycycle of a
planar differential system (1.1) consists of finitely many regular orbits and singular-
ities, which form a set Γ , i.e. Γ = {ψi (t)}ki=1 ∪ {qi }ki=1 with the ψi ’s regular orbits
and the qi ’s singularities of the system, which satisfy
• limt→∞ ψi (t) = qi+1 , limt→−∞ ψi (t) = qi , i = 1, . . . , k, where qk+1 = q1 ; and
• Γ has an inner (or outer) neighborhood either filled up with periodic orbits or in
which any orbit has Γ as its ω or α limit set.
In particular, Γ is called a homoclinic cycle if k = 1; or a heteroclinic cycle if k = 2.
A polycycle Γ is compact if all singularities on Γ are located in the finite plane.
Ensiso and Peralta-Salas [139, Theorem 1.4] extended Theorem 2.6 to polycycles,
and obtained the next result.

Theorem 2.11 Assume that the smooth differential system (1.1) has a compact poly-
cycle Γ ⊂ Ω, which is a limit set of some orbits located either in its inner or outer
neighborhood. If system (1.1) has a C 1 inverse integrating factor V in Ω, then
Γ ⊂ V −1 (0).

Ensiso and Peralta-Salas [139] proved the above theorem by using the fact that the
universal covering manifold of a manifold is simply connected, and so a closed dif-
ferential one-form on it is exact. For the definition of the universal covering manifold,
see e.g. Sharpe [392].

Remark Berrone and Giacomini [31] proved in 2000 that if a planar smooth differ-
ential system (1.1) has an inverse integrating factor V defined in a neighborhood of a
2.2 Inverse Jacobian Multipliers and Their Vanishing Sets 55

hyperbolic saddle p0 satisfying V (p0 ) = 0, then V vanishes on all four separatrices

of p0 . This result does not hold in general for nonhyperbolic singularities, see e.g.
[168].
Next we discuss the regularity of the inverse integrating factors in a neighborhood
of a polycycle. Assume that the eigenvalues of the planar differential system (1.1) at
a singularity S are λ and μ.
• If r := −λ/μ = q/p is a positive rational number, we call S a q : p resonant
saddle.
García et al. [160, Proposition 11] in 2010 proved the nonexistence of analytic
inverse integrating factors in a neighborhood of a homoclinic cycle.
Proposition 2.6 Assume that the planar analytic differential system (1.1) has a
compact homoclinic cycle Γ with the singularity a strong resonant saddle. If system
(1.1) is not orbitally linearizable (formal or analytic) in some neighborhood of
the singularity, then it cannot have an analytic inverse integrating factor in any
neighborhood of Γ .
García et al. [160, p. 3603] verified that the system

3
ẋ = −x + 2y + x 2 , ẏ = 2x − y − 3x 2 + xy
2
satisfies all conditions of Proposition 2.6, and so it has no analytic inverse integrating
factor defined in a neighborhood of the homoclinic loop, which is contained in the
invariant algebraic curve x 2 (1 − x) − y2 = 0, and is homoclinic to the origin.
In the proof of Proposition 2.6 the authors made use of the normal form of system
(1.1) at a resonant saddle and the expression of the inverse integrating factors at the
saddle. For the reader’s convenience, to understand this last proposition, we briefly
introduce some fundamental results on normal forms. The details will be given in
Chap. 7.
A formal series is a series of the form
∞

h(x) = Cx + hk (x), (2.26)
k=2

where hk (x) is a vector-valued homogeneous polynomial of degree k. The formal

series (2.26) is (locally) invertible if C is an invertible matrix. A map y = h(x)
is an invertible formal transformation if h(x) is an invertible formal series. If C is
the unit matrix, we call y = h(x) a near identity formal transformation or a formal
transformation tangent to the identity. If the formal series h(x) is convergent, then
y = h(x) is an invertible analytic transformation.
The differential systems (1.1) and (1.6) are
• formally equivalent if there exists an invertible formal transformation y = h(x)
such that
56 2 Jacobian and Inverse Jacobian Multipliers

∂x h(x)f (x) = g ◦ h(x).

• formally orbitally equivalent if system (1.1) is formally equivalent to a system of

the form ẏ = a(y)g(y), where a(y) is an invertible formal series.
The differential system (1.1) is
• formally (orbitally) linearizable if system (1.1) is formally (orbitally) equivalent
to a linear system.
Let q be a singularity of system (1.1). Without loss of generality, let q be the
origin, and assume system (1.1) has the form
∞

ẋ = Ax + fk (x), (2.27)
k=2

where the fk ’s are vector-valued homogeneous polynomials of degree k. We call

system (2.27) a formal differential system. If the series in (2.27) is convergent, then
the system is analytic.
Let λ1 , . . . , λn be the eigenvalues of the matrix A.
• The eigenvalues of A satisfy a resonant relation if there exists some j ∈ {1, . . . , n}
and k = (k1 , . . . , kn ) ∈ Zn+ , |k| ≥ 2, such that

λj = k, λ,

where |k| = k1 +· · ·+kn . Recall that Z+ = N∪{0} and k, λ = k1 λ1 +· · ·+kn λn .

• Otherwise the eigenvalues of A do not satisfy any resonant relation.
Example Let A be a real or complex matrix of order 5 with eigenvalues

0, 2, −2, 4 + π i, 4 − π i.

Then the resonant relations that the eigenvalues of A satisfy are

0 = (k3 − 4k4 )2 + k3 (−2) + k4 (4 + π i) + k4 (4 − π i),

with k3 ∈ N, k4 ∈ Z+ , k3 ≥ 4k4 ; and

2 = (k3 − 4k4 + 1)2 + k3 (−2) + k4 (4 + π i) + k4 (4 − π i),

with k3 ∈ N, k4 ∈ Z+ , k3 ≥ 4k4 − 1.
Theorem 2.12 (Poincaré linearization theorem) If the eigenvalues of A in (2.27) do
not satisfy any resonant relation, then system (2.27) is linearizable.
For m = (m1 , . . . , mn ) ∈ Zn+ , m ≥ 2, set x m = x1m1 . . . xnmn . Let ej be the n-
dimensional unit vector with its jth entry equal to 1 and the others all vanishing. We
2.2 Inverse Jacobian Multipliers and Their Vanishing Sets 57

call ej the jth unit vector. A nonlinear monomial x m ej in system (2.27) is resonant
if λj = m, λ. System (2.27) is in the Poincaré normal form (or simply normal
form) if its linear part is in the Jordan normal form, and all nonlinear monomials are
resonant.
If the nonlinear term ψ(x) of a near identity transformation y = x + ψ(x) send-
ing system (2.27) to its Poincaré normal form consists of nonresonant monomials,
then the transformation is called a distinguished normalization. Correspondingly the
normal form system is a distinguished normal form. A monomial, say ym , of ψ(y)
is nonresonant if m, λ = 0; resonant if m, λ = 0. Readers should take care of
the difference between the resonances of the monomials in the systems and in the
transformations.

Example Set A = diag(0, 2, −2, 4 + π i, 4 − π i). If an analytic or a formal differ-

ential system
ẋ = Ax + f (x)

is in the Poincaré normal form, then its general form is

⎛ k1 k3 −4k4 k3 k4 k4 ⎞
k3 ∈N, k1 , k4 ∈Z+ , k3 ≥4k4 ak1 ,k3 ,k4 y1 y2 y3 y4 y5

ẏ = Ay + ⎝ k3 ∈N, k1 , k4 ∈Z+ , k3 ≥4k4 −1 bk1 ,k3 ,k4 y1k1 y2k3 −4k4 +1 y3k3 y4k4 y5k4 ⎠ ,
0

where 0 is the 3-dimensional zero vector.

Theorem 2.13 (Poincaré–Dulac normal form theorem) An analytic or a formal

differential system (2.27) is always formally equivalent to its Poincaré normal form
via a near identity formal transformation.

The proof of the Poincaré linearization theorem and the Poincaré–Dulac normal
form theorem can be found in [213, 230]. Clearly, the Poincaré linearization the-
orem is a special case of the Poincaré–Dulac normal form theorem, because if the
eigenvalues of the linear part of system (2.27) do not satisfy any resonant relation,
then its Poincaré–Dulac normal form is linear.
Problem: is an analytic differential system (2.27) analytically equivalent to its
Poincaré normal form? This is a classical problem, and there are extensive studies
on it, see e.g. [30, 33, 42, 83, 91, 92, 97, 132, 134, 138, 192, 195, 197, 198, 213,
230, 232, 339, 344, 367, 390, 400, 401, 408, 409, 412, 413, 420, 444, 466] and the
references therein. We will not discuss this problem in the general case. In Chap. 7
we will restrict our study to the analytic integrable (or partial integrable) differential
systems.
We go back to the inverse integrating factors. Proposition 2.6 provided a sufficient
condition under which an analytic differential system does not have an analytic
inverse integrating factor defined in a neighborhood of a homoclinic cycle. The
following result provides a link between two analytic inverse integrating factors
defined in a neighborhood of a polycycle. See García et al. [160, Theorem 6].
58 2 Jacobian and Inverse Jacobian Multipliers

Proposition 2.7 Assume that system (1.1) is analytic and has a compact polycycle
Γ . If all orbits located in a unilateral neighborhood of Γ have Γ as either their α
or ω limit set, then in a sufficiently small neighborhood of Γ the ratio of any two
analytic inverse integrating factors is a constant.
Similar to Theorem 2.7, García et al. [160, Theorem 6] studied the cyclicity of a
homoclinic loop with a hyperbolic saddle on it in terms of the vanishing multiplicity
of an inverse integrating factor. Recall that the cyclicity of a homoclinic loop Γ of
system (1.1) is the maximal number of limit cycles which can bifurcate from it under
the prescribed class of perturbations of system (1.1). We note that there are extensive
studies on the cyclicity of the homoclinic and the heteroclinic loops of a polynomial
differential system under polynomial perturbations, see e.g. [378].
Let Γ be a homoclinic loop consisting of a regular orbit ϕ(t) and a weak hyperbolic
saddle p0 . Without loss of generality we can assume that p0 is at the origin. After
an invertible real linear change of variables and a time rescaling (if necessary), by
applying the Poincaré–Dulac normal form theory we get a near identity formal change
of coordinates which brings system (1.1) to
   
k−1
   
ẋ x ak x
= 1+ i i
ci x y + x k yk + h.o.t., (2.28)
ẏ −y −bk y
i=1

where h.o.t. denotes the higher-order terms. If ak = bk , we call dk+1 = ak − bk

the first nonvanishing saddle quantity. In this case we can take a sufficiently higher-
order cut off of the formal transformation under which system (1.1) is analytically
equivalent to (2.28). If there does not exist a k such that ak = bk , it follows from
Zhang [466] that system (1.1) is analytically orbitally linearizable. Similar to the
vanishing multiplicity of an inverse integrating factor over a limit cycle, we can
define the vanishing multiplicity of an inverse integrating factor over a homoclinic
loop.
Theorem 2.14 Let Γ be a compact homoclinic loop of the planar analytic sys-
tem (1.1), which is homoclinic to a weak hyperbolic saddle, and whose associated
Poincaré map is not the identity. Assume that system (1.1) has an analytic inverse
integrating factor defined on a neighborhood of Γ , whose vanishing multiplicity over
Γ is m. The following statements hold.
(a) m ≥ 1 and the first possible nonvanishing saddle quantity is dm .
(b) If dm = 0, the cyclicity of Γ is 2m − 1.
(c) If dm = 0, the cyclicity of Γ is 2m.
We will not prove this theorem. We refer the reader to its original proof given in
[160, Theorem 6]. These last results also show the importance of inverse integrating
factors in the study of the dynamics of planar differential systems.

Remark Previously we have studied the existence and the regularity of the inverse
integrating factors of a planar analytic differential system in a neighborhood of
2.2 Inverse Jacobian Multipliers and Their Vanishing Sets 59

a limit cycle, or of an elementary singularity. But it is still an open problem to

characterize the existence and the regularity of a planar analytic differential system
in a neighborhood of a polycycle, or of a homoclinic loop, or of a heteroclinic loop.

2.3 Inverse Jacobian Multipliers and the Center-Focus

Problem

In this section we will study the existence of analytic or C ∞ inverse Jacobian mul-
tipliers of an n-dimensional analytic differential system at a singularity having a
two-dimensional center manifold with a pair of pure imaginary eigenvalues.
In this direction the first result is the classical Poincaré center theorem, which
reveals the essential character of the center-focus problem for planar analytic dif-
ferential systems. Recall that the center-focus problem is to determine whether a
singularity of a planar nonlinear differential system with a pair of pure imaginary
eigenvalues is a center or a focus.

Theorem 2.15 (Poincaré center theorem) Assume that a planar analytic differential
system has a singularity and its linear part at the singularity has a pair of pure
imaginary eigenvalues. Then the singularity is a center if and only if the system has
an analytic first integral in a neighborhood of the origin.

The proof of Theorem 2.15 will be obtained as a consequence of Theorem 7.18,

which will be proved later on. So we omit its proof here.
Lyapunov extended the Poincaré center theorem from two-dimensional differen-
tial systems to higher-dimensional systems. For the real analytic differential system
in Rn

ẋ = −y + f1 (x, y, z) = F1 (x, y, z),

ẏ = x + f2 (x, y, z) = F2 (x, y, z), (2.29)
ż = Az + f (x, y, z) = F(x, y, z),

where A is an (n − 2) × (n − 2) real matrix, and w = (w3 , . . . , wn )τ , w ∈ {z, f , F},

with τ the transpose of a matrix, we denote by

∂ ∂  n
∂
X = F1 (x, y, z) + F2 (x, y, z) + Fj (x, y, z)
∂x ∂y j=3
∂z j

its associated vector field. Set f := (f1 , f2 , f ) = O(|(x, y, z)|2 ).

Recall that a set I ⊂ Ω is an r-dimensional invariant manifold of system (1.1) if
it is invariant under the flow of system (1.1) and it is an r-dimensional manifold. For
r ∈ {∞, ω}, it is well known that if a C r differential system (1.1) has a singularity
q, it can be written, after a translation (if necessary), as
60 2 Jacobian and Inverse Jacobian Multipliers

ẋ = Lx + p(x), (2.30)

where L is the Jacobian matrix of f (x) at q and p(x) = O(|x|2 ) is an n-dimensional

C r vector-valued function.
Let s, u and c be the numbers of eigenvalues of L which have negative, positive
and zero real parts, respectively. Then system (2.30) can be written, after an invertible
linear transformation (if necessary), as

ẏ = Sy + g1 (y, z, w), ż = Uz + g2 (y, z, w), ẇ = Cw + g3 (y, z, w), (2.31)

where y, z and w are s, u and c-dimensional coordinates respectively, S, U and C

are square matrices of orders s, u and c respectively whose eigenvalues have only
negative, positive and zero real parts, and g1 , g2 , g3 = O(|(y, z, w)|2 ) are C r smooth,
r ∈ {∞, ω}. Clearly the linear differential system

ẏ = Sy, ż = Uz, ẇ = Cw (2.32)

respectively has the invariant hyperplanes Hs := {z = w = 0}, Hu := {y = w = 0}

and Hc := {y = z = 0}, which are respectively called the stable, unstable and
center hyperplanes of the linear differential system (2.32). Of course, these invariant
hyperplanes are s, u and c-dimensional manifolds, respectively.
The stable (resp. unstable) manifold theorem shows that for r ∈ {∞, ω}, the
C r differential system (2.31) has a unique s-dimensional (resp. u-dimensional) C r
invariant manifold W s (resp. W u ) tangent to the stable (resp. unstable) hyperplane
Hs (resp. Hu ), which is called the stable manifold (resp. unstable manifold) of sys-
tem (2.31) at the origin. The orbits starting on W s (resp. W u ) will positively (resp.
negatively) approach the origin.
The center manifold theorem states that for r ∈ {∞, ω}, the C r differential system
(2.31) has a c-dimensional C k invariant manifold W c tangent to the center hyperplane
Hc for any k ∈ N, which is called the center manifold at the origin. It is well known
that the center manifold of a C ω differential system (if it exists) may not be C ω and
even not be C ∞ . In general, center manifolds at a singularity may not be unique. For
more details, see e.g. Chicone [89, Chap. 4].
For system (2.29), to fix our objective we assume that A has no eigenvalues with
vanishing real parts. Under this hypothesis we get from the center manifold theorem
that system (2.29) has a two-dimensional center manifold, which is tangent to the
(x, y) plane at the origin. Moreover, this center manifold has the expression


n
Mc = {zi = ci (x, y)}, (2.33)
i=3

where the ci (x, y)’s, i = 3, . . . , n, are C r functions for some r ∈ N.

Now we can state the Lyapunov center theorem.
2.3 Inverse Jacobian Multipliers and the Center-Focus Problem 61

Theorem 2.16 (Lyapunov center theorem) Assume that A has no eigenvalues with
vanishing real parts. The following statements hold.
(a) System (2.29) restricted to M c has a center at the origin if and only if it admits
a real analytic local first integral of the form

Φ(x, y, z) = x 2 + y2 + higher-order terms,

which is defined in a neighborhood of the origin in Rn .

(b) If statement (a) holds, then the center manifold is unique and analytic.

For a proof of the Lyapunov center theorem, we refer the reader to Bibikov [33,
Theorems 3.1, 3.2 and Sect. 5] and Sijbrand [402].
In the next subsection we will show how to use the inverse integrating factors or
the inverse Jacobian multipliers to characterize the center-focus problem of a two or
higher-dimensional analytic differential system when restricted to a two-dimensional
center manifold.

2.3.1 The Center-Focus Problem via Inverse Integrating

Factors or Inverse Jacobian Multipliers

Making use of the inverse integrating factors, this subsection provides a character-
ization of when a singularity of a two-dimensional differential system with a pair
of pure imaginary eigenvalues is a center. For higher-dimensional systems having a
singularity with a pair of pure imaginary eigenvalues and a two-dimensional center
manifold, we prove the existence and the regularity of the inverse Jacobian multi-
pliers of the systems at the singularity, and we also characterize the singularity to
be a center or a focus when restricted to the center manifold via inverse Jacobian
multipliers.
In this direction Reeb [370] conducted pioneering work in 1952. He provided a
characterization of the center-focus problem via the inverse integrating factors.

Theorem 2.17 (Reeb center theorem) Assume that a real planar analytic differen-
tial system has a singularity with a pair of pure imaginary eigenvalues. Then the
singularity is a center if and only if the system has a real analytic and nonvanishing
inverse integrating factor in a neighborhood of the singularity.

One can prove this theorem by making use of the Poincaré center theorem. The
details are left to the reader as an exercise.
The Reeb center theorem was extended to three-dimensional differential systems
by Buică et al. [49] in 2012. Several new ideas were involved in the proofs of their
results.
62 2 Jacobian and Inverse Jacobian Multipliers

Theorem 2.18 (Buică, García and Maza center-focus theorem) Assume that system
(2.29) is a 3-dimensional real analytic differential system, and that A is a nonzero
real number. The following statements hold.
(a) System (2.29) restricted to M c has a center at the origin if and only if it admits
an analytic inverse Jacobian multiplier of the form

J(x, y, z) = z + higher-order terms,

which is defined in a neighborhood of the origin. Furthermore, if such an inverse

Jacobian multiplier exists, the center manifold is unique and analytic, and it is
contained in J −1 (0).
(b) If system (2.29) restricted to M c has a focus at the origin, then there exists a
C ∞ and nonflat inverse Jacobian multiplier of the form

J(x, y, z) = z(x 2 + y2 )k + higher-order terms, k ≥ 2,

which is defined in a neighborhood of the origin. Furthermore, there exists a

local C ∞ center manifold M such that M ⊂ J −1 (0).

Recall that a C ∞ function is nonflat at a point if its Taylor expansion at this point
is not identically zero.
We will not prove Theorem 2.18, which can be obtained from the proof of the
next Theorems 2.19 and 2.20. These two theorems describe the further extension of
Theorem 2.18 from three-dimensional differential systems to any finite-dimensional
system (2.29) by Zhang [475], where some ideas and techniques different from those
of [49, 50] are involved.
To state the results, we need to introduce some notation. Let λ3 , . . . , λn be the
eigenvalues of the matrix A. Then the √linear part of system (2.29) has the eigenvalues
λ = (i, −i, λ3 , . . . , λn ), where i = −1. For  = (1 , . . . , n ) ∈ Zn , we have


n
, λ = 1 i − 2 i + j λj .
j=3

Set  
R :=  ∈ Zn | , λ = 0,  + ej ∈ Zn+ , j = 3, . . . , n ,

where ej is the jth unit vector of Rn . Assume that

(H) R is one-dimensional and A is diagonalizable in C.
We remark that the assumption (H) implies that Re λj = 0 for j = 3, . . . , n.
So it follows from the center manifold theorem that system (2.29) has a center
manifold tangent to the (x, y) plane at the origin, which can be expressed in the form
(2.33). If the eigenvalues of A either all have positive real parts or all have negative
real parts, then R admits a unique Z+ -linearly independent element with (1, 1, 0)
2.3 Inverse Jacobian Multipliers and the Center-Focus Problem 63

as its generator, where 0 is the (n − 2)-dimensional zero vector. Note that in the
three-dimensional case, if A is a nonzero real number, the assumption (H) holds
automatically.
If A has complex eigenvalues, since A is real, its complex eigenvalues appear in
conjugate pairs. For convenience, when stating our next results we assume without
loss of generality that λ3+2j and λ3+2j+1 , j = 0, 1, . . . , m − 1, are the conjugate
complex eigenvalues of A, where m ∈ Z+ with 2m ≤ n − 2. Obviously if A has no
complex eigenvalues then m = 0.
The next result, due to Zhang [475], characterizes the center of system (2.29) at
the origin on the center manifold M c via inverse Jacobian multipliers.
Theorem 2.19 Assume that the differential system (2.29) is analytic and satisfies
the assumption (H). Let λ3 , . . . , λn be the eigenvalues of A, which satisfy either
Reλj > 0 for all j = 3, . . . , n or Reλj < 0 for all j = 3, . . . , n, and λ3+2s+1 = λ3+2s ,
s = 0, 1, . . . , m − 1, with m ∈ Z+ and 2m ≤ n − 2, where the bar denotes the
conjugate of a complex number. The following statements hold.

(a) System (2.29) restricted to the center manifold M c has a center at the origin
if and only if the system has a local analytic inverse Jacobian multiplier of the
form


m−1
 
J(x, y, z) = (z3+2j − p3+2j (x, y, z))2 + (z3+2j+1 − p3+2j+1 (x, y, z))2
j=0

n
× (zl − pl (x, y, z))V (x, y, z), if m > 0, (2.34)
l=3+2m

or


n
J(x, y, z) = (zl − pl (x, y, z))V (x, y, z), if m = 0,
l=3

in a neighborhood of the origin, where pj = O(|(x, y, z)|2 ) for j = 3, . . . , n, and

V (0, 0, 0) = 1.
(b) If system (2.29) has the inverse Jacobian multiplier in statement (a), then the
center manifold M c is unique and analytic, and M c ⊂ J −1 (0).

We note that the set of matrices satisfying (H) forms a full Lebesgue measure
subset of the set of real matrices of order n. The assumption (H) provides only a
sufficient condition. If (H) does not hold, it is still an open problem to determine
whether the origin restricted to the center manifold is a center or a focus via inverse
Jacobian multipliers.
When system (2.29) restricted to the center manifold has a focus at the origin,
the following result due to Zhang [475] verifies the existence of a C ∞ smooth local
inverse Jacobian multiplier around the singularity.
64 2 Jacobian and Inverse Jacobian Multipliers

Theorem 2.20 Assume that the differential system (2.29) is analytic and satisfies
the assumption (H). The following statements hold.
(a) If system (2.29) restricted to M c has a focus at the origin, then the system has
a local C ∞ inverse Jacobian multiplier of the form


m−1 
J(x, y, z) = (z3+2j − p3+2j (x, y, z))2 + (z3+2j+1 − p3+2j+1 (x, y, z))2
j=0

n  l
× (zl − pl (x, y, z)) (x − q1 (x, y, z))2 + (y − q2 (x, y, z))2 (2.35)
l=3+2m

× h (x − q1 (x, y, z))2 + (y − q2 (x, y, z))2 V (x, y, z), if m > 0,

or


n  l
J(x, y, z) = (zl − pl (x, y, z)) (x − q1 (x, y, z))2 + (y − q2 (x, y, z))2
l=3

× h (x − q1 (x, y, z))2 + (y − q2 (x, y, z))2 V (x, y, z), if m = 0,

in a neighborhood of the origin, with l ≥ 2, pj , qi = O(|(x, y, z)|2 ), and h(0) =

V (0, 0, 0) = 1.
(b) There exists a local C ∞ center manifold M such that M ⊂ J −1 (0).

The number l in Theorem 2.20 is called the vanishing multiplicity of the inverse
Jacobian multiplier at the origin.

2.3.1.1 Preparation for the Proof of Theorems 2.19 and 2.20

To prove Theorems 2.19 and 2.20 we need some technical results, which can also
be used to study some related problems. To simplify the notation we will replace the
real coordinates by the corresponding conjugate complex coordinates. The details
are as follows.
For the (n − 2)-tuple of eigenvalues λ3 , . . . , λn of A in the Jordan normal form,
if λj , λj+1 are two complex conjugate eigenvalues, and its associated coordinates
are zj , zj+1 , we choose a pair of conjugate complex coordinates
√ ζj = zj + izj+1
and ζj+1 = zj − izj+1 instead of zj and zj+1 , where i = −1. Moreover, we set
ξ = x + iy, η = x − iy. By the assumption (H) we can write system (2.29) in these
new coordinates as

ξ̇ = −iξ + p1 (ξ, η, ζ ) = P1 (ξ, η, ζ ),

η̇ = iη + p2 (ξ, η, ζ ) = P2 (ξ, η, ζ ), (2.36)
ζ̇ = Λζ + p(ξ, η, ζ ) = P(ξ, η, ζ ),
2.3 Inverse Jacobian Multipliers and the Center-Focus Problem 65

with Λ = diag(λ3 , . . . , λn ), where we have used the fact that A is in the real Jordan
normal form, where ζj = zj for λj real, j ∈ {3, . . . , n}. We note that these new
coordinates can greatly simplify the notation in the following calculations. Denote
by X the vector field associated to system (2.36).
Writing Proposition 2.3 according to system (2.29) gives the next result.

Lemma 2.1 Let J be an inverse Jacobian multiplier of system (2.29). Then an invert-
ible smooth change of coordinates (x, y, z) = Ψ (u, v, w) sends system (2.29) to

ẇ = (∂Ψ (w))−1 F ◦ Ψ (w),

with F = (F1 , F2 , F)τ and w = (u, v, w)τ , which has the inverse Jacobian multiplier


J(w) = J(Ψ (w))(DΨ (w))−1 ,

where ∂Ψ (w) and DΨ (w) denote respectively the Jacobian matrix and the Jacobian
of Ψ .

Applying the Poincaré–Dulac normal form theorem to system (2.36), we get the
next result.

Lemma 2.2 Assume that (H) holds. System (2.36) is formally equivalent to

u̇ = −u(i + ρ1 (uv)),
v̇ = v(i + ρ2 (uv)), (2.37)
ẇj = wj (λj + ρj (uv)), j = 3, . . . , n,

via a distinguished normalization, where the ρi ’s are analytic functions or formal

series without constant terms.

Lemma 2.2 assures only the existence of the formal normalization of system (2.36)
to its distinguished normal form. What about the regularity of the normalization? The
next results provide some sufficient conditions.

Lemma 2.3 Under the assumption (H), for the distinguished normalization trans-
forming system (2.36) to its Poincaré–Dulac normal form (2.37) the following state-
ments hold.
(a) If system (2.36) restricted to the center manifold M c has a focus at the origin,
then the distinguished normalization is C ∞ .
(b) If system (2.36) restricted to M c has a center at the origin, and the eigenval-
ues of A have either all positive real parts or all negative real parts, then the
distinguished normalization is analytic.

Proof (a) In (2.37) the coordinates u and v are conjugate, which forces ρ2 = ρ 1 .
By the assumption that the origin of system (2.37) restricted w = 0 is a focus, we
must have Re ρ1 = 0. Otherwise, the singularity on the two-dimensional manifold
66 2 Jacobian and Inverse Jacobian Multipliers

will be a center, a contradiction. So by Belitskii [29, Theorem 1] it follows that the

distinguished normalization transforming system (2.36) to its normal form (2.37)
is C ∞ .
(b) According to [33] we write system (2.36) as
       
ξ̇ ξ p1 (ξ, η, ζ ) −i 0
= A1 + , A1 = ,
η̇ η p2 (ξ, η, ζ ) 0 i
ζ̇ = A2 ζ + p(ξ, η, ζ ), A2 = Λ.

By the assumption (H) it follows that the eigenvalues of A1 and A2 are not resonant
with each other, i.e.

λj = 1 (−i) + 2 i = (2 − 1 )i, 1 , 2 ∈ Z+ for j = 3, . . . , n.

So we get from Bibikov [33, Theorem 10.1] that there exists a distinguished normal-
ization
ξ = u + ϕ1 (u, v), η = v + ϕ2 (u, v), ζ = r + ϕ(u, v),

with r = (r3 , . . . , rn ), ϕ = (ϕ3 , . . . , ϕn ) and ϕ1 , ϕ2 , ϕ = O(|(u, v, r)|2 ), which sends

system (2.36) formally to

u̇ = −u(i + ρ1 (uv)), v̇ = v(i + ρ2 (uv)), ṙ = Λr + h(u, v, r), (2.38)

with ρ1 , ρ2 = o(1), h = (h3 , . . . , hn )τ = O(|(u, v, r)|2 ) and h(u, v, 0) = 0. Accord-

ing to Bibikov [33], system (2.38) is called a quasi-normal form of system (2.36).
Since system (2.38) restricted to the center manifold r = 0 has a center at the
origin, it has a formal first integral. Recall that a formal first integral of an analytic
or a formal differential system (1.1) is a formal series which satisfies the Eq. (1.2).
Then we get from Zhang [466] that ρ1 (uv) = ρ2 (uv) in (2.38). Applying Bibikov
[33, Theorems 10.2, 3.2 and Sect. 5] shows that the distinguished normalization
from system (2.36) to (2.38) is convergent. Hence systems (2.36) and (2.38) are
analytically equivalent.
Our next aim is to find an analytic change of coordinates which sends system
(2.38) to system (2.37). Let

u = u, v = v, r = w + ψ(u, v, w)

be such a transformation. Some calculations show that

∂w ψΛw − iu∂u ψ + iv∂v ψ − Λψ (2.39)

= Λh(u, v, w + ψ(u, v, w)) − ∂w ψwρ + uρ1 ∂u ψ − vρ2 ∂v ψ,

where wρ = (w3 ρ3 , . . . , wn ρn )τ , ψ is a column vector and ∂w ψ is the Jacobian matrix

of ψ with respect to w. In the linear space H l+p+q formed by (n − 2)-dimensional
2.3 Inverse Jacobian Multipliers and the Center-Focus Problem 67

vector-valued homogeneous polynomials of degree l in w and of degrees p and q in

u and v, respectively, the linear operator

∂ ∂ ∂
L= Λw − iu + iv − Λ
∂w ∂u ∂v
has the spectrum
 
k, λ − pi + qi − λj | k ∈ Z+
n−2
, |k| = l, p, q ∈ Z+ , j = 3, . . . , n ,

see Bibikov [33] for more details.

Expanding ψ, ρ1 , ρ2 , ρ and h as Taylor series and using the assumption (H), we
can prove by induction that Eq. (2.39) have a formal series solution ψ consisting of
nonresonant monomials. Furthermore, the assumption that the eigenvalues of A have
their real parts either all positive or all negative guarantees that there exists a number
σ > 0 such that if k, λ − pi + qi − λj = 0 for (k, p, q) ∈ Zn+ then
 
k, λ − pi + qi − λj  ≥ σ.

Using the techniques in the proof of the classical Poincaré–Dulac normal form the-
orem (see e.g. Bibikov [33] or Zhang [466]), we can prove the convergence of ψ.
We complete the proof of statement (b), and consequently the lemma. 

The next result verifies the existence of analytic integrating factors of system
(2.36) restricted to the center manifold.

Lemma 2.4 Assume that system (2.36) has an analytic inverse Jacobian multiplier
of the form

J(ξ, η, ζ ) = (ζ3 − ψ3 (ξ, η, ζ )) . . . (ζn − ψn (ξ, η, ζ ))V (ξ, η, ζ ),

where ψs = O(|(ξ, η, ζ )|2 ) for s ∈ {3, . . . , n}, V is analytic and satisfies V (0, 0, 0) =
0. The following statements hold.

n
(a) M = {ζs = ψs (ξ, η, ζ )} is an invariant analytic center manifold of the
s=3
 in a neighborhood of the origin.
vector field X
|M .
(b) V |M is an analytic inverse integrating factor of X

Proof (a) Since J is an inverse Jacobian multiplier of 

X, we have

(J) = JdivX
X .
68 2 Jacobian and Inverse Jacobian Multipliers

This implies that there exist analytic functions

K0 (ξ, η, ζ ), K3 (ξ, η, ζ ), . . . , Kn (ξ, η, ζ )

such that
(ζs − ψs (ξ, η, ζ )) = Ks (ξ, η, ζ )(ζs − ψs (ξ, η, ζ )), s = 3, . . . , n,
X
(V (ξ, η, ζ )) = K0 (ξ, η, ζ )V (ξ, η, ζ ). (2.40)
X

This shows that each hypersurface Ss := {ζs = ψs (ξ, η, ζ )}, s = 3, . . . , n, is

invariant under the flow of X .
In addition, the invariant hypersurface Ss can be expressed via the Implicit Func-
tion Theorem as
ζs = γs (ξ, η), s = 3, . . . , n,

in a neighborhood of the origin, where γs (ξ, η) is analytic, and it satisfies γs (0, 0) = 0

and ∂ξ γs (0, 0) = ∂η γs (0, 0) = 0 for s = 3, . . . , n. This proves that


n
M = {ζs = γs (ξ, η)},
s=3

is an analytic center manifold of X  in a neighborhood of the origin, and is tangent

to the (ξ, η) plane at the origin.
(b) Set γ (ξ, η) = (γ3 (ξ, η), . . . , γn (ξ, η)) with γs (ξ, η) obtained in the proof of (a)
for s = 3, . . . , n. From the first equation of (2.40) we have

(ζs − ψs (ξ, η, ζ )) = 0
X on M , s = 3, . . . , n.

These n − 2 equations can be written in vector form

 
E − ∂ζ ψ P = P1 ∂ξ ψ + P2 ∂η ψ on M , (2.41)

where ∂σ ψ = (∂σ ψ3 , . . . , ∂σ ψn )τ , σ ∈ {ξ, η}, and ∂ζ ψ is the Jacobian matrix of

ψ = (ψ3 , . . . , ψn ) with respect to ζ . Moreover, it follows from the construction of
γ (ξ, η) that
γ (ξ, η) = ψ(ξ, η, γ (ξ, η)).

This implies    
E − ∂ζ ψ ∂ξ γ = ∂ξ ψ, E − ∂ζ ψ ∂η γ = ∂η ψ, (2.42)

where ∂σ γ = (∂σ γ3 , . . . , ∂σ γn )τ , σ ∈ {ξ, η}.

Set I(ξ, η) = V (ξ, η, γ (ξ, η)). Some calculations show that
2.3 Inverse Jacobian Multipliers and the Center-Focus Problem 69

|M (I(ξ, η)) = P1 (ξ, η, γ (ξ, η))∂ξ I + P2 (ξ, η, γ (ξ, η))∂η I

X (2.43)

= P1 [w]∂ξ V + P2 [w]∂η V + ∂ζ V P[w]M = X (V )|M = K0 V |M = K0 |M I,

where [w] = (ξ, η, γ (ξ, η)), and we have used (2.42) and (2.41) in the proof.
Finally, we only need to prove K0 |M = div(X |M ). First, from (2.40) and the
expression of J we have

= X
JdivX (J) = (K0 + K3 + · · · + Kn )J.

This implies
− K3 − · · · − Kn .
K0 = divX (2.44)

Second, from (2.40) and the fact that

(ζs − ψs (ξ, η, ζ )) = Ps − P1 ∂ξ ψs − P2 ∂η ψs − ∂ζ ψs P, s = 3, . . . , n,
X

we have
 
diag(K3 , . . . , Kn ) (ζ − ψ(ξ, η, ζ )) = E − ∂ζ ψ P − P1 ∂ξ ψ − P2 ∂η ψ. (2.45)

Differentiating (2.45) with respect to ζ , and writing the resulting equations in vector
form, we get
   −1
diag(K3 , . . . , Kn ) = E − ∂ζ ψ ∂ζ P E − ∂ζ ψ
 −1  −1
− ∂ξ ψ ∂ζ P1 E − ∂ζ ψ − ∂η ψ ∂ζ P2 E − ∂ζ ψ , (2.46)

where we have used the fact that

∂ξ ∂ζs ψj = ∂η ∂ζs ψj = ∂ζs ∂ζl ψj = 0 on M , for all 3 ≤ s, j, l ≤ n,

which are obtained from ψj (ξ, η, ζ ) = ζj on M , j = 3, . . . , n.

Since similar matrices have the same trace, one has

   −1 
n
trace E − ∂ζ ψ ∂ζ P E − ∂ζ ψ = trace(∂ζ P) = ∂ζj Pj ,
j=3
  −1  −1
trace ∂ξ ψ ∂ζ P1 E − ∂ζ ψ = trace E − ∂ζ ψ ∂ξ φ ∂ζ P1
 −1
= ∂ζ P1 E − ∂ζ ψ ∂ξ ψ,
  −1  −1
trace ∂η ψ ∂ζ P2 E − ∂ζ ψ = trace E − ∂ζ ψ ∂η ψ ∂ζ P2
 −1
= ∂ζ P2 E − ∂ζ ψ ∂η ψ.
70 2 Jacobian and Inverse Jacobian Multipliers

These last three equalities together with (2.46) show that


n
 −1  −1
K3 + · · · + Kn = ∂ζj Pj − ∂ζ P1 E − ∂ζ ψ ∂ξ ψ − ∂ζ P2 E − ∂ζ ψ ∂η ψ.
j=3
(2.47)
Combining (2.44), (2.47) and (2.42) yield

K0 |M = ∂ξ P1 + ∂η P2 + ∂ζ P1 ∂ξ γ + ∂ζ P2 ∂η γ M
|M ).
= ∂ξ P1 (ξ, η, γ (ξ, η)) + ∂η P2 (ξ, η, γ (ξ, η)) = div(X

This together with (2.43) verifies statement (b).

The lemma is proved. 

The proof of Lemma 2.4 shows that this last lemma also holds for C ∞ smoothness.
We now study the relation between center manifolds and C ∞ inverse Jacobian
multipliers.

Lemma 2.5 Assume that system (2.29) satisfies (H) and has a C ∞ inverse Jacobian
multiplier, which is expressed in the complex conjugate coordinates of the form


n
J(ξ, η, ζ ) = (ζs − ψs (ξ, η, ζ ))V (ξ, η, ζ ),
s=3

where ψs = O(|(ξ, η, ζ )|2 ) and V |ζ =ψ (ξ,η,ζ ) ≡ 0 for any  ∈ {3, . . . , n}. The
following statements hold.

n
(a) M∗ = {ζs = ψs (ξ, η, ζ )} is a center manifold of system (2.29) at the origin.
s=3
(b) For any smooth center manifold M of system (2.29) at the origin, if X |M has
a center at the origin, then J|M = 0.

Proof (a) The proof is similar to that of Lemma 2.4. The details are omitted.
(b) For any C0 ∈ M , which is located in a sufficiently small neighborhood of the
origin, let ϕt be the periodic orbit of period T of (2.29) passing through P0 . By the
definitions we have
dJ(ϕt )
= X (J)|ϕt = JdivX |ϕt .
dt
Integrating this equation from 0 to T yields

T 
J(C0 ) = J(C0 ) exp divX |ϕs ds . (2.48)
0
2.3 Inverse Jacobian Multipliers and the Center-Focus Problem 71

Restricting system (2.29) to the center manifold M , and then writing it in polar
coordinates (x, y) = (r cos θ, r sin θ ), one gets

dt = (1 + O(r))−1 dθ .

Integrating this last equation along the periodic orbit yields T = 2π + O(r). Hence
we have

T
T
divX |ϕs ds = (ι + O(|C0 |))ds = 2π ι + O(|C0 |) = 0,
0 0

in a sufficiently small neighborhood of the origin, where

ι = divX |at the origin = λ3 + · · · + λn = 0.

This together with (2.48) forces

J(C0 ) = 0.

By the arbitrariness of C0 ∈ M one has J|M ≡ 0. This proves statement (b).

This completes the proof of the lemma. 

2.3.1.2 Proof of Theorem 2.19

In the proof of Theorem 2.19 we will also use the notations given in the previous
subsection.
(a) Sufficiency. As in the last subsection we write system (2.29) in the form (2.36)
with complex conjugate coordinates. Lemma 2.4 and its proof show that system
(2.36) has an analytic center manifold with the expression


n
M = {ζs = γs (ξ, η)},
s=3

which is tangent to the (x, y) plane (i.e. (ξ, η) plane), where the γs ’s are analytic func-
tions satisfying γs (0, 0) = 0 and ∂γs (0, 0) = 0. Moreover, system (2.36) restricted
to M has an analytic inverse integrating factor I(ξ, η) = V (ξ, η, γ (ξ, η)), where V 
is V (x, y, z) written in the complex conjugate coordinates (ξ, η, ζ ).
Going back to the real coordinates, the center manifold M can be expressed in
the form
 n
M = {zs = hs (x, y)}.
s=3
72 2 Jacobian and Inverse Jacobian Multipliers

Set h(x, y) = (h3 (x, y), . . . , hn (x, y)). Since I(0, 0) = V (0, 0, 0) = 0, and
V (x, y, h(x, y)) is an inverse integrating factor of system (2.29) when restricted to
M , the differential one-form

1
((x + f2 (x, y, h(x, y)))dx + (y − f1 (x, y, h(x, y)))dy)
V (x, y, h(x, y))

is well defined in a neighborhood of the origin and it is exact. So it has an analytic first
integral of the form H(x, y) = x 2 + y2 + higher-order terms. By the Poincaré center
theorem it follows that the origin of the vector field X when restricted to the center
manifold M is a center. Furthermore, we get from Sijbrand [402, Theorems 6.3 and
7.1] that the center manifold at the origin is unique. So M c = M and consequently
system (2.29) restricted to M c has a center at the origin. This proves the statement.
Necessity. As in the proof of sufficiency, we still write system (2.29) in the form
(2.36). By Lemmas 2.2 and 2.3 there exists a distinguished normalization which
transforms system (2.36) to system (2.37).
By the proof of Lemma 2.3, we get that system (2.37) satisfies g1 = g2 , and so it
has the inverse Jacobian multiplier J = w3 . . . wn . Applying Lemma 2.1 to systems
(2.36) and (2.37) and combining the distinguished normalization, we deduce that
system (2.36) has the analytic inverse Jacobian multiplier

J ∗ = (ζ3 − ψ3 (ξ, η, ζ )) . . . (ζn − ψn (ξ, η, ζ ))(D(ξ, η, ζ ))−1 ,

where ψs = O(|(ξ, η, ζ )|2 ) is analytic, s = 3, . . . , n, and D(ξ, η, ζ ) is the Jacobian

of the transformation from (2.36) to (2.37), and satisfies D(0, 0, 0) = 1. Writing J ∗
in the (x, y, z) coordinates provides an analytic inverse Jacobian multiplier of system
(2.29), which is of the form (2.34).
(b) The first statement, i.e. the analyticity and the uniqueness of the center manifolds,
follows from the proof of sufficiency of statement (a). The second statement, i.e.
M c ⊂ J −1 (0), can be obtained from Lemma 2.5 (b).
This completes the proof of the theorem. 

2.3.1.3 Proof of Theorem 2.20

Here we will also use the notation introduced in Sect. 2.3.1.1.

(a) Under the assumption of the theorem, by Lemma 2.3 (a) it follows that system
(2.36) is locally C ∞ equivalent to its Poincaré normal form (2.37) with ρ1 = ρ2 .
Clearly system (2.37) has the C ∞ inverse Jacobian multiplier


J(u, v, w) = w3 . . . wn uv(ρ2 (uv) − ρ1 (uv)).
2.3 Inverse Jacobian Multipliers and the Center-Focus Problem 73

Since ρ2 (s) − ρ1 (s) is nonflat at s = 0 and ρ2 (0) − ρ1 (0) = 0, we deduce that


J(u, v, w) = w3 . . . wn (uv) δ(uv),

with  ≥ 2 and δ(0) = 0. Hence system (2.37) has a C ∞ inverse Jacobian multiplier
of the form


n
J(ξ, η, ζ ) = (ζs − ψs (ξ, η, ζ ))((ξ − ψ1 )(η − ψ2 ))l
s=3
× δ((ξ − ψ1 )(η − ψ2 ))(D(ξ, η, ζ ))−1 ,

where ψs = O(|(ξ, η, ζ )|2 ) is C ∞ for s = 1, . . . , n. Then the statement follows from

the same arguments as in the proof of necessity of statement (a) of Theorem 2.19.
(b) The proof follows from statement (a) and Lemma 2.5.
This completes the proof of the theorem. 

2.3.2 Hopf Bifurcation via Inverse Jacobian Multipliers

Using the inverse Jacobian multipliers to study Hopf bifurcation, the first result was
due to Buică et al. [50] for a three-dimensional differential system. Roughly speaking,
Hopf bifurcation is a bifurcation phenomena which provides a mechanism to produce
limit cycles from a singularity, i.e. for a smooth differential system with parameters,
assume that at given values of the parameters the system has a singularity with a pair
of pure imaginary eigenvalues, the others having nonvanishing real parts, if under
a sufficiently small perturbation of the parameters the singularity is fixed and its
stability changes when restricted to the two-dimensional center manifold, then the
system will either produce or lose one limit cycle in a sufficiently small neighborhood
of the singularity.
To study the Hopf bifurcation of system (2.29), we consider its analytic pertur-
bation of the form

ẋ = −y + g1 (x, y, z, ε) = G 1 (x, y, z, ε),

ẏ = x + g2 (x, y, z, ε) = G 2 (x, y, z, ε), (2.49)
ż = Az + g(x, y, z, ε) = G(x, y, z, ε),

where ε ∈ Rm is an m-dimensional small parameter, g1 and g2 are analytic func-

tions, and g is an (n − 2)-dimensional vector-valued function. Set g(x, y, z, ε) :=
(g1 , g2 , g). We assume that

g(x, y, z, ε) = O(|(x, y, z)|) and g(x, y, z, 0) = f(x, y, z),

74 2 Jacobian and Inverse Jacobian Multipliers

where f is defined in (2.29), and that the Jacobian matrix of G = (G 1 , G 2 , G) with

respect to (x, y, z) at the origin has the eigenvalues

α(ε) ± i, λs + βs (ε) satisfying α(0) = βs (0) = 0, s = 3, . . . , n. (2.50)

The condition (2.50) together with the assumption (H) ensure that system (2.49) has
a two-dimensional center manifold, and its origin is monodromic when restricted to
this center manifold. By definition, a singularity of a two-dimensional smooth vector
field is monodromic if all orbits near the singularity spiral around the singularity.
We now study the Hopf bifurcation of system (2.49) at the origin when the parame-
ters ε vary near 0 ∈ Rm . By definition, the Hopf-cyclicity (or cyclicity for simplicity)
of system (2.29) at the origin is the maximal number of limit cycles which can bifur-
cate from the origin of system (2.49) via Hopf bifurcation. This maximal number is
denoted by Cycl(Xε , 0), where Xε is the vector field associated to (2.49).
The next result, due to Zhang [475], characterizes the Hopf-cyclicity via the
vanishing multiplicity of the inverse Jacobian multipliers at the origin, which is an
extension of the main result of [50] from three-dimensional differential systems to
any finite-dimensional system. Here the main techniques are those of [50].

Theorem 2.21 Assume that the analytic differential system (2.29) satisfies the
assumption (H). If system (2.29) restricted to the center manifold M c has a focus
at the origin, then Cycl(Xε , 0) =  − 1, where  is the vanishing multiplicity of the
inverse Jacobian multiplier defined in Theorem 2.20.

In what follows, for simplicity we assume without loss of generality that the matrix
A of system (2.29) is in the real Jordan normal form.

2.3.2.1 Preparation for the Proof of Theorem 2.21

By Theorem 2.20, together with the assumption of Theorem 2.21, we get that system
(2.29) has a C ∞ inverse Jacobian multiplier of the form


m−1
 
J(x, y, z) = (z3+2j − ψ3+2j (x, y, z))2 + (z3+2j+1 − ψ3+2j+1 (x, y, z))2
j=0

n
× (zs − ψs (x, y, z))(x 2 + y2 ) V (x, y, z),
s=3+2m

where m is the number of pairs of the complex conjugate eigenvalues of A, or


n
J(x, y, z) = (zs − ψs (x, y, z))(x 2 + y2 ) V (x, y, z),
s=3
2.3 Inverse Jacobian Multipliers and the Center-Focus Problem 75

where ψ3 , . . . , ψn , V ∈ C ∞ satisfy ψs (x, y, z) = O(|(x, y, z)|2 ) and V (0, 0, 0) = 1.

Moreover, by Lemmas 2.4 and 2.5 and their proofs system (2.29) has a C ∞ center
manifold M c at the origin, which is the intersection of the invariant hypersurfaces

zj = ψj (x, y, z), j = 3, . . . , n. (2.51)

As in Sect. 2.3.1, if m > 0 we choose the complex conjugate coordinates

(w3+2s , w3+2s+1 ) instead of the real ones (z3+2s , z3+2s+1 ). In these new coordinates
(x, y, w) we can write system (2.29) as

ẋ = −y + ρ1 (x, y, w),
ẏ = x + ρ2 (x, y, w), (2.52)
ẇs = ws (λs + ρs (x, y, w)), s = 3, . . . , n,

where ρ1 , ρ2 = O(|(u, v, w)|2 ) and ρj = O(|(u, v, w)|), j = 3, . . . , n. Obviously

system (2.52) has the two-dimensional center manifold w = 0, because Reλs = 0
for s = 3, . . . , n. By Lemma 2.1 system (2.52) has the inverse Jacobian multiplier

 (x, y, w),
J(x, y, w) = w3 . . . wn (x 2 + y2 ) V

 ∈ C ∞ satisfying V
with V (0, 0, 0) = 0.
Note that systems (2.29) and (2.52) are C ∞ equivalent near the origin, so we
can consider without loss of generality that system (2.49) is a small perturbation of
system (2.52). In the cylindrical coordinates

x = r cos θ, y = r sin θ, z = rσ, r ≥ 0,

system (2.49) can be written as

θ̇ = 1 + Θ(θ, r, σ, ε), ṙ = R(θ, r, σ, ε) σ̇ = As + Γ (θ, r, σ, ε), (2.53)

where Θ, R and Γ satisfy

Θ(θ, r, σ, 0) = O(r), R(θ, r, σ, 0) = O(r 2 ),

R(θ, 0, σ, ε) = 0, Γ (θ, r, σ, 0) = O(r).

Correspondingly, system (2.53) when ε = 0 has the inverse Jacobian multiplier


J(r cos θ, r sin θ, rσ )/r n−1 = σ3 . . . σn r 2−1 V ∗ (θ, r, σ ),

with V ∗ (θ, 0, 0) = constant = 0.

76 2 Jacobian and Inverse Jacobian Multipliers

In a sufficiently small neighborhood of the origin, for |ε| 1 we have 1 +

Θ(θ, r, σ, ε) > 0. So we can rewrite system (2.53) in an equivalent way as

dr dσ
= a(θ, r, σ, ε), = Aσ + b(θ, r, σ, ε), (2.54)
dθ dθ

where b = (b3 , . . . , bn )τ , and bs has the factor σs for ε = 0. Moreover, a and b are
defined on the cylinder
 
C = (θ, r, σ, ε) ∈ R/(2π R) × R × Rn−1 × Rm : |r|, |ε| 1

and satisfy

a(θ, 0, σ, ε) = 0, a(θ, r, σ, 0) = O(r 2 ), b(θ, r, σ, 0) = O(r). (2.55)

Let
Yε = ∂θ + a(θ, r, σ, ε)∂r + Aσ + b(θ, r, σ, ε), ∂σ 

be the vector field associated to system (2.54), where ∂σ = (∂σ3 , . . . , ∂σn ). From the
inverse Jacobian multiplier of system (2.53) when ε = 0 we get the inverse Jacobian
multiplier of the vector field Y0

J(r cos θ, r sin θ, rσ )

J0 (θ, r, σ ) = = σ3 . . . σn r 2−1 V0 (θ, r, σ ), (2.56)
r n−1 (1 + Θ(θ, r, σ, 0))

with V0 = 1 + O(r) for r sufficiently small.

Now the study of the periodic solutions of system (2.49) is equivalent to the study
of the 2π periodic solutions of system (2.54) on the cylinder C . To apply the Poincaré
map of system (2.54), we denote by

ϕθ (r0 , σ0 , ε) := (rθ (r0 , σ0 , ε), σθ (r0 , σ0 , ε))

the solution of system (2.54) satisfying the initial condition ϕ0 (r0 , σ0 , ε) = (r0 , σ0 ) ∈
C . Since the flow of system (2.54) is transversal to the section θ = 0 on the cylinder
C , and system (2.54) is 2π periodic in θ , we can define its Poincaré map in a
neighborhood of the origin as

P(r0 , σ0 ; ε) = ϕ2π (r0 , σ0 , ε),

which is C ∞ because system (2.54) is C ∞ . Set

Pr (r0 , σ0 , ε) = r2π (r0 , σ0 , ε) and Pσ (r0 , σ0 , ε) = σ2π (r0 , σ0 , ε).

2.3 Inverse Jacobian Multipliers and the Center-Focus Problem 77

We have

2π
Pr (r0 , σ0 , ε) = r0 + p(v, rv (r0 , σ0 , ε), σv (r0 , σ0 , ε), ε)dv,
0

2π 
Pσ (r0 , σ0 , ε) = eA2π Eσ0 + e−Av q(v, rv (r0 , σ0 , ε), σv (r0 , σ0 , ε), ε)dv ,
0

where E is the unit matrix of order n − 2. Now the periodic orbits of system (2.54)
are uniquely determined by the zeros of the displacement function

D(r0 , σ0 , ε) := P(r0 , σ0 , ε) − (r0 , σ0 ).

Setting Dr (r0 , σ0 , ε) = Pr (r0 , σ0 , ε) − r0 , Dσ (r0 , σ0 , ε) = Pσ (r0 , σ0 , ε) − σ0 ,

we have D = (Dr , Dσ ). By (2.55) we can check that

Dσ (0, 0, 0) = 0, ∂σ Dσ (0, 0, 0) = e2πA − E.

So by the implicit function theorem it follows that the functional equation

Dσ (r0 , σ0 , ε) = 0

has a unique C ∞ solution, say σ0 = δ(r0 , ε), defined in a neighborhood of (r0 , ε) =

(0, 0). Then the zeros (r0 , σ0 ) of the displacement function D(r0 , σ0 , ε) are uniquely
determined by the zeros r0 , in terms of ε, of

d(r0 , ε) := Dr (r0 , δ(r0 , ε), ε),

through σ0 = δ(r0 , ε).

Next, to prove Theorem 2.21 we only need to study the zeros r0 of the reduced
function d(r0 , ε).

2.3.2.2 Proof of Theorem 2.21

The previous subsection reduces the proof of Theorem 2.21 to the study of the zeros
of d(r0 , ε) in r0 .
In [48] Buică and García provided a relation between inverse Jacobian multipliers
and the Poincaré map, which in system (2.54) with ε = 0 reads as

J0 (0, P(r0 , σ0 , 0)) = J0 (0, r0 , σ0 )DP(r0 , σ0 , 0), (2.57)

where DP is the Jacobian of P with respect to (r0 , s0 ). Using the expression of J0

in (2.56), the equality (2.57) reads
78 2 Jacobian and Inverse Jacobian Multipliers

Pσ 3 . . . Pσ n Pr2−1 V0 (0, Pr , Pσ ) = σ03 . . . σ0n r02−1 V0 (0, r0 , σ0 )DP(r0 , σ0 , 0),

(2.58)
where σ0 = (σ03 , . . . , σ0n ), and

Pσ (r0 , σ0 , 0) = (Pσ 3 (r0 , σ0 , 0), . . . , Pσ n (r0 , σ0 , 0)) = σ0 , Pσ∗ (r0 , σ0 ),

with Pσ∗ a suitable (n − 2)-dimensional C ∞ function, because each hyperplane

σs = 0, s = 3, . . . , n, is invariant under the flow of system (2.54) when ε = 0. Some
calculations show that

DP(r0 , σ0 , 0)|σ0 =0 = Pσ∗3 . . . Pσ∗n ∂r Pr |σ0 =0 , δ(r0 , 0) ≡ 0.

Via this expression the equality (2.58) simplifies to

Pr2−1 V0 (0, Pr , 0) = r02−1 V0 (0, r0 , 0)∂r Pr for σ0 = 0, (2.59)

and
d(r0 , 0) = D(r0 , 0, 0).

Set d(r0 , 0) = κk r0k + O(r0k+1 ) with κk = 0. We have

Pr (r0 , 0, 0) = r0 + κk r0k + O(r0k+1 ),

and consequently
V0 (0, Pr , 0) = V0 (0, r0 , 0) + O(r0k ).

So the equality (2.59) can be written as


V0 (0, r0 , 0) (2 − 1)κk r02−2+k + O(r 2−1+k ) + O(r02−1+k ) = V0 (0, r0 , 0)kκk r02−2+k .

This together with V0 (0, 0, 0) = 1 shows that

k = 2 − 1.

Moreover k ≥ 3, because  ≥ 2 by Theorem 2.20.

Using the expression of d(r0 , 0), we get from the Weierstrass preparation theorem
that d(r0 , ε) has at most 2 − 1 zeroes. Since r0 = 0 is a solution of d(r0 , ε) = 0,
and system (2.53) is symmetric with respect to (θ, r, s) → (θ + π, −r, −s), we see
that d(r0 , ε) = 0 has at most  − 1 positive roots in r0 . This proves that system (2.49)
has at most  − 1 small amplitude limit cycles which bifurcate from the origin of
system (2.29) on the two-dimensional center manifold.
2.3 Inverse Jacobian Multipliers and the Center-Focus Problem 79

The next example

u̇ = −v + g1 (u, v, w) + uh(u, v, ε),

v̇ = u + g2 (u, v, w) + vh(u, v, ε), (2.60)
ẇj = wj (λj + gj (u, v)) + wj h(u, v, ε), j = 3, . . . , n,


−1
shows the existence of  − 1 limit cycles, where h(u, v, ε) = ε−s as (u2 + v2 )s ,
s=1
Reλj = 0, j = 3, . . . , n, and wj and its conjugation appear in pairs. For more details,
see [50, 161].
In short the theorem is proved. 

2.4 Inverse Jacobian Multipliers via Lie Groups

In this section we introduce some fundamental results on the study of inverse inte-
grating factors and of inverse Jacobian multipliers via Lie groups.
An r-parameter Lie group G is by definition
• not only an r-dimensional smooth manifold,
• but also a group with its group operator denoted by ‘·’, such that
– the group operation

m : G × G → G, m(g, h) = g · h, g, h ∈ G,

– and the inverse operation

i : G → G, i(g) = g −1 , g ∈ G,

are both smooth on the manifold, where g −1 is the inverse element of g under
the group action.

Example G = Rn is an n-parameter Lie group with its group operator defined by

the summation of two vectors in Rn because Rn is a trivial n-dimensional manifold,
and the summation and its inverse operation (i.e. subtraction) are both analytic.

Another simple example is

  !
cos θ − sin θ
G = SO(2) := | θ ∈ [0, 2π ) ,
sin θ cos θ

which is a one-parameter Lie group (called the rotation group on the plane), where θ
denotes the angle rotated counterclockwise and the group operator is the multiplica-
tion of two matrices. Indeed, the structure of a one-dimensional manifold of SO(2)
80 2 Jacobian and Inverse Jacobian Multipliers

can be viewed by identifying G with the unit circle in R2

S1 = {(cos θ, sin θ )| θ ∈ [0, 2π )}.

Clearly SO(2) is a group under the action of multiplication of matrices, and the
product and inverse operations of matrices in SO(2) are analytic.
Let V ⊂ Rr be an open subset containing the origin, and V0 ⊂ V be a connected
subset having the origin in its interior. An r-parameter local Lie group consists of
the connected open subset V0 and a smooth group operator

m : V × V → Rr ,

together with its smooth inverse operator

i : V0 → V.

Example V = (−ε, ε) ⊂ R is a one-parameter local Lie group with group operation

the summation of numbers in V .

Let M be a smooth manifold, and let G be a Lie group with the group operator
‘·’. A transformation group acting on M consists of the Lie group G and a smooth
map ϕ : G × M → M satisfying
• ϕ(g, ϕ(h, x)) = ϕ(g · h, x) for all x ∈ M and all g, h ∈ G,
• ϕ(e, x) = x for all x ∈ M, where e is the unit element of the group G,
• ϕ(g −1 , ϕ(g, x)) = x for all x ∈ M and all g ∈ G.
A local transformation group acting on M consists of the local Lie group V0 ⊂ G
with the unit element e ∈ V0 and a smooth map ϕ : V0 × M → M satisfying
• ϕ(g, ϕ(h, x)) = ϕ(g · h, x) for all x ∈ M and all g, h ∈ V0 such that g · h ∈ V0
and ϕ(h, x) ∈ M,
• ϕ(e, x) = x for all x ∈ M,
• ϕ(g −1 , ϕ(g, x)) = x for all x ∈ M and all g ∈ V0 such that g −1 ∈ V0 and
ϕ(g, x) ∈ M.

Examples We now illustrate some transformation groups.

1. Translation groups in Rn . For arbitrary 0 = a ∈ Rn and G = R,

ϕ : G × Rn −→ Rn
(t, x) −→ x + at

is called a translation group. Obviously the group acts on the whole space Rn ,
and its orbits are all straight lines parallel to the vector a.
2. Rescaling groups in Rn . Let G = R+ := {t ∈ R| t > 0} be the multiplication
group. Set non-zero (α1 , . . . , αn ) ∈ Rn . Then
2.4 Inverse Jacobian Multipliers via Lie Groups 81

ϕ : G × Rn −→ Rn
(λ, x) −→ (λα1 x1 , . . . , λαn xn ),

is called a rescaling group. Each orbit generated by this group is a one-dimensional

smooth manifold.
3. For the smooth differential system (1.1) defined in Ω, or its associated vector field
X , denote by ϕt (x) (or ϕ(t, x)) its flow satisfying ϕ(0, x) = x, t ∈ J, where J is
either R or an open interval centered at the origin. Since J is a single-parameter
(local) Lie group, we get from the properties of solutions of the autonomous
differential systems that ϕt (x) is a one-parameter transformation group acting on
Ω. In this sense we call the vector field X an infinitesimal generator of this
group action.
For example, the infinitesimal generator of the one-parameter Lie group SO(2)
acting on the plane is
ẋ = −y, ẏ = x.

The symmetry group of system (1.1) is a local transformation group G acting on

the subset M of the product space R × Ω of the time and the phase space, such that
each integral curve of system (1.1) in M under the action of the group G is still an
integral curve of the system provided that its image is still in M.

d2x
Example SO(2) is a symmetry group of the differential equation = 0 because
dt 2
the set of integral curves of the equation is formed by all straight lines in the (t, x)
plane, and the action of SO(2) is a rotation in the plane and it transforms each straight
line to a straight line.

Next we need the Lie bracket of two vector fields. Let X and Y be two smooth
vector fields defined on a domain Ω. Their Lie bracket is defined as

[X , Y ](f ) = X Y (f ) − Y X (f ), for all f ∈ C 1 (Ω).

If the vector fields X and Y have local expressions


n
∂ 
n
∂
X = fi (x) , Y = gj (x) ,
i=1
∂xi j=1
∂xj

then


n n  n  
∂ ∂gi ∂fi ∂
[X , Y ] = (X (gi ) − Y (fi )) = fj − gj .
i=1
∂xi i=1 j=1
∂xj ∂xj ∂x i

Direct calculations show that for arbitrary smooth vector fields X , Y , Z defined
on the domain Ω, their Lie bracket satisfies
82 2 Jacobian and Inverse Jacobian Multipliers

• bilinearity: for arbitrary constants a, b,

[aX + bY , Z ] = a[X , Z ] + b[Y , Z ],

[X , aY + bZ ] = a[X , Y ] + b[X , Z ].

• anti-symmetry: [X , Y ] = −[Y , X ].
• Jacobi identity: [X , [Y , Z ]] + [Z , [X , Y ]] + [Y , [Z , X ]] = 0.
The next results provide a relation between the Lie bracket of vector fields and
their flows.
Theorem 2.22 Assume that X and Y are two smooth vector fields defined in an
open subset Ω ⊂ Rn .
(a) If Y is an infinitesimal generator of some symmetry group of the vector field
X , then the Lie bracket of X and Y satisfies

[X , Y ] = α(x)X ,

where α(x) is a scalar function, and is a common first integral of X and Y .

(b) Denote by ϕt and ψs the flows of the vector fields X and Y , respectively. Then
the two flows commute, i.e.

ϕt ◦ ψs (x) = ψs ◦ ϕt (x),

if and only if [X , Y ] = 0.
Theorem 2.22 (a) can be found in Berrone and Giacomini [32] and Stephani [411],
and (b) can be found in Olver [351, Theorem 1.34].
We now discuss the relation between the existence of a symmetry group and an
inverse integrating factor (Jacobian multiplier) of a differential system.

Theorem 2.23 Assume that the planar smooth differential system (1.1) has a sym-
metry group whose infinitesimal generator is V = g1 ∂x1 + g2 ∂x2 . Then system (1.1)
has an inverse integrating factor of the form V (x) = f1 g2 − f2 g1 .

Proof See Olver [351, Theorem 2.48] and its proof. This result is a special case of
the following Theorems 2.24 and 2.25, where we will give a unified proof. 

Example Consider the planar homogeneous differential system

ẋ = P(x, y), ẏ = Q(x, y),

where P and Q are homogeneous polynomials of the same degree. We can easily
check that this system has a symmetry group whose infinitesimal generator is

ẋ = x, ẏ = y.
2.4 Inverse Jacobian Multipliers via Lie Groups 83

So we get from Theorem 2.23 that this last homogeneous differential system has an
inverse integrating factor V (x, y) = xQ(x, y) − yP(x, y).
We now extend Theorem 2.23 to higher-dimensional differential systems.
Theorem 2.24 Assume that the n-dimensional smooth differential system (1.1) has
n − 1 symmetry groups, and their infinitesimal generators are

Yi = gi1 ∂x1 + · · · + gin ∂xn , i = 1, . . . , n − 1.

Then  
 f1 f2 · · · fn 

 g11 g · · · g1n 
 12
V (x) =  . .. .. . 
 .. . . .. 

 gn−1,1 gn−1,2 · · · gn−1,n 

is an inverse Jacobian multiplier of system (1.1).

Theorem 2.24 is also a special case of the following Theorem 2.25. The proof of
Theorem 2.24 will be obtained as an application of Theorem 2.25.
For simplicity of notation we will use differential forms. Here we only introduce
some basic notions and fundamental results on differential forms, which will be used
later on. For more details, we refer to [1, 13, 351].
Any real smooth function f (x) is a smooth differential 0-form. Its differential


n
df = ∂xi f dxi
i=1

is a differential 1-form. Generally, a smooth differential 1-form in Rn is

ω = a1 (x)dx1 + · · · + an (x)dxn ,

where the ai (x)’s are real smooth functions. The differential form dxi acting on the
vector ∂xj becomes a real function, i.e.

1, i = j,
dxi ; ∂xj  = δij =
0, i = j.

The differential 1-form ω acting on the vector field X associated with system (1.1)
becomes a real function, i.e.


n
ω; X  = ai (x)fi (x).
i=1

For differential 1-forms ω1 , . . . , ωk , their wedge product ω1 ∧ · · · ∧ ωk is by

definition
84 2 Jacobian and Inverse Jacobian Multipliers

ω1 ∧ · · · ∧ ωk ; X1 , . . . , Xk  = det(ωi ; Xj ),

for arbitrary vectors X1 , . . . , Xk . From the properties of determinants we easily see

that the wedge product satisfies
• multi-linearity: for an arbitrary differential 1-form ω1 , . . . , ωi , ωi , . . . , ωk and arbi-
trary constants a, b, we have

ω1 ∧ · · · ∧ (a ωi + b ωi ) ∧ · · · ∧ ωk = a ω1 ∧ · · · ∧ ωi ∧ · · · ∧ ωk
+ b ω1 ∧ · · · ∧ ωi ∧ · · · ∧ ωk ,

• anti-symmetry: ωi ∧ ωi = 0, ωi ∧ ωj = −ωj ∧ ωi for i = j.

In the coordinates x = (x1 , . . . , xn ) of Rn , a smooth differential k-form is by
definition 
ω= aI (x)dx I ,
I

where I := {(i1 , . . . , ik )} is a set of strictly monotone and increasing indices of

multiplicity k, i.e. 1 ≤ i1 < i2 < · · · < ik ≤ n, dx I = dxi1 ∧ · · · ∧ dxik and the aI (x)’s
are real smooth functions.
Generally, the wedge product of differential forms is defined by satisfying the
following properties:
• bilinearity: for arbitrary differential forms ω, ω and θ, θ  , and arbitrary constants
a, b,

(a ω + b ω ) ∧ θ = a ω ∧ θ + b ω ∧ θ, ω ∧ (a θ + b θ  ) = a ω ∧ θ + b ω ∧ θ  .

• associative law: for arbitrary differential forms ω, θ, ζ ,

ω ∧ (θ ∧ ζ ) = (ω ∧ θ ) ∧ ζ.

• anti-commutative law: for an arbitrary differential k-form ω and differential l-

form θ ,
ω ∧ θ = (−1)kl θ ∧ ω.

Clearly any differential form of order larger than n in an n-dimensional

space is zero.
The differential or exterior derivative of a differential k-form ω = aI (x)dx I is
I
defined as
 
n
dω = daI (x)dx I = ∂xj aI (x)dxj ∧ dx I ,
I I j=1

and it is a differential (k + 1)-form.

The Lie derivative of a differential form with respect to a vector field X , denoted
by LX , is defined by satisfying the following properties:
2.4 Inverse Jacobian Multipliers via Lie Groups 85

• linearity: for arbitrary differential forms ω, θ and arbitrary constants a, b,

LX (a ω + b θ ) = a LX (ω) + b LX (θ ).

• Leibniz identity: for arbitrary differential forms ω, θ ,

LX (ω ∧ θ ) = LX (ω) ∧ θ + ω ∧ LX (θ ).

• commutativity with differential: LX (dω) = dLX (ω).

From the definition we easily see that the Lie derivative of a differential k-form
with respect to the vector field X satisfies
  ⎛ ⎞
  
k
LX aI (x)dx I
= ⎝LX (aI )dx + I
aI dxi1 ∧ · · · ∧ LX (dxij ) ∧ · · · ∧ dxik ⎠ .
I I j=1


n
For the vector field X = fi ∂xi , the Lie derivative of a smooth function a(x)
i=1
with respect to X is by definition

LX (a(x)) = f1 ∂x1 a(x) + · · · + fn ∂xn a(x).

In particular,
LX (dxi ) = dLX (xi ) = dfi .

The Lie derivative of a vector field Y with respect to X is by definition

LX (Y ) = [X , Y ].

The Lie derivative of vector fields and of differential forms with a vector field can
be defined in a unified way, see e.g. Olver [351, Definition 1.63].

Theorem 2.25 Let X be the vector field associated to system (1.1). Assume that
the n vector fields X1 , . . . , Xn defined in Ω satisfy


n
LX (Xi ) = [X , Xi ] = aij (x)Xj , aij (x) ∈ C 1 (Ω), 1 ≤ i, j ≤ n,
j=1

and the trace of the matrix (aij (x)) formed by the coefficients of these last expressions
vanishes, i.e. tr(aij (x)) ≡ 0. Then

V (x) = det(dxi ; Xj )

is an inverse Jacobian multiplier of system (1.1).

86 2 Jacobian and Inverse Jacobian Multipliers

Proof The proof follows from Berrone and Giacomini [31]. By the properties of
determinants, it follows that the Lie derivative of a differential k-form ω acting on
vector fields X1 , . . . , Xk with respect to X is

LX (ω; X1 , . . . , Xk ) =LX (ω); X1 , . . . , Xk 


k
+ (ω; X1 , . . . , LX (Xi ), . . . , Xk ).
i=1

Next we compute the three parts by using the volume form σ = dx1 ∧ · · · ∧ dxn
to replace ω. From the wedge product of differential 1-forms we get

LX (σ ; X1 , . . . , Xn ) = LX det(dxi , Xj ) = LX V = X (V ).

Then it follows from the properties of the Lie derivative and the linearity of the wedge
product of differential forms that


n
LX (σ ); X1 , . . . , Xn  = dx1 ∧ · · · ∧ LX (dxi ) ∧ · · · ∧ dxn ; X1 , . . . , Xn 
i=1
n
= dx1 ∧ · · · ∧ dfi ∧ · · · ∧ dxn ; X1 , . . . , Xn 
i=1
n
= dx1 ∧ · · · ∧ ∂xi fi dxi ∧ · · · ∧ dxn ; X1 , . . . , Xn 
i=1
n
= ∂xi fi dx1 ∧ · · · ∧ dxi ∧ · · · ∧ dxn ; X1 , . . . , Xn  = V divf ,
i=1

and
⎛" #⎞

n 
n 
n
(σ ; X1 , . . . , LX (Xi ), . . . , Xn ) = ⎝ σ ; X1 , . . . , aij (x)Xj , . . . , Xn ⎠
i=1 i=1 j=1
⎛ ⎞

n n
= ⎝ aij (x)σ ; X1 , . . . , Xj , . . . , Xn ⎠
i=1 j=1
n
= (aii (x)σ ; X1 , . . . , Xi , . . . , Xn )
i=1
= tr(aij )V = 0.

These verify X (V ) = V divf = V divX , and consequently V is an inverse Jacobian

multiplier of system (1.1). This proves the theorem. 
2.4 Inverse Jacobian Multipliers via Lie Groups 87

Proof of Theorem 2.24. Set X1 = X , Xi = Yi−1 , i = 2, . . . , n. Then we have

V (x) = det(dxi ; Xj ).

By the assumption and Theorem 2.22 (a) we get that there exist functions ai (x), i =
2, . . . , n, such that

LX (X1 ) = [X , X1 ] = [X , X ] = 0,
LX (Xi ) = [X , Xi ] = [X , Yi−1 ] = ai (x)X = ai (x)X1 , i = 2, . . . , n.

Obviously the trace of the matrix formed by the coefficients of the expressions of
these Lie derivatives is zero. It follows from Theorem 2.25 that V (x) is an inverse
Jacobian multiplier of system (1.1). This proves the theorem. 
Some of the results on the relation between inverse integrating factors and limit
cycles for planar differential systems can be extended to inverse Jacobian multipliers
for higher-dimensional differential systems. Roughly speaking, a limit cycle of a
higher-dimensional differential system is an isolated periodic orbit.

Theorem 2.26 Assume that V (x) is a smooth inverse Jacobian multiplier of the
smooth differential system (1.1) in an open set Ω0 ⊂ Ω, and that Γ ⊂ Ω0 is a limit
cycle of system (1.1), which is an ω (resp. α) limit set of all nearby orbits. Then
Γ ⊂ V −1 (0).

Proof The main idea of the proof comes from [31]. Let ϕt be the flow of system
(1.1). Since the inverse Jacobian multiplier V satisfies

X (V ) = V divf ,

if ϕt (x) ∈ Ω0 we have

t 
V (ϕt (x)) = V (x) exp divf ◦ ϕs (x)ds .
0


This shows that for any orbit γ ⊂ Ω0 , we have either γ ⊂ V −1 (0) or γ V −1 (0) =
∅. Without loss of generality we assume that Γ is the ω limit set of all its nearby
orbits, and so it is asymptotically stable.
On the contrary, we assume that Γ  V −1 (0). Then there exists a closed tubular
neighborhood Bε (Γ ) centered at Γ of radius ε > 0, such that

0 < m ≤ V (x) ≤ M, for x ∈ Bε (Γ ),

where m, M ∈ R. Since Γ is asymptotically stable, we assume without loss of

generality that
ϕt (Bε (Γ )) ⊂ Bε (Γ ), t > 0.
88 2 Jacobian and Inverse Jacobian Multipliers

For D0 ⊂ Rn , denote by D0  its volume. It follows from Theorem 2.2 (i.e. Poincaré’s
theorem) that the integral

dx
ϕt (Bε (Γ )) V (x)

is invariant with respect to the time t. Hence we have

ϕt (Bε (Γ )) dx dx Bε (Γ )

≥ = ≥ , t > 0.
m ϕt (Bε (Γ )) V (x) Bε (Γ ) V (x) M

On the other hand, since

lim ϕt (Bε (Γ )) ⊂ Γ,
t→∞

we have
lim ϕt (Bε (Γ )) = 0.
t→∞

This contradiction implies that Γ ⊂ V −1 (0). 

For higher-dimensional differential systems it is still an open problem to charac-

terize the existence and regularity of the inverse Jacobian multipliers near a singu-
larity or a limit cycle, except in the cases studied in Sect. 2.3.
Chapter 3
Darboux and Liouvillian Integrability

Darboux and Liouvillian integrability is mainly concerned with algebraic aspects of

the integrability of differential systems, which is related to many subjects, such as real
and complex analysis, algebraic geometry, differential algebra, differential Galois
theory, and so on. The Darboux theory of integrability is classical, and is related to
the Poincaré problem, which asks to find the upper bound of invariant algebraic curves
of planar polynomial differential systems. This theory is also involved in the study
of Hilbert’s 16th problem [204]. There have appeared richer results on this subject,
see for instance Darboux [122, 123], Jouanolou [217], Schlomiuk [385], Carnicer
[59], Chavarriga et al. [66], Llibre [248], Dumortier et al. [134], Christopher et al.
[106], Llibre and Zhang [302], Zhang [477] and the references therein.
This chapter will introduce some basic concepts and fundamental results on the
algebraic aspects of integrability, for instance Darboux integrability and its exten-
sions, Liouvillian and elementary integrability, and the relations among these three
kinds of integrability.

3.1 The Classical Darboux Theory of Integrability

The Darboux theory of integrability was named for Jean–Gaston Darboux (1842–
1917). In 1878 he provided a method to construct first integrals of polynomial dif-
ferential systems via a sufficient number of invariant algebraic curves, or surfaces,
or hypersurfaces, see Darboux [122, 123] and Poincaré [362].
Consider the polynomial differential system

ẋ = P(x), x ∈ Kn , (3.1)

© Springer Nature Singapore Pte Ltd. 2017 89

X. Zhang, Integrability of Dynamical Systems: Algebra and Analysis,
Developments in Mathematics 47, DOI 10.1007/978-981-10-4226-3_3
90 3 Darboux and Liouvillian Integrability

where K = C or R, and P(x) = (P1 (x), . . . , Pn (x)) is a vector-valued polynomial

function. We call m := max{degP1 , . . . , degPn } the degree of the polynomial dif-
ferential system (3.1). In this chapter we will always use m to denote the degree of
system (3.1). Denote by

∂ ∂
XP = P1 (x) + · · · + Pn (x)
∂x1 ∂xn

the polynomial vector field associated to system (3.1).

Let K[x] be the ring of polynomials in x with coefficients in K. A polynomial
f (x) ∈ C[x] is a Darboux polynomial of the polynomial differential system (3.1) if
there exists a k(x) ∈ C[x] such that

XP (f ) = kf . (3.2)

The polynomial k(x) is called the cofactor of the Darboux polynomial f (x). Clearly,
if f (x) is a Darboux polynomial of system (3.1), the set V (f ) := {x ∈ Kn | f (x) = 0}
is invariant under the flow of system (3.1), i.e. any orbit of system (3.1) with initial
point in V (f ) will always remain in V (f ). The set V (f ) is called the variety of f . For
simplicity of notation we will also use f = 0 to represent the variety V (f ).

Remark For a real polynomial vector field XP , its Darboux polynomials and cofac-
tors (if they exist) may be complex. Of course, if they appear, they will be in conjugate
pairs.
If f is a Darboux polynomial of system (3.1), we say f = 0 is an invariant algebraic
curve of system (3.1) for n = 2, or an invariant algebraic surface for n = 3, or an
invariant algebraic hypersurface for n > 3. In what follows, if the dimension of a
system is not specified, we will call f = 0 an invariant algebraic hypersurface without
distinguishing the three cases.

3.1.1 The Existence of Darboux First Integrals

This subsection will discuss the existence of Darboux first integrals, Darboux inte-
grating factors and Darboux Jacobian multipliers. We first present a result on Darboux
polynomials and their irreducible factors, which will be useful in our later analysis.

Proposition 3.1 Assume that f ∈ C[x] has an irreducible decomposition, say f =

f1m1 . . . frmr in C[x]. The following statements hold.
(a) f is a Darboux polynomial of system (3.1) if and only if each fj , j = 1, . . . , r, is
a Darboux polynomial of the system.
(b) If k(x), k1 (x), . . . , kr (x) are the cofactors of f (x), f1 (x), . . . , fr (x) respectively,
then
k(x) = m1 k1 (x) + · · · + mr kr (x).
3.1 The Classical Darboux Theory of Integrability 91

Proof This follows from some easy calculations. 

By Proposition 3.1, in what follows a Darboux polynomial will be irreducible

unless otherwise specified. The next result characterizes some preliminary properties
of Darboux polynomials and their cofactors.

Proposition 3.2 For irreducible f ∈ C[x], x ∈ Cn , the following statements hold.

(a) If f = 0 is an invariant set of system (3.43), i.e. XP (f )|f =0 = 0, then there exists
a k(x) ∈ C[x] such that the equality (3.2) holds.
(b) If k(x) is a cofactor of f (x), then deg k ≤ m − 1, where m is the degree of system
(3.43).

Proof The proof of (b) can be obtained from (3.2) by comparing the highest order
homogeneous terms.
Statement (a) can be proved by using Fulton [157, p.18, Corollary 1], or the next
result (see Olver [351, Proposition 2.10]).

Proposition 3.3 Let M ⊂ Rn be an m-dimensional smooth submanifold, and F =

(f1 , . . . , fl ) : M → Rl be a smooth map. Assume that F has the maximal rank on
V (F) := {x ∈ M| F(x) = 0}, i.e. for all x ∈ V (F), the rank of the Jacobian matrix
of F at x is min{m, l}. Then a real smooth function g : M → R vanishes on V (F) if
and only if there exist smooth functions q1 , . . . , ql such that

g(x) = q1 (x)f1 (x) + · · · + ql (x)fl (x), x ∈ M.

Since f ∈ C[x] is an irreducible polynomial, its derivative has at most finitely

many zeros on V (f ), and so has maximal rank. Applying Proposition 3.3 we get that
there exists a smooth function k(x) such that

XP (f )(x) = k(x)f (x), x ∈ Rn .

Since XP (f ) and f (x) are both polynomials, it follows that k(x) ∈ C(x). Moreover,
k(x) is smooth in Rn . This forces k(x) ∈ C[x]. The proof of Proposition 3.2 is com-
plete. 
Assume that f , g ∈ C[x]. If there exists a L(x) ∈ Cm−1 [x] of degree no more than
m − 1 such that
XP (exp(g/f )) = L exp(g/f ),

then we call exp(g/f ) an exponential factor of system (3.1), and L is the cofactor
associated to exp(g/f ). Recall that m is the degree of the vector field XP . In what fol-
lows, for an exponential factor exp(g/f ) we always require that g and f are relatively
prime, i.e. gcd(g, f ) = 1.

Remark An exponential factor can be seen as the coincidence of two invariant alge-
braic curves or hypersurfaces. That is, for two invariant algebraic curves
92 3 Darboux and Liouvillian Integrability

f =0 and f + g = 0,  ∈ C,

we have  1/
f + g
exp(g/f ) = lim .
→0 f

For more information, we refer to Christopher [94], Christopher and Llibre [98, 99],
and Llibre and Zhang [302–305] and the references therein.
We now discuss the relation between exponential factors and invariant algebraic
hypersurfaces, see e.g. [98, 99].

Proposition 3.4 A function exp(g/f ) is an exponential factor of the polynomial

differential system (3.1) if and only if f is a Darboux polynomial of (3.1), and

XP (g) = fL + gk,

where L and k are cofactors of exp(g/f ) and f , respectively.

Proof Necessity. From the definitions of cofactors it follows that

XP (g)f − gXP (f )
= L,
f2

i.e.
XP (g)f − Lf 2 = gXP (f ).

Since gcd(g, f ) = 1, there exists a k ∈ C[x] such that XP (f ) = kf . Hence we have

XP (g) = Lf + gk.

Sufficiency. By the proof of the necessity we get

      
g g XP (g)f − gXP (f ) g
XP exp = exp 2
= L exp .
f f f f

This proves the proposition. 

For g, f , f1 , . . . , fr ∈ C[x], s1 , . . . , sr ∈ C, a function of the form

 
g
H := f1s1 . . . frsr exp
f

is called a Darboux function. We remark that Darboux functions are in general multi-
valued.
If a first integral (or an integrating factor or a Jacobian multiplier) of the polynomial
differential system (3.1) is a Darboux function, it is called a Darboux first integral (or
3.1 The Classical Darboux Theory of Integrability 93

a Darboux integrating factor or a Darboux Jacobian multiplier). If a first integral is

a polynomial function (or a rational function), it is called a polynomial first integral
(or a rational first integral).

Proposition 3.5 Assume that the polynomial differential system (3.1) has p irre-
ducible Darboux polynomials f1 , . . . , fp with the associated cofactors k1 , . . . , kp ,
and q pairwise different exponential factors E1 , . . . , Eq with the associated cofac-
tors L1 , . . . , Lq . For s1 , . . . , sp , r1 , . . . , rq ∈ C, the following statements hold.
s r
(a) H = f1s1 . . . fp p E1r1 . . . Eqq is a Darboux first integral of system (3.1) if and only if
s1 k1 + · · · + sp kp + r1 L1 + · · · + rq Lq = 0.
s r
(b) M = f1s1 . . . fp p E1r1 . . . Eqq is a Jacobian multiplier (or integrating factor) of sys-
tem (3.1) if and only if s1 k1 + · · · + sp kp + r1 L1 + · · · + rq Lq = −divP.

Proof The proof of statement (a) follows from

 p  ⎛ ⎞

p
 
q

p

q
q
−1
XP (H) = si fisi −1 XP (fi )
r
Ej j + fisi ⎝ rj Ej j XP (Ej )⎠
r

i=1 i=1 j=1 i=1 j=1 j=1

⎛ ⎞
p

q
=⎝ si ki + rj Lj ⎠ H.
i=1 j=1

The proof of statement (b) follows from (a) and the fact that M is a Jacobian
multiplier if and only if XP (M) = −Mdiv(P). This proves the proposition. 

Remark If a real polynomial vector field XP has a complex Darboux first integral, it
must have a real Darboux first integral. This follows from the facts that the product
of two first integrals is also a first integral, and that if f is a Darboux polynomial of
XP , then its conjugate f is also a Darboux polynomial of XP . In fact, for all k ∈ C
we have
 
k k

2 Re k Imf
f f = (Re f ) + (Im f )
2
exp −2 Im k arctan ,
Ref

where Re and Im are respectively the real and imaginary parts of a complex function.
This last equality can be proved by using
 i/2 
1−iz
arctan z = ln , z ∈ C,
1+iz
√
where i = −1.
94 3 Darboux and Liouvillian Integrability

3.1.2 The Darboux–Jouanolou Integrability Theorem

In this subsection we prove the classical Darboux–Jouanolou integrability theorem.

Theorem 3.1 (Darboux–Jouanolou integrability theorem) Assume that the n-

dimensional polynomial differential system (3.1) of degree m has p irreducible Dar-
boux polynomials. Set  
m+n−1
N= .
n

(a) If p ≥ N + 1, system (3.1) has a Darboux first integral.

(b) System (3.1) has a rational first integral if and only if p ≥ N + n.

Statement (a) of Theorem 3.1 was obtained by Darboux [122, 123] in 1878.
Statement (b) was proved by Jouanolou [217] in 1979 using the sophisticated tools
of algebraic geometry. Christopher and Llibre [99] in 2000 provided an elementary
proof of the Jouanolou result in the two-dimensional case. For higher-dimensional
systems an elementary and simple proof of the Jouanolou result was given by Llibre
and Zhang [304] in 2010.
The next proof of Theorem 3.1 (b) is due to Llibre and Zhang [304]. Its proof
needs a technique result given in Lemma 3.1, see [304].
By definition, m functions H1 , . . . , Hm (m < n) are k-functionally independent in
D1 ⊂ Kn if k elements of H1 , . . . , Hm are functionally independent in D1 , and any
k + 1 elements of them are not functionally independent in any positive Lebesgue
measure subset of D1 .

Lemma 3.1 Assume that H1 , . . . , Hm are k (<m) functionally independent ana-

lytic first integrals of the polynomial vector field XP . Without loss of generality, let
H1 , . . . , Hk be functionally independent. The following statements hold.
(a) For each s ∈ {k + 1, . . . , m}, there exist analytic functions Cs1 (x), . . ., Csk (x)
defined in a full Lebesgue measure subset of Kn such that

∇Hs (x) = Cs1 (x)∇H1 (x) + · · · + Csk (x)∇Hk (x), (3.3)

where ∇ denotes the gradient of the functions Hj in x.

(b) For each s ∈ {k + 1, . . . , m}, j ∈ {1, . . . , k}, if Csj (x) is not a constant, it is a
first integral of the vector field XP .

We remark that Lemma 3.1 continues to hold if we replace ‘analytic’ by ‘smooth’.

Proof (a) Let D1 be a full Lebesgue measure subset of Kn in which H1 , . . . , Hm are

k-functionally independent. By the assumption there exists a full Lebesgue measure
subset D2 ⊂ D1 such that for each x ∈ D2

∇H1 (x), . . . , ∇Hk (x)

3.1 The Classical Darboux Theory of Integrability 95

are linearly independent, and ∇Hs (x), s ∈ {k + 1, . . . , m}, is linearly dependent on

∇H1 (x), . . . , ∇Hk (x) in Cn . Hence, for each x ∈ D2 there exist Cs1 (x), . . . , Csk (x)
such that the equality (3.3) holds. Solving Eq. (3.3) via Cramer’s rule, we get the
solutions Cs1 (x), . . . , Csk (x) of Eq. (3.3), which can be expressed as functions of
∇H1 , . . . , ∇Hk , ∇Hs on D2 . Hence they must be analytic on D2 . Statement (a)
follows.
(b) For each , m ∈ {1, . . . , n}, the th and mth component of (3.3) are respectively

∂Hs ∂H1 ∂Hk ∂Hs ∂H1 ∂Hk

= Cs1 (x) + · · · + Csk (x) , = Cs1 (x) + · · · + Csk (x) .
∂x ∂x ∂x ∂xm ∂xm ∂xm

Taking the derivatives of the first equation with respect to xm and the second one with
respect to x , and subtracting these two resulting equations, we obtain

∂Cs1 ∂H1 ∂Cs1 ∂H1 ∂Csk ∂Hk ∂Csk ∂Hk

− + ··· + − = 0. (3.4)
∂x ∂xm ∂xm ∂x ∂x ∂xm ∂xm ∂x

For k = n − 1, combining (3.4) with , m ∈ {1, . . . , n} together with some com-

putations yield

  
∂Cs1 ∂H1 ∂Cs1 ∂H1 ∂Csk ∂Hk ∂Csk ∂Hk
− + ··· + − ·
∂x ∂xm ∂xm ∂x ∂x ∂xm ∂xm ∂x
1≤<m≤n ⎞
 (3.5)
∂H ∂H ∂H ⎠ = 0,
(−1)τ (ijk1 k2 ...,kn−2 )
2 3 n−1
...
∂xk1 ∂xk2 ∂xkn−2
σ (k1 ,k2 ...,kn−2 )

where σ runs over all permutations of {1, . . . , n} \ {i, j}, and τ evaluated on a per-
mutation of {1, . . . , n} is the minimal number of transpositions needed to transform
the permutation to the identity. Equation (3.5) is essentially
 
 ∂Cs1 ∂Cs1 ∂Cs1 
 ... 
 ∂x1 ∂x2 ∂xn 
 
 
 ∂H ∂H1 ∂H1 
 1 
 ...  = 0. (3.6)
 ∂x1 ∂x2 ∂xn 
 . .. .. .. 
 .. . . . 
 
 ∂Hn−1 ∂Hn−1 ∂Hn−1 
 ... 
 
∂x1 ∂x2 ∂xn

Since ∇H1 (x), . . . , ∇Hn−1 (x) are linearly independent in D2 , they span an (n −
1)-dimensional vector space in D2 , denoted by Vn−1 (x). Equation (3.6) means that
the vector ∇Cs1 (x) belongs to Vn−1 (x). In addition, for j = 1, . . . , n − 1, we have

∇Hj (x), P(x)

= 0, x ∈ D2 ,
96 3 Darboux and Liouvillian Integrability

where P(x) = (P1 (x), . . . , Pn (x)). These two facts verify that the vectors P(x) and
∇Cs1 (x) are orthogonal in D2 , i.e.

∇Cs1 (x), P(x)

= 0, for all x ∈ D2 .

This proves that Cs1 (x) (if not a constant) is a first integral of the vector field XP on
D2 .
Applying the same arguments as above to the functions Csj (x) shows that all
functions Csj (x) (if not constants), j = 2, . . . , k, are first integrals of XP . Statement
(b) is proved for k = n − 1.
For k < n − 1, similar to the proof in the case k = n − 1, we can show that
for arbitrary 1 , . . . , k+1 satisfying 1 ≤ 1 < 2 < · · · < k+1 ≤ n, the following
equality holds:  
 ∂Cs1 ∂Cs1 ∂Cs1 
 . . . 
 ∂x ∂x ∂xk+1 
 1 2
 
 
 ∂H1 ∂H1 ∂H1 
 . . . 
 ∂x ∂x
 1 2
∂xk+1  = 0, x ∈ D2 .
 .. .. .. .
..  
 . . .
 
 ∂Hk ∂Hk ∂Hk 
 
 ∂x ∂x . . . ∂x 
1 2 k+1

This implies that for each x ∈ D2 , ∇Cs1 (x) belongs to the k-dimensional vector space
spanned by {∇H1 (x), . . . , ∇Hk (x)}, denoted by Vk (x). Then XP (x) is orthogonal to
∇Cs1 (x) in D2 . That is, Cs1 (x) (if not a constant) is a first integral of the vector field
XP in D2 . The same arguments work for Cs2 , . . . , Csk . This proves statement (b)
and consequently the lemma. 

Remark Stating Lemma 3.1 in the differential form:

Under the assumptions of Lemma 3.1, the following statements hold.
(a) For each s ∈ {k + 1, . . . , m}, there exist analytic functions Cs1 (x), . . ., Csk (x)
defined in a full Lebesgue measure subset of Kn such that

dHs (x) = Cs1 (x)dH1 (x) + · · · + Csk (x)dHk (x). (3.7)

(b) For each s ∈ {k + 1, . . . , m}, j ∈ {1, . . . , k}, if Csj (x) is not a constant, it is a first
integral of the vector field XP .
We can provide an easy proof of Lemma 3.1 as follows. The proof of (a) is the
same as that of Lemma 3.1. To prove (b), differentiating the two sides of (3.8) one
gets
0 = dCs1 (x)dH1 (x) + · · · + dCsk (x)dHk (x). (3.8)
3.1 The Classical Darboux Theory of Integrability 97

By the functional independence of H1 , . . . , Hk one has dCsj ≡ 0 for j = 1, . . . , k.

Hence all Csj are first integrals of the vector field XP provided that they are not
constant. 

Proof of Theorem 3.1. Let f1 (x), . . . , fp (x) be the Darboux polynomials of the vector
field XP , and k1 (x), . . . , kp (x) be the associated cofactors.
(a) We treat Cm−1 [x] as the N-dimensional vector space CN , which is formed by
the coefficients of the polynomials in Cm−1 [x]. Since p ≥ N + 1, the cofactors
k1 (x), . . . , kp (x) are linearly dependent in Cm−1 [x]. So there exist 1 , . . . , p ∈ C
such that
1 k1 (x) + · · · + p kp (x) = 0.


By Proposition 3.5, it follows that H = f11 . . . fp p is a first integral of the vector field
XP , which ends the proof of the statement.
(b) The necessity is obvious. We prove the sufficiency. Denote by V the vector space
spanned by {k1 (x), . . ., kN+n (x)}. Then its dimension ρ ≤ N.
For ρ = N, we assume without loss of generality that k1 (x), . . . , kN (x) ∈ V are
linearly independent. Then for each s ∈ {1, . . . , n}, k1 (x), . . . , kN (x), kN+s (x) are
linearly dependent. So there exist vectors (s1 , . . . , sN , 1) ∈ CN+1 such that

s1 k1 (x) + · · · + sN kN (x) + kN+s (x) = 0. (3.9)

This together with the fact that ki (x) = XP (fi )(x)/fi (x) shows that
  
XP log f1s1 . . . fNsN fN+s = 0.

That is, each 

Hs = log f1σs1 . . . fNσsN fN+s , s = 1, . . . , n,

is an analytic first integral of the vector field XP in a full Lebesgue measure subset D3
of Cn . Moreover, the first integrals H1 , . . . , Hn are functionally dependent in some
positive Lebesgue measure subset of D3 .
Define

r(x) = rank{∇H1 (x), . . . , ∇Hn (x)}, d = max{r(x) : x ∈ D3 }.

Then d < n and there exists an open subset O ⊂ D3 such that d = r(x), x ∈ O.
Without loss of generality we assume that {∇H1 (x), . . . , ∇Hd (x)} has rank d in O.
By Lemma 3.1 (a) there exist Ck1 (x), . . . , Ckd (x) such that for all x ∈ O

∇Hk (x) = Ck1 (x)∇H1 (x) + · · · + Ckd (x)∇Hd (x), k = d + 1, . . . , n. (3.10)

By Lemma 3.1 (b) the functions Ckj (x) (if not constants), j ∈ {d + 1, . . . , n}, are first
integrals of the vector field XP on O.
98 3 Darboux and Liouvillian Integrability

From the expressions of H1 , . . . , Hn and Cramer’s rule, we get from (3.10) that
Ckj (x), j ∈ {1, . . . , d}, are rational. This together with the equation

∇Ckj (x), P(x)

= 0, x ∈ O,

shows that the Ckj (x)’s (if not constants) are rational first integrals of the vector field
XP .
Finally, we prove that there exists some Ckj which is not a constant. Assuming
the contrary, we get from (3.10) that

Hk (x) = Ck1 H1 (x) + · · · + Ckd Hd (x) + log Ck , k ∈ {d + 1, . . . , n},

where the Ck ’s are constants. Substituting the expressions of the Hj (x)’s into this last
equation gives
 Ck1  Ckd
f1k1 . . . fNkN fN+k = Ck f111 . . . fN1N fN+1 . . . f1d1 . . . fNdN fN+d ,

for k ∈ {d + 1, . . . , n}. This equality is impossible, because f1 , . . . , fN+d , fN+k are

irreducible and pairwise distinct. So there must exist at least one nonconstant Ckj0 (x)
for some j0 ∈ {1, . . . , d}.
The above proofs verify that in the case p = N system (3.1) has at least one
rational first integral.
For p < N, the proof follows from the same arguments as those in the case p = N.
This proves Theorem 3.1. 

3.2 Generalization of the Classical Darboux Theory

of Integrability

The classical Darboux theory of integrability has many kinds of extensions. This
section will present its extensions taking into account the independence of singular-
ities and the multiplicity of invariant algebraic curves (surfaces or hypersurfaces).

3.2.1 Taking into Account Independent Singularities

On the generalization of the Darboux–Jouanolou integrability theorem, Christopher

[94] in 1994 introduced exponential factors (mentioned in the previous section), and
Chavarriga et al. [77] in 1997 took into account independent singularities.
A finite set Λ ⊂ Kn is independent with respect to Cm−1 [x] if

ρ := dim (∩w∈Λ mw ∩ Cm−1 [x]) = N − d,

3.2 Generalization of the Classical Darboux Theory of Integrability 99

where N is the number defined in Theorem 3.1, i.e. the dimension of the vector space
Cm−1 [x], d = Λ (the number of elements in Λ), and mw is the maximal ideal in
C[x] associated to w (i.e. mw := {f ∈ C[x]| f (w) = 0}).

Remark The independence in the above definition is equivalent to saying that the
intersection of the d hypersurfaces


m−1
j
aj wl = 0, wl ∈ Λ, l = 1, . . . , d, (3.11)
|j|=0

is an (N − d)-dimensional subspace of CN , where

j j j
wl = wl11 . . . wlnn , wl = (wl1 , . . . , wln ), and j = (j1 , . . . , jn ).

In fact,     
mwl Cm−1 [x] = mwl Cm−1 [x] ,
wl ∈Λ wl ∈Λ

and ⎧ ⎫
 ⎨  
m−1 ⎬
 j
mwl Cm−1 [x] = (aj )|j|=0
m−1
∈ CN  aj wl = 0 .
⎩ ⎭
|j|=0

So we have
⎧ ⎫
  ⎨  

m−1
j
⎬
mwl Cm−1 [x] = (aj )|j|=0
m−1
∈ CN  aj wl = 0 .
⎩ ⎭
wl ∈Λ wl ∈Λ |j|=0

 
 
This implies that ρ := dim mwl Cm−1 [x] is the dimension of the linear
wl ∈Λ
space formed by the solutions of the linear algebraic equations (3.11) in the variables
(aj )|j|=0
m−1
. Hence, ρ = N − d indicates that the dimension of the linear space formed
by the solutions of (3.11) is N − d. In addition, the solutions of each equation in
(3.11) forms a hyperplane in CN , and so ρ = N − d implies that the intersection of
the d hypersurfaces in (3.11) is an (N − d)-dimensional subspace.

Theorem 3.2 Assume that the polynomial differential system (3.1) has p invariant
algebraic hypersurfaces, q exponential factors, and r independent singularities with
respect to Cm−1 [x] which are not on the invariant algebraic hypersurfaces. The
following statements hold.
(a) If p + q ≥ N − r + 1, system (3.1) has a Darboux first integral.
(b) If p + q ≥ N − r + n, system (3.1) has a rational first integral.
100 3 Darboux and Liouvillian Integrability

(c) If p + q ≥ N − r, and the independent singularities are all located on div(P) =

0, system (3.1) has either a Darboux first integral or a Darboux Jacobian mul-
tiplier.

Proof Let w1 , . . . , wr be the independent singularities of the vector field XP with

respect to Cm−1 [x], let fi (x) and Ej (x) be respectively the Darboux polynomials and
the exponential factors of XP , i = 1, . . . , p, j = 1, . . . , q, and ki (x) and Lj (x) be the
respective associated cofactors. Since

XP (fi ) = ki fi , XP (Ej ) = Lj Ej ,

it follows from the assumption of the theorem that

ki (ws ) = 0, Lj (ws ) = 0, i ∈ {1, . . . , p}, j ∈ {1, . . . , q}, s ∈ {1, . . . , r}.

The independence of the r singularities implies that

k1 (x), . . . , kp (x), L1 (x), . . . , Lq (x) (3.12)

generate an at most (N − r)-dimensional linear subspace of Cm−1 [x]. Then the next
proofs of (a) and (b) are the same as those of Theorem 3.1 (a) and (b).
(c) If the cofactors in (3.12) are linearly dependent, by Proposition 3.5 (a) system
(3.1) has a Darboux first integral. 
If the cofactors in (3.12) are linearly independent, since div(P) ∈ mw ∩
w∈Λ
Km−1 [x], it follows that div(P) must be a linear combination of (3.12). We get from
Proposition 3.5 (b) that system (3.1) has a Darboux Jacobian multiplier.
This completes the proof of the theorem. 

3.2.2 Taking into Account Algebraic Multiplicities

Next we consider the extension of the Darboux theory of integrability taking into
account the algebraic multiplicity of the invariant algebraic curves or hypersurfaces.
Let Cd [x] be the linear space formed by polynomials with complex  coefficients

n+d
of degree no more than d. It is clear that Cd [x] has dimension κ := . Let
n
υ1 , . . . , υκ be a basis of Cd [x], and define
⎛ ⎞
υ1 υ2 ... υκ
⎜ XP (υ2 ) . . . XP (υκ ) ⎟
XP (υ1 )
⎜ ⎟
Mκ := ⎜ .. .. .. .. ⎟, (3.13)
⎝ . . . . ⎠
κ−1 κ−1 κ−1
XP (υ1 ) XP (υ2 ) . . . XP (υκ )
3.2 Generalization of the Classical Darboux Theory of Integrability 101

where XP+1 (υs ) = XP (XP (υs )), s = 1, . . . , κ. By definition det Mκ is a dth extac-
tic polynomial, whereas det Mκ (x) = 0 is a dth extactic surface of the vector field
XP .

Remark The extactic surface is independent of the choice of the basis of Cd [x]. This
claim follows from the fact that if f = 0 is an invariant algebraic surface of degree d
of the vector field XP , then f divides det Mκ . This is because if we take f instead of
some υi in (3.13), then f divides the whole column on which f is located.
An invariant algebraic surface f = 0 of degree d of the vector field XP has alge-
braic multiplicity  if det Mκ ≡ 0 and  is the maximum positive number such that
f  divides det Mκ , and f = 0 has no defined algebraic multiplicity if det Mκ ≡ 0.

Remark The notion of an ‘extactic point’ was introduced by Arnold [14] in his
study of planar curves. The matrix Mκ had already appeared in the works of M.N.
Lagutinskii (1871–1915), see Dobrovol’skii et al. [127]. We refer to Pereira [357]
for a modern definition and explanation of extactic curves.
Christopher et al. [106] studied the algebraic multiplicity of the invariant alge-
braic curves of planar polynomial vector fields with the help of extactic curves, and
proved its equivalence with the infinitesimal multiplicity, integral multiplicity and
geometric multiplicity. Llibre and Zhang [302] discussed the algebraic multiplicity
of invariant algebraic hypersurfaces for higher-dimensional differential systems via
extactic surfaces, and made use of it to develop the Darboux theory of integrability.
The next result is an extension of the Darboux theory of integrability, taking into
account not only the number of invariant algebraic curves (surfaces) but also their
multiplicity.
Theorem 3.3 Assume that the polynomial vector field XP of degree m has p different
irreducible invariant algebraic surfaces f1 = 0, . . . , fp = 0.
(i) If one of the invariant algebraic surfaces has no defined algebraic multiplicity,
then the vector field XP has a rational first integral.
(ii) Assume that all the invariant algebraic surfaces fi = 0 have defined algebraic
multiplicity, denote by qi their multiplicities for i = 1, . . . , p, and assume that
the vector field XP restricted to each surface fi = 0 with multiplicity larger than
1 has no rational first integrals. The following statements hold.

p
(a) If qi ≥ N + 1, the vector field XP has a Darboux first integral, where N
i=1
is the number defined in Theorem 3.1.
p
(b) If qi ≥ N + n, the vector field XP has a rational first integral.
i=1

Remark Statement (i) is due to Pereira, see the second part of Theorem 3 of Pereira
[357], and an elementary proof in the two-dimensional case can be found in Theorem
5.3 of Christopher et al. [106]. Statement (ii) was also proved in [106] in the two-
dimensional case, and proved by Llibre and Zhang [302] in the higher-dimensional
case. For the proof of Theorem 3.3, we need the following result.
102 3 Darboux and Liouvillian Integrability

Theorem 3.4 Assume that the polynomial vector field XP has an irreducible invari-
ant algebraic surface f = 0 of degree d, and restricted to f = 0 the vector field XP
has no rational first integrals. Then f has algebraic multiplicity  if and only if XP
has  − 1 exponential factors of the form exp(gs /f s ), where gs is a polynomial of
degree at most s · m and gcd(gs , f ) = 1, s = 1, . . . ,  − 1.

Before proving Theorem 3.4 we would like to present the following remarks:
• If XP is a planar polynomial vector field, Theorems 3.4 and 3.3 hold without
the additional condition that the vector field XP restricted to f = 0 has no ratio-
nal first integrals. The two theorems are precisely the corresponding theorems of
Christopher et al. [106].
• For higher-dimensional polynomial differential systems the additional condition
is necessary, as the following example shows. The real polynomial vector field

ẋ = 1, ẏ = −y(y − 2z), ż = z(y − z)

has z = 0 as an invariant algebraic plane of multiplicity 2, but it has no exponential

factors associated to z = 0. The proof of this last claim can be found in Llibre and
Zhang [302, 305], so we omit the details. We can check that the vector field
restricted to z = 0 has a rational first integral x + 1/y.
To prove Theorem 3.4 we need the following lemma, which is a key point in the
proof of the theorem. Denote by C[x][ε] the ring of polynomials in ε with coefficients
in C[x], and by C[x][ε]/ εk
the set formed by elements of C[x][ε] modulo the ideal
generated by εk , i.e. εk
:= {εk p(x, ε)| p ∈ C[x][ε]}.

Lemma 3.2 Let X be a polynomial vector field of degree m, and let f = 0 be

an irreducible invariant algebraic hypersurface of degree d of X . The following
statements hold.
(a) If for any ε > 0 there exist

F = f0 + f1 ε + · · · + f−1 ε−1 ∈ C[x][ε]/ ε

,
LF = L0 + L1 ε + · · · + L−1 ε−1 ∈ C[x][ε]/ ε
,

with f0 = f , f1 , . . . , f−1 ∈ Cd [x], f1 not a multiple of f , and L0 , L1 , . . . , L−1 ∈

Cm−1 [x] such that
X (F) = FLF in C[x][ε]/ ε
, (3.14)

then X has  − 1 exponential factors of the form exp(gs /f s ) with gs coprime

with f and of degree at most sd for s = 1, . . . ,  − 1.
(b) If X restricted to f = 0 has no rational first integrals, the inverse of statement
(a) holds.

Lemma 3.2 was proved by Llibre and Zhang [302]. For planar polynomial vector
fields it was proved in [106] without the additional assumption of statement (b).
3.2 Generalization of the Classical Darboux Theory of Integrability 103

We note that for higher-dimensional differential systems the additional assumption

in statement (b) of Lemma 3.2 is necessary, as the next example shows. The system

ẇ = (y + z)w2 , ẏ = (1 + ay + y2 + yz)w, ż = z + bw2 − cwz + (yz + z2 )w,

(3.15)
when a = 0, has the exponential factors

E1 = exp(y/x) and E2 = exp (1/2 + y2 /2 + wz)/w2 ,

and the associated cofactors

LE1 = 1 and LE2 = bw − cz.

Let X1 be the vector field associated to system (3.15), and choose

F = f0 + f1 ε and LF = L0 + L1 ε,

with
f0 = w, L0 = w(y + z), f1 = y and L1 = LE1 .

We have
X1 (F) = FLF in C[w, y, z][ε]/ ε2
.

In addition, direct calculations show that there are no f2 ∈ C[w, y, z] and L2 ∈

C1 [w, y, z] such that F = f0 + f1 ε + f2 ε2 and LF = L0 + L1 ε + L2 ε2 satisfy X1 (F) =
FLF in C[w, y, z][ε]/ ε3
.
Proof of Lemma 3.2 (a) Denote by C(x){ε} the ring of all series in ε with coefficients
in C(x), the field of rational functions in x, and denote by Qf the field formed by all
quotients of two elements in C[x]/ f
, where f
is the ideal generated by f in C[x].
Taking the logarithm of F and expanding it in Taylor series in ε gives

log F = log f0 + εg1 f0 −1 + · · · + ε−1 g−1 f0−(−1) in C(x){ε}/ ε

,

where
gs = (−1)s−1 s−1 f1s + f0 G s (f0 , . . . , fs ), (3.16)

with G s a homogeneous polynomial of degree s − 1 in f0 , . . . , fs . Obviously deg gs ≤

sd and gcd(gs , f ) = 1. Equating the coefficients of εs , s = 1, . . . ,  − 1, in the two
sides of the equality (3.14) we get the  − 1 exponential factors, exp(gs /f s ), s =
1, . . . ,  − 1, of X .
(b) Let Fs := exp(gs /f s ), s = 1, . . . ,  − 1, be the  − 1 exponential factors of X
with cofactors LFs .
For k = 2, statement (b) holds for F = f0 + f1 ε and LF = L0 + L1 ε with f0 = f ,
f1 = g1 , L0 = Lf and L1 = LF1 .
104 3 Darboux and Liouvillian Integrability

In order to apply induction we assume that for the κ − 1 (∈ {2, . . . ,  − 2}) expo-
nential factors exp(gs /f s ), s = 1, . . . , κ − 1, there exist

F (κ) = f0 + f1 ε + · · · + fκ−1 εκ−1 in C[x][ε]/ εκ

,
LF (κ) = L0 + L1 ε + · · · + Lκ−1 εκ−1 in C[x][ε]/ εκ
,

such that Eq. (3.14) holds in C[x][ε]/ εκ

, where fi ∈ Cd [x] and Ls ∈ Cm−1 [x] for
s = 0, 1, . . . , κ − 1, and gcd(f1 , f ) = 1.
For the κ exponential factors exp(gs /f s ), s = 1, . . . , κ, set

κ−1

LF∗ (κ) = L0 + L1 ε + · · · + Lκ−1 εκ−1 + Lκ∗ εκ , Lκ∗ = −f0−1 fs Lκ−s .
s=1

We can check that in C(x){ε}/ εκ+1

X (F (κ) ) = F (κ) LF∗ (κ) ,

log F (κ) = log f0 + g1 f0 −1 ε + · · · + gκ−1 f0−(κ−1) εκ−1 + gκ∗ f0−κ εκ ,

with gs , s = 1, . . . , κ − 1, and gκ∗ being of the form (3.16). These last two equalities
show that X (gκ∗ /f0κ ) = Lκ∗ , and consequently

κ−1

X (gκ∗ ) = κL0 gκ∗ − f0κ−1 fs Lκ−s . (3.17)
s=1

Since X (f1 ) = f1 L0 + f0 L1 and exp(gκ /f κ ) is an exponential factor of X , we

can prove that gκ /f1κ is a first integral of X when restricted to f = 0. This together
with the assumption forces gκ /f1κ to be a nonzero constant on f = 0, denoted by cκ .
Since gκ∗ has the expression (3.16), we have gκ∗ = (−1)κ−1 κ −1 f1κ on f = 0. These
two facts imply
gκ − (−1)κ−1 κcκ gκ∗ = 0 on f = 0.

Since f0 = f is irreducible, it follows from Hilbert’s Nullstellensatz (see e.g. Artin [20])
∗
that there exists a polynomial gκ−1 such that

gκ = (−1)κ−1 κcκ (gκ∗ + f0 gκ−1

∗
).

∗
Obviously, deg gκ−1 ≤ (κ − 1)d. Since X (gκ ) = κL0 gκ + LFκ f κ , we get from (3.17)
that
 κ−1 

∗
X (gκ−1 ∗
) = (κ − 1)L0 gκ−1 + f0κ−2 fs Lκ−s + (−1)κ−1 f0 LFκ /(κcκ ) .
s=1
3.2 Generalization of the Classical Darboux Theory of Integrability 105

∗
Consequently, X (gκ−1 /f1κ−1 ) = 0 on f = 0. So there exists a constant cκ−1 and a
∗
polynomial gκ−2 of degree no more than (κ − 2)d such that

∗
gκ−1 = (−1)κ−2 (κ − 1)cκ−1 gκ−1 + f0 gκ−2
∗
.

Furthermore, we have

∗ ∗
X (gκ−2 ) = (κ − 2)L0 gκ−2 +
κ−1 

κ−3 κ−1 κ−2
f0 fs Ll−s + (−1) f0 LFκ /(κcκ ) − (−1) (κ − 1)cκ−1 f0 Lκ−1 .
s=1

Repeating the above process, we get that for s = 3, . . . , κ − 1

∗
gκ+1−s = (−1)κ−s (κ + 1 − s)cκ+1−s gκ+1−s + f0 gκ−s
∗
,

∗
with cκ+1−s a constant and gκ−s a polynomial of degree at most (κ − s)d satisfying
⎛
κ−1

X ∗
(gκ−s ) = (κ ∗
− s)L0 gκ−s + f0κ−1−s ⎝ fj Lκ−j + (−1)κ−1 f0 LFκ /(κcκ ) −
j=1

(−1)κ−2 (κ − 1)cκ−1 f0 Lκ−1 − · · · − (−1)κ−s (κ + 1 − s)cκ+1−s f0 Lκ+1−s .

Set

F κ+1 = f0 + εf1 + · · · + εκ fκ , LF κ+1 = L0 + εL1 + · · · + εκ Lκ ,

with

fκ = gκ∗ ,
Lκ = (−1)κ−1 LFκ /(κcκ ) − (−1)κ−2 (κ − 1)cκ−1 Lκ−1 − · · · − (−1)2c2 L2 .

We can check that

X (F κ+1 ) = F κ+1 LF κ+1 in C[x][ε]/ εκ+1

.

This proves statement (b) by induction, and consequently the lemma. 

Proof of Theorem 3.4. The main idea of the proof follows from that of Christo-
pher et al. [106].
Sufficiency. By Lemma 3.2 there exist

F = f0 + f1 ε + · · · + f−1 ε−1 in C[x][ε]/ ε

,
106 3 Darboux and Liouvillian Integrability

of degree d with f0 = f and gcd(f1 , f0 ) = 1, and

LF = L0 + L1 ε + · · · + L−1 ε−1 in C[x][ε]/ ε

,

of degree at most m − 1, such that

X (F) = FLF in C[x][ε]/ ε

.
 
n+d
Since Cd [x] is an R := -dimensional vector space, we choose one of its
n
bases, say {f , v2 , . . . , vR }. Then we can replace F by an equivalent one such that
f1 , . . . , f−1 belong to the span of v2 , . . . , vR .
Set

G = f0 F −1 in C(x){ε}/ ε
,
G s = GF (s) with F (s) = X s (F),

and denote by [G]i the coefficient of εi in G. Then

[G]−1 = (−f1 )−1 in C[x]/ f

, X (G) = (L0 − LF )G,

and for s = 0, 1, . . . , R − 1

X (G s ) = G s+1 + (L0 − LF )G s , (3.18)

−1

[G s ]−1 = [G]−1 X s (f0 ) + [G]−1−i X s (fi ). (3.19)
i=1

Moreover, f0 divides all [G s ]−1 for s = 0, 1, . . . , R − 1, which follows from G 0 =

f0 and f0 dividing all G s for s = 0, 1, . . . , R − 1, obtained by recursive calculations
from (3.18).
Rewriting (3.19) in the basis f0 , v2 , . . . , vR gives


R
[G s ]−1 = [G]−1 X s (f0 ) + αi X s (vi ), s = 0, 1, . . . , R − 1,
i=2


−1
where αi = [G]−1−j αji with αji the coefficient of vi in fj . From the construction of
j=1
[G]−1 and the extactic polynomial det MR , it follows that for some r ∈ N, f r divides
det MR if and only if it divides ([G]−1 )R det MR . Adding the product of αi /[G]−1
with the ith column of the matrix [G]−1 MR to its first column, the sth component
of the first column of the resulting matrix is [G s−1 ]−1 and the other columns are the
3.2 Generalization of the Classical Darboux Theory of Integrability 107

same as those of [G]−1 MR . The first column of this last matrix is divisible by f0 .
This proves that f has algebraic multiplicity at least .
Necessity. We will prove it by induction.
For  = 2, by the definition of algebraic multiplicity it follows that f 2 divides
det MR . We will prove that X has the exponential factor exp(g1 /f ).
We choose the basis {f , v2 , . . . , vR } of Cm [x] in order to define the extactic hyper-
surface, and define (X + Lf )r = (X + Lf )(X + Lf )r−1 , with Lf acting as a con-
stant operator and non-commutative with X . Then we have X r (f ) = f (X + Lf )r 1,
and consequently
det MR = f det MR∗ ,

where the ith component of the first column of MR∗ is given by (X + Lf )i−1 1, and the
other columns are the same as MR , denoted by Vi for i = 2, . . . , R. Since f divides
det MR∗ by the assumption, we have

det MR∗ = 0, in C[x]/ f

.

Hence the columns of MR∗ are linearly dependent on V (f ) = {f = 0} ⊂ Cn . By the

choice of v2 , . . . , vR , there exist α1 , . . . , αR ∈ Qf not all zero such that for s =
0, 1, . . . , R − 1


R
As := α1 (X + Lf )s 1 + αj X s (vj ) = 0 on f = 0. (3.20)
j=2

We claim that α1 = 0. If not, the last R − 1 columns of MR∗ are linearly depen-
dent. Assume that their rank is q with 1 < q < R − 1, and that {Vi1 , . . . , Viq } ⊂
{V2 , . . . , VR } are linearly independent. The fact that q > 1 follows from the assump-
tion that X (vi ) does not have the factor f for i ∈ {2, . . . , R}. For any Viq+1 ∈
{V2 , . . . , VR } \ {Vi1 , . . . , Viq }, we have


q+1
Bs := αij X s (vij ) = 0 on f = 0, s = 0, 1, . . . , R − 1,
j=1

with αiq+1 = 1. It follows that


q
X (Bs ) − Bs+1 = X (αij )X s (vij ) = 0 on f = 0, s = 0, 1, . . . , R − 2.
j=1

By the linear independency of Vi1 , . . . , Viq we must have X (αij ) = 0 on f = 0 for

j = 1, . . . , q. So the assumption forces αij = constant on f = 0, denoted by cij . Then
we have
ci1 vi1 + · · · + ciq viq + viq+1 = 0 on f = 0.
108 3 Darboux and Liouvillian Integrability

We get from Hilbert’s Nullstellensatz that f divides ci1 vi1 + · · · + ciq viq + viq+1 . This
contradicts the fact that deg vij ≤ d and {f , v2 , . . . , vR } is a basis of Cd [x]. Hence we
must have α1 = 0. The claim follows.
Since α1 = 0 on f = 0 and we are working in Qf , without loss of generality we
can take α1 = 1. Then for s = 0, 1, . . . , R − 1 we have


q
As := (X + Lf )s 1 + αij X s (vij ) = 0 on f = 0, (3.21)
j=1

with at least one of the αij not zero, say αir = 0 for some r ∈ {1, . . . , q}. Then for
s = 0, 1, . . . , R − 2 we have


q
(X + Lf )As − As+1 = (X (αij ) + Lf αij )X s (vij ) = 0 in Qf .
j=1

It follows that X (αij ) + Lf αij = 0 in Qf . So we have X (αij /αir ) = 0 in Qf . This

implies that there exist constants cij for j = r such that αij = cij αir on f = 0 for
j ∈ {1, . . . , r − 1, r + 1, . . . , q}. Hence from Eq. (3.21) with s = 0 and 1 we have


R 
R
1 + αir cj vj ≡ 0 and Lf + αir cj X (vj ) ≡ 0 on f = 0,
j=2 j=2

where cir = 1 and cj = 0 for j ∈

/ {i1 , . . . , iq }. Set


R
g1 = − cj vj .
j=2

We have g1 = 0 and X (g1 ) − g1 Lf ≡ 0 on f = 0. By Hilbert’s Nullstellensatz and

the fact that f is irreducible, there exists a polynomial L1 of degree necessarily no
greater than m − 1 such that

X (g1 ) = g1 Lf + fL1 .

This proves that exp(g1 /f ) is an exponential factor of X with cofactor L1 . So the

necessary part follows for  = 2.
Assume that for 2 < s <  such that f s divides det MR we have s − 1 expo-
nential factors Fi := exp(gi /f i ) with cofactors LFi of degree at most m − 1 for
i = 1, . . . , s − 1, where gi is a polynomial of degree no more than id and gcd
(gi , f ) = 1. By Lemma 3.2 there exist

F = f0 + εf1 + · · · + εs−1 fs−1 ,

LF = L0 + εL1 + · · · + εs−1 Ls−1 ,
3.2 Generalization of the Classical Darboux Theory of Integrability 109

such that X (F) = FLF in C[x][ε]/ εs

, where f0 = f , gcd(f1 , f ) = 1, fi ∈ Cd [x] for
i = 1, . . . , s − 1, and L0 = Lf . Working in a similar way as in the proof of Lemma
3.2 we can rewrite F in the basis {f , v2 , . . . , vR } as


R
F = f0 + ε qj (ε)vj ,
j=2


s−2
j 
R
j
where qj (ε) = εi qi . Observe that fi+1 = qi vj .
i=0 j=2
s+1
Assume that f divides det MR . We will prove the existence of s exponential
factors. Set

G = f1 + εf2 + · · · + εs−2 fs−1 , G r = G −1 G (r) , r = 0, 1, . . . , R − 1.

Recall that G (r) = X r (G). Then

G 0 = 1, G 1 = LF , X (G) = GLF ,

and
X (G r ) = G r+1 − G r LF in C[x][ε]/ f , εs−1
,

where f , εs−1
is the ideal generated by f and εs−1 in C[x, ε]. Moreover, we have

[G r+1 ]q = X ([G r ]q ) + L0 [G r ]q + L1 [G r ]q−1 + · · · + Lq [G r ]0 in C[x][ε]/ f , εs−1

.

Recall that [G r+1 ]q denotes the coefficient of εq of G r+1 . Set

H r = F −1 F (r) , S r = f0s H r .

Then

H 0 = 1, H 1 = LF , H r+1 = X (H r ) + LF H r , X (S r ) = S r+1 + (sL0 − LF )S r ,

and f0s divides S r for all r ∈ {0, 1, . . . , R − 1}. It follows easily that

G r+1 − H r+1 = X (G r − H r ) + LF (G r − H r ) in C[x][ε]/ f , εs−1

.

By induction we can prove that

G r+1 − H r+1 = 0 in C[x][ε]/ f , εs−1

,

for all r ∈ {0, 1, . . . , N − 2}, because G 0 − H 0 = 0 and G 1 − H 1 = 0.

110 3 Darboux and Liouvillian Integrability

Set T = f0s F −1 , then [T ]s−1 = (−f1 )s−1 = 0 in C[x]/ f

. We get from S r = TF (r)
that

s−1
[S r ]s−1 = [T ]s−1 f0(r) + [T ]s−1−i fi(r) .
i=1

As in the proof of the sufficient part, we have that f0s divides S r . Set S r = f0s Dr .
Writing [S r ]s−1 in the basis {f , v2 , . . . , vR } yields


R
[S ]s−1 =
r
[T ]s−1 f0(r) + βj X r (vj ),
j=2


s−1
j
where βj = [T ]s−1−i qi−1 .
i=1
Let MR∗ be an R × R matrix whose first column has its ith component given by
[S ]s−1 and the other columns are the same as the last R − 1 columns of MR .
i−1

Then for each j ∈ N, f j divides det MR∗ if and only if it divides det MR , because
det MR∗ = [T ]s−1 det MR . Consider a matrix MR which is different from MR∗ only in
the first column where we replace [S r ]s−1 of MR∗ by [Dr ]s−1 .
Since f s+1 divides det MR , it follows from the assumption that f divides det MR .
So the columns of MR are linearly dependent in Qf for all x ∈ C. Working in a
similar way as in the proof of Lemma 3.2, we can assume without loss of generality
that the last R − 1 columns of MR are linearly independent, and that there exist
α2 , . . . , αR ∈ Qf such that


R
C r := [Dr ]s−1 + αi X r (vi ) = 0 in Qf .
i=2

It follows from X ([Dr ]s−1 ) = −[LF Dr ]s−1 + [Dr+1 ]s−1 that

C r+1 − (X + L0 )C r (3.22)

s−1 
R
= Li [Dr ]s−1−i − (X (αi ) + L0 αi )X r (vi ) = 0 in Qf ,
i=1 i=2

for r = 0, 1, . . . , R − 2. Equation (3.22) is equivalent to


R
f1 [(LF − L0 )G r ]s−1 − X (f1 αi )X r (vi ) = 0 in Qf , (3.23)
i=2

where we have used the facts that

3.2 Generalization of the Classical Darboux Theory of Integrability 111


s−1
[Dr ]i = [H r ]i = [G r ]i and Li [Dr ]s−1−i = [(LF − L0 )G r ]s−1 .
i=1

Hence we have


R 
R
X (f1 αi )X r (vi ) = [(LF − L0 )f1 G −1 qi (ε)X r (vi )]s−1 in Qf .
i=2 i=2

Rearranging this last equation we get


R

X (f1 αi ) − [(LF − L0 )f1 G −1 qi (ε)]s−1 X r (vi ) = 0 in Qf ,
i=2

for r = 0, 1, . . . , R − 2. It is equivalent to

X (f1 αi ) = [(LF − L0 )f1 G −1 qi (ε)]s−1 in Qf , i = 2, . . . , R.

In addition, since X (G) = GLF and X (f1 ) = L0 f1 + L1 f0 we have for pi (ε) :=

qi (ε) − q0i
X (G −1 f1 pi (ε)) = G −1 f1 (L0 − LF )pi (ε) in Qf .

These last two equations yield

X (f1 αi + [G −1 f1 pi (ε)]s−1 ) = [G −1 f1 (LF − L0 )q0i ]s−1 in Qf . (3.24)


R
Since f1 = q0i vi , there must exist some q0i = 0, because f1 and f are relatively
i=2
coprime. We assume without loss of generality that q0R = 0. Then Eq. (3.24) with
i = R minus itself with i = R times q0i /q0R gives
      
q0i −1 q0i R
X f1 αi − R αR + G f1 p (ε) − R p (ε)
i
= 0 in Qf .
q0 q0 s−1

So there exist constants csi such that

  
csi q0i −1 q0i R
αi = + R αR − G p (ε) − R p (ε)
i
in Qf .
f1 q0 q0 s−1


R
Set fs = − csi vi . Then for r = 0, 1, . . . , R − 1 we have
i=2
112 3 Darboux and Liouvillian Integrability


R
[Dr ]s−1 = − αi X r (vi )
i=2
(r)   
fs αR (r) −1 (r) pR (ε) (r)
= − f + G G − f in Qf .
f1 q0R 1 q0R 1 s−1

These last equations for r = 0 and r = 1 can be written respectively as

  
fs αR −1 pR (ε)
0 = − R f1 + G G − R f1 ,
f1 q0 q0 s−1
  
X (fs ) αR pR (ε)
Ls−1 = − R L0 f1 + G −1 GLF − [GLF ]s−1 εs−1 − R L0 f1 ,
f1 q0 q0 s−1

in Qf , where we have used the facts that [D]s−1 = 0 and [D1 ]s−1 = Ls−1 , because
S 0 = f0s and S 1 = f0s LF . These last two equations yield
 
f1 Ls−1 = X (fs ) − L0 fs + f1 (LF − L0 ) − G −1 f1 [LF G]s−1 εs−1 s−1 on f = 0.

Since [LF − L0 ]s−1 = Ls−1 and [G −1 fs [LF G]s−1 εs−1 ]s−1 = [LF G]s−1 we have

X (fs ) = L0 fs + L1 fs−1 + · · · + Ls−2 f2 + Ls−1 f1 on f = 0.

So there exists an Ls ∈ Cm−1 [x] such that

X (fs ) = L0 fs + L1 fs−1 + · · · + Ls−1 f1 + Ls f0 .

Set
F ∗ = f0 + εf1 + · · · + εl fs , L ∗ = L0 + εL1 + · · · + εl Ll .

We have X (F ∗ ) = F ∗ LF ∗ in C[x][ε]/ εs+1

.
By Lemma 3.2 we get s exponential factors exp(gr /f r ) for r = 1, . . . , s. Hence
by induction we have proved that if the irreducible invariant hypersurface f = 0
of the vector field X has algebraic multiplicity , then the vector field has  − 1
exponential factors. This proves the necessary part.
This completes the proof of Theorem 3.4. 
Proof of Theorem 3.3. (i) This can be found in the proof of the second part of
Theorem 3 of Pereira [357].
(ii) Let fi (x) = 0 for i = 1, . . . , p be the irreducible invariant algebraic hypersurfaces
with algebraic multiplicity qi . Theorem 3.4 shows that for each fi we have qi − 1
p
exponential factors exp(gij /f j ), j = 1, . . . , qi − 1. Thus we have a total of K := qi
i=1
cofactors of degree at most m − 1 associated to the p invariant algebraic hypersurfaces
and K − p exponential factors.
3.2 Generalization of the Classical Darboux Theory of Integrability 113

The remainder of the proof is the same as that of Theorem 3.1, so the details are
omitted. This finishes the proof of the theorem. 

3.2.3 Taking into Account the Multiplicity of the Hyperplane

at Infinity

The Darboux theory of integrability was extended by Llibre and Zhang [303] to
take into account the multiplicity of the invariant line or surface at infinity. Here we
consider the extension in Rn . To use the hyperplane at infinity of Rn to study the
integrability of the vector field XP , we need the Poincaré compactification of XP ,
which was introduced by Cima and Llibre in [113]. !
In the local coordinate chart U1 = x = (1, x2 , . . . , xn )| (x2 , . . . , xn ) ∈ Rn−1 , the
transformation
1 y2 yn
x1 = , x2 = , . . . , xn = (3.25)
z z z

brings the vector field XP to

∂ ∂ ∂
X P = −zP1 (y) + (P2 (y) − y2 P1 (y)) + · · · + (Pn (y) − yn P1 (y)) ,
∂z ∂y2 ∂yn

where Pi = zm Pi (1/z, y2 /z, . . . , yn /z), i = 1, . . . , n, y = (z, y2 , . . . , yn ).

Since z = 0 is an invariant hyperplane of the vector field X P , the hyperplane
at infinity of Rn in the chart U1 is z = 0 of the vector field X P . The hyperplane at
infinity of the vector field XP has algebraic multiplicity  if z = 0 as an invariant
hyperplane of the vector field X P has algebraic multiplicity . By definition the
hyperplane at infinity of the vector field XP has no defined algebraic multiplicity
if z = 0 as an invariant hyperplane of the vector field X P has no defined algebraic
multiplicity.

Remark The above definition of the algebraic multiplicity of the hyperplane at infin-
ity was first introduced by Llibre and Zhang [303]. Schlomiuk and Vulpe [386] also
gave a definition on the algebraic multiplicity of the hyperplane at infinity using a
limit process. In fact, these two definitions are equivalent, but Llibre and Zhang’s
definition is convenient for computations.
Similar to Lemma 3.4, we have the next result, see [303].

Lemma 3.3 Assume that the vector field X P restricted to z = 0 has no rational first
integrals. Then z = 0 has algebraic multiplicity  if and only if X P has  − 1 expo-
nential factors exp(g s /zs ), s = 1, . . . ,  − 1, where g s ∈ Cs [y] cannot be divided
by z.
114 3 Darboux and Liouvillian Integrability

Remark
• For planar differential systems, it is not necessary in Lemma 3.3 that the vector
field X P restricted to z = 0 has no rational first integrals.
• The next example shows that the additional assumption in Lemma 3.3 is necessary
for higher-dimensional differential systems, see [303, Proposition 7]. The three-
dimensional polynomial differential system

ẋ = (−1 + cx − y − bwx)w − x 2 ,
ẏ = (x + (a + c)y − bwy)w − xy, (3.26)
ẇ = w(cw − x − bw ) 2

has an invariant algebraic plane w = 0 of multiplicity 4, where a, b, c are real

constants. But the exponential factors of system (3.26) associated to w = 0 are
either only E1 = exp((u1 x + u2 y)/w) if a = 0, with u1 , u2 constants and not all
zeros, or only E1 and E2 = exp((w + x 2 /2 + y2 /2)/w2 ) if a = 0. Note that system
(3.26) restricted to w = 0 has the rational first integral H = y/x.
Using Lemma 3.3 Llibre and Zhang [303, Theorem 1] extended the Darboux
theory of integrability, taking into account the multiplicity of the invariant hyperplane
at infinity.

Theorem 3.5 Assume that the polynomial vector field XP is of degree m in Rn ,

and has pairwise distinct irreducible invariant algebraic hypersurfaces fi = 0, i =
1, . . . , p.
(i) If one of the invariant algebraic hypersurfaces or the hyperplane at infinity has
no defined algebraic multiplicity, the vector field XP has a rational first integral.
(ii) Assume that each invariant algebraic hypersurface fi = 0 has an algebraic mul-
tiplicity qi , i = 1, . . . , p, and that the hyperplane at infinity has algebraic mul-
tiplicity . If the vector field XP restricted to the hyperplane at infinity or any
invariant hypersurface with multiplicity larger than 1 has no rational first inte-
grals, the following statements hold.

p
(a) If qi +  ≥ N + 2, the vector field XP has a Darboux first integral, where
i=1
N was defined in Theorem 3.1.

p
(b) If qi +  ≥ N + n + 1, the vector field XP has a rational first integral.
i=1

Remark
• For two-dimensional polynomial vector fields, the additional condition in Theorem
3.5 on the nonexistence of rational first integrals of the vector field restricted to
the invariant algebraic curves including the line at infinity is not necessary.
3.2 Generalization of the Classical Darboux Theory of Integrability 115

• For higher-dimensional polynomial vector fields, the additional condition in

Theorem 3.5 is necessary, see the remark after Lemma 3.3.
• There does not exist a version of Theorem 3.5 in Cn . This is an interesting question
to study.

3.2.4 On Nonautonomous Differential Systems

via the Wronskian Matrix

The Darboux theory of integrability was recently extended to nonautonomous differ-

ential systems. In 2009 Llibre and Pantazi [262] extended the Darboux–Jouanolou
integrability theorem to planar nonautonomous differential systems, which are poly-
nomial differential systems in the dependent variables where the coefficients are
smooth functions in the independent variable. Later, Blázquez-Sanz and Pantazi
[35] and Giné, Grau and Llibre [179] further extended the results of Llibre and Pan-
tazi [262] to higher-dimensional differential systems. Next we introduce the results
of these two papers.
Consider the nonautonomous polynomial differential system

ẋ = P(t, x), (t, x) ∈ U × Kn , (3.27)

where U is a connected subset of C, K = C or R, and P(t, x) = (P1 (t, x), . . . ,

Pn (t, x)) is of the form


m
Ps (t, x) = psi (t)x i , s = 1, . . . , n, (3.28)
|i|=0

where psi : U −→ C is a C  function for some  ∈ N ∪ {∞, ω}, x i = x1i1 · · · · · xnin for
x = (x1 , . . . , xn ) and i = (i1 , . . . , in ) ∈ Zn+ , and |i| = i1 + · · · + in . We say system
(3.27) has degree m if there exists some s ∈ {1, . . . , n} and i ∈ Zn+ satisfying |i| = m
such that psi (t) = 0. In what follows we write Ps (t, x) ∈ C  (U, C)[x], the ring of
polynomials in x with C  smooth coefficients in terms of t, and we call the elements
of C  (U, C)[x] generalized polynomials. Of course, if psi ∈ C system (3.27) is an
autonomous (polynomial) differential system. We denote by XN the vector field
associated to the nonautonomous differential system (3.27), i.e.

XN = ∂t + P1 (t, x)∂x1 + · · · + Pn (t, x)∂xn .

A function f (t, x) ∈ C  (U, C)[x] is a Darboux invariant polynomial of system

(3.27) if there exists a k(t, x) ∈ C  (U, C)[x] such that

XN (f ) = kf .
116 3 Darboux and Liouvillian Integrability

The function k(t, x) is called a cofactor of the Darboux invariant polynomial f (t, x).
The hypersurface {(t, x) ∈ U × C n | f (t, x) = 0} (or simply, f (t, x) = 0) is called a
polynomial invariant hypersurface.
Let g, h ∈ C  (U, C)[x] and F = exp(g/h). If there exists a L(t, x) ∈ C  (U, C)
[x] such that
XN (F) = LF,

we call F an exponential factor, and L its associated factor. For f1 , . . . , fr , g, h ∈

C  (U, C)[x], the function
f11 · · · · · frr exp(g/h)

is called a generalized Darboux function, where 1 , . . . , r ∈ C.

Let O be an open subset of U × Cn whose closure is the full space. A smooth
function H : O −→ C is a first integral of system (3.27) if XN (H) ≡ 0 on O. In
particular, if a first integral H of system (3.27) is a generalized Darboux function, it is
called a generalized Darboux first integral of system (3.27). If the first integral is of
the form H = G(t, x)/F(t, x) with G, F ∈ C  (U, C)[x] for some  ∈ N ∪ {∞, ω},
we call it a generalized rational first integral. If the first integral H ∈ C  (U, C)[x]
for some  ∈ N ∪ {∞, ω}, we call it a generalized polynomial first integral.
A smooth function J : O −→ C is a Jacobian multiplier of system (3.27) if
XN (J) = −divXN J on O, where divXN = ∂x1 P1 + · · · + ∂xn Pn . If a Jacobian mul-
tiplier J is a generalized Darboux function, it is called a generalized Darboux Jaco-
bian multiplier of system (3.27).
In this subsection the extension of the Darboux theory of integrability will depend
on the Wronskian matrix of the cofactors of the Darboux invariant polynomials and
of the exponential factors. Let K1 (t, x), . . . , Kr (t, x) ∈ C  (U, C)[x] be cofactors of
the Darboux invariant polynomials and of the exponential factors of system (3.27).
Define the Wronskian matrix of these cofactors as
⎛ ⎞
K1 K2 ... Kr
⎜ XN (K1 ) XN (K2 ) . . . XN (Kr ) ⎟
⎜ ⎟
W = W (K1 , . . . , Kr ) := ⎜ .. .. .. .. ⎟, (3.29)
⎝ . . . . ⎠
XNr−1 (K1 ) XNr−1 (K2 ) . . . XPr−1 (Kr )

where XN+1 (Ks ) = XN (XN (Ks )), s = 1, . . . , r.

The Wronskian W = detW of K1 , . . . , Kr will play a key role in the extension of
the Darboux theory of integrability. Let K = {K1 , . . . , Kr } be a subset of C  (U, C)[x],
and let Ks be a subset of K which has s (≤r) elements. Denote by W(Ks ) the Wron-
skian of Ks . The notation Ws = 0 will mean that W(Ks ) ≡ 0 for all t ∈ U and
all subsets Ks of K, and Ws = 0 will mean that there exists a subset Ks ⊂ K such
that W(Ks ) = 0 on a full Lebesgue measure subset of U. The set K satisfies con-
dition W∗ if there exists an s ∈ {2, 3, . . . , r} such that Ws = 0, and Wj = 0 for
j = 1, . . . , s − 1.
3.2 Generalization of the Classical Darboux Theory of Integrability 117

The next result is an improvement of that of Llibre and Pantazi [262], which
characterizes the C-linearly independence of the polynomials in C  (U, C)[x].
Lemma 3.4 For K1 , . . . , Kr ∈ C  (U, C)[x] \ {0}, with the coefficients being ele-
mentary functions or analytic functions, the following statements hold.
(a) If there is a (t0 , x0 ) ∈ U × Cn such that W(K)|(t0 ,x0 ) = 0, then K1 , . . . , Kr are
linearly independent over C.
(b) If the set K = {K1 , . . . , Kr } satisfies the condition W∗ for some s ∈ {2, . . . , r},
then either system (3.27) has a generalized rational first integral, or there exists
a subset of K of s elements which are linearly dependent over C.
Proof (a) Let c1 , . . . , cr ∈ C such that

c1 K1 + · · · + cr Kr ≡ 0 on U × Cn .

Letting the vector field XN act on this last equation r − 1 times, we get a system
of linear equations in c1 , . . . , cn with coefficient matrix equal to the Wronskian
matrix of K1 , . . . , Kr . Then under the assumption of the statement we obtain that
c1 = · · · = cn = 0. This proves the statement.
(b) By the assumption of the statement we assume without loss of generality that
K1 , . . . , Ks satisfy W(K1 , . . . , Ks−1 ) = 0 and W(K1 , . . . , Ks ) = 0. Then there exist
functions c1 (t, x), . . . cs−1 (t, x) such that

Ks = c1 (t, x)K1 + · · · + cs−1 (t, x)Ks−1 , (3.30)

where Kj = (Kj , XN (Kj ), . . . , XNs−1 (Kj ))τ , j = 1, . . . , s. Since there exists at least
one (t0 , x0 ) ∈ U × Cn such that W(K1 , . . . , Ks−1 )(t0 , x0 ) = 0, and
W(K1 , . . . , Ks−1 ) is a generalized polynomial with coefficients being either ele-
mentary or analytic, it follows that {(t, x) ∈ U × Cn | W(K1 , . . . , Ks−1 )(t, x) = 0} is
a surface of dimension at most n. So there exists an open and dense subset, say O,
of U × Cn such that W(K1 , . . . , Ks−1 )(t, x) = 0 on O.
Solving the first s − 1 equations of (3.30) on O using Cramer’s rule, we get a
unique solution c(t, x) = (c1 (t, x), . . . , cs−1 (t, x))τ . Clearly all the numerators and
denominators of the cj (t, x)’s, j ∈ {1, . . . , s − 1}, belong to C k (U, C) × Cn for some
k ∈ {∞, ω}. That is, all cj (t, x)’s are generalized rational functions with either ele-
mentary or analytic coefficients, and they are well defined on O.
Letting the vector field XN act on the jth equations of (3.30) for j = 1, . . . , s − 1,
and subtracting the resulting equation from the j + 1th equation of (3.30), we get

W (K1 , . . . , Ks−1 )XN (c(t, x)) = 0,

where XN (c(t, x)) = (XN (c1 (t, x)), . . . , XN (cs−1 (t, x)))τ . Since W (K1 , . . . ,
Ks−1 )(t, x) = 0 on O, we must have XN (c(t, x)) = 0 on O. Here we have two cases.
• c(t, x) is a constant vector. It follows from (3.30) that K1 , . . . , Ks are C-linearly
independent.
118 3 Darboux and Liouvillian Integrability

• c(t, x) is not a constant vector. We assume without loss of generality that cj (t, x) ≡
constant for some j ∈ {1, . . . , s − 1}. Then cj is a generalized rational first integral
of system (3.27).
This proves the statement and consequently the lemma. 
From the proof of Lemma 3.4 we know that this lemma also holds when sys-
tem (3.27) and Kj , j = 1, . . . , s, are in C  (U, C)[x] with each coefficient having its
derivative vanishing either identically or only in a zero Lebesgue measure subset.
To extend the Darboux theory of integrability, we also need the next result, see
[35, Lemma 7], which is an extension of Lemma 3.1 from autonomous differential
systems to nonautonomous differential systems.
Lemma 3.5 Let O be an open subset of U × Cn . Assume that H1 , . . . , Hn , Hn+1 are
analytic first integrals of XN defined on O, and that H1 , . . . , Hn are functionally
independent on O. The following statements hold.
(a) There exist analytic functions C1 , . . . , Cn defined on O such that

dHn+1 = C1 dH1 + · · · + Cn dHn . (3.31)

(b) If some function Cs , s = 1, . . . , n, is not a constant, it is a first integral of XN .

Proof (a) Here the proof is a small improvement on the original one. Let L be the
linear space spanned by dH1 , . . . , dHn . Since the vector field XN is orthogonal to
dH1 , . . . , dHn+1 and L has dimension n by the assumption, it follows that dHn+1
belongs to the linear space L, because we are in an (n + 1)-dimensional space. So
(3.31) holds. The analyticity of the Cs ’s follows by direct calculation using Cramer’s
rule.
(b) Taking the Lie derivative of the vector field XN acting on the two sides of
Eq. (3.31), we get

0 = LXN (C1 )dH1 + · · · + LXN (Cn )dHn ,

where we have used the facts that LXN (dHk ) = dLXN (Hk ) and LXN (Hk ) = 0, k =
1, . . . , n + 1, because the Hk ’s are first integrals of XN . By the functional indepen-
dence of H1 , . . . , Hn , it follows that LXN (Ck ) ≡ 0. This proves the
statement. 
Now we can state the main results of this subsection, which combine the
results of Giné, Grau and Llibre [179] and of Blázquez-Sanz and Pantazi [35],
together with some improvements. For the functions K1 . . . , Kr ∈ C  (U, C)[x],
ρ := rank(W (K1 , . . . , Kp+q )) means that there exists a (t0 , x0 ) ∈ U × Cn such that
rank(W (K1 , . . . , Kp+q ))(t0 , x0 ) = ρ and rank(W (K1 , . . . , Kp+q ))(t, x) ≤ ρ for all
(t, x) ∈ U × Cn .
Theorem 3.6 Assume that system (3.27) has elementary or analytic coefficient func-
tions, and it admits
3.2 Generalization of the Classical Darboux Theory of Integrability 119

(i) p Darboux invariant polynomials f1 , . . . , fp with cofactors K1 , . . . , Kp , where

f1 , . . . , fp ; K1 , . . . , Kp ∈ C k (U, C)[x] and their coefficient functions are elemen-
tary or analytic functions, and
(ii) q exponential factors Fp+1 , . . . , Fp+q with cofactors Kp+1 , . . . , Kp+q , where
Fi = exp(gi /hi ) with gi , hi , Ki ∈ C k (U, C)[x], i = p + 1, . . . , p + q, having ele-
mentary or analytic coefficient functions.
If d log f1 , . . . , d log fp , d log Fp+1 , . . . , d log Fp+q are C-linearly independent 1-
forms in C k−1 (U, C)[x], the following statements hold.
(a) If system (3.27) has a generalized Darboux first integral of the form
λ δ
f1λ1 . . . fp p Fp+1
δ1
. . . Fp+q
q
, (3.32)

where λi , δj ∈ C, i = 1, . . . , p, j = 1, . . . , q, then p + q > rank(W (K1 , . . . ,

Kp+q )).
(b) If p + q > rank(W (K1 , . . . , Kp+q )), then system (3.27) has either a generalized
rational first integral or a generalized Darboux first integral of the form (3.32).
(c) If system (3.27) has a generalized Darboux first integral of the form (3.32), then
p + q ≥ rank(W (K1 , . . . , Kp+q , divXN )).
(d) If p + q ≥ rank(W (K1 , . . . , Kp+q , divXN )), then system (3.27) has either a gen-
eralized Darboux Jacobi multiplier, or a generalized Darboux first integral, or
a generalized rational first integral. Concrete examples of each of these three
cases can be found.
(e) If p + q > rank(W (K1 , . . . , Kp+q )) + n, then system (3.27) has a generalized
rational first integral.

Similar to Lemma 3.4, Theorem 3.6 also holds when system (3.27) and its Dar-
boux invariant polynomials and exponentials have coefficients with derivatives either
identically vanishing or null only in a zero Lebesgue measure subset.

Proof (a) Direct calculations show that for a generalized Darboux function of the
form (3.32)
 
λ δ1 δq
XN f1λ1 . . . fp p Fp+1 . . . Fp+q (3.33)
⎛ ⎞
  p

q
λ δ1 δq
= f1λ1 . . . fp p Fp+1 . . . Fp+q ⎝ λj Kj + δj Kp+j ⎠ .
j=1 j=1

By the assumption of the statement we have


p

q
λj Kj + δj Kp+j ≡ 0, and U × Cn .
j=1 j=1

It follows that K1 , . . . , Kp+q are C-linearly dependent, and consequently

120 3 Darboux and Liouvillian Integrability

rank(W (K1 , . . . , Kp+q )) < p + q.

(b) Set ρ := rank(W (K1 , . . . , Kp+q )). Then ρ < p + q. To simplify the notation
we do not distinguish between the Darboux invariant polynomials and the exponen-
tial factors, denoting them in a unified way as G 1 (t, x), . . . , G p+q (t, x), and their
cofactors are also named K1 , . . . , Kp+q . We assume without loss of generality that
K1 , . . . , Kρ are linearly independent at some point of U × Cn and K1 , . . . , Kρ+1 are
linearly dependent in U × Cn . Recall that
 τ
p+q−1
Kj = Kj , XN (Kj ), . . . , XN (Kj ) , j = 1, . . . , ρ + 1.

As in the proof of Lemma 3.4 we can show that K1 , . . . , Kρ are linearly indepen-
dent on an open and dense subset, say O, of U × Cn . Then there exist functions
c1 (t, x), . . . , cρ (t, x) such that

Kρ+1 = c1 (t, x)K1 + · · · + cρ (t, x)Kρ on O.

Working in a similar way to the proof of Lemma 3.4 together with some technical
treatments we can prove that XN (cj ) = 0 on O, j = 1, . . . , ρ. Now we have two
cases.
• If there exists some j ∈ {1, . . . , ρ} such that cj is not a constant, then cj is a
generalized rational first integral.
• If all the cj ’s are constants, we have that G c11 . . . G ρρ G −1
c
ρ+1 is a Darboux first integral
of system (3.27).
This proves statement (b).
(c) Let J be the Darboux function defined in (3.32). By the assumption of the state-
ment, we have XN (J) = JdivXN . So we get from (3.33) that


p

q
λj Kj + δj Kp+j − divXN = 0.
j=1 j=1

This shows that

rank(W (K1 , . . . , Kp+q , divXN )) ≤ p + q.

(d) Set ρ := rank(W (K1 , . . . , Kp+q , divXN )). By the assumption of the statement
we have ρ ≤ p + q. Now we have

W (K1 , . . . , Kp+q , divXN ) = K1 , . . . , Kp+q , DN ,
3.2 Generalization of the Classical Darboux Theory of Integrability 121

where
 τ
p+q
Kj = Kj , XN (Kj ), . . . , XN (Kj ) , j = 1, . . . , p + q,
 τ
p+q
DN = divXN , XN (divXN ), . . . , XN (divXN ) .

Case 1. Assume that there exist Ki1 , . . . , Kiρ ∈ {K1 , . . . , Kp+q } with i1 < · · · < iρ
such that rank(Ki1 , . . . , Kiρ ) = ρ. Then it follows from the assumption of the state-
ment that rank(Ki1 , . . . , Kiρ , DN ) = ρ. Similar to the proof of statement (b) we obtain
that
DN = c1 (t, x)Ki1 + · · · + cρ Kiρ ,

where the cj (t, x)’s are the generalized rational functions and are defined in an open
and dense subset of U × Cn . Moreover, if some cj (t, x) is not a constant, it is a
generalized rational first integral. If all the cj (t, x)’s are constants, system (3.27) has
a generalized Darboux Jacobian multiplier.
Case 2. Assume that for any Ki1 , . . . , Kiρ ∈ {K1 , . . . , Kp+q } with i1 < · · · < iρ we
have rank(Ki1 , . . . , Kiρ ) < ρ. Then rank(K1 , . . . , Kp+q ) < ρ. We are in the case (b).
Hence system (3.27) has either a generalized rational first integral or a generalized
Darboux first integral.
This finishes the proof of statement (d).
(e) As in the proof of statement (b), set ρ := rank(W (K1 , . . . , Kp+q )), let
G 1 (t, x), . . . , G p+q (t, x) be the unified expressions of the Darboux invariant poly-
nomials fi and the exponential factors Fj , and let their corresponding cofactors be
K1 , . . . , Kp+q .
Again as in the proof of statement (b), we assume without loss of generality that
K1 , . . . , Kρ are linearly independent on an open and dense subset, say O, of U × Cn ,
and K1 , . . . , Kρ , Kρ+s are linearly dependent in U × Cn , s = 1, . . . , n + 1. Then we
have

Kρ+s = c1,s (t, x)K1 + · · · + cρ,s (t, x)Kρ , on O, s = 1, . . . , n + 1, (3.34)

where the cj,s (t, x)’s are generalized rational functionals or constants. If some
cj0 ,s0 (t, x), j0 ∈ {1, . . . , ρ}, s0 ∈ {ρ + 1, . . . , ρ + n}, is not a constant, it is a gener-
alized rational first integral of system (3.27).
If all cj,s (t, x)’s are constants, as in the proof of statement (b) we get from (3.34)
the n generalized Darboux first integrals

Hs (t, x) := G 11,s . . . G ρρ,s G −1

c c
ρ+s , s = 1, . . . , n + 1.

Since system (3.27) is n-dimensional, the first integrals log H1 (t, x), . . . ,
log Hn+1 (t, x) must be functionally dependent. Hence their differential one-forms
d log H1 (t, x), . . . , d log Hn+1 (t, x) are linearly dependent in U × Cn , and are gen-
eralized rational functions. Then we get from Lemmas 3.5 and 3.1, Theorem 3.1,
122 3 Darboux and Liouvillian Integrability

and their proofs that system (3.27) has a generalized rational first integral, where we
have used the facts that d log H1 (t, x), . . . , d log Hn+1 (t, x) are generalized rational
one-forms.
This completes the proof of statement (e) and consequently of the theorem. 

3.2.5 Differential Systems in the Sparse Case

In the previous subsections we extended the classical Darboux–Jouanolou integra-

bility theorem by taking into account the exponential factors, the independent singu-
larities, the algebraic multiplicities, and the Wronskian of cofactors, and to nonau-
tonomous differential systems. In this subsection we will further improve the classical
Darboux–Jouanolou integrability theorem of the polynomial differential system (3.1)
by taking into account the case when system (3.1) has sparse terms. The results of
this subsection are due to Chéze [87], where the main notion is the Newton polytope.
For a Laurent polynomial f (x) = cs x s with cs ∈ C, the convex hull
s∈Zn , |s| finite
in Rn of the vectors s with fs = 0, is called the Newton polytope of the Laurent
polynomial f (x). We denote by N (f ) this Newton polytope of f . For example, the
Newton polytope of the Laurent polynomial x 2 − y2 + x −1 y−1 − 2x + y is the closed
region limited by the triangle in R2 with vertexes (0, 2), (2, 0) and (−1, −1).
The main result of this subsection is the following, which is Theorem 1 of Chéze
[87].

Theorem 3.7 For the polynomial differential system (3.1) and its associated
 vec-
 n
tor field X , we denote by N the Newton polytope N yi Pix(x)
i
, where y =
i=1
(y1 , . . . , yn ) ∈ Cn is a generic point. Let N be the number of integer points in N ∪ Zn+ ,
where Z+ = N ∪ {0}. The following statements hold.
(a) If X has at least N + 1 irreducible Darboux polynomials, then X has a Dar-
boux first integral.
(b) If X has at least N + n irreducible Darboux polynomials, then X has a rational
first integral.

To prove Theorem 3.7 we need several technical lemmas, which are also from
Chéze [87]. Let P be a polytope in Rn . A hyperplane H is a supporting hyperplane
of P if H ∩ P = ∅, and P is located completely in one half-space separated by
H. If the polytope is in R2 , each supporting hyperplane can be chosen as a line
with one of the sides of the polytope belonging to it. If the polytope is in R3 , each
supporting hyperplane can be chosen as a plane with one of the surfaces of the
polytope belonging to it.
We claim that each supporting hyperplane H in Rn can be expressed as v, y
= c,
where v is an n-dimensional vector with integer coefficients, and c is an integer.
Indeed, since each supporting hyperplane is uniquely determined by a surface of
3.2 Generalization of the Classical Darboux Theory of Integrability 123

the polytope, each surface is uniquely defined by n linearly independent vectors

with integer entries. Let Xj = (x1j , . . . , xnj ), j = 1, . . . , n, be the n vectors and let
a, y
= c with a = (a1 , . . . , an ) ∈ Rn and c ∈ R be the equation of the supporting
hyperplane. Substituting Xj into the equation in place of y, we get a system of n linear
algebraic equations. Then the claim follows from the solution of this linear algebraic
system.
By this last claim it follows that a Newton polytope of a Laurent polynomial f
can be represented by the equations of its supporting hyperplanes, i.e.
!
N (f ) =  ∈ Zn | vj , 
≤ cj , j = 1, . . . , m , (3.35)

where m is the number of supporting hyperplanes of the polytope.

Let f be a Laurent polynomial, and v ∈ Zn . The degree of f related to v is by
definition max v, z
, and is denoted by degv f . The next result, taken from [87,
z∈N (f )
Proposition 4], describes some properties of the degree.
Proposition 3.6 Let f , g ∈ C[x], v ∈ Zn and y1 , y2 be two generic points in C. The
following statements hold.
(a) degv (f + g) ≤ max(degv (f ), degv (g)),
(b) degv (x1 f + x2 g) = max(degv (f ), degv (g)),
(c) degv (f g) = degv (f ) + degv (g),
(d) degv (∂xi f ) ≤ degv (f ) − vi = degv (f /xi ), where vi is the ith component of v.
Proof (a) Denote by Conv(N (f ) ∪ N (g)) the convex hull of N (f ) ∪ N (g). We
can check easily that N (f + g) ⊂ Conv(N (f ) ∪ N (g)). Hence we have

degv (f + g) = max v, z
≤ max v, z

z∈N (f +g) z∈Conv(N (f )∪N (g))

" #
= max v, z
= max max v, z
, max v, z

z∈N (f )∪N (g) z∈N (f ) z∈N (g)

!
= max degv (f ), degv (g) ,

where in the second equality we have used the fact that the maximum taken in
Conv (N (f ) ∪ N (g)) is attained at the points in N (f ) ∪ N (g), which follows
from (3.35).
(b) Since N (y1 f + y2 g) = Conv (N (f ) ∪ N (g)), it follows that

degv (y1 f + y2 g) = max v, z

= max v, z
.
z∈N (y1 f +y2 g) z∈Conv(N (f )∪N (g))

The next proof is the same as that of (a).

(c) Ostrowski [352, Theorem V] shows that N (f g) = N (f ) + N (g) with + being
understood in the sense of Minkowski. Then the proof can be completed following
similar arguments as in the proof of (a).
(d) Since N (∂xi f ) ⊂ N (f /xi ), it follows that
124 3 Darboux and Liouvillian Integrability

degv (∂xi f ) = max v, z

≤ max v, z − ei

z∈N (∂xi f ) z∈N (f )

= max v, z
− vi = degv (f /xi ).
z∈N (f )

The proof of the proposition is complete. 

The next proposition, due to Chéze [87, Proposition 5], provides a relation between
two Newton polytopes via their degrees.

Proposition 3.7 Let f , g be two Laurent polynomials in Cn , and let N (f ) be given

in (3.35). If degvj (g) ≤ degvj (f ), then N (g) ⊂ N (f ).

Proof Direct calculations show that for each integer vector vj in (3.35)

max vj , z
= degvj g ≤ degvj f = max vj , z
= cj .
z∈N (g) z∈N (f )

This implies that N (g) ⊂ N (f ). 

Now we study the Newton polytope of the cofactors of the Darboux polynomials,
which is due to Chéze [87, Proposition 6].

Lemma 3.6 Let X be the vector field associated to system (3.1), and suppose it
has a Darboux polynomial f with cofactor g. Then

N (g) ⊂ N ∩ Nn ,

where N is defined in Theorem 3.7.

Proof Since g is a polynomial, we have N (g) ⊂ Rn+ , where R+ is the set of non-
negative real numbers. In addition, by X (f ) = gf it follows from Proposition 3.6
that for any v ∈ Zn

degv (g) + degv (f ) = degv (gf ) = degv P1 ∂x1 f + · · · + Pn ∂xn f
≤ max {degv (Pi ∂xi f )} = max {degv Pi + degv ∂xi f }
i=1,...,n i=1,...,n

≤ max {degv Pi + degv f − vi }.

i=1,...,n

This reduces to
 
P1 Pn
degv (g) ≤ max {degv Pi − vi } = max {degv (Pi /xi )} = degv y1 + · · · + yn .
i=1,...,n i=1,...,n x1 xn

Hence, it follows from Proposition 3.7 that N (g) ⊂ N. This proves the lemma. 

Proof of Theorem 3.7. Let f be a Darboux polynomial of the vector field X with
corresponding cofactor g. By Lemma 3.6 one has
3.2 Generalization of the Classical Darboux Theory of Integrability 125

N (g) ⊂ N ∩ Rn+ .

Let nj = (nj1 , . . . , njn ), j = 1, . . . , N, be N points in N ∩ Zn+ . Then the cofactor g is

a C-linear combination of x n1 , . . . , x nN , which forms an N-dimensional linear space.
Hence any N + 1 cofactors are C-linearly dependent.
The following proofs of (a) and (b) are the same as those of Theorem 3.1.
The theorem is proved. 
As an application of Theorem 3.7, we consider the Lorenz system

ẋ = s(y − x) = P(x, y, z), ẏ = rx − y − xz = Q(x, y, z), ż = −bz + xy = R(x, y, z).

The Newton polytope, denoted by NL , of w1 Px −1 + w2 Qy−1 + w3 Rz−1 with the

generic points w1 , w2 , w3 ∈ R is the convex hull in R3 of the points

(−1, 1, 0), (0, 0, 0), (1, −1, 0), (1, −1, 1), (1, 1, −1).

Moreover, we have NL ∩ Z3+ = {(0, 0, 0), (1, 0, 0)}. This implies that any cofactor
is of the form k0 + k1 x. Using Theorem 3.7 we can reduce the calculation of the
cofactors in two dimensions.
The associated Newton polytope of the next example

ẋ = −y + axy, ẏ = x + bx 2 + cy2

is the convex hull in R2 of the points (−1, 1), (0, 1), (1, −1), (2, −1). The nonneg-
ative integer points in this Newton polytope are (0, 0), (1, 0), (0, 1). So the cofactor
of any Darboux polynomial of this system is of the form k0 + k1 x + k2 y. In this
example, Theorem 3.7 cannot simplify the calculations of the Darboux polynomials
of the system.

3.2.6 Other Extensions

The Darboux theory of integrability also has several other extensions along different
lines. We list some of them:
• On regular surfaces. Llibre and Rodríguez [267] established a Darboux integra-
bility theorem in R3 on regular surfaces. Llibre and Zhang [300] obtained a cor-
responding Darboux integrability theorem in Rn on regular hypersurfaces.
• The Local Darboux theory of integrability. Recently the Darboux theory of inte-
grability was extended to a neighborhood of a singularity by Llibre et al. [294],
which is called the local Darboux theory of integrability. We will present this local
theory of integrability in Chap. 7, where we will study the existence of analytic
normalizations of analytic integrable differential systems.
126 3 Darboux and Liouvillian Integrability

• Discrete dynamical systems. Gasull and Mañosa [170] extended the classical Dar-
boux theory of integrability to discrete dynamical systems.
As a special case of invariant algebraic curves or hypersurfaces, the problem
on the existence and the number of invariant planes or hyperplanes for polynomial
differential systems is relatively simple, and there are plenty of results. But there
are also lots of open problems to be solved. We refer to Artés et al. [17, 18], Artés
and Llibre [19], Dai [121], Llibre and Medrado [255], Llibre and Rodríguez [266],
Sokulski [405], Ye [452], and Zhang et al. [462, 478, 479] and the references therein.
The inverse problem of the Darboux theory of integrability is, for a given set of
algebraic curves or hypersurfaces, to characterize the polynomial differential systems
for which these prescribed algebraic curves or hypersurfaces are invariant. In this
direction, there have appeared a series of results, see Christopher et al. [101–105]
and the references therein.
At the present time the extension of the Darboux theory of integrability does not
take into account the geometry and topology of the invariant algebraic curves, or
surfaces, or hypersurfaces. This would be an interesting subject to study.

3.3 Liouville and Elementary First Integrals

Prelle and Singer [365] in 1983 provided a reduction of the elementary first integrals
of polynomial differential systems. Singer [403] in 1992 together with Christopher
[95] in 1999 proved that for a planar polynomial differential system, the existence
of Liouvillian first integrals is equivalent to the existence of a Darboux integrating
factor. Consequently the Liouvillian integrability of a two-dimensional polynomial
differential system is equivalent to the Darboux integrability. Before stating Prelle
and Singer’s results, we first introduce some fundamental results on differential field
extensions, which will be used in the proofs.

3.3.1 Background on Differential Field Extensions

Let K be a ring. Denote by ⊕ and ⊗ the summation and multiplication operators of

the ring respectively, and by 0K and 1K respectively the zero and the unit elements
of K. The ring K is of characteristic zero if for arbitrary r ∈ N

1 ⊕ · · · ⊕ 1K = 0K
$ K %& '
r

does not hold. If there exists an r ∈ N such that the above equality holds, the smallest
such r is called the characteristic of K, denoted by char(K). For example, R and Z
3.3 Liouville and Elementary First Integrals 127

are respectively a field and a ring of characteristic zero, whereas Z/(rZ) is a ring of
characteristic r.
A derivative on the ring K is an operator δ : K → K which satisfies

δ(x + y) = δx + δy, δ(xy) = (δx)y + x(δy), for all x, y ∈ K.

A differential field (K, Δ) consists of the field K and the set Δ of commutative
derivatives defined on K. In this paper all mentioned fields have characteristic 0.
For example, in Rn with the coordinates x = (x1 , . . . , xn ) set Δ = {∂x1 , . . . , ∂xn },
then (R(x), Δ) is a differential field. Here R(x) is the field of rational functions in x
with coefficients in R.
A differential field extension of a differential field (K, Δ) is a differential field
(L, Δ ) with the properties that K ⊂ L and for all δ  ∈ Δ we have δ  K ∈ Δ. Because
of the relation between the derivatives of the differential field (K, Δ) and its field
extension (L, Δ ), we also use Δ to represent Δ without causing confusion. To
simplify the notation we also use L/K to denote the differential field extension
(L, Δ) of (K, Δ).
Example
 
• R, dxd
is a differential field (a trivial one). R(x), dx d
is a differential field consist-
ing of the field of real rational functions in the single
variable x and its derivative,
and it is a differential field extension of R, dx d
.
• The differential  field (R(x,
y), {∂x , ∂y }) is a differential field extension of the dif-
ferential field R(x), dx d
.
 √
• The differential field R(x, x), dx d
is a differential field extension of the differ-

ential field R(x), dx d
.
For a field extension L/K,
• α ∈ L is called
– an algebraic element of K if there exists a polynomial F with coefficients in K
such that F(α) = 0.
– a transcendental element of K if α is not an algebraic element over K.
• If each element of L is algebraic over K, we call L/K an algebraic extension of
the field K.
• S ⊂ L is algebraically independent over K if for any s1 , . . . , sr ∈ S, there does not
exist a polynomial P(Z1 , . . . , Zr ) with coefficients in K such that P(s1 , . . . , sr ) ≡ 0.
• The maximal cardinality of the algebraically independent elements of L over K is
called the transcendental degree of L/K.
For example, K = Q, L = Q(x), where x is a single variable. Then the transcenden-
tal degree of L/K is 1. Indeed, the elements of Q are algebraically dependent, hence
the algebraically independent elements of L over Q can only be W := Q(x) \ Q.
Clearly any element of W is algebraically independent, but any two elements are
algebraically dependent (because they are functionally dependent). This shows
that the transcendent degree of L/K is 1.
128 3 Darboux and Liouvillian Integrability

Example Since the field under consideration is of characteristic zero, for the algebraic
extension L of a differential field K, the derivative over K can be uniquely extended
to L.
For a field K,
• The splitting field of a polynomial p(x) over K is the smallest field extension
( of K
such that p(x) splits into a product of linear factors on it. That is, p(x) = (x − ai ),
ai ∈ L, and L/K is the smallest field extension of K such that the factorization
can√happen. For example,
√ the splitting field of the polynomial x 2 − 2 over Q is
Q( 2) := {a + b 2| a, b ∈ Q}.
• The field K is algebraically closed if each nonconstant polynomial in K[x] has a
zero in K. For example, C is algebraically closed, whereas R is not.
• The algebraic closure of K is the smallest algebraic field extension L over K such
that L is algebraically closed. For example, the algebraic closure of R is C. In fact,
every field K has an algebraic closure.
• A polynomial p(x) over K is separable if all roots of p(x) in the algebraic closure
of K are pairwise distinct, i.e. without repeated roots.
• An algebraic field extension L/K of K is normal if whenever an irreducible poly-
nomial of K[x] has zeros in L, it can be factorized as a product of linear factors
over L.
• A normal closure of an algebraic field extension L/K is a field extension L of
L such that L/K is normal, and L is the smallest field extension satisfying this
property. In fact, each algebraic field extension has a normal closure.
• A field automorphism over K is a bijective map ϕ : K → K preserving all alge-
braic properties of K. Preserving algebraic properties means that ϕ(0K ) = 0K ,
ϕ(1K ) = 1K , ϕ(a + b) = ϕ(a) + ϕ(b) and ϕ(ab) = ϕ(a)ϕ(b). For example, the
conjugacy of the complex numbers is a field automorphism over the field C.
• All field automorphisms over K fixing all elements of a subfield K  ⊂ K form a
group under the composition of maps called the Galois group. Denote by G (K/K  )
this Galois group.
√
Example Set K  = Q, and K = Q( 2). The field automorphism over K fixing Q is

ϕ: K√ −→ K√
a + 2b −→ a − 2b.

• The number of elements of a group G is called order of the group, denoted by |G |.

For a field extension L/K,
• L can be considered as a vector space over K: the elements of L are treated as
vectors, and elements of K are treated as scalars, whereas the sum of the vectors
is that of the elements in the field L and the product of the elements of L and K is
that of the elements in the field L.
– The dimension of this vector space is called the degree of the differential field
extension, denoted by [L : K].
3.3 Liouville and Elementary First Integrals 129
√ √
Example [Q[ 2] : Q] = 2. [Q[ 3] : Q] = 3. These arguments can be checked
by direct calculations.
• If [L : K] ∈ N, we call L/K a finite field extension.
• Let S ⊂ L,
– K(S) denotes the minimal subfield of L including K and S.
– If S contains only one element, we call K(S) the minimal field extension of K.

Example Assume that L/K is a differential field extension and t ∈ L is a transcen-

dental element of K. Then K(t) is the smallest field extension of K which contains
K and t. The derivative over (K(t), Δ) can be obtained by the extension of the deriv-
atives of (K, Δ). For instance, for arbitrary δ ∈ Δ there exists an xδ ∈ K(t) such that
δt = xδ .
A differential field extension is elementary if this extension can be written in the
tower form
K = K0 ⊂ K1 ⊂ · · · ⊂ Kr = L

such that each field extension satisfies one of the following three properties
(a) Ki+1 is a finite algebraic extension of Ki ; or
(b) Ki+1 = Ki (t), where t satisfies that for each δ ∈ Δ, there exists an x ∈ Ki such
δt
that = δx; or
t
(c) Ki+1 = Ki (t), where t satisfies that for each δ ∈ Δ, there exists an x ∈ Ki such
δx
that δt = .
x
In fact, an elementary extension of a field consists of finitely many smallest field
extensions, each step adding either finitely many algebraic elements, or finitely many
exponential elements or finitely many logarithmic elements. Each Ki is called a tower
element of the elementary field extension, i = 0, 1, . . . , r. The elementary functions
of a single variable were introduced by Liouville from 1833 to 1841 in his study of
the integration of functions.
A differential field extension is Liouvillian if the extension can be written in the
above tower form such that each extension in the tower is of one of the following
three forms:
(a) Ki+1 is a finite algebraic extension of Ki ; or
(b) Ki+1 = Ki (t), where t satisfies that for each δ ∈ Δ, there exists an x ∈ Ki such
δt
that = δx; or
t
(c) Ki+1 = Ki (t), where t satisfies that for each δ ∈ Δ, δt ∈ Ki .
In fact, the Liouvillian extension of a field includes the elementary extension and
the smallest field extension in a few steps by adding finitely many integrations of
elements of the basic field. We remark that the logarithmic elements can be obtained
by integration.
130 3 Darboux and Liouvillian Integrability

Example Consider the field C(x) of rational functions in the single variable x with
coefficients in C.
√
• x∈ / C(x) is an√algebraic function over C(x), and it belongs to the algebraic field
extension C(x, x).
• ln x ∈/ C(x) is a logarithmic function over C(x), and it belongs to the elementary
field extension C(x, ln x).
• tan−1 x ∈/ C(x) is also a logarithmic function over C(x) because tan−1 x = − 21 i ln

1+ix
, and so C x, tan−1 x also belongs to the elementary field extension over
1−ix √
C(x), where i = −1.
2
• ex ∈/ C(x) is an exponential function over C(x), and it belongs to the elementary
2
field extension C(x, ex ).
We note that the functions in the second and third examples are both integrals of
2
functions in C(x). The function ex in the fourth example can also be obtained by
)x 2 )x
integrating 2ses ds. Whereas √ 0 e−s ds, called the Error function and denoted
2 2

π
by Erf(x), is not an elementary function. The fact that the Error function is not
elementary was first proved by Liouville (see Lützen [314]). A proof can also be
found in Rosenlicht [377].
When is the integral of an elementary function also an elementary function?
Liouville proved the following result, see [314, 377]. Let (K, Δ) be a differential
field. Set
Con(K, Δ) = {k ∈ K| δk = 0, for all δ ∈ Δ}.

Without risk of confusion we also denote the latter by Con(K).

Theorem 3.8 (Liouville) Let (L, δ) be an elementary field extension of (K, δ).
Assume that Con(K) = Con(L). If x ∈ K and y ∈ L satisfy δy = x, then there exist
c1 , . . . , cm ∈ Con(K) and w0 , w1 , . . . , wm ∈ K such that

m
δwi
x = δw0 + ci .
j=1
wi

This theorem provides an essential property describing those functions in (K, δ)

whose integral is an elementary function.
In what follows, a function H(x) is a Liouvillian (or elementary) first integral of
system (1.1) if H belongs to the Liouvillian (or elementary) field extension of C(x)
and it is a first integral of system (1.1). System (1.1) is Liouvillian (or elementary)
integrable if it has n − 1 functionally independent Liouvillian (elementary) first
integrals.
3.3 Liouville and Elementary First Integrals 131

3.3.2 The Prelle and Singer Integrability Theorems

In this section we will introduce the theorems of Prelle and Singer, who characterized
the ordinary differential systems which have an elementary or a Liouvillian first
integral. These results were proved respectively by Prelle and Singer [365] in 1983
and by Singer [403] in 1992.
Before introducing Prelle and Singer’s results on elementary first integrals, we
illustrate a simple example: the predator and prey system

ẋ = ax − bxy, ẏ = −cy + dxy,

with a, b, c, d positive constants, has the first integral

H(x, y) = dx + by − c log x − a log y.

By this first integral we can prove that all orbits in the first quadrant of the (x, y)
plane are closed, see e.g. Ding and Li [126], or Jost [216] and Zhang [473]. Note
that this first integral belongs to the elementary extension of the field C(x, y), and it
can be written as 
w0 (x, y) + ci log wi (x, y),

where wj is an algebraic function over C(x, y). The result of Prelle and Singer [365]
in 1983 shows that if an ordinary differential system has an elementary first integral,
it can be expressed in this last form.

Theorem 3.9 (Prelle–Singer theorem) Assume that (L, Δ) is an elementary field

extension of the differential field (K, Δ), and Con(L, Δ) = Con(K, Δ). Choose
δ1 , . . . , δn ∈ Δ, y1 , . . . , yn ∈ K, and

X = y1 δ1 + · · · + yn δn .

If Con(L, Δ) is a proper subset of Con(L, {X }), then there exist c1 , . . . , cm ∈

Con(K, Δ), and algebraic elements w0 , w1 , . . . , wm over K, such that

m
X wi m
δwi
X w0 + ci = 0, but δw0 + ci = 0, for some δ ∈ Δ.
i=1
wi i=1
wi

Remark If K = C(x) with x = (x1 , . . . , xn ), and Δ = {∂x1 , . . . , ∂xn } is the set of all
partial derivatives. Then Con(K, Δ) = C. The vector field X in Theorem 3.9 is
either rational or polynomial. The assumption Con(L, Δ) = Con(L, {X }) implies
that the vector field X has a nonconstant elementary first integral. The Prelle–Singer
theorem indicates that if the vector field X has an elementary first integral, it must
have a first integral of the form
132 3 Darboux and Liouvillian Integrability

w0 + ci log wi ,
i


where is a finite summation, ci ∈ C, and the wi ’s are algebraic functions
( over
C(x). Moreover, the vector field X has a Darboux-like first integral ew0 wici .
i
Theorem 3.9 has the following consequence.

Corollary 3.1 If a planar polynomial vector field X has an elementary first integral,
it must have an integrating factor of the form
n
f1n1 . . . fp p , fi ∈ C[x, y], ni ∈ Z,

where the fi ’s are Darboux polynomials of the system, i = 1, . . . , p.

Proof Let the polynomial vector field X be

X = P(x, y)∂x + Q(x, y)∂y ,

where ∂x and ∂y denote the partial derivatives with respect to x and y. By Theorem 3.9
the vector field X has a nontrivial first integral of the form


n
H= wici ew0 ,
i=1

where the ci ’s ∈ C and the wi ’s are the algebraic functions over C(x, y). Then we
have
P(x, y)∂x H + Q(x, y)∂y H ≡ 0.

Since P and Q are not simultaneously identically equal to zero, assuming Q is not
identically zero, set
∂x H
R= .
Q

Then R is an integrating factor of X .

We claim that for each algebraic function wi , ∂x wi is also an algebraic function.
Indeed, assume that wi is a solution of the algebraic equation

C m + a1 (x, y)C m−1 + · · · + am−1 (x, y)C + am (x, y) = 0, (3.36)

where ai ∈ C(x, y) for i = 1, . . . , m, and the degree m is minimal in the sense that no
smaller degree polynomial has wi as a solution. This polynomial is called the minimal
polynomial of wi . Since the ai (x, y)’s are rational, so are the ∂x ai (x, y)’s. Let K be
the minimal algebraic field extension containing the wi ’s of C(x, y). Differentiating
Eq. (3.36) with respect to x gives
3.3 Liouville and Elementary First Integrals 133

wim−1 ∂x a1 (x, y) + · · · + ∂x am−1 wi + ∂x am

∂x wi = − .
mwim−1 + (m − 1)a1 wim−2 + · · · + am−1

This shows that ∂x wi ∈ K.

From the form of H we get that ∂x H ∈ K, and consequently R ∈ K. That is, R is
an algebraic function on C(x, y).
Since R is an integrating factor of X , we have

X (R) = −R div X . (3.37)

Let K be the normal closure of K, and G be the Galois group formed by the auto-
morphisms over K fixing C(x, y). By one of Emil Artin’s results, see e.g. Lang [224,
Theorem 1.1] and its
 proof, it follows
 that the group G is of finite order, denoted by
N = |G |, and N = K : C(x, y) .
For any automorphism σ ∈ G , we get from (3.37) that

X (σ (R)) = −σ (R)div X .

It follows that
X (V ) = −V div X

where
1 
V (x, y) = σ (R).
N
σ ∈G

Clearly, for any σ0 ∈ G we have

1  1 
σ0 (V ) = σ0 (σ (R)) = σ (R) = V (x, y),
N N
σ ∈G σ ∈G

because σ0 is an automorphism on K. This shows that V (x, y) ∈ C(x, y). We complete

the proof of the corollary. 

In 1992 Singer [403, Theorem 1] studied those planar polynomial differential

systems
ẋ = P(x, y), ẏ = Q(x, y), (3.38)

with P, Q ∈ C[x, y], whose orbits are contained in the vanishing set of a Liouvillian
function. By definition a Liouvillian function is an element in the Liouvillian field
extension of C(x, y).

Theorem 3.10 (Singer’s theorem) Assume that the polynomial differential system
(3.38) has an analytic solution (x, y) = (ϕ(t), ψ(t)) defined in an open subset V ⊂
C. If there is a Liouvillian function F(x, y) which is analytic in an open subset
containing S := {(ϕ(t), ψ(t))| t ∈ V }, and F(x, y)|S = 0, then either S is an
134 3 Darboux and Liouvillian Integrability

algebraic solution, or system (3.38) has an integrating factor

* (x,y) 
R(x, y) := exp U(x, y)dx + V (x, y)dy ,
(x0 ,y0 )

where U, V ∈ C(x, y) satisfy ∂y U = ∂x V .

This theorem verifies the importance of invariant algebraic curves in studying the
dynamics of polynomial differential systems.
An important consequence of this last theorem is the equivalence between Liou-
ville and Darboux integrability for planar polynomial differential systems, see the
corollary of Singer [403, Theorem 1] and Christopher [95].

Theorem 3.11 (Singer–Christopher theorem) A planar polynomial differential

system (3.38) has a Liouvillian first integral if and only if it has a Darboux inte-
grating factor.

Proof The main idea of the proof follows from Christopher [97] with small revisions.
For the differential field (K, Δ) = (C(x, y), {∂x , ∂y }), δ ∈ Δ is δ = ∂x or δ = ∂y .
Sufficiency. Assume that the differential system (3.38) has a Darboux integrating fac-
tor R. Then by definition of the Darboux functions we can check that δR/R ∈ C(x, y).
This shows that R satisfies the condition (b) in the tower form of the Liouvil-
lian field extension. Hence RP and RQ belong to an element of the tower because
RP, RQ ∈ C(x, y)(R). By the condition (c) in the tower form of the Liouvillian field
extension we get that
* (x,y)
H(x, y) = RPdy − RQdx
(x0 ,y0 )

is a Liouvillian function, and consequently it is a Liouvillian first integral of sys-

tem (3.38).
Necessity. Denote by X the vector field associated to system (3.38). Without loss
of generality we assume that gcd(P, Q) = 1.
The proof will be separated into two parts. We first prove

Lemma 3.7 If the polynomial differential system (3.38) has a Liouvillian first inte-
gral, it has an integrating factor
* 
R = exp Udx + V dy ,

where U, V ∈ C(x, y) and ∂y U = ∂x V .

Proof To prove this proposition, we only need to prove that there exist U, V ∈
C(x, y) such that
∂y U = ∂x V, PU + QV = ∂x P + ∂y Q. (3.39)
3.3 Liouville and Elementary First Integrals 135

Let H(x, y) be an element in a Liouvillian field extension L of C(x, y) such that

X (H) ≡ 0, i.e. P∂x H + Q∂y H ≡ 0.

Since gcd(P, Q) = 1, there exists an h ∈ L such that

h ∂x H = Q, h ∂y H = −P. (3.40)

We can check that h is an integrating factor of system (3.38). Set

∂x h ∂y h
A= , B= .
h h
Then A, B ∈ L, and

∂y A = ∂x B, PA + QB = ∂x P + ∂y Q, (3.41)

where the second equality can be obtained by differentiating the first equation in
y and the second equation in x of (3.40), and then subtracting these two resulting
equations.
Some easy calculations show that the integrating factor h of system (3.38) can be
rewritten as * (x,y) 
R(x, y) := exp A(x, y)dx + B(x, y)dy .
(x0 ,y0 )

Clearly we can treat A and B in the same element in the tower form of the Liouvillian
field extension. Next we prove that if A, B ∈ Ki+1 , we can find C, D ∈ Ki replacing
A, B for which (3.41) holds. Then by induction we can find U, V ∈ C(x, y) satisfying
(3.41).
According to the tower form of the Liouvillian field extension we discuss three
different cases (a)–(c). In cases (b) and (c), we assume without loss of generality
that t is a transcendental element. Otherwise Ki (t) is an algebraic field extension of
Ki , and so we are in case (a).
(a) Assume that Ki+1 is a finite algebraic field extension of Ki . Let K i+1 be the normal
closure of Ki+1 , and G be the Galois group formed by the automorphisms over K i+1
fixing Ki . By Lang [224, Theorem 1.1] and  its proof, the group G is of finite order,
denoted by N = |G |, and N = K i+1 : Ki .
Since all elements of G preserve both Ki and the algebraic structures of Ki+1 , and
P, Q ∈ C(x, y) ⊂ Ki , we get from (3.41) that
  
ϕ(PA + QB) = ϕ ∂x P + ∂y Q ,
ϕ∈G ϕ∈G
136 3 Darboux and Liouvillian Integrability

which implies
   
P ϕ(A) + Q ϕ(B) = ∂x ϕ(P) + ∂y ϕ(Q) = N ∂x P + ∂y Q ,
ϕ∈G ϕ∈G ϕ∈G

and that
ϕ(∂y A) = ϕ(∂x B),

implies
∂y ϕ(A) = ∂x ϕ(B), for all ϕ ∈ G .

Since A and B are algebraic functions over Ki , there exist minimal polynomials
f (z), g(z) with coefficients in Ki , such that

f (A) = 0, g(B) = 0.

In addition, since ϕ ∈ G fixes Ki , this verifies that

f (ϕ(A)) = 0, g(ϕ(B)) = 0,

i.e.
ϕ(A), ϕ(B) ∈ Ki+1 , for all ϕ ∈ G .

Set
1  1 
C := ϕ(A), D := ϕ(B).
N N
ϕ∈G ϕ∈G

Then the equalities (3.41) holds for C, D instead of A, B. Since C and D are respec-
tively the average of all elements of the Galois group G acting on A and B, we
have
ϕ(C) = C, ϕ(D) = D, for all ϕ ∈ G .

Consequently C, D ∈ Ki .
(b) Assume that Ki+1 = Ki (t) with t a transcendental element of Ki , and δt/t ∈ Ki
for δ ∈ {∂x , ∂y }.
Since A, B ∈ Ki (t), we set A = a(t), B = b(t) ∈ Ki (t), which are the rational
functions in t with their coefficients in Ki . Expanding a(t) and b(t) in Laurent series
with respect to the transcendental element t
 
a(t) = C + ai t i , b(t) = D + bi t i ,
i =0 i =0

whose coefficients are in Ki . Direct calculations show that

3.3 Liouville and Elementary First Integrals 137

∂y A = ∂y a(t) = ∂y C + (∂y ai + iai p)t i ,
i =0

∂x B = ∂x b(t) = ∂x D + (∂x bi + ibi q)t i ,
i =0

where p, q ∈ Ki satisfy p = ∂y t/t, q = ∂x t/t. Substituting the expansions of A =

a(t), B = b(t) and the last two equalities into (3.41), and comparing the coefficients
of t 0 gives
∂y C = ∂x D, PC + QD = ∂x P + ∂y Q.

(c) Assume that Ki+1 = Ki (t) with t a transcendental element of Ki , and δt ∈ Ki for
δ ∈ {∂x , ∂y }.
Set A = a(t), B = b(t) ∈ Ki (t), and expand a(t) and b(t) as Laurent series with
respect to 1/t
 r 
r
a(t) = ai t i , b(t) = bi t i ,
i=−∞ i=−∞

where ai , bi ∈ Ki . Direct calculations show that


r
∂y A = ∂y a(t) = (∂y ai−1 + iai p)t i−1 + ∂y ar t r ,
i=−∞
r
∂x B = ∂x b(t) = (∂x bi−1 + ibi q)t i−1 + ∂x br t r ,
i=−∞

where p, q ∈ Ki satisfy p = ∂y t, q = ∂x t. Substituting the expansions of A = a(t)

and B = b(t) and these last equalities into (3.41), and comparing the coefficients of
t r gives

if r = 0, ∂y ar = ∂x br , Par + Qbr = 0; or
if r = 0, ∂y a0 = ∂x b0 , Pa0 + Qb0 = ∂x P + ∂y Q.

If the latter holds, C = a0 and D = b0 are suitable for the choice.

If the former holds, since Par + Qbr = 0 and gcd(P, Q) = 1, there exists an h ∈ Ki
such that
P = −br h, Q = ar h.

So we get from ∂y ar = ∂x br that

∂x h ∂y h
∂x P + ∂y Q = −br ∂x h + ar ∂y h = P +Q .
h h
138 3 Darboux and Liouvillian Integrability

Set
∂x h ∂y h
C := , D := .
h h
Then C, D ∈ Ki , and ∂y C = ∂x D and ∂x P + ∂y Q = PC + QD.
By induction, there exist U, V ∈ K0 = C(x, y) such that the equalities (3.39) hold.
This proves the lemma. 
The second step is to prove
Lemma 3.8 If the polynomial differential system (3.38) has the integrating factor
* 
R = exp Udx + V dy , with U, V ∈ C(x, y), and ∂y U = ∂x V,

then it has a Darboux integrating factor

 
g
exp fili ,
f i

where g, f , fi ∈ C[x, y], and li ∈ C.

Proof Since U, V ∈ C(x, y), take their numerators and denominators as polynomials
in x with coefficients in C(y). Let K be the smallest normal algebraic field extension
of C(y) such that K is the splitting field of the numerators and the denominators
of U and V . Then over the field K, the functions U and V can be expanded in the
fractional forms


r 
ni
αij  p
U= + ξi x i ,
i=1 j=1
(x − βi )j i=0


s 
mi
γij  q
V = + ηi x i ,
i=1 j=1
(x − βi )j i=0

where αij , γij , βi , ξi , ηi ∈ K, some of which can be zero. Since

 ni 
r   
αij jαij βi
p
∂y U = − + ξi x i ,
i=1 j=1
(x − βi )j (x − βi )j+1 i=0


s 
mi
−jγij  q
∂x V = + (i + 1)ηi+1 x i ,
i=1 j=1
(x − βi ) j+1
i=0

where ‘’ denotes the partial derivative with respect to y, comparing the coefficients
in ∂y U = ∂x V yields

ξi = (i + 1)ηi+1 , 
αi,j+1 − jβi αij + jγij = 0. (3.42)
3.3 Liouville and Elementary First Integrals 139

In the second equality, setting j = 0, we get αi1 ∈ C.

Set
  −1 αij  ξi x i+1 *
Φ(x, y)= αi1 log(x − βi ) + + + η0 dy,
i i,j,j>1
j − 1 (x − βi )j−1 i
i+1

where the last integral is a primitive function of η0 . Direct calculations show that
∂x Φ = U. In addition, since
 −βi
∂y Φ(x, y) = αi1
i
x − βi
 −1  αij (j − 1)αij
   i+1
ξi x
+ − + + η0 ,
i,j,j>1
j − 1 (x − βi ) j−1 (x − βi ) j
i
i+1

by the equalities in (3.42) we get ∂y Φ = V . Hence

*
Φ(x, y) = Udx + V dy.

Let G be the Galois group formed by the automorphisms over K fixing C(y). Then
it follows from the choice of K that G is a finite group. Denote by N = |G | the order
of G . Set
1 
Ψ = σ (Φ).
N
σ ∈G

Since σ ∈ G preserves all the algebraic structures of K, we get from the properties
of the automorphisms over the field extensions that
*  *
σ (αi1 log(x − βi )) = αi1 log(x − σ (βi )), σ η0 dy = σ (η0 )dy,

where the integral equality may be different by a constant. Since σ ∈ G fixes C(x, y),
we have

σ (∂x Φ) = σ (U) =⇒ ∂x σ (Φ) = σ (U) = U,

σ (∂y Φ) = σ (V ) =⇒ ∂y σ (Φ) = σ (V ) = V.

It follows that
∂x Ψ = U, ∂y Ψ = V.

Furthermore, we get from the expression of Φ and the definition of Ψ that

140 3 Darboux and Liouvillian Integrability

 *
Ψ (x, y) = li log Ri (x, y) + R(x, y) + S(y)dy,
i

where Ri , R ∈ C(x, y), and S ∈ C(y). Note that Ri ∈ C(x, y) comes from

Ri = (x − ϕ(βi ))
ϕ∈G

and the fact that Ri does not change under the action of each ϕ ∈ G . Since S(y) has
a fractional expansion in C, we have
* 
S(y)dy = kj log(Sj (y)) + S0 (y),
j

where Sj ∈ C[y], j = 0, and S0 ∈ C(y).

Taking the exponential of Ψ yields a Darboux integrating factor
  k
exp (Ψ ) = exp (R(x, y) + S0 (y)) Rili (x, y) Sj j (y).
i j

This proves Lemma 3.8. 

Lemmas 3.7 and 3.8 provide a proof of Theorem 3.11. 

Theorem 3.10 shows that if a planar polynomial differential system has an invari-
ant Liouvillian curve (by definition a curve contained in the zero set of a Liouvillian
function), then it is either an invariant algebraic curve, or the system has a Liouvillian
first integral. Zernov and Scárdua [459] proved under some conditions that a Liouvil-
lian integrable real polynomial differential system has an invariant algebraic curve.
Hence one might conjecture that a planar Liouvillian integrable real polynomial
differential system has an invariant algebraic curve.
Giné and Llibre [184] in 2012 found a counterexample to this conjecture by
considering the real quadratic differential system

ẋ = −1 − x(2x + y), ẏ = 2x(2x + y).


This system has the Darboux integrating factor R = exp −(2x + y)2 /4 , and so is
Liouvillian integrable. This last system also has the Liouvillian first integral
   
(2x + y)2 √ 2x + y
H(x, y) = 2x exp − − π Erf ,
4 2

where Erf(z) is the Error function mentioned above. The first integral H is Liouvillian
but not elementary. In addition, Giné and Llibre [184, Proposition 5] proved that this
system has no invariant algebraic curves located in the finite plane.
3.3 Liouville and Elementary First Integrals 141

The above example indicates that there exist quadratic differential systems which
have a Darboux integrating factor and a Liouvillian but not elementary first integral.
In addition, the two-dimensional Lotka–Volterra system

ẋ = x(a + bx + cy), ẏ = y(d + ex + fy)

has a Darboux integrating factor x α yβ if af (e − b) = bd(c − f ) and an elementary

first integral x α+1 yα+1 L(x, y), where α is a suitable constant, and L is a polynomial
of degree one. Note that this elementary first integral is a Darboux integral.
For systems having a Darboux integrating factor with first integrals not of Darboux
type, Żoła̧dek [485] introduced the Darboux–Schwartz–Christoffel and the Darboux
hyperelliptic first integrals. These first integrals can be distinguished by holonomy
group. Note that the Darboux hyperelliptic first integrals are elementary, whereas the
Darboux–Schwartz–Christoffel first integral is not elementary. See also Christopher
[95].
Recently, Zhang [477] generalized the Singer–Christopher Theorem 3.11 of
Singer [403] and Christopher [97] to any finite-dimensional polynomial differen-
tial system. Consider an n-dimensional polynomial differential system

ẋ = P(x), x ∈ Cn , (3.43)

where P(x) = (P1 (x), . . . , Pn (x)) is a vector-valued polynomial function. The next
two results, due to Zhang [477], generalize Theorem 3.11 to higher-dimensional
differential systems.

Theorem 3.12 For the polynomial differential system (3.43), the following state-
ments hold.
(a) If system (3.43) has n − 1 functionally independent Darboux Jacobian multipli-
ers, it is Liouvillian integrable.
(b) If system (3.43) is Liouvillian integrable, it has a Darboux Jacobian multiplier.

Proof (a) Let J1 (x), . . . , Jn−1 (x) be the n − 1 functionally independent Darboux
Jacobian multipliers of system (3.43). Then they satisfy

XP (J ) = −J divXP ,  = 1, . . . , n − 1, (3.44)

where XP (x) is the vector field associated to system (3.43).

Since each J is a Darboux function, direct computations show that

∂s J /J ∈ C(x),  ∈ {1, . . . , n − 1}, s ∈ {1, . . . , n}.

Here as before, ∂s J denotes the partial derivative of the function J with respect to
xs . By the condition (b) of the Liouvillian field extension it follows that J1 , . . . , Jn−1
are Liouvillian functions. Hence we get from Proposition 2.2 that J /Jk for 1 ≤  =
k ≤ n − 1 are non-trivial Liouvillian first integrals of the vector field XP .
142 3 Darboux and Liouvillian Integrability

We claim that the n − 2 Liouvillian first integrals

J1 /Jn−1 , . . . , Jn−2 /Jn−1

are functionally independent. In fact, if

c1 ∇ (J1 /Jn−1 ) + · · · + cn−2 ∇ (Jn−2 /Jn−1 ) = 0

holds (recall that ∇ is the gradient of a function), then

c1 Jn−1 ∇J1 + · · · + cn−2 Jn−1 ∇Jn−2

−(c1 J1 + · · · + cn−2 Jn−2 )∇Jn−1 = 0.

This forces that

c1 Jn−1 = · · · = cn−2 Jn−1 = c1 J1 + · · · + cn−2 Jn−2 = 0

in a full Lebesgue measure subset of Cn , where we have used the functional inde-
pendence of J1 , . . . , Jn−1 . Consequently one has

c1 = · · · = cn−2 = 0

in a full Lebesgue measure subset of Cn . The claim follows.

By this last claim, we can assume without loss of generality that

y1 = J1 /Jn−1 , . . . , yn−2 = Jn−2 /Jn−1 , yn−1 = xn−1 , yn = xn ,

denoted by y = G(x), describe an invertible change of coordinates defined in a full

Lebesgue measure subset of Cn . This transformation sends system (3.43) in an equiv-
alent way to the system

ẏ1 = 0, . . . , ẏn−2 = 0, ẏn−1 = Pn−1 ◦ G −1 (y), ẏn = Pn ◦ G −1 (y), (3.45)

which has the first integrals Is (y) = ys , s = 1, . . . , n − 2, and the Jacobian multiplier

M(y) := Jn−1 ◦ G −1 (y)DG −1 (y),

where DG −1 (y) is the Jacobian of G −1 . Then as in the proof of the Jacobian

theorem we get that system (3.45), when restricted to any level hypersurfaces of
I1 (y), . . . , In−2 (y), has the first integral
*
In−1 (yn−1 , yn ) = V gn dyn−1 − V gn−1 dyn ,
3.3 Liouville and Elementary First Integrals 143

where V is the restriction of M(y) to any level hypersurfaces of I1 (y), . . . , In−2 (y),
i.e.
V := Jn−1 ◦ G −1 (I1 , . . . , In−2 , yn−1 , yn )DG −1 (I1 , . . . , In−2 , yn−1 , yn ),

and we treat I1 , . . . , In−2 as constants

From the construction of In−1 one gets that In−1 is a Liouvillian function, and that
I1 , . . . , In−2 , In−1 are functionally independent, because I1 , . . . , In−2 are indepen-
dent of yn−1 and yn−2 . Hence system (3.43) has the n − 1 functionally independent
Liouvillian first integrals

H1 (x) := I1 ◦ G(x), . . . , Hn−1 (x) = In−1 ◦ G(x).

This proves the statement.

(b) As in the proof of Theorem 3.11, this statement can be proved by the next two
lemmas. The first one presents an expression of a Jacobian multiplier of a Liouvillian
integrable system.

Lemma 3.9 If system (3.43) is Liouvillian integrable, it has a Jacobian multiplier

of the form * 
J = exp U1 dx1 + · · · + Un dxn ,

with Us ∈ C(x), s = 1, . . . , n, and

∂k U = ∂ Uk , 1 ≤ k <  ≤ n.

The second lemma shows the existence of Darboux Jacobian multipliers via the
Jacobian multipliers of the form given in Lemma 3.9.

Lemma 3.10 If the polynomial differential system (3.43) has a Jacobian multiplier
* 
J = exp U1 dx1 + · · · + Un dxn ,

with Us ∈ C(x), and ∂k U = ∂ Uk for 1 ≤ k,  ≤ n, then it has a Darboux Jacobian

multiplier g
exp f ci ,
h i i

where g, h, fi ∈ C[x], ci ∈ C, and the product is finite.

The proofs of these two lemmas are mostly similar to Theorem 3.11 with some
technical treatments. We refer the reader to Zhang [477] for details.
Lemmas 3.9 and 3.10 verify statement (b), and consequently the theorem. 
Finally, as an application of Singer’s theorem, we show that any planar quasi-
homogeneous polynomial differential system is Liouvillian integrable, see García
144 3 Darboux and Liouvillian Integrability

et al. [159]. A planar polynomial differential system

ẋ = P(x, y), ẏ = Q(x, y), (3.46)

with P, Q ∈ C[x, y], is quasi-homogeneous if there exist s1 , s2 , d ∈ N such that for

arbitrary α ∈ R+ = {a ∈ R| a > 0},

P(α s1 x, α s2 y) = α s1 −1+d P(x, y), Q(α s1 x, α s2 y) = α s2 −1+d P(x, y).

We call s1 and s2 the weight exponents of system (3.46), and d the weight degree with
respect to the weight exponents s1 and s2 . In particular, for s1 = s2 = 1 system (3.46)
is the classical homogeneous polynomial differential system of degree d. Recall that
a polynomial F(x, y) is quasi-homogeneous of weight degree d if

F(α s1 x, α s2 y) = α d F(x, y).

Of course, if s1 = s2 = 1 then F is a homogeneous polynomial of degree d.

A quasi-homogeneous polynomial F(x, y) of weight degree m with respect to the
weight exponents s1 and s2 satisfies the generalized Euler formula

s1 x∂x F(x, y) + s2 y∂y F(x, y) = mF(x, y).

This argument can be proved by differentiating F(α s1 x, α s2 y) = α m F(x, y) with

respect to α and then taking α = 1.

Theorem 3.13 Assume that system (3.46) is quasi-homogeneous of weight degree

d with the weight exponents s1 and s2 . Then V (x, y) = s1 xQ(x, y) − s2 yP(x, y) is
an inverse integrating factor of system (3.46), and consequently system (3.46) is
Liouvillian integrable.

Proof The first statement follows from the facts that V (x, y) is an inverse integrating
factor of system (3.46) if and only if M(x, y) := P∂x V + Q∂y V − V (∂x P + ∂y Q) ≡
0, and that

M = (s1 x∂x Q − s2 y∂x P + s1 Q)P + (s1 x∂y Q − s2 y∂y P − s2 P)Q

− (s1 xQ − s2 yP)(∂x P + ∂y Q)
 
= s1 x∂x Q + s2 y∂y Q + s1 Q P − s1 x∂x P + s2 y∂y P + s2 P Q
= ((s1 + s2 + d − 1)Q) P − ((s1 + s2 + d − 1)P) Q ≡ 0.

The second statement is a consequence of Theorem 3.11. 

3.4 Liouvillian Integrability Versus Darboux Polynomials 145

3.4 Liouvillian Integrability Versus Darboux Polynomials

In Sect. 3.3 we proved that Liouvillian integrability implies the existence of Darboux
Jacobian multipliers. In this section we will prove that the existence of a Darboux
Jacobian multiplier together with some additional conditions implies the existence
of Darboux polynomials.
Consider the polynomial differential system (3.43), i.e. the system

ẋ = P(x), x ∈ Cn ,

where P(x) = (P1 (x), . . . , Pn (x)) is an n-dimensional vector-valued polynomial.

The next result shows that a Liouvillian integrable polynomial differential system
with conditions on the degree of the system implies the existence of Darboux poly-
nomials, see Llibre et al. [293].
Theorem 3.14 Let Kj be the degree of the polynomial Pj (x1 , . . . , xn ) with respect to
the variable xk for some k ∈ {1, . . . , n}. If system (3.43) has a Darboux Jacobian mul-
tiplier and Kk > max{K1 , . . . , K̂k , . . . , Kn } + 1 for some k ∈ {1, . . . , n}, then system
(3.43) has a Darboux polynomial, where K̂k denotes the absence of Kk .
The proof of Theorem 3.14 needs the following result, which is well known to
specialists working in the Darboux theory of integrability. A published proof was
presented only recently in [293].
Proposition 3.8 If a Darboux function
g
F = f11 . . . fmm exp , (3.47)
h

with g and h coprime and fs ’s irreducible, is a Jacobian multiplier of system (3.43),

then g and fs for s = 1, . . . , m are all Darboux polynomials of system (3.43).
Proof Let XP be the vector field associated to system (3.43). Writing XP (F) =
−FdivXP concretely gives


m
 j −1 j+1
h2 j f11 . . . fj−1
j−1
fj fj+1 . . . fmm XP (fj ) (3.48)
j=1

+ f11 . . . fmm (XP (g)h − gXP (h)) = −f11 . . . fmm h2 divXP .

It follows clearly that fs divides h2 XP (fs ) in C[x] for each s ∈ {1, . . . , m}.
If fs divides XP (fs ), it is a Darboux polynomial.
If fs divides h, set h(x) = h0 (x)fs (x) with h0 a polynomial coprime with fs and 
a positive integer. Then fs divides gXP (h) by (3.48), and consequently fs divides
XP (h), where we have used the fact that fs and g are coprime because g and h are
coprime. The equality XP (h) = XP (h0 )fs + h0 fs−1 XP (fs ) shows that fs divides
XP (fs ), and so it is a Darboux polynomial of system (3.43).
146 3 Darboux and Liouvillian Integrability

Finally, the fact that h is a Darboux polynomial follows easily from (3.48) and
the fact that all fs ’s are Darboux polynomials. 
Proof of Theorem 3.14. Without loss of generality we prove the theorem for k = n
only. Set z = (x1 , . . . , xn−1 ), and write system (3.43) in the form

(j) (j) (j) K

ẋj = p0 (z) + p1 (z)xn + · · · + pKj (z)xn j , j = 1, . . . , n (3.49)

(j)
where the pi (x)’s are polynomials for j = 1, . . . , n and i = 0, 1, . . . , Kj . By the
assumption of the theorem we have pK(n)n ≡ 0.
Let J be a Darboux Jacobian multiplier of system (3.43). On the contrary,
we assume that system (3.49) has no Darboux polynomial. Then it follows from
Proposition 3.8 that J must be of the form J = eg(x) with g(x) a polynomial. Set

g(x) = g0 (z) + g1 (z)x1 + · · · + g (z)xn ,

with  ≥ 0 and g (z) ≡ 0. We obtain from XP (J) = −JdivXP that

⎛ ⎞

Kn 
 
n−1 Kj

r
ps(n) (z)xns igi (z)xni−1 + ⎝ ps (z)xn ⎠
(j) s
∂xj gi (z)xni
s=0 i=0 j=1 s=0 i=0


 ⎛ K ⎞
  n 
n−1 
Kj
=− gi (z)xni ⎝ sps(n) (z)xns−1 + ∂xj p(j) s⎠
s (z)xn . (3.50)
i=0 s=0 j=1 s=0

Since Kn > max{K1 , . . . , Kn−1 } + 1, comparing the coefficients of xnKn +−1 in the
Eq. (3.50) gives (Kn + )pK(n)n (z)g (z) ≡ 0, a contradiction. This shows that system
(3.43) has a Darboux polynomial, which completes the proof of the theorem. 
Combining Theorems 3.14 and 3.12 (b), we easily obtain the next result.
Corollary 3.2 Using the notations of Theorem 3.14, if system (3.43) is Liouvillian
integrable and Kk > max{K1 , . . . , K̂k , . . . , Kn } + 1 for some k ∈ {1, . . . , n}, then
system (3.43) has a Darboux polynomial.
We remark that Theorem 3.14 and Corollary 3.2 were first proved in [179] by
Giné et al. for two-dimensional polynomial differential systems, and then proved in
[293] for any finite-dimensional polynomial differential systems.
We remark that if Kk ≤ max{K1 , . . . , K̂k , . . . , Kn } + 1 for all k ∈ {1, . . . , n}, then
Theorem 3.14 and Corollary 3.2 may not hold.
Theorem 3.14 and Corollary 3.2 provide a sufficient condition for the existence
of Darboux polynomials of polynomial differential systems. The next result shows
that the condition is not necessary.
Proposition 3.9 If Kk > max{K1 , . . . , K̂k , . . . , Kn } + 1 does not hold for all k =
1, . . . , n, the following statements hold.
3.4 Liouvillian Integrability Versus Darboux Polynomials 147

(a) There exist examples of system (3.43) which are Liouvillian integrable and have
no Darboux polynomials.
(b) There also exist examples of system (3.43) which are Liouvillian integrable and
have Darboux polynomials.

Proof (a) The differential system in R3

ẋ = 2xz − zk , ẏ = 3yz − z , ż = −1, (3.51)

with k,  > 1 even, provides an example of statement (a), which has the two func-
tionally independent Liouvillian first integrals
* *
z e 2 z dz,
2 2 3 2 3 2
H1 (x, y, z) = xez − zk ez dz, H2 (x, y, z) = ye 2 z −

but has no Darboux polynomials. For details, see [293].

(b) The differential system in R3

ẋ = 2xz − zk , ẏ = 2yz − z , ż = −1, (3.52)

with k,  ∈ N, has the functionally independent Liouvillian first integrals

* *
z2 k z2 z2
z ez dz,
2
H1 (x, y, z) = xe − z e dz, H2 (x, y, z) = ye −

and also has a Darboux polynomial. For details, again see [293]. 

We remark that the polynomial differential system in Rn

ẋ1 = k1 x1 xn − xn1 , . . . , ẋn−1 = kn−1 xn−1 xn − xnn−1 , ẋn = −1, (3.53)

with kj , j ∈ N \ {1}, is Liouvillian integrable. But for different choices of kj and j ,

system (3.53) may or may not have a Darboux polynomial.
Applying Theorem 3.14 and Corollary 3.2 to Liénard differential systems, one
gets the next result.

Theorem 3.15 The polynomial Liénard differential systems

ẋ = y, ẏ = −f (x)y − g(x), (3.54)

with f , g ∈ C[x], are not Liouvillian integrable provided that deg f > deg g and
g(x) ≡ 0.

We remark that Llibre and Valls in [280, 285] characterized the Liouvillian inte-
grability of the Liénard polynomial differential systems ẋ = y, ẏ = −cx − f (x)y,
and of system (3.54) with deg f ≥ deg g ≥ 2, respectively. In [287] Llibre and Valls
148 3 Darboux and Liouvillian Integrability

proved that system (3.54) with deg g = deg f + 1 is not Liouvillian integrable. Here
we provide a different and easier proof than [280, 285] of the non-Liouvillian inte-
grability of Liénard differential systems, see [293].
Proof of Theorem 3.15. As is well known, we can write the Liénard differential
systems (3.54) in the form

ẋ1 = x2 − F(x1 ), ẋ2 = −g(x1 ), (3.55)

via the change of variables

* x
x1 = x, x2 = y + F(x), F(x) = f (s)ds. (3.56)
0

On the contrary, we assume that system (3.54) is Liouvillian integrable, and H(x, y)
is a Liouvillian first integral of the system. Then G(x1 , x2 ) := H(x1 , x2 − F(x1 )) is
a Liouvillian first integral of system (3.55).
The assumption deg f > deg g implies that deg F > deg g + 1. From
Corollary 3.2 it follows that system (3.55) has a Darboux polynomial. Let M(x1 , x2 )
be a Darboux polynomial of system (3.55) with cofactor K(x1 , x2 ). Then sys-
tem (3.54) has the Darboux polynomial N(x, y) := M(x, y + F(x)) with cofactor
L(x, y) := K(x, y + F(x)), which contradicts the result of Odani [350], which states
that if f , g ≡ 0, deg f ≥ deg g and g/f ≡ constant, system (3.54) has no Darboux
polynomial. This contradiction implies that system (3.54) is not Liouvillian inte-
grable.
This completes the proof of the theorem. 
Chapter 4
Existence and Degree of Darboux
Polynomials

This chapter presents results on the degree and existence of Darboux polynomials
with an emphasis on invariant algebraic curves. We also introduce some tools and
methods for characterizing the Darboux polynomials of polynomial vector fields.

4.1 The Degree of Invariant Algebraic Curves

The Darboux theory of integrability depends on the existence of Darboux polyno-

mials. However, as Poincaré noted, there is no general method to compute Darboux
polynomials. For polynomial differential systems of a given degree, if we know the
upper bound of the degrees of the Darboux polynomials, it is possible to compute
these Darboux polynomials. So we first study the degree of the invariant algebraic
curves of the planar polynomial differential systems. This is one of the classical
Poincaré problems, and it is also one of the central problems in the modern theory
of dynamical systems and algebraic geometry.
Open problem. For a planar polynomial differential system of a given degree, are
the degrees of its invariant algebraic curves (if they exist) bounded?
This problem is in fact the affine version of the problem posed by Poincaré [362]
in 1891 for the upper bound of invariant algebraic curves of singular holomorphic
foliations in the projective plane. The details will be given in Sect. 4.1.2.
We will see that the answer to this open problem depends on concrete differential
systems.

© Springer Nature Singapore Pte Ltd. 2017 149

X. Zhang, Integrability of Dynamical Systems: Algebra and Analysis,
Developments in Mathematics 47, DOI 10.1007/978-981-10-4226-3_4
150 4 Existence and Degree of Darboux Polynomials

4.1.1 Examples of Invariant Algebraic Curves

of Arbitrary Degree

The linear differential system

ẋ = kx, ẏ = y, k,  ∈ N and gcd(k, ) = 1,

has a rational first integral H = y k /x  , and so it has invariant algebraic curves

y k − cx  = 0, c ∈ C. This example shows that linear differential systems may have
invariant algebraic curves of arbitrary degree.
In 2001 Llibre and Świrszcz [272] constructed the Liénard differential system of
degree 3

4n 8 16
ẋ = y, ẏ = −2(2x + 1)y + x − x 2 − x 3 , n ∈ N \ {1},
(n − 1)2 3 9

which has an invariant algebraic curve f = 0 of degree 2n + 2 with

  n
f = (1 − n)1+n −6x + 4(n − 1)x 2 + 3(n − 1)y 6nx + 4(n − 1)x 2 + 3(n − 1)y
 n+1
− 9n − (n − 1)2 (6x + 4x 2 + 3y) ,

and a rational first integral H = f (x, y)/g(x, y)n+1 , where

g := 9n − (n − 1)2 (6x + 4x 2 + 3y) = 0

is also an invariant algebraic curve.

In 2002 Christopher and Llibre [100] provided the quadratic differential system

ẋ = x(1 − x), ẏ = −λy + Ax 2 + Bx y + y 2 ,

with
ab(c − a)(c − b) 2ab
λ = c − 1, A = 2
, B =a+b−1− ,
c c
and
a = 1 − k, k ∈ N, b ≤ a, c ∈ R \ Q,

which
• has no rational first integrals; but
• has a Darboux integrating factor x c f 2 (1 − x)a+b+1−c ; and
• an irreducible invariant algebraic curve of degree k
4.1 The Degree of Invariant Algebraic Curves 151
 
ab
f := y− x F + x(1 − x)F  = 0,
c

where F = F(a, b; c; x) is the hypergeometric function

k−1
(a) j (b) j x j
F(a, b;c; x) = ,
j=0
(c) j j!

1, j = 0,
(c) j =
c(c + 1)(c + 2) . . . (c + j − 1), j > 0,

which is a solution of the hyperbolic geometric differential equation

x(1 − x)y  + (c − (a + b + 1)x)y  − aby = 0.

From these last examples, Christopher and Llibre [100] conjectured: For a
polynomial differential system of degree m, there exists a positive integer N (m)
such that if the system has an irreducible invariant algebraic curve of degree greater
than N (m), then it has a Darboux first integral or a Darboux integrating factor.
Note that the integrating factor in the example of Christopher and Llibre [100]
does not contain exponential factors. Chavarriga and Grau [73] in 2003 constructed
another example of a quadratic differential system

ẋ = 1, ẏ = 2n + 2x y + y 2 , n ∈ N,

which has a unique irreducible invariant algebraic curve of degree n + 1

f (x, y) := Hn (x)y + 2n Hn−1 (x) = 0, with the cofactor k = 2x + y,

and a Darboux integrating factor e−x f (x, y)−2 , where Hn (x) is a Hermite polyno-
2

mial of degree n, and satisfies the following properties (see e.g. Abramowitz and
Stegun [2]):

Hn (x) = 2n Hn−1 (x),

Hn (x) = 2x Hn (x) − 2n Hn (x),
Hn (x) = 2x Hn−1 (x) − 2(n − 1)Hn−2 (x).

The above examples indicate that if we do not have any restriction on a polynomial
differential system, it is not possible to bound the degrees of its invariant algebraic
curves.
152 4 Existence and Degree of Darboux Polynomials

4.1.2 Invariant Algebraic Curves in the Projective Plane

To apply the tools of algebraic geometry, we first recall some preliminary facts
from projective planar geometry. Since the real projective plane is a subspace of the
complex projective plane, in what follows we only consider the complex projective
plane, denoted by CP2 .
The complex projective plane is the quotient space formed by the straight lines in
C3 passing through the origin but without the origin, i.e.

CP2 = [Z ]| 0  = Z ∈ C3 , Z  ∈ [Z ] if and only if ∃ 0  = λ ∈ C such that Z  = λZ .

We call Z = (Z 1 , Z 2 , Z 3 ) homogeneous coordinates of CP2 , which we also write as

[Z 1 : Z 2 : Z 3 ].
The complex n-dimensional projective space CPn can be defined similarly. Note
that CPn is a complex n-dimensional manifold, which can be covered by the n + 1
local coordinate charts


Ui = Z ∈ CPn | Z i = 0 , i = 1, . . . , n + 1.

In Ui , we define the local coordinate system

Z1 Z i−1 Z i+1 Z n+1

z1 = , . . . , z i−1 = , zi = , . . . , zn = .
Zi Zi Zi Zi

This establishes a holomorphic diffeomorphism between Ui and Cn , which follows

from the facts that for Z i = 0,

(Z 1 , . . . , Z i , . . . , Z n ) and (Z 1 /Z i , . . . , Z i−1 /Z i , 1, Z i+1 /Z i , . . . , Z n /Z i )

denote the same point in Ui ⊂ CPn , and that

ϕ: Ui −→ Cn
(Z 1 , . . . , Z i , . . . , Z n ) −→ (Z 1 /Z i , . . . , Z i−1 /Z i , Z i+1 /Z i , . . . , Z n /Z i )

is a holomorphic diffeomorphism between Ui and Cn . In addition, the transformation

between the two coordinate charts Ui and U j is also holomorphic (in fact it is a
fractional linear transformation). This shows that CPn is a complex n-dimensional
manifold.
According to Cerveau and Lins Neto [61], a singular holomorphic foliation F in
CP2 can be given in at least three different ways:
• In the affine coordinates (x, y) ∈ C2 , F is given by a polynomial vector field

∂ ∂
XA := p(x, y) + q(x, y) , gcd( p, q) = 1. (4.1)
∂x ∂y
4.1 The Degree of Invariant Algebraic Curves 153

Its singular set is the variety VA := {(x, y) ∈ C2 | p(x, y) = q(x, y) = 0}. The
holomorphic solutions of the vector field (4.1) define the leaves of F in C2 \ VA .
• In affine coordinates (x, y) ∈ C2 , F can also be given as a polynomial differential
one-form
ωA := p(x, y)dy − q(x, y)d x, gcd( p, q) = 1.

The singular set is VA := {(x, y) ∈ C2 | p(x, y) = q(x, y) = 0}. The leaves of F

in C2 \ VA are the holomorphic solutions of the differential equation ωA = 0.
• In homogeneous coordinates [X : Y : Z ] ∈ CP2 , F is given by a polynomial dif-
ferential one-form

ω := P(X, Y, Z )d X + Q(X, Y, Z )dY + R(X, Y, Z )d Z , (4.2)

which satisfies the projective condition:

X P + Y Q + Z R = 0,

where P, Q and R are homogeneous polynomials of the same degree and have
no common factors. The singular set is

SingF = {(X, Y, Z ) = 0| P(X, Y, Z ) = Q(X, Y, Z ) = R(X, Y, Z ) = 0}.

We will call ω a projective holomorphic one-form.

In what follows we will say that the holomorphic foliation F is represented by XA ,
or ωA , or ω.
For the projective one-form ω, there exist homogeneous polynomials L , M and
N such that

P = Z M − Y N, Q = X N − Z L, R = Y L − X M.

So the one-form ω can be written as

ω = L(Y d Z − Z dY ) + M(Z d X − X d Z ) + N (X dY − Y d X ).

Thus we get a vector field in CP2 associated to F

∂ ∂ ∂
X =L +M +N . (4.3)
∂X ∂Y ∂Z

If the homogeneous polynomials (L  , M  , N  ) also define ω, i.e.

P = Z M  − Y N , Q = X N  − Z L , R = Y L  − X M ,
154 4 Existence and Degree of Darboux Polynomials

then

(M  − M)Z − (N  − N )Y = (N  − N )X − (L  − L)Z
= (L  − L)Y − (M  − M)X = 0.

This implies that there exists a homogeneous polynomial R of degree m − 1 such

that
L  = L + X R, M  = M + Y R, N  = N + Z R.

Hence, studying the projective one-form (4.2) of degree m + 1 in CP2 is equivalent

to studying the projective vector field (4.3) of degree m.

Remark From the projective condition of the projective one-form ω we can get the
integrability condition: ω ∧ dω = 0. The projective condition can also be written as

ω; R = 0,

where
∂ ∂ ∂
R=X +Y +Z
∂X ∂Y ∂Z
is the radial vector field.
If a holomorphic foliation F is represented as an affine differential one-form


ω = q(x, y)d x − p(x, y)dy, m = max{deg p, deg q},

then in homogeneous coordinates [X : Y : Z ] the affine one-form 

ω, via the trans-
formation x = X/Z , y = Y/Z , can be written as the projective one-form

ω = Z Q(X, Y, Z )d X − Z P(X, Y, Z )dY + (Y P(X, Y, Z ) − X Q(X, Y, Z ))d Z ,

where

P(X, Y, Z ) = Z m p(X/Z , Y/Z ), Q(X, Y, Z ) = Z m q(X/Z , Y/Z ).

Note that the components of ω are all homogeneous polynomials of degree m + 1.

For a given singular holomorphic foliation F , its degree is defined as follows:
taking the affine coordinates (x, y), we write F as the affine one-form


ω = P(x, y)dy − Q(x, y)d x = 0.

Let L be a straight line in C2 which is not invariant under the flow of F . Then the
number of points on L which are either singularities of F or at which the flow of F
is tangent to L is finite, denoted by m, and it does not exceed max{deg P, deg Q}.
Lins Neto [243] in 1988 proved that
4.1 The Degree of Invariant Algebraic Curves 155

• for a generic straight line L, the number m is a constant, which is called the degree
of F ,
• if F is of degree m, and has the expression  ω, then

P(x, y) = p(x, y) + xr (x, y), Q(x, y) = q(x, y) + yr (x, y),

where either r (x, y) ≡ 0 and max{deg p, deg q} = m; or r (x, y) is a homoge-

neous polynomial of degree m and max{deg p, deg q} ≤ m.
Let F(X, Y, Z ) be a reduced homogeneous polynomial. Here reduced means that
F has no repeated factors. By definition F = 0 is an invariant algebraic curve of F
given by a projective holomorphic one-form ω if and only if

d F ∧ ω = Fθ,

where θ is a differential two-form with coefficients homogeneous polynomials of

the same degree.
In what follows, all invariant algebraic curves are defined by the reduced poly-
nomials (if not otherwise specified). In algebraic geometry, an invariant algebraic
curve is also called a separatrix.
Let F be a singular holomorphic foliation of degree m represented by an affine
expression

∂ ∂
XA = ( p(x, y) + xr (x, y)) + (q(x, y) + yr (x, y)) .
∂x ∂y

Set
     
X Y X Y X Y
P(X, Y, Z ) = Z m p , , Q(X, Y, Z ) = Z m q , , R(X, Y, Z ) = Z m r , .
Z Z Z Z Z Z

Then by Chavarriga and Llibre [74] and Zhang [463] the projective vector field
associated to XA is

∂ ∂ ∂
XP = P(X, Y, Z ) + Q(X, Y, Z ) − R(X, Y, Z ) ,
∂X ∂Y ∂Z
and its projective differential one-form is

ω = (−Y R − Z Q)d X + (Z P + X R)dY + (X Q − Y P)d Z .

Denote by ∂ 0 the degree of polynomials, or of vector fields, or of singular holomorphic

foliations, or of projective holomorphic one-forms. Then ∂ 0 ω = ∂ 0 F + 1, where
ω is the projective holomorphic one-form representing the singular holomorphic
foliation F .
156 4 Existence and Degree of Darboux Polynomials

Zhang [463, Theorem 1] in 2003 presented an equivalent characterization of the

invariant algebraic curves of holomorphic foliations in the projective plane.
Proposition 4.1 Let F(X, Y, Z ) = 0 be an invariant algebraic curve of degree n of
the singular holomorphic foliation F in CP2 . The following statements hold.
(a) There exists a differential two-form θ for which d F ∧ ω = Fθ if and only if
there exists a homogeneous polynomial K (X, Y, Z ) such that

XP (F) = K F.

(b) The above equality holds if and only if

XA ( f ) = k f,

where f (x, y) = F(X, Y, Z )| Z =1 and k(x, y) = n R(X, Y, Z ) + K (X,

Y, Z )| Z =1 .
Proof For a proof, see [463]. 

In the next section we will discuss the degree of invariant algebraic curves of
singular holomorphic foliations in CP2 .

4.1.3 The Degree of Invariant Algebraic Curves in the Nodal

and Nondicritical Cases

A point on a curve in a real or complex plane is nodal if at the point the curve, as a
complex curve, is of normal crossing type, that is, it has exactly two branches and
intersects transversally at this point. A planar curve is nodal if all its singularities are
nodal. Recall that a singularity of a curve is a point where the curve is not smooth or
the tangent vector of the curve at this point is not well defined.
The next result, due to Cerveau and Lins Neto [61, Theorem 1], provided an upper
bound on the degree of invariant nodal algebraic curves.
Theorem 4.1 (Cerveau–Lins Neto theorem) Let F be a singular holomorphic foli-
ation of degree m in CP2 . Assume that F = 0 is a reduced invariant nodal algebraic
curve of degree n of the foliation F . The following statements hold.
(a) n ≤ m + 2.
(b) If n = m + 2, then F is reducible and the foliation F is of logarithmic type,

k
that is, it can be expressed in the rational closed form λi dFFi i , where λi ∈ C,
i=1
and the Fi ’s are the irreducible factors of F, i.e. F = F1 . . . Fk .
Remark From statement (b) it follows easily that any irreducible invariant nodal
algebraic curve has degree at most m + 1.
4.1 The Degree of Invariant Algebraic Curves 157

The upper bound in (a) is optimal. The following example shows that the
upper bound can be achieved by a concrete foliation. Take F = F1 . . . Fk with
F1 , . . . , Fk ∈ C[X, Y, Z ] irreducible and pairwise coprime homogeneous polyno-
mials, such that each curve Fi = 0 has no singularity in CP2 , and they intersect
transversally (if they intersect) and without three curves intersecting at the same
point. Choose λ1 , . . . , λk ∈ C such that the λi /λ j ’s are nonnegative rational num-
bers, and satisfy
 k
λi ∂ 0 Fi = 0.
i=1

k
Then the homogeneous differential form F λi d Fi
Fi
is projective, and the sum of
i=1
the degrees of the invariant algebraic curves of the foliation F associated to the
differential form is ∂ 0 F + 2 because the holomorphic foliation has a Darboux first
integral
H = F1λ1 . . . Fkλk ,

and the invariant curves defined by H = c = 0 are not algebraic. For instance, let f 1
be a nonsingular curve of degree m, and f 2 and f 3 be two different straight lines. For
λ1 = 1, λ2 = π, λ3 = −m − π , the curves defined by H = f 1 f 2π f 3−m−π = c = 0,
are not algebraic.
The Cerveau–Lins Neto theorem provided a uniform upper bound on the degree
of the invariant nodal algebraic curves. In 1994 Carnicer [59] also obtained the same
upper bound for the invariant algebraic curves which can have only nondicritical
singularities of the holomorphic foliation.
A singularity of a holomorphic foliation is dicritical if there are infinitely many
invariant curves of the foliation passing through the singularity. Note that a singularity
of an invariant algebraic curve of a holomorphic foliation will be a singularity of the
foliation, whereas a singularity of a foliation located on an invariant algebraic curve
is not necessary a singularity of the invariant curve.

Theorem 4.2 (Carnicer’s theorem) Let F be a singular holomorphic foliation of

degree m in CP2 . If F = 0 is a reduced invariant algebraic curve of degree n of the
foliation F , and has no dicritical singularities on it, then

n ≤ m + 2.

In the proofs of Theorems 4.1 and 4.2, the main techniques are the blow up, the
relation between the multiplicities of the singularities of a foliation and of the singu-
larities of the invariant algebraic curves, and the expression of the Euler characteristic
of these curves via their multiplicities. For more information on these notions and
techniques, see e.g. Ilyashenko and Yakovenko [213] and Carnicer [59].
Next we shall state the main ideas and steps which will be used in the proof of
Theorem 4.2.
158 4 Existence and Degree of Darboux Polynomials

For a given singular holomorphic foliation F and its invariant algebraic curve S,
assume that p ∈ S is a singularity of F , and B is a branch of S passing through p.
Since S is algebraic, its branch B is analytic. The next result, due to Cerveau and
Lins Neto [61], shows the existence of the Puiseux parametrization of B, see e.g.
Chirka [90] and Dimca [125].

Proposition 4.2 Assume that B ⊂ C2 is an irreducible complex one-dimensional

analytic subset containing the origin 0. Then there exists an analytic (holomorphic)
one to one map φ : D → φ(D) ⊂ B, where D ⊂ C is the unit disc, such that φ(0) =
0, φ(D) is a neighborhood of the origin, and φ : D \ {0} → φ(D) \ {0} is an analytic
diffeomorphism. Moreover, there exists a system of local coordinates (z, w) in C2
such that φ can be expressed as

φ : D → B,
 ∞


s → (z, w) = (φ1 (s), φ2 (s)) = s , k
an s n
, m > k.
n=m

Recall that a Puiseux series is a series of the form

∞

f (z) = ck z k/n ,
k=k0

where n ∈ N, k0 ∈ Z. In particular, if n = 1, the Puiseux series is in fact the Laurent

series; if further k0 ≥ 0 then it is the usual power series. The local expression of the
irreducible analytic curve B in Proposition 4.2 can be written as the Puiseux series
∞

w= an z n/k ,
n=m

which is called the Puiseux parametrization of the analytic curve B.

A simple example of the Puiseux parametrization of analytic curves is for y 2 = x 3 .
It has the Puiseux parametrization s → (s 2 , s 3 ). One of the two branches at the origin
has the Puiseux series expression: y = x 3/2 .
Let S be an invariant algebraic curve of a holomorphic foliation F , and B be
a branch of S passing through a singularity p. Choose a minimal Puiseux parame-
trization of B as φ(t) = (φ1 (t), φ2 (t)) : D → C2 satisfying φ(0) = 0, where D is
the unit disk centered at the origin 0 in C. The multiplicity of the foliation F on B
at p is the order of
d
φ ∗ (X ) = R(t)
dt

at t = 0 ∈ D, denoted by μ p (F , B), where X is a vector field representing the

foliation F in a neighborhood of p. Furthermore, we have
4.1 The Degree of Invariant Algebraic Curves 159

1 d R(t)
μ p (F , B) = ,
2πi γ (B) R(t)

where γ (B) is the homology class of the curve θ → φ(r eiθ ) (0 < r < 1), 0 ≤ θ ≤
2π , in H1 (B \ { p}, t). For more information on the homology class, see e.g. Hatcher
[200, pp. 105–106].
Next we define the multiplicity of a vector field at a singularity. Let M be a two-
dimensional smooth surface, and V be a smooth vector field on M. Assume that
q ∈ M is an isolated singularity of V , and D is a closed ball centered at q, in which
there is no other singularities of V . The index of V at q, denoted by indV (q), is the
V (z)
degree of the map u(z) = : ∂ D → S1 , where ∂ D denotes the boundary of
V (z)
D. Roughly speaking, the degree of a continuous mapping between two compact
oriented manifolds of the same dimension is the number of times that the domain
manifold wraps around the range manifold under the mapping. For more information,
see e.g. Hirsch [206] and Milnor [333].
Poincaré–Hopf theorem. Let M be a compact oriented surface of dimension two,
and V be a smooth vector field on M with only isolated singularities. If M has a
boundary, we assume that the vector field V points in the outer normal direction
along the boundary. Then the Euler characteristic of M is equal to the sum of the
indices of all singularities of the vector field V on M, i.e.

χ (M) = indV (q),
q

where the sum runs over all singularities of V on M.

The Euler characteristic of a two-dimensional compact surface M (closed or
having a boundary) can be defined as follows: Taking any triangular splitting of M
(that is, splitting M into triangles with curved line boundaries), we denote by α0 , α1
and α2 respectively the number of vertices, boundaries and triangles of those curved
line triangles. Then α0 − α1 + α2 is independent of the splitting of M. This constant
is called the Euler characteristic of the surface M, denoted by χ (M).
In algebraic topology, the Euler characteristic of the surface M is determined by
the first, second and third Betti numbers, denoted by b0 , b1 and b2 , of the surface,
i.e. χ (M) = b0 − b1 + b2 . Recall that the ith Betti number is the dimension of the
ith homology group of the surface. For more information, see e.g. Hirsch [206].
For an irreducible invariant algebraic curve S in CP2 , using the blow up technique
on singularities of S (see e.g. Laufer [225] and Ilyashenko and Yakovenko [213] for
details on blow up) one can prove (see e.g. Lins Neto [243, p.195]) that there exists
a compact Riemann surface, say S  , and a holomorphic map π : S  → S such that π
restricted to S  \ π −1 (Sing(S)) is a diffeomorphism. By definition the intrinsic Euler
characteristic of S is the Euler characteristic of S  .

Proposition 4.3 Assume that F is a singular holomorphic foliation of degree m

in CP2 , and S is a reduced irreducible invariant algebraic curve of degree n of F .
160 4 Existence and Degree of Darboux Polynomials

Denote by χ (S) the intrinsic Euler characteristic of S, and by g(S) the topological
genus of S. Then

χ (S) = 2 − 2g(S) = μ p (F , B) − n(m − 1), (4.4)
B

where the sum runs over all the local branches B of S passing through the singu-
larities of F , and μ p (F , B) denotes the multiplicity of F at B passing through a
singularity p.

Proof The main part of the proof follows from Cerveau and Lins Neto [61] with
some corrections by Zhang [471].
Since S is an invariant algebraic curve of degree n, we can choose affine coor-
dinates (x, y) such that S intersects the line l∞ at infinity at exactly n points and
transversally. Let the vector field

∂ ∂
X = P(x, y) + Q(x, y)
∂x ∂y

be an expression of F in affine coordinates (x, y). We assume without loss of

generality that p = [1 : 0 : 0] ⊂ S ∩ l∞ . Take the change of coordinates u = y/x
and v = 1/x around the singularity p, under which the vector field X can be written
as  
 
X = v−m+1 −u P(u,  v) + Q(u, v) ∂ − v P(u, v) ∂ ,
∂u ∂v

where
 v) = vm P (1/v, u/v) ,
P(u,  v) = vm Q (1/v, u/v) .
Q(u,

Since S and l∞ intersect transversally, the local branch B∞ of S passing through the
singularity [1 : 0 : 0] has an analytic expression u = ψ(v) in the coordinates (u, v).
This means that h(u, v) := u − ψ(v) = 0 is an analytic solution of the vector field
X. So there exists a local analytic function k(u, v) such that
 
 v) + Q(u,
−u P(u,  v)∂v h = kh.
 v) ∂u h − v P(u, (4.5)

in the new coordinates ξ = u − ψ(v) and η = v gives

Writing the vector field X
 
  ∂ ∂
X ∗ = η−(m−1) −(ξ + ψ(η))P ∗ + Q ∗ + ηψ  (η)P ∗ − η P∗ ,
∂ξ ∂η

where
 + ψ(η), η),
P ∗ (ξ, η) = P(ξ  + ψ(η), η),
Q ∗ (ξ, η) = Q(ξ

and equation (4.5) in the (ξ, η) coordinates becomes

4.1 The Degree of Invariant Algebraic Curves 161

−(ξ + ψ(η))P ∗ + Q ∗ + η P ∗ ψ  (η) = k ∗ ξ,

where k ∗ (ξ, η) = k(ξ + ψ(η), η). This proves that along B∞ the vector field X ∗
has the expression  
∗ −(m−1) ∗ ∂ ∗ ∂
X =η ξk − ηP .
∂ξ ∂η

∗ (η) with P
Set P ∗ |ξ =0 = η P ∗ (0) = 0 and  ≥ 0, and set

∗
P
∗ η2+−m =
ζ := P ηm−2− ,
|η|2(m−2−)

where η is the conjugate

 of η. Then the multiplicity or the order of the pole of the
∗ 
vector field X at η = 0 is
ξ =0

 
1 dζ m−2− dη
= = 2 +  − m,
2πi C ζ 2πi C η

where C is the homology class H1 (B∞ , η) of the curve θ → φ(r eiθ ) on ξ = 0,

0 ≤ θ ≤ 2π . It follows from the expression of X ∗ that μ p (X , B∞ ) =  + 1.
Let π :  S → S be the σ process which blows up the singularities of S, see e.g.
[213]. Then  S is smooth and compact, and the Euler characteristic of  S satisfies
χ ( S) = 2 − 2g(S) (see e.g. Cerveau and Lins Neto [61]). Then 
 S is invariant under
the flow of the pull back vector field X  := π ∗ ( X | S ) of X under the action of π .
Let p be the singularity of the vector field X  on  S corresponding to the singularity
p of F on S, and let B be the unique branch of 
 S passing through 
p corresponding
to the branch B of S passing through p. Then the Poincaré–Hopf index of X  on  B
p is μ p (F , B).
at 
Let p1 , . . . , pn be n points in l∞ ∩ S. For each pi , we denote by li the quantity l
associated to the singularity pi as chosen in the above proof. By the Poincaré–Hopf
index theorem we have

 
n
χ (S) = μ p (F , B) − (m − 2 − li ),
B i=1

where B is the set of the branches of S passing through the singularities in the finite
plane. Taking into account μ pi (X , B∞ ) = li + 1 we get

χ (S) = μ p (F , B) − n(m − 1),
B

where B runs over all the local branches of S passing through the singularities in the
projective plane.
This completes the proof of the proposition. 

162 4 Existence and Degree of Darboux Polynomials

The next example illustrates Proposition 4.3, see Zhang [471]. The foliation F3
in CP2 given by the homogeneous differential form

(X 3 − 2Y 2 Z )Z d X − X (Y 2 + Z 2 )Z dY − (X 4 − 2X Y 2 Z − X Y Z 2 − X Y 3 )d Z

has X = 0 as an invariant line, on which there are exactly two singularities P1 =

[0 : 1 : 0] and P2 = [0 : 0 : 1] of F . The foliation F3 | X =0 at P1 and P2 can be
∂
represented respectively by the vector fields −2z 2 ∂z and −2y 2 ∂∂y . These show that
μ P1 (F3 , X = 0) = 2 and μ P2 (F3 , X = 0) = 2. In addition, the foliation F3 is of
degree 3. Then Proposition 4.3 verifies χ (x = 0) = 2.
The equality (4.4) has a well-known consequence.

Corollary 4.1 If S is an irreducible nonsingular invariant algebraic curve of degree

n, then the Euler characteristic of S satisfies χ (S) = −n(n − 3).

Proof The proof follows an idea of Zhang [471]. Taking a system of local coor-
dinates (x, y) in CP2 such that S intersects the line at infinity l∞ transversally. Let
p1 , . . . , pn be the intersection points of S with l∞ , and let B1 , . . . , Bn be respectively
the branches of S passing through the n points. Take f ∈ C[x, y] as an expression
of S in the affine plane, and denote by G f the holomorphic foliation induced by the
projectivization in CP2 of the differential 1-form d f . Clearly G f is of degree n − 1.
Applying the equality (4.4) to the foliation G f one gets


n
μ pi (G f , Bi ) = χ (S) + n(n − 2). (4.6)
i=1

Since S is a nonsingular curve, and μ pi (G f , Bi ) = 1, i = 1, . . . , n, it follows from

(4.6) that n = χ (S) + n(n − 2). This proves the corollary. 

In the proof of Theorem 4.2 we will also need the next result. For a proof, see
Carnicer [59, Lemma 2.2].

Lemma 4.1 Let S be a two-dimensional analytic manifold, and F be a singular

holomorphic foliation formed by curves of S. Assume that p ∈ S is a nondicritical
singularity of F , and C is an invariant analytic curve of F passing through p. Let
f (x, y) be the reduced equation of C at p, and G be the foliation defined by the
differential 1-form d f in a neighborhood of p. Then on any branch B of C at p it
holds that
μ p (F , B) ≥ μ p (G , B).

Proof of Theorem 4.2. The main idea follows from Carnicer [59]. Let C be an invariant
algebraic curve of F defined by F = 0. Choose an affine coordinate system (x, y)
of CP2 for which C intersects transversally the line at infinity l∞ at exactly n points.
Let q1 , . . . , qn ∈ C ∩ l∞ be the intersection points. Then the branch of C passing
through qi is an irreducible analytic curve, denoted by Bi .
4.1 The Degree of Invariant Algebraic Curves 163

Let f ∈ C[x, y] be reduced and represent C in the affine plane. Denote by G the
foliation induced by the projectivization in CP2 of the differential form d f in C2 .
Then ∂ 0 G = n − 1 and μqi (G , C) = 1. If q ∈ C, we denote by C{q} the set of all
irreducible branches of C passing through q. For q ∈ C and q = qi , i = 1, . . . , n,
and B ∈ C{q}, Lemma 4.1 shows that μq (F , B) ≥ μq (G , B). Note that if q is not
a singularity of C then μq (F , B) ≥ μq (G , B) = 1.
Let C1 , . . . , Cr be all the branches of C. We get from Proposition 4.3 that
⎛ ⎞

r  
χ (Ci ) + n(m − 1) = ⎝ μ p (F , B)⎠
i=1 q∈C∩Sing(F ) B∈C{ p}
⎛ ⎞
 
≥ ⎝ μ p (G , B)⎠ − n
q∈C∩Sing(F ) B∈C{ p}


r
= χ (Ci ) + n(n − 3),
i=1

where the inequality follows from Lemma 4.1 and μqi (G , C) = 1. This proves n ≤
m + 2, and consequently the theorem. 

We remark that if a singular holomorphic foliation F of given degree in CP2 has
an invariant algebraic curve S, then the degree of S depends on the concrete type
of singularities on S. For instance, Campillo et al. [54, 55] obtained the following
estimate
∂ 0 C ≤ ∂ 0 F + 2 + a,

where a is an integer determined by the singularities on S.

Walcher [437] in 2000 derived some restrictions on the degree of invariant alge-
braic curves, and also on the possible form of the algebraic integrating factors, for
planar polynomial vector fields whose singularities at infinity satisfy a certain generic
condition.

4.2 Existence of Darboux Polynomials

When applying the Darboux theory of integrability, the key point is to find a suf-
ficient number of Darboux polynomials (or invariant algebraic curves, surfaces or
hypersurfaces). However, detecting the existence of Darboux polynomials of a given
polynomial differential system is extremely difficult, as Poincaré pointed out. This
section introduces some methods for studying the existence of Darboux polynomials
for some kinds of classical differential systems.
164 4 Existence and Degree of Darboux Polynomials

4.2.1 Polynomial Vector Fields Without Darboux Polynomials

Jouanolou [217] in 1979 proved that the planar system

   
ω = x n − y n+1 d x − 1 − x y n dy


has no invariant algebraic curves for n ≥ 2. Cerveau and Lins Neto [61] in 1991
provided a simple proof of this result.
Lins Neto and Soares [244, Theorems I and II] in 1996 proved the following
results on the nonexistence of invariant algebraic hypersurfaces.

Theorem 4.3 For the n (≥ 2)-dimensional vector field of degree d (≥ 2)


n−1
  ∂   ∂
X0 = d
xi+1 − xi x1d + 1 − xn x1d ,
i=1
∂ xi ∂ xn

n
∂
Xμ = μ xi + X0 , μ ∈ C,
i=1
∂ xi

let F0 and Fμ be the foliations in CPn defined by the projectivization of X0 and of

Xμ , respectively. The following statements hold.
• If n is even, F0 has no invariant algebraic hypersurfaces (or algebraic solutions).
• If n is odd and 0 < |μ| ≤ 1, Fμ has no invariant algebraic hypersurfaces.

Theorem 4.4 Denote by ℵd the set of one-dimensional holomorphic foliations of

degree d in CPn (n ≥ 2). If d ≥ 2, there exists an open and dense subset Jd of ℵd ,
in which no foliation has invariant algebraic hypersurfaces.

Related to this last result, Falla Luza [143] in 2011 proved that in the set of second-
order differential equations in CP2 , those having no algebraic solution form a dense
subset.
In 2008 Moulin-Ollagnier and Nowicki [341] provided a new method of studying
the nonexistence of Darboux polynomials of a given vector field. We now recall their
results.
Let R be a commutative ring and let d : R −→ R be a derivation. The ring of
constants of d is by definition C(d) := { f ∈ R| d( f ) = 0}. An element f ∈ R is a
Darboux element of d if f = 0 and it is not invertible over R, and satisfies d( f ) = k f
for some k ∈ R. Clearly, if R is a polynomial ring, then a Darboux element is a
Darboux polynomial.
The next result, due to [341, Theorem 2.1], provides an equivalent characterization
of the nonexistence of Darboux elements.

Theorem 4.5 Let R be a commutative ring containing Q, and let R[t] be the poly-
nomial ring in one variable t over R. Assume that
4.2 Existence of Darboux Polynomials 165

• d : R −→ R is a derivation without Darboux elements,

• D : R[t] −→ R[t] is a derivation such that D( f ) = d( f ) for f ∈ R.
The following statements hold.
(i) If D(t) = at + b for some a, b ∈ R, then the derivation D has no Darboux
element if and only if there does not exist an element f ∈ R such that d( f ) =
a f + b.
(ii) If D(t) = b for some b ∈ R, then the derivation D has no Darboux element if
and only if b ∈
/ d(R).

Proof Statement (ii) is a special case of statement (i) with a = 0. So we only prove
(i) following [341].
Necessity. On the contrary, we assume that there exists an f ∈ R such that d( f ) =
a f + b. Then D(t − f ) = a(t − r ). This shows that t − f is a Darboux element of
D, a contradiction.
Sufficiency. On the contrary, we assume that D has a Darboux element f ∈ R[t].
Then f = 0 is not invertible in R[t] and D( f ) = k f for some k ∈ R[t]. Since d
does not admit a Darboux element, it follows that f ∈ / R. Let n be the degree of f
in t. Then n ≥ 1. Set

f = f n t n + f n−1 t n−1 + · · · + f 1 t + f 0 ,

where f n , . . . , f 1 , f 0 ∈ R and f n = 0. We have

D( f ) = (d( f n ) + na f n )t n + gn−1 t n−1 + · · · + g1 t + g0 ,

for some gn−1 , . . . , g1 , g0 ∈ R. Since deg(D( f )) ≤ n in t, it follows from D( f ) =

k f that k ∈ R. Equating the coefficients of t n in D( f ) = k f gives that d( f n ) +
na f n = k f n , i.e.
d( f n ) = (k − na) f n .

Since f n = 0, if f n is not invertible in R, it is a Darboux element of d, a contra-

diction. So f n must be invertible in R. Set F = f f n−1 . Then F ∈ R[t] is a monic
polynomial of degree n, and it is a Darboux element of D because D(F) = ν F with
ν = f n d( f n−1 ) + k.
Set
F := t n + cn−1 t n−1 + cn−2 t n−2 + · · · + c1 t + c0 ,

where cn−1 , . . . , c1 , c0 ∈ R. Easy calculations show that

D(F) = nat n + μn−1 t n−1 + · · · + μ1 t + μ0 ,

with μn−1 = d(cn−1 ) + nb + (n − 1)acn−1 . Set

G := D(F) − na F.
166 4 Existence and Degree of Darboux Polynomials

Then in one way G = ν F − na F = (ν − na)F and in another way

G = (μn−1 − nacn−1 )t n−1 + σn−2 t n−2 + · · · + σ1 t 1 + σ0 ,

with σn−2 , . . . , σ1 , σ0 ∈ R. Since G is of degree at most n − 1 and F is of degree n,

it follows that G ≡ 0. Hence, we have d(cn−1 ) + nb − acn−1 = 0, i.e.
 c   c 
n−1 n−1
d − =a − + b.
n n
This contradicts the sufficient assumption. Hence D has no Darboux element.
The proof of the theorem is complete. 

Using Theorem 4.5, Moulin-Ollagnier and Nowicki [341] proved that the vector
fields
∂ ∂ ∂ ∂ s−1 ∂
x2s + x3s + x1s + x2 x3s−1 + · · · + xn−2 xn−1 , (4.7)
∂ x1 ∂ x2 ∂ x3 ∂ x4 ∂ xn

with n ≥ 4, and

∂ ∂ ∂ ∂ s−2 ∂
x2s + x3s + x1s + x1 x2 x3s−2 + · · · + xn−3 xn−2 xn−1
∂ x1 ∂ x2 ∂ x3 ∂ x4 ∂ xn

both have no Darboux polynomials. The proof was done by induction using Jouano-
lou’s result [217], which shows that the vector field

∂ ∂ ∂
X J := x2s + x3s + x1s , s ≥ 2,
∂ x1 ∂ x2 ∂ x3

has no Darboux polynomial. In fact, denote by δn the vector field associated to (4.7),
and set δ3 = X J . By induction we assume that δn−1 has no Darboux polynomial. Set
t = xn , and

R = R[x1 , . . . , xn−1 ], d = δn−1 ,

R[t] = R[x1 , . . . , xn−1 , t], D = δn .

Then δn (t) = b := xn−2 xn−1

s−1
, and D( f ) = d( f ) for all f ∈ R. Direct calculations
show that b ∈ / d(R). Then the conclusion follows from Theorem 4.5 (ii).
In addition to the results on the nonexistence of Darboux polynomials, there are
also lots of results on the existence of invariant algebraic curves or surfaces of real
polynomial differential systems. However, a complete characterization has only been
achieved for some concrete differential systems, such as the Lorenz, Rikitake and
Rabinovich systems, and so on. In the next subsections we introduce some methods
for studying the existence and the nonexistence of invariant algebraic curves or
surfaces of some classical polynomial differential systems.
4.2 Existence of Darboux Polynomials 167

4.2.2 Liénard Differential Systems: Invariant Algebraic

Curves

For the Liénard differential system

ẋ = y, ẏ = − f m (x)y − gn (x), (4.8)

where

m 
n
f m (x) = ai x i and gn (x) = bi x i
i=0 i=0

are respectively real polynomials of degrees m and n (i.e. am bn = 0), Odani [350]
in 1995 proved the following result.

Theorem 4.6 If n ≤ m and f m gn ( f m /gn ) ≡ 0, the Liénard differential system (4.8)

has no algebraic solutions (or no invariant algebraic curves).

Proof If F(x, y) = 0, F ∈ C[x, y], is an invariant algebraic curve of system (4.8),

then there exists a k(x, y) ∈ C[x, y] such that

y∂x F − ( f m (x)y + gn (x))∂ y F = k(x, y)F(x, y). (4.9)

Set

k 

F(x, y) = Fi (x)y i , k(x, y) = k j (x)y j ,
i=0 j=0

where  ≤ min{m, n − 1}. Then equating the coefficients of y i in (4.9), we get the
equations that Fi (x) and k j (x) satisfy:

k j (x) = 0, 2 ≤ j ≤ ,
Fk (x) − k1 (x)Fk (x) = 0,

Fs−1 (x) − (k0 + s f m (x))Fs (x) − (s + 1)gm (x)Fs+1 (x) = 0, s = k, k − 1, . . .

Inductively solving these equations gives

Fs (x) = 0, s = 0, 1, . . . , k.

The details are left to the reader as an exercise. This means that system (4.8) has no
invariant algebraic curves under the given conditions. The theorem follows. 

In the proof of Theorem 4.6, we expanded the polynomials F(x, y) and k(x, y)
in y. This method of expanding a polynomial or an analytic function of multiple
variables in one of its variables is one of the valid tools in the study of Darboux
polynomials and of polynomial first integrals. Using the characteristic method for
168 4 Existence and Degree of Darboux Polynomials

solving linear partial differential equations one can simplify the proof of Theorem 4.6.
The characteristic method will be introduced in the next subsection, where we apply
it to characterize the Lorenz system which has an invariant algebraic surface.
As a corollary of Theorem 4.6, Odani [350] demonstrated that the Van der Pol
equation
ẍ + μ(x 2 − 1)ẋ + x = 0, μ = 0,

or its equivalent system

ẋ = y, ẏ = −μ(x 2 − 1)y − x, μ = 0,

has no algebraic limit cycles. This was the first result on the nonexistence of algebraic
limit cycles of planar polynomial differential systems. Recall that an algebraic limit
cycle of an autonomous differential system is a limit cycle which is contained in the
set of zeros of a polynomial. Of course the set of zeros of the polynomial is invariant
under the flow of the system.
The Van der Pol equation was introduced by Van der Pol [429] in 1926 to describe
the oscillation of constant amplitude of a triode vacuum tube. This equation has
numerous applications in physics and biology. It plays a key role in the qualitative
theory of planar polynomial differential systems. As is well known, this equation
has at most one limit cycle, and it has a limit cycle for suitable choices of μ, see e.g.
Ye [451] and Zhang et al. [480].
If the condition of Odani’s theorem does not hold, the Liénard differential system
may have algebraic limit cycles. Wilson [442] in 1964 posed an example

μ2 3 2
ẋ = y, ẏ = −μ(x 2 − 1)y − x (x − 4) − x,
16
which has the invariant algebraic curve
 2 
μ μ 2 2
y + x(x − 4)y + (x − 4)
2 2 2
x (x − 4) + 1 = 0,
2 16
μ
with cofactor k(x, y) = − x 2 . In fact, if 0 < |μ| < 2 the invariant algebraic curve
2
contains a limit cycle.
Related to the result of Odani [350] and Wilson’s example, Odani conjectured
that if m < n ≤ 2m, the Liénard differential system (4.8) has no limit cycles.
Using a method similar to that of Odani [350], Hayashi [201] in 1996 proved the
following result.

Theorem 4.7 For n = m + 1, the Liénard system has an invariant algebraic curve
of the form y = P(x) if and only if gn (x) = −( f m (x) + P  (x))P(x).
4.2 Existence of Darboux Polynomials 169

Żoła̧dek et al. [65, 486] pursued this thread, and obtained the following classifi-
cation of Liénard differential systems (4.8) which have an invariant algebraic curve
or an algebraic limit cycle.

Theorem 4.8 Denote by Am,n the set of Liénard polynomial differential systems of
type (4.8) with am bn = 0. Since system (4.8) is uniquely determined by the coefficients
of f m and gn , we have Am,n  Cm+n × (C∗ )2 , where C∗ = C \ {0}. For any given
(m, n), the following statements hold.
(a) If n > m, there exists a system in Am,n which has an invariant algebraic curve.
(b) If n > m, and (m, n) = (0, 1), the generic systems in Am,n have no algebraic
limit cycles.
(c) If m > 1, n > m + 1, and (m, n) = (2, 4), there exists a system in Am,n which
has an algebraic limit cycle.
(d) If either m = 0, or m = 1, or (m, n) = (2, 4), or n = m + 1, no system in Am,n
has algebraic limit cycles.

These results show that Odani’s conjecture is false. Except for the case (m, n) =
(1, 3) in (d), all results in Theorem 4.8 were stated by Żoła̧dek [486, Theorem 1] in
1998. The result in case (m, n) = (1, 3) was obtained by Chavarriga et al. in 2006.
When Żoła̧dek proved the existence of algebraic limit cycles in statement (c), in
fact he proved the existence of hyperelliptic limit cycles. A hyperelliptic limit cycle
is a limit cycle which is contained in an elliptic branch of a hyperelliptic curve

F(x, y) := (y + P(x))2 − Q(x) = 0, (4.10)

where P(x) and Q(x) are polynomials.

In 2008 Llibre and Zhang [301] proved that if (m, n) = (3, 5), the Liénard system
(4.8) has no hyperelliptic limit cycles. So Theorem 4.8 does not hold, at least in case
(c). We note that the proof of Theorem 4.8 (c) and (d) is too technical to be checked.
The proof of (b) is shorter (see [486, §5]), but it depends on the proof of (c) and (d).
For other values of (m, n) in Żoła̧dek’s theorem (c), there do exist Liénard systems
in Am,n which have hyperelliptic limit cycles, see Llibre and Zhang [301], Yu and
Zhang [457] and Liu et al. [245].
Next we present a proof of statement (a), which is due to Żoła̧dek.
Proof of Theorem 4.8. (a) Direct calculations show that a rational curve y = P(x),
P(x) ∈ C[x], is an invariant algebraic curve of the Liénard differential system if and
only if
gn (x) = −( f m (x) + P  (x))P(x). (4.11)

Case 1. m < n ≤ 2m + 1. For an arbitrary polynomial P(x) of degree n − m, we

get a Liénard differential system in Am,n with gn satisfying (4.11), which has the
invariant algebraic curve y = P(x).
Case 2. n > 2m + 1. If n = 2k − 1, any polynomial P(x) of degree k is suitable for
the theorem.
170 4 Existence and Degree of Darboux Polynomials

If n = 2k > 2m + 1, we cannot apply (4.11) to obtain invariant algebraic curves

of the Liénard differential systems. Consider the hyperelliptic function

F(x, y) := (y + Pm+1 (x))2 − Q(x) = 0, (4.12)

where Pm+1 , Q ∈ C[x], and the subscripts denote the degrees of the polynomials.
If F(x, y) is an invariant algebraic curve of the Liénard differential system, then we
have  
 1 Q 1  Q
f m = Pm+1 + Pm+1 , gn = − Q − Pm+1
2
. (4.13)
2 Q 2 Q

Set

 

Pm+1 = a (x − xi )m i , m i = 0, m i = m + 1, xi = x j , i = j
i=1 i=1

 

Q=a (x − xi )ni , n i = n + 1.
i=1 i=1

Since n > 2m + 1, we get

   
 1 Q Q
deg Pm+1 + Pm+1 = m, deg Q  − Pm+1
2
= n.
2 Q Q

This proves statement (a). 

The next result, due to Llibre and Zhang [301], exhibits concrete examples of
Liénard systems (4.8) with m > 1 and n ≥ 2m + 1, or m > 2 and n = 2m, which
have algebraic limit cycles.

Theorem 4.9 For the Liénard systems of type (m, n), the following statements hold.
(a) Assume that m > 1 and n > 2m + 1. If

1  
f m (x) = (x + 1)m−2 (3 + 2m + n)x 2 + (6 − 2m − n)x − 3 ,
2
1    
gn (x) = (n + 1)x 2 − (n − 2)x − 1 x 2 − x (x + 1)2m−3 + (x + 1)n−2 ,
2
then the Liénard system (4.8) has an algebraic limit cycle contained in the
hyperelliptic curve
 2
Fa (x, y) := y + x(x − 1)(x + 1)m−1 + x(x − 1)(x + 1)n−1 = 0.

(b) Suppose that m > 1 and n = 2m + 1. If

4.2 Existence of Darboux Polynomials 171

2m + 1 m−1
f m (x) = (m + 2)x m − x ,
  2
1  
g2m+1 (x) = x − 1 + (x − 1)x 2m−1 ,
2

then the Liénard system (4.8) has an algebraic limit cycle contained in the
hyperelliptic curve
 2
Fb (x, y) := y + (x − 1)x m + x(x − 1) = 0.

(c) Assume that m ≥ 3 and n = 2m. If

1
f m (x) = 4(x − 1)x m−2 ((m + 1)x − m + 1)
2  
+ ε (4m + 1)x m−1 −4(m − 1)x m−2 − (m + 10)x 2 + 2(m + 6)x − m − 2

− ε 2 ((m + 7)x − m − 2) ,
ε  
g2m (x) = x(x − 1) x m−2 − x + 1 − ε ×
2  
2(x − 1)((m + 1)x − m + 1)x m−2 + ε x m−1 + 2x m−2 − 4x 2
  
+ 4x − m(x − 1) 2x m−2 − x + 1 − ε2 ((m + 3)x − m) ,

where 0 < ε < (1/2)m−2 , then the Liénard system (4.8) has an algebraic limit
cycle contained in the hyperelliptic curve
   2
Fc (x, y) := y + x m−2 − ε (x − 1 + ε)(x − 1)x
 
− x m−2 − ε (x − 1 + ε)(x − 1)3 x m = 0.

The proof of Theorem 4.9 needs the following result which characterizes the
boundary of a maximum period annulus, see for instance Ye [451, Theorem 1.6].

Lemma 4.2 If a planar analytic vector field has a maximum period annulus, then
its outer and inner boundaries must contain singularities of the vector field.

Recall that a period annulus of a planar polynomial vector field is an annulus

filled up with periodic orbits of the vector field. A maximum period annulus is a
period annulus which cannot be contained inside any other period annulus.
Proof of Theorem 4.9. (a) We can check that Fa (x, y) = 0 passes through the points
(0, 0) and (1, 0), which are regular points of the vector field. Moreover, the branch
passing through these two points of Fa (x, y) = 0 is closed and contained in the strip
0 ≤ x ≤ 1. So this branch is a periodic orbit, denoted by C. Inside the region
 limited
√
by C there is a unique singularity (x0 , 0) with x0 = n − 2 + n + 8 /(2(n +
2

1)) =: h(n). In addition, we can show that gn (x0 ) > 0, and
172 4 Existence and Degree of Darboux Polynomials
 
f m (x0 ) = (x0 + 1)m−2 (m + 1)x02 − (m − 2)x0 − 1 > 0,

because h  (n) > 0. This verifies that (x0 , 0) is a stable focus or node, and conse-
quently C is a limit cycle. Otherwise, there will be a maximum period annulus of
the vector field containing C in its interior, but this contradicts Lemma 4.2 and the
fact that the inner boundary of the annulus is a stable focus or node.
(b) Similar to the proof of statement (a) we can prove that the hyperelliptic curve
Fb (x, y) = 0 contains a limit cycle passing through the regular points (0, 0) and
(1, 0).
(c) As in the proof of statement (a), we obtain that Fc (x, y) = 0 contains a periodic
orbit, denoted by C, which passes through the regular points (ε(1/(m−2)) , 0) and
(1 − ε, 0), and is located in the strip ε(1/(m−2)) ≤ x ≤ 1 − ε. Furthermore, we can
check that the system has a unique singularity, say (x0 , 0), in the interior of the region

limited by C, with x0 a simple zero of g2m (x) satisfying g2m (x0 ) > 0 and f m (x0 ) = 0.
These facts verify that the periodic orbit C is an algebraic limit cycle of the system.
This completes the proof of Theorem 4.9. 

The next result shows that for all m > 4 and n = 2m − 1, or n = 2m − 2, there
exist Liénard systems (4.8) of type (m, n) which have [m/2] − 1 algebraic limit
cycles, see [301, Theorem 4].

Theorem 4.10 For any m > 4 and 1 , . . . , m+1 ∈ R satisfying i < i+1 for i =
1, . . . , m − 1 and m+1 ∈ / {1 , . . . , m }, set P(x) = (x − 1 ) . . . (x − m )
(x − m+1 ). The following statements hold.
(a) If Q(x) = (x − 1 ) . . . (x − m )(x − m+1 )m+2 , and m+1 = (1 + · · · + m )
/m, then the Liénard system (4.8) with f m and gn given in (4.13) having the
prescribed polynomials P and Q is of type (m, 2m − 1), and it has [m/2] − 1
algebraic limit cycles, which are contained in the hyperelliptic curve (4.12).
(b) If Q(x) = (x − 1 ) . . . (x − m−1 )(x − m )3 (x − m+1 )m , and

m−1
(m − 2)m+1 + m = i ,
  i=1
(4.14)
m−2 2
m+1 + (m − 2)m m+1 = i  j ,
2 1≤i< j≤m−1

then the Liénard system (4.8) with f m and gn given in (4.13) having the pre-
scribed polynomials P and Q can be of type (m, 2m − 2), and may have
[m/2] − 1 algebraic limit cycles, which are contained in the hyperelliptic curve
(4.12).

Proof Here we provide a sketch of the proof. For details, see [301, Theorem 4].
(a) Under the condition, we can check that the Liénard differential system is of type
(m, 2m − 1). Without loss of generality we can set 1 < m+1 < 2 . Some calcula-
tions show that gn (i ) = 0 for i = 1, . . . , m and Q(x) > 0 for x ∈ (m−2i , m−2i+1 ),
i = 1, . . . , [m/2] − 1. Moreover, the Liénard system has a singularity (si , 0) with
4.2 Existence of Darboux Polynomials 173

m−2i < si < m−2i+1 for each i ∈ {1, . . . , [m/2] − 1}. Similar arguments as in the
proof of Theorem 4.9 verify that the hyperelliptic curve (4.13) contains [m/2] −
1 periodic orbits Si , for i ∈ {1, . . . , [m/2] − 1}, which pass through (m−2i , 0)
and (m−2i+1 , 0) and are located in the strip m−2i < x < m−2i+1 for each i ∈
{1, . . . , [m/2] − 1}.
Let si ∈ [m−2i , m−2i+1 ] be the point where Q(x) takes its maximal value in the
interval. Then we can show that gn (si ) > 0 for i = 1, . . . , [m/2] − 1, and it follows
that (si , 0) is a focus or node. Hence Si is a limit cycle for each i ∈ {1, . . . , [m/2] −
1}. This proves statement (a).
(b) For any fixed 1 , . . . , m−1 satisfying i < i+1 , the Eq. (4.14) always have a
solution (m , m+1 ) satisfying m > m−1 and m+1 < 1 . We have

gn = (x − m )(x − m+1 )K (x)F(x)/2,

where

m−1
K (x) = (x − 1 ) . . . (x
− i ) . . . (x − m+1 )
i=1


m−1
+ (x − i ) (3(x − m+1 ) + m(x − m )) ,
i=1
m−1 

F(x) = (x − i ) − (x − m )(x − m+1 )m−2 = q x m−4 + G(x),
i=1

with     
m−2 m−2 3
q= m 2m+1 + m+1 − i  j k ,
2 3
1≤i< j<k≤m−1

and G(x) a polynomial of degree at most m − 5. Note that the Liénard differential
system is of type (m, n) with n ≤ 2m − 2. Clearly we can choose the values of
1 , . . . , m−1 such that q = 0. In this case the system is of type (m, 2m − 2).
We can check that gn (i ) = 0 for i = 1, . . . , m − 1, and Q(x) > 0 for x ∈
(m−2i , m−2i+1 ), i = 1, . . . , [m/2] − 1. Moreover, the Liénard differential system
has the singularities (si , 0) with si ∈ (m−2i , m−2i+1 ) satisfying K (si ) = 0 and
K  (si ) < 0 for i ∈ {1, . . . , [m/2] − 1}. This shows that the hyperelliptic curve (4.13)
contains [m/2] − 1 periodic orbits, denoted by Si for i = 1, . . . , [m/2] − 1, which
pass through (m−2i , 0) and (m−2i+1 , 0), and are located in the strip m−2i ≤ x ≤
m−2i+1 for each i ∈ {1, . . . , [m/2] − 1}. Furthermore,  we can prove  that for i =

m−1
1, . . . , [m/2] − 1, gn (si ) > 0 and f m (si ) = − (si −  j ) ((m − 1)m +
j=1
2rm+1 − (m + 1)si ) = 0 for some values of 1 , . . . , m satisfying (4.14). These facts
force that the singularity (si , 0) is either a focus or a node, and consequently Si is a
limit cycle of the Liénard differential system for each i ∈ {1, . . . , [m/2] − 1}. This
proves statement (b).
174 4 Existence and Degree of Darboux Polynomials

The proof of the theorem is complete. 

At the end of this subsection we summarize the results on the existence of hyper-
elliptic limit cycles due to Llibre and Zhang [301], Yu and Zhang [457] and Liu
et al. [245].

Theorem 4.11 On the hyperelliptic limit cycles of the Liénard system (4.8), the
following statements hold.
(a) For (m, n) = (3, 5), system (4.8) has no hyperelliptic limit cycles.
(b) For m ≥ 4 and m + 1 < n < 2m, there exist systems (4.8) of type (m, n) which
do have at least one hyperelliptic limit cycle.

Proof (a) In order that system (4.8) is of type (3, 5), by (4.10) and (4.13) we can
assume that

P(x) = a(x − 1 )(x − 2 )(x − 3 )(x − 4 ),

Q(x) = a(x − 1 )n 1 (x − 2 )n 2 (x − 3 )n 3 (x − 4 )n 4 ,

with n i ∈ Z+ and n 1 + n 2 + n 3 + n 4 = 8. For any possible choice of n i ’s one can

show that the hyperelliptic curve (4.10) cannot be a limit cycle of system (4.8) of
type (3, 5). The details can be found in the proof of Theorem 2 of Llibre and Zhang
[301].
(b) Case 1. For m = 4 and m + 1 < n < 2m there are only two cases (m, n) = (4, 6),
(4, 7).
Set P(x) = (x − 1 )x(x − 1)(x − 4 )(x − 5 ) and Q(x) = (x − 1 )7 x(x − 1)
(x − 4 ) with 1 < 0, 4 > 1 and 5 = (51 − 4 − 1)/2, then system (4.8) is of
type (4, 7) and has a limit cycle contained in the hyperelliptic curve (4.10).
Set P(x) = (x − 1 )(x − 2 )x(x − 1)(x − 5 ) and Q(x) = (x − 1 )5 (x − 2 )3
√ − 1) with 1 < 0, 2 > 1, 5 = (31 + 2 − 1)/2 and r1 = 1 − 2 −
x(x
2 32 (2 − 1)/3, then system (4.8) is of type (4, 6), and has a limit cycle contained
in the hyperelliptic curve (4.10) for suitable choices of 1 and 2
Case 2. For m = 5 and m + 1 < n < 2m there are only three cases (m, n) = (5, 7),
(5, 8) and (5, 9). The proof can be done using similar techniques as in (a) and (b)
with m = 4. The details are omitted, we refer to the proof of Theorem 1.1 of Yu and
Zhang [457].
Case 3. For m > 5,  the proof follows from Theorem 1.1 of Liu et al. [245].
4m + 1
Subcase 3.1. 1 + ≤ n < 2m − 2 with odd n.
3
2m + 1 − n i
Let k = ≥ 2, choose xi = for 1 ≤ i ≤ k, and y j ’s for j = 1, . . . ,
2 k
2k satisfying

y1 < x1 < y2 < y3 < x2 < y4 < · · · < y2k−1 < xk < y2k ,
(x − y1 )(x − y2 ) . . . (x − y2k ) = (x − x1 )2 (x − x2 )2 . . . (x − xk )2 − ε
4.2 Existence of Darboux Polynomials 175

3n − 4m − 1
with 0 < ε  1, and z i = y1 − i for i = 1, . . . ,  = > 0. Set
2


k 
 
2k
P(x) = (x − xi ) (x − z i ) (x − yi ),
i=1 i=1 i=1


k 
 
2k
Q(x) = (x − xi ) 4
(x − z i ) 2
(x − yi ).
i=1 i=1 i=1

Then system (4.8) is of type (m, n) and it has a limit cycle contained in the hyperel-
liptic curve (4.10).
4m + 2
Subcase 3.2. 1 + ≤ n < 2m − 2 with even n.
3
2m − n i
Let k = ≥ 2, choose xi = for 0 ≤ i ≤ k, and y j ’s for j = 0, 1, . . . , 2k
2 k
satisfying

x0 < y0 < y1 < x1 < y2 < y3 < x2 < y4 < · · · < y2k−1 < xk < y2k ,
(x − y0 )(x − y1 ) . . . (x − y2k ) = (x − x0 )(x − x1 )2 (x − x2 )2 . . . (x − xk )2 − ε,

3n − 4m − 2
with 0 < ε  1, and z i = x0 − i for i = 1, . . . ,  = > 0. Set
2


k 
 
2k
P(x) = (x − x0 ) (x − xi ) (x − z i ) (x − yi ),
i=1 i=1 i=0


k 
 
2k
Q(x) = (x − x0 )3 (x − xi )4 (x − z i )2 (x − yi ).
i=1 i=1 i=0

Then system (4.8) is of type (m, n) and it has a limit cycle contained in the hyperel-
liptic curve (4.10). 
4m + 1
Subcase 3.3. m + 1 < n ≤ with odd n.
3
2m + 1 − n 4m − 3n + 3
Let k = ≥ 2 and  = ≥ 1. We have s = k −  = n −
2 2
m − 1 ≥ 1. Define
G(x) = c0 (T2k (x) − 1),

where c0 = 1/(22k k) and T2k (cos θ ) = cos(2kθ ) is the 2k-th Chebyshev polynomial.
Then there exist numbers x1 < x2 < · · · < x2k with xi = xi (ε) continuously so that
(2k − i)π
lim xi (ε) = cos , and
ε→0 2k
176 4 Existence and Degree of Darboux Polynomials

G(x1 ) = G(x3 ) = · · · = G(x2−1 ) = −2c0 ,

G(x2+1 ) = G(x2+3 ) = · · · = G(x2k−1 ) = −2c0 − ε,
G(x2 ) = G(x4 ) = · · · = G(x2k−2 ) = 0,
 x
G(x) = (t − x1 )(t − x2 ) . . . (t − x2k−1 )dt.
1

Moreover, G(x) satisfies


 
s 
s
2k(G(x) + 2c0 ) = (x − x2i−1 ) 2
(x − yi ) (x − z i ),
i=1 i=1 i=1


k−1
2kG(x) = (x − y0 )(x − 1) (x − x2i )2 ,
i=1

where y0 < x1 < x2 < z 1 < x2+1 < y1 < x2+2 < z 2 < · · · < x2k−2 < z s < x2k−1
< ys < 1. Set


k−1 
 
s 
s
P(x) = (x − y0 )(x − 1) (x − x2i ) (x − x2i−1 ) (x − yi ) (x − z i ),
i=1 i=1 i=1 i=1


k−1 
s 
s
Q(x) = (x − y0 )3 (x − 1)3 (x − x2i )4 (x − yi ) (x − z i ).
i=1 i=1 i=1

Then system (4.8) is of type (m, n) and it has a limit cycle contained in the hyperel-
liptic curve (4.10). 
4m + 2
Subcase 3.4. m + 1 < n ≤ with even n.
3
2m − n 4m − 3n + 2
Let k = ≥ 2 and  = ≥ 0. We have s = k −  = n − m −
2 2
1 ≥ 1. Define
G(x) = c0 (T2k+1 (x) − 1),

where c0 = 1/(22k (2k + 1)) and T2k+1 (cos θ ) = cos((2k + 1)θ ) is the 2k + 1-th
Chebyshev polynomial. Then there exist numbers x1 < x2 < · · · < x2k−1 with xi =
(2k + 1 − i)π
xi (ε) continuously so that lim xi (ε) = cos , and
ε→0 2k + 1

G(x1 ) = G(x3 ) = · · · = G(x2k−1 ) = 0,

G(x2 ) = G(x4 ) = · · · = G(x2 ) = −2c0 ,
G(x2+2 ) = G(x2+4 ) = · · · = G(x2k ) = −2c0 − ε,
 x
G(x) = (t − x1 )(t − x2 ) . . . (t − x2k )dt.
1
4.2 Existence of Darboux Polynomials 177

Moreover, G(x) satisfies


 
s 
s
(2k + 1)(G(x) + 2c0 ) = (x − y0 ) (x − x2i )2 (x − yi ) (x − z i ),
i=1 i=1 i=1


k
(2k + 1)G(x) = (x − 1) (x − x2i−1 )2 ,
i=1

where y0 < x1 < x2+1 < z 1 < x2+2 < y1 < x2+3 < z 2 < · · · < x2k−1 < z s < x2k
< ys < 1. Set


k 
 
s 
s
P(x) = (x − 1) (x − x2i−1 ) (x − x2i ) (x − yi ) (x − z i ),
i=1 i=1 i=0 i=1


k 
s 
s
Q(x) = (x − 1)3 (x − x2i−1 )4 (x − yi ) (x − z i ).
i=1 i=1 i=1

Then system (4.8) is of type (m, n) and it has a limit cycle contained in the hyperel-
liptic curve (4.10).
This completes the proof of the theorem. 

4.2.3 Lorenz Systems: Invariant Algebraic Surfaces

The Lorenz system is a simplified mathematical model for atmospheric convection

[306] which was constructed by the American mathematician and meteorologist
Edward Lorenz (1917–2008) in 1963:

ẋ = s(y − x), ẏ = r x − y − x z, ż = −bz + x y.

When s = 10, r = 28, b = 8/3, this system has a strange attractor, which can be
seen via numerical simulations. This example is a milestone in chaos theory. From
the dynamical point of view, the Lorenz system has been intensively investigated.
However, its dynamics has not been completely characterized for all values of
the parameters, see e.g. Sparrow [406]. From the integrability point of view, the
Lorenz system has also been studied by many mathematicians and physicists, see e.g.
Giacomini and Neukirch [174], Giacomini et al. [175], Goriely [189], Gradshteyn
and Ryzhik [191], Gupta [193], Kús [222], Llibre and Zhang [296], Schwarz [387],
Segur [388], Steeb [407], Strelcyn and Wojciechowski [414], Tabor and Weiss [419]
and the references therein.
The study of the characterization of invariant algebraic surfaces of the Lorenz
system was initiated by Segur [388] in 1982, who found three irreducible invariant
178 4 Existence and Degree of Darboux Polynomials

algebraic surfaces using the Painlevé method. Kús [222] in 1983 found another three
irreducible invariant algebraic surfaces using the Carlemann embedding. Since then,
many mathematicians and physicists tried to develop different methods either to
find other irreducible invariant algebraic surfaces or to prove that there are none
besides the known six. They did not succeed until 2000, when Llibre and Zhang
[298] completed the characterization of Lorenz systems which have an invariant
algebraic surface by using weight homogeneous polynomials and the method of
characteristic curves for solving linear partial differential equations. Here the weight
homogeneous polynomial is the quasi-homogeneous polynomial defined previously.
In this direction, Swinnerton-Dyer [417] presented a different approach to complete
the characterization of Lorenz systems which have an invariant algebraic surface.
The method introduced by Llibre and Zhang [296] solved not only the problem
of the characterization of invariant algebraic curves of the Lorenz system but also
some similar problems on many other differential systems, which will be mentioned
later on. Since this method is so useful, as an illustration we introduce and apply it
to study the invariant algebraic surfaces of the generalized Lorenz system.
For three-dimensional differential systems exhibiting chaotic phenomena, Chen
and Ueta [81] in 1999 obtained a three-dimensional chaotic system dual to the Lorenz
system, called the Chen system:

ẋ = a(y − x), ẏ = (c − a)x − x z + cy, ż = x y − bz,

which has a chaotic attractor when a = 35, b = 3, c = 28. Lü and Chen [311] in
2002 reported another three-dimensional chaotic system, called the Lü system:

ẋ = a(y − x), ẏ = −x z + cy, ż = x y − bz.

In the sense of Vanecek and Celikovsksý [430], the Lü system is a bridge between
the Lorenz and the Chen systems. Lü and Zhang [313] in 2007, and Lü [312] in 2009
characterized respectively the Chen and the Lü systems which have an invariant
algebraic surface using the method introduced by Llibre and Zhang in [296].
Unifying the Lorenz, Chen and Lü systems gives the following system

ẋ = a (y − x) = P(x, y, z),
ẏ = bx + cy − x z = Q(x, y, z), (4.15)
ż = dz + x y = R(x, y, z),

where x, y, z are real variables, and a, b, c, d are real parameters. This unified sys-
tem is called the generalized Lorenz system, which has been intensively studied
from the dynamical point of view. Next we will complete the characterization of
the generalized Lorenz system having an invariant algebraic surface using weight
homogeneous polynomials and the method of characteristic curves for solving linear
partial differential equations.
4.2 Existence of Darboux Polynomials 179

Let S be a set of Darboux polynomials of the generalized Lorenz system (4.15).

If T ⊂ S is a minimal subset such that each element of S is either a product
of finitely many elements of T or a finite sum of these products having the same
cofactor, we call the elements of T generators of S .
The following result, due to Wu and Zhang [445, Theorem 1.1], characterizes the
generalized Lorenz system which has an invariant algebraic surface.

Theorem 4.12 Assume that a = 0. The generators of Darboux polynomials of the

generalized Lorenz system (4.15) are formed by the following six, together with a
seventh generator which has a polynomial first integral:

Darboux polynomials cofactors parameters

x 2 − 2az −2a d = −2a
y2 + z2 2c d = c, b = 0
cy 2 + cz 2 + bx 2 2c a = −c, d = c
4 4 2 2 8 2 4 4 c
x + cx z − c y − c x y − cbx
4 2 2 c a = − ,d = 0
3 9 9 3 3 3
x 4 + 4cx 2 z − 4c2 y 2 + 8c2 x y + 4cbx 2 + 16(c + b)c2 z 4c a = −c, d = 4c
b = 2a + c,
x 4 − 4ax 2 z − 4a 2 y 2 + 4a(4a + 2c)x y − (4a + 2c)2 x 2 −4a
d = −6a − 2c
y + z − 2bz
2 2 0 c=d=0

Remark If a = 0, the generalized Lorenz system (4.15) is linear. Its dynamics is

simple, so we omit this trivial case in the theorem. The last case in Theorem 4.12
provides a polynomial first integral but not a proper Darboux polynomial, which
was obtained in [4] by Algaba et al. Recall that a proper Darboux polynomial of a
polynomial differential system is a Darboux polynomial which is not a first integral
of the system.

In the proof of Theorem 4.12, we will need weight homogeneous polynomials and
the method of characteristic curves for solving linear partial differential equations.
The weight homogeneous polynomial has the same definition as that of the quasi-
homogeneous polynomial, which was defined previously. Here we recall the method
of characteristic curves for solving linear partial differential equations in dimension
three.
A. The Method of Characteristic Curves for Solving Linear Partial Differential
Equations
The method of characteristic curves for solving linear partial differential equations
has already been introduced in Chap. 1. Here we state it in a slightly different way
which is more suitable for our application, see Bleecker and Csordas [37, Chap. 2].
Consider the linear partial differential equation

∂u ∂u ∂u
a(x, y, z) + b(x, y, z) + c(x, y, z) + d(x, y, z)u = e(x, y, z), (4.16)
∂x ∂y ∂z
180 4 Existence and Degree of Darboux Polynomials

where a(x, y, z), b(x, y, z), c(x, y, z), d(x, y, z), e(x, y, z) are C 1 functions, and
u = u(x, y, z) is an unknown function. A curve (x(t), y(t), z(t)), t ∈ I , is a char-
acteristic curve of the partial differential equation (4.16) if it is an integral curve of
the characteristic equation

dx dy dz
= a(x(t), y(t), z(t)), = b(x(t), y(t), z(t)), = c(x(t), y(t), z(t)).
dt dt dt

Observe that for any point (x0 , y0 , z 0 ) on the characteristic curve, the vector (a(x0 , y0 ,
z 0 ), b(x0 , y0 , z 0 ), c(x0 , y0 , z 0 )) is tangent to the curve at the point. In practice, if
c(x, y, z) = 0, we usually write the characteristic equations as

dx a(x, y, z) dy b(x, y, z)
= , = . (4.17)
dz c(x, y, z) dz c(x, y, z)

Assume that Eq. (4.17) has two functionally independent implicit solutions
A(x, y, z) = c1 and B(x, y, z) = c2 , where c1 , c2 are arbitrary constants. Consider
the coordinate change of variables

ξ = A(x, y, z), η = B(x, y, z), ζ = z, (4.18)

and its inverse transformation x = p(ξ, η, ζ ), y = q(ξ, η, ζ ) and z = r (ξ, η, ζ )

(sometimes, an explicit inverse transformation cannot be obtained, or is not well
defined). The linear partial differential equation (4.16) becomes an ordinary differ-
ential equation in ζ (for fixed ξ and η):

c(ξ, η, ζ )u ζ + d(ξ, η, ζ )u = e(ξ, η, ζ ), (4.19)

where c, d, e and u are c, d, e and u written in ξ , η and ζ .

If u = u(ξ, η, ζ ) is a general solution of (4.19), then

u(x, y, z) = u(A(x, y, z), B(x, y, z), z)

is a general solution of (4.16).

B. Proof of Theorem 4.12.
The idea follows Llibre and Zhang [298]. We first consider the case c = 0.
Taking the change of variables

x = ρ −1 ξ, y = ρ −2 η, z = ρ −2 η, t = ρτ, (4.20)

with ρ > 0 an arbitrary constant, the generalized Lorenz system (4.15) is transformed
to
4.2 Existence of Darboux Polynomials 181

ξ  = a(η − ρξ ),
η = −ξ ζ + cρη + bρ 2 ξ, (4.21)
ζ  = ξ η + dρζ,

where ‘’ denotes the derivative with respect to τ .

Assume that f (x, y, z) is a Darboux polynomial of system (4.15) with cofactor

k(x, y, z) = k0 + k1 x + k2 y + k3 z.

Let  be the weight degree of the highest order weight homogeneous polynomial of
f with the weight exponents (1, 2, 2) associated to x, y, z. Set

F(ξ, η, ζ ) = ρ  f (ρ −1 ξ, ρ −2 η, ρ −2 ζ ),
K (ξ, η, ζ ) = ρ 2 k(ρ −1 ξ, ρ −2 η, ρ −2 ζ ).

Then f = F |ρ=1 , and F is a Darboux polynomial of system (4.21) with cofactor

ρ −1 K .
Expanding F as a sum of weight homogeneous polynomials with respect to the
weight exponent (1, 2, 2)

F = F0 + ρ F1 + ρ 2 F2 + · · · + ρ m Fm ,

where Fi is a weight homogeneous polynomial in ξ, η, ζ of weight degree  − i,

i = 0, 1, . . . , m,  ≥ m. Since F is a Darboux polynomial of system (4.21) with
cofactor ρ −1 K , we have


m
∂ Fi m
∂ Fi m
∂ Fi
a(y − xρ) ρi + (cyρ + bxρ 2 − x z) ρi + (dzρ + x y) ρi
i=0
∂x i=0
∂y i=0
∂z


m
= (k1 x + k2 yρ −1 + k3 zρ −1 + k0 ρ) ρ i Fi ,
i=0

where we still use x, y, z instead of ξ, η, ζ .

In the above equations, comparing the coefficients of ρ −1 we get k2 = k3 = 0.
Furthermore, equating the coefficients of ρ i , i = 0, 1, . . . , m + 2, gives

L [F0 ] = k1 x F0 ,
∂ F0 ∂ F0 ∂ F0
L [F1 ] = k1 x F1 + k0 F0 + ax − cy − dz , (4.22)
∂x ∂y ∂z
∂ F j−1 ∂ F j−1 ∂ F j−1 ∂ F j−2
L [F j ] = k1 x F j + k0 F j−1 + ax − cy − dz − bx ,
∂x ∂y ∂z ∂y
j = 2, 3, . . . , m + 2,
182 4 Existence and Degree of Darboux Polynomials

where F j = 0 if j > m, and L is the linear partial differential operator

∂ ∂ ∂
L = ay − xz + xy .
∂x ∂y ∂z

The characteristic equations of the homogeneous partial differential equations

associated to (4.22) are
dx ay dy xz
= , =− ,
dz xy dz xy

which have the general solutions

x 2 − 2az = c1 , y 2 + z 2 = c2 ,

where c1 , c2 are constants of integration. We take the change of variables

u = x 2 − 2az, v = y 2 + z 2 , w = z, (4.23)

which has the inverse transformation

√ !
x = ± u + 2aw, y = ± v − w2 , z = w. (4.24)

To simplify the notation, without loss of generality we consider only the plus case.
By the transformations (4.23) and (4.24), the first partial differential equation in
(4.22) becomes the ordinary one (for fixed u and v):

! d F0
v − w2 = k1 F 0 , (4.25)
dw

where F 0 is F0 written as a function of u, v and w. Equation (4.25) has the general

solution  
z
F 0 = C 0 (u, v) exp k1 arcsin ! ,
y2 + z2

where C 0 is an arbitrary smooth function of u and v. Hence we have

 
z
F0 (x, y, z) = F 0 (u, v, w) = C 0 (x − 2az, y + z ) exp k1 arcsin !
2 2 2
.
y2 + z2

Since F0 is a weight homogeneous polynomial, it forces k1 = 0, and C 0 is a polyno-

mial in u and v.
The above proof tells us the following.
4.2 Existence of Darboux Polynomials 183

• Any cofactor associated to a Darboux polynomial of the generalized Lorenz system

(4.15) is a constant.
Substituting the expressions of u and v into C 0 gives either


n
F0 = ai (x 2 − 2az)2i−1 (y 2 + z 2 )n−i , (4.26)
i=1

with weight degree 4n − 2, or


n
F0 = ai (x 2 − 2az)2i (y 2 + z 2 )n−i , (4.27)
i=0

with weight degree 4n.

B1 F0 Has the Form (4.26)
Substituting (4.26) into the second equation of (4.22) gives

∂F ∂F ∂F
L [F1 ] = ax − cy − dz + k0 F0
∂x ∂y ∂z

n
= (a(4i − 2) − 2c(n − i) + k0 ) ai (x 2 − 2az)2i−1 (y 2 + z 2 )n−i
i=1

n
+ 2a(2a + d)(2i − 1)ai (x 2 − 2az)2i−2 (y 2 + z 2 )n−i z
i=1

n
+ 2(c − d)(n − i)ai (x 2 − 2az)2i−1 (y 2 + z 2 )n−i−1 z 2 .
i=1

Similar to solving F 0 , together with the transformations (4.23) and (4.24), from this
last partial differential equation we get the ordinary one (for fixed u and v):

√ ! d F1 n
u + 2az v − w2 = (a(4i − 2) − 2c(n − i) + k0 ) ai u 2i−1 vn−i
dw i=1
n
+ 2a(2a + d)(2i − 1)ai u 2i−2 vn−i w
i=1

n
+ 2(c − d)(n − i)ai u 2i−1 vn−i−1 w2 .
i=1

This ordinary differential equation has a general solution

184 4 Existence and Degree of Darboux Polynomials


n 
dw
F 1 (u, v, w) = (a(4i − 2) − 2c(n − i) + k0 ) ai u 2i−1 vn−i √ !
i=1 u + 2aw v − w2

n 
wdw
+ 2a(2a + d)(2i − 1)ai u 2i−2 vn−i √ !
i=1 u + 2aw v − w2

n 
w2 dw
+ 2(c − d)(n − i)ai u 2i−1 vn−i−1 √ ! + C 1 (u, v),
i=1 u + 2aw v − w2

where we have used the formulas in [298, Appendix (A1 )], or [37], and C 1 (u, v) is
an arbitrary smooth function of u and v.
Applying [298, Appendix (A2 ) and (A3 )] to the integrals in F 1 , and the facts that
C1 (x, y, z) = C 1 (x 2 − 2az, y 2 + z 2 ) and F1 is a weight homogeneous polynomial
of weight degree 4n − 3, we must have C 1 (u, v) = 0, and

(a(4i − 2) − 2c(n − i) + k0 ) ai
+ a(4a 2 + 2ad)(2i + 1)ai+1 = 0, i = 0, 1, . . . , n, (4.28)
(3a(4i − 2) + 3k0 + (n − i)(−2d − 4c)) ai = 0, i = 1, 2, . . . , n,

with a0 = an+1 = 0. Some calculations show that the condition (4.28) is equivalent
to one of the following three:
(i) d = c, a = −c/2, k0 = c(2n − 1), and there exists an i 0 ∈ {1, . . . , n − 1} such
that ai0 = 0;
(ii) d = −2a, k0 = −(4n − 2)a, F0 = an (x 2 − 2az)2n−1 ;
(iii) d = −2a, F0 ≡ 0.
We now discuss the existence of Darboux polynomials of the generalized Lorenz
system in each of the three cases.
Case (i) In this case, we have F1 = 0, and Eq. (4.22) with j = 2 becomes


n
L [F2 ] = − 2(n − i)bai x y(x 2 − 2az)2i−1 (y 2 + z 2 )n−i−1 .
i=1

Its associated ordinary differential equation is

d F2 
n−1
=− 2(n − i)bai u 2i−1 vn−i−1 ,
dw i=1

which has a general solution


n−1
F 2 (u, v, w) = − 2(n − i)bai u 2i−1 vn−i−1 w + C 2 (u, v),
i=1
4.2 Existence of Darboux Polynomials 185

where C 2 is a constant of integration, and it is an arbitrary smooth function of u and

v. The fact that F2 (x, y, z) = F 2 (u, v, w) is a weight homogeneous polynomial of
weight degree 4n − 4 implies


n−1
F2 (x, y, z) = − 2(n − i)bai (x 2 − 2az)2i−1 (y 2 + z 2 )n−i−1 z
i=1


n−1
+ ai(2) (x 2 − 2az)2i (y 2 + z 2 )n−i−1 ,
i=1

where the ai(2) ’s are arbitrary constants, i = 0, 1, . . . , n − 1, which will be deter-

mined later on.
Similar to the above calculations, we get from Eq. (4.22) with j = 3, together
with the expressions of F1 and F2 , that

F3 (x, y, z) = F 3 (u, v, w)
n−1 
dw
=− ai(2) u 2i vn−i−1 √ √
i=0
u + 2aw v − w2

n−1 
wdw
+ 2ai b(n − i)u 2i−1 vn−i−1 √ √ + C 3 (u, v),
i=1
u + 2aw v − w2

where C 3 is an arbitrary smooth function. The fact that F3 is a weight homogeneous

polynomial of weight degree 4n − 5 forces

C 3 = 0; ai(2) = 0, i = 0, 1, . . . , n − 1; b(n − i)ai = 0, i = 1, 2, . . . , n − 1.

The last condition shows that b = 0 (because there exists an i 0 ∈ {1, . . . , n − 1}

such that ai0 = 0). Hence, we have F2 = F3 = 0. Moreover, by similar recursive
calculations as above, we have Fi = 0, i = 4, . . . , m. These proofs show that the
generalized Lorenz system in case (i) has only the Darboux polynomials of the form


n
ai (x 2 − 2az)2i−1 (y 2 + z 2 )n−i .
i=1

Obviously, it consists of x 2 − 2az and y 2 + z 2 . We can check that x 2 − 2az and

y 2 + z 2 are two Darboux polynomials of the generalized Lorenz system (4.15).
Case (ii) In this case, one has F0 = (x 2 − 2az)2n−1 and F1 = 0, where we choose
an = 1 without loss of generality. Similar to the treatments in Case (i), one gets that
Eq. (4.22) with j = 2 and j = 3 has respectively the solutions
186 4 Existence and Degree of Darboux Polynomials


n−1
(2)
F2 (x, y, z) = ai (x 2 − 2az)2i (y 2 + z 2 )n−i−1 ,
i=0
F3 (x, y, z) =F 3 (u, v, w)

n−1
1√ !
(2)
= (4ai − 2(n − i − 1)c + k0 )ai u 2i vn−i−1 u + 2aw v − w2
av
i=0

n−1 
(2) u wdw
+ (4ai − 2(n − i − 1)c + k0 )ai u 2i vn−i−1 √ √
i=0
av u + 2aw v − w2

n−1 
3 w2 dw
+ (4ai − 2(n − i − 1)c + k0 )ai(2) u 2i vn−i−1 √ √
i=0
v u + 2aw v − w2

n−1 
(2) w2 dw
+ 2(n − i − 1)(c − d)ai u 2i vn−i−2 √ √ + C 3 (u, v),
i=0 u + 2aw v − w2

where the ai(2) ’s are real numbers to be determined, i = 0, 1, . . . , n − 1, and C 3 is

an arbitrary smooth function.
By the fact that F3 is a weight homogeneous polynomial of weight degree 4n − 5,
one has C 3 = 0, and

(4ai − 2(n − i − 1)c − (4n − 2)a) ai(2) = 0,

i = 0, 1, 2, . . . , n − 1.
(n − i − 1)(c + 2a)ai(2) = 0,

These conditions imply that F2 (x, y, z) = F3 (x, y, z) = 0. Furthermore, inductive

calculations show that Fi = 0, i = 4, . . . , m. Hence, the generalized Lorenz system
has only the Darboux polynomials of the form (x 2 − 2az)2n−1 with cofactor −2(2n −
1)a, whose generator is x 2 − 2az.
Case (iii) In this case, we have F0 (x, y, z) = 0. Consequently, the generalized
Lorenz system has no Darboux polynomial.
B2 F0 Has the Form (4.27)
Now the second equation of (4.22) with F0 given in (4.27) is


n
L [F1 ] = (k0 + 4ai − 2c(n − i)) ai (x 2 − 2az)2i (y 2 + z 2 )n−i
i=0

n
+ 4ia(2a + d)ai (x 2 − 2az)2i−1 (y 2 + z 2 )n−i
i=0

n
+ 2(c − d)(n − i)ai (x 2 − 2az)2i (y 2 + z 2 )n−i−1 .
i=0

Some calculations show that

4.2 Existence of Darboux Polynomials 187

n
1 √ !
F1 = (k0 + 4ai − 2c(n − i)) ai u 2i vn−i u + 2az v − w2
i=0
a
n−1 
 
1
+ (k0 + 4ai − 2c(n − i)) ai + 4a(2a + d)(i + 1)ai+1
i=0
a

wdw
× u 2i−1 vn−i √ √
u + 2az v − w2

n−1 
w2 dw
+ (3k0 + 12ai + 2(−d − 2c)(n − i)) ai u v 2i n−i−1
√ √
i=0
u + 2az v − w2
1
+ (k0 + 4an)an u 2n+1 v−1 + C 1 (u, v),
a

with C 1 an arbitrary smooth function.

Since F1 (x, y, z) = F 1 (u, v, w) is a weight homogeneous polynomial of weight
degree 4n − 1, we must have C 1 = 0, and

(k0 + 4an)an = 0,
4a (2a + d)(i + 1)ai+1 + (k0 + 4ai − 2c(n − i)) ai = 0,
2
(4.29)
(2(−d − 2c)(n − i) + 12ai + 3k0 )ai = 0,

for i = 0, 1, . . . , n − 1. One can check that the Eq. (4.29) are equivalent to one of
the four conditions:
(i) d = −2a, k0 = −4an, an = 0, ai = 0, i = 0, 1 . . . , n − 1,
(ii) d = c, a = −c/2, k0 = 2nc, and ai = 0 for some i ∈ {0, 1, . . . , n − 1},
(iii) d = c, k0 = 2cn, a = −c/2, a0 = 0, and ai = 0, i = 1, .. . ,n,
n
(iv) d = −6a − 2c, k0 = −4an, a = −c/2, an−i = (−4a 2 )i an , an = 0.
i
Next we study the existence of Darboux polynomials of the generalized Lorenz
system in each of the four cases.
Case (i) In this case we have F0 = an (x 2 − 2az)2n and F1 = 0. Similar to the com-
putations showing that F0 has the form (4.26), we get Fi = 0, i = 2, . . . , m. Hence,
the generalized Lorenz system (4.15) has only the Darboux polynomials of the form
an (x 2 − 2az)2n with cofactors −4an and arbitrary constants an = 0. These Darboux
polynomials have the same generator x 2 − 2az with cofactor −2a.
Case (ii) In this case one has F1 = 0. By induction we can prove that F j = 0 for all
j > 1. So any Darboux polynomial is of the form


n
F = F0 = ci (x 2 + cz)2i (y 2 + z 2 )n−i ,
i=0
188 4 Existence and Degree of Darboux Polynomials

with cofactor 2cn, where the ci ’s are arbitrary constants and are not all zero. Their
generators are {x 2 + cz, y 2 + z 2 }.
Case (iii) In this case one has F0 = a0 (y 2 + z 2 )n and F1 = 0. Some calculations
show that Eq. (4.22) with j = 2 has the solution


n
F2 = −2bna0 (y 2 + z 2 )n−1 z + ai(2) (x 2 − 2az)2i−1 (y 2 + z 2 )n−i .
i=1

Now from Eq. (4.22) with j = 3 we get


wdw
F 3 = −2bna0 cv n−1
√ √
u + 2aw v − w2
n 
dw
+ [2i(2a + c) − 2a]ai(2) u 2i−1 vn−i √ √
i=1
u + 2aw v − w2
n 
(2) 2i−2 n−i wdw
+ (4a + 2ac)(2i − 1)ai u
2
v √ √ + C 3 (u, v),
i=1
u + 2aw v − w2

with C 3 an arbitrary smooth function.

By the fact that F3 (x, y, z) = F 3 (u, v, w) is a weight homogeneous polynomial
of weight degree 4n − 3 in x, y, z, it follows that C 3 = 0 and

(2i(2a + c) − 2a) ai(2) = 0, i = 0, 1, . . . , n,

−2bnca0 + (4a 2 + 2ac)ai(2) = 0,
i = 2, . . . , n.
(4a 2 + 2ac)(2i − 1)ai(2) = 0,

Hence F3 (x, y, z) = 0. Some easy calculations show that these last conditions are
the same as either
(iii 1 ) b = 0, ai(2) = 0, i = 1, . . . , n; or
(iii 2 ) b = 0, a = −c, a1(2) = −bna0 /c, ai(2) = 0, i = 2, . . . , n.
Subcase (iii 1 ). In this subcase one has F2 = 0. By induction we can prove that

Fi = 0, i = 4, . . . , m.

Hence the generator of the Darboux polynomials is y 2 + z 2 with cofactor 2c.

Subcase (iii 2 ). In this subcase one has
 
n bx 2
F2 = a0 (y 2 + z 2 )n−1 .
1 c

Moreover, Eq. (4.22) with j = 4 has the solution

4.2 Existence of Darboux Polynomials 189

2b2 n(n − 1)a0 2  (4) n−1

F4 = − (x + cz)(y 2 + z 2 )n−2 z + ai (x 2 + 2cz)2i (y 2 + z 2 )n−i−1 ,
c
i=0

where the ai(4) ’s are constants to be determined.

Further calculations show that Eq. (4.22) with j = 5 has the solution
 
2b2 n(n − 1) 
n−1
(4) 2i−1 n−i−1 wdw
F5 = − a0 uv n−2
+ 4icai u v √ √
c i=0
u + 2aw v − w2

n−1 
dw
+ 2(i − 1)cai(4) u 2i vn−i−1 √ √ + C 5 (u, v),
i=0
u + 2aw v − w2

with C 5 a smooth function to be determined. For F5 (x, y, z) = F 5 (u, v, w) to be

a weight homogeneous polynomial of weight degree 4n − 5 in x, y, z, one has
C 5 (u, v) = 0, and

n−1
−2b2 n(n − 1)c−1 a0 uvn−2 + 4icai(4) u 2i−1 vn−i−1 = 0,
i=0 i = 0, 1, . . . , n − 1.
(i − 1)cai(4) = 0,

These conditions can be further simplified to

n(n − 1)b2
a1(4) = a0 , a (4)
j = 0, j = 0, 2, . . . , n − 1.
2c2
These show that
   2 2
n 2 n−2 bx
F4 = a0 (y + z )
2
, F5 = 0.
2 c

Moreover, by induction one can prove that

   2 i
n bx
F2i = a0 (y 2 + z 2 )n−i , F2i+1 = 0, where i ≥ 5, 2i ≤ m.
i c

Hence, under the prescribed conditions, any Darboux polynomial of the generalized
Lorenz system (4.15) is of the form
n  
  i  n
n  n−i bx 2 bx 2
f (x, y, z) = a0 y 2 + z 2 = a0 y 2 + z 2 + ,
i c c
i=0

bx 2
with cofactor 2nc, whose generator is y 2 + z 2 + c
with cofactor 2c.
Case (iv) In this case, by setting an = 1 one has
190 4 Existence and Degree of Darboux Polynomials
 n
F0 (x, y, z) = (x 2 − 2az)2 − 4a 2 (y 2 + z 2 ) ,

and
 
n  2 n−1
F1 (x, y, z) = (x − 2az)2 − 4a 2 (y 2 + z 2 ) 4a(4a + 2c)x y.
1

Now we get from the second equation of (4.22) that

   n−1
F2 = (4a − 12a 2 )(4a + 2c) + 8a 2 b n (x 2 − 2az)2 − 4a 2 (y 2 + z 2 ) z
 2 
2 n−2
+ 16a (4a + 2c) n(n − 1)(y + z ) (x − 2az) − 4a (y + z )
3 2 2 2 2 2 2
z
 n−2
− 8a 2 (4a + 2c)2 n(n − 1)x 2 (x 2 − 2az)2 − 4a 2 (y 2 + z 2 ) z2

n
+ ai(2) (x 2 − 2az)2i−1 (y 2 + z 2 )n−i ,
i=1

where the ai(2) ’s are constants to be determined.

From Eq. (4.22) with j = 3, and applying calculations similar to [298, Appendix
(A4 )–(A6 )] and [298, Appendix B], we get

F 3 = 8a(4a + 2c) (−2(4a + 2c)b(u + 2aw) + (6s + 2c)(4ar − 2a(4a + 2c))w)

 
n
× (u 2 − 4a 2 v)n−2
2
  !
n √
+ 64a 3 (4a + 2c)3 (u + 2aw)(v − w2 )(u 2 − 4a 2 v)n−3 u + 2aw v − w2
3
 
n−1
+ 16a(3a + c)(a + c) (b − (2a + c)) n (−4a 2 )n−i−1
i

wdw
× u 2i vn−i−1 √ !
u + 2aw v − w2
n      
(2) n−1 n−2
− 6a ai + 2(4a + 2c)n b + (4a + 2c)(n − 1)
i −1 i −1
i=1
" 
w2 dw
× (−4a 2 )n−i u 2i−1 vn−i−1 √ ! + C 3 (u, v),
u + 2aw v − w2

with C 3 a smooth function to be determined. Again the fact that F3 (x, y, z) =

F 3 (u, v, w) is a weight homogeneous polynomial of weight degree 4n − 3 in x, y, z
forces C 3 = 0, and

(3a + c)(a + c)(b −(2a+ c)) =

 0,  
(2) n−1 n−2
ai = −2(4a + 2c) b + (4a + 2c)(n − 1) , (4.30)
i −1 i −1
 n−i
× n −4a 2 , i = 1, . . . , n.
4.2 Existence of Darboux Polynomials 191

Hence we have
 
n  2 n−2
F2 = (x − 2az)2 − 4a 2 (y 2 + z 2 ) (4a(4a + 2c)x y)2
2
 
n  2 n−1
+ (x − 2az)2 − 4a 2 (y 2 + z 2 )
1
   
× −2(4a + 2c)bx 2 + (4a − 12a 2 )(4a + 2c) + 4a(6a + 2c)b z ,
 
n  2 n−3
F3 = (x − 2az)2 − 4a 2 (y 2 + z 2 ) (4a(4a + 2c)x y)3
3
 
n  2 n−2
+ (x − 2az)2 − 4a 2 (y 2 + z 2 ) 2 (4a(4a + 2c)x y)
2
   
× −2(4a + 2c)bx 2 + (4a − 12a 2 )(4a + 2c) + 4a(6a + 2c)b z .

Since the first equation of (4.30) is equivalent to a = −c/3, or a = −c or b =

2a + c, we discuss each of the three cases.
If a = −c/3, then d = 0 and k0 = 4nc/3. The generalized Lorenz system has the
Darboux polynomial

4 4 8 4
f 2 = x 4 + cx 2 z − c2 y 2 − c2 x y − cbx 2 ,
3 9 9 3
with cofactor 4c/3. Direct calculations show that F0 , F1 , F2 and F3 are respectively
the weight homogeneous polynomials of weight degrees 4n, 4n − 1, 4n − 2 and
4n − 3 with the weight exponent s = (1, 2, 2) of
 n
4 4 8 4
g = x + cx 2 z − c2 y 2 − c2 x y − cbx 2
4
.
3 9 9 3

Set f := F|ρ=1 . Note that if f − g ≡ 0, it is also a Darboux polynomial of the

generalized Lorenz system (4.15) with the cofactor 4nc/3. But f − f ∗ has its highest
order term with the weight degree less than 4n − 3, a contradiction. Hence f 2 (x, y, z)
is a generator of the Darboux polynomials under the given conditions.
If a = −c, then d = 4c and k0 = 4cn. A similar proof shows that the Darboux
polynomial is
 n
x 4 + 4cx 2 z − 4c2 y 2 + 8c2 x y + 4cbx 2 + 16(c + b)c2 z ,

with cofactor 4cn.

If b = 2a + c, since k0 = −4an, d = −6a − 2c and a = −c/2, the Darboux
polynomial is
 n
x 4 − 4ax 2 z − 4a 2 y 2 + 4a(4a + 2c)x y − (4a + 2c)2 x 2 ,
192 4 Existence and Degree of Darboux Polynomials

with cofactor 4an.

Combining the above cases we get a proof of the theorem for c = 0.
If c = 0, the calculation is very easy, see [4, 5].
This completes the proof of the theorem. 

The above methods have also been successively applied to solve similar problems
for Belousov–Zhabotinskii systems, Rössler systems, Rikitake systems, Rabinovich
systems, and so on, see for instance Llibre and Zhang [296, 297, 299], Xie and
Zhang [448], Zhang [460, 465], and Llibre and Valls [276]. The characterizations
of Darboux polynomials for the Chen system and the Lü system were completed by
Lü and Zhang [313] (see also Deng and Chen [124]), and Lü [312] and Llibre et al.
[250], using the same methods as those in [298].
The above mentioned systems are all three-dimensional. Their global dynamics
have been completely characterized provided that they have an invariant algebraic
surface, see for instance Cao and Zhang [57] and Llibre et al. [260] for the Lorenz
systems, Chen et al. [80] and Llibre et al. [259] for the Rabinovich systems, Llibre
et al. [258] for the Rikitake systems, Liu and Yang [247] for the Lü system, Llibre
et al. [261] for the Chen system, Algaba et al. [5] and Wu and Zhang [446] for
the generalized Lorenz system, and so on. Since the study of the global dynamics
of differential systems is not our objective in this book, we refer the reader to the
original papers mentioned above.
In addition, Valls [426] characterized invariant algebraic surfaces of the general-
ized Raychaudhuri equation using a different method.

4.3 Other Results on Darboux Polynomials

As mentioned in the last section, after the work of Llibre and Zhang [298] the study
of the characterization of Darboux polynomials of concrete models has made great
progress. In this subsection we describe some other methods and results in the search
for Darboux polynomials of polynomial differential systems.

A. Simplifying Cofactors of Darboux Polynomials

The results of this subsection are due to Ferragut and Gasull [147], who provided
methods to simplify the form of cofactors, and consequently simplify calculations
in seeking Darboux polynomials.
Consider the polynomial differential systems of degree m

ẋ = P(x) = (P1 (x), . . . , Pn (x)), x ∈ Rn , (4.31)

where P1 (x), . . . , Pn (x) ∈ R[x] is coprime, and the dot denotes the derivative with
respect to t. As before, we denote by X P the associated vector field of system (4.31).
The first tool of Ferragut and Gasull [147] is the use of transformations. To do so,
we need the next result. The proof is easy, see [147, Lemma 2.1].
4.3 Other Results on Darboux Polynomials 193

Proposition 4.4 Let x = ϕ(u) be a change of coordinates with inverse u = ϕ −1 (x).

Assume that f is a Darboux polynomial of X P with cofactor k. The following state-
ments hold.
(a) The new system u̇ = Q(u) := ∂x ϕ −1 (ϕ(u))P(ϕ(u)) has the new Darboux poly-
nomial f = f ◦ ϕ(u) with the new cofactor 
k = k ◦ ϕ(u).
(b) After the change of time dt/ds = ν(u) applied to system u̇ = Q(u), the new
system u  = ν(u)Q(u) has the Darboux polynomial  f with cofactor ν
k.

Tool 1. The monomial maps. Let A = (ai j ) be an n × n matrix with integer coeffi-
cients and det A = ±1. Take the change of coordinates
 
y = g A (x) := x A = x a1 , . . . , x an ,
a a
where a j = (a j1 , . . . , a jn ) and x a j = x1 j1 , . . . , xn jn for j = 1, . . . , n, which is called
a monomial map. It follows from [28] that g A is a birational map, and g −1 A = g A−1 .
Set A−1 = (bi j ). One has

x = g A−1 (y) = (y b1 , . . . , y bn ),

where b j = (b j1 , . . . , b jn ) is the jth row of A−1 , j = 1, . . . , n.

From Proposition 4.4, this last change of coordinates sends system (4.31) to the
system  −1 
 n y j Ps y A
ẏ j = a js , j = 1, . . . , n. (4.32)
s=1
y bs

In the case when system (4.32) is rational, we can take a time rescaling

dt
= y α , α = (α1 , . . . , αn ),
dτ
to send system (4.32) to a coprime polynomial differential system
 −1 

n y j Ps y A
y j = Q(y) := y α a js , j = 1, . . . , n, (4.33)
s=1
y bs

whose associated vector field is denoted by Y .

Let f (x) be a Darboux polynomial of system (4.31) with cofactor k(x). Then we
can choose y β with β ∈ Zn such that
 −1 
F(y) = y β f y A

is a Darboux polynomial of system (4.33) and F has no factor yi for any i ∈

{1, . . . , n}. From Proposition 3.1 we get that F has the associated cofactor
194 4 Existence and Degree of Darboux Polynomials


n
Q i (y)  −1 
K (y) = βi + yα k y A , (4.34)
i=1
yi

where Q i is the ith component of Q. Note that K (y) has only possibly a denominator
y r with r ∈ Zn+ . Since Y is a polynomial vector field, F is a polynomial, and F has
no factor yi , we get from
Y (F)(y) = K (y)F(y)

that the coefficients of K (y) with rational monomial must be zero. Hence, we can
simplify the calculation of cofactors of Darboux polynomials.
Next we illustrate an application of this tool on a monomial map.
Consider the Bogdanov–Takens system

ẋ1 = x2 , ẋ2 = a1 + a2 x2 + x12 + x1 x2 , (4.35)

with a1 , a2 ∈ R, see [147, Example 3.1]. Assume that f is a Darboux polynomial

of system (4.35) with cofactor k = k0 + k1 x1 + k2 x2 . Taking A to be a 2 × 2 matrix
satisfying det A = 1, we choose
   
a1 a11 a12
A := = .
a2 a21 a22

Then its inverse is    

a22 −a12 b1
A−1 = =: .
−a21 a11 b2

Under the change of variables (y1 , y2 ) = (x a1 , x a2 ) and the time rescaling dt/dτ =
y1σ1 y2σ2 , system (4.35) is transformed to a new polynomial differential system in y1
and y2 with the time in terms of τ . Correspondingly, the cofactor k is transformed to
a rational function k ∗ in y1 and y2 . In order for k ∗ to be a polynomial, we get k2 = 0.
This reduces further calculations in the search for Darboux polynomials. For more
details, see [147, Example 3.1].
Tool 2. Reduction by symmetry. In the case when the polynomial differential sys-
tems have symmetry, one can obtain new Darboux polynomials from the known ones.
The next result, taken from [147, Lemma 2.2], provides one such construction.
Proposition 4.5 Assume that system (4.31) admits a symmetry in the sense that there
exists a linear invertible map ρ such that

ρ ◦ P(x) = c P ◦ ρ(x), c ∈ R a nonzero constant.

If f (x) is a Darboux polynomial of system (4.31) with cofactor k(x), then f ◦ ρ(x)
is also a Darboux polynomial of system (4.31) with cofactor ck ◦ ρ(x).
Recall that a product of Darboux polynomials of a given system is also a Darboux
polynomial with a cofactor which is the sum of the cofactors of the provided poly-
4.3 Other Results on Darboux Polynomials 195

nomials. So by Proposition 4.5 one gets a Darboux polynomial f (x) f ◦ ρ(x) with
cofactor k(x) + c f ◦ ρ(x). This observation is easy, but it simplifies the search for
Darboux polynomials in the symmetric case.
For example, the cubic differential systems

ẋ = a1 y + a2 y 3 , ẏ = a3 x + a4 x 2 + a5 y 2 + a6 x y 2 ,

which includes the Dolov system [130], has the symmetry ρ(x, y) = (x, −y) with
c = −1. If f is a Darboux polynomial of this last system with cofactor k(x, y), then
f (x, y) f (x, −y) is a Darboux polynomial with the cofactor k(x, y) − k(x, −y) =
k01 y + k11 x y. This greatly simplifies the calculations of the Darboux polynomials
together with their cofactors.

B. Other Results on Darboux Polynomials.

This subsection collects some other results on Darboux polynomials, which were not
stated in the previous sections.
In 2010 Ripoll and Sebag [372] studied the set of Darboux polynomials for a
special class of differential vector fields. Let C{x} and C{x, y} be the differential
ring of germs of univariate and bivariate holomorphic functions respectively, and let
C be the field of quotients of the ring C{x}. Consider the vector field

∂ ∂ ∂
Z =R + Ry  − V ,
∂x ∂y ∂y

where R ∈ C[y] and V ∈ C[y, y  ]. For such a vector field, by [372] a Darboux poly-
nomial is an element f in C{x, y}[y  ] such that Z ( f ) = k f for some k ∈ C[y, y  ].
For the vector field Z , Ripoll and Sebag [372] obtained the following result.

Theorem 4.13 The set of Darboux polynomials having a bounded degree and a
given cofactor in C[y, y  ] of the vector field Z is a finite-dimensional C-vector
space.

Note that in [372] the authors also provided an explicit upper bound on the dimen-
sion of the vector space formed by the Darboux polynomials.
Chapter 5
Algebraic, Analytic and Meromorphic
Integrability

This chapter studies the polynomial and analytic integrability of some concrete phys-
ical models, the polynomial and rational integrability of Hamiltonian systems, and
the meromorphic integrability of differential systems near a given orbit via the dif-
ferential Galois group. Finally, we present an algorithm to compute the rational first
integrals and the Darboux polynomials of polynomial differential systems.

5.1 Algebraic First Integrals

In this section we consider the relation of Darboux polynomials with algebraic

first integrals, including polynomial and rational first integrals. By the definition of
Darboux polynomials, a polynomial first integral is a Darboux polynomial with van-
ishing cofactor. The next result can be checked by easy calculations.

Proposition 5.1 The polynomial differential system (3.1) has a rational first integral
if and only if it has two relatively prime Darboux polynomials with the same cofactor.

5.1.1 Algebraic and Rational Integrability: Their

Equivalence

An algebraic function H (x) is a solution of the following algebraic function in C

f 0 + f 1 C + f 2 C 2 + . . . + f k−1 C k−1 + C k = 0, (5.1)

where k ∈ N, and f i (x) ∈ C(x) are not all constants. Here C(x) is the field of rational
functions in x with their coefficients in C. In the following each algebraic function
© Springer Nature Singapore Pte Ltd. 2017 197
X. Zhang, Integrability of Dynamical Systems: Algebra and Analysis,
Developments in Mathematics 47, DOI 10.1007/978-981-10-4226-3_5
198 5 Algebraic, Analytic and Meromorphic Integrability

is a root of an algebraic equation of the form (5.1) with the lowest degree. In this
sense the polynomial in the left-hand side of the algebraic equation (5.1) is called
the minimal monomic polynomial of the algebraic function H , and Eq. (5.1) is called
the minimal algebraic equation of H (x).
Obviously, polynomial and rational functions are all algebraic functions. If a
function H (x) is not algebraic, it is called a transcendental function. If a first integral
is an algebraic function, it is called an algebraic first integral.
An n-dimensional autonomous differential system is called algebraic integrable
(rational integrable, or polynomial integrable) if it has n − 1 functionally indepen-
dent algebraic first integrals (rational first integrals, or polynomial first integrals).
The next result, due to Bruns [47], shows that the study of algebraic first integrals
of a polynomial (or rational) differential system can be reduced to the study of rational
first integrals.

Proposition 5.2 For a polynomial differential system (3.1), or its associated vector
field X P , the following statements hold.
(a) If H (x) is an algebraic first integral of the vector field X P , then all coefficients
(if not constants) of its minimal algebraic equation are first integrals of X P .
Consequently, if X P has an algebraic first integral, it must have a rational first
integral.
(b) The vector field X P has  (1 ≤  < n) functionally independent algebraic first
integrals if and only if it has  functionally independent rational first integrals.

Proof Some parts of the proof follow from Forsyth [155, pp. 323–335].
(a) Assume that (5.1) is a minimal algebraic equation of H (x). Then

f 0 + f 1 H + f 2 H 2 + . . . + f k−1 H k−1 + H k = 0.

It follows that
 
0 = X P f 0 + f 1 H + f 2 H 2 + . . . + f k−1 H k−1 + H k
= X P ( f 0 ) + X P ( f 1 )H + X P ( f 2 )H 2 + . . . + X P ( f k−1 )H k−1 .

Since X P is a polynomial vector field, and the f i ’s are rational functions, X P ( f i ) is

also rational. By the minimal property we must have

X P ( f i ) ≡ 0, i = 0, 1, . . . , k − 1.

Hence if f i is not a constant, it should be a rational first integral of the vector field
XP .
In addition, the f i ’s are not all constants, otherwise H is a constant. This shows that
the vector field X P has at least one rational first integral. This proves the statement.
(b) Sufficiency is obvious. We prove the necessity by induction on the number  of
the functionally independent algebraic first integrals.
5.1 Algebraic First Integrals 199

If  = 1 the conclusion follows from (a). For  = 2, assume that H1 (x) and H2 (x)
are two functionally independent algebraic first integrals. Their minimal equations
are respectively

f 10 + f 11 H1 + f 12 H12 + . . . + f 1,k1 −1 H1k1 −1 + H1k1 = 0,

f 20 + f 21 H2 + f 22 H22 + . . . + f 2,k2 −1 H2k2 −1 + H2k2 = 0.

Clearly Hi is a function of f i j , j = 0, 1, . . . , ki − 1, i = 1, 2. If any two elements of

the set R2 := { f 1r , f 2s | r ∈ {0, 1, . . . , k1 − 1}, s ∈ {0, 1, . . . , k2 − 1}}, of rational
functions are functionally dependent, it follows from Theorem 1.1 that H1 and H2
can be locally expressed as functions in terms of a unique function which depends
on one of the coefficients of the minimal equations defining H1 and H2 . Hence H1
and H2 are functionally dependent, contradicting the assumption. So there exist two
elements in R2 , which are functionally independent rational first integrals of system
(3.1).
By induction, we assume that the conclusion holds for 2 <  < n − 1. That is,
for  functionally independent algebraic first integrals H1 (x), . . . , H (x), and the
 associated minimal algebraic equations, there are  coefficients of these minimal
equations which are  functionally independent rational first integrals of system (3.1).
For  + 1 functionally independent algebraic first integrals H1 (x), . . . , H (x),
H+1 , and their associated minimal algebraic equations

f i0 + f i1 Hi + f i2 Hi2 + . . . + f i,ki −1 Hiki −1 + Hiki = 0, i = 1, . . . , ,  + 1.

(5.2)
By the inductive assumption, the coefficients of the minimal equations of H1 , . . . , H
contain  functionally independent rational first integrals of the vector field X P ,
denoted by { f 1∗ , . . . , f ∗ }. Denote by R+1 the set of coefficients of the  + 1 minimal
equations.
On the contrary, we assume that R+1 does not contain  + 1 functionally inde-
pendent elements. Then each element of R+1 is locally a function of f 1∗ , . . . , f ∗ .
Consequently, H1 , . . . , H+1 are locally functions of f 1∗ , . . . , f ∗ . This implies that
H1 , . . . , H+1 are functionally dependent, a contradiction. Hence in the set of coef-
ficients of the minimal Eq. (5.2) there exist  + 1 elements which are functionally
independent first integrals of the vector field X P .
By induction we complete the proof of statement (b) and consequently of the
proposition. 

From the relation between Darboux polynomials and rational first integrals, and
the relation between algebraic and rational first integrals, we obtain the following
results on the generalized Lorenz system.
Corollary 5.1 For the generalized Lorenz system (4.15) with a = 0, the following
statements hold.
(a) System (4.15) has a polynomial first integral if and only if c = d = 0. Any poly-
nomial first integral is a polynomial function of y 2 + z 2 − 2bz.
200 5 Algebraic, Analytic and Meromorphic Integrability

(b) System (4.15) has a rational first integral if and only if b = 0 and c = d = −2a.
Any rational first integral is a rational function of (x 2 − 2az)2 /(y 2 + z 2 ).
(c) System (4.15) is not algebraically integrable.

Proof (a) The proof follows from Theorem 4.12 and the fact that system (4.15) has
a polynomial first integral if and only if it has a Darboux polynomial with the zero
cofactor.
(b) The proof follows from Theorem 4.12 and the fact that system (4.15) has a rational
first integral if and only if it has two relatively prime Darboux polynomials with the
same cofactor.
(c) This follows from (b) and Proposition 5.2. 


The above theory has lots of applications to the polynomial and rational integra-
bility of some physical models. For instance, the 2D Lotka–Volterra system [52], the
Rikitake system [296], the Lorenz system [298], the Rössler system [299], the Chen
system [313], and so on.
In the next three subsections we study the polynomial, rational and Darboux
integrability of three higher-dimensional fluid mechanic models.

5.1.2 Kirchoff Equations: Polynomial and Rational First

Integrals

In this subsection we will study the integrability of the Kirchoff equations in R6

1 1
ṗ1 = p2 π3 − p3 π2 ,
I3 I2
1 1
ṗ2 = p3 π1 − p1 π3 ,
I1 I3
1 1
ṗ3 = p1 π2 − p2 π1 ,
I I
2 1  
1 1 1 1
π̇1 = − π2 π3 + − p2 p3 , (5.3)
I3 I2 m3 m2
   
1 1 1 1
π̇2 = − π3 π1 + − p3 p1 ,
I1 I3 m1 m3
   
1 1 1 1
π̇3 = − π1 π2 + − p1 p2 ,
I2 I1 m2 m1

where the dot denotes the derivative with respect to the time t. The Kirchoff equations
were derived by Holmes et al. [208] in 1998, and describe the motion of a rigid body in
an ideal incompressible fluid that is at rest at infinity. The variables p = ( p1 , p2 , p3 )
and π = (π1 , π2 , π3 ) in system (5.3) are respectively the linear and angular momen-
tum vectors with respect to the body-fixed frame; I = diag (I1 , I2 , I3 ) is the sum of
5.1 Algebraic First Integrals 201

the body inertia matrix plus the added inertia matrix, and M = diag (m 1 , m 2 , m 3 ) is
the sum of the mass matrix for the body alone plus the added mass matrix. Due to
the physical constraints the six parameters I1 , I2 , I3 , m 1 , m 2 , m 3 of system (5.3) are
all positive. For more information on the Kirchoff equations, see e.g. [208].
System (5.3) is Hamiltonian with the Hamiltonian function consisting of the total
kinetic energies:
1  πi2 1  pi2
3 3
H̄ ( p, π ) = + . (5.4)
2 i=1 Ii 2 i=1 m i

This Hamiltonian vector field can be written as

M(∇ H̄ )τ ,

with the structure matrix

⎛ ⎞
0 0 0 0 − p3 p2
⎜ 0 0 0 p3 0 − p1 ⎟
⎜ ⎟
⎜ 0 0 0 − p2 p1 0⎟
M =⎜
⎜ 0 − p3 p2
⎟,
⎜ 0 −π3 π2 ⎟ ⎟
⎝ p3 0 − p1 π3 0 −π1 ⎠
− p2 p1 0 −π2 π1 0

where ∇ H̄ is the gradient of H̄ , i.e.

 
∇ H̄ = ∂ p1 H̄ ( p, π ), ∂ p2 H̄ ( p, π ), ∂ p3 H̄ ( p, π ), ∂π1 H̄ ( p, π ), ∂π2 H̄ ( p, π ), ∂π3 H̄ ( p, π ) .

Recall that an nth order matrix M(x) = (m i j (x)) in Rn is a structure matrix of

a Hamiltonian vector field M(x)(∇ H (x))τ if M(x) satisfies
• anti-symmetry: m i j = −m ji , i, j = 1, . . . , n;
• Jacobi identity:


n
 
m i ∂x m jk + m k ∂x m i j + m j ∂x m ki = 0, i, j, k = 1, . . . , n.
=1

A function F(x) is a Casimir function of the Hamiltonian vector field or of its

structure matrix if ∇ F(x)M(x) ≡ 0. Obviously each Casimir function is a first in-
tegral of the associated Hamiltonian vector field.
The Hamiltonian vector field M(x)(∇ H (x))τ in R2m with a nondegenerate struc-
ture matrix M(x) is Liouvillian integrable if it has m functionally independent first in-
tegrals which are in involution. Recall that the m first integrals H1 = H, H2 , . . . , Hm
of M(x)(∇ H (x))τ are in involution if ∇ Hi (x)M(x)(∇ H j (x))τ ≡ 0 for 1 ≤ i = j ≤
m.
We can check that system (5.4) has two Casimir functions
202 5 Algebraic, Analytic and Meromorphic Integrability

H̄2 ( p, π ) = p1 π1 + p2 π2 + p3 π3 , H̄3 ( p, π ) = p12 + p22 + p32 . (5.5)

So system (5.3) has the three functionally independent first integrals H̄ , H̄1 , H̄2 , and
consequently it is Liouvillian integrable.
Moreover, by Bogoyavlenskii [39] and Holmes et al. [208] if

m 1 I1 (m 3 − m 2 ) + m 2 I2 (m 1 − m 3 ) + m 3 I3 (m 2 − m 1 ) = 0, (5.6)

system (5.3) has the fourth polynomial first integral

H̄4 ( p, π ) = (I1 − I3 )(I1 − I2 ) p12 + I1 I3 (I1 − I2 )π22 + I1 I2 (I1 − I3 )π32 .

This last first integral depends only on the dependent variables p1 , π2 , π3 . We de-
note it by Φ( p1 , π2 , π3 ). We can check that its permutation Φ( p3 , π1 , π2 ) and
Φ( p2 , π3 , π1 ) are also first integrals of system (5.3) under the condition (5.6). Here
the fourth first integral is a first integral of system (5.3) which is functionally inde-
pendent of the known three given in (5.4) and (5.5).
To unify the notation, as suggested in [283] we set

1 1
μi = , νi = , xi = pi , xi+3 = πi , i = 1, 2, 3.
Ii mi

Then the Kirchoff equations (5.3) can be written as

ẋ1 = μ3 x2 x6 − μ2 x3 x5 =: P1 (x),
ẋ2 = μ1 x3 x4 − μ3 x1 x6 =: P2 (x),
ẋ3 = μ2 x1 x5 − μ1 x2 x4 =: P3 (x),
ẋ4 = (μ3 − μ2 )x5 x6 + (ν3 − ν2 )x2 x3 =: P4 (x), (5.7)
ẋ5 = (μ1 − μ3 )x6 x4 + (ν1 − ν3 )x3 x1 =: P5 (x),
ẋ6 = (μ2 − μ1 )x4 x5 + (ν2 − ν1 )x1 x2 =: P6 (x).

Its associated vector field is

∂ ∂
X := P1 (x) + . . . + P6 (x) ,
∂ x1 ∂ x6

where x = (x1 , . . . , x6 ), which has the three functionally independent first integrals

1 1 1 1 1 1
H1 (x) = μ1 x42 + μ2 x52 + μ3 x62 + ν1 x12 + ν2 x22 + ν3 x32 ,
2 2 2 2 2 2
H2 (x) = x1 x4 + x2 x5 + x3 x6 , (5.8)
H3 (x) = x12 + x22 + x32 .
5.1 Algebraic First Integrals 203

Now the condition (5.6) is

μ2 μ3 (ν2 − ν3 ) + μ1 μ2 (ν1 − ν2 ) + μ1 μ3 (ν3 − ν1 ) = 0. (5.9)

Llibre and Valls [283, Theorems 1 and 3] in 2012 proved the following results.

Theorem 5.1 For the Kirchoff equations (5.7), the following statements hold.
(a) Under the condition (5.9), system (5.7) is completely integrable.
(b) System (5.7) either satisfies (5.9) or has at most four functionally independent
polynomial first integrals.

Statement (a) was proved by using the four known functionally independent first
integrals and Theorem 2.4. To prove statement (b) they used the resonant conditions
of system (5.7) at singularities, see Zhang [464]. This method was first suggested by
Llibre, Yu and Zhang [295] to study the Euler equations. We will not prove (b) here,
but we refer to the next section for a proof for the Euler equations.
Set
ρ(x1 , x2 , x3 , x4 , x5 , x6 ) = (−x1 , −x2 , x3 , −x4 , −x5 , x6 ),

and K ρ := K ◦ ρ for a given function K (x). If K ρ = −K , we say K is ρ -

antisymmetric. Llibre and Valls [283, Theorem 2] stated that
If F is a Darboux polynomial of the vector field X with cofactor K being ρ-
antisymmetric, then F F ρ is a polynomial first integral of X .
This result was improved by Zhang [472] in 2013, who obtained the following.

Theorem 5.2 If the vector field X associated to system (5.7) has a Darboux poly-
nomial, then it has a fourth polynomial first integral.

For example, the Kirchoff equations (5.7) satisfying

ν1 = ν2 = ν3 , μ1 − μ3 = μ3 − μ2 = σ = 0

has a Darboux polynomial P = x4 − x5 with cofactor K = −σ x6 , and so by

Theorem 5.2 it has a fourth polynomial first integral.
Theorem 5.2 has the next consequence, which can be proved from the relations
between the algebraic and the rational first integrals, and between the rational first
integrals and the Darboux polynomials.

Corollary 5.2 For the Kirchoff vector field X , the following statements hold.
(a) If X has a fourth rational first integral, it has a fourth polynomial first integral.
(b) If X has a fourth algebraic first integral, it has a fourth polynomial first integral.

Proof of Theorem 5.2. The main steps of the proof of the theorem follow from
those of Zhang [472, Theorem 1], which are composed of the following lemmas and
propositions.
204 5 Algebraic, Analytic and Meromorphic Integrability

Proposition 5.3 For homogeneous vector fields, the following statements hold.
(a) Any homogeneous part of a Darboux polynomial of a homogeneous polynomial
vector field is a Darboux polynomial of the vector field.
(b) The cofactor of a homogeneous Darboux polynomial of a homogeneous polyno-
mial vector field is a homogeneous polynomial.

The proof is an easy exercise. By this proposition we only consider the homoge-
neous Darboux polynomials of the Kirchoff equations.
Assume that F(x) is a homogeneous Darboux polynomial of the Kirchoff vector
field X with cofactor K (x) := k1 x1 + . . . + k6 x6 ≡ 0. Then

X (F)(x) = (k1 x1 + . . . + k6 x6 )F(x) (5.10)

and at least one of the following conditions

|k1 |2 + |k4 |2 = 0, |k2 |2 + |k5 |2 = 0, |k3 |2 + |k6 |2 = 0

holds. By symmetry of the vector field X with respect to x1 , x2 , x3 , and respectively

x4 , x5 , x6 , we assume without loss of generality that

|k1 |2 + |k4 |2 = 0. (5.11)

Take the change of coordinates

ρ14 (x1 , x2 , x3 , x4 , x5 , x6 ) = (−x1 , x2 , x3 , −x4 , x5 , x6 ),

ρ25 (x1 , x2 , x3 , x4 , x5 , x6 ) = ( x1 , −x2 , x3 , x4 , −x5 , x6 ), (5.12)
ρ36 (x1 , x2 , x3 , x4 , x5 , x6 ) = ( x1 , x2 , −x3 , x4 , x5 , −x6 ),

and set for any function V in R6

Vi j := V ◦ ρi j , i ∈ {1, 2, 3}, j ∈ {4, 5, 6}.

Lemma 5.1 If the vector field X has a Darboux polynomial F with cofactor K ,
then the polynomial Fi j for i j = 14, or 25, or 36 is also a Darboux polynomial of
X with cofactor −K i j .

Proof We prove the lemma only for the case i j = 14. The others other cases follow
by similar arguments. Set

y = (y1 , . . . , y6 ) = ρ14 (x1 , . . . , x6 ).

By the definition and some easy calculations we obtain F14 (x) = F(y), and
5.1 Algebraic First Integrals 205

P1 (y) = (P1 )14 (x) = P1 ◦ ρ14 (x) = P1 (x),

P2 (y) = (P2 )14 (x) = P2 ◦ ρ14 (x) = −P2 (x),
P3 (y) = (P3 )14 (x) = P3 ◦ ρ14 (x) = −P3 (x),
P4 (y) = (P4 )14 (x) = P4 ◦ ρ14 (x) = P4 (x),
P5 (y) = (P5 )14 (x) = P5 ◦ ρ14 (x) = −P5 (x),
P6 (y) = (P6 )14 (x) = P6 ◦ ρ14 (x) = −P6 (x).

It follows that
∂ F(y) ∂ F(y) ∂ F(y)
X (F14 ) = −P1 (x) + P2 (x) + P3 (x)
∂ y1 ∂ y2 ∂ y3
∂ F(y) ∂ F(y) ∂ F(y)
−P4 (x) + P5 (x) + P6 (x)
∂ y4 ∂ y5 ∂ y6
6
∂ F(y)
=− Pl (y) = −K (y)F(y) = −K 14 (x)F14 (x).
l=1
∂ yl

The lemma follows. 



As a consequence of Lemma 5.1 we have the next results.

Corollary 5.3 If the Kirchoff vector field X has a polynomial (or an analytic) first
integral V , then Vi j is also a first integral of X for i j = 14, or 25, or 36.

The next result provides a new construction of Darboux polynomials from a given
one.

Lemma 5.2 Assume that the Kirchoff vector field X has a proper homogeneous
Darboux polynomial F with cofactor K = k1 x1 + . . . + k6 x6 . The following state-
ments hold.
(a) F F14 is a Darboux polynomial of X with cofactor 2(k1 x1 + k4 x4 ).
(b) (F F14 )25 and (F F14 )36 are both Darboux polynomials of X with the same
cofactor −2(k1 x1 + k4 x4 ).

Proof The proof follows from Lemma 5.1 together with some easy calculations, and
is left to the reader as an exercise. 


The next result is a local version of the so-called Frobenius integrability theorem,
which will be used in the proof of Theorem 5.2.

Theorem 5.3 Consider the set of vector fields


n
∂
Vi = A(i)
j (z) , i = 1, . . . , k < n, z = (z 1 , . . . , z n ) ∈ Cn ,
j=1
∂z j
206 5 Algebraic, Analytic and Meromorphic Integrability

where A(z) = A(i)
j (z) is a C  smooth k × n matrix function defined on
1≤i≤k,1≤ j≤n
an open subset U of Cn ,  ∈ {2, 3, . . . , ∞, ω}. Assume that
• there exists a z 0 ∈ U such that rankA(z 0 ) = k,
• the vector fields V1 , . . . , Vk have the common C  first integrals G, G 1 , . . . , G n−k
in U , and the last n − k first integrals are functionally independent at z 0 .
Then there exists a C  function Φ in n − k variables defined in a neighborhood of
(G 1 (z 0 ), . . . , G n−k (z 0 )) and a neighborhood U0 of z 0 such that

G(z) = Φ(G 1 (z), . . . , G n−k (z)), z ∈ U0 .

Proof The holomorphic setting of this theorem can be found in Narasimhan [347,
Sect. 2.11.20] and Ilyashenko and Yakovenko [213, Theorem 2.9] with a proof from
the geometric point of view. A smooth version of this theorem may be found in [53].
Here we provide an analytic proof, which is due to Weng and Zhang [440]. Set
z = (u, v) with
u = (z 1 , . . . , z n−k ), v = (z n−k+1 , . . . , z n ),

and set
F = (G 1 (z), . . . , G n−k (z)), z = (z 1 , . . . , z n ) ∈ Cn .

Since the vector fields V1 , . . . , Vk , i.e. A(z), have rank k at z = z 0 , we assume

without loss of generality that the last k columns of A form an invertible matrix at
z 0 . Set
⎛ ⎞ ⎛ (1) ⎞
A(1)
1 . . . A1(k) (k)
An−k+1 . . . An−k+1
⎜ .. ⎟ , A (z) = ⎜ .. .. ⎟ .
A1 (z) = ⎝ ... . ⎠ 2 ⎝ . . ⎠
(1) (k) (1)
An−k . . . An−k An . . . An(k)

Then A2 (z) is invertible at z = z 0 .

Since G(z), G 1 (z), . . . , G n−k (z) are the common first integrals of the C  smooth
vector fields Vi in U , i = 1, . . . , k, we have

∂u F(z)A1 (z) = −∂v F(z)A2 (z), z ∈ U, (5.13)

where ∂u F(z) and ∂v F(z) are respectively the Jacobian matrices of F(z) with respect to
u and v. Since A2 (z) is invertible at z 0 , we get from (5.13) that ∂v F(z) can be expressed
as a linear function of ∂u F(z) in a neighborhood N0 of z 0 . In addition, ∂z F(z) has rank
n − k at z 0 , which forces ∂u F(z) to have rank n − k in some neighborhood N1 ⊂ N0
of z 0 .
According to the decomposition z = (u, v), set z 0 = (u 0 , v0 ). Applying the Im-
plicit Function Theorem to the functional equations

F(u, v) = c, c = (c1 , . . . , cn−k )

5.1 Algebraic First Integrals 207

in N1 , we get n − k C  smooth functions u = W(v, c) in a neighborhood M0 of

(u 0 , F(z 0 )) such that
F(W(v, c), v) ≡ c, (v, c) ∈ M0 .

Differentiating this equation with respect to v gives

∂u F(W(v, c), v)∂v W(v, c) + ∂v F(W(v, c), v) ≡ 0, (v, c) ∈ M0 .

Since ∂u F(W(v, c), v) is invertible, we have

∂v W(v, c) = A1 (W(v, c), v)(A2 (W(v, c), v))−1 , in M0 ,

where we have used (5.13).

For the common first integral G of the vector fields V1 , . . . , Vk , set

Φ(v, c) = G(W(v, c), v).

Then Φ is C  smooth, and satisfies

 
∂v Φ(v, c) = ∂u G(z ∗ )A1 (z ∗ ) + ∂v G(z ∗ )A2 (z ∗ ) (A2 (z ∗ ))−1 ≡ 0,

where z ∗ = (W(v, c), v), and we have used (5.14) and the fact that G is a common
first integral of the vector fields V1 , . . . , Vk .
The above proof shows that Φ is independent of v. Hence

G(z) = Φ(c) = Φ(G 1 (z), . . . , G n−k (z))

in some neighborhood U0 of z 0 . This proves the theorem. 



We now apply the Darboux polynomial F of X to construct a fourth polynomial

first integral.

Lemma 5.3 Assume that the Kirchoff vector field X has the homogeneous Darboux
polynomial F with cofactor K . The following statements hold.
(a) H4 := F F14 (F F14 )25 and H5 := F F14 (F F14 )36 are two polynomial first inte-
grals of X .
(b) Each of H4 and H5 is functionally independent of the three polynomial first
integrals H1 , H2 and H3 given in (5.8).

Proof (a) Set

R1 (x) = F(x)F14 (x), R2 (x) = (F F14 )25 (x), R3 (x) = (F F14 )36 (x). (5.14)

By Lemma 5.2, R1 and R2 are two Darboux polynomials of X with cofactors

2(k1 x1 + k4 x4 ) and −2(k1 x1 + k4 x4 ), respectively. This shows that H4 = R1 (x)R2 (x)
208 5 Algebraic, Analytic and Meromorphic Integrability

is a polynomial first integral of X . Similar arguments verify that H5 = R1 R3 is also

a polynomial first integral of X .
(b) Since F is irreducible, it follows that F14 , F25 and (F25 )14 are all irreducible.
We claim that R1 and R2 have no common factors. On the contrary we must have:
• either F = F25 , or F = −F25 , or F = (F25 )14 , or F = −(F25 )14 .
Note that

X (F) = (k1 x1 + k2 x2 + k3 x3 + k4 x4 + k5 x5 + k6 x6 )F,

X (F25 ) = −(k1 x1 − k2 x2 + k3 x3 + k4 x4 − k5 x5 + k6 x6 )F25 .

If either F = F25 or F = −F25 , we must have

k1 = k3 = k4 = k6 = 0,

contradicting |k1 |2 + |k4 |2 = 0. In addition, we have

X ((F25 )14 ) = (−k1 x1 − k2 x2 + k3 x3 − k4 x4 − k5 x5 + k6 x6 )(F25 )14 .

So in each of the cases F = (F25 )14 and F = −(F25 )14 we obtain

k1 = k2 = k4 = k5 = 0,

again contradicting |k1 |2 + |k4 |2 = 0. This proves the claim.

The next step is to prove that H1 , H2 , H3 and H4 = R1 R2 are functionally inde-
pendent. With this aim in mind we define the vector fields
⎛ ⎞
∂ H1 
∂ H1 
∂ H1 ∂ H1
⎜ ... ... ... ⎟
⎜ ∂ x1 ∂ xi ∂x j ∂ x6 ⎟
⎜   ⎟
⎜ ∂ H2 ∂ H2 ∂ H2 ∂ H2 ⎟
⎜ ... ... ... ⎟
⎜ ∂ x1 ∂ xi ∂x j ∂ x6 ⎟
Zi j = det ⎜
⎜ ∂H
⎟ , 1 ≤ i < j ≤ 6,
⎟
⎜ 3

∂ H3 
∂ H3 ∂ H3 ⎟
⎜ ... ... ... ⎟
⎜ ∂ x1 ∂ xi ∂x j ∂ x6 ⎟
⎜ ⎟
⎝ ∂  ∂  ∂ ∂ ⎠
... ... ...
∂ x1 ∂ xi ∂x j ∂ x6

where the hat denotes the absence of the element under it. Clearly H1 , H2 and H3
are the first integrals of each of the vector fields Zi j for 1 ≤ i < j ≤ 6.
On the contrary we assume that H1 , H2 , H3 and H4 are functionally dependent.
Then H4 is also a first integral of Zi j for all 1 ≤ i < j ≤ 6. It follows that

Zi j (R1 )R2 + Zi j (R2 )R1 ≡ 0, for all 1 ≤ i < j ≤ 6. (5.15)

Since R1 and R2 are relatively prime, there should exist polynomials pi j (x) such that
5.1 Algebraic First Integrals 209

Zi j (R1 ) = pi j (x)R1 (x), Zi j (R2 ) = − pi j (x)R2 (x), 1 ≤ i < j ≤ 6. (5.16)

Next we prove that p25 (x) = p36 (x) ≡ 0. Indeed, calculations via Mathematica
give
∂ ∂ ∂ ∂
Z25 = 2 A1 (x) + 2 A3 (x) + 2 A4 (x) + 2 A6 (x) ,
∂ x1 ∂ x3 ∂ x4 ∂ x6

with

A1 (x) = x3 (μ3 x1 x6 − μ1 x3 x4 ),
A3 (x) = x1 (μ1 x3 x4 − μ3 x1 x6 ),
A4 (x) = (ν1 − ν3 )x1 x32 + μ3 x6 (x1 x6 − x3 x4 ),
A6 (x) = (ν3 − ν1 )x12 x3 + μ1 x4 (x3 x4 − x1 x6 ),

and
∂ ∂ ∂ ∂
Z36 = 2B1 (x) + 2B2 (x) + 2B4 (x) + 2B5 (x) ,
∂ x1 ∂ x2 ∂ x4 ∂ x5

with

B1 (x) = x2 (μ2 x1 x5 − μ1 x2 x4 ),
B2 (x) = x1 (μ1 x2 x4 − μ2 x1 x5 ),
B4 (x) = (ν1 − ν2 )x1 x22 + μ2 x5 (x1 x5 − x2 x4 ),
B5 (x) = (ν2 − ν1 )x12 x2 + μ1 x4 (x2 x4 − x1 x5 ).

Note that R2 = R1 ◦ ρ25 by (5.14), and Z25 is independent of x2 and x5 . We get from
Z25 (R1 ) ◦ ρ25 = p25 R1 ◦ ρ25 and (5.16) that

Z25 (R2 ) = p25 ◦ ρ25 R2 .

From the second equality of (5.16) for i j = 25 we obtain

p25 ◦ ρ25 = − p25 . (5.17)

In another way, one can show that p25 is independent of x2 and x5 . It follows that
p25 ◦ ρ25 = p25 , which together with (5.17) verifies p25 ≡ 0. The same arguments
also show that p36 ≡ 0.
The above proof together with (5.16) shows that R1 is a polynomial first integral
of the vector fields Z25 and Z36 . Recall that H1 , H2 and H3 are also first integrals of
Z25 and Z36 .
Let Z = [Z25 , Z36 ] be the Lie bracket of the vector field Z25 and Z36 . Then
H1 , H2 , H3 and R1 are all first integrals of Z . Some calculations via Mathematica
yield
210 5 Algebraic, Analytic and Meromorphic Integrability


6
∂
Z = 4Ci (x) ,
i=1
∂ xi

where

C1 = μ1 (ν3 − ν2 )x1 x22 x32 + 2μ1 x2 x3 x4 (μ3 x2 x6 − μ2 x3 x5 ) + μ1 x1 μ2 x32 x52 − μ3 x22 x62 ,

C2 = μ1 (ν1 − ν3 )x12 x2 x32 + 2μ2 x1 x3 x5 (μ1 x3 x4 − μ3 x1 x6 ) + μ1 x2 μ3 x12 x62 − μ1 x32 x42 ,

C3 = μ1 (ν2 − ν1 )x12 x22 x3 + 2μ3 x1 x2 x6 (μ2 x1 x5 − μ1 x2 x4 ) + μ1 x3 μ1 x22 x42 − μ2 x12 x52 ,

C4 = μ1 (ν2 − ν3 )x22 x32 x4 + 2μ3 (ν1 − ν2 )x1 x22 x3 x6 + 2μ2 (ν3 − ν1 )x1 x2 x32 x5
+ μ1 x4 (μ3 x22 x62 − μ2 x32 x52 ) + 2μ2 μ3 x1 x5 x6 (x3 x5 − x2 x6 ),
C5 = (2(ν2 − ν1 )x1 x2 + μ1 x4 x5 )(μ1 x3 x4 − μ3 x1 x6 )x3

− μ1 (2x2 x4 − x1 x5 ) (ν3 − ν1 )x1 x32 + μ3 x6 (x3 x4 − x1 x6 ) ,
C6 = (2(ν1 − ν3 )x1 x3 + μ1 x4 x6 )(μ2 x1 x5 − μ1 x2 x4 )x2

− μ1 (2x3 x4 − x1 x6 ) (ν1 − ν2 )x1 x22 + μ2 x5 (x1 x5 − x2 x4 ) .

We now prove that the vector fields Z25 , Z36 and Z are linearly dependent in C6 .
Indeed, if not, by Theorem 5.3 the first integral R1 of the three vector fields should be
a C  smooth function of H1 , H2 , H3 at least in some open subset of C6 . Hence R1 is
a first integral of X , contradicting the fact that R1 is a proper Darboux polynomial
of X . This proves the claim.
Set ⎛ ⎞
Ai A j A k
Ji jk = ⎝ Bi B j Bk ⎠ , 1 ≤ i < j < k ≤ 6.
Ci C j Ck

The linear dependence of the vector fields Z25 , Z36 and Z in C6 implies that

det Ji jk ≡ 0, for all 1 ≤ i < j < k ≤ 6.

Some calculations via Mathematica show that

J124 = 16μ1 x3 ((ν1 − ν2 )x12 x22 + μ1 x22 x42 − 2μ2 x1 x2 x4 x5 + μ2 x12 x52 )
× (μ2 x3 x5 − μ3 x2 x6 )((ν1 − ν3 )x12 x32 + μ1 x32 x42 − 2μ3 x1 x3 x4 x6 + μ3 x12 x62 ).

Hence we get from det J124 ≡ 0 that

μ1 = 0, or ν1 = ν2 and μ1 = μ2 = 0,
or μ2 = μ3 = 0, or ν1 = ν3 and μ1 = μ3 = 0,
5.1 Algebraic First Integrals 211

which contradicts the fact that μi = 1/Ii = 0 for i = 1, 2, 3.

The above proofs verify that H1 , H2 , H3 and H4 = R1 R2 are functionally inde-
pendent first integrals of X .
The same arguments guarantee that H1 , H2 , H3 and H5 = R1 R3 are also func-
tionally independent first integrals of X . This completes the proof of the lemma.



Now Theorem 5.2 follows from Lemmas 5.1, 5.2 and 5.3. 


Remark From Theorem 5.2 and its corollary, it follows that the existence of a fourth
algebraic first integral of the Kirchoff equations forces the existence of a fourth
polynomial first integral. The question of whether the Kirchoff equations have other
polynomial first integrals functionally independent of the known ones is still open.

Another open problem is: does the existence of a nontrivial rational first integral
of the Kirchoff equations imply the existence of the fourth and fifth functionally
independent polynomial first integrals?

5.1.3 Euler Equations on the Lie Algebra so(4): Polynomial

First Integrals

Set
so(n, C) := {A ∈ Mn (C)| Aτ = −A},

where Mn (C) is the set of nth order matrices with entries in C, and τ denotes the
transpose of a matrix. Then so(n, C) under the Lie bracket defined by

[A, B] = AB − B A, for all A, B ∈ so(n, C),

forms a Lie algebra. Recall that a real or complex finite-dimensional Lie algebra is
a real or complex finite-dimensional linear space g endowed with a bracket operator

[·, ·] : g × g −→ g,

satisfying bilinearity, anti-symmetry and the Jacobian identity. For more details, see
e.g. Olver [351].
Taking a suitable basis of the Lie algebra so(4) (see Adler and van Moerbeke [3]),
the Euler equations corresponding to the Hamiltonian

1
6
H= λi xi2
2 i=1

can be written as
212 5 Algebraic, Analytic and Meromorphic Integrability

d x1
= (λ3 − λ2 )x2 x3 + (λ6 − λ5 )x5 x6 =: P1 (x1 , x2 , x3 , x4 , x5 , x6 ),
dt
d x2
= (λ1 − λ3 )x1 x3 + (λ4 − λ6 )x4 x6 =: P2 (x1 , x2 , x3 , x4 , x5 , x6 ),
dt
d x3
= (λ2 − λ1 )x1 x2 + (λ5 − λ4 )x4 x5 =: P3 (x1 , x2 , x3 , x4 , x5 , x6 ),
dt (5.18)
d x4
= (λ3 − λ5 )x3 x5 + (λ6 − λ2 )x2 x6 =: P4 (x1 , x2 , x3 , x4 , x5 , x6 ),
dt
d x5
= (λ4 − λ3 )x3 x4 + (λ1 − λ6 )x1 x6 =: P5 (x1 , x2 , x3 , x4 , x5 , x6 ),
dt
d x6
= (λ2 − λ4 )x2 x4 + (λ5 − λ1 )x1 x5 =: P6 (x1 , x2 , x3 , x4 , x5 , x6 ),
dt

where λ1 , . . . , λ6 ∈ C6 are parameters. For more information on the background of

the Euler equations we refer to Adler and van Moerbeke [3], Fomenko and Trofimov
[153, 154, 422], and Perelomov [358] and the references therein.
The Euler equations (5.18) always have the two polynomial first integrals


6
H1 = x1 x4 + x2 x5 + x3 x6 , H2 = xi2 .
i=1

Obviously H, H1 and H2 are functionally independent except in the case λ1 = . . . =

λ6 . Moreover, under either the product condition

λ 4 = λ1 , λ5 = λ2 , λ6 = λ3 ,

or the Manakov condition

M = λ1 λ4 (λ23 + λ56 ) + λ2 λ5 (λ31 + λ64 ) + λ3 λ6 (λ12 + λ45 ) = 0, (5.19)

the Euler equations have either the fourth polynomial first integrals (see
Maciejewski et al. [316])

H4 = λ1 x1 x4 + λ2 x2 x5 + λ3 x3 x6

or the fourth polynomial first integral

H3 = λ16 λ24 x42 + λ15 λ26 x52 + λ16 λ26 x62 ,

where
λi j = λi − λ j , i, j ∈ {1, . . . , 6}.

In 2008 Popov and Strelcyn [363] obtained the next result (for a simple proof, see
Popov et al. [364]).
5.1 Algebraic First Integrals 213

Theorem 5.4 If the Euler equations have a fourth rational first integral, then they
have a fourth polynomial first integral.

The following problem is still open:

Are there other conditions different from the product and Manakov conditions
under which the Euler equations have a fourth polynomial first integral?
More generally, we have the open problem:
Characterize all Euler equations which have a fourth polynomial first integral.
These open problems are related to the following conjecture posed by Popov and
Strelcyn in [363]:
The Euler equations can have a fourth polynomial first integral only if either the
Manakov condition or the product condition holds.
Corresponding to the conjecture, Llibre, Yu and Zhang [295] obtained the next
result.

Theorem 5.5 The Euler equations (5.18) either satisfy the Manakov condition, or
have at most four functionally independent polynomial first integrals.

The proof of Theorem 5.5 needs the following result, which is due to Zhang [464].

Theorem 5.6 Assume that the vector field (1.1) has a singularity at the origin. Let
λ1 , . . . , λn be the eigenvalues of the linear part of the vector field at the origin. Set
 
 
n 
n
L = k = (k1 , . . . , kn ) ∈ Zn+  ki λi = 0, |k| := ki > 0 .
i=1 i=1

If the vector field (1.1) has r < n functionally independent and analytic local first
integrals F1 (x), . . . , Fr (x) at the origin, and the Q-linear space generated by L is
r -dimensional, then any nontrivial analytic first integral of the vector field (1.1) in a
neighborhood of the origin is an analytic function in terms of F1 (x), . . . , Fr (x).

We remark that the functional independence in Theorem 5.6 is in the sense that
the gradients of the functions at the origin are of full rank.
The proof of Theorem 5.5 can also be obtained by using the next result, which is
due to Chen et al. [82, Theorem 1.1].

Theorem 5.7 For an analytic vector field (1.1) defined in (Rn , 0) with f (0) = 0, let
λ1 , . . . , λn be the eigenvalues of ∂x f (0), the Jacobian matrix of f at 0. Set
 

n
L = k = (k1 , . . . , kn ) ∈ (Z+ ) | n
ki λi = 0, |k| > 0 .
i=1

Then the number of functionally independent analytic local first integrals in (Rn , 0)
is no more than the maximal number of Q+ -linearly independent elements of L .
214 5 Algebraic, Analytic and Meromorphic Integrability

The element (k1 , . . . , kn ) in L of Theorem 5.7 is called the resonant lattice

of λ1 , . . . , λn . Recall that λ1 , . . . , λn are Q+ -linearly dependent if there exist
k1 , . . . , kn ∈ Q+ such that k1 λ1 + . . . + kn λn = 0. Otherwise they are called Q+
-linearly independent, where Q+ is the set of nonnegative rational numbers.
Proof of Theorem 5.5. Direct calculations show that the Euler equations have singu-
larities filled up by the three planes:

(x1 , 0, 0, x4 , 0, 0) , (0, x2 , 0, 0, x5 , 0) , (0, 0, x3 , 0, 0, x6 ) .

At the singularities S1 = (x1 , 0, 0, x4 , 0, 0), the resonant lattices (k1 , . . . , k6 ) of

the corresponding eigenvalues satisfy
   
α − β (k3 − k4 ) + α + β (k5 − k6 ) = 0, (5.20)

where

α = − (λ12 λ13 + λ15 λ16 ) x12 + (λ24 λ46 + λ34 λ45 ) x42 ,
β = −4Δ + α 2 ,
  
Δ = λ12 λ15 x12 + λ24 λ45 x42 λ13 λ16 x12 + λ34 λ46 x42 .

Clearly, the Eq. (5.20) has linearly independent nonnegative integer solutions

(k1 , . . . , k6 ) := (1, 0, 0, 0, 0, 0), (0, 1, 0, 0, 0, 0), (0, 0, 1, 1, 0, 0) and (0, 0, 0, 0, 1, 1).

In order for (5.20) to have more linearly independent nonnegative integer solutions,
we must have either
√ √
(i) (α − β)(α + β) = 0; or 
√ √ √  √
(ii) (α − β)(α + β) = 0, and α − β/ α + β is a rational number.
We next distinguish these two cases.
Case (i) Some calculations show that the condition in (i) is equivalent to one of the
following 8 different conditions (here the computations have taken into account all
the singularities S1 ):
• λ1 = λ2 , λ4 = λ5 ;
• λ1 = λ5 , λ2 = λ4 ;
• λ1 = λ3 , λ4 = λ6 ;
• λ1 = λ6 , λ3 = λ4 ;
• λ1 = λ2 = λ4 ;
• λ1 = λ4 = λ5 ;
• λ1 = λ3 = λ4 ;
• λ1 = λ4 = λ6 .
One can check that the first four conditions satisfy the Manakov condition, and
that M restricted to the last four conditions gives
5.1 Algebraic First Integrals 215

• M|λ1 =λ2 =λ4 = λ13 λ15 λ16 ,

• M|λ1 =λ4 =λ5 = λ12 λ13 λ16 ,
• M|λ1 =λ3 =λ4 = λ21 λ15 λ16 ,
• M|λ1 =λ4 =λ6 = λ21 λ13 λ15 ,
where M is the polynomial defined in the Manokov condition.
Similar arguments acting on the singularities (0, x2 , 0, 0, x5 , 0) and (0, 0, x3 ,
0, 0, x6 ) show that under any of the conditions λ1 = λ2 = λ4 , or λ1 = λ4 = λ5 , or
λ1 = λ3 = λ4 , or λ1 = λ4 = λ6 , the Euler equations either satisfy the Manakov con-
dition, or the number of linearly independent resonant lattices of the eigenvalues at
some singularity is at most 4. This proves the theorem in the case (i).
Case (ii) Now we have Δ = 0 and α = 0. Set
   
α− β/ α + β = m/n, m, n ∈ N coprime.

We have
Δ m 2n2
= .
α2 (m 2 + n 2 )2

This implies that Δ must be the square of λ12 λ15 x12 + λ24 λ45 x42 , or of λ13 λ16 x12 +
λ34 λ46 x42 modulo a positive constant. Without loss of generality we set
 
λ13 λ16 x12 + λ34 λ46 x42 = L 2 λ12 λ15 x12 + λ24 λ45 x42 ,
 
α = K λ12 λ15 x12 + λ24 λ45 x42 ,

where K is a constant satisfying L/K = mn/(n 2 + m 2 ). These two conditions are

equivalent to

λ13 λ16 = λ12 λ15 L 2 , (5.21)

λ34 λ46 = λ24 λ45 L 2 , (5.22)
−λ12 λ13 − λ15 λ16 = λ12 λ15 K , (5.23)
λ24 λ46 + λ34 λ45 = λ24 λ45 K . (5.24)

If λ5 = λ6 , Eqs. (5.21) and (5.22) have two solutions

λ2 = (L 2 λ1 + λ31 )/L 2 , λ6 = λ4 ,

and
λ2 = (L 2 λ4 + λ34 )/L 2 , λ6 = λ1 .

They both satisfy the Manakov condition.

If λ5 = λ6 , Eqs. (5.21) and (5.22) have a unique solution
216 5 Algebraic, Analytic and Meromorphic Integrability

λ16 λ46 − L 2 (λ1 λ46 − λ6 λ45 )

λ2 = − ,
L 2 λ56
λ1 λ45 − λ5 λ46 − L 2 λ15 λ45
λ3 = − .
λ56

They also satisfy the Manakov condition. This proves the theorem in case (ii).
This completes the proof of Theorem 5.5. 


5.1.4 The 5-Dimensional Lorenz Systems: Darboux

and Analytic Integrability

The following 5-dimensional differential system in R5

dU
= −V W + bV Z ,
dT
dV
= U W − bU Z ,
dT
dW
= −U V, (5.25)
dT
dX
= −Z ,
dT
dZ
= bU V + X,
dT
was constructed by Lorenz [307] in 1986 to describe coupled Rosby waves and
gravity waves, where b ∈ R is a parameter. Lorenz [307] was mainly concerned
with the invariant slow manifold of system (5.25). In [269, 291] the authors studied
its Darboux and global integrability. Here we summarize the main results of [291],
which solved completely the global analytic and Darboux integrability of system
(5.25).
Recall that a function H is a global analytic first integral of system (5.25) if the
domain where H is defined is R5 .
Theorem 5.8 For the 5-dimensional Lorenz differential system (5.25), the following
statements hold:
(a) If b = 0, system (5.25) has three functionally independent global analytic first
integrals, namely H1 = U 2 + V 2 , H2 = V 2 + W 2 and H3 = X 2 + Z 2 . Further-
more, any other global analytic first integral is an analytic function in terms of
H1 , H2 and H3 .
(b) If b = 0, one has the next results.
(b1 ) Any global analytic first integral is a function in terms of H1 and H4 =
H2 + H3 ;
5.1 Algebraic First Integrals 217
√
(b2 ) The unique
√ proper irreducible Darboux
√ √ U + −1V and
polynomials are
U − −1V with the cofactors −1(W − bZ ) and − −1(W − bZ ), re-
spectively. All other proper Darboux polynomials are compositions of these
two.
(b3 ) The system has only two functionally independent Darboux first integrals.

Statement (a) was stated and proved in [269], and statement (b) was proved in
[291], which solved an open problem posed in [269].

Proof The main ideas of the proofs follow from those of Llibre and Zhang [298] using
the method of characteristic curves for solving linear partial differential equations
together with some new techniques.
Statement (a) can be proved using the idea of (b1 ), but it is much easier than that
of (b1 ). The details are omitted.
We now present a sketch of the proof of statement (b).
(b1 ) Let H be a global analytic first integral of system (5.25), and suppose it has the
Taylor expansion
∞

H= H j (U, V, W, X, Z ),
j=m

with m ∈ N, and H j homogeneous polynomials of degree j for j = m, m + 1, . . ..

By the fact that H is a first integral of system (5.25) one gets

∂ Hm ∂ Hm
−Z +X = 0, (5.26)
∂X ∂Z
∂ H j+1 ∂ H j+1 ∂ Hj ∂ Hj
−Z +X = V (W − bZ ) − U (W − bZ ) (5.27)
∂X ∂Z ∂U ∂V
∂ Hj ∂ Hj
+ UV − bU V ,
∂W ∂Z
for j = m, m + 1, . . ..
Applying the characteristic method to the linear partial differential equation
(5.26), one obtains the general solution of (5.26)

Hm (U, V, W, X, Z ) = G m (H3 , U, V, W ),

with G m a polynomial function in its variables. For j ≥ m under the change of

variables
H3 = X 2 + Z 2 , Z = Z , (5.28)

the linear partial differential equation (5.27) becomes the ordinary differential
equation
218 5 Algebraic, Analytic and Meromorphic Integrability

 j+1
dH j
∂H j
∂H
H3 − Z 2 = V (W − bZ ) − U (W − bZ )
dZ ∂U ∂V

∂ Hj 
∂ Hj
+ UV − 2bU V Z , (5.29)
∂W ∂ H3

where Hj is H j written in the variables H3 , U, V, W, Z instead of the variables

U, V, W, X, Z .
m = G m (H3 , U, V, W ) some calculations show that
For j = m, since H

m+1
dH A1 A2
= + Z, (5.30)
dZ H3 − Z 2 H3 − Z 2

where
∂ ∂ ∂
A1 = L1 (G m ), L1 = V W − UW + UV ,
∂U ∂V ∂W
∂ ∂ ∂
A2 = L2 (G m ), L2 = −bV + bU − 2bU V .
∂U ∂V ∂ H3

Integrating the Eq. (5.30) in Z gives


m+1 (H3 , U, V, W, Z ) = A1 arctan  Z
H − A2 H3 − Z 2 + G m+1 (H3 , U, V, W ),
H3 − Z 2

with G m+1 a constant of integration. In order for Hm+1 to be a polynomial we must

have A1 = 0, i.e.
∂G m ∂G m ∂G m
VW − UW + UV = 0.
∂U ∂V ∂W

Its general solution is of the form g(H1 , H2 ), with g any continuous differentiable
function. So we must have

Hm = G m (H3 , U, V, W ) = Rm (H1 , H2 , H3 ),

with Rm a homogeneous polynomial in terms of H1 , H2 , H3 . This implies that m is

even.
Now we have
 ∂ Rm ∂ Rm
m+1 = 2bU V Sm H3 − Z 2 + G m+1 (H3 , U, V, W ),
H Sm = − .
∂ H3 ∂ H2

Consequently, Eq. (5.27) with j = m + 1 becomes

 m+2  
dH Z
H3 − Z 2 = A3 + A4 Z + A5 H3 − Z 2 + A6 Z H3 − Z 2 − A7  ,
dZ H3 − Z 2
5.1 Algebraic First Integrals 219

where

A3 = L1 (G m+1 ),
A4 = L2 (G m+1 ),
   
∂ Sm ∂ Sm ∂ Sm
A5 = 2bV 2 W Sm + U − 2bU 2 W Sm + V + 2bU 2 V 2 ,
∂U ∂V ∂W
   
∂ Sm ∂ Sm
A6 = −2b2 V 2 Sm + U + 2b2 U 2 Sm + V ,
∂U ∂V
A7 = −2b2 U 2 V 2 Sm .

m+2 with respect to Z gives

Integrating this last equation on H

Z  A6 2
Hm+2 =A3 arctan  − A4 H3 − Z 2 + A5 Z + Z
H3 − Z 2 2
S7  
+ log H3 − Z 2 + G m+2 (H3 , U, V, W ),
2
with G m+2 a constant of integration. This forces that A3 = 0 and A7 = 0.
The equation A3 = 0 implies that

G m+1 = Rm+1 (H1 , H2 , H3 ),

with Rm+1 a polynomial function. Consequently, Hm+1 = 0 because Hm+1 is a

homogeneous polynomial of odd degree.
The equation A7 = 0 is the same as

∂ Rm ∂ Rm
= . (5.31)
∂ H3 ∂ H2

We will use this equation to obtain the expression of Rm . To do so, set

 
m m−
Rm (H1 , H2 , H3 ) = ck H1 H2m−−k H3k .
=0 k=0

Substituting this last expression into Eq. (5.31), and equating the coefficients of H1
and of H2m−−k H3k , we arrive at
 
m−
ck = c0 .
k
220 5 Algebraic, Analytic and Meromorphic Integrability

It follows that


m
Rm = c0 (H1 + H2 )m− H3 =: Rm∗ (H1 + H2 , H3 ),
=0

with Rm∗ a homogenous polynomial.

By recursive calculations one gets the proof of statement (b1 ). √
(b2 ) According
√ to [291], introduce the new variables Y1 = U + −1V and Y2 =
U − −1V . Then system (5.25) can be written as
√
Y1 = −1Y1 (W − bZ ),
√
Y2= − −1Y2 (W − bZ ),
√
 −1  2 
W = Y1 − Y22 , (5.32)
4
X  = −Z ,
√
 −1b  2 
Z =− Y1 − Y22 + X.
4
Let f be a proper Darboux polynomial of system (5.32) with cofactor k of the form
k = a0 + a1 Y1 + a2 Y2 + k1 , where k1 = a2 W + a4 X + a5 Z .
Distinguishing the two cases: k1 = 0 and k1 = 0, Llbire and Valls [291] obtained
the following results.
• The unique irreducible proper Darboux polynomials of system (5.32) with cofactor
satisfying k1 = 0 are Y1 and Y2 .
• System (5.32) has no irreducible proper Darboux polynomials with cofactor sat-
isfying k1 = 0.
Their proofs involve recursive computations and restriction to the invariant hy-
perplanes Y1 = 0 and Y2 = 0. For details, see the original proof of Llibre and
Valls [291].
(b3 ) To prove this statement, one needs to compute the exponential factors of system
(5.32). By the properties of the exponential factors, one can set the exponential factor
to be of the form  
g
E = exp ,
Y1n 1 Y2n 2 H4n 3

with n 1 , n 2 , n 3 ∈ Z+ and g a polynomial which is coprime with Y1 if n 1 > 0, or

coprime with Y2 if n 2 > 0, or coprime with H4 if n 3 > 0. As shown in [291], if E is
an exponential factor, it forces n 1 = n 2 . Hence, any exponential factor (if it exists)
must be of the form  
g
E = exp .
H1n 1 H4n 3
5.1 Algebraic First Integrals 221

In terms of the definition of the Darboux first integral and statement (b2 ), a
Darboux first integral of system (5.32) (if it exists) should be of the form
 
g
H= Y11 Y22 H43 exp 4 .
H1n 1 H4n 3

As proved in [291] one gets that either 1 = 2 and 4 = 0, or 1 = 2 and g =

g(H1 , H4 ) with g a polynomial in H1 and H4 . This shows that any Darboux first
integral is composed of H1 and H4 . The statement follows.
This completes the proof of the theorem. 


5.2 Natural Hamiltonian Systems: Polynomial

and Rational Integrability

In this section we consider the Hamiltonian systems with their Hamiltonians con-
sisting of kinetic and potential energies, called natural Hamiltonian systems.

5.2.1 Hamiltonian Systems in the Canonical Form

Assume that the Hamiltonian function is of the form

1
m
H (q, p) = T ( p) + V (q) = μi pi2 + V (q),
2 i=1

where q = (q1 , . . . , qm ) are the location coordinates, p = ( p1 , . . . , pm ) are the mo-

menta, and μ = (μ1 , . . . , μm ) ∈ Cm are the parameters. The canonical Hamiltonian
system associated to the Hamiltonian function H is

q̇i = μi pi , ṗi = −∂qi V, i = 1, . . . , m. (5.33)

Denote by X H the Hamiltonian vector field associated to system (5.33). If the poten-
tial V (q) is a polynomial function, system (5.33) is called a polynomial Hamiltonian
system. A second first integral of X H is a first integral which is functionally inde-
pendent of the Hamiltonian function H .
Maciejewski and Przybylska [317] in 2004 studied the polynomial integrability
of the natural polynomial Hamiltonian systems (5.33) and some related questions
(see e.g. Nakagawa et al. [345]), where the authors posed a series problems:
222 5 Algebraic, Analytic and Meromorphic Integrability

• Does there exist a natural polynomial Hamiltonian system which has a second
rational first integral?
• If a natural Hamiltonian system has a second rational first integral, does it imply
the existence of a second polynomial first integral?
• Does there exist a natural polynomial Hamiltonian system which has an analytic
first integral but not a polynomial first integral?
• Does the existence of a second analytic first integral for a natural Hamiltonian
system imply the existence of a second polynomial first integral?
• Does there exist a natural polynomial Hamiltonian system which has a meromor-
phic first integral but not a rational first integral?
• Does the existence of a second meromorphic first integral for a natural Hamiltonian
system imply the existence of a second rational first integral?
For the first problem, García et al. [165, Proposition 5] in 2010 constructed a
quadratic Hamiltonian function

p12 p2 q2 q2
H (q, p) = + 2 − 1 − 2.
2 2 2 2
Its associated Hamiltonian vector field has the rational first integral

p1 + q 1
F(q, p) = .
p2 + q 2

Note that this Hamiltonian system is linear. The problem is still open for nonlinear
natural polynomial Hamiltonian systems.
García et al. [165, Theorem 3 and Proposition 6] answered the second problem.
Theorem 5.9 For the Hamiltonian vector field X H associated to the Hamiltonian
system (5.33), assume that the potential V is a polynomial of degree > 2.
(a) If F(q, p) is a Darboux polynomial of X H , then HF = F F ρ is a polynomial
first integral of X H , where ρ(q, p) = (q, − p), F ρ = F ◦ ρ.
(b) If m ≥ 2, and at least two μi are not zeros, then HF is a second first integral of
XH .
(c) If X H has a rational first integral, then it has a polynomial first integral. Fur-
thermore, if m ≥ 2, and at least two μi are not zeros, then X H has a second
polynomial first integral.
Proof The proof follows from the ideas of [165], where the key tool is the use of
quasi-homogeneous polynomials. For a polynomial P(x) ∈ C[x], we denote by P +
its highest order weight homogeneous part with respect to a weight exponent γ , and
by degγ (P) the weight degree of P, which is in fact the weight degree of P + .
For a polynomial vector field X , we denote by X + its highest order weight
homogeneous part with respect to the weight exponent γ , and by degγ (X ) the
weight degree of X , which is in fact the weight degree of X + .
Recall that for the weight homogeneous polynomials in Rn or Cn with respect to
a weight exponent γ = (γ1 , . . . , γn ) ∈ Nn , the following conditions are equivalent:
5.2 Natural Hamiltonian Systems: Polynomial and Rational Integrability 223

• F is weight homogeneous of degree r with respect to γ ;

• F(t γ1 x1 , . . . , t γn xn ) = t r F(x1 , . . . , xn ), for all t > 0;
• γ1 x1 ∂x1 F + . . . + γn xn ∂xn F = r F.
For the Darboux polynomials, the next result can be checked easily. Assume that
F is a Darboux polynomial of the polynomial vector field X with cofactor K . Then
• F + is a Darboux polynomial of X + with cofactor either K + or 0.
• degγ (K ) ≤ degγ (X ).
For the Hamiltonian vector field X H , set

γ ∗ = (γ1 , . . . , γm , γm+1 , . . . , γ2m ) = (2, . . . , 2, r, . . . , r ),

where r = deg V > 2. Then degγ ∗ T = degγ ∗ V = 2r , H + = T ( p) + V + (q), and

the following statements hold:
• If V is a homogeneous polynomial of degree r ≥ 2, then the Hamiltonian vector
field X H is weight homogeneous with respect to the weight exponent γ ∗ , and
degγ ∗ (X H ) = r − 2.
• Assume that F is a Darboux polynomial of X H , and deg V > 2. Then K ∈ C[q].
The second statement can be proved by observing that

degγ ∗ (K ) ≤ degγ ∗ (X H ) = r − 2 and degγ ∗ ( pi ) = r, i = 1, . . . , m.

Next for the transformation ρ : (q, p) → (q, − p), the following statements hold.
• The Hamiltonian vector field X H is ρ-antisymmetric, i.e. ρ∗ (X H ) := (∂ρX H ) ◦
ρ −1 = −X H , where ∂ρ denotes the Jacobian matrix of ρ.
• If F is a Darboux polynomial of X H with cofactor K , then F ρ is also a Darboux
polynomial of X H with cofactor −K .
The first result follows from the fact that ∂ρX H = −X H ◦ ρ. The second one
can be proved by observing that K ρ = K , and

X H F = K F =⇒ ρ∗ (X H F) = ρ∗ (K F) =⇒ ρ∗ (X H )ρ∗ (F) = ρ∗ (K )ρ∗ (F)

=⇒ ρ∗ (X H )F ρ = K ρ F ρ = K F ρ =⇒ X H F ρ = −K F ρ .

(a) The above proof shows that

X H (F F ρ ) = X H (F)F ρ FX H (F ρ ) = (K − K )F F ρ = 0.

This means that HF = F F ρ is a first integral of the Hamiltonian vector field.

(b) The proof needs the following result, due to Baider et al. [22, Proposition 1.16],
which shows the equivalence of functional independence and algebraic indepen-
dence. Recall that the rational functions F1 , . . . , Fm ∈ K(x) are algebraically de-
pendent over K if there exists a nonzero polynomial W ∈ K[Z 1 , . . . , Z m ] such that
224 5 Algebraic, Analytic and Meromorphic Integrability

W (F1 (x), . . . , Fm (x)) ≡ 0. Otherwise they are called algebraically independent,

that is for any nonzero polynomial W ∈ K[Z 1 , . . . , Z m ] we have W (F1 (x), . . . ,
Fm (x)) ≡ 0.

Proposition 5.4 In C(x) the functional and algebraic independences are equivalent.

The proof of this last proposition needs a field extension. We refer to Cong et al.
[117, Lemma 5] and its proof, see also the following Lemma 7.4 and its proof.
We now continue the proof of statement (b). First we claim that the Hamiltonian
function H is irreducible. Indeed, we assume without loss of generality that μ1 , μ2 =
0. On the contrary, if H is reducible then
  m 

m 
m 
2H = μi pi2 + 2V (q) = αi pi + W1 βi pi + W2 ,
i=1 i=1 i=1

where αi , βi ∈ C, i = 1, . . . , m, W1 , W2 , ∈ K[q], W1 W2 = 2V . Comparing the co-

efficients of p1 and p2 , together with some calculations, we get μ1 μ2 = 0, a contra-
diction. So, H is irreducible.
Second we prove that H and HF are functionally independent. Since the Hamil-
tonian vector field X H is weight homogeneous with respect to the weight exponent
γ ∗ , we can assume that F is weight homogeneous with respect to γ ∗ . On the contrary,
if H and HF are functionally dependent, then by Proposition 5.4 they are algebraically
dependent. So there exists a W ∈ K[X, Y ] such that

 := W (H, HF ) ≡ 0.
W

Since H and HF are weight homogeneous with respect to γ ∗ , for any weight homo-
geneous part of W , say W , we also have W (H, HF ) ≡ 0. This shows that W can be
chosen as a lowest order weight homogeneous polynomial such that W (H, HF ) ≡ 0.
Set
W (X, Y ) = X i + Y W1 (X, Y ), 0 = W1 ∈ K[X, Y ].

Then
HF W1 (H, HF ) = −H i .

This implies that HF divides H , contradicting the fact that H is irreducible and HF
is reducible. Hence H and HF are functionally independent.
(c) Since the numerator and the denominator of a rational first integral are both
Darboux polynomials with the same cofactors, the proof follows from (b).
This completes the proof of the theorem. 

On the relation between the Darboux polynomial and the polynomial and rational
first integrals of the natural Hamiltonian system in the canonical form (5.33), under
the assumption that system (5.33) is symmetric or antisymmetric with respect to an
involution ρ(q, p), Llibre et al. [270] in 2012 proved that the existence of a proper
5.2 Natural Hamiltonian Systems: Polynomial and Rational Integrability 225

Darboux polynomial f (q, p) implies either a polynomial first integral of the form
fρ ∗ f or a rational first integral of the form f /ρ ∗ f .

5.2.2 Hamiltonian Systems in a Generalized Form

This part introduces the results obtained by García and Hernández-Bermejo [166] in
2013. They considered a generalization of Theorem 5.9.
Proposition 5.5 Let A be an m × m invertible real matrix, and μ1 , . . . , μm be real
constants. Consider the Hamiltonian systems of m degrees of freedom defined in R2m
with the Hamiltonian
⎛ ⎞2
1  ⎝
m m
H (q, p) = μi ai j p j ⎠ + V (A T q).
2 i=1 j=1

Assume that V (q) is a polynomial of degree at least 3, and that the Hamiltonian
system associated to H admits a proper Darboux polynomial F(q, p). The following
statements hold.
(a) HF (q, p) = F(q, p)F(q, − p) is a polynomial first integral of the system.
(b) If m ≥ 2 and at least two μi are not zero, then HF is a second polynomial first
integral.
The proof uses the linear canonical (symplectic) transformation
 
(q, p)τ = diag (A−1 )τ , A (Q, P)τ

which sends the Hamiltonian H (q, p) to

1
m
H ∗ (Q, P) = μi Pi2 + V (Q),
2 i=1

and Theorem 5.9.

In [166] this last proposition was generalized to a more general case.
Theorem 5.10 Consider a Poisson Hamiltonian system of m degrees of freedom
with a structure matrix J (x) and a smooth Hamiltonian

1
m
H (x) = μi Φm+i
2
(x) + V (Φ1 (x), . . . , Φm (x)) , x ∈ Ω ⊂ R2m ,
2 i=1

that is
ẋ = J (x)∇ H (x).
226 5 Algebraic, Analytic and Meromorphic Integrability

Assume that
• the map y = Φ(x) = (Φ1 (x), . . . , Φ2m (x)) is a C1 diffeomorphism
τ in Ω,
• the structure matrix J (x) = ∂ y Φ −1 ◦ Φ(x)S2m ∂ y Φ −1 ◦ Φ(x), where S2m is
the canonical symplectic matrix of order 2m, i.e.
 
0m E m
S2m = ,
−E m 0m

where E m is the unit matrix of order m, and 0m is the matrix of order m with entries
all vanishing,
• the potential V (q) is a polynomial of degree at least 3 in terms of q,
• the Poisson Hamiltonian has a proper Darboux polynomial

F(x) = G(Φ(x)),

with G a polynomial in 2m variables.

Then

HF (x) := F(Φ(x))F (Φ1 (x), . . . , Φm (x), −Φm+1 (x), . . . , −Φ2m (x))

is a first integral of the Poisson Hamiltonian system. Moreover, if at least two μi ’s

are not zero, then HF (x) is a second first integral of the system.
This last theorem can be proved using the change of variables

(q, p) = (Φ1 (x), . . . , Φm (x), Φm+1 (x), . . . , Φ2m (x)) ,

which is symplectic.
The structure matrix in this last theorem is nondegenerate. In the degenerate case,
García and Hernández-Bermejo [166, Theorem 3] presented a characterization as
follows.
Theorem 5.11 Consider the Poisson Hamiltonian system

ẋ = J (x)∇ H (x), x ∈ Ω ⊂ R2m+s ,

with the smooth Hamiltonian

1
m
H (x) = 2 (x) + V (Φ (x), . . . , Φ (x)) + W Φ
μi Φm+i

1 m 2m+1 (x), . . . , Φ2m+s (x) ,
2
i=1
5.2 Natural Hamiltonian Systems: Polynomial and Rational Integrability 227

which satisfies
• the map

y = Φ(x) : = Φ H (x) + Φ D (x)

= (Φ1 (x), . . . , Φ2m (x)) ⊕ (Φ2m+1 (x), . . . , Φ2m+s (x))

is a C 1 diffeomorphism in Ω, where ⊕ is the direct sum of two linear spaces,

τ
• the structure matrix J (x) = ∂ y Φ −1 ◦ Φ(x)S2m,s ∂ y Φ −1 ◦ Φ(x), where
 
S2m 0
S2m,s := S2m ⊕ 0s = ,
0 0s

• the potential V is a polynomial of degree at least 3 in terms of its variables,

• the Poisson Hamiltonian system has a proper Darboux polynomial F(x) =
G(Φ H (x)), with G a polynomial in the 2m variables.
Then

HF (x) := F(Φ H (x))F (Φ1 (x), . . . , Φm (x), −Φm+1 (x), . . . , −Φ2m (x))

is a first integral of the Poisson Hamiltonian system. Moreover, if at least two μi ’s are
not zero, then HF (x) is a first integral functionally independent of the Hamiltonian
H and the Casimir functions Φ2m+1 (x), . . . , Φ2m+s (x).
The proof can be completed by using the change of variables

qi = Φi (x), i = 1, . . . , m,
pi = Φm+i (x), i = 1, . . . , m,
z j = Φ2m+ j (x), j = 1, . . . , s,

which send the Poisson Hamiltonian system to the Darboux canonical form. For
detailed information on the Darboux canonical form, see e.g. Weinstein [439].

5.3 Hamiltonian Systems: Integrability via the Differential

Galois Group

In this subsection we mainly recall some results on the integrability of Hamiltonian

systems with some special potentials, including some extensions to general systems.
We will not prove these results here, instead referring the reader to the original proofs.
These results provide either some necessary conditions or concrete characterizations
on whether a given Hamiltonian system is integrable or not. One of the main tools
for studying the integrability of natural Hamiltonian systems with a homogeneous
228 5 Algebraic, Analytic and Meromorphic Integrability

potential is the Morales–Ramis theory, which was derived from the differential Galois
group of linear differential systems.

5.3.1 Hamiltonian Systems Having Homogeneous Potentials

5.3.1.1 Morales–Ramis Theory via the Differential Galois Group

In this part, for the reader’s convenience we recall the necessary results on
Morales–Ramis theory, which depends on differential Galois theory. For more de-
tailed information on differential Galois theory, see e.g. van der Put and Singer [368],
or Morales–Ruis [334].
First we recall the definition of a Darboux point. Consider the natural Hamiltonian
systems (5.33) with a homogeneous potential V (q) of degree k ∈ Z, where the case
when k is a negative integer means the potential is an inverse of a homogeneous
polynomial of degree −k. A nonzero vector d ∈ Cm is a Darboux point of system
(5.33) if it is a solution of the gradient equation

∇V (d) = γ d,

where γ ∈ C∗ := C \ {0}. Since V is homogeneous of degree k, it follows that for

any α ∈ C∗ , d ∗ = αd is also a Darboux point of system (5.35), because ∇V (d ∗ ) =
α k−1 ∇V (d) = α k−2 γ d ∗ . But d and d ∗ are the same point in the complex projective
space CPm−1 . So, in the definition of the Darboux point we can take γ = 1.
Let d be a Darboux point of the potential V in system (5.33). Define
 
P(d) := (q, p) ∈ C2m | (q, p) = (μd, νd), (μ, ν) ∈ C2 .

Then P(d) is a two-dimensional invariant plane under the flow of system (5.33).
Restricted to this invariant plane system (5.33) becomes

μ̇ = ν, ν̇ = −μk−1 , (5.34)

whose phase curves are all given by

 
1 1
γk,ρ := (μ, ν) ∈ C2 | μk + ν 2 = ρ , ρ ∈ C,
k 2

provided that k ∈ Z := Z \ {0}.

For each solution (μ(t), ν(t)) of (5.34), one can check that

(q(t), p(t)) := (μ(t)d, ν(t)d)

5.3 Hamiltonian Systems: Integrability via the Differential Galois Group 229

is a solution of system (5.33). Denote this phase curve by

 
Γk,ρ := (q, p) ∈ C2m | (q, p) = (μd, νd), (μ, ν) ∈ γk,ρ .

Let λ be the eigenvalues of the Hessian matrix of V at d

⎛ ⎞
∂ 2 V (d) ∂ 2 V (d) ∂ 2 V (d)
⎜ ∂q 2 ...
⎜ 1 ∂q1 ∂q2 ∂q1 ∂qm ⎟ ⎟
⎜ .
.. .
.. . .. .. ⎟
∂q V (d) = ⎜
2
. ⎟.
⎜ ⎟
⎝ ∂ 2 V (d) ∂ 2 V (d) ∂ V (d) ⎠
2
...
∂qm ∂q1 ∂qm ∂q2 ∂qm2

Then one of the eigenvalues is k − 1, which is called the trivial eigenvalue. The
other eigenvalues are called nontrivial eigenvalues, and satisfy the next result when
m = 2.
Theorem 5.12 Assume that V (q) = V (q1 , q2 ) is a homogeneous potential of
degree k ∈ Z, and has k different Darboux points d1 , . . . , dk . Let λi be the non-
trivial eigenvalue of the Hessian matrix of V (q) at di . Then


k
1
= −1.
i=1
λi − 1

This result was proved by Maciejewski and Przybylska [319] in 2005 for positive
degree, and by Llibre et al. [254] in 2011 for negative degree.
We now recall the Morales–Ramis theory, see [334–336], which is one of the most
beautiful applications of differential Galois theory to the theory of integrability.
Theorem 5.13 Assume that the Hamiltonian system (5.33) with a homogeneous
potential V of degree k ∈ Z \ {0} is integrable in the Liouvillian sense and whose
first integrals are meromorphic in a connected neighborhood of Γk,ρ , ρ ∈ C∗ , then
any eigenvalue λ of the Hessian matrix of V (q) at any Darboux point must satisfy
the conditions given in Table 5.1, where p is an integer.
The proof of this last theorem depends on an analysis of the linear variational
equation of system (5.33) along the phase curves Γk,ρ with ρ = 0, and the next
result, which is due to Morales-Ruiz and Ramis [334, 335].
Theorem 5.14 Assume that the Hamiltonian system (5.33) is Liouvillian integrable
with meromorphic first integrals in a neighbourhood of a phase curve Γ correspond-
ing to a particular solution. Then, the identity component G0 of the differential Galois
group G of the variational equations of system (5.33) along Γ is Abelian.
As mentioned in [60], Theorem 5.13 could be proved, without using differential
Galois theory, by analyzing the monodromy group of the variational equations, see
Ziglin [484].
230 5 Algebraic, Analytic and Meromorphic Integrability

Table 5.1 List of the degree and the associated nontrivial eigenvalues
Degree Eigenvalue λ
k 1  
k p+ p( p − 1), k − 1 + k 2 p( p + 1)
2 2k
9 5 9 1
5 − + (1 + 3 p)2 , − + (2 + 5 p)2
40 18 40 10
1 2
4 − + (1 + 3 p)2
8 9
1 1 1 3
− + (1 + 3 p)2 , − + (1 + 4 p)2
24 6 24 32
3
1 3 1 3
− + (1 + 5 p)2 , − + (2 + 5 p)2
24 50 24 50
±2 arbitrary
25 1 25 3
− (1 + 3 p)2 , − (1 + 4 p)2
24 6 24 32
−3
25 3 25 3
− (1 + 5 p)2 , − (2 + 5 p)2
24 50 24 50
9 2
−4 − (1 + 3 p)2
8 9
49 5 49 1
−5 − (1 + 3 p)2 , − (2 + 5 p)2
40 18 40 10

5.3.1.2 Homogeneous Potentials of Two Degrees of Freedom of Degree

−3, −2, . . . , 3, 4

As we know, a systematic study of the integrability of natural Hamiltonian systems

with homogeneous potentials was initiated by Yoshida in [453, 454], with the help
of Ziglin’s theory [482, 483]. In this part we consider the two degrees of freedom of
Hamiltonian systems

q̇i = pi , ṗi = −∂qi V, i = 1, 2, (5.35)

with the Hamiltonian

1 2
2
H (q, p) = p + V (q1 , q2 ),
2 i=1 i

where (q, p) ∈ C4 , V is a homogeneous polynomial of degree k (called the homoge-

neous potential of degree k), or an inverse of a homogeneous polynomial of degree 
(called the homogeneous potential of degree −). Their polynomial integrability has
5.3 Hamiltonian Systems: Integrability via the Differential Galois Group 231

a long history. All systems (5.35) having a polynomial potential of degree ≤ 5 and a
second polynomial first integral of degree ≤ 4 were characterized in the 1980s, see
e.g. [253] and the references therein.
As we know, the polynomial Hamiltonian systems (5.35) whose complete in-
tegrability has been characterized are very few. Firstly, all systems (5.35) with a
homogeneous potential of degree −1, 0, 1, 2 are completely integrable, see [253,
Theorem 2]. The next theorem summarizes these results.
Theorem 5.15 Systems (5.35) having a homogeneous potential of degree −1, 0, 1, 2
are completely integrable. Moreover, the potentials and the first integrals are the
following:
• the potential V = aq12 + bq1 q2 + cq22 and the first integral I = b2 q12 + 4bcq1 q2 +
(b2 + 4c2 − 4ac)q22 − 2(a − c) p22 + 2bp1 p2 ,
• the potential V = aq1 + bq2 and the first integral I = ap2 − bp1 ,
• the potential V = a and the first integral I = p1 ,
• the potential V = 1/(aq1 + bq2 ) and the first integral I = ap2 − bp1 .
For systems (5.35) with a homogeneous potential of degree −2, −3, 3, their inte-
grability was studied by Llibre et al. in [253, 254, 289] respectively, who transformed
the homogeneous potentials to the equivalent simple ones. To state their results, for
convenience we recall this type of transformation.
Let
PO2 (C) := {A ∈ M (2, C)| A Aτ = α E},

where α ∈ C \ {0}, and M (2, C) is the set of matrices of order 2 with entries in
C. It is easy to check that PO2 (C) forms a group under the multiplication of matri-
ces. Two potentials V1 (q1 , q2 ) and V2 (q1 , q2 ) are equivalent if there exists a matrix
M ∈PO2 (C) such that V1 (q1 , q2 ) = V2 (M(q1 , q2 )τ ). As mentioned in [253, 254],
if V1 and V2 are two equivalent potentials, then system (5.35) with the potential V1
is integrable if and only if system (5.35) with the potential V2 is integrable. For a
proof, see [203]. So, in order to study the integrability of systems (5.35), one only
needs to study them with their equivalent potentials.
For systems (5.35) with a homogeneous potential of degree −2

1
V (q) = , a, b, c ∈ C not all zero, (5.36)
aq12 + bq22 + bq32

Llibre et al. [253] characterized their polynomial and analytic integrability.

Theorem 5.16 For systems (5.35) with a potential (5.36), the following statements
hold.
(a) The potential V in (5.36) is equivalent to the potential V ∗ = 1/(aq12 + cq22 ).
(b) System (5.35) with the potential V ∗ is integrable with a second polynomial first
integral if and only if either c = 0, or c = 0 and a ∈ {0, c}. Moreover, in the
former the first integral is p2 ; and in the latter the first integral is p1 if a = 0,
or q1 p2 − q2 p1 if a = c.
232 5 Algebraic, Analytic and Meromorphic Integrability

(c) System (5.35) with the potential V ∗ is integrable with a second analytic first
integral if and only if either c = 0, or c = 0 and a ∈ {0, c}.
The proofs of this last theorem mainly depend on the change of variables
(q1 , q2 , p1 , p2 ) → (q1 , q2 , p1 , T ) with T = q2 p1 − q1 p2 , and a recursive calcula-
tion of a first integral of the form


n
j
F(q1 , q2 , p1 , T ) = f j (q2 , p1 , T )q1 ,
j=−n

of the new system in these coordinates, which are obtained from system (5.35) and
its possible polynomial first integrals via the transformation.
For systems (5.35) with a homogeneous potential of degree −3

1
V (q) = , a, b, c, d ∈ C not all zero, (5.37)
aq13 + bq12 q2 + cq1 q22 + dq32

Llibre et al. [254] characterized their polynomial integrability.

Theorem 5.17 For system (5.35) with a potential (5.37), the following statements
hold.
(a) After a rotation and a rescaling (if necessary) the potential V in (5.37) is reduced
to one of the following forms:

1 1 1
V0 = , V1 = , V2,3 = 2 √ ,
(aq12+ cq2 )
2
aq1 + q2
3 3
(q1 + q2 )(q2 ± −1q1 )
2

1 1
V4,5 = √ , Vgen = .
(q2 ± −1q1 ) 3 aq1 + q1 q2 + dq23
3 2

(b) System (5.35) with the potential Vgen does not admit a polynomial first integral.
(c) System (5.35) with the potential V0 is integrable with the polynomial first integral
p2 .
(d) System (5.35) with the potential V1 admits a polynomial first integral if and only
if a = 0, in which the first integral is p1 .
(e) System (5.35) with the potential V2,3 does not admit a polynomial first integral.
√ with the potential V4,5 is integrable with the polynomial first inte-
(f) System (5.35)
gral p1 ∓ −1 p2 .
In [254] the authors posed the conjecture:
System (5.35) with the potential

1
V (q1 , q2 ) = , a j ∈ C,
a0 q1m + a1 q1m−1 q2 + . . . + am−2 q1 q2m−2 + am q2m
(5.38)
with m ≥ 2 and a1 . . . am−2 am = 0, does not admit a polynomial first integral.
5.3 Hamiltonian Systems: Integrability via the Differential Galois Group 233

This conjecture was verified only for m = 2, 3 in [253, 254].

In [254] the authors also posed the next conjecture:
System (5.35) with the potential (5.38) with m ≥ 1 is integrable with a second
polynomial first integral if and only if the potential (5.38) is equivalent to one of
the following potentials

V = f (q1 ) + g(q2 ), V = f (aq1 + bq2 ), V = f (q12 + q22 ),

where f and g are suitable functions.

This conjecture was verified only for m = 1, 2, 3 in [253, 254].
The main tool in the proofs of Theorem 5.17 is the Morales–Ramis theory, i.e.
Theorem 5.13.
Using the Morales–Ramis theory, Maciejewski and Przybylska [318] in 2004 also
provided the necessary and sufficient conditions for system (5.35) with a homoge-
neous potential of degree 3 to be meromorphically integrable.

Theorem 5.18 System (5.35) with a homogeneous potential V of degree 3 has a

second meromorphic first integral if and only if the potential V is equivalent to one
to the following:
√ √
Vi = (q2 − −1q1 )i (q2 + −1q1 )3−i , i = 0, 1, . . . , 3,
V4 = αq13 ,
q13 c
V5 = + q23 ,
3 3
q12 q2
V6 = + q23 ,
2
q 2 q2 8
V7 = 1 + q23 ,
2√ 3
−3 3 1 2
V8 = ± q + q q2 + q23 .
18 1 2 1
Maciejewski and Przybylska [319] in 2005 and Llibre et al. [252] in 2011 com-
pleted the meromorphic integrability for system (5.35) with a homogeneous potential
of degree 4.
Theorem 5.19 System (5.35) with a homogeneous potential V of degree 4 has a
second meromorphic first integral if and only if the potential V is equivalent to one
to the following:
√ √
Vi = α(q2 − −1q1 )i (q2 + −1q1 )4−i , i = 0, 1, . . . , 4,
V5 = αq24 ,
α 4
V6 = q + q24 ,
4 1
234 5 Algebraic, Analytic and Meromorphic Integrability

1
V7 = 4q14 + 3q12 q22 + q24 ,
4
3 2 2 1 4
V8 = 2q1 + q1 q2 + q2 ,
4
2 4
1 2
V9 = (q1 + q2 ) ,
2 2
4
√ 1
V10 = −q12 (q1 + −1q2 )2 + (q12 + q22 )2 .
4
In addition to the results in [318], Llibre and Valls [289] in 2014 studied the
Darboux integrability of system (5.35) with a homogeneous potential of degree 3,
and obtained the following results.

Theorem 5.20 System (5.35) with a homogeneous potential V of degree 3 has a

second Darboux first integral if and only if the potential V is equivalent to one of
the following:
√ √
Vi = α(q2 − −1q1 )i (q2 + −1q1 )3−i , i = 0, 1, . . . , 3,
V4 = αq13 ,
q13 c
V5 = + q23 ,
3 3
q12 q2
V6 = + q23 ,
2
q 2 q2 8
V7 = 1 + q23 ,
2√ 3
−3 1
V8± = ± q13 + q12 q2 + q23 .
18 2
Moreover, the corresponding first integrals are respectively
√
V0 H0 = p1 − −1 p2 ,
√ √ √ √
V1 H1 = 9 −1 p12 + 6 p1 p2 + 3 −1 p22 − 16αq13 + 24 −1αq12 q2 + 8 −1αq23 ,
√ √ √ √
V2 H2 = −9 −1 p12 + 6 p1 p2 − 3 −1 p22 − 16αq13 − 24 −1αq12 q2 − 8 −1αq23 ,
√
V3 H3 = p1 + −1 p2 ,
V4 H4 = p2 ,
V5 H5 = 3 p12 + 2q13 ,
V6 H6 = −8 p1 p2 q1 + 8 p12 p2 − q12 (q12 + 4q22 ),
V7 H7 = 72 p14 − 36 p1 p2 q13 − 3q16 + 2(3 p22 + 16q23 )(3 p22 + 6q12 q2 + 16q23 )
+ 12 p12 (3 p22 + 12q12 q2 + 16q23 ),
√ √
V8± H8± = −q16 ± 6 −3q15 q2 + 27q14 q22 ± 6 −3q13 ( p12 + p22 + 2q23 )
+ 54q12 q2 ( p12 + p22 + 2q23 ) + 27( p12 + p22 + 2q23 )2 .
5.3 Hamiltonian Systems: Integrability via the Differential Galois Group 235

5.3.1.3 Other Necessary Conditions for the Integrability

of Hamiltonian Systems (5.35)

In this part we summarize some other necessary conditions for the Hamiltonian
system (5.33) to be integrable in the Liouvillian sense.
Theorem 5.13 gave some necessary conditions for the Hamiltonian system (5.33)
to be Liouvillian integrable when the homogeneous potentials have degree nonzero.
In the case when the homogeneous potential is of degree zero, Casale et al. [60] in
2010 obtained the following results.
Theorem 5.21 Assume that the potential V ∈ C(q) of system (5.33) is homoge-
neous of degree k = 0, and that it has a Darboux point d ∈ Cm . If system (5.33) is
Liouvillian integrable with rational first integrals, then
(a) all eigenvalues of the Hessian matrix of V (q) at q = d are integers,
(b) the Hessian matrix ∂q2 V (d) is diagonalizable.
We remark that a homogeneous potential V of degree zero could be nonconstant.
For instance, V (q1 , q2 ) = q1 q2−1 − 2q1−2 q22 is of degree zero. In this situation, system
(5.33) has the particular solution
 
Γρ := (q, p) ∈ C2m | (q, p) = (μd, νd) ,

where d ∈ Cm is a Darboux point of V (q), and (μ, ν) satisfies

 
1
γρ := (μ, ν) ∈ C2 | ν 2 + ln μ = ρ , ρ ∈ C,
2

which is a solution of the two-dimensional system

μ̇ = ν, ν̇ = −μ−1 .

The proof of this last theorem strongly depends on an analysis of the variational
equation along Γk,ρ
1
ẋ = y, ẏ = − 2 ∂q2 V (d)x.
μ

Another necessary condition for integrability was established in [135] by Duval

and Maciejewski via the diagonolizability of the Hessian matrix ∂q2 V (q) at a Darboux
point.
Theorem 5.22 Let V (q) be a homogeneous potential of degree k ∈ Z \ {−2, 0, 2},
and suppose it has a Darboux point. Assume that the Hamiltonian system (5.33) is
Liouvillian integrable with meromorphic first integrals in a connected neighborhood
of Γk,ρ , ρ ∈ C. Then the following statements hold.
(a) The Hessian matrix ∂q2 V (d) does not have an elementary Jordan block of size
 ≥ 3.
236 5 Algebraic, Analytic and Meromorphic Integrability

(b) If ∂q2 V (d) admits an elementary Jordan block of size m = 2 with a correspond-
ing eigenvalue λ, then (k, λ) satisfies one of the conditions in Table 5.1 except
(k, λ) = (k, p + k2 p( p − 1), (±2, arbitrary).

In this last theorem, an elementary Jordan block of size  with an eigenvalue λ is

a matrix of the form ⎛ ⎞
λ 0 0 ··· 0 0
⎜1 λ 0 ··· 0 0⎟
⎜ ⎟
⎜0 1 λ ··· 0 0⎟
⎜ ⎟
⎜ .. .. . . . . .. .. ⎟ ∈ M (C).
⎜. . . . . . ⎟
⎜ ⎟
⎝0 0 0 ··· λ 0⎠
0 0 0 ··· 1 λ

Applying the arguments used to prove Theorems 5.13, 5.21 and 5.22, one can get
some similar results:
• Maciejewski and Przybylska [321] in 2016 obtained some necessary conditions
on Liouvillian integrability with meromorphic first integrals of the Hamiltonian
system (5.33) with an algebraic potential. Recall that an algebraic potential is an
algebraic function of q, which is a root of a minimal polynomial with coefficients
in C(q).
• Szumiński et al. [418] in 2015 presented some necessary conditions on the
Liouvillian integrability of the two degrees of freedom of Hamiltonian system
(5.35) with a special potential of the form H = T + V with
 
1 pϕ2
T = r m−k pr2 + , V = r m U (ϕ),
2 r2

where m, k are integers and k = 0, and U is a complex meromorphic function of

ϕ ∈ C. Here (r, ϕ) are treated as the polar coordinates associated to q = (q1 , q2 ),
and ( pr , qϕ ) are related to p = ( p1 , p2 ).

5.3.2 Hamiltonian Systems: Integrability via Other

Extensions

5.3.2.1 Integrability via Higher-Order Variational Equations

The above results are all related to the analysis of the first-order variational equations
of system (5.33) along a particular solution in the framework of differential Galois
theory. The approach using the higher-order variational equations can be found in
the work of Baider et al. [22]. Some recent approaches via higher-order variational
equations include:
5.3 Hamiltonian Systems: Integrability via the Differential Galois Group 237

• Morales-Ruiz et al. [337] in 2007 showed that if a meromorphic Hamiltonian sys-

tem is integrable in the Liouville sense with meromorphic first integrals defined in a
connected neighbourhood U of a special phase curve Γ and functionally indepen-
dent in U \ Γ , then for any p ∈ N, the identity component (G p )0 of the differential
Galois group G p of the pth order variational equation of the Hamiltonian system
along Γ is Abelian.
• Duval and Maciejewski [136] in 2014 pursued their study of the integrability of
system (5.33) with a homogeneous potential of degree k ∈ Z \ {−1, 0, 1} via an
analysis of the differential Galois group of the higher-order variational equations
of system (5.33) along a particular solution. They obtained the following results.
– In [136] Duval and Maciejewski proved the solvability of the higher-order varia-
tional equations provided that the first-order variational equation has a virtually
Abelian differential Galois group.
– Using this result together with other hypotheses, Duval and Maciejewski [136,
137] obtained some necessary conditions in order for system (5.33) with homo-
geneous potentials of degree k ∈ Z \ {−1, 0, 1} to be integrable in the Liouvil-
lian sense.
The notions mentioned above can be found, for example, in van der Put and Singer
[368], or Morales-Ruiz [334], where there are more detailed descriptions.

5.3.2.2 Part Integrability of Hamiltonian Systems

The works of Maciejewski et al. [322] show that the methods of Morales–Ramis
theory are also valid for part integrability, which are stated in the following two
theorems. These two results present necessary conditions for the existence of just
one additional first integral and of r ∈ {2, . . . , m − 1} functionally independent first
integrals of system (5.33), respectively. Here part integrability means that the number
of functionally independent first integrals is less than that required in the sense of
Liouvillian integrability.

Theorem 5.23 Assume that the Hamiltonian system (5.33) has a homogeneous po-
tential V ∈ C(q) of degree k ∈ Z∗ , and that the potential V has a Darboux point
d ∈ Cm . If the Hessian matrix ∂q2 V (d) is diagonalizable and has the eigenvalues
λ1 , . . . , λm−1 , λm = k − 1, and system (5.33) admits a second meromorphic first in-
tegral in a connected neighbourhood U of a particular solution Γk,ρ , ρ = 0, then
either
• there exists an r ∈ {1, . . . , m − 1} such that the pair (k, λr ) is in the list of Table 5.1,
or
• there exist 1 ≤ i < j < m such that

1 1
(k − 2)2 + 8kλi = (k − 2)2 + 8kλ j + p,
2k 2k

for some p ∈ Z.
238 5 Algebraic, Analytic and Meromorphic Integrability

Theorem 5.24 Assume that the Hamiltonian system (5.33) has a homogeneous po-
tential V ∈ C(q) of degree k ∈ Z∗ , and that the potential V has a Darboux point
d ∈ Cm . If the Hessian matrix ∂q2 V (d) is diagonalizable and has the eigenvalues
λ1 , . . . , λm−1 , λm = k − 1, and system (5.33) admits an additional  ∈ {1, . . . , m −
1} functionally independent and commuting meromorphic first integrals in a con-
nected neighbourhood U of a particular solution Γk,ρ , ρ = 0, then
• at most  pairs (k, λi ), say (k, λm−i ), i = 0, 1, . . . ,  − 1, belong to the list in
Table 5.1.
Moreover, if system (5.33) has one more meromorphic first integral F+1 on U such
that F1 , . . . , F+1 pairwise commute and are functionally independent in U \ Γk,ρ ,
then either
• there exists an i ∈ {1, . . . , n − m} such that the pair (k, λi ) belongs to the list in
Table 5.1, or
• there exist 1 ≤ i < j < m −  such that

1 1
(k − 2)2 + 8kλi = (k − 2)2 + 8kλ j + p,
2k 2k

for some p ∈ Z.

On the characterization of the part integrability of the Hamiltonian system (5.33)

Maciejewski and Przybylska [320] also obtained some necessary conditions, which
are partially similar to those in these last two theorems.
B3 . Integrability of Hamiltonian Systems via Improper Darboux Points
The above results on the necessary condition of integrability depend on the Darboux
points of the potentials. In the definition of a Darboux point d, if ∂q V (d) = 0, i.e.
γ = 0, then d is called an improper Darboux point.
Przybylska [366] and Studziński and Przybylska [416] studied the integrability
of system (5.33) with polynomial and rational potentials, respectively. In the latter,
the results are more general. We state them here.

Theorem 5.25 Assume that the Hamiltonian system (5.33) has a homogeneous po-
tential V (q) which admits an improper Darboux point d. If V ∈ C(q) is of degree
k ∈ Z \ {−2, 0, 2}, and system (5.33) is Liouvillian integrable with rational first in-
tegrals, then the Hessian matrix ∂ 2 q V (d) is nilpotent, i.e., all its eigenvalues vanish.

On the integrability of the natural Hamiltonian systems having a special potential,

there are also many other results, see e.g. [346, 431, 455, 456] and the references
therein.
On the results applying differential Galois theory to study the integrability of
some concrete systems, see e.g. [396] for fourth Painlevé equations, [239] for a
three-particle Toda lattice Hamiltonian system with two degrees of freedom, [397]
for a generalized Yang–Mills Hamiltonian system with two degree of freedom, and
[238] for the Hénon–Heiles system, and the references therein.
5.3 Hamiltonian Systems: Integrability via the Differential Galois Group 239

5.3.3 Extension of the Morales–Ramis Theory to General

Systems

Here we introduce the results obtained by Li and Shi [240] in 2012, in which they
extended the Morales–Ramis theory from Hamiltonian systems with a homogeneous
potential to general systems.
Consider the complex holomorphic differential system

ẋ = f (x), x ∈ M, (5.39)

where the dot denotes the derivative with respect to the complex time t, and M is
a connected n-dimensional complex analytic manifold. Let ϕ(t), t ∈ V ⊂ C, be a
nontrivial solution of Eq. (5.39), i.e. it is not an equilibrium. Then the phase curve
Γ = {ϕ(t)| t ∈ V } is a Riemannian surface.
It is known that the variational equation of system (5.39) along the solution Γ is

ẏ = B(t)y, B(t) = ∂x f (ϕ(t)). (5.40)

Note that y in (5.40) belongs to the tangent space TΓ M of M along Γ , and TΓ M

is a sub-vector bundle of the tangent bundle T M of M .
Let N := TΓ M /T Γ be the normal bundle, and let π : TΓ M → N be the natural
projection. Reducing the variational Eq. (5.40) to the normal bundle N gives

ż = π∗ (B(t)π −1 z), z ∈ N, (5.41)

where π ∗ is the tangent map induced by π . The linear differential system (5.41)
is called the normal variational equation of system (5.39) along the Riemannian
surface Γ .
Denote by G and GN respectively the Galois groups of the linear differential
equations (5.40) and (5.41), see [334, 368]. These groups may be identified as al-
gebraic subgroups of the general linear group G L(C, n). We denote by G 0 and GN0
respectively the identity components of G and GN .

Theorem 5.26 For the complex holomorphic differential system (5.39), the follow-
ing statements hold.
(a) If system (5.39) has m, m ∈ {1, . . . , n − 1}, functionally independent first inte-
grals holomorphic or meromorphic in a neighborhood of Γ , then the normal
variational system (5.41) has m functionally independent homogeneous polyno-
mial or rational first integrals.
(b) If system (5.39) has m ∈ {1, . . . , n − 1} functionally independent meromorphic
first integrals in a neighborhood of Γ , then
(b1 ) the Lie algebra gN of the differential Galois group GN of system (5.41) has
m meromorphic invariants,
240 5 Algebraic, Analytic and Meromorphic Integrability

(b2 ) the identity component GN0 of GN has k ≤ n − m − 1 generators, i.e.

 s 
GN0 := eI1 t1 eI2 t2 . . . eIk tk | (t1 , . . . , tk ) ∈ V k ⊂ Ck , s ∈ N ,

where {I1 , . . . , Ik } is a basis of gN with k ≤ n − m − 1.

Moreover,

(I1 ) if m = n − 1 then gN = {0} and GN0 = {1},

(I2 ) if m = n − 2 then gN and GN0 are both commutative,
(I3 ) if m = n − 3 then gN and GN0 are both solvable.

We remark that statement (a) can be found in [22, 471, 481]. Statement (b) is
[240, Theorem 7], whose proof is based on (a), and the expressions of the identity
component of a Lie group and its Lie algebra via the first integrals of systems (5.40)
and (5.41). For more details on Lie groups, their Lie algebras and their expressions,
see e.g. [351, 368].
The next results are consequences of Theorem 5.26, see Li and Shi [240, Corol-
lary 1].

Corollary 5.4 For system (5.39), the following statements hold.

(a) If the Lie algebra gN and the identity component GN0 of the differential Galois
group GN of system (5.41) are not trivial, then system (5.39) has at most n − 2
(in case n > 1) functionally independent meromorphic first integrals.
(b) If gN and GN0 are not commutative, then system (5.39) has at most n − 3 (in case
n > 2) functionally independent meromorphic first integrals.
(c) If gN and GN0 are not solvable, then system (5.39) has at most n − 4 (in case
n > 3) functionally independent meromorphic first integrals.

5.4 Calculations of Rational First Integrals and Darboux

Polynomials of Planar Polynomial Vector Fields

This section presents some faster computational methods for calculating Darboux
polynomials and rational first integrals with an emphasis on planar polynomial vector
fields. These results are due to Bostan et al. [40], and Ferragut and Giacomini [148].
Consider the planar polynomial differential system of degree m

ẋ = P(x, y), ẏ = Q(x, y), (x, y) ∈ C2 , (5.42)

whose associated vector field is denoted by

∂ ∂
X := P(x, y) + Q(x, y) .
∂x ∂y
5.4 Calculations of Rational First Integrals and Darboux Polynomials… 241

A rational function P/Q with P, Q ∈ C[x, y] is reduced if P and Q are coprime.

The degree of a reduced rational function F = P/Q is the maximum of degP and
degQ. As in [40, 148], the objective of this section is to give an efficient algorithm
to handle the problem
PN : Given a degree bound N ∈ N, either compute a rational first integral F ∈
C(x, y) of (5.42) of degree at most N, or prove that no such F exists.
To do so, we assume that N ≥ m + 1.
As we know, a reduced rational function H = f /g is a first integral of (5.42)
if and only if f and g are both Darboux polynomials with the same cofactor. The
next result, due to Singer [403, Lemma A.1], provides further information on two
invariant algebraic curves passing through a regular point.

Proposition 5.6 Assume that (x0 , y0 ) is a regular point of system (5.42). If f and g
are two Darboux polynomials of system (5.42) satisfying f (x0 , y0 ) = g(x0 , y0 ) = 0,
and f is irreducible, then f divides g.

Proof For the reader’s convenience we provide here a proof due to Singer [403].
Without loss of generality, we assume that P(x0 , y0 ) = 0. Let R(x) be the resultant
of f and g in terms of y. Recall that for two univariate monic polynomials ξ(z) and
η(z) over C, their resultant, denoted by res(ξ, η), is the product of the differences of
their roots in an algebraic closure of C, i.e.

(x − y),
(x,y): ξ(x)=η(y)=0

where in the case of multiple roots, the factors are repeated according to their mul-
tiplicities.
By the properties of resultants there exist A0 , B0 ∈ C[x, y] such that

R(x) = A0 f + B0 g.

Then R(x0 ) = 0. Acting on both sides of this last equation by X gives

P R  (x) = A1 f + B1 g, (5.43)

where A1 , B1 ∈ C[x, y], and we have used the facts that f and g are two Dar-
boux polynomials of X . This shows that R  (x0 ) = 0, because P(x0 , y0 ) = 0, and
f (x0 , y0 ) = g(x0 , y0 ) = 0.
Once again the vector field X acts on the two sides of (5.43), and we get

X (P)R  (x) + P R  (x) = A2 f + B2 g.

It follows clearly that R  (x0 ) = 0.

242 5 Algebraic, Analytic and Meromorphic Integrability

Continuing this calculation yields that all derivatives of R(x) vanish at x0 . This
implies that R(x) ≡ 0. Consequently, f and g have a common factor. Then the
irreducibility of f forces that f divides g. 

A. Composite Rational First Integrals
A rational function R(x, y) ∈ C(x, y) is composite if it can be written as

R = r ◦ W (x, y), with W ∈ C(x, y), r ∈ C(s) and deg r ≥ 2.

Otherwise, F is said to be non-composite.

The next result characterizes the structure of the set of rational first integrals of a
planar polynomial differential system (5.42), see [40, Theorem 10].
Theorem 5.27 If system (5.42) has a rational first integral, then

{R ∈ C(x, y)| X (R) = 0} = C(F),

for some non-composite reduced rational first integral F of the vector field X .
Proof The proof is given in [40]. Here we repeat it with a little more detail for the
reader’s convenience.
Define
D := {R ∈ C(x, y)| X (R) = 0}.

Then C  D  C(x, y). By Schinzel [384, Theorem 1, p. 12], which says that
for fields K ⊂ K ⊂ K[z] with z = (z 1 , . . . , z n ) and K characteristic zero, so that
K = K(g1 , . . . , gr ), where r ≤ 1 + tr.deg.K /K, it follows from tr.deg.D/C ≤
tr.deg.C(x, y)/C = 2 that
D = C( f 1 , f 2 , f 3 ),

with f 1 , f 2 , f 3 ∈ C(x, y).

Since f 1 , f 2 , f 3 are all first integrals of X , i.e. X ( f i ) = 0 for i = 1, 2, 3, and
we are in two-dimensional space, f 1 , f 2 , f 3 are forced to be pairwise functionally
dependent. They must be algebraically dependent. This shows that tr.deg.D/C = 1.
By Schinzel [384, Theorem 3, p. 15], which says that for fields K ⊂ K ⊂ K[z] with
z = (z 1 , . . . , z n ), if tr.deg.K /K = 1 then K = K(g) for g ∈ K(z), we have that
D = C(F).
We now prove that F is non-composite. If not, we should have F = r (G) with deg
r ≥ 2. Then it follows from Gutierrez et al. [194, Proposition 2.2] that C(F)  C(G).
Moreover, G is also a rational first integral of X because X (F) = r  (G)X (G) and
r  (G) ≡ 0. That is, G ∈ D. It follows that C(G) ⊂ D. Consequently, D = C(F) 
C(G) ⊂ D, a contradiction. So F is non-composite. This completes the proof of the
theorem. 

As an easy consequence of Theorem 5.27 a non-composite reduced rational first
integral of X has minimal degree inside the set D of the rational first integrals of
5.4 Calculations of Rational First Integrals and Darboux Polynomials… 243

X . This observation together with the spectrum of a rational function will play a
crucial role in constructing algorithms for computing rational first integrals.
B. The Spectrum of Rational Functions
For a reduced rational function R = A/B ∈ C(x, y) of degree N , the set

σ (A, B) = {λ ∈ C ∪ {∞}| A − λB is reducible in C[x, y], or deg(A − λB) < N }

is called the spectrum of R. Here by convention A − λB = B if λ = ∞. Note that

the spectrum is also called the set of remarkable values in the case when A − λB is
reducible in C[x, y], see e.g. Ferragut and Llibre [149].
By Bodin [38, Theorem 2.2, Corollary 2.3], the spectrum σ (A, B) is finite if
and only if A/B is non-composite. On the cardinality of the spectrum, denoted
by |σ (A, B)|, there are lots of results (see [88] and the references therein), which
can be traced back to Poincaré [362]. Chéze [88, Theorem 2, Corollary 3] improved
the previous results and obtained the following.

Theorem 5.28 If A/B ∈ C(x, y) is a reduced non-composite rational first integral

of the vector field X of degree m, then

m(m + 1)
|σ (A, B)| ≤ B(m) + 5, where B(m) = .
2
As consequences of Theorem 5.28 one has the following results, which will be
used in the construction of algorithms for computing rational first integrals.

Corollary 5.5 Assume that A/B ∈ C(x, y) is a reduced non-composite rational first
integral of degree N for the vector field X of degree m. The following statements
hold.
(a) For all but B(m) + 5 constants λ ∈ C ∪ {∞}, the polynomial A − λB has
degree N and it is an irreducible Darboux polynomial of X with the same
cofactor as that of A and B.
(b) If X has a rational first integral, then there exist infinitely many irreducible
Darboux polynomials which all have the same degree and the same cofactor.
(c) The number of values of c ∈ C for which A(0, c)/B(0, c) ∈ σ (A, B) is bounded
by N (B(m) + 5).

Proof (a) This is a direct consequence of Theorem 5.28.

(b) This follows from Theorem 5.28 and the facts that, except for finitely many values
in C, for all other λ ∈ C, the polynomials A − λB are irreducible and have degree
equal to max{deg A, degB}, and that A and B have the same cofactor.
(c) From Theorem 5.28, the number of elements A(0, c)/B(0, c) ∈ σ (A, B) with
c ∈ C is at most B(m) + 5. For each λ ∈ σ (A, B), A(0, c) − λB(0, c) = 0 has at
most N roots. These two facts verify the statement.
This completes the proof of the corollary. 

244 5 Algebraic, Analytic and Meromorphic Integrability

C. Algebraic Power Series via Rational First Integrals

Much literature has been generated on the subject of computing Darboux polynomials
and rational first integrals, some of which were mentioned in the previous sections.
As we have seen, the main difficulties are computational, see also Chéze [86]. In
2010 Ferragut and Giacomini [148] posed a new approach to computing rational
first integrals. Here, following the steps of Bostan et al. [40] in constructing the
computational algorithms, we introduce algebraic power series.
A formal power series y(x) ∈ C[[x]] is algebraic if there exists a non-zero poly-
nomial W ∈ C[x, y] such that W (x, y(x)) = 0. In this case we call y(x) algebraic
over C(x). If W has the lowest degree inside the set of polynomials M ∈ C[x, y]
satisfying M(x, y(x)) = 0, it is called a minimal polynomial of y(x) in C[x, y].
Associated to system (5.42), we consider the differential equation

dy Q(x, y)
= . (5.44)
dx P(x, y)

Here without loss of generality we assume that P(0, y) ≡ 0. Otherwise we take a

translation in x, which will achieve the desired result. Then for any c ∈ C such that
P(0, c) = 0, Eq. (5.44) has a unique solution y = yc (x) satisfying the initial condi-
tion yc (0) = c. Since the equation is rational and hence analytic in a neighborhood
of (0, c), the solution yc (x) can be expanded as a convergent power series.
The next result, due to Bostan et al. [40, Proposition 16], will play a fundamental
important role in the construction of computational algorithms.
Proposition 5.7 For the polynomial system (5.42) satisfying P(0, y) ≡ 0, and c ∈ C
such that P(0, c) = 0, let yc (x) ∈ C[[x]] be the unique solution of (5.44) satisfying
the initial condition yc (0) = c. The following statements hold.
(a) If system (5.42) admits a non-composite rational first integral A/B, then yc (x) is
algebraic. More precisely, yc (x) is a root of the non-zero polynomial λA − μB,
where λ = Q(0, c) and μ = P(0, c).
(b) If A/B is a reduced non-composite rational first integral of degree at most N for
system (5.42) of degree m, then for all but at most N (B(m) + 5) values of c ∈ C,
the polynomial λA − μB is a minimal polynomial of yc (x), where λ = B(0, c)
and μ = A(0, c).
Proposition 5.7 asserts that if system (5.42) has a rational first integral, then all
solutions of Eq. (5.44) are algebraic. The next result, due to [148] and [40, Proposi-
tion 17], verifies its converse.
Proposition 5.8 For system (5.42) and its associated vector field X and Eq. (5.44),
the following statements hold.
(a) If f ∈ C[x, y] is an irreducible Darboux polynomial of X , then all roots y(x) ∈
C(x) of f (x, y) = 0 such that P(0, y(0)) = 0 are solutions of Eq. (5.44).
(b) The minimal polynomial of an algebraic solution y(x) ∈ C[[x]] of Eq. (5.44)
such that P(0, y(0)) = 0 is a Darboux polynomial of X .
5.4 Calculations of Rational First Integrals and Darboux Polynomials… 245

(c) System (5.42) admits a rational first integral if and only if all the solutions of
Eq. (5.44) are algebraic.
These two propositions can be proved using the previous results together with
the fact that for each Darboux polynomial f , the variety {(x, y)| f (x, y) = 0} is
invariant under the flow of the vector field X . The details are omitted.
Propositions 5.7 and 5.8 provide information on the search for rational first inte-
grals A/B via minimal polynomials of the form λA − μB of a solution of (5.44).
As we mentioned above, the solution yc (x) of (5.44) with the initial point (0, c)
such that P(0, c) = 0 can be expanded as a Taylor series. In practice, it is difficult
to find an explicit expression for this power series solution yc (x). One only uses a
truncation of yc (x). The next result, due to Bostan et al. [40, Proposition 18] (a small
improvement of [15, Lemma 2.4]), shows that computing yc (x) mod x N +1 is enough
2

to search for rational first integrals of degree no greater than N ∈ N.

Proposition 5.9 Let y(x) ∈ C[[x]] be an algebraic power series whose minimal
polynomial W ∈ C[x, y] has degree at most N . The following statements hold.
(a) If V ∈ C[x, y] is a polynomial of degree at most N such that
2
+1
V (x, y(x)) ≡ 0 mod x N , (5.45)

then V (x, y(x)) ≡ 0.

(b) If V has minimal degree in y among the polynomials satisfying (5.45), then
V = f W for some f ∈ C[x].
Proof Let R(x) be the resultant of W and V with respect to y. Then there exist
K , L ∈ C[x, y] such that

R(x) = K (x, y)W (x, y) + L(x, y)V (x, y).

By the condition (5.45) it follows that R(x) ≡ 0 mod x N +1 . In addition, we get from
2

the expression of the resultant via the determinant of the Sylvester matrix associated
to W and V that
degR ≤ degW degV ≤ N 2 .

This implies that R ≡ 0. Hence W and V have a non-trivial common factor. The fact
that W is irreducible verifies that W divides V . That is, there exists an f ∈ C[x, y]
such that V = f W . Hence, V (x, y(x)) = f (x, y(x))W (x, y(x)) ≡ 0. Now V has
minimal degree in y among the set of polynomials satisfying (5.45), which forces
f ∈ C[x]. This proves the proposition. 

A polynomial V (x, y) ∈ C[x, y] is a minimal solution of Eq. (5.45) if it is a solu-
tion of Eq. (5.45) with minimal degree in y among the set of polynomials satisfying
(5.45).
Proposition 5.9 suggests an ansatz concerning the minimal polynomial of a solu-
tion of (5.44) using the truncation of this solution.
246 5 Algebraic, Analytic and Meromorphic Integrability

D. Algorithms for Computing Rational First Integrals

In this subsection we recall the algorithms for computing rational first integrals posed
in [40].
Algorithm I GenericRationalFirstIntegral, see [40, Sect. 3.3].
Input: P, Q ∈ C[x, y] with P(0, y) = 0 and a bound N ∈ N.
Output: “A non-composite rational first integral of (5.42) of degree at most N ”, or
“None”.

(I1 ) For an indeterminate c, compute the polynomial yc ∈ C(c)[x] of degree at most

N 2 + 1 such that

dyc Q(x, yc (x))

mod x N +1 .
2
yc (0) = c and =
dx P(x, yc (x))

(I2 ) Compute all non-trivial polynomials V ∈ C(c)[x, y] of degree ≤ N such that

2
+1
V (c, x, yc (x)) ≡ 0 mod x N . (5.46)

If no such V exists, then Return None.

Else, among the solutions of (5.46), pick a minimal solution W ∈ C[c][x, y].
(I3 ) Let W0 be the primitive part of W relative to y. Note that W0 is W divided by
the common factors of the coefficients of W .
Set A(x, y) := W0 (0, x, y).
Pick any c1 ∈ C such that

W0 (c1 , x, y)
∈
/ C,
A(x, y)

and set B(x, y) := W0 (c1 , x, y).

(I4 ) If X (B/A) = 0, then Return B/A. Else, Return None.

In Algorithm I, the output “None” means that there is no rational first integral of
degree at most N . But there may exist a rational first integral of degree > N .
The reasonability of Algorithm I was proved in [40, Theorem 20]. This algorithm
is entirely based on linear operations. However, the computations working over C(c)
are not very efficient in practice. The authors in [40] posed an efficient alternative.
Algorithm II ProbabilisticRationalFirstIntegral, see [40, Sect. 4.1].
Input: P, Q ∈ C[x, y] such that P(0, y) ≡ 0, and P(0, ci ) = 0 for two different
c1 , c2 ∈ C, and a bound N ∈ N.
Output: “A non-composite rational first integral of degree at most N for system
(5.42)”, or “None”, or “I do not know”.

(I I1 ) For i = 1, 2 do:
5.4 Calculations of Rational First Integrals and Darboux Polynomials… 247

(a) Compute the polynomial yci ∈ C[x] of degree at most N 2 + 1 such that

dyci Q(x, yci (x))

mod x N +1 .
2
yci (0) = ci and =
dx P(x, yci (x))

(b) Compute all non-trivial polynomials Vi ∈ C[x, y] of degree ≤ N such that

2
+1
Vi (x, yci (x)) ≡ 0 mod x N . (5.47)

(c) If no such Vi exists, then Return None.

Else, among the solutions of (5.47), pick a minimal solution Wi ∈ C[x, y].
Let A and B be respectively the primitive parts of W1 and W2 relative to y.
(d) If i = 1, then while (A(0, c2 ) = 0 or P(0, c2 ) = 0), do c2 = c2 + 1.
(I I2 ) If X (A/B) = 0, then Return A/B. Else, Return [“I do not know”,[c2 ]].

In [40, Theorem 22] the authors proved that Algorithm II should terminate, and in
Algorithm II
• the output “None” means that there is no rational first integral of degree at most
N for system (5.42).
• the output “A/B” is a non-composite rational first integral of degree at most N
for system (5.42).
• if system (5.42) has a non-composite rational first integral K /L of degree at most
N , and L(0, ci )/K (0, ci ) ∈
/ σ (K , L), the algorithm returns such a first integral.
In the opposite case to the third one, i.e.

L(0, c1 )/K (0, c1 ) ∈ σ (L , K ) or L(0, c2 )/K (0, c2 ) ∈ σ (L , K ),

the algorithm may compute A and B, but A/B is not a rational first integral. The
algorithm will return “I do not know”.
In the case when system (5.42) does not have a rational first integral of degree
≤ N , the algorithm will behave as follows:
• If Eq. (5.47) has no non-trivial solution, then the algorithm returns “None”.
• If Eq. (5.47) has a non-trivial solution, then the algorithm returns “I do not know”.
The next algorithm shows that if system (5.42) has a rational first integral of degree
≤ N , then after running ProbabilisticRationalFirstIntegral at most N (B(m) + 5) +
1 times, we will get a non-composite rational first integral of degree ≤ N .
Algorithm III DeterministicRationalFirstIntegral, see [40, Sect. 4.2].
Input: P, Q ∈ C[x, y] such that P(0, y) ≡ 0, and a bound N ∈ N.
Output: “A non-composite rational first integral of degree ≤ N for system (5.42)”,
or “None”.

(I I I1 ) Let Ω := ∅
248 5 Algebraic, Analytic and Meromorphic Integrability

(I I I2 ) While |Ω| ≤ 2N (B(m) + 5) + 2 do

(a) Choose two random but different elements c1 , c2 ∈ C \ Ω such that

P(0, ci ) = 0, i = 1, 2.
(b) F := Pr obabilisticRational Fir st I ntegral(P, Q, (c1 , c2 ), N ).
(c) If F =“None”, then Return “None”.
(d) Else if F = [“I do not know”,[e2 ]], then Ω := Ω ∪ {c1 , e2 }, and go to Step
(I I I2 ).
(e) Else Return F.

(I I I3 ) Return “None”

In [40, Theorem 25] the authors proved that Algorithm III is correct in the sense
that it returns a rational first integral of degree ≤ N if and only if it exists, and that
it returns “None” if and only if there is no rational first integral of degree ≤ N .
The next algorithm computes the minimal polynomial in (5.47).
Algorithm IV GuessMinimalPolynomial, see [40, Sect. 5.3].
Input: A polynomial z(x) ∈ C[x] such that deg z ≤ N 2 + 1, with a bound N ∈ N.
Output: “A minimal solution of (5.47) with degree ≤ N ”, or “None”.

N N
−i
(I V1 ) Let V (x, y) = m i j x j y i be the bivariate polynomial that we are
i=0 j=0
searching for.
(I V2 ) Construct the linear system for the m i j ’s by

 N −i
N 
2
+1
L : V (x, z(x)) = m i j x j z(x)i ≡ 0 mod x N .
i=0 j=0

(I V3 ) If L does not have a non-trivial solution, then Return “None”.

(I V4 ) Else compute a row-echelon form of a basis of solution of L to find a solution
V (x, y) of minimal degree in y and Return it.

In [40] the authors estimate the total costs and the complexity of the algorithms.
They also present examples showing applications of these algorithms.
E. Algorithms for Computing Darboux Polynomials
In this subsection we recall the algorithms for computing Darboux polynomials posed
in [40].
Algorithm I IrreducibleDarbouxPolynomialsPartial, See [40, Sect. 7.1].
Input: P, Q ∈ C[x, y] such that P(0, y) ≡ 0, P(0, y), Q(0, y) coprime, and a bound
N ∈ N such that the vector field X has no rational first integral of degree ≤ N .
Output: The set of all irreducible Darboux polynomials F of X satisfying deg F ≤ N
and F(0, y) ∈/ C.
5.4 Calculations of Rational First Integrals and Darboux Polynomials… 249

(a) E := ∅.
(b) For c ∈ C such that P(0, c) = 0, compute the polynomial yc ∈ C(c)[x] of degree
≤ N 2 + 1 satisfying

dyc Q(x, yc )
mod x N +1 .
2
yc (0) = c and ≡
dx A(x, yc )

(c) For c ∈ C such that Q(0, c) = 0, compute the polynomial xc ∈ C(c)[y] of degree
≤ N 2 + 1 satisfying

d xc P(xc , y)
mod y N +1 .
2
xc (c) = 0 and ≡
dy Q(xc , y)

(d) Let
 N −i
N 
V (x, y) = mi j x j yi
i=0 j=0

be the Darboux polynomials that we are searching for.

(e) Construct the linear system L1 (c) for the m i j ’s by
2
+1
L1 (c) : V (x, yc (x)) ≡ 0 mod x N .

( f ) Construct the linear system L2 (c) for the m i j ’s by

2
+1
L2 (c) : V (xc (y), y) ≡ 0 mod y N .

(g) For k = 1, 2 do:

(g1 ) Clear the denominator in Lk (c).
(g2 ) Compute the Smith normal form of Lk (c). Let Pk (c) be the last invariant
factor of Lk (c).
(g3 ) Factorize Pk (c) over C:

k

Pk (c) = Pki (c).
i=1

(g4 ) For i from 1 to k do:

(g41 ) Set Cci := C[c]/Pki (c).
(g42 ) Compute a solution Vki of Lk (ci ) with minimal degree in y, and which
is primitive with respect to y.
(g43 ) If gcd(X (Vki ), Vki ) = Vki , then

E := E ∪ {[Vki (c, x, y), Pki (c)]}.

250 5 Algebraic, Analytic and Meromorphic Integrability

(h) Return E.

This last algorithm deals with only the Darboux polynomials F satisfying
F(0, y) ∈/ C. For the Darboux polynomials F satisfying F(0, y) ∈ C, the invari-
ant curve V (x, y) = 0 intersects the line at infinity. Bostan et al. [40] posed another
algorithm.
Set
     
1 y 1 y 1 y
P ∗ (y, z) = P , zm , Q ∗ (y, z) = Q , zm , F ∗ (y, z) = F , zm .
z z z z z z

Then
• if F(0, y) ∈ C \ {0}, then F ∗ (0, y) ∈
/ C;
• if F is a Darboux polynomial of X , then F ∗ is a Darboux polynomial of the vector
field
  ∂ ∂
X ∗ = −y P ∗ (y, z) + Q ∗ (y, z) − P ∗ (y, z)z .
∂y ∂z

Algorithm II IrreducibleDarbouxPolynomials, See [40, Sect. 7.1].

Input: P, Q ∈ C[x, y] such that P(0, y) ≡ 0, P(0, y) and Q(0, y) coprime, Q ∗ (0, z)
≡ 0, P ∗ (0, z) and Q ∗ (0, z) coprime, and a bound N ∈ N such that the vector field
X has no rational first integral of degree ≤ N .
Output: The set of all irreducible Darboux polynomials F of X satisfying deg F ≤
N.

(a) E := IrreducibleDarbouxPolynomialsPartial (P, Q, N ).

(b) E∗ := IrreducibleDarbouxPolynomialsPartial (−y P ∗ + Q ∗ , −P ∗ z, N ).
(c) For all [F ∗ (c, y, z), P(c)] ∈ E∗ do:
 
y 1
∗
(c1 ) F(c, x, y) := F c, , x deg F .
x x
(c2 ) Add [F(c, x, y), P(c)] to E.
(d) Return E.

Note as mentioned in [40] that ‘P(0, y) ≡ 0, P(0, y) and Q(0, y) coprime,

Q ∗ (0, z) ≡ 0, P ∗ (0, z) and Q ∗ (0, z) coprime’ in Algorithm I is not a restriction.
It can be satisfied by a number of shifts.
The next algorithm avoids distinguishing whether F(0, y) ∈ C or not.
Algorithm III ProbabilisticIrreducibleDarbouxPolynomials, See [40, Sect. 7.1].
Input: P, Q ∈ C[x, y], a bound N ∈ N such that the vector field X has no rational
first integral of degree ≤ N , and two elements x0 , α ∈ C.
Output: The set of all irreducible Darboux polynomials F of X satisfying deg F ≤
N.
5.4 Calculations of Rational First Integrals and Darboux Polynomials… 251

(a) E := ∅.
(b) Set

Pα (x, y) = P(x + αy, y) − α Q(x + αy, y),

Q α (x, y) = Q(x + αy, y),
∂ ∂
Xα = Pα (x, y) + Q α (x, y) .
∂x ∂y

(c) For c ∈ C, compute the polynomial yc ∈ C(c)[x] of degree ≤ N 2 + 1 satisfying

dyc Q α (x, yc )
mod y N +1 .
2
yc (x0 ) = c and ≡
dx Pα (x, yc )

(d) Let
 N −i
N 
V (x, y) = mi j x j yi
i=0 j=0

be the Darboux polynomials that we are searching for.

(e) Construct the linear system L(c) for the m i j ’s by
2
+1
L(c) : V (x, yc (c, x)) ≡ 0 mod x N .

( f ) Clear the denominator in L(c).

(g) Compute the Smith normal form of L(c). Let P(c) be the last invariant factor
of L(c).
(h) Factorize P(c) over C:


P(c) = Pi (c).
i=1

For i from 1 to  do:

(h 1 ) Set Cci := C[c]/Pi (c).
(h 2 ) Compute a solution Vi of L(ci ) with minimal degree in y, and which is
primitive with respect to y.
(h 3 ) If gcd(Xα (Vi ), Vi ) = Vi , then

E := E ∪ {[Vi (c, x − αy, y), Pi (c)]}.


k
(i) Factorize Pα (x, 0) over C: Pα (x, 0) = Pi (x).
i=1
( j) For i from 1 to k do:

If gcd(Xα (Pi ), Pi ) = Pi , then E := E ∪ {[Pi (x − αy, y), c − 1]}.

252 5 Algebraic, Analytic and Meromorphic Integrability

(k) Return E.

Bostan et al. [40, Proposition 33] provided the probability that Algorithm III suc-
cessfully obtains an irreducible Darboux polynomial.

Proposition 5.10 The Algorithm III: ProbabilisticIrreducibleDarbouxPolynomials

is correct. Moreover, if x0 and α are chosen uniformly at random in a finite set Ω ⊂ C
such that |Ω| > N m(B(m) + 1), then the probability that this algorithm returns all
irreducible Darboux polynomials is at least
  
N (B(m) + 1) N d(B(m) + 1)
1− 1− .
|Ω| |Ω|
Chapter 6
Applications of the Darboux Theory
of Integrability

The Darboux theory of integrability has been successively applied to study the
dynamics of polynomial differential systems. This chapter will focus on the center-
focus and the limit cycle problems of planar polynomial differential systems. In the
last section we will also present applications of this theory to specific models in two
and higher dimensions. Its applications in dynamical analysis will not be described
here, instead we refer to Cao and Zhang [57], Chen et al. [80], Gasull et al. [169],
Llibre et al. [260, 261], Zhang [470] and the reference therein.

6.1 The Center Problem via the Darboux Theory

of Integrability

For a real planar polynomial differential system we assume that the origin is a sin-
gularity and the linear part of the system at the origin has a pair of pure imaginary
eigenvalues, then after an invertible linear transformation and a rescaling of the time
or of the independent variable the system can be written as

ẋ = −y + p(x, y), ẏ = x + q(x, y), (6.1)

where p and q are real polynomials without constant and linear terms. As is well
known, if p and q are quadratic homogeneous polynomials, the center-focus problem
for system (6.1) has been solved, see e.g. Bautin [27], Li [228], Ye [451], and Zhang
et al. [480]. If p and q are higher-order polynomials, even though there are plenty of
results, in general the center-focus problem is still open (even for cubic differential
systems), see e.g. Dumortier et al. [134], Li [229], Roussarie [378], Romanovski and
Shafer [373] and the references therein.

© Springer Nature Singapore Pte Ltd. 2017 253

X. Zhang, Integrability of Dynamical Systems: Algebra and Analysis,
Developments in Mathematics 47, DOI 10.1007/978-981-10-4226-3_6
254 6 Applications of the Darboux Theory of Integrability

There are two main difficulties in solving this problem. One is the calculation of
the Lyapunov constants, and another is how to find analytic first integrals at the origin
provided that the Lyapunov constants vanish. Usually it is not possible to compute
all Lyapunov constants of a given system. We should try to compute the first N
Lyapunov constants for some N ∈ N, then we get the conditions that the coefficients
of system (6.1) satisfy for which the first N Lyapunov constants vanish, and prove
that under these conditions the system has an analytic first integral in a neighborhood
of the origin. If this is the case, the system has a center at the origin by the Poincaré
center theorem. Therefore, the search for first integrals plays an extremely important
role in the study of the center-focus problem.
The past studies on the center-focus problem showed that the Darboux theory of
integrability is a useful tool in finding first integrals of planar polynomial differential
systems. See e.g. Cairó et al. [51], Chavarriga and Giné [71], Han et al. [196],
Rousseau and Schlomiuk [379], Schlomiuk [385] and the references therein.
Theorem 6.1 If p(x, y) and q(x, y) are either quadratic homogeneous polynomials,
or cubic homogeneous polynomials, or of the form

p = p2 + x f, q = q2 + y f,

where p2 , q2 and f are quadratic homogeneous polynomials, then system (6.1) has
a center at the origin if and only if it has a Darboux first integral.
We will not prove this theorem. When p and q are quadratic homogeneous poly-
nomials there is a stronger theorem called the Kapteyn–Bautin theorem, which we
prove below. For other cases we will present some related remarks.
A. If p and q are quadratic homogeneous polynomials, system (6.1) is the general
form of quadratic differential systems which have the origin as a weak focus or
a center, which is called a nondegenerate monodromy singularity. Through an
affine change of coordinates, system (6.1) can be written as (see Kapteyn [218,
219])
ẋ = −y − bx 2 − cx y − dy 2 , ẏ = x + ax 2 + Ax y − ay 2 . (6.2)

Kapteyn [218, 219] and Bautin [27] proved the next results:

Theorem 6.2 (Kapteyn–Bautin theorem) System (6.2) has a center at the origin if
and only if one of the following conditions holds:
(III) c = a = 0;
(II) b + d = 0;
(I) c + 2a = A + 3b + 5d = a 2 + bd + 2d 2 = 0;
(H) A − 2b = c + 2a = 0.

Under each of the above conditions Lunkevich and Sibirskii [310], see also
Schlomiuk [385, Theorem 3.1], obtained a Darboux first integral of system (6.2).
Under the conditions (III) and (II), the Darboux first integrals can be found in [385,
6.1 The Center Problem via the Darboux Theory of Integrability 255

Tables 1 and 2, p.807]. Under the condition (I) the system has a rational first inte-
gral given in [385, (4.24), p.819]. Whereas under the condition (H) the system is
Hamiltonian.
Besides the normal form (6.2) of quadratic differential systems having a non-
degenerate monodromy singularity at the origin, there appeared other normal forms
and the corresponding conditions (called center conditions) for which the singularity
is a center (in this case it is a nondegenerate center), see e.g. [228, 308, 451, 480].

B. If p and q are cubic homogeneous polynomials, Malkin [327] obtained the

necessary and sufficient conditions for system (6.1) to have a center at the origin.
To simplify the center conditions, Sibirsky [398] wrote the cubic differential
systems with a cubic homogeneous nonlinearity and a nondegenerate monodromy
singularity at the origin as

ẋ = −y − (ω + θ − a)x 3 − (η − 3μ)x 2 y − (3ω − 3θ + 2a − ξ )x y 2 − (μ − ν)y 3 ,

ẏ = x + (μ + ν)x 3 + (3ω + 3θ + 2a)x 2 y + (η − 3μ)x y 2 + (ω − θ − a)y 3 . (6.3)

A necessary and sufficient condition for system (6.3) to have a center at the origin
is one of the following three:
(H) ξ = a = 0;
(II) ξ = ν = θ = 0;
(III) ξ = ν = ω = η = 4(μ2 + θ 2 ) − a 2 = 0.
Lunkevich and Sibirsky [309] applied the algebraic invariants obtained by them-
selves in [399] to prove the integrability of the systems under four conditions.

Under the condition (H) system (6.3) is Hamiltonian. In 1995 Rousseau and
Schlomiuk [379] proved the integrability of system (6.3) under the conditions (II)
and (III) via the Darboux theory of integrability.
C. If p = p2 + x f, q = q2 + y f , system (6.1) is a projective quadratic system,
whose line at infinity is composed of singularities. The center problem at the
origin can also be characterized using the Darboux theory of integrability, see
Cairó et al. [51], and Chavarriga and Giné [71].

In the proof of the above results we need the Lyapunov constants, which can be
found in some books on the qualitative theory of ordinary differential equations (see
e.g. [134, 373]). For system (6.1), which has a nondegenerate monodromy singularity
at the origin, there always exists a function of the form
 
V (x, y) = x 2 + y 2 + O (x 2 + y 2 )2 ,

such that its derivative along system (6.1) is of the form

dV ∂V ∂V
= (−y + p(x, y) + (x + q(x, y)) = η4 (x 2 + y 2 )2 + η6 (x 2 + y 2 )3 + · · ·
dt ∂x ∂y
256 6 Applications of the Darboux Theory of Integrability

We call η2k , k ∈ N, the kth Lyapunov constant of the monodromy singularity. The
origin is a center if and only if all η2k = 0, k ∈ N\{1}. If there exists some η2m+2 = 0,
but η2+2 = 0,  < m, we call the origin a weak focus of order m.
Proof of the Kapteyn–Bautin theorem. Necessity. Some calculations show that the
first three Lyapunov constants of system (6.2) at the origin are

η4 = c2 (2a + c)(c + d),

η6 = c3 (A − 2b)(b + d)(A + 3b + 5d),
η8 = c4 (A − 2b)(b + d)2 (a 2 + bd + 2d 2 ),

where c2 , c3 , c4 are nonzero real constants. Hence the necessary conditions for the
origin of system (6.2) to be a center are η4 = η6 = η8 = 0. These are equivalent to
the conditions (I)–(III) and (H).
Sufficiency. Under the condition (H), system (6.2) is Hamiltonian with Hamiltonian
function
1 2  a d
H (x, y) = x + y 2 + x 3 + bx 2 y − ax y 2 + y 3 .
2 3 3
So the origin is a center.
Under the condition (III), system (6.2) can be simplified to

ẋ = −y − bx 2 − dy 2 , ẏ = x + Ax y.

It is invariant under the transformation (x, y, t) → (−x, y, −t). So the origin is a

center.
Under the condition (II), system (6.2) is invariant under rotations with respect
to the origin. If a = 0, we take a suitable rotation such that a = 0. Then c = 0,
otherwise we are in case (III).
Case 1. A = 0. System (6.2) has the Darboux polynomials and cofactors

f 1 = 1 + Ay, k1 = Ax,
f 2 = (1 − by)2 + c(1 − by)x − b(A + b)x 2 , k2 = −2bx − cy.

Since the divergence of system (6.2) is equal to k1 + k2 , by the Darboux integrability

theorem system (6.2) has the Darboux integrating factor f 1−1 f 2−2 , and consequently
it is Darboux integrable. Clearly this integrating factor is analytic at the origin. This
shows that system (6.2) has an analytic first integral in a neighborhood of the origin,
and so the origin is a center.
Case 2. A = 0. The divergence of system (6.2) is k2 , and so the system has the
integrating factor f 2−2 . Hence system (6.2) is Darboux integrable, and has an analytic
first integral at the origin. Consequently the origin is a center of system (6.2).
6.1 The Center Problem via the Darboux Theory of Integrability 257

Under the condition (I), if d = 0, we are in the case (III), system (6.2) has a center
at the origin. Assume that d = 0, then system (6.2) can be written as

a 2 + 2d 2 2 3a 2 + d 2
ẋ = −y + x + 2ax y − dy 2 , ẏ = x + ax 2 + x y − ay 2 .
d d
It has the Darboux polynomial and cofactor

  2(a 2 + d 2 )
f 1 = (a 2 + d 2 ) (dy − ax)2 + 2dy + d 2 , k1 = x.
d
The divergence of the system is 5k1 /2. This gives us the Darboux integrating factor
−5/2
f 1 . Obviously this integrating factor is analytic at the origin. Hence the system is
Darboux integrable with an analytic first integral at the origin, and consequently the
origin is a center.
This completes the proof of the theorem. 
In 1995, Cozma and Subă [119] provided a condition on the existence of Darboux
integrating factors or of Darboux first integrals via the vanishing Lyapunov constants
of system (6.1) at the origin.
Theorem 6.3 Assume that the origin is a weak focus of the polynomial differential
system (6.1) of degree m, and its first q Lyapunov constants vanish, 0 < q ≤
(m − 1)/2.
(a) If system (6.1) has at least m(m + 1)/2 − 1 − q invariant algebraic curves
not passing through the origin, and exponential factors, then the system has a
Darboux integrating factor.
(b) If system (6.1) has at least m(m + 1)/2 − q invariant algebraic curves not
passing through the origin, and exponential factors, then the system has a
Darboux first integral.
(c) In either case (a) or case (b), the origin is a center of system (6.1).
Christopher and Llibre [98, Theorem 3.2] in 1999 provided a different proof of
this last theorem.
In 2008, Christopher and Schlomiuk [110] provided an algebraic mechanism for
producing centers of the systems

ẋ = P3 (x)y, ẏ = P0 (x) + P1 (x)y + P2 (x)y 2 , (6.4)

with Pi ∈ R[x]. They showed that systems (6.4) with a nondegenerate center fall
naturally into two classes: those with a Darboux first integral, and those which arise
from simpler systems via singular algebraic transformations. To state their results
we need the next definitions.
A planar polynomial vector field (P(x, y), Q(x, y)) is algebraically reducible at a
singularity p if there exists a map (u, v) = Φ(x, y) := ( f (x, y), g(x, y)) with f and
g analytic functions (real or complex) in a neighborhood of p which are also algebraic
258 6 Applications of the Darboux Theory of Integrability

over C(x, y), such that the differential equation Pdy − Qd x = 0 is the pull-back of
a differential equation A(u, v)dv − B(u, v)du = 0 without singularities around the
image of p. If the map Φ(x, y) is rational, we call the vector field (P(x, y), Q(x, y))
rational reducible.
We now state the main results of Christopher and Schlomiuk [110]. The first one
concerns the system (6.4) with P3 (x) ≡ 1.
Theorem 6.4 System (6.4) with P3 (x) = 1, P0 (0) = 0 and P0 (0) < 0 has a center
at the origin if and only if one of the following conditions is satisfied.
• System (6.4) with P3 (x) = 1 is algebraically reducible via the map (u, v) = (x, y 2 )
and thus it has a symmetry in the x-axis.
• System (6.4) with P3 (x) = 1 is algebraically reducible via the map (u, v) =
(r (x), y) for some polynomial r (x) = x 2 + o(x 2 ) over R.
• There is a local first integral of Darboux type.
The second main result of [110] is for the general system (6.4).
Theorem 6.5 System (6.4) with P0 (0) = 0 and P3 (0)P0 (0) < 0 has a center at the
origin if and only if one of the following conditions is satisfied.
• System (6.4) is algebraically reducible via the map (u, v) = (x, y 2 ) and thus it
has a symmetry in the x-axis. 
• System (6.4)
 is algebraically reducible via the map (u, v) = (M(x))1/r , y
(R(x))1/q , where M(x) ∈ R(x) with (M(x))1/r of order 2 at x = 0 and R(0) = 0.
• There is a local first integral of Darboux type.

6.2 Algebraic Limit Cycles: Existence and Uniqueness

In this section we will use results on singular holomorphic foliations to study limit
cycles. Recall from Sect. 2.1 that for a projective holomorphic one-form

 
ω = P(X, 
Y, Z )d X + Q(X, 
Y, Z )dY + R(X, Y, Z )d Z ,

there exist homogeneous polynomials   N

L, M,  such that

= ZM
P −YN
,  = XN
Q  − Z
L, = Y
R 
L − X M.

Hence

=  ∂  ∂ +N
 ∂
X L +M
∂X ∂Y ∂Z
, see Cerveau and
is a holomorphic vector field in CP2 representing the foliation F
Lins Neto [61].
6.2 Algebraic Limit Cycles: Existence and Uniqueness 259

A holomorphic foliation F of degree m can be represented by a vector field in

affine coordinates as

ẋ = p(x, y) + xr (x, y), ẏ = q(x, y) + yr (x, y), (6.5)

where either r is a homogeneous polynomial of degree m and max{deg p, deg q} ≤

m; or r ≡ 0 and max{deg p, deg q} = m. The projective vector field and the projec-
tive differential one-form of system (6.5) are respectively

∂ ∂ ∂
X = P(X, Y, Z ) + Q(X, Y, Z ) − R(X, Y, Z ) , (6.6)
∂X ∂Y ∂Z
ω = (−Y R − Z Q)d X + (Z P + X R)dY + (X Q − Y P)d Z , (6.7)

where
   
X Y X Y
P(X, Y, Z ) = Z p m
, , Q(X, Y, Z ) = Z q
m
, ,
Z Z Z Z
 
X Y
R(X, Y, Z ) = Z m r , .
Z Z

6.2.1 The Existence of Rational First Integrals

via Intersection Numbers

In this section one of the key tools is the intersection number of two planar algebraic
curves. Let Ck be the k-dimensional affine space over C with coordinates x. For
P ∈ Ck , we define O P (Ck ) to be the set of rational functions on Ck that are defined
at P, i.e. those rational functions whose denominators do not vanish at P. We can
check that O P (Ck ) is a subring of C(x), and it is called a local ring of Ck at P.
Let F and G be two planar curves in C2 , and P ∈ C2 . Assume that F and G have
no common component that passes through P. From Fulton [157, Sect. 3.3], the inter-
section number of F and G at P, denoted by I (P, F ∩ G), is defined through seven
properties, and is in fact the dimension of the quotient space O P (C2 )/(F, G), i.e.
 
I (P, F ∩ G) = dim O P (C2 )/(F, G) ,

where (F, G) is the ideal in O P (C2 ). Note that O P (C2 )/(F, G) is a vector space
over C. From Fulton [157, Page 81] we have

I (P, F ∩ G) = dim (C[x]/(F, G)) , (6.8)
P∈Λ

where Λ is the set of points in C2 where F and Q intersect.

260 6 Applications of the Darboux Theory of Integrability

Example For F = x 2 − y 2 and G = y − x 2 , their intersection points in C2 are

(x, y) = (0, 0), (1, 1) and (−1, 1).
At P0 = (0, 0), we have


U (x, y)

O P0 (C2 ) = U, V ∈ C[x, y], V (0, 0) = 0 .

1 + V (x, y)

In O P0 (C2 ), the ideal

(F, G) = {R F + SG| R, S ∈ O P0 (C2 )}.

We claim that (F, G) = (y, x 2 ) in O P0 (C2 ). Indeed, the polynomials x 2 − y 2 , y −

x 2 ∈ (F, G) induce that y − y 2 = (x 2 − y 2 ) + (y − x 2 ) ∈ (F, G). In addition,
1/(1 − y) ∈ O P0 (C2 ) implies y ∈ (F, G). Consequently, x 2 ∈ (F, G). The claim
follows. By this claim we get that O P0 (C2 )/(F, G) is a two-dimensional vector space
with the basis {1, x}. Hence I (P0 , F ∩ G) = 2.
At P1 = (1, 1), we take the change of variables u = x − 1, v = y − 1. Then


U (u, v)

O P1 (C2 ) = U, V ∈ C[u, v], V (0, 0) = 0 ,

1 + V (u, v)

and F = 2u − 2v + u 2 − v2 and G = v − 2u − u 2 . In O P1 (C2 ), the ideal


(F, G) = R F + SG| R, S ∈ O P1 (C2 ) .

We claim that (F, G) = (u, v) in O P1 (C2 ). In fact, F, G ∈ (F, G) induce that

v + v2 ∈ (F, G), and so v ∈ (F, G). Consequently, 2u + u 2 ∈ (F, G). This shows
that u ∈ (F, G). The claim follows. Hence O P1 (C2 )/(F, G) is one-dimensional
vector space with the basis {1}. This proves that I (P1 , F ∩ G) = 1.
At P2 = (−1, 1), the same arguments as those used in treating P1 verify that
I (P2 , F ∩ G) = 1.
By the equality (6.8) we have dim(C[x, y]/(F, G)) = 4. Moreover, C[x, y]/
(F, G) has the basis {1, x, y, x y}. This follows from
 
(F, G) = P F + QG| P, Q ∈ C[x, y]} = {(y − x 2 )P + (y − y 2 )Q| P, Q ∈ C[x, y] ,

and the fact that any H (x, y) ∈ C[x, y] can be expressed as

H (x, y) = a0 + a1 x + b1 y + c11 x y + (y − x 2 ) f (x, y) + (y − y 2 )g(x, y),

for some f, g ∈ C[x, y]. This last fact can be obtained from x 2 = y − (y − x 2 ),
y 2 = y − (y − y 2 ), x 3 = x y − (y − x 2 )x, x 2 y = y − (y − y 2 ) − (y − x 2 )y,
x y 2 = x y − (y − y 2 )x, and y 3 = y − (y − y 2 )(1 + y), and so on.
6.2 Algebraic Limit Cycles: Existence and Uniqueness 261

The next result, called Bézout’s theorem, provides the intersection number of two
algebraic curves in the complex projective plane, see e.g. Fulton [157]. Let F and G
be two projective planar curves in CP2 with the coordinates [X : Y : Z ]. Assume that
F and G have no common factor. Let P = [X 0 : Y0 : Z 0 ] ∈ CP2 be an intersection
point of F and G. If Z 0 = 0, set p = (x, y) with x = X 0 /Z 0 and y = Y0 /Z 0 ,
and f (x, y) = F(x, y, 1) and g(x, y) = G(x, y, 1). Then the intersection number
I (P, F ∩ G) is I ( p, f ∩ g). If Z 0 = 0, we have X 0 = 0 or Y0 = 0. If X 0 = 0, we
set p = (y, z) with y = Y0 / X 0 and z = Z 0 / X 0 = 0, and f (y, z) = F(1, y, z) and
g(y, z) = G(1, y, z). Then the intersection number I (P, F ∩ G) is I ( p, f ∩ g). So
the intersection number of F and G in CP2 is

I (F, G) = I (P, F ∩ G),
P∈P

where P is the set of points in CP2 where F and G intersect. In the following we
also call P the set of intersection points of F and G in CP2 .
Bézout’s theorem. Assume that F = 0 and G = 0 are two algebraic curves in CP2
with no common factors, then their intersection number isI (F, G) = deg Fdeg G.
For a holomorphic foliation F defined by system (6.5), the next result provides a
sufficient condition for F to have a rational first integral, see Chavarriga and Llibre
[74] and Zhang [463].

Theorem 6.6 Assume that a holomorphic foliation F of degree m in CP2 has an

irreducible invariant algebraic curve F(X, Y, Z ) = 0 of degree n > 1 with a cofactor
K , that is

∂F ∂F ∂F
X (F) = P(X, Y, Z ) + Q(X, Y, Z ) − R(X, Y, Z ) = K F,
∂X ∂Y ∂Z

where X is the vector field (6.6) representing the foliation F . If m 2 is the intersection
number of the curves

n P − X K = 0, n Q − Y K = 0, n R + Z K = 0,

and m of them are at infinity, then F has a rational first integral.

Remark If R ≡ 0, i.e. r ≡ 0, Theorem 6.6 was proved by Chavarriga and Llibre in

[74, Theorem 2], whereas if R ≡ 0, the theorem was proved by Zhang [463].

Theorem 6.6 will be used in the proof of the result that any quadratic differential
system has no cubic limit cycle, which will be presented in the coming subsection.
Proof of Theorem 6.6. According to (6.6), we write P, Q, R as polynomials in Z ,
i.e.
262 6 Applications of the Darboux Theory of Integrability

P(X, Y, Z ) = p0 Z m + p1 (X, Y )Z m−1 + · · · + pm (X, Y ),

Q(X, Y, Z ) = q0 Z m + q1 (X, Y )Z m−1 + · · · + qm (X, Y ),
K (X, Y, Z ) = k0 Z m−1 + k1 (X, Y )Z m−2 + · · · + km−1 (X, Y ),

where ρi (X, Y ), ρ ∈ { p, q, k}, is a homogeneous polynomial of degree i. Since

n P − X K = 0, n Q − Y K = 0, n R + Z K = 0

have the intersection number m at infinity, there exist constants λ1 , λ2 such that

npm (X, Y ) − X km−1 (X, Y ) = λ1 R(X, Y ),

(6.9)
nqm (X, Y ) − Y km−1 (X, Y ) = λ2 R(X, Y ).

This shows that equation (6.5) can be written as

λ1 x
ẋ = p0 + p1 (x, y) + · · · + pm−1 (x, y) + r (x, y) + (km−1 + nr (x, y)),
n n
λ2 y
ẏ = q0 + q1 (x, y) + · · · + qm−1 (x, y) + r (x, y) + (km−1 + nr (x, y)).
n n
(6.10)
Case 1: λ1 = λ2 = 0. Then

npm (x, y) ≡ xkm−1 (x, y), nqm (x, y) ≡ ykm−1 (x, y).

These imply that np(x, y) − xk(x, y) and nq(x, y) − yk(x, y) are of degree at most
m − 1. By the assumption we get that

np − xk = 0, nq − yk = 0, and nr + k = 0

have the intersection number m 2 − m. Hence the intersection number of

np − xk = 0 and nq − yk = 0

is at least m 2 − m. By Bézout’s theorem, np − xk and nq − yk must have a common

factor. Otherwise, the intersection number of np − xk = 0 and nq − yk = 0 is at
most (m − 1)2 , a contradiction.
Set C := gcd(np − xk, nq − yk), and  := deg C. Then 0 <  ≤ m − 1. Set

np − xk = uC, nq − yk = vC.

Then system (6.10) can be written as

1 1 1 1
ẋ = uC(x, y) + x(k + nr ), ẏ = vC(x, y) + y(k + nr ). (6.11)
n n n n
6.2 Algebraic Limit Cycles: Existence and Uniqueness 263

We claim that u and v are both constant. Indeed, again by the fact that np − xk = 0,
nq − yk = 0 and nr + k = 0 have the intersection number m 2 − m, we get that

C =0 and nr + k = 0

have the intersection number m, and that

u = 0, v = 0 and nr + k = 0

have the intersection number m(m −  − 1). But u and v are relatively prime, and
have degrees at most m −  − 1, so it follows that  = m − 1. Consequently, u and
v are constants.
If u = 0 (resp. v = 0), then the line x = 0 (resp. y = 0) is invariant with cofactor
k/n. By Proposition 3.5 system (6.5) has the rational first integral
 
H (x, y) = f (x, y)x −n , resp. H (x, y) = f (x, y)y −n .

Consequently
 
H = F(X, Y, Z )X −n , resp. H = F(X, Y, Z )Y −n ,

is a rational first integral of the foliation F in CP2 .

If u, v = 0, system (6.11) has the invariant line vx − uy = 0 with cofactor k/n.
Hence system (6.5) has the rational first integral

H (x, y) = f (x, y)(vx − uy)−n .

Consequently, the foliation F has the rational first integral

H = F(X, Y, Z )(vX − uY )−n .

Case 2: λ1 and λ2 are not both zero. We assume without loss of generality that λ2 = 0.
Taking the change of coordinates

w = λ2 x − λ1 y, y = y,

system (6.10) becomes

1
ẇ = b(x, y) + wk(x, y),
n
1 λ2 1
ẏ = q0 + q1 (x, y) + · · · + qm−1 (x, y) − y(k0 + · · · + km−2 ) + r + yk,
n n n
(6.12)
where x = (w + λ1 y)/λ2 , and
264 6 Applications of the Darboux Theory of Integrability
 
1
b(x, y) = λ2 p0 + p1 + · · · + pm−1 − x(k0 + · · · + km−2 )
n
 
1
−λ1 q0 + q1 + · · · + qm−1 − y(k0 + · · · + km−2 ) .
n

Similar arguments as those used in Case 1 verify that the polynomials b and nq −
yk − λn2 k have a common factor.
If b ≡ 0, system (6.12) has the invariant line λ2 x − λ1 y = 0 with cofactor k/n.
So system (6.12) has the first integral

H (x, y) = f (x, y)(λ2 x − λ1 y)−n .

Consequently the foliation F has the rational first integral

H = F(X, Y, Z )(λ2 X − λ1 Y )−n .

If b ≡ 0, since
 
λ2 1 λ2
q + yr − k = nq − yk − k ≡ 0,
n2 n n

set C := gcd(b, nq − yk − λn2 k),  := deg C and set

λ2
b = uC, nq − yk − k = vC.
n
Then system (6.12) can be written as
 
1 1 1 λ2
ẇ = uC + wk(x, y), ẏ = vC + + y k(x, y), (6.13)
n n n n

where we have used the fact that

1 λ2 1
q0 + q1 + · · · + qm−1 − y(k0 + · · · + km−2 ) + r + yk
n n n
1 1 λ2
= vC + ( + y)k, by (6.9).
n n n
We claim that u and v are both nonzero constants. Indeed, by Bézout’s theorem it
follows that C = 0 and k = 0 have the intersection number m; and that u = 0, v = 0
and k = 0 have the intersection number m(m −−1). Since deg u, deg v ≤ m −−1,
and u and v are coprime, we get  = m − 1. The claim follows.
We can check that system (6.13) has the invariant algebraic curve f 1 = vw −
nuy − uλ2 = 0 with cofactor k/n. It follows that system (6.5) has the first integral
6.2 Algebraic Limit Cycles: Existence and Uniqueness 265

H (x, y) = f (x, y)(vλ2 x − vλ1 y − nuy − uλ2 )−n .

Consequently, the foliation F has the first integral

H (X, Y, Z ) = F(X, Y, Z )(vλ2 X − vλ1 Y − nuY − uλ2 Z )−n .

This proves the theorem. 

6.2.2 Quadratic Differential Systems: Algebraic Limit Cycles

In this subsection we will apply Theorem 6.6 to prove that quadratic foliations have
no cubic limit cycles.

Theorem 6.7 For the quadratic holomorphic foliation F in CP2 , the following
statements hold.
(a) Any irreducible invariant algebraic curve with an elliptic branch of the folia-
tion F has no singularity.
(b) The foliation F has no cubic limit cycles.

Remark The three papers [140–142] by Evdokimenco from 1970 to 1974 proved
that a quadratic differential system has no cubic algebraic limit cycles. Shen [393] in
1991 proved the same result for affine quadratic differential systems using a different
method. Here the proof is due to Zhang [463, Theorem 2.2], but the main idea comes
from Chavarriga et al. [75].

To prove Theorem 6.7 we need two fundamental results on algebraic curves, see
Fulton [157].

Lemma 6.1 Any irreducible cubic algebraic curve in CP2 either has no singularities
or has at most one singularity of multiplicity 2.

Lemma 6.2 If F(X, Y, Z ) = 0 is an algebraic curve of degree n in CP2 , and

L = 0 is a projective line not contained in F = 0, then their intersection number is
I (F, L) = n.

In addition, we need the following Darboux lemma, a simple proof of which can
be found in [75].

Lemma 6.3 (Darboux lemma) Let A, A , B, B  , C and C  be homogeneous poly-

nomials in CP2 of degree ,  , m, m  , n and n  respectively, and satisfying

 +  = m + m  = n + n  := λ and A A + B B  + CC  ≡ 0.
266 6 Applications of the Darboux Theory of Integrability

Assume that the curves A = 0, B = 0 and C = 0; and also A = 0, B  = 0 and

C  = 0, have no common factors. Then their intersection number satisfies

mn +  m  n 
I (A, B, C) + I (A , B  , C  ) ≥ .
λ
Furthermore, if the six curves have no common intersection point, equality holds.

Proof of Theorem 6.7. (a) According to Lemma 6.1, we only need to prove that any
irreducible cubic curve having an elliptic branch has no multiple point.
On the contrary, we assume that there is an irreducible cubic curve F = 0 which
has an elliptic branch and a multiple point. Then the multiple point is not on the
elliptic branch. Furthermore, for a projective line L = 0 passing through the interior
of the elliptic branch and the multiple point, we have I (F, L) ≥ 4, a contradiction
with Lemma 6.2. So it follows from Lemma 6.1 that any irreducible cubic curve
having an elliptic branch is nonsingular. This proves the statement.
(b) We claim that if F has an irreducible algebraic curve with an elliptic branch,
then it has a rational first integral.
In the following we continue to use the notations given in Theorem 6.6 and its
proof. Let F be an irreducible invariant cubic curve of F with cofactor K . Then

∂F ∂F ∂F
P +Q −R = K F,
∂X ∂Y ∂Z
or equivalently
     
1 ∂F 1 ∂F 1 ∂F
P − XK + Q − YK − R + ZK = 0,
n ∂X n ∂Y n ∂Z

where we have used the Euler formula that the homogenous polynomials satisfy.
Since F has an elliptic branch, it has no singularities by statement (a). Hence F has
no multiple points, and consequently
 
∂F ∂F ∂F
h := I , , = 0. (6.14)
∂ X ∂Y ∂ Z

Since p+xr and q + yr have no common factor, P − n1 X K , Q − n1 Y K and R + n1 Z K

are forced to have no common factor. By the Darboux lemma and (6.14) one gets
 
1 1 1
h  := I P− X K, Q − Y K, R + Z K = 4.
n n n

Hence it follows from Theorem 6.6 that the foliation F has a rational first integral.
The claim follows.
6.2 Algebraic Limit Cycles: Existence and Uniqueness 267

This last claim implies that a quadratic foliation F in CP2 has no cubic limit cycle.
Otherwise, the foliation has both limit cycles and rational first integrals, which is not
possible. This completes the proof of the theorem. 
Besides the result on the nonexistence of cubic limit cycles for quadratic differen-
tial systems, there are also classifications of quadratic differential systems that have
either a quadratic limit cycle or a quartic limit cycle.
Qin [369] in 1958 proved that if a quadratic differential system has a quadratic limit
cycle, then through an affine change of coordinates the limit cycle can be expressed
as Γ := x 2 + y 2 − 1 = 0, and the quadratic differential system can be written as

ẋ = −y(ax + by + c) − (x 2 + y 2 − 1), ẏ = x(ax + by + c).

Moreover, the system has a unique limit cycle, i.e. the algebraic one.
For quadratic differential systems having a quartic limit cycle, Yablonskii [449]
in 1966 found a family of quadratic systems that have a quartic limit cycle. Filiptsov
[152] in 1973 found another family of quadratic systems that have a quartic limit
cycle. Chavarriga in 1999 found a third family of quadratic systems having a quar-
tic limit cycle (which has not been published). Applying the intersection numbers,
Chavarriga et al. [76, Theorem 1] in 2004 completed the classification of quadratic
systems that have a quartic limit cycle, and proved that there are exactly four families
of quadratic differential systems that can have a quartic limit cycle. The next results
exhibit the classification mentioned above.

Theorem 6.8 A quadratic differential system with a quartic limit cycle is affinely
equivalent to one of the four systems
(a) (Yablonskii) The system

ẋ = −4abcx − (a + b)y + 3(a + b)cx 2 + 4x y,

 
3
ẏ = (a + b)abx − 4abcy + 4abc2 − (a + b)2 + 4ab x 2 + 8(a + b)cx y + 8y 2 ,
2

where abc = 0, a = b, ab > 0 and 4c2 (a − b)2 + (3a − b)(a − 3b) < 0. This
system has the quartic irreducible invariant algebraic curve
 2
y + cx 2 + x 2 (x − a)(x − b) = 0,

which has two branches: an elliptic one (algebraic limit cycle) and an isolated
singularity.
(b) (Filipstov) The system

ẋ = 6(1 + a)x + 2y − 6(2 + a)x 2 + 12x y,

ẏ = 15(1 + a)y + 3a(1 + a)x 2 − 2(9 + 5a)x y + 16y 2 ,
268 6 Applications of the Darboux Theory of Integrability

3
where 0 < a < . This system has the quartic irreducible invariant algebraic
curve 13

3(1 + a)(ax 2 + y)2 + 2y 2 (2y − 3(1 + a)x) = 0,

which has two branches: one topologically equivalent to an ellipse and the other
topologically equivalent to a straight line.
(c) The system

ẋ = 5x + 6x 2 + 4(1 + a)x y + ay 2 ,
ẏ = x + 2y + 4x y + (2 + 3a)y 2 ,
√
−71 + 17 17
where < a < 0. This system has the quartic irreducible invariant
32
algebraic curve

x 2 + x 3 + x 2 y + 2ax y 2 + 2ax y 3 + a 2 y 4 = 0,

which has three branches: one topologically equivalent to an ellipse and the
other two topologically equivalent to straight lines. And each of the topological
lines contains a singularity of the system.
(d) The system

ẋ = 2(1 + 2x − 2kx 2 + 6x y),

ẏ = 8 − 3k − 14kx − 2kx y − 8y 2 ,

1
where 0 < k < . This system has the quartic irreducible invariant algebraic
curve 4

1
+ x − x 2 + kx 3 + x y + x 2 y 2 = 0,
4
which has three branches: one topologically equivalent to an ellipse and the
other two topologically equivalent to straight lines. And each of the topological
lines contains a singularity of the system.

The proof of this theorem is very complicated, and so it will be omitted here. We
refer the reader to [76] for its original proof.

Remark Chavarriga et al. [69] proved the uniqueness of the limit cycles of the systems
in Theorem 6.8. That is, each of the four systems in Theorem 6.8 has no other limit
cycles than the algebraic one.
6.2 Algebraic Limit Cycles: Existence and Uniqueness 269

Besides the above results, there are also lots of other results on algebraic limit
cycles. However, they are only partial results for concrete systems, see e.g. [56, 107,
181, 249, 271] and the references therein.
On the relation between limit cycles and invariant algebraic curves, we have the
following unsolved conjecture: A quadratic differential system having an invariant
algebraic curve has at most one limit cycle.
This conjecture has been verified only in the cases of quadratic differential systems
having an invariant line or an invariant quadratic algebraic curve. These results were
proved by Cherkas and Zhilevich [84, 85], and Rychkov [381], see also Ye [452].
For quadratic differential systems having an invariant algebraic curve of degree ≥ 3,
there are no general results concerning this conjecture.

6.3 Hilbert’s 16th Problem: A Weak Version on Algebraic

Limit Cycles

In this section we will discuss the maximal number of algebraic limit cycles of
polynomial differential systems using the Darboux theory of integrability. Hilbert’s
16th problem consists of two parts: the first concerns the configuration of the maximal
number of branches that the planar algebraic curves of a given degree can have, and
the second concerns the maximal number of limit cycles and their distribution for
polynomial differential systems of a given degree. There has been great progress in
the past century on these two different parts of Hilbert’s 16th problem. But they still
remain open in general, see Smale [404].
Weakening this problem, Llibre et al. posed a weak version of Hilbert’s 16th
problem on algebraic limit cycles, see Llibre et al. [264, p.1402].
A weak version of Hilbert’s 16th problem For real planar polynomial vector fields
of a given degree, does the number of algebraic limit cycles have a uniform upper
bound depending only on the degree of the vector fields?
This weak version of Hilbert’s 16th problem provides a beautiful relation between
the two parts of Hilbert’s 16th problem.
Consider real planar polynomial vector fields of degree m

∂ ∂
X = p(x, y) + q(x, y) , (6.15)
∂x ∂y

where p(x, y), q(x, y) ∈ R[x, y] and max{deg p, deg q} = m. Llibre et al. [264]
solved this weakened Hilbert’s 16th problem in the case when all invariant algebraic
curves of the vector fields are generic. Recall that the invariant algebraic curves
f j = 0, j = 1, . . . , k, of the vector field are generic if all of the following conditions
hold.
270 6 Applications of the Darboux Theory of Integrability

• All the curves f j = 0 are nonsingular, that is there do not exist points on the curve
at which all partial derivatives of f j vanish.
• The highest order homogeneous term of each f j has no repeated factors.
• If two curves intersect, they must intersect transversally.
• There are no three curves intersecting at the same point.
• The higher-order homogeneous terms of any two curves have no common factor.
The results of Llibre et al. [264] can be stated as follows:
Theorem 6.9 Assume that all irreducible invariant algebraic curves of a real planar
polynomial vector field X of degree m are generic, then the following statements
hold.
• The maximal number of algebraic limit cycles of X does not exceed 1 + (m −
1)(m − 2)/2 if m is even, or (m − 1)(m − 2)/2 if m is odd.
• There exist real planar polynomial vector fields of degree m which have the max-
imal number of limit cycles.
Inspired by this last result, Llibre et al. [264] posed the following conjecture.

Conjecture Is 1 + (m − 1)(m − 2)/2 the maximal number of limit cycles that the
real planar polynomial vector fields of degree m can have?

Llibre et al. [265] obtained another upper bound when all invariant algebraic
curves of the vector fields are nonsingular. Note that both of their results in [264,
265] require that all invariant algebraic curves are nonsingular.
Zhang [471] in 2011 obtained the next result in which the invariant algebraic
curves can be singular. This result verifies the above conjecture in the case when all
invariant algebraic curves are of nodal type.
Theorem 6.10 If all invariant algebraic curves including the line at infinity of a
real planar polynomial vector field (6.15) of degree m are nodal, then
(a) the maximal number of algebraic limit cycles of the vector field (6.15) is at
most 1 + (m − 1)(m − 2)/2 if m is even, or (m − 1)(m − 2)/2 if m is odd;
(b) there exist polynomial vector fields (6.15) of degree m satisfying the given
condition which have the maximal number of algebraic limit cycles.
To prove Theorem 6.10 we need the next result by Harnack. Its proof can be found
in Coolidge [118], Wilson [443] and Viro [435].
Theorem 6.11 (Harnack’s theorem) The number of elliptic branches of a real irre-
ducible algebraic curve S of degree n is at most

1 + (n − 1)(n − 2)/2 − O p (S)(O p (S) − 1), if n is even,
p

(n − 1)(n − 2)/2 − O p (S)(O p (S) − 1), if n is odd,
p
6.3 Hilbert’s 16th Problem: A Weak Version on Algebraic Limit Cycles 271

where p runs over all singularities of S, and O p (S) is the order of the curve S at the
singularity p. Furthermore, there exist real irreducible algebraic curves of degree n
which can have the maximal number of elliptic branches.

The next result, due to Llibre et al. [264, Lemma 6], will also be used in the proof
of Theorem 6.10.

Lemma 6.4 For , s ∈ N, and  ≥ s, set

⎧ ⎫
⎨

s ⎬
D := (x1 , . . . , xs ) ∈ Rs
x j ≥ 1, j = 1, . . . , s, xj ≤  ,
⎩ ⎭
j=1

and

k: D −→ R
s  
1 1
s
3 2 1
(x1 , . . . , xs ) −→ (x j − 1)(x j − 2) = xj − − s.
2 j=1 2 j=1 2 8

Then the maximal value of k on D is ( − s)( − s − 1)/2 ≥ 0, and it is obtained at

the vertex ( + 1 − s, 1, . . . , 1) of D.

Proof of Theorem 6.10. (a) Write the vector field (6.15) as the differential one-form

q(x, y)d x − p(x, y)dy.

Its projective one-form is

ω0 = Z Qd X − Z PdY + (Y P − X Q)d Z ,

where X, Y, Z are the homogeneous coordinates, and

P = Z m p(X/Z , Y/Z ) and Q = Z m q(X/Z , Y/Z ).

Let F0 be the singular holomorphic foliation defined by the projective one-form

ω0 . Clearly, the line at infinity of F0 is invariant. By the assumption we get from
Theorem 4.1 that the sum n of the degrees of all irreducible invariant algebraic curves
of the foliation F0 is no more than m + 2.
Case 1. n = m + 2. Theorem 4.1 shows that F is reducible, say with irreducible
decomposition F = F1 · · · · · Fk with k ≥ 2, and that the one-form ω0 has the
expression


k
d Fi
ω0 = F σi ,
i=1
Fi
272 6 Applications of the Darboux Theory of Integrability

where σi ∈ C, and Fi can be complex, as mentioned before. The one-form ω0 has

the inverse integrating factor F, and consequently it is Darboux integrable with the
Darboux first integral

H (X, Y, Z ) = F1σ1 · · · · · Fkσk .

Since the one-form ω0 is projective, we should have σ1 deg F1 + · · · + σk deg Fk = 0.

For k = 2, the foliation F0 has the rational first integral

H (X, Y, Z ) = F1μ F2−ν with μ, ν ∈ N, (μ, ν) = 1 and μ/ν = deg F2 / deg F1 ,

and so it has infinitely many invariant algebraic curves. This contradicts the fact that
n ≤ m + 2.
For k ≥ 3, by Harnack’s theorem we get that the total number of ovals which are
contained in Fi , for i = 1, . . . , k, is no more than
k 
 
(m + 2 − k)(m + 1 − k) 
k
(deg Fi − 1)(deg Fi − 2)
+ δi ≤ M(k) := + δj,
2 2
i=1 i=1

where either δi = 1 for deg Fi even or δi = 0 for deg Fi odd, and we have used
Lemma 6.4 and deg F1 + · · · + deg Fk = m + 2. Note from Lemma 6.4 that equality
holds if and only if one of the Fi ’s has degree m + 3 − k and the others all have
degree 1. We can check that the maximum of the M(k) for k ∈ {3, . . . , m + 2} is
k
attained at k = 3, because δ j ≤ [m/2] + 1, where [·] denotes the integer part
j=1
function. Moreover, if k = 3 and the three invariant algebraic curves respectively
have degrees 1, 1 and m, then the maximum is

(m − 1)(m − 2)
+ δ,
2
where δ = 1 if m is even and δ = 0 if m is odd.
Case 2. n ≤ m + 1. Since the line at infinity is invariant under the foliation F0 , the
total degree of the invariant algebraic curves in the affine plane is no more than m.
We get from Harnack’s theorem and the proof of Case 1 that the number of algebraic
limit cycles is less than or equal to the maximal value that the theorem stated. This
proves statement (a).
(b) We only need to provide an example of a real planar polynomial differential
system (6.15) of degree m which has the maximal number of algebraic limit cycles
and whose invariant algebraic curves in the affine plane have total degree m and
m + 1 respectively.
Case 1. The number m + 1 is the total degree of the invariant algebraic curves in
the affine plane. By Harnack’s theorem there exists a nonsingular algebraic curve of
6.3 Hilbert’s 16th Problem: A Weak Version on Algebraic Limit Cycles 273

degree m, denoted by f 1 , which has the maximal number, i.e. (m − 1)(m − 2)/2 + δ,
of ovals, where δ = 1 if m is even or δ = 0 if m is odd.
Choose a straight line, namely f 2 , as the line at infinity which is located outside the
ovals of f 1 and intersects f 1 transversally. Choose another straight line, denoted by
f 3 , such that it is located outside the ovals of f 1 and intersects f 1 and f 2 transversally
and these three curves do not meet at the same point.
Let F1 , F2 and F3 be the projectivization of f 1 , f 2 and f 3 , respectively. Taking
σ1 , σ2 , σ3 ∈ R not zero such that

σ1 m + σ2 + σ3 = 0, σi /σ j ∈
/ {r ∈ Q; r < 0},

the foliation Fm induced by the projective one-form

σ1 F2 F3 d F1 + σ2 F1 F3 d F2 + σ3 F1 F2 d F3

has only the three invariant algebraic curves F1 , F2 and F3 , because it has the Darboux
first integral F1σ1 F2σ2 F3σ3 . In addition, Fm has the inverse integrating factor F1 F2 F3 .
So we get from Theorem 2.6 that all limit cycles of Fm must be contained in the
vanishing set of F1 F2 F3 , i.e. of F1 . Hence Fm has exactly (m − 1)(m − 2)/2 + δ
algebraic limit cycles. Note that Fm is a singular holomorphic foliation of degree m,
and that the line at infinity is invariant by Fm . So its affine expression should be a
polynomial differential system of degree m and it has the form (6.15).
Case 2. The number m is the total degree of the invariant algebraic curves in the
affine plane. The proof of this case follows from Christopher [96].
By Harnack’s theorem, there exists a nonsingular algebraic curve of degree m
having the maximal number M of elliptic branches, where M = (m −1)(m −2)/2+δ
with either δ = 1 for m even or δ = 0 for m odd. Denote by g(x, y) this nonsingular
curve. Taking a linear function h(x, y) such that h = 0 does not intersects the elliptic
branches of g = 0, choose c, d ∈ R satisfying ch x + dh y = 0. Then the real planar
polynomial differential system

ẋ = cg − hg y , ẏ = dg + hgx (6.16)

is of degree m, and it has the elliptic branches of g = 0 as hyperbolic limit cycles.

Furthermore, system (6.16) has no other limit cycles. This proves statement (b).
This completes the proof of the theorem. 
Theorem 6.10 confirms the conjecture when the algebraic curves have only nodal
singularities. Note that the assumption in Theorem 6.10 is on the singularities of the
invariant algebraic curves but not on those of the vector field.
We now discuss the case when all invariant algebraic curves have only nondicrit-
ical singularities of the vector field. Consider a vector field
274 6 Applications of the Darboux Theory of Integrability

∂ ∂
X = ( p(x, y) + xr (x, y)) + (q(x, y) + yr (x, y)) , (6.17)
∂x ∂y

where p, q, r ∈ R[x, y], max{deg p, deg q, deg r } = m, and r is either a homoge-

neous polynomial of degree m or identically zero. As we mentioned previously, m
is the degree of the vector field (6.17). We remark that if r ≡ 0, when working in
affine space we usually treat the vector field (6.17) as having degree m + 1.
The next result, due to Zhang [471, Theorem 1.2], solves the weak version of
Hilbert’s 16th problem on algebraic limit cycles when all invariant algebraic curves
have only nondicritical singularities of the vector field.

Theorem 6.12 Assume that a real planar polynomial vector field (6.17) of degree
m has all its invariant algebraic curves nondicritical. The following statements hold.
(a) If r (x, y) ≡ 0, the maximal number of algebraic limit cycles of the vector field
is at most 1 + m(m − 1)/2 when m is even, or m(m − 1)/2 when m is odd.
(b) If r (x, y) ≡ 0, the maximal number of algebraic limit cycles of the vector field
is at most 1 + (m + 1)m/2 when m is even, or (m + 1)m/2 when m is odd.

Proof Write system (6.17) as the differential one-form

(q(x, y) + yr (x, y))d x − ( p(x, y) + xr (x, y))dy.

Its projective one-form is

ω1 = (Z Q + Y R)d X − (Z P + X R)dY + (Y P − X Q)d Z ,

where X, Y, Z are the homogeneous coordinates in CP2 and

P = Z m p(X/Z , Y/Z ), Q = Z m q(X/Z , Y/Z ), R = Z m r (X/Z , Y/Z ).

Denote by F1 the singular holomorphic foliation induced by ω1 . By the assump-

tion we get from Theorem 4.2 that the total degree of the invariant algebraic curves
of F1 is no more than m + 2.
(a) If r (x, y) ≡ 0, the line at infinity is invariant under the foliation F1 . So the total
degree n of all the invariant algebraic curves in the affine plane is at most m + 1.
Then the proof follows from that of Theorem 6.10.
(b) If r (x, y) ≡ 0, the line at infinity is composed of singularities of the foliation
F1 . Then the total degree n of the invariant algebraic curves of the vector field (6.17)
in the affine plane is no more than m + 2.
We claim that there exist vector fields (6.17) of degree m with r (x, y) ≡ 0, whose
invariant algebraic curves have total degree m + 2. Indeed, take reduced polynomials
f 1 , . . . , f k ∈ C[x, y], k ≥ 3, satisfying deg f 1 + · · · + deg f k = m + 2, such that
their projectivizations F1 , . . . , Fk in CP2 are nonsingular, and that they intersect
transversally, and any three of them cannot intersect at the same point. Furthermore,
6.3 Hilbert’s 16th Problem: A Weak Version on Algebraic Limit Cycles 275

choose the nonzero constants σ1 , . . . , σk ∈ C satisfying σi /σ j ∈

/ {r ∈ Q; r < 0},
1 ≤ i = j ≤ k such that

σ1 deg F1 + · · · + σk deg Fk = 0.

Then the foliation F ∗ induced by the one-form

⎛ ⎞

k 
k
ω∗ = σj ⎝ Fi d F j ⎠
j=1 i=1,i= j

is of degree m, and has only the invariant algebraic curves F1 , . . . , Fk . Moreover, the
singularities of F ∗ are all nondicritical. This fact follows from Seidenberg’s theorem
[389] on the reduction of singularities and the fact that all invariant curves of F ∗
passing through these singularities are the branches of the curves Fi , i = 1, . . . , k.
By Theorem 4.2 the total degree of the invariant algebraic curves of the foliation F ∗
is at most m + 2. Meanwhile, f 1 , . . . , f k have total degree m + 2. Hence, the line at
infinity of F ∗ is not invariant. This implies that the affine expression of F ∗ must be
of the form (6.17) with r (x, y) ≡ 0. The claim follows.
If F1 has an irreducible invariant algebraic curve of degree m +2, and the number
of elliptic branches of this curve is equal to the maximal number stated by Harnack’s
theorem, then the foliation has the maximal number of algebraic limit cycles. In all
other cases there do not exist vector fields (6.17) of degree m which have the maximal
number of algebraic limit cycles.
This completes the proof of the theorem.


Remark
• By the proof of Theorems 6.10 and 6.12, if a vector field has the maximal number
of algebraic limit cycles, then it is Darboux integrable.
• We conjecture that in the nondicritical case the minimal upper bound is still that
given in Theorem 6.10.
• Does there exist a foliation of degree m whose nondicritical irreducible invariant
algebraic curve is of degree either m + 1 or m + 2?
We note from Theorem 4.1 that this is not possible in the nodal case.

Theorem 6.12 and its proof have the next consequences.

Corollary 6.1 Assume that a polynomial vector field (6.17) of degree m has no
dicritical singularities. The following statements hold.
(a) If r (x, y) ≡ 0, the maximal number of algebraic limit cycles of the vector
field (6.17) is at most 1 + m(m − 1)/2 if m is even, or m(m − 1)/2 if m is odd.
(b) If r (x, y) ≡ 0, the maximal number of algebraic limit cycles of the vector
field (6.17) is at most 1 + m(m + 1)/2 if m is even, or m(m + 1)/2 if m is odd.
276 6 Applications of the Darboux Theory of Integrability

Corollary 6.2 Assume that a real planar polynomial vector field (6.17) of degree m
has no dicritical singularities, and has at least three irreducible invariant algebraic
curves. The following statements hold.
(a) The maximal number of algebraic limit cycles of the vector field (6.17) is at
most 1 + (m − 1)(m − 2)/2 if m is even, or (m − 1)(m − 2)/2 if m is odd.
(b) There exists a polynomial vector field (6.17) satisfying r (x, y) ≡ 0 which has
the maximal number of limit cycles. Furthermore this kind of vector field has
exactly three invariant algebraic curves.

Remark Corollary 6.2 verifies the conjecture posed by Llibre et al. in [264] in the
nondicritical case, with an additional condition on the number of irreducible invariant
algebraic curves.

6.4 Applications to Concrete Models

The theory of algebraic integrability has been widely applied to concrete models to
show the existence of Darboux, elementary and Liouvillian first integrals of these
models and their dynamics, where the models have the mentioned kind of invariants,
including polynomial or rational first integrals. In this section we collect these kinds
of results, which are known from the literature, but have not appeared in the previous
sections and chapters of this book.

6.4.1 Concrete Two-Dimensional Models in Applications

In 2013 Valls [427] studied the Liouvillian integrability of the continuous determin-
istic business model

x  + (1 − a)x  + (1 + a)(x  )3 + bx = 0, (6.18)

where the prime denotes the derivative with respect to time t, a > 0 and b ≥ 0. For
more details on the background of this model related to economics, see [427] and
the references therein. Equation (6.18) can be readily written as a planar differential
system
x  = y, y  = (a − 1)y − bx − (a + 1)y 3 . (6.19)

As we know, a planar polynomial differential system is Liouvillian integrable if and

only if it has a Darboux integrating factor. By computing the Darboux polynomials
and the exponential factors of this model, Valls [427] obtained the next results.

Theorem 6.13 For system (6.19), the following statements hold.

6.4 Applications to Concrete Models 277

(a) When b = 0, the system (6.19) has the first integral

⎧ √ √ 
⎪
⎪ 1 − a2 1 + ay
⎨ −(1 − a)x − arctan √ , if a = 1,
F= 1+a 1−a
⎪
⎪ 2x y − 1
⎩ , if a = 1.
y

(b) When b = 0,
(b1 ) system (6.19) has no polynomial first integrals.
(b2 ) system (6.19) has no proper Darboux polynomials.
2
(b3 ) system (6.19) has two exponential factors e x with cofactor y and e x
with cofactor 2x y.
(b4 ) system (6.19) has no Liouvillian first integrals.

For a generalized Liénard differential system

ẋ = y, ẏ = −cx − f (x)y,

Llibre and Valls [280, Theorem 1] provided some sufficient conditions on the exis-
tence of Liouvillian first integrals via the existence of Darboux integrating factors.
For a Riccati differential system

ẋ = 1, ẏ = a(x)y 2 + b(x)y + c(x),

under the additional condition c(x) = κ(b(x) − κa(x)) for κ ∈ C, Llibre and
Valls [288] verified the existence of Darboux polynomials, and gave a necessary and
sufficient condition for the existence of algebraic first integrals of this last system.

6.4.2 Concrete Three-Dimensional Models in Applications

In 2013 Llibre and Valls studied the Darboux integrability of the Falkner–Skan
equation  
f  + f f  + λ 1 − ( f  )2 = 0, (6.20)

where λ ∈ R is a parameter. This equation was derived by Falkner and Skan in [144]
to describe the steady two-dimensional flow of a slightly viscous incompressible
fluid past a wedge. In the case λ = 0, the wedge reduces to a flat plate, and equation
(6.20) is called the Blasius equation, which was first considered by Blasius in [34].
From the dynamical point of view, there is an extensive literature on the bifurca-
tions, the existence of oscillation and periodic orbits, and so on. See [286] and the
reference therein. Writing equation (6.20) as the differential system
278 6 Applications of the Darboux Theory of Integrability

ẋ = y, ẏ = z, ż = −x z − λ(1 − y 2 ), (6.21)

together with some recursive calculations, Llibre and Valls [286] obtained the fol-
lowing results.

Theorem 6.14 For the Falkner–Skan and Blasius system the following statements
hold.
(a) Both systems have no polynomial first integrals.
(b) The unique irreducible proper Darboux polynomial of the Blasius system is z.
(c) The unique proper Darboux polynomial of the Falkner–Skan system is 1 −
y 2 + 2x z if λ = 1/2.
(d) The unique exponential factors of both systems are e x and e y , except if λ = −1
then the systems have the additional exponential factor e z+x y .
(e) Both systems have no Darboux first integrals.

The Hide–Skeldon–Acheson dynamo model

ẋ = x(y − 1) − βz, ẏ = α(1 − x 2 ) − κ y, ż = x − λz, (6.22)

with α, β, λ, κ real parameters, was proposed by Hide, Skeldon and Acheson in

1996 for modeling a self-exciting dynamo action in which a Faraday disk and coil
are arranged in series with either a capacitor or a motor. For more details on the
physical meaning of this model, see [202], and on the dynamical analysis, see the
references in [326]. Here we report the results in [326, Theorem 1] concerning its
Liouvillian integrability.
Note that if α = 0, system (6.22) is essentially linear in x and z. For α = 0, set

α 1
F1 = y + α ln x − x 2 − y 2 ,
 2 2
F2 = zv(x) − v(x)w(x)d x,
 
2(κ − 1 + y + T)
F3 = κ log − T,
−x
√ √  √ √ 
κ −1− κ κ −1x + y
F4 = √ √ exp 2 κ κ − 1 z ,
κ −1+ κ κ −1x + y

where
  α −1/2
w(x) = 1 − 2 F3 + x 2 − α ln x ,
  2 
v(x) = exp λ xw(x)d x ,

T= κ 2 (1 − x 2 ) + (1 − y)2 + κ(x 2 + 2y − 2).

6.4 Applications to Concrete Models 279

Mahdi and Valls [326] obtained the next results.

Theorem 6.15 For system (6.22) with α = 0, the following statements hold.
(a) β=0
(a1 ) If κ = 0 then F1 and F2 are the first integrals.
(a2 ) If α = κ(1 − κ) and κ = 0, then F3 is a first integral. Additionally, if
λ = 0 then F4 is also a first integral.
(a3 ) If κ = 0 and α = κ(1 − κ), system (5.44) has no polynomial first
integral; has only Darboux polynomials generated by x; has only expo-
nential factors generated by e z ; and it is not Darboux integrable.
(b) β = 0
(b1 ) System (5.44) has no polynomial first integral; and no proper Darboux
polynomial.
(b2 ) System (5.44)
 has only exponential factors generated
 by e z and addi-
tionally exp −x /2 + y/α − y /(2α) − βz /2 if κ = λ = 0.
2 2 2

(b3 ) System (5.44) is not Darboux integrable.

The main tools in the proof of this last theorem are recursive calculation and the
characteristic method for solving linear differential equations, see [298].
For the Nosé–Hoover system
 
ẋ = −y − x z, ẏ = x, ż = α x 2 − 1 , (6.23)

which models the interaction of a particle with a heat-bath, Mahdi and Valls [323]
characterized its elementary and Darboux integrability.

Theorem 6.16 For the Nosé–Hoover system (6.23), the following statements hold.
(a) If α = 0, system (6.23) is integrable with the elementary first integrals:

(x + y)e y/(x+y) , if z = 2,
H1 = z, H2 =
(2x − 2λ1 y))λ1 (−2x + 2λ2 y))λ2 , if z = 2,

where
1  1 
λ1 = z+ z 2 − 4 , λ1 = −z + z2 − 4 .
2 2
(b) If α = 0, system (6.23) does not have a Darboux first integral. More precisely,
(b1 ) system (6.23) does not admit a proper Darboux polynomial,
(b2 ) system (6.23) has only exponential factors generated by e y and
e z +α(x +y ) with respectively the cofactors x and −2αz.
2 2 2
280 6 Applications of the Darboux Theory of Integrability

For the three-wave interaction system

ẋ = γ x + δy + z − 2y 2 ,
ẏ = γ y − δx + 2x y, (6.24)
ż = −2z − 2zx,

with γ , δ real parameters, Mahdi and Valls [324] gave a classification of system
(6.24) having a Darboux polynomial.
Theorem 6.17 Besides the Darboux polynomial z with cofactor −2(1 + x), system
(6.24) has an additional Darboux polynomial if and only if γ and δ satisfies one of
the following conditions: Moreover, system (6.24) has a rational first integral if and
only if either

Parameters Darboux polynomials cofactors

2
γ = −2, δ = 0 x 2 + y 2 + yz -4
δ
δ=0 y 2x + γ
δ
γ = 0, δ = 0 y− 2x
2
γ = −1 x 2 + y2 + z -2

• γ = 2 and δ = 0, in which case the first integral is yz, or

• γ = −1 and δ = 0, in which case the first integral is y 2 z 2 /(x 2 + y 2 + z)3 .
On the characterization of Darboux polynomials, Christodoulides and Damianou
[93] presented some properties of Darboux polynomials of the three-dimensional
Lotka–Volterra systems

ẋ1 = x1 (r x2 + sx3 ), ẋ2 = x2 (−r x1 + t x3 ), ẋ3 = x3 (−sx1 − t x2 ).

Bao and Yang [23] applied the methods developed by Llibre and Zhang in [298]
to the stretch-twist-fold system

ẋ = αz − 8x y,
ẏ = 11x 2 + 2y 2 + z 2 + βx z − 3,
ż = −αx + 2yz − βx y,

and completed the characterization of its Darboux polynomials, exponential factors,

and consequently its Darboux and algebraic integrability.
Lima et al. [241] also used the methods in [298], and proved that the Rucklidge
system

ẋ = −ax + by − yz, ẏ = x, ż = −z + y 2 ,
6.4 Applications to Concrete Models 281

has neither polynomial first integrals nor Darboux polynomials. They also obtained
some sufficient conditions on the nonexistence of analytic first integrals.
The Michelson system

x2
ẋ = y, ẏ = z, ż = c2 − y − , (6.25)
2
was derived by Michelson [332] in the study of the traveling wave solutions of
the Kuramoto–Sivashinsky equation. Llibre and Valls [281] proved that this system
admits neither a polynomial first integral nor a Darboux polynomial, and that the
unique exponential factors are eb2 x+b2 y with the cofactor b2 y + b3 z when (0, 0) =
(b2 , b3 ) ∈ C2 , and consequently the system is not Darboux integrable. Moreover,
the authors also proved the nonexistence of global analytic first integrals.
The virus model

ẋ = λ − d x − βx z, ẏ = −ay + βx z, ż = ky − uz, (6.26)

with λ, β, a, d, k, u nonnegative parameters, has been broadly applied to describe

interactions among healthy cells, infected cells and free viruses, see [349]. But it
is very difficult to characterize its dynamics for all values of the parameters. Valls
[428] investigated its integrability, and obtained the next results.

Theorem 6.18 For the virus system (6.26), the following statements hold.
• If βk = 0 and λ = d = 0, system (6.26) has the analytic first integral
 
βk  a 
H1 = x exp x+y− z .
ua k

If in addition u = a, system (6.26) is integrable with the second first integral

k  a  ua k 1
H2 = − x + y − z log x − 2 log2 x + x − z 2 .
β k 2β β 2

If u = a, system (6.26) is not completely integrable.

• If βk(λ2 + d 2 ) = 0, system (6.26) has no rational first integrals.
• If βk(λ2 + d 2 ) = 0, all Darboux polynomials of system (6.26) are generated by

Parameters Darboux polynomials Cofactors

λ=0 x −d − βz
a = d = 0 x + y − λ/d −d
u = d = 0 x + y − λ/d + (a − d)z/k -d
282 6 Applications of the Darboux Theory of Integrability

6.4.3 Concrete Higher-Dimensional Models in Applications

Ferragut and Valls [151] studied the Darboux integrability of the 5-dimensional
Chemical Reaction Network (CRN)

ẋ1 = −c1 x1 x22 + c2 x4 + c4 x5 ,

ẋ2 = −2c1 x1 x22 + c4 x5 ,
ẋ3 = c2 x4 − c3 x3 x4 , (6.27)
ẋ4 = −c2 x4 + − c3 x3 x4 ,
c1 x1 x22
ẋ5 = −c4 x5 + c3 x3 x4 ,

where c1 , c2 , c3 , c4 are positive constants. They obtained the next results.

Theorem 6.19 For system (6.27), the following statements hold.

(a) Any polynomial first integral is a polynomial function of H1 = x1 + x4 + x5 .
(b) Any proper Darboux polynomial is generated by F = c2 − c3 x3 with the
cofactor −c3 x4 .
(c) There are six exponential factors:

F1 = e x3 , F2 = e x2 −2x1 , F3 = e x1 +x4 ,
F4 = e(x2 −2x1 ) , F5 = e(2x1 −x2 )(x1 −x3 +x4 ) , F6 = e(x1 −x3 +x4 ) .
2 2

If e g/ h is another exponential factor, then h ∈ C[H1 ] and

g = a1 x3 + a2 (x2 − 2x1 ) + a3 (x1 + x4 ) + a4 (x2 − 2x1 )2

+ a5 (x2 − 2x1 )(x1 − x3 + x4 ) + a6 (x1 − x3 + x4 )2 ,

with a1 , . . . , a6 ∈ C.
(d) It has the Darboux (non-rational) first integral

H2 = F 3c2 /c3 e−(x1 +x4 ) e−(x2 −2x1 ) e x3 .

(e) It is not Darboux integrable. That is, it does not have four functionally inde-
pendent Darboux first integrals.

The static, spherically symmetric Einstein–Yang–Mills equations with a cosmo-

logical constant a ∈ R
6.4 Applications to Concrete Models 283

ṙ = r N ,
Ẇ = rU,
Ṅ = (k − N )N − 2U 2 , (6.28)
 
k̇ = s 1 − 2ar 2 + 2U 2 − k 2 ,
U̇ = sW T + (N − k)U,
Ṫ = 2U W − N T,

have been extensively studied from both dynamical and integrability points of view,
where s ∈ {−1, 1} with s = 1 when considering physical time as a temporal variable,
while s = −1 when considering physical time as a spatial variable. This system
always has the invariant algebraic surface f = 0 with
 
f = 2k N − N 2 − 2U 2 − s 1 − T 2 − ar 2 , with the associated cofactor − 2N .

We can check that system (6.28) is a homogeneous system of degree 2 when restricted
to f = 0.
Llibre and Valls [274, 275] characterized the Darboux integrability of system
(6.28) as shown in the next theorem.

Theorem 6.20 System (6.28) restricted to f = 0 has the following properties.

• It has the polynomial first integrals F = 2k N − N 2 + s(ar 2 + T 2 ) − 2U 2 and
G = W2 − rT.
• Any polynomial first integral belongs to C[F, G].
• Any rational first integral belongs to C(F, G).
• Any Darboux first integral is of the form F1c1 . . . Frcr e A/B , where Fi , A, B ∈
C[F, G] and ci ∈ C for i = 1, . . . , r .

6.4.4 Abel Equations and Foliations

In the search for Darboux polynomials and of first integrals, Cariñena et al. [58]
introduced some new ideas to study the higher-order Riccati equations and the higher-
order Abel equations. Here the mth order Riccati equations are DmR x = 0 with
 m
d
DmR x = + kx x,
dt

and the mth order Abel equations are DmA x = 0 with

 m
d
DmA x = + kx 2 x.
dt
284 6 Applications of the Darboux Theory of Integrability

Note that any general form of a Riccati (Abel) equation of order m is a superposition
of the previous Riccati (Abel) equations.
Cariñena et al. [58] show that
• the second-order Abel equation D2A x = 0 as a planar vector field

∂ ∂
v − (4kx 2 v + k 2 x 5 )
∂x ∂v

always has the Darboux polynomial v + kx 3 with cofactor −kx 2 and also the first
 −1
integral x v + kx 3 − t; and
• the third order Abel equation D3A x = 0 as a three-dimensional vector field always
has three Darboux polynomials, and two time dependent (polynomially) function-
ally independent first integrals.
For any nth order Abel equation DnA x = 0 as an n-dimensional vector field, its n Dar-
boux polynomials and n −1 time dependent (polynomially) functionally independent
first integrals are also constructed in [58] by an inductive method.
On the integrability of foliations, Galindo and Monserrat [158] in 2006 gave three
algorithms to decide whether an algebraic plane foliation has a rational first integral
and to compute it in the affirmative cases.
Pinheiro and Reis [361] show that the existence of two independent holomorphic
first integrals for foliations by curves on (C3 , 0) is not a topological invariant. The
precise results are formulated in the next theorem.

Theorem 6.21 For the foliations F and G foliated by the phase curves of the vector
fields X and Y respectively,

∂   ∂ ∂
X = 2x y + x 3 + 2y 2 − 2yz ,
∂x ∂y ∂z
  ∂   ∂   ∂
Y = x x − y − 2y 2 + y x − y − y2 − z x − y − y2 ,
∂x ∂y ∂z

the following statements hold.

(a) The foliations F and G are topologically equivalent.
(b) F 2 has 3the
 two functionally independent holomorphic first integrals H1 =
y − x z 2 and H2 = x z, while G does not have two functionally independent
first integrals.

In [361] the authors posed the following conjecture:

If two foliations F and G by curves in (C3 , 0) are topologically equivalent, and
do not admit invariant surfaces over which the induced foliation are dicritical,
then F has two holomorphic first integrals if and only if the same is true of G .
Related to this conjecture, Pinheiro and Reis [361] proved the next results.
6.4 Applications to Concrete Models 285

Theorem 6.22 Let F be a foliation by curves in (C3 , 0) with the origin being an
isolated singularity. Suppose that
• F admits two functionally independent holomorphic first integrals,
• the foliation F ∗ , obtained from F by the punctual blow-up centered at the origin,
has only isolated singularities, which are all simple.
Then the foliation F possesses an invariant surface over which the induced foliation
is dicritical.

Recall that a simple singularity is one which has at least one eigenvalue different
from zero. A foliation F by curves in (C3 , 0) restricted to an invariant surface S is
dicritical if F has infinitely many separatrices on S passing through the origin. A
separatrix of the foliation F in (C3 , 0) is a (germ of an) analytic curve passing through
the origin and invariant under the flow of a germ of the vector fields defining F .
Chapter 7
Local Integrability of Differential Systems

In Chap. 1 we proved that for k ∈ N ∪ {∞, ω}, an n-dimensional C k autonomous

differential system always has n − 1 functionally independent C k first integrals in
a neighborhood of a regular point, where the first integrals are independent of the
independent variable of the system. This chapter will concentrate on the existence of
analytic or formal first integrals of analytic differential systems in a neighborhood
of a singularity, with an emphasis on the varieties and the existence of analytic
normalizations of analytic integrable (or partially integrable) differential systems.
We will also introduce the local theory of Darboux integrability of local analytic or
formal differential systems.
Before introducing the main results we first review some properties of formal and
convergent power series rings. More detailed information can be found in Zariski
and Samuel [458, Chap. VII].
Denote by C[[x]], x = (x1 , . . . , xn ), the ring of formal series in n variables. Its
elements can be expressed as

ψ(x) = ψk (x),
k≥0

where the ψk ’s are homogeneous polynomials in x of degree k. If ψ ≡ 0, the smallest

 for which ψ ≡ 0 is called the order of ψ, denoted by o(ψ). The ring C[[x]] has
a unique maximal ideal m := {φ ∈ C[[x]] : φ0 = 0}.
A formal series ψ is multiplicatively invertible in C[[x]] if and only if ψ0 ≡ 0.
In this case o(ψ) = 0. Moreover, every nonzero noninvertible formal series can be
written as a product of irreducible series, uniquely up to the ordering of factors and
multiplication by invertible series. Mutatis mutandis, these properties carry over to
the subring C[[x]]c of C[[x]], which consists of convergent power series in some
neighborhood of 0, or analytic functions at x = 0.

© Springer Nature Singapore Pte Ltd. 2017 287

X. Zhang, Integrability of Dynamical Systems: Algebra and Analysis,
Developments in Mathematics 47, DOI 10.1007/978-981-10-4226-3_7
288 7 Local Integrability of Differential Systems

Denote by C((x)) the field of formal meromorphic functions in (Cn , 0), i.e. the
ratio of two formal series in C[[x]], and denote by C((x))c the field of meromorphic
functions in (Cn , 0), i.e. the ratio of two analytic functions.
Let ψ ∈ C[[x]] and d be a complex number. If ψ is invertible, we assume without
loss of generality that ψ0 = 1, then ψ d ∈ C[[x]] in view of the binomial series
expansion
d 
(1 + (ψ − 1)) = d
(ψ − 1)k .
k
k≥0

This series is convergent in C[[x]] in the m-adic topology, see e.g. Zariski and
Samuel [458, Chap. VIII, Sect. 2]. The m-adic
 topology is induced by the norm
||ψ|| := 2−o(ψ) for ψ ≡ 0. Hence a series γk of formal series is convergent in the
m-adic topology if and only if the orders of the γk ’s tend to infinity as k → ∞.
Consider n-dimensional smooth differential systems

ẋ = F(x), x ∈ Ω ⊂ Cn , (7.1)

with a singularity x = x0 ∈ Ω, where Ω is an open subset. Since a real vector field

can be viewed as a vector field in Cn , in what follows we only consider vector fields
in Cn . Without loss of generality we assume that x0 = 0 and that system (7.1) can
be written as  
ẋ = Ax + f (x), x ∈ Cn , 0 , (7.2)

where (Cn , 0) denotes a neighborhood of the origin in Cn , and f (0) = 0 and

∂x f (0) = 0. Recall that ∂x f denotes the Jacobian matrix of f with respect to x.
If f is analytic at x = 0, then f can be expanded as a convergent series in a neigh-
borhood of x = 0. If f is a formal series in x, we call (7.2) a formal differential
system. Of course, if f is convergent, system (7.2) is analytic.
A formal series H (x) is a formal first integral of an analytic or a formal differential
system (7.2) if

∂x H (x), F(x) ≡ 0, x ∈ (Cn , 0),

where ∂x H is the gradient of H , which is the sum of the gradients of all homogeneous
parts of H , and the equality holds in the sense that all homogeneous terms of the
same degree are identically zero.
In the next section we review the Poincaré–Dulac normal form theorem of
differential systems (7.2). For more information, we refer to Chow et al. [92], Li [230],
Ilyashenko and Yakovenko [213] and the references therein.
7.1 The Foundations of Poincaré Normal Form Theory 289

7.1 The Foundations of Poincaré Normal Form Theory

For an analytic or a formal differential system (7.2), let λ = (λ1 , . . . , λn ) be the

n-tuple of eigenvalues of A, and set

Mλ := m ∈ Zn+ |
m, λ = 0, |m| ≥ 2 ,

where Z+ is the set of nonnegative integers, m = (m 1 , . . . , m n ) and |m| = m 1 +

· · · + m n . Denote by m λ the maximal number of Q+ -linearly independent elements
in Mλ . If λ = 0 then m λ ≤ n − 1. If Mλ is not empty, we call the eigenvalues of A
resonant. Otherwise they are called nonresonant.
Set

Rλ := m ∈ Zn+ | ∃ j ∈ {1, . . . , n} such that λ j =
m, λ, |m| ≥ 2 .

If Rλ is a nonempty and finite set, we say that A satisfies finite resonant relations.

Example Assume that λ is the 3-tuple of eigenvalues of a real matrix of third order.
(a) If λ = (1, 2, −2), then all resonant relations are

1 = (2 + 1) × 1 + k × 2 + ( + k) × (−2), , k ∈ Z+ ,  + k ≥ 1,
2 = 2 × 1 + k × 2 + ( + k − 1) × (−2), , k ∈ Z+ ,  + k ≥ 1, (k, ) = (1, 0),
−2 = 2 × 1 + k × 2 + ( + k + 1) × (−2), , k ∈ Z+ ,  + k ≥ 1.

(b) If λ = (3, π, −2), then all resonant relations are

3 = (2 + 1) × 3 + 3 × (−2),  ∈ N,
−2 = 2 × 3 + (3 + 1) × (−2),  ∈ N.

(c) If λ = (1/3, 2, 5), then all resonant relations are

1
2=6× ,
3
1 1 1
5 = 3 × + 2 × 2 = 9 × + 1 × 2 = 15 × .
3 3 3
(d) If λ = (2, π, 7), then there are no resonant relations.

Note that the eigenvalues in (a) and (b) satisfy infinitely many resonant relations,
whereas the eigenvalues in (c) satisfy finitely many resonant relations.
Denote by X the vector field associated to system (7.2). Set

X = X1 + Xh ,
290 7 Local Integrability of Differential Systems

where X1 and Xh are respectively the linear and higher-order terms of the vector
field. Moreover, we have A = A1 + A2 , where A1 is semisimple and A2 is nilpotent.
Correspondingly we get
X1 = X1s + X1n ,

where X1s =
A1 x, ∂x  is the semisimple part of X1 , and X1n =
A2 x, ∂x  is the
nilpotent part of X1 . Without loss of generality we assume that


n
∂
X1s := λi xi .
i=1
∂ xi

The vector field X or system (7.2) is in the Poincaré normal form if the Lie
bracket of X1s and Xh is identically zero, i.e.

[X1s , Xh ] = 0.

By the definition of the Lie bracket the vector field X is in the Poincaré normal form
if and only if all its nonlinear monomials x k e j satisfy

λ j −
λ, k = 0, j ∈ {1, . . . , n},

where e j is the jth unit vector. This is due to the bilinearity of the Lie bracket and
the fact that
[X1s , x k ∂x j ] = (
λ, k − λ j )x k ∂x j .

Under the above conditions we call x k e j a resonant monomial of X . Hence the

vector field X or system (7.2) is in the Poincaré normal form if and only if its
nonlinear terms consist of resonant monomials.

Example Assume that the analytic differential system

ẋ = Ax + f (x), A = diag{λ1 , λ2 , λ3 }, x ∈ R3 , f (x) = O(|x|2 ),

is in the Poincaré normal form. Then the following statements hold.

• If the eigenvalues of A are λ = (1, 2, −2), then f (x) is of the form

⎛  ⎞
al,k x12l+1 x2k x3l+k
⎜ l,k∈Z+ ,l+k≥1
 ⎟
⎜ bl,k x12l x2k x3l+k−1 ⎟
f (x) = ⎜
⎜ l,k∈Z+ ,l+k≥1,(l,k)=(0,1)
⎟.
⎟
⎝  2l k l+k+1 ⎠
cl,k x1 x2 x3
l,k∈Z+ ,l+k≥1

• If the eigenvalues of A are λ = (1/3, 2, 5), then f (x) is of the form

7.1 The Foundations of Poincaré Normal Form Theory 291
⎛ ⎞
0
f (x) = ⎝ bx16 ⎠.
c1 x1 x2 + c2 x1 x2 + c3 x1
3 2 9 15

• If the eigenvalues of A are λ = (2, π, 7), then f (x) = 0.

For a given analytic or formal differential system (7.2), the Poincaré–Dulac normal
form theorem shows that there always exist an invertible formal transformation, say
y = Φ(x), under which system (7.2) can be transformed to a system in the Poincaré
normal form.
• The transformation y = Φ(x) is called a normalization, and it is taken to be tangent
to the identity or near identity. That is, Φ(x) = x+ higher-order terms.
• If the invertible transformation is only formal, we say that system (7.2) is formally
equivalent to its normal form.
• If the invertible transformation is analytic, system (7.2) is said to be analytically
equivalent to its normal form.
Generally the normalization is not necessary unique. If its nonlinear term consists
of only nonresonant monomials, we call the normalization distinguished, and corre-
spondingly the normal form is called a distinguished normal form. The distinguished
normalization is unique, and consequently the distinguished normal form is unique.
This will be seen in the proof of the Poincaré–Dulac normal form theorem.
To prove the Poincaré–Dulac normal form theorem, we need the next result, which
is due to Bibikov [33], obtained in 1979.
Lemma 7.1 Denote by G r (C) the vector space formed by the n-dimensional vector-
valued homogeneous polynomials in x of degree r with coefficients in C. Let A and
B be nth order matrices in C with eigenvalues respectively λ and κ. Define a linear
operator L on G r (C) as follows:

Lh =
∂x h, Ax − Bh, h ∈ G r (C).

Then the spectrum of L is

 
σ (L) :=
, λ − κ j   ∈ Zn+ , || = r, j = 1, . . . , n .

Proof Here the proof follows from Li et al. [233], see also Li [230]. The linear space
G r (C) has the basis

Ω := x k e j | k ∈ Zn+ , |k| = r, j = 1, . . . , n .

Since L is a linear operator on the vector space G r (C), we only need to prove that
the set of eigenvalues of L on G r (C) is σ (L).
First we assume that A and B are diagonalizable. Let S and T be invertible matrices
such that S −1 AS and T −1 BT are diagonal. Set x = Sy, then it is a linear bijection
on G r (C). Set
292 7 Local Integrability of Differential Systems

g(y) = T −1 h(Sy) =⇒ h(x) = T g(S −1 x).

Then
   
(Lh)(x)= T ∂ y g(S −1 x)S −1 , Ax − BT g(S −1 x) = T ∂ y g(y), S −1 ASy − BT g(y).

Define a linear operator

 
(L )(y) = ∂ y g(y), S −1 ASy − T −1 BT g(y), g(y) ∈ G r (C).

Then ρ is an eigenvalue of L if and only if it is an eigenvalue of L .

The above treatment allows us to assume without loss of generality that the matri-
ces A and B are diagonal, i.e.

A = diag{λ1 , . . . , λn }, B = diag{κ1 , . . . , κn }.

For an arbitrary x k e j ∈ Ω, we have

L(x k e j ) = (
, λ − κ j )x k e j .

This proves that the spectrum of L is σ (L).

Assume that A and B are not diagonalizable (at least one of them). Then there
exist diagonalizable matrices A(ε), B(ε) such that

lim A(ε) = A, lim B(ε) = B.

ε→0 ε→0

Let λ(ε) and κ(ε) be the eigenvalues of A(ε) and B(ε), respectively. Then

lim λ(ε) = λ, lim κ(ε) = κ.

ε→0 ε→0

The above proof shows that

• the linear operator

(L ε h)(x) :=
∂x h, A(ε)x − B(ε)h, h ∈ G r (C)

has the spectrum


σ (L ε ) :=
, λ(ε) − κ j (ε);  ∈ Zn+ , || = r, j = 1, . . . , n .

• lim L ε = L.
ε→0

Therefore the linear operator L has the spectrum σ (L). The proof of the lemma is
complete. 
7.1 The Foundations of Poincaré Normal Form Theory 293

Theorem 7.1 (Poincaré–Dulac normal form theorem) An analytic or a formal dif-

ferential system (7.2) can always be transformed to its distinguished normal form by
a distinguished normalization.

Proof We assume without loss of generality that A is in the lower triangular Jordan
normal form. Let
x = y + Φ(y) (7.3)

be a near identity analytic or formal diffeomorphism which sends system (7.2) to

ẏ = Ay + g(y), (7.4)

where g(y) and Φ(y) are analytic functions or formal series without the constant
and linear terms. Expand V ∈ { f, g, Φ} as the series
∞

V (z) = Vs (z),
s=2

where the Vs (z) is a vector-valued homogeneous polynomials of degree s. Then gs

and Φs satisfy


s−1
L(Φs ) :=
∂ y Φs , Ay − AΦs = [ f ]s − ∂ y Φ j (y)gs+1− j (y) − gs (y), (7.5)
j=2

where the [ f ]s ’s are the vector-valued homogeneous polynomials of degree s

obtained by re-expanding f (y + Φ(y)) as a series in y.
By Lemma 7.1, the spectrum of the linear operator L =
∂ y , Ay − A over G s (C)
is

σ (L) =
m, λ − λ j | λ = (λ1 , . . . , λn ), λ j ∈ σ (A), m ∈ Zn+ , |m| = s .

Decompose G s (C) = G0s (C) G1s (C) in such a way that the former contains only
the resonant terms and the latter consists of the nonresonant terms.
According to the decomposition of G s (C), we separate the right-hand side of
Eq. (7.5) into two parts. Then Eq. (7.5) can be written as two equations whose
right-hand sides belong to G0s (C) and G1s (C), respectively. For the equation with its
right-hand side in G0s (C), we set


s−1
gs (y) = [ f ]s − ∂ y Φ j (y)gs+1− j (y).
j=2

Then we get a corresponding solution Φs (y) = 0. For the equation with its right-
hand side in G1s (C), since L restricted to G1s (C) is invertible, setting gs (y) = 0 we
294 7 Local Integrability of Differential Systems

get a unique solution Φs (y) of the equation. This proves that Eq. (7.5) has a unique
solution with Φs nonresonant and gs resonant.
The above proof shows that the normalization (7.3) is distinguished, and system
(7.4) is a distinguished normal form of system (7.2). This finishes the proof of the
theorem. 

Remark
• If the eigenvalues of A are nonresonant, then system (7.2) is formally equivalent
to the linear system ẏ = Ay.
• If A satisfies finitely many resonant relations, then system (7.2) is formally equiv-
alent to a polynomial differential system.
• The convergence of the normalization in the Poincaré–Dulac normal form theo-
rem depends on the eigenvalues λ of A. We will not discuss this problem in gen-
eral, instead referring to Belitskii and Tkachenko [30], Bibikov [33], Ilyashenko
and Yakovenko [213], Li and Lu [235, 237], Stolovitch [412], Takens [420],
Walcher [437], Wu and Li [444], Zung [487] and the references therein.
The eigenvalues λ of A belong to the Poincaré domain if the convex hull of
λ1 , . . . , λn in C does not contain the origin of C. Otherwise the eigenvalues of A
belong to the Siegel domain.
Theorem 7.2 (Poincaré normal form theorem) If the eigenvalues of the linear part
A of the analytic differential system (7.2) belong to the Poincaré domain, then system
(7.2) is analytically equivalent to a polynomial differential system.
The proof of Theorem 7.2 needs the next result, see e.g. Li [230].
Proposition 7.1 If the eigenvalues λ of A belong to the Poincaré domain, then λ
satisfy finitely many resonant relations, that is Mλ contains finitely many elements.
Proof Since λ belong to the Poincaré domain, there exists a line  in C which
separates the n eigenvalues λ1 , . . . , λn from the origin. Let δ be the distance from the
origin to , and let k be a line passing through the origin and orthogonal to . Then
the projection δi of each λi on k is larger than δ. In addition, each resonant relation
satisfies
λ j =
λ, m.

Their projections on k satisfy

δ j = m 1 δ1 + · · · + m n δn ≥ |m|δ, (7.6)

where |m| = m 1 + · · · + m n . Clearly, the equality (7.6) holds for only finitely many
m ∈ Zn+ . This proves that λ satisfy only finitely many resonant relations. 
Here we will not present a proof of the Poincaré normal form theorem. Its proof
is similar to Theorem 7.18 in Sect. 7.5, which will be proved.
7.2 Local Analytic and Formal First Integrals 295

7.2 Local Analytic and Formal First Integrals

In this section we will discuss the existence of analytic or formal first integrals of
system (7.2) in a neighborhood of the origin.
Before introducing the general theory we illustrate an easy example. For the planar
linear differential system
ẋ = αx, ẏ = βy,

with α, β ∈ R and α, β = 0, we can check that

• if α/β = −m/n with m, n ∈ N, the planar differential system has the analytic first
integral x n y m ;
/ Q, the planar differential system has the first integral |x|β |y|−α , but it has
• if α/β ∈
no analytic first integral.
This example shows that the existence of analytic first integrals of an analytic differ-
ential system in a neighborhood of a singularity depends on the eigenvalues of the
linear part of the system at the singularity.
Let λ be the n-tuple of eigenvalues of the matrix A, the linear part of system (7.2).
Recall that 
Mλ := m ∈ Zn+ |
m, λ = 0, |m| ≥ 2 .

Note that the linearly dependent elements in the set Mλ may not be expressed in terms
of each other over Z+ . For instance, if λ = (1, 1, −2), each element of Mλ can be
expressed as a Z+ -linear combination of (1, 1, 1), (2, 0, 1) and (0, 2, 1). Clearly the
last three are Q+ -linearly dependent, but they cannot be expressed in terms of each
other over Z+ . Of course, the first one can be expressed as a Q+ -linear combination
of the other two.
One of Poincaré’s classical results [362] provided a necessary condition on the
existence of analytic first integrals of system (7.2).

Theorem 7.3 (Poincaré nonintegrability theorem) If the eigenvalues of A do not

satisfy any resonant relations, i.e. Mλ = ∅, then system (7.2) has neither analytic
nor formal first integrals in any neighborhood of the origin.

Furta [156] in 1996 gave another proof of Theorem 7.3 with the additional assump-
tion that the matrix A can be diagonalized. Shi and Li [394] in 2001 presented a
different proof of Theorem 7.3 without Furta’s additional assumption. Here we will
not prove Theorem 7.3, which can be obtained as a consequence of a general result
that will be stated and proved later on.
The condition of Theorem 7.3 implies that all eigenvalues are not equal to zero.
If A has zero eigenvalues, what can we get? In the next subsection we will partially
answer this problem.
296 7 Local Integrability of Differential Systems

7.2.1 Local Integrability via Partial Nonresonances

Li et al. [233, Theorem 1] in 2003 generalized the Poincaré nonintegrability theorem

to the case when A has a zero eigenvalue and the others are nonresonant.
Theorem 7.4 Assume that the eigenvalues λ = (λ1 , . . . , λn ) of A satisfy


n 
n
λ1 = 0, ki λi = 0, ki ∈ Z+ , ki ≥ 2.
i=2 i=2

The following statements hold.

(a) For n > 2, system (7.2) has a formal first integral in a neighborhood of x = 0 if
and only if the singularity x = 0 is not isolated. In particular, if the singularity
x = 0 is isolated, system (7.2) has no analytic first integrals in any neighborhood
of x = 0.
(b) For n = 2, system (7.2) has an analytic first integral in a neighborhood of x = 0
if and only if the singularity x = 0 is not isolated.
Proof The assumption of the theorem implies that λi = 0, i = 2, . . . , n. So the
Jordan normal form of A can be of the form diag(0, B), where B is an n − 1th
Jordan normal form. We assume without loss of generality that A = diag(0, B).
(a) By the Poincaré–Dulac normal form theorem, there exists a near identity formal
transformation x = z + ϕ(z) which sends system (7.2) to its Poincaré normal form

ż 1 = h 1 (z 1 ), ẇ = Bw + G 2 (z 1 , w), (7.7)

with z = (z 1 , w) and w = (z 2 , . . . , z n ), where G 2 = (h 2 , . . . , h n ) consist of resonant

monomials, and satisfy
G 2 (z 1 , 0) ≡ 0.

To prove statement (a), we need the next result.

Lemma 7.2 The singularity z = 0 of system (7.7) is not isolated if and only if
h 1 (z 1 ) ≡ 0.
Proof Necessity. Choose a sufficient higher cut of the normalization from system
(7.2)–(7.7), say
x = z + ψ(z),

with ψ a vector-valued polynomial of degree sufficiently high, under which system

(7.2) becomes
ż 1 =  2 (z 1 , w).
h 1 (z 1 , w), ẇ = Bw + G (7.8)

Note that  2 and G 2 , differ only in the sufficiently higher-order terms.

h 1 and h 1 , and G
Clearly the singularity x = 0 of system (7.2) is isolated if and only if the singularity
z = 0 of system (7.8) is isolated.
7.2 Local Analytic and Formal First Integrals 297

Set
h 1 (z 1 ) = cz 1m + higher-order terms.

By the nondegeneracy of the matrix B, we get from the Implicit Function Theorem
that the functional equation

2 (z 1 , w) = 0
Bw + G

has the unique solution

w = w(z 1 ) = O(z 1M ), M  m.

Substituting it into 
h 1 yields


h 1 (z 1 , w(z 1 )) = cz 1m + higher-order terms.

Since the singularity z = 0 of system (7.8) is not isolated, we must have c = 0. This
proves that h 1 (z 1 ) ≡ 0.
Sufficiency. On the contrary, we assume that the singularity x = 0 of system (7.2) is
isolated. The functional equation

B X 2 + F2 (x1 , X 2 ) = 0, where X 2 = (x2 , . . . , xn ), F2 = ( f 2 , . . . , f n ),

has a unique solution in a neighborhood of x = 0, denoted by

X 2 = X 2 (x1 ).

Substituting it into the first equation of system (7.2) yields

 
f 1 (x1 , X 2 (x1 )) = q0 x1m + O y1m+1 ,

where m ∈ N \ {1} and q0 = 0 because the singularity x = 0 is isolated. The integer

m is called the multiplicity of the singularity x = 0.
Since h 1 ≡ 0, for any k > m there exists a sufficiently higher cut x = z + ψ(z)
of the normalization from system (7.2)–(7.7), under which system (7.2) is trans-
formed to
ż 1 = O(z 1k ), ẇ = Bw + G 2 (z 1 , w).

This indicates that the multiplicity of the singularity z = 0 is at least k. This contra-
diction shows that the singularity x = 0 of system (7.2) is not isolated.
This completes the proof of the lemma. 
298 7 Local Integrability of Differential Systems

Now we continue to prove Theorem 7.4. Note that system (7.2) has a formal first
integral in a neighborhood of x = 0 if and only if system (7.7) has a formal first
integral in a neighborhood of z = 0. So we only need to prove statement (a) for
system (7.7).
Necessity. Let H (z 1 , w) be a formal first integral of system (7.7) in a neighborhood
of z = 0. Set
∞
 ∞

H (z 1 , w) = ai (w)z 1i , G 2 (z 1 , w) = Bi (w)z 1i ,
i=0 i=0

where the ai (w)’s and Bi (w)’s are formal series in w with the order of B0 (w) at
least 2. Using these expressions and equating the coefficients of z 1 of the equation
that the formal first integral satisfies, we get
 
∂a0 (w)
, Bw + B0 (w) ≡ 0. (7.9)
∂w

Set a0 (w) = c0 + ck (w) + O(|w|k+1 ), where c0 is a constant and the ck (w)’s are
homogeneous polynomials of degree k. Then we get from (7.9) that
 
∂ck (w)
, Bw ≡ 0.
∂w

Lemma 7.1 shows that the homogeneous equation in ck has only the zero solution
ck = 0. Hence a0 (w) = c0 a constant. By induction we can prove that all ai (w)’s are
constants. So H (z 1 , w) = H (z 1 ) is a function of z 1 .
That the series H (z 1 ) is a formal first integral of system (7.7) reduces to

h 1 (z 1 )∂z1 H (z 1 ) ≡ 0,

which implies
h 1 (z 1 ) ≡ 0.

By Lemma 7.2 the singularity z = 0 of system (7.7) is not isolated.

Sufficiency. Since the singularity x = 0 of system (7.2) is not isolated, we get from
Lemma 7.2 that h 1 (z 1 ) ≡ 0 in system (7.7). Hence H (z 1 , w) = z 1 is a first inte-
gral of system (7.7), and consequently system (7.2) has a formal first integral in a
neighborhood of x = 0. This proves the statement.
(b) The necessity follows from (a). We prove the sufficiency. Now x = (x1 , x2 ).
Solving λ2 x2 + f 2 (x1 , x2 ) = 0 gives a unique solution x2 = ψ(x1 ). Since the sin-
gularity x = 0 is not isolated, it forces f 1 (x1 , ψ(x1 )) ≡ 0. Taking the new coordinates

z 1 = x1 , z 2 = x2 − ψ(x1 ), (7.10)
7.2 Local Analytic and Formal First Integrals 299

system (7.2) can be written as

ż 1 = q1 (z 1 , z 2 )z 2 , ż 2 = q2 (z 1 , z 2 )z 2 , (7.11)

with q2 (0, 0) = λ2 = 0, and q1 , q2 analytic. Since z = 0 is a regular point of the

system
ż 1 = q1 (z 1 , z 2 ), ż 2 = q2 (z 1 , z 2 ), (7.12)

Theorem 1.2 shows that system (7.12) has an analytic first integral in a neighborhood
of z = 0, and so is system (7.11). The analyticity of the inverse of the transformation
(7.10) verifies that system (7.2) has an analytic first integral in a neighborhood of
x = 0.
This proves Theorem 7.4. 

Theorem 7.4 can be extended to nonlinear periodic differential systems via the
characteristic multipliers of the monodromy operators of the systems, see [233, The-
orem 5], and to the diffeomorphisms near a fixed point, see [233, Theorems 6 and 8].
Theorem 7.4 can also be extended to analytic differential systems in a neighborhood
of periodic orbits, see [233, Corollary 7].
For the planar analytic diffeomorphisms

F(x) = Ax + f (x), f (x) = O(|x|2 ),

the method in the proof of Theorem 7.4 (b) is not suitable. It remains an open problem
whether there exists a result similar to Theorem 7.4 (b) for planar analytic diffeo-
morphisms. We state it in detail as follows.
An analytic function H (x) is a first integral of the local diffeomorphism F(x) if

H (x) = H (F m (x)), for all x ∈ (C2 , 0), m ∈ N.

If H is only a formal series and satisfies this last equality, we call H a formal first
integral of the diffeomorphism F(x). Let λ1 and λ2 be the two eigenvalues of A.
Open problem Does there exist a local diffeomorphism F(x) in (C2 , 0) satisfying
λ1 = 1, |λ2 | = 1, such that x = 0 is not an isolated fixed point of F(x), and such
that F(x) has no analytic first integrals in any neighborhood of x = 0.
Next we further extend Poincaré’s result, i.e. Theorem 7.3. See [233, Theorem 3].

Theorem 7.5 Assume that system (7.2) has k ∈ {1, . . . , n − 1} functionally inde-
pendent formal first integrals H1 , . . . , Hk at the origin, and that n − k eigenvalues
λ1 , . . . , λn−k of A satisfy


n−k 
n−k
ki λi = 0, ki ∈ Z+ , ki = 0.
i=1 i=1

Then any formal first integral of system (7.2) is a formal series of H1 , . . . , Hk .

300 7 Local Integrability of Differential Systems

We stress that here the functional independence of H1 , . . . , Hk at the origin is in

the sense that the gradients of H1 , . . . , Hk at x = 0 have full rank.
Proof Taking a formal change of variables
⎛ ⎞
yk+1
⎜ .. ⎟
yi = Hi (x), i = 1, . . . , k, ⎝ . ⎠ = Bx,
yn

with B a constant matrix for which the last transformation is invertible, system (7.2)
is transformed to
Ẏk = 0, Ẏn−k = MYn−k + Fn−k (y), (7.13)

where Yk = (y1 , . . . , yk ), Yn−k = (yk+1 , . . . , yn ), Fn−k = O(|y|2 ) with y = (Yk ,

Yn−k ), and M is an n − kth constant matrix with the eigenvalues λ1 , . . . , λk .
If H (y) is a formal first integral of system (7.13), we can set
∞

H (y) = am (Yn−k )Ykm .
|m|=0

Using the arguments given in the proof of Theorem 7.4 (a), we get that all am (Yn−k )’s
are constants. Hence H (y) is a formal series in terms of Yk = (H1 , . . . , Hk ). This
proves the theorem. 
Similar to Theorem 7.5, we have
Theorem 7.6 Assume that the analytic differential system (7.2) has k ∈ {1, . . . , n −
1} functionally independent analytic first integrals H1 , . . . , Hk in a neighborhood
of the origin, and that the Q+ -linear space generated by the set Mλ = {m ∈
Zn+ |
m, λ = 0, |m| ≥ 2} is of dimension k. Then any analytic first integral of system
(7.2) is an analytic function of H1 , . . . , Hk in a neighborhood of the origin.
We remark that the functional independence of H1 , . . . , Hk in this last theorem is
also in the sense that their gradients at the origin are linearly independent. The same
is applied to the following Theorem 7.7.
Theorem 7.6 and its proof can be found in Zhang [464, Theorem 1]. Kwek et al.
[223, Theorem A] also provided a proof of Theorem 7.6 with the restriction that A
is diagonalizable.
Theorem 7.6 has many applications to the local integrability of dynamical systems,
see e.g. Blé et al. [36] and Llibre and Valls [273, 275, 283, 284] and the references
therein.
The next result, due to Zhang [464, Theorem 2], extends the Poincaré noninte-
grability theorem and Theorem 7.4. Set
 
 
n
n−1 
Mλ := k = (k2 , . . . , kn ) ∈ Z+ ki λi = 0 .
i=2
7.2 Local Analytic and Formal First Integrals 301

Theorem 7.7 Assume that the analytic differential system (7.2) has m ∈ {1, . . . , n −
2} functionally independent analytic first integrals Φ1 , . . . , Φm in a neighborhood of
the origin, and that λ1 = 0 and the Q+ -linear space generated by Mλ is of dimen-
sion m. If the characteristic subspace associated to λ1 is tangent to the (n − m)-
dimensional space S = {Φ1 (x) = 0} ∩ . . . ∩ {Φm (x) = 0}, the following statements
hold.
(a) For m < n − 2, system (7.2) has the formal first integral
∞

H (x) = h s (x)Φ1s1 (x) . . . Φmsm (x), s = (s1 , . . . , sm ) ∈ Zm
+, (7.14)
|s|=1

in a neighborhood of the origin, with the h s (x)’s not all constants, if and only
if the singularities of system (7.2) form a surface which is tangent to S at the
origin.
(b) For m = n − 2, system (7.2) has an analytic first integral of the form (7.14)
in a neighborhood of the origin, with the h s (x)’s not all constants, if and only
if the singularities of system (7.2) form a surface which is tangent to S at the
origin.

7.2.2 Local Integrability via Resonances

We will show in this subsection that the number of functionally independent analytic
or formal first integrals depends on the number of resonances of the eigenvalues of
the linear part of an analytic or a formal differential system.
In 2008, Chen et al. [82, Theorem 1.1] improved the Poincaré nonintegrability
theorem taking into account the number of resonances. Consider the quasiperiodic
vector field

θ̇ = ω + Ω(θ, x),
(θ, x) ∈ Fm × Fn , F = R or C, (7.15)
ẋ = Ax + f (θ, x),

where Ω = O(x), and f = O(x2 ) is a vector-valued analytic function periodic

in θ of period 2π . Denote by X the vector field associated to system (7.15), i.e.

X =
ω + Ω(θ, x), ∂θ  +
Ax + f (θ, x), ∂x .

A nonconstant function H (θ, x) is an analytic first integral (or formal first inte-
gral) of the vector field X if H is an analytic function (or a formal series) periodic
in θ of period 2π , and the derivative of H along the vector field X is identically
equal to zero, i.e. X (H ) ≡ 0.
Let λ = (λ1 , . . . , λn ) be the n-tuple of eigenvalues of the matrix A, and let ρ be
the number of Q+ -linearly independent elements of the set
302 7 Local Integrability of Differential Systems

Rω,λ := {(k, )| i

k, ω +
, λ = 0, k ∈ Zm ,  ∈ Zn+ }.

Theorem 7.8 The number of functionally independent analytic first integrals of the
analytic vector field (7.15) in a neighborhood of the constat solution x = 0 is less
than or equal to ρ.
We remark that ρ is the minimal upper bound of the functionally independent
analytic first integrals that the vector field (7.15) can have. If ρ = 0, Theorem 7.8 is
simply the Poincaré nonintegrability theorem, i.e. Theorem 7.3.
As a consequence of Theorem 7.8 we have
Theorem 7.9 The number of functionally independent analytic first integrals of the
analytic autonomous vector field (7.15) in a neighborhood of x = 0 is less than or
equal to the number of Q+ -linearly independent elements of

Mλ := { ∈ Zn+ |
, λ = 0}.

The next example exhibits a relation in Theorem 7.8 between the numbers of
functionally independent first integrals and of resonances. Consider the quasiperiodic
differential system

θ̇1 = ω1 , θ̇2 = ω2 , ẋ1 = 3x1 − 2x2 x32 , ẋ2 = −ω3 ix2 , ẋ3 = ω3 ix3 , (7.16)
√
where i = −1, and the frequencies ω1 , ω2 do not satisfy the relation: k1 ω1 +
k2 ω2 = 0 for k1 , k2 ∈ Z and |k1 | + |k2 | = 0. If

k1 ω1 + k2 ω2 + k3 ω3 = 0, for all k1 , k2 , k3 ∈ Z, |k1 | + |k2 | + |k3 | = 0,

then Rω,λ has the basis (0, 0, 0, 1, 1). Hence the function x2 x3 is a generator of the
first integrals of the vector field (7.16). If

ω3 = m 1 ω1 + m 2 ω2 , for some m 1 , m 2 ∈ Z,

then Rω,λ has a basis formed by (m 1 , m 2 , 0, 1, 0) and (−m 1 , −m 2 , 0, 0, 1). Hence

any analytic first integral of the vector field (7.16) is an analytic function of
ei(m 1 θ1 +m 2 θ2 ) x2 and e−i(m 1 θ1 +m 2 θ2 ) x3 .
To prove Theorem 7.8, we need some preparation. Similar to the autonomous
differential systems (7.2), we introduce the normal form for the quasiperiodic vector
field X or its associated quasiperiodic differential system (7.15).
The n-tuple of eigenvalues λ of the matrix A is nonresonant if for all k ∈ Zm ,
 ∈ Zn+ , || > 1, the following conditions hold.

, λ = −i
k, ω ,
, λ − λ j = −i
k, ω , j = 1, . . . , n. (7.17)

The n-tuple of eigenvalues λ are weakly nonresonant if the condition (7.17) holds
except for k = 0.
7.2 Local Analytic and Formal First Integrals 303

Set

X = X1 + X h , X1 =
ω, ∂θ  +
Ax, ∂x , Xh higher-order terms.

Decompose A into

A = A1 + A2 , with A1 semisimple, A2 nilpotent.

Correspondingly

X1 = X1s + X1n , X1s =

ω, ∂θ  +
A1 x, ∂x , X1n =
A2 x, ∂x .

Without loss of generality, we assume that

X1s =
ω, ∂θ  +
λx, ∂x ,

where λx = (λ1 x1 , . . . , λn xn ).
The vector field X or system (7.15) is in normal form if the Lie bracket of X1s
and Xh is zero, i.e.
[X1s , Xh ] = 0.

Note that if system (7.15) is in normal form, all its pseudomonomials ei

k,θ x  are
resonant. That is, if ei
k,θ x  is in the component ∂θ j , then i
k, ω +
, λ = 0, called
first resonant; if ei
k,θ x  is in the component ∂x j , then i
k, ω +
, λ = λ j , called
second resonant.
A pseudomonomial ei
k,θ x  in an analytic or a formal quasiperiodic function
is resonant if i
k, ω +
, λ = 0. The reader should take care of the difference
between the resonances of vector fields and of functions.
Recall that a near identity transformation reducing the vector field X to its normal
form is called a normalization. Note that the normalization reducing a quasiperiodic
vector field to its normal form is not unique in general. A normalization is dis-
tinguished if it consists of nonresonant terms. A normal form obtained through a
distinguished normalization is called a distinguished normal form.
The next result, due to Chen et al. [82, Lemma 2.2], is a generalization of the
Poincaré–Dulac normal form theorem, i.e. Theorem 7.1, from autonomous differen-
tial systems to quasiperiodic differential systems, which provides the existence and
the uniqueness of the distinguished normal form.

Theorem 7.10 Any analytic or formal vector field X (or its associated differential
system (7.15)) can always be reduced to its normal form through a distinguished
normalization. Moreover, the distinguished normalization and the normal form are
unique.

Proof Let
θ = ϑ + ϕ(ϑ, y), x = y + ψ(ϑ, y),
304 7 Local Integrability of Differential Systems

be a transformation which sends the vector field X to

ϑ̇ = ω + Λ(ϑ, y), ẏ = Ay + g(ϑ, y),

where Λ, ϕ = O(y) and g, ψ = O(y2 ) are 2π periodic functions in ϑ. Expand

the nonlinear functions w ∈ {Λ, g, ϕ, ψ, Ω, f } as Taylor series in y

w(ϑ, y) = wr (ϑ, y),
r

where the wr ’s are homogeneous polynomials in y of degree r with the coefficients

2π periodic functions in ϑ. Then we have

∂ϑ ϕr , ω +
∂ y ϕr , Ay = Ωr − Λr − pr ,
r = 1, 2, . . .

∂ϑ ψr +1 , ω +
∂ y ψr +1 , Ay − Aψr +1 = fr +1 − gr +1 − qr +1 ,
(7.18)
where pr , qr +1 are successively determined with pr a polynomial in ϕs , Λs , gs+1 ,
s = 1, . . . , r − 1; and qr +1 a polynomial in ψs , Λs−1 , gs , s = 2, . . . , r .
Expand wr as a Fourier series in ϑ

wr (ϑ, y) = Vrk (y)ei
k,ϑ , w ∈ {Λ, g, ϕ, ψ, Ω, f }.
k∈Zm

The Eq. (7.18) are reduced to

K0 ϕrk = Ωrk − Λrk − prk ,

r = 1, 2, . . . (7.19)
K1 ψrk+1 = frk+1 − grk+1 − qrk+1 ,

where

K0 = i
k, ϑ + L0 , L0 h(y) =
∂ y h, Ah, h ∈ G r (y)
K1 = i
k, ϑ + L1 , L1 h(y) =
∂ y h, Ah − Ah, h ∈ G r +1 (y).

By Lemma 7.1 the spectrums of K0 and K1 are respectively

{i
k, ϑ +
, λ;  ∈ Zn+ , || = r },
{i
k, ϑ +
, λ − λ j ;  ∈ Zn+ , || = r + 1, j = 1, . . . , n}.

According to the operator K0 (resp. K1 ), we separate the space G r (y) into the
direct sums
r +1 r +1
G r (y) = G0,1
r
(y) ⊕ G0,2
r
(y) (resp. G r +1 (y) = G1,1 (y) ⊕ G1,2 (y)),
7.2 Local Analytic and Formal First Integrals 305

such that K01 := K0 |G0,1 r

(y) is invertible, whereas K0 := K0 |G0,2
2 r
(y) is degenerate
(that is, K0 has only the zero spectrum). Meanwhile, K1 := K1 |G1,1
2 1 r +1
(y) is invertible,
and K12 := K1 |G1,2r +1
(y) has only the zero spectrum. Correspondingly we separate the
right-hand sides Ωr − Λr − pr and fr +1 − gr +1 − qr +1 of (7.19) respectively as
k k k k k k

   
Ωrk,1 − Λrk,1 − prk,1 + Ωrk,2 − Λrk,2 − prk,2 ∈ G0,1
r
(y) ⊕ G0,2
r
(y),
   
r +1 r +1
frk,1 k,1 k,1 k,2 k,2 k,2
+1 − gr +1 − qr +1 + f r +1 − gr +1 − qr +1 ∈ G1,1 (y) ⊕ G1,2 (y).

Then Eq. (7.19) can be written as

K0s ϕrk,s = Ωrk,s − Λrk,s − prk,s ,

r = 1, 2, . . . , s = 1, 2. (7.20)
K1s ψrk,s k,s k,s k,s
+1 = f r +1 − gr +1 − qr +1 ,

For s = 1, by the invertibility of the operators K01 and K11 systems (7.20) have
a unique solution for any given Λrk,s and grk,s
+1 . So we choose

Λrk,1 = grk,1
+1 = 0,

and the unique solutions of Eq. (7.20) are denoted by ϕrk,1 and ψrk,1
+1 , which are
nonresonant.
When s = 2, we set

Λrk,2 = Ωrk,2 − prk,2 , grk,2 k,2 k,2

+1 = f r +1 − qr +1 .

Then Eq. (7.20) have the solutions ϕrk,2 = ψrk,2

+1 = 0.
The above process provides a distinguished formal normalization
 
θ =ϑ+ ϕrk,1 (y)ei
k,ϑ , x =y+ ψrk,1 (y)ei
k,ϑ ,
+ , r ≥1
k∈Zm + , r ≥2
k∈Zm

which transforms the vector field X to its distinguished formal normal form
 
ϑ̇ = ω + Λrk,2 (y)ei
k,ϑ , ẏ = Ay + grk,2 (y)ei
k,ϑ .
k∈Zm , r ≥1 k∈Zm , r ≥2

This finishes the proof of the theorem. 

Theorem 7.10 has an easy consequence.

Corollary 7.1 For the vector field X , the following statements hold.
(a) If the n eigenvalues of A are not resonant, then X is formally equivalent to
X1 .
306 7 Local Integrability of Differential Systems

(b) If the n eigenvalues of A are weakly nonresonant, then X is formally equivalent

to an autonomous vector field.
Proof (a) The proof is easy.
(b) By the assumption Eq. (7.19) with k = 0 has a unique solution for any choices
of Λrk and grk+1 . We choose Λrk = grk+1 = 0 for k = 0. The statement follows. 
To prove Theorem 7.8 we also need the next result, see [82, Lemma 2.3].
Lemma 7.3 Assume that H (θ, x) is an analytic or a formal first integral of the
vector field X , and is 2π periodic in θ . Let Y be the distinguished normal form
vector field of X , and K (ϑ, y) be the expression of H (θ, x) in the coordinates (ϑ, y)
given in the proof of Theorem 7.10. Then
(a) K (ϑ, y) is a first integral of Y .
(b) K (ϑ, y) consists of the resonant terms. That is, if we expand K as a Fourier
series 
K (ϑ, y) = K (k,) y  ei
k,ϑ ,
k∈Zm , ∈Zn+

we have i
k, ω +
, λ = 0.
Proof (a) This follows from Proposition 1.2.
(b) To simplify the notation we assume without loss of generality that X is in the
distinguished normal form. Expand H as a Taylor series in x
∞

H (θ, x) = Hs (θ, x),
s=r

where the Hs ’s are homogeneous polynomials in x of degree s, s = r, r + 1, . . .,

with Hr ≡ 0. Then we have


s−r
 
L Hs = −
∂θ Hs− j , Ω j  +
∂x Hs− j , f j+1  , s = r, r + 1, . . . (7.21)
j=1

where L is the linear operator defined by

L Hs =
∂θ Hs , ω +
∂x Hs , Ax.

Equation (7.21) with s = r is linear homogeneous. From the proof of Theorem

7.10 we get that the nontrivial solution of Eq. (7.21) with s = r must consist of
resonant monomials.
Some calculations show that Eq. (7.21) with s = r + 1 has its right-hand side
formed by the first resonant pseudomonomials. Hence the solution Hs+1 of Eq. (7.21)
must consist of the first resonant pseudomonomials.
By induction we can prove that all solutions Hs of Eq. (7.21) consist of resonant
terms. This proves the lemma. 
7.2 Local Analytic and Formal First Integrals 307

Proof of Theorem 7.8. By Lemma 7.3 and its proof, we assume without loss of gener-
ality that the vector field X is in the distinguished normal form, and the functionally
independent analytic or formal first integrals are F1 , . . . , Fq . Since F j , j = 1, . . . , q,
consist of the resonant monomials, it follows that X1s (F j ) = 0. That is, each F j is
also a first integral of X1s . Clearly, all analytic and formal first
  i
k,θ
s
integrals of X1 are
generated by x e | i
k, ω +
, λ = 0,  ∈ Z+ , k ∈ Z , which has the max-
m n

imum number ρ of functionally independent elements. This proves that X has no

more than ρ functionally independent analytic and formal first integrals in a neigh-
borhood of the origin. The proof of the theorem is complete. 

7.3 Local (Formal) Meromorphic First Integrals

This section turns to the study of the functionally independent meromorphic and
formal meromorphic first integrals of the analytic or formal differential system (7.2)
in a neighborhood of the singularity x = 0.
In 2007, Shi [395, Theorem 1] generalized Theorem 7.3 and obtained a necessary
condition for system (7.2) to have a rational first integral.

Theorem 7.11 If system (7.2) has a rational first integral, there exists a 0 = k ∈ Zn
such that
λ, k = 0.

Cong et al. [117, Theorem 1] in 2011 extended the Poincaré nonintegrability

theorem and Shi’s result to a more general case.
A function H (x) is a (formal) meromorphic first integral of system (7.2) if H (x) ∈
C((x))c (H (x) ∈ C((x))) and it is a first integral of the system.
The eigenvalues λ of the matrix A are Z-resonant if there exists a k ∈ Zn , k = 0,
such that

λ, k = 0.

Note that the Z-resonance is different from the previous resonance in the sense that
here we permit k to be negative integers.

Theorem 7.12 Let λ = (λ1 , . . . , λn ) be the eigenvalues of A. The number of func-

tionally independent (formal) meromorphic first integrals of system (7.2) in (Cn , 0) is
no greater than the dimension of the Q-linear space generated by {k ∈ Zn |
k, λ =
0, k = 0}.

We remark that the methods in the proof of Theorems 7.3, 7.8 and 7.11 are not
sufficient to prove Theorem 7.12. Here we present some other techniques to treat
this problem, which were given in Cong et al. [117].
If the eigenvalues of A are all zero, Theorem 7.12 is trivial. This case was discussed
for semi-quasihomogeneous vector fields in Cong et al. [117, Theorem 1].
308 7 Local Integrability of Differential Systems

7.3.1 The Equivalence Between Algebraic and Functional

Independence

The proof of Theorem 7.12 needs Proposition 5.4, that is, the next result.
Lemma 7.4 The functions F1 (x), . . . , F (x) ∈ C(x) are algebraically independent
if and only if they are functionally independent.
Proof The proof follows from that of Cong et al. [117, Lemma 5] with ideas from
Ito [214, Lemma 9.1].
Sufficiency. On the contrary, if F1 (x), . . . , F (x) are algebraically dependent, there
exists a minimal polynomial P(z 1 , . . . , z  ) with complex coefficients such that

P(F1 (x), . . . , F (x)) ≡ 0.

Then we have

∂x j P(F1 (x), . . . , F (x)) ≡ 0, j = 1, . . . , n.

This is equivalent to
⎛ ⎞⎛ ⎞
∂x1 F1 (x) . . . ∂x1 F (x) ∂z1 P(F1 (x), . . . , F (x))
⎜ .. .. .. ⎟⎜ .. ⎟
⎝ . . . ⎠⎝ . ⎠ ≡ 0. (7.22)
∂xn F1 (x) . . . ∂xn F (x) ∂z P(F1 (x), . . . , F (x))

Since P(z 1 , . . . , z  ) is minimal, and ∂zs P(z 1 , . . . , z n ) = 0 for at least one s ∈

{1, . . . , n}, this implies that at least one of the equalities

∂z1 P(F1 (x), . . . , F (x)) ≡ 0, . . . , ∂z P(F1 (x), . . . , F (x)) ≡ 0

cannot hold. Hence the rank of the n ×  matrix in (7.22) is less than . Conse-
quently F1 (x), . . . , F (x) are functionally dependent, a contradiction. This proves
the algebraic independence of F1 (x), . . . , F (x).
Necessity. Since F1 , . . . , F are algebraically independent, it follows that C(F1 , . . . ,
F ) is a finite field extension over C of transcendental degree  (see e.g. Hartshorne
[199, Theorem 4.8A]). By the extension of the derivatives over field extensions (see
e.g. [224, Theorem 5.1, p. 370]), there exist  derivatives D j ( j = 1, . . . , ) over
C(F1 , . . . , F ) satisfying

0, j = s,
D j Fs = δ js =
1, j = s.

By the algebraic independence of F1 , . . . , F , one gets that C(x) is a finite

field extension over C(F1 , . . . , F ) of transcendental degree n − . It follows that
7.3 Local (Formal) Meromorphic First Integrals 309

there exist n derivatives Y1 , . . . , Yn over C(x) such that Y j = D j , j = 1, . . . , ,

when restricted to C(F1 , . . . , F ). Furthermore, all derivatives
 over C(x) form an
∂ 
n-dimensional vector space over C(x) with a basis { ∂ x j  j = 1, . . . , n}. Hence we
have
n
∂
Yj = q jν ,
ν=1
∂ xν

with q jν ∈ C(x). The derivative Ys acting on C(F1 , . . . , F ) gives


n
∂ Fs
δ js = D j Fs = Y j Fs = q jν , j, s ∈ {1, . . . , }.
ν=1
∂ xν

This proves that ∂x F1 , . . . , ∂x F have rank , and consequently F1 , . . . , F are func-

tionally independent. This completes the proof of the lemma. 

7.3.2 The Lowest Order Parts of Functionally Independent

First Integrals

Let F(x) be an analytic or a polynomial function defined in (Cn , 0). Denote by

F 0 (x) the lowest order homogeneous term of F(x). Let F(x) = G(x)/H (x) be a
meromorphic or a formal meromorphic function defined in (Cn , 0). Denote by F 0 (x)
the rational function G 0 (x)/H 0 (x). Expand the analytic functions G(x) and H (x)
as Taylor series
∞
 ∞

G 0 (x) + G s (x) and H 0 (x) + H s (x),
s=1 s=1

where G s (x) and H s (x) are the homogeneous polynomials of degrees deg G 0 (x) + s
and deg H 0 (x) + s, respectively. Then F can be written as
∞
G 0 (x)  V s (x)
F(x) = + , (7.23)
H 0 (x) s=1 W s (x)

where V s (x) and W s (x) are homogeneous polynomials. Some easy calculations
show that

deg G 0 (x) − deg H 0 (x) < deg V s (x) − deg W s (x) for all s ≥ 1.

Denote by deg V s (x) − deg W s (x) the degree of the rational homogeneous function
V s (x)/W s (x), and set
310 7 Local Integrability of Differential Systems

o(G) = deg G 0 (x), o(F) = o(G) − o(H ) = deg G 0 (x) − deg H 0 (x).

Recall that o(F) is the order of F.

Lemma 7.5 Assume that

F1 (x) = G 1 (x)/H1 (x), . . . , Fm (x) = G m (x)/Hm (x)

are functionally independent (formal) meromorphic first integrals of the analytic

differential system (7.2) in (Cn , 0). Then there exist polynomials Pi (z 1 , . . . , z m ),
i = 2, . . . , m, such that

R1 = F1 (x), R2 (x) = P2 (F1 (x), . . . , Fm (x)), . . . , Rm (x) = Pm (F1 (x), . . . , Fm (x))

are functionally independent (formal) meromorphic first integrals, and

R10 (x), R20 (x), . . . , Rm0 (x)

are functionally independent rational functions.

Proof This result was first proved by Ziglin [481] in 1983. In 1996 Baider et al. [22]
provided another proof. Here the proof follows from Cong et al. [117, Lemma 6]
with ideas from Ito [214, Lemma 2.1].
If F10 (x), . . . , Fm0 (x) are functionally independent, the proof is done.
We assume without loss of generality that F10 (x), . . . , Fk0 (x) for some k ∈
{1, . . . , m − 1} are functionally independent, whereas F10 (x), . . . , Fk+1
0
(x) are func-
tionally dependent. Lemma 7.4 shows that F1 (x), . . . , Fk+1 (x) are algebraically
0 0

dependent. So there exists a minimal polynomial P(z), z = (z 1 , . . . , z k+1 ), such that

P(F10 (x), . . . , Fk+1

0
(x)) ≡ 0 and ∂zk+1 P(z) ≡ 0.

Since F1 (x), . . . , Fm (x) are functionally independent, it follows that the matrix

∂(F1 (x), . . . , Fk+1 (x))

(7.24)
∂(x1 , . . . , xn )

has rank k + 1. Let

∂(F1 (x), . . . , Fk+1 (x))
M :=
∂(xi1 , . . . , xik+1 )

be a (k + 1) × (k + 1) minor of (7.24), with a nonvanishing determinant M. Denote

by M0 the lowest order rational term of M. Then

∂(F10 (x), . . . , Fk+1

0
(x))
M0 = detM 0 := det .
∂(xi1 , . . . , xik+1 )
7.3 Local (Formal) Meromorphic First Integrals 311

Define

k+1
ν(F1 , . . . , Fk+1 ) := o(M) + k + 1 − o(Fs ).
s=1

Since the degree of the lowest order term of

 
∂x j Fi = Hi ∂x j G i − G i ∂x j Hi (Hi )−2

is greater than or equal to o(Fi ) − 1 (the former happens in the case Hi0 ∂x j G i0 −
G i0 ∂x j Hi0 ≡ 0), it follows that

ν(F1 , . . . , Fk+1 ) ≥ 0.

Note that ν(F1 , . . . , Fk+1 ) = 0 if and only if o(M0 ) = o(M), and if and only if
M0 = 0, which is equivalent to the functional independence of F10 (x), . . . , Fk+1
0
(x).
Hence we have
ν(F1 , . . . , Fk+1 ) > 0.

Set
Rk+1 (x) := P(F1 (x), . . . , Fk+1 (x)).

Then F1 (x), . . . , Fk (x), Rk+1 (x) are functionally independent. Moreover, we claim
that
ν(F1 , . . . , Fk , Rk+1 ) < ν(F1 , . . . , Fk+1 ).

Indeed, write the polynomial P as

 ρ ρ
P(z) = pρ z ρ , z ρ = z 1 1 · . . . · z k+1
k+1
, ρ = (ρ1 , . . . , ρk+1 ) ∈ (Z+ )k+1 ,
ρ

and define

μ = min{
ρ, (o(F1 ), . . . , o(Fk+1 )| pρ = 0, ρk+1 = 0}.

Since Rk+1 (x) contains Fk+1 , and P(F10 (x), . . . , Fk+1

0
(x)) ≡ 0, it follows that μ <
o(Rk+1 ). Hence the lowest order term of P(z) must contain Fk+1 (x). This implies
that the lowest order term of Rk+1 (x) has degree larger than that of P(z). Set

 := det(∂(F1 (x), . . . , Fk (x), Rk+1 (x))/∂(xi1 , . . . , xik , xik+1 ).

M

We have
 
 = o(M) + o ∂zk+1 P(F1 , . . . , Fk+1 ) = o(M) + μ − o(Fk+1 ).
o(M)
312 7 Local Integrability of Differential Systems

Hence


k
 +k+1−
ν(F1 , . . . , Fk , Rk+1 ) = o(M) o(F j ) − o(Rk+1 )
j=1

= ν(F1 , . . . , Fk+1 ) + μ − o(Rk+1 ) < ν(F1 , . . . , Fk+1 ).

This proves the claim.

If ν(F1 , . . . , Fk , Rk+1 ) = 0, then F10 (x), . . . , Fk0 (x), Rk+10
(x) are functionally
independent.
If ν(F1 , . . . , Fk , Rk+1 ) > 0, then F10 (x), . . . , Fk0 (x), Rk+1
0
are functionally depen-
dent. Continuing the above process, we obtain a polynomial P(z),  z = (z 1 , . . . , z k+1 ),
such that
k+1 (x) = P(F
F1 (x), . . . , Fk (x), R  1 (x), . . . , Fk+1 (x))

are functionally independent, and ν(F1 , . . . , Fk , R k+1 ) = 0. This means that the
0 0
rational functions F1 , . . . , Fk , Rk+1 are functionally independent.
0

By induction and the functional independence of the formal rational first integrals

k+1 (x), Fk+2 (x), . . . , Fm (x),

F1 (x), . . . , Fk (x), R

we obtain a proof of the lemma. The details are left to the reader as an exercise. 

7.3.3 Proof of the Results on Functionally Independent

First Integrals

In this subsection we will prove Theorem 7.12, which is taken from [117]. We
call the ratio of two monomials a rational monomial. A rational monomial x k /x  ,
k,  ∈ (Z+ )n , is resonant if
λ, k −  = 0, where λ is the eigenvalues of A. A rational
function is homogeneous if its numerator and denominator are both homogeneous
polynomials. A rational homogeneous function is resonant if the ratio of any two
elements in the set of all monomials in the numerator and denominator is a resonant
rational monomial.
For the vector field X associated to system (7.2), set

X = X1 + Xh :=
∂x , Ax +
∂x , f (x).

Lemma 7.6 If F(x) = G(x)/H (x) is a formal meromorphic first integral of the
vector field X , then
F 0 (x) = G 0 (x)/H 0 (x)
7.3 Local (Formal) Meromorphic First Integrals 313

is a resonant rational homogeneous first integral of the linear vector field X1 , where
F 0 is not a constant, otherwise if F 0 (x) ≡ a ∈ C, we can consider (F − a)0 .

Proof Write F(x) as

∞

F(x) = F 0 (x) + F i (x),
i=1

where F 0 (x) is the lowest order rational homogeneous function, and the F i (x)’s,
i ∈ N, are rational homogeneous functions of degree larger than deg(F 0 (x)). Since

X (F(x)) ≡ 0, x ∈ (Cn , 0),

we get that
X1 (F 0 (x)) ≡ 0. (7.25)

We claim that F 0 (x) is resonant. Indeed, without loss of generality we assume

that G 0 (x) and H 0 (x) are relative prime, and write Eq. (7.25) as
   
H 0 (x) ∂x G 0 (x), Ax − G 0 (x) ∂x H 0 (x), Ax ≡ 0.

Then there exists a constant c such that

Lc (G 0 (x)) ≡ 0, Lc (H 0 (x)) ≡ 0, (7.26)

where
Lc =
∂x , Ax − c.

It has the spectrum over G p (C) with p = degG 0 (x)

S p := {
, λ − c|  ∈ (Z+ )n , || = p},

and the spectrum over G q (C) with q = degH 0 (x)

Sq := {
m, λ − c| m ∈ (Z+ )n , |m| = q}.

So the solution G 0 (x) of the first equation of (7.26) consists of the monomials
x  satisfying
, λ − c = 0, whereas the solution H 0 (x) of the second equation of
(7.26) consists of the monomials x m satisfying
m, λ − c = 0. These together verify
that F 0 (x) = G 0 (x)/H 0 (x) is a resonant rational homogeneous first integral of X1 .
This proves the claim, and consequently the lemma. 

Proof of Theorem 7.12. Let

F1 (x) = G 1 (x)/H1 (x), . . . , Fm (x) = G m (x)/Hm (x)

314 7 Local Integrability of Differential Systems

be m functionally independent formal meromorphic first integrals of the vector field

X . By Lemma 7.5 we can assume without loss of generality that

F10 (x) = G 01 (x)/H10 (x), . . . , Fm0 (x) = G 0m (x)/Hm0 (x)

are functionally independent. Lemma 7.6 shows that F10 (x), . . . , Fm0 (x) are resonant
rational homogeneous first integrals of the linear vector field X1 . Decompose X1
into semisimple and nilpotent parts

X1 = X1s + X1n , X1s =

∂x , As x, X1n =
∂x , An x,

where As and An are the semisimple and the nilpotent parts of A. Then F10 (x), . . . ,
Fm0 (x) must be first integrals of X1s . This implies that m is less than or equal to the
number of functionally independent resonant rational monomials, which is equal to
the maximal number of Q-linearly independent elements of {k ∈ Zn :
k, λ = 0}.
This completes the proof of the theorem. 
Applying Theorem 7.12 and its proof, we can extend the result of Theorem 7.12
to periodic differential systems (see [117, Theorem 4] and [215, Theorem 1]), to
analytic differential systems near a periodic orbit (see [117, Theorem 3]), and also
to semi-quasihomogeneous differential systems (see [117, Theorem 2]).
On the existence of meromorphic first integrals, Li et al. [234] presented an equiv-
alent characterization for planar analytic differential systems around an elementary
singularity.

7.4 The Local Theory of Darboux Integrability

In this section we further extend the results on local analytic and meromorphic first
integrals to local Darboux first integrals. The main results were proved by Llibre
et al. [294] in 2014.
Assume that system (7.2) is an analytic or a formal one. As before, denote by X
the vector field associated to system (7.2).
A noninvertible element ψ ∈ C[[x]] \ {0} is a formal Darboux semi-invariant
of system (7.2) or of the vector field X if there exists a formal series K , called a
cofactor, such that
X (ψ) = K ψ.

If ψ and K are convergent in a neighborhood of the origin, we call ψ a local analytic

Darboux semi-invariant.
Similar to the Darboux polynomials of the polynomial differential systems, we
have the next results.
Proposition 7.2 For the analytic or formal differential system (7.2), the following
statements hold.
7.4 The Local Theory of Darboux Integrability 315

(a) Each irreducible factor of a formal Darboux semi-invariant is also a formal

Darboux semi-invariant.
(b) A noninvertible analytic function is a local analytic Darboux semi-invariant of
X if and only if its zero set is invariant under the flow of X .
(c) Each invertible series is automatically a formal Darboux semi-invariant.
(d) Multiplying a formal Darboux semi-invariant by an invertible series will yield
another formal Darboux semi-invariant.
Proof Statement (a) follows from the unique factorization theorem of the formal
series. Statement (b) can be proved by the Hilbert–Rückert Nullstellensatz (see
Shafarevich [391]). Statements (c) and (d) follow from some easy
calculations. 
We remark that in statement (c) the invertible analytic functions are of no geo-
metric interest in view of statement (b), because their local zero set at the origin is
empty.

7.4.1 Local Darboux First Integrals

For the formal series or analytic functions ψ1 , . . . , ψr , and nonzero σ1 , . . . , σr ∈ C, if

 
X ψ1σ1 · · · ψrσr = 0,

then H := ψ1σ1 · · · ψrσr is called a local Darboux first integral of X . Of course, if all
the σi ’s are integers then H is a formal meromorphic first integral.
Similar to the classical Darboux first integral, we have the next result.
Proposition 7.3 Assume that the ψi ’s, i = 1, . . . , r , are formal Darboux semi-
invariants with cofactors K i . Then ψ1σ1 · · · ψrσr is a local Darboux first integral of
X if and only if
σ1 · K 1 + · · · + σr · K r = 0.

The next result characterizes some important properties of formal Darboux semi-
invariants and of formal Darboux first integrals, which is the main result of this
section.
Theorem 7.13 For system (7.2) or its associated vector field X , let λ = (λ1 , . . . , λn )
be the eigenvalues of A, and let Xs be the semisimple vector field of the linear part
of X . The following statements hold.
(a) If ψ is a formal Darboux semi-invariant of system (7.2) with cofactor K , and ψ 0
is the lowest order homogeneous term of ψ, then ψ 0 is a Darboux polynomial
of Xs , and there exists an m ∈ Zn+ , |m| > 0 such that

K (0) =
m, λ.
316 7 Local Integrability of Differential Systems

(b) If ψ1σ1 · · · ψrσr is a formal Darboux first integral of system (7.2), then (ψ10 )σ1 · · ·
(ψr0 )σr is a Darboux first integral of Xs .

Remark Statement (a) is a generalization of Theorem 7.3 and of Shi

[395, Theorem 1], because each Darboux semi-invariant with zero cofactor is a
first integral of the system. Statement (b) generalizes Cong et al. [117, Lemma 7] on
formal meromorphic first integrals. We note that the method in [117] is not suitable
for the proof of Theorem 7.13.

To simplify the proof of Theorem 7.13, we will apply the Poincaré–Dulac normal
form theorem. We restate it here with some additional information on local Darboux
semi-invariants.

Proposition 7.4 For the analytic or formal differential system (7.2) with A in the
Jordan normal form, the following statements hold.
(a) There always exist a near identity formal transformation

x = Φ(y) = y + higher-order terms

and a Poincaré–Dulac normal form system

ẏ = G(y) := Ay + higher-order terms (7.27)

such that
∂Φ(y)G(y) = F(Φ(y)), for all y ∈ (Cn , 0).

(b) ψ is a formal Darboux semi-invariant of system (7.2) with cofactor K if and

only if ψ ◦ Φ is a formal Darboux semi-invariant of system (7.27) with cofactor
K ◦ Φ.

By statement (b) we only need to prove Theorem 7.13 for the normal form systems.
We will see that it would be very convenient to use normal form systems to study
formal Darboux semi-invariants.
Assume that system (7.2) is in the Poincaré normal form, and let X be its asso-
ciated vector field. A formal Darboux semi-invariant ψ of X with cofactor K is
canonical if
Xs (ψ) = K (0)ψ and Xs (K ) = 0.

Walcher [437, Lemma 2.2] in 2000 provided a relation between formal Darboux
semi-invariants and canonical formal Darboux semi-invariants.

Lemma 7.7 Assume that system (7.2) is in the Poincaré normal form, and ϕ is a
formal Darboux semi-invariant of the system. The following statements hold.
(a) There exist an invertible series ρ and a canonical formal Darboux semi-
invariant ψ of system (7.2) such that ϕ = ρψ.
7.4 The Local Theory of Darboux Integrability 317

(b) In particular, each formal Darboux semi-invariant of system (7.2) is also a

formal Darboux semi-invariant of Xs .

The proof of this last lemma can be obtained from the definition of formal Darboux
semi-invariants. Lemma 7.7 (b) can be used to obtain necessary conditions for the
existence of a formal Darboux semi-invariant.

Lemma 7.8 Let ψ be a formal Darboux semi-invariant of Xs with cofactor K such

that Xs (K ) = 0. Then K is a constant.

Proof Let ψ () be the lowest order homogeneous term of ψ, and let it have degree
. Write ψ and K as
 
ψ= ψ (+ j) , K = k (i) ,
j≥0 i≥0

with ψ (+ j) and k (i) homogeneous polynomials of degree  + j and i, respectively.

Then we have

  (i) (+ j−i)

j

Xs ψ (+ j) = k ψ , j = 0, 1, . . .
i=0

 
By induction we can prove Xs ψ (+ j) = k (0) ψ (+ j) for all j. The lemma
follows. 

The next result, due to Walcher [436], is a characterization of the canonical

Darboux semi-invariants of Xs . It can be proved by using Lemma 7.8 via some
calculations.

Lemma 7.9 For a linear semisimple vector field Xs with eigenvalues λ, the follow-
ing statements hold.
(a) Any homogeneous canonical Darboux polynomial of the vector field Xs is a
linear combination of monomials x m , and its cofactor is
m, λ.
(b) The cofactor of each canonical Darboux semi-invariant of Xs is a Z+ -linear
combination of eigenvalues λ.

The following example illustrates an application of the above lemma. For n = 2,

assume that λ2 = 3 · λ1 = 0. Then the Poincaré normal form of system (7.2) is

ẋ1 = λ1 x1 , ẋ2 = λ2 x2 + cx13 , c ∈ C.

Clearly ψδ := x2 + δx13 for all δ ∈ C is an irreducible Darboux polynomial of Xs



with cofactor 3λ1 . So for any nonzero constants σ1 , . . . , σ ∈ C satisfying σ j = 0,
j=1
and pairwise distinct constants δ1 , . . . , δ ∈ C,
318 7 Local Integrability of Differential Systems

(x2 + δ1 x1 )σ1 · · · (x2 + δ x1 )σ

is a Darboux first integral of Xs .

The final result is the main technique used to prove Theorem 7.13.

Proposition 7.5 Assume that system (7.2) is in the Poincaré normal form, and
ϕ1σ1 · · · ϕrσr is a Darboux first integral of the system. Then there exist canonical
Darboux semi-invariants ψ1 , . . . , ψr such that

ψ1σ1 · · · ψrσr = ϕ1σ1 · · · ϕrσr .

In particular, ψi is also a Darboux semi-invariant of Xs , and ψ1σ1 · · · ψrσr is a Darboux

first integral of Xs .

Proof By Lemma 7.7, for each ϕi , i ∈ {1, . . . , r − 1}, there exists an invertible
formal series ρi such that ψi (x) = ρi−1 (x)ϕi (x) is a canonical semi-invariant of
system (7.2). Let L i (x) be the cofactor associated to ψi (x). Then X s (L i ) = 0 for
1 ≤ i ≤ r − 1.
Define
σ /σ σr −1 /σr
ψr = ρ1 1 r . . . ρr −1 ϕr .

Then ψ1 , . . . , ψr satisfy the requirement of the lemma. For details, we refer the
reader to the original proof. 

Proof of Theorem 7.13. The proof follows from Propositions 7.4 and 7.5, and Lemmas
7.8 and 7.9. 

7.4.2 Applications of the Local Theory of Darboux

Integrability

The first application is to the formal meromorphic first integrals, which is an inter-
esting consequence of Proposition 7.5.

Corollary 7.2 Assume that system (7.2) is in the Poincaré normal form, and ϕ is a
formal meromorphic first integral of the system. Let X be the vector field associated
to system (7.2) and Xs the semisimple part of the linearization of X . The following
statements hold.
• There exist canonical Darboux semi-invariants ψ1 , ψ2 and a constant γ such that

ψ1
ϕ= , Xs (ψ1 ) = γ ψ1 , Xs (ψ2 ) = γ ψ2 ,
ψ2

and ψ1 /ψ2 is a formal meromorphic first integral of Xs .

7.4 The Local Theory of Darboux Integrability 319

• The set of formal meromorphic first integrals of X is a subset of the meromorphic

first integrals of Xs .

This result provides a necessary condition for finding formal meromorphic first
integrals of the vector field X via Xs .
The next application of the formal Darboux first integral is to the classical Darboux
theory of integrability.

Proposition 7.6 Assume that the polynomial differential system

ẋ = P(x), x ∈ Cn , P(x) ∈ (C[x])n , (7.28)

has a Darboux first integral

ϕ1σ1 · · · ϕrσr

where ϕ1 , . . . , ϕr are pairwise relatively prime and irreducible polynomials, and the
σ j ’s are nonzero constants. Let z be a singularity of system (7.28), and B := ∂z P(z)
have the eigenvalues μ1 , . . . , μn with B = Bs + Bn in the semisimple-nilpotent
decomposition. The following statements hold.
(a) If system (7.28) has no analytic first integral in a neighborhood of z, then there
exists some j ∈ {1, . . . , r } such that ϕ j (z) = 0.
(b) Assume that ϕ1 (z) = · · · = ϕq (z) = 0 for some q ≥ 1, and ϕk (z) = 0 for all
k > q. Let ϕ 0j (x − z) be the lowest order term of ϕ j (x − z), j ∈ {1, . . . , q}.
Then there exist nonnegative integers  jk such that
 

X Bs (ϕ 0j ) =  jk μk ϕ 0j , 1 ≤ j ≤ q,
k

 σ  σ
and ϕ10 1 · · · ϕq0 q is a first integral of ẋ = Bs x, where X Bs is the vector field
associated to system ẋ = Bs x.

Proof (a) If all ϕ j (z) = 0, then ϕ1σ1 · · · ϕrσr is an analytic first integral in a neighbor-
hood of z. This contradiction verifies statement (a).
σq+1
(b) Since the function ρ := ϕq+1 · · · ϕrσr is analytic at z, and ρ(z) = 0, there exists
an analytic function g in a neighborhood of z such that g σq = ρ. Applying Theorem
7.13 to the local Darboux first integral
σq−1  σ
ϕ1σ1 · · · ϕq−1 gϕq q

of system (7.28), we get statement (b), which completes the proof of the
proposition. 

We believe it is a worthwhile and important project to find further applications of

the local theory of Darboux integrability to both integrability and dynamics.
320 7 Local Integrability of Differential Systems

7.5 Analytic Normalization of Analytic Integrable Systems

In this section we will discuss the existence of analytic normalizations of analytic

integrable differential systems near a singularity. Consider the analytic differential
system
ẋ = Ax + f (x), x ∈ (Kn , 0) (7.29)

where K = C or R, and f (x) = O(|x|2 ) is analytic in (Kn , 0). Assume that the
eigenvalues of A are λ = (λ1 , . . . , λn ).
If n = 2, K = R, and A has the purely imaginary eigenvalues λ = ±iω = 0, by
a linear change of variables and a time rescaling system (7.29) can be written as
∞
 ∞

ẋ = −y + f i (x, y), ẏ = x + gi (x, y), (7.30)
i=2 i=2

where f i , gi are homogeneous polynomials of degree i. The origin of the analytic

differential system (7.30) is either a center or a weak focus. If the origin is a center,
it is called an elementary or nondegenerate center of system (7.29).
On the local integrability of the planar analytic differential systems at a singularity
there are lots of results, see e.g. [213, 452] and the references therein. Here we mainly
concentrate on the equivalent characterization of analytic integrable systems, and the
existence of embedding flows of analytic integrable dynamical systems.

7.5.1 Equivalent Characterization of Analytic

Integrable Systems

We first recall one of Poincaré’s classical results, which provides an equivalent char-
acterization of elementary centers [362].
Theorem 7.14 For the real planar analytic differential system (7.30), the following
statements hold.
(a) The origin of system (7.30) is a center if and only if system (7.30) has an analytic
first integral in a neighborhood of the origin.
(b) The origin of system (7.30) is a center if and only if system (7.30) is analytically
orbitally equivalent to its linear part in a neighborhood of the origin. That is,
there exists a near identity analytic change of coordinates which transforms
system (7.30) to
   
u̇ = −v 1 + q(u 2 + v2 ) , v̇ = u 1 + q(u 2 + v2 ) ,

where q(z) = o(1) is an analytic function.

7.5 Analytic Normalization of Analytic Integrable Systems 321

Remark Theorem 7.14 (b) can also be stated as

(b ) System (7.30) has a center at the origin if and only if when written in conjugate
complex coordinates system (7.29) can be transformed to

u̇ = iu(1 + q(uv)), v̇ = −iv(1 + q(uv))

by a near identity analytic change of coordinates, where q(z) = o(1) is an

analytic function.
Theorem 7.14 (a) has been generalized to C k differential systems, k ∈ N ∪ {∞},
see Lyapunov [315] and Mazzi and Sabatini [331].
Theorem 7.15 Assume that system (7.29) is a real planar C k differential system,
k ∈ N ∪ {∞}. Then the origin is a center if and only if system (7.29) has a C k first
integral in a neighborhood of the origin, which attains the minimal value at the
origin.
Mardes̆ić et al. [329] in 1995 characterized the isochronous center of the planar
analytic differential system (7.30) at the origin. Their result extends Theorem 7.14 (b)
of Poincaré. Recall that a center is isochronous if all the periodic orbits inside the
center annulus have the same period. A center annulus is formed by a continuous
family of periodic orbits around a center.
Theorem 7.16 A real planar analytic differential system (7.30) has an isochronous
center at the origin if and only if it can be analytically linearized in a neighborhood
of the origin. That is, there exists a near identity analytic transformation sending
system (7.30) to
u̇ = −ku, v̇ = kv,

where k is a nonzero constant.

Remark By Theorem 7.16 the necessary condition for a center of a planar analytic
differential system to be isochronous is that the center is elementary. This condition
was generalized by Villarini [432] in 1992 to C ∞ differential systems. For more
information on isochronous centers, we refer to the survey paper [78] by Chavarriga
and Sabatini.
For planar hyperbolic saddles, Moser [338] in 1956 provided an equivalent char-
acterization of the existence of an area preserving analytic normalization of a planar
Hamiltonian system in a neighborhood of a hyperbolic saddle.
Theorem 7.17 Assume that (7.29) is a real planar analytic Hamiltonian system.
Then the origin is a hyperbolic saddle if and only if there exists a real analytic area
preserving change of coordinates which transforms system (7.29) to

u̇ = −u (1 + q(uv)) , v̇ = v (1 + q(uv)) ,

where q(z) = o(1) is an analytic function.

322 7 Local Integrability of Differential Systems

For the integrability of complex quadratic systems in C2 , Christopher et al. [108]

studied their orbital normalizability, normalizability (i.e., convergence of the normal-
ization), integrability (i.e., orbital linearizability), and linearizability of the systems
at saddle points. In this direction there are also some results for certain special three-
dimensional systems, see e.g. Christopher and Rousseau [109] and Romanovski
et al. [375].
On the existence of analytically equivalent normalizations of local analytic inte-
grable systems, using a geometric method Zung [488] in 2002 proved that any ana-
lytic integrable differential system is analytically conjugate to its normal form. Zung
[487] in 2005 obtained the existence of analytic normalization for integrable Hamil-
tonian systems, that is, the existence of analytic symplectic normalizations which
transform the analytic integrable symplectic Hamiltonian systems to their Birkhoff
normal forms. Zhang [466, Theorem 1.1] in 2008 not only proved the existence of
analytic normalization for general n-dimensional analytically integrable differential
systems (7.29) to their Poincaré normal forms at a nondegenerate singularity, but
also provided the concrete expressions of the normal forms. Here nondegeneracy
means that A has no zero eigenvalues.
We now state the result of Zhang [466]. Set

Mλ := m = (m 1 , . . . , m n ) ∈ Zn+ |
m, λ = 0, |m| ≥ 1 ,

whose elements are called resonant lattices, as above. Denote by m λ the number of
Q+ -linearly independent elements of Mλ . If λ = 0 then m λ ≤ n − 1.

Theorem 7.18 Assume that the origin of system (7.29) is a nondegenerate singu-
larity. Then system (7.29) has n − 1 functionally independent analytic first integrals
in a neighborhood of the origin if and only if m λ = n − 1, and system (7.29) is
analytically equivalent to a distinguished normal form of the form

ẏi = λi yi (1 + g(y)), i = 1, . . . , n, (7.31)

where g(y) = o(1) is an analytic function in y m , m ∈ Mλ and gcd(m 1 , . . . , m n ) = 1.

Theorem 7.18 is a little different from that of [466, Theorem 1.1]. Here we erase
the assumption that A is diagonalizable.
Proof of Theorem 7.18. Sufficiency. Since m λ = n − 1, there exist n − 1 Q+ -linearly
independent elements in Mλ , say

m i = (m i1 , . . . , m in ) ∈ Zn+ , i = 1, . . . , n − 1.

Since
m i , λ = 0, it follows that the y m i ’s, i = 1, . . . , n − 1, are functionally inde-
pendent analytic first integrals of system (7.42). Consequently, system (7.29) has the
n − 1 functionally independent analytic first integrals Ψ m i , i = 1, . . . , n − 1, where
y = Ψ (x) is the inverse of the near identity analytic transformation which sends
system (7.29)–(7.42).
7.5 Analytic Normalization of Analytic Integrable Systems 323

Necessity. The proof follows from the following three lemmas. Let

ẏ = Ay + g(y) (7.32)

be the distinguished normal form of system (7.29) with the associated distinguished
normalization
x = y + Φ(y). (7.33)

Lemma 7.10 If system (7.29) has n − 1 functionally independent analytic first inte-
grals, then its distinguished normal form has the following form

ẏi = λi yi (1 + q(y)), i = 1, . . . , n, (7.34)

where q(y) is a formal series without the constant term.

Proof Let Y be the vector field associated to the distinguished normal form (7.32),
and let Ys be the semisimple component of the linear part of Y . Assume that
H1 (x), . . . , Hn−1 (x) are the n − 1 functionally independent analytic first integrals
of system (7.29). By Lemma 7.3 the vector field Y has n − 1 functionally inde-
pendent first integrals K 1 (y), . . . , K n−1 (y) with all their monomials resonant. So
K 1 (y), . . . , K n−1 (y) are also first integrals of Ys . This proves that the two vector
fields Y and Ys are parallel in a neighborhood of the origin. Hence there exists a func-
tion 1 + q(y) such that Y = (λ1 y1 (1 + q(y)), . . . , λn yn (1 + q(y))). This proves the
lemma. 

The next result provides a condition that the eigenvalues of the linear part of an
integrable system at the singularity satisfy.

Lemma 7.11 If system (7.29) has n − 1 functionally independent analytic first inte-
grals in a neighborhood of the origin, then there exists a σ > 0 such that for all

m, λ − λi = 0, m ∈ Zn+ , |m| ≥ 2, we have

|
m, λ − λi | > σ.

Proof By Theorem 7.8 and the assumption, the eigenvalues of A must satisfy the
n − 1 resonant relations:

m i,1 λ1 + · · · + m i,n λn = 0, i = 1, . . . , n − 1, (7.35)

where the n − 1 vectors (m 1,1 , . . . , m 1,n ), . . . , (m n−1,1 , . . . , m n−1,n ) ∈ Zn+ have rank
n − 1. We assume without loss of generality that
⎛ ⎞
m 1,1 . . . m 1,n−1
⎜ ⎟
det ⎝ ... ..
. ⎠ = 0.
m n−1,1 . . . m n−1,n−1
324 7 Local Integrability of Differential Systems

Then Eq. (7.35) have a unique solution, which is of the form

λ1 = q1 p1−1 λn , . . . , λn−1 = qn−1 pn−1

−1
λn ,

where pi , qi ∈ Z \ {0}, and gcd( pi , qi ) = 1, i = 1, . . . , n − 1.

These last expressions of the λi ’s show that for any m ∈ Zn+ and |m| ≥ 2 such
that
m, λ − λi = 0, we have

|
m, λ − λi | = σi |λn |( p1 . . . pn−1 )−1 ≥ |λn |( p1 . . . pn−1 )−1 ,

where σi ∈ N. Then the lemma holds with

 
σ = min |λi |( p1 . . . pn−1 )−1  i = 1, . . . , n .

Lemma 7.10 shows that condition A of Bruno [42–44] holds, i.e. the Poincaré
normal form of system (7.29) has the form

ẋ = As x(1 + q(x)), q(0) = 0.

Lemma 7.11 indicates that condition ω of Bruno [42–44] holds, which is one of the
Diophantine conditions on the eigenvalues of A.

Theorem 7.19 (Bruno’s theorem) If the analytic differential system (7.29) satisfies
conditions ω and A, then it is analytically equivalent to its Poincaré normal form.

By Lemmas 7.10 and 7.11, and Bruno’s theorem, we can finish the proof of the
necessary part. Here for completeness we provide a proof which is independent of
Bruno’s theorem.

Lemma 7.12 Under the necessary conditions of Theorem 7.18, the distinguished
normalization (7.33) reducing system (7.29) to its normal form (7.34) is convergent.

Proof By Lemma 7.10, the normal form of system (7.29) is of the form (7.31) via the
distinguished normalization (7.33). Let f i and ϕi be respectively the ith components
of the n-dimensional vector-valued functions f and Φ. We write v(z) ∈ { f i , ϕi , q}
as the summation of monomials

v(z) = v(m) z m ,
m∈Zn+
7.5 Analytic Normalization of Analytic Integrable Systems 325

where the v(m) ’s are the coefficients of the monomials z m = z 1m 1 . . . z nm n . The proof
of Theorem 7.1 shows that the ϕs(m) ’s satisfy

(
m, λ − λs )ϕs(m) = [ f s (y+Φ(y))](m) − λs q (m−es ) (7.36)

n 
− ϕs()  j λ j q (m−) ,
j=1 ≺m,∈Zn+

where [ f s ](m) := [ f s (y + Φ(y))](m) is the coefficient of y m obtained by re-expanding

f s (y + Φ(y)) in y and es is the sth unit vector. Hereafter,  ≺ m means m −  ∈ Zn+ .
Since the normalization is distinguished, if
m, λ − λs = 0 then Eq. (7.36) has
the solution
⎛ ⎞

ϕs(m) = 0, q (m−es ) = λ−1
s
⎝[ f s ](m) −
, λϕs() q (m−) ⎠ . (7.37)
≺m, l∈Zn+

If
m, λ − λs = 0, according to the proof of the Poincaré–Dulac normal form the-
orem, taking q (m−es ) = 0, Eq. (7.36) has the solution
⎛ ⎞

ϕs(m) = ⎝[ f s ](m) −
, λϕs() q (m−) ⎠ (
m, λ − λs )−1 . (7.38)
≺m, ∈Zn+

We claim that in (7.37)

q (m−es ) = λ−1
s [ fs ]
(m)
.

Indeed, since the q (m−) ’s are the coefficients of the resonant terms, we must have

m − , λ = 0. Hence
, λ =
m, λ = λs . This implies that the ϕs() ’s are the
coefficients of the resonant terms, and so ϕs() = 0 because the normalization is
distinguished. The claim follows.
The above arguments verify that the distinguished normalization is

xs = ys + ϕs(m) y m ,
m∈Zn+ ,|m|>1

with ϕs(m) satisfying (7.38), and that the distinguished normal form is
⎛ ⎞

ẏs = λs ys ⎝1 + q (m−es ) y m−es ⎠ , s = 1, . . . , n,
0=m−es ∈Zn+

with q (m−es ) satisfying (7.37), and

m − es , λ = 0.
326 7 Local Integrability of Differential Systems

We now estimate ϕs(m) . By Lemma 7.11 there exists a positive number δ such that
for |
m, λ − λ j | = 0, m ∈ Zn+ , |m| ≥ 2, we have

|
m, λ − λ j |−1 ≤ δ.

Hence we get from

, λ =
m, λ = λs that
  () (m−) 

 ≺m, ∈Zn
, λϕs q   
|λs |

 () (m−) 
 +  ϕ q .
 ≤ 1+

m, λ − λs  |
m, λ − λs | s
  ≺m, ∈Zn+

Set
ρ = max{1 + δ|λs |, s = 1, . . . , n}.

One has  (m)      () (m−) 

ϕ  ≤ δ [ f s ](m)  + ρ ϕ q .
s s
≺m, ∈Zn+

From the Cauchy inequality there exists a polydisc P:={|xs | < r, s = 1, . . . , n},
in which the following inequalities hold
 
[ f s ](m)  ≤ Mr −|m| , n
M = max sup{| f s |},
s=1 ∂P

where ∂P denotes the boundary of P. Define

∞

fˆ(x) = M r −|m| x m .
|m|=2

Clearly fˆ is analytic in the interior of P, and is a majorant series of all f s , s =

1, . . . , n. In what follows, we denote by v̂ the majorant series of a given series v, and
represent it as v  v̂, see e.g. Hille [205].
Direct calculations show that


n 
n
ϕs + q  ϕ̂s + q̂
s=1 s=1

n
 (nδ + κ) fˆ(y + ϕ̂) + (ρ + 1) ϕ̂s q̂, (7.39)
s=1

where κ = max1≤s≤n {|λs |−1 }. Since the coefficients of ϕ̂s and q̂ are positive, the
n
convergence of the series ϕ̂s (y) + ĝ(y) is equivalent to its convergence when
s=1
7.5 Analytic Normalization of Analytic Integrable Systems 327

taking y1 = . . . = yn = w. Set


n 

ξ(w) = ϕ̂s (y) + q̂  .

s=1 y1 =...=yn =w

Then ξ(w) = wη(w) with η(w) a series. Moreover, we get from (7.39) that

wη(w)  (nδ + κ)w2 fˆ∗ (1 + η(w)) + (ρ + 1)w2 η(w)2 , (7.40)

where fˆ∗ (1 + η(w)) = fˆ(w + ϕ̂1 (w, . . . , w), . . . , w + ϕ̂(w, . . . , w))/w2 .

Set
L(w, h) := h − (nδ + κ)w fˆ∗ (1 + h) − (ρ + 1)wh 2 . (7.41)

We can check that L(w, h) is analytic in a neighborhood of the origin, and satisfies

L(0, 0) = 0, ∂h L|(0,0) = 1.

By the Implicit Function Theorem, L(w, h) = 0 has a unique analytic solution h(w)
in a neighborhood of the origin. Comparing (7.40) and (7.41) we get that h(w) is a
majorant series of η(w). Hence η(w) is convergent in a neighborhood of the origin,

n
and consequently ξ(w) is convergent. This proves the convergence of ϕ̂s + q̂, and
s=1
consequently of ϕs and q. This completes the proof of the lemma. 
Combining Lemmas 7.10–7.12, one can get a proof of the necessity. The proof
of Theorem 7.18 is complete. 
Theorem 7.18 required the nondegeneracy of system (7.29) at the origin. This
additional condition was removed by Zhang [474, Theorem 1.3].
Theorem 7.20 Assume that the eigenvalues of A are not all equal to zero. Then
system (7.29) has n − 1 functionally independent analytic first integrals in a neigh-
borhood of the origin if and only if m λ = n − 1, and system (7.29) is analytically
equivalent to its distinguished normal form

ẏi = λi yi (1 + q(y)), i = 1, . . . , n, (7.42)

where q(y) = o(1) is an analytic function in y m , m ∈ Mλ and gcd(m 1 , . . . , m n ) = 1.

Recently Theorem 7.18 was extended to analytic differential systems with a peri-
odic orbit by Wu and Zhang in [447].
Theorem 7.21 Assume that the analytic differential system

ẋ = f (x), x ∈ Ω ⊂ Rn ,

with Ω an open subset, has a periodic orbit, and it is analytically integrable in a

neighborhood of the periodic orbit. The following statements hold.
328 7 Local Integrability of Differential Systems

• The system is analytically equivalent to its distinguished normal form in a neigh-

borhood of the periodic orbit.
• The characteristic exponents of the periodic orbit satisfy

λ1 = ν1 i, . . . , λn−1 = νn−1 i, λn = νn i, ν j ∈ Q.

Proof See the original proof of Wu and Zhang in [447], where they used Floquet’s
theorem, Fourier expansion, the properties of characteristic multipliers of the periodic
orbit, and so on. The details are omitted. 

The next result generalizes Theorems 7.18, 7.20 and the result in [263] to formal
Darboux integrable systems, see Llibre et al. [294].

Theorem 7.22 For the analytic or formal differential system (7.29) in Cn , assume
that As = 0. The following statements hold.
(a) System (7.29) has n − 1 functionally independent formal Darboux first integrals
if and only if it satisfies condition A of Bruno. Furthermore, condition ω of
Bruno and the existence of n − 1 functionally independent formal Darboux first
integrals ensures the convergence of the normalization.
(b) If system (7.29) is analytic and has n − 1 functionally independent formal
meromorphic first integrals, then the normalization from system (7.29) to its
Poincaré normal form is convergent.

Proof (a) The necessity follows from Lemma 7.10, whereas the sufficiency follows
from the proof of the sufficiency of Theorem 7.18.
(b) Similar to the proof given in Zhang [466] and [471], there exist a nonzero constant
α and integers λ∗j for which the eigenvalues of A satisfy λ j = α · λ∗j , j = 1, . . . , n.
This ensures condition ω of Bruno. By statement (a) an integrable system satisfies
condition A of Bruno. Hence statement (b) follows from Theorem 7.19 of Bruno.
This completes the proof of the theorem. 

The following example, due to Llibre et al. [294], provides an application of

Theorem 7.22. The polynomial differential system

ẋ1 = x1 + a1 x12 + a2 x1 x2 + a3 x22 + higher-order terms,

ẋ2 = 2x2 + b1 x12 + b2 x1 x2 + b3 x22 + higher-order terms,

has a formal Darboux first integral if and only if b1 = 0, because its Poincaré normal
form is
ẋ1 = x1 , ẋ2 = 2x2 + b1 x12 ,

and a planar analytic or formal differential system has a formal Darboux first integral
if and only if its Poincaré normal form vector field is parallel to its linear part.
7.5 Analytic Normalization of Analytic Integrable Systems 329

7.5.2 Integrable Discrete Dynamical Systems

and Embedding Flows

In the last subsection we discussed the convergence of the normalization of an analytic

integrable differential system. In this subsection we establish a relation between
integrable differential systems and integrable discrete systems. That is, we discuss
the existence of embedding flows of integrable discrete dynamical systems. In the
last part of this subsection we will also present a result on the existence of analytic
normalizations of analytic integrable discrete dynamical systems to their Poincaré
normal forms.
Continuous dynamical systems are usually connected to discrete dynamical sys-
tems though the Poincaré maps. Given an n-dimensional manifold M , and a C k
dynamical system defined on it, k ∈ Z+ ∪ {∞, ω}, we denote by ϕt (x) the flow of
the system. For any given time T , ϕT (x) defines a C k diffeomorphism on M , which
provides a C k discrete dynamical system on M .
Conversely, given a C  ,  ≤ k, discrete dynamical system, say ϕ(x), on an n-
dimensional C k manifold, the problem is: Does there exist a smooth vector field X
on M such that ϕ(x) is the time 1 map of the vector field? That is

X ◦ F(x) = ∂x F(x)X (x), for all x ∈ M .

Recall that ∂x F is the Jacobian matrix of F.

We remark that the time 1 is not essential, and it can be chosen as any prescribed
time which depends on the concrete problem. If the vector field X exists, it is
called an embedding vector field of ϕ on M . In this circumstance, we also speak of
embedding the diffeomorphism ϕ in the vector field X . If the embedding vector field
X is autonomous, we say that ϕ can be embedded in a flow. Then the mentioned
problem is called an embedding flow problem of the diffeomorphism.
Generally speaking, it is hopeless to study this problem on a given manifold. Until
now most of the results have been local, that is in a neighborhood of a singularity.
Progress on the study of the embedding flow problem is slow, and the results are few.
We refer to the following references on this subject.
• Arnold [12, p. 200]: the existence of C ∞ embedding periodic differential systems
of local C ∞ diffeomorphisms in a finite-dimensional space.
• Belitskii and Tkachenko [30], and Li [230]: embedding one-dimensional analytic
and smooth diffeomorphisms in one-dimensional differential equations.
• Culver [120]: the existence of embedding autonomous linear differential systems
of finite-dimensional linear diffeomorphisms.
• Ilyashenko and Yakovenko [213]: the existence of formal embedding differential
systems of finite-dimensional unipotent diffeomorphisms. Recall that a unipotent
diffeomorphism is a diffeomorphism whose linear part has eigenvalues all equal
to 1.
330 7 Local Integrability of Differential Systems

• Kuksin and Pöschel [221]: embedding finite-dimensional and nearly integrable

analytic (or C ∞ ) symplectic diffeomorphisms into analytic (or C ∞ ) Hamiltonian
systems.
• Palis [353]: the diffeomorphisms which can be embedded in a flow are rare in the
Baire sense.
• Li et al. [232]: the existence of C ∞ embedding autonomous differential systems
of finite-dimensional local C ∞ hyperbolic diffeomorphisms with weak nonreso-
nance.
Recall that a local hyperbolic diffeomorphism is a diffeomorphism whose linear
part at its fixed point has no eigenvalues with vanishing real parts.
• Zhang [466, 468, 469]: the existence of C ∞ or analytic embedding autonomous
differential systems of finite-dimensional C ∞ or analytic local hyperbolic diffeo-
morphisms with weak nonresonance.
• Zhang [467]: the existence of C ∞ embedding autonomous differential systems of
infinite-dimensional C ∞ local hyperbolic diffeomorphisms with weak nonreso-
nance.
• Ríbon [371]: the necessary conditions on the existence of formal embedding
autonomous differential systems of finite-dimensional formal hyperbolic diffeo-
morphisms with weak resonance.
• Zhang [476]: the sufficient conditions on the existence of C ∞ embedding
autonomous differential systems of 3-dimensional C ∞ hyperbolic diffeomor-
phisms with weak resonance.
Focusing on our subject, here we will consider the existence of an embedding
flow of integrable diffeomorphisms. Given an n-dimensional C k manifold M , k ∈
Z+ ∪ {∞, ω}, we consider a C k diffeomorphism F : M → M . A smooth function
H is a first integral of F if

H (x) = H (F(x)), for all x ∈ M .

If F has n − 1 functionally independent first integrals, it is called an integrable

diffeomorphism. If in addition the first integrals are C k , we call F C k integrable. If
k = ω, we call F analytic integrable.

Remark There exists a diffeomorphism F which has the maximal number, i.e. n, of
functionally independent first integrals. Cima et al. [111] in 2006 proved that if F
has n functionally independent first integrals, then in most cases there exists a k ∈ N
such that F k is the identity.

Zhang [466, Theorem 1.3] proved the existence of an embedding flow of integrable
diffeomorphisms.

Theorem 7.23 Any volume preserving analytic integrable diffeomorphism on a real

or complex analytic manifold M can be embedded in an analytic autonomous
differential system.
7.5 Analytic Normalization of Analytic Integrable Systems 331

Proof Let
H1 (x), . . . , Hn−1 (x)

be n − 1 functionally independent analytic first integrals of F. Then except per-


n−1
haps in a zero Lebesgue measure subset each orbit of F(x) is contained in {x ∈
i=1
M | Hi (x) = ci } =: γc for some c = (c1 , . . . , cn ) ∈ Rn .
Set
Z (x) = ∇ H1 (x) × . . . × ∇ Hn−1 (x), x ∈ M ,

where ∇ denotes the gradient of a function, and × the cross product of n-dimensional
vectors. Recall that the cross product of n − 1 column vectors v1 , . . . , vn−1 in Rn is
also a vector v, which is defined via

v, w =
v1 × . . . × vn−1 , w = det(v1 , . . . , vn−1 , w), for all w ∈ Rn .

Obviously, Z (x) is an analytic vector field, and is tangent to each γc at x ∈ γc .

This implies that γc is an orbit of Z (x). Consequently, the orbits of the vector field
Z (x) are formed by {γc | c ∈ Rn }, and any orbit of F(x) is contained in the orbit of
Z.
To prove that Z is an embedding vector field of F(x), we need the techniques
from Cima et al. [112].
First we claim that

D F(x)Z (x) = ∂x F(x) ((∂x F(x))τ ∇ H1 (x) × . . . × (∂x F(x))τ ∇ Hn−1 (x)) ,
(7.43)
where D F(x) = det(∂x F(x)). Indeed, for any w(x) ∈ Tx M , the tangent space of
M at x, we have

w(x), ∂x F(x) ((∂x F(x))τ ∇ H1 (x) × . . . × (∂x F(x))τ ∇ Hn−1 (x))
=
∂x F(x))τ w(x), (∂x F(x))τ ∇ H1 (x) × . . . × (∂x F(x))τ ∇ Hn−1 (x)
= det ((∂x F(x)τ w(x), (∂x F(x)τ ∇ H1 (x), . . . , (∂x F(x)τ ∇ Hn−1 (x))
= D F(x) det (w(x), ∇ H1 (x), . . . , ∇ Hn−1 (x)) .

The claim follows.

Second we claim that

∂x F(x)Z (x) = D F(x)Z (F(x)). (7.44)

Indeed, by Hi (F(x)) = Hi (x), i = 1, . . . , n − 1, we have (∂x F(x))τ ∇ Hi ◦ F(x) =

∇ Hi (x). So by the definition of Z and the equality (7.43), one has
332 7 Local Integrability of Differential Systems

 
∂x F(x)Z (x) = ∂x F(x) (∂x F(x))τ (∇ H1 ◦ F(x)) × . . . × (∂x F(x))τ (∇ Hn−1 ◦ F(x))
= D F(x)Z ◦ F(x).

This proves the claim.

Since the diffeomorphism F(x) is volume preserving, i.e. D F(x) = 1, we get
from (7.44) that X is an analytic embedding vector field of F(x). This proves the
theorem. 

Remark Theorem 7.23 also holds for any volume preserving C k integrable diffeo-
morphism, k ∈ Z+ ∪ {∞, ω}. That is, each finite-dimensional volume preserving C k
integrable diffeomorphism can be embedded in a C k−1 autonomous integrable vec-
tor field, where if k = ∞ (resp. ω), we mean k − 1 = ∞ (resp. ω). The embedding
vector field in Theorem 7.23 is also integrable.

In 2013, Zhang [474, Theorem 1.4] solved the embedding flow problem for inte-
grable diffeomorphisms without the volume preserving restriction.

Theorem 7.24 Any C k integrable diffeomorphism on a real or complex C k manifold

M can be embedded in a C k−1 autonomous vector field on M .

Proof With the help of local coordinates of the manifold, by suitably modify-
ing the embedding vector field constructed in the proof of Theorem 7.23 we can
obtain a proof of the theorem. The details are omitted. We refer to the proof of
[474, Theorem 1.4]. 

Cima et al. [112, Theorem 2] studied the relation between the dynamics of inte-
grable diffeomorphisms and of integrable vector fields. A diffeomorphism F on
a differentiable manifold M satisfies condition μ if there exists a smooth map
μ : M → R such that

μ(F(z)) = D F(z)μ(z), for all z ∈ M .

Denote by γz the orbit of a continuous or a discrete dynamical system passing

through z.

Theorem 7.25 Let M ⊂ Rn be an open subset. Assume that F : M → M is a

diffeomorphism satisfying the condition μ, and has n − 1 functionally independent
first integrals H(x) = (H1 (x), . . . , Hn−1 (x)). Set
  
μ(x) −∂x2 H1 , ∂x1 H1 , if n = 2,
Zμ (x) =
μ(x) (∇ H1 (x) × . . . × ∇ Hn−1 (x)) , if n > 2.

For z ∈ M , if γz is a regular orbit of the vector field Zμ , and the number of connected
components of {x ∈ M | H(x) = H(z)} is  < ∞, then there exists a natural number
m ∈ [1, ] such that γz is an invariant set of F m .
7.5 Analytic Normalization of Analytic Integrable Systems 333

Remark The functionally independent first integrals H1 , . . . , Hn−1 of F are also the
functionally independent first integrals of the vector field Xμ . Theorem 7.25 provides
an implicit relation between the integrable diffeomorphisms and their embedding
flows.

Finally, we turn to the existence of analytic normalization of analytic integrable

diffeomorphisms. Let F(x) = Bx + f (x) be a local analytic or formal diffeomor-
phism with f having no constant and linear parts, and let μ = (μ1 , . . . , μn ) be the
n-tuple of eigenvalues of the matrix B. An element of

D = { m ∈ Zn+  μm = 1, |m| ≥ 2

is called a resonant lattice. A resonant lattice is simple if it is not a Q+ -linear combina-

tion of the other resonant lattices. Denote by dμ the maximal number of Q+ -linearly
independent elements of D. We call this number the resonant rank of B.
An analytic or formal diffeomorphism F(x) = Bx + f (x) is in the Poincaré
normal form if B is in the Jordan normal form, and the nonlinear term f (x) consists
of resonant monomials. A monomial x m e j in the jth component of f (x) is resonant
if μm = μ j .
An analytic or formal diffeomorphism G(y) is a Poincaré normal form of an
analytic or formal diffeomorphism F(x) if G(y) is in the Poincaré normal form,
and F(x) and G(y) are conjugate, i.e. there exists a near identity analytic or formal
transformation y = Φ(x) such that

G ◦ Φ(x) = Φ ◦ F(x).

The conjugacy y = Φ(x) is called a normalization from F(x) to G(y). Furthermore,

• if Φ(x) is analytic, we say y = Φ(x) is an analytic normalization of F(x), and
G(y) is an analytic normal form of F(x), or F(x) is analytically equivalent to its
normal form G(y);
• if the transformation Φ(x) consists of nonresonant terms, we call y = Φ(x) a
distinguished normalization of F(x). A monomial x m e j in the jth component of
Φ(x) is nonresonant if μm = 1.
Readers should take note of the difference between the resonances of monomials in
the normal forms and in the normalizations.

Theorem 7.26 For a local analytic diffeomorphism F(x) = Bx + f (x) in (Cn , 0)

with f (x) = O(|x|2 ), assume that B has at least one eigenvalue not on the unit cycle
of C. Then F(x) is analytic integrable if and only if the following conditions hold:
(a) the resonant rank of B is dμ = n − 1, where μ is the n-tuple of eigenvalues of
B; and
(b) F(x) is analytically equivalent to its normal form

G(y) = (μ1 y1 (1 + p1 (y)), . . . , μn yn (1 + pn (y))) ,

334 7 Local Integrability of Differential Systems

where p1 (y), . . . , pn (y) are analytic, and satisfy the functional equations

(1 + p1 (y))m k1 . . . (1 + pn (y))m kn = 1, k = 1, . . . , n − 1,

where m k = (m k1 , . . . , m kn ) ∈ D, k = 1, . . . , n − 1, are n − 1 linearly inde-

pendent simple resonant lattices.

Proof For a detailed proof, see that of Zhang [474, Theorem 1.1]. 

7.6 Varieties and Normalization of Partially Integrable

Systems

This section contributes to the study of varieties of integrable and partially integrable
differential systems, and the convergence and generic divergence of normalizations
of partially integrable differential systems.

7.6.1 Varieties of Partially Integrable Systems

A real planar analytic differential system with a nondegenerate monodromy singu-

larity can be written as

u̇ = −v + p(u, v), v̇ = u + q(u, v), (7.45)

where p, q = o(|(x, y)) are analytic functions. Poincaré proved that there exists an
analytic function or a formal series W (u, v) such that

 ∞
∂W ∂W
(−v + p(u, v)) + (u + q(u, v)) = vs (u 2 + v2 )s ,
∂u ∂v s=

where  ∈ N is greater than or equal to 2, and thevs ’s are polynomials in the coef-
ficients of p(u, v) and q(u, v). Recall that the set {vs = 0} in the parameter space
s
formed by the coefficients of system (7.45) is called a variety of system (7.45) at the
origin.
Note that the origin of system (7.45) is a center if and only if vs = 0 for all s. It
is also equivalent to system (7.45) having an analytic first integral.
These results were extended to any finite-dimensional integrable systems by
Romanovski et al. [376] in 2014. Du et al. [131] in 2016 further extended these
results to the partially integrable case. Now we state these results.
Let X be the vector field associated to the n-dimensional analytic differential
system (7.29), i.e. the system
7.6 Varieties and Normalization of Partially Integrable Systems 335

ẋ = Ax + f (x), x ∈ (Kn , 0)

and let λ be the n-tuple of eigenvalues of the matrix A. Set


Mλ = m ∈ Zn+ |
λ, m = 0, |m| ≥ 2 .

As before we call elements of Mλ resonant lattices associated to λ.

Theorem 7.27 Let m λ ∈ {1, . . . , n − 1} be the maximum number of Q+ -linearly

independent elements of Mλ . The following statements hold.
(a) There exists a series Ψ (x) with its resonant monomials arbitrary such that

X (Ψ (x)) = vσ x σ , (7.46)
σ ∈Zn+

where vσ are polynomials in terms of the coefficients of X .

(b) If the vector field X has m λ functionally independent analytic or formal first
integrals, then for any Ψ satisfying (7.46) we have

vσ = 0, for all σ ∈ Mλ . (7.47)

(c) Under the condition of (b), the vector field X has m λ functionally independent
first integrals of the form

H1 (x) = x σ1 + h 1 (x), . . . , Hm λ (x) = x σd + h m λ (x), (7.48)

where σ1 , . . . , σd are Q+ -linearly independent elements of Mλ , and each h j (x),

j = 1, . . . , m λ , consists of nonresonant monomials in x of degree larger than
|σ j |.

Note that if Mλ = ∅, statement (a) is trivial, because Ψ should be a constant and

X (Ψ ) ≡ 0. So we assume without loss of generality that Mλ = ∅. The vσ ’s are
called generalized Lyapunov quantities.
Statement (a) of Theorem 7.27 was proved in [376, Theorem 1.1], in which they
also proved statement (b) in the completely integrable case (i.e. m λ = n − 1). The
version in (b) was proved in [131]. Note that the ideas and tools for proving (b) are
different in [131, 376].
In case (b) of Theorem 7.27, that is, when the number of functionally independent
first integrals is equal to m λ , system (7.29) is called partially integrable if m λ <
n − 1. Meanwhile the variety mentioned above is called a partial integrability variety
of system (7.29). In the case m λ = n − 1, as usual we call system (7.29) completely
integrable.
By statement (b) of Theorem 7.27, in the (partially) integrable case the functions
vσ have the same set of common zeros for any choice of the resonant coefficients in
Ψ (x). So when we study the (partial) integrability of an analytic differential system,
336 7 Local Integrability of Differential Systems

we can set the resonant coefficients in Ψ (x) to be zero. When system (7.29) is a
polynomial one, by the Hilbert basis theorem [157] the variety {vσ = 0} is finitely
σ
generated.
Theoretically, Theorem 7.27 can be used to provide an algorithm for computing
both generalized Lyapunov quantities and analytic first integrals of analytic (par-
tially) integrable differential systems (7.29) at the origin. In practice, the concrete
calculations will be very complicated.
If the number of functionally independent first integrals is less than m λ , the situa-
tion will be different, see [133]. That is, some vσ ’s depend on the resonant coefficients
of Ψ , which cannot be eliminated.

7.6.1.1 Preparation for the Proof of Theorem 7.27

In this subsection we present some known results, which will be used in the proof of
Theorem 7.27.
The next result, due to Llibre et al. [263, Proposition 5], characterizes functionally
independent first integrals of the vector field X via the semisimple part Xs of its
linear term when X is in the Poincaré–Dulac normal form.

Proposition 7.7 Assume that the vector field X associated to system (7.29) is in
the Poincaré–Dulac normal form, and that Xs admits m λ functionally independent
polynomial first integrals. Then X admits m λ functionally independent formal first
integrals if and only if it admits every polynomial first integral of Xs .

In [131] Du et al. provided a new proof of this proposition, which corrected a

small gap in its original proof, where they used a generalized Puiseux series. Recall
that a Puiseux series is one of the form
∞
 k
ρ(z) = ck z  , z ∈ K,
k=k0

where  ∈ N, k0 ∈ Z. Note that if  = 1, a Puiseux series is a Laurent series. A

generalized Puiseux series in x = (x1 , . . . , xn ) is by definition a Laurent series in
x1k1 , . . . , xnkn with k1 , . . . , kn ∈ Q+ . For more information on generalized Puiseux
series, see e.g. [383].
Next we study the properties of the matrix A in the case when system (7.29) is
partially integrable. Without loss of generality we assume that A is in the Jordan
normal form, and that A = diag(A , A ) with eigenvalues λ = (λ , λ ) such that
(H ) if k = (k  , k  ) ∈ Mλ then k  = 0; and there exists a k ∈ Mλ such that k  = 0.
Clearly it is possible that A = A . The assumption (H ) is not a restriction. It
only says that the resonance of λ is independent of λ . This can be realized by
reorganizing the coordinates (if necessary). For instance, if A ∈ M (3, C) is in the
7.6 Varieties and Normalization of Partially Integrable Systems 337

Jordan normal form and has the eigenvalues −1 − 2i, −1 + 2i, 3 then A = A ; or
if it has the eigenvalues −3i, 3i, 1 then A = diag(−3i, 3i).

Proposition 7.8 Assume that the vector field X admits m λ functionally independent
formal first integrals. If A = (A , A ) satisfies (H ) then A is diagonal.

Proof Let
ẏ = Ay + g(y) (7.49)

be the Poincaré–Dulac normal form of system (7.29), i.e. X . By the assumption

of the theorem one gets that system (7.49) has m λ functionally independent first
integrals. Let σ1 , . . . , σm λ ∈ Mλ be Q+ -linearly independent. Then y σ1 , . . . , y σmλ are
functionally independent first integrals of Xs . Since we are under the assumptions of
Proposition 7.7, it follows that y σ1 , . . . , y σmλ are the first integrals of system (7.49),
and consequently are the first integrals of ẏ = Ay.
According to the decompositions of A = (A , A ) and y = (y  , y  ), set corre-
spondingly σ j = (σ j , σ j ) for j = 1, . . . , m λ . By the assumption (H ) and the fact

that y σ1 , . . . , y σmλ are the first integrals of system ẏ = Ay, one gets y σ j = (y  )σ j for
j = 1, . . . , m λ , and


∂ y  (y  )σ j , A y   = 0, j = 1, . . . , m λ .

This forces A to be diagonal, which ends the proof.

In the proof of Theorem 7.27, we also need normal forms on an invariant manifold.
Let λ = (λ , λ ) be the eigenvalues of A with λ = (λ1 , . . . , λk ), λ = (λk+1 , . . . , λn )
for some k ∈ {2, . . . , n}, and λ j = λs for j ∈ {1, . . . , k} and s ∈ {k + 1, . . . , n}. We
can write system (7.29) possibly with an invertible real linear change of coordinates
as
ẋ  = A x  + f  (x  , x  ), ẋ  = A x  + f  (x  , x  ), (7.50)

where A and A have respectively the eigenvalues λ and λ , and f  and f  consist
of the higher-order terms.
System (7.50) is in normal form on an invariant manifold if f  (x  , 0) ≡ 0 and
 
f (x , 0) consists of resonant monomials. See Bibikov [33], or Bruno [44]. In other
words, system (7.50) is in normal form on an invariant manifold if and only if system
(7.50) restricted to the invariant hyperplane x  = 0 is in the Poincaré–Dulac normal
form. If after a near identity analytic (or formal) change of coordinates system (7.50)
is transformed to a normal form on an invariant manifold, we say that system (7.50)
is analytically (or formally) equivalent to its normal form on an invariant manifold.
The next result shows the existence of a normal form on an invariant manifold,
see [33, Theorem 3.1].

Theorem 7.28 For an analytic or a formal differential system (7.50), if

λs −
λ , m   = 0, for all s ∈ {k + 1, . . . , n} and m  ∈ Zk+ ,
338 7 Local Integrability of Differential Systems

then it is formally equivalent to its normal form on an invariant manifold through a

near identity change of variables

x  = y  + ϕ  (y  ), x  = y  + ϕ  (y  ), (7.51)

with the coefficients of resonant monomials of ϕ  (y  ) arbitrary and the other coeffi-
cients of ϕ  and ϕ  uniquely determined.

If the condition in Theorem 7.28 holds, we say the eigenvalues λ and λ are
nonresonant with each other. If h  (y  ) consists of the nonresonant monomials only,
the transformation (7.51) is called distinguished. Correspondingly the normal form
on an invariant manifold of system (7.50) is called distinguished.

7.6.1.2 Proof of Theorem 7.27

(a) Let Ψ (x) be a series, and write it in the form

∞

Ψ (x) = Ψk (x),
k=

for some  ∈ N, where Ψk is a homogeneous polynomial of degree k. Write f (x) of

(7.29) as
∞

f (x) = f k (x),
k=2

with f k (x) an n-dimensional vector-valued homogeneous polynomial of degree k.

Then we have

X (Ψ ) =
∂x Ψ (x), Ax + f (x)
=
∂x Ψ (x), Ax + (
∂x Ψ+1 (x), Ax +
∂x Ψm (x), f 2 (x))
∞
 
 
k−1
+
∂x Ψ+k (x), Ax +
∂x Ψ+s (x), f k+1−s (x) .
k=2 s=0

By Lemma 7.1 and working in a similar way to the proof of Theorem 7.1, one
can choose Ψ (x) such that
∂x Ψ (x), Ax consists of resonant monomials. Then
we can choose Ψ+1 for which
∂x Ψ+1 (x), Ax +
∂x Ψ (x), f 2 (x) is composed of
resonant monomials. Successively, one can obtain Ψ+k for k = 2, 3, . . . such that

k−1

∂x Ψ+k (x), Ax +
∂x Ψ+s (x), f k+1−s (x) contains only the resonant monomi-
s=0
als. This proves statement (a).
(b) By statement (a), there exists an analytic function or a formal series Ψ (x) such
that (7.46) holds. On the contrary, if the statement does not hold, there exists some
7.6 Varieties and Normalization of Partially Integrable Systems 339

σ0 ∈ Mλ such that vσ0 = 0, and vσ = 0 for all σ ∈ Mλ satisfying |σ | < |σ0 |. Under
this hypothesis one has

X (Ψ (x)) = vσ0 x σ0 + vσ x σ + h.o.t. (7.52)
σ ∈Mλ , |σ |=|σ0 |, σ =σ0

where h.o.t is the sum of the resonant monomials of the form vσ x σ with |σ | > |σ0 |.
We assume without loss of generality that A is in the Jordan normal form.
By the classical Poincaré–Dulac normal form theorem there exists a distinguished
normalization
x = y + ϕ(y) = Λ(y) (7.53)

under which the vector field X is transformed to its Poincaré–Dulac normal form
vector field Y , with its associated differential system in the form

ẏ = Ay + g(y) = G(y).

Then we have

X (Ψ ) ◦ Λ(y) = vσ0 y σ0 + vσ y σ + h.o.t. (7.54)
σ ∈Mλ , |σ |=|σ0 |

and Y (y) = (∂ y Λ(y))−1 X ◦ Λ(y).

Set
Φ(y) = Ψ ◦ Λ(y).

Some calculations show that

Y (Φ(y)) = (∂ y Λ(y))−1 X ◦ Λ(y)(Φ(y)) = X (Ψ ) ◦ Λ(y). (7.55)

Write Φ(y) as
∞

Φ(y) = Φ j (y),
j=m

with m ∈ N and the Φ j ’s homogeneous polynomials of degree j. Clearly m ≤ |σ0 |.

If m < |σ0 |, using the methods in the proof of the classical Poincaré–Dulac normal
form theorem or of Theorem 7.18 we can prove that the homogeneous polynomials
Φ j for j < |σ0 | are composed of resonant monomials.
This last result shows that Φm (y) is a first integral of the vector field Ys =

∂ y , As y. Moreover, by the assumption of the theorem we get that the normal

form vector field Y has m λ functionally independent first integrals. It follows from
Proposition 7.7 that Φm (y) is a first integral of Y . By induction we obtain
340 7 Local Integrability of Differential Systems
⎛ ⎞ ⎛ ⎞
∞
 ∞

Y ⎝ Φ j (y)⎠ = Y ⎝ Φ j (y)⎠ .
j=m j=|σ0 |

This together with (7.52) and (7.54) yields


Y1 (Φ|σ0 | (y)) = vσ0 y σ0 + vσ y σ , (7.56)
σ ∈Mλ , |σ |=|σ0 |

where Y1 is the linear part of Y .

Since A is in the Jordan normal form, one has A = diag(A , A ) with the resonant
eigenvalues of A determined by A . Then by Proposition 7.8 and the assumption (H ),
one can prove that

Y1 (Φ|σ0 | (y)) = Y1 (Φ|σ0 |n (y)) =

∂ y Φ|σ0 |n (y), Ay,

where W|σ0 |n (y) is the nonresonant part of Φ|σ0 | (y). The last expression of these
equalities is either identically zero or composed of the nonresonant monomials,
whereas the right-hand side of (7.56) consists of the nonvanishing resonant mono-
mials, a contradiction. This shows that X (Φ(x)) ≡ 0, which concludes the proof of
the statement.
(c) As in the proof of (b) we can write A = diag(A , A ) and λ = (λ , λ ). Under
the assumption of the statement, we can prove that λ and λ are not resonant with
each other. Then by Theorem 7.28 it follows that the vector field X , i.e. system
(7.29), is transformed to its normal form Y on an invariant manifold, i.e.

ẏ  = A y  + g  (y  , y  )), ẏ  = A y  + g  (y  , y  ), (7.57)

through a near identity change of coordinates x = y + ϕ(y  ) = Φ(y), where y =

(y  , y  ) is the corresponding decomposition associated to λ.
Since Mλ has exactly m λ Q+ -linearly independent elements, let α1 , . . . , αm λ be
linearly independent elements of Mλ . Then

y α1 , . . . , y αmλ ,

are m λ functionally independent first integrals of Ys , the semisimple part of the linear
vector field Y1 in Y . By Proposition 7.7 it follows that the y α j ’s are the first integrals
of the vector field Y . Then
 −1 α j
Φ (x) = x α j + h.o.t., j = 1, . . . , m λ ,

are m λ functionally independent (analytic or formal) first integrals of X .

To prove that X has m λ functionally independent first integrals satisfying the
nonresonant conditions, we choose by Proposition 7.8
7.6 Varieties and Normalization of Partially Integrable Systems 341

Ψ j (x) = x α j + ψ j (x), j = 1, . . . , m λ , (7.58)

with ψ j (x) = o(|x||α j | ) and their resonant monomials arbitrary, such that
 ( j)
X (Ψ j (x)) = ψk x k , j = 1, . . . , d,
k∈Mλ

( j)
where the ψk ’s are polynomials in terms of the coefficients of X and of the ψi (x)’s
of degrees less than |k|. By the assumption of the statement and (b) one gets that
( j)
ψk ≡0 for all k ∈ Mλ and j ∈ {1, . . . , m λ }.

This proves that Ψ1 , . . . , Ψm λ are the functionally independent first integrals of system
(7.29). Taking all resonant monomials of the Ψ j (x)’s in (7.58) to be zero will achieve
our aim. This proves the statement.
This completes the proof of the theorem. 

7.6.2 Analytic Normalization of Partially Integrable Systems

In Sect. 7.5.1 we presented an equivalent characterization between the existence of

analytic normalizations and analytically integrable systems. Here we will study the
existence of analytic normalizations of partially integrable differential systems to
their normal forms on an invariant manifold.
Consider system (7.29) and, as in Sect. 7.6.2, let A = diag(A , A ) and λ =
(λ , λ ) with λ = (λ1 , . . . , λk ) and λ = (λk+1 , . . . , λn ) respectively the eigenvalues


of A and A and λ j = λs for j ∈ {1, . . . , k} and s ∈ {k + 1, . . . , n}. Write system

(7.29) in the form (7.50), i.e.

ẋ  = A x  + f  (x  , x  ), ẋ  = A x  + f  (x  , x  ).

Let X ∗ be the analytic vector field associated to system (7.50). Set


Mλ = m  ∈ Zk+ |
m  , λ  = 0, |m| ≥ 2 .

Assume that
(Hλ ) For any j ∈ {1, . . . , k} there exists an m  ∈ Mλ such that its jth component
does not vanish.
We remark that this assumption does not cause any restriction on λ.
The next result is Theorem 2 of Du et al. [131], which verifies the existence of
analytic normalization of partial analytically integrable systems with an additional
condition.
342 7 Local Integrability of Differential Systems

Theorem 7.29 Assume that λ = 0 and (Hλ ) holds, and that m  = 0 for any m =
(m  , m  ) ∈ Mλ . If the vector field X ∗ has k − 1 functionally independent analytic
first integrals, then Mλ has exactly k − 1 Q+ -linearly independent elements, and the
vector field X ∗ is analytically equivalent to its normal form on invariant manifolds
of the form
ẏ  = diag(λ1 , . . . , λk )y  (1 + q(y  )) + g ∗ (y  , y  ),
(7.59)
ẏ  = A y  + g  (y  , y  ),

with g ∗ (y  , 0) = 0, g  (y  , 0) = 0, where all monomials, say y q , of q(y  ) are reso-
nant, i.e.
q  , λ  = 0.

Note that Theorem 7.29 is an extension of Theorem 7.18 from analytic integrable
differential systems to partially analytic integrable systems.
In the proof of Theorem 7.29, Theorem 7.28 and the next result will play central
roles. For a proof of Theorem 7.30, see [230, Theorem 2.24], or [33].

Theorem 7.30 Assume that

•
m  , λ  − λs = 0 for m  ∈ Zk+ with |m  | ≥ 2 and s ∈ {k + 1, . . . , n},

• there exists a σ > 0 such that for any nonresonant monomials y q in ϕ  (y  ) and
ϕ  (y  ) of the normalization (7.51) sending system (7.50) to its distinguished normal
form on invariant manifold

ẏ  = A y  + g  (y  , y  ), ẏ  = A y  + g  (y  , y  ), (7.60)

one has
|
q  , λ  − λk | > σ, k ∈ {1, . . . , n},

• A y  + g  (y  , 0) = diag(λ1 , . . . , λk )y  (1 + q(y  )) with q = o(1) a scale function.

If system (7.50) is analytic, then the normalization (7.51) is convergent in a neigh-
borhood of the origin.

Proof of Theorem 7.29. To apply Theorems 7.28 and 7.30, the first step is to prove
that

q  , λ  = λ j , for q  ∈ Zk+ , |q  | ≥ 2 and j ∈ {k + 1, . . . , n}. (7.61)

The second step is to show that the partially integrable systems (7.50) have the special
normal form (7.59). The third step is to verify that there exists a σ > 0 such that if

q  , λ  − λ = 0 for q  ∈ Zk+ and  ∈ {1, . . . , n}, then

|
q  , λ  − λ | ≥ σ, for q  ∈ Zk+ and  ∈ {1, . . . , n}.

Thus the proof of Theorem 7.29 follows from Theorem 7.30. The details are omitted.
We refer the reader to the original proof given in [131, Theorem 2]. 
7.6 Varieties and Normalization of Partially Integrable Systems 343

7.6.3 Generic Divergence of Normalizations of Partially

Integrable Systems

In this subsection we continue to consider the convergence of the partially inte-

grable differential systems. As an example we introduce the results of Chen et al. in
[82, Theorem 1.2], which provide further relations among the first integrals, the
resonance of the eigenvalues and the normal forms.

Theorem 7.31 Given an analytic vector field Z in C2n with the origin a singu-
larity, let (λ, μ) = (λ1 , . . . , λn , μ1 , . . . , μn ) be the 2n-tuple of eigenvalues of Z at
the origin. Assume that λ j , μ j are nonzero and pairwise resonant, j = 1, . . . , n,
whereas λ1 , . . . , λn are not resonant. If Z has n functionally independent analytic
first integrals in a neighborhood of the origin, the following statements hold.
(a) The vector field Z is formally equivalent to

u̇ j = λ j u j (1 + q j (z 1 , . . . , z n )),
j = 1, . . . , n, (7.62)
v̇ j = μ j v j (1 + q j (z 1 , . . . , z n )),

where q j , j ∈ {1, . . . , n}, is a formal series in z 1 , . . . , z n with z s = u ns s vsm s and

n s , m s ∈ N pairwise relatively prime, and m s /n s = −λs /μs .
(b) If either the n eigenvalues λ belong to the Poincaré domain, or the formal series
q j are all the same and for all |
m, λ − λ j | = 0, |
m, λ − λ j | ≥ σ > 0, where
σ is a fixed constant, m ∈ Qn+ , then the equivalence in (a) is analytic.
(c) A formal series is a first integral of (7.62) in u 1 , . . . , u n , v1 , . . . , vn if and only
if it is a series in the n variables z 1 , . . . , z n .
(d) Let V be the set of vector fields which have the same linear part as Z . If there
is a vector field in V with a divergent normalization, then the generic vector
fields in V have this property.

Proof Statements (a), (b) and (c) can be proved using the arguments given in the
proof of Theorems 7.2 and 7.18, see Chen et al. [82, Theorem 1.2]. The details are
omitted.
(d) To prove this statement we need some knowledge of pluripolar sets. A set Λ ⊂ Cm
is called pluripolar if, for each z ∈ Λ, there exist a neighborhood
 U of z and a
plurisubharmonic function g defined on U such that Λ U ⊂ g −1 (−∞), see e.g.
Klimek [220] and Tsuji [423]. Given an open subset Ω ⊂ Cm , a function g : Ω →
[−∞, ∞) is plurisubharmonic if
• it is upper semicontinuous, i.e. {z ∈ Ω| g(z) < σ } is open for each σ ∈ R; and
• it is not identically equal to −∞ in any connected component of Ω; and
• for arbitrary z ∈ Ω
2π
1
g(z) ≤ g(z + eit y)dt,
2π 0
√
where i = −1, y ∈ Cm is any number such that z + cy ∈ Ω whenever |c| ≤ 1.
344 7 Local Integrability of Differential Systems

We remark that a pluripolar set in Cm is an m-dimensional zero Lebesgue measure

set. The countable union of pluripolar sets is again a pluripolar set.

A subharmonic function g defined on Cm has minimal growth if g(z) − log z

is bounded above as z → ∞. Let L be the set of plurisubharmonic functions
defined on Cm with minimal growth. For arbitrary Λ ⊂ Cm , define

VΛ (z) = sup{g(z)| g ∈ L , g|Λ ≤ 0}, z ∈ Cm .

This function is called the pluri-Green function over Λ. We can use it to estimate
polynomial functions, see e.g. Pérez-Marco [359, 360].

Lemma 7.13 (Bernstein–Walsh theorem). Assume that Λ ⊂ Cm is not pluripolar.

If P(z) is a polynomial of degree d, then for z ∈ Cm we have

|P(z)| ≤ PΛ exp (d VΛ (z)) ,

where PΛ is the supremum of the absolute values of P on Λ.

The inequality in Lemma 7.13 is a reformulation of the Bernstein–Walsh inequal-

ity, see e.g. Klimek [220, p. 182].
Applying Lemma 7.13 one obtains the next result, see Chen et al.
[82, Lemma 3.1].

Lemma 7.14 For the set V of vector fields given in Theorem 7.31 (d), let F be
any finite-dimensional affine space of V , then either all vector fields in F have
convergent distinguished normalizations, or only the vector fields in an exceptional
pluripolar subset of F have this property.

Proof The main idea of the proof comes from Pérez-Marco [360]. If the second
statement of the lemma holds, we are done.
Assume that the second statement is false. That is, there exists a non-pluripolar
subset of F in which every vector field has a convergent normalization. Let
X1 , . . . , Xm be m = 2n linearly independent vector fields without constant and
linear parts, and let Z0 be the linear part of Z . Let G be the m-dimensional linear
space spanned by

{Z0 + t1 X1 + . . . + tm Xm | t = (t1 , . . . , tm ) ∈ Cm }.

Clearly G ⊂ V and it is isomorphic to Cm . For each t ∈ Cm , we denote by Zt the

vector field in G .
Let S ⊂ Cm be a non-pluripolar set, and for arbitrary t ∈ S let the vector field
Zt ⊂ G have a convergent distinguished normalization. Set
!
S = Sr ,
r ≥1
7.6 Varieties and Normalization of Partially Integrable Systems 345

where Sr is a subset of S such that for each t ∈ Sr , the vector field Zt has a
convergent distinguished normalization at least in the polydisk Pr of radius 1/r , and
the normalization is bounded by 1 in Pr . Then at least one Sr is a non-pluripolar
set. Otherwise S is a pluripolar set, a contradiction.
Assume that the vector field Zt is of the form

ẋ j = λ j x j + p j (t, x, y), ẏ j = μ j y j + q j (t, x, y), j = 1, . . . , n,

where p j , q j are polynomials in t, and analytic in x, y. Let (x, y) = Φt (u, v) be the

normalization sending the vector field Zt to its normal form

u̇ j = λ j u j + ξ j (t, u, v), v̇ j = μ j v j + η j (t, u, v),

and the components of the transformation are

x j = u j + ϕ j (t, u, v), y j = v j + ψ j (t, u, v), j = 1, . . . , n.

Expanding w ∈ { p, q, ϕ, ψ, ξ, η} as Taylor series in (u, v)


w j (t, u, v) = w(k,)
j (t)u k v , j = 1, . . . , n,
k,∈Zn+

where u k = u k11 . . . u knn and v = v11 . . . vnn . Then similar to the proof of Theorems
7.8 and 7.1 we get

(
k, λ +
, μ − λ j )ϕ (k,)
j (t) = [ p j ](k,) − ξ (k,)
j (t)

n 
 
− ϕ (r,s)
j (t) rσ ξσ(k+eσ −r,−s) (t) + sσ ησ(k−r,+eσ −s) (t) ,
σ =1 (r,s)

(
k, λ +
, μ − μ j )ψ (k,)
j (t) = [q j ](k,) − η(k,)
j (t) (7.63)

n 
 (k+eσ −r,−s) 
− ψ (r,s)
j r σ ξσ + sσ ησ(k−r,+eσ −s) ,
σ =1 (r,s)

where [h j ](k,) , h ∈ { p, q}, are the coefficients of u k v in the re-expansion

 of
h j (t, u + ϕ, v + ψ) in u and v, eσ is the σ th unit vector in Rn , and runs over
+ such that (k − r,  − s) ∈ Z+ .
all integer vectors (r, s) in Z2n 2n

For (k, ) resonant, i.e.

k, λ +
, μ − λ j = 0, or
k, λ +
, μ − μ j = 0,

we have
ϕ (k,)
j (t) = 0, or ψ (k,)
j (t) = 0.
346 7 Local Integrability of Differential Systems

For (k, ) nonresonant, we choose ξ (k,)

j (t) = η(k,)
j (t) = 0. Then equations (7.63)
have a unique solution


n   
[ p j ](k,) − ϕ (r,s)
j (t) rν ξν(k+eν −r,−s) (t) + sν ην(k−r,+eν −s) (t)
ν=1 (r,s)
ϕ (k,) (t) = ,
j

k, λ +
, μ − λ j
(7.64)

n  (r,s)  
[q j ](k,) − ψ j (t) rν ξν(k+eν −r,−s) (t) + sν ην(k−r,+eν −s) (t)
ν=1 (r,s)
ψ (k,) (t) = .
j

k, λ +
, μ − μ j
(7.65)

So the distinguished normalization Φt (u, v) is

 
xj = uj + ϕ (k,)
j (t)u k v , yj = vj + ψ (k,)
j (t)u k v ,
(k,)∈Zn+ (k,)∈Zn+

with (k, ) nonresonant, and ϕ (k,)

j (t) and ψ (k,)
j (t) being defined respectively by
(7.64) and (7.65).
Write Φt (u, v) as 
Φt (u, v) = Φ j (t)(u, v) j ,
j∈Z2n
+

where the (u, v) j ’s are multiplicative indices. The above proof shows that Φ j (t) is a
vector-valued polynomial of degree no greater than | j|. In addition, it follows from
the choice of Sr that Φt is analytic. So, by the Cauchy inequality there exists a
ρ0 > 0 such that
−| j|
Ψ (t) := sup Φ j (t)∞ ρ0 < ∞, t ∈ Sr ,
j

where the norm  · ∞ is taken over the sum of absolute values of all components of
a vector. Now the non-pluripolar set Sr can be expressed as
∞
!
Sr = Ts , Ts := {t ∈ Sr | Ψ (t) ≤ s},
s=1

in which at least one set, say Ts0 , is non-pluripolar. Let Ω ⊂ Ts0 be a non-pluripolar
and compact set such that there is a ρ1 > 0 such that for arbitrary t ∈ Ω and all
j ∈ Z2n
+ we have
| j|
Φ j (t)∞ ≤ ρ1 .
7.6 Varieties and Normalization of Partially Integrable Systems 347

Thus we get from the Bernstein–Walsh lemma that for any compact subset S ⊂ Cm
and | j| ≥ 2, there exists a ρ2 > 0 depending only on S for which
 
| j| | j|
Φ j S ≤ Φ j Ω exp | j| max VΩ (t) ≤ ρ1 ρ2 ,
t∈S

where Φ j Ω = max Φ j (t)∞ . This implies that for any element t in a compact
t∈Ω
subset of Cm , Φt (u, v) is convergent in the polydisk
 
(u, v)| |u i |, |vi | ≤ min ρ1−1 , ρ2−1 , i = 1, . . . , n .

Hence for all t ∈ Cm , Φt (u, v) is an analytic diffeomorphism in a neighborhood of

the origin of the (u, v) space. This proves the lemma. 
From Lemma 7.14 we can finish the proof of statement (d). 
This completes the proof of Theorem 7.31 

7.7 Other Results on Local Integrability

In this last section of the chapter we summarize some other results on local integra-
bility of differential systems, which were not mentioned in the previous sections.
A. Local Integrability of Planar Differential Systems
In 2001 Li et al. [234] presented necessary and sufficient conditions on the existence
of a formal meromorphic first integral for planar analytic differential systems which
have an elementary singularity.
Llibre and Valls [290] in 2016 studied the analytic Liénard differential system

ẋ = y, ẏ = −g(x) − f (x)y, (7.66)

pq
with g(0) = 0 and f (0) = 0, and under the assumption g  (0) = − f (0)2
( p − q)2
they provided necessary and sufficient conditions for system (7.66) to have an analytic
first integral in a neighborhood of the origin.
In case g(x) = x and f (x) ≡ 0, Llibre and Valls [290] showed that system (7.66)
" x in a neighborhood of the origin if and only if F(x) =
has an analytic first integral
F(−x), where F(x) = 0 f (s)ds.
For g(x) = ±x, Llibre and Valls [278] in 2009 provided necessary and sufficient
conditions for the Liénard analytic different system (7.66) to have an analytic first
integral in a neighborhood of the origin.
In [6] the authors investigated the analytic integrability of the planar analytic
differential system
348 7 Local Integrability of Differential Systems
 
ẋ = ∂ y h(x, y)K (h, y n ) + y n−1 Ψ (h, y n ) ξ(x, y), ẏ = −∂x h(x, y)K (h, y n )ξ(x, y),
(7.67)
with n ∈ N and the origin O a singularity, where h, K , Ψ, and ξ are analytic functions
defined in a neighborhood of the origin satisfying ∂x h ≡ 0, K (O) = 0 or Ψ (O) = 0.
In the proof of this result they applied the following result, obtained in the same paper.
A planar analytic vector field X has an analytic first integral defined in a neigh-
borhood of a singularity if and only if there exist two analytic functions u and v
∂(u, v)
with (x, y) = 0 in a neighborhood of 0, except possibly in a null Lebesgue
∂(x, y)
measure subset, such that X (u) = F(u, v)X (v), where F is an analytic function
in a neighborhood of (u(0), v(0)).
∂(u, v)
We remark that if (0) = 0, then the transformation (u, v) = (u(x, y),
∂(x, y)
v(x, y)) is not defined at the origin as used in the proof of [6], so the proof needs
more consideration than the original.
For the local integrability of planar analytic differential systems, Ferčeca et al.
[146] studied the existence of the local analytic first integrals of the complex planar
systems
ẋ = x − X (x, y), ẏ = −y + Y (x, y), (x, y) ∈ C2 ,

with X and Y homogeneous polynomials of degree m ∈ N \ {1}. With the help of

computer algebra they provided sufficient conditions for the existence of a local
analytic first integral of the systems with m = 5 by using the Darboux theory of
integrability and reversibility. They also proposed an approach to find reversible
Lotka–Volterra systems with homogeneous nonlinearities.
Recall that a vector field X is time reversible if there exists a diffeomorphism ψ
satisfying ψ ◦ ψ = I d such that ψ∗ X = −X ◦ ψ, where I d denotes the identity
map, and as before ψ∗ represents the tangent map. The map ψ is called an involution.
For the local integrability of the special planar differential systems

ẋ = P1 (x, y), ẏ = Q m (x, y), (7.68)

where P1 is linear and Q m is a polynomial of degree m, Llibre and Valls [282]

characterized the existence of their local analytic first integrals when m = 2. For
m = 3, Pan et al. in 2014 obtained a necessary and sufficient condition for system
(7.68) to have a C ∞ first integral provided P1 (x, y) = x. For m = 3 and P1 (x, y) =
y, Giné and Valls [188] separated the space of parameters of system (7.68) into
different regions, then they presented either necessary or sufficient conditions in
each of the regions for systems (7.68) to have a local analytic first integral.
There have been numerous studies of planar analytic differential systems having
an isolated nilpotent singularity at the origin

ẋ = y + X 2 (x, y), ẏ = Y2 (x, y), (7.69)

7.7 Other Results on Local Integrability 349

with X 2 , Y2 = o(|(x, y)|) analytic. Andreev [7] provided a necessary and sufficient
condition for the origin of system (7.69) to be monodromic, which is stated in nearly
all textbooks on the qualitative theory of ordinary differential equations. We do not
state it here.
After the change of variables (x, y) → (x, y − F(x)) with F(x) the unique solu-
tion of y + X 2 (x, y) = 0, and a rescaling, systems (7.69) having a monodromic
nilpotent singularity at the origin can be brought to

ẋ = −y(1 + X 1 (x, y)), ẏ = f (x) + yg(x) + y 2 Y0 (x, y),

where X 1 = o(1) and Y0 = O(1). Takens [421] in 1974 proved that this system can
be further formally transformed to a generalized Liénard system

ẋ = −y, ẏ = a(x) + yb(x),

with a and b formal series. In 2002 Strózyna and Zoła̧dek verified that this normal
form could be obtained through an analytic change of variables, and this normal form
can be further reduced to

ẋ = −y, ẏ = x 2n−1 + yb(x), (7.70)

with b(x) = bm x m +h.o.t. an analytic function.

Moussu [343] in 1982 presented a necessary and sufficient condition for system
(7.70) to have a center at the origin, which is that if the origin of system (7.70) is
monodromic, i.e. either m > n − 1, or m = n − 1 and bm − 4n < 0, then the origin
is a center if and only if b(x) is an odd function.
Note that for a nilpotent center there cannot exist an analytic first integral defined
in a neighborhood of the center, see Moussu [342], or Villarini [433]. In fact, Moussu
[343] also proved that general systems (7.70) having a nilpotent center do not have
analytic first integrals in a neighborhood of the center, see also Villarini [434].
Strózyna and Zoła̧dek [415] in 2002 presented a characterization for a nilpotent
center to have an analytic first integral. This result can be stated as that system (7.69)
with a center at the origin has an analytic first integral in a neighborhood of the origin
if and only if it is analytically orbitally equivalent to the system

ẋ = y, ẏ = −x 2n−1 .

However, Villarini [434] in 1995 proved that there exist examples for which no finite
order jet can provide sufficient analyticity conditions on the existence of an analytic
first integral around the nilpotent center. So in general Strózyna and Zoła̧dek’s results
are difficult to use to characterize the existence of analytic first integrals for a nilpotent
center of an analytic differential system.
García and Giné [162] in 2010 obtained a sufficient condition on the existence
of analytic first integrals: If system (7.70) with a center at the origin has an analytic
first integral, then either m = 2n − 1 with  ∈ N or b(x) ≡ 0.
350 7 Local Integrability of Differential Systems

In [67] Chavarriga et al. characterized the form of first integrals (formal or analytic)
of system (7.69), which has a nilpotent center at the origin. The concrete results are
the following.
• If X 2 = y f (x, y 2 ) and Y2 = g(x, y 2 ), then X has a local analytic first integral of
the form H (x, y) = y 2 + h(x, y) with h = o(|(x, y)|).
• If X has a formal (resp. analytic) first integral, then it has a first integral of the
form H (x, y) = y 2 + h(x, y) where h has no terms of order less than 2.
As an application of these results they characterized the nilpotent centers of X with
X 2 and Y2 homogeneous polynomials of degree three, which has an analytic first
integral.
Ferčeca et al. [145] studied the center of the system

ẋ = px + P(x, y), ẏ = −qy + Q(x, y), (x, y) ∈ C2 , (7.71)

with p, q ∈ N and P, Q ∈ C[x, y] without constant and linear terms. The origin of
system (7.71) is a p : −q resonant singularity, and it is called a resonant center if
there exists a local analytic first integral of the form

H (x, y) = x q y p + h.o.t.

In case p = 1 and q = 4, the authors obtained some sufficient conditions for system
(7.71) to have a 1 : −4 resonant center.
B. The Local Integrability of Higher-Dimensional Differential Systems
In [21] Aziz and Christopher investigated the local integrability and linearizability
of three-dimensional Lotka–Volterra equations at the origin

ẋ = P(x, y, z) = x(λ + ax + by + cy),

ẏ = Q(x, y, z) = y(μ + d x + ey + f y), (7.72)
ż = R(x, y, z) = z(ν + gx + hy + ky),

with λ, μ, ν = 0. If the eigenvalues (λ, μ, ν) satisfy k1 λ + k2 μ + k3 ν = 0 for

(k1 , k2 , k3 ) ∈ Z3+ not zero, we say that the origin of system (7.72) is (λ : μ : ν)
resonant.
For the (1 : −1 : 1), (2 : −1 : 1) and (1 : −2 : 1) resonances Aziz and Christo-
pher [21] obtained necessary and sufficient conditions for both the integrability and
linearizability of system (7.72). In their proof they used the Darboux theory of inte-
grability together with inverse Jacobian multipliers.
In 2013 Hu et al. [210] studied the systems

ẋ1 = P1 (x1 , x2 , x3 ), ẋ2 = x2 + P2 (x1 , x2 , x3 ), ẋ3 = −x3 + P3 (x1 , x2 , x3 ),

(7.73)
with P1 , P2 , P3 polynomials of degree m and without constant and linear terms, and
obtained some conditions on the existence of two first integrals of the form
7.7 Other Results on Local Integrability 351

H1 (x) = x1 + h.o.t., H2 (x) = x2 x3 + h.o.t.

and also conditions on the reversibility and Darboux integrability of system (7.73).
As an application they considered a quadratic system of the form (7.73).
In 2016 Romanovski and Shafer [374] pursued their study of the complete integra-
bility and time-reversibility of the three-dimensional polynomial differential systems
(7.73). Note that the origin of system (7.72) is (0 : 1 : −1) resonant.
For the well-known Biachi VI0 and VII0 system

ẋ = −x 2 + (z + 1)y 2 , ẏ = −4(z + 1) + x yz, ż = −yz(z + 2),

Llibre and Valls [279] proved that it has no first integral which is a polynomial,
rational, Darboux, or analytic function. Consequently the system is not integrable
inside these classes. In the same paper they also proved that this system has no
exponential factors and has only the irreducible Darboux polynomials z and z + 2.
Combining these results with Zhang [477] one can easily verify that the system is
not Darboux integrable.
In [277] Llibre and Valls analyzed the existence of local and global analytic first
integrals of the Chua system
 
7 41 7 6
ẋ = a z − x 3 − x 2 − b , ẏ = −z, ż = − x+y+ ,
10 2 10 25

for different choices of the values of a and b.

For other related results we refer to [45, 46, 62, 114, 176, 177, 187, 325] and
the references therein.
References

1. Abraham, R., Marsden, J.E.: Foundations of Mechanics. The Benjamin/Cummings Publishing

Company INC, London (1978)
2. Abramowitz, M., Stegun, I.A.: Handbook of Mathematical Functions with Formulas. Graphs,
and Mathematical Tables. Dover, New York (1992). Reprint of the 1972
3. Adler, M., van Moerbeke, P.: The algebraic integrability of geodesic flow on SO(4). Invent.
Math. 67, 297–331 (1982)
4. Algaba, A., Fernández-Sánchez, F., Merino, M., Rodríguez-Luis, A.J.: On Darboux poly-
nomials and rational first integrals of the generalized Lorenz system. Bull. Sci. Math. 138,
317–322 (2014)
5. Algaba, A., Fernández-Sánchez, F., Merino, M., Rodríguez-Luis, A.J.: Comments on Global
dynamics of the generalized Lorenz systems having invariant algebraic surfaces. Phys. D 266,
80–82 (2014)
6. Algaba, A., García, C., Reyes, M.: A note on analytic integrability of planar vector fields.
Eur. J. Appl. Math. 23, 555–562 (2012)
7. Andreev, A.F.: Solution of the problem of the center and the focus in one case. Akad. Nauk
SSSR. Prikl. Mat. Meh. 17, 333–338 (1953). (in Russian)
8. Andreev, A.F.: Investigation of the behaviour of the integral curves of a system of two differ-
ential equations in the neighbourhood of a singular point. Trans. Am. Math. Soc. 8, 183–207
(1958)
9. Anosov, D.V., Arnold, V.I.: Dynamical Systems I. Springer, Berlin (1988)
10. Apostol, T.M.: Mathematical Analysis, 2nd edn. Addison-Wesley Publishing Company Inc,
Amsterdam (1974)
11. Arnold, V.I.: Ordinary Differential Equations. Springer, Berlin (2006)
12. Arnold, V.I.: Geometric Methods in the Theory of Ordinary Differential Equations, 2nd edn.
Springer, New York (1988)
13. Arnold, V.I.: Mathematical Methods of Classical Mechanics. Springer, New York (1989)
14. Arnold V.I.: Remarks on the ecstatic points of plane curves. The Gelfand Mathematical Sem-
inars (1993–1995)
15. Aroca J.M., Cano J., Feng R., Gao, X.S.: Algebraic general solutions of algebraic ordinary
differential equations. In: ISSAC’05, pp. 29–36. ACM, New York (2005) (electronic)
16. Athanassopoulos, K.: Divergence of C 1 vector fields and nontrivial minimal sets on 2-
manifolds. J. Differ. Equ. 243, 24–35 (2007)
17. Artés, J.C., Grünbaum, B., Llibre, J.: On the invariant straight lines of polynomial differential
systems, Panar nonlinear dynamical systems (Delft, 1995). Diff. Equ. Dyn. Syst. 5(1997),
317–327 (1997)
18. Artés, J.C., Grünbaum, B., Llibre, J.: On the number of invariant straight lines for polynomial
differential systems. Pac. J. Math. 184, 207–230 (1998)

© Springer Nature Singapore Pte Ltd. 2017 353

X. Zhang, Integrability of Dynamical Systems: Algebra and Analysis,
Developments in Mathematics 47, DOI 10.1007/978-981-10-4226-3
354 References

19. Artés, J.C., Llibre, J.: On the number of slopes of invariant straight lines for polynomial
differential systems. J. Nanjing Univ. Math. Biquarterly 13, 143–149 (1996)
20. Artin, M.: Algebra. Prentice Hall Inc, Englewood Cliffs (1991)
21. Aziz, W., Christopher, C.: Local integrability and linearizability of three-dimensional Lotka-
Volterra systems. Appl. Math. Comp. 219, 4067–4081 (2012)
22. Baider, A., Churchill, R.C., Rod, D.L., Singer, M.F.: On the infinitesimal geometry of inte-
grable systems. Mechanics Day (Waterloo. ON, 1992), Fields Institute Communications, vol.
7, pp. 5–56. American Mathematical Society, Providence, RI (1996)
23. Bao, J., Yang, Q.: Darboux integrability of the stretch-twist-fold flow. Nonlinear Dyn. 76,
797–807 (2014)
24. Barreira, L., Valls, C.: Conjugacies between linear and nonlinear non-uniform contractions.
Ergod. Theory Dyn. Syst. 28, 1–19 (2008)
25. Barreira, L., Valls, C.: A Grobman-Hartman theorem for nonuniformly hyperbolic dynamics.
J. Differ. Equ. 228, 285–310 (2006)
26. Barreira, L., Valls, C.: Conjugacies for linear and nonlinear perturbations of nonuniform
behavior. J. Funct. Anal. 253, 324–358 (2007)
27. Bautin N.N.: On the number of limit cycles which appear with the variation of the coefficients
from an equilibrium position of focus or center type. Mat. Sb. 30(72), 181–196 (1952); English
transl. in Math. USSR-Sb. 100, 397–413 (1954),
28. Bedford, E., Kim, K.: Linear recurrences in the degree sequences of monomial mappings.
Ergod. Theory Dyn. Syst. 28, 1369–1375 (2008)
29. Belitskii, G.R.: Smooth equivalence of germs of vector fields with one zero or a pair of purely
imaginary eigenvalues. Funct. Anal. Appl. 20(4), 253–259 (1986)
30. Belitskii, G.R., Tkachenko, V.: One-dimensional Functional Equations. Birkhäuser, Berlin
(2003)
31. Berrone, L.R., Giacomini, H.J.: On the vanishing set of inverse integrating factors. Qual.
Theory Dyn. Syst. 1, 211–230 (2000)
32. Berrone, L.R., Giacomini, H.J.: Inverse Jacobian multipliers. Rend. Circ. Mat. Palermo (Serie
II) LII, 77–130 (2003)
33. Bibikov, YuN: Local Theory of Nonlinear Analytic Ordinary Differential Equations. Lecture
Notes in Math, vol. 702. Springer, Berlin (1979)
34. Blasius, H.: Grenzschichten in flussigkeiten mit kleiner reibung. Z. Math. Phys. 56, 1–37
(1908)
35. Blázquez-Sanz, D., Pantazi, Ch.: A note on the Darboux theory of integrability of non-
autonomous polynomial differential systems. Nonlinearity 25, 2615–2624 (2012)
36. Blé, G., Castellanos, V., Llibre, J., Quilantán, I.: Integrability and global dynamics of the
May-Leonard model. Nonlinear Anal. Real World Appl. 14, 280–293 (2013)
37. Bleecker, D., Csordas, G.: Basic Partial Differential Equations. Van Nostrand Reinhold, New
York (1992)
38. Bodin, A.: Reducibility of rational functions in several variables. Israel J. Math. 164, 333–347
(2008)
39. Bogoyavlenskii, O.I.: Integrable Euler equations on Lie algebras arising in problems of math-
ematical physics. Math. USSR Izvestiya 25, 207–257 (1985)
40. Bostan, A., Chéze, G., Cluzeau, T., Weil, J.-A.: Efficient algorithms for computing rational
first integrals and Darboux polynomials of planar polynomial vector fields. Math. Comput.
85, 1393–1425 (2016)
41. Bott, R., Tu, L.W.: Differential Forms in Algebraic Topology, Graduate Texts in Mathematics,
vol. 82. Springer, New York (1982)
42. Bruno, A.D.: Analytic form of differential equations I. Trans. Moscov. Math. Soc. 25, 131–288
(1971)
43. Bruno, A.D.: Analytic form of differential equations II. Trans. Moscov. Math. Soc. 26, 199–
239 (1972)
44. Bruno, A.D.: Local Methods in Nonlinear Differential Equations. Springer, Berlin (1989)
References 355

45. Bruno, A.D., Edneral, V.F.: On integrability of the Euler-Poisson equations. J. Math. Sci. 152,
479–489 (2008)
46. Bruno, A.D., Edneral, V.F.: On the integrability of a planar system of ODE’s near a degenerate
stationary point. J. Math. Sci. 168, 326–333 (2010)
47. Bruns, H.: Uber die Integrale des Vielkörper-Problems. Acta Math. 11, 25–96 (1887)
48. Buică, A., García, I.A.: Inverse Jacobian multipliers and first integrals for nonautonomous
differential systems. Z. Angew. Math. Phys. 66, 573–585 (2015)
49. Buică, A., García, I.A., Maza, S.: Existence of inverse Jacobian multipliers around Hopf
points in R3 : emphasis on the center problem. J. Differ. Equ. 252, 6324–6336 (2012)
50. Buică, A., García, I.A., Maza, S.: Multiple Hopf bifurcation in R3 and inverse Jacobian
multipliers. J. Differ. Equ. 256, 310–325 (2014)
51. Cairó, L., Feix, M.R., Llibre, J.: Integrability and algebraic solutions for planar polynomial
differential systems with emphasis on the quadratic systems. Resenhas 4, 127–161 (1999)
52. Cairó, L., Llibre, J.: Integrability of the 2D Lotka-Volterra system via polynomial first integrals
and polynomial inverse integrating factors. J. Phys. A 33, 2407–2417 (2000)
53. Camacho, C., Lins, Neto A.: Geometric Theory of Foliations. Translated from the Portuguese
by Sue E. Goodman. Birkhäuser, Boston Inc, Boston, MA (1985)
54. Campillo, A., Carnicer, M.M.: Proximity inequalities and bounds for the degree of invariant
curves by foliations of PC2 . Trans. Am. Math. Soc. 349, 2211–2228 (1997)
55. Campillo, A., Carnicer, M.M., de la Fuente, J.G.: Invariant curves by vector fields on algebraic
varieties. J. Lond. Math. Soc. 62(2), 56–70 (2000)
56. Cao, J., Llibre, J., Zhang, X.: Darboux integrability and algebraic limit cycles for a class of
polynomial differential systems. Sci. China Math. 57, 775–794 (2014)
57. Cao, J., Zhang, X.: Dynamics of the Lorenz system having an invariant algebraic surface. J.
Math. Phys. 48, 082702 (2007)
58. Cariñena1, J.F., Guha, P., Rañada1, M.F.: Higher-order Abel equations: Lagrangian formalism,
first integrals and Darboux polynomials. Nonlinearity 22, 2953–2969 (2009)
59. Carnicer, M.M.: The Poincaré problem in the nondicritical case. Ann. Math. 140, 289–294
(1994)
60. Casale, G., Duval, G., Maciejewski, A.J., Przybylska, M.: Integrability of Hamiltonian systems
with homogeneous potentials of degree zero. Phys. Lett. A 374, 448–452 (2010)
61. Cerveau, D.: Lins, Neto A.: Holomorphic foliations in CP(2) having an invariant algebraic
curve. Ann. Inst. Fourier 41, 883–903 (1991)
62. Chandrasekar, V.K., Senthilvelan, M., Lakshmanan, M.: On the complete integrability and
linearization of nonlinear ordinary differential equations. V. Linearization of coupled second-
order equations. Proc. R. Soc. A 465, 2369–2389 (2009)
63. Chavarriga J.: A new example of a quartic algebraic limit cycle for quadratic systems, Uni-
versitat de Lleida, preprint (1999)
64. Chavarriga, J., García, I.A., Giné, J.: On Lie’s symmetries for planar polynomial differential
systems. Nonlinearity 14, 863–880 (2001)
65. Chavarriga, J., García, I.A., Llibre, J.: Żoła̧dek, H.: Invariant algebraic curves for the cubic
Liénard system with linear damping. Bull. Sci. Math. 130, 428–441 (2006)
66. Chavarriga, J., Giacomini, H., Giné, J., Llibre, J.: Darboux integrability and the inverse inte-
grating factor. J. Differ. Equ. 194, 116–139 (2003)
67. Chavarriga, J., Giacomini, H., Giné, J., Llibre, J.: Local analytic integrability for nilpotent
centers. Ergod. Theory Dyn. Syst. 23, 417–428 (2003)
68. Chavarriga, J., Giacomini, H., Giné, J., Llibre, J.: On the integrability of two-dimensional
flows. J. Differ. Equ. 157, 163–182 (1999)
69. Chavarriga, J., Giacomini, H., Llibre, J.: Uniqueness of algebraic limit cycles for quadratic
systems. J. Math. Anal. Appl. 261, 85–99 (2001)
70. Chavarriga, J., Giné, J.: Integrable systems via inverse integrating factor. Extracta Math. 13,
41–60 (1998)
71. Chavarriga, J., Giné, J.: Integrability of cubic systems with degenerate infinity. Diff. Eqns.
Dyn. Syst. 6, 425–438 (1998)
356 References

72. Chavarriga, J., Giné, J., Grau, M.: Integrable systems via polynomial inverse integrating
factors. Bull. Sci. Math. 126, 315–331 (2002)
73. Chavarriga, J., Grau, M.: A family of non-Darboux integrable quadratic polynomial differen-
tial systems with algebraic solutions of arbitrarily high degree. Appl. Math. Lett. 16, 833–837
(2003)
74. Chavarriga, J., Llibre, J.: Invariant algebraic curves and rational first integrals for planar
polynomial vector fields. J. Differ. Equ. 169, 1–16 (2001)
75. Chavarriga, J., Llibre, J., Moulin-Ollagnier, J.: On a result of Darboux. LMS J. Comput. Math.
4, 197–210 (2001)
76. Chavarriga, J., Llibre, J., Sorolla, J.: Algebraic limit cycles of degree 4 for quadratic systems.
J. Differ. Equ. 200, 206–244 (2004)
77. Chavarriga, J., Llibre, J., Sotomayor, J.: Algebraic solutions for polynomial systems with
emphasis in the quadratic case. Expos. Math. 15, 161–173 (1997)
78. Chavarriga, J., Sabatini, M.: A survey of isochronous centers. Qual. Theory Dyn. Syst. 1,
1–70 (1999)
79. Cheb-Terrab, E.S., Roche, A.D.: Abel ODE’s: equivalence and integrable classes. Comput.
Phys. Commun. 130, 204–231 (2000)
80. Chen, C., Cao, J., Zhang, X.: The topological structure of the Rabinovich system having an
invariant algebraic surface. Nonlinearity 21, 211–220 (2008)
81. Chen, G., Ueta, T.: Yet another chaotic attractor. Int. J. Bifur. Chaos 9, 1465–1466 (1999)
82. Chen, J., Yi, Y., Zhang, X.: First integrals and normal forms for germs of analytic vector
fields. J. Differ. Equ. 245, 1167–1184 (2008)
83. Chen, K.T.: Equivalence and decomposition of vector fields about an elementary critical point.
Am. J. Math. 85, 693–722 (1963)
84. Cherkas, L.A., Zhilevich, L.I.: Some tests for the absence or uniqueness of limit cycles. Differ.
Equ. 6, 891–897 (1970)
85. Cherkas, L.A., Zhilevich, L.I.: The limit cycles of certain differential equations. Differ. Equ.
8, 924–929 (1972)
86. Chéze, G.: Computation of Darboux polynomials and rational first integrals with bounded
degree in polynomial time. J. Complex. 27, 246–262 (2011)
87. Chéze, G.: Darboux theory of integrability in the sparse case. J. Differ. Equ. 257, 601–609
(2014)
88. Chéze, G.: Bounding the number of remarkable values via Jouanolou’s theorem. J. Differ.
Equ. 258, 3535–3545 (2015)
89. Chicone, C.: Ordinary Differential Equations with Applications. Springer, New York (2006)
90. Chirka, E.M.: Complex Analytic Sets. Kluwer Academic Publishers, Dordrecht, The Nether-
lands (1989)
91. Chow, S.N., Hale, J.K.: Methods of Bifurcation Theory. Springer, New York (1982)
92. Chow, S.N., Li, C., Wang, D.: Normal Forms and Bifurcation of Planar Vector Fields. Cam-
bridge University Press, Cambridge (1994)
93. Christodoulides, Y.T., Damianou, P.A.: Darboux polynomials for Lotka-Volterra systems in
three dimensions. J. Nonlinear Math. Phys. 16, 339–354 (2009)
94. Christopher, C.: Invariant algebraic curves and conditions for a center. Proc. R. Soc. Edinb.
124A, 1209–1229 (1994)
95. Christopher, C.: Liouvillian first integrals of second order polynomial differential equations.
Electron. J. Differ. Equ. 49, 1–7 (1999)
96. Christopher, C.: Polynomial vector fields with prescribed algebraic limit cycles. Geom. Ded-
icata 88, 255–258 (2001)
97. Christopher, C., Li, C.: Limit Cycles of Differential Equations. Birkhäuser, Basel (2007)
98. Christopher, C., Llibre, J.: Algebraic aspects of integrability for polynomial systems. Qual.
Theory Dyn. Syst. 1, 71–95 (1999)
99. Christopher, C., Llibre, J.: Integrability via invariant algebraic curves for planar polynomial
differential systems. Ann. Differ. Equ. 16, 5–19 (2000)
References 357

100. Christopher, C., Llibre, J.: A family of quadratic polynomial differential systems with invariant
algebraic curves of arbitrarily high degree without rational first integrals. Proc. Am. Math.
Soc. 130, 2025–2030 (2002)
101. Christopher, C., Llibre, J., Pantazi, Ch., Walcher, S.: Inverse problems in Darboux’ theory of
integrability. Acta Appl. Math. 120, 101–126 (2012)
102. Christopher, C., Llibre, J., Pantazi, Ch., Walcher, S.: Darboux integrating factors: inverse
problems. J. Differ. Equ. 250, 1–25 (2011)
103. Christopher, C., Llibre, J., Pantazi, Ch., Walcher, S.: Inverse problems for invariant algebraic
curves: explicit computations. Proc. R. Soc. Edinb. Sect. A 139, 287–302 (2009)
104. Christopher, C., Llibre, J., Pantazi, Ch., Walcher, S.: Inverse problems for multiple invariant
curves. Proc. R. Soc. Edinb. Sect. A 137, 1197–1226 (2007)
105. Christopher, C., Llibre, J., Pantazi, Ch., Zhang, X.: Darboux integrability and invariant alge-
braic curves for planar polynomial systems. J. Phys. A 35, 2457–2476 (2002)
106. Christopher, C., Llibre, J., Pereira, J.V.: Multiplicity of invariant algebraic curves in polyno-
mial vector fields. Pac. J. Math. 229, 63–117 (2007)
107. Christopher, C., Llibre, J., Świrszcz, G.: Invariant algebraic curves of large degree for quadratic
system. J. Math. Anal. Appl. 303, 450–461 (2005)
108. Christopher, C., Mardeěic, P., Rousseau, C.: Normalizable, integrable and linearizable saddle
points for complex quadratic systems in C2 . J. Dyn. Control Syst. 9, 311–363 (2003)
109. Christopher, C., Rousseau, C.: Normalizable, integrable and linearizable saddle points in the
Lotka-Volterra system. Qual. Theory Dyn. Syst. 5, 11–61 (2004)
110. Christopher, J., Schlomiuk, D.: On general algebraic mechanisms for producing centers in
polynomial differential systems. J. Fixed Point Theory Appl. 3, 331–351 (2008)
111. Cima, A., Gasull, A.: Ma nosa, V.: Global periodicity and complete integrability of discrete
dynamical systems. J. Differ. Equ. Appl. 12, 697–716 (2006)
112. Studying discrete dynamical systems through differential equations: Cima, A., Gasull, A.,
Ma nosa, V.J. J. Differ. Equ. 244, 630–648 (2008)
113. Cima, A., Llibre, J.: Bounded polynomial systems. Trans. Am. Math. Soc. 318, 557–579
(1990)
114. Colak, I.E., Llibre, J., Valls, C.: Local analytic first integrals of planar analytic differential
systems. Phys. Lett. A 377, 1065–1069 (2013)
115. Coll, B., Ferragut, A., Llibre, J.: Polynomial inverse integrating factors for quadratic differ-
ential systems. Nonlinear Anal. 73, 881–914 (2010)
116. Coll, B., Ferragut, A., Llibre, J.: Phase portraits of the quadratic systems with a polynomial
inverse integrating factor. Int. J. Bifur. Chaos 19, 765–783 (2009)
117. Cong, W., Llibre, J., Zhang, X.: Generalized rational first integrals of analytic differential
systems. J. Differ. Equ. 251, 2770–2788 (2011)
118. Coolidge, J.L.: A Treatise on Algebraic Plane Curves. Dover Publications, New York (1959)
119. Cozma, D., Subă, A.: Partial integrals and the first focal value in the problem of centre.
Nonlinear Diff. Equ. Appl. 2, 21–34 (1995)
120. Culver, W.J.: On the existence and uniqueness of the real logarithm of a matrix. Proc. Am.
Math. Soc. 17, 1146–1151 (1966)
121. Dai, G.: Two estimations of the number of invariant straight lines for nth polynomial differ-
ential systems. Acta Math. Sci. 16, 232–240 (1996)
122. Darboux, G.: Mémoire sur les équations différentielles algébriques du premier ordre et du
premier degré (Mélanges). Bull. Sci. Math. 2, 60–96; 123–144; 151–200 (1878)
123. Darboux, G.: Del’emploi des solutions particulières algébriques dans l’intégration des sys-
tèmes d’équations différentielles algébriques. C.R. Math. Acad. Sci. Paris 86, 1012–1014
(1878)
124. Deng, X., Chen, A.: Invariant algebraic surfaces of the Chen system. Int. J. Bifur. Chaos 21,
1645–1651 (2011)
125. Dimca, A.: Topics on Real and Complex Singularities: An Introduction. Advanced Lectures
in Math. Vieweg, Braunschweig (1987)
358 References

126. Ding, T.R., Li, C.: Ordinary Differential Equations (in Chinese). Higher Education Press,
Beijing (2005)
127. Dobrovol’skii, V.A., Lokot’, N.V., Strelcyn, J.-M.: Mikhail Nikolaevich Lagutinskii (1871–
1915): un mathématicien méconnu. Historia Math. 25, 245–264 (1998)
128. do Carmo, M.P.: Riemannian Geometry. Birkhäuser, Boston (1992)
129. do Carmo, M.P.: Differential Geometry of Curves and Surfaces. Prentice-Hall Inc., New Jersey
(1976)
130. Dolov, M.V.: Limit cycles in the case of the center. Differ. Uravn. 8, 1691–1692 (1972)
131. Du, Z., Romanovski, V.R., Zhang, X.: Varieties and analytic normalizations of partially inte-
grable systems. J. Differ. Equ. 260, 6855–6871 (2016)
132. Dufour, J.P., Zung, N.T.: Poisson Structures and Their Normal Forms. Progress in Math.
Birkhäuser, Basel (2005)
133. Dukarić, M., Oliveira, R., Romanovski, V.G.: Local integrability and linearizability of a
(1 : −1 : −1) resonant quadratic system. J. Dyn. Differ. Equ. doi:10.1007/s10884-015-
9486-2
134. Dumortier, F., Llibre, J., Artés, J.C.: Qualitative Theory of Planar Differential Systems. Uni-
versiText. Springer, New York (2006)
135. Duval, G., Maciejewski, A.J.: Jordan obstruction to the integrability of Hamiltonian systems
with homogeneous potentials. Ann. Inst. Fourier (Grenoble) 59, 2839–2890 (2009)
136. Duval, G., Maciejewski, A.J.: Integrability of Hamiltonian systems with homogeneous poten-
tials of degrees ±2. An application of higher order variational equations. Discret. Contin. Dyn.
Syst. 34, 4589–4615 (2014)
137. Duval, G., Maciejewski, A.J.: Integrability of potentials of degree k = ±2. Second order
variational equations between Kolchin solvability and Abelianity. Discret. Contin. Dyn. Syst.
35, 1969–2009 (2015)
138. Écalle, J.: Introduction aux fonctions analysables et preuve constructive de la conjecture du
Dulac. Actualities Math, Hermann, Paris (1992)
139. Enciso, A., Peralta-Salas, D.: Existence and vanishing set of inverse integrating factors for
analytic vector fields. Bull. Lond. Math. Soc. 41, 1112–1124 (2009)
140. Evdokimenco, R.M.: Construction of algebraic paths and the qualitative investigation in the
large of the properties of integral curves of a system of differential equations. Differ. Equ. 6,
1349–1358 (1970)
141. Evdokimenco, R.M.: Behavior of integral curves of a dynamic system. Differ. Equ. 9, 1095–
1103 (1974)
142. Evdokimenco, R.M.: Investigation in the large of a dynamic system with a given integral
curve. Differ. Equ. 15, 215–221 (1979)
143. Falla, Luza M.: On the density of second order differential equations without algebraic solu-
tions on P2 . Publ. Math. 55, 163–183 (2011)
144. Falkner, G., Skan, S.W.: Solutions of the boundary layer equations. Phil. Mag. 7, 865–896
(1931)
145. Ferčeca, B., Giné, J., Mencinger, M., Oliveira, R.: The center problem for a 1 : −4 resonant
quadratic system. J. Math. Anal. Appl. 420, 1568–1591 (2014)
146. Ferčeca, F., Giné, J., Romanovski, V.G., Edneral, V.F.: Integrability of complex planar systems
with homogeneous nonlinearities. J. Math. Anal. Appl. 434, 894–914 (2016)
147. Ferragut, A., Gasull, A.: Seeking Darboux Polynomials. Acta Appl. Math. 139, 167–186
(2015)
148. Ferragut, A., Giacomini, H.: A new algorithm for finding rational first integrals of polynomial
vector fields. Qual. Theory Dyn. Syst. 9, 89–99 (2010)
149. Ferragut, A., Llibre, J.: On the remarkable values of the rational first integrals of polynomial
vector fields. J. Differ. Equ. 241, 399–417 (2007)
150. Ferragut, A., Llibre, J., Mahdi, A.: Polynomial inverse integrating factors for polynomial
vector fields. Discret. Contin. Dyn. Syst. 17, 387–395 (2007)
151. Ferragut, A., Valls, C.: On the Darboux integrability of a cubic CRN model in R5 . Chaos
Solitons Fractals 82, 131–138 (2016)
References 359

152. Filipstov, V.F.: Algebraic limit cycles. Differ. Equ. 9, 983–988 (1973)
153. Fomenko, A.T.: Integrability and Nonintegrability in Geometry and Mechanics. Kluwer Aca-
demic, Dordrecht (1988)
154. Fomenko, A.T., Trofimov, V.V.: Integrable System on Lie Algebras and Symmetric Spaces.
Gordon and Breach, New York (1988)
155. Forsyth, A.R.: Theory of Differential Equations. Cambridge University Press, Cambridge
(1900). Chapter 17
156. Furta, S.D.: On non-integrability of general systems of differential equations. Z. Angew. Math.
Phys. 47, 112–131 (1996)
157. Fulton, W.: Algebraic Curves: An Introduction to Algebraic Geometry. The Ben-
jamin/Cummings Publisheing Company, INC., London (1969)
158. Galindo, C., Monserrat, F.: Algebraic integrability of foliations of the plane. J. Differ. Equ.
231, 611–632 (2006)
159. Planar quasi-homogeneous polynomial differential systems and their integrability: García,
B., Llibre, J., Pérez del Río, J.S. J. Differ. Equ. 255, 3185–3204 (2013)
160. García, I.A., Giacomini, H., Grau, M.: The inverse integrating factor and the Poincaré map.
Trans. Am. Math. Soc. 362, 3591–3612 (2010)
161. García, I.A., Giacomini, H., Grau, M.: Generalized Hopf bifurcation for planar vector fields
via the inverse integrating factor. J. Dynam. Diff. Equ. 23, 251–281 (2011)
162. García, I.A., Giné, J.: Analytic nilpotent centers with analytic first integral. Nonlinear Anal.
72, 3732–3738 (2010)
163. García, I.A., Giné, J., Sorolla, J.: On the existence of polynomial inverse integrating factors
in quadratic systems with limit cycles. Dyn. Contin. Discret. Impuls. Sys. Ser. A Math. Anal.
12, 151–173 (2005)
164. García, I.A., Grau, M.: A survey on the inverse integrating factor. Qual. Theory Dyn. Syst. 9,
115–166 (2010)
165. García, I.A., Grau, M., Llibre, J.: First integrals and Darboux polynomials of natural polyno-
mial Hamiltonian systems. Phys. Lett. A 374, 4746–4748 (2010)
166. García, I.A., Hernández-Bermejo, B.: Poisson systems as the natural framework for additional
first integrals via Darboux invariant hypersurfaces. Bull. Sci. Math. 137, 242–250 (2013)
167. García, I.A., Maza, S.: A new approach to center conditions for simple analytic monodromic
singularities. J. Differ. Equ. 248, 363–380 (2010)
168. García, I.A., Shafer, D.S.: Integral invariants and limit sets of planar vector fields. J. Differ.
Equ. 217, 363–376 (2005)
169. Gasull, A., Llibre, J., Zhang, X.: One-dimensional quaternion homogeneous polynomial dif-
ferential equations. J. Math. Phys. 50, 082705 (2009). 17 pp
170. Gasull, A.: Mañosa, V.: A Darboux-type theory of integrability for discrete dynamical systems.
J. Differ. Equ. Appl. 8, 1171–1191 (2002)
171. Giacomini, H., Giné, J., Grau, M.: Linearizable planar differential systems via the inverse
integrating factor. J. Phys. A 41, 135205 (2008). 19 pp
172. Giacomini, H., Llibre, J., Viano, M.: On the nonexistence, existence and uniqueness of limit
cycles. Nonlinearity 9, 501–516 (1996)
173. Giacomini, H., Llibre, J., Viano, M.: Semistable limit cycles that bifurcate from centers. Int.
J. Bifur. Chaos 13, 3489–3498 (2003)
174. Giacomini, H.J., Neukirch, S.: Integrals of motion and the shape of the attractor for the Lorenz
model. Phys. Lett. A 227, 309–318 (1997)
175. Giacomini, H.J., Repetto, C.E., Zandron, O.P.: Integrals of motion for three-dimensional
non-Hamiltonian dynamical systems. J. Phys. A 24, 4567–4574 (1991)
176. Giné, J.: Analytic integrability and characterization of centers for nilpotent singular points.
Z. Angew. Math. Phys. 55, 725–740 (2004)
177. Giné, J.: Reduction of integrable planar polynomial differential systems. Appl. Math. Lett.
25, 1862–1865 (2012)
178. Giné, J.: The nondegenerate center problem and the inverse integrating factor. Bull. Sci. Math.
130, 152–161 (2006)
360 References

179. Giné, J., Grau, M., Llibre, J.: Polynomial and rational first integrals for planar quasi-
homogeneous polynomial differential systems. Discret. Contin. Dyn. Syst. 33, 4531–4547
(2013)
180. Giné, J., Grau, M., Llibre, J.: On the extensions of the Darboux theory of integrability. Non-
linearity 26, 2221–2229 (2013)
181. Giné, J., Llibre, J.: Integrability and algebraic limit cycles for polynomial differential systems
with homogeneous nonlinearities. J. Differ. Equ. 197, 147–161 (2004)
182. Giné, J., Llibre, J.: A family of isochronous foci with Darboux first integral. Pac. J. Math.
218, 343–355 (2005)
183. Giné, J., Llibre, J.: On the planar integrable differential systems. Z. Angew. Math. Phys. 62,
567–574 (2011)
184. Giné, J., Llibre, J.: A note on Liouvillian integrability. J. Math. Anal. Appl. 387, 1044–1049
(2012)
185. Giné, J., Maza, S.: The reversibility and the center problem. Nonlinear Anal. 74, 695–704
(2011)
186. Giné, J., Peralta-Salas, D.: Existence of inverse integrating factors and Lie symmetries for
degenerate planar centers. J. Differ. Equ. 252, 344–357 (2012)
187. Giné, J., Santallusia, X.: Essential variables in the integrability problem of planar vector fields.
Phys. Lett. A 375, 291–297 (2011)
188. Giné, J., Valls, C.: Analytic integrability of cubic-linear planar polynomial differential sys-
tems. J. Differ. Equ. 260, 1690–1716 (2016)
189. Goriely, A.: Integrability, partial integrability, and nonintegrability for systems of ordinary
differential equations. J. Math. Phys. 37, 1871–1893 (1996)
190. Goriely, A.: Integrability and Nonintegrability of Dynamical Systems. World Scientific, Sin-
gapore (2001)
191. Gradshteyn, I.S., Ryzhik, I.M.: Table of Integrals, Series, and Products. Academic Press, New
York (1980)
192. Grobman, D.M.: Topological classification of neighborhoods of a singularity in n-space. Mat.
Sb. (N.S.) 56, 77–94 (1962). (in Russian)
193. Gupta, N.: Integrals of motion for the Lorenz system. J. Math. Phys. 34, 801–804 (1993)
194. Gutierrez, J., Rubio, R., Sevilla, D.: On multivariate rational function decomposition. J. Symb.
Comput. 33, 545–562 (2002)
195. Han, M.A.: Periodic Solutions and Bifurcation Theory of Dynamical Systems (in Chinese).
Science Press, Beijing (2002)
196. Han, M.A., Romanovski, V.G., Zhang, X.: Integrability of a family of 2-Dim cubic systems
with degenerate infinity. Romania J. Phys. 61, 157–166 (2016)
197. Hartman, P.: On local homeomorphisms of Euclidean spaces. Bol. Soc. Mat. Mexicana 5,
220–241 (1960)
198. Hartman, P.: On the local linearization of differential equations. Proc. Am. Math. Soc. 14,
568–573 (1963)
199. Hartshorne, R.: Algebraic Geometry. Springer, New York (1997)
200. Hatcher, A.: Algebraic Topology. Cambridge University Press, Cambridge (2002)
201. Hayashi, M.: On polynomial Liénard systems which have invariant algebraic curves. Funkcial.
Ekvac. 39, 403–408 (1996)
202. Hide, R., Skeldon, A.C., Acheson, D.J.: A study of two novel self-exciting single-disc homopo-
lar dynamos: theory. Proc. R. Soc. Lond. Ser. A 452, 1369–1395 (1996)
203. Hietarinta, J.: Direct methods for the search of the second invariant. Phys. Rep. 147, 87–154
(1987)
204. Hilbert D.: Mathematische Probleme, Lecture, Second Int. Congr. Math. Paris (1900), Nachr.
Ges. Wiss. G"ttingen Math. Phys. KL. pp. 253–297 (1900); English Translation Bull. Am.
Math. Soc. 8, 437–479 (1902)
205. Hille, E.: Ordinary Differential Equations in the Complex Domain. Wiley, New York (1976)
206. Hirsch, M.: Differential Topology. Springer, New York (1976)
References 361

207. Hirsch, M., Smale, S.: Differential Equations, Dynamical Systems, and Linear Algebra. Aca-
demic Press, San Diego (1974)
208. Holmes, P., Jenkins, J., Leonard, N.E.: Dynamics of the Kirchoff equations I: cocident centers
of gravity and buoyancy. Phys. D 118, 311–342 (1998)
209. Hu, Y.X., Xue, C.Z.: One-parameter Lie groups and inverse integrating factors of n-th order
autonomous systems. J. Math. Anal. Appl. 388, 617–626 (2012)
210. Hu, Z., Han, M.A., Romanovski, V.G.: Local integrability of a family of three-dimensional
quadratic systems. Phys. D 265, 78–86 (2013)
211. Huebsch, W., Morse, M.: Diffeomorphisms of manifolds. Rend. Circ. Mat. Palermo 11(2),
291–318 (1962)
212. Ilyashenko Yu, S.: Finiteness Theorems for Limit Cycles. Translation of Mathematical Mono-
graph, vol. 94. American Mathematical Society, Providence (1991)
213. Ilyashenko, YuS, Yakovenko, S.: Lectures on Analytic Differential Equations. American
Mathematical Society, Providence (2008)
214. Ito, H.: Convergence of Birkhoff normal forms for integrable systems. Comment. Math.
Helvetici 64, 412–465 (1989)
215. Jiao, J., Shi, S., Zhou, Q.: Rational first integrals for periodic systems. Z. Angew. Math. Phys.
62, 233–243 (2011)
216. Jost, J.: Dynamical Systems. Springer, Berlin (2005)
217. Jouanolou, J.P.: Equations de Pfaff algébriques. Lectures Notes in Math, vol. 708. Springer,
New York (1979)
218. Kapteyn W.: On the midpoints of integral curves of differential equations of the first degree.
Nederl. Akad. Wetensch. Verslag. Afd. Natuurk. Konikl. Nederland, pp. 1446–1457 (1911)
219. Kapteyn, W.: New investigations on the midpoints of integrals of differential equations of
the first degree. Nederl. Akad. Wetensch. Verslag Afd. Natuurk. 20, 1354–1365; 21, 27–33
(1912)
220. Klimek, M.: Pluripotential Theory. London Mathematical Society Monographs, vol. 6. Oxford
University Press, Oxford (1991)
221. Kuksin, S., Pöschel, J.: On the inclusion of analytic symplectic maps in analytic Hamiltonian
flows and its applications. In: Kuksin, S., Lazutkin, V., Pöschel, J. (eds.) Seminar on Dynamical
Systems, pp. 96–116. Birkhäuser, Basel (1978)
222. Kús, M.: Integrals of motion for the Lorenz system. J. Phys. A 16, L689–L691 (1983)
223. Kwek, K.H., Li, Y., Shi, S.: Partial integrability for general nonlinear systems. Z. Angew.
Math. Phys. 54, 26–47 (2003)
224. Lang, S.: Algebra, Graduate Texts in Mathematics 211, vol. 211, 3rd edn. Springer, New York
(2002)
225. Laufer, H.B.: Normal Two-Dimensional Singularities. Princeton University Press, Princeton
(1971)
226. Laura-Guarachi, L., Osuna, O., Villaseor-Aguilar, G.: Non-existence of limit cycles via inverse
integrating factors. Electron. J. Differ. Equ. 2011(124), 1–6 (2011). 6 pp
227. Lefschetz, S.: Differential Equations: Geometric Theory. Dover Publications Inc, New York
(1977)
228. Li, C.: Two problems of planar quadratic systems (in Chinese). Sci. China Math. 12, 1087–
1096 (1982)
229. Li, J.B.: Hilbert’s 16th problem and bifurcations of planar polynomial vector fields. Int. J.
Bifur. Chaos 13, 47–106 (2003)
230. Li, W.G.: Normal Form Theory and its Application (in Chinese). Science Press, Beijing (2000)
231. Li, W.G., Llibre, J., Nicolau, M., Zhang, X.: On the differentiability of first integrals of two-
dimensional flows. Proc. Am. Math. Soc. 130, 2079–2088 (2002)
232. Li, W.G., Llibre, J., Zhang, X.: Extension of floquet’s theory to nonlinear periodic differential
systems and embedding diffeomorphisms in differential flows. Am. J. Math. 124, 107–127
(2002)
233. Li, W.G., Llibre, J., Zhang, X.: Local first integrals of differential systems and diffeomor-
phisms. Z. Angew. Math. Phys. 54, 235–255 (2003)
362 References

234. Li, W.G., Llibre, J., Zhang, X.: Planar analytic vector fields with generalized rational first
integrals. Bull. Sci. Math. 125, 341–361 (2001)
235. Li, W.G., Lu, K.: Sternberg theorems for random dynamical systems. Commun. Pure Appl.
Math. 58, 941–988 (2005)
236. Li, W.G., Lu, K.: Poincaré theorems for random dynamical systems. Ergod. Theory Dyn.
Syst. 25, 1221–1236 (2005)
237. Li, W.G., Lu, K.: A Siegel theorem for dynamical systems under random perturbations.
Discret. Contin. Dyn. Syst. Ser. B 9, 635–642 (2008)
238. Li, W., Shi, S.: Non-integrability of Hénon-Heiles system. Celest. Mech. Dyn. Astron. 109,
1–12 (2011)
239. Li, W., Shi, S., Liu, B.: Non-integrability of a class of Hamiltonian systems. J. Math. Phys.
52, 112702 (2011). 12 pp
240. Li, W., Shi, S.: Galoisian obstruction to the integrability of general dynamical systems. J.
Differ. Equ. 252, 5518–5534 (2012)
241. Lima, M.F.S., Llibre, J., Valls, C.: Integrability of the Rucklidge system. Nonlinear Dyn. 77,
1441–1453 (2014)
242. Lins, Neto A.: Some examples for the Poincaré and Painlevé problems. Ann. Sci. École Norm.
Sup. 35, 231–266 (2002)
243. Lins, Neto A.: Algebraic solutions of polynomial differential equations and foliations in
dimension two, in holomorphic dynamics. Lect. Notes Math. 1345, 192–232 (1988)
244. Algebraic solutions of one-dimensional foliations: Lins, Neto A., Soares, M.G. J. Diff. Geom.
43, 652–673 (1996)
245. Liu, C., Chen, G., Yang, J.: On the hyperelliptic limit cycles of Liénard systems. Nonlinearity
25, 1601–1611 (2012)
246. Liu, Y., Li, J.: Some Classical Problems of Planar Vector Fields (in Chinese). Science Press,
Beijing (2010)
247. Liu, Y., Yang, Q.: Dynamics of the L system on the invariant algebraic surface and at infinity.
Int. J. Bifur. Chaos 21, 2559–2582 (2011)
248. Llibre, J.: Integrability of Polynomial Differential Systems. Handbook of Differential Equa-
tions. Elsevier, Amsterdam (2004)
249. Llibre, J.: Open problems on the algebraic limit cycles of planar polynomial vector fields.
Bul. Acad. Stiinte Repub. Mold. Mat. 1, 19–26 (2008)
250. Llibre, J., Mahdi, A., Valls, C.: Darboux integrability of the Lü system. J. Geom. Phys. 63,
118–128 (2013)
251. Llibre, J., Mahdi, A., Valls, C.: Polynomial integrability of the Hamiltonian systems with
homogeneous potential of degree 3. Phys. D 240, 1928–1935 (2011)
252. Llibre, J., Mahdi, A., Valls, C.: Analytic integrability of Hamiltonian systems with homoge-
neous polynomial potential of degree 4. J. Math. Phys. 52, 012702 (2011). 9 pp
253. Llibre, J., Mahdi, A., Valls, C.: Polynomial integrability of the Hamiltonian systems with
homogeneous potential of degree −2. Phys. Lett. A 375, 1845–1849 (2011)
254. Llibre, J., Mahdi, A., Valls, C.: Polynomial integrability of the Hamiltonian systems with
homogeneous potential of degree −3. Phys. D 240, 1928–1935 (2011)
255. Llibre, J., Medrado, J.C.: On the invariant hyperplanes for d-dimensional polynomial vector
fields. J. Phys. A 40, 8385–8391 (2007)
256. Llibre, J., Medrado, J.C.: On the invariant hyperplanes for d-dimensional polynomial vector
fields. J. Phys. A 40, 8385–8391 (2007)
257. Llibre, J., Medrado, J.C., da Silva, P.R.: Limit cycles for singular perturbation problems via
inverse integrating factor. Bol. Soc. Parana. Mat. 26(3), 41C52 (2008)
258. Llibre, J., Messias, M.: Global dynamics of the Rikitake system. Physica D 238, 241–252
(2009)
259. Llibre, J., Messias, M., da Silva, P.R.: On the global dynamics of the Rabinovich system. J.
Phys. A 41(27), 275210 (2008). 21 pp
260. Llibre, J., Messias, M., da Silva, P.R.: Global dynamics of the Lorenz system with invariant
algebraic surfaces. Int. J. Bifur. Chaos 20(10), 3137–3155 (2010)
References 363

261. Llibre, J., Messias, M., da Silva, P.R.: Global dynamics in the Poincaré ball of the Chen system
having invariant algebraic surfaces. Int. J. Bifur. Chaos 22(6), 1250154 (2012). 17 pp
262. Llibre, J., Pantazi, Ch.: Darboux theory of integrability for a class of nonautonomous vector
fields. J. Math. Phys. 50, 102705 (2009). 19 pp
263. Llibre, J., Pantazi, C., Walcher, S.: First integrals of local analytic differential systems. Bull.
Sci. Math. 136, 342–359 (2012)
264. Llibre, J., Ramírez, R., Sadovskaia, N.: On the 16th Hilbert problem for algebraic limit cycles.
J. Differ. Equ. 248, 1401–1409 (2010)
265. Llibre, J., Ramírez, R., Sadovskaia, N.: On the 16th Hilbert problem for limit cycles on
nonsingular algebraic curves. J. Differ. Equ. 250, 983–999 (2011)
266. Llibre, J., Rodríguez, G.: Invariant hyperplanes and Darboux integrability for d-dimensional
polynomial differential systems. Bull. Sci. Math. 124, 599–619 (2000)
267. Llibre, J., Rodríguez, G.: Darboux integrability of polynomial vector fields on 2-dimensional
surfaces. Int. J. Bifur. Chaos 12, 2821–2833 (2002)
268. Llibre, J., Rodríguez, G.: Configurations of limit cycles and planar polynomial vector fields.
J. Differ. Equ. 198, 374–380 (2004)
269. Llibre, J., Saghin, R., Zhang, X.: On the analytic integrability of the 5-dimensional Lorenz
system for the gravity-wave activity. Proc. Am. Math. Soc. 142, 531–537 (2014)
270. Llibre, J., Stoica, C., Valls, C.: Polynomial and rational integrability of polynomial Hamil-
tonian systems. Electron. J. Differ. Equ. 2012(108), 1–6 (2012). 6 pp
271. Llibre, J., Świrszcz, G.: Relationships between limit cycles and algebraic invariant curves for
quadratic systems. J. Differ. Equ. 229, 529–537 (2006)
272. Llibre, J., Świrszsz G.: An example of a cubic Liénard system with linear damping having
invariant algebraic curves of arbitrary degree, preprint
273. Llibre, J., Valls, C.: Integrability of the Bianchi IX system. J. Math. Phys. 46(072901), 1–13
(2005)
274. Llibre, J., Valls, C.: Formal and analytic first integrals of the Einstein-Yang-Mills equations.
J. Phys. A 38, 8155–8168 (2005)
275. Llibre, J., Valls, C.: On the integrability of the Einstein-Yang-Mills equations. J. Math. Anal.
Appl. 336, 1203–1230 (2007)
276. Llibre, J., Valls, C.: Darboux integrability and algebraic invariant surfaces for the Rikitake
system. J. Math. Phys. 49(3), 032702 (2008). 17 pp
277. Llibre, J., Valls, C.: Analytic integrability of a Chua system. J. Math. Phys. 49, 102701 (2008)
278. Llibre, J., Valls, C.: On the local analytic integrability at the singular point of a class of Liénard
analytic differential systems. Proc. Am. Math. Soc. 138, 253–261 (2010)
279. Llibre, J., Valls, C.: Nonintegrability of a class of the Bianchi VI0 and VII0 models. J. Geom.
Phys. 60, 815–822 (2010)
280. Llibre, J., Valls, C.: Liouvillian first integrals for Liénard polynomial differential systems.
Proc. Am. Math. Soc. 138, 3229–3239 (2010)
281. Llibre, J., Valls, C.: The Michelson system is neither global analytic, nor Darboux integrable.
Phys. D 239, 414–419 (2010)
282. Llibre, J., Valls, C.: Analytic integrability of quadratic-linear polynomial differential systems.
Ergod. Theory Dyn. Syst. 31, 245–258 (2011)
283. Llibre, J., Valls, C.: On the polynomial integrability of the Kirchoff equations. Phys. D 241,
1417–1420 (2012)
284. Llibre, J., Valls, C.: On the integrability of a Muthuswamy–Chua system. J. Nonlinear Math.
Phys. 19, 1250029 (2012). 12 pp
285. Llibre, J., Valls, C.: Liouvillian first integrals for generalized Liénard polynomial differential
systems. Adv. Nonlinear Stud. 13, 825–835 (2013)
286. Llibre, J., Valls, C.: On the Darboux integrability of Blasius and Falkner-Skan equation.
Comput. Fluids 86, 71–76 (2013)
287. Llibre, J., Valls C.: The generalized Liénard polynomial differential systems x  = y, y  =
−g(x) − f (x)y with deg g = deg f + 1 are not Liouvillian integrable, preprint (2014)
364 References

288. Llibre, J., Valls, C.: Algebraic invariant curves and first integrals for Riccati polynomial
differential systems. Proc. Am. Math. Soc. 142, 3533–3543 (2014)
289. Llibre, J., Valls, C.: Darboux integrability of 2-dimensional Hamiltonian systems with
homogenous potentials of degree 3. J. Math. Phys. 55, 033507 (2014). 12 pp
290. Llibre, J., Valls, C.: On the analytic integrability of the Liénard analytic differential systems.
Discret. Contin. Dyn. Syst. B 21, 557–573 (2016)
291. Llibre, J., Valls, C.: On the integrabilitly of the 5-dimensional Lorenz system for the gravity-
wave activity. Proc. Am. Math. Soc. 145(2), 665–679 (2017)
292. Llibre, J., Valls, C., Zhang, X.: Completely integrable differential systems are essentially
linear. J. Nonlinear Sci. 25, 815–826 (2015)
293. Llibre, J., Valls, C., Zhang, X.: Liouvillian integrability versus Darboux polynomials. Qual.
Theory Dyn. Syst. 15(2), 503–515 (2016)
294. Llibre, J., Walcher, S., Zhang, X.: Local Darboux first integrals of analytic differential systems.
Bull. Sci. Math. 138, 71–88 (2014)
295. Llibre, J., Yu, J., Zhang, X.: On polynomial integrability of the Euler equations. J. Geom.
Phys. 96, 36–41 (2015)
296. Llibre, J., Zhang, X.: Invariant algebraic surfaces of the Rikitake system. J. Phys. A 33,
7613–7635 (2000)
297. Llibre, J., Zhang, X.: Invariant algebraic surfaces of the Belousov-Zhabotinskii systems. The
geometry of differential equations and dynamical systems. Comput. Appl. Math. 20, 195–219
(2001)
298. Llibre, J., Zhang, X.: Invariant algebraic surfaces of the Lorenz systems. J. Math. Phys. 43,
1622–1645 (2002)
299. Llibre, J., Zhang, X.: Darboux integrability for the Rössler system. Int. J. Bifur. Chaos 12,
421–428 (2002)
300. Llibre, J., Zhang, X.: Darboux integrability of real polynomial vector fields on regular alge-
braic hypersurfaces. Rend. Circ. Mat. Palermo (Serie II) LI, 109–126 (2002)
301. Llibre, J., Zhang, X.: On the algebraic limit cycles of Liénard systems. Nonlinearity 21,
2011–2022 (2008)
302. Llibre, J., Zhang, X.: Darboux theory of integrability in Cn taking into account the multiplicity.
J. Differ. Equ. 246, 541–551 (2009)
303. Llibre, J., Zhang, X.: Darboux theory of integrability for polynomial vector fields in Cn taking
into account the multiplicity at infinity. Bull. Sci. Math. 133, 765–778 (2009)
304. Llibre, J., Zhang, X.: Rational first integrals in the Darboux theory of integrability in Cn . Bull.
Sci. Math. 134, 189–195 (2010)
305. Llibre, J., Zhang, X.: On the Darboux integrability of polynomial differential systems. Qual.
Theory Dyn. Syst. 11, 129–144 (2012)
306. Lorenz, E.N.: Deterministic nonperiodic flow. J. Atmos. Sci. 20, 130–141 (1963)
307. Lorenz, E.N.: On the existence of a slow manifold. J. Atmos. Sci. 43, 1547–1557 (1986)
308. Luo, D.J., Zhang, X., Dong, M.F.: Qualitative Theory and Bifurcation of Dynamical Systems
(in Chinese). Science Press, Beijing (2001)
309. Lunkevich, V.A., Sibirsky, K.S.: Conditions of a centre in homogeneous nonlinearities of third
degree. Differ. Equ. 1, 1164–1168 (1965)
310. Lunkevich, V.A., Sibirskii, K.S.: Integrals of a general quadratic differential system in cases
of a center. Differ. Equ. 18, 786–792 (1982)
311. Lü, J., Chen, G.: A new chaotic attractor coined. Int. J. Bifur. Chaos 12, 659–661 (2002)
312. Lü, T.: Darboux polynomials and non-algebraic integrability of the Lü system. Ann. Differ.
Equ. 25, 420–431 (2009)
313. Lü, T., Zhang, X.: Darboux polynomials and algebraic integrabililty of the Chen systems. Int.
J. Bifur. Chaos 17, 2739–2748 (2007)
314. Lützen, J.: Joseph Liouville 1809–1882: Master of Pure and Applied Mathematics. Springer,
New York (1990)
315. Lyapunov, M.: Problème Général de la Stabilté du Movement. Princeton University Press,
Princeton (1947)
References 365

316. Maciejewski, A.J., Popov, S.I., Strelcyn, J.-M.: The Euler equations on Lie algebra so(4): an
elementary approach to integrability condition. J. Math. Phys. 42, 2701–2717 (2001)
317. Maciejewski, A.J., Przybylska, M.: Darboux polynomials and first integrals of natural poly-
nomial Hamiltonian systems. Phys. Lett. A 326, 219–226 (2004)
318. Maciejewski, A.J., Przybylska, M.: All meromorphically integrable 2D Hamiltonian systems
with homogeneous potential of degree 3. Phys. Lett. A 327, 461–473 (2004)
319. Maciejewski, A.J., Przybylska, M.: Darboux points and integrability of Hamiltonian systems
with homogeneous polynomial potential. J. Math. Phys. 46, 062901 (2005). 33 pp
320. Maciejewski, A.J., Przybylska, M.: Partial integrability of Hamiltonian systems with homo-
geneous potential. Regul. Chaotic Dyn. 15, 551–563 (2010)
321. Maciejewski, A.J., Przybylska, M.: Integrability of Hamiltonian systems with algebraic poten-
tials. Phys. Lett. A 380, 76–82 (2016)
322. Maciejewski, A.J., Przybylska, M., Yoshida, H.: Necessary conditions for the existence of
additional first integrals for Hamiltonian systems with homogeneous potential. Nonlinearity
25, 255–277 (2012)
323. Mahdi, A., Valls, C.: Integrability of the Nosé-Hoover equation. J. Geom. Phys. 61, 1348–1352
(2011)
324. Mahdi, A., Valls, C.: Invariant algebraic surfaces for the reduced three-wave interaction sys-
tem. J. Math. Phys. 52, 122702 (2011). 10 pp
325. Mahdi, A., Valls, C.: On the nonintegrability of magnetic field lines. Phys. D 251, 60–62
(2013)
326. Mahdi, A., Valls, C.: Integrability of the Hide-Skeldon-Acheson dynamo. Bull. Sci. Math.
138, 470–482 (2014)
327. Malkin, K.E.: Criteria for center of a differential equation. Volg. Matem. Sbornik 2, 87–91
(1964)
328. Man, Y.-K., Maccallum, M.A.H.: A rational approach to the Prelle-Singer algorithm. J. Sym-
bolic Comput. 24, 31–43 (1997)
329. Linearization of isochronous centers: Mardes̆ić, P., Rousseau, C., Toni, B. J. Differ. Equ. 121,
67–108 (1995)
330. Markus, L.: Global structure of ordinary differential equations in the plane. Trans. Am. Math.
Soc. 76, 127–148 (1954)
331. Mazzi, L., Sabatini, M.: A characterization of centers via first integrals. J. Differ. Equ. 76,
222–237 (1988)
332. Michelson, D.: Steady solutions for the Kuramoto-Sivashinsky equation. Phys. D 19, 89–111
(1986)
333. Milnor, J.W.: Topology from the Differentiable Viewpoint. Princeton University Press, Prince-
ton (1997)
334. Morales-Ruiz, J.J.: Differential Galois Theory and Non-integrablity of Hamiltonian Systems.
Progress in Mathematics. Birkhäuser, Basel (1999)
335. Morales-Ruiz, J.J., Ramis, J.P.: Galoisian obstructions to integrability of Hamiltonian systems.
I, II. Methods Appl. Anal. 8(33–95), 97–111 (2001)
336. Morales-Ruiz, J.J., Ramis, J.P.: A note on the non-integrability of some Hamiltonian systems
with a homogeneous potential. Methods Appl. Anal. 8, 113–120 (2001)
337. Morales-Ruiz, J.J., Ramis, J.P., Simó, C.: Integrability of Hamiltonian systems and differential
Galois groups of higher variational equations. Ann. Sci. Éc. Norm. Supér 40, 845–884 (2007)
338. Moser, J.: The analytic invariants of an area-preserving mapping near a hyperbolic fixed point.
Commun. Pure Appl. Math. 9, 673–692 (1956)
339. Moser, J.: A rapidly covergent iteration method and nonlinear differential equations II. Ann.
Scuola Norm. Sup. Pisa. 20(3), 499–535 (1966)
340. Moulin-Ollagnier, J.: About a conjecture on quadratic vector fields. J. Pure Appl. Algebra
165, 227–234 (2001)
341. Moulin-Ollagnier, J., Nowicki, A.: Derivations of polynomial algebras without Darboux poly-
nomials. J. Pure Appl. Algebra 212, 1626–1631 (2008)
366 References

342. Moussu, R.: Une demonstration d’un théoréme de Lyapunov-Poincaréé. Asterisque 98–99,
216–223 (1982)
343. Moussu, R.: Symétrie et forme normale des centres et foyers dégénéré. Ergod. Theory Dyn.
Syst. 2, 241–251 (1982)
344. Nagumo, M., Isé, K.: On the normal forms of differential equations in the neighborhood of
an equilibrium point. Osaka J. Math. 9, 221–234 (1957)
345. Nakagawa, K., Maciejewski, A.J., Przybylska, M.: New integrable Hamiltonian systems with
first integrals quartic in momenta. Phys. Lett. A 343, 171–173 (2005)
346. Nakagawa, K., Yoshida, H.: A necessary condition for the integrability of homogeneous
Hamiltonian systems with two degrees of freedom. J. Phys. A 34, 2137–2148 (2001)
347. Narasimhan, R.: Analysis on Real and Complex Manifolds. North-Holland Mathematical
Library, vol. 35. North-Holland Publishing Co., Amsterdam (1985)
348. Neumann, D.A.: Classification of continuous flows on 2-manifolds. Proc. Am. Math. Soc. 48,
73–81 (1975)
349. Nowak, M.A., May, R.M.: Virus Dynamics. Mathematical Principles of Immunology and
Virology. Oxford University Press, New York (2000)
350. Odani, K.: The limit cycle of the van der Pol equation is not algebraic. J. Differ. Equ. 115,
146–152 (1995)
351. Olver, P.J.: Applications of Lie Groups to Differential Equations. Graduate Texts in Mathe-
matics, vol. 107. Springer, New York (1993)
352. Ostrowski, A.M.: On multiplication and factorization of polynomials, I. Lexicographic order-
ings and extreme aggregates of terms. Aequ. Math. 13, 201–228 (1975)
353. Palis Jr., J.: Vector fields generate few diffeomorphisms. Bull. Am. Math. Soc. 80, 503–505
(1974)
354. Palis Jr., J., de Melo, W.: Geometric Theory of Dynamical Systems An Introduction. Springer,
New York (1982)
355. Pan, Y., Wang, C., Zhang, X.: Global C ∞ integrability of cubic-linear polynomial differential
systems. Qual. Theory Dyn. Syst. 13, 73–87 (2014)
356. Pearson, J.M., Lloyd, N.G., Christopher, C.J.: Algorithmic derivation of centre conditions.
SIAM Rev. 38, 619–636 (1996)
357. Pereira, J.V.: Vector fields, invariant varieties and linear systems. Ann. Inst. Fourier 51, 1385–
1405 (2001)
358. Perelomov, A.M.: Integrable Systems of Classical Mechanics and Lie Algebras, vol. I.
Birkhäuser, Basel (1990)
359. Pérez-Marco, R.: Total convergence or general divergence in small divisions. Commun. Math.
Phys. 223, 451–464 (2001)
360. Pérez-Marco, R.: Convergence or generic divergence of the Birkhoff normal form. Ann. Math.
157, 557–574 (2003)
361. Pinheiro, S., Reis, H.: Topological aspects of completely integrable foliations. J. Lond. Math.
Soc. 89(2), 415–433 (2014)
362. Poincaré, H.: Sur l’intégration des équations différentielles du premier ordre et du premier
degré I and II. Rend. Circ. Mat. Palermo 5, 161–191 (1891); 11 193–239 (1897)
363. Popov, S.I., Strelcyn, J.-M.: On rational integrability of Euler equations on Lie algebra so(4,C).
Israel J. Math. 163, 263–283 (2008)
364. Popov, S.I., Respondek, W., Strelcyn, J.-M.: On rational integrability of Euler equations on
Lie algebra so(4,C), revisited. Phys. Lett. A 373, 2445–2453 (2009)
365. Prelle, M.J., Singer, M.F.: Elementary first integrals of differential equations. Trans. Am.
Math. Soc. 279, 215–229 (1983)
366. Przybylska, M.: Darboux points and integrability of homogeneous Hamiltonian systems with
three and more degrees of freedom. Nongeneric cases. Regul. Chaotic Dyn. 14, 349–388
(2009)
367. Pugh, C.: On a theorem of P. Hartman. Am. J. Math. 91, 363–367 (1969)
368. van der Put, M., Singer, M.F.: Galois Theory of Linear Differential Equations. Grundlehren
der Mathematischen Wissenschaften, vol. 328. Springer, Berlin (2003)
References 367

369. Qin,
 Y.X.: iOnj quadratic
 algebraic limit cycles of differential system dy/d x =
ai j x y / bi j x i y j (in Chinese). Sci. China 7, 934–945 (1958)
0≤i+ j≤2 0≤i+ j≤2
370. Reeb G.: Sur certaines propiétés topologiques des variétés feuilletées, In: Wu, W.T., Reeb, G.
(Eds.), Sur les espaces fibrés et les variétés feuilletées, Tome Xi, in: Actualités Sci. Indust.
1183, Hermann et Cie, Paris (1952)
371. Ribón, J.: Embedding smooth and formal diffeomorphisms through the Jordan-Chevalley
decomposition. J. Differ. Equ. 253, 3211–3231 (2012)
372. Ripoll, O., Sebag, J.: Tissus du plan et polynômes de Darboux [Planar webs and Darboux
polynomials]. Ann. Fac. Sci. Toulouse Math. 19(6), 1–11 (2010)
373. Romanovski, V.G., Shafer, D.S.: The Center and Cyclicity Problems: A Computational Alge-
braic Approach. Birkhäuser, Boston (2009)
374. Romanovski, V.G., Shafer, D.S.: Complete integrability and time-reversibility of some 3-dim
systems. Appl. Math. Lett. 51, 27–33 (2016)
375. Romanovski, V.G., Wu, Y.H., Christopher, C.: Linearizability of three families of cubic sys-
tems. J. Phys. A 36(9), L145–L152 (2003)
376. Romanovski, V.G., Xia, Y., Zhang, X.: Varieties of local integrability of analytic differential
systems and their applications. J. Differ. Equ. 257, 3079–3101 (2014)
377. Rosenlicht, M.: Integration in finite terms. Am. Math. Mon. 79, 963–972 (1972)
378. Roussarie, R.: Bifurcation of Planar Vector Fields and Hilbert’s Sixteenth Problem.
Birkhäuser, Basel (1998)
379. Rousseau, C., Schlomiuk, D.: Cubic vector fields symmetric with respect to a center. J. Differ.
Equ. 123, 388–436 (1995)
380. Royden, H.L.: The analytic approximation of differentiable mappings. Math. Ann. 139, 171–
179 (1960)
381. Rychkov, G.S.: The limit cycles of the equation u(x +1)du = (−x +ax 2 +bxu+cu+du 2 )d x.
Differ. Equ. 8, 1748–1750 (1972)
382. Sard, A.: The measure of the critical values of differentiable maps. Bull. Am. Math. Soc. 48,
883–890 (1942)
383. Sathaye, A.: Generalized Newton-Puiseux expansion and Abhyankar-Moh semigroup theo-
rem. Invent. Math. 74, 149–157 (1983)
384. Schinzel, A.: Polynomials with Special Regard to Reducibility. Encyclopedia of Mathematics
and its Applications, vol. 77. Cambridge University Press, Cambridge (2000)
385. Schlomiuk, D.: Algebraic particular integrals, integrability and the problem of the center.
Trans. Am. Math. Soc. 338, 799–841 (1993)
386. Schlomiuk, D., Vulpe, N.: Planar quadratic vector fields with invariant lines of total multi-
plicity at least five. Qual. Theory Dyn. Syst. 5, 135–194 (2004)
387. Schwarz, F.: An algorithm for determining polynomial first integrals of autonomous systems
of ordinary differential equations. J. Symb. Comput. 1, 229–233 (1985)
388. Segur, H.: Soliton and the inverse scattering transform. In: Osborne, A.R., Malanotte Rizzoli,
P. (eds.) Topics in Ocean-Physics, pp. 235–277. North-Holland, Amsterdam (1982)
389. Seidenberg, A.: Reduction of singularities of the differential equation Ady = Bd x. Am. J.
Math. 90, 248–269 (1968)
390. Sell, G.R.: Smooth linearization near a fixed point. Am. J. Math. 107, 1035–1091 (1985)
391. Shafarevich, I.R.: Basic Algebraic Geometry. Springer, New York (1977)
392. Sharpe, R.W.: Differential Geometry. Springer, New York (1997)
393. Shen, B.Q.: The existence of cubic limit cycles and separatrix loops of quadratic systems (in
Chinese). Chin. Ann. Math. 12A, 382–389 (1991)
394. Shi, S., Li, Y.: Non-integrability for general nonlinear systems. Z. Angew. Math. Phys. 52,
191–200 (2001)
395. Shi, S.: On the nonexistence of rational first integrals for nonlinear systems and semiquasi-
homogeneous systems. J. Math. Anal. Appl. 335, 125–134 (2007)
396. Shi, S., Li, W.: Non-integrability of a class of Painlevé IV equations as Hamiltonian systems.
J. Math. Phys. 54, 102703 (2013). 9 pp
368 References

397. Shi, S., Li, W.: Non-integrability of generalized Yang-Mills Hamiltonian system. Discret.
Contin. Dyn. Syst. 33, 1645–1655 (2013)
398. Sibirsky, K.S.: On the number of limit cycles in the neighborhood of a singular point. Differ.
Equ. 1, 36–47 (1965)
399. Sibirsky, K.S.: Introduction to the Theory of Invariants of Differential Equations. Manchester
University Press, Manchester (1988)
400. Siegmund, S.: Reducibility of nonautonomous linear differential equations. J. Lond. Math.
Soc. 65, 397–410 (2002)
401. Siegmund, S.: Normal forms for nonautonomous differential equations. J. Differ. Equ. 178,
541–573 (2002)
402. Sijbrand, J.: Properties of center manifolds. Trans. Am. Math. Soc. 289, 431–C469 (1985)
403. Singer, M.F.: Liouvillian first integrals of differential equations. Trans. Am. Math. Soc. 333,
673–688 (1992)
404. Smale, S.: Mathematical problems for the next century. Math. Intell. 20(2), 7–15 (1998)
405. Sokulski, J.: On the number of invariant lines of polynomial vector fields. Nonlinearity 9,
479–485 (1996)
406. Sparrow, C.: The Lorenz Equations. Springer, New York (1982)
407. Steeb, W.H.: Continuous symmetries of the Lorenz model and the Rikitake two-disc dynamo
system. J. Phys. A 15, L389–L390 (1982)
408. Sternberg, S.: On the behavior of invariant curves near a hyberbolic point of a surface trans-
formation. Am. J. Math. 77, 526–534 (1955)
409. Sternberg, S.: Local contractions and a theorem of Poincaré. Am. J. Math. 79, 809–824 (1957)
410. Sternberg, S.: Lectures on Differential Geometry. Prentice-Hall, Englewood Cliffs (1964)
411. Stephani, H.: Differential Equations. Cambridge University Press, Cambridge (1989)
412. Stolovitch, L.: Progress in normal form theory. Nonlinearity 22, R77–R99 (2009)
413. Stowe, D.: Linearization in two dimensions. J. Differ. Equ. 63, 183–226 (1986)
414. Strelcyn, J.M., Wojciechowski, S.: A method of finding integrals for three-dimensional sys-
tems. Phys. Lett. A 133, 207–212 (1988)
415. Strózyna, E.: Zoła̧dek, H.: The analytic and formal normal form for the nilpotent singularity.
J. Differ. Equ. 179, 479–537 (2002)
416. Studziński, M., Przybylska, M.: Darboux points and integrability analysis of Hamiltonian
systems with homogeneous rational potentials. Phys. D 249, 1–15 (2013)
417. Swinnerton-Dyer, S.P.: The invariant algebraic surfaces of the Lorenz system. Math. Proc.
Camb. Phil. Soc. 132, 385–393 (2002)
418. Szumiński, W., Maciejewski, A.J., Przybylska, M.: Note on integrability of certain homoge-
neous Hamiltonian systems. Phys. Lett. A 379, 2970–2976 (2015)
419. Tabor, M., Weiss, J.: Analytic structure of the Lorenz system. Phys. Rev. A 24, 2157–2167
(1981)
420. Takens, F.: Normal forms for certain singularities of vector fields. Ann. Inst. Fourier 23,
163–195 (1973)
421. Takens, F.: Singularities of vector fields. Publ. Math. Inst. Hautes Ètudes Sci. 43, 47–100
(1974)
422. Trofimov, V.V.: Introduction to the Geometry of Manifolds with Symmetry. Moscow Uni-
versity Press, Moscow (1989). in Russian, p. 1994. English translation Kluwer Academic,
Dordrecht
423. Tsuji, M.: Potential Theory in Modern Function Theory. Maruzen Co., Ltd, Tokyo (1959)
424. Tsygvintsev, A.: Algebraic invariant curves of plane polynomial differential systems. J. Phys.
A 34, 663–672 (2001)
425. Valls, C.: Rikitake system: analytic and Darbouxian integrals. Proc. Roy. Soc. Edinb. Sect. A
135, 1309–1326 (2005)
426. Valls, C.: Invariant algebraic surfaces for generalized Raychaudhuri equations. Commun.
Math. Phys. 308, 133–146 (2011)
427. Valls, C.: Liouvillian integrability of a deterministic business cycle model. Anal. Appl. (Sin-
gap.) 11(6), 1350032 (2013). 9 pp
References 369

428. Valls, C.: Invariant algebraic surfaces for a virus dynamics. Z. Angew. Math. Phys. 66, 1315–
1328 (2015)
429. Van der Pol, B.: On relaxation-oscillations. Lond. Edinb. Dublin Phil. Mag. J. Sci 2, 978–992
(1927)
430. Vanecek, A., Celikovsksý, S.: Control Systems: From Linear Analysis to Synthesis of Chaos.
Prentice-Hall, London (1996)
431. Vigo-Aguiar, M.I., Sansaturio, M.E., Ferrándiz, J.M.: Integrability of Hamiltonians with poly-
nomial potentials. J. Comput. Appl. Math. 158, 213–224 (2003)
432. Villarini, M.: Regularity properties of the period function near a center of planar vector fields.
Nonlinear Anal. 19, 787–803 (1992)
433. Villarini, M.: On the problem of existence of holomorphic first integrals. J. Math. Anal. Appl.
194, 799–822 (1995)
434. Villarini, M.: Algebraic nonsolvability of the problem of existence of holomorphic first inte-
grals. J. Dyn. Control Syst. 1, 403–425 (1995)
435. Viro, O.: From the sixteenth Hilbert problem to tropical geometry. Jpn. J. Math. 3, 185–214
(2008)
436. Walcher, S.: On differential equations in normal form. Math. Ann. 291, 293–314 (1991)
437. Walcher, S.: On the Poincaré problem. J. Differ. Equ. 166, 51–78 (2000)
438. Warner, F.W.: Foundations of differentiable manifolds and Lie groups. Foresman and Co.,
London, Scott (1971)
439. Weinstein, A.: The local structure of Poisson manifolds. J. Diff. Geom. 18, 523–557 (1983)
440. Weng, S.L., Zhang, X.: Integrability of vector fields versus inverse Jacobian multipliers and
normalizers. Discret. Contin. Dyn. Syst. 36, 6539–6555 (2016)
441. Whitney, H.: Differentiable manifolds. Ann. Math. 37, 645–680 (1936)
442. Wilson, C.: Algebraic periodic solutions of Liénard equations. Contrib. Differ. Equ. 3, 1–20
(1964)
443. Wilson, G.: Hilbert’s sixteenth problem. Topology 17, 53–74 (1978)
444. Wu, H., Li, W.G.: Poincaré type theorems for non-autonomous systems. J. Differ. Equ. 245,
2958–2978 (2008)
445. Wu, K., Zhang, X.: Darboux polynomials and rational first integrals of the generalized Lorenz
systems. Bull. Sci. Math. 136, 291–308 (2012)
446. Wu, K., Zhang, X.: Global dynamics of the generalized Lorenz systems having invariant
algebraic surfaces. Phys. D 244, 25–35 (2013)
447. Wu, K., Zhang, X.: Analytic normalization of analytically integrable differential systems near
a periodic orbit. J. Differ. Equ. 256, 3552–3567 (2014)
448. Xie, F., Zhang, X.: Invariant algebraic surfaces of the Rabinovich system. J. Phys. A 36,
499–516 (2003)
449. Yablonskii, A.I.: Limit cycles of certain differential equation. Differ. Equ. 2, 335–344 (1966)
450. Yakovenko, S.: Smooth normalization of a vector field near a semistable limit cycle. Ann.
Inst. Fourier, Grenoble 43, 893–903 (1993)
451. Ye, Y.Q.: Theory of Limit Cycles. Translation of Mathematical Monographs, vol. 66, 2nd
edn. American Mathematical Society, Providence (1986)
452. Ye, Y.Q.: Qualitative Theory of Polynomial Differential Systems. Shanghai Science & Tech-
nology Pub, Shanghai (1995)
453. Yoshida, H.: A criterion for the nonexistence of an additional integral in Hamiltonian systems
with a homogeneous potential. Phys. D 29, 128–142 (1987)
454. Yoshida, H.: A criterion for the nonexistence of an additional analytic integral in Hamiltonian
systems with n degrees of freedom. Phys. Lett. A 141, 108–112 (1989)
455. Yoshida, H.: A new necessary condition for the integrability of Hamiltonian systems with a
two-dimensional homogeneous potential. Phys. D 128, 53–69 (1999)
456. Yoshida, H.: Nonexistence of an additional analytic integral in Hamiltonian systems with an
n-dimensional homogeneous potential, Singular Behavior and Nonlinear Dynamics, Vol. 1
(Sámos,1989): 114–117. World Scientific Publ, Teaneck, NJ (1988)
370 References

457. Yu, X., Zhang, X.: The hyperelliptic limit cycles of the Liénard systems. J. Math. Anal. Appl.
376, 535–539 (2011)
458. Zariski, O., Samuel, P.: Commutative Algebra, vol. II. Springer, New York (1960)
459. Zernov, A.E., Scárdua, B.A.: Integration of polynomial ordinary differential equations in the
real plane. Aequ. Math. 61, 190–200 (2001)
460. Zhang, X.: Integrals of motion of the Rabinovich system. J. Phys. A 33, 5137–5155 (2000)
461. Zhang, X.: Exponential factors and Darbouxian first integrals of the Lorenz system. J. Math.
Phys. 43, 4987–5001 (2002)
462. Zhang, X.: Invariant hyperplanes and Darboux integrability of polynomial vector fields. J.
Phys. A 35, 9931–9941 (2002)
463. Zhang, X.: Invariant algebraic curves and rational first integrals of holomorphic foliations in
C P(2). Sci. China Ser. A 46, 271–279 (2003)
464. Zhang, X.: Local first integrals for systems of differential equations. J. Phys. A 36, 12243–
12253 (2003)
465. Zhang, X.: Exponential factors and Darboux integrability for the Rössler system. Int. J. Bifur.
Chaos 14, 4275–4283 (2004)
466. Zhang, X.: Analytic normalization of analytic integrable systems and the embedding flows.
J. Differ. Equ. 244, 1080–1092 (2008)
467. Zhang, X.: Embedding diffeomorphisms in flows in Banach spaces. Ergod. Theory Dyn. Syst.
29, 1349–1367 (2009)
468. Zhang, X.: Embedding smooth diffeomorphisms in flows. J. Differ. Equ. 248, 1603–1616
(2010)
469. Zhang, X.: The embedding flows of C ∞ hyperbolic diffeomorphisms. J. Differ. Equ. 250,
2283–2298 (2011)
470. Zhang, X.: Global structure of quaternion polynomial differential equations. Commun. Math.
Phys. 303, 301–316 (2011)
471. Zhang, X.: The 16th Hilbert problem on algebraic limit cycles. J. Differ. Equ. 251, 1778–1789
(2011)
472. Zhang, X.: Comment on On the polynomial integrability of the Kirchoff equations, Physica
D 241 (2012) 1417–1420. Phys. D 250, 47–51 (2013)
473. Zhang, X.: Ordinary Differential Equations (in Chinese). Science Press, Beijing (2015)
474. Zhang, X.: Analytic integrable systems: analytic normalization and embedding flows. J. Differ.
Equ. 254, 3000–3022 (2013)
475. Zhang, X.: Inverse Jacobian multipliers and Hopf bifurcation in higher dimensional differen-
tial systems. J. Differ. Equ. 256, 3278–3299 (2014)
476. Zhang, X.: The embedding flow of 3-dimensional locally Hyperbolic C ∞ diffeomorphisms.
J. Dyn. Differ. Equ. 27, 29–54 (2015)
477. Zhang, X.: Liouvillian integrability of polynomial differential systems. Trans. Am. Math.
Soc. 368, 607–622 (2016)
478. Zhang, X., Ye, Y.Q.: On the number of invariant lines for polynomial systems. Proc. Am.
Math. Soc. 126, 2249–2265 (1998)
479. Zhang, X.K.: Number of integral lines of polynomial systems of degree three and four, pp.
209–212. J. Nanjing Univ. Math. Biquarterly, Suppl (1993)
480. Zhang, Z., Ding, T.R., Huang, W.Z., Dong, Z.X.: Qualitative Theory of Differential Equa-
tions. Translation of Mathematical Monographs, vol. 101. American Mathematical Society,
Providence (1992)
481. Ziglin, S.L.: Branching of solutions and nonexistence of first integrals in Hamiltonian mechan-
ics I. Funct. Anal. Appl. 16, 181–189 (1983)
482. Ziglin, S.L.: Bifurcation of solutions and the nonexistence of first integrals in Hamiltonian
mechanics. I (Russian), Func. Anal. Appl. 16(3), 30–41 (1982)
483. Ziglin, S.L.: Bifurcation of solutions and the nonexistence of first integrals in Hamiltonian
mechanics. II (Russian), Func. Anal. Appl. 17(1), 8–23 (1983)
484. Ziglin, S.L.: On first integrals of groups of complex linear transformations and of natural
mechanical systems with a homogeneous potential (Russian). Mat. Zametki 70, 839–844
(2001)
References 371

485. The classification of reversible cubic systems with center: Żoła̧dek. H. Topol. Methods Non-
linear Anal. 4, 79–C136 (1994)
486. Algebraic invariant curves for Liénard equation: Żoła̧dek. H. Trans. Am. Math. Soc. 350,
1681–1701 (1998)
487. Zung, N.T.: Convergence versus integrability in Birkhoff normal form. Ann. Math. 161, 141–
156 (2005)
488. Zung, N.T.: Convergence versus integrability in Poincar-Dulac normal form. Math. Res. Lett.
9(2–3), 217–228 (2002)
Index

A invertible, 55
Algebraic Annular flow, 11, 15
dependent, 223 Anti-commutative law, 84
equation, 99, 198 Anti-symmetry, 82, 84, 201
first integral, 197, 198 Associative law, 84
formal power series, 244 Atlas, 4
function, 197
independent, 224
invariant, 255 B
limit cycle, 168 Bendixson compactification, 18
topology, 159 Bendixson sphere, 18
Algebraically closed, 128 Betti number, 159
Algebraically independent, 127, 308 Bézout’s theorem, 19, 20, 261
Algebraically reducible, 257 Bifurcation, 10, 46
Algebraic closure, 128 Bilinearity, 82, 84
Algebraic element, 127, 129 Blasius equation, 277
Algebraic extension, 128 Blow up, 157, 159
of the field, 127 Bruno’s theorem, 324
Algebraic field extension, 128, 135
Algebraic geometry, 152
Algebraic integrable, 198 C
Algebraic limit cycle, 269 Canonical Hamiltonian system, 35
Algebraic multiplicity, 100, 101, 113 Canonical region, 13, 14, 19, 20
α limit set, 11, 18, 19, 54, 58 C r parallel, 14, 15
Analytic, 1 Carnicer’s theorem, 157
diffeomorphism, 347 Casimir function, 201
function, 1 Cauchy inequality, 326, 346
integrable, 327 Center, 18, 53
normalization, 329, 333 isochronous, 321
perturbation, 73 Center annulus, 321
submanifold, 15 Center condition, 255
Analytically equivalent Center-focus problem, 59, 254
to normal form, 291, 333 Center-focus theorem, 62
Analytic first integral Center hyperplane, 60
of periodic differential system, 301 Center manifold, 60, 61
Analytic integrable, 330 C ∞ , 75
Analytic transformation analytic, 61, 68
© Springer Nature Singapore Pte Ltd. 2017 373
X. Zhang, Integrability of Dynamical Systems: Algebra and Analysis,
Developments in Mathematics 47, DOI 10.1007/978-981-10-4226-3
374 Index

smooth, 70 Cyclicity, 58, 74

Center manifold theorem, 60, 62 of a singularity, 35
Cerveau–Lins Neto theorem, 156 Cylindrical change of coordinates, 75
Characteristic
equation, 21, 23
of a field, 126 D
zero, 126 Darboux element, 164
Characteristic curve, 178, 180 Darboux first integral, 92
Characteristic exponent, 328 local, 315
Characteristic method, 167 Darboux function, 92
Characteristic multiplier, 299 Darboux integrability, 276, 277
Characteristic subspace, 301 Darboux integrating factor, 93
Chart, 113 Darboux invariant polynomial, 115
of a manifold, 4 Darboux Jacobian multiplier, 93
Chemical Reaction Network, 282 Darboux–Jouanolou integrability theorem,
Chen system, 178 94, 98
C k completely integrable, 6 Darboux lemma, 265, 266
C k diffeomorphism, 4 Darboux point, 228
C k discrete dynamical system, 329 Darboux polynomial, 90
C k dynamical system, 329 canonical, 317
C k equivalence, 8 Darboux semi-invariant
local, 8 analytic, 314
C k equivalent, 7, 8 canonical, 316, 317
C k first integral, 6 canonical formal, 316
C k integrable, 6, 330 formal, 314
C k orbital equivalence, 8 Darboux theory of integrability, 89
C k parallel flow, 11 local, 125
C k parallel neighborhood, 12 Degree, 123
C k smooth function, 4 Degree of a continuous map, 159
C k smooth manifold, 4 Degree of a holomorphic foliation, 155
C k smooth map, 4 Degree of a polynomial DS, 11, 90
Closed form, 36, 156 Degree of a vector field, 11
Cofactor, 90, 91, 116, 314, 316 Degree of differential field extension, 128
Commutativity with differential, 85 Deterministic business model, 276
Compactified vector field, 18, 19 Dicritical, 285
Complete, 15 Dicritical singularity, 157
Complete C r flow, 15 Diffeomorphism
Complete flow, 14 analytic, 329
Completely integrable, 6, 30, 44 smooth, 329
locally C k , 6 Differential 0-form, 83
Complete metric Differential 1-form, 83
geodesically, 15 Differential k-form, 84
Composite, 242 Differential equation
Condition ω of Bruno, 324, 328 hyperbolic geometric, 151
Condition W∗ , 116 Differential field, 127
Condition A of Bruno, 324, 328 Differential field extension, 127
Conjugacy, 7 elementary, 129
Convex hull, 123 Liouvillian, 129
Coordinate chart, 113 Differential form, 35, 36
of a manifold, 4 Differential Galois group, 228
C r manifold, 15 Differential Galois theory, 228, 229
Critical point, 7, 39 Differential of differential form, 84
Critical value, 39 Differential one-form, 35, 153
Index 375

Differential system, 1 Embedding flow problem

analytic, 1, 11, 49 of diffeomorphism, 329
autonomous, 1, 3, 330 Embedding vector field, 329
C ∞, 1 analytic, 332
C k smooth, 1 of a map, 50
completely integrable, 44 Equilibrium, 7
cubic, 253 Equivalent potential, 231
finite-dimensional, 44 Error function, 130
formal, 56, 288 Euler characteristic, 157, 159–162
higher-dimensional, 59 intrinsic, 159
homogeneous, 27 Euler equation, 211, 212
integrable, 57, 125 Euler formula
Liénard, 147, 167 generalized, 144
linear, 2 Exact, 35
Liouvillian integrable, 145 Exact form, 54
nonautonomous, 3 Exponential factor, 91, 116
nonautonomous polynomial, 115 Extactic curve, 101
ordinary, 131 Extactic point, 101
periodic, 299, 329 Extactic polynomial, 101
planar, 11, 35 Extactic surface, 101
polynomial, 11, 89, 294 Extended phase plane, 2
quadratic, 150, 254 Extended phase space, 2
quasiperiodic, 302, 303 Exterior derivative, 84
rational, 198
smooth, 1
two-dimensional, 44 F
Discrete dynamical system Falkner–Skan equation, 277
analytic integrable, 329 Fibre, 15
Displacement function, 77 Fibre bundle, 15
Distinguished normal form, 57, 291, 293, Field automorphism, 128
303, 338 Finite field extension, 129, 308
Distinguished normalization, 57, 65, 291, Finite resonant relation, 289
303 First integral, 2, 116
analytic, 14
analytic local, 61
E continuous, 13
Eigenvalue, 16, 28, 52 continuously differentiable, 5
Eigenvector, 29 Darboux, 90, 99
Einstein–Yang–Mills equations, 282 elementary, 130
Elementary center, 320 formal, 66, 288
Elementary extension of a field, 129 formal meromorphic, 307
Elementary field extension, 129, 130 functionally independent, 3, 6
Elementary first integral, 126, 132 holomorphic, 206
Elementary integrability, 89 Liouvillian, 130
Elementary integrable, 130 local Darboux, 314
Elementary singularity, 52, 53 meromorphic, 222, 314
Elliptic sector, 18 of diffeomorphism, 330
Embedding diffeomorphism of differential form, 35
in vector field, 329 of Hamiltonian system, 35
Embedding DS of local diffeomorphism, 299
formal, 329 polynomial, 93
Embedding flow, 50, 329 rational, 93, 99
of integrable diffeomorphism, 330 weak, 13
376 Index

First-order Heteroclinic cycle, 54

variational equations, 236 Hide–Skeldon–Acheson dynamo, 278
Flat, 62 Higher-order
Flow, 7 variational equations, 236, 237
Flow box theorem, 7, 44 Higher-order Abel equation, 283
Focus, 16, 18, 53 Higher-order Riccati equation, 283
Foliation, 284 Hilbert–Rückert Nullstellensatz, 315
Formal first integral Hilbert’s 16th problem, 11, 269
of diffeomorphisms, 299 a weak version, 269
of periodic differential system, 301 Hilbert’s Nullstellensatz, 104, 108
Formally equivalent, 55 Holomorphic foliation, 149, 153, 154
to normal form, 291 Homeomorphism, 4, 7
Formally linearizable, 56 Homoclinic cycle, 54
Formally orbitally equivalent, 56 Homogeneous coordinates, 152
Formally orbitally linearizable, 56 Homogeneous potential, 230
Formal series, 55 Homology class, 161
invertible, 56 Hopf bifurcation, 73
local invertible, 55 Hopf-cyclicity, 74
Formal transformation Hopf–Rinow theorem, 15
invertible, 55, 291 Hyperbolic diffeomorphism, 330
near identity, 55 Hyperbolic geometric function, 151
tangent to identity, 55 Hyperbolic node, 53
Fourth first integral, 202 Hyperbolic saddle, 53
Frobenius integrability theorem, 205 Hyperbolic sector, 12, 18
Functionally independent, 4, 308 Hyperbolic singularity, 53
Hyperelliptic limit cycle, 169

G
Galois group, 128, 133, 135 I
Generalized Darboux function, 116 Implicit Function Theorem, 22, 24, 206
Generalized Lyapunov quantity, 335 Improper Darboux point, 238
Generalized polynomial, 115 Independent set, 98
Generalized polynomial first integral, 116 Independent singularities, 98, 99
Generalized Puiseux series, 336 Infinitesimal generator, 81
Generalized rational first integral, 116 Initial point, 2, 3
Generator Integrability condition, 154
of Darboux polynomials, 179 Integrable, 6
Generic invariant algebraic curves, 269 locally C k , 6
Genus, 17 Integrable diffeomorphism, 330
Global analytic analytic, 330, 333
first integral, 216 Integrating factor, 36
analytic, 67
Intersection number, 259, 261
H Invariant
Hamiltonian function, 36 algebraic curve, 89, 90, 155
Hamiltonian system, 35 algebraic hypersurface, 89, 90
analytic, 330 algebraic surface, 89, 90
C ∞ , 330 integral, 37–39
natural, 221 set, 90
polynomial, 221 Invariant manifold, 59, 60
Hamiltonian vector field, 201 Invariant set, 3, 6
Harnack’s theorem, 270, 272 Inverse function theorem, 9
Hermite polynomial, 151 Inverse integrating factor, 46
Index 377

analytic, 49 Liouvillian curve, 140

multiplicity of zero, 48 Liouvillian extension
vanishing multiplicity, 48 of a field, 129
Inverse Jacobian multiplier, 45 Liouvillian field extension, 130
C ∞ , 59 Liouvillian function, 140
Inverse map, 15 Liouvillian integrability, 147, 278
Inverse transformation, 180, 182 Liouvillian integrable, 130
Invertible transformation, 5 Local Lie group, 80
Involution, 201 r -parameter, 80
Local ring, 259
Local transformation group, 80, 81
J
Lorenz system, 177
Jacobian multiplier, 37, 116
Jacobi identity, 82, 201 generalized, 178, 179
Jacobi’s theorem, 40 Lyapunov center theorem, 61
Jordan normal form, 57 Lyapunov constant, 254, 256

K
M
Kapteyn–Bautin theorem, 254
m-adic topology, 288
k-functionally independent, 94
Kirchoff equations, 200 Majorant series, 326
Kirchoff vector field, 203, 204 Manakov condition, 212, 213
Manifold, 4
analytic, 162
L orientable, 4
(λ : μ : ν) resonant, 350 simply connected, 54
Laurent series, 136, 137 with boundary, 4
Lax pair, 26, 27 Maximal ideal, 287
equation, 27 Maximum period annulus, 171
Lebesgue measure, 97 Meromorphic function, 288
Leibniz identity, 85 formal, 288
Level curve, 3 Michelson system, 281
Level hypersurface, 2 Minimal algebraic equation, 198
Level surface, 3 Minimal field extension, 129
Lie algebra, 211 Minimal growth, 344
finite-dimensional, 211 Minimal monomic polynomial, 198
Lie bracket, 81, 82 Minimal polynomial, 244
Lie derivative, 84
Minimal solution, 245
of a smooth function, 85
Monodromic, 74
of a vector field, 85
Monodromy group, 229
of differential form, 85
Monodromy operator, 299
Lie group, 80
r -parameter, 79 Morales–Ramis theory, 227–229, 233
Lie symmetry, 46 Multiplicatively invertible, 287
Limit cycle, 11 Multiplicity
hyperbolic, 48, 273 of a homoclinic loop, 35
of multiplicity m, 48 of a limit cycle, 35
Limit set, 17, 46, 54, 87 of a singularity, 53, 297
Linearity, 85, 86 of limit cycle, 47
Linearly independent, 29 Multiplicity of a foliation
Linear transformation on a invariant curve at a point, 158
invertible, 60 Multiplicity of a vector fields at a singularity,
Liouville’s theorem, 37, 38, 46 159
378 Index

N Phase orbit, 9
Neumann’s theorem, 14, 15 Phase plane, 2
Newton polytope, 122 Phase space, 2
Nilpotent matrix, 290 Pluri-Green function, 344
Nilpotent part, 290 Pluripolar set, 343
Nilpotent singularity, 53 Plurisubharmonic function, 343
Nodal curve, 156 Poincaré–Bendixson theorem, 17, 18
Nodal point, 156 Poincaré center theorem, 59, 72
Nonalgebraic limit cycle, 168 Poincaré compactification, 18
Non-composite, 242 Poincaré domain, 294
Nondegenerate Poincaré–Dulac normal form theorem, 57,
monodromy singularity, 254 291, 293
Nondegenerate center, 53, 255 Poincaré–Hopf index theorem, 161
Nondegenerate singularity, 53 Poincaré–Hopf theorem, 159
Nondicritical singularity, 157, 162, 274 Poincaré linearization theorem, 56, 57
Nonregular Lax pair, 27 Poincaré map, 47
Nonresonant, 57 Poincaré nonintegrability theorem, 295, 296,
Nonresonant eigenvalues, 289, 302 300, 301
Nonresonant monomial, 67, 333 Poincaré normal form, 57, 290, 329, 333
Nonsingular curve, 157, 273 of a diffeomorphism, 333
Normal closure, 128, 133, 135 Poincaré normal form theorem, 294
Normal form, 50, 57, 303 Poincaré section, 47
analytic, 333 Poincaré theorem, 38
Normal form on an invariant manifold, 337 Polycycle, 17
Normalization, 291, 303, 333 Polynomial
analytic, 125, 333 first integral, 197
formal, 305 function, 309
Normal variational equation, 239 Polynomial differential system
Nosé–Hoover system, 279 quasi-homogeneous, 143, 144
Polynomial integrable, 198
Polynomial invariant hypersurface, 116
O Polynomial vector field, 17, 94
ω limit set, 11, 17–19, 54, 58 Positively invariant set, 3
Orbit, 2 p : −q resonant
Orbital equivalence singularity, 350
local C k , 8 Prelle–Singer theorem, 131
Order Product condition, 212, 213
of a curve at a singularity, 271 Projective condition, 153, 154
of a formal series, 287 Projective holomorphic one-form, 258
of a group, 128 Projective planar curve, 20
of differentiability, 14 of degree m, 20
Projective plane, 20, 149, 152, 156
Projective space, 152
P Proper Darboux polynomial, 179, 210
Partial differential equation Pseudomonomial, 303
linear homogeneous, 21 Puiseux parametrization, 158
quasilinear, 23 Puiseux series, 158, 336
Partial integrability variety, 335
Partially integrable, 335
Period annulus, 12, 49, 171 Q
Periodic orbit, 11 Q-linearly independent, 314
Pfaffian form, 35 Q+ -linearly independent, 213, 214, 301
Phase flow, 38 Quasi-homogeneous polynomial, 144
Index 379

of weight degree d, 144 Semi-quasihomogeneous vector field, 307

Quasi-normal form, 66 Semisimple, 29
Quasiperiodic, 303 Semisimple-nilpotent decomposition, 319
function, 303 Semisimple part, 290
Quasiperiodic vector field, 301 Separatrix, 12, 155
Quotient space, 14, 15, 20, 152 Siegel domain, 294
σ -antisymmetric, 203
σ process, 161
R Simple monodromy singularity, 54
Radial flow, 11 Singer’s theorem, 133, 143
Radial vector field, 154 Singular holomorphic foliation, 152, 155,
Rational 258
first integral, 197 Singularity, 7, 10
Rational flow, 11 Singular point, 7
Rational function, 93 Singular set, 153
Rational integrable, 198 Smooth manifold, 81
Rational monomial, 312 n-dimensional, 4
Rational reducible, 258 Spectrum, 243
Reduced Spiral flow, 11
rational function, 241 Splitting field, 128, 138
Reduced polynomial, 155 Stable hyperplane, 60
Reeb center theorem, 61 Stable manifold, 60
Regular Lax pair, 27 Stable manifold theorem, 60
Regular point, 4, 7, 39 Stereographic projection, 18, 19
Regular surface, 125 Strip flow, 11
Regular value, 4 Strong focus, 53
Remarkable values, 243 Strong hyperbolic saddle, 53
Rescaling group, 80, 81 Strong resonant saddle, 55
Resonant, 57, 303 Structure matrix, 201
of a Hamiltonian vector field, 201
first, 303
Sturm–Liouville operator, 26
rational homogeneous function, 312
Supporting hyperplane, 122
rational monomial, 312
Symmetry group, 81–83
second, 303
Symplectic diffeomorphism, 330
Resonant eigenvalues, 289
Resonant lattice, 214, 322, 333
Resonant monomial, 57, 290, 333 T
Resonant pseudomonomial, 306 Time reversible, 348
Resonant rank of a matrix, 333 Topologically equivalent, 7
Resonant relation, 56 Toral flow, 11, 16
Resonant saddle, 55 Tower element, 129
Resonant with each other, 338, 340 Tower form, 129
Resultant, 241 Trajectory, 2
Riccati differential system, 277 Transcendental degree, 127, 308
Riemann surface, 159 Transcendental element
Rotation group, 79 of a field, 127
Transcendental function, 198
Transformation group, 80
S Translation group, 80
Saddle, 16 Transversal section, 15, 76
Saddle quantity, 58
Second first integral, 221
Seidenberg’s theorem, 275 U
Semihyperbolic singularity, 53 Unipotent diffeomorphism, 329
380 Index

Universal covering manifold, 54 of order m, 256

Unstable hyperplane, 60 Weak hyperbolic saddle, 53
Unstable manifold, 60 Weakly nonresonant, 302, 306
Unstable manifold theorem, 60 Weak nonresonance, 330
Upper semicontinuous, 343 Wedge product, 83
of differential form, 84
Weierstrass preparation theorem, 78
V
Vanishing multiplicity Weight degree, 144
of inverse Jacobian multiplier, 35, 64, 74 Weight homogeneous polynomial, 178
Variational equation, 229, 235 Whitney’s embedding theorem, 15
Variety, 334 Wronskian, 116
Vector field, 2, 10 Wronskian matrix, 116
analytic, 17
autonomous, 302
quasiperiodic, 303
Z
Virus model, 281
Zero of multiplicity m, 49
Żoła̧dek, 141, 169
W Z-resonance, 307
Weak focus, 53, 320 Zung, 294