Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

International Journal of Bifurcation and Chaos, Vol. 16, No.

4 (2006) 887–910
c World Scientific Publishing Company

DIFFERENTIAL GEOMETRY AND MECHANICS:


APPLICATIONS TO CHAOTIC
DYNAMICAL SYSTEMS
JEAN-MARC GINOUX∗ and BRUNO ROSSETTO†
PROTEE Laboratory,
I.U.T. of Toulon, University of South,
B.P. 20132, 83957, LA GARDE Cedex, France
∗ginoux@univ-tln.fr
†rossetto@univ-tln.fr

Received May 16, 2005; Revised July 1, 2005

The aim of this article is to highlight the interest to apply Differential Geometry and Mechanics
concepts to chaotic dynamical systems study. Thus, the local metric properties of curvature and
torsion will directly provide the analytical expression of the slow manifold equation of slow-fast
autonomous dynamical systems starting from kinematics variables (velocity, acceleration and
over-acceleration or jerk).
The attractivity of the slow manifold will be characterized thanks to a criterion proposed
by Henri Poincaré. Moreover, the specific use of acceleration will make it possible on the one
hand to define slow and fast domains of the phase space and on the other hand, to provide
an analytical equation of the slow manifold towards which all the trajectories converge. The
attractive slow manifold constitutes a part of these dynamical systems attractor. So, in order to
propose a description of the geometrical structure of attractor, a new manifold called singular
manifold will be introduced. Various applications of this new approach to the models of Van der
Pol, cubic-Chua, Lorenz, and Volterra–Gause are proposed.

Keywords: Differential geometry; curvature; torsion; slow-fast dynamics; strange attractors.

1. Introduction the eigenmode associated in the “fast” eigenvalue is


There are various methods to determine the evanescent. Thus, the tangent linear system approx-
slow manifold analytical equation of slow-fast imation method, presented in the appendix, pro-
autonomous dynamical systems (S-FADS), or vides the slow manifold analytical equation of a
autonomous dynamical systems considered as slow- dynamical system according to the “slow” eigenvec-
fast (CAS-FADS). A classical approach based on tors of the tangent linear system, i.e. according to
the works of Andronov [1966] led to the famous the “slow” eigenvalues. Nevertheless, according to
singular approximation. For ε = 0, another method the nature of the “slow” eigenvalues (real or com-
called: tangent linear system approximation, devel- plex conjugated) the plot of the slow manifold ana-
oped by Rossetto et al. [1998], consists in using lytical equation may be difficult even impossible.
the presence of a “fast” eigenvalue in the func- Also to solve this problem it was necessary to make
tional jacobian matrix of a (S-FADS) or of a (CAS- the slow manifold analytical equation independent
FADS). Within the framework of application of of the “slow” eigenvalues. This could be carried out
the Tihonov’s theorem [1952], this method uses by multiplying the slow manifold analytical equa-
the fact that in the vicinity of the slow manifold tion of a two-dimensional dynamical system by a

887
888 J.-M. Ginoux & B. Rossetto

“conjugated” equation, that of a three-dimensional and


dynamical system by two “conjugated” equations. (X) = [f1 (X), f2 (X), . . . , fn (X)]t ∈ E ⊂ Rn
In each case, the slow manifold analytical equation
independent of the “slow” eigenvalues of the tangent The vector (X) defines a velocity vector field
linear system is presented in the appendix. in E whose components fi which are supposed to be
The new approach proposed in this article is continuous and infinitely differentiable with respect
based on the use of certain properties of Differential to all xi and t, i.e. are C ∞ functions in E and with
Geometry and Mechanics. Thus, the metric proper- values included in R, satisfy the assumptions of the
ties of curvature and torsion have provided a direct Cauchy–Lipschitz theorem. For more details, see for
determination of the slow manifold analytical equa- example [Coddington & Levinson, 1955]. A solu-
tion independently of the “slow” eigenvalues. It has tion of this system is an integral curve X(t) tangent
been demonstrated that the equation thus obtained to  whose values define the states of the dynami-
is completely identical to that which the tangent cal system described by Eq. (1). Since none of the
linear system approximation method provides. components fi of the velocity vector field depends
The attractivity or repulsivity of the slow man- here explicitly on time, the system is said to be
ifold could be characterized while using a crite- autonomous.
rion proposed by Henri Poincaré [1881] in a report Note: In certain applications, it would be sup-
entitled “Sur les courbes définies par une equation posed that the components fi are C r functions in
différentielle”. E and with values in R, with r ≥ n.
Moreover, the specific use of the instantaneous
acceleration vector allowed a kinematic interpreta- 2.2. Slow-fast autonomous dynamical
tion of the evolution of the trajectory curves in the system (S-FADS )
vicinity of the slow manifold by defining the slow
and fast domains of the phase space. A (S-FADS) is a dynamical system defined under
At last, two new manifolds called singular were the same conditions as above but comprising a small
introduced. It was shown that the first, the singu- multiplicative parameter ε in one or several compo-
lar approximation of acceleration, constitutes a first nents of its velocity vector field:
approximation of the slow manifold analytical equa- dX
tion of a (S-FADS) and is completely equivalent to = (X) (2)
dt
the equation provided by the method of the succes-
with
sive approximations developed by Rossetto [1986]  
while the second, the singular manifold, proposes an dX dx1 dx2 dxn t
= ε , ,..., ∈ E ⊂ Rn
interpretation of the geometrical structure of these dt dt dt dt
dynamical systems attractor. 0<ε1

2. Dynamical System, Slow-Fast The functional jacobian of a (S-FADS) defined


by (2) has an eigenvalue called “fast”, i.e. with a
Autonomous Dynamical System
large real part on a large domain of the phase space.
(S-FADS), Considered as Slow-Fast Thus, a “fast” eigenvalue is expressed like a polyno-
Autonomous Dynamical System mial of valuation −1 in ε and the eigenmode which
(CAS-FADS) is associated in this “fast” eigenvalue is said:
The aim of this section is to recall definitions and • “evanescent” if it is negative,
properties of (S-FADS) and of (CAS-FADS). • “dominant” if it is positive.
2.1. Dynamical system The other eigenvalues called “slow” are
expressed like a polynomial of valuation 0 in ε.
In the following we consider a system of differential
equations defined in a compact E included in R:
2.3. Dynamical system considered
dX as slow-fast (CAS-FADS )
= (X) (1)
dt
It has been shown [Rossetto et al., 1998] that
with a dynamical system defined under the same con-
X = [x1 , x2 , . . . , xn ]t ∈ E ⊂ Rn ditions as (1) but without small multiplicative
Differential Geometry and Mechanics: Applications to Chaotic Dynamical Systems 889

parameters in one of the components of its veloc- the total derivative of V(t) is the vector function
ity vector field, and consequently without singu- γ(t) of the scalar variable t which represents the
lar approximation, can be considered as slow -fast if instantaneous acceleration vector of the mobile M
its functional jacobian matrix has at least a “fast” at the instant t; namely:
eigenvalue, i.e. with a large real part on a large
dV
domain of the phase space. γ(t) = (4)
dt
3. New Approach of the Slow Since the functions fi are supposed to be C ∞
Manifold of Dynamical functions in a compact E included in Rn , it is pos-
Systems sible to calculate the total derivative of the vec-
tor field V(t) defined by (1) or (2). By using the
This approach consists in applying certain con-
derivatives of composite functions, we can write the
cepts of Mechanics and Differential Geometry to
derivative in the sense of Fréchet:
the study of dynamical systems (S-FADS or CAS-
FADS). Mechanics will provide an interpretation of dV d dX
= (5)
the behavior of the trajectory curves, integral of dt dX dt
a (S-FADS) or of a (CAS-FADS), during the var- By noticing that d/dX is the functional
ious phases of their motion in terms of kinemat- jacobian matrix J of the system (1) or (2), it follows
ics variables: velocity and acceleration. The use of from Eqs. (4) and (5) that we have the following
Differential Geometry, more particularly the local equation which plays a very important role:
metric properties of curvature and torsion, will
make it possible to directly determine the analytical γ = JV (6)
equation of the slow manifold of (S-FADS) or of
(CAS-FADS). 3.1.3. Tangential and normal components
of the instantaneous acceleration
3.1. Kinematics vector functions vector
Since our proposed approach consists in using the By making the use of the Frénet [1847] frame, i.e. a
Mechanics formalism, it is first necessary to define frame built starting from the trajectory curve X(t)
the kinematics variables needed for its development. directed towards the motion of the mobile M . Let
Thus, we can associate the integral of the system (1) us define τ the unit tangent vector to the trajectory
or (2) with the coordinates, i.e. with the position, curve in M , ν the unit normal vector, i.e. the prin-
of a moving point M at the instant t. This inte- cipal normal in M directed towards the interior of
gral curve defined by the vector function X(t) of the concavity of the curve and β the unit binormal
the scalar variable t represents the trajectory curve vector to the trajectory curve in M so that the tri-
of the moving point M . hedron (τ , ν, β) is direct. Since the instantaneous
velocity vector V is tangent to any point M to the
3.1.1. Instantaneous velocity vector trajectory curve X(t), we can construct a unit tan-
As the vector function X(t) of the scalar variable t gent vector as follows:
represents the trajectory of M , the total derivative V
τ = (7)
of X(t) is the vector function V(t) of the scalar vari- V
able t which represents the instantaneous velocity In the same manner, we can construct a unit
vector of the mobile M at the instant t; namely: binormal, as:
dX V∧γ
V(t) = = (X) (3) β= (8)
dt V ∧ γ
The instantaneous velocity vector V(t) is sup- and a unit normal vector, as:
ported by the tangent to the trajectory curve.
τ̇ V⊥
ν =β∧τ = = (9)
3.1.2. Instantaneous acceleration vector τ̇  V⊥ 
with
As the instantaneous vector function V(t) of the
scalar variable t represents the velocity vector of M , V = V⊥  (10)
890 J.-M. Ginoux & B. Rossetto

where the vector V⊥ represents the normal vec- 3.2.1. Parametrization of the
tor to the instantaneous velocity vector V directed trajectory curve
towards the interior of the concavity of the tra- The trajectory curve X(t) integral of the dynami-
jectory curve and where the dot (·) represents cal system defined by (1) or (2), is described by the
the derivative with respect to time. Thus, we can motion of a current point M position which depends
express the tangential and normal components of on a variable parameter: the time. This curve can
the instantaneous acceleration vector γ as: also be defined by its parametric representation
γ·V relative in a frame:
γτ = (11)
V x1 = F1 (t), x2 = F2 (t), . . . , xn = Fn (t)
γ ∧ V where the Fi functions are continuous, C ∞ func-
γν = (12)
V tions (or C r+1 according to the above assumptions)
By noticing that the variation of the Euclidean in E and with values in R. Thus, the trajectory
norm of the instantaneous velocity vector V can be curve X(t) integral of the dynamical system defined
written: by (1) or (2), can be considered as a plane curve
or as a space curve having certain metric proper-
dV γ·V
= (13) ties like curvature and torsion which will be defined
dt V below.
And while comparing Eqs. (11) and (12) we
deduce that 3.2.2. Curvature of the trajectory curve
dV Let us consider the trajectory curve X(t) having
= γτ (14)
dt in M an instantaneous velocity vector V(t) and
Taking account of Eq. (10) and using the def- an instantaneous acceleration vector γ(t), the cur-
initions of the scalar and vector products, the vature, which expresses the rate of changes of the
expressions of the tangential (11) and normal (12) tangent to the trajectory curve, defined by:
components of the instantaneous acceleration vec-
1 γ ∧ V γν
tor γ can be finally written: = = (17)
 V3 V2
γ·V dV 
γτ = = = γcos(γ, V) (15) where  represents the radius of curvature.
V dt
Note: The location of the points where the local
γ ∧ V 
γν = = γ|sin(γ, V)| (16) curvature of the trajectory curve is null represents
V the location of the points of analytical inflexion, i.e.
Note: While using the Lagrange identity: the location of the points where the normal compo-
γ ∧ V2 + (γ · V)2 = γ2 · V2 , one finds easily nent of the instantaneous acceleration vector γ(t)
the norm of the instantaneous acceleration vector vanishes.
γ(t).
3.2.3. Torsion of the trajectory curve
2 γ ∧ V2 (γ · V)2
γ = γτ2 + γν2 = + Let us consider the trajectory curve X(t) hav-
V2 V2
ing in M an instantaneous velocity vector V(t),
γ ∧ V2 + (γ · V)2 an instantaneous acceleration vector γ(t), and an
= = γ2
V2 instantaneous over-acceleration vector γ̇, the tor-
sion, which expresses the difference between the
3.2. Trajectory curve properties trajectory curve and a plane curve, defined by:
In this approach the use of Differential Geometry 1 γ̇ · (γ ∧ V)
=− (18)
will allow a study of the metric properties of the γ ∧ V2
trajectory curve, i.e. curvature and torsion whose
definitions are recalled in this section. One will where represents the radius of torsion.
find, for example, in [Delachet, 1964; Struik, 1934; Note: A trajectory curve whose local torsion is
Kreyzig, 1959] or [Gray, 2006] a presentation of null is a curve whose osculating plane is stationary.
these concepts. In this case, the trajectory curve is a plane curve.
Differential Geometry and Mechanics: Applications to Chaotic Dynamical Systems 891

3.3. Application of these properties This shows that the instantaneous velocity and
to the determination of the acceleration vectors are collinear, which results in:
slow manifold analytical γ∧V =0 
equation
3.3.2. Slow manifold equation of a
In this section it will be demonstrated that the use
three-dimensional dynamical system
of the local metric properties of curvature and tor-
sion, resulting from Differential Geometry, provide Proposition 3.2. The location of the points where
the analytical equation of the slow manifold of a the local torsion of the trajectory curves integral of a
(S-FADS) or a (CAS-FADS) of dimension two or three-dimensional dynamical system defined by (1)
three. Moreover, it will be established that the slow or (2) is null, provides the slow manifold analytical
manifold analytical equation thus obtained is com- equation associated to this system.
pletely identical to that provided by the tangent Analytical Proof of Proposition 3.2. The vanishing
linear system approximation method presented in condition of the torsion provides:
the appendix.
1 γ̇ · (γ ∧ V)
=− = 0 ⇔ γ̇ · (γ ∧ V) = 0 (20)
γ ∧ V2
3.3.1. Slow manifold equation of a
two-dimensional dynamical system The first corollary inherent in the tangent linear
system approximation method implies to suppose
Proposition 3.1. The location of the points where
that the functional jacobian matrix is stationary.
the local curvature of the trajectory curves integral
That is to say
of a two-dimensional dynamical system defined by
(1) or (2) is null, provides the slow manifold ana- dJ
=0
lytical equation associated to this system. dt
Analytical Proof of Proposition 3.1. The vanishing Derivative of the expression (6) provides:
condition of the curvature provides: dV dJ dJ
γ̇ = J + V = Jγ + V
1 γ ∧ V dt dt dt
= =0⇔γ∧V =0 (19)
 V3 dJ
= J 2V + V ≈ J 2V
By using the expression (6), the coordinates of dt
the acceleration vector are written: The equation above is written as:
   
ẍ aẋ + bẏ (J 2 V) · (γ ∧ V) = 0
γ =
ÿ cẋ + dẏ
By developing this equation one finds in the
The equation above is written: long term Eq. (A.34) obtained by the tangent linear
cẋ2 − (a − d)ẋẏ − bẏ 2 = 0 system approximation method. The two equations
are thus absolutely identical. 
This equation is absolutely identical to Eq.
Geometrical Proof of Proposition 3.2. The tangent
(A.27) obtained by the tangent linear system
linear system approximation makes it possible to
approximation method. 
write that:
Geometrical Proof of Proposition 3.1. The vanishing V = αYλ1 + βYλ2 + δYλ3 ≈ βYλ2 + δYλ3
condition of the curvature provides:
While replacing in the expression (6) we obtain:
1 γ ∧ V
= =0⇔γ∧V =0 γ = JV = J(βYλ2 + δYλ3 ) = βλ2 Yλ2 + δλ3 Yλ3
 V3
According to what precedes, the over-
The tangent linear system approximation acceleration vector is written as:
makes it possible to write that:
γ̇ ≈ J 2 V
V = αYλ1 + βYλ2 ≈ βYλ2
While replacing the velocity by its expression
While replacing in the expression (6) we obtain: we have:
γ = JV = J(βYλ2 ) = βλ2 Yλ2 = λ2 V γ̇ ≈ J 2 (βYλ2 + δYλ3 ) = βλ22 Yλ2 + δλ23 Yλ3
892 J.-M. Ginoux & B. Rossetto

Thus it is noticed that: • If it is null, the trajectory curve is tangent to this


manifold.
V = βYλ2 + δYλ3 • If it is negative, the manifold is said repulsive.
γ = βλ2 Yλ2 + δλ3 Yλ3
This scalar product which represents the total
γ̇ = βλ22 Yλ2 + δλ23 Yλ3 derivative of φ constitutes a new manifold (V̇) which
This demonstrates that the instantaneous is the envelope of the manifold (V).
velocity, acceleration and over-acceleration vectors
Proof. Let us consider a manifold (V) defined by
are coplanar, which results in:
the implicit equation φ(x, y, z) = 0.
γ̇ · (γ ∧ V) = 0 The normal vector directed towards the outside
of the concavity of the curvature of this manifold is
This equation represents the location of the written as:
 
points where torsion is null. The identity of the two ∂φ
methods is thus established.   ∂x 
 
 
Note: Main results of this study are summarized  ∂φ 
η = ∇φ =    (21)
in Table 1 presented below. Abbreviations mean: 
 ∂y 
 
— T.L.S.A.: Tangent Linear System Approxima-  ∂φ 
tion ∂z
— A.D.G.F.: Application of the Differential Geom- The instantaneous velocity vector of the trajec-
etry Formalism tor y curve is defined by (1):
 
In the first report entitled the “Courbes définies dx
par une équation différentielle” Henri Poincaré  dt 
 
[1881] proposed a criterion making it possible to  
 dy 
characterize the attractivity or the repulsivity of V=  

a manifold. This criterion is recalled in the next  dt 
 
section.  dz 
dt
3.3.3. Attractive, repulsive manifolds The scalar product between these two vectors
is written as:
Proposition 3.3. Let X(t) be a trajectory curve
∂φ dx ∂φ dy ∂φ dz
having in M an instantaneous velocity vector V(t) V · ∇φ = + + (22)
and let (V) be a manifold (a curve in dimension two, ∂x dt ∂y dt ∂z dt
a surface in dimension three) defined by the implicit By noticing that Eq. (22) represents the total
equation φ = 0 whose normal vector η = ∇φ is derivative of φ, the envelope theory makes it possi-
directed towards the outside of the concavity of this ble to state that the new manifold (V̇) defined by
manifold. this total derivative constitutes the envelope of the
manifold (V) defined by the equation φ = 0. The
• If the scalar product between the instantaneous demonstration in dimension two of this Proposition
velocity vector V(t) and the normal vector η = results from what precedes. 
∇φ is positive, the manifold is said attractive
with respect to this trajectory curve
3.3.4. Slow, fast domains
Table 1. Determination of the slow manifold analytical In the Mechanics formalism, the study of nature
equation. of motion of a mobile M consists in being inter-
T.L.S.A. A.D.G.F.
ested in the variation of the Euclidean norm of its
instantaneous velocity vector V, i.e. in the tangen-
1 γ ∧ V tial component γτ of its instantaneous acceleration
Dimension 2 V ∧ Yλ2 = 0 = V3
=0
 vector γ. The variation of the Euclidean norm of the
Dimension 3 V · (Yλ2 ∧ Yλ3 ) = 0
1 γ̇·(γ ∧ V)
= − γ ∧ V2 = 0 instantaneous velocity vector V depends on the sign

of the scalar product between the instantaneous
Differential Geometry and Mechanics: Applications to Chaotic Dynamical Systems 893

velocity vector V and the instantaneous accelera- in ε, i.e. the singular approximation, of the slow
tion vector γ, i.e. the angle formed by these two manifold equation associated in a (S-FADS) com-
vectors. Thus if, γ · V > 0, the variation of the prising a small multiplicative parameter ε in one
Euclidean norm of the instantaneous velocity vector of the components of its velocity vector field V. In
V is positive and the Euclidean norm of the instan- this section, it will be demonstrated that the sin-
taneous velocity vector V increases. The motion is gular approximation associated in the acceleration
accelerated, it is in its fast phase. If γ · V = 0, vector field γ constitutes the first-order approxima-
the variation of the Euclidean norm of the instan- tion in ε of the slow manifold equation associated
taneous velocity vector V is null and the Euclidean with a (S-FADS) comprising a small multiplicative
norm of the instantaneous velocity vector V is con- parameter ε in one of the components of its velocity
stant. The motion is uniform, it is in a phase of vector field V and consequently a small multiplica-
transition between its fast and slow phases. More- tive parameter ε2 in one of the components of its
over, the instantaneous velocity vector V is perpen- acceleration vector field γ.
dicular to the instantaneous acceleration vector γ.
Proposition 4.1. The manifold equation associated
If, γ · V < 0, the variation of the Euclidean norm
in the singular approximation of the instantaneous
of the instantaneous velocity vector V is negative
acceleration vector γ of a (S-FADS ) constitutes the
and the Euclidean norm of the instantaneous veloc-
first-order approximation in ε of the slow manifold
ity vector V decreases. The motion is decelerated.
equation.
It is in its slow phase.
Definition 3.1. The domain of the phase space in Proof of Proposition 4.1 for Two-Dimensional
which the tangential component γτ of the instan- (S-FADS). In dimension two, Proposition 3.1
taneous acceleration vector γ is negative, i.e. the results in a collinearity condition (19) between the
domain in which the system is decelerating is called instantaneous velocity vector V and the instanta-
slow domain. neous acceleration vector γ. While posing:
The domain of the phase space in which the dx dy
tangential component γτ of the instantaneous accel- = ẋ and = ẏ = g
dt dt
eration vector γ is positive, i.e. the domain in which
the system is accelerating is called fast domain. The slow manifold equation of a (S-FADS) is
written as:
Note: On the one hand, if the (S-FADS) studied      
∂g 2 1 ∂f ∂g 1 ∂f
comprises only one small multiplicative parameter ẋ − g − ẋ − g2 = 0
ε in one of the components of its velocity vectors ∂x ε ∂x ∂y ε ∂y
field, these two domains are complementary. The (23)
location of the points belonging to the domain of This quadratic equation in ẋ has the following
the phase space where the tangential component γτ discriminant:
of the instantaneous acceleration vector γ is can-     
2 1 ∂f ∂g 2 ∂g 1 ∂f
celled, delimits the boundary between the slow and ∆=g − +4 g2 = 0
fast domains. On the other hand, the slow manifold ε ∂x ∂y ∂x ε ∂y
of a (S-FADS) or a (CAS-FADS) necessarily belongs The Taylor series of its square root up to terms
to the slow domain. of order 1 in ε is written as:


4. Singular Manifolds    

   
√ 1 ∂f ε ∂g ∂f
∆≈ g 1 +  2 2
The use of Mechanics made it possible to introduce ε ∂x   ∂f ∂x ∂y
a new manifold called singular approximation of the 

∂x
acceleration which provides an approximate equa- 


tion of the slow manifold of a (S-FADS).    

∂g ∂f 2
− + O(ε )
4.1. Singular approximation of ∂y ∂x 



the acceleration
The singular perturbations theory [Andronov et al., Taking into account what precedes, the solution
1966] have provided the zero order approximation of Eq. (23) is written as:
894 J.-M. Ginoux & B. Rossetto

 
∂f dx d2 x
∂y 3
d z dt dt2
ẋ ≈ −g   + O(ε) (24) + 3 =0 (26)
∂f dt dy d2 y
∂x dt dt2
This equation represents the second-order
In order to simplify, let us replace the three
approximation in ε of the slow manifold equa-
determinants by:
tion associated with the singular approximation.
According to Eq. (6), the instantaneous accelera- dy d2 y d2 x dx
tion vector γ is written as:
 2    dt dt2 dt2 dt
d x ∂f dx ∂f dy ∆1 = ; ∆2 = ;
 ε dt2   ∂x dt + ∂y dt  dz d2 z d2 z dz
    dt dt2 dt2
γ= =  dt
 d2 y   ∂g dx ∂g dy 
+
dt2 ∂x dt ∂y dt dx d2 x
The singular approximation of the acceleration dt dt2
∆3 =
provides the equation: dy d2 y
∂f dx ∂f dy dt dt2
+ =0
∂x dt ∂y dt Equation (26) then will be written as:
While posing: ... ... ...
(x)∆1 + ( y )∆2 + ( z )∆3 = 0 (27)
dx dy
= ẋ and = ẏ = g While posing:
dt dt
We obtain: dx dy dz
  dt
= ẋ,
dt
= ẏ = g,
dt
= ż = h
∂f
∂y By dividing Eq. (27) by (z̈), we have
ẋ = −g   (25)
∂f
1 ... ...
∂x (ẋÿ − ẍẏ) + [(x)∆1 + ( y )∆2 ] = 0
... (28)
(z )
By comparing this expression with Eq. (24)
which constitutes the second-order approximation First term of Eq. (28) is written as:
in ε of the slow manifold equation, we deduce from    
Eq. (25) that it represents the first-order approxi- ∂g 2 1 ∂f ∂g ∂g h
(ẋÿ − ẍẏ) = ẋ − g − − ẋ
mation in ε of the slow manifold equation. It has ∂x ε ∂x ∂y ∂z g
 
also been demonstrated that the singular approxi- g2 ∂f ∂f h
mation of the acceleration constitutes the first of the − + (29)
ε ∂y ∂z g
successive approximations developed in [Rossetto,
1986].  For homogeneity reasons let us pose:
Proof of Proposition 4.1 for Three-Dimensional (S- 1 ... ...
FADS). In dimension three, Proposition 3.2 results g2 G = ... [(x)∆1 + ( y )∆2 ]
(z )
in a coplanarity condition (20) between the instan-
taneous velocity vector V, the instantaneous accel- Equation (28) is written:
eration vector γ and the instantaneous over-    
acceleration vector γ̇. The slow manifold equation ∂g 2 1 ∂f ∂g ∂g h
ẋ − g − − ẋ
of a (S-FADS) is written as: ∂x ε ∂x ∂y ∂z g
 
d2 y d2 x g2 ∂f ∂f h
dy dx − + + g2 G = 0 (30)
d3 x dt dt2 d3 y dt2 dt ε ∂y ∂z g
+
dt3 dz d2 z dt3 d2 z dz This quadratic equation in ẋ has the following
dt dt2 dt2 dt discriminant :
Differential Geometry and Mechanics: Applications to Chaotic Dynamical Systems 895

  
1 ∂f
2 ∂g ∂g h 2 By comparing this expression with Eq. (31)
∆=g − + which constitutes the second-order approximation
ε ∂x ∂y ∂z g
     in ε of the slow manifold equation, we deduce from
2 ∂g 1 ∂f ∂f h Eq. (32) that it represents the first-order approxi-
+ 4g + −G
∂x ε ∂y ∂z g mation in ε of the slow manifold equation.
The Taylor series of its square root up to terms It has also been demonstrated that the singular
of order 1 in ε is written as: approximation of the acceleration constitutes the
 first of the successive approximations developed in

   
    [Rossetto, 1986]. 
√ 1 ∂f ε ∂g ∂f
∆≈ g 1 +  2 2
ε ∂x   ∂f ∂x ∂y

 The use of the criterion proposed by H.
∂x Poincaré (Proposition 3) made it possible to char-

 acterize the attractivity of the slow manifold of a
    

∂f h ∂g ∂g h ∂f  (S-FADS) or a (CAS-FADS). Moreover, the pres-
2
+ − + + O(ε ) ence in the phase space of an attractive slow man-
∂z g ∂y ∂z g ∂x 


 ifold, in the vicinity of which the trajectory curves
converge, constitutes a part of the attractor.
Taking into account what precedes, the solution The singular manifold presented in the next
of Eq. (30) is written as: section proposes a description of the geometrical
  structure of the attractor.
∂f ∂f h
+
∂y ∂z g
ẋ ≈ −g   + O(ε) (31) 4.2. Singular manifold
∂f
∂x The denomination of singular manifold comes from
the fact that this manifold plays the same role with
This equation represents the second-order
respect to the attractor as a singular point with
approximation in ε of the slow manifold equa-
respect to the trajectory curve.
tion associated with the singular approximation.
According to Eq. (6), the instantaneous accelera-
tion vector γ is written as: Proposition 4.2. The singular manifold is defined
    by the intersection of slow manifold of equation
d2 x ∂f dx ∂f dy ∂f dz φ = 0 and an unspecified Poincaré section (Σ) made
 ε 2   ∂x dt + ∂y dt + ∂z dt 
 dt    in its vicinity. Thus, it represents the location of the
   
 d y   ∂g dx ∂g dy ∂g dz 
2 points satisfying:
γ=    + + 
= ∂z dt  φ∩Σ=0
 dt2   ∂x dt ∂y dt  (33)
   
 d2 z   ∂h dx ∂h dy ∂h dz 
+ + This manifold of codimension one is a subman-
dt2 ∂x dt ∂y dt ∂z dt
ifold of the slow manifold.
The singular approximation of the acceleration
In dimension two, the singular manifold is
provides the equation:
reduced to a point.
∂f dx ∂f dy ∂f dz In dimension three, it is a “line” or more exactly
+ + =0 a “curve”.
∂x dt ∂y dt ∂z dt
The location of the points obtained by integra-
While posing: tion in a given time of initial conditions taken on
dx dy dz this manifold constitutes a submanifold also belong-
= ẋ, = ẏ = g and = ż = h
dt dt dt ing to the attractor generated by the dynamical sys-
tem. The whole of these manifolds corresponding
We obtain:
  to various points of integration makes it possible to
∂f ∂f h reconstitute the attractor by deployment of these
+
∂y ∂z g singular manifolds.
ẋ = −g   (32)
∂f The concept of deployment will be illustrated
∂x in Sec. 5.4.
896 J.-M. Ginoux & B. Rossetto

5. Applications We obtain the following implicit equation:


1
5.1. Van der Pol model [9y 2 + (9x + 3x3 )y + 6x4 − 2x6 + 9x2 ε] = 0
9ε2
The oscillator of B. Van der Pol [1926] is a second-
(36)
order system with nonlinear frictions which can be
written as: Since this equation is quadratic in y, we can
solve it in order to plot y according to x.
ẍ + α(x2 − 1)ẋ + x = 0
x3 x x  4
y1,2 = − − ± x − 2x2 + 1 − 4ε (37)
The particular form of the friction which can be 6 2 2
carried out by an electric circuit causes a decrease Figure 1 shows the plot of the slow manifold
of the amplitude of the great oscillations and an equation (37) of the Van der Pol system with ε =
increase of the small. There are various ways of writ- 0.05 by using Proposition 3.1, i.e. the collinear-
ing the previous equation like a first-order system. ity condition between the instantaneous velocity
One of them is: vector V and the instantaneous acceleration vec-
  
 x3 tor γ, i.e. the location of the points where the

ẋ = α x + y − 3 curvature of the trajectory curves is cancelled.
 Moreover, Definition 1 makes it possible to delimit

ẏ = − x the area of the phase plane in which, the scalar
α product between the instantaneous velocity vector
When α becomes very large, x becomes a “fast” V and the instantaneous acceleration vector γ is
variable and y a “slow” variable. In order to analyze negative, i.e. where the tangential component γτ
the limit α → ∞, we introduce a small √ parameter of its instantaneous acceleration vector γ is nega-
ε = 1/α2 and a “slow time” t = t/α = εt. Thus, tive. We can thus graphically distinguish the slow
the system can be written as: domain of the fast domain (in blue), i.e. the domain
  of stability of the trajectories.
dx  
 ε dt    x3 The blue part of Fig. 1 corresponds to the
  f (x, y) x + y − 3  domain where the variation of the Euclidean norm
V = =  (34)
 dy  g(x, y) of the instantaneous velocity vector V is posi-
−x tive, i.e. where the tangential component of the
dt
with ε a positive real parameter
ε = 0.05 1

where the functions f and g are infinitely differen-


tiable with respect to all xi and t, i.e. C ∞ functions
in a compact E are included in R2 and with val- 0.5
ues in R. Moreover, the presence of a small multi-
plicative parameter ε in one of the components of
its velocity vector field V ensures that the system
(34) is a (S-FADS). We can thus apply the method
-2 -1 1 2
described in Sec. 3, i.e. Differential Geometry. The
instantaneous acceleration vector γ is written as:
    
d2 x 1 dx dy 2 dx
 2  d  ε dt + dt − x dt 
-0.5
 dt   
γ =   (35)
 d2 y  dt  dx 

dt2 dt
-1
Proposition 3.1 leads to:
1 γ ∧ V
= = 0 ⇔ γ ∧ V = 0 ⇔ ẍẏ − ẋÿ = 0 Fig. 1. Slow manifold (in blue) and stable domain of the
 V3 Van der Pol system (in white).
Differential Geometry and Mechanics: Applications to Chaotic Dynamical Systems 897

Y Y
3

0.75

0.5

0.25
X
-3 -2 -1 1 2 3

X
1.2 1.4 1.6 1.8
-1

-0.25

-2

-0.5

-3

-0.75 Fig. 3. Singular approximation of the acceleration of the B.


Van der Pol system (in magenta).

Fig. 2. Zoom of the stable part of the B. Van der Pol system. Y
3

instantaneous acceleration vector γ, is positive. Let


us take note that, as soon as the trajectory curve,
2
initially outside this domain, enters inside, it leaves
the slow manifold to reach the fast foliation.
The slow manifold equation provided by Propo- 1
sition 4.1 leads to the following implicit equation:
1
[3x − 4x3 + x5 + (3 − 3x2 )y − 3xε] = 0 (38)
3ε2 -3 -2 -1 1 2 3
X

Starting from this equation we can plot y


according to x: -1

x5 − 4x3 + 3x(1 − ε)
y= (39)
3(−1 + x2 ) -2
Figure 3 shows the plot of the slow mani-
fold equation (39) of the Van der Pol system with
ε = 0.05 by using Proposition 4.1, i.e. the singular -3

approximation of the instantaneous acceleration Fig. 4. Isoclines of the acceleration vector of the B. Van der
vector γ in magenta. Blue curve represents the slow Pol system.
manifold equation (37) provided by Proposition 3.1.
In order to illustrate the principle of the method
presented above, we have plotted in Fig. 4 the iso- which tend to the slow manifold defined by Propo-
clines of acceleration for various values: 0.5, 0.2, 0.1, sition 3.1 are plotted.
0.05.
The very large variation rate of the acceleration 5.2. Chua model
in the vicinity of the slow manifold can be noticed The Chua circuit [1986] is a relaxation oscillator
in Fig. 4. Some isoclines of the acceleration vector with a cubic nonlinear characteristic elaborated
898 J.-M. Ginoux & B. Rossetto

from a circuit comprising a harmonic oscillator for with values in R. Moreover, the presence of a small
which the operation is based on a field-effect tran- multiplicative parameter ε in one of the compo-
sistor, coupled to a relaxation-oscillator composed nents of its instantaneous velocity vector V ensures
of a tunnel diode. The modeling of the circuit uses that the system (40) is a (S-FADS). We can thus
a capacity which will prevent abrupt voltage drops apply the method described in Sec. 3, i.e. Differen-
and makes it possible to describe the fast motion tial Geometry. In dimension three, the slow mani-
of this oscillator by the following equations which fold equation is provided by Proposition 3.2, i.e. the
constitute a slow-fast system. vanishing condition of the torsion:
 
dx   
 dt  1 γ̇ · (γ ∧ V)
  1 44 3 41 2 =− = 0 ⇔ γ̇ · (γ ∧ V) = 0
   z − x − x − µx  γ ∧ V2
 dy   ε 3 2 
V= =
 dt   −z


    Within the framework of the tangent linear sys-
  tem approximation, Corollary 1 leads to Eq. (34).
 dz  −0.7x + y + 0.24z
Using Mathematica Fig. 5 shows a plot of
dt the phase portrait of Chua model and its slow
(40) manifold.
with ε and µ are real parameters Without the framework of the tangent lin-
ear system approximation, i.e. considering that
ε = 0.01
the functional jacobian varies with time, Propo-
µ = 6.94 sition 3.2 provides a surface equation which
where the functions f , g and h are infinitely dif- represents the location of points where torsion is
ferentiable with respect to all xi , and t, i.e. C ∞ cancelled, i.e. the location of points where the
functions in a compact E are included in R3 and osculating plane is stationary and where the slow

Fig. 5. Slow manifold of Chua model.


Differential Geometry and Mechanics: Applications to Chaotic Dynamical Systems 899

Fig. 6. Attractive part of the slow manifold of Chua model.

Fig. 7. Singular approximation of the acceleration of Chua model.


900 J.-M. Ginoux & B. Rossetto

manifold is attractive. Thus, the attractive part the impredictible behavior of weather. After hav-
of the slow manifold of Chua model is plotted ing developed nonlinear partial derivative equations
in Fig. 6. starting from the thermal equation and Navier–
We deduce, according to Proposition 3.3, that Stokes equations, Lorenz truncated them to retain
the location of the points where the torsion is only three modes. The most widespread form of the
negative corresponds to the attractive parts of Lorenz model is as follows:
the slow manifold. Thus, the attractive part of  
dx
the slow manifold of the Chua model is plotted  dt   
 
in Fig. 6.   σ(y − x)
 dy   
Slow manifold equation provided by Proposi- V= 
 dt  =  −xz + rx − y  (41)
tion 4.1 leads to the following implicit equation:   xy − βz
 
1  dz 
(5043x3 + 9020x4 + 3872x5 − 246xz
6ε2 dt
− 264x2 z − 4.2xε + 6yε + 1.44zε with σ, r and β are real parameters: σ = 10,
+ 369x2 µ + 352x3 µ − 6zµ + 6xµ2 ) = 0 β = 8/3, r = 28, where the functions f , g and h are
infinitely differentiable with respect to all xi and t,
The surface plotted in Fig. 7 constitutes a quite i.e. C ∞ functions in a compact E are included in
good approximation of the slow manifold of this R3 and with values in R. Although this model has
model. no singular approximation, it can be considered as a
(S-FADS), according to Sec. 1.3, because it has been
numerically checked [Rossetto et al., 1998] that its
6. Lorenz Model functional jacobian matrix possesses at least a large
The purpose of the model established by Edward and negative real eigenvalue in a large domain of
Lorenz [1963] was in the beginning to analyze the phase space. Thus, we can apply the method

Fig. 8. Slow manifold of Lorenz model.


Differential Geometry and Mechanics: Applications to Chaotic Dynamical Systems 901

Fig. 9. Attractive part of the slow manifold of Lorenz model.

described in Sec. 3, i.e. Differential Geometry. In slow manifold. Thus the attractive part of the slow
dimension three, the slow manifold equation is pro- manifold of the Lorenz model is plotted in Fig. 9.
vided by Proposition 3.2, i.e. the vanishing condi-
tion of the torsion:
7. Volterra–Gause Model
1 γ̇ · (γ ∧ V) Let us consider the model elaborated by Ginoux
=− = 0 ⇔ γ̇ · (γ ∧ V) = 0
γ ∧ V2 et al. [2005] for three species interacting in a
predator–prey mode.
Within the framework of the tangent linear system  
dx  
approximation, Corollary 1 leads to Eq. (34). Using  dt  1 1
 
Mathematica Fig. 8 shows the plot of the phase   (x(1 − x) − x y) 
2

portrait of Lorenz model and its slow manifold.  dy   ξ 
V=   
 dt  =  −δ1 y + x 12 y − y 12 z  (42)
Without the framework of the tangent lin-    
ear system approximation, i.e. considering that   1
εz(y 2 − δ2 )
 dz 
the functional jacobian varies with time, Proposi-
tion 3.2 provides a surface equation which repre- dt
sents the location of the points where torsion is with ξ, ε, δ1 and δ2 are real parameters: ξ = 0.866,
cancelled, i.e. the location of the points where the ε = 1.428, δ1 = 0.577, δ2 = 0.376.
osculating plane is stationary and where the slow And where the functions f, g and h are infinitely
manifold is attractive. Thus, the attractive part differentiable with respect to all xi , and t, i.e. C ∞
of the slow manifold of Lorenz model is plotted functions in a compact E are included in R3 and
in Fig. 9. with values in R.
We deduce, according to Proposition 3.3, that This model consisting of a prey, a preda-
the location of the points where the torsion is tor and top-predator has been named Volterra–
negative corresponds to the attractive parts of the Gause because it combines the original model of
902 J.-M. Ginoux & B. Rossetto

Fig. 10. Deployment of the singular manifold of the Volterra–Gause model.

V. Volterra [1926] incorporating a logisitic limita- the evolution of the trajectory curves on the surface
tion of P. F. Verhulst [1838] type on the growth generated by this attractor. Indeed, the location of
of the prey and a limitation of G. F. Gause [1935] the points of intersection of the slow manifold with
type on the intensity of the predations of the preda- a Poincaré section carried out in its vicinity con-
tor on the prey and top-predator on the preda- stitutes a “line” or more exactly a “curve”. Then
tor. The equations (42) are dimensionless, remarks by using numerical integration this “curve” (resp.
and details about the changes of variables and “line”) is deployed through the phase space and its
the parameters have been extensively presented in deployment reconstitutes the attractor shape. The
[Ginoux et al., 2005]. Moreover, the presence of result is plotted in Fig. 10.
a small multiplicative parameter ξ in one of the
components of its instantaneous velocity vector V 8. Discussion
ensures that the system (42) is a (S-FADS). So, the
Considering the trajectory curves integral of
method described in Sec. 3, i.e. Differential Geome-
dynamical systems as plane or space curves evolv-
try would have provided the slow manifold equa-
ing in the phase space, it has been demonstrated
tion thanks to Proposition 3.2. But, this model
in this work that the local metric properties of
exhibits a chaotic attractor in the snail shell shape
curvature and torsion of these trajectory curves
and the use of the algorithm developed by Wolf
make it possible to directly provide the analytical
et al. [1985] have made it possible to compute
equation of the slow manifold of dynamical systems
what can be regarded as its Lyapunov exponents:
(S-FADS or CAS-FADS) according to kinematics
(+0.035, 0.000, −0.628). Then, the Kaplan–Yorke
variables. The slow manifold analytical equation is
[1983] conjecture provided the following Lyapunov
thus given by the following:
dimension: 2.06. So, the fractal dimension of this
chaotic attractor is close to that of a surface. The • vanishing condition of the curvature in dimension
singular manifold makes it possible to account for two,
Differential Geometry and Mechanics: Applications to Chaotic Dynamical Systems 903

• vanishing condition of the torsion in dimension Chua, L. O., Komuro, M. & Matsumoto, T. [1986] “The
three. double scroll family,” IEEE Trans. Circuits Syst. 33,
1072–1097.
Thus, the use of Differential Geometry concepts Coddington, E. A. & Levinson, N. [1955] Theory
has made it possible for the analytical equation of of Ordinary Differential Equations (MacGraw Hill,
the slow manifold to be completely independent of NY).
the “slow” eigenvectors of the functional jacobian Delachet, A. [1964] La Géométrie Différentielle, “Que
of the tangent linear system, and it was demon- sais-je”, n◦ 1104 (PUF, Paris).
strated that the equation thus obtained is com- Frenet, F. [1852] “Sur les courbes à double cour-
pletely identical to that providing the tangent bure,” Thèse Toulouse, 1847. Résumé dans J. de
linear system approximation method [Rossetto Math., 17.
et al., 1998] presented below in the appendix. So Gause, G. F. [1935] The Struggle for Existence (Williams
and Wilkins, Baltimore).
characterization of its attractivity is possible while
Ginoux, J. M., Rossetto, B. & Jamet, J. L. [2005]
using a criterion proposed by Henri Poincaré [1881] “Chaos in a three-dimensional Volterra–Gause model
in his report entitled “Sur les courbes définies par of predator–prey type,” Int. J. Bifurcation and Chaos
une equation différentielle”. 5, 1689–1708.
Moreover, the specific use of the instantaneous Gray, A., Salamon, S. & Abbena, E. [2006] Modern
acceleration vector, inherent in Mechanics, allows Differential Geometry of Curves and Surfaces with
on the one hand a kinematics interpretation of Mathematica (Chapman & Hall/CRC).
the evolution of the trajectory curves in the vicin- Kaplan, J. & Yorke, J. A. [1979] “Chaotic behavior
ity of the slow manifold by defining the slow and of multidimensional difference equations, in func-
fast domains of the phase space and on the other tional differential equations and approximation of
hand, to approach the analytical equation of the fixed points,” Lecture Notes in Mathematics 730,
slow manifold thanks to the singular approxima- 204–227.
Kreyszig, E. [1991] Differential Geometry (Dover, NY).
tion of acceleration. The equation thus obtained
Lorenz, E. N. [1963] “Deterministic non-periodic flows,”
is completely identical to that which provides the J. Atmos. Sci. 20, 130–141.
successive approximations method [Rossetto, 1986]. Poincaré, H. [1881] “Sur les courbes définies par une
Thus, it has been established that the presence in équation différentielle,” J. Math. Pures et Appl.,
the phase space of an attractive slow manifold, in Série III 7, 375–422.
the vicinity of which the trajectory curves converge, Poincaré, H. [1882] “Sur les courbes définies par une
defines part of the attractor. So, in order to propose équation différentielle,” J. Math. Pures et Appl.,
a qualitative description of the geometrical struc- Série III 8, 251–296.
ture of attractor a new manifold called singular has Poincaré, H. [1885] “Sur les courbes définies par une
been introduced. équation différentielle,” J. Math. Pures et Appl.,
Various applications to the models of Van der Série IV 1, 167–244.
Pol, cubic-Chua, Lorenz and Volterra–Gause have Poincaré, H. [1886] “Sur les courbes définies par une
équation différentielle,” J. Math. Pures et Appl.,
made it possible to illustrate the practical inter-
Série IV 2, 151–217.
est of this new approach for the dynamical systems Rossetto, B. [1986] “Singular approximation of chaotic
(S-FADS or CAS-FADS) study. slow-fast dynamical systems,” Lecture Notes in
Physics 278, 12–14.
Acknowledgments Rossetto, B., Lenzini, T., Ramdani, S. & Suchey, G.
[1998] “Slow-fast autonomous dynamical systems,”
The authors would like to thank Dr. C. Gérini for Int. J. Bifurcation and Chaos 8, 2135–2145.
his bibliographical and mathematical remarks and Struik, D. J. [1988] Lecture on Classical Differential
comments. Geometry (Dover, NY).
Tihonov, A. N. [1952] “Systems of differential equations
containing small parameters in the derivatives,” Mat.
References Sbornik N.S. 31, 575–586.
Andronov, A. A., Khaikin, S. E. & Vitt, A. A. [1966] Van der Pol, B. [1926] “On ‘relaxation-oscillations’,”
Theory of Oscillators (Pergamon Press, Oxford). Phil. Mag. 7, 978–992.
Chlouverakis, K. E. & Sprott, J. C. [2004] “A compari- Verhulst, P. F. [1838] “Notice sur la loi que suit la popu-
son of correlation and Lyapunov dimensions,” Physica lation dans son accroissement,” Corresp. Math. Phys.
D 200, 156–164. X, 113–121.
904 J.-M. Ginoux & B. Rossetto

Volterra, V. [1926] “Variazioni e fluttuazioni del numero velocity vector field (X) of this system is
d’individui in specie animali conviventi,” Mem. Acad. overshadowed by the fast dynamics of the lin-
Lincei III 6, 31–113. ear part.
Volterra, V. [1931] Leçons sur la Théorie Mathématique
de la Lutte pour la Vie, Gauthier-Villars, Paris. d(X)
Wolf, A., Swift, J. B., Swinney, H. L. & Vastano, J. A. (X) = (X0 ) + (X − X0 )
dX X0
[1985] “Determining Lyapunov exponents from a time
series,” Physica D 16, 285–317. + O((X − X0 )2 ) (A.1)

Corollaries
Appendix To the dynamical system defined by (1) or (2)
Formalization of the Tangent Linear is associated a tangent linear system defined as
System Approximation Method follows:
dδX
The aim of this appendix is to demonstrate that the = J(X0 )δX (A.2)
approach developed in this work generalizes the tan- dt
gent linear system approximation method [Rossetto where
et al., 1998]. After having pointed out the neces- δX = X − X0 , X0 = X(t0 ) and
sary assumptions to the application of this method
d(X)
and the corollaries which result from this, two con- = J(X0 )
ditions (of collinearity/coplanarity and orthogonal- dX X0
ity) providing the analytical equation of the slow
manifold of a dynamical system defined by (1) or Corollary 1. The nonlinear part condition implies
(2) will be presented in a formal way. Equivalence the stability of the slow manifold. So, according to
between these two conditions will then be estab- Proposition 3.3, the velocity varies slowly on the
lished. Lastly, while using the sum and the prod- slow manifold. This involves that the functional
uct (also the square of the sum and the prod- jacobian J(X0 ) varies slowly with time, i.e.
uct) of the eigenvalues of the functional jacobian of dJ
the tangent linear system, the equation of the slow (X0 ) = 0 (A.3)
dt
manifold generated by these two conditions will be
made independent of these eigenvalues and will be The solution of the tangent linear system (A.2)
expressed according to the elements of the func- is written as:
tional jacobian matrix of the tangent linear system. δX = eJ(X0 )(t−t0 ) δX(t0 ) (A.4)
It will be thus demonstrated that this analytical
So,
equation of the slow manifold is completely iden- n
tical to that provided by the Propositions 3.1 and 
δX = ai Yλi (A.5)
3.2 developed in this article.
i=1
where n is the dimension of the eigenspace, ai rep-
resents coefficients depending explicitly on the coor-
Assumptions dinates of space and implicitly on time and Yλi the
The application of the tangent linear system eigenvectors associated in the functional jacobian of
approximation method requires that the dynamical the tangent linear system.
system defined by (1) or (2) satisfies the following
assumptions: Corollary 2. In the vicinity of the slow manifold
the velocities of the dynamical system defined by
(H1 ) The components fi , of the velocity vector (1) or (2) and that of the tangent linear system (4)
field (X) defined in E are continuous, C ∞ merge.
functions in E and with values included in R.
dδX
(H2 ) The dynamical system defined by (1) or = VT ≈ V (A.6)
(2) satisfies the nonlinear part condition dt
[Rossetto et al., 1998], i.e. the influence of where VT represents the velocity vector associated
the nonlinear part of the Taylor series of the in the tangent linear system.
Differential Geometry and Mechanics: Applications to Chaotic Dynamical Systems 905

The tangent linear system approximation functional jacobian of the tangent linear system are
method consists in spreading the velocity vector written as:
field V on the eigenbasis associated to the eigenval-  
∂g
ues of the functional jacobian of the tangent linear λ −
 i ∂y 
system. 
Yλi   (A.9)

Indeed, by taking account of (A.2) and (A.5) ∂g
we have according to (A.6): ∂x

dδX with
= J(X0 )δX i = 1, 2
dt
n
 The projection of the velocity vector field V on
= J(X0 ) ai Yλi the eigenbasis is written according to Corollary 2:
i=1
n  n
 dδX
= ai J(X0 )Yλi = VT ≈ V = ai λi Yλi
dt
i=1 i=1
n = αYλ1 + βYλ2
= ai λi Yλi (A.7)
i=1 where α and β represent coefficients depending
explicitly on the coordinates of space and implicitly
Thus, Corollary 2 provides: on time and where Yλ1 represents the “fast” eigen-
n
vector and Yλ2 the “slow” eigenvector. The exis-
dδX 
= VT ≈ V = ai λi Yλi (A.8) tence of an evanescent mode in the vicinity of the
dt slow manifold implies according to Tihonv’s theo-
i=1
rem [1952]: α  1. We deduce:
Equation (A.8) constitutes in dimension two
(resp. dimension three) a condition called collinear- Proposition A.1. A necessary and sufficient con-
ity (resp. coplanarity) condition which provides dition of obtaining the slow manifold equation of a
the analytical equation of the slow manifold of a two-dimensional dynamical system is that its veloc-
dynamical system defined by (1) or (2). ity vector field V is collinear to the slow eigenvector
An alternative proposed by Rossetto et al. Yλ2 associated to the slow eigenvalue λ2 of the func-
[1998] uses the “fast” eigenvector on the left asso- tional jacobian of the tangent linear system. That is
ciated to the “fast” eigenvalue of the transposed to say:
functional jacobian of the tangent linear system.
In this case the velocity vector field V is then V ≈ βYλ2 (A.10)
orthogonal with the “fast” eigenvector on the left. While using this collinearity condition, the
This constitutes a condition called orthogonality equation constituting the first-order approximation
condition which provides the analytical equation of in ε of the slow manifold of a two-dimensional
the slow manifold of a dynamical system defined by dynamical system is written as:
(1) or (2).
These two conditions will be the subject of V ∧ Yλ2 = 0
a detailed presentation in the following sections.

Thereafter it will be supposed that the assumptions      
(H1) and (H2) are always checked. ∂g dx ∂g dy
− λ2 − =0 (A.11)
∂x dt ∂y dt

Collinearity/coplanarity condition
Slow manifold equation of a
Slow manifold equation of a three-dimensional dynamical system
two-dimensional dynamical system Let us consider a dynamical system defined under
Let us consider a dynamical system defined the same conditions as (1) or (2). The eigenvec-
under the same conditions as (1) or (2). The tors associated to the eigenvalues of the func-
eigenvectors associated to the eigenvalues of the tional jacobian of the tangent linear system are
906 J.-M. Ginoux & B. Rossetto

written as: written as:


     
1 ∂f ∂g 1 ∂f ∂g ∂g
 + λi −   λi − ∂y 
 ε ∂y ∂z ε ∂z ∂y  t  
    Yλi   (A.15)
 1 ∂f ∂g ∂g 1 ∂f   1 ∂f 

Yλi  + λi − 

 ε ∂z ∂x ∂z ε ∂x  ε ∂y
   
 1 ∂f ∂g 1 ∂f ∂g  with
− + λi − λi −
ε ∂y ∂x ε ∂x ∂y i = 1, 2
(A.12) tY represents the “fast” eigenvector on the
λ1
with left associated to the dominant eigenvalue, i.e. the
largest eigenvalue in absolute value and t Yλ2 is
i = 1, 2, 3 the “slow” eigenvector on the left.
The projection of the velocity vector field V on But since according to Rossetto et al. [1998],
the eigenbasis is written according to Corollary 2: the velocity vector field V is perpendicular to the
 n “fast” eigenvector on the left t Yλ1 , we deduce:
dδX
= VT ≈ V = ai λi Yλi Proposition A.3. A necessary and sufficient con-
dt
i=1
dition of obtaining the slow manifold equation of
= αYλ1 + βYλ2 + δYλ3 a two-dimensional dynamical system is that its
where α, β and δ represent coefficients depending velocity vector field V is perpendicular to the fast
explicitly on the coordinates of space and implic- eigenvector t Yλ1 on the left associated to the fast
itly on time and where Yλ1 represents the “fast” eigenvalue λ1 of the transposed functional jacobian
eigenvector and Yλ2 , Yλ3 the “slow” eigenvectors. of the tangent linear system. That is to say:
The existence of an evanescent mode in the vicinity V⊥t Yλ1 (A.16)
of the slow manifold implies according to Tihonv’s
theorem [1952]: α  1. We deduce: While using this orthogonality condition, the
equation constituting the first-order approximation
Proposition A.2. A necessary and sufficient con- in ε of the slow manifold of a two-dimensional
dition of obtaining the slow manifold equation of dynamical system is written as:
a three-dimensional dynamical system is that its
velocity vector field V is coplanar to the slow eigen- V · t Yλ1 = 0
vectors Yλ2 and Yλ3 associated to the slow eigen- ⇔
values λ2 and λ3 of the functional jacobian of the      
∂g dx 1 ∂f dy
tangent linear system. That is to say: λ1 − + =0 (A.17)
∂y dt ε ∂y dt
V ≈ βYλ2 + δYλ3 (A.13)
While using this coplanarity condition, the Slow manifold equation of a
equation constituting the first-order approximation three-dimensional dynamical system
in ε of the slow manifold of a three-dimensional Let us consider a dynamical system defined under
dynamical system is written as: the same conditions as (1) or (2). The eigenvec-
det(V, Yλ2 , Yλ3 ) = 0 ⇔ V · (Yλ2 ∧ Yλ3 ) = 0 tors associated to the eigenvalues of the transposed
(A.14) functional jacobian of the tangent linear system are
written as:
   
∂g ∂h ∂h ∂g
Orthogonality condition  + λi − 
 ∂x ∂y ∂x ∂y 
Slow manifold equation of a    
 1 ∂f ∂h ∂h 1 ∂f 
two-dimensional dynamical system t 
Yλi  + λi − 
ε ∂y ∂x ∂y ε ∂x 
 
Let us consider a dynamical system defined under    
 1 ∂f ∂g 1 ∂f ∂g 
the same conditions as (1) or (2). The eigenvec- − + λi − λi −
tors associated to the eigenvalues of the transposed ε ∂y ∂x ε ∂x ∂y
functional jacobian of the tangent linear system are (A.18)
Differential Geometry and Mechanics: Applications to Chaotic Dynamical Systems 907

with While using the trace and determinant of the


i = 1, 2, 3 functional jacobian of the tangent linear system, we
have:
tY
λ1 represents the “fast” eigenvector on the  
t 1 ∂f ∂g ∂f ∂g
left associated to the dominant eigenvalue, i.e. the Yλ1 · Yλ2 = −
ε ∂x ∂y ∂y ∂x
largest eigenvalue in absolute value and t Yλ2 , t Yλ3   
are the “slow” eigenvectors on the left. ∂g 1 ∂f ∂g
− +
But since according to Rossetto et al. [1998], ∂y ε ∂x ∂y
the velocity vector field V is perpendicular to the  2   
∂g 1 ∂f ∂g
“fast” eigenvector on the left t Yλ1 , we deduce: + +
∂y ε ∂y ∂x
Proposition A.4. A necessary and sufficient con- =0
dition of obtaining the slow manifold equation of
a three-dimensional dynamical system is that its So,
velocity vector field V is perpendicular to the fast t
Yλ1 ⊥Yλ2 (A.21)
eigenvector t Yλ1 on the left associated in the fast
eigenvalue λ1 of the transposed functional jacobian Thus, collinearity and orthogonality conditions
of the tangent linear system. That is to say: are completely equivalent. 
Proof of Proposition 5 in Dimension Three. In dimen-
V⊥t Yλ1 (A.19)
sion three, the slow manifold equation may
While using this orthogonality condition, the be obtained while considering that the velocity
equation constituting the first-order approximation vector field V is:
in ε of the slow manifold of a three-dimensional
dynamical system is written as: • either coplanar to the “slow” eigenvectors Yλ2
and Yλ3
V⊥t Yλ1 ⇔ V · t Yλ1 = 0 (A.20) • either perpendicular to the “fast” eigenvector on
the left t Yλ1
Equivalence of both conditions There is equivalence between both conditions
Proposition A.5. Both necessary and sufficient provided that the “fast” eigenvector on the left t Yλ1
collinearity/coplanarity and orthogonality condi- is orthogonal to the plane containing the “slow”
tions providing the slow manifold equation are eigenvectors Yλ2 and Yλ3 . While using the coor-
equivalent. dinates of these eigenvectors defined in the above
section, the sum and the product (also the square
Proof of Proposition 5 in Dimension Two. In dimen- of the sum and the product) of the eigenvalues of
sion two, the slow manifold equation may be the functional jacobian of the tangent linear system,
obtained while considering that the velocity vector the following equality is demonstrated:
field V is: Yλ2 ∧ Yλ3 = t Yλ1 (A.22)
• either collinear to the “slow” eigenvector Yλ2 Thus, coplanarity and orthogonality conditions
• either perpendicular to the “fast” eigenvector on are completely equivalent. 
the left t Yλ1
Note: In dimension three, numerical studies
There is equivalence between both conditions shown significant differences in the plot of slow
provided that the “fast” eigenvector on the left manifold according to whether one or the other
tY
λ1 is orthogonal to the “slow” eigenvector Yλ2 . of both conditions (coplanarity (A.14) or orthog-
Both co-ordinates of these eigenvectors defined in onality (A.20)) were used. These differences come
the above section make it possible to express their from the fact that each of these two conditions
scalar product: uses one or two eigenvectors whose coordinates are
∂g expressed according to the eigenvalues of the func-
t
Yλ1 · Yλ2 = λ1 λ2 − (λ1 + λ2 ) tional jacobian of the tangent linear system. Since
∂y
 2    these eigenvalues can be complex or real according
∂g 1 ∂f ∂g to their localization in the phase space the plot of
+ +
∂y ε ∂y ∂x the analytical equation of the slow manifold can be
908 J.-M. Ginoux & B. Rossetto

difficult even impossible. Also to solve this prob-     2 


dy 1 ∂f dy
lem it is necessary to make the analytical equation × − =0 (A.24)
dt ε ∂y dt
of the slow manifold independent of the eigenval-
ues. This can be carried out by multiplying each Both Eqs. (A.23) and (A.24) are equal provided
equation of the slow manifold by one or two “con- that:
jugated” equations. The equation obtained will be    
∂g 1 ∂f
presented in each case (dimension two and three) in = 0 and = 0
∂x ε ∂y
the next section.
These two last conditions are, according to the
definition of a dynamical system, satisfied because if
Slow manifold equation independent of the they were not both differential equations that make
eigenvectors the system completely uncoupled, it would not be
Proposition A.6. Slow manifold equations of a a system anymore. Thus, the equations obtained
dynamical system obtained by the collinearity/ by the collinearity and orthogonality conditions are
coplanarity and orthogonality conditions are equivalent.
equivalent. (V ∧ Yλ1 ) · (V ∧ Yλ2 ) = 0
Proof of Proposition 6 in Dimension Two. In order to ⇔
demonstrate the equivalence between the slow man- (V · Yλ1 )(V · t Yλ2 ) = 0
t (A.25)
ifold equations obtained by each condition, they 
should be expressed independently of the eigenval-
ues. So let us multiply each equation (A.11) then Equations (A.23) and (A.24) provide the slow
(A.17) by its “conjugated” equation, i.e. an equa- manifold equation of a two dimensional dynamical
tion in which the eigenvalue λ1 (resp. λ2 ) is replaced system independently of the eigenvalues of the func-
by the eigenvalue λ2 (resp. λ1 ). Let us take note tional jacobian of the tangent linear system. In order
that the “conjugated” equation (A.11) corresponds to express them, we adopt the following notations
to the collinearity condition between the velocity for:
vector field V and the eigenvector Yλ1 . The prod-
• the velocity vector field
uct of Eq. (A.11) by its “conjugated” equation is  
written as: ẋ
V
(V ∧ Yλ1 ) · (V ∧ Yλ2 ) = 0 ẏ
So, while using the trace and the determinant of • the functional jacobian
the functional jacobian of the tangent linear system  
a b
we have: J=
   2    c d
∂g ∂g dx 1 ∂f ∂g dx
− − • the eigenvectors coordinates (A.9)
∂x ∂x dt ε ∂x ∂y dt  
    2  λi − d
dy 1 ∂f dy Yλi
× − =0 (A.23) c
dt ε ∂y dt
with
In the same manner, the product of Eq. (A.17)
i = 1, 2
by its “conjugated” equation which corresponds to
the orthogonality condition between velocity vector Equations (A.23) and (A.24) providing the slow
field V and the eigenvector t Yλ2 is written as: manifold equation of a two-dimensional dynamical
(V · t Yλ1 )(V · t Yλ2 ) = 0 system independently of the eigenvalues of the func-
tional jacobian of the tangent linear system are then
So, while using the trace and the determinant of written as:
the functional jacobian of the tangent linear system
we have: cẋ2 − (a − d) ẋẏ − bẏ 2 = 0 (A.26)
   2    2

1 ∂f ∂g dx 1 ∂f ∂g dx αij ẋi ẏ j = 0 with
− − φ=
ε ∂y ∂x dt ε ∂x ∂y dt i,j=0
Differential Geometry and Mechanics: Applications to Chaotic Dynamical Systems 909


 2
= 0 si i + j = • the velocity vector field
αij = (A.27)  
= 0 si i + j = 2 ẋ
 
 V ẏ 

Proof of Proposition 6 in Dimension Three. In order
to demonstrate the equivalence between the slow • the functional jacobian
manifold equations obtained by each condition,  
the same step as that exposed in the above sec- a b c
tion is applied. The slow manifold equation should  
J = d e f
be expressed independently of the eigenvalues. In g h i
dimension three, each equation (A.14) then (A.20)
must be multiplied by two “conjugated” equations • the “slow” eigenvectors coordinates (A.12)
obtained by circular shifts of the eigenvalues. Let us  
take note that the first of the “conjugated” equa- bf + c(λi − e)
 
tions (A.14) corresponds to the coplanarity condi- Yλi cd + f(λi − a) 
tion between the velocity vector field V and the −bd + (λi − a)(λi − e)
eigenvectors Yλ1 and Yλ2 , the second corresponds
to the coplanarity condition between the velocity with
vector field V and the eigenvectors Yλ1 and Yλ3 .
The product of Eq. (A.14) by its “conjugated” i = 2, 3
equation is written as: • the “fast” eigenvectors coordinates on the left
[V · (Yλ1 ∧ Yλ2 )] · [V · (Yλ2 ∧ Yλ3 )] (A.18)
 
· [V · (Yλ1 ∧ Yλ3 )] = 0 (A.28) hd + g(λ1 − e)
t  
In the same manner, the product of Eq. (A.20) Yλ1  bg + h(λ1 − a) 
by its “conjugated” equation which corresponds to −bd + (λ1 − a)(λ1 − e)
the orthogonality condition between the velocity
Starting from the coplanarity condition (A.14)
vector field V and the eigenvector t Yλ2 and, the
and while replacing the eigenvectors by their coordi-
orthogonality condition between the velocity vector
nates (A.12) and while removing all the eigenvalues
field V and the eigenvector t Yλ3 is written as:
λ2 and λ3 thanks to the sum and the products (and
(V · t Yλ1 )(V · t Yλ2 )(V · t Yλ3 ) = 0 (A.29) also the square of the sum and the products) of the
eigenvalues of the functional jacobian of the tangent
By using Eq. (A.22) and all the circular linear system, we obtain the following equation:
shifts which result from this we demonstrate that
Eqs. (A.28) and (A.29) are equal. Thus, the equa- A1 ẋ − B1 ẏ + C ż = 0 (A.31)
tions obtained by the coplanarity and orthogonality
with
conditions are equivalent.
A1 = fλ21 − (ef + if + cd)λ1 + efi
[V · (Yλ1 ∧ Yλ2 )] · [V · (Yλ2 ∧ Yλ3 )]
+ cdi − cfg − f 2h
· [V · (Yλ1 ∧ Yλ3 )] = 0
B1 = cλ21 − (ac + ic + bf )λ1 + aci (A.32)
⇔ (A.30)
+ bfi − c2g − cfh
(V · Yλ1 )(V · Yλ2 )(V · t Yλ3 ) = 0
t t
C = bf 2 − c2d + cf(a − e)

Equation (A.31) is absolutely identical to
Equations (A.28) and (A.29) provide the slow that one would obtain by the orthogonality con-
manifold equation of a three-dimensional dynami- dition (A.20). Let us multiply Eq. (A.31) by
cal system independently of the eigenvalues of the its “conjugated” equations in λ2 and λ3 , i.e.
functional jacobian of the tangent linear system. In by (A2 ẋ − B2 ẏ + C ż) and (A3 ẋ − B3 ẏ + C ż). The
order to express them, we adopt the following nota- coefficients Ai , Bi are obtained by replacing in
tions for: Eq. (A.32) the eigenvalue λ1 by eigenvalue λ2
910 J.-M. Ginoux & B. Rossetto

Table 2. Slow manifold analytical equation of a two-


dimensional dynamical system.
(
P2
i j = 0 si i + j = 2
φ= αij ẋ ẏ = 0 with αij =
i,j=0 = 0 si i + j = 2
α20 = c
α11 = −(a − d)
α02 = −b

Table 3. Slow manifold analytical equation of a three-dimensional dynamical system.


(
P3
i j k = 0 si i + j + k = 3
φ= αijk ẋ ẏ ż = 0 with αijk =
i,j,k=0 = 0 si i + j + k = 3
α300 = d2h + dgi − fg 2 − dge
α030 = ch2 + abh − b2g − ibh
α003 = c2d + cfe − bf 2 − cfa
α210 = bdg + aeg − e2g + cg 2 − 2adh − 2fgh + deh − agi + egi + dhi
α120 = −abg + 2beg + a2h − fh2 − bdh − aeh + 2cgh − bgi − ahi + ehi
α201 = −bd2 + ade − cdg + 2afg + 2dfh − efg − adi − dei − fgi + di2
α102 = acd + cde − a2f − 2bdf + aef + cfg + f 2h − 2cdi + afi − efi
α021 = b2d − abe − 2bcg − 2ceh + ach + bfh + abi + bei + chi − bi2
α012 = 2bcd − ace + ce2 − abf − bef − c2g − cfh + aci − cei + 2bfi
α111 = abd − a2e − bde + ae2 − acg + 3bfg − 3cdh + efh + a2i − e2 i + cgi − fhi − ai2 + ei2

then by eigenvalue λ3 respectively for i = 2, 3. eigenvalues and also of the square of sum and prod-
We obtain: uct of eigenvalues and which are directly connected
to the elements of the functional jacobian matrix of
(A1 ẋ − B1 ẏ + C ż)(A2 ẋ − B2 ẏ + C ż) the tangent linear system. The equation obtained is
× (A3 ẋ − B3 ẏ + C ż) = 0 (A.33) the result of a demonstration (available by request
to the authors) which establishes a relation between
So,
the coefficients of this polynomial and the elements
3
 of the functional jacobian matrix of the tangent lin-
φ= αijk ẋi ẏ j ż k = 0 with ear system.
i,j,k=0 The expression (A.34) represents the slow man-
 ifold equation of a three-dimensional dynamical
= 0 si i + j + k = 3
αijk = (A.34) system independently of the eigenvalues of the func-
= 0 si i + j + k = 3
tional jacobian of the tangent linear system. Both
By developing this expression we obtain a poly- expressions (A.27) and (A.34) are also available at
nomial comprising terms of the sum and product of the address: http://ginoux.univ-tln.fr.

You might also like