You are on page 1of 14

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/324393581

First integrals and integrating factors of second order autonomous systems

Preprint · April 2018

CITATIONS READS

0 44

2 authors:

Tamas Kalmar-Nagy Balázs Sándor


Budapest University of Technology and Economics
2 PUBLICATIONS   0 CITATIONS   
76 PUBLICATIONS   891 CITATIONS   
SEE PROFILE
SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Nonlinear Dynamics View project

Porcolation: an invasion percolation model for mercury porosimetry View project

All content following this page was uploaded by Tamas Kalmar-Nagy on 10 April 2018.

The user has requested enhancement of the downloaded file.


First integrals and integrating factors of second order autonomous
systems
Tamás Kalmár-Nagy∗
Department of Fluid Mechnics, Faculty of Mechanical Engineering
Budapest University of Technology and Economics
Balázs Sándor†
Department of Hydraulic and Water Resources Engineering
Faculty of Civil Engineering
Budapest University of Technology and Economics
Water Management Research Group - Hungarian Academy of Sciences

Abstract
We present a new approach to the construction of first integrals for second order autonomous systems without
invoking a Lagrangian or Hamiltonian reformulation. We show and exploit the analogy between integrating
factors of first order equations and their Lie point symmetry and integrating factors of second order autonomous
systems and their dynamical symmetry.
We connect intuitive and dynamical symmetry approaches through one-to-one correspondence in the frame-
work proposed for first order systems. Conditional equations for first integrals are written out, as well as
equations determining symmetries. The equations are applied on the simple harmonic oscillator and a class of
nonlinear oscillators to yield integrating factors and first integrals.

1 Introduction
The importance of first integrals of differential equations have long been recognized. First integrals are conserved
quantities of the underlying dynamical system, providing qualitative information about its behavior. A first integral
can be used to reduce the order of the differential equation, and a complete set of first integrals determines the
solution of the equation. First integrals can prove useful in the stability analysis of the system: a first integral may
be used to construct a Lyapunov function [1] to be used with the second method of Lyapunov [2]. First integrals
can many times be related to the symmetries of the problem, but symmetries are not always necessarily related
to first integrals. González-López [3] discusses the connection between point symmetries and the integrability
by quadratures of second-order ordinary differential equations. He provides an example of a family of second-
order ordinary differential equations integrable by quadratures whose point symmetry group is trivial, refuting the
widespread belief that the existence of nontrivial point symmetries is a necessary condition for the integrability by
quadratures of ordinary differential equations. The knowledge symmetries is important, sometimes just as important
as the conserved quantity itself. Excellent textbooks for symmetry methods include [4] and [5].
Many different techniques have been published for deriving first integrals. Ad hoc construction of first integrals
is quite widespread, some less rigorous than others. Direct construction of first integrals is shown, for example, by
Sarlet and Bahar [6]. Sarlet and Bahar’s method consists of trying to construct a first integral in a manner similar
to the one used in obtaining the energy integral for conservative systems, namely that of multiplying the equation of
motion by an appropriate integrating factor. The most systematic, group-theoretical approaches include Noether’s
theorem and its various generalizations and analysis by one-parameter families of infinitesimal transformations.
The theory of Lie groups can be applied directly to any differential equation.
The purpose of this paper is to present a new approach to the construction of a first integral for second order
autonomous systems without invoking a Lagrangian or Hamiltonian reformulation. This paper is strongly influenced
by the work of Olver [7, 8]. The book by Bluman and Anco [9] describes the connection between the integrating
∗ jcnd@kalmarnagy.com
† sandor.balazs@epito.bme.hu

1
factor and the Lie point symmetry of first order systems, helping us to recognize the similar relation between the
integrating factor and dynamical symmetries of second order autonomous systems. In particular, the natural space
for studying symmetries is the 3-dimensional jet space (t, x, ẋ) for first order equations and (x, ẋ, ẍ) for second order
autonomous equations. In the first order case a first integral is a family of solution curves and symmetries map
solution curves into solution curves. For second order systems a first integral is a family of phase space curves
and dynamical symmetries map phase space curves to phase space curves. A paper by Muriel and Romero [10]
introduces a new class of symmetries (λ-symmetries) that strictly include Lie symmetries, but is more general. Later,
Muriel and Romero [11] investigated the relationship between integrating factors and λ-symmetries for ordinary
differential equations. In a 2009 paper Muriel and Romero [12] studied first integrals, integrating factors and λ-
symmetries of second-order differential equations. The knowledge of a λ-symmetry permits the determination of an
integrating factor or first integral via coupled first-order linear partial differential equations. These methods include
and complete other methods to find integrating factors or first integrals that are based on variational derivatives or
the Prelle–Singer method [13].
Cheb-Terrab and Roche [14] presented what they termed a systematic algorithm for the construction of inte-
grating factors for second order ordinary differential equations. They showed that there were instances of ordinary
differential equations without Lie point symmetries which were solvable with this algorithm. Leach and Bouquet
[15] demonstrate that the existence of integrating factors is paralleled by the existence of suitable Lie symmetries
which enable one to reduce the equations to quadratures thereby emphasising the fact that integrability relies upon
symmetry.
In this paper we consider dynamical systems of the form
ẍ = f (x, ẋ). (1)
Hale and Kocak [16] defines a first integral ω (x (t) , ẋ (t)) as a real-valued C1 function that is not constant on any
open subset of R2 if the function ω is constant along every solution of the planar differential equation (1). That is,
for any solution x (t) satisfying x (0) = x0 , ẋ (0) = v0 , the composite function satisfies ω (x (t) , ẋ (t)) = ω (x0 , v0 )
for all t for which the solution is defined. In this paper, ω and f are required to be sufficiently smooth: we assume
that their second derivatives exist, i.e. ω, f ∈C2 .
We present two approaches to calculate integrating factors and first integrals for Eq. (1). The first method gives a
simple conditional equation for the integrating factors in the form
Λ = Λ(x, ẋ), (2)
which has a straightforward connection with similar equations resulting from variational derivatives or λ-symmetries.
The second approach deduces direct relation between the integrating factors and a generalised Lie-symmetry (called
dynamical symmetry) of Eq. (1). Both of these approaches are similar with procedures for first order systems
ẋ = f (t, x). In particular, we exploit the analogy between first order and second order autonomous systems
regarding their natural “embedding” spaces (their so-called jet spaces), both of them isomorphic to R3 . Therefore
we first present the methods with first order systems (3) first, then we turn our attention to the second order sytems
(1).

2 Intuitive determination of integrating factors and first integrals:


first order systems
Let us consider the first order system (the independent variable is placed first)
ẋ = f (t, x). (3)
The family of solution curves
ω(t, x) = const. (4)
serve as first integrals of Eq. (3).
The total derivative Dt of the first integral ω (t, x) must be zero, i.e.
dω (t, x)
Dt ω (t, x) = = ωt + ẋωx = 0, (5)
dt
where partial differentiation is denoted with a subscript. Since ẋ − f = 0, we can write the following:
ωt + ẋωx = Λ(t, x)(ẋ − f ), (6)

2
where Λ(t, x) is the integrating factor. Rearranging Eq. (6) we get

(ωx − Λ) ẋ + ωt + f Λ = 0. (7)

A sufficient condition for Eq. (7) to hold is

ωx = Λ,
ωt = −f Λ. (8)

Equating the mixed second derivatives (i.e. ωxt = ωtx ) results in a partial differential equation for the integrating
factor:
Λt + (f Λ)x = 0. (9)

Remark 1 The more general formalism on writing the determining equations of the integrating factors of ordinary
differential equations is based on the variational derivative of the differential algebraic expression [9, 17, 18, 19]

θ(t, x, ẋ) = Λ(ẋ − f ). (10)

This variational derivative is also called Euler-operator and has the following form for one independent and one
dependent variable:

X ∂ ∂ ∂ ∂
E= (−Dt )j j = − Dt + Dt2 − ... . (11)
j=0
∂x ∂x ∂xt ∂xtt

Applying this operator to θ(t, x, ẋ) truncates the infinite sum at j = 1, and we arrive directly to expression (9):

∂θ ∂θ
E(θ) = − Dt = −Λt − (f Λ)x = 0. (12)
∂x ∂ ẋ
In other words, the Euler operator annihillates the total derivatives, i.e.

E(θ) = E(Dt ω) ≡ 0. (13)

2.1 Connection between integrating factors and Lie-point symmetries


This Section is based on the books of Olver [8], Bluman and Anco [9], and Cohen [20]. Let us first look at the
geometrical context of first order equations. The surface ẋ = f (x, t) is embedded in the jet space

J (1) = ((t, x), ẋ) ' R3 , (14)

where (t, x) is the total space of the independent and dependent variables, and ẋ spans the first jet space [9, 7].
This natural embedding will provide the connection with the treatment of second order autonomous equations.
A one-parameter Lie transformation group of the plane [t, x] (the plane of the independent and dependent
variables) transforms points (t, x) into points (t̃, x̃) according to

(t̃, x̃) = g · (t, x) , (15)



where g = g t (t, x, ε), g x (t, x, ε) is a group element and ε is the parameter. The infinitesimal generator of the
corresponding Lie-algebra has the form
∂ ∂
v = ξ(t, x) + η(t, x) , (16)
∂t ∂x
where
dg t dg x

ξ= and η = . (17)
dε ε=0 dε ε=0
For the interpretation of Lie-point symmetries of Eq. (3) we need to characterise the jet space (14) with the
corresponding first (because the highest derivative is first order - for a second-order system the second prolongation
will be used) extension/prolongation of the group action or the infinitesimal generator (16):

∂ ∂ ∂
v(1) = ξ +η + η (1) , (18)
∂t ∂x ∂ ẋ

3
where
η (1) = Dt η − ẋDt ξ = ηt + (ηx − ξt )ẋ − ξx ẋ2 (19)
expresses the transformation of the ẋ derivative in the first jet under the group action.
A group g, with generator (16) is a Lie-point symmetry group of Eq. (3) if it maps solution curves into solution
curves. Or equivalently, if the differential algebraic expression

F(t, x, ẋ) = ẋ − f (t, x) (20)

vanishes under the extended generator (18):


v(1) (F) = 0, (21)
provided
F = 0. (22)
Condition (21) means that the graph of f (t, x) (or the 0 level set of F) is invariant under the extended group
action, while mapping solutions into solutions on the plane [t, x]. Since the equation is first order, the first integrals
(4) are also families of solution curves ω(t, x) =const., i.e. these are conform-invariants to the generator (16)

v (ω) = Ω(ω), (23)

where Ω is an arbitrary function of the first integrals.


The connection between integrating factors and Lie-point symmetries arising from the pairing of condition (6)
on the total derivative of the first integrals, and condition (23) on the conformal invariance of the first integrals -
as families of solution curves - on the [t, x] plane, with Ω ≡ 1 (without loss of generality):

Dt ω = ωt + ẋωx = ωt + f ωx = 0, (24)
v (ω) = ξωt + ηωx = 1. (25)

From this system of equations we get (η 6= ξf )

1 f
ωx = , ωt = − , (26)
η − ξf η − ξf

and substituting Eq. (26) into Eq. (24) results in


1
Dt ω = (ẋ − f ). (27)
η − ξf

Comparing this with Eq. (6) we find the integrating factor in terms of the coefficient functions of the symmetry
generator as
1
Λ(t, x) = . (28)
η − ξf
The Lie-point symmetries of (3) can be calculated through the coefficients ξ and η of the generators with
symmetry condition (21), which can be expanded (with ẋ = f ) as

v(1) (F) = η (1) − ξft − ηfx = ηt + (ηx − ξt )f − ξx f 2 − ξft − ηfx = 0. (29)

Now we are ready to state the following: there is a one-to-one correspondance between the Lie-point symmetries
and the integrating factors of (3), because substituting Eq. (28) into the determining equation of the integrating
factors (9) results in
1 
2

Λt + (f Λ)x = − η t + (η x − ξt )f − ξx f − ξf t − ηfx
(η − ξf )2
1
=− v(1) (F) = 0. (30)
(η − ξf )2

Remark 2 The choice


η = ξf (31)

4
with arbitrary ξ solves the condition equation of symmetries (29) and hence provides infinite number of symmetries.
However, these are trivial symmetries with generators of the form
 
A = ξ ∂t + f ∂x , (32)

with which
A (ω) = 0, (33)
so each solution curve is invariant with respect to these symmetries, moreover no integrating factor corresponds to
these. Dealing with these types of symmetries does not help to solve the original equation.

Remark 3 The vector field (32) with ξ = 1 is called as the vector field of the equation. The commutator (or
Lie-bracket) of the vector field of the equation and a Lie-point symmetry generator (16) is the following vector field:
 
[A, v] = [∂t + f ∂x , ξ∂t + η∂x ] = A(ξ) − v(1) ∂t + A(η) − v(f ) ∂x . (34)

This new vector field acts on the first integral ω as


 
[A, v](ω) = A(ξ) − v(1) ωt + A(η) − v(f ) ωx , (35)

as well as    
[A, v](ω) = A v(ω) − v A(ω) = A(1) − v(0) ≡ 0. (36)

Since ωt and ωx are not necessarily zero, the coefficients in Eq. (34) must be, so these are proportional to the
coefficients in
Dt ω = ωt + f ωx = 0, (37)
hence (with v(1) ≡ 0)
A(η) − v(f )
A(ξ) = = ρ(t, x). (38)
f
Considering this in the commutator (34) results in

[A, v] = ρA, (39)

hence the Lie-point symmetry generators and the vector field of the equations are conform-invariants due to the
connection between the first integrals and Lie-point symmetries.

3 Analysis of second order autonomous equations


Here we try to construct integrating factors Λ(x, ẋ) for second order systems of the form (1)

ẍ = f (x, ẋ) . (40)

Analogously with the first order case (6), with ω = ω(x, ẋ) and Λ = Λ(x, ẋ) we can write the following formula:

Dt ω = ẋωx + ẍωẋ = Λ(ẍ − f ) = 0, (41)

or equivalently
(ωẋ − Λ)ẍ + ẋωx + f Λ = 0. (42)
Sufficient conditions for this to hold are

ωẋ = Λ,
ẋωx = −f Λ. (43)

The conditional equation (equality of the mixed derivatives) ωẋx = ωxẋ yields

ẋ2 Λx + ẋ(f Λ)ẋ − f Λ = 0, (44)

which is the counterpart of (9) for second order systems (1).

5
Theorem 4 The conditional equation (44) of the integrating factors is consistent the ones resulting from the more
general formalism of Euler-operators.

Proof. Applying the Euler-operator (11) on the differential algebraic expression

θ(x, ẋ, ẍ) = Λ(ẍ − f ) (45)

results in
∂θ ∂θ ∂θ
E(θ) = − Dt + Dt2 = A(ẋ, x)ẍ + B(ẋ, x) = E(Dt ω) ≡ 0. (46)
∂x ∂ ẋ ∂ ẍ
A sufficient condition for this equality to hold is that the expressions A and B are zero, i.e.

A(ẋ, x) = 2Λx + ẋΛxẋ + (f Λ)ẋẋ = 0, (47)

and
B(ẋ, x) = −(f Λ)x + ẋ(f Λ)xẋ + ẋ2 Λxx = 0. (48)
It can be easily checked that expressions (47) and (48) are derivatives of the left-hand side of Eq. (44):

(ẋ2 Λx + ẋ(f Λ)ẋ − f Λ)ẋ = ẋA(ẋ, x),


(ẋ2 Λx + ẋ(f Λ)ẋ − f Λ)x = B(ẋ, x). (49)

Thus expression (44) is consistent with the ones arising from Euler-operator, moreover it is easier to solve than
system (47)-(48) for the integrating factor.

3.1 Connection between integrating factors and various symmetries


From the aspect of Lie-groups, the two-jet
J (2) = (x, ẋ, ẍ) ' R3 (50)
3
spanned by the variable and its derivatives of the equation is isomorphic to R , just like the one-jet (14) in the
first order case. The main difference from the first order case is that we considered invariant families of solution
curves as first integrals in the first order case and now we are considering invariant families of phase space curves.
The symmetries considered are not Lie-point symmetries, but a special type of generalised symmetries, namely
dynamical symmetries.

3.1.1 Consistency with a λ-symmetry


The concept of λ-symmetries is based on a special generator field, which is similar to a Lie-point symmetry generator,
but its extension on the jet space is more general than the prolongation of Lie-point symmetries [11]. For example,
for the Lie-point symmetry generator
∂ ∂
v = ξ(t, x) + η(t, x) (51)
∂t ∂x
the first extension has the following form:
∂ ∂ ∂
v(1) = ξ +η + η (1) , (52)
∂t ∂x ∂ ẋ
with
η (1) = Dt η − ẋDt ξ + λ(η − ẋξ), (53)
which, compared with Eq. (19), is more general through the λ-part.
In [12] a connection had been deduced between the integrating factors and λ-symmetries, which are connected
to certain Lie-point symmetries. The conditional equation is

Λx + (λΛ)ẋ = 0, (54)

where
Q = η − ξ ẋ (55)
is the characteristic of v and
Dt Q
λ= . (56)
Q

6
Theorem 5 The conditional equation (44) for the integrating factors is a special form of (54), corresponding to
the v = ∂t Lie-point symmetry.

Proof. The time translation is a Lie-point symmetry of every equation of the form (1),

f
Q = −ẋ, λ= , (57)

the conditional equation is
1 2 
Λx + (λΛ)ẋ = ẋ Λx + ẋ(f Λ)ẋ − f Λ . (58)
ẋ2

3.1.2 One-to-one correspondence with the dynamical symmetries


Instead of the Lie-point symmetry generator (16) which we considered in the first order case, let us introduce the
infinitesimal generator of dynamical symmetries of (1), which has the general form [3, 8, 9]

∂ ∂ ∂
v = ξ(t, x, ẋ) + η(t, x, ẋ) + ζ(t, x, ẋ) . (59)
∂t ∂x ∂ ẋ
The vector field of Eq. (1) is
∂ ∂ ∂
A= + ẋ +f , (60)
∂t ∂x ∂ ẋ
and the expression (59) is a dynamical symmetry, if

[A, v] = ρ(t, x, ẋ)A. (61)

We consider the following vector field (this is a special case of a dynamical symmetry which exploits our analogy
between jet spaces for first and second order equations):


v = η(x, ẋ) , (62)
∂x
with its first extension
∂ ∂
v(1) = η + η (1) , (63)
∂x ∂ ẋ
where
η (1) = Dt η = A(η) = ẋηx + f ηẋ (64)
is a dynamical symmetry generator of Eq. (1).

Theorem 6 The differential equation (1)


ẍ = f (x, ẋ)
provides dynamical symmetries with generator (63) if η(x, ẋ) satisfies the following parabolic partial differential
equation:
ẋ2 ηxx + 2ẋf ηxẋ + f 2 ηẋẋ + ẋfx ηẋ + (f − ẋfẋ )ηx − fx η = 0. (65)
The corresponding transformation maps phase space curves into phase space curves, i.e. the first integrals

ω = ω(x, ẋ) = const.

are invariant families of phase space curves with the corresponding integrating factors with determining equation
(44)
ẋ2 Λx + ẋ(f Λ)ẋ − f Λ = 0.
Further, there is a one-to-one correspondence between these symmetries and integrating factors of the following form

Λ(x, ẋ) = . (66)
ηẋ ẋf + ηx ẋ2 − ηf

7
Proof. In analogy with the first order case and connection equations of the invariant family of curves (25) we can
write the following formulas:

Dt ω = ẋωx + f ωẋ = 0, (67)


(1) (1)
v (ω) = ηωx + η ωẋ = 1. (68)

Equation (68) provides that the space space curves are invariant family of curves of the dynamical symmetry
generator (63). Using this equation and formula (64) for η (1) we can write the following:


Dt ω = (ẍ − f ) = Λ(ẍ − f ) (69)
ηẋ ẋf + ηx ẋ2 − ηf

which proves the formula (66) and the invariance of the families of space space curves.
The dynamical symmetries of (1) corresponding to the generator (63) can be calculated from (second prolonga-
tion, because this is a second order system)
v(2) (F) = 0, (70)
where
F = ẍ − f, (71)
and
∂ ∂ ∂
v(2) = η + η (1) + η (2) . (72)
∂x ∂ ẋ ∂ ẍ
Here
η (2) = Dt η (1) = A(η (1) ) = ηxx ẋ2 + 2ηxẋ ẋf + ηẋẋ f 2 + ηẋ (ẋfx + fẋ f ) + ηx f. (73)
Expanding the condition of symmetry (70) results in

ẋ2 ηxx + 2ẋf ηxẋ + f 2 ηẋẋ + (f − ẋfẋ )ηx + ẋfx ηẋ − fx η = 0,

which is the symmetry condition (65) stated in the Theorem. This is a parabolic equation because its discriminant
2
is (2ẋf ) − 4ẋ2 f 2 = 0.
The one-to-one correspondence between the dynamical symmetries and integrating factors can be proven with
substituting (66) into the integrating factor determining equation (44)

ẋ2 Λx + ẋ(f Λ)ẋ − f Λ = ẋΛ2 v(2) (F) = 0. (74)

Remark 7 The geometrical interpretations in the jet space do not necessarily follow straightforwardly when one
considers dynamical symmetries. We found one-to-one correspondence betweeen dynamical symmetries and inte-
grating factors with direct connection of the families of phase space curves. However, we had considered a special
form of generators, namely Eq. (63). However, since time translation is a point symmetry of every equation of the
form (1), the graph of the differential algebraic expression F = ẍ − f (x, ẋ) in the two-jet (50) simply consists of the
same sections translated along the t axis. The transformations corresponding to (63) are maps on the (x, ẋ) phase
space and the first integrals are invariant families of phase space curves through (68) and they are also symmetries
of Eq. (1) through the condition (70). The simple structure of this two-jet is also revealed by the commutator
 ∂  ∂
[A, v(1) ] = A(η) − v(1) (ẋ) + A(η (1) ) − v(1) (f )
∂x ∂ ẋ
(1) (1) ∂ (2) ∂

= η −η + v (F) ≡ 0, (75)
∂x ∂ ẋ
if the symmetry condition (70) holds. It means that the flows of the vector fields A and v(1) commute on the surface
of the graph of first integrals ω(x, ẋ).

8
4 Application for oscillators
There are numerous papers dealing with the important topic of determining elementary first integrals of two-
dimensional autonomous systems of differential equations.
Prelle and Singer [21] proposed a procedure for solving nonlinear first-order ordinary differential equations where
the right-hand side is a rational function of the dependent and independent variables. It assumes a quasi-polynomial
form for the integrating factor with a specified polynomial degree. The application of this procedure for several
dynamical systems is given in [22]. Duarte et al. [13] have extended the Prelle-Singer procedure to second-order
ODEs where the right hand side is a rational function of the independent variable, the dependent variable and its
first derivative. Chandrasekar et al. [23] undertake the study of a generalized second order nonlinear equation using
a generalized extended Prelle-Singer procedure. Ibragimov [19] introduces the concept of the adjoint for nonlinear
equations and constructs Lagrangians and integrating factors. Integrating factors and the associated first integrals
for Lienard-type and frequency-damped oscillators were considered in [24] where first integrals are derived from the
method to compute λ-symmetries and the associated reduction algorithm. The knowledge of a λ-symmetry of the
equation permits the determination of an integrating factor or a first integral by means of coupled first-order linear
systems of partial differential equations.
Here we apply our proposed approaches to second-order oscillators ẍ = f (x, ẋ), i.e. we solve Eq. (44) for
determining the integrating factor
ẋ2 Λx + ẋ(f Λ)ẋ − f Λ = 0, (76)
and calculate the first integral ω from

ωẋ = Λ,
f
ωx = − Λ. (77)

We also solve the symmetry determining equation (65)

ẋ2 ηxx + 2ẋf ηxẋ + f 2 ηẋẋ + ẋfx ηẋ + (f − ẋfẋ )ηx − fx η = 0 (78)

with the formula (66) for the integrating factor


Λ= . (79)
ηẋ ẋf + ηx ẋ2 − ηf

We start out with the simple undamped harmonic oscillator

ẍ + x = 0. (80)

The PDE for the integrating factor is (f = −x)

ẋ2 Λx − ẋ(xΛ)ẋ + xΛ =
(ẋΛx − xΛẋ ) ẋ + xΛ =
ẋ2 Λx − xẋΛẋ + xΛ = 0. (81)

We find
x2 ẋ2
 
Λ = ẋg + (82)
2 2
where g is an arbitrary function. Taking g = 1 (so that Λ = ẋ) the PDE system (77) determining the first integral
is

ωẋ = ẋ,
ωx = x, (83)

from which we we recover the expected first integral

x2 ẋ2
ω= + . (84)
2 2

9
The symmetry determining equation

ẋ2 ηxx − 2ẋxηxẋ + x2 ηẋẋ − ẋηẋ − xηx + η = 0 (85)


is satisfied by the linear function (dilation)
η = ax + bẋ, (86)
with which
ẋ 1
Λ= = ẋ . (87)
−ηẋ ẋx + ηx ẋ2 + ηx a (x2 + ẋ2 )
This yields 
ln x2 + ẋ2
ω= + c. (88)
2a
2 2
With the choice a = 1/2, c = − ln 2 this results ω = ln x +
2

, a function of the simple first integral (84).
Next, we investigate the integrating factors and first integrals for the damped harmonic oscillator

ẍ + 2ζ ẋ + x = 0. (89)

Here
f = −2ζ ẋ − x. (90)
We first look at the symmetry determining equation
2
ẋ2 ηxx − 2ẋ (2ζ ẋ + x) ηxẋ + (2ζ ẋ + x) ηẋẋ − ẋηẋ − xηx + η = 0. (91)

This is again satisfied by the dilation


η = ax + bẋ, (92)
This yields
ẋ ẋ
Λ= 2
= . (93)
ηẋ ẋf + ηx ẋ − ηf a (x + ẋ2 + 2ζ ẋx)
2

While Eq. (76) can not easily be solved for Λ, it can easily be shown that this integrating factor solves the PDE
for the integrating factor

ẋ2 Λx − ẋ((2ζ ẋ + x) Λ)ẋ + (2ζ ẋ + x) Λ =


(ẋΛx − (2ζ ẋ + x) Λẋ ) ẋ + xΛ = 0. (94)

The “first integral” can then be calculated as (with the choice a = 1/2)

2ζ x + ζ ẋ
ω = ln x2 + ẋ2 + 2ζ ẋx − p

atanh p . (95)
2
ζ −1 ẋ ζ 2 − 1

We note here that this first integral is formal in the sense that it has an obvious singularity at x = ẋ = 0 and
hence it does not satisfy the smoothness criterion of our definition for first integrals. The function (93) appears
in Cantwell’s book ([25], p. 116). While we found this function from the original equation through the dynamical
symmetry, Cantwell rewrites the oscillator equation into Hamiltonian form. He then makes this nonexact first
order equation exact with an integrating factor (with the method described in 2.1). This is the crucial point:
the dynamical symmetry of the second order autonomous ODE and the Lie-point symmetry of the corresponding
Hamiltonian system are the same dilation symmetry in the same phase space.
We now turn our attention to nonlinear oscillators. For the Duffing oscillator

ẍ + x + δx3 = 0, f = −x − δx3 , (96)

we get the integrating factor by solving Eq. (76) as


 2
ẋ2 x4

x
Λ = ẋg + +δ . (97)
2 2 4

10
With g = 1 the PDE system (77) determining the first integral is
ωẋ = Λ = ẋ,
f
ωx = − Λ = x + δx3 , (98)

from which we recover the well-known first integral
x2 ẋ2 x4
ω= + +δ . (99)
2 2 4
The nonlinear oscillator
ẍ + (1 + ẋ) x = 0, (100)
has been studied in the context of relaxation oscillations [26] and laser oscillations [27]. Beatty and Mickens [28]
and later Mickens [29] and Kalmár-Nagy and Erneux [30] investigated the nonlinear oscillator
ẍ + 1 + ẋ2 x = 0.

(101)
These oscillators are characterized by a velocity-dependent stiffness coefficient and depend on only one parameter.
Their general form is
ẍ + ẋn x + x = 0, (102)
and thus
f = −x (1 + ẋn ) . (103)
The PDE for the integrating factor is
(ẋΛx − x (1 + ẋn ) Λẋ ) ẋ + (x + (1 − n) xẋn ) Λ = 0,
which yields  !
ẋ x2 +12 F 1, n2 , 1 + n2 , −ẋn ẋ2
Λ= g , (104)
1 + ẋn 2
where 12 F denotes the hypergeometric function and
∞ 2

(1)n −ẋkn
 
1 2 2 X
n n
2F 1, , 1 + , −ẋn = 2
 . (105)
n n 1+ n n
k!
k=0
Here ()n denotes the Pochhammer symbol

1 n=0
(q)n = . (106)
q (q + 1) .. (q + n − 1) n>0
Taking g = 1, the integrating factor becomes

Λ= , (107)
1 + ẋn
and the PDE system (77) determining the first integral is

ωẋ = Λ = ,
1 + ẋn
f
ωx = − Λ = x. (108)

The corresponding first integral is
x2
 
1 2 2
ω= + F 1, , 1 + , −ẋn ẋ2 . (109)
2 4 n n
For n = 1, 2, 4 we get the explicit first integrals
x2
ω= + ẋ − ln (1 + ẋ) , (110)
2
x2 1
+ ln 1 + ẋ2 ,

ω= (111)
2 2
x2 1
ω= + arctan ẋ2 . (112)
2 2

11
5 Conclusions
We presented an intuitive as well as a more formal approach to determine first integrals of second order systems.
The geometrical analogy between first order nonautonomous systems and second order autonomous systems was
elucidated by showing that in both cases the jet spaces were isomorphic to R3 . Partial differential equations for
first integrals and for determining symmetries were written out. Application of these equations were shown on the
simple harmonic oscillator (both in the undamped and damped case) and a class of nonlinear oscillators.

References
[1] Nikolai Gurevich Chetaev. The stability of motion. Pergamon Press, 1961.
[2] Nicolas Rouche, Patrick Habets, and Michel Laloy. Stability theory by Liapunov’s direct method, volume 4.
Springer, 1977.

[3] Artemio González-López. Symmetry and integrability by quadratures of ordinary differential equations. Physics
Letters A, 133(4-5):190–194, 1988.
[4] Peter E Hydon and Peter Ellsworth Hydon. Symmetry methods for differential equations: a beginner’s guide,
volume 22. Cambridge University Press, 2000.

[5] Daniel J Arrigo. Symmetry analysis of differential equations: an introduction. John Wiley & Sons, 2015.
[6] W Sarlet and LY Bahar. A direct construction of first integrals for certain non-linear dynamical systems.
International Journal of Non-Linear Mechanics, 15(2):133–146, 1980.
[7] Peter J Olver. Equivalence, invariants and symmetry. Cambridge University Press, 1995.

[8] Peter J Olver. Applications of Lie groups to differential equations, volume 107. Springer Science & Business
Media, 2000.
[9] George Bluman and Stephen Anco. Symmetry and integration methods for differential equations, volume 154.
Springer Science & Business Media, 2008.

[10] C Muriel and JL Romero. New methods of reduction for ordinary differential equations. IMA Journal of
Applied mathematics, 66(2):111–125, 2001.
[11] C Muriel and JL Romero. Integrating factors and λ–symmetries. Journal of Nonlinear Mathematical Physics,
15(3):300–309, 2008.
[12] C Muriel and JL Romero. First integrals, integrating factors and λ-symmetries of second-order differential
equations. Journal of Physics A: Mathematical and Theoretical, 42(36):365207, 2009.
[13] LGS Duarte, SES Duarte, LACP da Mota, and JEF Skea. Solving second-order ordinary differential equations
by extending the Prelle-Singer method. Journal of Physics A: Mathematical and General, 34(14):3015, 2001.
[14] Edgardo S Cheb-Terrab and Austion D Roche. Integrating factors for second-order odes. Journal of Symbolic
Computation, 27(5):501–519, 1999.
[15] PGL Leach and S É Bouquet. Symmetries and integrating factors. Journal of Nonlinear Mathematical Physics,
9(sup2):73–91, 2002.
[16] Jack K Hale and Hüseyin Koçak. Dynamics and bifurcations, 2012.

[17] Stephen C Anco and George Bluman. Integrating factors and first integrals for ordinary differential equations.
European Journal of Applied Mathematics, 9(3):245–259, 1998.
[18] Nail H Ibragimov. Invariant Lagrangians and a new method of integration of nonlinear equations. Journal of
mathematical analysis and applications, 304(1):212–235, 2005.
[19] Nail H Ibragimov. Integrating factors, adjoint equations and lagrangians. Journal of Mathematical Analysis
and Applications, 318(2):742–757, 2006.

12
[20] Abraham Cohen. An Introduction to the Lie Theory of one-parameter groups: With Applications to the solution
of Differential Equations. DC Health & Company, 1911.
[21] Myra Jean Prelle and Michael F Singer. Elementary first integrals of differential equations. Transactions of
the American Mathematical Society, 279(1):215–229, 1983.
[22] YK Man. First integrals of autonomous systems of differential equations and the Prelle-Singer procedure.
Journal of Physics A: Mathematical and General, 27(10):L329, 1994.
[23] VK Chandrasekar, SN Pandey, M Senthilvelan, and M Lakshmanan. A simple and unified approach to identify
integrable nonlinear oscillators and systems. Journal of mathematical physics, 47(2):023508, 2006.
[24] Emrullah Yaşar. Integrating factors and first integrals for Liénard type and frequency-damped oscillators.
Mathematical Problems in Engineering, 2011, 2011.
[25] Brian Cantwell. Introduction to Symmetry Analysis, volume 29. Cambridge University Press, 2002.
[26] Steven M Baer and Thomas Erneux. Singular Hopf bifurcation to relaxation oscillations. SIAM Journal on
Applied Mathematics, 46(5):721–739, 1986.

[27] Thomas Erneux, SM Baer, and Paul Mandel. Subharmonic bifurcation and bistability of periodic solutions in
a periodically modulated laser. Physical Review A, 35(3):1165, 1987.
[28] John Beatty and Ronald E Mickens. A qualitative study of the solutions to the differential equation. Journal
of sound and vibration, 283(1):475–477, 2005.
[29] Ronald E Mickens. Investigation of the properties of the period for the nonlinear oscillator. Journal of Sound
and Vibration, 292(3):1031–1035, 2006.
[30] Tamás Kalmár-Nagy and Thomas Erneux. Approximating small and large amplitude periodic orbits of the
oscillator. Journal of Sound and Vibration, 313(3):806–811, 2008.

13

View publication stats

You might also like