Ibragimov - 1992 - Group Analysis of Ordinary Differential Equations

Uspekhi Mat Nauk 47:4 (1992), 83-144 Russian Math.
Surveys 47:4 (1992), 89-156
Group analysis of ordinary differential equations and

the invariance principle in mathematical physics
(for the 150th anniversary of Sophus Lie)
N.Kh. Ibraglmov
CONTENTS
Preface 90
Chapter I. Definitions and elementary applications 92
§1.1. One-parameter transformation groups 93
§1.2. Prolongation formulae 96
§1.3. Groups admissible by differential equations 98
§1.4. Integration and reduction of order using one-parameter groups 99
1.4.1. Integrating factor 99
1.4.2. Method of canonical variables 100
1.4.3. Invariant differentiation 101
§1.5. Defining equations 102
§1.6. Lie algebras 104
§1.7. Contact transformations 105
Chapter II. Integration of second-order equations admitting a two-dimensional 107
algebra
§2.1. Consecutive reduction of order 108
2.1.1. An instructive example 108
2.1.2. Solvable Lie algebras 109
§2.2. The method of canonical variables 111
2.2.1. Changes of variables and basis in an algebra 111
2.2.2. Canonical form of two-dimensional algebras 112
2.2.3. An integration algorithm 113
2.2.4. An example of implementation of the algorithm 114
Chapter III. Group-theoretical classification of second-order equations 115
§3.1. Equations admitting a three-dimensional algebra 115
3.1.1. Classification in the complex domain 116
3.1.2. Classification over the reals. Isomorphism and similarity 117
§3.2. The general classification result 120
§3.3. Two remarkable classes of equations 121
3.3.1. The equation 121
y" + F3(x,y)(y')3 + F2(x,y)(yf + Fl(x,y)y' + F(x,y) = 0.
Linearizability criteria
3.3.2. Equations y" + a(x)y' + b(x)y = c(x)y~3 122
90 N.Kh. Ibragimov
Chapter IV. Ordinary differential equations with a fundamental system of 124

solutions (following Vessiot—Guldberg—Lie)
§4.1. The main theorem 124
§4.2. Examples 125
§4.3. Projective interpretation of the Riccati equation 127
§4.4. Linearizable Riccati equations 128
Chapter V. The invariance principle in problems of mathematical physics 129
§5.1. Spherical functions 129
§5.2. A group-theoretical touch to Riemann's method 131
§5.3. Symmetry of fundamental solutions, or the first steps in group analysis 139
in the space of distributions
5.3.1. Something about distributions 139
5.3.2. Laplace's equation 142
5.3.3. The heat equation 144
5.3.4. The wave equation 145
Chapter VI. Summary of results 146
References 153
Preface
"The extreme importance of the work of Lie for the general development
of geometry cannot be overestimated: I am sure that in the near future it will
be even more so", wrote F. Klein in his reaction [1] to S. Lie's results on
group-theoretical foundations of geometry, which Lie submitted as his entry
for the first N.I. Lobachevskii prize. Without any doubt, this is also a fair
valuation of Lie's work in the area of differential equations.
Marius Sophus Lie was born on 17 December 1842 in the village
Nordfjordeid in Norway. He was the sixth (youngest) child in the family of
Johann Hermann Lie, a Lutheran pastor. Starting in 1857 he studied in
Christiania (now Oslo), first in a grammar school and then (1859 — 1865) in
the university. Among other events that determined the choice of his scientific
direction, of importance were: self-study in 1868 of the geometric works
of Chasles, Poncelet, and Plucker, a journey to Germany and France
(1869 — 1870) and a meeting there with Klein (which grew into a close
friendship and a long term partnership), Chasles, Jordan, and Darboux.
During 1872 — 1886 Lie worked in the university of Christiania, and from
1886 to 1898 in the university of Leipzig. Lie died on 18 February 1899
in Christiania. The life, development of ideas and work of the great
Norwegian mathematician are discussed in detail in the excellent book of
E.M. Polishchuk, "Sophus Lie" (Nauka, Leningrad 1983). This book reflects
in sufficient detail all the main aspects of Lie's scientific work, and enables the
reader to get close to Lie the man.
One of the more remarkable achievements of Lie was the discovery that the
majority of known methods of integration of ordinary differential equations,
which seemed up to that time artificial and internally disconnected, could be
derived in a unified manner using the theory of groups (see the book [2]).
Group analysis of differential equations 91
Sophus Lie (1842-1899)
Moreover, Lie [3] provided a classification of all ordinary differential equations

of arbitrary order in terms of the symmetry groups they admit, and thus
described the whole collection of equations for which integration or lowering
of the order could be effected by group-theoretical methods. However, these
and other very valuable results he obtained could not for a long time be
widely disseminated and remained known to only a few. It could be said that
this is the state of affairs today with methods of solution of problems of
mathematical physics: many of these are of a group-theoretical nature, but
are presented as a result of a lucky guess.
The present survey, written to mark the coming 150th anniversary of
Sophus Lie's birth, was planned as a possible way of presenting group-
theoretical methods in an advanced course on differential equations. It
represents a significant expansion of the chapter on group-theoretical analysis
of the course on mathematical physics I teach in the Moscow Physical-
Technical Institute. In this work I present the main algorithms and results of
Lie, as well as related results of other authors, including some of my own.
92 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
N.Kh. Ibragimov
C H AP T ER I
DEFINITIONS AND ELEM ENTARY APPLICATIONS

The concept of a differential equation has two components. In the case of
first- order ordinary differential equations, for example, we have to
F(x, y, y') = 0 in the space of the three variables
1) define the surfacezyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
x, y, y'; we shall call this surface the hull of the differential equation;
2) determine a class of solutions; thus, for example, a smooth solution is a
continuously differentiable function φ(χ) such that the curve y = φ(χ),
y' = 9φ(χ)/ 9χ belongs to the hull, that is, F(x, φ(χ), d<p(x)/ dx) = 0
identically for all χ; passage to discontinuous or generalized solutions of the
equation (with the same hull) changes the situation drastically.
In the process of integrating differential equations, the crucial step is the
simplification of the hull by a suitable change of variables. To this end we
use the symmetry group of the equation (or the admissible group), which is
defined as the group of transformations of the (JC, j>)- plane, the prolongation
of which to the derivatives y', ... leaves the hull of the equation under
consideration invariant.
In this chapter we concentrate mainly on finding and using one- parameter
symmetry groups of ordinary differential equations.
Fig. 1. The hull of the Ricatti equation^ .y' + >>2—2/x2 = 0 is_a surface invariant under the
group of non- homogeneous stretchings χ = xe", y = ye~", y' = y'e~ 2a.
Group analysis of differential equationszyxwvutsrqponmlkjihgfedcbaZYXWV
93
Example. The Riccati equation y'+y1—2/ x2 = 0 admits the transformation

a
group χ = xe", y = ye~ , since the hull of this equation (see F ig. 1) is
invariant with respect to non- homogeneous stretching χ = xea, y = ye~",
y' = y'e~ 2a, obtained by the prolongation of this group to the first
derivative y'.
After the change of variables / = In x, u = xy, we have the equation
W' + M2 — M—2 = 0, which defines a cylindrical surface (a parabolic cylinder;
see F ig. 2) as the stretchings are transformed into the group of translations
/ = ί + β, « = «, «' = «'.
Fig. 2. The hull of the equation u' + u2- u- 2 = 0 obtained from the Ricatti equation
y' + y2—2/ x2 = 0 by the change of variables t = In x, u = xy.
§1.1. One- parameter transformation groups

We consider invertible transformations in the (x, j>)- plane
(1.1) x=<p(x, y, a), y=$(x, y, a ) ,
depending on a real parameter a, such that
(1- 2) <p|«- o= a;, ψ|ο= ο= ίΛ
We say that these transformations form a one- parameter group G if consecutive
application of two of these transformations is equivalent to the application of
a third transformation of the form (1.1). By a suitable choice of parameter a
this group property can be written in the following way:
φ ( ϊ , y, b)*=q>(x, y, a+b),
(1.3)
ψ(χ, jy, δ ) = ψ ( χ , y, a+b).
N.Kh. Ibragimov
If the conditions (1.3) hold n ot for all values of the parameter (in some
a and b, we call G a
fixed interval), but only for sufficiently small values ofzyxwvutsrqponmlkjihgfedcbaZYX
local one- parameter group. In group analysis of differential equations one deals
precisely with such local groups, which for brevity are called simply groups.
Transformations (1.1) are called point transformations (unlike, for example,
contact transformations, in which the transformed values x, y" depend also on
the derivative y', see §1.7), an d the group G is called a group of point
transformations. F rom (1.2), (1.3) we see that the inverse transformation to
(1.1) is obtained by changing the sign of the group parameter a:
D enoting by T a the transformation (1.1) which takes a point (x, y) into a

point (x, y), by / the identity transformation, by TO"1 the transformation
inverse to T a, which takes a the point (x, y) into (x, y), and by T aT b a
composition of two transformations, we summarize the properties (1.2) —(1.4):
Definition 1.1. The collection G of transformations T a is called a local one-
parameter group if
1) T o = / e G, 2) T aT b = T a+b e G, 3) T~ x = T.a e G for a and b
sufficiently small.
Let us expand the functions φ , ψ in Taylor series in the parameter α in a
neighbourhood of a = 0 and write down the infinitesimal (infinitely small)
transformation (1.1), with (1.2) taken into account, as
(1.1') x»x+l(x, y)a, y^y+r\ (x, y)a,
where
(1.5) y, a)
da da
The vector (ξ, η) with components (1.5) is the tangent vector (at the point
(x, y)) to the curve described by the transformed points (x, y), and is
therefore called the tangent vector field of the group.
F or example, for the group of rotations
χ =x cos a+y sin a, y=y cos a—x sin a
the infinitesimal transformation has the form
x~x+ya, yt&y—xa.
A one- parameter group can be completely recovered if we know its
infinitesimal transformation (1.1') by using the following L ie equation with the
correct initial conditions:
act Ψ), ΦΙ«=* = Χ,
(1.6)
i$. = η(φ , Ψ). Ψ α= = ο =y-
da
Group analysis of differential equationszyxwvutsrqponmlkjihgfedcbaZYXW
95
The tangent vector field (1.5) is sometimes also written as a first- order
differential operator
(1.7)
scalar under an arbitrary change of variables, unlike the

which behaves as azyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
vector (ξ, η ). Lie called the operator (1.7) the symbol of the infinitesimal
transformation (1.Γ) or the symbol of the group; later on the term
infinitesimal operator entered into usage (or, for short, the operator of the
group), while in physical literature one frequently encounters the term
generator of the group.
Definition 1.2. A function F(x, y) is called an invariant of the group of

transformations (1.1) if for each point (x, y) the function F is constant on the
trajectories described by the transformed points (x, y):
Theorems 1.1 —1.4 quoted below are well known. We omit the proofs of
the first three, while the proof of the fourth one is included as it uses the
method of passing to canonical variables we shall use in what follows.
Theorem 1.1. A function F(x, y) is an invariant if and only if it satisfies the
partial differential equation
(1.8) X F ^ H x , y ) - M ) ^
Hence it follows that every one- parameter group of point transformations

in the plane has one independent invariant, which can be taken to be the left-
hand side of the first integral J(x, y) = C of the ordinary differential
equation conjugate to (1.8) (characteristic equation):
dx dy
(1.8')
l(x, y) η(χ, y)
Any other invariant is then a function of J.
The concepts we introduced above can be generalized in an obvious way to
the multidimensional case, in which instead of transformations in the
(x, j)- plane we consider groups of transformations
(1- 9) x' - fi *. β ), i=i, . . . , «,
1
in the «- dimensional space of points χ = (χ , ..., χ"). Let us concentrate on
this higher- dimensional case and consider a system of equations
(1.10) F f ( *) - 0 / ".(*)«<>, s<n.
N.Kh. Ibragimov
5Fk/ dx'\ \ is equal to s at all

We shall assume that the rank of the matrix ||zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQP
points χ satisfying the system of equations (1.10). The system of equations
(1.10) defines an («—j)- dimensional surface M.
Definition 1.3. We say that the system of equations (1.10) is invariant with
respect to the group G of transformations (1.9) if each point χ on the surface
Μ is moved along the surface, that is, if from JC e Μ it folllows that χ e M.
Theorem 1.2. The system of equations (1.10) is invariant with respect to the
group of transformations G (1.9) with symbol
dxl ' Έ α~
if and only if
(1.12)
Theorem 1.3. Lei iAe system of equations (1.10) ac/mii α grow/» G, enrf suppose
that the tangent vector ξ(χ) of the group does not vanish on the surface Μ
defined by the equations (1.10). Then the system (1.9) can be equivalently
rewritten in the form
(1.10') Φ ,,(Α(.τ), . . . , / , - , ( *) ) «0 , fc- 1 s,
wAere 7i(x) / n _ i(x) w α basis of invariants {the collection of all functionally
independent invariants) of the group G. Equations (1.10) and (1.10') are
equivalent in the sense that they define the same surface M.
Theorem 1.4. Every one- parameter group of transformations (1.1) reduces to the
group of translations t — t + a, u = u with symbol X = 9/9/ by a suitable
change of variables
ί= ί( ζ , y), u= u( x, y).
Such variables /, u are called canonical variables.

Proof. U nder a change of variables the differential operator (1.7) transforms
according to the formula
(1.13) * = X( , , £ + *< «) - !- .
Therefore canonical variables are found from the equations
(1.14) X(O = 1, *( u ) - 0 .
§1.2. Prolongation formulae

Let us write down the rules for the transformation of y', y" under the
action of the point transformations (1.1), regarded as a change of variables.
H ere it is convenient to use the "t o t al" differentiation operator
D erivatives are transformed according to the formulae
dy DP Px + y'Pv + y"Pv.
(1.16) :
~dx D<f>
If we start from a group of point transformations G, then after adding the

formula (1.15), we obtain the prolonged group G, which acts in the space of the
ι
three variables (x, y, y'); if we add the formula (1.16), we have the twice
prolonged group G, acting in the space (x, y, y', y").
2
Substituting into (1.15), (1.16) the infinitesimal transformation (1.Γ)
χ = χ + αξ, y = y + ar\ and neglecting all terms of order o(a), we obtain
infinitesimal transformations of derivatives:
aD(r\ )
i+aD®
y' + [D (η) — y' D (ξ)] a == y' - f alh
/ + [Ζ>(ζι) — y"D(l)]a = y"
Therefore the symbols of the prolonged groups G, G are

1 2
These are called the first and second prolongations of the infinitesimal
operator (1.7). The term prolongation formulae is frequently used to denote the
expressions for the additional coordinates:
(1.17') = D (η) - y'D (ξ) = η , + ( η , — y' — y'%,

(1.18') = D (ζ,) - y"D (ξ) = i w +
( η » - 2 U ) y' 2 - y'%v
N.Kh. Ibragimov
§1.3. Groups admissible by differential equations

G be a group of point transformations and let G, G be its first and
Let zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
second prolongations, defined in §1.2.
Definition 1.4. We say that a first- order ordinary differential equation admits
a group of point transformations (1.1) G if the hull of the given equation
F(x, y, y')=0
(that is, the two- dimensional surface that this equation defines in the three-
dimensional space of the variables x, y, y') is invariant in the sense of
Definition 1.3 with respect to the prolonged group G.
ι
Similarly, a second- order differential equation admits a group G if its hull
(the three- dimensional surface
(119) F(x, y, y', y")=0
in the space x, y, y', y") is invariant with respect to the twice prolonged
group G.
2
This definition can be extended naturally to differential equations of higher
order, as well as to systems of partial differential equations.
It is not hard to construct differential equations admitting a given group by
using Theorem 1.3 on the representation of invariant equations in terms of
invariants. H ere it is useful to keep in mind that every one- parameter group
of transformations in the (x, j)- plane has exactly one independent invariant;
prolonging the group to the first derivative y' adds another invariant which
necessarily depends on y' and is therefore called a first- order differential
invariant. A twice prolonged group already has three functionally independent
invariants; the added invariant depends on the second derivative and is called
a second- order differential invariant.
It is convenient to compute second- order differential invariants in
accordance with the following theorem of Lie ([2], Ch. 16, §5).
Theorem 1.5. If for a group G we know the invariant u(x, y) and the first- order
differential invariant o(x, y, y'), then the derivative
u
. _ do _ vx - f y'vy + y"vy, _ Dv
du ux - f y'uy ~ Du
is a second- orderdifferential invariant.
Any (second or lower order) differential invariant of the group G is a
function of w, υ, w. By further differentiation we can obtain differential
2 2 3 3
invariants of higher orders d o/ du , d o/ du , . . . .
( 1)
F or reference, we quote in Tables 6.2 and 6.3 some first- order and second-
order differential equations together with the symbol of the one- parameter
(1)
The notation Table 6.« means Table η in Chapter 6.
Group analysis of differential equationszyxwvutsrqponmlkjihgfedcbaZ
99
group they admit. These tables have been constructed using Theorems 1.3
and 1.5.
§1.4. Integration and reduction of order using one- parameter groups

G roup theory sheds light on a whole range of approaches and results
concerning integration of specific types of equations that are widely used in
practice; it allows us to understand better the interconnections between
various particular methods of solution. In this section we discuss the simplest
applications of one- parameter groups in problems of integration and lowering
of the order of ordinary differential equations.
1.4.1. Integrating factor.
This method is applicable only to first- order equations. Let us consider a
first- order ordinary differential equation written in a form symmetric with
respect to χ and y:
(1.20) Q(x, y)dx- P(x, y)dy=0.
Theorem 1.6 (Lie [5]). Equation (1.20)
Q(x, y)dx- P(x, y)dy=0
admits a one- parameter group with the symbol (1.7)
if and only if the function
(1.21) µ = - Έ 7Γ3
{for ξ,ζ) — ϊ\ Ρ Φ 0) is an integrating factor for the equation (1.20).

Example. Let us consider the Riccati equation
Since it is an equation of power type, it is natural to expect that it admits a

group of stretchings. Substituting χ = ax, y = by, we have
2 2
For this equation to be invariant, we must have b/ a — b = I/ a . Hence
b = I/ a. Therefore the equation admits a one- parameter group of stretchings
(which can be written in the form χ = xe", y = ye~°) with symbol
(1.23) X=x J- - yJ>- .

Writing (1.22) in the form (1.20),
(1.22')
Κ»zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
N.Kh. Ibragimov
and applying the formula (1.21), we obtain the integrating factor
After multiplication by this factor, (1.22') is brought to the following form:

xdy - {- (zy* — 2jx)dx xdy + ydx dx
2 8 2
x*y — xy — 2 ζ ^ —x y — 2 χ
from which we obtain the solution in the form

*y- 2 =
c
zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLK
xy + l xs * χ (a:3 — C) '
1.4.2. M ethod of canonical variables.
If we know the admissible group, we can use Theorem 1.4 on reducing a one-
parameter group to a group of translations by changing to canonical variables.
Since the property of invariance of an equation with respect to a group is
independent of the choice of variables, by chan gin g t o can on ical variables we
obtain an equation which does not depend on one of the variables so that (in
the case of a first- order equation) it can be integrated in quadratures or (in
the case of a higher- order equation) we can lower its order. The following
examples clarify these statements.
Example 1. Let us solve the Riccati equation (1.22) by the method of
canonical variables. The group with operator (1.23) that this equation admits
is a particular case of a group of non- homogeneous stretching of Table 6.1 for
k - —1. In this case the canonical variables are
i= ln x, u=xy.
Changing to these variables reduces (1.22) to a clearly integrable form
u'+ u2- u- 2= 0.
Example 2. The linear second- order equation
(1.24) y"+f(x)y=0
admits a group of stretchings in y with the operator
Xz=y
(1- 25) 4y-
Reduction to translations is effected by the change of variables « = x, t = In y,
after which (1.24) takes the form
u "- u ' + / ( u ) u ' 3 = 0 .
101
The resulting equation does not contain the independent variable. Therefore
u' = p(u), and the
its order can be lowered by the well- known substitutionzyxwvutsrqponmlkjihgfedcbaZYXWVUT
problem reduces to the integration of the Riccati equation
1.4.3. Invariant differentiation.

If a second- order or higher- order equation admits a one- parameter group, then
by Theorems 1.3 and 1.5 its order can be lowered by one. This method is the
invariant formulation of well- known methods of lowering the order of an
equation, which does not contain either the dependent or the independent
variable explicitly.
F or definiteness, let us deal with second- order equations. By Theorem 1.3
any second- order equation admitting a group G can be written in terms of the
differential invariants u, o, w of, respectively, zeroth, first, and second order.
By Theorem 1.5 the second- order differential invariant can be chosen to be
of the form w = do/ du and the invariant second- order differential equation
under consideration can be written as
(1.26) lOr= F( u' p>·

Thus we can lower the order: if we have found an integral
(1.27) O (u, υ, C ) = 0
of the first- order equation (1.26), the solution of the original second- order
equation can be reduced to quadratures. Indeed, substituting known
expressions u(x, y) and o(x, y, y') into (1.27) leads to a first- order differential
equation, which admits the group G in view of the invariance of u and o, and
which can therefore be integrated in quadratures.
Let us use this second method to lower the order of the equation (1.24).
Writing down the first prolongation of the operator (1.25)
we find the invariants u = χ, ο = y' / y. F rom Theorem 1.5 we find the

second- order differential invariant
dv =
y" y"- ^V" v%
du y y* y
2
whence y "jy — do/ du + o . On substituting this expression into (1.24) we
obtain the first- order equation (1.26) in the form of the following Riccati
equation:
102 N.Kh. Ibragtmov
§1.5. Defining equations

Let us move on now to the question of construction of an admissible
group. Suppose we have a second- order equation (first- order equations are
Fy> = 0). By Definition 1.4 and
included as a particular case, in which zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJI
Theorem 1.2, the infinitesimal invariance criterion has the form of equation
(1.12) applied to the twice prolonged operator X:
2
.28) r\ Fv = 0,
where ζι and ζ 2 are computed from the prolongation formulae (1.17') and
(1.18'). Equation (1.28) is called the defining equation for the group admitted
by the ordinary differential equation (1.19).
In what follows we shall consider differential equations written in the
explicit form
(1.29) = / ( * , y, y')·
In this case, after substituting the values of ζ 1 ( ζ 2 from (1.17'), (1.18') with y"
given by the right- hand side of (1.29), the defining equation (1.28) assumes the
form
(1.30)
H ere/ (x, y, y') is a known function (we are dealing with a given differential
equation (1.29)), while the coordinates ξ and η of the required admissible
operator (1.7) are unknown functions of x, y. Since the left- hand side of
(1.30) contains in addition to x, y also the quantity y' considered as an
independent variable, the defining equation decomposes into several independent
equations, thus becoming an overdetermined system of differential equations
for ξ, η . Solving this system, we find all the operators admitted by the
differential equation (1.29) under consideration.
Example. Let us find the operators
dy
admitted by the second order- equation
(1.31) ff- if- e-

Group analysis of differential equationszyxwvutsrqponmlkjihgfe
103
H e r e / = ey—y' and the defining equation (1.30) has the form
— y'%
%y
The left- hand side of this equation is a third- degree polynomial in the
variable y'. Therefore the defining equation decomposes into the following
four equations, obtained by setting the coefficients of the various powers of y'
equal to zero:
(1.32) (y'r- . | M y = 0,
(1.33) G,')·: ^ - 4
(1.34) „': 2^- ^ + (- 1- ^- 3^= 0,
(1.35) (y')·: ^ I + ± n j c + ( ^ - _ 2^ —η ) ^ = ϋ.
From equations (1.32) and (1.33) by integration with respect to y we obtain
Let us substitute these expressions for ξ, η into (1.34), (1.35). As the

dependence of ξ and η on y ispolynomial, while the left- hand sides of
equations (1.34), (1.35) contain ey, we must have
!„= (), η , - 2 ξ , - η - 0 .
The first of these gives us ρ = 0, that is, the equality ξ = α(χ); taking this
into account, the second condition can be written in the form
Hence 2(a'- a/ x) + q = 0,2a' + b = 0. Therefore

ξ= α (ζ), η= - 2α '(ζ).
Substituting these expressions into (1.34), we have
from which a = C\ x hi x- VC^x; here equation (1.35) is satisfied identically.

N.Kh. Ibragimov
As a result, we have obtained the general solution of the defining equations

(1.32)- (1.35) in the form
C\ , C2. In view of the linearity of the defining

with constant coefficientszyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJ
equations, the general solution can be represented as a linear combination of
two independent solutions
η,= —
ξ2= χ , η2= - 2.
This means that (1.31) admits two linearly independent operators
(4.36) ^zlnx- z- — 2(1 + lu x) —, X l

dy dx dy
and that the set of all admissible operators is a two- dimensional vector space
with basis (1.36).
§1.6. Lie algebras

Let us return now to general properties of defining equations. As can be
seen from (1.30), a defining equation is a linear partial differential equation
for the functions ξ and η in the variables χ and y. Therefore the set of its
solutions forms a vector space, which was already noted in the example
above. However, in addition there is another property which is a specific
characteristic of defining equations. I t turns out that the set of solutions of a
defining equation forms a very special type of vector space called a Lie
algebra (the term is due to Weyl; Lie himself referred to an infinitesimal
group).
Definition 1.5. A L ie algebra of operators (1.7) is a vector space L with the
following property: if the operators
(1.37) Χ, - ^ ^ + η »^ - . Χ^ Ι,- L + ru- ^

are elements of it, so is their commutator
(1.38) [Xv A%] = (X, ( y - X2 (£,)) - L. + (Χ, (η2) - Xt (η,)) ~
The Lie algebra is denoted by the same letter L , and we take its dimension to
be the dimension of the vector space L .
A general statement about properties of defining equations for second- order
ordinary differential equations is the following result due to Lie [2].
105
1.6. The set of all solutions of a defining equation (1.30) for second-
TheoremzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
order equations (1.29) forms an algebra L of dimension r ^ 8. The maximal
dimension r = 8 is attained if and only if equation (1.29) either is linear or can
be linearized by a change of variables.
§1.7. Contact transformations

In addition to point transformations (1.1), groups of contact or tangent
transformations [6], [7] have been found to be of use in mechanics, geometry,
and the theory of differential equations. We shall discuss straight away the
multidimensional case with an arbitrary number η of independent variables
χ = (x1, ..., x") and one dependent variable u (for a larger number of
dependent variables there are no contact transformations that are not point
transformations; for a proof of this see, for example, [8], §28.3, [9] §9, or [10]
§14.1).
Let us denote by u' the collection of first derivatives ui = 9w/ 9x' and
consider a one- parameter group of transformations
χ'=ψ'(χ, u, α, α), δ = ψ(.τ, u. «', α ) . ΰ ί = ω ί (Λ:, a, u, a)
in the (2n+ l)- dimensional space of variables (x, u, «')·

Theorem 1.7. An operator
• d d ft
is a symbol of a group of contact transformations if and only if

ti dW ... dW Y dW dW
diii * Mill ' dx1 ' du
for some function W — W(x, u, u').
Lie [6] calls the function W defined in this theorem the characteristic
function of the group of contact transformations. Construction of a group of
contact transformations admissible for an ordinary differential equation thus
reduces to the determination of the characteristic function, the function
W = W(x, y, p) of three variables x, y and ρ = y', by substitution into the
defining equation of an operator of the form
(1.39) X = — Wµ 4- 4- (W — pWp) 4- - f (Wx + pWv)^- ·

ox oy op
Example. Let us find the operators (1.39) admissible by the equation
(1.40) i, '"= 0.
Prolongation of the operator (1.38) to third derivatives has the form
Λ = A ~i- £·> ——= • + L« —- —r— ,

N.Kh. Ibragimov
where the coordinate ζ 2 is computed from the formula (1.18'), while for ζ 3 we
have the analogous third prolongation formula
The defining equation has the form
and decomposes into the following four equations:
y" :W
( I f") ·: W
The system of differential equations in W thus obtained is easy to solve;
the result is
2
> (xp - 2yp) - t- C.o ( Xy-
^ d e p e n d s on ten arbitrary constants Ct, so equation (1.40) admits a ten-
parameter group of contact transformations. The Lie algebra of this group is
obtained by substituting the function W in the formula (1.39); it has the
following basis:
d
χ _. χ— XJL Y—x*JL X — u— X —
l 2 Λ x A J 5
~~diT* ~ dy ' 3— d y ' *~ dy ' ~ dx '
The operators ΛΊ, ..., Λ7 generate a 7- dimensional algebra of a group of

point transformations (that is the maximal group of point transformations
admitted by the equation (1.40)), and their prolongations to the derivative
y' = ρ are, naturally, not quoted here. The three remaining operators
X%, X9, X\ a are symbols of contact transformations proper.
This seemingly simple example has a deep geometrical meaning:
Lie established a remarkable connection between the group of contact
transformations for the equation (1.40) and confonnal mappings in three-
dimensional space (see [6], Ch. 10, §2).
Remark. F or every second- order differential equation there are infinitely many
contact transformations transforming it into any other given second- order
equation, for example into y" = 0. In particular, every second- order ordinary
107
differential equation admits an infinite group of contact transformations. The

situation changes drastically for higher- order equations. In this connection, in
addition to the example considered above, it is useful to keep in mind that for
y"' = 0 it is
a third- order equation to be reducible to the simplest form zyxwvutsrqponmlkjihgfedcbaZYXW
necessary that it be of the form
y~+A (y"V+B(y"y+Cif"+D=Q
with coefficients A, B, C, D dependent only on x, y, y' ([6] Ch. 3, §3).
Lie also indicated connections between contact transformations and the
theory of waves and was groping towards a group- theoretic approach to the
H uygens principle (see [7] and the book [6] p. 97), from which Fig. 3 is taken.
«*„
Fig. 3. The Huygens principle in the construction of a wave front coincides with the
statement that contact transformations χ - x+pt/ y/ l+p2, y = y- t\ \ J\ +p2, ρ = ρ
with characteristic function W = —y/ l +p2 forms a one- parameter group.
D evelopment of this idea of Lie together with the above- mentioned relation
of contact and conformal transformations has become subsequently the basis
of the solution of the problem on the H uygens principle (in the sense of
H adamard) for equations with a non- trivial conformal group (see [11] or [10],
Ch. 2, as well as the survey [12]).
CHAPTER II
INTEGRATION OF SECOND- ORDER EQUATIONS ADM ITTING

A TWO- DIMENSIONAL ALGEBRA
"I noticed that most ordinary differential equations integrable by the old
methods remain invariant under certain transformations, and these integration
methods consist in making use of this property. H aving thus presented the
old methods from a general viewpoint, I set myself a natural problem: to
develop a general theory of integration for all ordinary differential equations
admitting finite or infinitesimal transformations." (S. Lie [13], Ch. IV).
N.Kh. Ibragtmov
§2.1. Consecutive reduction of order

F rom §1.4 we know (see also [14]—[22]) that using an admissible one-
dimensional algebra we can lower the order of a second- order equation once.
Therefore it is natural to expect that if there is a two- dimensional algebra, we
can lower the order twice, that is, we can integrate the equation. We shall
start with a simple example.
2.1.1. An instructive example.
Let us consider the following linear equation:
f/ - +y'- ylx=O.
(2.1)zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
Clearly, it has the solution y = χ and by the superposition principle it admits

the group of transformations y = y + ax with symbol Χχ = xQ/ Qy. In
addition, equation (2.1) is homogeneous and therefore admits the group of
stretchings in y with symbol X2 = yd/ dy- Thus, for equation (2.1) we know
at least two admissible operators (out of the eight operators possible by
Theorem 1.6):
λ
(2.2) ' =
Their commutator is
(2.3) [Xt. X,)=X,.
Therefore the vector space with basis (2.2) is a two- dimensional Lie
algebra L 2.
We write down the first prolongation of the operator X\ \
ι dy dy
and find its invariants, u — χ, ο = y'—y/ x. By Theorem 1.5 we have a
second- order differential invariant
y Λ
du ~ χ ~ τ* ~~ y u *
from which we have y" = do/ du + v/ u. Expressing the left- hand side of (2.1)
in the variables u, υ, we obtain the following first- order equation (1.26):
(2.4) *. ()
N ext let us find out how the operator X2 acts in the («, o)- plane. To that
end, let us prolong it in y' and pass to the variables u, 0 according to (1.13):
L = y ^^-
w
JLr^Y
+ yJL = XX.(u)
r^Y = J- +X.(u)- Jl- .
,- dy dy .. t - v du ' s ' dv
We have obtained the sretching operator Υ = υδ/δΉ, which is admitted by
equation (2.4). Therefore the operator Xx has enabled us to lower the order of
109
equationzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
(2.1) and to reduce it to the form (2.4), while the operator X2 ensures
the ability to further integrate the resulting first- order equation. Construction
of the solution of the original equation is done in reverse order: first we find
the solution of equation (2.4):
which, by substituting ο - y'—yjx, is rewritten as a first- order non-

homogeneous equation
Above we started the reduction of order using the first of the operators (2.2).
This was a completely random choice. Let us work out next whether anything
changes if we start with the operator X2. Its differential invariants are
u=x. v=y'/ y, y y y y
Using these, equation (2.1) assumes the form
(2.5) ~ + ν"· + ν - = O.
du u
U nlike equation (2.4), this Riccati equation does not have an obviously
integrable form. Let us see if the so far unused operator X\ provides us with
the simplifying change of variables. To that end we find how it acts in the
(w, o)- plane. We have
Since y = e'vdx, the resulting operator can be written in the form

du
(2.6) r = , i ' ( i _ M , i;)J L.
ov
This is not an operator of a group of point transformations, as it contains an
integral term, but rather an operator of non- local symmetry of equation (2.5).
If we try to find a simplifying change of variables by reducing (2.6) to a
translation operator in accordance with §1.4.2, it becomes clear that the idea of
starting the reduction of order of equation (2.1) using the operator X2 is not a
good one. Why!
In order to understand the situation arising here and to formulate a general
principle of integration using a known group, we need some facts about the
structure of Lie algebras.
2.1.2. Solvable Lie algebras.
Let L r be a finite- dimensional Lie algebra of dimension r and let Ν be a linear
subspace in L r.
Definition 2.1. A subspace Ν is called a subalgebra if [X, Y] e Ν for all
X, Ye Ν (that is, if this subspace is a Lie algebra in its own right) and an
ideal of the algebra L T if [X, Y] e Ν for all X e Ν and all Ye L r.
N.Kh. Ibragimov
Ν is an ideal, then we can introduce an equivalence relation in the

IfzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
algebra L r: operators X and Υ in L r are taken to be equivalent if Y—X e N.
The set of all operators equivalent to a given operator X is called the residue
class represented by X; every element Υ of this residue class has the form
Υ = X+Z for some Ζ e N. The set of residue classes forms a Lie algebra,
which is called the factor algebra of the L r with respect to the ideal Ν and is
denoted by L r/ N. As elements of the factor algebra we can take
representatives of the corresponding residue classes.
If all the constructions are to be considered in the complex domain, then
we have the following theorem.
Theorem 2.1. Any algebra L r, r ^ 2, contains a two- dimensional subalgebra.
Moreover, every operator X can be included in a two- dimensional algebra.
Definition 2.2. An algebra L r is called solvable if there exists a sequence
(2.7)
of subalgebras of dimensions r, r — 1, ..., 1 respectively, in which each
subalgebra £ , _ ] is an ideal in L s (s = 2, ..., r).
Definition 2.3. Let X\ , ..., Xr be a basis of the algebra L r. The subspace
spanned by the commutators [Χµ, Χν] of all possible pairs of basis operators
forms an ideal, denoted by L'r, and is called the derivative algebra. D erivative
n + 1)
algebras of higher order are defined recursively: Z,£ = (L^)', n= 1, 2, ... .
Theorem 2.2. An algebra L r is solvable if and only if a derivative algebra of L r
of some order is identically zero: L^ = 0 for some η > 0.
Corollary. Every two- dimensional algebra is solvable.
In the case of a two- dimensional algebra L^, to construct the sequence (2.7)
we have to choose the basis X\ , X^ so that the equality [ΑΊ, Xi\ = ctX\ holds.
Then the one- dimensional algebra L\ spanned by X\ forms an ideal in La,
while the factor algebra L2/ L1 can be identified with the one- dimensional
algebra spanned by X%.
N ow we can give the answer to the question posed at the end of §2.1.1.
As can be seen from the equality (2.3), the operator X\ generates an ideal L\
in the algebra L 2 with basis (2.2). H aving lowered the order of equation (2.1)
using this ideal L\ , we have obtained a first- order equation (2.4) which admits
the factor algebra L2/L1 (which we identified with the one- dimensional algebra
spanned by X2) with a natural definition of its action on the
(M, o)- plane. But when, so as to lower the order, we used the one- dimensional
algebra with basis X2, which is not an ideal, we lost this additional symmetry.
In our attempt to restore the symmetry, we found it necessary to extend the
concept of symmetry by introducing the non- local symmetry operator (2.6).
Therefore, if we want to deal exclusively with point symmetries, reduction of
order must be done using ideals, which is what Lie himself did.
111
F rom Theorem 2.1 and the corollary of Theorem 2.2 it becomes clear why
the basis of methods of integration of second- order equations is taken to be
two- dimensional algebras (or, which amounts to the same, two- parameter
groups). It is sufficiently obvious also that if we want to integrate an
η > 2 by the method of consecutive order reduction, the
equation of orderzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
equation must admit a solvable «- dimensional algebra. F or a detailed
discussion of this question see, for example, [19], Theorems 2.60, 2.61,
and 2.64. H ere we restrict ourselves to the two- dimensional case.
§2.2. The method of canonical variables

If a second- order equation admits a two- dimensional algebra, then instead
of consecutive reduction of order we can directly make a change of variables
x, y (introduce canonical variables in the sense of Lie), in which the equation
under consideration assumes an integrable form. By Theorem 2.1 this also
includes equations that admit algebras of higher dimensions (in this case
complex- valued changes of variable may be necessary). At the same time,
there is a special method of integrating second- order equations using a
three- dimensional algebra [2], Ch. 24, but we do not discuss it here.
2.2.1. Changes of variables and basis in an algebra.
F irst we consider structural peculiarities of two- dimensional Lie algebras. One
of the properties, solvability of every two- dimensional algebra, was already
mentioned in the corollary of Theorem 2.2. The property of solvability is,
obviously, invariant with respect to changes of variables x, y and is
independent of the choice of basis.
Let us now discuss two other invariant properties, which lead to a partition
of all two- dimensional Lie algebras into four types, and form the basis of the
method of canonical variables for integration using two- dimensional algebras.
These properties are related to the rule of transformation of the commutator
(1.38) of the operators (1.37) and of their pseudo- scalar product
(2.8)
under a non- degenerate change of variables
d{
(2.9) t = t{x,y). u = u(x,y): f
and a change of basis in L 2:

(2.10) Xt'~O,Xt+aLjit, Χί'- β , Χ, + β Α; Δ
Lemma 2.1. When changing to a new basis (2.10) the commutator of the
operators (1.37) transforms according to
while under a change of variables (2.9) it transforms covariantly:

N.Kh. Ibraghnov
In every two- dimensional L ie algebra we can choose a basis

Proposition 2.1. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
X\ , X2 so that either of the two following commutation relations holds:
[X,,Xi)=O orzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQ
[X,,X,]= X,,
moreover, each of these relations is invariant with respect to a change of
variables (2.9).
Lemma 2.2. The pseudo- scalar product (2.8) transforms under (2.9) and (2.10)
according to
Proposition 2.2. The equation

x.vxô
is invariant with respect to any change of variables (2.9) and any change of
basis (2.10).
2.2.2. Canonical form of two- dimensional algebras.
Propositions 2.1 and 2.2 lead to the following result concerning the partition
of the set of all two- dimensional Lie algebras into four main types.
Theorem 2.3. Every two- dimensional L ie algebra can be reduced, by choosing an
appropriate basis X\ , X2, to one of the four different types defined by the
following structure relations:
I. [Xl,X1)=0, X.VXÔ;
II. [X,, X2 ]= 0, X,VX2= 0;
III. [X,, X2]= X,, X.VXÔ;
IV. [ Xi,Xt)= Xl,
These structure relations are invariant with respect to the change of variables (2.9).
U sing the invariance of structure relations I—IV with respect to the change
of variables x, y, we can, by a judicious change of variables, simplify the form
of basis operators of the two- dimensional algebras of all the above typesand
arrive at the following result:
Theorem 2.4. By a suitable change of variables (2.9) a basis of an algebra L 2
can be reduced to one of the following forms:
I X = — X =— •
1
dx ' * dy '
II. X = — . X = * — •
d d
χ — χ - ^~
1 4
dy dx ' ' dy
d
IV X - =— X
The corresponding variables x, y are called the canonical variables.

2.2.3. An integration algorithm.

N ow let us find all second- order ordinary differential equations that admit
two- dimensional Lie algebras of indicated types and integrate them.
Type I. To construct all second- order equations admitting an algebra L 2 with
the basis Xx = 8/9x and X2 = 9/o>, we must find a basis of second- order
differential invariants of these operators. In this case the prolongations of the
operators coincide with the operators themselves, so that the desired
differential invariants are y' and y". Therefore, a general second- order
equation, which admits an algebra L 2 of the first type, has the form
It can be integrated in quadratures:

y
= χ + Ct or explicitly y' = φ {χ - f Q ,
1 \y )
whence // = \ Φ ( x - f î) ^ ( x + Cx) + C 8 .
Type I I . A basis of differential invariants of an algebra L 2 of type II is given

by χ and y', while an invariant differential equation has the form
Its solution reduces to two quadratures:
Type I I I . H ere an invariant equation has the form
=4- / 00.
It is also solved by two quadratures:
S - ~- rr — In χ - f C1 or explicitly y' = φ (In χ - f C\ ),

t(y)
whence y = \ φ (In x 4- CJ dx • + Cv
Type IV. H ere we have an invariant equation
with the general solution
2/ = <
N.Kh. Ibragimov
The results of Theorems 2.3 and 2.4, as well as the corresponding invariant
equations, are given in Table 6.4, while the integration algorithm is presented
as Table 6.6.
2.2.4. An example of the implementation of the algorithm.
Let us apply the group algorithm to the equation
(2.11) iT = 4 - - •
' y2 xy
1st step. Determination of the admissible algebra L 2. Solving the defining
equation (1.30), we find that
, r)=(CiX+Ci/ 2)y.
Therefore equation (2.11) admits an algebra L 2 with basis
dx dy dx 2 dy
According to Table 6.6 we can advance directly to the third step.
3rd step. Determination of the type of the algebra L 2. We have
Therefore the algebra L 2 is of type III in Table 6.4; for the correspondence to
be complete, we must change the sign of the operator X2, and then the basis
(2.12).Y —χ *— [ xii χ = χ d V ^
2
dx dy ' Ox 2 dy
has exactly the structure of type III in Table 6.4.
4th step. Determination of the integrating change of variable. H ere X\ is the
operator of the group of projective transformations of Table 6.1. We shall
use the change of variables given there (interchanging t and u):
(2.13) < = — , u= - >

χ χ
After this change of variable, the operators (2.12) have the form
d
x —JL —J LJ L- L
χzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJ
the difference between them and the corresponding operators of type III in
Table 6.4 (the factor 1/2 in X2) are of no importance. Excluding the solutions
of (2.11) of the form
(2.14) y=Cx,
we write equation (2.11) in the form
• ±1 + J- = 0.
115
t\ (C\ t— 1). If Ci = 0 here, then

F rom this, integrating once, we obtain «' = zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPO
(2.15) U __£ .4.C,
and if Cx φ 0, then
(2.16) U
= - L - | 1 i n l d t —l l + C,.
5th step. Solution in the original variables. Substituting into (2.15), (2.16) the
values (2.13) of the variables t, u, we obtain
(2.17) y
(2.18) C\ y 4 C2x 4- χ In + 6 7 = r 0.
CH AP TER III
GROUP- THEORETICAL CLASSIFICATION OF SECOND- ORDER EQUATIONS
"I n a short report to the Scientific Society of G ottingen (3 December 1874)

I presented, inter alia, an enumeration of all continuous transformation groups
in two variables x, y and emphasized, in particular, that this can be used as a
basis for classification and for a rational theory of integration of all
differential equation s/ (x, y, y', ..., y(m)) = 0 that admit a continuous group
of transformations. Later on I carried out the details of the programme
outlined there." (S. Lie [3], p. 187).
Below we present the results of the realization of that programme as they
pertain to second- order equations. Our restriction to second- order equations
is not due to any limitation of the method, but to the desire to concentrate on
specific material and to provide exhaustive results.
§3.1. Equations admitting a three- dimensional algebra

The basis of group- theoretical classification of ordinary differential
equations is the enumeration of all possible Lie algebras (1.7) on the
(x, j)- plane. The basis operators of all the algebras are maximally simplified
by means of a suitable change of variables (2.14); algebras related by such a
change of variables are called similar. Enumeration of algebras is made up to
similarity, since equations admitting similar algebras are equivalent in the sense
that they can be reduced to each other by a change of variables.
The group- theoretical classification [3] has a particularly simple form in the
case of second- order equations of interest to us (the classification result for
second- order equations is presented briefly and clearly in [24], §3). In this case
the admissible algebra (by which we mean the maximal algebra admitted by a
given equation) can be only of dimension 0, 1, 2, 3, or 8. D imension 0
means that the equation does not admit a group of point transformations;
N.Kh. Ibragimov
y" = ey + xy, y" = (y') 2 + xy, y" = 6y2 + x,

examples of such equations arezyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFE
3
y" = 2y + xy+C. D imensions 1 and 2 were considered in the previous
chapter. Therefore we start with a discussion of equations that admit a three-
dimensional algebra.
3.1.1. Classification in the complex domain.
Lie performed his classification in the complex domain, using, where necessary,
complex changes of variables or of algebra bases. Enumeration of all non-
similar (with respect to a complex change of variables) algebras in the plane,
derivation of the corresponding invariant second- order differential equations,
and methods of their solution, are all discussed in detail in the book [2],
Ch. V. In Ch. 22, §3 and Ch. 24, §3 Lie gives a summary of the results of
classification of three- dimensional algebras and the invariant equations; these
results are reproduced here as Table 6.8. Let us see how an equation in that
table is constructed, given that it admits a fixed three- dimensional algebra.
F or all three- dimensional algebras the construction of an invariant equation
is the same; it requires the solution of the defining equation (1.30) with
respect to the unknown function/ (x, y, y') for known coordinates ξ and η of
basis operators of the algebra L 3 being considered. Therefore it suffices to
demonstrate the construction for one of the algebras. As such an example, let
us choose in Table 6.8 an algebra with the basis
(3.1) X l = _
F or the operator Χχ we have ξ = 1, η = 1. On substituting these values

of ξ, η , the defining equation (1.30) assumes the form df/ dx + df/ dy = 0,
whence
f=f(x- y, y')-
Now we need to substitute into the defining equation the obtained value of/
and the coordinates ξ = χ, η = y of the second of the operators (3.1). As a
result, we obtain the equation zfx+f = 0, where ζ = x—y. The solution of
this equation has the form
ir
x— y
with an arbitrary function g(y'). Finally, the substitution into (1.30) of the
function / of (3.2) and of the coordinates ξ = χ2, η = y2 of the operator X
leads to the following equation for the function g(y')'-
Group analysis of differential equationszyxwvutsrqponmlkjihgfedcbaZY
117
F rom this we have

n
(3.3) g"- 2(s'+ Cy" + y"h C= const.
F ormulae (3.2), (3.3) give us the desired right- hand side of the invariant
equation. Thus, the algebra L 3 with basis (3.1) admits the equation
y
v
- +
• l'y—- - %—
2zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIH
= o
x—y
with an arbitrary constant C.
3.1.2. Classification over the reals. Isomorphism and similarity.
The complex classification of Lie is completely sufficient from the point of
view of integration methods. However, it is also of interest to consider the
results of the classification over the reals, since we are dealing with real
differential equations. We state these results as they pertain to second- order
equations with a three- dimensional algebra. They supplement the complex
classification of Lie; details can be found in [25].
Table 6.7 gives a feeling of the differences between classifications in the real
and the complex domains. There we quote the well- known enumeration, due
to Bianchi (see [26], or, for example, [27], §10 and [28], §24), of all possible
structures of three- dimensional Lie algebras that are non- isomorphic over the
reals. In that table we indicate the values of the three different commutators
of the basis, which is what defines the structure of each algebra L 3 .
Two algebras are isomorphic if we can choose bases for the algebras such
that the structure constants of the two algebras in these bases are equal. As
can be seen from Table 6.7, two real non- isomorphic algebras can turn out to
be isomorphic over the field of complex numbers (that is, when we allow
complex- valued basis transformations). In the case of two- dimensional
algebras there are two non- isomorphic structures (both over the reals and over
the complex numbers), which can be taken to be [ΑΊ, X2] = 0 and
[Xu X2] = Xi (see Table 6.4).
The concept of isomorphism is independent of the particular realization of
a Lie algebra as, say, an algebra of vector fields (or of differential
operators (1.7)). I n the latter case there is an additional way of transforming
one algebra into another, by a change of variables x, y. Algebras related by a
change of variables are called similar. Similar algebras are isomorphic (see
[8], §7.9); the converse, in general, is not true, as can be seen from Tables 6.4
and 6.8. I t is precisely similarity (and not simply isomorphism) that is of use
in group analysis, as a criterion of reducibility of one differential equation to
another by a suitable change of variables. N onetheless, establishing
isomorphism is important as a first (necessary) step for the determination of
similarity of algebras.
N.Kh. Ibragimov
It is both interesting and useful to give more detailed instructions for the
construction of similarity transformations. The main theorems on similarity of
(1)
groups of algebras were obtained by Lie ([13], p. 256]) and Eisenhart
([29], [30]). We shall now consider one of these in detail. First we need to
introduce the necessary notation.
L r and L r be two r- dimensional algebras of operators of the form
1
(1.11) in the «- dimensional space of variables χ = (x , ..., x"). As we have
said already, for algebras to beîmilar they have to be isomorphic. Therefore
let us take the algebras L r and L r to be isomorphic, write them down in the
variables χ and x, respectively, and choose bases
so that the structure constants of the two algebras in these bases are the same.
We shall be interested in the case when r > n, while the ranks of the matrices
II ξαΙΙ, ΙΙξαΙΙ are the same and equal to n. Renumbering the indices, if
necessary, we shall assume that this is the value of the rank of the square
matrices || %f, \ \ , ||ξ£ |[ (i, h = 1, ..., ri). Then the following conditions hold:
(3- 4) Ι Ζ - φ , ν . Ι»'- Φ*V (»- 1 "; / "- «+ 1 r) ,
where (pph, φ ρΗ are functions of χ and x, respectively, and where we sum over
h from 1 to n. Under the assumptions made, we have the following theorem.
Theorem 3.1. For algebras L r and L r with the same structure and the same
number of variables x', x' to be similar, it is necessary and sufficientthat the
functions <pph, <pph satisfy the system of equations
(3.5) ψ / ( i )=<(>/(• *) (Λ= 1 //; p= w+ l, . . . . r)
and that this system does not lead to any relations either amongthe variables x'
l
or amongthe variables x .
Sketch of the proof Similarity of the algebras under consideration means that
there is a non- degenerate change of variables x' - f\ x), such that
(3.6) l^^la"- ^- (*= • ! n;a= 1 r).
From this and (3.4) we have n{r—n) equations (3.5).

Necessity. If the algebras L r and L r are similar, then, clearly, equations (3.5)
are compatible; by eliminating xl from them it is impossible to obtain a
relation among the x\ and vice versa.
(1)
Eisenhart used the term "equivalent groups". We have used "similar groups" (and
"similar algebras") in view of what is said in the first paragraph of §3.1. Two different
words are used in the Russian text for "similar" and "equivalent". (Ed.)
Group analysis of differential equationszyxwvutsrqponmlkjihgfedcbaZYXWVU
119
Sufficiency. F or α = 1, ..., η let us write the equations (3.6) in the explicit

form
(3.6') - ^îSWlSix) ( *, *~ 1 «).
where || ξ*Α || is the square matrix inverse to || ξ Α ' ||, that is,
U sing the conditions (3.5), it can be shown that the system (3.6'), and thus
the system (3.6), is compatible (a detailed treatment of this non- trivial step in
the argument can be found in [30], §22). This means that there is a change of
variables that establishes the similarity of the algebras L r and L T .
Determination of the change of variables. The similarity transformation is
determined by solving the mixed system of equations comprising functional
equations (3.5) and differential equations (3.6') (which are integrable in view
of (3.5)) with respect to x'.
Example 1. Let L 3 and L 3 be the first and the second algebras of Table 6.8.
They are complex isomorphic: their structures are the same and are of
type IX in Table 6.7 if for L 3 we use the basis
*i — — (Χι + Λ Ά ) , X2 — iX.t, AT3 —- ^ - ( Xj — ΑΛ3).
Thus, let us consider three- dimensional algebras L 3 and £ 3 in the plane with
the same structure, having, respectively, the bases
Jt1==(l +
* ) * + «„ " χ, = , * _ „ *
ox dy - dy * dx
x + { i + t )
and
Equations (3.4), written in the form
gives us in this case

_ i s ι * . *
2 2
i H|- ι/* = χι/ φ* + χφϋ; _L (1 _·^*) ~ T _ (j .4.
N.Kh. Ibragimov
Hence we have
With these functions φ Λ and φ Α , the system (3.5) can be written in the form
£y _ y x- j- y __ 1
. 1 - fzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
1 — ί^ ζ * 1 — £g χ
and we have the change of variable
l £
,._ —V „- ,±±IL·
which establishes the similarity of the algebras L 3 and L3 in the complex
plane. In this example, in order to find the change of variable, it was enough
to solve equations (3.5).
Example 2. Let us consider the algebras L 3 and L 3 with bases
Xl
~'W i'~dy~ > χ
*- ^ χ- δχ"τι'Ί Ϊ' " ~ " ~dx~~ r!f"dy~
Λ 3
and
They have the same structure (see Table 6.8). Equations (3.4), written as in
the previous example, have the form
Hence
φ ' = — xy, <ft
With these functions φ Α and cpA, the equations (3.5) assume the form
x- =xy, 2x—x+y
while after eliminating χ they lead to the relation
By Theorem 3.1 the existence of this relation between χ and y shows that
these algebras L3 and £,3 are not similar (neither over the reals, nor over the
field of complex numbers), though they are isomorphic.
§3.2. The general classification result

By going through all non- similar Lie algebras of all possible dimensions
and through all the invariant equations, it is found that if a second- order
equation admits an algebra of dimension more than 4, then in fact it
admits an 8-dimensional algebra. But all such equations are linearizable

(Theorem 1.6) and equivalent to the equation y" = 0. This classification
result is quoted in Table 6.9. See also [31].
All the equations of that table, apart from those in the first line, and all
the equations obtained from them by an arbitrary change of variables (2.9)
can be integrated in quadratures by the group method. Therefore the set of
such equations is infinite and depends on four arbitrary constants, two
arbitrary functions of one variable (see Table 6.9), and on arbitrary functions
of two variables which arise as a result of the change of variables (2.9).
§3.3. Two remarkable classes of equations

The two classes of equations we consider below arise in a natural way from
the foregoing classification; they deserve to be studied in a more detailed
manner than we have done up till now.
3.3.1. The equation y" + F3(x, y)(y')3 + F2(x, yKy' f + Ft(x, y)y' + F(x, y) = 0.
Linearizability criteria.
In practical situations of solving differential equations it is useful to have
simple criteria of their linearizability. In such questions, the following result
of Lie can be used with success.
Theorem 3.2. The following statements are equivalent:
(1) The secondorder ordinary differential equation
(3-7) </'WU. y. y')
is linearizable by a point change of variables (2.9);
(2) equation (3.7) admits an %dimensional Lie group;
(3) equation (3.7) has the form
(3.8) #"+f,(x, y)i)':'+f<'Ax. y)y'+F,(x, n)y'+F(x, y)0
with coefficients F3, F2, Fu F satisfing the compatibility conditions of the
auxiliary system
l 2>
dx " * dy '
dz . „„ 1 dF. . 2 dFt
(4) equation (3.7) admits a twodimensional Lie algebra with a basis X\,
such that
(3.10)
N.Kh. Ibragimov
1. Let us prove that the equation

ExamplezyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
(3.11) y"+F{x,
where the function F is non- linear in y, cannot be linearized.
Indeed, in this case equations (3.9) have the form
zx—z2—Fw+Fv. Zy=—zw,
(3.12)
wx=zw, wv=—w*.
One of the compatibility conditions, namely zxy—zyx = 0, leads to Fyy = 0.
Therefore for any function F that is non- linear in y the system (3.12) is
incompatible.
Example 2. Let us determine when can we linearize equations of the form
(3.13) yy ' ^
By Theorem 3.2 we must restrict ourselves to the equations
(3.14) y" 4- 4 (A3y* + A2y'* + Aty' + Ao) = 0,

where the coefficients At are constant, since the function / depen ds only on y'.
F or the equation (3.14) the auxiliary system (3.9) has the form
A0Aa 2 A» „ . A. . Α-
Α.,, Aa A
w-| 2- 3
ar* 3 x* " xx x2 x2
x
F rom the condition zxy = zyx we have A2(2—Ai) + 9AoA3 = 0. Similarly,
from the condition wxy = wyx we obtain 3A3(l + A\ )—A\ = 0. Writing
A3 = —a, A2 = —b, we have then
Therefore we have the following assertion (see [25] and the references
contained therein).
Equation (3.13) is linearizable if and only if it has the form
with α and b constant.

3.2.2. Equations^" + a(x)y'+ b(x)y = c(x)y~ 3.
In the equation
(3.15) y"=Cy\ C= const,
Group analysis of differential equationszyxwvutsrqponmlkjihgfedcbaZYX
123
of Table 6.9 let us replace y" by a general second- order differential expression
with variable coefficients. Dividing by the coefficient of the highest- order
derivative, we obtain the equation
(3.16) y"+a(x)y'+b(x)y=c(x)y- \
Its particular case,
(3.17) y'+bW y^ky- *, fc= const,
has long been discussed in the literature [32] —[35] as an example of a non-
linear equation with exceptional properties. It turns out, first of all, that the
general solution of (3.17) can be expressed in terms of solutions of the
corresponding linear equation
(3.17') y"+b(x)y- Q.
N amely, the solution of equation (3.17) with arbitrary initial data
y(xo) = }Ό Φ 0, y'(x0) = yi has the form
where u, υ are linearly independent solutions of equation (3.17') such that

= =
K(*O) JO> «'(*<)) Ji> °(*o) = 0, o'(x0) Φ 0, while W is the Wronskian:
W ·- »· MO' — t)t/ = const Φ 0. Secondly, since equation (3.17) is in this manner
"linearizable", it satisfies a "superposition principle", but of course in a non-
linear form (see [35], pp. 154, 389). Equation (3.17) is also of importance in
applications; it appears in the study of travelling wave type solutions of the
Schrodinger equation [36], in the theory of elasticity (see [35], §3.3), and so on.
The peculiarities of equation (3.17) mentioned above are related to the fact
that for an arbitrary coefficient b{x) and constant k Φ 0 it admits a three-
parameter group, and, moreover, reduces to equation (3.15) by a change of
variables. The latter equation is easily integrated with the help of invariant
solutions. Such a group- theoretical viewpoint allows us to generalize the class
of (3.17) and to find in (3.16) all the equations having the same properties
as (3.17).
3.3. Equation (3.16) admits a L ie algebra
TheoremzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCB
(i)Lsifc(x) = 0,
(3.18) (ii) L 3 ifc(x) = k exp(- 2ja(x)dx), k = const Φ 0,
(in) LQ in all the other cases.
Theorem 3.4. Equation (3.16) can be reduced to canonical form (3.15) if
condition (3.18) holds. T he corresponding change of variables has the form
where fix) is any non- zero solution of the equation

f"+a(x)f'+b(x)f=0.
124zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
N.Kh. Ibragimov
CHAPTER IV
ORDINARY DIFFERENTIAL EQUATIONS WITH A FUNDAM ENTAL SYSTEM

OF SOLUTIONS (FOLLOWING VESSIOT- GULDBERG- UE)
Continuing our discussion of problems of integration of ordinary

differential equations, let us dwell on the following question: what equations
(apart from linear ones) have a fundamental system of solutions, so that the
problem of construction of the general solution reduces to finding a finite
number of particular solutions? This problem has no direct relation to the
admissible group, but, as Vessiot [41], G uldberg [42], and Lie [43] showed, it is
also solved using group theory.
§4.1. The main theorem

We say that a system of ordinary differential equations
(4.1) - ^ = zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
F* (i, x\ · ·. , x"), < = 1 n,
has a fundamental system of solutions if the general solution of the system

can be expressed in terms of a finite number m of arbitrarily chosen particular
solutions
(4.2) xk=(xkl,...,xk·), ft- 1,..., m,
by expressions of the form
(4.3) x'= fp'(x,, .... xm,C,, C),
which contain η arbitrary constants C\ , ..., Cn. In this case the particular
solutions (4.2), which are taken to be functionally independent, are called a
fundamental system of solutions of (4.1). It is required that the form of the
expression (4.3) be independent of the choice of the specific particular
solutions (4.2). However, this does not exclude the possibility that a certain
system of equations (4.1) admits a number of distinct representations (4.3) of
the general solution with a different number m of the required particular
solutions (see Example 6 below). The general form of equations with a
fundamental system of solutions was determined by Lie, who proved the
following basic theorem (its proof, together with a thorough preliminary
discussion and examples, can be found in [43] Ch. 24).
Theorem 4.1. The system of equations (4.1) has a fundamental system of
solutions if it can be represented in the form
(4 4)
· Τ =
such that the operators
(4.5) Χα = 1
Group analysis of differential equationszyxwvutsrqponmlkjihgfedcbaZY
125
form an r- dimensionalL ie algebra. Moreover the number m of the necessary

particular (fundamental) solutionszyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONML
(4.2) satisfies the condition
(4.6) nm&tr.
§4.2. Examples
Example 1. F or a linear homogeneous equation
dx/ dt=A{t)x
we have η = I, m = I, r = I, X = xd/ dx. The representation (4.3) of the

general solution in terms of the particular solution x\ has the form χ = Cx\ .
Clearly, condition (4.6) is satisfied as an equality.
Example 2. F or the linear non- homogeneous equation
dx/ dt=A(t)x+B(t)
we have η = 1, m = 2, r = 2, so condition (4.6) is also an equality here.
The operators (4.5) here are X\ = d/ dx, X2 = xd/ dx, and their commutator is
[X\ , X2] = X\ , so they do indeed generate a two- dimensional Lie algebra.
(4.3) here is the representation χ = x\ + Cx of the general solution JC in terms
of the particular solution x\ of the non- homogeneous equation and a
particular solution 5c, written as a difference of two solutions of the non-
homogeneous equation: χ = x2—x\ .
Example 3. An example of a non- linear equation with fundamental solutions
is the Riccati equation
(4.7) dx/ dt=P(t)+Q(t)x+R(t)x2.
It has the special form (4.4) with r = 3 and the operators (4.5) given by
(4.8) 1 l * X2 * *3 « *
X
dx ' dx ' dx '
which generate a Lie algebra L3. F ormula (4.6) means that m ^ 3, so to
represent the general solution of the Riccati equation with arbitrary
coeffcieints we need at least three particular solutions. I n reality it suffices to
know three solutions, as any four solutions of the Riccati equation are related
by the condition that their cross- ratio be constant. In §4.3 we shall return to
a discussion of th e properties of the Riccati equation from the group-
theoretical point of view.
Example 4. Let us consider a homogeneous system of two linear equations,

dxldt=au(t)x+ati(.t)y,
dyldt=all(t)x+aa(t)y.
N.Kh. Ibragimov
It has the special form (4.4) with coefficients
£ .= (*, 0). | 2 = ( j/ , 0 ) , |, =
Therefore in this case the operators (4.5) are
which, as is easily verified, generate an algebra L 4 . The representation of the

general solution
in terms of the two particular solutions (JCI, zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONM

y\ ) and (x2, y2) is the form that
(4.3) takes in this case, so we have m = 2. Condition (4.6) is an equality, as
η = 2, r = 4.
Example 5. In the case of a non- homogeneous system
dzldt=a, t (Ι)τΑ- α,Λί) y+bt (<),
dy/ dt=a2i (t)x+aiZ(t)y+bt(t)
one has to add to the coeffcients T a, ξ α of the previous example
7> = 6, ( O t T,- bt{t), |r, = ( 1. 0) . £ e = ( 0, 1),
so that the algebra £4 extends to an algebra L(, by adding the operators

X5 = 9/8x and X6 = Q/ Qy. Thus « = 2, r = 6, and it follows from (4.6)
that to represent the general solution of the above non- homogeneous system
with arbitrary coefficients a*/ (i) and £,- (/) we have to know three particular
solutions (xk, yk) k = 1, 2, 3; the corresponding formula (4.3) is well known:
Example 6. In his book [43] Lie gives an example of a non- homogeneous

system of two linear equations
dxldtâtl(t)y+bs(t),
(4.9)
dyldt=- at.M)x+bAt),
in which one can make do with two particular solutions. This pecularity is of
group- theoretical nature and consists in the following. By Theorem 4.1,
corresponding to the system (4.9) we have the operators ΛΊ = yd/ Qx—xd/ Qy,
X2 = 3/9x, X3 = d/ Qy, which generate a group of motions (rotations and
translations) in the (x, j)- plane. As the motions conserve distances between
any two points, any three solutions (xh y\ ), (x2, y2), (x, y) of ths system are
conneted by the relations
127
Thus, the system (4.9) admits two representations of the general solution: as a
linear superpositon of three solutions (Example 5), and as a "non- linear
superposition" of two solutions if the equalities (4.10) are solved for x, y.
§4.3. Projective interpretation of the Riccati equation

Theorem 4.1 allows us to find all the ordinary differential equations (4.1)
with a system of fundamental solutions by reducing the problem to that of
enumeration of all possible transformation groups with a finite number of
parameters (or of the corresponding finite- dimensional Lie groups of operators
(4.5)) in the «- dimensional space of variables χ = (χ1, ..., χ"). Such an
enumeration has been performed by Lie for the case of the real line (n = 1)
and of the plane (n = 2) (these results can also be found in [44], Ch. IV).
H ere we confine ourselves to the case η = 1. In this case the choice is very
restricted: every group of transformations on the real line coincides (up to a
change of variables) with the group of projective transformations generated by
the three- dimensional Lie group with basis (4.8), or with one of its subgroups.
This means that for η = 1 the Riccati equation (4.7) is (up to a change of the
variable x) the most general equation with a fundamental system of solutions.
For η ^ 2 such equations are much more numerous (see [45] —[47]).
Thus the Riccati equation is in a sense a realization of the group of
projective transformations. A reflection of this fact is the theorem of the
constancy of the cross- ratio of any four solutions of a Riccati equation.
To prove it, we introduce homogeneous coordinates u, D by setting
χ = u/ υ. Then the Riccati equation assumes the form
As the definition χ = u/ υ contains two functions, we can constrain them by

one additional condition, for example, by requiring that the equation
du/ dt- (l/ 2)Qu- Pu = 0 is to hold. Then (4.7') reduces to the system of two
linear equations
=...1 Q(/ ) u. t P{t)v, - ^ = - Λ ( 0 »- 4 Q ( ' ) " ·

*LzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
which is the form the Riccati equation takes in projective space. Let (u\ , v>\)
and (w2, υ 2) be two particular solutions of this system, chosen so that the
ratios «i/ i)i and u2/ o2 are not equal to the same constant. Then
so the general solution of the Riccati equation (4.7) is given by the formula
c u
(4.11) j —- i i- rC2ut __ ut + Kut
N.Kh. Ibragimov
X\ , ..., Χ4 be four solutions of the Riccati equation corresponding to some
particular values K\ , ..., AT4 of the constant Κ in (4.11). Then we have the
equality
x
,/ A0, *ι — *Ϊ . i — x* A\ — K 2 Κ ι— Κ4
4 1
ί · ^) ——— · r
__ , — κ —κ ' κ — κ '
which proves that the cross- ratio of any four solutions of the Riccati equation
is constant.
§4.4. Linearizable Riccati equations

Thus, by Theorem 4.1, to each Riccati equation (4.7) there corresponds a
Lie algebra L r, which is either a three- dimensional algebra with basis (4.8), or
(for particular values of the coefficients P, Q, R) some subalgebra of that
algebra (in this case r = 2 or r = 1). This algebra allows us to find out
easily whether there is a linearizing change of the variable χ for a given
Riccati equation.
Theorem 4.2. If a Riccati equation (4.7) has one of the following four
properties, it also has the other three:
(1) equation (4.7) is linearizable by a change of variable x;
(2) equation (4.7) can be written in the two- term form
(4.13) dx/ dt=T t (t) ξ, (χ)+ 7 1 , (01= (*)
so that the operators Χχ = ξι(χ)ά/ άχ, Χ2 = ξ2(χ)ά/ αχ generate a two-
dimensional Lie algebra, that is, [Xu X2] = at,Xi + $X2 (for [Χι, X2] = 0 the
algebra is one- dimensional, and in the Riccati equation we can separate the
variables);
(3) equation (4.7) has either the form
(4.14) dx/ dt=Q{t)x+R(t)x\
or the form
(4.15) dxldt=P(l)+Q(t)x+k[Q(t)- kP(t))xz
with some (in general complex) coefficient k;
(4) equation (4.7) admits a constant (in general complex) solution.
Remark. Equation (4.15) has a constant solution χ = —\ / k. Therefore the
linear equation, which is a particular case of (4.15) for k = 0, can be
considered as a Riccati equation, which has the point at infinity as its
constant solution.
129
C H AP TER V
THE INVARIANCE PRINCIPLE IN PROBLEM S OF M ATHEM ATICAL PHYSICS
Special types of exact solutions, widely known now as invariant solutions,

have long been successfully applied to the analysis of specific problems and
were in accepted use in mechanics and physics prior to the development of
group theory, becoming in the process a sort of folklore. Lie uncovered the
group- theoretical meaning of these solutions and considered the possibility of
integrating partial differential equations using invariant solutions, given a
sufficiently large group of symmetries ([38], Chapters III and IV).
Subsequently group theory made it possible to understand, clarify, and extend
many intuitive ideas, as a result of which the method of invariant solutions
could be included as an important integral part of contemporary group
analysis. It was precisely through the concept of an invariant solution that
the theatre of action of group analysis shifted in the sixties from ordinary
differential equations to problems of mechanics and mathematical physics
(largely due to the books [38], [39], [27], [18], [40]).
In the passage from ordinary to partial differential equations it becomes
impossible (barring rare exceptional cases), as well as practically futile, to write
down a general solution of the equation. Instead, mathematical physics deals
with solutions that satisfy a priori fixed side (initial, boundary, and so on)
conditions. In the solution of many problems of mathematical physics the
following semi- empirical principle (it has been rigorously justified so far only
in limited cases; see [8], §29, [38], §89, [48], [49]) may be used with success:
Invariance principle: if a boundary- value problem is invariant with respect to a
group, then one should look for a solution in the class of functions invariant with
respect to that group.
Invariance of a boundary- value problem presupposes invariance of the
differential equation, as well as of the manifold containing the data, and of
the data given on that manifold. If the boundary conditions violate invariance
(which happens frequently), the invariance principle is useful in conjunction
with other methods. Thus, for example, as we show below in §5.2, Riemann's
method reduces the Cauchy problem with arbitrary (and therefore not
invariant) data to a particular G oursat problem, which turns out to be
invariant, and can therefore be solved by applying the invariance principle.
§5.1. Spherical functions

Let us consider Laplace's equation Au - 0 in three independent variables
x, y, z, and let us find its solutions that are invariant with respect to rotations
and stretchings in the independent variables and the function u.
It is convenient to change to spherical coordinates
x—r sin θ cos φ, y=r sin θ sin φ, z=r cos 0
N.Kh. Ibragimov
(0 =ξ θ < π , 0 =ζ φ < 2π) and write Laplace's operator in the form
where we denote by A2 the Laplace—Beltrami operator on the unit sphere:
Δ2 sin 0 - ι
~ sine ae """ *ίηΗ)Ί ψ'
A solution that is invariant with respect to spatial rotations (spherically
symmetric) has the form « = R(r). F or it Laplace's equation leads to
(r2R')' — 0, whence we have
R= - ^L + C2.
F or the group of stretchings in u with symbol X = «9/ 6M we have as
invariants any function of x, y, z. Therefore no solution u = φ (χ, y, z) can
be invariant with respect to this group, as u is n ot an invariant.
Let us find invariant solutions for the group of stretchings in the
independent variables and the function u:
(5.2) x—ax, y=ay, z—az, u=a"u.
A function u = φ (χ, y, z) that is invariant with respect to the stretchings (5.2)
is called homogeneous of degree n. Thus, let us consider solutions of Laplace's
equation that are homogeneous of degree n. In this case, keeping in mind the
expansion of an arbitrary solution in homogeneous solutions, we shall take η to
be an integer.
Writing the operator of the group of stretchings (5.2) in spherical
coordinates in the form
d d
ν ,
Λ = r ——f nu - r— ,
or ou
we determine from the equation XJ = 0 a basis of invariants λ = ur~ n, θ, φ .
The invariant equation λ = Y n(Q, φ ) (the index η means that for each choice
of η we have a different function Y) provides us with the following general
form of an invariant solution:
(5.3) u=rnY n(Q, φ ).
Then
ΔΗ = ~ I - ^ r (nrnnY,,) + Γ "Δ 2 7 „ 1 = r"- * [n (n + 1) Υ n + Δ 2 Υη ],
and Laplace's equation assumes the form
(5.4) A 2 y u + n ( n + l ) F »= 0 .
Therefore the coefficient Y n of the invariant solution (5.3) is an eigenfunction of
the Laplace —Beltrami operator Δ2 (with eigenvalue n(n+ 1)); it is called a
spherical function of order n.
131
Equation (5.4) is invariant with respect to the change η = — (m + 1), since

n(n+ 1) = m(m+ 1). Therefore, if (5.3) is a solution, we also have a solution
of Laplace's equation given by the function
+ 1)
(5.3') r- <" y,.(e,<p).
A series whose terms are invariant solutions (5.1), (5.3), (5.3') with constant
coefficients,
(5.5) «( Γ, β , φ )= = ^ β + A) + J ^ n r
» +
_^L- jΚη(θ, φ),
is a (formal) solution of Laplace's equation in view of the superposition

principle (also a group property). The formula (5.5) gives a representation of
the general solution and is usually derived by the method of separation of
variables [50].
§5.2. A group- theoretical touch to Riemann's method

This section is of a synthetic nature, being a result of combining the
method suggested by Riemann [51] for integrating second- order linear
hyperbolic equations with Lie's [52] classification of such equations. In [53]
there is a detailed analysis of known methods of constructing a Riemann
function (called in [53] the Riemann—G reen function) for particular types of
equation. Six such methods are described there. The group- theoretical
approach presented below is a seventh.
Riemann's method reduces the question of integrating the equation
(5.6) L\ u] ssm^+aix, y)ux+b(x, y)ut+c(x, tf)u=- f{x, y)
to that of constructing the auxiliary function o, which satisfies the adjoint
equation
(5.7) L*\ v)s*i>xt-
and the following conditions on the characteristics:
ν χ
| α(*., η)<ίη j
(5.8) v\ x=Xt=e* , » | ^ . = e*.
If we find the function υ, then the solution of the Cauchy problem
(5.6') £ ( «] = / , »|τ
with data on an arbitrary non- characteristic curve γ is obtained from the
formula (see F ig. 4 on p. 132)
+ u(B)v(B)]+ C - ίΓ —ι;,^ + ibv jvx\ u\ dx —
AB ^ ' ·*
r 1 / 1 \ I , 1 , ff , , .
— \ - T vuv- \ - lav — - T vu)u\ dy\ + \ \ vjdxdij,
132 N.Kh. Ibragimov
where the double integral is taken over the domain bounded by the
characteristics χ = XQ, y = JO and the curve γ. The function υ is called the
Riemann function, while the boundary- value problem (5.7), (5.8) which defines
this function is called the characteristic Cauchy problem, or the Goursat
problem.
if ι
Fig. 4. On Riemann's method.
Example 1. F or the wave equation uxy = 0 the corresponding G oursat

problem has the very simple form
Clearly its solution (solutions of the G oursat problem are unique) is the
function ο = 1.
Example 2. The telegraph equation
(5.9)
is one of the simplest types of equations (after uxy = 0) to which Riemann's
method is applicable. The G oursat problem (5.7), (5.8) in this case has the
form
(5.10) ν (, 1, • 1.
Usually, textbooks offer the following "m ethod" for solving it: let us look for
a solution of the problem (5.10) in the form
(5.11) v^V(z), 2
= (x- ar 0 ) ( # - ». ) -
Then we obtain the ordinary differential equation
which is a form of the Bessel equation: passing to the standard form

µν"+ V + µν = 0 is done by using the change of variable µ = \ / 4i.
Group analysis of differential equationszyxwvutsrqponmlkjihg
133
Thus the Riemann function for the telegraph equation (5.9) is expressed in
terms of a Bessel function in the form
(5.12)
Example 3. Riemann [51], §9 applied the method he suggested to the equation
(5.13) i;xy - ; ———- ν = 0, I = const.
{% ~i" y)
For this equation the conditions (5.8) on the characteristics are
(5.13') r ' U *. = l, P | I - * = 1 ·
Riemann reduces the resulting characteristic Cauchy problem to an ordinary
differential equation (which defines a special G auss hypergeometric function)
by assuming that Ό is a function of the one variable
(bii) (χ—*0)(y- y0)
( 5 14 ) l
' - (ζ, + νΜχ + ν) '
In the last two examples we could reduce the construction of the Riemann
function to the solution of an ordinary differential equation by having guessed
correctly the form of the solution of the G oursat problem. It turns out that
the ability to perform such a reduction is related to the fact that the equation
admits a sufficiently large group, and to the invariance of the conditionson
the characterstics with respect to a one- parameter subgroup of that group. I t
is the invariance principle, and not a lucky guess, that gives us the correct
form of the solution. Let us consider the equations of Examples 2 and 3
from this point of view.
Example 2'. The telegraph equation (5.9) admits a three- parameter group (in
addition to stretchings of the variable u and the infinite group consisting of
addition to u of any solution of the equation; this group is common to all
linear equations) with generators
Let us find a linear combination of these operators
admissible for the G oursat problem (5.10). Let us insist first on the
invariance of the characteristics χ = x0, y = jo- This invariance criterion can
be written in the form
[X(x- xt) ]s= rts,â+ â:o= 0, [X (y- y0
Hence it can be seen that γ Φ 0 (otherwise α = β = 0). Therefore we can
set γ = 1 to obtain α = — x0, β = y0. The resulting operator
(5.15) X^- xJ- JL- ^- ^- A
N.Kh. Ibragimov
is obviously admitted by the G oursat equation (5.10). Therefore we can use

the invariance principle and look for a solution to the G oursat problem in the
class of functions invariant with respect to the one- parameter group with
the symbol (5.15). This group has two independent invariants, zyxwvutsrqponmlkjihgfedc
ο and
ζ = (x—xo)(y—yo)- Therefore the invariant solution has the form (5.11).
Example 3'. Equation (5.13) also admits three operators,
χ _d _d_ χ
_d_ J_ v- __ 2 _jL _ 2 "*
1 2 3
dx dy dx dy dx dy
As in the previous example, we can form from them a (unique) linear
combination
dx dy
which leaves characteristics and the conditions on the characteristics invariant.

Therefore we shall look for the solution of the problem (5.13), (5.13') in the
class of invariant functions. Since for the operator (5.16) the invariants are ο
and the expression
an invariant solution has the form ο = ο(µ). This is the invariant solution
found by Riemann; the variable ζ he used is related to the invariant (5.14) by
the functional relation ζ = µ/(1 — µ) and therefore is also an invariant.
Let us now move on to general statements. The starting point here will be
Lie's result [52] on the group classification of homogeneous equations (5.6).
Most convenient for our aims is the invariant (with respect to equivalence
transformations of linear equations) formulation of this result due to
Ovsyannikov [54] (see also [8], §9.6). Let us observe that in the assertions
below the admissible group is considered modulo addition to the dependent
variable of an arbitrary solution cp(;t, y) of the linear equation being
considered, that is, by an admissible algebra we mean a factor- algebra modulo
the ideal generated by the operators X = <p(x, j)6/ 8«. So let us start with the
following result of group classification.
Theorem 5.1. The equation
(5.18) t |i i ] e t t , + ( i ( i : , y)ux+b(x, y)uv+c(x, y)u=0
admits a four- dimensional Lie algebra if and only if the quantities
(5.19) p=k/ h, q=(ln\ h\ )Jh
(if h = 0, h and k have to be interchanged) are constant; it admits at most a
two- dimensional algebra if at least one of these quantities is non- constant. Here
we denote by h and k the Laplace invariants of equation (5.18):
(5.20) h
135
(5.18) reduces either to the form (if q = 0)

In the case of constant ρ and qzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFED
(5.21) uxy+xux- r- pyull+pxyu=O
and admits the operator
d d d
X = (a,./ - f- a 2 ) - j
X = (a,./ - f- a 2 ) - j^ - γ {— a,y + a s ) - ^ - — (aty + a3/ w ·+• a 4 ) u — ,
or to the form (if q Φ 0)
(5 22) U
· **~WT7^- ,1ϊΤ7) Β·+ TW+
and admits the operator
X = (ai- r- 1 + · a 2 j 4- a 3 ) - ^ - + ( ~ «ii/ 2 4- a 2 x — a s )
Corollary. The adjoint equation (5.7) admits a four- dimensional algebra if and
only if so does the original equation (5.18).
Indeed, in the mutually adjoint equations L[u] = 0 and L*[u] = 0 the
Laplace invariants h, k and Λ», fc, are related by a permutation: h = fc»,
k = Α.. This follows immediately from (5.7), (5.20). Therefore the
corresponding invariants (5.19) p, q and />*, q* are either constant for both
equations, or non- constant.
Remark. A general equivalence transformation of equations (5.18) has the
form
(5.23) *= / ( *) , y=g{y), a=>.{x,y)u.
The Laplace invariants are in reality invariant only with respect to a linear
transformation of the dependent variable u, without a change of the
coordinates x, y. On the other hand, the quantities ρ and q defined by (5.19)
are invariants of the general equivalence transformation (5.23) and therefore
arise naturally in the statement of Theorem 5.1.
N ow we can prove the main result of this section.
Theorem 5.2 [23]. Let the equation (5.6) have constant invariants (5.19). Then
the Goursat problem (5.7), (5.8) admits a one- parameter group and the Riemann
function can be determined from a second- orderordinary differential equation.
Proof. By Theorem 5.1 and its corollary, it suffices to consider equations (5.6)
for which the adjoint equation has already been reduced by a suitable
equivalence transformation to either of the forms (5.21) or (5.22), depending
on whether the invariant q* of the conjugate equation is or is not zero.
Suppose that q* = 0 and the adjoint equation has been reduced to the
form (5.21):
(5.21*) L*[u\ ^vxv+xvx+pyvt+p^xyv=Q, p*= con st.
N.Kh. Ibragimov
Then conditions (5.8) have the form

(5.21') «Ι,- »- **""·- ", v^- e"»1**- ",
and the Goursat problem (5.21*), (5.21') admits the operator
A" = (χ — Η) - ^ — (y — y0) - Q£ + ixo (y—y<>)—

This operator has the invariants
V== .w>t- «.<»- ».)- Kv*.u- x1.it a n d µ = ( χ —; e , , # )
By the invariance principle, the solution of the G oursat problem (5.21*),

(5.21') is sought in the invariant form
By substituting into (5.21*) we obtain the ordinary differential equation

(5.24) , t V"+ ( 1
for the function ν(µ), while the data (5.2Γ) on the characteristics beome the
initial condition F (0) = 1. So the construction of the Riemann function
reduces to solving equation (5.24) with this initial condition.
Suppose that q* Φ 0 and the adjoint equation has been reduced to the
form (5.22):
/ ?*. q+ —c o n st .
In this case conditions (5.8) are of the form
and the G oursat problem (5.22*), (5.22') admits the operator
X « ( r —x0) (x + i/ u) — - — [y — // 0) (:r(, - j- //) - | - +

2
τ x
Ρ* (· — o)—
+ — ΙzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLK
The invariants are
and an invariant solution is written in the form
^ - ι- ,v«) (χ. + y)'

137
As a result, the G oursat problem (5.22*), (5.22') is reduced to the solution of

the ordinary differential equation
(5.25) ,« (1 - µ ) η - + (1 - µ) [ 1 + j 2 ( P » + 1 > _zyxwvutsrqponmlkjihgfedcbaZYXWVU

ή µ ]V ' - f
under the condition F (0) = 1.

Example 4. Let us consider the equation
- — u — {), I s= const ?t 0,
which holds an "intermediate" position between equations (5.9) and (5.13).

The Laplace invariants (5.20) are h = k = —l/ (x + y), and the formulae (5.19)
give us ρ - 1, q = I/ [I (x + y)]. As q is not constant, Theorem 5.2 cannot be
applied.
F or practical use of Theorem 5.2 in the case of constant ρ and q, there is
no need to reduce the adjoint equation to the corresponding standard form
(5.21*) or (5.22*). It can turn out that the computations are easier in the
original variables. F or that we only need to find the admissible operators;
the corresponding algorithm is described in Table 6.10, and it will be easy to
remember after the following additional clarifications.
If for an equation (5.18) we are looking for an admissible operator of the
form
then, after a preliminary analysis of the defining equations, it turns out that
this operator must have the following special form:
(5.26) X = ξ (χ) - L. + η (,,) - ^ + a{x, y) u - jL .
Therefore the group admissible by equation (5.18) belongs in the class of

equivalence transformations (5.23). This is the first useful observation.
Subsequent analysis of the defining equations shows [8], §9.4 that the
operator (5.26) is admitted by (5.18) if and only if the following equations
hold:
(5.27)
The second useful observation has to do with these equations and allows us to
give them a clear geometric meaning.
Let us start by noting that up till now we have actively used the tangent
vector field of a one- parameter group in the form of a first- order linear
differential operator (1.7). If we now consider the close connection between
N.Kh. Ibragimov
linear partial differential equations and equations in differentials (equations of

characteristics), then it becomes natural to use, in addition to the operator (1.7),
itszyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
dual differential form
(5.28) 3 - i | dx- l dy.
Furthermore, the differentialform (5.28) will be considered as dual also with
respect to the operator (5.26); in this way we emphasize the fact that
transformations of x, y have a geometrical meaning, while the transformation
rule for the dependent variable u is defined by its physical interpretation, and
not by geometry.
Using the differential form (5.28), the defining equations (5.27) are written
in the form
(5.27') <ί(ΑΞ)= 0, βί(&Ξ)= 0, d(o+ai\ +bl)- *{h~k)3.
As every closed form is locally exact, the interpretation of the first two
equations (5.27) is that both Laplace invariants h and k (5.20) are integrating
factors for the differential form (5.28) dual to the operator (5.26). In view of
this, in particular, the form (A —k)Z is exact. Therefore the last of the
equations (5.27') can be integrated and gives us the coefficient σ of the
operator (5.26).
Example 5. Let us consider the equation
(5.29) uxv— —l— (ux f u,,) — / [x, y), Ι Φ 0, / Φ —1 .

·'· \ " If
For this equation ρ = 1, q = 2/ [l(l+ 1)] and Theorem 5.2 is applicable.

Let us write down the adjoint equation
( fts°) zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
+ {+ ) °
and the conditions on the characteristics
(5.31)
/ + y0 V / *o +
Let us compute the algebra admissible by equation (5.30). The Laplace

invariants for (5.30) are equal; they coincide with the invariants h = k for
(5.29) (see the corollary of Theorem 5.1). Therefore the first two equations
(5.27) reduce to the single equation
from which
139
After this, by integrating the third equation of (5.27'), which has the form
ά(σ + (ξ + τ\ )1/ (χ + γ)) = 0, we obtain σ = Ct+C\ l(y- x). Therefore the
operator admissible by equation (5.30) is equal to
[ClX* + C2x + Q - A. — {C^ — C,y + Q -

A' = zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA
and contains four arbitrary constants Q , ..., Cj, in accordance with

Theorem 5.1. Writing down the invariance condition for the values (5.31) on
the characteristics, we obtain
λ a . (X — Xe) (X + y0) _ _ (j, _ y0) (y 4. X() ) _ - i- / [(y _ y 0 ) _
Thus, the G oursat problem (5.30), (5.31) admits the one- parameter group with
the symbol (5.32).
An invariant solution has the form
Substitution into (5.30) gives us the following equation:
The two conditions (5.31) on the characteristics become one initial condition
V(0) = 1.
§5.3. Symmetry of fundamental solutions, or the first steps in group

analysis in the space of distributions
Continuing in the direction of the previous section, namely, application of
the invariance principle to boundary- value problems with arbitrary data by
reduction to an invariant problem of a special type, let us consider the
potential of group theory in the construction of fundamental solutions for the
three classical equations of mathematical physics. This natural path of
development of group analysis venturing into the space of distributions
(generalized functions) has been sketched in the pamphlet [22], where we can
find some additional heuristic considerations, which we omit here. On the
other hand, here we use infinitesimal techniques to a greater extent than
in [22].
5.3.1. S omething about distributions.
H ere we are considering the general «- dimensional case, so test functions φ (χ)
1
are infinitely differentiable functions of χ = (χ , ..., χ") with compact support.
N.Kh. Ibragimov
The action of a distribution / on zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFE

φ(χ) is denoted by < / , φ > or by
( / (*)» <p(x) > when different variables are in use.
Transformation of distributions.
n
Let there be given a transformation on fl? ,
(5.32) χ= Φ (χ),
00
where Φ is a C diffeomorphism. We denote the Jacobian of this
J
transformation by / = aet(Qx'/ Qx ). Let us define a transformation of
distributions by the equality
(5.33) <f(x), φ (*)> = < (7°Φ - 4 )(ί), (φ«φ- *)(χ)>,
which is an invariance condition for a functional. Let us start, as usual, with
the regular case, when / is locally integrable and
x)> = jj f(x)v(x)dz.
From the usual change of variables formula in the integral we have

/ (χ) φ (χ) dx = J (/ οφ- ΐ) (ΐ) (φ οφ ΐ) (*) | / 1 "i
n
R"
that is (see [57], p. 456)
(5.34) </ (*)
Comparison of (5.33) and (5.34) gives us the following formula for
transforming arbitrary distributions:
(5.35) 7 —Ι / Γ / or / =
Taking, in particular, / = δ , we obtain from (5.35) the following
transformation of the δ- function:
(5.36)
Let there be given now a one- parameter transformation group. Writing

instead of (5.32) the infinitesimal transformation χ' « χ' + αξ'(χ) of this
group and using the rule for differentiating Jacobians,
— «a. (ή,
we obtain the infinitesimal transformation (5.35):
(5.35') f«/ -
For the δ- function this has the form
(5.36')
Group analysis of differential equationszyxwvutsrqponmlkjihgfedcbaZ
141
The Leray form.

We consider a hypersurface in B?" defined by the equation
(5.37) P(x)= 0
with a continuously differentiable function zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQ

P(x), where grad Ρ Φ 0 on the
surface Ρ = 0. The Leray form ([55], Ch. IV, §1; see also [56], Ch. I ll, §1,
[57], p. 439) for this hypersurface is defined as an (n— l)- form ω such that
dPA&=dxlA., .Ada:".
At the points where Pj = dP/ dxJ Φ 0 it can be represented in the form
Using the Leray form on the surface (5.37) we can define the Dirac
measure δ (Ρ ) by
(5.38) <fi(P), φ> = [ φω.
If we denote by Q(p) the Heaviside function, or the characteristic function of

the domain Ρ ^ 0, that is,
0 for Ρ < 0,
ΐ for Ρ > 0,
then the distribution
(5.39) <Θ(Ρ), φ > = J φ (ι) da:
satisfies the differential equation

(5.40) θ'( Ρ )= δ ( Ρ ).
Differential equations with distributions.

H ere we do not deal with the question of describing the collection of all
solutions of arbitrary differential equations in the space of distributions (on
this see, for example, [56], Ch. I. §2.6 or [57], pp. 450- 452); we only note
those simplest equations that we need in what follows. One of them is (5.40).
F or η = 1 let us consider the first- order equation
(5.41) xf'=0.
Its only classical solution is / = const. In distributions it has two linearly
independent solutions: / i = 1 a n d / 2 = Q(x). Therefore the general solution
of (5.41) in the space of distributions has the form
(5.41') / -
N.Kh. Ibragimov
The D irac measure (5.38) satisfies the following equations:

(5.42) P6(P)= 0,
(5.43) i>6'(P)+ 6(P)= 0,
as well as an equation of arbitrary orderzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPO
m = 2, 3, ... :
(5.44) / )5(tn )( P ) + m 6< m -
Indeed, from the definition (5.38) we have
A·
that is, equation (5.42). F urthermore, if dP/ dx* Φ 0 (for some i it must be
true in view of the condition grad Ρ Φ 0), then from (5.42) by differentiation
with respect to χ' we obtain the equality
™6(Ρ) + Ρ6'(Ρ)Ά=>0,
ox ox
which on division by dP/ 5x' gives us (5.43). F urther differentiation leads to
the equations (5.44).
F rom (5.44) we can see that the first- order equation
(5.45) Pf(P)+ mf(P)= 0
has the general solution
(m t) m
(5.46) f*= Ci6 - (P)+ C2P- .
5.3.2. Laplace's equation.
It is well known that Laplace's equation
Διι= 0
is conformally invariant (moreover, it is the only conformally invariant linear

elliptic second- order equation [10]) and admits an algebra L with basis
d d d
ν γ J J
l l 1
dx ' dx dx '
(5.47) Y i = (2*V _ | χ |*6y) ±, + (2 - n) x'u ±- .
We shall regard the equation of a fundamental solution,

(5.48) Δΐι= δ ( *) ,
as a boundary- value problem, in which at a fixed point (at the origin) we are
given a δ- function singularity. Let us apply to it the invariance principle.
143
To that end, let us determine first which operators in L leave invariant the
singular point χ = 0. It is easy to see that this property is shared by all
operators (5.47) apart from the translation operators Xt (i = 1,...,»). F or
rotation operators X^ and for the stretchings Z\ and Z 2 it is obvious, while
the fact that Y t leaves the origin fixed, Y txk \ x=o = 0, is clear from the
k 2
equality Y tx = 2 χ ' χ *- |*1 δ ' *.
Let us return now to the equation (5.48). F irst of all, it admits the
operators Xtj, since the Laplacian and the δ- function are invariant with
respect to rotations (see (5.36)). As far as the operators Z\ and Z 2 are
concerned, we can form from them an admissible linear combination
(5.49) Z ^'jL + fruJ- .
To find the value of the constant coefficient k, let us write down the
infinitesimal invariance criterion of equation (5.48). F or that we need to
prolong the operator (5.49) to second derivatives, and also to add a
transformation of the δ- function according to (5.36'), in which in this case we
have Α- (ξ') = η. As a result we obtain the operator
(5.49') J W ^ +
/ ^ + (
*_ 1)U iJ L +
Therefore Ζ(Δ« — δ) = (k—2)Δ» + ηδ, and the invariance condition of the

equation has the form
Ζ (Δα - δ ) |Δ «- .= (*
that is, k = 2—n. Thus equation (5.48) admits the following operators:
(5.50) X ^ x * * - : ' * , zW- A. - t - t f- ft ) *.

dx dx dx "U
In a similar manner we can verify that the operators 7; are also admitted.
However, we do not need them and will regard them as excess symmetry of
the fundamental solution.
In accordance with the invariance principle, we shall look for a
solution of equation (5.48) that is invariant with respect to the group
with generating symbols (5.50). The rotation group has two invariants:
r = γ ( Λ 1 ) 2 + . . . + (xn) 2 and u. In these invariants the stretching operator of
(5.50) is written in the form
Z=r- ?-ar + (2 — n)u-

' du 4-
and supplies us with the invariant
N.Kh. Ibragimov
common to all the operators (5.50). By Theorem 1.3 an invariant solution can
be written in the form / = const, that is,
(5.51) a^C r 8 - ".
C = - .
By substitution into (5.48) we find the value of the constantzyxwvutsrqponmlkjihgfedcbaZY
r—- ,
(2 - n)iln
where Ω π is the surface area of the unit sphere in «- dimensional space.
Therefore the fundamental solution
has been determined from the invariance condition up to a constant multiple,

while equation (5.48) played the role of a normalizing condition.
5.3.3. The heat equation.
In this case instead of (5.48) we consider the equation
(5.53) Η / - Δΐϊ= δ ( ί, χ)
with η- dimensional Laplacian Δ in the variables x'. Obvious analogues of
(5.50) are the operators of rotations, G alilean transformations, and stretchings:
(5.54) ^ w J L - ^ J L ,
4r + Ά« » 4-
ι
dt βχ du
As in the case of Laplace's equation they are singled out from the full algebra
of symmetry of the heat equation by being admissible for the equation (5.53).
Now let us use the invariance principle. Invariants for Xy are t, r, u. In
the space of these invariants the operators Yj are written in the form
Therefore rotation and G alilean transformations have two common invariants,

t and ρ = er ^'. Rewriting the last of the operators (5.54) in these terms,
we obtain the following invariant:

/ = u t " V/ 4 ' .
Therefore an invariant solution is of the form
(5.55) B = C r ""β - "' ".
Equation (5.53) plays the role of a normalizing condition in this case also:
substituting the function of (5.55) into (5.53) we determine the value of the
constant C and find the fundamental solution of the heat equation:
(5.56) u= - (2fi5)- "< r r l/ 41 , *> 0.
145
5.3.4. The wave equation.

For
(5.57) Β, ( - ΔΒ= δ(ί, χ)
the analogues of the operators (5.50) are
d d
X- . = x i_JL- .xiJL, y , —f |χ '
V dx 9χ1 δχ1 d t
(5.58) * ' ' '
Clearly, they can be obtained from the (n+ l)- dimensional version of (5.50) by
making one of the variables imaginary. From this analogy it is completely
clear that for the rotations Χί} and for the Lorentz transformations 7, there
are two invariants: u and τ = t2 — r2, and that in these invariants the last
operator of (5.58) has the form
(5.59) A +
( i_ n ) u . i-
From this it is also clear that the invariant common to all the operators (5.58)
is J = ta^- "~ 1^ 2, so an invariant solution can be sought in the form
1
(5.60) a - C x"- "" .
However, substitution of (5.60) into (5.57) shows that there is no such
solution. The reason is very simple: equation (5.57) has no classical
solutions. N onetheless, we shall find an invariant solution by moving into the
space of distributions. To this end, let us look for an invariant solution of
the form
(5.61) α = / (τ).
As this equality is written in terms of the invariants u and τ of the operators
Xij and Yi, it remains to write down the invariance condition with respect to
the operator (5.59). This condition has the form
or
(5.62)
A classical solution of this ordinary differential equation is given by (5.60).

Let us find the general solution of (5.62) in the space of distributions.
It suffices to treat the case of odd n, since for the construction of the
fundamental solution for even η we can use H adamard's method of descent.
Thus, let η = 2m + 1, m = 0, 1, ... . Then equation (5.62) has the form
(5.63) τ / '( τ ) + η ι/ ( τ ) = 0,
146 N.Kh. Ibragimov
which is the same as (5.45). Therefore its general solution for m = 0 is given
by (5.41'):
(5.64) / (τΗ
while for m Φ 0 it is given by (5.46):
(5.65)
Substitution into (5.57) leads to 2C\ = n~ m, C2 = 0. Returning to the
original notation, we obtain a fundamental solution of the wave equation for
any odd number η of spatial variables:
(5.66) ΐ ί = θ ( ί 2 - ζ *) / 2 for w==l,
(5.67) for
The group method of constructing fundamental solutions presented above
was applied recently [62] to wave equations in Lorentz spaces with a non-
trivial conformal group.
CHAPTER VI
SUMMARY OF RESULTS
Table 1
Frequently encountered groups on the plane
Group Symbol Invariant Canonical variables

in which X = d/ dt
Translations d
in χ: χ = x + a, y = y X- t — x, u = y
in y. χ = x, y = y + b dx
d
X t = y, u = χ
along the straight line dy
kx+ly = 0: d d
χ = x+la, y = y—ka X J = kx+ ly ί = χ/ 1, u= kx + ly
~ ldx ~~ <JF
Rotation
χ = χ cos a+y sin a, X = y- L - xJ- / = ar»+ y* t — arctan — ,
dx dy
y = y cos a — x sin a u — f x* + y*
Lorentz transformation
χ - χ cosh a+y sinh a, J = y* 1«
1
tt = _ — In
y
ID. + * ,
—zyxwvutsrqponmlkjihgfedcbaZYXWV
y = y cosh a + χ sinh a dx dy 2 y —x
Galilean transformation
χ = x + ay, y = y
Table 1 (contd.)
Canonical variables
G roup Symbol Invariant
in \ vti;li X = δ/ θί
Homogeneous stretching X= x ° y ° ( = In x, u = —
+
χ = xe", y = ye" dx dy y y
Non- homogeneous stretching
X = xJL + kyA- t = In x, u = .fi
χ = xe", y = ye1" ox dy y y
Projective transformation
x
- y
X
l - a x ' ) 1- ax * = *·- £- + *?.*. I
x
χ y
dx dy y
Table 2
Some first- order equations with a known admissible operator
(1) d .
x= d .
ν ' = F(x),
y' = F(kx+ly), x = l- L- - ii.

dx
(Π ) yf- xf(r) „ d
y V » A • —
x — yF(r) ' ' ' dx
(I I I ) ίί' = F (yfx). d d
4- w
dx dy
(IV) k 1 lc
ν"= x - F(ylx ), x _ χ
d
dx r y dy '
(V) xy' — ρ ( xe- V), x _ T
d J_ β
dx ^ % '
(VI) y' = yF (jre- *), x _ d , d
dx dy
(VII) y' = - i- + xF (y/ x), x
_ _ d , y d
X dx χ dy
(VIU ) * / = y- \ - F (yfx). x = x*±
dx dy
(IX) v'- y g a
y
+ *
x + F{yfx) • dx dy
(Χ) y' — y d
x- hF(y) • " dx
(Χΐ) * !/ '= . V + F(x), x= T d
dv
(XII) *y'~ y d
In χ - {· r (y) A = Xy
dx
••
(XIII) x= d
xy' = y [In y - f- F(χ)], ••
(XVI) y' = P(x)y + Q (x), X- e iP (oc)d * d

By '
n
(XV) ?(x)y + Q(χ) y . ifV1- 1 i) fP(x)dx _£ _
(XVI) t = Ρ (χ) y. x=
N.Kh. Ibragimov
Table 3
Some second- order equations with a known admissible operator
(I ) y = F( y, y') , x- ^;
= F( x, y') ,
— F(kx- \ - li X = l— — kJL·
dx dy
y" = ( i + y'*)' V ^ + ifv' )· ~
X^y- L .
~dx "dy '
(I I I )
dx
(VI) y" + q'(v) rV· 9'(y) T \
?(v) J(V) I OX
(V) V" = F( x, y- xy'),
(VI) P( x) y"- p"( x ) y = / · (ζ, ρ (ζ) y'- p'( x) y) . X = p(z)^_.
(VII) P* (x) y" + P(x ) Ρ' ( ι) V' = F(y ,p(x)y'h Χ Ρ(Χ)

~ 4Ϊ·
(VI I I ) xy w( JL. X = X +y
lx- W-
x x + ky
(IX) y
= - k W'
(X) y = V
F(y*- *, y'lvh x= +y
i - bj-
(XI) y" = VF(z, y'Jy),
~~ "dx'
i'{y) n y
' \. V - l,,\
d
( Χ") y" y
dy
9(y)
(XIII) yy "= y'*+y*f ' f z, ^ 1 - In !/
wr X = x»J- .
> dy
(XIV) xy " - f y' = x*y'

\ ' z»' - ). ~~ ~dx '
(XV) *»i ~~ ~dx dy

y y
(XVI) z»i -1 ~^ j. ~~ "dx dy
+1
(XVII) z* ) ζ * ίί' = F( z- · y, ztf' — ky),
(XVIII) V + O- Dxpy'* = ji'F ( χι

Table 4
Canonical form of second- order equations admitting a two- dimensional Lie algebra
Basis of Li in Equation
Type Structure ai La
canonical variables
d d
1 (Χι. * , ] = ο, Χι\ / Χιφ Ο λr m
^^mm γ __ y' = / (sO
Π [Χι, Χ,} = 0, . λf 0 γ 0
y" = / (*)
d » Q d
III [Χι. ^ι VzyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGF
Χ* Φ ο χ » =
~dy"' dx dy X
_ d y _ d
IV (Χι. Xt)**Xl, Λ] γ Λ J ^= U y
« = / (x) y'
dy ' dy
Table 5
Computation of Lie algebras of operators admissible by a second- order equation
Differential equation (1.29) V = / (s, y, v')
Admissible operator
Χ^ Χ + ζ , / , + ζ , - ^ - ,
Second prolongation of
ζτ = - 0 (η) - y'0 (ξ) = η , + (η ν - ξ,) / - »"ξρ,
the operator ζ» = D (ζ,) - y·/ ) (|) = η«χ + (2η«ν - ξβ ) • +
+ (%ν - 2ξ«ν) ** - / n»» + (η» ~ 2ξχ - 3»^ ) ν β
η « + (2Tlg, - ξχ.) »' + (η^, - 21»,,) »* - y'»|. +

Defining equation (1.30)
150 N.Kh. Ibragimov
Table 6
Algorithm for integration of second- order equations
using a two- dimensional Lie algebra by the method of canonical variables
Step Actions Result
1 Find the algebrazyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCB

L T of all admissible operators Basis of L T : Xu ..., Xr
X = ξ(χ, y)d/ dx + T[(x, yfi/ dy by solving the defining
equation (1.30) of Table 5
2 If r — 2, go to the next step, while if r > 2, first find Basis of La'. X\ , X2

in L T any two- dimensional subalgebra La using
Theorem 2.1. F or r = 1 the order of the equation
can be lowered (§1.4), while for r = 0 the group-
theoretical approach is inapplicable
3 Determine the type of the resulting algebra La by Reduction of the

Theorem 2.3; compute the commutator \ X\ , X2] of structure of La to one
basis operators and their pseudo- scalar product of the four types of
Χι ν X2 (see (2.8)); if [Xu X2] Φ 0 and is not equal Table 4
to X\ , then by the linear change of variables (2.10)
pass to a new basis X{, X2 with [X[, X2] = X{
4 Using Theorem 2.4, reduce the basis of the algebra La Determination of the
to one of Table 4 by a suitable change of variables integrating change of
x, y. By this change of variables reduce the equation variables
under consideration to canonical form of
corresponding type from Table 4 and integrate it
(§2.2.3)
5 Rewrite the solution in the original variables Solution of the equation
Table 7
Real non- isomorphic structures of three- dimensional Lie algebras
(classification by Bianchi [26])
Type IX,, X,] [X,, X,)
I 0 0 0
II 0 0 - Xt
III 0 0 - Xi
IV 0 Xt - (Xt + Xt)
V 0 Xt — Xi
VI 0 kXt, k φ 0 Η 1 — Xi
VI I 0 Xl+ qXt Xt — QXI
VI I I Xi χ» - 1Xt
IX X* Χι Xt
Remark. In the complex domain there remain only seven non- isomorphic structures:
type VII assumes the form of VI with k = {q+i)l(q- i) if we change to the complex
basis ΑΙ = X2 + iX\ , X2 = X2—iXu X3 = X3/ (q- i), while type IX becomes the same
as type VII in the basis X\ γ iY Υ — rY Υ = Xi + iXs.
Table 8
Real non- similar three- dimensional Lie algebras an d invariant equations
Type in
Ν Basis of L 3 Equation
Table 7
dx dy '
d d y =
1· IX
A — —χ ' i » ·/
dx dy
2
dx dy
dx ' dy
1
VI I I dx~' dx 2 dy'
3
*~~ ~dx dy
d d
X — X —
4··
V
*~ ~dy~
„ a Y
a
A, = _ _ , Ai = _ — ,
5· VI I dx dy
d x
Xt = (qx + v)- g
j + (w — )- gg-
Χι = _ _ , Χ2 = _ _ ,
οχ oy
6 VI
for dx dy
III d d
ν v
for Al= _ — , Aj= Χ - τ- ι
7 y" = Cx *"*1
dy Oy
dx dy
γ 8 γ θ
Αι —. , Αϊ _ ,
8 dx dy
V Χ — d
d
y' = 0 .
σχ σι/
θ θ d
9 Χ — Χ —χ Χ —
dy dy dy
152 Table 8 (contd.)
Type in
Ν Basis of L 3 Equation
Table 7
10
X, «*
ax dy
IV
v d - τ d
Λ] = , A2 = X ——
x
11 y' = C e
dx
II d v
d v
d
= - — , Λ2 = _ _ , A S = I _ _
σι d <ty d dy
= zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFED
v
- r - , A2 = χ _ _ ,
Sy oy
13··
Remarks. * In the complex domain, case l_reduces to case 2 by ^hanging to variables
x, y using the relations χ — (xy—l)fcx+y), y — —i(xy+l)/ (x+y), while 5 reduces
to 6 with k - (q + i)l(q—i) bythe change of variables χ - (y- ix)jl, y - (y+ix)[2.
• • In cases 4 and 13 there are no invariant second- order equations, so third- order
equations are quoted.
Table 9
Lie's group- theoretical classification of second- order equations
G roup Basis operators Equation
d
Gi γ
Ai = ——
** = / (». ιΟ
d dx d
Υ Υ
dx dy
G·
y« = J_/ (y')
dx dy dx dy
χ —v
dx dy
ν & ν ι d * .. & ν md . d y* = Cy~*
Gs =_ , A-s = 2 *_ + y _ , A - *- _ + * , _
X l s = I
x i ~ J L , x 2 = ^_, x9 = x J L + ( X + y )° y'=Ce~V
fc- a
dx dy &x oy y'=Cy'*- 1, ΗφΟ, 4- .1.2
, Xi = - s—dj
Xl = ——dx > X* = Xdy
ιfX»= X - 3—dx - j— .
dx dy dy dx
Xt = y _ _ , Xt = y — , X, = x *_ _ + x y - _ , y" = 0
dx dy dx dy
* — ' £ + • "£
Table 10
Construction of the algebra admitted by a linear hyperbolic equation
in two variables
Equation u
xy +
α (
*·
v) u
x +
b ( x>
"' y
u + c (x tt
· V) = °
Laplace invariants A = ax + 06 — c, k = by - f ab — c
The general form of an

admissible operator OX dy On
Differential form dual Z=r\ dx~ ldy

to the operator X
Defining equations d (AS) = 0, d (fcB) = 0, d (σ + βη + 6|) = (h — k) S
Coordinate form of the • {Αξ), + (Αη)ν = 0, (*ξ)« + (*η), = Ο,

defining equations (σ.+ «η Η- 6ξ), = (Α - k) η , (σ + αη + b\ ) y =
= (Α — Α) Ι
References
[1] F . Klein, Review of Theorie der Transformationgruppen, Bd. I ll, at the award of the
first N .I. Lobachevskii Prize, 22 October 1897, Typolithography of the Imperial
University, Kazan 1989, pp. 10- 28.
[2] S. Lie, Vorlesungen uber Differentialgleichungen mit bekannten infinitesimalen
Transformationen (revised and edited by D r G . Scheffers), B.G. Teubner, Leipzig 1893.
[3] , Klassifikation und Integration von gewonlichen Differentialgleichungen
zwischen x, y, die eine G ruppe von Transformationen gestatten, Arch, for M ath. VIII
(1883), 187- 453 (parts I — III), IX (1884), 431- 448 (part IV); see [4], Vol. 5, articles
x, xi, xrv, xvi.
[4] — , G esammelte Abhandlungen, B.G . Teubner, Leipzig; H . Aschehoug & Co.,
Oslo, Vol. 1 (1934); Vol. 2 (Part 1) (1935); Vol. 2 (Part 2) (1937); Vol. 3 (1922);
Vol. 4 (1929); Vol. 5 (1924); Vol. 6 (1927).
[5] - , Zur Theorie des Integrabilitatsfactors, Christ. F orh., Aar 1874,
Christiania 1875, pp. 242- 254; see [4], Vol. 3, article XIII.
[6] , G eometrie der Beruhrungstransformationen (Presented by Sophus Lie and
G eorg Scheffers), B.G . Teubner, Leipzig 1896.
[7] , Die infinitesimalen Beruhrungstransformationen der Optik, Leipz. Berichte 1,
131- 133 (1896); see [4], Vol. 6, article XXIV.
[8] L.V. Ovsyannikov, Gruppovoi analiz differentsial'nykh uravnenii, N auka, Moscow 1978.
M R 80d:58072.
Translation: G roup analysis of differential equations, Academic Press, New York —
London 1982. M R 83m:58082.
[9] R.L. Anderson and N .H . Ibragimov, Lie —Backlund transformations in applications,
SIAM, Philadelphia, PA 1979. M R 80c:58048.
N.Kh. Ibragimov
[10] N .Kh. Ibragimov, zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA

Gruppy preobrazovanii ν matematicheskoi fizike, N auka,
Moscow 1983. M R 85j:58003.
Translation: Transformation groups applied to mathematical physics, Reidel,
D ordrecht 1985. M R 86e:58001.
[11] , Conformal invariance and the Huygens principle, Dokl. Akad. N auk SSSR
194(1970), 24- 27. M R 42 # 4881.
= Soviet M ath. D okl. 11 (1970), 1153- 1157.
[12] and A.O. Oganesyan, The hierarchy of Huygens equations in spaces with a
non- trivial conformal group, Uspekhi Mat. N auk 46:3 (1991), 111- 146.
= Russian Math. Surveys 46:3 (1991), 137- 176.
[13] S. Lie, Theorie der Transformationgruppen, Bd. I (revised with the help of F . Engel),
B.G. Teubner, Leipzig 1888.
[14] V. Anisimov, Kurs teorii obyknovennykh differentsial'nykh uravnenii (A course on
ordinary differential equations), Warsaw 1906.
[15] E. G oursat, Cours d'analyse mathematique, 3rd ed., Gauthier- Villars,
Paris 1918- 1923.
Translation: Kurs matematicheskogo analiza, G TTI, Moscow—Leningrad 1933
(Vol. I I , 4.11, Ch. XIX, sect. IV).
[16] E.L. Ince, Ordinary differential equations, Dover, New York 1944. MR 6- 65.
Translation of 1st ed.: Obyknovennye dijferentsila'nye uravneniya, ON TI,
Kharkov 1939.
[17] L.E. Dickson, Differential equations from the group standpoint, Ann. of Math. (2) 25
(1925), 287- 378.
[18] L.V. Ovsyannikov, Gruppovye svoistva differentsial'nykh uravnenii (G roup properties of
differential equations), Izdat. Sibirsk. Otdel. Akad. N auk SSSR, Novosibirsk 1962.
M R 26 # 264.
[19] P.J. Olver, Applications of Lie groups to differential equations, Springer- Verlag,
New York- Berlin 1986. M R 88f:58161.
Translation: Prilozhenie grupp Li k differentsial'nym uravneniyam, Mir,
Moscow 1989.
[20] G .W. Bluman and S. Kumei, Symmetries and differential equations, Springer- Verlag,
New York 1989. M R 91b: 58284.
[21] H . Stephani, Differential equations: Their solution using symmetries, Cambridge
University Press, Cambridge 1989. MR 91a:58001.
[22] N .Kh. Ibragimov, Azbuka gruppovogo analiza (ABC of group analysis), Znanie 1989,
no. 8. M R 91h:35020.
[23] , Opyt gruppovogo analiza obyknovennykh differentsial'nykh uravnenii
(An attempt at group analysis of ordinary differential equations), Znanie,
Moscow 1991. MR92i:34001.
[24] S. Lie, D ie infinitesimalen Beriihungstransformationen der Mechanik, Leipz. Berichte
(1889); (see [4], Vol. 6, article VI).
[25] F.M. Mahomed and P.G .L. Leach, Lie algebras associated with scalar second- order
ordinary differential equations, J. Math. Phys. 30 (1989), 2770- 2777. MR 90k:58193.
[26] L. Bianchi, Lezioni sulla teoria dei gruppi continui finiti di transformazioni, Spoerri,
Pisa 1918.
[27] A.Z. Petro, Prostranstva Einsteina (Einstein spaces), Fizmatgiz, Moscow 1961.
M R 25 # 4897. (Or see his Novye metodyi ν obshchei teorii otnositel'nosti (New
methods in the general theory of relativity), N auka, Moscow 1966. MR 34 # 7181.)
[28] B.A. Dubrovin, S.P. Novikov, and A.T. Fomenko, Sovremennaya geometria, N auka,
Moscow 1979 (2nd edition 1986). M R 81f: 53001.
Translation: Modern geometry—methods and applications. I, II, Springer- Verlag,
New York- Berlin 1984, 1985. M R 85a:53003; 86m:53001.
Group analysis of differential equationszyxwvutsrqponmlkjihgfedcbaZYXWVUTS
155
[29] L.P . E isen h art, E quivalen t c o n t in u o u s gro u ps, An n . of M a t h . (2) zyxwvutsrqponmlkjihgfedcbaZYX

33 (1932), 6 6 5 - 6 7 0 .
[30] , C o n t in u o u s gro u p s of t ran sfo rm at io n s, P rin cet o n U n iversity P ress, P rin ceton ,
N J 1933.
T ran slat io n : Nepreryvnye gruppy preobrazovanii, I L, M o sco w 1947.
[31] J. K r a u se a n d L. M ich el, C lassification of th e sym m etries of ordin ary differential
equ at io n s, in : P ro c . Sem in ar "G r o u p - t h eo r et ic a l m et h o d s in ph ysics" (M oscow,
J u n e 1990), Sprin ger- Verlag, N ew Yo r k 1991.
[32] V.P . E r m a ko v, U n iversitetskie I zvestiya (Kiev) 20:9 (1880), 1 - 2 5 .
3
[33] E. P in n ey, T h e n o n lin ear differential equ at io n y" +p(x)y+Cy~ = 0, P ro c . Am er.
M a t h . Soc. 1 (1950), 681. M R 1 2 - 3 3 6 .
[34] L . M . Berkovich a n d N . K h . R o zo v, Som e rem arks o n differential equ at io n s of th e
a
form y" + ao(x)y = <f>(x)y , D ifferen tsial'n ye U ravn en iya 8 (1972), 2 0 7 6 - 2 0 7 9 .
M R 47 # 8946.
= D ifferen tial E q u a t io n s 8 (1972), 1609- 1612.
[35] C . R o gers a n d W. F . Am es, N o n lin ea r bo u n d a r y value pro blem s in science a n d
en gin eerin g, Ac ad em ic P ress, Bo st o n , M A 1989. M R 90i:35001.
[36] M . Boiti, C . L a d d o m a n a , a n d F . P em pin elli, An equivalen t real form of th e n o n lin ear
Sch rodin ger eq u a t io n a n d t h e perm u t abilit y for Backlu n d t ran sfo rm at io n s, N u o vo
C im en t o Β (11) 62 (1981), 3 1 5 - 3 2 6 . M R 82j:81013.
[37] S. Lie, Z u r allgem ein en T h eo rie der partiellen D ifferen tialgleichun gen beliebiger
O r d n u n g, Leipz. Berich te 1 (1985), 5 3 - 1 2 8 ; see [4], Vol. 4, article I X.
[38] G . Birkhoff, H yd r o d yn a m ic s, P rin cet o n U n iversity P ress, P rin cet o n , N J 1950 (revised
edition 1960). M R 1 2 - 3 6 5 .
T ran slat io n : Gidrodinamika, I L , M o sc o w 1954 (revised edition 1963).
[39] L.I . Sedov, Metody podobiya i razmernostei ν mekhanike, 4t h ed., G I T T L ,
M o sc o w 1957.
T ran slat io n : Sim ilarity a n d dim en sion al m et h o d s in m ech an ics, Academ ic P ress,
N e w Yo r k - L o n d o n 1959. M R 21 # 6840.
[40] W. F . Am es, N o n lin e a r p art ial differential equ at io n s in en gin eerin g, Vols. I, I I ,
Acad em ic P ress, N e w Yo r k - L o n d o n 1965, 1972. M R 35 # 1235; 57 # 13108.
[41] E. Vessiot, Sur u n e classe d 'o qu at io n s differentielles, An n . Sci. E cole N o r m . Su p . 10
(1893), 53.
[42] A. G u ld berg, Sur les equ at io n s differentielles que posseden t u n system e fu n dam en t al
d'in tegrales, C .R . Ac a d . Sci. P aris 116 (1893), 964.
[43] S. Lie, Vorlesun gen iiber con tin uerlich e G r u p p e n m it geom etrisch en u n d an d eren
An wen d u n gen (revised a n d edited by D r G . Scheffers), B.G . T eu bn er, Leipzig 1893.
[44] N . G . C h ebo t a r ev, Teoriya grupp Li (Lie gr o u p th eory), G I T T L , M o sco w —
Len in grad 1940.
[45] W. F . Am es, N o n lin e a r su perpo sit io n for o p er a t o r equ at io n s, in : N o n lin ear equ at io n s
in abst ract spaces ( E d . V. L a ksh m ika n t h a m ) , Acad em ic P ress, N ew Yo rk 1978,
p p . 4 3 - 6 6 . M R 81k:34001.
[46] R .L. An d erso n , A n o n lin ear superposition prin ciple ad m it t ed by coupled R iccati
equ at io n s of t h e projective type, Lett. M a t h . P h ys. 4 (1980), 1 - 7 . M R 81g:93048.
[47] a n d P . Win tern itz, A n o n lin ear superposition prin ciple for R iccati equ at io n s
of co n fo rm al type, in : G r o u p th eoretical m e t h o d s in physics (P roceedin gs, C ocoyoc,
M exico 1980), L ect u re N o t e s in P hysics 135 (1980), 1 6 5 - 1 6 9 . P h ys. Abs. 81- 89770.
[48] V. M . M en 'sh iko v, C o n t in u o u s con jugation of in varian t solution s, D in a m ika Splosh n .
Sredy N o . 10 (1972), 7 0 - 8 4 . M R 56 # 16114.
[49] V.V. P u kh n a c h ev, I n va r ia n t solution s of th e N avier- Stokes equ at io n s describin g
m o t io n s with a free bo u n d a r y, D o kl. Akad . N a u k SSSR 202 (1972), 3 0 2 - 3 0 5 .
= Soviet P h ys. D o kl. 17 (1972), 2 4 - 2 7 .
156 N.Kh. Ibragimov
[50] A.N. Tikhonov and A.A. Samarskii, Uravneniya matematicheskoi fiziki, 5th ed.,
Nauka, Moscow 1977.
Translation: Equations of mathematical physics, McMillan, New York 1963.
MR 29 # 2498.
[51] G.F.B. Riemann, Uber die Fortpflanzung ebener Luftwellen von endlicher
Schwingungsweite, Abhandl. Koningl. Ges. Wiss. Gottingen 8 (1860).
[52] S. Lie, Uber die Integration durch bestimme Integrate von einer Classe linearer
partieller Differentialgleichungen, Arch. Math. Naturvid. 6 (1882), 328-368 (see [4],
Vol. 3, article XXXV).
[53] E.T. Copson, On the Riemann - Green function, Arch. Rational Mech. Anal. 1
(1958), 324-348. MR 20 # 4088.
[54] L.V. Ovsyannikov, Group properties of the S.A. Chaplygin equations, Zh. Prikl.
Mekh. i Tekhn. Fiz. 1960, no. 3, 126-145.
[55] J. Leray, Hyperbolic differential equations, Inst. for Advanced Study, Princeton,
NJ 1953. MR 16-139.
Translation: Giperbolicheskie differentsial'nye uravneniya, Nauka, Moscow 1984.
[56] I.M. Gel'fand and G.E. Shilov, Obobshchennye funktsii i deistviya nadnimi, Fizmatgiz,
Moscow 1959.
Translation: Generalized functions, Vols. I, II, Academic Press, New York-
London 1964, 1968. MR 29 # 3869; 37 # 5693.
[57] Y. Choquet-Bruhat, C. DeWitt-Morette, and M. Dillard-Bleick, Analysis, manifolds,
and physics, Parts I, II, North-Holland Publ. Co., Amsterdam-New York 1982,
1989. MR84a:58002; 91e:58001.
[58] Yu.Yu. Berest, Construction of fundamental solutions for Huygens' equations as
invariant solutions, Dokl. Akad. Nauk SSSR 317 (1991), 786-789.
= Soviet Math. Dokl. 43 (1991), 496-499.
Translated by M. Grinfeld Institute of Mathematical Modelling

of the Russian Academy of Sciences
Received by the Editors 12 March 1992

Ibragimov - 1992 - Group Analysis of Ordinary Differential Equations

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ibragimov - 1992 - Group Analysis of Ordinary Differential Equations

Uploaded by

Copyright:

Available Formats

Uspekhi Mat Nauk 47:4 (1992), 83-144 Russian Math.

Surveys 47:4 (1992), 89-156

Group analysis of ordinary differential equations and

Chapter IV. Ordinary differential equations with a fundamental system of 124

Sophus Lie (1842-1899)

Moreover, Lie [3] provided a classification of all ordinary differential equations

DEFINITIONS AND ELEM ENTARY APPLICATIONS

Example. The Riccati equation y'+y1—2/ x2 = 0 admits the transformation

§1.1. One- parameter transformation groups

D enoting by T a the transformation (1.1) which takes a point (x, y) into a

act Ψ), ΦΙ«=* = Χ,

scalar under an arbitrary change of variables, unlike the

Definition 1.2. A function F(x, y) is called an invariant of the group of

Hence it follows that every one- parameter group of point transformations

5Fk/ dx'\ \ is equal to s at all

Such variables /, u are called canonical variables.

§1.2. Prolongation formulae

H ere it is convenient to use the "t o t al" differentiation operator

D erivatives are transformed according to the formulae

If we start from a group of point transformations G, then after adding the

/ + [Ζ>(ζι) — y"D(l)]a = y"

Therefore the symbols of the prolonged groups G, G are

(1.17') = D (η) - y'D (ξ) = η , + ( η , — y' — y'%,

§1.3. Groups admissible by differential equations

§1.4. Integration and reduction of order using one- parameter groups

admits a one- parameter group with the symbol (1.7)

if and only if the function

{for ξ,ζ) — ϊ\ Ρ Φ 0) is an integrating factor for the equation (1.20).

Since it is an equation of power type, it is natural to expect that it admits a

(1.23) X=x J- - yJ>- .

and applying the formula (1.21), we obtain the integrating factor

After multiplication by this factor, (1.22') is brought to the following form:

from which we obtain the solution in the form

1.4.3. Invariant differentiation.

(1.26) lOr= F( u' p>·

we find the invariants u = χ, ο = y' / y. F rom Theorem 1.5 we find the

§1.5. Defining equations

(1.31) ff- if- e-

H e r e / = ey—y' and the defining equation (1.30) has the form

From equations (1.32) and (1.33) by integration with respect to y we obtain

Let us substitute these expressions for ξ, η into (1.34), (1.35). As the

Hence 2(a'- a/ x) + q = 0,2a' + b = 0. Therefore

Substituting these expressions into (1.34), we have

from which a = C\ x hi x- VC^x; here equation (1.35) is satisfied identically.

As a result, we have obtained the general solution of the defining equations

C\ , C2. In view of the linearity of the defining

(4.36) ^zlnx- z- — 2(1 + lu x) —, X l

§1.6. Lie algebras

(1.37) Χ, - ^ ^ + η »^ - . Χ^ Ι,- L + ru- ^

(1.38) [Xv A%] = (X, ( y - X2 (£,)) - L. + (Χ, (η2) - Xt (η,)) ~

§1.7. Contact transformations

in the (2n+ l)- dimensional space of variables (x, u, «')·

is a symbol of a group of contact transformations if and only if

(1.39) X = — Wµ 4- 4- (W — pWp) 4- - f (Wx + pWv)^- ·

Prolongation of the operator (1.38) to third derivatives has the form

Λ = A ~i- £·> ——= • + L« —- —r— ,

The defining equation has the form

and decomposes into the following four equations:

The operators ΛΊ, ..., Λ7 generate a 7- dimensional algebra of a group of

differential equation admits an infinite group of contact transformations. The

INTEGRATION OF SECOND- ORDER EQUATIONS ADM ITTING

§2.1. Consecutive reduction of order

Clearly, it has the solution y = χ and by the superposition principle it admits

(3.6') - ^^iSWlSix) ( , ~ 1 «).