23.05.2014 Views

symmetries and conservation laws obtained by lie group analysis for ...

symmetries and conservation laws obtained by lie group analysis for ...

symmetries and conservation laws obtained by lie group analysis for ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

DIPLOMA THESIS<br />

SYMMETRIES AND<br />

CONSERVATION LAWS OBTAINED<br />

BY LIE GROUP ANALYSIS FOR<br />

CERTAIN PHYSICAL SYSTEMS<br />

MARCUS ERIKSSON<br />

Uppsala School of Engineering<br />

<strong>and</strong><br />

Department of Astronomy <strong>and</strong> Space Physics, Uppsala University, Sweden<br />

APRIL 22, 2008


Abstract<br />

The <strong>for</strong>mal model of physical systems is typically made in terms of<br />

differential equations. Conservation <strong>laws</strong> can be found <strong>for</strong> any system of<br />

linear or non-linear differential equations, based on the concept of adjoint<br />

equations. Provided that the number of equations is equal to the number<br />

of variables, the adjoint equations <strong>for</strong> a system is used <strong>for</strong> construction of<br />

Lagrangians. Lie <strong>group</strong> <strong>analysis</strong> provides us with a way of finding all the<br />

<strong>symmetries</strong> of the original mathematical model. The <strong>symmetries</strong> are inherited<br />

<strong>by</strong> the adjoint equations. By considering the total system of differential<br />

equations together with its adjoint equations, we construct the Lagrangian<br />

<strong>and</strong> once the <strong>symmetries</strong> are found, Noether’s theorem is used <strong>for</strong> determining<br />

<strong>conservation</strong> <strong>laws</strong>. Consequently, <strong>for</strong> any symmetry found <strong>by</strong> Lie <strong>group</strong><br />

<strong>analysis</strong> a <strong>conservation</strong> law can be found.


The mass of mathematical truth is obvious <strong>and</strong> imposing; its practical<br />

applications, the bridges <strong>and</strong> steamengines <strong>and</strong> dynamos, obtrude<br />

themselves on the dullest imagination. The public does not need to<br />

be convinced that there is something in mathematics.<br />

C. G. Hardy


CONTENTS<br />

Contents<br />

Acknowledments<br />

v<br />

vii<br />

I Group Theory 1<br />

1 Introduction 3<br />

2 Notation <strong>and</strong> the total differential operator 5<br />

2.1 Functions of one independent variable x <strong>and</strong> one dependent variable<br />

y 5<br />

2.2 Functions of several variables 9<br />

2.3 Important equations 11<br />

3 Basic Lie <strong>group</strong> theory 13<br />

3.1 Trans<strong>for</strong>mation <strong>group</strong>s 13<br />

3.2 One-parameter <strong>group</strong> 15<br />

3.2.1 Canonical parameter 18<br />

3.3 The infinitesimal trans<strong>for</strong>mation <strong>and</strong> Lie’s equations 18<br />

3.4 Invariant functions <strong>and</strong> equations 22<br />

3.4.1 Invariant function 22<br />

3.4.2 Invariant equations 23<br />

3.5 Multi-parameter <strong>group</strong>s <strong>and</strong> Lie algebras 25<br />

3.5.1 Multi–parameter <strong>group</strong>s 25<br />

3.5.2 Short outline of Lie algebra 25<br />

v


CONTENTS<br />

4 Differential Algebra 27<br />

4.1 The space A 27<br />

4.2 Extended point trans<strong>for</strong>mation <strong>group</strong> 28<br />

4.2.1 Extended pointtrans<strong>for</strong>mation in the plane 28<br />

4.2.2 Point trans<strong>for</strong>mation involving many variables 32<br />

4.2.3 Extended generators <strong>for</strong> differential equations with two<br />

independent <strong>and</strong> one dependent variable 34<br />

5 Operators 37<br />

5.1 Euler-Lagrange operator 37<br />

5.2 Lie-Bäcklund operators 39<br />

5.3 Operators N i associated with Lie-Bäcklund operators <strong>and</strong> the<br />

fundamental identity 40<br />

II Symmetries <strong>and</strong> Conservation Laws 43<br />

6 Hamilton’s Variational Principle 45<br />

6.1 Hamilton´s variational principle in the plane 45<br />

6.1.1 Euler-Lagrange equation with several variables 50<br />

7 Noether’s Theorem <strong>and</strong> Conservation Laws 53<br />

7.1 Noether´s Theorem 53<br />

7.2 Many <strong>symmetries</strong> 60<br />

8 Adjoint equations 63<br />

8.1 Linear adjoint equations 63<br />

8.2 Adjoint equations of arbitrary differential equations 66<br />

8.3 Constructing Lagrangians <strong>by</strong> the use of adjoint equation 68<br />

9 Symmetries <strong>and</strong> The Main Conservation Theorem 71<br />

9.1 Determining equations 74<br />

9.2 The main <strong>conservation</strong> theorem 79<br />

9.3 Summary 85<br />

Bibliography 87<br />

vi


ACKNOWLEDMENTS<br />

First, I would like to thank my supervisor Professor Bo Thidè <strong>for</strong> giving me the<br />

opportunity to discover the bordercountry between mathematics <strong>and</strong> physics. I<br />

would also like to thank him <strong>and</strong> everyone else how has supported me <strong>and</strong> given<br />

me ideas as a result of many profound discussions, especially Dr Jan Bergman,<br />

Siavoush Mohammadi, Jonny Jansson, Lars Daldorff <strong>and</strong> professor Nail Ibragimov.<br />

You have all been of great importance <strong>for</strong> my work.<br />

vii


Part I<br />

Group Theory<br />

1


1<br />

INTRODUCTION<br />

To confirm physical theories we need to make observations <strong>and</strong> measurements of<br />

the physical system in interest. To be able to make a measurement of a physical<br />

quantity, we need to know if that quantity is conserved in time <strong>and</strong> space. For<br />

example;<br />

We have a ruler of lenght l at one side of a room. During a time ∆t we move<br />

it ( <strong>by</strong> throwing it ) to to the other side of the room. Now, if the ruler, after the<br />

transport, would have a length diffrent from l we cannot measure its length since<br />

it is changing in time <strong>and</strong> space, i.e the physical quantity l is not conserved <strong>and</strong><br />

there<strong>for</strong>e not very useful as a measurable. Of course, in reality l is conserved, <strong>and</strong><br />

this was just an example to illustrate the meaning of a conserved quantity.<br />

In radio comunications we transmit in<strong>for</strong>mation encoded in radiowaves. This<br />

transmission of in<strong>for</strong>mation is only useful if the in<strong>for</strong>mation we transmit will be<br />

the same at the reciever point. Thus, the in<strong>for</strong>mation in radiowaves is a conserved<br />

quantity.<br />

The motion of the ruler <strong>and</strong> the transmission of radiowaves are examples of physical<br />

systems. In mathematics we describe physical systems in terms of differential<br />

equations. The motion of the ruler is described <strong>by</strong> the equation of motion <strong>for</strong> a<br />

projectile [1] <strong>and</strong> the behavior of radiowaves, i.e electromagnetic waves, is described<br />

<strong>by</strong> Maxwell’s equations [2].<br />

In 2006 the mathematican Prof. Nail Ibragimov proved a main <strong>conservation</strong> theorem<br />

<strong>for</strong> any well-posed system of differential equations [3]. The purpose of this<br />

3


CHAPTER 1. INTRODUCTION<br />

work is to present the mathematical metod <strong>and</strong> the underlying mathematics to find<br />

<strong>conservation</strong> <strong>laws</strong> <strong>for</strong> physical systems described <strong>by</strong> differential equations. The<br />

mathematical tools we need orginate from Lie <strong>group</strong> theory, the variational principle,<br />

<strong>and</strong> the concept of adjoint equations. These three areas will be presented<br />

<strong>and</strong> tied togheter to <strong>for</strong>mulate the main <strong>conservation</strong> theorem.<br />

This work is divided into Part I <strong>and</strong> Part II. Part I will focus on the notations<br />

that will be used ( chapter 2 ) <strong>and</strong> serve as an introduction to Lie <strong>group</strong> theory<br />

(chapter 3 <strong>and</strong> 4). In chapter 5 some useful definitions, lemmas <strong>and</strong> theorems will<br />

be presented. In Part II, chapter 6-9, we will develop the method to find <strong>conservation</strong><br />

<strong>laws</strong> based on the Lie <strong>group</strong> theory from Part I. In chapter 6 we present<br />

the variational principle, which togheter with Part I leed us to a <strong>for</strong>mulation of<br />

Noethers theorem in chapter 7. In chapter 8 we present the method <strong>for</strong> construction<br />

of Lagrangians, based on the concept of adjoint equations. In chapter 9 we<br />

show how symetries can be found <strong>by</strong> Lie <strong>group</strong> theory <strong>and</strong> how this togheter with<br />

the concept of adjoint equations results in a main <strong>conservation</strong> theorem.<br />

4


2<br />

NOTATION AND THE TOTAL<br />

DIFFERENTIAL OPERATOR<br />

In this chapter we introduce some of the basic mathematical notations <strong>and</strong> the<br />

total differentiation operator that will be used throughout the thesis. When dealing<br />

with differential equations these notations comprise a convinient language<br />

adopted from differential algebra. It will be explaind how the notations are defined<br />

<strong>and</strong> how they simplify large equations.<br />

2.1 Functions of one independent variable x <strong>and</strong> one dependent<br />

variable y<br />

In order to obtain in<strong>for</strong>mation about a system we analyse the differential equations<br />

which describe it. In most cases it is very complicated or impossible to find<br />

a solution. Howerever, there are cases in which we know the general solutions to<br />

a differential equation. This solutions is a family of curves depending on a number<br />

of parameters (unknown constants), see [4] <strong>and</strong> [5]. We will start to find a<br />

differential equation <strong>for</strong> a family of curves, <strong>and</strong> see how this procedure motivates<br />

the introduction of a total differentiation operator [4].<br />

Consider a family of curves in the xy-plane,<br />

y = f (x,C 1 ,...,C n ), (2.1)<br />

5


CHAPTER 2. NOTATION AND THE TOTAL DIFFERENTIAL OPERATOR<br />

where the C k are parameters <strong>and</strong> k = 1,...,n. Equation (2.1) can be <strong>for</strong>mulated<br />

implicitly <strong>by</strong><br />

φ(x,y,C 1 ,...,C n ) = 0. (2.2)<br />

Differentiating equation (2.2) with respect to x yields,<br />

∂φ<br />

∂x + ∂φ ∂y<br />

∂y ∂x ≡ ∂φ<br />

∂x<br />

∂φ<br />

+ y′ = 0. (2.3)<br />

∂y<br />

A differentiation of equation (2.3) with respect to x yields<br />

∂ 2 φ<br />

∂x 2 + ∂2 φ<br />

∂y∂x<br />

∂y<br />

∂x + ∂2 φ ∂y<br />

∂x∂y ∂x + ∂φ ∂ 2 y<br />

∂y ∂x 2 + ∂2 φ ∂y ∂y<br />

∂y 2 ∂x ∂x<br />

≡ ∂2 φ<br />

∂x 2 + 2 ∂2 φ<br />

∂y∂x y′ + ∂2 φ<br />

∂y 2 (y′ ) 2 + ∂φ<br />

∂y y′′ = 0. (2.4)<br />

Proceeding the differentiation in this way, we will eliminate one of the parameters<br />

C k k = 1,...,n from φ(x,y,C 1 ,...,C n ). After n differentiations all C k are<br />

eliminated <strong>and</strong> we arrive at an nth-order ordinary differential equation of the <strong>for</strong>m<br />

F(x,y,y ′ ,y ′′ ,...,y (n) ) = 0. (2.5)<br />

We say, according to [4], that the function (2.1), y = f (x,C 1 ,...,C n ) provides a<br />

solution to the differential equation (2.5). The equation F(x,y,y ′ ,y ′′ ,...,y (n) ) = 0<br />

is termed the differential equation of a family of curves to y = f (x,C 1 ,...,C n ) or<br />

φ(x,y,C 1 ,...,C n ) = 0.<br />

Example 2.1<br />

The solution set <strong>for</strong> a simple harmonic oscillator without damping in one dimension<br />

is given <strong>by</strong> the family of curves [1]<br />

y = A 1 cost + A 2 sint, (2.6)<br />

where y is the position coordinate <strong>and</strong> A 1,2 are arbitrary constants (which may be<br />

found <strong>by</strong> using boundary <strong>and</strong> initial conditions <strong>for</strong> a specific problem). Eq.(2.6)<br />

is given implicit <strong>by</strong><br />

φ(t,y, A 1 , A 2 ) = 0, (2.7)<br />

6


2.1. FUNCTIONS OF ONE INDEPENDENT VARIABLE X AND ONE DEPENDENT<br />

VARIABLE Y<br />

i.e, a function of two constants. Differentiation of (2.6) once with respect to time<br />

yields<br />

ẏ = −A 1 sint + A 2 cost. (2.8)<br />

From (2.6) we get<br />

A 1 = y − A 2 sint<br />

. (2.9)<br />

cost<br />

Inserting (2.9) in (2.8) yields<br />

ẏ = A 2 (sint tant + cost) − ytant, (2.10)<br />

i.e, implicitly we have<br />

φ(t,y,ẏ, A 2 ) = 0. (2.11)<br />

Thus, one constant is eliminated. Differentiate (2.10) one more time yields<br />

sint<br />

ÿ = A 2<br />

cos 2 t − y 1<br />

cos 2 − ẏtant. (2.12)<br />

t<br />

From (2.10) we get<br />

A 2 =<br />

ẏ + ytant<br />

sint tant + cost . (2.13)<br />

Inserting (2.13) in (2.12) we get<br />

( )<br />

ẏ + ytant sint 1<br />

ÿ =<br />

sint tant + cost cos 2 − y<br />

t cos 2 − ẏtant. (2.14)<br />

t<br />

Thus, we get an implicit function of the <strong>for</strong>m<br />

φ(t,y,ẏ,ÿ) = 0, (2.15)<br />

where now all constants are eliminated.<br />

(2.14) is reduced to<br />

Using some trigonometric relations,<br />

ÿ = −y (2.16)<br />

7


CHAPTER 2. NOTATION AND THE TOTAL DIFFERENTIAL OPERATOR<br />

or, we get a differential function of the <strong>for</strong>m<br />

F(t,y,ẏ,ÿ) = ÿ + y = 0. (2.17)<br />

Equation (2.17) is termed the differential equation to the family of curves (2.6).<br />

Note that in general the equation of motion <strong>for</strong> a simple harmonic oscillator in one<br />

dimension is given <strong>by</strong> ÿ + ω 2 0 y = 0, where ω2 0 is a constant which in this example<br />

is chosen to 1.<br />

End of Example 2.1<br />

To simplify the notation, we introduce the total differentiation operator D x . It acts<br />

on functions depending on a finite number of variables x,y,y ′ ,y ′′ ,.... The total<br />

differential operator is defined as [4]:<br />

D x = ∂ ∂x + ∂ y′ ∂y + ∂<br />

∂<br />

y′′ + ··· + y(s+1)<br />

∂y ′ ∂y (s) (2.18)<br />

Using the total differentiation operator D x , we can write equation (2.3) <strong>and</strong> (2.4)<br />

in a compact <strong>for</strong>m as D x φ = 0 <strong>and</strong> D x D x φ ≡ D 2 xφ = 0, respectivly, up to D (n)<br />

x φ = 0.<br />

The set of all finite differential functions will be denoted <strong>by</strong> A, as in [6] <strong>and</strong><br />

[4]. The total derivate D x acts on differential functions, F(x,y,y ′ ,y ′′ ,...,y (n) ), of<br />

any finite number of the variables x,y,y ′ ,y ′′ .... To underst<strong>and</strong> the difference between<br />

the total differentiation D x <strong>and</strong> the partial differentiation ∂/∂x, consider the<br />

following example.<br />

Example 2.2<br />

Let f <strong>and</strong> g be two functions from A, defined as<br />

f = xy (2.19)<br />

g = y ′ (2.20)<br />

To find the total derivitive of f <strong>and</strong> g we let (2.18) operate on them.<br />

D x ( f ) = D x (xy) ≡ ( ∂ ∂x + y′ ∂ ∂y )(xy)<br />

= ∂ ∂x (xy) + y′ ∂ ∂y (xy) = y + y′ x. (2.21)<br />

8


2.2. FUNCTIONS OF SEVERAL VARIABLES<br />

If we would only take the partial derivative, we would instead get<br />

∂<br />

(xy) = y. (2.22)<br />

∂x<br />

And <strong>for</strong> g we have<br />

while<br />

D x (g) = D x (y ′ ) = ( ∂ ∂x + y′ ∂ ∂y + y′′ ∂<br />

∂y ′ )(xy) =<br />

= ∂ ∂x y′ + y ′ ∂ ∂y y′ + y ′′ ∂<br />

∂y ′ y′ = y ′′ , (2.23)<br />

∂<br />

∂x (y′ ) = 0. (2.24)<br />

End of Example 2.2<br />

2.2 Functions of several variables<br />

So far we have considered functions of one independent variable i.e, x,y,y ′ ,y ′′ ...<br />

we will now consider functions of several variables. The notation used here is<br />

adopted from the calculus of differential algebra <strong>and</strong> has turned out to be very<br />

convenient. The notations are as follows [4],<br />

x = {x i } ≡ (x 1 ... x n ) (2.25)<br />

u = {u α } ≡ (u 1 ...u m ) (2.26)<br />

where u α = u α (x i ). The partial derivatives will be denoted<br />

u (1) ≡ {u α i } ≡ ∂uα<br />

∂x i (2.27)<br />

u (2) ≡ {u α i j} ≡ ∂2 u α<br />

∂x i ∂x j (2.28)<br />

etc. <strong>for</strong> higher order derivatives.<br />

The total differential operator D i denotes total differentiation <strong>and</strong> is defined as [4]<br />

D i = ∂<br />

∂x i + uα i<br />

∂<br />

∂u α + uα i j<br />

∂<br />

∂u α j<br />

+ u α i jk<br />

∂<br />

∂u α jk<br />

+ ··· (2.29)<br />

9


CHAPTER 2. NOTATION AND THE TOTAL DIFFERENTIAL OPERATOR<br />

Recall that we assume summation over repeteded indicies,i.e., Einsteins summation<br />

convention. Differential algebra suggests that we treat (2.26)–(2.28) as independent<br />

variables even if they are connected <strong>by</strong> the relations<br />

u α i = D i (u α )<br />

u α i j = D j (u α i ) = D j D i (u α ). (2.30)<br />

The variables (2.26)–(2.28) are also referred to as the differential variables (we<br />

differentiate them with respect to x = (x 1 ,... x n )).<br />

Example 2.3<br />

Consider a function f in A, such that f = f (x 1 , x 2 ,y,z) where y = y(x 1 , x 2 ) <strong>and</strong><br />

z = z(x 1 , x 2 ). According to the notation <strong>and</strong> terminology presented above, we<br />

have that<br />

α = 1,2, i = 1,2, u 1 = y, u 2 = z<br />

Now, let f = y + z. Then<br />

D 1 ( f ) = ( ∂<br />

∂x 1 + ∂<br />

u1 1<br />

∂u 1 + ∂<br />

u2 1<br />

} {{<br />

∂u 2 )(y + z) =<br />

}<br />

u α ∂<br />

i ∂u α<br />

= ∂ ∂y ∂ ∂z ∂<br />

(y + z) +<br />

∂x1 ∂x 1 (y + z) +<br />

∂y ∂x 1 (y + z) =<br />

∂z<br />

= 0 + ∂y<br />

∂x 1 · 1 + ∂z<br />

∂x 1 · 1 = ∂y<br />

∂x 1 + ∂z<br />

∂x 1 (2.31)<br />

For D 2 ( f ) we get, in a similar way,<br />

D 2 ( f ) = ( ∂<br />

∂x 2 + ∂<br />

u1 2<br />

∂u 1 + ∂<br />

u2 2<br />

} {{<br />

∂u 2 )(y + z) =<br />

}<br />

u α ∂<br />

i ∂u α<br />

= ∂ ∂y ∂ ∂z ∂<br />

(y + z) +<br />

∂x2 ∂x 2 (y + z) +<br />

∂y ∂x 2 (y + z) =<br />

∂z<br />

= 0 + ∂y<br />

∂x 2 · 1 + ∂z<br />

∂x 2 · 1 = ∂y<br />

∂x 2 + ∂z<br />

∂x 2 . (2.32)<br />

End of Example 2.3<br />

10


2.3. IMPORTANT EQUATIONS<br />

2.3 Important equations<br />

The main notations presented <strong>and</strong> needed <strong>for</strong> the subsequent chapters are equations<br />

(2.25)–(2.29).<br />

x = {x i } ≡ (x 1 ... x n ), u = u α ≡ (u 1 ...u m )<br />

u (1) ≡ {u α i } ≡ ∂uα<br />

∂x i ,<br />

D i = ∂<br />

∂x i + uα i<br />

u (2) ≡ u α i j ≡ ∂2 u α<br />

∂x i ∂x j<br />

∂<br />

∂u α + ∂<br />

uα i j + u α ∂<br />

i jk + ···<br />

∂u α j<br />

∂u α jk<br />

11


3<br />

BASIC LIE GROUP THEORY<br />

In app<strong>lie</strong>d <strong>group</strong> <strong>analysis</strong> we will consider continuous trans<strong>for</strong>mation <strong>group</strong>s. A<br />

trans<strong>for</strong>mation of the variables in a differential equation <strong>for</strong>m a symmetry <strong>group</strong><br />

if it leaves the differential equation invariant [4]. These continuous symmetry<br />

<strong>group</strong>s are represented <strong>by</strong> Lie <strong>group</strong> theory e.g [6, 4, 7]. We will employ Lie’s<br />

theory (Lie, Marius Sophus, 1842–1899) which reduces the <strong>analysis</strong> from large<br />

<strong>group</strong>s to a determination of infinitesimal trans<strong>for</strong>mations. In this chapter we introduce<br />

the basic theory of Lie <strong>group</strong> <strong>analysis</strong> needed <strong>for</strong> finding <strong>symmetries</strong> of<br />

differential equations. Lie’s theory will be presented <strong>and</strong> the meaning of invariance<br />

will be clarified. We will also show how Lie <strong>group</strong>s lead to the concept of<br />

Lie algebras [6, 7].<br />

3.1 Trans<strong>for</strong>mation <strong>group</strong>s<br />

Trans<strong>for</strong>mations of variables can be explained in the following general way [8];<br />

Let the vector f be f = ( f 1 ,..., f m ) of m functions, each depending on n variables<br />

x = (x 1 ,... x n ) in R n . Then f(x) defines a trans<strong>for</strong>mation from R n to R m . If y =<br />

(y 1 ,...y m ) is the point in R m that corresponds to x under the trans<strong>for</strong>mation f, we<br />

have the relation that y = f(x).<br />

Consider the following trans<strong>for</strong>mations in R n [6]:<br />

¯x i = f i (x) (3.1)<br />

13


CHAPTER 3. BASIC LIE GROUP THEORY<br />

where x = (x 1 ,..., x n ), ¯x = ( ¯x 1 ,..., ¯x n ) denotes the trans<strong>for</strong>mation. The <strong>group</strong> action<br />

f = ( f 1 ,..., f n ) <strong>and</strong> its derivatives are continuous. Also let det‖ ∂ f i<br />

‖≠ 0. This<br />

∂x i<br />

means that we can solve x i as a function of f i according to the implicit function<br />

theorem [8, 9], i.e the trans<strong>for</strong>mations are invertible,<br />

x i = ( f −1 ) i ( ¯x). (3.2)<br />

Let us denote trans<strong>for</strong>mation (3.1) <strong>by</strong> T <strong>and</strong> the inverse trans<strong>for</strong>mation (3.2) <strong>by</strong><br />

T −1 .<br />

Geomtrically T is carring a point x ∈ R n to a point ¯x ∈ R n , <strong>and</strong> T −1 carries ¯x back<br />

to x. The identical trans<strong>for</strong>m I is just ¯x i = x i . Let T 1 <strong>and</strong> T 2 be two trans<strong>for</strong>mations<br />

in R n , T 1 : f i 1 (x), T 2 : f i 2 (x) respectively. Then the successive trans<strong>for</strong>mation, T 1T 2<br />

is defined as follow [6]:<br />

¯x = f i 1(x)<br />

¯x = f i 2( ¯x) = f i 2( f i 1(x)) (3.3)<br />

This means that T 1 T 2 (x) carries the point x directly to ¯x. It follows that TT −1 =<br />

T −1 T = I.<br />

Definition 3.1<br />

A <strong>group</strong> of trans<strong>for</strong>mations in R n is a set G of trans<strong>for</strong>mations (3.1), such that<br />

it contains the identity I <strong>and</strong> all trans<strong>for</strong>mations T s <strong>and</strong> Ts<br />

−1 within G <strong>and</strong> all<br />

products T i T j , compare [6].<br />

I ∈ G<br />

T −1<br />

s ∈ G T i T j ∈ G whenever T s ,T i ,T j ∈ G (3.4)<br />

<br />

Exempel 3.1<br />

Consider the translation <strong>group</strong> G of trans<strong>for</strong>mations T a (x) defined as<br />

¯x = f (x) = x + a (3.5)<br />

The successive trans<strong>for</strong>mation T a T b (x) is<br />

¯x = T a T b (x) = T b ( ¯x) = T b (x + a) = x + a + b (3.6)<br />

14


3.2. ONE-PARAMETER GROUP<br />

The identity trans<strong>for</strong>m exists, ¯x = x <strong>for</strong> a = 0, i.e T 0 ∈ G. Since both<br />

a,b ∈ R are arbitrary T a T b ∈ G. For the inverse trans<strong>for</strong>m Ts<br />

−1 we have<br />

x = f −1 ( ¯x) = ¯x − a. (3.7)<br />

That is Ta<br />

−1 = T −a <strong>and</strong> Ta<br />

−1<br />

<strong>group</strong> of trans<strong>for</strong>mations.<br />

∈ G. According to Definition 1, equation (3.5) <strong>for</strong>ms a<br />

End of Example 3.1<br />

3.2 One-parameter <strong>group</strong><br />

Now we consider the one-parameter <strong>group</strong>s. They consist of trans<strong>for</strong>mations of<br />

the <strong>for</strong>m [6]<br />

¯x = f (x,a). (3.8)<br />

In component <strong>for</strong>m, trans<strong>for</strong>mation (3.8) is written as<br />

¯x i = f i (x,a) (3.9)<br />

where x ∈ R n <strong>and</strong> a is a real parameter. Let trans<strong>for</strong>mation (3.8) be the trans<strong>for</strong>m<br />

T a <strong>and</strong> let the identity trans<strong>for</strong>m T a0 be<br />

f (x,a 0 ) = x (3.10)<br />

where a 0 (in general not zero) is the only value of a reducing T a to the identity<br />

trans<strong>for</strong>m.<br />

Definition 3.2<br />

A set G of trans<strong>for</strong>mations T a in R n given <strong>by</strong> (3.9) is called a one-parameter local<br />

<strong>group</strong> if there exist a subinterval U ′ ⊂ U containing a 0 , such that the functions<br />

f i (x,a) satisfy [6],<br />

f i ( f (x,a),b) = f i (x,c) i = 1,...,n (3.11)<br />

∀a, b ∈ U ′ , c ∈ U <strong>and</strong> c = φ(a,b) such that<br />

φ(a,b) = a 0 (3.12)<br />

15


CHAPTER 3. BASIC LIE GROUP THEORY<br />

has a unique solution b <strong>for</strong> any a ∈ U ′ . Equation (3.11) is called the composition<br />

rule.<br />

For a given a, the solution b of equation (3.12) is denoted a −1 (in general not 1/a).<br />

Then Ta<br />

−1 is given <strong>by</strong><br />

f i ( ¯x,a −1 ) = x i (3.13)<br />

<br />

Definition 3.2 may seem quite complicated at first glance. Let us make a little<br />

explanation:<br />

Let T a <strong>and</strong> T b be two trans<strong>for</strong>mations of the type (3.8). Then, <strong>for</strong> the successive<br />

trans<strong>for</strong>m T a T b = T c we must have a <strong>and</strong> b chosen from a smaller interval than c to<br />

assure that the trans<strong>for</strong>m T c is still a trans<strong>for</strong>mation of the trans<strong>for</strong>mations (3.8).<br />

In this smaller interval (a subinterval) we must have the value a 0 that reduces the<br />

trans<strong>for</strong>mations T a <strong>and</strong> T b to the identity trans<strong>for</strong>m, otherwise they alone would<br />

not <strong>for</strong>m a <strong>group</strong> of trans<strong>for</strong>mations according to Definition 3.1. This means that<br />

<strong>for</strong> any parameter a in the subinterval there is only one value of b that reduces<br />

c to a 0 . I.e, there is only one value of b that reduces the successive trans<strong>for</strong>m<br />

T a T b = T c to the identity trans<strong>for</strong>m. This value of b is denoted <strong>by</strong> a −1 <strong>and</strong> is,<br />

consequently, the inverse trans<strong>for</strong>m.<br />

The function<br />

c = φ(a,b) (3.14)<br />

is termed as the <strong>group</strong> composition law [6]. From the equations (3.9) <strong>and</strong> (3.11)<br />

it follows that<br />

φ(a,a 0 ) = a <strong>and</strong> φ(a 0 ,b) = b (3.15)<br />

Indeed, if we, in equation (3.11), set a = a <strong>and</strong> b = a 0 , then we get<br />

f i ( f (x,a),a 0 ) = f (x,a) ⇒ c = φ(a,a 0 ) = a<br />

where equation (3.10) was used at the first step, <strong>and</strong> the result followed <strong>by</strong> identification<br />

with the composition rule. Similarly <strong>for</strong> the second equation in (3.15).<br />

Exampel 3.2<br />

Consider the two trans<strong>for</strong>mations<br />

a) ¯x = ax<br />

b) ¯x = x + ax. (3.16)<br />

16


3.2. ONE-PARAMETER GROUP<br />

We wish to find their composition law. For a) we have:<br />

¯x = f (x,a) = ax (3.17)<br />

Using equation (3.11) we get <strong>for</strong> the LHS<br />

f ( f (x,a),b) = f (ax,b) = abx (3.18)<br />

<strong>and</strong> <strong>for</strong> the RHS<br />

f (x,c) = f (ax,b) = cx. (3.19)<br />

Thus, we get the <strong>group</strong> composition law<br />

c = φ(a,b) = ab (3.20)<br />

To find a −1 , we use equation (3.13) f ( ¯x,a −1 ) = x. Then we get<br />

a −1 ¯x = x<br />

⇒ a −1 (ax) = x<br />

⇒ a −1 = 1 a<br />

(3.21)<br />

For b) we have:<br />

¯x = f (x,a) = x + ax (3.22)<br />

Using equation (3.11) we get <strong>for</strong> the LHS<br />

f ( f (x,a),b) = f (x + ax,b) = x + (a + b + ab)x (3.23)<br />

<strong>and</strong> <strong>for</strong> the RHS<br />

f (x,c) = f (x + ax,b) = x + cx. (3.24)<br />

Thus, we get the <strong>group</strong> composition law<br />

c = φ(a,b) = a + b + ab (3.25)<br />

To find a −1 we use (3.13), f ( ¯x,a −1 ) = x as be<strong>for</strong>e. Then we get<br />

¯x + a −1 ¯x = x(1 + a + a −1 (1 + a))<br />

⇒ x = x(1 + a + a −1 (1 + a))<br />

⇒ a −1 = −<br />

a<br />

(3.26)<br />

1 + a<br />

Here we see that a −1 is not necessarily 1/a.<br />

End of Example 3.2<br />

17


CHAPTER 3. BASIC LIE GROUP THEORY<br />

3.2.1 Canonical parameter<br />

If we choose the <strong>group</strong> parameter a to be canonical, the composition law c =<br />

φ(a,b) reduces <strong>by</strong> definition to, see [6] <strong>and</strong> [4],<br />

c = φ(a,b) = a + b. (3.27)<br />

Consider the <strong>group</strong> trans<strong>for</strong>mation (3.8) with a canonical parameter a.<br />

¯x i = f i (x,a) (3.28)<br />

it is assumed that f i (x,a) are defined in a neighborhood of a = 0, <strong>and</strong> that<br />

f i (x,0) = x i (3.29)<br />

Thus, <strong>for</strong> a canonical parameter a we get the composition rule<br />

f i ( f (x,a),b) = f i (x,c) = f i (x,a + b) i = 1,...,n (3.30)<br />

where a, b are any numerical values in a neighborhood of a = 0. The inverse<br />

trans<strong>for</strong>m (3.13), when using a canonical parameter, is simply<br />

Indeed,<br />

x i = f i ( ¯x,−a) (3.31)<br />

x i = f i ( ¯x,a −1 ) = f i ( f (x,a),a −1 ) = f i (x,a + a −1 ) = f i (x,0) ⇒ a −1 = −a<br />

where in the second step we used (3.30) <strong>and</strong> in the last step we used (3.29). Thus,<br />

local <strong>group</strong>s with a canonical parameter are well defined <strong>by</strong> the initial condition<br />

(3.29), <strong>and</strong> the composition rule (3.30).<br />

3.3 The infinitesimal trans<strong>for</strong>mation <strong>and</strong> Lie’s equations<br />

We will now present the basic theory which reduces the <strong>analysis</strong> of a symmetry<br />

<strong>group</strong> of trans<strong>for</strong>mations to <strong>analysis</strong> of its infintesimal trans<strong>for</strong>m. This is one of<br />

the main ideas in Lie <strong>group</strong> theory, due to the fact that an element in a Lie <strong>group</strong><br />

is a continous differentiable trans<strong>for</strong>mation with an inverse with the same properties;<br />

i.e., a diffeomorphism. See [7, 9].<br />

18


3.3. THE INFINITESIMAL TRANSFORMATION AND LIE’S EQUATIONS<br />

The infintesimal trans<strong>for</strong>mation of a <strong>group</strong> G, where G is a one-parameter <strong>group</strong><br />

of trans<strong>for</strong>mations (3.28), is <strong>obtained</strong> <strong>by</strong> Taylor expansion of (3.28) with respect<br />

to the parameter a in a neigborhood a = 0 [6]. Only keeping the linear part in a<br />

yields,<br />

¯x i = f i (x,a) ≈ f i (x,0) + ∂ f i (x,a)<br />

| a=0 (a − 0) + ··· (3.32)<br />

∂a<br />

thus, we get the infinitesimal trans<strong>for</strong>m<br />

¯x i ≈ x i + aξ i (x) (3.33)<br />

where ξ i (x) is defined as<br />

ξ i (x) = ∂ ∂a [ f i (x,a)] a=0 (3.34)<br />

Geometrically (3.34) defines the tangent vector ξ(x) = (ξ 1 (x),...ξ n (x)) at the point<br />

x. There<strong>for</strong>e, ξ is called the tangent vector field of <strong>group</strong> G.<br />

The tangent vector field is associated with the first order differential operator<br />

X = ξ i ∂<br />

∂x i (3.35)<br />

called the generator of <strong>group</strong> G.<br />

Theorem 3.1<br />

Let G be a local <strong>group</strong> defined <strong>by</strong> (3.8), where the functions f i obey (3.29) <strong>and</strong><br />

(3.30). Let (3.33) be the infintesimal trans<strong>for</strong>mation of <strong>group</strong> G. Then ¯x i = f i (x,a)<br />

solves the system of first order ordinary differential equation, called the Lie equations<br />

[4, 6, 7],<br />

d ¯x i<br />

da = ξi ( ¯x) with initial condition ¯x i | a=0 = x i (3.36)<br />

i.e the solution to (3.36) provides a one-parameter local <strong>group</strong> with a given infintesimal<br />

trans<strong>for</strong>mation.<br />

<br />

19


CHAPTER 3. BASIC LIE GROUP THEORY<br />

For a given infinitesimal trans<strong>for</strong>mation (3.33) or generator (3.35), the trans<strong>for</strong>mations<br />

(3.8) are defined <strong>by</strong> integrating the Lie equations (3.36).<br />

Proof of theorem 3.1<br />

Set b = △a in (3.30). Then f i ( f (x,a),△a) = f i (x,a + △a). Now, exp<strong>and</strong>ing RHS,<br />

f i (x,a + △a) in a Taylor series in a neighborhood where a + △a ≈ a, yields<br />

f i (x,a+△a) ≈ f i (x,a)+ ∂ ∂a [ f i (x,a)]·(a+△a−a) = f i (x,a)+ ∂ ∂a [ f i (x,a)]·△a.<br />

Now, exp<strong>and</strong>ing LHS, f i ( f (x,a),△a) in a Taylor series in a neighborhood where<br />

△a ≈ 0, yields<br />

f i ( f (x,a),△a) ≈ f i ( f (x,a),0) +<br />

∂<br />

∂△a [ f i ( f (x,a),△a)]| △a=0 · (△a − 0)<br />

= f i (x,a) + ∂<br />

∂△a [ f i ( f (x,a),△a)]| △a=0 · (△a).<br />

Setting LHS = RHS <strong>and</strong> using (3.34), we get<br />

∂<br />

∂△a [ f i ( f (x,a),△a)]| △a=0 = ξ i ( f (x,a))<br />

This leads to the result that<br />

∂<br />

∂a [ f i (x,a)] = ξ i ( f (x,a)).<br />

We also need to show that the solution ¯x i = f i (x,a) of (3.36) satisfies the <strong>group</strong><br />

property (3.30). Let<br />

u(b) = f ( ¯x,b) ≡ f ( f (x,a),b), (3.37)<br />

v(b) = f (x,a + b) (3.38)<br />

considered <strong>for</strong> a fixt a. We have that u(0) = v(0) = f (x,a), which solves (3.36).The<br />

property (3.30) is staisfied if we can show that u(b) = v(b) in a neighborhood of<br />

b = 0. Consider th Lie equations <strong>for</strong> u <strong>and</strong> v,<br />

du<br />

db ≡ f ( ¯x,b) = ξ(u)<br />

db<br />

(3.39)<br />

dv f (x,a + b)<br />

≡ = ξ(v)<br />

db db<br />

(3.40)<br />

(3.41)<br />

20


3.3. THE INFINITESIMAL TRANSFORMATION AND LIE’S EQUATIONS<br />

with the initial conditions u(0) = f (x,a), v(0) = f (x,a) respectively. Let z = u(b)<br />

<strong>and</strong> z = v(b), then z solves<br />

dz<br />

= ξ(z) (3.42)<br />

db<br />

with the initial condition z| b=0 = f (x,a). According to the uniqueness theorem<br />

<strong>for</strong> initial value problems [10] we must have that u(b) = v(b). Thus, the <strong>group</strong><br />

property (3.30) is satisfied [6].<br />

□<br />

Exampel 3.3<br />

Consider the infinitesimal trans<strong>for</strong>mation <strong>group</strong> [4],<br />

¯x ≈ x + ax 2 ȳ ≈ y + axy. (3.43)<br />

It has generator<br />

X = ξ 1 ∂ ∂x + ξ2 ∂ ∂y , (3.44)<br />

where ξ 1 (x,y) = x 2 <strong>and</strong> ξ 2 (x,y) = xy. The Lie equations are<br />

d ¯x<br />

da = ¯x2 with initial condition ¯x| a=0 = x<br />

dȳ<br />

da = ¯xȳ with initial condition ȳ| a=0 = y. (3.45)<br />

Solving this system of differential equations <strong>by</strong> seperation of variables yields,<br />

¯x −1 d ¯x = da ⇒ − 1¯x = a +C 1<br />

⇒ ¯x = − 1<br />

a +C 1<br />

dȳ<br />

ȳ = ¯xda = − 1 da<br />

a +C 1<br />

⇒ ȳ = C 2<br />

a +C 1<br />

⇒<br />

¯x = − 1<br />

a +C 1<br />

,ȳ = C 2<br />

a +C 1<br />

(3.46)<br />

21


CHAPTER 3. BASIC LIE GROUP THEORY<br />

Using the initial conditions yields<br />

C 1 = − 1 x<br />

C 2 = − y x . (3.47)<br />

We get the one-parameter local <strong>group</strong><br />

¯x =<br />

x<br />

1 − ax<br />

ȳ =<br />

y<br />

1 − ax<br />

(3.48)<br />

End of Example 3.3<br />

3.4 Invariant functions <strong>and</strong> equations<br />

The meaning of invariant functions <strong>and</strong> invariant equations will now be clarified.<br />

3.4.1 Invariant function<br />

Definition 3.3<br />

A function F(x) is called invariant of a <strong>group</strong> G of trans<strong>for</strong>mations (3.28) if<br />

F( f (x,a)) = F(x) (3.49)<br />

identically in x <strong>and</strong> a in a neighborhood a = 0 [6].<br />

Theorem 3.2<br />

A function F(x) is an invariant of a <strong>group</strong> G with generator X, (3.35), if <strong>and</strong> only<br />

if it solves the homogeneous linear partial differential equation [6]<br />

X(F) ≡ ξ i ∂F(x)<br />

∂x i = 0. (3.50)<br />

<br />

Proof of Theorem 3.2<br />

Let F(x) be invariant. Consider F( f (x,a)) exp<strong>and</strong>ed in a Taylor series in a neighborhood<br />

around a = 0, i.e<br />

F( f (x,a)) ≈ F(x + aξ(x)) ≈ F(x) + aξ i ∂F(x)<br />

∂x i<br />

<br />

22


3.4. INVARIANT FUNCTIONS AND EQUATIONS<br />

The invariance condition F(x) = F( f (x,a)) yields (3.50).<br />

Conversely, let F(x) be any solution to the differential equation (3.50). Since<br />

(3.50) is valid at any point, consider it at ¯x = f (x,a), i.e,<br />

ξ i ∂F( ¯x)<br />

( ¯x)<br />

∂ ¯x i = 0<br />

Hence, invoking the Lie equations:<br />

dF( f (x,a))<br />

da<br />

=<br />

∂F( ¯x) d f i (x,a)<br />

∂ ¯x i = ξ i ∂F( ¯x) ¯x<br />

da ∂ ¯x i = 0<br />

Since F( f (x,0)) = F(x) we can conclude that u(a) = F( f (x,a)) solves the initial<br />

value problem<br />

du<br />

da = 0<br />

u| a=0 = F(x)<br />

The solution to the initial value problem is, <strong>by</strong> direct calculation, given <strong>by</strong> u =<br />

F(x) <strong>and</strong> this solution is unique according to the uniqueness theorem [10]. Hence,<br />

the two solutions u(a) = F( f (x,a)) <strong>and</strong> u = F(x) are identical <strong>for</strong> any x, i.e F( f (x,a)) =<br />

F(x) which is the invariance condition.<br />

Remark:<br />

From equation (3.50) we can see that ξ i really is a tangent vector field. Rewrite<br />

X(F) ≡ ξ i ∂F(x) = 0, using vector notation, i.e ξ i ≡ ξ <strong>and</strong> ∂ ≡ ∇, we write (3.50)<br />

∂x i<br />

∂x i<br />

as ξ • ∇F = 0. Recall from basic calculus that ∇F defines a normal vector to<br />

the surface described <strong>by</strong> F. If ξ • ∇F = 0 we must have that ξ is orthogonal to the<br />

normal vector to F, i.e ξ must be the tangent vector to F at the point x = (x 1 ... x n ).<br />

See e.g. [8].<br />

□<br />

3.4.2 Invariant equations<br />

Consider a system of equations [6]<br />

F σ (x) = 0 σ = 1... s, (3.51)<br />

23


CHAPTER 3. BASIC LIE GROUP THEORY<br />

where x = (x 1 ... x n ) ∈ R n <strong>and</strong> s < n. Impose that <strong>for</strong> all x satisfying (3.51) we have<br />

that the Jacobian matrix [8] is of rank s, i.e<br />

rank‖ ∂F σ(x)<br />

∂x i ‖ = s. (3.52)<br />

Recall that <strong>for</strong> a system of s equations with n variables where s < n, we get a<br />

solution involving (n − s) parameters if the rank of the system is equal to the<br />

number of equation. These parameters will span the set of solutions. Thus, the<br />

solution set x satisfying (3.51), F σ (x) = 0 is an (n − s)-dimensional surface in R n<br />

or, more precise, an (n − s)-dimensional manifold M ⊂ R n [6],[7], [9].<br />

Definition 3.4<br />

The system (3.51) is said to be invariat, or admitted <strong>by</strong> a <strong>group</strong> G of trans<strong>for</strong>mations<br />

¯x = f (x,a) if<br />

F σ ( ¯x) = 0 σ = 1... s (3.53)<br />

whenever x solves F σ (x) = 0. This means that any point in the manifold M is<br />

carried to another point in M. Thus, any path curve of G passing through a point<br />

x ∈ M <strong>lie</strong>s in M. M is an invariant manifold [6].<br />

Theorem 3.3<br />

The system (3.51) is invariant under the <strong>group</strong> G with infinitesimal generator X if<br />

<strong>and</strong> only if<br />

XF σ | (3.51) = 0 σ = 1... s (3.54)<br />

where | (3.51) means evaluated on its manifold [6].<br />

<br />

<br />

Provided that F σ <strong>and</strong> XF σ are analytic functions in a neigborhood of M, the<br />

invariance test (3.54) can be written as<br />

XF σ (x) = λ ν σ(x)F ν (x) σ = 1... s, (3.55)<br />

where λ ν σ are coefficients bounded in M [6]. For a more rigorous treatment of the<br />

connection between Lie <strong>group</strong>s <strong>and</strong> Manifolds see [7].<br />

24


3.5. MULTI-PARAMETER GROUPS AND LIE ALGEBRAS<br />

3.5 Multi-parameter <strong>group</strong>s <strong>and</strong> Lie algebras<br />

In previous sections we considered one–parameter <strong>group</strong>s. We will now extend<br />

the theory to multi–parameter <strong>group</strong>s, i.e instead of one parameter a, we have r–<br />

parameters, a = (a 1 ,...,a r ). This will lead to the concept of Lie algebras, which is<br />

frequently used in theoretical physics, especially in quantum mechanics. See e.g.<br />

[11].<br />

3.5.1 Multi–parameter <strong>group</strong>s<br />

Multi–parameter <strong>group</strong>s can be understood as one–parameter <strong>group</strong>s with the difference<br />

that the parameter a is now a vector–parameter, a = (a 1 ,...,a r ) [6]. An<br />

invertible trans<strong>for</strong>mation T a : R n → R n , or more precics T a r ≡ T a 1T a 2 ···T a r, in a<br />

neighborhood of a = 0 ≡ (0,...,0) is<br />

¯x = f (x,a) (3.56)<br />

We have the same defintion <strong>for</strong> a local r–parameter <strong>group</strong> as <strong>for</strong> a local one–<br />

parameter <strong>group</strong>, Definition 3.2. I.e, the trans<strong>for</strong>mation (3.56) is said to <strong>for</strong>m a<br />

local r–parameter <strong>group</strong> if<br />

f ( f (x,a),b) = f (x,c) (3.57)<br />

where b = (b 1 ,...,b r ) <strong>and</strong> the composition function c = c(a,b) is now a vector<br />

function with components<br />

c α = φ α (a,b), α = 1,...,r. (3.58)<br />

with the same properties as in Defintion 3.2, see [6],[7].<br />

3.5.2 Short outline of Lie algebra<br />

A Lie algebra is a vector space spanned <strong>by</strong> infinitesimal generators X. Generators<br />

of an r-parameter local <strong>group</strong>s, i.e multi-parameter <strong>group</strong>s leads to the concept of<br />

Lie algebras. Almost all in<strong>for</strong>mation in a <strong>group</strong> is contained in its Lie algebra [7].<br />

We define a Lie algebras of operators as follow [6]:<br />

25


CHAPTER 3. BASIC LIE GROUP THEORY<br />

Definition 5.2<br />

A Lie algebra is a vector space L of spanned <strong>by</strong> operators X α = ξα(x) i ∂<br />

following property: If the operators<br />

X 1 = ξ i 1(x) ∂<br />

∂x i ,<br />

are elements of L, then their commutator<br />

∂x i<br />

with the<br />

X 2 = ξ i 2(x) ∂<br />

∂x i (3.59)<br />

[X 1 , X 2 ] ≡ X 1 X 2 − X 2 X 1 = (X 1 (ξ i 2) − X 2 (ξ i 1)) ∂<br />

∂x i (3.60)<br />

is also an element of L. We denote the dimension of the Lie algebra <strong>by</strong> L r [6].<br />

The commutator is bilinear, skew-symetric <strong>and</strong> satisfies the Jaco<strong>by</strong> Identity [6, 7].<br />

Consider a basis of a Lie algebra L r given <strong>by</strong><br />

X α = ξ i α(x) ∂<br />

∂x i α = 1,...,r. (3.61)<br />

Then, any operator X ∈ L r can be written as a linear combination of this basis,<br />

i.e X = c α X α . Let X α <strong>and</strong> X β be basis vectors of L r then [ X α , X β<br />

]<br />

∈ Lr . Thus,<br />

due to the bilinearity property of the commutator (which is a consequence of its<br />

definition), we have that [6]<br />

[X α , X β = c γ αβ X γ. (3.62)<br />

This provides a simple test <strong>for</strong> the linear span of a Lie algebra. Thus, given r<br />

independent operators, if (3.62) hold they span an r-dimensional Lie algebra. The<br />

constants c γ αβ is called the structure constants [6],[7] of that Lie algebra.<br />

Consider the composition of r one-parameter <strong>group</strong>s T a = T a 1 ···T a r, generated <strong>by</strong><br />

the base operators X α , via the Lie equations<br />

d ¯x i<br />

da α = ξi α( ¯x), ¯x i | a α =0 = x i (3.63)<br />

where i = 1,...,n <strong>and</strong> α = 1,...,r. T a is an r-parameter local <strong>group</strong> G r if the operators<br />

X α <strong>for</strong>m an r-dimensional Lie algebra L r .<br />

The invariance conditions stated in previous sections is the same <strong>for</strong> multiparameter<br />

<strong>group</strong>s. This is due to the fact that we may consider a multi-parameter<br />

<strong>group</strong> as a composition a several one-parameter <strong>group</strong>s [6].<br />

<br />

26


4<br />

DIFFERENTIAL ALGEBRA<br />

Differential algebra furnishes us with a convenient language to use in modern<br />

<strong>group</strong> <strong>analysis</strong>. Some of the notations that will be presented here have already<br />

been introduced in chapter 2. Here, we will extend <strong>and</strong> generlaize the theory<br />

presented ear<strong>lie</strong>r.<br />

4.1 The space A<br />

Let us denote an arbitrary sequence <strong>by</strong> z [6],<br />

z = (x,u,u (1) ,u (2) ,...) (4.1)<br />

with elements z ν , ν 1. Here<br />

z i = x i<br />

i = 1...n<br />

z n+α = u α α = 1...m (4.2)<br />

<strong>and</strong> the remaning elements are derivatives of u.<br />

A finite subsequence of z will be denoted <strong>by</strong> [z].<br />

Definition 4.1<br />

A differential function f ([z]) is a locally analytic function. The order of f ([z]) is<br />

the highest order of derivatives in the differential function f ([z]). We denote the<br />

set of all differential functions of finite order <strong>by</strong> A [6].<br />

27


CHAPTER 4. DIFFERENTIAL ALGEBRA<br />

<br />

Properties in the space A:<br />

If f ([z]) ∈ A <strong>and</strong> g([z]) ∈ A then,<br />

1. a f + bg ∈ A <strong>for</strong> any constants a,b.<br />

2. f g ∈ A<br />

3. D i ( f ) ∈ A<br />

The first <strong>and</strong> second property follow from the fact that both f <strong>and</strong> g are analytic<br />

functions. For the third property we know that the derivative of an analytic<br />

function is analytic. Furthere more, from the expression <strong>for</strong> the infinite operator<br />

D i , (2.29), we see that it truncates when acting on a differential function, <strong>and</strong><br />

increasing the order of the differential function <strong>by</strong> one. i.e, if ord( f ) = s then<br />

ord(D i ( f )) = s + 1 is still finite. According to Definition 4.1, statement three follows<br />

[6].<br />

4.2 Extended point trans<strong>for</strong>mation <strong>group</strong><br />

Extended point trans<strong>for</strong>mation means that we in the usual point trans<strong>for</strong>mation<br />

(3.8) take the derivities into account. In that way we arrive at a trans<strong>for</strong>mation<br />

including n + s variables, where n is the number of variables <strong>and</strong> s is its derivities<br />

up to order s. We will consider extended point trans<strong>for</strong>mations in the plane <strong>and</strong><br />

then generilize it to many variables [6].<br />

Definition 4.2<br />

An extended point trans<strong>for</strong>mation up to the sth-derivative is called the sth-prolongation.<br />

<br />

4.2.1 Extended pointtrans<strong>for</strong>mation in the plane<br />

We start <strong>by</strong> looking at the point trans<strong>for</strong>mations in the plane [6], [4].<br />

¯x = f (x,y,a) ≈ x + aξ(x,y)<br />

ȳ = ϕ(x,y,a) ≈ y + aη(x,y) (4.3)<br />

28


4.2. EXTENDED POINT TRANSFORMATION GROUP<br />

where x is the independent variable <strong>and</strong> y the differential variable. we have that<br />

ȳ ′ ≡ dȳ<br />

d ¯x = ϕ xdx + ϕ y dy<br />

f x dx + f y dy = ϕ x + y ′ ϕ y<br />

f x + y ′ = D xϕ<br />

f y D x f<br />

(4.4)<br />

By adding (4.4) to the system (4.3) of point trans<strong>for</strong>mation, we arrive at a trans<strong>for</strong>mation<br />

<strong>group</strong> <strong>for</strong> three variables x,y,y ′ , where x is independent <strong>and</strong> y is the<br />

differential variable. We treat them as if they where all independent of each other,<br />

as be<strong>for</strong>e. If, <strong>for</strong> a canonical parameter a, (4.3) <strong>for</strong>ms a one-parameter <strong>group</strong> with<br />

the <strong>group</strong> property (3.30), i.e<br />

¯x = f ( ¯x,ȳ,b) = f (x,y,a + b)<br />

ȳ = ϕ( ¯x,ȳ,b) = ϕ(x,y,a + b) (4.5)<br />

<strong>and</strong><br />

ȳ ′ = ψ(x,y,y ′ ,a) ≡ D x(ϕ(x,y,a))<br />

D x ( f (x,y,a))<br />

(4.6)<br />

then ȳ ′ ≡ ψ( ¯x,ȳ,ȳ ′ ,a) must be equal to ψ(x,y,y ′ ,a + b) <strong>for</strong> (4.3) <strong>and</strong> (4.4) to <strong>for</strong>m<br />

a one-parameter <strong>group</strong>. i.e we need to show that<br />

ȳ ′ ≡ ψ( ¯x,ȳ,ȳ ′ ,a) = ψ(x,y,y ′ ,a + b) (4.7)<br />

To do that, consider the following expression <strong>for</strong> ȳ ′ :<br />

ȳ ′ D<br />

= ¯ x (ϕ( ¯x,ȳ,a))<br />

D¯<br />

x ( f ( ¯x,ȳ,a))<br />

(4.8)<br />

multiplying the numerator <strong>and</strong> the denominator of equation (4.8) <strong>by</strong> D x ( f (x,y,a)),<br />

we get<br />

ȳ ′ = D x( f (x,y,a)) D¯<br />

x (ϕ( ¯x,ȳ,a))<br />

D x ( f (x,y,a)) D¯<br />

x ( f ( ¯x,ȳ,a)) = D x(ϕ( ¯x,ȳ,a))<br />

D x ( f ( ¯x,ȳ,a))<br />

= D x(ϕ(x,y,a + b))<br />

D x ( f (x,y,a + b)) = ψ(x,y,y′ ,a + b) (4.9)<br />

where we in the first step used the chain rule [4],<br />

D x = D x ( f (x,y,a)) ¯ D x (4.10)<br />

29


CHAPTER 4. DIFFERENTIAL ALGEBRA<br />

<strong>and</strong> in the second step we used the <strong>group</strong> property (4.5).<br />

Consider the one-parameter <strong>group</strong> (4.3) of infinitesimal trans<strong>for</strong>mations. The generator<br />

to (4.3) is<br />

X = ξ(x,y) ∂ ∂x + η(x,y) ∂ ∂y<br />

(4.11)<br />

The first order extended generator has the <strong>for</strong>m<br />

where<br />

X (1) = ξ(x,y) ∂ ∂x + η(x,y) ∂ ∂y + ζ ∂<br />

(1)<br />

∂y ′ (4.12)<br />

ζ (1) = D x (η) − y ′ D x (ξ) (4.13)<br />

indeed, consider (4.3) in (4.4)<strong>and</strong> exp<strong>and</strong> <strong>for</strong> the parameter a in a neighborhood<br />

of 0, <strong>and</strong> only keeping linear terms in a, we get<br />

ȳ ′ = D xϕ<br />

D x f ≈ D x(y + aη(x,y))<br />

D x (x + aξ(x,y)) = y′ + aD x (η)<br />

1 + aD x (ξ)<br />

≈ [y ′ + aD x (η)] · [1 − aD x (ξ)] ≈ y ′ + aD x (η) − y ′ aD x (ξ)<br />

= y ′ + a[D x (η) − y ′ D x (ξ)] = y ′ + aζ (1) (4.14)<br />

The (s + 1)-times extended infintesimal trans<strong>for</strong>mations, <strong>and</strong> their generators are<br />

<strong>obtained</strong> recursively:<br />

ȳ (s+1) ≈ [y (s+1) +aD x (ζ s )]·[1−aD x (ξ)] ≈ y (s+1) +a[D x (ζ s )−y (s+1) D x (ξ)] (4.15)<br />

By setting ȳ (s+1) ≈ y (s+1) + aζ s+1 , we get the (s + 1)-times prolongation <strong>for</strong>mula<br />

ζ s+1 = D x (ζ s ) − y (s+1) D x (ξ) s = 1,2,.... (4.16)<br />

Exampel 4.1<br />

The spiral trans<strong>for</strong>m has the generator [4]<br />

X = (x − y) ∂ ∂x + (x + y) ∂ ∂y . (4.17)<br />

30


4.2. EXTENDED POINT TRANSFORMATION GROUP<br />

Its twice extended generator has the <strong>for</strong>m<br />

X (2) = ξ(x,y) ∂ ∂x + η(x,y) ∂ ∂y + ζ (1)(x,y,y ′ ) ∂<br />

∂y ′<br />

+ ζ (2) (x,y,y ′ ,y ′′ ) ∂ , (4.18)<br />

∂y ′′<br />

where ξ(x,y) = x − y, η(x,y) = x + y <strong>and</strong> ζ (1) ,ζ (2) is given <strong>by</strong>:<br />

ζ (1) = D x (η) − y ′ D x (ξ)<br />

= D x (x + y) − y ′ D x (x − y)<br />

= 1 + y ′ − y ′ (1 − y ′ ) = 1 + (y ′ ) 2 . (4.19)<br />

Thus, the first extended generator is<br />

X (1) = (x − y) ∂ ∂x + (x + y) ∂ ∂y + (1 + (y′ ) 2 ) ∂<br />

∂y ′ . (4.20)<br />

To get the second one we need to find ζ (2) .<br />

ζ (2) = D x (ζ (1) ) − y ′′ D x (ξ)<br />

= D x (1 + (y ′ ) 2 ) − y ′′ D x (x − y)<br />

= 0 + 2y ′ y ′′ − y ′′ (1 − (y ′ ) 2 ) = y ′′ ((y ′ ) 2 + 2y ′ − 1). (4.21)<br />

Hence, the second order extended generator is<br />

X (2) = (x − y) ∂ ∂x + (x + y) ∂ ∂y + (1 + (y′ ) 2 ) ∂<br />

∂y ′<br />

+ (y ′′ ((y ′ ) 2 + 2y ′ − 1)) ∂ . (4.22)<br />

∂y ′′<br />

End of Example 4.1<br />

In physical applications we often consider functions involving many differential<br />

variables <strong>and</strong> one independent variable, e.g , three spatial coordinates depending<br />

on time is a common case. More general, consider point trans<strong>for</strong>mations<br />

¯x = f (x,u,a) ≈ x + aξ(x,u)<br />

¯ u α = ϕ(x,u,a) ≈ u α + aη α (x,y) (4.23)<br />

31


CHAPTER 4. DIFFERENTIAL ALGEBRA<br />

with one independent variable x <strong>and</strong> m differential variables u = (u 1 ,,u m ). The<br />

trans<strong>for</strong>mation of u¯<br />

α 1<br />

is <strong>obtained</strong> in precisely the same way as (4.4) where <strong>obtained</strong>.<br />

We arrive at the prolongation <strong>for</strong>mula [6],<br />

ζ1 α = D x (η α ) − u α 1 D x (ξ)<br />

ζs+1 α = D x (ζs α ) − u α s+1D x (ξ) s = 1,2,.... (4.24)<br />

4.2.2 Point trans<strong>for</strong>mation involving many variables<br />

In the previous section we considered point trans<strong>for</strong>amtions in the plane <strong>and</strong> point<br />

trans<strong>for</strong>mations involving several differential variables <strong>and</strong> one independent variable,<br />

<strong>and</strong> found their prolongation <strong>for</strong>mulas. In this section we generlize these<br />

results into functions depending on several differential <strong>and</strong> several independent<br />

variables [6]. This case is very common in physical applications in the study<br />

of vector fields. For instance, physical systems described <strong>by</strong> Maxwell’s equations<br />

[2]. In that case we have the components of the electromagentic fields as<br />

the differential variables, <strong>and</strong> the spatial coordiantes <strong>and</strong> time as the independent<br />

variables.<br />

Let G be a one-parameter <strong>group</strong> of trans<strong>for</strong>mations, with independent variables<br />

x = (x 1 ,..., x n ) <strong>and</strong> differential variables u = (u 1 ,...,u m ):<br />

¯x i = f i (x,u,a), f i | a=0 = x i (4.25)<br />

ū α = ϕ α (x,u,a), ϕ α | a=0 = u α (4.26)<br />

with generator<br />

where<br />

X = ξ i (x,u) ∂<br />

∂x i + ηα (x,u) ∂<br />

∂u α (4.27)<br />

ξ i = ∂ f i<br />

∂a | a=0<br />

η α = ∂ϕα<br />

∂a | a=0 (4.28)<br />

Let ¯D i denote the total derivatives in the new variables ¯x i . The chain rule yields:<br />

D i = ( ∂ f j<br />

∂x i + ∂ f j<br />

uβ i<br />

∂u β ) D¯<br />

j ≡ D i ( f j ) D¯<br />

j (4.29)<br />

32


4.2. EXTENDED POINT TRANSFORMATION GROUP<br />

The connection between the new differential variables <strong>and</strong> their derivatives are<br />

ū α i = ¯D i (ū α ),<br />

ū α i j = ¯ D j ¯D i (ū α ) ≡ ¯ D j (ū α i ) (4.30)<br />

To find the first prolongation, consider the following relation,<br />

D i (ϕ α ) ≡ D i (ū α ) = D i ( f j ) ¯ D j (ū α ) = (ū α j )D i ( f j )<br />

⇒ ū α j = D i(ϕ α )<br />

D i ( f j )<br />

togheter with the following infinitesimal trans<strong>for</strong>mations<br />

(4.31)<br />

f i = x i + aξ i ,ϕ α = u α + aη α <strong>and</strong> set ū α i = u α i + aζ α i (4.32)<br />

then, substituting the first two equations of (4.32) into (4.31) one obtains the first<br />

prolongation <strong>for</strong>mula [6],<br />

ζ α i = D i (η α ) − u α j D i (ξ j ). (4.33)<br />

The first extended generator X (1) to (4.27) is written<br />

X (1) = ξ i (x,u) ∂<br />

∂x i + ηα (x,u) ∂<br />

∂u α + ζα i<br />

∂<br />

∂u α i<br />

The second prolongation <strong>for</strong>mula is given <strong>by</strong><br />

. (4.34)<br />

ζ α i 1 i 2<br />

= D i2 (ζ α i 1<br />

) − u α ji 1<br />

D i2 (ξ j ), (4.35)<br />

<strong>and</strong> the twice extended generator is written<br />

X (2) = ξ i (x,u) ∂<br />

∂x i + ηα (x,u) ∂<br />

∂u α + ζα i<br />

∂<br />

∂u α i<br />

The higher order prolongations are defined recursively as<br />

+ ζ α i 1 i 2<br />

∂<br />

∂u α i 1 i 2<br />

. (4.36)<br />

ζ α i 1 ...i s<br />

= D is (ζ α i 1 ...i s−1<br />

) − u α ji 1 ...i s−1<br />

D is (ξ j ), (4.37)<br />

where we have summation in repeteded indicies <strong>for</strong> every s.<br />

The extension of generators is invariant under any change of variables [6]. For<br />

a Lie algebra L r with r linearly independent operators of the <strong>for</strong>m (4.27), their extension<br />

up to the sth-prolongation also span a Lie algebra with the same structure<br />

constants as L r . This means that the relations between the elements in the two<br />

algebras are the same, i.e they are isomorphic [7].<br />

33


CHAPTER 4. DIFFERENTIAL ALGEBRA<br />

4.2.3 Extended generators <strong>for</strong> differential equations with two independent<br />

<strong>and</strong> one dependent variable<br />

In this section we will find the general extended generators <strong>for</strong> the special case of<br />

two independent variables x <strong>and</strong> y, <strong>and</strong> one dependent variable u = u(x,y) as done<br />

in [6]. We will consider partial differential equations up to second order. This<br />

kind of partial differential equations arises alot in physical applications, <strong>for</strong> example<br />

in gas dynamics where motion of gases are studied, particulary when they<br />

interact with moving objects.<br />

We consider a generator of a point trans<strong>for</strong>mation <strong>group</strong> of the <strong>for</strong>m<br />

X = ξ 1 (x,y,u) ∂ ∂x + ξ2 (x,y,u) ∂ ∂y + η(x,y,u) ∂ ∂u<br />

(4.38)<br />

The twice extension of (4.38) has the <strong>for</strong>m<br />

X = ξ 1 ∂ ∂ ∂x +ξ2 ∂y +η ∂ ∂u +ζ ∂ ∂ ∂ ∂ ∂<br />

1 +ζ 2 +ζ 11 +ζ 12 +ζ 22 . (4.39)<br />

∂u x ∂u y ∂u xx ∂u xy ∂u yy<br />

The extendent tangent vector field components ζ i <strong>and</strong> ζ i j are given <strong>by</strong> (4.33) <strong>and</strong><br />

(4.36) respectively.<br />

<strong>and</strong><br />

ζ 1 = D x (η) − u x D x (ξ 1 ) − u y D x (ξ 2 )<br />

ζ 2 = D y (η) − u x D y (ξ 1 ) − u y D y (ξ 2 ) (4.40)<br />

ζ 11 = D x (ζ 1 ) − u xx D x (ξ 1 ) − u xy D x (ξ 2 )<br />

ζ 12 = D y (ζ 1 ) − u xx D y (ξ 1 ) − u xy D y (ξ 2 )<br />

ζ 22 = D y (ζ 2 ) − u xy D y (ξ 1 ) − u yy D y (ξ 2 ). (4.41)<br />

Using the definitions of D x <strong>and</strong> D y one obtains:<br />

ζ 1 = η x + u x η u − u x ξ 1 x − (u x ) 2 ξ 1 u − u y ξ 2 x − u x u y ξ 2 u, (4.42)<br />

ζ 2 = η y + u y η u − u x ξ 1 y − u x u y ξ 1 u − u y ξ 2 y − (u y ) 2 ξ 2 u. (4.43)<br />

34


4.2. EXTENDED POINT TRANSFORMATION GROUP<br />

For the second order prolongations we get,<br />

ζ 11 = η xx + 2u x η xu + u xx η u + (u x ) 2 η uu − 2u xx ξ 1 x − u x ξ 1 xx<br />

− 2(u x ) 2 ξ 1 xu − 3u x u xx ξ 1 u − (u x ) 3 ξ 1 uu − 2u xy ξ 2 x − u y ξ 2 xx<br />

− 2u x u y ξ 2 xu − (u y u xx + 2u x u xy )ξ 2 u − (u x ) 2 u y ξ 2 uu, (4.44)<br />

ζ 12 = η xy + u y η xu + u x η yu + u xy η u + u x u y η uu − u xy (ξ 1 x + ξy)<br />

2<br />

− u x ξ 1 xy − u xx ξy 1 − u x u y (ξ 1 xu + ξuy) 2 − (u x ) 2 ξyu<br />

1<br />

− (2u x u xy + u y u xx )ξu 1 − (u x ) 2 u y ξuu 1 − u y ξ 2 xy − u yy ξ 2 x<br />

− (u y ) 2 ξ 2 xu − (2u y u xy + u x u yy )ξu 2 − u x (u y ) 2 ξuu, 2 (4.45)<br />

ζ 22 = η yy + 2u y η yu + u yy η u + (u y ) 2 η uu − 2u yy ξ 2 y − u y ξ 2 yy<br />

− 2(u y ) 2 ξ 2 yu − 3u y u yy ξ 2 u − (u y ) 3 ξ 2 uu − 2u xy ξ 1 y − u x ξ 1 yy<br />

− 2u x u y ξ 1 yu − (u x u yy + 2u y u xy )ξ 1 u − u x (u y ) 2 ξ 1 uu. (4.46)<br />

This is the general prolongation <strong>for</strong>mulas <strong>for</strong> a second order partial differential<br />

equation with one dependent variable <strong>and</strong> two independent variables. Examples<br />

on extended generators <strong>and</strong> the use of prolongation <strong>for</strong>mulas can be seen in chapter<br />

9.<br />

35


5<br />

OPERATORS<br />

In this chapter some differential operators that play a central role in the development<br />

of <strong>conservation</strong> <strong>laws</strong> will be presented. We will mainly state their definitions,<br />

<strong>and</strong> some useful Theorems <strong>and</strong> Lemmas. This is the last tools we need<br />

be<strong>for</strong>e we procced to Part II. However, one could move on directly to Part II <strong>and</strong><br />

just look up the definitions, theorems <strong>and</strong> lemmas when needed.<br />

5.1 Euler-Lagrange operator<br />

We start <strong>by</strong> defining the Euler-Lagrange operator in the space A.<br />

Definition 5.1<br />

The Euler-Lagrange operator is defined <strong>by</strong> the <strong>for</strong>mal sum [6],<br />

δ<br />

δu α =<br />

∂ ∞<br />

∂u α + ∑ (−1) s ∂<br />

D i1 ··· D is<br />

s=0<br />

∂u α α = 1...m (5.1)<br />

i 1 ...i s<br />

where we have summation over repeted indicies i 1 ...i s <strong>for</strong> every s.<br />

Exampel 5.1<br />

Consider the Euler-Lagrange operator acting on a function f = f (x,u,u (1) ,u (2) )<br />

where x = (x,y,z) are the independent variables, u is the only differential variable<br />

<br />

37


CHAPTER 5. OPERATORS<br />

<strong>and</strong> ( f ) = 2. I.e in (5.1) we have α = 1,i = 3 <strong>and</strong> s = 2. The Euler-Lagrange<br />

operator takes the <strong>for</strong>m<br />

δ<br />

δu = ∂ [<br />

]<br />

∂u + (−1) ∂ ∂ ∂<br />

D x + D y + D z<br />

∂u x ∂u y ∂u z<br />

+ (−1) 2[ D 2 ∂<br />

x + D 2 ∂<br />

y + D 2 ∂<br />

z<br />

∂u xx ∂u yy ∂u zz<br />

∂ ∂ ∂<br />

]<br />

+ D x D y + D x D z + D y D z . (5.2)<br />

∂u xy ∂u xz ∂u yz<br />

End of Example 4.1<br />

Lemma 5.1<br />

The total derivative <strong>and</strong> ∂/∂u α commutes [6], i.e<br />

∂<br />

[D i ,<br />

∂u α ] ≡ D ∂<br />

i<br />

∂u α − ∂<br />

∂u α D i = 0 (5.3)<br />

<br />

This is veryfied <strong>by</strong> direct calculation,<br />

D i<br />

∂<br />

∂u α ≡ ( ∂<br />

∂x i + ∂<br />

uα i<br />

∂u α + uα i j<br />

= ∂2<br />

∂x i ∂u α + uα i<br />

∂<br />

∂u α D i ≡<br />

∂<br />

∂u α ( ∂<br />

∂x i + uα i<br />

= ∂2<br />

∂ 2<br />

∂<br />

∂u α j<br />

∂ 2 u α + uα i j<br />

∂<br />

∂u α + uα i j<br />

+ ···) ∂<br />

∂u α<br />

∂u α j<br />

∂<br />

∂ 2<br />

∂u α j<br />

∂ 2<br />

∂uα<br />

+ ···<br />

+ ···)<br />

∂u α ∂x i + ∂uα i ∂<br />

}{{}<br />

∂u α ∂u α + uα i<br />

∂ 2 u α + ∂uα i j<br />

}{{}<br />

∂u α<br />

0<br />

0<br />

= ∂2<br />

∂x i ∂u α + uα i<br />

∂ 2<br />

∂ 2 u α + uα i j<br />

∂u α j<br />

∂ 2<br />

∂uα<br />

+ ···<br />

∂<br />

∂u α j<br />

∂ 2<br />

+ u α i j<br />

∂u α ∂u α j<br />

+ ···<br />

<strong>and</strong> Lemma 5.1 follows.<br />

38


5.2. LIE-BÄCKLUND OPERATORS<br />

Lemma 5.2<br />

δ<br />

δu α D i = 0 ∀α (5.4)<br />

Lemma 5.1 is verified <strong>by</strong> direct calculation. The following theorem is a direct<br />

consequense of Lemma 5.2 [6].<br />

Theorem 5.1<br />

A function f (x,u,u (1) ...u (s) ∈ A with independent variables x = (x 1 ... x n ) an differential<br />

variables u = (u 1 ...u m is the divergence of a vector field H = (h 1 ...h n ),<br />

where h i = h i (x,u,u (1) ...u (s−1) ) ∈ A i.e.<br />

if <strong>and</strong> only if<br />

f = div(H) ≡ D i (h i ) (5.5)<br />

δ f<br />

= 0. (5.6)<br />

δuα <br />

<br />

5.2 Lie-Bäcklund operators<br />

Definition 5.3<br />

A Lie-Bäcklund operator is defined <strong>by</strong> the <strong>for</strong>mal sum [6],<br />

where<br />

with<br />

X = ξ i ∂<br />

∂x i + ∂<br />

ηα<br />

∂u α + ζα i<br />

ξ i ([z]),η α ([z]) ∈ A<br />

∂<br />

∂u α i<br />

+ ζ α i 1 i 2<br />

∂<br />

∂u α i 1 i 2<br />

+ ··· , (5.7)<br />

ζ α i = D i (W α ) + ξ j u α i j (5.8)<br />

ζ α i 1 i 2<br />

= D i1 D i2 (W α ) + ξ j u α ji 1 i 2<br />

,...,<br />

W α = η α − ξ j u α j . (5.9)<br />

39


CHAPTER 5. OPERATORS<br />

Note that in the Lie-Bäcklund operator the functions ξ <strong>and</strong> η also depend on the<br />

derivatives of the differential variables, which is not the case when dealing with<br />

point trans<strong>for</strong>mations.<br />

The Lie-Bäcklund operator is the infinite order extension of X = ξ i ∂ + η α ∂<br />

∂x i<br />

The Lie-Bäcklund operator (5.7), is also written as [6]<br />

<br />

∂u α .<br />

X = ξ i D i + W α ∂<br />

∂u α + D i(W α ) ∂<br />

∂u α i<br />

+ D i1 D i2 (W α ∂<br />

)<br />

∂u α + ... (5.10)<br />

i 1 i 2<br />

The Lie Bäcklund operator truncates when acting on any differential function, i.e<br />

it belongs to A.<br />

Indeed, let<br />

X ν = ξν<br />

i ∂<br />

∂x i + ∂<br />

ηα ν<br />

∂u α ν = 1,2 (5.11)<br />

be two Lie-Bäcklund operators (where the extension to infinite order is understood).<br />

Their commutator is then given <strong>by</strong><br />

[X 1 , X 2 ] = X 1 X 2 − X 2 X 1 = (X 1 (ξ i 2)− X 2 (ξ i 1)) ∂<br />

∂x i +(X 1(η α 2)− X 2 (η α 1)) ∂<br />

∂u α (5.12)<br />

Wich also is a Lie-Bäcklund operator. This means that all Lie-Bäcklund operators<br />

<strong>for</strong>m an infinte–dimensional Lie algebra, which will be denoted <strong>by</strong> L B [6].<br />

5.3 Operators N i associated with Lie-Bäcklund operators <strong>and</strong><br />

the fundamental identity<br />

Here we present the two last operators needed <strong>for</strong> our <strong>analysis</strong> to find <strong>conservation</strong><br />

<strong>laws</strong>.<br />

40


5.3. OPERATORS N I ASSOCIATED WITH LIE-BÄCKLUND OPERATORS AND<br />

THE FUNDAMENTAL IDENTITY<br />

Definition 5.4<br />

Given a Lie-Bäcklund operator X, (5.10), we define n operators N i ,i = (1...n) <strong>by</strong><br />

the <strong>for</strong>mal sum [6]:<br />

N i = ξ i + W α<br />

δ<br />

δu α i<br />

+<br />

∞<br />

∑<br />

s=1<br />

D i1 ··· D is (W α δ<br />

)<br />

δu α . (5.13)<br />

ii 1 ...i s<br />

The fundamental identity is given in the following theorem, which is verified <strong>by</strong><br />

direct calculation.<br />

Theorem 5.3<br />

The Euler-Lagrange (5.1), Lie- Bäcklund (5.10) <strong>and</strong> the associated operators (5.13)<br />

are connected <strong>by</strong> the identity [6],<br />

X + D i (ξ i ) = W α δ<br />

δu α + D iN i . (5.14)<br />

<br />

<br />

41


Part II<br />

Symmetries <strong>and</strong> Conservation<br />

Laws<br />

43


6<br />

HAMILTON’S VARIATIONAL PRINCIPLE<br />

6.1 Hamilton´s variational principle in the plane<br />

We start with the basic problem of variational calculus, <strong>and</strong> treat it purly mathematically<br />

in two dimensions only. The problem is to determine the functions y(x)<br />

such that the integral<br />

∫ x2<br />

S = f (y(x),y ′ (x), x)dx (6.1)<br />

x 1<br />

is an extremum [1].<br />

The quantity f depends on the dependent variable y(x) <strong>and</strong> is considered as given<br />

<strong>for</strong> the particular problem. We will consider the limits of integration as fixed (note<br />

that is not necessary, if we allow the limits to vary we need to find, as said above,<br />

y(x) <strong>and</strong> x 1 , x 2 such that S is an extremum).<br />

To find an extremum when y(x) is varied we may define a neigbouring function<br />

as follow: We represent all neighbouring functions y in parametric <strong>for</strong>m as y(α, x)<br />

such that y(0, x) = y(x), we can then write<br />

y(α, x) = y(0, x) + αη(x) (6.2)<br />

where η(x) is an arbitrary function with a continuous first derivative <strong>and</strong> where<br />

η(x 1 ),η(x 2 ) vanish. Equation (6.1) will now take the <strong>for</strong>m<br />

∫ x2<br />

S (α) = f (y(α, x),y ′ (α, x), x)dx (6.3)<br />

x 1<br />

45


CHAPTER 6. HAMILTON’S VARIATIONAL PRINCIPLE<br />

where y ′ (α, x) = ∂y ′ /∂x. A necessary condition <strong>for</strong> S (α) to have an extremum is<br />

∂S<br />

∂α | α=0 = 0, (6.4)<br />

i.e that S is independent of α in first order along the path.<br />

Per<strong>for</strong>ming the imp<strong>lie</strong>d differentiation <strong>and</strong> assuming fixed boundaries yields<br />

∂S<br />

∂α = ∂ ∫ x2<br />

∫ x2<br />

( ∂ f<br />

f (y(α, x),y ′ ∂y<br />

(α, x), x)dx =<br />

∂α x 1 ∂y ∂α + ∂ f ∂y ′ )<br />

∂y ′ dx. (6.5)<br />

∂α<br />

Using (6.2) we see that<br />

∂y<br />

∂α = η(x)<br />

∂y ′<br />

∂α = ∂<br />

∂α (y′ (x) + α dη(x)<br />

dx ) = dη(x)<br />

dx . (6.6)<br />

Thus, (6.5) becomes<br />

∂S<br />

∂α = ∫ x2<br />

x 1<br />

( ∂ f<br />

∂y η(x) + ∂ f dη(x)<br />

∂y ′ dx<br />

x 1<br />

)<br />

dx. (6.7)<br />

Using integration <strong>by</strong> parts of the second term [8] in (6.7), i.e ∫ udv = uv − ∫ vdu<br />

(with ∂ f<br />

∂y<br />

= u <strong>and</strong> dη(x)<br />

′ dx<br />

= dv), yields<br />

∫ x2<br />

x 1<br />

∂ f dη(x)<br />

∂y ′ dx dx = ∂ f<br />

∂y ′ η(x)|x 2<br />

x 1<br />

−<br />

∫ x2<br />

x 1<br />

η(x) d ∂ f<br />

dx. (6.8)<br />

dx ∂y ′<br />

Using the fact that η(x) vanishes on the boundaries, we can write (6.7) as<br />

∫<br />

∂S x2<br />

( ∂ f<br />

∂α = x 1 ∂y η(x) − η(x) d ) ∫<br />

∂ f<br />

x2<br />

( ∂ f<br />

dx ∂y ′ dx =<br />

x 1 ∂y − d )<br />

∂ f<br />

dx ∂y ′ η(x)dx. (6.9)<br />

The condition that ∂S<br />

∂α | α=0 = 0 <strong>and</strong> the fact that η(x) is an arbitrary function, we<br />

must have that<br />

( ∂ f<br />

∂y − d )<br />

∂ f<br />

dx ∂y ′ | α=0 = 0 (6.10)<br />

which is a necessary condition <strong>for</strong> S to have an extremum. Equation (6.10) is the<br />

Euler-Lagrange equation in the plane.<br />

46


6.1. HAMILTON´S VARIATIONAL PRINCIPLE IN THE PLANE<br />

Example 6.1<br />

We begin with a quite mathematical example. Suppose that we want to find the<br />

shortest distance between two points in a plane, x 1 <strong>and</strong> x 2 . An arc lenght ds in a<br />

plane is [8]<br />

ds = √ √<br />

( ) dy 2<br />

dx 2 + dy 2 = 1 + dx. (6.11)<br />

dx<br />

The total lenght of a curve between x 1 <strong>and</strong> x 2 is then given <strong>by</strong><br />

√<br />

∫ x 2 ∫ x 2 ( ) dy 2<br />

I = ds = 1 + dx. (6.12)<br />

x 1 dx<br />

x 1<br />

We have an extremum problem where we want to find the curve which gives the<br />

minimum path between x 1 <strong>and</strong> x 2 , that is, the minimum of I. The function in (6.1)<br />

is here given <strong>by</strong><br />

√<br />

( ) dy 2<br />

f = 1 + . (6.13)<br />

dx<br />

Using equation (6.10) we have<br />

∂ f<br />

∂y = 0,<br />

∂ f<br />

∂y ′ = y ′<br />

√ . (6.14)<br />

1 + y ′2<br />

Thus, we arrive at the equation<br />

( )<br />

d y ′<br />

√ = 0, (6.15)<br />

dx 1 + y ′2<br />

or<br />

y ′<br />

√<br />

1 + y ′2 = c<br />

⇒ y ′ =<br />

c<br />

√ = a, (6.16)<br />

1 − c 2<br />

where a is some other constant related to the constant c according to equation<br />

(6.16). It is easy to see, <strong>by</strong> direct integration, that this is simply the equation of a<br />

straight line<br />

y = ax + b. (6.17)<br />

Thus, the shortest distance between two points in a plane is a straight line!<br />

47


CHAPTER 6. HAMILTON’S VARIATIONAL PRINCIPLE<br />

Example 6.2: The brachistochrone problem<br />

End of Example 6.1<br />

Consider a particle falling from rest from a higher point to a lower point along a<br />

curve, under the influence of gravity. The problem is to find the curve <strong>for</strong> which<br />

the fall takes the shortest time [1].<br />

Let the speed of the particle along the curve be v <strong>and</strong> an arc lenght of the curve<br />

is ds as in example 6.1. Since the only external <strong>for</strong>ce acting on the particle i the<br />

gravity, we get from <strong>conservation</strong> of energy that<br />

mgy = 1 2 mv2 . (6.18)<br />

I.e, gravitational potential energy trans<strong>for</strong>ms to kinetic energy. Solving this equation<br />

<strong>for</strong> v, we get<br />

v = √ 2gy. (6.19)<br />

Now, the time it takes <strong>for</strong> a particle to travel from a point P 1 to P 2 is give <strong>by</strong> the<br />

integral<br />

√<br />

∫ P2<br />

∫<br />

ds P2<br />

t 12 =<br />

P 1 v = 1 + y ′2<br />

dx. (6.20)<br />

P 1 2gy<br />

Thus, equation (6.20) is a extremum problem <strong>and</strong> the function to be varied is<br />

f (x,y,y ′ ) =<br />

√<br />

1 + y ′2<br />

2gy . (6.21)<br />

Now we would normaly directly use the Euler-Lagrange equation, (6.10) which is<br />

the condition <strong>for</strong> (6.20) to have an extremum. However, f (x,y,y ′ ) does not have<br />

any explicit dependence on x, there<strong>for</strong>e ∂ f /∂x = 0, which we can take advantage<br />

of to simplify the calculations in the following way:<br />

Take the total derivative of f with respect to x yields,<br />

D x f ≡ d f<br />

dx = ∂ f<br />

∂x + y′ ∂ f<br />

∂y + y′′ ∂ f<br />

∂y ′ . (6.22)<br />

48


6.1. HAMILTON´S VARIATIONAL PRINCIPLE IN THE PLANE<br />

Solving this <strong>for</strong> the ∂ f /∂y term gives<br />

y ′ ∂ f<br />

∂y = d f<br />

dx − ∂ f<br />

∂x − y′′ ∂ f<br />

∂y ′ . (6.23)<br />

Now, multiplying (6.10) with y ′ gives<br />

y ′ ∂ f<br />

∂y − d ∂ f<br />

y′ = 0. (6.24)<br />

dx ∂y ′<br />

Substitute (6.23) into (6.24) yields,<br />

d f<br />

dx − ∂ f<br />

∂x − y′′ ∂ f<br />

− ∂ f<br />

∂x + d dx<br />

In the case ∂ f /∂x = 0 we get<br />

∂y ′ − d ∂ f<br />

y′<br />

dx ∂y<br />

(<br />

′<br />

f − y ′ ∂ f )<br />

∂y ′ = 0 (6.25)<br />

f − y ′ ∂ f = C, (6.26)<br />

∂y ′<br />

where C is a constant of integration. Equation (6.26) was dicovered 1868 <strong>by</strong><br />

Beltrami <strong>and</strong> is known as the Beltrami identity in the plane, [12] Using this in our<br />

case we get<br />

1<br />

√ √ = C. (6.27)<br />

2gy 1 + y ′2<br />

Squaring both sides <strong>and</strong> rearranging, we get<br />

(1 + y ′2 )y = 1<br />

2gC 2 = k2 , (6.28)<br />

where kr is a new constant. The solution to this equation is given <strong>by</strong> the parametric<br />

equations<br />

x = 1 2 k2 (t − sint)<br />

y = 1 2 k2 (1 − cost). (6.29)<br />

This is the equations of a cycloid. Thus, the particle will fall from point P 1 to P 2<br />

in the shortest time if it follows the path of a cycloid.<br />

End of Example 6.2<br />

49


CHAPTER 6. HAMILTON’S VARIATIONAL PRINCIPLE<br />

6.1.1 Euler-Lagrange equation with several variables<br />

Now we consider the case of several independent variables x = (x 1 ... x n ) <strong>and</strong> differential<br />

variables u = (u 1 ...u m ). Consider a Lagrangian L ∈ A that is a differential<br />

function of first order an an arbitrary volume V ⊂ R n , spanned <strong>by</strong> x = (x 1 ... x n )<br />

with boundary ∂V. An action is the integral [6]<br />

∫<br />

l[u] = L(x,u,u (1) )dx. (6.30)<br />

V<br />

Consider a small variation h(x) in u which leads to a Lagrangian L = L(x,u +<br />

h,u (1) + h (1) ). Taylor exp<strong>and</strong>ing this <strong>for</strong> small variations up to the linear part in h<br />

yields<br />

L(x,u+h,u (1)+h(1) ) = L(x,u+0,u (1) +0)+ ∂L<br />

∂u α (hα −0)+ ∂L<br />

∂u α (h α i −0)+.... (6.31)<br />

i<br />

The variation δl[u] caused <strong>by</strong> the variation u+h(x), defined as the principal linear<br />

part in h of the integral<br />

∫<br />

V<br />

takes the <strong>for</strong>m<br />

[L(x,u + h,u (1) + h (1) ) − L(x,u,u (1) )]dx (6.32)<br />

∫<br />

δl[u] =<br />

V<br />

[ ∂L<br />

∂u α (hα ) + ∂L ]<br />

∂u α (h α i ) dx. (6.33)<br />

i<br />

Integration of second term <strong>by</strong> parts yields<br />

∫<br />

∫ [ ] ∫ [ ]<br />

∂L<br />

∂L<br />

∂L<br />

V ∂u α (h α i )dx = D i<br />

i<br />

V ∂u α (h α ) dx − D i<br />

i<br />

V ∂u α (h α )dx. (6.34)<br />

i<br />

Leading to<br />

∫ ( ∂L<br />

δl[u] =<br />

V ∂u α − D ∂L<br />

i<br />

∂u α i<br />

) ∫ [ ] ∂L<br />

h α dx + D i<br />

V ∂u α (h α ) dx. (6.35)<br />

i<br />

Remark: The first term on RHS of equation (6.34) might seem strange, but remeber<br />

that in basic calculus of one variable we have ∫ d f (x)<br />

dx<br />

dx = f (x), same rule<br />

apply here but in this case f = f (x,u,u (1) ) = ∂L<br />

∂u α (h α ) <strong>and</strong> d i<br />

dx is replaced <strong>by</strong> D i.<br />

50


6.1. HAMILTON´S VARIATIONAL PRINCIPLE IN THE PLANE<br />

Recall the divergence theorem [8]<br />

∫<br />

∫<br />

D i f (x,u,u (1),... )dx = f (x,u,u (1),... )ν i dx i = 1...n (6.36)<br />

V<br />

∂V<br />

where f is an arbitrary function within a volume V with boundary ∂V <strong>and</strong> unit<br />

outer normal ν = ν 1 ...ν n .<br />

Using the divergence theorem, (6.35) can be written as<br />

∫<br />

δl[u] =<br />

V<br />

( ∂L<br />

∂u α − D ∂L<br />

i<br />

∂u α i<br />

) ∫<br />

h α dx +<br />

∂V<br />

∂L<br />

(h α )ν i dx. (6.37)<br />

Provided that the functions h α (x) vanish on the boundary ∂V, we get<br />

∫ ( )<br />

∂L<br />

δl[u] =<br />

V ∂u α − D ∂L<br />

i h α dx. (6.38)<br />

∂u α i<br />

A function u = u(x) is an extremum of the variational integral (6.30) if δl[u] = 0.<br />

For an arbitrary volume V <strong>and</strong> increment h(x), that vanish on ∂V, we must have<br />

that<br />

∂L<br />

∂u α − D ∂L<br />

i = 0. (6.39)<br />

∂u α i<br />

∂u α i<br />

The variational derivate<br />

δL<br />

δu α ≡ ∂L<br />

∂u α − D i<br />

∂L<br />

∂u α i<br />

(6.40)<br />

is called the Euler-Lagrange equations.<br />

51


7<br />

NOETHER’S THEOREM AND<br />

CONSERVATION LAWS<br />

In this chapter we will <strong>for</strong>mulate Noether’s theorem using the mathematical tools<br />

we have developed so far, see also [13]. In the next chapter we will also explain<br />

how one can find a Lagrangian <strong>for</strong> any system of differential equations <strong>by</strong> using<br />

the theory of adjoint equations, leading to a more general <strong>conservation</strong> theorem.<br />

7.1 Noether´s Theorem<br />

Definition 7.1<br />

Let G be a one-parameter <strong>group</strong> of trans<strong>for</strong>mations [6]<br />

¯x i = f i (x,u,a) ū α = ϕ(x,u,a) (7.1)<br />

where i = 1...n <strong>and</strong> α = 1...m, with generator<br />

X = ξ i (x,u) ∂<br />

∂x i + ηα (x,u) ∂<br />

∂u α . (7.2)<br />

A variational integral (6.30), l[u] = ∫ V L(x,u,u (1))dx is said to be invariant under<br />

G if<br />

∫<br />

∫<br />

L( ¯x,ū,ū (1) )d ¯x = L(x,u,u (1) )dx (7.3)<br />

¯V<br />

V<br />

where ¯V ⊂ R n is the volume <strong>obtained</strong> from V <strong>by</strong> trans<strong>for</strong>mation (7.1) [6].<br />

53


CHAPTER 7. NOETHER’S THEOREM AND CONSERVATION LAWS<br />

<br />

Lemma 7.1<br />

The variational integral (6.30) is invariant under the <strong>group</strong> G if <strong>and</strong> only if ([6])<br />

X(L) + LD i (ξ i ) = 0, (7.4)<br />

where X is the extension of (7.2), i.e<br />

X(L) = ξ i ∂L ∂L<br />

+ ηα<br />

∂xi ∂u α + ζα i<br />

where<br />

ζ α i = D i (η α ) − u α j D i (ξ j )<br />

∂L<br />

∂u α i<br />

(7.5)<br />

<br />

Proof of Lemma 7.1:<br />

Recall from calculus of several variables that <strong>for</strong> a change of variables, which<br />

have a one-to-one trans<strong>for</strong>mation from a volume ¯V to a volume V, the volume<br />

elements given <strong>by</strong><br />

d ¯V = JdV (7.6)<br />

where J is the Jacobian <strong>for</strong> the trans<strong>for</strong>mation. Using this we can rewrite LHS of<br />

(7.3) in the <strong>for</strong>m<br />

∫<br />

∫<br />

L( ¯x,ū,ū (1) )d ¯x = L( ¯x,ū,ū (1) )Jdx, (7.7)<br />

¯V<br />

V<br />

where J = det‖D j ( ¯x i )‖ is the Jacobian of the first trans<strong>for</strong>mation in (7.1). Since<br />

both integrals now is taken over the same arbitrary volume V, the integral equation<br />

(7.3) is eqvivalent to<br />

L( ¯x,ū,ū (1) )J = L(x,u,u (1) ). (7.8)<br />

Consider the infinitesimal trans<strong>for</strong>mations<br />

¯x i = x i + aξ i ū α = u α + aη α ū α i = u α i + aζ α i . (7.9)<br />

54


7.1. NOETHER´S THEOREM<br />

Taylor expansion of L( ¯x,ū,ū (1) ) around a = 0 yields,<br />

L( ¯x,ū,ū (1) ) ≈ L( ¯x,ū,ū (1) )| a=0 + ∂L<br />

∂ ¯x i ∂ ¯x i<br />

∂a | a=0 · (a − 0)<br />

+ ∂L<br />

∂ū α ∂ū α<br />

∂a | a=0 · (a − 0) + ∂L<br />

∂ū α i<br />

∂ū α i<br />

∂a | a=0 · (a − 0)<br />

= L(x,u,u (1) ) + ∂L<br />

∂x i ξi a + ∂L<br />

∂u α ηα a + ∂L<br />

∂u α ζi α a<br />

i<br />

= L(x,u,u (1) ) + a[X(L)]. (7.10)<br />

Taylor expansion of J around a = 0 yields<br />

J ≈ J| a=0 + ∂J<br />

∂a | a=0 · a<br />

= det‖D j (x i )‖ + det‖D j ( ∂ ¯xi<br />

∂a )‖ a=0<br />

= 1 + aD i (ξ i ). (7.11)<br />

Thus, the LHS of equation (7.8) yields<br />

L( ¯x,ū,ū (1) )J ≈ ( L(x,u,u (1) ) + a[X(L)] )( 1 + aD i (ξ i ) )<br />

≈ L(x,u,u (1) ) + a[X(L)] + aL(x,u,u (1) )D i (ξ i )<br />

= L(x,u,u (1) ) + a [ [X(L)] + LD i (ξ i ) ] . (7.12)<br />

This should be equal to RHS of (7.8). It follows that<br />

[X(L)] + LD i (ξ i ) = 0.<br />

□<br />

Definition 7.2<br />

A vector T i = (T 1 ...T n ) is a conserved vector if ([6])<br />

D i (T i ) = 0 (7.13)<br />

<br />

55


CHAPTER 7. NOETHER’S THEOREM AND CONSERVATION LAWS<br />

Theorem 7.1<br />

Let the variational integral (6.30) be invariant under the <strong>group</strong> G, with generator<br />

(7.2). Then the vector T i ∈ A defined <strong>by</strong><br />

T i = Lξ i + (η α − ξ j u α j ) ∂L<br />

∂u α i<br />

i = (i,...n) (7.14)<br />

is a conserved vector of the Euler-Lagrange equation, i.e D i (T i ) = 0 on the solutions<br />

of Euler-Lagrange equations [6].<br />

Proof of Theorem 7.1:<br />

If the variational integral (6.30) is invariant it fulfills [X(L)] − LD i (ξ i ) = 0 according<br />

to Lemma 7.1. By rewriting the operator (7.2) according to (5.10), i.e<br />

X = ξ i D i + W α ∂<br />

∂u α + D i(W α ) ∂<br />

Straight<strong>for</strong>ward calculations yield<br />

∂u α i<br />

. (7.15)<br />

X(L) + LD i (ξ i ) = ξ i D i (L) + W α ∂L<br />

∂u α + D i(W α ) ∂L<br />

∂u α + LD i (ξ i )<br />

i<br />

= ξ i D i (L) + LD i (ξ i ) + W α ∂L<br />

∂u α + D i(W α ∂L<br />

∂u α )<br />

i<br />

− W α D i ( ∂L<br />

∂u α )<br />

i<br />

= D i (Lξ i ) + D i (W α ∂L<br />

∂u α ) + W α ( ∂L<br />

i ∂u α − D ∂L<br />

i<br />

∂u α )<br />

i<br />

= D i (Lξ i + W α ∂L<br />

∂u α ) + W α δL<br />

i δu α . (7.16)<br />

We know that <strong>for</strong> the solutions to the Euler-Lagrange equations we have that δL<br />

δu α =<br />

0 then, since<br />

Hence<br />

X(L) + LD i (ξ i ) ≡ D i (Lξ i + W α ∂L<br />

∂u α ) + W α δL<br />

i δu α = 0.<br />

D i (Lξ i + W α ∂L<br />

∂u α ) = 0. (7.17)<br />

i<br />

<br />

56


7.1. NOETHER´S THEOREM<br />

According to (7.13) in definition 7.2 it follows that T i = Lξ i + (η α − ξ j u α j ) ∂L<br />

∂u α i<br />

conserved vector.<br />

is a<br />

According to Theorem 5.1 one can add any divergence type function F to the<br />

Lagragian. The invariance condition (7.4) is then replaced <strong>by</strong> the divergence relation,<br />

[6],<br />

X(L) + LD i (ξ i ) = D i (B i ). (7.18)<br />

Example 7.1<br />

The differential equation, taken from [4]<br />

y ′′ + 1 x (y′ + y ′3 ) = 0 (7.19)<br />

has the Lagrangian<br />

L = x √ 1 + y ′2 (7.20)<br />

<strong>and</strong> admits the two-parameter <strong>group</strong> with generators<br />

X 1 = ∂ ∂y , X 2 = x ∂ ∂x + y ∂ ∂y . (7.21)<br />

In order to apply Noether’s theorem to find the <strong>conservation</strong> <strong>laws</strong>, we need to<br />

check that the invariance condition (7.4) in lemma 7.1 is fullfilled. I.e, that<br />

XL − LD i (ξ i ) = 0, (7.22)<br />

where the extension of X is understood. For X 1 we have the twice extended generator<br />

∂<br />

Y 1 = X 1 + ζ 1<br />

∂y ′ + ζ ∂<br />

2<br />

∂y ′′ (7.23)<br />

The prolongation <strong>for</strong>mulas is given <strong>by</strong> equation (4.24) in chapter 4. In this case<br />

the prolongations are<br />

ζ 1 = D x (η) − y ′ D x (ξ) = D x (1) − y ′ D x (0) = 0<br />

ζ 2 = D x (ζ 1 ) − y ′′ D x (ξ) = D x (0) − y ′ D x (0) = 0. (7.24)<br />

□<br />

57


CHAPTER 7. NOETHER’S THEOREM AND CONSERVATION LAWS<br />

The invariance test yields,<br />

∂<br />

∂y (L) − LD x(0) = 0 (7.25)<br />

For X 2 we have the twice extended generator<br />

∂<br />

Y 2 = X 2 + ζ 1<br />

∂y ′ + ζ ∂<br />

2 . (7.26)<br />

∂y ′′<br />

The prolongations are in this case<br />

ζ 1 = D x (y) − y ′ D x (x) = y ′ − y ′ = 0<br />

ζ 2 = D x (ζ 1 ) − y ′′ D x (x) = o − y ′′ = −y ′′ . (7.27)<br />

Thus, the extended generator is<br />

Y 2 = x ∂ ∂x + y ∂ ∂y − ∂<br />

y′′ . (7.28)<br />

∂y ′′<br />

The invariance test yields<br />

(<br />

x ∂ ∂x + y ∂ )<br />

∂y − ∂ ( y′′<br />

∂y ′′ x √ )<br />

1 + y ′2 −<br />

(x √ )<br />

1 + y ′2 = L − L = 0. (7.29)<br />

Thus, Noether’s theorem is applicabible to the two-parameter <strong>group</strong>.<br />

Example 7.2<br />

The nonlinear equation of motion <strong>for</strong> a pendulum is given <strong>by</strong> [14]<br />

End of Example 7.1<br />

y ′′ + ω 2 siny = 0, (7.30)<br />

where ω 2 is a constant <strong>and</strong> y = y(t). The Lagrangian to equation (7.30) is given <strong>by</strong><br />

L = y′2<br />

2 + ω2 cosy. (7.31)<br />

Equation (7.30) does not contain any explicit dependence on t, consecuently, it is<br />

invariant under time translation, <strong>and</strong> there<strong>for</strong>e has the generator<br />

X = ∂ ∂t<br />

(7.32)<br />

58


7.1. NOETHER´S THEOREM<br />

The conserved quantity associated with this symmetry is given <strong>by</strong> equation (7.15)<br />

in theorem 7.1, thus<br />

T i = Lξ i + (η α − ξ j u α j ) ∂L<br />

∂u α , (7.33)<br />

i<br />

where, in this case, ξ i = ξ = 1, η α = 0,∀α, u α = y <strong>and</strong> u α i = y ′ hence,<br />

T = L − y ′ ∂L<br />

∂y ′ . (7.34)<br />

Substituting in the expression <strong>for</strong> the Lagrangian <strong>and</strong> the fact that ∂L/∂y ′ = y ′ , we<br />

get the conserved quantity<br />

T = ω 2 cosy − y′2<br />

2 . (7.35)<br />

Theorem 7.2<br />

End of Example 7.2<br />

Let L(x,u,u (1) ,...,u (s) ) ∈ A be a differential function of any order. The Euler-<br />

Lagrange equations are, using equation (5.1),<br />

δL<br />

= 0 α = 1,...,m. (7.36)<br />

δuα Let X be a symmetry such that<br />

X(L) + LD i (ξ i ) = D i (B i ), B i ∈ A. (7.37)<br />

Then equations (7.36) admit a <strong>conservation</strong> law, D i (C i ) = 0, defined <strong>by</strong><br />

C i = N i (L) − B i , i = 1,...,n, (7.38)<br />

where the operators N i are defined <strong>by</strong> (5.13) [6].<br />

Proof of Theorem 7.2:<br />

This follows directly from the fundamental identity (5.14). Indeed,<br />

D i (C i ) = D i N i (L) − D i B i = 0<br />

⇒ D i N i (L) = D i B i = X(L) + LD i (ξ i )<br />

where the fact that δL<br />

δu α = 0 was used.<br />

<br />

59


CHAPTER 7. NOETHER’S THEOREM AND CONSERVATION LAWS<br />

□<br />

7.2 Many <strong>symmetries</strong><br />

If a differential equation has r <strong>symmetries</strong> X 1 ,..., X r of the <strong>for</strong>m (7.2),i.e<br />

X µ = ξ i µ(x,u) ∂<br />

∂x i + ηα µ(x,u) ∂<br />

∂u α µ = 1,...,r (7.39)<br />

then (7.14) provides r conserved vectors T 1 ,...,T r with components<br />

T i µ = Lξ i µ + (η α µ − ξ i µu α j ) ∂L<br />

∂u α i<br />

µ = 1,...r. (7.40)<br />

See [6].<br />

Example 7.3<br />

This example is taken from [14].<br />

The motion of a free particle in special relativity is described <strong>by</strong> the relativistic<br />

Lagragian<br />

L = −mc 2√ 1 − β 2 , (7.41)<br />

where β 2 = |v|2 , |v| 2 =<br />

c<br />

∑ 3 2 i=1 (vi ) 2 . The space coordinates of the particle is x =<br />

(x 1 , x 2 , x 3 ) <strong>and</strong> the velocity is v = (v 1 ,v 2 ,v 3 ) such that v = dx<br />

dt<br />

, where t is time.<br />

We have that x is our differential variables <strong>and</strong> t is our only independent variable.<br />

The generators of the Lorentz <strong>group</strong> (see e.g Classical dynamics) is given <strong>by</strong><br />

X 0 = ∂ ∂t<br />

X i = ∂<br />

∂x i<br />

X i j = x j ∂<br />

∂x i − ∂<br />

xi<br />

∂x j<br />

X 0i = t ∂<br />

∂x i + 1 c 2 xi ∂ ∂t . (7.42)<br />

60


7.2. MANY SYMMETRIES<br />

By applying Noether´s theorem to each of these generators with the given Lagragian<br />

we wish to find the corresponding <strong>conservation</strong> law. We have generators of<br />

the <strong>for</strong>m<br />

X = ξ(t,x) ∂ ∂t + ηα (t,x) ∂<br />

∂x α α = 1,2,3. (7.43)<br />

We will get a conserved quantity of the <strong>for</strong>m<br />

where<br />

T = ξL + (η α − ξv α ) ∂L<br />

∂v α , (7.44)<br />

∂L<br />

∂v α = 1<br />

mvα γ γ = √ . (7.45)<br />

1 − β 2<br />

For X 0 = ∂ ∂t we have ξ = 1,η1 = η 2 = η 3 = 0. Equation (7.44) yields<br />

T = L − v α ∂L<br />

∂v α<br />

= −mc 2 γ −1 − m|v| 2 γ<br />

= −mc 2 γ. (7.46)<br />

Putting T = −E we get E = mc 2 γ, i.e <strong>conservation</strong> of energy.<br />

For X i = ∂<br />

∂x i we get a conserved vector T = T i = (T 1 ,T 2 ,T 3 ). Here we have ξ = 0<br />

<strong>for</strong> all i,while η α = δ iα . Equation (7.44) yields<br />

T 1 = ∂L<br />

∂v 1 = mv1 γ<br />

T 2 = ∂L<br />

∂v 2 = mv2 γ<br />

T 3 = ∂L<br />

∂v 3 = mv3 γ (7.47)<br />

Leading to the conserved vector T = mvγ, i.e <strong>conservation</strong> of relativistic momentum.<br />

61


CHAPTER 7. NOETHER’S THEOREM AND CONSERVATION LAWS<br />

For the generator X i j = x j ∂ − x i ∂ , we have a rotational symmetry. With ξ = 0<br />

∂x i ∂x j<br />

<strong>and</strong> η α = x j δ αi − x i δ α j . Equation (7.44) yields<br />

T 12 = x 2 ∂L ∂L<br />

− x1<br />

∂v1 ∂v 2<br />

= mγ(x 2 v 1 − x 1 v 2 )<br />

T 23 = x 3 ∂L ∂L<br />

− x2<br />

∂v2 ∂v 3<br />

= mγ(x 3 v 2 − x 2 v 3 )<br />

T 31 = x 1 ∂L ∂L<br />

− x3<br />

∂v3 ∂v 1<br />

= mγ(x 1 v 3 − x 3 v 1 ). (7.48)<br />

We get the conserved vector T = (T 23 ,T 31 ,T 12 ). Putting T = −L we get the conserved<br />

vector L = m(x×vγ) = x×p rel , i.e <strong>conservation</strong> of angular momentum.<br />

For the generator X 0i = t ∂<br />

∂x i + 1 c 2 x i ∂ ∂t , we get a vector T = (T 1 ,T 2 ,T 3 ). For the<br />

first component we have ξ = 1 c 2 x 1 ,η 1 = t,η 2 = η 3 = 0. Equation (7.44) yields<br />

T 01 = L 1 c 2 x1 + (t − 1 c 2 x1 v 1 ) ∂L<br />

∂v 1<br />

= L 1 c 2 x1 + tmv 1 γ − 1 c 2 x1 (v 1 ) 2 γ<br />

= m γ (−x1 + tv 1 γ 2 − x 1 [βγ] 2 )<br />

= mγ(tv 1 − x 1 ). (7.49)<br />

A similar procedure <strong>for</strong> the remaining two components <strong>and</strong> putting T = Q, leads<br />

to the conserved vector Q = mγ(tv−x), i.e <strong>conservation</strong> of relative center of mass.<br />

End of Example 7.3<br />

62


8<br />

ADJOINT EQUATIONS<br />

In this chapter the very basic theory of adjoint equations is discussed. We start<br />

with linear adjoint equations <strong>and</strong> then we consider adjoint equations <strong>for</strong> arbitrary<br />

differential equations, as made in [15]. The latter will be used to find a Lagrangian<br />

<strong>for</strong> any arbitrary physical system described <strong>by</strong> arbitrary differential equations.<br />

8.1 Linear adjoint equations<br />

To get a feeling <strong>for</strong> the concept of adjoint equations we start <strong>by</strong> considering linear<br />

adjoint equations be<strong>for</strong>e we move on to arbitrary differential equations.<br />

Definition 8.1<br />

Let L be a linear differential operator of any order. An adjoint operator L ∗ to L is<br />

given <strong>by</strong> [4]<br />

vL[u] − uL ∗ [v] = D i (P i ) (8.1)<br />

∀u,v where P = (p 1 ,..., p n ) is a vector field. The equation L ∗ [v] = 0 is termed the<br />

adjoint equation to L[u] = 0.<br />

<br />

63


CHAPTER 8. ADJOINT EQUATIONS<br />

Consider a general second-order operator L [15],<br />

L = a i j (x)D i D j + b i (x)D i + c(x) (8.2)<br />

where i, j = 1,...,n <strong>and</strong> the coefficients are symetric in i, j, i.e a i j = a ji .<br />

Theorem 8.1<br />

The adjoint operator L ∗ to L is uniquely determined <strong>and</strong> has the <strong>for</strong>m, see [15],<br />

L ∗ [v] = D i D j (a i j v) − D i (b i v) + cv (8.3)<br />

Proof of Theorem 8.1:<br />

Let u,v be two differential functions. Then<br />

vL[u] = va i j (x)D i D j u+vb i (x)D i u+vc(x)u<br />

} {{ } } {{ } } {{ }<br />

1<br />

2<br />

3<br />

= D i (va i j (x)D j u) − D i (va i j (x))D j u<br />

} {{ }<br />

1<br />

+ D i (vb i (x)u) − uD i (vb i (x)) +vc(x)u.<br />

} {{ } } {{ }<br />

2<br />

3<br />

The second term in 1 can be rewritten as<br />

−D i (va i j (x))D j u = −D j (uD i (va i j (x))) + uD i D j (a i j (x)v)<br />

= −D i (uD j (va i j (x))) + uD i D j (a i j (x)v).<br />

<br />

(8.4)<br />

In the last step we changed the index of i, j <strong>and</strong> used the fact that the coefficients<br />

are symmetric. Hence<br />

vL[u] = D i (va i j (x)D j u) − D i (uD j (va i j (x))) + uD i D j (a i j (x)v)<br />

+ D i (vb i (x)u) − uD i (vb i (x)) + vc(x)u<br />

= u [ D i D j (a i j (x)v) − D i (vb i (x)) + c(x)v ]<br />

+ D i<br />

[<br />

va<br />

i j (x)u j − uD j (va i j (x)) + vb i (x)u ]<br />

⇒<br />

vL[u] − u [ D i D j (a i j (x)v) − D i (vb i (x)) + c(x)v ]<br />

= D i<br />

[<br />

va<br />

i j (x)u j − uD j (va i j (x)) + vb i (x)u ] . (8.5)<br />

64


8.1. LINEAR ADJOINT EQUATIONS<br />

Comparing this with (8.1), we see that the adjoint operator L ∗ has the <strong>for</strong>m<br />

L ∗ [v] = D i D j (a i j (x)v) − D i (vb i (x)) + c(x)v<br />

<strong>and</strong> it follows that<br />

P i = va i j (x)u j − uD j (va i j (x)) + vb i (x)u. (8.6)<br />

The same result app<strong>lie</strong>s if u,v are m-dimensional vector functions, <strong>and</strong> the coefficients<br />

are m × m-matrices according to [15].<br />

Definition 8.2<br />

An operator is said to be self adjoint if, <strong>for</strong> any function u,<br />

L[u] = L ∗ [u], (8.7)<br />

see [15] or [4].<br />

Theorem 8.2<br />

The operator L = a i j (x)D i D j + b i (x)D i + c(x)is self adjoint if <strong>and</strong> only if<br />

b i (x) = D j (a i j ) (8.8)<br />

Proof of Theorem 8.2:<br />

From Definition 8.1 we have that if L is self-adjoint, i.e according to Definition<br />

8.2 L[u] = L ∗ [v] when we put v = u, we must have that D i (P i ) vanish. Thus, <strong>for</strong><br />

v = u<br />

D i (P i ) ≡ D i (ua i j (x)u j − uD j (ua i j (x)) + ub i (x)u)<br />

= D i u [ a i j D j u − D j (ua i j ) + b i u ]<br />

= D i u [ a i j D j u − a i j D j u − uD j a i j + b i u ]<br />

= D i u [ u ( b i − D j a i j)] = 0 (8.9)<br />

has to be true <strong>for</strong> any u, <strong>and</strong> equation (8.8) follows. For alternative proof see [4].<br />

□<br />

<br />

<br />

□<br />

65


CHAPTER 8. ADJOINT EQUATIONS<br />

8.2 Adjoint equations of arbitrary differential equations<br />

We will now consider adjoint equations <strong>for</strong> arbitrary differential equations, to be<br />

used in the construction of Lagrangians as made in [15] <strong>and</strong> [3].<br />

Definition 8.3<br />

Consider a system of sth-order partial differential equations,<br />

F α (x,u,u (1) ,...,u (s) ) = 0 α = 1,...,m (8.10)<br />

where F α (x,u,u (1) ,...,u (s) ) ∈ A are differential functions with n independent variables<br />

x = (x 1 ,..., x n ) <strong>and</strong> m dependent variables u = (u 1 ,...,u m ),u = u(x). The<br />

system of adjoint equations to (8.10) is defined <strong>by</strong><br />

F ∗ α(x,u,v,...,u (s) ,v (s) ) ≡ δ(vβ F β )<br />

δu α = 0 α = 1,...,m (8.11)<br />

where v = (v 1 ,...,v m ) are new dependent variables v = v(x) [15].<br />

Definition 8.4<br />

A system of equations (8.10) is said to be self adjoint if the system <strong>obtained</strong> from<br />

the adjoint equations (8.11) <strong>by</strong> the substitution v = u:<br />

F ∗ α(x,u,u,...,u (s) ,u (s) ) = 0 α = 1,...,m (8.12)<br />

is identical with the original system (8.10) [15].<br />

Exampel 8.1<br />

The heat equation is<br />

u t − c(x)u xx = 0, (8.13)<br />

where c(x) is some arbitrary function with correct dimensions. To find the adjoint<br />

equation we use Definition 8.3. Here we have that<br />

F = F(x,t,u,u (1) ,u (2) ) = c(x)u xx − u t = 0, (8.14)<br />

<br />

<br />

66


8.2. ADJOINT EQUATIONS OF ARBITRARY DIFFERENTIAL EQUATIONS<br />

where x,t are independent variables, <strong>and</strong> u is the differential variable. By definition<br />

we obtain the adjoint equation as<br />

F ∗ = F ∗ (x,t,u,v,u (1) ,u (2) ,v (1) ,v (2) ) = δ<br />

δu (v[c(x)u xx − u t ])<br />

[ ]<br />

∂<br />

≡<br />

∂u − D ∂ ∂<br />

t − D x + D 2 ∂<br />

x + D 2 ∂ ∂<br />

t + D x D t<br />

∂u t ∂u x ∂u xx ∂u tt ∂u xt<br />

· (v[c(x)u xx − u t ])<br />

[<br />

]<br />

∂<br />

= −D t + D 2 ∂<br />

x (v[c(x)u xx − u t ])<br />

∂u t ∂u xx<br />

= −D t (−v) + D 2 x(c(x)v) = v t + (cv) xx = 0, (8.15)<br />

which is the adjoint equation to u t − c(x)u xx = 0 with new dependent variable v.<br />

End of Example 8.1<br />

Exampel 8.2<br />

The Kortweg-de Vries equation is a mathematical model of waves on shallow<br />

water surfaces. It has the <strong>for</strong>m, see [6] <strong>and</strong> [15],<br />

u t = u xxx + uu x . (8.16)<br />

Here we have that<br />

F = F(t, x,u,u (1) ,u (2) ,u (3) ) = u t − u xxx − uu x = 0. (8.17)<br />

The adjoint equation is<br />

F ∗ (t, x,u,v,u (1) ,u (2) ,u (3) ,v (1) ,v (2) ,v (3) ) = δ<br />

δu (v[u t − u xxx − uu x ])<br />

[ ]<br />

∂<br />

=<br />

∂u − D ∂ ∂<br />

t − D x − D 3 ∂<br />

x (v[u t − u xxx − uu x ])<br />

∂u t ∂u x ∂u xxx<br />

= −vu x − D t v + D x (vu) + D 3 xv<br />

= −vu x − v t + v x u + vu x + v xxx<br />

= −v t + v x u + v xxx . (8.18)<br />

67


CHAPTER 8. ADJOINT EQUATIONS<br />

The adjoint equation to u t = u xxx +uu x is v t = v x u+v xxx . Furthermore, if we replace<br />

v <strong>by</strong> u in the adjoint equation, we have that<br />

F ∗ (t, x,u,u,u (1) ,u (2) ,u (3) ,u (1) ,u (2) ,u (3) )<br />

= −(u t − u xxx − uu x ) = −F(t, x,u,u (1) ,u (2) ,u (3) ). (8.19)<br />

Thus, according to Definition 8.4, Korteweg-de Vries equation is self adjoint.<br />

End of Example 8.2<br />

8.3 Constructing Lagrangians <strong>by</strong> the use of adjoint equation<br />

The following theorem provides us with a powerful method to construct a Lagrangian<br />

<strong>for</strong> a system of arbitrary differential equations, if they are considered<br />

with its adjoint system. The only restriction is that the number of equations has to<br />

be equal to the number of dependent variables. This is the main result in [15].<br />

Theorem 8.4<br />

Any system of sth-order differential equations (8.10),<br />

F α (x,u,u (1) ,...,u (s) ) = 0<br />

α = 1,...,m<br />

considered together with its adjoint equation (8.11),<br />

F ∗ α(x,u,v,...,u (s) ,v (s) ) ≡ δ(vβ F β )<br />

δu α = 0 α = 1,...,m<br />

has a Lagrangian. The simultaneous system (8.10) <strong>and</strong> (8.11) with 2m dependent<br />

variables, u = (u 1 ,...,u m ) <strong>and</strong> v = (v 1 ,...,v m ), is the system of Euler-Lagrange<br />

equations with a Lagrangian L defined <strong>by</strong><br />

L = v β F β . (8.20)<br />

This is easy to verify, since we have<br />

δL<br />

δv α = F α(x,u,u (1) ,...,u (s) ) = 0, (8.21)<br />

<br />

68


8.3. CONSTRUCTING LAGRANGIANS BY THE USE OF ADJOINT EQUATION<br />

<strong>and</strong><br />

δL<br />

δu α = F∗ α(x,u,v,...,u (s) ,v (s) ) = 0, (8.22)<br />

i.e The Euler-Lagrange equation is satisfied.<br />

Example 8.3<br />

Consider example 8.1. The heat equation considered togheter with its adjoint<br />

equation has a Lagrangian, namely<br />

L = vu t − c(x)u x . (8.23)<br />

Indeed, the Euler-Lagrange are<br />

δL<br />

δv = u t − c(x)u x = 0<br />

δL<br />

δu = δv[u t − c(x)u x ]<br />

≡ F ∗ (t, x,u,v,u (1) ,u (2) ,v (1) ,v (2)<br />

δu<br />

= v t + (cv) xx = 0, (8.24)<br />

i.e the Lagrangian (8.23) satifies the Euler-Lagrange equation.<br />

End of Example 8.3<br />

69


9<br />

SYMMETRIES OF DIFFERENTIAL<br />

EQUATIONS AND THE MAIN<br />

CONSERVATION THEOREM<br />

In this chapter we clarify the important step of finding <strong>symmetries</strong> from a physical<br />

system described <strong>by</strong> differential equation. Some <strong>symmetries</strong> can easily be<br />

found <strong>by</strong> using the invariance condition, that is <strong>for</strong> exampel the symmetry <strong>group</strong>s<br />

that consists of translatation <strong>and</strong> scaling trans<strong>for</strong>mations. However, the main concern<br />

is to find a method that determines all <strong>symmetries</strong> of differential equations.<br />

This is done via the determining equation. First we consider some examples of<br />

simple <strong>symmetries</strong> <strong>obtained</strong> <strong>by</strong> using the fact that the <strong>group</strong> in interest leaves our<br />

differential equation invariant.<br />

Example 9.1<br />

Consider The Riccati equation [6]<br />

y ′ + y 2 − 2 x 2 = 0 (9.1)<br />

i.e, a function F(x,y,y ′ ) = 0, y ′ = dy<br />

dx<br />

. We set our symetry <strong>group</strong> G to be a dilatiation<br />

<strong>group</strong> ¯x = kx <strong>and</strong> ȳ = ly. Then equation (9.1) takes the <strong>for</strong>m<br />

ȳ ′ + ȳ 2 −<br />

2¯x 2<br />

= 0 (9.2)<br />

71


CHAPTER 9. SYMMETRIES AND THE MAIN CONSERVATION THEOREM<br />

i.e, F( ¯x,ȳ,ȳ ′ ) = 0. The invariance condition states that<br />

ȳ ′ + ȳ 2 − 2¯x 2 = λ(y′ + y 2 − 2 ). (9.3)<br />

x2 Using our dilatation <strong>group</strong> in the expression above yields,<br />

ȳ ′ + ȳ 2 −<br />

2¯x 2<br />

≡<br />

dȳ<br />

d ¯x + ȳ2 − 2¯x 2<br />

= l k y′ + l 2 y 2 − 2<br />

k 2 x<br />

(9.4)<br />

i.e, we must have l k = l2 = 1 . It follows that l = 1 k 2 k<br />

, where k is an arbitrary constant.<br />

Let k = e a , we then get the dilatation <strong>group</strong><br />

¯x = e a x, ȳ = e −a y. (9.5)<br />

The Lie equations yields<br />

d ¯x<br />

da | a=0 = x<br />

dȳ<br />

da | a=0 = −y. (9.6)<br />

It follows that the Riccati equation admitts the generator (symmetry)<br />

X = x ∂ ∂x − y ∂ ∂y . (9.7)<br />

Example 9.2<br />

Consider the equation of motion <strong>for</strong> a simple harmonic oscillator [1]<br />

End of Example 9.1<br />

ẍ + ω 2 x = 0, (9.8)<br />

where ω is a physical constant. It does not have an explicit dependence on t, <strong>and</strong><br />

is there<strong>for</strong>e invariant under the <strong>group</strong> of time translations, ¯t = t + a. It follows that<br />

(9.8) admits the generator (symmetry)<br />

X = ∂ ∂t . (9.9)<br />

End of Example 9.2<br />

72


We shall now present a more general method <strong>for</strong> finding all <strong>symmetries</strong> of differential<br />

equations. The main requirement <strong>for</strong> finding <strong>symmetries</strong> is that the differential<br />

equations are invariant under a <strong>group</strong> G of point trans<strong>for</strong>mations.<br />

Consider a kth-order differential function<br />

F σ (x,u,u (1) ...u (k) ) = 0 σ = 1... s, (9.10)<br />

where F σ ∈ A, x = (x 1 ,..., x n ) <strong>and</strong> u = (u 1 ,...,u m ).<br />

Definition 9.1<br />

Given any differential function F ∈ A, ord(F) = s, the equation<br />

F(x,u,u (1) ...u (s) ) = 0 (9.11)<br />

defines a manifold in the space of variables x,u,u (1) ...u (s) . This manifold is called<br />

the frame of the sth-order partial differential equation [6].<br />

This can be seen as a space spanned <strong>by</strong> the variables x,u,u (1) ...u (s) , i.e a (s + 2)-<br />

dimensional space, in which we have a (s + 1)-dimensional manifold [6].<br />

Definition 9.2<br />

A system of partial differential equations (9.10), is said to be locally solvable if,<br />

<strong>for</strong> any generic point z 0 = (x 0 ,u 0 ,u 0 (1) ...,u0 (k) ) lying on its frame F σ(x 0 ,u 0 ,u 0 (1) ...,u0 (k) ) =<br />

0, there exist a solution u = φ(x) of the system passing through z 0 such that φ(x)<br />

<strong>and</strong> its derivities assumes the values u 0 ,u 0 (1) ...,u0 (k) at x = x0 [6].<br />

The symmetry of the differential equation can be found <strong>by</strong> concerning its solution<br />

or its frame [6]. In the first case we need to know the solution of the differential<br />

equation. If the solutions are merely permuted among themselves under a <strong>group</strong><br />

G, we say that the differential equation is invariant under the <strong>group</strong> G of point<br />

trans<strong>for</strong>mations. The solution set is described <strong>by</strong> a family of curves. [5]<br />

In the other case, we say that the differential equation is invariant under <strong>group</strong> G<br />

if its frame is an invariat manifold <strong>for</strong> the extension of the <strong>group</strong> G to the same<br />

order as the differential equation. In the second case, where no solutions of the<br />

differential equation is needed, we use determining equations to find <strong>symmetries</strong>.<br />

<br />

<br />

73


CHAPTER 9. SYMMETRIES AND THE MAIN CONSERVATION THEOREM<br />

9.1 Determining equations<br />

The fact that a differential equation is invariant if its frame is an invariant manifold,<br />

togheter with Theorem 3.3 gives us a criterion <strong>for</strong> finding a symmetry <strong>group</strong><br />

[6].<br />

Theorem 9.1<br />

The system (9.10) is invariant under the <strong>group</strong> with generator X, extended to all<br />

its derivatives, if <strong>and</strong> only if<br />

XF σ | (9.10) = 0 (9.12)<br />

This equation determines all infinitesimal <strong>symmetries</strong> of the system (9.10) <strong>and</strong> is<br />

there<strong>for</strong>e called the determining equation. As in equation (3.55) the determining<br />

equation can be written as<br />

XF σ = φ ν σF ν , (9.13)<br />

where φ ν σ are differential functions bounded on the frame (9.10)[6].<br />

The left-h<strong>and</strong> side of the determining equation is a function of all variables in<br />

F, i.e x,u,u (1) ...u (k) togheter with the functions ξ i <strong>and</strong> η α which are functions of<br />

x,u (they are the functions in our extended generator X). Since this should be<br />

equal to zero we must have that the coefficients <strong>for</strong> the partial derivities u (i) must<br />

equal zero <strong>for</strong> each i = 1...k. Thus, the determining equation breaks down to a<br />

system of differential equation involving derivities of ξ i , η α with respect to x,u.<br />

This system is often overdetermined <strong>and</strong> can there<strong>for</strong>e be solved analytically <strong>for</strong><br />

the functions ξ i <strong>and</strong> η α . When we know ξ i <strong>and</strong> η α , we also know the infintesimal<br />

symmetry of our differential equation (9.10).<br />

<br />

Theorem 9.2<br />

The solutions of any determining equations <strong>for</strong>m a Lie algebra [6].<br />

<br />

74


9.1. DETERMINING EQUATIONS<br />

Example 9.3<br />

Consider the time independent partial differential equation from transonic gas<br />

dynamics [6]<br />

u x u xx + u yy = 0. (9.14)<br />

To find the <strong>symmetries</strong> of equation (9.14) we use the determining equation (9.12),<br />

where X is the twice extended generator given <strong>by</strong> (4.39) in chapter 4. Thus, from<br />

the determining equation we get<br />

u xx ζ 1 + u x ζ 11 + ζ 22 = 0. (9.15)<br />

Inserting the expressions <strong>for</strong> ζ 1 ,ζ 11 ,ζ 22 from equation (4.42),(4.44) <strong>and</strong> (4.46) in<br />

chapter 4, <strong>and</strong> set u yy = −u x u xx . Thus, the partial differential equation to solve <strong>for</strong><br />

ξ 1 ,ξ 2 ,η becomes<br />

u xx η x + u xx u x η u − u xx u x ξ 1 x − u xx (u x ) 2 ξ 1 u − u xx u y ξ 2 x − u xx u x u y ξ 2 u<br />

+ u x η xx + 2(u x ) 2 η xu + u x u xx η u + (u x ) 3 η uu − 2u x u xx ξ 1 x − (u x ) 2 ξ 1 xx<br />

− 2(u x ) 3 ξ 1 xu − 3(u x ) 2 u xx ξ 1 u − (u x ) 4 ξ 1 uu − 2u x u xy ξ 2 x − u x u y ξ 2 xx<br />

− 2(u x ) 2 u y ξ 2 xu − u x u y u xx ξ 2 u − 2(u x ) 2 u xy ξ 2 u − (u x ) 3 u y ξ 2 uu<br />

+ η yy + 2u y η yu − u x u xx η u + (u y ) 2 η uu + 2u x u xx ξ 2 y − u y ξ 2 yy<br />

− 2(u y ) 2 ξ 2 yu + 3u y u x u xx ξ 2 u − (u y ) 3 ξ 2 uu − 2u xy ξ 1 y − u x ξ 1 yy<br />

− 2u x u y ξ 1 yu − (u x ) 2 u xx ξ 1 u − 2u y u xy ξ 1 u − u x (u y ) 2 ξ 1 uu = 0. (9.16)<br />

This might seem quite complicated but this is a function of x,y,u,u x ,u y ,u xx ,u xy .<br />

Since ξ 1 ,ξ 2 ,η only depends on x,y,u we may isolate the terms containing u x ,u y ,u xx ,u xy<br />

<strong>and</strong> those free of these variables, <strong>and</strong> set each term equal to zero. For the terms<br />

containing u xy we get,<br />

Leading to<br />

u xy<br />

(<br />

−2ux ξ 2 x − 2u 2 xξ 2 u − 2ξ 1 y − 2u y ξ 1 u)<br />

= 0. (9.17)<br />

ξ 1 y = 0, ξ 1 u = 0, ξ 2 x = 0, ξ 2 u = 0. (9.18)<br />

For the terms containing u xx ,<strong>and</strong> insertion of (9.18), we get,<br />

u xx<br />

(<br />

ηx + u x<br />

(<br />

ηu − 3ξ 1 x + 2ξ 2 y))<br />

. (9.19)<br />

75


CHAPTER 9. SYMMETRIES AND THE MAIN CONSERVATION THEOREM<br />

Leading to<br />

η x = 0, η u − 3ξ 1 x + 2ξ 2 y = 0. (9.20)<br />

Now, using the results from equations (9.18) <strong>and</strong> (9.20), equation (9.16) reduces<br />

to<br />

Leading to<br />

(u x ) 3 η uu − (u x ) 2 ξ 1 xx + η yy + u y (η yu − ξ 2 yy) + (u y ) 2 η uu = 0 (9.21)<br />

η uu = 0, η yy = 0, ξ 1 xx = 0, η yu − ξ 2 yy = 0. (9.22)<br />

Thus, the system (9.16) is now split into the overdetermined system of linear<br />

partial differential equations (9.18),(9.20) <strong>and</strong> (9.22). This system is solved <strong>by</strong><br />

straight <strong>for</strong>ward calculation. From equation (9.18) it follows that ξ 1 = ξ 1 (x) <strong>and</strong><br />

ξ 2 = ξ 2 (y). Using the fact that ξ 1 xx = 0 yields<br />

ξ 1 = C 1 x +C 2 . (9.23)<br />

From equation (9.20) we have that η = η(y,u) <strong>and</strong><br />

η u = 3ξ 1 x + 2ξ 2 y = 3C 1 + 2ξ 2 y, (9.24)<br />

where equation (9.23) was used in the last step. From the first two equations in<br />

(9.22) we also know that η is linear in y <strong>and</strong> u. Now, differentiate (9.24) with<br />

respect to y <strong>and</strong> inserting it into the last equation in (9.22) yields,<br />

thus,<br />

ξ 2 yy = 0, (9.25)<br />

ξ 2 = C 3 y +C 4 . (9.26)<br />

Substituting (9.26) into (9.24) <strong>and</strong> integrate with respect to u, keeping in mind<br />

that η is linear in y, yields,<br />

η = (3C 1 + 2C 3 )u +C 5 y +C 6 . (9.27)<br />

Thus, the general solution to (9.15) is<br />

ξ 1 = C 1 x +C 2 , ξ 2 = C 3 y +C 4 , η = (3C 1 + 2C 3 )u +C 5 y +C 6 . (9.28)<br />

76


9.1. DETERMINING EQUATIONS<br />

This solution depends on six arbitrary constants (C 1 ,...,C 6 ). Thus, we arrive at a<br />

six-dimensional Lie algebra spanned <strong>by</strong> the operators<br />

X 1 = ∂ ∂x ,<br />

X 3 = ∂ ∂u ,<br />

X 2 = ∂ ∂y<br />

X 4 = y ∂ ∂u<br />

X 5 = x ∂ ∂x + 3u ∂ ∂u ,<br />

X 6 = y ∂ ∂y − 2u ∂ ∂u<br />

(9.29)<br />

This is <strong>obtained</strong> <strong>by</strong> successively letting one of the constants C i be equal to one,<br />

while the others are zero, <strong>for</strong> all i = 1,...,6.<br />

Example 9.4<br />

End of Example 9.3<br />

The motion of a free particle (i.e no external <strong>for</strong>ces are acting on it) in space is<br />

described <strong>by</strong> the differential equation<br />

∂ 2 x α<br />

= 0 α = 1,2,3 (9.30)<br />

∂t2 where x α = x α (t) is the dependent, or differential variables (x 1 , x 2 , x 3 ) <strong>and</strong> t is the<br />

independent variable. To find the <strong>symmetries</strong> of the partial differential equation<br />

(9.30) we look <strong>for</strong> a generator of the <strong>for</strong>m<br />

X = ξ(t, x 1 , x 2 , x 3 ) ∂ ∂t + ηα (t, x 1 , x 2 , x 3 ) ∂<br />

∂x α (9.31)<br />

<strong>and</strong> determining the tangent vector fields <strong>by</strong> solving equation (9.12) <strong>for</strong> ξ <strong>and</strong> η α .<br />

F σ in equation (9.12) is equation (9.30) here, <strong>and</strong> the twice extended generator is<br />

given <strong>by</strong><br />

X = ξ ∂ ∂t + ∂<br />

ηα<br />

∂x α + ζα 1<br />

∂<br />

∂x α 1<br />

+ ζ α 2<br />

∂<br />

∂x α 2<br />

, (9.32)<br />

where (x1 α, xα 2<br />

) denotes the first <strong>and</strong> second derivatives of component α respectively.<br />

The prolongations ζ1 α <strong>and</strong> ζα 2<br />

is given <strong>by</strong> (4.24),<br />

ζ α 1 = D t (η α ) − x α 1 D t (ξ)<br />

ζ α 2 = D t (ζ α 1 ) − x α 2 D t (ξ) (9.33)<br />

77


CHAPTER 9. SYMMETRIES AND THE MAIN CONSERVATION THEOREM<br />

Now, when (9.32) acting on (9.30), i.e,<br />

(<br />

ξ ∂ ∂t + ∂<br />

) ηα<br />

∂x α + ∂<br />

ζα 1 + ζ2<br />

α ∂ ∂ 2 x α<br />

= 0, (9.34)<br />

∂t2 ∂x α 1<br />

∂x α 2<br />

we arrive at the following system of equations to be solved:<br />

ζ α 2 = 0. (9.35)<br />

We use the prolongation <strong>for</strong>mulas (9.33) to calculate ζ2 α <strong>and</strong> ζα 1<br />

(note that the<br />

prolongations are defined recursively, so in order to calculate ζ2 α we first need to<br />

calculate ζ1 α).<br />

ζ1 α = ∂ηα + x β ∂η α ( )<br />

∂ξ<br />

t<br />

∂t ∂x β − xα t<br />

∂t + ∂ξ<br />

xγ t<br />

∂x γ<br />

= ∂ηα + x β ∂η α<br />

t<br />

∂t ∂x β − ∂ξ<br />

xα t<br />

∂t − xα t xt<br />

γ ∂ξ<br />

∂x γ , (9.36)<br />

where the lowercase letter t on x denotes the derivatives, i.e x t ≡ ∂x/∂t. α is a<br />

component index, while β <strong>and</strong> γ are dummy summation indicies. When calculating<br />

ζ α 2 we use the fact that x tt ≡ ∂ 2 x/∂t 2 = 0. Note that we directly invoking the<br />

chain rule.<br />

ζ α 2 = D t (ζ α 1 )<br />

( ∂η<br />

α<br />

= D t<br />

∂t<br />

) ( )<br />

+ D t xt<br />

β ∂η<br />

α<br />

(<br />

( )<br />

− D t x<br />

α ∂ξ ∂ξ<br />

t<br />

∂t − xα t D t<br />

∂t<br />

( ) ∂η<br />

α<br />

∂x β + xβ t D t<br />

∂x<br />

)<br />

β ( )<br />

− D t x<br />

α γ ∂ξ<br />

t xt<br />

∂x γ<br />

− x α t D t<br />

(<br />

x<br />

γ<br />

t<br />

) ∂ξ<br />

∂x γ − xα t x γ t D t<br />

( ∂ξ<br />

∂x γ )<br />

= ∂2 η α<br />

∂t 2 + ∂ 2 η α<br />

xβ t<br />

∂t∂x β + ∂x xγ t<br />

β ∂η α<br />

tt<br />

∂xt<br />

γ ∂x β<br />

+ xt<br />

β ∂ 2 η α<br />

∂t∂x β + xβ t xt<br />

γ ∂ 2 η α<br />

∂x γ ∂x β − ∂ξ<br />

xα tt<br />

∂t<br />

− xt<br />

α ∂ 2 ξ<br />

∂t 2 − xα t xt<br />

β ∂ 2 ξ<br />

∂t∂x β − xα ttxt<br />

γ ∂ξ<br />

∂x γ<br />

− xt α xtt<br />

γ ∂ξ<br />

∂x γ − xα t xt<br />

γ ∂ 2 ξ<br />

∂t∂x γ − xα t xt γ xt<br />

β ∂ 2 ξ<br />

∂x γ ∂x β . (9.37)<br />

78


9.2. THE MAIN CONSERVATION THEOREM<br />

Putting all x tt = 0 yields,<br />

ζ α 2 = ∂2 η α<br />

∂t 2<br />

+2xβ t<br />

∂ 2 η α<br />

∂t∂x β + xβ t x γ t<br />

∂ 2 η α<br />

∂x γ ∂x β − ∂ 2 ξ<br />

xα t<br />

∂t 2 −2xα t xt<br />

β<br />

∂ 2 ξ<br />

∂t∂x β − xα t x γ t x β t<br />

∂ 2 ξ<br />

∂x γ ∂x β<br />

(9.38)<br />

This system is split into several systems <strong>by</strong> equating to zero the terms free of first<br />

order derivatives x t , the terms linear, quadratic <strong>and</strong> cubic in x t . Thus, we arrive at<br />

the following system of equations:<br />

∂ 2 η α<br />

∂t 2 = 0<br />

∂<br />

∂t<br />

(2 ∂ηα<br />

∂x β − δα β<br />

(<br />

∂ ∂η<br />

α<br />

∂x γ ∂x β − 2δα β<br />

)<br />

∂ξ<br />

∂t<br />

∂ξ<br />

∂t<br />

= 0<br />

)<br />

= 0<br />

∂ 2 ξ<br />

∂x β ∂x γ = 0 (9.39)<br />

The solution to the system (9.39) depends on 24 arbitrary constants <strong>and</strong> there<strong>for</strong><br />

possess 24 infinitesimal <strong>symmetries</strong>. They <strong>for</strong>m a Lie algebra spanned <strong>by</strong> the<br />

following operators (see [6]):<br />

X 0 = ∂ ∂t , X α = ∂<br />

∂x α , S = t ∂ ∂t , P α = x α ∂ ∂t , Q ∂<br />

α = t<br />

∂x α<br />

Y αβ = x α ∂<br />

∂x β , Z 0 = t 2 ∂ ∂t + ∂ txβ ∂x β , Z α = x α t ∂ ∂t + xα x β ∂<br />

∂x β (9.40)<br />

End of Example 9.4<br />

9.2 The main <strong>conservation</strong> theorem<br />

In chapter 4 we introduced the concept of extended operators due to the fact that<br />

we treat all varibles x i ,u α <strong>and</strong> all derivities as independent even if they are connected<br />

<strong>by</strong> the relations given in (2.30) in chapter 2. To find a Lagrangian <strong>for</strong> an<br />

arbitrary system of differential equation we used the concept of adjoint equations<br />

<strong>and</strong> the lagrangian were constructed accordingly to theorem 8.4 in chapter 8, <strong>by</strong><br />

79


CHAPTER 9. SYMMETRIES AND THE MAIN CONSERVATION THEOREM<br />

considering the simultanious system. By introducing the adjoint equations we<br />

arrived at a system of 2m equations with a total of 2m dependent variables, m dependent<br />

variables from the original system <strong>and</strong> m new dependent variables from<br />

the system of adjoint equations. Now we need to show that the system of adjoint<br />

equations inherit the <strong>symmetries</strong> of the original system.<br />

Consider a system of m differential equations of order s,<br />

F α (x,u,u (1) ,...,u (s) ) = 0 α = 1,...,m, (9.41)<br />

with n independent variables x = (x 1 ,..., x n ) <strong>and</strong> m dependent variables<br />

u = (u 1 ,...,u m ). The adjoint system is<br />

F ∗ α(x,u,v,...,u (s) ,v (s) ) ≡ δ(vβ F β )<br />

δu α = 0 α = 1,...,m (9.42)<br />

where v = (v 1 ,...,v m ) are new dependent variables v = v(x). Consider a point<br />

symmetry <strong>for</strong> the original system (9.42) with a generator of the <strong>for</strong>m<br />

X = ξ i ∂<br />

∂x i + ∂<br />

ηα<br />

∂u α , (9.43)<br />

where the prolongation to all derivities involved in equation (9.42) is understood.<br />

The condition <strong>for</strong> equation (9.42) to be invariant is given <strong>by</strong> (3.55) in chapter 3,<br />

XF α = λ β αF β (9.44)<br />

The Lagrangian <strong>for</strong> the simultanious system 9.41 <strong>and</strong> 9.42 is given <strong>by</strong> [15]<br />

L = v α F α . (9.45)<br />

The extension of the generator (9.43) to the adjoint variables <strong>and</strong> their derivities<br />

(which is understood) is the generator [3]<br />

Y = ξ i ∂<br />

∂x i + ∂<br />

ηα<br />

∂u α + ∂<br />

ηα ∗<br />

∂v α . (9.46)<br />

The Lagrangian should remain invariant under a certian point trans<strong>for</strong>mation, we<br />

can find a condition on the infinitesimal tangent vector field associated with the<br />

adjoint variables, η α ∗ , <strong>by</strong> require the the invariance condition <strong>for</strong> the variational<br />

integral, according to lemma 7.1, should be satisfied, i.e<br />

Y(L) + LD i (ξ i ) = 0 (9.47)<br />

80


9.2. THE MAIN CONSERVATION THEOREM<br />

Direct calculations of equation (9.47), yields,<br />

Y(L) + LD i (ξ i ) = Y(v α F α ) + v α F α D i (ξ i )<br />

= Y(v α )F α + v α Y(F α ) + v α F α D i (ξ i )<br />

= Y(v α )F α + v α X(F α ) + v α F α D i (ξ i )<br />

= η α ∗ F α + v α λ β α(F β ) + v α F α D i (ξ i )<br />

= [η α ∗ + v β λ α β + v α D i (ξ i )]F α = 0, (9.48)<br />

where λ α β is defined <strong>by</strong> equation (9.44). To guarantee invariance of the simultanious<br />

system we arrive at the following condition <strong>for</strong> tangent vectorfield associated<br />

with the adjoint variables:<br />

η α ∗ = −[v β λ α β + v α D i (ξ i )] (9.49)<br />

Thus, if the original system admitts the operator (9.43), the adjoint system admitts<br />

the operator<br />

Y = ξ i ∂<br />

∂x i + ∂<br />

ηα<br />

∂u α − [vβ λ α β + v α D i (ξ i )] ∂<br />

∂v α . (9.50)<br />

Since a operator of this <strong>for</strong>m (9.50) provids a solution <strong>for</strong> the invariance condition<br />

(9.47), which is a determining equation, we have accordingly to theorem 9.2 that<br />

this operator <strong>for</strong>m a Lie algebra. Thus, it follows that the adjoint system inherit<br />

the <strong>symmetries</strong> of the original system. This can be <strong>for</strong>mulated in a theorem.<br />

Theorem 9.3<br />

For a system described <strong>by</strong> equation (9.41) that admits a generator (9.43), its adjoint<br />

system (9.42) inherits the <strong>symmetries</strong> of the original system, i.e the adjoint<br />

system admitts the generator (9.43) extended to the adjoint variables according to<br />

(9.50) [3].<br />

Theorem 9.4: The main <strong>conservation</strong> theorem<br />

Every Lie point, Lie-Bäcklund <strong>and</strong> non-local symmetry<br />

X = ξ i (x,u,u (1) ,...) ∂<br />

∂x i + ηα (x,u,u (1) ,...) ∂<br />

∂u α (9.51)<br />

<br />

81


CHAPTER 9. SYMMETRIES AND THE MAIN CONSERVATION THEOREM<br />

of differential equations<br />

F α (x,u,u (1) ,...,u (s) ) = 0, α = 1,...,m, (9.52)<br />

provides a <strong>conservation</strong> law <strong>for</strong> the system of differential equation (9.52) considered<br />

togheter with its adjoint system of equations [3].<br />

By using the determinig equation we can find the <strong>symmetries</strong> <strong>for</strong> any system of m<br />

differential equations. If the system of differential equations is well-posed, i.e the<br />

number of equations is equal to the number of dependent variables, we construct a<br />

Lagrangian <strong>by</strong> considering the original system togheter with the adjoint system of<br />

equations. Since the adjoint system inherit the <strong>symmetries</strong> of the original system,<br />

according to theorem 9.3, we now have a total system of 2m equations with known<br />

<strong>symmetries</strong> <strong>and</strong> a Lagrangian. Thus, we can now use Noethers theorem to find a<br />

<strong>conservation</strong> law. We illustrate the proccedure with an example.<br />

<br />

Example 9.5<br />

The mathematical <strong>for</strong>mulation <strong>for</strong> the motion of a free particle in space is given<br />

<strong>by</strong> the differential equations<br />

mẍ = 0, (9.53)<br />

where x = (x 1 , x 2 , x 3 ) is the spatial coordinates, <strong>and</strong> m is the mass of the particle.<br />

Thus, we have a system of differential equations of the <strong>for</strong>m<br />

F α (t, x α , ẋ α , ẍ α ) = 0 α = 1,2,3. (9.54)<br />

Here we have that t is the independent variable <strong>and</strong> x = x(t) is the dependent<br />

variable. The differential variables are x α , ẋ α , ẍ α <strong>and</strong> we treat the as if they were<br />

independent according to the definitions given in chapter 2.<br />

Equation (9.54) is invariant under rotational symmetry with generators of the <strong>for</strong>m<br />

X i j = x j ∂<br />

∂x i − ∂<br />

xi<br />

∂x j . (9.55)<br />

82


9.2. THE MAIN CONSERVATION THEOREM<br />

Note that this generators is the linear combination Y αβ − Y βα from example 9.3.<br />

Thus, we have the following three generators:<br />

X 12 = x 2 ∂<br />

∂x 1 − ∂<br />

x1<br />

∂x 2<br />

X 13 = x 3 ∂<br />

∂x 1 − ∂<br />

x1<br />

∂x 3<br />

X 23 = x 3 ∂<br />

∂x 2 − ∂<br />

x2<br />

∂x 3 . (9.56)<br />

Consider rotation around the x 3 -axis, that is generator X 12 . For that one we have<br />

a tangent vector field with components ξ = 0, η 1 = x 2 , η 2 = −x 1 <strong>and</strong> η 3 = 0. The<br />

twice extended generator, i.e the prolongation to the differential variables involved<br />

in (9.53) is given <strong>by</strong><br />

X 12,(2) = x 2 ∂<br />

∂x 1 − ∂<br />

x1<br />

∂x 2 + ∂<br />

ζα 2<br />

∂ẍ α α = 1,2,3, (9.57)<br />

where ζ α 1 <strong>and</strong> ζα 2<br />

is given <strong>by</strong><br />

ζ α 1 = D t<br />

(<br />

ζ<br />

α<br />

1<br />

)<br />

− ẋ α D t (ξ)<br />

ζ α 2 = D t<br />

(<br />

ζ<br />

α<br />

2<br />

)<br />

− ẍ α D t (ξ). (9.58)<br />

First we calculate ζ α 1<br />

<strong>and</strong> use the fact that ξ = 0. For α = 1,2,3 we get<br />

ζ 1 1 = D t (η 1 ) = D t (x 2 ) = ẋ 2<br />

ζ 2 1 = D t (η 2 ) = −D t (x 1 ) = −ẋ 1<br />

ζ 3 1 = 0. (9.59)<br />

For ζ α 2<br />

we now get in a similar way<br />

ζ2 1 = D t (ζ1) 1 = D t (ẋ 2 ) = ẍ 2<br />

ζ2 2 = D t (ζ1) 2 = −D t (ẋ 1 ) = −ẍ 1<br />

ζ2 3 = 0. (9.60)<br />

Thus, the twice extended generator is given <strong>by</strong><br />

X 12,(2) = x 2 ∂<br />

∂x 1 − ∂<br />

x1<br />

∂x 2 + ∂<br />

ẍ2<br />

∂ẍ 1 − ∂<br />

ẍ1<br />

∂ẍ 2 . (9.61)<br />

83


CHAPTER 9. SYMMETRIES AND THE MAIN CONSERVATION THEOREM<br />

The simultanious system (9.53) considered togheter with its adjoint system<br />

has the Lagrangian, according to (9.45)<br />

L = y β F β = my 1 ẍ 1 + my 2 ẍ 2 + my 3 ẍ 3 . (9.62)<br />

The extension of X 12 to the new adjoint differential variables y α is<br />

Y = x 2 ∂<br />

∂x 1 − ∂<br />

x1<br />

∂x 2 + ∂<br />

ηα ∗<br />

∂y α , (9.63)<br />

where η α ∗ is given <strong>by</strong> equation (9.49). In this case, since ξ = 0 it is simply<br />

λ α β<br />

η α ∗ = −λ α βy β . (9.64)<br />

is obtaind from the invariance condition (9.44). In our case we have<br />

X 12,(2) F α = λ β αF β . (9.65)<br />

Per<strong>for</strong>ming this calculation yields<br />

ẍ 2 = λ 1 1ẍ 1 + λ 2 1ẍ 2 + λ 3 1ẍ 3<br />

−ẍ 1 = λ 1 2ẍ 1 + λ 2 2ẍ 2 + λ 2 1ẍ 3<br />

0 = λ 1 3ẍ 1 + λ 2 3ẍ 2 + λ 3 3ẍ 3 , (9.66)<br />

<strong>for</strong> α = 1,2,3 respectivley. For these equations to hold we must have that λ 2 1 = 1<br />

<strong>and</strong> λ 1 2 = −1 <strong>and</strong> the others must be zero. Using equation (9.64) we get<br />

η 1 ∗ = −λ 1 1y 1 − λ 1 2y 2 − λ 1 3y 3 = y 2<br />

η 2 ∗ = −λ 2 1y 1 − λ 2 2y 2 − λ 2 3y 3 = −y 1<br />

η 3 ∗ = 0 (9.67)<br />

We now have all components of the extended tangent vector field <strong>for</strong> our simultanious<br />

system. To find the conserved quantity associated with the extended generator<br />

(9.63) we apply Noether’s theorem <strong>for</strong> a second order Lagrangian. Thus,<br />

we will get a conservd quantity C of the <strong>for</strong>m<br />

C = ξL + W α [ ∂L<br />

∂u α t<br />

]<br />

∂L<br />

− D t<br />

∂u α + D t (η α ) ∂L<br />

tt ∂u α , (9.68)<br />

tt<br />

84


9.3. SUMMARY<br />

where u α = (x 1 , x 2 , x 3 ,y 1 ,y 2 ,y 3 ),u α t = (ẋ 1 , ẋ 2 , ẋ 3 ,ẏ 1 ,ẏ 2 ,ẏ 3 ),u α tt = (ẍ 1 , ẍ 2 , ẍ 3 ,ÿ 1 ,ÿ 2 ,ÿ 3 ).<br />

For the tangent vector field components we have ξ = 0 <strong>and</strong> η α = (x 2 ,−x 1 ,0,y 2 ,−y 1 ,0).<br />

Since ξ = 0 we have that W α = η α , recall definition from chapter 5.3. In this case<br />

we also have that ∂L/∂u α t = 0,∀α. The conserved quantity (9.68) reduces to<br />

C = −η α ∂L<br />

D t<br />

∂u α + D t (η α ) ∂L<br />

tt ∂u α tt<br />

= −x 2 D t (my 1 ) − (−x 1 )D t (my 2 ) − 0 − y 2 D t (0) − (−y 1 )D t (0) − 0<br />

+ D t (x 2 )my 1 + D t (−x 1 )my 2 + D t (0)my 3 + D t (y 2 ) · 0 + D t (−y 1 ) · 0 + 0<br />

= −mx 2 ẏ 1 + mx 1 ẏ 2 + my 1 ẋ 2 − my 2 ẋ 1 . (9.69)<br />

Using the fact that the system (9.53) is self-adjoint, i.e we can replace the adjoint<br />

variable y with x <strong>and</strong> also that ẋ ≡ v, the velocity, we get<br />

C = −mx 2 v 1 + mx 1 v 2 + mx 1 v 2 − mx 2 v 1<br />

= 2m ( x 1 v 2 − x 2 v 1) . (9.70)<br />

Now, putting M = C/2, we recognice this as the third component of angular momentum.<br />

Doing the same calculations <strong>for</strong> X 13 <strong>and</strong> X 23 we arrive at the conserved<br />

quantity<br />

M = m(x×v). (9.71)<br />

I.e, <strong>conservation</strong> of angular momentum!<br />

End of Example 9.5<br />

This example illustrates the mathematical method to find a <strong>conservation</strong> law from<br />

a known symmetry.<br />

9.3 Summary<br />

For an arbitrary system of differential equations with a known symmetry <strong>and</strong> a<br />

known Lagrangian, we may use Noethers theorem to determinig a <strong>conservation</strong><br />

law, i.e Noethers theorem gives a connection between <strong>symmetries</strong> <strong>and</strong> <strong>conservation</strong><br />

<strong>laws</strong> <strong>and</strong> provides a simple method <strong>for</strong> calculating them <strong>for</strong> variational<br />

85


CHAPTER 9. SYMMETRIES AND THE MAIN CONSERVATION THEOREM<br />

problems.<br />

The <strong>symmetries</strong> of differential equations may be found <strong>by</strong> using Lie <strong>group</strong> theory,<br />

presented in Part I. By solving the determining equation we have the <strong>symmetries</strong>.<br />

At this point we have two cases to consider. First, if we have a Lagrangian, then<br />

we simply use Noethers theorem to calculate the <strong>conservation</strong> law <strong>for</strong> the <strong>symmetries</strong><br />

<strong>obtained</strong> <strong>by</strong> Lie <strong>group</strong> theory. On the other h<strong>and</strong>, if we don’t have a<br />

Lagrangian but the number of equations is equal to the number of dependent variables,<br />

the Main <strong>conservation</strong> theorem states that we have a <strong>conservation</strong> law if we<br />

consider our original system with its adjoint system. From the adjoint system we<br />

can construct a Lagrangian <strong>and</strong> hence, Noethers theorem can be used to calculate<br />

the associated <strong>conservation</strong> law. For more examples on this I recommend Refs.<br />

16 <strong>and</strong> 3.<br />

86


BIBLIOGRAPHY<br />

[1] Jerry B. Marion <strong>and</strong> Stephen T. Thornton. Classical dynamics of particles<br />

<strong>and</strong> systems. Thomson Brooks/Cole, United States, 1995. 4th Edition.<br />

[2] Bo Thidé. Electromagnetic Field Theory. Upsilon Books, 2007.<br />

http://www.plasma.uu.se/CED/Book.<br />

[3] Nail H. Ibragimov. A new <strong>conservation</strong> theorem. J. Math. Anal. Appl.,<br />

333:311–328, 2007.<br />

[4] Nail.H. Ibragimov. A practical course in differential equations <strong>and</strong><br />

mathematical modelling. ALGA publications, Blekinge Institute of<br />

Technology, Karlskrona, Sweden, 2005. 2nd edition.<br />

[5] Robert C. MacOwen. Partial Differential Equations. Pearson Education,<br />

United States, 2003. 2th Edition.<br />

[6] Nail.H. Ibragimov. Elementary Lie Group Analysis <strong>and</strong> Ordinary<br />

Differential Equations. Wiley, Chichester, 1999.<br />

[7] Peter J. Olver. Applications of Lie Groups to Differential Equations.<br />

Springer-Verlag, New York Inc, 1986.<br />

[8] Robert Adams. Calculus: A Complete Course. Addison Wesley, 2003. 5th<br />

Edition.<br />

[9] F. Brickell <strong>and</strong> R.S. Clark. Differentiable Manifolds. Van Nostr<strong>and</strong><br />

Reinhold Company, London, 1970.<br />

[10] R.P. Agarwal <strong>and</strong> V Lakshmikantham. Uniqueness <strong>and</strong> Nonuniqueness<br />

Criteria <strong>for</strong> Ordinary Differential Equations. World Scientific Publishing,<br />

1993.<br />

[11] J.J. Sakurai. Modern Quantum Mechanics. Addison Wesley, 1994. 2nd.<br />

edition.<br />

87


BIBLIOGRAPHY<br />

[12] http://mathworld.wolfram.com/BeltramiIdentity.html.<br />

[13] Emmy Noether <strong>and</strong> M.A. Tavel. Invariant variation problems, 2005.<br />

[14] Nail H. Ibragimov. Exam on symmetry <strong>and</strong> <strong>conservation</strong> <strong>laws</strong>, 2005.<br />

[15] Nail H. Ibragimov. Integrating factors, adjoint equations <strong>and</strong> lagrangians.<br />

J.Math.Anal.Appl., 318:742–757, 2006.<br />

88

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!