LIE CARTAN

Lessons on
Integral Invariants

Translated by
D.H. Delphenich

HERMANN
Editors for the sciences and arts, 156, boulevard Saint-Germain, Paris VI
1922
 TABLE OF CONTENTS

Page

Introduction i

FIRST CHAPTER.

Hamiltons least-action principle and the quantity of motion-energy tensor.

I. - Case of the free material point. 1

II. - General case. 7
III. - Transformation of the canonical equations. - Jacobis theorem. 15

CHAPTER II.

The two-dimensional integral invariants of dynamics.

I. - Formation of two-dimensional integral invariants in dynamics. 18

II. - Application to the theory of vortices. 20

CHAPTER III.

Integral invariants and invariant differential forms.

I. - General notion of an integral invariant. 26

II. - First integrals. 27
III. - Absolute integral invariants and invariant differential forms. 28
IV. - Relative integral invariants. - Hamiltons function. 30
V. - Examples. - The element of matter form. 33

CHAPTER IV.

The characteristic system of a differential form.

I. - The class of a differential form. 39

II. - The characteristic system of a differential form. 40

CHAPTER V.

Systems of Pfaff invariants and their characteristic systems.

I. - The notion of an invariant Pfaff system. 44

II. - The characteristic system of a Pfaff system. 47
- The rank of an algebraic form and its associated system. 48
ii Lessons on integral invariants

CHAPTER VI.

Forms with exterior multiplication.

I. - The associated system of a quadratic form. 50

II. - Alternating bilinear forms and quadratic exterior forms. 51
III. - Exterior forms of degree greater than two. 56
IV. - The associated system of an exterior form. 60
V. - Formulas that relate to quadratic exterior forms. 61

CHAPTER VII.

Exterior differential forms and their derived forms.

I. - The bilinear covariant of a Pfaff form. 68

II. - Exterior derivation. 70
III. - Exact exterior differential forms. 75

CHAPTER VIII.

The characteristic system of an exterior differential form.

Formation of integral invariants.

I. - The characteristic system of an exterior differential form. 79

II. - Formation of integral invariants. 80

CHAPTER IX.

Differential systems that admit an infinitesimal transformation.

I. - The notion of an infinitesimal transformation. 86

II. - Formation of integral invariants when starting with infinitesimal transformations. 88
III. - Examples. 89
IV. - Applications to the n-body problem. 92
V. - Application to the kinematics of rigid bodies. 97
VI. - Differential equations that admit an infinitesimal transformation. 98
VII. - Expressing the idea that a given system of differential equations admits 101
a given infinitesimal transformation.
VIII. - Equations of variation. 102

CHAPTER X.

Completely integrable Pfaff systems.

I. - Frobeniuss theorem. 104

II. - Forming the characteristic system of a Pfaff system. 106
III. - The integration of a completely integrable Pfaff system. 107
IV. Complete systems. 108
 Table of contents iii

CHAPTER XI.

The theory of the last multiplier.

I. - Definition and properties. 110

II. - Generalizations. 113
III. - Case where the independent variable is not specified. 115
IV. - Case where the given equations admit an infinitesimal transformation. 115
V. - Applications. 119

CHAPTER XII.

Equations that admit a relative linear integral invariant.

I. - General method of integration. 126

II. - The Poisson brackets and the Jacobi identity. 130
III. - Use of known first integrals. 132
IV. - Generalization of the Poisson-Jacobi theorem. 135

CHAPTER XIII.

Equations that admit an absolute linear integral invariant.

I. - General method of integration. 137

II. - Generalization of the Poisson-Jacobi formulas. 139
III. - Use of known first integrals. 142

CHAPTER XIV.

Differential equations that admit an invariant Pfaff equation.

I. - General method of integration. 150

II. - Use of known integrals. 152
III. - Application to first order partial differential equations. 154
IV. - Cauchys method. 156
V. - Lagranges method. 157
VI. First-order partial differential equations that admit an infinitesimal transformation. 158
VII. - Jacobis first method. 160
VIII. - Reducing certain differential equations to a first-order partial differential equation. 161
IX. - Remarks on the nature of the principal practical applications of Jacobis method. 162

CHAPTER XV.

Differential equations that admit several linear integral invariants.

I. - Case in which one knows as many integral invariants as there are unknowns. 164
II. - The group that preserves the given invariants. 167
III. - Examples. 169
IV. - Generalizations. 170
iv Lessons on integral invariants

CHAPTER XVI.

Differential equations that admit given infinitesimal transformations.

I. - Reduction of the problem. 172

II. - Case in which there are as many infinitesimal transformations as unknown functions. 175
III. - Application to second-order differential equations. 176
IV. - Generalizations. - Examples. 178

CHAPTER XVII.

Application of the preceding theories to the n-body problem.

I. - Reduction of the number of degrees of freedom. 183

II. - Equations of motion with respect to a moving reference frame. 188
III. - Case in which the area constants are all null. 192
IV. - Case in which the vis viva constant is null. 195

CHAPTER XVIII.

Integral invariants and the calculus of variations.

I. - Extremals attached to a relative integral invariant. 198

II. - The least-action principle of Maupertuis. 200
III. - Generalizations. 202
IV. - Application to the propagation of light in an isotropic medium. 203

CHAPTER XIX.

Fermats principle and the invariant Pfaff equation of optics.

I. - Fermats principle. 209

II. - The invariant Pfaff equation of optics. 212
III. - Fermats principle is independent of the spacetime framing. 215

BIBLIOGRAPHY.

- Bibliography. 218
 INTRODUCTION

This work is the reproduction of a professional course that was held during the summer
semester of 1920-1921 at the Facult des Sciences in Paris.
The theory of integral invariants was founded by H. Poincar and examined by him in tome
III of his Mthodes nouvelles de la Mchanique cleste.
In his notes to the Comptes rendus de lAcadmie des Sciences (16 and 30 June, 1902) the
author was led, in the study of differential equations that admit given transformations, to
consider certain differential forms that are called integral forms. They are the ones that are
characterized by the property that they are expressible in terms of only first integrals of the given
differential equations and their differentials. It was in contemplating research along the same
lines that the author arrived, on the one hand, to found his method of integration of systems of
partial differential equations that admit characteristics that depend only on arbitrary constants
(Cauchy characteristics), and, on the other hand, to found his theory of the structure of
continuous groups of transformations, both finite and infinite.
Now, one finds that the notion of integral form does not differ essentially from that of
integral invariant. It is the confrontation of these two notions that forms the basis for this work.
For example, consider a system of three first-order differential equations in the three
unknown functions x, y, z of the independent variable t. One may regard them as defining an
infinitude of trajectories of a moving point. A differential form, such as P dx + Q dy + R dz + H
dt, for example, may be envisioned as a quantity that is attached to a state (x, y, z, t) of the
moving point and an infinitely close state (x + dx, y + dy, z + dz, t + dt). We say that this form is
integral (or invariant, following the terminology that will be adopted in these Lessons), which
obviously signifies that this quantity depends only on the trajectory that contains the first state
and the infinitely close trajectory that contains the second state. In other words, an invariant
form does not change its value if one displaces the two states (x, y, z, t) and (x + dx, y + dy, z +
dz, t + dt) along their trajectories in an arbitrary manner. If one then considers a continuous,
linear collection of trajectories, and if one takes the integral P dx + Q dy + R dz along the arc of
the curve of positions taken by the moving point on its trajectories at the same instant t then this
integral will be independent of t; it is an integral invariant in the sense of H. Poincar.
Conversely, by a method that is quite simple to exhibit, there exists an invariant form P dx + Q
dy + R dz + H dt that corresponds to the integral invariant P dx + Q dy + R dz of H. Poincar.
These considerations are not limited to linear differential forms. Any invariant differential
form that is susceptible to being placed under an integration sign simple or multiple gives
rise to an integral invariant, in the sense of H. Poincar, if one suppresses the terms that contain
linear differential forms or the differentials of the independent variables (1).
By definition, the quantity under the integration sign in an integral invariant, in the sense of
H. Poincar, is nothing but a truncated invariant differential form. The invariant character of
the complete integral is preserved if it is taken over an arbitrary set of states, whether
simultaneous or not.

(1) R. Hargreaves, in a paper in the Transactions of the Cambridge Philosophical Society (v. XXI, 1912), has
already considered integrals that contain differentials of the independent variables. However, his point of view is
completely different from that of this book, and it will always play a role on the part of the independent variable.
vi Lessons on integral invariants

The consequences of this agreement between the two notions of integral invariant and
invariant differential form are numerous. In the first place, all of the properties that relate to the
formation of these integral invariants and the derivation of one from the other are obviously
important in their own right. Similarly, the same is true for the applications to the integration of
differential equations.
Another consequence has been pointed out that relates to the principles of Mechanics. H.
Poincar has shown that the general equations of dynamics possess the property that they admit a
(relative) linear integral invariant, namely:

(1) p1 q1 + p2 q2 + ... + pn qn ,

in which the qi and pi denote Hamiltons canonical variables. If one completes the differential
form under the sign then the integral invariant will take the form:

(2) p1 q1 + p2 q2 + ... + pn qn H t,

in which H denotes Hamiltons function. One thus sees the energy H of the material system
considered along with its quantity of motion (p1, ..., pn). The form under the sign thus acquires
an extremely important mechanical significance. One may give it the name of the quantity of
motion-energy tensor (1). The elementary Hamilton action is nothing but this tensor, when
considered along a trajectory. The notion of action is therefore related to the notion of the
quantity of motion and energy.
There is more: Not only do the differential equations of motion admit the integral invariant
(2), but they are, moreover, the only differential equations that enjoy this property. One may
then place the following principle at the basis of mechanics, which one may give the name of
the principle of conservation of motion and energy:

The motion of a material system (with perfect holonomic constraints and subject to forces
that are derived from a function of position) is regulated by first-order differential equations in
time, the parameters of position, and the parameters of velocity, and these differential equations
are characterized by the property that when the integral of the quantity of motion-energy
tensor is taken over an arbitrary closed, continuous, linear collection of states of the system it
will not change value when one displaces these states in an arbitrary manner along their
respective trajectories.

In this statement, the word state refers to the set of quantities that define the position of the
system in space, the instant where it is considered, and its velocity at that instant.
The preceding statement is more abstract and less intuitive than Hamiltons principle of least
action, for example. It nevertheless has an advantage that is important to point out. Lagranges
equations permit us to give the laws of mechanics a form that is independent of the framing
adopted for space, and this is what makes them important; however, time occupies a privileged

1
( ) The indicated form presents itself in a completely natural way when one calculates the variation of Hamiltons
action integral. This point of view has already been pointed out, which is why it is introduced in these Lessons.
 Introduction vii

position in them. On the contrary, the principle of the conservation of the quantity of motion and
energy gives the laws of mechanics a form that is independent of the framing adopted for the
universe (space-time). If one effects a change of variables that involves both the parameters of
position and time for the system then it suffices to know the form that will be taken by the
quantity of motion energy tensor in the new system of coordinates in order to deduce the
equations of motion. One thus obtains a schema to which all of the mechanical theories will be
subordinate, and to which, indeed, relativistic mechanics itself is subordinate.
It is important to remark that this schema applies only to material systems that depend on a
finite number of parameters.

The present work omits a great number of applications of the theory of integral invariants. In
particular, some that are extremely important to celestial mechanics, and are related to the theory
of periodic solutions of the three-body problem and the theory of Poisson stability are
systematically omitted. One is principally limited to applications that relate to the integration of
differential equations, although the problem is only begun along those lines.
Meanwhile, one is forced to show that this problem cannot be considered in isolation. One
only makes it narrow in scope if one does not regard it as a particular aspect of a more general
problem in which not only the consideration of integral invariants enter into it, but also that of
invariant Pfaff equations for the given differential equations, as well as the infinitesimal
transformations that preserve these differential equations. A complete exposition of the problem
is beyond the scope of these Lessons and would demand some knowledge of the theory of
continuous groups. On several occasions, one is limited to showing the fundamental role that is
played in the final analysis by the group G of transformations that, when applied to the integrals
of the given differential equations, leave all of the information that is known a priori about these
integrals invariant (1). Any system of differential equations may be converted into a system of
this type with a corresponding a simple group G. If this simple group is finite then one will
obtain systems of differential equations that have been studied especially by S. Lie and Vessiot,
who gave them the name of Lie systems. They are attached to the theory of integral invariants, in
the sense that they admit as many linear integral invariants as they have unknown functions, if
necessary by the adjunction of auxiliary unknown functions. One will find some general
indications of why one would take this latter viewpoint in chapter XV of these Lessons.
If the group G is infinite, and if one abstracts to the case where this is the most general group
in n variables a case in which one knows nothing about the corresponding system of
differential equations then it admits either an integral invariant of maximum degree (the theory
of the Jacobi multiplier), a relative linear integral invariant (the theory of equations that are
reducible to canonical form), or an invariant Pfaff equation (equations that amount to a first-
order partial differential equation). Chapters XI-XIV are dedicated to these classical theories.
The notion of an integral invariant may be envisioned from a viewpoint that is slightly
different from the usual one, which is that of H. Poincar, and which is, in summary, the one that
is used in these Lessons. Instead of considering a system of differential equations as being
attached to a multiple integral, relative to which they enjoy the property of invariance, one may
consider the system as being attached to a group of transformations with respect to which they
are invariant. These viewpoints are, moreover, related. The latter is the one that S. Lie took and
the one that he has wagered on several occasions to be the only truth. There again, the notion of

(1) Cf. . Cartan. - Les sous-groups des groups continus de transformations; Ann. Ec. Norm. (3), t. XXV (1908),
pp. 57-194 (Chap. I).
viii Lessons on integral invariants

integral invariant plays an important role since, as the author has shown (1), any group of
transformations may be defined as a set of transformations that admit a certain number of linear
integral invariants, by the adjunction of auxiliary variables, if necessary. This aspect of the
notion of integral invariant is completely omitted in these Lessons.
Several chapters are dedicated to the rules of the calculus of the differential forms that
present themselves under multiple integration signs. Goursat has given these forms the name of
symbolic expressions. I propose to call them differential forms with exterior multiplication, or,
more briefly, exterior differential forms, because they obey the rules of exterior multiplication of
H. Grassmann. Similarly, I propose to call the operation that permits us to pass from a multiple
integral of degree p 1 that is taken over a manifold of dimension p 1 to a multiple integral of
degree p that is taken over a manifold of dimension p that is bounded by the latter, exterior
derivation (2). This operation, which reduces to the classical operations when the coefficients of
the differential form under the sign admit first-order partial derivatives, may be preserved in a
sense when this is no longer true. In this regard, it poses interesting problems that have not been
systematically studied, although they deserve to be.
This work terminates in two chapters, which are very sketchy moreover, on the relations
between the theory of integral invariants, the calculus of variations, and the principles of optics.
One will find a list at the end of this volume, which makes no pretense of being complete, of
the principal works that relate to the theory of integral invariants. Papers that relate to the
classical theory of Jacobi multipliers, canonical equations, and first-order partial differential
equations are cited only when they are directly concerned with the theory of integral invariants.

Le Chesnay, 24 November 1921.

1
( ) . Cartan. - Sur la structure des groupes infinis de transformations; Ann. Norm. (3), t. XXI (1904), pp. 153-
206; t. XXII (1905), pp. 219-308.
2
( ) This is the D operation of Goursat.
 CHAPTER I

HAMILTONS LEAST-ACTION PRINCIPLE

AND THE QUANTITY OF MOTION-ENERGY TENSOR

I. - Case of the free material point.

1. One may base all of analytical mechanics on a principle that reduces the determination of
the motion of a material system to the solution of a problem in the calculus of variations; this
principle is Hamiltons least-action principle. We shall first discuss it in the case of a free
material point that is subject to a force that derives from a force function U that is a given
function of the rectangular coordinates x, y, z of the point and time t.
In this simple case, Hamiltons least-action principle is stated as:

Amongst all of the possible motions that take a material point with a given position (x0, y0, z0)
at the instant t0 to another position (x1, y1, z1) at the instant t1 the true motion is the one that
minimizes the definite integral:
t1 1
W = m( x2 + y2 + z 2 ) + U dt .
t0 2

In this expression, m denotes the mass of the point and x , y , z denote the components of its
velocity. The quantity under the integral sign is called the elementary action, and the integral W
is the action over the interval of time (t0, t1).
In order to prove this principle, one must regard x, y, z as functions of t and an arbitrary
parameter , and calculate the variation of W when one gives an increase , while supposing
that x, y, z reduce to x0, y0, z0 for t = t0, and to x1, y1, z1 for t = t1, and that this is true for any .
One has:

t1 U U U
W = m( x x + y y + z z ) + x+ y+ z dt .
t0
x y z

Now, one has:

x x ( x)
x = = = ,
t t t

so an integration by parts then gives, if one remarks that x, y, z vanish at the limits:

t1
U d2x U d2y U d2z
W = m x + m y + m z dt .
t0 x dt 2 y dt 2 z dt 2
2 Lessons on integral invariants

If one makes W zero for = 0 and all of the functions x, y, z of t zero at the limits then,
by a classical argument, it is necessary and sufficient that one have for = 0:

d 2 x U
m 2 = ,
dt x
d 2 y U
(3) m 2 = ,
dt y
d 2 z U
m = .
dt 2 z

It then results from this that the motions that the material point makes under the action of the
given force realize the extremum for the integral W with respect to all of the infinitely close
possible motions that correspond to the same initial and final positions of the point, and,
moreover, these motions are the only ones that enjoy this property.
To be rigorous, one may speak only of the extremum of the action and not of the minimum,
because the condition that the first variation W vanish is a necessary, but not sufficient,
condition for a minimum.

2. The elementary action:

12 m( x2 + y2 + z2 ) + U dt

seems to have been introduced here in a purely artificial way in order to state the laws of motion
in a condensed form. We shall see that one may replace Hamiltons principle with another
principle that is equivalent to it, in which a linear expression in dx, dy, dz, dt also appears, but
where all of the coefficients have a simple mechanical significance.
Indeed, continue to use the same action W, but now suppose that t0 and t1 are themselves
functions of the parameter , while the corresponding values x0, y0, z0, x1, y1, z1 are also
functions of . If one applies the preceding derivation to the definite integral then the
calculation of W would give:

W = 12 m( x2 + y2 + z2 ) + U t =t t1 12 m( x2 + y2 + z 2 ) + U t =t t0
1 0

+ [mx x + my y + mz z ]t =t1 [mx x + my y + mz z ]t =t0

t1 U d 2x U d2y U d2z
+ m 2 x + m 2 y + m 2 z dt .
x dt y dt z dt
t0

We now remark that one has:

 Hamiltons least-action principle 3

x x x
[ x]t =t1 = , x1 = t1 + ,
t =t1 t t =t1 t =t1
and, as a result:
[ x]t =t1 = x1 x1 t1 .
The formula that gives W is thus:

W = mx1 ( xi xi t1 ) + my1( y1 y1 t1 ) + mz1 ( zi z1 t1 )

+ 12 m( x12 + y12 + z12 ) + U1 t1
{mx0 ( x0 x0 t0 ) + my0 ( y0 y0 t0 ) + mz0 ( z0 z0 t0 )

(4)
+ 12 m( x0 2 + y0 2 + z0 2 ) + U 0 t0 }

t1 U d2x U d2y U d 2z
+ m 2 x + m 2 y + m 2 z dt.

x dt y dt z dt
t0

Set:
= mx( x x t ) + my( y y t ) + mz ( z z t )


(5) + 12 m( x2 + y 2 + z 2 ) + U t

= mx x + my y + mz z 12 m( x + y + z ) U t.
2 2 2

The differential expression that is thus introduced has for its components, first:

mx, my, mz ,

i.e., the components of the quantity of motion of the moving point, and then:

1
2
m ( x2 + y 2 + z 2 ) U ,
i.e., the energy E.
Thanks to this notation, one may write:

t1 U d2x U d2y U d2z

W = [ ]10 + m x + m y + m z dt .
x dt 2 y dt 2 z dt 2
t 0

Now suppose that one considers a collection of real trajectories that depend on a parameter
, and that one limits each trajectory to an interval of time (t0, t1) that varies with . The formula
that gives the variation of the action along these variable trajectories reduces to

W = ()1 ()0 .
4 Lessons on integral invariants

Finally, suppose that we consider a tube of trajectories, i.e., a closed continuous linear
collection of trajectories, each of which is limited to a time interval (t0, t1). When one returns to
the initial trajectory the total variation of the action is obviously zero, in such a way that, if one
integrates with respect to then one will have:
()1 = ()0.

3. In order to interpret the result we obtained, we agree to call the seven quantities:

x, y, z, x , y , z , t

the state of the material point, in which the first three quantities define the position of the point,
the next three define its velocity, and the last one defines the instant when the point is
considered. One may regard a state as a point in a seven-dimensional space: the state space. A
trajectory may be defined as the collection of all states that correspond to exactly one real
motion of the point, i.e., in summation, it is a solution of the system of differential equations:

dx dx U
dt = x, m
dt
=
x
,

dy dy U
(6) = y, m = ,
dt dt y
dz dz U
= z, m = .
dt dt z

From this, when the curvilinear integral:

= mx dx + my y + mz z E t

is taken along an arbitrary closed curve in the state space it does not vary if one displaces each
of the states that comprise it along the trajectory that corresponds to that state in an arbitrary
manner.
One may then say that, given an arbitrary tube of trajectories, if the integral is taken
along a closed curve that makes a circuit of the tube then that integral will be independent of
that curve and will depend only upon the tube.
One can remark that the expression may be regarded as the elementary work done on a
vector in a universe of four dimensions (x, y, z, t). This vector will have the three ordinary
components of the quantity of motion for its spatial components and energy for the component
that corresponds to time.

4. If one considers a simultaneous collection of states i.e., if one supposes that t = 0

then the integral will reduce to:
 Hamiltons least-action principle 5

mx dx + my y + mz z.

If we assume the latter viewpoint then we will obtain the following theorem:
If one considers a closed collection of trajectories, and if one takes the states that correspond
to an arbitrary fixed instant t then if the integral mx dx + my y + mz z is taken over the
closed collection of states thus obtained then that integral will be independent of t.

This theorem is due to H. Poincar, who characterized the property thus obtained by giving
the name of integral invariant to the integral:

mx dx + my y + mz z

when it is taken over a closed contour.

In Poincars way of thinking, the notion of energy is not involved. It necessarily appears if,
instead of considering a closed collection of simultaneous states, one considers a closed
collection of arbitrary states.
We say that the integral of the quantity of motion-energy is a complete integral
invariant or, more simply, an integral invariant, when there is no danger of confusion for the
differential equations of motion. Poincars integral invariant is thus the complete integral
invariant of the quantity of motion-energy when it is viewed in a particular light.
It is remarkable that if, instead of considering a collection of simultaneous states, one
considers a collection of states that satisfy the relations:

x = x t, y = y t, z = z t

then the tensor will reduce to Hamiltons elementary action:

12 m( x2 + y 2 + z2 ) + U t .

As a result, H. Poincars integral invariant and Hamiltons action are different aspects of
the integral of the quantity of motion-energy, although on first glance there is no relation
between these two notions.

5. In the preceding section, we simply deduced a property of the quantity of motion-

energy tensor from Hamiltons principle, namely, that the integral of this tensor along a closed
curve of states does not change when one deforms this closed curve without changing the
trajectories over which it is taken. We shall now show that this property may replace Hamiltons
principle, i.e., the differential equations of motion are the ones that admit the integral as an
integral invariant when it is taken over an arbitrary closed contour.
Let:
dx dy dz dx dy dz dt
= = = = = =
X Y Z X Y Z T
6 Lessons on integral invariants

be an arbitrary system of differential equations whose denominators are particular functions of

the seven variables x, y, z, x , y , z , t. Imagine a tube of integral curves of this system that
depend upon one parameter . This parameter will vary, for example, from 0 to l, such that the
integral curve that corresponds to = l coincides with the one that corresponds to = 0. In
order to express the idea that when the integral is taken over a closed curve that makes a
circuit around this tube it does not depend on the chosen closed curve, we imagine that the
coordinates x, y, z, x , y , z , t of an arbitrary state of the tube are functions of the parameter and
another parameter u. If one gives a fixed value to u then one will have a closed curve that makes
a circuit of this tube. If one displaces that closed curve along an integral curve of the tube then
one will have:
dx dy dt
du = = = ... = ,
X Y T

in which denotes an arbitrary factor that one may always choose in such a manner as to obtain
an arbitrary succession, given in advance, of closed contours that go around the tube for u =
const.
Having said this, when one gives a definite value to u the integral I = as a function of
(C )

u and, if one reserves the notation d for a displacement that makes only u vary, then one will
have:

dI = mdx x + m dy y + m dz z dE t + m x d(x) + m y d(y) + m z d(z) E d(t),

(C )

or, upon changing the order of differentiations of d and and integrating by parts:

dI = [ mx dx + my dy + mz dz E dt]C
+ ( mdx x + mdy y + mdz z dE t
(C )

mdx x mdy y mdz z + dt E).

The total integral part is obviously zero since the integration contour is closed. As for the
integral that remains in the right-hand side, in order for to be an integral invariant for the
differential system considered, it is necessary and sufficient that the integral must vanish when
one replaces:
dx, dy, dz, dx , dy , dz , dt
with:
X, Y, Z, X , Y , Z , T,

respectively, and that this must be true for any closed contour (C) and any function . One easily
deduces from this that the coefficients of:

dx, dy, dz, dx , dy , dz , dt

 Hamiltons least-action principle 7

become identically zero. As a result, in order for a system of differential equations to admit the
integral invariant , it is necessary and sufficient that the equations:

E
mdx + x dt = 0,
U
mdy + E dt = 0, mdx x dt = 0,
y
mdy U dt = 0,
mdz + E dt = 0, y
z
mdz U dt = 0,
E
(6) mdx + dt = 0, or z
x mdx + mxdt = 0,
E
mdy + y dt = 0, mdy + my dt = 0,
mdz + mz dt = 0,
E
mdz + z dt = 0, U
m( xdx + y dy + z dz ) + dU t dt = 0
dE + E dt = 0,
t

be consequences of the differential equations of the system.

The first six of these equations are nothing but the classical differential equations of motion.
As for the seventh one, it has the vis viva theorem as a consequence.

6. In the preceding, one sees the fundamental role that is played by the quantity of motion-
energy tensor. If one assumes that a trajectory is defined to be a succession of states that
constitute a solution for a system of ordinary differential equations then this system will be
characterized by the property that, among all of the imaginable systems of differential equations,
it will admit the curvilinear integral of the quantity of motion-energy tensor, when taken over
an arbitrary closed contour of states, as an integral invariant.
One thus obtains a new principle that may be called the principle of the conservation of the
quantity of motion and energy.

II. - General case.

7. All of the foregoing may be extended to the material systems that one habitually considers
in analytical mechanics. We suppose that these systems satisfy three conditions:

1. The constraints to which they are subject are perfect, i.e., at each instant t, the sum of the
elementary works done by the constraint forces is zero for any virtual displacement that is
8 Lessons on integral invariants

compatible with the constraints that exist at that the instant t. Under these conditions,
dAlemberts principle is valid and may be stated in the form:

dALEMBERTS PRINCIPLE. If one considers the motion of a material system that is

subject to perfect constraints under the action of given forces then at each instant the sum of the
elementary works done by the given forces and the forces of inertia is zero for any virtual
displacement of the system that is compatible with the constraints that exist at the instant t.

dAlemberts principle translates into the formula:

d2x d2y d 2z
(7) X m dt 2 x + Y m dt 2 y + Z m dt 2 z = 0 ,

in which X, Y, Z denote the components of the given force that is applied to the point (x, y, z) of
mass m and x, y, z denote the components of the most general elementary displacement that is
compatible with the constraints.
Amongst all of the systems with perfect constraints, we now consider the ones that have
holonomic constraints, i.e.:

2. We suppose that the constraints can be translated into a finite number of equations
between the coordinates of the points of the system and time t. Again, this amounts to saying that
it is possible to express the coordinates of the different points of the system by formulas such as

xi = fi(q1, ..., qn , t),

yi = gi(q1, ..., qn , t), (i =1, 2, ...)
zi = hi(q1, ..., qn , t),

with n arbitrary parameters q. To each system of values of q and t there corresponds one and
only one position of the system that is compatible with the constraints that exist at the instant t.
Any virtual displacement that is compatible with the constraints that exist at the instant t may be
obtained by giving arbitrary increments q1, ..., qn to q1, ..., qn .
We make one last hypothesis:

3. For any arbitrary virtual displacement that is compatible with the constraints that exist at
the instant t, the sum of the elementary works that are done by the given forces is the total
differential of a certain function U of the q and t, i.e.:

U U
( X x + Y y + Z z ) = q q1 + ... +
qn
qn .
1

U
The term t does not appear in the right-hand side because the virtual displacements to
t
which dAlemberts principle refer suppose that t remains constant.
 Hamiltons least-action principle 9

8. Hamiltons least-action principle may be extended without difficulty to the preceding

systems.
Set:
t1

W = 12 m( x2 + y2 + z2 ) + U dt .
t0

Regard the parameters q1, ..., qn , as functions of t and a parameter ; while the lower and
upper limits of the integral may depend on . A calculation that is identical to the one that was
given above (sec. 2) gives us the variation W of the action when one gives a variation to .
One obtains:
t1
d2x d2y d 2 z
(8) W = [ ]1 [ ]0 + U m 2 x + 2 y + 2 z dt ,
t0 dt dt dt
if one sets:

= m( x x + y y + z z ) 12 m( x2 + y2 + z 2 ) U t ,


[ ]1 = m( x1 x1 + y1 y1 + z1 z1 ) 2 m( x1 + y1 + z1 ) U1 t1 ,
' ' ' 1 '2 '2 '2
(9)

[ ]0 = m( x0 x0 + y0 y0 + z0 z0 ) 21 m( x0 + y0 + z0 ) U 0 t0 .
' ' ' '2 '2 '2

Having said this, dAlemberts principle immediately shows that when one is given a real
motion of the system, and one considers this motion in an arbitrary interval of time (t0, t1) then
this motion will realize an extremum of the action W with respect to all of the imaginable
infinitely close motions that correspond to the same initial position and the same final position of
the system. Conversely, the only motions that enjoy this property are the real motions of the
system. This is Hamiltons least-action principle.
Formula (8) shows, moreover, that when the integral is taken over a closed contour of
states of the system (that is compatible with the constraints), it does not change if one deforms
this closed contour by displacing each of the states that constitute it in an arbitrary manner along
the corresponding system trajectory. In other words, the integral is an integral invariant for
the differential equations of motion.
If one supposes that one considers only states of the system that are compatible with the
constraints then the differential form may again be called the quantity of motion-energy
tensor of the system.

9. The differentials x, y, z, t that enter into the expression are not arbitrary, in
general, because they must satisfy the equations that are obtained by totally differentiating the
constraint equations of the system. One may also express them in terms of

q1 , q2 , ..., qn , t

if one has introduced the n position parameters of the system. We shall take this viewpoint and
determine, on the one hand, the differential equations of motion, and, on the other, the quantity
10 Lessons on integral invariants

of motion-energy tensor. It will suffice for us to calculate W while supposing that the
elementary action is expressed by means of the parameters of q and time t. Set:

T = 12 m( x2 + y2 + z 2 ).
dqi
T, the kinetic energy, is a function of the second degree in the derivatives , which we write as
dt
qi' , and which we regard as independent arguments of qi and t. We temporarily set:

t1

F = T + U, W = Fdt.
t0

A simple calculation gives:

F F
q q + q q dt .
t1
W = F1 t1 F0 t0 + i '
'
i

t0
i i

However:
qi
qi' dt = dt = ( qi )dt = d ( qi ) ,
t t

so one has, after integrating by parts:

t
F
1 t1
F d F
W = F1 t1 F0 t0 + ' qi' + ' dt.
qi t0 t0 qi dt qi

In conclusion, we remark that one has:

qi (t0 , ) q (t , )
[ qi ]t0 = qi (t0 , ) and (qi0 ) = t0 + i 0 ,
t
hence:
[ qi ]t0 = (qi0 ) qi t0 and [ qi ]t1 = (qi (1) ) qi(1) t .

Finally, one has

F F
W = (qi ) qi
(1)
F t1
qi 1 qi 1

F F
(10) (qi ) qi
0
F t0
qi 0 qi 0

t1 F d F
+ dt.

qi dt qi
t0

 Hamiltons least-action principle 11

Hamiltons principle then leads us to the following equations of motion, which are nothing
but the Lagrange equations:

d T T U
(11) =0 (i = 1, 2, ..., n).
dt qi qi qi

Comparing the two expressions (8) and (10) that we found for W leads to the following
expression for the tensor :
T
(12) = qi H t ,
qi
in which we set:
T
(13) H = qi T U .
qi

T
The quantities are the quantities of generalized motion (with respect to the chosen
qi
system of coordinates); the quantity H is the generalized energy.

10. A simple remark permits us to simplify the calculation of the generalized energy H in
practice. In general, the kinetic energy T may contain terms of second degree, first degree, and
zero degree in q1 , q2 , ..., qn , namely:
T = T2 + T1 + T0 .

An application of Eulers formula concerning homogeneous functions then immediately gives:

H = T2 T0 U.

In the generalized energy, the term T2 may be regarded as of kinetic origin since the term T0
U is of dynamic origin.
For example, take the case of a free material point whose axes rotate around Oz with angular
velocity r. One has:
2T = m [( x ry)2 + ( y + rx)2 + z 2 ],

and, as a result, the energy, when referred to the chosen reference system, is:

H = 12 m ( x2 + y 2 + z 2 ) 12 mr2 (x2 + y2) U.

The part of the energy that is dynamical in origin may be decomposed into two terms, one of
which provides the given forces and the other provides the centrifugal forces. As for the
components of the quantity of motion, they are:

m( x ry), m( y + rx), m z ,
12 Lessons on integral invariants

i.e., they are the projections of the quantity of absolute motion onto the chosen coordinate axes

11. Hamiltons canonical variables. When the equations of motion are regarded as first-
order differential equations in qi, qi , t they take an extremely simple form if one introduces the
variables:
T
(14) pi = .
qi

The new variables, which one substitutes for qi , are all simply the components of the
quantity of motion of the system. The tensor then takes the simple form:

(15) = pi qi H t,

in which H may be regarded as a function of the qi , the pi , and t.

We shall search for the equations of motion directly after expressing the idea that they must
admit the integral as an integral invariant when it is taken over an arbitrary closed curve
of states of the system.
Let:
dq1 dq2 dp dt
(16) = = ... = n =
Q1 Q2 Pn T

be an arbitrary system of differential equations. In order to express the idea that it admits the
integral invariant , we need only to repeat the argument of sec. 5, word for word. We
consider a tube of integral curves of system (16) and express the 2n + 1 coordinates pi , qi , t of a
state of the tube as functions of the two parameters and u, the first of which is constant on an
integral curve and varies in the interval (0, l) in such a manner that the integral curve that is
defined by = l coincides with the integral curve that is defined by = 0. We let d denote the
symbol for differentiation with respect to the variable u and set:

I =
(C )

We then have, by an immediate integration by parts:

dI = (dpi qi dqi pi) dH t + dt H .

(C )

In order for the system (16) to admit the integral invariant , it is necessary and sufficient
that the coefficients of:
q1, q2, ..., qn , t

in the quantity under the sign vanish everywhere, if one takes the equations of the system into
account. Now, if one annuls these coefficients then one will obtain the 2n + 1 equations:
 Hamiltons least-action principle 13

H
dpi + q dt = 0,
i

H
(17) dqi + dt = 0,
pi
H
dH + dt = 0.
t

This shows that there is only one system of differential equations that admits the integral
invariant , and that would be the system that gives us both of the equations of motion in the
canonical form of Hamilton:

dqi H
= ,
dt pi
(18)
dpi H
= .
dt qi
The last equation:
H
dH dt = 0
t

is the analytical translation of the vis viva theorem; it is a consequence of the first 2n equations.

12. In the general case of the material systems in analytical mechanics, we thus arrive at the
generalized principle of the conservation of the quantity of motion and energy:

If one assumes that any motion of a system that is subject to given forces is a continuous
succession of states that satisfy a system of first-order differential equations then these
differential equations will be characterized by the property that they admit the integral of the
quantity of motion-energy tensor as an integral invariant when it is taken over an arbitrary
closed contour of states.

The quantity of motion-energy tensor can arbitrarily take any one of the forms:

= m( x x + y y + z z ) 12 m( x2 + y2 + z2 ) U t ,
T T
= qi H t ( H = qi T U ),
qi' qi
= pi qi H t.

If one displaces oneself in the space of states in such a way as to satisfy the relations:

qi = qi dt
14 Lessons on integral invariants

then the expression will reduce to Hamiltons elementary action (T + U)t. On the contrary,
if one considers only a collection of simultaneous states (t = 0) then one will obtain the
expression:
pi qi
that constitutes the element under the sign in the integral invariant that was proposed by H.
Poincar.

13. The principle of conservation of the quantity of motion and energy permits us to form
equations of motion no matter how we choose the parameters q1, ..., qn , t that serve to localize
the system in space and time. In other words, it gives us the laws of mechanics, since that fact
rests implicitly upon Hamiltons principle in a form that is independent of any particular mode of
spacetime framing. This property becomes analytically obvious if, instead of introducing the
derivatives q1 , ..., qn of the spatial parameters with respect to the temporal parameter, one
introduces n + 1 quantities:
q1 , q2 ,..., qn , t ,

whose mutual relations are defined by the equalities:

q1 q2 qn t
= = ... = = .
q1' q2' qn' 1
If one sets:
F = t(T + U ) ,

the right-hand side of which is homogeneous of the first degree in q1 ,..., qn , t and is expressed in
terms of the qi , t , qi , t , then the quantity of motion-energy tensor will take the form:

F F F
= q1 + ... + qn + t .
q1 qn t

In the general theory of relativity, the motion of a point that is subject to gravitational forces
obeys the preceding principle. The function F will then be of the form:

F= a q q ,
ik i k

in which the four variables qi serve to localize the point in space and time.
 Hamiltons least-action principle 15

III. - Transformation of the canonical equations. Jacobis theorem.

14. An important application of the preceding considerations relates to the transformation of

the canonical equations and the method for integrating the dynamical equations that is due to
Jacobi.
When the integral is taken over a closed contour, it obviously does not change if one
adds an exact differential to . Conversely, if another linear differential form enjoys the
same property of giving the same integral as when one takes it over an arbitrary closed
contour then will differ from only by an exact differential.
Suppose then that one must find 2n new variables ri , si , and a function K such that the two
expressions:
= pi qi Ht,
= ri si Kt

differ only by an exact differential. The differential equations of motion may be characterized by
the property that they admit the integral invariant , and, as a result, they may be written:

dsi K dri K
= , = ,
dt ri dt si

so the canonical form of these equations is preserved.

The hypothesis may be translated by an identity into the form:

(19) pi qi ri si (H K)t = V.

It is easy to realize such an identity. Indeed, start with an arbitrary function V of 2n + 1

arguments qi , si , t, and set:

V V V
(20) pi = , ri = , K= +H .
qi si t

If these equations define a change of variables i.e., if the first n are solvable with respect to
the s1, s2, ..., sn then the following n will give r1, ..., rn , the latter will give the function K, and
the new variables that are thus obtained will preserve the canonical form of the dynamical
equations. It is important to remark that if equations (20) are solvable in terms of the ri and the si
then, conversely, they will be solvable in terms of the pi and qi . In both cases, the possibility
condition is that the determinant:
2V
qi s j
must not be identically zero.
16 Lessons on integral invariants

The solution thus obtained from identity (19) is not the most general solution. It leaves aside
the case in which the 2n + 1 quantities qi , pi , t are related by one or more relations. However,
this singular case is easy to treat directly by giving the relations that exist between qi , pi , and t a
priori.

15. The applications of the preceding general theory become particularly interesting in two
cases.
The first one is the case in which the function K is identically zero. The canonical equations
become:
dsi dri
=0, =0.
dt dt

The equations of the trajectories reduce to:

si = ai , ri = bi ,

in which ai and bi are 2n arbitrary constants. From (20), in order for this to be the case, it is
sufficient to find a function V(t; q1, ..., qn ; a1 , ..., an) that satisfies the partial differential
equations:
V V
(21) + H t , qi , =0.
t qi

If this function V, into which n arbitrary constants a1, ..., an enter, is such that the
determinant:
2V
,
qi a j

is not identically zero then the equations of motion will be:

V V
pi = , bi = ;
qi ai

this is Jacobis theorem. The condition on the determinant amounts to saying that the function V
is a complete integral of Jacobis first-order partial differential equation (21).
The second application appears in the context of the theory of perturbations. Suppose that
the function H is the sum of two terms H1 and H2, the second of which is very small with respect
to the first one. This amounts to dividing the given forces into two groups, one of which is of
little importance with respect to the other and is composed of what one calls perturbing forces.
The method that is employed in celestial mechanics in this case consists of searching for a
complete integral V of the Jacobi equation:
V
+ H1 = 0
t
 Hamiltons least-action principle 17

that involves only the principal term of the function H. The 2n new variables ri , si thus
introduced will be constant if the perturbing forces do not exist. They are thus the unperturbed
trajectory parameters. The introduction of these new variables will preserve the canonical form
of the equations with the new function K = H2 , i.e., the part of H that relates only to the
perturbing forces.
Nonetheless, we do not insist at least for the moment upon the canonical equations and
Jacobis theorem. In particular, the relation that exists between the integration of the dynamical
equations and the integration of a first-order partial differential equation that does not explicitly
contain the unknown function will be clarified some other time when we have shown that one
may associate a linear integral invariant to any partial differential equation of that type or,
more generally, to any first-order partial differential equation that admits a known infinitesimal
transformation.
 CHAPTER II

THE TWO-DIMENSIONAL INTEGRAL INVARIANTS OF DYNAMICS

I. - Formation of the two-dimensional integral invariant in dynamics.

16. We have seen that Hamiltons elementary action can be obtained by supposing that in
the expression:
= pi qi H t
one has
qi = qi dt.

It is remarkable that the trajectories of a material system will again realize an extremum for
the integral:
t1
W = pi qi H dt
t0

if we simply suppose that the qi and qi are arbitrary functions of t that are subject only to the
condition that the qi take given values at the limits. As in Hamiltons principle, one thus
supposes only that the qi are the derivatives of the qi with respect to time. Similarly, one may
suppose, more generally, that the qi and qi are functions of the same parameter u, which varies
from 0 to 1, in such a way that the quantities qi and t take given values at the limits.
An easy calculation gives:

( ( p dq q dp ) H dt + t dH )
u =1 u =1
W = pi qi H t u=0 + i i i i
u =1

The total integral part is zero, by hypothesis. The extremal equations are thus obtained by
annulling the coefficients of:
q1, q2, ..., pn , t

in the quantity under the summation sign, but this calculation was done in sec. 11, and we
correctly gave the equations of motion in their canonical form.

17. The expression:

(dpi qi dqi pi) dH t + dt H,

that we have encountered twice is linear with respect to both types of differentials. One may
write it in the simpler form:
 Two-dimensional integral invariants 19

d d

if one assumes that the two differentiation symbols are interchangeable. The expression that we
denote by (d, ) enjoys the property of being zero whenever the symbol d defines an
elementary displacement in the direction of a trajectory in the state space, and the symbol
defines an arbitrary elementary displacement. Moreover, upon expressing that property, we
obtain relations between dq1, dq2, ..., dpn , dt that define the differential equations of the
trajectories, or from another point of view, the differential equations that admit the integral
invariant .
More generally, we now consider two arbitrary elementary displacements that are defined by
the differentiation symbols and , and we propose to look for the significance of the bilinear
form ( , ) . In order to do this, imagine a continuous two-dimensional set of states. One will
realize such a set by taking functions of two parameters and for the qi , pi , and t. Each state
of the set can be represented in the plane by a point with coordinates (, ), and the set will be
represented by an area in the plane. The symbols and refer to increments in alone and
alone, respectively. Therefore, consider four states A, B, C, D in the state space that correspond
to the parameter values:

, ,
+ , ,
, + ,
+ , + ,

respectively, and form the integral , which is taken over the closed contour ABCD. One
obviously has:

= , = , = + , = + ,
AB AC CD AC

and, as a result
= = ( , ) .
ABCD

18. The bilinear form ( , ) that appears when summing an arbitrary state and two
infinitely close states, and represents, from the foregoing, the value of the integral when it is
taken over a closed contour, is an invariant of the system of differential equations of the
corresponding trajectories, in the sense that it will not change value if one displaces each of the
states along the corresponding trajectory. This form is also a double integral element.
Moreover, if one regards, for example, p1 and q1 as the coordinates (which depend on two
parameters and ) of a point in the plane then the expression p1 q1 q1 p1 will be the
area element of this plane with respect to the curvilinear coordinates and , which one
habitually writes as:
dp1 dq1 or p1 q1 .
20 Lessons on integral invariants

This leads us to the notion of a new integral invariant:

(1) = pi qi H t.
This double integral, when taken over a two-dimensional area in the state space, is
reproduced if one displaces each of the states of that area along the corresponding trajectory.
This double integral is then obtained by the generalized Stokes formula as the expression of the
curvilinear integral:
= ( pi qi H t ) ,
when it is taken over the closed contour that bounds the area.
In Poincars way of thinking, one considers only areas that are composed of simultaneous
states. One may then state the result obtained in the following form:

Given a two-dimensional set of trajectories, if one takes the set of states on each trajectory
that correspond to a given instant t then the double integral:

pi qi
will be independent of t when it is taken over these states.

As one sees, this theorem expresses a particular aspect of the property that was proved above.

19. Poincar calls the two-dimensional integral invariant absolute, as opposed to the
invariant , which he calls relative. This signifies that the double integral possesses an
invariant character for any domain of integration whether open or closed, whereas the
integral possesses an invariant character only if it is taken over a closed contour.
Since the integral is nothing but the integral when it is taken over a closed contour,
one may confirm that the differential equations of motion are the only ones that admit the
integral invariant . The invariance of the integral is the analytic translation of a new
generalized principle in the form of the conservation of the quantity of motion and energy.

II. - Applications to vortex theory.

20. Up until now, we have considered only sets of trajectories that are realized in our
imagination. There is one case where such sets have a concrete existence. It is the case of a
perfect fluid that is subject to forces that are derived from a force function U. In hydrodynamics,
one proves the following equations:
 Two-dimensional integral invariants 21

U 1 p
x = x x ,

U 1 p
(2) y = ,
y y
U 1 p
z = ,
z z

in which x , y , z denote the components of the acceleration of the molecule that occupies the
position (x, y, z), and p and denote the pressure and density at that point, respectively.
We add to these hypotheses that there is a relation between p and that is given in advance,
which is certainly true if the motion is isothermal
If we direct our attention to a given motion of the fluid then we may regard p as a given
function of x, y, z, t, and by setting:
dp
q=

we will see that each molecule moves like a material point of mass 1 that has been placed in a
force field that is derived from a force function U q.
We thus have a concrete realization of an infinitude of trajectories of a moving point subject
to given forces. We remark that the q part of the force function represents the action that is
exerted on the molecule by the molecules in its environment.

21. The trajectory of each molecule can be regarded as a particular solution of the system of
differential equations:
dx du (U q)
= u, = ,
dt dt x
dy dv (U q)
(3) = v, = ,
dt dt y
dz dw (U q)
=w = .
dt dt z

Thus, if one considers a closed collection of molecules in the fluid (each taken at an arbitrary
instant) then the integral:
(4) u x + v y + w z E t

will not change in value if one displaces each molecule along its trajectory when it is taken over
that closed collection. In this expression, one has set:

(5) E= 1
2 (u2 + v2 + w2) U + q.
22 Lessons on integral invariants

E is the energy (per unit mass) of the fluid. This energy is the sum of the kinetic energy 12 (u2
+ v2 + w2), the potential energy U, and the internal hydrodynamic energy q.
In particular, if one considers a closed collection of molecules, all of which are considered at
the same instant t i.e., a closed streamline then the integral u x + v y + w z will keep the
same value if one takes the same fluid line (i.e., the streamline that is composed of the same
molecules) at different instants of its motion. This is the classical theorem of the conservation of
circulation. One gives the name of circulation to the integral u x + v y + w z.

22. Now let us take a different point of view. Let us always consider a particular motion
of the fluid mass. In this motion, the components u, v, w of the velocity are given functions of x,
y, z, t, and the trajectories of the different molecules may be regarded as solutions of the system
of differential equations:
dx
dt = u,

dy
(6) = v,
dt
dz
dt = w,

on the right-hand side of which u, v, w are assumed to be replaced by their values as functions of
x, y, z, t. The integral:
u x + v y + w z E t

is again obviously a relative integral invariant for these new differential equations. Upon
transforming it into a double integral, we obtain an absolute integral for the system (6).
If we form the expression then we will obtain:

( , ) = u x x u + v y y v + w z z w E t + t E .

The right-hand side is linear with respect to the 6 combinations:

y z z y , z x x z , x y y x ,
x t t x , y t t y , z t t z .

A simple calculation, which is nothing but an application of Stokess formula, gives the
following coefficients for the first three terms:

w v u w v u
= , = , = ;
y z z x x y

these are the components of the vorticity vector. In order to calculate the other three coefficients,
we can use the fact that since the expression is invariant for equations (6), the equations that
 Two-dimensional integral invariants 23

are obtained by annulling the coefficients of x, y, z, t in (d, ) can be consequences of

equations (6). Therefore, set:

(d , ) = (dy z dz y) + (dz x dx z) + (dx y dx y)

+ P (dx t dt x) + Q (dy t dt y) + R (dz t dt z).

The equations considered are:

dz dy Pdt = 0,
dx dz Qdt = 0,

(7)
dy dx Rdt = 0,
Pdx + Qdy + Rdz = 0.

If we say that they are a consequence of equations (6) then we will obtain:

P = wv
Q = u w,
R = v u.

As a result, the desired double integral invariant is:

y z + x y + x y + ( w v) x t + ( u w) y t + ( v u) z t.

When this integral is taken over a closed area of molecules that are all taken at the same instant t
we obtain the vorticity flux across that area. From this, we recover the theorem of the
conservation of vorticity flux through a fluid surface.

23. We will now calculate the expression (d , ) directly. In particular, the coefficient P of
dxt is obviously:
u E u u v w U 1 p
P= = u v w + ,
t x t x x x x x

and if one sets it equal to the preceding value:

u w v u
P = w v = w v
z x x y
then one will obtain the equation:

u u u u U 1 p
+u +v +w = ,
t x y z x x
24 Lessons on integral invariants

which is nothing but the main equation of hydrodynamics. Indeed, the left-hand side is the
expression that we developed for x .
This result reminds us that the integral u x + v y + w z E t is invariant for the
differential equations (6) only if u, v, w are the components of the velocity of a molecule of a
perfect fluid that is subject to a force that is derived from a force function, or again, if there is an
acceleration potential.

24. Equations (7), which may also be written:

(dz wdt ) (dz vdt ) = 0,

(7) (dx udt ) (dz wdt ) = 0,
(dy vdt ) (dx udt ) = 0,

are a consequence of the differential equations (6), but they are not equivalent to those
differential equations. In other words, the equations (6) of the trajectories are not the only ones
that admit the integral invariant . In particular, this is also the case for the equations:

dx dy dz dt
(9) = = = ;
0

equations (7) are obviously a consequence of this. The solutions of these equations are what one
calls vortex lines. Imposing the requirement that the differential equations of the trajectories
and the differential equations of the vortex lines admit the same integral invariant leads us to the
fundamental theorem of vortex theory.
Indeed, one may characterize an elementary displacement (dx, dy, dz, 0) (in the four-
dimensional universe of x, y, z, t) in the direction of a vortex line by the property that the bilinear
form (d , ) is zero for any displacement ; this results immediately from equations (7).
Having said this, consider a vortex line () at an instant t. The molecules that comprise it form a
line ( ) at another instant t . We shall show that ( ) is a vortex line for the instant t . Indeed,
let ( dx , dy , dz , 0) be an elementary displacement along ( ) and associate it with an arbitrary
displacement ( x , y , z , t ) . If we displace the three states

( x , y , z , t ) , ( x + dx , y + dy , z + dz , t ) , ( x + x , y + y , z + z , t + t )

along their respective trajectories the first two, up to the instant t and the second two, up to the
instant t + t then we will obtain a two-dimensional element for which (dx, dy, dz, 0) represents
a displacement along the vortex line (). The form ( , ) will thus have the value zero. It is
therefore zero for the original element as well, and as a result ( ) is a vortex line. This is the
celebrated theorem of Helmholtz.
 Two-dimensional integral invariants 25

25. Consider a vortex tube at the instant t and two closed curves (C) and (C ) that encircle
the tube. The circulation around these two closed curves is the same since is an integral
invariant for the differential equations (9) of the vortex lines. At another instant t , the vortex
tube will take another position in space, but the circulation around any closed line that encircles
the new tube will not have changed since is an integral invariant for the differential
equations of the trajectories. We recover the notion of what one calls the moment or the intensity
of a vortex tube in hydrodynamics, a quantity that is conserved through the duration of the
motion. This property is only a particular aspect of the invariance of the integral:

u x + v y + w z E t,

for the differential equations of the trajectories and for those of the vortex lines.
Furthermore, we recover all of these results as a particular case of a general theorem that
concerns differential forms that are simultaneously invariant for several systems of differential
equations.
It is trivial to remark that in all of the preceding we have essentially supposed that not all
three of , , were zero; i.e., that the motion of the fluids was rotational.
 CHAPTER III

INTEGRAL INVARIANTS AND INVARIANT DIFFERENTIAL FORMS

I. - General notion of an integral invariant.

26. The preceding chapters have shown us the importance of the notion of integral invariant
in mechanics. We shall now discuss this notion in its full generality.
Consider an arbitrary system of first-order ordinary differential equations (one knows that
one may always reduce to this case), which we write:

dx1
dt = X1 ,

dx2 = X2,

(1) dt
...

dxn = Xn.
dt

We have distinguished the independent variable t from the dependent variables x1, x2, ..., xn ,
but, as one can verify, that distinction is not essential, so we continue to say that t represents
time. The set of values of x1, ..., xn , t that correspond to a solution will be said to constitute a
trajectory, which we may regard as a curve in the n + 1-dimensional space of (x1, ..., xn , t).
Having said this, H. Poincar gave the name of integral invariant to an integral (simple or
multiple) that, when taken over an arbitrary set of simultaneous (i.e., ones that correspond to the
same value of t) points, does not change value when one displaces the points of that set along the
corresponding trajectories up to another arbitrary instant t . An integral invariant is called
absolute if its invariance property is true for any integration domain. It is called relative if the
invariance property is only true for a closed integration domain. The linear integral invariant of
mechanics:
pi qi
is relative; the double integral invariant of mechanics:

pi qi
is absolute.
The simplest forms for integral invariants are:

a x + a x + ... + a x ,
1 1 2 2 n n

a x + a x + ... + 2a x x + ... ,
2 2
11 1 22 2 12 1 2
 Integral invariants and invariant differential forms 27

a x x + a x x
12 1 2 12 1 3 + ... + an1, n xn1 xn ,

a x x x + ...
123 1 2 3

27. The quantity under the summation sign in an integral invariant is a differential form into
which the variables both dependent and independent and their differentials (or, similarly,
several sets of differentials) enter. That form F may be considered by itself, and it enjoys the
property that when it is calculated for an arbitrary point and one or more infinitely close but
simultaneous points it will not change values if one displaces these points along their respective
trajectories while keeping all of the points simultaneous. It is quite clear that from this viewpoint
one may consider more general forms F than the ones that are susceptible to appearing under an
integration sign; for example, an arbitrary rational (homogenous) function of x1, ..., xn .
As we showed in the examples that were treated in the first two chapters, there is no point in
restricting ourselves to the consideration of simultaneous points. We shall see that any
elementary integral invariant, in the sense of H. Poincar, may be regarded as resulting from
suppressing all of the terms that contain the differential or differentials of the independent
variable t in a more complete elementary integral invariant.
However, in order to arrive at this essential result, which will give us the key to almost all of
the properties of integral invariants, it is necessary to briefly recall the classical properties of first
integrals of a system of differential equations.

II. - First integrals.

28. As one knows, one calls a function u(x1, ..., xn , t) a first integral of system (1) if it enjoys
the property that if one replaces the x1, ..., xn with their values as functions of t along an
arbitrary trajectory then the function u of t thus obtained can be reduced to a constant. These
first integrals are solutions of the first-order linear partial differential equation:

u u u u
(2) + X1 + X2 + ... + X n = 0.
t x1 x2 xn

Imagine that one has integrated equations (1), and that one has expressed the dependent
variables x1, ..., xn as functions of time t and their initial values x10 , x20 , ..., xn0 for t = 0, namely
x x

x1 = f1 (t ; x10 ,..., xn0 ),

...
x = f (t ; x 0 ,..., x 0 ).
n n 1 n

When these equations are solved for x10 , ..., xn0 they will give functions of x1, ..., xn , t that are
obviously first integrals for these n quantities. One thus obtains a system of n first integrals that
are obviously independent; i.e., ones that are not related by any identity relation in x1, ..., xn , t.
28 Lessons on integral invariants

It is clear that any function of the first integrals x10 , ..., xn0 will be a first integral, and,
conversely, if u is an arbitrary first integral then its numerical value for an arbitrary trajectory
will be, by the same property, equal to u( x10 , ..., xn0 , 0).
The total differential of any function u of x1, x2, ..., xn , t may obviously be put into the form:

du = 1 (dx1 X1 dt) + 2 (dx2 X2 dt) + ... + n (dxn Xndt) + dt.

The necessary and sufficient condition for this to be a first integral is that the coefficient must
be identically zero. One may easily account for this by a direct argument. One may also verify
this by remarking that is nothing but the left-hand side of equation (2). Thus, the differential of
any first integral is a linear combination of n linear differential forms:

dx1 X1 dt, dx2 X2 dt, ..., dxn Xndt,

and, conversely, each of these forms is a linear combination of the differentials of n given
independent first integrals.

III. - Absolute integral invariants and invariant differential forms.

29. Having said this, we first occupy ourselves with the absolute integral invariants. The
element of any absolute integral invariant is a differential form F(x1, ..., xn, t; x1, ..., xn) that
does not change value if one displaces the point (x1, ..., xn , t) and the infinitely close point (x +
x1, ..., xn+ xn , t) along their respective trajectories while always considering them at the same
instant. In particular, consider them at the instant t = 0. We have:

F ( x1 ,..., xn , t ; x1 ,..., xn ) = F ( x10 ,..., xn0 , 0; x10 ,..., xn0 ) .

Now regard the xi0 in the right-hand side as first integrals of the system (1) and replace them
with their values as functions of x1, ..., xn , t. We obtain a new identity:

F ( x10 ,..., xn0 , 0; x10 ,..., xn0 ) = ( x1 ,..., xn , t ; x1 ,..., xn , t ) .

The left-hand side of this identity is obviously a quantity whose numerical value is of interest to
the trajectory that is defined by the initial values x10 ,..., xn0 and the infinitely close trajectory. Its
value is thus independent of the particular point (x1, ..., xn , t) that was taken from the first
trajectory and the particular point (x1+ x1, ..., xn + xn , t + t) that was taken from an infinitely
close trajectory. It is therefore also an element of an integral invariant, but of a more complete
integral invariant than the one that served as our point of departure, since now we are no longer
obliged to restrict ourselves to the consideration of simultaneous points.
 Integral invariants and invariant differential forms 29

We now remark that it is easy to pass from the initial form F to the final form . Indeed, if
one regards t as a constant in the calculation of x10 ,..., xn0 , x10 ,..., xn0 then one will obviously
revert to the form F. One thus has:

(x1, ..., xn , t; x1, ..., xn , 0) = F(x1, ..., xn , t; x1, ..., xn).

Now, t appears only by the intermediary of the x10 ,..., xn0 , and these n differentials are
linear combinations of
x1 X1 t, x2 X2 t, x3 X3 t.

As a result, depends only on these n linear combinations, and when one has its expression
for t = 0 one immediately has its expression for arbitrary t by replacing x1 with x1 X1 t,
etc.
Finally, one has:

(3) (x1, ..., xn , t; x1, ..., xn , t) = F(x1, ..., xn , t; x1 X1t, ..., xn Xn t).

30. We summarize the results that we just obtained; they are two in number.

1. The form F, which constitutes the element of an absolute integral invariant, in the sense of
H. Poincar, and in which only the differentials of the dependent variables appear, is associated
with a more complete form , in which the differential (or differentials) of the independent
variable t is also involved. One passes from the form to the form F by suppressing the terms
that contain t, and, conversely, one passes from the form F to the form by replacing:

x1, x2, ..., xn

with:
x1 X1 t, x2 X2 t, x3 X3 t.

2. The form may be expressed in terms of first integrals of the system (1) and their
differentials.
This latter property makes the invariant character of the form self-evident.
A simple example will make the relation between the two forms F and comprehensible. If
one starts with an arbitrary integral u then the total differential u will obviously be a form F.
The corresponding form is:
u u u
F= x1 + x2 + ... + xx ,
x1 x2 xn
and one has indeed:

u u u
= u = ( x1 X1 t ) + ( x2 X 2 t ) + ... + ( xn X n t ) .
x1 x2 xn
30 Lessons on integral invariants

31. We agree to say that a differential form that can be expressed in terms of first integrals of
the system (1) and their differentials is an invariant form for the system (1). The quantity under
the integration sign in an absolute integral invariant is obtained by annulling t in an invariant
form. This is the double integral invariant of mechanics that corresponds to the invariant form:

pi qi H t,
or, if one prefers, after introducing two types of differentials:

= (pi qi qi pi ) H t + t H .

Its expression in terms of the first integrals pi0 , qi0 is obviously:

= ( pi0 qi0 qi0 pi0 ).

IV. - Relative integral invariants. Hamiltons function.

32. One part of these results can be extended to the theory of relative integral invariants. It
is the fact that the linear integral invariant of dynamics that was considered by H. Poincar:

pi qi
does not change value when one displaces each state along its trajectory at the instant t up to an
arbitrary instant t and equals the integral:
pi qi .
0 0

Any relative integral invariant may thus be given an expression in terms of only first
integrals and their differentials, and, in that form, it may taken over any closed domain of
simultaneous or non-simultaneous states without losing its invariant character.

However, if one replaces the first integrals by their expressions as functions of the dependent
and independent variables in the new expression then one will obtain a form inside the
summation sign that is not derived from the initial form F by the same process as in the case of
an absolute invariant.
The equality:
F(x1, ..., xn, t; x1, ..., xn) = (x1, ..., xn , t; x1, ..., xn , 0)
holds for any closed integration domain that is formed from simultaneous points, but the term-
by-term equality of the two sums does not result from this, and one necessarily has only the
identity:
 Integral invariants and invariant differential forms 31

F(x1, ..., xn , t; x1, ..., xn) = (x1, ..., xn , t; x1, ..., xn , 0),

which will be necessary in order for one to deduce, conforming to the formula (3), that:

(x1, ..., xn , t; x1, ..., xn , t) = F(x1, ..., xn , t; x1 X1t, ..., xn Xn t).

This is true in the simple case of a free material point. The element:

F = m ( x x + y y + z z)

that appears inside the summation sign in the expression of the linear integral invariant of H.
Poincar leads to the form:

m( x x + y y + z z) m ( x 2 + y 2 + z 2 ) t,

which is nothing but a complete integral invariant, and differs from the form:

= m( x x + y y + z z) [ 12 m ( x 2 + y 2 + z 2 ) U]t

by only a simple exact differential, if necessary.

We should remark that the difficulty that presents itself here in passing from a relative
integral invariant, in the sense of H. Poincar, to a complete integral invariant does not have
great practical importance, because any relative integral invariant may be reduced to an absolute
integral invariant. Indeed, one knows that an integral taken over a closed contour, closed
surface, etc., may be reduced to an integral over an area bounded by the closed contour, a
volume bounded by the closed surface, etc.

33. It is useful to illustrate the preceding considerations with some examples.

Recall the (complete) linear integral invariant of dynamics, i.e., the quantity of motion-
energy tensor:
= pi qi Ht.
One has the equality:
pi qi H t = pi qi ,
0 0

(C ) ( C0 )

in which one supposes that the closed contour (C0) is composed of the states that constitute (C),
but are displaced along their trajectories up to the instant t = 0. One may furthermore consider
the integral of the right-hand side as being extended over the same contour (C) as the integral on
the left-hand side, on the condition that one regards the pi0 and the qi0 as functions of the pi , the
qi , and t. From that point of view, the two expressions:

p q H t
i i and p q 0
i
0
i
32 Lessons on integral invariants

give the same integral along an arbitrary closed contour, or one that differs only by an exact
form, and one has:
(4) pi qi H t = S + pi0 qi0 .
The function S is what one calls the Hamilton function, and it has a simple concrete
interpretation. If we refer to formula (10) of chapter I, which gives the variation of the action
along a variable trajectory, then we will see that S may be interpreted as representing the action
between the instant 0 and the instant 1 along the trajectory that leads up to the state (p, q, t).
The function S was considered by Hamilton, and it has a certain importance from a historical
perspective because it was the remarks that Hamilton made on that subject that led Jacobi along
the path of his discoveries that relate to the equations of mechanics. Indeed, Hamilton remarked
that if one knows how to express the function S, not as a function of the pi , qi , and t, but as a
function of the pi0 , qi0 , and t, then one will get the equations of motion by the same integration.
Indeed, when the identity (4) is put into the form:

S = pi qi H t pi0 qi0 ,
it will give:
S S S
(5) pi = , pi0 = , + H = 0.
qi qi0 t

The second set of equations will give the pi as functions of t and the 2n initial values. The
first one will give the quantities of motion pi . Finally, the last one shows that the function S is a
solution of the partial differential equation:

S S
(6) + H t , qi , = 0.
t qi

The difficulty with this way of looking at things is not merely in integrating that partial
differential equation, but in finding a solution for which the arbitrary constants qi0 agree
precisely with the initial values of the qi . Jacobi resolved this difficulty by showing that this
condition was totally unnecessary for the purpose of making the integration of the partial
differential equation (6) serve as the integration of the equations of motion; this is what we
briefly discussed already in sec. 14.

34. It is quite instructive to effect the calculation of the Hamilton function S in a simple case;
for example, the case of a free point of mass 1 that is not subject to any force. Here the
equations of motion are:
x = x0 t + x0 , x = x0 ,
y = y0 t + y0 , y = y0 ,
z = z0 t + z0 , z = z0 .
The difference:
 Integral invariants and invariant differential forms 33

S = ()0 = x x + y y + z z 1
2 ( x2 + y 2 + z 2 )t ( x0 x0 + y0 y0 + z0 z0)

is equal to:

S = x x + y y + z z 12 ( x2 + y 2 + z 2 )t x (x t x ) y (y t y ) z (z t z )
= 12 ( x2 + y 2 + z 2 )t + t ( x x + y y + z z ),

from which, upon taking into account that S must be annulled with t, we will get:

S= 1
2 ( x2 + y 2 + z 2 )t.

Upon expressing S by means of x, y, z, x0, y0, z0, t, one obtains:

( x x0 ) 2 + ( y y0 ) 2 + ( z z0 ) 2
1
S= 2 .
t

Hamiltons formulas (5) permit us to deduce the equations of motion from this function:

S x x0 S x x0
x = = , x0 = = ,
x t x0 t
S y y0 S y y0
y = = , y0 = = ,
y t y0 t
S z z0 S z z0
z = = z0 = = .
z t z0 t

V. - Examples. The element of matter form.

35. Now that we have made the preceding parenthetical remarks, we return to absolute
integral invariants.
In the simplest case, it is good to directly account for the invariant character of the
differential forms that are deduced, as was said above, from forces F by replacing xi with
xi Xit in them.
To simplify matters, take a system of two differential equations in two unknown functions:

dx dy
= X, = Y,
dt dt

and start with an absolute linear integral invariant:

I= a(x, y, z)x + b(x, y, t) y.

34 Lessons on integral invariants

The complete integral invariant that is associated with it is:

J= a(x, y, z)(x X t) + b(x, y, t) (y Yt).

Start with an arc of a curve A0 B0 in the xy-plane and guide the different points of that arc
along the corresponding trajectories. One thus obtains a type of cylindrical surface whose (non-
rectilinear) generators will be the trajectories. Trace two curved arcs MN and M N on the
surface that connect the trajectory that issues from A0 to the trajectory that issues from B0 . We
shall show that one has:

JMN = J M N .

The two curved arcs MN and M N , together

T with the trajectory arcs MM and NN , bound a
closed area of the surface. On the other hand,
M when the integral J is taken over each of these
P Q N latter two arcs the result is obviously zero since
M P upon displacing along one of these arcs one
Q constantly has:
x = X t, y = Y t.
O X
N As a result, when the integral J is taken over
Y the closed contour MNN M , one gets:
A0
B0 J MNN M = JMN J M N ,

and everything comes down to showing that this integral is zero. Now, from Stokess formula
this integral amounts to a surface integral that is taken over the area MNN M . We shall show
that this surface integral has a zero element. Indeed, in order to do this, we decompose the
surface into surface elements that are formed from small parallelograms that are defined, on the
one hand, by arcs of trajectories, and, on the other hand, by sections by the planes t = const. Let
PQQP be one of these surface elements. The element of the corresponding surface integral is
equal to
JPQ J PQ ,

but, since the points of PQ are simultaneous, as well as those of PQ , JPQ reduces to IPQ,
and J PQ to I PQ . Now, from the fact that I is an integral invariant, the two integrals IPQ
and I PQ are equal.
Hence, any integral element of the surface is precisely zero, and the theorem is proved.

36. An analogous argument will be made in the case of a double integral invariant:
 Integral invariants and invariant differential forms 35

I= a(x, y, t) xy .

Here, the passage from the form F to the form is a little more difficult than it was in the
preceding case.
One arrives at it by associating the surface element x y with a bilinear fom x y y
x . In order to do this, it suffices to imagine an indeterminate system of curvilinear coordinates
(, ), and regard x, y as the elementary displacement relative to an increment in the first
coordinate , and regard x , y as the elementary displacement relative to an increment
in the second one . One then has:
x y
F=a .
y y
From this, one deduces that:

x X t y Y t x y y t t x
= a =a + aX + aY ,
x X t y Y t x y y t t x

or, upon returning to the notations in the theory of surface integrals:

= a x y + aX y t + aY t x.

Therefore, consider a surface element:

J= axy + aX y t + aY t x,

and let us try to account for its invariant character

directly. In order to do this, imagine an arbitrary T
area S0 in the xy-plane and construct the
S
trajectories that issue from different points of that
area. We thus obtain an indefinite volume
bounded by a type of cylindrical lateral surface v
that is generated by the trajectories that begin on S
the contour of S0 . Cut this volume with two
arbitrary surfaces. We thus obtain two (plane or O X
curved) areas S and S that are in the interior of
the volume, but extend up to the lateral surface.
We would like to show that one has: S0
Y
JS = J S .

The areas S and S , along with a portion of the lateral surface of the cylinder, define a
volume V. On the other hand, when the integral J is taken over the area that bounds this volume,
the result is obviously zero, since, if we call the area element d and the direction cosines of the
normal , , then we will have:
36 Lessons on integral invariants

J = a( + X + Y) d,

and the fact that the direction (X, Y, 1) is normal to the direction (, , ), since it is the direction
of the tangents to the trajectories that generate the lateral surface considered. It results from this
that the difference J S JS can be regarded as the surface integral J taken over the closed area
that bounds the volume V. Everything amounts to showing that the volume integral is obviously
zero. In order to account for this, it suffices to take the elementary volume to be the volume that
is laterally bounded by small arcs of the trajectory and two plane areas that are parallel to the xy-
plane at the extremities, because then when the surface integral J is taken over each of the bases
it will reduce to the integral I, and, by hypothesis, the value of the integral I is the same for both
bases.

37. The kinematics of continuous media provides us with a concrete illustration of the
considerations that were developed in this chapter.
In a continuous medium in motion, the trajectory of each molecule can be regarded as a
solution to the system of differential equations:

dx dy dz
= u, = v, = w,
dt dt dt

in which the components of velocity u, v, w are assumed to be expressed as functions of x, y, z, t.

On the other hand, let (x, y, z, t) be the density at the instant t at the point (x, y, z). The mass
that occupies an arbitrary volume V at the instant t will be given by the triple integral:

x y z.
V

This integral obviously constitutes an absolute integral invariant, in the sense of H. Poincar.
It was the first example of an integral invariant that was given by H. Poincar. If the molecules
that occupy the volume V at the instant t occupy the volume V at another instant t then one will
obviously have:
(x, y, z, t) x y z = ( x, y, z, t ) x y z .
V V

The form that is associated with the form F = x y z can be calculated, as in the
preceding example, by writing F in the form:

x y z
F = x y z .
x y z
One deduces from this that:
 Integral invariants and invariant differential forms 37

x u t y v t z w t
= x u t y v t z w t ,
x u t y v t z w t

from which, by an easy calculation one obtains:

= (x y z u y z t v z x t w x y t).

The form represents the element of matter, when envisioned in its complete kinematical
aspect. If one considers an arbitrary three-dimensional set of molecules, and if one takes each
molecule of the set at an arbitrary instant t of its motion, then one will obtain a three-dimensional
domain in a four-dimensional universe (x, y, z, t). The triple integral of taken over this domain
will be equal to the total mass of the set of molecules considered. If the molecules are all taken
at the same instant t then they will occupy a certain volume V at that instant, and the integral of
will reduce to the integral x y z. However, this is entirely peculiar to the case at
V
hand.
Consider, to be specific, an area S in space, for example, and the set of all of molecules that
cross that area between an instant t0 and an instant t1. Take each of these molecules at the
moment when it traverses the area S. We will then get a three-dimensional domain in the
universe of (x, y, z, t). The states of this domain are easily expressed by means of the three
parameters , , . In order to do this, it suffices to express the coordinates of a point of S as
functions of the two parameters , , and to take t = . One then will have formulas such as:

x = f(, ),
y = g(, ),
z = h(, ),
t = ,

in which the parameters and take all of the values that correspond to the different points of
the area S and the parameter takes all of the possible values in the interval (t0, t1). When the
integral F is taken over this domain, the result will obviously be, ignoring the sign:

t u y z v z x + w x v .
t1
t0 (S )

The surface integral between brackets represents the matter flux that traverses the surface S at
the instant t. When it is multiplied by t, it will represent the quantity of matter that traverses the
surface S during the interval (t, t + t). The total integral thus represents what we have been
waiting for: viz., the total mass that crosses S in the interval (t0, t1).

38. Analogous remarks apply to the double integral invariant that we encountered in
hydrodynamics (Chap. II, formula (8)), viz.:
38 Lessons on integral invariants

J= y z + z x + x y + (w v) x t + ( u w) y t + ( v u) z t.

We saw (sec. 25) that when this integral is taken over a two-dimensional set of molecules
that are taken at the same instant t, the result will represent the moment or the intensity of the
vortex tube that is formed from the vortex lines that start at these molecules. Therefore, consider
the set of molecules that traverse an arc of the curve C in an interval of time (t0, t1). Instead of
taking these molecules at the same instant t, take each of them at the instant when it crosses the
arc of the curve C. At an arbitrary instant t, the moment of the vortex tube that originates from it
will be equal to the integral:
x y z
t1
t .
t0 (C )
u v w
 CHAPTER IV

THE CHARACTERISTIC SYSTEM OF A DIFFERENTIAL FORM

I. - The class of a differential form.

39. In all of this chapter, we will consider systems of differential equations in n variables x1,
x2, ..., xn without distinguishing the independent variable by a special notation; it will be any of
the variables x1, x2, ..., xn , arbitrarily. In other words, we will consider systems of differential
equations of the form:
dx1 dx2 dx
(1) = = ... = n .
X1 X 2 Xn

One of the first problems that presents itself in the theory of integral invariants is the
following one: to recognize whether a given differential form is invariant for a given system of
differential equations, and, more generally, to determine all of the systems of equations that
admit a given differential form as an invariant form.
Before commencing the solution of this problem for the differential forms that habitually
present themselves in applications we make several general remarks that lead us to an extremely
important theorem.
In order for a form to be an invariant of the system (1) it is necessary and sufficient that it
be expressible in terms of first integrals of (1) and their differentials. Thus, a necessary
condition for a given form to be invariant for a conveniently chosen system of differential
equations is that this form can be expressed by means of at most n 1 quantities and their
differentials.

40. Suppose then that the form can be expressed by means of r < n quantities y1, ..., yr
(i.e., functions of xi) and their differentials. Suppose, in addition, that they may not be expressed
in an analogous manner by means of less than r quantities. With these conditions, we shall
prove the following theorem:

In order for such a system of differential equations to admit as an invariant form it is

necessary and sufficient that y1, ..., yr be first integrals of the system.

The condition is obviously sufficient. To prove that it is necessary, consider a system of

differential equations that admit as an invariant form, and write the equations of that system
by taking y1, ..., yr , and n r other independent quantities yr+1 , ..., yn to be the new variables.
Let:
dy1 dy2 dy dy dy
(2) = = ... = r = r +1 = ... = n
Y1 Y2 Yr Yr +1 Yn
40 Lessons on integral invariants

be the equations of the system. If y1, ..., yr are not all first integrals then the first r denominators
Y1, ..., Yr will not all be zero. Suppose, for example, that Yr 0. One may then take yr to be the
independent variable, and the form will not change value if one replaces yr and yr by 0
everywhere. One then replaces:
y1, ..., yr1 , yr+1 , ..., yn
with their initial values:
y10 ,..., yr01 , yr0+1 ,..., yn0 ,

which are regarded as first integrals of system (2), and finally one replaces the differentials:

y1, ..., yn ,
with:
y10 ,..., yn0 .

But then, as contains neither yr+1, ..., yn nor their differentials, the new form thus
obtained depends upon only the y10 ,..., yr01 and their differentials. In other words, one can find r
1 functions z1, ..., zr1 of the xi such that can be expressed in terms of these r 1 functions
and their differentials. This result is contrary to the hypothesis. The number r will be called the
class of the form .

II. - The characteristic system of a differential form.

41. This extremely general theorem leads to some important consequences that help us to
better understand its scope.
From the foregoing, the most general system of differential equations that admit the form ,
when written in terms of the variables y1, ..., yn , as an invariant form is:

dy1 dy2 dy dy dy
(3) = = ... = r = r +1 = ... = n ,
0 0 0 Yr +1 Yn

where Yr+1, ..., Yn are arbitrary functions. We immediately deduce that any first integral that is
common to these systems is a function of the y1, ..., yr . As a result, if the form can be
expressed in a second manner by means of the r quantities z1, ..., zr and their differentials then
the zi will be functions of the yi and conversely, since the zi are first integrals that are common to
all of the differential systems that admit as an invariant form. This amounts to saying that
there is essentially only one manner of expressing the form in terms of a minimum number of
variables and their differentials, in the sense that if one has an expression that involves the
minimum number r of quantities y1, ..., yr then all of the others can be obtained by perfomring an
arbitrary change of variables on the y. This conclusion will obviously be invalid if r is not the
minimum number of variables.
 The characteristic system of a differential form 41

42. Another consequence is the following one: agree to say that a certain number three,
perhaps of the systems of differential equations in n variables:

dx1 dx2 dx
= = ... = n ,
X1 X 2 Xn
dx1 dx2 dx
= = ... = n ,
X 1 X 2 X n
dx1 dx2 dx
= = ... = n
X 1 X 2 X n

are linearly independent if it is impossible to find three coefficients , , that are not all zero
such that one has:
X 1 + X1 + X1 = 0,
X 2 + X 2 + X 2 = 0,
...
X n + X n + X n = 0.

In the contrary case, we say that they are linearly dependent.

The property of several systems being linearly independent or not obviously persists under an
arbitrary change of variables.
Among the systems (3) that admit as an invariant form, one can obviously find n r
linearly independent systems, namely, the ones that one obtains by setting all of the
denominators Yr+1 , ..., Yn equal to zero, except for one. Furthermore, all of the systems (3)
depend linearly upon these n r particular systems.
We thus see that if a form is invariant for n r and only n r linearly independent systems
of differential equations then it will be invariant for any system that depends linearly upon it,
and, moreover, all of these systems will have r independent first integrals in common.

43. For example, suppose n r = 2. There exist two systems of differential equations,
namely:
dx1 dx
= ... = n ,
X1 Xn

that admit as an invariant form, and any other system that enjoys this property will depend
linearly upon these two. Call the trajectories of the first system (C) and those of the second
system (). For any arbitrary point M in n-dimensional space, take the trajectories (C) and ()
that pass through that point. Take an arbitrary point P on (C) and an arbitrary point Q on ().
Finally, construct the trajectory ( ) that passes through P and the trajectory (C ) that passes
through it. These two new trajectories intersect. If y1, ..., yn2 are first integrals that are
common to the two systems considered, and if a1, ..., an2 are the numerical values of these
integrals at the point M then their numerical values at the point P and at the point Q will again be
42 Lessons on integral invariants

the same. As a result, the curves (C), (), ( ) , (C ) will be all situated on the same two-
dimensional manifold:
y1 = a1, y2 = a2, ..., yn2 = an2 ;

hence, the last two intersect each other.

44. The preceding case presents itself precisely in the case of the double integral invariant of
vortex theory, which corresponds to the differential form:

(5) = y z + z y + x y + ( w v) xt + ( u w) y t + ( v u) z t.

We have seen (sec. 24) that the systems of differential equations that admit as an invariant
form are the ones that imply, as a consequence, the three equations:

(dz wdt ) (dy vdt ) = 0,

(6) (dx udt ) (dz wdt ) = 0,
(dy vdt ) (dx udt ) = 0.

The most general of these systems may be written:

dx dy dz dt
= = = ,
u + v + w +

and from this we linearly derive the two systems:

dx dy dz dt
= = = ,
u v w 1
dx dy dz dt
= = = ,
0

that define the trajectories of the molecules of fluid and the vortex lines. The first system defines
the curves (C), and the second system defines the curves () at any time, and the properties that
one obtains in the general case may be expressed here by saying that the molecules that form a
vortex line () at the instant t again form a vortex line ( ) at the instant t . Helmholtzs
theorem is thus a very special consequence of the theorem that was proved at the beginning of
this chapter.

45. In the preceding two sections we supposed that n r = 2. Analogous geometric

considerations may be developed for any values of n and r. They are based upon the existence of
manifolds that are defined by r equations of the form:

y1 = a1, y2 = a2, ..., yr = ar ,

 The characteristic system of a differential form 43

and are such that any trajectory of a differential system (3) that contains one point of the
manifold will be contained completely in the manifold. Each of these manifolds, which are n
r-dimensional, can be obtained by starting with an arbitrary point M and passing a trajectory of
one of the systems that admit as an invariant form through that point, and then passing a
trajectory of either system through an arbitrary point P of this trajectory, and so on. One can
generate any n r-dimensional manifold by these operations and never escape it.
We give such a manifold the name of characteristic manifold for the form .
Characteristic manifolds may be regarded as resulting from the equations:

dy1 = 0, dy2 = 0, ..., dyr = 0.

However, if one returns to the original variables x1, ..., xn then these equations will be
composed of the set of linear relations in dx1, ..., dxn that are consequences of the equations of
any differential system that admits as an invariant form.
More simply, one may say: the necessary and sufficient condition for the elementary
displacement (dx1, ..., dxn) to be performed in the direction of a trajectory of a differential system
that admits as an invariant form translates analytically into a certain number of linear
equations in dx1, ..., dxn . These equations, r of which are assumed to be independent, define n
r-dimensional manifolds that depend upon r arbitrary constants, such that one and only one of
them passes through any point of space; these are the characteristic manifolds. The linear
system of total differentials itself is called the characteristic system of the form .

46. To abbreviate, call an equation that is linear in dx1, ..., dxn a Pfaff equation, and a system
of Pfaff equations, a Pfaff system. A system of r Pfaff equations in n variables can always be
regarded as defining r variables, which are considered to be dependent variables, that are
functions of the other n r, which are considered to be independent variables. In general, such a
system is impossible. For example, a classical result is that a Pfaff equation in three variables:

P dx + Q dy + R dz = 0,

in which one regards z as an unknown function of x and y, admits a solution that corresponds to
arbitrary given initial values only if a certain integrability condition, namely:

R Q P R Q P
P +Q + R = 0,
y z z x x y

is satisfied. In this case, one says that the system is completely integrable.

Similarly, one says that a Pfaff system of r equations with r unknown functions in n r variables
is completely integrable if it always admits a solution that corresponds to arbitrary given initial
values of these variables. That is what happens for the characteristic Pfaff system of a form .
The fundamental theorem of this chapter may be stated as follows:
44 Lessons on integral invariants

The characteristic Pfaff system of an arbitrary differential form is always completely

integrable.

47. Let us return one last time to the form of vortex theory. The characteristic Pfaff
system of that form is defined by equations (5) or, what amounts to the same thing:

dx udt dy vdt dz wdt

= = .

If we know how to express the idea that such a system is completely integrable then we will
necessarily arrive at the analytical translation of Helmholtzs theorem. As for the characteristic
manifolds, they are composed of the set of all states of the molecules that constitute such a
vortex line.
The characteristic Pfaff system for the double integral invariant of dynamics reduces to the
equations of motion, and the characteristic manifolds reduce to the trajectories.
The situation could be otherwise if, as we did in vortex theory, we were to consider only one
type of trajectory; for example, trajectories that satisfy some system of relations between the
variables.
 CHAPTER V

SYSTEMS OF PFAFF INVARIANTS

AND THEIR CHARACTERISTIC SYSTEMS

I. - The notion of an invariant Pfaff system

48. Instead of invariant forms for a system of differential equations one may also consider
invariant equations. In particular, H. Poincar used a finite system of invariant equations. They
enjoy the property that if a point satisfies such a system then all of the points that can be obtained
by displacing along the corresponding trajectory will also satisfy such a system. To use
geometric language: The manifold that is represented by a system of invariant equations is
generated by the trajectories.
One may also consider invariant differential equations. First of all, we restrict our point of
view to the simple case of two differential equations:

dx dy
(1) = X, = Y.
dt dt
The equation:
(2) y m(x, y, t)x = 0,

will be called invariant in the sense of H. Poincar if, given two arbitrary infinitely close
simultaneous points (x, y, t) and (x + x, y + y, t) that satisfy relation (2), the
points ( x , y , z , t ) and ( x + x, y + y, z + z, t ) that are obtained by displacing them along
their respective trajectories up to another arbitrary instant t also satisfy relation (2), i.e., one has
once more that:
y m( x , y , t ) x = 0.

If equation (2) is invariant, in the sense that we just made precise, then it will be equivalent to
the equation:
(3) y0 m(x0, y0, 0) x0 = 0,

in which one denotes the initial values of x, y along the trajectory that passes through the point
(x, y, t) by x0, y0. Now, if one regards x0, y0 in equation (3) as functions of (x, y, t), and if one
replaces x0, y0 with their numerical values then this equation will obviously take the form:

(4) y Y t m(x X t) = 0.
46 Lessons on integral invariants

This new equation (4), on account of its origin, has an invariant significance, in the complete
sense of the term, because it expresses an intrinsic property of the two trajectories that
correspond to the points (x, y, t) and (x + x, y + y, t + t).
Geometrically, equation (4) makes a plane (P) that passes through any point M = (x, y, t)
correspond to that point. The invariance property signifies that the line MM1 that joins a point M
to an infinitely neighboring point M1 is situated in the plane (P) that corresponds to the point M,
and if one displaces M and M1 along their respective trajectories (while always keeping them
infinitely close to each other) into MN and M1N then the line MNM1N will be situated in the
plane (PN) that corresponds to the point MN. We remark that the plane (P) is tangent to the
trajectory that passes through the point M.
From the preceding, it is obvious that if a curve (C) satisfies equation (4) at each of its points,
i.e., if it is an integral curve, then the surface that is generated by the trajectories that pass
through the different points of (C) will also be an integral surface of equation (4). This also
results analytically from the form (3) of equation (4).

49. The preceding considerations are easily generalized. Given a system of differential
equations:
dxi
(5) = Xi (i = 1, 2, ..., n),
dt
a system of Pfaff equations:
a11 x1 + ... + a1n xn + a1 t = 0,

(6) .......
a x + ... + a x + a t = 0,
h1 1 hn n h

will be called invariant for system (5) if equations (6) can be expressed uniquely in terms of first
integrals of (5) and their differentials; for example, ones that have the form:

a110 x10 + ... + a10n xn0 = 0,

(7) ...
a 0 x 0 + ... + a 0 x 0 = 0.
h1 1 hn n

This demands that equations (6) must be verified identically when one replaces xi by Xit.
However, this condition is obviously not sufficient. Be that as it may, if the Pfaff system (6) is
invariant then it will enjoy the important geometric property that given an arbitrary integral
manifold of the system (6), the manifold that is obtained by guiding each point of the given
manifold along the trajectory that corresponds to equations (7) will again be integral. Indeed,
this results if one displaces it onto this new manifold at an arbitrary point while equations (7) do
not cease to be verified.
(It should be understood that one calls a manifold an integral manifold if it is such that
equations (6) are verified if one displaces on that manifold in an arbitrary direction (x1,..., t).)
From this, it results that any integral manifold of an invariant Pfaff system (6) enjoys either
the property of being generated by the trajectories of the given equations (5) or the property of
 Systems of Pfaff invariants 47

being a part of an integral manifold of very large dimension that is itself generated by the
trajectories.

II. - The characteristic system of a Pfaff system.

50. Given an arbitrary Pfaff system of n variables (x1, ..., xn):

a11 x1 + ... + a1n xn = 0,

(8) ...
a x + ... + a x = 0,
h1 1 hn n

one may propose to determine all systems of differential equations:

dx1 dx2 dx
(9) = = ... = n
X1 X 2 Xn

for which system (8) is invariant. This is a problem that we will solve later on, but, without
solving it, one may still prove an important theorem that concerns all of these systems, a theorem
that is identical to one that was proved in the preceding chapter in the context of a given
differential form.
Suppose that the given equations (8) can be written in terms of the r quantities y1, ..., yr ,
which are functions of x and its derivatives, in the form:

b11 ( y ) y1 + ... + b1r yr = 0,

(10) ...
b ( y ) y + ... + b ( y ) y = 0,
h1 1 hr r

and suppose moreover that they cannot be written in terms of less than r quantities and their
differentials. The number r will be called the class of the system. The necessary and sufficient
condition for the Pfaff system (8) to be invariant for the equations (9) is that these equations (9)
must admit y1, y2 , ..., yr as first integrals.
The proof is exactly the same as in the preceding chapter, and the consequences that one
deduces are also the same. In particular, the equations that express that the given Pfaff system
(8) is invariant for the system of differential equations (9) can be reduced to r linear equations in
X1, ..., Xn , or what amounts to the same thing to r linear equations in dx1, ..., dxn , and these r
equations form a Pfaff system that is equivalent to:

dy1 = 0, ..., dyr = 0;

i.e., they are completely integrable. This Pfaff system is called the characteristic system of the
given Pfaff system (8). The equations of the characteristic system can be ultimately obtained by
adjoining r h other equations to the given system of h equations (8).
48 Lessons on integral invariants

The necessary and sufficient condition for such a Pfaff system (8) to be completely integrable
is obviously that it must coincide with its own characteristic system, in such a way that if one
knows how to form the characteristic system of an arbitrary Pfaff system then one will know
how to express that it is completely integrable by the same means.

51. It is obvious that a Pfaff system (8) can be regarded as invariant for its characteristic
system. Any integral manifold of the system (8) is either generated by its characteristic
manifolds or defines a subset of an integral manifold of much larger dimension that is itself
generated by characteristic manifolds.
If one considers an arbitrary differential form, and if this form is invariant for a certain
system of differential equations then the characteristic Pfaff system of the form will be invariant
for the same system of differential equations.
It follows that in hydrodynamics the Pfaff system:

x u t y v t z w t
= = ,

is invariant for the differential equations of the trajectories of the fluid molecules (and also for
the differential equations of the vortex lines).
All of these theorems, and some others that one may easily imagine, are immediate
consequences of the characteristic property of an invariant system that it involves only first
integrals of the differential equations for which it is invariant.

52. Consider either a differential form or a Pfaff system, or likewise, a set of several
differential forms and a Pfaff system, and let y1, ..., yr denote either the first integrals of the
characteristic Pfaff system, the given differential form, or the given Pfaff system, etc. It is
obvious that if one directs ones attention uniquely to the manner by which the differentials x1,
..., xn figure in the differential form, or the in the Pfaff system, etc., without being preoccupied
with the coefficients then these differentials will enter only in combinations of y1, ..., yr .
However, it might also be the case that they enter only as linear combinations of a number less
than r. In any case, if one knows the linear combinations in a minimum number of xi by means
of which the form (or Pfaff system, etc.) can be expressed then the equations that are obtained
by annulling these linear combinations will be a subset of the characteristic system.

III. - The rank of an algebraic form and its associated system.

53. The preceding considerations increase in clarity if we prove a theorem for algebraic
forms that is analogous to the one that led to the notion of a characteristic system:

If an algebraic form in n variables u1, ..., un can be expressed in terms of r linearly

independent combinations v1, ..., vr of variables without being expressible in terms of a smaller
number, and if, moreover, one has found another expression for the form by means of r other
 Systems of Pfaff invariants 49

linear combinations w1, ..., wr of variables then the wi will be independent linear combinations
of the vi.
Indeed, consider the 2r linear forms:

v1, ..., vr; w1, ..., wr

of the given variables. Suppose that among these forms there are 2r independent ones (0
r). This amounts to saying that there exist linearly independent combinations of the v that
are, at the same time, linear combinations of the w; call them t1, ..., t . Suppose, moreover as
is legitimate that the t1, ..., t are independent linear combinations of both the v1, ..., v and the
w1, ..., w . One then has a double equality of the form:

F(x1, ..., xn) = (t1, ..., t ; v+1 , ..., vr) = (t1, ..., t ; w+1 , ..., wr).

Since the quantities t1, ..., t , v+1 , ..., vr , w+1 , ..., wr are independent, this is possible only if
F, for example, does not depend upon the v+1 , ..., vr . This is not compatible with the hypothesis
that = r, and the theorem is thus proved.
The system of linear equations:
v1 = v2 = ... = vr = 0

will be called the associated system of the given form. The notion of associated system
obviously extends to a set of forms, or again to a system of algebraic equations. We say that the
integer r is the rank of the form.
From this, we infer that the characteristic system of a differential form always contains the
associated system of that form, which is considered to be an algebraic form in x1, ..., xn .
However, it may contain other equations besides the equations of the associated system.
 CHAPTER VI

FORMS WITH EXTERIOR MULTIPLICATION

I. - The associated system of a quadratic form.

54. We must say a word about ordinary algebraic, quadratic, cubic, forms, etc.
A quadratic form:
1,,n
(1) F(x) = aij ui u j = a11u12 + a22 u22 + + 2a12 u1u2 +
i, j

is, as one knows, reducible to a sum of squares. If the discriminant of the form is different from
zero then there will be as many squares as the number n of independent variables. We propose
to determine the minimum number of variables by means of which the form may be expressed
(by a convenient substitution). In order to obtain these variables it will suffice to consider the
system of linear equations:
F F F
(2) = 0, ..., = 0.
u1 u2 un

It is immediately obvious that this system is independent of the choice of variables. Suppose
that it reduces to r independent equations, which one may always suppose to be:

x1 = 0, x2 = 0, ..., xr = 0.

Having said this, the form F can be expressed in terms of r variables x1, ..., xr and cannot be
expressed in terms of less than r variables.
Indeed, express F by means of x1, ..., xr and of n r other independent forms xr+1, ..., xn . The
variable xr+1, for example, does not enter in F because if it did enter into a term such as A xr+1 x
then the equation:
F
= 0,
x

would contain xr+1, which is contrary to the hypothesis.

Conversely, suppose that the form F can be expressed by means of r variables y1, y2, ...,
yr . The system (2) that is formed by starting with the variables y1, ..., y , ..., yn manifestly
contains only the variables y1, ..., yr . It is therefore necessary that = r, and system (2) will then
reduce to:
y1 = y2 = ... = yr = 0.

The y1, ..., yr are thus independent linear combinations of the x1, ..., xr .
 Forms with exterior multiplication 51

The last part of the proof shows, as we already know, that expressing F in terms of a
minimum number of variables is possible in essentially one manner, up to a linear substitution of
a minimum number of these variables.
System (2) is the associated system of the form F.
The foregoing can be extended to a form that is integer and homogeneous of arbitrary degree.
For example, if F is a cubic form then the associated system of linear equations will be obtained
by annulling all of the second derivatives of F:

2F
= 0.
ui u j

This system gives the minimum number of variables by means of which F may be expressed.

II. - Alternating bilinear forms and quadratic exterior forms.

55. The forms with which we shall now occupy ourselves are the ones that appear under a
multiple integral sign when one considers the differentials to be variables. These are forms that
have special rules of calculation, upon which it is not pointless to insist.
We start with a bilinear form:
f(u, v) = aij ui uj ,
in two series of variables:
u1, ..., un ; v1, ..., vn .

Such a form is called symmetric if it is preserved when one exchanges the two series of
variables:

f(u, v) = f(v, u)

and alternating if it is preserved with a sign change under the same conditions:

f(u, v) = f(v, u).

The conditions that the coefficients of the symmetric form must satisfy are:

aij = aji .

The conditions for it to be alternating are:

aij + aji = 0, aii = 0.

52 Lessons on integral invariants

If one subjects the two series of variables ui and vi to the same linear substitution then the
form
f(u, v) will be changed into a new bilinear form F(U, V) in the new variables Ui, Vi, and it is
obvious that the form F(U, V) will again be symmetric if f is, and alternating if f is. This says
that the exchange of the two series of new variables U and V amounts to the exchange of the
original two series of variables u and v.
One may make a quadratic form, namely f(u, u), correspond to any symmetric bilinear
function
f(u, v), and the correspondence is invertible. If one sets:

f(u, u) = F(u)
then one will have:
1 F F F
f(u, v) = v1 + v2 + + vn .
2 u1 u2 un

An analogous correspondence can no longer be established for the alternating forms because
f(u, u) becomes identically zero in that case. This is an inconvenience that one may obviate in a
manner that we shall now describe.

56. We first remark that the coefficients of the terms ui ui are all zero for an alternating
bilinear form, and the coefficients of the terms ui uj and uj ui will have opposite signs. One may
then write:
f(u, v) = aij (ui u j u j ui ),
( ij )

in which the summation on the right-hand side is taken over all combinations of n index pairs, in
n(n 1)
such a way that there are terms in this summation. Since the expression ui vj uj vi is
2
nothing but the determinant:
ui u j
,
vi v j

one may, by a notational convention, denote it by the notation ():

ui vj uj vi = [ui uj]

by writing first the one and then the other of the two elements of the first row of the determinant
and then putting them between brackets. With this notation, one has:

() [DHD] This notation is misleading, compared to the more modern notation of u v , insofar as it makes no
mention of v; however, since Cartan uses this notation throughout, we shall remain faithful to the original.
 Forms with exterior multiplication 53

f(u, v) = aij [ui uj].

Similarly, we agree to denote the alternating bilinear form that is defined by the determinant:

f (u ) f (u )
f (v ) f (v)

by the notation [f(u) f (u)], in which f and f denote two arbitrary linear forms:

f(u) = a1 u1 + a2 u2 + ... + an un ,
f (u ) = a1 u1 + a2 u2 + ... + an un .

If one develops the preceding determinant then one will immediately find that:

f (u ) f (u ) ui uj
[ f (u ) f (u )] = = ai aj = ai aj [ui u j ].
f (v ) f (v) i j vi vj i j

Comparing the left and right-hand sides shows that the development of [f(u) f (u)] can be
obtained by regarding that expression as a product and then developing that product according
to the ordinary rules of algebra but without changing the order of the factors in the partial
products and agreeing that any partial product that contains two identical variables will be
zero, and that any partial product of two different variables will change sign when one changes
the order of the factors.
The multiplication whose rules that we have just stated is due to H. Grassmann, who gave it
the name of exterior multiplication.
Upon using that operation, one sees that one can make a form of second degree in only one
series of variables but with exterior multiplication correspond to any alternating bilinear
form, and conversely, to any quadratic form with exterior multiplication, there corresponds an
alternating bilinear form.
To abbreviate, we say exterior form instead of form with exterior multiplication.

57. If one performs a linear substitution of the variables in an exterior form F(u) then the
new form will be obtained simply by developing each partial product [ui uj] as a function of the
new variables.
The partial derivative F / u1 of an exterior quadratic form is defined simply as the sum of
the partial derivatives of its terms. A term that does not contain u1 will naturally have a zero
derivative. As for a term that does contain u1, one can always suppose that u1 appears as the first
term in the partial product. The derivative of A[u1 uj] will then be A ui . For example, one has:

[u1u2 ] [u1u2 ] [u1u2 ] [u u ]

= u2, = u1, = 0, , 1 2 = 0 .
u1 u2 u3 un
54 Lessons on integral invariants

With these conventions:

F F F
2F(u) = u1 + u2 + + un ,
u1 u 2 u n

in which the partial products of the right-hand side are exterior products.
If F(u) corresponds to the alternating form f(u, v) then one will obviously have:

F F F
f(u, v) = v1 + v2 + + vn ,
u1 u2 un

in which the partial products will be formed according to the rules of ordinary multiplication.
Finally, we remark that if one performs a linear substitution:

ui = hi1U1 + ... + hin Un (i =1, ..., n)

on the ui, and if F(u) becomes (U) by that substitution then one will have:

F F F
= h1k + h2 k + + hnk ,
U k u1 u2 un

since F is an ordinary algebraic form.

58. The system of linear equations:

F F F
(3) =0 = 0, , = 0,
u1 u2 un

in which F is a given exterior quadratic form, is obviously independent of the choice of

variables. One may then suppose that it reduces to the equations:

u1 = 0, u2 = 0, ..., ur = 0 ( n).

It is then the case that the form F will depend upon only ur+1 , ..., un . Indeed, if it contains a
term such as A[ur+1 u] then the equation:
F
=0
u

will not be a consequence of equations (3). The form F can thus be expressed uniquely by
means of the right-hand sides of equations (3).
Conversely, suppose that the form F can be expressed by means of r variables v1 , v2 , ...,
v . The left-hand sides of the equations of the system:
 Forms with exterior multiplication 55

F F
= 0, , =0
v1 vn

will depend upon only v1, ..., v . This system will thus contain at most independent equations.
As a result, one will have = r, and the vi will be linear combinations of the ui .

The associated system of an exterior quadratic form is therefore obtained by annulling all of
its first-order partial derivatives.

59. This result can made be made more precise. We shall show that the rank r is necessarily
even, and at the same time, find a reduced form for the exterior quadratic form that will play the
same role that the sum of the squares does for ordinary quadratic forms.
Suppose, to fix ideas, that the coefficient a12 of F(u) are not zero, and consider the form:

1 F F 1
= [(a12u2 + a13u3 + + a1nun )(a21u1 + a23u3 + + a2nun )].
a12 u1 u2 a12

This form has the same coefficients as F for the terms in:

[u1 u2], [u1 u3], ..., [u1 un], [u2 u3], ..., [u2 un],

i.e., for the terms that contain at least one of the variables u1 or u2. As a result, the form:

1 F F
F (u ) = F (u )
a12 u1 u 2

of that form is
will contain only the variables u3, u4, ..., un . Suppose then that the coefficient a34
not zero. Similarly, one will verify that the form:

1 F F
F (u ) = F (u )

a34 u3 u4

contains only the variables u5, u6, ..., un . One may thus continue, step by step, until one arrives at
a form that is identically zero. For example, suppose that one has:

1 F F 1 F F 1 F F
F(u) = + + .
a12 u1 u2 a34 u3 u4 a36 u5 u6

The six linear forms:

56 Lessons on integral invariants

F F F F F F
, , , , , ,
u1 u2 u3 u4 u5 u6

will obviously be independent. Upon setting:

F F
= U1, = a12U 2 ,
u1 u2
F F
= U3, U4 ,
= a34
u3 u4
F F
= U5, U 6 ,
= a36
u5 u6

the form F will be reduced to desired canonical form:

F(U) = [U1U2] + [U3U4] + [U5U6].

This argument is obviously general and leads to the canonical form:

F(U) = [U1U2] + ... + [U2s1U2s] (2s n).

The associated system is obviously:

U1 = U2 = ... = U2s = 0.

This result will have great significance later on.

60. The reduction of an exterior quadratic form to its canonical form is obviously possible in
an infinitude of ways. The set of linear substitutions that make a canonical form pass into
another constitute an important group that depends on s (2s + 1) arbitrary parameters. If s = 1
then these substitutions of two variables will be characterized by the condition that their
determinant must be equal to unity.

III. - Exterior forms of degree greater than two.

61. One may imagine exterior forms of arbitrary degree. One arrives at them most naturally
upon starting with a linear form in p series of variables ui , vi , ..., wi :

f(u, v, ..., w)

that satisfies the condition that exchanging two series of variables will reproduce the form, but
with a sign change. In the case p = 3, for example, this hypothesis entails the consequence that
 Forms with exterior multiplication 57

any term in which the same index appears twice will have a zero coefficient and that the set of
terms in which three distinct indices appear for example, 1, 2, 3 will be of the form:

u1 u2 u3
a123 v1 v2 v3 .
w1 w2 w3

The same notational convention as above leads to a distributive, but not commutative,
multiplication law, by which each product changes sign if one exchanges two of the variables
that appear in it. Consequently, one will have:

[u1 u2 u3] = [u2 u1 u3] = [u1 u3 u2] = [u3 u2 u1] = [u2 u3 u1] = [u3 u1 u2].

From this, one may define an exterior product such as:

[F ],

in which F, , , are exterior forms of arbitrary degree. The degree of the product is the sum of
the degrees of the factors. The product will necessarily be zero if the sum of the degrees exceeds
n. One easily confirms that if one exchanges two factors in the product then the product will not
change sign if at least one of these factors is of even degree, and it will undergo a sign change if
both of the degrees are odd. Similarly, one may define a sum of products of this nature.
In particular, the product of a form with itself will be zero if this form is of odd degree, but it
will not necessarily be zero if it is of even degree. For example, take a quadratic form F that has
been reduced to its canonical form:

F = [u1 u2] + [u3 u4] + ... + [u2s1 u2s].

One has:
1 2
[ F ] = [u1u2 u3u4 ] + [u1u2 u5 u6 ] + + [u2 s 3u2 s 2 u2 s 1u2 s ] ,
2
1 3
[ F ] = [u1u2 u3u4 u5u6 ] + ,
3!
1 s
[ F ] = [u1u2 u3u4 u2 s 1u2 s ] ,
s!
1
[ F s +1 ] = 0.
( s + 1)!

The rank 2s of a quadratic form F is therefore twice the largest power to which one may raise F
without annihilating it.
A simple application of the theory of determinants is the following one. Let:

F = a12[u1 u2] + a13[u1 u3] + a14[u1 u4] + a23[u2 u3] + a24[u2 u4] + a34[u3 u4]
58 Lessons on integral invariants

be a form in four variables. One has:

1 2
[ F ] = (a12 a34 a13 a24 + a14 a23 )[u1u2 u3u4 ] .
2

On the other hand, the associated system of F is:

a12u2 + a13u3 + a14u4 = 0,

a21u1 + + a23u3 + a24u4 = 0,
a31u1 + a32u2 + + a34u4 = 0,
a41u1 + a42u2 + a43u3 + = 0.

The condition for the form to be expressed by means of at least four variables is, on the one
hand, that one must have [F2] = 0, i.e.:

a12 a34 a13 a24 + a14 a23 = 0,

and, on the other hand, that the determinant of the associated system must be zero, i.e.:

0 a12 a13 a14

a21 0 a23 a24
= 0.
a31 a32 0 a34
a41 a42 a43 0

Despite appearances, these two equations are equivalent. Indeed, one may prove that the
determinant, which is anti-symmetric of even degree, is the square of the expression:

a12 a34 a13 a24 + a14 a23 .

62. Any exterior form of degree n (which is equal to the number of variables) is of the form:

A[u1 u2 ... un].

One may obtain canonical forms when the degree is n 1 or n 2. One then easily arrives at the
notion of the adjoint form to a given form.
Consider a form F of degree p, and let , , ..., ( n p 1) denote the linear forms with
indeterminate coefficients:
= 1 u1 + ... + n un ,
= 1 u1 + ... + n un ,
...

The exterior product [F ... ( n p 1) ] is of degree n, and so, as a result, is the form:
 Forms with exterior multiplication 59

[ u1 u2 ... un].

The coefficient F is linear with respect to each series of coefficients . Furthermore, it is

alternating. Therefore it corresponds to an exterior form of degree n p in the variables 1 , ...,
n ; this is the definition of the form that is adjoint to F.
If one performs a linear substitution on the u and if, at the same time, one performs a linear
substitution on the that preserves the expression 1 u1 + ... + n un then the expression [u1 ...
un] will obviously be preserved. In other words, the adjoint form is reproduced, but multiplied
by the determinant of the substitution that was performed on the variables u.
The adjoint form to a form F = F1 + F2 is obviously the sum of the adjoint forms 1 and 2.
Similarly, the adjoint form to a, where a is a numerical coefficient, is aF. From this, in order
to calculate the adjoint form to an arbitrary form it will suffice to know how to calculate the
adjoint form to a monomial form such as:

F = [u1 u 2 u n ] .

Upon applying the given definition, one finds that:

= [ p +1 , , n ] ,

since the indices p+1, ..., n are the ones among the indices 1, 2, ..., n that do not appear in the
set of 1, 2, ..., p . These indices are supposed to range in such an order that the total
sequence:
1 , ..., p , p+1 , ..., n
is even.

63. From this, suppose that F is a form of degree n1. The adjoint form will be of first
degree. One may thus suppose that it is reduced to n ; for example, in such a way that F may
always be converted into an expression of the form:

F = [u1 u2 ... un1].

Now suppose that F is of degree n 2. The form F will be of second degree. Therefore, one
may always suppose that it is given by formula:

= [1 2] + ... + [2s12s].
As a result, one will have:

F = [u3 u4 u5 ... un] + [u1 u2 u5 u6 ... un] + ... + [u1 u2 ... u2s2 u2s+1 ... un].

If s = 1 then F will be reduced to a monomial form.

For example, if n = 5 then any form F of degree 5 2 = 3 will be reducible to one of the
canonical forms:
60 Lessons on integral invariants

F = [u3 u4 u5],
F = [u1 u2 u5] + [u3 u4 u5].

If n = 6 then any form F of degree 4 will be reducible to one of the forms:

F = [u3 u4 u5 u6],
F = [u3 u4 u5 u6] + [u1 u2 u5 u6] = [([u1 u2] + [u3 u4]) u5 u6],
F = [u3 u4 u5 u6] + [u3 u4 u5 u6] + [u3 u4 u5 u6] = 12 [[u1 u2] + [u3 u4] + [u5 u6]]2.

The notion of adjoint form permits us to define the product of two forms when the sum of
their degrees exceeds n. This is the operation that H. Grassmann called regressive exterior
multiplication, but we shall not use it.

64. Again, we point out several applications of exterior multiplication. Suppose that f1, f2,
..., fh are h independent linear forms. The equation:

[F f1 f2 ... fh] = 0,

in which F is an arbitrary exterior form, gives the necessary and sufficient condition for F to be
annulled when one establishes relations:

f1 = 0, f2 = 0, ..., fh = 0
between the variables.
Indeed, one may consider the case in which one has fi = ui . In that case, if every term of F
contains one or more of the variables u1, .., uh then it will be obvious that the product [F u1 ... uh]
must be zero. Conversely, if this product is zero then an arbitrary term of F will contain one or
more of the variables u1, ..., uh as a factor, except that the multiplication of this term by [u1...uh]
will give a non-zero product that cannot be reduced to any other.

IV. - The associated system of an exterior form.

65. The determination of the associated system may performed just as easily for an exterior
form of arbitrary degree as it is for a quadratic form. If the form is of degree p then the
associated system will be obtained by annulling all of the partial derivatives of F of order p 1.
F
One will define a first order derivative, such as , to be the coefficient of u1 in the set of terms
u1
of F that contain that variable, after first taking the precaution of insuring that u1 has rank one in
F
each of these terms. We remark that this derivative no longer depends on u1. By definition,
u1
2F F
the derivative will be the derivative of with respect to u2 . One obtains it as a result
u1u2 u1
 Forms with exterior multiplication 61

of taking the set of terms of F that contain both the variables u1 and u2 , while making sure that u1
is of first rank and u2 is of second rank in each of these terms and finally suppressing the
variables u1 and u2 from all of these terms. From this, one has:

2 F 2F
= .
u1u2 u2u1

The higher-order partial derivatives are defined in the same fashion; they are necessarily
taken with respect to variables that are all different.
The rank of a form of degree n that is not identically zero is obviously equal to n. The rank
of a form of degree n 1 is equal to n 1. The rank of a form of degree n 2 is equal to n 2 if
it is reducible to a monomial form and to n in any other case. If the degree is less than n 2 then
one can say nothing a priori about the rank.

V. - Formulas that relate to exterior quadratic forms.

66. We return to the case of an exterior quadratic form F in n variables u1, u2, ..., un . It can
happen that one supposes that the variables are coupled by a linear relation:

f a1 u1 + a2 u2 + ... + an un = 0.

The form F, in which we will suppose, for example, that un is expressed as a function of u1,
..., un1 by means of the given relation, will have a certain rank that corresponds to the number
of linearly independent equations in its associated system. The latter obviously has:

F a1 F F a F
= 0, , n 1 = 0, f =0
u1 an un un1 an un

for its equations, or again:

F F F
u1 u2 u
= = = n , f = 0.
a1 a2 an

More generally, one may suppose that these variables are coupled by an arbitrary number of
relations:
f a1 u1 + a2 u2 + ... + an un = 0,
g b1 u1 + b2 u2 + ... + bn un = 0,
...
h l1 u1 + l2 u2 + ... + ln un = 0.

The associated system of F will then be defined by the formulas:

62 Lessons on integral invariants

F F F

u1 u2 un
a1 a2 an
= 0, f = 0, g = 0, ..., h = 0.
b1 b2 bn

l1 l2 ln

Equating the matrix above means annulling all of the determinants that are formed from the
rows of that matrix and the same number of columns.
One may remark that when one supposes that the variables are coupled by given relations,
the rank 2s of the form F will be twice the largest exponent such that the form:

[f g ... h F s ]
is not zero.

67. In particular, suppose that n = 2s and that the form F is of rank n. If one couples the
variables by only one relation then it will be obvious that the rank of the form cannot exceed n
1 = 2s 1, and since this rank is even it will be equal to at most 2s 2. Furthermore, it is easy
to see that it cannot descend below that limit.
From this, it results that if one couples the variables by p independent linear relations then
the rank of F will be diminished by at most 2p units. We look for the case in which the
maximum reduction will be obtained. If the relations are:

f1 = 0, f2 = 0, ..., fp = 0

then it will be necessary and sufficient that one must have:

(4) [f1 f2 ... fp Fsp+1] = 0.

This condition can be replaced by other simpler conditions. Indeed, we remark that if one
takes any two of the p given relations then these two relations will necessarily diminish the rank
of F by 4 units. Therefore, one will have:

(5) [f ifjFs1] = 0 (i, j = 1, 2, ..., p).

p ( p 1)
We shall prove that these necessary equations are also sufficient.
2
Indeed, suppose that these conditions are satisfied and make a change of variables in such a
manner as to take fi to ui . One will thus have

[ui uj Fs1] = 0,
 Forms with exterior multiplication 63

which shows that there is no term in [i j] in the form () that is adjoint to Fs1(u). Now, the
form that is adjoint to Fsq is q ; this is easily recognized by supposing that F is reduced to its
canonical form. As a result, each term of the adjoint form to Fsp+1, which is Fp1, will contain at
most p 1 of the variables 1, ..., p . As a result, the form that is adjoint to Fp1 will contain at
least one of the variables u1, ..., up . This amounts to saying that one will have:

[u1 u2 ... up Fsp+1] = 0.

Q.E.D.

The significance of the preceding theorem is easy to point out. Since the forms [fi fj Fs1] are
p ( p 1)
of degree n, the equations to be written will be in number, whereas, since the form [f1 f2
2
... fp Fsp+1] is of degree n p + 2, the number of equations that express that it is zero will be
Cnp 2 , and furthermore, each of them will contain the coefficients of all the given relations.
For example, if one has:
F = [u1 u2] + [u3 u4] + [u5 u6],
f1 = a1 u1 + ... + a6 u6,
f2 = b1 u1 + ... + b6 u6,
f3 = c1 u1 + ... + c6 u6

then, on account of the three relations f1 = f2 = f3 = 0, the condition that F be of rank 6 6 = 0

will be, by the first procedure:
[f1 f2 f3 F] = 0,
which gives:

a1 a3 a4 a1 a5 a6 a2 a3 a4 a2 a5 a6
b1 b3 b4 + b1 b5 b6 = 0, b2 b3 b4 + b2 b5 b6 = 0,
c1 c3 c4 c1 c5 c6 c2 c3 c4 c2 c5 c6

a3 a5 a6 a3 a1 a2 a4 a5 a6 a4 a1 a2
b3 b5 b6 + b3 b1 b2 = 0, b4 b5 b6 + b4 b1 b2 = 0,
c3 c5 c6 c3 c1 c2 c4 c5 c6 c4 c1 c2

a5 a1 a2 a5 a3 a4 a6 a1 a2 a6 a3 a4
b5 b1 b2 + b5 b3 b4 = 0, b6 b1 b2 + b6 b3 b4 = 0.
c5 c1 c2 c5 c3 c4 c6 c1 c2 c6 c3 c4

On the contrary, the theorem above puts the required conditions into the very simple form:

b1 c2 c1 b2 + b3 c4 c3 b4 + b5 c6 c5 b6 = 0,
c1 a2 a1 c2 + c3 a4 a3 c4 + c5 a6 a5 c6 = 0,
64 Lessons on integral invariants

a1 b2 b1 a2 + a3 b4 b3 a4 + a5 b6 b5 a6 = 0.

68. One may point out a theorem that is much more precise than the preceding one, and
which permits us to find, in the simplest fashion, the rank of the form to which F reduces when
one supposes that the variables are coupled by p given relations. In order to do this, we define
the alternating bilinear form:
(, ) = aij i j
by the equality:
s[Fs1(1 u1 + ... + n un)( 1 u1 + ... + n un)] = (, )[ ].

The exterior quadratic form:

() = ai j [i j]

F s 1
is (up to a factor) the adjoint form to . It is the absolute covariant of F, in the sense that if
( s 1)!
one performs an arbitrary linear substitution on the variables u1, ..., un , and the linear substitution
that preserves 1 u1 + ... + n un on the variables 1, ..., n , and if these two substitutions take the
two forms F(u) and () to F (u ) and ( ) then one will also have:

s[ F s 1 (1u1 + + nun )(1u1 + + nun )] = ( , )[ F ] .

In particular, if F has been reduced to its canonical form:

F = [u1 u2] + ... + [u2s1 us]

then one will immediately find for the canonical form of :

= [1 2] + ... + [2s1 s] .

From this, there easily results the general identity:

F s p
(1u1 + + nun )(1u1 + + nun ) (1(2 p 1)u1 + + n( 2 p 1)un
( s p )!
(6)
( p ) ( , , , (2 p 1) F s
= s! ,
p!

in which the exterior form of degree p that corresponds to the alternating multilinear form (p) is
equal to [p()]. This identity is obvious when F has been reduced to its canonical form and is
therefore true in the general case. This basically amounts to the property that was invoked in the
preceding section that the adjoint form to [Fsp] is equal to [p], up to a scalar factor.
 Forms with exterior multiplication 65

In particular, upon setting p = 2, and upon taking the terms in [i , j , k, l] in identity (6)
one will obtain:
F s2 Fs
(7) u u u u
i j k l = ( a a + a a + a a )
il jk ,
( s 2)!
ij kl ik lj
s!

in which the coefficients aij are defined by:

F s2 Fs
u u
i j = aij .
( s 2)! s!

Finally, one may deduce another identity that we will use later on. Consider the form:

F s2 F s 1
u i j k
u u (aij uk + a jk ui + aki u j ;
( s 2)! ( s 1)!

it is of degree 2s 1. If one exterior multiplies it by any of the variables u1, ..., u2s say ul
then one will immediately confirm from (7) that the product is zero. As a result, the form itself
will be identically zero. Since ui , uj , uk can be replaced by three arbitrary linear forms in these
variables, we arrive at the following theorem:

If one considers an arbitrary number of linear forms f1, f2, ..., fp , and if one sets:

F s 1 Fs
f f
i j = aij (i, j = 1, 2, ..., p)
( s 1)! s!

then one will have the identities:

F s2 F s 1
(8) f i j k =
f f (aij f k + a jk fi + aki f j ) .
( s 2)! ( s 1)!

69. We shall now go on to the problem that was stated above, which consists of finding the
rank of the form to which F is reduced when one supposes that the variables are coupled by p
independent linear relations:
f1 = 0, f2 = 0, ..., fp = 0.

We may suppose that these relations are:

u1 = 0, u2 = 0, ..., up = 0.
66 Lessons on integral invariants

They permit us to perform an arbitrary linear substitution on the u, with the one condition
that the first p variables u1, ..., up must be exchanged amongst themselves. It results from this
that we can perform an arbitrary linear substitution on the variables , with the one condition that
the last 2s p variables p+1, ..., 2s must be exchanged amongst themselves. We then set:

F s 1 Fs
ui u j = aij , (i, j = 1, 2, ..., p).
( s 1)! s!

If one suppresses the terms in p+1, ..., 2s from then one will obviously obtain:

1,, p
= a [
( ij )
ij i j ].

Let 2q be the rank of the form . One may, by a convenient linear substitution that is
performed on the 1, ..., p , reduce to:

= [1 2 ] + + [2 q 12 q ] .

As a result, one may, by removing linear combinations of the p+1, ..., 2s from the 1, ...,
p , if necessary and this is permissible reduce to:

= [1 2] + ... + [2q1 2q] + [2q+1 p+1] + ... + [p 2p2q]

+ [2p2q+1 2p2q+2] + ... + [2s1 2s],

but then the form F will be changed into:

F = [u1 u2] + ... + [u2q1 u2q] + [u2q+1 up+1] + ... + [up u2p2q] + ... + [u2s1 u2s] .

One sees that if one now takes the relations:

u1 = 0, u2 = 0, ..., up = 0

into account then the rank of F will be reduced by 2p 2q units.

We therefore arrive at the following theorem:

p ( p 1)
Consider the independent linear forms f1, f2, ..., fp , the quantities aij that are defined
2
by the equalities:
F s 1 Fs
f f
i j = aij ,
( s 1)! s!

and the exterior quadratic form in p variables 1, ..., p :

 Forms with exterior multiplication 67

1,, p
( ) = a [
( ij )
ij i j ].

If that form is of rank 2q then the rank of the form F will be reduced by 2p 2 q units when one
supposes that the variables are coupled by p relations:

f1 = 0, f2 = 0, ..., fp = 0.

Furthermore, if one performs a linear substitution on the p given linear forms such that is
reduced to its canonical form:
= [1 2] + ... + [2q1 2q]

then the form F will be reduced to the canonical form:

F = [f1 f2] + ... + [f2q1 f2q] + [f2q+1 fp+1] + ... + [fp f2p2q] + ... + [f2s1 f2s],

in which fp+1, ..., f2s denote new forms that are conveniently chosen to be mutually independent
and independent of the given forms.

In particular, if q = 0 then one recovers the preceding theorem that was stated and proved in
sec. 67.
 CHAPTER VII

EXTERIOR DIFFERENTIAL FORMS AND THEIR DERIVED FORMS

I. - The bilinear covariant of a Pfaff form.

70. Now consider a linear differential form (Pfaff form):

= a1x1 + a2x2 + ... + an xn .

One can derive an alternating bilinear form in two types of differentials from that form,
namely:
= ai ( xi xi ) + ( ai xi xi ai ) .

Suppose that the two differentiation symbols are interchangeable, i.e., that one has:

xi = xi .

As for the right-hand side, which one calls the bilinear covariant of the form , one may make it
correspond to an exterior quadratic differential form that we write, with the conventions made
above:
a j ai
= [ ai xi ] = [ x x ].
i

( ij ) xi x j i j

This form will be called the exterior derivative of the form .

This derivation procedure has a significance that is independent of the choice of variables.
Furthermore, it is the one that takes place when one passes from a curvilinear integral taken over
a closed contour to a double integral taken over a surface bounded by the contour.
For example, if one has three variables x, y, z and one sets:

= Px + Qy + Rz
then one will have:

P P P
= [Px] + [Q y] + [R z] = [ x x] + [ y x] + [ z x ]
x y z
Q Q Q
+ [ x y ] + [ y y ] + [ z y ]
x y z
R R R
+ [ x z ] + [ y z ] + [ z z ]
x y z
 Exterior differential forms and their derived forms 69

R Q P R Q P
= [ y z ] + [ z x ] + [ x y ],
y z z x x y

and Stokess formula can be written:

= ,
C S

in which S denotes a surface that is bounded by the contour C.

The necessary and sufficient condition for to be annihilated is that the form must be an
exact differential.

REMARK. The permutability of the two differentiation symbols and must take place
in the case where the differentiations are applied to an arbitrary function of independent variables
or else the differentiations would not have a covariant character. This is easy to verify. If one
sets:
y y
= y = x1 + + xn
x1 xn

then will be an exact differential, and one will have:

= 0,
i.e.:
y = y .

II. - Exterior derivation.

71. The same derivation procedure can be applied to an exterior differential form of any
degree. For example, let:
= aij [xi xj]

be a quadratic form, and consider the alternating bilinear form:

(, ) = aij (xi x j xj xi )

that corresponds to it, and introduce three differentiation symbols , , that are mutually
interchangeable. Finally, consider the expression:

( , ) (, ) + (, ),
70 Lessons on integral invariants

which obviously has an intrinsic significance that is independent of the choice of variables.
Upon doing this calculation, one easily confirms that it is reduced to an alternating trilinear
expression:

(, , ) = [aij ( xi x j x j xi ) aij (xi x j xj xi )

+ aij (xi x j xj xi ).
Now, a cubic exterior differential form:

a a a
= [ aij xi x j ] = ij + jk + ki [ xi x j xk ] ,
xk xi x j

which we shall call the derived form of , corresponds to that trilinear form.

72. In the case examined, it is important to account for the relation that exists between the
derivation of a quadratic exterior form and the operation that consists of passing from a double
integral taken over a closed surface to the triple integral taken over the volume bounded by
surface.
In order to do this, imagine that x1, ..., xn are functions of three parameters , , , and
consider an elementary parallelepiped in n-dimensional space whose edges are portions of the
coordinate lines, and whose vertices A, B, C, D, E, F, G, H correspond to the curvilinear
coordinates:

(, , ), (+, , ), (, + , ), (, , + ),
(+, + , ), (+, , + ), (, + , + ), (+, +, + ).

As one knows, the symbols , , refer to differentiations with respect to the three parameters
, , , respectively.
Now consider the curvilinear integral , which is taken over the surface that bounds this
parallelepiped.
The integrals that are taken over the three faces that contain A are, up to a sign:

( , ), ( , ), (, ),

and, in order for these integrals to be taken over all of the internal faces or all of the external
faces, it is necessary to take them to be either equal to the three preceding expressions or equal
and opposite. If we take them to be equal and opposite then the sum of the integrals taken over
the six faces will be:

( , ) ( , ) (, ) + [( , ) + ( , )]
+ [( , ) + ( , )] + [(, ) + (, )]
 Exterior differential forms and their derived forms 71

= ( , ) + ( , ) + (, ) = (, , ).

The surface integral is thus transformed into the volume integral .

In the simple case of three variables, if one sets:

= P [y z] + Q [z x] + R [x y]
then one will have:
P Q R
= [P y z] + [Q z x] + [R x y] = + + [x y z].
x y z

73. These considerations can be extended to exterior forms of arbitrary degree. Any exterior
form admits a derived form whose degree is greater by one unit and whose calculation is
extremely easy, since each term of the form:

A [xi xj ... xl]

gives rise to the derived term:
[A xi xj ... xl].

We note several formulas that are useful and easy to prove. If m is a coefficient that is a
finite function of the variables and is an arbitrary exterior form then one will have:

(m ) = [dm ] + m .

If and are two arbitrary exterior differential forms then one will have:

[ ] = [ ] [ ];

the + sign refers to the case in which is of even degree and the sign, to the case in which
is of odd degree. In particular, if is of even degree then the derived form of [p] will be given
by the ordinary formula:
[p ] = p [p1 ].

74. In the preceding, we supposed that the coefficients of the forms under consideration
were continuous functions that admitted partial derivatives of the first order. However, it is also
the case that one may still define an exterior derivative when the coefficients of a form do
not admit derivatives. A classic example is provided by potential theory.
We consider a material volume V that is bounded by a surface S. Let be the density at a
point of V. We suppose that the function is continuous. The potential U of that mass is a
continuous function on all of space that everywhere admits continuous first order derivatives.
72 Lessons on integral invariants

Regarding that function, there exists a theorem (Gausss theorem), which is expressed by the
formula:
U U U
x dy dx + y dz dx + z dx dy = 4 dx dy dz ;
the integral on the left-hand side is taken over an arbitrary closed surface and the one on the
right-hand side is taken over the volume that is bounded by that surface. From this, it results that
upon setting:
U U U
= [dy dz ] + [dz dx ] + [dx dy ]
x y z

one can define the exterior derivative of by:

= 4 [dx dy dz].

If the function U admits second-order partial derivatives then this will amount to the
classical Poisson formula, because the derivation procedure defined above will immediately
give:
2U 2U 2U
= 2 + 2 + 2 [dx dy dz ] .
x y z

However, one can still define the derivative if the function U does not admit second-order
partial derivatives, which is the general case when one does not make supplementary hypotheses
on the function .
One thus concedes the possibility of defining the exterior derivative as an autonomous
operation that is independent of classical derivation. There is then a direct proof of the formula
from the preceding section:
(1) [ ] = [ ] [ ] ,

in which one simply supposes that and are exterior-differentiable.

75. Take the simplest case of a linear form in two variables:

= P x + Q y
that admits an exterior derivative:
= R[x y] .

Suppose that the functions P and Q are continuous, and consider a function m that admits
continuous first-order partial derivatives. Here, the formula:

(m ) = m + [m ]
amounts to:
 Exterior differential forms and their derived forms 73

m m
m( P x + Q y) = mR + Q x P y x y.
The proof of this formula can be carried out very simply. Let A be the integration area, and
let C be the contour that bounds it. Divide the area A into a large number of partial areas; for
example, areas parallel to the axes. Take a point (x0, y0) in each of the partial areas, and call the
m m
values of the functions m, P, Q, , at that point:
x y

m m
m0 , P0 , Q0 , , .
x 0 y 0

In the interior, or on the contour of that area, one may set:

m m
m = m0 + ( x x0 ) + 1 + ( y y0 ) + 2 ,
x 0 y 0

P = P0 + 3, Q = Q0 + 4.

When the integral m(P x + Q y) is taken over the contour of that partial area, it will be
equal to:
m m
m ( P x + Q y ) +
( x x ) + ( y y ) ( P0 x + Q0 y ),
x 0 y 0
0 0 0

plus a quantity that is less than M l, upon denoting the upper limit of 1, 2, 3, 4, by , a fixed
number by M, the diameter of the area by , and the length of its contour by l. The sum of all
these supplementary quantities can obviously be made as small as one pleases, because l is of
order of the total area. As for the sum of the two integrals written above, it is equal to:

m m
m R + Q x P0 x y.
y 0
0 0
0

One easily derives the proof of the formula in question from this.
With analogous hypotheses, this proof may be extended to the case of a quadratic form:

= P [y z] + Q [z x] + R [x y].

The existence of the equality:

P y z + Q z x + R x y = H x y z
74 Lessons on integral invariants

implies that:

m m m
m( P y z + Q z x + R x y) = mH + P x + Q y + R z x y z.
The proof seems very difficult in the case of two linear forms in three variables:

= A x + B y + C z,
1 = A x + B y + C z.

Suppose that these two forms are differentiable and that one has, for example:

A x + B y + C z P y z + Q z x + R x y,
=

A x + B y + C z = P y z + Q z x + R x y.

Formula (1) then becomes:

(B C C B ) y z + (C A A C ) z x + (A B B A ) x y
= (P A + Q B + R C P A Q B R C) x y z.

It does not seem possible to prove this by the same procedure as in the preceding case
without adding supplementary hypotheses in order to this; for example, the hypothesis that the
functions A, B, C, A , B , C must satisfy a condition that is analogous to the Lipshitz condition.
It is interesting to study this question and see if it is really true that the differentiability of an
exterior product always results in the differentiability of its factors.
As for the question of knowing the conditions under which an exterior differentiable form is
differentiable, it is related, at least for forms of degree n 1 in n variables, to the theory of
additive functions on the set of C. de la Vale-Poussin (1). For example, the form = P [y z]
+ Q [z x] + R[x y] is differentiable if the sum of the integrals , when it is taken over the
surfaces that bound a finite number of cubes that are composed of planes that are parallel to the
coordinate planes, tends to zero when the sum of the volumes of these cubes tends to zero. The
function H that appears in this expression is the derivative:

= H [x y z],

which is not naturally continuous, in general.

In what follows, we shall always assume the legitimacy of the operations performed.

(1) See the book entitled: Integrales de Lebesgue, fonctions densemble, classes de Baire; Paris, Gauthier-
Villars, 1916.
 Exterior differential forms and their derived forms 75

III. - Exact exterior differential forms.

76. Now, here is an important theorem:

The derivative of the derivative of an arbitrary exterior differential form is identically

zero.
Indeed, take an arbitrary term in , say:

a [x1 x2 ... xp].

The corresponding term in is:
[a x1 x2 ... xp].

If a depends upon only x1, ..., xp then the latter term will be zero, and its derivative as well.
If, on the contrary, a is independent of x1, ..., xp then one can make a change of variables such
that a becomes equal to xp+1. The derivative of the term:

[xp+1 x1 x2 ... xp]

will then be zero, since the coefficient of that term is unity, so its derivative will contribute
nothing to the exterior derivative.
This theorem has a converse, namely:

If the derivative of a differential form is zero then the form can be regarded as the
derivative of a form whose degree is less than that of by one unit.

In order to prove this theorem, we shall appeal to the following lemma, which we will use
later on, moreover:

If the derivative of a form is zero, and if that form does not contain the differential xn then
its coefficients will be independent of xn .

Indeed, take a term in such as:

A [x1 x2 ... xp].

In the derivation, it will furnish the term:

[A x1 x2 ... xp],

which, upon developing, will give several terms, such as:

A
[ xn x1 x2 x p ] .
xn
76 Lessons on integral invariants

Obviously, the last term cannot be reduced any further, since no other term of contains xn
. Since = 0, one necessarily has:
A
= 0.
xn

Having thus proved the lemma, we return to our theorem. Let us call 0 what the form
becomes when one makes xn = xn0 and xn = 0. The derivative of 0 will obviously be zero if
that of is. Suppose then that the theorem has been proved for n 1 variables. It will be
possible to find a form 0 that is constructed from the variables x1, ..., xn1 , such that 0 is its
derivative:
0 = 0 .

If this is the case then separate the terms in the given form and the unknown form that
do not contain xn from those that do. One may write:

= 1 + [xn 2], = 1 + [xn 2].

If we calculate the terms in that contain xn then we will find that:

1
= xn [ xn 2 ] +
xn

Choose the form 2 arbitrarily, and determine 1 by the conditions:

1) 1 reduces to 0 for xn = xn0 .

2) One has:
1
= 2 + 2 .
xn
One thus obtains 1 by quadratures.

When the form is chosen in the manner that we just described, it will enjoy the following
properties:

1) The difference will not contain xn when one reduces the similar terms.
2) It will reduce to zero when one makes xn = xn0 in its coefficients.

We now remark that the derivative of that form is zero; consequently, from the lemma, all of
its coefficients will have values that are independent of xn . It is therefore identically zero, and
the theorem is proved.
 Exterior differential forms and their derived forms 77

The same proof shows that one may arbitrarily choose the terms in the form that contain
xn , arbitrarily choose the values for xn = xn0 in the terms that do not contain xn , but do contain
xn1 , arbitrarily choose the values for xn = xn0 , xn 1 = xn01 in the terms that contain neither xn
nor xn1 , but do contain xn2 , and so on.
Furthermore, it is quite clear that if one has a solution of the problem then all of the others
can be deduced from it by adding the derivative of an arbitrary form (whose degree is two units
less than that of ) to .

77. If is a linear form then, from the preceding theorem, the hypothesis that its exterior
derivative is zero will thus lead to the conclusion that was pointed out before that is an exact
differential. If is a quadratic form in three variables:

= P [y z] + Q [z x] + R [x y]
then the condition:
P Q R
+ + =0
x1 y z

will be necessary and sufficient for to be regarded as the derivative of a linear form, i.e., for
one to find three functions A, B, C that satisfy:

C B
= P,
y z
A C
= Q,
z x
B A
= R.
x y

Remark. If the coefficients of the form are uniform in a certain domain then the
condition = 0 will not always be sufficient to assure the existence of a form that is uniform
in that domain and has for its exterior derivative. For example, consider the two-dimensional
domain (closed and without boundary) that is composed of the points of a sphere , and let be
a form of degree 2 that is uniform in that domain (and has coefficients that admit continuous
partial derivatives of the first order). The derivative will obviously be zero. Nevertheless, if
there exists a linear form whose derivative is equal to then when one integrates
twice around the same great circle of the sphere in the two different senses one will obtain:

= 0,
78 Lessons on integral invariants

in which the integral is taken over the entire surface of the sphere. The preceding equation gives
a supplementary condition for to be regarded as the exact derivative of a form that is
uniform over all of the sphere.
 CHAPTER VIII

THE CHARACTERISTIC SYSTEM

OF AN EXTERIOR DIFFERENTIAL FORM.
FORMATION OF INTEGRAL INVARIANTS

I. - The characteristic system of an exterior differential form.

78. The results of the preceding chapter permit us to easily form the characteristic Pfaff
system of a given exterior differential form .
In order to do this, we remark that if is invariant for the system of differential equations:

dx1 dx2 dx
(1) = = ... = n
X1 X 2 Xn

then can be expressed in terms of n 1 independent first integrals y1, ..., yn1, and their
differentials; as a result, this will also be true for its derivative . As a consequence, the
system of linear equations (in total differentials) that is associated with the two exterior forms
and will be a consequence of the equations:

(2) dy1 = 0, ..., dyn1 = 0,

and, as a result, of equations (1).
In other words, in order for system (1) to admit () as an invariant form it is necessary that
the associated system of and be verified when one replaces the variables x1, ..., xn
with X1, ..., Xn .
Conversely, suppose this condition is true. Since the associated system of () is satisfied
on account of equations (1) it will also be satisfied on account of the equivalent equations (2).
Hence, when () is considered as an exterior form in the quantities dxi , it can be uniquely
expressed in terms of dy1, ..., dyn1 . Since the coefficients are functions of x, one may always
suppose that they are expressed by means of y1, ..., yn1 , and xn (if Xn 0). One will thus have:

= A i1 ...i p [dyi1 ...dyi p ].

Ai1 ...ip
When one forms , one will find that the only term in [dxn dyi1 ...dyi p ] has for its
xn
coefficient. Now, by hypothesis, can also be expressed by means of only the quantities dyi .
One thus has
Ai1 ...ip
= 0.
xn
80 Lessons on integral invariants

As a result, can be expressed by means of first integrals of the given system and their
differentials; it is therefore an invariant form.
It results immediately from this that the equations of the characteristic system of reduce to
the equations of the associated system of , combined with the equations of the associated
system of .

79. We shall now examine some important particular cases.

Suppose that is an exact form; i.e., = 0. In this case, the characteristic system of the
differential form coincides with the associated system of the form .
As an application, we look for the characteristic system of the (complete) relative integral

invariant . That relative invariant reduces to the absolute invariant . However is an
exact derivative. Hence, the characteristic system of a relative integral invariant coincides
with the associated system of the derived form .
This is just what happens with the linear integral invariant of dynamics:

= pi qi H t.
Here one has:
= [pi qi] [H t].

The associated system of the form is:

= 0, = 0, =0;
( qi ) ( pi ) ( t )

i.e., if we use the symbol d instead of :

H
dpi dt = 0,
qi
H
dqi dt = 0,
pi
H
dH dt = 0;
t

this is the same as what we found in chapter I (sec. 11).

Here, the differential form is quadratic. As a result, we know in advance that the number
of linearly independent equations of the associated system will be even; this explains why the 2n
+ 1 equations of the characteristic system reduce to 2n. Similarly, we will have an explanation
for what happens in hydrodynamics in relation to the invariant form:

[y z] + [z y] + [x y] + ( w v)[x t] + ( u w)[y t] + ( v u)[z t].

 Formation of integral invariants 81

Here n = 4; the characteristic system thus contains 4 or 2 or 0 independent equations. Now,

there cannot be 4, since the form is invariant for the differential equations of the trajectories of
the molecules. Hence, there are 2 or 0. There are 0 if = = = 0, i.e., if the motion is
irrotational. In the contrary case, one may expect, a priori, that the trajectories will not be the
only characteristic curves of the form.

80. One last important case is the one in which the form is of degree n 1. If it is
invariant for a system of differential equations then this system will necessarily be unique,
because the associated Pfaff system of will be composed of n 1 independent equations. In
order for the associated system of to contain no more than n 1 independent equations, it is
obviously necessary that must be zero. As a result, in order for a form of degree n 1 to
be invariant for a system of differential equations, it is necessary and sufficient that its derivative
must be identically zero.
A simple example is furnished by the integral invariant of the kinematics of continuous
media:
(x u t)(y v t)(z w t).
Here, the form is:

= [x y z] u [y z t] v [z x t] w [x y t].

The condition = 0 gives:

( u ) ( v ) ( w)
+ + + [ t x y z ] = 0 .
t x y z

This is the condition of continuity, or the law of conservation of matter:

( u ) ( v ) ( w)
+ + + = 0.
t x y z

One sees that this law of conservation of matter translates into the simple condition that the
derivative of the form that defines the quantity of elementary matter must be zero.

81. The conservation laws of physics can often be translated into analogous conditions. The
law of conservation of flux for a force field X, Y, Z translates into the condition that divergence
of that force field be zero, i.e.:
X Y Z
+ + = 0.
x y z

This simply expresses the idea that the derivative of the elementary force flux:
82 Lessons on integral invariants

= X [y z] + Y[z x] + Z[x y]
is zero.
Any (static) magnetic field satisfies this condition. The electromagnetic field, defined by
means of the exterior form:

= Hx [y z] + Hy [z x] + Hz [x y] + Ex [x t] + Ey [y t] + Ez [z t]

also satisfies the condition that is zero. One has:

H x H y H z H x E z E y
= + + [ x y z ] + + [ y z t ]
x y z t y z

H y E x Ez H z E y E z
+ + [ z x t ] + + [ x y t ] .
t z x t x y

Upon annulling the four coefficients of , one obtains the four classical equations, which
one may write, in vector notation:
div H = 0,
H
+ curl E = 0.
t

The form from hydrodynamics, which we have often considered already:

= [y dz] + [z y] + [x y] + ( w v)[x t] + ( u w)[y t] + ( v u)[z t],

also has a zero derivative, since is the derivative of the linear form of the quantity of motion-
energy, and the vectors (, , ) and ( w v, u v, v u) satisfy the same relations
as the magnetic field and the electric field. These two vectors are the vorticity, which plays the
role of magnetic field, and the vector product of vorticity with velocity, which plays the role of
electric force.
We remark that the electromagnetic field (or rather, the form that represents it) might not
be invariant for any system of differential equations, since [ 2] is not zero, in general. The
exception is when the magnetic field is perpendicular to the electric field. The characteristic
system will then be defined by the equations:

Hz dy Hy dz + Ex dt = 0,
Hx dz Hz dx + Ey dt = 0,
Hy dx Hx dy + Ez dt = 0,
Ex dx Ey dy Ez dz = 0,

which reduce to three. The system of differential equations that admits as invariant form will
then be:
 Formation of integral invariants 83

dx dy dz dt
= = = .
Hx Hy Hz 0

At any instant, it defines the lines of force of the magnetic field. Another is:

dx dy dz dt
= = = 2 .
E y H z Ez H y Ez H x Ez H x Ex H y E y H x H x + H y2 + H z2

If the magnetic field is zero then the characteristic manifolds will be defined by the
equations:
dt = 0, Ex dx + Ey dy + Ez dz = 0;

they will be the equipotential surfaces when considered at each instant t.

II. - Formation of integral invariants.

82. It is obvious that the exterior product of two invariant exterior forms is also an invariant
form. From this, the knowledge of an invariant exterior form implies the knowledge of any
other series of invariant forms, namely and all of the forms that are deduced from and
by exterior multiplication.
First, suppose that is an absolute invariant form of even degree. One will then have two
series of absolute invariant forms:

[ p], [ p1 ] (p = 1, 2, ...).

The derivative of a form of the first series will be a form of the second series. The derivative
of a form of the second series will be zero.
Now, suppose that one has a relative integral invariant , and suppose first that is of
even degree. One can deduce only one new invariant from that: viz., the absolute invariant .
On the contrary, if is of odd degree then one will have a series of relative integral
invariants p1 and a series of absolute integral invariants p . Moreover, the relative
integral invariant p 1 will reduce to the absolute invariant p by differentiation.
For example, this is true for the relative invariant of dynamics:

i =n

= pi qi H t.
i =1

The relative integral invariants that one deduces from it are:

84 Lessons on integral invariants

p 1
(p = 1, ..., n).

The absolute integral invariants are:

p
(p = 1, ..., n).

There thus exists a (relative or absolute) invariant of an arbitrary given degree that is less
than or equal to 2n.

83. One must not assume that the new invariants whose existence that we have just pointed
out are always the only ones that can be deduced (without integration) from a given invariant.
For example, suppose that one knows an invariant form that is reducible to the form:

= [1 2 3] + [4 5 6],

in which 1, 2, ..., 6 are six linearly independent (Pfaff) forms. Introduce six indeterminates
1, ..., 6, and consider the auxiliary quadratic form:

= 1 + 2 + ... + 6 .
1 2 6

It is obvious that if one regards the as quantities that are covariant in the then the form
will be covariant in . We say that this form is of rank 2. We obtain the conditions:

i = 0, (i = 1, 2, 3; = 4, 5, 6).

This gives two possible solutions: one for:

1 = 2 = 3 = 0,
and one for:
4 = 5 = 6 = 0.

The existence of two systems of three equations in Pfaff covariants results from this, namely:

1 = 2 = 3 = 0,
1 = 2 = 3 = 0.

As a result, the form [ 1 2 3] and the form [ 4 5 6] will themselves also be covariant. The
first one is obtained by taking into account in the equations of the second covariant system,
and the second one is obtained by taking the equations of the first covariant system into account.
Now, suppose that is expressed by means of first integrals of the system of equations, as
well as their differentials, and that is an invariant form. The formation of the two systems of
 Formation of integral invariants 85

Pfaff covariants will also work for the reduced form, and each of them will contain only first
integrals and their differentials. This will also be true for the two forms [ 1 2 3] and
[4 5 6], which are, as a result, invariant forms.

The existence of the integral invariant therefore implies the existence of each of the
integral invariants 1 2 3 and 4 5 6.

One verifies by an analogous argument that the existence of an invariant form of degree p > 2
that is reducible to a sum of h monomial terms such that the hp factors that enter into these terms
are linearly independent implies the property that each of these monomial terms must be an
invariant form.
This theorem is not true if p = 2.

84. In certain cases, the existence of an invariant form implies the existence of an invariant
equation. For example, consider the form:

= [1 2 5] + [3 4 5],

in which 1, ..., 5 denote five independent Pfaff forms. The only linear relation between these
forms that annihilates is obviously:
5 = 0.

This latter equation is thus invariant. It may be expressed by means of first integrals of
differential equations for which is an invariant form.
In a general manner, if is an invariant form, and if the associated system of is not
identical to its characteristic system then the associated system will be a system of Pfaff
invariants.
One may alter these considerations in various ways.

85. Once more, take the case of two quadratic invariant forms 1 and 2 that have the same
associated system. Let 2s be their common rank. The equation of degree s in :

[(1 2)] = 0,

which expresses the idea that the rank of the form 1 2 is less than 2s, obviously has an
invariant significance. The roots of the equation in are thus first integrals of the differential
equations that admit 1 and 2 as first integrals. One may show that, in the general case, 1
and 2 are reducible to the forms:

1 = 1 [1 2] + 2 [3 4] + ... + s [2s-1 2s],

2 = [1 2 ] + [3 4] + ... + [2s1 2s].

Each of the monomial forms [1 2], [3 4], ..., [2s-1 2s] is invariant.
 CHAPTER IX

DIFFERENTIAL SYSTEMS THAT ADMIT

AN INFINITESIMAL TRANSFORMATION

I. - The notion of an infinitesimal transformation.

85. A transformation in n variables is defined by a system of equations:

(1) xi = fi (x1, ..., xn) (i = 1, ..., n)

that can be solved for x1, ..., xn . Geometrically, if one regards x1, ..., xn as the coordinates of a
point M in the space of n dimensions then the transformation (1) takes an arbitrary point M of the
space to another point M by a well-defined law. The transformations are the ones that are
currently used in geometry (homothety, similarity, inversion, or, more simply, rotation,
translation, etc.).
The transformation (1) is called the identity when the right-hand sides reduce to x1, ..., xn ,
respectively; any point is then transformed into itself.
Given a system of differential equations:

dx1 dx2 dx
(2) = = ... = n ,
X1 X 2 Xn

this system is said to admit the transformation (1) when the application of this transformation to
different points of an arbitrary integral curve of (2) gives points that all belong to the same new
integral curve.
Consider a transformation that depends upon a parameter a that reduces to the identity for a
certain numerical value a0 of this parameter. Set a a0 = , and suppose that the right-hand side
can be developed in powers of :
xi = xi + i(x1, ..., xn) + ...

One will have what one may call an infinitesimal transformation by paying attention to only
the terms of first order in . An infinitesimal transformation is thus completely determined by
the n functions i of x1, ..., xn . One obtains the same infinitesimal transformation by multiplying
all of these functions by the same constant factor. We say that the function i represents the
increment of the variable xi by the infinitesimal transformation (in reality, the increment is i,
but the coefficient plays only an auxiliary role).
Given a function f(x1, ..., xn), the increment to which the infinitesimal transformation subjects
this function is, up to a factor of , the first term in the development:

f( x1 , ..., xn ) f(x1, ..., xn) = f(x1 + i , ..., xn + n) f(x1, ..., xn).

It is therefore:
 Differential systems and infinitesimal transformations 87

f f f
1 + 2 + ... + n .
x1 x2 xn

We denote this expression by the symbol Af:

f f f
(3) Af = 1 + 2 + ... + n .
x1 x2 xn

We agree to say that Af is the symbol of the infinitesimal transformation in question.

86. Formula (3) is analogous to the one that gives the total differential of a function f:

f f f
f = x1 + x2 + ... + xn .
x1 x2 xn

The only difference is that is the symbol of an undetermined operation, whereas A is the
symbol of a determined operation. The symbol of differentiation becomes the symbol of an
infinitesimal transformation when one gives the x1, ..., xn definite values (i.e., given functions
of the variables).
The operation symbolized by A is susceptible to being applied not only to finite functions,
but also to differential forms. For example, take the principal part (divided by ) of the
increment of dxi for A(dxi). Now, one has:

dxi dxi = dxi + ...

One is thus led to suppose that:
A(dxi) = dxi = d(Axi).

One sees by this that the operation A can be considered to be interchangeable with the
operation of (undetermined) differentiation.

87. Return now to the system of differential equations (2). This system will admit the
infinitesimal transformation (3) if the application of that transformation to the different points of
an arbitrary integral curve takes the points that are situated along it onto the same new integral
curve, up to second-order infinitesimals.
It is quite obvious that if the equations (2) admit a transformation that depends upon a
parameter a for any numerical value of this parameter then it will admit an infinitesimal
transformation that corresponds to the values of a that are infinitely close to the value a0 (if it
exists) that gives the identity transformation.
If there is a first integral of equations (2), and if these equations (2) admit an infinitesimal
transformation Af then it is clear that A(y) will also be a first integral. Indeed, y has some
numerical value c at any point M of an arbitrary integral curve (C). The function y is augmented
by A(y) at the point M that is the transform of M. This augmentation will be the same for any
88 Lessons on integral invariants

point M of (C). It is therefore necessary that A(y) have the same numerical value at all of the
points of (C). In other words, A(y) is a first integral.
Conversely, if the application of the operation A to an arbitrary first integral gives a first
integral again then the system (2) will admit the infinitesimal transformation Af. Indeed, if:

c1, c2, ..., cn1

are constant numerical values that are taken by n 1 independent first integrals:

y1, y2, ..., yn1

at the different points M of an integral curve (C) then the values that these integrals take at the
transformed points M will be the values that the functions:

y1 + A(y1), y2 + A(y2), ..., yn1 + A(yn1)

take at the points M themselves; they are therefore constant. As a result, the points M will truly
generate an integral curve.

II. - Formation of integral invariants when starting with infinitesimal transformations.

88. The preceding property show us that knowing an infinitesimal transformation Af that is
admitted by the differential equations (2) will permit us to deduce another invariant form from
the invariant differential form , namely A(). If the form is exterior then so is the form
A(), and the new form will have the same degree as the old one did.
There exists a second operation that permits us to deduce another invariant form from an
invariant exterior form . Suppose to fix ideas that is of third degree, and consider the
corresponding trilinear differential form (, , ) . Replace the symbol of undetermined
differentiation in this form with the symbol of the infinitesimal transformation. We obtain an
alternating linear form (A, , ) with two types of differentials , , to which there
ultimately corresponds a quadratic exterior form, which we designate by (A, ). This new form
is deduced from the first one by an operation that makes sense independently of the choice of
variables. If is expressed by means of first integrals yi of the equations (2) and their
differentials then the expression (A, ) will also be expressed in terms of the yi and the yi . As
a consequence, the operation that we just defined will permit us to deduce from any invariant
form another invariant form whose degree is diminished by one.
From this, one has:

(4) ( A, ) = 1 + 2 + ... + n .
( x1 ) ( x2 ) ( xn )
 Differential systems and infinitesimal transformations 89

89. The two new operations that we just defined are not independent of each other. Suppose
to fix ideas that is of second degree, and recall the definition of the exterior derivative .
One has (sec. 71):
(, , ) = ( , ) (, ) + (, )

for the condition that the three symbols , , be interchangeable. Replace the symbol with
the one for the infinitesimal transformation Af. We then have:

(A, , ) = A(( , ) ) (A, ) + (A, ) ;

i.e., upon passing to the exterior forms:

(, ) = (()) [(, ) ] ,
or, finally:
(5) (()) = (, ) + [(, ) ] .

This fundamental formula contains the result of the first operation that was performed on A
in its left-hand side. As for the two terms in the right-hand side, the first one can be obtained by
first applying the operation of exterior derivation to , followed by the second operation, which
is associated with Af. As for the second term, it is deduced from by the same operations, but
in the reverse order.
By definition, knowing an infinitesimal transformation Af that equations (2) admit puts us in
possession of an essentially new operation, which is defined by formula (4) and permits us to
deduce a new invariant form (, ) from an invariant form ().
In particular, we may remark that if y is a first integral then the first integral A(y) can be
obtained, first by differentiation, which gives () = y, and then by the application of operation
(4), which gives:
(A) = A(y).

III. - Examples.

90. Consider a continuous material medium in motion whose density is and whose velocity
components are u, v, w. As we saw in (sec. 37), the differential equations of motion for a
molecule:
dx dy dz
(6) = u, = v, =w
dt dt dt
admit the integral invariant:

(x y z u y z t v z x t w x y t),

which corresponds to the invariant form:

90 Lessons on integral invariants

= [x y z] u [y z t] v [z x t] w [x y t].

Suppose that the motion is permanent; i.e., that , u, v, w are independent of t. Equations (6)
do not contain time explicitly i.e., they do not change when one replaces t with t + so they
admit the infinitesimal transformation:

( A, ) = = u [ y z ] v [ z x] w [ x y ] .
( t )

The property of this form that it is invariant is physically obvious. Indeed, consider a tube of
trajectories, and cut this tube by two arbitrary surfaces that determine two areas S and S in the
interior of the tube. The quantity of matter that fills the volume situated between the lateral
surface of the tube and the two surfaces S and S is always the same, so the algebraic flux of the
matter traversing the surface that bounds this volume will be zero. Now, the flux that traverses
the lateral surface is zero. One will thus have:

(u x z + v z x + w x y) = (u y z + v z x + w x y).
S S

We remark that the invariant form (A, ) is an exact derivative. Indeed, its derivative if it
is not zero may differ from by only a finite factor. Now, this derivative does not contain t.
One thus has:
[(A, ) ] = 0.

As a result, the characteristic system of (A, ) reduces to its associated system. It is thus
given by the equations:
dx dy dx
= = .
u v w

It defines the trajectories of the molecules, but independently of the manner in which these
trajectories are described in time.
Formula (5) also shows that the form (A, ) has zero derivative. Indeed, here the form
is identically zero. On the other hand, since does not contain t explicitly, it does not change
when one changes t into t + , so A() is zero. This remark will be applied to the following
examples.

91. Now consider a perfect fluid in motion under the action of forces that are derived from a
potential. We have seen (sec. 22) that there exists an absolute invariant form:

= [y z] + [z x] + z [x y] + ( w v)[x t] + ( u w)[y t] + ( v u)[z t].

It is given by the exterior derivative of a linear form:

= u x + v y + w z E t,
 Differential systems and infinitesimal transformations 91

in which the coefficient E viz., the energy per unit mass is expressed by:

dp
E = 12 (u 2 + v 2 + w2 ) U + .

Suppose the motion is permanent; i.e., that u, v, w, p, are independent of t. As before, one
will have a new invariant form:

( A, ) = = (v w ) x + (w u ) y + (u v ) z
( t )

x y z
= u v w .

On the other hand, if one starts with the expression:

= [u x] + [v y] + [w z] [E t],
then one will find that:
(, ) = E .

As a result, E will be a first integral of the equations of motion. We recover Bernoullis

theorem, according to which the quantity:

1 2 2 dp
(u + v + w2 ) U +
2

remains constant along each streamline for a perfect fluid in permanent motion.
However, the form E is invariant, not only for the differential equations of motion of
molecular fluids, but also for the vortex lines, which also admit the invariant form . As a
result, the quantity E remains constant along not only each streamline, but also along each
vortex line.
If the motion is irrotational then the form (A, ), as we originally wrote it, will obviously
be identically zero. In this case, energy will be constant over all the fluid mass at any instant.
The equality:
x y z
E = u v w

permits us to represent the (spatial) variation of energy at each point M by means of a vector MH
that has this point for its origin, and which will be the vector product of the velocity vector (u, v,
w) with the vorticity vector (, , ). The derivative of energy in a given direction will be equal
to the projection of the vector MH on that direction.
92 Lessons on integral invariants

92. Another very general application relates to the problems of dynamics in which the
f
constraints and the forces are independent of time. The infinitesimal transformation Af =
t
that the equations of motion admit permits us to deduce from the fundamental integral invariant
of dynamics:
= pi qi H t,
the new integral invariant:
H

that is obtained by partial differentiation with respect to t. One therefore obtains the
generalized energy integral:
H=h

under the condition that the function H must be independent of time.

More generally, suppose that the function H does not contain one of the variables pi and qi
f
say qn , for example. The equations of motion thus admit the infinitesimal transformation ,
qn
from which one deduces the invariant linear form:

= pn .
( qn )

Therefore, if the function H does not contain one of the canonical variables then the
conjugate variable will be a first integral of the equations of motion.

IV. - Applications to the n-body problem.

93. Consider n mutually attracting material points that are subject to forces that are
proportional to their masses and inversely proportional to a given power of their distance. There
then exists a function of the forces:
mi m j
U = f p ,
i, j rij

in which the exponent p is a given (which is equal to 1 in the case of celestial mechanics), and
the quantity rij denotes the distance between two points Mi and Mj of masses mi and mj .
The equations of motion of the system admit a certain number of obvious infinitesimal
transformations. First, time is not explicitly contained in these equations. In addition, for any
solution of the problem, one may deduce another one from it by displacing the set in space, and
also by communicating a supplementary uniform rectilinear motion (which is the same for all
points) to each of the n points. One immediately deduces the existence of the following
infinitesimal transformations from this:
 Differential systems and infinitesimal transformations 93

f
A0 f = ,
t

f f f
A1 f = A2 f = , A3 f = ,
i xi yi zi

f f f f
A4 f = yi zi + yi zi , A5f = ..., A6f = ...,
zi yi zi yi

f f f f f f
A7 f = +t , A8 f = +t , A9 f = +t .
xi xi yi yi zi zi

The transformation A1 f corresponds to a translation parallel to Ox, the transformation Ai f, to

a rotation around Ox, and the transformation A7 f, to a supplementary motion of constant velocity
that is parallel to Ox.
Finally, one can point out one last infinitesimal transformation that is based upon
considerations of homogeneity. The equations:

d 2 xi U d 2 yi U d 2 zi U
mi = , mi = , mi =
dt 2 xi dt 2 yi dt 2 zi

remain unaltered if one multiplies all of the coordinates xi, yi, zi by the same constant factor ,
on the condition that t must be multiplied by 1+ p / 2 . The components xi, yi, zi of the velocity
are then multiplied by p / 2 . Upon taking = 1 + , one will arrive at the new infinitesimal
transformation:
f f f p f f p f
A10 f = xi + yi + zi xi + yi + 1 + t .
xi yi zi 2 xi yi 2 t

If we make the same definition of U then we remark that one will have:

A0U = A1U = ...= A9U = 0, A10U = pU.

94. Recall the fundamental second-degree integral invariant:

= m [ x xi] + mi [ y yi] + mi [ z zi] [ m ( x x + y y + z z )t] + [U t].

i i i i i i i i i i

Denote the linear form (Ai, ) by i. There then exist eleven invariant linear forms
0, 1, ..., 10 . This is easy to see, a priori, from formula (5), since the first ten are exact
differentials, because does not change under any given one of the first ten infinitesimal
transformations. As for 10, formula (5) gives:
94 Lessons on integral invariants

(10 ) = A10( ).

Now, has a degree of homogeneity (in the sense above) that is equal to 1 p / 2. One thus
has:
p
(10 ) = 1 ,
2

and 10 will be an exact differential only if p = 2; i.e., if the attraction is proportional to the cube
of the distance.
The calculation of the eleven forms i does not offer any difficulty and gives:

0 = m ( x x + y y + z z )t U = H,
i i i i i i

1 = m x = H1,
i i

2 = m y = 2,
i i

3 = m z = 3,
i i

4 = m (zi y yi z + y zi z yi
i i = H4,
i i i

5 = m ( xi z zi x + z xi x zi = 5 ,
i i i i i

6 = m ( yi x xi y + x yi y xi = 6 ,
i i i i i

7 = m (xi t x x t) = H7 ,
i i i

8 = m (yi t y y t) = 8 ,
i i i

9 = m (zi t z z t) = 9 ,
i i i

= m ( x x + y y + z z + x x + y y + z z )
10 i i i i i i i
p
2 i i
p
2 i i
p
2 i i

p
+ 1 + t H + pH t .
2
We have set:
H= m ( x + y + z
1
2 i i
2
i
2
i
2
) U,
H1 = m x , i i

H2 = m y , i i

H3 = m z , i i

H4 = m (y z z y ),i i i i i

H5 = m (z x x z ),i i i i i

H6 = m (x y y x ),i i i i i

H7 = m (x t x ), i i i

H8 = m (y t y ), i i i

H9 = m (z t z ). i i i
 Differential systems and infinitesimal transformations 95

p
One easily verifies that the bilinear covariant of 10 is equal to 1 + . If p = 2 then one
2
will have the new first integral:

m (xi x + yi y + zi z ) 2Ht = C,
i i i i

which will give:

m (x
i
2
i + yi2 + zi2 ) = 2 Ht 2 + 2Ct + C

upon integrating; this is the Jacobi integral.

The first integrals H1, H2, H3, H7, H8, H9 are the ones that give us the center of gravity
theorem. The first integrals H4, H5, H6 are the ones that give us the law of areas.

95. In the preceding section, we directly obtained only the differentials of the first integrals
H and not the integrals themselves. They must be given to us by applying the operation that
corresponds to the infinitesimal transformation Ai f to each of the invariant forms i. We have
therefore obtained invariant functions, i.e., first integrals:

ij = i(Aj) = (Ai, Aj) = ij ,

which we shall write in the form of a matrix with two indices that will be manifestly
antisymmetric. The calculations offer no difficulty. The quantity ij is found at the intersection
of the row i and the column j. The letter M denotes the sum of the masses of the n bodies.
96 Lessons on integral invariants

0 1 2 3 4 5 6 7 8 9 10

0 0 0 0 0 0 0 0 0 0 0 pH

1 0 0 0 0 0 H H2 M 0 0 3p H1

2 0 0 0 0 H3 0 H1 0 M 0 3p H 2

3 0 0 0 0 H2 H1 0 0 0 M 3p H 3

4 0 0 H3 H2 0 H6 H5 0 H3 H8 p
(1 )H 4
2

5 0 H3 0 H1 H6 0 H4 H9 0 H7 p
(1 )H 5
3

6 0 H2 H1 0 H5 H4 0 H8 H7 0 p
(1 ) H 6
3

7 0 M 0 0 0 H9 H8 0 0 0 H7

8 0 0 M 0 H9 0 H7 0 0 0 H8

9 0 0 0 M H8 H7 0 0 0 0 H9

10 pH p
3
H1
p
3
H2 p
3
H3 ( 3p 1)H4 ( 3p 1) H 5 ( 3p 1)H6 H7 H8 H9 0

We remark that the determinant of the elements of the preceding matrix is zero since it is an
antisymmetric determinant of odd degree. There thus exist eleven coefficients i that are not all
zero such that the expression i i becomes zero when one applies the operation to it that
relates to any of the transformations Ai f. One easily sees that 10 is zero. Calculation gives us
the expression i i , which is defined up to a factor of:

K 2 p H
+ ,
K p H
upon setting:

K = (MH4 + H2 H9 H3 H8)2 + (MH5 + H3H7 H1H9)2 + (MH6 + H1H8 H2H7)2.

K H
In the case of celestial mechanics, p = 1. The expression +
is the logarithmic
K H
differential of HK. This quantity HK is therefore invariant under all of the transformations Ai f.
 Differential systems and infinitesimal transformations 97

It is then easy to find an interpretation for it by making a convenient choice of coordinate axes.
If we take the center of gravity to be the origin, which is permissible, since it is animated with a
uniform rectilinear motion, then one will see that H1, H2, H3, H7, H8, H9 are annulled. The
aformentioned quantity is then H ( H 42 + H 52 + H 62 ) , up to a constant factor; i.e., it is the product
of the square of the kinetic moment of the system in its motion around the center of gravity with
the total energy of the system in this motion. This quantity is evidently independent of the choice
of axes and the choice of units.

V. - Application to the kinematics of rigid bodies.

96. Consider the motion of a rigid body with respect to three fixed rectangular axes. One
knows that at each instant it is defined by a system of vectors with the general resultant (p, q, r)
and the moment with respect to the origin (, , ). Suppose that these six quantities are given
functions of time. The differential equations of motion of a point of a rigid body are:

dx
= + qz ry = X ,
dt
dy
= + rx pz = Y ,
dt
dz
= + py qx = Z .
dt

These equations admit an obvious integral invariant. If one considers two infinitely close
points:
(x, y, z), (x + x, y + y, z + z)

of the rigid body at the instant t then the distance between these two points will not vary with
time. One thus has a differential form:
x2 + y2 + z2

that is invariant if one considers only points at the same instant, and which becomes invariant in
an absolute manner if one completes it by replacing:

x, y, z
with:
x X t, y Y t, z Z t,
respectively.
Let:
F = (x X t)2 + (y Y t)2 + (z Z t)2

be this invariant form, to which there corresponds the bilinear invariant:

F(, ) = (x X t)( x X t ) + (y Y t)( y Y t ) + (z Z t)( z Z t ).

98 Lessons on integral invariants

This bilinear form is not alternating, but symmetric. Nevertheless, the differential equations
of motion admit the infinitesimal transformation:

f
Af = .
t

As a result, one may deduce another invariant form from the form F, namely:

1 F
= X ( x X t ) Y ( y Y t ) Z ( z Z t ) .
2 ( t )

The same process may be repeated here, and this time it gives a first integral:

1 2 F
= X 2 +Y 2 + Z2 .
2 ( t ) 2

This first integral is obviously geometric. The motion of the rigid body is helicoidal, and the
preceding integral is equal to the square of the velocity of the point considered a velocity that
remains unchanged throughout the motion.

VI. - Differential equations that admit an infinitesimal transformation.

97. In the preceding examples, we supposed that an integral invariant was known. Now,
suppose that one knows only an invariant equation for example, the equation:

() a1x1 + a2x2 + ... + an xn = 0.

We assume that this equation is invariant; i.e., that it may be written in such a manner that it
contains only first integrals y1, ..., yn1 of the given differential equations and their derivatives. In
other words, one has:
() [b1(y)y1) + b2(y)y2 + ... + bn1(y)yn1],

in which the bi depend upon only the y1, ..., yn1 , and is an arbitrary function. If one replaces
the symbol of indefinite differentiation with the symbol for the infinitesimal transformation Af
then one will immediately have:

( ) b1 ( y ) y1 + b2 ( y ) y2 + ... + bn1 ( y ) yn1

= ,
( A) b1 ( y ) Ay1 + b2 ( y ) Ay2 + ... + bn 1 ( y ) Ayn1

and the right-hand side will obviously be an invariant linear form.

 Differential systems and infinitesimal transformations 99

The knowledge of an infinitesimal transformation Af that a given system of differential

equations admits and the knowledge of a Pfaff equation () = 0 that is invariant for this system
( )
implies the knowledge of a linear integral invariant .
( A)

For example, suppose that one is dealing with an ordinary differential equation:

dx dy
= .
X Y

It is invariant by itself. As a result, if it admits an infinitesimal transformation:

f f
Af = +
x y

then it will also admit the invariant linear form:

X y Y x
.
X Y

Here, since there is only one first integral, that form is necessarily an exact differential. In other
words, one knows an integrating factor for the equation; this result is classical.
Most of the differential equations that one knows to be integrable follow from the preceding
remark. This is the case for the equations:

dy dy dy y
= f ( x ), = f ( y ), = f .
dx dx dx x

For example, the last of these equations does not change if one multiplies x and y by the same
constant factor 1 + . It thus admits the infinitesimal transformation:

f f
Af = x +y .
x y
As a result, the expression:
y
dy f dx
x
y
y f x
x

is an exact differential. This property becomes obvious if one sets:

y = ax,
because then the expression becomes:
100 Lessons on integral invariants

du dx
+ .
u f (u ) x

The integration of that exact differential leads to the same calculations as in the classical method.

98. Finally, if one knows nothing a priori about a given system of differential equations then
the knowledge of an infinitesimal transformation that this system admits allows us to obtain an
invariant Pfaff system. Indeed, we look for all of the Pfaff equations = 0 that are
consequences of the given differential equations and are such that (A) is zero. If one sets:

() = 1 x1 + 2 x2 + ... + n xn

then the coefficients i will be given by the two conditions:

1 X1 + 2 X2 + ... + n Xn = 0,
1 1 + 2 2 + ... + n n = 0.

The desired set of equations thus forms a Pfaff system that is obtained by annulling all of the
determinants with three lines and three columns in the matrix:

x1 x2 xn
X1 X2 Xn .
1 2 n

This system has a significance that is independent of the choice of variables. Now, if one
takes the n 1 first integrals y1, ..., yn1, and one nth variable to be the variables then the
equations will reduce to:
y1 y2 yn1
= = = (i = Ayi ) .
1 2 n1

The Pfaff system under consideration is therefore invariant, and obviously it is completely
integrable, since it reduces to a system of ordinary differential equations in y1, ..., yn1.
For example, if the equations:
dx dy dz
= =
X Y Z
admit the infinitesimal transformation:

f f f
Af = + +
x y z
then the total differential equation:
 Differential systems and infinitesimal transformations 101

dx dy dz
X Y Z =0

will be completely integrable. Upon integrating it, one will obtain a first integral of the given
equations. After equating this first integral to a constant, one will be left with an ordinary
differential equation that admits a known infinitesimal transformation, which can be integrated
by quadrature.

VIII. - Expressing the idea that a given system of differential equations

admits a given infinitesimal transformation.

99. We have not yet indicated the analytical conditions that express whether a given system
of differential equations:
dx1 dx2 dx
(2) = = = n
X1 X 2 Xn

admits a given infinitesimal transformation:

f f f
(3) Af = 1 + 2 + + n .
x1 x2 xn
Set:
f f f
(5) Xf = X1 + X2 ++ Xn .
x1 x2 xn

Our problem is then basically one of expressing the idea that if f is a first integral, ie., if it
satisfies the equation:
Xf = 0,

then Af must also be a first integral. In other words, it amounts to expressing the idea that the
equation:
X(Af) = 0
must be a consequence of the equation:
Xf = 0.
We may substitute the equation:
X(Af) A(Xf) = 0,

for the first equation, which contains the second-partial derivatives of f. As an easy calculation
f f
shows, this new equation is linear and homogeneous with respect to , , . The desired
x1 xn
condition is therefore, quite simply, the existence of an identity of the form:
102 Lessons on integral invariants

(7) X(Af) A(Xf) = Xf,

in which denotes a conveniently chosen coefficient.

This condition is obviously verified if one uses the infinitesimal transformation whose
symbol is Xf. This transformation displaces each point M of space along the integral curve that
passes through that point; it thus leaves each integral curve invariant. If one directs ones
attention to the effect that is produced on the integral curves, which are considered as indivisible,
then this particular infinitesimal transformation will play the same role as the identity
transformation. One will easily see that the effect of applying the infinitesimal transformations
that were defined in this chapter vanishes in this particular case. The same remark applies to the
infinitesimal transformation Xf, in which is a given arbitrary factor.

VIII. - Equations of variation.

100. The notion of an equation of variation is due to H. Poincar. One may relate it to the
notion of infinitesimal transformation.
Consider a system of differential equations, which we write as:

dx1 dx
(8) = X 1 , , n = X n ,
dt dt

in which the right-hand sides are given functions of x1, ..., xn, t. Let:

(9) x1 = f1(t), x2 = f2(t), ..., xn = fn(t)

be a particular solution of this system. Take an infinitely close solution:

x1 = f1(t) + x1 , x2 = f2(t) + x2 , ..., xn = fn(t) + xn ,

in which is an infinitely small constant, and the are unknown functions of t. If we neglect the
infinitely small terms of second order then we will obtain the following as a definition of the
unknown functions:
d i X i X X
(10) = 1 + i 2 + + i n (i = 1, 2, , n) .
dt x1 x2 xn

These are the equations of variation relative to a particular solution under consideration.
It may be the case that one knows a particular solution of the equations of variation
independently of the particular solution of the given equations that serve to define the equations
of variations. The quantities 1, ..., n are then, in reality, definite functions of x1, ..., t that
satisfy the partial differential equations:

1 X X X
(11) + X 1 i + + X n i = 1 i + 2 i + + n i .
t x1 x2 x1 x2 xn
 Differential systems and infinitesimal transformations 103

In this case, the given equations obviously admit the infinitesimal transformation:

f f f
Af = 1 + 2 + + n .
x1 x2 xn

This transformation is expressed by the equations:

x1 = x1 + 1,
...
xn = xn + n ,
t = t.

The transformed curve of the integral curve (9) has for its equations:

xi + i = fi(t)
or
xi = fi(t) i .

It is also an integral curve since (1, ..., n) constitutes a solution of the equations of
variation.
More generally, any solution (i) of equations (11) corresponds to an infinitude of
infinitesimal transformations that leave the given system (8) invariant, namely, the
transformations:
f f f f f
(12) Bf = 1 + + n + + X1 + + X n ,
x1 xn t x1 xn

in which is an arbitrary function.

Conversely, suppose that one knows an infinitesimal transformation that leaves the given
system invariant; it may always put into the form (12). The integral curve (9) is then changed by
this transformation into the curve:
xi + i + Xi = fi(t + ),
or
xi = fi(t) i + [ fi (t ) Xi] = fi(t) i .

As a consequence, the equations of variation (11) will admit the solution (1, ..., n).
All of these properties ultimately result from the fact that equations (11) define an analytical
f
translation of relation (7) only when the coefficient of in Af is zero.
t
 CHAPTER X

COMPLETELY INTEGRABLE PFAFF SYSTEMS

I. - Frobeniuss theorem.

101. A system of h Pfaff equations:

1 a11dx1 + a12 dx2 + + a1n dxn = 0,

(1)
a dx + a dx + + a dx = 0,
h h1 1 h2 2 hn n

is completely integrable when it may be put into the form:

(2) dy1 = dy2 = ... = dyh = 0.

From this, it follows that each form i is linear in dy1, ..., dyh , and, as a result, that i is
zero, on account of (2), i.e., on account of (1).

In order for a Pfaff system to be completely integrable it is necessary and sufficient that the
derivatives of its left-hand sides all be annulled on account of the system equations.

In order to prove the converse, we first remark that the property that we just stated does not
depend on the choice of variables and does not depend on the choice of the r left-hand sides. In
other words, if one writes the equations of the system in the form:

i ii 1 + 2 2 + ... + ih h = 0,
i

...
h hi 1 + h2 2 + ... + hh h = 0

then the derivatives 1, 2 , , h will also be annihilated on account of the system equations.
One has:
i = i1 1 + i2 2 + ... + ih h + [di1 1] + [di2 2] + ... + [dih h] ,

and each of the terms on the right-hand side is annihilated, by hypothesis, under the indicated
conditions.
Having said this, we suppose that the converse has been proved up to n 1 variables, and
prove it for n variables. The i are annulled on account of equations (1), and a fortiori if one
also makes dxn = 0. As a result, if one regards xn as a fixed parameter then system (1) is
reducible to the form:
 Completely integrable Pfaff systems 105

d y1 = 0,
...
dyh = 0,

in which y1, ..., yn are h independent functions of x1, ..., xn1, but which might also contain the
parameter xn . Now, if one no longer regards xn as constant then the system will obviously be
reducible to the form:
1 dy1 + b1 dxn = 0,

(3)
dy + b dx = 0,
h h h n

in which b1, ..., bh are functions of y1, ..., yh , and, for example, of xh+1, ..., xn . One has,
moreover:
1 = [db1 dxn], ..., h = [dbh dxh].

On account of equations (3), these formulas reduce to:

bi bi
i = [dxh +1 dxn ] + + [dxn1 dxn ] .
xh+1 xn1

The hypotheses thus imply as a consequence that the coefficients bi depend upon only the y1,
..., yh , xn . But then equations (3) constitute a system of ordinary differential equations that may
be reduced to the form:
dz1 = 0, ..., dzh = 0.

The symbols z1, ..., zh denote h independent first integrals.

102. The preceding theorem, which is due to Frobenius, permits us (sec. 64) to express the
necessary and sufficient conditions for the complete integrability of the given system by means
of the relations:
[1, ..., h 1 ] = 0, ..., [1, ..., h h ] = 0.
.
For example, take a Pfaff equation in three variables:

P dx + Q dy + R dz = 0.

The condition for complete integrability is:

R Q P R Q P
[ ] ( Pdx + Qdy + Rdz ) dydz + dzdx + dxdy
y z z x x y
106 Lessons on integral invariants

R Q P R Q P
P +Q + R [dx dy dz ] = 0 .
y z z x x y

II. - Forming the characteristic system of a Pfaff system.

103. One may give another form to the argument that was presented above by seeking the
characteristic Pfaff system of a given arbitrary system (1) in a general manner.
In order for a system such as (1) to be invariant for the differential equations:

dx1 dx
(4) == n ,
X1 Xn

it is necessary and sufficient that the equations of (1) can be expressed by means of first integrals
of (4) and their differentials. Therefore, it is first necessary that 1, ..., h are annihilated on
account of (4), and it is therefore necessary that the forms:

[1, ..., h 1 ] , ..., [1, ..., h h ]

can be expressed by means of differentials of the first integrals of (4). In other words, it is
necessary that the associated system of the forms:

1, 2..., h , [1, ..., h 1 ], ..., [1, ..., h h ]

be a consequence of equations (4).

Conversely, if this condition is realized, and if y1, ..., yn1 are first integrals of (4) then one
will be left with equations of the form:

1 dy1 + b1, h+1 dyh+1 + ... + b1, n1 dyn1 = 0,

...
h dyh + bh, h+1 dyh+1 + ... + bh, n1 dyn1 = 0.

bi , h+1 bi , n1
The form [i , ..., h i ] does not involve dxn , so the derivatives , ,
will all
xn xn
be zero. As a result, equations (1) can be written in such a manner as to involve only the first
integrals of system (4) and their differentials. Hence, system (1) is indeed invariant for equations
(4).
It results from this that the characteristic system of (1) is nothing but the associated system of
the forms:
1, 2..., h , [1 ... h 1 ], ..., [1 ... h h ] .

In particular, in order for the system to be completely integrable, it is necessary and sufficient
that this system be identical with (1), i.e., that the forms:
 Completely integrable Pfaff systems 107

[1 ... h 1 ] , ..., [1 ... h h ]

be identically zero.

104. One may also obtain the equations of the characteristic system of (1) in the form:

1 = 0, 2 = 0, , h = 0,

i i

i
(dx1 ) (dx2 ) (dxn )
=0.
ai1 ai 2 ain


ah1 ah 2 ahn

In particular, the characteristic system of a Pfaff equation:

a1 dx1 + a2 dx2 + ... + an dxn = 0

is given by the equations:

a1dx1 + a2 dx2 + + an dxn = 0,
a12 dx2 + a13 dx3 + + a1n dxn a21 dx1 + a23 dx3 + + a2n dxn
= =
a1 a n
an1 dx1 + + an ,n 1dxn 1
= ,
an

upon setting:
a j ai
aij = .
xi x j

We shall return to this system later in Chapter XIV.

III. - The integration of a completely integrable Pfaff system.

105. We now return to a completely integrable Pfaff system, which we write:

dz1 = a11dx1 + + a1q dxq ,

(5)
dz = a dx + + a dx .
h h1 1 hq q
108 Lessons on integral invariants

The integration of this system amounts to the integration of a system of ordinary differential
equations in h unknowns z1, ..., zh of one independent variable x1.
Indeed, we know that the system admits one and only one solution that corresponds to the
given initial values ( xi0 , z 0j ) . In order to find the values of the unknown functions z1, ..., zh that
correspond to a given system of numerical values x11 , , x1q of the independent variables, we
displace ourselves on the integral manifold from the point ( xi1 ) to the point ( xi1 ) , and follow the
variation of the zj . For any such succession of intermediary values of the independent variables,
the result will always be the same. For example, set:

x2 x20 = m2 ( x1 x10 ), , xq xq0 = mq ( x1 x10 ) ,

in which m2, ..., mq denote the q 1 quantities:

xi1 xi0
mi = 1 (i = 2, , q) .
x1 x10
We have:
dz1 = (a11 + a12 m2 + + a1q mq ) dx1 ,

(6)
dz = (a + a m + + a m ) dx .
h h1 h2 2 hq q 1

It will suffice for us to integrate this system of ordinary differential equations and determine
the solution that corresponds to the values z10 , , zh0 of the unknown functions for x1 = x10 . Once
this solution is determined, we will obtain z11 , , z1h by replacing the parametric quantities m2 , ...,
mq with the values that were indicated above.
We remark that the knowledge of one first integral of a system of ordinary differential
equations (6) in q 1 parameters mi does not unavoidably imply the knowledge of a first integral
of the Pfaff system (5).

IV. - Complete systems.

106. Return to the completely integrable system (1), in which we denote a system of h
independent first integrals by y1, ..., yh . Arbitrarily choose n h linear differential forms:

h+1 , ..., n

that are mutually independent and independent of the forms 1, ..., h . Any linear form in dx1,
..., dxn may be expressed in only one manner as a linear function of the 1, ..., n . Now take an
undetermined function f and consider its total differential:
 Completely integrable Pfaff systems 109

f f f
df = dx1 + dx2 + + dxn .
x1 x2 xn

One may express it linearly by means of 1, ..., h , whose coefficients are obviously linear
f f
and homogeneous in , , . Let:
x1 xn

(7) df = X1f 1 + X2f 2 + ...+ Xnf n .

f f
The n expressions Xif are linearly independent in , , .
x1 xn
Having said this, any first integral of the completely integrable system (1) is characterized by
the property that its differential, which is considered to be a linear form in dx1, ..., dxn , is
annulled under the one condition that equations (1) are verified; in other words, by the property
of annulling:
Xh+1 f, ..., Xn f.

The system of n h linearly independent partial differential equations:

(8) Xh+1 f = 0, ..., Xn f = 0

thus admits h linearly independent solutions y1, ..., yh .

Identity (7) gives us that:

(9) dyi = X1yi 1 + X2 yi 2 + ...+ Xn yi n (i = 1, 2, ..., h).

Since the yi are independent functions, the right-hand sides of equations (9) will be linearly
independent combinations of the 1, 2, ..., h . As a result, system (1) will be equivalent to (2).
Hence, it will be completely integrable.
We agree to say that the system of equations (8) forms a complete system if it admits the
maximum number h of linearly independent solutions. We see that there is a complete system
that corresponds to any completely integrable Pfaff system, and conversely. The correspondence
is such that if the equations of the Pfaff system are:

1 = 2 = ... = h = 0

then the equations of the complete system will be:

Xh+1 f = 0, Xh+2 f = 0,..., Xn f = 0.

107. It is easy to find conditions for a given system of linear first-order partial differential
equations to be complete.
Start with identity (7) and exterior differentiate it. We easily get
110 Lessons on integral invariants

h=n i = n j =n
(9) X
h =1
h f k + X i ( X j f )[i j ] = 0.
i =1 j =1

The n covariants k can be expressed as quadratic exterior forms in the 1, ..., n . Let:

1,..., n
(10) k = c
( ij )
ijk [ i j ].

If we equate all of the [i j] terms in identity (10) to zero then we will find that:

k =n
(11) X i ( X j f ) X j ( X i f ) + cijk X k f = 0.
k =1

Note the duality between formulas (10) and (11).

Now suppose that system (1) is completely integrable. From Frobeniuss theorem, this
signifies that 1 , ..., h will be annihilated, along with 1, ..., h . In other words, one will have:

ch+1, h+j, k = 0 (i, j = 1, ..., n h; k = 1, ..., h).

As a result, from (12), the combinations:

Xh+i (Xh+j f) Xh+j (Xh+i f)

will depend linearly upon only the Xh+1 f, ..., Xn f. The converse is obvious.
We agree to denote the combination X(Yf) Y(Xf) by (XY). One sees that the necessary and
sufficient condition for such a system to be complete is that the brackets of all the left-hand sides,
when taken two at a time, must be linear combinations of these left-hand sides.
 CHAPTER XI

THE THEORY OF THE LAST MULTIPLIER

I. - Definition and properties.

108. Consider a system of differential equations:

dxi dyi dzi

(1) = Xi , = Yi , = Zi
dt dt dt

that admits an integral invariant of maximum degree n:

= M [(x1 X1 t)(x2 X2 t) ... (xn Xn t)].

As we have seen (sec. 80), the condition for this to be the case is that the exterior derivative
must be zero, which gives, by an easy calculation:

M ( MX 1 ) ( MX 2 ) ( MX 3 )
(2) + + + ... + =0.
t x1 x2 x3

The coefficient M is known by the name of the Jacobi multiplier.

As we have seen, condition (2) expresses the idea that the form can be expressed by means
of n independent first integrals:
y1 , y2 , ..., yn

of the system (1) and their differentials. In other words, that one has an identity:

(3) M [(x1 X1 t)(x2 X2 t) ... (xn Xn t) = H(y1, y2, ..., yn)[y1, y2, ..., yn].

It is now possible for us to recover the classical theorems that relate to the Jacobi multiplier.

THEOREM I. The quotient of two multipliers M and M is a first integral.

Indeed, the two identities (3) that relate to the two multipliers M and M give:

M H ( y1 ,..., yn )
= .
M H ( y1 ,..., yn )
112 Lessons on integral invariants

THEOREM II. - If one knows p independent first integrals of equations (1) then one can
determine a multiplier of the system of n p differential equations that reduces the integration
of the given system.

Suppose that one knows p independent first integrals y1, y2, ..., yp , and suppose, as is always
permissible, that they are independent functions of p variables x1, x2, ..., xp , i.e.:

D ( y1 ,..., y p )
0.
D ( x1 ,..., x p )
Equations (1) may then be written:

dy1 dy p
(4) = 0, , = 0,
dt dt

dx p +1 dxn
(5) = X p +1 , , = Xn ,
dt dt

and if one equates y1, ..., yp to arbitrary constants C1, ..., Cp then the integration of system (1)
will reduce to that of system (5), in the right-hand side of which one is supposed to replace x1, ...,
xp with their values as functions of xp+1, ..., xn , t, C1, ..., Cp .
This suggests that the form , which is invariant for equations (4) and (5), can obviously be
written:
= N [(y1 ... yp (xp+1 Xp+1 t) ... (xn Xn t)].

In order to obtain the value of the coefficient N, it is necessary to identify that expression
with the original expression. For example, by equating the terms in:

[x1 x2 ... xn],

one will obtain:
D( y1 ,..., y p )
M =N .
D ( x1 ,..., x p )

Having thus determined the quantity N, one has the identity:

N [(y1 ... yp (xp+1 Xp+1 t) ... (xn Xn t)] = H [y1...yp yp+1...yn],

i.e.:
[y1 ... yp {N(xp+1 Xp+1 t) ... (xn Xn t) Hyp+1...yn}] = 0.

This identity expresses the idea (sec. 64) that if one takes into account the linear relations:

y1 = 0, y2 = 0, ..., yp = 0
then one will have that:

(6) N [(xp+1 Xp+1 t) ... (xn Xn t)] = H (y1, ..., yn)[yp+1...yn].

 The theory of the last multiplier 113

The left-hand side of this equality is therefore an invariant form for the system of differential
equations (5). In other words, the system (5) admits the multiplier:

M
N= .
D ( y1 ,..., y p )
D ( x1 ,..., x p )

THEOREM III. - If one knows n 1 independent first integrals of equations (1) then the
integration of the equations is achieved by a quadrature.

It suffices to apply Theorem II in the case of p = n 1. One then sees that the linear
differential form:
M
( xn X n t )
D ( y1 , , yn1 )
D( x1 , , xn1 )

is an exact differential when one supposes that the variables are coupled by relations:

y1 = C1, y2 = C2, ..., yn1 = Cn1 .

The general solution of equations (1) is thus obtained by equating the integral of a total
differential:
M
D( y1 ,..., yn 1 ) (dxn X n dt )
D( x1 ,..., xn 1 )
to a constant.

II. - Generalizations.

109. The theorem of the last multiplier can be generalized to the considerably more general
case in which one knows an invariant form of arbitrary degree r < n. Suppose that one knows
n 1 independent first integrals y1, ..., yn1 . Choose n r of these integrals:

y1 , y 2 , , y n r

in all possible ways, and consider the obviously invariant forms:

[ y1 y 2 y n r ] .
114 Lessons on integral invariants

All of these forms are of degree n. If they are not all zero then we shall revert to the
preceding case that we just studied. We have a multiplier. We may even have several, and in
certain cases Theorem I can give the last first integral by dividing two of these multipliers.
The exceptional case is the one for which all of the preceding forms are zero. Now, imagine
that is expressed by means of y1, ..., yn1 and the differential of an (unknown) nth first
integral yn . The hypothesis made amounts to saying that does not contain yn , because if
contains, for example, a non-zero term such that:

A[y1 ... yr1 yn]

then the exterior product of with yr+1yr+2 ... yn1 would not be zero
If this is the case then will be an exterior form in y1, ..., yn1 , which is an exterior form
whose coefficients that one may calculate. Each of these coefficients will be a first integral. If at
least one of these coefficients is independent of y1, ..., yn1 then one will achieve the integration
by equating that coefficient to an arbitrary constant. The only doubtful case is the one for which
all of the coefficients are functions of y1, ..., yn1 . Now it is clear that in this case the knowledge
of the invariant form might not be of any help in achieving this integration. We simply
remark that in this case the given equations do not constitute the characteristic system of .
We may therefore state the following general theorem:

The knowledge of an invariant differential form that admits the given system (1) of
differential equations as its characteristic system permits us, in the most unfavorable case, to
achieve the integration of this system by a quadrature when one already knows n
1 independent first integrals.

110. Another generalization of Jacobis theory of the last multiplier relates to completely
integrable Pfaff systems. Let:
1 = 0, 2 = 0, ..., r = 0

be a completely integrable system for which one knows an invariant form of maximum degree r:

= M [1 2 ... r].

The knowledge of r 1 first integrals y1, ..., yr1 of the system permits us to achieve the
integration by a quadrature. Indeed, upon equating y1, ..., yr1 to arbitrary constants, the given
system will reduce to only one equation for example, r = 0 and one will have a formula
such as:
= N [y1 ... yr1r],

in which the coefficient N can be deduced from M by an easy identification. It results from this
that Nr is an invariant form for the single equation that remains to be integrated; viz., r = 0. In
other words, Nr is an exact differential. The integration is then achieved by a quadrature.
 The theory of the last multiplier 115

Finally, the completely general theorem that summarizes all of the cases envisioned is the
following one:

The knowledge of a differential form permits us, in the most unfavorable case, to achieve
the integration of the characteristic system of that form by quadrature when one already knows r
1 independent first integrals, where we have denoted the class of that form by r.

III. - Case where the independent variable is not specified.

111. If the given system of differential equations were put into the form:

dx1 dx2 dx
= = = n
X1 X 2 Xn

then any integral invariant of degree n 1 would be of the form:

= MX1 [dx2 dx3 ... dxn] MX2 [dx1 dx3 ... dxn] + ... (1)nMXn [dx1 dx2 ... dxn1],

and the condition = 0 would become:

(MX1 ) ( MX 2 ) ( MX n )
+ + + =0.
x1 x2 xn

Apart from the difference described, this theory would be identical to the one that was described
above.

IV. - Case in which the given equations admit an infinitesimal transformation.

112. Take the general case of a completely integrable system:

(7) 1 = 0, 2 = 0, ..., r = 0

that admits an invariant form of degree r, which one may suppose is reduced to the form:

= [1 2 ... r].

Suppose that this system admits a known infinitesimal transformation Af, and form the
quantities:
1(A), 2(A), ..., r(A),
116 Lessons on integral invariants

which we suppose are all zero. One may always suppose that the equations of the system are
written in such a manner as to make:

(8) 1(A) = 1, 2(A) = 3(A) = ... = r(A) = 0,

while keeps the same form.

As we have seen (sec. 88), the knowledge of the infinitesimal transformation Af permits us to
deduce another invariant form (A, ) from the form (), which, with the hypotheses made
here, reduces to:
(A, ) = [2 ... r].

We denote this new form by the letter . We have that:

= [1 ].

The associated (and non-characteristic) system of is:

(9) 1 = 0, 2 = 0, ..., r = 0.

It is completely integrable. This results from an earlier theorem (sec. 98), but also from the fact
that since is expressible by means of r first integrals y1, ..., yr of the given system and their
differentials, the associated system to (, ), like , contains only y1, ..., yr , and their
differentials. As a result, it will be a system of ordinary differential equations. Hence, it is
completely integrable.
Now, form the exterior derivative of the form . It is a new invariant form of degree r.
One therefore has:
= m = m [1 ].

The coefficient m is a first integral. However, there are some cases that we must discuss:

1. m = 0: is zero, and the system (9) that is associated to is its characteristic system.
One therefore knows a multiplier of system (9). As a result, when one knows r 2 independent
first integrals of this system, the integration will be achieved by a quadrature. A second
quadrature will then achieve the integration of the given system (7). This quadrature is
obviously 1.
It is obvious that here and are reducible to:

= [y2 y3 ... yr], = [y1 y2 ... yr].

When the transformation Af is applied to the first integrals of the given system, it will reduce
to:
f
Af = .
y1
 The theory of the last multiplier 117

There are an infinitude of ways of choosing the first integrals in such a way that the givens
remain the same, i.e., in such a manner that and Af do not change. One may perform an
arbitrary transformation on y2, y3, ..., yr of functional determinant 1, and add an arbitrary function
of y2, ..., yr to y1 . This explains the nature of the simplifications that are presented in the
integration.

2. m is a non-zero constant: In this case, we suppose that one has integrated system (9), and
let y1, ..., yr be a system of r 1 independent integrals. One will have:

= H[y2 y3 ... yr],

in which the coefficient H is independent of y2, ..., yr (otherwise would be zero), but H is still a
first integral of the given system. One thus obtains an rth integral of the given system by simple
differentiations.
If we write y1 in place of H then we have that:

1 f
= [ y1 y2 yr ] , = y1[ y2 yr ] , Af = my1 .
m y1

The most general transformation in y1, ..., yr that preserves the given data is obtained by
performing an arbitrary transformation on y2, ..., yr and setting:

y1
y1 = .
D( y2 , , yr )
D( y2 , , yr )

This explains why the integration of system (9) cannot be simplified and also why, once that
integration has been performed, the integration of the system (7) can be deduced from it.

3. The coefficient m is not constant, but A(m) is zero: The function m is a first integral of
system (9). The integration of this system amounts to that of a system of differential equations in
r 2 unknown functions. The integration of the given system is deduced from it as in the
preceding case.
The form is reducible to:
= y1[my3 ... yr],
and one has:
1 f
= [ y1 m y3 yr ] , Af = my1 .
m y1

The transformations that preserve the givens are:

118 Lessons on integral invariants

m = m, y3 = f3 (m, y3 yr ), , yr = f r (m, y3 , , yr ),
y1
y1 = .
D ( f 3 , , f r )
D ( y3 , , yr )

This explains the simplifications that are presented in the integration.

4. The coefficient m is not constant and Am = m1 0 : In all of what follows, take the
general case:
Am = m1, Am1 = m2, ..., Ami-1 = mi ,

and suppose that m, m1, ..., mi1 are i independent first integrals of the given system, and that mi is
a function of m, ..., mi1 .
The given system then admits i known independent first integrals, and its integration
amounts to the integration of a system of differential equations in r 1 unknown functions when
one knows a multiplier.
We look for the reduced form of and Af. One may always set:

= H [m m1 ... mi1 yi+1 ... yr],

in which yi+1, ..., yr are r i first integrals of the system (9) and H is a function of m, ..., mi1, yi+1,
..., yr . Obviously, one has:
f f f
Af = m1 + m2 + + mi .
m m1 mi 1

We express the fact that the exterior derivative of = (, ) is equal to m, or, what
amounts to the same thing:
A() = m.
One has:
mi
A() = ( A( H ) + H )[ m m1 mi 1 yi +1 yr ] .
mi 1
One will thus have:
mi
A( H ) + H = mH .
mi 1

Let h(m, m1, ..., mi1) be a particular solution of that partial differential equation. The latter
may be written:
H
A = 0 .
h

In other words, H / h is an integral of equations (9). One may then choose yi+1, ..., yr in such
a manner as to reduce that function to unity. One will thus have:
 The theory of the last multiplier 119

= h(m, ..., mi1)[m m1 ... mi1 yi+1 ... yr],

f f f
Af = m1 + m2 + + mi .
m m1 mi 1

The transformations that preserve the given data are obviously:

m = m, m1 = m1 , , mi 1 = mi 1 ,
yi +1 = f i +1 (1 , , i 1 , yi +1 , , yr ), ,
yr = f r ( 1 , , i 1 , yi +1 , , yr ),
with:
D ( yi +1 , , yr )
= 1.
D( yi +1 , , yr )

(We have let 1, ..., i1 denote i 1 independent functions of m, m1, ..., mi1 that satisfy A = 0.)
The nature of the preceding transformations explains the simplifications that are presented by
the integration.
However, it may be the case that i = r. In this case, no integration needs to be performed,
since one has r independent first integrals by definition.

V. - Applications.

113. The theory of the last multiplier can be applied to all of the indicated preceding
examples that involve an invariant form whose degree is equal to the number of unknown
functions. Recall these cases:

1. The equations that give the motion of the molecules in a continuous medium when one
knows the density and the components u, v, w of the velocity as functions of x, y, z, t:

dx dy dz
= u, = v, = w.
dt dt dt

Since the integral invariant is:

= [(x u t)(y v t)(z w t)] ,

the multiplier will be . Therefore, if one knows two independent first integrals then the
integration is achieved by a quadrature.
If the motion is steady then the invariant form:

= ( , ) = u [y z] v [z x] w [x y]
120 Lessons on integral invariants

has zero derivative. The equations that give the geometric trajectories:

dx dy dz
= =
u v w

admit a multiplier r. As a result, if one knows a first integral then the determination of the
trajectories will require only a quadrature, and a final quadrature gives t.

2. The equations that give the vortex lines of a given vector field (X, Y, Z) are the
characteristic equations of the form:

[X x] + [Y y] + [Z z]

Z Y X Z Y X
= [ y z ] + [ z x ] + [ x y ] .
y z z x x y
The equations:
dx dy dz
= =
Z Y X Z Y X

y z z x x y

thus admit a known multiplier, which is unity.

3. The equations of dynamics, in their canonical form:

dqi H dpi H
= , = ,
dt pi dt qi

admit the multiplier 1. This results from a direct calculation. It also results from this that the
existence of the invariant form:

i =n
H H
= [( qi t )( pi + t )]
i =1 pi qi

implies the existence of the invariant form n:

1 n i =n H H
= [( qi t )( pi + t )] .
n! i =1 pi qi
 The theory of the last multiplier 121

114. However, the preceding theory of the last multiplier applies not only to material
systems for which the canonical equations of Hamilton are valid, but also to any system with
perfect holonomic constraints whose given forces depend upon only the position of the system.
For such a system, one has Lagranges equations:

d T T
= Qi (q1 , , qn , t ) .
dt qi qi

If the Qi are zero then the introduction of Hamiltons canonical variables will lead to the
equations:
dqi H dpi H
= , = .
dt pi dt qi

It results from this that the complete equations of motion are susceptible to being put into the
form:
dqi H dpi H
= , = + Qi .
dt pi dt qi

They admit the multiplier 1. In other words, they admit the invariant form:

H H H H
= [( q1 t ) ( qn t )( pi + Q1 t ) ( pn + Qn t )] ,
p1 pn qi qn

which can be written in terms of the Lagrange variables:

i=n
T T
= [( qi qi t )( + Qi t )] .
i =1 qi qi

If the constraints, as well as the given forces, are independent of time then the equations of
motion will admit the infinitesimal transformation:

f
Af = ,
t

and, as a result, they will also admit the invariant form = (, ), whose derivative is zero.
From the general theorem, the integration of the equations of motion is reduced to the integration
of the (geometric) equations of trajectories:
122 Lessons on integral invariants

T
d
dqi qi dqi dpi
= or = ,
qi T H H
+ Qi + Qi
qi pi qi

to which the theory of the last multiplier may be applied, and a quadrature gives time. Indeed,
one obviously has, for example:
q
= t i ,
qi
since (, ) is equal to .

115. As an example of forces that depend on time, but still admit a known infinitesimal
transformation, consider the simple case of a point that is moving on a fixed line and is attracted
to a fixed point on the line according to a force that is proportional to the distance, with the
proportionality factor being a known function of time. The motion is given by the second-order
differential equation:
d2x
= k (t ) x ,
dt 2
or by the system:
dx dx
(10) = x , + k (t ) x = 0 .
dt dt

The second-order equation does not change if one changes x into x, with being an
arbitrary constant factor. As a result, the system to which it is equivalent admits the infinitesimal
transformation whose effect is to change:
x, x , t
into
(1 + ) x, (1 + ) x , t,

respectively. The symbol of this transformation is:

f f
Af = x + x .
x x
System (10) admits the invariant form:

= [(x x t)( x + kx t)],

which corresponds to the multiplier 1. Here, the derived form (, ) is:

= x( x + kx t) x (x x t) = x x x x + ( x2 + kx2)t.
 The theory of the last multiplier 123

It is an invariant form. Its exterior derivative is:

= 2[x x ] + 2 x [ x t] + 2kx [x t] = 2.

Since the coefficient of in the right-hand side is constant, we have (sec. 112) that it
suffices to integrate the completely integrable equation = 0 in order to deduce the general
solution of the given system by differentiations. The form is indeed reducible to y1 y2. This
form may be written, upon changing into d:

x x 2
= x2 d + 2 + k dt .
x x
If one sets:
x
=u
x

then one will thus be led to integrate the Ricatti equation:

du
+ u2 + k = 0 .
dt

Suppose we have integrated that equation. We have a first integral in the form:

(t )u + (t ) (t ) x + (t ) x
y2 = = .
(t )u + (t ) (t ) x + (t ) x

After identifying with y1 dy2, we find, upon taking, for example, the terms in dx :

x = y1 x ,
( x + x )2
from which:
( x + x) 2
y1 = .

If one supposes as is permissible that the determinant is equal to 1 (or simply

constant) then the general solution of the system will be furnished by the equations:

x + x = C1,
x + x = C2,
and one has that:
x = C2 (t) C1 (t).
124 Lessons on integral invariants

In other words, the coefficients (t) and (t) that present themselves in the general integral of
the Ricatti equation constitute a system of fundamental solutions to the given second-order
equation.
Other situations may also present themselves. Suppose we know the general solution of the
Ricatti equation, expressed in terms of t and the integration constant y2 . The identity:

y1 dy2 = x2 [du + (u2 + k) dt]

gives:
u
y1 = x 2 ,
y2

from which we obtain x as a function of y1, y2, and t. Since we have:

y2
u= ,
y2 +
we will obtain:
x2 = y1 ( y2)2,
from which, we will get:
x = C1 + C2 .

116. REMARK. The theory of Jacobis last multiplier can be applied to other problems in
mechanics than the ones that were pointed out above. For example, take the motion of a material
point that is subject to a force that is a function of only its position in space, but the system of
reference is associated with a uniform rotational motion around Oz. The equations of motion
are of the form:
d2x dy
2
+ 2 X = 0,
dt dt
d2y dx
2
2 Y = 0,
dt dt
d2z
Z = 0,
dt 2

in which X, Y, Z are given functions of x, y, z, t. If we write them as:

dx dx
= x = 2 y + X ,
dt dt
dy dy
= y = 2 x + Y ,
dt dt
dz dz
= z = +Z
dt dt
 The theory of the last multiplier 125

then we will obtain a system that obviously admits the multiplier 1.

117. The final application that we shall envision is furnished by the integral invariant of
hydrodynamics:

= [y z] + [z x] + [x y] + ( w v)[x t] + ( u w)[y t] + ( v u)[z t].

The characteristic system of this invariant is composed of the two Pfaff equations:

dx wdt dy vdt dz wdt

= = .

The integral manifolds in spacetime (x, y, z, t) are two-dimensional manifolds that are generated,
for example, by a vortex line in its different successive positions.
The integration of this system amounts to that of a system of two differential equations in
two unknown functions when one knows a multiplier. The search for molecular trajectories (i.e.,
the streamlines) requires, in addition, the integration of an ordinary differential equation that
might be arbitrary.
If the motion is steady then the characteristic manifolds will be given by two quadratures,
namely:
dx dy dz
u v w =C,

and then if we take the preceding equation into account, which we assume to be solvable in terms
of z, then we will get:
dx dy
t+ = C .
v u
 CHAPTER XII

EQUATIONS THAT ADMIT

A RELATIVE LINEAR INTEGRAL INVARIANT

I. - General method of integration.

118. Consider a Pfaff form and the characteristic system of the relative integral
invariant . This is the associated system of the form .
First, suppose that has 2n + 1 variables. Since the form has even rank (sec. 59), its
characteristic system will be composed of 2n equations, in general. As a result, there will exist,
in general, one and only one system of differential equations that admits a relative integral
invariant , where is an arbitrary Pfaff form in 2n + 1 variables; this is the case for the
integral invariant of dynamics.
In a general manner, let 2n be the rank (or the class) of the form . It is easy to indicate a
method of integration for the characteristic equations of .
Indeed, let y1 be a first integral of these equations (it is obtained by an operation of order 2n).
If one couples the variables by the relation y1 = C1, i.e., if one couples the differentials by the
relation dy1 = 0, then the rank of will diminish, and as it always remains even, it will reduce to
2n 2. Let y2 be a first integral of the new characteristic system. If we suppose that:

y1 = C1, y2 = C2

then the rank of reduces to 2n 4, and so on. Thus, by successive operations of order:

2n, 2n 2, ..., 4, 2,
one will find n first integrals:
y1, y2, ..., yn ,

such that if one supposes that the variables are coupled by the relations:

y1 = C1, y2 = C2, ..., yn = Cn

then the rank of will become zero. At that point, since is identically zero, the form will
be an exact differential. A quadrature will then put it into the form:

= dS.

The function S depends on n constants C1, ..., Cn . If one no longer supposes that the
variables are related by the n indicated relations then one will obviously have:
 Equations with a relative linear integral invariant 127

= dS + z1 dy1 + z2 dy2 + ... + zn dyn ,

and, as a result:
= [dz1 dy1] + [dz2 dy2] + ... + [dzn dyn].

Since is of rank 2n, the 2n differentials dyi and dzi will be linearly independent. Hence, the
2n functions yi and zi will constitute a system of independent first integrals of the given
equations, and the integration is thus accomplished.
By definition, the integration requires n + 1 operations of order:

2n, 2n 2, ..., 4, 2, 0,
according to the differentiations.

REMARK I. The quantity S serves as only an intermediary here. In general, it is not a first
integral of the characteristic equations of the invariant .

REMARK II. One sees from the result that was just obtained that any quadratic exterior
form with zero exterior derivative can be put into the form:

[dz1 dy1] + [dz2 dy2] + ... + [dzn dyn] .

119. It is important for us to account for the indeterminacy in the choice of functions yi and
zi that enter into the canonical form. The equality:

[dz1 , dy1] + [dz2 , dy2 ] + ... +[dzn , dyn ] = [dz1 dy1] + ... + [dzn dyn]

implies the property that the difference:

z1 dy1 + z2 dy2 + ... + zn dyn (z1 dy1 + z2 dy2 + ... + zn dyn)

must be an exact differential dV. Suppose, as is the general case, that y1 , ..., yn are independent
functions of z1, ..., zn . There is then no relation between the yi and the yi . If one expresses V as
a function of the yi and the yi then one will deduce that:

V V
zi = , zi = .
yi yi

These equations, which involve an arbitrary function of 2n arguments, permit us to express

the y and the z as functions of the y and z. As a result, the latter n give y1 , ..., yn , and the first n
give z1 , ..., zn . This supposes that one does not have:
128 Lessons on integral invariants

V V V
D , , ,
y1 y2 yn =0.
D( y1, y2 , , yn )

Under the same condition, one can express the y as a function of the y and the z by means of
the first n equations and thus obtain the z by means of the last equations.
One will similarly treat the case where there exist one or several relations between the y and
the y .
The set of transformations thus defined on the variables y and z i.e., on the integral curves
of the given equations defines an infinite group that plays the same role in this theory that the
group of transformations of functional determinant equal to 1 plays in the theory of the Jacobi
multiplier.

120. Recall the integration of the characteristic equations of . Suppose that, by an

arbitrary procedure, we come to know N > n independent first integrals y1, y2, ..., yn , such that
upon equating them to arbitrary constants, the rank of becomes zero, i.e., becomes an exact
differential. A quadrature that corresponds to the differentiations then gives:

= dS + z1 dy1 + z2 dy2 + ... + zn dyn .

It is easy to see that z1, z2, ..., zn are first integrals.

Indeed, suppose that among the functions yi and zi there are N + r independent ones. One
may then express the functions zi as functions of the yi and r of the others, which we call t1, ..., tr .
Having assumed this, we will then have:

= [dz1 dy1] + [dz2 dy2] + ... + [dzN dyN].

By hypothesis, the characteristic system of is composed of the equations:

dy1 = 0, dy2 = 0, ..., dyN = 0.

It also consists of the equation:

z z z
dzi + 1 dy1 + 2 dy2 + + N dy N = 0 ;
[dyi ] yi yi yi
hence, the equation:
dzi = 0.

One sees from this that the zi are first integrals, and, on the other hand, that N + r will be
equal to 2n.
 Equations with a relative linear integral invariant 129

Finally, the knowledge of N first integrals makes an exact differential when one equates
them to arbitrary constants and permits us accomplish the integration by one quadrature and
differentiations.

121. In practice, it may happen that one does not seek all of the solutions of the given
differential equations, but only the ones for which the N first integrals y1, ..., yN , have given
numerical values. One may then proceed in the following manner: Since the form is zero
when one annulls dy1, ..., dyN it can be put into the form:

= [dy11] + [dy22] + ... + [dyN N]

in an infinitude of ways, with the i being conveniently chosen linear forms. Among these N
forms i , there are 2n N of them that are independent of each other and independent of the dyi
. Suppose that the same is true for 1, ..., 2nN . The characteristic system of is obviously
composed of the equations:
dy1 = dy2 = ... = dyN = 0,
1 = 2 = ... = 2nN = 0.

If we express the idea that the exterior derivative of is zero then we will obtain:

[dy1 1 ] + [dy2 1 ] + ... + [dyN N ] = 0,

from which, in particular, upon exterior multiplying by [dy2dy3 ... dyN], we will obtain:

[dy1 dy2 ... dyN 1 ] = 0.

The form 1 (as well as the forms 2, ..., 2nN) is thus an exact differential when one gives
fixed numerical values to the yi . As a result, the desired solutions are obtained by 2n N
independent quadratures:
1 = 1, ..., 2nN = 2nN .

There is no reason to be surprised at the fact that we have encountered 2n N quadratures

here when the search for the general solution requires only one quadrature. Indeed, after
performing the 2n N quadratures that were indicated above, we then perform the single
quadrature:
11 + 22 + ... + 2nN 2nN = const.

that has 2n N arbitrary parameters, 1, ..., 2nN .

The preceding integration procedure uses only the invariant form , and does not involve
the form . Thus, the things that play an essential role here are the knowledge of the absolute
integral invariant of the second degree and the property of the form that it is an exact
derivative. The form (or the forms ), whose derivative is , plays only an auxiliary role.
130 Lessons on integral invariants

II. - The Poisson brackets and the Jacobi identity.

122. Let 2n be the rank of the exterior derivative , and let f and g be two first integrals of
its characteristic system. The two differential forms:

[ n 1 df dg] and [ n ]

are invariants of maximum degree 2n. They differ by only a factor, and this factor is a first
integral. We set:
1 1
[ n 1df dg ] = ( f g )[ n ]
(n 1)! n!
or
(f g)[ n ] = n [ n 1 df dg].

The quantity (f g) thus defined bears the name of Poisson bracket. It is an alternating bilinear
form in the partial derivatives of f and g.

The bracket of two first integrals is again a first integral.

This theorem, which is due to Poisson in the particular case of canonical equations, had its
importance pointed out by Jacobi.
Before going on to the applications of this theorem, we make several remarks.
The condition (f g) = 0 expresses the idea that the rank of is equal to 2n 4. In this case,
one says that the integrals f and g are in involution.
If this condition is not satisfied then the defining formula of (f g) will express the idea that
the form:
[df dg ]

( f g)

is of rank 2n 2. Indeed, the nth power of this form is:

n
[ n ] [ n 1df dg ] = 0 .
( f g)

Again, we remark that if one reduces to its normal form:

= [1 2] + [ 3 4] + ... + [2n1 2n],

and if one sets:
df = f1 1 + f2 2 + ... + f2n 2n ,
dg = g1 1 + f2 2 + ... + g2n 2n ,
then one will have:
(f g) = f1 g2 f2 g1 + f3 g4 f4 g3 + ... + f2n1 g2n f2n g2n1 .
 Equations with a relative linear integral invariant 131

We finally remark that from what was said above (sec. 118), one may always suppose that
the i are exact differentials. An easy calculation then gives the following identity, which is due
to Jacobi:
((f g) h) + ((g h) f) + ((h f) g) = 0,

which is applied to three arbitrary first integrals f, g, h.

However, the verification can be carried out without any restrictions on the linear forms 1,
2, ..., 2n . It is based upon the identity:

1 1
(1) [ n1 (( f g ) dh + ( g h) df + (h f ) dg ] = [ n 2 df dg dh] ,
(n 1)! (n 2)!

which is nothing but identity (8) that was proved in sec. 68. Upon exterior differentiating and
remarking that the exterior derivative of the right-hand side is zero, we obtain:

[ n 1 d(f g) dh] + [ n 1 d(g h) df] + [ n 1 d(h f) dg] = 0,

which is nothing but the Jacobi identity.

123. The method of integration that was pointed out at the beginning of this chapter can be
stated in terms of the Poisson brackets. Let:
Xf = 0

be the equation that expresses the idea that f is a first integral. One first seeks a particular
solution y1 of that equation. One then seeks a particular solution y2 of the system:

Xf = 0, (y1 f) = 0,

then a particular solution y3 of the system:

Xf = 0, (y1 f) = 0, (y2 f) = 0,

and so on, up to a particular solution yn of the system:

Xf = 0, (y1 f) = 0, (y2 f) = 0, ..., (yn1 f) = 0.

In the case of the canonical equations of dynamics:

dqi H dpi H
= , = ,
dt pi dt qi

which correspond to the invariant form:

132 Lessons on integral invariants

i =n
= [ pi qi ] [ H t ],
i =1

the partial differential equation in the first integrals of the given equation is:

f i = n f H f H
+ = 0.
t i =1 qi pi pi qi

As for the Poisson brackets (f g) of the two first integrals, they are defined by the equality:

n[ n 1 df dg] = (f g)[ n ].
Equating the terms in:
[p1 q1 p2 ... pn qn]
on both sides, we obtain:
i=n
f g g f
( f g) = .
i =1 pi qi pi qi

The partial differential equation in the first integrals can also be written by means of the
definition of the bracket (f g) of two arbitrary functions of qi , pi , and t as:

f
( H f ) = 0.
t

III. - Use of known first integrals.

124. Now, we shall once more take up the problem of integrating the characteristic equations
of the differential form by supposing that a certain (arbitrary) number of first integrals y1 , y2 ,
..., yp are known. When one equates these integrals to arbitrary constants C1, C2, ..., Cp , the
form will have its rank reduced by a certain even number 2 p 2p of units. It then suffices to
integrate the characteristic equations of that new form or rather, to find n p first integrals in
involution. One is then reduced to the problem of sec. 120.
The preceding method does not generally give us all the possible sets of known integrals.
Indeed, from the Poisson-Jacobi theorem, the brackets of p given integrals, when taken two at a
time, are themselves first integrals of the equations being integrated. One will then form the
brackets (yi yj), and then, if they provide new integrals, one forms the brackets of these integrals
amongst themselves and the given integrals, and so on, until the operation gives us nothing new.
This amounts to saying that one may always, by prior differentiations, suppose that the brackets
(yi yj) are functions of the first integrals y1, y2, ..., yp .
Now, in order to know how many units it takes to reduce the rank of when one supposes
that the variables are coupled by relations:
 Equations with a relative linear integral invariant 133

y1 = C1 , y2 = C2 , ..., yp = Cp ,

it will suffice to apply the theorem of sec. 69 to the exterior quadratic form that is constructed
from the variables x1, ..., x2n+1, which are coupled by the relations:

y1 = 0, y2 = 0, ..., yp = 0.

In this case, the coefficients aij of sec. 69 are the brackets (yi yj), and the quadratic form is:

= ( yi y j )[i j ] .

The number of units by which the rank of is reduced is equal to the maximum number 2p
diminished by the rank of the form .

125. One may account for the fact that all of the possible sets have been obtained from the
given first integrals in the following manner.
Perform a linear substitution (with coefficients in y1, ..., yp) on the p variables x1, ..., xp in
such a way as to reduce to its normal form:

= [1 2 ] + + [ 2 q 1 2 q ] (2q p).

This amounts to replacing the linear forms y1, ..., yp with new differential forms:

1, ..., p ,

which are linear in y1, ..., yp with coefficients that are functions of y1, ..., yp , and are such that
one has identically:
1 y1 + 2 y2 + + p y p = 1 1 + 2 2 + + p p .

The exterior quadratic form will then take the form:

= [12] + ... + [2q12q] + [2q+11] + ... + [pp2q] + [p2q p2q+2] + ... + [2np12np],

upon introducing 2n p new linear forms 1, ..., 2np .

Let denote the form:
= [12] + ... + [2q12q],

and express the idea that the exterior derivative of is zero. If we neglect all of the terms that
contain one of the linear forms:
2q+1 , ..., p ; p2q+1 , ..., 2np
then we will get:
(2) + [ 2 j +11 ] + + [ p p 2 q ] = 0 .
134 Lessons on integral invariants

As the form is constructed from only the functions yi and their differentials, the same is
true for . As a result, no reduction of the similar terms can be made between the different
parts of the left-hand side of (2). In particular, it results from this that each of the forms:

2q +1 , , p

will be zero (upon supposing that the forms 2q+1, ..., p are zero). As a result, the Pfaff system:

2q+1 = ... = p = 0

is completely integrable. We denote a system of first integrals of these equations by:

y2q +1 , , y p .
Furthermore, one always has:
= 0

if one regards the forms 2q+1, ..., p as zero. In other words, if one supposes that
the y2q +1 , , y p are constant then the form will be an exact derivative, and, as a result (sec.
118), it will be reducible to:
= [dy1 dy2 ] + + [dy2 q1 dy2q ] .

Finally, one easily sees that one can put into the form:

(3) = [dy1 dy2 ] + + [dy2 q1 dy2 q ] + [dy2 q+1 1 ] + +[dy p p 2 q ] + [ p 2 q +1 p 2 q + 2 ] +

This ultimately amounts to the following theorem:

One can find p functions:

y1 , y2 , , y p

of the given p first integrals that satisfy the conditions:

( y1 y2 ) = = ( y2 q 1 y2 q ) = 1,

while all of the other brackets ( yi y j ) are zero.

126. Apart from the intrinsic interest of this theorem, its form clarifies the fact stated above
that the indicated method of integration has educed all of the possible consequences of the given
integrals. The form (3) that was found for permits us to write:

= dS + y1 dy2 + + y2 q 1 dy2 q + w1 dy2q +1 + + w p 2 q dy p + v1 du1 + + vn p+ q dun p + q ,

 Equations with a relative linear integral invariant 135

= [dy1dy2 ] + + [dy2q 1 dy2 q ] + [dw1 dy2 q +1 ] + +[ w p 2 q dy p ] + [dv1 du1 ] + + [dvn p + q dun p + q ] .

The most general group of transformations on the integral curves that preserves the givens,
i.e., that leaves , y1, ..., yp invariant, is defined by the following equations, in which the primed
letters indicate the transformed variables and V denotes an arbitrary function of the arguments ui ,
ui, y2 q +1 , , y p :
yi = yi (i = 1, 2, , p ),
V V
v1 = , , vn p+ q =
u1 un p + q
V V
v1 = , , vn p + q =
u1 un p + q
V V
w1 = w1 + , , wp 2q = w p 2q + .
y2q +1 y p 2 q

Any unambiguous procedure that permits us, upon starting with and p first integrals y1, ...,
yp , to deduce another first integral by operations that have a significance that is independent of
the choice of variables necessarily leads to a first integral that is invariant under the most general
group of transformations that preserve , y1, ..., yp . However, the only functions that are
invariant under this group are obviously arbitrary functions of y1, ..., yp .

IV. - Generalization of the Poisson-Jacobi theorem.

127. The Poisson-Jacobi theorem is immediately generalized if, instead of two first integrals,
one knows two invariant linear forms 1 and 2. The quantity that is defined by the equality:

(4) n [ n 1 12] = [ n ]

is obviously a first integral. It reduces to (y1 y2) if 1 and 2 are the differentials of the two first
integrals y1 and y2.
We apply this remark to the case in which, assuming that the characteristic equations of
admit two infinitesimal transformations A1f and A2f, one has:

1 = (A1, ), 2 = (A2, ).

In order to calculate the quantity in this case, we apply the operation that takes an invariant
form () into the form (A1, ) to both sides of equality (4). One obtains:

n (n 1) [ n 2 i12] n [ n 1 12(A1)] = n [ n 1 1],

from which, since the form [ n 1 1] is certainly not zero:

136 Lessons on integral invariants

= 2(A1) = (A1, A2) = 1(A2).

When the generalized Poisson-Jacobi theorem is applied to the two invariant forms (A1,
) and (A2, ) it leads to the first integral (A1, A2) that is furnished by twice applying the
operations that correspond to the infinitesimal transformations A1f and A2f.
 CHAPTER XIII

EQUATIONS THAT ADMIT

AN ABSOLUTE LINEAR INTEGRAL INVARIANT

I. - General method of integration.

128. Let be a linear differential form. Its bilinear covariant is of even rank, namely 2n.
Two cases may be presented, according to whether the equation = 0 does or does not belong to
the characteristic system of . We shall first occupy ourselves with the latter case.

I. Obviously, one may set:

= [1 2] + ... + [2n1 2n].

The 2n + 1 forms , 1, ..., 2n are independent. In this case, the characteristic equations of
are (sec. 78):
= 1 = 2 = ... = 2n = 0.

One may easily exhibit a reduced form for . Indeed, from operations of order:

2n, 2n 2, ..., 2,
we successively find n first integrals:
y1, y2, ..., yn

of the characteristic equations of , and reduce its rank to zero when we equate them to
arbitrary constants. A quadrature puts into the form:

= du + z1 dy1 + z2 dy2 + ... + zn dyn .

This is the desired reduced form, which is obtained by operations of order:

2n, 2n 2, ..., 2, 0,

and which, once obtained, gives the general solution of the characteristic equations of .
In this case, one sees that the integration of the characteristic equations of and the
integration of the characteristic equations of are two equivalent problems, and the fact that
is an absolute integral invariant has no more importance in the integration than if were a
relative integral invariant. This is true at least when one follows the method indicated in sec.
118. This will no longer be the case when one applies the method of sec. 121.

II. In the former case, one may express:

= [ 1] + [2 3] + ... + [2n2 2n1]

138 Lessons on integral invariants

in terms of the 2n linearly independent forms , 1, ..., 2n1 . The equations:

= 0, 2 = 3 = ... = 2n1 = 0

that are obtained by writing the equations of the associated system of , in addition to the
equation = 0, in which one supposes that the differentials are coupled by the relation = 0,
have an intrinsic significance. This is the associated system of the two forms and [ ], and,
as a result (sec. 103), it is the characteristic system of the Pfaff equation = 0. We call that
system (), and denote the characteristic system of by (S), which contains the equation 1 = 0,
as well.
One may obtain a first integral y1 of the system () by an operation of order 2n 1. Upon
equating the system () of the new form i.e., the characteristic system of the new equation
= 0 to an arbitrary constant, the number of its equations will be reduced by two units. One may
thus find new integrals:
y2, ..., yn1
by operations of order:
2n 3, ..., 3,

such that upon equating them to new arbitrary constants the new system () that corresponds to
will contain only one equation, which will obviously be = 0. This says that this equation is
completely integrable, and a new operation of order 1 will give a new integral yn , which permits
us to write:
= z1 dy1 + z2 dy2 + ... + zn dyn .

One will thus arrive at the reduced form of , which effectively involves the minimum
number 2n of variables, since 2n is the number of equations in the characteristic system (S) of ;
i.e., the class of .
One finds, with no difficulty, the most general transformation that one can perform on the
characteristic variables yi and zi that preserves the form . The equality:

z1 dy1 + ... + zn dyn = z1 dy1 + ... + zn dyn

gives, upon keeping the most general case:

V( y1 , ..., yn , y1, ..., yn) = 0,

z1 z z z z
= 2 = = n = 1 = = p .
V V V V V
y1 y2 yn y1 yn
 Equations with an absolute linear integral invariant 139

z2 z
These formulas show that the variables y1 , ..., yn , , , n are transformed into each other.
z1 z1
They are the minimum number of variables by means of which the equation = 0 may be
written. They are the first integrals of the characteristic system () of that equation.
If there exist p independent relations:

V1 = 0, V2 = 0, ..., Vp = 0

between the yi and the yi then the formulas that define the transformation will be:

V1 V V
zi = 1 + 2 2 + + p p ,
yi yi yi
V1 V V p
zi = 1 + 2 2 + + p ,
yi yi yi

with p auxiliary unknowns 1, ..., p .

129. One must remark that, from the point of view of integration, the difference between the
two cases where the characteristic system of is odd (2n + 1) or even (2n) is that in the first case
the integration requires operations of order:

2n, 2n 2, ..., 2, 0,

whereas, in the second case it requires operations of order:

2n 1, 2n 3, ..., 1.

One must also remark that the two cases are practically distinguished from each other in the
following manner: Let 2n be the rank of , i.e., let n be the largest exponent such that the form
[ n ] is non-zero. In the first case, [ n ] is non-zero. In the second case, [ n ] is zero.

II. - Generalization of the Poisson-Jacobi formulas.

130. I. Suppose that the form is of the first type. Let f be an arbitrary first integral of its
characteristic system. The form [ n df] is invariant and of maximum order 2n + 1. One may
thus set:
[ n df] = {f}[ n ],
140 Lessons on integral invariants

in which {f} is a finite quantity that is linear with respect to the first order partial derivatives of
the function f. The quantity {f} is either a constant or a first integral of the characteristic
equations of .
Now let f and g be two first integrals of the characteristic equations of . One may define a
quantity (f g) by the relation:
n [ n df dg] = (f g)[ n ].

The quantity (f g) is again a first integral.

If the form has been reduced:

= du + z1 dy1 + ... + zn dyn
then one has:
f
{f}= ,
u
i=n
f g g g f f
( f g) = zi zi .
i =1 zi yi u yi yi u

From this, one may deduce the following important identities without difficulty:

{(f g)} = ({f} g) + (f {g}),

((f g) h) + ((g h) f) + ((h f) g) = (f g){h} + (g h){f} + (h f){g}.

131. In order to prove these identities directly, we remark that the form df {f} is a linear
combination of the 2n linearly independent forms by means of which may be expressed, since
one has:
[ n (df {f})] = 0.

From this, one immediately derives an identity of the form:

n [ n 1 (df {f})(dg {g} )] = [ n ],

and exterior multiplication by gives = (f g). One then has:

(1) n [ n 1 df dg] n{f}[ n 1 dg] + n{g}[ n 1 df} = (f g)[ n ].

When the identity (8) of sec. 68 is applied to the three linear forms df {g}, dg {g}, dh
{h} , that gives:

(f g)[ n 1 (dh {h})] + (g h)[ n 1 (df {f})] + (hf)[ n 1 (dg {g})]

= (n 1)[ n 2 (df {f})(dg {g})(dh {h})],

from which one deduces, upon multiplying by , that:

 Equations with an absolute linear integral invariant 141

(2) [ n 1 ((f g) dh + (g h) df + (h f) dg)] = (n 1)[ n 2 df dg dh].

Having said that, taking the exterior derivative of identity (1) shows that:

n[ n 1 d{f}dg] + n[ n 1 df d{g}] = [ n d(f g)],

i.e., the first identity shows that:

({f} g) + (f {g}) = {(f g)}.

Exterior derivation of identity (2) then gives:

[ n (f g) dh + (g h) df + (h f) dg)] [ n 1 d(f g) dh] [ n 1 d(g h) df] [ n 1 d(h f) dg]

= (n 1)[ n 1 df dg dh].

On the other hand, exterior multiplication of (1) by dh gives:

n [ n 1 df dg dh] n{f}[ n 1 dg dh] + n{g}[ n 1 df dh] = (f g)[ n dh].

From this one deduces the final formula:

n [ n 1 df dg dh] = [{f}(g h) + {g}(h f) + {h}(f g)][ n ],

and, from the preceding, the identity has shown that:

(f g){h} + (g h){f} + (h f){g} = ((f g) h) + ((g h) f + ((h f) g).

132. Now suppose that the form is of the second type. Similarly, if one is given two first
integrals f and g of the characteristic equations of then one will define the quantities {f} and (f
g) by the formulas:
n [ n 1 df] = {f}[ n ]
n [ n 1 df dg] = (f g)[ n ].

If is of the reduced form:

= z1 dy1 + z2 dy2 + ... + zn dyn
then one will have:
f f f
{ f } = z1 + z2 + + zn ,
z1 z 2 z n

f g f g
( f g) = .
z
i iy yi z i

One then verifies the following formulas without difficulty:

142 Lessons on integral invariants

(3) {(f g)} = (f g) + ({f} g) + (f {g})

(4) ((f g) h) + ((g h) f) + ((h f) g) = 0,

in which the second one is nothing but the Jacobi identity, since f, g, h are first integrals of the
characteristic equations of .
In order to prove the first identity directly we apply the identity (8) of sec. 68 to the three
linear forms, , df, dg. The relations:

n [ n 1 df] = {f}[ n ],
n [ n 1 df dg] = (f g)[ n ],
n [ n 1 dg ] = {g}[ n ],
lead to the identity:
[ n 1 ({f} dg + (f g) {g} df)] = (n 1)[ n 2 df dg],

which, when exterior differentiated, gives:

[ n 1 d{f} dg + [ n 1 df d{g}] + (f g)[ n ] [ n 1 d(f g)]

= (n 1)[ n 1 df dg].

Upon replacing each term by its value and simplifying, one obtains the identity that was to be
proved.

III. - Use of known first integrals.

133. Suppose that the form is of the first type and that we know p independent first
integrals y1, ..., yp of these characteristic equations. We form the quantities {yi}, (yi yj). If they
introduce new integrals then we adjoin them to the given ones and repeat the operation until it
gives no new integrals. We may thus suppose that the former result is obtained, i.e., that the
quantities {yi} = ai , (yi yj) = aij are functions of y1, ..., yp .
If we now introduce auxiliary variables x1, ..., xp then we will obtain two forms, one of which
is linear:
= a1 x1 + a2 x2 + ... + ap xp ,

and the other of which is quadratic:

= aij [i j] .

The first form indicates the value of the quantity {f} when f is any one of the variables y1, ...,
yp that admits 1, ..., p for its partial derivatives. The second form, or rather the alternating
bilinear form:
aij i j
to which it corresponds, indicates the value of the parenthesis (f g).
 Equations with an absolute linear integral invariant 143

Having said this, we shall reduce the preceding two forms by a convenient linear substitution
in the variables i .
Three cases are possible: If the form has been reduced to:

= [ 1 2 ] + ... + [ 2q 1 2q ]
then one might have:
= 0,
= 1 ,
= 2q +1 .

When a linear substitution with coefficients that are functions of the yi is performed on the
yi, one will obtain p differential forms 1, ..., p that satisfy the identity:

1 1 + 2 2 + ... + p = 1 y1 + 2 y2 + ... + p yp .
p

Having said this, all of the forms in case a):

[ n i], [ n 1 i j]
are zero, except for:
n[ n 1 12] = ... = n n1 [2q12q] = [ n ].

One easily deduces that:

= [1 2] + ... + [2q1 2q] + [2q+1 1] + ... + [p p2q] + [p2q+1 p2q+2] + ...

If one equates the yi to arbitrary constants then the form will remain a form of the first
type, and the rank of will be reduced by 2p 2q units. This case is identical to the one that
was studied in the preceding chapter, in which the given first integrals are the integrals of the
characteristic system of .

In case b), one has:

[ n (1 )] = [ n 2] = ... = [ n p] = 0,

and n is reducible to the form:

n = [(1)2] + [3 4] + ... + [2q12q] + [2q+11] + ... + [pp2q] + [p2q+1p2q+2] + ...

If one equates the yi to arbitrary constants then the form will again remain of the first type,
and the rank of will be reduced by 2p 2q units.

In case c), one has:

[ n 1] = ... = [ n 2q] = [ n (2q+1 )] = ... = [ n p] = 0.

144 Lessons on integral invariants

The form is reducible to:

= [1 2] + ... + [2q12q] + [(2q+1 ) 1] + ... + [pp2q] + [p2q+1p2q+2] + ...

If one equates the yi to arbitrary constants then the form will become a form of the second
type, and the rank of will be reduced by 2p 2q 2 units. The characteristic system of the
new equation = 0 will be composed of 2n 2p + 2q + 1 equations. In this case, the
integration will require operations of order:

2n 2p + 2q + 1, ..., 3, 1,

whereas, in cases a) and b) it will require operations of order:

2n 2p + 2q, ..., 2, 0,
resp.
In summation, if the exterior product [] is zero then the form will remains a form of the
first type, and it will become a form of the second type in the contrary case.

134. Now suppose that the form is of the second type. We again have two forms:

= a11 + ... + ap p ,
= aij [i j] .

Since the coefficients aij are given by the equations:

n [ n 1 dyi dyj] = aij [ n ],

the rank of will be reduced by 2p 2q units when one equates the integrals yi to arbitrary
constants, if 2q is the rank of the form .
Since the form is reduced to its normal form:

= [ 1 2 ] + ... +[ 2q 1 2q ],

one can suppose, at the same time, that one has one of the following forms for :

a) = 0,
b) = 1 ,
c) = 2q +1 .

From identity (3), case a) requires that all of the brackets (yi yj) must be zero, i.e., that the
form must be identically zero. One thus has q = 0. In this case, one obviously has:

= [ 1] + [1 2] + ... + [p p+1] + [p+2 p+3] + ...

 Equations with an absolute linear integral invariant 145

The form remains a form of the second type, since the number n is diminished by p units.

In case b), one has:

n [ n 1 1] = n [ n 1 1 2] = ... = n [ n 1 2q12q] = [ n ],

and is reducible to the form:

= [1 2] + ... + [2q12q] + [( + 2) 1]
+ [2q+1 2] + ... + [p p2q+1] + [p2q+2 p2q+3] + ...

If one equates the yi to arbitrary constants then the form will remain a form of the second
type since the rank of will be diminished by 2p 2 q units.

In case c), is reducible to the form:

= [1 2] + ... + [2q1 2q] + [ 2q+1] + [2q+2 1] + ...+ [pp2q1] + [p2q p2q+1] + ...

If one equates the yi to arbitrary constants then the form will become a form of the first
type since the rank of will be diminished by 2p 2q units.
In summation, if the product [] is zero then the form will remain a form of the second
type, and in the contrary case it will become a form of the first type.

135. In summation, we have obtained four essentially distinct reduced problems by

abstracting from two cases of a), one of which was treated in the preceding chapter, and the other
of which corresponds to knowing p first integrals in involution for the characteristic system of
the equation = 0.
One may examine the four reduced problems a little more carefully and ask whether all of the
possible results of the known first integrals have been obtained. The method of answering this
question is the same as the one that was used in the preceding chapter. It is based upon the
reduction of to a canonical form that involves p conveniently chosen functions of y1, ..., yp ,
and other independent first integrals. Once this canonical form is obtained, one deduces from it
the equations of the largest group of transformations that preserve the given data when they act
on the integral curves
We shall rapidly indicate the canonical forms of and in each of the four cases, with the
calculations that bring us to that form being made in the same manner as in the preceding chapter
(sec. 125).

1. The form is of the first type, and the forms and are reducible to:

= 1 , = [ 1 2 ] + ... + [ 2q 1 2q ].
In this case, one has:

= [(1) 2] + [34] + ... + [2q12q] + [2q+1 1] + ... + [p p2q] + [p2q+1 p2q+2] + ...
146 Lessons on integral invariants

If one sets:
= [1 2] + ... + [2q1 2q]

then the exterior derivative of will give, if one neglects terms in:

2q+1 , ..., p , p2q+1 , ..., 2np ,

the identity:
[ 2] + [ 2 ] + [ 2q +1 1] + ... + [ p p2q] = 0.

One may then set:

dy2
2 =, 2 q +1 = dy2q +1 , , p = dy p .
y2
1
Since the exterior derivative of the form is zero, one may then set:
y2

= y2 ([dy1 dy2 ] + + [dy2 q 1 dy2 q ]) .

Finally, one has:

dy2
= + y2 [dy1 dy2 ] + + y2 [dy2 q 1 dy2 q ] +[dy2q +1 1 ] + + [dy p p 2 q ] +
y2

This result may be put into a more intuitive form by setting:

1
= y1 dy2 y2 q1 dy2q .
y2
Indeed, one then gets:

= [dy2q +11 ] + + [dy p p 2 q ] + [ p 2 q +1 p 2 q + 2 ] + + [2 n p 12 n p ] .

In this form, one sees quite clearly that all of the possible results have been deduced from the
known integrals.
In addition, one has the canonical relations:

{ y1} = 1, { yi } = 0, (i = 2, , p );
( y1 y2 ) = ( y3 y4 ) = = ( y2q 1 y2 q ) = y2 ,

and all of the other brackets are zero.

2. The form is of the first type, and the forms and are reducible to:

= 2q +1 , = [ 1 2 ] + ... + [ 2q 1 2q ].
In this case, one has:
 Equations with an absolute linear integral invariant 147

= [1 2] + ... + [2q1 2q] + [(2q+1 ) 1] + ...+ [p p2q] + [p2q+1 p2q+2] + ...

Exterior derivation of the right-hand side easily shows that one may set:

2q + 2 = dy2q + 2 , , p = dy p ,
2 q +1 = dy2 q +1 + y1 dy2 + + y2 q 1 dy2 q ,
= [dy1 dy3 ] + [dy3 dy4 ] + + [dy2 q 1 dy2 q ] .
Upon setting:
= + dy2q +1 + y1 dy2 + + y2 q 1 dy2q ,
one obtains:
= [ 1 ] + [dy2 q + 2 2 ] + + [dy p p 2q ] + [ p 2 q +1 p 2q + 2 ] + ,

a formula that makes it obvious that all of the possible results have been derived from the given
integrals.
In addition, one has obtained the canonical relations:

{ y2q +1 } = 1,
( y1 y2 ) = = ( y2 q 1 y2 q ) = 1,

( y1 y2 q +1 ) = y1 , ( y3 y2q +1 ) = y3 , ...,

in which all of the other quantities { yi }, ( yi y j ) are zero.

3. The form is of the second type, and the forms and are reducible to:

= 1 , = [ 1 2 ] + ... + [ 2q 1 2q ].
In this case, one has:

= [1 2] + ...+ [2q1 2q] + [( + 2) 1] + [2q+1 2] + ... + [pp2q+1] + ...

Again, set:
= [12] + ... + [2q12q],

and exterior differentiate , while neglecting the terms in:

+ 2 , 2q+1 , ..., p, p2q+2 , ...

We get:
= [ 1] + [ 2 1] + [ 2q +1 2 ] + ... + [ p p2q+1] = 0.

This identity permits us to set:

2 q +1 = dy2 q+1 , , p = dy p .
148 Lessons on integral invariants

One then sees that if we regard y2q +1 , , y p as constants then 2 will be equal to , which
is of rank 2q, since the equation 2 = 0 is a part of the associated system to 2 . One may then
suppose that:
2 = ( y1 dy2 + + y2q 1 dy2 q ) .
Finally, if one sets:
= + 2 = y1 dy2 y2q 1 dy2q
then one will get:
= [ 1 ] + [dy2 q +1 2 ] + + [dy p p 2 q+1 ] +

One sees that all of the possible results of the known integrals have been deduced, and one
arrives, in addition, at the canonical relations:

{ y1 } = y1 , { y3 } = y3 , ..., { y2 q1} = y2q 1 ,

( y1 y2 ) = ( y3 y4 ) = = ( y2q 1 y2 q ) = 1 ,

with all of the other quantities { yi }, ( yi y j ) being zero.

4. The form is of the second type, and the forms and are reducible to:

= 2q +1 , = [ 1 2 ] + ... + [ 2q 1 2q ].
In this case, one has:

= [1 2] + ... + [2q1 2q] + [ 2q+1] + [2q+2 1] + ... + [pp2q1] + ...

If the notation retains its same significance that it had in part II then one will have, by
neglecting terms in:
2q+3 , ..., p , p2q , ..., 2n1p ,
the identity:
+ [ 2q+1] [ 2q +1 ] + [ 2q + 2 1] + ... + [ p p2q1] = 0.

The exterior derivatives 2q +1 , 2q + 2 , ..., p are zero, along with 2q+1 , ..., p . One may
therefore set:
dy
2 q +1 = 2 q +1 , 2 q + 2 = dy2 q + 2 , , p = dy p .
y2 q +1

The exterior derivative of the form y2 p +1 will then be zero when one regards y2q + 2 , , y p
collectively as constants. One may therefore set:

y2q +1 = [dy1 dy2 ] + + [dy2q 1 dy2 q ] .

Finally, upon setting:
 Equations with an absolute linear integral invariant 149

= y2 q+1 y1 dy2 y2 q 1 dy2 q ,

one gets:
= [dy2 q + 2 1 ] + + [dy p p 2 q 1 ] +

One sees clearly that all of the possible results have been deduced from the known integrals.
In addition, one obtains the canonical relations:

{ y2q +1} = y2q +1 ,

( y1 y2 ) = ( y3 y4 ) = = ( y2q 1 y2 q ) = y2 q +1 ,

with all of the other quantities {yi}, (yi yj) being zero.
 CHAPTER XIV

DIFFERENTIAL EQUATIONS THAT ADMIT

AN INVARIANT PFAFF EQUATION

I. - General method of integration.

136. We have already (sec. 104) encountered the characteristic system of a Pfaff equation:

(7) a1dx1 + a2dx2 + ... + ar dxr = 0.

It is composed of the equations:

(dx1 ) (dx2 ) (dxr )
(8) = 0, = = = ,
a1 a2 ar

of which the last r 1 provide the associated system of the quadratic exterior form when one
supposes that the variables are coupled by the relation = 0.
This characteristic system has likewise been encountered in the preceding chapter (sec. 128)
in the context of a Pfaff expression of the second type.
The number of independent equations of the characteristic system (2) is always odd. Indeed,
on account of the relation = 0, one may put into the form:

= [1 2] + ... + [2n1 2n] (mod ),

if we denote the linear differential forms, which are mutually independent and independent of ,
by 1, 2, ..., 2n . The characteristic system of equation (1) is then defined by the equations:

= 1 = 2 = ... = 2n = 0.

The integer n is, as one knows, the greatest integer such that the form [ n ] is non-zero. The
class of the equation = 0 is equal to the degree of that form.

137. It is easy to recover a canonical form for equation (1). Indeed, let y1 be an arbitrary
first integral of the characteristic system (2). If one equates y1 to an arbitrary constant C1 and dy1
to zero then the rank of the characteristic system of the new equation (1) will be reduced by at
least one unit, and, since its rank is odd, it will be reduced by at least two units. Let y2 be a first
integral of the new characteristic system. If one sets:
 Differential equations and invariant Pfaff equations 151

y1 = C1 , y2 = C2 , dy1 = 0, dy2 = 0

then the rank of the characteristic system of the given equation will be reduced by at least four
units, and so on. Finally, after at least n + 1 equations the equation = 0 will be verified
identically. In other words, that equation will be of the form:

z1 dy1 + z2 dy2 + ... + zq dyq + dyq+1 = 0 (q n).

The integer q is, moreover, equal to n; otherwise equation (1) could be written by means of less
than 2n + 1 variables.

Hence, if the characteristic system of equation (1) is of rank 2n + 1 then this equation will be
reducible to the form:
dyn+1 + z1 dy1 + z2 dy2 + ... + zn dyn = 0,
and the quantities:
y1, ..., yn+1 ; z1, ..., zn

will constitute a system of independent first integrals of the characteristic equations.

One sees that by this method the reduction of system (1) to its canonical form and, as a
result, the integration of its characteristic system requires n+1 successive operations of order:

2n + 1, 2n 1, ..., 3, 1
and differentiations.

138. One may remark, as in chapter XII (sec. 120), that knowing N n+1 first integrals:

y1, y2, ..., yN

such that equation (1) is verified identically by equating these integrals to arbitrary constants will
permit us to accomplish the integration of the characteristic equation by differentiation. Indeed,
equation (1) may in one and only one manner be put into the form:

dyN + z1 dy1 z2 dy2 + ... + zN-1 dyN1 = 0,

and one proves that the coefficients z1, ..., zN1 are again first integrals of the characteristic
equations.
More generally, one may propose to see to what the integration of the characteristic system
reduces when one knows a certain number r of independent first integrals of this system.

139. First integrals in involution. We say that two first integrals f and g of the characteristic
system of equation (1) are in involution if one has:

(9) [ n 1 df dg] = 0.
152 Lessons on integral invariants

This definition is obviously independent of the choice of variables, and is also independent of
the arbitrary factor by which one may multiply the left-hand side of equation (1).
The property of two first integrals being in involution implies the important consequence that
the rank of the characteristic system will be reduced by four units when one supposes that the
variables are coupled by two relations:
f = C, g = C ,

in which C and C are two arbitrary constants. Indeed, if one supposes that df = dg = 0 as well as
= 0, then condition (3) will express the idea that the rank of is less than 2n 2, and as a
result, is equal to 2n 4.

II. - Use of known integrals.

140. Case where one knows p first integrals y1, ..., yp that are pair-wise independent and in
involution. - In this case, from the developments of chapter VI (sec. 67), it results that when one
supposes the differentials are coupled by the relations:

= 0, dy1 = 0, ..., dyp = 0,

the rank of will be reduced by 2n 2p. If one supposes that the variables are coupled by the
relations:
y1 = C1, y2 = C2 = , ..., yp = Cp ,

then the characteristic system of equation (1) will be of rank 2n 2p + 1, and its integration will
require operations of order:
2n 2p + 1, 2n 2p 1, ..., 3, 1,
along with its differentiations.
The case that we shall now examine is the one in which the rank of the characteristic system
is instantly reduced by the maximum number of units 2p.

141. Case where the given first integrals are not all in pair-wise involution. In this case,
when one equates the given first integrals to arbitrary constants, the reduction of rank of the
characteristic system will not attain its upper limit 2p. On the other hand, one may determine an
absolute linear integral invariant for the characteristic equations, which, in certain cases, may
produce a reduction of the integration problem that is at least as large as in the first case, which is
apparently the most favorable case.
Indeed, suppose that y1 and y2 are two first integrals of the characteristic equations that are
not in involution. One will have:
n [ n 1 dy1 dy2] = A[ n ],

in which the coefficient A is non-zero. There exist an infinitude of (unknown) functions m such
that:
 Differential equations and invariant Pfaff equations 153

= m

is an invariant form; i.e., it may be expressed by means of first integrals of the characteristic
equations and their differentials. For such a form, one has:

= m + [dm],
and, as a result:
[ n1 dy1 dy2] = mn [ n 1 dy1 dy2],
[ n ] = mn+1[ n ].
By comparison, one thus has:
A
n [ n1dy1 dy2 ] = [ n ] .
m

The two forms between brackets are obviously invariant. As a result, A / m is a first integral.
Hence
A
= A
m

is an invariant form. This is the conclusion that we hoped to reach:

If one knows two first integrals y1, y2 such that the function A, which is defined by the
equality:

n [ n 1 dy1 dy2] = A [ n ],

is not zero then the linear form A will be an absolute invariant form.

In addition, we remark that the minimum number of variables by means of which the form
A can be expressed are obviously the 2n + 1 first integrals of the given characteristic equations.
The characteristic system of the form A thus agrees with that of equation (1) and, as a result, it
is of odd rank. The form A is then of the first type.

142. One may attach the preceding theorem to a method of integration that is susceptible to a
vast generalization and consists of integrating the characteristics of the form u , where u is an
auxiliary variable. Indeed, it is obvious that to any solution of these equations there corresponds
a solution of the characteristic equations of the equation = 0, namely, the one that is obtained
by eliminating the auxiliary variable u from the relations that define the solution.
The form u is obviously of the second type, and the general method of integration of its
characteristic equations that was discussed in sec. 128 is identical to the one that was recalled in
sec. 137 for the characteristic equations of the equation = 0. However, the advantage
becomes obvious if one knows, a priori, first integrals of the characteristic equations, because
one may apply the method that was discussed in sec. 134 to the integration of the characteristic
154 Lessons in integral invariants

equations of the form u . In particular, if one knows two first integrals y1 and y2 of the
characteristic equations of the equation = 0 then one will have:

{y1} = 0, {y2} = 0,

and if one then calculates the bracket (y1 y2) that was defined by (sec. 132):

(n + 1)[ (u )n dy1 dy2] = (y1 y2) (u )n +1

then this will give, upon developing and equating the terms which contain du:

A
( y1 y2 ) = ,
u

in which A is the quantity that was defined in the preceding section. One may continue to apply
A
the general method while preserving the variable u by forming the quantity , which is the
u
bracket of that quantity with y1 and y2, and so on. One may thus remark that the form A is itself
invariant since A / u is a first integral of the characteristic equations of the form u.

III. - Application to first-order partial differential equations.

143. The problem of integrating the characteristic equations of a Pfaff equation finds an
immediate application in the theory of first-order partial differential equations. Indeed, to
integrate an equation:
z z z
F z , x1 , x2 , , xn ; , , , , = 0,
x1 x2 xn

or, if we employ the classical notation:

(10) F(z, x1, x2, ..., xn ; p1, p2, ..., pn) = 0,

is to determine (n + 1) functions z, p1, p2, ..., pn of x1, x2, ..., xn that satisfy equation (4) and the
Pfaff equation:
(11) dz p1 dx1 p2 dx2 ... pn dxn = 0.

Now, if one imagines that one of the 2n + 1 arguments z, xi, pi has been expressed in
equation (4) as a function of the 2n others then the Pfaff equation (5) will contain only 2n
variables, and its characteristic system will necessarily be of odd rank 2n 1. As a result, on
account of equation (4) the Pfaff equation (5) will be reducible to the canonical form:
 Differential equations and invariant Pfaff equations 155

(12) dZ P1 dX1 ... Pn1 dXn1 = 0,

in which Z, X1, ..., Xn1, P1, ..., Pn1, are 2n 1 independent functions. They are first integrals of
the characteristic equations of equation (5).
Having said this, suppose that one wants to bring equation (5) into the canonical form (6).
As the integration of equation (4) amounts, in essence, to the determination of a number, which
is equal to n + 1 (with the given relation (4)), of independent relations between z, x1, ..., xn , p1, ...,
pn such that equation (5) results, it will suffice to establish a number, which is equal to n and
implies equation (6) as well, of independent relations between Z, X1, ..., Pn1, in order to arrive at
this result. But this is possible in a general manner by taking:

Z = f(X1, ..., Xn1),

f f
P1 = , , Pn 1 = ,
X 1 X n 1

in which f denotes an arbitrary function of its arguments. More generally, one will establish an
arbitrary number h n of independent relations:

1(Z, X1, ..., Xn1) = 0, 2 = 0, ..., h = 0,

between Z, X1, ..., Xn1, and combine them with the relations that are obtained by eliminating the
homogenous parameters 1, ..., h from the equations:

1 1 2 2 h h
1 + P1 + 2 + P1 + + h + P1 = 0,
X 1 Z X 1 Z X 1 Z
...
1 1 2 2 h h
1 + Pn 1 + 2 + Pn1 + + h + Pn1 = 0.
X n1 Z X n 2 Z X n1 Z

144. The equations:

X1 = a1, ..., Xn1 = an1 ,
P1 = b1, ..., Pn1 = bn1 ,
Z = c,

define one-dimensional multiplicities that are characteristic multiplicities of the Pfaff equation
(5) (when one supposes that the variables are coupled by relation (4)). These are what one calls
the characteristics of the partial differential equation (4). One sees immediately that any surface
integral is generated by characteristics.
It is easy to form the differential equations of the characteristics. Indeed, they are the
equations of the associated system of if one supposes that the differentials of the variables are
156 Lessons in integral invariants

coupled by the relation = 0 and also by the relation dF = 0. One thus obtains the equations
(sec. 104) by adding the equations:

(dz ) (dx1 ) (dx2 ) (dp1 ) (dpn )
1 p1 p2 0 0 = 0,
F F F F F

z x1 x2 p1 pn

which may be written:

dx1 dx dp1 dpn
(13) = = n = = = ,
F F F F F F
+ p1 + pn
p1 pn x1 z xn z

to equation (5). One then recovers the classical equations.

IV. - Cauchys method.

145. The method that we just presented amounts, in essence, to the integration of the
characteristic equations and the reduction of equation (5) to its canonical form (6). Moreover,
this reduction will result in the integration if it is itself directed in a convenient manner (sec.
137). Whatever procedure is used to integrate the characteristic equations, it is easy to see that
the reduction of equation (5) to its normal form is always possible once the integration of the
characteristic equations has been performed. It suffices to determine the first integrals that, for a
given numerical value xn0 of xn reduce them to:

z, x1, ..., xn1, p1, ..., pn ,

respectively.
If one denotes these first integrals by:

Z, X1, ..., Xn1, P1, ..., Pn ,

which are necessarily coupled by the relation:

F(Z, X1, ..., Xn1, xn0 ; P1, ..., Pn) = 0,

then relation (5), which may, as one knows, be expressed by means of first integrals, will
obviously reduce to:
dZ P1 dX1 ... Pn1 dXn1 = 0.
 Differential equations and invariant Pfaff equations 157

This is the principle of Cauchys method.

V. - Lagranges method.

146. Lagranges method of the complete integral also adheres easily to the preceding
viewpoint. The equation:
(14) V(z, x1, ..., xn ; a1, ..., an) = 0

will define a complete integral if it defines a function of z that satisfies equation (4) for any n
arbitrary constants a1, ..., an . Equation (4) is, moreover, the only one that satisfies all of the
functions of z that are defined by (8), because the elimination of the a1, ..., an , between equation
(8) and the equations:
V V
x + p1 z = 0,
1
(15)
V V
+ pn = 0,
xn z

which are derived from them, leads in general (and this is what we have supposed) to only one
relation, which is naturally equation (4).
Since equation (4) is the result of the elimination of the a1, ..., an , from the (n + 1) equations
(8) and (9) the integration of equation (4) will amount to satisfying the Pfaff equation (5) while
supposing that the 3n + 1 variables z, xi , pi , ai are coupled by the (n + 1) relations (8) and (9).
Now, on account of these relations, one has:

V V V
0= dz + dx1 + + da1 +
z x1 a1
V V V
= (dz p1dx1 pn dxn ) + da1 + + dan .
z a1 an

The Pfaff equation (5) is itself reduced to its normal form by setting:

X1 = a1, ..., Xn1 = an1, Z = an ,

V V
a a
P1 = 1 , , Pn1 = n1 .
V V
an an

One sees that the characteristics are defined by the equations:

158 Lessons in integral invariants

V = 0,

V V V V
+ b1 = 0, , + bn 1 =0.
a1 an an1 an
This is a classical result.

147. We now apply the theorem of sec. 141 to the particular case of an equation in two
independent variables:
(16) F(x, y, z, p, q) = 0.

The knowledge of two independent first integrals u and v of the characteristic equations
leads, when they are not in involution, to the determination of a linear integral invariant for the
characteristic equations. This invariant is A, where A is defined the equality:

[ du dv] = A[ ],

or rather, since we have supposed that the variables are coupled by relation (10) here:

[dF du dv] = A[dF ].

If we take the terms in [dx dz dp dq] in both sides of this equation in particular then we will
find that:
F F F F
+p
x z p q
1 u u u u
A= +p .
F x z p q
q v v v v
+p
x z p q

Hence, if the determinant of the right-hand side is non-zero then the expression A(dz p dx
q dy) will be an invariant form for the characteristic equations.

VI. First-order partial differential equations that admit an infinitesimal transformation.

148. If the first-order partial differential equation:

F(z, x1, ..., xn , p1, ..., pn) = 0

admits an infinitesimal transformation Af that acts on the variables z, x1, ..., pn then this would
signify that any system of n + 1 relations between these 2n + 1 variables that defines an integral
 Differential equations and invariant Pfaff equations 159

multiplicity is changed by the transformation into another system of n + 1 relations that also
defines an integral multiplicity. As a result, on account of equation (4), the Pfaff equation:

dz p1 dx1 ... pn dxn = 0

will admits the infinitesimal transformation Af. It then immediately results from this (sec. 97)
that the linear form:
( ) z p1 x1 p2 x2 pn xn
= ,
( A) A( z ) p1 A( x1 ) p2 A( x2 ) pn A( xn )

will be invariant for the system of differential equations of the characteristics.

The knowledge of one infinitesimal transformation thus implies the knowledge of a linear
integral invariant for the characteristic equations and, as a result, the integration of the given
equation, which, since it is a problem of the second type that requires operations of order:

2n + 1, 2n 1, ..., 3, 1,

is converted into a problem of the first type that requires operations of order:

2n, 2n 2, ..., 2, 0.

149. A classic example of this is the case in which the given equation (1) does not depend
explicitly on z. It is obvious then that from any solution of the equation one can deduce another
solution by adding an arbitrary constant to z. In other words, the given equation admits the
infinitesimal transformation:
f
Af = .
z

The absolute integral invariant that admits the characteristic equations is then:

= z p1 x1 ... pn xn.

The method of integration of these equations is, of its nature, a result of the theory of chapter
XII. Here, the characteristic equations of are:
 dx1 dx dp1 dpn
= = n = = = .
F F F F
p1 pn x1 xn

Once we have determined n 1 first integrals that are pair-wise in involution the integration
of the characteristic equations of will be converted into a quadrature, since the expression
will become an exact differential when on equates the n 1 first integrals to arbitrary constants.

V11. - Jacobis first method.

150. Jacobis first method for integrating first-order partial differential equations is related to
the preceding considerations. Jacobi converted equation (4), which was supposed arbitrary, into
an equation in which the unknown function no longer figured, namely:

V V
x x
F z , x1 , , xn , 1 , , n = 0.
V V
z z

To abbreviate, we set:
V
=u,
z

and the characteristic equations to be integrated are those of the absolute invariant form:

V u (z p1 x1 ... pn xn),

in which the 2n + 3 variables are related by the relation (4). They admit the relative integral
invariant:
(17) u(z p1 x1 ... pn xn),
and they are the characteristic equations of the integral invariant that one integrates with the
methods of chapter XII.
Jacobis method approaches the one that we indicated in sec. 142, with the difference that the
latter method used the integral (11) as absolute integral invariant, but Jacobis method uses it as
a relative integral invariant. Moreover, Jacobis method leads to operations of order:

2n + 2, 2n, ..., 2, 0,
instead of:
2n + 1, 2n 1, ..., 1.
 Differential equations and invariant Pfaff equations 161

Its advantage is that it permits us to use the knowledge of the given first integrals by
applying the Poisson-Jacobi theorem. However, this advantage is preserved by the method of
sec. 142, which deduced all of the possible consequences of given first integrals.

VIII. - Reducing certain differential equations to a first-order partial differential equation.

151. One may now place oneself at a viewpoint that is the opposite of the one that was taken
in the preceding sections.
First, consider a Pfaff equation with an even number 2s of variables, but suppose that only s
+ 1 of these coefficients are non-zero:

a1 dx1 + a2 dx2 + ... + as+1 dxs+1 = 0.

The characteristic equations of this Pfaff equation are obviously the same as those of the
first-order partial differential equation in s independent variables x1, x2, ..., xs that is obtained by
setting:
xs+1 = a, a1 + p1 as+1 = 0, ..., as+1 ps as+1 = 0,

and eliminating xs+1, xs+2, ..., x2s, between these s + 1 equations. Of course, it is necessary to
suppose that the elimination is possible and that it gives only one relation.

152. Now consider a system of differential equations that admit a relative linear integral
invariant , such that the form is in 2s + 1 variables and [ s ] is non-zero. The differential
equations considered are the characteristic equations of . Their integration can be converted
into the integration of a first-order partial differential equation that does not contain the unknown
function explicitly if the s coefficients of the differentials are zero in :

a1 dx1 + a2 dx2 + ... + as+1 dxs+1 .

Indeed, consider the Pfaff equation:

dV dV a1 dx1 a2 dx2 ... as+1 dxs+1 = 0,

and set:
p1 = a1, p2 = a2, ..., ps+1 = as+1 .

The elimination of the xs+2 , ..., x2s+1 between these s + 1 equations leads to one relation:

(18) F(x1, ..., xs+1 ; p1 , ..., ps+1) = 0,

which is nothing but the partial differential equation that was alluded to above. The differential
equations of the characteristics of that equation are formed from the characteristic equations
of , to which one adds the equation:
dV = 0.
162 Lessons in integral invariants

One easily accounts for the fact that the method of integration that was indicated in chapter
XII of the characteristic equations of leads to the same operations as the search for the
characteristics of the partial differential equation (12).
If the invariant is that of the equations of dynamics:

= p1 q1 + ... + pn qn H t

then equation (12) will be nothing but the Jacobi equation:

V V V
+ H t , q1 , qn ; , , =0.
t q1 qn

153. Jacobis method for integrating the equations of dynamics is therefore based
fundamentally upon the equivalence of two integration problems, the problem of integrating the
characteristic system of a relative linear integral invariant , and the problem of integrating the
characteristic equations of a first-order partial differential equation that admits an infinitesimal
transformation (for example, one that does not explicitly contain the unknown function). The
nature of the problem is determined in the two cases by the existence of an integral invariant .
This method of reducing to one partial differential equation is useful only if the form has
2s + 1 variables with s zero coefficients, but it is hard to believe that in the case where that
peculiarity is not present the integration of the characteristic equations of would be a problem
that is more complicated than the search for the characteristics of a first-order differential
equation that does not explicitly contain the unknown functions, or, what amounts to the same
thing, the integration of a system of canonical differential equations. Basically, the importance
of the canonical equations stems uniquely from their property of admitting an integral invariant
, and not on the simplicity of their form. The existence of the integral invariant is the
fundamental property from which everything else is derived.

IX. - Remarks on the nature of the principal practical applications of Jacobis method.

154. In fact, most of the rich variety of applications of Jacobis method in dynamics have
their origin in the simplifications that the search for a complete integral of Jacobis partial
differential equation presents, an equation that is obtained as a sum of functions in each of which
only one of the variables q1, ..., qn other than t appears. However, these simplifications may be
exhibited independently of any recourse to the theory of first-order partial differential equations
and complete integrals.
Indeed, let be a linear differential form in 2s + 1 variables that we denote by:

x1 , ..., x2s , t.
 Differential equations and invariant Pfaff equations 163

Suppose that may be decomposed into a sum of p terms:

= 1 + 2 + ... + p ,

such that the form is constructed from a certain number 2h of variables x and the variable t in
such a manner that the variables x that enter into the formation of any two of the forms 1, ..., p
are different. As a result, one will have:

s = h1 + h2 + ... + hp .

If one supposes that quadratic exterior form is of rank 2s then it will be necessary that the
forms 1 , 2 , ..., p must be of rank 2h1, 2h2, ..., 2hp , resp. The reduction of each of these p
forms to its canonical form will thus imply the same reduction for . As a result, integrating
the characteristic equations of amounts to integrating the characteristic equations of 2 , ...,
p , and the p corresponding problems may be solved independently of each other.
An even greater simplification is produced if the numbers ki of the x variables (that are
different for different forms i) that enter into the composition of these forms at the same time as
t are not all even. In this case, the variable t will be a first integral of the characteristic equations
of . Indeed, if one gives an arbitrary constant value to t then the rank of the quadratic form i
will be reduced to at most:

ki for even ki ,
ki1 for odd ki .

Now, 2s is equal to the sum of all the ki . The rank of will therefore be less than 2s for
constant t, which was to be proved. Furthermore, one sees that there cannot be two numbers ki
that are both odd, and when one makes t constant, the reduction of to its normal form will be
furnished by the reductions of 1 , ..., p to their normal forms when one likewise makes t
constant.
 CHAPTER XV

DIFFERENTIAL EQUATIONS THAT ADMIT

SEVERAL LINEAR INTEGRAL INVARIANTS

I. - Case in which one knows as many integral invariants as there are unknowns.

155. We will not go into the general problem of integrating differential equations that admit
an arbitrary number of integral invariants in these lessons. We will restrict ourselves to the
particularly simple case in which a system of n first-order ordinary differential equations in n
unknown functions admits n (independent) linear invariant forms:

1, 2, ..., n ;
i.e., n absolute linear integral invariants:

,
1 2 , ..., n .

In this particular case, the given differential equations can be written:

(1) 1 = 2 = ... = n = 0.

Since the quadratic exterior forms 1 , 2 , ..., n are invariant, they may be expressed in
terms of 1, 2, ..., n by formulas such as:

1,...,n
(2) s = c
( ik )
iks [i k ] (s = 1, 2, , n).

The coefficients ciks are obviously first integrals of the given differential equations. We shall see
that one may always convert them into a form where they are constant.
Indeed, suppose that among the integrals ciks there are a certain number r of independent
ones, which we denote by:
y1 , y2 , ..., yr .

The ciks are thus well-defined functions of these r integrals. Each differential dyi is, in turn,
an invariant form that can be expressed linearly by means of 1, 2, ..., n :

dyi = bi1 1 + bi2 2 + ... + bin n (i = 1, 2, ..., r).

The coefficients bik are, in turn, first integrals. If there are r of them that are mutually
independent and independent of the yi then their differentials dyr+1, ..., dyr + r can also be
 Equations with several linear integral invariants 165

expressed linearly as functions of the i , and the coefficients might provide new first integrals,
and so on. We will arrive at a point where these operations cease, and one will arrive at a certain
number n of first integrals y1, ..., y such that the coefficients ciks in formula (2) and the
coefficients bik in the formula:

(3) dyi = bi1 1 + bi2 2 + ... + binn (i = 1, 2, ..., )

are well-defined functions of the y1, y2, ..., y .

Having said this, suppose, for the sake of specificity, that the determinant that is obtained by
taking the first columns of the matrix of bik is non-zero. One may then substitute the invariant
forms dy1, ..., dy for the invariant forms 1, ..., . If one attributes arbitrary constant values
to the y1, ..., y then the system of equations to be integrated will admit n invariant forms
+1 , ..., n , and one will have:
+1,..., n
s =
( ik )
ciks [i k ] ( s = + 1, , n),

in which the coefficients ciks are now constant.

156. Therefore, take the case in which the coefficients ciks in formula (2) are all constant.
Conversely, it is easy to see that the existence of relations such as (2) implies, as a consequence,
the property of the forms 1, ..., n that they must be invariant with respect to the differential
equations (1). The characteristic system of the set of forms 1, 2,..., n is obtained (sec. 78) by
adding the equations of the associated system to 1 , 2 , ..., n equations that are all a
consequence of equations (1) to equations (1).
If one substitutes linear combinations with constant coefficients:

1 = a111 + a12 2 + + a1n n ,

(4)
= a + a + + a ,
n n1 1 n2 2 n1n n

for the forms 1, ..., n then these n new forms will again be invariant, and one will again have
relations:
1,..., n
s = ciks [i k ],
( ik )

with new constants ciks . We say that the matrix of ciks has the same structure as the matrix of ciks
. It is possible that one can choose the constant coefficients aij in the substitution (4) in such a
manner that only the 1 , , appear in the expression of the first < n derivatives 1, , ;
i.e., in such a manner that one will have:
166 Lessons on integral invariants

c + i , k , s = ck , +i , s = c +i , + j , s = 0 (k, s = 1, 2, ..., ; i, j = 1, ..., n ).

In this case, the forms 1, ..., are invariant for the completely integrable system of Pfaff
equations:
1 = 2 = ... = = 0.

If one knows how to integrate this system, and if one equates these first integrals to arbitrary
constants then the given system will be converted into a system that is analogous to the first one,
except that n will be replaced by n .
We say that the matrix of ciks is simple if it is impossible to find a linear substitution with
constant coefficients (4) that brings about the preceding reduction. We then say that the given
system of differential equations can be converted into successive systems such that the matrix of
ciks is simple for each of them. A particular integration problem corresponds to each matrix.

157. Leaving aside this method of reduction for the moment, we imagine a second system of
differential equations:
( 1 ) 1 = 2 = ... = n = 0

that admit the n invariant forms i with the relations:

1,...,n
( 2 ) s = ciks [ i k ],
( ik )

where the coefficients ciks have the same numerical values as in formula (2). Let

y1, y2, ..., yn ,

z1, z2, ..., zn ,

respectively, be two systems of independent first integrals, the first one being associated with
equations (1), and the second one, with equations (1 ). 1, 2, ..., n can then be expressed by
means of the dyi and their differentials. It is possible to choose the first integrals zi in such a
fashion that the i can be expressed by means of the zi and the dzi in the same way that the i
are expressed by means of the yi and the dyi. This amounts to saying that if this condition is not
realized then one can, at least, find functions:

f1(y1, ..., yn), ..., fn(y1, ..., yn)

such that upon setting:
z1 = f1(y1, ..., yn), ..., zn = fn(y1, ..., yn),

the i will become equal to the i, respectively. In order to this, it suffices to integrate the total
differential equations:
 Equations with several linear integral invariants 167

1 1 = 0,
(5)
n n = 0,

in which the zi are unknown functions of the independent variables yi . This Pfaff system (5) is
completely integrable (sec. 101), because, if one takes equations (5) into account then the
exterior derivatives:
si s = ciks [ i k ] ciks [i k ],

of the left-hand sides will all be zero. It is therefore possible to satisfy the stated conditions, and
in an infinitude of ways (that depend on n arbitrary constants).
In particular, this proves that the integrations of the two systems (1) and ( 1 ) are two
problems of essentially the same nature, in the sense that any method that uses only the given
property of the 1, ..., n that they are invariant forms for the integration can be applied to
systems (1) and ( 1 ) in a parallel fashion, such that any progress in the integration of (1) implies
an equivalent progress in the integration of ( 1 ).

II. - The group that preserves the given invariants.

158. Return to system (1), and imagine that a choice of n independent first integrals:

y1, ..., yn
has been made.
It is possible to find an infinitude of other systems of n first integrals:

y1 , y2 , , yn

such that the forms i may be expressed by means of the yi and their differentials in the same
manner as they are by means of the yi and their differentials. In order to do this, it suffices to
integrate the Pfaff system:
1 1 = 0,
(6)
n n = 0,

in which s denotes the same function of the yi and dyi that s is of yi and dyi . In this Pfaff
system, we regard the arguments y1 , ..., yn as unknown functions of the independent variables y1,
..., yn . Such a system is completely integrable for the same reason that was indicated in relation
to system (5). Therefore, there exist functions:

(7) ys = f s ( y1 , , yn ; C1 , , Cn ) ( s = 1, , n),
168 Lessons on integral invariants

that depend on n arbitrary constants and satisfy the conditions stated above.
Equations (7) define an infinitude of transformations that act on the first integrals y1, ...,
yn and preserve the givens of the problem; i.e., they leave the forms 1, ..., n invariant. These
transformations form a group G because, since they are characterized by the property of
preserving 1, ..., n , it will be obvious that if one performs a transformation of the form (7),
followed by another one, then one will again obtain a transformation of the same form. The
group G is a finite group of n parameters; it is the largest group that preserves the given
invariant forms when it is applied to the first integrals of the given system. As one easily
concedes, just what the knowledge of these n forms gives us depends on the nature of that group.
Moreover, this is a general fact that applies to all the cases where one knows, a priori, integral
invariants, systems of invariant equations, infinitesimal transformations, etc. The nature of the
largest group of transformations that preserves the known information when applied to the first
integrals of given differential equations (or, what amounts to the same thing, to their integral
curves, which are regarded as being indivisible) has an overshadowing importance in the
integration of that system.
In the case that occupies us, one sees, in particular, that it is impossible to obtain any first
integral without integration (7) by starting only with the fact that 1, ..., n are invariant forms.
Nevertheless, the fact that the forms 1, ..., n are invariant will, by itself, permit us to
individualize a first integral y1, for example which, as a result, becomes equal to one of the
integrals yi that are defined by formula (7). However, this is obviously impossible, because
equations (6) always admit a solution such that given numerical values of y1, ..., yn will
correspond to arbitrary numerical values of y1 , , yn .

159. The constants ciks play an important role in relation to the group G. They are what one
calls the structure constants of that group in group theory. The method of reduction that was
indicated above (sec. 156) is based precisely upon the decomposition of G into a normal series of
subgroups. The case where the matrix of ciks is simple corresponds to the simple groups.
One knows that the structure constants of a group are not arbitrary. One may verify this here
by saying that the exterior derivatives of the 1 , ..., n are zero. If we use the expressions (2)
for 1 , ..., n then the exterior derivative of s will be (sec. 73):

1,..., n
i=n
1,..., n

ciks ([i k ] [i k ]) =
( ik )
c i ci s + c i ci s + c i ci s [ ].
( ) i =1

One thus has the necessary relations:

i=n

c
i =1
i ci s + c i ci s + c i ci s = 0 (, , , s = 1, 2, ..., n).

7
( ) This is also true for an arbitrary sequence of invertible operations that are applied to 1, ..., n and which are
capable of being performed no matter what the nature of the coefficients of these forms.
 Equations with several linear integral invariants 169

In group theory, one proves that they are sufficient for the existence of a group that admits
the ciks for its structure constants.

III. - Examples.

160. Suppose that all of the constants ciks are zero. It is obvious then that since the forms 1,
..., n are exact differentials, the integration will require only n independent quadratures. Since
the forms 1, ..., n are reducible to:

1 = dy1 , 2 = dy2 , ..., n = dyn ,

the group G will have:
ys = ys + Cs (s = 1, 2, ..., n)
for its equations.
The preceding case presents itself when n = 1.
We look for all possible cases when n = 2. Along with the case that we just examined, one
may have:
1 = a [1 2],
2 = b [1 2],

in which the coefficients a and b are not both zero. Suppose, for example, that b 0 . If one
takes a2 b 1 to be a new form 1 then one will immediately see that one has:

1 = 0,
2 = [1 2 ] .
One quadrature gives:
1 = dy1 .

If one equates y1 to an arbitrary constant then 2 will become an exact differential, and a
second quadrature will complete the integration. By changing the notations slightly, one may
suppose that:
dy
1 = 1
y1
dy
2 = 2 .
y1

The group G will then have the equations:

y1 = C1 y1 ,
y2 = C1 y2 + C2 .
170 Lessons on integral invariants

161. We shall not discuss the general case for n = 3. We only point out the most interesting
case, in which one reduces formulas (2) to:

1 = [1 2],
2 = [1 3],
3 = [2 3].

In this case, the integration of equations (1) amounts to the integration of a Ricatti equation.
Indeed, consider the Pfaff equation:

(8) dt + 1 + t 2 + 12 t 2 3 = 0,

in which t is regarded as an unknown function of the variables both dependent and independent
that appear in the given differential equations. This equation is completely integrable. Indeed,
one verifies without difficulty that the exterior derivative of its left-hand side is zero if one takes
the equation itself into account (and if one uses the expressions for 1 , 2 , 3 ). As a result, as
one knows, one may convert this integration into the integration of an ordinary differential
equation that is obviously a Ricatti equation. Now, if one denotes a system of independent first
integrals of the given equations (1) by y1, y2, y3 then the expressions 1, 2, 3 may be expressed
by means of three quantities y1, y2, y3 and their differentials. The solution of equation (1) for
general t is thus a function of y1, y2, y3 (and an arbitrary constant C). As a result, if one has
integrated the Ricatti equation (8) in its classical form:

+ C
t=
+ C

then the mutual relations between the four functions , , , will furnish three first integrals of
the given equations, and one easily shows that they are independent.

IV. - Generalizations.

162. We nevertheless do not insist on the use of the foregoing theory, which, in order to be
appropriately developed, requires a very extended knowledge of the theory of groups. One sees
how the latter is necessarily introduced if one pushes the method of integrating differential
equations that admit given integral invariants to its limit. We point out only that the method that
was indicated in sec. 142 can be generalized to an arbitrary system of differential equations that
admit an invariant form, invariant Pfaff equations, etc. It consists of forming any linear integral
invariants that the given system of equations implies as independent first integrals by the
introduction of auxiliary variables. An example will suffice to make the spirit of this method
comprehensible.
Suppose that one has integrated a system of differential equations () in 4 variables:

1 = 2 = 3 = 0,
 Equations with several linear integral invariants 171

and each of these equations 1 = 0, 2 = 0, 3 = 0 are invariant for the given system. One
introduces three new variables u1, u2, u3, and one will consider the three forms:

1 = u11 , 2 = u2 2 , 3 = u3 3 .

The integration of the characteristic equations () of these three forms implies that of the
given differential equations by the elimination of u1, u2, u3 between the relations that define an
arbitrary solution of (). We then form the exterior derivatives 1, 2 , 3 . If one supposes that
one has:
1 a1[ 23 ] (mod 1 ),
2 a2 [ 21 ] (mod 2 ),
3 a3 [1 2 ] (mod 3 ),

with coefficients a1, a2, a3 that are functions of the original variables, then one will have:

a1u1
1 [ 2 3 ] (mod 1 ),
u2 u3
au
2 2 2 [ 31 ] (mod 2 ),
u3 u1
au
3 3 3 [1 2 ] (mod 3 ) .
u1u2
The coefficients:
a3 u1 au au
a1 = , a2 = 2 2 , a3 = 3 3
u2 u3 u3u1 u1u2

are thus first integrals of the system (), which is characteristic for the forms 1 , 2 , 3 . As a
result, the same is true for:
a2 a3
a2 a3 = ,
u1
and the form:
a2 a3 1 = a2 a3 1

is again an invariant form. However, it does not contain the auxiliary variables u1, u2, u3. It is
thus an invariant form for the given equations (), and the same is true for a3 a1 2 , a1a2 3 .
As a result, if any of the coefficients a1, a2, a3 are zero then the given differential system will
admit three invariant linear forms, and one is confronted with the problem that was treated in
this chapter.
Naturally, this will not always be the case, but in any case one will have the means to deduce
all of the possible consequences of the known information about the given equations.
 CHAPTER XVI

DIFFERENTIAL EQUATIONS THAT ADMIT

GIVEN INFINITESIMAL TRANSFORMATIONS

I. - Reduction of the problem.

163. We have already considered differential equations that admit infinitesimal

transformations, but these equations were assumed to admit an integral invariant or a Pfaff
invariant. We shall now take the most general viewpoint, which will furnish us, moreover, with
an illustration of the theories that were sketched out in the preceding chapter.
Consider a system of n ordinary differential equations (or a completely integrable system of n
Pfaff equations):
(1) 1 =2 = ... = n = 0,

and suppose that this system admits a certain number r n of infinitesimal transformations:

A1f, A2f, ..., Ar f .

We then look for the consequences that one might deduce from the knowledge of these r
infinitesimal transformations by integrating. This is a problem that has been solved by S. Lie.
We confine ourselves to its essential generalities.
Consider the matrix of quantities i (Ak) that is obtained by replacing the differentiation
symbol in the form i with the symbol of the infinitesimal transformation Ak f. Suppose that the
determinant that is formed from the first r rows and the first r columns of the matrix:

1 ( A1 ) 1 ( A2 ) 1 ( Ar )
2 ( A1 ) 2 ( A2 ) 2 ( Ar )
(2)

n ( A1 ) n ( A2 ) n ( Ar )

is zero. One may then substitute linear combinations of the left-hand sides of equations (1) for
those left-hand sides in such a way that the matrix becomes:
 Differential equations that admit given infinitesimal transformations 173

1 0 0
0 1 0
. . . .
(3) 0 0 1
0 0 0
. . . .
0 0 0

i.e., in such a manner that all of the i (Aj) are zero, except for:

1(A1) = 2(A2) = ... = r(Ar) = 1.

It is obvious that if n is greater than r then the new forms 1, ..., n will not be perfectly
determined; one may once more perform an arbitrary linear substitution on:

r+1, ..., n ,

and one may add an arbitrary linear combination of the r+1 , ..., n to each of the forms 1, ...,
r .
If equations (1) have been put into the form:

dy1 = dy2 = ... = dyn = 0

then it will be obvious that, since the quantities i (Aj) = Aj(yi) are presumably first integrals, the
new forms 1, ..., n that are obtained by reducing the matrix of i(Aj) to its canonical form can
always be presumed to have been constructed from the yi and their differentials. The following
two consequences result from this, along with what was said above:

1. Whenever the matrix of i(Ak) is reduced to its normal form (3), the Pfaff system:

(4) r+1 = ... = n = 0

will be an invariant system.

2. Each of the linear forms 1, ..., n is an invariant form, up to a linear combination of the
left-hand sides of the preceding invariant Pfaff system.

164. Before we proceed, we first remark that if the system (1) admits two infinitesimal
transformations Af and Bf then it will admit the infinitesimal transformation Cf whose symbol is
defined by:
Cf = A(Bf) B(Af).

Assume, with no loss of generality, that the symbols of the infinitesimal transformations
that one may deduce from the r given transformations, when taken pair-wise, are linear
174 Lessons on integral invariants

combinations of the A1f, ..., Ar f. In other words, suppose that one has:

s=r
(5) Ai ( Ak f ) Ak ( Ai f ) = iks As f (i, k = 1, 2, , r ).
s =1

With this hypothesis, we shall prove that the Pfaff system (4) is completely integrable.
In order to prove this, it is necessary that we return to the definition of the bilinear
covariant ( , ) of a linear form in the case that we have not considered up till now in which
the two differentiation symbols , are not interchangeable. If one sets:

() = a1x1 + a2 x2 + ... + an xn
then one will have:

( ) () = a1 ( x1 x1 ) + ... + an ( xn xn )
a a
+ k i ( xi xk xk xi ) ,
xi xk
or rather, if we agree to set:
=
then we will have:
(6) ( ) () = ( ) + ( , ) .

We apply this formula to the case in which the symbols and are replaced by the symbols
Ai f and Akf. It will then be convenient to replace with the symbol:

Ai (Ak f) Ak (Ai f) = iks As f.

Finally, suppose that one takes any one of the forms r+1, ..., n to be , which we may, as
we have seen, presume to be expressed in terms of y1, ..., yn and their differentials. One will
have:
r+ = c,u,r+ [ ] .
However:
r+ (Ai) = r+ (Ak) = r+ (As) = 0.

Therefore, one has the relation:

c,,r+ [ (Ai) (Ak) (Ai) (Ak)] = 0.

It results from this that the coefficients c,,r+ , will be zero when the indices , are both less
than or equal to r, since the preceding relation then obviously reduces to:

ci,k,r+ = 0 (i, k = 1, 2, ..., r).

 Differential equations that admit given infinitesimal transformations 175

As a consequence, since the exterior derivatives r+1, ..., n are all zero when one takes
equations (4) into account, system (4) will be completely integrable (sec. 101

II. - Case in which there are as many infinitesimal transformations as unknown functions.

165. Now suppose that the system (4), which is an absolutely arbitrary completely integrable
Pfaff system, has been integrated. Similarly, simply suppose that a solution of this system (4) is
known. An infinitude of solutions in the given system will correspond to that solution that are
obtained by integrating the equations:

(7) 1 = 2 = ... = r = 0.

This is a system for which one knows r invariant forms 1, ..., r . One is then reduced to the
problem that was treated in the preceding chapter.
Here, it is easy to determine, a priori, the coefficients ciks that enter into the expressions 1 ,
..., r :
1,,n
s = c
( ik )
iks [i k ] .

Indeed, we apply formula (6) after replacing the symbol with the symbol A , the
=r
symbol with the symbol A , and the symbol with

A
=1
in it. As all of the i (Ak) are

equal to 0 or to 1, i.e., to constants, formula (6) will reduce to:

0 = s + cs .
Therefore, one will have:
cs = s .

166. We now restrict ourselves to the case in which the coefficients iks are constants. In
this case, one proves that the given infinitesimal transformations A1f, ..., Ar f will generate an r-
parameter group whose structure coefficients are the iks . One sees that the system (7) falls
into the category of the two systems that we studied in the preceding chapter (sec. 156), and the
group G that corresponds to it has the same structure as the group that admits the given
differential system (7). This group G is the largest group that, when applied to the first integrals
y1, ..., yr , preserves the following rule: These integrals are interchangeable with the given
infinitesimal transformations.
Indeed, denote an arbitrary function of y1, ..., yr by f. It is obvious that one may determine
in one and only one manner r Pfaff expressions 1, ..., r such that one has identically i.e.,
f f
for any differentials dy1, ..., dyr and any arguments ,, :
y1 yr
176 Lessons on integral invariants

f f
df dy1 + + dyr = 1 A1 f + + r Ar f .
y1 yr

In this identity, if we replace the symbol of indeterminate differentiation d with the symbol
Akf then we will have:
Akf = 1(Ak) A1f + ... + k(Ak) Arf.

As a result, all of the i(Ak) will be zero except for:

1(A1) = 2(A2) = ... = r(Ar) = 1.

Finally, it results that the forms i are identical with the forms i . We then perform a
transformation of the group G on the y1, ..., yr, and these quantities become y1 , , yr . The
function f of the y1, ..., yr becomes a function f of the y1 , , yr , the symbols A1f, ..., Arf
become A1 f , , Ar f , and one will have:

df = 1 A1 f + + r Ar f .

However, since the i are forms in the yi and their differentials, just as the i are forms in
f
the yi and their differentials, the coefficient of in Ai f will be the same function of the
yk
f
y1 , , yr that the coefficient of in Aif is as a function of the y1, ..., yr . In other words, the
yk
given infinitesimal transformations transform the yi in the same way that they transform the yi .
Here, one sees once more that the group G is the largest group of transformations that
preserves the given data when applied to the first integrals.

III. - Application to second-order differential equations.

167. We have already treated the case n = r = 1 directly; let us take several other examples.
A second-order differential equation of the form:

d2y dy
2
=F
dx dx
is equivalent to the system:
dy y dx = 0,
dy F ( y) dx = 0,

which admits the two infinitesimal transformations:

 Differential equations that admit given infinitesimal transformations 177

f f
Af = , Bf = .
x y

In order to reduce this matrix of quantities i (Ak) to its normal form, it is necessary to take:

dy
1 = dx ,
F ( y )

ydy
2 = dy .
F ( y)

These two invariant forms are exact differentials, and one gets the desired general solution by
two independent quadratures:
dy ydy
x = C1 , y = C2 .
F ( y) F ( y)

Now, take a second-order differential equation of the form:

d2y dy
y 2 = F .
dx dx

It admits a translation that is parallel to the x-axis and a homothety with a center O that
correspond to the two infinitesimal transformations:

f f f
Af = , Bf = x +y .
x x y

The given equation is equivalent to the system:

dy y dx = 0,
y dy F ( y) dx = 0.

In order to make the matrix of i (Ak) normal, it is necessary to take:

x y xy
1 = dx dy dy ,
y F ( y)
dy y
2 = dy .
y F ( y)
Since one has:
A(Bf) B(Af) = Af

here, one will have, as is easily verified:

178 Lessons on integral invariants

1 = [1 2],
2 = 0.

As a result, the integration is carried out by means of two quadratures:

y dy
F ( y )
y = C1 e ,
y dy
1
x = C1 e F ( y) dy + C2 .
F ( y )

IV. - Generalizations. Examples.

168. In the case of system (1), one may arrive at n Pfaff equations that admit r infinitesimal
transformations:
A1f, ..., Ar f,

such that the rank of the matrix of i (Ak) becomes less than r (this is certainly the case if r > n).
Therefore let be the rank of this matrix, and suppose, as is permissible, that the determinant
that is constructed from the first rows and the first columns is not zero. For any index s, one
will then have r relations of the form:

s (Ar+1) = 11s (A1) + ... + 1 s (A),

...
s (Ar) = r,1 s (A1) + ... + = r, s (A).

The coefficients ij that were introduced into these relations are first integrals. Indeed, for
any linear combination of 1, ..., n in particular, for the differentials dy1, ..., dyn of n
independent first integrals one will have the same relations:

A+1 (ys) = 11 A1 (ys) + ... + 1 A (ys),

which implies that the values of 11, ..., 1 depend upon only the Ai (yk) ; i.e., upon the y1, ..., yn .
We will not pursue the general case further; it is based upon the same principles as the
preceding one.

169. EXAMPLE I. Consider the differential equation:

(1 + y2 )3/ 2 = Ry

of plane curves that have a ray of given curvature. It is equivalent to system:

 Differential equations that admit given infinitesimal transformations 179

1 dy y dx = 0,
2 R dy (1 + y 2 )3/ 2 dx = 0.

This system admits the three infinitesimal transformations that correspond to a translation
parallel to Ox, a translation parallel to Oy, and a rotation around O. One may calculate the
effects of these transformations on not only x and y, but also on y . One finds without difficulty
that:
f
A1 f = ,
x
f
A2 f = ,
y
f f f
A3 f = y + x (1 + y2 ) .
x y y

The matrix of quantities i (Ak) is the following one:

y 1 x + yy
.
(1 + y 2 )3/ 2 0 y (1 y ) + R(1 + y2 )
2 3/ 2

One therefore has:

R Ry
s ( A3 ) = y + s ( A1 ) + x s ( A2 ) .
2
1 + y 1 + y2

As a result, one obtains two first integrals of the given system by simple differentiations, and
the general solution is furnished by the formulas:

Ry
x= + C1 ,
1 + y 2
R
y= + C2 ,
1 + y 2
or
(x1 C1)2 + (y C2)2 = R2.

170. EXAMPLE II. Consider the third-order differential equation:

3 y y 2
y =
1 + y 2

that defines the plane curves of constant curvature. It is equivalent to the system:
180 Lessons on integral invariants

1 dy y dx = 0,
2 dy y dx = 0,
3 y y
3 dy dy = 0 .
1 + y 2

This admits four infinitesimal transformations, which correspond to a translation parallel to

Ox, a translation parallel to Oy, a rotation around O, and a homothety with center O. The
symbols of these transformations, which are regarded as operating on x, y, x , y , and y , are:

f
A1 f = ,
x
f
A2 f = ,
y
f f f f
A3 f = y + x + (1 + y2 ) + 3 y y ,
x y y y
f f f
A4 f = x + y y .
x y y

Here, the matrix of quantities i (Ak) is:

y 1 x + yy y xy
y 0 1 + y 2 + yy xy .
0 0 0 y

It is of rank 3, and the determinant that is obtained by taking the first, the second, and the
fourth column, for example, is non- zero. From this, one deduces the relations:

1 + y2 1 + y 2
s ( A3 ) = y +
s 1
( A ) + x y s ( A2 ) ,
y y

which lead to the two first integrals:

1 + y 2
u = x y ,
y
1 + y 2
v= y+ .
y

In order to continue the integration, choose linear combinations 1 , 2 , 3 such that the
principal determinant of the matrix of the i ( Ak ) is reduced to its normal form. In order to do
this, one may take:
 Differential equations that admit given infinitesimal transformations 181

1 3xy dy
1 = dx + dy + x ,
y 1 + y y
2

y 3 yy dy
2 = dy + dy + y ,
y 1 + y y
2

3 y dy dy
3 = .
1 + y 2 y
On the other hand:
du = 1 + u3 ,
dv = 2 + v 3 ,
and:
1 = [1 3 ] ,
2 = [2 3 ] ,
3 = 0 .

As a result, 3 is an exact differential, and one gets the missing first integral by a quadrature.
The general solution of the given equation is furnished by the formulas:

(1 + y2 )3/ 2
= C3 ,
y
C3 y
x= + C1
1 + y 2
C3
y= + C2 .
1 + y2

Here, one sees that the group G that preserves the given data is:

dC3
C1 = C1 , C2 = C2 , C3 = aC3 , because 3 = ,
C3

with an arbitrary constant a. It is because of the fact that there is only one parameter that the
integration reduces to a quadrature. In the preceding exercise, the group G reduced to the
identity transformation, and the solution was obtained without integration

171. REMARK. For all of the examples in which one arrives at n invariant linear forms,
one obtains integral invariants of all degrees by constructing an arbitrary exterior form with
constant coefficients from the 1, ..., n . This is why one has the invariant integral 1 2 3
182 Lessons on integral invariants

in the latter exercise, which, if one is limited to sets of states that correspond to the same value of
x, reduces to:
dy dy dy
y2 .
As a result, if one considers an arbitrary family of circumferences that depend on three
parameters, and if one cuts the circles of that family by an arbitrary parallel to the y-axis then the
dy dy dy
integral , when taken over the family of circles under consideration, will be
y2
dC dC dC
independent of x. It will be, moreover, equal to 1 2 3 if one denotes the coordinates
C3
of the center by C1 and C2 and those of the ray by C3 .
 CHAPTER XVII

APPLICATION OF THE PRECEDING THEORIES

TO THE n-BODY PROBLEM

I. - Reduction of the number of degrees of freedom.

172. We have already seen (sec. 123) how the method of integration that was discussed in
Chapter XII is applied to the canonical equations of dynamics:

dqi H dpi H
= , = .
dt pi dt pi

We suppose the function H is arbitrary. If that function is independent of time then the
function H will be a first integral (sec. 92), and one is left with the integration of the equations:

dqi dp
= i ,
H H
pi qi

whose first integrals are solutions of the equation:

(H f) = 0
and a quadrature.

173. A little later on, we shall study the reduction in the integration of the n-body problem
that is produced by accounting for the previously determined (sec. 93) infinitesimal
transformations that the equations of motion admit. We suppose, as is permissible, that the
system of n bodies is referred to its center of gravity, i.e., that the 3n coordinates xi , yi , zi , and
the 3n velocity components xi , yi , zi are coupled by the relations:

mi xi = 0, mi yi = 0, mi zi = 0,
mi xi = 0, mi yi = 0, mi zi = 0.

Let U be a function of forces that is assumed to be homogeneous and of degree p with respect
to the coordinates. The equations of motion then admit the five infinitesimal transformations:

f
A0 f =
t
184 Lessons on integral invariants

f f f f
A1 f = yi zi + yi zi ,
zi yi zi yi

f f f f
A2 f = zi xi + zi xi ,
xi zi xi zi

f f f f
A3 f = xi yi + xi yi ,
yi xi yi xi

f f f p f f f p f
A4 f = xi + yi + zi xi + yi + zi + 1 + t .
xi yi zi 2 xi zi zi 2 t

On the other hand:

= {mi [ xi xi] + mi [ yi yi] + mi [ zi zi]} [H t],

if one sets:
1
H=
2
mi ( xi2 + yi2 + zi2 ) U .

174. The five invariant linear forms:

i = (Ai , )
are:
0 = H ,
=H ,
1 1

2 = H 2 ,

3 = H 3 ,
(1)
= m ( x x + y y + z z + p x x + p y y + p z z )
4 i i i i i i i 2 i i 2 i i 2 i i

p
+ 1 + t H + pH t ,
2

in which one sets:

H1 = mi ( yizi zi yi ),

(2) H 2 = mi ( zixi xizi ),
H = m ( x y y x ).
3 i i i i i

Finally, one has:

 Application of the preceding theory to the n-body problem 185

p
4 = A4 ( ) = 1 .
2

The matrix of quantities aij = (Ai , Aj) has already been addressed in the most general case
(sec. 95). We reproduce it below.

0 1 2 3 4

0 0 0 0 0 pH

1 0 0 H3 H2 (1 ) H
p
2 1

2 0 H3 0 H1
(1 ) H
p
2 2

3 0 H2 H1 0 (1 ) H
p
2 3

4 pH ( p
2 )
1 H1 ( p
2 1 H2) ( p
2 1 H3) 0

175. We now recognize five invariant linear forms, and the matrix of coefficients aij is
defined by the operation of the generalized Poisson brackets:

N [ N 1 i j] = aij [ N ].

We apply the theory of Chapter XII (sec. 125). Construct the auxiliary form:

0,1...,4
( ) = a [ ].
( ij )
ij i j

It may be written:

p2
() = pH [4 0] + [4 (H1 1 + H2 2 + H3 3)] + H1 [2 3] + H2 [3 1] + H3 [1 2].
2

It is of rank 4, and its reduction to normal form:

= [ 4 0 ] + [ 1 2 ]
can be accomplished by setting:

p2
0 = pH0 + (H1 1 + H2 2 + H3 3),
2
186 Lessons on integral invariants

4 = 4 ,
1 = 1 1 + 2 2 + 3 3 ,
2 = 1 1 + 2 2 + 3 3 ,
3 = 0 ,

in which the i and i are chosen as is always possible in such a manner as to make:

2 3 2 3 = H1 , 3 1 3 1 = H2 1 2 1 2 = H3 .

One may add the supplementary conditions:

1 1 + 2 2 + 3 3 = 0,
12 + 22 + 32 = 12 + 22 + 32 = H12 + H 22 + H 32 .

However, if one defines five linear forms 0, 1, 2, 3, 4 by the identity:

0 0 + 1 1 + 2 2 + 3 3 + 44 = 0 0 + 1 1 + 2 2 + 3 3 + 4 4

then one will obtain, without difficulty:

4 = 4 ,

2 H1dH1 + H 2 dH 2 + H 3dH 3
0 = ,
p2 H12 + H 22 + H 32

2 pH H1dH1 + H 2 dH 2 + H 3dH 3
3 = dH ,
p2 H12 + H 22 + H 32

1dH1 + 2 dH 2 + 3dH 3
1 = ,
H12 + H 22 + H 32

1dH1 + 2 dH 2 + 3dH 3
2 = .
H12 + H 22 + H 32

When the auxiliary form has been reduced to [ 4 0 ] + [ 1 2 ], we will have:

= [40] + [12] + [53] + [ 6 7] + ...;

i.e., when we perform the calculations:

 Application of the preceding theory to the n-body problem 187

2 H1dH1 + H 2 dH 2 + H 3dH 3
= 4
p2 H12 + H 22 + H 32
(3)
H [dH 2 dH 3 + H 2 [dH 3 dH1 ] + H 3[dH1dH 2 ]
+ 1 + ,
H 2
+ H 2
+ H 2
1 2 3

in which we have set:

2 pH H1dH1 + H 2 dH 2 + H 3dH 3
(4) = 5 dH + [67 ] +
p2 H12 + H 22 + H 32

176. If one equates the four first integrals H, H1, H2, H3 to arbitrary constants then the rank
of will be reduced by six units. It thus passes from 6n 6 to 6n 12, which corresponds to a
problem with 3n 6 degrees of freedom (which will be 3 in the case of the three-body problem)
as a consequence. However, the corresponding characteristic system contains arbitrary
parameters.
There is a (theoretical) procedure for reducing the number of degrees of freedom while
completely avoiding the introduction of arbitrary parameters. After annulling the exterior
derivative in the right-hand side of equation (3) and taking into account the relation:

2 p
4 = ,
2
one will obtain:
H dH + H dH + H dH
= 1 1 2 2 2 2 2 3 3 .
H1 + H 2 + H 3

This relation expresses the idea that the exterior derivative of the quadratic form:

1 1 2 H1 dH1 + H 2 dH 2 + H 3 dH 3
= 4
H12 + H 22 + H 32 H12 + H 22 + H 32 p2 ( H12 + H 22 + H 32 )3/ 2

H1[dH 2 dH 3 ] + H 2 [dH 3 dH1 ] + H 3[dH1 dH 2 ]

( H12 + H 22 + H 32 )3/ 2

2 1 H [dH dH ] + H [dH dH ] + H [dH dH ]
= 4 1 2 3 2 3 1 3 1 3

2 p ( H1 + H 2 + H 3 )
2 2 2 3/ 2
H1 + H 2 + H 3
2 2 2

188 Lessons on integral invariants

is zero. This property is, moreover, evident in the right-hand side of the preceding equality,
whose first term has a zero exterior derivative since it is an exact exterior derivative. We shall
see that the same is true for the second term.
In order to interpret the second term, consider the vector (OS) of length = H12 + H 22 + H 32 ,
which represents the kinetic moment of the system with respect to the origin, and which has H1,
H2, H3 for its projections. If one imagines a surface element d that is described by the point S,
and if one calls the direction cosines of the normal to that surface element 1, 2, 3 then one
will have:
[dH2 dH3] = 1 d, [dH3 dH1] = 2 d, [dH1 dH2] = 3 d,

and, as a result:

H1[dH 2 dH 3 ] + H 2 [dH 3 dH1 ] + H 3[dH1 dH 2 ]

( H12 + H 22 + H 32 )3/ 2
H + 2 H 2 + 3 H3 cos d
= 1 1 d = = d ,
2 2

in which represents the angle that OS makes with the normal to the element, and d
represents the solid angle that the surface element subtends at the origin. If one denotes the co-
latitude (viz., the angle with Oz) and the longitude (viz., the angle between the plane zOS and the
plane xOz) of the point S by and , respectively, then one will have, on the one hand:

d = sin [d d] .

As a result, the form under consideration can be regarded as the exterior derivative of the
linear form cos d.
We then set:
2 4
(5) = + cos d .
2 p

We see that the characteristic equations of the relative integral invariant are:

2p d
dH H = 0,
(6) p2
= 0, = 0, = 0,
5 6 7

II. - Equations of motion with respect to a moving reference frame.

177. It is easy to interpret this system. In order to fix ideas, we place ourselves in the case of
the three-body problem of celestial mechanics (p = 1). First, we calculate the quantities 5 (Ai),
 Application of the preceding theory to the n-body problem 189

6 (Ai), etc. In order to do this, we shall apply the operation that corresponds to Ai f to both sides
of formula (3), while writing it in terms of dH, 6, 7, ..., 11 . We immediately have:

dH = 5(A0) dH 7(A0) 6 + 6(A0) 7 + ... + 10(A0) 11,

0 = 5(A1) dH 7(A1) 6 + 6(A1) 7 + ... + 10(A1) 11,
...
0 = 5(A4) dH 7(A4) 6 + 6(A4) 7 + ... + 10(A4) 11.

We therefore find that all of the (Ai) are zero for 5, except for 5(A0), which is equal
to 1.

Ai
On the other hand, we remark that the quantities Ai H + 2 H are all zero, since the function K

= H2 will be invariant for each of the infinitesimal transformations under consideration. We see
1
that the characteristic system (6) of can be defined by all linear combinations of the

equations of motion that enjoy the property of being verified identically when one replaces the
symbol of undetermined differentiation in them with the symbol of any one of the infinitesimal
transformations A1 f, A2 f, A3 f, A4 f.

178. This result is what permits us to geometrically interpret the system (6).
In order to do that, imagine different possible reference systems, each of which is defined by
three rectangular coordinate axes, a time origin, and units of length, time, and mass. We fix the
unit of mass once and for all, and impose the condition on the other units that the constant of
universal attraction must have a fixed numerical value. The unit of length is also arbitrary.
Finally, we fix the origin of the axes, which will be the center of gravity of the three-body
system, as well as the time origin.
The remaining reference systems depend on four arbitrary parameters. Three of them fix the
orientation of the axes, and the fourth one fixes the units.
One may make a reference system correspond to each state of the three bodies (as defined by
their positions, velocities, and time, and depending on 13 variables) according to a law that is
determined in advance, in such a manner as to reduce the number of quantities that fix the state
of the three bodies with respect to that reference system by 4 units. For example, one may
choose the line that joins the center of gravity to the body A1 to be the x-axis, the plane of the
three bodies to be the xy-plane, and the distance OA1 to be the unit of length. The state of the
three bodies is then defined by the two coordinates of A2, the six projections of the velocities of
A1 and A2 onto the three axes, and finally, the time t. One may fix the choice of moving
reference system that corresponds to a given state by another law. While always taking Oz to be
perpendicular to the plane of the three bodies, one may take Ox to be parallel to A1 A2, and take
the length of the side A1 A2 to be the unit of length. One may also choose the axes according to
either of the preceding laws, but choose the units in such a manner as to make the measure of the
kinetic moment OS equal to 1. One may also take Oz to be perpendicular to the plane of the
three bodies, take the plane zOS to be the xz-plane, and choose the units in such a way as to make
190 Lessons on integral invariants

the measure of OS equal to 1. Under this latter hypothesis, the nine quantities that determine the
state of the three bodies with respect to the moving reference system will be the two coordinates
of A1, the two coordinates of A2, time, and finally, the six components of the velocities of A1 and
A2, which makes 11 quantities, although they will be coupled by the two relations = 1, = 0.
Now suppose that we have made a choice of correspondence between each three-body state
and a moving reference frame by one of the preceding laws or any other law imaginable, and let:

q1, q2, ..., q9

be the nine quantities that determine the state of the three bodies with respect to the
corresponding moving reference system. The state of the three bodies will be determined with
respect to a fixed reference system if one knows, along with q1, ..., q9, the four parameters u1, u2,
u3, u4 that define the position of the moving reference system with respect to the fixed reference
system. These four parameters will be, for example, the three parameters that depend on the nine
direction cosines and the ratio of the moving unit of length to fixed unit of length. Finally, the
quantities (19 in number, but reducing to 13):

xi , yi , zi , xi , yi , zi , t

that fix the state of the three bodies with respect to the fixed reference system are definite
functions of the 13 quantities:
q1, q2, ..., q9, u1, u2, u3, u4.

Conversely, the latter are definite functions of the former. However, when the 9 quantities
q1, ..., q9 are considered to be functions of the xi , yi , zi , xi , yi , zi , t they will obviously be
invariant under each of the infinitesimal transformations A1 f, ..., A4 f, because performing one of
these transformations amounts to changing the fixed system of reference and then altering the
quantities u1, u2, u3, u4 that define the relation between the moving reference system and the
fixed reference system, but without altering the quantities qi that define the state of the three
bodies with respect to the moving reference system.
One may also say that if one looks for all of the linear differential forms in dxi , dyi , ..., dyi
that enjoy the property of being annulled when one replaces the symbol d in them with the
symbols A1 f, ..., A4 f then one will find all of the linear combinations in dq1, ..., dq9, and uniquely
at that.
In particular, equations (8) for the characteristic system of have left-hand sides that are
linear in dq1, ..., dq9. Since they are 8 in number, these equations (6) may be put into the form:

dqi Ci dq9 = 0 (i = 1, 2, ..., 8),

and, since they are completely integrable the Ci depend only on the qi . In other words, we say
that system (6) is a system of ordinary differential equations in q1, ..., q8 . It therefore defines the
motion of the three bodies with respect to the moving reference frame.
 Application of the preceding theory to the n-body problem 191

179. It is now easy to effectively form the equations of system (6). In order to do this, we
start with the relative integral invariant , in which we have set:

2
= 4 + cos d ,

and imagine that we have expressed all of the quantities xi , yi , ..., zi , t in terms of the q1, ..., q9,
u1, ..., u4. First of all, we know that the differentials du1, du2, du3, du4 cannot appear in the
defining expression for , which can be constructed in terms of the linear forms dqi Ci dq9 ,
uniquely. As a result, in order to calculate we can regard u1, ..., u4 as fixed parameters.
Moreover, when the coefficients of the form are expressed in terms of the dq1, ..., dq9 they
cannot contain the variables u1, ..., u4, since then the exterior derivative would not be zero
then. In order to carry out the calculation, one can therefore not only regard u1, ..., u4 as fixed
parameters, but one can also give them arbitrary numerical values. In particular, one may thus
give them numerical values that correspond to the case in which the fixed reference system
agrees with the moving reference system. In other words, in order to form , one may give the
values Xi, Yi, ..., Z i , T, which define the state of the three bodies with respect to the moving
reference system, to the quantities xi , yi , ..., zi , t in . As we have seen, these thirteen quantities
reduce down to nine.

180. In particular, we examine the case in which the moving unit of length is chosen in such
a manner as to reduce by one unit (the moving unit of length is then found to be fixed). In this
case, one has:
= 24 + cos d.

If one adds an exact differential then one will get the relative integral invariant + cos d
for the desired differential equations.
Since the z-axis is assumed to be normal to the plane of the three bodies, and the x-axis is
assumed to be parallel to A1A2 , for example, the position of the triangle will depend on three
quantities 1, 2, 3. However, one will have:

+ cos d = mi ( X i dXi + Yi dYi) H dT + cos d

= 1 d1 + 2 d2 + 3 d3 + 4 d4 H dt,
in which we have set:
4 = , 4 = cos .

The desired equations of relative motion are then:

di H di H
= , = (i = 1, 2, 3, 4).
dt i dt i
192 Lessons on integral invariants

They are canonical and admit the first integral H = const.

For example, one may take1, 2, 3 to be the lengths of the sides of the triangle A1 A2 A3 .
If one supposes that the motion is planar then will be zero, and there will be no more than
six unknown functions 1, 2, 3, 1, 2, 3 .

181. Once the motion of the three bodies with respect to the moving reference system is
known, the absolute motion will be determined by a quadrature. Indeed, if we first know the
projections of the kinetic moment OS onto the moving axes then we will know the ratio of the
moving units to the fixed ones by giving it the constant number C that measures OS with respect
to the fixed reference. One may then take OS to be the fixed z-axis, where the position of the
fixed axes depends on an unknown angle. This angle will be given by a quadrature. Indeed, it
suffices to remark that the invariant form 4 (when expressed by means of the fixed coordinates)
is an exact differential when one takes into account the relations that are assumed to be obtained,
and which define the relative motion. Indeed, the formula:

d H [dH 2 dH 3 ] + [ H 2 dH 3 dH1 ] + H 3 [dH1 dH 2 ]

24 = = 2 4 + 1 +
2

shows that under these conditions 4 is zero (since H1 and H2 are zero). The integration is thus
accomplished by means of the formula:
4 = const.

One may remark that this quadrature may be performed when one has determined the motion
from only the geometric viewpoint without having found the time (by a quadrature, as one
knows). In other words, the two quadratures that give time (in the relative motion) and the
orientation that defines the fixed axes with respect to the moving axes can be performed
independently of each other.

III. - Case in which the area constants are all zero.

182. The preceding theory essentially assumed that H12 + H 22 + H 32 0 . We study the
motions for which all three area constants will be zero. In this case, it is necessary to suppose
that the 18 quantities:
xi , yi , zi , xi , yi , zi
are not only coupled by the relations:

mi xi = 0, mi yi = 0, mi zi = 0,
mi xi = 0, mi yi = 0, mi zi = 0,

but also by the relations:

 Application of the preceding theory to the n-body problem 193

mi ( yi zi zi yi) = 0, mi ( zi xi xi zi) = 0, mi ( xi yi yi xi) = 0.

It is easy to show that the plane of the triangle of the three bodies remains fixed, because the
components of the three velocities that are normal to the plane are all zero, at least if the three
bodies are not all in a straight line.
We may thus suppose that the zi and the zi are zero, and there then remain five relations
between the 12 quantities xi , yi , xi , yi , namely:

mi xi = 0, mi yi = 0,
mi xi = 0, mi yi = 0,
mi ( xi yi yi xi) = 0.

All totalled, there are then seven dependent variables and one independent variable (time).
However, , which is of even rank, cannot have a rank that is equal to the number of
differential equations of motion. Therefore, the characteristic system of does not agree with
that of the equations of motion.
Here we have three infinitesimal transformations A0 f, A3 f, A4 f with:

(0, ) = H, (3, ) = 0, (4, ) = 4.

The seven differential equations of motion can then be put into the form of Pfaff equations:

1 = 0, 2 = 0, ..., 7 = 0,
and one may suppose that:
1(A3) = ... = 6(A3) = 0, 7(A3) = 1.

The form , which can be expressed by means of the 1, ..., 7, certainly does not contain
7, since otherwise the form (A3, ) would not be identically zero. As a result, the
characteristic system of will be the completely integrable system:

1 = 2 = = 6 = 0.

It gives the motion of the three bodies independently of the orientation of the triangle of the
three bodies around the center of gravity. Once this system is integrated, the integration will be
given by a quadrature. Indeed, the relation 7(A3) = 1 assures us that 7 will have the property
of being an invariant form for the differential equation 7 = 0.
Now, return to the form of rank 6. Its characteristic system admits the two infinitesimal
transformations A0 f and A4 f, which give rise to two invariant linear forms 0 = H, which we
denote by 1, and 4, which we denote by 2. We assume, as is permissible, that one has (A0)
= (A4) = 0 for each of the forms 3, , 6. A calculation that is analogous to the one that was
made in the general case gives:
1 1
= [ 1 2 ] + = [ H 2 ] + ,
H H
194 Lessons on integral invariants

in which has rank 4 and is formed from 2, 4, 5, 0 .

As we saw above, one has = 2 4 = 2 2 , and, as a result:

H 2
= 2 2 + 2 = ( H 2 ) .
H H

The form 1
2
H is therefore an exact derivative; as a result it admits the following equations:

(7) 3 = 4 = 5 = 6 = 0

for its characteristic system.

This system may be integrated by some equations of order:

4, 2, 0.

By definition, the equations of motion will be given by operations of order 4 and 2 after two
quadratures.
We remark that from the expression (1) for 4 = 2, the form H 2 , which plays the role of
relative invariant for system (7), is equal to:

1 1
H 2 = H mi ( xi xi + yi yi + xi xi + yi yi ) + ( H 3/ 2t ) .
2 2

System (7) is easy to interpret: It gives the motion of the three bodies with respect to a
moving reference system, which one may make to correspond with each state of the three bodies
according to a determined law for which the origin of time no longer is necessarily fixed. For
example, one may choose the actual instant for the moving time origin, and fix the unit of length
by the condition that the energy H will have a given fixed numerical value. The equations of the
system are obtained by starting with the form H 2 in which one includes the moving
coordinates. Obviously, under the hypotheses envisioned, one may substitute the form:

mi ( xi xi + yi yi).

Here, the quantities of motion of the three bodies form a system of vectors that is equivalent
to zero. One may thus regard the quantity of motion of the bodies Ai as the resultant of two
vectors ui and uk that are directed along the sides Ai Ak and Ai Aj and counted positively when they
are projections of Ak Ai and Aj Ai . One then has, upon denoting the three sides of the triangle by
r1, r2, r3:
= u1 r1 + u2 r2 + u3 r3 .
One has, moreover:
 Application of the preceding theory to the n-body problem 195

1 2 r22 + r32 r12 m m m m mm

H= 2
u + u3
2
+ u u
2 3 + f 2 3+ 3 1+ 1 2 = h.
2m1 r2 r3 r1 r2 r3

The equations of relative motion are then:

dr1 dr dr du1 du2 dr3

= 2 = 3 = = = .
H H H H H H
u1 u2 u3 r1 r2 r3

IV. - Case in which the vis viva constant is zero.

183. The preceding theory implicitly supposed that the vis viva constant was non-zero. If we
suppose that it is zero then the variables will be subject to a new relation:

1
2
mi ( xi2 + yi2 ) U = 0 .
There are now only six dependent variables and one independent variable. The invariant
form (A0 , ), is identically zero here, just as the form (A3, ) is.
The system of equations of motion may be put into the form:

1 = 2 = ... = 6 = 0,
and one may suppose (sec. 163) that:

1(A0) = 2(A0) = ... = 4(A0) = 0, 5(A0) = 1, 6(A0) = 0,

1(A3) = 2(3) = ... = 4(3) = 0, 5(3) = 1, 6(3) = 0.

When the form is expressed in terms of the i , it will obviously contain neither 5 nor 6 .
Finally, suppose that:
2(A4) = 3(4) = 4(4) = 0, 1(A4) = 1,

and (A4, ) = 2 . One will have:

= 2 2 = [12] + [34].

The form 2 is of the second type, and the equations:

2 = 3 = 4 = 0

form a completely integrable system that is characteristic for the equation 2 = 0. They define
the motion of the three bodies with respect to a moving reference system whose time origin is
196 Lessons on integral invariants

variable. Here, for example, one may choose the side r3 of the triangle to be a unit of length.
The equations to be integrated then constitute the characteristic system of the Pfaff equation:

2 d(u1 r1 + u2 r2 + u3) u1 dr1 u2 dr2 = 0,

in which the quantities u1, u2, u3, r1, r2 are coupled by the relation:

m m mm
1
2m
( )
u22 + u32 + 2u2u3 cos A1 + f 2 3 + 3 1 + m1m2 = 0 .
r1 r2
If one sets:
r1 = x, r2 = y, u1 r1 + u2 r2 + u3 = z, u1 = 2p, u2 = 2q

then one will be reduced to the integration of the first order partial differential equation:

1 1 2 1 1 2 2 x2 + y 2 1
2 + p + 2 + q + pq
m2 m3 m1 m3 m3 xy
p x2 + 1 y 2 q y 2 + 1 x2
+( z 2 px 2qy ) +
m2 x m1 y
1 1 1 m2 m3 m1 m3
+ + ( z 2 px 2qy ) f
2
+ + m1 m2 = 0 .
2 m1 m2 x y

Once this equation has been integrated, one will obtain the general solution of the
characteristic system of by differentiations because once 2 has been put into the form Z1 dY1
+ Z2 dY2 , one will deduce the first integrals Y1, Y2, Z1, Z2 of this system from it by differentiation.
However, the equations of motion are not, moreover, completely integrable. It is necessary
to integrate the equations:
5 = 6 = 0.

They constitute a system of differential equations that admit the two infinitesimal
transformations A0 f, A1 f, and the matrix:

5 (A 0 ) 5 (A3 )
6 (A 0 ) 6 (A 3 )
is reduced to its normal form:
1 0
,
0 1

precisely. On the other hand, because one has:

A0 (A3 f) A3 (A0 f) = 0,
 Application of the preceding theory to the n-body problem 197

the two transformations A0 f, A2 f will be interchangeable, and one will then have:

5 = 0,
6 = 0.

As a result, the integration is accomplished by means of two independent quadratures: One

gives the orientation of the triangle A1 A2 A3 , and the other one gives the time.
 CHAPTER XVIII

INTEGRAL INVARIANTS AND THE CALCULUS OF VARIATIONS

I. - Extremals attached to a relative integral invariant.

184. We have already seen in chapter 1 (sec. 9) that the differential equations for the
extremals of the integral:
I = F(q1, ..., qn ; q1 , ..., qn ; t) dt,

agree with the characteristic equations of the relative integral invariant , when one sets:

i =n
F i = n F
= qi qi F t ,
i =1 qi i =1 qi

and regards q1, ..., qn , q1 , ..., qn , t, as 2n + 1 independent variables.

In the calculus of variations, one regards q1, ..., qn as arbitrary functions of t and q1 , ..., qn ,
as their derivatives. In the n + 1-dimensional space of (q1, ..., qn , t), any extremal will enjoy the
property that when the integral I is taken over a given arc of that curve it will be stationary when
compared to the arcs of infinitely close curves that admit the same origin and the same extremity.
However, one may also place oneself in the 2n + 1-dimensional space of (q1, ..., qn , q1 , ..., qn ,
t). An extremal curve will then enjoy the property that when I is taken over a given arc of that
curve it will be stationary when compared to infinitely close curves for which the initial and final
values of only the coordinates q1, ..., qn , t are the same as for the given extremal. If one takes the
second point of view then q1 , ..., qn will be functions of t that have no relationship to the
derivatives of q1, ..., qn with respect to t, a priori.

185. More generally, start with a linear differential form in 2n + 1 variables. Suppose that
the form is of rank 2n and finally suppose that n of the coefficients of the differentials in
are identically zero. We may therefore set:

= a1 t + a2 x2 + ... + an tn b t,

since the quantities a1, ..., an , b are functions of the 2n + 1 independent variables x1, ..., xn , y1, ...,
yn , t.
The characteristics of the relative integral invariant or, what amounts to the same thing,
the quadratic exterior form are given by a system of ordinary differential equations:
 Integrl invariants and the calculus of variations 199

dxi dyi
(1) = Xi , = Yi ,
dt dt

if one supposes, as we do, that t is not a first integral of the characteristic equations.
Having said this, consider an arc of a curve in 2n + 1-dimensional space that connects the
point M ( xi0 , yi0 , t 0 ) to the point M ( xi1 , yi1 , t1 ) , and form the integral:

M1
I = a1 dx1 + a2 dx2 + b dt .
M0

We calculate the variation of that integral when one passes from the arc of the curve under
consideration to an arc of a curve infinitely close to the given one that connects the point
( xi0 + xi0 , yi0 + yi0 , t 0 + t 0 ) to the point ( x1i + xi1 , y1i + yi1 , t1 + t1 ) . We will have:

M1 M1
I = [ ]10 + (d d ) = [ ]10 + (d , ) .
M0 M0

If we would like the integral to be stationary relative to all of the curves that are infinitely
neighboring close to the given curve then it will be necessary to have:

(d, ) = 0

when one displaces along the arc of the given curve for any xi , yi , t. In other words, it is
necessary that the arc of the curve must belong to a characteristic of the form . The value of
the integral will be stationary for all the arcs of the infinitely close curves on which is zero at
the origin and the curve extremity; i.e., the arcs on which x1, ..., xn , t has the same initial and
final values as the arc of the given curve.

186. Now suppose that we restrict the field of curves that are infinitely close to the given
curve to curves for which the functions xi , yi , and t satisfy the first n characteristic equations:

dxi
(2) = X i.
dt

We assume that the n functions X1, ..., Xn are independent of the y1, ..., yn , which permits us
to take arbitrary functions of t for the xi . Finally, we suppose that the initial and final values of
y1, ..., yn , t are the same for the varied curves as for the primitive curve. With these conditions,
one has:
I = (d, ).

It is easy to see that there is no term in dy1, ..., dyn in (d, ). Indeed, the coefficient of y1
will be:
200 Lessons on integral invariants

a1 a a b
dx1 + 2 dx2 + + n dxn dt .
y1 y1 y1 y1

If one takes the characteristic equations (2) into account then this coefficient will necessarily
be annulled. It is therefore zero when one displaces along the extremal. Since only x1, ..., xn
enter under the sign in the expression for I, the coefficients of x1, ..., xn , t will become
zero. As a result, the extremals will be given by the characteristic equations of .

II. - The least-action principle of Maupertuis.

187. Suppose that the Hamiltonian function H is independent of time. Consider the set of
motions for which the function H has a given constant value h. The corresponding trajectories
are the characteristics of the linear integral invariant , with:

i=n
= pi qi h t ,
i =1

or, what amounts to the same thing, of the integral invariant , with:

i=n
= pi xi .
i =1

Indeed, the form differs from only by an exact differential. This form is constructed
from 2n variables, which are coupled by the relation:

H = h,

and only n of these coefficients are zero. The characteristic equations are:

dq1 dq dp1 dpn

= = n = == .
H H H H
p1 pn q1 qn

As a result, the trajectories are extremals of the integral:

p1 dq1 + ... + pn dqn

in the 2n 1 dimensional space (qi , pi), whether one considers all of the curves for which the
initial and final values of q1, ..., qn are given or only those curves that satisfy the equations:
 Integrl invariants and the calculus of variations 201

dq1 dq
= = n ,
H H
p1 pn
and, of course, the condition that H = h.

188. Take the second viewpoint, for example. Suppose that the qi are the position
parameters of the system, and that the pi are the components of the quantity of motion. If one
denotes the vis viva by 2T and decomposes it into terms of second degree, first degree, and
degree zero, in q1 , ..., qn then one will have:

T
H = qi T U = T2 T0 U .
qi

Substitute the variables qi for the pi . By hypothesis, one has:

T2( q ) = T0 + U + h,

i =n
T i =n
T
= qi = 2 qi + T1 ( qi ) .
i =1 qi i =1 qi

Finally, suppose that one has:

dq1 dq2 dq T (dq) T2 (dq )

= = = n = 2 = .
q1 q2 qn T2 (q) T0 + U + h

The quantity under the sign in the integral I is:

T2 T2 (dq)
d = qi + T1 (dq) = 2(T0 + U + h) 2T2 (dq) + T1 (dq).
qi T0 + U + h

Therefore, if one sets:

2T = aij qi qj + 2 bi qi + 2T0

then one will arrive at the following theorem, which constitutes the principle of least action in
the sense of Maupertuis:

Trajectories are extremals of the integral:

( 2(T0 + U + h) aij dqi dq j + bi dqi )

202 Lessons on integral invariants

relative to all of the infinitely close trajectories, subject to the constraints that they have the
same initial and final position of the system and satisfy the vis viva theorem H = h, with a given
vis viva constant.

One recovers the principle in its classical form when T1 = T0 = 0

189. EXAMPLE. In the case of a free moving point that is referred to fixed axes, the
trajectories will be extremals of the integral:

2(U + h) ds.

If the point is referred to axes that are rotating around Oz with a constant angular velocity ,
and if, moreover, the time-independent force field is directed along the axes then one will have:

2T = m [( x + y)2 + ( y x)2 + z 2 ]
= m ( x2 + y 2 + z 2 ) 2m (x y y x ) + m2 (x2 + y2).

For a point of mass 1, the trajectories will be extremals of the integral:

2 ( x 2 + y 2 ) + 2U + 2h ds ( x dy y dx ).

III. - Generalizations.

190. Everything that was done in the case where time does not explicitly enter into H can
also be done in the case where H does not contain one of the other variables qi and pi . To fix
ideas, take the case of a free material point of mass 1 that is subject to a central force that is a
function of the distance. Consider all of the motions that it makes around the given plane, which
we take to be the xy-plane, and which obey the law of areas with a given constant C. The real
motions are given by a system of differential equations that admit the relative integral invariant:

= r r + r ( r 2 + 12 r 2 2 U ) t ,
2 1
2

which obviously reduces to:

C2
= r r 12 r + 12 U dt.
2
2
r

The form depends upon only the variables r, r , and t, and one of its characteristic
equations is:
 Integrl invariants and the calculus of variations 203

dr
= r.
dt

Iresults from this that if one is given the values r0 and t0 for initial conditions and r1 and t2 for
final conditions then the actual motion that satisfies these conditions will be the one that makes
the integral:
t1
1 2 1 C2
t1
1 dr 2 1 C 2
L = r dr 2 r + 2 2 U dt = 2 2 2 U dt
t0 r dt
t0
r

stationary relative to all of the infinitely close motions that satisfy the same conditions at the
limits and verify the law of areas with the area constant C

IV. - Application to the propagation of light in an isotropic medium.

191. Consider an isotropic medium whose index of refraction n is known at each point.
Fermats principle then leads us to define light rays as the extremals of the integrals:

n ds = n dx 2 + dy 2 + dz 2 .

If one introduces an auxiliary variable t then one will be dealing with the case of an integral:

F(x, y, z; x , y , z , t),
with:
F = n x2 + y 2 + z 2 .

The linear relative integral invariant , whose light rays are then the characteristics, is
defined by the formula:
F F F F F F
= x+ y+ z x + y + z F t ,
x y z x y z
which becomes:
x y z
= n x+ y+ z ,
x 2 + y 2 + z 2 x 2 + y 2 + z 2 x 2 + y 2 + z 2

or furthermore:
= n ( x + y + z),

in which , , denote the directions cosines of an arbitrary direction. The form therefore
depends upon 5 variables, in reality. It is easy to form its characteristic equations and to show
that they contain, in particular, the equations:
204 Lessons on integral invariants

dx dy dz
= = .

The direction (, , ) is obviously nothing but the tangent to the light ray under consideration.

192. The property of that it is a relative integral invariant leads to the property of a
pencil of light rays that says that if one describes a closed curve (C) that encircles the pencil then
when the integral n cos s is taken over that curve it will be independent of the chosen curve,
in which we have let denote the angle between the tangent to (C) at a point M and the tangent
to the pencil of light that passes through M. One may easily prove that the necessary and
sufficient condition for the rays of a congruence to all be normal to the same surface is that this
integral must be zero for any pencil of rays that is taken from the congruence. This corresponds
to the theorem of Malus that the rays of a congruence that are normal to a surface will be normal
to an infinitude of surfaces. The condition for this to be the case is that the quadratic exterior
form must be zero, or, more precisely, that the alternating bilinear form (, ) must be zero,
in which one considers to be the symbol of differentiation with respect to one of the parameters
of the congruence and to be the symbol of differentiation with respect to the other parameter.
The light rays that propagate in the medium under consideration depend on the four
parameters u1, u2, u3, u4 . One calls a transformation that is performed on these parameters that
changes any congruence of rays that are normal to one surface into another congruence of rays
that are normal to another surface a Malus transformation. As we have seen, the form is
expressible in terms of the ui and their differentials. The most general Malus transformation is
obviously defined by the equation:
( u , du ) = k (u, du),

in which k is an unknown function. Taking the exterior derivative of both sides of this equation
immediately gives:
[dk ] = 0.

Since is of rank 4, this is possible only if dk = 0; i.e., if k is a constant. As a result, the

desired transformation can be obtained by expressing the idea that the linear form:

( u , du ) k (u, du)
must be an exact differential:

n( x , y , z )( dx + dy + dz ) = k n(x, y, z)( dx + dy + dz) + dV.

For example, define a light ray by the coordinates (x0, y0) of a point where it intersects the xy-
plane, and the direction cosines of its tangent at this point by 0, 0, 0. One will have:

n( x0 , y0 , z0 )( 0 dx0 + 0 dy0 ) kn(x0, y0, 0)(0 dx0 + 0 dy0) = dV.

 Integrl invariants and the calculus of variations 205

1st Case. There is no relation between x0 , y0 , z0 . In this case, V is a known function of x0,
y0, x0 , y0 , and one has:
V V
kn( x0 , y0 , z0 ) 0 = , kn( x0 , y0 , 0) 0 = ,
x0 y0
V V
n( x0 , y0 , 0) 0 = , n( x0 , y0 , 0) 0 = .
x0 y0

The first two equations give x0 and y0 ; the last two then give 0 and 0 .
2nd Case. There is one and only one relation between x0 , y0 , x0 , and y0 . Let:

F(x0 , y0 ; x0 , y0 ) = 0

be that relation. If one denotes an arbitrary function V of x0 , y0 , x0 , y0 and introduces an

auxiliary parameter then one will have:

V F V F
kn( x0 , y0 , 0) 0 = + , kn( x0 , y0 , 0) 0 = + ,
x0 x0 y0 y0

V F V F
n( x0 , y0 , 0) 0 = + , n( x0 , y0 , 0) 0 = + .
x0 x0 y0 y0

When the first two of these four equations are combined with the equation F = 0 they will
give x0 , y0 , and ; the last two then give 0 and 0 .

3rd Case. x0 and y0 are given functions of x0 and y0 :

x0 = f(x0, y0), y0 = g(x0, y0).

V is then a function of x0, y0, and one has:

f g V
n( x0 , y0 , 0) 0 + 0 = kn( x0 , y0 , 0) 0 + ,
x0 x0 x0

f g V
n( x0 , y0 , 0) 0 + 0 = kn( x0 , y0 , 0) 0 + ;
y0 y0 y0

these equations then determine 0 and 0 .

206 Lessons on integral invariants

193. The form is invariant, and we have seen above the characteristic property of
congruences of rays for which that form is identically zero. The invariant form 12 2 has
applications in optics. Its developed expression is:

= n [dn ( x + y + z)( x + y + z)]

n2 ([ y z] + [ z x] + [ x y]).

For example, take all of the light rays that traverse a given surface element d and whose
tangents at the points of intersection are parallel to the lines that are interior to an infinitely small
nappe of a cone d . The rays under consideration will depend on four parameters u1, u2, u3, u4,
where the first two, for example, define the position of the point of intersection of the element
d, and the last two define the orientation of the tangent of that point. Take the state each light
ray to be characterized by the corresponding point of intersection (x, y, z) and the direction
cosines (, , ) of the tangent to that point. The first three quantities x, y, z depend upon only u1
and u2 , so any cubic exterior form in x, y, z is zero. As a result, up to a sign, the invariant
1
2
2 reduces to:

1
2 2 = n2 ([ y z] + [ z x] + [ x y]).

However, if one denotes the direction cosines of the normal to the element d by , , then
one will have:
[y z] = d, [z x] = d, [x y] = d,
[ ] = d, [ ] = d, [ ] = d.

As a result:
1
2
2 = n2( + + ) d d = n2 cos d d,

in which denotes the angle between the normal to the surface and the (mean) direction of the
light rays that traverse the surface.
Having said this, if one considers an arbitrary set of light rays that depend upon four
parameters then one can take the point on each ray where that ray pierces a given surface (S).
All of the rays that pass through this same point form a solid cone, and the integral invariant
12 that relates to the given set may be given by the formula:
2

I = n2 cos d d,

in which d denotes the surface element of S, and d denotes the nappe of an elementary cone of
rays that starts from the same point of S and makes an angle with the normal to S.
For example, take the set of light rays that traverse a volume bounded by a closed surface (S),
and take each ray at the point where it leaves the volume. For that set, one will have:
 Integrl invariants and the calculus of variations 207

I = n2 d cos d.

However, if one takes the longitude and co-latitude on a sphere of radius 1 to be the
coordinates then the integral cos d will be equal to:

sin cos d d,

which is taken over the hemisphere 0 ; it is thus equal to . As a result, one has:
2

I= n2 d.

If the medium has index 1 then the rays will be rectilinear, and the integral I will be equal
to the product of the area of the surface with
194. As an application of the general methods of integration that were discussed in chapter
XVI, we propose to determine the path of light rays in an isotropic medium in which the index of
refraction n depends upon only one of the rectangular coordinates z. Here, one knows the
invariant , as well as three infinitesimal transformations that correspond to a translation
parallel to Ox, a translation parallel to Oy, and a rotation around Oz:

f f f f f f
A1 f = , A2 f = , A3 f = x y + .
x y y x

Since these three transformations leave the form:

= n ( x + y + z)

invariant, the linear invariant forms (, Ai) will reduce to:

(A1) = (n ),
(A2) = (n ),
(A3) = [n( x y)].

One thus has three first integrals:

n, n, n(x y).
Set:
n = a, n = b, x y = c.

The last relation shows that any light ray is in a plane parallel to Oz. As a result, one has
that:
208 Lessons on integral invariants

(
= ax + by + n 2 a 2 b 2 dz )
dz dz
x a a y b 2 b .
n2 a 2 b2 n a 2 b2
As a result, one has:

dz dz
x = a + a , y = b + b
n2 a2 b2 n2 a2 b2

for the trajectories of the light rays.

 CHAPTER XIX

FERMATS PRINCIPLE
AND THE INVARIANT PFAFF EQUATION OF OPTICS

I. - Fermats principle.

195. In the preceding chapter, we considered an integral invariant of the optics of isotropic
media, while supposing that the index of refraction was independent of time.
Now take an arbitrary medium, in which we suppose that the propagation of light waves is
defined by a Monge equation:

(1) F(x, y, z, t; dx, dy, dz, dt) = 0

that is homogeneous in dx, dy, dz, dt. This signifies that the wave emanating from a light signal
that is emitted at the instant t at the point (x, y, z) will have:

F(x, y, z, t; X x, Y y, Z z, dt) = 0

for its equation at the instant t + dt.

As one knows, relative to the point (x, y, z) and at the instant t the wave surface has:

F(x, y, z, t; X x, Y y, Z z, 1) = 0
for its equation.
In such a medium, a light ray is defined by taking x, y, z to be three functions of t that satisfy
equation (1) and, moreover, a supplementary condition that constitutes what one calls Fermats
principle. Among all of the curves that satisfy equation (1) or, as on says among all of the
integral curves of the Monge equation (1), the light ray that emanates from a given point (x1, y1,
z1) is the one that minimizes the time t1 t0 that is necessary for the light to go from the first
point to the second. In other words, light rays are extremals of the Mayer problem that is
defined by the Monge equation (1).

196. We briefly recall how Fermats principle leads to the formation of differential equations
that define the light rays. Imagine a light ray that starts from a point (x0, y0, z0) at the instant t0
and reaches the point (x1, y1, z1) at the instant t1. If one is given an arbitrary integral curve of
equation (1) that is infinitely close to the light ray then one can suppose that x, y, z, t are
expressed as functions of a parameter u for the light ray, as well as the integral curve, such that
the values 0 and 1 of that parameter correspond to the instants t0 and t1 for the light ray,
respectively. Let:
x + x, y + y, z + z, t + t

be functions of u that relate to the varied curve. Denote the derivatives of x, y, z, t with respect to
210 Lessons on integral invariants

u by x , y , z , t . If one writes equation (1) in the form:

(2) F(x, y, z, t; x , y , z , t ) = 0

and varies that equation then one will obtain:

F F F F F F F F
x+ y+ z + t + x + y + z + t = 0 .
x y z t x y z t

We multiply the left-hand side of the latter equation by du, with being a given function of
u, and integrate from 0 to 1. We have:

F F F F F dx F dy
1

x x + y y + z z +
0
t
t +
x du
+
y du
F dz F dt
+ + du = 0,
z du t du
or, after integrating by parts:

F F F F
1

x + y + z + t
x y z t 0
(3)
1 F d F F d F
+ x x + + t du = 0.
du x t du t
0

If the integral curve in a neighborhood of the light ray satisfies the initial and final conditions
that were imposed then one will have:

(x)0 = (y)0 = (z)0 = (x)1 = (y)1 = (z)1 = 0,

and, as a result:

F
1
F d F F d F
1 ( t ) + x + + t du = 0.
t 1
1
0 x du x t du t

One may specify the functions x, y, z arbitrarily, provided that they are annihilated at the
limits of the interval. We then determine the function by the condition that the coefficient of t
in the quantity under the sign must be zero. In order for (t)1 to be zero for any varied integral
curve, it is necessary and sufficient that the coefficients of x, y, z in the quantity under
the sign must also be zero.
In other words, if one introduces an auxiliary quantity then the light rays will be given
by equation (2), when combined with the equations:
 Fermats principle and the Pfaff equation of optics 211

F d F
x = 0,
du x

F d F
= 0,
y du y
(4)
F d F
z
= 0,
du z

F d F
t = 0.
du t

Moreover, besides equation (2), the elimination of gives the three equations:

F d F F d F F d F F d F

x du x y du y z du z t du t
( 4 ) = = = ,
F F F F
x z z t

to which, we agree to append:

dx dy dz dt
= x , = y , = z , = t. = t .

du du du du

One immediately sees that when equation (2) is differentiated with respect to u then along
dx dy dz dt
with equations ( 4 ), this will give , , , as four equations of the first degree, and
dt dt dt du
the values that one derives will not depend on u. As is natural, the parameter u appears only in
the final equations, which are of the form:

dx dy dz dt dx dy dz dt
= = = = = = = ,
x y z t X Y Z T

in which X, Y, Z, T are given functions of x, y, z, t, x , y , z , t that are homogeneous of second

degree in x , y , z , t and satisfy:

F F F F F F F F
x + y + z + t +X +Y +Z +T = 0.
x y z t x y z t

In reality, the differential equations of light rays are first-order ordinary differential
x y z
equations in x, y, z, t, , , , and these seven quantities are assumed to be coupled by
t t t
relation (2).
212 Lessons on integral invariants

II. - The invariant Pfaff equation of optics.

197. Now consider a family of light rays that depend upon a parameter , and take each of
the light rays in an interval of time (t0, t1) that depends on and correspond to a point of
departure (x0, y0, z0) that varies with and a point of arrival (x1, y1, z1) that likewise varies with
. If, for each light ray, one denotes the arbitrary auxiliary function that figures in equations (4)
by , and applies formula (3) to it then one will obtain:

F F F F
1 x1 + y1 + z1 + t1
x 1 y 1 z 1 t 1
F F F F
= 0 x0 + y0 + z0 + t0 .
x 0 y 0 z 0 t 0

From this, it results that when the differential equation of the light rays is considered to be a
x y z
first-order system of differential equations in x, y, z, t, , , that are coupled by (2), it will
t t t
admit the invariant Pfaff equation:

F F F F
x+ y+ z+ t = 0 .
x y z t

This Pfaff equation, which also depends upon only the relations between x , y , z , t , is a
function of six variables. Its characteristic system is a system of ordinary differential equations,
which, as a result, can only be identical to the equations of the light rays.
We thus arrive at the conclusion that the light rays are characteristics of the Pfaff equation:

F F F F
(5) x+ y+ z + t = 0.
x y z t

This is the invariant Pfaff equation of optics.

198. In practice, the Monge equation (1) is written in the form:

dx dy dz dx dy dz
x, y , z , t ; , , F ( x, y , z , t ; , , ,1) = 0 .
dt dt dt dt dt dt
By setting:

dx dy dz
= x , = y , = z ,
dt dt dt
 Fermats principle and the Pfaff equation of optics 213

it is easy to form the invariant Pfaff equation. Indeed, one has:

F F F F
x + y + z + t = 0.
x y z t

As a result, equation (5) can be written:

F F F F F F
t x + t y + t z x + y + z t = 0.
x y z x y z

Since the left-hand side is homogeneous in x , y , z , t , one can replace its arguments
by x , y , z , 1, respectively. One thus has the form:

(6) x+ y+ z x + y + z t = 0
x y z x y z

for the invariant Pfaff equation.

For example, take a medium in which the wave surface is a sphere, and let c / n be the
velocity of propagation of light, where c is the velocity in a vacuum and n is the index of
refraction (which is a function of x, y, z, t). Here, the Monge equation is:

dx 2 dy 2 dz 2 2
n + + c = 0,
2

dt dt dt

and the invariant Pfaff equation is:

n 2 ( x x + y y + z z ) c 2 t = 0 .
Upon setting:
nx ny nz
= , = , =
c c c
it becomes:
n( x + y + z) c t = 0.

, , are then the direction cosines of the tangent to the light rays.

If n does not depend on time then the law of propagation of light will admit the infinitesimal
f
transformation , and, as a result, the differential equations that give the light rays will admit
t
the invariant form:
n
t ( x + y + z ) .
c
214 Lessons on integral invariants

The differential equations that give the (geometric) curves that are described by the light rays
admit, in turn, the relative integral invariant:

n( x + y + z).

We thus recover the viewpoint of the preceding chapter (sec. 191).

199. As one knows (sec. 152), the characteristic equations of the Pfaff invariant of optics can
be converted into the characteristic equations of a first-order partial differential equation. (The
converse is also true, but we shall leave it at that.)
The existence of an integral invariant will be assured any time the law of propagation of light
admits an infinitesimal transformation. In all of these cases, one may convert the search for light
rays into a problem in the ordinary calculus of variations.
For example, take the case in which the law of propagation of light is given by the Monge
equation:
n2 (dx2 + dy2 + dz2) c2 dt2 = 0,

where the index of refraction can depend on x, y, t, but not on z. One then has the infinitesimal
transformation:
f
Af = .
z
The form:
( ) n( y + y + z ) c
= = z + x + y t
( A) n n

is an invariant form. Once the coordinates x and y are known as a function of t, one will get z by
a quadrature. As for the differential equations that give x and y as functions of t, they will admit
the relative integral invariant:
c
x + y n t ,
or, what amounts to the same thing, the integral invariant:

x + y t,

in which , , are three quantities that are coupled by the relation:

n2 2
1 + 2 + 2 = .
c2

In particular, the equations of the characteristics consist of the equations:

 Fermats principle and the Pfaff equation of optics 215

n2
dt 2 (dx 2 + dy 2 )
dx dy dt c 2
= = = .
n2 n

c2 c

The light rays render the integral:

c2 2
dx dy + dt = n2
dt dx 2 dy 2
stationary.

III. - Fermats principle is independent of the spacetime framing.

200. It is important to remark that the invariant Pfaff equation of optics is coupled to the
Monge equation that defines the law of propagation of light in a manner that is independent of
the framing chosen for space and time. In other words, the equation:

F F F F
x+ y+ z + t = 0
x y z t
is a covariant of the equation:
F(x, y, z, t; x , y , z , t ) = 0

relative to any change of variables performed on x, y, z, t. This result is basically Fermats

principle itself. However, one also recover this equation in the following manner, in which
nothing distinguishes one of the variables x, y, z, t from the others.
Consider the Pfaff system:
dx dy dz dt
(19) = = = ,
x y z t
x y z
such that x, y, z, t, , , are assumed to be coupled by relation (2), and look for the derived
t t t
system of (7). It is the system that is composed of Pfaff equations that are linear combinations of
equations (7), and which enjoy the property that the exterior derivative of their left-hand side is
zero on account of equations (7). If one sets:

x y z t
= = = ,
x y z 1

as above, then any linear combination of equations (7) will be of the form:

u (dx x dt ) + v (dy y dt ) + w (dz z dt ) = 0 .

216 Lessons on integral invariants

If one takes equations (7) into account then the exterior derivative of the left-hand side will
reduce to:
[dt (u dx + v dy + w dz)] .

On account of equations (7) and the derivative of equation (2), the condition for this to be
zero is:
[dt (u dx + v dy + w dz )(dx x dt)(dy y dt)(dz z dt) dF] = 0,

or, upon simplifying:

F F F
dx dy dz dt (u dx + v dy + w dz) dx + dy + dz = 0.
x y z
This gives:
u v w
= = .
F F F
x y z

The system derived from system (7) is therefore simply the Pfaff equation:

F F F
(dx x dt ) + (dy y dt ) + (dz z dt ) = 0,
x y z

whose characteristics are the light rays.

It results from this that, even in optics, the time coordinate does not play a role that is
essentially different from the one that is played by the spatial coordinates. The fundamental laws
of optics are not necessarily related to the classical notions of space and time, and they behave
just as they do in the theory of relativity.

201. For example, if one chooses a convenient framing for the universe (i.e., spacetime) then
the laws of propagation of light in a gravitational field that is produced by a unique mass (which
is reduced to a point) will be furnished by the Schwarzschild equation:

dr 2 2m 2
+ r 2 (d 2 + sin 2 d 2 ) 1 dt = 0 .
2m r
1
r

f
These laws admit the infinitesimal transformation . When the light rays are considered
t
from only the space point of view, they will thus be defined as realizing an extremum of the
integral:
 Fermats principle and the Pfaff equation of optics 217

dr 2 r 2 (d 2 + sin 2 d 2 )
2m
2
+
2m
.
1
1 r
r

Propagation takes place in a plane that passes through the center of attraction, and if one
supposes that this plane is defined by = 0 then one will have to realize an extremum for the
integral:
dr 2 r 2 d 2
2
+
2m 1 2m
.
1 r
r

f
By taking advantage of the existence of the infinitesimal transformation , the integration

offers no difficulty, and gives:
dr
C
= r2 .
2 2m
C 1
1 r
r2
 BIBLIOGRAPHY

H. POINCAR. - Les mthodes nouvelles de la Mcanique cleste, t. III, Paris, Gauthier-Villars,

1899.
P. APPELL. - Trait de Mcanique rationelle, t. II, chap. XXV, Paris, Gauthier-Villars.
H. POINCAR. Sur les rsidus des intgrales doubles, Acta Mathem., t. IX, 1887, pp. 321-
380.
- Analysus situs, Journal Ec. Polyt., 1895.
TH. DE DONDER. tude sur les invariants intgraux, Rend. Circ. Mat. Palermo, t. XV
(1901), pp 66-131; t. XVI (1902), pp 155-179.
- Sur les invariants intgraux, Atti del IV Congresso internazionale dei Matematici, t. II,
Roma, 1909, pp. 129-137.
- Sur le multiplicateur de Jacobi gnralis, Bulletin de lAcad. Royale de Belgique (classes
des sciences), 1908, pp. 129-137.
- Sur les invariants intgraux relatifs, ibid., 1909, pp. 66-83.
- Application du multiplicateur gnralis, ibid., 1909, pp. 610-621.
- Sur le multiplicateur gnralis, ibid., pp. 268-286.
- Sur les invariants intgraux relatifs et leurs applications la Physique mathmatique, ibid.,
1911, pp. 50-70.
- Quelques remarques sur le multiplicateur de Jacobi et le multiplicateur gnralis, ibid.,
1911, pp. 740-749.
- Introduction la thorie des invariants intgraux, ibid., 1913, pp. 1043-1073.
- Applications nouvelles des invariants intgraux, Mmoires de lAcad. Royale de Belgique
(classes des sciences) (2), I, (1904).
- Sur les quations canonique de Hamilton-Volterra, ibid., (2), III.
G. KOENIGS. Application des invariants intgraux la rduction au type canonique dun
systme quelconque dquations differentielles, C.R. Acad. des Sc. Paris, t. CXXI, (1895),
pp. 875-878.
- Sur les invariants intgraux, ibid., t. CXXII, (1896), pp. 25-27.
S. LIE. Ueber Integralinvarianten und Differentialgleichungen, Videnskabsselskabets
Skrifter, Christiania, 1902, no. 1 (73 pp.).
K. ZORAWSKI. Ueber gewisse Transformationseigenschaften der vielfachen Integrale,
Bull. Acad. des sciences de Cracovie (sc. math. et natur.), 1909, pp. 483-542.
R. HARGREAVES. Integral forms and their connexion with physical equations, Trans.
Cambridge Philosoph. Society; t. XXI (1912), pp. 107-122.
R. DONTOT. Sur les invariants intgraux de la propagation par ondes, Bull. Soc. Math. de
France, t. XLII (1914), pp. 53-91.
E. VESSIOT. Sur les invariants intgraux et quelques points doptique gomtrique, Bull.
Soc. Math. de France, t. XLII (1914), pp. 142-167.
- Sur un invariant intgral de lHydrodynamique et son application la relativit, C.R. Acad.
des Sc. De Paris, 30 Dec. 1918.
E. GOURSAT. Sur les invariants intgraux, Journal Math. pures et appliqes (5), t. IV,
(1908), pp. 331-365.
- Sur quelques points de la thorie des invariants intgraux, ibid., (7), t. I, (1915), pp. 241-
250.
 Biblography 219

- Sur certains systmes dequations aux diffrentiels totales et sur une gnralisation du
problme de Pfaff, Ann. Fac. Sc. Toulouse, t. VII, (1915).
E. CARTAN. Sur lintgration des systmes diffrentiels compltement intgrables, C.R.
Acad. des Sc. Paris, t.. CXXXV, (1902), pp. 1415-1417; 1564-1566.
T. CHELLA. Vantaggiche si possono trarre da noti invarianti e differenziali in alcuni
problemi dintegrazione, Annali R. Scuola norm. sup. Pisa (sc. fis.-mat.), t. XI, (1910), pp.
1-137.

On the symbolic calculus that is applied to exterior differential forms, and on certain related
symbolic calculi, one may consult, in addition to the preceding:

E. CARTAN. Sur certains expressions diffrentiels et le problme de Pfaff, Ann. Ec. Norm.,
(3), t. XVI, (1899), pp. 239-332.
A. BOREL. Sur les transformations et extensions de la formule de Stokes, Ann. Fac. Sc.
Toulouse, t. IV, (1912); t. VI, (1914), t. VII, (1915)

Finally, on the subject of integral invariants of continuous groups of transformations, where

the viewpoint is a little different from that of H. Poincar, one may consult:

S. Lie. Die Theorie der Integralinvarianten ist ein corollar der Theorie der
Differentialinvarianten, Ber. Schs. Gesellsch., Leipzig, 1897, pp. 342-357.
Ueber die Integralinvarianten und ihre Verwertung fr die Theorie der
Differentialgleichungen, ibid., 1897, pp. 369-410.
K. ZORAWSKI. Ueber Integralinvarianten der continuerlichen Transformationsgruppen,
Bull. Acad. Sc. Cracovie (sc. math. et nat.), 1895, pp. 127-130.
E. CARTAN. Le principe de dualit et certains intgrales multiples de lespace tangentiel et
de lespace rgl, Bull. Soc. Math. France, t. XXIV (1896), pp. 140-177.
TH. DE DONDER. Sur un problme relatif aux invariants intgraux, Bull. Acad. royale de
Belgique (classe des sc.), 1912, pp. 583-590.
R. DELTHEIL. Sur la thorie des probabilits gomtriques, Thesis, Toulouse, Private Ed.,
1920.