Introduction To The Geometry of Classical Dynamics

Introduction to the
Geometry of Classical Dynamics

Renato Grassini
Dipartimento di Matematica e Applicazioni
Universit`a di Napoli Federico II
H1//11 LT T
HIKARI LTD
Hikari Ltd is a publisher of international scientic journals and books.
www.m-hikari.com
Renato Grassini, Introduction to the Geometry of Classical Dynamics, First
published 2009.
No part of this publication may be reproduced, stored in a retrieval system,
or transmitted, in any form or by any means, without the prior permission of
the publisher Hikari Ltd.
ISBN 978-954-91999-4-9
Typeset using L
A
T
E
X.
Mathematics Subject Classication: 70H45, 58F05, 34A09
Keywords: smooth manifolds, implicit dierential equations, dAlemberts
principle, Lagrangian and Hamiltonian dynamics
Published by Hikari Ltd
iii
Preface
The aim of this paper is to lead (in most elementary terms) an under-
graduate student of Mathematics or Physics from the historical Newtonian-
dAlembertian dynamics up to the border with the modern (geometrical ) Lagrangian-
Hamiltonian dynamics, without making any use of the traditional (analytical )
formulation of the latter.
1
Our expository method will in principle adopt a rigorously coordinate-free
language, apt to gain from the very historical formulation the conscious-
ness (at an early stage) of the geometric structures that are intrinsic to the
very nature of classical dynamics. The coordinate formalism will be conned to
the ancillary role of providing simple proofs for some geometric results (which
would otherwise require more advanced geometry), as well as re-obtaining the
local analytical formulation of the theory from the global geometrical one.
2
The main conceptual tool of our approach will be the simple and general
notion of dierential equation in implicit form, which, treating an equation
just as a subset extracted from the tangent bundle of some manifold through a
geometric or algebraic property, will directly allow us to capture the structural
core underlying the evolution law of classical dynamics.
3
1
Such an Introduction will cover the big gap existing in the current literature between
the (empirical) elementary presentation of Newtonian-dAlembertian dynamics and the (ab-
stract) dierential-geometric formulation of Lagrangian-Hamiltonian dynamics. Standard
textbooks on the latter are [1][2][3][4], and typical research articles are [5][6][7][8][9][10].
2
The dierential-geometric techniques adopted in this paper will basically be limited to
smooth manifolds embedded in Euclidean ane spaces, and are listed in Appendix (whose
reading is meant to preceed that of the main text). More advanced geometry can be found
in the textbooks already quoted, as well as in a number of excellent introductions, e.g.
[11][12][13][14].
3
Research articles close to the spirit of this approach are, among others, [15][16] (on
implicit dierential equations) and [17][18][19][20] (on their role in advanced dynamics).
Contents
Preface iii
1 From Newton to dAlembert 1
1.1 The data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Conguration space . . . . . . . . . . . . . . . . . . . . . . . . . 1
Mass distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Force eld . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Mechanical system . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 The question . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Smooth motions . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Dynamically possible motions . . . . . . . . . . . . . . . . . . . . 3
1.3 The answer after Newton . . . . . . . . . . . . . . . . . . . . . . 4
Newtons law of constrained dynamics . . . . . . . . . . . . . . . . 4
1.4 dAlemberts reformulation . . . . . . . . . . . . . . . . . . . . . 4
dAlemberts principle of virtual works . . . . . . . . . . . . . . . . 4
1.5 dAlemberts implicit equation . . . . . . . . . . . . . . . . . . . 5
Tangent dynamically possible motions . . . . . . . . . . . . . . . . 5
dAlembert equation . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 From dAlembert to Lagrange 8
2.1 Integrable part of dAlembert equation . . . . . . . . . . . . . . 8
Restriction of dAlembert equation . . . . . . . . . . . . . . . . . . 9
Extraction of the integrable part . . . . . . . . . . . . . . . . . . . 9
2.2 Lagrange equation . . . . . . . . . . . . . . . . . . . . . . . . . 10
Covector formulation . . . . . . . . . . . . . . . . . . . . . . . . . 10
Riemannian geodesic curvature eld . . . . . . . . . . . . . . . . . 11
Normal form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Integral curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Classical Lagrange equations . . . . . . . . . . . . . . . . . . . . . 17
iv
Table of Contents v
2.3 Euler-Lagrange equation . . . . . . . . . . . . . . . . . . . . . . 23
Conservative system . . . . . . . . . . . . . . . . . . . . . . . . . 23
Lagrangian geodesic curvature eld . . . . . . . . . . . . . . . . . . 24
Integral curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Classical Euler-Lagrange equations . . . . . . . . . . . . . . . . . . 25
2.4 Variational principle of stationary action . . . . . . . . . . . . . 26
Variational calculus . . . . . . . . . . . . . . . . . . . . . . . . . 26
Variational principle . . . . . . . . . . . . . . . . . . . . . . . . . 29
Riemannian case . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3 From Lagrange to Hamilton 33
3.1 Legendre transformation . . . . . . . . . . . . . . . . . . . . . . 33
Lagrangian function and Legendre transformation . . . . . . . . . . 33
Coordinate formalism . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2 Hamiltonian function . . . . . . . . . . . . . . . . . . . . . . . . 35
Energy and Hamiltonian function . . . . . . . . . . . . . . . . . . 35
3.3 Hamiltonian vector eld . . . . . . . . . . . . . . . . . . . . . . 36
Canonical symplectic structure and Hamiltonian vector eld . . . . . 36
3.4 Hamilton equation . . . . . . . . . . . . . . . . . . . . . . . . . 37
Cotangent dynamically possible motions . . . . . . . . . . . . . . . 37
Hamilton equation . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Classical Hamilton equations . . . . . . . . . . . . . . . . . . . . . 38
4 Concluding remarks 39
4.1 Inertia and force . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Gauge transformations . . . . . . . . . . . . . . . . . . . . . . . 40
4.3 Geometrizing physical elds . . . . . . . . . . . . . . . . . . . . 41
4.4 Geometrical dynamics . . . . . . . . . . . . . . . . . . . . . . . 41
5 Appendix 42
5.1 Submanifolds of a Euclidean space . . . . . . . . . . . . . . . . 42
Ane subspaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Embedded submanifolds . . . . . . . . . . . . . . . . . . . . . . . . 43
Atlas of charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Locally Euclidean topology . . . . . . . . . . . . . . . . . . . . . . 44
Smoothness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Smooth curves and tangent vectors . . . . . . . . . . . . . . . . . . 44
Tangent vector spaces . . . . . . . . . . . . . . . . . . . . . . . . 45
Open submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 47
vi Table of Contents
Implicit function theorem . . . . . . . . . . . . . . . . . . . . . . 47
5.2 Tangent bundle and dierential equations . . . . . . . . . . . . . 49
Tangent bundle and canonical projection . . . . . . . . . . . . . . . 49
Vector bre bundle structure . . . . . . . . . . . . . . . . . . . . . 49
Tangent lift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Dierential equations in implicit form . . . . . . . . . . . . . . . . 50
Integral curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Integrable part . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Vector elds and normal form . . . . . . . . . . . . . . . . . . . . 51
Cauchy problems and determinism . . . . . . . . . . . . . . . . . . 51
5.3 Second tangent bundle and second-order dierential equations . 52
Second tangent lift and second tangent bundle . . . . . . . . . . . . 52
Ane bre bundle structure . . . . . . . . . . . . . . . . . . . . . 53
Second-order dierential equations in implicit form . . . . . . . . . 55
Integral curves and base integral curves . . . . . . . . . . . . . . . 56
Semi-sprays and normal form . . . . . . . . . . . . . . . . . . . . 57
Cauchy problems and determinism . . . . . . . . . . . . . . . . . . 58
5.4 Cotangent bundle and dierential forms . . . . . . . . . . . . . 58
Cotangent vector spaces . . . . . . . . . . . . . . . . . . . . . . . 58
Cotangent bundle and canonical projection . . . . . . . . . . . . . . 59
Vector bundle structure . . . . . . . . . . . . . . . . . . . . . . . . 59
Dierential 1-forms . . . . . . . . . . . . . . . . . . . . . . . . . 59
Semi-basic dierential 1-forms . . . . . . . . . . . . . . . . . . . . 61
Semi-Riemannian metric . . . . . . . . . . . . . . . . . . . . . . . 62
Almost-symplectic structure . . . . . . . . . . . . . . . . . . . . . 63
5.5 Intrinsic approach to smooth manifods . . . . . . . . . . . . . . 64
Intrinsic geometry of embedded submanifolds . . . . . . . . . . . . . 64
Intrinsic geometry of smooth manifolds . . . . . . . . . . . . . . . 66
References 67
Chapter 1
From Newton to dAlembert
In this chapter, we shall recall the problem of classical particle dynamics,
Newtons answer to the problem and dAlemberts reformulation of the answer.
The latter will then be shown to correspond to a dierential equation in implicit
form on a Euclidean space.
1.1 The data
Classical particle dynamics deals with an empirical problem, whose data in
the simplest cases can be described in mathematical terms as follows.
Conguration space
A reference space mathematical extension of a rigid body carrying an ob-
server is conceived as a 3-dimensional, Euclidean, ane space c
3
, modelled
on a vector space E
3
with inner product (time will be conceived as an ori-
ented, 1-dimensional, Euclidean, ane space, classically identied with the
real line R).
Particle is synonymous with point-like body, i.e. a body whose position
in c
3
is dened by a single point of c
3
.
Therefore, for a given (ordered) system of particles, a position or con-
guration in c
3
is dened by a single point of c := c
3
(3-dimensional,
Euclidean, ane space, modelled on E := E
3
).
1
1
Recall that the inner product in E is dened by putting, for all v = (v
1
, . . . , v
) and
w = (w
1
, . . . , w
) belonging to E ,
v w :=
i=1
v
i
w
i
1
2 1 From Newton to dAlembert
The particle system may generally be subject in c
3
to some holonomic (or
positional ) constraints, owing to which it is virtually allowed to occupy only
the positions belonging to an embedded submanifold
Q c
called conguration space of the system in c
3
.
Q generally consists of all the points p c satisfying a number < 3
of independent scalar equalities f
(p) = 0 ( = 1, . . . , ), called two-sided

constraints, and/or some strict scalar inequalities g
(p) > 0 ( = 1, . . . , ),
called strict one-sided constraints. Under usual hypotheses of regularity on
the constraints, Q is an embedded submanifold of c , whose dimension n :=
dimQ = 3 called number of the degrees of freedom of the system is
given by the dimension of the Euclidean environment minus the number of
the two-sided constraints (in absence of two-sided constraints, Q is an open
submanifold).
2
Mass distribution
The response of the system to any internal or external inuence, will generally
depend on how massive its particles are, the inertial mass of a particle being
conceived as a positive scalar quantity.
The inertial mass distribution carried by the system will be denoted by
m := (m
1
, . . . , m
)
Force eld
The force resultant of all the internal and/or external inuences
acting in c
3
on the particles and not depending on their being constrained
or unconstrained, is generally described as a smooth vector-valued mapping,
called force eld,
F : U E Tc E : (p, v) F(p, v) = (F
1
(p, v), . . . , F
(p, v))
where U denotes an open subset of c containing Q.
Remark that, if F[
{p}E
= const. for all p U , then F called positional
force eld can be regarded as a smooth mapping f : U c E : p f(p) ,
where f(p) denotes the constant value of F[
{p}E
.
3
2
We shall not consider non-strict one-sided constraints, which would give rise to a mani-
fold Q with boundary (nor shall we consider time-dependent constraints, which would give
rise to a manifold Q bred over the real line).
3
We shall not consider time-dependent force elds, which would later give rise to time-
dependent dierential equations.
Mechanical system
The above empirical mechanical system will briey be denoted by the triplet
o := (Q, m, F)
1.2 The question
With reference to such a mechanical system o , the basic problem of dynamics
can be expressed in the following terms.
Smooth motions
A smooth motion of the particle system in the reference space c
3
is described
as a smooth curve of c , say
: I R c : t (t) = p(t)
Such a curve is meant to establish a conguration p(t) at each instant t
of a time interval I R, along which the positions p(t) = (p
1
(t), . . . , p
(t))
of the particles, their velocities p(t) = ( p
1
(t), . . . , p
(t)) and their accelera-

tions p(t) = ( p
1
(t), . . . , p
(t)) (as well as the derivatives of any order) vary

continuously.
Dynamically possible motions
Smooth dynamics basically deals with the time-evolution problem in the
unknown expressed by the following question:
For the above constrained point-mass system, what are the smooth mo-
tions in the chosen reference space that are possible under the action of the
given ?
Such motions will briey be called the dynamically possible motions (DPMs)
of o (whereas the smooth motions which would be possible in absence of force,
i.e. F = 0, will be said to be the inertial motions of o ).
1.3 The answer after Newton
After Newton, the answer to the above predictive question is given by the
following law.
Newtons law of constrained dynamics
A smooth curve
: I c : t (t) = p(t)
is a DPM of o , i, for all t I ,
4
p(t) Q
m p(t) = F(p(t), p(t)) + (t)
(t) T
p(t)
Q
The rst condition just exhibits the kinematical eects of the constraints,
which only allow motions living in Q.
The second condition is the classical Newtons law with a right hand side
encompassing the possible dynamical eects of the contraints, expressed by
an unknown constraint reaction (t) E .
The third condition expresses the only known empirical requisite of the
constraint reaction, which apart from possible frictions tangent to Q is
always orthogonal to Q.
5
1.4 dAlemberts reformulation
After dAlembert, the unknown constraint reaction can be cancelled from
Newtons law of constrained dynamics as follows.
dAlemberts principle of virtual works
The last two of the above conditions can obviously be expressed in the form
F(p(t), p(t)) m p(t) T
p(t)
Q
4
For any = (
1
, . . . ,
) R
and w = (w
1
, . . . , w
) E , we shall put w :=
(
1
w
1
, . . . ,
w
n
) . Moreover, for any p Q, T
p
Q will denote the orthogonal complement
in E of the tangent vector space T
p
Q.
5
If the empirical law of friction is to be taken into consideration, then you will embody
it in F.
that is,
_
F(p(t), p(t)) m p(t)
_
p = 0 , p T
p(t)
Q
called dAlemberts principle of virtual works (since the inner product therein
denes the work of active force F(p(t), p(t)) and inertial force m p(t) along
any virtual displacement p , i.e. any innitesimal displacement tangent to
Q and therefore virtually allowed by the constraints).
6
So a smooth curve
: I c : t (t) = p(t)
is a DPM of o , i it satises, for all t I the time-evolution law
p(t) Q, m p(t) p = F(p(t), p(t)) p , p T
p(t)
Q ()
1.5 dAlemberts implicit equation

From the mathematical point of view, condition ()
shows that determining

the DPMs of o is a second-order dierential problem, whose unknown is a
smooth curve of c . It turns into a rst-order dierential problem, whose
unknown is a smooth curve of Tc , as follows.
Tangent dynamically possible motions
If
: I c : t (t) = p(t)
is DPM of o , its tangent lift
: I Tc : t (t) = (p(t), p(t))
will be called a tangent dynamically possible motion (TDPM) of o .
6
Owing to footnotes
1
and
4
, dAlemberts principle reads
i=1
_
F
i
(p(t), p(t)) m
i
p
i
(t)
_
p
i
= 0 , p = (p
1
, . . . , p
) T
p(t)
Q
Remark that, if Q is an open submanifold of c (absence of two-sided constraints), one
has T
p
Q = E
= 0 for all p Q (absence of constraint reaction) and then dAlemberts

principle simply reads
m
i
p
i
(t) = F
i
(p(t), p(t)) , i = 1, . . . ,
DPMs and TDPMs bijectively correspond to one another, since the tangent
lift operator is obviously inverted by the base projection operator
. Trough such a bijection, the problem of determining the DPMs
proves to be naturally equivalent to that of determining the TDPMs.
Owing to ()
, a smooth curve
c : I Tc : t c(t) = (p(t), v(t))
is a TDPM of o , i it satises, for all t I , the time-evolution law
p(t) Q, p(t) = v(t) , m v(t) p = F(p(t), v(t)) p , p T
p(t)
Q ()
dAlembert equation
Condition () will now be seen to correspond to a rst-order dierential equa-
tion in implicit form on Tc (second-order on c ),
7
namely dAlembert equa-
tion
T
d
Al
:= (p, v; u, w) TTc [ p Q, u = v , F(p, v) mw T
p
Q
= (p, v; u, w) TTc [ p Q, u = v , mw p = F(p, v) p , p T
p
Q
T
2
c
Proposition 1 The TDPMs of o are the integral curves of T
d
Al
(and then
the DPMs are its base integral curves).
Proof Recall that a smooth curve
c : I Tc : t c(t) = (p(t), v(t))
is an integral curve of T
d
Al
, i its tangent lift
c : I TTc : t c(t) = (p(t), v(t); p(t), v(t))
satises condition
Im c T
d
Al
that is, for all t I ,
c(t) = (p(t), v(t); p(t), v(t)) T
d
Al
which is exactly condition (), characterizing the TDPMs.
7
Recall that Tc = c E is a Euclidean ane space modelled on E E. Its tangent
bundle is therefore TTc = (c E) (E E) .
Also recall that a smooth curve
: I Tc : t (t) = p(t)
is a base integral curve of T
d
Al
, i its second tangent lift
: I TTc : t (t) = (p(t), p(t); p(t), p(t))
satises condition
Im T
d
Al
that is, for all t I ,
(t) = (p(t), p(t); p(t), p(t)) T
d
Al
which is exactly condition ()
, characterizing the DPMs.
Chapter 2
From dAlembert to Lagrange
Dynamics is now a problem of integration, i.e. determination and/or qualita-
tive analysis of the integral curves of dAlembert equation (implicit dierential
equation on Euclidean space Tc ). In this connection, the latter will be shown
to be equivalent to a Lagrange equation (implicit dierential equation on
manifold TQ), which will naturally be obtained and thoroughly discussed.
2.1 Integrable part of dAlembert equation
As to the integration of T
d
Al
, the rst step is to extract its integrable part,
i.e. the region
T
(i)
d
Al
T
d
Al
swept by the tangent lifts of all its integral curves (i.e. covered by the orbits
of such lifts). As to the extraction of T
(i)
d
Al
, it is quite natural to start from
the following remark.
Owing to Prop.1, the base integral curves (if any) of T
d
Al
are constrained
to live in Q (see condition ()
) and then their tangent lifts, i.e. the integral

curves, live in TQ. As a consequence, the tangent lifts of the integral curves
live in TTQ, that is to say,
T
(i)
d
Al
TTQ
So we obtain
T
(i)
d
Al
T
d
Al
TTQ
8
Restriction of dAlembert equation
The above result suggests focusing on the restriction of T
d
Al
obtained via
intersection with TTQ, i.e. on the rst-order dierential equation in implicit
form on TQ (second-order on Q)
T
Lagr
:= T
d
Al
TTQ
which will be called Lagrange equation.
Note that Lagrange equation is an eective restriction of dAlembert equa-
tion (i.e. T
Lagr
T
d
Al
) in presence, and only in presence, of two-sided
constraints (when dimQ < dimc ), as is shown by the following proposition.
Proposition 2 T
Lagr
T
d
Al
, i dimQ < dimc .
Proof Let dimQ < dimc (whence T
p
Q E for all p Q). We shall prove
that, under the above hypothesis, T
Lagr
T
d
Al
. Indeed, we can choose
p Q, v E T
p
Q (whence (p, v) , TQ), u = v and w =
1
m
(F(p, v) +
) with T
p
Q (whence F(p, v) mw T
p
Q) and clearly we obtain
(p, v; u, w) T
d
Al
, (p, v; u, w) , TTQ and then (p, v; u, w) , T
Lagr
.
Conversely, let dimQ = dimc (whence T
p
Q = T
2
(p,v)
Q = E for all (p, v)
TQ). We shall prove that, under the above hypothesis, T
d
Al
T
Lagr
.
Indeed, for any (p, v; u, w) T
d
Al
, we have p Q, v E = T
p
Q,
u = v and w E = T
2
(p,v)
Q, that is, (p, v; u, w) T
2
Q TTQ and
then (p, v; u, w) T
Lagr
.
Extraction of the integrable part

By extracting T
Lagr
from T
d
Al
, via intersection of the latter with TTQ, we
just obtain T
(i)
d
Al
,= , as is shown by the following proposition.
Proposition 3 T
(i)
d
Al
= T
Lagr
,=
Proof First we remark that T
d
Al
and T
Lagr
are equivalent equations i.e.
they have the same integral curves since, owing to inclusions T
(i)
d
Al
T
Lagr

T
d
Al
, condition Im c T
d
Al
(i.e. Im c T
(i)
d
Al
) implies Im c T
Lagr
and,
conversely, Im c T
Lagr
implies Im c T
d
Al
.
That amounts to saying T
(i)
d
Al
= T
(i)
Lagr
.
Then we anticipate that T
Lagr
is integrable, i.e. , = T
Lagr
= T
(i)
Lagr
.
Hence our claim.
10 2 From dAlembert to Lagrange

So the focal point is now to prove the integrability of T
Lagr
. That will
follow from the stronger property of T
Lagr
being reducible to normal form, as
will be shown in the sequel.
2.2 Lagrange equation
In order to prove the reducibility of Lagrange equation to normal form, we
shall need to give a deeper insight into its algebraic formulation and the un-
derlying geometric structures. Its integral curves will then be given a global
characterization in terms of the above geometric structures (the traditional
local characterization in coordinate formalism will nally be deduced).
Covector formulation
From the set-theoretical point of view, Lagrange equation can be expressed in
the form
T
Lagr
= (p, v; u, w) TTQ [ u = v , mwp = F(p, v)p , p T
p
Q T
2
Q
From the algebraic point of view, the condition on virtual works charac-
terizing T
Lagr
can be given the form of a covector equality, as will now be
shown.
Associated with the inertial mass distribution m, there is a semi-basic
1-form
[m] : T
2
Q T
Q : (p, v; v, w) (p, [m](p, v, w))

on T
2
Q, called covector inertial eld, whose value
[m](p, v, w) := (mw) [
TpQ
T
p
Q
at any (p, v; v, w) T
2
Q is the virtual work of mw.
1
Associated with the force eld F, there is a semi-basic 1-form
F : TQ T
Q : (p, v) (p, F(p, v))

on TQ, called covector force eld, whose value
F(p, v) := F(p, v) [
TpQ
T
p
Q
at any (p, v) TQ is the virtual work of F(p, v) .
2
1
For any u E , we dene u E
by putting u : v E u v R. Then, for any

p Q, the restriction of u to T
p
Q yields u [
TpQ
T
p
Q.
2
The virtual work of a positional force eld f can be regarded as an ordinary 1-form on
Q, namely f : p Q (p, f(p)) T
Q , f(p) := f (p) [
TpQ
T
p
Q.
Proposition 4 Lagrange equation can be given the covector formulation
T
Lagr
= (p, v; u, w) TTQ [ u = v , [m](p, v, w) = F(p, v)
Proof Just notice that condition
mw p = F(p, v) p , p T
p
Q
means
(mw) [
TpQ
= F(p, v) [
TpQ
that is,
[m](p, v, w) = F(p, v)
Riemannian geodesic curvature eld

The reducibility of T
Lagr
to normal form requires that, for any choice of the
data (p, v) TQ, the algebraic equation [m](p, v, w) = F(p, v) should be
uniquely solvable with respect to the unknown w T
2
(p,v)
Q. As the latter
only appears in the left hand side of the equation, the above property is to be
checked through a thorough investigation of [m] .
The following geometric considerations showing that such a semi-basic
1-form on T
2
Q is the transformed of a suitable semi-basic 1-form on TQ
through a distinguished semi-spray will prove to be crucial.
Remark that [m] is the semi-basic 1-form induced on T
2
Q by the Eu-
clidean metric
g
m
: E E
: u g
m
(u) := (mu)
(positive denite, symmetric, linear map), since its value at any (p, v; v, w)
T
2
Q is
[m](p, v, w) = g
m
(w)
TpQ
T
p
Q
In the same way, there is a semi-basic 1-form
g : TQ T
Q : (p, v) (p, g
p
(v))
induced on TQ by g
m
, whose value at any (p, v) TQ is
g
p
(v) := g
m
(v)
TpQ
T
p
Q
g is a Riemannian metric on Q, characterized by the quadratic form or
Lagrangian function
K : TQ R : (p, v) K(p, v) :=
1
2
g
p
(v) [ v) =
1
2
mv v
(the kinetic energy of the mechanical system).
Riemannian manifold (Q, K) carries a distinguished semi-spray
K
: TQ T
2
Q : (p, v) (p, v; v,
K
(p, v))
called Riemannian spray, uniquely determined by the following property.
Proposition 5 There exists one, and only one, semi-spray
K
on TQ s.t.,
for all (p, v) TQ,
g
m
(
K
(p, v))
TpQ
= 0
Proof (i) Unicity First remark that, for any (p, v) TQ, the map
w T
2
(p,v)
Q
()
g
m
(w)
TpQ
T
p
Q
is injective, since
g
m
(w
1
)
TpQ
= g
m
(w
2
)
TpQ
with w
1
, w
2
T
2
(p,v)
Q and then w
1
w
2
T
p
Q implies
g
p
(w
1
w
2
) = g
m
(w
1
w
2
)
TpQ
= g
m
(w
1
)
TpQ
g
m
(w
2
)
TpQ
= 0
that is, recalling that g
p
: T
p
Q T
p
Q is a linear isomorphism,
w
1
= w
2
So, if there exists a vector in T
2
(p,v)
Q whose image through () is zero, it is
unique.
(ii) Existence For any (p, v) TQ, put
K
(p, v) := w+u T
2
(p,v)
Q
with
w T
2
(p,v)
Q , u := g
1
p
_
g
m
(w)
TpQ
_
T
p
Q
The image of
K
(p, v) through () is zero, since
g
m
(
K
(p, v))
TpQ
= g
m
(w+u)
TpQ
= g
m
(w)
TpQ
+g
m
(u)
TpQ
= g
m
(w)
TpQ
+g
p
(u)
= g
m
(w)
TpQ
g
m
(w)
TpQ
= 0
K
transforms g (semi-basic 1-form on TQ) into
[K] : T
2
Q T
Q : (p, v; v, w) (p, [K](p, v, w))

(semi-basic 1-form on T
2
Q) by putting, for any (p, v; v, w) T
2
Q,
[K](p, v, w) := g
p
(w
K
(p, v)) T
p
Q
[K] will be called Riemannian geodesic curvature eld.
Actually [K] does not dier from [m] , as is shown in the following propo-
sition.
Proposition 6 Lagrange equation, in covector formulation, reads
T
Lagr
= (p, v; u, w) TTQ [ u = v , [K](p, v, w) = F(p, v)
Proof Owing to Prop. 4, it will suce to show that
[K] = [m]
To this end, note that, for any (p, v; v, w) T
2
Q, from Prop. 5 we obtain
[K](p, v, w) := g
p
(w
K
(p, v))
= g
m
(w
K
(p, v))
TpQ
= g
m
(w)
TpQ
g
m
(
K
(p, v))
TpQ
= g
m
(w)
TpQ
= [m](p, v, w)

Normal form
The reducibility of Lagrange equation to normal form immediately follows from
Prop. 6.
Consider the vertical force eld
F
:= g
1
F : TQ TQ : (p, v) (p,
F
(p, v))
F
(p, v) := g
1
p
(F(p, v)) T
p
Q
and the semi-spray
:=
K
+
F
: TQ T
2
Q : (p, v) (p, v; v, (p, v))
(p, v) :=
K
(p, v) +
F
(p, v) T
2
(p,v)
Q
Proposition 7 Lagrange equation can be put in the normal form
T
Lagr
= Im = (p, v; u, w) TTQ [ u = v , w = (p, v)
Proof Just notice that covector equality
[K](p, v, w) = F(p, v)
reads
g
p
(w
K
(p, v)) = F(p, v)
w
K
(p, v) = g
1
p
(F(p, v))
w =
K
(p, v)) +
F
(p, v)
w = (p, v)
Integral curves
The condition characterizing the integral curves of Lagrange equation can now
be formulated as follows.
Let
c : I TQ : t c(t) = (p(t), v(t))
be a smooth curve of TQ and
Q
c : I Q : t (
Q
c)(t) = p(t)
its projection onto Q.
3
3
The tangent lift of
Q
c will be denoted by (
Q
c)
.
Proposition 8 c is an integral curve of T
Lagr
, i
(
Q
c)
= c , [K] c = F c ()
or, in normal form,
c = c ()
Proof As is known, c is an integral curve of T
Lagr
, i
Im c T
Lagr
that is to say, for all t I ,
() c(t) = (p(t), v(t); p(t), v(t)) T
Lagr
(i) Owing to Prop. 6, condition () reads
p(t) = v(t) , [K](p(t), v(t), v(t)) = F(p(t), v(t))
that is,
(
Q
c)
(t) = (p(t), p(t))

= (p(t), v(t))
= c(t)
and
([K] c)(t) = ( p(t), [K](p(t), v(t), v(t)) )
= ( p(t), F(p(t), v(t)) )
= (F c)(t)
That proves our rst claim.
(ii) Owing to Prop. 7, condition () also reads
p(t) = v(t) , v(t) = (p(t), v(t))
that is,
c(t) = (p(t), v(t); p(t), v(t))
= (p(t), v(t); v(t), (p(t), v(t)))
= ( c)(t)
That proves our second claim.

As a consequence, the condition characterizing the base integral curves of
Lagrange equation will be formulated as follows.
Let
: I Q : t (t) = p(t)
be a smooth curve of Q and
[K] : I T
Q
its Riemannian geodesic curvature, whose vector (rather than covector) ex-
pression is the covariant derivative
4

dt
:= g
1
[K] : I TQ
Proposition 9 is a base integral curve of T
Lagr
, i
[K] = F ()
or, in vector formulation,

dt
=
F

or, in normal form,
= ()
Proof Recall that a base integral curve of Lagrange equation is the projection
=
Q
c of an integral curve c (smooth curve of TQ satisfying condition
() or the equivalent ()).
(i) Now, if is a base integral curve, condition () implies = (
Q
c)
= c
whence = c and then ()
. Conversely, if satises condition ()
, then it
is obviously a base integral curve (projection of c := satisfying ()). Clearly,
condition ()
is equivalent to
g
1
[K] = g
1
F
that is,

dt
=
F

which is the above mentioned vector formulation of ()
.
4
Covariant derivative is also related to an important geometric structure, called Levi-
Civita connection of Riemannian manifold (Q, K) .
(ii) In the same way, through condition (), one can show that the base integral
curves are characterized by ()
.
Alternatively, from

dt
: t I

(p(t), p(t); p(t), p(t)) T
2
Q
[K]
_
p(t), g
p(t)
( p(t)
K
(p(t), p(t)))
_
T
Q
g
1
_
p(t), p(t)
K
(p(t), p(t))
_
TQ
and

K
: t I (p(t), p(t)
K
(p(t). p(t))) TQ
that is,

dt
=
K

we deduce that the vector formulation of ()
(which has already been seen to

characterize the base integral curves) is equivalent to
=
K
+
F

which is condition ()
From the dynamical point of view, some remarks are now in order.
For F = 0 , the base integral curves characterized by a vanishing Rie-
mannian geodesic curvature [K] = 0 coincide with the inertial motions
of o (i.e. the motions which would be possible if F were zero).
The eect of a covector force eld F ,= 0 is then that of deviating the
DPMs of o from the inertial trend, by giving them a non-vanishing Rieman-
nian geodesic curvature, namely [K] = F .
Classical Lagrange equations
The scalar equations obtained with the aid of a chart by orderly equalling the
components of the covector or vector-valued functions which appear in the left
and right hand sides of ()
or ()
, are the classical Lagrange equations of

Analytical Dynamics.
They will prove to be only locally equivalent to the geometric Lagrange
equation, in the sense that they only characterize the base integral curves of
the latter which live in the coordinate domain of the given chart.
Preliminaries
Consider the coordinate domain | = Im of a chart : q W (q) | ,
expressing the points p | Q in function of coordinates q W R
n
(with n := dimQ).
5
To any smooth curve : t I (t) = p(t) Q living in | , i.e.
satisfying
p = p(t) |
for all t I , there corresponds in a smooth coordinate expression
q = q(t) W
related to by p(t) = (q(t)) , also denoted
p = (q) (1)
(dependence from time t is understood).
To the rst tangent lift c = , that is,
p = p(t) | , v = v(t) T
p(t)
Q
with
v = p
there corresponds in a smooth coordinate expression
q = q(t) W , v = v(t) R
n
related to by (1) and the time derivative of (1)
6
v = v(q, v) := v
h
p
q
h
q
(2)
where
v = q
is the n-tuple of linear components of v in . Remark that, for all h =
1, . . . , n,
v
v
h
(q,v)
=
p
q
h
q
,
v
q
h
(q,v)
= v
k

2
p
q
h
q
k
q
=
d
dt
p
q
h
q
(3)
5
Recall that, for any q =
1
(p) W , the partial derivatives
_
p
q
1
q
, . . . ,
p
q
n
q
_
provide a basis of T
p
Q.
6
A repeated index, in upper and lower position, denotes summation over (1, . . . , n).
To the second tangent lift c = , that is,
p = p(t) | , v = v(t) T
p(t)
Q , w = w(t) T
2
(p(t),v(t))
Q
with
v = p , w = v = p
there corresponds in a smooth coordinate expression
q = q(t) W , v = v(t) R
n
, w = w(t) R
n
related to by (1), (2) and the time derivative of (2)
w = w(q, v, w) := w
h
p
q
h
q
+v
h
v
k

2
p
q
h
q
k
q
where
w = v = q
is the n-tuple of ane components of w in .
In the sequel, the components of F in will be denoted by
F
h
(q, v) := (F )
h
= (F(p, v))
h
=
_
F(p, v) [
p
q
h
q
_
= F(p, v)
p
q
h
q
and the components of
F
= g
1
F will be denoted by
F
i
(q, v) := (
F
)
i
= (
F
(p, v))
i
= (g
1
p
(F(p, v)))
i
= g
ih
(q) (F(p, v))
h
= g
ih
(q) F
h
(q, v)
where [g
hk
(q)] is the inverse of the nonsingular matrix [g
hk
(q)] dened by
g
hk
(q) :=
_
g
p
_
p
q
h
q
_
[
p
q
k
q
_
= m
p
q
h
p
q
k
q
Lagrange equations
The above coordinate formalism will now be adopted for the characterization
of the base integral curves of T
Lagr
living in the given coordinate domain.
Proposition 10 A smooth curve : t I p = (q(t)) | Q living
in the coordinate domain | of a chart of Q is a base integral curve of
T
Lagr
, i its coordinate expression q = q(t) satises (for all h, i = 1, . . . , n)
the classical Lagrange equations
q = v ,
d
dt
K
v
h
(q,v)
K
q
h
(q,v)
= F
h
(q, v) ()
h
or, in normal form,
7
q = v , v
i
=
_
i
jk
_
q
v
j
v
k
+F
i
(q, v) ()
i
Proof (i) Recall that is a base integral curve of T
Lagr
, i it satises equation
()
. As lives in the coordinate domain of , equation ()
is equivalent to
the n scalar equations obtained by orderly equalling the components in of
its left and right hand sides, i.e.
([K] )
h
= (F )
h
()
h
The components F
h
(q, v) := (F )
h
have already been shown in the above
preliminaries, where we have put v = q . The components ([K] )
h
will now
be evaluated. To this end, by making use of (3) and usual rules of derivation,
we obtain
([K] )
h
:= ([K](p, v, w))
h
=
_
[K](p, v, w) [
p
q
h
q
_
= mw(q, v, w)
p
q
h
q
=
d
dt
_
mv(q, v)
p
q
h
q
_
mv(q, v)
d
dt
p
q
h
q
=
d
dt
_
mv(q, v)
v
v
h
(q,v)
_
mv(q, v)
v
q
h
(q,v)
7
Here we encounter the Christhoel symbols (of the Levi-Civita connection) associated
with a Riemannian manifold, dened on the coordinate domain of any chart by
_
i
jk
_
:=
1
2
g
ih
(
j
g
kh
+
k
g
hj

h
g
jk
)
with
i
g
hk
:=
g
hk
q
i
=
d
dt
v
h
(q,v)
_
1
2
mv v
_

q
h
(q,v)
_
1
2
mv v
_
=
d
dt
K
v
h
(q,v)
K
q
h
(q,v)
So equations ()
h
just take the form ()
h
.
(ii) Also recall that is a base integral curve of T
Lagr
, i it satises equation
()
is equivalent to
the n scalar equations obtained by orderly equalling the ane components in
of its left and right hand sides, i.e.

i
= ( )
i
()
i
where

i
= w
i
= v
i
and
( )
i
= (
K
+
F
)
i
= (
K
)
i
+ (
F
)
i
=
i
K
(q, v) +F
i
(q, v)
The components F
i
(q, v) := (
F
)
i
have already been shown in the above
preliminaries. The components
i
K
(q, v) := (
K
)
i
will now be evaluated.
To that purpose we need the coordinate expression of K , which is given by
K(p, v) =
1
2
_
g
p
_
v
h
p
q
h
q
_
[ v
k
p
q
k
q
_
=
1
2
g
hk
(q) v
h
v
k
whence
K
q
h
(q,v)
=
1
2
(
h
g
jk
)
q
v
j
v
k
K
v
h
(q,v)
= g
hk
(q)v
k
and
2
K
v
h
q
k
(q,v)
= (
k
g
hj
)
q
v
j
2
K
v
h
v
k
(q,v)
= g
hk
(q)
As a consequence, the above components ([K] )
h
can more explicitly be
expressed in the form
([K] )
h
= ([K](p, v, w))
h
=

2
K
v
h
v
k
(q,v)
w
k
+

2
K
v
h
q
k
(q,v)
v
k
K
q
h
(q,v)
= g
hk
(q)w
k
+ (
k
g
hj
)
q
v
j
v
k
1
2
(
h
g
jk
)
q
v
j
v
k
= g
hk
(q)w
k
+
1
2
(
k
g
hj
)
q
v
j
v
k
+
1
2
(
j
g
hk
)
q
v
j
v
k
1
2
(
h
g
jk
)
q
v
j
v
k
= g
hk
(q)w
k
+
1
2
(
k
g
hj
+
j
g
kh

h
g
jk
)
q
v
j
v
k
whence
_

dt
_
i
= (g
1
([K] ))
i
= (g
1
p
([K](p, v, w)))
i
= g
ih
(q) ([K](p, v, w))
h
= w
i
+
_
i
jk
_
q
v
j
v
k
Therefore, identities
8
_

dt
_
i
= (
K
)
i
=
i
(
K
)
i
read
w
i
+
_
i
jk
_
q
v
j
v
k
= w
i
i
K
(q, v)
Hence we obtain
i
K
(q, v) =
_
i
jk
_
q
v
j
v
k
So equations ()
i
i
.
8
See the proof, part (ii), of Prop. 9.
2.3 Euler-Lagrange equation
A special mention, for its primary role in both mathematical and theoretical
physics, is to be given to the dynamics of a conservative system.
Conservative system
Such a name refers to a system o = (Q, m, f) carrying a conservative eld f ,
i.e. a positional force eld whose virtual work
f = dV
is an exact 1-form, deriving from a smooth potential energy
V : Q R
(determined up to a locally constant function).
The name of conservative is due to the following conservation law of
mechanical energy
E := K +V : TQ R : (p, v) E(p, v) := K(p, v) +V (p)
(kinetic energy plus potential energy).
Proposition 11 Along each integral curve c of
T
Lagr
= (p, v; u, w) TTQ [ u = v , [K](p, v, w) = d
p
V
the mechanical energy E keeps constant, i.e.
E c = const.
Proof If c = (p, v) denotes an integral curve and w := v , from p = v and
[K](p, v, w) = d
p
V it follows that
d
dt
(E c) =
d
dt
E(p, v)
=
d
dt
K(p, v) +
d
dt
V (p)
=
d
dt
(
1
2
mv v) +
d
dt
V (p)
= mw v +
d
dt
V (p)
= [K](p, v, w) [ v) +d
p
V [ v)
= d
p
V [ v) +d
p
V [ v)
= 0
Hence our claim.

Lagrangian geodesic curvature eld
In conservative dynamics, the kinetic energy and the potential energy which
are the two ingredients generating T
Lagr
can be merged into a unique object,
as follows.
Dene a new Lagrangian function by putting
L := K V : TQ R : (p, v) L(p, v) := K(p, v) V (p)
(kinetic energy minus potential energy).
Associated with L, there is a Lagrangian geodesic curvature eld given by
[L] := [K] +dV : T
2
Q T
Q : (p, v; v, w) (p, [L](p, v, w))

with
[L](p, v, w) := [K](p, v, w) +d
p
V T
p
Q
(Riemannian geodesic curvature eld minus conservative eld).
From the above denitions, it follows that
T
Lagr
= D
EulLagr
that is, T
Lagr
does not dier from the Euler-Lagrange equation
D
EulLagr
:= (p, v; u, w) TTQ [ u = v , [L](p, v, w) = 0
generated by L.
Integral curves
The conditions characterizing the integral curves and the base integral curves
of Euler-Lagrange equation can now be formulated as follows.
Proposition 12 A smooth curve c of TQ is an integral curve of D
EulLagr
,
i
(
Q
c)
= c , [L] c = 0 ()
Proof The same as the proof, part (i), of Prop.8.
Proposition 13 A smooth curve of Q is a base integral curve of D

EulLagr
,
i its Lagrangian geodesic curvature vanishes, i.e.
[L] = 0 ()
Proof The same as the proof, part (i), of Prop.9.

Classical Euler-Lagrange equations
The local scalar equations obtained with the aid of a chart by equalling to
zero the components of the covector-valued function which appears in the left
hand side of ()
, are the classical Euler-Lagrange equations of Analytical

Dynamics.
Proposition 14 A smooth curve : t I p = (q(t)) | Q living
in the coordinate domain | of a chart of Q is a base integral curve of
D
EulLagr
, i its coordinate expression q = q(t) satises (for all h = 1, . . . , n)
the classical Euler-Lagrange equations
q = v ,
d
dt
L
v
h
(q,v)
L
q
h
(q,v)
= 0 ()
h
Proof Recall that is a base integral curve of D
EulLagr
, i it satises equa-
tion ()
is equivalent
to the n scalar equations obtained by equalling to zero the components in
of its left hand side, i.e.
([L] )
h
= 0 ()
h
The above components are given by
([L] )
h
:= ([L](p, v, w))
h
=
_
[L](p, v, w) [
p
q
h
q
_
=
_
[K](p, v, w) [
p
q
h
q
_
+
_
d
p
V [
p
q
h
q
_
=
d
dt
K
v
h
(q,v)
K
q
h
(q,v)
+
V
q
h
q
=
d
dt
(K V )
v
h
(q,v)
(K V )
q
h
(q,v)
=
d
dt
L
v
h
(q,v)
L
q
h
(q,v)
where we have put v = q .
So equations ()
h
h
.

Remark For both L := K and L := K V , the semi-basic 1-form [L] :
T
2
Q T
Q is a mapping which takes any (p, v; v, w) T

2
Q to the covector
(p, [L](p, v, w)) T
Q characterized, in a chart, by components

([L](p, v, w))
h
=

2
L
v
h
v
k
(q,v)
w
k
+

2
L
v
h
q
k
(q,v)
v
k
L
q
h
(q,v)
Such a coordinate technique can be used to dene [L] for an arbitrary La-
grangian function L : S TQ R (smooth on an open subset S of TQ),
since also in such a case the value [L](p, v, w) (for any (p, v) S ), dened
by the above components ([L](p, v, w))
h
in a given chart, turns out to be an
invariant covector.
9
2.4 Variational principle of stationary action
Euler-Lagrange equation takes conservative dynamics into the classical area of
variational principles, as will now be shown.
Variational calculus
Let
: I Q : t p(t)
be a smooth curve of Q.
A smooth variaton of with xed end-points in a closed sub-interval of I,
is a smooth mapping
: (, ) J Q : (s, t) p(s, t)
(with > 0 and J I ) satisfying
p(0, t) = p(t) , t J
and, at the end-points of a closed interval [t
1
, t
2
] J ,
p(s, t
1
) = p(t
1
) , p(s, t
2
) = p(t
2
) , s (, )
can be thought of as a one-parameter family

s
: J Q : t p
s
(t) := p(s, t)
s(,)
9
Actually, through higher geometric methods, [L] can be given a coordinate-free de-
nition.
of varied curves near [
J
=
o
with xed end-points in [t
1
, t
2
] J I ,
whose tangent lifts
s
s(,)
are all included in
: (, ) J TQ : (s, t)
_
p(s, t),
p
t
(s,t)
_
also dene a one-parameter family

t
: (, ) Q : s p
t
(s) := p(s, t)
tJ
of isocronous curves, whose tangent vectors at the points of [
J
are the values
of
o
: J TQ : t
_
p(t),
p
s
(0,t)
_
Now consider, in a closed sub-interval [t
1
, t
2
] of I , the action of , i.e.
the integral
10
1
:=
_
t
2
t
1
(L ) dt
For any smooth variation of with xed end-points in [t
1
, t
2
] , the
action of is then the real-valued function
1
:=
_
t
2
t
1
(L ) dt : (, ) R : s 1
s
:=
_
t
2
t
1
(L
s
) dt
At s = 0 , the value of 1
is 1
and its derivative
:=
d1
ds
0
is called the rst variation of 1
at .
Proposition 15 For every smooth variation of with xed end-points in
a closed sub-interval [t
1
, t
2
] of I , the rst variation of 1
at is related to
the Lagrangian geodesic curvature of by
=
_
t
2
t
1
[L] [
o
) dt
10
All of the following considerations and results of variational theory, hold true for an
arbitrary smooth Lagrangian function as well (on this matter, also recall the nal Remark
of the previous section).
Proof First remark that
:=
d
ds
0
_
t
2
t
1
(L ) dt =
_
t
2
t
1
(L )
s
s=0
dt
Now, for any given t
o
J , consider a chart on a neighbourhood | of the
point p(t
o
) Im . In a suitably small neighbourhood of (0, t
o
) (, ) J ,
takes values in | (by continuity) and will then have a local coordinate ex-
pression (q
h
(s, t)). As a consequence, will have local coordinate expression
_
q
h
(s, t),
q
h
t
(s,t)
_
, denoted, for s = 0 , by (q = q(t), v = v(t)) with v = q ,
and
o
will have components
o
h
:=
q
h
s
s=0
. With the aid of , we obtain
(L )
s
(0,to)
=
L
q
h
(q(to),v(to))
q
h
s
(0,to)
+
L
v
h
(q(to),v(to))
2
q
h
st
(0,to)
=
L
q
h
(q(to),v(to))
q
h
s
(0,to)
+
d
dt
to
_
L
v
h
(q,v)
q
h
s
s=0
_
_
d
dt
to
L
v
h
(q,v)
_
q
h
s
(0,to)
=
_
L
q
h
(q(to),v(to))
d
dt
to
L
v
h
(q,v)
_
o
h
(t
o
) +
d
dt
to
_
L
v
h
(q,v)
o
h
_
that is,
11
(L )
s
(0,to)
= [L] [
o
)(t
o
) +
d
dt
to
FL [
o
)
So, on the whole J, we have
(L )
s
s=0
= [L] [
o
) +
d
dt
FL [
o
)
As
o
(t
1
) = 0 and
o
(t
2
) = 0 , we also have
_
t
2
t
1
d
dt
FL [
o
) dt = FL [
o
)(t
2
) FL [
o
)(t
1
) = 0
Hence our claim.
11
FL : I T
Q will denote the (invariant) covector-valued map whose components

in are given by
(FL )
h
=
L
v
h
(q,v)
(see the nal Remark of Legendre transformation in Chap. 3).
Variational principle
A smooth curve : I Q is said to be a geodesic curve of (Q, L) , if it satises
Hamiltons variational principle of stationary action, owing to which, for every
smooth variation of with xed end-points in a closed sub-interval of I,
1
is required to be stationary at , that is to say,
= 0 ,
The above variational principle is completely equivalent to Euler-Lagrange
equation.
Proposition 16 The geodesic curves of (Q, L) are the base integral curves of
D
EulLagr
.
Proof (i) If : I Q is a base integral curve of D
EulLagr
, i.e.
[L] = 0
we clearly obtain, for every smooth variation of with xed end-points in
a closed sub-interval [t
1
, t
2
] of I ,
_
t
2
t
1
[L] [
o
) dt = 0
and hence, owing to Prop.15, is a geodesic curve of (Q, L) .
(ii) Conversely, if : t I p(t) Q is not a base integral curve of
D
EulLagr
, say
([L] )(t
o
) ,= 0 , t
o
I
we shall prove that, for a suitable smooth variation of with xed end-
points in a closed sub-interval [t
1
, t
2
] of I ,
_
t
2
t
1
[L] [
o
) dt ,= 0
which, owing to Prop.15, means that is not a geodesic curve of (Q, L) .
To this end, consider a chart on a neighbourhood
| p(t
o
)
Owing to our hypothesis, at least one of the components ([L] )
h
is non-null
at t
o
, say
([L] )
1
(t
o
) > 0
By continuity, there exists a suitably small open interval J I containing t
o
s.t., for all t J ,
p(t) | , ([L] )
1
(t) > 0
Now, for each s (, ) with a suitably small > 0 and each t J , we
consider the point p(s, t) | of coordinates
q
1
(s, t) := q
1
(t) +s(cos(t t
o
) cos)
q
2
(s, t) := q
2
(t)
. . . . . .
q
n
(s, t) := q
n
(t)
where (q
h
(t)) is the n-tuple of coordinates of p(t) and
0 < <

2
s.t. [t
o
, t
o
+] J
By doing so, we dene a map
: (, ) J Q : (s, t) p(s, t)
which is immediately seen to be a smooth variation of with xed end-points
in
[t
1
, t
2
] := [t
o
, t
o
+]
The components of
o
are , for all t J,
o
1
(t) = cos(t t
o
) cos
o
2
(t) = 0
. . . . . .
o
n
(t) = 0
and, for all t (t
1
, t
2
) ,
o
1
(t) > 0
Hence
_
t
2
t
1
[L] [
o
) dt =
_
t
2
t
1
([L] )
h
o
h
dt =
_
t
2
t
1
([L] )
1
o
1
dt > 0
which is our claim.

Riemannian case
For L = K , the above variational theory characterizes the geodesic curves of
Riemannian manifold (Q, K) through condition
[K] = 0
Owing to the positive deniteness of K , from the conservation law of
energy E = K it follows that, on the one hand, a geodesic curve satisfying
K = const. = 0
degenerates into a singleton and, on the other hand, a non-degenerate geodesic
curve is a uniform motion, that is,
K = const. > 0
The geometric meaning of non-degenerate geodesic curves will now be
shown.
Let us consider one more Lagrangian function, namely
:=
2K
which is clearly smooth on TQ K
1
(0) .
12
If : t I Q : t p(t) Q is a smooth curve satisfying
Im TQ K
1
(0)
(i.e. p(t) ,= 0 for all t I ) and [t
1
, t
2
] I , the integral
L
:=
_
t
2
t
1
( ) dt
denes the length of the arc of with end-points (p(t
1
), p(t
2
)) .
13
12
is smooth only on TQ K
1
(0) , since, in any chart, its partial derivatives
q
h
(q,v)
=
1
_
2K(q, v)
K
q
h
(q,v)
,

v
h
(q,v)
=
1
_
2K(q, v)
K
v
h
(q,v)
are not even dened for v = 0 , i.e. for v = 0 or, equivalently, (p, v) K
1
(0).
13
For any dt > 0 , (p(t), p(t)) dt =
_
g
p(t)
(dp) [ dp) > 0 denes the length, in the
given Riemannian metric, of the innitesimal arc dp := p(t)dt .
The variational theory previously sketched, applied to ,
14
shows that
is a curve of stationary length, that is,
= 0 ,
i
[] = 0
Proposition 17 is a non-degenerate geodesic curve of (Q, K) , i it is a
uniform motion of stationary length.
Proof The above result is an immediate consequence of the following
Lemma If is a uniform motion, that is,
K = = const. > 0
then
[] =
1
2
[K]
In order to prove the lemma, recall that, along the uniform motion : t
I p(t) Q, we have (for any given t I and any chart at p(t) )
15
([] )
h
=
d
dt
v
h
(q,v)

q
h
(q,v)
=
d
dt
_
1
2
K
v
h
(q,v)
_
2
K
q
h
(q,v)
=
1
2
_
d
dt
K
v
h
(q,v)
K
q
h
(q,v)
_
=
1
2
([K] )
h
which is our claim.
14
See footnote
10
.
15
See footnote
12
.
Chapter 3
From Lagrange to Hamilton
The Lagrangian dynamics of a conservative system o whose geometrical
arena is the velocity phase space TQ, supporting Euler-Lagrange equation
will now be taken into the range of Hamiltonian dynamics, where the geo-
metrical arena is the momentum phase space T
Q, supporting a Hamilton
equation which, dened in terms of the canonical symplectic structure of T
Q,
directly arises in normal form.
3.1 Legendre transformation
The classical transition from velocity to momentum phase space is provided
by the well known map which takes any velocity (p, v) TQ onto the corre-
sponding momentum (p, mv [
TpQ
) T
Q.
Lagrangian function and Legendre transformation
The above map is indeed the vector bundle isomorphism
g : TQ T
Q : (p, v) (p, g
p
(v)) , g
p
(v) := mv [
TpQ
owing to which the conguration space of the system has been given the struc-
ture of a Riemannian manifold (Q, K) .
1
Such an isomorphism is also said to be the Legendre transformation de-
termined by Lagrangian function L = K V (in terms of which it can be
expressed)
2
and is denoted by
FL : TQ T
Q : (p, v) (p, F
p
L(v)) , F
p
L(v) := g
p
(v)
(whence
Q
FL =
Q
).
1
See Riemannian geodesic curvature eld in Chap. 2.
2
See the next Coordinate formalism.
33
34 3 From Lagrange to Hamilton
Coordinate formalism
Recall that T
Q is a 2n-dimensional manifold, where each element (p, ) is

completely characterized, in a natural chart, by two n-tuples of coordinates
(q, p) , namely the coordinates q = (q
1
, . . . , q
n
) of p = (q) , in a chart of
Q, and the components p = (p
1
, . . . , p
n
) of in , given by p
h
= [
p
q
h
q
) .
In natural coordinate formalism, Legendre transformation
(p, v) (p, ) , = F
p
L(v)
is expressed by
3
(q, v) (q, p) , p
h
= g
hk
(q) v
k
=
L
v
h
(q,v)
(1)
The inverse transformation
(p, ) (p, v) , v = (F
p
L)
1
()
is in turn expressed by
(q, p) (q, v) , v
h
= g
hk
(q) p
k
=:
h
(q, p) (2)
Remark Such a coordinate technique can be used to dene a Legendre
morphism FL for an arbitrary Lagrangian function L : S TQ T
Q
(smooth on an open subset S of TQ), since also in such case the value
= F
p
L(v) T
p
Q (for any (p, v) S ), dened by the above components
p
h
=
L
v
h
(q,v)
in a given chart, turns out to be an invariant covector.
4
L is then said to be a regular or a singular Lagrangian function, according
to whether FL is injective or not.
3
Recall that
p
h
=

g
p
(v) [
p
q
h
q
_
=

g
p
_
p
q
h
q
_
[ v
_
=

g
(q)
_
p
q
h
q
_
[
p
q
k
q
_
v
k
= g
hk
(q) v
k
=
K
v
h
(q,v)
=
L
v
h
(q,v)
4
Actually, through higher geometric methods, FL can be given a coordinate-free def-
inition.
3.2 Hamiltonian function
Legendre transformation also determines transition from the Lagrangian func-
tion (on TQ) to a Hamiltonian function (on T
Q) .
Energy and Hamiltonian function
Recall that, associated with Lagrangian function L = K V , there is the
energy function
5
E := K +V = 2K L = FL [ id
TQ
) L : TQ R
The Hamiltonian function
H := E (FL)
1
: T
Q R
is the push forward of E by Legendre transformation FL.
In natural coordinate formalism, the Hamiltonian function
(p, ) H(p, ) = E((FL)
1
(p, )) = E(p, (F
p
L)
1
()) = E(p, v) = F
p
L(v) [ v) L(p, v)
= [ v) L(p, v) , v = (F
p
L)
1
()
is expressed by
(q, p) H(q, p) = p
k
v
k
L(q, v) , v = (q, p)
Hence, owing to (1) and (2), we obtain
H
q
h
(q,p)
= p
k
v
k
q
h
(q,p)
L
v
k
(q,v)
v
k
q
h
(q,p)
L
q
h
(q,v)
=
L
q
h
(q,v)
(3)
and
H
p
h
(q,p)
= v
h
+p
k
v
k
p
h
(q,p)
L
v
k
(q,v)
v
k
p
h
(q,p)
=
h
(q, p) (4)
5
We put
FL [ id
TQ
) : (p, v) TQ F
p
L(v) [ id
TQ
(v)) = F
p
L(v) [ v) = g
p
(v) [ v) = 2K(p, v) R
3.3 Hamiltonian vector eld
The canonical symplectic geometry of T
Q will now be taken into considera-

tion.
Canonical symplectic structure and Hamiltonian vector eld
Let
: TM T
M : (, X) (,
(X))
be the canonical symplectic structure of cotangent bundle M := T
Q.
6
As is a vector bundle isomorphism of TM onto T
M , it transforms
the dierential
dH : M T
M
of Hamiltonian function H into the Hamiltonian vector eld
X
H
:=
1
dH : M TM
Recall that, on the 2n-dimensional manifold M = T
Q, the canonical sym-

plectic structure is characterized, in any natural chart, by the 2n 2n
matrix of components
_

(1)(1)

(1)(2)
(2)(1)

(2)(2)
_
with
(1)(1)
=
(2)(2)
= 0 (nn zero matrix) and
(1)(2)
=
(2)(1)
= (nn
identity matrix).
If, for any M ,
_
d
H
(1)
d
H
(2)
and
_
X
H
()
(1)
X
H
()
(2)
_
are the 2n-tuples of components of d
H T
M and X
H
() T
M , respec-
tively, the linear mapping
d
H =
(X
H
())
is expressed by
7
d
H
(1)
=
()(1)
X
H
()
()
, d
H
(2)
=
()(2)
X
H
()
()
6
Any point (p, ) T
Q is now referred to as M .
7
Summation over () = (1), (2) is understood.
that is,
d
H
(1)
= X
H
()
(2)
, d
H
(2)
= X
H
()
(1)
As
d
H
(1)
=
_
H
q
h
(q,p)
_
, d
H
(2)
=
_
H
p
h
(q,p)
_
we obtain
X
H
()
(1)
=
_
H
p
h
(q,p)
_
, X
H
()
(2)
=
_
H
q
h
(q,p)
_
(5)
3.4 Hamilton equation
Remark that Legendre transformation FL : TQ T
Q transforms each
smooth curve c : I TQ of velocity phase space into a smooth curve FLc :
I T
Q of momentum phase space.

Cotangent dynamically possible motions
If c is an integral curve of D
EulLagr
, i.e. a TDPM of o , then FL c will
be said to be a cotangent dynamically possible motion (CDPM) of o .
As FL is an isomorphism, TDPMs and CDPMs bijectively correspond
to one another. Through such a bijection, the problem of determining the
TDPMs proves to be naturally equivalent to that of determining the CDPMs.
Hamilton equation
Proposition 18 The CDPMs are the integral curves of Hamilton equation
D
Ham
:= ImX
H
Proof Let
k : I T
Q , c : I TQ
be smooth curves corresponding to each other through Legendre transforma-
tion, i.e.
k = FL c
and then both projecting down onto the same curve
Q
k =
Q
FL c =
Q
c : I Q
We shall prove that
() (
Q
c)
= c , [L] c = 0

k = X
H
k ()
For any t I , the natural coordinates (q, p) = (q(t), p(t)) of k = k(t) are
then related to the natural coordinates (q, v) = (q(t), v(t)) of c = c(t) by (1)
and (2), i.e.
p
h
=
L
v
h
(q,v)
, v
h
=
h
(q, p).
So condition (), that is,
q
h
= v
h
,
d
dt
L
v
h
(q,v)
L
q
h
(q,v)
= 0
is equivalent, owing to (1) and (2), to
q
h
=
h
(q, p) , p
h
=
L
q
h
(q,v)
and then, owing to (3) and (4), to
q
h
=
H
p
h
(q,p)
, p
h
=
H
q
h
(q(p)
()
h
which, owing to (5), is just condition ().
8
Remark the second-order character of the above Hamilton equation, ex-

hibeted by the fact that any integral curve k of D
Ham
is the Legendre lift of
its own projection
Q
k , i.e.
k = FL c = FL (
Q
c)
= FL (
Q
k)
The integral curves and the base integral curves of D

Ham
are then bi-
jectively related to one another by projection
Q
and, owing to the above
bijection, the rst-order problem of determining the integral curves of D
Ham
proves to be naturally equivalent to the second-order problem of determining
its base integral curves, i.e. the solution curves of
(FL )
= X
H
(FL )
Classical Hamilton equations
The local scalar equations ()
h
, obtained in a natural chart by orderly equalling
the components of the left and right hand side of (), are the classical Hamil-
ton equations of Analytical Dynamics.
8
Recall that
_
q
h
p
h
_
are the components of

k .
Chapter 4
Concluding remarks
Some comments on the previous results and some perspectives on further de-
velopments, now follow.
4.1 Inertia and force
Propositions 1, 3 and 6 have shown that the mathematical model actually
underlying classical dynamics is given by the triplet
/ := (Q, K, F)
which will still be called mechanical system.
The corresponding Lagrange equation
T
Lagr
= (p, v; v, w) TTQ [ u = v , [K](p, v, w) = F(p, v)
is then the dynamics of /, the DPMs of / being dened as the base
integral curves of T
Lagr
and hence characterized by condition
[K] = F
If F = 0 , the DPMs of / characterized by a Riemannian geodesic
curvature [K] identically vanishing coincide with the geodesic curves of
Riemannian manifold (Q, K) , which can therefore be called inertial motions
in (Q, K) .
If F ,= 0 , the DPMs of / characterized by a Riemannian geodesic
curvature [K] diering from zero as much as is imposed by the covector
force F are then to be called forced motions in (Q, K) , since they appear
to be perturbed with respect to the above inertial trend.
So K and F seem to correspond to the empirical notions of inertia
(dening inertial motions) and force (dening forced motions), respectively.
39
40 4 Concluding remarks
4.2 Gauge transformations
However the above notions of inertia and force are not uniquely determined
by dynamics.
In fact, there exist gauge transformations of /, which alter the geo-
metrical ingredient K and the dynamical ingredient F without altering the
dynamics itself, as will now be shown.
For any real-valued smooth function V : Q R, consider the gauge
transformation
K L := K V , F F := F +dV
On the one hand, such a transformation gives rise to a new mechanical
system
M:= (Q, L, F)
with Lagrange equation
D
Lagr
= (p, v; v, w) T
2
Q [ u = v , [L](p, v, w) = F(p, v)
and DPMs the base integral curves of D
Lagr
characterized by condition
[L] = F
If F = 0 , the DPMs of M characterized by a Lagrangian geodesic
curvature [L] identically vanishing coincide with the geodesic curves of
Lagrangian manifold (Q, L) , which will therefore be called inertial motions
in (Q, L) .
If F ,= 0 , the DPMs of M characterized by a Lagrangian geodesic
curvature [L] diering from zero as much as is imposed by the covector
force F are then to be called forced motions in (Q, L) , since they appear
to be perturbed with respect to the above inertial trend.
On the other hand, from the very denitions of
[L](p, v, w) := [K](p, v, w) +d
p
V
and
F(p, v) := F(p, v) +d
p
V
for all (p, v; v, w) T
2
Q, it follows that
[L](p, v, w) = F(p, v)
i
[K](p, v, w) = F(p, v)
4 Concluding remarks 41
whence
D
Lagr
= T
Lagr
So transition from / to M just leads to dierent specications of the
conventional notions of inertia and force, without altering the observable class
of DPMs.
4.3 Geometrizing physical elds
From the physical point of view, the meaning of the above kind of gauge
transformations can be illustrated as follows.
Think of V as the potential energy of a conservative component of F,
that is to say, put F =

F + f , f being a conservative eld with virtual work
f = dV .
Now look at the ingredients L and F of M.
On the one hand, F = F f is the virtual work of

F = F f , that
is to say, the conservative eld does not appear any more in the dynamical
ingredient of M.
On the other hand, L embodies V , that is to say, the conservative eld
is encompassed in the geometrical ingredient of M as a potential function,
whose eect is that of transforming the natural Riemannian geometry K of
the particle system into a perturbed Lagrangian geometry L = K V .
That anticipates, in a sense, Einsteins idea of geometrizing physical elds.
4.4 Geometrical dynamics
From the mathematical point of view, gauge transformations suggest a gener-
alized geometrical dynamics.
In such a theory, a mechanical system will be conceived as an arbitrary
triplet M consisting of a smooth manifold Q equipped with a (regular or a
singular) Lagrangian function L and a semibasic 1-form F (both smooth on
an open subset of TQ).
The dynamics of M will then be dened by the corresponding Lagrange
equation D
Lagr
(which will or will not be reducible to normal form, according
to whether L is a regular or a singular Lagrangian function, respectively).
The global analysis of D
Lagr
, its Hamiltonian formulation (or Hamilton-
Dirac formulation, in the general case of L being a singular Lagrangian) and
further generalizations, as well as applications to classical and relativistic dy-
namics, are the object of (part of) the current research work on geometrical
dynamics.
Chapter 5
Appendix
The geometry of smooth manifolds embedded in Euclidean ane spaces, is
the main topic of this Appendix.
1
Tangent bundles and dierential equa-
tions, cotangent bundles and dierential forms, will be made to fall into its
range. The above geometry will nally be re-read, so as to lead to the mod-
ern approach to smooth manifolds (which does away with any reference to
embeddings into Euclidean environments).
5.1 Submanifolds of a Euclidean space
A submanifold of a Euclidean ane space will be presented as a locus which
locally exhibits the same dierential-topological behaviour as an ane sub-
space (that is suggested from the familiar circumstance of smooth curves or
surfaces locally resembling straight lines or planes, respectively).
Ane subspaces
Let c be an m-dimensional Euclidean ane space, modelled on a vector space
E (acting freely and transitively on c as an Abelian group of operators).
2
A subset / c is said to be an n-dimensional Euclidean ane subspace
of c , if it is the orbit of a point o c under the action of an n-dimensional
vector subspace A E , i.e. / = o +A := o +v : v A .
3
1
The main (linear, metric, topological) properties of Euclidean ane spaces and the
fundamentals of dierential calculus on such spaces are the prerequisites for reading this
Appendix.
2
E is meant to be a vector space on the eld R of real numbers. The elements of E
will be called free vectors, whereas for any point p c the elements of p E will be
said to be vectors attached at p. The action of E on c will be denoted by +.
3
The name ane subspace is clearly due to the fact that condition / = o + A is
42
5 Appendix 43
Such a subspace / can be described in c by means of global ane
parametrizations (or systems of ane coordinates), say
/ = Im
with
4
: R
n
c : q p = (q) = o + (q) = o +q
h
e
h
(where o / and : v R
n
(v) = v
h
e
h
E is the injective linear map
determined by a basis (e
1
, . . . , e
n
) of A).
From the dierential-topological point of view, the above kind of parametriza-
tion exhibits the following well known properties:
(1) : R
n
/ is a homeomorphism;
5
(2) : R
n
c is C
- dierentiable, with injective dierential d

q
=
at each q R
n
.
6
Embedded submanifolds
Let c be an m-dimensional Euclidean ane space (modelled on E ).
A subset Q c carrying the structure of an n-dimensional embedded
submanifold of c (or smooth manifold embedded in c ), is one s.t. each point
p Q admits an open neighbourhood | , in the subspace topology of Q,
which is the image
| = Im
of a local parametrization
: W R
n
c : q p = (q)
dened on an open subset W of R
n
and satisfying the following topological-
dierential properties:
necessary and sucient for a subset / of c to inherit from the ane environment a
structure of ane space modelled on a vector subspace A of E . Familiar examples are the
straight lines and the planes (i.e. the ane subspaces of dimension 1 and 2, respectively).
4
A repeated index, in lower and upper position, will denote summation.
5
Here we refer to the subspace topology of /.
6
Recall that a mapping like is C
- dierentiable, i it is continuous and admits

continuous partial derivatives of any order. Its dierential at q is the linear map
d
q
: R
n
E : v = (v
h
) v = v
h
p
q
h
q
44 5 Appendix
1. : W | is a homeomorphism;
2. : W c is C
- dierentiable, with injective dierential d

q
at each
q W .
Remark that d
q
is injective, i its image
Imd
q
= Span
_
p
q
h
q
_
(spanned by the rst-order partial derivatives of ) is an n-dimensional vector
subspace of E .
Hence n m.
Atlas of charts
Owing to the denition, a manifold Q c is covered by an atlas of open
subsets | described in c by means of local parametrizations . Each point
p | is then given a unique n-tuple of coordinates q =
1
(p) W by the
inverse map
1
: | W , which is called an n-dimensional local chart (or
local coordinate system) on Q (it is said to be global in the special case of its
coordinate domain | being equal to the whole Q).
7
Locally Euclidean topology
From a topological point of view, Q is a locally Euclidean subspace of c , in
the sense that it is covered by open subsets (namely, the coordinate domains)
which, owing to property 1, are all homeomorphic to open subsets of Euclidean
space R
n
.
Smoothness
From a dierential point of view, Q is a smooth subspace of c , in the sense
that it admits, owing to property 2, an n-dimensional tangent vector space
at each one of its points, as will now be shown.
Smooth curves and tangent vectors
Recall that tangency is a local (innitesimal) concept, linked to derivation
as follows.
7
In the sequel, too will be called a chart.
5 Appendix 45
Consider a smooth curve of c , i.e. a C
- dierentiable c-valued map

: I c : t (t) = p(t)
dened on an open interval I R ( Im will be called the orbit of ).
The vector p(t) E obtained from by derivation at any t I , is said
to be the free vector tangent at p(t) to .
8
Tangent vector spaces
Now consider the smooth curves of c with orbits lying on Q, i.e.
Im Q
(also referred to as smooth curves of Q and denoted by : I Q), and
passing through a given point p Q, i.e.
p Im
A vector of E will be said to be a free vector tangent at p to Q, if it is
tangent at p to one of the above curves.
The set
T
p
Q E
of all the free vectors tangent at p to Q will be called tangent vector space of
Q at p , by virtue of the following result.
Proposition 19 T
p
Q is an n-dimensional vector subspace of E .
Proof Follows from property 2, owing to the following
Lemma For any chart : W | at p (i.e. | p = (q) , with q W ),
T
p
Q = Imd
q
()
(i) Let v T
p
Q, i.e.
v = p(t
o
)
for some smooth curve : t I p(t) Q through p(t
o
) = p (with t
o
I ).
8
Think of the graphic representation of derivative p(t) := lim
t0
1
t
_
p(t +t) p(t)
_
when it does not vanish as an oriented segment, limit of a secant segment, attached at
p(t) .
46 5 Appendix
As | is an open neighbourhood of p(t
o
) = p in Q, there exists (by continuity)
an open neighbourhood I
o
of t
o
in I s.t. (I
o
) | . As a consequence, ca
be given the local C
coordinate expression
=
1
[
Io
: I
o
W : t q(t) =
1
(p(t))
By derivation at t
o
of p(t) = (q(t)) , we obtain
v = p(t
o
)
= q
h
(t
o
)
p
q
h
q(to)
= d
q(to)
( q(t
o
))
= d
q
(v) , v := q(t
o
) R
n
Hence v Imd
q
.
(ii) Now let v Imd
q
, i.e.
v = d
q
(v) , v R
n
Choose t
o
R and consider the ane map
: R R
n
: t q(t) := q + (t t
o
)v
As W is an open neighbourhood of q(t
o
) = q in R
n
, there exists (by continu-
ity) an open neighbourhood I
o
of t
o
in R s.t. (I
o
) W . As a consequence,
= [
Io
: I
o
| : t p(t) = (q(t))
is a smooth curve of Q through
p(t
o
) = (q(t
o
)) = (q) = p
whose derivative at t
o
is
p(t
o
) = q
h
(t
o
)
p
q
h
q(to)
= d
q(to)
( q(t
o
))
= d
q
(v)
= v
Hence v T
p
Q.
Remark that
v T
p
Q v = (d
q
)
1
(v) R
n
is the linear isomorphism which takes any vector v T
p
Q to its components
v R
n
in the basis
_
p
q
h
q
_
of T
p
Q (called linear components of v in ).
5 Appendix 47
Open submanifolds
Trivial examples of submanifolds of c are the open subets.
Any system of ane coordinates on c restricted to an open subset U c
determines an m-dimensional global chart on U , which is therefore an m-
dimensional embedded submanifold of c and, for all p U ,
T
p
U = E
For instance, if
g = (g
1
, . . . , g
) : c R
: p g(p) = (g
1
(p), . . . , g
(p))
is a continuous map, then
U := g
1
(R
+
)
= p c [ g(p) (R
+
)
= p c [ g
1
(p) > 0, . . . , g
(p) > 0
(inverse image by g of the open subset (R
+
)
) is an open submanifold
of c .
Implicit function theorem
Non-trivial examples of submanifolds of c arise from the well known Implicit
Function Theorem, concerning loci described by means of algebraic equations.
Proposition 20 Let
f = (f
1
, . . . , f
) : U c R
: p f(p) = (f
1
(p), . . . , f
(p))
be a dierentiable map, dened on an open subset U c (say U = g
1
(R
+
)
)
and taking values in R
( < m := dimc ), with surjective dierential d

p
f :
E R
at each point p of
Q := f
1
(0) = p U [ f(p) = 0 = p U [ f
1
(p) = 0, . . . , f
(p ) = 0
= p c [ f
1
(p) = 0, . . . , f
(p ) = 0, g
1
(p) > 0, . . . , g
(p) > 0
Q is then an n-dimensional embedded submanifold of c with
n := m
and, for all p Q,
T
p
Q = Ker d
p
f
48 5 Appendix
Proof (i) The above mentioned theorem of Analysis states that, for any p Q,
the fact of d
p
f being surjective implies the existence of an open neighbourhood
| of p in Q, an open subset W R
n
and a -tuple of real-valued functions
1
: W R, . . . ,
: W R s.t.
| = Im
where
: W c
is a map dened (with a suitable choice of ane coordinates : R
m
c ) by
(x
1
, . . . , x
n
)

( x
1
, . . . , x
n
,
1
(x
1
, . . . , x
n
), . . . ,
(x
1
, . . . , x
n
) )
and satisfying properties 1 and 2, i.e. an n-dimensional chart on Q.
9
(ii) As is composable with f and f = 0 , we have
d
p
f d
q
= d
q
(f ) = 0
(with p = (q) ) and then
Imd
q
Ker d
p
f
Moreover, the surjectivity of d
p
f and the injectivity of d
q
imply the dimen-
sional result
dimKer d
p
f = dimE dimImd
p
f = m = n = dimImd
q
So we obtain
Imd
q
= Ker d
p
f
which, owing to (), completes our proof.
9
So, on
1
(|) , the algebraic equations
f
1
_
(x
1
, . . . , x
n
, x
n+1
, . . . , x
n+
)
_
= 0, . . . , f
_
(x
1
, . . . , x
n
, x
n+1
, . . . , x
n+
)
_
= 0
implicitly dene x
n+1
, . . . , x
n+
as functions of (x
1
, . . . , x
n
) W , i.e.
x
n+1
=
1
(x
1
, . . . , x
n
), . . . , x
n+
=
(x
1
, . . . , x
n
)
5 Appendix 49
5.2 Tangent bundle and dierential equations
The tangent bundle of a manifold is the rst geometric arena of dierential
equations.
Tangent bundle and canonical projection
The tangent bundle of an n-dimensional smooth manifold Q c is the disjoint
union of its tangent vector spaces, i.e.
TQ =
_
pQ
p T
p
Q
(for any p Q, the set p T
p
Q is made up of all the attached vectors
tangent at p to Q).
The canonical projection of TQ onto its base manifold Q is the map
Q
: TQ Q : (p, v) p
Vector bre bundle structure
TQ is a vector bre bundle over Q, in the sense that its bre p T
p
Q over
each p Q is a vector space (canonically isomorphic to T
p
Q).
Such bres are glued together by natural charts (arising from the charts
: W | of Q)
1
: (q, v) W R
n
(p, v) = ((q), d
q
(v))
1
Q
(|)
which give TQ the structure of a 2n-dimensional smooth manifold embedded
in Tc = c E .
10
Tangent lift
The tangent lift of a smooth curve
: I Q : t (t) = p(t)
living in Q c , is the smooth curve
: I TQ : t (t) = (p(t), p(t))
living in TQ Tc .
10
Remark that, for any open subset U c (and then for U = c as well), TU = U E .
Recall that Tc = c E is a Euclidean ane space, modelled on E E .
50 5 Appendix
Dierential equations in implicit form
From the very denition of tangent vector spaces, it follows that TQ is the
region of Tc swept by the tangent lifts of all the smooth curves of Q (i.e.
covered by the orbits of such lifts).
As a consequence, if a subset
D TQ
is assigned (by prescribing some geometric or algebraic property), the problem
may arise of determining the smooth curves of Q whose tangent lifts live
in D, i.e.
Im D ()
that is to say,
(t) = (p(t), p(t)) D , t I ()
Such a D will be called (autonomous) rst-order dierential equation in
implicit form on Q.
11
Integral curves
Any solution to the above problem, i.e. any smooth curve of Q satisfying
condition (), is called an integral curve of D (or maximal integral curve, if
it is not restriction of any other integral curve).
Integrable part
The problem of establishing the existence of integral curves is the same as that
of determining the integrable part of D, which is the region
D
(i)
D
swept by the tangent lifts of all its integral curves.
D is said to be integrable, if
, = D = D
(i)
11
We shall not consider the case of a non-autonomous equation D J
1
Q, subset of
the jet bundle J
1
Q := R TQ, whose solutions are the smooth curves of Q with jet
extension j
1
: t I j
1
(t) := (t, p(t), p(t)) J
1
Q living in D, i.e. Im j
1
D.
5 Appendix 51
Vector elds and normal form
The main case of integrability is that of a dierential equation D reducible to
normal form, that is, one for which there exists a smooth vector eld
X : Q TQ : p (p, X(p))
on Q
12
s.t.
D = ImX = (p, v) TQ [ v = X(p)
For such an equation, condition () reads
= X
that is to say,
p(t) = X(p(t)) , t I
Cauchy problems and determinism
The integrability of
D = ImX ,=
is a consequence of the following fundamental property of normal equations
(here stated without proof).
Determinism theorem: For any t
o
R and p
o
Q, there exists a unique
maximal solution to Cauchy problem (D, t
o
, p
o
) , i.e. a unique maximal inte-
gral curve : t I p(t) Q of D, with I t
o
, satisfying initial condition
p(t
o
) = p
o
(all of the other solutions to the problem being restrictions of the
maximal one).
Indeed, for any (p
o
, v
o
) D, i.e. p
o
Q and v
o
= X(p
o
) , we obtain
(p
o
, v
o
) = (p(t
o
), X(p(t
o
))) = (p(t
o
), p(t
o
)) Im
o
D
(i)
whence
D = D
(i)
12
X is said to be smooth, if in any chart : W | it admits C
linear components
q W (d
q
)
1
_
X((q))
_
R
n
52 5 Appendix
5.3 Second tangent bundle and second-order
dierential equations
The second tangent bundle of a manifold is the geometric arena of second-order
dierential equations.
Second tangent lift and second tangent bundle
Consider the tangent bundle
TTQ TTc
of TQ Tc .
Recall that TTQ is the region of TTc = (c E) (E E) swept by the
tangent lifts of all the smooth curves of TQ, the tangent lift of any such curve
c : I TQ : t c(t) = (p(t), v(t))
being given by
c : I TTQ : t c(t) = (p(t), v(t); p(t), v(t))
In particular, the second tangent lift of a smooth curve
: I Q : t (t) = p(t)
living in Q, is the tangent lift of
: I TQ : t (t) = (p(t), p(t))
i.e. the smooth curve
13
: I T
2
Q : t (t) = (p(t), p(t); p(t), p(t))
living in the region T
2
Q TTQ, called second tangent bundle of Q, dened
by
T
2
Q = (p, v; u, w) TTQ [ u = v
13
p(t) will denote the derivative of p(t) .
5 Appendix 53
Ane bre bundle structure
Remark that
T
2
Q =
_
(p,v)TQ
(p, v)
_
v T
2
(p,v)
Q
_
with
T
2
(p,v)
Q := w E [ (p, v; v, w) TTQ
is an ane bre bundle over TQ, in the sense that its bre (p, v)
_
v T
2
(p,v)
Q
_
over each (p, v) TQ is an ane space (canonically iso-
morphic to T
2
(p,v)
Q), as will now be shown.
Proposition 21 For any (p, v) TQ, T
2
(p,v)
Q is an n-dimensional ane
subspace of E modelled on T
p
Q, i.e.
T
2
(p,v)
Q = z +T
p
Q
for some z E .
Proof Follows from property (), owing to the following
Lemma For any natural chart
1
: WR
n

1
Q
(|) at (p, v) (i.e.
1
Q
(|)
(p, v) = ((q), d
q
(v)) with (q, v) W R
n
),
T
2
(p,v)
Q = z(q, v) + Imd
q
()
where z(q, v) := v
h
v
k
2
p
q
h
q
k
q
.
(i) Let w T
2
(p,v)
Q, that is,
(p, v; v, w) TTQ
or
(p, v; v, w) = c(t
o
) = (p(t
o
), v(t
o
); p(t
o
), v(t
o
))
for some smooth curve c : t I c(t) = (p(t), v(t)) TQ and t
o
I .
As | is an open neighbourhood of p = p(t
o
) in Q, there exists (by continuity)
an open neighbourhood I
o
of t
o
in I s.t. p(t) | for all t I
o
. Then c[
Io
can be given the C
coordinate expression
p(t) = (q(t)) , v(t) = d
q(t)
(v(t)) = v
h
(t)
p
q
h
q(t)
with
q(t
o
) = q , q(t
o
) = v(t
o
) = v
54 5 Appendix
By derivation at t
o
, we obtain
w = v(t
o
)
= v
h
(t
o
) q
k
(t
o
)

2
p
q
h
q
k
q(to)
+ v
h
(t
o
)
p
q
h
q(to)
= v
h
v
k

2
p
q
h
q
k
q
+w
h
p
q
h
q
, w := v(t
o
)
= z(q, v) +d
q
(w)
Hence w z(q, v) + Imd
q
.
(ii) Now let w z(q, v) + Imd
q
, that is,
w
=
z(q, v) +d
q
(w)
for some w R
n
.
Choose t
o
R and consider the map
: R R
n
: t q(t) := q +v(t t
o
) +
1
2
w(t t
o
)
2
satisfying
q(t
o
) = q , q(t
o
) = v , q(t
o
) = w
As W is an open neighbourhood of q(t
o
) = q in R
n
, there exists (by continu-
ity) an open neighbourhood I
o
of t
o
in R s.t. (I
o
) W . As a consequence,
= [
Io
: I
o
| : t p(t) = (q(t))
is a smooth curve of Q through
p(t
o
) = (q(t
o
)) = (q) = p
whose derivatives at t
o
are
p(t
o
) = q
h
(t
o
)
p
q
h
q(to)
= d
q(to)
( q(t
o
))
= d
q
(v)
= v
and
p(t
o
) = q
h
(t
o
) q
k
(t
o
)

2
p
q
h
q
k
q(to)
+ q
h
(t
o
)
p
q
h
q(to)
= v
h
v
k

2
p
q
h
q
k
q
+w
h
p
q
h
q
= z(q, v) +d
q
(w)
= w
5 Appendix 55
So
(p, v; v, w) = (p(t
o
), p(t
o
); p(t
o
), p(t
o
)) = (t
o
) Im T
2
Q TTQ
Hence w T
2
(p,v)
Q.
Remark that
w T
2
(p,v)
Q w = (d
q
)
1
(wz(q, v)) R
n
is the ane isomorphism which takes any vector w T
2
(p,v)
Q to its ane
coordinates w R
n
in the frame
_
z(q, v),
_
p
q
h
q
_
_
of T
2
(p,v)
Q (called ane
components of w in ).
Second-order dierential equations in implicit form
Part (ii) of the above proof shows as well that T
2
Q is the region of TTQ
swept by the second tangent lifts of all the smooth curves of Q.
As a consequence, if a subset
T T
2
Q
is assigned (by prescribing some geometric or algebraic property), the problem
may arise of determining the smooth curves : t I p(t) Q whose
second tangent lifts live in T, i.e.
Im T ()
that is to say,
(t) = (p(t), p(t); p(t), p(t)) T , t I ()
Such a T will be called (autonomous) second-order dierential equation
in implicit form on Q.
56 5 Appendix
Integral curves and base integral curves
A second-order dierential equation on Q, say T T
2
Q TTQ, is a rst-
order dierential equation on TQ as well, whose integral curves exhibit the
following peculiar property.
Proposition 22 Each integral curve c of T T
2
Q is the tangent lift of its
own projection
Q
c , i.e.
(
Q
c)
= c
Proof Let
c : t I c(t) = (p(t), v(t)) TQ
be a smooth curve of TQ, and
Q
c : t I
Q
(c(t)) = p(t) Q
its projection onto Q.
If c is an integral curve of T, it satises, for all t I ,
c(t) = (p(t), v(t); p(t), v(t)) T
whence
p(t) = v(t)
that is,
(
Q
c)
(t) = (p(t), p(t)) = (p(t), v(t)) = c(t)

which is our claim.
The base integral curves =

Q
c of T projections of the integral curves
c of T are the solution curves of problem (), as will now be shown.
Proposition 23 A smooth curve of Q is a base integral curve of T, i it
is a solution curve of problem ().
Proof (i) Let Im T. Putting c := , whence
Q
c =
Q
and c = ,
we obtain
Q
c = and Im c T.
(ii) Conversely, let =
Q
c and Im c T. From Prop. 22, we obtain
= (
Q
c)
= c , whence = c , and then Im T.
The integral curves and the base integral curves of T are then bijectively
related to one another by projection
Q
(Prop. 22) and, owing to the above bi-
jection, the rst-order problem of determining the integral curves of T proves
to be naturally equivalent to the second-order problem of determining its base
integral curves, i.e. the solution curves of () (Prop. 23).
5 Appendix 57
Semi-sprays and normal form
A second-order equation T is then reducible to normal form, if there exists a
smooth semi-spray
: TQ T
2
Q TTQ : (p, v) (p, v; v, (p, v))
on TQ
14
s.t.
T = Im = (p, v; u, w) TTQ [ u = v , w = (p, v)
Proposition 24 Condition Im c T, characterizing the integral curves c of
T = Im, reads
c = c
Condition Im T, characterizing the base integral curves of T = Im,
reads
=
Proof (i) For a smooth curve c : t I c(t) = (p(t), v(t)) TQ, condition
c(t) = (p(t), v(t); p(t), v(t)) T = Im , t I
reads
p(t) = v(t) , v(t) = (p(t), v(t)) , t I
that is,
c(t) = (p(t), v(t); p(t), v(t))
= (p(t), v(t); v(t), (p(t), v(t)))
= ( c) (t) , t I
(ii) For a smooth curve : t I p(t) Q, condition
(t) = (p(t), p(t); p(t), p(t)) T = Im , t I
reads
p(t) = (p(t), p(t)) , t I
that is,
(t) = (p(t), p(t); p(t), p(t))
= (p(t), p(t); p(t), (p(t), p(t)))
= ( ) (t) , t I
14
(vector eld on TQ taking values in T
2
Q) is said to be smooth, if in any chart
: W | it admits C
ane components
(q, v) W R
n
(d
q
)
1
_
((q), d
q
(v)) z(q, v)
_
R
n
58 5 Appendix
Cauchy problems and determinism
If T = Im, then owing to determinism theorem Cauchy problem (T, t
o
, (p
o
, v
o
)) ,
for any t
o
R and (p
o
, v
o
) TQ, admits a unique maximal solution, which
amounts to saying that there exists a unique maximal base integral curve
: t I p(t) Q of T, with t
o
I , satisfying initial conditions
(p(t
o
), p(t
o
)) = (p
o
, v
o
) (all of the other solutions to the problem being restric-
tions of the maximal one).
5.4 Cotangent bundle and dierential forms
The cotangent bundle of a manifold is the geometric arena of dierential forms.
Cotangent vector spaces
For any p Q, to the tangent vector space T
p
Q there corresponds, by duality,
the cotangent vector space T
p
Q made up of all the covectors
: T
p
Q R : v [ v)
(linear maps of T
p
Q in R).
Covectors can naturally be summed to one another and multiplied by
scalars, so that T
p
Q turns into a vector space.
To the basis
_
p
q
h
q
_
determined in T
p
Q by a chart at p = (q) , there
corresponds a dual basis (d
p
q
h
) in T
p
Q, dened by
d
p
q
h
: T
p
Q R : v = v
k
p
q
k
q
d
p
q
h
[ v) := v
h
The components (p
h
) which characterize any covector T
p
Q in the
above basis called components of in are given by
p
h
:=
_
[
p
q
h
q
_
since, by pairing with all v T
p
Q, we obtain
[ v) =
_
[ v
h
p
q
h
q
_
=
_
[
p
q
h
q
_
v
h
= p
h
v
h
= p
h
d
p
q
h
[ v) = p
h
d
p
q
h
[ v)
that is,
= p
h
d
p
q
h
5 Appendix 59
Cotangent bundle and canonical projection
The cotangent bundle of an n-dimensional smooth manifold Q is the disjoint
union of its cotangent vector spaces, i.e.
T
Q :=
_
pQ
p T
p
Q
The canonical projection of T
Q onto its base Q is the map
Q
: T
Q Q : (p, ) p
Vector bundle structure
T
Q is a vector bre bundle over Q, since its bre p T
p
Q over each
p Q is a vector space (canonically isomorphic to T
p
Q).
Such bres are glued together by natural charts (arising from the charts
: W | of Q)
1
: (q, p) W R
n
(p, ) = ((q), p
h
d
p
q
h
)
1
Q
(|)
which give T
Q the structure of a 2n-dimensional smooth manifold.

15
Dierential 1-forms
A smooth real-valued function V : Q R
16
gives rise to a covector eld
dV : Q T
Q : p (p, d
p
V )
called dierential of V or exact 1-form on Q, where the dierential d
p
V
T
p
Q of V at p is dened by putting, for any vector v = p(t
o
) T
p
Q tangent
to a smooth curve : t I p(t) Q at p = p(t
o
) ,
d
p
V [ v) :=
d
dt
(V )
to
15
Unlike TQ Tc = c E , cotangent bundle T
Q is not naturally embedded in

some Euclidean space related to c , and therefore its structure of smooth manifold is to be
meant in the generalized sense explained in the last subsection Intrinsic geometry of smooth
manifolds of this Appendix.
16
V is said to be smooth, if in any chart : W | it admits a C
coordinate
expression
V
:= V : q W V ((q)) R
60 5 Appendix
and then, by suitably restricting to the coordinate domain of a chart where
p = (q) and v = d
q
(v) ,
d
p
V [ v) =
d
dt
(V

1
)
to
=
V
q
h
q(to)
q
h
(t
o
)
=
V
q
h
q
v
h
=
V
q
h
q
d
p
q
h
[ v)
=
_
V
q
h
q
d
p
q
h
[ v
_
that is to say,
d
p
V =
V
q
h
p
d
p
q
h
with
V
q
h
p
:=
V
q
h
q
The above considerations can all be extended to real-valued functions de-
ned on open subsets of Q (check, for instance, that d
p
q
k
is the dierential at
p of the k-th projection q
k
: | R of
1
).
More generally, a dierential 1-form on Q is dened as a smooth covector
eld
f : Q T
Q : p (p, f(p))
Its value f(p) T
p
Q is expressed in a chart at p = (q) by
f(p) = f
h
(q) d
p
q
h
in terms of components
17
f
h
(q) :=
_
f(p) [
p
q
h
q
_
17
f is said to be smooth, if in any chart : W | it admits C
components
f
h
: q W f
h
(q) R
5 Appendix 61
Semi-basic dierential 1-forms
The concept of 1-form furtherly generalizes into that of semi-basic 1-form on
TQ, dened as a smooth bundle morphism
F : TQ T
Q : (p, v) (p, F(p, v))

Its value F(p, v) T
p
Q is expressed in a chart where p = (q) and
v = d
q
(v) by
F(p, v) = F
h
(q, v) d
p
q
h
18
F
h
(q, v) :=
_
F(p, v) [
p
q
h
q
_
The concept of semi-basic 1-form on T
2
Q is dened in the same way, as a
smooth bundle morphism
: T
2
Q T
Q : (p, v; v, w) (p, (p, v, w))

Its value (p, v, w) T
p
Q is expressed in a chart where p = (q) ,
v = d
q
(v) and w = z(q, v) +d
q
(w) by
(p, v, w) =
h
(q, v, w) d
p
q
h
19
h
(q, v, w) =
_
(p, v, w) [
p
q
h
q
_
Special kinds of semibasic 1-forms will now be described.
18
F is said to be smooth, if in any chart : W | it admits C
components
F
h
: (q, v) W R
n
F
h
(q, v) R
19
is said to be smooth, if in any chart : W | it admits C
components
h
: (q, v, w) W R
n
R
n

h
(q, v, w) R
62 5 Appendix
Semi-Riemannian metric
A semi-Riemannian metric on an n-dimensional manifold Q is a symmetric
vector bundle isomorphism of TQ onto T
Q, that is, a semi-basic 1-form

g : TQ T
Q : (p, v) (p, g
p
(v))
such that, for each p Q, its restriction
g
p
: T
p
Q T
p
Q : v g
p
(v)
is a linear isomorphism (non-degenerateness) and the bilinear inner product
g
p
(u) [ v) of any two vectors u, v T
p
Q satises the symmetry condition
(commutativity)
g
p
(u) [ v) = g
p
(v) [ u)
If non-degenerateness is strengthened by requiring positive deniteness, i.e.
g
p
(v) [ v) > 0 for all (p, v) TQ with v ,= 0, g is said to be a Riemannian
metric.
Owing to polarization identity
g
p
(u) [ v) =
1
2
_
g
p
(u +v) [ u +v) g
p
(u) [ u) g
p
(v) [ v)
_
(for all p Q and u, v T
p
Q), g is characterized by its quadratic form
K : TQ R : (p, v) K(p, v) :=
1
2
g
p
(v) [ v)
A manifold equipped with a (semi-)Riemannian metric will then be denoted
by (Q, K) and called (semi -)Riemannian manifold.
Remark that, any semi-spray
K
: TQ T
2
Q : (p, v) (p, v; v,
K
(p, v))
associated with a (semi-)Riemannian manifold (Q, K) , naturally transforms
g : TQ T
Q : (p, v) (p, g
p
(v))
(semi-basic 1-form on TQ) into
[K] : T
2
Q T
Q : (p, v; v, w) (p, [K](p, v, w))

(semi-basic 1-form on T
2
Q) by putting
[K](p, v, w) := g
p
(w
K
(p, v)) T
p
Q
5 Appendix 63
In a chart at p = (q) , the inner product g
p
(u) [ v ) is expressed
(owing to bilinearity) by
g
p
(u) [ v ) =
_
g
(q)
_
u
h
p
q
h
q
_
[ v
k
p
q
k
q
_
= u
h
_
g
(q)
_
p
q
h
q
_
[
p
q
k
q
_
v
k
and therefore g
p
is characterized by the (non-singular, symmetric) nn matrix
of components
g
hk
(q) :=
_
g
(q)
_
p
q
h
q
_
[
p
q
k
q
_
Almost-symplectic structure
An almost-symplectic structure on a 2n-dimensional manifold M is a skew
symmetric vector bundle isomorphism of TM onto T
M , that is, a semi-

basic 1-form
: TM T
M : (, X) (,
(X))
such that, for each M , its restriction
: T
M T
M : X
(X)
is a linear isomorphism (non-degenerateness) and the bilinear product
(X) [
Y ) of any two vectors X, Y T
M satises the skew-symmetry condition

(anticommutativity)
(X) [ Y ) =
(Y ) [ X)
is said to be a symplectic structure, if it is characterized in a suit-
able atlas of charts of M by the (non-singular, skew-symmetric) 2n 2n
symplectic matrix of components
_

(1)(1)

(1)(2)
(2)(1)

(2)(2)
_
with
(1)(1)
=
(2)(2)
= 0 (zero n n matrix) and
(1)(2)
=
(2)(1)
=
(identity n n matrix).
On a cotangent bundle M = T
Q there exists a canonical symplectic

structure, characterized by the symplectic matrix in the atlas of the natural
charts of M .
64 5 Appendix
5.5 Intrinsic approach to smooth manifods
A deeper insight into the structure of an embedded submanifold of a Eu-
clidean ane space, will provide useful suggestions for a generalized denition
of smooth manifold and an intrinsic approach to its geometry (without any
reference to embeddings into Euclidean environments).
Intrinsic geometry of embedded submanifolds
Let Q be an n-dimensional, smooth manifold embedded into a Euclidean
ane space c .
The n-dimensional atlas / made up of all the charts of Q, could be shown
to be
1. C
, i.e. any two charts

=
1
: | Q W R
n
,
=
1
: |
Q W
R
n
belonging to / are bijections (between subsets of Q and R
n
, respectively)
related to each other by C
transition functions
q (| |
)

1
p | |
(| |
)
q
(| |
)

1
p | |

q (| |
)
2. complete, i.e. any bijection
:

| Q

W R
n
which is C
-related to all the charts of /, belongs to /.

From the topological point of view, the domains of the charts belonging to
a complete, C
, n-dimensional atlas (on any set) turn out to be the base of a

locally Euclidean manifold topology, where all the charts are homeomorphisms
between open subsets (actually, the manifold topology determined by / on
Q coincides with the subspace topology of Q).
From the dierential point of view, for each p Q, the tangent vector
space T
p
Q can be identied through a canonical isomorphism with the
quotient vector space /
p
R
n
/
p
, where
/
p
/
5 Appendix 65
denotes the set of all the charts with domains containing p and
(, v)
p
(
, v
) i v
= d
(p)
(

1
) (v) ()
is the equivalence relation dened in /
p
R
n
by the transformation law for
the components of tangent vectors.
20
The vector space structure of the quotient is the invariant one that makes
an isomorphism of the bijection
21
p
: R
n
/
p
R
n
/
p
: v
p
(v) := [(, v)]
p
()
determined by any =
1
/
p
.
The above mentioned identication is due to the invariant isomorphism
p
: T
p
Q /
p
R
n
/
p
: v = d
(p)
(v)
p
(v) :=
p
(v)
In particular, for the basis of T
p
Q determined by a chart =
1
/
p
,
we put
22
p
q
h
(p)
= d
(p)
(
h
)
p

q
h
p
:=
p
(
h
)
As a consequence, for the vector p(t) tangent to a smooth curve : t
I (t) Q at p(t) = (t) , we have in any chart =
1
/
(t)
(where
q(t) = ((t)) )
p(t) = d
q(t)
( q(t))
p
(t) :=
(t)
( q(t))
=
(t)
( q
h
(t)
h
)
= q
h
(t)
(t)
(
h
)
= q
h
(t)

q
h
(t)
20
For any two charts =
1
and
=
1
in /
p
where p = (q) =
(q
) and any
tangent vector v = d
q
(v) = d
q

(v
) T
p
Q, we have
d
q

(v
) = d
q
(v) = d
q
(

1
) (v) = d
q

_
d
q
(

1
) (v)
_
whence, owing to the injectivity of d
q

,
v
= d
q
(

1
) (v)
21
[(, v)]
p
will denote the complete equivalence class of (, v) /
p
R
n
under
p
.
22
(
h
)
h=1,...,n
will denote the canonical basis of R
n
.
66 5 Appendix
Intrinsic geometry of smooth manifolds
On a non-empty set M , a structure of m-dimensional smooth manifold will
then be dened as a complete, C
, m-dimensional atlas /.
From the topological point of view, the manifold topology of M i.e. the
topology determined by the domains of the charts belonging to / is locally
Euclidean.
From the dierential point of view, the smoothness of M i.e. the dier-
entiability of the transition functions of / determines, at each M , an
m-dimensional tangent vector space given by
T
M := /
R
m
/
(where the equivalence relation
is dened as in () ).
Each chart (or coordinate system x = (x
i
)
i=1,...,m
) belonging to /
determines an isomorphism
: R
m
T
M (dened as in () ) and then a

basis in T
M given by
x
i
:=
(
i
)
Moreover, an invariant vector (t) tangent to a smooth curve : t I
(t) M at (t) is dened by putting in any chart /
(t)
(where the
coordinate expression x(t) = ((t)) is meant to be C
)
(t) :=
(t)
( x(t))
=
(t)
( x
i
(t)
i
)
= x
i
(t)
(t)
(
i
)
= x
i
(t)

x
i
(t)
From the above denitions, all of the further developments of the theory
(tangent bundles and dierential equations, cotangent bundle and dierential
forms) follow in quite a natural way.
67
References
[1] C. Godbillon, Geometrie Dierentielle et MecaniqueAnalytique, Hermann,
Paris (1969).
[2] R. Abraham and J.E. Marsden, Foundations of Mechanics, Benjamin/Cummings,
Mass. (1978).
[3] P. Libermann and C. Marle, Symplectic Geometry and Analytical Mechan-
ics, Reidel Publ., Dordrecht (1987).
[4] M. de Leon and P.R. Rodrigues, Methods of Dierential Geometry in An-
alytical Mechanics, North-Holland, Amsterdam (1989).
[5] M.J. Gotay, J.M. Nester and G. Hinds, Presymplectic Manifolds and the
Dirac-Bergmann Theory of Constraints, J. Math. Phys. 19 (1978), 2388-2399.
[6] M.J. Gotay and J.M. Nester Presymplectic Lagrangian Systems I and II,
Ann. Inst. Henri poincare A 30 (1979), 129-142 and 32 (1980), 1-13.
[7] F. Cantrijn, J.F. Carinena, M. Crampin and L.A. Ibort, Reduction of De-
generate Lagrangian systems, J. Geom. Phys. 3 (1986), 353-400.
[8] X. Gr`acia and J.M. Pons, A Generalized Geometric Framework for Con-
strained Systems, Di. Geom. Appl. 2 (1992) 223-247.
[9] M. de Leon and D. Martin de Diego, Symmetries and Constants of the
Motion for Singular Lagrangian Systems, Int. J. Theor. Phys. 35 (1996),
975-1011.
[10] G. Blankenstein and A.J. van der Schaft, Symmetry and Reduction in
Implicit Generalized Hamiltonian Systems, Rep. Math. Phys. 47 (2001), 57-
100.
[11] S. Kobayashi and K. Nomizu, Foundations of Dierential Geometry I,
Wiley Intersc, Publ., New York (1963).
[12] W.M. Boothby, Introduction to Dierentiable Manifolds and Riemannian
Geometry, Academic Press, New York (1975).
[13] F.W. Warner, Foundations of Dierentiable Manifolds and Lie Groups,
Springer-Verlag, New York (1983).
68
[14] J.M. Lee, Introduction to Smooth Manifolds, Springer-Verlag, New York
(2003).
[15] P.J. Rabier and W.C. Rheinboldt, A Geometric Treatment of Implicit
Dierential-Algebraic Equations, J. Di. Eqs 109 (1994),110-146.
[16] F. Barone and R. Grassini, Geometry of Implicit Dierential Equations
on Smooth Manifolds, Bull. Math.Soc. Sc. Math. Roumanie 47(95) No. 3-4
(2004), 119-145.
[17] F. Barone and R. Grassini, Generalized Dirac Equations in Implicit Form
for Constrained Mechanical Systems, Ann. Henri Poincare 3 (2002), 297-330.
[18] F. Barone and R. Grassini, A Note on Godbillon Dynamics and Generalized
Potentials, Ric. Mat. 51, 2 (2002), 265-284.
[19] F. Barone and R. Grassini, Generalized Hamiltonia Dynamics after Dirac
and Tulczyjew, Banach Center Publ. 59 (2003), 77-97.
[20] R. Grassini, Noether Symmetries and Conserved Momenta of Dirac Equa-
tion in Presymplectic Dynamics, Int. Math. Forum, 2, no.45 (2007), 2207-
2220.

Introduction To The Geometry of Classical Dynamics

Загружено:

Сведения о документе

Исходное описание:

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Introduction To The Geometry of Classical Dynamics

Загружено:

Авторское право:

Доступные форматы

Introduction to the

Geometry of Classical Dynamics

(p) = 0 ( = 1, . . . , ), called two-sided

(t)) and their accelera-

(t)) (as well as the derivatives of any order) vary

1.5 dAlemberts implicit equation

shows that determining

= 0 for all p Q (absence of constraint reaction) and then dAlemberts

, characterizing the DPMs.

) and then their tangent lifts, i.e. the integral

Extraction of the integrable part

10 2 From dAlembert to Lagrange

Q : (p, v; v, w) (p, [m](p, v, w))

Q : (p, v) (p, F(p, v))

by putting u : v E u v R. Then, for any

Riemannian geodesic curvature eld

Q : (p, v; v, w) (p, [K](p, v, w))

14 2 From dAlembert to Lagrange

(t) = (p(t), p(t))

16 2 From dAlembert to Lagrange

or, in vector formulation,

. Conversely, if satises condition ()

(which has already been seen to

, are the classical Lagrange equations of

. As lives in the coordinate domain of , equation ()

. As lives in the coordinate domain of , equation ()

24 2 From dAlembert to Lagrange

Q : (p, v; v, w) (p, [L](p, v, w))

Proposition 13 A smooth curve of Q is a base integral curve of D

Proof The same as the proof, part (i), of Prop.9.

2 From dAlembert to Lagrange 25

, are the classical Euler-Lagrange equations of Analytical

. As lives in the coordinate domain of , equation ()

26 2 From dAlembert to Lagrange

Q is a mapping which takes any (p, v; v, w) T

Q characterized, in a chart, by components

and its derivative

Q will denote the (invariant) covector-valued map whose components

is required to be stationary at , that is to say,

2 From dAlembert to Lagrange 31

Q is a 2n-dimensional manifold, where each element (p, ) is

Q will now be taken into considera-

Q, the canonical sym-

Q of momentum phase space.

Remark the second-order character of the above Hamilton equation, ex-

The integral curves and the base integral curves of D

- dierentiable, with injective dierential d

- dierentiable, i it is continuous and admits

- dierentiable, with injective dierential d

- dierentiable c-valued map

( < m := dimc ), with surjective dierential d

(t) = (p(t), p(t)) = (p(t), v(t)) = c(t)

The base integral curves =

= c , whence = c , and then Im T.

Q onto its base Q is the map

Q is a vector bre bundle over Q, since its bre p T

Q the structure of a 2n-dimensional smooth manifold.

Q is not naturally embedded in

Q : (p, v) (p, F(p, v))

Q : (p, v; v, w) (p, (p, v, w))

Q, that is, a semi-basic 1-form

Q : (p, v; v, w) (p, [K](p, v, w))