Univerrsity of Trento - Constrained Calculus of Variations and Geometric Optimal Control Theory

Faculty of Mathematical, Physical and Natural Sciences
Doctoral School in Mathematics - XXII Cicle

Philosophi Doctor Thesis
Constrained Calculus of Variations
and
Geometric Optimal Control Theory
Candidate:
Dr. Gianvittorio Luria
Advisors:
Prof. Enrico Massa
Prof. Enrico Pagani
To my mother, in her loving memory.
If every individual student follows the same current fashion
in expressing and thinking [..], then the variety of hypothesis
being generated [..] is limited. Perhaps rightly so, for possibly
the chance is high that the truth lies in the fashionable
direction. But, on the off-chance that it is in another
direction - - a direction obvious from an unfashionable view
of field-theory - - who will find it? Only someone who has
sacrificed himself by teaching himself [..] from a peculiar and
unusual point of view; one that he may have to invent for
himself. I say sacrificed himself because he most likely will get
nothing from it, because the truth may lie in another direction,
perhaps even the fashionable one.
But, if my own experience is any guide, the sacrifice is
really not great because if the peculiar viewpoint taken is truly
experimentally equivalent to the usual in the realm of the known
there is always a range of applications and problems in this realm
for which the special viewpoint gives one a special power and
clarity of thought, which is valuable in itself. Furthermore,
in the search for new laws, you always have the psychological
excitement of feeling that possible nobody has yet thought of
the crazy possibility you are looking at right now.
(Richard P. Feynman)
Le vritable voyage de dcouverte ne consiste pas
chercher de nouveaux paysages, mais avoir de nouveaux yeux.
(Marcel Proust)
Preface
The present work provides a fresh approach to the calculus of variations in the
presence of nonholonomic constraints.
The whole topic has been extensively studied since the beginning of the twen-
tieth century and has been recently revived by its close links with optimal control
theory. It is actually of great interest because of its several applications in a
wide range of elds such as Physics, Engineering [24] and Economics [12]. Among
others, we mention here the pioneering works of Bolza and Bliss [5], the contribu-
tion of Pontryagin [17] and the more recent developments by Sussman, Agrachev,
Hsu, Montgomery and Griths [35, 1, 27, 15, 9], characterized by a dierential
geometric approach.
Consider an abstract system B subject to a set of dierentiable conditions,
restricting the set of both its admissible congurations and velocities. We shall
tackle the following problem: how do we pick out among all the admissible evolu-
tions of B connecting two xed congurations, the ones (if any) that minimize a
given action functional?
In broaching the matter, we will make use of the tools provided by jetbundle
geometry, nonholonomic geometry and gauge theory. The abstract system B is
viewed as a dynamical system whose state can be specied by a nite number of
degrees of freedom. Denoted by 1
n+1
its conguration spacetime, having local
coordinates t, q
1
, . . . , q
n
, the admissible evolutions of B are then characterized by
the solutions of the parametric system of dierential equations
dq
i
dt
=
i
(t, q
1
, . . . , q
n
, z
1
, . . . , z
r
) , r n (1)
expressing the derivatives of the state variables in terms of a smaller number of
control variables.
Equations (1) are interpreted as the local representation of a set of kinetic con-
straints. More precisely, they are regarded as the local expression of the condition
under which an evolution is kinematically admissible. Geometrically, the request
is that the jetextension of must belong to a submanifold i : / j
1
(1
n+1
) which
describes the totality of admissible kinetic states. Given the system (1), by Cauchy
ii Preface.
theorem, every assignment of the functions z
A
(t) and of a point in 1
n+1
determines
an evolution of B as the solution of the given ordinary dierential equations with
the given initial conditions. However, in the absence of specic assumptions on
the nature of /, the functions z
A
(t), in themselves, have no invariant geometri-
cal meaning. To pursue the idea of the z
A
s as the controllers of the evolution,
attention should be rather shifted on sections : 1
n+1
/. Hence, every such
section is called a control .
Besides the constraints (1), it is also given an action functional
1[] :=
_
t
1
t
0
L(t, q
1
(t), . . . , q
n
(t), z
1
(t), . . . , z
r
(t)) dt (2)
expressed as the integral of a suitable cost function, or Lagrangian L(t, q, z)
along the admissible evolutions of the system. As stated above, our goal is to
nd, among these, the ones connecting the xed endpoints q
i
(t
0
), q
i
(t
1
) which
minimize the functional (2). Exactly as in ordinary function theory, the rst step
in the solution of the problem consists in investigating the stationarity conditions
for the action functional through the analysis of its rst variation.
The innitesimal deformations of an admissible section are discussed via a
revisitation of the familiar variational equation. The novelty of the approach relies
on the introduction of a transport law for vertical vector elds along , yielding a
covariant characterization of the true degrees of freedom.
The analysis is subsequently extended to arbitrary piecewise dierentiable evo-
lutions consisting of families of contiguous closed arcs
(s)
: [a
s1
, a
s
] 1
n+1
.
No restrictions are posed on the deformability of the intervals or on the mobility
of the corners (a
s
), s = 1, . . . , N 1.
The argument allows to assign to every admissible evolution a corresponding
abnormality index, rephrasing in a geometrical context the traditional attributes
of normality and abnormality commonly found in the literature [10].
Furthermore, the abnormality index of an evolution is seen to be related to its
ordinariness, that is to the property that every admissible innitesimal deforma-
tion vanishing at the endpoints is tangent to some nite deformations with xed
endpoints.
Within the stated framework, the search for the (local) stationary curves of 1
with respect to the admissible deformations leaving the endpoints xed results
in a fully covariant algorithm, summarizing the content of Pontryagins maximum
principle. The resulting equations are shown to provide sucient conditions for
any evolution, and necessary and sucient conditions for an ordinary evolution to
be an extremal.
A major breakthrough consists in the possibility of lifting the given constrained
variational problem to a corresponding free one in the contact bundle ((/) /,
Preface. iii
dened as the pullback of the dual space V

(1
n+1
) over /.
This solution method relies on the capability to establish a canonical corre-
spondence between the input data of the problem, namely the kinetic constraints
and the Lagrangian, and a distinguished 1form
PPC
in the contact manifold such
that every stationary curve of the variational problem based on it projects onto a
stationary curve of the corresponding problem in / related to the functional (2).
The canonical characterization of the form
PPC
called the Pontryagin
PoincareCartan form within the manifold ((/) is actually intimately con-
nected with the gauge structure of the whole theory: as it is well known, two dif-
ferent Lagrangians diering by a term
df
dt
, being f = f(t, q) any smooth function
over the conguration manifold, give rise to two equivalent variational problems.
In this sense, the real information isnt brought so much by the Lagrangian as by
the action functional.
In order to analyze the implications of this fact, keeping all dierences into
account, we take advantage of the geometrical setting introduced some years ago
for a gaugeinvariant formulation of Classical Mechanics [31, 32].
The construction is based on the introduction of a principal bre bundle over
the conguration spacetime 1
n+1
, with structural group (R, +), referred to as the
bundle of ane scalars. This is seen to induce two principal bundles L(1
n+1
) and
L
c
(1
n+1
) over the velocity space j
1
(1
n+1
), respectively called the Lagrangian and
coLagrangian bundle, as well as the further Hamiltonian and coHamiltonian
bundles over the phase space (1
n+1
) . In the presence of nonholonomic con-
straints, the Lagrangian bundles are easily adapted to the submanifold /, through
a straightforward pullback procedure.
Gaugeequivalent Lagrangians are then naturally interpreted as dierent rep-
resentations of one and the same section : j
1
(1
n+1
) L(1
n+1
) of the Lagrangian
bundle, dened up to an action of the gauge group.
A crucial role in the construction of the canonical PontryaginPoincareCartan
form over the contact manifold ((/) is then seen to be played by the locus of zeroes
of a distinguished pairing in the product manifold L(1
n+1
)
V
n+1
H(1
n+1
).
In the resulting scheme, a gaugeindependent free variational problem over
((/) is proved to be equivalent to the original constrained one.
The last part of the present work is devoted to establishing whether a given
piecewise dierentiable extremal , which is supposed to be normal even on closed
subintervals, gives rise to a minimum for the action functional (2).
The issue is worked out analyzing the socalled second variation of 1 . Actually,
the subject proves to be much harder than one could ever expect. First of all, the
expression in local coordinates of the second variation evidently involves the second
derivatives of the Lagrangian function, evaluated along the extremal curve. These
iv Preface.
last are easily seen to undergo a nontensorial transformation law whenever the
rst derivatives of L dont vanish along . This, of course, represents an actual
obstruction to a geometric approach. Apparently, the natural way out should
consist in making use of the gauge structure of the theory, by means of which it is
possible to replace the original Lagrangian by an equivalent one, characterized by
its being critical along the curve.
However, this adaptation method looks beforehand to be strictly connected
with the time intervals over which the arcs constituting the evolution are indi-
vidually dened. Therefore, it unavoidably fails whenever the deformation process
varies such intervals.
The combination of both the request for the tensorial nature of all results and
the will to deal with piecewise dierentiable curves made up of closed arcs whose
reference intervals are possibly changed by the deformation process is thus the
cause of much trouble.
Even so, it is actually possible to get over this stando by resorting to a family
of local gauge transformations instead of a single global one. Pursuing this strategy
enables to get a plainly covariant expression of the second variation in terms of a
quadratic form made up of an integral part and an algebraic one, related only to
the jumps of the curve.
It is now possible to break up the remaining part of the problem into consec-
utive logical steps. First of all, each single closed arc constituting the evolution is
requested to give rise to a minimum with respect to the special class of deforma-
tions which leave its own endpoints xed. This involves uniquely the behaviour
of the integral part of the quadratic form.
Focussing attention on a single arc, well rst prove a sucient condition for
minimality. This will turn out to be intimately related to the solvability of a
nonlinear dierential equation throughout the denition interval of the arc itself.
In the second instance, Jacobi vector elds are taken into account. They rep-
resent a special class of innitesimal deformations such that each of them links
families of extremal curves. They are used to investigate the processes of focaliza-
tion and, by means of the further concept of conjugate point, to give a necessary
condition for minimality.
Both the sucient and necessary conditions are eventually glued together,
showing that the lack of conjugate points along the arc implies the solvability
of the above nonlinear dierential equation on the whole of it.
At this point, it only remains to establish how the previous results can be
converted into a global one, applicable to the whole evolution.
We will show how this can be done by investigating the deniteness property of
the second variation restricted to the innitesimal deformations vanishing at the
corners and of a further quadratic form, dened on a suited quotient space.
Contents
1 Geometric setup 1
1.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Nonholonomic constraints . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Fibre bundles along sections . . . . . . . . . . . . . . . . . . . . . . 9
1.4 The gauge setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4.1 The Lagrangian bundles . . . . . . . . . . . . . . . . . . . . 13
1.4.2 The nonholonomic Lagrangian bundles . . . . . . . . . . . 16
1.4.3 The Hamiltonian bundles . . . . . . . . . . . . . . . . . . . 18
1.4.4 Further developments . . . . . . . . . . . . . . . . . . . . . 20
1.5 The variational setup . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5.1 Deformations . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.5.2 Innitesimal controls . . . . . . . . . . . . . . . . . . . . . . 25
1.5.3 Corners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.5.4 The abnormality index . . . . . . . . . . . . . . . . . . . . . 35
2 The rst variation 39
2.1 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2 The PontryaginPoincareCartan form . . . . . . . . . . . . . . . . 42
2.3 The Pontryagins maximum principle . . . . . . . . . . . . . . . . 43
2.4 Hamiltonian formulation . . . . . . . . . . . . . . . . . . . . . . . . 54
3 The second variation 59
3.1 Adapted Lagrangians . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.2 The second variation of the action functional . . . . . . . . . . . . 69
3.3 The associated singlearc problem . . . . . . . . . . . . . . . . . . 71
3.3.1 The matrix Riccati equation and the sucient conditions . 75
3.3.2 Jacobi elds . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.3.3 Conjugate points and the necessary conditions . . . . . . . 82
3.3.4 The necessary and sucient conditions . . . . . . . . . . . . 84
3.4 The induced quadratic form . . . . . . . . . . . . . . . . . . . . . . 87
A Adapted local charts 91
vi CONTENTS
B Finite deformations with xed endpoints: an existence theorem 95
C Admissible angular deformations 105
D A touch of theory of quadratic forms 109
Bibliography 113
Chapter 1
Geometric setup
1.1 Preliminaries
For the sake of convenience, we review here a few basic aspects of jetbundle
geometry [20, 29] which will play a major role in the subsequent discussion. The
terminology is borrowed from Mechanics
1
.
Let 1
n+1
t
R denote an (n+1)dimensional bre bundle, henceforth called
the event space and referred to local bred coordinates t, q
1
, . . . , q
n
. Every section
: R 1
n+1
, locally described as q
i
= q
i
(t), will be interpreted as an evolution
of an abstract system B, parameterized in terms of the independent variable t.
The rst jetspace j
1
(1
n+1
)

1
n+1
is then an ane bundle over 1
n+1
,
modelled on the vertical space V (1
n+1
) and called the velocity space. Both
spaces j
1
(1
n+1
) and V (1
n+1
) may be viewed as submanifolds of the tangent space
T(1
n+1
) according to the identications
2
j
1
(1
n+1
) = z T(1
n+1
) z, dt) = 1 (1.1.1a)
V (1
n+1
) = v T(1
n+1
) v, dt) = 0 (1.1.1b)
In view of equation (1.1.1a), every z j
1
(1
n+1
) determines a projection operator
T
z
: T
(z)
(1
n+1
) V
(z)
(1
n+1
), sending each vector X T
(z)
(1
n+1
) into the
vertical vector
T
z
(X) := X
X, (dt)
(z)
_
z (1.1.2)
Given any set of local coordinates t, q
1
, . . . , q
n
on 1
n+1
, the corresponding lo-
cal jetcoordinate system on j
1
(1
n+1
) is denoted by t, q
1
, . . . , q
n
, q
1
, . . . , q
n
, with
1
Although this is a natural choice, it may be somehow misleading. Just to avoid any possible
misunderstanding, it is therefore advisable to recall that, although formulated making use of
mechanical terms, constrained calculus of variations doesnt satisfy the principle of determinism
and, as such, it cant by no means be considered as belonging under Classical Mechanics.
2
Property (1.1.1a) is peculiar of those jetspaces which are built on bre bundles having a
1dimensional base space.
2 Chapter 1. Geometric setup
transformation laws
t = t +c , q
i
= q
i
(t, q
1
, . . . , q
n
) ,

q
i
=
q
i
t
+
q
i
q
k
q
k
(1.1.3)
The vertical bundle V (1
n+1
) is similarly referred to coordinates t, q
1
, .., q
n
, v
1
, .., v
n
.
In this way, the content of equations (1.1.1a,b) is summarized into the relations
z =
_

t
+ q
i
(z)

q
i
_
(z)
z j
1
(1
n+1
) (1.1.4a)
v = v
i
(v)
_

q
i
_
(v)
v V (1
n+1
) (1.1.4b)
while the projection operator (1.1.2) is expressed in coordinates as
T
z
_
X
0
_

t
_
(z)
+ X
i
_

q
i
_
(z)
_
=
_
X
i
X
0
q
i
(z)
_
_

q
i
_
(z)
=
=
_
X,
_
dq
i
q
i
(z)dt
_
(z)
_
_

q
i
_
(z)
(1.1.5)
By the very denition of jetbundle, every section : R 1
n+1
may be lifted
to a section j
1
(): R j
1
(1
n+1
), simply by assigning to each t R the tangent
vector to , namely
: q
i
= q
i
(t) j
1
() :
_
q
i
= q
i
(t)
q
i
=
dq
i
dt
(1.1.6)
The section j
1
() will be called the jetextension of on j
1
(1
n+1
). The annihilator
of the distribution tangent to the totality of the jetextensions of sections is a
subspace ((j
1
(1
n+1
)) of T
(j
1
(1
n+1
)), called the contact bundle. The tangent
space to the curve j
1
() T(j
1
(1
n+1
)) is spanned by the vector eld
(j
1
())
_

t
_
=

t
+
_
dq
i
dt
_

q
i
+
_
d q
i
dt
_

q
i
=

t
+ q
i

q
i
+
_
d
2
q
i
dt
2
_

q
i
The request for the curve j
1
() to pass through an arbitrarily chosen point z
in j
1
(1
n+1
) xes exclusively the values of the functions q
i
(t) and of their rst
derivatives but it doesnt aect the second derivatives
d
2
q
i
dt
2
. Therefore, a vector
Y T
z
(j
1
(1
n+1
)) is tangent to the jetextension of some section if and only if
it is represented in coordinate as
Y = Y
0
__

t
_
z
+ q
i
(z)
_

q
i
_
z
_
+ Y
i
_

q
i
_
z
Y
0
, Y
i
R (1.1.7)
1.1 Preliminaries 3
From this it is easily seen that the contact bundle is locally generated by the
1forms
i
= dq
i
q
i
dt (1.1.8)
Every section : j
1
(1
n+1
) ((j
1
(1
n+1
)) is called a contact 1form.
4
We now address ourselves to the vertical bundle
3
V (j
1
(1
n+1
))

j
1
(1
n+1
).
Given any jetcoordinate system t, q
i
, q
i
in j
1
(1
n+1
), we refer V (j
1
(1
n+1
)) to
bred coordinates t, q
i
, q
i
, v
i
according to the prescription
V V (j
1
(1
n+1
)) V = v
i
(V)
_

q
i
_
(V)
The ane character of the bration j
1
(1
n+1
) 1
n+1
provides a canonical iden-
tication of V (j
1
(1
n+1
)) with the pullback of V (1
n+1
) under the projection
: j
1
(1
n+1
) 1
n+1
, giving rise to the vector bundle homomorphism
V (j
1
(1
n+1
))

V (1
n+1
)
j
1
(1
n+1
)

1
n+1
(1.1.9)
For each z j
1
(1
n+1
), the bre
z
=
1
((z)) through z is actually an ane
submanifold of j
1
(1
n+1
), modelled on the vertical space V
(z)
(1
n+1
). Every pair
(z, v), v V
(z)
(1
n+1
) is therefore an applied vector at z in
z
, that is an
element of the tangent space T
z
(
z
). On the other hand, by denition, T
z
(
z
) is
canonically isomorphic to the vertical space V
z
(J
1
(1
n+1
)). By varying z, we con-
clude that the totality of pairs (z, v) j
1
(1
n+1
) V (1
n+1
) satisfying (z) = (v)
is in bijective correspondence with the points of V (j
1
(1
n+1
)), thereby establishing
diagram (1.1.9).
In bre coordinates, the representation of the map takes the simple form
_
V
i
_

q
i
_
z
_
= V
i
_

q
i
_
(z)
v
i
(V) = v
i
(V) V V (j
1
(1
n+1
))
(1.1.10)
3
Since j1(Vn+1) is bred on both Vn+1 and the real line R, there exist two vertical bre
bundles over j1(Vn+1). In the following, V (E; B) will stand for the bundle of vertical vectors
associated with the bration E B. Moreover, in order to make the notation as easy as possible,
the symbol V (j1(Vn+1)) will denote by a little abuse of language the vertical bundle with
respect to the bration j1(Vn+1) Vn+1.
In the same manner, every vertical vector v at (z) determines a correspond-
ing vertical vector v
v
at z, which is tangent to the curve z + v. The
correspondence v v
v
is known as vertical lift of vectors and is expressed in local
coordinates as
v = V
i
_

q
i
_
(z)
v
v
= V
i
_

q
i
_
z
(1.1.11)
4
On account of equation (1.1.1b), the dual space of V (1
n+1
) under the pairing
, ) : T(1
n+1
)
V
n+1
T
(1
n+1
) F(1
n+1
) is a vector bundle, henceforth denoted
by V
(1
n+1
), which is canonically isomorphic to the quotient of the cotangent
space T
(1
n+1
)

1
n+1
by the equivalence relation

_
() = (
dt
()
(1.1.12)
Every local coordinate system t, q
i
in 1
n+1
induces bred coordinates t, q
i
, p
i
in
V
(1
n+1
), with
p
i
( ) :=
_
,
_

q
i
_
( )
_
V
(1
n+1
)
and transformation laws
t = t +c , q
i
= q
i
(t, q
1
, . . . , q
n
) , p
i
= p
k
q
k
q
i
(1.1.13)
The pullback of V
(1
n+1
) through the map : j
1
(1
n+1
) 1
n+1
provides
another equivalent denition of the contact bundle ((j
1
(1
n+1
)). By this point of
view, a contact 1form is essentially a pair (z, ) j
1
(1
n+1
) V

(1
n+1
), with
V

(z)
(1
n+1
). Now, by equation (1.1.2), every such pair determines a linear
functional on the tangent space T
z
(j
1
(1
n+1
)) according to the prescription
, X) := , T
z
(X)) X T
z
(j
1
(1
n+1
)) (1.1.14)
In coordinates, recalling equation (1.1.5), the denition of p
i
( ) and making use
of the identication p
i
() = p
i
( ), equation (1.1.14) takes the explicit form
, X) =
_
,
_

q
i
_
(z)
_
i
| z
, X
_
=
p
i
( )
i
| z
, X
_
=
p
i
()
i
| ()
, X
_
1.2 Nonholonomic constraints 5
From this it easily seen that the knowledge of the functional (1.1.14) is mathemat-
ically equivalent to the knowledge of . Moreover, we nd again that the contact
bundle is identical to the vector subbundle of the cotangent space T
(j
1
(1
n+1
))
locally generated by the forms (1.1.8), while the coordinates p
i
coincide with the
components involved in the representation
= p
i
()
i
| ()
((j
1
(1
n+1
)) (1.1.15)
The situation is conveniently summarized into the commutative diagram
((j
1
(1
n+1
))

V
(1
n+1
)
j
1
(1
n+1
)

1
n+1
(1.1.16)
Notice that, by construction, ((j
1
(1
n+1
)) is at the same time a vector bundle over
j
1
(1
n+1
) and an ane bundle over V
(1
n+1
).
At each z j
1
(1
n+1
) the duality between V
(z)
(1
n+1
) and V

(z)
(1
n+1
) de-
termines a bilinear pairing | ) : V
z
(j
1
(1
n+1
)) ((j
1
(1
n+1
)) R based on the
prescription
V | ) := (V), ()) V V
z
(j
1
(1
n+1
)), ((j
1
(1
n+1
)) (1.1.17)
In coordinates, setting V = v
i
(V)
_

q
i
_
z
, = p
i
()
i
| z
, equations (1.1.10),
(1.1.17) yield the expression
V | ) = v
i
(V)
__

q
i
_
z
, ()
_
= v
i
(V) p
i
() (1.1.18)
By varying z, we extend it to a bilinear pairing between vertical vectors and
contact 1forms on j
1
(1
n+1
), fullling the duality relations
_

q
i
_
_
_
_
j
_
=
i
j
(1.1.19)
1.2 Nonholonomic constraints
Let / denote an embedded submanifold of j
1
(1
n+1
), bred over 1
n+1
. The situ-
ation, summarized into the commutative diagram
/
i
j
1
(1
n+1
)
1
n+1
1
n+1
(1.2.1)
provides the natural setting for the study of nonholonomic constraints.
The manifold / is referred to local bred coordinates t, q
1
, . . . , q
n
, z
1
, . . . , z
r
with transformation laws
t = t +c , q
i
= q
i
(t, q
1
, . . . , q
n
) , z
A
= z
A
(t, q
1
, . . . , q
n
, z
1
, . . . , z
r
) (1.2.2)
while the imbedding i : / j
1
(1
n+1
) is locally expressed as
q
i
=
i
(t, q
1
, . . . , q
n
, z
1
, . . . , z
r
) i = 1, . . . , n (1.2.3)
with rank
_
_
_
(
1

n
)
(z
1
z
r
)
_
_
_ = r. Alternatively, one may adopt an implicit representa-
tion
g
_
t, q
1
, . . . , q
n
, q
1
, . . . , q
n
_
= 0 = 1, . . . , n r (1.2.4)
with rank
_
_
_
(g
1
g
nr
)
( q
1
q
n
)
_
_
_ = n r. For simplicity, in the following we shall not
distinguish between the manifold / and its image i(/) j
1
(1
n+1
).
A section : R 1
n+1
will be called /admissible (admissible for short) if
and only if its rst jetextension is contained in /, namely if there exists a section
: R / satisfying j
1
( ) = i . With this notation, given any section
described in coordinates as q
i
= q
i
(t) , z
A
= z
A
(t) , the admissibility requirement
takes the explicit form
dq
i
dt
=
i
(t, q
1
(t), . . . , q
n
(t), z
1
(t), . . . , z
r
(t)) (1.2.5)
Equations (1.2.5) indicates that, for any admissible evolution of the system,
the knowledge of the functions z
A
(t) determines q
i
(t) up to initial data. On the
other hand, in the absence of specic assumptions on the nature of the manifold
/, the functions z
A
(t), in themselves, have no invariant geometrical meaning.
To pursue the idea of the z
A
s as the controllers of the the evolution of the
system, attention should rather be shifted on sections : 1
n+1
/. Henceforth,
every such section will be called a control for the system; the composite map
i : 1
n+1
j
1
(1
n+1
) will be called an admissible velocity eld.
In local coordinates we have the representations
: z
A
= z
A
(t, q
1
, . . . , q
n
) (1.2.6a)
i : q
i
=
i
(t, q
1
, . . . , q
n
, z
A
(t, q
1
, . . . , q
n
)) (1.2.6b)
conrming that the knowledge of does actually determine the evolution of the
system from any given initial event in 1
n+1
, through a well posed Cauchy problem.
A section : R 1
n+1
and a control : 1
n+1
/ will be said to belong to
each other if and only if the lift : R / factors into = , i.e. if and only if
the jetextension j
1
() coincides with the composite map i : R j
1
(1
n+1
).
1.2 Nonholonomic constraints 7
4
The concepts of vertical vector and contact 1form are easily extended to the
submanifold /: as usual, the vertical bundle V (/) is the kernel of the push
forward
: T(/) T(1
n+1
) while the contact bundle ((/) is the pullback on
/ of the bundle ((j
1
(1
n+1
)), as expressed by the commutative diagram
((/)

((j
1
(1
n+1
))
/
i
j
1
(1
n+1
)
(1.2.7)
The manifolds V (/) and ((/) will be referred to local coordinates t, q
1
, . . . , q
n
,
z
1
, . . . , z
r
, w
1
, . . . , w
r
and t, q
1
, . . . , q
n
, z
1
, . . . , z
r
, p
1
, . . . , p
n
respectively.
In this way, setting

i
:= i
(
i
) = dq
i
i
(t, q
1
, . . . , q
n
, z
1
, . . . , z
r
) dt (1.2.8)
we have the representations
X V (/) X = w
A
(X)
_

z
A
_
(X)
(1.2.9a)
((/) = p
i
()
i
(1.2.9b)
The restriction to V (/) of the pushforward i
: T(/) T(j
1
(1
n+1
)) determines
a vector bundle homomorphism
V (/)
i
V (j
1
(1
n+1
))
/
i
j
1
(1
n+1
)
(1.2.10)
Composing the last one with diagram (1.1.9) and introducing the simplied nota-
tion := i
, we get a homomorphism
V (/)

V (1
n+1
)
/

1
n+1
(1.2.11)
In coordinates, the previous argument provides the representation

_
V
A
_

z
A
_
z
_
=
_
V
A
_
i
z
A
_
z
_

q
i
_
i(z)
_
= V
A
_
i
z
A
_
z
_

q
i
_
(z)
written more synthetically as
v
i
(V ) =
_
i
z
A
_
z(V )
w
A
(V ) (1.2.12)
In a similar way, composing diagrams (1.1.16) and (1.2.7) and setting := ,
we get a bundle morphism
((/)

V
(1
n+1
)
/

1
n+1
(1.2.13)
described in coordinates as
t ( ()) = t() , q
i
( ()) = q
i
() , p
i
( ()) = p
i
()
The latter allows to regard the contact bundle ((/) as a bre bundle over the space
V
(1
n+1
), identical to the pull-back of V
(1
n+1
) through the map /

1
n+1
.
At each z /, diagrams (1.2.11), (1.2.13) determine a bilinear pairing between
V
z
(/) and (
z
(/), essentially identical to the restriction of the pairing (1.1.17),
based on the prescriptions
V | ) := (V ), ()) = p
i
()
_
i
z
A
_
z
w
A
(V ) (1.2.14)
Once again, by varying z, we get a bilinear pairing between vertical vectors and
contact 1forms satisfying the relations
_

z
A
_
_
_
_

i
_
z
=
_
i
z
A
_
z
z / (1.2.15)
It should not pass unnoticed that, unlike the original pairing (1.1.17), the map
V (/)
A
((/) F(/), based on equation (1.2.15), has now a singular char-
acter. A simple dimensionality argument actually shows that no duality can be
established between the spaces V (/) and ((/), it being selfevident that any
contact 1form =
i

i
fullling
i
_
i
z
A
_
()
= 0, A = 1, . . . , r annihilates
all vertical vectors. The totality of these 1forms generates a vector subbundle
(/) ((/), called the Chetaev bundle [30]. Every element (/) is called a
Chetaev 1form on /.
At last, it is worth remarking the presence on ((/) of a distinguished 1form
L
, called the Liovulle 1form, dened by the relation
X,
L|
_
=
(X), ) ((/), X T
(((/)) (1.2.16)
1.3 Fibre bundles along sections 9
and expressed in coordinates as
L
= p
i

i
= p
i
(dq
i
i
dt) (1.2.17)
1.3 Fibre bundles along sections
Let us now see how the geometric setup developed so far looks like when restricted
to a given section. The argument will play an important role in the variational
context as it provides a suitable framework for dealing with deformations.
The pullback over the section of the vertical space V (1
n+1
) determines a
vector bundle V ()
t
R, called the vertical bundle over . Given any local
coordinate system t, q
i
in 1
n+1
, we shall refer V () to bred coordinates t, v
i
according to the representation
X V () X = v
i
(X)
_

q
i
_
(t(X))
(1.3.1)
Likewise, the dual bundle V
()
t
R is identical to the pullback on of
the space V
(1
n+1
) . With the notation of 1.1, the situation is expressed by the
commutative diagram
V
() V
(1
n+1
)
t
R

1
n+1
(1.3.2)
The elements of V
() will be called the virtual 1forms along .

More generally, every element belonging to a bred tensor product of the form
V ()
R
V
()
R
will be called a virtual tensor along .
Notice that, according to the stated denition, a virtual 1form

at a point
(t) is not a 1form in the ordinary sense, but an equivalence class of 1forms
under the relation
(dt)
(t)
(1.3.3)
For simplicity, we preserve the notation , ) for the pairing between V () and
V
(). Also, given any local coordinate system t, q

i
in 1
n+1
, we refer V
() to
ber coordinates t, p
i
, with p
i
(
) =
,
_

q
i
_
(t(
))
_
.
The virtual 1forms along determined by the dierentials dq
i
will be denoted
by
i
, i = 1, . . . , n. In this way, every section W : R V ()
R
V

()
R
is
locally expressed as
W = W
i
.....
j
(t)
_

q
i
_

j
(1.3.4)
We remark that, according to diagram (1.3.2), each ber V
()
|t
is isomorphic
to the subspace of the cotangent space T
(t)
(1
n+1
) annihilating the tangent vector
to the curve at the point (t). This had to be expected as it was implicit in
the two equivalent denitions of the contact bundle we stated early. Formally, this
viewpoint is implemented by setting
i
=
_
dq
i
dq
i
dt
dt
_
. Although apparently
simpler, this characterization of V
() has some drawbacks in the case of piecewise

dierentiable sections and so we shall preferably stick to the original denition.
4
We recall from 1.1 that every section : R 1
n+1
admits a jetextension
j
1
(): R j
1
(1
n+1
), expressed in coordinates as q
i
= q
i
(t), q
i
=
dq
i
dt
. In a
similar way, every vertical vector eld X = X
i
q
i
over 1
n+1
may be lifted to a
eld J(X) = X
i
q
i
+
_
X
i
t
+
X
i
q
k
q
k
_

q
i
:= X
i
q
i
+

X
i
q
i
over j
1
(1
n+1
). The
argument is entirely standard (see, for instance, [20]) and is based on the following
construction:
the local 1parameter group of dieomorphisms
: 1
n+1
1
n+1
generated
by X induces, by pushforward, a one parameter group of dieomorphisms
(
: T(1
n+1
) T(1
n+1
)
the innitesimal generator of (
is a vector eld T(X) over T(1

n+1
)
the eld T(X) is tangent to the submanifold j
1
(1
n+1
) T(1
n+1
) locally
described by the equation

t = 1. As such, it denes a vector eld J(X) over
j
1
(1
n+1
)
Proposition 1.3.1. The rst jet space j
1
(V ()) is canonically isomorphic to the
vector bundle over R formed by the totality of vectors Z along j
1
() annihilat-
ing the 1form dt. With this identication, the bration
: j
1
(V ()) V ()
coincides with the restriction to j
1
(V ()) of the pushforward of the projection
: j
1
(1
n+1
) 1
n+1
.
Proof. Fix any t
R and a section X: R V (), then choose any vector eld

Y dened in a neighborhood U (t
) and such that Y

| (t)
= X(t) t
1
(U).
1.3 Fibre bundles along sections 11
In coordinates, setting : q
i
= q
i
(t), X = X
i
_

q
i
_
, the lift of the eld Y at the

point j
1
()(t
) takes the form

J(Y )
| j
1
()(t
)
= X
i
(t
)
_

q
i
_
j
1
()(t
)
+
dX
i
dt
t=t
_

q
i
_
j
1
()(t
)
(1.3.5)
Its now an easy matter to verify that all assertion of Proposition 1.3.1 follow as a
direct result of equation (1.3.5).
Consistently with equation (1.3.5), given any section X: R V (), the jet
extension j
1
(X) will be called the lift of X to the curve j
1
(). In local coordinates,
we have the representation
j
1
(X) = X
i
_

q
i
_
j
1
()
+
dX
i
dt
_

q
i
_
j
1
()
(1.3.6)
Both manifolds j
1
(V ()) and V () have an obvious nature of vector bundles
over R. With respect to this structure, the map
: j
1
(V ()) V () is clearly
an homomorphism with kernel identical to the restriction of the vertical bundle
V (j
1
(1
n+1
)) to the curve j
1
(). We set ker(
) := V (j
1
()) and call it the vertical
subbundle of j
1
(V ()).
The manifold j
1
(V ()) will be referred to jetcoordinates t, v
i
, v
i
, based on
the identication
Z j
1
(V ()) Z = v
i
(Z)
_

q
i
_
j
1
()(t(Z))
+ v
i
(Z)
_

q
i
_
j
1
()(t(Z))
(1.3.7)
In terms of these, the jetextension of a section v
i
= v
i
(t) takes the standard form
v
i
= v
i
(t) , v
i
=
dv
i
dt
, while the projection
: j
1
(V ()) V () is described by
v
i
(
(Z)) = v
i
(Z) . In particular, the vertical subbundle V (j
1
()) coincides with
the submanifold of j
1
(V ()) locally described by the equation v
i
= 0, i = 1, . . . , n.
Corollary 1.3.0.1. The vector bundles V (j
1
())
t
R and V ()
t
R are
canonically isomorphic
Proof. As pointed out in 1.1, for each z j
1
(1
n+1
) the ane character of
the bration j
1
(1
n+1
)

1
n+1
determines an isomorphism between the vertical
spaces V
z
(j
1
(1
n+1
)) and V
(z)
(1
n+1
), expressed in coordinates as
_
V
i
_

q
i
_
z
_
= V
i
_

q
i
_
(z)
In particular, for z = j
1
()(t), our previous denitions imply the identications
(z) = (t) , V
(z)
(1
n+1
) = V ()
|t
, V
z
(j
1
(1
n+1
)) = V (j
1
())
|t
. By varying t, this
gives rise to a vector bundle isomorphism
V (j
1
())

V ()
t
_t
R R
(1.3.8)
expressed in coordinates as
_
V
i
_

q
i
_
j
1
()
_
= V
i
_

q
i
_
(1.3.9)
4
In the presence of nonholonomic constraints, given any admissible section ,
let A( )
t
R denote the vector bundle formed by the totality of vectors along
annihilating the 1form dt. On account of Proposition 1.3.1, the pushforward
i
: T(/) T(j
1
(1
n+1
)) gives rise to a bundle morphism
A( )
i
j
1
(V ())
V () V ()
(1.3.10)
making A( ) into a subbundle of j
1
(V ()) bred over V ().
Once again all arrows in diagram (1.3.10), regarded as maps between vector
bundles over R, have the nature of homomorphisms. The kernel of the projection
A( )

V (), clearly identical to the restriction of the vertical bundle V (/) to
the curve , will be denoted by V ( ), and will be called the vertical subbundle
along .
Every bred coordinate system t, q
i
, z
A
in / induces coordinates t, v
i
, w
A
in
A( ) according to the prescription
X = v
i
(

X)
_

q
i
_
(t(

X))
+ w
A
(

X)
_

z
A
_
(t(

X))

X A( ) (1.3.11)
In terms of these, and of the jetcoordinates t, v
i
, v
i
on j
1
(V ()) , the morphism
(1.3.10) is locally described by the system
t = t , v
i
= v
i
, v
i
=
_
i
q
k
_

v
k
+
_
i
z
A
_

w
A
(1.3.12)
1.4 The gauge setup 13
while the vertical subbundle V ( ) coincides with the slice v
i
= 0 in A( ).
For later use, let us nally observe that the morphism (1.3.10) maps V ( )
into the vertical subbundle V (j
1
()) j
1
(V ()) . Composing with the morphism
(1.3.8), and recalling the denition of the composite map := i
, this gives
rise to an injective homomorphism
V ( )

V ()
t
_t
R R
(1.3.13)
In coordinates, equations (1.3.9), (1.3.12) provide the representation

_
Y
A
_

z
A
_

_
= Y
A
_
i
z
A
_

_

q
i
_
(1.3.14)
1.4 The gauge setup
In the sequel we will take advantage of a geometrical setting that was initially
introduced almost ten years ago in order to develop a gaugeinvariant formulation
of Lagrangian Mechanics. Hence, for convenience of the reader, we outline here its
main features.
1.4.1 The Lagrangian bundles
Given any system subject to (smooth) positional constraints, we introduce a double
bration P

1
n+1
t
R, where:
i) 1
n+1
t
R is the conguration spacetime of the system;
ii) P

1
n+1
is a principal bre bundle with structural group (R, +) .
As a consequence of the stated denition, each bre P
x
:=
1
(x), x 1
n+1
is an ane 1space. The total space P is therefore a trivial bundle, dieomorphic
in a non canonical way to the Cartesian product 1
n+1
R, called the bundle of
ane scalars over 1
n+1
.
The action of (R, +) on P results into a 1parameter group of dieomorphisms
: P P , conventionally expressed through the additive notation
() := + R, P (1.4.1)
Every map u: P R satisfying the requirement
u( +) = u() +
is called a (global) trivialization of P . If u, u
is any pair of trivializations, the

dierence u u
is then (the pullback of) a function over 1

n+1
. Moreover, every
section : 1
n+1
P determines a trivialization u
F(P) and conversely, the

relation between and u
being expressed by the condition

= (()) + u
() P (1.4.2)
Therefore, once a (global) trivialization u: P R has been chosen, every sec-
tion : 1
n+1
P is completely characterized by the knowledge of the function
f =
(u) F(1
n+1
).
The assignment of u allows to lift every local coordinate system t, q
1
, . . . , q
n
over 1
n+1
to a corresponding bred one t, q
1
, . . . , q
n
, u over P . The most general
transformation between bred coordinates has the form
t = t +c , q
i
= q
i
(t, q
1
, . . . , q
n
) , u = u +f(t, q
1
. . . . , q
n
)
The action of the group (R, +) on the manifold P is expressed in bred coordinates
by the relations
t( +) = t() , q
i
( +) = q
i
() , u( +) = u() +
As a result, the generator of the group action (1.4.1), usually referred to as the
fundamental vector eld of P, is canonically identied with the eld

u
.
The (pullback of) absolute time function determines a bration P
t
R
whose associated rst jetspace will be indicated by j
1
(P, R)

P . As usual,
this will be referred to local jetcoordinates t, q
i
, u, q
i
, u subject to transformation
laws
t = t +c , q
i
= q
i
(t, q
1
, . . . , q
n
) , u = u +f(t, q
1
. . . . , q
n
) (1.4.3a)
q
i
=
q
i
q
k
q
k
+
q
i
t
,

u = u +
f
q
k
q
k
+
f
t
:= u +

f (1.4.3b)
The manifold j
1
(P, R) is naturally embedded into the tangent space T(P)
through the identication
j
1
(P, R) = z T(P) z, dt) = 1
expressed in local coordinate as
z j
1
(P, R) z =
_

t
+ q
i
(z)

q
i
+ u
i
(z)

u
_
(z)
(1.4.4)
Every section : R P may be lifted to a section : R j
1
(P, R) by assigning
to each t R the tangent vector to , namely
:
_
q
i
= q
i
(t)
u = u(t)
:
_
_
q
i
= q
i
(t)
u = u(t)
q
i
=
dq
i
dt
u
i
=
du
dt
(1.4.5)
In addition to the jet attributes, the space j
1
(P, R) inherits from P two dis-
tinguished actions of the group (R, +), related in a straightforward way to the
identication (1.4.4).
The rst one is simply the pushforward of the action (1.4.1), restricted to the
submanifold j
1
(P, R) T(P). In jetcoordinates, a comparison with equation
(1.4.4) provides the local representation
(
(z) =
_

t
+ q
i
(z)

q
i
+ u
i
(z)

u
_
(z)+
(1.4.6a)
expressed symbolically as
(
: (t, q
i
, u, q
i
, u
i
) (t, q
i
, u +, q
i
, u
i
) (1.4.6b)
The quotient of j
1
(P, R) by this action is a (2n+2)dimensional manifold, hence-
forth denoted by L(1
n+1
). As shown in [31], the quotient map makes j
1
(P, R)
into a principal bre bundle over L(1
n+1
), with structural group (R, +). Further-
more, equation (1.4.6b) shows that L(1
n+1
) is an ane bre bundle over 1
n+1
with local coordinates t, q
i
, q
i
, u.
The second action of (R, +) on j
1
(P, R) follows from the invariant character
of the eld

u
and is expressed in local coordinates by the addition
(z) := z +
_

u
_
(z)
=
_

t
+ q
i
(z)

q
i
+
_
u
i
(z) +
_

u
_
(z)
(1.4.7a)
summarized into the symbolic relation
: (t, q
i
, u, q
i
, u
i
) (t, q
i
, u, q
i
, u
i
+) (1.4.7b)
The quotient of j
1
(P, R) by this action is once again a (2n+2)dimensional man-
ifold, henceforth denoted by L
c
(1
n+1
). As before, equation (1.4.7b) points out
that L
c
(1
n+1
) is a bre bundle over P (as well as on 1
n+1
), with coordinates
t, q
i
, u, q
i
. The quotient map makes j
1
(P, R) L
c
(1
n+1
) into a principal bre
bundle, with structural group (R, +) and group action (1.4.7a).
The last step in the construction relies on the observation that the group actions
(1.4.6a), (1.4.7a) do commute. Each of them may be then used to induce a group
action on the quotient space generated by the other one. As illustrated in [31],
this makes both L(1
n+1
) and L
c
(1
n+1
) into principal bre bundles over a common
double quotient space, canonically dieomorphic to the velocity space j
1
(1
n+1
).
The situation is summarized into the commutative diagram
j
1
(P, R) L
c
(1
n+1
)
_
L(1
n+1
) j
1
(1
n+1
)
(1.4.8)
in which all arrows denote principal brations, with structural groups isomor-
phic to (R, +) and group actions obtained in a straightforward way from equa-
tions (1.4.6b), (1.4.7b). The principal bre bundles L(1
n+1
) j
1
(1
n+1
) and
L
c
(1
n+1
) j
1
(1
n+1
) are respectively called the Lagrangian and the coLagrangian
bundle over j
1
(1
n+1
).
The advantage of this framework is exploited to the utmost by giving up
the traditional approach, based on the interpretation of the Lagrangian function
L(t, q
i
, q
i
) as the representation of a (gaugedependent) scalar eld over j
1
(1
n+1
)
and introducing instead the concept of Lagrangian section, meant as a section
: j
1
(1
n+1
) L(1
n+1
) of the Lagrangian bundle.
For each choice of the trivialization u of P, the description of takes the local
form
u = L(t, q
i
, q
i
) (1.4.9)
and so it does still rely on the assignment of a function L(t, q
i
, q
i
) over j
1
(1
n+1
).
However, as soon as the trivialization is changed into u = u+f , the representation
(1.4.9) undergoes the transformation law
u = u +

f = L(t, q
i
, q
i
) +

f := L
(t, q
i
, q
i
) (1.4.10)
involving a dierent, gaugeequivalent Lagrangian.
1.4.2 The nonholonomic Lagrangian bundles
Let us return to diagram (1.2.1), with the base manifold explicitly identied with
the conguration spacetime 1
n+1
of an abstract system B and with the imbed-
ding i : / j
1
(1
n+1
) taken as a description of the kinetic constraints acting on
it [28, 30]. The construction of the Lagrangian bundles is easily adapted to the
submanifold /, through a straightforward pull-back process.
The situation is conveniently illustrated by means of a commutative diagram
-
-
-
-
? ?
? ?
*
L(/) /
L
c
(/) j
A
1
(P, R)
j
1
(1
n+1
)
L
c
(1
n+1
) j
1
(P, R)
L(1
n+1
)
(1.4.11)
where:
L(/) and L
c
(/) are respectively the pullback of L(1
n+1
) and L
c
(1
n+1
) on
the submanifold / j
1
(1
n+1
);
j
A
1
(P, R) may be alternatively seen as the pullback of j
1
(P, R) L(1
n+1
)
on the submanifold L(/) L(1
n+1
) or as the pullback of
j
1
(P, R) L
c
(1
n+1
) on L
c
(/) L
c
(1
n+1
).
As usual, we refer / to local bred coordinates t, q
1
, . . . , q
n
, z
1
, . . . , z
r
with
transformation laws
t = t +c , q
i
= q
i
(t, q
1
, . . . , q
n
) , z
A
= z
A
(t, q
1
, . . . , q
n
, z
1
, . . . , z
r
)
(1.4.12)
and express the imbedding i : / j
1
(1
n+1
) in the form
q
i
=
i
(t, q
1
, . . . , q
n
, z
1
, . . . , z
r
) (1.4.13)
The geometrical properties of the abovedened pullback bundles are straight-
forwardly inherited from their respective holonomic counterparts. In particular:
Every choice of a trivialization u of P allows to lift any coordinate system of
/ to coordinates t, q
i
, z
A
, u on L
c
(/), t, q
i
, z
A
, u on L(/) and t, q
i
, u, z
A
, u
on j
A
1
(P, R). The resulting coordinate transformations are obtained by com-
pleting equations (1.4.12) with (the signicant part of) the system
u = u +f(t, q
i
) ,

u = u +
f
t
+
f
q
k

k
(t, q
i
, z
A
) := u +

f (1.4.14)
Equation (1.4.13) locally describes all the embeddings L(/) L(1
n+1
),
L
c
(/) L
c
(1
n+1
) and j
A
1
(P, R) j
1
(P, R).
Both actions (1.4.6a), (1.4.7a) of the group (R, +) on j
1
(P, R) preserve the
submanifold j
A
1
(P, R) thereby inducing two corresponding actions (
and
on j
A
1
(P, R), expressed in coordinate as
(
:
_
t, q
i
, u, z
A
, u
_

_
t, q
i
, u +, z
A
, u
_
(1.4.15a)
:
_
t, q
i
, u, z
A
, u
_

_
t, q
i
, u, z
A
, u +
_
(1.4.15b)
Acting in the same way as before, it is easily seen that the manifold j
A
1
(P, R)
is a principal bre bundle over L(/) under the action (
, as well as a
principal bre bundle over L
c
(/) under the action
. Moreover, both L(/)

and L
c
(/) are principal bre bundles over / under the (induced) actions
(
and
respectively. Accordingly, all arrows in the front and rear faces

of the diagram (1.4.11) express principal brations, while those in the left
and righthand faces are principal bundle homomorphisms.
Preserving the terminology, the principal bre bundles L(/) / and
L
c
(/) / will be respectively called the nonholonomic Lagrangian and
nonholonomic coLagrangian bundle over /. A section : / L(/) will
be called a (nonholonomic) Lagrangian section. Once a trivialization u of
P has been xed, any such section is locally expressed as
u = L(t, q
i
, z
A
) (1.4.16)
Under an arbitrary change u u+f of the trivialization, the representation
(1.4.16) undergoes the transformation law
u = u +

f = L(t, q
i
, z
A
) +
f
t
+
f
q
i

i
:= L
(t, q
i
, z
A
) (1.4.17)
1.4.3 The Hamiltonian bundles
Parallelling the discussion in 1.4.1, we shall now deal with the construction of the
Hamiltonian bundles on 1
n+1
. To this end, we focus on the bration P 1
n+1
,
and denote by : j
1
(P, 1
n+1
) P the associated rst jetspace.
Every bred coordinate system t, q
i
, u on P induces local coordinates t, q
i
, u, p
0
, p
i
on j
1
(P, 1
n+1
), with transformation group
t = t +c , q
i
= q
i
(t, q
1
, . . . , q
n
) , u = u +f(t, q
1
, . . . , q
n
) (1.4.18a)
p
0
= p
0
+
f
t
+
_
p
k
+
f
q
k
_
q
k
t
, p
i
=
_
p
k
+
f
q
k
_
q
k
q
i
(1.4.18b)
The manifold j
1
(P, 1
n+1
) is naturally imbedded into the cotangent space T
(P)
through the identication
j
1
(P, 1
n+1
) =
_
T
(P)
_
,

u
_
= 1
_
expressed in local coordinate as
j
1
(P, 1
n+1
) =
_
du p
0
()dt p
i
()dq
i
()
(1.4.19)
Furthermore, equations (1.4.18a,b) ensure the invariance of the contact 1form
= du p
0
dt p
i
dq
i
(1.4.20)
henceforth referred to as the Liouville 1form of j
1
(P, 1
n+1
).
Exactly as in the Lagrangian case, one can easily establish two distinguished
actions of the group (R, +) on j
1
(P, 1
n+1
), expressed locally as
(
() := (
() =
_
du p
0
()dt p
i
()dq
i
()+
(1.4.21a)
() := (dt)
()
=
_
du (p
0
() +) dt p
i
()dq
i
()
(1.4.21b)
Referring once again to [31] for the necessary details, we point out that:
The direct product of the actions (1.4.21a,b) makes j
1
(P, 1
n+1
) into a prin-
cipal bre bundle over a (2n + 1)dimensional base space (1
n+1
), with
coordinates t, q
i
, p
i
, called the phase space.
In view of equations (1.1.13), (1.4.18a,b), it is readily seen that the phase
space (1
n+1
) is an ane bundle over 1
n+1
, modelled on V
(1
n+1
).
The quotient of j
1
(P, 1
n+1
) by the action (1.4.21a), denoted by H(1
n+1
), is
an ane bundle over 1
n+1
, modelled on the cotangent space T
(1
n+1
) and
called the Hamiltonian bundle.
Any trivialization u: P R allows to lift every local coordinate system
t, q
1
, . . . , q
n
on 1
n+1
to a corresponding one t, q
1
, . . . , q
n
, p
0
, p
1
, . . . , p
n
on
H(1
n+1
), subject to the transformation law
p
0
= p
0
+
f
t
, p
i
= p
i
+
f
q
i
(1.4.22)
further to a change of u into u = u +f(t, q).
The quotient map makes j
1
(P, 1
n+1
) into a principal bre bundle over
H(1
n+1
), with structural group (R, +) and fundamental vector

u
.
The canonical 1form (1.4.20) endows j
1
(P, 1
n+1
) H(1
n+1
) with a dis-
tinguished connection, called the canonical connection. At the same time,
the action (1.4.21b) passes to the quotient, thereby making H(1
n+1
) into
a principal bre bundle over the phase space (1
n+1
).
The quotient of j
1
(P, 1
n+1
) by the action (1.4.21b), denoted by H
c
(1
n+1
), is
a (2n + 2)dimensional manifold, with coordinates t, q
i
, u, p
i
, called the co
Hamiltonian bundle. The quotient map makes j
1
(P, 1
n+1
) into a principal
bre bundle over H
c
(1
n+1
). At the same time, the action (1.4.21a), suitably
transferred to H
c
(1
n+1
), makes the latter into a principal bre bundle over
(1
n+1
) .
The previous discussion is summarized into the commutative diagram
j
1
(P, 1
n+1
) H
c
(1
n+1
)
_
H(1
n+1
) (1
n+1
)
(1.4.23)
in which all arrows denote principal brations, with structural group isomorphic
to R. As implicit in the notation, it may be easily showed that the manifold
j
1
(P, 1
n+1
) is indeed identical to the pullback of H
c
(1
n+1
) over H(1
n+1
), as well
as the pullback on H(1
n+1
) over H
c
(1
n+1
).
1.4.4 Further developments
The identications (1.4.4), (1.4.19) provide a natural pairing between the bres of
the rst jetspaces j
1
(P, R)

P and j
1
(P, 1
n+1
)

P , expressed in coordinate
as
z, ) =
_
_

t
+ q
i
(z)

q
i
+ u
i
(z)

u
_
(z)
,
_
du p
0
()dt p
i
()dq
i
()
_
(1.4.24)
for all z j
1
(P, R), j
1
(P, 1
n+1
) satisfying (z) = ().
In view of equations (1.4.6a), (1.4.21a), the correspondence (1.4.24) satises
the invariance property
_
(
(z) , (
()
_
=
z,
_
(1.4.25)
thereby inducing an analogous pairing operation between the bres of the bundles
L(1
n+1
) 1
n+1
and H(1
n+1
) 1
n+1
, or just the same giving rise to a
biane map of the bred product L(1
n+1
)
V
n+1
H(1
n+1
) onto R, expressed in
coordinates as
, F(, ) := u() p
0
() p
i
() q
i
() (1.4.26)
Let o denote the submanifold of L(1
n+1
)
V
n+1
H(1
n+1
) described by the equation
o =
_
(, ) L(1
n+1
)
V
n+1
H(1
n+1
) F(, ) = 0
_
(1.4.27)
A straightforward argument, based on equation (1.4.26), shows that the subman-
ifold o is at the same time a bre bundle over L(1
n+1
) and over H(1
n+1
). The
former case is made explicit by referring o to local coordinates t, q
i
, q
i
, u, p
i
, the
p
i
s been regarded as bre coordinates. The latter circumstance is similarly ac-
counted for by referring o to coordinates t, q
i
, q
i
, p
0
, p
i
, related to the previous
ones by the transformation
u = p
0
+ p
i
q
i
and with the q
i
s playing the role of bre coordinates.
Recalling the denition of the contact bundle ((j
1
(1
n+1
)), the situation is
summarized into the following commutative diagram
o
//
uuk
k
k
k
k
k
k
k
k
k
k
k
k
k
k
k
k
H(1
n+1
)
yys
s
s
s
s
s
s
s
s
s
L(1
n+1
)
//
1
n+1
L(1
n+1
)
V
n+1
H(1
n+1
)
//
uuk
k
k
k
k
k
k
k
k
k
k
k
k
k
H(1
n+1
)
yyt
t
t
t
t
t
t
t
t
t
L(1
n+1
)
//
1
n+1
((j
1
(1
n+1
))
//
uuk
k
k
k
k
k
k
k
k
k
k
k
k
k
V
(1
n+1
)
yys
s
s
s
s
s
s
s
s
s
j
1
(1
n+1
)
//
1
n+1
(1.4.28)
in which the nature of o as a principal bre bundle over ((j
1
(1
n+1
)) stands out.
Depending on the choice of the local coordinates over o, the group action may be
expressed symbolically either as
: (t, q
i
, q
i
, u
i
, p
i
) (t, q
i
, q
i
, u
i
+, p
i
) (1.4.29a)
or
: (t, q
i
, q
i
, p
0
, p
i
) (t, q
i
, q
i
, p
0
+, p
i
) (1.4.29b)
Furthermore, its worth pointing out that the canonical contact 1form (1.4.20)
of j
1
(P, 1
n+1
) can be pulledback onto the bred product j
1
(P, R)
P
j
1
(P, 1
n+1
).
The principal bre bundle j
1
(P, R)
P
j
1
(P, 1
n+1
) L(1
n+1
)
V
n+1
H(1
n+1
) is
consequently endowed with a canonical connection.
For every choice of the trivialization u of P 1
n+1
, the dierence du

is (the
pullback of) a 1form

u
on L(1
n+1
)
V
n+1
H(1
n+1
), locally expressed as
u
= p
0
dt +p
i
dq
i
(1.4.30)
and subject to the transformation law
u
=
_
p
0
+
f
t
_
dt +
_
p
i
+
f
q
i
_
dq
i
=

u
+df (1.4.31)
under an arbitrary transformation u u = u +f(t, q).
Eventually, the form

u
can be once again pulledback onto o. In this last
step, depending on the choice of the coordinates over o, the resulting 1form can
be locally expressed as
u
= p
0
dt +p
i
dq
i
udt +p
i
_
dq
i
q
i
dt
_
(1.4.32)
Hence, the submanifold o is provided with a distinguished 1form
u
which is
dened up to the choice of the trivialization of P .
1.5 The variational setup
1.5.1 Deformations
Given a section : R 1
n+1
, locally described as q
i
= q
i
(t), a nite deforma-
tion of is, by denition, a continuous map : R R 1
n+1
, dened
on the subset = (t, ) [ t
0
t t
1
, < < and satisfying the condition
(t, 0) = (t). By varying the parameter within its denition domain, we get a
1parameter family of sections
, satisfying
0
= .
Actually, it is usually made a distinction between the so called weak and strong
variations. In order to understand this dierence we need to introduce some topol-
ogy in the space of sections of 1
n+1
.
Denition 1.5.1. Let : (c, d) 1
n+1
be a dierentiable section, [a, b] (c, d)
be any closed interval and (U, h), h = (t, q
1
, . . . , q
n
) a corresponding bred local
chart such that (t) U for any t [a, b]. Let also and be a positive number
and a nonnegative integer respectively. Then ^
(,)
() is the set of all dieren-
tiable sections
: R 1
n+1
such that the following two conditions hold for any
t [a, b]:
1.5 The variational setup 23
1)
(t) U
2)
d
k
(q
i

(t))
dt
k

d
k
(q
i
(t))
dt
k
< k = 0, . . . ,
We let the reader verify that the sets ^
(,)
() form a system of neighborhoods
of for a topology on the space of sections of 1
n+1
. In particular, the topology
related to the sets ^
(,0)
() is called the strong topology while the one related to
the sets ^
(,1)
() is referred to as the weak topology.
By abuse of language, a deformation
is also said to be weak (or strong ) if,

for any > 0, there exists an > 0 such that
^
(,1)
() (or
^
(,0)
())
for any < . We point up that, as a consequence of the previous denitions, any
weak deformation is also always a strong one while the converse may not occur.
Example 1.5.1: In the onedimensional case, consider the variation
(t) : q(
(t)) = q(t) + sin

_
t
2
_
As goes to zero,
tends to by the squeeze rule. However, we have

dq(
(t))
dt
=
dq
dt
+
1
cos
_
t
2
_
and so
1
tends to innity while the cosine oscillates, generating increasingly large varia-
tions in the slope a typical strong, not weak, variation.
For each t R, the curve
(t) is called the orbit of the deformation
through the point (t). The vector eld along tangent to the orbits at = 0,
whenever dened, is called the innitesimal deformation associated with
.
4
In the presence of nonholonomic constraints, care must be taken of the re-
quirement of kinematical admissibility. A deformation
is called admissible if
and only if each section
: R 1
n+1
is admissible. In a similar way, a defor-
mation
of an admissible section : R / is called admissible if and only if

all sections
: R / are admissible.
As pointed out in 1.2, the admissible sections : R 1
n+1
are in 11
correspondence with the admissible sections : R / through the relations
= , j
1
() = i (1.5.1)
Every admissible deformation of may therefore be expressed as

: R / denoting an admissible deformation of .

In coordinates, preserving the representation : q
i
= q
i
(t) , z
A
= z
A
(t) , the
admissible deformations of are described by equations of the form

: q
i
=
i
(, t) , z
A
=
A
(, t) (1.5.2)
subject to the conditions
i
(0, t) = q
i
(t) ,
A
(0, t) = z
A
(t) (1.5.3a)
i
t
=
i
(t,
i
(, t),
A
(, t)) (1.5.3b)
We now dwell upon the fact that any (admissible) not weak nite deformation
: 1
n+1
can by no means be lifted to a corresponding (admissible) deforma-
tion : /, since the continuity of is lacking. On this account, from now
on we will restrict ourselves to consider weak variations only. Variational problems
with respect to strong variations can be dealt by means of a more general method,
based on the socalled Weierstrass Excess function. The argument is beyond the
purposes of the present work and will not be pursued.
Setting
X
i
(t) :=
_
_
=0
,
A
(t) :=
_
_
=0
Z
i
(t) :=
_
2
_
=0
, K
A
(t) :=
_
2
_
=0
(1.5.4)
the innitesimal deformation tangent to
is described by the vector eld
X = X
i
(t)
_

q
i
_

+
A
(t)
_

z
A
_

(1.5.5)
while equation (1.5.3b) is reected into the relations
dX
i
dt
=

t
=0
=
_
i
q
k
_

X
k
+
_
i
z
A
_

A
(1.5.6a)
dZ
i
dt
=

t
=0
=
_

2
i
q
k
q
r
_

X
k
X
r
+ 2
_

2
i
q
k
z
A
_

X
k
A
+
+
_

2
i
z
A
z
B
_

B
+
_
i
q
k
_

Z
k
+
_

i
z
A
_

K
A
(1.5.6b)
the rst of which is commonly referred to as the variational equation.
The innitesimal deformation tangent to the projection
is similarly
described by the eld
X =

X =
_
_
=0
_

q
i
_
= X
i
(t)

q
i
(1.5.7)
Collecting all previous results and recalling the denitions of the vector bundles
V () and A( ) we get the following
Proposition 1.5.1. Let : R 1
n+1
and : R / denote two admissible
sections, related by equation (1.5.1). Then:
i) the innitesimal deformations of and of are respectively expressed as
sections X : R V () and

X : R A( );
ii) a section X : R V () represents an admissible innitesimal deforma-
tion of if and only if its rst jetextension factors through A( ), i.e. if
and only if there exists a section

X : R A( ) satisfying j
1
(X) = i

X;
conversely, a section

X : R A( ) represents an admissible innitesimal
deformation of if and only if it projects into an admissible innitesimal
deformation of , i.e. if and only if i

X = j
1
(

X).
The proof is entirely straightforward, and is left to the reader.
From a structural viewpoint, Proposition 1.5.1 establishes a complete symmetry
between the roles of diagram (1.2.1) in the study of the admissible evolutions
and of diagram (1.3.10) in the study of the admissible innitesimal deformations,
thus enforcing the intuitive idea that the latter context is essentially a linearized
counterpart of the former one.
1.5.2 Innitesimal controls
According to Proposition 1.5.1, the admissible innitesimal deformations of an
admissible section : R 1
n+1
are in 11 correspondence with the sections
X : R A( ) satisfying the consistency requirement i

X = j
1
(

X).
In local coordinates, setting

X = X
i
(t)

q
i
+
A
(t)

z
A
, the stated requirement
is expressed by the variational equation
dX
i
dt
=

i
q
k
X
k
+

i
z
A

A
(1.5.8)
all coecients being evaluated along the curve .
Exactly as it happened in 1.2 with regard to the admissibility of evolutions,
equation (1.5.8) indicates that, for each admissible

X, the knowledge of the func-
tions
A
(t) determines the remaining X
i
(t) up to initial data, through the solution
of a well posed Cauchy problem.
Once again, however, the drawback is that the components
A
, in themselves,
have no invariant geometrical meaning, but obey the nonhomogeneous transfor-
mation law
A
=
_

X, d z
A
_
=
z
A
t
_

X, dt
_
+
z
A
q
i
_

X, dq
i
_
+
z
A
z
B
_

X, z
B
_
=
=
z
A
q
i
X
i
+
z
A
z
B

B
(1.5.9)
under an arbitrary coordinate transformation. Therefore, if is covered by several
local charts, assigning the functions
A
(t) on each of them doesnt even allow to
verify if they link up properly except by integrating the variational equation.
The diculty is overcome introducing a linearized version of the idea of control .
Referring to diagram (1.3.10), we thus state the following
Denition 1.5.2. Let : R 1
n+1
denote an admissible evolution. Then:
a linear section h : V () A( ), meant as a vector bundle homomorphism
satisfying
h = id, is called an innitesimal control along ;

the image H( ) := h(V ()), viewed as a vector subbundle of A( ) R,
is called the horizontal distribution along induced by h; every section
X : R A( ) satisfying

X(t) H( ) t R is called a horizontal section.
Remark 1.5.1: The term innitesimal control is intuitively clear: given an admissible
section , let : 1
n+1
/ denote any control belonging to , that is satisfying = .
Then, on account of the identity
= ( )
= id, the restriction to V () of

the tangent map
: T(1
n+1
) T(/) determines a linear section
: V () A( ).
The innitesimal controls may therefore be thought of as equivalence classes of ordinary
controls belonging to the same curve and having a rst order contact along it.
Given an innitesimal control h : V () A( ), on account of Denition 1.5.2
and of the canonicity of the vertical subbundle V ( ) = ker
, it is easily seen
that the horizontal distribution H( ) does indeed provide a splitting of the vector
bundle A( ) into the bred direct sum
A( ) = H( )
R
V ( ) (1.5.10)
This gives rise to a couple of homomorphisms T
H
: A( ) H( ) (horizontal
projection) and T
V
: A( ) V ( ) (vertical projection), uniquely dened by the
relations
T
H
= h
; T
V
= id T
H
(1.5.11)
In bre coordinates, preserving the notation (1.3.1), (1.3.11), every innitesimal
control h : V () A( ) is represented by a linear system of the form
w
A
= h
i
A
(t) v
i
(1.5.12)
In this way:
the horizontal distribution H( ) is locally spanned by the vector elds
i
:= h
_ _

q
i
_
_
=
_

q
i
_

+ h
i
A
_

z
A
_

(1.5.13)
every vertical vector eld X = X
i
(t)
_

q
i
_
along may be lifted to a

horizontal eld h(X) along , expressed in components as
h(X) = X
i
(t)

i
= X
i
(t)
_ _

q
i
_

+ h
i
A
_

z
A
_

_
(1.5.14)
every vector

X = X
i
_

q
i
_

+
A
_

z
A
_

A( ) admits a unique represen-
tation of the form

X = T
H
(

X) +T
V
(

X) , with
T
H
(

X) = X
i

i
, T
V
(

X) =
_
A
X
i
h
i
A
_
_

z
A
_

:= Y
A
_

z
A
_

(1.5.15)
denoting by (( ) the contact bundle along , meant as the restriction of the
contact bundle ((/) to the curve , in view of equation (1.5.13) we have
the relations
_
i
,
k
_
=
__

q
i
_

+ h
i
A
_

z
A
_

,
_
dq
k

k
dt
_

_
=
k
i
(1.5.16)
showing that the bundle (( ) and H( ) are dual of each other under ordinary
pairing.
Remark 1.5.2: As implicit in the previous discussion, the advantage of the newer formu-
lation comes from the fact that, unlike
A
_

z
A
_

, the vector eld T
V
(

X) = Y
A
_

z
A
_

has now an invariant geometrical meaning.
It follows from the request

i
=
q
j
q
i
j
that
q
i
+

h
A
i
z
A
=
q
j
q
i
_

q
j
+h
B
j
z
B
_

z
B
q
i
z
B
+

h
A
i
z
B
z
A
z
B
=
q
j
q
i
h
B
j
z
B
that is
h
A
i
=
z
A
z
B
q
j
q
i
h
B
j

z
A
z
B
z
B
q
i
(1.5.17)
Therefore, the components Y
A
undergo the homogeneous transformation law
Y
A
=

X
i

h
A
i
= X
j
z
A
q
j
+
B
z
A
z
B
X
j
q
i
q
j
_
z
A
z
B
q
j
q
i
h
B
j

z
A
z
B
z
B
q
i
_
=
=
_
B
X
j
h
B
j
_
z
A
z
B
+X
j
_
z
A
q
j
+
q
i
q
j
z
A
z
B
z
B
q
i
_
=
= Y
B
z
A
z
B
(1.5.18)
the cancelation coming from the identity
z
A
q
i
= 0
z
A
z
B
z
B
q
i
=
z
A
q
j
q
j
q
i
The role of Denition 1.5.2 in the study of the variational equation (1.5.8) is
further enhanced by the following
Denition 1.5.3. Let h be an innitesimal control for the (admissible) section
. A section X : R V () is said to be htransported along if and only if its
horizontal lift h(X) : R A( ) is an admissible innitesimal deformation of ,
namely if and only if i
h(X) = j
1
(X).
In view of equations (1.5.8), (1.5.14), setting X = X
i
(t)
_

q
i
_
, the condi-
tion for htransport is expressed in coordinates by the linear system of ordinary
dierential equations
dX
i
dt
=
__
i
q
k
_

+ h
k
A
_
i
z
A
_

_
X
k
= X
k

i
(1.5.19)
From the latter, recalling Cauchy theorem, we conclude that the htransported
sections of V () form an ndimensional vector space V
h
, isomorphic to each bre
V ()
|t
through the evaluation map X X(t) . We have thus proved:
Proposition 1.5.2. Every innitesimal control h : V () A( ) determines a
trivialization of the vector bundle V ()
t
R.
Proposition 1.5.2 provides an identication between sections X : R V ()
and vector valued functions X : R V
h
and therefore by duality also
an identication between sections

: R V
() and vector valued functions
: R V
h
, thus allowing the introduction of an absolute time derivative
D
Dt
for
vertical vector elds and virtual 1forms along .
The algorithm is readily implemented in components. To this end, let
_
e
(a)
_
,
_
e
(a)
_
denote any pair of dual bases for the spaces V
h
, V
h
. By denition, each
e
(a)
is a vertical vector eld along , obeying the transport law (1.5.19).
In coordinates, setting e
(a)
= e
i
(a)
_

q
i
_
, this implies the relation

de
i
(a)
dt
= e
k
(a)
i
(1.5.20a)
In a similar way, each e
(a)
is a virtual 1form along , expressed on the basis
i
as e
(a)
= e
(a)
i

i
, with e
(a)
i
e
i
(b)
=
a
b
.
On account of equation (1.5.20a), the components e
(a)
i
obey the transport law
d
dt
_
e
(a)
i
e
j
(a)
_
= 0 =
de
(a)
i
dt
= e
(a)
j
j
(1.5.20b)
The functions
i
j
:=
de
(a)
i
dt
e
j
(a)
= e
(a)
i
de
j
(a)
dt
(1.5.21a)
will be called the temporal connection coecients associated with the innitesimal
control h in the coordinate system t, q
i
. Comparison with equations (1.5.13),
(1.5.20a,b) provides the representation
i
j
=

j
=
_
j
q
i
_

h
i
A
_
j
z
A
_

(1.5.21b)
Given any vertical vector eld X = X
i
_

q
i
_
along , the denition of the operator

D
Dt
is then summarized into the expression
DX
Dt
=
d
dt
_
X, e
(a)
_
e
(a)
=
d
dt
_
X, e
(a)
i

i
_
e
(a)
=
d
dt
_
X
i
e
(a)
i
_
e
j
(a)
_

q
j
_
=
=
_
dX
j
dt
+ X
i
i
j
__

q
j
_
(1.5.22a)
with the coecients
i
j
given by equation (1.5.21b). In a similar way, given any
virtual 1form

=
i

i
, the same argument provides the evaluation
D
Dt
=
d
dt
_
, e
(a)
_
e
(a)
=
d
dt
_
i
e
i
(a)
_
e
(a)
j

j
=
_
d
j
dt

i
j
i
_

j
(1.5.22b)
By a little abuse of notation we shall henceforth systematically use the following
symbology:
DX
i
Dt
:=
_
DX
Dt
_
i
,
D
i
Dt
:=
_
D
Dt
_
i
The operation
D
Dt
is immediately extended to a derivation of the algebra of
virtual tensor elds along , commuting with contractions. In coordinates, we
have the representation
D
Dt
_
W
i
j
(t)
_

q
i
_

j

_
:=
DW
i
j
Dt
_

q
i
_

j
(1.5.23a)
with
DW
i
j
Dt
=
dW
i
j
dt
+
k
i
W
k
j

j
k
W
i
k
+ (1.5.23b)
After these preliminaries, let us go back to the variational equation (1.5.8). By
means of the projections (1.5.11), every section

X = X
i
_

q
i
_

+
A
_

z
A
_

splits
into the sum
X = T
H
(

X) + T
V
(

X) = h(X) + Y (1.5.24)
with
X =
(X) , Y = T
V
(

X) =
_
A
h
i
A
X
i
_
_

z
A
_

On the other hand, on account of equation (1.5.13), the variational equation
(1.5.8) is mathematically equivalent to the relation
dX
i
dt

k
_
i
_
X
k
=
_
h
k
A
X
k
+
A
_
_
i
z
A
_

Recalling equations (1.5.21b), (1.5.22a), (1.5.24), as well as the representation
(1.3.14) of the homomorphism V ( )

V (), the latter may be written syntheti-
cally as
DX
Dt
=
_
Y
_
=
_
T
V
(

X)
_
(1.5.25a)
or also, setting X = X
a
e
(a)
, Y = Y
A
_

z
A
_

, and expressing everything in com-
ponents in the basis e
(a)
dX
a
dt
=
_
e
(a)
,
_
Y
_
_
= e
(a)
i
_
i
z
A
_

Y
A
(1.5.25b)
Exactly as its original counterpart (1.5.8), equation (1.5.25a) points out that
every innitesimal deformation X is determined by the knowledge of a vertical
vector eld Y = Y
A
_

z
A
_

through the solution of a well posed Cauchy problem.
As we noticed earlier, the advantage is that, in the newer formulation, all
quantities have a precise geometrical meaning relative to the horizontal distribution
H( ) induced by the innitesimal control h. On the other hand, one should not
overlook the fact that, in the standard formulation of the problem, no distinguished
section h : V () A( ) is provided, and none is needed in order to formulate the
results. In this respect, the innitesimal control h plays the role of a gauge eld,
useful for covariance purposes, but unaecting the evaluation of the extremals.
Accordingly, in the subsequent analysis we shall employ h as a userdened object,
eventually checking the invariance of the results under arbitrary changes h h
.
1.5.3 Corners
In order to address a more and more vast class of problems, we actually shall not
deal with sections in the ordinary sense but with piecewise dierentiable evolutions,
dened on closed intervals. To account for this aspect, we adopt the following
standard terminology:
an admissible closed arc
_
, [a, b]
_
in 1
n+1
is the restriction to a closed
interval [a, b] of an admissible section : (c, d) 1
n+1
dened on some
open interval (c, d) [a, b];
a piecewise dierentiable evolution of the system in the interval [t
0
, t
1
] is a
nite collection
_
, [t
0
, t
1
]
_
:=
__
(s)
, [a
s1
, a
s
]
_
, s = 1, . . . , N, t
0
= a
0
< a
1
< < a
N
= t
1
_
of admissible closed arcs satisfying the matching conditions
(s)
(a
s
) =
(s + 1)
(a
s
) s = 1, . . . , N 1 (1.5.26)
On account of equation (1.5.26), the image (t) is well dened and continuous for
all t
0
t t
1
, thus allowing to regard the map : [t
0
, t
1
] 1
n+1
as a section in
a broad sense. The points (t
0
), (t
1
) are called the endpoints of , while the
points c
s
:= (a
s
) , s = 1, . . . , N 1 are called the corners of .
Consistently with the stated denitions, the lift of an admissible closed arc
_
, [a, b]
_
is the restriction to [a, b] of the lift : (c, d) /, while the lift
of a piecewise dierentiable evolution
__
(s)
, [a
s1
, a
s
]
__
is the family of lifts

(s)
, each restricted to the interval [a
s1
, a
s
]. The image (t) is well dened for all
t ,= a
1
, . . . , a
N1
, thus allowing to regard : [t
0
, t
1
] / as a (generally discontin-
uous) section of the velocity space. In particular, since the map i : / j
1
(1
n+1
)
is an imbedding of / into an ane bundle over 1
n+1
, each dierence
_

as
= i
_

(s + 1)
(a
s
)
_
i
_

(s)
(a
s
)
_
, s = 1, . . . , N 1
identies a vertical vector in T
cs
(1
n+1
), henceforth called the jump of at the
corner c
s
.
In local coordinates, setting q
i
(
(s)
(t)) := q
i
(s)
(t), equations (1.2.5), (1.5.26)
provide the representation
_

as
=
__
dq
i
(s + 1)
dt
_
as
_
dq
i
(s)
dt
_
as
__

q
i
_
cs
=
_
i
( )
_
as
_

q
i
_
cs
(1.5.27)
with
_
i
( )
as
:=
i
(
(s + 1)
(a
s
))
i
(
(s)
(a
s
)) denoting the jump of the function
i
( (t)) at t = a
s
.
Pursuing the generalization process, an admissible deformation of an admis-
sible closed arc
_
, [a, b]
_
is a 1parameter family
_
, [a(), b()]
_
, [[ < , of
admissible closed arcs depending continuously on and satisfying the condition
_
0
, [a(0), b(0)]
_
=
_
, [a, b]
_
. Notice that the denition explicitly includes possi-
ble variations of the reference intervals [a(), b()].
In a similar way, an admissible deformation of a piecewise dierentiable evo-
lution
_
, [t
0
, t
1
]
_
is a collection
__
(s)
, [a
s1
(), a
s
()]
__
of deformations of the
various arcs, satisfying the matching conditions
(s)
(a
s
()) =
(s + 1)
(a
s
()) [[ < , s = 1, . . . , N 1 (1.5.28)
Under the stated circumstances, the lifts
and
(s)
, respectively restricted
to the intervals [a(), b()] and [a
s1
(), a
s
() ] are easily recognized to provide
deformations for the lifts : [a, b] / and
(s)
: [a
s1
, a
s
] /.
Unless otherwise stated, we shall only consider deformations leaving the interval
[t
0
, t
1
] xed, namely those satisfying the conditions a
0
() t
0
, a
N
() t
1
. No
restriction will be posed on the functions a
s
(), s = 1, . . . , N 1.
Each curve c
s
() :=
(a
s
()) will be called the orbit of the corner c
s
under
the given deformation.
In local coordinates, setting q
i
(
(s)
(t)) =
i
(s)
(, t) , the matching conditions
(1.5.26) read
i
(s)
(, a
s
()) =
i
(s + 1)
(, a
s
()) (1.5.29)
while the representation of the orbit c
s
() takes the form
c
s
() : t = a
s
() , q
i
=
i
(s)
(, a
s
()) (1.5.30)
The previous arguments are naturally reected into the denition of the in-
nitesimal deformations. Thus, an admissible innitesimal deformation of an ad-
missible closed arc
_
, [a, b]
_
is a triple (, X, ), where X is the restriction to
[a, b] of an admissible innitesimal deformation of : (c, d) 1
n+1
, while ,
are the derivatives
=
da
d
=0
, =
db
d
=0
(1.5.31)
expressing the speed of variation of the interval [a(), b()] at = 0.
Likewise, an admissible innitesimal deformation of a piecewise dierentiable
evolution
_
, [t
0
, t
1
]
_
is a collection
_

s1
, X
(s)
,
s

_
of admissible innites-
imal deformations of each single closed arc, with
s
=
das
d
=0
, and, in particular,
with
0
=
N
= 0 whenever the interval [t
0
, t
1
] is held xed.
At the same time, whenever a corner c
s
is shifted by the deformation process,
the tangent vector to c
s
() is given by
W
(s)
=
_
_
c
s
()
_
d
d
_
=0
=
s
_

t
_
cs
+
_
i
+ X
i
_
cs
_

q
i
_
cs
(1.5.32)
The quantities
s
, X
i
(s)
, Z
i
(s)
arent actually independent: equations (1.5.29)
imply the identities
i
(s)
i
(s)
t
da
s
d
=
i
(s + 1)
i
(s + 1)
t
da
s
d
; (1.5.33a)
i
(s)
2
+ 2
i
(s)
t
da
s
d
+
i
(s)
t
2
_
da
s
d
_
2
+
i
(s)
t
d
2
a
s
d
2
=
=
i
(s + 1)
2
+ 2
i
(s + 1)
t
da
s
d
+
i
(s + 1)
t
2
_
da
s
d
_
2
+
i
(s + 1)
t
d
2
a
s
d
2
(1.5.33b)
From these, evaluating everything at = 0, recalling denitions (1.5.4) and intro-
ducing the notation
s
=
d
2
as
d
2
=0
, we get the jump relations
_
X
i
(s + 1)
X
i
(s)
_
as
=
s
_
dq
i
(s + 1)
dt

dq
i
(s)
dt
_
as
=
s
_
i
( )
_
as
(1.5.34a)
_
Z
i
(s + 1)
Z
i
(s)
_
as
= 2
s
_
dX
i
(s)
dt

dX
i
(s + 1)
dt
_
as
+
s
_
dq
i
(s)
dt

dq
i
(s + 1)
dt
_
as
+
+
2
s
_
d
2
q
i
(s)
dt
2

d
2
q
i
(s + 1)
dt
2
_
as
(1.5.34b)
whence also, in view of the variational equation (1.5.8),
_
Z
i
(s + 1)
Z
i
(s)
_
as
= 2
s
_
_
i
q
k
_

(s)
X
k
(s)

_
i
q
k
_

(s + 1)
X
k
(s + 1)
+
+
_
i
z
A
_

(s)
A
(s)

_
i
z
A
_

(s + 1)
A
(s + 1)
_
as
+
+
s
_
i
|
(s)
i
|
(s + 1)
_
as
+
2
s
_
d
i
|
(s)
dt

d
i
|
(s + 1)
dt
_
as
(1.5.34c)
Moreover, the admissibility of each single innitesimal deformation X
(s)
re-
quires the existence of a corresponding lift

X
(s)
= X
i
(s)
_

q
i
_

(s)
+
A
(s)
_

z
A
_

(s)
satisfying the variational equation (1.5.8).
Both aspects are conveniently accounted for by the assignment to each
(s)
of an (arbitrarily chosen) innitesimal control h
(s)
: V (
(s)
) A(
(s)
) . In this
way, proceeding as in 1.5.2 and denoting by
_
D
Dt
_
(s)
the absolute time derivative
along
(s)
induced by h
(s)
, we get the following
Proposition 1.5.3. Every admissible innitesimal deformation of an admissible
evolution
_
, [t
0
, t
1
]
_
over a xed interval [t
0
, t
1
] is determined, up to initial data,
by a collection of vertical vector elds
_
Y
(s)
= Y
A
(s)
_

z
A
_

(s)
_
, s = 1, . . . , N and
by N 1 real numbers
1
, . . . ,
N1
through the covariant variational equations
_
DX
(s)
Dt
_
(s)
= (Y
(s)
) = Y
A
(s)
_
i
z
A
_

(s)
_

q
i
_
(s)
s = 1, . . . , N (1.5.35)
completed with the jump conditions (1.5.34a). The lift of the deformation is de-
scribed by the family of vector elds
X
(s)
= h
(s)
(X
(s)
) + Y
(s)
, s = 1, . . . , N (1.5.36)
The proof is entirely straightforward, and is left to the reader. Introducing n piece-
wise dierentiable vector elds

1
, . . . ,

n
along according to the prescription
i
(t) = h
(s)
_

q
i
_
(s)
(t)
t (a
s1
, a
s
) , s = 1, . . . , N
equation (1.5.36) takes the explicit form
X
(s)
= h
(s)
_
X
i
(s)
_

q
i
_
(s)
_
+ Y
(s)
= X
i
(s)
i
+ Y
A
(s)
_

z
A
_

(s)
(1.5.37)
on each open arc
(s)
: (a
s1
, a
s
) /.
To discuss the implications of equation (1.5.35), resuming the notation V () for
the totality of vertical vectors along
4
, we dene a transport law in V (), hence-
forth called htransport, gluing h
(s)
transport along each arc
_
(s)
, [a
s1
, a
s
]
_
and
continuity at the corners, namely continuity of the components at t = a
s
.
In view of Proposition 1.5.2, the htransported elds form an ndimensional
vector space V
h
, isomorphic to each bre V ()
|t
. This provides a canonical iden-
tication of V () with the cartesian product [t
0
, t
1
] V
h
, thus allowing to regard
every section X : [t
0
, t
1
] V () as a vector valued function X : [t
0
, t
1
] V
h
.
Exactly as in 1.5.2, the situation is formalized referring V
h
to a basis e
(a)
related to the basis

_

q
i
_
by the transformation
_

q
i
_
= e
(a)
i
(t) e
(a)
, e
(a)
= e
i
(a)
(t)
_

q
i
_
(1.5.38)
Given any admissible innitesimal deformation
__
X
(s)
, [a
s1
, a
s
]
__
, we now
glue all sections X
(s)
: [a
s1
, a
s
] V (
(s)
) into a single, piecewise dierentiable
4
Notice that this makes perfectly good sense also at the corners (as).
function X : [t
0
, t
1
] V
h
, with jump discontinuities at t = a
s
expressed in
components by equation (1.5.34a). For each s = 1, . . . , N this provides the repre-
sentation
X
(s)
= X
a
(t) e
(a)
,
_
DX
(s)
Dt
_
(s)
=
dX
a
dt
e
(a)
t (a
s1
, a
s
) (1.5.39)
In a similar way, we collect all elds Y
(s)
into a single object Y , henceforth
conventionally called a vertical vector eld along .
By abuse of language, we also denote by Y = Y
A
_

z
A
_

the vector eld along
the open arcs of dened by the prescription
Y
A
(t) = Y
A
(s)
(t) a
s1
< t < a
s
, s = 1, . . . , N (1.5.40)
In this way, the covariant variational equation (1.5.35) takes the form
dX
a
dt
= Y
A
e
(a)
i
_
i
z
A
_

t ,= a
s
(1.5.41a)
completed with the jump conditions
_
X
a
as
=
_
X
i
as
e
(a)
i
(a
s
) =
s
e
(a)
i
(a
s
)
_
i
( )
_
as
s = 1, . . . , N 1 (1.5.41b)
1.5.4 The abnormality index
A deeper insight into the algorithm discussed in 1.5.3 is gained denoting by V
the innite dimensional vector space formed by the totality of vertical vector elds
Y =
_
Y
(s)
, s = 1, . . . , N
_
along , and setting W := V R
N1
. On account
of equations (1.5.41a,b), every admissible innitesimal deformation of is then
determined, up to initial data, by an element (Y,
1
, . . . ,
N1
) W.
In the following we shall be mainly interested in innitesimal deformations
X : [t
0
, t
1
] V () vanishing at the endpoints. Setting X(t
0
) = 0, equations
(1.5.41a,b) provide the evaluation
X(t) =
_
_
t
t
0
Y
A
e
(a)
i
_
i
z
A
_

dt
as<t
s
e
(a)
i
(a
s
)
_
i
( )
_
as
_
e
(a)
(1.5.42)
The vanishing of both X(t
0
) and X(t
1
) is therefore expressed by the condition
_
_
t
1
t
0
Y
A
e
(a)
i
_
i
z
A
_

dt
N1
s=1
s
e
(a)
i
(a
s
)
_
i
( )
_
as
_
e
(a)
= 0 (1.5.43)
The left hand side of equation (1.5.43) denes a linear map : WV
h
whose
kernel is therefore isomorphic to the vector space of the admissible innitesimal
deformations vanishing at the endpoints of .
Depending on the nature of the inclusion (W) V
h
, the evolutions of the
system will be classied into normal , when (W) = V
h
, and abnormal , when
(W) V
h
5
.
The dimension of the annihilator
_
(W)
_
0
V
h
will be called the abnormality
index of .
On this point, a useful characterization is provided by the following
Proposition 1.5.4. The annihilator
_
(W)
_
0
V

h
coincides with the totality
of htransported virtual 1forms

=
i

i
satisfying the conditions
i
_
i
z
A
_

= 0 A = 1, . . . , r (1.5.44a)
i
(a
s
)
_
i
( )
as
= 0 s = 1, . . . , N 1 (1.5.44b)
Proof. In view of equation (1.5.43), the subspace
_
(W)
_
0
V

h
consists of the
totality of elements

=
a
e
(a)
=
a
e
(a)
i

i
satisfying the relation
a
_
_
t
1
t
0
Y
A
e
(a)
i
_
i
z
A
_

dt
N1
s=1
s
e
(a)
i
(a
s
)
_
i
( )
_
as
_
= 0
(Y,
1
, . . . ,
N1
) W, clearly equivalent to equations (1.5.44a,b).
By equations (1.5.21b), (1.5.22b), the condition of htransport of

along each
arc
(s)
is expressed in coordinates as
d
i
dt
+
k
_
k
q
i
_

+ h
i
A
k
_
k
z
A
_

= 0 (1.5.45)
the cancellation arising from the requirement (1.5.44a).
The content of Proposition 1.5.4 is therefore independent of the choice of the
innitesimal controls h
(s)
: V (
(s)
) A(
(s)
) .
Remark 1.5.3: According to Proposition 1.5.4, the abnormality index of a piecewise
dierentiable section cannot exceed the abnormality index of each single arc
(s)
. Thus,
for example, if one of the arcs is normal, is necessarily normal. More generally, because of
the additional restrictions posed by equations (1.5.44b) and by the continuity requirements
[
]
as
= 0, an evolution may happen to be normal even if all its arcs
(s)
are abnormal.
Typical examples are:
5
As we shall see, when applied to the extremals of an action functional, this terminology agrees
with the current one (see, among others, [10] and references therein).
1
n+1
= RE
2
, referred to coordinates t, x, y. Constraint: x
2
+ y
2
= v
2
. Imbedding
/ j
1
(1
n+1
) expressed in coordinates as x = v cos z, y = v sinz. Piecewise
dierentiable evolution consisting of two arcs:
(1)
: x = 0, y = vt t
0
t 0
(2)
: x = vt, y = 0 0 t t
1
Equation (1.5.44a) admits htransported solutions

(1)
=
2
along
(1)
and
(2)
=
1
along
(2)
, , R. Both arcs are therefore abnormal. Notwith-
standing, is normal, since no pair

(1)
,
(2)
matches into a continuous nonnull
virtual 1form along .
1
n+1
= R E
2
. Coordinates t, x, y. Constraint: v
3
x = ( y
2
a
2
t
2
)
2
. Imbedding
/ j
1
(1
n+1
) expressed in coordinates as x = v
3
(z
2
a
2
t
2
)
2
, y = z. Piecewise
dierentiable evolution consisting of two arcs:
(1)
: x = 0, y =
1
2
a(t
2
t
2
) t
0
t t
(2)
: x =
a
4
5v
3
(t
5
t
5
), y = 0 t
t t
1
(t
,= 0). Equation(1.5.44a) admits htransported solutions of the form

=
1
along the whole of . Both arcs
(1)
,
(2)
are therefore abnormal. Notwithstanding,
is normal, since no solution satises condition (1.5.44b).
Remark 1.5.4: Even in the dierentiable case, the normality of an evolution is a global
property. In this sense, a normal arc : [t
0
, t
1
] 1
n+1
may happen to be abnormal
when restricted to a subinterval [t
0
, t
1
] [t
0
, t
1
] . An illustrative example may be given
by means of a bump function:
1
n+1
= R E
3
. Coordinates t, q
1
, q
2
, q
3
. Imbedding / j
1
(1
n+1
) expressed in
coordinates as q
1
= z
1
, q
2
= z
2
, q
3
= g(t)z
2
, being g(t) a C
function dened
as g(t) :=
2t
(t
2
1)
2
e
1
t
2
1
for any [t[ < 1 and g(t) := 0 otherwise. Dierentiable
evolution consisting of the single arc:
: q
1
= vt
2
, q
2
= vt , q
3
= vf(t) t
0
t t
1
, t
0
< 1 , t
1
> 1
being
f(t) :=
_
e
1
t
2
1
[t[ < 1
0 [t[ 1
For any R, equation(1.5.44a) admits therefore htransported solutions of the
form

=
3
when restricted to the subinterval [t
0
, 1] . Notwithstanding, is
normal, since no solution may be found along the whole of it.
In view of the contents of Remark 1.5.4, an evolution : [t
0
, t
1
] 1
n+1
will be
called locally normal if its restriction to any closed subinterval [t
0
, t
1
] [t
0
, t
1
] is
a normal arc, namely if and only if, along any such subinterval, equations (1.5.44)
admit the one trivial solution
i
(t) = 0.
As a concluding remark, its worth pointing out that, although geometrically
signicant, the arguments discussed so far provide only a partial picture of the
situation. Actually, rather than the totality of admissible innitesimal deforma-
tions vanishing at the endpoints here identied with the kernel of the map
: W V
h
a variational context involves the (possibly smaller) subfamily X
of innitesimal deformations tangent to admissible nite deformations with xed
endpoints.
The linear span of X, henceforth denoted by (), will be called the variational
space of . The evolutions of the system will be classied into ordinary, when
() = ker() and exceptional , when () ker().
A hierarchy between the various typologies is provided by the following
Proposition 1.5.5. The normal evolutions form a subset of the ordinary ones.
The result is proved in Appendix B. In this connection, see also [27].
Chapter 2
The rst variation
2.1 Problem statement
Let L F(/) denote a dierentiable function on the velocity space /, hence-
forth called the Lagrangian. Also, let
_
, [t
0
, t
1
]
_
( for short) denote an admis-
sible piecewise dierentiable evolution of the system, dened on a closed interval
[t
0
, t
1
] R. Indicating by the lift of to /, dene the action functional
1[] :=
_

Ldt :=
N
s=1
_
as
a
s1
_

(s)
_
(L) dt (2.1.1)
As it was already outlined in the Introduction, the problem we intend to deal
with is the one of characterizing, among all the admissible evolutions connecting
a given pair of points in 1
n+1
, the ones (if any) which minimize
1
the functional
(2.1.1). More precisely, recalling Denition 1.5.1, we state the following
Denition 2.1.1. An evolution
_
, [t
0
, t
1
]
_
is called a weak local minimum for the
functional (2.1.1) if there is a neighborhood ^
(,1)
() of , such that 1[] 1[
]
for all admissible piecewise dierentiable
^
(,1)
() joining the endpoints
of . The evolution is likewise called a strong local minimum for the functional
(2.1.1) if all previous properties hold, with ^
(,1)
() systematically replaced by
^
(,0)
().
As a direct result of Denitions 1.5.1, 2.1.1, we see that every strong extremum
is also a weak one while the converse is generally false. Therefore, once the nec-
essary and sucient conditions for a weak minimum will have been found out, it
will be possible to try to supplement them in such a way as to guarantee a strong
minimum as well. However, this will not be carried out in the present work.
1
For the sake of explicitness, we shall consider only conditions for a minimum. In order to
obtain the conditions for a maximum, it is only needed to reverse the direction of all inequalities.
40 Chapter 2. The rst variation
Given an admissible evolution , we keep in line with Denition 2.1.1 by con-
sidering all weak deformations
with xed endpoints.

The rst step for the solution of the problem is now to study the stationarity
conditions for the functional (2.1.1), through the analysis of its socalled rst
variation.
Denition 2.1.2. An admissible evolution is called an extremal for the func-
tional (2.1.1) if and only if, for all admissible deformations with xed endpoints
=
__
(s)
, [a
s1
(), a
s
()]
__
, the function
1[
] :=
_

Ldt =
N
s=1
_
as()
a
s1
()
_

(s)
(L) dt
has a stationarity point at = 0.
Remark 2.1.1 (The gauge group): As it is well known, given any pair of 1forms L dt
and L
dt over /, their respective action integrals 1[] =

_

L dt and 1
[] =
_

L
dt
give rise to the same extremal curves if the dierence
_
L
L)dt is an exact dierential.

Under this circumstance, the equality
_
L dt =
_
L
dt holds along any closed curve,

thereby entailing the relation
1
] 1[
] =
_

_
L
L
_
dt
_

_
L
L
_
dt
for any deformation
vanishing at the endpoints, whence also

d
d
_
1
] 1[
]
_
0
In this particular sense, as far as a variational problem based on the functional (2.1.1)
is concerned, the Lagrangian function L F(/) is dened up to an equivalence relation
of the form
L L
L =
df
dt
, f F(1
n+1
) (2.1.2)
Otherwise stated, the real information isnt brought so much by L in itself as by a whole
family of Lagrangians, equivalent to each other in the sense expressed by equation (2.1.2).
The signicance of the arguments developed in 1.4.2 relies actually on the fact, ex-
plicitly pointed out by equations (1.4.16), (1.4.17), that the representation of an arbitrary
section : / L(/) involves exactly this family of Lagrangians, henceforth denoted by
(). A straightforward check shows that a necessary and sucient condition for two
sections and
to full () = (
) is that the dierence
, viewed as a function
over /, be itself of the form
=
df
dt
, f F(1
n+1
) (2.1.3)
Thus we see that, within our geometrical framework, the equivalence relation (2.1.2) be-
tween functions is replaced by the almost identical relation (2.1.3) between sections. Intu-
itively, the latter is a sort of active counterpartof the transformation law (1.4.17) for the
representation of a given section under arbitrary changes of the trivialization u: P R.
2.1 Problem statement 41
This viewpoint is formalized through the introduction of the concept of gauge group
2
.
By denition, a gauge transformation of the bundle P 1
n+1
is an isomorphism
P
g
P
_
1
n+1
1
n+1
bred over the identity map, and equivariant with respect to the action of the structural
group, namely fullling
g ( +) = g () + P , (2.1.4)
On the basis of equation (2.1.4), it is easily recognized that the group of gauge transforma-
tions over P is in 1-1 correspondence with the ring of dierentiable functions over 1
n+1
,
the relation f g
f
being given explicitly by
f F(1
n+1
) g
f
() := + f(()) P (2.1.5)
In local coordinates, the action of the map g
f
is expressed synthetically as
g
f
: (t, q
i
, u) (t, q
i
, u +f)
Every gauge transformation (2.1.5) may be lifted in a canonical way to a dieomor-
phism g
f
: j
A
1
(P, R) j
A
1
(P, R), expressed in coordinates as
g
f
: (t, q
i
, u, z
A
, u) (t, q
i
, u +f, z
A
, u +

f)
From this it is easily seen that the map g
f
commutes with both group actions (1.4.15a),
(1.4.15b), thus inducing maps g
f
: L(/) L(/) , and g
c
f
: L
c
(/) L
c
(/) , expressed
symbolically as
g
f
: (t, q
i
, z
A
, u) (t, q
i
, z
A
, u +

f)
g
c
f
: (t, q
i
, u, z
A
) (t, q
i
, u +f, z
A
)
The situation is summarized into the commutative diagrams
j
A
1
(P, R)
g
f
j
A
1
(P, R)
_
L(/)
g
f
L(/)
_
/ /
j
A
1
(P, R)
g
f
j
A
1
(P, R)
_
L
c
(/)
g
c
f
L
c
(/)
_
/ /
in which all horizontal arrows denote bundle isomorphisms.
It is now an easy matter to verify that equation (2.1.3) is mathematically equivalent
to the condition
= g
f
(2.1.6)
The geometrical counterpart of an equivalence class of Lagrangians on / is therefore a
section : / L(/) , dened up to the action of the gauge group.
2
See, for example, [4]
2.2 The PontryaginPoincareCartan form
To begin with, we focus on the lefthand face of diagram (1.4.28)
o ((j
1
(1
n+1
))
L(1
n+1
) j
1
(1
n+1
)
(2.2.1)
and we complete the state of the play with the two missing ingredients that are
needed to address the problem, namely
the nonholonomic constraints (sometimes improperly called the dynam-
ics), described by the imbedding i : / j
1
(1
n+1
) and locally expressed by
the equations
q
i
=
i
(t, q
1
, . . . , q
n
, z
1
, . . . , z
r
)
the nonholonomic Lagrangian section : u = L(t, q
i
, z
A
).
We next pullback the diagram (2.2.1) through the imbedding /
i
j
1
(1
n+1
),
giving rise to the analogous diagram
o
A
((/)
L(/) /
(2.2.2)
By construction, the manifold o
A
is then a principal bre bundle over ((/) under
the (induced) action
: (t, q
i
, z
A
, u
i
, p
i
) (t, q
i
, z
A
, u
i
+, p
i
) (2.2.3)
By means of the pullback procedure, the canonical form (1.4.32) determines
a distinguished 1form on o
A
, locally expressed by
3
u
= p
0
dt +p
i
dq
i
udt +p
i
_
dq
i
i
dt
_
(2.2.4)
Every nonholonomic Lagrangian section : / L(/) determines a trivial-
ization
: L(/) R of the bundle L(/) /. Let
:=
S
(
) denote the
pullback of
to o
A
, locally expressed as

(t, q
i
, z
A
, u, p
i
) =
(t, q
i
, z
A
, u) = u L(t, q
i
, z
A
) (2.2.5)
3
Aiming for easiness, the same symbol u will stand for both the form (1.4.32) and its
pullback on A.
2.3 The Pontryagins maximum principle 43
From this, taking equation (2.2.3) into account, it is an easy matter to check that
the function
is a trivialization of the bundle o

A
((/) and that, as such, it
determines a section

: ((/) o
A
, locally described by the equation
u = L(t, q
i
, z
A
) (2.2.6)
In brief, every section : / L(/) may be lifted to a section

: ((/) o
A
.
The local representations of both sections are formally identical and they obey the
transformation law (1.4.17) for an arbitrary change of the trivialization u: P R.
The section

: ((/) o
A
may now be used to pullback the form (2.2.4) onto
((/), hereby getting the 1form
PPC
:=

(
u
) = L dt + p
i
_
dq
i
i
dt
_
:= H dt + p
i
dq
i
(2.2.7)
henceforth referred to as the PontryaginPoincareCartan form.
Needless to say, the dierence H := p
i
i
L , known in the literature as the
Pontryagin Hamiltonian, is not an Hamiltonian in the traditional sense but a
function on the contact bundle.
2.3 The Pontryagins maximum principle
To understand the role of the PontryaginPoincareCartan form in the solution of
the addressed variational problem, we focus on the bration ((/)

1
n+1
, given
by the composite map := . A piecewise dierentiable section
_
, [t
0
, t
1
]
_
consisting of a nite family of closed arcs

(s)
: [a
s1
, a
s
] ((/) , s = 1, . . . , N, t
0
= a
0
< a
1
< < a
N
= t
1
will be called continuous if and only if the composite map is continuous,
namely if and only if projects onto a continuous, piecewise dierentiable section
: [t
0
, t
1
] 1
n+1
. A deformation
=
__

(s)
, [a
s1
(), a
s
()]
__
will similarly
be called continuous if and only if all sections
are continuous. A necessary

and sucient condition for this to happen is the validity of the matching conditions
(1.5.28), synthetically written as
lim
ta
+
s
()

(t) = lim
ta
s
()

(t) s = 1, . . . , N 1 (2.3.1)
A continuous deformation
is said to preserve the endpoints of if and

only if
is a deformation with xed endpoints. A vector eld along tangent

to the orbits of a continuous deformation is called an innitesimal deformation.
Notice that, since the stated denitions do not include any admissibility re-
quirement for the sections
, the only condition needed in order for a vector

eld X
i
_

q
i
_

+
A
_

z
A
_

+
i
_

p
i
_

to represent an innitesimal deformation of
is the consistency with the matching conditions (2.3.1), expressed in components
by the jump relations
lim
ta
+
s
()
_
X
i
+
s
dq
i
dt
_
= lim
ta
s
()
_
X
i
+
s
dq
i
dt
_
s = 1, . . . , N 1 (2.3.2)
with
s
=
_
das
d
_
=0
. On the same line as in 1.2, any section : [t
0
, t
1
] ((/),
locally described as
q
i
= q
i
(t), z
A
= z
A
(t), p
i
= p
i
(t)
and satisfying
dq
i
dt
=
i
_
t, q
1
(t), . . . , q
n
(t), z
1
(t), . . . , z
r
(t)
_
will henceforth be called admissible.
By means of
PPC
we now dene an action integral over ((/), assigning to
each continuous section : q
i
= q
i
(t), z
A
= z
A
(t), p
i
= p
i
(t) the real number
1[ ] :=
_

PPC
=
_
t
1
t
0
_
p
i
dq
i
dt
H
_
dt (2.3.3)
From the foregoing discussion, it should be clear that two dierent forms
PPC
and
PPC
linked together by a change of the trivialization u of P give rise to
two distinct representations of the same variational problem. In other words, the
extremal curves of two variational problems diering by the action of the gauge
group project onto the very same curve in 1
n+1
. In this connection, the study
of the consequences of both the impositions u =

f and in an extreme case
u = 0 gains some relevance.
For any continuous deformations
preserving the endpoints of we

have the relation
d1[
]
d
=0
=
_
t
1
t
0
__
dq
i
dt

H
p
i
_
i

_
dp
i
dt
+
H
q
i
_
X
i
H
z
A

A
_
dt +
+
N
s=1
_
lim
ta
s
_
s
_
p
i
dq
i
dt
H
_
+p
i
X
i
_
lim
ta
+
s1
_
s1
_
p
i
dq
i
dt
H
_
+p
i
X
i
_
_
From the latter, taking equations (2.3.1) and the conditions X
i
(t
0
) = X
i
(t
1
) = 0
into account, we conclude that the vanishing of
dI
d
=0
under arbitrary deforma-
tions of the given class is mathematically equivalent to the system
dq
i
dt
=
H
p
i
=
i
(t, q
i
, z
A
) (2.3.4a)
dp
i
dt
=
H
q
i
= p
k
k
q
i
+
L
q
i
(2.3.4b)
H
z
A
= p
i
i
z
A

L
z
A
= 0 (2.3.4c)
completed with the continuity conditions
_
p
i
as
=
_
H
as
= 0 s = 1, . . . , N 1 (2.3.4d)
where, as usual, we are denoting by [f ]
as
the jump of the function f(t) at t = a
s
.
Equation (2.3.4a) shows that the extremal curves for the functional (2.3.3) are
admissible. Therefore, whenever any of them is concerned, we have the identica-
tion
1[ ] :=
_

PPC
=
_
t
1
t
0
_
L + p
i
_
dq
i
dt

i
_
_
dt =
_
t
1
t
0
L
_
t, q
i
(t), z
A
(t)
_
dt
(2.3.5)
Moreover, their being extremals with respect to arbitrary deformations vanish-
ing at the endpoints automatically makes them extremals with respect to the
narrower class of admissible deformations as well. As a consequence, we can
state that every free extremal for the functional (2.3.3) gives rise to an extremal
: q
i
= q
i
(t) of the original problem.
Conversely, it is a much more awkward matter to establish if and under which
hypotheses an admissible evolution is an extremal for the functional (2.1.1) which
can be obtained from an extremal for the functional (2.3.3). Heuristically, the
variational problem (2.3.3) can be viewed as the study of the functional (2.1.1)
in which the kinematical admissibility condition (1.2.5) plays no more the role of
an a priori request upon sections but it is retrieved afterwards by the method of
Lagrange multipliers. It is therefore reasonable that, under suitable hypotheses,
one can prove the equivalence between the variational problem in / and the one
in ((/). Let us investigate this point.
Given an admissible piecewise dierentiable evolution , denoting by

X
(s)
the
innitesimal deformation associated with each single
(s)
and recalling the deni-

tion
s
=
das
d
=0
, the search for the extremality conditions for passes through
the evaluation
d1[
]
d
=0
=
N
s=1
_
d
d
_
as()
a
s1
()
L
_

(s)
_
dt
_
=0
=
=
N
s=1
_ _
as
a
s1
X
(s)
(L) dt +
_
s
L(
(s)
(a
s
))
s1
L(
(s)
(a
s1
))
_
_
(2.3.6a)
On account of the assumption
0
=
N
= 0, recalling equation (1.5.37) and
denoting by
_
L( )
as
:=
_
L(
(s + 1)
(a
s
)) L(
(s)
(a
s
))
the jump of the function L( (t)) at t = a

s
, equation (2.3.6a) may be concisely
written as
d1[
]
d
=0
=
N
s=1
_
as
a
s1
_
X
i
(s)
i
(L) +Y
A
(s)
L
z
A
_
dt
N1
s=1
s
_
L( )
as
(2.3.6b)
Equation (2.3.6b) is further elaborated by means of the introduction of N virtual
1forms

(s)
= p
(s)
i
(t)
i
(one for each arc
(s)
) satisfying the transport law
_
D
(s)
Dt
_
(s)
=
_
i
L
_

(s)

i
(2.3.7a)
as well as the matching conditions
(s)
as
=

(s + 1)
as
s = 1, . . . , N 1 (2.3.7b)
In order to make the notation as easy as possible we collect all

(s)
into a
continuous, piecewise dierentiable section

: [t
0
, t
1
] V
() according to the
prescription
(t) =

(s)
(t) t [a
s1
, a
s
] (2.3.8)
On account of equations (2.3.7a,b),

is then uniquely determined by L , up
to initial data at t = t
0
.
Taking the covariant variational equation (1.5.35) as well as the duality rela-
tions
_ _

q
i
_
(s)
,
k
_
=
k
i
into account, by equation (2.3.7a) we get the expres-
sion
X
i
(s)
i
L =
_
X
(s)
,
_
D
(s)
Dt
_
(s)
_
=
d
dt
_
X
(s)
,

(s)
_
__
DX
(s)
Dt
_
(s)
,

(s)
_
=
=
d
dt
_
X
i
(s)
p
(s)
i
_
p
(s)
i
_
i
z
A
_

(s)
Y
A
(s)
whence also
_
as
a
s1
X
i
(s)
i
(L) dt =
_
X
i
(s)
p
(s)
i
_
as
a
s1
_
as
a
s1
p
(s)
i
_
i
z
A
_

(s)
Y
A
(s)
dt
Summing over s, restoring the notations (1.5.40), (2.3.8) and recalling equations
(1.5.34a), (2.3.7b) as well as the conditions X(t
0
) = X(t
1
) = 0, this implies the
relation
N
s=1
_
as
a
s1
X
i
(s)
i
(L) dt =
_
t
1
t
0
p
i
_
i
z
A
_

Y
A
dt +
N1
s=1
s
_
i
( )
_
as
p
i
(a
s
)
In this way, omitting all unnecessary subscripts, equation (2.3.6b) gets the nal
form
d1[
]
d
=0
=
_
t
1
t
0
_
L
z
A
p
i
i
z
A
_
Y
A
dt +
N1
s=1
s
_
p
i
(t)
i
( ) L( )
_
as
(2.3.9)
In the algebraic environment introduced in 1.5.4, the previous discussion is
naturally formalized regarding the right hand side of equation (2.3.9) as a linear
functional d1
: W R on the vector space W = V R

N1
. A necessary
and sucient condition for to be an extremal for the functional (2.1.1) is then
the vanishing of d1
on the subset X W formed by the totality of elements

Y,
1
, . . . ,
N1
arising from nite deformations with xed endpoints. By linear-
ity, the previous condition is mathematically equivalent to the requirement
() ker(d1
) (2.3.10)
with () = Span(X) ker() denoting the variational space of .
As we shall see, equation (2.3.10) provides an algorithm for the determination
of all the extremals of the functional (2.1.1) within the class of ordinary evolutions.
The exceptional case is considerably more complicated, because of the lack
of an explicit characterization of the space () in terms of the local properties
of the section . In this respect, the simplest procedure and, quite often, the
only available one, is checking equation (2.3.10) separately on each exceptional
evolution.
In what follows we shall adopt an intermediate strategy, namely, rather than
dealing with equation (2.3.10) we shall discuss the implications of the stronger
requirement
ker() ker(d1
) (2.3.11a)
According to the classication introduced in 1.5.4, the latter is necessary and
sucient for an ordinary evolution to be an extremal of the functional (2.1.1),
but merely sucient for an exceptional evolution to be an extremal.
In the exceptional case, condition (2.3.11a) is sucient
Ker(dI)
Ker
()
but not necessary
Ker(dI)
Ker
()
In the ordinary case condition (2.3.11a) is instead both
necessary and sucient
Ker(dI)
()
By elementary algebra, the requirement (2.3.11) is equivalent to the existence
of a (possibly nonunique) linear functional K : V
h
R satisfying the relation
-
?
K
Q
Q
Q
Q
Qs
d1
W V
h
R
(2.3.11b)
Setting K = K
a
e
(a)
, and recalling equations (1.5.43), (2.3.9), the requirement
(2.3.11b) is expressed in components as
_
t
1
t
0
_
L
z
A
p
i
i
z
A
_
Y
A
dt +
N1
s=1
s
_
p
i
(t)
i
( ) L( )
_
as
=
K
a
_
_
t
1
t
0
Y
A
e
(a)
i
_
i
z
A
_

dt
N1
s=1
s
e
(a)
i
(a
s
)
_
i
( )
_
as
_
By the arbitrariness of Y,
1
, . . . ,
N1
, the latter condition splits into the
system
L
z
A

_
p
i
+K
a
e
(a)
i
_

i
z
A
= 0 A = 1, . . . , r (2.3.12a)
__
p
i
+K
a
e
(a)
i
_
i
( ) L( )
_
as
= 0 s = 1, . . . , N 1 (2.3.12b)
Collecting all results, and recalling Propositions 1.5.4, 1.5.5 we conclude
Theorem 2.3.1. Given an admissible evolution , let () denote the totality
of piecewise dierentiable virtual 1forms

= p
i
(t)
i
along satisfying equa-
tions (2.3.7a,b), (2.3.8) as well as the nite relations
p
i
i
z
A
=
L
z
A
A = 1, . . . , r (2.3.13a)
and the matching conditions
_
p
i
i
( ) L( )
_
as
= 0 s = 1, . . . , N 1 (2.3.13b)
Then:
a) the condition () ,= is sucient for to be an extremal for the functional
(2.1.1);
b) if is an ordinary evolution, the same condition is also necessary for to
be an extremal;
c) is a normal extremal, namely an extremal belonging to the class of normal
evolutions, if and only if the set () consists of a single element.
Proof. In view of equations (2.3.9), (2.3.13a,b), whenever the ansatz

()
is allowed, it implies
dI
d
=0
= 0 for all admissible innitesimal deformations
vanishing at the endpoints of . Assertion a) is then a direct consequence of
Denition 2.1.2.
In particular, according to our previous discussion, if is an ordinary extremal,
there exists at least one htransported 1form K = K
a
e
(a)
satisfying equations
(2.3.12a,b) in correspondence with any continuous virtual 1-form

= p
i

i
obey-
ing the transport law (2.3.7a). The sum

+ K =
_
p
i
+ K
a
e
(a)
i
_

i
is hence
automatically in the class (), thus proving assertion b).
Finally, as pointed out in 1.5.2, the normal evolutions form a subclass of
the ordinary ones, uniquely characterized by the requirement
_
(W)
_
0
= 0.
Therefore, according to assertion b), a normal evolution is an extremal if and
only if the class () is nonempty. Moreover, by equations (2.3.7a), (2.3.12a), if
is any pair of elements in the class (), the dierence

is automatically
an htransported 1form satisfying equations. (1.5.44a,b). By Proposition 1.5.4
this implies

_
(W)
_
0
, thus establishing assertion c).

In view of equations (1.5.21b), (1.5.22b)), for any

() the transport law
(2.3.7a) simplies to
dp
i
dt
+ p
k
_
k
q
i
_

+ h
i
A
p
k
_
k
z
A
_

=
_
L
q
i
_

+
h
i
A
_
L
z
A
_

the cancellation being due to equation (2.3.13a). Exactly as it happened with
Proposition 1.5.4, all assertions of Theorem 2.3.1 have therefore an intrinsic mean-
ing, irrespective of the choice of the innitesimal controls h
(s)
: V (
(s)
) A(
(s)
) .
The previous arguments provide an algorithm for the determination of the ordinary
extremals of the functional (2.1.1), relying on 2n +r equations
dq
i
dt
=
i
(t, q
i
, z
A
) (2.3.14a)
dp
i
dt
+

k
q
i
p
k
=
L
q
i
(2.3.14b)
p
i
i
z
A
=
L
z
A
(2.3.14c)
for the unknowns q
i
(t), p
i
(t), z
A
(t), completed with the continuity requirements
_
q
i
as
=
_
p
i
as
=
_
p
i
i
L
as
= 0 s = 1, . . . , N 1 (2.3.15)
Collecting all results, we can now state the following
Theorem 2.3.2. Every ordinary extremal for the functional (2.1.1) is the pro-
jection of at least one extremal for the functional (2.3.3). Moreover, the nor-
mality of implies the uniqueness of .
Proof. It is easily seen from the previous discussion that every ordinary extremal
: R 1
n+1
for the functional (2.1.1) determines both a unique admissible section
: R / and a section

: R V
() belonging to (). Because of the nature

of the contact bundle ((/) of bre bundle over the space V
(1
n+1
), identical to
the pull-back of the latter through the map /

1
n+1
,
((/)

V
(1
n+1
)
/

1
n+1
the pair ( ,

) characterizes a vcontinuous section : R ((/) satisfying
= , =

The thesis follows now directly from the observation that the equations (2.3.4)
coincide exactly with the equations (2.3.14), (2.3.15). The section is therefore
an extremal for the functional (2.3.3) which projects onto .
Eventually, whenever is normal, the uniqueness of is a straightforward
consequence of the fact that in this case the set () consists of a single
element, as shown in Theorem 2.3.1.
As far as the ordinary extremals are concerned, the original constrained vari-
ational problem in the event space is therefore equivalent to a free variational
problem in the contact manifold. This is precisely the essence of Pontryagins
maximum principle.
As already pointed out, all equations (2.3.14), (2.3.15) are independent of the
choice of the innitesimal controls, and involve only thetruedata of the problem,
namely the Lagrangian section and the constraint equations (1.2.5). In particu-
lar, the last pair of equations (2.3.15) extend to the ordinary evolutions the well
known ErdmannWeierstrass corner conditions of holonomic variational calculus
[8, 19].
Remark 2.3.1 (Same problem, equivalent solution): There is another possible approach
to the problem, slightly dierent but completely equivalent to the one outlined so far.
Apparently, it complicates matters without giving any signicant advantage. On the other
hand, it seems to be the most faithful translation of the original Pontryagins treatment
of the subject ([17]) into the geometrical context. Hence, at least for historical reasons, it
is worth telling about.
A variational problem, based on the functional
1[] :=
_

udt (2.3.16)
is introduced in the manifold L(1
n+1
), where stands for the jetextension of a section
: [t
0
, t
1
] P. As the 1form udt is well dened in L(1
n+1
) up to a term

f dt, the
functional (2.3.16) is independent of a particular choice of the gauge.
Setting : q
i
= q
i
(t) , u = u(t), it follows that
_

udt = u(t
1
) u(t
0
)
and so, assuming the values of q
i
(t
0
) and q
i
(t
1
) as xed, the problem consists in nding
a curve which minimizes the increment u(t
1
) u(t
0
) and whose projection onto 1
n+1
leaves the endpoints xed.
We now require the section to belong to the submanifold

/ of L(1
n+1
) locally
described by the equations
q
i
=
i
(t, q
i
, z
A
) , u = L(t, q
i
, z
A
) (2.3.17)
In other words, we are making use of the simultaneous assignment of both the kinetic
constraints and the Lagrangian section to express the submanifold

/ as the image (/)
4
.
In this way, every admissible section q
i
= q
i
(t) in 1
n+1
determines, up to a constant, an
admissible section q
i
= q
i
(t), u = u(t) of P.
Compared with the main approach outlined in 2.2, the present formulation just re-
places the section : / L(/) with the image space

/ = (/), considered as a sub-
manifold of L(1
n+1
). This submanifold, and consequently the section , is regarded as
set and therefore the representation u = L(t, q
i
, z
A
) is aected only by passive gauge
transformations. The same variational problem is bred by dierent submanifolds related
together by the action of the gauge group.
As well as in 2.2, the submanifold

/ L(1
n+1
) is lifted up onto a submanifold
((

/) o both by identifying C(

/) with the image space

(((/)) and by pulling back o
onto

/ by means of the commutative diagram
((

/)

o
/

L(1
n+1
)
(2.3.18)
All the same, the imbedding ((

/)

o is bred onto V
(1
n+1
) and its expression in
coordinate is formally identical to equations (2.3.17) which are involved in the represen-
tation of the submanifold

/.
It is now possible to make use of the form (1.4.32) to induce on ((

/) the 1form

(
u
) = L dt +p
i
_
dq
i
i
dt
_
(2.3.19)
and, consequently, to dene an action integral by the integration of the form (2.3.19) along
any section : [t
0
, t
1
] ((

/). Once again, this merely reproduces in the image space
((

/) the construction carried on in 2.2. Namely, the 1form(2.3.19) is simply the image of
the PontryaginPoincareCartan form (2.2.7) under the dieomorphism

: ((/) ((

/).
As previously mentioned, a signicant role in the study of the variational prob-
lem based on the functional (2.3.3) is played by the choice of the Lagrangian section
as u =

f(t, q). In this particular situation, the problem under discussion can be
made, by means of a gauge transformation, into the study of the action functional
1
0
[ ] :=
_

L
=
_
t
1
t
0
p
i
_
dq
i
dt

i
_
dt (2.3.20)
induced by the Liouville form (1.2.17) of ((/). The corresponding extremal curves
4
Needless to say, the holonomic case is automatically included in the present scheme, the
constraints (2.3.17) being reduced to the single request u = L(t, q
i
, q
i
).
are easily seen to satisfy the EulerLagrange equations
dq
i
dt
=
i
(t, q
i
, z
A
) (2.3.21a)
dp
i
dt
+

k
q
i
p
k
= 0 (2.3.21b)
p
i
i
z
A
= 0 (2.3.21c)
_
p
i
as
=
_
p
i
as
= 0 s = 1, . . . , N 1 (2.3.21d)
Equation (2.3.21a) is the admissibility requirement for the section . For this
reason, if an extremal of the functional (2.3.20) satises = , its projection
under the map : ((/) / coincides with the lift : [t
0
, t
1
] /.
For any admissible , the extremals projecting onto are therefore in 11
correspondence with the solutions p
i
(t) of the homogeneous system (2.3.21b,c,d),
with the functions q
i
(t), z
A
(t) regarded as given. On the other hand, according to
Proposition 1.5.4, equations (2.3.21b,c,d) are precisely the relations characterizing
the totality of virtual 1forms p
i
(t)
i
belonging to the annihilator
_
(W)
_
0
.
We have thus proved the following
Proposition 2.3.1. Let : [t
0
, t
1
] 1
n+1
denote any continuous, piecewise
dierentiable section. Then:
a) is admissible if and only if the functional (2.3.20) admits at least one
extremal projecting onto , namely satisfying = ;
b) for any such , the totality of extremals of 1
0
projecting onto form a nite
dimensional vector space over R, with dimension equal to the abnormality
index of .
In the language of 1.5.4, Proposition 2.3.1 asserts that a section : [t
0
, t
1
] 1
n+1
describes a normal evolution of the system if and only if the functional (2.3.20)
admits exactly one extremal projecting onto , namely the one corresponding to
the trivial solution p
i
(t) = 0. If the extremals projecting onto are more than one,
represents an abnormal evolution; if no such extremal exists, is not admissible.
We now come back to the study of the variational problem based on functional
(2.3.3) and we state
Proposition 2.3.2. The totality of extremals of the functional (2.3.3) projecting
onto a section : [t
0
, t
1
] 1
n+1
is an ane space, modelled on the vector space
formed by the extremals of the functional (2.3.20) projecting onto .
Proof. The proof is entirely straightforward and is based on the observation that
if : q
i
= q
i
(t), z
A
= z
A
(t), p
i
= p
i
(t) and
: q
i
= q
i
(t), z
A
= z
A
(t), p
i
=
i
(t)
are both extremals of the functional (2.3.3) projecting onto , then the contempo-
raneous validity of the EulerLagrange equations
dq
i
dt
=
i
(t, q
i
, z
A
) ,
dp
i
dt
+ p
k
k
q
i
=
L
q
i
, p
k
k
z
A
=
L
z
A
dq
i
dt
=
i
(t, q
i
, z
A
) ,
d
i
dt
+
k
k
q
i
=
L
q
i
,
k
k
z
A
=
L
z
A
implies that the curve q
i
= q
i
(t), z
A
= z
A
(t), p
i
= p
i
(t)
i
(t) is an extremal for
the functional (2.3.20).
The previous arguments provide a restatement of Theorem 2.3.1 in the envi-
ronment ((/). In particular, it is worth remarking that, in general, the projection
algorithm , applied to the totality of extremals of the functional (2.3.3),
does not yield back all the extremals of the functional (2.1.1), but only a subclass,
wide enough to include the ordinary ones. The missing extremals may be obtained
determining the abnormal evolutions by means of Proposition 2.3.1, nding out
which ones have an exceptional character, and analyzing each of them individually.
2.4 Hamiltonian formulation
Temporarily leaving aside all aspects related to the presence of corners, we observe
that a dierentiable curve in ((/) is at the same time a section with respect to
the bration ((/)
t
R and an extremal for the functional (2.3.3) if and only if
its tangent vector eld Z :=
_

t
_
satises the properties
Z, dt
_
= 1 , Z d
PPC
= 0 (2.4.1)
On account of equation (2.2.7), at any ((/) a necessary and sucient
condition for the existence of at least one vector Z T
(((/)) satisfying equa-

tions (2.4.1) is the validity of the relations
_
H
z
A
_
= 0 (2.4.2a)
Points at which equations (2.4.1) admit a unique solution Z will be called
regular points for the functional (2.3.3). In coordinates, the regularity requirement
is expressed by the condition
det
_

2
H
z
A
z
B
_
,= 0 (2.4.2b)
2.4 Hamiltonian formulation 55
In view of equation (2.4.2b), in a neighborhood of each regular point equations
(2.4.2a) may be solved for the z
A
s, giving rise to a representation of the form
z
A
= z
A
(t, q
1
, . . . , q
n
, p
1
, . . . , p
n
) (2.4.3)
The regular points form therefore a (2n+1)dimensional submanifold
j
((/),
locally dieomorphic to the space V
(1
n+1
).
When restricted to the submanifold , the pullback of the form (2.2.4) by
means of the section

: ((/) o
A
provides the 1form
PPC
:=
_
j

(
u
) = Hdt + p
i
dq
i
(2.4.4)
having denoted by H := j
(H ) the pullback of the Pontryagin Hamiltonian,

expressed in coordinates as
H = H (t, q
r
, z
A
(t, q
i
, p
i
), p
r
) = p
k
k
(t, q
r
, z
A
(t, q
i
, p
i
)) L(t, q
r
, z
A
(t, q
i
, p
i
))
In view of equations (2.2.7), (2.4.2a) we have then the identications
H
p
i
=
H
p
i
+
H
z
A
z
A
p
i
=
i
(2.4.5a)
H
q
i
=
H
q
i
+
H
z
A
z
A
q
i
= p
k
k
q
i

L
q
i
(2.4.5b)
On account of these, equations (2.3.14a,b) gives rise to the following system of
ordinary dierential equations in normal form for the unknowns q
i
(t), p
i
(t)
dq
i
dt
=
H
p
i
(2.4.6a)
dp
i
dt
=
H
q
i
(2.4.6b)
The original constrained Lagrangian variational problem has thus been reduced
to a free Hamiltonian problem on the submanifold j : ((/), with Hamiltonian
H(t, q
1
, . . . , q
n
, p
1
, . . . , p
n
) identical to the pullback H = j
(H )
5
. Once again,
all this is in full agreement with Pontryagins principle.
Remark 2.4.1: By virtue of Cauchy theorem, equations (2.4.6a, b) require the assignment
of 2n initial data in order to give rise to a unique solution. This indicates that, as far as
the calculus of variations is concerned, a xed endpoints problem is always wellposed,
regardless of its holonomic or nonholonomic nature. In the latter case, however, it is easily
seen that the contemporaneous knowledge of both the initial position and velocity of the
5
Conversely, setting H = j
(H), the inverse Legendre transformation q

i
=
H
p
i
, together with
equation (2.4.5a), yields back the constraint equations q
i
=
i
(t, q
k
, z
A
).
system (which corresponds to the assignment of n + r data) is not enough to determine
its future evolution. In this sense, it becomes apparent that the constrained calculus of
variations cant be considered as a branch of Mechanics for the principle of determinism
is not fullled.
All the same, the subject of the present work is commonly classied in the literature
as vakonomic Mechanics. This terminology, rst introduced by Arnold [2], is an abbre-
viation for Mechanics of variational axiomatic kind and is often thought as a sort of
nonholonomic Mechanics that diers from the latter inasmuch as the constraints are a
priori given. It is in our opinion that, although largely diused, such a terminology is
most inappropriate and misleading for, by denition, any theory which aspire to belong
under Mechanics must be deterministic.
A continuous extremal of the functional (2.3.3) consisting of a nite family
of closed arcs
(s)
: [a
s1
, a
s
] ((/), each contained in (a connected component
of) the submanifold will be called a regular extremal .
Singular extremals, partly, or even totally lying outside may also exist. In
fact, while equation (2.4.2a) is part of the system (2.3.14), and must therefore
be satised by any extremal, the requirement (2.4.2b) has only to do with the
wellposedness of the Cauchy problem for the subsystem (2.3.14).
On the other hand, by construction, the Hamilton equations (2.4.6a,b) deter-
mines only the regular extremals. The singular ones, if at all, have therefore to be
dealt with directly, looking for solutions of equations (2.3.14) not arising from a
well posed Cauchy problem.
In principle, this could be done extending to the nonholonomic context the
concepts and methods commonly adopted in the study of singular Lagrangians
[26]. The argument is beyond the purposes of the present work, and will not be
pursued.
To complete our analysis, let us nally discuss the role of equations (2.3.15) in
the study of corners. To this end, we consider the rear face of diagram (1.4.28),
now suitably pulled-back onto /:
o
A
H(1
n+1
)
_
((/) V
(1
n+1
)
(2.4.7)
and we recall that, for each choice of the trivialization u: P R, the Liouville
1form (1.4.20) of j
1
(P, 1
n+1
) provides the space H(1
n+1
) with the form
u
= p
0
dt + p
i
dq
i
(2.4.8)
By making use of it, as well as of the PontryaginPoicareCartan form (2.2.7), we
now introduce a morphism ((/)

H(1
n+1
) bred over V
(1
n+1
) and based on
the prescription
u
) =
PPC
2.4 Hamiltonian formulation 57
In coordinates, we have the explicit representation
: t = t, q
i
= q
i
, p
i
= p
i
, p
0
= H
_
t, q
i
, p
i
, z
A
_
(2.4.9)
The content of equations (2.3.4d) is then summarized into the following
Proposition 2.4.1. For any continuous extremal
_
, [t
0
, t
1
]
_
of the functional
(2.3.3), the composite map : [t
0
, t
1
] H(1
n+1
) is necessarily continuous.
The previous arguments provide a simple characterization of the jumps that
may possibly occur along a regular extremal :
__

(s)
, [a
s1
, a
s
]
__
. To this end
we observe that the restriction of the map (2.4.9) to the submanifold ((/)
determines an immersion : H(1
n+1
) and that, as already pointed out, at
each there exists, locally, one and only one dierentiable extremal of the
functional (2.3.3) through .
On the other hand, by Proposition 2.4.1, for each s = 1, . . . , N 1, the arcs

(s)
and
(s + 1)
are related by the condition
_

(s)
(a
s
)
_
=
_

(s + 1)
(a
s
)
_
. From
this, it is readily seen that the admissible discontinuities of or, all the same, the
admissible corners in the projection := : [t
0
, t
1
] 1
n+1
may only occur at
points in which the immersion : H(1
n+1
) is not injective.
Chapter 3
The second variation
The object of the present Chapter is to establish whether a given locally normal
extremal gives rise to a local minimum for the functional (2.1.1). As the totality of
these extremals has already been characterized, we will now address ourselves to
the analysis of the second derivative
d
2
I
d
2
=0
, commonly referred to as the second
variation of the action functional at .
In local coordinates, a simple calculation yields the result
d
2
1[
]
d
2
=0
=
N
s=1
_
_
as
a
s1
_
_

2
L
q
i
q
j
_

(s)
X
i
(s)
X
j
(s)
+ 2
_

2
L
q
i
z
A
_

(s)
X
i
(s)
A
(s)
+
+
_

2
L
z
A
z
B
_

(s)
A
(s)
B
(s)
+
_
L
q
i
_

(s)
Z
i
(s)
+
_
L
z
A
_

(s)
K
A
(s)
_
dt+
+ 2
s
_
L
q
i
X
i
(s)
+
L
z
A

A
(s)
_

(s)
(as)
+
2
s
_
dL
dt
_

(s)
(as)
+
2
s1
_
L
q
i
X
i
(s)
+
L
z
A

A
(s)
_

(s)
(a
s1
)

2
s1
_
dL
dt
_

(s)
(a
s1
)
+
+
s
L(
(s)
(a
s
))
s1
L(
(s)
(a
s1
))
_
(3.0.1)
Besides the serious diculties which lie in the determination of its denite-
ness, the previous expression hasnt apparently a tensorial character because of
the contemporaneous presence of both the rst and the second derivatives of the
Lagrangian, which entails these last to undergo a transformation law like the fol-
60 Chapter 3. The second variation
lowing one
2
L
q
i
q
j
=

2
L
q
k
q
r
q
k
q
i
q
r
q
j
+ 2

2
L
q
k
z
A
q
k
q
i
z
A
q
j
+

2
L
z
A
z
B
z
A
q
i
z
B
q
j
+
+
L
q
k
2
q
k
q
i
q
j
+
L
z
A
2
z
A
q
i
q
j
This, of course, makes it unt to be dealt with in a geometrical framework.
Therefore, before getting to the heart of the matter, we ought to take the necessary
steps in order to guarantee the tensorial character of all results.
3.1 Adapted Lagrangians
Generally speaking, a function f on a dierentiable manifold M is said to be
critical at a point x M if and only if its dierential vanishes at x.
Furthermore, the Hessian of f at a critical point x is a symmetric bilinear
functional (d
2
f)
x
: T
x
(M) T
x
(M) R which is dened by the following con-
struction: for any X, Y T
x
(M), denoting by

X,

Y their respective extensions to
vector elds, we let
(d
2
f)
x
, X Y
_
:=

X
x
(

Y (f) ), where

X
x
is of course just
X. Its symmetry is a direct consequence of f being critical at x, as we can readily
see from the relation
X
x
(

Y (f) )

Y
x
(

X(f) ) = [

X ,

Y ]
x
(f) = L
X
(
Y )
| x
(f) = 0
It is also clearly welldened inasmuch as

X
x
(
Y (f)) = X(
Y (f)) is independent
of the extension

X of X, while

Y
x
(

X(f)) is independent of

Y .
If the manifold M is referred to a local coordinate system x
1
, . . . , x
n
and
X = X
i
_

x
i
_
x
, Y = Y
i
_

x
i
_
x
, we can set

X = X
i
_

x
i
_
, with X
i
= const.
Then
(d
2
f)
x
, X Y
_
= X(
Y (f)) = X
_
Y
j
f
x
j
_
= X
i
Y
j
_

2
f
x
i
x
j
_
x
so we have the representation
(d
2
f)
x
=
_

2
f
x
i
x
j
_
x
(dx
i
)
x
(dx
j
)
x
Under the stated circumstance, the Hessian of f at x has therefore a tensorial
character. Similar conclusions hold if f is critical at each point of a submanifold
N M, in which case we write (df)
N
= 0 and denote by (d
2
f)
N
the Hessian of f
along N. Given any function f = f(t, q
i
) F(1
n+1
) we have then the following
properties:
3.1 Adapted Lagrangians 61
i) if f is critical on an admissible section : R 1
n+1
, its symbolic time
derivative

f :=
f
t
+
f
q
k

k
F(/) is itself critical on the lift of , and
satises

f
|
= 0;
ii) under the same assumption, for any admissible deformation X : R V ()
the quadratic form associated
1
to the Hessian (d
2
f)
fulls the relation

d
dt
(d
2
f)
, X X
_
=
(d
2

f)

,

X

X
_
(3.1.1)
Remark 3.1.1: Both properties may be easily veried by observing that the condition
(df)
= 0 implies the identities

_
d

f
_

=
_

f
t
dt +

f
q
k
dq
k
+

f
z
A
dz
A
_

=
d
dt
_
f
t
_
dt +
d
dt
_
f
q
k
_
dq
k
= 0
_

2
f
q
i
q
j
_

=
_

3
f
q
i
q
j
t
+

3
f
q
i
q
j
q
k

k
+

2
f
q
i
q
k
k
q
j
+

2
f
q
j
q
k
k
q
i
_

=
=
d
dt
_

2
f
q
i
q
j
_
+
_

2
f
q
i
q
k
_
k
q
j
_

+
_

2
f
q
j
q
k
_
k
q
i
_

_

2
f
q
i
z
A
_

=
_

2
f
q
i
q
k
_
k
z
A
_

,
_

2
f
z
A
z
B
_

= 0
The conclusion then follows by direct computation, expressing the derivatives
dX
i
dt
in terms
of the components X
i
,
A
through the variational equation (1.5.8).
The previous arguments may avail in our variational context. In this respect,
we recall the following results from the previous Chapters:
as far as the ordinary evolutions are concerned, the variational problem in
the event space based on the functional (2.1.1) is equivalent to the (free) one
in the contact manifold based on the functional (2.3.3);
for each normal extremal =
__
(s)
, [a
s1
, a
s
]
_
, s = 1, . . . , N,
_
of the
action integral (2.1.1) there exists a unique extremal : [t
0
, t
1
] ((/) of
the functional (2.3.3) projecting onto , i.e. satisfying = , whence also
= = ;
in coordinates, setting
(s)
: q
i
= q
i
(s)
(t), z
A
= z
A
(s)
(t), p
i
= p
(s)
i
(t), the
algorithm for the determination of relies both on Pontryagins equations
dq
i
(s)
dt
=
i
(t, q
i
(s)
, z
A
(s)
) ,
dp
(s)
i
dt
+ p
(s)
k
k
q
i
=
L
q
i
, p
(s)
k
k
z
A
=
L
z
A
1
See Appendix D.
and on ErdmannWeierstrass matching conditions
q
i
(s + 1)
(a
s
) = q
i
(s)
(a
s
) , p
(s + 1)
i
(a
s
) = p
(s)
i
(a
s
) , (H )

(s + 1) (a
s
) = (H )

(s) (a
s
)
under an arbitrary change of the trivialization u of the bundle P into
u
= uf(t, q
1
, . . . , q
n
), the PontryaginPoincareCartan form (2.2.7) obeys
the transformation law
PPC

PPC
=
_
L(t, q
i
, z
A
)

f
_
dt +
_
p
i
f
q
i
_

i
=
PPC
df
the extremals of the functional
_

PPC
dier from those of
_

PPC
by a
translation p
i
(t) p
i
(t) = p
i
(t)
f(t,q
i
(t))
t
along the bres of ((/)

/;
as it was to be expected on account of the gauge invariance of the projections
= and = , the corresponding action integrals
_

L
dt and
_

Ldt have actually the same extremals with respect to xed endpoints
deformations; in particular, every extremal yielding a minimum for the
rst integral, does the same for the second one and conversely.
The idea is now to take advantage of the gauge structure of the theory so as
to make every point of the section into a critical point for the Lagrangian.
However, in pursuing this strategy, we should not overlook we are extending
the class of admissible sections to piecewise dierentiable ones. Furthermore, as
far as these are concerned, our denition of deformation of an admissible evolution
of the system explicitly includes possible variations of the reference intervals.
Whenever both of the previous circumstances occur, the intention of replacing
the original Lagrangian by a gauge equivalent and critical one, becomes extremely
awkward. This is because, in order to achieve its goal, the function f F(1
n+1
)
which takes part in the gauge transformation u u f(t, q
1
, . . . , q
n
), should
be tailored along the section and, therefore, with respect to the intervals
[a
s1
, a
s
]. On the other hand, the evaluation of the second variation of the action
integral passes through integrations on the dierent intervals [a
s1
() , a
s
()]. In
this connection, it is even thinkable an extreme case in which, as varies, the
value t = a
s
() swings between the intervals [a
s1
, a
s
] and [a
s
, a
s+1
].
Remark 3.1.2: These kind of troubles instantly vanish whenever at most only one of
the abovenamed circumstances occurs, namely every time we happen to be in one of the
following particular situations:
a) section is dierentiable and so is
for any ;
b) section is dierentiable while
is just piecewisedierentiable for any ; time

intervals [a
s1
() , a
s
()] may be modied by the deformation process
2
;
2
The reader is referred to Appendix C for the proof of the actual existence of this kind of
deformations.
c) section is piecewisedierentiable and so is
for any ; time intervals [a

s1
, a
s
]
remain unchanged during the deformation process.
Whenever b) occurs, the function f is well-dened and dierentiable along the entire
interval [t
0
, t
1
] and, as such, may be easily restricted to any interval [a
s1
(), a
s
()], no
matter how the values a
s1
, a
s
vary with . On the other hand, in the circumstance c),
the tailoring on the function f along the section holds good along every deformation
. Needless to say, situation a) is the easiest one, as it combines all the simplications
brought by b) and c).
Remark 3.1.3: A further pleasantness regarding the particular circumstances described
in the previous Remark lies in the fact that, in all cases a), b), and c), the expression of
the second variation turns out to be quite simplied. In order to see this, taking equations
(1.5.6b), (1.5.34c) into account, we rst rewrite relation (3.0.1) more suitably as
d
2
1[
]
d
2
=0
=
N
s=1
_
as
as1
_
_

2
H
q
i
q
j
_

(s)
X
i
(s)
X
j
(s)
+ 2
_

2
H
q
i
z
A
_

(s)
X
i
(s)
A
(s)
+
+
_

2
H
z
A
z
B
_

(s)
A
(s)
B
(s)
+
_
H
q
i
_

(s)
Z
i
(s)
+
_
H
z
A
_

(s)
K
A
(s)
_
dt +
+
N1
s=1
_
2
s
_
dH
dt
+
dp
i
dt

i
_
as
2
s
_
X
i
+
s
i
_
as
_
dp
i
dt
_
as
_
(3.1.2)
where, as usual,
_
g
as
stands for the jump of the function g at the corner c
s
. It is now
readily seen that both in the situation b), in which
dH
dt
,
dpi
dt
and
i
dont jump at any of
the points (a
s
) , and in the situation c), in which
s
= 0 for any s, the above expression
reduces to the only integral term.
In order to cope with these intricacies, we will try a slightly dierent approach,
in line with the nature of the evolution as a nite collection of admissible closed
arcs
(s)
, each viewed as the restriction to the closed interval [a
s1
, a
s
] of an
admissible section (still denoted by
(s)
) dened on some open neighborhood
(b
s1
, b
s
) [a
s1
, a
s
].
We begin by introducing a family (U
s
, h
s
), s = 1, . . . , N of local charts
in 1
n+1
such that each U
s
is an open neighborhood of the admissible section
(s)
: (b
s1
, b
s
) 1
n+1
. Then, careless about P being a trivial bundle, for any s
we make use of a dierentiable function f
(s)
: U
s
R to change, in each
1
(U
s
),
the global trivialization u into a local one u
(s)
= u f
(s)
.
As a consequence, the Lagrangian section (1.4.16) is now locally expressed as
u
(s)
= u

f
(s)
= L(t, q
i
(s)
, z
A
(s)
)

f
(s)
:= L
(s)
(t, q
i
(s)
, z
A
(s)
) (3.1.3)
and so it relies on the assignment of s dierent functions L
(s)
, each of them
dened over the open set
1
(U
s
), here denoting the projection /

1
n+1
.
Likewise, instead of a unique and globally dened PontryaginPoincareCartan
form (2.2.7), we have now a collection of local 1forms
(s)
PPC
whose representation
in coordinates reads
(s)
PPC
=
PPC
df
(s)
= L
(s)
dt +
_
p
(s)
i

f
(s)
q
i
_

i
(s)
(3.1.4)
The idea is to make good use of the above construction, simply by choosing
suitable functions f
(s)
. In this regard we state
Denition 3.1.1. Given a normal extremal , a function S
(s)
F(U
s
) is said
to be adapted to the section
(s)
if and only if it fulls the condition
3
(dS
(s)
)

(s) = (
(s)
PPC
)

(s) (3.1.5)
By a little abuse of language, whenever a function f
(s)
: U
s
R is adapted to
(s)
, the same terminology will be used to denote the corresponding Lagrangian
function L
(s)
which takes part in the representation (3.1.3).
Theorem 3.1.1. For any s = 1, . . . , N, there exists (at least) a dierentiable
function S
(s)
F(U
s
) adapted to the section
(s)
: (b
s1
, b
s
) 1
n+1
.
Proof. As it is showed in Appendix A, each arc
(s)
may be locally made into the
coordinate line q
i
(s)
(t, q
1
, . . . , q
n
) = 0, for instance by setting q
i
(s)
:= q
i
q
i
(s)
(t).
A possible local solution of equation (3.1.5) is now easily recognized to be
S
(s)
0
(t, q
i
) = p
(s)
k
(t) q
k
(s)
+
_
t
t
0
L
|
dt (3.1.6)
p
(s)
k
(t) being any functions satisfying p
(s)
k
(t)
q
k
(s)
q
i
(s)
(t)
= p
(s)
i
(t).
Then, as a direct consequence of the vanishing of q
i
(s)
along
(s)
, we have:
0 =
d
dt
_
q
i
(s) |
(s)
_
=
q
i
(s)
t
(s)
+
q
i
(s)
q
k
(s)
k
|
(s)
(3.1.7a)
S
(s)
0
q
i
(s)
= p
(s)
k
(t)
q
k
(s)
q
i
(s)
= p
(s)
i
(t) (3.1.7b)
S
(s)
0
t
(s)
= p
(s)
k
(t)
q
k
(s)
t
(s)
+ L
|
(s) = p
(s)
k
(t)
q
i
(s)
q
k
(s)
k
|
(s)
+ L
|
(s) =
=
_
L p
(s)
k
(t)
k
_

(s)
(3.1.7c)
3
As usual, we are not distinguish between functions on Vn+1 and their pullback on C(A).
By linearity, the most general solution of equation (3.1.5) has therefore the local
expression
S
(s)
= S
(s)
0
+ C
(s)
(t, q
i
) (3.1.8)
being C
(s)
(t, q
i
) any function satisfying (dC
(s)
)
(s) = 0. Henceforth, every trans-

formation S
(s)
S
(s)
+ C
(s)
will be called a restricted gauge transformation.
Theorem 3.1.2. Equation (3.1.5) implies the relations
L
(s) |
(s) = dL
(s) |
(s) = 0 (3.1.9)
Proof. Because of Theorem 3.1.1, we have only to check the validity of equations
(3.1.9) for the function (3.1.6). In view of equations (3.1.7b,c), denoting by

S
(s)
0
its symbolic time derivative, we have the evaluation
S
(s)
0 |
(s) =
S
(s)
0
t
(s)
+ p
(s)
k

k
|
(s) = L
|
(s) (3.1.10)
yielding back the relation L
(s) |
(s) =
_
L

S
(s)
0
_
|
(s)
= 0.
Moreover, we have the further relations

S
(s)
0
q
k

(s)
=

2
S
(s)
0
q
k
q
r

(s)
r
|
(s) +

2
S
(s)
0
q
k
t

(s)
+ p
(s)
r
r
q
k

(s)
=
=
dp
(s)
k
dt
+ p
(s)
r
r
q
k

(s)
=
L
q
k

(s)
(3.1.11a)

S
(s)
0
z
A

(s)
= p
(s)
k
k
z
A

(s)
=
L
z
A

(s)
(3.1.11b)
whence also

S
(s)
0
t

(s)
=
d

S
(s)
0 |
(s)
dt

S
(s)
0
q
k

(s)
k
|
(s)

S
(s)
0
z
A

(s)
dz
A
dt
=
=
dL
|
(s)
dt

L
q
k

(s)
k
|
(s)
L
z
A

(s)
dz
A
dt
=
L
t

(s)
(3.1.11c)
On account of Theorems 3.1.1, 3.1.2, the Hessian

_
d
2
L
(s)
_

(s)
provides a
tensorial eld along
(s)
for any s = 1, . . . , N. Its local representation is most easily
expressed in terms of nonholonomic bases. Denoting by the symmetrized tensor
product and setting
i
(s)
=
_
dq
i

i
dt
_

(s)
=
i
|
(s)
,
A
(s)
=
_
dz
A
dz
A
dt
dt
_

(s)
, a
straightforward calculation yields the result
_
d
2
L
(s)
_

(s)
=
_
2
L
(s)
q
r
q
k
_

(s)

r
(s)

k
(s)
+ 2
_
2
L
(s)
z
A
q
k
_

(s)

A
(s)

k
(s)
+
+
_
2
L
(s)
z
A
z
B
_

(s)

A
(s)

B
(s)
(3.1.12)
Remark 3.1.4: The components G
(s)
AB
:=
_
2
L
(s)
z
A
z
B
_

(s)
are invariant under arbitrary
restricted gauge transformations and may therefore be evaluated arbitrarily choosing S
(s)
within the class of solutions of equation (3.1.5). Making use of the ansatz (3.1.6), we
obtain the representation
G
(s)
AB
=
_
2
(L

S
(s)
)
z
A
z
B
_

(s)
=
_

2
L
z
A
z
B
_

(s)
p
(s)
i
(t)
_

2
i
z
A
z
B
_

(s)
or equivalently
G
(s)
AB
=
_

2
K
(s)
z
A
z
B
_

(s)
(3.1.13)
with K
(s)
:= p
(s)
i
(t)
i
(t, q
i
, z
A
) L(t, q
i
, z
A
), henceforth referred to as the restricted
Pontryagin Hamiltonian.
In view of the identication
_
2
K
(s)
z
A
z
B
_

(s)
(t)
=
_
2
H
(s)
z
A
z
B
_

(s)
(t)
, the matrix (3.1.13) is
automatically non singular along any regular extremal.
Remark 3.1.5: Whenever det G
(s)
AB
,= 0, the Hessian (3.1.12) determines an innitesimal
control along
(s)
, namely a linear section h
(s)
: V (
(s)
) A(
(s)
), uniquely dened by
the condition
_
_
d
2
L
(s)
_

(s)
, h
(s)
(X
(s)
) Y
(s)
_
= 0 X
(s)
V (
(s)
) , Y
(s)
V (
(s)
) (3.1.14a)
In view of equations (1.5.13), (3.1.12), the requirement (3.1.14a) is locally expressed the
relations
_
_
d
2
L
(s)
_

(s)
,

_

z
A
_

(s)
_
=
_
2
L
(s)
q
i
z
A
_

(s)
+ G
(s)
AB
h
i
(s)
B
= 0 (3.1.14b)
Under the assumption det G
(s)
AB
,= 0, these may be solved for the components h
i
(s)
B
, thereby
providing the representation
h
i
(s)A
= G
AB
(s)
_
2
L
(s)
q
i
z
B
_

(s)
whence also
i
:= h
(s)
_ _

q
i
_
(s)
_
=
_

q
i
_

(s)
G
AB
(s)
_
2
L
(s)
q
i
z
B
_

(s)
_

z
A
_

(s)
(3.1.15)
with G
(s)
AB
G
BC
(s)
=
C
A
.
The absolute time derivative along
(s)
induced by h
(s)
will be denoted by
_
D
Dt
_
(s)
.
The expression (1.5.21b) for the temporal connection coecients takes now the form
i
j
=

j
=
_
j
q
i
_

(s)
+ G
AB
(s)
_
j
z
A
_

(s)
_
2
L
(s)
q
i
z
B
_

(s)
(3.1.16)
Unlike the components G
(s)
AB
, the full Hessian (3.1.12) and therefore also the associated
innitesimal control and its corresponding time derivative, are not gauge invariant, but
explicitly depend on the particular choice of the Lagrangian L
(s)
.
In view of Erdmann-Weierstrass conditions (2.3.15), the following identity is
easily seen to hold at each corner c
s
S
(s)
q
i
cs
=
S
(s + 1)
q
i
cs
,
S
(s)
t
cs
=
S
(s + 1)
t
cs
= d
_
S
(s + 1)
S
(s)
_
cs
= 0
and so the Hessian of the dierence S
(s + 1)
S
(s)
, evaluated at the point
c
s
=
_
a
s
,
(s)
(a
s
)
_
, is itself a tensor, hereby denoted by
_
d
2
S
cs
.
We now introduce the quantity
s
() :=
_
S
(s + 1)
S
(s)
_
cs()
=
= S
(s + 1)
_
a
s
(),
i
(s + 1)
_
a
s
(),
_
_
S
(s)
_
a
s
(),
i
(s)
_
a
s
(),
_
_
(3.1.17)
and, in view of (1.5.32) and (1.5.33b), we point up the relation
d
2
s
()
d
2
=0
=
2
s
2
_
S
(s + 1)
S
(s)
_
t
2
cs
+ 2
s
_
i
+ X
i
_
cs
2
_
S
(s + 1)
S
(s)
_
t q
i
cs
+
+
_
i
+ X
i
_
cs
_
j
+ X
j
_
cs
2
_
S
(s + 1)
S
(s)
_
q
i
q
j
cs
(3.1.18)
written more suitably as
d
2
s
()
d
2
=0
=
_
_
d
2
S
cs
, W
s
W
s
_
(3.1.19)
From this, collecting all the previous results, we get the following identity
N
s=1
_
as()
a
s1
()
L
(s)|
dt
N1
s=1
s
() =
_

L dt
N
s=1
_
as()
a
s1
()
S
(s)
|
dt +
N1
s=1
_
S
(s + 1)
S
(s)
_
cs()
=
_

L dt S
(N)
(t
1
) + S
(0)
(t
0
) =
=
_

L dt
_

L dt
(3.1.20)
For later use, before moving on to the analysis of the second variation of the
action functional, we observe that, whenever is a normal extremal, the use of an
adapted Lagrangian provides a canonical splitting of T(((/)).
As already remarked, if the trivialization u of P is changed in each
1
(U
s
) into
the local one u
(s)
= uf
(s)
(t, q
1
, . . . , q
n
), then the extremals of the corresponding
functional
N
s=1
_

(s)

(s)
PPC
dier locally from those of
_

PPC
by a translation
p
i
(t) p
(s)
i
(t) = p
i
(t)
f
(s)
(t,q
i
(t))
q
i
along the bres of ((/). In particular:
if f
(s)
is adapted to
(s)
, the local representation of
(s)
satises the condition
p
(s)
i
(t) 0;
because of the condition (dC
(s)
)
(s) = 0, the lift is evidently invariant

under restricted gauge transformations.
Moreover, in view of the nature of ((/)

/ as a vector bundle over /,
each local section O
(s)
:
1
(U
s
) / ((/) given by p
(s)
i
(t, q
i
, z
A
) = 0 has an
invariant geometrical meaning.
For each
(s)
O
(s)
(
1
(U
s
)), every vector

Z
(s)
T
(s)
(((/)) may therefore
be split into a horizontal part

Z
h
(s)
O
(s)
Z
(s)
) tangent to the submanifold
O
(s)
(
1
(U
s
)), and a vertical part

Z
v
(s)
, tangent to the ber
1
( (
(s)
)) at
(s)
.
In coordinates, we have the explicit representation
Z
v
(s)
=
_

Z, (dp
(s)
i
)
(s)
_
_

p
i
_
(s)
,

Z
h
(s)
=

Z
(s)

Z
v
(s)
The previous algorithm can be get interacted with another intrinsic attribute
of ((/), represented by the Liouville 1form (1.2.17). In this way, to each vector
Z
(s)
T
(s)
(((/)) we may associate a 1form in T
(s)
(((/)) according to the
prescription
Z
(s)

Z
v
(s)

Z
v
(s)
_
d
L
_
(s)
=
_

Z
(s)
, (dp
(s)
i
)
(s)
_
_

p
i
d
L
_
(s)
=
=
_

Z
(s)
, (dp
(s)
i
)
(s)
_

i
(s)
(3.1.21)
At last, we observe that every element of the cotangent space T
(s)
(((/))
generated by the above process may be uniquely expressed as the pullback of a
1form
(s)
T
(
(s)
)
(1
n+1
).
Collecting all results and recalling the denition of virtual 1forms introduced
in 1.3, we conclude that the lift algorithm determines a bijective corre-
spondence between vector elds

Z
(s)
along each
(s)
and pairs (

Z
(s)
,
(s)
), in which
Z
(s)
=
Z
(s)
) is a vector eld along
(s)
=
(s)
, while
(s)
=

Z
(s)
, (dp
i
)

(s)
_

i
3.2 The second variation of the action functional 69
is a virtual 1form along
(s)
. This will play a crucial role in the subsequent dis-
cussion.
3.2 The second variation of the action functional
Let : [t
0
, t
1
] 1
n+1
be a normal (not necessarily regular) extremal of the action
functional (2.1.1). In view of the identity (3.1.20), the analysis of the second
variation of 1[] may be better carried out by evaluating the second derivative
d
2
1[
]
d
2
=0
=
d
2
d
2
_
N
s=1
_
as()
a
s1
()
L
(s)|
dt
N1
s=1
s
()
_
=0
In this connection, being each L
(s)
adapted to the corresponding arc
(s)
, a
simple calculation yields the result
d
2
d
2
_
_
as()
a
s1
()
L
(s)|
dt
_
=0
=
_
as
a
s1
_
_

2
L
q
i
q
j
_

(s)
X
i
(s)
X
j
(s)
+
+ 2
_

2
L
q
i
z
A
_

(s)
X
i
(s)
A
(s)
+
_

2
L
z
A
z
B
_

(s)
A
(s)
B
(s)
_
=
=
_
as
a
s1
_
_
d
2
L
(s)
_

(s)
,

X
(s)

X
(s)
_
dt
(3.2.1)
which, together with equation (3.1.18), provides the nal (plainly covariant) ex-
pression
d
2
1[
]
d
2
=0
=
N
s=1
_
as
a
s1
_
_
d
2
L
(s)
_

(s)
,

X
(s)

X
(s)
_
dt +
N1
s=1
_
_
d
2
S
cs
, W
s
W
s
_
(3.2.2)
Remark 3.2.1: In view of equation (3.1.8), the Lagrangian L
(s)
is not unique, but is
dened up to a restricted gauge transformation L
(s)
L
(s)

C
(s)
, with (dC
(s)
)
(s) = 0.
Therefore, as an internal consistency check, we ought to prove that the expression (3.2.2)
does not depend on any specic choice of the functions S
(s)
(t, q
i
).
We start by noticing the following identities, which are a straightforward consequence
of the condition (dC
(s)
)
(s)
= 0:
0 =
d
dt
_
C
(s)
q
i
_
(s)
=
_
2
C
(s)
q
i
t
_
(s)
+
_
2
C
(s)
q
i
q
j
_
(s)
j
|
(s)
0 =
d
dt
_
C
(s)
t
_
(s)
=
_
2
C
(s)
t
2
_
(s)
+
_
2
C
(s)
tq
j
_
(s)
j
|
(s)
=
=
_
2
C
(s)
t
2
_
(s)
2
C
(s)
q
i
q
j
_
(s)
i
|
(s)

j
|
(s)
In view of these, denoting by
_
d
2
C
cs
the tensor provided at the point c
s
by the Hessian
of the dierence C
(s + 1)
C
(s)
, we now evaluate
_
_
d
2
C
cs
, W
s
W
s
_
=
2
s
_
2
C
t
2
_
cs
+ 2
s
_
i
+ X
i
_
cs
_

2
C
t q
i
_
cs
+
+
_
i
+ X
i
_
cs
_
j
+ X
j
_
cs
_

2
C
q
i
q
j
_
cs
=
=
2
s
_

2
C
q
i
q
j

i
j
_
cs
2
s
_
i
+ X
i
_
cs
_

2
C
q
i
q
j

j
_
cs
+
+
_
i
+ X
i
_
cs
_
j
+ X
j
_
cs
_

2
C
q
i
q
j
_
cs
but, by the jump relations (1.5.34a),
_

2
C
q
i
q
j

j
_
cs
_
i
+ X
i
_
cs
=
_

2
C
q
i
q
j
_
i
+ X
i
_
j
_
cs
=
=
_

2
C
q
i
q
j
X
i
j
_
cs
+
s
_

2
C
q
i
q
j

i
j
_
cs
and also
_

2
C
q
i
q
j
_
cs
_
i
+ X
i
_
cs
_
j
+ X
j
_
cs
=
_

2
C
q
i
q
j
_
i
+ X
i
__
j
+ X
j
_
_
cs
=
=
_

2
C
q
i
q
j
X
i
X
j
_
cs
+ 2
s
_

2
C
q
i
q
j
X
i
j
_
cs
+
2
s
_

2
C
q
i
q
j

i
j
_
cs
whence nally
N1
s=1
_
_
d
2
C
cs
, W
s
W
s
_
=
N1
s=1
_

2
C
q
i
q
j
X
i
X
j
_
cs
On the other hand, by equation (3.1.1), we have
N
s=1
_
as
as1
_
_
d
2

C
(s)
_

(s)
,

X
(s)

X
(s)
_
dt =
N
s=1
_
as
as1
d
dt
_
_
d
2
C
(s)
_
(s)
, X
(s)
X
(s)
_
dt =
=
N1
s=1
_

2
C
q
i
q
j
X
i
X
j
_
cs
3.3 The associated singlearc problem 71
and so we see that each single term of the righthand side of equation (3.2.2) actually
depends on how the function S
(s)
has been chosen, while the entire expression is (as
hoped) gaugeinvariant.
The problem of establishing whether a locally normal extremal constitutes a
minimum for the functional (2.1.1), now based on the analysis of the expression
(3.2.2), may be conveniently broken up into two consecutive logical steps:
i) rst of all, each single arc
_
(s)
, [a
s1
, a
s
]
_
is requested to give rise to a
minimum with respect to the special class of deformations which leave the
points
(s)
(a
s1
),
(s)
(a
s
) xed;
ii) afterwards, it still remains to gure out how to link up the previous results
in order to make them globally applicable to the entire evolution .
This way of going about the matter surely makes the treatment a little bit
longer than what it would be in case the problem is tackled as a whole at once.
However, in return, the discussion will turn out to be more clear as various dicul-
ties are faced separately. Moreover, the analysis of i), that will henceforth called
the associated singlearc problem, is evidently equivalent to the one that would be
drawn when dealing with the (not infrequent) situation
4
in which the section is
dierentiable as well as
for any .
3.3 The associated singlearc problem
From now on we shall thus momentarily focus our attention on a single specic
admissible closed arc
_
(s)
, [a
s1
, a
s
]
_
, which is supposed to represent a normal
extremal of the action functional
_

(s)
L dt. Collecting all the previous results,
we see that the analysis of its second variation involves uniquely the behavior of
the integral
d
2
1 [
(s)
]
d
2
=0
=
_
as
a
s1
_
_
d
2
L
(s)
_

(s)
,

X
(s)

X
(s)
_
dt (3.3.1)
In particular, when
(s)
is a regular extremal, introducing the horizontal basis
(3.1.15) associated with the hessian
_
d
2
L
(s)
_

(s)
and expressing

X
(s)
in compo-
nents as

X
(s)
= X
i
(s)
i
+Y
A
(s)
_

z
A
_

(s)
, equation (3.1.14b) provides the identica-
tion
_
d
2
L
(s)
_

(s)
,

X
(s)

X
(s)
_
= N
(s)
kr
X
k
(s)
X
r
(s)
+ G
(s)
AB
Y
A
(s)
Y
B
(s)
(3.3.2)
4
See Remark 3.1.2.
with
N
(s)
kr
:=
_
d
2
L
(s)
_

(s)
,

r
_
=
__
2
L
(s)
q
k
q
r
_
G
AB
(s)
_
2
L
(s)
q
k
z
A
__
2
L
(s)
q
r
z
B
__

(s)
(3.3.3)
As already pointed out, unlike the integral (3.3.1), the Hessian
_
d
2
L
(s)
_

(s)
is not a gaugeinvariant object. The eect of the restricted gauge group on the
representation (3.3.2) is therefore reected into the fact that the integrand at the
right-hand-side of equation (3.3.1) is dened up to an arbitrary transformation of
the form
_
_
d
2
L
(s)
_

(s)
,

X
(s)

X
(s)
_

__
d
2
_
L
(s)

C
(s)
_
_

(s)
,

X
(s)

X
(s)
_
=
=
_
d
2
L
(s)
_

(s)
,

X
(s)

X
(s)
_
d
dt
_
_
d
2
C
(s)
_
(s)
, X
(s)
X
(s)
_
=
=
_
N
(s)
ij

DC
(s)
ij
Dt
_
X
i
(s)
X
j
(s)
2 C
(s)
ij
_
i
z
A
_

(s)
X
j
(s)
Y
A
(s)
+ G
(s)
AB
Y
A
(s)
Y
B
(s)
(3.3.4)
where we have introduced the simplied notation C
(s)
ij
:=
_
2
C
(s)
q
i
q
j
_
(s)
and with
the components
DC
ij
Dt
expressed by equation (1.5.23) in terms of the ordinary
derivatives
dC
ij
dt
and of the temporal connection coecients
i
k
.
On this basis we state
Theorem 3.3.1. Let
(s)
: [a
s1
, a
s
] 1
n+1
be a normal extremal. Then, if
the matrix G
(s)
AB
(t) is non singular at a point t
(a
s1
, a
s
), there exist > 0
and a restricted gauge transformation L
(s)
L
(s)

C
(s)
such that the Hessian
_
d
2
(L
(s)

C
(s)
)
_

(s)
(t)
has algebraic rank equal to r for t (t
, t
+).
Proof. By continuity, there exists an interval [c, d ] t
where det G
(s)
AB
,= 0.
We focus on that interval, and apply equation (3.3.4) to the arc
(s)
_
[c, d ]
_
. Setting
Y
A
(s)
:= Y
A
(s)
G
AB
(s)
C
(s)
ir
_
r
z
B
_

(s)
X
i
(s)
and taking the symmetry of C
ij
into account, equation (3.3.4) may be rewritten
as
_
d
2
(L
(s)

C
(s)
)
_

(s)
,

X
(s)

X
(s)
_
=
=
_
N
(s)
ij

DC
ij
Dt
G
AB
(s)
_
r
z
A
_

(s)
_
l
z
B
_

(s)
C
(s)
ir
C
(s)
lj
_
X
i
(s)
X
j
(s)
+ G
(s)
AB
Y
A
(s)
Y
B
(s)
The thesis is therefore established as soon as we prove that the Riccatitype
dierential equation
DC
(s)
ij
Dt
+ G
AB
(s)
_
r
z
A
_

(s)
_
l
z
B
_

(s)
C
ir
C
lj
N
(s)
ij
= 0 (3.3.5)
admits at least one symmetric solution C
(s)
ij
= C
(s)
ij
(t) in a neighborhood of t = t
.
To this end, we set
M
rl
(s)
:= G
AB
(s)
_
r
z
A
_

(s)
_
l
z
B
_

(s)
(3.3.6)
and denote by C
(S)
ij
and C
(A)
ij
respectively the symmetric and antisymmetric part
of C
(s)
ij
. Due to the symmetry of M
ij
(s)
and N
(s)
ij
, equation (3.3.5) then splits into
the system
DC
(S)
ij
Dt
+ M
rl
(s)
_
C
(S)
ir
C
(S)
lj
+ C
(A)
ir
C
(A)
lj
_
N
(s)
ij
= 0 (3.3.7a)
DC
(A)
ij
Dt
+ M
rl
(s)
C
(S)
ir
C
(A)
lj
+ M
rl
(s)
C
(A)
ir
C
(S)
lj
= 0 (3.3.7b)
Being the second equation linear and homogeneous in C
(A)
ij
, by Cauchy theorem we
conclude that, if we choose C
(s)
ij
symmetric at t = t
, there exists > 0 such that

the solution of the Cauchy problem for equation (3.3.5) exists and is symmetric
for [t t
[ < .
In view of Theorem 3.3.1, whenever det G
(s)
AB
(t
) ,= 0, by a proper choice of the

gauge around the point (t
), the quadratic polynomial (3.3.4) can be reduced to

the canonical form
_
_
d
2
(L
(s)

C
(s)
)
_

(s)
,

X
(s)

X
(s)
_
= G
(s)
AB
Y
A
(s)
Y
B
(s)
=
_

2
K
(s)
z
A
z
B
_

(s)
Y
A
(s)
Y
B
(s)
(3.3.8)
in a neighborhood of t
, K
(s)
denoting the restricted Pontryagin Hamiltonian.
Unfortunately, the purely local validity of equation (3.3.8) makes it unsuited
to the study of the second variation (3.3.1), which involves an integration over
the whole interval [a
s1
, a
s
]. We shall return on this point later. At present, we
shall concentrate on the role of Theorem 3.3.1 in the identication of sucient
conditions for a regular extremal to yield a (relative) minimum for the action
functional. In this connection, a preliminary result is provided by the following
Corollary 3.3.1.1. Under the same assumptions as in Theorem 3.3.1, given
any vertical vector eld V
(s)
= V
A
(s)
_

z
A
_

(s)
along
(s)
with compact support
[a, b] (t
, t
+), there exist a dierentiable function g = g(t) not identically

zero on [a, b] and an innitesimal deformation

X
(s)
= X
i
(s)
_

q
i
_

(s)
+Y
A
(s)
_

z
A
_

(s)
with support contained in [a, b] satisfying the relation
Y
A
(s)
G
AB
(s)
C
(s)
rs
_
s
z
B
_

(s)
X
r
(s)
= gV
A
(s)
Proof. Using the variational equation in the form (1.5.25a), the required condi-
tions are summarized into the pair of relations
DX
i
(s)
Dt
=
_
i
z
A
_

(s)
_
gV
A
(s)
+ G
AB
(s)
C
(s)
rl
_
l
z
B
_

(s)
X
r
(s)
_
(3.3.9a)
X
i
(s)
(a) = X
i
(s)
(b) = 0 (3.3.9b)
For any choice of g(t), equation (3.3.9) is a rst order linear dierential equation
for the unknowns X
i
(s)
(t); integrating it with initial data X
i
(s)
(a) = 0 yields the
solution
X
i
(s)
(t) = W
i
k
(t)
_
t
a
(W
1
)
k
r
_
r
z
A
_

(s)
gV
A
(s)
d
W
i
k
being the Wronskian of the equation. In order to ensure X
i
(s)
(b) = 0 it is
then sucient to choose g(t) within the (innitedimensional) vector space of
dierentiable functions over (t
, t
+) satisfying the conditions

_
b
a
(W
1
)
k
r
_
r
z
A
_

(s)
gV
A
(s)
d = 0 , k = 1 . . . n
Corollary 3.3.1.2. The positive semideniteness of the matrix G

(s)
AB
(t) at all
t [a
s1
, a
s
] is a necessary condition for a normal extremal
(s)
: [a
s1
, a
s
] 1
n+1
to yield a minimum for the action functional.
Proof. Suppose that G
(s)
AB
is not positive semidenite at some t
[a
s1
, a
s
].
Depending on the value of det G
(s)
AB
(t
) we have then two possible alternatives:

i ) if det G
(s)
AB
(t
) ,= 0, on account of Theorem 3.3.1 there exist a restricted gauge

transformation L
(s)
L
(s)

C
(s)
such that a representation like (3.3.8) holds
in a neighborhood (t
, t
+).
Then, given any vertical vector eld V
(s)
with support contained in the interval
(t
, t
+ ) and satisfying G
(s)
AB
V
A
(s)
V
B
(s)
< 0 (for instance, the eigenvector
corresponding to the negative eigenvalue of G
(s)
AB
in (t
, t
+), multiplied by
a suitable function with compact support), Corollary 3.3.1.1 implies the existence
of at least one innitesimal deformation

X
(s)
satisfying
d
2
1
d
2
=0
=
_
as
a
s1
_
d
2
(L
(s)

C
(s)
)
_

(s)
,

X
(s)

X
(s)
_
dt =
_
b
a
g
2
G
(s)
AB
V
A
(s)
V
B
(s)
dt < 0
Therefore, does not provide a minimum for the action functional.
ii) if det G
(s)
AB
(t
) = 0, choose > 0 in such a way that

is not a root of the secular equation det(G
(s)
AB

AB
) = 0;
at least one root of the secular equation is smaller than .
Let M F(/) be a dierentiable function globally dened on / and having
local expression
5
M =
AB
(z
A
z
A
(t))(z
B
z
B
(t)) in a neighborhood U of the
point
(s)
(t
). Also, let [c, d ] t
be a closed interval, satisfying

(s)
([c, d ]) U.
Setting L
(s)
:= L
(s)
+M, one can then easily verify the properties:
a) the section
(s)
: [c, d ] 1
n+1
is a normal extremal for the action integral
_

(s)
L
(s)
dt;
b) the matrix
_
2
L
(s)
z
A
z
B
_

(s)
(t
)
= G
(s)
AB
+
AB
is both non singular and non
positive (semi)denite.
In view of a) and b), the analysis developed in point i) ensures the existence of
at least one innitesimal deformation

X
(s)
having support in [a, b] [c, d ] and
satisfying
_
d
c
(d
2
L
(s)
)

(s) ,

X
(s)

X
(s)
_
dt < 0. On the other hand, by construc-
tion, this implies also
_
d
c
_
_
d
2
L
(s)
_

(s)
,

X
(s)

X
(s)
_
dt =
=
_
d
c
_
_
d
2
L
(s)
_

(s)
,

X
(s)

X
(s)
_
dt
_
d
c
AB
_
dz
A
,

X
(s)
__
dz
B
,

X
(s)
_
dt
_
d
c
_
_
d
2
L
(s)
_

(s)
,

X
(s)

X
(s)
_
dt < 0
once again proving that does not yield a minimum for the action functional.
3.3.1 The matrix Riccati equation and the sucient conditions
From now on we shall concentrate on the class of regular normal extremals. The
role of regularity in the solution of the Pontryagin equations (2.3.4) more specif-
ically, in the conversion of these into a system of ordinary dierential equations
in Hamiltonian forms for the unknowns q
i
(t), p
i
(t) should be well known from
2.4. However, when the problem is not nding the extremals, but working with
a given extremal
(s)
: [a
s1
, a
s
] 1
n+1
, regularity is merely an attribute of ,
5
As usual, we are writing z
A
(t) for z
A
( (t)).
ensuring the existence of an expression of the form (3.3.8) in a neighborhood of
each t
[a
s1
, a
s
].
On the other hand, as already pointed out, the purely local validity of equation
(3.3.8) is of little help in the evaluation of the second variation (3.3.1): it should
therefore be investigated to what extent equation (3.3.8) may be converted into
a global result, valid over the whole interval [a
s1
, a
s
]. On account of equations
(3.3.5), (3.3.6), this means analyzing the interval of existence of the solutions of
the Riccatilike dierential equation
6
DC
(s)
ij
Dt
+ M
rl
(s)
C
(s)
ir
C
(s)
lj
N
(s)
ij
= 0 (3.3.10)
The main diculty with the latter comes from its nonlinearity. To overcome
this aspect, we introduce two auxiliary virtual tensors E
(s)
ij
(t) and K
i
(s)
j
(t) along
(s)
, satisfying the transport laws
DK
i
(s)
j
Dt
= M
ir
(s)
E
(s)
rj
(3.3.11a)
DE
(s)
ij
Dt
= N
(s)
ir
K
r
(s)j
(3.3.11b)
On any interval (a, b) on which det K
i
(s)j
(t) ,= 0, the (generally non symmetric)
tensor
C
(s)
ij
= E
(s)
ir
_
K
1
(s)
_
r
j
(3.3.12)
is then well dened, and satises the relation
DE
(s)
ip
Dt
=
DC
(s)
ir
Dt
K
r
(s)
p
+ C
(s)
ir
DK
r
(s)p
Dt
Substituting from equations (3.3.11a), (3.3.12) and multiplying by (K
1
(s)
)
p
j
the
latter may be rewritten in the form
N
(s)
ij
=
DC
(s)
ij
Dt
+ C
(s)
ir
M
rl
(s)
E
(s)
lp
(K
1
(s)
)
p
j
=
DC
(s)
ij
Dt
+ C
(s)
ir
M
rl
(s)
C
(s)
lj
(3.3.10)
formally identical to equation (3.3.10).
Needless to say, the symmetry property C
(s)
ij
= C
(s)
ji
is also needed in order
for the tensor (3.3.12) to represent the Hessian of a function C
(s)
along
(s)
. An
argument similar to the one exploited in the proof of Theorem 3.3.1 shows that
this aspect relies entirely on the choice of the initial data. Indeed, on account
of equation (3.3.10), the antisymmetric part of C
(s)
ij
obeys a linear homogeneous
6
The regularity assumption is once again crucial in ensuring the global character of the absolute
time derivative
D
Dt
induced by the hessian (d
2
L
(s)
)

(s)
along
(s)
.
system of the form (3.3.7b). Once again, by Cauchy Theorem we conclude that
if C
(s)
ij
turns out to be symmetric say at t = a
s1
(as it happens e.g. choosing
E
(s)
ir
(a
s1
) = 0, K
r
(s)j
(a
s1
) =
r
j
), it will remain symmetric up to the rst value
t
> a
s1
(if any) at which det K
r
(s)j
(t
) = 0.
Remark 3.3.1: The analysis of equations (3.3.10), (3.3.11a, b) is considerably simplied
referring the virtual tensor algebra along
(s)
to an h
(s)
transported basis
_
e
(a)
, e
(a)
_
and recalling that, in this way, the components of the absolute time derivative of a eld
T coincide with the ordinary derivatives
dT
a
b
dt
. Equation (3.3.10) reduces then to the
ordinary matrix Riccati equation
dC
(s)
ab
dt
+ M
rs
(s)
C
(s)
ar
C
(s)
sb
N
(s)
ab
= 0 (3.3.13)
while equations (3.3.11a, b) take the simpler form
dK
a
(s)b
dt
= M
ac
(s)
E
(s)
cb
(3.3.14a)
dE
(s)
ab
dt
= N
(s)
ac
K
c
(s)
b
(3.3.14b)
Collecting all results, we can now state
Theorem 3.3.2 (Sucient conditions). Let
(s)
: [a
s1
, a
s
] 1
n+1
be a nor-
mal extremal for the action functional. Also, let H
(s)
= p
(s)
i

i
L denote the
Pontryagin Hamiltonian associated with the given Lagrangian. Then, if the matrix
G
(s)
AB
(t) :=
_

2
K
(s)
z
A
z
B
_

(s)
=
_

2
H
(s)
z
A
z
B
_

(s)
is positive denite at each t [a
s1
, a
s
] and if the system (3.3.11a, b) admits at
least one solution E
(s)
ij
(t), K
i
(s)
j
(t) satisfying the conditions
E
(s)
ir
_
K
1
(s)
_
r
j
symmetric,
det K
i
(s) j
,= 0 everywhere on [a
s1
, a
s
],
the section
(s)
yields a weak local minimum for the action functional.
Proof. The stated assumptions imply both the regularity of the extremal
(s)
and the existence of a global solution of the Riccatilike equation (3.3.10) along
(s)
, thus ensuring the validity of an expression like equation (3.3.8) on the whole
interval [a
s1
, a
s
]. So, if the matrix G
(s)
AB
is positive denite on [a
s1
, a
s
], this
provides the evaluation
d
2
1[
(s)
]
d
2
=0
=
_
as
a
s1
_
_
d
2
(L
(s)

C
(s)
)
_

(s)
,

X
(s)

X
(s)
_
dt =
=
_
as
a
s1
G
(s)
AB
Y
A
(s)
Y
B
(s)
dt > 0
for every nonzero admissible deformation

X : [a
s1
, a
s
] A( ) vanishing at the
endpoints.
A deeper insight into the meaning of the condition det K
i
(s) j
,= 0 is provided
by the study of the Jacobi vector elds, reviewed and adapted to the present
geometrical context.
3.3.2 Jacobi elds
Given a regular normal extremal
(s)
: [a
s1
, a
s
] 1
n+1
, we now consider the
(unique) extremal
(s)
: [a
s1
, a
s
] ((/) of the functional
_

(s)

(s)
PPC
projecting
onto
(s)
. Also, as usual, we preserve the notation = : ((/) 1
n+1
for
the composite bration ((/) / 1
n+1
.
Let us now introduce a special class of deformations
(s)
of
(s)
in which every
section
(s)
: [a
s1
, a
s
] ((/) is itself an extremal of
_

(s)

(s)
PPC
.
In this way, the 1parameter family
(s)
:=
(s)
is a deformation of the
original section
(s)
, consisting of extremals of the functional
_

(s)
L dt.
At this stage, we do not impose any restriction on the behavior of the end
points
(s)
(a
s1
),
(s)
(a
s
). In coordinates, setting

(s)
: q
i
=
i
(s)
(t, ) , z
A
=
A
(s)
(t, ) , p
i
=
(s)
i
(t, ) a
s1
() t a
s
()
(3.3.15)
our assumptions are summarized into the request that, for each value of , the
functions at the righthandside of equations (3.3.15) satisfy Pontryagins equa-
tions
i
(s)
t
=
i
(t,
i
(s)
,
A
(s)
) (3.3.16a)

(s)
i
t
+
(s)
k
_
k
q
i
_

(s)
=
_
L
(s)
q
i
_

(s)
(3.3.16b)

(s)
i
_
i
z
A
_

(s)
=
_
L
(s)
z
A
_

(s)
(3.3.16c)
As a check of inner consistency it is worth observing that, in view of the
condition (dL
(s)
)

(s) = 0, equations (3.3.16b, c) and the normality of
(s)
yield
back the relation
(s)
i
(t, 0) = 0.
Strictly associated with
(s)
is a corresponding innitesimal deformation, lo-

cally expressed as

X
(s)
= X
i
(s)
_

q
i
_

(s)
+
A
(s)
_

z
A
_

(s)
+
(s)
i
_

p
i
_

(s)
, with
X
i
(s)
=
_
i
(s)
_
=0
,
A
(s)
=
_
A
(s)
_
=0
,
(s)
i
=
_

(s)
i
_
=0
(3.3.17)
Taking equations (3.3.16) and the relation
(s)
i
(t, 0) = 0 into account, it is
easily seen that the components (3.3.17) satisfy the following system of dierential
algebraic equations
dX
i
(s)
dt
=
_
i
q
k
_

(s)
X
k
(s)
+
_
i
z
A
_

(s)
A
(s)
(3.3.18a)
d
(s)
i
dt
+
(s)
k
_
k
q
i
_

(s)
=
_
2
L
(s)
q
i
q
k
_

(s)
X
k
(s)
+
_
2
L
(s)
q
i
z
A
_

(s)
A
(s)
(3.3.18b)

(s)
i
_
i
z
A
_

(s)
=
_
2
L
(s)
z
A
q
k
_

(s)
X
k
(s)
+
_
2
L
(s)
z
A
z
B
_

(s)
B
(s)
(3.3.18c)
Given any vector eld

X
(s)
satisfying equations (3.3.17), its pushforward

X
(s)
will be called a Jacobi eld along
(s)
. By denition, a Jacobi eld
X = X
i
(s)
_

q
i
_
(s)
is therefore the innitesimal deformation tangent to a nite
deformation consisting of a 1parameter family of extremals of the action func-
tional.
Remark 3.3.2 (The accessory problem): The resemblance between equations (3.3.18)
and Pontryagins ones (2.3.4) sticks out a mile. This aspect can be made explicit by
replacing the imbedding (1.2.3) by its linearized counterpart (1.3.12) , namely regarding
the vector bundle V (
(s)
) as the conguration spacetime of an abstract system B
, and
the bundle A(
(s)
) V (
(s)
) as the associated space of admissible velocities. In this way,
the admissible evolutions of B
are in 1-1 correspondence with the admissible innitesimal

deformations of
(s)
.
Referring V (
(s)
) and A(
(s)
) to coordinates t, v
i
and t, v
i
, w
A
respectively, according
to the prescriptions (1.3.1) and (1.3.11), the imbedding i
: A(
(s)
) j
1
(V (
(s)
) is locally
expressed by
v
i
=
_
i
q
k
_

(s)
v
k
+
_
i
z
A
_

(s)
w
A
:=
i
(t, v
i
, w
A
) (3.3.19)
To complete the picture, we adopt the quadratic form
L
(s)
(

X
(s)
) :=
1
2
_
_
d
2
L
(s)
_

(s)
,

X
(s)

X
(s)
_
(3.3.20)
as a Lagrangian on A(
(s)
), whose representation in coordinates reads
L
(s)
(t, v
i
, w
A
) =
1
2
__
2
L
(s)
q
i
q
j
_

(s)
v
i
v
j
+2
_
2
L
(s)
q
i
z
A
_

(s)
v
i
w
A
+
_
2
L
(s)
z
A
z
B
_

(s)
w
A
w
B
_
and denote by I the functional assigning the action integral I [X
(s)
] :=
_
X
(s)
L
(s)
dt to
each admissible section X
(s)
: [a
s1
, a
s
] V (
(s)
). In this way, for any nite deformation
(s)
of
(s)
tangent to X
(s)
, equation (3.2.1) provides the identication
I [X
(s)
] =
1
2
d
2
1[
(s)
]
d
2
=0
It may be now easily veried that the equations (3.3.18) involved in the denition of
the Jacobi elds, now more suitably rewritten as
dX
i
(s)
dt
= X
k
(s)
k
v
i
+
A
(s)
k
w
A
d
(s)
i
dt
+
(s)
k
k
v
i
=
L
(s)
v
i

(s)
i
i
w
A
=
L
(s)
w
A
are formally identical to the Pontryagins equations for the determination of the extremals
v
i
= X
i
(s)
(t), w
A
=
A
(s)
(t) of the functional I subject to the constraints (3.3.19), which
is commonly referred to as the accessory variational problem.
Coming back to the system (3.3.18) and recalling the discussion at the end of
3.1, we decompose the eld

X
(s)
into the pair
X
(s)
= X
i
(s)
_

q
i
_

(s)
+
A
(s)
_

z
A
_

(s)
,
(s)
=
(s)
i

i
|
(s)
in which

X
(s)
is a vector eld along
(s)
while
(s)
is a virtual 1form along
(s)
. By a little abuse of language, this will be called a Jacobi pair belonging to
X
(s)
=

X
(s)
.
Under the further (crucial) hypothesis of regularity of
(s)
, we next make use
of the innitesimal control h
(s)
: V (
(s)
) A(
(s)
) induced
7
by the Lagrangian
L
(s)
to express the eld

X
(s)
in terms of the Jacobi eld X
(s)
and of a vertical
vector Y
(s)
in the form

X
(s)
= h
(s)
(X
(s)
) +Y
(s)
= X
i
(s)
i
+Y
A
(s)
z
A
. On account of
equations (3.1.15) we have then the relation
A
(s)
= Y
A
(s)
G
AB
(s)
_
2
L
(s)
q
i
z
B
_

(s)
X
i
(s)
(3.3.22)
7
See Remark 3.1.5
Together with the identication G
(s)
AB
=
_
2
L
(s)
z
A
z
B
_

(s)
, the latter allows to write
equation (3.3.18c) into the form

(s)
i
_
i
z
A
_

(s)
= G
(s)
AB
Y
B
(s)
= Y
A
(s)
= G
AB
(s)

(s)
i
_
i
z
B
_

(s)
(3.3.23)
From this, substituting into equations (3.3.18a,b), recalling the denitions
(3.3.3), (3.3.6) and expressing everything in terms of the absolute time deriva-
tive, we eventually obtain the following system of dierential equations for the
unknowns X
i
(s)
(t),
(s)
i
(t):
_
DX
i
(s)
Dt
_
(s)
= G
AB
(s)
_
i
z
A
_

(s)
_
j
z
B
_

(s)

(s)
j
= M
ij
(s)

(s)
j
(3.3.24a)
_
D
(s)
i
Dt
_
(s)
=
__
2
L
(s)
q
i
q
j
_

(s)
G
AB
(s)
_
2
L
(s)
q
i
z
A
_

(s)
_
2
L
(s)
q
j
z
B
_

(s)
_
X
j
(s)
=
= N
(s)
ij
X
j
(s)
(3.3.24b)
As the attentive reader will have certainly already noticed, these are formally
identical to the linearized form of Riccati equation (3.3.11): a result that will prove
to be fundamental in the sequel in order to determine the necessary and sucient
conditions for a weak local minimum.
Remark 3.3.3: Keeping in line with Remark 3.3.1, if the virtual tensor algebra along
(s)
is referred to an h
(s)
transported basis
_
e
(a)
, e
(a)
_
, the set of 2n dierential equations
(3.3.24) are written in the form
dX
a
(s)
dt
= M
ab
(s)

(s)
b
(3.3.25a)
d
(s)
a
dt
= N
(s)
ab
X
b
(s)
(3.3.25b)
Once again, these are easily seen to represent the Hamilton equations for the function
H
(s)
(t, X
a
(s)
,
(s)
b
) =
a

a
L
(s)
=
1
2
M
ab

(s)
a

(s)
b

1
2
N
(s)
ab
X
a
(s)
X
b
(s)
(3.3.26)
which is the linearized counterpart of the Hamiltonian function on the extremal curve
(s)
.
The relationship between Jacobi elds and the second variation is made evident
by the following
Proposition 3.3.1. Given a Jacobi pair
_

X
(s)
,
(s)
_
, for any arbitrary vector eld
Z
(s)
A(
(s)
), the following identity holds:
_
_
d
2
L
(s)
_

(s)
,

X
(s)

Z
(s)
_
=
d(
(s)
i
Z
i
(s)
)
dt
(3.3.27)
Proof. The thesis immediately follows by direct computation, in view of equations
(3.1.12), (3.3.18). Setting

Z
(s)
= Z
i
_

q
i
_

(s)
+Z
A
_

z
A
_

(s)
, we have
d(
(s)
i
Z
i
(s)
)
dt
=
_

2
L
q
i
q
j
_

(s)
X
j
(s)
Z
i
(s)
+
_

2
L
q
i
z
A
_

(s)
A
(s)
Z
i
(s)
+
+
_

2
L
z
A
q
i
_

(s)
X
i
(s)
Z
A
(s)
+
_

2
L
z
A
z
B
_

(s)
B
(s)
Z
A
(s)
Remark 3.3.4: Hitherto, our treatment of Jacobi elds has uniquely involved the adapted
Lagrangian L
(s)
. This choice was suggested both by consistency with the previous anal-
ysis and also by the simplied calculations. However, it goes without saying that its not
at all necessary in order to cover the subject. We could actually have considered
(s)
as
extremal of the functional
_

(s)

PPC
instead of
_

(s)

(s)
PPC
. In this way equations (3.3.18)
would have been directly written in terms of the Pontryagin Hamiltonian H , with the
quantities
(s)
i
replaced by
(s)
i
:=
_
(s)
i
_
=0
, related to the previous ones by the relation

(s)
i
(t) =
_

(s)
i
(t, )
_
=0
=

(s)
i
(t, )
S
(s)
q
i
_
=0
=
(s)
i
(t)

2
S
(s)
q
i
q
j
X
j
(s)
The argument is almost identical to the one developed so far and will be omitted.
3.3.3 Conjugate points and the necessary conditions
Jacobi elds are related to the necessary conditions for (local) minimality through
the concept of conjugate point.
Denition 3.3.1 (Conjugate point). A point
(s)
(), (a
s1
, a
s
], along a given
extremal curve
(s)
is said to be conjugate to
(s)
(a
s1
) if there exists a nonzero
Jacobi eld X
(s)
: [a
s1
, a
s
] V (
(s)
) such that X
(s)
(a
s1
) = X
(s)
() = 0.
It is easily seen that the search for conjugate points can be performed by looking
for a solution of equations (3.3.24) with X
i
(s)
(a
s1
) = 0 and
(s)
i
(a
s1
) varying
amongst all the possible values in R
n
.
Because of the linearity of equations (3.3.24), their solution will depend on the
initial data through a set of timedependent matrices in the form
X
i
(s)
(t) = A
i
j
(t, a
s1
) X
j
(s)
(a
s1
) + B
ij
(t, a
s1
)
(s)
j
(a
s1
) (3.3.28a)

(s)
i
(t) = C
ij
(t, a
s1
) X
j
(s)
(a
s1
) + D
j
i
(t, a
s1
)
(s)
j
(a
s1
) (3.3.28b)
with A
i
j
(a
s1
, a
s1
) =
i
j
, B
ij
(a
s1
, a
s1
) = 0 , C
ij
(a
s1
, a
s1
) = 0 and
D
j
i
(a
s1
, a
s1
) =
j
i
. Conjugate points can be therefore determined by means of
equation (3.3.28a) restricted to the choice X
i
(s)
(a
s1
) = 0, namely
X
i
(s)
(t) = B
ij
(t, a
s1
)
(s)
j
(a
s1
) (3.3.29)
Hence, a point
(s)
() is conjugate to
(s)
(a
s1
) whenever
(s)
j
(a
s1
) belongs to the
kernel of B
ij
(, a
s1
) and this can only happen when det(B
ij
(, a
s1
)) vanishes.
The link between conjugate points and the analysis of the second variation is
claried by the following generalization of a classical result of Bliss ([19]):
Theorem 3.3.3. Consider an extremal closed arc
(s)
: [a
s1
, a
s
] 1
n+1
and
suppose there exists a value (a
s1
, a
s
) such that the point
(s)
() is conjugate
to
(s)
(a
s1
). Then the quadratic form
d
2
1 [
(s)
]
d
2
=0
=
_
as
a
s1
_
_
d
2
L
(s)
_

(s)
,

X
(s)

X
(s)
_
dt
is necessarily indenite.
Proof. Let us dene a symmetric bilinear functional
_
d
2
1
_

(s)
over A(
(s)
) as
_
_
d
2
1
_

(s)
,

V
(s)

W
(s)
_
:=
_
as
a
s1
_
_
d
2
L
(s)
_

(s)
,

V
(s)

W
(s)
_
dt
for any

V
(s)
,

W
(s)
in A(
(s)
). Then, by a wellknown result in the theory of
quadratic forms
8
, the thesis is proved as soon as we show that, in the presence of
a point
(s)
() conjugated to
(s)
(a
s1
), the kernel of
_
d
2
1
_

(s)
does not coincide
with the locus of zeroes of its associated quadratic form.
Under the stated hypothesis, there exists a Jacobi eld J
(s)
V (
(s)
) such
that J
(s)
(a
s1
) = J
(s)
() = 0. By means of this, we now dene a continuous
innitesimal deformation vanishing at the endpoints X
(s)
: [a
s1
, a
s
] V (
(s)
)
in the following manner:
X
(s)
(t) :=
_
J
(s)
(t) a
s1
t
0 t a
s
8
See Appendix D, Lemma D.1.
Then, denoting by

X
(s)
A(
(s)
) the lift of X
(s)
and by
_

J
(s)
,
(s)
_
a Jacobi pair
belonging to J
(s)
, in view of equation (3.3.27) we have
_
_
d
2
1
_

(s)
,

X
(s)

X
(s)
_
=
_

a
s1
_
_
d
2
L
(s)
_

(s)
,

J
(s)

J
(s)
_
dt =
=
_

(s)
i
J
i
(s)
_
a
s1
= 0
At the same time, if W
(s)
is any innitesimal deformation of
(s)
vanishing at the
endpoints and such that W
(s)
() ,= 0, we have also
_
_
d
2
1
_

(s)
,

X
(s)

W
(s)
_
=
_

a
s1
_
_
d
2
L
(s)
_

(s)
,

J
(s)

W
(s)
_
dt =
=
(s)
i
()

W
i
(s)
()
Since, by hypothesis, J
(s)
(t) ,= 0 for every t (a
s1
, ), the uniqueness of the
solution of the timereversed Cauchy problem (3.3.24) in
(s)
() implies that

(s)
i
() ,= 0 for at least one value of the index i. Therefore
_
_
d
2
1
_

(s)
,

X
(s)

W
(s)
_
,= 0
showing that

X
(s)
does not belong to the kernel of
_
d
2
1
_

(s)
.
As a direct consequence of Theorem 3.3.3, we can now state the following
Proposition 3.3.2 (Necessary conditions). Suppose the extremal closed arc
(s)
: [a
s1
, a
s
] 1
n+1
is a (local) minimum for the functional
_

(s)
Ldt. Then,
for every (a
s1
, a
s
), there cannot be any point
(s)
() conjugate to
(s)
(a
s1
).
3.3.4 The necessary and sucient conditions
So far we have separately proved a sucient and a necessary condition for a given
extremal
(s)
to be a minimum; we shall now glue them together into a necessary
and sucient one. However, in order to do so, we shall need to strengthen the
hypothesis of normality of
(s)
by requiring the latter to be locally normal.
In the event, we will prove that, whenever no conjugate point is present, the
solutions of equations (3.3.24) can be used to build a global solution of the Riccati
equation (3.3.5), valid along the whole interval [a
s1
, a
s
], thus satisfying some
of the hypothesis of Theorem 3.3.2. To this purpose, we rst need a technical
argument.
Lemma 3.3.3.1. Let
(s)
: [a
s1
, a
s
] 1
n+1
9
be a locally normal extremal and
suppose the matrix G
(s)
AB
is nonsingular at each t [a
s1
, a
s
]. If along
(s)
there
is no point conjugate to
(s)
(a
s1
), then there exists a t
> a
s
such that the
absence of conjugate points may be extended over a wider interval [a
s1
, t
].
Proof. Consider the family of Jacobi pairs
_

X
(s)
,
(s)
_
(k)
, k = 1, . . . , n, obtained
as solutions of equations (3.3.24) with initial data
(X
(s)
)
i
(k)
(a
s1
) = 0 , (
(s)
)
i(k)
(a
s1
) =
ik
The nonexistence of conjugate points along
(s)
is easily seen to be equivalent to
the condition det
_
(X
(s)
)
i
(k)
(t)
_
,= 0 for all t (a
s1
, a
s
].
If that is not the case, there would be some (a
s1
, a
s
] at which the
homogenous system a
k
(X
(s)
)
i
(k)
() = 0 would admit a nonnull solution a
1
, .., a
n
.
The elds

X
(s)
:= a
k
(

X
(s)
)
(k)
,
(s)
= a
k
(
(s)
)
(k)
would then constitute a Jacobi
pair satisfying the conditions
(s)
(a
s1
) ,= 0,

X
(s)
(a
s1
) =

X
(s)
() = 0. On the
other hand,

X
(s)
cannot be identically zero over the whole interval [a
s1
, ]: if it
were so, the 1form
(s)
would satisfy the equations
_
DX
i
(s)
Dt
_
(s)
= M
ij
(s)

(s)
j
= 0 =
(s)
j
_
j
z
B
_

(s)
= 0
_
D
(s)
i
Dt
_
(s)
= 0
a
s1
t
contradicting the local normality of
(s)
.
To sum up,

X
(s)
would be a nonzero Jacobi vector eld vanishing at both
a
s1
and , which clashes with the assumption of nonexistence of conjugate
points along
(s)
.
By continuity, this implies det
_
(X
(s)
)
i
(k)
(t)
_
,= 0 for all t (a
s1
, t
] with
t
(a
s
, b
s
) suciently close to a
s
. The absence of conjugate points holds there-
fore in a wider interval [a
s1
, t
].
We are now ready to take the conclusive step towards the formulation of the nec-
essary and sucient conditions for minimality, which is provided by the following
Proposition 3.3.3. Let
(s)
: [a
s1
, a
s
] 1
n+1
be a locally normal extremal
and suppose the matrix G
(s)
AB
is nonsingular at each t [a
s1
, a
s
]. If no pair of
conjugate points exists on
(s)
, the Riccati equation (3.3.5) admits a symmetric
solution throughout the interval [a
s1
, a
s
].
9
We recall that the closed arc
(s)
is the restriction to the closed interval [as1, as] of an
admissible section dened on some open neighborhood (bs1, bs) [as1, as] .
Proof. As usual, we regard
(s)
as the restriction of an extremal dened on an
open interval (b
s1
, b
s
) [a
s1
, a
s
]. Let t
(a
s
, b
s
) and consider a family of so-
lutions
_

X
(s)
,
(s)
_
(k)
of equations (3.3.24), obtained imposing the initial conditions
(X
(s)
)
i
(k)
(t
) = 0 and (
(s)
)
i(k)
(t
) =
ik
.
In view of Lemma 3.3.3.1, whenever t
is chosen suciently close to a

s
, the
absence of conjugate points implies the requirement det
_
(X
(s)
)
i
(k)
(t)
_
,= 0 for all
t [a
s1
, t
).
A comparison between the Hamiltonian system (3.3.24) and the linearization
(3.3.11) of Riccati equation shows that we can now assume the identications
K
i
(s)j
(t) := (X
(s)
)
i
(j)
(t) , E
(s)
ij
(t) := (
(s)
)
i(j)
(t) (3.3.30)
As a consequence, the matrix K
i
(s)j
(t) is nonsingular everywhere on [a
s1
, t
) and
therefore, as weve seen in 3.3.1, the tensor C
(s)
ij
= E
(s)
ir
_
K
1
(s)
_
r
j
represents a
solution of the Riccati equation (3.3.5) all over the interval [a
s1
, t
) [a
s1
, a
s
].
In order to complete the proof, we now only need to show that this C
(s)
ij
is
also symmetric. To this end we observe that the matrix R
ip
(s)
:= K
i
(s)
j
_
E
1
(s)
_
jp
is
perfectly meaningful in a neighborhood (t
, t
] and satises the relations

R
ip
(s)
(t
) = 0 , R
ip
(s)
C
(s)
pq
= K
i
(s)
j
_
E
1
(s)
_
jp
E
(s)
pr
_
K
1
(s)
_
r
q
=
i
p
t < t
The matrix R
ip
(s)
is therefore symmetric at t = t
. Moreover, on account of equa-

tions (3.3.24), it satises the equation
DR
ip
(s)
Dt
=
DK
i
(s)j
Dt
_
E
1
(s)
_
jp
+ K
i
(s)
j
D
_
E
1
(s)
_
jp
Dt
= M
ir
(s)
R
il
(s)
N
(s)
lk
R
kp
(s)
which is again of the Riccatitype (3.3.5), with the roles of the matrices M
ij
(s)
, N
(s)
ij
interchanged. Exactly as in Theorem 3.3.1, this establishes the symmetry of R
ip
(s)
in a neighborhood of t = t
.
For each t (t
, t
) the matrix C
(s)
ij
(t) =
_
R
ij
(s)
(t)
_
1
is therefore symmetric.
Once again, on account of the linearity of equation (3.3.7b), we conclude that
C
(s)
ij
(t) is symmetric over the whole interval [a
s1
, t
) [a
s1
, a
s
].
Collecting all the above arguments, we are now able to state the following
Theorem 3.3.4 (Necessary and sucient conditions). Suppose the closed arc
(s)
: [a
s1
, a
s
] 1
n+1
is a locally normal extremal of the functional
_

(s)
L dt
with respect to the class of deformations vanishing at the endpoints and let
(s)
be its (unique) lift to ((/) solving Pontryagins equations (2.3.4a,b,c). Denote
3.4 The induced quadratic form 87
by H
(s)
(t, q
i
, z
A
, p
i
) = p
(s)
i

i
(t, q
i
, z
A
) L(t, q
i
, z
A
) the Pontryagin Hamiltonian
associated with the given Lagrangian and let
G
(s)
AB
(t) :=
_

2
H
(s)
z
A
z
B
_

(s)
Then, the arc
(s)
is a minimum for the action functional if and only if, for every
t [a
s1
, a
s
], the matrix G
(s)
AB
is positive denite and there is no point conjugate
to
(s)
(a
s1
).
The proof should, at this time, be quite straightforward and is left to the reader.
3.4 The induced quadratic form
With Theorem 3.3.4, the former step of the stated resolution strategy can be said
brought o. From now onwards, we shall thus embrace the hypothesis of each arc
(s)
being a locally normal extremal of the functional
_

(s)
L dt and a minimum
with respect to the xed endpoints deformations and well devote ourselves to the
further task of nding out whether it is possible to combine the previous results in
order to make them globally applicable to the entire evolution . This will involve
the study of the deniteness properties of the quadratic form (3.2.2) and will be
carried out by making use of the results of Appendix D and, in particular, along
the lines of Theorem D.1.
To start with, we observe that, under the present hypothesis, we are supposed
to be able to nd N restricted gauge transformations in such a way that
_
_
d
2
L
(s)
_

(s)
,

X
(s)

X
(s)
_
= G
(s)
AB
Y
A
(s)
Y
B
(s)
s = 1, . . . , N
and so the quadratic form (3.2.2) can be written more suitably as
d
2
1[
]
d
2
=0
=
N
s=1
_
as
a
s1
G
(s)
AB
Y
A
(s)
Y
B
(s)
dt
N1
s=1
_
_
d
2
S
cs
, W
s
W
s
_
(3.4.1)
Moreover, being each matrix G
(s)
AB
positive denite, the Lagrangian functions
L
(s)
, s = 1, . . . , N provide their respective arcs
(s)
with an innitesimal control
h
(s)
, therefore assigning a transport law to the vertical space V () or, all the same,
a canonical trivialization of the latter into the cartesian product RV
h
.
We recall that, in the algebraic environment developed in 1.5.4, the vector
space of the admissible innitesimal deformations vanishing at the endpoints of
was seen to be isomorphic to the kernel of the linear map : W V
h
whose
representation in an htransported basis e
(a)
of V
h
reads
_
Y,
1
, . . . ,
N1
_
=
_
_
t
1
t
0
Y
A
e
(a)
i
_
i
z
A
_

dt
N1
s=1
s
k
a
(s)
_
e
(a)
being k
a
(s)
:= e
(a)
i
(a
s
)
_
i
( )
as
.
In view of this, we now introduce one more linear map
J: ker() T
c
1
(1
n+1
) T
c
N1
(1
n+1
)
which maps each element
_
Y,
1
, . . . ,
N1
_
ker() into the corresponding col-
lection W
(1)
, . . . , W
(N 1)
of tangent vectors to the orbits of the corners. The
subspace ker(J) ker() is therefore formed by the totality of the admissible
innitesimal deformations for which W
(1)
= = W
(N 1)
= 0, namely the ones
that vanish at the corners.
Setting
a
(s)
_
W
(1)
, . . . , W
(N 1)
_
:=
X
(s)
, e
(a)
_
cs
, well henceforth refer the
space T
c
1
(1
n+1
) T
c
N1
(1
n+1
) to the coordinate system
s
,
a
(s)
. Re-
calling the expression (1.5.32), this results in the representation
W
(s)
=
s
_

t
_
cs
+
_
a
(s)
+
s
i
|
(s)
(as)
e
(a)
i
(a
s
)
_
(e
(a)
)
cs
(3.4.2)
Theorem 3.4.1. If each arc
(s)
is normal, the map J is surjective.
Proof. We rst observe the following identity
X
a
(s + 1)
(a
s
) =
a
(s)
_
W
(1)
, . . . , W
(N 1)
_
+
_
X
as
, e
(a)
(a
s
)
_
=
=
a
(s)
_
W
(1)
, . . . , W
(N 1)
_

s
k
a
(s)
which is a direct consequence of the jump conditions (1.5.41b). The request for any
arbitrary element
_
W
(1)
, . . . , W
(N 1)
_
to be the image under J of a corresponding
_
Y,
1
, . . . ,
N1
_
ker() makes the vertical vector elds Y to be subject to
the conditions
_
as
a
s1
Y
A
(s)
e
(a)
i
_
i
z
A
_

(s)
dt = X
a
(s)
(a
s
) X
a
(s)
(a
s1
) =
a
(s)

a
(s 1)
+
s1
k
a
(s 1)
(3.4.3)
The conclusion follows at once simply by observing that the above equation admits
(at least) a solution Y for any possible choice of the variables
s
,
a
(s)
if and only
if the mappings
Y
(s)

_
as
a
s1
Y
A
(s)
e
(a)
i
_
i
z
A
_

(s)
dt s = 1, . . . , N
3.4 The induced quadratic form 89
are surjective, which is totally equivalent to the normality of each arc
(s)
.
On account of Theorem 3.4.1, under the stated hypothesis, the quotient space
ker()
/ker(W) coincides with the cartesian product T
c
1
(1
n+1
) T
c
N1
(1
n+1
) .
Each element
_
W
(1)
, . . . , W
(N 1)
_
, thought as an equivalence class in ker(), is
then formed by the totality of
_
Y,
1
, . . . ,
N1
_
such that, for any s, Y
(s)
fulls
the condition (3.4.3) while
s
=
W
(s)
, dt
|cs
_
.
Coming back to the study of the quadratic form (3.4.1), it is readily seen that
its restriction to the subspace ker(J) is positive denite, being the sum of N
positive denite quadratic forms. Moreover, its restriction to any equivalence class
J
1
_
W
(1)
, . . . , W
(N 1)
_
has a single stationarity point. In order to nd it out, it
is possible to make use of the method of Lagrange multipliers by considering the
functional
N
s=1
_
as
a
s1
G
(s)
AB
Y
A
(s)
Y
B
(s)
dt
N1
s=1
_
_
d
2
S
cs
, W
s
W
s
_
+
+
N
s=1
(s)
a
_
_
as
a
s1
Y
A
(s)
e
(a)
i
_
i
z
A
_

(s)
dt
a
(s)
+
a
(s 1)

s1
k
a
(s 1)
_
(3.4.4)
with independent variables Y
A
(s)
,
(s)
a
and xed
s
,
a
(s)
.
The vanishing of the rst derivatives with respect to the
(s)
a
s obviously gives
back the constraints (3.4.3), while the variation with respect to the Y
A
(s)
s provides
the relations
2 G
(s)
AB
Y
B
(s)
+
(s)
a
e
(a)
i
_
i
z
A
_

(s)
= 0 =

Y
A
(s)
=
1
2
G
AB
(s)

(s)
a
e
(a)
i
_
i
z
B
_

(s)
(3.4.5)
Substituting into equations (3.4.3), we get
1
2

(s)
b
_
as
a
s1
G
AB
(s)
_
i
z
A
_

(s)
_
j
z
B
_

(s)
e
(a)
i
e
(b)
j
dt =
=
1
2

(s)
b
_
as
a
s1
M
ab
(s)
dt =
a
(s)

a
(s 1)
+
s1
k
a
(s 1)
(3.4.6)
Because of the nonsingularity of G
(s)
AB
, the local normality of each arc
(s)
implies
the positive deniteness of the corresponding matrix
g
ab
(s)
:=
_
as
a
s1
M
ab
(s)
dt s = 1, . . . N
and therefore, denoting by g
(s)
ab
its inverse matrix, we can solve equations (3.4.6)
for the unknowns
(s)
a
s in the form
(s)
a
= 2 g
(s)
ab
_
b
(s)

b
(s 1)
+
s1
k
b
(s 1)
_
(3.4.7)
The expression of the stationarity point

Y
A
(s)
can now be rewritten as
Y
A
(s)
= G
AB
(s)
g
(s)
ab
_
b
(s)

b
(s 1)
+
s1
k
b
(s 1)
_
e
(a)
i
_
i
z
B
_

(s)
which is actually a minimum point, once again on account of the positive denite-
ness of the matrixes G
(s)
AB
.
At last, we may induce a quadratic form f :
ker()
/ker(W) R by mapping each
equivalence class
_
W
(1)
, . . . , W
(N 1)
_
in ker() into the real number given by the
evaluation of the quadratic form (3.4.1) at the corresponding (unique) minimum
point

Y
A
(s)
. In local coordinates, we have the representation
f
_
s
,
a
(s)
_
=
N
s=1
g
(s)
ab
_
a
(s)

a
(s 1)
+
s1
k
a
(s 1)
__
b
(s)

b
(s 1)
+
s1
k
b
(s 1)
_
+
N1
s=1
_
_
d
2
S
cs
, W
s
W
s
_
(3.4.8)
being the vectors W
s
implicitly expressed in terms of the variables
s
,
a
(s)
by
means of equation (3.4.2).
Collecting all previous results, we have thus proved
Theorem 3.1. Let
_
, [t
0
, t
1
]
_
:=
__
(s)
, [a
s1
, a
s
]
_
, s = 1, . . . , N
_
be a piecewise
dierentiable locally normal extremal of the functional
_

L dt. Suppose the matrix
G
(s)
AB
is positive denite along each arc
(s)
and suppose there is no point conjugate
to
(s)
(a
s1
). Then, a necessary and sucient condition for the minimality of
is the positive deniteness of the quadratic form (3.4.8).
Appendix A
Adapted local charts
The aim of the present Appendix is to single out a distinguished nite family of
local charts in / that covers the section and makes its representation as easy as
possible. The use of these charts will turn out to be most useful especially when
the discussion itself is already rather entangled, as it helps in easing the notation
and reduces the eort needed to carry out all calculations. It goes without saying
that, in order to preserve the generality of all results, one should always take care
of checking their independence of any particular choice of coordinates.
Lemma A.1. Let : (c, d) 1
n+1
be a dierentiable section and m, n (c, d).
Then, for every closed interval [a, b] (c, d) there exist an open neighborhood
(m, n) [a, b] and a dierentiable vector eld X such that
_

t
_
= X
| (t)
for
any t (m, n).
Proof. Let m (c, a) and n (b, d). Being compact, the arc ([m, n]) is
covered by a nite family of local charts with compact closure (V
1
, k
1
), . . . (V
r
, k
r
)
that we order timewise. In each local chart, where is represented in coordinates
by q
i
=
i
(t), it is always possible to arrange a straightforward transformation
q
i
= q
i
i
(t) such that reduces to the coordinate line q
i
= 0, which is therefore
tangent to the eld

t
. We now sort out, among all the partitions of unity that
are subordinate to the covering V
1
, . . . , V
r
, 1
n+1
([c, d]), the (nite) family
of functions whose supports intersect ([c, d]) and dene as g
, = 1, . . . , r, the
sum of the ones whose supports are contained in V
but not in V
, < . In
this way, weve provided every open set V
with a function g
having support in
V
and globally dened on 1

n+1
in such a way that
((t)) = 1 for every

t [m, n] .
It is now an easy matter to see that, if we dene a eld X
()
as
X
()
x
=
_
g
(x)
_

t
_
x
x V
0 x / V
92 Appendix A. Adapted local charts

the vector eld X :=
r
=1
X
()
fulls all the required properties.
According to Lemma A.1, the integral line of X passing through the point
(m) is dened at least up to t = n. By wellknown theorems in dierential
equations (see e.g. [11, 22]), this in turn implies that the same will happen if
the initial data are chosen in an open neighborhood of (m). In particular, if we
denote by W the intersection of this open set with the hyperplane
a
: t = a and
by the ow tube containing all the integral lines of X that spit out of W, then:
all the lines contained in are dened (at least) up to m t n;
every local coordinate system q
1
, . . . , q
n
on W may be used to refer to
local coordinates t, q
1
, . . . , q
n
. Moreover, it is always possible, without any
loss of generality, to make the choice q
i
((a)) = 0 which makes the curve
into the coordinate line q
i
= 0 .
In the presence of piecewise dierentiable sections it is possible to apply the
previous construction in each single arc and then to combine the results into a
global one. We rst provide the arc
(1)
with a local chart (
1
, t, q
1
(1)
, . . . , q
n
(1)
) as
above. We then choose W
1

1
a
1
and refer it to local coordinates q
1
(1)
, . . . , q
n
(1)
.
In doing so, we should be wise enough to take it as small as to be used as initial
data set for a second ow tube
2
which will contain (not strictly) the closed
interval [a
1
, a
2
]. Pursuing this process till the end, we obtain a nite family of
local charts (one for every dierentiable arc
(s)
) with the following properties:
(i) each single arc
(s)
is contained in
s
and is represented there as the coor-
dinate line q
i
(s)
= 0 ;
(ii) in the intersection
s

s+1
, the transformation
q
i
(s + 1)
= q
i
(s + 1)
(t, q
1
(s)
, . . . , q
n
(s)
)
is such that
q
i
(s + 1)
(a
s
, q
1
(s)
, . . . , q
n
(s)
) = q
i
(s)
(A.1)
Lemma A.2. Let : (c, d) / be the lift of an admissible dierentiable section
: (c, d) 1
n+1
. Then, for any closed interval [a, b] (c, d) there exists a bred
local chart (
U,
h),

h = (t, q
1
, . . . , q
n
, z
1
, . . . , z
r
) satisfying the properties
(i) (t)

U t [a, b]; (A.2a)
(ii)
_
(c, d)
_

U coincides with the curve q
i
= z
A
= 0; (A.2b)
(iii)
i
_
(t)
_
=
_
i
q
k
_
(t)
= 0 (t)

U. (A.2c)
93
Proof. The construction carried out at the end of Lemma A.1 ensures the ex-
istence of tubular local charts (, h), h = (t, q
1
, . . . , q
n
) in 1
n+1
and (
),
= (t, x
1
, . . . , x
n+r
) in / satisfying the conditions
([a, b]) , q
i
((t)) = 0 t (c, d)
1
()
([a, b])

, x
( (t)) = 0 t (c, d)
1
(
)
Without loss of generality we may assume (
) . The restriction to

of
the projection : / 1
n+1
is then described in coordinates as
q
i
= q
i
(t, x
1
, . . . , x
n+r
)
with rank
_
_
_
( q
1
q
n
)
(x
1
x
n+r
)
_
_
_ = n. In particular, the dierentials dt, d q
1
, . . . , d q
n
are
linearly independent everywhere on

.
Let
A
:=
A
(t) dx
| (t)
denote r linear dierential forms along , depending
dierentiably on t, and completing dt
| (t)
, d q
i
| (t)
to a basis of T
(t)
(/) .
Dene r dierentiable functions on

by
z
A
=
n+r
=1
(t) x
Then, by construction, the Jacobian

_
_
_
( q
1
q
n
z
1
z
r
)
(x
1
x
n+r
)
_
_
_ is non singular at each point
(t). The functions t, q
i
, z
A
form therefore a coordinate system in a neighborhood
U of the intersection
_
(c, d)
_

. The system is automatically bred over ,
and satises both properties (A.2a, b), and the rst condition (A.2c).
To complete the proof, let

q
i
=

i
(t, q
i
, z
A
) denote the representation of the
imbedding / j
1
(1
n+1
) in the coordinates t, q
i
, z
A
. Under an arbitrary linear
transformation q
i
=
i
j
(t) q
j
, z
A
= z
A
we have then the transformation laws
i
=
d
i
j
dt
q
j
+
i
j

j
,

i
q
k
=
_
d
i
j
dt
+
i
r
r
q
j
_
_
1
_
j
k
In particular, if the matrix
i
j
(t) is a solution of the dierential equation
d
i
j
dt
+
i
r
_
r
q
j
_
(t)
= 0
the coordinates t, q
i
, z
A
satises all stated requirements.
Every local chart (
U,
h) satisfying equations (A.2a, b, c) will be said to be adapted

to the closed arc
_
, [a, b]
_
.
94 Appendix A. Adapted local charts
Corollary A.1. Let =
__

(s)
, [a
s1
, a
s
]
_
, s = 1, . . . , N
_
be the lift of an ad-
missible piecewise dierentiable section
_
, [t
0
, t
1
]
_
. Then, there exist bred lo-
cal charts (
U
s
,
h
s
),

h
s
=
_
t, q
1
(s)
, . . . , q
n
(s)
, z
1
(s)
, . . . , z
r
(s)
_
adapted to the arcs
(s)
such that, in each intersection (
U
s
) (
U
s+1
), the coordinate transformation
q
i
(s + 1)
= q
i
(s + 1)
(t, q
1
(s)
, . . . , q
n
(s)
) satises the condition (A.1)
q
i
(s + 1)
(a
s
, q
1
(s)
, . . . , q
n
(s)
) = q
i
(s)
Proof. The result follows at once by applying Lemma A.2 arc by arc and setting
(s)
i
j
(a
s
) =
(s + 1)
i
j
(a
s
)
for all s = 1, . . . , N 1.
Every family of local charts
_
(
U
s
,
h
s
), s = 1, . . . , N
_
satisfying the require-
ments of Corollary A.1 will be said to be adapted to the lift .
Assigning an adapted family of local charts automatically singles out a dis-
tinguished innitesimal control h
(s)
along each arc
(s)
, uniquely dened by the
requirement
h
(s)
_
_

q
i
(s)
_
(s)
(t)
_
=
_

q
i
(s)
_

(s)
(t)
h
i
A
(t) = 0
In view of equations (1.5.21b), (1.5.22a) and (A.2c), the absolute time deriva-
tive associated with h
(s)
is described in coordinates as
D
Dt
_

q
i
(s)
_
(s)
(t)
= 0 s = 1, . . . , N (A.3)
Since, by Corollary A.1, the elds
_

q
i
(s)
_
(s)
(t)
are continuous at the corners,
then the sections e
(i)
: [t
0
, t
1
] V () given by
e
(i)
(t) =
_

q
i
(s)
_
(s)
(t)
t [a
s1
, a
s
] , s = 1, . . . , N (A.4)
form a basis for the space V
h
of htransported vector elds along .
On account of equation (A.2c), the corresponding dual basis for the space V
h
is given by e
(i)
(t) =
i
|
(s)
(t)
= dq
i
(s)|
(s)
(t)
t [a
s1
, a
s
] , s = 1, . . . , N. By
denition, together with equations (A.3) we have therefore the dual relations
D
Dt

i
|
(s)
(t)
= 0 (A.5)
Appendix B
Finite deformations with xed
endpoints: an existence
theorem
According to Proposition 1.5.1, the admissible innitesimal deformations of an
admissible, piecewise dierentiable section : [t
0
, t
1
] 1
n+1
are in bijective cor-
respondence with the sections

X : R A( ) fullling the consistency requirement
locally expressed by the variational equation (1.5.8).
In the event, this bijective correspondence is actually considered as a full iden-
tication between them. It was just in this particular sense that in 1.5.4 we
claimed that the most general admissible innitesimal deformation X of vanish-
ing at t = t
0
is determined by an element (Y,
) W, namely by a vertical vector

eld Y along and by a collection of real numbers
= (
1
, . . . ,
N1
) and that,
in particular, a necessary and sucient condition for X to satisfy X(t
1
) = 0 is
expressed by the requirement (1.5.43) which, in adapted coordinates, reads
_
t
1
t
0
Y
A
_
i
z
A
_

dt
N1
s=1
s
_
i
( )
as
= 0 (B.1)
This is, for the most part, a right way of acting but care must be taken inasmuch
there now may be pathological circumstances in which one can nd admissible
innitesimal deformations of vanishing at its endpoints that are not tangent to
any admissible nite deformation
with xed endpoints.

Example B.1. Consider a system B in 1
n+1
= RE
2
(referred to coordinates
t, x, y ) and subject to the constraint x
2
+ y
2
= v
2
. We seek those evolutions which
join the endpoints (t
0
= 0, x
0
= 0, y
0
= 0) and (t
1
=

t, x
1
= v
t, y
1
= 0) and
minimize a given action functional.
It is now apparent that, regardless of the nature of the functional, the problem
has a unique solution, represented by the curve : x(t) = vt, y(t) = 0.
96 Appendix B. Finite deformations with xed endpoints: an existence theorem
Such a solution is therefore a rigid curve, completely lacking in admissible
nite deformations with xed endpoints. Even so, there could be admissible in-
nitesimal deformations vanishing at the endpoints. To see this, we express the
imbedding i : / j
1
(1
n+1
) in the form
i :
_
x = v cos z
y = v sin z
We then require the admissibility of by making the condition
_
v = v cos z
0 = v sin z
whence we get z = 0. A possible lift of is therefore represented by the curve
: x(t) = vt, y(t) = 0, z(t) = 0.
The variational equations (1.5.8) are now expressed by
_
_
_
dX
1
dt
= v (sin z)

= 0
dX
2
dt
= v (cos z)

= v
the rst of which, together with the request X
1
(t
1
) = 0, entails
X
1
(t) 0
In like manner, the second one becomes
X
2
(t) = v
_
t
0
() d
completed by the condition
X
2
(t
1
) = v
_
t
1
0
() d = 0
To sum up, a possible particular solution is given by
X
1
(t) = 0 , X
2
(t) = v sin
_
t
t
1
_
, (t) =

t
1
cos
_
t
t
1
_
and so weve found an admissible innitesimal deformation which vanishes at the
endpoints of , regardless of the latter being a rigid curve that admits no nite
deformations.
97
Therefore, given an admissible, piecewise dierentiable section , a crucial
question is establishing under what circumstances every admissible innitesimal
deformation vanishing at its endpoints is tangent to an admissible nite deforma-
tion
with xed endpoints. If this is the case, the evolution is called ordinary,
otherwise exceptional. We will now try to get sucient conditions for ordinariness.
For this purpose, recalling the contents of Appendix A, we introduce a family of
local charts (U
s
, k
s
) adapted to and denote by e
(i)
, e
(i)
the corresponding
dual bases for the spaces V
h
, V
h
.
We also bring in, as an auxiliary tool, a positive metric on V
h
, described by
a symmetric tensor = g
ij
e
(i)
e
(j)
. In view of the identication of V () with
[t
0
, t
1
] V
h
, this automatically sets up a scalar product along the bres of V ()
which, in turn, determines a scalar product between vertical vector elds along ,
based on the prescription
_
Y, Z
_
:=
_
(Y ), (Z)
_
(B.2)
: V ( ) V () denoting the homomorphism (1.3.13). In adapted coordinates,
equations (1.3.14), (B.2) provide the evaluation (Y, Z) = G
AB
Y
A
Z
B
, with
G
AB
=
_

_

z
A
_

,
_

z
B
_

_
= g
ij
_
i
z
A
_

_
j
z
B
_

(B.3)
As usual, the inverse of the matrix G
AB
will be denoted by G
AB
.
In a similar manner, by the ane character of the bration j
1
(1
n+1
) 1
n+1
,
assigning induces an orthogonal projection from the bers of V (j
1
()) to the
ones of V ( ) whose representation in local coordinates reads
_

q
i
_
j
1
()
G
AB
_
_

q
i
_
j
1
()
, i
_

z
A
_

_
_

z
A
_

=
= G
AB
_
_

q
i
_
j
1
()
,
_

q
j
_
j
1
()
_
_
j
z
B
_

_

z
A
_

=
= G
AB
g
ij
_
j
z
B
_

_

z
A
_

(B.4)
being now : V (j
1
()) V () the homomorphism (1.3.8).
By means of , to every
= (
1
, . . . ,
N1
) R
N1
we associate N 1
functions a
s
() according to the prescription
a
s
() := a
s
+
s

1
2

2
s

2
g
ij

i
_
j
( )
as
s = 1, . . . , N 1 (B.5)
For notational convenience, the family is completed by the constant functions
a
0
() = t
0
, a
N
() = t
1
.
In a similar way, given any vertical vector eld Y along , meant as a family
of elds Y
(s)
= Y
A
(s)
_

z
A
_

(s)
along the arcs of , for each V
h
we denote by
(s)
(, )
: (U
s
) U
s
, s = 1, . . . , N the (n + 1)parameter families of sections
described in coordinates as
z
A
(s)
= Y
A
(s)
(t) +
1
2

2
A
(s) i
(t)
i
(B.6)
with
A
(s) i
(t) := g
ik
G
AB
_
k
z
B
_

(B.7)
It goes without saying that, being strictly coordinatedependent, equation
(B.7) has no invariant geometrical meaning, but is merely a technical tool, whose
usefulness will be clear in the subsequent discussion.
Theorem B.1. Let be an admissible, piecewise dierentiable evolution and de-
note by (Y,
) an admissible innitesimal deformation of which vanishes at the

endpoints. Dene the metric and the functions
A
(s) i
(t), a
s
() as above. Then,
given any open subset V
h
with compact closure, there exist an > 0 and
a family
(, )
=
__
(s)
(, )
, [a
s1
(), a
s
()]
__
of piecewise dierentiable admissible
sections dened for [[ < , and fullling the following properties:
a)
(0, )
(t) = (t) ;
b)
(, )
(t
0
) = (t
0
) , ;
c)
(s)
(, )
(a
s
()) =
(s + 1)
(, )
(a
s
()) s = 1, . . . , N 1
d) each arc
(s)
(, )
(t), expressed in coordinates as q
i
(s)
=
i
(s)
(t, ,
i
), satises
the control equation
i
(s)
t
=
i
_
t ,
i
(s)
, Y
A
(s)
(t) +
1
2

2
A
(s) i

i
_
(B.8)
Proof. Let (U
s
, k
s
) be a family of local charts adapted to and A V
h
denote
an open set with compact closure containing

. A straightforward argument shows
the existence of an m > 0 such that the image
(s)
(,)
((U
s
)) is entirely contained
in U
s
for all A, [[ < m, s = 1, . . . , N.
We choose such an m R
+
and examine the situation separately in each chart
(U
s
, k
s
). There, solving equation (B.8) amounts to determining the integral curves
of the (n + 1)parameter family of vector elds Z
(s)
(, )
=

t
+ Z
i
(s)
q
i
on (U
s
),
with Z
i
(s)
=
i
_
t, q
k
, Y
A
(s)
(t) +
1
2

2
A
(s) h
(t)
h
_
.
99
This, in turn, is equivalent to determining the integral curves of a single vector
eld

Z
(s)
=

t
+ Z
i
(s)
q
i
in the product manifold (m, m) A (U
s
).
Let
(s)
(, )
(t; x) denote the integral curve of

Z
(s)
through the point (, , x).
Also, let c
s1
denote the corner (a
s1
). Then, on account of equations (A.2c),
chosen any
A, the curve
(s)
(0,
)
(t; c
s1
) coincides with the coordinate line
q
i
= 0, = 0, =
and is therefore dened for all t in an open interval

(b
s1
, b
s
) [a
s1
, a
s
].
By wellknown theorems in ordinary dierential equations [11, 22] this im-
plies the existence of an open neighborhood W
s1
(0,
, c
s1
) such that the
curve
(s)
(, )
(t; x) is dened for all (, , x) W
s1
and all t in the closed interval
_
t(x), a
s
()
(b
s1
, b
s
).
In particular, denoting by
s
the slice t = a
s
() in (m, m) A(U
s
), we
conclude that the 1parameter group of dieomorphisms determined by the eld
Z
(s)
maps the intersection W
s1

s1
into an open neighborhood of the point
(0,
, c
s
) in
s
. Without loss of generality we may always arrange for the image
of each W
s1

s1
to be contained in W
s

s
, s = 1, . . . , N.
The rest is now entirely straightforward: let U and
U
> 0 respectively
denote an open neighborhood of
in A and a positive number such that

1
(, , x
0
) W
0

0
[[ <
U
, U. For each [[ <
U
, U consider
the sequence of closed arcs
(s)
(, )
: [a
s1
(), a
s
()] (U
s
) dened inductively by
(1)
(, )
(t) =
(1)
(, )
(t; x
0
) t [t
0
, a
1
()]
(s + 1)
()
(t) =
(s + 1)
(, )
_
t;
(s)
()
(a
s
())
_
t [a
s
(), a
s+1
()]
The collection
(, )
:=
__
(s)
(, )
, [a
s1
(), a
s
()]
_
, s = 1, . . . , N
_
is then easily
recognized to dene an (n + 1)parameter family of continuous, piecewise dier-
entiable sections fullling all Theorems requirements. To complete our proof let
us nally recall that, for any
A, the family
(, )
exists for all in an open
neighborhood U
and all [[ <

U
. On the other hand, by the assumed com-
pactness of

, the subset A may be covered by a nite number of subsets
U
1
, . . . , U
k
of the required type.
The conclusion thus follows by choosing = min
U
1
, . . . ,
U
k
.
According to Theorem B.1, for any open subset V
h
with compact clo-
sure, the correspondence
(, )
(t
1
) sets up a 1parameter family of dieren-
tiable maps of into the hypersurface t = t
1
, with values in a neighborhood
of the point (t
1
). Moreover, given any dierentiable curve = () in ,
the 1parameter family of sections
(, ())
(t), [[ < , t [t
0
, t
1
] is a defor-
1
Notice that, according to our thesis, we are freezing the choice of the point x0 .
mation of tangent to the original innitesimal deformation X determined by
(Y,
1
, . . . ,
N1
) and leaving the rst endpoint (t
0
) xed.
Therefore, in order to nd an answer for our opening question, it just remains
to establish the existence of a curve () satisfying
(, ())
(t
1
) (t
1
) in some
open neighborhood of = 0.
In adapted coordinates, setting for simplicity
i
(, ) :=
i
(N)
(t
1
, , ), the
required condition reads
i
_
,
1
(), . . . ,
n
()
_
= 0 i = 1, . . . , n (B.9a)
Taking the relations
i
(0, ) = q
i
(N)
((t
1
)) = 0,
_
_
=0
= X
i
(t
1
) into ac-
count, a straightforward application of Taylors theorem shows that, whenever
the condition X(t
1
) = 0 holds true, namely whenever the eld Y and the co-
ecients
s
full equation (B.1), the functions
i
are necessarily of the form
i
(, ) =
2
i
(, ), with
i
(, ) regular at = 0. Under the stated assump-
tions, equation (B.9) is therefore equivalent to the condition
i
(,
1
, . . . ,
n
) = 0 i = 1, . . . , n (B.9b)
We will now discuss its solvability for the
i
s as functions of in a neigh-
borhood of = 0. To start with, we observe that the matching conditions c) of
Theorem B.1 give rise to relations of the form
i
(s + 1)
(a
s
(), , ) = q
i
(s + 1)
_
a
s
() ,
1
(s)
_
a
s
(), ,
_
, . . . ,
n
(s)
_
a
s
(), ,
_
_
q
i
(s + 1)
= q
i
(s + 1)
(t, q
1
(s)
, . . . , q
n
(s)
) denoting the transformation between adapted co-
ordinates in the intersection (U
s
U
s+1
). From these, deriving with respect to
we get the expressions
i
(s + 1)
t
da
s
d
+
i
(s + 1)
=
q
i
(s + 1)
t
da
s
d
+
q
i
(s + 1)
q
k
(s)
_
k
(s)
t
da
s
d
+
k
(s)
_
(B.10)
At = 0, recalling equations (1.5.34a), (A.1), (B.5) as well as the identication
X
i
(s)
=

i
(s)
=0
the latter provide the relation
X
i
(s + 1)
(a
s
) =
s
q
i
(s + 1)
t
cs
+ X
i
(s)
(a
s
) =
q
i
(s + 1)
t
cs
=
_
i
( )
_
as
(B.11)
In a similar way, on account of equations (A.1), (B.5), (B.11), deriving equation
(B.10) with respect to and evaluating everything at = 0, a straightforward
101
calculation yields the result
_
i
(s + 1)
i
(s)
2
_
cs
=
=
2
s
2
q
i
(s + 1)
t
2
+ 2
s
2
q
i
(s + 1)
t q
k
(s)
X
k
(s)
+
2
q
i
(s + 1)
q
h
(s)
q
k
(s)
X
h
(s)
X
k
(s)
+
2
s
_
dX
i
(s + 1)
dt

dX
i
(s)
dt
_
cs
+
2
s
_
i
( )
_
as
g
rk
_
r
( )
_
as
k
(B.12)
expressing the jumps
_
i
(s + 1)
i
(s)
2
_
cs
in terms of the section , of the
innitesimal deformation and of the variables
i
.
In addition to this let us now make use of the fact that, exactly as it happened
in 1.5.1 with equation (1.5.6b), in each adapted chart, by derivation of (B.8), we
get the evolution equations
t
_
i
(s)
2
_
=0
=
_

2
i
q
k
q
r
_

(s)
X
k
X
r
+ 2
_

2
i
q
k
z
A
_

(s)
X
k
Y
A
+
+
_

2
i
z
A
z
B
_

(s)
Y
A
Y
B
+
i
q
k
_

(s)
_
k
(s)
2
_
=0
+
_
i
z
A
_

(s)
A
(s) k

k
the cancelation arising from equation (A.2c).
From the latter, restoring the notation
i
(, ) for
i
(N)
(t
1
, , ) and recalling
equations (B.7), (B.12), as well as the components g
ij
being by denition
constant along , we get an expression of the form
=0
=
_
2
_
=0
=
= b
i
+
_
N
s=1
_
as
a
s1
_
i
z
A
_

(s)
(t)
A
(s) k
(t) dt +
N1
s=1
2
s
_
i
( )
_
as
g
jk
_
j
( )
_
as
_
k
=
= b
i
+
_
_
t
1
t
0
G
AB
_
i
z
A
_

_
j
z
B
_

dt +
N1
s=1
2
s
_
i
( )
_
as
_
j
( )
_
as
_
g
jk
k
(B.13)
with b
i
R depending solely on the section and on the original innitesimal
deformation. Collecting all results we can therefore state
Proposition B.1. Let : [t
0
, t
1
] 1
n+1
be a continuous, piecewise dierentiable,
admissible section. Then, if the matrix
S
ij
:=
_
t
1
t
0
G
AB
_
i
z
A
_

_
j
z
B
_

dt +
N1
s=1
2
s
_
i
( )
_
as
_
j
( )
_
as
(B.14)
is nonsingular, every innitesimal deformation of vanishing at the endpoints
is tangent to a nite deformation with xed endpoints.
Proof. The conclusion follows at once simply by observing that, on account of
equation (B.13), the nonsingularity of the matrix (B.14) ensures the solvability
of equations (B.9b) in a neighborhood of = 0 .
Proposition B.1 may be rephrased in the language of 1.5.4: whenever the
section is abnormal, Proposition 1.5.4 and equation (A.4) imply actually the
existence of at least one nonzero virtual 1form
i

i
|
with constant components
i
fullling the relations
i
_
i
z
A
_
(t)
= 0 ,
i
_
i
( )
as
= 0 (B.15)
and therefore automatically satisfying
i
S
ij
= 0, completely equivalent to the
singularity of the matrix (B.14).
More specically, denoting by p the abnormality index of , we have the fol-
lowing
Theorem B.2. The matrix (B.14) has rank n p.
Proof. By denition, the index p coincides with the dimension of the annihilator
_
(W)
_
0
V
h
, which is identical to the dimension of the space of constant

solutions of equations (B.15).
On the other hand, by equations (B.3), (B.14), the matrix S
ij
is positive
semidenite. Its kernel is therefore identical to the totality of zeroes of the quadratic
form
2
S
ij
j
, that is to the totality of ntuples (
1
, . . . ,
n
) R
n
fullling the
relation
0 =
_
_
t
1
t
0
G
AB
_
i
z
A
_

_
j
z
B
_

dt +
N1
s=1
2
s
_
i
( )
_
as
_
j
( )
_
as
_
j
=
=
_
t
1
t
0
G
AB
_
i
_
i
z
A
_

__
j
_
j
z
B
_

_
dt +
N1
s=1
2
s
_
i
_
i
( )
_
as
_
2
Because of the positive deniteness of G
AB
(t), the last condition is equiva-
lent to equations (B.15). This proves dim
_
ker(S
ij
__
= p which, in turn, entails
rank
_
S
ij
_
= n p.
In the language of 1.5.4, Proposition B.1 and Theorem B.2 show that the normal
evolutions form a subset of the ordinary ones, thus establishing Proposition 1.5.5.
Along the same lines, a deeper result is provided by the following
2
See Appendix D, Lemma D.1.
103
Theorem B.3. Let p ( 0) denote the abnormality index of the evolution .
Then a sucient condition for the ordinariness of is the existence of both an
(np)dimensional submanifold S 1
n+1
, contained in the slice t = t
1
and
including the point (t
1
), and an > 0 such that every deformation
which
leaves (t
0
) xed fulls the relation
(t
1
) S for all [[ < .
Proof. We assume the existence of both a submanifold S
i
1
n+1
and an > 0
with the stated properties. We also denote by (V,
1
, . . . ,
np
) a local chart in S
centered at the point (t
1
) and by
t = t
1
, q
i
(N)
=
i
_
1
, . . . ,
np
_
(B.16)
the representation of S in adapted coordinates.
By hypothesis, the correspondence (, )
(, )
(t
1
) factors through S for
any open subset V
h
with compact closure and for any (, ). This
gives rise to a dierentiable map g : (, ) S satisfying the relation
(, )
(t
1
) = i g(, ).
In coordinates, setting
(g(, )) = g
(,
1
, . . . ,
n
) and resuming the nota-
tion
i
(,
1
, . . . ,
n
) for q
i
(N)
(
(, )
(t
1
), this provides the identication
i
(,
1
, . . . ,
n
) =
i
_
g
1
(,
1
, . . . ,
n
) , . . . , g
np
(,
1
, . . . ,
n
)
_
(B.17)
From this, recalling the relation g
(0,
1
, . . . ,
n
) =
((t
1
)) = 0 as well as the
rank of the Jacobian
(
1

n
)
(
1

np
)
being maximal, it is easily seen that the equalities
i
(0,
1
, . . . ,
n
) =

i
(0,
1
, . . . ,
n
) = 0 are reected into analogous properties
of the functions g
.
By Taylors theorem we have therefore an expression of the form
g
=
2
(,
1
, . . . ,
n
) (B.18)
with the functions
regular at = 0.
The proof is thus reduced to establishing the solvability of the system
(,
1
, . . . ,
n
) = 0 (B.19)
for the
i
s as functions of in a neighborhood of = 0.
To this end, by direct computation, from equations (B.17), (B.18) we derive
the relation
_
2
_
=0
=
_

2
_
=0
+
_

i
2
g
2
_
=0
=
= 2
_
_
(0,...,0)
| =0
= 2
_
_
(t
1
)
(0,
1
, . . . ,
n
)
Together with equations (B.13), (B.14), the latter provides the identication
b
i
+ S
ir
g
rk

k
= 2
_
_
(t
1
)
(0,
1
, . . . ,
n
) (B.20)
In view of this, the functions
(0,
1
, . . . ,
n
) are therefore linear polynomials
(0,
1
, . . . ,
n
) = M
k

k
+ c
(B.21)
with coecients M
k
, c
uniquely determined in terms of b

i
, S
ir
, g
rk
and of the
imbedding (B.16). In particular, by equation (B.20), the rank of the matrix M
k
cannot be smaller than the one of S
ij
and, of course, cannot exceed n p. Ac-
cording to Theorem B.2, we have therefore rankM
k
= n p.
Collecting all results, we conclude:
the system (B.19) admits
p
solutions of the form (0,
1
, . . . ,
n
);
on account of equation (B.21), the Jacobian
_
_
_
(
1

np
)
(
1

n
)
_
_
_ has rank n p
at each point (0,
1
, . . . ,
n
). By continuity, it has therefore rank n p in a
neighborhood of every solution (0,
1
, . . . ,
n
) of equations (B.19).
By the implicit function theorem, this proves that the system (B.19) admits at
least a solution of the form
i
=
i
() in a neighborhood of = 0 (actually,
innitely many solutions whenever p > 0).
Appendix C
Admissible angular
deformations
Let : [t
0
, t
1
] 1
n+1
be a normal dierentiable evolution. If : [t
0
, t
1
] / is
the lift of we can refer / to a system of local bred coordinates (
U, t, q
i
, z
A
)
adapted to , as discussed in Appendix A.
Chosen both an arbitrary point t
(t
0
, t
1
) as well as point z = (t
, 0, z
A
) on
the bre
1
((t
)) /, for every
(0, t
t
0
) we can take into account the
control : U /, locally described as:
z
A
(t, q) =
_
_
0 t
0
t < t
z
A
t
t < t
0 t
t t
1
Theorem C.1. There exists > 0 such that for every < the equation
dq
i
dt
=
i
(t, q
i
, z
A
(t, q
i
))
with initial data q
i
(t
0
) = 0 admits a unique solution q
i
(t, ) which is continuous
over the interval [t
0
, t
1
] and piecewisedierentiable over (t
0
, t
1
), with corners lo-
cated in t
and t
.
Proof. As far as the interval [t
0
, t
) is concerned, the required solution is

evidently q
i
(t, ) = 0. Then, moving onto [t
, t
) and here considering the

dierential equation
dq
i
dt
=
i
(t, q
i
, z
A
) (C.1)
we can readily prove the existence of an > 0 such that equation (C.1) admits a
unique solution fullling the condition q
i
(t
, ) = 0 for every < . The values

q
i
taken by this solution when evaluated in t = t
can be assumedsmall (namely

106 Appendix C. Admissible angular deformations
of the same order as ) and may be used as initial data in t
for the dierential

equation
dq
i
dt
=
i
(t, q
i
, 0)
Therefore, by wellknown theorems in ordinary dierential equations, such equa-
tion is solvable up to the point t = t
1
, taking care of decreasing the value of
if necessary. As a result, we are given an admissible deformation q
i
=
i
(t, )
of the curve that is irreversible (since it is dened for > 0 only), that fulls
the condition lim
0
+
= and that, unlike the original evolution , is endowed

with a pair of corners.
A great improvement of Theorem C.1 is provided by the following:
Corollary C.1. If is a normal curve, then it is possible to alter the control
in the interval [t
, t
1
] in such a way that all the curves
pass through the same

point
(t
1
) = (t
1
) .
Proof. Let t = t
, q
i
= q
i
() be the orbit of the second corner of the deformation
and let

X = X
i
(t)
_

q
i
_

+Y
A
(t)
_

z
A
_

be an innitesimal deformation of the
arc ( , [t
, t
1
]), such that X
i
(t
) =
d q
i
d
=0
. Chosen a system of local coordinates
adapted to , the variational equation reads
X
i
(t) = X
i
(t
) +
_
t
t
i
z
A
_

Y
A
dt
Therefore, among the above described innitesimal deformations, the ones which
vanish in t = t
1
are in bijective correspondence with the vector elds Y
A
(t)
_

z
A
_

satisfying:
_
t
1
t
i
z
A
_

Y
A
dt = X
i
(t
)
Now let

X be an innitesimal deformation with the above properties. Following
the guidelines provided in Appendix B, in the interval [t
, t
1
] we substitute the
original control z
A
(t, q
i
) = 0 with
z
A
(t, q
i
) = Y
A
(t) +
1
2

2
A
i
(t)
i
where, passing over all the useless details,
A
i
(t) is an n r matrix while
= (
1
, . . . ,
n
) is a vector in R
n
. The quantities q
i
(t) are required to fulll
the dierential equation
dq
i
dt
=
i
(t, q
i
, Y
A
+
1
2

2
A
i
(t)
i
) , (C.2)
107
with initial data q
i
(t
, ) = q
i
(). Recalling the results of Appendix B, for su-
ciently small values of , the solution of the system (C.2) exists up to t = t
1
thus
determining a trajectory q
i
= q
i
(t
1
, ,
1
, . . . ,
n
) :=
i
(,
1
, . . . ,
n
).
Once again, we only need to determine a set of functions
i
=
i
() such that
i
(,
1
(), . . . ,
n
()) = 0. By Dinis theorem, this is only possible if the Jacobian
matrix

i
j
is nonsingular. In this connection, the following facts can be proved:
the relations
i
(0,
) = q
i
((t
1
)) = 0,
_
_
=0
= X
i
(t
1
) = 0, entail that
i
(,
) =
2
i
(,
),
i
(,
) being regular for 0

+
. The required identity can
be therefore expressed in the form:
i
(,
1
, . . . ,
n
) = 0 (C.3)
in a system of adapted coordinates, equation (C.2) yields the evolution equation
t
_
2
q
i
2
_
=0
=
_

2
i
q
k
q
r
_

X
k
X
r
+ 2
_

2
i
q
k
z
A
_

X
k
Y
A
+
_

2
i
z
A
z
B
_

Y
A
Y
B
+
_
i
z
A
_

A
k
k
whence
t
_
k
_
=0
=
_
i
z
A
_

A
k

i
k
=
_

i
z
A

A
k
dt (C.4)
the solvability of (C.3) is then equivalent to the nonsingularity of the matrix
(C.4) for at least one choice of the functions
A
i
, which is automatically guaranteed
by the normality of .
Appendix D
A touch of theory of quadratic
forms
Let V be a linear space
1
over R and let : V V R be a symmetric bilin-
ear functional. The mapping v (v, v) of V into R is called the quadratic
form associated to . If V is referred to a basis e
1
, . . . , e
n
, we then have the
representation
(v, v) = v
i
v
j
(e
i
, e
j
) =
ij
v
i
v
j
which actually shows the nature of (v, v) as a quadratic form in the variables v
i
.
Depending on the properties of their associated quadratic form, symmetric
bilinear functionals are classied as
indenite if, varying v, the quantity (v, v) may assume arbitrary real values;
positive (negative) semidenite if (v, v) 0 ( 0) v V .
In this connection, we have the following
Lemma D.1. A symmetric bilinear functional : V V R is semidenite if
and only if then its kernel
2
ker() coincides with the locus of zeroes of the quadratic
form (v, v) .
Proof. The vanishing of (u, u) for all u ker() is quite obvious. Lets see the
converse. If u V is such that (u, u) = 0, then
(u +v, u +v) = 2 (u, v) + (v, v)
1
For the time being, we suppose V to be nitedimensional. In case of need, however, we may
straightforwardly make all results that are drawn here t an innitedimensional context.
2
We recall that the kernel of a bilinear functional : V V R is dened as the set
ker() = {u| u V, (u, v) = 0 v V }.
110 Appendix D. A touch of theory of quadratic forms
for all v V , R. Because of the arbitrariness of , if the functional has
to be semidenite as it is by hypothesis the quantity (u, v) is necessarily
zero for all v V . This in turn implies u ker() .
Another possible way of looking at Lemma D.1 is that if we are given a sym-
metric bilinear functional on V and if we nd u, v V such that (u, u) = 0
but (u, v) ,= 0, then we can assert that is necessarily indenite.
A not singular semidenite symmetric bilinear functional is said to be denite.
According to Lemma D.1, this entails
positive (negative) denite (v, v) > 0 (< 0) v V, v ,= 0
We now conclude this brief Appendix by proving how the knowledge of the
denite character of the functional on both a subspace and a quotient space
enables to give a statement about its deniteness on the entire space.
Theorem D.1. Let K V be a linear subspace and W :=
V
/K the quotient space
of V by K. If the restriction of the symmetric bilinear functional : V V R
onto the subspace K is not singular, then:
i) for any v V , the restriction to the equivalence class [v] of the quadratic
form associated with has a single stationarity point v
;
ii) dening a map f : W R as f([v]) := (v
, v
) automatically sets up a
quadratic form on the quotient space W;
iii) if is positive denite, so is f ; conversely, the positive deniteness of both
f on W and on K implies the positive deniteness of on the whole of
V .
Proof. We consider a basis
, = 1, . . . , r = dimK, in the subspace K and

complete it to a basis
, e
i
of V . Every element v V is then represented in
components as v =
+ v
i
e
i
, while its equivalence class [v] is the ane space
formed by the totality of vectors u =
+ v
i
e
i
with xed v
i
s and arbitrary
s. The restriction to [v] of the quadratic form associated to the functional is

thus written in coordinates as
(u, u) =
+ 2
i

v
i
+
ij
v
i
v
j
whilst the search for its stationarity points is carried out by means of the equation
0 =

= 2
_
+
i
v
i
_
(D.1)
111
Hence, because of the nonsingularity of the matrix
, denoting by
its
inverse, we nd out
v
i
v
i
+ v
i
e
i
:=
+ v
i
e
i
(D.2)
This proves i). Assertion ii) is then easily seen to be selfevident simply by pointing
out that each element [v] has components v
i
with respect to the basis [e
i
] of
W and that the function f is represented in coordinates as
(v
, v
) =
+ 2
i

v
i
+
ij
v
i
v
j
=
_
ij

j
_
v
i
v
j
(D.3)
At last, if is positive denite, then
(v, v) > 0 v ,= 0 (v
, v
) > 0 v
,= 0 f([v]) > 0 [v] ,= 0

showing the positivity of f .
Conversely, if is positive denite when restricted to K, the stationarity point
v
that we worked out by means of equations (D.1), (D.2) is clearly a minimum,

the Hessian

2
being positive denite by hypothesis. Thus, if f is also positive

denite, for any v V , (v, v) (v
, v
) = f([v]) which, in particular, entails

(v, v) > 0 v / K. On the other hand, by hypothesis, (v, v) > 0 v K0
whence the conclusion.
Bibliography
[1] A. A. Agrachev and Yu.L. Sachov, Control Theory from the Geometric View-
point, Springer-Verlag, Berlin Heidelberg New York (2004).
[2] V. I. Arnold, Dynamical Systems III, Encyclopaedia of Mathematical Sciences,
Springer-Verlag, Berlin Heidelberg New York (1985).
[3] S. Benenti, Relazioni simplettiche, Pitagora Editrice, Bologna (1988).
[4] D. Bleecker, Gauge Theory and Variational Principles, Addison-Wesley Pub-
lishing Company, London, (1981).
[5] G.A. Bliss, Lectures on the calculus of the variations, The University of
Chicago Press, Chicago (1946).
[6] M. de Leon and P.R. Rodrigues, Methods of Dierential Geometry in Analyt-
ical Mechanics, North Holland, Amsterdam (1989).
[7] I.M. Gelfand and S.V. Fomin, Calculus of variations, Prentice-Hall Inc., En-
glewood Clis (1963).
[8] M. Giaquinta and S. Hildebrandt, Calculus of variations I, II , Springer-
Verlag, Berlin Heidelberg New York (1996).
[9] P. Griths, Exterior dierential systems and the calculus of variations,
Birkhauser, Boston (1983).
[10] M.R. Hestenes, Calculus of variations and optimal control theory, Wiley, New
York London Sydney (1966).
[11] W. Hurewicz, Lectures on ordinary dierential equations, John Wiley & Sons,
Inc., and MIT Press, New York and Cambridge, Mass. (1958). (Reprinted by
Dover Publ. (1990)).
[12] M.D. Intriligator, Mathematical Optimization and Economic Theory,
PrenticeHall, Inc., Englewood Clis, N.J. (1971).
114 Bibliography
[13] C. Lanczos, The variational principles of mechanics, University of Toronto
Press, Toronto (1949) (Reprinted by Dover Publ. (1970)).
[14] J.W. Milnor, Morse Theory, Annals of Mathematics Studies, Princeton Uni-
versity Press (1963).
[15] R. Montgomery, A Tour of Subriemannian Geometries, Their Geodesics and
Applications, AMS, Math. Surveys and Monographs, Vol. 91 (2000).
[16] J.F. Pommaret, Systems of Partial Dierential Equations and Lie Pseu-
dogroups, Gordon & Breach, New York (1978).
[17] L.S. Pontryagin, V.G. Boltyanskii, R.V. Gamkrelidze and E.F. Mishchenko,
The mathematical theory of optimal process, Interscience, New York (1962).
[18] H. Rund, The Hamilton-Jacobi theory in the calculus of variations, Van Nos-
trand, London (1966).
[19] H. Sagan, Introduction to the calculus of variations, McGrawHill Book Com-
pany, New York (1969).
[20] D.J. Saunders, The Geometry of Jet Bundles, London Mathematical Society,
Lecture Note Series 142, Cambridge University Press (1989).
[21] S. Sternberg, Lectures on Dierential Geometry, Prentice Hall, Englewood
Clis, New Jersey (1964).
[22] F. W. Warner, Foundations of Dierential Manifolds and Lie Groups,
SpringerVerlag, New York (1983).
[23] L. C. Young Lectures on the Calculus of Variations and Optimal Control The-
ory (second edition), AMS Chelsea Publishing, New York (1980).
[24] M.

Zefran, Continuous Methods for Motion Planning, Ph.D Thesis, University
of Pennsylvania, (1996).
[25] M. Crampin, Tangent Bundle Geometry for Lagrangian Dynamics, J. Phys. A:
Math. Gen., 16, 37553772 (1983).
[26] M.J. Gotay and J.M. Nester, Presymplectic Lagrangian systems I: the con-
straint algorithm and the equivalence theorem Ann. Inst. Henri Poincare,
Physique theorique, 30, 12942 (1979).
[27] L. Hsu, Calculus of Variations via the Griths Formalism, J. Dierential
Geometry 36, 551-589 (1992).
Bibliography 115
[28] E. Massa and E. Pagani, Classical Dynamics of non-holonomic systems: a
geometric approach, Ann. Inst. Henri Poincare, Physique theorique, 55,
511544 (1991).
[29] E. Massa and E. Pagani, Jet bundle geometry, dynamical connections, and
the inverse problem of Lagrangian Mechanics, Ann. Inst. Henri Poincare,
Physique theorique, 61, 1762 (1994).
[30] E. Massa and E. Pagani, A new look at Classical Mechanics of constrained
systems, Ann. Inst. Henri Poincare, Physique theorique, 66, 136 (1997).
[31] E. Massa, E. Pagani and P. Lorenzoni, On the gauge structure of Classical
Mechanics, Transport Theory and Statistical Physics, 29, 6991 (2000).
[32] E. Massa, S. Vignolo and D. Bruno, Nonholonomic Lagrangian and Hamilto-
nian Mechanics: an intrinsic approach, J. Phys. A: Math. Gen., 35, 67136742
(2002).
[33] E. Massa, D. Bruno and E. Pagani, Geometric control theory I: mathematical
foundations, Rev. Math. Phys., Vol. ??, ???? (2009).
[34] W. Sarlet, F. Cantrijn and M. Crampin, A New Look at Second Order Equa-
tions and Lagrangian Mechanics, J. Phys. A: Math. Gen., 17, 19992009
(1984).
[35] H. Sussmann and W.S. Liu Shortest paths for sub-Riemannian metrics on
rank-2 distributions, Memoirs of the American Mathematical Society, No. 564,
Vol. 118 (1995).

Univerrsity of Trento - Constrained Calculus of Variations and Geometric Optimal Control Theory

Загружено:

Сведения о документе

Исходное описание:

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

Univerrsity of Trento - Constrained Calculus of Variations and Geometric Optimal Control Theory

Загружено:

Авторское право:

Доступные форматы

Faculty of Mathematical, Physical and Natural Sciences

Doctoral School in Mathematics - XXII Cicle

() will be called the virtual 1forms along .

(). Also, given any local coordinate system t, q

() has some drawbacks in the case of piecewise

is a vector eld T(X) over T(1

R and a section X: R V (), then choose any vector eld

) and such that Y

, the lift of the eld Y at the

) takes the form

: P P , conventionally expressed through the additive notation

is any pair of trivializations, the

is then (the pullback of) a function over 1

F(P) and conversely, the

being expressed by the condition

. Moreover, both L(/)

respectively. Accordingly, all arrows in the front and rear faces

is also said to be weak (or strong ) if,

(t)) = q(t) + sin

tends to by the squeeze rule. However, we have

(t) is called the orbit of the deformation

of an admissible section : R / is called admissible if and only if

24 Chapter 1. Geometric setup

: R / denoting an admissible deformation of .

is described by the vector eld

X : R A( ) satisfying the consistency requirement i

h = id, is called an innitesimal control along ;

= id, the restriction to V () of

along may be lifted to a

() and vector valued functions

, this implies the relation

along , the denition of the operator

related to the basis

,= 0). Equation(1.5.44a) admits htransported solutions of the form

with xed endpoints.

dt over /, their respective action integrals 1[] =

L)dt is an exact dierential.

dt holds along any closed curve,

vanishing at the endpoints, whence also

) is that the dierence

: L(/) R of the bundle L(/) /. Let

is a trivialization of the bundle o

are continuous. A necessary

is said to preserve the endpoints of if and

is a deformation with xed endpoints. A vector eld along tangent

, the only condition needed in order for a vector

preserving the endpoints of we

and recalling the deni-

the jump of the function L( (t)) at t = a

: W R on the vector space W = V R

on the subset X W formed by the totality of elements

is any pair of elements in the class (), the dierence

, thus establishing assertion c).

() belonging to (). Because of the nature

(((/)) satisfying equa-

(H ) the pullback of the Pontryagin Hamiltonian,

(H), the inverse Legendre transformation q

fulls the relation

= 0 implies the identities

is just piecewisedierentiable for any ; time

for any ; time intervals [a

(s) = 0. Henceforth, every trans-

On account of Theorems 3.1.1, 3.1.2, the Hessian