Gauge Mechanics-Sardanashvily

Gauge Mechanics
(World Scientific, Singapore, 1998 )
L. MANGIAROTTI, G. SARDANASHVILY
Preface
This book presents in a unified way modern geometric methods in analytical me-
chanics, based on the application of jet manifolds and connections. As is well known,
the technique of Poisson and symplectic spaces provide the adequate Hamiltonian
formulation of conservative mechanics. This formulation, however, cannot be ex-
tended to time-dependent mechanics subject to time-dependent transformations.
We will formulate non-relativistic time-dependent mechanics as a particular field
theory on fibre bundles over a time axis.
The geometric approach to field theory is based on the identification of classical
fields with sections of fibred manifolds. Jet manifolds provide the adequate mathe-
matical language for Lagrangian field theory, while the Hamiltonian one is phrased
in terms of a polysymplectic structure. The 1-dimensional reduction of Lagrangian
field theory leads us in a straightforward manner to Lagrangian time-dependent
mechanics. At the same time, the canonical polysymplectic form on a momentum
phase space of time-dependent mechanics reduces to the canonical exterior 3-form
which plays the role similar to a symplectic form in conservative mechanics. With
this canonical 3-form, we introduce the canonical Poisson structure and formulate
Hamiltonian time-dependent mechanics in terms of Hamiltonian connections and
Hamiltonian forms.
Note that the theory of non-linear differential operators and the calculus of vari-
ations are conventionally phrased in terms of jet manifolds. On the other hand, jet
formalism provides the contemporary language of differential geometry to deal with
non-linear connections, represented by sections of jet bundles. Only jet spaces enable
us to treat connections, Lagrangian and Hamiltonian dynamics simultaneously.
In fact, the concept of connection is the main link throughout the book. Con-
nections on a configuration space of time-dependent mechanics are reference frames.
Holonomic connections on a velocity phase space define non-relativistic dynamic
v
vi
equations which are also related to other types of connections, and can be writ-
ten as non-relativistic geodesic equations. Hamiltonian time-dependent mechanics
deals with Hamiltonian connections whose geodesics are solutions of the Hamilton
equations.
The presence of a reference frame, expressed in terms of connections, is the main
peculiarity of time-dependent mechanics. In particular, each reference frame defines
an energy function, and quantizations with respect to different reference frames are
not equivalent.
Another important peculiarity is that a Hamiltonian fails to be a scalar function
under time-dependent transformations. As a consequence, many well-known con-
structions of conservative mechanics fail to be valid for time-dependent mechanics,
and one should follow methods of field theory.
At the same time, there is the essential difference between field theory and time-
dependent mechanics. In contrast with gauge potentials in field theory, connections
on a configuration space of time-dependent mechanics fail to be dynamic variables
since their curvature vanishes identically. Following geometric methods of field the-
ory, we obtain the frame-covariant formulation of time-dependent mechanics. By
analogy with gauge field theory, one may speak about gauge time-dependent me-
chanics.
In comparison with non-relativistic time-dependent mechanics, a configuration
space of relativistic mechanics does not imply any preferable fibration over a time.
To construct the velocity phase space of relativistic mechanics, we therefore use
formalism of jets of submanifolds. At the same time, Hamiltonian relativistic me-
chanics is seen as an autonomous Hamiltonian system on the constraint space of
relativistic hyperboloids.
With respect to mathematical prerequisites, the reader is expected to be familiar
with the basics of differential geometry of fibre bundles. For the convenience of the
reader, several mathematical facts and notions are included as an Interlude, thus
making our exposition self-contained.
Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1 Interlude: bundles, jets, connections 9

1.1 Fibre bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Multivector fields and differential forms . . . . . . . . . . . . . . . . . 20
1.3 Jet manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.4 Connections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
1.5 Bundles with symmetries . . . . . . . . . . . . . . . . . . . . . . . . . 46
1.6 Composite fibre bundles . . . . . . . . . . . . . . . . . . . . . . . . . 53
2 Geometry of Poisson manifolds 59

2.1 Jacobi structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.2 Contact structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.3 Poisson structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.4 Symplectic structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.5 Presymplectic structure . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.6 Reduction of symplectic and Poisson structures . . . . . . . . . . . . 86
2.7 Appendix. Poisson homology and cohomology . . . . . . . . . . . . . 91
2.8 Appendix. More brackets . . . . . . . . . . . . . . . . . . . . . . . . . 98
2.9 Appendix. Multisymplectic structures . . . . . . . . . . . . . . . . . . 103
3 Hamiltonian systems 107

3.1 Dynamic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.2 Poisson Hamiltonian systems . . . . . . . . . . . . . . . . . . . . . . . 113
3.3 Symplectic Hamiltonian systems . . . . . . . . . . . . . . . . . . . . . 116
3.4 Presymplectic Hamiltonian systems . . . . . . . . . . . . . . . . . . . 121
vii
viii CONTENTS
3.5 Dirac Hamiltonian systems . . . . . . . . . . . . . . . . . . . . . . . . 126

3.6 Dirac constraint systems . . . . . . . . . . . . . . . . . . . . . . . . . 131
3.7 Hamiltonian systems with symmetries . . . . . . . . . . . . . . . . . . 136
3.8 Appendix. Hamiltonian field theory . . . . . . . . . . . . . . . . . . . 142
4 Lagrangian time-dependent mechanics 153

4.1 Fibre bundles over R . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.2 Dynamic equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.3 Dynamic connections . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
4.4 Non-relativistic geodesic equations . . . . . . . . . . . . . . . . . . . 172
4.5 Reference frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
4.6 Free motion equations . . . . . . . . . . . . . . . . . . . . . . . . . . 181
4.7 Relative acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
4.8 Lagrangian systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
4.9 Newtonian systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
4.10 Holonomic constraints . . . . . . . . . . . . . . . . . . . . . . . . . . 209
4.11 Non-holonomic constraints . . . . . . . . . . . . . . . . . . . . . . . . 214
4.12 Lagrangian conservation laws . . . . . . . . . . . . . . . . . . . . . . 221
5 Hamiltonian time-dependent mechanics 229

5.1 Canonical Poisson structure . . . . . . . . . . . . . . . . . . . . . . . 230
5.2 Hamiltonian connections and Hamiltonian forms . . . . . . . . . . . . 233
5.3 Canonical transformations . . . . . . . . . . . . . . . . . . . . . . . . 244
5.4 The evolution equation . . . . . . . . . . . . . . . . . . . . . . . . . . 249
5.5 Degenerate systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
5.6 Quadratic degenerate systems . . . . . . . . . . . . . . . . . . . . . . 264
5.7 Hamiltonian conservation laws . . . . . . . . . . . . . . . . . . . . . . 271
5.8 Time-dependent systems with symmetries . . . . . . . . . . . . . . . 273
5.9 Systems with time-dependent parameters . . . . . . . . . . . . . . . . 276
5.10 Unified Lagrangian and Hamiltonian formalism . . . . . . . . . . . . 285
5.11 Vertical extension of Hamiltonian formalism . . . . . . . . . . . . . . 288
5.12 Appendix. Time-reparametrized mechanics . . . . . . . . . . . . . . . 298
6 Relativistic mechanics 301

6.1 Jets of submanifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
6.2 Relativistic velocity and momentum phase spaces . . . . . . . . . . . 305
CONTENTS ix
6.3 Relativistic dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 309

6.4 Relativistic geodesic equations . . . . . . . . . . . . . . . . . . . . . . 313
7 Appendix A. Geometry of BRST mechanics 319
8 Appendix B. On quantum time-dependent mechanics 329
Bibliography 333
x CONTENTS
Introduction
The present book deals with first order mechanical systems, governed by the second
order differential equations in coordinates or the first order ones in coordinates
and momenta. Our goal is the description of non-conservative mechanical systems
subject to time-dependent transformations, including inertial and non-inertial frame
transformations and phase transformations.
Symplectic technique is well known to provide the adequate Hamiltonian for-
mulation of conservative (i.e., time-independent) mechanics where Hamiltonians are
independent of time [2, 6, 72, 116, 126]. The familiar example is a mechanical sys-
tem whose momentum phase space is the cotangent bundle T ∗ M of a configuration
space M . This fibre bundle is provided with the canonical symplectic form
Ω = dpi ∧ dq i , (0.0.1)
written with respect to the holonomic coordinates (q i , pi = q̇i ) on T ∗ M . A Hamil-

tonian H of a conservative mechanical system is defined as a real function on the
momentum phase space T ∗ M . Then a motion of this system is an integral curve of
the Hamiltonian vector field
ϑ = ϑi ∂ i + ϑi ∂i
on T ∗ M which fulfills the Hamilton equations
ϑcΩ = −dH,
ϑi = ∂ i H, ϑi = −∂i H.
Lagrangian conservative mechanics is usually seen as a particular Hamiltonian me-

chanics on the tangent bundle T M of a configuration space M , which is endowed
with the presymplectic form defined by a Lagrangian.
1
2 INTRODUCTION
The Hamiltonian formulation of conservative mechanics cannot be extended in a

straightforward manner to time-dependent mechanics because the symplectic form
(0.0.1) is not invariant under time-dependent transformations, including the inertial
frame transformations.
The existent formulation of time-dependent mechanics implies a preliminary
splitting of a configuration space
Q = R × M, (0.0.2)
where M is a manifold, while R is a time axis (see [29, 31, 50, 110, 136, 146, 166] and
references therein). ¿From the physical viewpoint, it means that a certain reference
frame is chosen. Then we have the corresponding splitting of the velocity phase
space
R × TM (0.0.3)
and that of the momentum phase space
R × T ∗ M. (0.0.4)
The momentum phase space (0.0.4) is provided with the presymplectic form
pr∗2 Ω = dpi ∧ dq i (0.0.5)
which is the pull-back of the canonical symplectic form Ω (0.0.1) on the cotangent
bundle T ∗ M [27]. By a time-dependent Hamiltonian H is meant a real function
on the momentum phase space R × T ∗ M , while trajectories of motion are integral
curves of the time-dependent vector field
ϑ : R × T ∗M → T T ∗M
which satisfies the Hamilton equations
ϑi = ∂ i H, ϑi = −∂i H.
The problem is that the splittings (0.0.2) – (0.0.4) are broken by any time-
dependent transformation, and so is the presymplectic form (0.0.5). Therefore the
familiar methods of conservative mechanics and their extensions to the product
spaces (0.0.2) – (0.0.4) fail to be valid for mechanical systems subject to time-
dependent transformations.
INTRODUCTION 3
We will formulate non-relativistic time-dependent mechanics as a particular field

theory whose configuration space is a fibred manifold over a time axis R [14, 55, 57,
106, 114, 132, 159, 161].
Geometric formalism of field theory is based on the identification of classical
fields with sections of a fibred manifold Y → X. The corresponding velocity phase
space is the first order jet manifold J 1 Y of sections of Y → X, while the momentum
one is the Legendre bundle
n−1
Π = V ∗ Y ⊗( ∧ T ∗ X), n = dim X, (0.0.6)
over Y [28, 56, 57, 73, 96, 158, 159].
In the case of X = R of time-dependent mechanics, its configuration space is a
fibred manifold
π : Q → R,
equipped with fibred coordinates (t, q i ). The base R is parameterized by the Carte-
sian coordinates t with the transition functions t0 = t+const. Relative to these
coordinates, the time axis R is provided with the standard vector field ∂t and the
standard 1-form dt which is also the volume element on R. Of course, this is not
the case of relativistic mechanics (see Chapter 6) nor of the models with a time
reparametrization (see Section 5.12).
The velocity phase space of non-relativistic time-dependent mechanics is the first
order jet manifold J 1 Q of sections of the fibred manifold Q → R. It is equipped
with the adapted coordinates (t, q i , qti ). There is the canonical imbedding
λ : J 1 Q ,→ T Q, (0.0.7)
Q
λ = ∂t + qti ∂i ,
of the velocity phase space J 1 Q into the tangent bundle T Q of the configuration
space Q. From now on we will identify J 1 Q with its image in T Q given by the
coordinate conditions
ṫ = 1, q̇ i = qti . (0.0.8)
This is an affine subbundle of T Q → Q modelled over the vertical tangent bundle
V Q → Q of the fibred manifold Q → R.
The morphism (0.0.7) plays a prominent role in our formulation of time-dependent
mechanics. It enables us to treat the jet manifold J 1 Q as a velocity phase space
4 INTRODUCTION
of a mechanical system. Due to this morphism, every connection Γ on the fibred

manifold Q → R can be identified with the nowhere vanishing vector field
Γ : Q → J 1 Q ⊂ T Q, (0.0.9)
Γ = ∂t + Γi ∂i ,
on Q which is the horizontal lift of the standard vector field ∂t on R by means of

this connection Γ. We will continue to call (0.0.9) a connection in order to refer
to the standard properties of connections without additional explanation. ¿From
the physical viewpoint, a connection (0.0.9) sets a tangent vector at each point
of the configuration space Q, which characterizes the velocity of an ”observer” at
this point. It follows that a connection Γ on the fibred manifold Q → R defines a
reference frame [57, 132, 161]. In particular, one can think of the difference qti − Γi
as being the relative velocity with respect to the reference frame Γ, whereas the
notion of a relative acceleration is more intricate (see Section 4.7).
The momentum phase space of non-relativistic time-dependent mechanics is the
Legendre bundle (0.0.6) where X = R. This phase space is isomorphic to the vertical
cotangent bundle Π = V ∗ Q of the fibred manifold Q → R, and is equipped with the
holonomic coordinates (t, q i , pi = q̇i ). It should be emphasized that this is not the
most general case of a momentum phase space of time-dependent mechanics, which
is defined as a fibred manifold Π → R provided with a Poisson structure such that
the corresponding symplectic foliation belongs to the fibration Π → R [74]. In fact,
putting Π = V ∗ Q, we restrict our consideration to Hamiltonian systems which have
the Lagrangian counterparts.
Note that Lagrangian and Hamiltonian formalisms are equivalent only if a La-
grangian is hyperregular, i.e., the Legendre map from the velocity phase space to the
momentum one is a diffeomorphism. In general, a degenerate Lagrangian involves a
set of associated Hamiltonians in order to exhaust solutions of the Lagrange equa-
tion (see Section 5.5). Nevertheless, there are physically interesting systems whose
phase spaces fail to be the cotangent bundles of configuration spaces, and they do
not admit any Lagrangian description [168]. The unified Lagrangian–Hamiltonian
formalism of the joint velocity-momentum phase space
Π = V ∗J 1Q ∼
= J 1V ∗Q
enables us to relate a Lagrangian system to any Hamiltonian one (see Section 5.10).
INTRODUCTION 5
Let us turn to the momentum phase space V ∗ Q of time-dependent mechanics.

It is endowed with the canonical exterior 3-form
Ω = dpi ∧ dq i ∧ dt (0.0.10)
which is the particular case of the canonical polysymplectic form on the Legendre
bundle (0.0.6), when X = R [57, 161]. The exterior form (0.0.10) is invariant under
all holonomic transformations of the momentum phase space V ∗ Q.
In time-dependent mechanics, the canonical 3-form Ω (0.0.10) plays a role sim-
ilar to the canonical symplectic form (0.0.1) in conservative symplectic mechanics.
The form (0.0.10) yields the canonical Poisson structure on the momentum phase
space V ∗ Q, and provides the Hamiltonian formulation of time-dependent mechanics
in terms of Hamiltonian connections and Hamiltonian forms. This formulation is
compatible with the Lagrangian formulation of time-dependent mechanics on the
velocity phase space J 1 Q, and is equivalent to the Lagrangian one in the case of
hyperregular Lagrangians.
The following peculiarities of Hamiltonian time-dependent mechanics should be
emphasized.
• The canonical Poisson structure defined by the 3-form Ω (0.0.10) on the mo-
mentum phase space V ∗ Q of time-dependent mechanics is degenerate.
• A Hamiltonian on a momentum phase space of time-dependent mechanics fails

to be a scalar function, but reads
H = pi Γi + H
f ,
Γ (0.0.11)
where Γ is a connection on the fibred manifold Q → R, while H f is a Hamil-

Γ
tonian function which is also an energy density with respect to the reference
frame Γ.
• A Poisson bracket of a Hamiltonian (0.0.11) with functions on a momentum

phase space is defined only locally. Being equal to zero with respect to some
coordinates, it does not necessarily vanish with respect to other ones.
• As a consequence, the evolution equation in time-dependent mechanics is not

reduced to a Poisson bracket, and integrals of motion are not functions in invo-
lution with a Hamiltonian. For the same reason, the familiar Dirac–Bergmann
6 INTRODUCTION
algorithm for describing constraint systems is not extended to time-dependent

mechanics.
• Quantizations with respect to different reference frames are not equivalent.
For the sake of simplicity, throughout the book, a configuration space of time-
dependent mechanics Q → R is assumed to be a fibre bundle with a typical fibre
M . We will call it a configuration bundle. It is always trivial. Note that, although
such a configuration space Q is diffeomorphic to a direct product R × M , in general,
it cannot be canonically identified to R × M . Its different trivializations
ψ:Q∼
=R×M (0.0.12)
differ from each other in fibrations Q → M , while the fibration Q → R is once for all.
Given a trivialization (0.0.12) of a configuration space, there are the corresponding
splittings of velocity and momentum phase spaces
J 1Q ∼
= R × T M,
V ∗Q ∼
= R × T ∗ M.
We show that every trivialization (0.0.12) of the fibre bundle Q → R defines a

(complete) connection (0.0.9) on this fibre bundle, and vice versa. Consequently,
every such a trivialization corresponds to a (complete) reference frame.
Note that a reference frame is one of the main ingredients in our picture of
time-dependent mechanics. For instance, each reference frame defines an energy
function. Following geometric methods of field theory, we obtain the formulation of
time-dependent mechanics which involves necessarily connections Γ on a configura-
tion bundle Q → R, and thus is covariant under reference frame transformations.
By analogy with the gauge field theory, one may speak about gauge time-dependent
mechanics, though the term ”gauge mechanics” also stands for mechanics of de-
formable bodies [120] (see Section 5.11).
At the same time, there is an essential difference between field theory and time-
dependent mechanics. Since a configuration space of time-dependent mechanics is a
fibre bundle Q → R over a 1-dimensional base, the curvature of any connection on a
configuration bundle vanishes identically. In contrast with gauge potentials of gauge
theories, these connections fail to be dynamic quantities because they can always be
brought into the standard vector field Γ = ∂t by time-dependent transformations.
INTRODUCTION 7
Therefore, Lagrangians in time-dependent mechanics are covariant, but not invariant

under reference frame transformations.
Connections play a prominent role in our formulation of time-dependent mechan-
ics. As was mentioned above, connections on a configuration bundle Q → R describe
non-relativistic reference frames. Holonomic connections on the jet bundle J 1 Q → R
define non-relativistic dynamic equation which, in turn, are associated with connec-
tions on the affine jet bundle J 1 Q → Q and the tangent bundle T Q → Q. As a
result, every non-relativistic dynamic equation can be seen as a geodesic equation on
the tangent bundle T Q → Q that furnishes the relationship between non-relativistic
and relativistic dynamics (see Section 6.4). Hamiltonian time-dependent mechanics
deals with Hamiltonian connections whose geodesics are solutions of the Hamilton
equations.
In comparison with non-relativistic mechanics, if a configuration space of a me-
chanical system has no preferable fibration Q → R, we obtain the general formula-
tion of relativistic mechanics, including Special Relativity on the Minkowski space
Q = R4 . The velocity phase space of relativistic mechanics is the first order jet
manifold J11 Q of 1-dimensional submanifolds of the configuration space Q [57, 161].
This notion of jets generalizes that of jets of sections of fibre bundles which we have
utilized in field theory and non-relativistic mechanics. The jet bundle J11 Q → Q is
projective, and one can think of its fibres as being spaces of the 3-velocities of a rela-
tivistic system. The 4-velocities of a relativistic system are represented by elements
of the tangent bundle T Q of the configuration space Q, while the cotangent bundle
T ∗ Q, endowed with the canonical symplectic form, plays the role of the momentum
phase space of relativistic theory. As a result, Hamiltonian relativistic mechanics
can be seen as a constraint Dirac system on the hyperboloids of relativistic momenta
in the momentum phase space T ∗ Q.
Formalism of jets of submanifolds provides the common description of non-
relativistic mechanics and relativistic theory. In particular, the tangent bundle T Q
of a configuration space Q plays the role of the space of the 4-velocities both in non-
relativistic and relativistic mechanics. The difference is only that, given a fibration
Q → R, the 4-velocities of a non-relativistic system live in the subbundle (0.0.8) of
T Q, whereas the 4-velocities of a relativistic theory belong to the hyperboloids
gµν q̇ µ q̇ ν = 1, (0.0.13)
where g is an admissible pseudo-Riemannian metric in T Q. Moreover, as was men-
tioned above, both relativistic and non-relativistic equations of motion can be seen
8 INTRODUCTION
as geodesic equations on the tangent bundle T Q, but their solutions live in its dif-
ferent subbundles (0.0.13) and (0.0.8).
Unless otherwise stated, we believe that all quantities are physically dimension-
less. Following field theory, we will sometimes refer to the universal unit system
where the velocity of light c and the Planck constant h̄ are equal to 1, while the
length unit is the Planck one
(Gh̄c−3 )1/2 = G1/2 = 1, 616 · 10−33 cm,
where G is the Newtonian gravitational constant. Relative to the universal unit

system, the physical dimension of the spatial and temporal Cartesian coordinates
is the [length], the physical dimension of a mass is the [length]−1 , while an action
functional and a metric tensor are physically dimensionless.
Chapter 1
Interlude: bundles, jets,

connections
This Chapter does not claim to be a survey on modern differential geometry. The
relevant material is presented in a fairly informal way. For details, we refer the
reader to [57, 100, 157, 164, 170, 185].
Throughout the book, all maps are smooth, i.e., of class C ∞ , while manifolds
are real, finite-dimensional, second-countable and, hence, paracompact. Unless oth-
erwise stated, we assume that manifolds are connected.
We use the standard symbols ⊗, ∨, and ∧ for the tensor, symmetric, and exterior
products, respectively. The interior product (contraction) of vectors and forms is
denoted by c. By ∂BA are meant the partial derivatives with respect to the coordinates
with indices A B . The symbol ◦ stands for a composition of maps.
1.1 Fibre bundles

Subsections: Fibre bundles, 9; Vector bundles, 12; Affine bundles, 14; Tangent and
cotangent bundles, 15; Tangent and cotangent bundles of fibre bundles, 16; Sheaves,
18.
Fibre bundles
By a fibre bundle is meant a locally trivial fibred manifold
π:Y →X (1.1.1)
9
10 CHAPTER 1. INTERLUDE: BUNDLES, JETS, CONNECTIONS
where a fibration (or a projection) π is a surjective submersion ¿from a manifold Y ,

called a total space, onto a base X. Unless otherwise stated, we put dim X = n. By
definition, a base X of a fibre bundle (1.1.1) admits an open covering {Uξ } so that
Y is locally isomorphic to the splittings
ψξ : π −1 (Uξ ) → Uξ × V,
called local bundle trivializations, together with the transition functions
ρξζ : (Uξ ∩ Uζ ) × V → (Uξ ∩ Uζ ) × V,

ψξ (y) = (ρξζ ◦ ψζ )(y), y ∈ π −1 (Uξ ∩ Uζ ),
where V is the typical fibre of the fibre bundle (1.1.1). The bundle trivializations
(Uξ , ψξ ) constitute an atlas
Ψ = {(Uξ , ψξ ), ρξζ }
of a fibre bundle. Given an atlas Ψ, a fibre bundle Y is provided with the associated
atlas of fibred coordinates (xλ , y i ), where
xλ (y) = (xλ ◦ π)(y), y ∈ Y,
are coordinates on the base X, and
y i (y) = (y i ◦ pr2 ◦ ψξ )(y)
are coordinates on the typical fibre V .

A fibre bundle Y → X is called trivial if Y is diffeomorphic to the product X ×V .
Different trivializations of a fibre bundle differ from each other in projections Y → V
of the total space Y onto the typical fibre V .
Theorem 1.1.1. [170]. Each fibre bundle over a contractible base is trivial. 2
By a section (or a global section) of a fibre bundle (1.1.1) is meant a manifold

morphism s : X → Y such that π ◦ s = Id X. A section s is an imbedding, i.e.,
s(X) ⊂ Y is a submanifold of a total space Y which is also a topological subspace
of Y . Similarly, a section s of a fibre bundle Y → X over a submanifold N ⊂ X is
a morphism s : N → Y , such that
π ◦ s = iN : N ,→ Y.
1.1. FIBRE BUNDLES 11
A section of a fibre bundle over an open subset of its base will be called simply a
(local) section. A fibre bundle, by definition, admits a local section over an open
neighbourhood of each point of its base.
Theorem 1.1.2. [170]. A fibre bundle Y → X whose typical fibre is diffeomorphic

to Rm has a global section. A (smooth) section over a closed subset of X can always
be extended to a global section. 2
A fibred morphism of two bundles π : Y → X and π 0 : Y 0 → X 0 is a pair of maps

Φ : Y → Y 0 and f : X → X 0 such that the diagram
Φ
Y −→ Y 0
? ?
(1.1.2)
0
X −→ X
f
is commutative, i.e., Φ sends fibres to fibres. In brief, we will say that (1.1.2) is a
fibred morphism
Φ : Y −→ Y 0
f
over f . If f = Id X, then
Φ : Y −→ Y 0
X
is called a fibred morphism over X.

Remark 1.1.1. Unless otherwise stated, by the rank of a fibred morphism Φ (1.1.2)
over a diffeomorphism f is meant its rank minus dim X. •
A fibred morphism Φ (1.1.2) over X (or its image Φ(Y )) is said to be a subbundle
of the fibre bundle Y 0 → X if Φ(Y ) is a submanifold of Y 0 .
We deduce from the implicit function theorem the following useful criteria for
an image and a pre-image of a fibred morphism to be a subbundle [149, 185].
Theorem 1.1.3. Let Φ : Y → Y 0 be a fibred morphism over X. Given a global

section s0 of the fibre bundle Y 0 → X such that s0 (X) ⊂ Im Φ, by the kernel of the
fibred morphism Φ with respect to the section s0 is meant the pre-image
Ker s0 Φ = Φ−1 (s0 (X))

of s(X) by Φ. If Φ : Y → Y 0 is a fibred morphism of constant rank over X, then

Im Φ and Ker s0 Φ are subbundles of Y 0 and Y , respectively. 2
An isomorphism of fibre bundles is a fibred morphism (1.1.2) such that Φ is a

diffeomorphism. A fibred morphism [isomorphism] of a fibre bundle Y → X to itself
is called an endomorphism [automorphism]. An automorphism over Id X is said to
be a vertical automorphism. Following physical terminology, automorphisms of fibre
bundles will also be called gauge transformations.
Given a fibre bundle π : Y → X and a manifold map f : X 0 → X, the pull-back
f ∗ Y of Y by f is the fibre bundle over X 0 whose total space is
def
f ∗ Y ={(x0 , y) ∈ X 0 × Y : π(y) = f (x0 )}
together with the natural projection (x0 , y) 7→ x0 . Roughly speaking, the fibre of
the pull-back f ∗ Y over a point x0 ∈ X 0 is that of Y over the point f (x0 ) ∈ X. If
X 0 ⊂ X is a submanifold of X and iX 0 is the corresponding natural injection, then
the pull-back
i∗X 0 Y = Y |X 0
is called the restriction of a fibre bundle Y to the submanifold X 0 ⊂ X.

Let π : Y → X and π 0 : Y 0 → X be fibre bundles over the same base X. Their
fibred product
Y ×Y 0
over X is defined as the pull-back

∗
Y × Y 0 = π∗Y 0 or Y × Y 0 = π 0 Y
X X
together with the natural projection onto X.
Vector bundles
A vector bundle is a fibre bundle Y → X such that:
• its typical fibre V and all the fibres Yx = π −1 (x), x ∈ X, are real finite-
dimensional vector spaces;
• there is a bundle atlas Ψ = {(Uα , ψα )} of Y → X whose trivialization mor-

phisms ψα restrict to linear isomorphisms
ψα (x) : Yx → V, ∀x ∈ Uα .
Dealing with a vector bundle Y , we always use linear bundle coordinates (xλ , y i )
associated with the above-mentioned bundle atlas Ψ. We have
(pr2 ◦ ψα )(y) = y i ei ,
y = y i ei (x) = y i ψα (x)−1 (ei ),
where {ei } is a fixed basis for the typical fibre V of Y , while {ei (x)} is the fibre
basis (or the frame) for the fibre Yx of Y , which is associated with the bundle atlas
Ψ.
By virtue of Theorem 1.1.2, vector bundles have global sections, e.g., the global
zero section 0(X).
b If there is no risk of confusion, we write 0,
b instead of 0(X).
b
A morphism of vector bundles Φ : Y → Y 0 is defined as a fibred morphism over
f : X → X 0 whose restriction Φx : Yx → Yf0(x) to each fibre of Y is a linear map. It
is called a linear bundle morphism over f .
The following assertion is a corollary of Theorem 1.1.3.
Proposition 1.1.4. If Y → X and Y 0 → X are vector bundles and Φ : Y → Y 0 is a

linear bundle morphism of constant rank over X, then the image of Φ and the kernel
Ker b0 Φ of Φ with respect to the zero section 0b of Y 0 → X are vector subbundles of
Y 0 → X and Y → X, respectively. Note that a vector subbundle of a vector bundle
is a closed imbedded submanifold. 2
Unless otherwise stated, by Ker Φ of a linear bundle morphism Φ is meant its

kernel with respect to the zero section 0.
b
There are the following standard constructions of new vector bundles from old.
• Let Y → X be a vector bundle with a typical fibre V . By Y ∗ → X is meant

the dual vector bundle with the typical fibre V ∗ , dual of V . The interior
product of Y and Y ∗ is defined as a fibred morphism
c : Y ⊗ Y ∗ −→ X × R.
X
• Let Y → X and Y 0 → X be vector bundles with typical fibres V and V 0 ,

respectively. Their Whitney sum Y ⊕ Y 0 is a vector bundle over X with the
X
typical fibre V ⊕ V 0 .
• Let Y → X and Y 0 → X be vector bundles with typical fibres V and V 0 ,

respectively. Their tensor product Y ⊗ Y 0 is a vector bundle over X with the
X
typical fibre V ⊗ V 0 . Similarly, the exterior product Y ∧ Y is defined.
X
In particular, let Φ : Y1 → Y2 be an injection of a vector bundle π1 : Y1 → X1 to

a vector bundle π2 : Y2 → X2 over a diffeomorphism f : X1 → X2 . Then, there is
the dual surjection
Φ∗ : Y2∗ → Y1∗ ,
def
hΦ∗ (u), vi =hu, Φ(v)i, ∀v ∈ π1−1 (f −1 ◦ π2∗ (u)) ⊂ Y1 , u ∈ Y2∗ , (1.1.3)
over the diffeomorphism f −1 .
Affine bundles
Let π : Y → X be a vector bundle with a typical fibre V . An affine bundle

modelled over the vector bundle Y → X is a fibre bundle π : Y → X whose typical
fibre V is an affine space modelled over V , while the following conditions hold.
• All the fibres Yx of Y are affine spaces over the corresponding fibres Y x of the
vector bundle Y .
• There is a bundle atlas Ψ = {(Uα , ψα )} of Y → X whose local trivializations

restrict to affine isomorphisms
ψα (x) : Yx → V, ∀x ∈ Uα .
In particular, every vector bundle has a natural structure of an affine bundle.

Dealing with an affine bundle, we use only affine bundle coordinates (xλ , y i )
associated with the above-mentioned bundle atlas Ψ. We have the fibred morphisms
Y × Y −→ Y, (y i , y i ) 7→ y i + y i ,
X X
Y × Y −→ Y , (y i , y 0i ) 7→ y i − y 0i ,
X X
where (y i ) are linear coordinates on the vector bundle Y .

By virtue of Theorem 1.1.2, affine bundles have global sections.
A morphism of affine bundles Φ : Y → Y 0 is a fibred morphism over f whose
restriction Φx : Yx → Yf0(x) to each fibre of Y is an affine map. It is called an affine
bundle morphism over f .
Every affine bundle morphism Φ : Y → Y 0 from an affine bundle Y modelled
0
over a vector bundle Y to an affine bundle Y 0 modelled over a vector bundle Y
determines uniquely the linear bundle morphism
0
Φ:Y →Y ,
∂Φi j
y 0i ◦ Φ = y ,
∂y j
called the linear derivative of Φ.
Let Y × Y 0 be the fibred product of two affine bundles Y → X and Y 0 → X
X
0
which are modelled over the vector bundles Y → X and Y → X, respectively. This
0
product, called the Whitney sum, is also an affine bundle modelled over Y ⊕ Y .
X
0
Furthermore, let Y 0 → X be an affine bundle modelled over a vector bundle Y → X.
Let Y ⊂ Y 0 be an affine subbundle modelled over a vector bundle Y → X. Assume
0
that Y is the Whitney sum of Y and a complementary vector bundle Z → X. Then
one can easily verify that the affine bundle Y 0 → X decomposes in the Whitney sum
Y 0 = Y ⊕ Z.
X
Tangent and cotangent bundles
The fibres of the tangent bundle
πZ : T Z → Z
of a manifold Z are tangent spaces to Z. Given a coordinate atlas
ΨZ = {(Uξ , φξ )}
of a manifold Z, the tangent bundle is provided with the (holonomic) atlas
Ψ = {(πz−1 Uξ ), ψξ = T φξ )},
where by T φξ is meant the tangent map to φξ . The associated linear bundle coor-
dinates (ż λ ) on T Z are called the induced (or holonomic) coordinates with respect
to the frames {∂λ } in tangent spaces Tz Z. Their transition functions read
∂z 0λ µ
ż 0λ = ż .
∂z µ
Every manifold morphism f : Z → Z 0 generates the fibred morphism of the tangent
bundles
T f : T Z −→ T Z 0 ,
f
∂f λ µ
ż 0λ ◦ T f = ż ,
∂z µ
called the tangent map to f .
The cotangent bundle of a manifold Z is the dual
π∗Z : T ∗ Z → Z
of the tangent bundle T Z → Z. It is equipped with the (holonomic) coordinates

(z λ , żλ ) with respect to the coframes {dz λ } dual of {∂λ }. Their transition functions
read
∂z µ
żλ0 = żµ .
∂z 0λ
Tangent and cotangent bundles of fibre bundles
Let πY : T Y → Y be the tangent bundle of a fibre bundle π : Y → X. Given

fibred coordinates (xλ , y i ) on Y , the tangent bundle T Y is equipped with the coor-
dinates (xλ , y i , ẋλ , ẏ i ).
The tangent bundle T Y is a fibre bundle
π ◦ πY : T Y → X
over X, while the tangent map T π to π defines the fibration
T π : T Y → T X.
There is the commutative diagram

Tπ
T Y −→ T X
? ?
Y −→ X
π
The tangent bundle T Y → Y of a fibre bundle Y → X has the vertical tangent

subbundle
def
V Y = Ker T π
of T Y , given by the coordinate relation ẋλ = 0. This subbundle consists of the

vectors tangent to fibres of Y . The vertical tangent bundle V Y is provided with the
coordinates (xλ , y i , ẏ i ) with respect to the frames {∂i }.
Let T Φ be the tangent map to a fibred morphism Φ : Y → Y 0 . Its restriction
V Φ = T Φ|V Y : V Y → V Y 0 ,
ẏ 0i ◦ V Φ = ∂V Φi = ẏ j ∂j Φi , (1.1.4)
to V Y is a linear bundle morphism of the vertical tangent bundle V Y to the vertical

tangent bundle V Y 0 , called the vertical tangent map to Φ.
Every vector bundle Y → X admits the canonical vertical splitting of the vertical
tangent bundle
VY ∼
= Y ×Y (1.1.5)
X
because the coordinates ẏ i on V Y have the same transformation law as the linear
coordinates y i on Y .
An affine bundle Y → X modelled over a vector bundle Y → X also admits the
canonical vertical splitting of the vertical tangent bundle
VY ∼
= Y ×Y (1.1.6)
X
because the coordinates ẏ i on V Y have the same transformation law as the linear
coordinates y i on the vector bundle Y .
The cotangent bundle T ∗ Y of a fibre bundle Y → X is equipped with the
coordinates (xλ , y i , ẋλ , ẏi ). There is its natural fibration T ∗ Y → X over X, but not
over T ∗ X.
The vertical cotangent bundle V ∗ Y → Y of a fibre bundle Y → X is defined

as the vector bundle dual of the vertical tangent bundle V Y → Y . It should be
emphasized that there is no canonical injection of V ∗ Y into the cotangent bundle
T ∗ Y of Y , but we have the canonical projection
ζ : T ∗ Y → V ∗ Y, (1.1.7)
Y
ζ : ẋλ dx + ẏi dy i 7→ ẏi dy i ,
λ
where {dy i } are the bases for fibres of V ∗ Y , which are dual of the frames {∂i } in
the vertical tangent bundle V Y .
With V Y and V ∗ Y , we have the following two exact sequences of vector bundles
over Y :
0 → V Y ,→ T Y → Y × T X → 0, (1.1.8)
X
0 → Y × T ∗ X ,→ T ∗ Y → V ∗ Y → 0. (1.1.9)
X
For the sake of simplicity, we will denote the pull-backs
Y × T X, Y × T ∗X
X X
simply by T X and T ∗ X.
Example 1.1.2. Let us consider the tangent bundle T T ∗ X of T ∗ X and the cotan-
gent bundle T ∗ T X of T X. Relative to coordinates (xλ , pλ = ẋλ ) on T ∗ X and (xλ , ẋλ )
on T X, these fibre bundles are provided with the coordinates (xλ , pλ , ẋλ , ṗλ ) and
(xλ , ẋλ , ẋλ , ẍλ ), respectively. By inspection of the coordinate transformation laws,
one can show that there is the isomorphism
α : T T ∗X ∼
= T ∗ T X, pλ ←→ ẍλ , ṗλ ←→ ẋλ (1.1.10)
of these bundles over T X [43, 96].

Given a fibre bundle Y → X, there is the similar isomorphism
αV : V V ∗ Y ∼
= V ∗ V Y, pi ←→ ÿi , ṗi ←→ ẏi (1.1.11)
over V Y , where (xλ , y i , pi , ẏ i , ṗi ) and (xλ , y i , ẏ i , ẏi , ÿi ) are coordinates on V V ∗ Y and
V ∗ V Y , respectively. •
Sheaves
There are several equivalent definitions of sheaves [16, 80]. We will start from
the following. A sheaf on a topological space X is a topological fibre bundle S →
X whose fibres, called the stalks, are Abelian groups Sx provided with discrete
topology.
A presheaf on a topological space X is defined if an Abelian group SU corresponds
to every open subset U ⊂ X (S∅ = 0) and, for any pair of open subsets V ⊂ U ,
there is the homomorphism
rVU : SU → SV
such that
rUU = Id SU ,
U V U
rW = rW rV , W ⊂ V ⊂ U.
Example 1.1.3. Let X be a topological space, SU the additive Abelian group of

all continuous functions on U ⊂ X, while the homomorphism
rVU : SU → SV
is the restriction of these functions to V ⊂ X. Then {SU , rVU } is a presheaf. •
Every presheaf {SU , rVU } on a topological space X yields a sheaf on X whose

stalk Sx at a point x ∈ X is the direct limit of the Abelian groups SU , x ∈ U , with
respect to the homomorphisms rVU . It means that, for each open neighbourhood U
of a point x, every element s ∈ SU determines an element sx ∈ Sx , called the germ
of s at x. Two elements s ∈ SU and s0 ∈ SV0 define the same germ at x if and only
if there is an open neighbourhood W 3 x such that
U V 0
rW s = rW s.
For instance, two real functions s and s0 on X define the same germ sx if they
coincide on an open neighbourhood of x. The sheaf generated by the presheaf in
Example 1.1.3 is called the sheaf of continuous functions. The sheaf of smooth
functions on a manifold X is defined in a similar way.
Two different presheaves may generate the same sheaf. Conversely, a sheaf de-
fines a presheaf of Abelian groups Γ(U, S) of local sections of the sheaf S. This
presheaf {Γ(U, S), rVU } is called the canonical presheaf of the sheaf S. It is eas-
ily seen that the sheaf generated by the canonical presheaf {Γ(U, S), rVU } of the
sheaf S coincides with S. Therefore, we will further identify sheaves and canonical
presheaves.
Example 1.1.4. Let Y → X be a vector bundle. The germs of its sections make
up the sheaf S(Y ) of sections of Y → X. The stalk Sx (Y ) of this sheaf at a point
x ∈ X consists of the germs of sections of Y → X in a neighbourhood of x ∈ X.
The stalk Sx (Y ) is a module over the ring Cx∞ (X) of the germs at x ∈ X of smooth
functions on X. If we deal with a tangent bundle T X → X, the stalk Sx (T X) is a
Lie algebra with respect to the Lie bracket of vector fields. •
1.2 Multivector fields and differential forms

Subsections: Vector fields, 20; Vector fields on fibre bundles, 21; Multivector fields,
23; The Schouten–Nijenhuis bracket, 24; Exterior forms, 26; Exterior forms on fibre
bundles, 27; Interior products, 28; Bivector fields and 2-forms, 29; The Lie derivative,
31; Tangent-valued forms, 32; Distributions, 33; Foliations, 35.
Vector fields
A vector field on a manifold Z is defined as a global section of the tangent bundle

T Z → Z. The set T (Z) of vector fields on Z is both a module over the ring C ∞ (Z)
of smooth functions on Z and a real Lie algebra with respect to the Lie bracket
[v, u] = (v λ ∂λ uµ − uλ ∂λ v µ )∂µ , u = uλ ∂λ , v = v λ ∂λ .
A curve c : () → Z, () ⊂ R, in Z is said to be an integral curve of a vector field

u on Z if
ċ = u ◦ c,
ċλ (t) = uλ (c(t)), t ∈ ().
Recall that, for every point z ∈ Z, there exists a unique integral curve
c : (−, ) → Z, > 0,
of a vector field u through z = c(0).

A vector field u on an imbedded submanifold N ⊂ Z is said to be a section of
the tangent bundle T Z → Z over N . It should be emphasized that this is not a
1.2. MULTIVECTOR FIELDS AND DIFFERENTIAL FORMS 21
vector field on a manifold N since u(N ) does not belong to T N ⊂ T X in general.

A vector field on a submanifold N ⊂ Z is called tangent to the submanifold N if
u(N ) ⊂ T N .
Let U ⊂ Z be an open subset and > 0. By a local 1-parameter group of local
diffeomorphisms of Z defined on (−, ) × U is meant a mapping
G : (−, ) × U 3 (t, z) 7→ Gt (z) ∈ Z
which possesses the following properties:
• for each t ∈ (−, ), the mapping Gt is a diffeomorphism of U onto the open
subset Gt (U ) ⊂ Z;
• Gt+t0 (z) = (Gt ◦ Gt0 )(z) if t + t0 ∈ (−, ).
If such a mapping G is defined on R × Z, it is called a 1-parameter group of diffeo-

morphisms of Z.
Theorem 1.2.1. [100]. Each local 1-parameter group of local diffeomorphisms G

on U ⊂ Z defines a local vector field u on U by setting u(z) to be the tangent vector
to the curve s(t) = Gt (z) at t = 0. Conversely, let u be a vector field on a manifold
Z. For each z ∈ Z, there exist a number > 0, a neighbourhood U of z and a unique
local 1-parameter group of local diffeomorphisms on (−, ) × U , which determines
u. 2
In brief, every vector field u on a manifold Z is the generator of a local 1-

parameter group of local diffeomorphisms. In particular, every exterior form φ on
a manifold Z is invariant under a local 1-parameter group of local diffeomorphisms
Gu with the generator u, i.e.,
g ∗ φ = φ, ∀g ∈ Gu ,
if and only if its Lie derivative Lu φ along u vanishes.

If a vector field u on a manifold Z is induced by a 1-parameter group of diffeo-
morphisms of Z, then u is called a complete vector field.
Vector fields on fibre bundles

A vector field u on a fibre bundle Y → X is said to be projectable if it projects

over a vector field uX on X, i.e., if the following diagram
u
Y −→ T Y
π Tπ
? ?
X −→
u
TX
X
is commutative. A projectable vector field has the coordinate expression
u = uλ (xµ )∂λ + ui (xµ , y j )∂i , uX = uλ ∂λ .
A vector field τ = τ λ ∂λ on a base X of a fibre bundle Y → X can give rise to

a vector field on Y , projectable over τ , by means of some connection on this fibre
bundle (see (1.4.7) below). Nevertheless, a tensor bundle
m k
T = (⊗ T X) ⊗ (⊗ T ∗ X),
admits the canonical lift

2 ···αm
∂
τe = τ µ ∂µ + [∂ν τ α1 ẋνα
β1 ···βk + . . . − ∂β1 τ ν ẋανβ12···α
···βk − . . .]
m
···αm (1.2.1)
∂ ẋαβ11···β k
of any vector field τ on X. In particular, there exist the canonical lift

∂
τe = τ µ ∂µ + ∂ν τ α ẋν (1.2.2)
∂ ẋα
of τ onto the tangent bundle T X, and its canonical lift
∂
τe = τ µ ∂µ − ∂β τ ν ẋν (1.2.3)
∂ ẋβ
onto the cotangent bundle T ∗ X. Hereafter, we will use the compact notation
∂
∂˙λ = . (1.2.4)
∂ ẋλ
A projectable vector field u = ui ∂i on a fibre bundle Y → X is said to be vertical
if it projects over the zero vector field uX = 0 on X.
Let Y → X be a vector bundle. Using the canonical vertical splitting (1.1.5),
we obtain the canonical vertical vector field
uY = y i ∂i (1.2.5)
on Y , called the Liouville vector field. For instance, the Liouville vector field on the
tangent bundle T X reads
uT X = ẋλ ∂˙λ . (1.2.6)
Accordingly, any vector field τ = τ λ ∂λ on a manifold X has the canonical vertical
lift
τV = τ λ ∂˙λ (1.2.7)
onto the tangent bundle T X.
Multivector fields
A multivector field ϑ of degree | ϑ |= r (or simply an r-vector field) on a manifold

Z is a section
1
ϑ = ϑλ1 ...λr ∂λ1 ∧ · · · ∧ ∂λr
r!
r
of the exterior product ∧ T Z → Z. Let us denote by Tr (Z) the vector space of
r-vector fields on Z. In particular, T1 (Z) is the space of vector fields on Z (denoted
by T (Z) for the sake of simplicity), while T0 (Z) is the vector space C ∞ (Z) of smooth
functions on Z. All multivector fields on a manifold Z make up the real Z-graded
vector space T∗ (Z) which is also a Z-graded exterior algebra with respect to the
exterior product of multivector fields.
Given a manifold Z, the tangent lift ϑe onto T Z of an r-vector field ϑ on Z is
defined by the relation
ϑ(
e σer, . . . , σ
e 1 ) = ϑ(σ r ,g
. . . , σ1) (1.2.8)
where: (i) σ k = σλk dxλ are arbitrary 1-forms on the manifold Z, (ii) by
σe k = ẋµ ∂µ σλk dxλ + σλk dẋλ
are meant their tangent lifts (1.2.24) onto the tangent bundle T Z of Z, and (iii)
the right-hand side of the equality (1.2.8) is the tangent lift (1.2.22) onto T Z of the
function ϑ(σ r , . . . , σ 1 ) on Z [67]. We then have the coordinate expression
1
ϑ = ϑλ1 ...λr ∂λ1 ∧ · · · ∧ ∂λr ,
r!
1
ϑe = [ż µ ∂µ ϑλ1 ...λr ∂˙λ1 ∧ · · · ∧ ∂˙λr + (1.2.9)
r!
r
ϑλ1 ...λr ∂˙λ1 ∧ · · · ∧ ∂λi ∧ · · · ∧ ∂˙λr ].
X
i=1
In particular, if τ is a vector field on a manifold Z, its tangent lift (1.2.9) coincides

with the canonical lift (1.2.2). If an r-vector field ϑ is simple, i.e.,
ϑ = τ 1 ∧ · · · ∧ τ r,
its tangent lift (1.2.9) reads

r
τV1 ∧ · · · ∧ τei · · · ∧ τVr ,
X
ϑe =
i=1
where τVk is the vertical lift (1.2.7) onto T Z of the vector field τ k .
Example 1.2.1. The tangent lift of a bivector field
1
w = wµν ∂µ ∧ ∂ν
2
is
1
we = (ż λ ∂λ wµν ∂˙µ ∧ ∂˙ν + wµν ∂µ ∧ ∂˙ν + wµν ∂˙µ ∧ ∂ν ).
2
•
Schouten–Nijenhuis bracket
The exterior algebra of multivector fields on a manifold Z is provided with the

Schouten–Nijenhuis bracket which generalizes the Lie bracket of vector fields as
follows [13, 181]:
[., .]SN : Tr (M ) × Ts (M ) → Tr+s−1 (M ), (1.2.10)

1 1
ϑ = ϑλ1 ...λr ∂λ1 ∧ · · · ∧ ∂λr , υ = υ α1 ...αs ∂α1 ∧ · · · ∧ ∂αs ,
r! s!
def rs
[ϑ, υ]SN = ϑ ? υ + (−1) υ ? ϑ,
r
ϑ?υ = (ϑµλ2 ...λr ∂µ υ α1 ...αs ∂λ2 ∧ · · · ∧ ∂λr ∧ ∂α1 ∧ · · · ∧ ∂αs ).
r!s!
There following relations hold:
[ϑ, υ]SN = (−1)|ϑ||υ| [υ, ϑ]SN , (1.2.11)

[ν, ϑ ∧ υ]SN = [ν, ϑ]SN ∧ υ + (−1)(|ν|−1)|ϑ| ϑ ∧ [ν, υ]SN , (1.2.12)
|ν|(|υ|−1) |ϑ|(|ν|−1)
(−1) [ν, [ϑ, υ]SN ]SN + (−1) [ϑ, [υ, ν]SN ]SN + (1.2.13)
|υ|(|ϑ|−1)
(−1) [υ, [ν, ϑ]SN ]SN = 0.
Example 1.2.2. Let

1
2
be a bivector field. Its Schouten–Nijenhuis bracket reads
[w, w]SN = wµλ1 ∂µ wλ2 λ3 ∂λ1 ∧ ∂λ2 ∧ ∂λ3 .
The Schouten–Nijenhuis bracket commutes with the tangent lift (1.2.9) of mul-
tivectors [67], i.e.,
[ϑ,
e υe]SN = [ϑ,
gυ] .
SN (1.2.14)
Remark 1.2.3. Let us point out another sign convention used in the definition of
the Schouten–Nijenhuis bracket [125]. This bracket, denoted by [., .]SN0 , is
[ϑ, υ]SN0 = −(−1)|ϑ| [ϑ, υ]SN . (1.2.15)
The relation (1.2.11) for this bracket reads
[ϑ, υ]SN0 = (−1)(|ϑ|−1)(|υ|−1) [υ, ϑ]SN0 . (1.2.16)
The relation (1.2.12) keeps its form, i.e.,
[ν, ϑ ∧ υ]SN0 = [ν, ϑ]SN0 ∧ υ + (−1)(|ν|−1)|ϑ| ϑ ∧ [ν, υ]SN0 , (1.2.17)
while the relation (1.2.13) is replaced by
(−1)(|ν|−1)(|υ|−1) [ν, [ϑ, υ]SN0 ]SN0 + (−1)(|ϑ|−1)(|ν|−1) [ϑ, [υ, ν]SN0 ]SN0 + (1.2.18)
(−1)(|υ|−1)(|ϑ|−1) [υ, [ν, ϑ]SN0 ]SN0 = 0.
The equalities (1.2.16) and (1.2.18) show that, with the modified Schouten–Nijenhuis
bracket (1.2.15), the Z-graded vector space T∗ (Z) of multivector fields on a manifold
Z is a graded Lie algebra, where the Lie degree of a multivector field ϑ is | ϑ | −1.
In particular,
def
adϑ(υ) =[ϑ, υ]SN0 (1.2.19)
is a graded endomorphism of degree | ϑ | −1 of the graded Lie algebra T∗ (Z). If ϑ

is a vector field, the endomorphism (1.2.19) is the Lie derivative
adϑ(υ) = Lϑ υ (1.2.20)
of the multivector field υ along ϑ. •
Exterior forms
An exterior r-form on a manifold Z is a section

1
φ = φλ1 ...λr dz λ1 ∧ · · · ∧ dz λr
r!
r
of the exterior product ∧ T ∗ Z → Z. We denote by Or (Z) the vector space of exterior
r-forms on a manifold Z. This is also a module over the ring O0 (Z) = C ∞ (Z).
¿From now on we will use the notation C ∞ (Z) for the ring of smooth functions on
a manifold Z, while O0 (Z) stands for the vector space of these functions as a rule.
All exterior forms on Z constitute the exterior Z-graded algebra O∗ (Z) with
respect to the exterior product. The exterior differential is the first order differential
operator
d : Or (Z) → Or+1 (Z),

1
dφ = ∂µ φλ1 ...λr dz µ ∧ dz λ1 ∧ · · · dz λr ,
r!
on O∗ (Z). It obeys the relations
d ◦ d = 0,
d(φ ∧ σ) = d(φ) ∧ σ + (−1)|φ| φ ∧ d(σ),
where | φ | is the degree of φ.

Given a manifold map f : Z → Z 0 , by f ∗ φ is meant the pull-back on Z of an
r-form φ on Z 0 by f , which is defined by the condition
f ∗ φ(v 1 , . . . , v r )(z) = φ(T f (v 1 ), . . . , T f (v r ))(f (z)), ∀v 1 , · · · v r ∈ Tz Z.
We have the relations
f ∗ (φ ∧ σ) = f ∗ φ ∧ f ∗ σ,
df ∗ φ = f ∗ (dφ).
For instance, if iN : N → Z is a submanifold, the pull-back i∗N φ onto N is called

the restriction of an exterior form φ to N .
Exterior forms on fibre bundles
Let π : Y → X be a fibre bundle with fibred coordinates (xλ , y i ). The pull-back

on Y of exterior forms on X by π provides the inclusion
π ∗ : O∗ (X) → O∗ (Y ).
Exterior forms
r
φ : Y → ∧ T ∗ X,
1
φ = φλ1 ...λr dxλ1 ∧ · · · ∧ dxλr ,
r!
on Y such that ϑcφ = 0 for arbitrary vertical vector field ϑ on Y are said to be
horizontal forms. A horizontal n-form is called a horizontal density. We will use the
notation
ω = dx1 ∧ · · · ∧ dxn , ωλ = ∂λ cω. (1.2.21)
In the case of the tangent bundle T X → X, there is a different way, besides

the pull-back, to lift onto T X the exterior forms on X [67, 110, 189]. Let f be a
function on X. Its tangent lift onto T X is defined as the function
fe = ẋλ ∂λ f. (1.2.22)
Let σ be an r-form on X. Its tangent lift onto T X is said to be the r-form σe given
by the relation
σe (τe1 , . . . , τer ) = σ(τ1 ,g

. . . , τr ), (1.2.23)
where τi are arbitrary vector fields on T X, and τei are their canonical lifts (1.2.2)
onto T X. We have the coordinate expression
1
σ= σλ ···λ dxλ1 ∧ · · · ∧ dxλr ,
r! 1 r
1
σe = [ẋµ ∂µ σλ1 ···λr dxλ1 ∧ · · · ∧ dxλr + (1.2.24)
r!
r
σλ1 ···λr dxλ1 ∧ · · · ∧ dẋλi ∧ · · · ∧ dxλr ].
X
i=1
The following equality holds:
dσe = dσ.
f
Example 1.2.4. Given a 2-form

1
Ω = Ωµν dxµ ∧ dxν
2
on a manifold X, its tangent lift (1.2.24) onto T X reads
e = 1 (ẋλ ∂ Ω dxµ ∧ dxν + Ω dẋµ ∧ dxν + Ω dxµ ∧ dẋν ).

Ω (1.2.25)
λ µν µν µν
2
•
Interior products
The interior product (or the contraction) of a vector field u = uµ ∂µ and an

exterior r-form φ is given by the coordinate expression
r
(−1)k−1 λk
uλk φλ1 ...λk ...λr dz λ1 ∧ · · · ∧ dz ∧ · · · ∧ dz λr =
X
ucφ = c (1.2.26)
k=1 r!
1
uµ φµα2 ...αr dz α2 ∧ · · · ∧ dz αr .
(r − 1)!
It satisfies the relations
φ(u1 , . . . , ur ) = ur c · · · u1 cφ, (1.2.27)

uc(φ ∧ σ) = ucφ ∧ σ + (−1)|φ| φ ∧ ucσ, (1.2.28)
0 0 0 0 1
[u, u ]cφ = ucd(u cφ) − u cd(ucφ) − u cucdφ, φ ∈ O (Z). (1.2.29)
The generalization of the interior product (1.2.26) for multivector fields is the
left interior product
ϑcφ = φ(ϑ), | ϑ |≤| φ |, φ ∈ O∗ (Z), ϑ ∈ T∗ (Z),
of multivector fields and exterior forms, which is derived from the equality
φ(u1 ∧ · · · ∧ ur ) = φ(u1 , . . . , ur ), φ ∈ O∗ (Z), ui ∈ T (Z),

for simple multivector fields. We have the relation

ϑcυcφ = (υ ∧ ϑ)cφ = (−1)|υ||ϑ| υcϑcφ, φ ∈ O∗ (Z), ϑ, υ ∈ T∗ (Z).
Example 1.2.5. The formula (1.2.29) can be generalized for multivector fields as
follows [13]:
[ϑ, υ]SN cφ = (−1)|υ|(|ϑ|−1) ϑcd(υcφ) + (−1)|ϑ| υcd(ϑcφ) − υcϑcdφ,
where | φ |=| ϑ | + | υ | −1. •
The right interior product

ϑbφ = ϑ(φ), | φ |≤| ϑ |, φ ∈ O∗ (Z), ϑ ∈ T∗ (Z),
of exterior forms and multivector fields is given by the equalities
ϑ(φ1 , . . . , φr ) = ϑbφr · · · bφ1 , φi ∈ O1 (Z), ϑ ∈ Tr (Z),
1
ϑbφ = ϑα1 ...αr−1 µ φµ ∂α1 ∧ · · · ∧ ∂αr−1 , φ ∈ O1 (Z).
(r − 1)!
It satisfies the relations
(ϑ ∧ υ)bφ = ϑ ∧ (υbφ) + (−1)|υ| (ϑbφ) ∧ υ, φ ∈ O1 (Z),
ϑ(φ ∧ σ) = ϑbσbφ, φ, σ ∈ O∗ (Z).
In particular, if | ϑ |=| φ |, we have the natural pairing
h, i : Tr (Z) × Or (Z) → C ∞ (Z),
hϑ, φi = ϑcφ = ϑbφ = ϑ(φ) = φ(ϑ). (1.2.30)
Bivector fields and 2-forms
Each bivector field

1
2
on a manifold Z defines the linear fibred morphism
w] : T ∗ Z → T Z,
Z
def
]
w (α) = −w(z)bα, α ∈ Tz∗ Z, (1.2.31)
] µν
w (α) = w (z)αµ ∂ν ,
which fulfills the relation
w(z)(α, β) = w] (α)cβ = w(z)bβbα, z ∈ Z, α, β ∈ Tz∗ Z.
One says that a bivector field w is of rank r at a point z ∈ Z if the morphism

(1.2.31) has rank r at z. If this morphism is an isomorphism at all the points z ∈ Z,
the bivector field w is said to be non-degenerate. Such a bivector field can exist only
on an even-dimensional manifold.
The morphism (1.2.31) can be generalized to the homomorphism of graded al-
gebras O∗ (Z) → T∗ (Z) in accordance with the relation
def
w] (φ)(σ1 , . . . , σr ) =(−1)r φ(w] (σ1 ), . . . , w] (σr )), (1.2.32)
φ ∈ Or (Z), σi ∈ O1 (Z).
This is clearly an isomorphism if the bivector field w is non-degenerate.

Each 2-form
1
Ω = Ωµν dz µ ∧ dz ν
2
on a manifold Z defines the linear fibred morphism
Ω[ : T Z → T ∗ Z,
def
Ω[ (v) = −vcΩ(z), v ∈ Tz Z, (1.2.33)
[ µ ν
Ω (v) = −Ωµν (z)v dz .
One says that a 2-form Ω is of rank r at a point z ∈ Z if the morphism (1.2.33) has
k
rank r at z. This is the maximal number 2k such that ∧ Ω(z) 6= 0.
The kernel of a 2-form Ω is defined as the kernel
def [
Ker Ω = {v ∈ Tz Z : vcucΩ = 0, ∀u ∈ Tz Z} (1.2.34)
z∈Z
of the morphism (1.2.33). Its fibre Ker z Ω at a point z ∈ Z is a vector subspace of

the tangent space Tz Z whose codimension equals the rank of Ω at z. If a 2-form Ω
is of constant rank, its kernel (1.2.34) is a subbundle of the tangent bundle T Z in
accordance with Proposition 1.1.4.
A 2-form Ω is called non-degenerate if its rank is equal to dim Z at all points
z ∈ Z. A non-degenerate 2-form Ω can exist only on a 2m-dimensional manifold.
m
Then ∧ Ω is nowhere vanishing, and can play the role of a volume element on Z.
On a 2m-dimensional manifold Z, there is one-to-one correspondence between

the non-degenerate 2-forms Ωw and the non-degenerate bivector fields wΩ in accor-
dance with the equalities
wΩ (φ, σ) = Ωw (wΩ] (φ), wΩ] (σ)), (1.2.35)

Ωw (ϑ, ν) = wΩ (Ω[w (ϑ), Ω[w (ν)), (1.2.36)
φ, σ ∈ O1 (Z), ϑ, ν ∈ T (Z),
where the morphisms wΩ] (1.2.31) and Ω[w (1.2.33) obey the relations
wΩ] = (Ω[w )−1 ,

Ωwαβ wΩαν = δβν ,
i.e.,
wΩ] (Ω[w (ϑ)) = ϑ, Ω[w (wΩ] (φ)) = φ.
The Lie derivative
The Lie derivative of an exterior form φ along a vector field u is given by the
equality
Lu φ = ucdφ + d(ucφ).
In particular, if f is a function, then
Lu f = u(f ) = ucdf.
The relation
Lu (φ ∧ σ) = Lu φ ∧ σ + φ ∧ Lu σ
is fulfilled. Given the tangent lift φe (1.2.24) of an exterior form φ, we have
Lu (φ) = u∗ φe
[67, 147]. The Lie derivative (1.2.20) of a multivector field υ along a vector field u
is
Lu υ = [u, υ]SN0 = [u, υ]SN ,

and it obeys the equality
Lu (ϑ ∧ υ) = Lu ϑ ∧ υ + ϑ ∧ Lu υ
in accordance with the relation (1.2.17).
Tangent-valued forms
Elements of the tensor product Or (Z) ⊗ T (Z) are called the tangent-valued
r-forms
r
φ : Z → ∧ T ∗ Z ⊗ T Z,
1
φ = φµλ1 ...λr dz λ1 ∧ · · · ∧ dz λr ⊗ ∂µ .
r!
There is one-to-one correspondence between the tangent-valued 1-forms φ on a ma-
nifold Z and the linear bundle endomorphisms over Z:
φb : T Z → T Z,
φb : Tz Z 3 v 7→ vcφ(z) ∈ Tz Z, (1.2.37)
and
φb∗ : T ∗ Z → T ∗ Z,
φb∗ : Tz∗ Z 3 v ∗ 7→ φ(z)cv ∗ ∈ Tz∗ Z. (1.2.38)
In particular, the canonical tangent-valued 1-form
θZ = dz λ ⊗ ∂λ
on Z corresponds to the identity morphisms (1.2.37) and (1.2.38).

Let Z = T X. There is the fibred endomorphism J of the tangent bundle T T X
of T X such that, for every vector field τ on X, we have
J ◦ τe = τV , J ◦ τV = 0,
where τe is the canonical lift (1.2.2) and τV is the vertical lift (1.2.7) onto T T X of a
vector field τ on T X. This endomorphism reads
J(∂λ ) = ∂˙λ , J(∂˙λ ) = 0. (1.2.39)

It is readily observed that J ◦ J = 0, and the rank of J equals n. The endomorphism

J (1.2.39), called an almost tangent structure [110, 189], corresponds to the tangent-
valued form
φJ = dxλ ⊗ ∂˙λ (1.2.40)
on the tangent bundle T X.
Distributions
An n-dimensional smooth distribution on a k-dimensional manifold Z is an n-

dimensional subbundle T of the tangent bundle T Z. We will say that a vector field
v on Z is subordinate to a distribution T if it is a section of T → Z. An integral
curve of a vector field, subordinate to a distribution T, is called admissible with
respect to T.
A distribution T is said to be involutive if the Lie bracket [u, u0 ] is a section of
T → Z, whenever u and u0 are sections of the distribution T → Z.
A connected submanifold N of a manifold Z is called an integral manifold of a
distribution T on Z if the tangent spaces to N belong to the fibres of this distribution
at each point of N . Unless otherwise stated, by an integral manifold we mean an
integral manifold of maximal dimension, equal to dimension of the distribution T.
An integral manifold N is called maximal if there is no other integral manifold which
contains N .
Theorem 1.2.2. [185]. Let T be a smooth involutive distribution on a manifold Z.

For any point z ∈ Z, there exists a unique maximal integral manifold of T passing
through z. 2
In view of this fact, involutive distributions are also called completely integrable
distributions.
If a distribution T is not involutive, there are no integral submanifolds of di-
mension equal to the dimension of a distribution. However, integral submanifolds
always exist, e.g., the integral curves of vector fields, subordinate to T.
We refer the reader to [68] for a detailed exposition of differential and Pfaffian
systems.
A differential system S on a manifold Z is said to be a subbundle of the sheaf
S(T Z) of vector fields on Z whose fibre Sz at each point z ∈ Z is a submodule
of the Cz∞ (Z)-module Sz (T Z) (see Example (1.1.4)). The germs of sections of a

distribution T obviously make up a differential system S(T).
The flag of a differential system S is the sequence of differential systems
S1 = S, S2 = [S, S], ··· Si = [Si−1 , S].
Here [S, S0 ]z is the Cz∞ (Z)-module generated by [v, u], v ∈ Sz , u ∈ S0z . Let S(T) be
a differential system associated with a distribution T, and let
S(T) = S1 ⊂ S2 ⊂ · · ·
be its flag. In general, Si is not associated with a distribution. If this is the case for
all i, we may define the flag of a distribution
T = T1 ⊂ T2 ⊂ · · · . (1.2.41)
A distribution is called regular if its flag (1.2.41) is well defined. The sequence
(1.2.41) stabilizes, i.e., there exists an integer r such that Tr−1 6= Tr = Tr+1 [184];
moreover Tr is involutive. In particular, if r = 1, we are dealing with the integrable
case. If Tr = T Z, the distribution T is called totally non-holonomic.
A codistribution T∗ on a manifold Z is a subbundle of the cotangent bundle.
For instance, the annihilator Ann T of an n-dimensional distribution T is a (k − n)-
dimensional codistribution.
Theorem 1.2.3. [185]. Let T be a distribution and Ann T its annihilator. Let
∧Ann T be the ideal of the exterior algebra O∗ (Z) which is generated by elements of
Ann T. A distribution T is involutive if and only if the ideal ∧Ann T is a differential
ideal, i.e., d(∧Ann T) ⊂ ∧Ann T. 2
Corollary 1.2.4. Let T be a smooth involutive r-dimensional distribution on a

k-dimensional manifold Z. Every point z ∈ Z has an open neighbourhood U 3 z
which is a domain of a coordinate chart (z 1 , . . . , z k ) such that the restrictions of the
distribution T and its annihilator Ann T to U are generated by the r vector fields
∂ ∂
1
,..., r
∂z ∂z
and the (k − r) 1-forms dz k−r+1 , . . . , dz k , respectively. It follows that integral man-
ifolds of an involutive distribution make up a foliation. 2
Example 1.2.6. Every 1-dimensional distribution on a manifold Z is integrable.

Its section is a nowhere vanishing vector field u on Z, while its integral manifolds are
the integral curves of u. By virtue of Corollary 1.2.4, there exist local coordinates
(z 1 , . . . , z k ) around each point z ∈ Z such that u is given by
∂
u= .
∂z 1
•
A Pfaffian system S ∗ is a submodule of the C ∞ (Z)-module O1 (Z). In particular,

sections of a codistribution constitute a Pfaffian system. Any Pfaffian system S ∗
defines the ideal ∧S ∗ of the exterior algebra O∗ (Z) which is generated by elements
of S ∗ .
Given a flag (1.2.41) of a regular distribution, one can introduce the coflag of
the codistribution
Ann (T) ⊃ Ann (T2 ) ⊃ · · · . (1.2.42)
The coflag (1.2.42) stabilizes. In particular, a distribution T is totally non-holonomic

if and only if its coflag (1.2.42) shrinks to zero.
Foliations
An r-dimensional (regular) foliation on a k-dimensional manifold Z is said to be

a partition of Z into connected leaves Fι with the following property. Every point
of Z has an open neighbourhood U which is a domain of a coordinate chart (z α )
such that, for every leaf Fι , the connected components Fι ∩ U are described by the
equations
z r+1 = const., ··· z k = const.
[90, 150]. Note that leaves of a foliation fail to be imbedded submanifolds, i.e.,
topological subspaces in general.
Example 1.2.7. Submersions π : Y → X and, in particular, fibre bundles are
foliations with the leaves π −1 (x), x ∈ π(Y ) ⊂ X. A foliation is called simple if it is
a fibre bundle. Any foliation is locally simple. •
Example 1.2.8. Every real function f on a manifold Z with nowhere vanishing

differential df is a submersion Z → R. It defines a 1-codimensional foliation whose
leaves are given by the equations
f (z) = c, c ∈ f (Z) ⊂ R.
This is the foliation of level surfaces of the function f , called a generating function.
Every 1-codimensional foliation is locally a foliation of level surfaces of some function
on Z. •
The level surfaces of arbitrary function f 6= const. on a manifold Z define a sin-

gular foliation F on Z [90]. Its leaves are not submanifolds in general. Nevertheless
if df (z) 6= 0, the restriction of F to some open neighbourhood U of z is a foliation
with the generating function f |U .
1.3 Jet manifolds

Subsections: Jet manifolds, 36; Canonical horizontal splittings, 38; Second order jet
manifolds, 39; The total derivative, 41; Higher order jet manifolds, 41; Differential
operators and differential equations, 41.
Jet manifolds
Given a fibre bundle Y → X with bundle coordinates (xλ , y i ), let us consider

the equivalence classes jx1 s, x ∈ X, of its sections s, which are identified by their
values si (x) and the values of their first derivatives ∂µ si (x) at points x ∈ X. The
equivalence class jx1 s is called the first order jet of sections s at the point x ∈ X.
The set J 1 Y of first order jets is provided with a manifold structure with respect to
the adapted coordinate atlas
(xλ , y i , yλi ),
(xλ , y i , yλi )(jx1 s) = (xλ , si (x), ∂λ si (x)),
i ∂xµ
y0λ = (∂µ + yµj ∂j )y 0i . (1.3.1)
∂x0 λ
It is called the jet manifold of sections of the fibre bundle Y → X (or simply the jet
manifold of the fibre bundle Y → X).
1.3. JET MANIFOLDS 37
The jet manifold admits the natural fibrations
π 1 : J 1 Y 3 jx1 s 7→ x ∈ X, (1.3.2)
π01 : J 1 Y 3 jx1 s 7→ s(x) ∈ Y, (1.3.3)
where (1.3.3) is an affine bundle modelled over the vector bundle
T ∗ X ⊗ V Y → Y.
Y
For the sake of convenience, the fibration J 1 Y → X is further called a jet bundle,
while the fibration J 1 Y → Y is an affine jet bundle.
There are the following two canonical monomorphisms of the jet manifold J 1 Y
over Y :
λ : J 1 Y ,→ T ∗ X ⊗ T Y, (1.3.4)
Y
λ = dxλ ⊗ dλ = dxλ ⊗ (∂λ + yλi ∂i ),
where dλ is called the total derivative, and
θ1 : J 1 Y ,→ T ∗ Y ⊗ V Y, (1.3.5)
Y
θ1 = θi ⊗ ∂i = (dy − yλi dxλ ) ⊗ ∂i ,
i
where
θi = dy i − yλi dxλ (1.3.6)
is called the contact form. In accordance with these monomorphisms, every element
of the jet manifold J 1 Y can be represented by the tangent-valued forms
dxλ ⊗ (∂λ + yλi ∂i ) and (dy i − yλi dxλ ) ⊗ ∂i .
Each fibred morphism Φ : Y → Y 0 over a diffeomorphism f is extended to the

fibred morphism of the corresponding jet manifolds
J 1Φ : J 1Y → J 1Y 0,
Φ
1
J Φ: jx1 s 7→ jf1(x) (Φ ◦ s ◦ f −1 ),
i ∂(f −1 )µ
y 0 λ ◦ J 1 Φ = (∂j Φi yµj + ∂µ Φi ) ,
∂x0λ
called the jet prolongation of the morphism Φ.
Each section s of a fibre bundle Y → X has the jet prolongation to the section
def
(J 1 s)(x) = jx1 s,
(y i , yλi ) ◦ J 1 s = (si (x), ∂λ si (x)),
of the jet bundle J 1 Y → X. A section s of the jet bundle J 1 Y → X is said to be

holonomic if this is the jet prolongation of some section of the fibre bundle Y → X.
Any projectable vector field
u = uλ (xµ )∂λ + ui (xµ , y j )∂i
on a fibre bundle Y → X admits the jet prolongation to the vector field
u = r1 ◦ J 1 u : J 1 Y → J 1 T Y → T J 1 Y,
u = uλ ∂λ + ui ∂i + (dλ ui − yµi ∂λ uµ )∂iλ , (1.3.7)
on the jet manifold J 1 Y . One can show that the jet prolongation of vector fields
u 7→ u is the morphism of Lie algebras, i.e.,
[u, u0 ] = [u, u0 ].
In order to obtain (1.3.7), we have used the canonical fibred morphism
r1 : J 1 T Y → T J 1 Y,
ẏλi ◦ r1 = (ẏ i )λ − yµi ẋµλ .
In particular, there is the canonical isomorphism
V J 1 Y = J 1 V Y, (1.3.8)
ẏλi i
= (ẏ )λ .
Canonical horizontal splittings
The canonical morphisms (1.3.4) and (1.3.5) can be viewed as the morphisms
b : J 1 Y × T X 3 ∂ 7→ d = ∂ cλ ∈ J 1 Y × T Y
λ (1.3.9)
λ λ λ
X Y
and
θb1 : J 1 Y × V ∗ Y 3 dy i 7→ θi = θ1 cdy i ∈ J 1 Y × T ∗ Y, (1.3.10)

Y Y
where {dy i } are the bases for the fibres of the vertical cotangent bundle V ∗ Y . These
morphisms determine the canonical horizontal splittings of the pull-backs
J 1 Y × T Y = λ(T
b X) ⊕ V Y, (1.3.11)
Y J 1Y
ẋλ ∂λ + ẏ i ∂i = ẋλ (∂λ + yλi ∂i ) + (ẏ i − ẋλ yλi )∂i ,
and
J 1 Y × T ∗ Y = T ∗ X ⊕ θb1 (V ∗ Y ), (1.3.12)
Y J 1Y
ẋλ dx + ẏi dy i = (ẋλ + ẏi yλi )dxλ + ẏi (dy i − yλi dxλ ).
λ
Second order jet manifolds
Taking the first order jet manifold of the jet bundle J 1 Y → X, we come to the
repeated jet manifold J 1 J 1 Y , provided with the adapted coordinates
i
(xλ , y i , yλi , y(µ) i
, yµλ ),
µ
i ∂x j
y 0 (λ) = 0
0i
λ (∂µ + y(µ) ∂j )y ,
∂x
0i ∂xα j j ν 0i
y µλ = 0 µ (∂α + y(α) ∂j + yνα ∂j )y λ .
∂x
There exist two different affine fibrations of J 1 J 1 Y over J 1 Y :
• the familiar affine jet bundle (1.3.3)
π11 : J 1 J 1 Y → J 1 Y, yλi ◦ π11 = yλi , (1.3.13)
modelled over the vector bundle
T ∗ X ⊗ V J 1 Y → J 1 Y, (1.3.14)
J 1Y
• and the affine bundle
J 1 π01 : J 1 J 1 Y → J 1 Y, yλi ◦ J 1 π01 = y(λ)

i
, (1.3.15)
whose underlying vector bundle
J 1 (T ∗ X ⊗ V Y ) → J 1 Y (1.3.16)
differs from (1.3.14).

In general, there is no canonical identification of these fibrations, but it can be made

by means of a symmetric linear connection on X [57].
The points q ∈ J 1 J 1 Y , where π11 (q) = J 1 π01 (q), make up the affine subbundle
Jb2 Y → J 1 Y of J 1 J 1 Y , called the sesquiholonomic jet manifold. This is given by
the coordinate conditions
i
y(λ) = yλi ,
and is coordinated by (xλ , y i , yλi , yµλ

i
).
The second order jet manifold J 2 Y of a fibre bundle Y → X is the affine sub-
bundle π12 : J 2 Y → J 1 Y of the fibre bundle Jb2 Y → J 1 Y , given by the coordinate
conditions
i i
yλµ = yµλ
and coordinated by (xλ , y i , yλi , yλµ

i i
= yµλ ). It is modelled over the vector bundle
2
∨ T ∗ X ⊗ V Y → J 1 Y.
J 1Y
The second order jet manifold J 2 Y can also be seen as the set of the equivalence
classes jx2 s of sections s of the fibre bundle Y → X, which are identified by their
values and the values of their first and second order partial derivatives at points
x ∈ X:
yλi (jx2 s) = ∂λ si (x), i

yλµ (jx2 s) = ∂λ ∂µ si (x).
Let s be a section of a fibre bundle Y → X and J 1 s its jet prolongation to a

section of the jet bundle J 1 Y → X. The latter gives rise to the section J 1 J 1 s of
the repeated jet bundle J 1 J 1 Y → X. This section takes its values into the second
order jet manifold J 2 Y . It is called the second order jet prolongation of the section
s, and is denoted by J 2 s.
Proposition 1.3.1. Let s be a section of the jet bundle J 1 Y → X and J 1 s its jet
prolongation to the section of the repeated jet bundle J 1 J 1 Y → X. The following
three facts are equivalent:
• s = J 1 s where s is a section of the fibre bundle Y → X;
• J 1 s takes its values into Jb2 Y ;

• J 1 s takes its values into J 2 Y .
The total derivative
We will use the total derivative operator
dλ = ∂λ + yλi ∂i + yλµ
i
∂iµ .
It satisfies the equalities
dλ (φ ∧ σ) = dλ (φ) ∧ σ + φ ∧ dλ (σ),
dλ (dφ) = d(dλ (φ)).
Higher order jet manifolds
The k-order jet manifold J k Y of a fibre bundle Y → X comprises the equivalence

classes jxk s, x ∈ X, of sections s of Y identified by the k + 1 terms of their Tailor
series at the points x ∈ X. The jet manifold J k Y is provided with the adapted
coordinates
(xλ , y i , yλi , . . . , yλi k ···λ1 ),

yλi l ···λ1 (jxk s) = ∂λl · · · ∂λ1 si (x), 0 ≤ l ≤ k.
Every section s of a fibre bundle Y → X gives rise to the section J k s of the fibre
bundle J k Y → X such that
yλi l ···λ1 ◦ J k s = ∂λl · · · ∂λ1 si , 0 ≤ l ≤ k.
Differential operators and differential equations
Let J k Y be the k-order jet manifold of a fibre bundle Y → X and E → X a

vector bundle over X.
Definition 1.3.2. A fibred morphism
E : J kY → E (1.3.17)
X
is called a k-order differential operator on the fibre bundle Y → X. It sends each

section s(x) of Y → X onto the section (E ◦ J k s)(x) of the vector bundle E → X.
2
The kernel of a differential operator is the subset

Ker E = E −1 (0(X))
b ⊂ J k Y, (1.3.18)
where 0b is the zero section of the vector bundle E → X, and we assume that
0(X)
b ⊂ E(J k Y ).
Definition 1.3.3. A system of k-order partial differential equations (or simply a

differential equation) on a fibre bundle Y → X is defined as a closed subbundle E
of the jet bundle J k Y → X [20, 57, 104]. 2
Its (classical) solution is a (local) section s of the fibre bundle Y → X such that
its k-order jet prolongation J k s lives in E.
For instance, if the kernel (1.3.18) of a differential operator E is a closed sub-
bundle of the fibre bundle J k Y → X, it defines a differential equation
E ◦ J k s = 0.
The following condition is sufficient for a kernel of a differential operator to be
a differential equation.
Proposition 1.3.4. Let the morphism (1.3.17) be of constant rank. By virtue of

Theorem 1.1.3, its kernel (1.3.18) is a closed subbundle of the fibre bundle J k Y → X
and, consequently, is a k-order differential equation. 2
1.4 Connections
Subsections: Connections, 42; The curvature of connections, 44; Linear connections,
44; Affine connections, 45; Flat connections, 45.
Connections
A connection on a fibre bundle Y → X is defined as a global section

Γ : Y → J 1 Y,
Γ = dxλ ⊗ (∂λ + Γiλ (xµ , y j )∂i ),
1.4. CONNECTIONS 43
of the affine jet bundle J 1 Y → Y . Combining a connection Γ and the morphisms

(1.3.9) and (1.3.10) gives the splittings
b ◦ Γ : T X → T Y,
λ
θb1 ◦ Γ : V ∗ Y → T ∗ Y
of the exact sequences (1.1.8) and (1.1.9), respectively. Accordingly, substitution of

the section yλi = Γiλ into the expressions (1.3.11) and (1.3.12) leads to the familiar
splittings of the tangent bundle
T Y = Γ(T X) ⊕ V Y, (1.4.1)
Y
ẋλ ∂λ + ẏ i ∂i = ẋλ (∂λ + Γiλ ∂i ) + (ẏ i − ẋλ Γiλ )∂i ,
and the cotangent bundle
T ∗ Y = T ∗ X ⊕ Γ(V ∗ Y ), (1.4.2)
Y
ẋλ dxλ + ẏi dy = (ẋλ + Γiλ ẏi )dxλ + ẏi (dy i − Γiλ dxλ ),
i
of a fibre bundle Y → X with respect to the connection Γ. In an equivalent way,

the connection Γ defines the corresponding projection
Γ : T Y 3 ẋλ ∂λ + ẏ i ∂i 7→ (ẏ i − ẋλ Γiλ )∂i ∈ V Y (1.4.3)
and the corresponding section
Γ = (dy i − Γiλ dxλ ) ⊗ ∂i (1.4.4)
of the fibre bundle T ∗ Y ⊗ V Y → Y .

Y
Connections on a fibre bundle Y → X constitute an affine space modelled over
the linear space of soldering forms
σ : Y → T ∗ X ⊗ V Y,
Y
σ= σλi dxλ ⊗ ∂i .
Any connection Γ on a fibre bundle Y → X defines the first order differential

operator on Y
DΓ : J 1 Y 3 z 7→ [z − Γ(π01 (z))] ∈ T ∗ X ⊗ V Y, (1.4.5)

Y
DΓ = (yλi − Γiλ )dxλ ⊗ ∂i ,
called the covariant differential. Its action on sections s of the fibre bundle Y reads
∇Γ s = DΓ ◦ J 1 s = [∂λ si − (Γ ◦ s)iλ ]dxλ ⊗ ∂i . (1.4.6)
For instance, a section s is said to be an integral section for a connection Γ, if
∇Γ s = 0, i.e., Γ ◦ s = J 1 s. For any section s of a fibre bundle Y → X, there exists
a connection Γ on Y → X such that s is its integral section. This connection is an
extension of the section s(x) 7→ J 1 s(x) of the affine jet bundle J 1 Y → Y over the
closed submanifold s(X) ⊂ Y in accordance with Theorem 1.1.2.
A connection Γ on a fibre bundle Y → X defines the horizontal lift
Γτ = τ λ (∂λ + Γiλ ∂i ) (1.4.7)
onto Y of each vector field τ = τ λ ∂λ on X.
The curvature of connections
The curvature of a connection Γ on a fibre bundle Y → X is said to be the

2-form on Y
2
R : Y → ∧ T ∗ X ⊗ V Y,
Y
1 i
R = Rλµ dxλ ∧ dxµ ⊗ ∂i ,
2
i
Rλµ = ∂λ Γiµ − ∂µ Γiλ + Γjλ ∂j Γiµ − Γjµ ∂j Γiλ . (1.4.8)
Linear connections
Let Y → X be a vector bundle. A linear connection on Y → X reads

Γ = dxλ ⊗ [∂λ + Γλ i j (x)y j ∂i ].
It defines the dual linear connection
Γ∗ = dxλ ⊗ [∂λ − Γi j λ (x)yj ∂ i ]
on the dual vector bundle Y ∗ → X. For instance, a linear connection K on the
tangent bundle T X, and the dual linear connection K ∗ on the cotangent bundle
T ∗ X are given by the expressions
K = dxλ ⊗ (∂λ + Kλ α ν (x)ẋν ∂˙α ), (1.4.9)
K ∗ = dxλ ⊗ (∂λ − Kλ µ ν (x)ẋµ ∂˙ ν ). (1.4.10)
1.4. CONNECTIONS 45
Affine connections
Let Y → X be an affine bundle modelled over a vector bundle Y → X. An

affine connection on Y → X reads
Γ = dxλ ⊗ [∂λ + (Γλ i j (x)y j + Γiλ (x))∂i ].
It defines the linear connection

∂
Γ = dxλ ⊗ [∂λ + Γλ i j (x)y j ]
∂y i
on the vector bundle Y → X.
Flat connections
Each connection Γ on a fibre bundle Y → X, by definition, yields the horizontal

distribution Γ(T X) ⊂ T Y on Y , generated by the horizontal vector fields (1.4.7).
The following assertions are equivalent.
• The horizontal distribution is involutive.
• The connection Γ is flat (curvature-free), i.e., its curvature is equal to zero

everywhere.
• There is an integral section for the connection Γ through any point y ∈ Y .
Hence, a flat connection Γ on Y → X yields the integrable horizontal distribution,

i.e., the horizontal foliation on Y , transversal to the fibration Y → X. Its leaf
through a point y ∈ Y is defined locally by an integral section sy for the connection
Γ through y. Conversely, let a fibre bundle Y → X admit a horizontal foliation such
that, for each point y ∈ Y , the leaf of this foliation through y is locally defined by
a section sy of Y → X through y. Then the map
Γ : Y → J 1 Y,
Γ(y) = jx1 sy , π(y) = x,
introduces a flat connection on Y → X. Thus, there is one-to-one correspondence

between the flat connections and the horizontal foliations on a fibre bundle Y → X.
Given a horizontal foliation on a fibre bundle Y → X, there exists the associated
atlas of bundle coordinates (xλ , y i ) of Y such that every leaf of this foliation is locally
generated by the equations y i = const., and the transition functions y i → y 0 i (y j ) are

independent of the base coordinates xλ [23, 57]. This is called the atlas of constant
local trivializations. Two such atlases are said to be equivalent if their union is
also an atlas of constant local trivializations. They are associated with the same
horizontal foliation. Thus, we come to the following assertion.
Proposition 1.4.1. There is one-to-one correspondence between the flat connec-

tions Γ on a fibre bundle Y → X and the equivalence classes of atlases of constant
local trivializations of Y such that Γiλ = 0 relative to these atlases. 2
1.5 Bundles with symmetries

Subsections: Tangent and cotangent bundles of Lie groups, 46; Principal bundles,
48; The linear frame bundle, 52.
Tangent and cotangent bundles of Lie groups
Let G be a real Lie group with dim G > 0 and gl [gr ] its left [right] Lie algebra
of left-invariant vector fields ξl (g) = T Lg (ξl (e)) [right-invariant vector fields ξr (g) =
T Rg (ξr (e))] on the group G. Here, e is !
the unit element of G, while Lg and Rg denote the action of G on itself on
the left and on the right, respectively. Every left-invariant vector field ξl (g) [right-
invariant vector field ξr (g)] corresponds to the element v = ξl (e) [v = ξr (e)] of the
tangent space Te G provided with both left and right Lie algebra structures. For
instance, given v ∈ Te G, let vl (g) and vr (g) be the corresponding left-invariant and
right-invariant vector fields. There is the relation
vl (g) = T Lg ◦ T Rg−1 (vr (g)).
Let {m = m (e)} [{εm = εm (e)}] denote the basis for the left [right] Lie algebra,
and let ckmn be the right structure constants:
[εm , εn ] = ckmn εk .
The mapping g 7→ g −1 yields the isomorphism
ρ : gl 3 m 7→ εm = −m ∈ gr (1.5.1)
1.5. BUNDLES WITH SYMMETRIES 47
of left and right Lie algebras. For instance, we have
[m , n ] = −ckmn k .
The tangent bundle πG : T G → G of the Lie group G is trivial. There are the
isomorphisms
%l : T G 3 q 7→ (g = πG (q), T L−1
g (q)) ∈ G × gl ,
%r : T G 3 q 7→ (g = πG (q), T Rg−1 (q)) ∈ G × gr .
The left action Lg of a Lie group G on itself defines its adjoint representation g 7→
Adg in the right Lie algebra gr and its identity representation in the left Lie algebra
gl . Correspondingly, there is the adjoint representation
ε0 : ε 7→ ad ε0 (ε) = [ε0 , ε],

ad εm (εn ) = ckmn εk ,
of the right Lie algebra gr in itself.

An action
G × Z 3 (g, z) 7→ gz ∈ Z
of a Lie group G on a manifold Z on the left yields the homomorphism
gr 3 ε → ξε ∈ T (Z)
of the right Lie algebra gr of G into the Lie algebra of vector fields on Z such that
ξAd g(ε) = T g ◦ ξε ◦ g −1 (1.5.2)
[100]. Vector fields ξεm are said to be the generators of a representation of the Lie
group G in Z.
Let g∗ = Te∗ G be the vector space dual of the tangent space Te G. It is called the
dual Lie algebra (or the Lie coalgebra), and is provided with the basis {εm } dual of
the basis {εm } for Te G. The group G and the right Lie algebra gr act on g∗ by the
coadjoint representation
def
hAd∗ g(ε∗ ), εi =hε∗ , Ad g −1 (ε)i, ε∗ ∈ g∗ , ε ∈ gr , (1.5.3)
∗ 0 ∗ ∗ 0 0
had ε (ε ), εi = −hε , [ε , ε]i, ε ∈ gr ,
∗ n
ad εm (ε ) = −cnmk εk .
Remark 1.5.1. In the literature (see, e.g.,[2]), one can meet another definition of
the coadjoint representation in accordance with the relation
hAd∗ g(ε∗ ), εi = hε∗ , Ad g(ε)i.
•
An exterior form φ on the group G is said to be left-invariant [right-invariant] if
φ(e) = L∗g (φ(g)) [φ(e) = Rg∗ (φ(g))]. The exterior differential of a left-invariant [right-
invariant] form is left-invariant [right-invariant]. In particular, the left-invariant
1-forms satisfy the Maurer–Cartan equations
1
dφ(, 0 ) = − φ([, 0 ]), , 0 ∈ gl .
2
There is the canonical gl -valued left-invariant 1-form
θl : Te G 3 7→ ∈ gl
on a Lie group G. The components θlm of its decomposition θl = θlm m with respect
to the basis for the left Lie algebra gl make up the basis for the space of left-invariant
exterior 1-forms on G:
m cθln = δm
n
.
The Maurer–Cartan equation, written with respect to this basis, reads

1
dθlm = cm θn ∧ θlk .
2 nk l
Accordingly, the canonical gr -valued right-invariant 1-form
θr : Te G 3 ε 7→ ε ∈ gr
on the group G is defined. There are the relations
θl (vg ) = θl (T L−1 −1
g (vg )) = T Lg (vg ), vg ∈ Tg G,
θr (vg ) = θr (T Rg−1 (vg ))
= T Rg−1 (vg ),
ρ(θl (vg )) = −T Lg ◦ T Rg−1 θr (vg ) = −Adg(θr (vg )),
where ρ is the isomorphism (1.5.1).
Principal bundles
We refer the reader to [100, 170, 192] for the general theory of principal bundles.
Let πP : P → X be a principal bundle with a real structure Lie group G. There
is the canonical free transitive action
RG : P × G → P, (1.5.4)
X
Rg : p 7→ pg, p ∈ P, g ∈ G,
of G on P on the right.
A principal bundle P is equipped with a bundle atlas ΨP = {(Uα , ψαP )} whose
trivialization morphisms
ψαP : πP−1 (Uα ) → Uα × G
obey the condition
pr2 ◦ ψαP ◦ Rg = g ◦ pr2 ◦ ψαP , ∀g ∈ G.
Due to this property, every trivialization morphism ψαP uniquely determines a local
section zα : Uα → P such that
pr2 ◦ ψαP ◦ zα = e.
The transformation rules for zα read
zβ (x) = zα (x)ραβ (x), x ∈ Uα ∩ Uβ , (1.5.5)
where ραβ are the transition functions of the atlas ΨP . Conversely, the family
{(Uα , zα )} of local sections of P , which obey (1.5.5), uniquely determines a bundle
atlas ΨP of P .
A principal bundle P → X admits the canonical trivial vertical splitting
α :VP ∼
= P × gl
such that α−1 (m ) are fundamental vector fields on P corresponding to the basis
elements m for the left Lie algebra gl .
Taking the quotient of the tangent bundle T P → P and the vertical tangent
bundle V P of P by the tangent map T Rg , we obtain the vector bundles
TG P = T P/G and VG P = V P/G (1.5.6)

over X. Sections of TG P → X are G-invariant vector fields on P , while sections of

VG P → X are G-invariant vertical vector fields on P . Hence, the typical fibre of
VG P → X is the right Lie algebra gr of the right-invariant vector fields on the group
G. The group G acts on this typical fibre by the adjoint representation.
The Lie bracket of vector fields on P goes to the quotients (1.5.6) and defines the
Lie bracket of sections of the vector bundles TG P → X and VG P → X. It follows
that VG P → X is a fibre bundle of Lie algebras (the gauge algebra bundle in the
terminology of gauge theories) whose fibres are isomorphic to the right Lie algebra
gr of the group G.
Example 1.5.2. When P = X × G is trivial, we have
VG P = X × T G/G ∼
= X × gr .
Example 1.5.3. Given a local bundle splitting of P , there are the corresponding
local bundle splitting of TG P and VG P . Given the basis {εp } for the Lie algebra gr ,
we obtain the local fibre bases {∂λ , εp } for the fibre bundle TG P → X and {εp } for
the fibre bundle VG P → X. If
ξ, η : X → TG P,
ξ = ξ λ ∂λ + ξ p εp , η = η µ ∂µ + η q εq ,
are sections, the coordinate expression of their bracket is
[ξ, η] = (ξ µ ∂µ η λ − η µ ∂µ ξ λ )∂λ + (ξ λ ∂λ η r − η λ ∂λ ξ r + crpq ξ p η q )εr .
•
Let J 1 P be the first order jet manifold of a principal bundle P → X with a
structure Lie group G. Bearing in mind that the jet bundle J 1 P → P is an affine
bundle modelled over the vector bundle
T ∗ X ⊗ V P → P,
P
let us consider the quotient of the jet bundle J 1 P → P by the jet prolongation J 1 Rg
of the canonical action (1.5.4). We obtain the affine bundle
C = J 1 P/G → X (1.5.7)
modelled over the vector bundle

C = T ∗ X ⊗ VG P → X.
Hence, there is the canonical vertical splitting
VC ∼
= C × C.
X
In the case of a principal bundle P → X, the exact sequence (1.1.8) reduces to

the exact sequence
0 → VG P ,→ TG P → T X → 0. (1.5.8)
X
A principal connection A on a principal bundle P → X is defined as a section

A : P → J 1 P which is equivariant under the action (1.5.4) of the group G on P ,
i.e.,
J 1 Rg ◦ A = A ◦ Rg , ∀g ∈ G. (1.5.9)
Turning now to the quotients (1.5.6), such a connection defines the splitting of the
exact sequence (1.5.8). It is represented by the tangent-valued form
A = dxλ ⊗ (∂λ + Aqλ εq ), (1.5.10)
where Apλ are local functions on X.
On the other hand, due to the property (1.5.9), there is obviously one-to-one
correspondence between the principal connection on a principal bundle P → X and
the global sections of the fibre bundle C → X (1.5.7), called the bundle of principal
connections.
Let a principal connection on the principal bundle P → X be represented by
the vertical-valued form A (1.4.4). Then the form
A Id ⊗α
A : P −→ T ∗ P ⊗ V P −→ T ∗ P ⊗ gl
P
is the familiar gl -valued connection form on the principal bundle P . Given a local
bundle splitting (Uξ , zξ ) of P , this form reads
q
A = θP − Aλ dxλ ⊗ q ,
where θP is the canonical gl -valued 1-form on P , {p } is the basis of gl , and Apλ are
local functions on P such that
q q
Aλ (pg)q = Aλ (p)Adg −1 (q ).
The pull-back zξ∗ A of A over Uξ is the well-known local connection 1-form
Aξ = −Aqλ dxλ ⊗ q , (1.5.11)

q
where Aqλ = Aλ ◦ zξ are local functions on X.
It is readily observed that the coefficients Aqλ of this form are precisely the co-
efficients of the form (1.5.10). Moreover, given a bundle atlas of P , the bundle of
principal connections C is equipped with the associated bundle coordinates (xλ , aqλ )
such that, for any section A of C → X, the local functions
Aqλ = aqλ ◦ A
are again the coefficients of the local connection 1-form (1.5.11). In gauge theory,
these coefficients are treated as gauge potentials. We will use this term to refer to
sections A of the fibre bundle C → X.
Let now
Y = (P × V )/G (1.5.12)
be a fibre bundle associated with the principal bundle P → X whose structure

group G acts on the typical fibre V of Y on the left. Let us recall that the quotient
in (1.5.12) is defined by identification of the elements (p, v) and (pg, g −1 v) for all
g ∈ G. Briefly, we will say that (1.5.12) is a P -associated fibre bundle.
As is well known, the principal connection A (1.5.10) induces the corresponding
connection on the P -associated fibre bundle (1.5.12). If Y is a vector bundle, this
connection takes the form
A = dxλ ⊗ (∂λ + Apλ Ipi ∂i ),
where Ipi are generators of the representation of the right Lie algebra gr on V . This
is called the associated principal connection or simply a principal connection on
Y → X.
The linear frame bundle
Let X be an n-dimensional connected oriented manifold. Let
πLX : LX → X
be the principal bundle of oriented linear frames {sa } in the tangent spaces to X
(or simply the frame bundle). Its structure group is GL+ (n, R). The frame bundle
1.6. COMPOSITE FIBRE BUNDLES 53
is associated with the tangent bundle T X and with the cotangent bundle T ∗ X of
X. Given frames {∂µ } in the tangent bundle T X, every element {sa } of the frame
bundle LX takes the form
sa = sµ a ∂µ ,
where sµ a are matrix elements of the group GL+ (n, R). These matrix elements
constitute the bundle coordinates
∂x0µ λ
(xλ , sµ a ), s0µ a = s a,
∂xλ
on the frame bundle LX. With respect to these coordinates, the canonical action
of GL+ (n, R) on LX on the right reads
Rg : sµ a 7→ sµ b g b a , g ∈ GL+ (n, R).
The frame bundle LX is equipped with the canonical Rn -valued 1-form
θLX = sa µ dxµ ⊗ ta , (1.5.13)
where {ta } is a fixed basis for Rn , while sa µ are elements of the inverse matrix of
sµ a .
The frame bundle, like tensor bundles, admits the canonical lift of any diffeo-
morphism f of its base X to the automorphism
fe : (xλ , sλ a ) 7→ (f λ (x), ∂µ f λ sµ a ).
These automorphisms and the corresponding automorphisms of associated bundles

are called general covariant transformations or holonomic automorphisms. For in-
stance, in the case of the tangent bundle T X, the holonomic automorphisms fe = T f
are the tangent maps to the diffeomorphisms f . Generators of general covariant
transformations of tensor bundles are the canonical lifts τe (1.2.1) of vector fields τ
on X.
1.6 Composite fibre bundles

Let us consider the composition of fibre bundles
Y → Σ → X, (1.6.1)
where
πY Σ : Y → Σ (1.6.2)
and
πΣX : Σ → X (1.6.3)
are fibre bundles. This is called a composite fibre bundle. The composite fibre bundle
(1.6.1) is provided with an atlas of fibred coordinates (xλ , σ m , y i ), where (xµ , σ m )
are fibred coordinates on the fibre bundle (1.6.3) and the transition functions σ m →
σ 0m (xλ , σ k ) are independent of the coordinates y i .
The following two assertions on composite fibre bundles are useful in applications
to field theory and mechanics [57, 159].
Proposition 1.6.1. Given a section h of the fibre bundle Σ → X and a section sΣ

of the fibre bundle Y → Σ, their composition
s = sΣ ◦ h (1.6.4)
is a section of the composite fibre bundle Y → X (1.6.1). Conversely, every section

s of the fibre bundle Y → X is the composition (1.6.4) of the section h = πY Σ ◦ s
of the fibre bundle Σ → X and some section sΣ of the fibre bundle Y → Σ over the
submanifold h(X) ⊂ Σ. 2
Proposition 1.6.2. Given a composite fibre bundle (1.6.1), let h be a global

section of the fibre bundle Σ → X. Then the restriction
Yh = h∗ Y (1.6.5)
of the fibre bundle Y → Σ to h(X) ⊂ Σ is a subbundle
ih : Yh ,→ Y
of the fibre bundle Y → X. 2
Let us consider the jet manifolds J 1 Σ, JΣ1 Y , and J 1 Y of the fibre bundles Σ → X,
Y → Σ and Y → X, respectively. They are parameterized by the coordinates
(xλ , σ m , σλm ), (xλ , σ m , y i , yeλi , ym

i
), (xλ , σ m , y i , σλm , yλi ),
respectively. There is the canonical map

ρ : J 1 Σ × JΣ1 Y −→ J 1 Y, (1.6.6)
Σ Y
i m
yλi ◦ ρ = ym σλ + yeλi .
In particular, let
A = dxλ ⊗ (∂λ + Aiλ ∂i ) + dσ m ⊗ (∂m + Aim ∂i ) (1.6.7)
be a connection on the fibre bundle Y → Σ, and
Γ = dxλ ⊗ (∂λ + Γm
λ ∂m )
a connection on the fibre bundle Σ → X. Then the connection

B = ρ ◦ (A ◦ π01 × Γ ◦ π 1 ) = dxλ ⊗ [∂λ + Γm i m i
λ ∂m + (Am Γλ + Aλ )∂i ] (1.6.8)
on the composite fibre bundle Y → X is defined. This is called the composite
connection. In brief, we will write
B = A ◦ Γ.
For instance, let us consider a vector field τ on the base X, its horizontal lift Γτ
onto Σ by means of the connection Γ and, in turn, the horizontal lift A(Γτ ) of Γτ
onto Y by means of the connection A. Then A(Γτ ) coincides with the horizontal
lift Bτ of τ onto Y by means of the composite connection (1.6.8).
Given a composite fibre bundle Y (1.6.1), there are the following exact sequences
of vector bundles over Y :
0 → VΣ Y ,→ V Y → Y × V Σ → 0, (1.6.9)
Σ
0 → Y × V ∗ Σ ,→ V ∗ Y → VΣ∗ Y → 0, (1.6.10)
Σ
where VΣ Y and VΣ∗ Y are vertical tangent and cotangent bundles of the fibre bundle
Y → Σ, respectively. Every connection A (1.6.7) on the fibre bundle Y → Σ
determines the splittings
V Y = VΣ Y ⊕ A(Y × V Σ), (1.6.11)
Y Σ
ẏ i ∂i + σ̇ m ∂m = (ẏ i − Aim σ̇ m )∂i + σ̇ m (∂m + Aim ∂i ),
V ∗ Y = (Y × V ∗ Σ) ⊕ A(VΣ∗ Y ), (1.6.12)
Σ Y
ẏi dy i + σ̇m dσ m = ẏi (dy i − Aim dσ m ) + (σ̇m + Aim ẏi )dσ m ,
of the exact sequences (1.6.9) and (1.6.10), respectively. Using the splitting (1.6.11),
one can construct the first order differential operator
f : J 1 Y → T ∗ X ⊗ V Y,
D Σ
Y
f = dxλ ⊗ (y i − Ai − Ai σ m )∂ ,
D (1.6.13)
λ λ m λ i
called the vertical covariant differential, on the composite fibre bundle Y → X. This
operator can also be seen as the composition
f = pr ◦ D : J 1 Y → T ∗ X ⊗ V Y → T ∗ X ⊗ V Y ,
D 1 B Σ
Y Y
where DB is the covariant differential (1.4.5) relative to some composite connection

(1.6.8), but D
f does not depend on Γ.
The vertical covariant differential (1.6.13) possesses the following important

property. Let h be a section of the fibre bundle Σ → X and Yh the subbundle
(1.6.5) of the composite fibre bundle Y → X, which is the restriction of the fibre
bundle Y → Σ to h(X). Then the restriction of the vertical covariant differential
f (1.6.13) to J 1 i (J 1 Y ) ⊂ J 1 Y coincides with the familiar covariant differential on
D h h
Yh relative to the connection
Ah = dxλ ⊗ [∂λ + (Aim ∂λ hm + (A ◦ h)iλ )∂i ]
on Yh , which is the restriction of the connection A to h(X) in accordance with the

commutative diagram
J 1i
J 1 Yh −→
h
J 1Y
Ah 6 6 AΣ
Yh −→ Y
ih
Example 1.6.1. Let Γ : Y → J 1 Y be a connection on a fibre bundle Y → X.

In accordance with the canonical isomorphism V J 1 Y ∼
= J 1 V Y (1.3.8), the vertical
tangent map V Γ : V Y → V J 1 Y to Γ defines the connection
V Γ : V Y → J 1 V Y,
V Γ = dxλ ⊗ (∂λ + Γiλ ∂i + ∂j Γiλ ẏ j ∂˙i ), (1.6.14)
on the composite vertical tangent bundle V Y → X. The dual connection on the

composite vertical cotangent bundle V ∗ Y → X reads
V ∗ Γ : V ∗ Y → J 1 V ∗ Y,
V ∗ Γ = dxλ ⊗ (∂λ + Γiλ ∂i − ∂j Γiλ ẏi ∂˙ j ). (1.6.15)
If Y → X is an affine bundle, the connection V Γ (1.6.14) can be seen as the

composite connection (1.6.8) generated by the connection Γ on Y → X and the
linear connection
e = dxλ ⊗ (∂ + ∂ Γi ẏ j ∂˙ ) + dy i ⊗ ∂
Γ (1.6.16)
λ j λ i i
on the vertical tangent bundle V Y → Y . •

Chapter 2
Geometry of Poisson manifolds
This Chapter is devoted to the basic geometric structures on a manifold, which we

meet in mechanics. We start from a Jacobi structure whose particular case is a
Poisson structure.
Throughout this Chapter, by Z, unless otherwise stated, is meant a k-dimensional
manifold with coordinates (z λ ).
For the sake of convenience, the Schouten–Nijenhuis bracket [., .]SN is denoted
simply by [., .].
2.1 Jacobi structure

A Jacobi bracket (or a Jacobi structure) on a manifold Z is defined as a bilinear
map
O0 (Z) × O0 (Z) 3 (f, g) → {f, g} ∈ O0 (Z)
on the vector space O0 (Z) of real functions on Z. This map, by definition, satisfies
the following conditions:
(A1) {g, f } = −{f, g} (skew-symmetry),
(A2) {f, {g, h}} + {g, {h, f }} + {h, {f, g}} = 0 (Jacobi identity),
(A3) the support of {f, g} is contained in the intersection of the supports of f
and g.
A manifold Z endowed with a Jacobi structure is called a Jacobi manifold.
A Jacobi bracket provides the space O0 (Z) with a structure of a Lie algebra
because it is expressed by a bidifferential operator of not more than first order in
59
60 CHAPTER 2. GEOMETRY OF POISSON MANIFOLDS
each of its arguments [71, 98].
Proposition 2.1.1. [98, 118, 123]. Every Jacobi bracket on a manifold Z is

uniquely defined in accordance with the relation
{f, g} = w(df, dg) + uc(f dg − gdf ) (2.1.1)
by a pair (w, u) of a bivector field w and a vector field u on Z such that
[u, w] = 0, [w, w] = 2u ∧ w. (2.1.2)
2
Example 2.1.1. Taking w = 0 in (2.1.1), we find that every vector field u on a

manifold Z provides the Jacobi bracket
{f, g} = uc(f dg − gdf ).
The relations (2.1.2) are obviously satisfied. •
Example 2.1.2. The Jacobi bracket (2.1.1) with u = 0 is said to be a Poisson

bracket. According to Proposition 2.1.1, a bivector field w on a manifold Z yields a
Poisson bracket if it meets the condition
[w, w] = wµλ1 ∂µ wλ2 λ3 ∂λ1 ∧ ∂λ2 ∧ ∂λ3 = 0.
It is called a Poisson bivector field. •
Let us consider the following examples of Jacobi manifolds which are not Poisson
ones.
Example 2.1.3. Odd-dimensional contact manifolds are Jacobi manifolds in ac-
cordance with Proposition 2.2.6. •
Example 2.1.4. Let Ω be a non-degenerate 2-form and φ a closed 1-form on an

even-dimensional manifold Z such that
dΩ = φ ∧ Ω.
The triple (Z, Ω, φ) is called a locally conformally symplectic manifold. This is a
Jacobi manifold characterized by the bivector field wΩ (1.2.35) and the vector field
u = (Ω[ )−1 (φ) [98, 118, 123]. If φ = 0, we have a symplectic manifold. •
2.1. JACOBI STRUCTURE 61
Let Z1 and Z2 be Jacobi manifolds. A manifold map % : Z1 → Z2 is said to be a

Jacobi morphism if, for every pair (f, g) of functions on Z2 , we have
{f ◦ %, g ◦ %}1 = {f, g}2 ◦ %.
In particular, a vector field v on a Jacobi manifold (Z; w, u) is the generator of a
local 1-parameter group of Jacobi morphisms (or an infinitesimal Jacobi morphism)
if and only if
Lv w = 0, Lv u = 0.
Definition 2.1.2. Given a function f ∈ O0 (Z) on a Jacobi manifold (Z; w, u), the
vector field
def
ϑf = w] (df ) + f u
is called the Hamiltonian vector field for f . 2
We have the relation

[ϑf , ϑg ] = ϑ{f,g} . (2.1.3)
It follows that the map f 7→ ϑf is the Lie algebra homomorphism.
The values of all Hamiltonian vector fields at all points of Z constitute the char-
acteristic distribution T on the Jacobi manifold (Z; w, u). A glance at the formula
(2.1.3) shows that this distribution is involutive, but it has different dimensions at
different points z ∈ Z in general. Therefore, this distribution is not a subbundle of
the tangent bundle T Z and, consequently, is not a distribution as that in Section
1.2. A Jacobi manifold Z is said to be transitive if its characteristic distribution
coincides with the tangent bundle T Z. Transitive Jacobi manifolds are proved to
be either locally conformally symplectic manifolds, if dim Z is even, or the contact
ones, if dim Z is odd [98, 118].
Theorem 2.1.3. [44, 98]. The characteristic distribution on a Jacobi manifold is

completely integrable in the sense of Sussmann [173] and defines on Z a singular
Stefan foliation [171]. Each leaf of this foliation is endowed with a unique Jacobi
structure such that its canonical injection into Z is a Jacobi morphism. 2
The following definition generalizes for Jacobi manifolds the notion of coisotropic
and Lagrangian submanifolds of a Poisson manifold [83] (cf. Definition 2.3.5).
Definition 2.1.4. Let (Z; w, u) be a Jacobi manifold, with the characteristic

distribution T, and N a submanifold of Z. The submanifold N is said to be:
• coisotropic if
w] (Ann Tz N ) ⊆ Tz N, z ∈ N,
where Ann Tz N ⊂ Tz∗ Z is the annihilator of Tz N , and
• Lagrangian if
w] (Ann Tz N ) = Tz N ∩ Tz , z ∈ N.
Different Jacobi structures can lead to the same characteristic distribution as

follows. Let (Z; w, u) be a Jacobi manifold and a a nowhere vanishing function on
Z. Let us consider the bivector field wa and the vector field ua on Z given by
wa = aw, ua = w] (da) + au. (2.1.4)
Then the pair (wa , ua ) (2.1.4) is a Jacobi structure on Z. It is called conformally

equivalent to the Jacobi structure (w, u) because of the relation
1
{f, g}a = {af, ag}.
a
It is readily observed that the Hamiltonian vector field for a function f with respect
to the Jacobi structure (w, u) is also the Hamiltonian vector field for the function
1
a
f with respect to the conformally equivalent Jacobi structure (2.1.4). It follows
that conformally equivalent Jacobi structures on a manifold Z define the same char-
acteristic distribution on Z.
A Jacobi morphism from a Jacobi manifold to a conformally equivalent Jacobi
manifold is said to be a conformal Jacobi morphism. Such a morphism preserves
the characteristic distribution on a Jacobi manifold.
In particular, a vector field v on a Jacobi manifold (Z; w, u) is an infinitesimal
conformal Jacobi isomorphism if and only if there exists a function b on Z such that
Lv w = bw, Lv u = w] (b) + bu.

2.2. CONTACT STRUCTURE 63
2.2 Contact structure

We will consider contact structures on odd-dimensional manifolds. By a contact
structure is meant a strictly contact structure in the sense of [116].
Definition 2.2.1. An exterior 1-form θ on a (2m + 1)-dimensional manifold Z is

said to be a contact form if
θ ∧ (dθ)m 6= 0 (2.2.1)
everywhere on Z. 2
If θ is a contact form, so is aθ where a is a nowhere vanishing function on Z.

The pair (Z, θ) (or equivalently the pair (Z, aθ)) is termed a contact manifold (we
refer the reader to [15, 116] for the details of terminology).
If a manifold admits a contact structure, this manifold is orientable, and has the
volume element (2.2.1). It follows from (2.2.1) that the exterior differential dθ of a
contact form θ is a presymplectic form of rank 2m.
There is the following variant of well-known Darboux’s theorem [116].
Theorem 2.2.2. Every point z of a (2m + 1)-dimensional contact manifold (Z, θ)

has an open neighbourhood U which is the domain of a coordinate chart (z 0 , . . . , z 2m )
such that the contact form θ on U has the local expression
m
θ = dz 0 − z m+i dz i .
X
(2.2.2)
i=1
These coordinates are called the Darboux coordinates. 2
A contact form θ on a manifold Z defines the isomorphism

def
[(u) = ucdθ + (ucθ)θ, u ∈ T (Z), (2.2.3)
of the C ∞ (Z)-module T (Z) of vector fields on a manifold Z to the C ∞ (Z)-module

O1 (Z) of exterior 1-forms on Z [83]. This isomorphism can be extended to a mapping
from the exterior algebra T∗ (Z) of multivector fields to the exterior algebra O∗ (Z)
of exterior forms by putting
def
[(u1 ∧ · · · ∧ ur ) = [(u1 ) ∧ · · · ∧ [(ur ), ui ∈ T (Z). (2.2.4)
Proposition 2.2.3. Let θ be a contact form on a manifold Z. There exists the

unique nowhere vanishing vector field
E = [−1 (θ) (2.2.5)
on Z, called the Reeb vector field, such that
Ecθ = 1, Ecdθ = 0
[116]. It is readily observed that, relative to the local Darboux coordinates, this
vector field reads E = ∂0 . 2
Example 2.2.1. Let M be a manifold with coordinates (q i ) and Z the odd-

dimensional manifold R × T ∗ M , equipped with the coordinates (t, q i , pi = q̇i ). The
manifold R × T ∗ M is well known to admit the contact form
θ = pi dq i − dt. (2.2.6)
For the sake of convenience, this contact form is chosen to differ in minus sign from
the Darboux one (2.2.2). •
Example 2.2.2. To generalize Example 2.2.1, let Q → R be a fibre bundle, with

fibred coordinates (t, q i ), and Z = V ∗ Q the vertical tangent bundle of Q, equipped
with the coordinates (t, q i , pi = q̇i ). As mentioned above, V ∗ Q is a phase space of
time-dependent mechanics. Given a connection
Γ = dt ⊗ (∂t + Γi ∂i )
on the fibre bundle Q → R, the manifold V ∗ Q is provided with the contact form
θΓ = pi (dq i − Γi dt) − dt
(see Proposition 5.2.11 below). The corresponding Reeb vector field reads
E = −(∂t + Γi ∂i − pj ∂i Γj ∂ i ).
Since a connection Γ on a fibre bundle Q → R is flat, it defines the corresponding

atlas of local constant trivializations such that, with respect to this atlas, Γ = dt⊗∂t
and the contact form θΓ is brought into the form (2.2.6). •
Given a (2m + 1)-dimensional contact manifold (Z, θ), the tangent bundle T Z
of Z admits the splitting
T Z = Ker dθ ⊕ Ker θ,
where Ker dθ is the 1-dimensional vector subbundle generated by the Reeb vector
field E [116]. In particular, every vector field u on Z is decomposed in a unique
fashion into
u = (ucθ)E + (u − (ucθ)E).
By duality, the cotangent bundle T ∗ Z of the contact manifold Z is found to have

the splitting
T ∗ Z = Θ ⊕ Ker E,
where Θ is the 1-dimensional vector subbundle generated by the contact form θ. As

a consequence, every 1-form φ on the contact manifold Z is decomposed into
φ = (Ecφ)θ + (φ − (Ecφ)θ).
There is the corresponding splitting of the isomorphism (2.2.3). This sends Ker dθ
onto Θ and Ker θ onto Ker E. The restriction of [ to Ker θ coincides with the
morphism (−dθ)[ (1.2.33) defined by the presymplectic 2-form dθ.
The 2m-dimensional subbundle Ker θ of the tangent bundle T Z of a contact
manifold Z is called the contact distribution on Z. Any contact form aθ, where a
is a nowhere vanishing function on Z, defines the same contact distribution on Z.
As is well known, there exist integral submanifolds of the contact distribution Ker θ
of dimension m, but none of higher dimension [15]. It is readily observed that the
contact forms θ and aθ have the same contact distribution.
Definition 2.2.4. A submanifold N of a (2m + 1)-dimensional contact manifold Z

is said to be a Legendre submanifold if it is an m-dimensional integral submanifold
of the contact distribution on Z. 2
Let (Z1 , θ1 ) and (Z2 , θ2 ) be contact manifolds. A manifold map % : Z1 → Z2 is

said to be a contact transformation if
%∗ θ2 = aθ1
where a is a nowhere vanishing function on Z [116]. A contact transformation

preserves the contact distribution, but not a contact form in general. We consider
especially the contact automorphisms of a contact manifold (Z, θ), which keeps the
contact form θ.
By definition, the Reeb vector field E (2.2.5) is the generator of a local 1-
parameter group of local contact automorphisms (or an infinitesimal contact au-
tomorphism) of the contact manifold (Z, θ), i.e.,
LE θ = 0.
It is readily observed that, if a vector field u on Z is an infinitesimal contact auto-

morphism of the contact manifold (Z, θ), then u is a Hamiltonian vector field with
respect to the presymplectic form dθ on Z, i.e., the exterior form ucdθ is exact (see
Definition 2.5.2 below). The converse assertion is the following.
Proposition 2.2.5. Given a contact manifold (Z, θ), let ϑf be a Hamiltonian

vector field on Z for a function f with respect to the presymplectic form dθ, i.e.,
ϑf cdθ = −df.
Then the vector field
ϑf = [f − (ϑf cθ)]E + ϑf (2.2.7)
is an infinitesimal contact automorphism of the contact manifold Z, i.e.,
Lϑf θ = 0.
Proof. It follows from direct computation. QED
Note that the vector field (2.2.7), like ϑf , is a Hamiltonian vector field for the
function f with respect to the presymplectic form dθ.
A contact manifold is a Jacobi one as follows.
Proposition 2.2.6. [123]. Each contact form θ on a (2m+1)-dimensional manifold

Z yields the associated Jacobi bracket (2.1.2) on Z which is defined by the Reeb
vector field E (2.2.5) and the bivector field w such that
w] (φ)cθ = 0, w] (φ)cdθ = −(φ − (Ecφ)θ) (2.2.8)

for any 1-form φ on Z. 2
Relative to the local Darboux coordinates for the contact form θ, the above-
mentioned Jacobi bracket (2.1.2) reads
X ∂ m+i ∂ ∂ ∂
w= ( + z ) ∧ , u=E= ,
i ∂z i ∂z 0 ∂z m+i ∂z 0
m
X
{f, g} = (∂m+i g∂i f − ∂m+i f ∂i g) + ([g]∂0 f − [f ]∂0 g),
i=1
where
m
z m+i ∂m+i f − f,
X
[f ] =
i=1
m
z m+i ∂m+i g − g.
X
[g] =
i=1
In particular, let (Z, θ) be a contact manifold and (w, u) the associated Jacobi
structure on Z. A submanifold N of Z is a Legendre submanifold of the contact
manifold (Z, θ) if and only if it is the Lagrangian submanifold of the Jacobi manifold
(Z; w, u) [83]. Note that, in this case, the characteristic distribution of the Jacobi
structure coincides with T Z, and we have
wz] (Ann Tz N ) = Tz N.
Let (Z, θ) be a contact manifold, v a vector field on Z, and b a function on Z.

One can show [83] that the pair (v, b) is an infinitesimal conformal Jacobi morphism
if and only if the pair (v, −b) is an infinitesimal contact transformation, i.e.,
Lv θ = −bθ.
Example 2.2.3. Let M be a manifold with coordinates (q i ) and Z the odd-

dimensional manifold R×T ∗ M , equipped with the holonomic coordinates (t, q i , pi =
q̇i ). This manifold is provided with the contact form (2.2.6). Hence, it is a Jacobi
manifold where
w = −(∂i + pi ∂t ) ∧ ∂ i , E = −∂t . (2.2.9)
•
2.3 Poisson structure

In accordance with Example 2.1.2, a bivector field
1
2
on a manifold Z defines a Poisson bracket (or a Poisson structure)
{f, g} = w(df, dg) = wµν ∂µ f ∂ν g (2.3.1)
on Z if and only if
[w, w] = 0, wµλ1 ∂µ wλ2 λ3 + wµλ2 ∂µ wλ3 λ1 + wµλ3 ∂µ wλ1 λ2 = 0.
Besides the conditions (A1 – A3), the Poisson bracket (2.3.1) satisfies the Leibniz
rule
{h, f g} = {h, f }g + f {h, g}. (2.3.2)
A manifold Z endowed with a Poisson bivector field w is termed a Poisson manifold,

while a real vector space O0 (Z) of functions on Z forms a Poisson algebra, i.e., O0 (Z)
is a real associative and commutative algebra with unit with respect to pointwise
multiplication, a real Lie algebra with respect to the Poisson bracket, and these two
operations intertwine via the Leibniz rule (2.3.2).
Example 2.3.1. Each manifold admits a zero Poisson structure characterized by
the zero bivector field w = 0. •
Example 2.3.2. Let u and v be vector fields on a manifold Z such that [u, v] = 0
everywhere on Z. It is readily observed from the relation (1.2.12) that w = u ∧ v is
a Poisson bivector field. •
The Poisson structure (2.3.1) defined by a Poisson bivector field w is said to be

regular if the associated morphism w] : T ∗ Z → T Z (1.2.31) is of constant rank.
Hereafter, by a Poisson structure we mean only a regular Poisson structure.
The regular Poisson structure (2.3.1) defined by a Poisson bivector field w is
said to be non-degenerate if the associated morphism w] : T ∗ Z → T Z (1.2.31) is
of maximal rank. A non-degenerate Poisson structure can exist only on an even-
dimensional manifold.
2.3. POISSON STRUCTURE 69
Example 2.3.3. A non-degenerate Poisson manifold (Z, w) is a symplectic manifold

with the symplectic form Ωw (1.2.36) (see Proposition 2.4.4 below). •
Two functions f and g on a Poisson manifold (Z, w) are said to be in involution

with each other if their Poisson bracket {f, g} equals zero. A function C on a Poisson
manifold (Z, w) is called a Casimir function if it is in involution with any function
on Z, i.e.,
{C, f } = 0, ∀f ∈ O0 (Z).
Casimir functions make up the centre of the Poisson algebra (O0 (Z), w). For in-
stance, if a Poisson structure is non-degenerate, all the Casimir functions (on a
connected manifold Z) are constant.
Let (Z1 , w1 ) and (Z2 , w2 ) be Poisson manifolds. A manifold map % : Z1 → Z2 is
said to be a Poisson morphism if
{f ◦ %, g ◦ %}1 = {f, g}2 ◦ %, ∀f, g ∈ O0 (Z2 ),
or
w2 = T % ◦ w1 ,
where T % is the tangent map to %. If % is a Poisson morphism, the rank of w1 (z) is

grater or equal to that of w2 (%(z)). If a Poisson morphism % is an immersion, w1 (z)
and w2 (%(z)) are of equal rank.
A vector field u on a Poisson manifold (Z, w) is the generator of a local 1-
parameter group of local Poisson automorphisms (or an infinitesimal Poisson auto-
morphism) of (Z, w) if and only if
Lu w = [u, w] = 0. (2.3.3)
Such a vector field is called canonical.

Note that there are no pull-back or push-forward operations of Poisson structures
by manifold maps in general. The following assertion deals with Poisson projections
[181].
Proposition 2.3.1. Let (Z, w) be a Poisson manifold and π : Z → Y a projection.

The following properties are equivalent:
• for every pair of functions (f, g) on Y and for each point y ∈ Y , the restriction
of the function {f ◦ π, g ◦ π} to the fibre π −1 (y) is constant;
• there exists a Poisson structure w0 on Y for which π is a Poisson morphism,

i.e.,
{f ◦ π, g ◦ π} = {f, g}0 ◦ π.
One says that the Poisson structure w0 in Proposition 2.3.1 is coinduced by the
projection π.
The direct product Z × Z 0 of Poisson manifolds (Z, w) and (Z 0 , w0 ) has the
Poisson structure defined by the bivector field w + w0 on Z × Z 0 . It is called the
direct product of Poisson structures. Obviously, the projections pr1 and pr2 are
Poisson morphisms.
One can speak of a Poisson submanifold (Z 0 , w0 ) of a Poisson manifold (Z, w) if
Z 0 is a submanifold of Z and the natural injection Z 0 → Z is a Poisson morphism.
Definition 2.3.2. Given a real function f on a Poisson manifold (Z, w), the image
ϑf = w] (df ), (2.3.4)
µν
ϑf = w ∂µ f ∂ν ,
of its exterior differential df by the morphism w] (1.2.31) is called the Hamiltonian

vector field for f with respect to the Poisson structure w. 2
Example 2.3.4. The Hamiltonian vector field for a Casimir function equals zero.
•
It is readily observed that each Hamiltonian vector field is also a canonical vector
field.
The Hamiltonian vector field ϑf for a function f , by definition, obeys the rela-
tions
ϑf cdg = {f, g}, ∀g ∈ O0 (Z),

[ϑf , ϑg ] = ϑ{f,g} . (2.3.5)
Then it follows from (2.3.5) that f 7→ ϑf is the Lie algebra homomorphism.

In particular, we have
ϑf cdf = 0
and, consequently, ϑf is tangent to the leaves of the singular foliation of the level
surfaces of the function f at its regular points, where df 6= 0.
Remark 2.3.5. It is clear that not every vector field on a Poisson manifold Z is
a Hamiltonian vector field. Given a vector field u on a manifold Z, one can try to
construct a Poisson bracket and to find a function on Z such that u would be a
Hamiltonian vector field for this function. A closely related subject is the inverse
problem in Hamiltonian mechanics, which consists in trying to represent a given
system of first order dynamic equations
ż λ = uλ (z) (2.3.6)
on a manifold Z as the Hamilton equation with respect to some Poisson structure,

called the generating Poisson structure on Z. In Section 3.2, we will present a
general technique of constructing a generating Poisson structure for any dynamic
equation (2.3.6) at least locally [59, 82]. •
The values of all Hamiltonian vector fields at all points of a Poisson manifold
Z constitute an even-dimensional characteristic distribution T on the Poisson ma-
nifold (Z, w). By virtue of the relation (2.3.5), this distribution is involutive and,
consequently, completely integrable.
Theorem 2.3.3. [181]. A Poisson structure induces a symplectic structure on

leaves of the characteristic foliation on Z, called a symplectic foliation. 2
In particular, if a Poisson structure is non-degenerate, its characteristic distri-

bution T coincides with T Z, and Z is a symplectic manifold.
Remark 2.3.6. It should be recalled that we are considering a regular Poisson
structure, and the corresponding symplectic foliation is non-singular. •
A 2m-dimensional symplectic foliation on a k-dimensional Poisson manifold Z

admits the adapted coordinates (. . . , z 2m+1 , . . . , z k ) described in Corollary 1.2.4.
Moreover, one can choose these coordinates in such a way to bring the Poisson
structure into the following canonical form [181, 186].
Proposition 2.3.4. For any point z of a Poisson manifold, there exists a coordinate
system
(q 1 , . . . , q m , p1 , . . . , pm , z 2m+1 , . . . , z k ) (2.3.7)
in a neighbourhood of z such that
{pi , q j } = δij , {q i , q j } = {pi , pj } = {q i , z A } = {pi , z A } = {z A , z B } = 0.(2.3.8)
The coordinates (2.3.7) are called canonical coordinates. Given canonical coor-
dinates, one says that the coordinates q i and pi are canonically conjugate.
In canonical coordinates (2.3.7), the Poisson bracket (2.3.1) takes the form
w = ∂ i ∧ ∂i ,
{f, g} = ∂ i f ∂i g − ∂i f ∂ i ,
while the Casimir functions are arbitrary functions of the coordinates z A . Accord-
ingly, the Hamiltonian vector field for a function f reads
ϑf = ∂ i f ∂i − ∂i f ∂ i .
Let (Z, w) be a Poisson manifold with the characteristic distribution T, and N

a submanifold of Z. The familiar definitions of coisotropic and Lagrangian sub-
manifolds of a symplectic manifold are generalized for a Poisson manifold as follows
[181, 187].
Definition 2.3.5. The submanifold N is said to be
• coisotropic if
w] (Ann T N ) ⊆ T N,
• Lagrangian if
w] (Ann T N ) = T N ∩ T.
The following theorems generalize for Poisson manifolds the corresponding re-
sults on symplectic isomorphisms (see [178]).
Theorem 2.3.6. [187]. Let Φ : Z1 → Z2 be a manifold morphism between Poisson

manifolds (Z1 , w1 ) and (Z2 , w2 ). This is a Poisson morphism if and only if its graph
∆Φ = {(z, Φ(z)) z ∈ Z1 } ⊂ Z1 × Z2
is a coisotropic submanifold of the Poisson manifold (Z1 × Z2 , w1 − w2 ). 2
Theorem 2.3.7. [67]. Let u be a vector field on a Poisson manifold (Z, w).
(i) In accordance with the equality (1.2.14), the tangent lift we (1.2.9) of the
Poisson bivector field w is a Poisson bivector field on the tangent bundle T Z of Z.
(ii) A vector field u is an infinitesimal Poisson automorphism of the Poisson
manifold (Z, w) if and only if u(Z) is a Lagrangian submanifold of the Poisson
manifold (T Z, w).
e 2
A Poisson structure on a manifold Z can be defined entirely by its symplectic

foliation instead of by a Poisson bivector field [181].
Theorem 2.3.8. Let Z be a manifold and F a foliation on Z such that every leaf
Fι of F is endowed with a symplectic structure. Given a function f ∈ O0 (Z) on Z,
let ϑιf be a Hamiltonian vector field on a leaf Fι for the function f |Fι with respect
to the symplectic structure on this leaf. Put
def
ϑf (z) = ϑιf (z), z ∈ Fι .
If ϑf is a differentiable vector field on the manifold Z for arbitrary function f , then

Z has a unique Poisson structure given by the Poisson bracket
{f, g}(z) = ϑιf cdg
whose symplectic foliation is F . 2
Example 2.3.7. Let us consider the product Z = R × T ∗ M with the coordinates

(t, q i , pi ). If the cotangent bundle T ∗ M is provided with the canonical symplectic
structure (see Example 2.4.2 below), the fibres of the projection pr1 : Z → R
constitute the symplectic foliation on Z which satisfies the conditions of Theorem
2.3.8. Then, the manifold R × T ∗ M is provided with the Poisson structure given by
the Poisson bivector field
w = ∂ i ∧ ∂i . (2.3.9)
Obviously, this is the product of the zero Poisson structure on R and that defined
by the canonical symplectic structure on the cotangent bundle T ∗ M . •
Example 2.3.8. To generalize Example 2.3.7, let us consider a fibre bundle π :

Z → X, parameterized by fibred coordinates (xλ ; q i , pi ), whose typical fibre V is a
symplectic manifold with the Poisson bracket
{f, g} = ∂ i f ∂i g − ∂i f ∂ i . (2.3.10)
Each trivialization chart
ψξ : π −1 (Uξ ) → Uξ × V
is endowed with the Poisson structure described in Example 2.3.7 with the Poisson
bracket (2.3.10). To generalize construction of symplectic vector bundles [116], we
will say that Z → X is a symplectic bundle if it admits a bundle atlas Ψ whose
transition functions
ρξζ (x) : {x} × V → {x} × V, x ∈ Uξ ∩ Uζ ,
provide isomorphisms of the symplectic manifold V . Then the fibre bundle Z → X

is equipped with the Poisson structure given by the Poisson bracket (2.3.10).
For instance, let Z = V ∗ Q be the vertical cotangent bundle of a fibre bundle Q →
R. The typical fibre of V ∗ Q is the cotangent bundle T ∗ M of the typical fibre M of the
fibre bundle Q → R. The fibre bundle V ∗ Q → R is a symplectic bundle, provided
with the Poisson bracket (2.3.10) written with respect to the coordinates (t, q i , pi ) on
V ∗ Q. Indeed, it is readily observed that this Poisson bracket is invariant under all
holonomic coordinate transformations of the vertical cotangent bundle V ∗ Q. This is
the canonical Poisson structure on a phase space V ∗ Q of time-dependent mechanics.
In Section 5.1, we will obtain it in a different way.
The notion of a symplectic bundle is naturally extended to that of a Poisson
bundle. One should distinguish this notion from that of Jacobi bundles in the
2.4. SYMPLECTIC STRUCTURE 75
sense of Kirillov [98, 123] where a Jacobi bracket is generalized for sections of a
1-dimensional linear bundle, called a Jacobi bundle. •
Example 2.3.9. The Lie coalgebra g∗ of a Lie group G is provided with the
canonical Poisson structure, called the Lie–Poisson structure (see [2, 116, 181]).
This Poisson structure is given by the bracket
def
{f, g} =hε∗ , [df (ε∗ ), dg(ε∗ )]i, f, g ∈ O0 (g∗ ), (2.3.11)
on g∗ , where df (ε∗ ), dg(ε∗ ) ∈ gr since we can regard them as linear mappings from
Tε∗ g∗ = g∗ to R. The Lie–Poisson bracket (2.3.11) is defined by the Poisson bivector
field
wmn = ckmn zk ,
where ckmn are the structure constants of the Lie algebra gr and zk are coordinates
on g∗ with respect to a basis {εk }. We have the coordinate expression
{f, g} = ckmn zk ∂ m f ∂ n g.
There is the following well-known theorem.
Theorem 2.3.9. [186]. The symplectic leaves of the Lie–Poisson structure on the
coalgebra g∗ of a connected Lie group G are the orbits of the coadjoint representation
(1.5.3) of G on g∗ . 2 •
2.4 Symplectic structure

Non-degenerate Poisson manifolds are symplectic manifolds (see Proposition 2.4.4
below).
Definition 2.4.1. A non-degenerate exterior 2-form Ω on a manifold Z is said to

be symplectic if it is closed, i.e., dΩ = 0. A manifold equipped with a symplectic
form is called a symplectic manifold. 2
Every symplectic manifold (Z, Ω) is 2m-dimensional. It is orientable, and

1 m
V= ∧Ω (2.4.1)
m!
is the volume element on Z.
Definition 2.4.2. A morphism ζ : Z → Z 0 of a symplectic manifold (Z, Ω) to a

symplectic manifold (Z 0 , Ω0 ) is called a symplectic morphism if Ω = ζ ∗ Ω0 . 2
By very definition, a symplectic morphism is an immersion.

Remark 2.4.1. One should distinguish a symplectic morphism from a symplecto-
morphism in the terminology of [116]. The latter is a diffeomorphism. •
A vector field u on a symplectic manifold (Z, Ω) is said to be a generator of a

local 1-parameter group of symplectic automorphisms (or an infinitesimal symplectic
automorphism) of a symplectic manifold (Z, Ω) if and only if
Lu Ω = 0.
Such a vector field is called canonical.

Example 2.4.2. Let M be a manifold with coordinates (q i ) and π∗M : T ∗ M →
M its cotangent bundle, equipped with the holonomic coordinates (q i , pi ). The
cotangent bundle T ∗ M is endowed with the canonical Liouville form
θ = pi dq i .
This form is defined by the condition
vcθ(p) = T π∗M (v)cp, ∀v ∈ Tp T ∗ M, p ∈ T ∗ M.
Its exterior differential is the canonical symplectic form
Ω = dθ = dpi ∧ dq i (2.4.2)
on T ∗ M . All holonomic coordinates on the cotangent bundle T ∗ M are canonical for

this form.
It should be emphasized that the canonical symplectic form (2.4.2) is not a
unique symplectic form on the cotangent bundle T ∗ M . For every closed 2-form φ
on a manifold M , the form
∗
Ωφ = Ω + π∗M φ (2.4.3)
is also a symplectic form on T ∗ M [116]. •

The canonical symplectic form (2.4.2) plays the fundamental role in view of
Darboux’s theorem [116].
Theorem 2.4.3. Let (Z, Ω) be a symplectic manifold. Each point of Z has an open
neighbourhood U which is the domain of a canonical coordinate chart
(q 1 , . . . , q m , p1 , . . . , pm )
such that the symplectic form Ω on U is given by the coordinate expression (2.4.2).
2
This theorem is an immediate consequence of Proposition 2.3.4 and Proposition

2.4.4 below.
We have the following relationship between the non-degenerate Poisson struc-
tures and the symplectic ones.
Proposition 2.4.4. [117]. On a 2m-dimensional manifold Z, there is one-to-one

correspondence between the symplectic forms Ω and the non-degenerate Poisson
bivector fields w given by the equalities (1.2.35) – (1.2.36). 2
The canonical coordinates for a symplectic form Ω are also canonical for the
corresponding Poisson bivector field w, and we have
Ω = dpi ∧ dq i , w = ∂ i ∧ ∂i .
With respect to these coordinates, the bundle isomorphism Ω[ (1.2.33) reads
Ω[ : v i ∂i + vi ∂ i 7→ −vi dq i + v i dpi ,
while the isomorphism w] = (Ω[ )−1 (1.2.31) is
w] : vi dq i + v i dpi 7→ v i ∂i − vi ∂ i .
In view of Proposition 2.4.4, a Poisson structure is termed sometimes a cosym-
plectic structure (one should distinguish this notion from the cosymplectic structure
in Remark 4.8.8). A local structure of an arbitrary Poisson manifold in relation to
a symplectic one is described by the following two theorems [181, 186].
Theorem 2.4.5. Any point of a Poisson manifold has an open neighbourhood which
is Poisson equivalent to a product of a manifold with the zero Poisson structure and
a symplectic manifold. 2
Let (Z, w) be a Poisson manifold. Its symplectic realization is said to be a

symplectic manifold (Z 0 , Ω) together with a projection Z 0 → Z such that the Poisson
structure w on Z is coinduced from the Poisson structure wΩ on Z 0 .
Theorem 2.4.6. Any point of a Poisson manifold has an open neighbourhood

which is realizable by a symplectic manifold. 2
In local Darboux coordinates, this symplectic realization is seen as follows. The

Poisson structure given by the Poisson bracket (2.3.8) with respect to the canonical
coordinates is coinduced from the symplectic structure given by the symplectic form
Ω = dpi ∧ dq i + dz A ∧ dz A
with respect to the coordinates
(q 1 , . . . , q m , p1 , . . . , pm , z 2m+1 , . . . , z k , z 2m+1 , . . . , z k )
by the projection
(q 1 , . . . , q m , p1 , . . . , pm , z 2m+1 , . . . , z k , z 2m+1 , . . . , z k ) 7→
(q 1 , . . . , q m , p1 , . . . , pm , z 2m+1 , . . . , z k ).
Example 2.4.3. Let V ∗ Q be the vertical cotangent bundle of a fibre bundle Q → R

as in Example 2.3.8. It is provided with the Poisson structure given by the Poisson
bracket
{f, g} = ∂ i f ∂i g − ∂i f ∂ i g.
Its symplectic realization is the cotangent bundle T ∗ Q of Q endowed with the canon-
ical symplectic form
dΞ = dp ∧ dt + dpi ∧ dq i , (2.4.4)
written with respect to the holonomic coordinates (t, q i , p, pi ) (see Section 5.1). •
Let (Z1 , Ω1 ) and (Z2 , Ω2 ) be symplectic manifolds equipped with the associated
Poisson structures w1 and w2 , respectively. The map % : Z1 → Z2 is a Poisson
isomorphism if it is a symplectic isomorphism. Accordingly, a vector field on a
symplectic manifold is canonical if and only if it is canonical for the associated
Poisson structure. However, if dim Z1 > dim Z2 and % is a Poisson morphism, then
this is not a symplectic morphism, i.e.,
w2 = T % ◦ w1 , Ω1 6= %∗ Ω2 .
The notion of a Hamiltonian vector field on a Poisson manifold is restated for a

symplectic one as follows.
Definition 2.4.7. A vector field ϑ on a symplectic manifold (Z, Ω) is said to be

locally Hamiltonian [Hamiltonian] if the exterior form ϑcΩ is closed [exact]. 2
As an immediate consequence of Definition 2.4.7, we find the following.
• A vector field ϑ on a symplectic manifold (Z, Ω) is locally Hamiltonian if and

only if it is an infinitesimal symplectic automorphism, i.e.,
Lϑ Ω = d(ϑcΩ) = 0.
• A vector field ϑ on a symplectic manifold (Z, Ω) is Hamiltonian if and only if

it is a Hamiltonian vector field in accordance with Definition 2.3.2 for some
function f on Z, and then
ϑf cΩ = −df. (2.4.5)
• The Poisson bracket defined by a symplectic form Ω reads
{f, g} = ϑg cϑf cΩ.
In the literature (see, e.g., [2]), one may meet a definition of Hamiltonian vector
fields, which differs in the minus sign from (2.4.5).
Example 2.4.4. Let us consider the cotangent bundle T ∗ M equipped with the
canonical symplectic form Ω (2.4.2). Let u = ui ∂i be a vector field on M . Then its
canonical lift
ue = ui ∂i − pj ∂i uj ∂ i
(1.2.3) onto T ∗ M is a Hamiltonian vector field. We have
uecΩ = −d(ui pi ).
•
Let us turn now to submanifolds of a symplectic manifold.
Let (Z, Ω) be a 2m-dimensional symplectic manifold and w the corresponding
Poisson bivector field on Z. Let N be an n-dimensional submanifold of Z. The set
def [
OrthΩ T N = {v ∈ Tz Z : vcucΩ = 0, ∀u ∈ Tz N }, (2.4.6)
z∈N
is called the orthogonal of T N relative to the symplectic form Ω or simply Ω-

orthogonal space of T N .
Recall the useful formulas
OrthΩ (OrthΩ T N ) = T N,
Ann T N = Ω[ (OrthΩ T N ),
w] (Ann T N ) = OrthΩ T N,
Ann (OrthΩ T N ) = Ω[ (T N ).
The inclusion T N1 ⊂ T N2 is equivalent to OrthΩ T N1 ⊃ OrthΩ T N2 . We have also
OrthΩ (T N1 ∩ T N2 ) = OrthΩ T N1 + OrthΩ T N2
and, in particular,
T N ∩ OrthΩ T N = OrthΩ (T N + OrthΩ T N ).
Note that, in general,
T N ∩ OrthΩ T N 6= 0,
while T Z |N is not the sum T N + OrthΩ T N .
Henceforth, when there is no risk of confusion, we will write OrthT N instead of
OrthΩ T N .
Let ΩN = i∗N Ω be the restriction of the symplectic form Ω to the submanifold
N . This is a presymplectic form on N . Its kernel (1.2.34) is
Ker ΩN = T N ∩ OrthΩ T N.
The following special submanifolds of a symplectic manifold (Z, Ω) are usually
considered. The presymplectic form ΩN on any of such special manifolds is always
of constant rank.
Definition 2.4.8. A submanifold N ⊂ Z of a 2m-dimensional symplectic manifold

(Z, Ω) is said to be (cf. Definition 2.3.5):
• coisotropic if OrthΩ N ⊂ T N , n ≥ m;
• isotropic if T N ⊂ OrthΩ N , n ≤ m;
• Lagrangian if OrthΩ N = T N , n = m;
• symplectic if ΩN is a symplectic form on N .
As an immediate consequence of this definition, a symplectic form restricted to

isotropic or Lagrangian submanifolds is equal to zero.
One can classify the germs of special submanifolds of a symplectic manifold [48].
Recall that a germ of a submanifold N at a point z ∈ Z is the equivalence class
(N ; z) of submanifolds of the manifold Z which pass through z and coincide with
N in an open neighbourhood of z.
With respect to local canonical coordinates (q i , pi ), a Lagrangian submanifold of
a symplectic manifold is given by the equations
q b = ∂ b S, pa = −∂a S, (2.4.7)
where S(q a , pb ) is a function, called the generating function, of the m variables

{q a , pb ; a ∈ A, b ∈ B} for some partition (A, B) of the set (1, . . . , m). Then the
germ of a Lagrangian submanifold (N ; z) is symplectically equivalent to the germ
of the subspace
{(q, p) ∈ R2m : pi = 0, i = 1, . . . , m} (2.4.8)
at the point 0 ∈ R2m of the symplectic space R2m . The expression (2.4.8) results
from (2.4.7) by means of the symplectic automorphisms
q b 7→ −pb , pb 7→ q b
and
pi → pi + ∂i S(q a , q b ).
Example 2.4.5. Let the cotangent bundle T ∗ M of a manifold M be equipped

with the canonical symplectic form Ω. Then the image 0(M
b ) of the zero section
0b of T ∗ M → M is a Lagrangian submanifold {pi = 0} of the symplectic manifold

(T ∗ M, Ω). •
The germ of a (2m − r)-dimensional coisotropic submanifold (N ; z) is symplec-

tically isomorphic to the germ of the subspace
{(q, p) ∈ R2m : q i = 0, i = 1, . . . , r}
at the point 0 ∈ R2m of the symplectic space R2m .

The germ of a 2(m − r)-dimensional symplectic submanifold (N ; z) is symplec-
tically isomorphic to the germ of the subspace
{(q, p) ∈ R2m : q i = pi = 0, i = 1, . . . , r}
at the point 0 ∈ R2m of the symplectic space R2m .

We will need the following two constructions in the sequel [178].
Example 2.4.6. Let (Z1 , Ω1 ) and (Z2 , Ω2 ) be symplectic manifolds. The 2-form
Ω1 Ω2 = pr∗1 Ω1 − pr∗2 Ω2
is clearly a symplectic form on the product Z1 × Z2 . Then the graph of a sym-

plectic diffeomorphism of (Z1 , Ω1 ) onto (Z2 , Ω2 ) is a Lagrangian submanifold of the
symplectic manifold (Z1 × Z2 , Ω1 Ω2 ). •
Example 2.4.7. Let Ω be the canonical symplectic form (2.4.2) on the cotangent
bundle T ∗ M of an m-dimensional manifold M . Then its tangent lift Ω
e (1.2.23) is
the symplectic form
e = dṗ ∧ dq i + dp ∧ dq̇ i
Ω (2.4.9)
i i
on the tangent bundle T T ∗ M of T ∗ M , written with respect to the holonomic co-

ordinates (q i , q̇ i , pi , ṗi ) on T T ∗ M . Due to the isomorphism α : T T ∗ M ∼
= T ∗T M
(1.1.10), the form (2.4.9) is also the pull-back α∗ ΩT ∗ T M of the canonical symplectic
form ΩT ∗ T M on the cotangent bundle T ∗ T M of the manifold T M .
Let H be a function on the cotangent bundle T ∗ M . One can think of H as being
a Hamiltonian of conservative mechanics. The Hamiltonian vector field
ϑH = ∂ i H∂i − ∂i H∂ i
2.5. PRESYMPLECTIC STRUCTURE 83
for H defines the closed 2m-dimensional submanifold
q̇ i = ∂ i H, ṗi = −∂i H (2.4.10)
of the 4m-dimensional symplectic manifold (T T ∗ M, Ω).

e The restriction of the sym-
plectic form (2.4.9) to the submanifold (2.4.10) is equal to 0. Hence, this is a
Lagrangian submanifold with the generating function H(q i , pi ).
Let L be a function on the tangent bundle T M . On can think of L as being a
Lagrangian of conservative mechanics. The exterior differential dL of L yields the
morphism
α−1 ◦ dL : T M → T ∗ T M → T T ∗ M.
Its image, given by the coordinate relations
ṗi = ∂i L, pi = ∂˙i L, (2.4.11)
is a Lagrangian submanifold of the symplectic manifold (T T ∗ M, Ω).

e Its generating
function is L(q i , q̇ i ).
Furthermore, let φ be an exterior 1-form on the tangent bundle T M , treated as
a generalized Lagrangian [10]. Then we have the morphism
α−1 ◦ φ : T M → T ∗ T M → T T ∗ M
and the relation
(α−1 ◦ φ)∗ Ω
e = dφ.
It follows that (α−1 ◦φ)(T M ) is a Lagrangian submanifold of the symplectic manifold

e if and only if the 1-form φ on T M is closed. •
(T T ∗ M, Ω)
2.5 Presymplectic structure

As in the case of Poisson structures, there are no pull-back or push-forward opera-
tions of symplectic structures by manifold maps in general. Pull-backs of symplectic
forms are the presymplectic ones.
Definition 2.5.1. An exterior 2-form Ω on a manifold Z is said to be presymplectic

if it is closed, but not necessarily non-degenerate. A manifold equipped with a
presymplectic form is called a presymplectic manifold. 2
Using the formula (1.2.29), one can justify the fact that the kernel Ker Ω (1.2.34)
of a presymplectic form Ω of constant rank is an involutive distribution, called the
characteristic distribution [116]. It defines the characteristic foliation on a presym-
plectic manifold Z. The restriction of the presymplectic form Ω to the leaves of this
foliation is equal to zero.
Remark 2.5.1. In contrast with a symplectic structure, a presymplectic one is
not associated with any Poisson structure in a canonical way. Nevertheless, there
is a construction [49] which may make a Poisson bivector field w correspond to a
presymplectic form Ω on a manifold Z in accordance with the relations
w ] ◦ Ω[ ◦ w ] = w ] ,
Ker w] ∩ Im Ω[ = 0.
The notion of a Hamiltonian vector field on a symplectic manifold is extended

in a straightforward manner to a presymplectic manifold.
Definition 2.5.2. A vector field ϑ on a presymplectic manifold (Z, Ω) is said to

be locally Hamiltonian [Hamiltonian] if the exterior form ϑcΩ is closed [exact]. 2
As in a symplectic case, we find the following.
• A vector field ϑ on a presymplectic manifold (Z, Ω) is locally Hamiltonian if

and only if it is an infinitesimal automorphism of this presymplectic manifold,
i.e.,
Lϑ Ω = d(ϑcΩ) = 0.
• A Hamiltonian vector field ϑf for a function f on a presymplectic manifold

obeys the relation
ϑf cdf = 0
and, consequently, is tangent to the leaves of the singular foliation of the level
surfaces of the function f at regular points, where df 6= 0.
2.5. PRESYMPLECTIC STRUCTURE 85
Note however that, in contrast with Poisson manifolds, not each function on a
presymplectic manifold admits an associated Hamiltonian vector field.
The pull-back of a symplectic form by a manifold map is obviously a presym-
plectic form. The converse is the following.
Proposition 2.5.3. Any presymplectic form φ on a manifold M can be represented

as the pull-back of some symplectic form. 2
Proof. Let φ be a presymplectic form on a manifold M . Then it is the pull-back
φ = 0b∗ (Ω + π∗M
∗
φ)
of the symplectic form Ωφ (2.4.3) on the cotangent bundle T ∗ M of M by the global

zero section 0b of T ∗ M → M . QED
Moreover, it is readily observed that
OrthΩφ T 0(M
b ) = (T 0)(Ker
b φ) ⊂ T 0(M
b ).
It follows that the imbedding 0b of the presymplectic manifold (M, φ) into the sym-
plectic manifold (T ∗ M, Ωφ ) is coisotropic.
If a presymplectic form Ω on a manifold M is of constant rank, the stronger
result holds [62, 72].
Proposition 2.5.4. Given a presymplectic manifold (M, Ω) where Ω is of con-

stant rank, there exists a symplectic form on a tubular neighbourhood of the zero
section of the dual bundle to the characteristic distribution Ker Ω, where M can be
coisotropically imbedded. 2
There is another well-known pull-back construction, where a presymplectic form

is seen as a pull-back of a symplectic form by a surjective submersion.
Proposition 2.5.5. Let a presymplectic form Ω on manifold M be of constant rank

and its characteristic foliation be simple, i.e., its leaves are fibres of a fibre bundle
π : M → P . Then the base P of this fibre bundle is equipped with a symplectic
form ΩP such that Ω is the pull-back of ΩP by π. 2
Proof. The symplectic form ΩP on P is defined by the relation

def
ΩP (v, v 0 ) = Ω(v, v 0 )
for any v ∈ T π −1 (v), v 0 ∈ T π −1 (v 0 ). QED
This pull-back construction is the key point of what is called a reduction of a

symplectic structure.
2.6 Reduction of symplectic and Poisson structures

A submanifold N of a symplectic manifold (Z, Ω) fails to be a symplectic one in
general. At the same time, if the presymplectic form ΩN = i∗N Ω on N is of constant
rank and the characteristic foliation of ΩN is simple, i.e., its leaves are fibres of a
fibre bundle N → P , the base P of this fibre bundle is a symplectic manifold in
accordance with Proposition 2.5.5.
This construction is called a reduction of a symplectic structure. It is general-
ized for Poisson structures. Important reduction processes appear in connection,
e.g., with constraint Hamiltonian systems and Lie group actions on symplectic and
Poisson manifolds.
Definition 2.6.1. [2, 116]. A symplectic reduction of a symplectic manifold (Z, Ω)

is said to be a surjective submersion π : N → P of a submanifold N ⊂ Z onto a
symplectic manifold (P, ΩP ), which satisfies
π ∗ ΩP = i∗N Ω.
One says that (P, ΩP ) is the reduced symplectic manifold of the symplectic manifold
(Z, Ω) via the submanifold N . 2
The above speculations show that, given a submanifold N of a symplectic ma-

nifold (Z, Ω), there exists a symplectic reduction π : N → P with connected fibres
if and only if
• the presymplectic form ΩN = i∗N Ω on N is of constant rank,
• the characteristic foliation of ΩN is simple.
Proposition 2.6.2. [116]. Let (Z, Ω) be a symplectic manifold and N a submani-

fold of Z.
(i) If π : N → P is a symplectic reduction, then the rank of the pull-back
presymplectic form ΩN on N is constant and equal to dim P , Ker ΩN = V N , and
2.6. REDUCTION OF SYMPLECTIC AND POISSON STRUCTURES 87
the connected components of fibres of π are the leaves of the characteristic foliation
for ΩN . Furthermore, if π has connected fibres, the characteristic foliation of ΩN is
simple, and is exactly the fibration N → P .
(ii) Conversely, if the rank of the presymplectic form ΩN on a submanifold N of
a symplectic manifold (Z, Ω) is constant and the characteristic foliation for ΩN is
simple, the base P of the corresponding fibration N → P has a unique symplectic
form such that this fibration is a symplectic reduction of (Z, Ω) via N . This is a
unique symplectic reduction via N with connected fibres. Furthermore, if N → P 0
is another symplectic reduction, there exists a surjection P → P 0 which is a local
symplectic isomorphism. 2
A coisotropic submanifold N provides the most interesting case when the presym-
plectic form ΩN is of constant rank which is neither dim N nor 0.
Example 2.6.1. Let (M, φ) be a presymplectic manifold, where a presymplectic
form φ is of constant rank and its characteristic foliation is simple. Combining the
pull-back constructions in Propositions 2.5.3 and 2.5.5 gives the reduction of the
symplectic structure Ωφ (2.4.3) on the cotangent bundle T ∗ M via the coisotropic
submanifold 0(M
b ). •
The reduction procedure is extended to Poisson manifolds as follows.
Definition 2.6.3. [129]. By a reductive structure of a Poisson manifold (Z, w) is

meant a triple (Z, N, E) of a submanifold N of M and a vector subbundle E ⊂ T Z |N
together with a submersion π : N → P if the following conditions are satisfied:
(i) E ∩ T N is tangent to the fibres of the submersion N → P ;
(ii) if df and dg, where f, g ∈ O0 (Z), belong to Ann E, so is d{f, g}w .
A reductive structure is said to be a Poisson reduction if: (i) P above is a Poisson
manifold with a Poisson bivector field W and (ii) for any local functions f, g on
P and for any local extensions f , g onto Z of the pull-backs f ◦ π, g ◦ π such that
df , dg ⊂ Ann E, the relation
{f , g}w ◦ iN = {f, g}W ◦ π
holds good. One says that (P, W ) is the reduced Poisson manifold of (Z, w) via
(N, E). 2
The following assertion furnishes the necessary and sufficient conditions for a
reductive structure of a Poisson manifold to be a Poisson reduction.
Proposition 2.6.4. [129, 181]. Let (Z, N, E) be a reductive structure of a Poisson

manifold (Z, w). This is a Poisson reduction if and only if
w] (Ann E) ⊆ T N + E. (2.6.1)
The functorality property of the Poisson reduction is given by the following.
Proposition 2.6.5. [129]. Let (Z, N, E) and (Z 0 , N 0 , E 0 ) be Poisson reductions,

and let Φ : Z → Z 0 be a Poisson map such that Φ(N ) ⊂ N 0 , T Φ(E) ⊂ E 0 , and Φ
sends the leaves of N to the leaves of N 0 . Then Φ induces a unique Poisson map
b : P → P 0 , called the reduction of Φ, such that π 0 ◦ Φ = Φ
Φ b ◦ π. 2
Example 2.6.2. Let Z = T ∗ G, where G is a connected Lie group, N = Z, and E

comprise the tangent space to the right G-orbits. Then (Z, E) is a Poisson reduction,
where π(Z) = T ∗ G/G = g∗ is the Lie coalgebra of G provided with the Lie–Poisson
structure (2.3.11) [129, 130]. •
Example 2.6.3. To compare a Poisson reduction with a symplectic one, let (Z, Ω)
be a symplectic manifold and w the corresponding Poisson bivector field on Z.
Let N be a submanifold of Z such that the presymplectic form ΩN = i∗N Ω is of
constant rank. Put E = OrthT N . Then E ∩ T N = Ker ΩN is tangent to the
characteristic foliation of ΩN . If there exists a symplectic reduction π : N → P
of (Z, Ω) to (P, ΩP ), one can show that (Z, N, OrthT N ) is a reductive structure of
the Poisson manifold (Z, w). The condition (i) of Definition 2.6.3 holds since the
integral manifolds of the distribution Ker ΩN are connected components of fibres of
N → P . Furthermore, since we are on a symplectic manifold, df ∈ Ann (OrthT N )
if and only if the Hamiltonian vector field ϑf for f belongs to
w] (Ann (OrthT N )) = T N. (2.6.2)
Then, ϑf , ϑg ∈ T N implies
{ϑf , ϑg } = ϑ{f,g} ∈ T N.
Hence, condition (ii) of Definition 2.6.3 also holds. Moreover, the inclusion (2.6.1)
obviously takes place because of (2.6.2). It follows that (Z, N, OrthT N ) is a Poisson
2.6. REDUCTION OF SYMPLECTIC AND POISSON STRUCTURES 89
reduction and the corresponding reduced structure W on P is associated with the

symplectic form ΩP . On the other hand, if we take E = Ker ΩN , the inclusion
(2.6.1) requires N to be coisotropic, since
w] (Ann E) = T N + OrthT N,
and one may apply the previous result. •
One can extend this Example to the Poisson manifold case as follows.
Lemma 2.6.6. [181]. Let N be a submanifold of a Poisson manifold (Z, w) such

that
def
C(N ) = w] (Ann T N ) ∩ T N (2.6.3)
is a distribution, i.e., is of constant rank. Then C(N ) is involutive. Furthermore, if N

is transversal to the leaves of the symplectic foliation F for w on Z, the distribution
C(N ) defines a subfoliation of F ∩ T N which is transversally symplectic along each
leaf of the latter. 2
If the foliation C(N ) is simple and corresponds to a fibration π : N → P ,

then F ∩ T N projects to a symplectic foliation π(F ) on P , and P has a well-
defined Poisson structure given by this foliation. This is called the leafwise reduction
of (Z, w) via N . It follows in the same way as for the symplectic case discussed
earlier that, if w] (Ann T N ) has a constant dimension, then (Z, N, w] (Ann T N )) is
a reductive structure. Because of
w] (Ann (w] (Ann T N ))) ⊆ T N
and the relation (2.6.1), the reductive triple (Z, N, w] (Ann T N )) is also a Poisson
reduction, while the reduced Poisson manifold P is exactly the one defined by the
leafwise reduction.
Given a Poisson manifold (Z, w), let (P, W ) be a reduced Poisson manifold of
(Z, w) via a submanifold N . Then one can say that the Poisson algebra O0 (P ) of
smooth functions on P is the reduction of the Poisson algebra of O0 (Z) of smooth
functions on Z. If N is a closed imbedded submanifold of a Poisson manifold Z,
this reduction is described in an intrinsic way, i.e., in terms of ideals of the Poisson
algebra O0 (Z) as follows [97].
Let N be a closed imbedded submanifold of a Poisson manifold (Z, w). Let us

consider the set I of real functions f on Z which vanish on N , i.e., i∗N f = 0. This
set is the kernel I = Ker i∗N of the linear morphism
i∗N : O0 (Z) → O0 (N ),
and it is an ideal of the associative commutative algebra O0 (Z). Since N is a closed

and imbedded submanifold of Z, we have the isomorphism
O0 (Z)/I ∼
= O0 (N ) (2.6.4)
of associative commutative algebras.

Let us consider the space of all vector fields on Z which restrict to vector fields
on N . It is given by
def
TN ={u ∈ T (Z) : ucdf ∈ I, ∀f ∈ I}. (2.6.5)
It is clear that TN 6= T (Z) |N . Suppose that ϑf is a Hamiltonian vector field which

restricts to a vector field on N . By the very definition (2.6.5), this means that
ϑf cdg = {f, g} ∈ I, ∀g ∈ I.
Hence, the functions whose Hamiltonian vector fields restrict to vector fields on N
are those functions in the normalizer of I, denoted by
def
I(N ) ={f ∈ O0 (Z) : {f, g} ∈ I, ∀g ∈ I}. (2.6.6)
It follows from the Jacobi identity that the normalizer (2.6.6) is a Poisson subalgebra
of O0 (Z). Put
def
I 0 (N ) = I(N ) ∩ I. (2.6.7)
It is naturally a Poisson subalgebra of I(N ). This subalgebra is generally non-zero

since I 2 ⊂ I 0 (N ) due to the Leibniz rule.
The following Theorem leads us back to the reduction procedure.
Theorem 2.6.7. [97]. Let N be a closed imbedded submanifold of a Poisson

manifold Z. Let us assume that w] (Ann T N ) and C (2.6.3) are distributions of
constant rank, and the foliation C is simple. Then (Z, N, w] (Ann T N )) is a reductive
structure such that there is the associative commutative algebra isomorphism
O0 (P ) ∼
= I(N )/I 0 (N ). (2.6.8)
2.7. APPENDIX. POISSON HOMOLOGY AND COHOMOLOGY 91
Since the right-hand side is naturally a Poisson algebra, this isomorphism defines a
Poisson structure on P . 2
Remark 2.6.4. The result is based on the fact that all sections of w] (Ann T N ) are
restrictions to N of Hamiltonian vector fields for elements in I, while all sections of
C are Hamiltonian vector fields for elements in I 0 (N ) restricted to N . In particular,
if N is coisotropic, I ⊆ I(N ), i.e., I = I 0 (N ) is a Poisson subalgebra of O0 (Z). •
Theorem 2.6.7 shows that a Poisson reduction procedure can be seen as an

algebraic one; that leads to the following purely algebraic definition [97].
Definition 2.6.8. Let P be a Poisson algebra, J be an ideal of P, J(N ) its

normalizer (2.6.6), and J 0 (N ) = J(N ) ∩ J. One says that the Poisson algebra
J(N )/J 0 (N ) is the reduction of the Poisson algebra P via the ideal J. 2
In particular, an ideal J of a Poisson algebra P is said to be coisotropic if J is

a Poisson subalgebra of P.
The above algebraic reduction procedure has been generalized for Jacobi mani-
folds [87].
We meet the Poisson reduction in the sense of Definition 2.6.8 in connection with
the BRST scheme [97, 169].
2.7 Appendix. Poisson homology and cohomology

This Section is concerned with the Koszul–Brylinski–Poisson homology and the
Lichnerowicz–Poisson cohomology of a Poisson manifold, which find their applica-
tion in the quantization procedure.
Remark 2.7.1. Let us recall briefly the basic notions of homology and cohomology
of complexes [17, 122].
A sequence
∂
0 1 ∂ 2 ∂ ∂p ∂p+1
0 ←− B0 ←− B2 ←− · · · ←− Bp ←− · · · (2.7.1)
of Abelian groups Bp and homomorphisms ∂p is said to be a chain complex if
∂p ◦ ∂p+1 = 0, ∀p ∈ N,
i.e., Im ∂p+1 ⊂ Ker ∂p . The quotient

def
Hp (B∗ ) = Ker ∂p /Im ∂p+1
is called the pth homology group of the chain complex B∗ (2.7.1). The chain complex
(2.7.1) is said to be exact at an element Bp if Hp (B∗ ) = 0. It is called an exact
sequence if it is exact at each element.
A sequence
δ0 δ1 δ p−1 δp
0 ,→ B 0 −→ B 1 −→ · · · −→ B p −→ · · · (2.7.2)
of Abelian groups B p and homomorphisms δ p is said to be a cochain complex if
δ p ◦ δ p−1 = 0, ∀p ∈ N.
The pth cohomology group of the cochain complex B ∗ (2.7.2) is the quotient
def
H p (B ∗ ) = Ker δ p /Im δ p−1 .
The De Rham complex of exterior forms on a manifold Z

d d d d
· · · −→ Or−1 (Z) −→ Op (Z) −→ Or+1 (Z) −→ · · ·
exemplifies a cochain complex. Its cohomology group H p (Z), called the De Rham
pth cohomology group, is the quotient of the space of closed p-forms by the subspace
of exact p-forms. •
Given a Poisson manifold (Z, w), let us consider the operator
δw : Or (Z) → Or−1 (Z),

def
δw = wc ◦ d − d ◦ wc, (2.7.3)
on the exterior algebra O∗ (Z), called the Poisson codifferential [19, 102, 181]. It is
given by the coordinate expression
r
(−1)i+1 {f0 , fi }df1 ∧ · · · ∧ df
X
δw (f0 df1 ∧ · · · ∧ dfr ) = c ∧ · · · ∧ df +
i r
i=1
i+j
X
(−1) f0 d{fi , fj } ∧ df1 ∧ · · · ∧ df
c ∧ · · · ∧ df
i
c ∧ · · · ∧ df .
j r
1≤i<j≤r
Henceforth, when there is no risk of confusion, we shall write δ instead of δw .

The operator δ (2.7.3) obeys the equalities
δ ◦ δ = 0, (2.7.4)
d ◦ δ + δ ◦ d = 0. (2.7.5)
Due to the nilpotency property (2.7.4), we have the chain complex

δ δ δ δ
· · · ←− Op−1 (Z) ←− Op (Z) ←− Op+1 (Z) ←− · · · ,
called the canonical complex of the Poisson manifold (Z, w). The homology H∗can (Z)
of this complex is termed the canonical or the Koszul–Brylinski–Poisson homology.
Furthermore, due to the property (2.7.5), the periodic double complex
Er,l (Z) = Ol−r (Z),

δ : Er,l (Z) → Er,l−1 (Z), d : Er,l (Z) → Er−1,l (Z),
can also be defined.

Let (Z, Ω) be a 2m-dimensional symplectic manifold, provided with the volume
form V (2.4.1), and wΩ the corresponding Poisson bivector field. Imitating the
well-known star isomorphism for Riemannian manifolds, one introduces the star
operation
∗ : Or (Z) → O2m−r (Z),

def
∗φ = wΩ] (φ)cV,
where w] is the homomorphism (1.2.32). The (2m − r)-form ∗φ is called the adjoint
of the r-form φ relative to the symplectic form Ω. The following equalities
∗(∗φ) = φ, φ ∈ O∗ (Z),
∗(φ ∧ σ) = wΩ] (φ)c(∗σ) = (−1)|φ||σ| wΩ] (σ)c(∗φ)
are fulfilled [116].
Lemma 2.7.1. [19]. In the case of a symplectic manifold, the codifferential δ (2.7.3)
on Or (Z) is given by the formula
δ = (−1)r+1 ∗ d ∗ . (2.7.6)
2
The relation (2.7.6) establishes an isomorphism of the canonical homology group

Hican (Z)with the De Rham cohomology group H 2m−i (Z) of a symplectic manifold
as follows.
Theorem 2.7.2. [19, 181]. If (Z, Ω) is a 2m-dimensional symplectic manifold, then
Hican (Z) = H 2m−i (Z). (2.7.7)
Given a Poisson manifold (Z, w), let us now introduce the operation
wb : Tr (Z) → Tr+1 (Z),

def
w(ϑ)
b = −[w, ϑ], ϑ ∈ T∗ (Z), (2.7.8)
on the exterior algebra T∗ (Z) of multivector fields on Z, where [., .] is the Schouten–
Nijenhuis bracket (1.2.10). This operation satisfies the rules
wb ◦ wb = 0, (2.7.9)
b ∧ υ) = w(ϑ)
w(ϑ b ∧ υ + (−1)|ϑ| ϑ ∧ w(υ),
b (2.7.10)
and is sometimes called a contravariant exterior differential [181]. Its relation to the
familiar exterior differential is
b ] (φ)) = −w ] (dφ),
w(w φ ∈ O∗ (Z), (2.7.11)
while that to the codifferential δw (2.7.3) is given by the formula
w(ϑ)cφ
b = ϑcδw (φ) + (−1)|φ| δw (ϑcφ), | ϑ |=| φ | −1. (2.7.12)
Example 2.7.2. It is readily observed that
−w(f
b ) = [w, f ] = ϑf , f ∈ T0 (Z),
is the Hamiltonian vector field for a function f on Z. •
Due to the nilpotency rule (2.7.9), the operation wb (2.7.8) provides the cochain
complex
w w w w
· · · −→ Tr−1 (Z) −→ Tr (Z) −→ Tr+1 (Z) −→ · · · ,
b b b b
r
called the Lichnerowicz–Poisson cochain complex. The cohomology groups HLP (Z, w)
of this complex are said to be Lichnerowicz–Poisson or LP –cohomology groups of
the Poisson manifold (Z, w).
0
Example 2.7.3. The LP-homology group HLP (Z, w) is the centre of the Poisson
algebra (O0 (Z), w). It consists of functions f ∈ T0 (Z) modulo constant functions
whose Hamiltonian vector fields ϑf = −w(f b ) ∈ Ker wb vanish. •
1
Example 2.7.4. The first LP-homology group HLP (Z, w) is the space of canonical
vector fields u for the Poisson bivector field w (i.e., Lu w = −w(u)
b = 0) modulo
b ), f ∈ T0 (Z). •
Hamiltonian vector fields −w(f
2
Example 2.7.5. The second LP-homology group HLP (Z, w) has the distinguished
element {w} whose representative is the Poisson bivector field w itself. We have
{w} = 0 if there is a vector field u on Z such that w = w(u)
b = −Lu w. •
∗
It is readily observed that, due to the property (2.7.10), HLP (Z, w) makes up a
graded commutative algebra with respect to the product
def
{ϑ} ∧ {υ} ={ϑ ∧ υ}.
It is also provided with the bracket

def
[{ϑ}, {υ}] ={[ϑ, υ]}
in accordance with the relation (1.2.13).
Theorem 2.7.3. [181]. The relation (2.7.11) induces a homomorphism of the

graded algebras
H ∗ (Z) → HLP
∗
(Z) (2.7.13)
of the De Rham cohomology and the LP-cohomology. This is an isomorphism if the

Poisson bivector field w comes from a symplectic structure on Z. 2
In case of a symplectic manifold, the isomorphism (2.7.13) and the isomorphism

(2.7.7) lead to the isomorphism
2m−i
Hican (Z) = HLP (Z).
In general, we have the following relationship between the LP-cohomology and the
canonical homology of a Poisson manifold.
Proposition 2.7.4. By virtue of the relation (2.7.12), the natural pairing (1.2.30)
induces the corresponding pairing
i
h, i : HLP (Z) × Hican (Z) → H0can (Z),
def
h{ϑ}, {φ}i ={hϑ, φi},
of vector spaces of the LP-cohomology and the canonical homology of a Poisson

manifold. 2
The constructions of the LP-cohomology and the canonical homology for Poisson
manifolds can be extended to the Jacobi ones [115].
Given a Jacobi manifold (Z; w, u), let us define the operator
w
d u : Tr (Z) → Tr+1 (Z),
def
w
d u = −[w, ϑ] + ru ∧ ϑ, ϑ ∈ Tr (Z),
which generalizes the contravariant exterior differential (2.7.8). This operator sat-
isfies the rules
Lu ◦ w
d u=w
d u ◦ Lu ,
w
d u2 (ϑ) = −Lu ϑ ∧ w, ϑ ∈ T∗ (Z), (2.7.14)
|ϑ|
w
d u(ϑ ∧ υ) = w
d u(ϑ) ∧ υ + (−1) ϑ ∧ w
d u(υ).
Its connection with the exterior differential is given by the relations
Lu (w] (φ)) = w] (Lu φ), (2.7.15)

] ] ]
w
d u(w (φ)) = −w (dφ) + w (ucφ) ∧ w. (2.7.16)
A glance at the expression (2.7.14) shows that the operator w

d u becomes nilpotent
on the subspaces of u-invariant multivectors
def
T r (Z) ={ϑ ∈ Tr (Z) : Lu ϑ = [u, ϑ] = 0}.
This allows us to introduce the cochain complex

w
cu w
cu w
cu w
cu
· · · −→ T r−1 (Z) −→ T r (Z) −→ T r+1 (Z) −→ · · · .
r
The cohomology groups HLP (Z, w) of this complex are called Lichnerowicz–Jacobi
or LJ–cohomology groups of the Jacobi manifold (Z; w, u). It is clear that, if u = 0
and (Z, w) is a Poisson manifold, the LJ-cohomology is precisely the LP-cohomology.
1
Example 2.7.6. The first LJ-cohomology group HLP (Z, w) has the distinguished
element whose representative is the vector field u, while the second LJ-cohomology
r
group HLP (Z, w) has the distinguished element whose representative is w. •
In the same way that the LP-cohomology groups are connected with the De Rham
cohomology groups, the LJ-cohomology groups are connected with the cohomology
groups of the subcomplex of basic forms of the De Rham complex.
An r-form φ on a Jacobi manifold (Z; w, u) is said to be basic if
ucφ = 0, Lu φ = 0.
It is easily seen that, if φ is a basic form, so is its exterior differential dφ. Hence,
basic forms constitute a subcomplex
d d d d
· · · −→ Or−1 r
B (Z) −→ O (Z)B −→ O
r+1
(Z)B −→ · · · (2.7.17)
of the De Rham complex.

On subspaces OrB (Z) ⊂ Or (Z) of basic forms, the relation (2.7.16) takes the
form similar to (2.7.11):
w
d u(w] (φ)) = −w] (dφ), φ ∈ OrB (Z). (2.7.18)
Moreover, we deduce from the relation (2.7.15) that, if φ is a basic r-form, then
w] (φ) ∈ T r (Z). In the same way as in Theorem 2.7.3, the equality (2.7.18) yields a
homomorphism
HB∗ (Z) → HLJ

∗
(Z)
¿from the cohomology of the complex (2.7.17), called the basic De Rham cohomol-
ogy, to the LJ-cohomology.
To introduce the canonical homology of a Jacobi manifold (Z; w, u), let us con-
sider the codifferential δw (2.7.3) defined by the bivector field w. It was proved in
[38] that, if φ is a basic form, so is δw (φ), and δw ◦ δw = 0. Then the chain complex
δ δ δ δ
w
· · · ←− Op−1 w p w p+1 w
B (Z) ←− OB (Z) ←− OB (Z) ←− · · ·
is defined. The homology of this complex is called the canonical homology of a

Jacobi manifold. If u = 0, this is exactly the canonical homology of a Poisson
manifold.
Similarly to Proposition 2.7.4, we have the following relationship between the
LJ-cohomology and the canonical homology of a Jacobi manifold.
Theorem 2.7.5. [115]. There is the pairing

i
h, i : HLP (Z) × Hican (Z) → H0can (Z),
def
h{ϑ}, {φ}i ={hϑ, φi},
of vector spaces of the LJ-cohomology and the canonical homology of a Jacobi

manifold. 2
The result follows from the fact that, if ϑ ∈ T ∗ (Z) and φ is a basic form, then
their pairing hϑ, φi (1.2.30) is a basic function, and from the relation
w
d u(ϑ)cφ = ϑcδw (φ) + (−1)r δw (ϑcφ), φ ∈ OrB (Z), ϑ ∈ T r−1 (Z).
Note that, bearing additionally in mind the map (2.2.4), one can apply the
construction of the LJ-cohomology and the canonical homology of a Jacobi manifold
to contact manifolds as the particular case of Jacobi manifolds. We refer to [38, 115]
for more details.
2.8 Appendix. More brackets

We will give a brief survey of several extensions of the Poisson bracket to multivectors
and differential forms and its generalization to the bracket of n > 2 functions.
In this Section, we will return to the notation [., .]SN for the Schouten–Nijenhuis
bracket.
Let (Z, w) be a Poisson manifold. With the nilpotent operation wb (2.7.8), one
can introduce the bracket
def
[ϑ, υ]w = −[w(ϑ),
b υ]SN
on an exterior algebra T∗ (Z) of multivector fields on Z. This bracket has the property
[ϑ, υ]w = −(−1)|ϑ||υ| [υ, ϑ]w + w([ϑ,

b υ]SN ),
2.8. APPENDIX. MORE BRACKETS 99
and is a graded skew-commutative Lie bracket on the quotient T∗ (Z)/w(T

b ∗ (Z)),
where the Lie degree of a multivector field ϑ is | ϑ | −1.
Remark 2.8.1. Recall that multivector fields on a manifold Z constitute a graded
Lie algebra with respect to the Schouten–Nijenhuis bracket [., .]SN0 (1.2.15), where
the Lie degree of a multivector field ϑ is | ϑ | −1 (see Remark 1.2.3). •
The Poisson bracket can be extended to exterior forms as follows.

If the Poisson bivector w on a manifold Z is non-degenerate, i.e., (Z, w) is a
symplectic manifold, there is the isomorphism w] (1.2.32) of graded algebras O∗ (Z)
and T∗ (Z). This isomorphism extends the Schouten–Nijenhuis bracket to the algebra
of exterior forms [133]:
{., .}w : Or (Z) × Os (Z) → Or+s−1 (Z), (2.8.1)

def
w] ({φ, σ}w ) =[w] φ, w] σ]SN , φ, σ ∈ O∗ (Z). (2.8.2)
In the general case of a Poisson manifold, we have the homomorphism w] (1.2.32)

which satisfies the property (2.7.11). Therefore, in order to construct the bracket
(2.8.1), let us consider the codifferential operator δw (2.7.3). Then the Schouten–
Nijenhuis bracket of exterior forms (2.8.1) is defined as
def
{φ, σ}w =(δw φ) ∧ σ + (−1)|φ| φ ∧ (δw σ) − δw (φ ∧ σ) (2.8.3)
[102, 181]. This bracket has the properties
{φ, σ}w = (−1)|φ||σ| {σ, φ}w ,

(−1)|φ|(|θ|−1) {φ, {σ, θ}w }w + (−1)|σ|(|φ|−1) {σ, {θ, φ}w }w +
(−1)|θ|(|σ|−1) {θ, {φ, σ}w }w = 0.
Example 2.8.2. In particular, the bracket (2.8.3) of 1-forms reads
{φ, σ}w = Lw] φ σ − Lw] σ φ − d(w(φ, σ)) = (2.8.4)

] ]
w φcdσ − w σcdφ + d(w(φ, σ)).
It provides O1 (Z) with a Lie algebra structure such that
w] : O1 (Z) → T (Z)
is a Lie algebra homomorphism (2.8.2). The relationship between the bracket (2.8.4)
and the Poisson bracket (2.3.1) is
{df, dg}w = d{f, g}.
•
Using the nilpotency properties (2.7.4) and (2.7.5) of the codifferential δw , one
can obtain the formula
d{φ, σ}w = −{dφ, σ}w − (−1)|φ| {φ, dσ}w .
Then the bracket

def
{φ, σ}d = −{dφ, σ}w (2.8.5)
on an algebra of exterior forms O∗ (Z) can be introduced [21, 133]. This is a graded
skew-commutative Lie bracket
{φ, σ}d = (−1)|φ||σ| {σ, φ}d
on the quotient O∗ (Z)/dO∗ (Z), where the Lie degree of an exterior form φ is | φ | −1.
Let us turn now to an n-ary generalization of Poisson and Jacobi brackets.
Definition 2.8.1. [84, 85]. A generalized almost Jacobi bracket of order n on a

manifold Z is said to be an n-linear map
n
{...} : × O0 (Z) → O0 (Z) (2.8.6)
which is
• skew-symmetric
{..., fi , ..., fj , ...} = −{..., fj , ..., fi , ...}
with respect to any pair of arguments,
• and is a first order linear differential operator on Z with respect to each argu-
ment, i.e.,
{g1 f1 , ...} = g1 {f1 , ...} + f1 {g1 , ...} − g1 f1 {1, ...}.

2.8. APPENDIX. MORE BRACKETS 101
If {...} is a generalized almost Jacobi bracket of order n, then there exist an

n-vector field w and an (n − 1)-vector field e on a manifold Z such that
n
(−1)i+1 fi e(df1 , ..., df
X
{f1 , ..., fn } = w(df1 , ..., dfn ) + c , ..., df )
i n (2.8.7)
i=1
and
{1, f1 , ..., fn−1 } = e(df1 , ..., dfn−1 ).
Conversely, any pair (w, e) of an n-vector field w and an (n − 1)-vector field e

defines the generalized almost Jacobi bracket (2.8.6) given by the relation (2.8.7). A
manifold Z provided with the bracket (2.8.7) is called a generalized almost Jacobi
manifold. If e = 0, we obtain a generalized almost Poisson manifold.
In order to reproduce a Jacobi structure in the case of n = 2, one should add the
integrability conditions, generalizing the Jacobi identity for the generalized almost
Jacobi structure. In fact, two different types of integrability conditions have been
suggested. They are
• the fundamental identity
{f1 , . . . , fn−1 , {g1 , . . . , gn }} = {{f1 , . . . , fn−1 , g1 }, g2 , . . . , gn } + (2.8.8)

{g1 , {f1 , . . . , fn−1 , g2 }, . . . , gn } +
· · · + {g1 , . . . , gn−1 , {f1 , . . . , fn−1 , gn }}
• and the generalized Jacobi identity
{f1 , . . . , fn−1 , {fn , . . . , f2n−1 }} + (2.8.9)

{f2 , . . . , fn , {fn+1 , . . . , f2n−1 , f1 }} + · · · = 0.
The bracket (2.8.7) satisfying the fundamental identity (2.8.8) is called a Nambu–
Jacobi bracket and (Z; w, e) is a Nambu–Jacobi manifold [85, 128]. If e = 0, the
bracket (2.8.7) is said to be a Nambu–Poisson bracket, and (Z, w) is a Nambu–
Poisson manifold [9, 84, 175].
The bracket (2.8.7) satisfying the generalized Jacobi identity (2.8.9) is called a
generalized Jacobi bracket, and (Z; w, e) is a generalized Jacobi manifold [84, 145].
If e = 0, the bracket (2.8.7) is said to be a generalized Poisson bracket, and (Z, w)

is a generalized Poisson manifold [8, 9, 84].
The relationship between Nambu–Jacobi manifolds and generalized Jacobi man-
ifolds can be shown directly by noticing that the generalized Jacobi identity (2.8.9)
is the full antisymmetrization of the fundamental identity (2.8.8). It follows that
every Nambu–Jacobi manifold is a generalized Jacobi manifold [9, 85].
For n even, the integrability conditions of a generalized Jacobi structure (w, e)
are given by the following assertion [145].
Proposition 2.8.2. A generalized almost Jacobi structure (w, e) of even order

n = 2p is a generalized Jacobi structure if and only if the relations
[w, e]SN = 0, [w, w]SN = 2(2p − 1)w ∧ e,
generalizing the conditions (2.1.2), hold good. 2
At the same time, the integrability conditions of a Nambu–Jacobi structure (w, e)

in particular imply that
• the multivector field w is a Nambu–Poisson structure of order n,
• the multivector field e is a Nambu–Poisson structure of order n − 1 [128],
• and every Nambu–Poisson multivector w can be written as an exterior product

of vector fields [79].
Example 2.8.3. Let us consider the product Z = R × T ∗ M with the holonomic

coordinates (t, q i , pi ) as in Examples 2.2.1, 2.2.3 and 2.3.7. It is equipped with both
the Jacobi structure given by the pair (w1 , E1 ) (2.2.9) and the Poisson structure w2
(2.3.9). Then Z is also provided with the generalized Jacobi structure of order 4
given by the multivector fields
w = w1 ∧ w2 , e = E ∧ w2
(see [85] for details). •

2.9. APPENDIX. MULTISYMPLECTIC STRUCTURES 103
2.9 Appendix. Multisymplectic structures

We will give a brief exposition of multisymplectic and vector-valued generalizations
of a symplectic structure.
Similarly to that a Poisson bivector field is replaced by a multivector one for a
generalized Poisson structure or a Nambu–Poisson structure, one can consider an
n-ary generalization of a symplectic form. This is a multisymplectic form.
Given an m-dimensional manifold M with coordinates (z λ ), let us consider the
fibre bundle
k
ζ : ∧ T ∗M → M
whose sections are exterior k-forms on M . This fibre bundle is equipped with the
holonomic coordinates (z λ , pΛ ), where Λ = (λ1 < . . . < λk ) are multi-indices of the
k
length | Λ |= k. The manifold ∧ T ∗ M is provided with the canonical exterior k-form
Θ defined by the relation
k k
un c...u1 cΘ(p) = T ζ(un )c...T ζ(u1 )cp, p ∈ ∧ T ∗ M, ui ∈ Tp (∧ T ∗ M ).
Its coordinate expression is
pλ1 ...λk dz λ1 ∧ · · · ∧ dz λk ,
X
Θ= (2.9.1)
Λ
where the sum is over all multi-indices Λ of the length k.

The exterior differential dΘ of the form (2.9.1) is the (k + 1)-symplectic form
dpλ1 ...λk ∧ dz λ1 ∧ · · · ∧ dz λk
X
ΩM = dΘ = (2.9.2)
Λ
which belongs to the class of multisymplectic forms in the sense of Martin [131]. If
k = 1, the form ΩM (2.9.2) is the familiar symplectic form on the cotangent bundle
T ∗M .
Example 2.9.1. For instance, let Y → X be a fibre bundle over a (1 < n)-
dimensional base X. Let us consider the canonical form Θ (2.9.1) on the exterior
n
product ∧ T ∗ Y . The homogeneous Legendre bundle over Y is said to be the fibre
bundle
n−1
ZY = T ∗ Y ∧ ( ∧ T ∗ X), (2.9.3)
equipped with the holonomic coordinates (xλ , y i , pλi , p), where the coordinate p has
the transformation law
∂xε ∂y j ∂y 0 i
! !
0
p = det p − 0 i µ pµj .
∂x0 ν ∂y ∂x
The canonical bundle monomorphism over Y
n−1 n
iZ : T ∗ Y ∧ ( ∧ T ∗ X) → ∧ T ∗ Y
yields the pull-back form
Ξ = i∗Z Θ = pω + pλi dy i ∧ ωλ (2.9.4)
on the fibre bundle ZY , where we use the notation (1.2.21). Its exterior differential
ΩZ = dp ∧ ω + dpλi ∧ dy i ∧ ωλ (2.9.5)
is also called the multisymplectic form [63]. In particular, if Y → R, the multisym-

plectic form (2.9.5) leads to the canonical symplectic form (2.4.4) on the cotangent
bundle ZY = T ∗ Y . •
Let us touch on another generalization of a symplectic structure when an Rn -

valued 2-form is considered [53, 142]. Let X be an n-dimensional manifold and
LX → X the principal frame bundle coordinated by (xλ , sµ a ). The frame bundle
LX is provided with the canonical Rn -valued 1-form θLX (1.5.13) which reads
θLX = sa µ dxµ ⊗ ta ,
where {ta } is a basis for Rn . Let us take its exterior differential
dθLX = dsa µ ∧ dxµ ⊗ ta ,
called the n-symplectic form. It is readily observed that this form is non-degenerate,
i.e., given a vector field
u = uµ ∂µ + uaµ ∂aµ
on LX, the equality ucdθLX = 0 holds if and only if u = 0. By analogy with the
symplectic case, one can write
uf cdθLX = −df, (2.9.6)

2.9. APPENDIX. MULTISYMPLECTIC STRUCTURES 105
where f is an Rn -valued function on the frame bundle LX. There is the class
of functions (e.g., which are constant on the fibres of LX) such that vector fields
satisfying the equation (2.9.6) exist. Then their Jacobi bracket
{f, g} = uf cdg
is defined, and it also belongs to the above-mentioned class of functions.

Let now Y → X be an arbitrary fibre bundle. The Legendre bundle over Y is
said to be the fibre bundle
n n−1
Π = ∧ T ∗X ⊗ T X ⊗ V ∗Y ∼
= V ∗ Y ∧ ( ∧ T ∗ X), (2.9.7)
Y Y
equipped with the holonomic coordinates (xλ , y i , pλi ) possessing the transition func-
tions
∂y j ∂x0 λ ∂xα
!
λ
p0 i = 0 i µ det pµj . (2.9.8)
∂y ∂x ∂x0 β
It is provided with the T X-valued form
Λ = dpλi ∧ dy i ∧ ω ⊗ ∂λ , (2.9.9)
called the polysymplectic form. The Legendre bundle (2.9.7) plays the role of a
finite-dimensional phase space in the Hamiltonian formulation of classical field the-
ory [28, 56, 57, 73, 96, 158, 159].
In the case of X = R, the polysymplectic Hamiltonian formalism leads us to the
Hamiltonian formulation of time-dependent mechanics (see Chapter 5).
The main peculiarity of polysymplectic formalism is that Hamiltonian connec-
tions on the Legendre bundle Π → X play a role similar to Hamiltonian vector fields
on a symplectic manifold. A connection
µ i
γ = dxλ ⊗ (∂λ + γλi ∂i + γλi ∂µ )
on the fibre bundle Π → X is said to be locally Hamiltonian if the exterior form

γcΛ is closed (see Section 3.8 for details).
Remark 2.9.2. Note that generalization of the Poisson bracket for polysymplectic
manifolds meets the difficulty (see [91, 92] and references therein). Nevertheless, in
the particular case of an affine bundle Y → X whose base X is provided with a

non-degenerate metric g, the bracket of horizontal 1-forms on Π → X,
!
∂φα ∂σβ ∂σβ ∂φα q
{φ, σ} = g αβ − µ | g |dxµ ,
∂pµi ∂y i ∂pi ∂y i
φ = φλ (p)dxλ , σ = σλ (p)dxλ ,
can be globally defined. •

Chapter 3
Hamiltonian systems
The theory of Hamiltonian systems is a vast subject which can be studied from many
different viewpoints. This Chapter provides a brief exposition of different types of
Hamiltonian systems on Poisson, symplectic and presymplectic manifolds in conser-
vative mechanics. We leave aside many interesting constructions of a Hamiltonian
conservative mechanics and its straightforward extension to the direct product
[time] × [autonomous phase space]
that cannot be applied to time-dependent mechanics because a Hamiltonian is not

a function under time-dependent transformations. Nevertheless, any object defined
on a phase space V of conservative mechanics or on the product R × Z has the
counterpart on a phase space
ψ:W ∼
=R×Z (3.0.1)
of time-dependent mechanics, where different trivializations ψ (3.0.1) correspond to

different reference frames.
For instance, there is one-to-one correspondence between the connections on a
fibre bundle W → R (3.0.1) and the time-dependent vector fields
R × Z →TZ
Z
because of the diffeomorphism
J 1ψ : J 1W ∼
= J 1 (R × Z) = R × T Z.
107
108 CHAPTER 3. HAMILTONIAN SYSTEMS
However, this correspondence is not canonical, but depends on a trivialization

(3.0.1).
Note that Hamiltonian relativistic mechanics can be seen as an autonomous
Hamiltonian system on the hyperboloid of relativistic momenta, i.e., it exemplifies a
Dirac constraint system (see Section 6.3). In this Chapter, we pay special attention
to constraint systems, including symplectic Hamiltonian systems, Dirac Hamiltonian
systems, and Dirac constraint systems, seen as particular presymplectic Hamiltonian
systems.
3.1 Dynamic equations

There are several types of equations in conservative mechanics. These are first
and second order dynamic equations, Hamilton equations with respect to Poisson,
symplectic or presymplectic structures, and Lagrange equations. The relationship
between them is the following.
• The dynamic equations, by definition, are differential equations which can be

algebraically solved for the highest order derivatives.
• The second order dynamic equations on a manifold M , by definition, are

particular first order dynamic equations on the tangent bundle T M of M .
• The first order dynamic equations on a manifold Z can be represented as

Lagrange equations for Lagrangians on the tangent bundle T Z of Z, but they
are not Hamilton equations on Z in general.
• The Hamilton equations are not necessarily first order dynamic equations.
• The Lagrange equations on a manifold M are derived as particular Hamilton

equations on the tangent bundle T M , but they are not necessarily second
order dynamic equations and differential equations in any strict sense.
• The second order dynamic equations on a manifold M fail to be represented as

Lagrange equations for Lagrangians on the tangent bundle T M or as Hamilton
equations on T M in general.
This Section is devoted to the notions of first and second order dynamic equations
on a manifold, called autonomous dynamic equations.
3.1. DYNAMIC EQUATIONS 109
Definition 3.1.1. Let Z, dim Z > 1, be a manifold, coordinated by (z λ ), and u a

vector field on Z. The closed subbundle u(Z) of the tangent bundle T Z, given by
the coordinate relations
ż λ = uλ (z), (3.1.1)
is said to be an (autonomous) first order dynamic equation on a manifold Z. This

is a system of first order differential equations on the fibre bundle R × Z → R in
accordance with Definition 1.3.3 (see Example 5.2.1). 2
By solutions of the first order dynamic equation (3.1.1) are meant integral curves
of the vector field u.
Remark 3.1.1. Generalizing Definition 3.1.1, we will say that an (autonomous)
first order differential equation on a manifold Z is a closed submanifold E of the
tangent bundle T Z of Z. A solution of this differential equation is an integral curve
of a vector field on a submanifold N ⊂ Z, which takes its values into T N ∩ E. •
In this Chapter, we will deal with the first order dynamic equations (3.1.1) when
vector fields u are Hamiltonian vector fields for Poisson, symplectic or presymplectic
structure on a manifold Z, treated as a momentum phase space of conservative
mechanics. However, they are not equivalent to the Hamilton equations in general.
The Hamilton equations with respect to a Poisson structure, by definition, are first
order dynamic equations, and so are the Hamilton equations with respect to a
symplectic structure. This is not the case of Hamilton equations with respect to a
presymplectic structure.
Let us bear in mind the well-known inverse problem in conservative Hamiltonian
mechanics, which consists in trying to represent a first order dynamic equation on a
manifold Z as Hamilton equations with respect to some Poisson structure on Z [32]
(see the next Section). The important motivation of the inverse problem is that,
with a Poisson structure, one may find integrals of motion of a dynamic equation
and also quantize it.
Remark 3.1.2. As in the particular case of the general result [76, 81, 156], let us
point out that every first order dynamic equation on a manifold Z can be represented
as Lagrange equations for a Lagrangian on the tangent bundle T Z, but not in an
explicit form. •
Let M be a manifold, coordinated by (q i ). If it is a configuration space of

conservative mechanics, the corresponding velocity phase space is the tangent bundle
T M of M , while the cotangent bundle T ∗ M plays the role of the momentum phase
space.
Definition 3.1.2. An autonomous second order dynamic equation on a manifold

M is defined as a first order dynamic equation (3.1.1) on the tangent bundle T M ,
which is associated with a holonomic vector field
Ξ = q̇ i ∂i + Ξi (q j , q̇ j )∂˙i (3.1.2)
on T M . This vector field, by definition, obeys the condition
J(Ξ) = uT M ,
where J is the endomorphism (1.2.39) and uT M is the Liouville vector field (1.2.6)
on T M (see the notation (1.2.4)). 2
The vector field (3.1.2) is called a dynamic vector field. In the literature, it is
often termed a second order dynamic equation.
Let the double tangent bundle T T M be provided with the coordinates
(q i , q̇ i , q̇i , q̈ i ). (3.1.3)
With respect to these coordinates, the second order dynamic equation defined by
the holonomic vector field Ξ (3.1.2) reads
q̇i = q̇ i , q̈ i = Ξi (q j , q̇ j ). (3.1.4)
By solutions of the second order dynamic equation (3.1.4) are meant the curves
c : () → M in a manifold M whose tangent prolongations ċ : () → T M are
integral curves of the holonomic vector field Ξ or, equivalently, whose second order
tangent prolongations c̈ live in the subbundle (3.1.4). They satisfy the second order
differential equations
c̈i (t) = Ξ(cj (t), ċj (t)).
A particular case of second order dynamic equations on a manifold M is that of

geodesic equations on the tangent bundle T M .
Given a connection
K = dq j ⊗ (∂j + Kji ∂˙i )
on the tangent bundle T M → M , let

c : TM ×TM → TTM
K (3.1.5)
M
be the corresponding linear bundle morphism over T M which splits the exact se-
quence
0 −→ VM T M ,→ T T M −→ T M × T M −→ 0.
M
Definition 3.1.3. A geodesic equation on T M with respect to the connection K

is defined as the image
q̇i = q̇ i , q̈ i = Kji q̇ j (3.1.6)
of the morphism (3.1.5) restricted to the diagonal T M ⊂ T M × T M . 2
By a solution of a geodesic equation on T M is meant a geodesic curve c in M ,

whose tangent prolongation ċ is an integral section (a geodesic vector field) over
c ⊂ M for the connection K. The geodesic equation (3.1.6) can be written in the
form
q̇ j ∂j q̇ i = Kji q̇ j ,
where q̇ i (q j ) is a geodesic vector field (which exists at least on a geodesic curve),

while q̇ i ∂i is the formal operator of differentiation (along a curve).
It is readily observed that the morphism K c |
T M is a holonomic vector field
on T M . It follows that any geodesic equation (3.1.5) on T M is a second order
equation on M . The converse is not true in general. Nevertheless, we have the
following theorem.
Theorem 3.1.4. [136]. Every second order dynamic equation (3.1.4) on a manifold
M defines a connection KΞ on the tangent bundle T M → M whose components are
1
Kji = ∂˙j Ξi . (3.1.7)
2
2
However, the second order dynamic equation (3.1.4) fails to be a geodesic equa-
tion with respect to the connection (3.1.7) in general. In particular, the geodesic
equation (3.1.6) with respect to a connection K determines the connection (3.1.7)
on T M → M which does not necessarily coincide with K. A second order equation
Ξ on M is a geodesic equation for the connection (3.1.7) if and only if Ξ is a spray,
i.e.,
[uT M , Ξ] = Ξ,
where uT M is the Liouville vector field (2.4.9) on T M , i.e.,
Ξi = aij (q k )q̇ i q̇ j .
In Section 4.3, we will improve Theorem 3.1.4 (see Proposition 4.3.3 below).
An extensive literature is devoted to the inverse problem for second order dy-
namic equations in Lagrangian mechanics. In Section 4.9, we will touch on the
following two aspects of this problem in time-dependent mechanics:
• the condition for an Euler–Lagrange type operator to be an Euler–Lagrange
one as the particular inverse problem in field theory,
• and the corresponding relation between Newtonian and Lagrangian systems.

We refer the reader to [136] for a survey on the inverse problem in conservative
Lagrangian mechanics. Recall that, in accordance with Remark 3.1.2, a second
order dynamic equation on a manifold M , being a first order dynamic equation on
T M , can always be thought of as Lagrange equations for a Lagrangian on the double
tangent bundle T T M of M .
In this Chapter, we will give two examples of the inverse problem for second
order dynamic equations in conservative Hamiltonian mechanics.
• As is well known, the tangent bundle T M fails to possess any canonical Pois-
son, symplectic, or presymplectic structures. Nevertheless, there are proce-
dures to study a second order dynamic equation as the Hamilton ones with
respect to some Poisson structure on T M (see Remark 3.2.3 below).
• Lagrange equations for a Lagrangian L on the tangent bundle T M can be also

seen as Hamilton equations with respect to the presymplectic structure (or
the symplectic structure, if L is regular), defined by this Lagrangian on T M
(see Examples 3.3.2 and 3.4.2 below).
3.2. POISSON HAMILTONIAN SYSTEMS 113
3.2 Poisson Hamiltonian systems

Let (Z, w) be a k-dimensional Poisson manifold and H a real function on Z.
Definition 3.2.1. A Poisson Hamiltonian system (w, H) on a manifold Z for an

(autonomous) Hamiltonian H with respect to the Poisson structure w is the set
{v ∈ Tz Z : v − w] (dH)(z) = 0}.
[
SH = (3.2.1)
z∈Z
A solution of the Hamiltonian system (3.2.1) is a vector field ϑ on a submanifold

N ⊂ Z, which takes its values into T N ∩ SH .
It is readily observed that the Poisson Hamiltonian system SH (3.2.1) has a
unique solution which is the Hamiltonian vector field
ϑH = w] (dH) (3.2.2)
for the Hamiltonian H, that passes through any point of SH . Hence, SH is a first
order equation in accordance with Definition 3.1.1. It is called the Hamilton equation
for the Hamiltonian H with respect to the Poisson structure w.
With respect to the local canonical coordinates (q i , pi , z A ) (2.3.7) for the Poisson
structure w, the Hamilton equation (3.2.1) reads
q̇ i = ∂ i H, ṗi = −∂i H, ż A = 0,
while the vector field ϑH (3.2.2) takes the form
ϑH = ∂ i H∂i − ∂i H∂ i .
Its integral curves r : () → Z satisfy the equations
ṙi = ∂ i H ◦ r, ṙi = −∂i H ◦ r, ṙA = 0. (3.2.3)
Recall that, given a dynamic equation (3.1.1) associated with a vector field u on
a manifold Z, the Lie derivative Lu of a function f on Z along u can be treated as
the evolution of f along solutions of this dynamic equation. If
Lu f = 0, (3.2.4)
the function f is constant on any solution of the above-mentioned dynamic equation

or equivalently on integral curves of u. A function f 6= const. is called a first integral
of motion if it obeys the equation (3.2.4). ¿From the geometric viewpoint, this means
that the vector field u is tangent to the level surfaces of the function f at its regular
points, where df 6= 0.
If a dynamic equation is a Hamilton equation with respect to a Poisson structure,
one can find its first integrals of motion as functions in involution with a Hamiltonian
as follows.
Let H be a Hamiltonian of a Poisson Hamiltonian system (w, H). The Lie
derivative of an arbitrary real function f on Z along the Hamiltonian vector field
ϑH for H reads
LϑH f = ϑH cdf = {H, f }. (3.2.5)
The equality (3.2.5) is called the evolution equation. Substituting solutions r of the
Hamilton equations (3.2.3) in (3.2.5), we obtain the evolution
dt (f ◦ r) = {H, f } ◦ r
of a function f along the integral curves of the Hamiltonian vector field ϑH . In

particular, if
{H, f } = 0,
the function f is a first integral of motion of the Hamiltonian system (w, H).
Example 3.2.1. A Hamiltonian H itself is obviously a first integral of motion of
the Hamiltonian system (w, H). •
It is easily seen that, if f and f 0 are first integrals of motion of a Poisson Ha-
miltonian system, so is their Poisson bracket {f, f 0 }. It follows that first integrals
of motions constitute a Lie algebra.
A vector field v on Z is called an infinitesimal symmetry of a Hamiltonian H if
Lv H = 0.
In particular, if f is a first integral of motion of a Poisson Hamiltonian system

(w, H), the Hamiltonian vector field ϑf for f is an infinitesimal symmetry of the
Hamiltonian H, i.e.,
Lϑf H = ϑf cdH = −{H, f } = 0.

3.2. POISSON HAMILTONIAN SYSTEMS 115
Let us turn now to the inverse problem in conservative Hamiltonian mechanics,

mentioned in Remark 2.3.5. It has the following solution.
Proposition 3.2.2. For any first order dynamic equation (3.1.1) on a manifold Z,
there exists locally a generating Poisson Hamiltonian system (w, H) on Z such that
(3.1.1) is locally a Hamilton equation. 2
Proof. If u = 0, the statement is obvious. Let u(z) 6= 0 at a point z ∈ Z. There

exists a coordinate system (q 1 , . . . , q k ) on an open neighbourhood of z such that
∂ ∂
u= and [u, v] = 0, v= .
∂q 1 ∂q 2
Then u ∧ v is a local Poisson bivector field of rank 2 (see Example 2.3.2). It is
readily observed that u is locally a Hamiltonian vector field for the function q 2 (z λ )
with respect to the Poisson structure u ∧ v. QED
Proposition 3.2.2 leads us to the conditions for the inverse problem of a first
order dynamic equation (3.1.1) to have a global solution.
Proposition 3.2.3. [59, 82]. Let u be a nowhere vanishing vector field on a

manifold Z. If there exist
• a nowhere vanishing vector field v on Z such that [u, v] = 0 everywhere on Z,
• and a function f on Z such that Lu f = 0 and Lv f = 1,
then u is the Hamiltonian vector field for the function f with respect to the Poisson
structure u ∧ v. 2
Example 3.2.2. Every first order dynamic equation (3.1.1) on a manifold Z is a

Hamilton equation on the product R×Z with coordinates (t, z λ ) for the Hamiltonian
H = t and with respect to the Poisson structure w = u ∧ ∂t . We refer the reader to
[59] for the physically interesting examples. •
Remark 3.2.3. The method based on Propositions 3.2.2 and 3.2.3 can be applied
to study the existence of a generating Poisson structure for a second order dynamic
equation (3.1.4) on a manifold M defined by a holonomic vector field Ξ (3.1.2)
on the tangent bundle T M . In particular, for any second order dynamic equation
(3.1.4) on M , there exists locally a generating Poisson Hamiltonian system (w, H)

on T M such that the associated holonomic vector field Ξ is the Hamiltonian vector
field for the Hamiltonian H with respect to the Poisson structure w. The necessary
condition for a Poisson structure w on the tangent bundle T M to be a generating
Poisson structure for a holonomic vector field Ξ on T M is
LΞ (w) = [Ξ, w] = 0.
We refer the reader to [182] for a detailed analysis of this condition. •
3.3 Symplectic Hamiltonian systems

Let (Z, Ω) be a symplectic manifold. The notion of a symplectic Hamiltonian system
is a repetition of Definition 3.2.1 (see [127]).
Definition 3.3.1. A symplectic Hamiltonian system (Ω, H) on a manifold Z for a

Hamiltonian H with respect to the symplectic structure Ω is the set
def [
SH = {v ∈ Tz Z : vcΩ + dH(z) = 0}. (3.3.1)
z∈Z
As in the general case of Poisson systems, the symplectic Hamiltonian system

(Ω, H) has a unique solution which is the Hamiltonian vector field ϑH such that
ϑH cΩ = −dH
for the Hamiltonian H, that passes through any point of SH . Hence, SH is a first
order dynamic equation, called the Hamilton equation.
With respect to the local canonical coordinates (q i , pi ) for the symplectic struc-
ture Ω, the Hamilton equation (3.3.1) reads
q̇ i = ∂ i H, ṗi = −∂i H. (3.3.2)
In the symplectic case, there is one-to-one correspondence between the Hamil-

tonian systems and the Hamiltonian vector fields on a symplectic manifold (Z, Ω).
The first integrals of motion of a symplectic Hamiltonian system and infinitesimal
symmetries of a Hamiltonian are defined as a repetition of those for a Poisson
Hamiltonian system.
3.3. SYMPLECTIC HAMILTONIAN SYSTEMS 117
Recall that a Hamiltonian system on a 2m-dimensional symplectic manifold is

called completely integrable if there exist m first integrals of motion which are
pairwise in involution and whose differentials are linearly independent on a dense
open subset of that manifold.
Example 3.3.1. Let Z = T ∗ M be a symplectic manifold provided with the canon-
ical symplectic form Ω (2.4.2). In accordance with Example 2.4.7, any Hamiltonian
system (Ω, H) on the symplectic manifold (T ∗ M, Ω) is the Lagrangian submanifold
SH = ϑH (T ∗ M ) (2.4.10) of the symplectic manifold T T ∗ M equipped with the sym-
plectic form Ω
e (2.4.9). The generating function of this submanifold, given by the
coordinate relations
q̇ i = ∂ i H, ṗi = −∂i H, (3.3.3)
is the Hamiltonian H(q i , pi ). Any Hamiltonian H on the symplectic manifold T ∗ M

defines the fibred morphism over M
def ∗ ∗ ∗
H
c =π
∗T M ◦ α ◦ ϑH : T M → T T M → T T M → T M, (3.3.4)
q̇ i ◦ H
c = ∂ i H,
called the Hamiltonian map. •
Example 3.3.2. Any non-degenerate autonomous Lagrangian system can be seen

as a symplectic Hamiltonian system as follows.
An autonomous Lagrangian is defined as a real function L on the tangent bundle
T M of an event space M . Throughout this Chapter, we will use the compact
notation
πi = ∂˙i L, πji = ∂˙j ∂˙i L.
Using the tangent-valued form φJ (1.2.40) on the tangent bundle T M , let us intro-
duce the Poincaré–Cartan 1-form
def
HL = φJ cdL = πi dq i .
Its differential
ΩL = dπi ∧ dq i = πji dq̇ j ∧ dq i + ∂j πi dq j ∧ dq i (3.3.5)
is a symplectic form on T M if and only if the Lagrangian L is regular, i.e., its

Hessian det (πji ) is nowhere vanishing.
If a Lagrangian L is regular, one can consider the symplectic Hamiltonian system

(ΩL , EL ) with respect to the symplectic form ΩL for the Hamiltonian
EL = uT M cdL − L = q̇ i πi − L, (3.3.6)
called the energy function, where we make use of the Liouville vector field uT M
(1.2.6) on T M . Then we come to the Hamilton equation for the Hamiltonian vector
field ϑL such that
ϑL cΩL = −dEL . (3.3.7)
With respect to the coordinates (3.1.3) on the double tangent bundle T T M , the
equation (3.3.7), called the Cartan equation for the Lagrangian L, reads
πij (q̇i − q̇ i ) = 0, (3.3.8)

j j j j
∂i L − q̈ πji − q̇ ∂j πi + (q̇ − q̇ )∂i πj = 0. (3.3.9)
Since the Lagrangian L is regular, the equation (3.3.8) reduces to the equality
q̇i = q̇ i , (3.3.10)
while the equation (3.3.9) comes to the Lagrange equations
∂i L − q̈ j πji − q̇ j ∂j πi = 0 (3.3.11)
for the Lagrangian L.

The condition (3.3.10) means that, in the case of a regular Lagrangian L, the
Hamiltonian vector field ϑL is holonomic. It defines the second order dynamic
equation on T M which is equivalent to the Lagrange equations (3.3.11).
Let us consider the fibre bundles T T ∗ M and T ∗ T M provided with the coor-
dinates (q i , pi , q̇ i , ṗi ) and (q i , q̇ i , q̇i , q̈i ), respectively, and recall the isomorphism α
(1.1.10):
α : T T ∗M ∼
= T ∗ T M, pi ←→ q̈i , ṗi ←→ q̇i .
The symplectic form ΩL (3.3.5) yields the fibred morphism (1.2.33) over T M :
Ω[L : T T M → T ∗ T M,
q̇i ◦ Ω[L = −q̈ j πji + q̇j (∂i πj − ∂j πi ), q̈i ◦ Ω[L = q̇j πij ,
3.3. SYMPLECTIC HAMILTONIAN SYSTEMS 119
and the corresponding fibred morphism
α−1 ◦ Ω[L : T T M → T ∗ T M → T T ∗ M, (3.3.12)

ṗi ◦ α−1 Ω[L = −q̈ j πji + q̇j (∂i πj − ∂j πi ), pi ◦ α−1 ◦ Ω[L = q̇j πij .

−1
e = −Ω[∗ Ω ∗
Ω L L T T M = −(α ◦ Ω[L )∗ Ω,
e
e = −[d(−q̈ j π + q̇j (∂ π − ∂ π )) ∧ dq i + d(q̇j π ) ∧ dq̇ i )],
Ω L ij i j j i ij
where Ω e is the tangent lift (1.2.25) of Ω onto T T M and by Ω ∗

L L T T M is meant the
∗
canonical symplectic form on the cotangent bundle T T M of T M .
It is readily observed that the system of Lagrange equations (3.3.11) is a Lagran-
gian submanifold of the symplectic manifold (T T M, Ωe ). Accordingly, the image of
L
the Lagrange equations (3.3.11) into T T ∗ M by the morphism α−1 ◦ Ω[L (3.3.12) is
the Lagrangian submanifold
pi = q̇ j πji , ṗi = −∂i L + q̇ j ∂i πj
of the symplectic manifold (T T ∗ M, Ω).

e The generating function of this Lagrangian
submanifold is the energy function EL (3.3.6).
Every Lagrangian L on the configuration space T M yields the fibred morphism
over M
def
Lb = πT ∗ M ◦ α−1 ◦ dL : T M → T ∗ T M → T T ∗ M → T ∗ M ; (3.3.13)
pi = ∂˙i L, (3.3.14)
called the Legendre map. A Lagrangian L is regular if and only if the corresponding
Legendre map Lb is a local diffeomorphism. A Lagrangian L is called hyperregular
if the Legendre map Lb is a diffeomorphism.
The tangent map T Lb to the Legendre map Lb is the fibred morphism
T Lb : T T M → T T ∗ M,
pi ◦ T Lb = πi , ṗi = πij q̈ j + ∂j πi q̇j ,
over M . Then the image of the Lagrange equations (3.3.11) into T T ∗ M by T Lb is

another Lagrangian submanifold
pi = πi , ṗi = ∂i (3.3.15)
of the symplectic manifold (T T ∗ M, Ω).

e This is exactly the Lagrangian submanifold
(2.4.11), defined by the differential dL (see Example 2.4.7). Its generating function
is L.
One can find the conditions of a hyperregular Lagrangian L on the tangent
bundle T M and a Hamiltonian H on the cotangent bundle T ∗ M to define the same
Lagrangian submanifolds (3.3.15) and (3.3.3) of the symplectic manifold (T T ∗ M, Ω).
e
This takes place if and only if
uT ∗ M cdH − H = L ◦ H,
c (3.3.16)
pi ∂ i H − H = L(q i , ∂ i H),
where uT ∗ M is the Liouville vector field (1.2.5) on T ∗ M → M , or equivalently
uT M cdL − L = H ◦ L, b
q̇ i ∂˙i L − L = H(q i , ∂˙i L),
where uT M is the Liouville vector field (1.2.6) on T M → M . Taking the differential

of the equation (3.3.16), we find that
H b−1 ,
c= L
(∂i L) ◦ H
c = −∂ H.
i
Then we have the commutative diagram
L ϑ
T M −→ TTM
L
b TL
b
? ?
∗ ∗
T M −→ T T M
ϑH
The constructions in Examples 3.3.1, 3.3.2 lead to description of Hamiltonian and

Lagrangian dynamics of autonomous systems in terms of Lagrangian submanifolds
and to constructing unified Lagrangian-Hamiltonian formalism on the phase space
T T ∗ M [111, 176, 177] (see Example 3.4.2 below).
3.4. PRESYMPLECTIC HAMILTONIAN SYSTEMS 121
3.4 Presymplectic Hamiltonian systems

The notion of a Hamiltonian system is naturally extended to presymplectic manifolds
(see [12, 60, 138, 172]).
Definition 3.4.1. Let Z be a k-dimensional manifold equipped with a presym-

plectic form Ω, and let H be a real function on Z. A presymplectic Hamiltonian
system for a Hamiltonian H, like (3.3.1), is the set
def [
SH = {v ∈ Tz Z : vcΩ + dH(z) = 0}. (3.4.1)
z∈Z
A solution of the presymplectic Hamiltonian system (3.4.1) is a Hamiltonian

vector field ϑH for H which lives in SH . It satisfies the equality
ϑH cΩ = −dH. (3.4.2)
In comparison with the symplectic case, the form Ω is degenerate, the set (3.4.1)
fails to be a submanifold in general, and a solution of the equation (3.4.2) does not
necessarily exist everywhere on the manifold Z.
Note that, for any point z ∈ Z, the fibre
{v ∈ Tz Z : vcΩ + dH(z) = 0} (3.4.3)
of the set SH over z is an affine space modelled over the vector space
Ker z Ω = {v ∈ Tz Z : ucvcΩ = 0, ∀u ∈ Tz Z}
which is the fibre over z of Ker Ω (1.2.34) of the presymplectic form Ω. The fibre
(3.4.3) may be empty.
Proposition 3.4.2. The equation
vcΩ + dH(z) = 0, v ∈ Tz Z, (3.4.4)
has a solution only at the points of the subset
N2 = {z ∈ Z : ucdH(z) = 0, ∀u ∈ Ker z Ω} ⊂ Z, (3.4.5)
i.e., where Ker z Ω ⊂ Ker z dH [60, 138]. 2

Proof. Let a vector v ∈ Tz Z, satisfying the equation (3.4.4), exist. Then, the
contraction of the right-hand side of the equation (3.4.4) with an arbitrary element
u ∈ Ker z Ω leads to the equality ucdH(z) = 0. In order to prove the converse
assertion, it suffices to show that
dH(z) ∈ Im Ω[ ,
that results from the inclusions
dH(z) ∈ Ann(Ker dH(z)) ⊂ Ann(Ker z Ω) = Im Ω[ .
QED
It follows that, if Z 6= N2 a presymplectic Hamiltonian system is not a differential

equation on Z. If N2 = Z, it is a differential equation, but not a dynamic equation.
¿From now on, let us suppose that a presymplectic form Ω is of constant rank,
and that N2 (3.4.5) is a submanifold of Z, but not necessarily connected. Then, by
virtue of Proposition 1.1.4, the kernel Ker Ω is a closed vector subbundle (moreover,
an involutive distribution) of the tangent bundle T Z. Accordingly, SH |N2 is an affine
bundle over N2 . It admits a global section, but this section does not live necessarily
in T N2 ⊂ T Z, i.e., is not a vector field on the submanifold N2 in general.
To obtain a solution of the presymplectic Hamiltonian system (3.4.1), let us
consider a submanifold N ⊂ N2 ⊂ Z such that
SH |N ∩ Tz N 6= ∅, ∀z ∈ N,
or equivalently
dH(z) ∈ Ω[ (T N ), ∀z ∈ N.
If the morphism Ω[ |T N is of constant rank, then SH ∩ T N → N is an affine bundle

modelled over the vector bundle Ker Ω ∩ T N → N which may be 0. Sections of
the affine bundle SH ∩ T N → N are Hamiltonian vector fields ϑH (3.4.2) for the
Hamiltonian H on a submanifold N . If such a submanifold exists, it may be found
by the following well-known algorithm [12].
Let us take the intersection
SH |N2 ∩ T N2
and project it to Z. We obtain the subset

N3 = πZ (SH |N2 ∩ T N2 ) ⊂ Z.
If N3 is a submanifold, let us take the intersection
SH |N3 ∩ T N3 .
Its projection to Z gives a subset N4 ⊂ Z, and so on. Since a manifold Z is finite-
dimensional, the procedure is stopped after a finite number of steps by one of the
following results.
• There is i ≥ 2 such that a set Ni is empty. This means that a presymplectic
Hamiltonian system has no solution.
• A set Ni , i ≥ 2, fails to be a submanifold. It follows that a solution may exist,

but not everywhere on Ni .
• If Ni+1 = Ni for some i ≥ 2, this is the desired submanifold N . If the morphism

Ω[ |T N is of constant rank, a solution of the presymplectic Hamiltonian system
(3.4.1) exists everywhere on N .
Remark 3.4.1. Since N2 6= Z in general, one can consider another variant of how
to solve a presymplectic Hamiltonian system. Let N2 (3.4.5) be a submanifold of
Z. It can be provided with the pull-back presymplectic form Ω2 = i∗N2 Ω. Then the
pull-back presymplectic Hamiltonian system
[
SH2 = {v ∈ Tz N2 : vcΩ2 + dH2 (z) = 0}
z∈N2
for the pull-back Hamiltonian H2 = i∗N2 H on N2 can be examined. As for the initial
system SH , we find that the equation
vcΩ2 + dH2 (z) = 0, v ∈ Tz N2 ,
has a solution on the subset
N30 = {z ∈ N2 : uci∗N2 dH(z) = 0, ∀u ∈ Ker z Ω2 ⊂ T N2 } ⊂ N2 ,
and so on. We obtain the chain of submanifolds
N2 ⊃ N30 ⊃ N40 · · ·
which is finite, and is stopped by one of the following results.
• There is i ≥ 2 such that a set Ni0 is empty.
• A set Ni0 , i ≥ 2, fails to be a submanifold.

0
• If Ni+1 = Ni0 for some i ≥ 2, this is a desired submanifold N 0 ⊂ Z. If the
presymplectic form i∗N 0 Ω on N 0 is of constant rank, then there exists a solution
of the presymplectic Hamiltonian system (i∗N 0 Ω, ı∗N 0 H) everywhere on N 0 .
It is readily observed that any solution ϑH ⊂ T N of the equation (3.4.2) on a

submanifold N ⊂ Z also obeys the equation
ϑH ci∗N (Ω + dH) = 0
and, consequently, is a solution of the pull-back presymplectic system on N . It

follows that N ⊂ N 0 . •
Solutions of the presymplectic Hamiltonian system SH (3.4.1) on a submanifold

N constitute an affine space modelled over the linear space of sections of the vector
bundle Ker Ω ∩ T N → N . If this vector bundle is not 0, different solutions of the
Hamilton equation (3.4.1) correspond to different first order dynamic equations on
N . Therefore, a Hamilton equation with respect to a presymplectic structure on a
manifold Z is not equivalent to a first order dynamic equation on Z in general.
Sections of the vector bundle Ker Ω → Z are sometimes called gauge fields in
order to emphasize that, being solutions of the presymplectic Hamiltonian system
(Ω, 0) for the zero Hamiltonian, they do not contribute to a physical state, and are
responsible for the gauge freedom [12, 172]. At the same time, there are physically
interesting presymplectic Hamiltonian systems, e.g., in relativistic mechanics when
a Hamiltonian is equal to zero (see also Remark 4.8.9 below). In this case, Ker dH =
T Z and the Hamilton equation (3.4.4) has a solution everywhere on a manifold Z.
Another pull-back construction is connected with the above-mentioned gauge
freedom. Let a presymplectic form Ω on manifold Z be of constant rank and let its
characteristic foliation be simple, i.e., its leaves are fibres of a fibre bundle π : Z → P
over a symplectic manifold (P, ΩP ) in accordance with Proposition 2.6.2. Then Ω is
the pull-back π ∗ ΩP of a symplectic form ΩP on the base P by this fibration. Let a
Hamiltonian H be the pull-back π ∗ HP of a function HP on P . Then we have
Ker Ω = V N ⊂ Ker dH,

and the presymplectic Hamiltonian system (Ω, H) has solutions everywhere on the
manifold Z. Any such solution ϑH is projected onto a unique solution of the symplec-
tic Hamiltonian system (ΩP , HP ) on the manifold P , while gauge fields are vertical
vector fields on the fibre bundle Z → P . This is the case of a gauge invariant
Hamiltonian system, where P is called a physical phase space.
In the case of a presymplectic Hamiltonian system on a manifold Z, this manifold
fails to be provided with a Poisson structure in general (see Remark 2.5.1), that is an
essential problem for quantization. Nevertheless, by virtue of Propositions 2.5.3 and
2.5.4, every presymplectic form on a manifold Z can be represented as a pull-back of
a symplectic form by the coisotropic imbedding. It follows that each presymplectic
Hamiltonian system can be seen as a Dirac constraint system [26] that we will
investigate in Section 3.6.
Example 3.4.2. Autonomous Lagrangian systems with constraints are often treated
as presymplectic Hamiltonian systems (see [30, 61, 112, 113, 138] and references
therein). Every Lagrangian L on the tangent bundle T M of a configuration mani-
fold M provides T M with the presymplectic form ΩL (3.3.5). Let us consider the
presymplectic Hamiltonian system (ΩL , EL ) on T M for the Hamiltonian EL (3.3.6).
This is exactly the Cartan equation (3.3.8) – (3.3.9) for the Lagrangian L. Its re-
striction to the submanifold q̇i = q̇ i (3.3.10) of the double tangent bundle T T M
of M leads to the Lagrange equations (3.3.11). This means that solutions of the
Lagrange equations are solutions of the Cartan equation which are dynamic vector
fields on T T M .
Conversely, a symplectic Hamiltonian system (Ω, H) on the cotangent bundle
∗
T M can be seen as a degenerate Lagrangian system on the velocity phase space
T T ∗ M of the momentum phase space T ∗ M as follows. Put the Lagrangian
LH = pi q̇ i − H(q i , pi ) (3.4.6)
on this velocity phase space. It is readily observed that the pre-image in T T T ∗ M of

the Hamilton equation (3.3.2) on T T ∗ M for the Hamiltonian H by the projection
T T T ∗ M → T T ∗ M is exactly the Lagrange equations (3.3.11) on T T T ∗ M for the
Lagrangian (3.4.6). It follows that they have the the same solutions r : () → T ∗ M .
It should be emphasized, that, in general, neither Cartan equations nor Lagrange
equations are differential equations in a strict sense. They are differential equations,
e.g., when the Hessian matrix πij of L is of constant rank. •
3.5 Dirac Hamiltonian systems
There is an extensive literature devoted to autonomous mechanical systems with

constraints (see, e.g., [121, 127, 137, 166, 172] and references therein). In this and
the next Sections, we will deal only with autonomous Hamiltonian systems with con-
straints. We leave the constraints in time-dependent Lagrangian and Hamiltonian
mechanics for Chapters 4 and 5.
Let (Z, Ω) be a 2m-dimensional symplectic manifold and H a Hamiltonian on
Z. Let N be a (2m − n)-dimensional closed imbedded submanifold of Z, called the
primary constraint space or simply a constraint space. We aim to investigate the
following two problems:
(A) solutions of the Hamiltonian system (Ω, H) on a manifold Z, which live in
the constraint space N ,
(B) and the restriction of the Hamiltonian system (Ω, H) to the constraint space
N.
Remark 3.5.1. The following local relations will be useful in the sequel. Let the
constraint space N be given locally by the equations
fa (z) = 0, a = 1, . . . , n, (3.5.1)
where fa (z) are local functions on Z, called the primary constraints. Let us consider
the set
IN = Ker i∗N ⊂ O0 (Z) (3.5.2)
of real functions f on Z which vanish everywhere on N , i.e., i∗N f = 0. This is an

ideal of the associative commutative algebra O0 (Z) such that
O0 (Z)/IN ∼
= O0 (N )
(cf. (2.6.4)). Its elements f are written locally in the form

n
g a fa ,
X
f= (3.5.3)
a=1
where g a are functions on Z. This means that the ideal IN is locally generated by
the constraints fa . We will call {fa } a local basis for the ideal IN .
3.5. DIRAC HAMILTONIAN SYSTEMS 127
Let dIN be the submodule of the O0 (Z)-module O1 (Z), which is generated locally
by the exterior differentials df of functions f ∈ IN . Its elements are the finite sums
g i dfi , g i ∈ O0 (Z).
X
σ= fi ∈ IN ,
i
By virtue of (3.5.3), they are given by local expressions

n
(g a dfa + fa φa ),
X
σ= (3.5.4)
a=1
where g a are functions and φa are 1-forms on Z.

It should be emphasized that the expressions (3.5.3) and (3.5.4) are fulfilled
locally on a domain where the constraint space N is given by the equations (3.5.1).
•
The solution of problem (A) is obvious. It exists if a Hamiltonian vector field
ϑH , restricted to the constraint space N , is tangent to N , i.e.,
[
SH |N = {v ∈ Tz Z : vcΩ + dH(z) = 0} ⊂ T N. (3.5.5)
z∈N
Then integral curves of the Hamiltonian vector field ϑH do not leave N .

The condition (3.5.5) is satisfied if and only if
{H, IN } ⊂ IN , (3.5.6)
i.e., if and only if the Hamiltonian H belongs to the normalizer I(N ) (2.6.6) of the
ideal IN . With respect to the local basis {fa }, the condition (3.5.6) reads
n
gac fc ,
X
ϑH cdfa = {H, fa } = (3.5.7)
c=1
where g c are functions on Z.

If the relation (3.5.6) does not hold, let us introduce the secondary constraints
{H, fa } = 0.
If the collection of primary and secondary constraints is not closed with respect to
the relation (3.5.7), let us add the tertiary constraints
{H, {H, fa }} = 0,
and so on. If a solution exists anywhere on N , the procedure is stopped after a finite
number of steps by constructing the complete system of constraints. The complete
system of constraints defines the final constraint space, which do not include the
points of Z where the Hamiltonian vector field ϑH is transversal to the primary
constraint space N .
¿From the algebraic viewpoint, we find an ideal Ifin of O0 (Z) which is the minimal
extension of the ideal IN such that
{H, Ifin } ⊂ Ifin .
In the framework of the algebraic theory of constraints, problem (A) can be

reformulated as follows.
Let N be a closed imbedded submanifold of a symplectic manifold (Z, Ω) and IN
(3.5.2) the ideal of functions which vanish everywhere on N . All its elements are said
to be constraints. We aim to find a Hamiltonian, called an admissible Hamiltonian,
on Z such that the symplectic Hamiltonian system (Ω, H) has a solution everywhere
on the constraint space N . One can treat N as a final constraint space. Moreover,
we may take an ideal IN whether it is an ideal of functions vanishing on some
submanifold N ⊂ Z or not.
In accordance with the condition (3.5.6), any element of the normalizer I(N )
(2.6.6) is an admissible Hamiltonian.
Let us consider the intersection I 0 (N ) = I(N ) ∩ IN (2.6.7). Its elements are
called the first-class constraints, while the remaining elements of IN are the second-
class constraints. The first-class constraints make up a Poisson subalgebra of the
normalizer I(N ) which, in turn, is a Poisson subalgebra of the Poisson algebra O0 (Z)
on the symplectic manifold (Z, Ω) (see Section 2.6). Recall that IN2 ⊂ I 0 (N ), i.e.,
the products of second-class constraints are also first-class constraints.
Example 3.5.2. If N is a coisotropic submanifold of Z, then IN ⊂ I(N ) and
I 0 (N ) = IN . It follows that we have only the first-class constraints. •
The admissible Hamiltonians which are not first-class constraints are the rep-
resentatives of the non-zero elements of the quotient I(N )/I 0 (N ), which is the re-
duction of the Poisson algebra O0 (Z) via the ideal IN (see Definition 2.6.8). In
particular, let the presymplectic form i∗N Ω on N be of constant rank and its char-
acteristic foliation be simple, i.e., it defines a fibration π : N → P over a symplectic
manifold P . In view of the isomorphism (2.6.8), one can think of elements of the
3.5. DIRAC HAMILTONIAN SYSTEMS 129
quotient I(N )/I 0 (N ) as being the Hamiltonians on the physical phase space P . It
follows that the restriction of an admissible Hamiltonian H to the constraint space
N coincides with the pull-back onto N of some function f on P , i.e.,
i∗N H = π ∗ f, f ∈ O0 (P ).
Let us turn now to problem (B). Given a Hamiltonian H on a symplectic manifold

(Z, Ω) and a constraint space N ⊂ Z, a Dirac Hamiltonian system (Ω, H, N ) on the
constraint space N is defined as the subset
def
{v ∈ Tz Z : i∗N (vcΩ + dH(z)) = 0}
[
S∗H = (3.5.8)
z∈N
[127, 137]. Note that
SH |N ⊂ S∗H .
Therefore, the set (3.5.8) is not empty.

We will say that a Dirac Hamiltonian system on N has a solution if there is a
vector field on a neighbourhood of each point z ∈ N , which lives into the subset S∗H
(3.5.8). For instance, every solution of problem (A) is also a solution of problem
(B).
Example 3.5.3. Problem (B) has a complete solution in the germ terms in the
following case. Let Z = T ∗ M be a symplectic manifold equipped with the canonical
symplectic form Ω, and N its coisotropic submanifold. Then T N is a coisotropic
submanifold of the symplectic manifold T T ∗ M equipped with the symplectic form
e (2.4.9). Indeed, let u ∈ T T N and v ∈ T T T ∗ M \ T T N . There exists a vector field
Ω
τu on N such that its canonical lift τeu onto T N contains u. There is also a vector
field τv on T ∗ M such that its canonical lift τev onto T T ∗ M contains v. Let u and v
be projected over the same element q ∈ T N . By the virtue of the relation (1.2.23),
one can then write
vcucΩ
e = τe (q)cτe (q)cΩ
v u
e = (τ (z)cτ
v
g(z)cΩ),
u z = πZ (q).
In the case of an arbitrary u ∈ Tq T N , this expression differs from 0 because there

exists a vector field τv on a neighbourhood of z, which does not belong to OrthΩ T N
since N is coisotropic. It follows that OrthΩe T T N does not include elements of
T T T ∗M \ T T N .
Let H be a Hamiltonian on T ∗ M . Then the symplectic Hamiltonian system SH

(3.3.1) is a Lagrangian submanifold of the symplectic manifold (T T ∗ M, Ω)
e (see Ex-
ample 2.4.7). A Dirac Hamiltonian system (Ω, H, N ) on a (2m − n)-dimensional
coisotropic submanifold N has a solution if the coisotropic submanifold T N ⊂
T T ∗ M and the Lagrangian submanifold SH are not transversal. The germ of any
such pair (T N, S; .) is symplectically equivalent to the pair
({q 1 = · · · = q 2n = 0}, {q a = ∂ a S(p), a = 1, . . . , 2m}; 0), (q, p) ∈ R4m ,
where S is the germ of a function such that the rank of the tangent map to the
morphism
p → (∂1 S, . . . , ∂2n S)
at 0 is not maximal. We refer the reader to [48] for a detailed classification of the
germs of such generating functions S. •
A Dirac Hamiltonian system (Ω, H, N ) is called completely integrable if there

exists its solution through each point of the set S∗H . The obvious necessary condition
for a Dirac Hamiltonian system to be completely integrable is the inclusion S∗H ⊂
T N . In this case, a Dirac Hamiltonian system (Ω, H, N ) reduces to the symplectic
Hamiltonian system (i∗N Ω, i∗N H) on the constraint space N .
Let us inspect the above-mentioned necessary condition. Note that, for each
point z ∈ N , the fibre
{v ∈ Tz Z : i∗N (vcΩ + dH(z)) = 0}
of the set S∗H over z ∈ N is a non-empty affine space modelled over the vector space
{v ∈ Tz Z : ucvcΩ = 0, ∀u ∈ Tz N } (3.5.9)
which is the fibre of the Ω-orthogonal space OrthΩ T N (2.4.6) of T N . It follows that
OrthΩ T N ⊂ T N . Therefore, a Dirac Hamiltonian system on the constraint space
N is completely integrable only if N is a coisotropic submanifold of the symplectic
manifold (Z, Ω). Then we have
ucdH = 0, ∀u ∈ OrthΩ N.
In order to formulate a sufficient condition for a Dirac Hamiltonian system to be

completely integrable, let us assume that the vector spaces (3.5.9) for all z ∈ N are
3.6. DIRAC CONSTRAINT SYSTEMS 131
of the same dimension. Then the Ω-orthogonal OrthΩ T N of T N is a vector bundle

over N , while the Dirac Hamiltonian system S∗H → N is an affine bundle over N .
Proposition 3.5.1. [127]. Let a submanifold N of a symplectic manifold (Z, Ω)

be coisotropic. Then OrthΩ T N is an involutive distribution on N . If the exterior
differential dH is constant on the leaves of the corresponding foliation, then the
Dirac Hamiltonian system (3.5.8) is completely integrable. 2
Proof. Since the set S∗H is not empty, it suffices to show that this set belongs to
T N . Let v ∈ S∗H and z = πZ (v). Whenever u ∈ OrthΩ T N ⊂ T N over z, the
equality
vcucΩ = ucdH = 0
holds. It follows that v ∈ Tz N . Since S∗H ⊂ T N , there is a vector field ϑ on N

through any point of the Dirac Hamiltonian system S∗H . Such a solution is not
unique. If ϑ is the above-mentioned vector field, then any vector field ϑ + υ where υ
lives in OrthΩ T N ⊂ T N is also a solution of the Dirac Hamiltonian system. QED
Thus, one should consider completely integrable Dirac Hamiltonian systems on

coisotropic submanifolds N . In this case, a Dirac Hamiltonian system (Ω, H, N )
reduces to the presymplectic Hamiltonian system on the constraint space N for the
pull-back Hamiltonian i∗N H with respect to the pull-back presymplectic form i∗N Ω.
This is the case of a Dirac constraint system.
3.6 Dirac constraint systems

Let (Z, Ω) be a 2m-dimensional symplectic manifold, and let N be a closed imbedded
submanifold of Z. Let H be a Hamiltonian on Z.
Definition 3.6.1. A Dirac constraint system on the constraint space N is the set
def
{v ∈ Tz N : vci∗N (Ω + dH(z)) = 0}.
[
SN ∗ H = (3.6.1)
z∈N
In particular, we are dealing with Dirac constraint systems when a Hamiltonian

description of degenerate autonomous Lagrangian systems is investigated (see, e.g.,
[58, 172] and references therein). They are called generalized Hamiltonian systems.
In this case, a momentum phase space Z is the cotangent bundle T ∗ M of a config-
uration space M . This phase space is provided with the canonical symplectic form.
A primary constraint space N ⊂ Z is an image of the Legendre map Lb (3.3.13)
defined by a degenerate Lagrangian L on the velocity phase space T M . In fact, one
introduces some initial Hamiltonian H on T ∗ M such that the energy function EL
(3.3.6) for L is the pull-back L∗ H of this Hamiltonian H by the Legendre map L. b
Thus, we come to a Dirac constraint system on the Lagrangian constraint space N .
The goal consists in constructing a Hamiltonian H0 on the momentum phase space
T ∗ M (but not uniquely in general) such that the solution ϑH0 of the symplectic Ha-
miltonian system (Ω, H0 ) on T ∗ M provides a solution of the above-mentioned Dirac
constraint system on a final constraint space.
In fact a Dirac constraint system SN ∗ H is the pull-back presymplectic Hamil-
tonian system (i∗N Ω, i∗N H) on a submanifold N ⊂ Z for the Hamiltonian i∗N H and
with respect to the presymplectic form i∗N Ω on N . For instance, Proposition 3.4.2
shows that the equation
i∗N (vcΩ + dH(z)) = 0, v ∈ Tz N,
has a solution only at the points of the subset
N2 = {z ∈ N : uci∗N dH(z) = 0, ∀u ∈ Ker z ΩN }
which is assumed to be a manifold. Such a solution, however, fails to be tangent to

N2 . Then one should repeat the algorithm for presymplectic Hamiltonian systems
in Remark 3.4.1. Nevertheless, one can say more since the presymplectic system
SN ∗ H (3.6.1) on N is the pull-back of the symplectic system (Ω, H) on Z.
Let a (2m − n)-dimensional closed imbedded submanifold N of Z be a final
constraint space of the Dirac constraint system (3.6.1). This means that the equation
vcΩN + dHN (z) = 0, v ∈ Tz N,

ΩN = i∗N Ω, HN = i∗N H,
has a solution at each point z ∈ N . By virtue of Proposition 3.4.2, this is equivalent

to the condition
ucdHN (z) = 0, ∀u ∈ Ker z ΩN , ∀z ∈ N,

or
Ker ΩN = T N ∩ OrthΩ T N ⊂ Ker H. (3.6.2)
We will reformulate these condition in algebraic terms. Let IN be the ideal of

functions on Z which vanish everywhere on the final constraint space N . Let I(N )
be the normalizer (2.6.6) of IN and I 0 (N ) = I(N ) ∩ IN . It is readily observed that,
restricted to N , Hamiltonian vector fields ϑf for elements f of I 0 (N ) with respect
to the symplectic form Ω on Z take their values into T N ∩ OrthΩ T N [97]. Then the
condition (3.6.2) can be written in the form
{H, I 0 (N )} ⊂ IN , (3.6.3)
whereas
{H, IN } 6⊂ IN (3.6.4)
in general. A glance at the relations (3.6.3) and (3.6.4) shows that, though the
Dirac constraint system on N has a solution, the Hamiltonian vector field ϑH for
the Hamiltonian H with respect to the symplectic form Ω on Z is not tangent to
N , and its restriction to N is not such a solution in general.
Example 3.6.1. Let the final constraint space N be a coisotropic submanifold of
the symplectic manifold (Z, Ω). Then IN = I 0 (N ), i.e., there are only first-class
constraints. In this case, the Hamiltonian vector fields of the Hamiltonian H and
of other Hamiltonians H + f , f ∈ IN , provide solutions of the Dirac constraint
system on N . Note that, if a primary constraint space is coisotropic, we arrive at
a Dirac Hamiltonian system, where the relation (3.6.3) is exactly the condition of
Proposition 3.5.1. •
The goal of constructing a generalized Hamiltonian system is to find a constraint

f ∈ IN such that the modified Hamiltonian H + f would satisfy both the conditions
{H + f, I 0 (N )} ⊂ IN , (3.6.5)
{H + f, IN } ⊂ IN . (3.6.6)
The condition (3.6.5) is fulfilled for any f ∈ IN , while (3.6.6) is an equation for a
second-class constraint f .
Remark 3.6.2. Since IN is the ideal of functions vanishing on a manifold N ,

the normalizer of IN coincides with the normalizer of I 0 (N ) [97]. Therefore, the
conditions (3.6.5) and (3.6.6) may be combined into
{H + f, I 0 (N )} ⊂ I 0 (N ).
•
The equation (3.6.6) is solved in many concrete models, that implies separating
first- and second-class constraints. The general problem lies in the fact that the
collection of elements which generate IN2 ⊂ I 0 (N ) is necessarily infinite reducible
[97]. At the same time, the Hamiltonian vector fields for elements of IN2 vanish on
the constraint space N . Therefore, one can apply the following procedure [137].
Since N is a (2m−n)-dimensional closed imbedded submanifold of Z, the ideal IN
is locally generated by a finite basis {fa }, a = 1, . . . , n, while N is defined locally by
the equations (3.5.1). Let the presymplectic form Ωn be of constant rank 2m−n−k,
i.e., Ker z ΩQ has the dimension k ≤ n at all points z ∈ Q. It defines a k-dimensional
characteristic foliation on N . Since N ⊂ Z is closed, there exist locally k linearly
independent vector fields ub on Z which, restricted to N , are tangent to the leaves
of this foliation. They read
n
gba ϑfa ,
X
ub = b = 1, . . . , k,
a=1
where gba are functions on Z and ϑfa are Hamiltonian vector fields for constraints
fa . Then one can choose the collection of constraints φb , b = 1, . . . , n, where the
first k functions take the form
n
gba fa ,
X
φb = b = 1, . . . , k.
a=1
Let ϑφb be the Hamiltonian vector fields for these functions. One can easily justify
the fact that
ϑφb |N = ub |N , b = 1, . . . , k.
It follows that the constraints φb belong to I 0 (N ) \ IN2 . Therefore, they are called
the first-class constraints, while the remaining φk+1 , . . . , φn are the second-class con-
straints. We have the relations
n
a
X
{φb , φc } = Cbc φa , b = 1, . . . , k, c = 1, . . . , n,
a=1
a
where Cbc are local functions on Z.
It should be emphasized that the constraints φ1 , · · · , φk do not constitute any
local basis for I 0 (N ), though φ1 , · · · , φn make up a local basis for the ideal IN .
Now let us consider a local Hamiltonian on Z
n
H0 = H + λa φa ,
X
(3.6.7)
a=1
where λa are functions on Z. Since H obeys the condition (3.6.5), we find

n
Bba φa ,
X
{H, φb } = b = 1 . . . , k,
a=1
where Bba are functions on Z. Then, the equation (3.6.6) takes the form
n n
a
Dcb φb ,
X X
{H, φc } + λ {φa , φc } = c = k + 1, . . . n, (3.6.8)
a=k+1 b=1
where Dcb are functions on Z. This is the system of linear algebraic equations for
the coefficients λa , a = k + 1, . . . n, of second-class constraints. These coefficients
are defined uniquely by the equations (3.6.8), while the coefficients λa , a = 1, . . . , k,
of first-class constraints in the Hamiltonian (3.6.7) remain arbitrary.
Then, being restricted to the constraint space N , the Hamiltonian vector field
of the Hamiltonian H0 (3.6.7) on Z provides a local solution of the Dirac constraint
system on N .
We refer the reader to [137] for a global variant of the above procedure.
Example 3.6.3. If N is a symplectic submanifold of Z, then I 0 (N ) = IN2 . Therefore,

we have only second-class constraints, and the Hamiltonian (3.6.7) of a generalized
Hamiltonian system is defined uniquely. •
Remark 3.6.4. In concrete models, the final constraint space N fails to be given
in advance. Therefore, a different procedure of separating first- and second-class
constraints is usually applied (see, e.g., [39, 58, 172]), but its global treatment is
under discussion. •
3.7 Hamiltonian systems with symmetries

In this Section, we summarize the well-known results connected with a group action
on symplectic and Poisson manifolds [2, 116, 126, 130, 181]. By G throughout is
meant a real connected Lie group.
We start from a symplectic manifold case. Let (Z, Ω) be a symplectic manifold
on which a real connected Lie group G acts on the left so that, whenever g ∈ G, the
mapping z 7→ gz is a symplectic isomorphism of Z. Such an action of a group G is
called a symplectic action.
Remark 3.7.1. The classification of transitive symplectic actions of connected Lie
groups has been done (see [69, 72, 103]). •
Since G is connected, its action on a manifold Z is symplectic if and only if, for
any element ε of the right Lie algebra gr of a the group G, the corresponding vector
field ξε on Z is locally Hamiltonian, i.e.,
Lξε Ω = d(ξε cΩ) = 0.
Let us suppose that it is a Hamiltonian vector field, i.e.,
ξε cΩ = −dJε ,
where Jε is a real function on Z.
Definition 3.7.1. A symplectic action of a connected Lie group G on a symplectic

manifold Z is called a Hamiltonian action if, whenever ε ∈ gr , there exists a function
Jε on Z such that ξε is the Hamiltonian vector field for Jε . 2
Proposition 3.7.2. An action of a connected Lie group G on a symplectic manifold

Z is Hamiltonian if and only if there exists a mapping, called a momentum mapping,
from Z to the Lie coalgebra
Jb : Z → g∗
such that, whenever ε ∈ gr , the function
Jε (z) = hJ(z),
b εi, ∀ε ∈ gr , (3.7.1)
is that in Definition 3.7.1. 2

3.7. HAMILTONIAN SYSTEMS WITH SYMMETRIES 137
Proof. If Jb exists, we obviously have a Hamiltonian action. Conversely, if ξε for

any ε ∈ gr is the Hamiltonian vector field for a function Jε on a manifold Z, then
the momentum mapping Jb is defined by the relation (3.7.1) applied to the basis
{εm } of the Lie algebra gr . QED
It is readily observed that, if Jb and Jb0 are different momentum mapping for the
same action of G on Z, then
dhJ(z)
b − Jb0 (z), εi = 0, ∀ε ∈ gr ,
i.e., Jb − Jb0 = const. on Z.

Example 3.7.2. Let a symplectic form on Z be exact, i.e., Ω = dθ, and let θ be
G-invariant, i.e.,
Lξε θ = d(ξε cθ) + ξε cΩ = 0, ∀ε ∈ gr .
Then the map Jb : Z → g∗ , given by the relation
hJ(z),
b εi = (ξε cθ)(z),
is a momentum mapping. •
Proposition 3.7.3. If a Hamiltonian H on a symplectic manifold Z is invariant

under a group G acting on Z, a momentum mapping Jb defines a family of first
integrals of motion of the Hamiltonian system for H. 2
Proof. We have
Lξε H = ξε cdH = 0, ∀ε ∈ gr ,
and then
{H, Jε } = 0, ∀ε ∈ gr .
QED
Thus, the functions Jε (z) (3.7.1) are first integrals of motion. However, they are
not pairwise in involution in general. Let us find their Poisson brackets.
Definition 3.7.4. A momentum mapping Jb is called equivariant if
J(gz)
b = Ad∗ g(J(z)),
b ∀g ∈ G,
where Ad∗ g is the coadjoint representation (1.5.3). 2
Example 3.7.3. The momentum mapping in Example 3.7.2 is equivariant. In

accordance with the relation (1.5.3), it suffices to show that
Jε (gz) = JAd g−1 (ε) (z),
i.e.,
(ξε cθ)(gz) = (ξAd g−1 (ε) cθ)(z).
The latter results follow from the relation (1.5.2). •
Example 3.7.4. Let T ∗ M be a symplectic manifold equipped with the canonical

symplectic form Ω (2.4.2). Let a group G act on M on the left by the generators
ξm = εim (q)∂i .
The canonical lift of this action onto T ∗ M has the generators
ξem = εim ∂i − pj ∂i εjm ∂ i , (3.7.2)
and preserves the canonical Liouville form θ on T ∗ M . In accordance with Example

2.4.4, the vector fields (3.7.2) are Hamiltonian vector fields for the functions
Jm = εim (q)pi (3.7.3)
and, as follows from Example 3.7.3, there exists the equivariant momentum mapping
Jb = εim (q)pi εm .
In the case of a G-invariant Hamiltonian on T ∗ M , the functions (3.7.3) are first

integrals of motion. •
In the case of non-equivariant momentum mappings, let us consider the difference
σ(g) = J(gz)
b − Ad∗ g(J(z)).
b (3.7.4)
We refer the reader to [2] for proof that this difference is constant on a symplectic
manifold Z, and fulfills the equality
σ(gg 0 ) = σ(g) + Ad∗ g(σ(g 0 )). (3.7.5)
Definition 3.7.5. A map σ : G → g∗ is called the cocycle on a group G if the

condition (3.7.5) holds. A cocycle σ is said to be a coboundary on a group G, if
there exists an element µ ∈ g∗ such that
σ(g) = µ − Ad∗ g(µ). (3.7.6)
The equivalence classes of cocycles modulo coboundaries make up the cohomology
group of the group G. 2
By this definition, the difference (3.7.4) is the cocycle of the group G associated
with the action of G on Z.
Proposition 3.7.6. Each symplectic action of a group G on a symplectic manifold

Z defines a cohomological class [σ] of G. 2
Proof. Let Jb and Jb0 be different momentum mappings associated with a symplectic
b Jb0 is constant
action of a group G on a symplectic manifold Z. Since the difference J−
on Z, then the difference of the corresponding cocycles σ − σ 0 is the coboundary
(3.7.6) where µ = Jb − Jb0 . QED
In particular, an equivariant momentum mapping defines the zero cohomological

class of the group G.
Theorem 3.7.7. Given a momentum mapping Jb associated with a symplectic

action of a group G on a symplectic manifold Z, the following relation
{Jε , Jε0 } = J[ε,ε0 ] − hTe σ(ε0 ), εi (3.7.7)
takes place (we refer the reader again to [2], where, however, the left Lie algebra is
utilized and Hamiltonian vector fields differ in the minus sign from those we use).
2
In the case of an equivariant momentum mapping as in Example 3.7.4, the

relation (3.7.7) leads to the homomorphism
{Jε , Jε0 } = J[ε,ε0 ] (3.7.8)
of the Lie algebra gr to the Lie algebra of functions on a symplectic manifold Z

with respect to the Poisson bracket. These are the desired Poisson brackets of first
integrals of motion for a G-invariant Hamiltonian system.
We will refer to this result in connection with the conservation laws in time-
dependent mechanics (see Section 5.8).
Let now (Z, w) be a Poisson manifold, on which a connected Lie group G acts on
the left so that, whenever g ∈ G, the mapping z → gz is a Poisson automorphism
of Z. Such an action of a group G is called a Poisson action. Since G is connected,
its action on a manifold Z is a Poisson action if and only if, for any element ε of
the right Lie algebra gr of a group G, the corresponding vector field ξε on Z is an
infinitesimal Poisson automorphism, i.e., the condition (2.3.3) holds. The equivalent
conditions are
ξε ({f, g}) = {ξε (f ), g} + {f, ξε (g)},

ξε ({f, g}) = [ξε , ϑf ](g) − [ξε , ϑg ](f ),
[ξε , ϑf ] = ϑξε (f ) ,
where f, g ∈ O0 (Z), and by ϑf is meant the Hamiltonian vector field (2.3.4) for a
function f . Note that, by very definition of a Hamiltonian action, generators of a
Hamiltonian action of a group G on a Poisson manifold (Z, w) are tangent to the
leaves of the symplectic foliation on Z, and there is a Hamiltonian action of G on
every symplectic leaf in the sense of symplectic geometry.
Definition 3.7.1, Propositions 3.7.2, 3.7.3, and Definition 3.7.4 are naturally ex-
tended to a Poisson action of a connected Lie group G on a Poisson manifold. Let
us consider the case of an equivariant momentum mapping.
Proposition 3.7.8. [181] An equivariant momentum mapping Jb : Z → g∗ is a

Poisson morphism if the coalgebra g∗ is provided with the Lie–Poisson structure
(2.3.11). 2
There is the following reduction theorem.
Theorem 3.7.9. [181]. Let (Z, w) be a Poisson manifold endowed with a Hamilto-
nian action of a connected Lie group G and the equivariant momentum map J.
b Let
∗
q be a point of g such that
• q is a non-critical value for all the restrictions of Jb to the symplectic leaves

Fι of Z, i.e., the level set Zq = Jb−1 (q) is a submanifold of Z and, for any
symplectic leaf Fι , Fιq = Zq ∩ Fι is a submanifold of Fι ;
• intersections Fιq of the submanifold Zq with symplectic leaves of Z are clean,

i.e., T Fιq = T Fι ∩ T Zq , and so are its intersections with the orbits of G in Z.
Let a subgroup Gq ⊂ G be the stabilizer of the point q. Then intersections Fιq make
up a regular foliation on Zq , and its leaves are the orbits of the connected component
of the unit in Gq for the action of Gq on Zq . Furthermore, if this foliation is defined
by a submersion Zq → P , the base P has a well-defined Poisson structure whose
characteristic distribution is the projection of T ∩ T Zq , where T is the characteristic
distribution on (Z, w). This is exactly the reduced Poisson manifold of (Z, w) via
(Zq , E = T (orbits G)) (see Definition 2.6.3). 2
Note that if the momentum mapping Jb of a Hamiltonian action on a Poisson

manifold is not equivariant, we have the relation
J(gz)
b = Ad∗ g(J(z))
b + σ(g, z), (3.7.9)
similar to (3.7.4) in the symplectic case, where the function σ(g, z) is constant on
the leaves of the symplectic foliation on Z. Since now σ(g, z) depends on z, we have
different stabilizers of q for the action given by (3.7.9) on g∗ for every symplectic
leaf. Therefore, the foliation on Zq is no longer regular.
We end this Section with the theorem which shows the possibility of reducing a
Poisson Hamiltonian system possessing symmetries.
Theorem 3.7.10. [129, 181]. Let (Z, N, E) be a Poisson reduction and P a reduced
manifold of Z via (N, E). Let H be a Hamiltonian on Z such that the flow Φt of
its Hamiltonian vector field ϑH preserves the submanifold N and the bundle E, and
dH is an annihilator of E. Then Φt induces a flow Φ b of Poisson automorphisms of
t
P (see Proposition 2.6.5) along the Hamiltonian vector field ϑHb for the Hamiltonian
c on P , defined uniquely by the relation H | = π ∗ H.
H c Moreover, we have
N
ϑHb ◦ π = T π ◦ ϑH .
2
Thus, one can say that, under the hypotheses of Proposition 3.7.10, the Poisson
Hamiltonian system on a Poisson manifold Z reduces to a Poisson system on the
reduced Poisson manifold P .
Example 3.7.5. Let (Z, Zq , E = T (orbits G)) be a Poisson reduction as in Theorem

3.7.9. If H is a G-invariant Hamiltonian on the Poisson manifold Z, the Poisson
Hamiltonian system reduces to a Poisson system on the reduced Poisson manifold
of (Z, w) via (Zq , E = T (orbits G)). •
3.8 Appendix. Hamiltonian field theory
As is well-known, applied to a field theory on a fibre bundle Y → X, the famil-

iar symplectic techniques take the form of instantaneous Hamiltonian formalism
on an infinite-dimensional momentum phase space [64], in contrast with the finite-
dimensional velocity phase space J 1 Y . The Hamiltonian counterpart of the La-
grangian formulation of field theory is polysymplectic Hamiltonian formalism on
the Legendre bundle Π (2.9.7) provided with the polysymplectic form Λ (2.9.9) (see
[57, 159] for a survey). Here we aim to give a brief exposition of this formalism
because its restriction to fibre bundles Q → R leads in a straightforward manner to
the Hamiltonian formulation of time-dependent mechanics [57, 161].
Let Y → X be a fibre bundle over an n-dimensional manifold X, and
n−1
Π = V ∗ Y ∧ ( ∧ T ∗ X) (3.8.1)
the Legendre bundle (2.9.7). It admits the composite fibration
πΠX = π ◦ πΠY : Π → Y → X. (3.8.2)
For the sake of convenience, we will call Π → Y the Legendre vector bundle, while
the term Legendre bundle stands for the fibration Π → X.
Given fibred coordinates (xλ , y i ) on the fibre bundle Y → X, the Legendre bundle
(3.8.1) is equipped with the holonomic coordinates (xλ , y i , pλi ) together with the
transition functions (2.9.8). These coordinates are compatible with the composite
fibration (3.8.2) and are linear bundle coordinates on the vector bundle Π → Y . We
will call them the canonical coordinates.
3.8. APPENDIX. HAMILTONIAN FIELD THEORY 143
There is the canonical bundle monomorphism over Y

n+1
Θ : Π ,→ ∧ T ∗ Y ⊗ T X, (3.8.3)
Y Y
Θ : (x , y , pλi ) 7→ −pλi dy i ∧ ω ⊗ ∂λ ,
λ i
which defines the tangent-valued Liouville form on the Legendre bundle Π. Then
the polysymplectic form
Λ = dpλi ∧ dy i ∧ ω ⊗ ∂λ (3.8.4)
on Π is defined as a unique T X-valued (n + 2)-form on Π such that the relation
Λcφ = −d(Θcφ)
holds for any exterior 1-form φ on X.

Given the vector Legendre bundle Π over a fibre bundle Y → X, we have the
exact sequence
n
0 −→ Π × ∧ T ∗ X ,→ ZY −→ Π −→ 0, (3.8.5)
X
where ZY is the homogeneous Legendre bundle (2.9.3) and
πZΠ : ZY → Π (3.8.6)
is an affine bundle over Π with 1-dimensional fibres.
Definition 3.8.1. Let h be a section of the fibre bundle (3.8.6). Then the pull-back
n
H = h∗ Ξ : Π → ∧ T ∗ Y, (3.8.7)
H = pλi dy i ∧ ωλ − Hω,
of the canonical form Ξ (2.9.4) on ZY by h is called the polysymplectic Hamiltonian

form (or simply the Hamiltonian form). 2
The exterior differential of the Hamiltonian form (3.8.7) is the pull-back
dH = h∗ ΩZ
on Π of the multisymplectic form (2.9.5).

Example 3.8.1. Let
Γ = dxλ ⊗ (∂λ + Γiλ ∂i )
be a connection on the fibre bundle Y → X. Hence, we have the splitting
Γ : V ∗ Y → T ∗ Y,
Γ : dy i 7→ dy i − Γiλ dxλ ,
of the exact sequence (1.1.8). Then Γ also yields the splitting
hΓ : Π → ZY ,
hΓ : pλi dy i ⊗ ωλ 7→ pλi dy i ∧ ωλ − pλi Γiλ ω,
of the exact sequence (3.8.5). It follows that every connection Γ on the fibre bundle
Y → X defines the Hamiltonian form
HΓ = h∗Γ Ξ,
HΓ = pλi dy i ∧ ωλ − pλi Γiλ ω, (3.8.8)
on the Legendre bundle Π. •
Proposition 3.8.2. [57, 159]. Hamiltonian forms on Π constitute an affine space

modelled over the linear space of horizontal densities
n
H f : Π → ∧ T ∗X
f = Hω (3.8.9)
on the Legendre bundle Π → X. They are called Hamiltonian densities. 2
This means that, if H is a Hamiltonian form and Hf is a horizontal density (3.8.9),

then H − H f is also a Hamiltonian form. Conversely, if H and H 0 are Hamiltonian
forms, their difference H − H 0 is a Hamiltonian density (3.8.9).
Example 3.8.1 and Proposition 3.8.2 combine into the following.
Corollary 3.8.3. Every Hamiltonian form on the Legendre bundle Π admits the
decomposition
f = pλ dy i ∧ ω − pλ Γi ω − H
H = HΓ − H f ω, (3.8.10)
Γ i λ i λ Γ
where Γ is a connection on Y → X. 2
Furthermore, every Hamiltonian form admits a canonical decomposition as fol-

lows. We mean by a Hamiltonian map any fibred morphism
Φ : Π → J 1 Y,
Y
yλi ◦ Φ = Φiλ (q), q ∈ Π, (3.8.11)
over Y . Its composition with the canonical morphism λ (1.3.4) yields the bundle
morphism
λ ◦ Φ : Π → T ∗X ⊗ T Y
Y Y
represented by the T Y -valued 1-form
Φ = dxλ ⊗ (∂λ + Φiλ (q)∂i ) (3.8.12)
on the vector Legendre bundle Π → Y .

Example 3.8.2. Let Γ be a connection on Y → X. Then, the composition
b =Γ◦π 1
Γ ΠY : Π → Y → J Y,
b = dxλ ⊗ (∂ + Γi ∂ ),
Γ λ λ i
is a Hamiltonian map. Conversely, every Hamiltonian map Φ : Π → J 1 Y yields the

associated connection
ΓΦ = Φ ◦ 0b
on Y → X, where 0b is the global zero section of the vector Legendre bundle Π → Y .

ΓbΓ = Γ.
Proposition 3.8.4. [57, 159] Every Hamiltonian form H on the Legendre bundle
Π defines the associated Hamiltonian map
c : Π → J 1 Y,
H
c = ∂ i H.
yλi ◦ H λ (3.8.13)
2
Corollary 3.8.5. Every Hamiltonian form H on the Legendre bundle Π deter-

mines the associated connection
c◦0
ΓH = H b
on Y → X. 2
ΓHΓ = Γ,
where HΓ is the Hamiltonian form (3.8.8) associated with the connection Γ on

Y → X.
Corollary 3.8.6. Every Hamiltonian form (3.8.10) admits the canonical splitting
H = HΓH − H.
f
The following assertion generalizes Example 3.8.1.
Proposition 3.8.7. Every Hamiltonian map (3.8.11) represented by the form

(3.8.12) on Π defines the associated Hamiltonian form
HΦ = ΦcΘ = pλi dy i ∧ ωλ − pλi Φiλ ω,
where Θ is the tangent-valued Liouville form (3.8.3). 2
In particular, if
HHb = H,
then H = HΓ for some connection Γ on Y .

Hamilton equations in conservative mechanics are the equations of integral curves
of Hamiltonian vector fields. Hamilton equations in polysymplectic Hamiltonian for-
malism are the equations of integral sections for Hamiltonian connections as follows.
Let J 1 Π be the first order jet manifold of the Legendre bundle Π → X. It is
equipped with the adapted fibred coordinates
(xλ , y i , pλi , yµi , pλµi ).

We have the commutative diagram

J 1π
J 1 Π −→
ΠY
J 1Y
? πΠY
?
Π −→ Y
yµi ◦ J 1 πΠY = yµi .
Definition 3.8.8. A connection

µ i
γ = dxλ ⊗ (∂λ + γλi ∂i + γλi ∂µ )
on the Legendre bundle Π → X is said to be a locally Hamiltonian connection if
the exterior form
γcΛ = dpλi ∧ dy i ∧ ωλ − (γλi dpλi − γλi
λ
dy i ) ∧ ω (3.8.14)
is closed. 2
Example 3.8.3. Every connection Γ on a fibre bundle Y → X gives rise to the

connection
e = dxλ ⊗ [∂ + Γi (y)∂ +
Γ (3.8.15)
λ λ i
(−∂j Γiλ pµi + Kλ µ ν pνj − Kλ α α pµj )∂µj ]

on the Legendre bundle Π → X, where K is a symmetric linear connection (1.4.9)
on T X. The connection Γ
e (3.8.15) obeys the relation
ΓcΛ
e = d(ΓcΘ).
e is a locally Hamiltonian connection. •
It follows that Γ
Thus, locally Hamiltonian connections always exist on the Legendre bundle Π →

X, and every connection Γ on Y → X gives rise to a locally Hamiltonian connection
on Π → X.
It is easily observed that a connection γ on the fibre bundle Π → X is a locally
Hamiltonian connection if and only if γ fulfills the conditions
∂λi γµj − ∂µj γλi = 0, (3.8.16)
µ µ
∂i γµj − ∂j γµi = 0, (3.8.17)
µ
∂j γλi + ∂λi γµj = 0. (3.8.18)
Using the relation (3.8.18), we find that the second term in the right-hand side
of the expression (3.8.14) is a closed form. Then, in accordance with the relative
Poincaré lemma, this expression is brought locally into the form
γcΛ = d(pλi dy i ∧ ωλ − Hγ ω) = dHγ , (3.8.19)
where Hγ is a local function on Π such that
γλi = ∂λi Hγ , λ
γλi = −∂i Hγ .
Given a connection Γ on the fibre bundle Y → X, the local form Hγ in the expression
(3.8.19) can be written as
Hγ = HΓ − H
f ω,
Γ
where H f ω is a local horizontal density on Π → X. In accordance with Proposition

Γ
3.8.2, it follows that Hγ defines the local section
hγ : (xλ , y i , pλi ) 7→ (xλ , y i , pλi , p = −Hγ )
of the fibre bundle ZY → Π, i.e., Hγ is a local Hamiltonian form. Thus, we have

proved the following.
Proposition 3.8.9. For every locally Hamiltonian connection γ on the Legendre

bundle Π → X, there exists a local Hamiltonian form H in a neighbourhood of each
point q ∈ Π such that
γcΛ = dH.
Let us formulate the converse assertion.
Definition 3.8.10. The Hamilton operator EH for a Hamiltonian form H on the

Legendre bundle Π → X is defined to be the first order differential operator
n+1
EH : J 1 Π → ∧ T ∗ Π,
b = [(y i − ∂ i H)dpλ − (pλ + ∂ H)dy i ] ∧ ω,
EH = dH − Λ (3.8.20)
λ λ i λi i
where
b = dpλ ∧ dy i ∧ ω + pλ dy i ∧ ω − y i dpλ ∧ ω
Λ i λ λi λ i
is the pull-back of the polysymplectic form Λ (3.8.4) on J 1 Π. 2
A glance at the expression (3.8.20) shows that the Hamilton operator EH is an

affine morphism over Π of constant rank. Thereby, its kernel is an affine subbundle
EH = Ker EH → Π (3.8.21)
of the affine jet bundle J 1 Π → Π which is given by the coordinate relations
yλi = ∂λi H,
pλλi = −∂i H.
This affine bundle is modelled over the vector subbundle of the vector bundle
T ∗X ⊗ V Π → Π (3.8.22)
Π
which is defined by the coordinate relations
y iλ = 0, pλλi = 0 (3.8.23)
with respect to the fibre coordinates (y iλ , pµλi ) on (3.8.22).

Since EH (3.8.21) is an affine subbundle, it is a closed imbedded submanifold
of the jet bundle J 1 Π → X and, therefore, is a system of first order differential
equations on Π → X in accordance with Definition 1.3.3.
Definition 3.8.11. The first order differential equations EH (3.8.21) are called the
Hamilton equations for the Hamiltonian form H on the Legendre bundle Π. 2
Since the subbundle EH (3.8.21) is affine, it always admits a global section γ.

Any such section is a connection on Π → X which meets the condition
EH ◦ γ = 0.
This condition takes the form
γcΛ = dH. (3.8.24)
It follows that every connection on Π → X which takes its values into the Hamilton
equations EH is a locally Hamiltonian connection.
Definition 3.8.12. A locally Hamiltonian connection γ on the Legendre bundle

Π → X is said to be a Hamiltonian connection associated with a Hamiltonian form
H if γ obeys the relation (3.8.24). 2
Thus, we have proved the following.
Proposition 3.8.13. Every Hamiltonian form on the Legendre bundle Π has an

associated Hamiltonian connection. 2
We have the equations of a Hamiltonian connection associated with a given

Hamiltonian form:
γλi = ∂λi H, (3.8.25)

λ
γλi = −∂i H. (3.8.26)
By the equation (3.8.25), every Hamiltonian connection γ for a Hamiltonian form

H satisfies the relation
J 1 πΠY ◦ γ = H,
c (3.8.27)
µ i
γ = dxλ ⊗ (∂λ + ∂λi H∂i + γλi ∂µ ),
where H
c is the Hamiltonian map (3.8.13). It projects over the connection Γ on
H
Y → X associated with the Hamiltonian form H. We have the commutative diagram
J 1Π
γ % ↓ J 1 πΠY
H
Π −→ J 1 Y
b
0 -% ΓH
Y
A glance at the equations (3.8.26) shows that there is a set of Hamiltonian

connections associated with the same Hamiltonian form H. They differ from each
other in soldering forms σe on Π → X which obey the equations
σe cΩ = 0,
σeλi = 0, λ
σeλi = 0,
and take their values into the subbundle (3.8.23) of the vector bundle (3.8.22).
A classical solution of the Hamilton equations (3.8.21), by definition, is a section

r of the Legendre bundle Π → X such that its jet prolongation J 1 r takes its values
into the kernel of the Hamilton operator EH (3.8.20). Then, r satisfies the differential
equations
∂λ ri = ∂λi H ◦ r, (3.8.28)
∂λ riλ = −∂i H ◦ r. (3.8.29)
Every integral section J 1 r = γ◦r of a Hamiltonian connection γ associated with a

Hamiltonian form H is a classical solution of the corresponding Hamilton equations.
Conversely, if r is a global solution of the Hamilton equations (3.8.28) – (3.8.29) for
a Hamiltonian form H, there exists an extension of this solution J 1 r : r(X) → J 1 Π
to a Hamiltonian connection which has r as an integral section.
Substituting J 1 r in (3.8.27), we obtain the identity
J 1 (πΠY ◦ r) = H
c◦r
for every classical solution r of the Hamilton equations (3.8.28 – (3.8.29).

Remark 3.8.4. Note that the Hamilton equations (3.8.28) – (3.8.29) can be in-
troduced without appealing to the Hamilton operator. They are equivalent to the
relation
r∗ (ucdH) = 0
which is assumed to hold for any vertical vector field u on the Legendre bundle
Π → X. •
Chapter 4
Lagrangian time-dependent
mechanics
This Chapter is devoted to the formulation of time-dependent mechanics on a phase

space of coordinates and velocities. This phase space is the first order jet manifold
J 1 Q of a configuration bundle Q → R. The familiar case of the direct product
Q = R × M corresponds to the choice of a certain reference frame. For the sake
of brevity, the above-mentioned formulation is called Lagrangian time-dependent
mechanics although equations of motion are not necessarily Lagrange equations.
Connections provide the main ingredients in the Lagrangian formulation of time-
dependent mechanics. These are the reference frames seen as connections on a
configuration bundle Q → R, the dynamic equations represented by connections
on the jet bundle J 1 Q → R, the dynamic connections on the affine jet bundle
J 1 Q → Q, the connections on the tangent bundles T Q → Q and T J 1 Q → J 1 Q,
and the Lagrangian connections. For instance, we show that every non-relativistic
dynamic equation on a configuration space Q can be seen as a geodesic equation
with respect to some connection on the tangent bundle T Q → Q.
Reference frames are necessarily involved when we deal with dynamic equations,
forces, accelerations, and conservation laws in time-dependent mechanics.
The mass metric is another interesting object of Lagrangian and Newtonian
systems, of the inverse problem, and of systems with holonomic and non-holonomic
constraints.
Throughout this Chapter,
π:Q→R (4.0.1)
153
154 CHAPTER 4. LAGRANGIAN TIME-DEPENDENT MECHANICS
is a fibre bundle whose typical fibre M is an m-dimensional manifold, while its base
R is parameterized by the Cartesian coordinates t possessing the transition functions
t0 = t+ const. In the universal unit system, the physical dimension of t is equal to
[length]. A fibre bundle Q → R is endowed with bundle coordinates (t, q i ). For the
sake of convenience, we also use the compact notation q λ where q 0 = t.
4.1 Fibre bundles over R

In this Section, we point out the most important peculiarities of fibre bundles over
R, which we refer to in the sequel.
1.
The base R of a fibre bundle (4.0.1) is provided with the standard vector field ∂t
and the standard 1-form dt which are invariant under the coordinate transformations
t0 = t+ const. The same symbol dt also stands for any pull-back of the standard 1-
form dt onto a fibre bundle over R. Note on one-to-one correspondence between the
vector fields f ∂t and the real functions f on R. The similar correspondence between
the horizontal densities φdt and the real functions φ on a fibre bundle Q → R
takes place. Roughly speaking, we may neglect the contribution of T R and T ∗ R to
several expressions. At the same time, one should be careful with such simplification
in the framework of the universal unit system. For instance, the coefficient φ of a
horizontal density φdt has the physical dimension [length]−1 , whereas a function φ
is physically dimensionless.
2.
Since R is contractible, any fibre bundle over R is obviously trivial. Different
trivializations
ψ:Q∼
=R×M (4.1.1)
differ from each other in the fibrations Q → M , while the fibration Q → R is once
for all.
Let J 1 Q be the first order jet manifold of a fibre bundle Q → R (4.0.1). It is
provided with the adapted coordinates (t, q i , qti ). Given a direct product
Q=R×M →R
4.1. FIBRE BUNDLES OVER R 155
coordinated by (t, q i ), there is the canonical isomorphism
J 1 (R × M ) = R × T M, (4.1.2)
i
q it = q̇ ,
that one can justify by inspection of the transition functions of the coordinates q it and
i
q̇ , when the transition functions of q i are independent of t. Due to the isomorphism
(4.1.2), every trivialization (4.1.1) yields the corresponding trivialization of the jet
manifold
J 1Q ∼
= R × T M. (4.1.3)
3.
Given the jet manifold J 1 Q of a fibre bundle Q → R, the canonical imbedding
(1.3.4) takes the form
λ : J 1 Q ,→ T Q, (4.1.4)
λ : (t, q i , qti ) 7→ (t, q i , ṫ = 1, q̇ i = qti ).
For brevity, we will write
λ = dt = ∂t + qti ∂i , (4.1.5)
where by dt is meant the total derivative. ¿From now on, we will identify the jet
manifold J 1 Q with its image in T Q.
Remark 4.1.1. Following precisely the expression (1.3.4), we should write the
morphism λ (4.1.5) in the form
λ = dt ⊗ (∂t + qti ∂i ). (4.1.6)
With respect to the universal unit system, the physical dimension of λ (4.1.5) is
[length]−1 , while λ (4.1.6) is dimensionless. •
Using the morphism (4.1.4), one can define the contraction
J 1 Q × T ∗ Q → Q × R,
Q Q
i
(qt ; ṫ, q̇i ) 7→ λc(ṫdt + q̇i dq i ) = ṫ + qti q̇i , (4.1.7)
where (t, q i , ṫ, q̇i ) are the coordinates on the cotangent bundle T ∗ Q.
4.
A glance at the expression (4.1.4) shows that the affine jet bundle J 1 Q → Q is
modelled over the vertical tangent bundle V Q of the fibre bundle of Q → R. As a
consequence, we have the following canonical splitting (1.1.6) of the vertical tangent
bundle VQ J 1 Q of the affine jet bundle J 1 Q → Q:
α : VQ J 1 Q ∼
= J 1 Q × V Q, (4.1.8)
Q
α(∂it ) = ∂i ,
together with the corresponding splitting of the vertical cotangent bundle VQ∗ J 1 Q
of J 1 Q → Q:
α∗ : VQ∗ J 1 Q ∼
= J 1 Q × V ∗ Q, (4.1.9)
Q
α∗ (dqti ) = dq i ,
where dqti and dq i are the holonomic bases for VQ∗ J 1 Q and V ∗ Q, respectively. Then
the exact sequence (1.6.9) of vertical bundles over the composite fibre bundle
J 1Q → Q → R (4.1.10)
reads
α−1
?
i π
0 −→ VQ J 1 Q ,→ V J 1 Q −→
V
J 1 Q × V Q −→ 0.
Q
Hence, we obtain the following linear endomorphism over J 1 Q of the vertical tangent
bundle V J 1 Q of the jet bundle J 1 Q → R:
def
vb = i ◦ α−1 ◦ πV : V J 1 Q → V J 1 Q, (4.1.11)
vb(∂i ) = ∂it , vb(∂it ) = 0.
This endomorphism obeys the nilpotency rule
vb ◦ vb = 0. (4.1.12)
Combining the horizontal splitting (1.3.11), the corresponding projection
pr2 : J 1 Q × T Q → J 1 Q × V Q ∼
= VQ J 1 Q,
Q Q
i t t
∂t 7→ −qt ∂i , ∂i 7→ ∂i ,
and the inclusion V J 1 Q ,→ T J 1 Q, one can extend the endomorphism (4.1.11) to

the tangent bundle T J 1 Q:
vb : T J 1 Q → T J 1 Q,
vb(∂t ) = −qti ∂it , vb(∂i ) = ∂it , vb(∂it ) = 0. (4.1.13)
This is called the vertical endomorphism. It inherits the nilpotency property (4.1.12).
The transpose of the vertical endomorphism vb (4.1.13) is
vb∗ : T ∗ J 1 Q → T ∗ J 1 Q,
vb∗ (dt) = 0, vb∗ (dq i ) = 0, vb∗ (dqti ) = θi , (4.1.14)
where
θi = dq i − qti dt
are the contact forms (1.3.6). The nilpotency rule
vb∗ ◦ vb∗ = 0
is also fulfilled. The homomorphisms vb and vb∗ are associated with the tangent-valued
1-form
vb = θi ⊗ ∂it
in accordance with the relations (1.2.37) – (1.2.38).

With the endomorphism vb∗ , one can introduce the vertical exterior differential
dv = vb∗ ◦ d,
acting on the algebra O∗ (J 1 Q) of exterior forms on the jet manifold J 1 Q. For

example, if f is a function on J 1 Q, we have
dv f = ∂it f θi .
5.
In view of the morphism λ (4.1.4), any connection
Γ = dt ⊗ (∂t + Γi ∂i ) (4.1.15)
on a fibre bundle Q → R can be identified with a nowhere vanishing horizontal

vector field
Γ = ∂t + Γi ∂i (4.1.16)
on Q which is the horizontal lift Γ∂t (1.4.7) of the standard vector field ∂t on R
by means of the connection (4.1.15). Conversely, any vector field Γ on Q such that
dtcΓ = 1 defines a connection on Q → R. Of course, the integral curves
c : R ⊃ () → Q,
ṫ(τ ) = 1, ċi (τ ) = Γi ◦ c(τ ), τ ∈ (),
of the vector field (4.1.16) coincide with the integral sections
∂t ci (t) = (Γ ◦ c)i (t)
of the connection (4.1.15).

Connections on a fibre bundle Q → R constitute an affine space modelled over
the vector space of vertical vector fields on Q → R. Accordingly, the covariant
differential (1.4.5) associated with a connection Γ on Q → R takes its values into
the vertical tangent bundle V Q of Q → R:
DΓ : J 1 Q → V Q, (4.1.17)
Q
q̇ i ◦ DΓ = qti − Γi .
A connection Γ on a fibre bundle Q → R is obviously flat. It yields a horizontal

distribution on Q. The integral manifolds of this distribution are integral curves
of the vector field (4.1.16) which are transversal to the fibres of the fibre bundle
Q → R.
Corollary 4.1.1. By virtue of Proposition 1.4.1, each connection Γ on a fibre

bundle Q → R defines an atlas of local constant trivializations of Q → R such
that the associated bundle coordinates (t, q i ) on Q possess the transition function
q i → q 0i (q j ) independent of t, and
Γ = ∂t (4.1.18)
with respect to these coordinates. Conversely, every atlas of local constant trivial-
izations of the fibre bundle Q → R determines a connection on Q → R which is
equal to (4.1.18) relative to this atlas. 2
A connection Γ on a fibre bundle Q → R is said to be complete if the horizontal

vector field (4.1.16) is complete.
Proposition 4.1.2. Every trivialization of a fibre bundle Q → R yields a complete

connection on this fibre bundle. Conversely, every complete connection Γ on Q → R
defines its trivialization (4.1.1) such that the vector field (4.1.16) equals ∂t relative
to the bundle coordinates associated with this trivialization. 2
Proof. Every trivialization of Q → R defines the horizontal lift of the standard

vector field ∂t on R to the vector field Γ = ∂t on Q which is obviously complete.
Then the corresponding connection on Q → R is also complete. Conversely, let Γ
be a complete connection on a fibre bundle Q → R. The horizontal vector field Γ
(4.1.16) is the generator of a 1-parameter group GΓ acting freely on Q. The orbits
of this action are integral curves of the vector field Γ. Hence, we obtain a projection
Q → Q/GΓ = M (4.1.19)
of Q along the above-mentioned integral curves onto some fibre of Q, e.g., Qt=0 ∼
=
M . Combining the projection (4.1.19) and the projection Q → R gives a desired
trivialization (4.1.1) of Q. QED
6.
Let J 1 J 1 Q be the repeated jet manifold of a fibre bundle Q → R, provided with
the adapted coordinates
(t, q i , qti , q(t)

i
, qtti ).
For a fibre bundle Q → R, we have the canonical isomorphism k between the affine
fibrations π11 (1.3.13) and J 1 π01 (1.3.15) of J 1 J 1 Q over J 1 Q, i.e.,
π11 ◦ k = J01 π01 , k ◦ k = Id J 1 J 1 Q,
where
qti ◦ k = q(t)
i
, i
q(t) ◦ k = qti , qtti ◦ k = qtti . (4.1.20)
In particular, the affine bundle π11 (1.3.13) is modelled over the vertical tangent
bundle V J 1 Q of J 1 Q → R which is canonically isomorphic to the underlying vector
bundle J 1 V Q → J 1 Q of the affine bundle J 1 π01 (1.3.15).
By JQ1 J 1 Q throughout is meant the first order jet manifold of the affine jet bundle
J 1 Q → Q. The adapted coordinates on JQ1 J 1 Q are (q λ , qti , qλt
i
).
For a fibre bundle Q → R, the sesquiholonomic jet manifold Jb2 Q coincides with
the second order jet manifold J 2 Q, coordinated by
(t, q i , qti , qtti ).
The affine bundle J 2 Q → J 1 Q is modelled over the vertical tangent bundle
VQ J 1 Q ∼
= J 1Q × V Q → J 1Q (4.1.21)
Q
of the affine jet bundle J 1 Q → Q. There are the imbeddings

λ2 Tλ
J 2 Q ,→ T J 1 Q ,→ VQ T Q ∼
= T 2 Q ⊂ T T Q,
λ2 : (t, q i , qti , qtti ) 7→ (t, q i , qti , ṫ = 1, q̇ i = qti , q̇ti = qtti ), (4.1.22)
i
T λ ◦ λ2 : (t, q , qti , qtti ) i i i
7→ (t, q , ṫ = ṫ = 1, q̇ = q̇ = qti , ẗ i
= 0, q̈ = qtti ), (4.1.23)
where
(t, q i , ṫ, q̇ i , ṫ, q̇i , ẗ, q̈ i )
are the coordinates on the double tangent bundle T T Q, by VQ T Q is meant the

vertical tangent bundle of T Q → Q, and T 2 Q ⊂ T T Q is a second order tangent
space, given by the coordinate relation ṫ = ṫ.
Due to the morphism (4.1.22), any connection ξ on the jet bundle J 1 Q → R is
represented by a horizontal vector field on J 1 Q such that ξcdt = 1.
Example 4.1.2. A connection on the jet bundle J 1 Q → R is defined as a section of
the affine bundle π11 (1.3.13). A connection Γ (4.1.16) on a fibre bundle Q → R has
the jet prolongation to the section J 1 Γ of the affine bundle J 1 π01 . By virtue of the
isomorphism k (4.1.20), every connection Γ on Q → R gives rise to the connection
def
JΓ = k ◦ J 1 Γ : J 1 Q → J 1 J 1 Q,
JΓ = ∂t + Γi ∂i + dt Γi ∂it , (4.1.24)
on the jet bundle J 1 Q → R. •
A connection on the jet bundle J 1 Q → R is said to be holonomic if it is a section
ξ = dt ⊗ (∂t + qti ∂i + ξ i ∂it )

of the holonomic subbundle J 2 Q → J 1 Q of J 1 J 1 Q → J 1 Q. In view of the morphism

(4.1.22), a holonomic connection is represented by a horizontal vector field
ξ = ∂t + qti ∂i + ξ i ∂it (4.1.25)

on J 1 Q. Conversely, every vector field ξ on J 1 Q which fulfills the conditions
dtcξ = 1, vb(ξ) = 0,
where vb is the vertical endomorphism (4.1.13), is a holonomic connection on the jet
bundle J 1 Q → R.
Holonomic connections (4.1.25) make up an affine space modelled over the linear
space of vertical vector fields on the affine jet bundle J 1 Q → Q, i.e., which live in
VQ J 1 Q.
A holonomic connection ξ defines the corresponding covariant differential (4.1.17)
on the jet manifold J 1 Q:
Dξ : J 1 J 1 Q −→
1
VQ J 1 Q ⊂ V J 1 Q,
J Q
i
q̇ ◦ Dξ = 0, q̇ti ◦ Dξ = qtti − ξ i ,
which takes its values into the vertical tangent bundle VQ J 1 Q of the jet bundle
J 1 Q → Q. Then, by virtue of Proposition 1.3.1, any integral section c : () → J 1 Q
for a holonomic connection ξ is holonomic, i.e., c = ċ where c is a curve in Q.
4.2 Dynamic equations

We start our exposition of Lagrangian time-dependent mechanics ¿from the notion
of a second order dynamic equation on a configuration bundle Q → R.
Recall that a dynamic equation, by definition, is a differential equation which
can be algebraically solved for the highest order derivatives.
Definition 4.2.1. Let

Γ = ∂t + Γi ∂i
be a connection on a fibre bundle Q → R. The corresponding covariant differential
DΓ (4.1.17) is a first order differential operator on Q. Its kernel, given by the
coordinate relations
qti = Γi (t, q i ), (4.2.1)
is a closed subbundle of the jet bundle J 1 Q → R. Therefore, in accordance with

Definition 1.3.3, this is a first order differential equation on the fibre bundle Q → R,
called the first order dynamic equation on Q → R. 2
Solutions of the first order dynamic equation (4.2.1) are integral sections (geodesic
curves) for the connection Γ.
Definition 4.2.2. Let us consider the first order dynamic equation (4.2.1) on the
jet bundle J 1 Q → R, which is associated with a holonomic connection ξ (4.1.25)
on J 1 Q → R. This is a closed subbundle of the second order jet bundle J 2 Q → R,
given by the coordinate relations
qtti = ξ i (t, q j , qtj ). (4.2.2)
Consequently, it is a second order differential equation on the fibre bundle Q → R

in accordance with Definition 1.3.3. This equation is called a second order dynamic
equation, or simply a dynamic equation, if there is no danger of confusion. The
corresponding horizontal vector field ξ (4.1.25) is also termed a dynamic equation.
2
A solution of the dynamic equation (4.2.2), called a motion, is a curve c in Q

whose second order jet prolongation c̈ lives in (4.2.2). It is clear that any integral
section c for the holonomic connection ξ is the jet prolongation ċ of a solution c of
the dynamic equation (4.2.2), i.e.,
c̈i = ξ i ◦ ċ, (4.2.3)
and vice versa.

Remark 4.2.1. By very definition, the second order dynamic equation (4.2.2) on a
fibre bundle Q → R is equivalent to the system of first order differential equations
i
q(t) = qti , qtti = ξ i (t, q j , qtj ), (4.2.4)
on the jet bundle J 1 Q → R. Any solution c of these equations takes its values into
J 2 Q and, by virtue of Proposition 1.3.1, is holonomic, i.e., c = ċ. Therefore, the
equations (4.2.2) and (4.2.4) are equivalent. The equation (4.2.4) is said to be the
first order reduction of the second order dynamic equation (4.2.2) . •
One can easily find the transformation law of a second order dynamic equation
under bundle coordinate transformations q i → q 0i (t, q j ). This transformation law
reads
qtt0i = ξ 0i , ξ 0i = (ξ j ∂j + qtj qtk ∂j ∂k + 2qtj ∂j ∂t + ∂t2 )q 0i (t, q j ). (4.2.5)
Example 4.2.2. From the physical viewpoint, the most interesting dynamic equa-
tions are the quadratic dynamic equations, i.e., those which are polynomials in the
coordinates qti of degree ≤ 2. This property is global due to the transformation law
(4.2.5). The dynamic equation of a relativistic particle exemplifies a non-polynomial
dynamic equation. •
A dynamic equation ξ on a fibre bundle Q → R is said to be conservative if

there exists a trivialization (4.1.1) of Q whose corresponding trivialization (4.1.3)
of J 1 Q is such that the vector field ξ (4.1.25) on J 1 Q is projectable over M . Then
this projection
Ξξ = q̇ i ∂i + ξ i (q j , q̇ j )∂˙i
is a second order dynamic equation on the typical fibre M of Q → R in accordance
with Definition 3.1.2. Conversely, every autonomous second order dynamic equation
Ξ on a manifold M can be seen as a conservative dynamic equation
ξΞ = ∂t + q̇ i ∂i + ui ∂˙i (4.2.6)
on the fibre bundle R × M → R in accordance with the isomorphism (4.1.3).
We conclude this Section with the following theorem.
Theorem 4.2.3. Any dynamic equation on a fibre bundle Q → R is equivalent to

an autonomous second order dynamic equation on a manifold Q. 2
Proof. Given a dynamic equation ξ on a fibre bundle Q → R, let us consider the

diagram
J 2 Q −→ T 2 Q
ξ 6 6 Ξ (4.2.7)
λ
J 1 Q −→ T Q
where Ξ is a holonomic vector field on the tangent bundle T Q, and we use the
morphism (4.1.23). A glance at the expression (4.1.23) shows that the diagram
(4.2.7) can be commutative only if the component Ξ0 of a vector field Ξ vanishes.

Since the transition functions t → t0 are independent of q i , such a vector field
may exist on T Q. Now, the diagram (4.2.7) becomes commutative if the dynamic
equation ξ and a vector field Ξ fulfill the relation
ξ i = Ξi (t, q j , ṫ = 1, q̇ j = qtj ). (4.2.8)
It is easily seen that this relation holds globally because the substitution of q̇ i = qti
in the transformation law of a vector field Ξ restates the transformation law (4.2.5)
of the holonomic connection ξ. In accordance with the relation (4.2.8), the desired
vector field Ξ is an extension of the section T λ◦λ2 ◦ξ of the fibre bundle T 2 Q → T Q
over the closed submanifold J 1 Q ⊂ T Q to a global section. Such an extension always
exists by virtue of Theorem 1.1.2, but is not unique. Then the dynamic equation
(4.2.2) can be written in the form
qtti = Ξi |ṫ=1,q̇j =qj . (4.2.9)

t
This is equivalent to the autonomous second order dynamic equation
ẗ = 0, ṫ = 1, q̈ i = Ξi , (4.2.10)
on Q. Being a solution of the equation (4.2.10), a curve c in Q also satisfies the

equation (4.2.9), and vice versa. QED
It should be emphasized that, written in the bundle coordinates (t, q i ), the second
order equation (4.2.10) is well defined with respect to any coordinates on Q.
4.3 Dynamic connections

In order to say more than was said in Theorem 4.2.3, we will consider the relationship
between the holonomic connections on the jet bundle J 1 Q → R and the connections
on the affine jet bundle J 1 Q → Q (see Propositions 4.3.1 and 4.3.2 below).
Let
γ : J 1 Q → JQ1 J 1 Q
be a connection on the affine jet bundle J 1 Q → Q. It takes the coordinate form
γ = dq λ ⊗ (∂λ + γλi ∂it ), (4.3.1)

4.3. DYNAMIC CONNECTIONS 165
with the transformation law

∂q µ
γλ0i = (∂j q 0i γµj + ∂µ qt0i ) . (4.3.2)
∂q 0λ
Remark 4.3.1. In view of the canonical splitting (4.1.8), the curvature (1.4.8) of
a connection γ (4.3.1) reads
2
R : J 1 Q → ∧ T ∗ Q ⊗ V Q,
J 1Q
1 i 1 i k
R = Rλµ dq λ ∧ dq µ ⊗ ∂i = ( Rkj dq ∧ dq j + R0j
i
dt ∧ dq j ) ⊗ ∂i ,
2 2
i
Rλµ = ∂λ γµi − ∂µ γλi + γλj ∂j γµi − γµj ∂j γλi . (4.3.3)
Using the contraction (4.1.7), we obtain the soldering form

i k i i j
λcR = [(Rkj qt + R0j )dq j − R0j qt dt] ⊗ ∂i
on the affine jet bundle J 1 Q → Q. Its image by the canonical projection T ∗ Q →

V ∗ Q is the tensor field
R : J 1 Q → V ∗ Q ⊗ V Q,
Q
i k i
R= (Rkj qt + R0j )dq j ⊗ ∂i , (4.3.4)
and then we come to the scalar field

e : J 1 Q → R,
R
e = Ri q k + Ri ,
R (4.3.5)
ki t 0i
on the jet manifold J 1 Q. •
Proposition 4.3.1. Any connection γ (4.3.1) on the affine jet bundle J 1 Q → Q

defines the holonomic connection
ξγ = ρ ◦ γ : J 1 Q → JQ1 J 1 Q → J 2 Q, (4.3.6)
ξγ = ∂t + qti ∂i + (γ0i + qtj γji )∂it ,
on the jet bundle J 1 Q → R. 2

Proof. Let us consider the composite fibre bundle (4.1.10) and the morphism ρ
(1.6.6) which reads
i
ρ : JQ1 J 1 Q 3 (q λ , qti , qλt ) 7→ (q λ , qti , q(t)
i i
= qti , qtti = q0t + qtj qjt
i
) ∈ J 2 Q. (4.3.7)
A connection γ (4.3.1) and the morphism ρ (4.3.7) combine into a desired holonomic
connection ξγ (4.3.6) on the jet bundle J 1 Q → R. QED
It follows that every connection γ (4.3.1) on the affine jet bundle J 1 Q → Q

yields the dynamic equation
qtti = γ0i + qtj γji (4.3.8)
on the configuration bundle Q → R. This is precisely the restriction to J 2 Q of

the kernel Ker D
f of the vertical covariant differential D
γ
f (1.6.13) defined by the
γ
connection γ:
f : J 1 J 1 Q → V J 1 Q,
Dγ Q
f = qi − γ i − qj γ i .
q̇ti ◦ D (4.3.9)
γ tt 0 t j
Therefore, connections on the jet bundle J 1 Q → Q are called the dynamic connec-
tions. The corresponding equation (4.2.3) can be written in the form
c̈i = ρ ◦ γ ◦ ċ,
where ρ is the morphism (4.3.7).

Of course, different dynamic connections can lead to the same dynamic equation
(4.3.8).
Proposition 4.3.2. Any holonomic connection ξ (4.1.25) on the jet bundle J 1 Q →

R defines the dynamic connection
γξ = dt ⊗ [∂t + (ξ i − 21 qtj ∂jt ξ i )∂it ] + dq j ⊗ [∂j + 12 ∂jt ξ i ∂it ] (4.3.10)
on the affine jet bundle J 1 Q → Q. 2
Proof. Let ξ be a holonomic vector field (4.1.25). Given an arbitrary vertical vector
field
u = ai ∂i + bi ∂it
on the jet bundle J 1 Q → R, let us put

def
Iξ (u) =[ξ, vb(u)] − vb([ξ, u]),
Iξ (u) = −ai ∂i + (bi − aj ∂jt ξ i )∂it ,
where vb is the vertical endomorphism (4.1.13). Then, there is the endomorphism
Iξ : V J 1 Q →
1
V J 1 Q,
J Q
i
Iξ : q̇ ∂i + q̇ti ∂it 7→ −q̇ i ∂i + (q̇ti − q̇ j ∂jt ξ i )∂it .
It is readily observed that this endomorphism obeys the condition
Iξ ◦ Iξ = Iξ .
Then, one can construct the projection

def 1
Jξ = (Iξ + Id V J 1 Q) : V J 1 Q → VQ J 1 Q,
2 J 1Q
1
Jξ : q̇ i ∂i + q̇ti ∂it 7→ (q̇ti − q̇ j ∂jt ξ i )∂it .
2
Recall that a holonomic connection ξ on J 1 Q → R defines the projection (1.4.3)
which reads
ξ : T J 1 Q 3 ṫ∂t + q̇ i ∂i + q̇ti ∂it 7→ (q̇ i − ṫqti )∂i + (q̇ti − ṫξ i )∂it ∈ V J 1 Q.
Let us consider the composition of morphisms
Iξ ◦ ξ : T J 1 Q → V J 1 Q → VQ J 1 Q,
1 1
ṫ∂t + q̇ i ∂i + q̇ti ∂it 7→ [q̇ti − ṫ(ξ i − qtj ∂jt ξ i ) − q̇ j ∂jt ξ i ]∂it .
2 2
This corresponds to the connection γξ (4.3.10) on the affine jet bundle J 1 Q → Q.
QED
It is readily observed that the dynamic connection γξ (4.3.10), defined by a

dynamic equation, possesses the property
γik = ∂it γ0k + qtj ∂it γjk (4.3.11)
which implies the relation
∂jt γik = ∂it γjk .

Therefore, a dynamic connection γ, obeying the condition (4.3.11), is said to be

symmetric. The torsion of a dynamic connection γ is defined as the tensor field
T : J 1 Q → V ∗ Q ⊗ V Q,
Q
T = Tik dq i ⊗ ∂k ,
Tik = γik − ∂it γ0k − qtj ∂it γjk . (4.3.12)
It follows at once that a dynamic connection is symmetric if and only if its torsion
vanishes.
Let γ be a dynamic connection (4.3.1) and ξγ the corresponding dynamic equa-
tion (4.3.6). Then the dynamic connection (4.3.10) associated with the dynamic
equation ξγ takes the form
1
γξγ ki = (γik + ∂it γ0k + qtj ∂it γjk ), γξγ k0 = ξ k − qti γξγ ki .
2
It is readily observed that γ = γξγ if and only if the torsion T (4.3.12) of the dynamic
connection γ vanishes.
Example 4.3.2. Since the jet bundle J 1 Q → Q is affine, it admits an affine
connection
γ = dq λ ⊗ [∂λ + (γλ0
i
(q µ ) + γλj
i
(q µ )qtj )∂it ]. (4.3.13)
i i
This connection is symmetric if and only if γλµ = γµλ . One can easily justify that an
affine dynamic connection generates a quadratic dynamic equation, and vice versa.
Nevertheless, a non-affine dynamic connection, whose symmetric part is affine, also
define a quadratic dynamic equation.
An affine connection (4.3.13) on the affine jet bundle J 1 Q → Q yields the linear
connection
γ = dq λ ⊗ [∂λ + γλj
i
(q µ )q̇tj ∂˙i ]
on the vertical tangent bundle V Q → Q. •
Using the notion of a dynamic connection, we may modify Theorem 3.1.4 as

follows. Let Ξ be an autonomous second order dynamic equation on a manifold
M , and ξΞ (4.2.6) the corresponding conservative dynamic equation on the bundle
R×M → R. The latter yields the dynamic connection γ (4.3.10) on the fibre bundle
R × T M → R × M.
Its components γji are exactly those of the connection (3.1.7) on the tangent bundle
T M → M in Theorem 3.1.4, while γ0i make up a vertical vector field
1
e = γ0i ∂˙i = (Ξi − q̇ j ∂˙j Ξi )∂˙i (4.3.14)
2
on T M → M . Thus, we have shown the following.
Proposition 4.3.3. Every autonomous second order dynamic equation Ξ (3.1.4)

on a manifold M admits the decomposition
Ξi = Kji q̇ j + ei
where K is the connection (3.1.7) on the tangent bundle T M → M , and e is the

vertical vector field (4.3.14) on T M → M . 2
Remark 4.3.3. With a dynamic connection γξ (4.3.10), we also restate the well-
known linear connection on the tangent bundle T J 1 Q → J 1 Q, associated with a
dynamic equation ξ on Q [132], and can write it with respect to holonomic bases.
Given a holonomic connection ξ (4.1.25), we have the corresponding horizontal
splitting (1.4.1) of the tangent bundle T J 1 Q of J 1 Q:
T J 1 Q = Hξ ⊕ V J 1 Q, (4.3.15)
J 1Q
ṫ∂t + q̇ i ∂i + q̇ti ∂it = ṫξ + (q̇ i − ṫqti )∂i + (q̇ti − ξ i )∂it , (4.3.16)
and the corresponding horizontal splitting (1.4.2) of the cotangent bundle T ∗ J 1 Q of

J 1 Q:
T ∗ J 1 Q = R ⊕ V ∗ J 1 Q,
J 1Q
i
ṫdt + q̇i dq + q̇it dqti = (ṫ + qti q̇i + ξ i q̇it )dt + q̇i θi + q̇it (dqti − ξ i dt), (4.3.17)
Using these splittings, one can introduce the non-holonomic bases (ξ, ∂i , ∂it ) for the
tangent bundle T J 1 Q, and the non-holonomic bases
(dt, θi , dqti − ξ i dt)
for the cotangent bundle T ∗ J 1 Q of J 1 Q.

Furthermore, the dynamic connection γξ (4.3.10), associated with the dynamic

equation ξ, provides the corresponding splitting (1.6.11) of the vertical tangent
bundle V J 1 Q of J 1 Q:
V J 1 Q = VQ J 1 Q ⊕ Hγξ ,
J 1Q
1 1
q̇ i ∂i + q̇ti ∂it = (q̇ti − q̇ j ∂jt ξ i )∂it + q̇ j (∂j + ∂jt ξ i ∂it ). (4.3.18)
2 2
The splittings (4.3.16) and (4.3.18) combine into the splitting
T J 1 Q = Hξ ⊕ Hγξ ⊕ VQ J 1 Q (4.3.19)
J 1Q J 1Q
together with the non-holonomic bases

1
(ξ, hj = ∂j + ∂jt ξ i ∂it , ∂it ) (4.3.20)
2
for the tangent bundle T J 1 Q [55]. Note that the fibre bundle Hγξ in the splitting
(4.3.19), like VQ J 1 Q, is isomorphic to the pull-back
H γξ ∼
= J 1 Q × V Q → J 1 Q, hj ←→ ∂j . (4.3.21)
Q
Accordingly, the dynamic connection γξ (4.3.10) determines the corresponding

splitting (1.6.12) of the vertical cotangent bundle V ∗ J 1 Q of J 1 Q. Combining this
splitting with the decomposition (4.3.17) gives the splitting
T ∗ J 1 Q = R ⊕ V ∗ Q ⊕ VQ∗ J 1 Q
J 1Q J 1Q
together with the non-holonomic bases
(dt, θi , dqti − γαi dq λ )
for the cotangent bundle T ∗ J 1 Q of J 1 Q.

Every connection γ (4.3.1) on the affine jet bundle J 1 Q → Q gives rise to the
connection V γ (1.6.14) on the composite vertical tangent bundle
VQ J 1 Q → J 1 Q → Q.
This connection reads
V γ = dq λ ⊗ (∂λ + γλi ∂it + ∂jt γλi q̇tj ∂˙it ).

Since J 1 Q → Q is an affine bundle, the connection V γ, in turn, can be seen as the

composite connection (1.6.8) generated by the connection γ on the affine jet bundle
J 1 Q → Q, and the linear connection (1.6.16):
γe = dq λ ⊗ (∂λ + ∂jt γλi q̇tj ∂˙it ) + dqtk ⊗ ∂kt , (4.3.22)
on the vertical tangent bundle VQ J 1 Q → J 1 Q. Due to the splitting (4.1.8), the
connection γe (4.3.22) in fact is a linear connection
γe = dq λ ⊗ (∂λ + ∂jt γλi q̇ j ∂˙i ) + dqtk ⊗ ∂˙k (4.3.23)
on the pull-back (4.3.21). In particular, if γ = γξ is the connection (4.3.10), the
connection γeξ (4.3.23) yields a connection on the pull-back bundle Hγξ → J 1 Q in
the splitting (4.3.19).
Due to the splitting (4.3.19), the tangent bundle T J 1 Q → J 1 Q can be provided
with a linear connection A [132] which is the direct sum of
• the trivial connection on the linear bundle Hξ → J 1 Q,
• the connection γeξ (4.3.23) on the pull-back bundle Hγξ → J 1 Q,
• and the similar connection (4.3.22) on the vertical tangent bundle VQ J 1 Q →

J 1 Q.
With respect to the non-holonomic basis (4.3.20), where
ṫ∂t + q̇ i ∂i + q̇ti ∂it = ṫξ + ui hi + v i ∂it , (4.3.24)
the above-mentioned linear connection A reads
!
λ i j ∂ i j ∂
A = dq ⊗ ∂λ + ∂j γλ u + ∂j γλ v + dqti ⊗ ∂it . (4.3.25)
∂ui ∂v i
Recall the expression (4.3.10) for the components γλi of the dynamic connection γξ .
Substituting
q̇ i = ṫqti + ui , q̇ti = ṫξ i + uk γki + v i
¿from (4.3.24) in (4.3.25), we obtain the linear connection A written with respect
to the holonomic bases (∂λ , ∂it ) for T J 1 Q:
A = dq λ ⊗ (∂λ + Aλ ) + dqti ⊗ (∂it + Ati ),
Aλ = ṫ[−qtj ∂j γλi ∂˙i + (∂λ ξ i − qtk ∂λ γki − qtj γki ∂j γλk + qtk γkj ∂j γλi − ξ j ∂j γλi )∂˙it ] +
q̇ j [∂j γ i ∂˙i + (∂λ γ i + γ i ∂j γ k − γ k ∂k γ i )∂˙ t ] + q̇tj ∂j γ i ∂˙ t ,
λ j k λ j λ i λ i
Ati = ṫ[∂˙i + (∂it ξ j − k t j ˙t
qt ∂i γk )∂j ] + j t k ˙t
q̇ ∂i γj ∂k .
The covariant derivatives ∇ relative to the connection A fulfill the conditions
∇λ (ξ) = ∇ti (ξ) = 0, (4.3.26)

∇λ (∂jt ) = −∂jt γλk ∂kt , ∇λ (hj ) = −∂jt γλk hk ,
∇ti (∂jt ) = ∇ti (hj ) = 0, ∇ti (∂j ) = −∂it γjk ∂kt .
In particular, by virtue of the equality (4.3.26), every solution c of the dynamic

equation ξ is also a geodesic of the linear connection A. •
4.4 Non-relativistic geodesic equations

In this Section, we aim to show that every dynamic equation on a configuration
bundle Q → R is equivalent to a geodesic equation on the tangent bundle T Q → Q.
We start from the relation between the dynamic connections γ on the affine jet
bundle J 1 Q → Q and the connections
K = dq λ ⊗ (∂λ + Kλµ ∂˙µ ) (4.4.1)
on the tangent bundle T Q → Q of the configuration space Q. We will use the

notation (1.2.4).
Let us consider the diagram
J 1λ
JQ1 J 1 Q −→ JQ1 T Q
γ 6 6 K (4.4.2)
λ
J 1Q −→ T Q
where JQ1 T Q is the first order jet manifold of the tangent bundle T Q → Q, coordi-
nated by
(t, q i , ṫ, q̇ i , (ṫ)µ , (q̇ i )µ ).
The jet prolongation over Q of the canonical imbedding λ (4.1.4) reads
J 1 λ : (t, q i , qti , qµt

i
) 7→ (t, q i , ṫ = 1, q̇ i = qti , (ṫ)µ = 0, (q̇ i )µ = qµt
i
).
Then we have
J 1 λ ◦ γ : (t, q i , qti ) 7→ (t, q i , ṫ = 1, q̇ i = qti , (ṫ)µ = 0, (q̇ i )µ = γµi ),

K ◦ λ : (t, q i , qti ) 7→ (t, q i , ṫ = 1, q̇ i = qti , (ṫ)µ = Kµ0 , (q̇ i )µ = Kµi ).
4.4. NON-RELATIVISTIC GEODESIC EQUATIONS 173
It follows that the diagram (4.4.2) can be commutative only if the components Kµ0
of the connection K (4.4.1) on the tangent bundle T Q → Q vanish.
Since the transition functions t → t0 are independent of q i , a connection
f = dq λ ⊗ (∂ + K i ∂˙ )
K (4.4.3)
λ λ i
with Kµ0 = 0 may exist on the tangent bundle T Q → Q in accordance with the
transformation law
i ∂q µ
K 0 λ = (∂j q 0i Kµj + ∂µ q̇ 0i ) 0λ . (4.4.4)
∂q
Now the diagram (4.4.2) becomes commutative if the connections γ and K
f fulfill
the relation
γµi = Kµi ◦ λ = Kµi (t, q i , ṫ = 1, q̇ i = qti ). (4.4.5)
It is easily seen that this relation holds globally because the substitution of q̇ i = qti
in (4.4.4) restates the transformation law (4.3.2) of a connection on the affine jet
bundle J 1 Q → Q. In accordance with the relation (4.4.5), the desired connection K f
is an extension of the section J 1 λ ◦ γ of the affine jet bundle JQ1 T Q → T Q over the
closed submanifold J 1 Q ⊂ T Q to a global section. Such an extension always exists
by virtue of Theorem 1.1.2, but is not unique. Thus, we have proved the following.
Proposition 4.4.1. In accordance with the relation (4.4.5), every dynamic equa-
tion on a configuration bundle Q → R can be written in the form
qtti = K0i ◦ λ + qtj Kji ◦ λ, (4.4.6)
where K f is a connection (4.4.3) on the tangent bundle T Q → Q. Conversely, each

connection K f (4.4.3) on T Q → Q defines the dynamic connection γ (4.4.5) on the
affine jet bundle J 1 Q → Q and the dynamic equation (4.4.6) on a configuration
bundle Q → R. 2
Then we come to the following theorem.
Theorem 4.4.2. Every dynamic equation (4.2.2) on a configuration bundle Q → R

is equivalent to the geodesic equation
q̈ 0 = 0, q̇ 0 = 1,
q̈ i = Kλi (q µ , q̇ µ )q̇ λ , (4.4.7)
on the tangent bundle T Q relative to a connection K f with the components K 0 = 0

λ
and Kλi (4.4.5). Its solution is a geodesic curve in Q which also obeys the dynamic
equation (4.4.6), and vice versa. 2
In accordance with this theorem, the second order equation (4.2.10) in Theorem
4.2.3 can be chosen as a geodesic equation. It should be emphasized that, written
in the bundle coordinates (t, q i ), the geodesic equation (4.4.7) and the connection
K
f (4.4.5) are well defined with respect to any coordinates on Q.
As was mentioned, from the physical viewpoint, the most interesting dynamic
equations are the quadratic ones
ξ i = aijk (q µ )qtj qtk + bij (q µ )qtj + f i (q µ ). (4.4.8)
This property is global due to the transformation law (4.2.5). Then one can use the
following two facts.
Proposition 4.4.3. There is one-to-one correspondence between the affine con-

nections γ on the affine jet bundle J 1 Q → Q and the linear connections K (4.4.3)
on the tangent bundle T Q → Q. 2
Proof. This correspondence is given by the relation (4.4.5), written in the form
i j
γµi = γµ0
i
+ γµj qt = Kµ i 0 (q ν )ṫ + Kµ i j (q ν )q̇ j |ṫ=1,q̇i =qti = Kµ i 0 (q ν ) + Kµ i j (q ν )qtj ,
i.e.,
i
γµλ = Kµ i λ .
QED
In particular, if an affine dynamic connection γ is symmetric, so is the corre-

sponding linear connection K.
Corollary 4.4.4. Every quadratic dynamic equation (4.4.8) on a configuration

bundle Q → R of non-relativistic mechanics gives rise to the geodesic equation
q̈ 0 = 0, q̇ 0 = 1,
q̈ i = aijk (q µ )q̇ j q̇ k + bij (q µ )q̇ j q̇ 0 + f i (q µ )q̇ 0 q̇ 0 (4.4.9)
4.4. NON-RELATIVISTIC GEODESIC EQUATIONS 175
on the tangent bundle T Q with respect to the symmetric linear connection

1
Kλ 0 ν = 0, K0 i 0 = f i , K0 i j = bij , Kk i j = aikj (4.4.10)
2
The geodesic equation (4.4.9), however, is not unique for the dynamic equation
(4.4.8).
Proposition 4.4.5. Any quadratic dynamic equation (4.4.8), being equivalent to

the geodesic equation with respect to the symmetric linear connection K f (4.4.10),
is also equivalent to the geodesic equation with respect to an affine connection K 0
on T Q → Q which differs from K f (4.4.10) in a soldering form σ on T Q → Q with
the components
σλ0 = 0, σki = hik + (s − 1)hik q̇ 0 , σ0i = −shik q̇ k − hi0 q̇ 0 + hi0 ,
where s and hiλ are local functions on Q. 2
Proof. The proof follows from direct computation. QED
Proposition 4.4.5 can also be deduced from the following lemma.
Lemma 4.4.6. Every affine vertical vector field
σ = [f i (q µ ) + bij (q µ )qtj ]∂it (4.4.11)
on the affine jet bundle J 1 Q → Q is extended to the soldering form
σ = (f i dt + bik dq k ) ⊗ ∂˙i (4.4.12)
Proof. Similarly to Proposition 4.4.3, one can show that there is one-to-one corre-
spondence between the VQ J 1 Q-valued affine vector fields (4.4.11) on the jet manifold
J 1 Q and the VQ T Q-valued linear vertical vector fields
σ = [bij (q µ )q̇ j + f i (q µ )q̇ 0 ]∂˙i
on the tangent bundle TQ. This linear vertical vector field determines the desired
soldering form (4.4.12). QED
In Section 6.4, we will use Theorem 4.4.2, Corollary 4.4.4 and Proposition 4.4.5
in order to study the relationship between non-relativistic and relativistic equations
of motion.
Now let us extend our inspection of dynamic equations to connections on the
tangent bundle T M → M of the typical fibre M of a configuration bundle Q → R.
In this case, the relationship fails to be canonical, but depends on a trivialization
(4.1.1) of Q → R.
Given such a trivialization, let (t, q i ) be the associated coordinates on Q, where q i
are coordinates on M with transition functions independent of t. The corresponding
i i
trivialization (4.1.3) of J 1 Q → R takes place in the coordinates (t, q i , q̇ ), where q̇
are coordinates on T M . With respect to these coordinates, the transformation law
(4.3.2) of a dynamic connection γ on the affine jet bundle J 1 Q → Q reads
 
0i
i ∂q 0i ∂q 0i ∂ q̇ ∂q n
γ 0 0 = j γ0j γk0i =  j γnj + n  0 k .
∂q ∂q ∂q ∂q
It follows that, given a trivialization of Q → R, a connection γ on J 1 Q → Q defines

the time-dependent vertical vector field
j ∂
γ0i (t, q j , q̇ ) i : R × TM → V TM
∂ q̇
and the time-dependent connection

!
k ∂ j ∂
dq ⊗ k
+ γki (t, q j , q̇ ) i : R × T M → J 1T M ⊂ T T M (4.4.13)
∂q ∂ q̇
on the tangent bundle T M → M .

Conversely, let us consider a connection
!
∂ k i j ∂
K = dq ⊗ k
+ K k (q j , q̇ ) i
∂q ∂ q̇
on the tangent bundle T M → M . Given the above-mentioned trivialization of the

configuration bundle Q → R, the connection K defines the connection Kf (4.4.3),
with the components
i
K0i = 0, Kki = K k ,
4.5. REFERENCE FRAMES 177
on the tangent bundle T Q → Q. The corresponding dynamic connection γ on the

affine jet bundle J 1 Q → Q reads
i
γ0i = 0, γki = K k . (4.4.14)
Using the transformation law (4.3.2), one can extend the expression (4.4.14) to
arbitrary bundle coordinates (t, q i ) on the configuration space Q as follows:
∂q i j j r j r r ∂ 2 q i j ∂Γi
" #
γki = K (q (q ), q̇ (q , q )) + q̇ + n ∂k q n , (4.4.15)
∂q j n t
∂q n ∂q j ∂q
γ0i = ∂t Γi + ∂j Γi qtj − γki Γk ,
where
Γi = ∂t q i (t, q j )
is the connection on Q → R, corresponding to a given trivialization of Q, i.e., Γi = 0

relative to (t, q i ). The dynamic equation on Q defined by the dynamic connection
qtti = ∂t Γi + qtj ∂j Γi + γki (qtk − Γk ). (4.4.16)
By construction, it is a conservative dynamic equation. Thus, we have proved the

following.
Proposition 4.4.7. Any connection K on the typical fibre M of a configuration

bundle Q → R yields a conservative dynamic equation (4.4.16) on Q. 2
4.5 Reference frames

¿From the physical viewpoint, a reference frame in non-relativistic mechanics deter-
mines a tangent vector at each point of a configuration space Q, which characterizes
the velocity of an ”observer” at this point. This speculation leads us to suggest the
following mathematical definition of a reference frame in non-relativistic mechanics
[57, 132, 161].
Definition 4.5.1. In non-relativistic mechanics, a reference frame is a connection

Γ on the configuration bundle Q → R. 2
In accordance with this definition, one can think of the horizontal vector field
(4.1.16), associated with a connection Γ on Q → R, as being a family of ”observers”,
while the corresponding covariant differential
def
q̇Γi = DΓ (qti ) = qti − Γi
determines the relative velocities with respect to the reference frame Γ.

In particular, given a motion c : R → Q, its covariant derivative ∇Γ c (1.4.6)
with respect to a connection Γ is the velocity of this motion relative to the reference
frame Γ. For instance, if c is an integral section for the connection Γ, the velocity
of the motion c relative to the reference frame Γ is equal to 0. Conversely, every
motion c : R → Q, defines a reference frame Γc such that the velocity of c relative
to Γc vanishes. This reference frame Γc is an extension of the section c(R) → J 1 Q
of the affine jet bundle J 1 Q → Q over the closed submanifold c(R) ∈ Q to a global
section in accordance with Theorem 1.1.2.
Remark 4.5.1. It should be emphasized that the vertical tangent bundle V Q of a
configuration bundle Q, but not the jet manifold J 1 Q plays the role of the ”space
of coordinates and velocities”, while elements of J 1 Q may be termed the absolute
velocities. In the universal unit system, elements of V Q, however, have the same
physical dimension [q] as elements of Q, whereas absolute velocities are of physical
dimension [q] − 1. •
By virtue of Corollary 4.1.1, any reference frame Γ on a configuration bundle

Q → R is associated with an atlas of local constant trivializations, and vice versa.
The connection Γ reduces to
Γ = ∂t (4.5.1)
with respect to the corresponding coordinates (t, q i ), whose transition functions

q i → q 0i are independent of time. One can think of these coordinates as being
also the reference frame, corresponding to the connection (4.5.1). They are called
the adapted coordinates to the reference frame Γ. Thus, we come to the following
definition, equivalent to Definition 4.5.1.
Definition 4.5.2. In non-relativistic mechanics, a reference frame is an atlas of

local constant trivializations of a configuration bundle Q → R. 2
4.5. REFERENCE FRAMES 179
In particular, with respect to the coordinates q i adapted to a reference frame Γ,

the velocities relative to this reference frame are equal to the absolute ones
i
DΓ (q it ) = q̇ Γ = q it .
A reference frame is said to be complete if the associated connection Γ is com-
plete. By virtue of Proposition 4.1.2, every complete reference frame defines a
trivialization of a bundle Q → R, and vice versa.
Remark 4.5.2. Given a reference frame Γ, one should solve the equations
∂q i (t, q a )
Γi (t, q j (t, q a )) = , (4.5.2)
∂t
∂q a (t, q j ) i j ∂q a (t, q j )
Γ (t, q ) + =0 (4.5.3)
∂q i ∂t
in order to find the coordinates (t, q a ) adapted to Γ.
Let (t, q1a ) and (t, q2i ) be the adapted coordinates for reference frames Γ1 and
Γ2 , respectively. In accordance with the equality (4.5.3), the components Γi1 of the
connection Γ1 with respect to the coordinates (t, q2i ) and the components Γa2 of the
connection Γ2 with respect to the coordinates (t, q1a ) fulfill the relation
∂q1a i
Γ + Γa2 = 0.
∂q2i 1
•
Using the relations (4.5.2) – (4.5.3), one can rewrite the coordinate transforma-
tion law (4.2.5) of dynamic equations as follows. Let
a
q att = ξ (4.5.4)
be a dynamic equation on a configuration space Q, written with respect to a reference
frame (t, q n ). Then, relative to arbitrary bundle coordinates (t, q i ) on Q → R, the
dynamic equation (4.5.4) takes the form
∂q i ∂q a ∂q i a
qtti = dt Γi + ∂j Γi (qtj − Γj ) − (q j
− Γj
)(q k
− Γk
) + ξ , (4.5.5)
∂q a ∂q j ∂q k t t
∂q a
where Γ is the connection corresponding to the reference frame (t, q n ). The dynamic
equation (4.5.5) can be expressed in the relative velocities q̇Γi = qti − Γi with respect
to the initial reference frame (t, q a ). We have
∂q i ∂q a j k ∂q i a
dt q̇Γi = ∂j Γi q̇Γj − a j k
q̇Γ q̇Γ + a ξ (t, q j , q̇Γj ). (4.5.6)
∂q ∂q ∂q ∂q
Accordingly, any dynamic equation (4.2.2) can be expressed in the relative velocities
q̇Γi = qti − Γi with respect to an arbitrary reference frame Γ as follows:
dt q̇Γi = (ξ − JΓ)it = ξ i − dt Γ, (4.5.7)
where JΓ is the prolongation (4.1.24) of the connection Γ onto the jet bundle J 1 Q →
R.
For instance, let us consider the following particular reference frame Γ for a
dynamic equation ξ. The covariant derivative of a reference frame Γ with respect
to the corresponding dynamic connection γξ (4.3.10) reads
∇γ Γ = Q → T ∗ Q × VQ J 1 Q, (4.5.8)
∇γ Γ = ∇γλ Γk dq λ ⊗ ∂k ,
∇γλ Γk = ∂λ Γk − γλk ◦ Γ.
A connection Γ is called a geodesic reference frame for the dynamic equation ξ if
Γc∇γ Γ = Γλ (∂λ Γk − γλk ◦ Γ) = (dt Γi − ξ i ◦ Γ)∂i = 0. (4.5.9)
Proposition 4.5.3. Integral sections c of a reference frame Γ are solutions of a

dynamic equation ξ if and only if Γ is a geodesic reference frame for ξ. 2
Proof. The proof follows at once from substitution of the equality (4.5.9) in the
dynamic equation (4.5.7). QED
Remark 4.5.3. The left- and right-hand sides of the equation (4.5.7) separately are
not well-behaved objects. This equation will be brought below into the covariant
form (4.7.6). •
Reference frames play a prominent role in many constructions of time-dependent

mechanics. In particular, we obtain a converse of Theorem 4.4.2.
Theorem 4.5.4. Given a reference frame Γ, any connection K (4.4.1) on the

tangent bundle T Q → Q defines a dynamic equation
ξ i = (Kλi − Γi Kλ0 )q̇ λ |q̇0 =1,q̇j =qj .

t
2
4.6. FREE MOTION EQUATIONS 181
This theorem is a corollary of Proposition 4.4.1 and the following lemma.
Lemma 4.5.5. Given a connection Γ on the fibre bundle Q → R and a connection

K on the tangent bundle T Q → Q, there is the connection K
f on T Q → Q with the
components
f0 = 0,
K fi = K i − Γi K 0 .
K
λ λ λ λ
Proof. The proof follows from the inspection of transition functions. QED
4.6 Free motion equations

Let us point out the following interesting class of dynamic equations which we agree
to call the free motion equations.
Definition 4.6.1. We say that the dynamic equation (4.2.2) is a free motion
equation if there exists a reference frame (t, q i ) on the configuration space Q such
that this equation reads
q itt = 0. (4.6.1)
With respect to arbitrary bundle coordinates (t, q i ), a free motion equation takes
the form
∂q i ∂q m
qtti i
= dt Γ + ∂j Γ i
(qtj − Γ ) − m j k (qtj − Γj )(qtk − Γk ),
j
(4.6.2)
∂q ∂q ∂q
where Γi = ∂t q i (t, q j ) is the connection associated with the initial frame (t, q i ) (cf.
(4.5.5)). One can think of the right-hand side of the equation (4.6.2) as being the
general coordinate expression for an inertial force in non-relativistic mechanics. The
corresponding dynamic connection γξ on the affine jet bundle J 1 Q → Q reads
∂q i ∂q m
γki = ∂k Γi − (q j − Γj ), (4.6.3)
∂q m ∂q j ∂q k t
γ0i = ∂t Γi + ∂j Γi qtj − γki Γk .
It is affine. By virtue of Proposition 4.4.3, this dynamic connection defines a linear

connection K on the tangent bundle T Q → Q, whose curvature necessarily vanishes.
Thus, we come to the following criterion of a dynamic equation to be a free motion
equation.
Proposition 4.6.2. If ξ is a free motion equation on a configuration space Q,

it is quadratic, and the corresponding symmetric linear connection (4.4.10) on the
tangent bundle T Q → Q is a curvature-free connection. 2
This criterion is not a sufficient condition because it may happen that the com-
ponents of a curvature-free symmetric linear connection on T Q → Q vanish with
respect to the coordinates on Q which are not compatible with the fibration Q → R.
The similar criterion involves the curvature of a dynamic connection (4.6.3) of a
free motion equation.
Proposition 4.6.3. If ξ is a free motion equation, then the curvature R (4.3.3) of

the corresponding dynamic connection γξ is equal to 0, and so are the tensor field
R (4.3.4) and the scalar field R
e (4.3.5). 2
Proposition 4.6.3 also fails to be a sufficient condition. If the curvature R (4.3.3)

of a dynamic connection γξ vanishes, it may happen that components of γξ are
equal to 0 with respect to non-holonomic bundle coordinates on the affine jet bundle
J 1 Q → Q.
Nevertheless, we can formulate the necessary and sufficient condition of the ex-
istence of a free motion equation on a configuration space Q.
Proposition 4.6.4. A free motion equation on a fibre bundle Q → R exists if and

only if the typical fibre M of Q admits a curvature-free symmetric linear connection.
2
Proof. Let a free motion equation take the form (4.6.1) with respect to some atlas
of local constant trivializations of a fibre bundle Q → R. By virtue of Proposition
4.3.2, there exists an affine dynamic connection γ on the affine jet bundle J 1 Q → Q
whose components relative to this atlas are equal to 0. Given a trivialization chart
of this atlas, the connection γ defines the curvature-free symmetric linear connection
(4.4.13) on M . The converse statement follows at once from Proposition 4.4.7. QED
4.6. FREE MOTION EQUATIONS 183
The free motion equation (4.6.2) is simplified if the coordinate transition func-
tions q i → q i are affine in the coordinates q i . Then we have
qtti = ∂t Γi − Γj ∂j Γi + 2qtj ∂j Γi . (4.6.4)
Example 4.6.1. Let us consider a free motion on a plane R2 . The corresponding

configuration bundle is R3 → R, coordinated by (t, r). The dynamic equation of
this motion is
r̈ = 0. (4.6.5)
Let us choose the rotatory reference frame with the adapted coordinates
!
cos ωt − sin ωt
r = Ar, A= . (4.6.6)
sin ωt cos ωt
Relative to these coordinates, the connection Γ corresponding to the initial reference
frame reads
Γ = ∂t r = ∂t A · A−1 r.
Then the free motion equation (4.6.5) with respect to the rotatory reference frame
(4.6.6) takes the familiar form
!
0 −1
rtt = ω 2 r + 2 rt . (4.6.7)
1 0
The first term in the right-hand side of the equation (4.6.7) is the centrifugal force
(−Γj ∂j Γi ), while the second one is the Coriolis force (2qtj ∂j Γi ). •
The following lemma shows that the free motion equation (4.6.4) is affine in the
coordinates q i and qti .
Lemma 4.6.5. Let (t, q a ) be a reference frame on a configuration bundle Q → R

and Γ the corresponding connection. Components Γi of this connection with respect
to another coordinate system (t, q i ) are affine functions in the coordinates q i if and
only if the transition functions between the coordinates q a and q i are affine. 2
Proof. If
q i = ain (t)q n + bi (t),

then we have
Γi = (a−1 )nj ∂t ain (q j − bj ) + ∂t bi .
Conversely, let
Γi = cij (t)q j + si (t)
such that the homogeneous linear system of differential equations
∂t q i (t) = cij (t)q j (t) + si (t) (4.6.8)
has a solution on R. Let aij (t) be the m×m matrix whose columns are m fundamental
solutions q1 (t), · · · , qm (t) of the equations (4.6.8) (see, e.g., [22]). This matrix is
invertible for all t ∈ R. Then the connection Γ is associated with the coordinates
(t, q n ) such that
q i = ain (t)q n + bi (t),
where functions bi (t) fulfill the equations
∂t bi = (a−1 )kn ∂t aik bn + si .
QED
One can easily find the geodesic reference frames for the free motion equation
qtti = 0. (4.6.9)
They are Γi = v i = const. By virtue of Lemma 4.6.5, these reference frames define
the adapted coordinates
q i = kji q j − v i t − ai , kji = const., v i = const., ai = const. (4.6.10)
The equation (4.6.9) obviously keeps its free motion form under the transformations
(4.6.10) between the geodesic reference frames. It is readily observed that these
transformations are precisely the elements of the Galilei group.
4.7. RELATIVE ACCELERATION 185
4.7 Relative acceleration

In comparison with the notion of a relative velocity, that of a relative acceleration
is more intricate.
To consider a relative acceleration with respect to a reference frame Γ, one
should prolong the connection Γ on the configuration bundle Q → R to a holonomic
connection ξΓ on the jet bundle J 1 Q → R. Note that the jet prolongation JΓ (4.1.24)
of Γ onto J 1 Q → R is not holonomic. We can construct the desired prolongation
by means of a dynamic connection γ on the affine jet bundle J 1 Q → Q.
Lemma 4.7.1. Let us consider the composite bundle (4.1.10). Given a frame Γ
on Q → R and a dynamic connections γ on J 1 Q → Q, there exists a dynamic
connection γe on J 1 Q → Q with the components
γeki = γki , γe0i = dt Γi − γki Γk . (4.7.1)
Proof. Combining the connection Γ on Q → R and the connection γ on J 1 Q → Q

gives the composite connection (1.6.8) on J 1 Q → R which reads
B = dt ⊗ (∂t + Γi ∂i + (γki Γk + γ0i )∂it ).
Let JΓ be the jet prolongation (4.1.24) of the connection Γ on J 1 Q → R. Then the

difference
JΓ − B = dt ⊗ (dt Γi − γki Γk − γ0i )∂it
is a VQ J 1 Q-valued soldering form on the jet bundle J 1 Q → R, which is also a

soldering form on the affine jet bundle J 1 Q → Q. The desired connection (4.7.1) is
γe = γ + JΓ − B = dt ⊗ (∂t + (dt Γi − γki Γk )∂it ) + dq k ⊗ (∂k + γki ∂it ).
QED
Now, we construct a certain soldering form on the affine jet bundle J 1 Q → Q

and add it to this connection. Let us apply the canonical projection T ∗ Q → V ∗ Q
and then the imbedding Γ : V ∗ Q → T ∗ Q to the covariant derivative (4.5.8) of
the reference frame Γ with respect to the dynamic connection γ. We obtain the
VQ J 1 Q-valued 1-form
σ = [−Γi (∂i Γk − γik ◦ Γ)dt + (∂i Γk − γik ◦ Γ)dq i ] ⊗ ∂kt
on Q whose pull-back onto J 1 Q is the desired soldering form. The sum
def
γΓ = γe + σ,
called the frame connection, reads
γΓ i0 = dt Γi − γki Γk − Γk (∂k Γi − γki ◦ Γ), (4.7.2)
γΓ ik = γki i
+ ∂k Γ − γki ◦ Γ.
This connection yields the desired holonomic connection
ξΓi = dt Γi + (∂k Γi + γki − γki ◦ Γ)(qtk − Γk )
on the jet bundle J 1 Q → R.
Let ξ be a dynamic equation and γ = γξ the connection (4.3.10) associated with
ξ. Then one can think of the vertical vector field
def
aΓ = ξ − ξΓ = (ξ i − ξΓi )∂it (4.7.3)
on the affine jet bundle J 1 Q → Q as being a relative acceleration with respect to
the reference frame Γ in comparison with the absolute acceleration ξ.
For instance, let us consider a reference frame which is geodesic for the dynamic
equation ξ, i.e., the relation (4.5.9) holds. Then the relative acceleration of a motion
c with respect to the reference frame Γ is
(ξ − ξΓ ) ◦ Γ = 0.
Let ξ now be an arbitrary dynamic equation, written with respect to coordinates
(t, q i ) adapted to the reference frame Γ, i.e., Γi = 0. In these coordinates, the relative
acceleration with respect to the reference frame Γ is
1
aiΓ = ξ i (t, q j , qtj ) − qtk (∂k ξ i − ∂k ξ i |qj =0 ). (4.7.4)
2 t
Given another bundle coordinates (t, q 0i ) on Q → R, this dynamic equation takes

the form (4.5.6), while the relative acceleration (4.7.4) with respect to the reference
frame Γ reads
a0iΓ = ∂j q 0i ajΓ .
4.7. RELATIVE ACCELERATION 187
Then we can write a dynamic equation (4.2.2) in the form which is covariant under
coordinate transformations:
f qi = d qi − ξ i = a ,
D (4.7.5)
γΓ t t t Γ Γ
where Df is the vertical covariant differential (4.3.9) with respect to the frame
γΓ
connection γΓ (4.7.2) on the affine jet bundle J 1 Q → Q.
In particular, if ξ is a free motion equation which takes the form (4.6.1) with
respect to a reference frame Γ, then
f qi = 0
DγΓ t
relative to arbitrary bundle coordinates on the configuration bundle Q → R.

The left-hand side of the dynamic equation (4.7.5) can also be expressed in the
relative velocities such that this dynamic equation takes the form
dt q̇Γi − γΓ ik q̇Γk = aΓ (4.7.6)
which is the covariant form of the equation (4.5.7).

The concept of a relative acceleration is understood better when we deal with
the quadratic dynamic equation ξ, and the corresponding dynamic connection γ is
affine.
Lemma 4.7.2. If a dynamic connection γ is affine, i.e.,
γλi = γλ0
i i
+ γλk qtk ,
so is a frame connection γΓ for any frame Γ. 2
Proof. The proof follows from direct computation. We have
γΓ i0 = ∂t Γi + (∂j Γi − γkj
i
Γk )(qtj − Γj ),
γΓ ik = ∂k Γi + γkj
i
(qtj − Γj )
or
γΓ ijk = γjk
i
,
γΓ i0k = ∂k Γi − γjk
i
Γj , γΓ ik0 = ∂k Γi − γkj
i
Γj , (4.7.7)
γΓ i00 i j
= ∂t Γ − Γ ∂j Γ + i i
γjk Γj Γk .
QED
In particular, we obtain
γΓ ijk = γjk
i
, γΓ i0k = γΓ ik0 = γΓ i00 = 0
relative to the coordinates adapted to a reference frame Γ.

A glance at the expression (4.7.7) shows that, if a dynamic connection γ is
symmetric, so is a frame connection γΓ .
Corollary 4.7.3. If a dynamic equation ξ is quadratic, the relative acceleration

aΓ (4.7.3) is always affine, and it admits the decomposition
aiΓ = −(Γλ ∇γλ Γi + 2q̇Γλ ∇γλ Γi ), (4.7.8)
where γ = γξ is the dynamic connection (4.3.10), and
q̇Γλ = qtλ − Γλ , qt0 = 1, Γ0 = 1,
is the relative velocity with respect to the reference frame Γ. 2
Note that the splitting (4.7.8) gives a generalized Coriolis theorem. In particu-
lar, the well-known analogy between inertial and electromagnetic forces is restated.
Corollary 4.7.3 shows that this analogy can be extended to arbitrary quadratic dy-
namic equation.
4.8 Lagrangian systems

A velocity phase space J 1 Q of time-dependent mechanics does not admit any canon-
ical structure mentioned in the previous Chapter. Since J 1 Q is an odd-dimensional
manifold, it has no symplectic structure, whereas presymplectic and Poisson struc-
tures are defined only for Lagrangian systems, and depend on a Lagrangian. By
a Lagrangian system is meant a mechanical system whose motions are solutions of
Lagrange equations for some Lagrangian on a velocity phase space J 1 Q. Obviously,
a mechanical system is not necessarily Lagrangian, whereas Lagrange equations are
not necessarily dynamic equations. Moreover, it may happen that Lagrange equa-
tions fail to be differential equations in a strict sense (see Definition 1.3.3). Note that,
in the framework of Lagrangian formalism, one also meets Cartan and Hamilton–De
4.8. LAGRANGIAN SYSTEMS 189
Donder equations, besides the Lagrange one. These equations are equivalent to each
other and to Lagrange equations in the case of a regular Lagrangian.
A Lagrangian of a mechanical system is defined as a horizontal density
L = Ldt, (4.8.1)
1
L : J Q → R,
on the velocity phase space J 1 Q, where L is called a Lagrangian function or simply

a Lagrangian if there is no danger of confusion. With respect to the universal unit
system, a Lagrangian (4.8.1) is physically dimensionless.
Here, we do not study the calculus of variations in depth, but apply in a straight-
forward manner the first variational formula [57].
Let us consider a projectable vector field
u = ut ∂t + ui ∂i , ut = 0, 1, (4.8.2)
on a configuration bundle Q → R and calculate the Lie derivative of a Lagrangian

(4.8.1) along the jet prolongation
u = ut ∂t + ui ∂i + dt ui ∂it ,
dt = ∂t + qti ∂i + qtti ∂it ,
(1.3.7) of u. We obtain
Lu L = (ucdL)dt = (ut ∂t + ui ∂i + dt ui ∂it )Ldt. (4.8.3)
The first variational formula provides the following canonical decomposition of

the Lie derivative (4.8.3) in accordance with the variational problem:
ucdL = (ui − ut qti )Ei + dt (ucHL ), (4.8.4)
where
HL = vb∗ (dL) + L = πi dq i − (πi qti − L)dt (4.8.5)
is the Poincaré–Cartan form (see (4.1.14)), and
EL : J 2 Q → V ∗ Q,
EL = Ei dq i = (∂i − dt ∂it )Ldq i (4.8.6)
is the Euler–Lagrange operator for a Lagrangian L. The latter can be seen as a

2-form
EL = (∂i − dt ∂it )Ldq i ∧ dt.
Its coefficients Ei are called variational derivatives. We will use the notation
πi = ∂it L, πji = ∂jt ∂it L.
The kernel Ker EL ⊂ J 2 Q of the Euler–Lagrange operator (4.8.6) defines the

system of second order differential equations on Q
(∂i − dt ∂it )L = 0, (4.8.7)
called the Lagrange equations. Its solutions are (local) sections c of the fibre bundle
Q → R whose second order jet prolongations c̈ live in (4.8.7). They obey the
equations
d
∂i L ◦ ċ − (πi ◦ ċ) = 0. (4.8.8)
dt
Remark 4.8.1. The kernel of an Euler–Lagrange operator Ker EL fails to be a

subbundle of the fibre bundle J 2 Q → R in general. Therefore, it may happen that, as
was mentioned above, the Lagrange equations (4.8.7) are not differential equations
in a strict sense. In particular, by virtue of Proposition 1.3.4, it is a differential
equation when the morphism (4.8.6) is of constant rank, e.g., if a Lagrangian L is
regular. Recall that a Lagrangian L is called regular (non-degenerate), if
det πji 6= 0
everywhere on the velocity phase space J 1 Q. If a Lagrangian L is non-degenerate,

the Lagrange equations can be solved algebraically for second order derivatives, and
are equivalent to a dynamic equation. •
Example 4.8.2. Let Q = R2 → R be a configuration space, coordinated by

(t, q). The corresponding velocity phase space J 1 Q is equipped with the adapted
coordinates (t, q, qt ). The Lagrangian
1
L = q 2 qt2 dt
2
on J 1 Q leads to the Euler–Lagrange operator
EL = [qqt2 − dt (q 2 qt )]dq
whose kernel is not a submanifold at the point q = 0. •
Every Lagrangian L on the jet manifold J 1 Q yields the Legendre map

b : J 1 Q → V ∗ Q,
L (4.8.9)
pi ◦ L
b =π,
i
where (t, q i , pi ) are coordinates on the vertical cotangent bundle V ∗ Q. Indeed, due
to the vertical splitting (4.1.8), the vertical tangent morphism V L to L yields the
linear morphism
V L : J 1Q × V Q → R
and, consequently, the morphism (4.8.9).

In time-dependent mechanics on a configuration space Q → R, the vertical
cotangent bundle V ∗ Q plays the role of a momentum phase space (see Chapter 5).
Remark 4.8.3. The Legendre map (4.8.9) is a local diffeomorphism, i.e., it is of
maximal constant rank if and only if a Lagrangian L is regular. A Lagrangian L is
b is a diffeomorphism. •
called hyperregular if the Legendre map L
In Section 5.1, we will show that the vertical cotangent bundle V ∗ Q is provided
with the canonical 3-form
def
Ω = dpi ∧ dq i ∧ dt, (4.8.10)
derived from the polysymplectic form (2.9.9). Let us consider the pull-back
b ∗ Ω = dπ ∧ dq i ∧ dt
ΩL = L (4.8.11)
i
on J 1 Q of the canonical form (4.8.10) by means of the Legendre map L b (4.8.9).

The form ΩL provides Lagrangian formalism with the construction similar to Ha-
miltonian mechanics, but which depends on the choice of a Lagrangian L. In the
framework of this construction, the Lagrangian counterpart of a Hamiltonian form
is the Poincaré–Cartan form (4.8.5).
If a Lagrangian L is regular, the form ΩL (4.8.11) defines a Poisson structure on

the jet manifold J 1 Q as follows. By means of ΩL , every vertical vector field
ϑ = ϑi ∂i + ϑ̇i ∂it
on the jet bundle J 1 Q → R corresponds to the 2-form
ϑcΩL = {[ϑ̇j πji + ϑj (∂j πi − ∂i πj )]dq i − ϑi πji dqtj } ∧ dt.
This is one-to-one correspondence if a Lagrangian L is regular. Indeed, given an

arbitrary 2-form
φ = (φi dq i + φ̇i dqti ) ∧ dt
on J 1 Q, the algebraic equations
ϑ̇j πji + ϑj (∂j πi − ∂i πj ) = φi ,

−ϑi πji = φ̇j
have a unique solution
ϑi = −(π −1 )ij φ̇j ,

ϑ̇j = (π −1 )ji [φi + (π −1 )kn φ̇n (∂k πi − ∂i πk )].
In particular, every function f on J 1 Q determines a vertical vector field
ϑf = (π −1 )ij ∂jt f ∂i − (π −1 )ji [∂i f + (π −1 )kn ∂nt f (∂k πi − ∂i πk )]∂jt (4.8.12)
on J 1 Q → R in accordance with the relation
ϑf cΩL = −df ∧ dt.
Then the Poisson bracket

def
{f, g}L dt = ϑg cϑf cΩL , f, g ∈ O0 (J 1 Q), (4.8.13)
−1 ij
{f, g}L = (π ) (∂it f ∂j g − ∂it g∂j f ) +
(∂n πk − ∂k πn )(π −1 )ki (π −1 )nj ∂it f ∂jt g,
can be defined on the space O0 (J 1 Q) of functions on the jet manifold J 1 Q. In

particular, the vertical vector field ϑf (4.8.12) is the Hamiltonian vector field for
the function f with respect to the Poisson structure (4.8.13). The Poisson structure
(4.8.13) defines the corresponding symplectic foliation on the jet manifold J 1 Q,

which coincides with the fibration J 1 Q → R. The symplectic form on the leaf Jt1 Q
of this foliation is
Ωt = dπi ∧ dq i
[57, 182].
Example 4.8.4. If a Lagrangian L is hyperregular, the Poisson structure (4.8.13) is
isomorphic to the canonical Poisson structure on the momentum phase space V ∗ Q
(see Section 5.1). Indeed, the Poisson bracket (4.8.13) reads
{πi , πj } = {q i , q j } = 0, {πi , q j } = δij .
Remark 4.8.5. If a Lagrangian is degenerate, one may try to introduce a Poisson

structure on J 1 Q corresponding to the presymplectic form dHL (see (4.8.25) below)
in accordance with the procedure suggested in [49] (see Remark 2.5.1). •
Definition 4.8.1. A connection
ξL = ∂t + f i ∂i + ξ i ∂it
on the jet bundle J 1 Q → R is said to be a Lagrangian connection for the Lagrangian

L if it obeys the equation
ξL cΩL = dHL (4.8.14)
which takes the coordinate form
(f i − qti )πji = 0, (4.8.15)

j j
∂i L − ∂t πi − f ∂j πi − ξ πji + (f − j
qtj )∂i πj = 0. (4.8.16)
Example 4.8.6. A glance at the equations (4.8.15) – (4.8.16) shows that, for a
regular Lagrangian L, a Lagrangian connection is necessarily holonomic and unique.
•
In order to clarify the meaning of the equation (4.8.14), let us consider the
Lagrangian
def
L = L(t, q i , qti ) + (q(t)
i
− qti )πi (t, q i , qti )
on the repeated jet manifold J 1 J 1 Q. The corresponding Euler–Lagrange operator,

called the Euler–Lagrange–Cartan operator, reads
EL : J 1 J 1 Q → T ∗ J 1 Q,
j
EL = [(∂i L + ∂i πj (q(t) − qtj ) − dbt πi )dq i + πij (q(t)
j
− qtj )dqti ] ∧ dt, (4.8.17)
i
dbt = ∂t + q(t) ∂i + qtti ∂it .
Then the equation (4.8.14) is equivalent to the condition
ξL (J 1 Q) ⊂ Ker EL
which is the first order differential equation on the jet manifold J 1 Q, called the
Cartan equations. They read
j
πij (q(t) − qtj ) = 0, (4.8.18)
j
∂i L − dbt πi + (q(t) − qtj )∂i πj = 0. (4.8.19)
Remark 4.8.7. In Section 5.2, we will show that the Euler–Lagrange–Cartan

operator (4.8.17) is the Lagrangian counterpart of a Hamilton operator. •
Solutions of the Cartan equations (4.8.18) – (4.8.19) are (local) integral sections
c : () → J 1 Q of Lagrangian connections ξL for a Lagrangian L, i.e., J 1 c = ξL ◦ c.
Furthermore, the equation (4.8.14) for these integral sections is equivalent to the
relation
c∗ (ϑcdHL ) = 0 (4.8.20)
which is assumed to hold for any vertical vector field ϑ on the jet bundle J 1 Q → R.
It is easily seen that the restriction
Ker EL = J 2 Q ∩ Ker EL (4.8.21)
of the Euler–Lagrange–Cartan operator EL (4.8.17) to the second order jet manifold

J 2 Q ⊂ J 1 J 1 Q recovers the Euler–Lagrange operator EL (4.8.6) for the Lagrangian
L. Therefore, the Lagrange equations (4.8.7) are equivalent to the Cartan equations
(4.8.18) – (4.8.19) on the holonomic sections c = J 1 c of the jet bundle J 1 Q → R.
It follows that solutions of the Lagrange equations (4.8.7) are integral sections of
holonomic Lagrangian connections for the Lagrangian L, which take their values into
the kernel of the Euler–Lagrange operator (4.8.21). Different holonomic Lagrangian
connections lead to different dynamic equations associated with the same system of
Lagrange equations.
In the case of a regular Lagrangian L, the Cartan equations (4.8.18) – (4.8.19)
are equivalent to the Lagrange equations (4.8.7), and lead to the unique dynamic
equation
qttj = (π −1 )ij [−∂i L + ∂t πi + qtk ∂k πi ] (4.8.22)
on the configuration space Q.

By very definition, the Poincaré–Cartan form HL (4.8.5) defines the fibred mor-
phism
c : J 1 Q −→ T ∗ Q,
H (4.8.23)
L
Q
c = (π , L − π q i ),
(pi , p) ◦ H L i i t
¿from the affine jet bundle J 1 Q → Q to the cotangent bundle T ∗ Q → Q of Q,

which plays the role of the homogeneous Legendre bundle (2.9.3), equipped with the
coordinates (t, q i , p, pi ). The morphism (4.8.23) is termed the Legendre morphism
associated with HL . There is the following relation between the Legendre map L b
(4.8.9) and the Legendre morphism H c (4.8.23):
L
b =ζ ◦H
L c , (4.8.24)
L
where ζ is the canonical projection T ∗ Q → V ∗ Q (1.1.7).

Furthermore, it is readily observed that the Poincaré–Cartan form HL is the
pull-back of the canonical Liouville form
Ξ = pdt + pi dq i
on the cotangent bundle T ∗ Q by the associated Legendre morphism (4.8.23). Ac-

cordingly, the presymplectic form
dHL = dπi ∧ dq i − d(πi q i − L) ∧ dt (4.8.25)

on the jet manifold J 1 Q is the pull-back of the canonical symplectic form
dΞ = dp ∧ dt + dpi ∧ dq i
on the cotangent bundle T ∗ Q of the configuration space Q.

Remark 4.8.8. The presymplectic form dHL (4.8.25), together with the 1-form dt,
define a copresymplectic structure on the configuration space J 1 Q [37, 86]. Let us
take the exterior product
(dHL )m ∧ dt = det(πij )( dqti ∧ dq i )m ∧ dt.

X
(4.8.26)
i
If it is nowhere vanishing, the pair (dHL , dt) is called a cosymplectic structure on

the (2m + 1)-dimensional manifold J 1 Q. A glance at the expression (4.8.26) shows
that the pair (dHL , dt) is cosymplectic if and only if the Lagrangian L is regular. A
cosymplectic structure (dHL , dt) on J 1 Q defines the isomorphism
T J 1 Q 3 v 7→ vcdHL + (vcdt)dt ∈ T ∗ J 1 Q.
In particular, let T be a subbundle of the tangent bundle T J 1 Q → J 1 Q. One can

define the orthocomplement
T⊥ = {v ∈ T J 1 Q : vcdHL + (vcdt)dt ∈ Ann (T)}
of T in T J 1 Q with respect to (dHL , dt).

Note that the classification of Lagrangians on the product R × M by the proper-
ties of the intersection Ker dHL ∩ Ker dt has been suggested [86]. This classification
remains also true for Lagrangians on an arbitrary fibre bundle Q → R, though not all
its constructions (e.g., the tangent-valued forms S and S (see [86])) are maintained
under time-dependent transformations. •
Given a Lagrangian L, let the image ZL of the configuration space J 1 Q by the

Legendre morphism Hc (4.8.23) be an imbedded subbundle
L
iL : ZL ,→ T ∗ Q
of the cotangent bundle T ∗ Q. It is provided with the pull-back De Donder form
ΞL = i∗L Ξ.
By analogy with the Cartan equations (4.8.20), the corresponding Hamilton–De

Donder equations for sections r of the fibre bundle ZL → R are written as the
condition
r∗ (ucdΞL ) = 0 (4.8.27)
where u is an arbitrary vertical vector field on ZL → R. To obtain these equations

in an explicit form, one should substitute the solutions
qti (t, q j , pj ), L(t, q j , pj , p)
of the equations
pi = πi (t, q j , qtj ),
p = L(t, q j , qtj ) − πi (t, q j , qtj )qti (4.8.28)
in the Cartan equations (4.8.20).

If a Lagrangian L is regular, the equations (4.8.28) have a unique solution. Then
the Hamilton–De Donder equations take the coordinate form
∂t ri = −∂ i r,
∂t ri = ∂i r, (4.8.29)
r = L(t, q i , qti (t, q j , pj )) − pi qti (t, q j , pj ),
and are equivalent to the Cartan equations. If a Lagrangian L is degenerate, the

equations (4.8.28) may admit different solutions, or no solutions at all. More can
be said in the following case.
Proposition 4.8.2. Let the Legendre morphism H c : J 1 Q → Z be a submersion.

L L
1
Then a section c of the jet bundle J Q → R is a solution of the Cartan equations
c ◦ c is a solution of the Hamilton–De Donder equations
(4.8.20) if and only if H L
(4.8.27) [63]. 2
Remark 4.8.9. In the case of a regular Lagrangian L, the Hamilton–De Donder

equations (4.8.29) can be seen as the equations of motion in variant (B) of the
Dirac system on the symplectic manifold T ∗ Q in the case of: (i) a zero Hamiltonian,
(ii) the primary constraint space N = ZL (4.8.28), and (iii) with the additional
condition that a solution v of the equation (3.6.1) has the component v 0 = 1. The
latter condition guarantees that the motion parameter is t. •
In conclusion, let us touch briefly on the case of conservative Lagrangians. Let

us suppose that, given a trivialization
ψ:Q∼
=R×M
in the coordinates (t, q i ), a Lagrangian L is independent of t. With respect to these
coordinates, the configuration space J 1 Q admits the presymplectic form
def
ωL = dπi ∧ dq i
(cf. (3.3.5)), and the equation (4.8.14) for a Lagrangian connection ξL may be
written in the form
ξL cωL = −d(πi q it − L).
Thus, a conservative Lagrangian system reduces to the presymplectic Hamiltonian

system whose Hamiltonian is the energy function (πi q it − L) (see Examples 3.3.2 and
3.4.2). In particular, if a Lagrangian is degenerate, one can apply the procedure from
Section 3.6 to investigate it [138]. We refer the reader to [31, 37, 111] for extension
of this procedure to time-dependent degenerate Lagrangian systems on the product
R × M . We will investigate degenerate Lagrangian systems in the framework of
Hamiltonian formalism (see Section 5.5).
4.9 Newtonian systems

Let L be a Lagrangian on a velocity phase space J 1 Q and L b the Legendre map
(4.8.9). Due to the vertical splitting (1.1.6) of V V ∗ Q, the vertical tangent map V L
b
to L
b reads
b : V J 1 Q → V ∗ Q × V ∗ Q.
VL Q
Q
It yields the linear fibred morphism

def b : V J 1 Q → V ∗ J 1 Q,
[ =(Id J 1 Q , pr2 ◦ V L) Q Q (4.9.1)
[: ∂it 7→ πij dqtj ,
where {dqtj } are bases for the fibres of the vertical tangent bundle VQ∗ J 1 Q → J 1 Q.
The morphism (4.9.1) defines the mapping
J 1 Q → VQ∗ J 1 Q ⊗ VQ∗ J 1 Q
J 1Q
4.9. NEWTONIAN SYSTEMS 199
and, due to the splitting (4.1.9), also the mapping
c : J 1 Q −→ V ∗ Q ⊗ V ∗ Q,
m
Q Q
mij = pij ◦ m
c = πij ,
where (t, q i , pij ) are holonomic coordinates on V ∗ Q ⊗ V ∗ Q. Thus, πij = mij are
Q
2
∗ c on the velocity phase space J 1 Q. It is
components of the ∨ V Q-valued field m
called the mass tensor.
Let a Lagrangian L be regular. Then the mass tensor is non-degenerate, and
defines a fibre metric, called mass metric, in the vertical tangent bundle VQ J 1 Q →
J 1 Q. Let us recall that, if a Lagrangian L is regular, there exists a unique Lagrangian
connection ξL for L which is holonomic in accordance with the equation (4.8.15),
and obeys the equation
mik ξLk = −∂t πi − ∂j πi qtj + ∂i L. (4.9.2)
This holonomic connection defines the dynamic equation (4.8.22). At the same time,
the equation (4.9.2) leads to the commutative diagram
[
VQ J 1 Q −→ VQ∗ J 1 Q
DξL -%EL
J 2Q
EL = [ ◦ DξL ,
Ei = mik (qttk − ξLk ), (4.9.3)
where DξL is the covariant differential (4.1.17) relative to the connection ξL . Fur-
thermore, the derivation of (4.9.2) with respect to qtj results in the relation
ξL cdmij + mik γjk + mjk γik = 0, (4.9.4)
where
1
γik = ∂it ξLk
2
are coefficients of the symmetric dynamic connection γξL (4.3.10) corresponding to
the dynamic equation ξL .
Thus, each regular Lagrangian L defines the dynamic equation ξL , related to the
Euler–Lagrange operator EL by means of the equality (4.9.3), and the non-degenerate
mass tensor mij , related to the dynamic equation ξL by means of the relation (4.9.4).
This is a Newtonian system in accordance with the following definition.
Definition 4.9.1. Let Q → R be a fibre bundle together with
c in the fibre bundle VQ J 1 Q → J 1 Q:

• a (non-degenerate) fibre metric m
c : J 1 Q → V ∗ Q ⊗ V ∗ Q,
m
Q
1
c = mij dq i ∨ dq j ,
m
2
satisfying the symmetry condition
∂kt mij = ∂jt mik , (4.9.5)
• and a holonomic connection ξ (4.1.25) on the jet bundle J 1 Q → R, related to

the fibre metric m
c by the compatibility condition (4.9.4).
The triple (Q, m

c, ξ) is called a Newtonian system. This is not the terminology of
[31]. 2
Note that the compatibility condition (4.9.4) can also be introduced in an in-
trinsic way as
∇ξ m
c = 0,
where by ∇ is the covariant derivative with respect to the connection γeξ∗ on the
vertical cotangent bundle VQ∗ J 1 Q → J 1 Q, which is dual of the connection γeξ (4.3.23)
on the vertical tangent bundle VQ J 1 Q → J 1 Q.
Definition 4.9.1 generalizes the second Newton law of point mechanics. Indeed,
the dynamic equation for a Newtonian system is equivalent to the equation
mik (qttk − ξ k ) = 0. (4.9.6)
There are two main reasons for considering Newtonian systems.

From the physical viewpoint, with a mass tensor, we can introduce the notion
of an external force. Note that, in the universal unit system, the mass tensor m
c
is dimensional. For instance, the dimension of a mass tensor of a point mass with
respect to Cartesian coordinates q i is [length]−1 , while that with respect to the angle
coordinates is [length].
Definition 4.9.2. An external force is defined as a section of the vertical cotangent

bundle VQ∗ J 1 Q → J 1 Q. Let us also bear in mind the isomorphism (4.1.9). 2
Note that there are no canonical isomorphisms between the vertical cotangent
bundle VQ∗ J 1 Q and the vertical tangent bundle VQ J 1 Q of J 1 Q. Therefore, one should
distinguish forces and accelerations which are related by means of a mass metric (see
also Remark 4.9.2 below).
Let (Q, m
c, ξ) be a Newtonian system and f an external force. Then
def
ξfi = ξ i + (m−1 )ik fk (4.9.7)
is a dynamic equation, but the triple (Q, m

c, ξf ) is not a Newtonian system in general.
As follows from direct computation, if and only if an external force possesses the
property
∂it fj + ∂jt fi = 0, (4.9.8)
then ξf (4.9.7) fulfills the relation (4.9.4), and (Q, m

c, ξf ) is also a Newtonian system.
Example 4.9.1. For instance, the Lorentz force
fi = eFλi qtλ , qt0 = 1, (4.9.9)
where
Fλµ = ∂λ Aµ − ∂µ Aλ (4.9.10)
is the electromagnetic strength, obeys the condition (4.9.8). Note that the Lorentz
force (4.9.9), just as other forces, can be expressed in the relative velocities q̇Γ with
respect to an arbitrary reference frame Γ:
∂q j ∂q n
!
fi = e i F nj q̇Γk + F 0j ,
∂q ∂q k
where q are the coordinates adapted to the reference frame Γ, and F is the electro-
magnetic strength, written with respect to these coordinates. •
Remark 4.9.2. The contribution of an external force f to a dynamic equation
qtti − ξ i = (m−1 )ik fk
of a Newtonian system obviously depends on a mass tensor. It should be empha-

sized that, besides external forces, we have a universal force which is a holonomic
connection
ξ i = Kµλ
i
qtµ qtλ , qt0 = 1,
associated with a symmetric linear connection K f (4.4.3) on the tangent bundle

T Q → Q. From the physical viewpoint, this is a non-relativistic gravitational force,
including an inertial force, whose contribution to a dynamic equation is independent
of a mass tensor. •
¿From the mathematical viewpoint, the equation (4.9.6) is the kernel of an Euler–
Lagrange-type operator (see (4.9.25) below). By an appropriate choice of a mass
tensor, one may hope to bring it into Lagrange equations. We have seen that a
non-degenerate Lagrangian system is necessarily a Newtonian one, while a converse
statement is not generally true.
Example 4.9.3. Let us consider a non-degenerate quadratic Lagrangian
1
L = mij (q µ )qti qtj + ki (q µ )qti + φ(q µ ), (4.9.11)
2
where the mass tensor mij is a Riemannian metric in the vertical tangent bundle
V Q → Q (see the isomorphism (4.1.8)). Then the Lagrangian L (4.9.11) can be
written as
1
L = − gαµ qtα qtµ , qt0 = 1, (4.9.12)
2
where g is the metric
g00 = −2φ, g0i = −ki , gij = −mij (4.9.13)
on the tangent bundle T Q. The corresponding Lagrange equations take the form
qtti = −(m−1 )ik {λkν }qtλ qtν , qt0 = 1, (4.9.14)
where
1
{λµν } = − (∂λ gµν + ∂ν gµλ − ∂µ gλν )
2
are the Christoffel symbols of the metric (4.9.13). Let us assume that this metric is
non-degenerate. By virtue of Corollary 4.4.4, the dynamic equation (4.9.14) gives
rise to the geodesic equation (4.4.9) on the tangent bundle T Q, which reads
q̈ 0 = 0, q̇ 0 = 1,
q̈ i = {λ i ν }q̇ λ q̇ ν − g i0 {λ0ν }q̇ λ q̇ ν .
Let us now bring the Lagrangian function (4.9.11) into the form
1
L = mij (q µ )(qti − Γi )(qtj − Γj ) + φ0 (q µ ), (4.9.15)
2
where Γ is a Lagrangian frame connection on Q → R which takes its values into the
kernel of the Legendre map L b (see Section 5.6). This connection defines an atlas
of local constant trivializations of the fibre bundle Q → R and the corresponding
coordinates (t, q i ) on Q such that the transition functions q i → q 0i are independent
of t, and Γi = 0 with respect to (t, q i ). In these coordinates, the Lagrangian (4.9.15)
reads
1
L = mij q it q jt + φ0 (q ν (q µ )). (4.9.16)
2
Let us assume that φ0 is a nowhere vanishing function on Q. Then the Lagrange
equations (4.9.14) take the form
q itt = {λ i ν }q λt q νt , q 0t = 1,
where {λ i ν } are the Christoffel symbols of the metric (4.9.13), whose components
with respect to the coordinates (t, q i ) read
g00 = −2φ0 , g0i = 0, gij = −mij . (4.9.17)
Then the spatial part of the corresponding geodesic equation

0 0
q̈ = 0, q̇ = 1,
i λ ν
q̈ = {λ i ν }q̇ q̇ (4.9.18)
on the tangent bundle T Q is precisely the spatial part of the geodesic equation with
respect to the Levi–Civita connection for the metric (4.9.17) on T Q. •
This example shows that a mass tensor may be treated sometimes as a field
variable.
A Newtonian system (Q, m c, ξ) is said to be standard, if m

c is the pull-back on
1
VQ J Q of a fibre metric in the vertical tangent bundle V Q → Q in accordance with
the isomorphisms (4.1.8) and (4.1.9), i.e., the mass tensor mc is independent of the
i
velocity coordinates qt .
c in V Q → Q can be seen as a mass
It is readily observed that any fibre metric m
metric of a standard Newtonian system, given by the Lagrangian
1
L = mij (q µ )(qti − Γi )(qtj − Γj ), (4.9.19)
2
where Γ is a reference frame. If m c is a Riemannian metric, one can think of the
Lagrangian function (4.9.19) as being a kinetic energy with respect to the reference
frame Γ.
Example 4.9.4. Let us consider a system of n distinguishable particles with masses
(m1 , . . . , mn ) in a 3-dimensional Euclidean space R3 . Their positions (r1 , . . . , rn )
span the configuration space R3n . The total kinetic energy is
n
1 X
Ttot = mA | ṙA |2 ,
2 A=1
that corresponds to the mass tensor
mABij = δAB δij mA , A, B = 1, . . . n, i, j = 1, 2, 3,
on the configuration space R3n . To separate the translation degrees of freedom, one
performs a linear coordinate transformation
→ →
(r1 , . . . , rn ) 7→ ( ρ 1 , . . . , ρ n−1 , R),
→ →
where R is the centre of mass, while the n − 1 vectors ( ρ 1 , . . . , ρ n−1 ) are mass-
weighted Jacobi vectors (see their definition below) [119, 120]. The Jacobi vectors
ρA are chosen so that the kinetic energy about the centre of mass has the form
1 n−1
X → ˙
T = | ρ A |2 , (4.9.20)
2 A=1
that corresponds to the Euclidean mass tensor
mABij = δAB δij , A, B = 1, . . . n − 1, i, j = 1, 2, 3,
on the translation-reduced configuration space R3n−3 . The usual procedure for defin-
ing Jacobi vectors involves organizing the particles into a hierarchy of clusters, in
which every cluster consists of one or more particles, and where each Jacobi vector
joins the centres of mass of two clusters, thereby creating a larger cluster. A Jacobi
vector, weighted by the square root of the reduced mass of the two clusters it joins,
is the above-mentioned mass-weighted Jacobi vector. For example, in the four-body
problem, one can use the following clustering of the particles:
→ √
ρ 1 = µ1 (r2 − r1 ),
→ √
ρ 2 = µ2 (r4 − r3 ),
√ m3 r3 + m4 r4 m1 r1 + m2 r2

→
ρ 3 = µ3 − ,
m3 + m4 m1 + m2
1 1 1 1 1 1 1 1 1
= + , = + , = + .
µ1 m1 m2 µ2 m3 m4 µ3 m1 + m2 m3 + m4
Different clusterings lead to different collections of Jacobi vectors, which are related
by linear transformations. Since these transformations maintain the Euclidean form
(4.9.20) of the kinetic energy, they are elements of the group O(n − 1), called the
”democracy group”. •
Now let us turn to the conditions for a Newtonian system to be a Lagrangian one.
This is the well-known inverse problem formulated for time-dependent mechanics.
We will investigate it as the particular inverse problem for dynamic systems on fibre
bundles [57].
The equation (4.9.6) is the kernel of the second order differential Euler–Lagrange
type operator
E : J 2 Q → V ∗ Q, (4.9.21)
k
E = mik (ξ − qttk )dq i .
One can write such an operator as a 2-form
E = Ei θi ∧ dt, (4.9.22)
where θi are contact forms (1.3.6). Obviously, the class of Euler–Lagrange type
operators includes the Euler–Lagrange operators (4.8.6). Thus, the inverse problem
reduces to the conditions for the Euler–Lagrange type operators to be the Euler–
Lagrange ones.
Given a fibre bundle Q → R, we have the so called variational sequence
d δ δ δ
0 → R → O0 (Q) →
H
O0,1 (J 1 Q) → O1,1 (J 2 Q) →
2
O2,1 (J 3 Q) →
3
···, (4.9.23)
where O0,1 (J 1 Q) is the space of horizontal densities Ldt on J 1 Q, while O1,1 (J 2 Q)

is the space of 2-forms Ei dθi ∧ dt on J 2 Q, and so on [46, 57, 179]. In particular,
the elements of O0,1 (J 1 Q) are the Lagrangians Ldt, while those of O1,1 (J 2 Q) are
the Euler–Lagrange-type operators written as 2-forms (4.9.22). The key point lies
in the fact that the variational sequence (4.9.23) is a complex. It implies that
δ ◦ dH = 0, (4.9.24)
δ2 ◦ δ = 0, (4.9.25)
where
dH (f ) = dt f dt, f ∈ O0 (J 1 Q),
is the horizontal exterior differential,
δ(Ldt) = (∂i − dt ∂it )Lθi ∧ dt (4.9.26)
is the Euler–Lagrange map (or the variational operator), and
δ2 (Ei θi ∧ dt) = [(2∂j − dt ∂jt + d2t ∂jtt )Ei θj ∧ θi +

(∂jt Ei + ∂it Ej − 2dt ∂jtt Ei )θti ∧ θj + (∂jtt Ei − ∂itt Ej )θttj ∧ θi ] ∧ dt = 0
is the Helmholtz–Sonin map.

A glance at the variational operator (4.9.26) shows that the image Im δ of the
variational operator (4.9.26) consists of the Euler–Lagrange operators, and Im δ ⊂
Ker δ2 in accordance with the equality (4.9.25). It follows that
δ2 (E) = 0 (4.9.27)
is the necessary condition for an Euler–Lagrange type operator E to be an Euler–

Lagrange one
E = δ(L) (4.9.28)
for some Lagrangian L. The condition (4.9.27) takes the coordinate form
1
∂j Ei − ∂i Ej + dt (∂it Ej − ∂jt Ei ) = 0, (4.9.29)
2
∂j Ei + ∂i Ej − 2dt ∂jtt Ei = 0,
t t
(4.9.30)
∂jtt Ei − ∂itt Ej = 0. (4.9.31)
The obstruction preventing the condition (4.9.27) from implying the equality (4.9.28)
is topological [46, 179]. If
Q = Rm+1 → R
(in particular, locally), the variational sequence (4.9.23) is exact, i.e., Im δ = Ker δ2 ,
and the equality (4.9.27) is also a sufficient condition for an Euler–Lagrange type
operator E to be an Euler–Lagrange one.
Remark 4.9.5. A glance at the equality (4.9.24) shows that the Lagrangian L in
the equality (4.9.28) is defined modulo the variationally trivial Lagrangians
L = dt f dt (4.9.32)
where f is a function on Q. •
Applying the condition (4.9.27) to the operator (4.9.21), we restate the well-
known Helmholtz conditions (see [34, 78, 81, 143, 156] and references therein). It
is readily observed, that the condition (4.9.31) is satisfied since the mass tensor is
symmetric. The condition (4.9.30) holds due to the equality (4.9.4) and the property
(4.9.5). Thus, it suffices to verify the condition (4.9.29) for a Newtonian system to
be a Lagrangian one.
Example 4.9.6. Let ξ be a free motion equation which takes the form (4.6.9)
with respect to a reference frame (t, q i ), and let m
c be a mass tensor which depends
only on the velocity coordinates q it . Such a mass tensor may exist in accordance
with affine coordinate transformations (4.6.10) which maintain the equation (4.6.9).
Then ξ and m c make up a Newtonian system. This system is a Lagrangian one if
c is constant with respect to the above-mentioned reference frame (t, q i ). Relative
m
to arbitrary coordinates on a configuration space Q, the corresponding Lagrangian
takes the form (4.9.19), where Γ is the connection associated with the reference
frame (t, q i ). •
Example 4.9.7. Let us consider the 1-dimensional motion of a point mass m0

subject to friction. It is described by the equation
m0 qtt = −kqt , k > 0, (4.9.33)

on the configuration space R2 → R, coordinated by (t, q). This mechanical system

is characterized by the mass function m = m0 and the holonomic connection
k
ξ = ∂t + qt ∂q − qt ∂ t , (4.9.34)
m q
but it is neither a Newtonian nor a Lagrangian system. The conditions (4.9.29)

and (4.9.31) are satisfied for an arbitrary mass function m(t, q, qt ), whereas the
conditions (4.9.4) and (4.9.30) take the form
−kqt ∂qt m − km + ∂t m + qt ∂q m = 0. (4.9.35)
The mass function m = const. fails to satisfy this relation. Nevertheless, the
equation (4.9.35) has a solution
" #
k
m = m0 exp t . (4.9.36)
m0
The mechanical system characterized by the mass function (4.9.36) and the holo-
nomic connection (4.9.34) is both a Newtonian and a Lagrangian system with the
Havas Lagrangian
" #
1 k
L = m0 exp t qt2 (4.9.37)
2 m0
[151]. The corresponding Lagrange equations are equivalent to the equation of

motion (4.9.33). •
Example 4.9.8. [143]. The dynamic equation
xtt − yt = 0, ytt − y = 0
on the configuration space R3 → R is not equivalent to any system of Lagrange

equations. •
4.10. HOLONOMIC CONSTRAINTS 209
4.10 Holonomic constraints

Let (Q, m
c, ξ) be a Newtonian system, where m
c is a Riemannian metric in the vertical
1 1
tangent bundle VQ J Q → J Q. This Section is devoted to the restriction of (Q, m
c, ξ)
to a closed imbedded fibred submanifold iN : N ,→ Q of the configuration bundle
Q → R, which is treated as a holonomic constraint. We refer the reader to [24]
and references therein for the geometric description of holonomic constraints in
conservative mechanics.
With a holonomic constraint N ⊂ Q, we have the following imbedding diagrams
J 2 N ,→ J 2 Q
V N ,→ V Q
? ?
J 1 N ,→ J 1 Q
? ?
N ,→ Q ,
? ?
&. N ,→ Q
R &.
R
Furthermore, there exists an open tubular neighbourhood U of the submanifold N ,
which is a fibred manifold over N [109]. Therefore, one can provide the configuration
space Q with the atlas of bundle coordinates (t, σ r , q i ) such that
q i |N = 0, qti |J 1 N = 0, qtti |J 2 N = 0.
We will continue to use the notation q λ for the whole coordinate collection (t, σ r , q i ).
Let nb be the induced Riemannian fibre metric in the vertical tangent bundle
VN J N → J 1 N . Its components are
1
nrs = mrs |J 1 N .
Since
∂pt nrs = ∂pt mrs |J 1 N ,
the Riemannian metric nb satisfies the symmetry condition (4.9.5).
Using the Riemannian metric m c in VQ J 1 Q → J 1 Q, we can define the orthocom-
plement V → J 1 N of the subbundle VN J 1 N ⊂ VQ J 1 Q |J 1 N and the corresponding
splitting
VQ J 1 Q |J 1 N = VN J 1 N ⊕ V (4.10.1)
J 1N
together with the canonical projections
pr1 : VQ J 1 Q |J 1 N → VN J 1 N,
pr1 (∂rt ) = ∂rt , pr1 (∂it ) = nrs msi ∂rt , (4.10.2)
and
pr2 : VQ J 1 Q |J 1 N → V,
pr2 (∂rt ) = 0, pr2 (∂it ) = ∂it − nrs msi ∂rt = ϑi .
Since the restriction J 2 Q |J 1 N → J 1 N is an affine bundle modelled over the vector

bundle VQ J 1 Q |J 1 N , the splitting (4.10.1) defines the corresponding decomposition
J 2 Q |J 1 N = J 2 N ⊕ V.
J 1N
Then the dynamic equation ξ : J 1 Q → J 2 Q of our Newtonian system splits in

the following way:
ξ |J 1 N = ξN + r, (4.10.3)
where
ξN : J 1 N → J 2 N,
r
ξN = (ξ r + nrs msi ξ i ) |J 1 N , (4.10.4)
and
r = ξ i ϑi : J 1 N → V. (4.10.5)
It follows that the dynamic equation ξ on the configuration space Q induces the
dynamic equation ξN (4.10.3) on the holonomic constraint N . In an equivalent way,
the induced dynamic equation ξN is characterized by the relation
s
nrs ξN = (mrs ξ s + mri ξ i ) |J 1 N . (4.10.6)
Proposition 4.10.1. Let (Q, m

c, ξ) be a Newtonian system on a configuration space
Q, and N ⊂ Q a holonomic constraint. Then (N, nb , ξN ) is a Newtonian system. 2
Proof. The proof of the compatibility condition (4.9.4) for ξN and n

b follows from
direct computation. QED
Let γξ be the dynamic connection (4.3.10) associated with a dynamic equation

ξ of the Newtonian system in question. It can be represented by the tangent valued
form (1.4.4) which reads
γξ : J 1 Q → T ∗ J 1 Q ⊗ VQ J 1 Q,
J 1Q
γξ = (dqti − γλi dq λ ) ⊗ ∂i + (dσtr − γλr dq λ ) ⊗ ∂r .
Let us consider the map

γξ
γeξ : J 1 N ,→ J 1 Q |J 1 N −→ T ∗ J 1 Q ⊗ VQ J 1 Q |J 1 N −→
J 1Q
T ∗ J 1 N ⊗ VN J 1 N,
J 1N
where the dual morphism (1.1.3)
(J 1 iN )∗ : T ∗ J 1 Q |J 1 N → T ∗ J 1 N
and the morphism pr1 (4.10.2) are utilized. We obtain
γeξ = (dσtr − γe0r dt − γesr dσ s ) ⊗ ∂r , (4.10.7)

γe0r = (γ0r + nrs msi γ0i ) |J 1 N , γepr = (γpr + nrs msi γpi ) |J 1 N .
It follows that the map γeξ is a connection on the affine jet bundle J 1 N → N .
Proposition 4.10.2. The dynamic connection γeξ (4.10.7), induced on the jet
bundle J 1 N → N of the holonomic constraint N by the dynamic connection γξ ,
defines a dynamic equation on N which is precisely ξN (4.10.4), induced on N by
the dynamic equation ξ. 2
Proof. The proof is straightforward. QED
It is readily observed that the dynamic connection γeξ differs in a torsion from
the symmetric dynamic connection γN , associated with the dynamic equation ξN .
1 1
γe0r = − T0r + γN r0 , γesr = Tsr + γN rs ,
2 2
Tsr = −∂st nrh mhi ξ i − nrh ∂st mhi ξ i . (4.10.8)
Let now a Newtonian system (Q, m c, ξ) be a Lagrangian system for a regular

Lagrangian L on the velocity phase space J 1 Q. The pull-back LN = i∗N L of L by
the imbedding J 1 iN is a Lagrangian on the holonomic constraint N . We come to

the following assertion.
Proposition 4.10.3. The dynamic equation ξLN , associated with the induced
Lagrangian LN on N , coincides with the dynamic equation ξLN (4.10.4) induced on
N by the dynamic equation ξL , associated with the Lagrangian L. 2
Proof. We obtain the relation
(mrs ξLs + mri ξLi ) |J 1 N = −∂t ∂rt LN − σts ∂s ∂r LN + ∂rt LN = nrs ξLs N
(see (4.9.2)), where
nrs = mrs |J 1 N = ∂rt ∂st LN .
This is precisely the relation (4.10.6). QED
It is readily observed that the Lagrange equations for a Lagrangian LN on the

holonomic constraint N are locally equivalent to those of the local Lagrangian
L + λi y i on Q, where λi are the Lagrange multipliers. Consequently, the dynamic
equation ξLN = ξLN describes a motion on an ideal holonomic constraint N . This
fact also remains true for an arbitrary Newtonian system. Indeed, the splitting
(4.10.3) shows that one can think of the dynamic equation ξN as being the dynamic
equation ξ at the points of N plus the additional acceleration −r (4.10.5), treated as
a constraint reaction acceleration. Then, bearing in mind the isomorphism (4.1.8),
it is readily observed that this acceleration is orthogonal to the holonomic constraint
N with respect to the mass metric m c, i.e.,
c(u, s) = mhj uh ξ j − mhr uh nrs msj ξ j = 0

m
for any local vertical tangent vector field uh ∂h on N → R.

Thus, we generalize the D’Alembert principle (the principle of virtual displace-
ments) for an arbitrary Newtonian system. In accordance with this principle, a
constraint reaction acceleration of a Newtonian system on an ideal holonomic con-
straint is orthogonal to the constraint manifold with respect to the mass metric of
this system.
One can say more when (Q, m c, ξ) is a standard Newtonian system, i.e., m
c is a
Riemannian metric in the vertical tangent bundle V Q → Q. This metric induces
b on the vertical tangent bundle V N → N . Its components

the Riemannian metric n
are
nrs = mrs |N .
We have the splitting
T Q |N = T N ⊕ V
N
with the canonical projections
pr1 (∂r ) = ∂r , pr1 (∂i ) = nrs msi ∂r ,

pr2 (∂r ) = 0, pr2 (∂i ) = ∂it − nrs msi ∂r = ϑi ,
and the corresponding splitting of the first order jet manifold
J 1 Q |N = J 1 N ⊕ V.
N
In particular, any reference frame Γ on a configuration space Q defines a reference

frame ΓN on the holonomic constraint N in accordance with the decomposition
Γ |N = ΓN + σV ,
nrs ΓsN = (mrs Γs + mri Γi ) |N . (4.10.9)
Moreover, the expression (4.10.8) shows that, in the case of a standard Newtonian
system, the dynamic connection γeξ (4.10.7) is precisely the dynamic connection γξN ,
associated with the dynamic equation ξN on the holonomic constraint N .
Example 4.10.1. Let (Q, m c, ξ) be a Newtonian system for a free motion equation
ξ. Then, ξN fails to be a free motion equation on a holonomic constraint N in
general. In particular, if L is a Lagrangian (4.9.19), its restriction to a holonomic
constraint N reads
1 1
LN = nrs (σtr − ΓrN )(σts − ΓsN ) + mc(σV , σV ),
2 2
where ΓN is the induced frame (4.10.9). •
4.11 Non-holonomic constraints

Here we are concerned with the geometric theory of non-holonomic constraints in
time-dependent mechanics. We refer the reader to [25, 30, 112, 113, 124, 183, 184]
and references therein for the geometric description of non-holonomic constraints
in conservative mechanics. The key problem is that the method of the Lagrange
multipliers is not appropriate to non-holonomic constraints in general (see [7, 25,
112, 114]).
Let the jet manifold J 1 Q be a velocity phase space of time-dependent mechanics
on a configuration bundle Q → R. The most general non-holonomic constraints
considered in the literature are given by codistributions S or, accordingly, by distri-
butions Ann (S) on the jet manifold J 1 Q [55, 132]. In conservative mechanics on a
configuration space M , this is the case of a codistribution on T M [183]. Submani-
folds of the jet manifold J 1 Q [88, 106] and distributions on a configuration space Q
[114] can also be seen as non-holonomic constraints.
In connection with non-holonomic constraints in time-dependent mechanics, one
usually studies the following problem. Given a configuration space Q, let ξ be a
dynamic equation on Q, and S a codistribution on J 1 Q whose annihilator Ann (S)
is treated as a non-holonomic constraint. Similarly to the case of holonomic con-
straints, the goal is a decomposition
ξ = ξe + r, (4.11.1)
where ξe is a dynamic equation obeying the condition
ξe ⊂ Ann (S).
Then one can think of the dynamic equation ξe as describing a mechanical system
subject to the non-holonomic constraint S, while (−r) is said to be the constraint
reaction acceleration.
Let us assume that the codistribution S has dimension n and is locally spanned
by the 1-forms
sa = sa0 dt + sai dq i + ṡai dqti
on the jet manifold J 1 Q. Then a dynamic equation ξe is compatible with the non-
holonomic constraint S if
sa (ξ) e a = sa + sa q i + ṡa ξei = 0.

e = ξcs
0 i t i
4.11. NON-HOLONOMIC CONSTRAINTS 215
This equation is algebraically solvable for n components of ξe if and only if the n × m

matrix ṡai (q λ , qti ) has everywhere maximal rank n ≤ m. Therefore, we will restrict
our consideration to the non-holonomic constraints, called admissible, such that
dim S = dim vb(S),
where vb∗ is the vertical endomorphism (4.1.14).

Example 4.11.1. Note that holonomic constraints can also be seen as the non-
holonomic ones which, however, are not admissible. •
If a non-holonomic constraint is admissible, there exists a local m × n matrix

ṡia (q λ , qti )
such that
ṡia ṡbi = δab . (4.11.2)
Then, the local decomposition (4.11.1) of a dynamic equation ξ can be written in

the form
ξ i = ξei + ṡia sa (ξ). (4.11.3)
The global decomposition (4.11.1) exists by virtue of the following lemma.
Lemma 4.11.1. [55]. The intersection
F = J 2 Q ∩ Ann (S)
is an affine bundle over J 1 Q, modelled over the vector bundle
F = VQ J 1 Q ∩ Ann (S).
Proof. The intersection F consists of the vertical vectors v i ∂it ∈ VQ J 1 Q which fulfill
the conditions
ṡai (q λ , qtj )v i = 0.
Since the non-holonomic constraint S is admissible, every fibre of F is of dimension

m − n, i.e., F is a vector bundle, while F is an affine bundle. QED
Since the intersection J 2 Q ∩ Ann (S) → J 1 Q is an affine bundle, it has always a

global section ξe by virtue of Theorem 1.1.2.
To construct the global decomposition (4.11.1), one should perform a splitting

of the vertical tangent bundle
VQ J 1 Q = F ⊕ V (4.11.4)
J 1Q
in order to obtain the corresponding splitting of the second order jet manifold
J 2Q = F ⊕ V (4.11.5)
J 1Q
which is modelled over (4.11.4). Here, V → J 1 Q should be interpreted as the bundle

of possible constraint reaction accelerations.
If an admissible non-holonomic constraint S is of dimension n = m, a dynamic
equation ξ is decomposed in a unique fashion. If n < m, the decomposition (4.11.1)
is indeterminate. Different variants of this decomposition lead to different con-
straint reaction accelerations which, from the physical viewpoint, characterize dif-
ferent types of non-holonomic constraints.
Let us turn now to some important particular cases of non-holonomic constraints.
Example 4.11.2. Let N be a closed imbedded submanifold of the velocity phase
space J 1 Q, defined locally by the equations
f a (q λ , qti ) = 0, a = 1, . . . , n < m. (4.11.6)
One can treat N as a non-holonomic constraint at points N ⊂ J 1 Q, given by the

codistribution S = Ann (T N ) on J 1 Q |N . This codistribution is locally spanned by
the 1-forms
sa = df a = ∂t f a dt + ∂j f a dq j + ∂jt f a dqtj .
The non-holonomic constraint N is admissible if and only if the matrix (∂jt f a ) is

of maximal rank n. It follows that N is a fibred submanifold of the affine jet
bundle J 1 Q → Q. Moreover, it is possible to separate locally some velocities qta ,
a = 1, . . . , n, expressed as functions of the remaining coordinates (q λ , qtn+1 , . . . , qtm )
such that the equations (4.11.6) are brought into the form
qta − g a (q λ , qtn+1 , . . . , qtm ) = 0. (4.11.7)
Then we have
sa0 = −∂t g a , sai = −∂i g a , ṡai = δia − ∂it g a , (4.11.8)

where ∂bt g a = 0 for all indices a, b = 1, . . . , n. Now, applying the above procedure to
the codistribution (4.11.8), one can obtain a dynamic equation on a submanifold N ,
compatible with the constraint Ann (T N ). For instance, one can use the particular
solution ṡia = δai of the equations (4.11.2), where the matrix ṡai is given by the
expression (4.11.8) [106]. Then the (local) dynamic equation ξe (4.11.3), compatible
with the constraint (4.11.7), reads
ξei6=a = ξ i , ξea = ξ a − sa (ξ) = ∂k g a ξ k + ∂t g a + qtk ∂k g a . (4.11.9)
Example 4.11.3. A non-holonomic constraint N is said to be linear if it is an affine

subbundle of the affine jet bundle J 1 Q → Q, which is given by the local equations
f a = f0a (q λ ) + fia (q λ )qti = 0, (4.11.10)
where the matrix fia is of maximal rank. A linear non-holonomic constraint is always
admissible. Since N is an affine subbundle of J 1 Q → Q, it has a global section Γ
which is a connection on the configuration bundle Q → R, called the constraint
reference frame. With this connection Γ, the constraints (4.11.10) take the form
fia (q λ )(qti − Γi ) = 0. (4.11.11)
We may say that the linear constraint is stationary with respect to the constraint
reference frame Γ. Then, one can think of
q̇Γi = qti − Γi ,
satisfying the equation (4.11.11), as virtual velocities relative to the linear constraint
N. •
Example 4.11.4. Let a configuration space Q admit a composite fibration Q →

Σ → R, where
πQΣ : Q → Σ
is a fibre bundle, and let (t, σ r , q a ) be coordinates on Q, compatible with this fibra-
tion. Given a connection
B = dt ⊗ (∂t + B a ∂a ) + dσ r ⊗ (∂r + Bra ∂a ) (4.11.12)

on the fibre bundle Q → Σ, we obtain the corresponding horizontal splitting (1.4.1)

of the tangent bundle T Q. Restricted to the jet manifold J 1 Q ⊂ T Q, this splitting
reads
∗
J 1 Q = B(πQΣ J 1 Σ) ⊕ VΣ Q,
Q
∂t + σtr ∂r + qta ∂a = [(∂t + B a ∂a ) + σtr (∂r + Bra ∂a )] +
[qta − B a − σtr Bra ]∂a ,
∗
where πQΣ J 1 Σ is the pull-back of the affine jet bundle J 1 Σ → Σ onto Q. It is readily
observed that
∗
N = B(πQΣ J 1 Σ)
is an affine subbundle of the affine jet bundle J 1 Q → Q, defined locally by the

equations
qta − σtr Bra (q λ ) − B a (q λ ) = 0
(cf. (4.11.7)). Then this subbundle yields a linear non-holonomic constraint, given
by the codistribution S = Ann (T N ) [162, 163]. This codistribution is locally
spanned by the 1-forms
sa = −(∂t B a + σtr ∂t Bra )dt − (∂s B a + σtr ∂s Bra )dσ s − (4.11.13)

(∂b B a + σtr ∂b Bra )dq b + dqta − Bra dσtr .
With the connection (4.11.12), we also have the splitting (1.6.11) of the vertical
tangent bundle V Q of Q → R and the corresponding splitting of the vertical tangent
bundle VQ J 1 Q (see the canonical isomorphism (4.1.21)). The latter splitting reads
VQ J 1 Q = F ⊕ V,
J 1Q
σ̇tr ∂rt + q̇ta ∂at = σ̇tr (∂rt + Bra ∂at ) + (q̇ta − Bra σ̇tr )∂at . (4.11.14)
It is readily observed that F |N consists of vertical vectors which are the annihilators
of the codistribution (4.11.13). The splitting (4.11.14) yields the corresponding
splitting (4.11.5) of the second order jet manifold J 2 Q. Then, given a dynamic
equation ξ on J 1 Q, we obtain the decomposition (4.11.1) which reads
ξer = ξ r , ξea = ξ a − sa (ξ)

(cf. (4.11.9)). •
Now we consider Newtonian systems because they provide the vertical tangent
bundle VQ J 1 Q with a non-degenerate fibre metric m
c. Let us assume that m c is a
Riemannian metric. With this metric, we immediately obtain the splitting (4.11.4),
where V is the orthocomplement of F . Then the decomposition (4.11.3) takes the
form
ξ i = ξei + m
fab mij ṡaj sb (ξ), (4.11.15)
where m
fab is the inverse matrix of
fab = ṡai ṡbj mij

m
[55]. By definition, the decomposition (4.11.15) satisfies the generalized D’Alembert

principle. The constraint reaction acceleration
−ri = −m
fab mij ṡaj sb (ξ) (4.11.16)
is orthogonal to every element of VQ J 1 Q∩Ann (S) with respect to the mass metric m
c.
1
Since elements of VQ J Q∩Ann (S) can be treated as the virtual accelerations relative
to the non-holonomic constraint S, the constraint reaction acceleration (4.11.16)
characterizes S as an ideal non-holonomic constraint.
The Gauss principle is also fulfilled as follows. Given a dynamic equation ξ and
the above-mentioned fibre metric m c, let us define a positive function G(w) on J 2 Q
as

c ξ(π12 (w)) − w, ξ(π12 (w)) − w ,
G(w) = m
G(q λ , qti , qtti ) = mij (q λ , qtk )(ξ i (q λ , qtk ) − qtti )(ξ j (q λ , qtk ) − qttj ).
We say that
q
kwk = G(w)
is a norm of w ∈ J 2 Q.
Proposition 4.11.2. [55]. Among all dynamic equations compatible with a

non-holonomic constraint, the dynamic equation ξe defined by the decomposition
(4.11.15) is that of least norm. 2
Proof. Let ζ be another dynamic equation which takes its values into F . Then
ξe − ζ ⊂ F and
c(ξe − ζ, ξ − ξ)
m e = 0.
Then we obtain
kζk = m
c(ξ − ξe + ξe − ζ, ξ − ξe + ξe − ζ) = kξk c(ξe − ζ, ξe − ζ),
e +m
i.e., kζk > kξk.

e QED
Now we will show that, in the case of non-degenerate Lagrangian system and lin-
ear non-holonomic constraints, the decomposition (4.11.15) satisfies the traditional
D’Alembert principle.
Let S be an admissible non-holonomic constraint on the velocity phase space
J Q, and L a regular Lagrangian on J 1 Q with a Riemannian mass metric mij = πij .
1
Since this is a particular Newtonian system, we obtain the dynamic equation
qtti = ξLi − m
fab mij ṡaj (ṡbk ξL
k
+ sbk qtk + sb0 ), (4.11.17)
ξLi = mij (−∂t πj − ∂k πj qtk + ∂j L),
which is compatible with the constraint S, treated as an ideal non-holonomic con-

straint. This is the system of Lagrange equations in the presence of a non-Lagrangian
external force
fab ṡai sb (ξL )
Fi = −m
(see the equation (4.12.9) below). It is a constraint reaction force. Let us consider
the energy-momentum conservation law in the presence of this force. It reads
LΓ L − q̇Γi fi = −dt TΓ ,
where LΓ L is the Lie derivative of the Lagrangian L along a reference frame Γ and
TΓ is the energy function relative to this reference frame (see (4.12.15), (4.12.16),
and (4.12.19) below). In particular, if a non-holonomic constraint is linear and Γ
is a constraint reference frame (see Example 4.11.3), the constraint reaction force
does not contribute to the energy conservation law. It follows that, in this case,
the standard D’Alembert principle holds, while the equation (4.11.17) describes a
motion in the presence of an ideal non-holonomic constraint in the sense of this
D’Alembert principle.
4.12. LAGRANGIAN CONSERVATION LAWS 221
The constrained motion equation (4.11.17) on a configuration space Q is neither

a system of Lagrange equations nor a dynamic equation of a Newtonian system.
In Section 5.11, we will show that this can be seen as Hamilton equations in the
framework of the Hamiltonian formalism extended to the configuration space V Q.
4.12 Lagrangian conservation laws

Given a Lagrangian system (J 1 Q, L), its integrals of motion can be found if a La-
grangian L possesses symmetries. In time-dependent mechanics, these integrals of
motion should be covariant under time-dependent transformations. For instance,
the canonical energy function
EL = πi qti − L (4.12.1)
fails to be such an integral of motion. There are different approaches in order to

obtain the conservation laws in Lagrangian dynamics (see [1, 29, 51, 52, 57, 160] for
the geometric methods). As in field theory, we will use the first variational formula
of the calculus of variations [57, 160].
Let L be a Lagrangian (4.8.1) on the velocity phase space J 1 Q and
u = ut ∂t + ui ∂i , ut = 0, 1, (4.12.2)
a projectable vector field (4.8.2) on the configuration bundle Q → R. By virtue of

Theorem 1.2.1, this vector field may be treated as the generator of a 1-parameter
group of local automorphisms, called gauge transformations, of the fibre bundle
Q → R. In particular, if ut = 0, the vertical vector field (4.12.2) is the generator
of vertical automorphisms of the fibre bundle Q → R projected over the identity
transformation of the base R. If ut = 1, the vector field u (4.12.2) is projected over
the standard vector field ∂t on the base R, which is the generator of the group of
translations of R.
Let us apply the first variational formula (4.8.4) to a Lagrangian L and to a
vector field u (4.12.2). On the shell (4.8.7), the identity (4.8.4) is brought into the
weak identity
ucdL ≈ −dt T, (4.12.3)

(ut ∂t + ui ∂i + dt ui ∂it )L ≈ −dt (πi (ut qti − ui ) − ut L),
where, by analogy with field theory,
T = −ucHL = πi (ut qti − ui ) − ut L (4.12.4)
is said to be the current along the vector field u. The symbol ”≈” stands throughout
for weak equalities fulfilled on-shell.
If the Lie derivative Lu L of a Lagrangian L along a vector field u vanishes,
i.e., L is invariant under the 1-parameter group generated by u, we have the weak
conservation law
0 ≈ −dt [πi (ut qti − ui ) − ut L]. (4.12.5)
It is brought into the differential conservation law

d
0≈− [(πi ◦ ċ)(ut ∂t ci − ui ◦ c) − ut L ◦ ċ]
dt
on solutions c of the Lagrange equations (4.8.8). A glance at this expression shows
that, in time-dependent mechanics, the conserved current (4.12.4) plays the role of
an integral of motion.
Remark 4.12.1. Let L be a Lagrangian on the velocity phase space J 1 Q. Every
integral of motion (4.12.4) defines a non-holonomic constraint dT = 0 on J 1 Q such
that any Lagrangian connection ξL for L is a dynamic equation compatible with this
constraint. •
Remark 4.12.2. The conservation law (4.12.5) is broken if a Lagrangian depends

on variable parameters which do not live in the dynamic shell (4.8.7) (see Section
5.9). In order to obtain a conservation law in the presence of such parameters, let
us consider a bundle product
Qtot = Q × Y (4.12.6)
of a configuration bundle Q → R with coordinates (t, q i ) and a fibre bundle Y → R

with coordinates (t, y A ), whose sections are the variable parameters which take the
background values
y B = φB (x), yλB = ∂λ φB (x).
A Lagrangian L is defined on the total velocity phase space J 1 Qtot .

Let u be a projectable vector field (4.12.2) on Qtot which projects also onto Y
because the transformations of parameters do not depend on the dynamic variables.
This vector field takes the coordinate form
u = ut ∂t + uA (t, y B )∂A + ui (t, y B , q j )∂i . (4.12.7)
Substitution of the vector field (4.12.7) into the first variational formula (4.8.4)
leads to the first variational formula in the presence of parameters
[ut ∂t + uA ∂A + ui ∂i + dt uA ∂At + dt ui ∂it ]L = (uA − ytA ut )∂A L + (4.12.8)

A
πA dt (u − ytA ut ) i
+ (u − qti ut )Ei − dt [πi (ut qti i t
− u ) − u L].
Then the weak identity
[ut ∂t + uA ∂A + ui ∂i + dt uA ∂At + dt ui ∂it ]L ≈ (uA − ytA ut )∂A L +

πA dt (uA − ytA ut ) − dt [πi (ut qti − ui ) − ut L]
is fulfilled on the dynamic shell (4.8.7).

If a Lagrangian L is invariant under gauge transformations of the product (4.12.6)
whose generator is the vector field u (4.12.7), we obtain the weak conservation law
in the presence of parameters
(uA − ytA ut )∂A L + πA dt (uA − ytA ut ) ≈ dt [πi (ut qti − ui ) − ut L].
Remark 4.12.3. The first variational formula (4.8.4) can also be utilized when
a Lagrangian possesses symmetries, but a motion equation is seen as Lagrange
equations plus additional non-Lagrangian external forces and reads
(∂i − dt ∂it )L + Fi (t, q j , qtj ) = 0. (4.12.9)
Let us substitute Ei = −Fi from this equality in the first variational formula (4.8.4)
and assume that the Lie derivative of the Lagrangian L along a vector field u van-
ishes. Then, we have the conservation law
(ui − qti )Fi ≈ −dt [πi (ut qti − ui ) − ut L]. (4.12.10)
•
It is easy to see that the weak identity (4.12.3) is linear in the vector field u.
Therefore, one can consider superposition of the weak identities (4.12.3) associated
with different vector fields.
For instance, if u and u0 are projectable vector fields (4.12.2), projected onto the
standard vector field ∂t on R, the difference of the corresponding weak identities
(4.12.3) results in the weak identity (4.12.3) associated with the vertical vector field
u − u0 . Conversely, every vector field u (4.12.2), projected onto ∂t , can be written
as the sum
u=Γ+ϑ (4.12.11)
of some reference frame
Γ = ∂t + Γi ∂i (4.12.12)
and a vertical vector field ϑ on Q.

It follows that the weak identity (4.12.3) associated with an arbitrary vector
field u (4.12.2) can be represented as the superposition of those associated with a
reference frame Γ (4.12.12) and some vertical vector field ϑ.
If u = ϑ is a vertical field, the weak identity (4.12.3) reads
(ϑi ∂i + dt ϑi ∂it )L ≈ dt (πi ϑi ).
If the Lie derivative of L along ϑ vanishes, we obtain from (4.12.5) the weak con-
servation law
0 ≈ dt (πi ϑi ) (4.12.13)
and the integral of motion
T = −πi ϑi . (4.12.14)
By analogy with field theory, (4.12.13) is called the Noether conservation law for
the Noether current (4.12.14).
Example 4.12.4. Let assume that, given a trivialization Q ∼ = R×M in coordinates
(t, q i ), a Lagrangian L is independent of a coordinate q 1 . Then the Lie derivative
of L along the vertical vector field ϑ = ∂1 equals zero, and we have the conserved
Noether current (4.12.14) which reduces to the momentum T = −π1 . With respect
to arbitrary coordinates (t, q 0i ), this conserved Noether current takes the form
∂q 0i 0
T = − 1 πi .
∂q
In particular, the free motion Lagrangian admits m conserved Noether currents. •
In the case of a reference frame Γ (4.12.12), where ut = 1, the weak identity

(4.12.3) reads
(∂t + Γi ∂i + dt Γi ∂it )L ≈ −dt (πi (qti − Γi ) − L), (4.12.15)
where
TΓ = πi (qti − Γi ) − L (4.12.16)
is said to be the energy function relative to the reference frame Γ [51, 57, 161].
With respect to the coordinates adapted to the reference frame Γ, the weak
identity (4.12.15) takes the form of the familiar energy conservation law
∂t L ≈ −dt (πi qti − L), (4.12.17)
and TΓ coincides with the canonical energy function EL (4.12.1). It follows that
the canonical energy function EL is not a unique existent energy function. Each
reference frame defines an energy function.
Example 4.12.5. Let us consider a free motion on a configuration space Q. It is
described by the Lagrangian
1
L = mij q it q jt , m = const., (4.12.18)
2
written with respect to a reference frame (t, q i ) such that the free motion dynamic
equation takes the form (4.6.1). Let Γ be the associated connection. Then the
conserved energy function TΓ (4.12.16) relative to this reference frame Γ is precisely
the kinetic energy of this free motion. Relative to arbitrary coordinates (t, q i ) on Q,
it takes the form
1
TΓ = πi (qti − Γi ) − L = mij (q µ )(qti − Γi )(qtj − Γj ).
2
Now we generalize this example for a motion described by the equation (4.12.9),
where L is the free motion Lagrangian (4.12.18) and F is an external force. The Lie
derivative of the Lagrangian (4.12.18) along the reference frame Γ vanishes, and we
have the weak equality (4.12.10) which reads
q̇Γi Fi ≈ dt TΓ , (4.12.19)
where q̇Γi is the relative velocity. This is the well-known physical law whose left-hand
side is the power of an external force. •
Example 4.12.6. Let us consider a 1-dimensional motion of a point mass m0

subject to friction on the configuration space R2 → R, coordinated by (t, q) (see
Example 4.9.7). It is described by the dynamic equation (4.9.33) which is the
system of Lagrange equations for the Lagrangian L (4.9.37). It is readily observed
that the Lie derivative (4.8.3) of this Lagrangian along the vector field
1 k
Γ = ∂t − q∂q (4.12.20)
2 m0
vanishes. Hence, we have the conserved energy function (4.12.16) with respect to
the reference frame Γ (4.12.20). This energy function reads
1 2 mk 2 2
" #
1 k k
TΓ = m0 exp t qt (qt + q) = mq̇Γ − q ,
2 m0 m0 2 8m20
where m is the mass function (4.9.36). •
Since any vector field u (4.12.2) can be represented as the sum (4.12.11) of a
reference frame Γ (4.12.12) and a vertical vector field ϑ, each current (4.12.4) along
a vector field u (4.12.2) is the sum of a Noether current (4.12.14) along the vertical
vector field ϑ and the energy function (4.12.16) relative to the reference frame Γ
[51, 57, 161]. Conversely, energy functions relative to different reference frames Γ
and Γ0 differ from each other in the Noether current along the vertical vector field
Γ − Γ0 .
In conclusion, we touch on gauge transformations with a generator u (4.12.2),
which preserve the Euler-Lagrange operator EL , but not necessarily a Lagrangian
L. They are called generalized invariant transformations [57, 105, 174]. We use the
formula
LJ 2 u EL = ELu L
[54, 57], where
J 2 u = ut ∂t + ui ∂i + dt ui ∂it + dt dt ui ∂itt
is the second order jet prolongation of a vector field (4.12.2).

The following two assertions are immediate corollaries of this formula.
Corollary 4.12.1. If a Lagrangian L is invariant under a 1-parameter gauge group

whose generator is a vector field u (4.12.2), so is the associated Euler–Lagrange
operator EL . 2
Corollary 4.12.2. If an Euler–Lagrange operator EL , associated with a Lagran-

gian L, is invariant under a 1-parameter gauge group whose generator is a vector
field u (4.12.2), the Lie derivative Lu L is a variationally trivial Lagrangian (4.9.32).
2
Let us consider conservation laws in the case of generalized invariant transfor-

mations. Let L be a Lagrangian and EL the associated Euler–Lagrange operator.
Let u be a vector field (4.12.2) which is the generator of a local 1-parameter group
of gauge transformations such that
LJ 2 u EL = 0. (4.12.21)
Then we have the equality
Lu L = dt f,
where f is a function on Q. In this case, the weak identity (4.12.3) reads
dt f ≈ dt (πi (ui − qti ) + ut L),
and we obtain the weak equality
0 ≈ dt (πi (ui − qti ) + ut L − f ). (4.12.22)
Example 4.12.7. Let L be the free motion Lagrangian (4.12.18). The correspond-
ing Euler–Lagrange operator
EL = −mij qttj dq i
is invariant under the Galilei transformations with the generator
u i = v i t + ai , v i = const., ai = const., (4.12.23)
(see (4.6.10)). At the same time, the Lie derivative of the free motion Lagrangian
(4.12.18) along the vector field (4.12.23) does not vanish, and we have
Lu L = mij v i qtj = dt (mij v i q j + c), c = const.,
[148]. Then the weak equality (4.12.22) shows that (qti t − q i ) is a constant of motion.
•
Remark 4.12.8. The invariance condition (4.12.21) is generalized as
LJ 2 u EL ≈ 0
if one deals with symmetry transformations of the differential equation EL = 0

[93, 94]. For instance, the Galilei transformations (4.6.10) are the symmetry trans-
formations of the free motion equation. •
Chapter 5
Hamiltonian time-dependent
mechanics
This Chapter is devoted to the Hamiltonian formulation of time-dependent mechan-

ics with respect to an arbitrary reference frame. Let us recall that a configuration
bundle Q → R of time-dependent mechanics is isomorphic to the product R × M ,
where M is a typical fibre of Q, but this isomorphism is not canonical in general.
Different trivializations Q ∼
= R × M correspond to different reference frames. For
this reason, many constructions of Hamiltonian conservative mechanics on a sym-
plectic manifold Z and its extension to the product R × Z can not be applied to
mechanical systems which are subject to time-dependent transformations, including
canonical transformations and reference-frame transformations.
Let us summarize the main peculiarities of Hamiltonian time-dependent mechan-
ics.
• A momentum phase space of time-dependent mechanics is provided with the

canonical degenerate Poisson structure. However, Hamiltonian time-dependent
mechanics does not reduce to a Poisson Hamiltonian system, since a Hamilto-
nian on a momentum phase space of time-dependent mechanics is not a scalar
function under time-dependent transformations (see Remark 5.1.1 below).
• As a consequence, the evolution equation of time-dependent mechanics is not

expressed in terms of a Poisson bracket, and integrals of motion cannot be
defined as functions in involution with a Hamiltonian. For the same reason,
the familiar procedure of describing constraint Hamiltonian systems in Section
229
230 CHAPTER 5. HAMILTONIAN TIME-DEPENDENT MECHANICS
3.6 cannot be applied to time-dependent mechanics.
• Hamiltonian and Lagrangian formulations of time-dependent mechanics are

equivalent in the case of hyperregular Lagrangians. A degenerate Lagrangian
requires a set of associated Hamiltonians in order to exhaust all solutions of
the Lagrange equations.
We follow the notation of the previous Chapter, where Q → R is a fibre bundle

(4.0.1), coordinated by (t, q i ), whose typical fibre M is an m-dimensional manifold,
while its base R is equipped with the Cartesian coordinate t possessing the transition
functions t0 = t+ const.
5.1 Canonical Poisson structure

As was mentioned, the momentum phase space of time-dependent mechanics is the
vertical cotangent bundle
πΠ : V ∗ Q → Q (5.1.1)
of a configuration bundle Q → R. This is the particular Legendre bundle Π (2.9.7)

over a fibre bundle Q → X when X = R. The vertical cotangent bundle (5.1.1) is
equipped with the coordinates (t, q i , pi = q̇i ).
The momentum phase space V ∗ Q is provided with the canonical Poisson struc-
ture as follows (cf. Example 2.3.8). Let us consider the cotangent bundle T ∗ Q of
the configuration space Q, equipped with the coordinates (t, q i , p, pi ). This is the
homogeneous Legendre bundle (2.9.3) over a fibre bundle Q → X when X = R.
The cotangent bundle T ∗ Q admits the canonical Liouville form
Ξ = pdt + pi dq i (5.1.2)
and the canonical symplectic form
dΞ = dp ∧ dt + dpi ∧ dq i . (5.1.3)
The corresponding Poisson bracket on the vector space O0 (T ∗ Q) of functions on

T ∗ Q reads
{f, g} = ∂ p f ∂t g − ∂ p g∂t f + ∂ i f ∂i g − ∂ i g∂i f. (5.1.4)

5.1. CANONICAL POISSON STRUCTURE 231
Let us consider the subspace of O0 (T ∗ Q) which comprises the pull-backs ζ ∗ f on

T ∗ Q of functions f on the vertical cotangent bundle V ∗ Q by the canonical projection
T ∗ Q → V ∗ Q (1.1.7). It is easily seen that this subspace is closed under the Poisson
bracket (5.1.4). By virtue of Proposition 2.3.1, there exists the canonical Poisson
structure
{f, g}V = ∂ i f ∂i g − ∂ i g∂i f (5.1.5)
on momentum phase space V ∗ Q induced by (5.1.4), i.e.,
ζ ∗ {f, g}V = {ζ ∗ f, ζ ∗ g}.
The corresponding Poisson bivector field
w(df, dg) = {f, g}V
on V ∗ Q → R is vertical with respect to the fibration V ∗ Q → R, and reads
wij = 0, wij = 0, wi j = 1. (5.1.6)
A glance at this expression shows that the holonomic coordinates of V ∗ Q are canon-
ical for the Poisson structure (5.1.5). Since the rank of the bivector field w (5.1.6) is
constant, the Poisson structure (5.1.5) is regular. This Poisson structure is obviously
degenerate.
Given the Poisson bracket (5.1.5), the Hamiltonian vector field ϑf for a function
f on the momentum phase space V ∗ Q is defined by the relation
{f, g}V = ϑf cdg, ∀g ∈ O0 (V ∗ Q).
It is the vertical vector field
ϑf = ∂ i f ∂i − ∂i f ∂ i (5.1.7)
on the fibre bundle V ∗ Q → R. Hence, the characteristic distribution of the Poisson

structure (5.1.5) is precisely the vertical tangent bundle V V ∗ Q ⊂ T V ∗ Q of the fibre
bundle V ∗ Q → R.
In accordance with Theorem 2.3.3, the Poisson structure (5.1.5) defines the sym-
plectic foliation on the momentum phase space V ∗ Q, which coincides with the fibra-
tion V ∗ Q → R. The symplectic forms on the fibres of V ∗ Q → R are the pull-backs
Ωt = dpi ∧ dq i
of the canonical symplectic form on the typical fibre T ∗ M of the fibre bundle V ∗ Q →
R with respect to trivialization morphisms [27, 74, 161]. Given such a trivialization
V ∗Q ∼
= R × T ∗ M, (5.1.8)
the Poisson structure (5.1.5) is isomorphic to the direct product of the zero Poisson
structure on R and the canonical symplectic structure on T ∗ M (see Example 2.3.7).
The Poisson structure (5.1.5) can be introduced in a different way [57, 161].
The polysymplectic form (2.9.9) on the Legendre bundle V ∗ Q → Q (5.1.1) reads
Λ = dpi ∧ dq i ∧ dt ⊗ ∂t ,
and reduces to the canonical exterior 3-form
Ω = dpi ∧ dq i ∧ dt. (5.1.9)
This is the exterior differential of the canonical 2-form
Θ = pi dq i ∧ dt (5.1.10)
on the momentum phase space V ∗ Q. The canonical forms (5.1.9) and (5.1.10) are
maintained under any holonomic coordinate transformation of V ∗ Q.
Given the canonical form (5.1.9), every function f on the momentum phase
space V ∗ Q determines the corresponding Hamiltonian vector field ϑf (5.1.7) by the
relation
ϑf cΩ = −df ∧ dt, (5.1.11)
while the Poisson bracket (5.1.5) is defined by the condition
{f, g}V dt = ϑg cϑf cΩ.
Remark 5.1.1. It should be emphasized that, though the momentum phase space
V ∗ Q of time-dependent mechanics is provided with the canonical Poisson structure,
non-conservative mechanical systems are not reduced to the Poisson Hamiltonian
systems of Section 3.2. Given a trivialization (5.1.8), i.e., a reference frame, one
can write the Hamilton equations (5.2.20) – (5.2.21) (see below) as the equations
of the Hamiltonian vector field ϑH (5.1.7) for a Hamiltonian H with respect to the
Poisson structure (5.1.5). Moreover, by very definition of the canonical form Ω
(5.1.9), the reference frame transformations are canonical transformations for the
5.2. HAMILTONIAN CONNECTIONS AND HAMILTONIAN FORMS 233
Poisson structure (5.1.5) (see Section 5.3). However, a Hamiltonian H (see (5.2.12)
below) is not a scalar function under reference frame transformations in general.
Therefore these transformations are not the equivalence transformation of a Poisson
Hamiltonian system. •
5.2 Hamiltonian connections and Hamiltonian forms

The dynamics of time-dependent mechanics on a phase phase space V ∗ Q is described
by first order dynamic equation.
Definition 5.2.1. Let
γ = ∂t + γ i ∂i + γi ∂ i (5.2.1)
be a connection on the fibre bundle V ∗ Q → R. In accordance with Definition 4.2.1,

it defines the first order dynamic equation on the momentum phase space V ∗ Q → R
written as
qti = γ i , pti = γi
with respect to the adapted coordinates
(t, q i , pi , qti , pti )
on the jet manifold J 1 V ∗ Q. 2
Example 5.2.1. Let us consider the product
Q = R × M → R,
coordinated by (t, q i ). We have the canonical isomorphism
V ∗ Q = R × T ∗ M,
with the coordinates pi = q̇i , and the corresponding isomorphism
J 1 V ∗ Q = R × T T ∗ M, (5.2.2)
with the coordinates qti = q̇ i , pti = ṗi . Every vector field
u = ui ∂i + ui ∂ i
on the cotangent bundle T ∗ M defines the first order dynamic equation
γ = ∂t + ui ∂i + ui ∂ i
on V ∗ Q (5.2.2). Its projection
q̇ i = ui , ṗi = ui
onto T ∗ M is an autonomous first order dynamic equation in conservative mechanics

(see Definition 3.1.1). •
Let J 2 Q be the second order jet manifold of a configuration space Q → R

provided with the adapted coordinates
(t, q i , qti , qtti ).
Recall that it coincides with the sesquiholonomic jet manifold Jb2 Q.

Following the general scheme of polysymplectic formalism in field theory (see
Section 3.8), we say that a connection (5.2.1) on the Legendre bundle V ∗ Q → R is
locally Hamiltonian if the exterior form γcΩ is closed, i.e.,
Lγ Ω = d(γcΩ) = 0. (5.2.3)
It is readily observed that a connection γ (5.2.1) on V ∗ Q → R is locally Hamiltonian

if and only if it obeys the conditions
∂ i γ j − ∂ j γ i = 0, (5.2.4)
∂i γj − ∂j γi = 0, (5.2.5)
∂j γ i + ∂ i γj = 0. (5.2.6)
Example 5.2.2. Every connection
Γ : Q → J 1 Q ⊂ T Q,
Γ = ∂t + Γi ∂i , (5.2.7)
on the configuration bundle Q → R gives rise to the locally Hamiltonian connection
e = V ∗ Γ = ∂ + Γi ∂ − p ∂ Γi ∂ j
Γ (5.2.8)
t i i j
(1.6.15) on the momentum phase space V ∗ Q → R such that
ΓcΩ
e = dHΓ ,
HΓ = pi dq i − pi Γi dt. (5.2.9)
•
Locally Hamiltonian connections constitute an affine space modelled over the
linear space of vertical vector fields ϑ on the Legendre bundle V ∗ Q → R, which
fulfill the same condition
d(ϑcΩ) = 0 (5.2.10)
as (5.2.3). These vector fields are locally the Hamiltonian vector fields in accordance
with the following lemma.
Lemma 5.2.2. Every closed form γcΩ on the fibre bundle V ∗ Q → R is exact. 2
Proof. Let us consider the decomposition
γ=Γ
e + ϑ, (5.2.11)
where Γ is a connection on Q → R and Γ e is its lift (5.2.8) onto the fibre bundle
V ∗ Q → R, while ϑ is a vertical vector field on V ∗ Q satisfying the relation (5.2.10).
It is easily seen that
ϑcΩ = σ ∧ dt,
where σ is a 1-form on V ∗ Q. Using the properties of the De Rham cohomology of a

manifold product, one can show that every closed 2-form σ ∧ dt on a manifold V ∗ Q,
diffeomorphic to the product R × T ∗ M , is exact, and so is γcΩ [57]. Moreover, in
accordance with the relative Poincaré lemma, we can write locally
ϑcΩ = df ∧ dt,
where f is a local function on V ∗ Q. QED
Definition 5.2.3. An exterior 1-form H on the momentum phase space V ∗ Q is

said to be a locally Hamiltonian form if
γcΩ = dH
for a connection γ on the fibre bundle V ∗ Q → R. 2
By virtue of Proposition 5.2.2, there is one-to-one correspondence between the

locally Hamiltonian connections and locally Hamiltonian forms, considered through-
out modulo closed forms. In particular, the exterior form HΓ (5.2.9) is a locally
Hamiltonian form.
Definition 5.2.4. An exterior 1-form H on the momentum phase space V ∗ Q is

called a Hamiltonian form if it is the pull-back
H = h∗ Ξ = pi dq i − Hdt (5.2.12)
of the Liouville form Ξ (5.1.2) on the homogeneous Legendre bundle T ∗ Q by a

section h of the fibre bundle
ζ : T ∗ Q → V ∗ Q. (5.2.13)
Remark 5.2.3. Note that, with respect to the universal unit system, a Hamiltonian
form is physically dimensionless. •
Remark 5.2.4. Given a trivialization Q ∼ = R×M , the Hamiltonian form (5.2.12) is

the well-known integral invariant of Poincaré–Cartan [6]. Therefore, the coefficient
H of the horizontal part of the Hamiltonian form (5.2.12) is said to be a Hamiltonian.
A glance at the expression (5.2.12) shows that Hamiltonians fail to be functions,
i.e., H 6∈ O0 (V ∗ Q), but make up an affine space modelled over the linear space of
functions on V ∗ Q. •
For instance, every connection Γ on a configuration bundle Q → R is an affine

section
p ◦ Γ = −pi Γi
of the fibre bundle (5.2.13) (see (1.3.12)), and defines the Hamiltonian form HΓ
(5.2.9) on the momentum phase space V ∗ Q, where its Hamiltonian is
H = pi Γi .
It follows that any Hamiltonian form on the momentum phase space V ∗ Q admits
the splitting
f dt = p dq i − (p Γi + H
H = HΓ − H f )dt, (5.2.14)
Γ i i Γ
where Γ is a connection on Q → R, and H f is a real function on V ∗ Q, called the

Γ
Hamiltonian function.
Since a connection Γ on a configuration space Q → R is treated as a refer-
ence frame, i.e., as a kinematic object, one may say that, in accordance with the
decomposition (5.2.14), every Hamiltonian form H is split into kinematic and dy-
namic parts, and so is its Hamiltonian H. In Section 5.7, we will show that the
Hamiltonian function H f in the splitting (5.2.14) is the Hamiltonian energy func-
Γ
tion relative to the reference frame Γ. Of course, the splitting (5.2.14) is not unique.
Nevertheless, every Hamiltonian form H admits the canonical splitting (5.2.14) as
follows.
We mean by a Hamiltonian map any fibred morphism
Φ : V ∗ Q → J 1 Q,
Q
qti i
◦ Φ = Φ (p), p ∈ V ∗ Q, (5.2.15)
over Q from the momentum phase space V ∗ Q to the velocity phase space J 1 Q. Its
composition with the canonical morphism λ (4.1.4) yields the fibred morphism
Φ : V ∗ Q → T Q,
Q
Φ = ∂t + Φi ∂i . (5.2.16)
In particular, every connection Γ on a configuration bundle Q → R defines the
Hamiltonian map
b = Γ ◦ π : V ∗ Q → Q → J 1 Q,
Γ Π
b = ∂ + Γi ∂ .
Γ t i
Conversely, every Hamiltonian map Φ (5.2.15) yields the associated connection

ΓΦ = Φ ◦ 0b
on the configuration bundle Q → R, where 0b is the global zero section of the vertical
cotangent bundle V ∗ Q → Q. For instance, we have
ΓbΓ = Γ.
Proposition 5.2.5. Every Hamiltonian map (5.2.16) defines the associated Ha-
miltonian form
HΦ = −ΦcΘ = (λ ◦ Φ)∗ Ξ = pi dq i − pi Φi dt
on V ∗ Q, where Θ is the canonical 2-form (5.1.10). 2
Proposition 5.2.6. Every Hamiltonian form H on the momentum phase space

V ∗ Q defines the associated Hamiltonian map
c : V ∗ Q → J 1 Q,
H
qti ◦ H
c = ∂ i H.
Proof. Let h : V ∗ Q ,→ T ∗ Q be a section of the fibre bundle (5.2.13), which

corresponds to the Hamiltonian form H, i.e., H = h∗ Ξ. The vertical tangent map
V h of the morphism h defines the linear fibred morphism
V h : V V ∗Q → V ∗Q × T ∗Q
Q
over V ∗ Q. Therefore, it can be represented by the section
V h = dpi ⊗ (dq i − ∂ i Hdt)
of the fibre bundle
V ∗V ∗Q ⊗
∗
T ∗ Q → V ∗ Q.
V Q
After natural contractions, this section is brought into the section
V h = (dq i − ∂ i Hdt) ⊗ ∂i
of the pull-back
V ∗ Q ×(T ∗ Q ⊗ V Q) → V ∗ Q,
Q
and takes its values into the image of the jet manifold J 1 Q by its canonical imbedding
(1.3.5) into T ∗ Q ⊗ V Q. QED
Corollary 5.2.7. Every Hamiltonian form H on the momentum phase space V ∗ Q

determines the associated connection
c◦0
ΓH = H b
on the configuration bundle Q → R. It is called a Hamiltonian frame connection.

2
ΓHΓ = Γ,
where HΓ is the Hamiltonian form (5.2.9) associated with the connection Γ on the
fibre bundle Q → R.
Corollary 5.2.8. Every Hamiltonian form H (5.2.12) on the momentum phase

space V ∗ Q admits the canonical splitting
H = HΓH − Hdt.
f
Let us turn now to the relationship between the locally Hamiltonian forms and
the Hamiltonian ones.
Proposition 5.2.9. Every locally Hamiltonian form on the momentum phase

space V ∗ Q is locally a Hamiltonian form. 2
Proof. Given two locally Hamiltonian forms Hγ and Hγ 0 on the momentum phase
space V ∗ Q, their difference
σ = Hγ − Hγ 0 ,
dσ = (γ − γ 0 )cΩ,
is a 1-form on V ∗ Q such that the 2-form σ ∧ dt is closed and, consequently, exact

by virtue of Lemma 5.2.2. Hence, in accordance with the relative Poincaré lemma,
we have
σ = f dt + dg,
where f and g are local functions on V ∗ Q. Then, we deduce from the splitting
(5.2.11) that, in a neighbourhood of every point p ∈ V ∗ Q, a locally Hamiltonian
form Hγ can be written as
Hγ = HΓ + f dt,
and coincides with the pull-back of the Liouville form Ξ on the cotangent bundle
T ∗ Q by the local section
(t, q i , pi ) 7→ (t, q i , pi , p = −pi Γi + f )
of the fibre bundle (5.2.13). QED
A converse of Proposition 5.2.9 is the following.
Proposition 5.2.10. Given a Hamiltonian form H on the momentum phase space

V ∗ Q, there exists a unique connection γH on the Legendre bundle V ∗ Q → R, called
the Hamiltonian connection, such that
γH cΩ = dH. (5.2.17)
Proof. As in the polysymplectic case, let us introduce the first order Hamilton
operator on the phase space V ∗ Q, which is associated with the Hamiltonian form
H. This Hamilton operator reads
2
EH : J 1 V ∗ Q → ∧ T ∗ V ∗ Q,
def
EH = dH − λcΩ = [(qti − ∂ i H)dpi − (pti + ∂i H)dq i ] ∧ dt, (5.2.18)
where λ is the canonical monomorphism (4.1.4) and J 1 V ∗ Q is the jet manifold of

the fibre bundle V ∗ Q → R, coordinated by
(t, q i , pi , qti , pti ).
It is readily observed that the kernel of the Hamilton operator EH (5.2.18) is an

image of the global section
γH = ∂t + ∂ i H∂i − ∂i H∂ i (5.2.19)
of the affine jet bundle J 1 V ∗ Q → V ∗ Q, which is precisely the desired Hamiltonian

connection (see Remark 5.2.6 below for a different proof). QED
Let us recall that Hamiltonian forms H on the momentum phase space V ∗ Q

make up an affine space modelled over the linear space of horizontal densities f dt
on the fibre bundle V ∗ Q → R. Then, as follows ¿from the relation (5.2.17), Hamil-
tonian connections γH constitute an affine space modelled over the linear space of
Hamiltonian vector fields (5.1.11):
H − H 0 = df, γH − γH 0 = −ϑf .
The kernel of the covariant differential DγH , associated with a Hamiltonian con-
nection (5.2.19), is a closed imbedded subbundle of the jet bundle J 1 V ∗ Q → R,
and defines the system of first order differential equations on the momentum phase
space V ∗ Q. These are called the Hamilton equations
qti = ∂ i H, (5.2.20)
pti = −∂i H (5.2.21)
for the Hamiltonian form H. Note that, in comparison with the Lagrange equations,
the Hamilton ones are always well defined differential equations. The integral sec-
tions r : R 3 () → V ∗ Q of the Hamiltonian connection (5.2.19) (or equivalently, the
integral curves of the horizontal vector field (5.2.19)) are solutions of the Hamilton
equations (5.2.20) – (5.2.21). Moreover, a glance at the equation (5.2.20) shows that
the relation
J 1 (πΠ ◦ r) = H
c◦r (5.2.22)
is fulfilled for any solution r of the Hamilton equations (5.2.20) – (5.2.21).
Remark 5.2.5. The Hamilton equations (5.2.20) – (5.2.21) are the particular case
of the first order dynamic equations
qti = γ i , pti = γi (5.2.23)
on the momentum phase space V ∗ Q → R, where γ is a connection (5.2.1) on the

fibre bundle V ∗ Q → R.
The first order reduction of the equation of a motion of a point mass m subject
to friction in Example 4.9.7 exemplifies first order dynamic equations which are not
Hamilton ones. These equations read
1 k
qt = p, pt = − p. (5.2.24)
m0 m0
The connection
1 k
γ = ∂t + p∂q − p∂p
m m
does not obey the condition (5.2.6) for a locally Hamiltonian connection. At the
same time, the equations (5.2.24) are equivalent to the Hamilton equations for the
Hamiltonian associated with the Lagrangian (4.9.37). •
Note that the Hamilton equations (5.2.20) – (5.2.21) can be introduced without
appealing to the Hamilton operator. On sections r of the fibre bundle V ∗ Q → R,
these equations
ṙi = ∂ i H ◦ r, ṙi = −∂i H ◦ r (5.2.25)
are equivalent to the relation
r∗ (ucdH) = 0 (5.2.26)
which is assumed to hold for any vertical vector field u on V ∗ Q → R.

Remark 5.2.6. Every Hamiltonian form H on a momentum phase space V ∗ Q
defines a presymplectic structure. A Hamiltonian form H = h∗ Ξ, by definition, is
a pull-back of the canonical Liouville form Ξ (5.1.2) by means of a section h of the
fibre bundle T ∗ Q → V ∗ Q. Accordingly, its exterior differential
dH = h∗ dΞ = (dpi + ∂i Hdt) ∧ (dq i − ∂ i Hdt) (5.2.27)
is the pull-back of the canonical symplectic form (5.1.3), and is a presymplectic

form. The presymplectic form (5.2.27) has constant rank 2m since the form
(dH)m = (dpi ∧ dq i )m − m(dpi ∧ dq i )m−1 ∧ dH ∧ dt (5.2.28)
is nowhere vanishing. It is also easily seen that
(dH)m ∧ dt 6= 0.
It follows that the pair (dH, dt) defines a cosymplectic structure on V ∗ Q (see Remark
4.8.8). Since the presymplectic form dH is of constant rank 2m, its kernel Ker dH is
a 1-dimensional distribution which is spanned by the vector field γH (5.2.19). Hence,
there is a unique vector field u = γH on V ∗ Q such that
ucdH = 0, ucdt = 1
(see [37]). This is a different proof of Proposition 5.2.10. •
Proposition 5.2.11. The Hamiltonian form (5.2.8) is a contact form on the mo-
mentum phase space V ∗ Q if the function
γH cH = pi ∂ i H − H = [H] (5.2.29)
nowhere vanishes [116]. 2
Proof. Since a Hamiltonian connection γH (5.2.19) is a nowhere vanishing vector

field, the condition
H ∧ (dH)m 6= 0
for a Hamiltonian form H to be a contact form is equivalent to the condition
γH c(H ∧ (dH)m ) = (γH cH)(dH)m = [H](dH)m 6= 0.
The result follows because the form (dH)m (5.2.28) is nowhere vanishing. QED
Note that one may try to add some exact form, e.g., the form cdt, c = const.,
to a Hamiltonian form H in order to make the function [H] nowhere vanishing. For
instance, the Hamiltonian form HΓ (5.2.9) fails to be a contact one since [HΓ ] = 0,
but the equivalent Hamiltonian form HΓ − dt, where [HΓ − dt] = 1, is a contact
form.
If a Hamiltonian form H is a contact one, the corresponding Reeb vector field
(2.2.5) reads
EH = [H]−1 γH . (5.2.30)
By virtue of Proposition 2.2.6, one can then introduce the Jacobi bracket defined
by the vector field (5.2.30) and by the bivector field wH on V ∗ Q derived from the
relations (2.2.8) which read
wH (φ, .)cH = 0,
wH (φ, .)cdH = −(φ − (EH cφ)H),
whenever φ is a 1-form on V ∗ Q. We find
wH (φ, σ) = φi σi − σ i φi + pi σ i EH cφ − pi φi EH cσ,
where φ, σ are arbitrary 1-forms on V ∗ Q. The corresponding Jacobi bracket on the

momentum phase space V ∗ Q is
{f, g}H = wH (φ, σ) + EH c(f dg − gdf ) =

{f, g}V + [H]−1 ([g]γH cdf − [f ]γH cdg),
where {f, g}V is the Poisson bracket (5.1.5) and
[f ] = pi ∂ i f − f, [g] = pi ∂ i g − g.
Remark 5.2.7. Given the Poisson bracket (5.1.5) on the momentum phase space
V ∗ Q, one can introduce the generalized Poisson bracket {., .}w (2.8.3) on the exterior
algebra O∗ (V ∗ Q), and the bracket {., .}d (2.8.5) on the quotient
O∗ (V ∗ Q)/dO∗ (V ∗ Q).
In particular, the generalized Poisson bracket (2.8.3) of Hamiltonian forms H and

H 0 reads
{H, H 0 }w = pi (∂ i H0 − ∂ i H)dt.
5.3 Canonical transformations

Canonical transformations in time-dependent mechanics are not compatible with
the fibration V ∗ Q → Q of the momentum phase space in general.
Definition 5.3.1. By a canonical automorphism is meant a vertical automorphism

ρ of the fibre bundle V ∗ Q → R, which preserves the canonical Poisson structure
(5.1.5) on the momentum phase space V ∗ Q, i.e.,
{f ◦ ρ, g ◦ ρ}V = {f, g}V ◦ ρ.
2
5.3. CANONICAL TRANSFORMATIONS 245
It is easily seen that an automorphism ρ of V ∗ Q → R is canonical if and only if

it preserves the canonical form Ω (5.1.9) on V ∗ Q, i.e.,
Ω = ρ∗ Ω.
Definition 5.3.2. The bundle coordinates of the fibre bundle V ∗ Q → R are called
canonical if they are canonical for the Poisson structure (5.1.5). 2
It is readily observed that canonical coordinate transformations satisfy the rela-

tion
∂p0 i ∂q 0 i ∂p0 i ∂q 0 i
− = 0,
∂pj ∂pk ∂pk ∂pj
∂p0 i ∂q 0 i ∂p0 i ∂q 0 i
− k j = 0,
∂q j ∂q k ∂q ∂q
0i
0
∂p i ∂q ∂p0 i ∂q 0 i
j
− j
= δjk .
∂pk ∂q ∂q ∂pk
By very definition of the canonical form Ω, the holonomic coordinates of the vertical
cotangent bundle V ∗ Q → Q are the canonical coordinates. Accordingly, holonomic
automorphisms
∂q j
(q i , pi ) 7→ (q 0i , p0i = pj ) (5.3.1)
∂q 0i
of V ∗ Q → Q, induced by the vertical automorphisms of the configuration bundle
Q → R are also canonical.
Proposition 5.3.3. Locally Hamiltonian connections are transformed into each

other by canonical automorphisms, and so are locally Hamiltonian forms. 2
Proof. If γ is a locally Hamiltonian connection for H, we have
T ρ(γ)cΩ = (ρ−1 )∗ (γcΩ) = d((ρ−1 )∗ H),
and T ρ(γ) is also a locally Hamiltonian connection. QED
Proposition 5.3.4. Let γ be a complete locally Hamiltonian connection on V ∗ Q →

R, i.e., the vector field (5.2.1) is complete. There exist canonical coordinates on V ∗ Q
such that γ = ∂t . 2
Proof. A glance at the relation (5.2.3) shows that each locally Hamiltonian con-
nection γ is the generator of a local 1-parameter group of canonical automorphisms
of the fibre bundle V ∗ Q → R. Let V0∗ Q be the fibre of V ∗ Q → R at 0 ∈ R. Then
canonical coordinates of the symplectic manifold V0∗ Q ∼ = T ∗ M dragged along inte-
gral curves of the complete vector field γ determine the desired canonical coordinates
on V ∗ Q. QED
In other words, a complete locally Hamiltonian connection γ on the momentum

phase space V ∗ Q in accordance with Proposition 4.1.2 defines a trivialization
ψ : V ∗ Q → R × V0∗ Q (5.3.2)
of the fibre bundle V ∗ Q → R such that the corresponding coordinates of V ∗ Q,

compatible with this trivialization, are canonical. However, it should be emphasized
that, although the fibre V0∗ Q is diffeomorphic to T ∗ M , the trivialization (5.3.2) fails
to be a trivialization of the type V ∗ Q ∼= R × T ∗ M since the trivialization morphism
ψ is not a bundle morphism of the fibre bundle V ∗ Q → Q.
In particular, let H be a Hamiltonian form (5.2.14) such that the correspond-
ing Hamiltonian connection γH (5.2.19) is complete. By virtue of Proposition 5.3.4,
there exists a trivialization of the phase space V ∗ Q with respect to the global canon-
ical coordinates (q A , pA ) (where q A are not coordinates on Q in general) such that
Ω = dpA ∧ dq A ∧ dt, γH = ∂t , dH = dpA ∧ dq A ,
and H reduces to the Hamiltonian form
H = pA dq A
with the Hamiltonian H = 0. Then the corresponding Hamilton equations take the
form of the equilibrium equations
qtA = 0, ptA = 0 (5.3.3)
such that q A (t, q i , pi ) and pA (t, q i , pi ) are constants of motion.

Accordingly, any Hamiltonian H can be locally brought into zero, and the cor-
responding Hamilton equations are reduced to the equilibrium ones (5.3.3) by local
canonical coordinate transformations.
5.3. CANONICAL TRANSFORMATIONS 247
Example 5.3.1. Let us consider the 1-dimensional motion with constant acceler-
ation a with respect to the coordinates (t, q). Its Hamiltonian on the momentum
phase space R3 → R reads
p2
H= − aq.
2
The associated Hamiltonian connection is
γH = ∂t + p∂q + a∂p .
This Hamiltonian connection is complete. The canonical coordinate transformations

at2
q 0 = q − pt + , p0 = p − at (5.3.4)
2
bring it into γH = ∂t . Then, the functions q 0 (t, q, p) and p0 (t, p) (5.3.4) are constants
of motion. •
Example 5.3.2. Let us consider the 1-dimensional oscillator with respect to the
same coordinates as in the previous Example. Its Hamiltonian on the momentum
phase space R3 → R reads
1
H = (p2 + q 2 ).
2
The associated Hamiltonian connection is
γH = ∂t + p∂q − q∂p ,
which is complete. The canonical coordinate transformations
q 0 = q cos t − p sin t, p0 = p cos t + q sin t (5.3.5)
bring it into γH = ∂t . Then, the functions q 0 (t, q, p) and p0 (t, q, p) (5.3.5) are con-
stants of motion. •
It should be emphasized that, in general, canonical automorphisms do not trans-

form Hamiltonian forms into Hamiltonian forms, but only locally.
Let H be a Hamiltonian form (5.2.12) on a momentum phase space V ∗ Q, equipped
with the coordinates (t, q i , pi ). Given a canonical automorphism ρ, we have
d(ρ∗ H − H) = 0,
where
ρ∗ H = ρi dρi − H ◦ ρdt. (5.3.6)
Therefore, we can write locally
H − ρ∗ H = dS, (5.3.7)
where S(t, q i , pi ) is a local function on a momentum phase space V ∗ Q.

It should be emphasized that, if a configuration space Q is a contractible mani-
fold, the equality (5.3.7) is fulfilled globally, and the function S is defined everywhere
on V ∗ Q.
Substituting (5.3.6) in (5.3.7), we obtain the relations
∂i S = pi − ρj ∂i ρj ,
∂ i S = −ρj ∂ i ρj ,
H − H0 = ρi ∂t ρi − ∂t S.
Taken on the graph
∆ρ = {(p, ρ(p)) ∈ V ∗ Q × V ∗ Q, p ∈ V ∗ Q}
of the canonical automorphism, the function S plays the role of a generating function
of canonical transformations.
Example 5.3.3. Given a canonical automorphism ρ, let
det(∂ i ρj ) 6= 0. (5.3.8)
Then, if the graph ∆ρ can be coordinated by

i
(t, q i , q 0 = q i ◦ ρ),
we obtain the familiar relations

∂S
pi = ,
∂q i
∂S
p0i = − 0i ,
∂q
i
H − H = ∂t S(t, q i , q 0 ).
0
(5.3.9)
•
5.4. THE EVOLUTION EQUATION 249
Example 5.3.4. For instance, the generating function of a holonomic automor-

phism (5.3.1), where det(∂i q 0j ) 6= 0, is
j j
S(t, q 0 , pi ) = −q i (t, q 0 )pi .
•
Let H be a Hamiltonian form whose associated Hamiltonian connection is com-
plete. By virtue of Proposition 5.3.4, there exists a canonical automorphism ρ which
bring its Hamiltonian into zero. If the condition (5.3.8) holds, then (5.3.9) is the
Hamilton–Jacobi equation.
5.4 The evolution equation

Given a Hamiltonian form H (5.2.14) and the corresponding Hamiltonian connection
γH (5.2.19) on the momentum phase space V ∗ Q, let us consider the Lie derivative
of a function f ∈ O0 (V ∗ Q) along the horizontal vector field γH , which reads
LγH f = γH cdf = (∂t + ∂ i H∂i − ∂i H∂ i )f. (5.4.1)
This equality is the evolution equation in time-dependent mechanics. Substituting a
solution r of the Hamilton equations (5.2.25) in (5.4.1), we obtain the time evolution
of f along this solution:
d
LγH f ◦ r = (f ◦ r).
dt
It should be emphasized that, in comparison with the evolution equation (3.2.5)
in symplectic mechanics, the right-hand side of the evolution equation (5.4.1) is not
reduced to the Poisson bracket since a Hamiltonian H on a momentum phase space
of time-dependent mechanics is not a function, i.e., H 6∈ O0 (V ∗ Q). Of course, one
can define locally the Poisson bracket {H, f }V of a Hamiltonian H with a function
f on V ∗ Q. However, being equal to zero with respect to some coordinates, this
Poisson bracket does not necessarily vanish with respect to other coordinates.
Given the splitting (5.2.14) of a Hamiltonian form H, the evolution equation
(5.4.1) is written as
LγH f = ∂t f + (Γi ∂i − ∂i Γj pj ∂ i )f + {H
f , f} .
Γ V (5.4.2)
In particular, the following two consequences of this form of the evolution equation
should be pointed out.
• Since the evolution equation (5.4.2) is not reduced to the Poisson bracket,
the integrals of motion in time-dependent mechanics cannot be defined as
functions on a momentum phase space, which are in involution with a Hamil-
tonian. Therefore, to obtain conservation laws in Hamiltonian time-dependent
mechanics, we follow the methods of field theory (see Section 5.7).
• The second (kinematic) term in the right-hand side of this equality plays the
essential role under quantization. It makes the quantization procedure depen-
dent on a reference frame. In Appendix B, we will show that quantizations
under different reference frames fail to be equivalent in general.
Note that the kinematic term in the evolution equation (5.4.2) can be eliminated
at least locally by means of canonical transformations. Let a connection Γ in the
splitting of a Hamiltonian form (5.2.14) be complete. With respect to the coordinate
system (t, q i ) adapted to the reference frame Γ, the configuration bundle Q is triv-
ialized, and the corresponding holonomic coordinates (t, q i , pi ) on the momentum
phase space V ∗ Q are canonical. With respect to these coordinates, the evolution
equation (5.4.2) takes the familiar form
LγH f = ∂t f + {H,
f f} .
V
5.5 Degenerate systems

This Section is devoted to the relationship between Lagrangian and Hamiltonian
formalisms of time-dependent mechanics. This relationship is characterized by the
diagram
H
V ∗ Q −→ J 1 Q
b
L
b 6 L
b
?
H
J 1 Q ←− V ∗ Q
b
which fails to commute in general. Its jet extension is

J 1H
J 1 V ∗ Q −→ J 1 J 1 Q
b
J 1L J 1L
b 6 b
?
J 1H
J 1 J 1 Q ←− J 1 V ∗ Q
b
5.5. DEGENERATE SYSTEMS 251
where the jet prolongations of Hamiltonian and Legendre maps read

c : J 1 V ∗ Q → J 1 J 1 Q,
J 1H
(q λ , qti , q(t)
i c = (q λ , ∂ i H, q i , d ∂ i H),
, qtti ) ◦ J 1 H t t
b : J 1 J 1 Q → J 1 V ∗ Q,
J 1L
(q λ , pi , qti , pti ) ◦ J 1 L
b = (q λ , π , q i , d π ).
i (t) t i
As we will show below, in the case of a hyperregular Lagrangian when the Leg-
endre map L b (4.8.9) is a diffeomorphism, Lagrangian and Hamiltonian formalisms
are equivalent. Conversely, let H be a Hamiltonian form and γH the corresponding
Hamiltonian connection (5.2.19) on the momentum phase space V ∗ Q → R. Let us
consider the composition of morphisms
c ◦ γ : V ∗ Q → J 2 Q ⊂ J 1 J 1 Q,
J 1H (5.5.1)
H
(qti , q(t)
i
, qtti ) ◦ J 1 H
c ◦ γ = (∂ i H, ∂ i H, γ cd (∂ i H)).
H H t
If the Hamiltonian map H c is a diffeomorphism, then J 1 H

c◦γ ◦H c−1 is a holonomic
H
connection on the jet bundle J 1 Q → R, which defines a dynamic equation on the
velocity configuration space J 1 Q. This dynamic equation is equivalent to the system
of Lagrange equations for the Lagrangian function
c−1∗ [H]
L=H
c−1 .
which is the pull-back on J 1 Q of the function [H] (5.2.29) by the morphism H
Let us consider more general conditions for solutions of Hamilton equations on
a momentum phase space to be solutions of Lagrange equations and second order
dynamic equations on a velocity phase space, and vice versa.
Following the general scheme of polysymplectic Hamiltonian formalism [57, 158,
159], we say that a Hamiltonian form H on the momentum phase space V ∗ Q is
associated with a Lagrangian L on the velocity phase space J 1 Q if H obeys the
conditions
b ◦H
L c◦L
b = L,
b (5.5.2)
H = HHb + L ◦ H
c (5.5.3)
b ◦H
[161]. A glance at the condition (5.5.2) shows that the composition L c is the
projection operator
pi (p) = ∂i L(t, q j , ∂ j H(p)), p ∈ Q, (5.5.4)
¿from V ∗ Q onto a submanifold

b 1 Q) ⊂ V ∗ Q,
N = L(J (5.5.5)
called the the Lagrangian constraint space. Given a Hamiltonian form H associ-
ated with the Lagrangian L, this constraint space is defined by the relation (5.5.4).
Accordingly, the composition Hc◦L b is the projection operator from J 1 Q onto a
c ) ⊂ J 1 Q. The relation (5.5.3) takes the form
submanifold H(N
L◦H
c = [H] (5.5.6)
everywhere on V ∗ Q.
Remark 5.5.1. Unless otherwise stated, we regard N as a subset of the momentum
phase space V ∗ Q, without a manifold structure. All objects are defined on the whole
phase space V ∗ Q, and their restriction to N means only that their values at the
points of N ⊂ V ∗ Q are considered. •
Proposition 5.5.1. A Hamiltonian form H associated with a Lagrangian L obeys

the relation
c∗ H | ,
H |N = H (5.5.7)
L N
where HL is the Poincaré–Cartan form (4.8.5). 2
Acting on both sides of the equality (5.5.6) by the exterior differential, we obtain
the relations
∂t H(p) = −(∂t L)(t, q j , ∂ j H(p)), p ∈ N, (5.5.8)
∂i H(p) = −(∂i L)(t, q j , ∂ j H(p)), p ∈ N, (5.5.9)
(pi − (∂it L)(t, q j , ∂ j H))∂ i ∂ a H = 0. (5.5.10)
A glance at the relation (5.5.10) shows that:
• the condition (5.5.2) is a corollary of the condition (5.5.3) if the Hamiltonian
map Hc is regular, i.e.,
det(∂ i ∂ j H) 6= 0
at all points of the Lagrangian constraint space N ;

• the Hamiltonian map H

c is always degenerate outside the Lagrangian con-
straint space N .
Example 5.5.2. Let L = 0 be the zero Lagrangian. In this case, the Lagrangian
constraint space is N = 0(Q),
b where 0b is the canonical zero section of the Legendre
∗
bundle V Q → Q. The condition (5.5.2) is fulfilled trivially for every Hamiltonian
map, while the condition (5.5.3) takes the coordinate form
H = pi ∂ i H.
Any Hamiltonian form HΓ (5.2.9) obeys this condition, and is associated with the
Lagrangian L = 0. •
If a Lagrangian L is hyperregular, there exists a unique associated Hamiltonian

form
b −1 )∗ L
H = HLb−1 + (L (5.5.11)
such that
H b −1 ,
c=L pi ≡ πi (q λ , ∂ j H(q λ , pk )), qti ≡ ∂ i H(q λ , πj (q λ , qtk )). (5.5.12)
As an immediate consequence of (5.5.12), we have
J 1H b −1 .
c = (J 1 L)
Proposition 5.5.2. Let L be a hyperregular Lagrangian, and H the associated

Hamiltonian form. The following relations hold:
b ∗ H,
HL = L (5.5.13)
b ∗E ,
EL = (J 1 L) (5.5.14)
H
c ∗E ,
EH = (J 1 H) (5.5.15)
L
where EH is the Hamilton operator (5.2.18) for H, and EL is the Euler–Lagrange–

Cartan operator (4.8.17) for L. 2
A glance at the relations (5.5.7) and (5.5.13) shows that the Poincaré–Cartan
form is the Lagrangian counterpart of a Hamiltonian form, whereas it follows from
(5.5.14) and (5.5.15) that the Lagrangian counterpart of the Hamilton operator is
the Euler–Lagrange–Cartan operator EL (4.8.17).
In particular, if γH is a Hamiltonian connection for the associated Hamiltonian
form H (5.5.11), then, by virtue of the equality (5.5.15), the composition J 1 H
c◦γ
H
(5.5.1) takes its values into the kernel of the Euler–Lagrange–Cartan operator EL or,
more exactly, in the kernel of the Euler–Lagrange operator EL . Then the composition
J 1H b : J 1 Q 3 (t, q i , q i ) 7→
c◦γ ◦L
H t
(t, q i , qti = ∂ i H ◦ L,
b q i = ∂ i H ◦ L,
(t)
b q i = γ cd(∂ i H) ◦ L)
tt H
b =
(t, q i , qti , q(t)

i
= qti , qtti = γH cd(∂ i H) ◦ L)
b ∈ J 2Q
is a holonomic Lagrangian connection for L. Conversely, if ξL is a Lagrangian

connection for L, then
γH = J 1 L
b ◦ξ ◦H
L
c
is a Hamiltonian connection for H. This proves the following assertion.
Proposition 5.5.3. Let L be a hyperregular Lagrangian and H the associated

Hamiltonian form (5.5.11).
(i) Let a section r of the fibre bundle V ∗ Q → R be a solution of the Hamilton
equations (5.2.20) – (5.2.21) for the Hamiltonian form H. Then the section
c = πΠ ◦ r
of the fibre bundle Q → R is a solution of the Lagrange equations (4.8.7) for the
Lagrangian L, while its first order jet prolongation ċ satisfies the Cartan equations
(4.8.18) – (4.8.19).
(ii) Conversely, if a section c of the jet bundle J 1 Q → R is a solution of the
Cartan equations for the Lagrangian L, the section
b ◦c
r=L
of the fibre bundle V ∗ Q → R satisfies the Hamilton equations (5.2.20) – (5.2.21) for
the Hamiltonian form H. 2
It follows that, given a hyperregular Lagrangian, there is one-to-one correspon-

dence between the solutions of the Lagrange equations (and, consequently, of the
Cartan equations) and the solutions of the Hamilton equations for the associated
Hamiltonian form.
In the case of regular Lagrangian L, the Lagrangian constraint space N is an
open subbundle of the Legendre bundle V ∗ Q → Q. If N 6= V ∗ Q, an associated
Hamiltonian form fails to be defined everywhere on V ∗ Q in general. At the same
time, an open constraint subbundle N can be provided with the appropriate pull-
back structure with respect to the imbedding N ,→ V ∗ Q so that we may restrict our
consideration to Hamiltonian forms on N . If a regular Lagrangian is additionally
semiregular (see Definition 5.5.4 below), the associated Legendre morphism is a
diffeomorphism of J 1 Q onto the Lagrangian constraint space N and, on N , we can
restate all results true for hyperregular Lagrangians.
Example 5.5.3. Let Q be the fibre bundle R2 → R with coordinates (t, q). Its jet
manifold J 1 Q ∼ = R3 and its Legendre bundle V ∗ Q ∼
= R3 are coordinated by (t, q, qt )
and (t, q, p), respectively. Put
L = exp(qt )dt. (5.5.16)
This Lagrangian is regular, but not hyperregular. The corresponding Legendre map
reads
p◦L
b = exp q .
t
It follows that the Lagrangian constraint space N is given by the coordinate relation
p > 0. This is an open subbundle of the Legendre bundle, and L b is a diffeomorphism
of J 1 Q onto N . Hence, there is a unique Hamiltonian form H on N which is
associated with the Lagrangian (5.5.16). Its Hamiltonian function reads
H = p(ln p − 1).
This Hamiltonian form, however, fails to be smoothly extended to V ∗ Q. •
Hereafter, we will restrict our consideration of degenerate systems described by

semiregular Lagrangians. Most physically interesting Lagrangians, including the
quadratic ones, are of this type. On the other hand, there is the comprehensive
relationship between Lagrangian and Hamiltonian formalisms in this case [57, 158,
159, 190].
Definition 5.5.4. A Lagrangian L is said to be semiregular, if the pre-image

b −1 (p) of any point p ∈ N is a connected submanifold of the velocity phase space
L
J 1 Q. 2
Proposition 5.5.5. All Hamiltonian forms H associated with a semiregular La-

grangian L coincide with each other at the points of the Lagrangian constraint space
N , i.e.,
H |N = H 0 |N .
Moreover, the Poincaré–Cartan form HL for the Lagrangian L is the pull-back

b ∗ H,
HL = L
(πi qti − L)dt = H(t, q i , πi )dt, (5.5.17)
of any associated Hamiltonian form H by the Legendre map L.

b 2
Proof. Let u be a vertical vector field on the jet bundle J 1 Q → Q. If u takes its
values into the kernel Ker T L
b of the tangent morphism to L,
b it is easy to see that
Lu HL = 0.
Hence, the Poincaré–Cartan form HL for a semiregular Lagrangian L is constant

b −1 (p) of each point p ∈ N . Then results follow from
on the connected pre-image L
(5.5.7). QED
Note that the Hamilton operators for Hamiltonian forms in Proposition 5.5.5
do not necessarily coincide at points of the Lagrangian constraint N because of the
derivatives of these forms.
Let H be a Hamiltonian form associated with a semiregular Lagrangian L. Act-
ing by the exterior differential on the relation (5.5.17), we obtain the equality
(qti − ∂ i H ◦ L)dπ
b b i
i ∧ dt − (∂i L + ∂i (H ◦ L))dq ∧ dt = 0 (5.5.18)
or, equivalently, the equalities
πij (qtj − ∂ j H ◦ L)
b = 0,
∂i πj (qtj − ∂ j H ◦ L)
b − (∂ L + (∂ H) ◦ L)
i i
b = 0.
Using the equality (5.5.18), one can extend the relation

b ∗E
EL = (J 1 L) (5.5.19)
H
(5.5.14), but not necessarily the relation (5.5.15) to semiregular Lagrangians. The
relation (5.5.19) enables us to generalize item (i) of Proposition 5.5.3 for Hamiltonian
forms associated with semiregular Lagrangians.
Proposition 5.5.6. Let a section r of the fibre bundle V ∗ Q → R be a solution

of the Hamilton equations for a Hamiltonian form H associated with a semiregular
Lagrangian L. If r lives in the Lagrangian constraint space N , the section
c = πΠ ◦ r
of the fibre bundle Q → R satisfies the Lagrange equations for L, while its first
order jet prolongation
c ◦ r = ċ
c=H
obeys the corresponding Cartan equations (4.8.18) – (4.8.19). 2
c ◦ r. Since r(R) ⊂ N , then

Proof. Put c = H
b ◦ c,
r=L ṙ = J 1 L
b ◦ J 1 c.
If r is a solution of the Hamilton equations, the exterior form EH vanishes at points

of ṙ(R). Hence, the pull-back form
b ∗E
EL = (J 1 L) H
vanishes at points of J 1 c(R). It follows that the section c of the fibre bundle J 1 Q →
R obeys the Cartan equations. By virtue of the equation (5.2.20), we have c = ċ, and
the section c is a solution of the Lagrange equations, which lives in the submanifold
H(N
c ). QED
In the case of semiregular Lagrangians, item (ii) of Proposition 5.5.3 can be

modified as follows.
Proposition 5.5.7. Given a semiregular Lagrangian L, let a section c of the jet

bundle J 1 Q → R be a solution of the Cartan equations (4.8.18) – (4.8.19). Let H
be a Hamiltonian form associated with L so that the corresponding Hamiltonian

map satisfies the condition
H b ◦ c = J 1 (π 1 ◦ c).
c◦L (5.5.20)
0
Then the section

b ◦ c,
r=L
ri = πi (t, cj , cjt ), r i = ci ,
of the fibre bundle V ∗ Q → R is a solution of the Hamilton equations (5.2.20) –

(5.2.21) for H, which lives in the Lagrangian constraint space N . 2
Proof. The Hamilton equations (5.2.20) hold by virtue of the condition (5.5.20).
Using the relations (5.5.18) and (5.5.20), the Hamilton equation (5.2.21) is brought
into the Cartan equation (4.8.19):
b ◦ c = −(cj − ∂ j H ◦ L
dt πi ◦ c = −(∂i H) ◦ L b ◦ c)∂ π ◦ c + ∂ L ◦ c =
t i j i
(∂t cj − cjt )∂i πj ◦ c + ∂i L ◦ c.
QED
Proposition 5.5.6 shows that, if H is a Hamiltonian form associated with a

semiregular Lagrangian L, every solution of the corresponding Hamilton equations,
which lives in the Lagrangian constraint space N , yields a solution of the Cartan
equations and that of the Lagrange equations for L. Thus, the counterpart of de-
generate Lagrangian systems are the constraint Hamiltonian systems.
We will restrict our consideration to the case of a Lagrangian constraint space
N . This plays the role of a primary constraint space. In order that local solutions
of the Hamilton equations for a Hamiltonian form H exists on N , the Hamiltonian
connection γH must be tangent to N . This condition is given by the equations
pi = ∂it L(t, q j , ∂ j H), (5.5.21)

(∂t + ∂ j H∂j − ∂j H∂ j )cd(pi − ∂it L(t, q j , ∂ j H)) = 0, (5.5.22)
where (5.5.21) is the equation (5.5.4) of a Lagrangian constraint space, while (5.5.22)
requires that the horizontal vector field γH is tangent to N at the point (t, q i , pi ). In
contrast with the case of autonomous constraint systems, the left-hand side of the
equation (5.5.22) is not reduced to the Poisson bracket. Therefore, we cannot follow
the familiar procedure of constructing the final constraint space in Section 3.5.
There is another possibility. Given a solution of the Lagrange equations, one can
try to find an associated Hamiltonian form H such that this solution is a solution
of the Hamilton equations for H. It may happen that different solutions of the
Lagrange equations require different Hamiltonian forms in general. Thus, degenerate
Lagrangian systems are described as multi-Hamiltonian systems.
Note that, in the case of semiregular Lagrangians, the condition (5.5.20) is the
obstruction for a solution c of the Cartan equations to be a solution of the Hamilton
equations and the Lagrange equations. At the same time, one can try to find a
family of associated Hamiltonian forms such that each solution of the Lagrange
equations is a solution of the Hamilton equations for some Hamiltonian form from
this family.
Definition 5.5.8. We will say that a family of Hamiltonian forms H, associated

with a Lagrangian L, is complete if, for each solution c of the Lagrange equations,
there exists a solution r of the Hamilton equations for a Hamiltonian form H from
this family so that
c = πΠ ◦ r, b ◦ ċ
r=L (5.5.23)
(see the relation (5.2.22)). 2
Let L be a semiregular Lagrangian. Then, by virtue of Proposition 5.5.7, such

a complete family of associated Hamiltonian forms exists if and only if, for every
solution c of the Lagrange equations for L, there is a Hamiltonian form H from this
family such that
c◦L
H b ◦ ċ = ċ. (5.5.24)
Example 5.5.4. Let L = 0. This Lagrangian is semiregular. Its Lagrange equations

reduce to the identity 0 ≡ 0. Every section c of the configuration bundle Q → R
is a solution of this equation. Given a section c, let Γ be a connection on the fibre
bundle Q → R such that c is its integral section. The Hamiltonian form HΓ (5.2.9)
is associated with L = 0, and the Hamiltonian map H c obeys the relation (5.5.24).
Γ
The corresponding Hamilton equations read
qti = Γi , pti = −pj ∂i Γj .

They have a solution

b ◦ ċ,
r=L
r i = ci , ri = 0,
which lives in the Lagrangian constraint space pi = 0. •
The example below shows that a complete family of associated Hamiltonian

forms may exist when a Lagrangian is not necessarily semiregular.
Example 5.5.5. Let Q be the fibre bundle R2 → R1 in Example 5.5.3 with coor-
dinates (t, q). Put
1
L = (qt )3 .
3
The associated Legendre map reads
b = q2.
p◦L (5.5.25)
t
The corresponding Lagrangian constraint space N is given by the coordinate relation

p ≥ 0. It fails to be a submanifold of the momentum phase space V ∗ Q. There exist
two associated Hamiltonian forms
2
H+ = pdq − p3/2 dt,
3
2 3/2
H− = pdq + p dt,
3
on N , which correspond to the two different solutions
√ √
qt = p, qt = − p
of the equation (5.5.25). The Hamiltonian forms H+ and H− make up a complete

family. •
Remark 5.5.6. Let γH be a Hamiltonian connection for a Hamiltonian form H

associated with a semiregular Lagrangian L. By virtue of the relation (5.5.19), the
composition J 1 H
c ◦ γ (5.5.1) takes its values into the kernel of the Euler–Lagrange
H
operator EL . Since
c ◦ L|
H b = Id H(Q),
c
H(Q)
b
the morphism
J 1H b : J 1 Q 3 (t, q i , q i ) 7→
c◦γ ◦L
t
(t, q i , qti = ∂ i H ◦ L,
b q i = ∂ i H ◦ L,
(t)
b q i = γ cd(∂ i H) ◦ L)
tt H
b
restricted to H(Q)
c is the holonomic section over H(Q)
c ⊂ J 1 Q of the affine jet
bundle J 2 Q → J 1 Q. Let H(Q)
c be a closed submanifold, e.g., when a Lagrangian L
is almost regular (see Definition 5.5.9 below). Then the section J 1 Hc◦γ ◦L
H
b can
be extended to a holonomic connection on the jet bundle J 1 Q → R, which defines
a dynamic equation on the configuration space Q. In this case, given a solution r
of the Hamilton equations for H which lives in the Lagrangian constraint space N ,
its projection πΠ ◦ r is a solution of both the Lagrange equations and the dynamic
one. •
Definition 5.5.9. A semiregular Lagrangian L is called almost regular if: (i) the
Lagrangian constraint space N is a closed imbedded subbundle iN : N ,→ V ∗ Q of
the Legendre bundle V ∗ Q → Q and (ii) the Legendre map Lb : J 1 Q → N is a fibre
bundle. 2
Proposition 5.5.10. [57, 190]. Let L be an almost regular Lagrangian. On an

open neighbourhood in V ∗ Q of each point q ∈ N , there exists a complete family of
local Hamiltonian forms associated with L. 2
Let L be an almost regular Lagrangian L. Given an associated Hamiltonian form

H, the Hamiltonian map
c | : N → J 1Q
H N
b : J 1 Q → N , and its image H(N

restricted to N is a section of the fibre bundle L c ) is
a closed imbedded submanifold of the configuration space J 1 Q. Therefore, it follows
¿from Remark 5.5.6 that each solution of the Lagrange equations for L, which is a
solution of Hamilton equations, is a solution of a dynamic equation.
Since solutions of the Lagrange equations for a degenerate Lagrangian may cor-
respond to solutions of different Hamilton equations, we can conclude that, roughly
speaking, the Hamilton equations involve some additional conditions in comparison
with the Lagrange equations. Therefore, let us separate a part of the Hamilton
equations which are defined on the Lagrangian constraint space Q in the case of an
almost regular Lagrangian.
Let
def
HN = i∗N H
be the restriction of a Hamiltonian form H associated with an almost regular La-
grangian L to the Lagrangian constraint space N . By virtue of Proposition 5.5.5,
this restriction, called the constrained Hamiltonian form, is uniquely defined. More-
over, the Poincaré–Cartan form is the pull-back HL = L b ∗ H of the constrained
N
Hamiltonian form HN . On sections r of the fibre bundle Q → R, we can write the
constrained Hamilton equations
r∗ (uN cdHN ) = 0, (5.5.26)
where uN is an arbitrary vertical vector field on the fibre bundle N → R [56, 57, 161].
For the sake of simplicity, we can identify a vertical vector field uN on N → R with
its image T iN (uN ) and can bring the constrained Hamilton equations (5.5.26) into
the form
r∗ (uN cdH) = 0, (5.5.27)
where r is a section of the fibre bundle N → R, and uN is an arbitrary vertical vector
field on N → R. It should be emphasized that these equations fail to be equivalent
to the Hamilton equations (5.2.26) restricted to the Lagrangian constraint space N .
If an almost regular Lagrangian admits associated Hamiltonian forms (see Propo-
sition 5.5.10), the following three assertions together with Proposition 5.5.7 give
the relationship between Cartan, Hamilton–De Donder, Hamilton, and constrained
Hamilton equations [57, 161].
Proposition 5.5.11. For any Hamiltonian form H associated with an almost

regular Lagrangian L, every solution r of the Hamilton equations which lives in the
Lagrangian constraint space N is a solution of the constrained Hamilton equations
(5.5.27). 2
Proof. For any vertical vector field uN on the fibre bundle N → R, the vector field
T iN (uN ) is obviously a local vertical vector field on the fibre bundle V ∗ Q → R.
Then we have
r∗ (uN cdHN ) = r∗ (uN ci∗N dH) = r∗ (T iN (uN )cdH) = 0
if r is a solution of the Hamilton equations (5.2.26) for the Hamiltonian form H.

QED
Proposition 5.5.12. A section c of the jet bundle J 1 Q → R is a solution of the

Cartan equations (4.8.20) if and only if L◦c
b is a solution of the constrained Hamilton
equations (5.5.27). 2
Proof. Let uN be a vertical vector field on the fibre bundle N → R. Since L b is a

1
submersion, there exists a vertical vector field v on the jet bundle J Q → R such
that
b ◦v =u .
TL N
For instance, v is the horizontal lift of u by means of a connection on the affine jet
bundle J 1 Q → Q. Let a section c : R ,→ J 1 Q be a solution of the Cartan equations
(4.8.20). Then we have
b ◦ c)∗ (u cdH ) = c∗ (vcdH ) = 0.
(L (5.5.28)
N N L
It follows that the section L b ◦ c : R ,→ N is a solution of the equations (5.5.27).

The converse is obtained by running (5.5.28) in reverse, bearing in mind that the
restriction of any vector field v on J 1 Q to c(R) is projectable by L.
b QED
Proposition 5.5.13. The constrained Hamilton equations (5.5.26) are equivalent

to the Hamilton-De Donder equations (4.8.27). 2
Proof. Let H be a Hamiltonian form associated with an almost regular Lagrangian

L, and let h be the corresponding section of the fibre bundle
ζ : T ∗Q → V ∗Q
(see Definition 5.2.4). This section yields the morphism
hN = h ◦ iN : N → T ∗ Q
which does not depend on the choice of H. This is a local section of the fibre bundle
T ∗ Q → V ∗ Q over N ⊂ V ∗ Q, i.e,
ζ ◦ hN = Id N. (5.5.29)
Moreover, we have
HN = i∗N H = i∗N (h∗ Ξ) = h∗N Ξ,
whenever H is a Hamiltonian form associated with the Lagrangian L. In accordance

with the relation (5.5.17),
c = h ◦ L,
H b (5.5.30)
L N
where H
c is the Legendre morphism (4.8.23) associated with the Poincaré–Cartan
L
form HL . Substituting (4.8.24) in (5.5.30), we obtain
c =h ◦ζ ◦H
H c .
L N L
It follows that
hN ◦ ζ |ZL = Id iL (ZL ), (5.5.31)
where iL (ZL ) = Hc (J 1 Q) is the image of the Legendre morphism H c . A glance at

L L
the relations (5.5.29) and (5.5.31) shows that there is the bundle isomorphism
ζ◦iL
ZL - N
i−1
L ◦hN
over Q. Since HN = h∗N Ξ and ΞL = i∗L Ξ, we have
HN = (i−1 ∗
L ◦ hN ) ΞL , ΞL = (ζ ◦ iL )∗ HN .
Hence, the Hamilton–De Donder equations (4.8.27) are equivalent to the constrained
Hamilton equations (5.5.26). QED
5.6 Quadratic degenerate systems

As an important illustration of Propositions 5.5.6 and 5.5.7, let us describe the
complete families of Hamiltonian forms associated with almost regular quadratic
Lagrangians. The Hamiltonians of these Hamiltonian forms are quadratic in mo-
menta.
5.6. QUADRATIC DEGENERATE SYSTEMS 265
Remark 5.6.1. Since Hamiltonians in time-dependent mechanics are not scalar

functions on a momentum phase space, one cannot apply to them the well-known
analysis of the normal forms [18] (e.g., quadratic Hamiltonians [6] in symplectic
mechanics). •
Given a configuration bundle Q → R, let us consider a quadratic Lagrangian

1
L = aij qti qtj + bi qti + c, (5.6.1)
2
where a, b and c are local functions on Q. This property is global due to the
transformation law of the velocity coordinates qti . The associated Legendre map
reads
b = a qj + b .
pi ◦ L (5.6.2)
ij t i
Lemma 5.6.1. The Lagrangian (5.6.1) is semiregular. 2
Proof. For any point p of the Lagrangian constraint space N (5.5.5), the system of
linear algebraic equations (5.6.2) for qti has solutions which make up an affine space
modelled over the linear space of solutions of the homogeneous linear algebraic
equations
0 = aij q̇ j ,
where q̇ j are the holonomic coordinates of the vertical tangent bundle V Q. This
affine space is obviously connected. QED
Let us assume that the Lagrangian L (5.6.1) is almost regular, i.e., the matrix
aij is of constant rank.
The Legendre map (5.6.2) is an affine morphism over Q. It defines the corre-
sponding linear morphism
L : V Q → V ∗ Q,
Q
pi ◦ L = aij q̇ j ,
whose image N is a linear subbundle of the Legendre bundle V ∗ Q → Q. Accordingly,

the Lagrangian constraint space N , given by the equations (5.6.2), is an affine
subbundle, modelled over N , of the momentum phase space V ∗ Q → Q. Hence, the
constraint fibre bundle N → Q has a global section. For the sake of simplicity, let us
assume that this is the canonical zero section 0(Q)
b of the vertical cotangent bundle
∗
V Q → Q. Then N = N .
The kernel
Ker L b −1 (0(Q))
b =L b
of the Legendre map is an affine subbundle of the affine jet bundle J 1 Q → Q, which
is modelled over the vector bundle
−1
Ker L = L (0(Q))
b ⊂ V Q.
Then there exists a connection
Γ : Q → Ker L,
b (5.6.3)
aij Γj + bi = 0, (5.6.4)
on the configuration bundle Q → R, which takes its values into Ker L.

b It is called
a Lagrangian frame connection. With this connection, the quadratic Lagrangian
(5.6.1) can be brought into the form
1
L = aij q̇Γi q̇Γj + c0 ,
2
q̇Γi = qti − Γi .
For instance, if the quadratic Lagrangian (5.6.1) is regular, there is a unique solution
(5.6.3) of the algebraic equations (5.6.4).
Proposition 5.6.2. There exists a linear map
σ : V ∗ Q → V Q, (5.6.5)
i ij
q̇ ◦ σ = σ pj ,
over Q such that
L ◦ σ ◦ iN = iN .
Proof. The map (5.6.5) is a solution of the algebraic equations
aij σ jk akb = aib . (5.6.6)

The matrix aij is symmetric. After diagonalization, let this matrix have non-
vanishing components aAA , A ∈ I. Then a solution of the equations (5.6.6) takes
the form
1
σAA = AA , σAA0 = 0, A, A0 ∈ I, A 6= A0 ,
a
while the remaining components are arbitrary. In particular, there is a solution
1
σAA = , σAA0 = 0, σCB = 0, B 6∈ I, (5.6.7)
aAA
which satisfies the relation
σ = σ ◦ L ◦ σ, (5.6.8)
ij ik jb
σ = σ akb σ .
QED
¿From now on, by σ is meant the solution (5.6.7). If the Lagrangian (5.6.1) is
regular, the map (5.6.5) is uniquely determined by the equation (5.6.6).
The Lagrangian frame connection (5.6.3) and the map (5.6.5) play a prominent
role in the construction below.
Proposition 5.6.3. The matrix σ defines the splitting
J 1 Q = Ker L
b ⊕ Im(σ ◦ L),
b (5.6.9)
Q
qti = [qti − σ ik (akj qtj + bk )] + [σ ik (akj qtj + bk )].
of the velocity phase space J 1 Q. 2
In particular, it follows that, since L and σ are linear morphisms, their compo-
sition L ◦ σ is a fibration of the momentum phase space V ∗ Q over the Lagrangian
constraint space N .
Proposition 5.6.4. There is also the following splitting of the momentum phase
space
V ∗ Q = Ker σ ⊕ N,
Q
pi = [pi − aij σ jk pk ] + [aij σ jk pk ].

Corollary 5.6.5. Every vertical vector field
u = ui ∂i + ui ∂ i
on the momentum phase space V ∗ Q → R admits the decomposition
u = (u − uN ) + uN , (5.6.10)
ui = [ui − aij σ jk uk ] + [aij σ jk uk ],
where
uN = ui ∂i + aij σ jk uk ∂ i
is a vertical vector field on the Lagrangian constraint space N → R. 2
Given the linear map σ (5.6.5) and the Lagrangian frame connection Γ (5.6.3),
let us consider the affine Hamiltonian map
Φ = Γ + σ : V ∗ Q → J 1 Q, (5.6.11)
i i ij
Φ = Γ (q) + σ pj ,
and the Hamiltonian form
H = HΦ + Ldt,
1 1
H = pi dq i − [Γi (pi − bi ) + σ ij pi pj − c]dt. (5.6.12)
2 2
It is readily observed that Hc = Φ and, by virtue of (5.6.8), the Hamiltonian form
H is associated with the quadratic Lagrangian (5.6.1).
We aim to show that the Hamiltonian forms (5.6.12), parameterized by the
Lagrangian frame connections Γ (5.6.3), constitute a complete family.
Given the Hamiltonian form (5.6.12), let us consider the Hamilton equations
(5.2.20) for sections r of the fibre bundle V ∗ Q → R. They read
b + σ) ◦ r,
ċ = (Γ c = πΠ ◦ r, (5.6.13)
or
∇t ri = σ ij rj ,
where
∇t ri = ∂t ri − (Γ ◦ c)i
is the covariant derivative relative to the connection Γ. With the splitting (5.6.9),
we have the surjections
S = pr1 : J 1 Q → Ker L,
b
S : qti → qti − σ ik (akj qtj + bk ),
and
F = pr2 : J 1 Q → Im(σ ◦ L),

b
F : qti → σ ik (akj qtj + bk ).
With respect to these surjections, the Hamilton equations (5.6.13) are split into the
following two parts:
S ◦ ċ = Γ ◦ c, (5.6.14)
i ik j
∇t r = σ (akj ∂t r + bk ),
and
F ◦ ċ = σ ◦ r, (5.6.15)
ik j ik
σ (akj ∂t r + bk ) = σ rk .
The Hamilton equations (5.6.14) are independent of the canonical momenta rk

and play the role of gauge conditions. Moreover, for every section c of the fibre
bundle Q → R (in particular, for every solution of the Lagrange equations for the
Lagrangian L), there exists a Lagrangian frame connection Γ (5.6.3) such that the
equation (5.6.14) holds. Indeed, let Γ0 be a connection on the fibre bundle Q → R
whose integral section is c. Put
def
Γ = S ◦ Γ0 ,
i j
Γ = Γ0 − σ ik (akj Γ0 + bk ).
In this case, the Hamiltonian map (5.6.11) satisfies the relation (5.5.24) for c, i.e.,
Φ◦L
b ◦ ċ = ċ.
Thence, the Hamiltonian forms (5.6.12) constitute a complete family. The Hamil-
tonian forms from this family differ from each other only in the Lagrangian frame
connections Γ (5.6.3) which lead to the different gauge conditions (5.6.14). It is
readily observed that a Lagrangian frame connection Γ is also a Hamiltonian frame
connection for the corresponding Hamiltonian form (5.6.12).
Proposition 5.6.6. For every Hamiltonian form H (5.6.12), the Hamilton equa-
tions (5.2.21) and (5.6.15) on sections of the constraint fibre bundle N → R are
equivalent to the constrained Hamilton equations (5.5.27). 2
Proof. In accordance with the decomposition (5.6.10) of a vertical vector field u on

the momentum phase space V ∗ Q → R, the constrained Hamilton equations (5.5.27)
take the form
r∗ (aij σ jk ∂ i cdH) = 0, (5.6.16)
r∗ (∂i cdH) = 0. (5.6.17)
The equations (5.6.17) are obviously the Hamilton equations (5.2.21) for H. Bearing
in mind the relations (5.6.4) and (5.6.8), one can easily bring the equations (5.6.16)
into the form (5.6.15). QED
It follows that the equations (5.6.14) are the additional conditions which make
the Hamilton equations differ from the constrained Hamilton equations (5.5.27) and
the Lagrange equations in the case of a quadratic Lagrangian.
Corollary 5.6.7. By virtue of Proposition 5.5.12, a section c of the jet bundle

J 1 Q → R is a solution of the Cartan equations (4.8.18) – (4.8.19) for the almost
regular Lagrangian (5.6.1) if and only if L◦c
b is a solution of the constrained Hamilton
equations (5.2.21) and (5.6.15). 2
It follows that the equations (5.6.14) are responsible for the obstruction condition
(5.5.20) for solutions c of the Cartan equations to be solutions of the Hamilton
equations and of the Lagrange equations for the Lagrangian (5.6.1). It is readily
seen that the equations (5.6.15) do not contribute to the obstruction condition
(5.5.20). If r = Lb ◦ c, these equations hold for solutions c of the Cartan equations
(4.8.18).
Proposition 5.6.8. Let c be a solution of the Cartan equations for the almost
regular quadratic Lagrangian L (5.6.1). Let c0 be a section of the fibre bundle
5.7. HAMILTONIAN CONSERVATION LAWS 271
V Q → R, which takes its values into Ker L and projects onto c = π01 ◦ c. Then the
sum c + c0 over Q is also a solution of the Cartan equations. 2
Proof. The proof is an immediate consequence of the relation

b ◦c=L
r=L b ◦ (c + c ).
0
QED
Since Hamiltonian forms associated with an almost regular quadratic Lagrangian

(5.6.1) constitute a complete family, that is, each solution of the Lagrange equations
is also a solution of some Hamilton equations, we can conclude that each solution
of the Lagrange equations is a solution of some dynamic equation.
5.7 Hamiltonian conservation laws

As was mentioned above, the notion of integrals of motion as functions in involution
with a Hamiltonian can not be extended to time-dependent mechanics because its
Hamiltonian is not a scalar function under time-dependent transformations.
In Section 4.12, we have studied the conservation laws in Lagrangian time-
dependent mechanics. Therefore, in order to discover the conservation laws in
Hamiltonian time-dependent mechanics, let us consider the following construction.
Given a Hamiltonian form H (5.2.12) on a momentum phase space V ∗ Q, let us
consider the Lagrangian
def
LH =(pi qti − H)dt (5.7.1)
(cf. (4.4.6)) on the jet manifold J 1 V ∗ Q of the momentum phase space V ∗ Q → R.

Note that the Lagrangian (5.7.1) plays a prominent role in the functional integral
approach to mechanics (see Remark 5.11.4 below). Here, our utilization of this
Lagrangian is based on the following two assertions.
Lemma 5.7.1. The Poincaré–Cartan form for the Lagrangian LH (5.7.1) coincides
with the pull-back onto J 1 V ∗ Q of the Hamiltonian form H by projection J 1 V ∗ Q →
V ∗ Q, while the Euler–Lagrange operator for LH coincides with the pull-back onto
J 2 V ∗ Q of the Hamilton operator for H by projection J 2 V ∗ Q → J 1 V ∗ Q. 2

It follows from this Lemma that the Lagrange equations for the Lagrangian LH
(5.7.1) are equivalent to the Hamilton equations for the Hamiltonian form H.
Lemma 5.7.2. Let u be a vector field (4.12.2) on a configuration space Q, and
ue = ut ∂t + ui ∂i − ∂i uj pj ∂ i (5.7.2)
its lift onto the momentum phase space V ∗ Q in accordance with the formula (1.6.15).
Then, there is the equality
Leu H = LJ 1 eu LH ,
where J 1 ue is the jet prolongation (1.3.7) of the vector field ue onto J 1 V ∗ Q. 2
Thus, we can use the general procedure of constructing Lagrangian conservation

laws in Section 4.12 for the Lagrangian LH in order to obtain the Hamiltonian
conservation laws.
Let us apply the first variational formula (4.8.4) to the Lagrangian (5.7.1). Then
the weak identity (4.12.3) reads
−ui ∂i H − ut ∂t H + pi dt ui ≈ −dt T,
e (5.7.3)
e = −p ui + ut H.
T (5.7.4)
i
In the case of a vertical vector field u, when ut = 0, the weak identity (5.7.3)
leads to the equality
LJ 1 eu LH ≈ dt (pi ui ).
If LJ 1 eu LH = 0, we have the weak conservation law
0 ≈ dt (pi ui )
of the Noether current

e = −p ui .
T (5.7.5)
i
For a reference frame Γ (5.2.7) on the configuration bundle Q → R, the weak

identity (5.7.3) takes the form
−∂t H − Γi ∂i H + pi dt Γi ≈ −dt H
f ,
Γ
5.8. TIME-DEPENDENT SYSTEMS WITH SYMMETRIES 273
where Hf = H − p Γi is the Hamiltonian function in the splitting (5.2.14).

Γ i
In the case of semiregular Lagrangians, there is the following relationship between
Lagrangian and Hamiltonian conserved currents.
Proposition 5.7.3. Let H be a Hamiltonian form on the momentum phase space

V ∗ Q, which is associated with a semiregular Lagrangian L on the velocity phase
space J 1 Q. Let r be a solution of the Hamilton equations (5.2.20) – (5.2.21) for
H, which lives in the Lagrangian constraint space N , and c the associated solution
of the Lagrange equations for the Lagrangian L so that the conditions (5.5.23) are
satisfied. Let u be a vector field (4.12.2) on the configuration space Q → R. Then
we have
T(r)
e c ◦ r),
= T(H
e L
T( b ◦ ċ) = T(ċ),
where T is the current (4.12.4) on the configuration space J 1 Q, and T

e is the current
(5.7.4) on the phase space V ∗ Q. 2
In accordance with Proposition 5.7.3, the Hamiltonian counterpart of the Lagran-

gian energy function TΓ (4.12.16) relative to a reference frame Γ is the Hamiltonian
function H
f in the splitting (5.2.14). Therefore, one can think of the Hamiltonian
Γ
function HΓ as being the Hamiltonian energy function relative to the reference frame
f
Γ. In particular, if Γi = 0, we obtain the well-known energy conservation law
∂t H ≈ dt H
relative to the coordinates adapted to the reference frame Γ. This is the Hamiltonian
variant of the Lagrangian energy conservation law (4.12.17).
5.8 Time-dependent systems with symmetries

In mechanics, just as in field theory, systems with symmetries can be described in
terms of bundles with structure groups. However, the strength of gauge potentials
in mechanics is always equal to zero. Therefore, the Yang-Mills Lagrangians in
mechanics cannot be constructed, and gauge potentials fail to be dynamic variables.
They define reference frames. Any gauge potential in time-dependent mechanics

can be brought into zero by gauge transformations.
Let πP : P → R be a principal bundle with a structure Lie group G. Let
Q = (P × M )/G (5.8.1)
be a P -associated fibre bundle.

By gauge transformations of a principal bundle P are meant its vertical auto-
morphisms Φ which are equivariant with respect to the canonical action (1.5.4) of
the structure group G on P on the right, i.e.,
Φ ◦ rg = rg ◦ Φ, ∀g ∈ G. (5.8.2)
Due to the property (5.8.2), gauge transformations Φ of a principal bundle yield the
gauge transformations of a P -associated fibre bundle
ΦQ : (P × M )/G → (Φ(P ) × M )/G.
Thus, one can treat a fibre bundle Q (5.8.1) as a configuration space of a mechanical
system with the gauge symmetry group G.
Every gauge transformation of the configuration bundle Q → R gives rise canoni-
cally to the holonomic vertical automorphisms (5.3.1) of the momentum phase space
V ∗ Q → R. They are canonical automorphisms for the canonical Poisson structure
(5.1.5) on V ∗ Q. Moreover, this group action is Hamiltonian as follows (see Definition
3.7.1).
Any 1-parameter group of gauge transformations of a principal bundle P defines
a vector field (its generator) on P which is invariant under the canonical action of
the group G on P . There is one-to-one correspondence between these right-invariant
vector fields on P , called principal vector fields, and the sections of the gauge algebra
bundle VG P (1.5.6), whose typical fibre is the right Lie algebra gr .
Accordingly, any 1-parameter group of gauge transformations of a configuration
space Q, associated with P , defines a principal vertical vector field
ξ = ξ i (q)∂i
on the configuration space Q. This vector field gives rise to the vector field
ξe = ξ i ∂i − ∂i ξ j pj ∂ i
5.8. TIME-DEPENDENT SYSTEMS WITH SYMMETRIES 275
on the momentum phase space (see (5.7.2)). It is readily observed that this is the
Hamiltonian vector field for the Noether current
J = pi ξ i (q) (5.8.3)
(5.7.5) which is defined globally on V ∗ Q (see Example 3.7.4). It follows that, given
a finite dimensional Lie group of gauge transformations, one may apply the familiar
theorems of reduction, e.g., Theorem 3.7.9 to time-dependent mechanics, but not
Theorem 3.7.10 and other results which involve the condition of an invariance of a
Hamiltonian under a group action. Hamiltonians in time-dependent mechanics fail
to be invariant under time-dependent transformations, including gauge transforma-
tions.
Remark 5.8.1. It should be emphasized that gauge transformations make up a
group whose suitable Sobolev completion is a Banach Lie group. The possible ex-
tension of the above-mentioned theorems to this infinite-dimensional Lie group in
the spirit of [36] seems promising. At the same time, we will show below that a Ha-
miltonian form may possess a finite-dimensional Lie group of gauge transformations.
•
Let us restrict our consideration to the case of a vector bundle Q → R. A
principal vector field on this vector bundle reads
ξ = αm (t)εm i j q j ∂i , (5.8.4)
where εm i j are generators of the action of the group G on the typical fibre M , and
αm (t) are local functions on R. The vector field (5.8.4) gives rise to the vector field
ξe = αm (t)εm i j q j ∂i − αm (t)εm i j pi ∂ j
on the momentum phase space V ∗ Q. It is readily observed that this vector field
is the Hamiltonian vector field for the Noether current pi ξ i (5.8.3) relative to the
canonical Poisson structure on V ∗ Q. Moreover, these Noether currents constitute
the Lie algebra with respect to the Poisson bracket
{pi ξ i , pj ξ 0j }V = pi [ξ, ξ 0 ]i . (5.8.5)
In particular, let pi ξ i and pi ξ 0i be conserved Noether currents for a Hamiltonian

form H on the momentum phase space V ∗ Q. Since
[ξ,
g e ξe0 ],
ξ 0 ] = [ξ, e ξe0 ] = [J 1 ξ,
J 1 [ξ, e J 1 ξe0 ]
and
LJ 1 ξeLH = 0, LJ 1 ξe0 LH = 0,
we obtain
LJ 1 [ξe,ξe0 ] LH = 0,
i.e., the Noether current (5.8.5) is also conserved. Thus, conserved Noether currents
in time-dependent mechanics constitute a Lie algebra (cf. Example 3.7.4). This
algebra obviously is finite-dimensional, and corresponds to some finite-dimensional
Lie group of gauge symmetries of a Hamiltonian form H. As was shown, an action
of this group on a momentum phase space V ∗ Q is Hamiltonian, with an equivariant
momentum mapping.
This result remains true for Noether currents (4.12.14) in Lagrangian mechanics.
Given a Lagrangian L, the Noether currents (4.12.14) [conserved Noether currents
(4.12.14)] along vector fields (5.8.4) constitute a Lie algebra with respect to the
Poisson bracket (4.8.13).
5.9 Systems with time-dependent parameters

Let us consider a configuration space which is a composite fibre bundle
Q → Σ → R, (5.9.1)
coordinated by (t, σ m , q i ) where (t, σ m ) are coordinates of the fibre bundle Σ → R.

We will treat sections h of the fibre bundle Σ → R as time-dependent parameters,
and say that the configuration space (5.9.1) describes a mechanical system with
time-dependent parameters. Let us call Σ → R the parameter bundle. Note that
the fibre bundle Q → Σ is not necessarily trivial.
Let us recall that, by virtue of Proposition 1.6.2, every section h of the parameter
bundle Σ → R defines the restriction
Qh = h∗ Q (5.9.2)
of the fibre bundle Q → Σ to h(R) ⊂ Σ, which is a subbundle ih : Qh ,→ Q of

the fibre bundle Q → R. One can think of the fibre bundle Qh → R as being a
configuration space of a mechanical system with the background parameter function
h(t).
5.9. SYSTEMS WITH TIME-DEPENDENT PARAMETERS 277
The velocity momentum space of a mechanical system with parameters is the

jet manifold J 1 Q of the composite fibre bundle (5.9.1) which is equipped with the
adapted coordinates
(t, σ m , q i , σtm , qti ).
Let the fibre bundle Q → Σ be provided with a connection
A = dt ⊗ (∂t + Ait ∂i ) + dσ m ⊗ (∂m + Aim ∂i ). (5.9.3)
Then the corresponding vertical covariant differential

f : J 1 Q → V Q,
D Σ
f = (q i − Ai − Ai σ m )∂ ,
D (5.9.4)
t t m t i
(1.6.13) is defined on the configuration space Q. Given a section h of the parameter

bundle Σ → R, its restriction to J 1 ih (J 1 Qh ) ⊂ J 1 Q is precisely the familiar covariant
differential on Qh corresponding to the restriction
Ah = ∂t + (Aim ∂t hm + (A ◦ h)it )∂i (5.9.5)
of the connection A to h(R) ⊂ Σ. Therefore, one may use the vertical covari-
ant differential D
f in order to construct Lagrangians for a mechanical system with
parameters on the configuration space Q (5.9.1).
We will suppose that a Lagrangian L depends on derivatives of parameters σtm
only through the vertical covariant differential D
f (5.9.4), i.e.,
L = L(t, σm , q i , qti − Ai − Aim σtm )dt. (5.9.6)
Such a Lagrangian is obviously degenerate because of the constraint condition
πm + Aim πi = 0.
As a consequence, the total system of the Lagrange equations
(∂i − dt ∂it )L = 0, (5.9.7)

t
(∂m − dt ∂m )L =0 (5.9.8)
admits a solution only if the very particular relation
(∂m + Aim ∂i )L + πi dt Aim = 0 (5.9.9)

holds. However, we believe that parameter functions are background, i.e., inde-
pendent of a motion. Therefore, only the Lagrange equations (5.9.7) should be
considered.
In particular, let us apply the first variational formula (4.12.8) in order to obtain
conservation laws of a mechanical system with time-dependent parameters. Let
u = ut ∂t + um (t, σ k )∂m + ui (t, σ k , q j )∂i , ut = 0, 1, (5.9.10)
be a vector field which is projectable with respect to both the fibration Q → R and
the fibration Q → Σ. Then the Lie derivative Lu L of a Lagrangian L along the
vector field u (5.9.10) reads
[ut ∂t + um ∂m + ui ∂i + dt um ∂m
t
+ dt ui ∂it ]L =
(um − σtm ut )∂m L + πm dt (um − σtm ut ) +
(ui − qti ut )Ei − dt [πi (ut qti − ui ) − ut L],
dt = ∂t + σtm ∂m + qti ∂i + σttm ∂m
t
+ qtti ∂it .
On the shell (5.9.7), the weak identity
[ut ∂t + um ∂m + ui ∂i + dt um ∂m
t
+ dt ui ∂it ]L ≈
(um − σtm ut )∂m L + πm dt (um − σtm ut ) − dt [πi (ut qti − ui ) − ut L]
is fulfilled. If the Lie derivative Lu L vanishes, we obtain the conservation law
0 ≈ (um − σtm ut )∂m L + πm dt (um − σtm ut ) − dt [πi (ut qti − ui ) − ut L] (5.9.11)
for a system with time-dependent parameters.

Now, let us describe such a system in the framework of Hamiltonian formalism.
Its momentum phase space is the vertical cotangent bundle V ∗ Q. Given a connection
(5.9.3), we have the splitting (1.6.12) which reads
V ∗ Q = A(VΣ∗ Q) ⊕(Q × V ∗ Σ),

Q Q
pi dq + pm dσ = pi (dq − Aim dσ m ) + (pm + Aim pi )dσ m .

i m i
Then V ∗ Q can be provided with the coordinates
pi = pi , pm = pm + Aim pi ,
compatible with this splitting. However, these coordinates fail to be canonical in

general. Given a section h of the parameter bundle Σ → R, the submanifold
{σ = h(t), pm = 0}
of the momentum phase space V ∗ Q is isomorphic to the Legendre bundle V ∗ Qh of

the subbundle (5.9.2) of the fibre bundle Q → R, which is the configuration space
of a mechanical system with the parameter function h(t).
Let us consider Hamiltonian forms on the momentum phase space V ∗ Q which
are associated with the Lagrangian L (5.9.6). Given a connection
Γ = ∂t + Γm ∂m (5.9.12)
on the parameter bundle Σ → R, the desired Hamiltonian form
H = (pi dq i + pm dσ m ) − [pi Ai + pm Γm + H(t,

f σ m , q i , p )]dt
i (5.9.13)
can be found, where
B = ∂t + Γm ∂m + B i ∂i , B i = Ai + Aim Γm ,
is the composite connection (1.6.8) on the fibre bundle Q → R, which is defined by

the connection A (5.9.3) on Q → Σ and the connection Γ (5.9.12) on Σ → R. The
Hamiltonian function Hf satisfies the conditions
πi (t, q i , σ m , ∂ i H(t,
f σ m , q i , π )) ≡ π ,
i i
pi ∂ i H f ≡ L(t, q i , σ m , ∂ i H)
f− H f
which are obtained by substitution of the expression (5.9.13) in the conditions (5.5.2)
– (5.5.3).
The Hamilton equations for the Hamiltonian form (5.9.13) read
qti = Ai + Aim Γm + ∂ i H,
f (5.9.14)
j
pti = −pj (∂i A + ∂i Ajm Γm ) − ∂i H,
f (5.9.15)
σtm = Γm , (5.9.16)
i n
ptm = −pi (∂m A + Γ ∂m Ain ) − ∂m H,
f (5.9.17)
whereas the Lagrangian constraint space is given by the equations
pi = πi (t, q i , σ m , ∂ i H(t,
f σ m , q i , p )),
i (5.9.18)
pm + Aim pi = 0. (5.9.19)
The system of equations (5.9.14) – (5.9.19) are related in the sense of Section 4.5
to the Lagrange equations (5.9.7) – (5.9.8). Because of the equations (5.9.17) and
(5.9.19), this system is overdetermined. Therefore, the Hamilton equations (5.9.14)
– (5.9.17) admit a solution living in the Lagrangian constraint space (5.9.18) –
(5.9.19) if the very particular condition, similar to the condition (5.9.9), holds. Since
the Lagrange equations (5.9.8) are not considered, we also can ignore the equation
(5.9.17). Note that the equations (5.9.17) and (5.9.19) determine only the momenta
pm and do not influence other equations.
Therefore, let us consider the system of equations (5.9.14) – (5.9.16) and (5.9.18)
– (5.9.19). Let the connection Γ in the equation (5.9.16) be complete and admit
the integral section h(t). This equation together with the equation (5.9.19) defines
a submanifold V ∗ Qh of the momentum phase space V ∗ Q, which is the momentum
phase space of a mechanical system with the parameter function h(t). The remaining
equations (5.9.14) – (5.9.15) and (5.9.18) are the equations of this system on the
momentum phase space V ∗ Qh , which correspond to the Lagrange equations (5.9.7)
in the presence of the background parameter function h(t).
Conversely, whenever h(t) is a parameter function, there exists a connection Γ on
the parameter bundle Σ → R such that h(t) is its integral section. Then the system
of equations (5.9.14) – (5.9.15) and (5.9.18) – (5.9.19) describes a mechanical system
with the background parameter function h(t). Moreover, we can locally restrict our
consideration to the equations (5.9.14) – (5.9.15) and (5.9.18).
The following examples illustrate the above construction.
Example 5.9.1. Let us consider the 1-dimensional motion of a probe particle in
the presence of a force field whose centre moves. The configuration space of this
system is the composite fibre bundle
R3 → R2 → R,
coordinated by (t, σ, q) where σ is a coordinate of the field centre with respect to

some inertial frame and q is a coordinate of the probe particle with respect to the
field centre. There is the natural inclusion
Q × T Σ 3 (t, σ, q, ṫ, σ̇) 7→ (t, σ, ṫ, σ̇, ẏ = −σ̇) ∈ T Q

Σ
which defines the connection
A = dt ⊗ ∂t + dσ ⊗ (∂σ − ∂q )
on the fibre bundle Q → Σ. The corresponding vertical covariant differential (5.9.4)

reads
D
f = (q + σ )∂ .
t t q
This is precisely the velocity of the probe particle with respect to the inertial frame.
Then the Lagrangian of this particle takes the form
1
L = [ (qt + σt )2 − V (q)]dt. (5.9.20)
2
In particular, we can obtain the energy conservation law for this system. Let us
consider the vector field u = ∂t . The Lie derivative of the Lagrangian (5.9.20) along
this vector field vanishes. Using the formula (5.9.11), we obtain
0 ≈ −πq σtt − dt [πq qt − L],
or
0 ≈ ∂q Lσt − dt [πq (qt + σt ) − L],
where
T = πq (qt + σt ) − L
is an energy function of the probe particle with respect to the inertial reference
frame. •
Example 5.9.2. Let us consider the 1-dimensional oscillator whose frequency is

a time-dependent parameter. In accordance with the adiabatic hypothesis, there
exists a coordinate q such that this oscillator differs from the standard one only in
kinetic energy.
Let us take the configuration space
Q = R × R+ × R → R+ × R → R,
provided with the coordinates (t, σ > 0, q), where σ is the frequency parameter. The
corresponding momentum phase space V ∗ Q is coordinated by (t, σ, q, pσ , pq ). The
oscillator is characterized by the Hamiltonian form
H = pσ dσ + pq dy − Hdt,
1
H = pσ Γσ + (σ 2 p2q + q 2 ). (5.9.21)
2
The corresponding Hamiltonian frame connection
ΓH = ∂t + Γσt ∂σ (5.9.22)
on the configuration bundle Q → R is the composite connection (1.6.8) generated

by some linear connection
Γ = dt ⊗ (∂t + Γ(t)σ∂σ )
on the parameter bundle
Σ = R+ × R → R
and by the trivial connection
A = dt ⊗ ∂t + dσ ⊗ ∂σ
on the fibre bundle Q → Σ.

The Hamiltonian form (5.9.21) is associated with the Lagrangian
1
L = (σ −2 qt2 − q 2 )dt
2
which describes the oscillator with time-dependent frequency σ = h(t) in accordance
with the adiabatic hypothesis.
The Hamilton equations (5.9.14) – (5.9.16) for the Hamiltonian form (5.9.21)
read
qt = σ 2 p q , (5.9.23)
pt = −q, (5.9.24)
σt = Γ(t)σ. (5.9.25)
They also describe the oscillator with the variable frequency σ = h(t) if we put
Γ = h−1 ∂t h. The Lagrangian constraint space equations (5.9.18) are trivial.
Let us consider the canonical transformation
q = σq 0 ,
1
pq = p0 q ,
σ
1 0 0
pσ = p0σ − qp .
σ q
Relative to the new coordinates (t, σ, q 0 , p0 σ , p0 q ), the connection (5.9.22) is written
ΓH = ∂t + Γσ (∂σ − q 0 ∂q0 ),
while the Hamiltonian form (5.9.21) reads
H = p0σ dσ + p0q dq 0 − Hdt,

1
H = p0σ Γσ − p0q q 0 Γ + (p02 + σ 2 q 02 ).
2 q
Accordingly, the Hamilton equations (5.9.23) - (5.9.25) take the form
1 0 σ
p0ty = p Γ − σ2q0, (5.9.26)
σ q
1
qt0 = − q 0 Γσ + p0q , (5.9.27)
σ
σt = Γσt . (5.9.28)
Substitution of σ = h(t) and the equation (5.9.28) in the equations (5.9.26) and
(5.9.27) leads to the Hamilton equations corresponding to the Hamiltonian form
H = p0q dq 0 − Hdt,
1 1
H = − ∂t σq 0 p0q + (p02 + σ 2 q 02 ).
σ 2 q
This Hamiltonian form describes the well-known Berry oscillator where the connec-
tion (5.9.5) is the classical variant of the Berry connection
1
Ah = ∂t − ∂t σq 0 ∂q0
σ
[135, 144]. •
Example 5.9.3. Let us consider an n-body as in Example 4.9.4. Let R3n−3 be the
→
translation-reduced configuration space of the mass-weight Jacobi vectors { ρ A }.
→ →0
Two configurations { ρ A } and { ρ A } are said to define the same shape of the n-
→0 →
body if ρ A = R ρ A for some rotation R ∈ SO(3). This introduces the equivalence
relation between configurations, and the shape space S of the n-body is defined as
the quotient
S = R3n−3 /SO(3)
[119, 120]. Then we have the composite fibre bundle

R × R3n−3 → R × S → R, (5.9.29)
where the fibre bundle
R3n−3 → S (5.9.30)
has the structure group SO(3). The composite fibre bundle (5.9.29) is provided with
the bundle coordinates(t, σ m , q i ), where q i , i = 1, 2, 3, are some angle coordinates,
e.g., the Eulerian angles, while σ m , m = 1, . . . , 3n − 6, are said to be the shape
→
coordinates on S. A section % A (σ m ) of the fibre bundle (5.9.30), called a gauge
convention, determines an orientation of the n-body in a space. Given such a section,
any point {ρα } of the translation-reduced configuration space R3n−3 is written as
→ →
ρA = R(q i ) % A (σ m ).
This relation yields the splitting
˙
→ →
ρ A = ∂i Rq̇ i %A + ∂m % A σ̇ m
of the tangent bundle T R3n−3 , which determines a connection A on the fibre bundle
(5.9.30). This is also a connection on the fibre bundle
R × R3n−3 → R × S,
with the components At = 0. Then the corresponding vertical covariant differential
(5.9.4) reads
f = (q i − Ai σ m )∂ .
D t m t i
With this vertical covariant differential, the total angular velocity of the n-body
takes the form
→ →
fi = ω + A σ m ,
Ω = ai D (5.9.31)
m t
→
where ai are certain kinematic coefficients and ω is the angular velocity of the n-
body as a rigid one. In particular, we obtain the phenomenon of a falling cat if
→
Ω = 0 so that
→
ω = −Am σtm .
The coefficients Am in the expression (5.9.31) are called gauge potentials, and the
mechanics of deformable bodies is sometimes termed gauge mechanics. •
5.10. UNIFIED LAGRANGIAN AND HAMILTONIAN FORMALISM 285
5.10 Unified Lagrangian and Hamiltonian formalism

The relationship between Lagrangian and Hamiltonian formalisms in Section 5.5 is
broken by canonical transformations if the transition functions q i → q 0 i depend on
momenta. The following construction enables us to overcome this difficulty.
Given a configuration bundle Q → R, let V ∗ J 1 Q be the vertical cotangent bundle
of the velocity phase space J 1 Q → R, with coordinates
(t, q i , qti , q̇i , q̇it ),
and let J 1 V ∗ Q be the jet manifold of the phase space V ∗ Q → R, with coordinates
(t, q i , pi , qti , pti ).
Proposition 5.10.1. There is the isomorphism

Π = V ∗J 1Q ∼
= J 1 V ∗ Q, q̇i ←→ pti , q̇it ←→ pi , (5.10.1)
over J 1 Q. 2
Proof. The isomorphism (5.10.1) can be proved by an inspection of transition

functions of the coordinates (q̇i , q̇it ) and (pi , pti ). QED
Due to the isomorphism (5.10.1), one can think of Π as being both the momen-
tum phase space over the velocity phase space J 1 Q and the velocity phase space
over the momentum phase space V ∗ Q. Hence, the space Π can be utilized as the
unified phase space of joint Lagrangian and Hamiltonian formalism.
Remark 5.10.1. In connection with this, note that, according to [10, 127, 180], the
dynamics of autonomous mechanical system described by a degenerate Lagrangian
L on T M is governed by a differential equation on T ∗ M generated by dL, due to
the canonical diffeomorphism between T T ∗ M and T ∗ T M . •
The unified phase space Π is equipped with the holonomic coordinates

(t, q i , qti , pti , pi ), (5.10.2)
where (q i , pti ) and (qti , pi ) are canonically conjugate pairs. It is endowed with the jet
prolongation of the canonical 3-form (4.8.10), which with respect to the coordinates
(5.10.2) reads
Ω = (dpti ∧ dq i + dpi ∧ dqti ) ∧ dt = dt (dpi ∧ dq i ∧ dt), (5.10.3)
where
dt = ∂t + qti ∂i + pti ∂ i
is the total derivative on Π. The corresponding Poisson bracket (5.1.5) takes the
form
∂f ∂g ∂f ∂g ∂g ∂f ∂g ∂f
{f, g}Π = + i
− − . (5.10.4)
∂pti ∂q i ∂pi ∂qt ∂pti ∂q i ∂pi ∂qti
It is readily observed that the canonical form (5.10.3) and, consequently, the
Poisson bracket (5.10.4) are invariant under the transformations of the unified phase
space Π, which are the jet prolongation of the canonical automorphisms of the phase
space V ∗ Q.
Let
H = pti dq i + pi dqti − HΠ (t, q i , qti , pti , pi )dt
be a Hamiltonian form on Π. The corresponding Hamilton equations (5.2.20) –

(5.2.21) read
∂HΠ
dt q i = , (5.10.5)
∂pti
∂HΠ
dt qti = , (5.10.6)
∂pi
∂HΠ
dt pi = − i , (5.10.7)
∂qt
∂HΠ
dt pti = − i . (5.10.8)
∂q
Substitution of (5.10.5) in (5.10.6) and of (5.10.7) in (5.10.8) leads to the equations
∂HΠ ∂HΠ
dt = , (5.10.9)
∂pti ∂pi
∂HΠ ∂HΠ
dt i
= (5.10.10)
∂qt ∂q i
which look like the Lagrange equations for the ”Lagrangian” HΠ . Though HΠ is
not a true Lagrangian, one can put
HΠ = −L + dt (pi Γi ), (5.10.11)
5.10. UNIFIED LAGRANGIAN AND HAMILTONIAN FORMALISM 287
whenever L is a Lagrangian on the velocity phase space J 1 Q. Then the equations

(5.10.9) – (5.10.10) are equivalent to the Lagrange equations for a Lagrangian L on
J 1 Q. However, their solutions fail to be solutions of the corresponding Hamilton
equations (5.10.5) – (5.10.8) for the Hamiltonian (5.10.11) in general. This illustrates
the fact that solutions of the Hamilton equations (5.10.5) – (5.10.8) are necessarily
solutions of the Lagrange equations (5.10.9) – (5.10.10), but the converse statement
is not true.
To construct the joint Lagrangian-Hamiltonian formalism, let us consider the
Hamiltonian form
H = pti dq i + pi dqti − (dt H + (pi qti − H) − L)dt (5.10.12)
on the unified phase space Π, where L is a semiregular Lagrangian on the velocity
phase space J 1 Q and H is an associated Hamiltonian form on the momentum phase
space V ∗ Q. The Hamilton equations (5.10.5) – (5.10.8) for H (5.10.12) read
∂H
dt q i = , (5.10.13)
∂pi
∂H ∂H
dt qti = dt + qti − , (5.10.14)
∂pi ∂pi
∂H ∂L
dt pi = − i − pi + i , (5.10.15)
∂q ∂qt
∂H ∂H ∂L
dt pti = −dt i + i + i . (5.10.16)
∂q ∂q ∂q
Using the relations (5.5.2) and (5.5.8), one can show that solutions of the Hamilton
equations (5.2.20) – (5.2.21) for the Hamiltonian form H, which live in the La-
b 1 Q) ⊂ V ∗ Q are solutions of the equations (5.10.13) –
grangian constraint space L(J
(5.10.16).
Now let us consider the Lagrange equations (5.10.9) – (5.10.10) for the Hamil-
tonian form (5.10.12). They read
∂H
dt q i − = 0, (5.10.17)
∂pi
∂LQ ∂H ∂L
dt pi − dt i = − i − i . (5.10.18)
∂qt ∂q ∂q
In accordance with Proposition 5.5.7, every solution of the Lagrange equations for
the Lagrangian L such that the relation (5.5.24) holds is a solution of the equations
(5.10.17) – (5.10.18).
In particular, if the Lagrangian L is hyperregular, the equations (5.10.17) –

(5.10.18) and the equations (5.10.13) – (5.10.16) are equivalent to the Lagrange
equations for L and the Hamilton equations for the associated Hamiltonian form.
5.11 Vertical extension of Hamiltonian formalism

Let Q → R, coordinated by (t, q i ), be a configuration space of time-dependent
mechanics. Let us consider the vertical tangent bundle V Q of the fibre bundle
Q → R, coordinated by (t, q i , q̇ i ). It can be seen as the new configuration space,
called the vertical configuration space. Here, we will be concerned with the following
three applications of Hamiltonian mechanics on this configuration space:
• mechanical systems with linear deviations (see Example 5.11.1 below),
• Hamiltonian description of non-holonomic constraints (see Example 5.11.3 be-

low),
• Hamiltonian BRST mechanics (see Appendix A).
Given a vertical configuration space V Q, the corresponding velocity phase space

is the first order jet manifold J 1 V Q of the fibre bundle V Q → R. It is canonically
isomorphic (see (1.3.8)) to the vertical tangent bundle V J 1 Q of the ordinary velocity
phase space J 1 Q → R, coordinated by
(t, q i , q̇ i , qti , q̇ti ).
We call V J 1 Q the vertical velocity phase space.

The Legendre bundle of the vertical configuration space V Q is the vertical cotan-
gent bundle V ∗ V Q of the fibre bundle V Q → R. It plays the role of a momentum
phase space, called the vertical momentum phase space, of mechanical systems on
the vertical configuration space V Q. The vertical momentum phase space is canon-
ically isomorphic (see (1.1.11)) to the vertical tangent bundle V V ∗ Q of the ordinary
momentum phase space V ∗ Q → R, coordinated by
(t, q i , pi , q̇ i , ṗi ).
It is easily seen from the transformation laws

∂q j
p0i = pj ,
∂q 0i
5.11. VERTICAL EXTENSION OF HAMILTONIAN FORMALISM 289
∂p0i ∂p0 ∂q j ∂ q̇ j
ṗ0i = ṗj + ji q̇ j = 0i ṗj + 0i pj
∂pj ∂q ∂q ∂q
that (q i , ṗi ) and (q̇ i , pi ) are the canonically conjugate pairs.

Thus, we have observed that the vertical velocity and momentum phase spaces,
corresponding to the vertical configuration space V Q, are the vertical tangent bun-
dles of ordinary velocity and momentum phase spaces, respectively.
The vertical momentum phase space V V ∗ Q is endowed with the canonical 3-form
ΩV = [dṗi ∧ dq i + dpi ∧ dq̇ i ] ∧ dt. (5.11.1)
For brevity, one can write
ΩV = ∂V Ω,
where Ω is the canonical 3-form (5.1.9) on the momentum phase space V ∗ Q, and
the formal operator of the vertical derivative
∂V = q̇ i ∂i + ṗi ∂ i
(1.1.4) is utilized. This operator possesses the properties
∂V (φ ∧ σ) = ∂V (φ) ∧ σ + φ ∧ ∂V (σ), φ, σ ∈ O∗ (V ∗ Q),

∂V (dφ) = d(∂V (φ)).
The canonical 3-form (5.11.1) provides the vertical momentum phase space
V V ∗ Q with the canonical Poisson structure, given by the Poisson bracket
{f, g}V V = ∂˙ i f ∂i g + ∂ i f ∂˙i g − ∂ i g ∂˙i f − ∂˙ i g∂i f.
Recall the notation (1.2.4).

The notions of a Hamiltonian vector field, a Hamiltonian connection, a Hamil-
tonian form and others on the vertical momentum phase space V V ∗ Q ∼ = V ∗ V Q are
introduced in accordance with the general formalism of Hamiltonian time-dependent
mechanics, expressed in Section 5.2. In particular, any Hamiltonian form on V V ∗ Q
reads
HV = ṗi dq i + pi dq̇ i − HV dt.

Since a Hamiltonian form is defined modulo exact forms and the function pi q̇ i is glob-
ally defined on V V ∗ Q, we will write a Hamiltonian form on the vertical momentum
phase space V V ∗ Q as
HV = ṗi dq i − q̇ i dpi − HV dt. (5.11.2)
The corresponding Hamilton equations read
γ i = qti = ∂˙ i HV , (5.11.3)
γi = pti = −∂˙i HV , (5.11.4)
γ̇ i = q̇ti = ∂ i HV , (5.11.5)
γ̇i = ṗti = −∂i HV , (5.11.6)
where
γ̇ = ∂t + γ i ∂i + γi ∂ i + γ̇ i ∂˙i + γ̇i ∂˙ i
is a Hamiltonian connection on the vertical momentum phase space V V ∗ Q → R.

We have the following relationship between Hamiltonian formalism on the mo-
mentum phase space V ∗ Q and that on the vertical momentum phase space V V ∗ Q.
Proposition 5.11.1. Let γH be a Hamiltonian connection on the ordinary mo-

mentum phase space V ∗ Q → R for a Hamiltonian form
H = pi dq i − Hdt. (5.11.7)
Then the connection V γH (1.6.14) on the vertical momentum phase space V V ∗ Q →

R is a Hamiltonian connection for the Hamiltonian form
HV = ṗi dq i − q̇ i dpi − ∂V Hdt, (5.11.8)

∂V H = (q̇ i ∂i + ṗi ∂ i )H,
called the vertical extension of the Hamiltonian form H. 2
Proof. A direct computation shows that, given a Hamiltonian connection
γH = ∂t + γ i ∂i + γi ∂ i ,
γ i = ∂ i H, γi = −∂i H,
on the fibre bundle V ∗ Q → R, the connection
V γH = ∂t + γ i ∂i + γi ∂ i + γ̇ i ∂˙i + γ̇i ∂˙ i ,
γ̇ i = ∂V γ i , γ̇i = ∂V γi ,
on the fibre bundle V V ∗ Q → R satisfies the Hamilton equations (5.11.3) – (5.11.6)

for the Hamiltonian form (5.11.8), i.e.,
γ i = ∂˙ i HV = ∂ i H, (5.11.9)
γi = −∂˙i HV = −∂i H, (5.11.10)
i i i
γ̇ = ∂ HV = ∂V ∂ H, (5.11.11)
γ̇i = −∂i HV = −∂V ∂i H. (5.11.12)
QED
In particular, given a splitting
H = pi Γi + H
f
Γ
of a Hamiltonian H with respect to a connection Γ on Q → R, there is the corre-

sponding splitting
HV = ṗi Γi − q̇ j (−pi ∂j Γi ) + ∂V H
f
Γ
of the Hamiltonian HV (5.11.8) with respect to the connection Γ e (5.2.8) on the fibre
bundle V ∗ Q → R.
It is easily seen that the Hamilton equations (5.11.9) – (5.11.10) for the Hamil-
tonian form HV (5.11.8) are precisely the Hamilton equations (5.2.20) – (5.2.21) for
the Hamiltonian form H.
Example 5.11.1. In order to clarify the physical meaning of the Hamilton equations
(5.11.11) – (5.11.12), let us consider Q → R a vector bundle. Due to the splitting
VQ∼
= Q ⊕ Q,
we can think of the vertical configuration space V Q as the configuration space of

the initial mechanical system and its small linear deviations
q i + εq̇ i , ε ∈ R,
where ε is a small parameter. Accordingly, the vertical momentum phase space
V V ∗Q ∼
= V ∗Q ⊕ V ∗Q
can be seen in a similar way as the momentum phase space of the initial Hamiltonian
mechanical system and its small linear deviations
q i + εq̇ i , pi + εṗi .
Let H be a Hamiltonian form on the momentum phase space V ∗ Q of the initial

Hamiltonian system, and let r(t) be a solution of the Hamilton equations (5.11.9) –
(5.11.10). Let r(t) be a Jacobi field, i.e.,
r(t) + εr(t)
is also a solution of the same Hamilton equations modulo terms of order two in ε.
Then it is readily observed that the Jacobi field r(t) satisfies the Hamilton equa-
tions (5.11.11) – (5.11.12) for the Hamiltonian form HV (5.11.8). Substituting a
solution r(t) of the Hamilton equations (5.11.9) – (5.11.10) in the Hamilton equa-
tions (5.11.11) – (5.11.12), we obtain the system of homogeneous linear differential
equations for the Jacobi field r(t), which are really the differential equations of linear
deviations. •
Returning to the definition of a Hamiltonian form HV (5.11.8), let us consider

the vertical tangent bundle V T ∗ Q of the cotangent bundle T ∗ Q → R. It is equipped
with holonomic coordinates
(t, q i , pi , p, q̇ i , ṗi , ṗ)
and endowed with the canonical form
ΞV = ṗdt + ṗi dq i − q̇ i dpi .
One may also utilize another canonical form
ΞV + d(q̇ i pi )
since the exact form d(q̇ i pλi ) is globally defined. By very definition, a Hamiltonian
form H (5.11.7) on the ordinary momentum phase space V ∗ Q is the pull-back H =
h∗ Ξ of the Liouville form Ξ (5.1.2) on the cotangent bundle T ∗ Q by a section h of

the fibre bundle T ∗ Q → V ∗ Q. Then we have
HV = (V h)∗ ΞV ,
where
V h : V V ∗Q → V T ∗Q
is the vertical tangent map to h.

We can also generalize Proposition 5.11.1 as follows.
Proposition 5.11.2. Any connection γ on the momentum phase space V ∗ Q →

R gives rise to a Hamiltonian connection on the vertical momentum phase space
V V ∗ Q → R. 2
Proof. Let us consider the Hamiltonian form
HV = ṗi (dq i − γ i dt) − q̇ i (dpi − γi dt) = ṗi dq i − q̇ i dpi − (ṗi γ i − q̇ i γi )dt
on the vertical momentum phase space V V ∗ Q. The corresponding Hamiltonian

connection on V V ∗ Q → R is given by the Hamilton equations (5.11.3) – (5.11.6)
and reads
γ i = γ i, γi = γi , γ̇ i = ṗj ∂ i γ j − q̇ j ∂ i γj , γ̇i = −ṗj ∂i γ j + q̇ j ∂i γj . (5.11.13)
In particular, if γ is a locally Hamiltonian connection on the fibre bundle V ∗ Q → R,

i.e., satisfies the relations (5.2.4) – (5.2.6), then the Hamiltonian connection (5.11.13)
is precisely the connection V γ (1.6.14). QED
It follows that every first order dynamic equation (5.2.23) on the momentum
phase space V ∗ Q can be seen as the Hamilton equations (5.11.3) – (5.11.4) for a
suitable Hamiltonian form on the vertical momentum phase space.
Example 5.11.2. The equations of a motion (5.2.24) of a point mass m subject
to friction in Example 5.2.5 are the Hamilton equations (5.11.3) – (5.11.4) for the
Hamiltonian
1 k
HV = p( ṗ + q̇)
m m
on the vertical momentum phase space. •
Similarly to the relationship between Hamiltonian formalisms on ordinary and

vertical momentum phase spaces, that between Lagrangian formalisms on ordinary
and vertical velocity phase spaces is stated.
Given a Lagrangian L = Ldt (4.8.1) on an ordinary velocity phase space J 1 Q,
let us consider the vertical tangent morphism
V L : J 1Q → R × R
to the Lagrangian function L and the Lagrangian
LV = (pr2 ◦ V L)dt = LV dt, (5.11.14)

LV = ∂V L = (q̇ i ∂i + q̇ti ∂it )L,
on the vertical velocity phase space V J 1 Q. It is called the vertical extension of the
Lagrangian L. The corresponding Lagrange equations read
(∂˙i − dt ∂˙it )LV = 0, (5.11.15)

(∂i − dt ∂it )LV = 0. (5.11.16)
The Lagrange equation (5.11.15) is precisely the Lagrange equation for the initial
Lagrangian L. If a configuration space Q → R is a vector bundle, the Lagrange
equation (5.11.16), just as the Hamilton equations (5.11.11) – (5.11.12) do, describes
small linear deviations of the initial mechanical system.
Let us turn now to the relationship between Lagrangian and Hamiltonian for-
malisms on the vertical configuration space.
Each Lagrangian LV (5.11.14) on the vertical velocity phase space V J 1 Q yields
the Legendre map
L b : V J 1 Q → V V ∗ Q,
b =VL (5.11.17)
V
VQ
pi = ∂˙it LV = πi , ṗi = ∂V πi , (5.11.18)
which is the vertical tangent map to the Legendre map L.

b
Each Hamiltonian form HV (5.11.8) on the vertical momentum phase space
V V ∗ Q yields the Hamiltonian map
H c : V V ∗ Q → V J 1 Q,
c =VH (5.11.19)
V
VQ
qti = ∂˙ i HV = ∂ i H, q̇ti = ∂V ∂ i H,
which is the vertical tangent map to the Hamiltonian map H.

c
Proposition 5.11.3. If a Hamiltonian form H on the momentum phase space V ∗ Q

is associated with a Lagrangian L on the velocity phase space J 1 Q, then its vertical
extension HV (5.11.8) on the vertical momentum phase space V V ∗ Q is associated
with the vertical extension LV (5.11.14) of the Lagrangian L on the vertical velocity
phase space V J 1 Q. 2
Proof. By virtue of the relations (5.11.17) and (5.11.19), the condition (5.5.2) for
the Hamiltonian form HV reads
b ◦VH
VL c◦VL
b = V L,
b
and follows at once from the condition (5.5.2) for the Hamiltonian form H. The con-
dition (5.5.3) for the Hamiltonian form HV , written in the coordinate form (5.5.6),
follows immediately from the condition (5.5.6) for the Hamiltonian form H. QED
The equalities (5.11.18) show that, if a Lagrangian L on the velocity phase space
1
J Q is semiregular, so is its vertical extension LV (5.11.14) on the vertical velocity
phase space V J 1 Q. Therefore, Proposition 5.11.3 can also be formulated for a
semiregular Lagrangian. In particular, if the relation (5.5.17) is fulfilled for H and
L, it is true for the vertical extensions HV and LV .
Example 5.11.3. Let us consider the Hamiltonian counterpart of the constrained
motion equation (4.11.15) in the case of a hyperregular Lagrangian and a non-
holonomic constraint. By virtue of Proposition 5.11.2, it can be represented as
the Hamilton equations (5.11.3) – (5.11.4) for a suitable Hamiltonian form on the
vertical phase space V V ∗ Q.
Let L be a hyperregular Lagrangian with a Riemannian mass metric m c and
1
S a codistribution on the velocity phase space J Q, whose annihilator Ann (S) is
an admissible non-holonomic constraint on J 1 Q. In the case of a hyperregular
Lagrangian, Hamiltonian and Lagrangian formalisms of time-dependent mechanics
are equivalent, and we have
γH = J 1 L
b ◦ ξ ◦ H.
L
c
Let introduce the notation M ij = ∂ i ∂ j H. There are the relations
M ik (mkj ◦ H)
c = δi ,
j mkj (M ik ◦ L)
b = δi ,
j mij = πij .
It follows that M is a fibre metric in the vertical tangent bundle VQ V ∗ Q of the fibre
bundle V ∗ Q → Q.
Given the above-mentioned codistribution S on J 1 Q, let us consider the pull-back
codistribution Hc∗ S on V ∗ Q spanned locally by 1-forms
c∗ sa = (sa + ṡa ∂ ∂ j H)dt + (sa + ṡa ∂ ∂ j H)dq i + ṡa M ij dp =

βa = H 0 j t i j i i j
β0a dt + βia dq i + β̇ ai dpi .

This codistribution defines a non-holonomic constraint on the momentum phase
space V ∗ Q.
Given a Hamiltonian connection γH (5.2.19), let us find its splitting
γH = γe + ϑ (5.11.20)
where γe is a connection on V ∗ Q → R which satisfies the condition
c∗ S).
γe ⊂ Ann (H (5.11.21)
The connection γe (5.11.21) obviously defines a first order dynamic equation on the
momentum phase space V ∗ Q, which is compatible with the non-holonomic con-
straint Hc∗ S. The decomposition (5.11.20) is not unique. Let us construct it as
follows.
Given a Hamiltonian connection γH , we consider the codistribution SH on V ∗ Q,
spanned locally by the 1-forms dq i − γHi
dt. Its annihilator Ann (SH ) is an affine
subbundle of the affine jet bundle J V Q → V ∗ Q modelled over the vertical tan-
1 ∗
gent bundle VQ V ∗ Q. The Hamiltonian connection γH is obviously a section of this

subbundle. Let us consider the intersection
c∗ S).
W = Ann (SH ) ∩ Ann (H (5.11.22)
Lemma 5.11.4. The intersection W (5.11.22) is an affine bundle over V ∗ Q, modelled

over the vector bundle
W = VQ V ∗ Q ∩ Ann (H
c∗ S).
Proof. The intersection W consists of elements v of VQ V ∗ Q which fulfill the con-

ditions
vi β̇ ai = 0.
Since the non-holonomic constraint S is admissible and the matrix M ij is non-

degenerate, every fibre of W is of dimension m − n, i.e., W is a vector bundle, while
W is an affine bundle. QED
Then, using the fibre metric M in VQ V ∗ Q, we obtain the splitting
VQ V ∗ Q = W ⊕ V,
where V is the orthocomplement of W , and the associated splitting
Ann (SH ) = W ⊕ V.
The corresponding decomposition (5.11.20) reads

f M β̇ ai β b (γ )∂ j ,
γe = γH − M (5.11.23)
ab ij H
where M
f is the inverse matrix of
ab
f ab = β̇ ai β̇ bj M .
M ij
The splitting (5.11.23) is the Hamiltonian counterpart of the splitting (4.11.15).

We have the relations
fab = M
m f ab ◦ H,
c β a (γH ) = sa (ξL ) ◦ H,
c
and as a consequence
γe = J 1 L ◦ ξe ◦ H.
c
Note that the above procedure can be extended in a straightforward manner to

any standard Newtonian system seen as a Lagrangian system with the Lagrangian
(4.9.19) and an external force. Following this procedure, one may also study a non-
holonomic Hamiltonian system, without appealing to its Lagrangian counterpart.
The connection (5.11.23) defines the system of first order dynamic equations
qti = ∂ i H, f M β̇ aj β b (γ )
pti = −∂i H − M ab ij H (5.11.24)
on the momentum phase space V ∗ Q, which are not Hamilton equations. Neverthe-
less, in accordance with Proposition 5.11.2, one can restate the constrained motion
equations (5.11.24) as the Hamilton equations (5.11.9) – (5.11.10) for the Hamilto-
nian form
HV = ṗi dq i − q̇ i dpi − ∂V Hdt − q̇ i M

f M β̇ aj β b (γ )dt
ab ij H
on the vertical momentum phase space V V ∗ Q. Its last term can be written in brief
as (−∂V cϑcΩ). To exclude the remaining Hamilton equations (5.11.11) – (5.11.12),
one can choose their solutions q̇ i = ṗi = 0. •
Remark 5.11.4. Let HV be a Hamiltonian form (5.11.2) on the vertical momen-

tum phase space V V ∗ Q. Similarly to the Lagrangian (5.7.1), let us consider the
Lagrangian
LH = ṗi qti − q̇ i pti − HV (5.11.25)
on the first order jet manifold J 1 V V ∗ Q of the fibre bundle V V ∗ Q → R. It is readily

observed that its Lagrange equation is precisely the Hamilton equations (5.11.3) –
(5.11.6) for the Hamiltonian form HV . In particular, let H be a Hamiltonian form
on an ordinary momentum phase space V ∗ Q and HV = ∂V H. In this case, the
Lagrangian (5.11.25) reads
LH = ṗi (qti − ∂ i H) − q̇ i (pti + ∂i H).
It is easily seen that this Lagrangian vanishes on solutions of the Hamilton equations
for the Hamiltonian form H. For this reason, it plays a prominent role for the
functional integral formulation of mechanics [65, 66] •
5.12 Appendix. Time-reparametrized mechanics

We have assumed above that the base R of a configuration space of time-dependent
mechanics is parameterized by a coordinate t with the transition functions t → t0 =
t+ const. Here, we consider an arbitrary reparametrization of time
t → t0 = f (t) (5.12.1)
which is discussed in some models of quantum mechanics [75, 153, 154].

In the case of an arbitrary time reparametrization (5.12.1), a configuration space
of time-dependent mechanics is a fibre bundle Q → R over a 1-dimensional base R,
diffeomorphic to R. Let R be coordinated by t with the transition functions (5.12.1).
In contrast with R, the base R does not admit either the standard vector field ∂t
or the standard 1-form dt. Therefore, we should strictly follow the polysymplectic
5.12. APPENDIX. TIME-REPARAMETRIZED MECHANICS 299
Hamiltonian formalism of Section 3.8. Nevertheless, Hamiltonian formalism of time-

reparametrized mechanics possesses some peculiarities because of a 1-dimensional
base R.
1. There exists the canonical tangent-valued 1-form
θR = dt ⊗ ∂t
on the base R of a configuration space of time-reparametrized mechanics.
2. The velocity phase space J 1 Q of time-reparametrized mechanics is not an
affine subbundle of the tangent bundle T Q, whereas the momentum phase space
Π → Q (3.8.1) is isomorphic to the vertical cotangent bundle V ∗ Q → Q. It fol-
lows that the momentum phase space of time-reparametrized mechanics is provided
with the canonical Poisson structure (5.1.5). Moreover, this Poisson structure is
invariant under time reparametrization (5.12.1) which, consequently, is a canonical
transformation.
3. The momentum phase space V ∗ Q is endowed with the canonical polysym-
plectic form
Λ = dpi ∧ dq i ∧ θR .
Then the notions of a Hamiltonian connection and a Hamiltonian form are the rep-
etitions of those in Section 3.8. At the same time, since the homogeneous Legendre
bundle (2.9.3) of time-reparametrized mechanics is the cotangent bundle T ∗ Q of
Q, Hamiltonian forms and Hamilton equations of time-reparametrized mechanics
follow the fashion of those in time-dependent mechanics. The difference is only that
a Hamiltonian function H f in the splitting (5.2.14) is a density, but not a function
Γ
under the transformations (5.12.1).
4. Since a Lagrangian and a Hamiltonian of time-reparametrized mechanics
are densities under the transformations (5.12.1), one should introduce a volume
element on the base R in order to construct them in explicit form. The key prob-
lem of models with time reparametrization lies in the fact that the time axis R of
time-reparametrized mechanics has no canonical volume element. Another prob-
lem is concerned with a mass tensor. Since the velocity phase space J 1 Q of time-
reparametrized mechanics is an affine bundle J 1 Q → Q modelled over the vector
bundle T ∗ R ⊗ V Q, a mass tensor fails to be invariant under time reparametrization
Q
(5.12.1).
Chapter 6
Relativistic mechanics
Let us note on the following two main peculiarities of relativistic mechanics in com-
parison with non-relativistic one.
• A configuration space Q of relativistic mechanics has no fibration Q → R

in general or, at least, it has no preferable fibration over R. Therefore, one
cannot use the familiar formalism of jets of sections of fibre bundles, but must
generalize it for jets of submanifolds.
• Hamiltonian relativistic mechanics is described as an autonomous constraint

Dirac system on the hyperboloids of relativistic momenta in the momentum
phase space T ∗ Q.
Relativistic mechanics can be formulated on an arbitrary configuration space Q.

If Q is a 4-dimensional manifold provided with a pseudo-Riemannian metric g, we
come to relativistic mechanics in the presence of a gravitational field g. If Q = R4
and g is the Minkowski metric, this is the case of Special Relativity.
6.1 Jets of submanifolds

We start from the general notion of jets of submanifolds [57, 104].
Definition 6.1.1. Let Z be a manifold of dimension m + n. The first order

jet manifold Jn1 Z of n-dimensional submanifolds of a manifold Z comprises the
equivalence classes [S]1z , z ∈ Z, of n-dimensional imbedded submanifolds of Z which
301
302 CHAPTER 6. RELATIVISTIC MECHANICS
pass through z ∈ Z and which are tangent to each other at z. There is the natural
fibration Jn1 Z → Z. 2
Remark 6.1.1. In fact, the definition of the jet [S]1z of submanifolds involves only
local properties of submanifolds around the point z ∈ Z. It is easily extended to
higher order jets of submanifolds. By definition, Jn0 Z = Z. •
The set of jets Jn1 Z is provided with a manifold structure as follows.

Let Y → X be an (m + n)-dimensional fibre bundle over an n-dimensional base
X and let Φ be an imbedding of Y into Z. Then there is the natural injection of
the jet manifold J 1 Y of the fibre bundle Y → X into Jn1 Z such that
J 1 Φ : J 1 Y → Jn1 Z, (6.1.1)
jx1 s 7→ [S]1Φ(s(x)) , S = Im (Φ ◦ s),
where s are sections of Y → X.
Proposition 6.1.2. The injection (6.1.1) defines a chart on Jn1 Z. Such charts
cover the set Jn1 Z, and transition functions between these charts are differentiable.
They provide the set Jn1 Z with the structure of a finite-dimensional manifold. 2
Proof. The proof is based on the fact that, given a submanifold S ⊂ Z which
belongs to the jet [S]1z , there exists a neighbourhood Uz of the point z and the
tubular neighbourhood US of Uz ∩ S so that the fibration
US → Uz ∩ S
takes place [109]. This means that every jet [S]1z belongs to a chart of the above-
mentioned type. We will describe these charts and the corresponding transition
functions in explicit form. QED
It is convenient to use the following coordinate atlases of the jet manifold Jn1 Z
of n-dimensional submanifolds of Z.
Let Z be equipped with a coordinate atlas
{(U ; z A )}, A = 1, . . . , n + m. (6.1.2)

6.1. JETS OF SUBMANIFOLDS 303
Though Jn0 Z, by definition, is diffeomorphic to Z, let us provide Jn0 Z with the atlas
where every chart (U ; z A ) on a domain U ⊂ Z is replaced with the
!
n+m (n + m)!
=
m n!m!
charts on the same domain U which correspond to different partitions of the collec-
tion (z 1 · · · z A ) in the collections of n and m coordinates, denoted by
(xλ , y i ), λ = 1, . . . , n, i = 1, . . . , m. (6.1.3)
The transition functions between the coordinate charts (6.1.3) of Jn0 Z, associated
with the same coordinate chart (6.1.2) of Z, are reduced simply to an exchange
between coordinates xλ and y i . Transition functions between arbitrary coordinate
charts (6.1.3) of the manifold Jn0 Z read
xeλ = geλ (xµ , y j ), yei = fei (xµ , y j ), (6.1.4)

xλ = g λ (xeµ , yej ), y i = f i (xeµ , yej ).
Given the coordinate atlas (6.1.3) of the manifold Jn0 Z, the jet manifold Jn1 Z is
provided with the adapted coordinates
(xλ , y i , yλi ), λ = 1, . . . , n, i = 1, . . . , m. (6.1.5)
Remark 6.1.2. If S ⊂ Z is an imbedded submanifold which belongs to the jet

[S]1z , there exists an open neighbourhood Uz of z ∈ Z and a coordinate chart (xλ , y i )
(6.1.3) which covers Uz so that S ∩ Uz is given by the coordinate relations
y i = S i (xλ )
and
(xλ , y i , yλi )([S]1z ) = (xλ (z), S i (xµ (z)), ∂λ S i (xµ (z))).
Using the operators of total derivatives
dλ = ∂λ + yλi ∂i ,
one can obtain the transition functions of the coordinates (6.1.5) under coordinate
transformations (6.1.4) in explicit form. Given coordinate transformations (6.1.4),
one can easily find that
dexλ = [dexλ g α (xeµ , yei )] dxα . (6.1.6)
Then we have
" ! # !
∂ ∂ ∂ ∂
yeλi = λ
+ yeλp p g α (xeλ , yei ) α
+ yαj j fei (xµ , y j ). (6.1.7)
∂ xe ∂ ye ∂x ∂y
It is readily observed that the transition functions (1.3.1) are a particular case of
the coordinate transformations (6.1.7) when the transition functions g α (6.1.4) are
independent of coordinates yei . Note that, in contrast with (1.3.1), the coordinate
transformations (6.1.7) are not affine. It follows that the fibration Jn1 Z → Z is not
an affine bundle.
There is one-to-one correspondence
e : [S]1 7→ ẋλ (∂ + y i ([S]1 )∂ )
λ (6.1.8)
z λ λ z i
between the jets [S]1z at a point z ∈ Z and n-dimensional vector subspaces of the
tangent space Tz Z. It follows that the jet bundle Jn1 Z → Z has the structure group
GL(n, m; R) ⊂ GL(m + n, R)
of linear transformations of the vector space Rm+n which preserve its subspace Rn .
Its typical fibre is the Grassman manifold
G(m + n, n) = GL(n + m; R)/GL(n, m; R)
of n-dimensional subspaces of the vector space Rm+n . Thus, the jet bundle Jn1 Z → Z
is a Grassman bundle [45, 70]. In particular, if n = 1, the fibre coordinates y0i on
J11 Z → Z with the transition functions (6.1.7) are precisely the standard coordinates
of the projective space RPm .
Example 6.1.3. Let Y → X be an (m + n)-dimensional fibre bundle over an n-
dimensional base X, and J 1 Y the first order jet manifold of its sections. Let Jn1 Y
be the first order jet manifold of n-dimensional subbundles of Y . Then the injection
J 1 Y ,→ Jn1 Y (6.1.1) is an affine subbundle of the jet bundle Jn1 Y → Y . Its fibre
at a point y ∈ Y consists of the n-dimensional subspaces of the tangent space Ty Y
whose intersections with the vertical subspace Vy Y of Ty Y are the zero vector. •
6.2. RELATIVISTIC VELOCITY AND MOMENTUM PHASE SPACES 305
Remark 6.1.4. Generalizing the notion of a connection on a bundle, one may

think of global sections of the jet bundle Jn1 Z → Z as being preconnections on the
manifold Z [134]. By virtue of the well-known theorem [80], if such a preconnection
Γ exists, its image Γ(Z) in the tangent bundle T Z → Z by the correspondence λ e
(6.1.8) is a vector subbundle of T Z with the structure group GL(n, R). The quotient
T Z/Γ(Z) is also a vector subbundle with the structure group GL(m, R). Thus, we
have the decomposition
T Z = Γ(Z) ⊕ T Z/Γ(Z)
of the tangent bundle T Z, which can be treated as a horizontal splitting with respect
to the preconnection Γ. However, it should be emphasized that, since Jn1 Z → Z is
not an affine bundle, preconnections do not make an affine space. Moreover, it may
happen that a preconnection does not exist on a manifold Z. •
6.2 Relativistic velocity and momentum phase spaces

Let us now turn to the case of relativistic mechanics, where Z = Q is an (m + 1)-
dimensional configuration space which has not necessarily a fibration or a preferable
fibration over the time axis R. We call Q a relativistic configuration space in compar-
ison with a non-relativistic configuration space Q endowed with a fibration Q → R.
This configuration space is provided with the coordinate atlas (6.1.3) together with
the transition functions (6.1.4), i.e.,
(q 0 , q i ), i = 1, . . . , m,
q 0 → qe0 (q 0 , q j ), q i → qei (q 0 , q j ). (6.2.1)
The coordinates q 0 in different charts of Z play the role of a temporal coordinate.

Note that, given a coordinate chart (U ; q 0 , q i ), we have a local fibre bundle
U 3 (q 0 , q i ) 7→ q 0 ∈ () ⊂ R (6.2.2)
which can be treated as a configuration space of a local non-relativistic mechanical

system.
The velocity phase space of relativistic mechanics on the configuration space Q
is the first order jet manifold J11 Q of 1-dimensional subbundles of Q. It is endowed
with the coordinates (q 0 , q i , q0i ) (6.1.5). Their transition functions are obtained as
follows.
Given coordinate transformations (6.2.1), the total derivative (6.1.6) reads
∂q 0 ∂q 0
!
0
deq0 = deq0 (q )dq0 = + e0k
q dq0 .
∂ qe0 ∂ qek
In accordance with the relation (6.1.7), we have
∂q 0 ∂q 0 ∂ qei ei
! !
j ∂q
qe0i 0
= deq0 (q )dq0 (qe ) = i
+ e0k
q + q 0 .
∂ qe0 ∂ qek ∂q 0 ∂q j
The solution of this equation is
∂ qei ei ∂ qe0 e0
! !
j ∂q k ∂q
qe0i = + q 0 / + q 0 . (6.2.3)
∂q 0 ∂q j ∂q 0 ∂q k
A glance at this transformation law shows that the velocity phase space J11 Z → Z
of relativistic mechanics is really a projective bundle.
Example 6.2.1. Let consider the configuration space Q = R4 , provided with the
Cartesian coordinates
(q 0 , q i ), i = 1, 2, 3.
This is the case of Special Relativity. For instance, let
qe0 = q 0 chα − q 1 shα,
qe1 = −q 0 shα + q 1 chα, (6.2.4)
2,3 2,3
qe =q .
be a Lorentz transformation of the plane (q 0 , q 1 ). Substituting these expressions in
the formula (6.2.3), we obtain
−shα + q01 chα
qe01 = ,
chα − q01 shα
q02,3
qe02,3 = .
chα − q01 shα
This is precisely the transformation law of 3-velocities in Special Relativity if we
put
1 v
chα = √ , shα = √ ,
1 − v2 1 − v2
6.2. RELATIVISTIC VELOCITY AND MOMENTUM PHASE SPACES 307
where v is the velocity of a reference frame moving along the axis q 1 . •
Thus, one can think of the velocity phase space J11 Q of relativistic mechanics
as the space of 3-velocities v of a relativistic system. For the sake of convenience,
we will call J11 Q the 3-velocity phase space, though the dimension of Q is not equal
necessarily to 3 + 1. Given a coordinate chart (U ; q 0 , q i ), 3-velocities v = (v0i ) of a
relativistic system can be seen as velocities of a local non-relativistic system (6.2.2)
with respect to the corresponding local reference frame Γ = ∂t on U . However, the
notion of a reference frame in non-relativistic mechanics has no meaning in relativis-
tic mechanics since the relativistic transformations q0i → q00i and Γi → Γi are not
affine and the relative velocity q0i −Γi is not maintained under these transformations.
To introduce the relativistic velocities, let us consider the tangent bundle T Q of
the configurations space Q. It is equipped with the holonomic coordinates
(q 0 , q i , q̇ 0 , q̇ i ).
In accordance with (6.1.8), there is the multivalued morphism λ

e ¿from the 3-velocity
phase space J11 Q to the tangent bundle T Q when a point (q 0 , q i , q0i ) ∈ J11 Q corre-
sponds to a line
e : q i 7→ (q̇ 0 , q̇ i = q̇ 0 q i ) ⊂ T Q
λ (6.2.5)
0 0
in the tangent space to Q at the point (q 0 , q i ). Conversely, there is the morphism
% : T Q → J11 Q,
Q
q̇ i
q0i ◦ % = , (6.2.6)
q̇ 0
such that
e = Id J 1 Q.
%◦λ 1
Indeed, it is readily observed that q0i and q̇ i /q̇ 0 have the same transformation laws.
It should be emphasized that, though the expressions (6.2.5) and (6.2.6) are singular
at q̇ 0 = 0, this point belongs to another coordinate chart, and the morphisms λ e and
% are well defined.
Thus, one can think of the tangent bundle T Q as being the space of relativistic
velocities or 4-velocities of a relativistic system. It is called the 4-velocity phase
space.
Remark 6.2.2. Note that, with respect to the universal unit system, the elements
of T Q are not physically dimensionless, in contrast with the physical relativistic
velocities. •
Since the morphism λ e of J 1 Q onto T Q is multivalued and the converse morphism

1
(6.2.6) is a surjection, one may try to find a subbundle W of the tangent bundle T Q
such that
% : W → JQ1
is an injection. Let us assume that Q is oriented and endowed with a pseudo-
Riemannian metric g with the signature (+, − . . . −). The pair (Q, g) is called a
relativistic system. This metric g defines the subbundle of velocity hyperboloids
Wg = {q̇ λ ∈ T Q : gµν (q)q̇ µ q̇ ν = 1}. (6.2.7)
Of course, Wg is neither vector nor affine subbundle of T Q. Let Q be time-oriented
with respect to the pseudo-Riemannian metric g. By definition, this means that
Wg is a disjoint union of two connected subbundles W + and W − [77, 155]. Then it
is readily observed that the restriction of the morphism % (6.2.6) to each of these
subbundles is an injection into J11 Q.
Let us consider the image of this injection in a fibre of J11 Q at a point q ∈ Q.
There are local coordinates (q 0 , q i ) on a neighbourhood of q ∈ Q such that the
pseudo-Riemannian metric g(q) at q comes to the Minkowski metric
g(q) = η = diag(1, −1, · · · , −1). (6.2.8)
With respect to these coordinates, the velocity hyperboloid Wq ⊂ Tq Q is given by
the equation
(q̇ 0 )2 − (q̇ i )2 = 1.
X
This is the union of the subset Wq+ , where q̇ 0 > 0, and Wq− , where q̇ 0 < 0. Restricted
to W + , the morphism (6.2.6) takes the familiar form of the relations between 3- and
4-velocities
1
q̇ 0 = q ,
1 − (q0i )2
P
q0i
q̇ i = q
1− (q0i )2
P
6.3. RELATIVISTIC DYNAMICS 309
in Special Relativity. The image of each of the hyperboloids W ± in the 3-velocity

phase space J11 Q by the morphism (6.2.6) is the open ball
(q0i )2 < 1.
X
(6.2.9)
i
This relation means that 3-velocities of a relativistic system (Q, g) are bounded in
accordance with the relativistic principle.
Let us turn now to the momentum phase space of relativistic mechanics. Given
a chart of the relativistic configuration space Q, the homogeneous Legendre bundle
corresponding to the local non-relativistic system (6.2.2) is the cotangent bundle
T ∗ Q. This fact motivates us to choose T ∗ Q as the relativistic momentum phase
space, equipped with the holonomic coordinates (q λ , pλ ). The cotangent bundle
T ∗ Q is endowed with the canonical symplectic form
Ω = dpµ ∧ dq µ , (6.2.10)
called the relativistic symplectic form.

Given a pseudo-Riemannian metric g on the relativistic configuration space Q,
we have the corresponding isomorphism between relativistic velocity and momen-
tum phase spaces T Q and T ∗ Q. ¿From the physical viewpoint, this isomorphism,
however, fails to be correct. With respect to the universal unit system, a metric ten-
sor g is dimensionless, whereas 4-velocities and relativistic momenta have different
physical dimension. The latter are of dimension [length]−1 . Therefore, we should
use a mass metric in order to perform the above-mentioned isomorphism.
6.3 Relativistic dynamics

In a straightforward manner, Lagrangian formalism fails to be appropriate to rela-
tivistic mechanics because a Lagrangian
L = Ldq 0 , (6.3.1)
by very definition, is defined only locally on a coordinate chart (6.1.5) of the 3-

velocity phase space J11 Q. For instance, the Lagrangian
s
(q0i )2 dq 0
X
Lm = −m 1 − (6.3.2)
i
of a free relativistic point mass m in Special Relativity, can be defined on each

coordinate chart of J11 Q, but it is not maintained in a straightforward manner by the
Lorentz transformations (6.2.4) because of the term dq 0 . Given a motion q i = ci (q 0 )
with respect to a coordinate chart (q 0 , q i ), its Lorentz transformation (6.2.4) reads
qe0 = q 0 chα − c1 (q 0 )shα,

ce1 (qe0 ) = −q 0 (qe0 )shα + c1 (q 0 (qe0 ))chα,
ce2,3 (qe0 ) = c2,3 (q 0 (qe0 )).
Then we have the Lorentz invariance of the pull-back c∗ Lm of the Lagrangian (6.3.2),
i.e.,
c∗ Lm = ce∗ L
e
m
where dqe0 = dq0 (qe0 )dq 0 and dq0 is the total derivative.
Therefore, we concentrate our consideration on Hamiltonian relativistic mechan-
ics.
Let L be a local regular Lagrangian (6.3.1) on a coordinate chart (q 0 , q i ) of the
3-velocity phase space J11 Q. In accordance with Remark 4.8.9, it defines a local
Dirac constraint system on the relativistic momentum phase space T ∗ Q in the case
of a zero Hamiltonian and the primary constraint space
∂L ∂L
p0 = L − q0i , pi = . (6.3.3)
∂q0i ∂q0i
Since the Lagrangian L is regular, the equations (6.3.3) are solvable for
p0 = p0 (q µ , pi ).
Then a solution of the Lagrange equation is an integral curve of the vector field
ϑ = ∂0 + ϑi ∂i + ϑi ∂ i
on the primary constraint space (6.3.3), which fulfills the equation
ϑcΩN = 0,
ϑi = ∂i p0 , ϑi = ∂ i p 0 ,
where ΩN is the restriction of the relativistic symplectic form (6.2.10) to the con-
straint space (6.3.3). For instance, the Lagrangian (6.3.2) (which is regular on the
6.3. RELATIVISTIC DYNAMICS 311
ball (6.2.9)) defines a Dirac constraint system on T ∗ Q in the case of a zero Hamil-
tonian and the primary constraint space
p20 − p2i = m2 .
X
(6.3.4)
i
Therefore, let us describe a relativistic system (Q, g) as an autonomous Hamil-

tonian system on the symplectic manifold T ∗ Q, characterized by a Hamiltonian
H : T ∗ Q → R, (6.3.5)
called a relativistic Hamiltonian.

Remark 6.3.1. It should be emphasized that the coordinate −p0 , but not the
relativistic Hamiltonian H, plays the role of a relativistic energy function. •
Any relativistic Hamiltonian H (6.3.5) defines the Hamiltonian map

c : T ∗ Q → T Q,
H
q̇ µ = ∂ µ H, (6.3.6)
(3.3.4) over Q from the relativistic momentum space T ∗ Q to the 4-velocity phase
space T Q. Since the 4-velocities of a relativistic system live in the velocity hyper-
boloids (6.2.7), we have the constraint subspace
c∗ W ,
N =H (6.3.7)
g
µ ν
gµν ∂ H∂ H = 1,
of the relativistic momentum phase space T ∗ Q. It follows that the relativistic system
(Q, g) can be described as an autonomous Dirac constraint system on the primary
constraint space N (6.3.7) [154]. Its solutions are integral curves of the Hamiltonian
vector field
u = ui ∂i + ui ∂ i
on N ⊂ T ∗ Q, that obeys the Hamilton equation
uci∗N Ω = −i∗N dH. (6.3.8)
Its solution does not necessarily exist.

Example 6.3.2. The relativistic Hamiltonian of a free relativistic point mass in

Special Relativity is
1 µν
H=− η pµ pν , (6.3.9)
2m
where η is the Minkowski metric (6.2.8), while the constraint space N (6.3.7) is given
by the equation (6.3.4). It is readily observed that the restriction of the Hamiltonian
(6.3.9) to this constraint space is a constant function, i.e.,
i∗N dH = 0.
Then the Hamilton equation (6.3.8) takes the form
uci∗N Ω = 0. (6.3.10)
We obtain its solution

u0 η ik pk
ṗi = ui = 0, q̇ i = ui = q
m2 − η jk pj pk
and then the familiar expression for 3-velocities
mη ik pk
qti = q , pi = const.
m2 − η jk pj pk
•
Example 6.3.3. Let us consider a point electric charge e in the Minkowski space in
the presence of an electromagnetic potential Aλ . Its relativistic Hamiltonian reads
1 µν
H=− η (pµ − eAµ )(pν − eAν ),
2m
while the constraint space N (6.3.7) is
η µν (pµ − eAµ )(pν − eAν ) = m2 .
As in the previous Example, the Hamilton equation (6.3.8) takes the form (6.3.10).
Its solution is
ṗk = uk = uµ ∂k Aµ , (6.3.11)
u0 η ik (pi − eAi )
q̇ k = uk = q . (6.3.12)
m2 − η ij (pi − eAi )(pj − eAj )
6.4. RELATIVISTIC GEODESIC EQUATIONS 313
The equality (6.3.12) leads to the usual expression for the 3-velocities
mηki qti
pk = q + Ak
1 + ηij qti qtj
Substituting this expression in the equality (6.3.11), we obtain the familiar equation
of motion of a relativistic charge in an electromagnetic field. •
Example 6.3.4. The relativistic Hamiltonian for a point mass m in a gravitational

field g on a 4-dimensional manifold Q reads
1 µν
H=− g (q)pµ pν ,
2m
while the constraint space N (6.3.7) is
g µν pµ pν = m2 .
As in previous Examples, the equation (6.3.8) takes the form (6.3.10). •
Examples (6.3.2) – (6.3.4) illustrate the relativistic systems where the restriction
of a relativistic Hamiltonian to the constraint space N (6.3.7) is a constant function.
In this case, the Hamilton equation takes the form (6.3.10). In accordance with
Proposition 3.4.2, this equation has a solution everywhere on the primary constraint
space N (6.3.7).
Remark 6.3.5. Given a symplectic manifold (T ∗ Q, Ω) and its submanifold N
(6.3.7), one may apply the reduction procedure described in Section 2.6. For the
relativistic systems in Examples 6.3.2 – 6.3.4, this procedure is not, however, use-
ful since the pull-back Hamiltonian i∗N H on the constraint space N is a constant
function and the associated Hamiltonian vector fields belong to the kernel of the
presymplectic form i∗N Ω on N . •
6.4 Relativistic geodesic equations

Let H be a relativistic Hamiltonian on the relativistic momentum phase space T ∗ Q
and
ξH = ∂ i H∂i − ∂i H∂ i (6.4.1)
the corresponding Hamiltonian vector field whose integral curves are solutions of
the Hamilton equation
ξH cΩ = −dH.
If the Hamiltonian map H c (6.3.6) is a diffeomorphism, the Hamiltonian vector field

(6.4.1) yields the holonomic vector field
ξ = TH c−1 = q̇ µ ∂ + ξ µ ∂˙ ,
c◦ξ ◦H (6.4.2)
H µ µ
c−1 ,
ξ µ = (∂α ∂ µ H∂ α H − ∂ α ∂ µ H∂α H) ◦ H
on the 4-velocity space T Q, where
(q µ , q̇ µ , q̇µ , q̈ µ ) ◦ T H
c = (q µ , ∂ µ H, q̇ µ , ∂ ∂ µ Hq̇ α + ∂ α ∂ µ Hṗ )
α α
is the tangent morphism to the Hamiltonian map H. c The holonomic vector field
(6.4.2) defines a conservative second order dynamic equation
c−1 ,
q̈ µ = (∂α ∂ µ H∂ α H − ∂ α ∂ µ H∂α H) ◦ H (6.4.3)
called the relativistic dynamic equation, on the relativistic configuration space Q

(see Definition 3.1.2).
For instance, the relativistic dynamic equation (6.4.3) for a point electric charge
in Example 6.3.3 takes the well-known form
q̈ µ = η µν q̇ λ Fνλ , (6.4.4)
where F is the electromagnetic strength (4.9.10). The relativistic dynamic equation

(6.4.3) for a point mass m in a gravitational field g in Example 6.3.4 reads
q̈ µ = {λ µ ν }q̇ λ q̇ ν , (6.4.5)
where
1
{λ µ ν } = − g µα (∂λ gαν + ∂ν gαλ − ∂α gλν )
2
are the Christoffel symbols of the pseudo-Riemannian metric g.
The equations (6.4.4) and (6.4.5) exemplify relativistic dynamic equations which
are geodesic equations
q̈ µ = Kλµ (q α , q̇ α )q̇ λ (6.4.6)

with respect to a connection
K = dq λ ⊗ (∂λ + Kλµ ∂˙µ ) (6.4.7)
on the tangent bundle T Q → Q (see Definition 3.1.6). We call (6.4.6) the relativistic
geodesic equation. For instance, a connection K in the equation (6.4.5), is the Levi–
Civita connection of the pseudo-Riemannian metric g. In the equation (6.4.4), K is
the zero Levi–Civita connection of the Minkowski metric plus the soldering form
σ = η µν Fνλ dq λ ⊗ ∂˙µ . (6.4.8)
We say that a relativistic geodesic equation (6.4.6) on the 4-velocity phase space
T Q describes a relativistic system (Q, g) if its geodesic vector field does not leave the
subbundle of velocity hyperboloids (6.2.7). It suffices to require that the condition
(∂λ gµν q̇ µ + 2gµν Kλµ )q̇ λ q̇ ν = 0 (6.4.9)
holds for all tangent vectors which belong to Wg (6.2.7). Obviously, the Levi–Civita
connection {λ µ ν } of the metric g fulfills the condition (6.4.9). Any connection K on
T Q → Q can be written as
Kλµ = {λ µ ν }q̇ ν + σλµ (q λ , q̇ λ ),
where
σ = σλµ dq λ ⊗ ∂˙λ
is a soldering form. Then the condition (6.4.9) takes the form
gµν σλµ q̇ λ q̇ ν = 0. (6.4.10)
It is readily observed that the soldering form (6.4.8) in the equation (6.4.4) obeys
this condition for the Minkowski metric η.
Let now compare relativistic and non-relativistic geodesic equations. In physical
applications, one usually thinks of non-relativistic mechanics as being an approxi-
mation of small velocities of relativistic theory. At the same time, the 3-velocities in
mathematical formalism of non-relativistic mechanics are not bounded. It has long
been recognized that the relation between the mathematical schemes of relativistic
and non-relativistic mechanics is not trivial.
Let a relativistic configuration space Q admit a fibration Q → R, where R is a

time axis. One can think of the fibre bundle Q → R as being a configuration space
of a non-relativistic mechanical system.
In order to compare relativistic and non-relativistic dynamics, one should con-
sider pseudo-Riemannian metric on T Q, compatible with the fibration Q → R.
Note that R is a time of non-relativistic mechanics. It is one and the same for all
non-relativistic observers. In the framework of a relativistic theory, this time can be
seen as a cosmological time. Given a fibration Q → R, a pseudo-Riemannian metric
on the tangent bundle T Q is said to be admissible if it is defined by a pair (g R , Γ)
of a Riemannian metric on Q and a non-relativistic reference frame Γ, i.e.,
2Γ ⊗ Γ
g= − gR, (6.4.11)
| Γ |2
| Γ |2 = gµν
R µ ν
Γ Γ = gµν Γµ Γν ,
in accordance with the well-known theorem [77]. The vector field Γ is a time-like
vector relative to the pseudo-Riemannian metric g (6.4.11), but not with respect to
other admissible pseudo-Riemannian metrics in general.
There is the canonical imbedding (4.1.4) of the velocity phase space J 1 Q of
non-relativistic mechanics into the affine subbundle
q̇ 0 = 1, q̇ i = q0i (6.4.12)
of the 4-velocity phase space T Q. Then one can think of (6.4.12) as the 4-velocities
of a non-relativistic system. The relation (6.4.12) differs from the familiar rela-
tion (6.2.6) between 4- and 3-velocities of a relativistic system. In particular, the
temporal component q̇ 0 of 4-velocities of a non-relativistic system equals 1 (rela-
tive to the universal unit system). It follows that the 4-velocities of relativistic and
non-relativistic systems occupy different subbundles of the 4-velocity space T Q.
Moreover, Theorem 4.4.2 shows that both relativistic and non-relativistic equations
of motion can be seen as the geodesic equations on the same tangent bundle T Q.
The difference between them lies in the fact that their solutions live in the different
subbundles (6.2.7) and (6.4.12) of T Q. At the same time, relativistic equations,
expressed in the 3-velocities q̇ i /q̇ 0 of a relativistic system, tend exactly to the non-
relativistic equations on the subbundle (6.4.12) when q̇ 0 → 1, g00 → 1, i.e., where
non-relativistic mechanics and the non-relativistic approximation of a relativistic
theory only coincide.
Let (q 0 , q i ) be a non-relativistic reference frame on Q compatible with the fibra-

tion of Q → R. Given a non-relativistic geodesic equation (4.4.7), we will say that
the relativistic geodesic equation (6.4.4) is the relativization of (4.4.7) if the spatial
parts of these equations are the same.
In accordance with Lemma (4.5.5), any relativistic geodesic equation with respect
to a connection K (6.4.7) is a relativization of a non-relativistic geodesic equation
with respect to the connection
f = dq λ ⊗ (∂ + (K i − Γi K 0 )∂ ),
K λ λ λ i
where Γi = 0 is the connection corresponding to the reference frame (q 0 , q i ). Of

course, for different reference frames, we have different non-relativistic limits of the
same relativistic equation. The converse procedure is more intricate.
Example 4.9.3 shows that the Lagrange equation for any non-relativistic quadratic
Lagrangian (4.9.11) admits a relativization with respect to the reference frame as-
sociated with a Lagrangian frame connection, where an inertial force is absent. It
follows that inertial forces do not admit relativization. Conversely, let us consider
a geodesic motion
q̈ µ = {λ µ ν }q̇ λ q̇ ν (6.4.13)
in the presence of a pseudo-Riemannian metric g on a relativistic configuration space
Q. Let (q 0 , q i ) be local hyperbolic coordinates such that g00 = 1, g0i = 0. These
coordinates make up a non-relativistic reference frame for a local fibration Q → R.
Then, by virtue of Lemma (4.5.5), the equation (6.4.13) has the non-relativistic limit
(4.9.18) which is the Lagrange equation for the Lagrangian (4.9.16) where φ0 = 0.
This Lagrangian describes a free non-relativistic mechanical system with the mass
tensor mij = −gij .
In view of Proposition 4.4.5, the relativization procedure, however, fails to be
unique. Therefore, the relativization (4.9.12) of an arbitrary quadratic Lagrangian
(4.9.11) may lead to confusion. In fact, this relativization procedure is appropriate
to a gravitational Lagrangian (4.9.15), where φ0 is a gravitational potential. An
arbitrary quadratic dynamic equation can be written in the form
i
q00 = −(m−1 )ik {λkµ }q0λ q0µ + biµ (q ν )q0µ , q00 = 1,
where {λkµ } are the Christoffel symbols of some pseudo-Riemannian metric g, whose
spatial part is the mass tensor (−mik ), while
bik (q µ )q0k + bi0 (q µ ) (6.4.14)
is an external force. With respect to the coordinates where g0i = 0, one may
construct the relativistic equation
q̈ µ = {λ µ ν }q̇ λ q̇ ν + σλµ q̇ λ , (6.4.15)
where the soldering form σ must fulfill the condition (6.4.10). It takes place only if
gik bij + gij bik = 0,
i.e., the external force (6.4.14) is the Lorentz-type force plus some potential one.
Then, we have
σ00 = 0, σk0 = −g 00 gkj bj0 , σkj = bjk .
The relativization (6.4.15) exhausts almost all familiar examples. To complete

our exposition, we will point out also another relativization procedure. Let a non-
relativistic acceleration ξ i (xµ ) be a spatial part of a 4-vector ξ λ in the Minkowski
space. Then one can write the relativistic equation
q̈ λ = ξ λ − ηαβ ξ β q̇ α q̇ λ . (6.4.16)
This is the case, e.g., for the relativistic hydrodynamics that we usually meet in
the literature on gravitation theory. However, the non-relativistic limit q̇ 0 = 1 of
(6.4.16) does not coincide with the initial non-relativistic equation. There are also
other variants of relativistic hydrodynamic equations [107].
Chapter 7
Appendix A. Geometry of BRST

mechanics
As is well-known, the BRST formalism is one of the corner stones of contem-

porary quantum field theory. This formalism has been extended to mechanics
[5, 65, 66, 140, 141]. One can think of BRST mechanics as being particular su-
persymmetric mechanics [3, 42, 108], where supercharges are the BRST charges, or
particular mechanics on supermanifolds [4, 33, 41, 95, 165]. We aim to show that
Hamiltonian time-dependent mechanics on a vertical configuration space V V ∗ Q ad-
mits the natural extension to BRST mechanics on graded manifolds.
We start from the naive BRST extension of Hamiltonian mechanics where the
odd variables are introduced in a formal way.
Let the configuration space Q → R be a vector bundle. Besides the familiar
coordinates (t, q i , pi , q̇ i , ṗi ) on the vertical configuration space V V ∗ Q, let us consider
additionally the odd variables
c i , ci , c i , c i (7.0.1)
which are assumed to be anticommutative. Following the physical terminology, we

will call c and c (7.0.1) the ghosts and antighosts, respectively. Let ci and ci with
respect to the indices i have the same transition functions as the linear coordinates
q i , while ci and ci have the transition functions of the coordinates pi , i.e.,
∂q 0i j ∂q j ∂q 0i j ∂q j
c0i = c, c0i = cj , c0i = c, c0i = cj . (7.0.2)
∂q j ∂q 0i ∂q j ∂q 0i
319
320 CHAPTER 7. APPENDIX A. GEOMETRY OF BRST MECHANICS
The formal products of the odd variables (7.0.1) make up the algebra of polyno-
mials over the ring of complex functions on V V ∗ Q. Provided with the unit element,
this is the Z2 -graded ring P(c, c) of polynomials of the odd variables. One may say
that such a polynomial is globally defined if it is invariant under the transformations
(7.0.2). We will follow the convention where the antighosts c are written on the left
of the ghosts c.
For the sake of convenience, we will use the compact notation y A for both even
and odd variables. Put {y} = 0 for the even variables and {y} = 1 for the odd ones.
Let us consider the left differentiations ∂/∂y of the graded ring P(c, c), which
possess the properties
∂ ∂ ∂
(φσ) = (φ)σ + (−1){φ}{y} φ (σ), (7.0.3)
∂y ∂y ∂y
∂ ∂ A B ∂ ∂
A B
= (−1){y }{y } B A . (7.0.4)
∂y ∂y ∂y ∂y
We introduce the exterior forms dy as the dual of the derivatives ∂/∂y such that
{dy} = {y}, (7.0.5)

∂
A
cdy B = δAB . (7.0.6)
∂y
One may define the exterior product of exterior forms dy by the rule
0
dy ∧ dy 0 = (−1){y}{y }+1 dy 0 ∧ dy. (7.0.7)
With this exterior product, the ring P(c, c) is extended to the (Z, Z2 )-bi-graded ring
of polynomials P(c, c, dy) which is defined as the algebra of exterior forms dy over
the ring P(c, c), where
0
ydy 0 = (−1){y}{y } (dy 0 )y,
φ ∧ σ = (−1)|φ||σ|+{φ}{σ} σ ∧ φ, φ, σ ∈ P(c, c, dy). (7.0.8)
Accordingly, the interior product (7.0.6) obeys the relation

∂ ∂ ∂
c(φ ∧ σ) = ( cφ) ∧ σ + (−1)|φ|+{φ}{y} φ ∧ ( cσ). (7.0.9)
∂y ∂y ∂y
The ring P(c, c, dy) is also provided with the exterior differential
def ∂
dy A
X
dφ = (φ), (7.0.10)
A ∂y A
321
where the derivation

∂
(φ), φ ∈ P(c, c, dy),
∂y
fulfills the conditions (7.0.3), (7.0.4) and
∂
(dy 0 ) = 0.
∂y
As follows from (7.0.4) and (7.0.7), the exterior differential (7.0.10) is a nilpotent
operator, i.e.,
d ◦ d = 0.
It is readily observed that, putting c = 0, c = 0, we restate the familiar exterior

algebra of even variables y.
The main principle of the BRST extension of mechanics is its invariance under
the BRST and anti-BRST transformations whose generators read
∂ ∂ ∂ ∂
ϑ = ci i
+ ci + iq̇ i i + iṗi , (7.0.11)
∂q ∂pi ∂c ∂ci
∂ ∂ ∂ ∂
ϑ = ci i + ci − iq̇ i i − iṗi (7.0.12)
∂q ∂pi ∂c ∂ci
[65, 66]. The generators (7.0.11) – (7.0.12) fulfill the nilpotency rule
ϑϑ = 0, ϑ ϑ = 0, ϑϑ + ϑϑ = 0,
i.e., for any function f (y), we have
ϑcd(ϑcdf ) = 0,
ϑcd(ϑcdf ) = 0,
ϑcd(ϑcdf ) + ϑcd(ϑcdf ) = 0.
Following the criterion of BRST and anti-BRST invariance, let us consider a

Hamiltonian form H (5.2.12) on the momentum phase space V ∗ Q and its vertical
extension HV (5.11.8) on the vertical momentum phase space V V ∗ Q. We aim to
find a 1-form HC ∈ P(c, c, dy) such that the sum
HS = HV + HC
is BRST-invariant, i.e., its Lie derivative along the vector field ϑ (7.0.11) vanishes:
Lϑ (HV + HC ) = ϑcd(HV + HC ) + d(ϑc(HV + HC )) = 0.
We obtain a desired Hamiltonian form
HS = ṗi dq i − q̇ i dpi + i(ci dci − ci dci ) − ∂V Hdt − HC dt, (7.0.13)

HC = i(cj ∂ j + cj ∂j )(cj ∂ j + cj ∂j )H,
called the BRST extension of the Hamiltonian form H. It is easily justified that
this form is also anti-BRST-invariant, i.e., its Lie derivative along the vector field ϑ
(7.0.12) also vanishes.
In particular, given a splitting of a Hamiltonian
H = pi Γi j q j + H
f (7.0.14)
with respect to some reference frame Γ on the configuration space Q → R, there is

the corresponding splitting of the BRST extended Hamiltonian
HS = ṗi Γi j q j − q̇ j (−pi Γi j ) + ci Γi j cj − cj (−Γi j ci ) + ∂V H

f+
i(cj ∂ j + cj ∂j )(cj ∂ j + cj ∂j )H
f
with respect to the connection Γe (5.2.8) on the momentum phase space V ∗ Q → R.

This splitting shows that the Hamiltonian form (7.0.13) is globally defined, and that
the connection is given on the odd variables (ci , ci ) as on the elements of the fibre
bundle V V ∗ Q → R.
We also can write the BRST extension of the canonical 3-form Ω (5.1.9) on the
momentum phase space V ∗ Q. This extension is deduced from the canonical 3-form
ΩV (5.11.1) on the vertical momentum phase space V V ∗ Q, and reads
ΩS = [dṗi ∧ dq i + dpi ∧ dq̇ i + i(dci ∧ dci − dci ∧ dci )] ∧ dt. (7.0.15)
The form (7.0.15) is globally defined. By construction, it is BRST- and anti-BRST-

invariant. The Hamiltonian connection
∂ ∂ ∂ ∂
γ = ∂t + γi ∂ i + γ i ∂i + γ̇i ∂˙ i + γ̇ i ∂˙i + g i + g i i + gi + gi i
∂ci ∂c ∂ci ∂c
323
for the Hamiltonian form HS (7.0.13) with respect to the canonical form ΩS satisfies
the Hamilton equations
γ i = qti = ∂ i H, (7.0.16)
γi = pti = −∂i H, (7.0.17)
γ̇ i = q̇ti = ∂ i HS , (7.0.18)
γ̇i = ṗti = −∂i HS , (7.0.19)
i
g = cit j j
= (cj ∂ + c ∂j )∂ H, i
(7.0.20)
j j
g i = cti = −(cj ∂ + c ∂j )∂i H, (7.0.21)
g i = cit = (cj ∂ j + cj ∂j )∂ i H, (7.0.22)
j j
gi = cti = −(cj ∂ + c ∂j )∂i H (7.0.23)
for the Hamiltonian form HS .
It is readily observed that the equations (7.0.16) – (7.0.17) are the Hamilton
equations for the initial Hamiltonian form H. To clarify the meaning of other
equations, let us note that, if y(t) is a solution of the Hamilton equations (7.0.16) –
(7.0.23), then the deviations
q i (t) + εεq̇ i (t) + εci (t) + εci (t),
pi (t) + εεṗi (t) + εci (t) + εci (t),
with the odd parameters ε and ε are solutions of the Hamilton equations for the
Hamiltonian form H extended to the case of even, but not necessarily real variables
q i , pi .
The following construction plays a prominent role for the BRST quantization
procedure. Let H f be a Hamiltonian function in some decomposition (7.0.14) of a
Hamiltonian H. Such a Hamiltonian function becomes an operator under quantiza-
tion. Let us consider the operators
Hβ = eβ H ◦ ϑ ◦ e−β H = ϑ − β(cj ∂ j + cj ∂j )H, β ∈ R,
e e f
−β H βH
Hβ = e ◦ϑ◦e = ϑ + β(cj ∂ j + cj ∂j )H,
e e f
called the BRST and anti-BRST charges, which act on functions f (y) by the law
ϑ(f ) = ϑcdf, H(f
f ) = Hf.
f
It easy to see that these operators are nilpotent, i.e.,

Hβ ◦ Hβ = 0, Hβ ◦ Hβ = 0. (7.0.24)
Let H
f be the BRST extension of the Hamiltonian function H,
S
f which is also treated
as an operator on functions f (y). Then we obtain the relations
Hβ ◦ H
f −H
S
f ◦ H = 0,
S β
Hβ ◦ H
f −H
S
f ◦ H = 0,
S β
iβ H
f =H ◦H +H ◦H .
S β β β β
These relations together with the relations (7.0.24) provide the operators Hβ , Hβ ,
and Hf with the structure of a Lie superalgebra [35], similar to what we have in
S
supersymmetric mechanics [42, 108].
Now let us restate the above construction in the strict mathematical terms. In
general, BRST mechanics can be developed as mechanics on supermanifolds [4, 33].
However, its naive variant shows that it suffices to consider the following particular
case of mechanics on supermanifolds.
• First of all, we should restrict our consideration to vector configuration bun-
dles Q → R in order to introduce the generators of BRST and anti-BRST
transformations.
• Secondly, the BRST extension HS of a Hamiltonian H is a polynomial with

respect to the odd variables. Therefore, we can narrow the class of superfunc-
tions under consideration.
Let us recall a few basic notions. A vector space Q is called a Z2 -graded vector
space or simply a graded vector space if there is the decomposition
Q = Q0 ⊕ Q1 ,
where Q0 is called its even subspace, while Q1 is said to be the odd subspace. An
element of a graded vector space is called homogeneous if it belongs to Q0 or Q1 .
The degree of a homogeneous element v ∈ Q is denoted by {v}. A graded vector
space is said to be (m, n)-dimensional if dim Q0 = m and dim Q1 = n.
By a graded commutative algebra B is meant an algebra which is a vector graded
space B = B0 ⊕ B1 such that
ar as = (−1)rs as ar ∈ Br+s , ar ∈ Br , as ∈ Bs , s, r = 0, 1.
A graded commutative Banach algebra is a graded commutative algebra if it is a
Banach algebra and the condition
ka0 + a1 k = ka0 k + ka1 k
325
is fulfilled.
Let V be a vector space, and
k
Λ = ∧V = R ⊕ ∧ V
k
its exterior algebra. This is a Z-graded commutative algebra provided with Z2 -

graded structure as follows:
2m
Λ0 = R ⊕ ∧ V,
m
2m+1
Λ1 = ⊕ ∧ V.
m
It is called the Grassman algebra. Given a basis {ci , i ∈ I} for the vector space V ,
the elements of the Grassman algebra take the form
ai1 ···ik ci1 · · · cik ,

X X
a= (7.0.25)
k (i1 ···ik )
where the sum is over all the collections of indices (i1 · · · ik ) such that no two of
them are the permutations of each other. For the sake of simplicity, we will omit
the symbol of the exterior product of elements of the Grassman algebra Λ. The
Grassman algebra becomes a graded commutative Banach algebra if its elements
(7.0.25) are endowed with the norm
X X
kak = | ai1 ···ik | .
k (i1 ···ik )
Remark 7.0.1. There is a different definition of a Grassman algebra [89], which is

equivalent to the above one only in the case of an infinite-dimensional vector space
V [40]. •
Let Q be a graded vector space and B a graded commutative algebra with a unit
element. If Q is both left and right B-modules such that
as vr = vr as ∈ Qs+r , v r ∈ Qr , as ∈ Bs ,
it is called a graded B-module. Each B-graded module is obviously a B0 -module.

Example 7.0.2. Every graded vector space Q can be extended to a graded B-
module which is its graded B-envelope
BQ = (BQ)0 ⊕ (BQ)1 = (B0 ⊗ Q0 ⊕ B1 ⊗ Q1 ) ⊕ (B1 ⊗ Q0 ⊕ B0 ⊗ Q1 ).

The graded envelope
B m+n = (B0m ⊕ B1n ) ⊕ (B1m ⊕ B0n )
of the (m, n)-dimensional graded vector space Qm,n is said to be a vector superspace.
Its B0 -submodule
B m,n = B0m ⊕ B1n
is also called a vector superspace. If B is a Banach algebra, the vector superspace

B m+n is a Banach space provided with the norm
kai ci k =
X
kai k, ai ∈ B,
i
where {ci } is a basis for Qm,n . •
The main ingredient in a theory of supermanifolds is the sheaf B of graded

commutative algebras on a manifold Z. Let (Z, B(U )) be its canonical presheaf
where by B(U ) is meant the graded commutative algebra of local sections of the
sheaf B over an open subset U ⊂ Z. If the sheaf B fulfills some conditions, called
the Rothstein axioms, it is said to be an R-supermanifold (we refer the reader to
[11, 40, 152] for details). The notion of an R-supermanifold includes Berezin’s graded
manifolds, the supermanifolds of A.Rogers and infinite-dimensional supermanifolds
described by A.Jadczyk and K.Pilch.
We restrict our consideration to the following class of R-supermanifolds [33]. Let
E → Z be a vector bundle with a typical fibre V , and E ∗ → Z the dual of E → Z.
We consider the Grassman fibre bundle
k
∧E ∗ = R ⊕(⊕ ∧ E ∗ ) (7.0.26)
Z k Z
over Z. Its typical fibre is the Grassman algebra Λ∗ = ∧V ∗ . The sections of the
Grassman fibre bundle make up the sheaf of Grassman algebras B on the manifold
Z, which belongs to the class of graded manifolds. Its sections, i.e., sections of the
fibre bundle (7.0.26) are called superfunctions. Given bundle coordinates (z λ , y i )
with respect to the bases {cj } in E → Z, superfunctions take the form
ai1 ···ik ci1 · · · cik ,

X X
f= (7.0.27)
k (i1 ···ik )
where {ci } are the dual bases in E ∗ .

327
Let us consider the sheaf der B of graded differentiations of the sheaf B. Let U
be an open subset of Z and B(U ) the restriction of the sheaf B to U . By a graded
differentiation of the sheaf B(U ) is meant its endomorphism ϑ such that
ϑ : B(U ) → B(U ),
ϑ(ab) = ϑ(a)b + (−1){ϑ}{a} aϑ(b), (7.0.28)
for the homogeneous elements ϑ ∈ der B and a, b ∈ B(U ). The graded differentia-
tions of B(U ) constitute the B-module der B(U), and the presheaf of such a B-module
generates the sheaf der B. One can show that der B is isomorphic to the sheaf of
sections of the fibre bundle
T E ⊗ ∧E → Z. (7.0.29)
E
On can think of its elements as being Λ-dependent vector fields, called vector su-
perfields, on the fibre bundle E. They read
∂ ∂
ϑ = aλ λ
+ ai i , (7.0.30)
∂z ∂c
where aλ , ai are superfunctions (7.0.27) and
∂ ∂
i
= i
∂c ∂y
are elements of the holonomic bases in T E. Due to the splitting V E ∼ = E × E,
these elements can be identified with the basis elements ci in E. It follows that the
operators ∂/∂ci act on superfunctions (7.0.27) by the law of the exterior product
of vectors and covectors. In accordance with the relation (7.0.28), they are odd.
The operators ∂/∂z λ are even, and act on the superfunctions (7.0.27) as familiar
derivatives.
The dual of the sheaf der B is the sheaf der∗ B generated by the linear morphisms
derB|U → B(U). (7.0.31)
This is the sheaf of sections of the fibre bundle
T ∗ E ⊗ ∧E → Z.
E
Its elements take the form
φ = aλ dz λ + ai dci , (7.0.32)
where aλ , ai are superfunctions (7.0.27), while dci = dy i are elements of the holo-
nomic bases in T ∗ E. One can think of these elements as being the Λ-dependent
1-forms on the fibre bundle E. Then the morphisms (7.0.31) can be treated as the
interior product of the vector fields (7.0.30) and the 1-forms (7.0.32).
One can introduce the exterior differential of superfunctions
d : B → der∗ B,
df (ϑ) = (−1){f }{ϑ} ϑ(f ) = (−1){f }{ϑ} ϑcdf,
which is extended to the exterior algebra of Λ-dependent exterior forms on E, called

exterior superforms. These superforms are sections of the fibre bundle
∧T ∗ E ⊗ ∧E → Z, (7.0.33)
E
where ∧T ∗ E is the Grassman fibre bundle induced by T ∗ E.

The vector superfields, seen as sections of the fibre bundle (7.0.29), and the
exterior superforms, represented by sections of the fibre bundle (7.0.33), are exactly
those mathematical objects which we have considered above on the naive level. They
satisfy the rules (7.0.3) – (7.0.10). To obtain the BRST extension of Hamiltonian
mechanics in terms of vector superfields and exterior superforms, one can put Z =
V V ∗ Q, while E is the vertical tangent bundle of the fibre bundle V V ∗ Q → R so
that P(c, c, dy) = der∗ B.
Chapter 8
Appendix B. On quantum
time-dependent mechanics
Here we are concerned only with a few aspects of quantizing time-dependent me-
chanics.
We will start from the following important remark. Let Q → R be a configura-
tion space of time-dependent mechanics, V ∗ Q its momentum phase space, and H
a Hamiltonian form on V ∗ Q. Let Γ be a (complete) reference frame together with
the adapted coordinates (t, q i ) on Q. Since Γi = 0 relative to these coordinates, the
energy function with respect to this reference frame coincides with the Hamiltonian
H in the decomposition (5.2.12) (see Section 5.7). We have the Hamilton equation
(5.2.20):
qti = ∂ i H, (8.0.1)
where one can think of q̇Γi = qti as being the relative velocities with respect to the
reference frame Γ. In the new reference frame coordinates (t, q 0i ), the Hamiltonian
H0 (t, q 0i , p0i ) = H(t, q j (t, q 0i , p0i ), pj (t, q 0i , p0i )) − pj (t, q 0i , p0i )∂t q j (t, q 0i , p0i ) =(8.0.2)
H(t, q j (t, q 0i , p0i ), pj (t, q 0i , p0i )) + p0k Γk (t, q 0i , p0i )
coincides with the energy function with respect to the new reference frame, while
the energy function with respect to initial reference frame is
f = H(t, q j (t, q 0i , p0 ), p (t, q 0i , p0 )) = H0 (t, q 0i , p0 ) − p0 Γk (t, q 0i , p0 )

H (8.0.3)
Γ i j i i k i
329
330CHAPTER 8. APPENDIX B. ON QUANTUM TIME-DEPENDENT MECHANICS
(see relation (4.5.2)). Relative to the coordinates (t, q 0i , p0i ), the Hamilton equation
(8.0.1) reads
∂H0 (t, q 0j , p0j )

qt0i = ,
∂p0i
where qt0i can be treated as the relative velocities with respect to the new reference
frame. At the same time, we can rewrite these equations as
f (t, q 0i , p0 )
∂HΓ i
q̇Γi = 0
,
∂pi
where q̇Γi are the relative velocities with respect to the initial reference frame. Thus,
we arrive at the following conclusion. Let we quantize a Hamiltonian system with
a Hamiltonian H, written relative to coordinates (t, q i ), and perform a coordinate
transformation. In new coordinates, we quantize the Hamiltonian system with re-
spect to the new reference frame if we use the Hamiltonian (8.0.2), and quantize
that with respect to the initial reference frame if we use the Hamiltonian function
(8.0.3). It follows that, in quantum mechanics, a passage from one set of coordi-
nates to another is not a technical procedure. This explains why most quantization
formulations lead to correct result only when a Cartesian coordinate system is used
[47].
Let us point out the two quantization schemes which are invariant under reference
frame transformations.
Let Q → R be a configuration space of time-dependent mechanics and V ∗ Q its
momentum phase space, provided with the canonical Poisson structure w (5.1.6).
Let H be a Hamiltonian form on V ∗ Q. Given a reference frame Γ together with the
adapted coordinates (t, q i ) on Q, the Hamiltonian connection γH (5.2.19) for H can
be represented as the sum
γH = ∂t + ϑH ,
where ϑH is the Hamiltonian vector field for the Hamiltonian H in the decompo-
sition (5.2.12) with respect to the above-mentioned canonical Poisson structure w.
It follows that, given a reference frame Γ, a time-dependent Hamiltonian system
with a Hamiltonian form H can be seen as a Poisson Hamiltonian system (w, H).
This Poisson Hamiltonian system can be quantized by the methods of geometric
331
quantization as follows (we refer the reader to [103, 167, 181, 188] for the geomet-
ric quantization technique). We restrict our consideration to the prequantization
procedure.
Let M be the typical fibre of the configuration space Q and T ∗ M the correspond-
ing typical fibre of V ∗ Q, provided with the canonical symplectic form Ω (2.4.2). This
form is exact. Therefore it belongs to the zero integral Chern class 0 ∈ H 2 (M, Z).
Then the bundle product
V ∗M × C → V ∗M
is the prequantization bundle over M , provided with the connection
A = dpj ⊗ ∂ j + dq j ⊗ (∂j − 2πipj ∂c ),
where c is a coordinate on C, such that its curvature is precisely the 2-form
R = dA = −2πiΩ. (8.0.4)
A complete reference frame Γ define a projection
πΓ : V ∗ Q → T ∗ M.
The pull-back of the canonical symplectic form Ω by this projection is the exact
2-form
πΓ∗ Ω = dpi ∧ dq i
such that
w = w] (πΓ∗ Ω).
Then the pull-back bundle
πΓ∗ (T ∗ M × C) = V ∗ Q × C → V ∗ Q (8.0.5)
is the prequantization bundle over V ∗ Q, provided with the connection
πΓ∗ A = dpj ⊗ ∂ j + dq i ⊗ (∂i − 2πipi ∂c )
such that its curvature obeys the relation similar to (8.0.4). It follows that the
Poisson algebra on the Poisson manifold (V ∗ Q, w) is quantized, i.e., any function
f ⊂ O0 (V ∗ Q) defines the operator
fb : s 7→ −∂j f ∂ j s + ∂ j f (∂j + 2πipj )s (8.0.6)

on sections s(t, q j , pj ) of the prequantization bundle (8.0.5). Since s are simply

complex functions on V ∗ Q, we can rewrite the operator (8.0.6) in the form
fb : s 7→ {f, s}V + 2πipj ∂ j f s. (8.0.7)
It is readily observed that this operator is invariant under holonomic transforma-
tions of the momentum phase space V ∗ Q. It follows that prequantization (8.0.7) is
independent of a reference frame. The result is obvious since the Poisson bivector w
(5.1.6) belongs to the zero element of the second LP-cohomology group HLP 2
(V ∗ Q, w)
(see Example 2.7.5).
Another quantization procedure invariant under holonomic time-dependent trans-
formations is based on the Schrödinger equation. Given a reference frame Γ together
with the adapted coordinates (t, q i ) on Q, let H(t, q j , pj ) be a Hamiltonian with re-
spect to this reference frame. Then the corresponding Schrödinger equation reads
∂Ψ ∂
ih̄ = H(t, q i , −ih̄ j )Ψ. (8.0.8)
∂t ∂q
Relative to another reference frame coordinates (t, q 0i ), the Schrödinger equation
∂Ψ ∂
ih̄ = H0 (t, q 0i , −ih̄ 0j )Ψ, (8.0.9)
∂t ∂q
where H0 is the Hamiltonian (8.0.2). It follows that the Schrödinger equations
(8.0.8) and (8.0.9) provide quantization with respect to different reference frames.
The following example shows that these quantizations are not equibvalent in general.
Example 8.0.3. Let us consider a 1-dimensional motion described by the Hamil-
tonian
p2
H= + U (q)
2m
on the momentum phase space R3 → R with respect to some inertial reference frame
(t, q, p). The corresponding Schrödinger equation is
2m
ψ 00 (q) + (E − U (q))ψ(q) = 0.
h̄2
Let us consider another reference frame given by a connection Γ(q i ). The energy
function with respect to the reference frame Γ reads
H
f = H − pΓ.
Γ
333
The corresponding Schrödinger equation is

2m 2im
ψ 00 (q) + 2 (E − U (q))ψ(q) − Γ(q)ψ 0 (q) = 0.
h̄ h̄
Let us substitute
im Z

ψ(q) = ρ(q) exp Γ(q)dq (8.0.10)
h̄
in this equation. We obtain
2m m 2

ρ00 (q) + 0
2 E − U (q) + ih̄Γ (q) + Γ (q) ρ(q) = 0. (8.0.11)
h̄ 2
For instance, let Γ = v = const. be an inertial reference frame. In this case, the
equation (8.0.11) takes the form
2m m

00
ρ (q) + 2 E + v 2 − U (q) ρ(q) = 0.
h̄ 2
It is readily observed that energy levels of the quantum system with respect to the
moving inertial reference frames are the shifts
m
E + v2
2
of those of the initial quantum system. Moreover, the transformation ψ → ρ (8.0.10)
is not an automorphism of the Hilbert space of states of the initial quantum system.
This means that quantum systems with respect to different inertial reference frames
are not equivalent. •
Bibliography
[1] M.Abbati, R.Cirelli, S. De Santis and E.Ruffini, The second Noether theorem
in the formalism of jet-bundles: Symmetries and degeneration, J. Geom.
Phys. 17 (1995) 321.
[2] R.Abraham and J.Marsden, Foundations of Mechanics, Second Edition (Ben-

jamin/Cummings Publ. Comp., London, 1978).
[3] A.Adrianov, F.Cannata, M.Ioffe and D.Nishnianidze, Matrix Hamiltonians:

SUSY approach to hidden symmetries, J. Phys. A 30 (1997) 5037.
[4] S.Albeverio and Shao-Ming Fei, BRST Structures and Symplectic Geometry
on a Class of Supermanifolds, Lett. Math. Phys. 33 (1995) 207.
[5] A.Aringazin, BRS and anti-BRS invariant states in path integral approach
to Hamiltonian and Birkoffian mechanics, Phys. Lett B 314 (1993) 333.
[6] V.Arnold, Mathematical Methods of Classical Mechanics (Springer, Berlin,

1978).
[7] V.Arnold, V.Kozlov and A.Neishtadt, Mathematical aspects of classical and

celestial mechanics, in Dynamical Systems III, (Springer, Berlin, 1988) p.1.
[8] J. de Azcárraga, A.Perelomov and J.Pérez Bueno, The Schouten–Nijenhuis

bracket, cohomology and generalized Poisson structures, J. Phys. A 29
(1996) 7993.
[9] J. de Azcárraga, J.Izquierdo and J. Pŕez Buenot, On the higher-order gen-

eralizations of Poisson structures, J. Phys. A 30 (1997) L607.
[10] F.Barone, R.Grassini and G.Mendella, A generalized Lagrange equation in

implicit form for non-conservative mechanics, J. Phys. A 30 (1997) 1575.
335
336 BIBLIOGRAPHY
[11] C.Bartocci, U.Bruzzo, D.Hernández Ruipérez, V.Pestov, Foundations of su-

permanifold theory: the axiomatic approach, Diff. Geom. and Appl. 3 (1993)
135.
[12] M.Bergvelt and E. de Kerf, The Hamiltonian structure of Yang–Mills theories

and instantons, Physica 139A (1986) 101.
[13] K.Bhaskara and K.Vismanath, Poisson Algebras and Poisson Manifolds,

Pithman Research Notes in Mathematics 174 (Longhman Sci., Harlow,
1988).
[14] E.Binz, H.Fischer and J.Śniatycki, Geometry of Classical Fields (North-

Holland, Amsterdam, 1988).
[15] D.Blair, Contact Manifolds in Riemannian Geometry, Lect. Notes in Math-

ematics 509 (Springer, Berlin, 1976).
[16] G. Bredon, Sheaf theory (McGraw-Hill, N.-Y., 1967).
[17] K.Brown, Cohomology of Groups (Springer-Verlag, Berlin, 1982).
[18] A.D.Bruno, The normal form of a Hamilton system, Uspehi Matemat. Nauk
43 (1988) N1, 23 (in Russian).
[19] J-L. Brylinski, A differential complex for Poisson manifolds, J. Diff. Geom.
28 (1988) 93.
[20] R.Bryant, S.Chern, R.Gardner, H.Goldschmidt, P.Griffiths, Exterior Differ-

ential Systems (Springer, Berlin, 1991).
[21] A.Cabras and A.Vinogradov, Extension of the Poisson bracket to differential

forms and multi-vectors, J. Geom. Phys. 9 (1992) 75.
[22] S.Campbell, An Introduction to Differential Equations and their Applica-

tions (Wadsworth Inc., Belmont, 1990).
[23] D.Canarutto, Bundle splittings, connections and locally principle fibred

manifolds, Bull. U.M.I. Algebra e Geometria Serie VI V-D (1986) 18.
BIBLIOGRAPHY 337
[24] F.Cardin and G.Zanzotto,

On constrained mechanical systems: D’Alembert’s and Gauss’ princoples,
J. Math. Phys. 30 (1989) 1473.
[25] F.Cardin and M.Favretti, On nonholonomic and vakonomic dynamics of me-

chanical systems with nonintegrable constraints, J. Geom. Phys. 18 (1996)
295.
[26] J.Cariñena, J.Gomis, L.Ibort and N.Román, Canonical transformation the-

ory for presymplectic systems, J. Math. Phys. 26 (1985) 1961.
[27] J.Cariñena and M.Rañada, Poisson maps and canonical transformations for
time-dependent Hamiltonian systems, J. Math.Phys. 30 (1989) 2258.
[28] J.Cariñena, M.Crampin and L.Ibort, On the multisymplectic formalism for

first order field theories, Diff. Geom. and Appl. 1 (1991) 345.
[29] J.Cariñena, J.Fernández-Núñez and E.Martı́nez, A geometric approach to

Noether’s second theorem in time-dependent Lagrangian mechanics, Lett.
Math. Phys. 23 (1991) 51.
[30] J.Cariñena and M.Rañada, Lagrangian systems with constraints. A geomet-

ric approach to the method of Lagrange multipliers, J. Phys. A 26 (1993)
1335.
[31] J.Cariñena and J.Fernández-Núñez, Geometric theory of time-dependent sin-

gular Lagrangians, Fortschr. Phys. 41 (1993) 517.
[32] J.Cariñena, L.Ibort, G.Marmo and A.Stern, The Feynman problem and the
inverse problem for Poisson dynamics, Phys. Rep. 263 (1995) 153.
[33] J.Cariñena and H.Figueroa, Hamiltonian versus Lagrangian formulations of

supermechanics, J. Phys. A 30 (1997) 2705.
[34] G.Caviglia, Helmholtz conditions, covariance, and invariance identities, Int.

J. Theor. Phys. 24 (1985) 377.
[35] S.Cecotti and C.Vafa, Topological anti-topological fusion, Nucl. Phys. B367
(1991) 359.
338 BIBLIOGRAPHY
[36] P.Chernoff and J.Marsden, Properties of Infinite Dimensional Hamiltonian

Systems, Lect. Notes in Mathematics 425 (Springer, Berlin, 1974).
[37] D.Chinea, M.de León, and J.Marrero, The constraint algoritm for time-
dependent Lagrangians, J. Math. Phys. 35 (1994) 3410.
[38] D.Chinea, M.de León, and J.Marrero, The canonical double complex for
Jacobi manifolds, C. R. Acad. Sci., Paris I 323 (1996) 637.
[39] N.Chitaia, S.Goglidze and Yu.Surovtsev, Dynamical systems with first- and
second-class constraints, Phys. Rev. D 56 (1997) 1135; 1142.
[40] R.Cianci, Introduction to Supermaniifolds (Bibliopolis, Naples, 1990).
[41] R.Cianci, M.Francaviglia and I.Volovich, Variational calculus and Poincaré–

Cartan formalism on supermanifolds, J. Phys. A 28 (1995) 723.
[42] F.Cooper, A.Khare and U.Sukhatme, Supersymmetry and quantum mechan-

ics, Phys. Rep. 251 (1995) 267.
[43] M.Crampin, F.Cantrijn and W.Sarlet, Lifting geometric objects to a cotan-

gent bundle, and the geometry of the cotangent bundle of a tangent bundle,
J. Geom. Phys. 4 (1987) 469.
[44] P.Dazord, A.Lichnerowicz and C-M.Marle, Structure locale des variétés de

Jacobi, J. Math. Pures et Appl. 70 (1991) 101.
[45] P.Dedecker, On the generalization of symplectic geometry to multiple in-

tegrals in the calculus of variations, in Differential Geometric Methods in
Mathematical Physics, Lect. Notes in Mathematics 570 (Springer, Berlin,
1977), p. 395.
[46] P.Dedecker and W.Tulczyjew, Spectral sequences and the inverse problem
of the calculus of variations, in Differential Geometric Methods in Mathe-
matical Physics, Lect. Notes in Mathematics 836 (Springer, Berlin, 1980),
p. 498.
[47] P.A.M.Dirac, The Principles of Quantum Mechanics (Oxford Univ. Press,

Oxford, 1976).
BIBLIOGRAPHY 339
[48] W.Domitrz and S.Janeczko, Normal forms of symplectic structures on the

stratified spaces, Colloquium Mathematicum LXVIII (1995) fasc.1, 101.
[49] B.Dubrovin, M.Giordano, G.Marmo and A.Simoni, Poisson brackets on

presymplectic manifolds, Int. J. Mod. Phys. 8 (1993) 3747.
[50] A.Echeverrı́a Enrı́quez, M.Muñoz Lecanda and N.Román Roy, Geometrical

setting of time-dependent regular systems. Alternative models, Rev. Math.
Phys. 3 (1991) 301.
[51] A.Echeverrı́a Enrı́quez, M.Muñoz Lecanda and N.Román Roy, Non-standard

connections in classical mechanics, J. Phys. A 28 (1995) 5553.
[52] F.Fatibene, M.Ferraris and M.Francaviglia, Noether formalism for conserved

quantities in classical gauge field theories, J. Math. Phys. 35 (1994) 1644.
[53] R.Fulp, J.Lawson and L.Norris, Generalized symplectic geometry as a cov-

ering theory for the Hamiltonian theories of classical particles and fields, J.
Geom. Phys. 20 (1996) 195.
[54] G.Giachetta and L.Mangiarotti, Gauge-invariant and covariant operators in

gauge theories, Int. J. Theor. Phys. 29 (1990) 789.
[55] G.Giachetta, Jet manifolds in non-holonomic mechanics, J. Math. Phys. 33

(1992) 1652.
[56] G.Giachetta and L.Mangiarotti, Constrained Hamiltonian systems and

gauge theories, Int. J. Theor.Phys. 34 (1995) 2353.
[57] G.Giachetta, L.Mangiarotti and G.Sardanashvily, New Lagrangian and Ha-

miltonian Methods in Field Theory (World Scientific, Singapore, 1997).
[58] D.Gitman and I.Tyutin, Canonical Quantization of Constrained Fields

(Nauka, M., 1986) (in Russian).
[59] A.Gomberoff and S.Hojman, Non-standard construction of Hamiltonian

structures, J.Phys. A 30 (1997) 5077.
[60] M.Gotay, J.Nester and G.Hinds, Presymplectic manifolds and the Dirac–
Bergman theory of constraints, J. Math.Phys. 19 (1978) 2388.
340 BIBLIOGRAPHY
[61] M.Gotay and J.Nester, Presymplectic Lagrangian systems, Ann. l’Inst. Henri
Poincaré 30 (1979) 129; 32 (1980) 1.
[62] M.Gotay, On coisotropic imbeddings of presymplectic manifolds, Proc.

Amer. Math. Soc. 84 (1982) 111.
[63] M.Gotay, A multisymplectic framework for classical field theory and the cal-
culus of variations. I. Covariant Hamiltonian formalism, in Mechanics, Anal-
ysis and Geometry: 200 Years after Lagrange, ed. M.Francaviglia (Elsevier
Science Publishers B.V., 1991), p. 203.
[64] M.Gotay, A multisymplectic framework for classical field theory and the
calculus of variations. II. Space + time decomposition, Diff. Geom. and Appl.
1 (1991) 375.
[65] E.Gozzi, M.Reuter and W.Thacker, Hidden BRS invariance in classical me-
chanics, Phys. Rev. D40 (1989) 3363.
[66] E.Gozzi, M.Reuter and W.Thacker, Symmetries of the classical path integral
on a generalized phase-space manifold, Phys. Rev. D46 (1992) 757.
[67] J.Grabowski and P.Urbański, Tangent lifts of Poisson and related structures,
J. Phys. A 28 (1995) 6743.
[68] P.Griffiths, Exterior Differential Systems and the Calculus of Variations

(Birkhäuser, Boston, 1983).
[69] D.Grigore and O.Popp, The complete classification of generalized homoge-

neous symplectic manifolds, J. Math. Phys. 30 (1989) 2476.
[70] D.Grigore, A generalized Lagrangian formalism in particle mechanics and

classical field theory, Fortschr. d. Phys. 41 (1993) 569.
[71] F.Guédira and A.Lichnerowicz, Géométrie des algèbres de Lie locales de

Kirillov, J. Math. Pures et Appl. 63 (1984) 407.
[72] V.Guillemin, S.Sternberg, Symplectic Techniques in Physics (Cambridge

Univ. Press., Cambridge, 1990).
BIBLIOGRAPHY 341
[73] C. Günther, The polysimplectic Hamiltonian formalism in field theory and

calculus of variations, J. Diff. Geom. 25 (1987) 23.
[74] A.Hamoui and A.Lichnerowicz, Geometry of dynamical systems with time-

dependent constraints and time-dependent Hamiltonians: An approach to-
wards quantization, J. Math. Phys. 25 (1984) 923.
[75] J.Harlet, Time and time functions in parametrized non-relativistic quantum

mechanics, Class. Quant. Grav. 13 (1996) 361.
[76] P.Havas, The connection between conservation laws and invariance groups:
Folklore, fiction and fact, Acta Phys. Austr. 38 (1973) 145.
[77] S.Hawking and G.Ellis, The Large Scale Structure of a Space-Time (Cambr.
Univ. Press, Cambridge, 1973).
[78] M.Henneaux, Equation of motion, commutation relations and ambiguities in

the Lagrangian formalism, Ann. Phys. 140 (1982) 1.
[79] J.Hietarinta, Nambu tensors and commuting vector fields, J. Phys. A 30

(1997) L27.
[80] F.Hirzebruch, Topological Methods in Algebraic Geometry (Springer, Berlin,

1966).
[81] S.Hojman and L.Urrutia, On the inverse problem of the calculus of varia-
tions, J. Math. Phys. 22 (1981) 1896.
[82] S.Hojman, The construction of a Poisson structure out of a symmetry and

a conservation law of a dynamical system, J. Phys. A 29 (1996) 667.
[83] R.Ibáñez, M.de León, J.Marrero and D.Martı́n de Diego, Co-isotropic and
Legendre-Lagrangian submanifolds and conformal Jacobi morphisms, J.
Phys. A 30 (1997) 5427.
[84] R.Ibáñez, M.de León, J.Marrero and D.Martı́n de Diego, Dynamics of Pois-
son and Nambu–Poisson brackets, J. Math. Phys. 38 (1997) 2332.
[85] R.Ibáñez, M.de León, J.Marrero and E.Padrón, Nambu–Jacobi and general-
ized Jacobi manifolds, J. Phys. A 31 (1998) 1267.
342 BIBLIOGRAPHY
[86] L.Ibort and J.Marı́n-Solano, A geometric classification of Lagrangian func-

tions and the reduction of evolution space, J. Phys. A 25 (1992) 3353.
[87] L.Ibort, M. de León and G.Marmo, Reduction of Jacobi manifolds, J. Phys.

A 30 (1997) 2783.
[88] A.Ibort, M. de León, E.Lacomba, J.Marrero, D.Martı́n de Diego and

P.Pitanga, Geometric formulation of mechanical systems subjected to time-
dependent one-sided constraints, J. Phys. A 31 (1998) 2655.
[89] A.Jadczyk and K.Pilch, Superspaces and Supersymmetries, Commun. Math.

Phys. 78 (1981) 391.
[90] F.Kamber and P.Tondeur, Foliated Bundles and Characteristic Classes, Lect.
Notes in Mathematics 493 (Springer, Berlin, 1975).
[91] I.Kanatchikov, On field theoretical generalization of a Poisson algebra, Rep.

Math. Phys. 40 (1997) 225.
[92] I.Kanatchikov, Towards to Born–Weyl quantization of fields, Int. J. Theor.

Phys. 37 (1998) 333.
[93] G.Katzin and J.Levine, Characteristic functional structures of infinitesimal

symmetry mappings of dynamical systems. I. Velocity-dependent mappings
of second-order differential equations, J. Math. Phys. 26 (1985) 3080.
[94] G.Katzin and J.Levine, Characteristic functional structures of infinites-

imal symmetry mappings of dynamical systems. IV. Classical (velocity-
independent) mappings of second-order differential equations, J. Math. Phys.
30 (1988) 2039.
[95] O.Khudaverdian and A.Nersessian, Canonical Poisson brackets of different

gradings and strange superalgebras, J. Math. Phys. 32 (1991) 1938.
[96] J.Kijowski and W.Tulczyjew, A Symplectic Framework for Field Theories

(Springer, Berlin, 1979).
[97] T.Kimura, Generalized classical BRST cohomology and reduction of Poisson

manifolds, Commun. Math. Phys. 151 (1993) 155.
BIBLIOGRAPHY 343
[98] A.Kirillov, Local Lie algebras, Russian Math. Surveys 31 (1976) 55.
[99] A.Kirillov, Geometric quantization, in Dynamical Systems IV, eds

V.I.Arnol’d and S.P.Novikov (Springer, Berlin, 1990) p.137.
[100] S.Kobayashi and K.Nomizu, Foundations of Differential Geometry, Vol.1.

(John Wiley, N.Y. - Singapore, 1963).
[101] I.Kolář, P.Michor and J.Slovák, Natural Operations in Differential Geometry

[102] Y.Kosmann-Schwarzbach and F.Magri, Poisson–Nijenhuis structures, Ann.

l’Inst. Henri Poincaré 53 (1990) 35.
[103] B.Kostant, Quantization and unitary representation, in Lectures in Modern

Analysis and Applications III, Lect. Notes in Mathematics, 170 (Springer,
Berlin, 1970), p.87.
[104] I.Krasil’shchik, V.Lychagin and A.Vinogradov, Geometry of Jet Spaces and

Nonlinear Partial Differential Equations (Gordon and Breach, Glasgow,
1985).
[105] D.Krupka, Some geometric aspects of variational problems in fibred mani-

folds, Folia Fac. Sci. Nat. UJEP Brunensis 14 (1973) 1.
[106] O.Krupkova, Mechanical systems with nonholonomic constraints, J. Math.

Phys. 38 (1997) 5098.
[107] B.Kupershmidt, The Variational Principles of Dynamics (World Scientific,

Singapore, 1992).
[108] A.Lahiri, P.K.Roy and B.Bagchi, Supersymmetry in quantum mechanics,

Int. J. Mod. Phys. A 5 (1990) 1383.
[109] S.Lang, Differential Manifolds (Addison–Weslay, Reading, Massachusetts,

1972).
[110] M.de León and P.Rodrigues, Methods of Differential Geometry in Analytical

Mechanics (North-Holland, Amsterdam, 1989).
344 BIBLIOGRAPHY
[111] M.de León and J. Marrero, Constrained time-dependent Lagrangian systems

and Lagrangian submanifolds, J. Math. Phys. 34 (1993) 622.
[112] M.de León and D. Martı́n de Diego, On the geometry of non-holonomic

Lagrangian systems, J. Math. Phys. 37 (1996) 3389.
[113] M.de León, J. Marrero and D. Martı́n de Diego, Mechanical systems with
nonlinear constraints, Int. J. Theor. Phys. 36 (1997) 979.
[114] M.de León, J. Marrero and D. Martı́n de Diego, Non-holonomic Lagrangian

systems in jet manifolds, J. Phys. A 30 (1997) 1167.
[115] M.de León, J. Marrero and E.Padrón, Lichnerowicz–Jacobi cohomology, J.

Phys. A 30 (1997) 6029.
[116] P.Libermann and C-M.Marle, Symplectic Geometry and Analitical Mechan-

ics (D.Reidel Publishing Company, Dordrecht, 1987).
[117] A.Lichnerowicz, Les variétés de Poisson et leurs algèbres de Lie associées, J.

Diff. Geom. 12 (1977) 253.
[118] A.Lichnerowicz, Les variétés de Jacobi et leurs algèbres de Lie associées, J.

Math. Pures et Appl. 57 (1978) 453.
[119] R.Littlejohn and M.Reinsch, Internal or shape coordinates in the n-body

problem, Phys. Rev. A 52 (1995) 2035.
[120] R.Littlejohn and M.Reinsch, Gauge fields in the separation of rotations and
internal motions in the n-body problem, Rev. Mod. Phys. 69 (1997) 213.
[121] L.Lusanna, An enlarged phase space for finite-dimensional constrained sys-

tems, unifying their Lagrangian, phase- and velocity-space descriptions,
Phys. Rep. 185 (1990) 1.
[122] S.Mac Lane, Homology (Springer, Berlin, 1967).
[123] C.-M.Marle, On Jacobi manifolds and Jacobi bundles, in Symplectic Ge-

ometry, Groupoids, and Integrable Systems, ed. P.Dazord and A.Weinstein
(Springer, Berlin, 1989), p. 227.
BIBLIOGRAPHY 345
[124] C.-M.Marle, Reduction of constrained mechanical systems and stability of

relative equilibria, Commun. Math. Phys. 174 (1995) 295.
[125] C.-M.Marle, The Schouten–Nijenhuis bracket and interior products, J.

Geom. Phys. 23 (1997) 350.
[126] G.Marmo, E.Saletan, A.Simoni and B.Vitale, Dynamical Systems. A Differ-

ential Geometric Approach to Symmetry and Reduction (John Wiley, N.Y.,
1985).
[127] G.Marmo, G.Mendella and W.Tulczyjew, Constrained Hamiltonian systems

as implicit differential equations, J. Phys. A 30 (1997) 277.
[128] G.Marmo, G.Vilasi and A.Vinogradov, The local structure of n-Poisson and
n-Jacobi manifolds and some applications, J. Geom. Phys (1998) (appear).
[129] J.Marsden and T.Ratiu, Reduction of Poisson manifolds, Lett. Math. Phys.
11 (1986) 161.
[130] J.Marsden and T.Ratiu, Introduction to Mechanics and Symmetries

[131] G.Martin, A Darboux theorem for multi-symplectic manifolds, Lett. Math.

Phys. 16 (1988) 133.
[132] E.Massa and E.Pagani, Jet bundle geometry, dynamical connections and
the inverse problem of Lagrangian mechanics, Ann. Inst. Henri Poincaré 61
(1994) 17.
[133] P.Michor, A generalization of Hamiltonian mechanics, J. Geom. Phys. 2

(1985) N2, 67.
[134] M.Modugno, A.Vinogradov, Some variations on the notion of connections,

Ann. Matem. Pura ed Appl. CLXVII (1994) 33.
[135] R. Montgomery, The connection whose holonomy is the classical adiabatic

angles of Hannay and Berry and its generalization to the non-integrable case,
Commun. Math. Phys. 120 (1988) 269.
346 BIBLIOGRAPHY
[136] G.Morandi, C.Ferrario, G.Lo Vecchio, G.Marmo and C.Rubano, The inverse
problem in the calculus of variations and the geometry of the tangent bundle,
Phys. Rep. 188 (1990) 147.
[137] M.Muñoz-Lecanda, Hamiltonian systems with constraints: A geometric ap-

proach, Int. J. Theor. Phys. 28 (1989) 1405.
[138] M.Muñoz-Lecanda and N.Román-Roy, Lagrangian theory for presymplectic

systems, Ann. Inst. Henrı́ Poincaré 57 (1992) 27.
[139] M.Muñoz-Lecanda and N.Román-Roy, Gauge systems: Presymplectic and

group action theory, Int. J. Theor. Phys. 32 (1993) 2077.
[140] Kh.Nirov and A.Razumov, Equivalence between Lagrangian and Hamilto-

nian BRST formalisms, J. Math. Phys. 34 (1993) 3933.
[141] Kh.Nirov and A.Razumov, Generalized Schrödinger representation in BRST

quantization, Nucl. Phys. B429 (1994) 389.
[142] L.Norris, Schouten–Nijenhuis brackets, J. Math. Phys. 38 (1997) 2694.
[143] F.Pardo, The Helmholtz conditions in terms of constants of motion in clas-

sical mechanics, J. Math. Phys. 30 (1989) 2054.
[144] P.Pereshogin and P.Pronin, Geometrical treatment of nonholonomic phase

in quantum mechanics and applications, Int. J. Theor. Phys. 32 (1993) 219.
[145] J.Pérez Bueno, Generalized Jacobi structures, J. Phys. A 30 (1997) 6509.
[146] V.Perlick, The Hamiltonization problem from a global viewpoint, J. Math.

Phys. 33 (1992) 599.
[147] G.Pidello and W.Tulczyjew, Derivations of differential forms on jet bundles,

Ann. Mat. Pura Appl. 147 (1987) 249.
[148] R.De Pietri, L.Lusanna and M.Pauri, Standard and generalized Newtonian
gravities as ”gauge” theories of the extended Galilei group: I. The standard
theory, Class. Quant. Grav. 12 (1995) 219.
[149] J.Pommaret, Systems of Partial Differential Equations and Lie Pseudogroups

(Gordon and Breach, Glasgow, 1978).
BIBLIOGRAPHY 347
[150] B.Reinhart, Differential Geometry and Foliations (Springer, Berlin, 1983).
[151] F.Riewe, Nonconservative Lagrangian and Hamiltonian mechanics, Phys.

Rev. E 53 (1996) 1890.
[152] M.Rothstein, The axioms of supermanifolds and a new structure arising from
them, Trans. Amer. Math. Society 297 (1986) 159.
[153] C.Rovelli, Quantum mechanics without time: A model, Phys. Rev. D42
(1990) 2638.
[154] C.Rovelli, Time in quantum gravity: An hypothesis, Phys. Rev. D43 (1991)
442.
[155] R.Sachs and H.Wu, General Relativity for Mathematicians (Springer, Berlin,
1977).
[156] R.Santilli, Foundations of Theoretical Mechanics I (Springer, N.Y., 1978).
[157] G.Sardanashvily, Gauge Theory in Jet Manifolds (Hadronic Press, Palm

Harbor, 1993).
[158] G.Sardanashvily, Constraint field systems in multimomentum canonical vari-

ables, J. Math. Phys. 35 (1994) 6584.
[159] G.Sardanashvily, Generalized Hamiltonian Formalism for Field Theory. Con-

straint Systems. (World Scientific, Singapore, 1995).
[160] G.Sardanashvily, Stress-energy-momentum tensors in constraint field theo-

ries, J. Math. Phys. 38 (1997) 847.
[161] G.Sardanashvily, Hamiltonian time-dependent mechanics, J. Math. Phys. 39

(1998) 2714.
[162] W.Sarlet, F.Cantrijn and D.Saunders, A geometric famework for the study
of non-holonomic Lagrangian systems, J. Phys. A 28 (1995) 3253.
[163] W.Sarlet, F.Cantrijn and D.Saunders, A geometric famework for the study
of non-holonomic Lagrangian systems: II, J. Phys. A 29 (1996) 4265.
348 BIBLIOGRAPHY
[164] D.Saunders, The Geometry of Jet Bundles (Cambr. Univ. Press, Cambridge,
1989).
[165] A.Schwarz, Geometry of Batalin–Vilkovisky quantization, Commun. Math.

Phys. 155 (1993) 249.
[166] J.Slawianowski, Geometry of Phase Spaces (Wiley, N.Y.,1991).
[167] J.Śniatycki, Geometric Quantization and Quantum Mechanics (Springer,

Berlin, 1980)
[168] J.Souriau, Structures des Systemes Dynamiques (Dunod, Paris, 1970).
[169] J.Stasheff, Homological reduction of constrained Poisson algebras, J. Diff.

Geom 45 (1997) 221.
[170] N.Steenrod, The Topology of Fibre Bundles (Princeton Univ. Press, Prince-
ton, 1972).
[171] P.Stefan, Accessible sets, orbits and foliations with singularities, Proc. Lon-
don Math. Soc. 29 (1974) 699.
[172] K.Sundermeyer, Constrained Dynamics (Springer, Berlin, 1982).
[173] H.Sussmann, Orbits of families of vector fields and integrability of distribu-

tions, Trans. Amer. Math. Soc. 180 (1973) 171.
[174] F.Takens, Symmetries, conservation laws and variational principles, in Ge-

ometry and Topology, Lect. Notes in Mathematics 597 (Springer-Verlag,
Berlin, 1977), p. 581.
[175] L.Takhtajan, On foundations of the generalized Nambu mechanics, Com-

mun. Math. Phys. 160 (1994) 295.
[176] W.Tulczyjew, Lagrangian submanifolds and Hamiltonian dynamics, C. R.

Acad. Sci. Paris A 283 (1976) 15.
[177] W.Tulczyjew, Lagrangian submanifolds and Lagrangian dynamics, C. R.

Acad. Sci. Paris A 283 (1976) 675.
BIBLIOGRAPHY 349
[178] W.Tulczyjew, The Legendre transformation, Ann. Inst. Henri Poincaré A

XXVII (1977) 101.
[179] W.Tulczyjew, The Euler–Lagrange resolution, in Differential Geomet-

ric Methods in Mathematical Physics, Lect. Notes in Mathematics 836
(Springer-Verlag, Berlin, 1980), p. 22.
[180] W.Tulczyjew, Geometric Formulation of Physical Theories (Bibliopolis,

Naples, 1989).
[181] I.Vaisman, Lectures on the Geometry of Poisson Manifolds (Birkhäuser Ver-

lag, Basel, 1994).
[182] I.Vaisman, Second order Hamiltonian vector fields on tangent bundles, Diff.
Geom. and Appl. 5 (1995) 153.
[183] A.Vershik, Classical and non-classical dynamics with constraints, in Global

Analysis – Studies and Applications 1, Lect. Notes in Mathematics 1108
(Springer, Berlin, 1984), p. 278.
[184] A.Vershik, V.Gershkovich, Nonholonomic dynamical systems, in Dynamical

Systems VII, eds V.I.Arnol’d and S.P.Novikov (Springer, Berlin, 1994) p.1.
[185] F.Warner, Foundations of Differential Manifolds and Lie Groups (Springer,

Berlin, 1983).
[186] A.Weinstein, The local structure of Poisson manifolds, J. Diff. Geom. 18

(1983) 523.
[187] A.Weinstein, Co-isotropic calculus and Poisson gruppoids, J. Math. Soc.

Japan 40 (1988) 705.
[188] N.Woodhouse, Geometric Quantization (Clarendon Press, Oxford, 1980).
[189] K.Yano and S.Ishihara, Tangent and Cotangent Bundles, Pure and Applied
Mathematics Ser. 16 (Marcel Dekker, N.Y., 1973).
[190] O. Zakharov, Hamiltonian formalism for nonregular Lagrangian theories in

fibered manifolds, J. Math. Phys. 33 (1992) 607.
350 BIBLIOGRAPHY
[191] Q.Zhao, The representation of Lie groups and geometric quantizations, Com-
mun. Math. Phys. 194 (1998) 135.
[192] R.Zulanke and P.Wintgen, Differentialgeometrie und

Faserbündel, Hochschulbucher fur Mathematik, Band 75 (VEB Deutscher
Verlag der Wissenschaften, Berlin, 1972).

Gauge Mechanics-Sardanashvily

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Gauge Mechanics-Sardanashvily

Uploaded by

Copyright:

Available Formats

Gauge Mechanics

(World Scientific, Singapore, 1998 )

1 Interlude: bundles, jets, connections 9

2 Geometry of Poisson manifolds 59

3 Hamiltonian systems 107

3.5 Dirac Hamiltonian systems . . . . . . . . . . . . . . . . . . . . . . . . 126

4 Lagrangian time-dependent mechanics 153

5 Hamiltonian time-dependent mechanics 229

6 Relativistic mechanics 301

6.3 Relativistic dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . 309

7 Appendix A. Geometry of BRST mechanics 319

8 Appendix B. On quantum time-dependent mechanics 329

written with respect to the holonomic coordinates (q i , pi = q̇i ) on T ∗ M . A Hamil-

on T ∗ M which fulfills the Hamilton equations

Lagrangian conservative mechanics is usually seen as a particular Hamiltonian me-

The Hamiltonian formulation of conservative mechanics cannot be extended in a

and that of the momentum phase space

pr∗2 Ω = dpi ∧ dq i (0.0.5)

which satisfies the Hamilton equations

We will formulate non-relativistic time-dependent mechanics as a particular field

of a mechanical system. Due to this morphism, every connection Γ on the fibred

on Q which is the horizontal lift of the standard vector field ∂t on R by means of

Let us turn to the momentum phase space V ∗ Q of time-dependent mechanics.

• A Hamiltonian on a momentum phase space of time-dependent mechanics fails

where Γ is a connection on the fibred manifold Q → R, while H f is a Hamil-

• A Poisson bracket of a Hamiltonian (0.0.11) with functions on a momentum

• As a consequence, the evolution equation in time-dependent mechanics is not

algorithm for describing constraint systems is not extended to time-dependent

• Quantizations with respect to different reference frames are not equivalent.

We show that every trivialization (0.0.12) of the fibre bundle Q → R defines a

Therefore, Lagrangians in time-dependent mechanics are covariant, but not invariant

(Gh̄c−3 )1/2 = G1/2 = 1, 616 · 10−33 cm,

where G is the Newtonian gravitational constant. Relative to the universal unit

Interlude: bundles, jets,

1.1 Fibre bundles

By a fibre bundle is meant a locally trivial fibred manifold

where a fibration (or a projection) π is a surjective submersion ¿from a manifold Y ,

called local bundle trivializations, together with the transition functions

ρξζ : (Uξ ∩ Uζ ) × V → (Uξ ∩ Uζ ) × V,

xλ (y) = (xλ ◦ π)(y), y ∈ Y,

are coordinates on the base X, and

y i (y) = (y i ◦ pr2 ◦ ψξ )(y)

are coordinates on the typical fibre V .

By a section (or a global section) of a fibre bundle (1.1.1) is meant a manifold

Theorem 1.1.2. [170]. A fibre bundle Y → X whose typical fibre is diffeomorphic

A fibred morphism of two bundles π : Y → X and π 0 : Y 0 → X 0 is a pair of maps

is called a fibred morphism over X.

Theorem 1.1.3. Let Φ : Y → Y 0 be a fibred morphism over X. Given a global

Ker s0 Φ = Φ−1 (s0 (X))

of s(X) by Φ. If Φ : Y → Y 0 is a fibred morphism of constant rank over X, then

An isomorphism of fibre bundles is a fibred morphism (1.1.2) such that Φ is a

is called the restriction of a fibre bundle Y to the submanifold X 0 ⊂ X.

over X is defined as the pull-back

together with the natural projection onto X.

A vector bundle is a fibre bundle Y → X such that:

• there is a bundle atlas Ψ = {(Uα , ψα )} of Y → X whose trivialization mor-

Proposition 1.1.4. If Y → X and Y 0 → X are vector bundles and Φ : Y → Y 0 is a

Unless otherwise stated, by Ker Φ of a linear bundle morphism Φ is meant its

• Let Y → X be a vector bundle with a typical fibre V . By Y ∗ → X is meant

• Let Y → X and Y 0 → X be vector bundles with typical fibres V and V 0 ,

• Let Y → X and Y 0 → X be vector bundles with typical fibres V and V 0 ,

In particular, let Φ : Y1 → Y2 be an injection of a vector bundle π1 : Y1 → X1 to

G : (−, ) × U 3 (t, z) 7→ Gt (z) ∈ Z

• Gt+t0 (z) = (Gt ◦ Gt0 )(z) if t + t0 ∈ (−, ).