Found Phys (2008) 38: 470–487

DOI 10.1007/s10701-008-9216-2

On Classical and Quantum Objectivity

Gabriel Catren

Received: 9 July 2007 / Accepted: 28 February 2008 / Published online: 15 March 2008
© Springer Science+Business Media, LLC 2008

Abstract We propose a conceptual framework for understanding the relationship

between observables and operators in mechanics. To do so, we introduce a postu-
late that establishes a correspondence between the objective properties permitting to
identify physical states and the symmetry transformations that modify their gauge de-
pendant properties. We show that the uncertainty principle results from a faithful—or
equivariant—realization of this correspondence. It is a consequence of the proposed
postulate that the quantum notion of objective physical states is not incomplete, but
rather that the classical notion is overdetermined.

Keywords Quantum mechanics · Symplectic mechanics

1 Introduction

It is commonly stated that quantum mechanics differs from classical mechanics in its
use of operators acting on physical states. According to this description, the transi-
tion from classical to quantum mechanics can be understood as a substitution of a
commutative algebra of functions—relative to pointwise multiplication—by a non-
commutative algebra of operators. Nevertheless, the use of operators acting on phys-
ical states is also an essential feature of classical mechanics. In fact, classical observ-
ables play two fundamental roles in mechanics, namely they are functions that can
be evaluated on states—and used for identifying them—and they define Hamiltonian
vector fields that act on states by means of infinitesimal canonical transformations
(see [1, 2, 5, 9, 10, 13]). For example, the temporal evolution of classical systems is
given by the integration of the infinitesimal classical action generated by the Hamil-
tonian vector field νH associated to the Hamiltonian function H (q, p). Nevertheless,

G. Catren ()
CREA, UMR 7656, École Polytechnique/CNRS, 1 rue Descartes, 75005, Paris, France
e-mail: gabrielcatren@yahoo.com.ar
Found Phys (2008) 38: 470–487 471

the Poisson algebra of classical observables and the Lie algebra of classical operators
(under the Lie bracket of vector fields) are not isomorphic (given that the correspond-
ing map is not injective). This means that classical mechanics fails to establish a
faithful correspondence between the two fundamental roles played by classical ob-
servables. According to Dirac’s quantization conditions [4], quantum mechanics can
be partially obtained by forcing an isomorphism between the algebraic structures as-
sociated to these two roles, namely between a subalgebra of the Poisson algebra of
classical observables and the commutator algebra of operators. In the framework of
the geometric quantization program, this task can be achieved by extending classical
operators νf —Hamiltonian vector fields associated to observables f —to quantum
operators ν̂f , such that the commutator algebra of the latter is isomorphic to the
Poisson algebra of classical observables. This quantum correction of classical oper-
ators can be performed by defining a complex line bundle L → M over the phase
space M, endowed with a hermitian connection such that its curvature is given by the
symplectic form on M. The polarized sections of this complex line bundle define the
corresponding quantum states (see [3, 8, 12–14]).
The fact that quantum mechanics can be obtained by establishing a faithful cor-
respondence between observables and operators suggests that a deeper analysis of
the conceptual meaning of this correspondence could shed some new light on the
rational necessity of quantum mechanics. In what follows, we will propose a con-
ceptual framework for understanding the relation between observables and operators.
We will argue that their correspondence can be understood in terms of the relation
between the objective properties that define physical states and the symmetry trans-
formations that interchange their gauge dependant properties. To do so, we will show
that, even tough it remains valid, the standard characterization of objective properties
as the invariants under symmetry transformations does not suffice for properly under-
standing the correspondence between observables and operators. Therefore, we will
propose a postulate that specifies the lacking fundamental aspect of their relation.
Roughly speaking, we will argue that the objective properties of a physical state play
a twofold role, namely (1) they permit to identify the physical state, and (2) they in-
duce the transformations between the different “aspects”, “profiles” or non-objective
properties of the state. From a conceptual point of view, we will then show that the
proposed postulate implies the necessity of an uncertainty principle. According to
this postulate, the quantum description of a physical state is not underdetermined, but
rather the classical description is overdetermined. This implies that quantum mechan-
ics should not be considered an obstruction to the validity of a well-defined classical
notion of objectivity, but rather as a consistent theory of physical objectivity that over-
comes the classical impasses. The quantum uncertainty principle can be considered
the signature of this achievement.
In this work we analyze the rational necessity of quantum mechanics by proposing
a new interpretation of one of its fundamental features, namely the uncertainty prin-
ciple. In Sect. 2, we propose a postulate—not satisfied by classical mechanics—in
order to understand the fundamental correspondence between observables and oper-
ators. In Sect. 3, we show that this postulate can be further clarified and justified by
giving a precise definition to the notion of physical states’ possible properties. To do
so, we consider the so-called momentum map as well as its equivariance properties.
472 Found Phys (2008) 38: 470–487

We show that the failure of classical mechanics to establish a faithful correspondence

between observables and operators is manifested through the non-equivariance of the
relevant momentum map. We then consider how quantum mechanics can be obtained
by forcing such an equivariance. In Sect. 4, we consider a dual version of this last
result by using the geometric quantization formalism. In Sect. 5, we summarize and
discuss the main results.

2 Objective Physical States

In general, the relevant variables allowing us to specify a physical state depend on

the arbitrary election of a coordinate system. The objective properties of the state are
then identified with the invariants under the symmetry transformations that modify
the coordinate systems. In this section, we argue that this standard characterization of
objective properties does not suffice for understanding the fundamental role played
by symmetry transformations in classical and quantum mechanics.
In classical mechanics, the transformations of the canonical variables used for
specifying a physical state are given by canonical transformations. In particular, we
are interested in considering infinitesimal canonical transformations induced by clas-
sical observables. In fact, the symplectic structure ω of the phase space M per-
mits to pass from an observable f ∈ C ∞ (M) to the symplectic diffeomorphism
f
φλ : M → M generated by the Hamiltonian vector field νf induced by f . The rela-
tion between the observable and the induced Hamiltonian vector field νf is given by
the expression iνf ω = df , where iνf ω denotes the contraction of the 2-form ω with
the vector νf .1 Therefore, the classical observables f ∈ C ∞ (M) play a twofold role.
Not only do they define local coordinates on the phase space M—which can be used
for identifying classical states—but they also induce infinitesimal canonical transfor-
mations. In principle, these transformations might be equivalently interpreted either
as passive transformations of the coordinate system or as active transformations of
physical states [5]. The formal equivalence between passive and active transforma-
tions will be called Leibniz equivalence. In general relativity, the invariance under
the group of general diffeomorphisms of space-time implies that—if Leibniz equiv-
alence is accepted—localization on space-time is pure gauge. In principle, we might
think that the invariance of classical mechanics under the group of simplectic diffeo-
morphisms can be interpreted in an analogous way. However, such is not the case. If
Leibniz equivalence were valid, we would be forced to accept that localization on M
is pure gauge. But, in that case the very notion of a space that parameterizes physi-
cally distinct states would be lost. Consequently, infinitesimal canonical transforma-
tions induced by classical observables are interpreted as active transformations be-
tween physically distinct states. For example, the canonical transformations induced
by the Hamiltonian function H (q, p) ∈ C ∞ (M) generates the temporal evolution of
physical states, that is to say their objective physical change in time. In other words,

1 In M = R2 , the Hamiltonian vector field associated to an observable f ∈ C ∞ (M) is given by the ex-

pression νf = ∂f ∂ ∂f ∂
∂p ∂q − ∂q ∂p .
Found Phys (2008) 38: 470–487 473

“[. . .] the motion of a mechanical system corresponds to the continuous evolution or

unfolding of a canonical transformation” [5]. More generally, through its induced
Hamiltonian vector field νp = ∂q , a momentum p generates canonical transforma-
tions of the conjugated coordinate q. Since H generates temporal evolutions of any
observable and p generates canonical transformations of q, the temporal evolution of
q is given by the dependence of H on p (through the Hamilton’s equation q̇ = ∂H ∂p ).
2

Since the temporal evolution of q is considered an effective physical change, general

transformations generated by p should also be interpreted as active changes of the
state.3
Classical mechanics is thus characterized by its denial of Leibniz equivalence:
states related by canonical transformations induced by classical observables are con-
sidered to be physically distinct states. In what follows we argue that, in order to
improve the comprehension of the foundations of mechanics, it is necessary to par-
tially maintain the validity of Leibniz equivalence. Moreover, quantum mechanics is
the formalism required to satisfy this prescription. According to this interpretative
scheme, we will consider states connected through certain canonical transformations
induced by physical observables to be gauge equivalent. This implies that the particu-
lar value of the coordinate affected by the canonical transformation is not an objective
property of the state. In other words, localization in the orbit of the corresponding
symplectic diffeomorphism is pure gauge. We will call objective reduction the iden-
tification of states by means of certain infinitesimal canonical transformations.
As we have seen, if we considered all possible infinitesimal canonical transforma-
tions as symmetry transformations, all states in M would be gauge equivalent. It is
then necessary to specify the particular subset of infinitesimal canonical transforma-
tions that should be considered for performing the objective reduction of each state.
To do so, we propose the following fundamental postulate:
‡ The objective properties of a physical state define the only generators of
symmetry transformations accepted for its objective reduction.
A set of objective properties that, according to this postulate, defines both the
identity of an objective physical state s and the correct symmetry transformations
needed for its objective reduction, will be called eidos εs of the physical state s.
In this way, the eidos that identifies a particular physical state also specifies which
properties should not be considered objective properties but “profiles” that depend on
the coordinate system.
In classical mechanics the set of objective properties of a physical state is given
by the 2n local coordinates (q i , pi )i=1,...,n of the phase space M (in what follows we
will restrict the analyses to the case M = R2n ; for the consideration of more general
cases see [6]). According to postulate ‡, objective properties are in correspondence
with the generators of the symmetry transformations required for the objective reduc-
tion of the state. In classical mechanics, this correspondence is provided by the map

2 This remark explains the fundamental conceptual difference between q̇ (the infinitesimal temporal evo-
lution of q) and p (the observable that generates infinitesimal canonical transformations of q).
3 For instance, the uniform rotation generated by a conserved angular momentum L is an effective physical
transformation of the system.
474 Found Phys (2008) 38: 470–487

f → νf between classical observables f ∈ C ∞ (M) and Hamiltonian vector fields

νf ∈ HM . In particular, the momentum p defines the generator νp = ∂q of infini-
tesimal translations of the conjugated coordinate q (Lieνp q = νp (q) = {q, p} = 1)
and viceversa. Since in classical mechanics both q and p are used for identifying
the state, postulate ‡ demands that both νq and νp should be used for the objective
reduction of the corresponding state. But this implies that both q and p are gauged
out, which is contradictory with the assumption that they define objective properties
of the state (since by definition objective properties have to be invariant under sym-
metry transformations). In other words, the action defined by both νq and νp reduces
the phase space M to a single point {∗} or, equivalently, the group G of translations
in both position and momentum acts transitively on M (i.e. M is a single G-orbit).
Clearly, this is a reductio ad absurdum. We began by supposing that a physical state
is identified by means of the 2n coordinates of the phase space M, that is to say that
each point in M defines a different physical state. We nevertheless found, guided by
postulate ‡, that there is only one possible physical state {∗}. We can thus conclude
that objective physical states defined by an eidos that satisfies postulate ‡ cannot be
identified with classical states in M.
This classical failure to implement postulate ‡ is a direct consequence of the fact
that classical states are specified by means of too many variables. We will refer to
this fact as the classical overdetermination of the eidos. According to postulate ‡, an
objective property f ∈ εs of the physical state s has to be invariant under the infini-
tesimal canonical transformations generated by the whole set of objective properties
in εs . This implies that νg f = {f, g} = 0 for all f, g ∈ εs . Therefore, the set of ob-
servables in the eidos εs have to define a commutative Poisson algebra. Hence, an
eidos that completely specifies an objective physical state is given by the maximal
number of mutually commuting observables (i.e. by a complete set of commuting ob-
servables). In order to satisfy postulate ‡ it is then necessary to reduce the eidos to
half the number of components. Therefore, an acceptance of the proposed postulate
implies that the quantum notion of objective physical states is not incomplete (when
compared to a hypothetical complete classical description), but rather that the classi-
cal notion is overdetermined. As a consequence of this overdetermination of classical
states, the relationship established between objective properties and symmetry trans-
formations by postulate ‡ cannot be elucidated within the framework of classical me-
chanics. In fact, the assumption of postulate ‡ implies what we could call the a priori
necessity of an uncertainty principle. If an objective physical state is identified by
means of an eidos composed of n components, there should be n other possible prop-
erties that are gauged out. This means that the values of the latter have no objective
physical meaning. For example, the momentum p of a system with a well-defined
position q is completely gauged out by the symmetry transformation induced by the
objective property q. Hence, the momentum p is necessarily undetermined. There-
fore, the possible properties of an objective physical state s can be separated into two
sets: the n objective properties {xi }i=1,...,n belonging to the eidos εs , and the n non-
objective properties {xj }j =n+1,...,2n that are gauged out by the symmetry transforma-
tions defined by the former. The fact that there is an even number of possible proper-
ties is a direct consequence of this intertwining between objective and non-objective
properties. Therefore, the set of possible properties has to be given by a twofold set
of 2n symplectically intertwined—or canonically conjugated—properties.
Found Phys (2008) 38: 470–487 475

Since general canonical transformations mix canonical variables, it should be pos-

sible to have properties which are neither objective properties nor profiles, but a mix-
ture of both. In fact, the flexibility of quantum mechanics’ formalism enables the
consideration of intermediate physical states where neither q nor p are sharp objec-
tive properties. If for example q is an unsharp objective property of a physical state,
the conjugated momentum p is not completely gauged out. Hence p is in turn an un-
sharp objective property that partially gauges the coordinate q. This means that, for
a given physical state, a certain property can be partially considered an unsharp ob-
jective property (that partially gauges the conjugated variable) and partially a gauged
variable. The resulting subtle equilibrium between unsharp objective properties and
unsharp non-objective profiles are formally governed by the uncertainty principle.

3 From Universal Symmetry Transformations to Objective Properties

Postulate ‡ establishes a conceptual correspondence between the objective properties

of a physical state and the induced symmetry transformations that connect its gauge
dependant “profiles”. In this section we will justify the necessity of such a correspon-
dence by giving a precise meaning to the notion of physical states’ properties. To do
so, we will consider the transition from a group of symmetry transformations acting
on states to the definition of certain quantities—the momenta—that define the possi-
ble properties of the states. This relation is provided by the so-called momentum map
(see [9–11]). In other words, instead of going from objective properties to (realized)
symmetry transformations—by means of the map f → νf —we will go from (univer-
sal) symmetry transformations to objective properties. Under certain conditions, the
momentum map permits to define the objective properties of a physical state as quan-
tities that specify how the physical state realizes in a particular way certain universal
operations. As we will show, these conditions are not satisfied in classical mechanics.
The action Φ : G × M → M of a Lie group G on a manifold M defines a map
ι between Lie algebra elements ξ ∈ g and fundamental vector fields νξ on M. This
map is defined by means of the expression νξ (x) = dλ d
(exp(−λξ ) · x)|λ=0 , for x ∈ M.
At each x ∈ M, the fundamental vector field νξ generates the infinitesimal transfor-
mation of the classical state x defined by the universal Lie algebra element ξ ∈ g.
We will say that νξ (x) ∈ Tx M is the realization by the particular state x of the uni-
versal infinitesimal generator ξ . The action Φ is said to be symplectic if G acts on
M by means of symplectic diffeomorphisms, i.e. if Φg∗ ω = ω for all g ∈ G (where
Φg := Φ(g, ·) : M → M). The relevant question is whether it is possible to obtain the
fundamental vector field νξ as a Hamiltonian vector field corresponding to a function
hξ ∈ C ∞ (M). If this is the case, the fundamental vector field νξ has to satisfy the
equation iνξ ω = dhξ . A symplectic G-action is said to be Hamiltonian if there exists
a map

μ̃ : g → C ∞ (M),
ξ → hξ
476 Found Phys (2008) 38: 470–487

(called co-momentum) such that the following diagram commutes

π
0 R C ∞ (M) HM 0

ι
μ̃
g

where the short sequence is exact.4 The co-momentum map μ̃ is defined up to a

constant, i.e. the function μ̃ (ξ ) = hξ + k (with k ∈ R) has also νξ as its induced
Hamiltonian vector field. The corresponding momentum map is the map

μ : M → g∗ ,

defined by
μ(x), ξ = μ̃(ξ )(x) = hξ (x),
with ξ ∈ g and ·, · : g∗ × g → R the natural duality pairing.5 Thanks to these
maps, the realization of universal operations in g as fundamental vector fields on
M can be factorized through functions in C ∞ (M). As we will see, if certain con-
ditions are satisfied—namely if the momentum map is equivariant and if the map
C ∞ (M) → HM is injective—then physical observables can be defined as quantities
that specify how different physical states realize differently universal operations in g.
The important fact is that these conditions are no satisfied in classical mechanics.
For a fixed x ∈ M, μ(x) is an element in g∗ that takes on each infinitesimal
generator ξ ∈ g a certain value μ(x), ξ ∈ R that depends, by definition, on the
state x. The quantity μ(x), ξ can then be used for identifying—at least partially—
the state x. By considering how this quantity depends on the states, this informa-
tion can be used to distinguish between states. To do so, one can fix a universal
generator ξ ∈ g and compare the values of μ(x), ξ for different states x. In fact,
each element ξ ∈ g defines a function hξ : M → R (called a momentum) given by
hξ (x) = μ(x), ξ . By construction, the Hamiltonian vector field associated to the
momentum hξ is the fundamental vector field νξ that realizes the universal transfor-
mation ξ . For example, the angular momentum L(q, p) of a particle in Euclidean
space is given by the momentum map μ(q, p) := L(q, p) = q × p. If one selects
a particular Lie algebra element ξ ∈ g = R3 , the function Lξ : M → R given by
= Lξ (q, p) = μ(q, p), ξ = ξ · (q × p) defines a possible property of classical states
in M. The Hamiltonian vector field νLξ associated to the observable Lξ generates
canonical transformations that can be identified with rotations around the axis de-
fined by ξ .

4 The exactness of this sequence simply means that the image of the inclusion R → C ∞ (M)—the constant
functions f = k ∈ R in C ∞ (M)—coincides with the kernel of the projection π . In other terms, νf =k = 0.
5 The ambiguity in the co-momentum map μ̃ induces an ambiguity in the corresponding momentum map.
The modified momentum map μ has to satisfy μ (x), ξ = μ̃ (ξ )(x) = μ̃(ξ )(x) + k. If one defines
μ (x) = μ(x) + a with a ∈ g∗ such that a(ξ ) = k, this relation is satisfied. This means that the momentum
map is defined up to a constant in g∗ .
Found Phys (2008) 38: 470–487 477

If the co-momentum map μ̃ is a homomorphism of Lie algebras—i.e. if μ̃([ξ, η]) =

{μ̃(ξ ), μ̃(η)}—the symplectic G-action is said to be strongly Hamiltonian. It can
be shown that this is the case if and only if the momentum map is infinitesimally
equivariant, i.e. if Tx μ(νξ (x)) = −ad∗ξ μ(x) for all ξ ∈ g, where Tx μ : Tx M →
T g∗ g∗ and ad∗ denotes the coadjoint action of g on g∗ (see [10, 11]).6 The action
of a group G on the manifold M is equivariant if Ad∗g −1 ◦ μ = μ ◦ Φg (where Ad∗
denotes the coadjoint action of G on g∗ ), i.e. if the following diagram commutes

μ
M g∗

Φg Ad∗ −1
g

M g∗
μ

It can be shown that equivariance implies infinitesimal equivariance and that the
converse is true only if G is connected [10]. For compact Lie groups, one can always
use the fact that momentum maps are defined up to a constant in g∗ in order to choose
them as equivariant [11].
We will now consider the significance of this equivariance property. The important
result is that, for equivariant momentum maps, the dual g∗ of the Lie algebra g defines
a final object in the category of Poisson G-manifolds, the momentum map μ being
the unique morphism from an object in the category—i.e. a phase space M—to the
universal object g∗ [7]. A Poisson G-manifold is a pair (M, μ̃), where M is a Poisson
manifold with a G-action and μ̃M a strong co-momentum map, i.e. a homomorphism
of Lie algebras μ̃M : g → C ∞ (M). It is then possible to define the category G of
all Poisson G-manifolds, where a morphism α : (M, μ̃M ) → (N, μ̃N ) is a smooth
map from M to N which preserves the Poisson bracket structure (i.e. which satisfies
α ∗ {f, g} = {α ∗ f, α ∗ g}) and such that the following diagram commutes

α∗
C ∞ (N ) C ∞ (M)

μ̃N μ̃M
id
g g

It can be shown that the dual vector space g∗ of the Lie algebra g defines a final
object in this category. The action of G on g∗ is the coadjoint action. Under this ac-

6 The adjoint representation of a Lie group G on its Lie algebra g is defined by Ad := T I : g → g,

g e g
where Ig : G → G is given by Ig (h) = ghg −1 . The infinitesimal adjoint action of g on g is defined by
the map adξ : g → g given by adξ η = [ξ, η] with ξ, η ∈ g. The coadjoint action of G on g∗ is defined
by using the dual map Ad∗g : g∗ → g∗ given by Ad∗g μ, ξ = μ, Adg (ξ ) , with μ ∈ g∗ and ξ ∈ g. The
coadjoint action of G on g∗ is then defined by the map Φ ∗ : G × g∗ → g∗ given by (g, μ) → Ad∗ −1 μ.
g
The infinitesimal coadjoint action of g on g∗ is defined by the map ad∗ξ : g∗ → g∗ given by the expression
ad∗ξ (μ) = μ, [ξ, ·] [10].
478 Found Phys (2008) 38: 470–487

tion, g∗ breaks up into so-called coadjoint orbits.7 The vector space g∗ has a canon-
ical Poisson bracket structure. This means that g∗ is a Poisson manifold, with the
coadjoint orbits as its symplectic leaves. The Poisson bracket on C ∞ (g∗ ) can be de-
fined as follows. If f, g ∈ C ∞ (g∗ ), then df (x), dg(x) ∈ Tx∗ g∗ g (for x ∈ g∗ ). The
Poisson bracket can then be defined as {f, g}(x) = x, [df (x), dg(x)] . In the Pois-
son G-manifold g∗ , it is possible to define a co-momentum map μ̃g∗ : g → C ∞ (g∗ )
given by μ̃g∗ (ξ )(x) = x, ξ .8 Therefore, the pair (g∗ , μ̃g∗ ) is a Poisson G-manifold.
Moreover, it is a final object in the category G of Poisson G-manifolds, being the
strong momentum map μ : M → g∗ the unique morphism from an object (M, μ̃M )
to (g∗ , μ̃g∗ ). In this way, the momentum map μ : M → g∗ establishes an identifica-
tion between states in M and elements of the universal model g∗ for the G-action.9
Since G also acts on g∗ , the characterization of a mechanical system as a realization
of the universal model g∗ is valid only if the momentum map μ : M → g∗ respects the
structure defined by the G-action. In other words, the G-action has to be equivariant.
This means that the realization of the universal model established by the momentum
map commutes with the G-action on both M and g∗ . The fact that an equivariant
momentum map preserves the relevant structures when passing from the universal
model g∗ to its realization by M is also manifested by the fact that such a map is a
Poisson map, i.e. it satisfies μ∗ {f, g} = {μ∗ f, μ∗ g}, for f, g ∈ C ∞ (g∗ ).
To sum up, we can say that universal operators in g act on concrete states in M by
means of realized operators in HM (fundamental or Hamiltonian vector fields). This
realization is mediated by classical observables by means of the co-momentum map
μ̃ : g → C ∞ (M). Since by definition the Lie algebra g acts on its dual g∗ , it is natural
to discover that g∗ constitues a universal model of the phase space M. This means
that g∗ is a final object in the category of Poisson G-manifolds. The identification
between the phase space M and the universal model g∗ is given by the momentum
map μ : M → g∗ . This correspondence is valid only if the maps respect the relevant
algebraic structures. This means that the G-action on both M and g∗ and the Poisson
structures have to be preserved by the corresponding maps.
We will now consider a potential obstruction to the momentum map’s infinitesimal
equivariance. This obstruction measures the extent to which the co-momentum map
μ̃ : g → C ∞ (M) fails to be a homomorphism of Lie algebras. The obstruction is then

7 The coadjoint orbit through μ ∈ g∗ is the subset of g∗ defined by O ≡ G · μ := {Ad ∗ (μ), ∀g ∈ G}.
μ g −1
The fundamental property of coadjoint orbits is that they are symplectic manifolds. In order to de-
fine the corresponding symplectic structures let’s consider the tangent vectors to coadjoint orbits. If
μ(t) = Ad∗ −1 μ is a curve in Oμ with μ(0) = μ and g(t) = exp(tξ ) for ξ ∈ g, it can be shown
g(t)
that μ (0) = −ad∗ξ μ. Thus Tμ Oμ = {ad∗ξ μ, ξ ∈ g}. The symplectic structure on Oμ can be defined as
ωOμ (μ)(−ad∗ξ μ, −ad∗η μ) = μ, [ξ, η] . It can be shown that the map Ad∗ −1 : Oμ → Oμ preserves ωOμ
g
(see [10] for more details).
8 It can be verified that the corresponding momentum map is the identity. Using that μ ∗ (x), ξ =
g
μ̃g∗ (ξ )(x) = x, ξ , it follows that μg∗ (x) = x.
9 In particular, a coadjoint orbit can be considered the universal model for the orbits of the G-action. It can
be shown that the image of μ is a union of coadjoint orbits. In particular, if G acts transitively on M, then
the momentum map is a covering of a unique coadjoint orbit.
Found Phys (2008) 38: 470–487 479

given by the map

Σ :g×g→R
defined by
Σ(ξ, η) = μ̃([ξ, η]) − {μ̃(ξ ), μ̃(η)}.
By using Jacobi’s identity in both {·, ·} and [·, ·], it can be shown that Σ satisfies

Σ(ξ, [η, ζ ]) + Σ(η, [ζ, ξ ]) + Σ(ζ, [ξ, η]) = 0.

This means that Σ defines a 2-cocycle in the Lie algebra cohomology H ∗ (g, R)
of g with values in R.10 The corresponding cohomology class [Σ] ∈ H 2 (g, R) is the
obstruction to the infinitesimal equivariance of the momentum map. The cocycle Σ is
a coboundary if there exists a 1-cochain λ : g → R such that Σ(ξ, η) = (δg λ)(ξ, η) =
−λ([ξ, η]). Two co-momentum maps differing in a constant element λ ∈ g∗ have as-
sociated cocycles differing in a coboundary: Σλ (ξ, η) = Σ(ξ, η) + λ([ξ, η]). This
means that they belong to the same cohomology class: [Σλ ] = [Σ]. If the cohomol-
ogy class is trivial—i.e. if [Σ] = 0 or equivalently Σ(ξ, η) = −λ([ξ, η])—then the
momentum map can be modified so as to be infinitesimally equivariant. In fact, the
new map μ̃ = μ̃ + λ is an infinitesimally equivariant co-momentum map.11 In partic-
ular, if g is semisimple, then the Second Whitehead Lemma states that H 2 (g, R) = 0.
When the cohomology class of Σ is not trivial, the momentum map can still be
adjusted to be infinitesimally equivariant by enlarging g to the central extension g =
g + R defined by Σ. The bracket in g is defined by

[(ξ, a), (η, b)] = ([ξ, η], Σ(ξ, η)).

10 Let C k (g, R) = {α : gk → R} be the set of R-valued k-cochains (where the maps α are skew k-linear

maps). Let’s define the differential δg : C k (g, R) → C k+1 (g, R) as


(δg α)(ξ0 , . . . , ξk ) = (−1)i+j α([ξi , ξj ], ξ0 , . . . , ξ̂i , . . . , ξ̂j , . . . , ξk ),
0≤i<j ≤k

where ξ̂i means that ξi has been suppressed. It can be shown that δg2 = 0. The corresponding cohomol-
ogy H ∗ (g, R) = ker(δg∗ )/im(δg∗−1 ) is the Lie algebra cohomology of g with values in R. In our case,
the element Σ is a skew bilinear map that satisfies (δg Σ)(η, ξ, ζ ) = −(Σ([η, ξ ], ζ ) + Σ([ζ, η], ξ ) +
Σ([ξ, ζ ], η)) = 0, i.e. it is a 2-cocycle in H 2 (g, R).
11 The obstruction associated to the co-momentum map μ̃ = μ̃ + λ is

Σ  (ξ, η) = μ̃ ([ξ, η]) − {μ̃ (ξ ), μ̃ (η)}

= μ̃([ξ, η]) + λ([ξ, η]) − {μ̃(ξ ) + λ(ξ ), μ̃(η) + λ(η)}
= μ̃([ξ, η]) + λ([ξ, η]) − {μ̃(ξ ), μ̃(η)}
= Σ(ξ, η) + λ([ξ, η])
= 0.
480 Found Phys (2008) 38: 470–487

The Lie algebra g acts on M by means of the Hamiltonian vector field νξ (x). The
induced momentum and co-momentum maps

μ : M → (g )∗ = g∗ ⊕ R,
μ̃ : g = g + R → C ∞ (M)

satisfy
νh(ξ,a) = νhξ ,
where h(ξ,a) = μ̃ (ξ, a) and hξ = μ̃(ξ ). This implies that h(ξ,a) − hξ = ρ(ξ, a), with
ρ(ξ, a) constant. The obstruction to the infinitesimal equivariance of μ̃ is

Σ  ((ξ, a), (η, b)) = μ̃ ([(ξ, a), (η, b)]) − {μ̃ (ξ, a), μ̃ (η, b)}
= μ̃ ([ξ, η], Σ(ξ, η)) − {μ̃(ξ ) + ρ(ξ, a), μ̃(η) + ρ(η, b)}
= μ̃([ξ, η]) + ρ([ξ, η], Σ(ξ, η)) − {μ̃(ξ ), μ̃(η)}
= Σ(ξ, η) + ρ([ξ, η], Σ(ξ, η)).

The 2-cocycle Σ  ((ξ, a), (η, b)) belongs to the trivial class if it is a coboundary,
i.e. if Σ  ((ξ, a), (η, b)) = −λ([(ξ, a), (η, b)]) = −λ([ξ, η], Σ(ξ, η)). Let’s define the
1-cochain τ : g → R given by τ (ξ, a) = a. Then Σ(ξ, η) = τ ([ξ, η], Σ(ξ, η)) and
thus

Σ  ((ξ, a), (η, b)) = (τ + ρ)([ξ, η], Σ(ξ, η))

= (τ + ρ)([(ξ, a), (η, b)])
= −λ([(ξ, a), (η, b)]),

with τ + ρ = −λ. Therefore, the co-comentum map μ̃ can be adjusted to be infini-
tesimally equivariant by choosing ρ(ξ, a) = −a. The resulting co-momentum map is
then
μ̃ (ξ, a) = μ̃(ξ ) − a.
We will now consider the action of g = R2 = {(a, b)} on M = R2 = {(q, p)} given
by Φ(a,b) (q, p) = (q +a, p +b). The realization of the Lie algebra element ξ = (a, b)
is given by the fundamental vector field νξ = a∂q + b∂p . This vector field is the
Hamiltonian vector field corresponding to the function hξ (q, p) = ap − bq. The co-
momentum map is then

μ̃ : g → C ∞ (M),
(a, b) → hξ (q, p) = ap − bq.

This co-momentum map is not infinitesimally equivariant. This means that it is

not a homomorphism of Lie algebras. The obstruction Σ is given by

Σ((a1 , b1 ), (a2 , b2 )) = μ̃([(a1 , b1 ), (a2 , b2 )]) − {μ̃(a1 , b1 ), μ̃(a2 , b2 )}

= b1 a2 − a1 b2 .
Found Phys (2008) 38: 470–487 481

Since [g, g] = 0, the only coboundary is zero. This means that Σ belongs to a
non-trivial cohomology class. In particular let’s consider the Lie algebra elements
ξq = (1, 0) and ξp = (0, 1). These elements induce on M the fundamental vector
fields νξq = ∂q and νξp = ∂p respectively. In this case the obstruction is

Σ(ξq , ξp ) = μ̃([ξq , ξp ]) − {μ̃(ξq ), μ̃(ξp )}

= μ̃(0) − {p, −q}
= −1.

Since the action of g = R2 on M = R2 is not infinitesimally equivariant, classi-

cal mechanics on M is not a satisfactory realization of the universal model g∗ for
the G-action. It is worth noting that the non-equivariance of μ̃ is a consequence of
the fact that the algebraic structure of g does not reflect the symplectic intertwining
between physical properties. In other words, [ξq , ξp ] = 0 even if {q, p} = 1. The in-
finitesimally equivariant co-momentum map is obtained through the central extension
of g = R2 defined by Σ . It is given by

μ̃ : g = R2 ⊕ R → C ∞ (M),
((a, b), c) → h(a,b) − c = ap − bq − c.

The central extension g = R2 ⊕ R is called Heisenberg algebra and the corre-

sponding connected and simply-connected Lie group is the Heisenberg group H .
The underlying manifold is H = R2 × S 1 (where R2 is considered an additive Lie
group), with multiplication given by the expression

 1 
(g1 , eiθ1 ) · (g2 , eiθ2 ) = g1 + g2 , ei[θ1 +θ2 + 2 Σ(g1 ,g2 )] ,

with identity (0, 1) and inverse (g, eiθ )−1 = (−g, e−iθ ).
As we have argued above, adopting postulate ‡ makes it impossible to identify the
space of states with the phase space M (since classical states x ∈ M are overdeter-
mined). We will now provide some heuristic justifications for the election of a new
space of states. If certain conditions were satisfied, it would be possible to use the mo-
mentum map to give a rigorous definition of physical properties. More precisely, if
the moment map were equivariant—and, as we will see in the next section, if the map
C ∞ (M) → HM were injective—the properties μ̃(ξ ) = hξ : M → R would spec-
ify how different physical states realize differently universal generators ξ ∈ g. This
characterization of physical properties would explain why the correspondence be-
tween observables and operators is a fundamental feature of mechanics. The fact that
physical properties hξ induce generators of infinitesimal canonical transformations—
through the map π : C ∞ (M) → HM —would be a consequence of the fact that hξ are
quantities—defined through the map μ̃ : g → C ∞ (M)—that specify how the univer-
sal generators ξ ∈ g are realized by states in M. The following diagram encompasses
482 Found Phys (2008) 38: 470–487

both sides of the relationship between observables and operators:

μ̃ π
g C ∞ (M) HM

The realization—as Hamiltonian vector fields in HM —of universal symmetry

transformations in g is factorized through physical properties in C ∞ (M), which can
be used for identifying the state. In this way, each physical property defines a realiza-
tion of a universal infinitesimal canonical transformation that acts on another possible
property. The phase space M is then characterized by a set of 2n simplectically inter-
twined possible properties. In particular, we have considered the transitive action of
the translation group R2 on M = R2 . The problem we found is that this group does
not reflect the symplectic intertwining between properties in M (since [ξq , ξp ] = 0).
This fact is manifested through the non-equivariance of the corresponding momentum
map (i.e. it does not preserve the relevant algebraic structures). In order to guaran-
tee the equivariance of the momentum map, the translation group R2 has to be ex-
tended to the Heisenberg group H = R2 ×S 1 . Nevertheless, even if the corresponding
Heisenberg algebra reflects the symplectic intertwining between canonically conju-
gated properties, [(ξq , 0), (ξp , 0)] = ([ξq , ξp ], Σ(ξq , ξp )) = (0, −1), its action on M
is still given by classical operators, i.e. by Hamiltonian vector fields. In other terms,
(ξ, a) ∈ g acts on M by means of the Hamiltonian vector field νξ (x). As we will
see in the next section, the Lie algebra of Hamiltonian vector fields is not isomorphic
to the Poisson algebra. This means that it does not reflect the intertwining between
coordinates q and momenta p. If one wants to define physical properties as quan-
tities that specify how different states realize differently universal operations, this
realization must be faithful. In other words, the realized operator algebra has to be
isomorphic to the universal operator algebra. This means that non-commuting univer-
sal operators should be mapped to non-commuting realized operators. Moreover, it is
also necessary to consider the following important factor. Since (q, p) is a complete
set of observables—every function that commutes with both q and p is necessarily
constant—the induced (quantum) operators ν̂q and ν̂p must also be a complete set.
This means that any operator commuting with both ν̂q and ν̂p is necessarily a multiple
of the identity. This conditions amounts to demand that the operators ν̂q and ν̂p act
irreducibly on the space of states.12 To sum up, the searched space of states has to be
endowed with an irreducible action of operators ν̂q and ν̂p , such that their commuta-
tor algebra is isomorphic to the Poisson algebra. In the case M = R2n , the Stone-Von
Neumann theorem guarantees all irreducible representations of the Heisenberg alge-
bra to be unitary equivalents to the Schrödinger representation (for a fixed value of
Planck’s constant). In this representation, operators associated to the observables q
and p act on functions in L2 (Rn ) by means of the expressions ν̂q ψ(q) = qψ(q) and
ν̂p ψ(q) = −i ∂ψ(q)
∂q . It is worth noting that the resulting quantum states satisfy pos-
tulate ‡ by construction. For example, if a quantum state has a well-defined position,

12 Irreducibility means that every closed subspace of the space of states which is invariant under the action
of this set is either {0} or the whole space of states.
Found Phys (2008) 38: 470–487 483

|ψ = |q , the transformation generated by the operator ν̂q (multiplication by q in

the coordinate representation) should not be an active transformation between phys-
ically distinct states, but rather a symmetry transformation that does not modify the
physical state. In fact, the transformed state ei ν̂q k |q = eiqk |q is gauge equivalent
to |q . On the other hand, since the observable p is not an objective property of the
state |q , transformations generated by ν̂p —i.e. translations in position—should not
be considered symmetry transformations, but rather active transformations between
physically different states. This is consistent with the fact that ei ν̂p k |q is not gauge
equivalent to |q .

4 From Objective Properties to Realized Symmetry Transformations

We will now consider the results described in the preceding section from a dual point
of view. We have seen that in order to define an equivariant co-momentum map
between universal operators and observables it is necessary to enlarge the translation
group that acts on M by adding an extra S 1 -degree of freedom (defining in this way
the Heisenberg group). Conversely, the prequantization formalism shows that in order
to define a faithful correspondence between observables and realized operators it is
necessary to enlarge the symplectic manifold M by adding an extra S 1 -degree of
freedom [3, 8, 12–14].
A fundamental feature of classical mechanics is that the surjective Lie algebra
homomorphism C ∞ (M) → HM —defined by the symplectic structure ω—between
classical observables f ∈ C ∞ (M) and classical operators νf ∈ HM is not injective,
being its kernel the set of constant functions f = k ∈ R. These properties of the map
C ∞ (M) → HM can be summed up by saying that the short sequence

i π
0 R C ∞ (M) HM 0

is exact. In other words, the image of the injection i—the constant functions in
C ∞ (M)—is the kernel of the projection π (i.e. νf =k = 0). This means that, as Lie al-
gebras, the Poisson algebra of classical observables C ∞ (M) and the Lie algebra HM
of Hamiltonian vector fields are not isomorphic. Consequently, while for example the
infinitesimal transformation of q generated by p is given by Lieνp q = {q, p} = 1, the
same infinitesimal action applied to νq yields Lieνp νq = [νp , νq ] = 0. In other words,
the classical differential operator νp acts non-trivially on q and trivially on νq . This
means that classical mechanics fails to establish a faithful correspondence between
observables and operators: a transformation of an observable f is not necessarily
reflected at the level of the induced operator νf . Nevertheless, the definition of a
physical property as a quantity that specifies how different states realize differently
universal operations is consistent only if a change in the value of the property is
faithfully reflected in the corresponding realized operators. It is worth noting that
[νp , νq ] = 0 (while {q, p} = 1) is the dual version of the fact—considered in the
preceding section—that [ξp , ξq ] = 0 (while {μ̃(ξp ), μ̃(ξq )} = {q, p} = 1). In other
words, both the universal operator algebra g and the realized operator algebra HM
484 Found Phys (2008) 38: 470–487

fail to reflect the symplectic intertwining defined by the Poisson algebra between
properties that are canonically conjugated.
The geometric arena of classical mechanics is insufficient for establishing a faith-
ful correspondence between observables and operators. As the prequantization for-
malism shows, in order to define an operator algebra isomorphic to the Poisson alge-
bra C ∞ (M), it is necessary to extend classical operators νf to quantum operators ν̂f
by adding additional components to the former. To do so, the symplectic manifold M
has to be extended by defining a complex line bundle over it. The quantum operators
ν̂f must satisfy the quantization conditions proposed by Dirac [4]:
Q1 ) If f = k ∈ R, then ν̂f =k = kI , where I is the identity operator.
Q2 ) If {f, g} = h, then [ν̂f , ν̂g ] = −iν̂h .
Q3 ) If {fi } is a complete set of classical observables, then the Hilbert space of
quantum states has to be irreducible under the action of the set {ν̂fi }.
The prequantization formalism shows that, in order to define an operator algebra
that satisfies Q1 and Q2 , it is necessary to define a complex line bundle L → M over
the phase space M with a hermitian connection ∇, such that its curvature is given
by the symplectic form ω. The additional “internal” dimensions defined by the fibers
let us extend the classical operators νf ∈ HM by means of vertical components ζf
tangent to the fibers. By construction, the extended quantum operators ν̂f = νfh + ζf
(where νfh is the horizontal lift of νf defined by the connection) satisfy commutation
relations isomorphic to a subalgebra of classical observables’ Poisson algebra. This
guarantees that [ν̂q , ν̂p ] = 0, even if [νq , νp ] = 0. By adding an internal “quantum”
dimension to the symplectic manifold M, it is possible to define a quantum operator
algebra isomorphic to the Poisson algebra. It can then be shown that the differential
operators ν̂f acts on sections s : M → L by means of the expression [3]

ν̂f · s = ∇νf s − f s.

Therefore, it might seem natural to identify the sections s : M → L with the new
notion of physical states. Nevertheless, since these sections depend on the 2n coor-
dinates of the phase space M, they do not correspond to the usual notion of quantum
states, which only depends on n coordinates. In fact, if quantum states were defined
by these sections, they could be localized in both q and p. This means that both q and
p could be objective properties of the physical state. However, as explained above,
such simultaneous localization would be contradictory with postulate ‡. The formal
counterpart to this conceptual objection is that the quantum operators associated to
a complete set of classical observables by the prequantification formalism, is not a
complete set (they do not act irreducibly on the space of quantum states [7]).13 This
problem can be solved by restricting the space of quantum states. To do so, it is nec-
essary to perform a second stage in the quantization process given by the election

13 For example, the prequantization of the cotangent bundle M = T ∗ R associates to the complete set of
∂ and ν̂ = −i ∂ respectively. Let’s
canonical variables q and p the quantum operators ν̂q = q + i ∂p p ∂q
∞ ∞
consider the subset C (R) ⊂ C (M) composed of states of the form ψ(q). The transformed states
ν̂q ψ(q) = qψ(q) and ν̂p ψ(q) = −i ∂ψ(q) ∞
∂q also depend only on q. This means that C (R) is a proper
subspace of quantum states invariant under the action of the quantum operators ν̂q and ν̂p .
Found Phys (2008) 38: 470–487 485

of a polarization (i.e. a foliation of M by Lagrangian submanifolds [14]). The re-

sulting polarized quantum states only depend on the n coordinates of a Lagrangian
submanifold of M. This means that they cannot be localized in both q and p.
In this way the geometric quantization formalism allow us to reobtain the results
described in the preceding section. In order to define physical properties by means of
an equivariant realization of universal operations, it is necessary to extend the transla-
tion group in M to the Heisenberg group. The Stone-Von Neumann theorem assures
that there is essentially a unique irreducible representation of this group, which is
given (modulo unitary equivalence) by the Schrödinger representation. On the other
hand, in order to define a quantum operator algebra that is isomorphic to the Poisson
algebra, it is necessary to extend the symplectic manifold by defining a fiber bundle
over it. The corresponding operator algebra acts irreducibly on the space of polar-
ized sections. By choosing a vertical polarization the Schrödinger representation is
recovered [14].

5 Conclusion

Beyond the empirical observation that quantum mechanics is, at least to our present
knowledge, the correct theory of mechanics, it still lacks a satisfactory explanation of
the rational necessity of substituting classical mechanics by quantum mechanics. In
this paper we proposed that an analysis of the relations between objective properties
of physical states and symmetry transformations could shed new light on this prob-
lem. The twofold role played by classical observables in mechanics—as functions
that can be evaluated on states and as generators of canonical transformations—is
considered here a fundamental feature that deserves further attention. Our first obser-
vation is that the classical correspondence between these two roles is not satisfactory.
This can be seen in two ways. On the one hand, the map C ∞ (M) → HM between
classical observables and Hamiltonian vector fields is not injective. On the other hand,
the co-momentum map μ̃ : g → C ∞ (M) is not equivariant (where g is the Lie alge-
bra of the translation group in M). This means that the Lie algebras HM and g do
not reflect the symplectic intertwining between properties that are canonically con-
jugated. In fact, quantum mechanics can be understood as the theory of mechanics
that bypasses these flaws. Dirac’s quantization conditions can be considered a for-
malization of this prescription. The geometric quantization formalism shows that in
order to satisfy Dirac’s conditions (at least for a certain subalgebra of observables),
classical operators (Hamiltonian vector fields) have to be extended by adding new
“vertical” components. In order to define these new components it is necessary to
extend the phase space M by defining a complex line bundle over it (endowed with
a hermitian connection of curvature defined by the symplectic form). Quantum states
are then given by polarized sections of the line bundle. From the dual point of view, in
order to define an equivariant momentum map, the translation group in M has to be
extended to the Heisenberg group. For M = R2n , the Stone-Von Neumann theorem
fixes (modulo unitary equivalence) its unique irreducible representation.
These formal results show that, to a certain extent, quantum mechanics can be
recovered by forcing a faithful—or equivariant—correspondence between observ-
486 Found Phys (2008) 38: 470–487

ables and operators. Hence, in order to understand the necessity of quantum me-
chanics, it is necessary to better establish the rational link between the two roles
played by observables. As a means of establishing this rational link, we proposed
postulate ‡. According to this postulate, the infinitesimal canonical transformations
induced by the objective properties of a physical state should not be considered active
transformations between different physical states—as in classical mechanics—but as
symmetry transformations that do not affect its objective properties, i.e. that do not
modify the physical state. This implies that states connected by these transforma-
tions have to be considered gauge equivalent. In other words, the objective proper-
ties that allow us to identify the state define also the symmetry transformations that
gauge its non-objective properties. Moreover, the momentum map formalism per-
mits a better understanding of the relationships between observables and operators.
According to this formalism, the objective properties of a physical state character-
ize the way the state realizes equivariantly certain universal transformations. Such
considerations give a conceptual scope to the correspondence between observables
and operators. Nevertheless, we showed that classical mechanics cannot be consis-
tently interpreted by means of postulate ‡. In fact, a determination of classical states
by both q and p is inconsistent with the interpretation of the induced infinitesimal
canonical transformations as effective symmetry transformations. According to the
proposed postulate, since each objective property characterizes the realization of a
symmetry transformation that gauges the canonically conjugated property, not all
properties can be consistently attributed to a particular physical state. This implies
that classical states are overdetermined, that is to say that they are defined by means
of too many canonical variables. On the contrary, the quantum uncertainty princi-
ple can be considered a faithful realization of postulate ‡. The momentum of a state
with a well-defined position q is completely gauged out by means of the symmetry
transformation induced by the observable q (which is an objective property of the
state). Therefore, we can conclude that it is not the case that the quantum notion of
physical objectivity is underdetermined—when compared to a hypothetical classi-
cal completeness—but that the classical notion is overdetermined. It is a remarkable
fact that non-commutativity of quantum operators is the condition of possibility for a
well-defined notion of physical objectivity. Quantum non-commutativity guarantees
that the symplectic intertwining between properties that are canonically conjugated
is faithfully reflected at the level of operators. Therefore, physical properties of states
can be consistently defined as the quantities that specify faithfully how different states
realize different universal operations.

Acknowledgements This work was partially supported by the grants PICT 04-17687 (ANPCyT), PIP
No 1478/01 (CONICET), UBACyT No X204 and the financial support of Fondation Maison des Sciences
de l’Homme (Paris).

References

1. Abraham, R., Marsden, J.E.: Foundations of Mechanics, 2nd edn. Addison-Wesley, Reading (1978)
2. Arnold, V.I.: Mathematical Methods of Classical Mechanics. Springer, New York (1989)
3. Brylinski, J.-L.: Loop Spaces, Characteristic Classes, and Geometric Quantization. Progress in Math-
ematics, vol. 107. Birkhäuser, Boston (1993)
Found Phys (2008) 38: 470–487 487

4. Dirac, P.A.M.: The Principles of Quantum Mechanics, revised 4th edn. Oxford Univ. Press, Oxford
(1967)
5. Goldstein, H.: Classical Mechanics. Addison-Wesley, Reading (1981)
6. Isham, C.J.: Topological and Global Aspects of Quantum Theory. In: DeWitt, B.S., Stora, R. (eds.)
Relativity, Groups and Topology II. Les Houches, Session XL. Elsevier, Amsterdam (1984)
7. Kirillov, A.A.: Lectures on the Orbit Method. Graduate Studies in Mathematics, vol. 64. Am. Math.
Soc., Providence (2004)
8. Kostant, B.: Quantization and Unitary representations. Lecture Notes in Mathematics, vol. 170.
Springer, New York (1970)
9. Libermann, P., Marle, C.-M.: Symplectic Geometry and Analytical Mechanics. Reidel, Dordrecht
(1987)
10. Marsden, J.E., Ratiu, T.S.: Introduction to Mechanics and Symmetry, 2nd edn. Springer, New York
(1999)
11. Ortega, J.-P., Ratiu, T.S.: Momentum Maps and Hamiltonian Reduction. Birkhäuser, Boston (2004)
12. Puta, M.: Hamiltonian Mechanical Systems and Geometric Quantization. Kluwer Academic, Dor-
drecht (1993)
13. Souriau, J.-M.: Structure of Dynamical Systems. A Symplectic View of Physics. Birkhäuser, Boston
(1997)
14. Woodhouse, N.: Geometric Quantization, 2nd edn. Oxford University Press, Oxford (1992)