Communications in

Commun. math. Phys. 65, 189-201 (1979) Mathematical

Physics
© by Springer-Verlag 1979

Geometrization of Quantum Mechanics

T. W. B. Kibble
Blackett Laboratory, Imperial College, London SW7 2BZ, England

Abstract. Quantum mechanics is cast into a classical Hamiltonian form in

terms of a symplectic structure, not on the Hubert space of state-vectors but on
the more physically relevant infinite-dimensional manifold of instantaneous
pure states. This geometrical structure can accommodate generalizations of
quantum mechanics, including the nonlinear relativistic models recently
proposed. It is shown that any such generalization satisfying a few physically
reasonable conditions would reduce to ordinary quantum mechanics for states
that are "near" the vacuum. In particular the origin of complex structure is
described.

1. Introduction

Geometrical ideas, especially symplectic structures, have come to play an increas-

ingly important role in classical mechanics [1, 2]. The geometry of the classical
phase space Γ also underlies the geometrical quantization programme of Souriau
[3, 4] (see also Kostant [5]). Moreover it is known that quantum dynamics can be
expressed in terms of a Hamiltonian structure on the Hubert space ffl of state-
vectors, where the imaginary part of the scalar product defines a symplectic
structure [6].
However if one is seeking an axiomatic basis for quantum mechanics, it seems
better to start from structures of direct physical significance, as in the operational
approach of Haag and Kastler [7] and others [8, 9] or the work on the geometry
of quantum logics [10, 11].
It is pointed out in Sect. 2 that this can be achieved by a slight modification of
the Hamiltonian formalism. We have to consider not the Hubert space tff itself
but the manifold Σ of "instantaneous pure states" which is (essentially but not
quite) a projective Hubert space. This formalism is closely akin to the work of
Mielnik [12] on the geometry of the space of quantum states. It provides a
convenient framework within which to discuss possible generalizations of quan-
tum mechanics. In particular it can readily accommodate the relativistic model
theories proposed in a recent paper [13], or at least a large class of them. It is
190 T. W. B. Kibble

worth noting that by formulating the theory on Σ rather than Jtf we automatically
ensure that it satisfies the scaling property shown in [13] to be necessary for a
consistent measurement theory (see also [14]).
From this point of view, the essential difference between classical and quantum
mechanics lies not in the set of states (save for the infinite dimensionality) nor in
the dynamic evolution, but rather in the choice of the class of observables, which is
far more restricted in quantum than in classical mechanics.
One motivation for the present work is the possibility that it might be of use in
the unification of quantum mechanics with general relativity. The idea that this
unification must be an essentially geometric one, so long championed by Einstein
in his search for a unified theory, has recently been coming back into favour. It
seems natural that as a prerequisite quantum mechanics itself should be cast in
geometrical language. Moreover there are good reasons for seeking to generalize
it, to free it from the restrictions of linearity just as general relativity has freed
space-time from the limitations of flatness [12, 15]. The geometrical structure
described in the present paper can easily be generalized to allow the space of
quantum states to be an arbitrary infinite-dimensional symplectic manifold.
Obviously, any viable generalization of quantum mechanics must reduce to
that theory for a wide range of phenomena. One of the main aims of this paper is
to discuss how this might happen. I shall show, on the basis of some rather natural
physical assumptions, that all the main features of conventional quantum
mechanics would emerge naturally for states that are in a suitable sense near the
vacuum, near enough to be represented by vectors in the tangent space.
In Sects. 2 and 3 we recall the main features of symplectic structures and
Hamiltonian dynamics, and introduce the quantum phase space Σ. For con-
ventional quantum mechanics, Σ is essentially a projective complex Hubert space
more precisely, a dense subspace thereof. However the formalism can be applied to
a much more general case, in which Σ is a real infinite-dimensional manifold with
(weak) symplectic structure. The dynamics is discussed in Sect. 3 in terms of a
Hamiltonian function E on Σ. For ordinary quantum mechanics E is the
expectation value of the Hamiltonian operator, and the evolution equation
reproduces Schrodinger's equation.
The concept of a symmetry is examined in Sect. 4 with particular emphasis on
the space-time symmetries associated with the Poincare group. Although an
eventual aim is the unification of quantum theory with gravity, for the present a
flat space-time is assumed. So too is the existence of a unique vacuum state v. The
tangent space Tv to Σ at the vacuum plays an important role.
The aim of Sect. 5 is to show that for states close enough to the vacuum to be
represented by vectors in the tangent space, the formalism necessarily reduces to
ordinary linear quantum mechanics. In particular, I shall show how, although Σ is
a real manifold, the complex structure of quantum mechanics would naturally
appear on the tangent space.
The conclusions are summarized in Sect. 6 and a number of unresolved
questions described.

2. The Quantum Phase Space

Axiomatic treatments of quantum mechanics often begin with the set of all mixed
states or ensembles [7, 11, 16, 17]. Pure states are regarded merely as extremal
Geometrization 191

elements of this convex set. However we shall deal exclusively with the set ίf of
pure states, and assume at the outset that all others can be expressed as
(presumably countable) mixtures of them.
Moreover instead of the Heisenberg-picture approach we shall use the
Schrodinger picture. We assume that, with respect to a given choice of time axis
each pure state can be represented by a path, or "history" in a space Σ of
"instantaneous pure states". For convenience, we reserve the term state for an
element of Σ; elements of £f will be called histories. Of course, the dynamics
establishes a one-to-one correspondence between elements of Σ at a specified time
and elements of ίf.
Although the formalism will be more general, it will be useful to begin by
discussing the structure of Σ in the special case of standard quantum mechanics.
Indeed this is the only space we shall actually need in this paper because the
generalized models considered use the same set of instantaneous states and differ
from the standard theory only in the dynamics. Later, however, more general
theories will be considered.
In classical mechanics, Σ is of course the phase space Γ, normally a finite-
dimensional manifold, but in quantum mechanics it is infinite-dimensional.
Essentially it is a projective Hubert space, the set of rays in a complex Hubert
space Jtf of state vectors. However there are technical problems associated with
the fact that the Hamiltonian operator H is unbounded. We have to choose
between two alternative formalisms, each with its peculiar advantages and
disadvantages. One is to work with the rays of ^ itself and accept that the
Hamiltonian vector field which specifies the dynamics is defined only on a dense
subspace [6]. The other, which we shall use here, is to work from the start with a
dense subspace ff of ffl equipped with a finer topology that makes H continuous.
The chief drawback of this second alternative is that Σ possesses only a weak
symplectic structure (see below). If we were trying to prove existence theorems for
solutions of the time-evolution equation, this might be a major defect, but for our
present purposes it is not important.
Let us assume then that there is a dense linear subspace $f QΪ 3F which forms a
common invariant domain for all the operators we wish to consider, including in
particular the Hamiltonian and other symmetry generators. For example, in a
nonrelativistic many-body theory we may take 3C to be the subspace of states
whose wave functions belong to some test-function space, say the Schwartz space
£f (see for instance [18]). In a field theory we might take it to be the subspace
generated by applying some algebra of observables to the vacuum state.
Physically, it is only states in this subspace that we can actually prepare, so JΓ is of
more direct physical relevance than #?..
The space jf is assumed to be equipped with a topology finer than that of J^
defined for example by a countable family of norms [18], which makes H and the
other symmetry generators continuous on Jf. It is possible to do this in such a way
that Jf becomes a Frechet-Schwartz space.
υ
Now let Jf ° be the set of all nonzero vectors in JΓ, and let <C be the
multiplicative group of all nonzero complex numbers. Then we choose as our
standard quantum phase space the projective space Z = JΓ°/<C°, which we shall
regard as a real manifold.
192 T. W. B. Kibble

For the general theory, all we shall assume is that Σ is a real infinite-
dimensional paracompact manifold modelled on some Frechet-Schwartz space Ύ*.
What this means [19] is that Σ is a Hausdorff topological space that can be
covered by a countable family of open sets UΛ on each of which a homeomorphism
φα is defined to an open set in ^ and that whenever UαnUβ=t0, the map φαφβl
restricted to φβ(UαnUβ) is twice continuously differentiable. In a general infinite-
dimensional non-Banach space, the definition of differentiability is quite proble-
matic [20]. However in the particular case of Frechet-Schwartz spaces, there is a
perfectly serviceable definition [21] which allows the construction of Ck manifolds
(see also [22, 23]).
It might not be unreasonable to require that Σ be a C°° manifold, but we shall
not need that assumption here.
In addition to its manifold structure, Σ is required to have a weak symplectic
structure. This means [6] that there is defined on Σ a closed two-form ω which is
weakly non-degenerate in the sense that if ωu(X9 Y) = 0 for all tangent vectors
YE TUΣ, then^ = 0. This form can be used to "lower" indices [1]. It defines a map
X\-^X]> from the tangent bundle TΣ to the cotangent bundle T*Σ : if X is any
tangent vector (or vector field) the corresponding cotangent vector (or differential
one-form) is
X* = iχω, (1)
where ix is the interior product (evaluation function) defined by

(Note that there is a difference of a factor 2 between conventions used here and in
references [1, 2, 6]. I follow the conventions of Choquet-Bruhat et al. [24]).
The map Xt-*X* is injective. However, ω is only weakly nondegenerate in the
sense that Xt-^17 is not in general bijective. There may be one-forms that are not
images of any vector field.
In the special case of ordinary quantum mechanics, ω is defined in terms of the
inner product. (As well its finer topology JΓ retains the inner product < , > of Jf .)
To define ω we need to introduce a local coordinate system in Σ. Let π be the
canonical projection of JΓ° onto Σ. Thus if u is any nonzero vector in JΓ, πu = u
denotes the corresponding state. Let v be any normalized vector in ctf. Then in a
neighbourhood of πveΣ we can represent states uniquely by vectors u in the
hyperplane

Thus we have a homeomorphism φ from this neighbourhood to an open set in JΓV.

We then define ω at πv by
ω πv (X,y) = 2Im<(^,<^y> (2)
for any X, Ye TπvΓ, or more generally at any point in this coordinate patch by

where X, YE TUΣ and Pu is the orthogonal projector onto the subspace normal to u.
Geometrization 193

This two-form ω is obviously skew-symmetric and weakly nondegenerate. But

while ω is readily seen to be closed, it is not exact. It is easy to construct a closed
surface over which its integral is nonzero, although locally ω = — dθ with
Im (dφu, φu)

This is another interesting difference between quantum and classical mechanics.

Classically, the phase space has the structure of a cotangent bundle and ω=—dθ
where θ is the canonical one-form θ = Σpdq. No such form exists globally on the
quantum phase space Σ.

3. Dynamics on the Quantum Phase Space

Let us now consider the specification of dynamics on Σ, assumed to be a real C2
Frechet-Schwartz manifold equipped with a weak symplectic structure.
The dynamics is described by a Hamiltonian flow, that is to say a one-
parameter group of diffeomorphisms τt:Σ-+Σ (with τ 0 = l and τt+s = τt°τs) which
preserve the symplectic structure, i.e.
τ*ω = ω . (4)
The τ f depend continuously and differentiably [21] on t and can thus be written
τt = exp tK ,
where K is a C1 vector field on Σ, which of course also leaves invariant the
symplectic structure :

£κω = 0 , (5)
where Lκ is the corresponding Lie derivative. (Had we adopted the alternative
approach mentioned in Sect. 2 we could not have required τt to be differentiable in
t and would have had to allow K to be defined only on a dense subset of Σ, see
[6].)
A history, i.e. an element of ^ is an integral curve of this flow in Σ, a curve c
everywhere tangent to K :

^ =*,,„, w
or equivalently
φ + ί) = τ f φ). (7)
By virtue of the identity Lκ = iκd 4- dίκ and the fact that ω is closed, it follows
from (5) and the definition (1) that
=0, (8)
i.e. X b is closed.
194 T. W. B. Kibble

In standard quantum mechanics, the space Σ defined in Sect. 2 is a simply

connected paracompact manifold, and so its first de Rham cohomology group is
trivial. (De Rham's theorem in general is hard to prove but here we need only the
comparatively straightforward result that a closed form in such a space is exact.)
This is an important feature, because it means that from (8) we can conclude that
K^ is exact, i.e. that a Hamiltonian function exists. We shall assume that the same
is true in any generalized quantum theory we consider, though it might indeed be
intriguing to consider a theory with a non-simply-connected quantum phase
space, a toroidal space say.
With this assumption it follows from (8) that there exists a Hamiltonian
function E, unique up to a constant, such that
K* = dE. (9)
It follows that any history is a solution of Hamilton's equations

Of course the function E uniquely determines the vector field K and therefore
the one-parameter group τt. What is much less obvious is whether, given E, any
such flow exists, or in other words to prove an existence theorem for solutions of
the differential Eq. (10). We shall not attempt to treat this problem here (see [6]).
In ordinary quantum mechanics, the function E is identified with the expec-
tation value of the Hamiltonian operator,

It is easy to verify that with the definition (3) for ω this reproduces Schrodinger's
equation. We seek a curve \p in 3f such that π°ψ = c. Equation (10) holds as an
equation for one-forms on Σ i.e. for arbitrary variations of c, but it can be pulled
back by π to yield an equation for one-forms on Jf°, i.e. for arbitrary variations of
t/λ The equation in fact involves only transverse variations, and thus yields

where P is the orthogonal projector defined in Sect. 2. Equivalently

. dψ(t)
dt

where α(ί) is an undetermined complex function. Its presence signifies our freedom
to multiply ψ by an arbitrary time-varying factor without changing c. Normally of
course we choose Im α = 0 to preserve normalization and fix Re α (equivalent to a
variable zero-point of energy) by some other convention.
A large class of generalized models can be described by this same formalism,
using the same set Σ and the same symplectic structure but with a more general
choice for the Hamiltonian function E. For example, let us consider as in [13] a
Geometrization 195

model of a scalar field φ. Let H be the Hamiltonian operator for the usual linear
4
quantum field theory with φ interaction. Let us take
2 2
£(«) = <Hχ + μKx<:</> (x):> (13)
2
in place of (11), where :φ : is defined by subtracting an (infinite) constant, as in
[13]. Then instead of (12) we obtain the Schrδdinger equation

2 2
^λ J d^:φ (x):yψ(t):φ (x):ψ(t) + a(t)ψ(t) . (14)

(We write <• •) for <• )πψ.) Except for the fact that the arbitrary time-
dependent factor is made explicit, (14) is precisely the evolution equation of one of
the models proposed in [13]. (There is also a difference of a factor of 2 in λ.) This
model can be thought of as one in which the particle mass becomes state-
dependent and position-dependent,

It is clear that many other generalized models can be constructed in the same
way. For example, if we add to (13) the extra term

we obtain a model in which both m 2 and the φ4 coupling constant g acquire a

state-dependence,

Not all the models of [13] can be expressed in this way because some of them
do not possess a conserved energy functional that can serve as our Hamiltonian
function E. However it seems natural to require the existence of such a function, so
this is not a severe restriction.
As we noted in the introduction, by formulating the theory on Σ rather than ffl
we automatically ensure the in variance under scaling transformations ip^λip
which was shown in [13] to be a necessary prerequisite for a consistent
measurement theory (see also [14]).

4. Symmetries
As a preliminary to the discussion of how conventional quantum mechanics might
emerge as a linear approximation to a more general theory, it will be useful to
examine the representation of symmetries.
A one-parameter group of (time-independent) symmetries is a group of
diffeomorphisms φs of Σ onto itself which leave invariant both the symplectic
structure and the Hamiltonian function,
φfω = ω, φfE = E. (15)
It follows that φs commutes with the time evolution τt. The generator of the group
is a C1 vector field X on Σ which also leaves invariant the symplectic structure and
196 T. W. B. Kibble

commutes with the generator of time evolution :

[X,K]=0. (16)
b
Just as for K itself, it follows thatX is closed, hence exact, so that there exists a
generating function Gx, unique up to an additive constant, such that
X» = dGx. (17)
If X and Y are symmetry generators, the effect of one on the generating
function of the other is
LXGY = G[XY], (18)

in the sense that the left hand side is a possible generating function for [X, Y] i.e.
this equation holds up to an arbitrary constant.
We can easily generalize the discussion to include time-dependent symmetries,
generated by vector fields that do not satisfy (16). Equation (18) is still valid if one
or other vector field is K.
We shall be interested in particular in space-time symmetries. We shall assume
that there is a realization by symmetries

of the connected component of the Poincare group, or rather 1 , if we wish to

accommodate fermions, of its two-fold universal covering group P. There is then a
corresponding realization by vector fields of its Lie algebra ^. Let us denote by Kμ
and Rμv the generators of translations and rotations, obeying the usual com-
mutation rules, for example

where ηλμ is the Minkowski-space metric tensor. Of course K0 is precisely the time-
translation vector field K introduced earlier.
The associated generating functions are the energy-momentum and angular
momentum functions,
P
μ =GKμ >
J
μv =GRμv '

P0 is the Hamiltonian function E. By virtue of (18) these functions transform in the

correct way under translations and rotations. For example,
Lκ μ
(19)

Next, let us assume the existence of a unique vacuum state veΣ invariant under
all operations of the Poincare group
τa>Λv = v. (20)

1 This point requires further justification. An argument for it was presented in an appendix to the
preprint version of this paper (ICTP/77-78/22), but cannot be included here for reasons of space
Geometrization 197

Equivalently
Kμ(v) = Q, Λ» = 0 . (21)
The uniqueness is not essential. We could easily accommodate a discretely
degenerate vacuum state. A continuous degeneracy, however, would be
unacceptable.
Because v is unique, it must be invariant under any symmetry. For any one-
parameter group φs of symmetries, φsv = v, or iϊX is the corresponding generator
*, = 0 . (22)
The generating function G = GX satisfies
dGv = 0. (23)
It is convenient to fix the arbitrary constants in generating functions by the
condition
G(υ) = Q . (24)
In particular
J» = 0, J» = 0 . (25)
Note that these conditions are consistent with the validity of Eq. (19) without extra
constant terms.
An important role will be played by the tangent space Tv = TVΣ to Σ at the
vacuum state. Because of (22) any symmetry generator X induces on Tv a linear
transformation X'υ which of course preserves the symplectic structure :

Because of (23) and (24), there is a quadratic approximation to G on the tangent

space. We may define the second derivative G"υ as a symmetric bilinear function of
vectors in Tυ. If c is any curve through v, with

then

From (17) there follows a relation between X'v and GJJ, namely
G'v=(X'vy (26)
in the sense that for all Y,Z<= Tυ,

In particular we may define the second derivative E"v of the energy function
E = P0. We shall need one further important assumption, relating to the positivity
of energy. (Remember that the zero point is fixed at the energy of the vacuum.) We
198 T. W. B. Kibble

shall assume specifically that the vacuum is a nondegenerate local minimum of the
energy function, i.e. that Έ'Ό is a positive definite bilinear function:
E^(Y,Y)>0 for all YΦO . (27)

Presumably the vacuum should also be a global minimum of E, but we shall not
need that condition here.

5. Complex Structure
As noted earlier, one of the most important questions to ask about any proposed
generalization of quantum mechanics is how the ordinary linear theory can
emerge as an approximation. The suggestion made here is that states that are, in a
sense to be defined, near the vacuum can be represented by vectors in the tangent
space Tυ, and that on Tv one has all the usual structure of linear quantum
mechanics, expressed of course in a particular local coordinate system like the one
used in defining ω.
To validate this suggestion, we have to do two things - to show how the
structures of linear quantum mechanics emerge on Tυ9 and to specify what is meant
by "nearness" to the vacuum. As far as the second question is concerned, no
general answer is possible it must depend on the specific model. In the generalized
field theories discussed earlier [13] the answer is reasonably clear. A state u is near
the vacuum if the expectation value <:φ2(x):>M is everywhere small compared to a
scale constant m2. For such states the nonlinear time-evolution equation can be
well approximated by its linear counterpart. Note however that there is no
guarantee that a state initially near the vacuum will remain so during its
subsequent history. In general, it will not.
In order for states near the vacuum to be represented by vectors in the tangent
space, one also needs of course a well defined map from some neighbourhood of v
in Σ into Tv. Here again, it is only in the context of a specific model that one can
expect to specify it. In our generalized models, it poses no problem because the
nonlinear and linear theories share the same space Σ and the same Tυ. In the linear
theory there is of course a natural map from Σ to Tv defined as in the definition of
ω in Sect. 2.
Let us now turn to the other outstanding problem, that of recovering the
structures of linear quantum theory on Tv. Much of the structure of course already
exists. In particular we have a linear representation of the Poincare group. What is
lacking however is the complex structure of ordinary quantum mechanics, and the
existence of a hermitean inner product.
A very important role in recovering this structure is played by the positive
definite symmetric bilinear function E'^. It is convenient for the moment to think of
it as defining a real inner product on Tv. (However, it is important to realise that
this is not directly related to the hermitean inner product we shall eventually
construct.) Thus Tv becomes a pre-Hilbert space.
Now any one-parameter group of time-independent symmetries φs leaves
invariant the energy function E, and so is represented on Tv by orthogonal
transformations φs^. In particular this is true of the time-translation operators τ fίis .
Geometrization 199

Consequently, the generator K'v is antisymmetric. We shall use it to introduce a

complex structure on Tv, in terms of a linear operator J satisfying J2 = — 1.
It is convenient (though not essential) to consider the complexified space
Tv = Tv@iTv and to define first a corresponding operator J on it. We may extend
K'v in an obvious way to a complex-linear operator K'v on Ύv. Then ίK'v is
essentially self-adjoint, and possesses a purely real spectrum. It may be regarded as
an energy operator. We seek to define an operator J which takes the value + i on
the subspace where ϊK'v is positive and — ί on that where it is negative, i.e.
J = isign(θg. (28)

There is a slight technical problem here. This construction is certainly possible

if the spectrum of iK'v does not include the point 0. If it does include it, J could still
be defined on the Hubert space obtained by completion of Ύv but there seems to be
no reason to suppose that it would necessarily leave Ύv invariant. Physically, the
condition that 0 is excluded from the spectrum corresponds to the requirement
that there be a nonzero minimum mass in the theory. For the moment we shall
assume that either 0 is excluded or, if it is not, that J can still be defined.
It is clear from its definition (28) that J commutes with the conjugation
operation X + iY^-X — ίY on Tϋ? and so defines a corresponding operator J on Tv.
In fact J could be defined directly on Tυ by setting
j^-x c-(κ ) 2 ]- 1 / 2 , (29)
where the positive square root of the operator is implied. It follows incidentally
that J commutes with all Poincare-group generators. Moreover
ω(JX,Y)=-ω(X,JY) . (30)

It may be helpful to describe one way of thinking about this operation in the case
of ordinary quantum mechanics. A vector XE Tv may be represented by a density
operator which in Dirac notation takes the form

where |v> is a representative of the vacuum, and <(v|x>— 0. Then J is defined by

Now that we have defined an operator J satisfying J2 = — 1 we may introduce

a complex structure on Tυ. We define a complex vector space Tcυ whose vectors are
in one-to-one correspondence X<->XC with those of Tv. T£ is
given a complex structure by defining
iX = (JX)c .
c
(31)

Note that Ύcv has half as many complex dimensions as Ύv.

In terms of the representation above, we may regard the passage to 7£ as
projecting out the "positive-energy" part, i.e.
200 T. W. B. Kibble

c
On T υ we can define a hermitean inner product by setting
2 <XC, 7 > = ω(X, JY) + iω(X, Ύ) .
C
(32)

It is easy to verify that it is indeed hermitean, linear in the second factor and
antilinear in the first.
On the tangent space we have then all the usual structures of linear quantum
theory, in particular the complex structure and hermitean inner product.
Moreover the symmetry generators are antihermitean, and multiplying by z makes
them essentially self-adjoint.
It should be remarked that the positivity of energy has played a vital role here.
Without it K'v would not have left invariant a positive symmetric form and so its
complexification would not have had a pure imaginary spectrum.

6. Discussion
The principal results of this paper are two. First, I showed in Sect. 2 that quantum
dynamics can be expressed in a simple and elegant Hamiltonian form in terms of a
symplectic structure on the manifold Σoϊ instantaneous pure states. Thus classical
and quantum mechanics, and generalizations thereof, can all be formulated in very
similar ways.
Second, I have shown that if one makes some rather natural physical
assumptions, such as existence of a unique vacuum and local positivity of the
energy, then in any of the proposed generalizations of quantum mechanics, the
conventional quantum theory will re-emerge as a linear approximation for states
near the vacuum. In particular, the complex structure appears naturally.
Many problems remain. An important feature of the method adopted here is
its use of the Schrόdinger picture, but this has the obvious disadvantage of lacking
manifest relativistic covariance. It would be useful to find a way of rewriting the
formalism in an explicitly covariant manner.
There are several technical points that need to be clarified. The mathematical
characterization of Frechet-Schwartz manifolds is not at present very simple. It is
not even clear that they constitute the correct class to choose. In particular the
cohomology of such manifolds is not well known and an existence theorem for
histories is needed. Also the introduction of complex structure in Sect. 5 required a
specific assumption about the spectrum of K'υ which needs to be eliminated or
better understood.
The only generalizations of quantum mechanics we have considered have used
the same quantum phase space Σ, but a different evolution law. However, one
might also examine alternative choices for Σ. In particular, it would be intriguing
to consider spaces with nontrivial first cohomology group.
Although the formalism was developed to permit generalizations of quantum
mechanics, it also provides an interesting starting point for axiomatisation of the
conventional theory. It would be useful to know what conditions imposed on the
symplectic structure and Hamiltonian function would allow one to identify the
theory with standard quantum mechanics.
Geometrization 201

One aim of this work was to rewrite quantum mechanics in a form better
suited to unification with general relativity, but so far we have considered only flat
space-time. The next step may be to examine what happens when both structures,
space-time and the space of quantum states, are simultaneously freed from the
restrictions of linearity. I hope to return to this question at a later date.
Acknowledgement. I am indebted to Dr. C. J. Isham and Mr. S. Randjbar-Daemi for criticisms of an
earlier version of this paper.

References
1. Abraham, R., Marsden, J.E.: Foundations of mechanics. New York, Amsterdam: Benjamin 1967
2. Maclane, S.: Geometrical mechanics (duplicated lecture notes). University of Chicago (1968)
3. Souriau, J.M.: Commun. math. Phys. 1, 374 (1966)
4. Souriau, J.M.: Ann. Inst. Henri Poincare A6, 311 (1967)
5. Kostant, B.: Quantization and unitary representations. Lecture notes in mathematics, Vol. 170.
Berlin, Heidelberg, New York: Springer 1970
6. Chernoff, P. R., Marsden, J. E.: Properties of infinite-dimensional Hamiltonian systems. Lecture
notes in mathematics, Vol. 425. Berlin, Heidelberg, New York: Springer 1974
7. Haag, R., Kastler, D.: J. Math. Phys. 5, 848 (1964)
8. Edwards, CM.: Commun. math. Phys. 20, 26 (1971)
9. Davis, E.B, Lewis, J.L.: Commun. math. Phys. 17, 239 (1970)
10. Birkhoff, G., Neumann, J.von: Ann. Math. 37, 823 (1936)
11. Varadarajan, V.S.: Geometry of quantum theory, Vol. 1. Princeton, NJ: van Nostrand 1968
12. Mielnik, B.: Commun. math. Phys. 37, 221 (1974)
13. Kibble, T.W.B.: Commun. math. Phys. 64, 73-82 (1978)
14. Haag, R., Bannier, U.: Commun. math. Phys. 60, 1 (1978)
15. Penrose, R.: Gen. Rel. Grav. 7, 31 (1976); 7, 171 (1976)
16. Jauch, J.M.: Foundations of quantum mechanics. Reading, Mass.: Addison-Wesley 1968
17. Piron, C.: Foundations of quantum physics. New York, Amsterdam: Benjamin 1976
18. Bogolubov, N.N., Logunov, A. A., Todorov, I.T.: Introduction to axiomatic quantum field theory.
Reading, Mass.: Benjamin 1975
19. Lang, S.: Introduction to differentiable manifolds. New York, London: Interscience 1962
20. Yamamuro, S.: Differential calculus in topological linear spaces. Lecture notes in mathematics,
Vol. 374. Berlin, Heidelberg, New York: Springer 1974
21. Kijowski, J, Szczyrba, W.: Studia Math. 30, 247 (1968)
22. Penot, J.-P.: Studia Math. 47, 1 (1973)
23. Lloyd-Smith, J.: Trans. Am. Math. Soc. 187, 249 (1974)
24. Choquet-Bruhat, Y., de Witt-Morette, C, Willard-Bleick, M.: Analysis, manifolds and physics.
Amsterdam: North-Holland 1977

Communicated by R. Haag

Received August 2, 1978; in revised form December 8, 1978