Constrained quantization and -angles

N. P. Landsman and K.K. Wren

Department of Applied Mathematics and Theoretical Physics
arXiv:hep-th/9706178v1 25 Jun 1997

University of Cambridge
Silver Street, Cambridge CB3 9EW, U.K.
February 1, 2008

Abstract
We apply a new and mathematically rigorous method for the quantization
of constrained systems to two-dimensional gauge theories. In this method,
which quantizes Marsden-Weinstein symplectic reduction, the inner prod-
uct on the physical state space is expressed through a certain integral over
the gauge group. The present paper, the first of a series, specializes to the
Minkowski theory defined on a cylinder. The integral in question is then con-
structed in terms of the Wiener measure on a loop group. It is shown how
-angles emerge in the new method, and the abstract theory is illustrated in
detail in an example.

E.P.S.R.C. Advanced Research Fellow

1
1 Introduction
1.1 Classical reduction

In Diracs theory of constrained dynamical systems [1, 2, 3] the so-called reduced

phase space is generically obtained by a two-step reduction procedure. In summary,
the two steps of the reduction of a classical constrained system are

1. Imposing the constraints i = 0; this restricts the phase space of the uncon-
strained system S to the constraint hypersurface C.

2. Quotienting by the null foliation N0 of the induced symplectic form on the

constraint hypersurface C.

The reduced phase space is then S 0 = C/N0 . This often, but not always, coincides
with the phase space S phys of physical degrees of freedom.
Roughly speaking, the second step undoes the underdetermination of the equa-
tions of motion on C; in gauge theories with connected gauge group physically
equivalent points are identied by this step. Indeed, in a gauge theory (formu-
lated in the temporal gauge for simplicity) the constraints are given by Gauss law,
and quotienting by the null foliation amounts to collapsing each orbit of the iden-
tity component G0 of the (time-independent) gauge group G to a point; one has
S 0 = C/G0 . If G is not connected, one needs to include a further step in order to
arrive at S phys , viz. quotienting C/G0 by the discrete group 0 (G) = G/G0 . Thus
S phys = S 0 /0 (G) = C/G.
More generally, consider the case that a Lie group G acts canonically on S (that
is, the action preserves the Poisson bracket). In the absence of certain topological
obstructions [4, 5] this action is then generated by functions i (chosen relative to
a basis {Ti } of the Lie algebra g of G) on S, whose Poisson brackets reproduce the
Lie bracket in g; i.e., {i , j } = Cijk k . Each i plays the role of a charge, and it
often happens that constraints are given by i = 0 for such charges. This setting,
indeed, applies in the case of a gauge group [6, 3]; also see section 2.2 below. We
will refer to this situation as the group case; the associated reduction of S is known
as Marsden-Weinstein reduction [4, 5, 3].

2
1.2 Diracs quantum reduction

In trying to nd a quantum analogue of the classical reduction procedure, Dirac

[1] saw that only one of the two classical steps needs to be quantized. Let us
restrict ourselves to the case where all constraints are rst-class (this means that all
Poisson brackets {i , j } vanish on C); this special case is the heart of the matter,
and includes gauge theories. Assume that, through some construction, a Hilbert
space H is given as the quantization of the (unconstrained) classical phase space
S. Along with H, which serves as the quantum state space of the unconstrained
i on
system, suppose the classical constraints have been quantized into operators
H.
Dirac, then, proposed that the quantization of S 0 be given by

i |i = 0 i};
H0D := {|i H| (1.1)

that is, H0D is the subspace of H which is annihilated by the quantum constraints.
It inherits the inner product from H, so that it becomes a Hilbert space in its own
right, in which physical amplitudes may be computed. Consistency of this proposal
i,
entails that each commutator [ j ] must annihilate HD (i.e., the quantum theory

is anomaly-free).
i form a representation of the Lie algebra
In the group case it suces that the
g. This Lie algebra representation usually corresponds to a unitary representation
U of G, in which case the space H0D may alternatively be characterized as

H0D = {|i H| U(g)|i = |i g G0 }. (1.2)

This evidently leaves open the question how, in case G is disconnected, the Hilbert
space Hphys is to be dened.
As we see, there is no analogue of the quotienting step of classical reduction,
which would, in a way, render quantum reduction a simpler procedure than its
classical counterpart. The reader will now remark that the quantum BRST proce-
dure, at least in its operator version, does mimic its classical counterpart in being
a two-step procedure as well. This is not the place to point out at what cost this

3
is achieved [7]; the relevant point is that the rst step in quantum BRST leads to
problems entirely similar to the ones encountered in the Dirac approach.
Diracs proposal has particularly dominated the literature on canonical quantum
gravity and quantum cosmology, where the so-called Hamiltonian constraint implies
the controversial Wheeler-DeWitt equation. The diculties this equation leads to
are by now widely known and acknowledged [8, 9], although it is not always appre-
ciated that most of these are merely a special instance of general problems with the
Dirac (and operator-BRST) approach. The main diculties are:

It is very rare that all quantum constraints have 0 in their discrete spectrum,
i |i = 0 often have no
with joint eigenspace. In other words, the equations
solution in H. This situation usually occurs when the group generated by the
constraints is not compact.

If one seeks solutions outside H, one has to construct an inner product on the
space of solutions afresh. While this is possible in certain cases, there is no
good prescription as to which (generalized) solutions to include.

In quantum cosmology the last problem lies behind the discussion what the wave
function of the universe should be [10].

1.3 A new method of constrained quantization

In view of these diculties, and also for purely mathematical reasons, alternatives
to Diracs quantization procedure (or its BRST version) have been sought. We shall
here make use of one such alternative [11, 12]1 , whose essential idea is to quantize the
second rather than the rst step of classical reduction. This new approach turns out
to work even when the Dirac (or BRST) method breaks down, reducing to it in those
cases where it happens to apply. Also, one has a clean denition and construction
of (weak) quantum observables (see below).
In its simplest version, this idea is implemented by manipulating the inner prod-
uct h | i on H (which by denition is positive denite) into a sesquilinear form h | iphys
1
Related methods will be mentioned at the end of this subsection.

4
which is positive semidenite. The construction of this form is dictated by the con-
straints. The form h | iphys will have a nonempty null space

N = {|i H| h|iphys = 0}; (1.3)

the physical state space is then given by

Hphys = H/N . (1.4)

The inner product h | iphys on Hphys is the one inherited from h | iphys ; it is positive
denite by construction. If V : H Hphys is the canonical projection, one therefore
has
h|iphys = hV |V iphys . (1.5)

The Hilbert space Hphys is the quantization of S phys . There is no need to pass through
an intermediate space H0 (quantizing S 0 ), although it often provides insight to do
so.
The set of bounded weak quantum observables consists of those bounded opera-
tors B on H which are self-adjoint with respect to the manipulated inner product,
i.e., which satisfy
h|B|iphys = h|B|iphys (1.6)

for all , H (here the bar stands for complex conjugation). Without the sub-
script phys this would, of course, be the condition that B be Hermitian. A weak
quantum observable B maps N into itself, so that its induced action on the quo-
tient Hphys species a well-dened physical observable B phys . By denition, one
has
B phys V |i = V B|i, (1.7)

and this property completely species B phys as an operator on Hphys .

In practice h | iphys is often only well-dened on a certain dense subspace D H;
this happens precisely when the Dirac procedure breaks down. In that case the
above construction of Hphys undergoes only minor modications: the null space N
is now dened as a subspace of D, and the quotient D/N has to be completed in the

5
inner product h | iphys to obtain Hphys . With this renement, the key mathematical
problems in the Dirac or BRST approaches are avoided. All this even works if all
constraints are second-class; in fact, the classication of the constraints into rst-
and second-class ones is unnecessary in our procedure.
In this more general case, a weak quantum observable B is a possibly unbounded
operator whose domain contains D, and which leaves D stable. As in the previous
paragraph, when B is a weak quantum observable the induced operator B phys on
Hphys is well-dened, and represents a physical quantum observable.
Let us return to the group case, supposing that the quantum constraints generate
a unitary representation U(G) on H. The construction of the manipulated inner
product for this situation is explained in detail in [11, 12], with the following result.
For the moment we assume that G is connected. If G is compact, one has D = H,
and
Z
h|iphys = dg h|U(g)|i, (1.8)
G0

where dg is the Haar measure; for later reference we have written G0 for G to reect its
connectedness. This equals h|Pid|i, where Pid is the projector on the subspace Hid
of H which transforms trivially under U(G). Hence N = H
id , so that H
phys
Hid .
This coincides with HD of the Dirac method; cf. (1.2). A crucial property of the
manipulated inner product, which is immediate from the above, is

h|U(h)|iphys = h|iphys (1.9)

for all h G and all , H. According to (1.6) and (1.9), each U(h) is a weak
quantum observable, which by (1.9) and (1.7) is represented by the unit operator
on Hphys . This suces to prove that G0 acts trivially in the physical space Hphys .
If G is merely locally compact, but not compact, (and here assumed unimodular
for simplicity, so that the left- and right- Haar measure coincide) the expression
(1.8), and its consequence (1.9) still follows, but is only dened on a suitable dense
domain D H. The projection Pid and the space Hid no longer exist (so that
the Dirac approach would break down). However, one can successfully proceed as
indicated earlier.

6
 The case were G is not even locally compact, e.g. if G is a gauge group, will be
faced in the present paper (also cf. [13, 14]). It turns out that one can still make good
mathematical sense of an expression of the above kind, despite the non-existence of
Haar measures on innite-dimensional groups.
The idea of group-averaging in the context of constrained quantization goes
back, at least, to Teitelboim [15, 16]; it is, of course, common practice in lattice
gauge theory. The constrained quantization procedure proposed in [9] also involves
expressions of the type (1.8).

1.4 Discrete reduction and -angles

As already remarked, the case where the gauge group G is disconnected is exceptional
in that the reduced phase space S 0 = C/G0 (although symplectic) does not coincide
with the physical phase space S phys = S 0 /0 (G). The passage from S to S 0 rather
than S phys can be mimicked in quantum theory by restricting the integral in (1.8)
to G0 ; this leads to a Hilbert space H0 , which is the quantization of S 0 . The passage
from H0 to Hphys would then involve a step quantizing the passage from S 0 to
S phys . One could, of course, postulate (1.8) also for disconnected G, but this would
overlook an important option one has available at this point.
We isolate the issue at stake by looking at the classical reduction of an arbitrary
symplectic manifold S 0 by a discrete group D [17]. Unlike in general reduction,
there is only one step, namely the passage to the quotient S 0 /D. When D acts
freely on S 0 this is a manifold, but if it doesnt S 0 /D is typically an orbifold (locally
a manifold) [18]. More importantly, S 0 /D is symplectic (away from its possible
singular points), quite unlike a quotient by a connected Lie group, which would not
be symplectic, being merely a Poisson space [17].
Accordingly, there are no constraints, and C = S 0 . The absence of constraints
leads to a certain freedom in the denition of the manipulated inner product h | iphys .
Namely, having quantized S 0 by a Hilbert space H0 , carrying a unitary represen-
tation U 0 of D, as before, we could pick an arbitrary one-dimensional unitary rep-
resentation U of D (where stands for a collection of parameters labelling such

7
representations), and dene the manipulated inner product on H (or on a suitable
dense subspace) by
h|iphys = U (g)h|U 0 (g)|i.
X
(1.10)
gD

The freedom to include U is due to the fact that in discrete classical reduction the
0 in i = 0, which would force U to be trivial, is absent.
In case that D is nonabelian, the above limitation to one-dimensional represen-
tations U yields -parameters only if D/[D, D] is nontrivial. In the framework of
[11, 12] it is, in fact, entirely possible to work with general unitary representations
of D (also cf. [19]). If U is dened on some Hilbert space H , one denes the
manipulated inner product h | iphys on H H rather than on H (or, if need be, on
D H ) by the obvious generalization of (1.10), viz. by sesquilinear extension of
X
h v| wiphys = hv|U (g)|wih|U(g)|i, (1.11)
gD

where , H and v, w H . The construction of Hphys then proceeds as before;

the null space N is now a subspace of H H (or D H ). This is, for example,
relevant for braid group statistics.
Returning to the general group case, with G disconnected, we can proceed at one
stroke, avoiding the intermediate space H0 , by replacing (1.8) by
Z
h|iphys = dg U (g)h|U(g)|i, (1.12)
G

(g) = U GG/G0 is dened through a one-dimensional unitary represen-

where U
tation U of G/G0 ; here GG/G0 is the canonical projection from G to G/G0 . The
inner product on the physical Hilbert space of states Hphys will then depend on ,
although this -dependence is usually undone by a unitary transformation which
puts it into the physical observables.
This procedure is equivalent to the following one, which in eect breaks the pro-
cess up into two separate steps. Hence one mimicks the classical reduction process,
and obtains an algorithm that in practice is often easier to implement. One rst
constructs H0 using (1.8); this only uses the restriction of U(G) to G0 . Since G0
acts trivially on H0 by construction, the full representation U(G) on H quotients

8
to a representation U 0 (G/G0 ) on H0 . One then puts the manipulated inner product
(1.10) (with D = 0 (G)) on H0 , and proceeds to construct the physical state space
Hphys as before.
In other treatments of phase factors [20, 21, 22, 19], restricted to the case
of a multiply connected conguration space Q, one encounters the fundamental
group 1 (Q). To relate this to 0 (G), note that a multiply connected space Q may
be written as Q = Q/D, with D = 1 (Q), and Q is the universal covering space
of Q. By construction, 1 (Q) = e, and if Q is connected one has the equality
1 (Q) = 0 (D). Hence we choose S 0 = T Q, on which D acts by pull-back; the
reduced space is T Q/D T Q (see below). In the opposite direction, we could
start from some connected and simply connected space X (the conguration space
of gauge elds, which is ane, being a case in point) and reduce S 0 = T X by the
action of a discrete group D on X (pulled back to T X). Since (T X)/D T (X/D)
for discrete D, the above-mentioned approaches would look at this as the problem
of quantizing the multiply connected space Q = X/D. By the same argument we
have 0 (D) = 1 (X/D).
In all other respects our way of introducing -angles is profoundly dierent from
others, and one goal of this paper is to explicitly illustrate how these angles emerge
in a mathematically rigorous constrained quantization method. In the context of
gauge theories the two methods of explaining such angles that are best known to eld
theorists (cf. [23, 24] for reviews) are, so to speak, orthogonal to ours. Firstly, in the
(Euclidean) path-integral method, where the -parameter enters through instantons,
one does not integrate over the gauge group: it is precisely the goal of the Faddeev-
Popov gauge xing procedure to avoid such an integration. In our approach, on
the other hand, all eects come from this integration. Secondly, in the Hamiltonian
approach one postulates that physical states transform like U(g)|i = U1 (g)|i;
this generalization of Diracs condition U(g)|i = |i (cf. (1.2)) is avoided here, for
the same reasons that Diracs original condition is bypassed.

9
1.5 Gauge theory on a circle

We will illustrate the new technique for gauge theories on a circle T R/2Z.
These resemble topological eld theories in that the physical phase space is nite-
dimensional; see [25] for the abelian case and [26, 27, 28] for the compact nonabelian
case. In fact, the physical conguration space of a pure Yang-Mills theory on a circle
is G/ad(G) (that is, the space of orbits of the adjoint action of G on itself); the
derivation of this result by Marsden-Weinstein reduction will be reviewed below. For
connected G this space is dieomeorphic to T /W , where T is a maximal torus in G
and W is the associated Weyl group; see [27] for a rigorous derivation in the present
context, and also cf. Theorem 4.44 in [29] for the isomorphism G/ad(G) T /W .
This space is singular (but note that T /W is not an orbifold in the sense of [18]),
and some care is needed in the denition of the cotangent bundle T (G/ad(G));
with the correct denition this is the physical phase space S phys .
To quantize according to our method, we have to face the full complexity of
the problem of dening the integral in (1.8) or (1.12). It turns out that the correct
choice of the gauge group G is to include all continuous loops in G with nite energy;
in analogy with the situation on at space [34] we might call G the Cameron-Martin
loop group. This choice, however, leads to two (apparent) diculties.
Firstly, being innite-dimensional, the gauge group G has no Haar measure.
It turns out that, heuristically speaking, the would-be Haar measure dg on G
combines with a Gaussian factor in the matrix element of U(g) to form a well-dened
measure. This combination closely resembles the way the non-existent Lebesgue
measure on the space of paths in R3 combines with the exponential of the kinetic
term in the Euclidean action to form the Wiener measure appearing in the Feynman-
Kac formula (cf., e.g., [30]). Hence one obtains essentially the Wiener measure W
(conditioned on loops). The Wiener measure on a loop group has appeared in the
literature before in various dierent contexts; see [31, 32, 33].
The second diculty is, then, that G has measure zero w.r.t. W . While this
may appear paradoxical to physicists, it is simply the well-known phenomenon that
paths with nite energy are too regular to be supported by the Wiener measure.

10
Instead of integrating over G in (1.8) and (1.12), we therefore integrate over the
closure G of G in a natural norm. This closure is simply the space of all continuous
loops. The representation U(G) cannot be extended to G, but such an extension is
not needed to dene the manipulated inner product.
Using the Wiener measure on G, the manipulated inner product can be computed
explicitly, and the structure of the physical Hilbert space Hphys and the action of
physical observables may be derived.
In the present paper we perform this computation when the structure group G is
U(1). In that case the gauge group G of maps from T to U(1) T is disconnected,
with 0 (G) = 1 (G) = Z (the gauge transformations are labelled by their winding
number). As far as 0 is concerned this mimicks the case where space is S 3 and
G = SU(2) (for here 0 (G) = 3 (G) = Z).
The construction of Hphys for compact semi-simple G requires special techniques
and involves fascinating mathematics, which is beyond the scope of the present pa-
per; see [35] for a detailed treatment. As expected, the physical Hilbert space comes
out to be Hphys = L2 (G/ad(G)) (cf. [26, 36] for other approaches to the quantization
of the Minkowski version of this model, and [37, 38, 39] for the Euclidean version),
but the point of the derivation lies not so much in the result as in the method.

1.6 Acknowledgement

The authors are profoundly indebted to Brian C. Hall for patiently clarifying L.
Gross approach to the Wiener measure, allowing them to correct a highly misleading
statement in the rst draft of this paper.

2 Classical reduction
In this section we perform the Marsden-Weinstein reduction of the unphysical phase
space S of Yang-Mills theory on a circle T to the physical phase space S phys . We
assume that the structure group G is a connected compact Lie group, whose Lie
algebra is denoted by g. Without loss of generality we take the principal G-bundle
over T, which denes the classcial setting, to be trivial, i.e., P = TG. We formulate

11
the theory in the temporal gauge A0 = 0 from the start; this partial gauge xing is
entirely innocent, and allows us to regard the gauge group G as consisting of maps
from T to G. The conguration space A consists of certain functions from T to g.
The action g : A gA of g G on A A is given by

gA() = Ad(g())A() dg() g 1(), (2.13)

where Ad(x)A = xAx1 stands for the adjoint action of G on g, and T.

For the basic mathematical structure of gauge theories we refer to [40]; functional-
analytic aspects are covered by [41, 42, 43]. Refs. [40] and [42] also contain most
other mathematical prerequisites for this chapter.

2.1 Choice of the gauge group

It is necessary to be quite precise about the nature of the spaces G and A. The gauge
group G, whose choice dictates that of A, should not be too large, in that a space
containing discontinuous gauge transformations would not reect the topology of the
bundle P . On the other hand, it should not be too small, since gauge transformations
and connections that are too smooth cannot be used as the basis of a quantum
theory. We will choose G to be the largest subspace of the group of all continuous
loops for which both classical reduction can be successfully carried out, and the
unitary representation U(G) lying at the heart of the construction of the quantum
theory is well dened.
To dene G we recall that a compact Lie group has a Riemannian structure
obtained by choosing an Ad-invariant Euclidean inner product ( , )g on g Rn ,
and translating this from g = Te G to the tangent space of other points by the
group action. Hence for a curve : [0, 2] G (and in particular for a loop g)
we can dene the function |g|
: [0, 2] R. The space H1 (T, G) by denition
consists of those g C(T, G) whose (weak) derivative g is square-integrable in
L2 ([0, 2], R). (Here and in what follows, L2 ([0, 2], . . .) is dened w.r.t
that |g|
the Lebesgue measure d, as distinct from L2 (T, . . .) which is dened w.r.t. the
normalized measure d/2.) In particular, the Riemannian length of g H1 (T, G)

12
exists. It can also be shown that such a g is absolutely continuous, and that g exists
almost everywhere; see [44]. Physically, one could say that H1 (T, G) consists of all
continuous loops with nite energy.
An alternative characterization of H1 (T, G) is to take the dening representation
Ud (G) on Hd ; the space Md of matrices on Hd is a normed space, so that one can
dene the Hilbert space H1 (T, Mn ) as the completion of C (T, Mn ) in the p = 1
Sobolev norm. Then H1 (T, G) is the subspace of H1 (T, Mn ) consisting of those
functions which take values in Ud (G). This endows H1 (T, G) with the structure of
a Hilbert manifold (cf. [40, 42]). The continuous inclusion H1 (T, G) C(T, G) is
then a consequence of the Sobolev embedding theorem (cf. [42]), from which it also
follows that H1 (T, G) is not contained in any C p (T, G) for p > 0.
The gauge group is the Hilbert Lie group

G = H1 (T, G) (2.14)

with Lie algebra2

g = H1 (T, g). (2.15)

The group operations in G are pointwise multiplication and inverse; these are smooth
with respect to the Hilbert manifold structure of G.
For the last point see [42, App. A]. Here H1 (T, g) is dened analogously to
H1 (T, G); it is a Hilbert space under the p = 1 Sobolev inner product
Z
d (f (), g())g + (f(), g())
 
hf, gi1 = g . (2.16)
T

One has the inclusion H1 (T, g) C(T, g), and the pointwise exponential map on g
is continuous [42].
The connectivity properties of G are determined by the following result.
With the gauge group G defined as in 2.1, and the structure group G equipped
with its usual topology as a Lie group, one has

0 (G) = 1 (G). (2.17)

2
Generically Hilbert spaces are over the complex numbers, unless a real vector space is explicitly
indicated, as in L2 ([0, 2], g) or H1 (T, g), which are real Hilbert spaces.

13
 To put this in perspective, note that one usually considers the loop group LG =
C(T, G), equipped with the topology of uniform convergence (with respect to the
metric topology of G inherited from the Riemannian structure, or from G Ud (G)
as above). This topology coincides with the compact-open topology, so that one has

0 (LG) = 1 (G) (2.18)

by denition of 1 .
For example, if G = U(1) it follows that 0 (LG) = Z; the members of a given
component LGn , n Z, are labelled by the winding number of the loop. More
generally, 1 (G) is isomorphic to a discrete subgroup D of the center of the universal
covering group G of G (i.e. G = G/D). Under this isomorphism an element []
1 (G) is the equivalence class of loops in G which are homotopic to the projection
(from G to G) of a path from e to in G.
Using (2.18), we label the components LG of LG by D. Since the inclusion
G LG is continuous with respect to the manifold topology on G, (2.17) will follow
if each intersection G = G LG is connected in the topology of G; we write G0
for Ge . To prove this, by the reasoning in the previous paragraph it suces to show
that any two H1 -paths in G between e and are homotopy-equivalent in H1 , which
is obvious. Hence (2.17) follows.
An explicit description of a component G of G is as follows. Using the fact that
the exponential map Exp : g G is surjective for compact connected Lie groups
[29], we can nd a X g for which Exp(X ) = . If [x]D denotes the equivalence
class in G = G/D of x G, we have

G = G0 g , (2.19)

where g () = [Exp(X /2)]D . In other words, any element g() of G is of the

form
g() () = [Exp(() + X /2)]D , (2.20)

where g; in particular, (2) = (0).

For example, if G = T one has D = 0 (G) = 2Z; one usually labels elements
of T by [0, 2). The Lie algebra t of T as well as of its covering group T = R

14
is identied with R; then Exp : t T is given by Exp(X) = exp(iX), whereas
Exp : t R is the identity map. Hence R is of the form 2n; we then have,
with slight abuse of notation,
gn () = ein . (2.21)

Finally, we determine the appropriate space of connections A; our choice is the

same as the one in [27]. If g H1 (T, G) then dg g 1 H0 (T, g) = L2 ([0, 2], g).
Hence we choose A = L2 ([0, 2], g) (a real Hilbert space). It can be shown that the
action of G on A is smooth [42, App. A]. Since A is a Hilbert space, the cotangent
bundle is T A = L2 ([0, 2], g ) L2 ([0, 2], g). We write elements of S as pairs
(E, A), where E and A take values in g and g, respectively. The G-action on A
(2.13) lifts to a smooth G-action on S given by

g : (E, A) (Co(g)E, Ad(g)A dg g 1), (2.22)

where we have omitted the argument , and Co stands for the co-adjoint action of
G on g . Note that dg g 1 is not, in general, an element of g. The innitesimal
transformation generated by g is

: (E, A) (E + Co()E, A DA ), (2.23)

where Co() stands for taken in the co-adjoint representation, and DA = d +

[A, ] = d Ad()A. We may identify g with its dual g through the choice of an
inner product on g; then Co()E is replaced by Ad()E = [, E].

2.2 Marsden-Weinstein reduction

The procedure of Marsden-Weinstein reduction is well-dened also for innite-di-

mensional (strongly) symplectic manifolds [5]; see in particular [6, 3] for Marsden-
Weinstein reduction in the context of gauge theories. We here take S = T A, and
reduce with respect of the group action (2.22). The Poisson bracket on C (S) is
given by !
F G F G
Z
{F, G} = d , (2.24)
T Ea () Aa () Aa () Ea ()

15
where A = Aa Ta and E = Ea a in terms of a basis {Ta } of g and its dual basis {a }
of g . For the linear functionals F (A) = A(f ) = hf |Ai and G(E) = E(g) = hg|Ei
on S, where f, g A are smearing functions, (2.24) yields

{A(f ), E(g)} = hf |gi. (2.25)

In particular, {A(1), E(1)} = 2.

It is clear that the action (2.22) preserves this Poisson bracket, so that it is
canonical. A momentum map J is a function from S to the dual Lie algebra g ,
which by denition satises
{J , f } = f ; (2.26)

we write J for hJ, i, where g. Here f is the innitesimal variation under

(2.23), i.e.,
!
Z
f f
f = d Co()E() DA () . (2.27)
T E() A()

Hence a possible choice, and the one one we adopt, is

J (E, A) = hE|DA i, (2.28)

where the pairing is, of course, between L2 ([0, 2], g) and L2 ([0, 2], g). This mo-
mentum map is innitesimally equivariant in the sense that

{J1 , J2 } = J[1 ,2 ] . (2.29)

The charges mentioned in the Introduction are therefore minus the components
of the momentum map.
An elegant way to compute the reduced space S phys = J 1 (0)/G was given by
Rajeev [26], and was further elaborated in [27]. All results until the end of this
subsection are taken from these references; we merely add the Marsden-Weinstein
reduction perspective.
Dene a map W : L2 ([0, 2], g) C([0, 2], G) by W (A) WA , given by
 Z 

WA () = P Exp d A( ) , (2.30)
0

16
where P denotes path-ordering, so that WA () is indeed an element of G; note that

WA (0) = e, (2.31)

so that W takes values in the subspace Ce ([0, 2], G) of functions satisfying3 f (0) =
e. The path-ordered exponential is ultimately dened as a product integral; see [45],
and [27] in the present context. In our context, it coincides with the solution of the
dierential equation4 !

+ A WA () = 0, (2.32)

with initial condition (2.31). The map W does not quite map A A into the gauge
group, since WA (2) is not necessarily equal to WA (0).
Although elements of S are not necessarily dierentiable, the constraints J (E, A) =
0 for all g force E in (E, A) J 1 (0) to have the form

E() = Co(WA ())E, (2.33)

where E g on the right-hand side is constant. For abelian G this simply means
that E() = E is independent of . The expression (2.33) implies that (E, A)
J 1 (0) satises Gauss law DA E = 0 (and vice versa).
To see the eect of passing from J 1 (0) to S phys we look at the cotangent bundle
T G, which is canonically isomorphic to g G [5]. Dene : J 1 (0) T G by

(E, A) = (E(0), WA (2)); (2.34)

here E(0) coincides with the E on the right-hand side of (2.33), and WA (2) is
the Wilson loop. The adjoint action of G on itself lifts to the action y : (, x)
(Co(y), Ad(y)x) on T G. With respect to this lifted adjoint action, the map
intertwines the G-action on S with the G-action on T G in that g = g(0) ,
where the G-action on the left-hand side is given by (2.22). Since is onto, the
3
R
R In probability theory the map A WA is seen as the composition I 0 of the primitive
0
: L2 ([0, 2], g) C([0, 2], g) and Itos map I = P Exp : C([0, 2], g) C([0, 2], G); cf.
[31, 32, 33].
4
Itos map is defined in terms of a stochastic differential equation similar to (2.32) but deals
with much more general function A, which in our case is essentially the derivative of absolutely
continuous functions.

17
physical phase space is
S phys = (T G)/Ad(G). (2.35)

All physical observables that only depend on A are functions of the Wilson loop; such
observables dene a certain commutative C -algebra [46]. All physical observables
that polynomially depend on E are expressible in terms of the invariant elements in
the universal enveloping algebra of G; the simplest such element corresponds to the
energy
1 Z 2
h(E, A) = E , (2.36)
4 T
where the notation E 2 includes the trace (in the co-adjoint representation).
It goes without saying that for abelian G the adjoint action is trivial, so that
S phys = T G in that case.

3 Quantum reduction
3.1 Quantization of the unconstrained system

We quantize the unconstrained phase space S = T A by the standard method of

second quantization. Hence we complexify the real Hilbert space A = L2 ([0, 2], g)
to
AC = L2 ([0, 2], gC ), (3.37)

and consider the Bosonic Fock space [47]

nS AC ;
M
H = exp(AC ) = (3.38)
n=0

here nS AC denotes the symmetrized tensor product of n copies of AC .

Of special interest are the coherent states | exp Ai in H, dened for |Ai AC
by the norm-convergent series [47]

| exp Ai = (n!)1/2 n |Ai;
X
(3.39)
n

the notation is motivated by the property that

h exp A| exp Bi = ehA|Bi , (3.40)

18
where hA|Bi stands for the inner product in AC . The importance of these vectors
lies partly in the fact that one can conveniently dene a unitary representation of
the gauge group G by

1 1 1 1
U(g)| exp Ai = e 2 hdg g |dg g i+hg dg|Ai | exp gAi, (3.41)

where gA is dened in (2.13). The main term | exp (Ad(g)A dg g 1 )i illustrates
that this is the second quantization of the action (2.13) of G on A, the other terms
being present in order to guarantee that U is a unitary group representation. Various
unitarily equivalent versions of this representation may be found in the literature
[31, 32, 50], and have been used in the present context [36]; for a three-dimensional
version cf. [13]. For later use we record the matrix element

1 1 1 1 1
h exp B|U(g)| exp Ai = e 2 hdg g |dg g i+hg dg|AihB|dg g i+hB|Ad(g)Ai . (3.42)

For f, g AC the usual creation- and annihilation operators a(f ), a(g) satisfy
the canonical commutation relations (CCR) [a(f ), a(g)] = hf |gi; note that a(f )
is antilinear in f , whereas (by implication) a(g) is linear in g. The linear span is
contained in the domain of these operators, and (in the Fock representation) one
has

a(f )| exp Ai = hf |Ai | exp Ai. (3.43)

The linear functions A(f ) and E(g) in C (S) (see text after (2.24)) are quantized
by
Q(A(f )) = 21 (a(f ) + a(f ) ) (3.44)

and
Q(E(g)) = i(a(g) a(g) ), (3.45)

respectively. From (2.25) and the CCR we see that

i[Q(A(f )), Q(E(g))] = Q({A(f ), E(g)}), (3.46)

as desired in quantization theory.

19
3.2 Intermezzo: Wiener measure on the gauge group

The subsequent construction involved in the quantisation procedure will make use
of the properties of the (conditioned) Wiener measure W on G. This measure was
constructed in [31, 32, 33], and, like the Wiener measure on Rn , is closely related to
Brownian motion and the heat equation. This relation is not very important for our
purpose; instead, the most ecient way to dene W is the following method due
to L. Gross [51] (also cf. [52]). For the theory of promeasures and general measure
theory in innite-dimensional spaces we refer to the reviews [52, 53, 40]; another
good reference for this subsection is section 5 of [32].
Any real Hilbert space K has a Gaussian promeasure c dened on it, which is
characterized by its Fourier transform
Z
1 1
dc ()eih|i = e 2 Q() = e 2 h|i , (3.47)
K

where Q() = kk2 is the covariance of c . With this covariance, c is the canonical
Gaussian measure on H; in general, any positive quadratic form Q can be the
covariance. If H is nite-dimensional, c is actually a measure, given by

1 2
dc (x1 , . . . , xn ) = dx1 . . . dxn e 2 kxk .

In general, only so-called cylindrical functions can be integrated with respect to a

promeasure. A cylindrical function f on a Hilbert space is of the form f = F p,
where F is an integrable function on a nite-dimensional subspace, and p is the
orthogonal projection onto that subspace. Eq. (3.47) provides an example: here the
cylindrical function is eih|i . A more detailed discussion may be found in [52, 53, 40].
Given a measurable map f : M N between two measure spaces M, N the
image (or push-forward) of a measure on M is the measure f on N, dened
by f (E) = (f 1 (E)) for all measurable subsets E N. In case that M and N
are innite-dimensional vector spaces and is merely a promeasure, this denition
of f initially only applies to cylinder subsets E of N. It may happen that f
thus dened has a countably additive extension to the -algebra generated by the
cylinder sets in N, so that it can be extended to a measure on N. But even in that

20
case, the volume of a non-cylindrical set E N must be computed by approximating
it with cylinder subsets, even when f 1 (E) is a cylinder set in M.
This comment applies to the case at hand. In terms of the map W (see (2.30))
and the promeasure c on K = L2 ([0, 2], g), the image W c is initially a promea-
sure on Ce ([0, 2], G), which can be extended to a measure . The image E of
L2 ([0, 2], g) in Ce ([0, 2], G) under W (which is the subspace of continuous paths
with nite energy) is not a cylinder set, and its volume should be evaluated through
the approximation procedure mentioned above. It then comes out that (E) = 0,
despite the fact that c (W 1 (E)) = 1.
Let Cex ([0, 2], G) be the space of continuous paths in G which start at e and
end at x; we abbreviate this as Cex . For each x G a measure x on Cex is
dened by the desintegration (A) = dx x ( Cex ), where dx is the Haar
R
G

measure on G, and is a measurable subset of Ce ([0, 2], G). The special case e
is then a measure on the space Ce (T, G) of continuous loops in G which start (and
end) at e. If we embed G in C(T, G) as the space of constant loops, we clearly have
C(T, G)/Ce (T, G) = G (as groups) and C(T, G) = Ce (T, G) G as measure spaces.
This factorization nally allows us to dene the Wiener measure W on C(T, G) as
the product e H .
Let G = C(T, G) be the space of all continuous loops in G; this is the completion
of G in the supremum norm (see [51, 52] for the general theory behind such comple-
tions in measure theory). It is clear that W (G) = 1, whereas the comments above
imply that W (G) = 0. We summarize this discussion by
The Wiener measure W on the extended gauge group G is a probability measure,
defined as the push-forward of the canonical Guassian promeasure on the real Hilbert
space L2 ([0, 2], g) by the Wilson loop map W in (2.30), conditioned on the space
of loops. The gauge group G of loops with finite energy has volume zero w.r.t. the
Wiener measure.
An important property of W is its behaviour under translations; this was rst
established in [34] for the original Wiener measure on Rn , and was proved in the

21
present context of loop groups by [31, 33, 54]. It is
 
dW (gh) = dW (g) exp 12 hdh h1 |dh h1i hg 1dg|dh h1i , (3.48)

where g G and h G (the translation property cannot be extended to all h G).

Another important property is that the measure is invariant with respect to
g 7 g 1 on all G [33]. These properties, as well as the denition of W , are
consistent with the heuristic formula
!
dg dg
dg() exp h g 1 | g 1i
Y
dW (g) = N 1
2 , (3.49)
T
d d
where N is an innite normalization constant. This formula does not make math-
ematical sense, since the Haar measure dg() on G or G does not exist.
Q
T

Nonetheless, it is sometimes useful in guessing the results of certain calculations.

3.3 The manipulated inner product

We now turn to the construction of the manipulated inner product h | iphys . As

explained in subsection 1.4, we may proceed in two stages, and rst perform the
quantum reduction with respect to the connected component G0 of the identity.
In any case, we need to determine a dense domain D H on which h | iphys is
dened; here H = exp(AC ). Many dierent choices of D lead to the same physical
Hilbert space; a guiding principle is computational convenience. It turns out to be
appropriate to choose the following domain.
The domain D exp(AC ) consists of the finite linear span of all coherent states

of the form | exp Ai, where A AC .
Following the proof of Prop. 2.2 in [47], one can show that D is dense exp(AC ).
Moreover, D is stable under the action of U(g) for any g G. The advantage of this
choice will become clear later on.
It so happens that the representation U dened in (3.41) cannot be extended
from G to G. Nonetheless, eqs. (1.8), (3.42), and (3.49), suggest, and almost imply,
that we should dene the manipulated inner product on D by sesquilinear extension
of

Z
1 dg|Ai
h exp B| exp Aiphys = dW (g) ehg ehB|gAi . (3.50)
G

22
Since g 1 dg is not necessarily in L2 , the expressions hg 1 dg|Ai and hB|gAi should
not be interpreted as inner products in L2 , but as stochastic integrals [55, S4.5].
In the present case these stochastic integrals reduce to Stieltjes integrals (see [34]
for this remark). We shall not dwell on this point, except by saying that the fol-
lowing manipulations are all justied in the context of this more general notion of
integration.
In any case, the postulate (3.50) is jusitied by the crucial property (1.9) (now
valid on D), which follows from (3.41), (3.50) and (3.48). Like the translation
formula (3.48), this property holds for all h G. It is important that D is stable
under U(G), since otherwise the left-hand side of (1.9) would not be dened.

4 The abelian case

4.1 Small gauge transformations

We will now look at the simplest case G = U(1). First, let us reduce with respect to
the space G0 of small gauge transformations. In the abelian case the product (3.50)
simplies to

Z
hB|Ai 1 dg|ABi
h exp B| exp Ai0 = e dW (g)ehg . (4.51)
G0

Write g = exp(i), where (2) = (0) for g G 0 . By the denition of G0 , the

set {g 1dg|g G0 } forms the Hilbert space P0 L2 ([0, 2], R) (this would no longer
be true in the non-abelian case). Here P0 is the projection onto the the constant
R 2
functions; we write P0 A = A0 1, where A0 = (2)1 0 d A() is the zeroth Fourier
mode of A, and 1 is the unit function in L2 ([0, 2], R). The symbol P0 will denote
the projection orthogonal to P0 .
Using the denition of the Wiener measure as the push-forward of the Gaussian
promeasure under the map W (cf. (2.30) or (2.32)), the right-hand side of (4.51)
becomes
Z
ehB|Ai dc ()ehd|ABi ,
P0 L2 ([0,2],R)

where c is the canonical Gaussian measure on P0 L2 ([0, 2], R). This step is justied
because G 0 is a cylinder set in G.

23
 The integral itself (without the prefactor) is computed from (3.47), and yields
exp[ 12 (A B )2 ], where A = P0 A etc. All in all, we obtain

1 2 2
h exp B| exp Ai0 = e 2 (A +B ) e2B 0 A0 , (4.52)

where A2 = hA|Ai, etc.

By denition, the induced Hilbert space H0 is the quotient D/N of D by the
null space N of the manipulated inner product, completed in the inherited norm.
A trick allows us to realize H0 in a more concrete way.
Dene V : D L2 (R) by linear extension of

x2
!
1/2 12 A2
hx|V | expAi = () e exp + 2xA0 A20 . (4.53)
2

It follows from (3.40) and a Gaussian integration that

h|i0 = hV |V i, (4.54)

for all , L; cf. (1.5). Here the inner product on the right-hand side is obviously
the one in L2 (R).
This property is the whole point behind introducing the map V . For it follows
that the map V has the same null space N as the manipulated product h | i0, so
that the quotient D/N is given by the image of D under V . Since the image of V
is dense in L2 (R), the closure of V D is obviously L2 (R). We may therefore identify
this space with H0 .
Recall the denition of a weak quantum observable; cf. (1.6) etc. Analogously
to (1.7), the induced action B 0 of a weak quantum observable B on H0 is given by

B 0 V |i = V B|i. (4.55)

In the present situation notable examples of weak quantum observables, at least

with respect to the modied inner product dened by (3.50), are Q(A(1)), and
Q(E(f )) for all f AC ; see (3.44), (3.45). The weak observability of Q(E(f ))
is a consequence of (3.42), (3.50), and the fact that it commutes with all gauge
transformations U(g). In fact, a calculation similar to the one leading to (4.52)

24
yields

1 2 2
h exp B|Q(E(f ))| exp Aiphys = 2if 0 (A0 B 0 )e 2 (A +B ) e2B 0 A0 . (4.56)

Hence Q(E(f ))0 = 0 for all f P0 AC , as was to be expected on the basis of Gauss
law. Writing the energy (2.36) as a mode expansion E = En En /4, this means
P
n

that only the zero mode contributes, leading to

1  2 0

Q(h)0 = (a(1) a(1) ) . (4.57)
8 2

Furthermore, the Wilson loop WA (2) (see (2.30)) is quantized by

1
Q(WA (2)) = e 2 i(a(1)+a(1) ) . (4.58)

These operators are evidently constructed from a(1) and a(1) . From (4.55) and
(4.53) we obtain
d
Q(a(1))0 = x + . (4.59)
dx
Since the induction procedure preserves the adjoint of a weak quantum observable,
it follows that
d
Q(a(1) )0 = x , (4.60)
dx
Hence in terms of the usual Schrodinger position q = x and momentum p = id/dx
we have

Q(WA (2))0 = eiq ; (4.61)

Q(h)0 = 1
2 p2 (4.62)

from (4.58) and (4.57), respectively. These are unbounded operators on L2 (R),
initially dened on the linear span of the usual coherent states, where they are
essentially self-adjoint (cf. [48, 49] for the theory of unbounded operators on Hilbert
space).

4.2 Large gauge transformations

Having arrived at the intermediate Hilbert space H0 = L2 (R), we now complete the
quantum reduction by the full group G. As explained in subsection 1.4, the discrete

25
group 0 (G) = G/G0 = 2Z acts on H0 through a unitary representation U 0 . To
compute this action, we write U(n) for U(gn ) and note that according to (2.21) eq.
(3.41) specializes to

2
U(n)| exp Ai = en +2nA0 | exp (A n1)i. (4.63)

From (4.53), (4.63), and (4.55) we then infer that the corresponding realization
U 0 (n) on L2 (R) is simply

hx|U 0 (n)|i = hx + 2n|i. (4.64)

The one-dimensional representations of U (G/G0 ) discussed in 1.4 are here given

by
U (n) = ein , (4.65)

of Z is Z
where [0, 2) (note that the unitary dual Z = T; one could consider any

R, and nd that all -dependent quantities are periodic in with period 2).
We then apply (1.10), which, with a convenient normalization factor, now reads
Z
h|iphys in
X
= 2 e dx h|xihx + 2n|i. (4.66)
nZ R

This is well-dened on D = V D, which, we recall, is the linear span of all coherent

states in L2 (R). (Other domains, such as Cc (R) or the Schwartz space S(R) would
be equally suitable, and lead to the same result.)
One then repeats the procedure that led from H to H0 . In the case at hand the
second step of the quantum reduction procedure is closely related to the description
of the Aharonov-Bohm eect in terms of induced representations [56].

4.3 Intermezzo: induced representations revisited

More generally, whenever H0 is of the form L2 (G), for some locally compact group
G, and G/G0 is a closed subgroup of G which acts on H0 in the right-regular rep-
resentation, the reduction from H0 to Hphys is itself a special case of the theory of
induced group representations (in the sense of Mackey; cf. [57]) as reformulated by
Rieel [58]. In this more general situation one is given a closed subgroup H G

26
(where G and hence H are assumed to be locally compact) and a unitary represen-
tation U of H in a Hilbert space H (here is some label). These data lead to
a unitary representation U of G on some Hilbert space H , said to be induced by
U (H) [57]. As shown in [58], one can construct U and H as follows (also cf. [11]).
For simplicity we assume that G and H are unimodular, so that left- and right-Haar
measures are the same; xing a normalization, we denote the Haar measure on G
and H by dx and dh, respectively. This denes L2 (G) = L2 (G, dx). The coset G/H
then has a G-invariant measure dq, which denes L2 (Q) = L2 (Q, dq).
Choose a dense subset D L2 (G) as D = Cc (G), and equip L2 (G) H with the
manipulated inner product, dened by sesquilinear extension of
Z Z
h v| wiphys = dh hv|U(h)|wi dx h|xihxh|i, (4.67)
H G

where h | i is the inner product in H . The expression (4.67) is well-dened on

D H . Then choose a (measurable) cross-section s : G/H G, and dene
Vs : D H L2 (G/H) H by
Z
hq|Vs | vi = dh U (h)|vihs(q)h|i, (4.68)
H

where q G/H. A simple computation, using the invariance of dh and the property
Z Z Z
dq dh f (s(q)h) = dx f (x) (4.69)
G/H H G

for all f Cc (G), leads to

h|iphys = hVs v|Vs i (4.70)

for all , D H , where the inner product on the right-hand side is in L2 (Q)
H ; cf. (4.54).
Therefore, by the argument that followed (4.54), the induced space Hphys (which
is the closure of D/N ) dened by (4.67) may be identied with L2 (Q) H . The
induced representation U (G) is then dened by the property U (x)Vs = Vs UL ,
where UL (G) is the left-regular representation on L2 (G), tensored with the identity
acting on H .

27
4.4 The physical Hilbert space

Comparing (4.67) with (4.66), it is clear that this general scheme applies to the case
at hand: one has G = R and H = 2Z, so that G/H = T = R/2Z, and U = U
on H = C. The Haar measure on Z is taken to be the counting measure times 2,
and the induced measure dq on G/H is just the Haar measure on T. It follows that

Hphys = L2 (Q) H = L2 (T). (4.71)

We choose s : T R to be s() = for [0, 2), upon which (4.68) reads

h|Vs |i = 2 ein h + 2n|i.

X
(4.72)
nZ

The condition for an operator B on L2 (R) to induce a well-dened physical

operator B phys on L2 (T) is (1.6), with h | iphys replaced by h | iphys , given by (4.66).
Explicitly, this condition is equivalent to
Z Z
dx h|xihx + 2n|B|i = dx h|B|xihx + 2n|i (4.73)
R R

for all , L. The physical observable B phys is then given by

B phys Vs = Vs B. (4.74)

Condition (4.73) is satisifed by all dierential operators with constant coe-

cients, such as p and p2 , but not by the position operator q. Instead, one must
consider periodic functions of x with period 2 (acting on L2 (R) as multiplication
operators). The quantization of the Wilson loop (4.61) is a case in point. From
(4.74) and (4.72) we then obtain

h|Q(WA (2))phys |i = ei h|i. (4.75)

For any power of the Schrodinger momentum p (such as the energy (4.62)) we nd
the formal expression
!n
d
h|(pn )phys |i = i h|i. (4.76)
d

28
This brings us to a crucial aspect of our technique, namely the fact that our method
of constructing Hphys automatically selects a domain of denition for unbounded
weak quantum observables. This domain is the image under V (or, in the present
case, Vs ) of the original domain (assuming the latter to be contained in the domain
D of the manipulated inner product). In the present case the pn were initially
dened on the domain D (i.e., the linear span of all coherent states in L2 (R)). One
easily veries that the closure of pn coincides with the closure of pn dened on the
domain C (R) L2 (R). It then follows e.g. from Theorem 11.2.3 in [57] that pn is
essentially self-adjoint on D for all n.
The image D of D under Vs is the domain D , consisting of the smooth functions
C ([0, 2]) for which all derivatives (n) , n = 0, . . . satisfy the twisted boundary
conditions (n) (2) = exp(i) (n) (0). As in Example X.1.1 in [49] one veries that
(pn )phys is essentially self-adjoint on D for all n. This particularly applies to the
enery (4.62), where n = 2. Hence one obtains a uniquely determined observable
Q(h)phys , dened as the self-adjoint closure of 12 p2 (initially dened on D ). Its
eigenfunctions n , n Z, are h|n i = exp(i(n /2)), with eigenvalues En =
1
2 (n /2)2 . This is one way of seeing how the -parameter enters the physical
theory.

References
[1] P.A.M. Dirac, Lectures on Quantum Mechanics (Yeshiva University, New York,
1964).

[2] M.J. Gotay, J.M. Nester, and G. Hinds, Presymplectic manifolds and the Dirac-
Bergmann theory of constraints, J. Math. Phys. 19 (1978) 2388-2399.

[3] E. Binz, J. Sniatycki, and H. Fischer, The Geometry of Classical Fields (North-
Holland, Amsterdam, 1988).

[4] V. Guillemin and S. Sternberg, Symplectic Techniques in Physics (Cambridge

University Press, Cambridge, 1984).

29
 [5] R. Abraham and J.E. Marsden, Foundations of Mechanics, 2nd ed. (Addison
Wesley, Redwood City, 1985).

[6] J.M. Arms, Symmetry and solution set singularities in Hamiltonian eld theo-
ries, Acta Phys. Polon. B17 (1986) 499-523.

[7] N.P. Landsman and N. Linden, Superselection rules from Dirac and BRST
quantization of constrained systems, Nucl. Phys. B371 (1992) 415-433.

[8] C.J. Isham, Canonical quantum gravity and the problem of time, in Integrable
Systems, Quantum Groups and Qauntum Field Theories, eds. L. A. Ibort and
M. A. Rodrguez, pp. 157-287 (Kluwer, Dordrecht, 1993).

[9] A. Ashtekar, J. Lewandowski, D. Marolf, J. Mourao, and T. Thiemann, Quan-

tization of dieomorphism invariant theories of connections with local degrees
of freedom, J. Math. Phys. 36 (1995) 6456-6493 (gr-qc/9504018).

[10] N.P. Landsman, Against the Wheeler-DeWitt equation, Class. Quantum Grav.
12 (1995) L119-L123 (gr-qc/9510033).

[11] N.P. Landsman, Rieel induction as generalized quantum Marsden-Weinstein

reduction, J. Geom. Phys. 15 (1995) 285-319, (hep-th/9305088); Err. ibid. 17
(1995) 298.

[12] N.P. Landsman, The quantization of constrained systems: from symplectic re-
duction to Rieel induction, in Proc. XIVth Workshop on Geometric Methods
in Physics, Bialowieza, 1995, eds. S.T. Ali, A. Odzijewicz, and A. Strasburger
(Polish Scientic Publishers, Warsaw, 1997) (dg-ga/9601009).

[13] N.P. Landsman and U.A. Wiedemann, Massless particles, electromagnetism,

and Rieel induction, Rev. Math. Phys. 7 (1995) 923-958 (hep-th/9411174).

[14] U.A. Wiedemann and N.P. Landsman, The Stueckelberg-Kibble model as an

example of quantized symplectic reduction, J. Math. Phys. 37 (1996) 2731-2747
(hep-th/9508134).

30
[15] C. Teitelboim, Quantum mechanics of the gravitational eld, Phys. Rev. D25
(1982) 3159- 3179.

[16] C. Teitelboim, Explicit evaluation of group-invariant measure as by-product of

path integration over Yang-Mills elds, J. Math. Phys. 25 (1984) 1093-1101.

[17] J.E. Marsden, Lectures on Mechanics, LMS Lecture Notes 174 (Cambridge
University Press, Cambridge, 1992).

[18] I. Satake, On a generalization of the notion of a manifold, Proc. Natl. Acad.

Sci. 42 (1956) 359-363.

[19] D. Giulini, Quantum mechanics on spaces with nite fundamental group, Helv.
Phys. Acta 68 (1995) 438-469.

[20] L.S. Schulman, Techniques and Applications of Path Integration (Wiley, New
York, 1981).

[21] C.J. Isham, Topological and global aspects of quantum theory, in Relativity,
Groups and Topology 2, eds. B.S. DeWitt and R. Stora, pp. 1059-1290 (North-
Holland, Amsterdam, 1983).

[22] P.A. Horvathy, G. Morandi, and E.C.G. Sudarshan, Inequivalent quantizations

in multiply connected spaces, Nuovo Cim. D11 (1989) 201-228.

[23] R. Jackiw, Topological investigations of quantized gauge theories, in Current

Algebra and Anomalies, pp. 211-359, eds. S.B. Treiman et al. (World Scientic,
Singapore).

[24] S. Weinberg, The Quantum Theory of Fields. Vol. II: Modern Applications
(Cambridge University Press, Cambridge, 1996).

[25] N. S. Manton, The Schwinger model and its axial anomaly, Ann. Phys. 159
(1985) 220-251.

[26] S.G. Rajeev, Yang-Mills theory on a cylinder, Phys. Lett. B212 (1988) 203-205.

31
[27] S.G. Rajeev and L. Rossi, Some rigorous results for Yang-Mills theory on a
cylinder, J. Math. Phys. 36 (1995) 3308-3319.

[28] K.S. Gupta, R.J. Henderson, S.G. Rajeev, and T. Turgut, Yang-Mills theory
on a cylinder coupled to point particles, J. Math. Phys. 35 (1994) 3845-3865.

[29] A.W. Knapp, Representation Theory of Semisimple groups. An Overview Based

on Examples (Princeton University Press, Princeton, 1986).

[30] B. Simon, Functional Integration and Quantum Physics (Academic Press, New
York, 1979)

[31] S. Albeverio and R. Hoegh-Krohn, The energy representation of Sobolev-Lie

groups, Composito Math. 36 (1978) 37-52.

[32] I. B. Frenkel, Orbital theory for ane Lie algebras, Inv. Math. 77 (1984) 301-
352.

[33] M.-P. Malliavin and P. Malliavin, Integration on loop groups. I. Quasi invariant
measures, J. Funct. Anal. 93 (1990) 207-237.

[34] R.H. Cameron and W.T. Martin, Transformation of Wiener integrals under
translations. Ann. Math. 45 (1944) 386-396.

[35] K.K. Wren, Constrained quantization and -angles II, submitted to Nucl. Phys.
B.

[36] J. Dimock, Canonical Quantization of Yang-Mills on a Circle, Rev. Math. Phys.

8 (1996) 85-102.

[37] E. Witten, On quantum gauge theories in two dimensions, Commun. Math.

Phys. 141 (1991) 153-209.

[38] E. Witten, Two dimensional gauge theories revisited, J. Geom. Phys. 9 (1992)
303-368.

32
[39] A. Sengupta, The moduli space of Yang-Mills connections over a compact sur-
face, Rev. Math. Phys. 9 (1997) 77-121.

[40] Y. Choquet-Bruhat, C. DeWitt-Morette, and M. Dillard-Bleick, Analysis, Man-

ifolds, and Physics, 2nd ed. (North-Holland, Amsterdam, 1982).

[41] P.K. Mitter and C.M. Viallet, On the bundle of connections and the gauge orbit
manifold in Yang-Mills theory, Commun. Math. Phys. 79 (1981) 457-472.

[42] D.S. Freed and K.K. Uhlenbeck, Instantons and Four-Manifolds (Springer, New
York, 1984).

[43] W. Kondracki and P. Sadowski, Geometric structure on the orbit space of gauge
connections, J. Geom. Phys. 3 (1986) 421-434.

[44] W. Klingenberg, Riemannian Geometry (de Gruyter, Berlin, 1982).

[45] J.D. Dollard and C.N. Friedman, Product Integration (Addison-Wesley, Lon-
don, 1979).

[46] A. Ashtekar and C.J. Isham, Representations of the holonomy algebras of grav-
ity and non-abelian gauge theories, Class. Quantum Grav. 9 (1992) 1433-1467.

[47] A. Guichardet, Symmetric Hilbert Spaces and Related Topics (Springer-Verlag,

Berlin, 1972).

[48] M. Reed and B. Simon, Methods of Modern Mathematical Physics. I: Functional

Analysis (Academic Press, New York, 1972).

[49] M. Reed and B. Simon, Methods of Modern Mathematical Physics. II: Fourier
Analysis, Self-adjointness (Academic Press, New York, 1975).

[50] R.S. Ismagilov, Representations of Innite-Dimensional Groups (American

Mathematical Society, Providence, 1996).

[51] L. Gross, Abstract Wiener spaces, Proc. Fifth Berkeley Symposium on Mathe-
matical Statistics and Probability, Vol. II part 1, pp. 31-42 (Univ. of California
Press, Berkeley, 1967).

33
[52] H.H. Kuo, Gaussian Measures in Banach Spaces, Lecture Notes in Mathematics
463 (Springer, Berlin, 1975).

[53] C. DeWitt-Morette, A. Maheshwari, and B. Nelson, Path integration in non-

relativistic quantum mechanics, Phys. Rep. 5 (1979) 255-372.

[54] E.P. Hsu, Quasi-invariance of the Wiener measure on the path space over a
compact Riemannian manifold, J. Funct. Anal. 134 (1995) 417-450.

[55] T. Hida, Brownian Motion (Springer, Heidelberg, 1980).

[56] N.P. Landsman, Quantization and superselection sectors II. Dirac monopole
and Aharonov-Bohm eect, Rev. Math. Phys. 2 (1990) 73-104.

[57] A. O. Barut and R. Raczka, Theory of Group Representations and Applications

(Polish Scientic Publishers, Warszaw, 1977).

[58] M.A. Rieel, Induced representations of C -algebras, Adv. Math. 13 (1974)

176-257.

34