TUW - 92 - 13

Canonical Quantization of Non-Einsteinian

arXiv:hep-th/9211054v1 12 Nov 1992

Gravity and the Problem of Time

PETER SCHALLER∗
THOMAS STROBL†

Institut für Theoretische Physik

Technische Universität Wien
Wiedner Hauptstr. 8-10, A-1040 Vienna
Austria

submitted to: Classical and Quantum Gravity

Abstract
For a 1+1 dimensional theory of gravity with torsion different approaches to
the formulation of a quantum theory are presented. They are shown to lead to
the same finite dimensional quantum system. Conceptual questions of quantum
gravity like e.g. the problem of time are discussed in this framework.

Vienna, November 1992

∗
e-mail: schaller@email.tuwien.ac.at
†
e-mail: tstrobl@email.tuwien.ac.at
1 Introduction
The canonical quantization of general relativity is, though considerably simplified by the
introduction of the Ashtekar variables [1] and the loop variables [2], still an unsolved
problem. Among the main open issues are [3]: regularizing and solving the quantum
constraints to get (all) the physical states, finding all Dirac observables and the correct
inner product, interpreting the obtained theory and, as part of this, (re)introducing the
notion of space and time.
In [4] the action for 2–dim gravity with torsion was given:

γ β
L = e (− R2 + T 2 − λ). (1)
4 2
The classical solutions of the equations of motion were calculated in [4], [5] and a Hamil-
tonian formulation of the theory was provided in [6], [7].
Quantizing the Hamiltonian system derived from (1), one faces the same conceptual
problems as in a four-dimensional theory of quantum gravity. The calculations, however,
are much simpler. In particular, all classical solutions of the theory are known locally,
and, similarly to 2 + 1–dim gravity [8], the phase space is found to be finite dimensional.
Thus we think this theory to provide a good scenario for testing general concepts of
quantum gravity. This is the motivation behind the present paper.
Those aspects of the classical theory, which are important for the quantization of the
system are comprised in section 2. In order to have the presentation selfcontained and to
avoid confusion about the notation, material is included in this section which is already
contained in previous publications on the subject. But also new aspects are provided,
among them local analytic solutions around points of vanishing torsion, which are missing
in the literature. Furthermore we show that the phase space is two dimensional if one
assumes the space time manifold M to be of the form M = S 1 × R. For γλ < 0 and
spacelike S 1 the reduced phase space has a simple topology and a quantum theory on it
can be formulated easily (section 3.1).
In section 3.2 the Dirac method [11] for the quantization of constrained systems is
employed: The variables of the unconstrained phase space are quantized in a canonical
way. The space of physical wave functions is then identified with the kernel of the quan-
tized constraints. Finally an inner product is to be introduced in the space of physical
wave functions. We present two ways to achieve that: One proposed in [10] starts from a
measure in the unconstrained phase space, which is reduced by gauge conditions. Since
the Faddeev-Popov determinant turns out to be inadequate to guarantee gauge inde-
pendence in our case, the method is altered somewhat. No gauge conditions are needed
for constraints which act multiplicatively. The gauge conditions implicitely introduce an
internal time into the system. A somewhat different approach [3] defines the inner prod-
uct without any reference to a measure in the unconstrained phase space by requiring a
sufficiently large set of Dirac observables to be hermitian. We conclude the subsection
by applying also the simple quantization scheme used in [8] to quantize 2 + 1 dim gravity
with zero cosmological constant.

1
 Under the restriction M = S 1 × R with spacelike S 1 the quantum theories resulting
from the reduced phase space quantization and the Dirac method are shown to be equiv-
alent. In the Dirac approach, however, one need not know the topology of the reduced
phase space. The method is still applicable, if the above restrictions are lessened.
An elegant alternative is presented in section 3.3: The constraints are abelianized
before quantization. It is most remarkable that this is possible in a relatively simple way.
Exploiting the fact that the abelianized constraints may serve as canonical variables,
the quantization becomes simple now. The resulting quantum system is completely
equivalent to the one obtained in the preceding sections.
Physical questions usually refer to space–time events characterized by coordinates
µ
x . To answer them in a quantum theory of gravity they have to be reformulated in
terms of Dirac observables — the space time coordinates xµ enter as parameters. As
in the classical theory the choice of a coordinate system is equivalent to the choice of a
gauge. This idea is realized in section 4: A one parameter family of gauge conditions
allows to express gauge dependent quantities in terms of the Dirac observables. Once
the parameter in this family is interpreted as an intrinsic time, these quantities become
dynamical (i.e. time dependent). With respect to this dynamics, a Schroedinger picture is
formulated. The gauge independent formulation of the quantum states then corresponds
to the according Heisenberg picture. In either picture it is possible to predict quantities
such as hgµν (xµ )i or hT 2 (xµ )i.

2 The Classical Theory

In the Lorentz-bundle over the space-time manifold we will use the metric
!
0 1
ηab = , (2)
1 0

with a, b ∈ {+, −}. In (1) R is the Ricci scalar and T a is the Hodge dual of the torsion
two–form (T 2 ≡ T a Ta ). Thus they are (Lorentz vector valued) functions on the space
time manifold. A simple rescaling allows us to set β = 1 in the following.
The calculation of the canonical conjugate momenta to the components eµ ± and ωµ ,
µ ∈ {0, 1}, of the zweibein and the connection yields the relations (ϕ̇ = ∂ϕ/∂x0 , ∂ϕ =
∂ϕ/∂x1 )
∂L
πa := = −Ta
∂ ė1 a
∂L
πω := = γR
∂ ω̇1
and the primary constraints
∂L ∂L
a
= = 0. (3)
∂ ė0 ∂ ω̇0 a

2
It is remarkable that torsion and curvature serve as canonical momenta for the 1-
components of zweibein and connection. The canonical Hamiltonian is
Z
H=− e0 a Ga + ω0 Gω (4)

with the secondary constraints (ε+− = 1)

Ga = εab E e1 b − εb a πb ω1 + ∂πa ≈ 0 (5a)

Gω = εa b πa e1 b + ∂ πω ≈ 0, (5b)
in which we used the abbreviations
1 1
E ≡ (πω )2 − π 2 − λ, π 2 ≡ π a πa . (6)
4γ 2

Anticipating our later restriction to x1 ∈ S 1 , we have omitted a surface term in

(4) resp. (5). Because of (3) e0 a and ω0 act as mere Lagrange multipliers within the
Hamiltonian (4). They resemble lapse and shift in 4d gravity. The constraints (5) are
first class:
{Ga , Gω } = −εb a Gb δ (7a)
1
{Ga , Gb } = εab (−π c Gc + πω Gω ) δ. (7b)
2γ
Let us analyze briefly the geometrical meaning of the constraints. Gω (x1 ) generates
local Lorentz transformations, thus corresponding exactly to the Gauss constraint in the
Ashtekar formulation of general relativity: The first term of (5b), e1 − π− − e1 + π+ , is
the generator of O(1,1)–transformations in the 4 dim. phasespace (e1 a , πa ), the sec-
ond term reflects the transformation property of the connection ω under local Lorentz
transformations on a slice of the space time manifold with x0 = const. The combination

e1 a Ga + ω1 Gω = e1 a ∂πa + ω1 ∂πω (8)

of the constraints can easily be shown to generate diffeomorphisms δx1 = ǫ(x1 ) [cf. also
(4)], thus being the analogue of the vector constraint of the usual 3 + 1–theory. The
combination
1
π a Ga − E Gω = ∂(π+ π− ) − E∂πω = exp(−πω ) ∂Q (9)
4γ
with
Q := exp(πω ) [2γ π 2 − (πω − 1)2 − 1 + Λ], Λ ≡ 4γλ (10)
generates, up to local Lorentz transformations, diffeomorphisms in the direction of con-
stant curvature and torsion squared, since it is a polynomial in the momenta only. The
quantity Q defined in (10) has vanishing Poisson brackets with all the constraints, and
therefore it is a constant on any classical solution of the field equations [5], [6]. As a
consequence of this and (10) lines of constant curvature always coincide with lines of
constant T 2 .

3
 Another combination of the constraints, generating diffeomorphisms in the x0 direc-
tion, is the Hamiltonian (4). The gauge choice e0 + = 1, e0 − = 0, ω0 = 0 in (4) (light
cone gauge) identifies it with −G+ . As this choice corresponds to a space time metric
with g00 = 0, it is obvious that G+ generates diffeomorphisms in a lightlike direction (up
to local Lorentz transformations, again). A similar argument holds for G− . To complete
the analogy to the 3 + 1 dim. theory, one can choose an appropriate linear combination
of G+ and G− to play the role of the scalar constraint.
To find the local behaviour of a solution in the neighbourhood of a point P on the
space time manifold where T+ (P ) = −π+ (xµ (P )) 6= 0, we can satisfy the constraints
by inverting them algebraically to express π− , ω1 , and e1 + in terms of πω , π+ , and
e1 − , as well as Q(x1 ) = Q0 = const [because of (9) using (10) instead of G− ]. The
general solution in a neighbourhood of P can then be determined by applying −G+ to
the remaining three fields, generating the x0 dependence of the solutionR in the light cone
gauge. Thus integrating {π+ , G+ } = 0, {πω , G+ } = −π+ , and {e1 − , G+ } = −e1 − π+ ,
R R

we end up with

πω = π+ x0 + B(x1 )
π+ = A(x1 )
1
π− = [Q0 exp(−πω ) + (πω − 1)2 + 1 − Λ]
4γπ+
1
ω1 = − (e1 − E + ∂π+ )
π+
1
e1 + = (e1 − π− + ∂πω )
π+
e1 − = D(x1 ) exp(πω ), (11)

in which A, B, D [A(x1 (P )) 6= 0] are still arbitrary functions. Nevertheless, locally it

is possible to gauge them away: Not changing the values of e0 a and ω0 , it is possible
to obtain A(x1 ) = 1 by the choice of an appropriate Lorentz frame, B(x1 ) = 0 by an
x1 –dependent shift of the origin of the x0 -variable, and D(x1 ) = 1 by an x0 independent
transformation of the x1 variable. This ’normalized solution’ may be taken also to
represent a solution in the neighbourhood of a point P with π+ (P ) = 0, π− (P ) 6= 0 since
we still have not made use of the discrete Lorentz transformation which exchanges ’+’
and ’-’. Thus locally the space of solutions to the field eqs. is parametrized completely
by the value of Q0 [6], [9].
The solution in the neighbourhood of a point with π+ (P ) = π− (P ) = 0 can be
obtained by first calculating the flow of G− starting from P (we label this null line
x0 = 0):
π+ (x1 ) = E0 x1 , π− (x1 ) = 0, (12a)
πω (x1 ) = B(x1 (P )) = const and E(x1 ) = E0 = const being determined implicitely by
Q0 [cf. (10) and (6)]. For E0 6= 0 the Gauss constraint then yields

e1 + (x1 ) = 0, (12b)

4
whereas locally a Lorentz gauge representative of the solutions to (5a) is

ω1 (x1 ) = −E0 , e1 − (x1 ) = E0 x1 − 1. (12c)

To avoid e = 0 at x1 = 1/E0 one can also ’deform’ π+ (x1 ) into E0 arctan x1 , when
changing (12c) correspondingly. Finally acting with −G+ on (12) gives the (local) x0 –
dependence.
Note that according to (10) such solutions (and all solutions with points where π 2 ≡
2π+ π− = 0) are possible only for a certain range of Q0 –values. An analysis of (10) shows
the existence of a function h(Λ) with the following property: π 2 has zeros on M for
Q0 < h(Λ) and does not for Q0 > h(Λ); one finds h to be zero for Λ ≤ 1 and to increase
monotonically for Λ > 1.
A solution √
with constant curvature and vanishing torsion (de Sitter or Liouville so-
lution, πω = ± Λ) is included in (12); it is obtained by choosing
√ √
Q0 = QdeS := 2 exp(± Λ) (∓ Λ − 1) ⇔ E0 = 0. (13)

In this case the eqs. (12b), (12c) are not a consequence of the constraints, which are
trivially fulfilled, but they represent a (local) gauge choice along the line x0 = 0. Note
that (11) gives a different solution for Q0 = QdeS .
For the case Q0 6= QdeS (and Q0 within the range of the existence of Ta = 0) one may
construct the solutions (12) from (11), if one allows for zeros of A(x1 ) (but ∂A(P ) 6= 0)
and chooses B and D such that singularities are avoided. Since in the coordinate system
chosen above the curvature has the form πω = −E0 x1 x0 + B0 (Q0 ), points with vanishing
Ta are saddle points of the curvature (and vice versa). Local solutions around such points
have not been obtained in the previous literature; nevertheless, within the conformal
gauge saddle points of R appeared by gluing together solutions [9].
To do quantum mechanics purely local considerations are not sufficient. A complete
analysis, however, treating all the topological aspects of the theory is beyond the scope
of the present paper. Instead we will restrict our considerations to the case that the
space time manifold M can be written as M = S 1 × R, the S 1 being spacelike. Further
we will regard the case Λ < 0 so as to exclude Q0 = QdeS .
Let us prove that under these assumptions there are no solutions to the field eqs.
which allow for a zero of T+ or T− . The cylinder we are to consider may be covered
by one chart with periodic boundary conditions in the x1 direction. Now, having e.g. a
point P with π− (P ) = 0, the constraint equation (5a) as well as (13) and the existence
of a spacelike S 1 through P tell us that

∂π− (P ) = E0 e1 + (P ) 6= 0. (14)

Since π− is a periodic function in x1 there has to exist at least another point P’ on

this line x0 = const with vanishing π− and sgn ∂π− (P ′ ) 6= sgn ∂π− (P ). But then eq.
(14) for P and P’ implies that the sign of e1 + changes along the curve x0 = const; this
contradicts the assumption that it is spacelike. Let us note in parenthesis that when
regarding M = Σ × R1 with Σ = R1 spacelike, the above argumentation shows us only

5
that there cannot exist more than one (null) line with π− = 0 for Q0 6= QdeS ; the same
is true for π+ = 0. For Q0 = QdeS the above reasoning does not go through as (14) is
not true.
Thus in our topological setting we know all the classical solutions. They are given by
(11) with A(x1 ) 6= 0 and Q0 ≥ 0; the latter restriction comes from the requirement π− 6= 0
(cf. the paragraph following eq. (12)). But how can we possibly do quantum mechanics,
if there is only one parameter Q0 labelling the gauge inequivalent solutions? Actually
for the case of our topology there is a second one as a simple consideration shows: For a
particular fixed value R0 of R the metric on the space time manifold induces a metric on
the curve generated by R = R0 . As this curve is compact (cf. the first eq. of (11)), it has
a finite length. This length is (though R0 dependent) by construction gauge-invariant
and obviously not determined by Q0 . (We can e.g. change the interval of periodicity at
will without changing the integrand, which can be made x1 –independent). Thus there
is a quantity P0 > 0 characteristic for the ’size of the universe’.
This fact and that there are no further gauge invariant quantities can be seen also
from a more formal point of view: As before it is always possible to find a gauge such
that A(x1 ) = 1 and B(x1 ) = 0. But now, normalizing the interval of periodicity of x1
to [0, 1], a diffeomorphism x1 → f (x1 ), D(x1 ) → ∂f D(f (x1 )) cannot change the zero
mode of the arbitrary (periodic) function D(x1 ); therefore it is possible only to make
D constant: D(x1 ) = D0 =: −P0 /4γ. The identification P0 ↔ −P0 then is obtained
by the gauge transformation x1 → −x1 ; and P0 6= 0 since we required the S 1 to be
spacelike. Thus in our topological setting the space of solutions of the eq.o.m. (and thus
the reduced phase space of the theory) is a two parameter space:

πω = x0 , π+ = 1
1
π− = [Q0 exp(−x0 ) + (x0 − 1)2 + 1 − Λ]
4γ
ω1 = −e1 − E, e1 + = e1 − π−
e1 − = −4γ P0 exp(x0 ), (15)
with
Q0 ≥ 0, P0 > 0. (16)
In this gauge the quantity E, defined in (6), becomes:
1 0 2 1
E= (x ) − π− − λ = [−Q0 exp(−x0 ) + 2(x0 − 1)]. (17)
4γ 4γ
Because of g11 = 2(e1 − )2 π− the requirement ’S 1 be spacelike’ is compatible only with
γ > 0, whereas for the case γ < 0 there exist no such solutions to the field equations.
Requiring M = S 1 × R1 , the S 1 being timelike, on the other hand, one obtains (15) for
γ < 0 and no solutions for γ > 0 (Λ < 0). This result holds irrespective of any gauge as
is obvious from (5b) multiplied by e1 − /π+ and the fact that ∂πω cannot be definite on
a circle. As will be shown in sec. 4, furthermore, the evolution parameter x0 in (15) can
be taken to be purely timelike.

6
 For the case Λ ≥ 0 the requirement of the existence of a spacelike closed section
leads to (15) with Q0 = QdeS or Q0 > h(Λ), h having been defined in the paragraph
following eq. (12), as well as to the Liouville solution. A gauge representative of the
latter depends also on one (topological) quantity P0 , as can be shown by considerations
similar to the ones leading to (15). Although we do know all the classical solutions in
this extended situation, too, the construction of a consistent quantum theory on these
classical solutions seems hardly manageable due to the existence of a discrete part in the
spectrum of Q0 (cf. sec. 3.1 below) — except when assigning these points the measure
zero, certainly. Thus for Λ ≥ 0 the classical requirement that the S 1 be spacelike cannot
be maintained within a quantum theory.

3 The Quantum Theory

For the present model the simplest and most straightforward quantization is the reduced
phase space quantization (sec. 3.1). It makes use of the fact that we know all the classical
gauge inequivalent solutions (under the assumptions made in the preceding section).
Nevertheless, in order to gain insight into theories where not all the classical solutions are
known, such as general relativity, it is instructive to apply also other standard methods of
quantization like for instance the Dirac procedure (sec. 3.2). Canonical transformations,
obtainable also without knowing the classical solutions, can dramatically simplify the
task of quantization. This shall be illustrated in sec. 3.3, where we succeed in describing
the constraint surface by the vanishing of canonical coordinates.

3.1 Reduced Phase Space Quantization

Having the reduced phase space at our disposal, which is the quarter of a plane under
our assumptions [cf. (16)], we need to find the symplectic structure it inherits from the
unconstrained phase space. This can be achieved by first finding the Dirac observables
which correspond to Q0 and P0 in (15) and then by calculating their Poisson bracket.
Q0 is obviously just the constant mode of (10), i.e. (as we fixed the length of period-
icity to one, factors 2π are avoided)
Z
Q0 = Q. (18a)
S1
To find the gauge independent quantity corresponding to P0 , one first makes the last eq.
of (15) exlicit in P0 . Deviding the obtained expression by π+ , which is one in the gauge
of (15), it becomes Lorentz invariant. Integration of the resulting one–form yields the
diffeomorphism invariant quantity
1 e1 −
Z
P0 = − exp(−πω ) (18b)
4γ S1 π+
as our second Dirac observable, commuting (weakly) with all the constraints. Now it is
straightforward to verify that the symplectic form on the reduced phase space equals
dQ0 ∧ dP0 . (19)

7
However, P0 in (18b) is not invariant against the discrete transformation x1 → −x1 ,
which is not included within the (continuous) flow of the constraints. Thus the completely
gauge independent quantity corresponding to the normalized solution (15) subject to the
restriction (16) is actually q
| P0 | = P0∗ P0 . (20)
Now the quantization is quite simple. The commutation relations (19), or better the
corresponding Weyl algebra, as well as the first eq. (16) as a restriction to the spectrum
of Q̂0 yield an L2 (R+ ) with Lebesgue measure as our Hilbert space (cf. e.g. [14]). In
this Q̂0 acts as a multiplicative operator
q and P̂0 as the usual derivative operator (up to
unitary equivalence). And since it is P̂0∗ P̂0 which corresponds to the classical quantity
P0 in (15), we also have no problem with self–adjointness and the second restriction (16)
(as we would have with P̂0 ).
Note that there is so far no ’dynamics’ present in this formulation of the quantum
theory. As typical for theories formulated in a reparametrization invariant way our
Hamiltonian (4) vanishes so that a priori there is no (naive) Schroedinger eq. or also no
(naive) path integral. How and in how far we can introduce some notion of time into
the canonical framework above shall be discussed in sec. 4. The corresponding problem
in the path integral formulation shall be tackled elsewhere [13].

3.2 Dirac quantization

In this section we shall quantize the unconstrained phase space and then calculate phys-
ical wave functions as the kernel of the quantized constraints. The form of the primary
constraints (3) allows to simply eliminate the zero components of our fields. So we are
left with a phase space Γ spanned by the variables ω1 , e1 a , πω , πa . Since our constraints
(5) are linear in the coordinates but quadratic in the momenta, we will work in the
momentum representation:
δ δ
ω1 → ih̄ , e1 a → ih̄ . (21)
δπω δπa
Thus the quantum constraints become

Ĝω Ψ = Ĝ+ Ψ = Ĝ− Ψ = 0 (22)

with ([ , ]+ denotes the anticommutator)

δ δ
Ĝω = ∂πω + ih̄(π− − π+ ) (23a)
δπ− δπ+
1 δ δ
Ĝ+ = ∂π+ + ih̄( [E, ]+ + π+ ) (23b)
2 δπ− δπω
1 δ δ
Ĝ− = ∂π− − ih̄( [E, ]+ + π− ). (23c)
2 δπ+ δπω

8
As already proven in [7] the quantized constraints (23) form a closed algebra. This
is crucial for the consistency of the simple quantization scheme used here. Otherwise
we would rely on more elaborate techniques like e.g. BRST quantization (cf. also [15]
and sec. 5). Note that the first replacement in (21) breaks the local Lorentz covariance
present in the classical theory: whereas the lefthand side of that eq. transforms as usual
for a connection of the Lorentz group (π± → exp[±α(x)]π± ⇒ ωµ → ωµ − ∂µ α), the
righthand side remains unchanged. This is a feature which should prevail also in the
Ashtekar formulation of the 3 + 1 theory.
Up to purely multiplicative terms our quantum constraints contain only Lie deriva-
tives. Thus the calculation of the kernel of the constraint operators will simplify con-
siderably, if, instead of some of the momenta, we use other variables which commute
strongly with the classical constraints. Because of (16) our wave functions have their
support only in an area where π+ and π− are different from zero [cf. (10) and remember
Λ < 0]. In such an area the map from either π− or π+ to Q is bijective. Therefore to
start with we will write our wave functions as

Ψ = Ψ[Q(π 2 , πω ), π+ , πω ]. (24)

With this general ansatz the integration of the first two eqs. (22) is straightforward,
yielding
i 1
Z Z
Ψ = exp(− ∂πω ln |π+ |) exp( δ(0) πω ) Ψ̃[Q], (25)
h̄ S 1 2 S1

whereas the last eq. (22) becomes

∂Q Ψ̃[Q] = 0, (26)

as is also clear from (9) which is valid also in the quantum case. The δ(0) is understood
to be defined in an appropriate regularization. We could e.g. discretize the x1 variable
according to xi − xi−1 = l. δ(0) appears to be 1l in this regularization.
The operator ordering in (23) guarantees that the quantum constraints are hermitian
with respect to the Lebesgue measure [dπω ][dπa ]. We could avoid the δ(0) term by a
R

different choice of the operator ordering in (22): Putting all derivatives to the right, the
constraint algebra still closes and as the constraints vanish on physical states, they are
automatically hermitian in the physical sector, whatever operator ordering we choose.
We will find, however, that the δ(0) term plays quite a crucial role in the reduction of
the Lebesgue measure to an inner product in the space of physical states.
Starting from (24) with ′ +′ and ′ −′ exchanged, we obtain analogously
i 1
Z Z
Ψ = exp( ∂πω ln |π− |) exp( δ(0) πω ) Ψ̃[Q] (27)
h̄ 2
as well as (26). Due to the latter eq., which is equivalent to setting (10) equal to some
constant Q0 , it is obvious that the transition amplitude
iZ
exp( ∂πω ln |π 2 | dx1 ) = 1 (28)
h̄ S 1

9
so that the Ψ̃[Q] in (25) and (27) do indeed coincide.
Note that in the above considerations we made use of our restrictions on Λ and the
topology only when we restricted the support of the physical wave functions to positive
values of (10). Within the Dirac quantization everything else is the same also in the
completely general case: (25) and (27) fulfilling (26) give the general solution to (22)
on charts of the phase space with π+ 6= 0 and π− 6= 0, respectively. Because of (28)
they can be patched together to give the wave functions fulfilling (22) on all of the phase
space except for points with simultaneous zeros of π+ and π− . To extend this solution
to all of the phase space except for points with πa = E = 0 (⇔ Q = QdeS ) by a further
ansatz Ψ = Ψ[Q, πa ] does not seem to be so easy though: A calculation analoguosly to
the above ones yield a difficult differential eq. of first order, and to make a good guess
is aggrevated by the fact that the expected phase factor will definitely be not locally
Lorentz covariant due to (21) or (23a), as can be seen explicitely from (25) or (27).
However, our solutions (25) or (27) extend to a much more general situation anyway:
The phase factors in Ψ can be integrated as long as the torsion does not vanish on an
interval of the S 1 . Thus with (25) we exclude only functional distributions solving the
quantum constraints (22) such as
√
δ[πa ] δ[πω ∓ Λ], (29)
which obviously corresponds to the Liouville or de Sitter solution.
Still we have to define an inner product in the space of wave functions (25). To
this end we may first realize that Ψ∗ Ψ gives a factor x1 exp(πω (x1 )) and that the
Q

product of this factor and the formal Lebesgue measure [dπω ][dπa ] on the unconstrained
momentum space yields an expression being invariant under the classical flow of the
constraints. The integral of Ψ∗ Ψ with the Lesbegue measure, however, will diverge, as
the wave functions are roughly speaking constant in the direction of G+ and Gω . Note
that having implemented these two constraints G− is purely multiplicative and is of
no relevance for the considerations at this stage. G+ and Gω are the generators of a
non–abelian group [cf. (7a)], the (infinite) volume of which has to be ’devided out’ from
the integral. As this group acts freely and transitively on the (πω , π+ )–plane (or more
strictly speaking the half plane with positive values of π+ ), it is suggestive to restrict the
values of πω and π+ by the gauge conditions
π+ (x1 ) = c(x1 ) πω (x1 ) = t(x1 ). (30)
These gauge conditions may be realized by the introduction of δ[π+ − c] δ[πω − t] into
the measure. This expression is, however, not invariant under the flow of G+ and Gω .
Thus the resulting expression for hΨ∗ Ψi will become gauge dependent. In our simple
model this is not really disturbing, as the gauge dependence can be reabsorbed in the
normalization of the wave function. Nevertheless, to get insight into similar problems in
more complicated theories, it is interesting how a gauge independent measure can be con-
structed. The introduction of a Faddeev-Popov determinant, which is the determinant
of (δ denotes the delta function)
! !
{πω , G+ } {πω , Gω } −π+ 0
= δ, (31)
{π+ , G+ } {π+ , Gω } 0 π+

10
will not lead to a satisfactory result. To our mind this seems to be correlated to the
fact that the group generated by Gω , G+ does not allow for an invariant, non-degenerate
bilinear form on its algebra.
To find an invariant measure let us calculate the action of Gω and G+ on Ω =
[dπ+ ][dπω ]. We find [cf. (31)]
Z Z
{Ω, Gω } = Ω, {Ω, G+ } = 0. (32)

As this coincides with the transformation of π+ , it is obvious that the expression (1/π+ ) Ω
Q

and thus its dual (π+ )δ[π+ − c]δ[πω − t] is invariant. Realizing the constraint (9) by a
Q

further delta functional, we end up with

Z
[exp(πω ) π+ ] δ[π+ − c] δ[πω − t]δ[∂Q] Ψ̃∗ Φ̃.
Y
hΨ, Φi ∝ [dπω ][dπ+ ][dπ− ] (33a)
x1

Changing the variables of integration from π− (x1 ) to Q(x1 ) we find

Z
hΨ, Φi = dQ0 Ψ̃∗ (Q0 ) Φ̃(Q0 ), (33b)

with the normalized Ψ̃(Q0 ) ∝ Ψ̃[∂Q = 0, Q0 ]. Note that all divergent factors are com-
pensated by the transformation of the variable of integration. We may now remove the
regularisation introduced after (26) and remain again with a one dimensional quantum
mechanical system described by an L2 (R+ ) [or L2 (R)] with Lebesgue measure. A solu-
tion such as (29) could be also implemented at this stage when assigning some (arbitrary)
weight to the point(s) Q0 = QdeS . This does not seem very rewarding, though. To com-
plete the equivalence with sec. 3.1 we have to apply the Dirac observable (18b) to our
wave function (25); we indeed find:
h̄ d
Z
hΨ, P̂0 Φi = dQ0 Ψ̃∗ (Q0 ) Φ̃(Q0 ). (34)
i dQ0
The
q constraint P0 > 0 is then implemented such as in the preceding subsection (P0 ↔
P̂0∗ P̂0 ).
An alternative way to formulate an inner product in the space of physical wave
functions is to first recognize that there is a natural bijective map Ψ ↔ Ψ̃(Q0 ) between
this space and the space of functions over the variable Q0 . An inner product on this
space L2 (R+ ) [or L2 (R)] is then implicitely defined by the condition that a basic set of
Dirac observables, in our model Q̂0 and P̂0 , should be hermitian with respect to this
inner product [3]. Because of P̂0 := (h̄/i) (d/dQ0 ) [cf. (19)] the hermiticity requirement
obviously fixes the measure µ within the general ansatz
Z
hΨ, Φi = dQ0 µ(Q0 ) Ψ̃∗ (Q0 ) Φ̃(Q0 ) (35)

to be independent of Q0 ; so again we end up with (33b). Whereas this approach to

find an inner product is more straightforward than the first one, a generalization of

11
it to models (such as general relativity), where a basic set of Dirac observables is not
(yet) known, seems to be difficult. The approach leading to (33), on the other hand, is
applicable whenever one finds good gauge conditions.
Choosing t to be x1 –independent, it is suggestive to regard it as an ’intrinsic’ time.
In analogy to the classical case one can then denote this time parameter by x0 . With
this interpretation the second equation of (22) can be regarded as a kind of Schroedinger
equation [with a time dependent Hamiltonian — cf. (23b)], and the (time dependent)
coordinate transformation from Q(πa , πω = t) to the variable Q in (33) as a shift to a
Heisenberg representation of quantum mechanics (cf. also sec. 4 for more details). An
obvious generalization would be π+ = c+ (xµ ), πω = cω (xµ ). Due to our construction of
(33) the resulting quantum theories are independent of the choice of c+ and cω . To get
’physical’ results we can e.g. calculate the expectation value of the torsion: Plugging the
multiplicative operator π 2 into (33a) we find
1
hπ 2 i(t) = [hQ0 i exp(−t) + (t − 1)2 + 1 − Λ]. (36)
4γ
Although we obtained some nontrivial dynamics by reinterpreting and generalizing the
gauge choice (30) and although we could calculate e.g. (36), we are not yet ready to
determine hg11 i etc., for having not fixed the corresponding gauge freedom. This will be
done in sec. 4.
There is also another related approach [8] leading to the correct Hilbert space: Since
the constraints are linear in the coordinates, the momenta are transformed into momenta
under the action of the constraints. Thus we could regard the functionals Ψ[πω , πa ] on
the constraint surface modulo the flow of the constraints as the physical wave functions.
With the general ansatz (24) (π+ 6= 0) the Lie derivative of the constraints yield the
dependence on Q(x1 ) which reduces to the dependence on its zero mode due to (9). The
inner product is constructed as two paragraphs above.

3.3 Abelianization
It is well known that any system of first class constraints allows a formulation, where the
constraints are abelian. A system of canonical coordinates may then be found such that
the abelianized constraints are part of it. Unfortunately, in general this canonical coor-
dinates are defined locally only and they are non polynomial in the original coordinates.
Moreover, it is difficult to find them. For these reasons they are of minor practical use in
most systems. In our system, however, the abelianization will turn out a powerful tool.
Again let us first assume π+ 6= 0. We already know the quantity Q to commute
with all the constraints and ∂Q to be a linear combination of the constraints. It is thus
clear that Q will play a crucial role in the abelianization. As Q is a combination of
the momenta, it commutes with all the momenta and the Poisson bracket with any of
the coordinates on the configuration space yields a function of the momenta times the
delta-function. So take a configuration space coordinate, devide it by the function of the
momenta on the right hand side of its commutator with Q to end up with a canonical

12
conjugate; e.g. for e1 + one obtaines in this way:

1 e1 − 1
{ exp(−πω ) (x ), Q(y 1 )} = δ(x1 − y 1). (37)
4γ π+

There are no obvious canonical conjugates to the other constraints. But a glance at (31)
suggests to reformulate Gω and G+ by multiplying them with a factor (1/π+ ). So we are
led to
(ω̃1 , ẽ1 + , Q; πω , π+ , P ) (38a)
with
G+ Gω
ω̃1 = , ẽ1 + = − (38b)
π+ π+
1 e1 −
P =− exp(−πω ) . (38c)
4γ π+
Since ω̃1 and P are Lorentz invariant and commute with π+ they obviously commute
with ẽ1 + . Checking finally that also ω̃1 and P commute, we indeed find that (38a) forms
a complete set of canonical coordinates (in a region where π+ 6= 0).
In a region where π− 6= 0 we may, up to signs, exchange the role of ’+’ and ’-’ in the
above considerations. We thus find
G− Gω 1 e1 +
(− , , Q; πω , π− , − exp(−πω ) ) (39)
π− π− 4γ π−
to form a set of canonical coordinates.
Within our topological framework it is near at hand to further Fourier transform
Q(x1 ) and P (x1 ). This then completes the canonical splitting of our theory into the
gauge sector andR the Dirac sector, the latter being spanned by the conjugates Q0 =
1
Q(x ), P0 = S 1 P (x1 ). [Note that the zero modes of (38c) and the corresponding
R
S1
variable in (39) coincide due to (5b) and (10)]. The quantization is now obvious. Any
quantization scheme will lead to a system equivalent to that of sec. 3.1.

4 Space–Time and Observables

As the symmetries of a theory of gravity include diffeomorphisms in space and time,
any Dirac observable (i.e. any function on the phase space invariant under the action
of the constraints) is space and time independent. This is the reason for the lack of
any dynamics within the (classical) reduced phase space or the corresponding quantum
system (cf. sec. 3.1). In order to reintroduce the notion of space and time into the
theory we have to break the according symmetries. This is most easily done by gauge
conditions. Measurable quantities are then defined by the requirement to be invariant
under the remaining symmetries. There is, of course, some arbitrariness in the choice of
gauge conditions. This arbitrariness reflects the fact that different observers may have
different means to measure quantities.

13
 We have already seen an example in section 3.2: πω is a function on the space time
manifold. Under the restrictions specified in section 2 the lines where πω is constant
provide a foliation of space-time. From (15) (or also (11)) we find the leaves to be
spacelike. This might encourage us to choose πω as a time variable x0 :≡ t. Functions
on the constraint surface which depend on t and the Dirac observables alone, like e.g.
π 2 [cf. (10)], are invariant under the flow of those linear combinations of the constraints
which leave the gauge condition
πω − t = 0 (40a)
invariant. They thus are measurable quantities in this setting and we may calculate their
expectation values etc. for a given quantum state. In this way we regain results like (36).
Let us mention that the choice of a time variable t does not determine its flow (∂/∂t):
A (t–dependent) diffeomorphism in the direction of constant time will leave the choice
of time unchanged while varying the flow of time and thus the Hamiltonian generating
it. In our example the condition (40a) implies

{πω , H} = 1. (41)

With (4) we find that the values of the Lagrange multipliers e0 a , ω0 are restricted by
(41), but certainly not completely determined.
In order to quantize quantities like e.g. the components of the metric, we have to
fix the coordinate system of the observer by further gauge conditions. The form of the
canonical coordinates (38) suggests the choice

π+ = 1, ∂P = 0. (40b)

Due to (40a) and the first eq. of (40b), which have been implemented already within
the approach of sec. 3.2 for the special case c = 1, the second eq. of (40b) is equivalent
to ∂e1 − = 0. These gauge conditions together with our Dirac observables uniquely de-
termine all the quantities (ω1 , e1 a , πω , πa ) on the constraint surface Γ̂. A simple algebraic
manipulation yields (15) with x0 = t. (Note that the choice of good gauge conditions,
turning all first class constraints into second class constraints, saves one the integration
of the flow of the Hamiltonian; in more complicated systems this can be a decisive ad-
vantage). Antisymmetrizing these classical relations, we obtain a one parameter family
of hermitian operators on our Hilbert space:

π− (t) = (1/4γ) [Q0 exp(−t) + (t − 1)2 + 1 − Λ] (42a)

e1 − (t) = −4γ P0 exp t (42b)

e1 + (t) = (1/2) [e1− (t), π− (t)]+ (42c)
ω1 (t) = −(1/2) [e1 − (t), E(t)]+ (42d)
Thus we can now predict mean values, standard deviations etc. for g11 and ω1 .
The time evolution of the operator relations (42) is unitary. To show this let us first
specify the relation between (33a) and (33b). By means of the second eq. (9) we can,

14
instead of performing the shift of variables from π− (x1 ) to Q(x1 ), also directly integrate
out all the delta functions within (33a). This yields:
Z
hΨ, Φi = dπ− Ψ∗S (π− , t) ΦS (π− , t) (43)

with
t
ΨS (π− , t) ≡ exp( ) Ψ̃(Q0 (π− , t)), (44)
2
the latter function coinciding with Ψ̃ in (33b) and Q0 (π− , t) being the function at the
righthand side of (10) in the gauge (40); π− is the zero mode of π− (x1 ), projected out
within the inner product (33). Due to our gauge choice (40) (with t ≡ x0 ) we have
been allowed to drop the phase factors in (25), provided we do not consider derivative
operators of some higher order in πω and π+ . Since the inner product hΨ, Φi does not
depend on t [cf. (33b)], there exists a unitary operator U(t) satisfying

ΨS (π− , t) = U(t)ΨS (π− , 0). (45)

1
Because of ΨH (π− ) = ΨS (π− , 0) = Ψ̃(−Q0 + 2γ − λ), (33b) differs from the usual Heisen-
berg representation only by the trivial bijection
1
π− ↔ −Q0 + − λ. (46)
2γ
Since (42a) is the result of the transition from (33a) to (33b) applied to π− , it is ob-
vious that π− (t) is a unitary evolution in t. (It differs from the Heisenberg operator
(π− )H (t) = U ∗ (t)π− U(t), corresponding to the time independent operator π− in the
effective Schroedinger picture, just by this bijection).
In the Dirac-approach of section 3.2 we had to partially fix the gauge in order to define
an inner product in the space of physical states. It was not necessary to formulate a
gauge condition for the multiplicative constraint ∂Q = 0. Nevertheless, wanting to obtain
unique results for he1 − (x1 )i etc., this is necessary. The corresponding gauge condition is
implemented as an operator condition on the wave functions in this approach:
δ
∂P Ψ = Ψ = 0. (47)
δ∂Q
This guarantees that when operators such as e1 − (x1 ) act on Ψ only its (physical) de-
pendence on Q0 (π− , t) contribute to the result in (33a). Thus effectively the operator
e1 − (x1 ) acting on wave functions in the Schroedinger picture can be replaced by its con-
stant mode e1 − . The eqs. (18b) and (34) show that the time independent Schroedinger
operator e1 − is transformed into the righthand side of (42b) within the transition from
(33a) to (33b). Since the latter has been shown to be practically a Heisenberg picture,
the unitary evolution of e1 − (t) is also obvious. Now the unitarity of the remaining two
eqs. of (42) is a trivial consequence: (42c) and (42d) are the ’Heisenberg evolution’ of
(1/2)[e1 − , π− ]+ and −(1/2) [e1 − , ES (t)]+ , respectively. The operator ES (t) is given by
the first eq. of (17); it is explicitely time dependent in the Schroedinger picture.

15
 So, having chosen a gauge (40), there is a natural Schroedinger picture associated
to the operator evolution in the Ψ̃(Q0 )–representation. The wave functions (44) obey a
usual Schroedinger equation [cf. second eq. of (22) in our gauge] with a hermitian, time
dependent Hamiltonian
1
hS (t) = − [e1 − , ES (t)]+ . (48)
2
With (46) and the replacement e1 − ↔ −4γP0 in the above, we obtain the explicit form
of the evolution operator U0 (t) generating the time dependence of (42):

i t 2 − t′2 1
Z
U0 (t) = T exp(− P0 − [Q0 , P0 ]+ dt′ ). (49)
h̄ 0 2γ 2
To find this operator the Dirac approach of sec. 3.2 was very helpful. Clearly, in order
to have a (non–trivial) Schroedinger picture it has been necessary to break at least some
of the gauge symmetries of the reparametrization invariant theory.
To calculate the zero components of the connection and the zweibein, we have to
investigate the flow of time: with (41) and {π+ , H} = {∂P, H} = 0 we find

e0 + = 1 + f π− , e0 − = f (x0 ), ω0 = −f E, (50)

in which f (t) is an arbitrary function of t, π− is determined by (42a), and E by the

expression on the righthand side of (17). It is remarkable that a complete set of gauge
conditions in the (ω1 , e1 a , πω , πa )-space Γ does not fix the flow of time completely. Irre-
spective of how we choose f the classical solutions can be always brought into the form
(15) (under the assumptions specified there). There is also another perspective to see
this: Since (15) is independent of x1 it is invariant under a coordinate transformation
x1 → x1 + F (x0 ), which is the most general invariance of (15). This transformation, on
the other hand, induces e0 + → e0 + − e1 + Ḟ , and analoguosly for the other zero compo-
nents. Starting from the light cone gauge e0 + = 1, e0 − = 0, ω0 = 0, this transformation
yields again (50) [with f = −e1 − Ḟ ], when restricting it to the cross section Γ̄ of the
constraints and the gauge conditions. This generalization of the ’light cone gauge’ allows
for a strictly timelike flow of time: at least under our assumptions it is always possible
to choose f such that g00 > 0. It is interesting that through our prescription leading
to (50) also the ’Lagrange multipliers’ e0 a , ω0 became operators in the Hilbert space for
any gauge choice f 6= 0. Again different choices of f , which may be operator valued,
correspond to different observers.
The mechanism described above also works in the opposite direction: One may first
choose the values of the Lagrange multipliers. The according flow of time then restricts
but not completely determines the choice of a gauge in the phase space Γ. For instance
the light cone gauge leading to (11) does not allow for any x0 –dependence of π+ , the x1 –
dependence of this function, however, is still arbitrary. Having chosen some combination
of the constraints as our classical Hamiltonian H and being able to integrate the flow of
it, we may also introduce the affine parameter of H as an ’extrinsic time’ [16] (having
factored out the action of the other constraints). This approach seems straightforward
and suggestive, but it has the drawback that one has to be able to completely integrate

16
the equations of motion, what usually is not the case; moreover, any ’extrinsic’ time is
clearly equivalent to some ’intrinsic’ one introduced by gauge conditions.
It is a special feature of our system that there exists a gauge such that the x1 –
dependence of the solutions drops out completely. To obtain explicitely space–time
dependent operators we could choose a gauge

πω = x0 , π+ = A(x1 ), ∂e1 − = 0 (51)

in which A(x1 ) is an arbitrary nonvanishing periodic function, e.g. A(x1 ) = 2+sin(x1 /2π).
Analoguosly to above the gauge conditions (51) determine uniquely the fields of Γ̄ in
terms of the Dirac observables:
1 1 1
π− (x1 , t) = − [Q0 exp(−t) − (t − 1) 2
+ λ − ],
A(x1 ) 4γ 4γ
P0 e1 − (t) π− (x1 , t)
e1 − (t) = R exp t, e1 + (x1 , t) = ,
S1 A A(x1 )
∂A(x1 ) + E(t) e1 − (t)
ω1 (x1 , t) = − ,
A(x1 )
where E is given by the righthandside of (17), and restrict the gauge choice for the zero
components to:
1 + e0 − π− f (x0 ) e0 − E
e0 + = , e0 − = , ω0 = − .
A A A
Choosing some gauge for f (x0 ) and A(x1 ), we could now calculate hgµν (xµ )i, hωµ (xµ )i,
h∆gµν (xµ )i, h∆ωµ (xµ )i, etc.

5 Conclusion
We have succeeded in quantizing the model (1) under the assumption that the corre-
sponding classical solutions lead to a space time M of the ’physical’ form M = S 1 × R1
with spacelike S 1 and the assumption Λ ≡ 4γλ < 0 (in sec. 3.2 also under more general
assumptions). The simple structure of the phase space in this restricted model allowed
us to apply different methods of quantization, to compare the results, and to elucidate
conceptual problems of quantum gravity like the relation between Dirac observables,
gauge conditions, and measurable quantities — in a framework, where the connection
between classical and quantum expressions becomes very clear.
From our considerations in chapter 2 we conjecture that the quantization of models
with other values of Λ and with a more general topology will come down to the quan-
tization of some finite dimensional phase space with a more complicated topology. In
this more general framework the Liouville theory, which is the de Sitter solution of our
theory, would be included. Interesting questions like the one of topology changing could
be addressed. We thus think that a detailed analysis of the general theory would be
desireable.

17
 It would be also interesting to compare our results to still further methods of quan-
tization like e.g. the BRST quantization. In [15] the nilpotency of a quantum version of
ci Gi + c̄i C i jk cj ck has been shown. [Gi = (Ga , Gω ), C are the structure functions, and c,
c̄ the ghosts and antighosts, respectively]. The study of the cohomolgy problem of this
operator would be the next step. Another promising area for investigations seems the
coupling of (1) to matter fields [17].

Acknowledgements
The authors are grateful to W. Kummer for drawing their attention to the model
and reading the manuscript and to F. Haider for helping with the calculations leading to
(25). Furthermore they want to thank H. Grosse for valuable comments and H. Balasin,
M.O. Katanaev, P. Presnajder, D. Schwarz, and H. Urbantke for discussions.

References
[1] A. Ashtekar, Phys. Rev. Lett. 57, 2244 (1886); Phys. Rev. D36, 1587 (1887).

[2] C. Rovelli and L. Smolin, Phys. Rev. Lett. 61, 1155 (1888); Nucl. Phys. B331, 80
(1890).

[3] A. Ashtekar Lectures on Non–Perturbative Canonical Gravity, World Scientific, Sin-

gapore, 1991; A. Ashtekar, Summer School in Les Houches 1992.

[4] M. O. Katanaev and I. V. Volovich, Phys. Lett. B175, 413 (1886); Ann. Phys.
(N.Y.) 197, 1 (1890); M. O. Katanaev, J. Math. Phys. 31, 882 (1890).

[5] W. Kummer and D. J. Schwarz, Phys. Rev. D45, 3628 (1892).

[6] H. Grosse, W. Kummer, P. Prešnajder and D. J. Schwarz, Novel Symmetry of Non–

Einsteinian Gravity in Two Dimensions, J. Math. Phys. 33, 3892 (1892).

[7] T. Strobl, All Symmetries of Non–Einsteinian Gravity in d = 2, preprint TUW–92–

07, to appear in Int. J. Mod. Phys. A.

[8] E. Witten, Nucl. Phys. B311, 46 (1888).

[9] M. O. Katanaev, J. Math. Phys. 32, 2483 (1891); M. O. Katanaev, All universal
coverings of two–dimensional gravity with torsion, preprint.

[10] M. Henneaux, C. Teitelboim, Quantization of Gauge Systems, Princeton University

Press 1993.

[11] P. A. M. Dirac, Lectures on Quantum Mechanics, Yeshiva University, New York

1964.

18
[12] C. J. Isham in Recent Aspects of Quantum Fields, ed. Gausterer et. al. Lecture Notes
in Physics 396, p.123, Springer Berlin Heidelberg 1991.

[13] W. Kummer, P. Schaller, T. Strobl, in preparation.

[14] W. Thirring, Lehrbuch der Mathematischen Physik, Band 3, Springer Wien New
York 1979.

[15] N. Ikeda, K.I. Izawa, Quantum Gravity with Dynamical Torsion in Two Dimensions,
preprint.

[16] R.S. Tate, An Algebraic Approach to the Quantization of Constrained Systems: Fi-
nite Dimensional Examples, Thesis, Syracuse University, NY 13244, USA, 1992.

[17] W. Kummer, Deformed iso(2,1) Symmetry and Non–Einsteinian 2d–Gravity with

Matter, talk presented at Hadron Structure ‘92, Stara Lesna, Slovakia, Sept. 1992,
preprint TUW–92–12; W. Kummer, T. Strobl, in preparation.

19