All Symmetries of Non-Einsteinian Gravity in D 2
All Symmetries of Non-Einsteinian Gravity in D 2
All Symmetries of Non-Einsteinian Gravity in D 2
All Symmetries
of Non-Einsteinian Gravity in d = 2
Thomas Strobl∗
Institut für Theoretische Physik
Technische Universität Wien
Wiedner Hauptstr. 8-10, A-1040 Wien
Austria
Abstract
∗ e-mail: [email protected]
1 Introduction
One of the unsolved problems of quantum field theory is to find a quantized ver-
sion of general relativity. A promising step in this direction is the investigation
of some simpler models in lower dimensions, which was one of the reasons1,2 for
considering the Lagrangian
γ β
L = − e ( Rabµν Rabµν + T aµν Taµν + λ) (1)
4 4
in d = 2. (1) is an example of a theory with quadratic terms in curvature and
torsion. In two dimensions it is the unique diffeomorphism invariant action, up
to surface terms, yielding second order differential equations for the zweibein
eµ a = gµ a and the spin connection ω a bµ . Supplementing it e.g. by the Einstein–
Hilbert term eR does not change the classical field equations since it is a trivial
total divergence here. In Katanaev et. al.1 the integrability of (1) was proven
using the conformal gauge, thereafter Kummer et. al.2 showed that the light–
cone (LC) gauge enables one to find the explicit form of the general solution.
Nevertheless, although the LC gauge is ideally apt to solve the field eqs. (cf.
also sec. 2 below), the study of properties concerning geodesics, such as global
completeness3 , is preferably done within the conformal gauge.
It is one of the purposes of this paper to show that the choice of the LC
gauge, as intuitively introduced in Kummer et. al.2 , is almost compelling when
starting from the covariant form of the field equations (sec. 2) and even more
so when using the Dirac–Hamiltonian formulation (sec. 3). As a by–product of
this we will find that there is no direct analogue of decoupling field eqs. for the
Euclidian version of the theory, except possibly when using complex coordinates
on an intermediary level. In contrast to previous work2,4 we shall try to stay as
covariant as possible in all of our calculations, keeping e.g. the covariance in the
Lorentz indices within the Hamiltonian formulation. In sec. 3 we also propose
a solution to the problem of surface terms on the Hamiltonian level11 , at least
for the case of this model.
The integrability of the system inevitabely raises the question of symmetries.
In a recent publication4 it was related to the fact that the Poisson bracket
relations between the first class (FC) constraints and the momenta may be
written as a deformed iso(2, 1)–algebra, which was said to correspond to some
’novel symmetry’ visible only on the Hamiltonian level.
It is, however, the point of view of the present paper that the Poisson bracket
relations between the constraints (and momenta) alone are of little relevance for
the integrability and the symmetries of the system. Due to a general theorem5
it is always possible to reformulate the first class constraints such that their
Poisson brackets vanish (strongly).b Then the information about the symmetries
b Such an abelianization of the contstraints of the present model has been actually accom-
plished in Ref. 7.
1
is completely hidden in the specific form of the constraints as functions on phase
space. To our mind the real content of the symmetries corresponding to FC
constraints becomes visible only when being translated to the Lagrangian level.
In sec. 4 we will show that there is a one–to–one correspondence between the
symmetry transformations generated by the FC constraints on the Hamiltonian
level and the diffeomorphisms and local Lorentz transformations leaving the
action (1) invariant. With the results of Henneaux et. al.6 it follows, moreover,
that there are no further local symmetry transformations hidden in the model.
The Hamiltonian formulation of sec. 3 also serves as a preparation for a
non–perturbative quantization of the model. In sec. 5 we will show that the
constraint algebra has no quantum anomalies. Therefore Dirac’s procedure of
imposing the constraints on a Hilbert space should be applicable. Some of the
related problems will be addressed in sec. 6; they will be tackled in a further
publication7 .
γ β
L = − e [ R2 + (−1)M T 2 + λ] (2)
4 2
with T 2 ≡ T c Tc . Using Cartan’s structural equations, the second one of which is
2
R = −2 εµν ωµ ,ν in d = 2, and relations such as δe = e ea µ δeµ a , ea µ = ga µ being
the inverse of eµ a , the variational principle yields for (2) (covariant derivatives
are denoted by indices after a semicolon):
with
γ 2 (−)M β 2
E ≡ R + T − λ. (5)
4 2
As a natural extension of the usual variational principle in classical point me-
chanics we have required the variation of the fields eµ a and ωµ to vanish at
the boundary ∂B of the parameter area B so that we were allowed to drop the
surface term
Z
e εµ ν [−γ R δων + (−)M+1 β Ta δeν a ] dxµ . (6)
∂B
Since (3) and (4) are tensor equations all Latin indices can be replaced by Greek
ones. The field equations have to be supplemented by the additional geometric
requirement e ≡ det eµ a 6= 0.
The equations (3), (4) immediately show p that for sgn(λ) = sgn(γ) there
exists a solution with T a ≡ 0 and R ≡ ± 4λ/γ, which has been called the
deSitter solution2 .
In order to find the complete solution, we write the equations of motion
(e.o.m.) explicitely in the ea –basis, still for an arbitrary (constant and normal-
ized) reference metric gab :
3
g0̂0̂ = 0. Staying with a real tangent space, this can only be achieved for the
case M = 1. Thus, at least without introducing complex coordinates, the field
eqs. corresponding to the Euclidean version of (1) cannot be decoupled as easily
as in the Minkowskian case. Therefore in the following we will consider only
M = 1.
What we have demanded so far is already equivalent to the LC gauge as
introduced in Kummer et. √al.2 . To show this, we first note that the coordi-
nate change2,4 x± = (1/ 2)(x0 ± x1 ), performed before having introduced
any gauge, comes down to a mere relabelling (since it is a diffeomorphism).
Further one notices that the above requirement e0̂ µ = δ(0µ) implies for the
inverse matrix, i.e. the zweibein, e0 a = δ(0̂a) and vice versa. This last equation
still corresponds to a reference metric with an arbitrary g1̂1̂ , but there exists a
(global) frame transformation from this gab to a light–cone metric
0 1
gab = (11)
1 0
so that e0 a = δ(0̂a) remains unchanged. This would be e.g. not the case, if it
originally corresponded to gab = diag(1, −1). Finally, ωab0 = 0 is invariant under
any global frame transformation (cf. (52) below). So, the natural requirement
of replacing a covariant derivative by a normal one and the requirement of an
obvious decoupling of the covariant field eqs. (3),(4) or (7) to (9) automatically
leads to the intuitively introduced light–cone gauge of Kummer et. al.2 .
At this point let us observe that the above gauge already fixes the space–time
character of the coordinate x0 to be a light–cone coordinate: g00 = e0 a e0 b gab =
g0̂0̂ = 0. So when referring the zweibein to (11), it is not so severe that the LC
gauge is not attainable for x0 ∈ S 1 , since this corresponds to an ’unphysical
world’ anyway. g11 = 2e1 0̂ e1 1̂ +(e1 1̂ )2 g1̂1̂ , on the other hand, is to be determined
by the field eqs. and in general will differ from zero.
For reasons of completeness and because our formulation differs from previ-
ous work, we will briefly sketch how to solve the remaining field equations. Since
there is no sense in keeping g1̂1̂ arbitrary for this purpose, the zweibein will refer
to the LC metric (11) for the rest of this section. Knowing the x0 –dependence
of T − = T+ as well as of R and T + , one first determines the dependence of
e− µ = (1/e1 − )(−e1 + , 1) and ω1 on this coordinate, in order to reformulate the
remaining eqs. of (7) to (10) as ordinary differential eqs. in x1 . To this end
one uses the defining eqs. for torsion and curvature which in this gauge are
T a = −(1/e1 − )[∂0 e1 a + δ(a+) ω1 ] and R = −2(∂0 ω1 )/e1 − . Thus one obtains
for A(x1 ) 6= 0:
β
R = A x0 + B +
γ
γ 1 λ − (β 2 /4γ)
T+ = C exp(− A x0 ) − (A x0 + B)2 +
β 4A γA
4
γ
T− = − A
β
β D γ
ω1 = F− exp( A x0 ) (A x0 + B)
2γ A β
βD β2 4λ γ
e1 + = G − (F + CD) x0 + [(A x0
+ B)2
+ − ] exp( A x0 )
4γ A2 γ2 γ β
γ
e1 − = D exp( A x0 ) (12)
β
with
λ γ β
I ≡ − (B + )2
β 4β γ
and with the arbitrary functions B, C̃, D 6= 0, F̃ , G̃. For the case A = 0 ev-
erywhere
p the remaining eqs. of (7) to (10) can be shown to lead to B =
± (4λ/γ) − (β/γ) (⇔ I = 0) and C̃ = 0 everywhere, which is the deSitter
part of the solution. For T − 6= 0 everywhere it suffices to regard only the first
eq. of (7), the other two eqs. being redundant. It yields in this case (the prime
denoting a derivative):
A′ + A(CD + F ) = 0 (14)
β
B ′ − γ CD − AG = 0 (15)
γ
C ′ − C(F − β AG) = 0 (16)
Plugging (14, 15) into (16) one can easily integrate the latter equation to give
C1 γ
C= exp[− B − 1] (17)
A β
for an arbitrary constant C1 . Eqs. (14) and (15) finally can be solved by
expressing F and G in terms of the remainig three functions.
5
Calculating the Lorentz invariant T 2 = 2 T + T − and making the result
explicit in C1 , one regains the quantity Q(R, T 2 ) ∝ C1
γ 2γ γ 4γλ
Q = exp( R) [ T 2 − ( R − 1)2 − 1 + 2 ≡
β β β β
γ 2γλ γ
≡ 2 exp( R) [− 2 E + R − 1] (18)
β β β
Q;a = 0, (19)
which also follows immediatly from taking the covariant derivative of the first
equation of (18) and the subsequent usage of (3), (4).
There exist also solutions with isolated zeros of A(x1 ) as outlined in Ref. 7.
3 Hamiltonian Formulation
Again gab will be restricted only by gab = const and det gab = −1 in the fol-
lowing. Further we will use the notation ∂0 f =: f˙, x1 =: x, ∂1 f =: ∂f and
choose, without loss of generality, x0 =: t as evolution parameter; clearly the
classical system is reproduced by the Hamiltonian formalism irrespective of the
space–time character of t ≡ x0 , which is not fixed at this stage. We observe first
that (1) or (2) is of the simple form
1 i
L= (ϕ̇ − f i ) Wij (ϕ̇j − f j ) − V (20)
2
with
6
is polynomial in the basic fields, in contrast to the Lagrangian and also in con-
trast to the conventional Hamiltonian of general relativity in d = 4 (but similar
to the Ashtekar formulation of the latter — cf. Isham12 ). This seems to indicate
an advantage of the Hamiltonian path integral formulation as opposed to the
Lagrangian one. However, when formally integrating out the momenta8 of a
BRST–version of the canonical Hamiltonian (with ϕ̄i interpreted as Lagrange
multipliers – cf. comments following (31) below), one just regainsc the La-
grangian as used in Kummer et. al.9 ; the four–vertex ghost couplings10 , typical
for theories with a non–closing constraint algebra, disappear in this model.
To get e.o.m. and constraints which are equivalent to the Lagrangian field
eqs., one has to use the primary Hamiltonian HP , which is the sum of HC and
the product of all primary constraints with Lagrange multiplier (LM) functions.
Usually this is proven for discrete systems (cf. Sundermeyer11 and references
therein), and a formal translation to the continuous case would imply HP =
HC + λ̄i π̄i . But in our opinion this is not enough to reproduce exactly what
we have done on the Lagrangian sector (sec. 2). There we have required the
variation of the basic fields to vanish at the boundary of the parameter space B in
order to get the field equations (3), (4). This can be viewed as a minimalization
of the corresponding action with ’temporarily fixed’ boundary conditions on
∂B, finally asking for the set of all solutions compatible with some boundary
condition. Now, any of the usual proofs working in a discrete version of a
continuous Lagrangian system correspond to a rectangular B ⊂ R2 . But for
such a B the ’temporary’ fixation of the fields at tmin and tmax have to be
supplemented by ’temporary’ boundary conditions for the field variables at xmin
and xmax . According to (6) it suffices to prescribe such boundary conditions
only for the ϕi (and not necessarily for the ϕ̄i ) in our case; they are explicitly
t–dependent and have to be added as additional primary constraints to the
Hamiltonian formulation of the system.
As a next step one has to calculate the secondary constraints. Considering
i
ϕ
R | x−boundary ≈ g i (t) first, the requirement that its Poisson brackets with HP ≡
dx HP shall vanish reproduces just πi |x−boundary ≈ (−β Ta , γ R) |x−boundary
(cf. (22)) when identifying ϕ̇i |x−boundary with ġ i (t). All further tertiary, etc.
constraints can be reformulated to be only restrictions to the possible choice for
g i (t). So we are left with a set of t–dependent second class constraints at the
boundary. The corresponding degrees of freedom can be eliminated by means of
Dirac brackets so that we will be allowed to set all surface terms strongly equal
to zero below.
A simple calculation, using e = ε(ab) e0 a e1 b , shows that the conservation of
π̄i ≈ 0 gives (Gi := {π̄i , HP }):
7
with
1 1 2
E ≡ (πω )2 − π − λ, π 2 ≡ π a πa . (25)
4γ 2β
On shell (using the e.o.m. for ϕ̇i ) (25) becomes just (5). It should
R be men-
tioned, furthermore, that on deriving (23, 24) we have replaced dy ∂y [δ(y −
x)] πi (y) by −∂πi (x), which is possible only since πi vanishes strongly at
xmin , xmax .
At this point it is convenient to calculate the brackets of the basic fields with
the Gi ; the only nonvanishing ones are (we suppress the arguments x and y and
write δ ′ for ∂δ(x − y) etc.):
If we had not employed the argument with the Dirac brackets, (31) would be
still valid since, as alluded to above, one would have to include the (not com-
pletely well–defined) term −δ(x − xboundary ) πi (x) in the formulas (23,24) for
the secondary constraints Gi .
That the Hamiltonian vanishes weakly is a feature in common to all parame-
trization invariant theories11 . In this specific case, though, the structure of
(31) reveals that from the Hamiltonian point of view one could also consider
H = µi Gi instead of HP , which is obtained from (31) in the gauge ϕ̄i ≈ −µi ,
dropping the barred fields by the introduction of Dirac brackets. This is also
completely analoguos to the usual Hamiltonian formulation of general relativity
in four dimensions (cf. e.g. Isham12 ), where one regards the lapse function
and the shift vector as mere Lagrange multipliers. Because of (31) the relations
(29,30) show that there are no further secondary constraints. Moreover, all of
the 6 ∞ constraints (Gi , π̄i ) =: GA are first class (FC).
d For a detailed comparison one has to set γ = 1/2, to replace β by 2β, and to replace
8
Before turning to the symmetries generated by the FC constraints, let us
comment on the flow generated by (31), i.e. the ’t–evolution’ of the coordinates.
Supplementing them by the constraints, we have (cf. (23,24)):
i
ϕ̄˙ ≈ λ̄i , π̄i ≈ 0 (32)
ϕ̇i ≈ {ϕi , HP } ⇔ πi ≈ (−β Ta , γ R) (33)
∂µ πω ≈ εab eµ a π b (34)
∂µ πa ≈ −εab (eµ b E + ωµ π b ) (35)
Multiplying the latter two eqs. by eµ b , one obviously regains just the covariant
e.o.m. (3), (4).
Beside this reformulation of the covariant field equations on the Lagrangian
level the Hamiltonian formulation (using HP ) provides some additional infor-
mation and insight. On the one hand, the equations (32) practically force one to
use a LC–like gauge. Furthermore, the constraint eqs. (23,24) serve as constants
of the motion, thus saving one the integration of three of the e.o.m. (cf. Ref. 4 or
Ref. 7 for more details). On the other hand, the knowledge that (34), (35) with
µ = 1 are first class constraints reveals11 that on shell the choice of the LC gauge
does not fix the gauge freedom completely. From the Hamiltonian point of view
the LC gauge turns only π̄i ≈ 0 into second class constraints. This shows that
it should be possible to gauge away still 3 ∞ phase space coordinates, i.e the
three arbitrary functions of x ≡ x1 in the general solution obtained at the end
of sec. 2 – except possibly for a finite number of constants. Since we will show
in the following section that the gauge symmetries generated by the FC con-
straints are just diffeomorphisms and Lorentz transformations, this elimination
could already have been carried through in sec. 2. The corresponding steps on
the Lagrangian level can be found in Grosse et. al.4 . According to their result
−4(γ/β)2 C1 = Q(R, T 2 ) is the only gauge independent (physical) parameter
left in the model. For the topology of a cylinder, however, the reduced phase
space is two dimesional (cf. Ref. 7).
One could also try to build up a Hamiltonian formulation with the extended
Hamiltonian HE as proposed by Dirac13 , which emerges from Hc by adding
all FC constraints via Lagrange multipliers (LM). When absorbing −ϕ̄i into
the definition of new LMs λi , which will prove to be a mathematical shortcut
especially in the following section since it formally corresponds to a strongly
vanishing canonical Hamiltonian, one obtains (λA := (λi , λ̄i )):
HE = λA GA . (36)
In this way the e.o.m. for the unbarred fields become proportional to LMs,
too. Setting all the λA zero as the simplest choice, one is left with ϕA ≈
ϕA (x), πA ≈ (πi (x), 0), in which the nine functions of x are restricted by the
three contraints Gi . The FC character of the six constraints GA will lead to a
gauge identification of the six functions needed as ’initial data’ on a hyperline
9
t = const so that the (superficial) degrees–of–freedom count gives the same as
in the case of HP . That the fixation of all LMs in H does not fix the gauge
completely can be understood by considering a hyperline for some smaller t:
Starting from such a line with different choices for the LMs leads to different but
physically equivalent points on the hyperline used above. Nevertheless, without
having gone into further calculational detail, we believe that it would be difficult
to extract geometrically interpretable results comparable to the results of sec. 2
when using HE = 0 as above. (Note also that the gauge choice λA = 0 in (3.17)
corresponds to a gauge with e = 0 in the original formulation). However, for
the gauge independent quantity Q(R, T 2 ) the e.o.m. are unchanged: Q̇ ≈ 0 is
trivial and ∂Q ≈ 0 follows from Gi ≈ 0.
in which f = f (ϕA (x, t), πA (x, t)) is an arbitrary function on the phase space and
ǫA (x, t) an arbitrary infinitesimal parameter, are a symmetry transformation of
SE (up to surface terms), iff
If we had not absorbed Hc by the definition of the LMs, we had to add −VBA ǫB
to (39) ({Hc (t), GA (x, t)} =: VBA GB ). Now, according to Henneaux et. al.6
these are all local transfomations leaving SE invariant.
In our case the above equations become (cf. (26) to (28)):
10
1
δe1 a = − εbc e1 c π a ǫb − εa b ω1 ǫb − ∂ǫa + εa c e1 c ǫω (41)
β
1
δω1 = εbc e1 c πω ǫb − ∂ǫω (42)
2γ
δπa = εab E ǫb + εac π c ǫω (43)
δπω = −εbc π c ǫb (44)
1
δλa = ǫ˙a + εbc e0 c π a ǫb + εa b ω1 ǫb − εa c e0 c ǫω (45)
β
1
δλω = ǫ˙ω + εbc e0 c πω ǫb (46)
2γ
Setting λi = −ϕ̄i , the action SE becomes equal to SP . Thus, all local sym-
metries leaving SP invariant are obtained when restricting (40) to (46) by
δλi = −δ ϕ̄i ≡ −ǭi . This gives among others:
1
δeµ a = −∂µ ǫa − εbc eµ c π a ǫb − εa b ωµ ǫb + εa c eµ c ǫω (47)
β
1
δωµ = εbc eµ c πω ǫb − ∂µ ǫω , (48)
2γ
whereas the transformations for the momenta remain unchanged. The transition
from SP to SL is accomplished by the second equations of (32) and (33). Using
the latter to eliminate the momenta from (47), (48) as well as the identitiese
(T a , R/2) ενµ = (T a µν , 2ω[ν ,µ] ) and the abbreviation eb ν ǫb ≡ ǫν , one verifies
easily:
11
or δαL ωµ = −α,µ . The usual transformation under diffeomorphisms x̃µ :=
xµ + ξ µ (xσ ) for a field vµ reads infinitesimally: δξD vµ = −vν ξ ν ,µ −vµ ,ν ξ ν .
Since obviously δξD eµ a = −(eν a ξ ν ),µ −2e[µ a ,ν] ξ ν the comparison of (49, 50) to
the above equations yields:
Thus it is found that the flow generated by Ga on the regular sector is a specific
combination of a diffeomorphism and a Lorentz transformation, whereas Gω
corresponds directly to the generator of the Lorentz group in this sector.
There are no momenta living on the purely Lagrangian sector. Nevertheless,
we do not see a general reason within the above proceduref why the transforma-
tion properties of the momenta on the SP –level could not lead to a restriction
of the parameters in (53) (similar to the transition from SE to SP , in which
the number of free parameters has been halved). To show explicitely that this
is not the case, we note that δπ̄i = 0 is consistent with the second equation of
(32). Furthermore, by means of (33) the eqs. (43) and (44) can be written as:
with δ̂ = 0 on shell (cf. (3,4))! This completes the proof for the one–to–one
correspondence of the gauge symmetries on the Lagrangian and the Hamiltonian
level, and, as consequence, the absence of any further local symmetry.
GA ψphys = 0, (57)
f Certainly we know already that SL is invariant under all of (53).
12
it is common to demand that one cannot produce further constraints by applying
GB to the lefthand side of (57). This is guaranteed, if the constraint algebra
(29,30) has no quantum anomalies, i.e. if the quantum version of (29,30) is
obtained from these equations by the mere replacement
{ , } → −i/h̄ [ , ] (58)
without reordering of the (hermitean) operators Gi on the r.h.s. so that some
Gi would be placed on the left of πi . To show that there are indeed no such
anomalies for the case of our model, one first observes that the constraints are
only linear in the fields ϕi (cf. (23,24)). It is easy to verify, by dropping a
similar term as the one in (56), that the commutator between two operator
valued functions
1 1
f = [ϕi (x), fi (πj (x))]+ and g = [ϕi (x), gi (πj (x))]+
2 2
gives
1 ∂gk ∂fk
[f (x), g(y)] = ih̄ [ϕk , ( fi − gi )]+ δ(x − y), (59)
2 ∂πi ∂πi
which, except for the factor ih̄, is just the antisymmetrization of the
classical result. Applying this to (29,30), we are left to show that
(1/2) [e1 b (x), π a (x)εab E(x)]+ = (1/2) π a (x) εab [e1 b (x), E(x)]+ , but this is true
because of (56).
6 Outlook
To carry through the Dirac quantization there are still two main problems to be
solved at this stage. The first one is the appearance of a term proportional to
δ(0) when one solves (57) with hermitean GA by representing e1 b as a functional
derivative operator. The other one is the well–known problem of ’frozen time’
(cf. Isham12 ): Due to (57) HP vanishes on all physical states so that a normal
Schroedinger equation does not make sense. These problems, among others,
are treated in Ref. 7. — What makes the quantization of the present model
interesting is that there appear similar conceptual problems as in d = 4, but
that the mathematical difficulties are much simpler to be overcome. Therefore it
is possible to explicitely check some of the basic approaches to quantum gravity.
Another promising step seems to be the coupling of matter fields to the
action (1), which, according to our results, is not restricted by any ’hidden’
symmetry.
Acknowledgements
I have profitted from numerous discussions with W. Kummer, D. J. Schwarz,
and especially P. Schaller, who also contributed the idea to the simple proof
13
following eq. (58). I am also grateful to these people, most of all W. Kummer,
for many suggestions regarding the manuscript.
14
with some arbitrary functions f and g.
In a manifold with topology S 1 × S 1 it is possible to use just one chart
when requiring that all fields are periodic in x0 and x1 (e.g. with the period
normalized to 2π). One concludes from (66) by integrating this equation over a
period in x0 that even if (60) is obtainable globally (61) is not.
References
15