Hamiltonian formulation of teleparallel gravity

Rafael Ferraro1, 2, ∗ and María José Guzmán1, †

1
Instituto de Astronomía y Física del Espacio (IAFE, CONICET-UBA),
Casilla de Correo 67, Sucursal 28, 1428 Buenos Aires, Argentina.
2
Departamento de Física, Facultad de Ciencias Exactas y Naturales,
Universidad de Buenos Aires, Ciudad Universitaria, Pabellón I, 1428 Buenos Aires, Argentina.
The Hamiltonian formulation of the teleparallel equivalent of general relativity (TEGR) is de-
veloped from an ordinary second-order Lagrangian, which is written as a quadratic form of the
coefficients of anholonomy of the orthonormal frames (vielbeins). We analyze the structure of
eigenvalues of the multi-index matrix entering the (linear) relation between canonical velocities and
momenta to obtain the set of primary constraints. The canonical Hamiltonian is then built with
the Moore-Penrose pseudo-inverse of that matrix. The set of constraints, including the subsequent
arXiv:1609.06766v1 [gr-qc] 21 Sep 2016

secondary constraints, completes a first class algebra. This means that all of them generate gauge
transformations. The gauge freedoms are basically the diffeomorphisms, and the (local) Lorentz
transformations of the vielbein. In particular, the ADM algebra of general relativity is recovered as
a sub-algebra.

I. INTRODUCTION Fλρ = ∂λ Aρ − ∂ρ Aλ ). Since Fλρ is anti-symmetric, then

The determination of the independent dynamical de-
grees of freedom is of the utmost importance in any field
theory, since it allows to exhibit the internal consistency
of the theory, and tackle the issue of the well-posedness
of the Cauchy problem. It also puts the theory into a dif-
ferent perspective, because it helps to find the minimal
number of variables specifying the state of the system,
so being vital for the quantization of the theory. Ac-
cording to the procedure due to Dirac [1], the number
of genuine degrees of freedom can be determined from
the algebra of the constraints among the canonical vari-
ables of the theory. The constraints firstly appear when
the canonical momenta are computed. These primary
constraints have to be consistent with the Hamiltonian
evolution of the system, which leads to secondary con-
straints, and so on. Finally, the set of all the constraints
is reclassified as first class and second class constraints,
depending whether their Poisson brackets are or not null
on the constraint surface in the phase space. First class
constraints generate gauge transformations; so, each of
them is related to a spurious degree of freedom. On the
other hand, second class constraints can be reorganized
as pairs of spurious conjugated variables. Thus, the num-
ber of genuine degrees of freedom can be computed as
# d.o.f. = # pairs of canonical variables
−# first class constraints
1 is captured in the canonical momenta π i ) [2].
− # second class constraints . (1)
2 The canonical formulation of general relativity (GR)
A nice example is the Maxwell potential, described by
four dynamical variables Aµ that are governed by the
Lagrangian L[Aµ ] ∝ Fλρ F λρ (the field tensor Fλρ is
∗ ferraro@iafe.uba.ar; member of Carrera del Investigador Cientí-
fico (CONICET, Argentina).
† mjguzman@iafe.uba.ar
∂0 A0 is not present in the Lagrangian. Thus the canon-
ical momentum π 0 = ∂L/∂(∂0 A0 ) identically vanishes;
it is a primary constraint. The consistency of the con-
straint π 0 = 0 with the evolution requires the vanishing
of the Poisson bracket between π 0 and the Hamiltonian;
this leads to the secondary constraint ∇i π i ∝ ∇i F 0i = 0
(Gauss’s law). Both constraints are first class, since the
Poisson brackets between canonical momenta are iden-
tically null. Therefore, according to Eq. (1), one real-
izes that the electromagnetic field has not four degrees of
freedom Aµ at each event, but only two (electromagnetic
waves are transversal). At the level of the initial data,
the existence of constraints imply a restriction on the
spectrum of allowed initial configurations. Besides, the
absence of kinetic term for A0 in the Lagrangian implies
that the evolution of this dynamical variable, conjugate
to the first class constraint π 0 , remains completely un-
determined. The same happens to the evolution of the
longitudinal component of the potential A|| , which also
remains undetermined as a consequence of the existence
of the first class constraint ∇i π i . Thus, A0 and A|| are
gauge freedoms. The former conclusions can also be de-
rived from a slightly modified Lagrangian. The integra-
tion by parts of one of the terms containing ∂i A0 leads
to a surface term, which can be eliminated, plus the term
A0 ∇i F 0i . In such way, the spurious degree of freedom
A0 becomes a Lagrange multiplier whose variation leads
to the Gauss’s law constraint (any other presence of A0
relies on the widely spread formalism by Arnowitt, Deser
and Misner (ADM) [3], in which the spacetime is fo-
liated into a family of spacelike hypersurfaces that in-
duces a proper decomposition of the metric tensor gµν .
The Einstein-Hilbert Lagrangian can be integrated by
parts to realize that the temporal sector of the metric
(the lapse N and the shift vector Ni ) is thrown into the
role of Lagrange multipliers associated to four first class
constraints (the super-Hamiltonian and super-momenta

constraints). So written, the Lagrangian gives dynamics
only to the six components of the 3-dimensional metric
gij on the spacelike hypersurfaces of the foliation; but
the canonical variables (gij , π ij ) are still constrained by
the four first class constraints. Thus the gravitational
field contains only two genuine degrees of freedom. In
fact, apart from the undetermined evolutions of the four
Lagrange multipliers (N, Ni ), there are also four gauge
freedoms among the six components of gij (gravitational
waves are transversal and trace-less). As a feature that
distinguishes GR from electromagnetism, the GR Hamil-
tonian vanishes because of the constraints. This feature
is typical of systems having a time hidden among their
canonical variables [4].
Early in the 1918, Weyl’s unsuccessful attempt of uni-
fying gravitation and electromagnetism introduced for
the first time the notion of gauge theories [5]. Einstein
himself tried ten years later the same unification idea,
but taking advantage of the sixteen components of the
tetrad field in order to include the electromagnetic field
[6]. Later he realized that the arbitrariness in the choice
of the tetrad comes from the set of local Lorentz trans-
formations that leave the metric unchanged, therefore
the extra degrees of freedom could not give account for
electromagnetism. However, he introduced the concepts
of teleparallelism that remain important until today, pre-
senting for the first time the teleparallel equivalent of gen-
eral relativity (TEGR), an equivalent formulation of gen-
eral relativity. In fact, although both theories have differ-
ent Lagrangian formulations, they are equivalent at the
level of the equations of motion. Nonetheless, they are
based on completely different Lagrangian constructions.
This is so because TEGR describes gravity as the effect
of torsion in the curvatureless Weitzenböck geometry;
the dynamical variables are not the components of the
metric gµν but those of the field of orthonormal frames
–tetrads or vierbeins– eaµ (a and µ are SO(3, 1) and coor-
dinate indices, respectively) [6, 7]. As a consequence, the
Hamiltonian formalisms of GR and TEGR are different
too. Among the articles treating the Hamiltonian for-
mulation of TEGR we specially mention Ref. [11], which
introduces a set of auxiliary variables in a first order ap-
proach that lowers the order of the Euler-Lagrange equa-
tions (cf. [8, 9, 12, 13]), and Ref. [10] that deals with an
enlarged set of variables and constraints to enforce the
vanishing of the curvature. The canonical formulation of
TEGR has been also stated in the geometric language of
differential forms [14, 15].
In this work we will put forward the Hamiltonian
formalism for TEGR in a way as close as possible to
the second order formalism of electrodynamics that was
sketched above. This work is organized as follows: in
Section II we introduce the standard TEGR dynamics,
which is governed by a Lagrangian quadratic in the tor-
sion. In Section III we show that the TEGR Lagrangian
can be reformulated as the quadratic inner product of
the anholonomy coefficients with respect to a superme-
tric that is defined in the tangent space. In Section IV
2
we obtain the set of primary and secondary constraints
that are equivalent to those of electrodynamics and GR
geometrodynamics. In Section V we study the gauge
transformations generated by these constraints (they will
prove to be first class). Compared with geometrodynam-
ics, TEGR has an additional gauge symmetry associ-
ated to local Lorentz transformations of frames, which
is the source of the constraints analyzed in Section VI.
In Section VII the (constrained) linear relations between
canonical momenta and velocities is inverted to build the
canonical TEGR Hamiltonian H; the procedure implies
a careful analysis of the eigenvector structure involved
in these linear relations, in order to build the respective
pseudo-inverse matrix. The entire set of n(n + 3)/2 con-
straints (n is the spacetime dimension) is consistent with
the evolution governed by H; besides, they are first class
as proven by the algebra of constraints computed in Sec-
tion VIII. In Section IX we summarize the main steps and
the achievements of the paper. The Appendix A shows
some useful computations that are needed throughout
the work.
II. TEGR AND STANDARD LAGRANGIAN
FORMULATION
TEGR is a theory of gravity where the field of or-
thonormal frames plays the role of dynamical variable.
Let M be a manifold, {ea } a basis in the tangent space
Tp (M ), and {Ea } its dual basis in the cotangent space
Tp∗ (M ) (i.e., if the 1-forms Ea are applied to the vectors
eb one obtains Ea (eb ) = δba ). They can be expanded in
a coordinate basis as ea = eµa ∂µ and Ea = Eµa dxµ ; so
duality means that
Eµa ebµ = δba , eµa Eνa = δνµ . (2)
Here and from now on, we will use Greek letters µ, ν, ... =
0, ..., n − 1 for spacetime coordinate indices, and Latin
letters a, b, ..., g, h = 0, ..., n − 1 for Lorentzian tangent
space indices. A vielbein (vierbein o tetrad in n = 4
dimensions) is a basis encoding the metric structure of
the spacetime:
g = ηab Ea ⊗ Eb , (3)
therefore,
Ea · Eb = g(Ea , Eb ) = ηab , (4)
which means that the vielbein is an orthonormal basis.
In component notation, the former expressions look
gµν = ηab Eµa Eνb , ηab = gµν eµa eνb , (5)
which implies that the relation between the metric vol-
ume and the determinant of the matrix Eµa is
p .
|g| = det[Eµa ] = E . (6)
 3

Since the vielbein encodes the metric structure of the this point, we use to say that TEGR is a theory where
spacetime, one can formulate a dynamical theory of the the gravitational effects are fully encoded in the torsion.
spacetime geometry by defining a Lagrangian for the viel- On the contrary, GR associates gravity to curvature; it
bein field. In particular, there is a Lagrangian which assumes that the spacetime is endowed with the torsion-
leads to dynamical equations for the vielbeins that are less Levi-Civita connection, whose curvature enters the
equivalent to Einstein equations for the metric [16]. The Einstein-Hilbert Lagrangian L = E R. The reason why
so called teleparallel equivalent of general relativity is TEGR is indeed equivalent to GR is traced to the fact
governed by the Lagrangian density that their respective Lagrangian densities differ in a sur-
face term:
L = ET , (7)
− E R = E T − 2 ∂ρ (E Tµµρ ), (12)
where T is the torsion scalar
. Even so the vielbein field contains n2 components, while
T = T ρ µν Sρ µν
, (8) the metric tensor has only n(n + 1)/2. However, TEGR
dynamical equations are invariant under local Lorentz
which is made up of transformations of the vielbein, which involve n2
gen-


.
T µνρ = eaµ (∂ν Eρa − ∂ρ Eνa ) , (9) erators. Such a gauge invariance means that n2 =
n(n − 1)/2 degrees of freedom cancels out, which allows
and that the theories turn out to be equivalent at the level of
the equations of motion.
. 1
 
Sρµν = K µνρ + Tλλµ δρν − Tλλν δρµ , (10)
2
III. TEGR LAGRANGIAN IN TERMS OF THE
where
VIELBEIN FIELD
. 1
K µνρ = (Tρ µν − T µνρ + T νµρ ) . (11)
2 With the aim of preparing the TEGR Lagrangian for
µ the study of its canonical structure, we will rewrite it
In Lagrangian (7), the strength field T νρ is the tor-
. completely in terms of eµa , Eνa and the derivatives ∂µ Eνa .
sion associated with the Weitzenböck connection Γµνρ =
µ a µν This imply the removing of any presence of the metric
ea ∂ν Eρ , and K ρ is the contorsion [17]. In geometric
. field, since such contributions hide a dependence on the
language, torsion is the 2-form Ta = dEa + ω ab ∧ Eb vielbein. We transform the scalar torsion into
a
, where the 1-form ω b is the spin connection. Weitzen-
böck connection is the choice ω ab = 0, because it leads to 1 1
T = Tρ µν T ρµν − T ρµν T µνρ − T ρµρ T νµν . (13)
(Ta )νρ = (dEa )νρ = ∂ν Eρa − ∂ρ Eνa = Eµa T µνρ . Weitzen- 4 2
böck connection is metric compatible, since ∇ν Eµa = We note that all terms in T are quadratic in the anti-
∂ν Eµa − Γ λνµ Eλa = 0. Besides, from Eq. (2) we also get symmetrized derivatives of the vielbein; writing term by
that ∇ν eµa = 0. This means that the vielbein is au- term one gets
tomatically parallel-transported along any curve. Fur-
thermore, the parallel-transport of any vector does not 1 1
T µν T ρµν = gρα g βµ g γν T αβγ T ρµν ; (14)
depend on the path (it is absolute), since Weitzenböck 4 ρ 4
connection has the remarkable feature that the curva- then one replaces the expressions for the torsion tensor
.
ture Rab = dω ab + ω ac ∧ ω cb is identically zero. The 9 and the metric in terms of the vielbein field and its
(Weitzenböck) covariant derivative of a vector is ∇ν U = inverse 5:
∇ν (U a ea ) = ea ∂ν U a ; thus, vector U will be parallel
transported if and only if its components U a are con- 1
T µν
T ρµν = ηab η c[d η f ]e E ∂µ E aν ∂ρ E bλ eµc eνe eρd eλf .
stant. 4 ρ
Although TEGR Lagrangian can be understood in (15)
terms of the Weitzenböck connection and its respective After this procedure has been performed in all the terms,
torsion, it should be emphasized that the TEGR La- the TEGR Lagrangian becomes
grangian neither fixes the connection nor the vielbein; it 1
only determines the metric, as it is well known. Further- L=E T = E ∂µ E aν ∂ρ E bλ eµc eνe eρd eλf Mab cedf , (16)
2
more, whenever matter couples minimally to the metric,
as usual, the free particles will follow geodesics of the
µ
where we call supermetric Mab cedf the emerging Lorentz
(torsionless) Levi-Civita connection Γνρ . 1 Setting aside invariant tensor given by
. e] f]
Mab cedf = 2 ηab η c[d η f ]e − 4 δa[d η f ][c δb + 8 δa[c η e][d δb .
(17)
1 However, Levi-Civita and Weitzenböck connections are related
µ
The supermetric is antisymmetric in the pairs of indices
through the contorsion: Γνρ = Γµ µ
νρ − K νρ . c− e and d− f , what implies that only the antisymmetric
 4

parts of ∂µ E aν and ∂λ E bρ take part in the Lagrangian From Eq. (21) we immediately get n trivial primary
(16). Other properties of the supermetric are summa- constraints
rized in the Appendix A 1. .
G(1)
a = Π0a ≡ 0 , (24)
We remark that the index structure of the supermetric which are derived by noticing that e0c e0e is symmetric in
is natural when we recognize in Eq. (16) the anholon- c − e but Mab cedf is antisymmetric. Although we can-
c
omy coefficients fab , which are defined by the commuta- not prove yet that they are first class (i.e. we do not
c
tor [ea , eb ] = fab ec . In fact, by using the equations (2) know yet whether they generate gauge transformations),
a
the coefficients fbc can be rewritten as the electromagnetic analogue tells us that they mean the
E0a ’s are spurious gauge dependent variables, that would
a
fbc = −eµb eνc (∂µ Eνa −∂ν Eµa ) = −2 eµb eνc ∂[µ Eν]
a
, (18)
become Lagrange multipliers if an integration by parts
which can be related to other geometrical magnitudes, were performed in the action. This is in line with the
as the Weitzenböck torsion and the Lie derivative of the spurious character of the temporal sector of the metric
vielbein: tensor we have commented in Section I.
The primary constraints must be satisfied at any time.
a
fbc = Ta (ec , eb ) = (Lec Ea )(eb ) . (19) In other words, if the system is on the constraint surface
at the initial time, it must remain there along the evo-
In terms of these coefficients, the Lagrangian density lution. If this consistency requirement were not accom-
looks in a very elegant form: plished, then it could be enforced by resorting to new
(secondary) constraints [20]. From a Hamiltonian per-
1
L = a
E fce b
fdf Mab cedf . (20) spective, the consistency of the primary constraints is
8 controlled by means of the primary Hamiltonian [2]
A similar expression for the Lagrangian can be found in Z
Ref. [18], where the anholonomy coefficients are identi- Hp = H + dx ua (t, x) φ(1) a (t, x) , (25)
fied with a Yang-Mills-like field strength; however, that
Lagrangian still mixed tangent space and coordinate in- where H is the canonical Hamiltonian, ua (t, x) are arbi-
dices. Instead, Lagrangian (20) does not involve coordi- (1)
trary functions, and φa are all the primary constraints.
nate indices; it shows that supermetric Mab cedf is a rele- The consistency will be fulfilled if the Poisson brackets
vant geometric object in the (co-) tangent space structure (1)
{φa , Hp } are null on the constraint surface. This re-
of the spacetime. We intend to analyze the Hamiltonian quirement could be satisfied by properly choosing the
structure of TEGR by starting from Lagrangian (16, 20), functions ua (t, x); if not, new (secondary) constraints
and following a canonical second-order procedure. will be needed to enforce it, and so on. Actually, in
TEGR we will find that all the Poisson brackets be-
tween constraints are null on the constraint surface. This
IV. SUPER-HAMILTONIAN AND means that primary and secondary constraints are all
SUPER-MOMENTA CONSTRAINTS
first class; they generate gauge transformations. Thus
the constraints will be consistent with the evolution if
We compute the canonical momenta by differentiating their Poisson brackets with H vanish on the constraint
the Lagrangian (16) with respect to the time derivative surface (i.e., if H is gauge invariant, as it should be ex-
of the canonical variable Eµa : pected).
∂L In spite of the entire set of primary constraints was
Πµa = = E ∂ρ E bλ e0c eµe eρd eλf Mab cedf
∂(∂0 Eµa ) not obtained yet, the evolution of constraints (24) can
1 be analyzed at the level of the Euler-Lagrange evolution
= − E e0c eµe fdfb
Mab cedf . (21) equations,
2
So, the Poisson brackets in TEGR are defined as ∂L ∂L
∂µ − = 0. (26)
. ∂(∂µ Eν ) ∂Eνa
a
{A(t, x), B(t, y)} =
Z 
δA(t, x) δB(t, y) δA(t, x) δB(t, y)
 By splitting the first term, one gets
dz − (. 22)
δEλa (z) δΠλa (z) δΠλa (z) δEλa (z) ∂L ∂L
∂0 Πνa + ∂i − = 0. (27)
∂(∂i Eνa ) ∂Eνa
The brackets between fundamental canonical variables
are Therefore, if the constraints (24) must be fulfilled at any
time, we obtain n equations – those having ν = 0 – which
{Eµa (t, x), Πνb (t, y)} = δba δµν δ(x − y) . (23) do not contain second-order temporal derivatives:
Additional fundamental Poisson brackets, including E, ∂L ∂L
∂i − = 0. (28)
eµa , etc., are summarized in Appendix A 4. ∂(∂i E0a ) ∂E0a
 5

(2)
Like the Gauss’s law in electromagnetism these equations Gµ . In general, the infinitesimal gauge transformation
do not contain dynamics, but they constrain the dynam- generated by a first class constraint G is [2]
ics. Since the derivatives of the vielbein enter the La- Z
grangian only in antisymmetric combinations, then it is a
δEµ (t, x) = dy ǫ(t, y) {Eµa (t, x), G(t, y)} . (34)
∂L ∂L
∂i = − ∂i = − ∂i Πia . (29) Any transformation of the vielbein has to be accompa-
∂(∂i E0a ) ∂(∂0 Eia )
nied by a transformation of the basis {ea }, in order to re-
Thus, we have found n secondary constraints: spect the duality relations Ea (eb ) = δba of Eq. (2). There-
fore
∂L
∂i Πia + = 0. (30)
∂E0a Ea (δeb ) + δEa (eb ) = 0 , (35)

We will prove that these constraints are consistent with or

the evolution; so they do not generate new constraints.
For this, we will apply the derivative ∂0 to the constraints δeνb = − eνa eµb δEµa . (36)
(30), and use the Eq. (27) to replace ∂0 Πia :
  According to Eq. (34), any linear combination of primary
i ∂L ∂L ∂L (1)
constraints ǫb (t, x) Gb generates a transformation that
∂0 ∂i Πa + = −∂i ∂j + ∂µ
∂E0a ∂(∂j Eia ) ∂Eµa only affects the temporal component of the 1-forms Ea ,
∂L ∂L
= −∂i ∂j a + ∂ν ∂µ ≡0 (31) δE0a (t, x) = ǫa (t, x) , (37)
∂(∂j Ei ) ∂(∂µ Eνa )
(we also use Eq. (26) in the last step). The null result (or δEa = ǫa dt) which also implies 2

comes from the fact that ∂µ Eνa enters the Lagrangian in

antisymmetric combinations but the operators ∂i ∂j and δeνb = − ǫa eνa e0b . (38)
∂ν ∂µ are symmetric. (2) (2)
So far, we have got a set of constraints which is consis- Instead, the transformations generated by G0 , Gk only
tent with the evolution. To write them in a fully canoni- affect the spatial components of the forms Ea (the canon-
cal way, we have to compute the derivative ∂L/∂E0a and ical Hamiltonian density H does not contain Π0a ). Then,
(2)
express it as a function of the momenta, the vielbein the infinitesimal gauge transformations generated by G0
and its spatial derivatives. This computation is made in (2)
and any arbitrary combination ξ k Gk are respectively
the Appendix A 2, where we obtain R that the canonical
Hamiltonian density H (i.e., H = dx H) takes part in δEia (t, x) = ξ Ėia (t, x) + E0a ∂i ξ
the results. These results are better understood when = ∂i (E0a ξ) + ξ 2 ∂[0 Ei]a , (39)
projected on E0a and Eka . Thus, we get the secondary
constraints written in canonical form:
(2) . δEia (t, x) = ξ k ∂k Eia + Eka ∂i ξ k
G0 = H − ∂i (E0c Πic ) ≈ 0 , (32)
= ∂i (Eka ξ k ) + ξ k 2 ∂[k Ei]a . (40)

(2) . In these results there is a term resembling the gauge

Gk = ∂k Eic Πic − ∂i (Ekc Πic ) ≈ 0 (33)
(the symbol ≈ stands for equalities that are valid on the
constraint surface). The constraints (32) and (33) are
equivalent to the super-Hamiltonian and super-momenta
constraints of the ADM formalism. While the ADM
Hamiltonian vanishes on the constraint surface, the
TEGR Hamiltonian does not. The reason can be traced
to the surface term in Eq. (12); in fact, according to
Eq. (32), H is not zero but a divergence (which became
a spatial divergence thanks to the constraints (24)).
transformation of the electromagnetic potential. How-
ever, they come together with a term related to the
Weitzenböck torsion Ta = dEa . Both terms are needed
because, differing from the electromagnetic Lagrangian,
TEGR Lagrangian depends not only on the exterior
derivative of the field Ea but on the field itself. Even
so, the whole result exhibits a clear geometric content,
which can be evidenced by means of the Lie derivative of
a p−form α along a vector ξ,
Lξ α = d[α(ξ)] + dα(ξ) . (41)

V. GAUGE TRANSFORMATIONS
g
2 Since E = εab...g E0a E1b ... En−1 , where εab...g is the Levi-
We have already anticipated –although not proven yet– g
Civita symbol, we also obtain eh δE = e0h ǫa εab...g E1b ... En−1
0

that all the constraints will be first class. So, let us con- a ν b g 0
= −Eν δeh εab...g E1 ... En−1 = −E δeh . Therefore eh E is 0
(1) invariant under the transformation (37).
sider the gauge transformations generated by Ga and

In fact, the r.h.s. of Eqs. (39) and (40) constitute the
spatial components of Lξ Ea , where ξ is the arbitrary vec-
tor field formed by the infinitesimal parameters ξ(t, x),
ξ k (t, x). We notice that Eqs. (39) and (40) can be ex-
tended to the temporal component of the 1-forms Ea ,
since any change of E0a is a gauge transformation. There-
fore, we have obtained that TEGR is insensitive to 2n in-
dependent gauge transformations of the vielbein on the
constraint surface, which are given by Eq. (37) and
δEa = Lξ Ea . (42)
The derivative character of transformation (42) together
with Eq. ( 35) imply that
δeb = Lξ eb = [ξ , eb ] . (43)
In turn, this last transformation leads to a change of the
anholonomy coefficients:
c
δfab c
= Lξ fab c
= ξ(fab ), (44)
as can be easily verified by using the Jacobi identity to
compute δ[ea , eb ] = [δea , eb ] + [ea , δeb ].
We remark that the Lie derivative of any Lagrangian
–understood as the n− form L = L dx0 ∧ ... ∧ dxn−1 ,
where L is the Lagrangian density– is always a boundary
term. In fact, if α is a n−form in Eq. (41), then its Lie
derivative Lξ α is the exact form d[α(ξ)]. But in a theory
of gravity, like TEGR, this kind of (quasi-) invariance of
the Lagrangian comes from a symmetry of its dynamical
variables generated by a proper combination of the trivial
primary constraints and the secondary ones. In fact, the
change of the TEGR Lagrangian n−form,
1 v ae Cab ef = 0 .
L = a
E fce b
fdf Mab cedf dx0 ∧ ... ∧ dxn−1
8 (1) .
1 a b
= f f M cedf E0 ∧ ... ∧ En−1 , (45)
8 ce df ab
(we used that the vielbein is orthonormal to rewrite the
volume) under the gauge transformation (42) is equal to
its Lie derivative by virtue of Eqs. (42) and (44):
1
δL = δf a f b M cedf E0 ∧ ... ∧ En−1
4 ce df ab
1 a b
+ fce fdf Mab cedf δE0 ∧ ... ∧ En−1 + ...
8 We will introduce an independent set of coefficients v ae
= Lξ L = d[L(ξ)] . (46)
VI. MORE PRIMARY CONSTRAINTS. THE
LORENTZ GAUGE GROUP
So far we have found the 2n constraints that reflect
the constraint structure of the ADM formulation of gen-
eral relativity. However, TEGR describes the n(n + 1)
components of the metric tensor through a n × n ma-
trix Eµa . The relation between both sets of dynamical
6
variables is given by the Eq. (5), which is invariant un-
der local Lorentz transformations of the vielbein. Since
we know that TEGR has dynamics only for the met-
ric, as is clear from the equivalence between TEGR and
Einstein-Hilbert Lagrangians expressed in Eq. (12), the
local Lorentz symmetry has to be a property not only of
the relation (5) but the set of dynamical equations. Then,
we should find that Lorentz transformations in the tan-
gent space constitute a gauge group in TEGR. Therefore,
we will search for more primary constraints in Eq. (21 ).
Eq. (21) is a system of n2 equations that are not lin-
early independent. In the previous Section we have al-
ready shown that they contain a set of n constraints that
trivially emerge for µ = 0. The existence of constraints
associated to the temporal coordinate index is a conse-
quence of the privileged character the temporal coordi-
nate plays in the canonical formalism. We expect that
the rest of the primary constraints are exclusively re-
lated to tangent space indices. Therefore, we will look
for constraints among the coordinate invariant combina-
tions Πµa Eµe ; according to Eq. (21) they are
Πµa Eµe = E Cab ef eλf ∂0 Eλb + E ∂i Eλb e0c eid eλf Mab cedf ,
(47)
ef
where Cab is defined as
.
Cab ef = e0c e0d Mab cedf . (48)
To find constraints (relations among the canonical vari-
ables) in Eq. (47), we should find (vielbein-depending)
coefficients v ae such that v ae Πµa Eµe does not contain
canonical velocities. In other words, since the square
matrix eλf is not singular, it should be
(49)
Notice that even the n trivial primary constraints Gg =
(1)
Π0g can be recovered in this way. In fact Gg requires
. 0 a
a
coefficients v|g| e = ee δg (the index between vertical bars
is a label for each independent set of coefficients), since
e0e δga Πµa Eµe = Π0g . On the other hand, these coefficients
satisfy Eq. (49), because Mgb cedf is antisymmetric in c−e:
a
v|g| ef
= e0e e0c e0d Mgb cedf ≡ 0 .
e Cab (50)
leading to the primary constraints associated with the
Lorentz group. Let be the set of coefficients v ae labeled
by gh
.
a
v|gh|ae = 2 δ[g ηh]e . (51)
Taking into account the form (17) of the supermetric, we
obtain
v|gh|ae Cab ef = 2 e0c e0d ηe[h Mg]bcedf = 4 e0c e0d δhgb
cdf
≡ 0,
(52)
 7

cdf
since δhgb is completely antisymmetric (see Eq. (A5) VII. TEGR CANONICAL HAMILTONIAN
for details of the calculation). The antisymmetric labels
gh classify n(n − 1)/2 new constraints. By combining We will fully exploit the multi-index notation intro-
Eqs. (47), (52) and ((A5)), one gets duced at the end of the previous Section. For this, we
define a set of objects of n2 components:
0 ≡ v|gh|ae (Πµa Eµe − E ∂i Eλb e0c eid eλf Mab cedf )
. .
= 2 ηe[h Πµg] Eµe + 4 E ∂i Eλb e0[h eig eλb] . (53) Ė B = eλf Ėλb , E0B = eif ∂i E0b ,
. .
In the last line, λ can be replaced with j due to the ΠA = Πµa Eµe , PA = E ∂i Ekb e0c eid ekf Mab cedf . (60)
antisymmetrization of the pair h − b. Besides, on the Thus the Lagrangian density (16) reads
constraint surface it is Π0g = 0. So, we define the primary
constraints 1
(1) .
L= (ΠA + PA )(Ė A − E0A ) − U , (61)
Ggh = 2 ηe[h Πig] Eie + 4 E ∂i Ejb e0[h eig ejb] ≈ 0 . (54) 2
where
In Section VIII we will prove that these n(n − 1)/2 con-
straints accomplish the Lorentz algebra. Besides, they . 1
will be consistent with the evolution. The entire set U = − E ∂i Eja ∂k Elb eic eje ekd elf Mabcedf . (62)
2
of constraints will prove to be first class. According to
Eq. (34), the gauge transformation of the vielbein gener- Therefore, the canonical Hamiltonian density turns out
(1)
ated by a combination ǫgh Ggh is to be
.
Z H = Πµa Ėµa − L = ΠA Ė A − L
δEja (t, x) = dy ǫgh (t, y) {Eja (t, x), 2 ηe[h Πig] (t, y) Eie } , 1 1
= (ΠA − PA ) Ė A + (ΠA + PA ) E0A + U. (63)
(55) 2 2
which can be extended to the component E0a by virtue
of the gauge transformation (37), so leading to the local To write H in a canonical way, the velocities Ė B must
Lorentz transformation be solved in terms of the momenta. Eq. (47) displays
the linear relation among velocities and momenta; this
δEa = ǫgh (t, x) ηeh δga − ηeg δha Ee .


(56) equation now reads

ΠA − PA = E CAB (Ė B − E0B ) . (64)

At this point, one could ask whether we have exhausted
the solutions to Eq. ( 49). We remark that Cab ef can In Eq. (64) Ė B cannot be straightforwardly solved be-
be rephrased as a symmetric n2 × n2 matrix by using a cause the matrix CAB is singular. Matrix CAB has
notation that take pairs of flat indices a, b, ... to define a n(n + 1)/2 null eigenvalues, since there are n(n + 1)/2
a
multi-index A = () e such that the Eq. (49) becomes primary constraints linear in the momenta. In spite of
v A CAB = 0 (57) the fact that CAB is not invertible, we can still solve
the subspace of velocities that is orthogonal to the sub-
For this, we use the following indexation formulas for space of null eigenvalues. In fact, by using a proper basis
a b
A = () e , B = () f 3 for splitting the subspace of null eigenvalues, CAB would
look like
A = (a − 1) n + e, B = (b − 1) n + f ; (58)  
2 ′ 0 0
so, A, B, ... = 1, ..., n . Eq. (A1) implies the symmetry of CAB = , (65)
0 C̃
CAB :
CAB = CBA . (59) In such a basis we would find n(n+1)/2 constraints ΠA −
PA = 0; besides, we would trivially solve n(n − 1)/2
Eq. (57) means that there are as many linear constraints relevant velocities,
as null eigenvalues the symmetric n2 ×n2 matrix CAB has.
The coefficients v A = v ae of the constrained combinations Ė A − E0A = E −1 D′AB (ΠB − PB ) , (66)
v ae Πµa Eµe are the components of the respective eigenvec-
tors. So far we have found n + n(n − 1)/2 = n(n + 1)/2 where the matrix D′ is
null eigenvalues. As we will see in the forthcoming sec-  
tions, the other n(n − 1)/2 eigenvalues are different from 0 0
D′ = , (67)
zero. 0 D̃

and satisfies
3
 
The formula can be inverted by taking a = [A/n], so e = A − ′ ′ ′ ′ 0 0
n[A/n] − 1, where [ ] means the integer part. D C =C D = . (68)
0 1
 8

Eq. (66) declares null the n(n + 1)/2 first velocities. This This is a consequence of the fact that the non-null eigen-
causes no harm, since these velocities enter the Hamil- values have the same absolute value. This means that
tonian (63) as the coefficients of the primary constraints the pseudo-inverse of C AB is DAB = λ−2 C AB . Therefore,
ΠA − PA = 0. So, the values of the n(n + 1)/2 first veloc- the matrix DAB in Eq. (71) is
ities are irrelevant, because different choices modify the
Hamiltonian by terms proportional to the constraints. DAB = λ−2 C AB = λ−2 e0g e0h M abge hf . (76)
Anyway this kind of terms are reintroduced in the pri-
mary Hamiltonian (25).
Let us use the matrix N of change of basis to return B. Dimension n > 3
to the original basis: C ′ = N CN −1 . Then, the previous
equation becomes In n = 4 dimensions, the matrix C AB has six non-null
  eigenvalues; they are
0 0
N −1 D′ N C = C N −1 D′ N = N −1 N. (69) λ1 = λ2 = λ3 = λ4 = λ5
0 1
.
= 2 [(e00 )2 − (e01 )2 − (e02 )2 − (e03 )2 ] = 2 g 00 = λ ,
The r.h.s. is not the identity, but is a symmetric matrix.
λ6 = −2 λ . (77)
Besides, the matrix
. Since their absolute values are not equal, the pseudo-
D = N −1 D′ N (70) inverse matrix DAB cannot be inferred in a so straight-
forward way as we did in n = 3 dimensions. In fact, ma-
satisfies that CDC = C and DCD = D. Therefore D is
trix C AB does not accomplishes the Eq. (75) when n > 3.
the Moore-Penrose pseudo-inverse of C. We will use the
The eigenvector related to the odd eigenvalue is
Eq. (66) in the original basis; so we must replace D′ with
D. Thus, we substitute Eq. (66) in Eq. (63) to obtain λ b
the canonical form of the Hamiltonian density: wB = wbf = − δ + e0f η bh e0h . (78)
2 f
1 In fact, in any dimension n, vector wB satisfies the eigen-
H= e (ΠA − PA )DAB (ΠB − PB ) + ΠA E0A + U, (71)
2 value equation
where e = E −1 = det(eµa ). The canonical Hamiltonian is C AB wB = e0g e0h M ab ge hf wbf
the integral of H. We can remind the form (32) of the = −(n − 2)λ wae = −(n − 2) λ wA . (79)
(2)
constraint G0 to write
Z Z Z We will show that the pseudo-inverse of C AB can be for-
(2) mulated as the matrix
H= dx H = dx G0 + E0c Πic dSi . (72)
DAB = λ−2 (C AB + α wA wB ) , (80)
Then, the canonical Hamiltonian is a constraint plus a
boundary term. As a consequence, the set of first class where α is a factor to be determined. The idea is to use
constraints will be automatically consistent with the evo- the projector associated to the odd eigenvalue to “im-
lution. prove” the matrix C AB and get the desired result. In
order that DAB be the pseudo-inverse of C AB , the r.h.s.
of the equation
A. Dimension n = 3
C AC DCD C DB = λ−2 C AC C CD C DB + α(n − 2)2 wA wB (81)
should be C AB . To find α, we will introduce the auxiliary
Let us work with the matrix C AB , matrix

C AB = C aebf = e0c e0d M ab ge hf , (73) C̃ AB = C AB + 4 λ−1 wA wB , (82)

which satisfies
where M ab ge hf = η ac ηde Mcb gdhf . CAB and C AB share
the eigenvectors of null eigenvalue (see the Appendix A 3 C̃ AB wB = C AB wB + 4 λ−1 wA wB wB
for the forms of these matrices). The non-null eigenvalues = −(n − 2) λ wA + (n − 1) λ wA = λ wA . (83)
of C AB are
. Besides, for any vector ℓB orthogonal to wB it is
λ1 = λ2 = 2 [(e00 )2 − (e01 )2 − (e02 )2 ] = 2 g 00 = λ , C̃ AB ℓB = C AB ℓB = λ ℓA . Then, C̃ AB is isotropic in the
λ3 = −λ . (74) subspace of non-null eigenvalues. 4 Since all the non-null

The case n = 3 is very simple because the matrix C AB

accomplishes
4 This is true not only for n = 4, but it has been checked for
2
C AB C BC C CD =λ C AD . (75) arbitrary n through the computer algebra program Cadabra [21].
 9

eigenvalues of C̃ AB are equal to λ, then C̃ AB accomplish However, the last one requires the knowledge of the
the Eq. (75). Therefore brackets between the momenta Π0a and the matrix DAB .
In the Appendix A 4 we summarize useful hints in order
λ2 (C AB + 4 λ−1 wA wB ) = λ2 C̃ AB = C̃ AC C̃ CD C̃ DB to simplify this calculation.
= C AC C CD C DB + 4 λ n2 − 5n + 7 wA wB ,


(84) The Poisson brackets between secondary constraints
(2)
i.e., Gµ reproduce the algebra of constraints of the ADM
formulation of general relativity:
λ−2 C AC C CD C DB = C AB −4 λ−1 (n−3)(n−2)wA wB . (85)
(2) (2)
{Gi (t, x), Gj (t, y)} = (94)
Substituting this result in Eq. (81), we get that DAB is
(2) (2)
the pseudo-inverse of C AB if α has the value −Gi (x) ∂jy δ(x − y) + Gj (y) ∂ix δ(x − y) ,
(2) (2) (2)
4 (n − 3) {G0 (t, x), G0 (t, y)} = g ij (x) Gi (x) ∂jy δ(x − y)
α = λ−1 . (86)
(n − 2) (2)
− g ij (y) Gi (y) ∂jx δ(x − y) , (95)
In n = 4 dimensions, α is equal to 2λ−1 . Thus the (2) (2) (2)
{G0 (t, x), Gi (t, y)} = G0 (x) ∂iy δ(x − y) . (96)
contravariant pseudo-inverse matrix DAB = λ−2 (C AB +
α wA wB ) in four dimensions is We have also verified that the Poisson brackets for the
(1)
[a 1 ab constraints Gab reproduces the Lorentz algebra:
DAB = Dabef = λ−1 (δf δeb] + η ηef )
2 {G(1)
(1)
−2 [a ac (t, x), Gf e (t, y)} = (97)
−λ (e0e e0f η ab + 4 e0g e0[e δf ] η b]g  
(1) (1)
ηec Gaf + ηaf G(1) (1)
ce − ηcf Gae − ηae Gcf δ(x − y) .
+e0g e0h η ag η bh ηef )
+2 λ−3 η ag η bh e0g e0h e0e e0f . (87) Besides it is
(1) (2)
{Gab (t, x), Gi (t, y)} = 0 . (98)

Finally the most intricate calculation is required by the

VIII. ALGEBRA OF CONSTRAINTS bracket
(2) (1) (2)
{G0 (t, x), Gab (t, y)} = E0c ηc[a e0b] G0 δ(x − y) . (99)
The Hamiltonian formalism for TEGR is not finished
without checking that the set of constraints is first class. In order to alleviate some difficult parts of it, some useful
For this, we have to compute the entire set of Poisson computations are summarized in Appendix A 4.
brackets between the constraints. The pseudo-inverse As a result we have got n trivial primary constraints
(1)
matrix DAB will enter the algebra of those Poisson brack- Ga , together with n(n − 1)/2 primary constraints that
(2)
ets involving the constraint G0 . It is worth mentioning come from the Lorentz algebra. Besides, we have ob-
(2)
that Eq. (78) can be replaced in the l.h.s of Eq. (79) to tained n secondary constraints Gµ that are equivalent to
obtain the super-Hamiltonian and super-momenta constraints of
1 the ADM formalism. Since we just proved that all con-
C abeb = wae = wA . (88) straints are first class, then the counting of degrees of
2(n − 2) freedom goes as
Therefore, matrix D in Eq. (80) can be written entirely
# d.o.f. = # (p,q) − # f.c.c.
in terms of the matrix C as
n(n + 3) n(n − 3)
λ−3 (n − 3) a c b d = n2 − = (100)
Dabef = λ−2 C abef + C ce C df , (89) 2 2
(n − 2)3
which is the number of degrees of freedom of general rel-
which will be useful to compute those brackets involving ativity in n dimensions.
(2)
G0 .
(1) IX. SUMMARY
The simplest brackets are those related to Ga :
(1)
{G(1)
a (t, x), Gb (t, y)} = 0, (90)
(2) The essence of a Hamiltonian constrained system lies
{Gi (t, x), G(1)
a (t, y)} = 0, (91)
(1) in the impossibility of solving all the canonical velocities
{Gab (t, x), G(1)
c (t, y)} = 0, (92) in terms of canonical momenta. This is because the mo-
(2) menta are not independent but satisfy constraint equa-
{G0 (t, x), G(1)
a (t, y)} =

0 (2) (2)
 tions, which in turn means that some dynamical vari-
ea G0 + eia Gi δ(x − y). (93) ables are spurious degrees of freedom. In the case of the
 10

teleparallel equivalent of general relativity (TEGR), such Appendix A

obstruction is expressed in the Eq. (47), since Cab ef can-
not be inverted. Cab ef is an object intimately linked to 1. Properties of the supermetric
the Lorentz invariant supermetric Mabcedf entering the
TEGR Lagrangian (20). In order to analyze how many There are many properties of the supermetric that were
constraints are involved in the Eq. (47), and how many used throughout this work, and can be deducted from its
canonical velocities can be solved, we have arranged the definition. Some of them are
components of Cab ef in a n2 × n2 symmetric matrix
CAB (the relation between the superindex A and the tan- Mabcedf = Mbadf ce = −Mabecdf = −Mabcef d . (A1)
gent space indices is given in Eq. (58)). We have shown
We can calculate “traces” of the supermetric, which de-
that the eigenvalues of C AB follow a very simple pattern:
pend on the dimension n. Some of them are
n(n + 1)/2 eigenvalues are null, n(n − 1)/2 − 1 of them
.
are equal to 2 g 00 = λ, and the remaining one is equal Mabaedf = Mbadf ae = 4(n − 2) η e[d δb ,
f]
(A2)
to (2 − n) λ. The primary constraints results from the d]
contraction of the Eq. (64) with each eigenvector of null Mabdf ae = Mbaaedf = 2(n − 2)η e[f δb , (A3)
(1)
eigenvalue; they include the n trivial constraints Ga (see Mabaebf = −2(n − 1)(n − 2) η ef
, (A4)
(1)
Eq. (24)) and the n(n−1)/2 Lorentz constraints Gab (see ghf
Eq. (54)). To build the canonical Hamiltonian we must The totally antisymmetric Kronecker delta δcab appears
identify the subset of canonical velocities that can be still in the antisymmetrized product
solved in terms of the momenta. For this, we employed h f g
ηe[c Ma]bgehf = 2 (δ[a g h f
δc] δb + δ[a f g h
δc] δb + δ[a δc] δb )
the Moore-Penrose pseudo-inverse of matrix C , which
can be sought in the form proposed in Eq. (80) thanks to . ghf
= −2 δcab . (A5)
the simple pattern of eigenvalues exhibited by the matrix
C. The so obtained matrix DAB is the piece we need to We also obtain
write the canonical Hamiltonian density H (see Eq. (71)).
M abge hf ηa[q δp]
e
= η bg ηf [q δq]
h g
+ η bh ηf [q δp] g h
+ δfb δ[q δp] , (A6)
Those terms associated with the unsolved velocities are
d]b
absorbed into the terms added to the primary Hamil- η e[c C ef = 4 e0f e0[c η d]b . (A7)
(1)
tonian Hp (25). Besides the primary constraints Ga ,
(1) (2) Some other combinations quadratic in M appear in the
Gab , we have also obtained n secondary constraints Gµ calculations, and it is useful to have them on hand
–the diffeomorphism constraints– that guarantee that
the primary constraints remain valid along the evolution C acec Magcedf = 4(n − 3)(n − 2) g 00 η c[d δgf ] (A8)
dictated by Hp (we have examined this consistency at
the level of the Euler-Lagrange equations). The consis- + 8 (n − 2) e0g e0[f η d]c + e0c e0[d δgf ]
tency under the evolution of the system must be checked Cabef M abce df = 6 (n − 3)(n − 2) η cd g 00 (A9)
with the secondary constraints too. Not surprisingly, the 0c 0d
(2) + 12 (n − 2) e e .
canonical Hamiltonian density H is equal to G0 except
for a boundary term. Thus, the consistency of the entire
set of constraints is guaranteed by the first class con- 2. Calculation of ∂L/∂E0a
straint algebra (90-99). Since the constraints are first
class, they generate gauge transformations. Therefore,
For computing ∂L/∂E0a , it is important to notice that,
there are n(n + 3)/2 spurious variables, what reduces the
contrarily to electromagnetism, E0a appears in the La-
number of degrees of freedom to n(n − 3)/2. The in-
grangian not just in the spatial derivatives ∂i E0a but also
dependent gauge transformations are those displayed in
as a part of eµa and E. First of all we need the quotient
Eqs. (37), (42), and (56).
∂eµc /∂Eλa , which is obtained from the duality relation:

∂eµb b
δνµ = eµb Eνb → 0= E + eµa δνλ . (A10)
∂Eλa ν

this implies that

ACKNOWLEDGMENTS ∂eµc
= −eµa eλc . (A11)
∂Eλa
The authors thank N. Deruelle and C. Bejarano for We will need also the expression ∂E/∂E0a , which is ob-
helpful discussions. This work was supported by Con- tained from the explicit formula for the determinant
sejo Nacional de Investigaciones Científicas y Técnicas
(CONICET) and Universidad de Buenos Aires. E = ǫabcd...g E0a E1b E2c E3d ...Eng , (A12)
 11

Then we obtain The Hamiltonian density can be extracted from the first
terms, to obtain
∂E ∂E ∂L
Eλa = δλ0 E → = E e0a . (A13) = −e0a H + e0a ∂i E0c Πic + 2 eja ∂[i Ej]
c
Πic (A16)
∂E0a ∂E0a ∂E0a
= e0a (∂i (E0c Πic ) − H) − E0c e0a ∂i Πic + 2 eja ∂[i Ej]
c
Πic .
In this way, This result is replaced in Eq. (30) to obtain n secondary
constraints:

∂L 1 Ejc eja ∂i Πic + e0a (∂i (E0c Πic ) − H) + 2 eja ∂[i Ej]
c
Πic ≈ 0 .
= E (e0a eµg eνe eρh eλf − eµa e0g eνe eρh eλf − eνa eµg e0e eρh eλf (A17)
∂E0a 2
We note that only spatial derivatives are present, and the
−eρa eµg eνe e0h eλf − eλa eµg eνe eρh e0f ) ∂µ Eνc ∂ρ Eλd Mcdgehf .(A14) canonical Hamiltonian takes part in the secondary con-
straints. We can isolate the contribution of the Hamilto-
nian by doing the contraction with E0a ; thus we get
In the last expression we identify the Lagrangian in the
(2)
first term, and different index combinations of the mo- G0 = H − ∂i (E0c Πic ) ≈ 0 . (A18)
menta. We rewrite it and continue with the algebraic
manipulation Besides, we perform the contraction with Eka , so yielding
(2)
Gk = ∂k Eic Πic − ∂i (Ekc Πic ) ≈ 0 . (A19)
∂L 1 1
a = e0a L − eµa ∂µ Eνc Πνc + eνa ∂µ Eνc Πµc
∂E0 2 2
3. Matrix C AB
1 ρ 1
− ea ∂ρ Eλd Πλd + eλa ∂ρ Eλd Πρd
2 2
0 We present the full expression for the matrix C AB in
= ea L + 2 ea ∂[µ Eν] Πµc
ν c
(A15)
n = 4, which appears in the definition of the canonical
= e0a L + 2 e0a ∂[i E0]
c
Πic + 2 eja ∂[i Ej]
c
Πic . momenta. It is

0 0 0 0 0 2c23 0 2c13 0 2c12

 
−2d12 −2d13 −2d12 −2d23 −2d13 −2d23
 0 −c23 d12 d13 c23 0 −d02 −d03 −d12 −d02 2d01 0 −d13 −d03 0 2d01 
 
 
 0 d12 −c13 d23 −d12 2d02 −d01 0 c13 −d01 0 −d03 −d23 0 −d03 2d02 
 
 0 d13 d23 −c12 −d13 2d03 0 0 2d03 c12 0 
 
−d01 −d23 −d02 −d01 −d02
 
 0 c23 −d12 −d13 −c23 0 d02 d03 d12 d02 −2d01 0 d13 d03 0 −2d01 
 
 2c23 0 0 0 0 0 2d02 0 −2c03 −2d23 2d03 0
 
−2d02 −2d03 −2d23 −2c02 
 
 −2d
 12 d02 d01 0 −d02 0 c03 d23 −d01 c03 0 d13 0 d23 d13 −2d12 

 
 −2d13 d03 0 d01 −d03 0 d23 c02 0 d23 −2d13 d12 −d01 c02 d12 0 
C AB = (A20)
 
 0 c13 d12 −2d02 d01 0 d01 0 d03 d23 0 d03

−d12 −d23 −c13 −2d02 
 
 −2d12
 d02 d01 0 −d02 0 c03 d23 −d01 c02 0 d13 0 d23 d13 −2d12 

 2c13 0 −2d03 2d01 −2c03 0 0 0 0 0 2d03 −2d13 0
 
−2d01 −2d13 −2c01 
 
 −2d
 23 0 d03 d02 0 −2d23 d13 d12 −d03 d13 0 c01 −d02 d12 c01 0  
 
 0 −d13 −d23 c12 d13 −2d03 0 d01 d23 0 −2d03 d02 −c12 d01 d02 0 
 
d03 0 d01 0 d23 c02 0 d23 d12 c02 d12 0 
 
 −2d13 −d03 −2d13 −d01
 
 −2d23
 0 d03 d02 0 −2d23 d13 d12 −d03 d13 0 c01 −d02 d12 c01 0  
2c12 −2d01 −2d02 0 2d01 −2c02 −2d12 0 2d02 −2d12 −2c01 0 0 0 0 0

where The matrices CAB and C AB are obtained by raising and

c01 = (e00 )2 − (e01 )2 , c02 = (e00 )2 − (e02 )2 , lowering indices with the corresponding η tensors. The
matrix DAB is obtained starting from (80).
c03 = (e00 )2 − (e03 )2 , c12 = (e01 )2 + (e02 )2 ,
c13 = (e01 )2 + (e03 )2 , c23 = (e02 )2 + (e03 )2 , (A21)
d01 = e00 e01 , d02 = e00 e02 , d03 = e00 e03 ,
d12 = e01 e02 , d13 = e01 e03 , d23 = e02 e03 .
 12

4. Poisson brackets tions, are the following

Some useful fundamental Poisson brackets between the {λ−γ , Π0c } = 2 γ λ−γ e0c , (A29)
−γ −(γ+1)
canonical variables and their derivatives are given below, {λ , Πic } = 4 γλ g , e0c 0i
(A30)
{E(t, x), Πµa (t, y)} = E eµa δ(x − y), (A22) {wA , Π0c } 0 A
= −2 ec w , (A31)
1
{e(t, x), Πµa (t, y)} = e eµa δ(x − y), (A23) {wA , Πic } = − e0 (ei e0 + eih e0g ) M ad ge hd .(A32)
2(n − 2) c g h

{eµa (t, x), Πνb (t, y)} = − eµb eνa δ(x − y) , (A24) Finally, we give some help to calculate the brackets of
the momenta and the matrix DAB . It is very simple to
get the brackets
{∂λ Eµa (t, x), Πνb (t, y)} = (A25)
−{Eµa (t, x), ∂λ Πνb (t, y)} = δba δµν ∂λx δ(x − y) , {DAB , Π0c } = 2 e0c DAB . (A33)

Z However for the spatial part of the momenta Πia , the

{∂µ Eνb (t, x), ∂λ Πλc (t, y)} = dz δcb ∂µx δ(x−z) ∂νy δ(y −z). brackets with the matrix D do not simplify so easily.
After using all the developed tools, we get
(A26)
These expressions are enough (together with patience {DAB , Πic } = 8 e0c g 0i λ−3 C AB + α wA wB


and a lot of calculations) to calculate those Poisson brack-
(2)
ets that do not involve G0 . For the remaining Poisson −λ−2 e0c (eig e0h + eih e0g ) M abge hf
brackets, we provide some easy-to-derive expressions α λ−2 e0c i 0
− (e e + eih e0g ) (M ad ge hd wB + M bd gf hd wA )
(2) 2(n − 2) g h
{G0 (t, x), Eic (t, y)} = (A27)
AB +4 α λ−3 e0c g 0i wA wB . (A34)
PB ) Eie δac ∂i E0c


eD (ΠB − + δ(x − y) ,
(2) In Eqs. (A29-A34) a factor δ(x − y) is understood. As
{G0 (t, x), ∂λ Eµc (t, y)} = (A28)
y a general advice, the raising and lowering of indices in the
e DAB (ΠB − PB ) Eµe δac + ∂µx E0c (x) ∂λ δ(x − y) .


supermetric Mabcedf must be carefully done, to keep the
Other combinations of brackets between canonical mo- original order of the indices and protect the symmetries
menta and some basic building blocks of the secondary of the object.
(2)
constraint G0 , that recurrently appear in the calcula-

[1] P.A.M. Dirac, Lectures on Quantum Mechanics. New (1994).

York: Belfer Graduate School of Science, Yeshiva Uni-
versity (1964).
[2] K. Sundermeyer, Constrained dynamics, vol. 169 of Lec-
ture Notes in Physics, Springer, Berlin, Germany (1982).
[3] R. Arnowitt, S. Deser and C.W. Misner, The Dynamics
of General Relativity, in Gravitation: an introduction to
current research, L. Witten, ed. (Wiley, New York, 1962);
reprinted in Gen. Relativ. Gravit. 40, 1997-2027 (2008).
[4] K.V. Kuchař, Time and interpretations of quantum grav-
ity, in the Proceedings of the 4th Canadian Conference
on General Relativity and Relativistic Astrophysics, eds.
G. Kunstatter, D. Vincent and J. Williams (World Sci-
entific, Singapore, 1992).
[5] H. Weyl, Gravitation und Elektrizität. Sitzungsber.
Preuss. Akad. Wissensch., p. 465 (1918).
[6] A. Einstein, Pruess. Akad. Wiss., 414 (1925); Sitzungs-
ber. Preuss. Akad. Wiss. Phys. Math. Kl. 217 (1928); 401
(1930); Math. Ann. 102, 685 (1930). English translation
in arXiv:physics/0503046.
[7] E. Cartan, C. R. Acad. Sci. Paris 174, 593 (1922); 174,
734 (1922).
[8] J.W. Maluf, Hamiltonian formulation of the teleparallel
description of general relativity, J. Math. Phys. 35, 335
[9] J.W. Maluf, J.F. Da Rocha-Neto, General Relativity on
a Null Surface: Hamiltonian Formulation in the Telepar-
allel Geometry, Gen. Relativ. Gravit. 31, 173-185 (1999).
[10] M. Blagojević and I.A. Nikolić, Hamiltonian structure of
the teleparallel formulation of general relativity, Phys.
Rev. D 62, 024021 (2000).
[11] J. Maluf and J.F. da Rocha-Neto, Hamiltonian formu-
lation of general relativity in the teleparallel geometry,
Phys. Rev. D 64, 084014 (2001).
[12] J.F. da Rocha Neto, J.W. Maluf and S.C. Ulhoa, Hamil-
tonian formulation of unimodular gravity in the telepar-
allel geometry, Phys. Rev. D 82, 124035 (2010).
[13] J.W. Maluf, The teleparallel equivalent of general rela-
tivity, Ann. Phys. (Berlin) 525, 339-357 (2013).
[14] A. Okołów, ADM-like Hamiltonian formulation of gravity
in the teleparallel geometry, Gen. Relativ. Gravit. 5, 2569
(2013).
[15] A. Okołów, ADM-like Hamiltonian formulation of grav-
ity in the teleparallel geometry: derivation of constraint
algebra, Gen. Relativ. Gravit. 6, 1636 (2014).
[16] K. Hayashi and T. Shirafuji, New General relativity,
Phys. Rev. D 19, 3524-3552 (1979).
[17] T. Ortín, Gravity and Strings (Cambridge University

Press, Cambridge, 2004).
[18] Y.M. Cho, Einstein Lagrangian as the translational
Yang-Mills Lagrangian, Phys. Rev. D 14, 2521 (1976).
[19] R. Weitzenböck, Invarianten Theorie (Noordhoff,
Groningen, 1923).
[20] J.L. Anderson and P.G. Bergmann, Constraints in co-
13
variant field theories, Phys. Rev. 83, 1018-1025 (1951).
[21] K. Peeters, Introducing Cadabra: A Symbolic computer
algebra system for field theory problems, hep-th/0701238
(2007).