Graviton propagator from background–independent quantum gravity

Carlo Rovelli
Centre de Physique Théorique de Luminy, Case 907, F-13288 Marseille, EU
Dipartimento di Fisica Università La Sapienza, P.A. Moro 2, I-00185 Roma, EU
(Dated: August 15, 2008)
We study the graviton propagator in euclidean loop quantum gravity, using the spinfoam formalism.
We use boundary–amplitude and group–field–theory techniques, and compute one component of
the propagator to first order, under a number of approximations, obtaining the correct spacetime
dependence. In the large distance limit, the only term of the vertex amplitude that contributes
is the exponential of the Regge action: the other terms, that have raised doubts on the physical
arXiv:gr-qc/0508124v2 23 Dec 2005

viability of the model, are suppressed by the phase of the vacuum state, which is determined by the
extrinsic geometry of the boundary.

An open problem in quantum gravity is to compute particle scattering amplitudes from the
background–independent theory and recover low–energy physics [1]. The difficulty is that general
covariance makes conventional n-point functions ill–defined in the absence of a background. A strat-
egy for addressing this problem has been suggested in [2]; the idea is to study the boundary amplitude,
namely the functional integral over a finite spacetime region, seen as a function of the boundary value
of the field [3]. In conventional quantum field theory, this boundary amplitude is well–defined (see
[4]) and codes the physical information of the theory; so does in quantum gravity, but in a fully
background–independent manner [5]. A generally covariant definition of n-point functions can then
be based on the idea that the distance between physical points –arguments of the n-point function– is
determined by the state of the gravitational field on the boundary of the spacetime region considered.
In this paper we implement this strategy.
We compute a first order term of the (connected) two-point function, starting from full non-
perturbative quantum general relativity. To this order, the computation produces a 4d version of
the “nutshell” 3d model studied in [6]. Using a natural gaussian form of the vacuum state, peaked on
the intrinsic as well as the extrinsic geometry of the boundary, we derive an expression for a compo-
nent of the graviton propagator. At low energy, this agrees with the conventional graviton propagator.
Other components and second order terms will be presented elsewhere [7]. Our main motivation is
to show that a technique for computing particle scattering amplitudes in background–independent
theories can be developed. (The viability of the notion of particle in a finite region is discussed in [8].
For the general relativistic formulation of quantum mechanics underlying this calculation, see [9]. On
the relation between graviton propagator and 3-geometries transition amplitudes in the conventional
perturbative expansion, see [10].)
We use some standard loop quantum gravity (LQG) results [9, 11], as well as a specific spinfoam
model [9, 12]: the GF T /B theory, in the terminology of [9], defined using group–field–theory methods.
The results do not change using the GF T /C model. These are tentative background–independent
quantizations of euclidean general relativity. The first was introduced in [13] and is favored by some
recent arguments [14, 15]. The second was introduced in [16] (see also [17]) and has good finiteness
properties [18].
The physical correctness of these theories has been questioned because in the low–energy limit their
interaction vertex (10j symbol, or Barrett-Crane vertex amplitude) has been shown to include –beside
the “good” term approximating the exponential of the Einstein-Hilbert action [19]– also two “bad”
terms: an exponential with opposite sign, giving the cosine of Regge action [19] (analogous to the
cosine in the Ponzano–Regge model) and a dominant term that depends on the existence of degenerate
four-simplices [20, 21]. We show here that only the “good” term contributes to the propagator. The
others are suppressed by the rapidly oscillating phase in the vacuum state that peaks the state on its
correct extrinsic geometry. Thus, the physical state selects the “forward” propagating [22] component
of the transition amplitude. This phenomenon was anticipated in [23].
 2

Following [2], the 2-point function of quantum general relativity can be extracted from a nonper-
turbative formalism using the expression
X
W abcd (x, y; q) = N W [s′ ] hs′ |hab (x)hcd (y)|si Ψq [s]. (1)
ss′

Here q is the intrinsic and extrinsic (a coherent state depends on both classical position and momen-
tum) geometry of a closed 3d surface, interpreted as the physical boundary of a 4d spacetime region,
and x and y are points on this (metric) surface. W [s] is the boundary amplitude, that codes the
quantum gravity dynamics, Ψq [s] is a quantum state of the geometry peaked on q, the sum is over a
basis s of states of the 3-geometry, namely s-knot states, and hab (x) is a linearized gravitational field
operator acting on the state Ψq (s). We refer to [2] for a motivation and a discussion of this expression
and to [9] for notation andP an introduction to the formalism and its ideas. N is the normalization
factor defined by 1 = N s W [s]Ψq (s). Eq (1) is well-defined if we have a dynamical model giving
W [s], a “vacuum” state Ψq [s] and a form for the operator hab (x).
We choose the boundary functional W [s] defined by the group field theory GF T /B. We recall the
definition of W [s], referring to [9] and [12] for motivations and details. The theory is defined for a
field φ : (SO(4))4 → R by an action of the form
λ
S[φ] = Skin [φ] + Sint [φ]. (2)
5!
The field φ can be expanded in modes φjα11...j 4
...α4 i (Eq. (9.71) of [9]). Indices jn , n = 1, ..., 4 label simple
SO(4) irreducible representations. Irreducible representations of SO(4) are labelled by a pair of spins
(j+ , j− ), corresponding to the split of so(4) = su(2)×su(2) into its self-dual and antiself-dual rotations;
simple representations are those where j+ = j− ≡ j, and are therefore labelled by a single spin j. The
index αn labels the components of vectors in the representation jn . The index i labels an orthonormal
basis of intertwiners (invariant vectors) on the tensor product of the four representations jn . We
choose a basis in which one of the basis elements is the Barrett-Crane intertwiner iBC (Eq. (9.99) of
[9]). SO(4)-invariant observables of the theory are computed as the expectation values
Z R 2 λR 5
W [s] = Dφ fs (φ) e− φ − 5! φ , (3)

where fs (φ) is the function of the field determined by a spinnetwork s = (Γ, jl , in ). Recall that
a spinnetwork is a graph Γ formed by nodes n connected by links l, colored with representations
jl associated to the links and intertwiners in associated to the nodes. We indicate by lnm a link
connecting the nodes n and m and by jnm ≡ jmn the corresponding color. The spinnetwork function
is defined by
XY j
fs (φ) = φαnm
nm in
; (4)
αnm n

here n runs over the nodes and, for each n, the index m runs over the four nodes that bound the
four links lnm joining at n. Each index αnm ≡ αmn appears exactly twice in the sum, and is thus
contracted. The perturbative expansion of (3) in λ leads to a sum over Feynman diagrams. The
Feynman rules are given by the propagator
j′
X Y
Pαjnn i αn′n i′ = δi,i′ δjn ,j ′π(n) δαn α′π(n) , (5)
π(n) n

where π(n) are the permutations of the four numbers n = 1, 2, 3, 4; and the vertex amplitude
Y  Y 
Vjαnm
nm in
= δin iBC δαnm αmn B(jnm ), (6)
n n<m
 3

where the index n = 1, ..., 5 labels the five legs of the five-valent vertex; while the index m 6= n
labels the four indices on each leg. B(jnm ) is the 10j symbol, or Barrett–Crane vertex amplitude
[12], whose large j expansion is discussed below. The sums over permutations in the propagator
give rises to a number of terms. Each of these can be interpreted as a spinfoam σ, by identifying
closed sequences of contracted deltas as faces. Hence the amplitudeP(3) can be written as a sum of
amplitudes of spinfoams bounded by a given spinnetwork W [s] = ∂σ=s A[σ], an expression that
is naturally interpreted (and can also be derived) as a sum over discretized 4-geometries bounded
by a given discretized 3-geometry, namely as a definition of the Misner-Hawking sum-over-geometries
formulation of quantum gravity, where R the amplitude of a 3-geometry 3g is given by the Feynman
3 iS[g]
integral over 4-geometries W [ g] = ∂g=3g Dg e . The amplitude A[σ] of a spinfoam σ with n
vertices is proportional to λn . Below we consider the first order of the expansion in λ. Since each
face f gives then a contribution dim jf (by contracting the deltas), in the large j limit the dominant
term is the one with the largest number of faces. To first order in λ, we have only one possibility: one
vertex and a five-valent boundary spinnetwork. The dominant contribution for large j is obtained by
choosing the spinfoam σ and the spinnetwork s = ∂σ defined by the dual of a four-simplex and its
boundary. The spinnetwork s has five nodes, connected by 10 links as follows
iq1
BQ j12
j51
  B QQ
i5 q Qq i2
l
CC l j14 j52 B #
s= j13B #  . (7)
j35j24
C  c
c #  j23
C   cB 
j45

q
C B
cBq
i4 j34 i3
P Q
The boundary function fs (φ) determined by this spinnetwork is fs (φ) = αnm n=1,5 φα nm in
jnm . This
is a an observable in the group field theory. The dominant term of its expectation value (3) is
! !
λ Y Y
W [s] = hin |iBC i dim(jnm ) B(jnm ). (8)
5! n n<m

The second ingredient we need is a boundary state Ψq [s]. To identify it, we need the geometrical
interpretation of the boundary spinnetwork s. To this aim, we use the fact that the spinfoam model
can be obtained from a discretization of general relativity on a triangulated spacetime. Introduce 4d
coordinates xµ and represent the gravitational field by means of the one-form tetrad field eI (x) =
eIµ (x)dxµ (related to Einstein’s metric by gµν (x) = eIµ (x)eI µ (x)). Assuming that the triangulation
is fine enough for this field to be approximately constant on a tetrahedron, with constant value eIµ ,
associate the 4d vector eIs = eIµ ∆xµs to the segment s of the triangulation, where ∆xµs is the coordinate
difference between the initial and final extremes of s. To each triangle t of the triangulation, associate
the bivector (the object with two antisymmetric indices)

BtIJ = eIs eJs′ − eJs eIs′ , (9)

where s and s′ are two sides of the triangle. (As far as orientation is kept consistent, the choice of the
sides does not matter). BtIJ is a discretization of the Plebanski two-form B IJ = eI ∧eJ . The quantum
theory is then formally obtained by choosing the quantities BtIJ as basic variables, and identifying
them with SO(4) generators JtIJ associated to each triangle of the triangulation, or, equivalently, to
each face of the corresponding dual spinfoam. The form (9) implies that ǫIJKL BtIJ BtKL ′ = 0 any time
t = t′ or t and t′ share an edge. Accordingly, the pseudo–scalar Casimir C̃ = ǫIJKL JtIJ JtKL = 0 is
required to vanish. This determines the restriction to the simple representations, which are the ones for
which C̃ = 0. The scalar Casimir C = 12 JtIJ JtIJ = 21 BtIJ BtIJ , on the other hand, is easily recognized,
using again (9), as the square of the area At of the triangle t. For simple representations, the value
 4
p
of C is j(j + 1). The quantization of the physical area, with eigenvalues proportional to j(j + 1) is
a key result of LQG [11], reappearing here in the context of the spinfoam models. The LQG result
assures us that we can interpret it as a physical quantization
p and not an artifact of the discretization
and fixes the proportionality constant: Aj = 8πh̄G j(j + 1). The geometrical interpretation of the
intertwiners can be obtained in a similar way and will be discusses elsewhere.
Let now q be the intrinsic and extrinsic geometry of the boundary Σq of a 4d (metric) sphere. We
want to construct the state Ψq [s]. (On the vacuum states in LQG, see [24, 25, 26, 27, 28].) Below we
shall only need the value of Ψq [s] for the spinnetworks defined on a graphs Γ, dual to 3d triangulations
∆. We identify each such ∆ with a fixed triangulation of Σq . We assume here for simplicity that
Ψq [s] is non vanishing only for s whose intertwiners are iBC intertwiners. This choice preserves the
symmetry of the sphere. (There is another simple symmetric choice, used in the first version of this
draft, which is to choose Ψq [s] independent from the intertwiners. This possibility will be explored
(0)
elsewhere.) The areas Anm of the triangles tnm of ∆ determine background values jnm of the spins,
(0) (0)
via A2nm = (8πh̄G)2 jnm (jnm + 1). We want a state Ψq [s] = Ψq (Γ, jnm ) peaked on these background
values. The simplest possibility is to choose a gaussian peaked on these values. However, this leaves
the possibility open of having a phase
( )
1 X
(0) (0) (0)
Ψq [s] = exp − α(nm)(pq) (jnm − jnm )(jpq − jpq ) + i Φnm jnm (10)
2jL n<m

where (n, m) runs on links of s, α(nm)(pq) is a numerical matrix that we will fix later on and the sum
P
n<m,p<q is understood. The phase factor in this state is important. As we know from elementary
quantum mechanics, it determines where the state Pis peaked in the variables conjugate to the spins
jmn . Recall the form of the Regge action SRegge = n<m Φnm (jmn )jnm , where Φnm (jmn ) are dihedral
angles at the triangles, which are function of the areas themselves and that ∂SRegge /∂jnm = Φnm . It
is then easy to see that these dihedral angles are the variables conjugate to the spins. Notice also that
they code the extrinsic geometry of the boundary surface, and in GR the extrinsic curvature is the
variable conjugate to the 3-metric. The value of Ψq [s] on the five–valent spinnetwork (7) can be deter-
mined by triangulating Σq with the 3d triangulation formed by the boundary of a regular four–simplex
√ (0) p
of side L. The area of the triangles is AL = 3L2 /4. Then jnm = jL where 8πh̄G jL (jL + 1) = AL .
(0)
In the large L limit we take jL = 8πh̄GAL . The dihedral angles Φnm = Φ of a regular tetrahedron
are given by cos(Φ) = −1/4. The 1/jL dependence of α (absent in the first version of this paper)
ensures that the relative uncertainties of areas and angles –that is: intrinsic and extrinsic geometry
of the boundary– become small in the large distance limit. Its need has been pointed out by Simone
Speziale in the 3d context [29] and by John Baez in the 4d case, following numerical investigation by
Dan Christensen and Greg Egan, that have shown that in the absence of this dependence the width of
the gaussian is not sufficient for the approximation taken above to hold [30]. To respect the symmetry
of the symmetry of the sphere, the covariance matrix α(nm)(pq) of the gaussian can depend only on
three numbers

α(nm)(pq) = α1 a(nm)(pq) + α2 δ(nm)(pq) + α3 b(nm)(pq) (11)

where δ(nm)(pq) = 1 if (nm) = (pq), a(nm)(pq) = 1 if just two indices are the same, and b(nm)(pq) = 1
if all four indices are different, and in all other cases these quantities vanish. (We will use this
notation, namely α(12)(13) = α1 , α(12)(12) = α2 , α(12)(34) = α3 repeatedly.) The state (10) is thus
completely determined up to the three numbers α1 , α2 , α3 . This amounts to select a vacuum state
which is a coherent state peaked both on the background values of the spins (the intrinsic geometry
of the boundary surface), and on the background values of the angles (the intrinsic geometry of the
boundary surface). See [6] for a similar construction in 3d.
The third ingredient we need is the graviton field operator. This is the fluctuation of the metric
operator over the flat metric hab (~x) = g ab (~x) − δ ab . It is more convenient to consider here the
 5

fluctuation of the densitized metric operator h̃ab (~x) = (det g)g ab (~x) − δ ab = E ai (~x)E bi (~x) − δ ab . In
the linear theory, the propagators of the two agree because of the trace-free condition. To determine
its action, we can equally use the geometrical interpretation discussed above, or LQG. We identify the
point ~x with one of the nodes n of the boundary spinnetwork s. Equivalently, with (the center of) one
of the tetrahedra of the triangulation. Four links emerge from this node. Say these are eI , I = 1, 2, 3, 4.
They are dual to the triangular faces of the corresponding tetrahedron. Let nIa be the oriented normal
to this face, defined as the vector product of two sides. Then E(n)Ii = E ai (~x)nIa can be identified
with the action of the SU (2) generator J i on the link eI . We have then immediately that the diagonal
terms define diagonal operators [9, 11]

E Ii (n)EiI (n)|si = (8πh̄G)2 jI (jI + 1)|si, (12)

where jI is the spin of the link I. Non–diagonal terms, that act on the intertwiners, are discussed
elsewhere.
We have now all the elements needed to compute the expression (1). Inserting (8), (10) and (12)
into (1) we obtain a well–defined expression for the propagator. We choose the points x and y to
be two distinct nodes of the boundary spinnetwork. Equivalently, these can be thought as points
located, say, in the centers of the corresponding dual tetrahedra: in the theory, of course, position is
not determined with precision lower that the individual “atoms of spaces” described by the individual
tetrahedra. We consider the ten by ten matrix W f (L) formed by the “diagonal” components of the
propagator
f (L)(ij)(kl) ≡ W abcd (x, y; q) na nb ñc ñd ,
W (13)

f (L)/|n|4 .
where n = nij and ñ = nkl , and nij is the normal to the triangle tij . We also write W (L) ≡ W
By symmetry

W (L)(ij)(kl) = W1 (L) a(ij)(kl) − W2 (L) δ(ij)(kl) + W3 (L) b(ij)(kl) . (14)

Before computing these quantity in the background independent theory, let us compute it in conven-
tional linearized quantum general relativity. In a flat background metric, two points in the center of ad-
jacent tetrahedra, in a surface with the boundary geometry chosen, are at a distance |xu1 − xd1 |q = L/4.
If the four indices i, j, k, l are all distinct, it is easy to see that n and ñ are orthogonal; then the prop-
agator is easily computed to be [10]

linearized 8πh̄G 1 32h̄G

W(ij)(kl) (L) = i 2
=i (15)
u d 2
4π |x1 − x1 |q πL2
linearized linearized
On the other hand, the components W(ij)(ij) and W(ij)(ik) are vacuum expectation values at fixed
“time”: the first is the flucuation of the area square of a triangle, and the second is the vacuum corre-
lation between the fluctuations of the area squares of two adjacent triangles in the same tetrahedron.
These are also proportional to L−2 . We can therefore write
32h̄G
W linearized(L) = W (16)
πL2
where W is a numerical matrix, with the same symmetry structure as in (14). In particular, W(12)(34) =
W3 = i, while W1 and W2 are real numbers of the order of unity, easily obtained from the linear theory.
We now compute the matrix W (L) in the full theory. Since this is a diagonal term in the propagator,
we can use (12) and (1) reads
X
f (L)(ij)(kl) =
W W [s] ((8πh̄G)2 jij (jij + 1) − |n|2 )((8πh̄G)2 jkl (jkl + 1) − |ñ|2 )Ψq [s]. (17)
s
 6

The terms |n|2 come from the background δ ab and are equal to the square of the area of the face,
namely to (8πh̄GjL )2 , for large j. Inserting (8) and (10) we have
!
λ(8πh̄G)4
X Y
f
W (L)(ij)(kl) = N dim(jnm ) (jij (jij + 1) − j 2 ) (jkl (jkl + 1) − j 2 )
5! L L
jnm n<m
( )
1 X
B(jnm ) exp − α(nm)(pq) (jnm − jL )(jpq − jL ) + iΦ jnm .(18)
2jL n,m

Since the vacuum exponential peaks the sum around jL , which is large, we can discard the +1. We
expand the summand in the fluctuations δjij = (jij − jL ), and keep only the lowest term, assuming
that the gaussian suppress the higher terms. We redefine N absorbing all the factors that cancel
with terms in (the expansion to this order of) the partition function. We assume that the dimj terms
vary slowly over the range where the gaussian is peaked, and can be considered constant. We change
summation variable from the spins to the fluctuation of the spins
X P
f (L)(ij)(kl) = 4N (8πh̄G)4 jL2 − 1 α δj δj +iΦ jnm
W δjij δjkl B(jL + δjnm ) e 2jL (nm)(pq) nm pq nm . (19)
δjnm
P
The rapidly oscillating term exp iΦ nm jnm tends to suppress the sum. To evaluate it, we need the
explicit form of B(jL + δjnm ) in the large j regime. This is of the form [20, 21]
X h π π
i
B(jnm ) = Pτ eiSRegge (τ )+kτ 4 + e−iSRegge (τ )+kτ 4 + D(jnm ) (20)
τ
p
Here τ labels the distinct 4-simplices having areas Anm = jnm (jnm + 1) and Pτ is a slowly varying
factor. Since the sum (19) is peaked around jnm = jL , let us expand the 10j symbol around this
point. To second order around jnm = jL , the Regge action reads
X 1
SRegge (jnm ) = Φ jnm + G(mn)(pq) δjmn δjpq , (21)
nm
2

∂f (j)
where, introducing the “discrete derivative” ∂j ≡ f (j + 1/2) − f (j), we have defined

∂Φmn (jrs )
G(mn)(pq) = . (22)
∂jpq jrs =jL

Thus, around jnm = jL , we have

h P 1 π
P 1 π
i
B(jnm ) = PτR ei(Φ nm jnm + 2 G(nm)(pq) δjnm δjpq + 4 ) + e−i(Φ nm jnm + 2 G(nm)(pq) δjnm δjpq + 4 ) + D(jnm )
(23)
where τR is the regular four simplex (for which kτR = 1), which is only non-degenerate four-simplex P
with these areas [20]. The key observation is now the fact that the rapidly oscillating exp{iΦ nm jnm }
term in the second term of this expression cancels with the rapidly oscillating term in (19). Therefore
the second term of the last expression contributes in a non-negligible
P way to the sum (19). The first
term is suppressed (by the rapidly oscillating factor exp 2iΦ nm jnm ) and it is reasonable to expect
that so is the degenerate term D(jnm ) because this corresponds to 4-simplices with different angles,
and should be dominated by different frequencies. Therefore (19) becomes, keeping only the first term
iπ
X i − 2j1 α(nm)(pq) δjnm δjpq
W (L)(ij)(kl) = 4N jL−2 P (τR )e 4 δjij δjkl e 2 G(nm)(pq) δjnm δjpq e L , (24)
δjnm
 7

f (L) = |n|4 W (L) = (8πh̄GjL )4 W (L). Approximating the sum by a gaussian

where we have also used W
integral gives

−1
W (L) = 4jL−2 jL−1 α − iG (25)

We only need to evaluate the derivatives (22) of the angles with respect to the spins. 4d geometry
gives [31]
 
∂Φnm 1 7 1
= √ a(nm)(pq) − 9δ(nm)(pq) − 4b(nm)(pq) ≡ 2 K(nm)(pq) . (26)
∂Apq 5L2 2 L

The ten by ten matrix K has purely numerical entries. From the relation between areas and spins, we
have
√
G(nm)(pq) = 8πh̄GL−2 K(nm)(pq) = 3(4jL )−1 K(nm)(pq) . (27)

Notice the jL factor that combines with the one in front of α in (25) to give the crucial overall 1/jL
dependence of the propagator. Using the reation between spins and areas, (25) reads

32πh̄G  √ −1
W (L) = √ α + i 3/4 K . (28)
3/4 L2

This is precisely the value (15-16) of the propagator computed from the linearized theory, with the cor-
rect 1/|x − y|2 spacetime√dependence.√The three numerical coefficients of the matrix α are completely
determined by α = 4π 2 / 3 W −1 − i 3/4 K.
The propagator we have computed approximates the perturbative one. This is correct for large
|x − y|, where the approximations used in evaluating the sum (18) hold. When |x − y| approaches the
Planck length, quantum gravitational corrections appear. In this regime, (18) can be easily computed
numerically. Given the discreteness of the representation, the propagator is likely to be cut-off at the
Planck scale. Therefore we obtain the result that, at least to this order, the (component we have
computed of the) propagator is the standard free-theory one at large distances, with corrections at
the Planck scale, wilch appear to make it finite.
Our main results are: (i) the “bad” terms of the Barrett-Crane vertex amplitude are cancelled
when the correct vacuum state, appropriately peaked on the extrinsic geometry, is used; (ii) the
correct propagator emerges to this order; and (iii) –for us the most interesting result– it appears to
be possible to compute n-point functions from background–independent quantum field theories.
——
I am strongly indebted to Leonardo Modesto, Simone Speziale, John Baez and Dan Christensen for
crucial criticisms, inputs and suggestions.

[1] S Carlip, Rept Prog Phys 64 (2001) 885

[2] L Modesto, C Rovelli, Phys Rev Lett 95 (2005) 191301
[3] R Oeckl, Phys Lett B575 (2003) 318-324; Class Q Grav 20 (2003) 5371-5380
[4] F Conrady, C Rovelli, Int J Mod Phys A 19, (2004) 4037 L Doplicher, Phys Rev D70 (2004) 064037
[5] F Conrady, L Doplicher, R Oeckl, C Rovelli, M Testa, Phys Rev D69 (2004) 064019
[6] D Colosi, L Doplicher, W Fairbairn, L Modesto, K Noui, C Rovelli, Class Q Grav 22 (2005) 2971
[7] L Modesto, C Rovelli, in preparation
[8] D Colosi, C Rovelli, gr-qc/0409054
[9] C Rovelli, Quantum Gravity, (Cambridge University Press, Cambridge, 2004)
[10] F Mattei, C Rovelli, S Speziale, M Testa, gr-qc/0508007
[11] C Rovelli, L Smolin, PRL 61 (1988) 1155; Nucl Phys B331 (1990) 80; B442 (1995) 593; B456 (1995) 734
 8

[12] A Perez, Class Q Grav 20 (2003) R43. JW Barrett, L Crane, J Math Phys 39 (1998) 3296
[13] R DePietri, L Freidel, K Krasnov, C Rovelli, Nucl Phys B574 (2000) 785
[14] M Bojowald, A Perez, gr-qc/0303026
[15] F Conrady, gr-qc/0504059
[16] A Perez, C Rovelli, Nucl Phys B599 (2001) 255-282
[17] D Oriti, RM Williams, Phys Rev D63 (2001) 024022
[18] A Perez, Nucl Phys B599 (2001) 427-434. L Crane, A Perez, C Rovelli, PRL 87 (2001) 181301
[19] JW Barrett, RM Williams, Adv Theor Math Phys 3 (1999) 209-215
[20] JC Baez, JD Christensen, G Egan, Class Q Grav 19 (2002) 6489
[21] L Freidel, D Louapre, Class Q Grav 20 (2003) 1267. J Barrett, C Steele, Class Q Grav 20 (2003) 1341
[22] ER Livine, D Oriti, Nucl Phys B663 (2003) 231
[23] D Colosi, C Rovelli, Phys Rev D68 (2003) 104008
[24] A Ashtekar, L Bombelli, A Corichi, PRD 72 (2005) 025008
[25] L Smolin, AIP Conf Proc 646 (2003) 59. F Conrady, Class Q Grav 22 (2005) 3261
[26] T Thiemann, gr-qc/0206037. T Thiemann, O Winkler, Class Q Grav 18 (2001) 4629
[27] A Corichi, JM Reyes, Int J Mod Phys. D10 (2001) 325. A Mikovic, Class Q Grav 21 (2004) 390
[28] A Ashtekar, C Rovelli, L Smolin, Phys Rev Lett 69 (1992) 237, C Rovelli, Helv Phys Acta 69 (1996) 582
[29] S Speziale, “Towards the graviton from spinfoams: the 3d toy model”, gr-qc/0512102
[30] J Baez, D Christensen, G Egan, private communication.
[31] C Rovelli, S Speziale, “Variation of the dihedral-angle with the area in a regular 4d simplex”, to appear