Fuzzy circle and new fuzzy sphere through confining

potentials and energy cutoffs

Gaetano Fiore, Francesco Pisacane
arXiv:1709.04807v1 [math-ph] 14 Sep 2017

Dip. di Matematica e applicazioni, Universita di Napoli Federico II,

& INFN, Sezione di Napoli,
Complesso Universitario M. S. Angelo, Via Cintia, 80126 Napoli, Italy

Abstract

Guided by ordinary quantum mechanics we introduce new fuzzy spheres Sd of

dimensions d = 1, 2: we consider an ordinary quantum particle in D = d+1 dimensions
subject to a rotation invariant potential well V (r) with a very sharp minimum on a
sphere of unit radius. Imposing a sufficiently low energy cutoff to freeze the radial
excitations makes only a finite-dimensional Hilbert subspace accessible and on it the
coordinates noncommutative a la Snyder; in fact, on it they generate the whole algebra
of observables. The construction is equivariant not only under rotations - as Madores
fuzzy sphere -, but under the full orthogonal group O(D). Making the cutoff and the
depth of the well dependent on (and diverging with) a natural number , and keeping
the leading terms in 1/, we obtain a sequence Sd of fuzzy spheres converging to the
sphere S d in the limit (whereby we recover ordinary quantum mechanics on
S d ). These models may be useful in condensed matter problems where particles are
confined on a sphere by an (at least approximately) rotation-invariant potential, beside
being suggestive of analogous mechanisms in quantum field theory or quantum gravity.

1 Introduction
In 1947 Snyder proposed [1] the first example of noncommutative spacetime hoping that
nontrivial (but Poincare equivariant) commutation relations among the coordinates, acting
as a fundamental regularization procedure, could cure ultraviolet (UV) divergencies in quan-
tum field theory (QFT)1 . He dubbed as distasteful arbitrary the UV regularization based on
momentum (or equivalently energy) cutoff, which had just been proposed in the literature
at the time, presumably as it broke manifest Lorentz equivariance and looked ad hoc. Iron-
ically, shortly afterwards this and other more sophisticated regularization procedures found
1
The idea had originated in the 30s from Heisenberg, who proposed it in a letter to Peierls [2]; the idea
propagated via Pauli to Oppenheimer, who asked his student Snyder to develop it.

1
widespread application within the renormalization method; as known, the latter has proved
to be extremely successful in extracting physically correct predictions from quantum electro-
dynamics, chromodynamics, and more generally the Standard Model of elementary particle
physics. The proposal of Snyder was thus forgotten for decades. On the other hand, it is
believed that any consistent quantum
p theory of gravitation will set fundamental bounds of
the order of Planck length lp = ~G/c3 1033 cm on the accuracy x of localization mea-
surements. The arguments, which in qualitative form go back at least to [3, 4, 5], are based
on a cutoff on the concentration of energy2 ; they were made more precise and quantitative by
Doplicher, Fredenhagen, Roberts [6], who also proposed that such a bound could follow from
appropriate noncommuting coordinates (for a review of more recent developments see e.g.
[7]). More generally, Connes Noncommutative Geometry framework [8] allows not only to
replace the commutative algebra A of functions on a manifold M by a noncommutative one,
but also to develop on it the whole machinery of differential geometry [8, 9]. Often one deals
with a family of noncommutative deformations A of A that become commutative when
the familys ruling parameter(s) go to some limit, exactly as the algebra of observables in
ordinary quantum mechanics becomes the algebra of functions on ordinary phase space as
~ 0.
Fuzzy spaces are particular examples parametrized by a positive integer n, so that the
algebra An is a finite-dimensional matrix algebra with dimension which increases and diverges
with n while An A (in a suitable sense). Since their introduction they have raised big
interest in the high energy physics community as a non-perturbative technique in QFT (or
string theory) based on an finite-discretization of space(time) alternative to the lattice one:
the advantage is that the algebras An can carry representations of Lie groups (not only of
discrete ones). The first and seminal fuzzy space is the Fuzzy Sphere (FS) of Madore and
Hoppe [10, 11]3 ; the first applications to QFT models are in [12, 13]. The FS is a sequence of
SO(3)-equivariant, finite noncommutative -algebras An isomorphic to Mn (the algebra of
n n matrices); each matrix represents the expansion in spherical harmonics of an element
of C(S 2 ) truncated at level n. An is generated by hermitean noncommutative coordinates
xi fulfilling
4i ijk k
[xi , xj ] = 2 x , r2 := xi xi = 1, n N \ {1} (1)
n 1
(here and below sum over repeated indices is understood). The Hilbert space is chosen as
H ' Cn so that it carry an irreducible representation of Uso(3), and the square distance r2
from the origin - which is central - be identically equal to 1. We note that however (1) are
equivariant only under SO(3), not O(3); in particular not under parity xi 7 xi . Fuzzy
spaces can be used also in extra dimensions to account for internal (e.g. gauge) degrees of
freedom, see e.g. [14].
2
In fact, by Heisenberg uncertainty relations to reduce the uncertainty x of the coordinate x of an
event one must increase the uncertainty px of the conjugated momentum component by use of high energy
probes; but by general relativity the associated concentration of energy in a small region would produce a
trapping surface (event horizon of a black hole) if it were too large; hence the size of this region, and x
itself, cannot be lower than the associated Schwarzschild radius, lp .
3
In [11] it was used in connection with the quantization of the membrane; its noncommutative geometry
was first described by Madore in [10].

2
 As the arguments leading to x lp suggest, imposing an energy cutoff E on an existing
theory can be physically justified by two reasons, at least. It may be a necessity when
we believe that E represents the threshold for the onset of new physics not accountable
by that theory. Or it may serve to yield an effective description of the system when we,
as well as the interactions with the environment, are not able to bring its state at higher
energies; this leads also to a lower bound in the accuracy with which our apparatus can
measure some observables (position, momentum, ...) of the system, which corresponds to the
maximum energy transferable to the system by the apparatus in the measurement process, or
by the envorinment during the interaction time. (Of course, the two reasons may co-exist.)
Mathematically, the cutoff is imposed by a projection on the Hilbert subspace characterized
by energies E E. If the Hamiltonian is invariant under some symmetry group, then the
projection is invariant as well, and the projected theory will inherit that symmetry.
That imposing such a cutoff can modify a quantum mechanical model by converting its
commuting coordinates into non-commuting ones is simply illustrated by the well-known
Landau model (see e.g. [15, 16]), which describes a charged quantum particle in 2D inter-
acting only with an uniform magnetic field (in the orthogonal direction) B. The energy
levels are En = ~|eB|mc
n, if we fix the additive constant so that the lowest level is E0 = 0;
~|eB|
choosing E mc (this may be physically justified e.g. by a very strong B), then the
Hilbert space of states is projected to the subspace H0 of zero energy, and ec Bx, y become
canonically conjugates, i.e. have a non-zero (but constant) commutator. The dimension of
H0 is approximately proportional to the area of the surface, hence is finite (resp. infinite) if
the area is.
Inspired by the projection mechanism in the Landau model, here we consider a quantum
particle in dimension D = 2 or D = 3 with a Hamiltonian consisting of the standard kinetic
term and a rotation invariant potential energy V (r) with a very deep minimum (a well) re-
spectively on a circle or on a sphere of unit radius; k V 00 (1)/4 > 0 plays the role of confining
parameter. Imposing an energy cutoff E makes only a finite-dimensional Hilbert subspace
HE accessible and the projected coordinates noncommutative on HE . We choose E 2k
so that HE does not contain excited radial modes, and on it the Hamiltonian reduces to the
square angular momentum; this can be considered as a quantum version of the constraint

r = 1. It turns out that the coordinates generate the whole algebra AE := End HE of
observables on HE . Their commutators are of Snyder type, i.e. proportional to the an-
gular momentum components Lij (apart from a small correction - depending only on the
square angular momentum - on the highest energy states), rather than some function of the
coordinates. Moreover, (AE , HE ) is equivariant under the full group O(D) of orthogonal
2 D
transformations, because both the starting quantum mechanical model  on L (R  ) and the
cut-off procedure are. Actually, we prove the realization AE = E Uso(D + 1) , with E
a suitable irreducible unitary representation of Uso(D + 1) on HE ; as a consequence, HE
carries a reducible representation of the subalgebra Uso(D) generated by the angular mo-
mentum components Lij , more precisely the direct sum of all irreducible representations
fulfilling E E; in the E limit this becomes the decomposition of the Hilbert space
L2 (S d ), d = D 1. This welcomed property is not shared by the FS [10, 11]. Similarly, the
decomposition of the subspace CE AE of completely symmetrized polynomials in the non-

3
commutative coordinates into irreducible Uso(D)-components becomes the decomposition of
the commutative algebra C(S d ) [which acts on L2 (S d ) and has the same decomposition4 ].
On HE the square distance R2 from the origin is not identically 1, but a function of the
square angular momentum such that nevertheless its spectrum is very close to 1 and collapses
to 1 in the k limit of an infinitely narrow and deep well; the latter limit is automatic
2
as we have to set k E for consistency. Thus the confining parameter k, or equivalently
the energy cutoff E, or a suitable natural number which we shall adopt to discretize both,
will also parametrize the noncommutativity of the coordinates. Finally, in a suitable sense
H L2 (S d ), C C(S d ) and A goes to the whole algebra of observables on L2 (S d ), as
.
We think our models are interesting not only as new toy-models of fuzzy geometries in
QFT and quantum gravity, but also in view of potential applications to quantum models in
condensed matter physics with an effective one- or two-dimensional configuration space in the
form of a circle, a cylinder or a sphere5 , because they respect parity, and the restriction to the
circle, cylinder or sphere is an effective one obtained a posteriori from the exact dynamics
in higher dimension. Moreover, our procedure can be generalized in a straightforward manner
to D > 3, as well as to other confining potentials; the dimension of the accessible Hilbert
space HE will be approximately B/hD , where h, B are Planck constant and the volume of
the classically allowed region in phase space (i.e. the one characterized by E E).
Our approach looks applicable to higher dimensions in a straightforward way. Fuzzy
spheres based on some Snyder-type commutation relations have already been proposed for
d = 4 in [17] (see e.g. also [18, 19]) and for all d 3 in [20]. In section 5 we mention how
we expect the results based on our approach would be related to the latter.
The plan of the paper is as follows. In section 2 we introduce the framework valid for
any D. In sections 3, 4 we treat the cases D = 2, 3 leading to S1 , S2 respectively. Section
5 contains final remarks, outlook and conclusions. In the Appendix (section 6) we have
concentrated lengthy computations and proofs.

2 General setting
As said, we consider a quantum particle in RD configuration space with Hamiltonian
1
H = + V (r), (2)
2
where r2 := xi xi , := i i (sum over repeated indices understood), i := /xi , i = 1, ..., D;
the cartesian coordinates xi , the momentum components ii , H itself are normalized so
as to be dimensionless. xi , ii generate the Heisenberg algebra O of observables. The
4
In fact, spherical harmonics make up a basis for both L2 (S d ) and C(S d ).
5
For instance, circular quantum waveguides, graphene nanotubes and fullerene. These are very thin wires
or layers of matter where electrons are confined by potential energies with very deep minima there and steep
gradients in the normal direction(s). The Hamiltonian on a cylinder can be written as the sum of the one
on the transverse section circle and of the kinetic term ~2 (/z)2 /2m in the direction of the axis.

4
canonical commutation relations

[xi , xj ] = 0, [ii , ij ] = 0, [xi , ij ] = iji (3)

as well as the Hamiltonian are invariant under all orthogonal transformations

xi 7 x0i = Qij xj , Q1 = QT (4)

(including parity Q = I). This implies [H, Lij ] = 0, where Lij := ixj i ixi j are the
angular momentum components. We shall assume that V (r) has a very sharp minimum at
r = 1 with very large k = V 00 (1)/4 > 0, and fix V0 := V (1) so that the ground state has
zero energy, i.e. E0 = 0. We choose an energy cutoff E fulfilling first of all the condition

V (r) ' V0 + 2k(r 1)2 if r fulfills V (r) E (5)

so that we can neglect terms of order higher than two in the Taylor expansion of V (r) around
1 and approximate the potential as a harmonic one in the classical region vE compatible with
the energy cutoff V (r) E. We are interested in finding the eigenfunctions of H

L2 RD


H = E, (6)

with eigenvalues E E and restricting quantum mechanics to the finite-dimensional Hilbert

subspace HE spanned by them. This means that we shall replace every observable A by A :=
PE APE , where PE is the projection on HE , and give to A the same physical interpretation.
In particular: xi will be intepreted as the observable associated to the measurement of the
i-th coordinate of the particle; H = H will still appear as the Hamiltonian in the original

Schrodinger equation. We shall also replace any Schrodinger equation i t = He , with
some extended Hamiltonian He = H + H 0 (containing a small extra term H 0 representing

some additional interaction), with the finite-dimensional one i t = He within HE .
q
By (5), vE RD is approximately the shell |r 1| EV 2k
0
; in the limit in which both
k, E diverge, but the right-hand side goes to zero, vE reduces to the unit sphere S D1 . We
expect that in this limit the dimension of HE diverges, and we recover standard quantum
mechanics on the sphere S D1 . As we shall see, this is the case.
PE commutes not only with H, but also with the Lij := ixj i ixi j , which are vector
fields tangent to every sphere r =const. The D derivatives i make up a globally defined
basis for the linear space of smooth vector fields. The set B = {r , Lij | i < j} (r := /r)
is an alternative complete set that is singular for r = 0, but globally defined elsewhere; for
D = 2 it is a basis, for D > 2 it is redundant, because of the relations

i1 i2 i3 ....iD xi1 Li2 i3 = 0. (7)

This redundancy is unavoidable. In D = 3 there are no two independent globally defined vec-
tor fields that are tangent to the sphere (S 2 is not parallelizable); for instance, the derivatives
, with respect to the polar angles are singular at the north and south poles. One needs
all three angular momentum components, which however are constrained by ijk xi Ljk = 0.

5
 Of course, the eigenfunctions of H can be more easily determined in terms of polar
coordinates r, , ..., recalling that the Laplacian in D dimensions decomposes as follows
1 1
= r2 + (D 1) r 2 L2 , (8)
r r
where r := /r and L2 = Lij Lij is the square angular momentum (in normalized units),
i.e. the Laplacian on the sphere S D1 . We know from the D-dimensional theory of angular
momentum that the eigenvalues of L2 are j (j + D 2); then replacing the Ansatz =
f(r)Y (, ...) (Y are eigenfunctions of L2 and of the elements of a Cartan subalgebra of
so(D); r, , ... are polar coordinates) transforms the PDE H = E into this auxiliary
ODE in the unknown f(r)
 
1 1
r (D 1) r + 2 j (j + D 2) + V (r) f(r) = E f(r).
2
(9)
r r

If V (r) keeps bounded or grows at most as /r2 (with p some 0) as r 0, then in the

same limit f (r) vanishes as f (r) = O (r ), with = + j(j + D 2). In fact, by Fuchs
theorem every solution of (9) is a combination of the two independent ones with r 0
asymptotic behaviour r , r ; but the coefficient of the second must vanish in order that
r+
L2 (RD ). Hence f (0) = 0. On the other hand, L2 (RD ) implies also f(r) 0.
In fact, condition (5) implies that f, become negligibly small outside the thin shell region
V (r) E (around r = 1), and that inside the latter we can approximate (9) by the eigenvalue
equation of a 1dimensional harmonic oscillator.

3 D=2: O(2)-equivariant fuzzy circle

We fixthe notation as follows: x x1 = r cos , y x2 = r sin , x := (x iy)/ 2 =
rei / 2; we abbreviate u := ei (whence u = ei ), /x , /; the
angular momentum L L12 can be expressed in the form L = i = x+ + x , and

[L, x ] = x , [L, ] = ,

i.e. the generators x , of the Heisenberg algebra O (of observables) are eigenvectors under
the adjoint action of L with eigenvalues 1. We look for of the form em (r, ) = fe(r)eim ,
with eigenvalues m Z and we can use the L-eigenvalue as a Z-grading for both H and O,
in a compatible way:
0 0 0 0
M M
H= Hm , O= Om , Om Om = Om+m , Om Hm = Hm+m
mZ mZ

(the last relation must be understood modulo domain restrictions); clearly L, r, h(r), r ,
O0 [for any funtion h(r)]. Equation (9) becomes

m2 e
 
00 1 e0
f (r) + f (r) + E V (r) 2 f (r) = 0.
e
r r

6
We change the radial variable r 7 := ln r and set f () := fe(r) = fe(e ), whereby the
previous equation is transformed
q into f 00 ()+{e2 [E V (e )]m2 }f () = 0. By condition
(5), in the region |r1| EV
2k
0
we can neglect the terms of order higher than two in the
Taylor expansions e = 1 + 2 + 22 + ..., V (e ) = V0 + 2k2 + ... and thus approximate
2

the above equation by

Hf () = em f (), H := 2 + km ( em )2 ,
E0 E 02 (10)
0 0
where km = 2(kE ), E = E V0 , em = , em = +E 0 m2 ,
km km
which has the form of the eigenvalue equation for a harmonic oscillator in 1 dimension with

equilibrium position at = em . In order that be square-integrable it must be f () 0,

f () 0, what selects as eigenfunctions of the auxiliary operator H
2
" #
( em ) km h p
4
i
fn,m () = exp Hn ( em ) km ,
2

(here Hnis the Hermite polynomial of order n), and as corresponding eigenvalues em,n =
(2n + 1) km . By (10) this implies that E 0 must fulfill the equation

E 02 0 2
p
+E m = (2n+1) 2(kE 0 ) (11)
2(kE 0 )
2
Squaring both sides of (11) and multiplying them by km one obtains a fourth degree equation
0
which determines E , and therefore E, in terms of V0 , m, n. As said, we fix V0 requiring that
the lowest energy level, which corresponds to n = m = 0, be E0 = 0. This implies that V0
must fulfill the equation
r r !3 3
V02

p 1 1 2 V0 2
V0 = 2(k + V0 ) V0 V0 = 1 + .
2(k + V0 ) 2k 2k k

P q n
1
Looking for the solution in the form V0 = n=1 vn 2k
we can determine the coefficients
q
1
vn , and therefore V0 , solving the latter equation order by order in 2k
. The solution is

V0 = 2k + 2 72 12k + O(1/k) and Figure 1 shows the appearance of the resulting
potential. Replacing this result
 in (11) one finds that at leading order E is given by En,m =

2n 2k + m 8n(n + 1) + O 1k . For sufficiently large k, E E can be fulfilled only with
2

n = 0, what we shall assume henceforth. Physically, this means that radial oscillations are
frozen, so that all corresponding classical trajectories are circles. The energies E below E
will therefore depend only on m, and will be denoted as Em . Consequently, km , em will be
determined by (10). Then at leading orders in 1/ k (11) yields as eigenvalues of L, H and

7
corresponding eigenfunctions

3m2
 
2 1
L = m, H = Em = m + O , (12)
2k k
2
" #
( m ) km
m (r, ) = Nm eim exp
e
, (13)
2

2 m2 m2
   
km 2 1 1 1
=1 + +O 3 , em = + +O 3 (14)
2k 2k k k2 2k 2k k2
we fix the normalization factor Nm so that Nm > 0 and all m have unit norm. The condition
E E is fulfilled if we project
h i the theory onto the Hilbert subspace H HE spanned by
the m with |m| := E (here [a] stands for the integer part of a > 0). For consistency
we must choose
2 < 2k (15)
so that all Em are smaller than the energy levels corresponding to n > 0, as we can see from
Figure 2; this is also sufficient to guarantee that km  1 for all |m| (by the way, km > 0
is a necessary condition for H to be the Hamiltonian of a harmonic oscillator).

Figure 1: Three-dimensional plot of V (r)

Figure 2: Two-dimensional plot of V (r) in-
cluding the energy-cutoff

8
 Let A O, then A O and O, HE H inherit the grading from O, H; for any Ah Oh

m+h hm+h , Ah m i

(
X if |m|, |m + h| ,
h h
A m = m0 hm0 , A m i =
m0 = 0 otherwise.

In formula (58) in the appendix we compute the matrix element hm+h , Ah m i for A of the
form A = f ()eih = f ()uh [if A contains also derivatives the result can be expressed
as a combination of matrix elements of the same type because of (23)]. In particular, up to
terms O(1/k 3/2 )
( q
m+1
+
a
m+1 2 1+ m(m+1)
k
if m 1,
um = , x m = (16)
0
0 otherwise,
q
where a := 1+ 94 12k + 64k
137
. Clearly 2 x+ = u a 1+ L(L+1)
k
. As u , x are the adjoints of
u, x+ , so are u , x respectively the adjoints of u, x+ . We can get rid of the m-independent
factor a reabsorbing it in the redefinitions

x
= .
a
Therefore we find
q
1 1+ m(m1) if m 1
2 k m1
Lm = mm , m = (17)
0 otherwise,

[the second relation is up to terms O(1/k 3/2 )]. Eq. (17) implies at leading order
m
m if |m| 1,
k



[ + , ]m = 1 (1) (18)
1+ m if m = ,


2 k

0 otherwise.

m2
 

1+ m if |m| 1,
k



R2 m

= 1 ( 1) (19)
1+ m if m = ,
2 k




0 otherwise;

in (19) we have introduced the square distance from the origin R2 := + + + , in analogy
with the classical definition. We see that R2 is not identically equal to 1 on all of H (as
in the standard quantization on the unit circle), but the m are eigenvectors of R2 with
eigenvalues depending only on m2 and growing with m2 (with the exception of the states

9
 , near the cutoff, which however play no role at lower energies); physically this is to be
expected because higher square angular momentum m2 is equivalent to a higher centrifugal
force, which classically yields a slightly more external circular trajectory. Moreover, all these
eigenvalues are close to 1 and go to 1 as k , while the eigenfunctions factorize as
m (r, ) (r1)eim .
If we now adopt (17) as exact definitions of + , , L, then (18-19) are exact as well, and
we easily find
Proposition 3.1. The + , , L defined by (17) generate the -algebra A := End(H ) '

M2+1 (C) of observables on H ; they fulfill ( + )2+1 = ( )2+1 = 0,
Q
(LmI) = 0,
m=

L Pe Pe (20)
+ = , L, = , +, = +
   
L = L, ,
k 2
where Pem stands for the projector on the 1-dimensional subspace spanned by m (clearly
P
P = m= Pm ) and L := P LP is the projection on H of the angular momentum
e
operator. All Pem can be expressed as polynomials in L using the spectral decomposition of
2
L. Moreover, the square distance from the origin can be expressed as the function of L
2
2 + + L Pe + Pe (+1)
R := + = 1 + , := 1+ . (21)
k 2 k

Of course, relations (20)4 , (21) only hold at leading order in 1/ k if also (17) do.
To obtain a fuzzy space depending only on one integer we can choose k as a function
of fulfilling (15); the commutative limit will be simply (which implies k ).
One possible choice is
 e e
2 2 + L 1 P P
k = (+1) ; then (20)4 becomes [ , ] = 2 2
+ 1+ . (22)
(+1) (+1) 2
Summarizing our results so far, we see that the combined effect of the confining potential
and of the energy cutoff is a non-vanishing commutator between the coordinates. Note that
relations (20), (22) are invariant not only under rotations (this includes parity a 7 a ,
L 7 L in D = 2), but also under orthogonal transformations with determinant 1, e.g.
1 7 1 , 2 7 2 , L 7 L, i.e. under the whole group O(2), as in the ordinary theory
without cutoff. This had to be expected, because both the commutation relations in the
original infinite-dimensional model and the Hamiltonian H (hence also the projectors PE )
are O(2)-invariant. Apart from the sign and from the last term containing the projections,
which plays no role far from the cutoff , relations (20), (22) are of the Snyders Lie algebra
type [1], because the commutator of the coordinates is a generator of rotations. +, (or
equivalently x+, x ) generate the whole -algebra A (also L can be expressed as a non-
ordered polynomial in +, ). To compute it is convenient to use polar coordinates. We
find
1 p
= ( i ) , m = ( em ) km m n=1,m (23)
2x

10
and = 0, because of P n=1,m = 0. Similarly, r , x1 go to zero in norm as k . As
 
consequences, neither nor the commutator + , vanish as k , as expected.
Only the vector field L tangent to the circle survives with the correct classical limit, as
desired. For completeness, the actions of + , are reported in (68-72).

3.1 Realization of the algebra of observables through U so(3)

For every n N the -algebra Mn (C) of endormophisms of Cn can be realized as the
n = (2+1)-dimensional unitary representation of so(3) ' su(2). In fact, the operators
on H , and in particular u, , are naturally realized in [Uso(3)], identifying m as the
vectors |, mi of the canonical basis, in standard ket notation. We denote as E + , E , E 0 the
Cartan-Weyl basis of so(3),
[E + , E ] = E 0 , [E 0 , E ] = E , E = E , E 0 = E 0 , (24)
as (with an abuse of notation also) its real structure, and as C the Casimir,
C = E a E a = 2E + E +E 0 (E 0 1) = 2E E + +E 0 (E 0 +1). (25)
The representation is characterized by the Casimir eigenvalue (C) = ( + 1). We
identify L = (E 0 ), which will still determine the grading. To simplify the notation in the
sequel we drop . For every Ah Oh one can determine a function of one variable fA (s)
such that
(E + )h if h > 0,
h 0 h h
A = fA (E )E , where E = 1 if h = 0,
h
(E ) if h < 0,

by requiring that hm+h , Ah m i = hm+h , fA (E 0 )E h m i and using (58). In particular it is

an easy exercise to check that from (16) and the adjoint relations it follows
L = E 0, = f (E 0 )E (26)
and u = fu (E 0 )E + , where at leading order6
s
1 1 1+s(s1)/k
fu (s) = p , f+ (s) = = f (s1) = f (s).
( + 1)s(s1) 2 ( + 1)s(s1)

Therefore (26) fulfill (20). The inverse of the change of generators (26) is clearly

1
E 0 = L, E = f L

.
The eigenvalue condition C = ( + 1) can be put more explicitly in either form

2
2E E + = ( + 1)E 0 (E 0 +1) = 2 + f+ L+1

. (27)
Summarizing, we have almost completely shown
6
Note that fu (E 0 ), f+ (E 0 ) are singular on , while fx (E 0 ) is singular on , but since their action
follows that of E + or E they can never act on such vectors, and the products at the right-hand side of (26)
are well-defined on all of H .

11
Proposition 3.2. Formulas (26) provide a O(2)-equivariant -algebra isomorphism between
the algebra A := End(H ) of observables (endomorphisms) on H and that on the C =
( + 1) irreducible representation of U so(3):

A := End(H ) ' MN (C) ' [U so(3)], N := 2+1. (28)

(The O(2)-equivariance of this realization is shown below.) Note also that every function
of E 0 = L, including f (E 0 ), can be expressed in polynomial form by spectral decomposition.
As consequences, the generators E + , E of Uso(3) [which are characterized by relations
(24), where E 0 has to be understood as an abbreviation for the commutator of E + , E ]
further constrained by (27)1 , or alternatively + , fulfilling (20) and (27)2 , generate all
the algebra A ; while ordered polynomials in E + , E , E 0 , or alternatively in + , , L, span
A . Therefore the above results for the action of the operators + , , ... on H can be
recovered determining the unique unitary representation of the -algebra generated by + ,
fulfilling relations (20), (27)2 , or more simply setting (26) and using our knowledge on the
representation of so(3).
As known, the group of -automorphisms of MN (C) ' A is inner and isomorphic to
SU (N ), i.e.
a 7 g a g 1 , a A , , (29)
with g an unitary N N matrix with unit determinant. A special role is played by the
subgroup SO(3) acting through the representation , namely g = [ei ], where so(3),
i.e. is a combination with real coefficients of E 0 , E ++E , i(E +E ). In particular, choosing
= E 0 [i.e. in the adopted so(2) Cartan subalgebra of so(3)] the automorphism amounts to
a rotation in the x1 x2 plane by an angle , i.e. E 0 7 E 0 and E 7 ei E , or equivalently
L 7 L and
 01
0 i x = x1 cos + x2 sin ,
x 7 x = e x
x02 = x1 sin + x2 cos ;

this a SO(2) transformation in the x1 x2 plane. Setting = (E + + E )/ 2 we obtain a
O(2) transformationwith determinant 1 in such a plane; this amounts to a rotation about
E 1 := (E + + E )/ 2 by an angle , i.e. to E 0 7 E 0 , E 7 E . As the functions f
fulfill f (s) = f (1 + s) = f (s), this is equivalent to x1 7 x1 , x2 7 x2 , L 7 L.
All other O(2) transformations with determinant 1 in the x1 x2 plane can be obtained by
composition with a SO(2) transformation.

4 O(3)-equivariant fuzzy sphere

When D = 3 one can associate a pseudovector Li = 12 ijk Ljk to the antisymmetric matrix
Lij of the angular momentum components. For all vectors v depending on x, i we shall

12
use either the components v i or the ones v a , a {, 0, +}, defined by
+ 1
i 0
1
v v
2 2
v = 1 i
2 2
0 v2 . (30)
0
v 0 0 1 v3
| {z }
U

(U is a unitary matrix) which fulfill

[La , v a ] = 0, [L0 , v ] = v , [L , v ] = v 0 , [L , v 0 ] = v . (31)

1 2 i a
In particular, x0 z, x = x ix
2
xiy

2
= r sine2 . We set ta := xr . Correspondingly,
the metric matrix ij = ij becomes eab = (U U T )ab = ab . Moreover, (8) becomes
= 1r r2 r r12 L2 . We look for of the form lm (r, , ) = f (r)
r
Ylm (, ), where Ylm (, )
are the spherical harmonics:

L2 Ylm (, ) = l(l + 1)Ylm (, ) , L3 Ylm (, ) = mYlm (, ) .

h i
Equation (9) becomes f 00 (r) + V (r) + l(l+1) r2
E f (r) = 0. By condition (5), in the
q
EV0
region |r 1| 2k
we can neglect the terms of order higher than two in the Taylor
1
expansion of r2 = 1 2(r 1) + 3(r 1)2 + ..., V (r) = V (1) + 2k(r 1)2 + ... and
thus approximate this equation by the eigenvalue equation for a 1dimensional harmonic
oscillator, that is

f 00 (r) + kl (r rel )2 f (r) = Ef

e (r) (32)

with kl := 2k + 3l(l + 1) ,
2k + 4l(l + 1) l(l + 1) 3l2 (l + 1)2 3


rel := =1+ + O k ,
2k + 3l(l + 1) 2k 4k 2

e := E V (1) l(l + 1) + l2 (l + 1)2

E .
2k + 3l(l + 1)
The square-integrable solutions of (32) have the form
2
(re
rl ) kl
 p 
fn,l (r) = Nl e 2 Hn (r rel ) 4 kl , n = 0, 1, ....,

and Ee = (2n + 1) kl . Choosing V (1) such that in the lowest level (characterized by n =

l = 0) E = 0, one has V (1) = k0 = 2k and, consequently,

l2 (l + 1)2
 
p 1
E En,l := (2n+1) 2k + 3l(l + 1)+V (1)+l(l+1) = 2n 2k+l(l+1)+O
2k + 3l(l + 1) k
To avoid diverging eigenvalues in the limit k we need to exclude n > 0. Therefore we
impose an energy cut-off E = (+1), that is we project the theory to the finite-dimensional

13
subspace HE H H spanned by the lm := 0,l,m with |m| l and l . We denote as
P the projection over H and abbreviate El = E0,l . For consistency one must choose

( + 1) 2 2k. (33)

Then the eigenfunction of L2 , L3 and H with eigenvalues respectively l(l + 1), m, El is

Nl (rerl )2 kl
lm (r, , ) = e 2 Ylm (, ),
r
p l2 (l + 1)2
El = 2k + 3l(l + 1) 2k + l(l + 1) ' l(l + 1),
2k + 3l(l + 1)
at the leading order in k. The action of the generators La (a {1, 0, +1}) is therefore
p
(l m)((l m + 1) m1
m m
L0 l = m l , m
L l = l =: l,m lm1 . (34)
2
In the appendix we compute the normalization factor Nl ; moreover, we show that the action
of xa on the vectors lm reads
a,m m+a
cl A l1 + cl+1 Bla,m l+1
m+a
if l < ,
l


xa lm = cl Aa,m
l
m+a
1 if l = , (35)


0 otherwise,

where Aa,m
l , Bla,m are the coefficients involved in the formula

ta Ylm = Aa,m
l
m+a
Yl1 + Bla,m Yl+1
m+a
(36)

and are explicitly reported in (73), while, up to terms O(1/k 3/2 ),

r
l2
cl = 1 + 1 l , c0 = c+1 = 0. (37)
k
a i
We now adopt (34-37) as exact definitions of xa , L . The xi , L can be obtained by the
inverse transformation of (30). In the appendix we prove
Proposition 4.1. The xi , Li defined by (34-37), (30) generate the -algebra A := End(H ) '
M(+1)2 (C) of observables on H . They fulfill
h i l
Y 2 Y

L l(l + 1)I = 0, L3 mI Pel = 0, (38)
l=0 m=l

xi = xi , [Li , xj ] = iijh xh , Li , Lj = iijh Lh ,
 
Li = Li , (39)
 
I
xi Li = 0, i j
[x , x ] = i ijh
+ C P Lh
e i, j, h {1, 2, 3} (40)
k

14
 2
where L := Li Li = La La is L2 projected on H , Pel is the projector on its eigenspace with
2
1 1+ k
eigenvalue l(l + 1), and C = k
+ 2+1
. Moreover, the square distance from the origin is
2
(+1)2 +1 e
 
2 i i a a L +1
R := x x = x x =1+ 1+ P . (41)
k k 2 + 1
2
By the last equation, again R2 can be expressed as a function of L only, grows with the
latter, and its spectrum collapses to 1 (apart from the highest eigenvalue) as k .

Of course, relations (40)4 , (41) hold only at leading order in 1/ k if also (34-37) do.
To obtain a fuzzy space depending only on one integer we can choose k as a function
of fulfilling (33); the commutative limit will be simply + (what implies k +).
One possible choice is k = 2 ( + 1)2 ; then (40) becomes
1
" ! #
 i j I 1 1 + (+1) 2
x , x = iijk 2 2
+ 2 2
+ Pe Lk (42)
( + 1) ( + 1) 2 + 1

and R2 1 as well.
We note that relations (39), (40), (42) are similar to those defining the Snyders Lie algebra,
because the commutator of the coordinates is a polynomial in the generator of rotations Li ,
more precisely proportional to the Li apart on H , and therefore are invariant under parity
xa xa (because L3 and L2 are), contrary to the fuzzy sphere of Madore.
The operators a = P x a P are such that
0
hlm0 , a lm i =
6 0 l0 = l 1, m0 = m a;

 
the explicit actions of the a and of the commutator a , b are in the appendix. As with
D = 2, the action of the a on the eigenfunction lm gives a vector which has a non trivial
projection on theHilbert subspace corrisponding to n = 1. Consequently, neither a a nor
the commutator a , b vanish as k +, i.e. a has not the usual commutative limit.

4.1 Realization of the algebra of observables through U so(4)

As the Lie algebra su(2) is spanned by {Ei }3i=1 fulfilling

[Ei , Ej ] = iijk Ek , (43)

3
so(4) ' su(2) su(2) is spanned by {Ei1 , Ei2 }i=1 , where we have abbreviated Ei1 := Ei 1,
Ei2 := 1 Ei , and

[Ei1 , Ej2 ] = 0, [Ei1 , Ej1 ] = iijk Ek1 , [Ei2 , Ej2 ] = iijk Ek2 . (44)

Li = Ei1 + Ei2 and Xi = Ei1 Ei2 make up an alternative basis of so(4) and fulfill

[Li , Lj ] = iijk Lk , [Li , Xj ] = iijk Xk , [Xi , Xj ] = iijk Lk . (45)

15
The Li close another su(2) Lie algebra. Applying the transformation (30) we obtain alter-
native generators labelled by a {, 0, +}, fulfilling
[L+ , L ] = L0 , [L0 , L ] = L = [X0 , X ], [X+ , X ] = L0 , (46)
[L , X ] = X0 , [L0 , X ] = X = [X0 , L ], [La , Xa ] = 0 (47)
(no sum over a), where we have abbreviated L2 = Li Li = La La , X 2 = Xi Xi = Xa Xa .
Let j be the unitary irreducible representation of U su(2) on the (2j + 1)-dimensional
Hilbert space Vj ; this is characterized by the eigenvalue j(j +1) of the Casimir C := Ei Ei .
The tensor product representation := of U so(4) ' Usu(2) Usu(2) on the
2 2
Hilbert space V := V V is characterized by the conditions C 1 := Ei1 Ei1 = 2 ( 2 + 1) =
2 2
Ei2 Ei2 =: C 2 , or equivalently
X L = L X = 0, X 2 +L2 = (+2); (48)
here and below we drop the symbol . V admits an orthonormal basis consisting of
common eigenvectors of L2 and L0 :
L0 |l, mi = m |l, mi L2 |l, mi = l(l + 1) |l, mi with 0 l and |m| l, (49)
in standard ket notation. V , H have the same dimension (+1)2 and the same decompo-
sition in irreducible representations of the Li subalgebra, and will be eventually identified.
We determine the action of the Xa P on the |l, mi. Because of the commutation relations
[L0 , Xa ] = aXa it must be Xa |l, mi = j=0 l,j a,m
|j, m + ai . In the appendix we show that
a,m
l,j = 0 unless j = l 1, and more precisely that the previous relations are fulfilled by
p
Xa |l, mi = dl Aa,m
l |l 1, m + ai + d l+1 Bl
a,m
|l + 1, m + ai , dl := (+1)2 l2 . (50)
The operators on H , and in particular La , xa , are naturally realized in [Usu(2) U su(2)],
identifying lm
as the vectors of the canonical basis |l, mi. For simplicity, we introduce the
operator := [ 4L2 + 1 1]/2; |l, mi is an eigenvector with eigenvalue l. The Ansatz
La = La , xa = g () Xa g(), (51)
automatically fulfills the the hermiticity relations xa = xa . Applying xa to |l, mi we find
xa |l, mi = g(l)g (l 1)dl Aa,m
l |l 1, m + ai + g (l + 1)g(l)dl+1 Bla,m |l + 1, m + ai ; (52)
this agrees with (35) if and only if for l > 1
q
l2
cl k
1+

g (l 1)g(l) = =p , (53)
dl (+1l)(+1+l)
which is solved by
v
l1
[Y2 ]
u
2
1 + (l2j)
u Ql1
u h=0 (+l2h) k
g(l) = t Ql (l12j)2
, (54)
h=0 (+l+12h) j=0 1 + k

16
where again [b] stands for the integer part of b. Alternatively, using the basic property
(z + 1) = z(z) of the Euler gamma function we can express a solution in the form
  
v 
l i k l i k
u
+l 2 +1+ 2 2 +1 2

l+1

u 2 +1 2
u
g(l) = t +1+l
l
    (55)
2 +1 2 +1 k l+1 + i k l+1 i k
2 2 2 2

(see the Appendix); this makes sense also for generic complex argument l. The inverse of
the transformation (51) is clearly Xa = [g ()]1 xa [g()]1 .
We have thus proved by an explicit construction

Proposition 4.2. Formulas (51), (55) define a O(3)-equivariant -algebra isomorphism

between the algebra A = End(H ) of observables (endomorphisms) on H and the C1 =



C2 = 2 2 + 1 irreducible representation of U so(4) ' Usu(2) Usu(2):

A := End(H ) ' MN (C) ' [U so(4)], N := (+1)2 . (56)

As already recalled, the group of -automorphisms of MN (C) ' A is inner and isomor-
phic to SU (N ), i.e. of the type (29) with g an unitary N N matrix with unit determinant.
A special role is played by the subgroup SO(4) acting in the representation , namely
g = [ei ], where so(4). In particular, choosing = i Li (i R) the automor-
phism amounts to a SO(3) SO(4) transformation (a rotation in 3-dimensional space).
Parity (Li , Xi ) 7 (Li , Xi ), or equivalently Ei1 Ei2 [the only automorphism of so(4) cor-
responding to the exchange of the two nodes in the Dynkin diagram], is a O(3) SO(4)
transformation with determinant 1 in the X1 X2 X3 space. This shows that (51) is O(3)-
equivariant.

5 Final remarks, outlook and conclusions

For both dimensions d = 1, 2 we have introduced a finite-dimensional approximation of
quantum mechanics on the sphere Sd by projecting below a suitable energy cutoff E quantum
mechanics of a particle in RD (D = d+1) configuration space subject to a rotation invariant
potential V (r) with a very sharp minimum on the sphere of radius r = 1. By parametrizing
both the confining parameter k and E by a positive integer we have obtained a sequence
of O(D)-equivariant such approximations. The following common features emerge. The
algebra of observables A on the Hilbert space H is isomorphic to [U so(D+1)], where
is a suitable irreducible unitary representation of Uso(D+1) on H . H carries a reducible
representation of the subalgebra Uso(D) generated by the projected angular momentum
components Lij , more precisely the direct sum of all irreducible representations fulfilling
the cutoff condition L2 ( + d 1); the same decompostion holds for the subspace
C A of completely symmetrized polynomials in the projected coordinates xi . In the

17
 limit these become the decompositions of the Hilbert space L2 (S d ) and of the
algebra of operators C(S d ) acting on L2 (S d ), respectively. The xi , or alternatively the
elements X i of a corresponding basis of so(D+1)\so(D), generate the algebra of observables
A ' [U so(D +1)] [the relation between them is of the form xi = g(L2 )X i g(L2 )]; their
commutators span so(D), the Lie algebra of angular momentum components Lij , as in the
Snyder algebra; a basis of A is made up of -images of mononomials - with a fixed ordering
- in the elements of a basis of so(D+1) , by the Poincare-Birkhoff-Witt theorem. The square
distance R2 = xi xi from the origin is not identically 1, but a function of L2 with spectrum
close to 1. The whole construction is SO(D+1)- and O(D) SO(D+1)-equivariant. In the
limit we recover ordinary O(D)-equivariant quantum mechanics on L2 (S d ), with the
observables Lij , xi going to the angular momentum components Lij and to the coordinates
xi of S d configuration space, because R2 1: H L2 (S d ), C C(S d ) and A goes to
the whole algebra of observables on L2 (S d ).
Our approach seems easily applicable with the same features to higher dimensions, where
comparison with previous proposals is possible. The fuzzy spheres of dimension d 3 of
[17, 20] are based on the algebra End(V ) of endomorphisms of the carrier space V of a
particular irreducible representation of SO(d + 1), so that the square distance form the
origin R2 be central and can be set strictly equal to 1. The commutation relations are also
of the Snyder type (although presumably slightly different from ours), hence equivariant
with respect to the full group O(d + 1). In [18, 19] Steinacker considers the possibility of a
4
fuzzy 4-sphere SN with a reducible representation of U so(5) on a Hilbert space V obtained
decomposing an irreducible representation of U so(6) characterized by a triple of highest
weights (N, n1 , n2 ); so End(V ) ' [U so(6)], in analogy with our scheme. The elements X i
of a basis of SO(6) \ SO(5) play the role of noncommutative cartesian coordinates. As a
consequence the SO(5)-scalar R2 = X i X i is no longer central, but its spectrum is still very
close to 1 if N  n1 , n2 ; note that in our approach this is guaranteed by adopting suitable
xi = g(L2 )X i g(L2 ) rather than X i as noncommutative cartesian coordinates. If n1 = n2 = 0
then R2 1, and one recovers the fuzzy 4-sphere [17]. Finally, the physical interpretation
of End(V ) is that it represents a fuzzy approximation of some fibre bundle on a sphere S 4
(rather than of the algebra of observables of a quantum particle on a S 4 ).

Acknowledgments

We are grateful to F. DAndrea and T. Weber for useful discussions. We acknowledge partial
support by COST Action MP1405 Quantum Structure of Spacetime.

6 Appendix
We shall repeatedly use the formula

+
r n/2    n2h
(2h 1)!!
Z
a2 +b n b2 /4a X n b
e d = e (57)
a h=0 2h 2a (2a)h

18
 R + 2
(for all a > 0 and b R), which can be easily derived from
ez dz = through
integration by parts and a linear change of integration variable.

6.1 Calculation of a rather general scalar product in D = 2

As a preliminary step, we prove a formula regarding a matrix element of a general form.
Proposition 6.1. For every entire function f () not depending on k and h Z the following
asymptotic expansion in 1/ k holds
2
   
f ih h
Tm,m0 := hm0 , f ()e m i = m0 m Km,m0 exp f () (58)
4cm,m0 = 0 m,m

3m2 m02
   
km + km0 1 1
cm,m0 := = 2k 1 + +O ,
2 2k 4k k

m2 +m02 +2
 
2+ km em + km0 em0 2 1
where m,m0 := = + +O , (59)
2cm,m0 2k 4k k 3/2

[2+ km em + km0 em0 ] km e2m +km0 e2m0
s 2
 
4 3 4cm,m0 2 1
Km,m0 := Nm Nm0 e =1+O 3 ,
cm,m0 k2
q
21k em
where Nm = 4 4km 3 e
m is the normalization of m (up to a phase). The powers in

1/ k arise from the Taylor expansion of the exponential and the definition of cm,m0 , e.g.
f 00 (m,m0 ) f (4) (m,m0 )
 
f 1
Tm,m0 = f (m,m0 ) + + +O .
4cm,m0 32(cm,m0 )2 k 3/2
If f () has continuous derivatives up to order 2h + 1 the above asymptotic expansion holds
up to order 2h.
As consequences, setting a := 1+ 49 12k + 64k
137


, we find up to terms O 1/k 3/2
0 0
2hm0 , um m m i = hm0 , ei(m m) m i = Km,m0 = 1, (60)
n2
0 nm,m0 +
hm0 , en+i(m m) m i = Km,m0 e 4cm,m0
, (61)
hm+1 , e+i m i

+ Km,m+1 m,m+1 + 4cm,m+1
1 1 9
hm+1 , x m i = = e = 1+
2 2 2 4 2k
   r
m(m+1)+137/32 a m(m + 1) a m(m + 1)
+ = 1+ = 1+ , (62)
2k 2 2k 2 k
 
1 9 m(m1)+137/32
hm1 , x m i = hm , x m1 i = 1+ +
+
2 4 2k 2k
r
a m(m 1)
= 1+ , (63)
2 k
a2
 
2 m(m 1)
|hm1 , x m i| = 1+ . (64)
2 k

19
 Proof of the proposition. From (13) we find
Z 2 Z +
f
Tm,m0 = d dr r m0 (r, )f ()eih m (r, )
0 0
Z +
2 km 2 km0
h
= Nm Nm0 2m0 m d f () e2(em ) 2 (em0 ) 2

Z +
km + km0 2 e2
km e2 0
m + km0
h +[2+ km em + km0 em0 ] m
= Nm Nm0 2m0 m d f () e 2 2

2
[2+ km

em + km0

em0 ]
e2
km e2 0
m + k m0 m
Z +
km +

km0
h 2( km + km0 ) 2 (m,m0 )2
= Nm Nm0 2m 0 m e d f () e 2

2
[
2+ km
em + km0
em0

km ]
e2 e2 0
m + k m0 m
Z +
2c
= h
Nm Nm0 2m 0 m e 4cm,m0 2
dz ez m,m0 f (m,m0 +z) ; (65)

in the last step we have changed the integration variable 7 z = m,m0 . Using the Taylor
expansion of f (m,m0 +z) and the vanishing of integrals of odd functions over R we find

Z+ Z+
2
z cm,m0 z 2 cm,m0
X zn
dz e f (m,m0 +z) = dz e f (n) (m,m0 )
n=0
n!

Z+ Z+
X
z 2 c z 2n X f (2n) (m,m0) 2 y
2n
= f (2n) (m,m0) dz e m,m0 = dy ey
n=0
(2n)! n=0 (cm,m0 )n+1/2 (2n)!


f (2n) (m,m0 ) 2n

X
r X r 
= = f ()
cm,m0 n=0 (4cm,m0 )n n! cm,m0 n=0 (4cm,m0 )n n! =m,m0

2
  

r 
= exp f () ;
cm,m0 4cm,m0 = 0 m,m

R + 2
here we have used the identity dy ey y 2n = (2n1)!!/2n , which can be proved iterating
integration by parts. Replacing in (65) we find (58), with
s 2
4 3 [ 2+ km
em + km0
em0 ]
e2
km e2 0
m + km0 m
Km,m0 := Nm Nm e
0 4ch
m
2
(66)
cm,m0

1
In particular choosing h = 0, f 1 and recalling the normalization condition Tm,0 =
2
km k = 1 we determine the normalization factors Nm :

s s
4 3 1 km 21k em
2 +2e
m 4
|Nm | e km =1 = Nm = e m
km 4 3

20
Hence,
s p
4 3 2 4 km km0 21km em 21k 0 em0
Nm Nm0 = p e m
cm,m0 km + km0

2 21k e
m 21k e m0 m 21k e
21k e m0
= rq q e m m0 = e m m0

4 km k
k 0
+ 4 kmm0
m

which replaced in (66) gives (59)3 . We now determine cm,m0 , m,m0 , Km,m0 at lowest order.
3
By (14), up to O(1/k 2 ),
3
m2 m2 1 3 4m2
 
p 1 2
p
km = 2k 1 + , km em = 1 + +
2k 2k 2k 4k
 
1 1 1 1 1 1 1
= + , = + ,
km 2k 2k km + km0 2k 2 2 2k

3m02 m2
 
km + km0 1
cm,m0 = = 2k 1 + ,
2 2k 4k

2 + km em + km0 em0 2 m2 + m02 + 2
m,m0 = = + .
2cm,m0 2k 4k

then using (10) and by algebraic manipulations one has

2
v [2+ em + km0
km em0 ]
e2
km e2 0
m + k m0 m 21k 21k e
m e
m0
u 2 (
2

km + km0 ) 2 m m0
Km,m0 =u
tq q e
km 4 km0
4
k m0
+ km
0 +k
2
E 0 0 + km0

v km km0 Em m
m
u 2 km km0
=u
tq q e 2( km + km0 )
km 4 km0
4
k m0
+ km
 2
v km km0
u 2 em + k1 e
m0 k1
e 2( )
km + km0 m m0
. =u
tq q
km 4 km0
4
k m0
+ km
.

and since
 2
km km0 1 1

em + em0 = 0,
2 km + km0 km km0
r r r
km m02 m2 4 km m02 m2 4 km km0
=1+ , =1+ , + 4 = 2,
km0 k km0 4k km0 km
 3
one has Km,m0 = 1, up to O k 2 .

21
6.2 Calculation of the action of operators in D = 2
We first compute + m :
(
X m1 hm1 , + m i if + 1 m
+ m = h hh , + m i = (67)
h 0 otherwise.

By (23) we can calculate this scalar product using (58) with h = 1 and

e h p i
f () = m ( em ) km ;
2
consequently,
3
m 3 m3 32 m2 + 79
63 
  
1 1 m 1
hm1 , + m i = m 4 8 32 64
+O 3 .
2 2 2k 2k k2
From (67) we have
m3 32 m2 + 79
(  
m
1 b+m 3m
32
m1 if + 1 m
+ m = 2 4 2k 2k (68)
0 otherwise.

where b = 12 + 3
8 2k
+ 63
128k
. Moreover

X m hm , m1 i if + 1 m
m1 = h hh , m1 i =
0 otherwise.
h

hm , m1 i = h+ m , m1 i = hm1 , + m i
m3 32 m2 + 79
   
1 3m 32
m 1
= b+m +O 3
2 4 2k 2k k2
whence (replacing m with m + 1)
(m+1)332 (m+1)2+79
(  
3(m+1) (m+1)
1 b+m+1 32
m+1 if m 1,
m = 2 4 2k 2k (69)
0 otherwise.

Eq. (68), (69) imply at leading order

(m+1) 2
( 
(m+1)3 32 (m+1)2 + 79

1 3(m+1)
b + m + 1 32
m if m 1,
+ m = 2 4 2k 2k
0 otherwise,
(  3 3 2 79
2
1 m 2 m + 32 m
b + m 43m
m if + 1 m ,
+ m = 2 2k 2k
0 otherwise,

22
whence
4m3+31
 m

3m


m 2 2k 2k
8
m if |m| 1,

m 2

m323 m2+79
 
1
  2 b+m 43m
2k
32
m if m = ,
+ , m = 2k (70)
(m+1)323 (m+1)2+79 (m+1) 2
 
1 3(m+1)

b+m+1 32
m if m = ,
2 4 2k 2k


0 otherwise.

and the analogous of laplacian

 3m2 3 2 27

2 1 +8 2m4+47m +32


m + 4 2k
2
2k
8
m if |m| 1,


m332 m2+95 m 2

1  
3m

b+m 32
m if m = ,


+ + + m = 2 4 2k 2k (71)
2
3 3
m +2 m +32 m+79
2 95

1 3(m+1)

b+m+1 4 2k
32
m if m = ,
2 2k


0 otherwise.

Then one can conclude that there exists 4 polinomyals P1 , P2 , Q1 , Q2 such that

     
+ , = P1 (L)+Q1 (L) Pe Pe and + + + = P2 (L)+Q2 (L) Pe Pe .
(72)

6.3 Spherical Harmonics

Let
+ x+iy r sin ei
x = r sin cos x = 2 = 2

i
y = r sin sin , x = xiy

2
= r sine2 ,
z = r cos

0
x = z = r cos
x+iy xiy
t0 = zr , t+1 =
2r
and t = ,
2r
then one has the following recurrence relations:
s s
(l + m)(l m) m (l + m + 1)(l m + 1) m
t0 Ylm = cos Ylm = Yl1 + Yl+1 ,
(2l + 1)(2l 1) (2l + 1)(2l + 3)
s s !
i
1 m sin e m 1 (lm)(lm1) m+1 (l+m+1)(l+m+2) m+1
t Yl = Yl = Y Yl+1 , (73)
2 2 (2l+1)(2l1) l1 (2l+1)(2l+3)
s s !
i
sin e 1 (l+m)(l+m1) (lm+1)(lm+2)
t1 Ylm = Ylm = Y m1 + m1
Yl+1 .
2 2 (2l+1)(2l1) l1 (2l+1)(2l+3)

The coefficients Aa,m

l , Bla,m are related by
a,m+a
Aa,m
l = hYl1 , t Yl i = hta Yl1
m+a a m m+a m+a
, Ylm i = hYlm , ta Yl1 i; = Bl1 (74)

23
and fulfill the properties (for all l 0, |m| l)
a,m+b+a
Ab,m
l Al + Ab,m+b
l+1
a,m+b
Al+1 = Aa,m
l Alb,m+a+b + Aa,m+a
l+1
b,m+a
Al+1 ,
X a,m a,m+a
Ab,m A
l+1 l
a,m+b
= Aa,m b,m+a
A
l+1 l , Al+1 Al = 0, (75)
a
X l X l+1 X a,m X
(Aa,m
l )2 = , a,ma 2
(Al+1 ) = , (Al )2 + (Aa,ma
l+1 )2 = 1.
a
2l+1 a
2l+1 a a

Actually, the latter are equivalent to the identities [ta , tb ] = 0, ta ta = 1 applied to Ylm .

6.4 Calculation of |Nl | in D = 3

|Nl | can be determined setting hlm , lm i = 1 and we will choose Nl = |Nl |; using the
orthonormality of the spherical harmonics, one obtains
Z + Z +
8

r
2 rl )2 kl
(re 2 rl )2 kl
(re 2 kl
1 = |Nl | e dr ' |Nl | e dr = |Nl | |Nl | =
0 kl 4

(76)

and the meaning of the approximation symbol ' is explained in subsection 6.9.

6.5 Calculation of a rather general scalar product in D = 3

We compute the useful scalar product
Z +
0 0
hLm , g(r)ta lm i = hYLm , ta Ylm i fl (r)fL (r)g(r)dr (77)
0

with a generic g(r) not depending on k; here we have used the decomposition lm (r, , ) =
Ylm (, ) flr(r) and factorized the integral over R3 into an integral over the angle variables
and the integral on the radial one. By (36) one finds
0
hYLm , ta Ylm i =
6 0 L = l 1 and m0 = m + a

The asymptotic expansion of the radial integral can be obtained from the general formula
2 +
+ kl kL (r
l L ) g (2n) (b
Z e e
r
X rl,L )
fl (r)fL (r)g(r)dr = e 2( kl + kL )

n , (78)
0 n=0
(2n)!! kl + kL

24
which we now prove:

el2 + 2

Z + Z + kl + kL kl r kL r
r2 +r( kl rel + kL reL )
eL
2 2
fl (r)fL (r)g(r)dr = Nl NL e g(r)dr
0 0
2 2 2
( kl +
L)
r
el + kL r kl r
el + kL r
eL Z 2
e kl + kL
= Nl NL e 2( kl + kL ) 2
e 2 (rbrl,L ) g(r)dr
0
2
kl kL (r +
l L )
e e
r Z kl + kL 2
(rbrl,L ) g(r)dr.
= Nl NL e 2( kl + kL )
e 2

kl rel + kL reL
with rel , kl as defined in (32) and rbl,L := kl + kL
. By Taylor expansion of g(r) around rbl,L ,
Z + Z +
kl + kL 2 kl + kL 2

e 2 (rbrl,L ) g(r)dr ' e (rbrl,L ) g(r)dr
2

0
Z + " +
X g (h) (b h
#
kl + kL 2 r l,L ) z
= e 2 z dr
h=0
h!
s +
(57) 2 X g (2n) (brl,L )
=
n ; (79)
kl + kL n=0 (2n)!! kl + kL

the equality ' holds up to terms vanishing exponentially as k +, as explained in

subsection 6.9. Now,
8
kl 2k + 4l(l + 1) l(l + 1) 3l2 (l + 1)2 3


Nl = 4 , kl = =1+ + O k ,
2k + 3l(l + 1) 2k (2k)2
p 3l(l + 1) 9l2 (l + 1)2  5
kl = 2k + 3 + O k 2 ,
2 2k 8(2k) 2
p 5l(l + 1) 21l2 (l + 1)2  5
kl rel = 2k + 3 + O k 2 ,
2 2k 8(2k) 2
2
kl kL (e r re ) [l(l + 1) L(L + 1)]2  5
l L
= 3 + O k 2
2 kl + kL 4 (2k) 2
3
l(l+1)+L(L+1) 4
l(l+1)L(L+1)+ 89 [l2 (l+1)2 +L2 (L+1)2 ] 3


rbl,L = 1+ + O k . (80)
4k 4k 2
But s
8
2 kl kL 2 2


= rq = 1 + O k ,
kl + kL 4 kl
q
4 kL
kL
+ kl

25
whence (78). Using relations (77), (78), one finds, for example

!
l2 g 00 (b
rl,l1 )
m+a
hl1 , g(r)ta lm i = Aa,m 1 + g (b
r )+ 2 + , (81)

l 3 l,l1 
(2k) 2 2
p
kl + kl1

!
(l+1) 2
g 00 (b
rl,l+1 )
m+a
hl+1 , g(r)ta lm i = Bla,m 1 + g (b
r )+ 2 + ) (82)

3
(2k) 2
l,l+1  p
2 kl + kl+1

6.6 Proof of (35) and of proposition 4.1

Using the decomposition xa = rta and the relations (36) one finds
m+a a m m+a m+a m+a

hl1 , rt l il1 + hl+1 , rta lm il+1 if l < ,
X
m+a m+a
xa lm = hhk , xa lm ihk = h1 , rta m i1 if l = , (83)

h,k
0 otherwise.

From (81) and (82) one has

m+a
hl1 , rta lm i = cl Aa,m
l , m+a
hl+1 , rta lm i = cl+1 Bla,m ,
 3 q (84)
l2 2
where, up to O k 2 , cl = 1 + 2k = 1 + lk 1 l , c0 = c+1 = 0.

cl is an integral over the r variable, while Aa,m

l is an integral over the angle variables; replacing
in (83) we find (35). The factorizations (84) are manifestations of the Wigner-Eckart theorem.
Using (35) one can calculate the action of the commutator on lm . For all l < and m
with |m| l, one has
 L0  L m
[x+ , x ]lm = (cl )2 (cl+1 )2 lm , [x0 , x ]lm = (cl )2 (cl+1 )2
 

2l + 1 2l + 1 l
Since
2
   
 2 2l+1 1 1
(cl ) (cl+1 )2 = c2

+O 3 if l < and = 1+ +O 3
k k2 k k2

we find at leading order in 1/ k
2
+ L0 0 L 1 1 + k
[x , x ] = + C Pe L0 , [x , x ] = C Pe L , C := + . (85)
k k k 2 + 1
+x
x+ x+x
Going back to the hermitean coordinates x x1 = 2
and y x2 = 2i
one finds
 
i j ijh I
[x , x ] = i + C P Lh
e
k

26
In this model the commutator is invariant under parity because the components of the
angular momentum are pseudo-vectors and they do not change sign under parity, while PE
is a projector, then a scalar.
Similarly, one has
h  i
1 + 1 (l2 + l + 1) + O 13 m if l <
R2 lm = h k 2   k 2 i l (86)

+ O 13 m if l =


1+
k 2+1 k2

or equivalently, at leading order in 1/ k,
1    2
 
2
X 1 2
R := 1 + (l + l + 1) Pel + 1 + Pe . (87)
l=0
k k 2 + 1
We see that the square distance from the origin is not central and equal to 1 on all the
representation (as in the standard quantization on the unit sphere), but its eigenvalues go
all to 1 as k +, except when l = . The last relation and (85) are exact if (34-37) are
adopted as exact definitions.
In order to prove that x L = 0, we must recall equation (30) and since (by definition)
L = 1i jkh xk h one has xL = 1i jkh xj xk h , but jkh is anti-symmetric and xj xk is symmetric,
j

then x L = 0. If one projects the previous operators on the Hilbert space H , since when
Lj acts on lm it preserves the square of the angular moment (because Lj doesnt increase
the value of l), that is PE Lj = Lj PE then one has PE xj Lj PE = PE xj PE Lj ; in conclusion,
the condition x L = 0 imples x L = 0. One can also verify this condition in this way, from
iLy
L = Lx 2
, then one has
x L = xi Lj ij = xa Lb eab = x+ L + x L+ + x0 L0 lm


 
1p +,m1 1p ,m+1 0,m
+cl l(l + 1) m(m 1)Al + l(l + 1) m(m + 1)Al + mAl l1,m
2 2
 
1p +,m1 1p ,m+1 0,m
+cl+1 l(l+1)m(m1)Bl + l(l+1)m(m+1)Bl +mBl l+1,m
2 2
and using (73) one obtains x L = 0.

6.7 Action and commutators of the a

In order to calculate the action of the derivation operators on the Hilbert space H , we first
note that
Ylm rl is a homogeneous polynomial of degree l in the xa variables. This implies
a Ylm rl = Cla,m Yl1 ma l1
, with some coefficients Cla,m . From the identities


r
[a , b ] Ylm rl = 0, a , xb ab Ylm rl = 0

 

one finds
Clb,m Cl1
a,mb
Cla,m Cl1
b,ma
= 0,
a,m+b
Ab,m a,m+b
l Cl1 + 2Ab,m
l Al1 Cla,m Al1
b,ma
= 0, (88)
2Ab,m
l Al
a,m+ba
+ Ab,m+b
l+1
a,m+b
Cl+1 Cla,m Ab,ma+b
l = ab .

27
By explicit calculations, one can conclude that Cla,m = (2l + 1)Aa,m l . On the other hand,
using
     
b xb b xb 0 b xb
fl (r) f l x l + 1 a f l x f l x 2xa
a l+1 = = 2x +
r xa (xb xb ) l+12 2 l+3
(xb xb ) 2
l+1
(xb xb ) 2 2 xb xb
 0 
fl (r) l + 1
= l+2 fl (r) ta ,
rl+1 r
and the results of section 6.5 we find
 
m fl (r) l m fl (r) a,m ma
a l (r, , ) = a r Y l (, ) + C Yl1 (, )
rl+1 r2 l
 0 
fl (r) l + 1 fl (r)
= 2 fl (r) ta Ylm (, ) + 2 Cla,m Yl1 ma
(, )
r r r
 0   0 
fl (r) fl (r) a,m ma fl (r) fl (r)
= (l+1) 2 Bl Yl+1 + + l 2 Aa,m l
ma
Yl1 .
r r r r
Hence
ma
a lm = l1 [Ml + lJl ] Aa,m
l
ma
l+1 [Ml+1 + (l+1)Jl+1 ] Bla,m , (89)
where
+
l2
Z  
fl (r)fl1 (r) 1 1
Jl := dr = 1 + +O 3 , (90)
0 r 8k 2k k2
Z +
18l3
 
0 l 1
Ml := fl1 (r)fl (r)dr = + + O 3 ; (91)
0 2k 8k 2k k2
 
Jl , Ml have been evaluated using (78). The commutator a , b on lm is
(l + 1)2 l2

a , b l = (Jl+1 )2 (Jl )2
  m
+ 2(l + 1)Jl+1 Ml+1 2lJl Ml
2l + 1 2l + 1

2 1 2 1
+ (Ml+1 ) (Ml ) a,b Lab lm
2l + 1 2l + 1

with

0 if a=b
1 if a = 0, b = +1

a,b = b,a =

1 if a = 0, b = 1
1 if a = 1, b = +1

6.8 Proof of proposition 4.2 and other results of subsection 4.1

Let i := iijk Xj Lk . Using (45) we easily find
L2 Xi = Xi L2 + 2(Xi + i ), L2 i = (i + 2Xi )L2 . (92)

28

This suggests to look for eigenvectors of L2 (and therefore of = [ 4L2 + 1 1]/2),
L2 i = i (), in the form i = Xi a() + i b(). The compatibility condition is a second
degree equation with the right eigenvalue () as the unknown. Up to -dependent factors
the solutions are

i = [Xi i ] (21)(1), := (1),
(93)
+
i = [Xi (+1) + i ] (2+3) + := (+1)(+2)
(here we have chosen the coefficients a(), b() so that
i = i ); this implies

i = i ( 1). (94)
Therefore a |l, mi |l 1, m + ai. Inverting (93) one can express the Xa (as well as the a )
a,m
as linear combinations of a with -dependent coefficients; hence l,j = 0 unless j = l 1,
and the Ansatz (50), as anticipated.
On the other hand, using (74-75) and the equalities
l,m
A,m
l
+,m+1
Bl1 A+,m
l
,m+1
Bl1 = m
2l+1
, A,m
l
0,m1
Bl1 A0,m
l
,m
Bl1 = 2l+1
,

d2l d2l+1 = 2l + 1, Aa,m

l
a,m+a
l1 = Ala,m+a la,m ,

,m
Bla,m l+1
a,m+a
= Bla,m+a la,m , A0,m
l l1 A0,m1
l l,m = A,m
l ,

,m
Bl0,m l+1 Bl0,m1 l,m = Bl,m , A,m
l
,m1
l1 A,m1
l l,m = A0,m
l ,

Bl,m l+1
,m1
Bl0,m1 l,m = Bl0,m ,
we obtain, for example,
[X+ , X ] |l, mi = dl dl1 A,m +,m1 +,m ,m+1


l A l1 A l A l1 |l 2, mi +
2 ,m +,m1 +,m ,m+1


dl Al Bl1 Al Bl1 |l, mi +
2 ,m +,m1 +,m ,m+1


dl+1 Bl Al+1 Bl Al+1 |l, mi +
dl+1 dl+2 Bl,m Bl+1 +,m1
Bl+,m Bl+1 ,m+1


|l + 2, mi =
,m +,m1 +,m ,m+1
d2l d2l+1 Al Bl1 Al Bl1



|l, mi =
m
2l + 1 |l, mi = m |l, mi = L0 |l, mi ,
2l + 1
[X+ , X0 ] |l, mi = dl dl1 A0,m A+,m +,m 0,m+1


l l1 Al Al1 |l 2, m + 1i +
0,m +,m +,m 0,m+1
d2l Al Bl1 Al Bl1


|l, m + 1i +
2 0,m +,m +,m 0,m+1


dl+1 Bl Al+1 Bl Al+1 |l, m + 1i +
0,m +,m +,m 0,m+1


dl+1 dl+2 Bl Bl+1 Bl Bl+1 |l + 2, m + 1i =
d2l d2l+1 A0,m +,m
A+,m 0,m+1



l Bl1 l Bl1 |l, m + 1i =
l+,m
2l + 1 |l, m + 1i = l+,m |l, mi = L+ |l, mi .
2l + 1

29
Since the Ansatz (50) differs from (36) only by the l-dependent coefficients dl , and l is not
changed by the action of the Lb , the fact that (31) holds for v a = ta guarantees that it holds
also for v a = Xa , i.e. proves the relations of the form [La , Xb ] = fab
c
Xc in (47).
In addition, because of

A+,m1
l l,m + A,m1
l l+,m + A0,m
l m = Bl+,m1 l,m + Bl,m1 l+,m + Bl0,m m = 0, (95)
,m+1
A+,m
l l1 + A,m
l
+,m1
l1 + A0,m
l
,m+1
m = Bl+,m l+1 + Bl,m l+1
+,m1
+ Bl0,m m = 0 (96)
and
l
A0,m
l
0,m
Bl1 + A,m
l
+,m1
Bl1 + A+,m
l
,m+1
Bl1 = (97)
2l + 1
we obtain
 
2 l
d2l+1 d2l d2l+1


X |l, mi = + |l, mi = [( + 2) l(l + 1)] |l, mi (98)
2l + 1
whence we can easily derive (48).
To prove (55) we first note that, using the basic property (z + 1) = z(z),
l1 l1   +l


+l l 2 +1
Y Y
l
(+l2h) = 2 h = 2 l

=: h1 (l)
h=0 h=0
2 2
+1
l l   +l+1


l+1 +1
Y Y +l+1
l+1 2
(+l+12h) = 2 h = 2 l+1

=: h2 (l).
h=0 h=0
2 2

Hence h1 (l)/h2 (l) gives the first ratio under the square root in (54), implying

+l+1

l+1

h1 (l) h1 (l1) 2
2 1
= l++1

l+1
= . (99)
h2 (l) h2 (l1) 4 2 +1 2 +1 (+1+l)(+1l)

l +1i k
 
l2
On the other hand, it is 1 + k
= k1 (l+i k)(li k). Setting h (l) := 2 2l+1 i2k  we find
2
2

   
l i k l+1 i k
2
+1 2
2
2 l2
h (l)h (l 1) = 2     = li k 1+ = f (l)f (l1)(100)
l+1 i 2k 2l i 2 k k
2

p
where f (l) := h+ (l)h (l)/ k. Therefore g(l) = h1 (l)f (l)/h2 (l), i.e. (55), solves (53).

6.9 Shifting the lower extreme of integration over r

Here we justify the approximation used in (76) and (79):
Z +
Z +
rl )2 kl
(re 2
e g(r)dr ' e(rerl ) kl g(r)dr. (101)
0

30
We first consider g(r) 1 and estimate for b > 0 and a + the following difference:
Z + Z +  Z + Z +
a(rb)2 a(r+b)2 z 2 dz

e dr = e dr = e = erfc b a
0 0 b a a 2 a
ab2
 
e 1
= 1 + . (102)
2ba 2ab2

Here we have used: the changes of integration variables r
7
R +z2 r, r
7 z = a(y + b) in the
2
first, second equalities; the definition erfc(x) = x e dz and the x asymptotic
x2 
expansion erfc(x) = ex 1 2x12 + of the complementary error function in the third and


fourth. We can apply the above formula to the integral (76) setting a = 2k+ , b = 1+ ;
consequently, the  error made shifting from 0 to the lower extreme in integrating over r
2k 
is of the order 1/ 2 2ke , which has zero asymptotic expansion in 1/ 2k, hence does
not contribute to the expansion of the integral: here are the meaning and the justification of
the symbol '. One obtains a similar result after a suitable number of integration by parts
also if g(r) is polynomial (or more generally analytic); this justifies (79).

References
[1] H. S. Snyder, Phys. Rev. 71 (1947), 38.
[2] Letter of Heisenberg to Peierls (1930), in: Wolfgang Pauli, Scientific Correspondence,
vol. II, 15, Ed. Karl von Meyenn, Springer-Verlag 1985.
[3] J. A. Wheeler, Ann. Phys. 2 (1957), 604.
[4] S. Deser, Rev. Mod. Phys. 29 (1957), 417.
[5] C. A. Mead, Phys. Rev. 135 (1964), B849.
[6] S. Doplicher, K. Fredenhagen, J. E. Roberts, Phys. Lett. B 331 (1994), 39-44; Commun.
Math. Phys. 172 (1995), 187-220;
[7] D. Bahns, S. Doplicher, G. Morsella, G. Piacitelli, Advances in Algebraic Quantum Field
Theory (2015), 289-330; and references therein.
[8] A. Connes, Noncommutative geometry, Academic Press, 1995.
[9] J. Madore, An introduction to noncommutative differential geometry and its physical
applications, Cambridge University Press, 1999.
[10] J. Madore, Journ. Math. Phys. 32 (1991) 332; Class. Quantum Grav. 9 (1992), 6947.
[11] J. Hoppe, Quantum theory of a massless relativistic surface and a two-dimensional bound
state problem, PhD thesis, MIT 1982; B. de Wit, J. Hoppe, H. Nicolai, Nucl. Phys. B305
(1988), 545.

31
[12] H. Grosse, J. Madore, Phys. Lett. B283 (1992), 218.

[13] H. Grosse, C. Klimcik, P. Presnajder, Int. J. Theor. Phys. 35 (1996), 231-244.

[14] P. Aschieri, H. Steinacker, J. Madore, P. Manousselis, G. Zoupanos SFIN A1 (2007)

25-42; and references therein.

[15] R. Jackiw. Int. Conf. Theor. Phys., Ann. Henri Poincare 4, Suppl. 2 (2003), pp. S913-
S919, Birkhauser Verlag, Basel, 2003.

[16] G. Magro, preprint quant-ph/0302001, 2003.

[17] H. Grosse, C. Klimcik, P. Presnajder, Commun. Math. Phys. 180 (1996), 429-438.

[18] H. Steinacker, J. High Energ. Phys. (2016) 2016: 156.

[19] H. Steinacker, arXiv:1704.02863

[20] S. Ramgoolam, Nucl. Phys. B610 (2001), 461-488; and references therein.

[21] F. DAndrea, F. Lizzi, P. Martinetti, J. Geom. Phys. 82 (2014), 18-45.

32