REGGE CALCULUS IN THE CANONICAL FORM

Vladimir Khatsymovsky1

———————————————————————- –

(3+1) (continuous time) Regge calculus is reduced to Hamiltonian form. The constraints are

classified, classical and quantum consequences are discussed. As basic variables connection

matrices and antisymmetric area tensors are used supplemented with appropriate bilinear con-
arXiv:gr-qc/9310004v1 3 Oct 1993

straints. In these variables the action can be made quasipolinomial with arcsin as the only

deviation from polinomiality. In comparison with analogous formalism in the continuum the-

ory classification of constraints changes: some of them disappear, the part of I class constraints

including Hamiltonian one become II class (and vice versa, some new constraints arise and some

II class constraints become I class). As a result, the number of the degrees of freedom coincides

with the number of links in 3-dimensional leaf of foliation. Moreover, in empty space classical

dynamics is trivial: the scale of timelike links become zero and spacelike links are constant.

———————————————————————- —-

1
Budker Institute of Nuclear Physics, Novosibirsk 630090, Russia

1
 1.INTRODUCTION.

Regge calculus [1] attracts much attention in connection with possibility to construct quan-

tum gravity theory free of ultraviolet divergences. Such the possibility is usually connected

with discrete nature of the set of field variables. The latter are link lengths of flat tetrahedrons

forming piecewiseflat Regge manifold. To introduce canonical quantisation we need continuous

time Hamiltonian formalism. It was studied in a number of works [2]-[7]. My strategy is that

of refs.[3, 8], in which required formalism is the limit of 4-dimensional Regge calculus while

distances between successive spacelike leaves tend to zero. The main problem is an adequate

choice of variables allowing one to describe in the continuous time limit all the degrees of free-

dom of an arbitrary Regge manifold and to pass to Hamiltonian formalism in the simplest

way.In refs. [11] tetrad-connection variables were used first considered by Bander in ref.[12].

In [11] formulation of Regge calculus was suggested analogous to Einstein-Cartan formalism in

the continuum general relativity (GR) and, by passing to the continuous time limit, Lagrangian

was found, although not for quite general Regge manifold. Using these results some trivial low-

dimensional models were considered in refs.[13] illustrating possible versions of arising finite

quantum theory.

In the given paper Einstein action for arbitrary Regge manifold is considered in the contin-

uous time limit and reduced to the canonical form.

2.DESCRIPTION OF THE SYSTEM.

Our main object is Regge manifold or simplicial complex [15]. The vertices or null-di-

mensional simplices σ 0 will be denoted by capital letters A, B, C, . . .; n-simplex σ n (unordered)

will be denoted by also unordered sequence of it’s n + 1 vertices: σ n = (A1 . . . An+1 ). Nn(d) is

the number of n-simplices in d-dimensional manifold (may be, infinite). In particular, the

2
number of n-simplices meeting at given m-simplex σ m will be denoted as Nn(d) (σ m ). Now local

frames are defined on 4-simplices σ 4 = (ABCDE). In these frames a, b, . . . = 0, 1, 2, 3 are

vector indices; metric is ηab = diag(−1, 1, 1, 1) and antisymmetric tensor ǫabcd corresponds to

ǫ0123 = +1. A ◦ B, A ∗ B are scalar and dual products of two matrices, respectively; ∗B is dual

matrix:

def 1 ab
A◦B = A Bab
2
def
A ∗ B = A ◦ ( ∗B) (1)

∗ ab def 1 ab cd
B = ǫ B
2 cd

In the local frames the following elements of SO(3, 1) are defined: connection matrices Ω(ABCD)

on 3-simplices (ABCD) and curvature matrices R(ABC) on 2-simplices (ABC). Besides, also

antisymmetric tensors (bivectors) v(ABC) are defined on 2-simplices: two vectors l1a , l2a form the

triangle with bivector

v ab = ǫabcd l1c l2d , (2)

def
whose norm |v| = (v ◦ v)1/2 is twice the area of the triangle.

Einstein action for Regge manifold is written in the form [11]:

X v(ABC)
S= |v(ABC)| arcsin ◦ R(ABC) . (3)
(ABC)
|v(ABC)|

Here function arcsin gives angle defect on a triangle in terms of curvature matrix R. The latter

is product of connection matrices:

ε(ABC)D (ABC)Dr ε
R(ABC) = Ω(ABCD11) . . . Ω(ABCD r)
, (4)

where ε(ABC)D = ±1 is sign function, whose argument is pair tetrahedron (ABCD) - triangle

(ABC). The only requirement imposed on this function is consistency of eqs. of motion for

3
Ωσ3 which is equivalent for particular Ω’s to closure condition for 2-surface of 3-face σ 3 . This

condition includes 2-face bivectors rotated by connection matrices required to transform these

bivectors to the same frame. In particular, in the neighbourhood of flat space Ω = 1 it takes

the form

ε(ABC)D v(ABC) + ε(DAB)C v(DAB) + ε(CDA)B v(CDA) + ε(BCD)A v(BCD) = O(Ω − 1). (5)

Consistency of such the conditions for 3-faces of 4-simplex (ABCDE) sharing common edge

(AB) requires that

ε(ABC)D ε(ABC)E ε(ABD)E ε(ABD)C ε(ABE)C ε(ABE)D = −1. (6)

Next some constraints on bivectors v are required ensuring their tetrad structure. The

difficulty is that neighbouring bivectors well may be defined in the different frames; namely, a

4-simplex (ABCD0 E0 ) exists for each (ABC) where v(ABC) is defined (to reflect this fact let us

introduce the more detailed notation

v(ABC) ≡ v(ABC)D0 E0 (7)

). Therefore it is natural to consider for each (ABC) the set of all (ABCDE) containing this

triangle and to define a priori arbitrary corresponding v(ABC)DE . Now, what conditions should

be fullfilled in order that this set of bivectors would correspond to some Regge manifold where

these bivectors are given by (2)? First, consider 4-simplex (ABCDE) and a vertex A in it.

The triangles sharing A satisfy relations on dual products of bivectors the same as those for

bivectors in the continuum theory at a given point [14]:

v(ABC)DE ∗ v(ABC)DE = 0, perm(B, C, D, E) (8)

v(ABC)DE ∗ v(ABD)CE = 0, perm(B, C, D, E) (9)

4
 ε(ABC)D ε(ADE)C v(ABC)DE ∗ v(ADE)BC +

ε(ABD)C ε(ACE)D v(ABD)CE ∗ v(ACE)BD = 0, perm(B, C, D, E). (10)

Second, the sum of bivectors in any tetrahedron is zero:

ε(ABC)D v(ABC)DE + ε(DAB)C v(DAB)CE + ε(CDA)B v(CDA)BE + ε(BCD)A v(BCD)AE = 0 (11)

in any of two 4-simplices (ABCDE) sharing the tetrahedron (ABCD). Not all of the relations

(8) - (11) are independent ones since modulo (11) validity of (8) - (10) at any vertex A means

their validity at remaining three vertices. If (8) - (11) hold, tensors v in the 4-simplex are

bilinears of it’s edges just as analogous continuum theory tensors are tetrad bilinears.

Finally, third, we need conditions ensuring unambiguity of linklengths recovered from v in

the different 4-simplices. In the continuum theory such the problem did not exist since the

tetrad was local function of the bivector. Now we can require continuity of scalar products of

bivectors on 3-face (ABCD) shared by 4-simplices (ABCDE) and (ABCDF ):

def
∆(v(ABC)D ◦ v(ABC)D ) = v(ABC)DE ◦ v(ABC)DE − v(ABC)DF ◦ v(ABC)DF = 0,

perm(A, B, C, D), (12)

def
∆(v(ABC)D ◦ v(ABD)C ) = v(ABC)DE ◦ v(ABD)CE − v(ABC)DF ◦ v(ABD)CF = 0,

perm(A, B, C, D). (13)

By (11) there are 6 such independent conditions for each 3-face. This number is sufficient for

continuity of it’s 6 edges. Eqs. (13), e.g., are sufficient. But the system (13) (modulo (8) - (11))

is still highly reducible: it is sufficient to require continuity of the edges of a triangle on only

all but one of tetrahedrons meeting at this triangle to get continuity on all such tetrahedrons.

In terms of only linklengths continuous symmetries of our system are absent since, generally

speaking, any change of linklengths means change of geometry. Extension of the set of variables

5
by inclusion of connection in our case is compensated by symmetry w.r.t. SO(3, 1) rotations

of local frames.

Thus, our formulation is characterised by action (3) and by the system of constraints (8) -

(13) of which we shall below extract irreducible ones.

3.CONTINUOUS TIME.

Here we derive the Lagrangian. In fact, it is generalisation on arbitrary Regge manifold of

the result of [11] written in bivector notations.

To pass to the continuous time let us divide the set of vertices of Regge manifold into 3-

dimensional leaves numbered by a parameter t which we call time and tend the step dt between

the leaves to zero. The points of the leaf will be denoted by indices i, k, l, . . .. Let us assume

the following consistency condition: each 4-dimensional simplex is formed by vertices of only

two neighbouring leaves and length of one of it’s edges is O(dt). This requires for each vertex

i at the leaf t occurence of it’s images i+ in the leaf t + dt and i− in the leaf t − dt such that

linklengths of (ii+ ) and (i− i) are of the order of dt. Any such 4-geometry is formed of given

3-leaf as follows. Let us choose any vertex i and consider it’s star in 3-leaf, i.e. the set of all

the simplices of the leaf containing this vertex. Connect the image i+ to all the vertices of this

star. Then analogous procedure can be repeated with the obtained ”mixed” leaf (where vertex

i is replaced by i+ ) and with some another vertex k. As a result, the leaf arises where two

vertices are taken at t + dt and others are at t. In analogous way all the rest of vertices can be

shifted in time untill we get the leaf all points of which are taken at the time t + dt. It is clear

that each thus obtained block of 4-geometry filling the space between the leaves t and t + dt is

specified by the consequence of the above defined time shifts of vertices. It is remarkable that

our Lagrangian will not depend on such the consequence.

6
 To pass to the limit dt → 0 let us choose sign function ε(ABC)D conveniently. In 3-

dimensional notations put

ε(i+ kl)i = −1, ε(ikl)i+ = +1 (14)

(this unify the form of the kinetic term)

Further, it is convenient when going to the continuous time to assume the continuity con-

dition: if (A1 A2 . . . An+1 ) → (B1 B2 . . . Bn+1 ) at dt → 0, then g(A1 A2 ...An+1 ) → g(B1 B2 ...Bn+1 ) for

a quantity g defined on n-simplices. (Convergence of one simplex to another is understood as

convergence of corresponding vertices Aj → Bj and of vectors of links (Aj Ak ) and (Bj Bk )).

In particular, let us choose for sign function

ε(i+ kl)m = ε(ikl)m , ε(ikl)m+ = ε(ikl)m . (15)

Also denote
def
εikl = ε(i+ ik)l . (16)

Then consistency condition for sign function (6) is equivalent to the following one:

εikl εikm ε(ikl)m ε(ikm)l = −1. (17)

Connection on spacelike tetrahedron at dt → 0 should describe parallel vector transport at

infinitesimal distance in time direction and thereby it takes the form

Ω(iklm) = 1 + f(iklm) dt. (18)

The same can be written for diagonal tetrahedrons with some vertices shifted to the next time

leaf. For continuity reasons corresponding antisymmetric matrices f do not change at such

shift (as those describing vector transport at infinitesimally close points and at infinitesimally

close directions). But this is even inessential since the resulting Lagrangian turns out to contain

7
([11]) only the sum h(iklm) of f ’s over all four types of tetrahedrons - (iklm) and it’s diagonal

images with different number of vertices shifted to the next time leaf; for example

h(iklm) = f(iklm) + f(iklm+ ) + f(ikl+ m+ ) + f(ik+ l+ m+ ) , (19)

where antisymmetric matrix h(iklm) is an analog of the continuum GR connection ω0 .

The tetrahedron connection is the discrete analog of continuum connection for transport

orthogonal to the tetrahedron. Let us denote timelike tetrahedron connection as

def
Ωi(kl) = Ω(i+ ikl) (20)

(and the same for tetrahedrons differing by time shifts of k, l).

For bivectors we denote

def def
nik(lm) dt = v(i+ ik)lm , π(ikl)m = v(ikl)i+ m (21)

Substituting the limiting form of variables into Regge action we get the Lagrangian where

analogs of the terms π ω̇, hDα π α and nα Rα of the continuum theory [14] can be viewed denoted

below LΩ̇ , Lh and Ln , respectively. Besides, some new terms appear due to the difference of

limiting curvature matrices R on spacelike and diagonal triangles from unity2 . Indeed, write

out the finite part of curvature matrix R(ikl) if, e.g., triangle (ikl) is common 2-face of the

timelike tetrahedrons (i− ikl) and (ik + kl):

ε − ε + †
R(ikl) = Ω(i(ikl)i (ikl)k
− ikl) Ω(ik + kl) + O(dt) = Ωi(kl) Ωk(li) + O(dt). (22)

Normals to the tetrahedrons (i− ikl) and (ik + kl) are, generally speaking, different, just as

vectors of links (i− i) and (kk + ) are (the latter being analogs of shift-lapse functions at different
2
The closure of these R to unity would be natural to assume for their contribution to L be finite [8]. However,

the finiteness can be achieved at noninfinitesimal R − 1 as well on condition that contributions of neighbouring

triangles cancel each other, just as in this work.

8
 +
✏k
✏✏✏ ❇❅
✏
✏✏✏ ❇❅
✏ ✏ ❇ ❅
✏
i+ ✏ ❇
❇ ✟✟
❅
l+
✟✟❇
✟✟ ❇
✟✟ ✏ k ❇
✟✟✏✏✏✑
✑ ❅ ❇
✟
✟✏✏✏ ✑ ❅ ❇
✟✏✟✏ ✑ ❅❇
✟
✏✏ ✑
✟
✏ ✑ ❅❇
✘
i
✑
✑
✘✘✘✘ l
✑ ✘✘✘
✑ ✘✘✘✘
✑ ✘
−✘✑✘✘
i

Figure 1: Infinitesimal 3-prism

points), so Ωi(kl) and Ωk(li) do not necessarily coincide. These matrices, however, are not quite

independent as follows from the equations of motion for connection; relation between them will

also ensure finiteness of the Lagrangian. Indeed, at the infinitesimal variation

†
δΩk(li) = wk(li) Ωk(li) dt, wk(li) = −wk(li) (23)

finite addends to the Lagrangian will arise only from potentially infinite terms (contribution

of the triangles with noninfinitesimal area and defects). There are two such terms containing

Ωk(li) - contributions of R(ikl) and R(ik+ l) .

Resulting variation of L is linear in wk(li) and leads to a constraint on Ω, π. Permuting

i, k, l we get the system which is solvable [11] and gives

Ωi(kl) = Ω(ikl) exp(φi(kl) π(ikl) + ∗φi(kl) ∗π(ikl) ), (24)

where φi(kl) , ∗φi(kl) are parameters. Noninfinitesimal contribution of (ikl) into action (and thus

infinite one into L) is proportional to φk(li) − φi(kl) . Contributions of diagonal triangles differs

by cyclic permutations of i, k, l so that the sum vanishes, e.g. (see Fig.1)

φk(li) − φi(kl) + φl(ik) − φk(li) + φi(kl) − φl(ik) = 0 (25)

9
Important is that these differences are to be multiplied by close up to O(dt) areas of images

of (ikl). In the next in dt order to get finite terms in the Lagrangian (below denoted as Lφ )

one should take into account infinitesimal area differences. The latter depend on n, the lateral

(timelike) 2-face bivectors.

The resulting Lagrangian reads

LRegge = LΩ̇ + Lh + Ln + Lφ (26)

X
LΩ̇ = π(ikl) ◦ Ω†(ikl) Ω̇(ikl) (27)
(ikl)
X X δ
(ikl)m −δ (ikl)m
Lh = h(iklm) ◦ ε(ikl)m Ω(ikl) π(ikl) Ω(ikl) (28)
(iklm) cycle perm iklm

def 1+ε
(δ = )
2
X nik
Ln = |nik| arcsin ◦ Rik (29)
ik |nik|
ikln ε ikl1 ε
(Rik = Ωi(kln)
. . . Ωi(kl1)
, εiklj = ε(iklj )lj−1 = −ε(iklj )lj+1 )
X X
Lφ = − π(ikl)m ◦ εikl φi(kl) nik(lm) (30)
(ikl) perm ikl

(entering last equation scalar products π(ikl)m ◦ nik(lm) do not depend on m due to the further

considered continuity of scalar products of bivectors). Appearing here in kinetic term bivector

π(ikl) is π(ikl)m at m = m− (ikl), i.e. it is bivector of a triangle (ikl) defined in the one of two

tetrahedrons (iklm± ) with the face (ikl) in 3-leaf whose vertices m+ (ikl), m− (ikl) (functions

of (ikl)) are defined according to ε(ikl)m± = ±1. We shall also write π(ikl)± or simply π± for

corresponding bivectors. Bivector nik is nik(lm) for some (lm). Thus, π− , Ω are dynamical

variables.

For varying in φ, ∗φ, Ω let us introduce matrices U = exp(φπ + ∗φ ∗π), so that

1 π
Ωi(kl) = Ω(ikl) Ui(kl) , φ= arcsin ◦ U. (31)
|π| |π|

It is convenient to treat Ω, U as matrices of general form and take into account the conditions

10
of orthogonality and required dependence of U on π with the help of Lagrange multipliers by

adding to LRegge the following terms:

X X
Lrot = B(ikl) ◦ (Ω†(ikl) Ω(ikl) − 1) + †
Pi(kl) ◦ (Ui(kl) π(ikl) Ui(kl) − π(ikl) )
(ikl) i(kl)
X †
+ Mi(kl) ◦ (Ui(kl) Ui(kl) − 1). (32)
i(kl)

Lagrange multipliers are symmetric (B, M) and antisymmetric (P ) matrices.

It remains to add to LRegge with the help of Lagrange multipliers constraints on bivectors (8)

- (13) where we shall pass to the notations π, n and extract irreducible constraints. Conditions

on the dual products v ∗ v ♮ where v ♮ is v or v ′ result in the constraints π ∗ π ♮ , π ∗ n and n ∗ n♮ .

Since algebraic sum of π in the tetrahedron in 3-leaf is zero, there are 6 independent constraints

π ∗ π ♮ in the tetrahedron. The number of constraints π ∗ n and n ∗ n♮ is 8 and 6, respectively,

in the tetrahedron at each vertex whose shift-lapse vector form given n’s.

The closure condition (11) for 3-leaf tetrahedrons reads

ε(ikl)m π(ikl)m + ε(mik)l π(mik)l + ε(lmi)k π(lmi)k + ε(klm)i π(klm)i = 0. (33)

For the timelike tetrahedrons conditions (11) allow us to express variations of bivectors π due

to time shift of any vertex of 3-leaf in terms of bivectors n. These conditions were already used

to express variations of π appearing when finding Lφ .

(3)
Subtracting from the number of components of π, n (which is 96N3 ) the number of con-

straints (33) and of those of v ∗ v ♮ type gives

28N (3) (34)

for the number of 4-prism parameters. This is natural since any 4-prism is defined by 22

linklengths; in addition, there are 6 rotational degrees of freedom.

11
 k+
✄❇❅
✄ ❇❅
✄ ❇ ❅
✄ ❇ ❅
✄ ❇ ✟✟✡ l+
✄ ✟✟❇ ✡
✄
✟ ✟ ❇
✡
✟✟✄ ✏ k ✡ ❇
✟✟✏✏✄✏ ✁❅✡ ❇
✟
✟ ✏✏ ✄ ✁ ✡ ❅ ❇
✟ ✟✏✏
✏ ✄ ✁✡ ❅❇
✏✟✏
✟✏ ✄ ✁✡ ❅❇
i ❍❍ ✄ ✁✡ ✟✟ l
❍❍ ✄ ✁✡ ✟✟
❍❍ ✄✁✡ ✟✟
❍❍✟
✁✄ ✟
✡
m

Figure 2: Diagonal 3-face (ik + lm) - common for 4-simplices (ikk + lm) and (ik + ll+ m).

Continuity conditions for scalar products (12) and (13) also take different form on spacelike

and timelike 3-faces. Namely, continuity on spacelike (and diagonal) faces means constraints

with derivatives, whose existence might change dynamical content of theory apart from

being simply analog of continuum GR. Fortunately, as it is proved below, given dynamical

constraints are consequences of the equations of motion for Lagrangian LRegge and

can be omitted

For example, consider continuity of the values like v ◦ v ♮ on 3-face (ik + lm) (see Fig.2).

The difference between bivectors of close spacelike (diagonal) 2-simplices in the same frame

on shifting the vertex k

def
Dk π(ikl)m /dt = (v(ik+ l)km − v(ikl)k+ m )/dt

= εkli nkl(im) − εkil nki(lm) , (35)

which is an analog of covariant derivative (in fact, already used when finding Lφ ). The differ-

ence of bivectors of the same 2-simplices in different frames (an analog of usual derivative)

def
δk π(ikl)m = v(ik+ l+ )km − v(ikl)k+ m . (36)

12
The continuity conditions connect δπ and Dπ:

π(ikl)m ◦ (δi π(ikl)m − Di π(ikl)m ) = 0, perm i, k, l; (37)

π(ilm)k ◦ (δi π(ikl)m − Di π(ikl)m )

+π(ikl)m ◦ (δi π(ilm)k − Di π(ilm)k ) = 0, perm i, k, l. (38)

However, δπ(ikl)m enter equations of motion for LRegge only in the form of full derivative

(and for such m that ε(ikl)m = −1)

π̇(ikl) dt = (δi + δk + δl )π(ikl) . (39)

Therefore constraints (37), (38) are equivalent to relations between π̇ and Dπ(ikl) = (Di + Dk +

Dl )π(ikl) :

π(ikl)m ◦ (π̇(ikl)m − Dπ(ikl) /dt) = 0 (40)

X
(Dπ(ikl)m /dt = εikl nik(lm) ),
perm ikl

π(ilm)k ◦ (π̇(ikl)m − Dπ(ikl)m /dt)

+π(ikl)m ◦ (π̇(ilm)k − Dπ(ilm)k /dt) = 0. (41)

Equations (40), (41) were earlier said to be consequences of the equations of motion for

Lagrangian LRegge supplemented with the rest of constraints on bivectors (without derivatives).

Indeed, (40) at ε(ikl)m = −1 arises immediately from LRegge at the following variation of con-

nection type variables:

Ω(ikl) → Ω(ikl) exp(ξ(ikl) π(ikl) ),

φi(kl) → φi(kl) − ξ(ikl) , perm i, k, l. (42)

Namely, the constraint (40) turns out to be added to Lagrangian multiplied by −ξ(ikl) . Further,

there is area continuity condition |π(ikl)m−|2 = |π(ikl)m+|2 among the scalar product continuity

13
constraints. Differentiating it will lead to (40) also at ε(ikl)m = +1. Besides, in the nondegener-
(3) (3)
ate Regge manifold N2 ≥ N1 (this follows from simple combinatorial discussion with taking

into account the fact that each edge is shared by no less then three 2-faces). Therefore all the

links in 3-leaf and, in turn, scalar products of different bivectors π can be expressed through

triangle areas. Corresponding relations of the type π ◦ π ′ = f (|π|) should follow from the below

written irreducible set of linear and bilinear constraints on bivectors π, n corresponding to

Regge manifold. Since we already know how to differentiate areas, using these relations will

give derivatives d(π ◦ π ′ )/dt in terms of π, n. The obtained relations are purely kinematic ones

valid for arbitrary Regge manifold and therefore these are no else than (41).

Thus we have shown that kinematic constraints with derivatives (40), (41) follows from

equations of motion for Lagrangian supplemented with constraints without derivatives and

should be omitted.

It remains to separate out irreducible conditions of continuity of scalar products (12), (13)

on timelike 3-faces. In 3-leaf this corresponds to continuity on (spacelike) triangles. On the

triangle (ikl) shared by tetrahedrons (iklm), (ikln) we have for the bivectors of timelike and

spacelike triangles meeting at vertex i:

def
∆(π(ikl) ◦ π(ikl) ) = π(ikl)m ◦ π(ikl)m − π(ikl)m ◦ π(ikl)m = 0, (43)

∆(π(ikl) ◦ nik ) = 0, (44)

∆(π(ikl) ◦ nil ) = 0, (45)

∆(nik ◦ nik ) = 0, (46)

∆(nil ◦ nil ) = 0, (47)

∆(nik ◦ nil ) = 0. (48)

By permutations of i, k, l we get additionally 5 analogous equations at vertices k, l (equation

14
(43) remains unchanged). Continuity of edges of tetrahedron (ii+ kl) and, in particular, of the

triangle (ikl) follows from (43) - (48). But continuity of the triangle (ikl) follows also from

the constraints at vertices k, l as well, that is, some constraints are superfluous. In any case,

it is sufficient to keep constraints ∆(π ◦ n) at only one vertex of each triangle which gives
(3)
their full number 2N2 . The latter is also abundant: the constraint ∆(π(ikl) ◦ nik ) at all others

fullfilled expresses continuity of the length of (ik) in 3-leaf which should be stated on all but
(3) (3)
one triangles (ikl) meeting at this edge. Their full number thus becomes 2N2 − N1 . The

constraint ∆(nik ◦nil ) can be associated with length continuity of edge (ii+ ). It suffices to impose
(3) (3) (3)
it on N3 (i) − 1 meeting at i triangles (ikl). Summation over vertices gives 4N3 − N0 for

the number of independent constraints of this type. Finally, constraints ∆(nik ◦ nik ) are given
(3) (3)
on all but one triangles meeting at (ordered) 1-simplex (ik); their full number is 6N2 − 2N1 .
(3)
The number of constraints ∆(π(ikl) ◦ π(ikl) ) is N2 . As a result, the full number of constraints
(3) (3) (3) (3) (3)
of the type of ∆(v ◦ v ♮ ) is 22N3 − 3N1 − N0 (with taking into account that N2 = 2N3 ).

This should be subtracted from (34) to give

(3) (3) (3)

6N3 + 3N1 + N0 (49)

(3)
degrees of freedom. Of this number in each 3-leaf 6N3 is the number of parameters of local
(3) (3)
rotations, N0 is the number of timelike lengths while N1 is the number of spacelike ones; since

we consider block of 4-geometry between the two leaves, we take into account here the number
(3) (3)
of spacelike edges in two leaves, 2N1 , plus the number of diagonal edges N1 . However, when
(3)
we glue different blocks together, we need N1 continuity conditions on 3-leaf between them.

These are just conditions contained in (40), (41) and shown above to follow from equations
(3) (3)
of motion. As a result, we have 2N1 + N0 independent linklengths at arbitrary time as it

should be for the Regge 4-manifold constructed of the most general 3-leaf.

15
 P
The constraints introduced, v ∗ v ♮ and v, v ◦ v ♮ , can be taken into account with the help

of Lagrange multipliers by adding to Lagrangian the terms Ldual and Lscal , respectively:

X ∗
X ∗
Ldual = µ(ik)(lm) π(ikl)m ∗ π(ikm)l + νi(kl)m nik(lm) ∗ nil(km)
(ik)(lm) i(kl)m
X X
∗ ∗
+ νik(lm) nik(lm) ∗ nik(lm) + λiklm π(ikl)m ∗ nik(lm)
ik(lm) iklm
X
∗
+ λik(lm) [εilk εimk π(ilm)k ∗ nik(lm) − ∗Λi(klm) ], (50)
ik(lm)
X X X
Lscal = η(iklm) ◦ ε(ikl)m π(ikl)m + µ(ikl)m (π(ikl)m ◦ π(ikl)m − A(ikl) )
(iklm) cycle perm iklm (ikl)m
X X
+ νik(lm) (nik(lm) ◦ nik(lm) − σik ) + χ1 (ikl)λiklm (π(ikl)m ◦ nik(lm) − Λikl )
ik(lm) iklm
X
+ χS (i(kl))[λiklm (π(ikl)m ◦ nik(lm) − Λikl ) + λilkm (π(ikl)m ◦ nil(km) − Λilk )]
i(kl)m
X
+ χΣ (i(kl))νi(kl)m (nik(lm) ◦ nil(km) − σi(kl) ). (51)
i(kl)m

Here ∗µ, ∗
ν, ∗
λ, µ, ν, λ, η and also ∗Λ, A, Λ, σ are sets of Lagrange multipliers;

χ1 , χS , χΣ are characteristic functions of some sets of simplices S1 , S, Σ, arising at

constructing irreducible set of constraints above. The S, Σ are sets of 2-simplices with marked

vertex, on each of which 2 constraints ∆(π◦n) and(or) 1 constraint ∆(n◦n′ ) are set, respectively.

S1 is the set of 2-simplices with marked both vertex and edge on which 1 constraint ∆(π ◦ n) is

set. It is convenient that the sets S, Σ be chosen so that continuity of π(ikl) ◦nik , π(ikl) ◦nil and

nik ◦ nil on necessary number of triangles were fullfilled simultaneously in order that continuity

of edges on these triangles would follow immediately. I have check possibility of such choice for

two simple examples of 3-leaf: the simplest periodic Regge manifold [16] and simplest closed

one - 3-surface of the 4-simplex.

As a result, quasipolinomial Lagrangian of Regge calculus takes the form of the sum of

expressions (26), (32), (50) and (51):

L = LRegge + Lrot + Ldual + Lscal (52)

16
 4.THE STRUCTURE OF CONSTRAINTS.

Proceeding to discussion of dynamics, consider full time derivative of some quantity f in

the system with Lagrangian (52) which can be written symbolically as

L = π ◦ Ω† Ω̇ − H. (53)

Here H is function of π, Ω and other variables. If f is function of π, Ω then it follows with

the help of equations of motion that

df
= {f, H}, (54)
dt

where Poisson brackets for specific form of the kinetic term in L prove to be

{f, H} = π ◦ [Hπ , fπ ] + Hπ ◦ Ω† fΩ − fπ ◦ Ω† HΩ . (55)

Here indices π, Ω mean corresponding derivative, which over π is assumed to be antisym-

metrised.

The Hamiltonian H, as in the continuum theory, turns out to be linear combination of con-

straints, i.e. it vanishes on their surface. Indeed, for Ldual , Lscal , Lrot it is so by construction.

It is also evident for Lh , while Ln + Lφ is the sum over vertices of the groups of terms −Hi

each of which is uniform function of degree 1 of the set nik for all possible k at given i. The n’s

of this set can be multiplied by some general factor without violating other constraints. This

variation leads to Hamiltonian constraint Hi :

X
Ln + Lφ = − Hi , Hi = 0. (56)
i

Requiring the constraints be conserved in time allows us to define Lagrange multipliers. Those

at II class constraints are defined uniquely and therefore in the absence of matter are zero.

Therefore classical dynamics is governed in this case by I class constraints.

17
 Proceeding to classification of constraints let us first establish continuous symmetries. The

latter correspond to occurence of I class constraints. Originally in the full discrete theory we

have symmetry w.r.t. SO(3, 1) rotations in the local frames in 4-simplices. In the continuous

time limit we have rotations in 4-prisms or, equivalently to say, in their tetrahedron bases; also

we have some transformations of φ, ∗φ. Tetrahedron rotations U(iklm) ∈ SO(3, 1) result in

†
v → U(iklm) vU(iklm) , (57)

ε †
(ikl)m
Ω(ikl) → (U(iklm) Ω(ikl) U(ikln) )ε(ikl)m , (58)

† †
h(iklm) → U(iklm) h(iklm) U(iklm) − U̇(iklm) U(iklm) , (59)

where v is bivector π or n in the tetrahedron (iklm); (ikln) is another tetrahedron in 3-leaf

with the same 2-face (ikl). It is easy to check that on functions of π, Ω infinitesimal rotations

U = 1 + u are generated by Gaussian constraint C(u) = −Lh |h→u by means of Poisson brackets

{C(u), ·} (see (55)).

The invariance at shifts φ, ∗φ is due to ambiguity when dividing Ωi(kl) into symmetric

in i, k, l part and rotation exp(φπ + ∗φ ∗π) not changing π(ikl) . In particular, symmetry

transformations at shift ∗φ take the form

∗ ∗
φi(kl) → φi(kl) − ζ(ikl) , (60)

Ω(ikl) → Ω(ikl) exp(ζ(ikl) ∗π(ikl) ), (61)

∗ ∗ 1
µ(ik)(lm) → µ(ik)(lm) + ζ̇(ikl) , . . . cycle perm i, k, l . . . (62)
2

(up to addition full time derivative to the Lagrangian). Generator here is the constraint π(ikl) ∗

π(ikl) , which, although not written explicitly in Lagrangian, is combination of constraints of the
P
type π ∗ π ′ and π.

Situation for shift of φ is complicated by occurence of linear in φ terms in the Lagrangian:

analogous transformations (42) lead, as we have seen, to constraints with generalised velocities

18
 (3) (3) (3) (3)
π̇. On the other hand, since N2 ≥ N1 , there exist N2 − N1 relations fα (|π|2 ) on scalar

squares of π. These constraints are consequences of our full set of constraints in Ldual , Lscal

and are I class constraints generating transformations (42) for the following particular choice

of parameters:
∂fα
ξ(ikl) = ξ α . (63)
∂(|π(ikl) |2 )

Then, up to the full derivative, the following term in the Lagrangian arises:

1 X ∂f Dπ(ikl)
∆ξ L = ξ˙α fα + ξ α π(ikl) ◦ (64)
2 (ikl)
∂(|π(ikl) |2 ) dt

(Dπ is defined in (40)). First term is here combination of constraints. In the second one the

differences of constraints fα between neighbouring 3-leaves arise. These differences are some

algebraic constraints on π, n and should be combinations from our full set in Ldual , Lscal as

well.

Thus, the I class constraints are encountered. These are Gaussian one and kinematic rela-

tions for scalar and dual squares of π. All other constraints, apart from those in Ldual , Lscal ,

should arise when varying L in nondynamical variables π+ , n, φ, ∗φ. Since the latter enter L

nonlinearly, the equations obtained do not give, generally speaking, any constraints on dynam-

ical variables π− , Ω, but rather simply allow one to express nondynamical variables in terms

of dynamical ones. However, an important exception exists: the scale of length of shift-lapse

vector at any vertex i enters L linearly. Therefore, first, bivectors nik at this vertex are defined

by given equations only up to the common scale, and, second, variation in this scale gives the

above mentioned Hamiltonian constraint (56) at this vertex. This constraint follows by acting

on L the following operator:

X ∂
nik ◦ . (65)
k ∂nik

Substituting into Hi the values of nondynamical variables in terms of dynamical ones gives a

19
constraint on π− , Ω. Nontrivial equations of gravity itself arise in Regge calculus at varying

edge lengths, the Hamiltonian constraint corresponding variation in timelike edges. Variation in

spacelike and diagonal edges means variation in π− and gives not the constraints but equations

of motion containing time derivatives.

As for the momentum constraints, these might arise, in analogy with continuum GR ([14]),

by acting on L the operator

∂ ∂
π(imk)l ◦ − π(iml)k ◦ . (66)
∂n(ik)lm ∂n(il)km

This operator cancels Ldual , but now we have also Lscal not cancelled by this operator. As a

result, there are no analogs of the momentum constraints of continuum GR: shift vectors enter

L nonlinearly, therefore variation in them allows one only to find these themselves.
(3)
Thus our system in the space of dynamical variables π− , Ω is described by N0 Hamiltonian
(3)
constraints Hi , 6N3 components of Gaussian constraint C and by additional kinematical
(3) (3)
constraints on bivectors π− . The I class constraints are C, π ∗ π, fα (|π|2). Since N3 = 2N2 ,

it is convenient to define each π− in any of two tetrahedrons so that each tetrahedron would

contain two bivectors defined in it, π and π ′ . Then other constraints, a priori II class ones, are
(3) (3)
Hi , N3 constraints π ∗ π ′ and N3 functions gA expressing scalar products π ◦ π ′ in terms of

squares |π|:

π ◦ π ′ = gA (|π|2). (67)

It is easy to see that all kinematical constraints mutually commute w.r.t. the brackets (55).
(3)
Nonzero Poisson brackets arise only between Hi ’s in different points and between Hi and 2N3
(3) (3)
constraints π∗π ′ , π◦π ′ −gA . This means that also 2N3 −N0 I class combinations of functions
(3) (3) (3)
π ∗ π ′ , π ◦ π ′ − gA exist. On the whole, there are 6N2 − N1 − N0 I class constraints. As
(3) (3)
2N0 II class ones we can take, in addition to Hi , also some N0 of products π ∗ π ′ . Without

20
taking into account the constraints the number of the degrees of freedom would coincide with
(3)
the number of canonical pairs 6N2 . Taking into account the constraints we get this number
(3)
coinciding with the number of edges N1 . This should be expectable since, generally speaking,

change of the length of any edge means change of geometry of 3-leaf.

We are faced also with some peculiarity connected with that Hamiltonian constraint is

II class one. As a result, the length of shift-lapse vector N being Lagrange multiplier at this

constraint in empty space is zero. However, in the presence of matter this singularity dissappear.

For example, contribution of electromagnetic field Fµν into action containes the terms of the

form

g 00 g αβ F0α F0β V, (68)

where V is the volume of 4-simplex, gµν is metric. Since V ∼ N, g 00 ∼ N −2 , the given terms

are proportional to N −1 , so that equations of motion give strictly nonzero value of N. One

can say that the matter fields prevent the collapse in time axis by developing the pressure from

within the 4-simplices.

Vanishing the timelike lengths in empty space leads also in some sence to triviality of classical

dynamics in this space. Indeed, in this case Hamiltonian reduces to linear combination of I

class constraints. Since all these commute with π ◦ π, the areas as well as links do not vary

in time. However, normalised bivectors n/|n| have quite complex dynamics. This means that

parameters of embedding the 3-leaf into 4-manifold have a nontrivial dynamics.

5.CONCLUSION.

Having got Regge calculus in canonical form we can write out puth integral as formal

solution to the canonical quantisation problem for this theory. The functional integral measure

is defined by volume element in phase space on hypersurface of constraints of the theory and

21
contains nonlocal factor which is determinant of the Poisson brackets of II class constraints.

The latter are original constraints of theory plus gauge conditions by the number of original I

class constraints. One of interesting feature of Regge theory is that (in the case of Euclidean

signature) integrations over connections (elements of SO(4), not of Lee algebra so(4), as in the

continuum theory) are finite and one does not need to fix the gauge, that is, to divide by the

gauge group volume. In this case the measure factor will be defined by simply the original II

class constraints of the theory (Hamiltonian and kinematical ones).

Another, unpleasant feature is that this measure is clearly singular in the vicinity of flat

manifold for which symmetry group is larger and classification of constraints changes. Therefore

in the vicinity of flat space the perturbation theory does not exist.

ACKNOWLEDGEMENTS.

The author is grateful to I.B.Khriplovich, V.N.Popov and G.A.Vilkovisky for useful discus-

sions.

22
References

[1] Regge, T.(1961) Nuovo Cimento, 19, 558.

[2] Lund, F., and Regge, T., unpublished.

[3] Collins, P.A., and Williams, R.M.(1973) Phys.Rev., D7, 965.

Collins, P.A., and Williams, R.M.(1974) Phys.Rev., D10, 3537.

[4] Williams, R.M.(1986) Class.Quantum Grav., 3, 853.

[5] Friedman, J.L., and Jack, I.J.(1986) J.Math.Phys., 27, 2973.

[6] Piran, T., and Williams, R.M.(1986) Phys.Rev., 33, 1622.

[7] Porter, J.(1987) Class.Quantum Grav., 4, 375.

Porter, J.(1987) Class.Quantum Grav., 4, 391.

Porter, J.(1987) Class.Quantum Grav., 4, 651.

[8] Brewin, L.(1987) Class.Quantum Grav., 4, 899.

Brewin, L.(1988) Class.Quantum Grav., 5, 839.

[9] Tuckey, P.A., and Williams, R.M.(1988) Class.Quantum Grav., 5, 155.

[10] Tuckey, P.A.(1987) ”Independent variables in 3+1 Regge calculus” Cambridge preprint

DAMTP 87-32.

[11] Khatsymovsky, V.(1989) Class.Quantum Grav., 6, L249.

Khatsymovsky, V.(1991) Class.Quantum Grav., 8, 1205.

23
[12] Bander, M.(1986) Phys.Rev.Lett., 57, 1825.

Bander, M.(1987) Phys.Rev., D36, 2297.

Bander, M.(1988) Phys.Rev., D38, 1056.

[13] Khatsymovsky, V.(1991) ”On the two-dimensional model of quantum Regge gravity”

Novosibirsk preprint INP 91-40.

Khatsymovsky, V.(1991) ”3D Regge gravity as an exactly soluble system” Novosibirsk

preprint INP 91-60.

[14] Khatsymovsky, V., ”On the polinomial formulations of general relativity”, in press.

[15] Sorkin, R.(1975) Journ.Math.Phys., 16, 2432.

Weingarten, D.(1977) Journ.Math.Phys., 18, 165.

[16] Rocek, M., and Williams, R.M.(1981) Phys.Lett., 104B, 31.

24