TUW - 94 - 03

hep-th/9405110
arXiv:hep-th/9405110v1 17 May 1994

Poisson Structure Induced (Topological) Field

Theories

PETER SCHALLER∗
THOMAS STROBL†

Institut für Theoretische Physik

Technische Universität Wien
Wiedner Hauptstr. 8-10, A-1040 Vienna
Austria

Abstract
A class of two dimensional field theories, based on (generically degen-
erate) Poisson structures and generalizing gravity-Yang-Mills systems, is
presented. Locally, the solutions of the classical equations of motion are
given. A general scheme for the quantization of the models in a Hamilto-
nian formulation is found.

Vienna, May 1994

∗ e-mail: schaller@email.tuwien.ac.at
† e-mail: tstrobl@email.tuwien.ac.at
 As we will show in the present letter, any Poisson structure on any finite di-
mensional manifold naturally induces a two dimensional topological field theory.
Further nontopological theories are obtained by adding nontrivial Hamiltonians
to the topological theories. Pure gravity and gauge theories in two dimensions
may be seen as special cases of this much larger class of models.
Choosing special coordinates on the Poisson manifold N , the equations of
motions of the models associated to the Poisson structure on N can be solved
locally.
The Poisson structure generates a foliation of N into symplectic leaves and
induces a connection in the loop space over each of these leaves. Reinterpreting
the constraints of the associated model as horizontality conditions on complex
line bundles over these loop spaces, a general scheme for the quantization of
the models in a Hamiltonian formulation (restricting the topology of the space
time manifold to the one of a cylinder) is obtained. It yields finite dimensional
quantum mechanical systems, generically including discrete degrees of freedom
of topological origin.
Denote by M3 a three dimensional manifold, whose boundary M = ∂M3
is the world sheet of a two dimensional field theory. Denote by N a manifold
of arbitrary dimension with a (generically degenerate) Poisson structure. The
latter may be expressed in terms of an antisymmetric tensor field P ∈ ∧2 T N .
Likewise P may be regarded as a map T ∗ N → T N mapping a one-form to
its contraction with the tensor field. As a consequence of the Jacobi identity
the flow of the Hamiltonian vector field Vf = P (df, ·) associated to an arbi-
trary function f on N leaves the Poisson structure invariant. Furthermore, the
Hamiltonian vector fields form a closed Lie algebra corresponding to an (infinite
dimensional) subgroup GH of the P-morphism group.
Let
(N, P ) ֒→ K
↓ (1)
M3
be a fiber bundle over M3 with fiber N and GH as the structure group. Let
A be a connection on K; denote by D and F the covariant derivative and the
curvature, respectively. As locally the connection is one-form valued on M and
a Hamiltonian vector field on N , we have in coordinates xµ on M3 and X i on
N
1
P = P ij ∂i ∧ ∂j , A = Aµ,i dxµ P ij ∂j (2)
2
where we used the abbreviations
∂
∂i = , Aµ,i = ∂i Aµ . (3)
∂X i
Expressing the square of the covariant derivative via a Hamiltonian vectorfield,

1
the curvature is calculated:
D = d + A, D ∧ D = F,i P ij ∂j ,
1 ij (4)
F = (Aν,µ + 2 Aµ,i Aν,j P + Cµν )dxµ ∧ dxν , Cµν,i P ij = 0 .

The appearance of the Cµν reflects an ambiguity in the definition of F .

With curvature and covariant derivative it is straightforward to formulate a
GH invariant action: Z
L= F, i ∧ DX i . (5)
M3

If d Cµν dxµ ∧ dxν = 0, the integrand is exact and the action may be rewritten
as a functional of fields Ai (x), X i (x) on M according to

L = M=∂M3 Ai ∧ dX i + 12 P ij Ai ∧ Aj + C ,
R
(6)
Ai (x) = Aµ,i (x, X(x))dxµ , C(x, X) = Cµν dxµ ∧ dxν

It is invariant under the symmetries induced by GH -transformations

δǫ X i = ǫi (x)P ij , δǫ Ai = dǫi + P lm ,i Al ǫm (7)

up to a total divergence only.

Generically, there is no natural choice of C except for C = 0, yielding a
topological field theory.1 Given a volume-form ε on M and a GH -invariant
function C̃ on N , however, a possible choice of C is given by C = εC̃.
From (6) the equations of motion follow immediately:

dX i + P ij Aj = 0 ,
(8)
dAi + 21 P lm ,i Al ∧ Am + C,i = 0 .

By construction the action is invariant under diffeomorphisms of M for C =

0. Indeed, in this case a diffeomorphism in the direction of a vectorfield ξ ∈ T M
is generated on-shell by the field dependent choice ǫ = iξ A in (7).
Though P is degenerate in general, it defines a bijective map T ∗ N/Ker(P ) →
Im(P ). The Jacobi identity guarantees Im(P ) to be an involutive system of
vector fields and (by the Frobenius theorem) to generate an integral surface
S (symplectic leave) through any point X ∈ N with T S ∼Im(P ). For the
antisymmetry of the Poisson structure Ker(P ) is orthogonal to Im(P ). On S
this allows the identification T ∗ S ∼ T ∗ N/Ker(P ). So P induces a bijective map
T ∗ S → T S. By virtue of the Jacobi identitity the two-form ΩS ∈ T ∗ S ∧ T ∗ S
associated with the inverse of this map is closed and thus symplectic.
Excluding singular points of P , we may always find local coordinates (Casimir-
Darboux coordinates) (X I , X α ), I = 1, ..,dim(ker(P )), such that the symplectic
leaves are described by X I = const and, in terms of the X α , ΩS is in Darboux
1 In this case an action equivalent to (6) has been proposed independently in [1].

2
form for each S. In these coordinates C is a function of the X I only and the
equations of motion simplify to

dX I = 0 , dAI = −C,I , Aα = (ΩS )βα dX β . (9)

So the X I (x) have to be constant on M , but otherwise arbitrary, whereas the

X α (x) remain completely undetermined by the field equations. Any choice of
the latter determines Aα uniquely through the last equation in (9). Each of the
AI is determined up to an exact one-form only.
Still one has not made use of the gauge freedom. As is obvious from (7) any
choice of the X α is gauge equivalent, and also AI ∼ AI + dhI , where the hI are
arbitrary functions. Thus locally any solution to the field equations is uniquely
determined by the constant values of the X I .
Additional structures evolve, if global apects are taken into account. These
will be dealt with elsewhere.
For a Hamiltonian formulation of the theory let us assume M to be of the
form S 1 × R parametrized by a 2π-periodic coordinate x1 and the evolution
parameter x0 . One then finds the zero components A0i of the Ai to play the
role of Lagrange multipliers giving rise to the system of first class constraints
(∂ = ∂/∂x1 )
Gi ≡ ∂X i + P ij A1j ≈ 0 . (10)
The fundamental Poisson brackets are given by {X i (x1 ), A1j (y 1 )} = −δji δ(x1 −
y 1 ) and the Hamiltonian reads (C = Ĉdx0 dx1 )
Z
H = dx1 (Ĉ − A0i Gi ) . (11)

To quantize the system in an X-representation we consider quantum wave

functions as complex valued functionals on the space ΓN of parametrized smooth
loops in N :
ΓN = {X : S 1 → N, x → X(x)} . (12)
Following the Dirac procedure, only such quantum states are admissible which
satisfy the quantum constraints
 
i i ij δ
Ĝ (x)Ψ[X ] = ∂X (x) + ih̄P (X) Ψ[X ] = 0 . (13)
δX j (x)
To find the solution of (13, let us again start with Casimir Darboux coordi-
nates: The I-components of the constraints ∂X I Ψ(X ) restricts the support of
Ψ to loops which are contained entirely in some symplectic leave S.
Let us denote the space of symplectic leaves by S. For S ∈ S let ΓS be the
space of loops on S. The ansatz Ψ|ΓS = exp Φ for the restriction of Ψ to ΓS
allows to rewrite the remaining constraint equations according to

A = −ih̄δΦ , (14)

3
where δ denotes the exterior derivative on ΓS and A is the one-form on ΓS given
by Z
A= δX α (x)(ΩS )αβ (X(x))∂X β (x)dx . (15)
S1
(Reinterpreting A as a connection in a U (1)-bundle over ΓS the constraint may
be seen as a horizontality condition on the section Ψ|ΓS .)
Equation (14) seems to suggest that A has to be exact. This is not true,
however, as Φ is determined up to an integer multiple of 2π only. Therefore, Ψ
is well defined, iff A is closed and integral, i.e. the intgral of A over any closed
loop in ΓS is an integer multiple of 2πh̄. To reformulate (15) in a coordinate
independent way consider a path γ in ΓS . As γ corresponds to a one parameter
family of loops in S, it spans a two dimensional surface σ(γ). A may now be
defined alternatively via Z Z
A= ΩS . (16)
γ σ(γ)

As any closed loop in ΓS generates a closed surface in S, A is closed and integral,

iff ΩS is closed and integral. The first condition holds, as ΩS is symplectic. The
integrality condition, however, may yield a restriction of the support of Ψ to
loops over elements of a (possibly discrete) subset of S, if the second homotopy
of the symplectic leaves is nontrivial. For S in this subset, Ψ is determined
up to a multiplicative constant on any connected component of ΓS . As the
space of connected components of ΓS is in one to one correspondence with the
first homotopy group of S, we may identify physical states with complex valued
functions on I defined via
[
I= Π1 (S) S̃ = {S ∈ S : ΩS integral} . (17)
S∈S̃

The Hamiltonian (11), being constant on each of the symplectic leaves, defines
a function on I and thus becomes a multiplicative operator upon quantization.
Obviously these results are coordinate independent, although we intermediately
used Darboux coordinates to derive them.
It would be interesting to reproduce the quantum mechanical system ob-
tained above via a BRST cohomology. This is out of the scope of the present
letter. Here we only want to mention that (despite the appearance of structure
functions P ij ,k in the constraint algebra) the BRS charge takes the minimal
form
1
Q = ci Gi − ci cj P ji ,k bk , (18)
2
where (ci , bi ) are canonically conjugate fermionic ghost variables. The nilpo-
tency of Q follows from the Jacobi identity.
Let us illustrate our formalism by the example of nonabelian gauge theories
(cf. also [2]). There N is the dual space g ∗ of the Lie algebra g of the gauge
group. The Lie product on g naturally induces a Poisson structure on g ∗ . The

4
symplectic leaves generated by the latter are the coadjoint orbits, i.e. the orbits
generated by the action of the gauge group in g ∗ equipped with the standard
symplectic form (Kirillov form) [3]. The coadjoint orbits are quantizable spaces,
iff this symplectic form is integral and their quantization yields the unitary irre-
ducible representations of g. This observation establishes a connection between
our representation and the connection representation of quantum mechanics
for non abelian gauge theories on a cylinder: In the connection representation,
where wave functions are functionals on the space of gauge connections, the
physical wave functions (i.e. the kernel of the constraints) can be identified with
functions on the space of unitary irreducible representations of g [4].
Choosing a three dimensional linear space N , gravitational theories are con-
structed from (5) via the identification

Aa = ea , A3 = ω , a = 1, 2 (19)

where ea and ω are the zweibein and the spin connection of the gravity theory.
The coordinates X a , X 3 become vector valued and scalar valued functions,
respectively, living on the two dimensional manifold M with the metric g =
ea eb κab . The latter is of Euclidian or Minkowski type, corresponding to the
respective signature of the frame metric κab = diag(1, ±1).
The action of a gravity theory has to be invariant under diffeomorphisms
on M and under frame rotations. The first condition yields C = 0 in (6). The
general solution to the Jacobi identity in three dimensions is given by [5]

P ij = f εijk Q,k , (20)

where ε is the antsymmetric symbol and f and Q are arbitrary functions on

N . Invariance under frame rotations restricts f and Q to be functions of X 3
and (X)2 = κab X a X b . Thus the gravity action we propose is given by (6) with
the Poisson structure defined by (20), where f and Q are subject to the above
restriction.
A class of examples is provided by

P ij = εijk uk , (21)

where ua = κab X b and u3 is an arbitrary function of X 3 and (X)2 . (It is

somewhat cumbersome, to derive (21) from (20). It is straightforward, however,
to convince oneself that (21) obeys the Jacobi identity). With (21) the action
(6) takes the form
Z
LGr = [X a (dea + εa b ω ∧ eb ) + X 3 dω + u3 (e1 ∧ e2 )] . (22)

Obviously the torsion Ta = dea + εa b ω ∧ eb vanishes on shell, if u3 depends on

X 3 only. For some special choices of u3 one may integrate out the X-fields to

5
reformulate LGr in terms of purely geometric quantities. E.g., u3 = (X 3 )2 − λ
leads to the action of torsionless R2 gravity.
A more detailed analysis of these and further examples is found in [6], where
many of the ideas underlying this paper have already been presented in a less
abstract form.

Acknowledgement
We are grateful to A. Alekseev for discussions and valuable comments and to S.
Shabanov for collaboration on related topics.

References
[1] N. Ikeda, Two-Dimensional Gravity and Nonlinear Gauge Theory, preprint
RIMS-953 (hep-th 9312059).
[2] D. Amati, S. Elitzur, E. Rabinovici, On Induced Gravity in 2-d Topological
Theories, preprint hep-th/9212003.
[3] Kirillov, A.A., Uspeki Mat. Nauk 17/4 (1962) 57. Kostant, B. (1970) Quan-
tization and unitary representations. In Lectures in modern analysis III
(ed. C.T. Taam). Lecture notes in mathematics, Vol. 170. Springer, Berlin.
Souriau, J.-M (1970) Structure des systmes dynamiques Dunod, Paris; Kir-
illov, A.A. (1976) Elements of the theory of representations Springer, Berlin.
Woodhouse, N.M.J. Geometric Quantization, second Ed. 1992, Clarendon
Press, Oxford.
[4] S.G. Rajeev, Phys. Lett. B 212 (1988) 203. A. Migdal, Sov. Phys. Jept. 42
(1976)
[5] J.Grabovski, G.Marmo. A.M.Perelomov; Poisson structures: towards a
classification, preprint ESI 17 (1993)

[6] P.Schaller, T.Strobl: Quantization of Field Theories Generalizing Gravity-

Yang-Mills Systems on the Cylinder; preprint TUW 94-02 (to be published
in LNP)