ANALYSIS AND GEOMETRY

ON MARKED CONFIGURATION SPACES

SERGIO ALBEVERIO, YURI KONDRATIEV
arXiv:math/0608344v1 [[Link]] 14 Aug 2006

EUGENE LYTVYNOV, AND GEORGI US

Abstract

We carry out analysis and geometry on a marked configuration space ΩM X over a Riemannian
manifold X with marks from a space M . We suppose that M is a homogeneous space M of a Lie
group G. As a transformation group A on ΩM M
X we take the “lifting” to ΩX of the action on X ×M of
the semidirect product of the group Diff 0 (X) of diffeomorphisms on X with compact support and
the group GX of smooth currents, i.e., all C ∞ mappings of X into G which are equal to the identity
element outside of a compact set. The marked Poisson measure πσ on ΩM X with Lévy measure σ
on X × M is proven to be quasiinvariant under the action of A. Then, we derive a geometry on
ΩMX by a natural “lifting” of the corresponding geometry on X × M . In particular, we construct a
gradient ∇Ω and a divergence divΩ . The associated volume elements, i.e., all probability measures
Ω Ω 2
µ on ΩM M
X with respect to which ∇ and div become dual operators on L (ΩX ; µ), are identified
as the mixed marked Poisson measures with mean measure equal to a multiple of σ. As a direct
consequence of our results, we obtain marked Poisson space representations of the group A and its
Lie algebra a. We investigate also Dirichlet forms and Dirichlet operators connected with (mixed)
marked Poisson measures.

1991 AMS Mathematics Subject Classification. Primary 60G57. Secondary 57S10, 54H15

0 Introduction
In recent years, stochastic analysis and differential geometry on configuration spaces have
been considerably developed in a series of papers [5–8], see also [37, 2, 3]. It has been
shown, in particular, that the geometry of the configuration space ΓX over a Rieman-
nian manifold X can be constructed via a simple “lifting procedure” and is completely
determined by the Riemannian structure of X. The mixed Poisson measures are then
exhibited as the “volume elements” corresponding to the differential geometry introduced
on ΓX . Intrinsic Dirichlet forms and operators, their canonical processes, as well as Gibbs
measures on configuration spaces, their characterization by integration by parts, and the
corresponding stochastic dynamics are among the problems which have been treated in
the above framework.
A starting point for this analysis, more exactly, for the definition of differentiation on
the configuration space, was the representation of the group of diffeomorphisms Diff 0 (X)
on X with compact support that was constructed by G. A. Goldin et al. [18] and A. M. Ver-
shik et al. [42] (see also [34, 38, 20]). The construction of this representation used, in turn,
the fact, following from the Skorokhod theorem, that the Poisson measure is quasiinvariant
with respect to the group Diff 0 (X).

1
 On the other hand, starting with the same work [42], many researchers consider rep-
resentations also on marked (in particular, compound) Poisson spaces. In statistical me-
chanics of continuous systems, marked Poisson measures and their Gibbsian perturbations
are used for the description of many concrete models, see e.g. [1]. Hence, it is natural to
ask about geometry and analysis on marked Poisson spaces. The first work in this direc-
tion was the paper [26], in which, just as in the case of the usual Poisson measure, the
action of the group Diff 0 (X) was used for the definition of the differentiation. However,
this group proved to be too small for reconstructing mixed marked Poisson measures as
“volume elements,” which means that Diff 0 (X) is to be extended in a proper way, which
we will describe in the present paper.
Let us recall that the configuration space ΓX is defined as the space of all locally finite
subsets (configurations) in X. Then, the marked configuration space ΩM X over X with
marks from, generally speaking, a manifold M is defined as

ΩMX := (γ, s) | γ ∈ ΓX , s ∈ M
γ
,

where M γ stands for the set of all maps γ ∋ x 7→ sx ∈ M . Let σ e be a Radon measure on
X × M such that σ e(K × M ) < ∞ for each compact K ⊂ X and σ e is nonatomic in X, i.e.,
σ M
e({x} × M ) = 0 for each x ∈ X. Then, one can define on ΩX a marked Poisson measure
πσe with Lévy measure σe.
Of course, one could consider πσe as a usual Poisson measure on the configuration space
ΓX×M over the Cartesian product of the underlying manifold X and the space of marks
M , and study the properties of this measure using the results of [2–5]. However, such
an approach does not distinguish between the two different natures of X and M and the
different roles that these play in physics. Thus, our aim is to introduce and study such
transformations of the marked configuration space which do “feel” this difference and lead
to an appropriate stochastic analysis and differential geometry.
In our previous paper [24], we were concerned with the model case M = R+ , which
corresponds, in fact, to the case of a compound Poisson measure. As has been promised
in [24], we generalize in the present paper the results of [24] to the case where M is a
homogeneous space of a Lie group G. This situation is natural from the physical point
of view. For example, one can take X = R3 and M to be the unit sphere S 2 in R3 , and
consider any marked configuration (γ, s) = {(x, sx )x∈γ } ∈ ΩMX as a system of particles in
3
R situated at the points x of γ and having spin sx at x ∈ γ. One has then to take G as
the rotation group, see e.g. [13].
Let GX denote the group of smooth currents, i.e., all C ∞ mappings X ∋ x 7→ η(x) ∈ G
which are equal to the identity element of G outside of a compact set (depending on
η). We define the group A as the semidirect product of the groups Diff 0 (X) and GX :
for a1 = (ψ1 , η1 ) and a2 = (ψ2 , η2 ), where ψ1 , ψ2 ∈ Diff 0 (X) and η1 , η2 ∈ GX , the
multiplication of a1 and a2 is given by

a1 a2 = (ψ1 ◦ ψ2 , η1 (η2 ◦ ψ1−1 )).

The group A acts in X × M as follows: for any a = (ψ, η) ∈ A

X × M ∋ (x, m) 7→ a(x, m) = (ψ(x), η(ψ(x))m) ∈ X × M ,

2
where, for g ∈ G and m ∈ M , gm denotes the action of g on m. Since each ω ∈ ΩM X can be
interpreted as a subset of X × M , the action of A can be lifted to an action in ΩM X . The
marked Poisson measure πσe is proven to be quasiinvariant under it. Thus, we can easily
construct, in particular, a representation of A in L2 (πσe ). It should be stressed, however,
that our representation of A is reducible, because so is the regular representation of A in
L2 (e
σ ), see subsec. 3.5 in [24] for details.
Having introduced the action of the group A on ΩM X , we proceed to derive analysis and
geometry on ΩM X in a way parallel to the works [7, 24], dealing with the usual configuration
R+
space ΓX and the marked configuration space ΩX , respectively. In particular, we note
that the Lie algebra a of the group A is given by a = V0 (X) × C0∞ (X; g), where V0 (X) is
the algebra of C ∞ vector fields on X having compact support and C0∞ (X; g) is the algebra
of C ∞ compactly supported functions from X into the Lie algebra g of the group G. For
each (v, u) ∈ a, we define the notion of a directional derivative of a function F : ΩM X →R
along (v, u), which is denoted by ∇Ω F . We obtain an explicit form of this derivative

(v,u)
on the special set FCb∞ D, ΩM X of smooth cylinder functions on ΩM X , which, in turn,
motivates our definition of a tangent bundle T (ΩM X ) of Ω M , and of a gradient ∇Ω F . We
X
note only that the tangent space Tω (ΩM X ) to the marked configuration space ΩMX at a point
M
ω = (γ, s) ∈ ΩX is given by

Tω (ΩM 2
X ) := L (X → T (X) ∔ g; γ),

where ∔ means direct sum.

Next, we derive an integration by parts formula on ΩM X , that is, we get an explicit
Ω
formula for the dual operator div of the gradient ∇ on ΩMΩ
X . We prove that the prob-
ability measures on ΩM for which ∇ Ω and divΩ become dual operators (with respect to
X
h·, ·iT (ΩM ) ) are exactly the mixed marked Poisson measures
X
Z
µκ,eσ = πzeσ κ(dz),
R+

where κ is a probability measure on R+ (with finite first moment) and πzeσ is the marked
Poisson measure on ΩM X with Lévy measure zeσ , z ≥ 0. This means that the mixed marked
Poisson measures are exactly the “volume elements” corresponding to our differential
geometry on ΩM X.
Thus, having identified the right volume elements on ΩM X , we introduce for each mea-
sure µκ,eσ the first order Sobolev space H01,2 (ΩM
X , µ σ
κ,e ) by closing the corresponding Dirich-
let form
Z
Ω


Eµκ ,eσ (F, G) = h∇Ω F, ∇Ω GiT (ΩM ) dπκ,eσ , F, G ∈ FCb∞ D, ΩM X ,
X
ΩM
X

on L2 (ΩM
X , µκ,e
σ ). Just as in the analysis on the usual configuration space, this is the step
where we really start doing real infinite dimensional analysis. The corresponding Dirichlet
operator is denoted by HµΩκ ,eσ ; it is a positive definite selfadjoint operator on L2 (ΩM
X , µκ,e
σ ).
Ω


The heat semigroup exp(−tHµκ ,eσ ) t≥0 generated by it is calculated explicitly. The results

3
on the ergodicity of this semigroup are absolutely analogous to the corresponding results
of [7]. Particularly, we have ergodicity if and only if µκ,eσ = πzeσ for some z > 0, i.e., µκ,eσ
is a (pure) marked Poisson measure.
We also clarify the relation between the intrinsic geometry on ΩM X we have constructed
with another kind of extrinsic geometry on ΩM X which is based on fixing the marked
Poisson measure πσe and considering the unitary isomorphism between L2 (ΩM X , πσ
e ) and
the corresponding Fock space
∞
M
2
F(L (X × M ; σ
e)) = L̂2 ((X × M )n , n! σ
e⊗n ),
n=0

where L̂2 ((X × M )n , n! σ e⊗n ) is the subspace of symmetric functions from

2 n
L ((X × M ) , n! σe ). Our main result here is to prove that HπΩσe is unitarily equiva-
⊗n

lent (under the above isomorphism) to the second quantization operator of the Dirichlet
operator HσeX×M on the L2 (X × M ; σ e) space.
As a consequence of the results of this paper, we obtain a representation on the marked
Poisson space L2 (πσe ) not only of the group A, but also of its Lie algebra a. Let us remark
that the groups of smooth (as well as measurable and continuous) currents are classical
objects in representation theory, see e.g. [4, 41, 11, 12, 43, 20] and references therein for
different representations of these groups. On the other hand, different representations of
the group A and its Lie algebra a, in the special case G = g = R, were constructed and
studied by G. Goldin et al. [17, 19, 16] from the point of view of nonrelativistic quantum
mechanics.
Finally, we note that, in a way parallel to the work [8], the results of the present
paper can be generalized to the interaction case where, instead of the Poisson measure πσe ,
describing a system of free particles, one takes its Gibbsian perturbation—more exactly,
a marked Gibbs measure on ΩM X of Ruelle type (see [28, 29]).

1 Marked Poisson measures

1.1 Marked configuration space
Let X be a connected, oriented C ∞ non-compact Riemannian manifold. The configuration
space ΓX over X is defined as the set of all locally finite subsets in X:

ΓX := γ ⊂ X | #(γ ∩ K) < ∞ for each compact K ⊂ X ,

where #(·) denotes the cardinality of a set. One can identify any γ ∈ ΓX with the positive
integer-valued Radon measure X
εx ∈ M(X),
x∈γ
P
where x∈∅ εx := zero measure and M(X) denotes the set of all positive Radon measures
on B(X).

4
 Let also M be a connected oriented C ∞ (compact or non-compact) Riemannian man-
ifold. The marked configuration space ΩM
X over X with marks from M is defined as

ΩMX := ω = (γ, s) | γ ∈ ΓX , s ∈ M
γ
,

where M γ stands for the set of all maps γ ∋ x 7→ m ∈ M . Equivalently, we can define ΩM X
as the collection of subsets in X × M having the following properties:
 
M a) ∀(x, m), (x′ , m′ ) ∈ ω : (x, m) 6= (x′ , m′ ) ⇒ x 6= x′
ΩX = ω ⊂ X × M ,
b) PrX ω ∈ ΓX

where PrX denotes the projection of the Cartesian product of X and M onto X. Again,
each ω ∈ ΩM
X can be identified with the measure
X
ε(x,m) ∈ M(X × M ).
(x,m)∈ω

It is worth noting that, for any bijection φ : X × M → X × M , the image of the

measure ω(·) under the mapping φ, (φ∗ ω)(·), coincides with (φ(ω))(·), i.e.,

(φ∗ ω)(·) = (φ(ω))(·), ω ∈ ΩM

X,

where φ(ω) = {φ(x, m) | (x, m) ∈ ω} is the image of ω as a subset of X × M .

Let Bc (X) and Oc (X) denote the families of all Borel, resp. open subsets of X that have
compact closure. Let also Bc (X × M ) denote the family of all Borel subsets of X × M
whose projection on X belongs to Bc (X).
Denote by C0,b (X × M ) the set of real-valued bounded continuous functions f on
X × M such that supp f ∈ Bc (X × M ). As usually, we set for any f ∈ C0,b (X × M ) and
ω ∈ ΩM
X Z X
hf, ωi = f (x, m) ω(dx, dm) = f (x, m).
X×M (x,m)∈ω

We note that, because of the definition of ΩMX , there are only a finite number of addends
in the latter series.
Now, we are going to discuss the measurable structure of the space ΩM X . We will use
a “localized” description of the Borel σ-algebra B(ΩM X ) over Ω M.
X
For Λ ∈ Oc (X), define

ΩM M
Λ := ω ∈ ΩX | PrX ω ⊂ Λ

and for n ∈ Z+ = {0, 1, 2, . . . }


ΩM
Λ (n) := ω ∈ ΩM
Λ | #(ω) = n .

It is obvious that
∞
G
ΩM
Λ = ΩM
Λ (n).
n=0

5
 Let Λmk := Λ × M (i.e., Λmk is the set of all “marked” elements of Λ) and let

e n := ((x1 , m1 ), . . . , (xn , mn )) ∈ Λn | xj 6= xk if j 6= k .
Λ mk mk

There is a bijection
(n) e n M
LΛ : Λ mk /Sn 7→ ΩΛ (n) (1.1)
given by
(n)
LΛ : ((x1 , m1 ), . . . , (xn , mn )) 7→ {(x1 , m1 ), . . . , (xn , mn )} ∈ ΩM
Λ (n),

where Sn is the permutation group over {1, . . . , n}. On Λnmk /Sn one introduces the related
metric
 
δ ((x1 , m1 ), . . . , (xn , mn )), ((x′1 , m′1 ), . . . , (x′n , m′n ))
 
= inf dn ((x1 , m1 ), . . . , (xn , mn )), ((x′σ(1) , m′σ(1) ), . . . , (x′σ(n) , m′σ(n) )) ,
σ∈Sn

where dn is the metric on Λnmk driven from the original metrics on X and M . Then,
e n /Sn becomes an open set in Λn /Sn and let B(Λ
Λ e n /Sn ) be the trace σ-algebra
mk mk mk
on Λe /Sn generated by B(Λ /Sn ). Let then B(Ω (n)) be the image σ-algebra of
n n M
mk mk Λ
B(Λe n /Sn ) under the bijection L(n) and let B(ΩM ) be the σ-algebra on ΩM generated by
mk Λ Λ Λ
the usual topology of (disjoint) union of topological spaces.
For any Λ ∈ Oc (X), there is a natural restriction map pΛ : ΩM M
X 7→ ΩΛ defined by

ΩM M
X ∋ ω 7→ pΛ (ω) := ω ∩ Λmk ∈ ΩΛ .

The topology on ΩM X is defined as the weakest topology making all the mappings pΛ
continuous. The associated σ-algebra is denoted by B(ΩMX ).
For each B ∈ Bc (X × M ), we introduce a function NB : ΩMX → Z+ = {0, 1, 2, . . . } such
that
NB (ω) := #(ω ∩ B), ω ∈ ΩMX. (1.2)
Then, it is not hard to see that B(ΩM M
X ) is the smallest σ-algebra on ΩX such that all the
functions NB are measurable.

1.2 Marked Poisson measure

In order to construct a marked Poisson measure, we fix:
(i) an intensity measure σ on the underlying manifold X, which is supposed to be a
nonatomic Radon one,
(ii) a non-negative function

X × B(M ) ∋ (x, ∆) 7→ p(x, ∆) ∈ R+

such that, for σ-a.a. x ∈ X, p(x, ·) is a finite measure on M .

6
 Now, we define a measure σ
e on (X × M , B(X × M )) as follows:
Z
σ
e(A) = p(x, dm) σ(dx), A ∈ B(X × M ). (1.3)
A
We will suppose that the measure σe is infinite and for any Λ ∈ Bc (X)
Z
σ
e(Λmk ) = 1Λ (x)p(x, M ) σ(dx) < ∞, (1.4)
X
i.e., p(x, M ) ∈ L1loc (σ).
Now, we wish to introduce a marked Poisson measure on ΩM X (cf. e.g. [23, 22]). To
e⊗n on (X × M )n , and for any Λ ∈ Oc (X), σ
this end, we take first the measure σ e⊗n can
n
be considered as a finite measure on Λmk . Since σ is nonatomic, we get
en ) = 0
e⊗n (Λnmk \ Λ
σ mk
e n /Sn , B(Λ
e⊗n as a measure on (Λ
and we can consider σ e n /Sn )) such that
mk mk

σ e n /Sn ) = σ
e⊗n (Λ e(Λmk )n .
mk
(n)
Denote by σ
eΛ,n := σe⊗n ◦ (LΛ )−1 the image measure on ΩM
Λ (n) under the bijection
(1.1). Then, we can define a measure λΛ
σ
e on Ω M by
Λ
X∞
1
λΛ
e :=
σ σ
eΛ,n ,
n=0
n!

eΛ,0 := ε∅ on ΩM
where σ Λ (0) = {∅}. The measure λσ
Λ is finite and λΛ (ΩM ) = eσ
e σ
e Λ
e (Λmk ) .

Hence, the measure

πσeΛ := e−eσ(Λmk ) λΛ
σ
e
is a probability measure on B(ΩM
Λ ). It is not hard to check the consistency property of the
family {πσe | Λ ∈ Oc (X)} and thus to obtain a unique probability measure πσe on B(ΩM
Λ
X)
such that
πσeΛ = p∗Λ πσe , Λ ∈ Oc (X).
This measure πσe will be called a marked Poisson measure with Lévy measure σ e.
For any function ϕ ∈ C0,b (X × M ), it is easy to calculate the Laplace transform of the
measure πσe
Z Z 
hϕ,ωi ϕ(x,m)
ℓπσe (ϕ) := e πσe (dω) = exp (e − 1) σ
e(dx, dm) . (1.5)
ΩM
X X×M

Example 1.1 Let p(x, ·) ≡ εm (·), where m is some fixed point of M and x ∈ X. Then,
σ
e = σ ⊗ εm and πσe = πσ is just the Poisson measure on (ΓX , B(ΓX )) with intensity σ.
Example 1.2 Let p(x, ·) ≡ τ (·), x ∈ X, where τ is a finite measure on (M, B(M )). Now,
σ
e = σ̂ = σ ⊗ τ and πσe coincides with the marked Poisson measure under consideration in
[26] (in the case where M is a manifold). Notice that the choice of σe = σ̂ as a product
measure means a position-independent marking, while the choice of a general σ e of the
form (1.3) leads to a position-depending marking.

7
2 Transformations of the marked Poisson measure
2.1 Group of transformations of the marked configuration space
We are looking for a natural group A of transformations of ΩM
X such that

(i) πσe is A-quasiinvariant;

da∗ πσe
(ii) A is big enough to reconstruct πσe by the Radon–Nikodym density , where a
dπσe
runs through A.

Let us recall that in the work [26] the group Diff 0 (X) was taken as A, just in the same
way as in the case of the usual Poisson measure [7]. Here, Diff 0 (X) stands for the group of
diffeomorphisms of X with compact support, i.e., each ψ ∈ Diff 0 (X) is a diffeomorphism
of X that is equal to the identity outside a compact set (depending on ψ). The group
Diff 0 (X) satisfies (i). However, unlike the case of the Poisson measure, the condition (ii)
is not satisfied, because, for example, in the case where σ e = σ ⊗ τ , there is no information
dψ ∗ πσe
about the measure τ that is contained in , see [26]. Therefore, just as in the case
dπσe
of [24], we need a proper extension of the group Diff 0 (X).
In what follows, we will suppose that M is a homogeneous space of a Lie group G (see
e.g. [10]). Let us recall that this means the existence of a C ∞ mapping θ : G × M → M
satisfying the following conditions:

(i) If e is the unity element of the group G, then

θ(e, m) = m for all m ∈ M ;

(ii) If g1 , g2 ∈ G, then

θ(g1 , θ(g2 , m)) = θ(g1 g2 , m) for all m ∈ M ;

(iii) For arbitrary m1 , m2 ∈ M , there exists g ∈ G such that θ(g, m1 ) = m2 .

For any g ∈ G, we will denote by θg : M → M the mapping given by θg (m) : = θ(g, m);
then θg defines a diffeomorphism of M .
Let us fix an arbitrary point m0 ∈ M and let H be the isotropy group of M :

H := g ∈ G | θg (m0 ) = m0 .

Then, the homogeneous space M can always be identified with the factor space G/H
(endowed with the unique corresponding C ∞ manifold structure), i.e., M = G/H.
Let us consider the group of smooth currents, i.e., all C ∞ mappings X ∋ x 7→ η(x) ∈ G,
which are equal to e outside a compact set (depending on η). A multiplication η1 η2 in
this group is defined as the pointwise multiplication of the mappings η1 and η2 . In the
representation theory this group is denoted by GX , or C0∞ (X; G).

8
 The group Diff 0 (X) acts in GX by automorphisms: for each ψ ∈ Diff 0 (X),
α
GX ∋ η 7→ α(ψ)η := η ◦ ψ −1 ∈ GX .

Thus, we can endow the Cartesian product of Diff 0 (X) and GX with the following multi-
plication: for a1 = (ψ1 , η1 ), a2 = (ψ2 , η2 ) from Diff 0 (X) × GX

a1 a2 = (ψ1 ◦ ψ2 , η1 (η2 ◦ ψ1−1 ))

and obtain a semidirect product

Diff 0 (X) × GX =: A
α

of the groups Diff 0 (X) and GX .

The group A acts in X × M in the following way: for any a = (ψ, η) ∈ A

X × M ∋ (x, m) 7→ a(x, m) = (ψ(x), θ(η(ψ(x)), m)) ∈ X × M . (2.1)

If id denotes the identity diffeomorphism of X and e is the function identically equal to e

on X, then we will just identify ψ with (ψ, e ) and η with (id, η). The action (2.1) of an
arbitrary a = (ψ, η) can be represented as

(x, m) 7→ a(x, m) = ηψ(x, m),

where

ψ(x, m) = (ψ(x), m),

η(x, m) = (x, θ(η(x), m)).

For any a = (ψ, η) ∈ A, denote Ka := Kψ ∪ Kη , where Kψ and Kη are the minimal

closed sets in X outside of which ψ = id and η = e , respectively. Evidently, Ka ∈ Bc (X),

a(Ka )mk = (Ka )mk ,

and a is the identity transformation outside (Ka )mk .

Now, let us recall some known facts concerning quasiinvariant measures on homoge-
neous spaces (see e.g. [45, 44]).

Theorem 2.1 Suppose G is a Lie group and H its subgroup, and let dg, δG and dh, δH
be fixed Haar measures and modular functions on G and H, respectively. Then:

(i) for every measure µ on G/H that is quasiinvariant with respect to the action of G on
G/H, there exists a measurable positive function ξ on G verifying

δH (h)
ξ(gh) = ξ(g), g ∈ G, h ∈ H, (2.2)
δG (h)

9
 and Z Z Z
f (g)ξ(g) dg = µ(d gH) f (gh) dh, f ∈ C0 (G), (2.3)
G G/H H

where C0 (G) denotes the set of continuous functions on G with compact support; for
each g ∈ G the Radon–Nikodym density is given by
dg∗ µ ξ(g−1 ge)
pµg (e
g H) := (e
g H) = , geH ∈ G/H;
dµ ξ(e
g)

(ii) there exists a quasiinvariant measure λ on G/H such that the function

pλ (g, geH) := pλg (e

g H)

is differentiable on G × G/H.

Remark 2.1 We recall that the modular function δG (·) of a Lie group G is defined from
the equality reg∗ dg = δG (e
g ) dg, where dg is the Haar measure on G (i.e., a fixed left-invariant
measure on G) and rg denotes the right translation on G, i.e., e g 7→ rg ge = ge
g.

We fix the measure λ on M = G/H from Theorem 2.1, (ii). As easily seen from
Theorem 2.1 (i), any quasiinvariant measure on M in equivalent to λ.

Remark 2.2 If H = {e}, i.e., M = G, then we can choose λ to be the Haar measure
dg on G. Moreover, if δG (h) = δH (h) for all h ∈ H (and only in this case) there exists
a λ being invariant with respect to the action of G on M . The latter condition holds
automatically if G is unimodular, that is, δG (g) ≡ 1 for all g ∈ G. This, in turn, holds for
all compact and simple Lie groups.

In what follows, we will suppose that the measure σ is equivalent to the Riemannian
volume ν on X: σ(dx) = ρ(x) ν(dx) with ρ > 0 ν-a.s., and that for ν-a.a. x ∈ X p(x, ·) is
equivalent to the measure λ:

p(x, dm) = p(x, m) λ(dm) with p(x, m) > 0 λ-a.a. m ∈ M.

Thus, the measure σ

e can be written in the form

σ
e(dx, dm) = ρ(x)p(x, m) ν(dx) λ(dm).

The condition σ
e(Λmk ) < ∞, Λ ∈ Bc (X), implies that the function

q(x, m) := ρ(x)p(x, m)

satisfies
q 1/2 ∈ L2loc (X; ν) ⊗ L2 (M ; λ). (2.4)
Noting that

a−1 (x, m) = (ψ, η)−1 (x, m) = (ψ −1 (x), θ(η −1 (x), m)),

we easily deduce the following

10
Proposition 2.1 The measure σ e is A-quasiinvariant and for any a = (ψ, η) ∈ A the
Radon–Nikodym density is given by

 σ
e d(a∗ σ
e) q(ψ −1 (x), θ(η −1 (x), m)) λ


 ap (x, m) := (x, m) = p (η(x), m) Jνψ (x),
de
σ q(x, m)

 if (x, m) ∈ {0 < q(x, m) < ∞} ∩ {0 < q(ψ −1 (x), θ(η −1 (x), m)) < ∞},

 σe
pa (x, m) = 1, otherwise,

where Jνψ is the Jacobian determinant of ψ (w.r.t. the Riemannian volume ν).

We give two examples of the above construction, which are important from the point
of view of the marked configuration space analysis. We refer the reader to e.g. [44, 45] for
further examples.

Example 2.1 Let G = R+ be the dilation group (e.g. [15]), i.e., the multiplication in this
group is given by the usual multiplication of numbers. As a homogeneous space M we
take G itself, by identifying the action of the group with the multiplication in it. As a
quasiinvariant measure λ on M we can take the restriction to R+ of the Lebesgue measure
on R.
R
The analysis and geometry on the marked configuration space ΩX+ were studied in our
previous work [24]. Here we only mention that the choice M = R+ leads (via a natural
isomorphism) to the class of compound Poisson measures. In other words, each mark
sx ∈ R+ corresponding to x ∈ X describes the charge of the measure
X
ω = (γ, s) = sx εx ∈ M(X)
x∈X

at the point x (or, in the case where X = R, the value of the jump of the process at x).

Example 2.2 Let G = O(d + 1) be the (d + 1)-dimensional orthogonal group and let
M = S d be the d-dimensional unit sphere in Rd+1 with the natural action of the group
O(d + 1) on S d , see e.g. [13, 44, 45]. As λ we take the surface measure on S d , which is
invariant w.r.t. the action of O(d + 1). From the point of view of statistical mechanics, a
mark sx ∈ S d describes in this example the spin of the particle at the point x.

2.2 A-quasiinvariance of the marked Poisson measure

Any a ∈ A defines by (2.1) a transformation of X × M , and, consequently, a has the
following “lifting” from X × M to ΩM X:

ΩM M
X ∋ ω 7→ a(ω) = a(x, m) | (x, m) ∈ ω ∈ ΩX . (2.5)

(Note that, for a given ω ∈ ΩM M

X , a(ω) indeed belongs to ΩX and coincides with ω for all
but a finite number of points.) The mapping (2.5) is obviously measurable and we can
define the image a∗ πσe as usually. The following proposition is an analog of a corresponding
fact about Poisson measures.

11
Proposition 2.2 For any a ∈ A, we have

a∗ πσe = πa∗ σe .

Proof. The proof is the same as for the usual Poisson measure πσ with intensity σ and
ψ ∈ Diff 0 (X) (e.g., [7]), one has just to calculate the Laplace transform of the measure
a∗ πσe for any f ∈ C0,b (X × M ) and to use the formula (1.5). 

Proposition 2.3 The marked Poisson measure πσe is quasiinvariant w.r.t. the group A,
and for any a ∈ A we have

d(a∗ πσe ) Y
(ω) = pσae (x, m). (2.6)
dπσe
(x,m)∈ω

Proof. The result follows from Skorokhod theorem on absolute continuity of Poisson
measures (see, e.g., [39, 40]). 

Remark 2.3 Notice that only a finite (depending on ω) number of factors in the product
on the right hand side of (2.6) are not equal to one.

3 The differential geometry of marked configuration spaces

3.1 The tangent bundle of ΩM
X

Let us denote by V0 (X) the set of C ∞ vector fields on X (i.e., smooth sections of T (X))
that have compact support. Let g denote the Lie algebra of G and let C0∞ (X; g) stand for
the set of all C ∞ mappings of X into g that have compact support. Then

a := V0 (X) × C0∞ (X; g)

can be thought of as a Lie algebra corresponding to the Lie group A. More precisely, for
any fixed v ∈ V0 (X) and for any x ∈ X, the curve

R ∋ t 7→ ψtv (x) ∈ X

is defined as the solution of the following Cauchy problem


 d ψ v (x) = v(ψ v (x)),
dt t t
(3.1)

ψ0v (x) = x.

Then, the mappings {ψtv , t ∈ R} form a one-parameter subgroup of diffeomorphisms in

Diff 0 (X) (see, e.g., [10]):

1)∀t ∈ R ψtv ∈ Diff 0 (X),

2)∀t1 , t2 ∈ R ψtv1 ◦ ψtv2 = ψtv1 +t2 .

12
Next, for each function u ∈ C0∞ (X; g), x ∈ X, and t ∈ R, we set ηtu (x) := exp(tu(x)),
where g ∋ Y 7→ exp Y ∈ G is the exponential mapping (see, e.g., [45]). Hence, for a fixed
x ∈ X, {ηtu (x), t ∈ R} is a one-parameter subgroup of G and

η0u (x) = e,
d u (3.2)
η (x) = u(x).
dt t t=0

Let us recall a fundamental theorem in the theory of Lie groups.

Theorem 3.1 There exists a neighborhood U of the zero in g and a neighborhood O of

the unit element e in G such that exp : U → O is an analytic diffeomorphism.

From this theorem, we conclude that, for each fixed u ∈ C0∞ (X; g), there exists ε > 0
such that for any t ∈ (−ε, ε) the mapping X ∋ x 7→ ηtu (x) ∈ G belongs to GX , which
yields, in turn, that ηtu ∈ GX for all t ∈ R, and moreover ηtu is a one-parameter subgroup
of GX .
Thus, for an arbitrary (v, u) ∈ a, we can consider the curve {(ψtv , ηtu ), t ∈ R} in A.
Hence, to any ω ∈ ΩM X there corresponds the following curve in ΩX :
M

R ∋ t 7→ (ψtv , ηtu )ω ∈ ΩM
X.

Define now for a function F : ΩM

X → R the directional derivative of F along (v, u) as

d
(∇Ω
(v,u) F )(ω) := F ((ψtv , ηtu )ω) t=0
,
dt
provided the right hand side exists. We will also denote by ∇Ω Ω
v and ∇u the directional
derivatives along (v, 0) and (0, u), respectively.
Absolutely analogously, one defines for a function ϕ : X × M → R the directional
derivative of ϕ along (v, u):

X×M d
(∇(v,u) ϕ)(x, m) = ϕ((ψtv , ηtu )(x, m)) t=0
. (3.3)
dt
Then, for a continuously differentiable function ϕ, we have from (2.1), (3.1), (3.2), and
(3.3)

X×M d
(∇(v,u) ϕ)(x, m) = ϕ((ψtv (x), θ(ηtu (ψtv (x)), m) t=0
dt
d d
= ϕ(ψtv (x), m) t=0 + ϕ(x, θ(ηtu (x), m)) t=0
dt dt
d
+ ϕ(x, θ(η0 (ψtv (x)), m)) t=0
u
dt
= h∇ ϕ(x, m), v(x)iTx (X) + h∇G ϕ(x, θ(e, m)), u(x)ig
X

= h∇X×M ϕ(x, m), (v(x), u(x))iT(x,m) (X×M ) . (3.4)

13
Here, T(x,m) (X × M ) := Tx (X) ∔ g and ∇X×M := (∇X , ∇ e M ), where ∇X denotes the
gradient on X and
∇e M f (m) = ∇G fˆ(e, m),
(3.5)
fˆ(g, m) := f (θ(g, m)), g ∈ G, m ∈ M,
∇G being the gradient on G.

Remark 3.1 Notice that upon (3.5) we have, for a fixed u ∈ g,

e M f (m), uig = h∇G f (θ(e, m)), uig
h∇
d
= f (θ(etu , m)) t=0
dt
= h∇M f (m), (Ru)(m)iTm (M ) , (3.6)

where ∇M denotes the usual gradient on M , and the vector field Ru on M is given by
d
M ∋ m 7→ (Ru)(m) := θ(etu , m) t=0
. (3.7)
dt

Let us introduce a special class of “nice functions” on ΩM

X . Denote by D the set of
∞
all C -functions ϕ on X × M such that the support of ϕ is in Bc (X × M ), and ϕ and
all its ∇X×M derivatives are bounded. Next, let Cb∞ (RN ) stand for the space of all C ∞ -
functions on RN which
together with all their derivatives are bounded. Then, we can
introduce FCb∞ D, ΩM M
X as the set of all functions F : ΩX 7→ R of the form

F (ω) = gF (hϕ1 , ωi, . . . , hϕN , ωi), ω ∈ ΩM

X, (3.8)


where ϕ1 , . . . , ϕN ∈ D and gF ∈ Cb∞ (RN ) (compare with [7]). FCb∞ D, ΩM
X will be called
the set of smooth cylinder functions

on ΩMX.
∞ M
For any F ∈ FCb D, ΩX of the form (3.8) and a given (v, u) ∈ a, we have, just as
in [7],

F ((ψtv , ηtu )ω) = gF (hϕ1 , (ψtv , ηtu )ωi, . . . , hϕN , (ψtv , ηtu )ωi)
= gF (hϕ1 ◦ (ψtv , ηtu ), ωi, . . . , hϕN ◦ (ψtv , ηtu ), ωi),

and therefore
N
X ∂gF
(∇Ω
(v,u) F )(ω) = X×M
(hϕ1 , ωi, . . . , hϕN , ωi)h∇(v,u) ϕj , ωi. (3.9)
∂rj
j=1

In particular, we conclude from (3.9) that

∇Ω Ω Ω
(v,u) = ∇v + ∇u . (3.10)

The expression of ∇Ω
(v,a) on smooth cylinder functions motivates the following defini-
tion.

14


Definition 3.1 The tangent space Tω ΩM M
X to the marked configuration space ΩX at a
point ω = (γ, s) ∈ ΩM
X is defined as the Hilbert space


Tω ΩM 2
X : = L (X → T (X) ∔ g; γ)
= L2 (X → T (X); γ) ⊕ L2 (X → g; γ)
M 
= Tx (X) ⊕ g
x∈γ

with scalar product

Z


hVω1 , Vω2 iTω (ΩM ) = hVω1 (x)Tx (X) , Vω2 (x)Tx (X) iTx (X) + hVω1 (x)g, Vω2 (x)gig γ(dx)
X
X
X

= hVω1 (x)Tx (X) , Vω2 (x)Tx (X) iTx (X) + hVω1 (x)g, Vω2 (x)gig , (3.11)
x∈γ


where Vω1 , Vω2 ∈ Tω ΩM X and Vω (x)Tx (X) and Vω (x)g denote the projection of


Vω (x) ∈ Tx (X)∔g onto Tx (X) and g, respectively. (Notice that the tangent space Tω ΩM
X
depends only on the γ coordinate of ω.) The corresponding tangent bundle is

[

T ΩM X = Tω ΩMX .
ω∈ΩM
X

As usually in Riemannian geometry, having directional derivatives and a Hilbert space

as a tangent space, we can introduce a gradient.

Definition 3.2 We define the intrinsic gradient ∇Ω of a function F : ΩM

X → R as the
mapping

ΩM Ω
X ∋ ω 7→ (∇ F )(ω) ∈ Tω ΩX
M

such that, for any (v, u) ∈ a,

(∇Ω Ω
(v,u) F )(ω) = h(∇ F )(ω), (v, u)iTω (ΩM ) . X

By (3.9) and (3.4) we have, for an arbitrary F ∈ FCb∞ D, ΩM
X of the form (3.8) and
each ω = (γ, s) ∈ ΩM
X,

N
X ∂gF
(∇Ω F )(ω; x) = (hϕ1 , ωi, . . . , hϕN , ωi)∇X×M ϕj (x, sx ), x ∈ γ. (3.12)
∂rj
j=1

3.2 Integration by parts and divergence on the marked Poisson space

Let the marked configuration space ΩM X be equipped with the marked Poisson measure
πσe . We strengthen the condition (2.4) by demanding that

q 1/2 ∈ H01,2 (X × M ). (3.13)

15
Here, H01,2 (X × M ) denotes the local Sobolev space of order 1 constructed with respect
to the gradient ∇X×M in the space L2loc (X; ν) ⊗ L2 (M ; λ), i.e., H01,2 (X × M ) consists of
functions f defined on X × M such that, for any set A ∈ Bc (X × M ), the restriction of
f to A coincides with the restriction to A of some function ϕ from the Sobolev space
H 1,2 (X × M ) constructed as the closure of D with respect to the norm
Z  
2
kϕk1,2 := e M ϕ(x, m)|2g + |ϕ(x, s)|2 ν(dx) λ(dm).
|∇X ϕ(x, m)|2Tx (X) + |∇
X×M

Additionally, we will suppose that, for each Λ ∈ Bc (X),

|∇G pλ (e, ·)|g ∈ L1 (Λmk , σ

e), (3.14)

where, as before,
dg ∗ λ
pλ (g, m) =
(m).
dλ


The set FCb∞ D, ΩM
X is a dense subset in the space

L2 (ΩM M
X , B(ΩX ), πσ
2
e ) =: L (πσ
e ).

For any (v,
u) ∈ a, we have a differential operator in L2 (πσe ) on the domain
FCb∞ D, ΩM
X given by


FCb∞ D, ΩM Ω 2
X ∋ F 7→ ∇(v,u) F ∈ L (πσ
e ).

Our aim now is to compute the adjoint operator ∇(v,u) Ω ∗ in L2 (π ). This corresponds, of
σ
e
course, to the deriving of an integration by parts formula with respect to the measure πσe .
But first we present the corresponding formula on X × M .

Definition 3.3 For any (v, u) ∈ a, the logarithmic derivative of the measure σ
e along
(v, u) is defined as the following function on X × M :
σ
e
β(v,u) := βvσe + βuσe

with  
∇X q(x, m)
βvσe (x, m) = , v(x) + divX v(x),
q(x, m) Tx (X)

divX = divX
ν being the divergence on X w.r.t. ν, and
 eM 
σ
e ∇ q(x, m)
βu (x, m) = , u(x) + h∇G pλ (e, m), −u(x)ig.
q(x, m) g

X×M
Upon (3.13), we conclude that, for each (v, u) ∈ a, the function ∇(v,u) log q is quadrat-
ically integrable with respect to the measure σ e, and therefore, since the support of
X×M
∇(v,u) log q belongs to Bc (X × M ), this function is from L1 (X × M , σe). Thus, in virtue
σ
e
of the condition (3.14), we get the inclusion β(v,u) ∈ L1 (X × M , σ
e).
By using standard arguments, one shows the following

16
Lemma 3.1 (Integration by parts formula on X × M ) For all ϕ1 , ϕ2 ∈ D, we have
Z
X×M
(∇(v,u) ϕ1 )(x, m)ϕ2 (x, m) σe(dx, dm) =
X×M
Z
X×M
=− ϕ1 (x, m)(∇(v,u) ϕ2 )(x, m) σe(dx, dm)
X×M
Z
σ
e
− ϕ1 (x, s)ϕ2 (x, s)β(v,u) (x, m) σ
e(dx, dm).
X×M

Remark 3.2 The function h∇G pλ (e, m), −u(x)ig, which appears in the definition of βuσe
is, for each fixed x ∈ X, the divergence on M with respect to the measure λ of the vector
field Ru(x) on M defined by (3.7), see Remark 3.1. Indeed, for any u ∈ g and for an
arbitrary f from C0∞ (M )—the space of all C ∞ functions on M with compact support, we
have
Z Z
e M
∇u f (m) λ(dm) = h∇M f (m), (Ru)(m)iTm (M ) λ(dm)
M M
Z
d
= f (θ(exp(tu), m)) t=0 λ(dm)
M dt
Z
d
= f (m) pλ (exp(tu), m) t=0 λ(dm)
dt
ZM
= f (m)h∇G pλ (e, m), uig λ(dm).
M

Definition 3.4 For any (v, u) ∈ a, the logarithmic derivative of the marked Poisson
measure πσe along (v, u) is defined as the following function on ΩM
X:
πσe
ΩM σ
e
X ∋ ω 7→ B(v,u) (ω) := hβ(v,u) , ωi. (3.15)

A motivation for this definition is given by the following theorem.

Theorem 3.2 (Integration by parts formula) For all F1 , F2 ∈ FCb∞ D, ΩM X and
each (v, u) ∈ a, we have
Z Z
Ω
(∇(v,u) F1 )(ω)F2 (ω) πσe (dω) = − F1 (ω)(∇Ω
(v,u) F2 )(ω) πσ
e (dω)
ΩM ΩM
X
Z X

πσe
− F1 (ω)F2 (ω)B(v,u) (ω) πσe (dω), (3.16)
ΩM
X

or
πσe
∇Ω ∗ Ω
(v,u) = −∇(v,u) − B(v,u) (ω) (3.17)


as an operator equality on the domain FCb∞ D, ΩM 2
X in L (πσ e ).

Proof. Because of (3.10), the formula (3.17) will be proved if we prove it first for the
operator ∇Ω Ω
v , i.e., when u(x) ≡ 0, and then for the operator ∇u , i.e., when v(x) = 0 ∈

17
Tx (X) for all x ∈ X. We present below only the proof for ∇Ω Ω
u , since the proof for ∇v is
basically the same as that of the integration by parts formula in case of Poisson measures
[7].
By Proposition 2.2, we have for all u ∈ C0∞ (X; g)
Z Z
u u
F1 (ηt (ω))F2 (ω) πσe (dω) = F1 (ω)F2 (η−t (ω)) πηtu ∗ σe (dω).
ΩM
X ΩM
X

Differentiating this equation with respect to t, interchanging d/dt with the integrals and
setting t = 0, the l.h.s. becomes the l.h.s. of (3.16). To see that the r.h.s. then also
coincides with the r.h.s. of (3.16), we note that

d u
F2 (η−t (ω)) t=0
= −(∇Ω
u F2 )(ω),
dt
and by Proposition 2.3
  X
d dπηtu ∗ σe d σe
(ω) = p u (x, m)
dt dπσe t=0 dt ηt t=0
(x,m)∈ω

= −hβuσe , ωi = −Buπσe (ω). 

Definition 3.5 For a vector field

V : ΩM M
X ∋ ω 7→ Vω ∈ Tω (ΩX ),

the divergence divΩ

πσe V is defined via the duality relation
Z Z
Ω
hVω , ∇ F (ω)i
F (ω)(divΩ
M πσ
Tω ΩX e (dω) = − πσe V )(ω) πσ
e (dω)
ΩM
X ΩM
X

for all F ∈ FCb∞ D, ΩM
X , provided it exists (i.e., provided
Z
F 7→ hVω , ∇Ω F (ω)i
e (dω)
M πσ Tω ΩX
ΩM
X

is continuous on L2 (πσe )).

A class of smooth vector fields on ΩM X for which the divergence can be computed in
an explicit form is described in the following proposition.

Proposition 3.1 For any vector field

N
X
Vω (x) = Fj (ω)(vj (x), uj (x)), ω ∈ ΩM
X , x ∈ X,
j=1

18


with Fj ∈ FCb∞ D, ΩM
X , (vj , uj ) ∈ a, j = 1, . . . , N , we have

N
X N
X

πσe
(divΩ
πσe V )(ω) = ∇Ω F
(vj ,uj ) j (ω) + B(vj ,uj )
(ω)Fj (ω)
j=1 j=1
N
X N
X
= h∇Ω Fj (ω), (vj , uj )i
+ σ
e
hβ(v , ωiFj (ω).
Tω ΩM j ,uj )
X
j=1 j=1

Proof. Due to the linearity of ∇Ω , it is sufficient to consider the case N
= 1, i.e., Vω (x) =
F1 (ω)(v(x), u(x)). By Theorem 3.2, we have for all F2 ∈ FCb∞ D, ΩM X
Z Z
− hVω , ∇Ω F2 (ω)i
e (dω) = − F1 (ω)∇Ω
M πσTω ΩX (v,u) F2 (ω) πσ
e (dω)
ΩM ΩM
ZX
Z X

πσe
= ∇Ω
(v,u) F1 (ω)F2 (ω) πσ
e (dω) + F1 (ω)F2 (ω)B(v,u) (ω) πσe (dω),
ΩM
X ΩM
X

which yields
πσe
(divΩ Ω
πσe V )(ω) = ∇(v,u) F1 (ω) + B(v,u) (ω)F1 (ω)

= h∇Ω F1 (ω), (v, u)i
+ hβ σe 

Tω ΩM (v,u) , ωiF1 (ω).
X

Remark
PN 3.3 Extending the definition of B πσe in (3.15) to the class of vector fields V =
j=1 Fj ⊗ (vj , uj ) by

N
X N
X

BVπσe (ω) := σ
e
hβ(vj ,uj )
, ωiFj (ω) + ∇Ω
(vj ,uj ) Fj (ω),
j=1 j=1

we obtain that
divΩ πσe
π σ • = B• .

In particular, if (v, u) ∈ a, it follows, for the “constant” vector field Vω ≡ (v, u) on ΩM

X,
that
X×M
divΩπσe (v, u)(ω) = hdiv σ
e (v, u), ωi,

where divX×M
σ
e
σ
e
(v, u) = β(v,u) is the divergence on X × M of (v, u) w.r.t. σ
e:
Z
h∇X×M ϕ(x, m), (v(x), u(x))iT(x,m) (X×M ) σ
e(dx, dm)
X×M
Z


=− ϕ(x, m) divX×M
σ
e (v, u) (x, m) σe(dx, dm), ϕ ∈ D.
X×M

19
3.3 Integration by parts characterization
In the works [7, 8] it was shown that the mixed Poisson measures are exactly the “volume
elements” corresponding to the differential geometry on the configuration space ΓX . Now,
we wish to prove that an analogous statement holds true in our case of ΩM X for mixed
marked Poisson measures.
We start with a lemma that describes σ e as the unique (up to a constant) measure on
X × M with respect to which the divergence divX×Mσ
e is the dual operator of the gradient
∇X×M .

Lemma 3.2 Let the conditions (3.13) and (3.14) hold. Then, for every Λ ∈ Oc (X) the
measures ze σ , z > 0, are the only positive Radon measures ξ on Λmk such that divX×M σ
e
is the dual operator on L2 (Λmk ; ξ) of ∇X×M when considered with the domains V0 (Λ) ×
∞ (Λ
C0∞ (Λ; g), resp. C0,b mk ) (i.e., the set of all (v, u) ∈ a, resp. ϕ ∈ D with support in Λ,
resp. Λmk ).

Proof. In virtue of the conditions (3.13) and (3.14), the lemma is obtained in complete
analogy with Remark 4.1 (iii) in [8]. Indeed, let q1 (x, m) and q2 (x, m) be two densities
w.r.t. ν ⊗ λ for which the logarithmic derivatives coincide. Then, we get

∇X X
v log q1 (x, m) = ∇v log q2 (x, m), v ∈ V0 (X),
eM
∇ eM
u log q1 (x, m) = ∇u log q2 (x, m), u ∈ C0∞ (Λ; g), ν ⊗ λ-a.s.,

which yields respectively

q1 (x, m) = q2 (x, m)c(m),

q1 (x, m) = q2 (x, m)e
c(x) ν ⊗ λ-a.s.

Therefore, q1 (x, m) = const q2 (x, m) ν ⊗ λ-a.s. 

Let κ be a probability measure on (R+ , B(R+ )). Then, we define a mixed marked
Poisson measure as follows: Z
µκ,eσ = πzeσ κ(dz). (3.18)
R+

Here, π0eσ denotes the Dirac measure on ΩM M

X with mass in ω = {∅}. Let Ml (ΩX ),
M M
l ∈ [1, ∞), denote the set of all probability measures on (ΩX , B(ΩX )) such that
Z
|hf, ωi|l µ(dω) < ∞ for all f ∈ C0,b (X × M ), f ≥ 0.
ΩM
X

Clearly, µκ,eσ ∈ Ml (ΩM

X ) if and only if
Z
z l κ(dz) < ∞. (3.19)
R+

We define (IbP)σe to be the set of all µ ∈ M1 (ΩM σ

e
X ) with the property that ω 7→ hβ(v,u) , ωi
is µ-integrable for all (v, u) ∈ a and which satisfy (3.16) with µ replacing πσe for all

20


F1 , F2 ∈ FCb∞ D, ΩM
X , (v, a) ∈ g. We note that (3.16) makes sense only for such measures
πσe X×M
and that B(v,u) depends only on σe not on πσe . Obviously, since ∇(v,u) obeys the product
rule for all (v, u) ∈ a, we can always take F2 ≡ 1. Furthermore, (IbP)σe is convex.

Theorem 3.3 Let the condition (3.13) and (3.14) be satisfied. Then, the following con-
ditions are equivalent:
(i) µ ∈ (IbP)σe ;
(ii) µ = µκ,eσ for some probability measure κ on (R+ , B(R+ )) satisfying (3.19) with l = 1.

Proof. The part (ii)⇒(i) is trivial. The proof of (i)⇒(ii) goes along absolutely analogously
to that in the particular case where G = M = R+ , see [24]. 
As a direct consequence of Theorem 3.3, we obtain

Corollary 3.1 The extreme points of (IbP)σe are exactly πzeσ , z ≥ 0.

3.4 A lifting of the geometry

Just as in the case of the geometry on the configuration space, we can present an inter-
pretation of the formulas obtained in subsections 3.1–3.3 via a simple “lifting rule.”
Suppose that f ∈ C0,b (X × M ), or more generally f is an arbitrary measurable func-
tion on X × M for which there exists (depending on f ) Λ ∈ Bc (X) such that supp f ⊂ Λmk .
Then, f generates a (cylinder) function on ΩMX by the formula

Lf (ω) := hf, ωi, ω ∈ ΩM

X.

We will call Lf the lifting of f .

As before, any vector field (v, u) ∈ a,

(v, u) : X ∋ x 7→ (v(x), u(x)) ∈ T(x,m) (X × M ) = Tx (X) ∔ g,

can be considered as a vector field on ΩM

X (the lifting of (v, u)), which we denote by L(v,u) :

L(v,u) : ΩM M 2
X ∋ ω = {γ, s} 7→ {x 7→ (v(x), u(x))} ∈ Tω (ΩX ) = L (X → T (X) ∔ g ; γ).

For (v1 , u1 ), (v2 , u2 ) ∈ a, the formula (3.11) can be written as follows:

L(v1 ,u1 ) , L(v2 ,u2 )
=L

Tω ΩM h(v1 ,u1 ),(v2 ,u2 )iT (X×M ) (ω),
X

i.e., the scalar product of lifted vector fields is computed as the lifting of the scalar product

h(v1 (x), u2 (x)), (v2 (x), u2 (x))iT(x,sx ) (X×M ) = f (x).

This rule can be used as a definition of the tangent space Tω ΩM
X .
The formula (3.9) has now the following interpretation:


∇Ω(v,u) Lϕ (ω) = L∇X×M ϕ (ω), ϕ ∈ D, ω ∈ ΩMX, (3.20)
(v,u)

21
and the “lifting rule” for the gradient is given by
(∇Ω Lϕ )(γ, s) : γ ∋ x 7→ ∇X×M ϕ(x, sx ). (3.21)
πσe
As follows from (3.15), the logarithmic derivative B(v,u) : ΩM
X → R is obtained via the
σ
e
lifting procedure of the corresponding logarithmic derivative β(v,u) : X × M → R, namely,
πσe
B(v,u) (ω) = Lβ σe (ω),
(v,u)

or equivalently, one has for the divergence of a lifted vector field:

divΩ
πσe L(v,a) = LdivX×M (v, a). (3.22)
e
σ

We underline that by (3.20) and (3.21) one recovers the action of ∇Ω Ω

(v,a) and ∇ on all


functions from FCb∞ D, ΩM X algebraically from requiring the product or the chain rule
to hold. Also, the action of divΩ
πσe on more general cylindrical vector fields follows as in
Remark 3.3 if one assumes the usual product rule for divπΩ to hold.
e
σ

4 Representations of the Lie algebra a of the group A

Using the A-quasiinvariance of πσe , we can define the unitary representation of the group
A = Diff 0 (X)× GX in the space L2 (πσe ). Namely, for a ∈ A, we define the unitary operator
α
s

da−1∗ πσe
Vπσe (a)F (ω) := F (a(ω)) (ω), F ∈ L2 (πσe ).
dπσe
Then, we have
Vπσe (a1 )Vπσe (a2 ) = Vπσe (a1 a2 ), a1 , a2 ∈ A.
As has been noted in Introduction, this representation is reducible, cf. [24]
As in subsec. 3.1, to any vector field v ∈ V0 (X) there corresponds a one-parameter
subgroup of diffeomorphisms ψtv , t ∈ R. It generates a one-parameter unitary group
Vπσe (ψtv ) := exp[itJπσe (v)], t ∈ R,
where Jπσe (v) denotes the selfadjoint generator of this group. Analogously, to a subgroup
ηtu , u ∈ C0∞ (X; g), there corresponds a one-parameter unitary group
Vπσe (ηtu ) := exp[itIπσe (u)]
with a generator Iπσe (u).

Proposition 4.1 For any v
∈ V0 (X) and u ∈ C0∞ (X; g), the following operator equalities
on the domain FCb∞ D, ΩM
X hold:

1 Ω 1 πσe
Jπσe (v) = ∇ + B ,
i v 2i v
1 1 πσe
Iπσe (u) = ∇Ω + B .
i u 2i u

22
Proof. These equalities follow immediately from the definition of the directional derivatives
∇Ω Ω v u
v and ∇a , Theorem 3.2, and the form of the operators Vπσe (ψt ) and Vπσe (θt ). 

For any (v, u) ∈ a, define an operator

Rπσe (v, u) := Jπσe (v) + Iπσe (u).
By Proposition 4.1,
1 1
Rπσe (v, u) = ∇Ω + B πσe .
i (v,u) 2i (v,u)
We wish to derive now a commutation relation between these operators.
Lemma 4.1 The Lie-bracket [(v1 , u1 ), (v2 , u2 )] of the vector fields (v1 , u1 ), (v2 , u2 ) ∈ a,
i.e., a vector field from a such that
X×M X×M X×M X×M X×M
∇[(v1 ,u1 ),(v2 ,u2 )]
= ∇(v1 ,u1 )
∇(v2 ,u2 )
− ∇(v2 ,u2 )
∇(v1 ,u1 )
on D,
is given by
[(v1 , u1 ), (v2 , u2 )] = ([v1 , v2 ], ∇X X
v1 u2 − ∇v2 u1 + [u1 , u2 ]),
where [v1 , v2 ] is the Lie-bracket of the vector fields v1 , v2 on X,
[u1 , u2 ](x) = [u1 (x), u2 (x)]
(the latter being the Lie-bracket on g of u1 (x), u2 (x) ∈ g), and ∇X
v u is the derivative in
direction v of a g-valued function u on X.
Proof. First, we have on D:
∇X X X X X
v1 ∇v2 − ∇v2 ∇v1 = ∇[v1 ,v2 ] , v1 , v2 ∈ V0 (X). (4.1)
Next, using (3.5),
Gˆ
∇eM
u f (x, m) = h∇ f (x, e, m), u(x)ig, fˆ(x, g, m) := f (x, θ(g, m)),
and so
eM∇
(∇ eM eM eM
u1 u2 − ∇u2 ∇u1 )f (x, m)
= h∇G fˆ(x, e, m), [u1 (x), u2 (x)]ig
eM
=∇ u1 , u2 ∈ C0∞ (X; g).
[u1 ,u2 ] f (x, m), (4.2)
Finally,
(∇X eM eM X
v ∇u − ∇u ∇v )f (x, m)
= h∇X h∇G fˆ(x, e, m), u(x)ig, v(x)iT x (X)

− h∇ h∇ fˆ(x, e, m), v(x)iTx (X) , u(x)ig

G X

= h∇X ∇G fˆ(x, e, m), v(x) ⊗ u(x)iT (X)⊗g + h∇G fˆ(x, e, m), ∇X u(x)ig
x v

− h∇ ∇ fˆ(x, e, m), u(x) ⊗ v(x)ig⊗Tx (X)

G X

= h∇G fˆ(x, e, m), ∇X u(x)ig = ∇

v
e MX f (x, m),
∇v u
∞
v ∈ V0 (X), u ∈ C0 (X; g). (4.3)
The equalities (4.1)–(4.3) yield the lemma. 

23
Proposition 4.2

For arbitrary (v1 , u1 ), (v2 , u2 ) ∈ a, the following operator equality holds
∞ M
on FCb D, ΩX :
[Rπσe (v1 , u1 ), Rπσe (v2 , u2 )] = Rπσe ([(v1 , u1 ), (v2 , u2 )]).
In particular,
[Jπσe (v1 ), Jπσe (v2 )] = −iJπσe ([v1 , v2 ]), v1 , v2 ∈ V0 (X),
[Iπσe (u1 ), Iπσe (u2 )] = −Iπσe ([u1 , u2 ]), u1 , u2 ∈ C0∞ (X; g),
[Jπσe (v), Iπσe (u)] = −iIπσe (∇X
v u), v ∈ V0 (X), u ∈ C0∞ (X; g).

Proof. First we note that Lemma 4.1 and (3.9) immediately imply


∇Ω Ω Ω Ω Ω
(v1 ,u1 ) ∇(v2 ,u2 ) − ∇(v2 ,u2 ) ∇(v1 ,u1 ) = ∇[(v1 ,u1 ),(v2 ,u2 )] on FCb∞ D, ΩM
X .

Therefore, by using the chain rule, we conclude that the lemma will be proved if we show
that
πσe πσe πσe
∇Ω Ω
(v1 ,u1 ) B(v2 ,u2 ) − ∇(v2 ,u2 ) B(v1 ,u1 ) = B[(v1 ,u1 ),(v2 ,u2 )] πσ
e -a.e. (4.4)
But upon the representation
πσe eM
B(v,u) (ω) = h∇X X G λ
v log q + ∇u log q + div v + h∇ p (e, m), −u(x)ig, ωi

and Remark 3.2, we easily derive (4.4) again from Lemma 4.1. 
Thus, the operators Rπσe (v, u), (v, u) ∈ a, give a marked Poisson space representation
of the Lie algebra a of the group A.

5 Intrinsic Dirichlet forms on marked Poisson spaces

5.1 Definition of the intrinsic Dirichlet form
From now on, the underlying space of “nice functions” on X × M will be instead of D
the space D0 := C0∞ (X × M ) consisting of all C ∞ functions with compact support in
X × M . Evidently, D0 is a subset of D and in
the case where M is itself compact D
0 =
D. Absolutely analogously to FC ∞ D, ΩM one constructs the set FC ∞ D , ΩM (⊂

b X b 0 X
FCb∞ D, ΩM X ), which is dense in L 2 (π ). By FP(D , ΩM ) we denote the set of all cylin-
σ
e 0 X
der functions of the form (3.8) in which the functions ϕ1 , . . . , ϕN belong to D0 and the
generating function gF is a polynomial on RN , i.e., gF ∈ P(RN ). Finally, in the same way
we introduce FCp∞ (D0 , ΩM ∞ N ∞
X ) where gF ∈ Cp (R ) (:=the set of all C -functions f on R
N

such that f and its partial derivatives of any order are polynomially bounded).
We have obviously


FCb∞ D0 , ΩM ∞ M
X ⊂ FCp (D0 , ΩX ),
FP(D0 , ΩM ∞ M
X ) ⊂ FCp (D0 , ΩX ),

and these are algebras with respect to the usual operations. The existence of the Laplace
transform ℓπσe (f ) for each f ∈ C0 (X × M ) implies, in particular, that FCp∞ (D0 , ΩM
X) ⊂
2
L (πσe ).

24
Definition 5.1 For F1 , F2 ∈ FCp∞ (D0 , ΩM
X ), we introduce a pre-Dirichlet form as
Z
Ω
Eπσe (F1 , F2 ) = h∇Ω F1 (ω), ∇Ω F2 (ω)iTω (ΩM ) πσe (dω). (5.1)
X
ΩM
X

Note that, for all F ∈ FCp∞ (D0 , ΩM X ), the formula (3.12) is still valid and therefore,
for F1 = gF1 (hϕ1 , ·i, . . . , hϕN , ·i) and F2 = gF2 (hξ1 , ·i, . . . , hξK , ·i) from FCp∞ (D0 , ΩM
X ), we
have

h∇Ω F1 (ω), ∇Ω F2 (ω)iTω (ΩM ) =

X
N X
X K
∂gF 1 ∂gF2
= (hϕ1 , ωi, . . . , hϕN , ωi) (hξ1 , ωi, . . . , hξK , ωi)×
∂rj ∂rk
j=1 k=1
Z
× h∇X×M ϕj (x, sx ), ∇X×M ξk (x, sx )iT(x,sx ) (X×M ) γ(dx)
X
N X
X K
∂gF 1 ∂gF2
= (hϕ1 , ωi, . . . , hϕN , ωi) (hξ1 , ωi, . . . , hξK , ωi)×
∂rj ∂rk
j=1 k=1

×hh∇X×M ϕj , ∇X×M ξk iT (X×M ) , ωi. (5.2)

Since for ϕ, ξ ∈ D0 , the function

h∇X×M ϕ(x, m), ∇X×M ξ(x, m)iT(x,m) (X×M ) =

e M ϕ(x, m), ∇
= h∇X ϕ(x, m), ∇X ξ(x, m)iTx (X) + h∇ e M ξ(x, m)ig

belongs to D0 , we conclude that

h∇Ω F1 (·), ∇Ω (·)F2 (·)iT (ΩM ) ∈ L1 (πσe ), F1 , F2 ∈ FCp∞ (D0 , ΩM

X ),
X

and so (5.1) is well defined.

We will call EπΩσe the intrinsic pre-Dirichlet form corresponding to the marked Poisson
measure πσe on ΩM Ω
X . In the next subsection we will prove the closability of Eπσe .

5.2 Intrinsic Dirichlet operators

We start with introducing the pre-Dirichlet operator corresponding to the measure σ
e on
X × M and to the gradient ∇X×M :
Z
X×M
Eσe (ϕ, ξ) := h∇X×M ϕ(x, m), ∇X×M ξ(x, m)iT(x,m) (X×M ) σe(dx, dm), (5.3)
X×M

where ϕ, ξ ∈ D0 . This form is associated with the Dirichlet operator

HσeX×M := HσeX + HσeM (5.4)

on D0 which satisfies

EσeX×M (ϕ, ξ) = (HσeX×M ϕ, ξ)L2 (eσ) , ϕ, ξ ∈ D0 . (5.5)

25
 e M , respectively. Evidently,
Here, HσeX and HσeM are the Dirichlet operators of ∇X and ∇

HσeX ϕ(x, m) = −∆X ϕ(x, m) − h∇X log q(x, m), ∇X ϕ(x, m)iTx (X) , (5.6)

where ∆X denotes the Laplace-Beltrami operator corresponding to ∇X .

Let us calculate the operator HσeM . Suppose f ∈ D0 and W ∈ C0 (X × M ; g). Analo-
gously to Remark 3.1, we conclude
e M f (x, m), W (x, m)ig = h∇M f (x, m), (RW )(x, m)iT (M ) ,
h∇ (5.7)
m

where RW ∈ C0∞ (X × M ; T M ) is given by

d
X × M ∋ (x, m) 7→ (RW )(x, m) := θ(exp(tW (x, m)), m) t=0
∈ Tm M. (5.8)
dt
Therefore, using the integration by parts formula on M for a vector field with a compact
support, we get
Z
h∇e M f (x, m), W (x, m)ig σ
e(dx, dm)
X×M
Z

=− f (x, m) divM (RW )(x, m)
X×M

+ h∇M log q(x, m), (RW )(x, m)iTm (M ) σ e(dx, dm)
Z
 
=− f (x, m) divM (RW )(x, m) + h∇ e M log q(x, m), W (x, m)ig σe(dx, dm),
X×M

where divM is the divergence on M with respect to the usual gradient ∇M and the measure
f M on X × M w.r.t. the gradient ∇
λ. Thus, the divergence div e M and the measure σ
e is
σ
e
given by
f M W (x, m) = divM (RW )(x, m) + h∇
div e M log q(x, m), W (x, m)ig.
σ
e

f M w.r.t. the measure ν(dx) λ(dm) equals

In particular, the divergence div

f M W (x, m) = divM (RW )(x, m).

div (5.9)

e M f ∈ C ∞ (X × M ; g), and so we have finally

It is easy to see that, for f ∈ D0 , W = ∇ 0

f M∇
HσeM f = div e M f = −∆
e M f − h∇
e M log q, ∇
e M f ig, f ∈ D0 , (5.10)

where
∆ f M∇
e M f = div e M f := divM (R(∇
e M f )). (5.11)
The closure of the form EσeX×M on

L2 (X × M ; σ
e) =: L2 (e
σ)

26
is denoted by (EσeX×M , D(EσeX×M )). This form generates a positive selfadjoint operator in
σ ) (the so-called Friedrichs extension of HσeX×M , see e.g. [9]). For this extension we
L2 (e
preserve the notation HσeX×M and denote the domain by D(HσeX×M ).

Let us introduce a differential operator HπΩσe on the domain FCb∞ D0 , ΩM X which is
∞ M


given on any F ∈ FCb D0 , ΩX of the form (3.8) by the formula
N
X ∂2F
(HπΩσe F )(ω) :=− (hϕ1 , ωi, . . . , hϕn , ωi)hh∇X×M ϕj , ∇X×M ϕk iT (X×M ) , ωi
∂rj ∂rk
j,k=1
N
X ∂F
+ (hϕ1 , ωi, . . . , hϕn , ωi)hHσeX×M ϕj , ωi. (5.12)
∂rj
j=1

Since
h∇X×M log q, ∇X×M ϕj iT (X×M ) ∈ L2 (e
σ ) ∩ L1 (e
σ)
(see condition (3.13)), the r.h.s. of (5.12) is well defined as an element of L2 (πσe ). The
following theorem implies, in particular, that HπΩσe is well defined as a linear operator on


FCb∞ D0 , ΩMX , i.e., independently of the representation of F as in (3.8).

Theorem 5.1 The operator HπΩσe is associated with the intrinsic Dirichlet form EπΩσe in the


sense that, for all F1 , F2 ∈ FCb∞ D0 , ΩM
X

EπΩσe (F1 , F2 ) = (HπΩσe F1 , F2 )L2 (πσe ) , (5.13)

or

HπΩσe = − divΩ
πσe ∇
Ω
on FCb∞ D0 , ΩM
X .

We call HπΩσe the intrinsic Dirichlet operator of the measure πσe .

Lemma 5.1 For any ϕ ∈ D0 and W ∈ C0∞ (X × M ; g), we have

f M (ϕW )(x, m) = h∇
div f M W (x, m).
e M ϕ(x, m), W (x, m)ig + ϕ(x, m)div

Proof. By (5.7), (5.8), and (5.9)

h i
f M (ϕW )(x, m) = divM d θ(exp(tϕ(x, m)W (x, m)), m)
div
dt t=0
h d i
= divM ϕ(x, m) θ(exp(tW (x, m)), m) t=0
dt
d
= h∇M ϕ(x, m), θ(exp(tW (x, m)), m) t=0 iTm (M )
dt
hd i
+ ϕ(x, m) div M θ(exp(tW (x, m)), m) t=0
dt
d f M W (x, m)
= ϕ(x, θ(exp(tW (x, m)), m)) t=0 + ϕ(x, m)div
dt
= h∇ f M W (x, m). 
e M ϕ(x, m), W (x, m)ig + ϕ(x, m)div

27
Proof of Theorem 5.1. For shortness

of notations we will prove the formula (5.13) in the
case where F1 , F2 ∈ FCb∞ D0 , ΩM
X are of the form

F1 = gF1 (hϕ, ωi), F2 = gF2 (hξ, ωi).

However, it is a trivial step to generalize the proof to general F1 , F2 .

Let Λ ∈ Oc (X) be chosen so that the supports of the functions ϕ and ξ are in Λmk .
Then, by (5.1), (5.2), and the construction of the marked Poisson measure
Z
EπΩσe (F1 , F2 ) = gF′ 1 (hϕ, ωi)gF′ 2 (hξ, ωi)hh∇X×M ϕ, ∇X×M ξiT (X×M ) , ωi πσe (dω)
ΩM
X
X∞ Z
σ
e (Λmk ) 1
= −e gF′ 1 (ϕ(x1 , m1 ) + · · · + ϕ(xn , mn ))
n=1
n! n
Λmk

×gF′ 2 (ξ(x1 , m1 ) + · · · + ξ(xn , mn ))

X
n 
× h∇X×M ϕ(xi , mi ), ∇X×M ξ(xi , mi )iT(x ,m ) (X×M ) σ e(dx1 , dm1 ) · · · σ
e(dx1 , dm1 )
i i
i=1
∞
X Z n
X
−e
σ(Λmk ) 1
=e h∇iX×M gF1 (ϕ(x1 , m1 ) + · · · + ϕ(xn , mn )),
n! Λnmk
n=1 i=1
X×M
∇i gF2 (ξ(x1 , m1 ) + · · · + ξ(xn , mn ))iT(x (X×M ) σ
e(dx1 , dm1 ) · · · σ
e(dxn , dmn ),
i ,mi )

where ∇iX×M denotes the ∇X×M gradient in the (xi , mi ) variables. Therefore, by using
(5.10) and Lemma 5.1, we proceed in the calculation of EπΩσe (F1 , F2 ) as follows:

X∞ Z X
n 
−e
σ (Λmk ) 1 (X×M )i
=e Hσe gF1 (ϕ(x1 , m1 ) + · · · + ϕ(xn , mn )) ×
n=1
n! Λnmk
i=1
×gF1 (ξ(x1 , m1 ) + · · · + ξ(xn , mn )) σe(x1 , m1 ) · · · σ
e(dxn , dmn )
∞
X Z  Xn
1
= −eσe (Λmk ) gF′′ 1 (ϕ(x1 , m1 ) + · · · + ϕ(xn , mn ))×
n! Λnmk
n=1 i=1
X×M
×h∇ ϕ(xi , mi ), ∇X×M ϕ(xi , mi )iT(x (X×M )
i ,mi )

+ gF′ 1 (ϕ(x1 , m1 ) + · · · + ϕ(xn , mn ))HσeX×M ϕ(xi , mi ) ×

×gF2 (ξ(x1 , m1 ) + · · · + ξ(xn , mn )) σ

e(dx1 , dm1 ) · · · σ
e(dxn , dmn )
Z
= HπΩσe F1 (ω)F2 (ω) πσe (dω). 
ΩM
X

Remark 5.1 The operator HπΩσe can be naturally extended to cylinder functions of the
form
F (ω) := ehϕ,ωi , ϕ ∈ D0 , ω ∈ ΩM
X,

28
since such F belong to L2 (πσe ). We then have
HπΩσe ehϕ,ωi = hHσeX×M ϕ − |∇X×M ϕ|2T (X×M ) , ωi ehϕ,ωi . (5.14)

As an immediate consequence of Theorem 5.1 we obtain

Corollary 5.1 (EπΩσe , FCb∞ D0 , ΩM 2
X ) is closable on L (πσ
Ω Ω
e ). Its closure (Eπσe , D(Eπσe )) is
associated with a positive definite selfadjoint operator, the Friedrichs extension of HπΩσe ,
which we also denote by HπΩσe (and its domain by D(HπΩσe )).

Clearly, ∇Ω also extends to D(EπΩσe ). We denote this extension by ∇Ω .

Corollary 5.2 Let

F (ω) := gF (hϕ1 , ωi, . . . , hϕN , ωi), ω ∈ ΩM
X,
(5.15)
ϕ1 , . . . , ϕN ∈ D(EσeX×M ), gF ∈ Cb∞ (RN ).

Then F ∈ D(EπΩσe ) and

N
X ∂gF
(∇Ω F )(ω; x) = (hϕ1 , ωi, . . . , hϕN , ωi)∇X×M ϕj (x, sx ).
∂rj
j=1

Proof. By approximation this is an immediate consequence of (3.12) and the fact that, for
all 1 ≤ i ≤ N , Z
h|∇X×M ϕi |2T (X×M ) , ωi πσe (dω) = EσeX×M (ϕi , ϕi ). (5.16)

Remark 5.2 Let µν,eσ ∈ M2 (ΩM X ) be given as in (3.18). Then, by Theorem 3.2, (ii)⇒(i),
all results above are valid with µν,eσ replacing πσe . By (5.12) we have


HπΩσe = HµΩν,eσ on FCb∞ D0 , ΩM X .

We note that the r.h.s. of (5.12) only depends on σ e and the Riemannian structure of
X × M . The respective Friedrichs extension on L2 (µν,eσ ) is again denoted by HµΩν,eσ , how-
ever it does necessarily not coincide with HπΩσe .

5.3 The heat semigroup and ergodicity

The results of this subsection are obtained absolutely analogously to the corresponding
results of the paper [7], so we omit the proofs.
For µκ,eσ ∈ M2 (ΩM Ω Ω
X ) let Tµκ ,eσ (t) := exp(−tHµκ ,eσ ), t > 0. Define

E(D1 , ΩMX ) = l. h. exp(hlog(1 + ϕ), ·i) | ϕ ∈ D1 ,

where l.h. means the linear hull and


D1 := ϕ ∈ D(HσeX×M ) ∩ L1 (e
σ ) | HσeX×M ϕ ∈ L1 (e
σ)
and − δ ≤ ϕ ≤ 0 for some δ ∈ (0, 1) .

29
Proposition 5.1 Let µκ,eσ be as in (3.18). Assume that HσeX×M is conservative, i.e.,
Z
(HσeX×M ϕ)(x, m) σ
e(dx, dm) = 0
X×M

for all ϕ ∈ D(HσeX×M ) ∩ L1 (e σ ) such that HσeX×M ϕ ∈ L1 (e

σ ), and suppose
that (HσeX×M , D0 ) is essentially selfadjoint on L2 (e
σ ). Then
X×M
TµΩκ ,eσ (t) exp(hlog(1 + ϕ), ·i) = exp(hlog(1 + e−tHσe ϕ), ·i), ϕ ∈ D1 , (5.17)

E(D1 , ΩM Ω
X ) ⊂ D(Hµκ ,eσ ), and

HµΩκ ,eσ exp(hlog(1 + ϕ), ·i)

= h(1 + ϕ)−1 HσeX×M ϕ, ·i exp(hlog(1 + ϕ), ·i), ϕ ∈ D1 .

Remark 5.3 (i) The condition of essential selfadjointness of HσeX×M on D0 is fulfilled if

X is complete and |β σe |T (X×M ) ∈ Lploc (X × M ; m ⊗ λ) for some p ≥ dim(X) + 1.
(ii) Since (exp(−tHσeX×M ))t>0 is sub-Markovian (i.e., 0 ≤ exp(−tHσeX×M )ϕ ≤ 1 for all
t > 0 and ϕ ∈ L2 (eσ ), 0 ≤ ϕ ≤ 1), because (EσeX×M , D(EσeX×M )) is a Dirichlet form, by a
simple approximation argument Proposition 5.1 implies that the equality (5.17) holds for
t > 0 and all ϕ ∈ L1 (e
σ ), −1 < ϕ ≤ 0.

Theorem 5.2 Let the conditions of Proposition 5.1 hold. Then E(D1 , ΩM X ) is an operator
Ω
core for the Friedrichs extension Hµκ ,eσ on L (µκ,eσ ). 2 (In other words:
(HµΩκ ,eσ , E(D1 , ΩM
X )) is essentially selfadjoint on L 2 (µ
σ
κ,e ).)

Theorem 5.3 Suppose that the conditions of Theorem 3.3 and Proposition 5.1 hold. Then
the following assertions are equivalent:
(i) µκ,eσ = πzeσ for some z > 0.
(ii) (EµΩκ ,eσ , D(EµΩκ ,eσ )) is irreducible (i.e., for F ∈ D(EµΩκ ,eσ ), EµΩκ ,eσ (F, F ) = 0 implies that
F = const).
(iii) (TµΩκ ,eσ (t))t>0 is irreducible (i.e., if G ∈ L2 (µκ,eσ ) such that TµΩκ ,eσ (t)(GF ) = GTµΩκ ,eσ (t)F
for all F ∈ L∞ (µκ,eσ ), t > 0, then G = const).
(iv) If F ∈ L2 (µκ,eσ ) such that TµΩκ ,eσ (t)F = F for all T > 0, then F = const .
(v) TµΩκ ,eσ (t) 6≡ 1 and ergodic (i.e.,
Z  Z 2
TµΩκ ,eσ (t)F − F dµκ,eσ dµκ,eσ → 0 as t → 0

for all F ∈ L2 (µκ,eσ )).

(vi) If F ∈ D(HµΩκ ,eσ ) with HµΩκ ,eσ = 0, then F = const .

30
Remark 5.4 Let us consider the diffusion process P on X × M associated to the Dirichlet
form (EσeX×M , D(EσeX×M )). This process can be interpreted as distorted Brownian motion
on the manifold X × M . More precisely, the diffusion of points x ∈ X is associated to
the Dirichlet form of the measure σ, so that it is distorted Brownian motion on X, and
the diffusion of marks sx , x ∈ X, is associated to the ∇ e M -Dirichlet form of the measure
p(x, dm) on M .
The existence of a diffusion process P corresponding to the Dirichlet form
(EµΩκ ,eσ , D(EµΩκ ,eσ )) follows from [31], and its identification with the independent infinite
particle process (on X × M ) may be proved by the same arguments as in [7]. By analogy
with the case of the process P on X × M , one can call P distorted Brownian motion on
ΩMX.

6 Intrinsic Dirichlet operator and second quantization

In this section, we want to describe the Fock space realization of the marked Poisson
spaces and show that HπΩσe is the second quantization of the operator HσeX×M .

6.1 Marked Poisson gradient and chaos decomposition

Let us define another “gradient” on functions F : ΩM
X → R, which has specific useful
properties on the marked Poisson space.

Definition 6.1 For any F ∈ FCp∞ (D0 , ΩM

X ) we define the marked Poisson gradient ∇
MP

as

(∇MP F )(ω, (x, m)) := F (ω + ε(x,m) ) − F (ω), ω ∈ ΩM

X , (x, m) ∈ X × M .

Let us mention that the operation

ΩM M
X ∋ ω 7→ ω + ε(x,m) ∈ ΩX

is a πσe -a.e. well-defined map because of the property

πσe {ω = (γ, s) ∈ ΩM
X | x ∈ γ} = 0

for an arbitrary x ∈ X (which easily follows from the construction of πσe ). We consider
∇MP as a mapping

∇MP : FCp∞ (D0 , ΩM

X ) ∋ F 7→ ∇
MP
F ∈ L2 (e
σ ) ⊗ L2 (πσe )

that corresponds to using the Hilbert space L2 (e

σ ) as a tangent space at any point ω ∈ ΩM
X.
Thus, for any ϕ ∈ D0 , we can introduce the directional derivative

(∇MP
ϕ F )(ω) = h∇
MP
F (ω), ϕiL2 (eσ)
Z
= (F (ω + ε(x,m) ) − F (ω))ϕ(x, m) σ
e(dx, dm).
X×M

31
 The most important feature of the marked Poisson gradient is that it produces (via
a corresponding “integration by parts formula”) the orthogonal system of Charlier poly-
nomials on (ΩM M
X , B(ΩX ), πσ
e ). Below, we describe this construction in detail using the
2
isomorphism between L (πσe ) and the symmetric Fock space (see [21, 25, 30])
Let F(L2 (e
σ )) denote the symmetric Fock space over L2 (e
σ ):
∞
M
F(L2 (e
σ )) := Fn (L2 (e
σ ))n!,
n=0

where
b
Fn (L2 (e σ ))⊗n = L̂2 ((X × M )n , σ
σ )) := (L2 (e e⊗n ), n ∈ N,
2
F0 (L (e
σ )) := R,

b denoting the symmetric tensor product. Thus, for each F = (f (n) )∞

⊗ 2 σ ))
n=0 ∈ F(L (e
∞
X
kF k2F (L2 (eσ)) = |f (n) |L̂2 (eσ⊗n ) n!.
n=0

By Ffin (D0 ) we denote the dense subset of F(L2 (e σ )) consisting of finite sequences
b
(n) N
(f )n=0 , n ∈ Z+ , such that each f (n) belongs to Fn (D0 ) := a.D⊗n
0 , the n-th symmetric
algebraic tensor power of D0 :
b
a.D⊗n b b
0 := l. h.{ϕ1 ⊗ · · · ⊗ϕn | ϕi ∈ D0 }.

In virtue of the polarization identity, the latter set is spanned just by the vectors of the
form ϕ⊗n with ϕ ∈ D0 .
Now, we define a linear mapping
N
X
Ffin (D0 ) ∋ F = (f (n) )N
n=0 7→ IF = (IF )(ω) = Qn (f (n) ; ω) ∈ FP(D0 , ΩM
X) (6.1)
n=0

by using the following recursion relation:

Qn+1 (ϕ⊗(n+1) ; ω) = Qn (ϕ⊗n ; ω)(hω, ϕi − hϕiσe )

− nQn (ϕ⊗(n−1) ⊗(ϕ
b 2 ), ω) − nQn−1 (ϕ⊗(n−1) ; ω)hϕ2 iσe ,
Q0 (1, ω) = 1, ϕ ∈ D0 . (6.2)
R
Here, we have set hϕiσe := ϕ de σ . Notice that, since D0 is an algebra under pointwise
multiplication of functions, the latter definition is correct.
It is not hard to see that the mapping (6.1) is one-to-one. Moreover, the following
proposition holds:

Proposition 6.1 The mapping (6.1) can be extended by continuity to a unitary isomor-
phism between the spaces F(L2 (e
σ )) and L2 (πσe ).

32
 For each ϕ ∈ D0 , let us define the creation and annihilation operators in F(L2 (e
σ )) by

a+ (ϕ)ψ ⊗n = ϕ⊗ψ
b ⊗n , a− (ϕ)ψ ⊗n = n(ϕ, ψ)L2 (eσ) ψ ⊗(n−1) , ψ ∈ D0 .

We will denote by the same letters the images of these operators under the unitary I.

Proposition 6.2 We have, for each ϕ ∈ D0 ,

a− (ϕ) = ∇MP
ϕ , a+ (ϕ) = ∇MP∗
ϕ .

In particular,
b n ; ω) = (∇MP∗
b · · · ⊗ϕ
Qn (ϕ1 ⊗ MP∗
ϕ1 · · · ∇ϕn 1)(ω), ω ∈ ΩM
X.

Finally, for each ϕ ∈ D0 we introduce the Poisson exponential

∞
X 1
e(ϕ; ·) := Qn (ϕ⊗n ; ·) = I(Exp ϕ),
n!
n=0

where  ∞
1 ⊗n
Exp ϕ = ϕ .
n! n=0
Then, one can show that, for ϕ > −1,
 
e(ϕ; ω) = exp hlog(1 + ϕ), ωi − hϕiσe , ω ∈ ΩM
X. (6.3)

6.2 Second quantization on the marked Poisson space

Let B be a contraction on L2 (e
σ ), i.e., B ∈ L(L2 (e
σ ), L2 (e
σ )), kBk ≤ 1. Then, we can define
the operator Exp B as the contraction on F(L (e 2 σ )) given by

Exp B ↾ Fn (L2 (e
σ )) := B ⊗ · · · ⊗ B (n times), n ∈ N,
2
Exp B ↾ F0 (L (e
σ )) := 1.

For any selfadjoint positive operator A in L2 (e σ ), we have a contraction semigroup

e−tA ,t ≥ 0, and it is possible to introduce a positive selfadjoint operator d Exp A as the
generator of the semigroup Exp(e−tA ), t ≥ 0:

Exp(e−tA ) = exp(−td Exp A). (6.4)

MP the
The operator d Exp A is called the second quantization of A. We denote by HA
image of the operator d Exp A in the marked Poisson space L2 (πσe ).

Theorem 6.1 Let D0 ⊂ Dom A. Then, the symmetric bilinear form corresponding to the
MP has the following representation:
operator HA
Z
MP
(HA F1 , F2 )L2 (πσe ) = (∇MP F1 , A∇MP F2 )L2 (eσ) πσe (dω) (6.5)
ΩM
X

for all F1 , F2 ∈ FP(D0 , ΩM

X ).

33
Remark 6.1 The bilinear form (6.5) uses the marked Poisson gradient ∇MP and a coef-
ficient operator A > 0. We will call
Z
MP
Eπσe ,A (F1 , F2 ) = (∇MP F, A∇MP G)L2 (eσ) πσe (dω)
ΩM
X

the marked Poisson pre-Dirichlet form with coefficient A.

Proof of Theorem 5.1. The proof is analogous to that of Theorem 5.1 in [7]. Using again
the fact that D0 is an algebra under pointwise multiplication, one easily concludes that,
for any F ∈ FP(D0 , ΩM M
X ) and any ω ∈ ΩX , the gradient ∇
MP F (ω, (x, m)) is a function

in D0 and hence
(∇MP F, A∇MP G)L2 (eσ) ∈ FP(D0 , ΩMX ),

so that the form (6.5) is well-defined. Then, one verifies the formula (6.5) by using Propo-
sitions 5.1, 5.2 and the explicit formula for d Exp A on Fn (D0 ):

d Exp A ϕ⊗n = n(Aϕ)⊗ϕ

b ⊗(n−1) , ϕ ∈ D0 . 

6.3 The intrinsic Dirichlet operator as a second quantization

The following two theorems are again analogous to the corresponding results (Theorems 5.2
and 5.3) in [7], so we omit their proofs.
Let us consider the special case of the second quantization operator d Exp A where the
operator A coincides with the Dirichlet operator HσeX×M .

Theorem 6.2 We have the equality

MP Ω
HH X×M = Hπσ
e
e
σ

on the dense domain FCp∞ (D0 , ΩM ∞ M

X ). In particular, for all F1 , F2 ∈ FCp (D0 , ΩX )
Z
h∇Ω F1 (ω), ∇Ω F2 (ω)iTω (ΩM ) πσe (dω)
X
ΩM
X
Z
= (∇MP F1 (ω), HσeX×M ∇MP F2 (ω))L2 (eσ) πσe (dω),
ΩM
X

or
∇Ω∗ ∇Ω = ∇MP∗ HσeX×M ∇MP
as an equality on FCp∞ (D0 , ΩM
X ).

Theorem 6.3 Suppose that the operator HσeX×M is essentially selfadjoint on the domain
D0 ⊂ Dom(HσeX×M ). Then, the intrinsic Dirichlet operator HπΩσe is essentially selfadjoint


on the domain FCb∞ D0 , ΩM
X .

34
Remark 6.2 Notice that in Theorem 6.3 we do not suppose the operator HσeX×M to be
conservative. So, this theorem is a generalization of Theorem 5.2 in the special case where
µκ,eσ = πσe .

Corollary 6.1 Suppose that the condition of Theorem 6.3 is satisfied and let TπΩσe (t) =
exp(−tHπΩσe ), t > 0. Then, for each ϕ ∈ D0 , ϕ > −1, we have
 X×M X×M 
TπΩσe (t) exp(hlog(1 + ϕ), ·i) = exp hlog(1 + e−tHσe ϕ), ·i − h(e−tHσe − 1)ϕiσe . (6.6)

Proof. The formula (6.6) follows from Proposition 6.1, (6.3), (6.4) and Theorems 6.2
and 6.3. 

Remark 6.3 If HσeX×M is conservative, then

Z
X×M
(e−tHσe − 1)ϕ de σ=0 for all t ≥ 0,

and so in this case (6.6) coincides with (5.17) for ϕ ∈ D0 , ϕ > −1.

ACKNOWLEDGMENTS

The authors were partially supported by the SFB 256, Bonn University. Support
by the DFG through Projects 436 113/39 and 436 113/43, and by the BMBF through
Project UKR-004-99 is gratefully acknowledged.

References
[1] M. Aizenman, S. Goldstein, and J. L. Lebowitz, Conditional equilibrium and the
equivalence of microcanonical and grandcanonical ensembles in the thermodynamic
limit, Comm. Math. Phys. 62 (1978), 279–302.

[2] S. Albeverio, A. Daletskii, and E. Lytvynov, Laplace operators and diffusions in

tangent bundles over Poisson spaces, Preprint SFB 256 No. 629, Universität Bonn,
1999, to appear in Proc. KNAW.

[3] S. Albeverio, A. Daletskii, and E. Lytvynov, Laplace operators on differential forms

over configuration spaces, Preprint SFB 256 No. 646, Universität Bonn, 2000, to
appear in J. Geom. Phys.

[4] S. Albeverio, R. J. Høegh-Krohn, J. A. Marion, D. H. Testard, B. S. Torrésani,

“Noncommutative Distributions—Unitary Representations of Gauge Groups and Al-
gebras,” Marcel Dekker, New York, Basel, Hong Kong, 1993.

[5] S. Albeverio, Yu. G. Kondratiev, and M. Röckner, Differential geometry of Poisson

spaces, C.R. Acad. Sci. Paris 323 (1996), 1129–1134.

35
 [6] S. Albeverio, Yu. G. Kondratiev, and M. Röckner, Canonical Dirichlet operator and
distorted Brownian motion on Poisson spaces, C.R. Acad. Sci. Paris 323 (1996),
1179–1184.

[7] S. Albeverio, Yu. G. Kondratiev, and M. Röckner, Analysis and geometry on config-
uration spaces, J. Func. Anal. 154 (1998), 444–500.
[8] S. Alberverio, Yu. G. Kondratiev, and M. Röckner, Analysis and geometry on con-
figuration spaces: The Gibbsian case, J. Func. Anal. 157 (1998), 242–291.

[9] V. I. Bogachev, N. V. Krylov, and M. Röckner, Elliptic regularity and essential self-
adjointness of Dirichlet operators on Rd , to appear in Ann. Scuola Norm. Pisa.
[10] W. M. Boothby, “An Introduction to Differentiable Manifolds and Riemannian Ge-
ometry,” Academic Press, San Diego, 1975.

[11] I. M. Gelfand, M. I. Graev, A. M. Vershik, Representations of the group of smooth

mappings of a manifold X into a compact Lie group, Compositio Math. 35 (1977),
299–334.
[12] I. M. Gelfand, M. I. Graev, A. M. Vershik, Representations of the group of functions
taking values in a compact Lie group, Compositio Math. 42 (1981), 217–243.

[13] I. M. Gelfand, R. A. Minlos, and Z. Ya. Shapiro, “Representations of the Rotation

and Lorentz Groups and their Applications,” Pergamon Press, Oxford, London, 1963.
[14] H. O. Georgii, “Canonical Gibbs Measures,” LNM 760, Springer-Verlag, Berlin, 1979.

[15] G. A. Goldin, What we have learned about local relativistic current algebra, in “Lo-
cal Currents and their Applications. Proceedings of an International Conference,”
pp. 100–114, North-Holland, Amsterdam, 1974.
[16] G. A. Goldin and R. Menikoff, Quantum-mechanical representations of the group of
diffeomorphisms and local current algebra describing tightly bound composite parti-
cles, J. Math. Phys. 26 (1985), 1880–1884.

[17] G. A. Goldin, R. Menikoff, and D. H. Sharp, Particle statistics from induced repre-
sentations of a local current group, J. Math. Phys. 21 (1980), 650–664.
[18] G. A. Goldin, J. Grodnik, R. T. Powers, and D. H. Sharp, Nonrelativistic current
algebra in the N/V limit, J. Math. Pgys. 15 (1974), 88–100.

[19] G. A. Goldin and D. H. Sharp, Particle spin from representations of the diffeomor-
phism group, Comm. Math. Phys. 92 (1983), 217–228.
[20] R. S. Ismagilov, “Representations of Infinite-Dimensional Groups,” American Math.
Soc., Providence, Rhode Island, 1996.

[21] Y. Ito and I. Kubo, Calculus on Gaussian and Poisson white noises, Nagoya Math.
J. 111 (1988), 41–84.

36
[22] O. Kallenberg, “Foundations of Modern Probability,” Springer-Verlag, New York,
Berlin, Heidelberg, 1997.

[23] J. F. C. Kingman, “Poisson Processes,” Clarendon Press, Oxford, 1993.

[24] Yu. G. Kondratiev, E. W. Lytvynov, and G. F. Us, Analysis and geometry on R+ -

marked configuration spaces, Meth. Func. Anal. and Geometry 5 (1999), no. 1, 29–64.

[25] Yu. G. Kondratiev, J. L. Silva, and L. Streit, Generalized Appell systems, Meth.
Func. Anal. and Topol. 3 (1997), no. 3, 28–61.

[26] Yu. G. Kondratiev, J. L. Silva, and L. Streit, Differential geometry on compound

Poisson space, Meth. Func. Anal. and Topol. 4 (1998), no. 1, 32–58.

[27] Yu. G. Kondrtatiev, J. L. Silva, L. Streit, and G. F. Us, Analysis on Poisson and
Gamma spaces, Infinite Dimensional Analysis, Quantum Probability and Related Top-
ics 1 (1998), 91–117.

[28] Yu. G. Kondratiev, T. Kuna, and J. L. da Silva, Marked Gibbs measures via cluster
expansion, Meth. Func. Anal. Topol. 4 (1998), no. 4, 50–82.

[29] T. Kuna, “Studies in Configuration Space Analysis and Applications,” Ph.D. Thesis,
Bonn University, 1999

[30] E. W. Lytvynov, A. L. Rebenko, and G. V. Shchepan’uk, Wick calculus on spaces of

generalized functions of compound Poisson white noise, Rep. Math. Phys. 39 (1997),
219–248.

[31] Z.-M. Ma and M. Röckner, Construction of diffusions on configuration spaces,

Preprint, Bielefeld, 1998, to appear in Osaka J. Math.

[32] K. Matthes, J. Kerstan, and J. Mecke, “Infinite Divisible Point Processes,” Akademie-
Verlag, Berlin, 1978.

[33] X. X. Nguyen and H. Zessin, Martin–Dynkin boundary of mixed Poisson processes.

Z. Wahrsch. verw. Gebiete 39 (1977), 191–200.

[34] N. Obata, Configuration space and unitary representations of the group of diffeomor-
phisms, RIMS Kokyoku 615 (1987), 129–153.

[35] N. Privault, A transfer principle from Wiener to Poisson space and applications, J.
Func. Anal. 132 (1995), 335–360.

[36] N. Privault, A pointwide equivalence of gradients on configuration spaces, C. R. Acad.

Sci. Paris 327 (1998) 677-682.

[37] M. Röckner, Stochastic analysis on configuration spaces: Basic ideas and recent re-
sults, In New Directions in Dirichlet Forms (eds. J. Jost et al.), Studies in Advanced
Mathematics, Vol. 8, American Math. Soc., 1998, 157–232.

37
[38] H. Shimomura, Poisson measures on the configuration space and unitary representa-
tions of the group of diffeomorphisms, J. Math. Kyoto Univ. 34 (1994), 599–614.

[39] A. V. Skorokhod, On the differentiability of measures which correspond to stochas-

tic processes I. Processes with independent increments, Theoria Verojat. Primen 2
(1957), 418–444.

[40] Y. Takahashi, Absolute continuity of Poisson random fields, Publ. RIMS Kyoto Univ.
26 (1990), 629–647.

[41] A. M. Vershik, I. M. Gelfand, and M. I. Graev, Irreducible representations of the

group GX and cohomologies, Funct. Anal. Appl. 8 (1974), 151–153.

[42] A. M. Vershik, I. M. Gelfand, and M. I. Graev, Representations of the group of

diffeomorphisms, Russian Math. Surveys 30 (1975), no. 6, 1–50.

[43] A. M. Vershik, I. M. Gelfand, and M. I. Graev, A commutative model of representation

of the group of flows SL(2, R)X that is connected with a unipotent subgroup, Funct.
Anal. Appl. 17 (1983), 137–139.

[44] N. Ja. Vilenkin and A. U. Klimyk, “Representations of Lie Groups and Special Func-
tions,” Vol. 1, Kluwer Acad. Publ., Dordrecht, 1991.

[45] A. Wawrzyńczyk, “Group Representations and Special Functions,” Kluwer Acad.

Publ., Dordrecht, 1983.

Institut für Angewandte Mathematik, Universität Bonn, Wegelerstr. 6, D 53115 Bonn;

and
SFB 256, Univ. Bonn; and
CERFIM (Locarno); Acc. Arch. (USI); and
BiBoS, Univ. Bielefeld
Institut für Angewandte Mathematik, Universität Bonn, Wegelerstr. 6, D 53115 Bonn;
and
SFB 256, Univ. Bonn; and
Institute of Mathematics, Kiev; and
BiBoS, Univ. Bielefeld
Institut für Angewandte Mathematik, Universität Bonn, Wegelerstr. 6, D 53115 Bonn;
and
BiBoS, Univ. Bielefeld
Radolfzellerstr. 9, D-81243 München, Germany

38