Japan. J. Math.

3, 19–47 (2008)
DOI: 10.1007/s11537-008-0752-5

Invariant or quasi-invariant probability measures for

infinite dimensional groups
Part II: Unitarizing measures or Berezinian measures

Paul Malliavin
Received: 4 November 2007 / Accepted: 20 February 2008
Published online: 22 March 2008
c The Mathematical Society of Japan and Springer 2008

Communicated by: Toshiyuki Kobayashi

Abstract. Some infinitesimal representations of Virasoro algebra are known to be unitarizable;

is it possible to realize the underlying Hilbert space as an L2 -space?
Keywords and phrases: infinite dimensional Siegel disk, univalent functions, Jordan curves, Vi-
rasoro algebra, Kirillov representation
Mathematics Subject Classification (2000): 11F46, 37K65, 60H07, 81R10

Introduction

Segal–Bargman (1959–1962) have constructed an L2 -holomorphic representa-

tion of the infinite dimensional Heisenberg group on the space of holomorphic
functions, square integrable relatively to a gaussian measure, the annihilation
operator being represented by ∂∂zk , the creation operator by the multiplication
by zk .
Our paradigm is the Berezin quantification of homogenous space which con-
tains as a particular case the Segal–Bargman construction; the most elementary
example of Berezinian quantification is, in the case of complex dimension 1,
constituted by the Poincaré disk with the volume element (1 − |z|2 )γ dz ∧ d z̄,
 This article is based on the 3rd Takagi Lectures that the author delivered at Graduate School
of Mathematical Sciences, the University of Tokyo on November 23, 2007.
P. M ALLIAVIN
10 rue Saint-Louis-en-l’Ile, 75004 Paris, France
(e-mail: sli@ccr.jussieu.fr)
20 P. Malliavin

which has a finite total mass for γ > −1. Then it is possible to construct unitary
representation of SL(2, R) on square integrable holomorphic functions, the ac-
tion involving now an automorphy factor, which has the shape (b̄z + ā)−γ where
the action is given by the homographic transformations. The particular case of
Jacobi θ function corresponds to the weight γ = 12 ; then the automorphy relation
correspond to the celebrated functional equation for the θ function discovered
by Poisson (J. Ecole Polytechnique 1822) and rediscovered ten years later by
Jacobi.
It is well known that this Berezin construction goes over in higher complex
dimension by replacing the Poincaré disk by the Siegel disk constituted by sym-
metric complex n × n matrices Z such that Z̄Z < Identity.
We will try to develop Berezin paradigm in an infinite dimensional setting.
Contents
1. Berezinian and unitarizing measures on Poncaré disk ..................... 20
2. Infinite dimensional Siegel disk ................................................ 23
3. Virasoro algebra ................................................................... 28
4. Representation of Virasoro algebra on the Siegel disk ..................... 30
5. Representation of Virasoro algebra on the space
of univalent functions ............................................................ 33
6. Unitarizing measure for Neretin representation of Virasoro algebra .... 39
7. Brownian motion on the space of Jordan curves ............................ 43

1. Berezinian and unitarizing measures on the Poncaré disk

 
Denote G the group SU(1, 1) constituted by the matrices g = ab̄ āb where |a|2 −
 iα β 
|b|2 = 1; its Lie algebra su(1, 1) is identified with β̄ −iα , where α is real; the
Poncaré disk D appears as D = SU(1, 1)/S1 .
The Kähler potential is defined as
(1.1a) K(z) := − log(1 − |z|2 ).
It satisfies the identity
(1.1b) K(gz) = 2ℜ log(b̄z + ā) + K(z).
It results from (1.1b) that the Kählerian metric ∂ ∂¯ K is invariant under the
action of G.
Theorem. Let H(D) be the space of holomorphic functions on Poincaré disk
D; let Hγ (D) be the subspace of holomorphic functions for which the following
Hilbertian norm is finite

(1.1c)  f 2γ := | f (z)|2 exp(−γ K(z)) dv(z),
D
Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 21

where dv is the hyperbolic volume measure on D: dv := (1 − |z|2 )−2 dz ∧ d z̄.

Define an action of G on H(D) by
 
γ 1 az + b
(1.1d) (Ta,b f )(z) = f ,
(b̄z + ā)γ b̄z + ā
γ
then Ta,b induces on Hγ (D) an unitary action;
1
is called the automorphy factor.
(b̄z + ā)γ

Proof.

γ
Ta,b ( f )2γ = exp(−γ (2ℜ log(b̄z + ā) + K(z))) × | f (g(z))|2 dv(z).
D

Make the change of variable z = g−1 (z ) we get by the invariance of the
hyperbolic measure

= exp(−γ (2ℜ log(b̄z + ā) + K(z))) × | f (z )|2 dv(z ).
D

Write (1.1b) for z = g(z) we get K(z ) = 2ℜ log(b̄z + ā) + K(z); therefore

γ
Ta,b ( f )2γ = exp(−γ K(z )) × | f (z )|2 dv(z ).
D


Bundle of orthonormal frame above D

We want to interpret the infinitesimal automorphy factor as an holomorphic con-

nection of a complex line bundle.
The automorphy factor expresses in terms of g ∈ G. The horizontal lift makes
possible to associate to a curve on D a lift to G and built in this way a map from
path space on D to automorphy factor. We shall compute these objects in the
framework of the formalism of orthonormal frame bundle. Let be the Lie
algebra of G; take for basis of :
     
i 0 01 0 i
e1 = , e2 = , e3 =
0 −i 10 −i 0
then [e1 , e2 ] = 2e3 , [e1 , e3 ] = −2e2 , [e2 , e3 ] = −2e1 .
We consider as a real Lie algebra define by the above bracket relations.
The vector e1 generates a group which is isomorphic to the circle S1 . The quo-
tient π : G → G/S1 is the Poincaré disk D which is a Kähler manifold, homoge-
neous under the left action of G. The transform of the base point 0 ∈ D by g ∈ G
22 P. Malliavin

is g(0) = b(ā)−1 . The Kähler potential is defined on D by − log(1 − |z|2 ); it lifts

to G as
(1.2a) K(g) := − log(1 − |b(ā)−1 |2 ) = log(1 + |b|2 ).
Remark that if
   
ab −1 ā −b
g= then g = ;
b̄ ā −b̄ a
therefore
(1.2b) K(g−1 ) = K(g).

Theorem. The orthonormal frame bundle O(D) of D is isomorphic to G; the
canonical parallelism associated to the Levi–Civita parallelism coincides with
the left invariant Maurer–Cartan form.
Proof. Denote Γ the tensor field describing the passage from the connection
associated to the right invariant parallelism to the Levi–Civita connection has
for expression
2Γ122 = ([e2 , e2 ] | e1 ) − ([e2 , e1 ] | e2 ) + ([e1 , e2 ] | e2 ) = 0.


Automorphy differential form

The differential dK of the Kähler potential can be written as the real part of an
holomorphic function; more precisely
Theorem. Let
d d
(∂eli K)(g) := K(exp(ε ei )g), (∂eri K)(g) := K(g exp(ε ei )),
d ε=0 d ε=0
then
(1.2c)
(∂el2 K)(g) = 2ℜ(ϕ (g)), where ϕ (g) := g(0) = b(ā)−1 ; (∂el3 K)(g) = 2ℑ(ϕ (g)).

Proof. To be precise denote K1 (g) = K(g(0)). Use (1.1b)

K(exp(ε e2 )z) = 2ℜ log(b̄ε z + āε ) + K(z),
(∂el2 K1 )(g) = 2ℜz, z = g(0).

Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 23

Infinitesimal generator of the modular representation

A function f on G is said holomorphic if it satisfies the equation

(1.3a) (∂er2 + i∂er3 ) f = 0.

Theorem. Define ϕ (g) = g(0) then ϕ is holomorphic. On the space  (G) of

holomorphic functions on G, the group G operates on the left f → g∗ f ; this ac-
tion defines the modular representation of G. The infinitesimal action on  (D)
of the modular representation is:
ρ (e2 ) f = (1 − z2 ) f  (z) − γ z f (z),
ρ (e3 ) f = i(1 + z2 ) f  (z) + iγ z f (z),
ρ (e1 ) f = 2iz f  (z) + iγ f .
We have
(1.3b) −ρ ([ei , e j ]) = ρ (ei )ρ (e j ) − ρ (e j )ρ (ei ).
√
Set νγ the Berezinian measure exp(−γ K) dv, then −1 × ρ (ei ) are Hermitian
operators on  L2νγ (D).

2. Infinite dimensional Siegel disk

The automorphic theory on the Poincaré disk corresponds to the complex di-
mension 1. In finite dimension it is generalized into Siegel automorphic forms
theory. We skip this intermediate step and go to the infinite dimensional object:
the infinite dimensional Siegel disk.
We consider the space V of real valued C1 -functions defined on the circle
with mean value equal to 0. On V we define a bilinear alternate form
 2π
1
(2.1a) ω (u, v) = uv d θ .
π 0

We introduce on V a complex structure defined by the Hilbert transform

 : sin(kθ ) → cos(kθ ), cos(kθ ) → − sin(kθ ).
We define on V an Hilbertian metric
u2 = −ω (u,  u).
We have  2
 
 ∑ ak cos(kθ ) + bk sin(kθ ) = ∑ k(a2 + b2 ).
  k k
k k
24 P. Malliavin

Then  is an orthogonal transformation of V .

We denote H = V ⊗ C; then H can be identified with complex valued func-
tion defined on the circle having mean value 0; on H we have the operation of
conjugation f → f¯.
√  can√be diagonalized in +H; as  = −1
2
The orthogonal transformation
only appears the eigenvalues √ −1 and − −1. We denote H the eigenspace
+
associated to the eigenvalue −1; then we can √ identify H to the vectors of
type (1, 0) that is the vectors of the form v − −1 (v), v ∈ V . We can also
identify H + with the functions having an holomorphic extension inside the unit
disk. Then define H − = H + ; then H − can be identified with the functions on
the circle which possess an holomorphic extension to the outside of the unit
disk and which are regular at the point at ∞ of the complex plane. The bilinear
form ω extends to a bilinear form ω̃ defined on H and we have

ω̃ (w, w ) = 0 if w, w ∈ H + or w, w ∈ H − .
We can express ω̃ in term of the Hilbertian structure

ω̃ (h+ , h− ) = (h+ | h̄− ), ∀h+ ∈ H + , h− ∈ H − ,

identity which is proved by checking on h+ = einθ , h− = eimθ , n > 0, m < 0.

We define a symmetric C-bilinear form on H × H by

(2.1b) h1 , h2 = (h1 |h̄2 ), then (h1 |h2 ) = h1 , h̄2 .

Then

(2.1c) ω̃ (h1 , h2 ) = h+ − − +
1 , h2 − h1 , h2 .

The restriction to H + ×H − of the bilinear form (∗) defines a duality coupling

h+ , h− := (h+ |h̄− ).

Given A ∈ End(H) we denote AT the transposed defined by

Ah1 , h2 = h1 , AT h2 .
 
Given a ∈ End(H + ), then the matrix a0 00 makes possible to identify
End(H + ) ⊂ End(H); then aT ∈ End(H) is the same in these two possible in-
terpretations; furthermore we have through the duality coupling (2.1b) makes

ah+ , h− = h+ , aT h− ;

which means that aT ∈End(H + ).

The adjoint a† ∈ End(H + ) is defined by

(aw1 |w2 ) = (w1 |a† w2 ), ∀w1 , w2 ∈ H + .

Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 25

The conjugation operator send H + → H − therefore ā ∈ End(H − ) and we have

the fact that the adjoint is obtained by conjugation followed by transposition
 T
a† = ā = aT .
An automorphism U of V extends to an endomorphism Ũ of H. Denoting π + ,
π − the projection of H on H + , H − we introduce
a := π +Ũ π + ; b := π +Ũ π − .
As the endomorphism Ũ commutes with the conjugation it is represented by the
matrix
 
ab
(2.1d) Ũg = .
b̄ ā

The Siegel disk in infinite dimension

The conservation of the symplectic form (2.1a) is equivalent to

(ā)T (a + b) − bT (ā + b̄) = π + , (b̄)T (a + b) − aT (ā + b̄) = −π − ;
we remark that the first relation is the conjugate of the second. Therefore we
have only to take care of the second relation which by splitting on the compo-
nents H + , H − which gives
(2.2a) aT ā − b† b = π + , aT b̄ − b† a = 0.
We call Symplectic Group of infinite order, let Sp(∞) the collection of bounded
operators a ∈ End(H + ), b ∈  (H − ; H + ) satisfying the relations (2.2a) and such
that
(2.2b) trace(b† b) < ∞.
Then Sp(∞) is a group for the composition of operators.
The infinite dimensional Siegel disk ∞ is defined as the set of operators
Z ∈  (H − , H + ) such that
Z T = Z, 1 − Z † Z > 0,
trace(Z † Z) < ∞.
 
We shall consider Z ∈ End(H) through the matrix 00 Z0 .
Theorem. The group Sp(∞) operates on ∞ by the law
(2.3a) Z → Y = (aZ + b)(b̄Z + ā)−1 .

Remark. In the above formula a, b are identified with elements of End(H).

26 P. Malliavin

Proof. We have firstly to show that Y ∈  (H − ; H + ): we have ā ∈ End(H − )

and b̄Z ∈ End(H − ); therefore (b̄Z + ā)−1 ∈ End(H − ).
We have secondly to show that Y T = Y :

Y T = (Zb† + a† )−1 (ZaT + bT );

therefore the identity Y T = Y is equivalent to

0 = (ZaT + bT )(b̄Z + ā) − (Zb† + a† )(aZ + b)

= Z(aT b̄ − b† a)Z + (bT b̄ − a† a)Z + Z(aT ā − b† b) + bT ā − a† b,

the first coefficient vanishes accordingly (2.2a); by conjugating (2.2a) we obtain

the vanishing of the fourth coefficient; we obtain

= π −Z + Z π +;
 
these two terms vanish according the fact that Z is in fact the matrix 00 Z0 .
We have to check that Δ := Y †Y − π − < 0. We introduce Δ := Y T Y − π − and
we denote D := (b̄Z + ā), then

D† ΔD = (Z † a† + b† )(aZ + b) − (Z † bT + aT )π − (b̄Z + ā)

= Z † (a† a − bT b̄)Z + Z † (a† b − bT ā) + (b† a − aT b̄)Z + (b† b − aT ā);

finally we get
Δ = −(D1 )† π + D−1 ≤ 0.
The orbit through Sp(∞) of the matrix Z0 = 0 is the space of matrices of the
form

(2.3b) Z = b(ā−1 ).

We remark that (a, 0) ∈ Sp(∞) iff a ∈ U(H + ) the unitary group of H. There-
fore the orbit of Z0 is contained in Sp(∞)/U(H + ).
It can be proved that the orbit of Z0 is  (∞).
We remark that a ∈ u(H + ) implies ā−1 = a† = aT ; therefore the action of
U(H + ) can be written as

Z → aZaT and Z̄Z → cZ̄Zc† ,

with c = ā. Therefore det(1 − Z † Z) is invariant under the action of U(H + ).

Define on ∞ the following function

(2.4a) K(Z) = − log det(1 − Z † Z) = −trace log(1 − Z † Z),

the last equality is intrinsic and do not depend upon a basis; it will be proved
using a basis diagonalizing Z † Z.
Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 27

The action (2.3b) of Sp(∞) on the matrix Z0 = 0 defines a lifting of K to

Sp(∞) by the formula
 
ab
(2.4b) K1 = log det(aT ā) = log det(1 + b† b) = trace log(1 + b† b).
b̄ ā
Under the action of Sp(∞) defined in (2.3a) we have (see [A-M01] formula
(3.2.5))
where appears the automorphy factor
K(Y ) − K(X) = 2ℜtrace(log D),
D := ā + b̄Z.
In order to compute the hessian of K we use the infinitesimal action of
  α β 
Sp(∞). The Lie algebra sp(∞) can be written as the 10 01 + ε β̄ ᾱ , where ε → 0
and where
α ∈ u(H + ) := {α ; α + α † = 0}, β ∈  := {β ; β − β T = 0},
the Lie brackett being given by
[α1 , α2 ] = (α1 α2 − α2 α1 ) ∈ u(H + ), [α , β ] = αβ − β ᾱ ∈  ,
[β1 , β2 ] = (β1 β̄2 − β2 β̄1 ) ∈ u(H + ).
The infinitesimal action on  (∞) is given by
(2.4c) exp(ε∂α )Z = Z + ε (α Z − Z ᾱ ), exp(ε∂β )Z = Z + ε (β − Z β̄ Z).
The infinitesimal action on K is given by
∂α K = 0, (∂β K)(Z) = 2ℜtrace(β̄ Z),
the first relation means that K is invariant under the left action of U(H + ) which
means that K is a radial function. The second derivatives are given by
∂z ∂α K = 0, ∂α ∂β K = ∂[α ,β ] K = 2ℜ(trace(ᾱ β̄ − β̄ α )Z),
(2.4d) ∂β1 ∂β2 K = 2ℜtrace(β̄2 (β1 − Z β̄1 Z)).
We use the underlying complex structure to split a real quadratic form in
components of type (1, 1) and (2, 0) (we can omit the component of type (0, 2)
which is imaginary conjugate to the component of type (2, 0)). The component
of type (0, 2) of the hessian is
(2.4e) −traceβ̄2 Z β̄1 Z .
The (1, 1) component of the hessian of K on  is trace(β2 β1 + β̄1 β2 ) that
is twice the scalar product associated to the Hilbert–Schmidt norm defining the
Riemann–Hilbert metric; so we prove:
(2.4f) K is the Kähler potential of an invariant Kähler metric on  (∞).

28 P. Malliavin

3. Virasoro algebra

The group of C∞ , orientation preserving diffeomorphism of the circle S1 will be

denoted by Diff(S1 ), its Lie algebra by diff(S1 ) the real valued C∞ functions φ
on S1 , the infinitesimal action being θ → θ + εφ (θ ).

[φ1 , φ2 ] = φ1 φ̇2 − φ2 φ̇1 .

We define on diff(S1 ) the bilinear antisymmetric form

  
dθ
(3.1a) ω ( f , g) := − f  + f (3) g .
S1 2π

Lemma.
ω ([ f1 , f2 ], f3 ) + ω ([ f2 , f3 ], f1 ) + ω ([ f3 , f1 ], f2 ) = 0.

Proof. We remark that [ f1 , f2 ] = f1 f¨2 − f2 f¨1 . As the wanted identity is linear

relatively to the bracket, it is sufficient to check it for the two following cases:
in the first case the first term of the identity is f1 f¨2 f3 − f2 f¨1 f3 and this term
cancelled with the others terms obtained by circular permutation; in the sec-
ond case we make an integration by part to replace the third derivative in the
expression of ω by a second derivative; then the first term of the wanted iden-
tity can be written as f1 f¨2 f¨3 − f2 f¨1 f¨3 and we get again cancellation by circular
permutation. 

Fixing a positive constant c, called the central charge, the Virasoro algebra
c is defined as vector space c := R ⊕ diff(S1 ); we denote by κ the central
element and define the bracket by
c
(3.1b) [ξ κ + f , ηκ + g]Îc := ω ( f , g)κ + [ f , g],
12
and according (1.2.2) the Jacobi identity is satisfied and we get a structure of
Lie algebra.
We denote by e0 the constant vector field which constitutes the Lie algebra
of the group S1 of rotations; We denote by diff0 (S1 ) the quotient of diff(S1 )/S1 .
We have e0 ∧ ζ , ω = 0 ∀ζ , therefore it is possible to quotient ω and we
get a well-defined 2-differential form on diff0 (S1 ) which will be by abuse of
notations still be denoted by ω .
Sometimes we shall identify diff0 (S1 ) with the functions having mean value
zero; using Fourier series this identification leads to write φ ∈ diff0 (S1 ) as
∞
φ (θ ) = ∑ ak cos kθ + bk sin kθ ,
k=1
Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 29

where ak , bk are rapidly decreasing sequences of real numbers. We define

∞
(3.1c) J(φ ) = ∑ −ak sin kθ + bk cos kθ .
k=1

Then J 2 = −Identity and we get a complex structure on diff0 (S1 ). On diff0 (S1 )⊗
√ type (1, 0) (resp.
C, the operator J diagonalizes; we call vector of √ of type (0, 1))
the eigenvectors associated to the eigenvalue −1 (resp. − −1). A vector of
type (1, 0) is of the form
√
φ = f − −1J( f ) = ∑ (ak − ibk ) cos kθ + (bk + iak ) sin kθ
k>0
= ∑ (ak − ibk ) exp(ikθ ),
k>0

which means that φ has a prolongation inside the unit disk as an holomorphic
function.
We take for basis of diff(S1 )⊗C the complex exponential ek (θ ) := exp(ikθ ),
k ∈ Z; in this basis the Lie bracket is
√ √ c
(3.2a) [em , en ] = −1 (n − m)em+n + −1 δ−mn
(m − m) κ .
3
12
The expression of ω can be written on for f1 , f2 ∈ diff0 (S1 ) ⊗C as
√
−1
(3.2b) ω ( f1 , f2 ) = ∑
2 k∈Z
(k3 − k)ck ( f1 )c−k ( f2 )
√
−1
= ∑
2 k>0
(k3 − k) f1 ∧ f2 , d ζk ∧ d ζ̄k ,

where f = ∑k∈Z ck ( f )ek and where d ζk denotes the R-linear map of diff0 (S1 )⊗C
defined by f → ck ( f ), k > 0; in the same way  f , d ζ̄k := c−k ( f ).
We have
√
d ζk , J f = −1d ζk , f ;
√
d ζ̄k , J f = − −1d ζ̄k , f , k > 0;
therefore
(3.2c) ω( f ,J f ) = ∑ (k3 − k)|ck ( f )|2 .
k>0
30 P. Malliavin

4. Representaion of Virasoro algebra on the Siegel disk

Theorem. If g is an orientation preserving diffeoomorphism of S1 then

ω (g∗ u, g∗ v) = ω (u, v).

Proof. Change of variable in the integral defining ω . 

We define an action of Diff(S1 ) on V by
 2π
1 ∗
(4.0) Ug−1 (u) = g u − (g∗ u) d θ .
2 0
Then
ω (Ug (u),Ug (v)) = ω (u, v).
We have in this way defined an embedment Ψ of Diff(S1 ) into the automor-
phism of V which preserves the symplectic form ω , that is the group Sp(∞):
(4.1a) Ψ : Diff(S1 ) → Sp(∞).
Define
(4.1b) K2 (g) := K(Ψ(g)).

Theorem. Given g ∈ Diff(S1 ) the operator bΨ(g) is given by a kernel

Bg (z, ζ ) = ∑ c j,k (g) z j ζ k ,

j,k>0

where
 2π
(4.2a) Bg (exp(−ikθ )) = exp(−ikθ )Bg (z, exp(iθ )) d θ = ∑ c j,k z j ,
0 j

and where
 2π
1
(4.2b) c j,k (g) = exp(−i( jθ + kg−1 (θ ))) d θ .
2π 0
Finally

1
(4.2c) b†g bg = ∑ η̄ j ζ l jl ∑ c̄k, j ck,l .
j,l>0 k k
j
(4.2d) trace(b†g bg ) = ∑ |c j,k |2 .
j,k>0 k

(4.2e) K2 (g) = trace log(I + b†g bg ), K2 (g) = K2 (g−1 ).

Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 31

−1
Proof. We consider the orthonormal basis αk exp(ikθ ), where αk = k 2 . Con-
sider the coefficients of the operator Ug in this basis: ĉ j,k = α j αk (exp(i jθ )|
Ug (exp −ikθ )); then
αk
ĉ j,k = c j,k .
αj
The operator b† b has for coefficients in this orthonormal basis
1
α j αl ∑ αk c̄ j,k αk cl,k .
k

By an integration by part

i
ck,l (g) = exp(−i(l θ + kg−1 (θ )))(−ik(g−1 ) (θ )) d θ ,
l
making the change of variable g−1 (θ ) = φ we get
k
= cl,k (g−1 ).
l
Therefore
l k
∑ k |ck,l (g)|2 = ∑ l |cl,k (g−1)|2 .


Remark. The functional K2 is relatively settled, another expression will be given

in (6.3a).

Ornstein–Uhlenbeck generator on the diffeomorphism group

In [M99] a diffusion process have been constructed through the associated SDE

(4.3a) d g̃x = dyx g̃x where dyx = ∑ αk ek (◦dxk )

k

where ek is the map R2 → diff(S1 ) defined by (ξ , η ) → ξ cos kθ + η sin kθ and

finally where {xk } is an infinite sequence of R2 valued independent Brownian
motions and where
1
αk = √ .
k3 + k
For the purpose of constructing a L2 representation of the Virasoro algebra,
we want to study the SDE

(4.3b) dgx (t) = dyx (t)gx (t) − c∇K2 (gx ) dt,

32 P. Malliavin

where the gradient is taken relatively to the metric α∗ and where c is a positive
constant, called the central charge and chosen sufficiently great. We have
(∇K)(g) = ∑ αk2 (∇k K)(g), where
k
(4.3c)
d
∇k K(g) = (K(exp(ek ((ε , 0))(ek (1, 0))))+K(exp(ek ((0, ε ))ek ((0, 1))))).
d ε|ε =0

Lemma. The differential of K2 is given by

(4.4a) dK2 = trace(1 + b† b)−1 ((db)† b + b† db).

Proof. Denote c := b† b then

d(c − c0 )n = dc(c − c0 )n−1 + (c − c0 )dc(c − c0 )n−2 + · · · + (c − c0 )n−1 dc
trace(d(c − c0 )n ) = trace(n(c − c0 )n−1 dc).
Developping c → log(1 + c) in Taylor series around the point c0 , we get the
result. 

Proposition.
(4.4b)
(∇s ck,l )(g)

−il 2π
= exp(−ikg(φ )−il φ )(es ((0, 1)) cos(sg(φ ))−es ((1, 0)) sin(sg(φ ))) d φ .
2π 0
Proof. Let z ∈ Diff(S1 ) then denote by ∂zl the derivative on the left associated to
z, then
 2π
−il
(∂zl ck,l )(g) = exp(−i(kθ + lg−1 (θ )))(g−1 ) (θ )z(θ ) d θ .
2π 0

Making the change of variables φ = g−1 (θ ), we get the result. 

Theorem 4.5. The function K2 is the Kählerian potential of the metric  2 .

3

Proof. We consider the left invariant metric on the diffeomorphism group de-
fined g−1 h 3 where h ∈ Tg (Diff(S1 )),
À 2

 2π
d
∂h c j,k = exp(−il θ − ik(−ε h(θ ) + g−1 (θ ))) d θ
d ε|ε =0 0
Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 33

 2π
=i exp(−il θ − ikg−1 (θ ))h(θ ) d θ .
0

We remark that denoting e the identity of the diffeomeorphism group we

have c j,k (e) = 0. Furthermore (∂h c j,k )(e) = 0 if h is antiholomorphic. Then

l 1
∂ ∂¯ K, einθ ∧ e−inθ = ∑ k2 1(k + l = n) = ∑ kl = (n3 − n).
k,l k k+l=n 6

In order to prove the theorem at every point of Diff(S1 ) we shall proceed

using homogeinity argument.
Hong–Rajeev [H-R] have constructed a map

Θ : Diff(S1 )/PSL(2, R) → ∞ ,

which satisfies the intertwinning formula

Θ(g0 g) = Ψ(g0 )Θ(g);

as on the Siegel disk ∞ the (1, 1) form ∂ ∂¯ K is invariant under the left action
of symplectic group, this intertwining formula proves the theorem. 

5. Representation of Virasoro algebra on the space of univalent functions

We denote by  the vector space of functions f which are holomorphic in the

unit disk D := {z; |z| < 1} and C∞ on its closure D̄ and such that f (0) = 0; we
denote by 0 the subspace of functions of f satisfying f  (0) = 0. We denote

:= { f ∈  ; f  (0) = 1 and f injective on D̄, f  (z) = 0, ∀z ∈ D̄}.

Then is an open set of the affine space f0 + 0 where f0 (z) = z, ∀z ∈ D.
As 0 is a complex vector space, inherits of an infinite dimensional structure
of complex manifold. The embedment → CN , defined by
 +∞ 
(5.1a) f (z) = z 1 + ∑ cn z ,n
n=1

introduces the affine coordinates f → {c∗ }. Granted De Branges Theorem,

is identified to an open subset of {|cn | < n + ε }.

Theorem 5.2. There exists a canonical identification of with Diff(S1 )/S1 . As

a consequence is an homogeneous space under the left action of Diff(S1 ).
Furthermore there exists a global section σ : → Diff(S1 ).
34 P. Malliavin

Proof. Given f ∈ , we denote Γ = f (∂ D). Then Γ is a smooth Jordan curve

which splits the complex plane into two connected open sets: Γ+ which contains
0 and Γ− which contains the point at infinity of the Riemann sphere (the map
z → z−1 sends Γ− onto a bounded domain countaining 0).
By the Riemann mapping theorem there exists an holomorphic map φ f :
D → Γ̄− such that φ f (∞) = ∞, φ being holomorphic nearby ∞.
c

A diffeomorphism g f ∈ Diff(S1 ) is defined by

g f (θ ) := ( f −1 ◦ φ f )(eiθ ).
We remark that φ f is uniquely defined up to a rotation of Dc , which means
up to an element of S1 ; therefore we get a canonical map

: → Diff(S1 )/S1 .
Furthermore it has been proved by Beurling–Ahlfors that
is bijective.
The stabilizer of the left action of S1 is the identity function f0 (z) = z.
We obtain the section σ by choosing φ 0f such that u(z) := φ 0f (z−1 ) satisfy
that u (0) is real and positive. 

Theorem 5.3. The fundamental cocycle (3.1a) defines a 2-differential closed

form Θ on which is invariant under the action of Diff(S1 ).

Proof. Given f ∈ denote S f := {g ∈ Diff(S1 ) such that g f = f0 }; define Θ f =

(g−1 )∗ ω for some g ∈ S f ; if γ also belongs to S f we have γ = gu with u ∈ S1 ;
then (u−1 g−1 )∗ ω = (g−1 )∗ ((u−1 )∗ ω ) expression equal, to (g−1 )∗ ω ; therefore
our definition is independent of the choice of g ∈ S f and Θ is a well-defined
2-differential form on , which is closed.
The invariance under the action of Diff(S1 ) results of the invariance of Mau-
rer–Cartan differential form or more elementary from the identities
Θγ g( f0 ) = (g−1 γ −1 )∗ ω = ((γ −1 )∗ Θ)γ g( f0 ) .


Kirillov infinitesimal action of diff(S1 ) on

Theorem. Given v ∈ diff(S1 ), and given f ∈ we define Kv ( f ) ∈ 0 by the

formula
   2
f 2 (z) t f (t) v(t) dt
(5.4) Kv ( f )(z) := ,
2π ∂ D f (t) f (t) − f (z) t
then Kv is the infinitesimal expression of the left action at the point f .
Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 35

Proof. Given v ∈ diff(S1 ) we consider the infinitesimal action

exp(ε v)( f −1 ◦ φ f ).

Denote by ∂v the first order differential operator associated to v; then we have

the intertwinning formula ∂v f −1 = f −1 ∂w where wτ = i f  (t)tvt , with t ∈ ∂ D and
τ = f (t) ∈ Γ; we can split w := w+ + w− where w+ has an holomorphic pro-
longement w++ to Γ+ satisfying 0 = w++ (0) = (w++ ) (0) and w− is meromor-
phic on Γ− with a unique pole at infinity which is simple; the Cauchy formula
gives

++ ζ2 w(τ )
w (ζ ) = dτ .
2π i Γ τ (τ − ζ )
2

Doing again an intertwinning by defining

w∗ defined by the relation

fε := exp(ε w++ ) ◦ f = f ◦ exp(ε w∗ ),
f  (z)w∗ (z) = w++ (ζ ),

we get then exp(ε v) f  fε−1 ◦ exp(ε w− )φ f which implies

fε exp(ε v)·
( f ).


Theorem 5.5. Fixing g0 ∈ Diff(S1 ) the map f → g0 f is an holomorphic map

→ .
Proof. We prove this fact when g0 = exp(v0 ); introducing hτ := exp(τ v0 ) f , we
have that
d
hτ = Kv0 (hτ ), h0 = f .
dτ
As the solution of an holomorphic differential equation depends holomorphi-
cally of its initial condition, we have only to prove that the map

→ 0 defined by f → Kv0 ( f )
depends holomorphically upon f , the fact which results from the formula given
in (5.4).
We denote by T ( ) the vector space of real first order differential operators
on (a real differential operator transforms real functionals into real function-
als). We call also the elements of T ( ) the real vector fields on .
Then the complex structure J on induces the splitting

T ( ) ⊗C = T (1,0) ( ) ⊕ T (0,1) ( ).
√
We have a map J˜ : T ( ) ⊗ C → T ( ) defined by (X + −1Y ) → X +
J(Y ); then T (0,1) ( ) can be characterized as the kernel of J;˜ furthermore for
an holomorphic functional Ψ we have Z · Ψ = (J(Z))˜ · Ψ. 
36 P. Malliavin

Theorem. Define
√ √
(5.6a) Lck := − −1Kcos kθ + Ksin kθ , ∀k = 0; Lc0 := − −1Ke0 ;
(5.6b) ˜ ck ),
Lk := J(L

then

(5.6c) Lk ( f ) = f  (z)zk+1 , k > 0; L0 = f  (z)z − f (z);

(1,0)
(5.6d) Lck ( f0 ) ∈ T f0 ( ) for k > 0;
(5.6e) [Lcj , Lck ] = ( j − k)Lcj+k , ∀ j, k ∈ Z;
(5.6f) [Lm , Ln ] = (m − n)Lm+n , ∀m, n ∈ Z.

Define Lhk as the (1, 0) component of the real vector field Lk , then ∀ k ∈ Z

(5.6g) Lhk Φ = Lck Φ, ∀Φ holomorphic; [Lhm , Lhn ] = (m − n)Lhm+n , ∀m, n ∈ Z.

Proof. The identity (5.6c) results from the computation of the integral (5.4) by
residue calculus.
Denote vl the right invariant vector field associated to v ∈ diff(S1 ); in partic-
ular to ek = exp(ikθ ) is associated elk , and by (3.2a) we have
√
[elm , eln ] = −1 (m − n)em+n , therefore
√
[Lcj , Lck ] = − −1 ( j − k)e j+k = ( j − k)Lcj+k .

We have the fact that the complex structure on , is invariant under the left
action of Diff(S1 ); therefore denoting φ the Lie derivative associated to the
left action of φ ∈ diff(S1 ) we have
 c 
φ J(L ˜ φ (Lck )).
˜ k ) = J(

Therefore
φ Lk = J([
˜ φ , Lck ])
which by C linearity implies that (5.6f).
Denote by  −1 the linear map identifying T f0 ( ) with diff0 (S1 ) on which
we put the complex structure defined in (3.1c). To prove that  is holomor-
phic is equivalent to verify that the map v → Kv ( f0 ) is C-linear or that, ∀k > 0,
Lk̄ ( f0 ) = 0; indeed
 
2π t −k dt dt
L ( f0 ) = = − ∑ zs = 0.
z2 k̄ ∂D z −t t s≥0 ∂D t s+k+1
Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 37

Denote Lan = Ln − Lhn the antiholomorphic component of Ln which is equal to

L̄hn . As we shall see later the operators Lhn in affine coordinates have holomorphic
coefficients which imply that [Lhn , Lam ] = 0; therefore
 h
[Lm , Ln ] = [Lhm , Lhn ] which establish (5.6g).

Theorem. The 2-differential form
(5.7a) Θ defined in (5.3) is of type(1, 1), positive definite, and
√
(5.7b) Θ f0 (Lcn ( f0 ), L̄cm ( f0 )) = −1δnm γn , where γn = (n3 − n).

Proof. As Θ is defined by transport through the holomorphic action of Diff(S1 ),

it is sufficient to check (5.7a) at f0 .
ω (−i exp(inθ ), i exp(−imθ )) = ω (exp(inθ ), exp(−imθ )) = −i(n − n3 )δnm .

Remark. Θg( f0 ) , vg( f0 ) ∧ w̄g( f0 ) = ω , g−1 vg ∧ g−1 w̄g = ω , v ∧ w̄ .

The vector fields Ln , n > 0, in affine coordinates

The formula (5.1a) gives a global chart → CN in terms of the Taylor coeffi-
cients f → {c∗ ( f )}. We want to express the vector fields Lhn in this chart.

Realization of the flow associated to Lhn , n > 0

Firstly we remark that the vector field Lcn , Lhn are not real vector fields; flows
of complex vector fields need to be defined some analytic prolongation; where
there exists they could be very singular. Consider on a neighbourhood of 0 the
following holomorphic function
(k) z
(5.8a) Mt (z) = 1 ,
(1 − tkzk ) k
where t is a small real parameter. Then for z small enough we have
 (k) (k)  (k)
(5.8b) Mt ◦ Mt  (z) = Mt+t  (z).
(k)
Then given f ∈ the composition f ◦ Mt ∈ / but it is an holomorphic
function defined for t small enough on an arbitrary large disk contained in the
unit disk.
38 P. Malliavin

We have ∀z0 , |z0 | < 1 the following identity:

(k)
f (Mt (z0 )) − f (z0 )
(5.8c) lim = f  (z0 )zk+1
0 .
t→0 t
We call cylindrical functional a map Φ : → R such that Φ( f ) = φ (. . . ,
f (zi ), . . .) where φ : Dr → R, φ being smooth, and i ∈ [1, r], |zi | < 1. Then

d  (k) 
(5.8d) (Lk Φ)( f ) = Φ f ◦ Mt , k > 0.
h
dt t=0
As the vector field Lhk is of type (1, 0) it is sufficient to check the identity
assuming furthermore that φ is holomorphic, then
(k)
Mt (z) = z + tzk+1 + o(t),
 ∞ 
(k)
(5.8e) f (Mt (z)) = (z + zt k ) 1 + ∑ cn zn (1 + kntzk ) + o(t).
n=1

Theorem. Denote by ∂k the holomorphic

√ partial derivative√relatively to the the
affine coordinate ck (if ck = ξk + −1ηk then 2∂k = ∂∂ξk − −1 ∂∂ηk ), then
+∞ 
(5.9a) Lk = ∂k + ∂k + ∑ (n + 1) cn ∂k+n + c̄n ∂¯k+n ),
¯ k > 0.
n=1

Proof. Consider a cylindrical function Φ, which is assumed furthermore to be

holomorphic, which means that φ is an holomorphic map Dr → C. Then all ∂¯
vanishes on such Φ. Any holomorphic cylindrical functional can be approxi-
mated by polynomials in the c∗ . We take Ψ := cqk0 . By [A-M01], formula (1.7.4)
we have:
⎧
 (k) ∗  ⎨ ck0 ( f ) if k > k0 ;
Ψ (Mt ) f = (t + ck0 ( f )) + o(t),
q
if k = k0 ;
⎩
(ck0 ( f ) + (k0 − k + 1)tck0 −k ( f ))q + o(t) if k < k0 .
Differentiating relatively to t and making t = 0 we obtain respectively 0 or
q−1
qck0 or q(k0 − k + 1)[ck0 ( f )]q−1 ck0 ( f ), relations which proves the theorem for
holomorphic functionals. The l.h.s. of (5.9a) can be written as the sum of a
(1, 0) vector field Z plus a (0, 1) vector field Y ; as Lk is a real vector field we
have Y = Z̄ relation which proves the theorem. 
Using the definition of Lh∗ we obtain
+∞
(5.9b) Lhk = ∂k + ∑ (n + 1)cn ∂k+n , k > 0, Lh0 = ∑ ncn∂n .
n=1 n≥1
Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 39

6. Unitarizing measure for Neretin representation of Virasoro algebra

In the affine coordinates {ck } on , Neretin introduced the sequence of poly-

nomials Pn defined by the following double indices recurrence relations:
(6.1a)
c
Lhk Pn = (n + k)Pn−k + γk δkn where γk = (k3 − k), P0 = P1 = 0, Pn (0) = 0,
12
where the central charge c has been fixed.
Lemma.
(6.1b) The polynomial Pn is of weight n.

Proof. The weight w(P) := ∑k k × (degree of P relatively ck ). Then the recur-

rence implies that Lh1 Pn = (n + 1)Pn−1 ; the lemma will result of the identities

w(Lh1 cα1 ) = α − 1 = w(cα1 ) − 1; w(Lh1 cαk ) = k(α − 1) + (k − 1) = w(cαk ) − 1.


Theorem. Denote  ( ) the vector space of holomorphic functionals defined
on ; we associate to Φ ∈  ( )
√ c √
(6.1c) ρ (κ )Φ = −1 Φ, ρ (e0 ) = −1 L0 Φ,
√ 12 √
ρ (exp(ikθ ))Φ = −1Lck Φ, ρ (exp(−ikθ )) = −1(Lc−k + Pk )Φ, ∀k > 0.
Then ρ is an anti-representation of c on  ( ).
Proof. Granted the holomorphy of Φ we can replace by Lcn by Lhn ; as Lhn are
operators with holomorphic coefficients we deduce that Lcv send an holomorphic
functional into an holomorphic functional; as Pn is holomorphic we obtain that
the operator ρ operates on  ( ).
The vector fields Lcv realize an anti-representation of diff(S1 ); Theorem is
proved when v1 , v2 are two exponential with positive frequencies.
Consider the case of v1 of positive frequency and v2 of negative frequency:
[ρ (exp ikθ ), ρ (exp −isθ )] = −(Lhk Ps + (k + s)Lck−s ) := Bk,s .
For k > s we deduce from (6.1b) and (5.9b) that Lhk Ps = 0.
For k = s we have
Bk,k = ρ (γk κ + 2ke0 ).
For k < s we have
Bk,s = (k + s)(Ps−k + Lck−s ) = ρ ((k + s)ek−s ).
40 P. Malliavin

Finally it remains to prove compute, for k, s > 0 the expression

(6.1d)
[ρ (exp(−ikθ )), ρ (exp(−isθ ))]−[Lhek , Lhes ] = Lh−k Ps −Lh−s Pk = (s−k)Pk+s , ∀s, k ≥ 0,
basic identity which is proved is [A-M01], formula (2.1.4). 

Differential form on Diff(S1 ) associated to Neretin polynomials

We associate to v ∈ diff(S1 ) the right invariant tangent vector field to Diff(S1 )

defined by (vl )g = exp(ε v)g; in particular to e2k = cos kθ , e2k+1 = sin kθ , e0 =
1 ∈ diff(S1 ) we associate the right invariant vector fields el∗ := exp(ε e∗ )g. We
consider the following first order
√ differential operator with complex
√ coefficient
defined on Diff(S ): L̃n := − −1e2s + e2s+1 s = 0 and L̃0 = − −1el0 . Denote
1 c l l

ψ−s is the dual basis of L̃cn that is

(6.2a) L̃cn , ψs = δn+s
0
.
It results from the duality that the differential of a function Ψ defined on
Diff(S1 ) has for expression
(6.2b) dΨ = ∑ (L̃c−k Ψ) ψk .
k∈Z

Furthermore the {ψs } satisfy the structural equation

−k
d ψk , L̃cq ∧ L̃cr = (r − q)ψk , L̃cq+r = (r − q)δq+r ,
or finally,
1
(6.2c) d ψk = − ∑ (k + 2s)ψ−s ∧ ψk+s .
2 s∈Z
A complex valued 1-differential form Ω on Diff(S1 ) will be built from Nere-
tin polynomials by the formula
(6.2d) Ω= ∑ (Pk ◦ π ) ψk ,
k>0

where π is the projection map Diff(S1 ) → Diff(S1 )/S1 .

Neretin Theorem.

1 dt
(6.3a) v , Ω g =
l
S f (t) v(logt) t 2 ,
2π ∂D t
where g = f −1 ◦ φ f and where S f is the Schwarzian derivative:
 
f (3) 3 f  2
(6.3b) S f :=  − .
f 2 f
Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 41

Proof. A proof can be founded in [A-M01], formula (A7.2). 

Theorem (Automorphy differential form). Denote by Θ̃ the image of the sym-
plectic form (5.7a) defined by the transformation g → g−1 , then
√
−1
(6.4a) Θ̃ = − ∑ γs ψ−s ∧ ψs ,
2 s>0
where γs is defined in (6.1a); we have
√
(6.4b) dΩ = −1 Θ̃.

Proof. The form Θ is left invariant. Therefore Θ̃ is right invariant. The forms
{ψq } constitutes a basis of right invariant differential forms; therefore there ex-
ists constants ck,s such that
1
Θ̃ = ∑ ck,s ψ−k ∧ ψ−s.
2 s,k∈Z

In order to compute the constants ck,s we look at this identity at the point f0 :
then for s > 0
√
Θ̃ , Lcs ∧ L̄cs f0 = −1 γs = cs,−s , L̄cs , ψs f0 = −cs,−s .
Define Ps∗ = Ps ◦ π for s > 0 and Ps∗ = 0 for s ≤ 0; with these notations
Ω = ∑s Ps∗ ψs . Using (6.2b)
dΩ = ∑ (L̃c−k Ps∗)ψk ∧ ψs + R,
k,s∈Z

where R, granted (6.2c), has the following expression

1 1
R=− ∑
2 r,t∈Z
Pr∗ (r + 2t)ψ−t ∧ ψr+t = − ∑ Ps+k
2 k,s∈Z
∗
(s − k) ψk ∧ ψs .

Therefore
1
dΩ = ∑ Bk,s ψk ∧ ψs ,
2 k,s∈Z
where
(6.4c) Bk,s := L̃c−k Ps∗ − L̃c−s Pk∗ − (s − k)Ps+k
∗
.
For k > 0 and s > 0 we obtain, granted (6.1d), Bk,s = 0.
For k < 0 and s < 0 we obtain that all the P∗ vanishes and Bk,s = 0.
Consider the last case k < 0 and s > 0; if −k > s the polynomial Ps being
of weight s do not depend the variables c j for j > s, therefore L−k Ps = 0; for
−k ≤ s the defining relations (6.1a) proves that B−k,s = γ−k δ−k
s
. 
42 P. Malliavin

Unitarizing measure

Definition. A probability measure μ on is an unitarizing measure for the

representation ρ of c in  L2μ if and only if for all v real (i.e., v ∈ c ) the
√
operator −1 × ρ (v) is Hermitian:

(6.5a) ρ (v) + (ρ (v))∗ = 0.

The relation (6.5a) is equivalent to

(6.5b) (ρ (exp(ikθ )))∗ = −ρ (exp(−ikθ )).

Theorem. A probability measure μ is unitarizing if it satisfies the following

relation
(6.5c)
divμ (Lck ) = P̄k , k ≥ 0, or divμ (Kcos kθ ) = ℑPk , divμ (Ksin kθ ) = ℜPk .

c
Proof. The unitarity condition (6.5b) is given in terms of the L√ k through the
formulas (6.1c); then the appearance in those formulas of a factor −1 changes
the sign of (6.5b):
   ∗ 
(6.5d) (Lk Φ)Ψ̄ − Φ (L−k + Pk )Ψ
c c
d μ = 0,
Å

where the ∗ above the parenthesis indicate that we take the imaginary conjugate.
Firstly we replace Lck → Lhk , Lc−k → Lh−k and we use the identities Lhk (ΦΨ̄) =
(Lhk Φ)Ψ̄ and ΦL̄h−k Ψ̄ = L̄h−k (ΦΨ̄); then (6.5d) takes the shape
  
(6.5e) (Lk − L−k − P̄k ) (ΦΨ̄) d μ = 0 or divμ (Zk ) = P̄k ,
h h
Å

where Zk = Lhk − Lh−k . The operator Lck = Lhk + Lak where Lak is a vector field of
type (0, 1). Using the fact that Lc−k = −L̄ck we get by conjugation that Lh−k = −Lak ,
therefore Zk = Lck and finally divμ (Lck ) = P̄k .
We explicit the differential operators in terms of real differential operators:
the decomposition
√
Lck = − −1Kcos kθ + Ksin kθ
implies, as the divergence of a probability measure is a real operator, that the
second part of (6.5c) holds true. 
Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 43

Definition. We call Berezinian measure a probability measure which, through

the divergence relatively to the left action of c , defines a differential form Ω
which is a multiple of the differential of the Kähler potential K.
Remark that specialized in finite dimension, as the Riemannian volume ele-
ment is invariant under the left action, we find back the definition of Berezinian
measure used in section 1.
Then (6.4b) and (6.5c) leads to
Every unitarizing measure is a Berezinian measure.

7. Brownian motion on the space of Jordan curves

For somebody having some practice with the theory of univalent functions the
formula (6.3a) is much more appealing that its equivalent formulation (4.3c) on
the Siegel disk. As a first step to the construction of an OU process as in (4.3),
it is therefore appealing to construct the Brownian motion associated to the rep-
resentation of Virasoro algebra on univalent functions; then a new construction
is needed; it has been done in [A-M-T].
A Jordan curve on the complex plane C is a closed subset Γ ⊂ C such
that there exists a continuous injective map of the circle φ : S1 → C satisfy-
ing φ (S1 ) = Γ; this parametrization φ is not unique: given two parametrizations
φ1 , φ2 of the same Jordan curve then there exists an homeomorphism h of S1
such that φ2 = φ1 ◦ h. The parametrizations define the same orientation of Γ if h
is an orientation preserving homeomorphism of the circle.
Denote  the set of all Jordan curves; it is a non-closed subset of the set of
all compact subsets of the plane.
When Γ is rectifiable, the arc length provides a canonical parametrization of
Γ.
More generally we can use holomorphic parametrizations, constructed in
the following way: the complement of Γ in C is the union of two connected
open subsets D+ −
Γ bounded and DΓ which is unbounded. The Riemann mapping
theorem gives the existence of a conformal map ψ + : D+ Γ onto the open unit
disk, defined up to the composition to an element of SU(1, 1), the group of
automorphisms of the Poincaré disk of the form z → az+bb̄z+ā
. Then by a theorem
of Caratheodory, ψ has a continous injective extension F + to D̄+
+
Γ , the closure
of D+ Γ , then (F + )−1 | gives a parametrization of Γ canonically defined up an
S1
element of SU(1, 1); the advantage to use holomorphic parametrization is that
its undeterminacy corresponds to the finite dimensional group SU(1, 1). Then
gΓ := F − ◦ (F + )−1 is an orientation preserving homemorphism of the circle S1 .
Denote  ∞ the space of C∞ Jordan curves. We take φ ∈ C∞ (S1 ; C) and
define Γ := Φ(S1 ); it can be proved then that gΓ ∈ Diff(S1 ) the group of C∞
orientation preserving diffeomorphisms of the circle; furthermore the classical
44 P. Malliavin

theory of C∞ conformal welding shows that the map Γ → gΓ is a surjective map

of  ∞ onto Diff(S1 ); finally we have that  ∞ is isomorphic to the following
double quotient
(7.1)  ∞  SU(1, 1)\Diff(S1 )/SU(1, 1).
Denote diff(S1 ) the Lie algebra of Diff(S1 ) of right invariant vector fields on
Diff(S1 ); we can identify diff(S1 ) with C∞ vector fields on S1 ; every such vector
field is of the form u ddθ , where u is a C∞ function on S1 ; the Lie algebra bracket
is given by [u, v] = uv̇ − vu̇. The complexified Lie algebra diff(S1 ) ⊗ C has a
natural basis ẽn := exp(inθ ) × ddθ in which the Lie bracket has the following
expression [ẽn , ẽm ] = i(m − n)ẽm+n , m, n ∈ Z.
The Poincaré automorphism group SU(1, 1) has for Lie algebra the Lie sub-
algebra of diff(S1 ) generated by 1, cos θ , sin θ : the vector field 1 × ddθ corre-
sponds to rotation of the disk, the vector field sin θ × ddθ corresponds to the fol-
lowing one parameter family of Poincaré automorphisms hλ where the relation

z = hλ (z) is zz −1
+1 = exp(iλ ) z+1 .
z−1

Theorem. There exists up to the multiplication by a constant an unique Rie-

mannian metric on  which is invariant under the action of SU(1, 1) on the
left and on the right. This metric is defined by one of its orthonormal basis given
in (7.2).
Proof. Consider the quotient diff(S1 )/sl(2, R), the adjoint action by ẽ j , j =
0, ±1 preserves sl(2, R) because SL(2, R) is a subgroup; therefore the adjoint
action passes to the quotient and acts infinitesimally on  ; let adr (ẽ j ) be this
action which by hypothesis is unitary as well the corresponding action adl (ẽ j )
on the left.
As adr (ẽ0 ) is unitary this means that the metric is invariant by rotation. Ex-
tending the metric as an hermitian metric on the complex valued function, the
invariance by rotation implies that the metric diagonalizes in the basis {ẽn }: we
denote {en } the orthonormal basis obtained by multiplying each ẽn by a normal-
izing constant.
Existence of the metric. Take
1
(7.2) en :=
exp(inθ ), |n| > 1,
|n|(n2 − 1)
Denote ρ j = adr (ẽ j ), j = 0, ±1; we have

ρ j (en ) = i ∑ βnk ( j)ek ,

k

(n − j)2 |n + j|(n + j)2 − 1) k
βnk ( j) := δn+ j ,
|n|(n2 − 1)
Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 45

where δ∗∗ denotes the Kronecker symbol. The unitarity condition (ρ j )∗ + ρ− j =

0 is equivalent to
βnk ( j) = βkn (− j), j = 0, 1.
Remark firstly that δn+ k
j = δk− j ; the verification in the case j = 0 is tauto-
n

logical. In the case of j = 1

(n − 1)2 |n + 1|((n + 1)2 − 1) (n + 1 + 1)2 |n|(n2 − 1)
= .
|n|(n2 − 1) |n + 1|((n + 1)2 − 1)
The computation above gives unitarity with the exception of the cases where the
denominators involved vanish
n(n2 − 1) = 0, (n + 1)((n + 1)2 − 1) = 0; n = −2, −1, 0, 1.
For n = 2, ρ1 (e−2 ) = ρ−1 (e−1 ) = 0; by passsage to the quotient we get the
unitarity for adr (e j ), j = −1, 0, 1.
Unicity of the metric. Any invariant metric diagonalizes in the basis {ẽn }; an-
other invariant metric will correspond to an orthonormal basis {en } where en =
an en where the an are positive constants. Such a transformation has for effect to
multiply
an
βnk (∗) → βnk (∗) = [β  ]kn (∗).
ak
The unitarity condition
[β  (1)]kn = [β  (−1)]kn
implies that
an ak
= , ∀n, k, βnk (1) = 0.
ak an
the last condition corresponds to n = k − 1; as the a∗ are positive we get
ak−1 = ak , ∀k.
By invariance of the metric: the passage from the left to the right has for
effect to change the sign of the operator ad. 
Remark. We find again the Kähler metric associated to a cocycle defining the
Virasoro algebra in (3.2).
Recall the construction of the canonical Brownian motion “on” Diff(S1 ).
The regularized canonical Brownian motion on the group of diffeomorphisms
of the circle is the stochastic flow on the circle S1 associated to the Itô SDE
dgrx,t (θ ) = dzrx,t (grx,t (θ )),
(7.3) rk
zrx,t (ψ ) := ∑ 3 (x2k (t) cos kψ + x2k+1 (t) sin kψ ),
√
k>1 k − k
46 P. Malliavin

where {x∗ } is a sequence of independent scalar valued Brownian motions and

where r ∈]0, 1[. It results from Kunita theory of stochastic flow that θ → grx,t (θ )
is a C∞ diffeomorphism. Then it is possible to prove that the limit limr→1 grx,t =
gx,t exists uniformly in θ and defines a random homeomorphism gx,t which is
the canonical Brownian motion “on” Diff(S1 ). This random homeomorphism
is furthermore Hölderian. See [A-R], [F].
The methodology used in [A-M-T] takes seventy pages; in one line it can be
summarized by saying: the non linear PDE of conformal geometry, called the
Ahlfors–Beltrami equation, is solved by the continuity method along a stochastic
flow of diffeomorphism of the complex plane.
Note added February 19, 2008. The references [K-S], [C-M08], which appeared
after the Takagi Lecture, contain new developments which are not reported in
this article.

References

[A] H. Airault, Stochastic analysis on finite dimensional Siegel disks, approach to the infi-
nite dimensional Siegel disk and upper half-plane, Bull. Sci. Math., 128 (2004), 605–
659.
[A-M00] H. Airault and P. Malliavin, Regularized Brownian motion on the Siegel disk of infinite
dimension, Ukrain. Mat. Zh., 52 (2000), 1158–1165.
[A-M01] H. Airault and P. Malliavin, Unitarizing probability measures for representations of
Virasoro algebra, J. Math. Pures Appl. (9), 80 (2001), 627–667.
[A-M06] H. Airault and P. Malliavin, Quasi-invariance of Brownian measures on the group of
circle homemorphisms and infinite-dimensional Riemannian geometry, J. Funct. Anal.,
241 (2006), 99–142.
[A-M-T] H. Airault, P. Malliavin and A. Thalmaier, Canonical Brownian motion on the space of
univalent functions and resolution of Beltrami eqautions by a continuity method along
stochastic flows, J. Math. Pures Appl. (9), 83 (2004), 955–1018.
[A-N] H. Airault and Yu. Neretin, On the action of Virasoro algebra on the space of univalent
functions, Bull. Sci. Math., 132 (2008), 27–39.
[A-R] H. Airault and J. Ren, Modulus of continuity for the canonic Brownian motion on the
group of diffeomorphisms of the circle, J. Funct. Anal., 196 (2002), 395–426.
[B] F.A. Berezin, Quantization in complex symmetric spaces, Izv. Akad. Nauk SSSR Ser.
Mat., 39 (1975), 363–402.
[C-M98] A.B. Cruzeiro and P. Malliavin, Nonperturbative construction of invariant measure
through confinement by curvature, J. Math. Pures Appl. (9), 77 (1998), 527–537.
[C-M08] A.B. Cruzeiro and P. Malliavin, Stochastic calculus of variations on complex line bun-
dle and construction of unitarizing measures for Poincaré disk, J. Funct. Anal., 2008.
[F] S. Fang, Solving stochastic differential equations on Homeo(S1 ), J. Funct. Anal., 216
(2004), 22–46.
[H-R] D.K. Hong and S.G. Rajeev, Universal Teichmuller space and Diff(S1 )/S1 , Comm.
Math. Phys., 135 (1991), 401–411.
[I] K. Itô, The Brownian motion and tensor fields on Riemannian manifolds, In: Proc.
Internat. Congr. Mathematicians, Stockholm, 1962, Inst. Mittag-Leffler, Djursholm,
1963, pp. 536–539.
Invariant or quasi-invariant probability measures for infinite dimensional groups: Part II 47

[Ki] A.A. Kirillov, Geometric approach to discrete series of unirreps for Vir., J. Math. Pures
Appl. (9), 77 (1998), 735–746.
[K-S] M. Kontsevich and Y. Suhov, On Malliavin measures, SLE, and CFT, Proc. Steklov
Inst. Math., 258 (2007), 100–146.
[Ki-Y] A.A. Kirillov and D.V. Yur’ev, Kähler geometry of the infinite-dimensional homoge-
neous manifold M=Diff+ (S1 )/Rot(S1 ), Funct. Anal. Appl., 20 (1986), 322–324.
[M99] P. Malliavin, The canonic diffusion above the diffeomorphism group of the circle, C.
R. Acad. Paris Sér. I Math., 329 (1999), 325–329.
[N] Yu. Neretin, Representations of Virasoro and affine Lie algebras, Encyclopedia Math.
Sci., 22 (1994), 157–234.
[W] A. Weil, Sur les groupes topologiques et les groupes mesurés, C. R. Math. Acad. Sci.
Paris, 202 (1936), 1147–1149.