Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces
This paper is a sequel to [Caine A., Pickrell D., Int. Math. Res. Not., to appear, arXiv:0710.4484], where we studied the Hamiltonian systems which arise from the Evens-Lu construction of homogeneous Poisson structures on both compact and noncompact type symmetric spaces. In this paper we consider l...
Збережено в:
| Дата: | 2008 |
|---|---|
| Автор: | |
| Формат: | Стаття |
| Мова: | English |
| Опубліковано: |
Інститут математики НАН України
2008
|
| Назва видання: | Symmetry, Integrability and Geometry: Methods and Applications |
| Онлайн доступ: | http://dspace.nbuv.gov.ua/handle/123456789/149016 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Цитувати: | Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces / D. Pickrell // Symmetry, Integrability and Geometry: Methods and Applications. — 2008. — Т. 4. — Бібліогр.: 12 назв. — англ. |
Репозитарії
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
irk-123456789-149016 |
|---|---|
| record_format |
dspace |
| spelling |
irk-123456789-1490162019-02-20T01:25:10Z Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces Pickrell, D. This paper is a sequel to [Caine A., Pickrell D., Int. Math. Res. Not., to appear, arXiv:0710.4484], where we studied the Hamiltonian systems which arise from the Evens-Lu construction of homogeneous Poisson structures on both compact and noncompact type symmetric spaces. In this paper we consider loop space analogues. Many of the results extend in a relatively routine way to the loop space setting, but new issues emerge. The main point of this paper is to spell out the meaning of the results, especially in the SU(2) case. Applications include integral formulas and factorizations for Toeplitz determinants. 2008 Article Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces / D. Pickrell // Symmetry, Integrability and Geometry: Methods and Applications. — 2008. — Т. 4. — Бібліогр.: 12 назв. — англ. 1815-0659 2000 Mathematics Subject Classification: 22E67; 53D17; 53D20 http://dspace.nbuv.gov.ua/handle/123456789/149016 en Symmetry, Integrability and Geometry: Methods and Applications Інститут математики НАН України |
| institution |
Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| collection |
DSpace DC |
| language |
English |
| description |
This paper is a sequel to [Caine A., Pickrell D., Int. Math. Res. Not., to appear, arXiv:0710.4484], where we studied the Hamiltonian systems which arise from the Evens-Lu construction of homogeneous Poisson structures on both compact and noncompact type symmetric spaces. In this paper we consider loop space analogues. Many of the results extend in a relatively routine way to the loop space setting, but new issues emerge. The main point of this paper is to spell out the meaning of the results, especially in the SU(2) case. Applications include integral formulas and factorizations for Toeplitz determinants. |
| format |
Article |
| author |
Pickrell, D. |
| spellingShingle |
Pickrell, D. Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces Symmetry, Integrability and Geometry: Methods and Applications |
| author_facet |
Pickrell, D. |
| author_sort |
Pickrell, D. |
| title |
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces |
| title_short |
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces |
| title_full |
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces |
| title_fullStr |
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces |
| title_full_unstemmed |
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces |
| title_sort |
homogeneous poisson structures on loop spaces of symmetric spaces |
| publisher |
Інститут математики НАН України |
| publishDate |
2008 |
| url |
http://dspace.nbuv.gov.ua/handle/123456789/149016 |
| citation_txt |
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces / D. Pickrell // Symmetry, Integrability and Geometry: Methods and Applications. — 2008. — Т. 4. — Бібліогр.: 12 назв. — англ. |
| series |
Symmetry, Integrability and Geometry: Methods and Applications |
| work_keys_str_mv |
AT pickrelld homogeneouspoissonstructuresonloopspacesofsymmetricspaces |
| first_indexed |
2025-07-12T20:53:02Z |
| last_indexed |
2025-07-12T20:53:02Z |
| _version_ |
1837475916683411456 |
| fulltext |
Symmetry, Integrability and Geometry: Methods and Applications SIGMA 4 (2008), 069, 33 pages
Homogeneous Poisson Structures
on Loop Spaces of Symmetric Spaces?
Doug PICKRELL
Department of Mathematics, University of Arizona, Tucson, AZ, 85721, USA
E-mail: pickrell@math.arizona.edu
Received June 14, 2008, in final form September 27, 2008; Published online October 07, 2008
Original article is available at http://www.emis.de/journals/SIGMA/2008/069/
Abstract. This paper is a sequel to [Caine A., Pickrell D., Int. Math. Res. Not., to appear,
arXiv:0710.4484], where we studied the Hamiltonian systems which arise from the Evens–
Lu construction of homogeneous Poisson structures on both compact and noncompact type
symmetric spaces. In this paper we consider loop space analogues. Many of the results
extend in a relatively routine way to the loop space setting, but new issues emerge. The
main point of this paper is to spell out the meaning of the results, especially in the SU(2)
case. Applications include integral formulas and factorizations for Toeplitz determinants.
Key words: Poisson structure; loop space; symmetric space; Toeplitz determinant
2000 Mathematics Subject Classification: 22E67; 53D17; 53D20
1 Introduction
The first purpose of this paper is to generalize the framework in [3] to loop spaces. This
generalization is straightforward, using the fundamental insight of Kac and Moody that finite
dimensional complex semisimple Lie algebras and (centrally extended) loop algebras fit into the
common framework of Kac–Moody Lie algebras.
Suppose that Ẋ is a simply connected compact symmetric space with a fixed basepoint.
From this, as we will more fully explain in Sections 2 and 3, we obtain a diagram of groups
G = L̂Ġ
↗ ↖
G0 = L̂Ġ0 U = L̂U̇
↖ ↗
K = L̂K̇
(1.1)
where U̇ is the universal covering of the identity component of the isometry group of Ẋ, Ẋ '
U̇/K̇, Ġ is the complexification of U̇ , Ẋ0 = Ġ0/K̇ is the noncompact type symmetric space dual
to Ẋ, LĠ denotes the loop group of Ġ, L̂Ġ denotes a Kac–Moody extension, and so on. This
diagram is a prolongation of diagram (0.1) in [3] (which is embedded in (1.1) by considering
constant loops).
We also obtain a diagram of equivariant totally geodesic (Cartan) embeddings of symmetric
spaces:
L(U̇/K̇)
φ→ L̂U̇
↓ ↓
L̃Ġ/L̃Ġ0
φ→ L̂Ġ
ψ← L̃Ġ/L̃U̇
↑ ↑
L̂Ġ0
ψ← L(Ġ0/K̇)
(1.2)
?This paper is a contribution to the Special Issue on Kac–Moody Algebras and Applications. The full collection
is available at http://www.emis.de/journals/SIGMA/Kac-Moody algebras.html
mailto:pickrell@math.arizona.edu
http://www.emis.de/journals/SIGMA/2008/069/
http://arxiv.org/abs/0710.4484
http://www.emis.de/journals/SIGMA/Kac-Moody_algebras.html
2 D. Pickrell
This is a prolongation of diagram (0.2) in [3].
Let Θ denote the involution corresponding to the pair (U̇ , K̇). We consider one additional
ingredient: a triangular decomposition
ġ = ṅ− ⊕ ḣ⊕ ṅ+ (1.3)
which is Θ-stable and for which ṫ0 = ḣ ∩ k̇ is maximal abelian in k̇. There is a corresponding
Kac–Moody triangular decomposition
L̂polġ =
(⊕
n<0
ġzn ⊕ ṅ−
)
⊕ h⊕
(
ṅ+ ⊕
⊕
n>0
gzn
)
extending (1.3).
This data determines standard Poisson Lie group structures, denoted πU and πG0 , for the
groups U = L̂U̇ and G0 = L̂Ġ0, respectively. By a general construction of Evens and Lu [4],
the symmetric spaces X = LẊ and X0 = LẊ0 acquire Poisson structures ΠX and ΠX0 , respec-
tively, which are homogeneous for the respective actions of the Poisson Lie groups (U, πU ) and
(G0, πG0). These spaces are infinite dimensional, and there are many subtleties associated with
Poisson structures in infinite dimensions (see [8]). Consequently in this paper we will always
display explicit decompositions and formulas, and we will avoid any appeal to general theory
(for “symplectic foliations”, for example).
The plan of this paper is the following. In Section 2 we introduce notation and recall some
well-known facts concerning loop algebras and groups.
In Section 3 we consider the case when Ẋ is an irreducible type I space. All of the results
of Sections 1–4 of [3] generalize in a relatively straightforward way to the loop context roughly
outlined above. The basic result is that ΠX0 has just one type of symplectic leaf, this leaf is
Hamiltonian with respect to the natural action of T0, there are relatively explicit formulas for this
Hamiltonian system, and in a natural way, this system is isomorphic to the generic Hamiltonian
system for ΠX . Although this system is infinite dimensional, a heuristic application of the
Duistermaat–Heckman exact stationary phase theorem to this system suggests some remarkable
integral formulas. This is discussed in Section 7 of [9]. These formulas remain conjectural.
In Section 4 I have attempted to do some calculations in the Ẋ = S2 case. The formulas are
complicated; I included them to give the reader a concrete feeling for the subject.
In Section 5, and Appendix A, we consider the group case. Again, the results of Sections
1–4 of [3] generalize in a straightforward way. However significant issues emerge when we try
to generalize the results of Section 5 of [3]. In the finite dimensional context of Ẋ = K̇, the
(negative of the) standard Poisson Lie group structure (K̇, πK̇) is isomorphic to (Ẋ,ΠẊ), by
left translation by a representative for the longest Weyl group element. In the loop context the
Poisson Lie group and Evens–Lu structures are fundamentally different: the symplectic leaves
for πK (essentially Bruhat cells) are finite dimensional, whereas the symplectic leaves for ΠX
(essentially Birkhoff strata) are finite codimensional. In finite dimensions Lu has completely
factored the symplectic leaves. Lu’s results, as formulated in [7] in terms of πK , do generalize
in a relatively straightforward way to the loop context. Some details of this generalization are
worked out in Appendix A, where we have extended this to the larger category of symmetrizable
Kac–Moody algebras.
The basic question is whether the Hamiltonian systems for ΠX , in this infinite dimensional
context, are solvable (in a number of senses). In Sections 5–7 we show that there is a natural
way to conjecturally reformulate and extend Lu’s results to suggest that the generic symplectic
leaves are integrable. However we have not succeeded in fully proving this conjecture: while we
can factor the momentum mapping, and the Haar measure relevant for Theorem 1 below, we
have not shown that the symplectic form Π−1
X factors.
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 3
In Section 6 we spell out the meaning of the results in Section 5 when K̇ = SU(2). One
consequence is the following integral formula.
Theorem 1. Given xj ∈ C, let
B
n∑
j=1
xjz
j
=
xn 0 . . . 0
xn−1 xn 0 . . . 0
...
. . . . . .
...
x2 . . . xn−1 xn 0
x1 x2 . . . xn−1 xn
. (1.4)
Then ∫ n−1∏
l=0
det
1 +B
n−l∑
j=1
xl+jz
j
B
n−l∑
j=1
(
xl+jz
j
)∗−pl
dλ(x1, . . . , xn)
= πn
1
(p1 − 1)
1
(2p1 + p2 − 3)
· · · 1
(np1 + (n− 1)p2 + · · ·+ pn − (2n− 1))
.
In particular, if we write Bn(x) for the matrix (1.4), for a general power series x =
∑
xjz
j,
then
1
det(1 +Bn(x)Bn(x)∗)p
dλ(x1, . . . , xn) (1.5)
is a finite measure if and only if p > 2− 1/n.
This result is important because it determines the critical exponents for the integrands in (1.5)
exactly, whereas I am not aware of any other way to even estimate these exponents in a useful
way. The relevance of this to the theory of conformally invariant measures, where one must
understand the limit as n→∞, is described in [10].
In Section 7 we consider the question of global solvability of the symplectic leaves, in the
SU(2) case. A consequence of the global factorization of the momentum mapping is the following
illustrative statement about block Toeplitz operators.
Theorem 2. Given complex numbers ηj, χj, ζj, let g : S1 → SU(2) be the product of SU(2)
loops
a(η0)
(
1 η0
−η̄0 1
)
· · · a(ηn)
(
1 ηnz
n
−η̄nz−n 1
)(
e
∑
χjz
j
0
0 e−
∑
χjz
j
)
,
a(ζn)
(
1 ζnz
−n
−ζ̄nzn 1
)
· · · a(ζ1)
(
1 ζ1z
−1
−ζ̄1z 1
)
,
where a(·) = (1 + | · |2)−1/2 and χ−j = −χ̄j. Let A(g) denote the Toeplitz operator defined by
the symbol g. Then
det(A(g)A(g)∗) =
∏
j
a(ηj)2ja(ζj)2je−|j||χj |2 .
When η and ζ vanish, this reduces to a well-known formula with a long history (e.g. see
Theorem 7.1 of [12]).
In Section 7, because the SU(2) loop space is infinite dimensional, it is necessary to take
a limit as n → ∞, so that the above product of loops is to be interpreted as an infinite
factorization of a generic g ∈ LSU(2). At a heuristic level, the invariant measures considered
in [9] factor in these coordinates. The conjectural integral formulas in Section 7 of [9] (in the
SU(2) case) follow immediately from this product structure. However changing coordinates in
infinite dimensions is nontrivial, and probabilistic analysis is required to justify this claim.
4 D. Pickrell
2 Loop groups
In this section we recall how (extended) loop algebras fit into the framework of Kac–Moody
Lie algebras. The relevant structure theory for loop groups is developed in [11], and for loop
algebras in Chapter 7 of [5].
Let U̇ denote a simply connected compact Lie group. To simplify the exposition, we will
assume that u̇ is a simple Lie algebra. Let Ġ and ġ denote the complexifications, and fix
a u̇-compatible triangular decomposition
ġ = ṅ− ⊕ ḣ⊕ ṅ+. (2.1)
We let 〈·, ·〉 denote the unique Ad(Ġ)-invariant symmetric bilinear form such that (for the dual
form) 〈θ, θ〉 = 2, where θ denotes the highest root for ġ, i.e. 〈·, ·〉 = 1
ġκ, where κ denotes the
Killing form, and ġ is the dual Coxeter number.
Let L̂ġ denote the real analytic completion of the untwisted affine Lie algebra corresponding
to ġ, with derivation included (the degree of smoothness of loops is essentially irrelevant for the
purposes of this discussion; any fixed degree of Sobolev smoothness s > 1/2 would work equally
well). This is defined in the following way. We first consider the universal central extension of
Lġ = Cω(S1, ġ),
0→ Cc→ L̃ġ→ Lġ→ 0.
As a vector space L̃ġ = Lġ⊕ Cc. In these coordinates, the L̃ġ-bracket is given by
[X + λc, Y + λ′c]L̃ġ = [X,Y ]Lġ +
i
2π
∫
S1
〈X ∧ dY 〉c. (2.2)
Then L̂ġ = Cd ∝ L̃ġ (the semidirect sum), where the derivation d acts by d(X + λc) = 1
i
d
dθX,
for X ∈ Lġ. The algebra generated by u̇-valued loops induces a central extension
0→ iRc→ L̃u̇→ Lu̇→ 0
and a real form L̂u̇ = iRd ∝ L̃u̇ for L̂ġ. We identify ġ with the constant loops in Lġ. Because
the extension is trivial over ġ, there are embeddings of Lie algebras
ġ→ L̃ġ→ L̂ġ.
The Lie algebra L̂ġ has a triangular decomposition
L̂ġ = n− ⊕ h⊕ n+, (2.3)
where
h = ḣ + Cc+ Cd,
n+ =
{
x =
∞∑
0
xnz
n ∈ H0(D; ġ) : x(0) = x0 ∈ ṅ+
}
and
n− =
{
x =
∞∑
0
xnz
−n ∈ H0(D∗; ġ) : x(∞) = x0 ∈ ṅ−
}
.
This is compatible with the finite dimensional triangular decomposition (2.1). We let N± denote
the profinite nilpotent groups corresponding to n±, e.g.
N− = H0(D∗,∞; Ġ, Ṅ−).
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 5
There is a unique Ad-invariant symmetric bilinear form on L̂ġ which extends the normalized
Killing form on ġ. It has the following restriction to h:
〈c1d+ c2c+ h, c′1d+ c′2c+ h′〉 = c1c
′
2 + c2c
′
1 + 〈h, h′〉.
This form is nondegenerate. The restriction of this form to L̂u̇ is also nondegenerate, although
this restriction is of Minkowski type, in contrast to the finite dimensional situation.
The simple roots for (L̂ġ, h) are {αj : 0 ≤ j ≤ rkġ}, where
α0 = d∗ − θ, αj = α̇j , j > 0,
d∗(d) = 1, d∗(c) = 0, d∗(ḣ) = 0, and the α̇j denote the simple roots for the triangular decompo-
sition of ġ (with α̇j vanishing on c and d). The simple coroots of h ⊂ L̂ġ are {hj : 0 ≤ j ≤ rkġ},
where
h0 = c− ḣθ, hj = ḣj , j > 0,
and the {ḣj} are the simple coroots of ġ. For i > 0, the root homomorphism iαi is iα̇i followed
by the inclusion ġ ⊂ L̂ġ. For i = 0
iα0
((
0 0
1 0
))
= eθz
−1, iα0
((
0 1
0 0
))
= e−θz, (2.4)
where {e−θ, ḣθ, eθ} satisfy the sl(2,C)-commutation relations, and eθ is a highest root for ġ.
Let Λj denote the fundamental dominant integral functionals on h. Any linear function λ
on h can be written uniquely as λ = λ̇+λ(h0)Λ0, where λ̇ can be identified with a linear function
on ḣ. In particular δ, the sum of the fundamental dominant integral functionals, is given by
δ = δ̇ + ġΛ0, where δ̇ is the sum of the fundamental dominant integral functionals for the finite
dimensional triangular structure (2.1).
For g̃ ∈ N− ·H ·N+ ⊂ L̃G,
g̃ = l · (diag)̃ · u, where (diag)̃(g̃) =
rkġ∏
0
σj(g̃)hj ,
where σj = σΛj is the matrix coefficient corresponding to Λj . If g̃ projects to g ∈ N− · Ḣ ·N+ ⊂
LG, then because σh0
0 = σc−ḣθ
0 projects to σ−ḣθ
0 , we have g = l · diag · u, where
diag(g) = σ0(g̃)−ḣθ
rkġ∏
1
σj(g̃)ḣj =
rkġ∏
1
(
σj(g̃)
σ0(g̃)ǎj
)ḣj
, (2.5)
and the ǎj are positive integers such that ḣθ =
∑
ǎj ḣj .
If g̃ ∈ L̃K, then |σj(g̃)| depends only on g, the projection of g̃ in LK. We will indicate this
by writing
|σj(g̃)| = |σj |(g). (2.6)
In this paper we will mainly deal with generic elements in L̃K having diagonal elements with
trivial T -component. Thus (2.6) has the practical consequence (important in Sections 6 and 7)
that we can generally work with ordinary loops in K. We record this for later reference.
Lemma 1. The restriction of the projection L̃K → LK to generic elements with diagonal terms
having trivial T -component is injective.
6 D. Pickrell
3 Type I case
In this section we assume that Ẋ is a type I simply connected and irreducible symmetric space.
We let U̇ denote the universal covering of the identity component of the group of automorphisms
of Ẋ, and so on, as in the Introduction. The irreduciblity and type I conditions imply that u̇
and ġ are simple Lie algebras.
Exactly as in the preceding section, we introduce the affine analogues g = L̂ġ and u = L̂u̇
of ġ and its compact real form u̇, respectively, and also the corresponding groups. We will write
the corresponding Lie algebra involution as −(·)∗, as we typically would in a finite dimensional
matrix context.
Let Θ denote the involution corresponding to the pair (u̇, k̇). We extend Θ complex linearly
to ġ, and we use the same symbol to denote the involution for the Lie group Ġ. We assume that
the triangular decomposition of the preceding section is Θ-stable. We extend Θ to an involution
of Lġ pointwise, and we then extend Θ to L̂ġ by
Θ(µd+ x+ λc) = µd+ Θ(x) + λc.
The triangular decomposition for L̂ġ is Θ-stable, and t0 = h ∩ L̂k̇ is maximal abelian in L̂k̇. We
let σ denote the Lie algebra involution −(·)∗Θ, we use the same symbol for the corresponding
group involution, and we let g0 = L̂ġ0 and G0 = L̂Ġ0 denote the corresponding real forms.
We have defined the various objects in the diagram (1.1). The Lie algebra analogue of the
diagram (1.1) is given by
L̂ġ = L̂u̇⊕ iL̂u̇
↗ ↖
L̂ġ0 = L̂k̇⊕ Lṗ L̂u̇ = L̂k̇⊕ iLṗ
↖ ↗
L̂k̇
where L̂k̇ = iRd ∝ L̃k̇ and L̃k̇ = Lk̇⊕ iRc. The sums in the diagram represent Cartan decompo-
sitions. In analogy with [3], we will write p = Lṗ, h0 = h ∩ g0 = t0 ⊕ a0 (relative to the Cartan
decomposition for g0), and t = h ∩ u = t0 ⊕ ia0.
Our next task is to explain the diagram (1.2). There are isomorphisms induced by natural
maps
L̂U̇/L̂K̇ → L̃U̇/L̃K̇ → LU̇/LK̇ → LẊ, (3.1)
and
L̂Ġ0/L̂K̇ → L̃Ġ0/L̃K̇ → LĠ0/LK̇ → LẊ0. (3.2)
In each case the first two maps are obviously isomorphisms. In the first and second cases the
third map is an isomorphism because Ẋ and Ẋ0 are simply connected, respectively.
We will take full advantage of these isomorphisms, and consequently there will be times
when we want to use the quotient involving hats, or tildes, and times when we want to use the
quotient not involving hats, or tildes. To distinguish when we are using hats, we will write our
group elements with hats, and similarly with tildes. Thus ĝ will typically denote an element
of L̂Ġ, whereas g will typically denote an element of LĠ, and unless stated otherwise, these two
elements will be related by projection.
For the natural maps
L̂Ġ/L̂Ġ0 → L̃Ġ/L̃Ġ0 → LĠ/LĠ0 → L(Ġ/Ġ0) (3.3)
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 7
and
L̂Ġ/L̂U̇ → L̃Ġ/L̃U̇ → LĠ/LU̇ → L(Ġ/U̇)
in each case the third map is an isomorphism, but the first two maps fail to be isomorphisms.
For example in (3.3) the second map is surjective, but there is a nontrivial fiber exp(Rc)L̃Ġ0
over the basepoint (represented by 1). This is the reason for the appearance of L̃Ġ/L̃Ġ0, rather
than L(Ġ/Ġ0), in the diagram (1.2).
There is an Iwasawa decomposition for L̂Ġ (see Chapter 8 of [11]), which we write as
L̂Ġ ' N− ×A× L̂U̇ : ĝ = l(ĝ)a(ĝ)u(ĝ), (3.4)
where A = exp(hR). In analogy with [3], we also write a = a0a1, relative to exp(hR) =
exp(a0) exp(it0). There is an induced right action
L̂U̇ × (T̂ × L̂Ġ0)→ L̂U̇ : (û, t̂, ĝ0)→ t̂−1u(ûĝ0) (3.5)
arising from the identification of L̂U̇ with N−A\L̂Ġ. We also write A0 = A ∩G0.
The Cartan embedding for the unitary type symmetric space is given by
φ : L(U̇/K̇)→ L̃U̇ ⊂ L̂U̇ : ũL̃K̇ → ũũ−Θ,
where we are using the isomorphism (3.1) in an essential way to express this mapping. There is
a corresponding embedding ψ in the dual case. More generally
φ : L̃Ġ/L̃Ġ0 → L̃Ġ ⊂ L̂Ġ : g̃L̃Ġ0 → g̃g̃∗Θ,
and the extension of ψ is similarly defined.
This explains the diagram (1.2). We should note that in what follows, in place of (1.1)
and (1.2), and the Kac–Moody triangular decomposition (2.3) for g, we could simply consider
the ordinary loop functor of the diagrams (0.1) and (0.2) of [3], and the analogue of the triangular
decomposition for Lġ. But in the process we would miss out on the interesting applications (such
as Theorem 1), and in analyzing the resulting Hamiltonian systems we would inevitably be led
to this Kac–Moody extended point of view.
We are now in a position to repeat verbatim the arguments in Sections 2–4 of [3], supple-
mented with remarks concerning Poisson structures in infinite dimensions. We will summarize
the main points.
Proposition 1. Relative to the extended real form Im〈·, ·〉 on g = L̂ġ,
(g, u, hR ⊕ n−) and (g, g0, t⊕ n−)
are Manin triples, extending the finite dimensional Manin triples (ġ, u̇, ḣR⊕ṅ−) and (ġ, ġ0, ṫ⊕ṅ−),
respectively.
We next apply the Evens–Lu construction to obtain global Poisson structures ΠX and ΠX0 on
the loop spaces X = LẊ and X0 = LẊ0, respectively, using the isomorphisms (3.1) and (3.2).
These Poisson structures are given by the same formulas as in the finite dimensional cases:
see (3.1) and (4.1) of [3]. As in the finite dimensional case, we have used the Ad-invariant
symmetric form on L̂ġ to identify p with a subspace of its dual (note the form is definite on p).
However, in this infinite dimensional context, the inclusion p→ p∗ is proper, so that this Poisson
structure must be understood in a weak sense. Consequently it is not clear that we can appeal
to any general theory (e.g. as in [8]) for the existence of a symplectic foliation, etc.
8 D. Pickrell
As in [3], the Hilbert transform H : g→ g associated to the triangular decomposition of g is
given by
x = x− + x0 + x+ 7→ H(x) = −ix− + ix+.
In the following statement, we can, and do, view a0 (defined following (3.4)) as a function
on X = G0/K.
Theorem 3.
(a) The Poisson structure ΠX0 has a regular symplectic foliation (by weak symplectic mani-
folds), given by the level sets of the function a0.
(b) The horizontal parameterization for the symplectic leaf through the basepoint is given by
the map s : A0\G0/K → G0/K
A0g0K → s(A0g0K) = a−1
0 g0K,
where g0 = la0a1u.
(c) If we identify T (G0/K) with G0 ×K p in the usual way, then
ω1([g0, x] ∧ [g0, y]) = 〈Ad
(
u(g0)−1
)
◦ H ◦Ad(u(g0))(x), y〉 (3.6)
is a well-defined two-form on G0/K.
(d) Along the symplectic leaves, Π−1
X0
agrees with the restriction of the closed two-form ω1.
Note that the facts that the form ω1 is closed and nondegenerate (on the double coset space
A0\G0/K → G0/K) is proven directly in Section 1 of [3].
Theorem 4.
(a) The Poisson structure ΠX has a symplectic foliation (by weak symplectic manifolds). The
symplectic leaves are identical to the projections of the LĠ0-orbits, for LĠ0 acting on LU̇ as
in (3.5), to L(U̇/K̇). Let S(1) denote the symplectic leaf containing the identity.
(b) The action of T̂0 = Rot(S1)×T0×exp(iRc) on Σφ(L(U̇/K̇))
1 is Hamiltonian with momentum
mapping
Σφ(L(U̇/K̇))
1 → (̂t0)∗ : ũ→ 〈− i
2
log(aφ(ũ), ·〉,
where ũ has the unique triangular decomposition ũ = lmãφl
∗Θ.
(c) The map ũ : G0 → U
g0 7→ u(g0),
where u is defined by (3.4), is equivariant for the right actions of K on G0 and U , invariant
under the left action of A0 on G0 and descends to a T0-equivariant diffeomorphism
ũ : A0\G0/K → S(1).
This induces an isomorphism of T0-Hamiltonian spaces
(A0\G0/K, ω1)→
(
S(1),Π−1
X
)
,
where ω1 is as in (3.6).
The symplectic foliation in part (a) can be described in a completely explicit way in terms
of triangular factorization and the Cartan embedding φ (see [2] for the finite dimensional case;
the arguments there extend directly).
Throughout this paper we will focus on the generic system S(1) in part (c). As we mentioned
in the Introduction, the main application which we envision is to use this Hamiltonian system to
generate useful integral formulas. In this loop context these integrals are infinite dimensional,
and more infrastructure and analysis are required to properly formulate and justify them (see [9],
especially Section 7). Even in finite dimensions, it is not known whether these type I systems
have any integrability properties (in sharp contrast to the type II case).
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 9
4 The S2 case
In this section we will do some illustrative calculations in the simplest Type I case
G = L̂SL(2,C)
↗ ↖
G0 = L̂SU(1, 1) U = L̂SU(2)
↖ ↗
K = L̂U(1)
If we identify Ẋ0 with ∆ (the unit disk) and Ẋ with Ĉ in the usual way, then from the preceding
section we have maps
L∆ ũ→ LĈ φ→ L̃SU(2), (4.1)
where the map ũ is covered by the map
L̂SU(1, 1) u→ L̂SU(2)
induced by the Iwasawa decomposition g0 = l(g0)a(g0)u(g0).
To orient the reader, we recall the nonloop case:
SU(1, 1) u→ SU(2)
↓ ↓
∆ → Ĉ φ→ SU(2)
g0 = 1
(1−ZZ̄)1/2
(
1 Z̄
Z 1
)
→ u(g0) = 1
(1+ZZ̄)1/2
(
1 Z̄
−Z 1
)
↓ ↓
Z → −Z → 1
1+|Z|2
(
1− |Z|2 2Z̄
−2Z 1− |Z|2
)
In this context u is obtained by a Gram–Schmidt process from the rows of g0, and
a =
(
1 + ZZ̄
1− ZZ̄
) 1
2
h1
.
To calculate the symplectic form note that[
g0, X =
(
0 x̄
x 0
)]
→ d
dt
∣∣∣
t=0
Z
(
g0e
tX
)
=
d
dt
∣∣∣
t=0
(
Z ch (tx) + sh (tx)
)(
ch (tx) + Z̄ sh (tx)
)−1 = (1− ZZ̄)x.
Thus a variation Ż of Z will correspond to [g0, X] with x = (1− ZZ̄)−1Ż. Thus
ω([g0, X] ∧ [g0, Y ])
= ω
([
1
(1− ZZ̄)1/2
(
1 Z̄
Z 1
)
,
(
0 x̄
x 0
)]
∧
[
1
(1− ZZ̄)1/2
(
1 Z̄
Z 1
)
,
(
0 ȳ
y 0
)])
= 〈H(Ad
(
1
(1 + ZZ̄)1/2
(
1 Z̄
−Z 1
))((
0 x̄
x 0
))
∧Ad
(
1
(1 + ZZ̄)1/2
(
1 Z̄
−Z 1
))((
0 ȳ
y 0
))
〉 =
i
(1− |Z|4)
( ¯̇ZZ ′ − ŻZ̄ ′
)
.
10 D. Pickrell
Thus
ω =
i
(1− |Z|4)
dZ ∧ dZ̄.
Returning to the loop case, we denote the maps in (4.1) by
f(θ)→ F (θ)→ φ(F )(θ).
We have written the argument as θ, as a reminder that these are functions on S1. To calculate
the map f → F , we need to find the Iwasawa decomposition
g0(θ) =
1
(1− f(θ)f̄(θ))1/2
(
1 f̄(θ)
f(θ) 1
)
= l(z)au(θ),
and remember that l(z) extends to a holomorphic function in the exterior of S1. In turn
φ(F )(θ) = uu∗Θ = l(z)au(z),
where l = a−1l−1a, a = a−2 = |σ0|h0 |σ1|h1 , u = l∗Θ. The image of φ(F ) in LSU(2) has the form(
α(θ) β(θ)
−β̄(θ) ᾱ(θ)
)
= l
(
z−1
)( |σ1|
|σ0|
)h1
u(z)
(see (2.5)).
The Iwasawa decomposition of g0 (the special self-adjoint representative above) is equivalent
to
g∗0l(g0)
−∗a(g0)−1 = g0l(g0)−∗a(g0)−1 = u(g0).
Write l−∗ =
(
a b
c d
)
, so that a, b, c, d are holomorphic functions in D, a(0) = d(0) = 1, and
c(0) = 0. Then(
1 f̄
f 1
)(
a b
c d
)(
a−1
0 0
0 a0
)
is of the form
(
A B
−B̄ Ā
)
. This implies
a+ f̄ c =
(
f̄ b̄+ d̄
)
a2
0, fa+ c = −
(
b̄+ fd̄
)
a2
0. (4.2)
As a reminder, these are equations for functions defined on S1.
Let H0 = P+ − P−, where for a scalar function g =
∑
gnz
n, P+g =
∑
n≥0
gnz
n. We take the
conjugate of the first equation in (4.2) and rewrite it as
−H0
(
ā+ da2
0
)
+ 2 = fH0
(
ba2
0 + c̄
)
.
This is equivalent to
−
(
ā+ da2
0
)
+ 2 = H0fH0
(
ba2
0 + c̄
)
.
The second equation in (4.2) is equivalent to
f̄
(
ā+ da2
0
)
= −
(
ba2
0 + c̄
)
.
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 11
These two equations imply(
ba2
0 + c̄
)
= −2(1− f̄H0fH0)−1(f̄) = −2f̄(1−H0fH0f̄)−1(1).
Note that the inverse on the right exists, because sup{|f(z)| : z ∈ S1} < 1. This determines ba2
0
and c, by applying P±.
We now see that
ā+ da2
0 = 2
(
1 +H0fH0f̄(1−H0fH0f̄)−1(1)
)
= 2(1−H0fH0f̄)−1(1).
Note that 1 + a2
0 is the zero mode of the right hand side, so that in principle we have determi-
ned a0, and l. This form of the solution does not explain in a clear way why the zero mode of
the right hand side is > 2.
To summarize, let
h = 2(1−H0fH0f̄)−1(1)
(this is a well-defined function on S1, and we do not know much more about it). Then
c̄ = −P−(f̄h), ba2
0 = −P+(f̄h), ā− 1 = P−(h), 1 + da2
0 = P+(h).
This implies the following
Proposition 2.
u = g0l−∗a−1 = (1− ff̄)−1/2a−1
0
(
(1 + P−h− fP−(f̄h))∗ −(f̄ + f̄P−h− P−(f̄h))
(f̄ + f̄P−h− P−(f̄h))∗ 1 + P−h− fP−(f̄h)
)
and
F =
(
f̄ + f̄P−h− P−(f̄h)
1 + P−h− fP−(f̄h)
)∗
∈ LĈ.
These general formulas are not especially enlightening. However, the SU(2) case considered
below suggests that there might be some special cases of these formulas which are tractable.
5 Type II case
In the type II case there is more than one reasonable interpretation of the diagram (1.1). The
differences between the possibilities are minor, but potentially confusing. We will briefly describe
a first possibility, which leads to diagram (1.2), but we will then consider a second possibility,
which is more elementary in a technical sense, and we will pursue this in detail.
Throughout this section K̇ denotes a simply connected compact Lie group with simple Lie
algebra k̇, Ẋ = K̇, viewed as a symmetric space, U̇ = K̇ × K̇, and ġ = k̇C ⊕ k̇C.
In the first interpretation of diagram (1.1), g = L̂ġ is defined in the following way. We first
define a central extension
0→ Cc→ L̃ġ→ Lġ→ 0.
As a vector space
L̃ġ = Lġ⊕ Cc;
12 D. Pickrell
the bracket is defined as in (2.2), where the form 〈·, ·〉 is the sum of the normalized invariant
symmetric forms for the two k̇C factors:
〈(x, y), (X,Y )〉 = 〈x, y〉+ 〈X,Y 〉.
Then L̂ġ = Cd ∝ L̃ġ, and
Θ(λd+ (x, y) + µc) = λd+ (y, x) + µc.
The Lie algebra analogue of diagram (1.1) is
g = L̂ġ
↗ ↖
L̂k̇⊕ {(x,−x) : x ∈ iLk̇} L̂k̇⊕ {(x,−x) : x ∈ Lk̇}
↖ ↗
L̂k̇
where L̂k̇ = iRd ∝ L̃k̇ and L̃k̇ = {(x, x) : x ∈ Lk̇} ⊕ iRc.
At the group level G = L̂Ġ = C ∝ L̃Ġ where L̃Ġ is an extension of LĠ by C∗; precisely, L̃Ġ
is a quotient
0→ C∗ → L̃K̇C × L̃K̇C → L̃Ġ→ 0,
where λ ∈ C∗ maps antidiagonally, λ→ (λc, λ−c).
As in the type I case, there are isomorphisms
L̂U̇/L̂K̇ → L̃U̇/L̃K̇ → LU̇/LK̇ → LK̇,
where the last map is given by (k1, k2)→ k1k
−1
2 . The Cartan embedding is given by
φ : X = LK̇ → L̃U̇ ⊂ L̂U̇ : k → ˜(k1, k2) ˜(k1, k2)
−Θ
,
where k = k1k
−1
2 . The dual map ψ is described in a similar way. This leads to the diagram (1.2)
in this type II case.
This first interpretation of diagram (1.1) is somewhat inconvenient, because unlike the finite
dimensional case, X 6= K, and U 6= K × K. In the remainder of this paper we will consider
a setup where these equalities do hold. It will be easier to compare this setup with the finite
dimensional case. The modest price we pay is that, in this second interpretation, X is a covering
of LK̇ (also, as a symmetric space, the invariant geometric structure is of Minkowski type, rather
than Riemannian type, but this geometric structure is irrelevant for our purposes).
From now on, we set K = L̂K̇, as in Section 2. We henceforth understand the diagram (1.1)
to be
G = KC ×KC
↗ ↖
G0 = {g0 = (g, g−∗) : g ∈ KC} U = K ×K
↖ ↗
∆(K) = {(k, k) : k ∈ K}
where ∆(K) = {(k, k) : k ∈ K}, G0 = {g0 = (g, g−∗) : g ∈ KC}, K = L̂K̇, G = L̂Ġ, and the
involution Θ is the outer automorphism Θ((g1, g2)) = (g2, g1). Also
X0 = G0/∆(K) ' KC/K, and X = U/∆(K) ' K,
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 13
where the latter isometry is (k1, k2)∆(K) 7→ k = k1k
−1
2 . As in [3] we will use superchecks to
distinguish structures for kC versus those for g.
We fix a triangular decomposition
ǧ = kC = ň− + ȟ + ň+. (5.1)
This induces a Θ-stable triangular decomposition for g
g = (ň− × ň−)︸ ︷︷ ︸
n−
+(ȟ× ȟ)︸ ︷︷ ︸
h
+(ň+ × ň+)︸ ︷︷ ︸
n+
. (5.2)
Let ǎ = ȟR and ť = iǎ. Then
t0 = {(x, x) : x ∈ ť}, and a0 = {(y,−y) : y ∈ ǎ}.
The standard Poisson Lie group structure on U = K×K induced by the decomposition in (5.2)
is then the product Poisson Lie group structure for the standard Poisson Lie group structure
on K induced by the decomposition (5.1).
Let us denote the Poisson Lie group structure on K by πK and the Evens–Lu homogeneous
Poisson structure on X = K by ΠX . The formal identification of k with its dual via the invariant
form allows us to view the Hilbert transform Ȟ associated to (5.1) as an element of k ∧ k. As
a bivector field
πK = Ȟr − Ȟl,
where Ȟr (resp. Ȟl) denotes the right (resp. left) invariant bivector field on K generated by Ȟ,
whereas ΠK = Ȟr + Ȟl.
Just as in the Type I case, the arguments of Sections 2–4 of [3] apply verbatim. We will focus
on the new issues which arise.
As we pointed out in the Introduction, the first thing to note is that Theorem 5.1 of [3] does
not hold in this context. The symplectic leaves for the Poisson Lie group structure on K are
finite dimensional, whereas the symplectic leaves for the Evens–Lu Poisson structure are finite
codimensional. Thus these structures are fundamentally different.
As in [3], we will write
k(ζ) =
(
1 0
ζ 1
)(
a(ζ) 0
0 a(ζ)−1
)(
1 −ζ̄
0 1
)
, (5.3)
where a(ζ) = (1 + |ζ|2)−1/2. Given a simple positive root γ, iγ : SU(2) 7→ K denotes the root
subgroup inclusion (as in (2.4)), and
rγ = iγ
((
0 i
i 0
))
,
a fixed representative for the corresponding Weyl group reflection.
Conjecture 1. Fix w ∈W .
(a) The submanifold Ň− ∩ w−1Ň+w ⊂ Ň− is Ť -invariant and symplectic.
Fix a representative w for w with minimal factorization w = rn · · · r1, in terms of simple
reflections rj = rγj corresponding to simple positive roots γj. Let wj = rj · · · r1.
(b) The map
Cn → N− ∩ w−1N+w : ζ = (ζn, . . . , ζ1)→ l(ζ),
14 D. Pickrell
where
w−1
n−1iγn(k(ζn))wn−1 · · ·w−1
1 iγ2(k(ζ2))w1iγ1(k(ζ1)) = l(ζ)au
is a diffeomorphism.
(c) In these coordinates the restriction of ω is given by
ω|N−∩w−1N+w =
n∑
j=1
i
〈γj , γj〉
1
(1 + |ζj |2)
dζj ∧ dζ̄j , (5.4)
the momentum map is the restriction of −〈 i2 log(a), ·〉, where
a(k(ζ)) =
n∏
j=1
(
1 + |ζj |2
)− 1
2
w−1
j−1hγjwj−1 ,
and Haar measure (unique up to a constant) is given by
dλN−∩w−1N+w(l) =
n∏
j=1
(
1 + |ζj |2
)δ̌(w−1
j−1hγjwj−1)−1
dλ(ζj)
=
∏
1≤i<j≤n
(
1 + |ζj |2
)−γi(wi−1w
−1
j−1hγjwj−1w
−1
i−1)
dλ(ζj).
where δ̌ =
∑
Λ̌j, the sum of the dominant integral functionals for ǧ, relative to (5.1).
(d) Let Cw denote the symplectic leaf through w, with respect to πK , with the negative of
the induced symplectic structure. Then left translation by w−1 induces a symplectomorphism
from Cw, with its image in (S(1), ω), which is identified with Ň− ∩ w−1Ň+w ⊂ Ň−.
Proposition 3. The following are true:
(1) part (b) of the conjecture;
(2) the formulas for a(k(ζ)) and Haar measure in part (c);
(3) the right hand side of (5.4) equals the image of the symplectic structure for Cw with
respect to the map in part (d); and
(4) the momentum maps for ω and the form in (3) do agree, and are given by the formula
in (c).
Thus the basic open question is whether (d) holds. This is known to be true in the finite
dimensional case (see Theorems 5.1 and 5.2 of [3]).
Proof. The proof uses a number of facts which are recalled in Appendix A. We will freely use the
notation which is used there. One technical point which emerges is that we are currently using K
to denote an extension of the real analytic completion of LK̇ (and similarly for its Lie algebra,
etc), whereas in the Appendix K is the restriction of the extension to the polynomial loop
group LpolK̇. Since all the root homomorphisms map into the extension over the polynomial
loop group, we will simply replace K by this small subgroup, rather than introducing more
notation.
Via the projection,
K → K/Ť = Ǧ/B̌+,
S(1) is identified with Σ1, the unique open Birkhoff stratum in the flag manifold. There is
a surjective map
SL(2,C)× · · · × SL(2,C)→ w−1C̄w :
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 15
(gn, . . . , g1)→ w−1
n−1iγn(gn)wn−1 · · ·w−1
1 iγ2(g2)w1iγ1(g1)B̌+.
This map is obtained by composing the Bott–Samelson desingularization of the Schubert varie-
ty C̄w by left translation by w−1, as in Corollary 3. This map has the remarkable property that
the notion of generic is compatible with factorization: the preimage of w−1Cw is precisely
SL(2,C)′ × · · · × SL(2,C)′,
where
SL(2,C)′ =
{(
a b
c d
)
: a 6= 0
}
.
This follows from Corollary 3.
In terms of the affine coordinate Ň− for Ǧ/B̌+,
Ň− ∩ w−1Ň+w ↔ w−1Cw.
We thus obtain a surjective map
SL(2,C)′ × · · · × SL(2,C)′ → Ň− ∩ w−1Ň+w : (gn, . . . , g1)→ l,
where
w−1
n−1iγn(gn)wn−1 · · ·w−1
1 iγ2(g2)w1iγ1(g1)B̌
+ = lB̌+.
By part (a) of Corollary 4, this map induces a parameterization
Cn ↔ S(1)SU(2) × · · · × S(1)SU(2) → Ň− ∩ w−1Ň+w,
(ζn, . . . , ζ1)↔ (kn(ζn), . . . , k1(ζ1))→ l(ζ),
where
w−1
n−1iγn(kn)wn−1 · · · iγ1(k1) = l(ζ)
n∏
1
a(ζj)
Ad(w−1
j−1)(hγj )u,
u ∈ Ň+. This is part (b) of the Theorem.
The formula for a(k(ζ)) in part (c) follows from Proposition 5. The formula for the Haar
measure follows from part (b) of Corollary 4.
Finally part (3) of the Proposition is equivalent to Lu’s factorization result, Theorem 3.4
of [7], and part (4) follows from the results in Sections 2–4 of [3] (which, as we have already
noted, are valid in the loop context), and Theorem 3.4 of [7]. �
6 The SU(2) case. I
To understand the significance of Proposition 3, we will now spell out its meaning in the simplest
case, where K̇ = SU(2).
In doing explicit calculations, it is convenient to work with ordinary loops, rather than lifts
in L̃SU(2). Thus in this section, and the next, we will identify S(1) with its projection to LSU(2)
(see Lemma 1). We will continue to denote this projection by S(1).
In this case there is an (outer) automorphism of L̃Ġ which interchanges the simple roots α0
and α1. At the level of loops, this automorphism is realized by conjugation by a multivalued
loop,
conj
((
0 iz1/2
iz−1/2 0
))
:
(
a b
c d
)
→
(
d cz−1
bz a
)
. (6.1)
16 D. Pickrell
The root subgroup corresponding to α1 is SL(2,C), the constants, and the root subgroup cor-
responding to α0 is the image of SL(2,C) under this automorphism (see chapter 5 of [11]).
In [11] there is a relatively explicit realization of the groups L̃SU(2,C) and L̃SL(2,C). In this
approach a loop g ∈ LSL(2,C) is identified with a multiplication operator on H = L2(S1,C2).
Relative to the Hardy polarization H = H+ ⊕H−,
g =
(
A B
C D
)
, (6.2)
where A (or D) is the classical Toeplitz operator and C (or B) is the classical Hankel operator
associated to g. The extension L̃SL(2,C) → LSL(2,C) is the C∗-bundle associated to the
pullback of the determinant line bundle, relative to the mapping
LSL(2,C)→ Fred(H+) : g → A(g).
The holomorphic function σ0 on L̃SL(2,C) is, viewed as a section of a line bundle, ‘detA(g̃)’.
Suppose that g ∈ LSU(2), and g̃ ∈ L̃SU(2) is a lift, which is uniquely determined up to
multiplication by an element of the unitary center exp(iRc). Then
|σ0(g̃)|2 = |σ0|2(g) = detA(g)∗A(g) = det(1 + Z(g)∗Z(g))−1,
where Z = CA−1, and
|σ1|2(g) = |σ0|2
((
z1/2 0
0 z−1/2
)
g
(
z−1/2 0
0 z1/2
))
. (6.3)
The simple reflections corresponding to the simple roots α0 and α1 are represented by the
group elements
s0 =
(
0 iz−1
iz 0
)
, and s1 =
(
0 i
i 0
)
,
respectively. We denote their images in the Weyl group by s̄i. The Weyl group (the affine Weyl
group of (ġ, ḣ)) has the structure
W = Z2s̄0 ∝ Z(s̄0s̄1) = Z2s̄1 ∝ Z(s̄0s̄1) = Ẇ ∝ Hom(T, Ṫ ).
Minimal factorizations in the Weyl group must simply alternate the s̄i. This leads to two possible
infinite minimal sequences of simple roots, the two possibilities depending upon whether one
begins with α0 or α1. These are equivalent via the automorphism above. In the following
theorem we will spell out Proposition 3 for the first possibility.
Theorem 5. Let w1 = s0, w2 = s1s0, w3 = s0s1s0, . . . . Then for n > 0,
(a)
N− ∩ w−1
n N+wn =
l =
1
n∑
j=1
xjz
−j
0 1
: xj ∈ C
.
(b) For the diffeomorphisms
Cn → Cn : (ζ1, . . . , ζn)→ x(n) =
n∑
j=1
x
(n)
j (ζ1, . . . , ζn)z−j
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 17
arising from the isomorphism in part (1) of Proposition 3, and the parameterization in part (a),
x
(n)
j (ζ1, . . . , ζn) = x
(N)
j (ζ1, . . . , ζn, 0, . . . , 0),
n < N (hence we will often suppress the superscript), and
x
(n)
j (ζ1, . . . , ζn) = x1(ζj , . . . , ζn, 0, . . . ).
(c) In terms of the correspondence of ζ with g ∈ LSU(2) and l ∈ N−, arising from the
isomorphism in part (1) of Proposition 3,
|σ0|2(g) =
1
det
(
1 +B
(
n∑
1
xjzj
)
B
(
n∑
1
xjzj
)∗) =
n∏
1
1
(1 + |ζj |2)j
and
|σ1|2(g) =
1
det
(
1 +B
(
n−1∑
1
xj+1zj
)
B
(
n−1∑
1
xj+1zj
)∗) =
n−1∏
1
1
(1 + |ζj+1|2)j−1
,
where B(·) is defined as in Theorem 1. In particular
|σ0|
|σ1|
=
n∏
1
1
(1 + |ζj |2)1/2
.
(d) More generally, for 0 ≤ l < n,
det
(
1 +B
(
n−l∑
1
xj+lz
j
)
B
(
n−l∑
1
xj+lz
j
)∗)
=
n−l∏
1
(
1 + |ζj+l|2
)j
.
(e)
n∏
j=1
dλ(xj) =
(
1 + |ζ2|2
)2(1 + |ζ2|2
)4 · · · (1 + |ζn|2
)2(n−1)
n∏
j=1
dλ(ζj).
Proof. Part (a) is a direct calculation. If n = 2m
wn = (s1s0)m = (−1)m
(
z 0
0 z−1
)m
.
Thus if u =
(
a b
c d
)
∈ N+, then
w−1
n uwn =
(
a z−2mb
z2mc d
)
∈ N−
implies a = d = 1, c = 0, and b =
2m−1∑
0
bjz
j . This implies (a) when n is even. The odd case is
similar.
Before taking on the other parts of the Theorem, we need to understand what part (a) says
in terms of the isomorphism of part (b) of Conjecture 1. It is straightforward to calculate that
w−1
j−1iγj (k(ζj))wj−1 = a(ζj)
(
1 ζjz
−j
−ζ̄jzj 1
)
.
18 D. Pickrell
This implies that
g = w−1
n−1iγn(k(ζn))wn−1 · · · iγ1(k(ζ1)) = a(ζn)
(
1 ζnz
−n
−ζ̄nzn 1
)
· · · a(ζ1)
(
1 ζ1z
−1
−ζ̄1z 1
)
.
If we write
g = a(ζn) · · · a(ζ1)
(
αn βn
γn δn
)
,
then there is a recursion relation(
βn+1
δn+1
)
=
(
1 ζnz
−n−1
−ζ̄nzn+1 1
)(
βn
δn
)
. (6.4)
In terms of the isomorphism in part (1) of Proposition 3, part (a) implies that1 −
n∑
1
x
(n)
j z−j
0 1
(αn βn
γn δn
)
=
αn − ( n∑
1
x
(n)
j z−j
)
γn βn −
(
n∑
1
x
(n)
j z−j
)
δn
γn δn
(6.5)
is an (entire) holomorphic function of z. In particular γn and δn must be holomorphic functions
of z, and
n∑
1
x
(n)
j (ζ1, . . . , ζn)z−j =
(
δ−1
n βn
)
−, (6.6)
where (·)− denotes the singular part (at z = 0). The holomorphicity of (6.5) can be checked
directly as follows. The recursion relation (6.4) shows that δn is of the form 1+
∑n
1 djz
j , and βn
is of the form
∑n
1 bjz
−j . Since γn = −β∗n on S1, this shows γn and δn are holomorphic functions
of z. It also shows the x(n)
j are well-defined by (6.6). The relation (6.6) implies the (1, 2) entry
of (6.5) is holomorphic. Also (6.6) implies the (1, 1) entry of (6.5) is of the form
αn − δ−1
n βnγn + holomorphic = δ−1
n (αnδn − βnγn) + hol.
= δ−1
n (αnα∗n + βnβ
∗
n) + hol. = (const)δ−1
n + hol. = holomorphic.
We now consider part (b). We will need several Lemmas.
Lemma 2.
(a) The x(n) satisfy the recursion relation
x(n+1) =
(x(n) + ζn+1z
−n−1
) n∑
p=0
(ζ̄n+1x
(n)zn+1)p
−
(6.7)
=
n∑
p=0
ζ̄pn+1
(
x(n)
)p+1(1 + |ζn+1|2
)
zp(n+1)
−
+ ζn+1z
−n−1. (6.8)
(b) x(n) can be replaced by x(n)+h(z), where h(z) is a holomorphic function, without changing
the recursion.
Proof. The recursion relation (6.4) and (6.6) imply that x(n+1) is the singular part of(
βn + δnζn+1z
−n−1
)(
δn − ζ̄n+1z
n+1βn
)−1 =
(
βnδ
−1
n + ζn+1z
−n−1
)(
1− ζ̄n+1βnδ
−1
n zn+1
)−1
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 19
=
(
βnδ
−1
n + ζn+1z
−n−1
) ∞∑
p=0
(
ζ̄n+1x
(n)zn+1
)p
. (6.9)
Since x(n)zn+1 is O(z), the singular part of (6.9) equals the right hand side of (6.7).
We now rewrite the right hand side of (6.7) as(x(n) + ζn+1z
−n−1
) n∑
p=0
(
ζ̄n+1x
(n)zn+1
)p
−
=
n∑
p=0
ζ̄pn+1
(
x(n)
)p+1
zp(n+1) + ζn+1z
−n−1 +
n∑
p=1
|ζn+1|2ζ̄p−1
n+1
(
x(n)
)p
z(p−1)(n+1)
−
+
n∑
p=0
ζ̄pn+1
(
x(n)
)p+1
zp(n+1) + ζn+1z
−n−1 +
n∑
p=1
|ζn+1|2ζ̄pn+1
(
x(n)
)p+1
zp(n+1)
−
+
n∑
p=0
ζ̄pn+1
(
x(n)
)p+1(1 + |ζn+1|2)zp(n+1)
−
+ ζn+1z
−n−1.
This completes the proof of part (a).
Part (b) is obvious. �
For small n the recursion implies x(1) = ζ1z
−1,
x(2) = ζ1
(
1 + |ζ2|2
)
z−1 + ζ2z
−2, (6.10)
x(3) =
(
ζ1
(
1 + |ζ2|2
)(
1 + |ζ3|2
)
+ ζ2
(
1 + |ζ3|2
)
ζ2ζ̄3
)
z−1 + ζ2
(
1 + |ζ3|2
)
z−2 + ζ3z
−3, (6.11)
x(4) =
(
ζ1
4∏
2
(
1 + |ζj |2
)
+ ζ2
4∏
3
(
1 + |ζj |2
)(
ζ2ζ̄3 + 2ζ3ζ̄4
)
+ ζ3
(
1 + |ζ4|2
)(
ζ3ζ̄4
)2)
z−1
+
(
ζ2
4∏
3
(
1 + |ζj |2
)
+ ζ3
(
1 + |ζ4|2
)
ζ3ζ̄4
)
z−2 + ζ3
(
1 + |ζ4|2
)
z−3 + ζ4z
−4. (6.12)
Lemma 3. y(n) = (zx(n+1))− depends only on ζ2, . . . , ζn+1, and satisfies the same recursion
as x(n), with the shifted variables ζ2, . . . in place of ζ1, . . . .
Proof. For small n the formulas above show that y(n) does not depend on ζ1. By Lemma 2
y(n) =
z
(x(n) + ζn+1z
−n−1
) n∑
p=0
(
ζ̄n+1x
(n)zn+1
)p
−
−
=
(zx(n) + ζn+1z
−n) n∑
p=0
(
ζ̄n+1zx
(n)zn
)p
−
=
(y(n−1) + ζn+1z
−n) n∑
p=0
(
ζ̄n+1y
(n−1)zn
)p
−
.
This establishes the recursion and induction implies y(n) does not depend on ζ1. �
20 D. Pickrell
We can now complete the proof of part (b). Lemma 3 implies that
(
zx(n+1)
)
− =
n+1∑
2
x
(n+1)
j (ζ1, . . . , ζn+1)z−j+1 =
n∑
1
x
(n)
i (ζ2, . . . , ζn)z−i.
This implies that for j > 1,
x
(n+1)
j (ζ1, . . . , ζn+1) = x
(n)
j−1(ζ2, . . . , ζn+1).
By induction this implies part (b). This also implies
x(n+1) = x
(n+1)
1 z−1 + x(n)(ζ2, . . . , ζn+1)z−1.
For future reference, note that there is a recursion for x1 of the form
x1(ζ1, . . . , ζn+1) = x1(ζ1, . . . , ζn)
(
1 + |ζn+1|2
)
+
∑
i+j=n+2
x1(ζi, . . . , ζn)x1(ζj , . . . , ζn)
ζ̄n+1
(
1 + |ζn+1|2
)
(6.13)
+
∑
i+j+k=2n+3
x1(ζi, . . . , ζn)x1(ζj , . . . , ζn)x1(ζk, . . . , ζn)
ζ̄2
n+1
(
1 + |ζn+1|2
)
+ · · · .
It would be highly desirably to find a closed form solution of this recursion for x1.
We now consider part (c). For g as in part (c), consider the Riemann–Hilbert factorization
g = g−g0g+, where
g− =
(
1 x
0 1
)
, g0 =
(
a0 b0
0 a−1
0
)
∈ SL(2,C),
and g+ ∈ H0(D, 0;SL(2,C), 1). Then
Z(g) = C(g)A(g)−1 = C(g−)A(g0g+)A(g0g+)−1A(g−) = Z(g−). (6.14)
(For use in the next paragraph, note that this calculation does not depend on the specific form
of g0.) Let ε1, ε2 denote the standard basis for C2. As in [11], consider the ordered basis
. . . , ε1z
j+1, ε2z
j+1, ε1z
j , . . . , j ∈ Z, for H. This basis is compatible with the Hardy polarization
of H. We claim that
Z(g−) = C(g−) =
. 0 xn . 0 x3 0 x2 0 x1
. 0 0 . 0 0 0 0 0 0
. . . x4 0 x3 0 x2
0 0 0 0 0
. 0 x3
. . . .
. . 0 xn
0 0 0 0 . . 0 0 0
. (6.15)
Let P± denote the orthogonal projections associated to the Hardy splitting of H. For example
P+
(
f =
∑
fkz
k
)
=
∑
k≥0
fkz
k. (6.16)
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 21
Suppose that
(
f1
f2
)
∈ H+. Then
C(g−)A(g−)−1
(
f1
f2
)
= P−g−P+g
−1
−
(
f1
f2
)
= P−g−
(
f1 − P+(xf2)
f2
)
,
P−
(
f1 + P+(xf2) + xf2
f2
)
=
(
P−(xf2)
0
)
= C(g−)
(
f1
f2
)
.
This is the first part of the claim. For the second part one simply calculates directly, using the
simple form for g−.
Comparing (6.15) with (1.4) proves the first part of (c).
Using the factorization
g =
(
1 x
0 1
)
g0g+,
and the specific form of g0, it is clear that
g−
((
z1/2 0
0 z−1/2
)
g
(
z−1/2 0
0 z1/2
))
=
(
1 x′
0 1
)
,
where x′ = (zx)− = x2z
−1 + · · ·+ xnz
−(n−1). We now use (6.3) and (6.15) to prove the second
part of (c).
Part (d) follows from (c). Part (e) can be read off from Lemma 2 (see (6.10)–(6.12)), or from
part (2) of Proposition 3. �
We are now in a position to prove Theorem 1 at the end of the Introduction.
Proof. By parts (d) and (e) of Theorem 5,∫
1
n−1∏
l=0
det
(
1 +B
(
n−l∑
j=1
xl+jzj
)
Bn
(
n−l∑
j=1
xl+jzj
)∗)pl
dλ(x1, . . . , xn)
=
∫ n−1∏
l=0
n−l∏
j=1
(
1 + |ζl+j |2
)−jpl
n∏
j=1
(
1 + |ζj |2
)−2(j−1)
dλ(ζj)
=
∫ (
1 + |ζ1|2
)−p1dλ(ζ1)
∫ (
1 + |ζ2|2
)2−(2p1+p2)
dλ(ζ2) · · ·
· · ·
∫ (
1 + |ζn|2
)2n−2−(np1+···+pn)
dλ(ζn)
= πn
1
p1 − 1
1
2p1 + p2 − 3
· · · 1
np1 + (n− 1)p2 + · · ·+ pn − (2n− 1)
. �
7 The SU(2) case. II
This is a continuation of Section 6. We first consider the limit n → ∞, in the context of
Theorem 5. From the point of view of analysis, this limit is naturally related to the critical
exponent s = 1/2 for the circle. We secondly show that there is a global factorization of the
momentum mapping for (S(1), ω), extending the formulas in (c) of Theorem 5. As in Section 6,
we will continue to view S(1) as a submanifold of LSU(2), rather than L̃SU(2).
22 D. Pickrell
As in (6.16), we let P± denote the orthogonal projections associated to the Hardy splitting
of L2(S1). Given f =
∑
fnz
n, we will write f∗ =
∑
c̄nz
−n. If we simply write z for the
multiplication operator corresponding to z, then
(·)∗ ◦ P− = z ◦ P+ ◦ z−1 ◦ (·)∗,
and
(·)∗ ◦ P+ = z ◦ P− ◦ z−1 ◦ (·)∗.
For a function F ∈ L∞(S1), viewed as a bounded multiplication operator on L2(S1), we will
write A(F ) = P+FP+, and so on, as in (6.2).
Suppose
l =
(
1 x
0 1
)
, where x =
n∑
1
xjz
−j .
There exists a unique g ∈ LpolSU(2) with unique triangular factorization g = la(g)h1u where
a(g) = |σ1|(g)/|σ0|(g), and
|σ0|2 = det(1 + C(x)∗C(x))−1, |σ1|2 = det(1 + C(zx)∗C(zx))−1 (7.1)
(see part (c) of Theorem 5).
Lemma 4. The triangular factorization of g ∈ LpolSU(2) is given by
g =
(
1 x
0 1
)(
a 0
0 a−1
)1 +
n−1∑
1
αjz
j
n−2∑
0
βjz
j
n∑
1
γjz
j 1 +
n−1∑
1
δjz
j
,
where
γ = −((1 + C(zx)C(zx)∗)−1(x))∗, δ∗ = C(x)γ,
1 + α =
1
a2
(1−A(x)γ), β = − 1
a2
A(x)(1 + δ).
Proof. Because
g =
(
a(1 + α) + a−1xγ aβ + xa−1(1 + δ)
a−1γ a−1(1 + δ)
)
has values in SU(2) (as a function of z ∈ S1),
a2(1 + α) + xγ = 1 + δ∗, and a2β + x(1 + δ) = −γ∗.
The first equation can be expressed in operator language as
A(x)γ = 1− a2(1 + α), C(x)γ = δ∗. (7.2)
The second equation is equivalent to
A(x)(1 + δ) = −a2β, C(x)(δ) = −x− γ∗. (7.3)
We can solve for δ, using the second equation of (7.2),
(·)∗ ◦ C(x)(γ) = zB(z−1x∗)(γ∗) = δ
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 23
The second equation of (7.3) now implies
C(x)zB
(
z−1x∗
)
(γ∗) = −x− γ∗
which is equivalent to
(1 + C(zx)C(zx)∗)(γ∗) = −x.
The Lemma follows from these equations. �
Let W 1/2 denote the Sobolev space of (Lebesgue equivalence classes of) functions having
half of a derivative, i.e. if f =
∑
fjz
j , then
∑
j|fj |2 < ∞. A class in W 1/2 is not in general
represented by a continuous function. Despite this, W 1/2(S1, SU(2)) is a connected topological
group (homotopy equivalent to LSU(2); see [1]), and it is the natural domain for the basic
factorization theorems in the theory of loop groups (see Chapter 8 of [11]). We let S(1) denote
the completion of S(1) in W 1/2(S1, SU(2)).
Theorem 6. By taking a limit as n → ∞ in Theorem 5, we obtain bijective correspondences
among the following three sets
(a)
{
ζ = (ζ1, ζ2, . . . ) :
∞∑
1
j|ζj |2 <∞
}
.
(b) x =
∞∑
1
xjz
−j ∈W 1/2, where (as in (b) of Theorem 5)
xj(ζ1, . . . ) = lim
n→∞
x1(ζj , . . . , ζn), (7.4)
and conversely
ζj(x1, . . . ) = lim
n→∞
ζ1(xj , . . . , xn). (7.5)
(c){
g =
(
d(z)∗ −c(z)∗
c(z) d(z)
)
∈ S(1) : c(z), d(z) ∈ H0(∆), c(0) = 0
}
,
where
g = lim
n→∞
a(ζn)
(
1 ζnz
−n
−ζ̄nzn 1
)
· · · a(ζ1)
(
1 ζ1z
−1
−ζ̄1z 1
)
,
a(ζj) = (1 + |ζj |2)−1/2, g has triangular factorization
g =
1
∞∑
1
xjz
−j
0 1
( |σ1|(g)
|σ0|(g) 0
0 |σ0|(g)
|σ1|(g)
)
u, (7.6)
and the entries of u are given by the same formulas as in Lemma 4.
Note that for k > 0 g =
(
z−k 0
0 zk
)
has a second row which is holomorphic in the disk, but
g is not in S(1). Thus in part (c) it is necessary to require that g ∈ S(1).
24 D. Pickrell
Proof. By Theorem 5, for ζ with a finite number of terms,∏(
1 + |ζj |2
)j = det(1 + C(x)C(x)∗), (7.7)
and obviously
tr(C(x)C(x)∗) =
∑
j|xj |2. (7.8)
For an arbitrary sequence ζ, the product on the LHS of (7.7) is finite iff
∞∑
1
j|ζj |2 <∞. Similarly,
for an arbitrary x, the determinant is finite iff (7.8) is finite. Thus given ζ as in part (a),
the partial sums for the series representing x will have limits in W 1/2. To understand why
there is a unique limit, and to prove the other statements in part (b), recall that the recursion
relation (6.13) implies that x1 has a series expansion, with nonnegative integer coefficients, in
terms of the variables ζj , ζ̄j , of the form
x1(ζ1, . . . ) = ζ1
∞∏
j=2
(1 + |ζj |2) + ζ2
∞∏
j=3
(
1 + |ζj |2
)(
ζ2ζ̄3 + 2ζ3ζ̄4 + · · ·
)
+ ζ3
∞∏
j=4
(
1 + |ζj |2
)(
(ζ3ζ̄4)2 + · · ·
)
+ · · · ,
Since |x1|2 is dominated by (7.8), this series will converge absolutely. Thus x1 is a well-defined
function of ζ. The relation (7.4) follows from (b) of Theorem (5). Thus all the xj , and hence
also x ∈W 1/2, are uniquely determined by ζ, assuming
∑
j|ζj |2 <∞.
Because
∞∏
j=k
(
1 + |ζj |2
)
=
det
(
1 + C(zkx)C(zkx)∗
)
det
(
1 + C(zk−1x)C(zk−1x)∗
)
and the triangular nature of the relation between the ζj and the xj , the map from the ζ to x can
be inverted, and the ζj will be expressible in terms of ζ1 as in (7.5). We have thus established
that there is a bijective correspondence between the sets in (a) and (b).
Now suppose that we are given x as in part (b). We claim that (7.1) and Lemma 4 imply that
we can obtain a g as in part (c), with triangular factorization as in (7.6). Because x ∈ W 1/2,
C(x) and C(zx) are Hilbert–Schmidt operators (viewed as operators on L2(S1)), so that the
formulas (7.1), and hence the formula for a(g), make sense. The formulas for γ and δ in Lemma 4
a priori show only that γ and δ are L2(S1) (not necessarily W 1/2), so that we obtain a SU(2)
loop g, expressed as in part (c), with L2 entries. But as in (6.14), Z(g) = Z
((
1 x
0 1
))
, and
this is a Hilbert–Schmidt operator. Because g ∈ L∞(S1), A(g) is a bounded operator, and hence
this implies that C(g) is a Hilbert–Schmidt operator. Thus g ∈W 1/2.
Conversely given g as in part (c), it is obvious that g has a triangular form as in (7.6), and
this determines x. The equality
det |A(g)|2 = det
(
1 + C(x)C(x)∗
)−1
implies that x ∈W 1/2. Because the sets in (a) and (b) are in correspondence, this also determi-
nes ζ. This completes the proof. �
Recall that in Theorem 5 we considered the minimal sequence α0, α1, . . . . By considering the
sequence α1, α0, . . . , or in other words, by applying the automorphism (6.1) which interchanges
the two simple roots, the proceeding Theorem can be reformulated in the following way.
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 25
Corollary 1. There are bijective correspondences among the following three sets
(a)
{
η = (η0, η1, . . . ) :
∞∑
0
j|ηj |2 <∞
}
.
(b) y =
∞∑
0
yjz
−j ∈W 1/2, where
yj(η0, . . . ) = lim
n→∞
y0(ηj , . . . , ηn),
and conversely
ηj(y0, . . . ) = lim
n→∞
η0(yj , . . . , yn).
(c){
h =
(
e(z) b(z)
−b∗(z) e(z)∗
)
∈ S(1) : b(z), e(z) ∈ H0(∆)
}
,
where
h = lim
n→∞
a(ηn)
(
1 ηnz
n
−η̄nz−n 1
)
· · · a(η0)
(
1 η0
−η̄0 1
)
,
and h has triangular factorization of the form
h =
(
1 0
y 1
)( |σ1|(h)
|σ0|(h) 0
0 |σ0|(h)
|σ1|(h)
)1 +
∞∑
1
αjz
j
∞∑
0
βjz
j
∞∑
1
γjz
j 1 +
∞∑
1
δjz
j
, (7.9)
β = −(1 + C(y)∗C(y))−1(y∗), α∗ = C(y)β,
1 + δ = a(h)2(1−A(y)β), γ∗ = −a(h)2D(y∗)(α∗).
Moreover
|σ0|2(h) =
∞∏
j=0
(
1 + |ηj |2
)−j
, |σ1|2(h) =
∞∏
j=0
(
1 + |ηj |2
)−(j+1)
and
a(h) =
|σ1|
|σ0|
=
∏(
1 + |ηj |2
)−1/2
.
In the following statement we will continue to view S(1) as a subset of W 1/2(S1, SU(2)), but
in the proof it will be necessary to consider lifts in the Kac–Moody extension (as in Lemma 1).
Theorem 7. Suppose that ζ, χ, and η are sequences such that
∞∑
j=1
j
(
|ζj |2 + |χj |2 + |ηj |2
)
<∞.
By slight abuse of notation, we identify χ with the function χ =
∞∑
j=1
χjz
j. Let g be defined as
in (c) of Theorem 6 and h as in (c) of Corollary 1.
(a) The product h−1e(χ−χ
∗)h1g ∈ S(1).
26 D. Pickrell
(b) The mapping
(η, χ, ζ) :
∑
j
(
|ηj |2 + |χj |2 + |ζj |2
)
<∞→ S(1) : (η, χ, ζ)→ h−1e(χ−χ
∗)h1g
is injective.
(c)
|σ0|2
(
h−1e(χ−χ
∗)h1g
)
=
∏
j=1
(
1 + |ηj |2
)−j exp
−2
∑
j=1
j|χj |2
∏
j=1
(
1 + |ζj |2
)−j
,
|σ1|2
(
h−1e(χ−χ
∗)h1g
)
=
∏
j=1
(1 + |ηj+1|2)−j exp
−2
∑
j=1
j|χj |2
∏
j=1
(
1 + |ζj+1|2
)−j
,
a2
(
h−1e(χ−χ
∗)h1g
)
=
∏
j=1
1 + |ζj |2
1 + |ηj |2
.
Proof. For part (a), we do a calculation at the level of loops. If we write h as in (7.9), then
the triangular factorization of h−1 is given by
h−1 = l
(
h−1
)(a(h−1) 0
0 a(h−1)−1
)(
1 y∗
0 1
)
,
where
l
(
h−1
)
= u(h)∗ =
(
1 + α∗ γ∗
β∗ 1 + δ∗
)
.
Then
h−1e(χ−χ
∗)h1g = l
(
h−1
)
a
(
h−1
)h1
(
1 y∗
0 1
)
e(χ−χ
∗)h1
(
1 x
0 1
)
a(g)h1u(g). (7.10)
The main point of the proof is that the middle three factors are upper triangular, and we can
find the (loop space) triangular factorization of the product with ease. Thus (7.10) equals
l
(
h−1
)
e−χ
∗h1
(
1 a(h−1)2P−(y∗e2χ
∗
+ xe2χ)
0 1
)
× a(h)h1a(g)h1
(
1 a(g)−2P+(y∗e2χ
∗
+ xe2χ)
0 1
)
eχh1u(g).
This triangular form implies part (a) of the Theorem.
This calculation also implies that a(h−1e(χ−χ
∗)h1g) = a(h−1)a(g). Given that a(h−1) =
a(h∗) = a(h), and formulas we have already established for a(g) and a(h) (see Corollary 1), this
also implies the third formula in part (c).
To obtain the first two formulas in part (c), we need to lift each of the factors, h−1, the
torus-valued loop, and g, into the extension (where the lift is determined by requiring that
the lift is in S(1)), and then repeat the preceding calculation in the extension. To do this
we replace a(g)h1 by |σ0|(g)h0 |σ1|(g)h1 , and similarly for h−1 (and recall that we have explicit
product formulas for the functions |σj |(g), and similarly for h). The torus-valued loop e(χ−χ
∗)h1
has a vanishing diagonal term in its triangular factorization. However, a well-known formula for
Toeplitz determinants implies that
|σj |
(
e(χ−χ
∗)h1
)
= exp
−2
∑
j=1
j|χj |2
.
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 27
Thus the lift of this torus-valued loop into the extension has diagonal term
exp
−2
∑
j=1
j|χj |2(h0 + h1)
.
It is now straightforward to repeat the calculation above and conclude, as we did at the level of
loops, that (i) the product of these lifts is in S(1) and (ii) the diagonal term of the product is
the product of the diagonal terms of the factors. This implies the formulas in part (c).
To prove that the mapping is injective, recall that
l
(
h−1e(χ−χ
∗)h1g
)
= l
(
h−1
)
e−χ
∗h1a(h)h1
(
1 P−(y∗e2χ
∗
+ xe2χ)
0 1
)
a(h)−h1
=
(
1 + α∗ γ∗
β∗ 1 + δ∗
)
e−χ
∗h1
(
1 a(h)2P−(y∗e2χ
∗
+ xe2χ)
0 1
)
=
(
(1 + α∗)e−χ
∗
(1 + α∗)e−χ
∗
a(h)2P−(y∗e2χ
∗
+ xe2χ) + γ∗eχ
∗
β∗e−χ
∗
β∗e−χ
∗
a(h)2P−(y∗e2χ
∗
+ xe2χ) + (1 + δ∗)eχ
∗
)
. (7.11)
We need to show that this matrix determines y (equivalently h or η), χ, and x (or g or ζ).
By the form of the triangular factorization of h in part (c) of Corollary 1, it is clear that
the first row of h, evaluated at a point z ∈ S1, is determined by 1 + α(z) and β(z). Because
h(z) ∈ SU(2), this also determines h(z). This means that the first row of h, as a holomorphic
function of z ∈ ∆ is determined up to a phase ambiguity by the ratio β/(1 + α).
Now suppose that we are given (7.11). The first column determines the pair of holomorphic
functions β and 1+α up to multiplication by a holomorphic function. This determines the ratio
β/(1 +α), and it also fixes the phase ambiguity. Thus the first column determines h. It is then
clear that χ is determined. We can then use the (1, 2) entry to find x. This proves injectivity,
and completes the proof of the Theorem. �
A Appendix
In this appendix we will review some of the ideas in [7], relevant to this paper, from a slightly
different perspective. The main rationale for including this appendix is that the basic arguments
are valid in the more general Kac–Moody category. Throughout this appendix, we will use the
notation and basic results in [6].
We start with the following data: A is an irreducible symmetrizable generalized Cartan ma-
trix; g = g(A) is the corresponding Kac–Moody Lie algebra, realized via its standard (Chevalley–
Serre) presentation; g = n−⊕h⊕n+ is the triangular decomposition; b = h⊕n+ the upper Borel
subalgebra; G = G(A) is the algebraic group associated to A by Kac–Peterson; H, N± and B
are the subgroups of G corresponding to h, n±, and b, respectively; K is the “unitary form”
of G; T = K ∩H the maximal torus; and W = NK(T )/T ' NG(H)/H is the Weyl group.
A basic fact is that (G,B,NG(H)) with Weyl group W is an abstract Tits system. This
yields a complete determination of all the (parabolic) subgroups between B and G. They are
described as follows.
Let Φ be a fixed subset of the simple roots. The subgroup of W generated by the simple
reflections corresponding to roots in Φ will be denoted by W (Φ). The parabolic subgroup
corresponding to Φ, P = P (Φ), is given by P = BW (Φ)B. Given w ∈ NK(T ), we will denote
its image in W/W (Φ) by w.
28 D. Pickrell
The basic structural features of G/P which we will need are the Birkhoff and Bruhat decom-
positions
G/P =
⊔
Σw, Σw = N−wP,
G/P =
⊔
Cw, Cw = BwP,
respectively, where the indexing set is W/W (Φ) in both cases. The strata Σw are infinite
dimensional if g is infinite dimensional, while the cells Cw are always finite dimensional. Our
initial interest is in the Schubert variety C̄w, the closure of the cell.
Fix w ∈ W/W (Φ). We choose a representative w ∈ NK(T ) of minimal length n, and we fix
a factorization
w = rn · · · r1, (A.1)
where rj = iγi
(
0 i
i 0
)
, and iγj : SL2 → G is the canonical homomorphism of SL2 onto the root
subgroup corresponding to the simple root γj .
Proposition 4. For w as in (A.1), the map
rn exp(g−γn)× · · · × r1 exp(g−γ1)→ G/P : (pj)→ pn · · · p1P
is a complex analytic isomorphism onto Cw.
This result is essentially (5) of [6] together with Tits’s theory. We will include a proof for
completeness.
Proof. Let ∆+ denote the positive roots, ∆+(Φ) the positive roots which are combinations of
elements from Φ. The “Lie algebra of P” is p = Σg−β ⊕ b where the sum is over β ∈ ∆+(Φ);
this is the Lie algebra of P in the sense that it is the subalgebra generated by the root spaces gγ
for which exp : gγ → G is defined and have image contained in P . The subgroups exp(gγ)
generate P . We also let p− denote the subalgebra opposite p: p− =
∑
g−γ , where the sum is
over γ ∈ ∆+ \∆+(Φ). The corresponding group will be denoted by P−.
The cell Cw is the image of the map N+ → G/P : u→ uwP . The stability subgroup at wP
is N+ ∩wPw−1.
At the Lie algebra level we have the splitting
n+ = n+ ∩Ad(w)(p)⊕ n+ ∩Ad(w)(p−). (A.2)
The second summand equals n+
w = ⊕ gβ , where the sum is over roots β > 0 with w−1β ∈
−(∆+ \ ∆+(Φ)). These roots β are necessarily real, so that exp : n+
w → N+
w ⊆ N+ is well-
defined.
For q ∈ Z+ let N+
q denote the subgroup corresponding to n+
q = span{gβ : height(β) ≥ q}.
Then N+/N+
q is a finite dimensional nilpotent Lie group, and it is also simply connected. By
taking q sufficiently large and considering the splitting (A.2) modulo n+
q , we conclude by finite
dimensional considerations that each element in N+ has a unique factorization n = n1n2, where
n1 ∈ N+
w and n2 ∈ N+ ∩wPw−1:
N+ ' N+
w ×
(
N+ ∩wPw−1
)
. (A.3)
The important point is that modulo N+
q , we can control N+ ∩wPw−1 by the exponential map.
The following lemma is standard.
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 29
Lemma 5. In terms of the minimal factorization w = rn · · · r1, the roots β > 0 with w−1β < 0
are given by
βj = rn · · · rj+1(γj) = rn · · · rj(−γj), 1 ≤ j ≤ n.
Because w is a representative of w ∈ W/W (Φ) of minimal length, all of these βj satisfy
w−1βj ∈ −(∆+ \∆+(Φ)). Otherwise, if say w−1βj ∈ −∆+(Φ), then
w−1rβj
w = r1 · · · rj−1rjrj−1 · · · r1 ∈ NK(T ) ∩ P
and w′ = w(w−1rβj
w) = rn · · · r̂j · · · r1 would be a representative of w of length < n (here we
have used the fact that W (Φ) = NK(T ) ∩ P/T , which follows from the Bruhat decomposition).
For future reference we note this proves that
N+
w = N+ ∩ (N−)w = N+ ∩ (P−)w (A.4)
and (A.3) shows that
N+
w × w ∼= Cw. (A.5)
Now for any 1 ≤ p ≤ q ≤ n,
⊕
p≤j≤p gβj
is a subalgebra of n+
w . Thus by (A.3)
exp(gβn)× · · · × exp(gβ1)× w ∼= Cw.
This completes the proof of Proposition 4, when we write
exp(gβj
) = rn · · · rj exp(gγj )rj · · · rn. �
For each j, let Pj denote the parabolic subgroup iγj (SL2)B. Let
γw = Pn ×B · · · ×B P1/B,
where
Pn × · · · × P1 ×B × · · · ×B → Pn × · · · × P1
is given by
(pj)× (bj)→
(
pnbn, b
−1
n pn−1bn−1, . . . , b
−1
2 p1b1
)
.
We have written “γw” instead of “γw” to indicate that this compact complex manifold depends
upon the factorization (A.1).
Corollary 2. The map
γw → C̄w : (pj)→ pn · · · p1P
is a (Bott–Samelson) desingularization of C̄w.
This is an immediate consequence of Proposition 4.
Let
SL′2 =
{
g =
(
a b
c d
)
∈ SL(2,C) : a 6= 0
}
.
30 D. Pickrell
Corollary 3. Consider the surjective map
SL2 × · · · × SL2 → C̄w : (gj)→ rniγn(gn) · · · r1iγ1(g1)P.
The inverse image of Cw is SL′2 × · · · × SL′2.
Proof. Let σ = rn−1 · · · r1. It suffices to show that for the natural actions
rniγn(SL′2)× Cσ̄ → Cw, (A.6)
rniγn(SL2 \ SL′2)× Cσ̄ → C̄w,
and
rniγn(SL2)× (C̄σ̄ \ Cσ̄)→ C̄w \ Cw.
The first line, (A.6), follows from Proposition 4, since iγn(SL′2)⊆exp(−g−γn)B and B×Cσ̄ ⊆ Cσ̄.
The second line follows from
rniγn
(
0 b
c d
)
· Cσ̄ = iγn
(
c b
0 d
)
· Cσ̄ ⊆ Cσ̄.
For the third line it’s clear that the image of the left hand side is a union of cells, since we can
replace rniγn(SL2) by Pn. This image is at most n− 1 dimensional. Therefore it must have null
intersection with Cw. �
Corollary 4.
(a) Let k(ζ) be defined as in (5.3). The map
Cn → Cw : (ζn, . . . , ζ1)→ rniγn(k(ζn)) · · · r1iγ1(k(ζ1))P
is a real analytic isomorphism.
(b) In terms of the parameterization in (a), and the parameterization of Cw by N+∩wPw−1
(see (A.4) and (A.5)), Haar measure (unique up to a constant) is given by
dλN+∩wPw−1 =
n∏
j=1
(
1 + |ζj |2
)δ(w−1
j−1hγjwj−1)−1
=
∏
1≤i<j≤n
(
1 + |ζj |2
)−γi(wi−1w
−1
j−1hγjwj−1w
−1
i−1)
,
where δ =
∑
Λj, the sum of the dominant integral functionals for g.
Proof. The proof of (a) is by induction on n. We write wn in place of w, and k(ζn) =
l(ζn)a(ζn)u(ζn) for its SL(2,C) triangular factorization.
The case n = 1 is obvious. Assume the result holds for n− 1. Suppose that
rniγn(k(ζn)) · · · r1iγ1(k(ζ1))P = rniγn(k(ζ ′n)) · · · r1iγ1(k(ζ ′1))P.
Since Cwn−1 is B+-stable, this equation implies
rniγn(l(ζn))xP = rniγn(l(ζ ′n))yP,
where xP, yP ∈ Cwn−1 . Proposition 4 implies that ζn = ζ ′n. Given this, the induction hypothesis
implies that ζj = ζ ′j for j < n. Thus the map is injective.
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 31
Since k(ζn) = l(ζn) modulo B+, and elements of B+ stabilize Cwn−1 , Proposition 4 also
implies the map is surjective. This proves (a).
Now consider (b). We will first establish the second formula by induction. We will then show
the two formulas are equivalent.
Recall the factorization
iγ(k(ζ)) = iγ
((
1 0
ζ 1
)(
a(ζ) 0
0 a(ζ)−1
)(
1 −ζ̄
0 1
))
.
From this the case n = 1 is clear. Now suppose the second formula in (b) is valid for n− 1 and
n > 1. We have
rniγn(k(ζn)) = rniγn
((
1 0
ζn 1
)(
a(ζn) 0
0 a(ζn)−1
)(
1 −ζ̄n
0 1
))
.
The action of
iγn
((
1 −ζ̄n
0 1
))
on Cwn−1 , in terms of the parameterization by N+ ∩ wn−1N
−w−1
n−1, is by translation. So this
action preserves Lebesgue measure. The action of
iγn
((
a(ζn) 0
0 a(ζn)−1
))
on Cwn−1 , in terms of the parameterization by N+ ∩ wn−1N
−w−1
n−1, is by conjugation, and we
can easily calculate the effect on volume. In a routine way this leads to the second formula
in (b).
To prove the formulas in (b) are equivalent, recall that for a simple positive root γ, with
corresponding reflection rγ , one has
rγδ = δ − γ.
This implies that
(wj−1δ)(hγj )− 1 = (rj−1 · · · r1δ)(hγj )− 1 = (rj−1 · · · r2(δ − γ1))(hγj )− 1
= (δ − γj−1 − rj−1γj−2 − · · · − rj−1 · · · r2γ1)(hγj )− 1
= −
j−1∑
i=1
(rj−1 · · · ri+1γi)(hγj ) = −
j−1∑
i=1
(
wj−1w
−1
i−1γi
)
(hγj ).
This completes the proof of (b). �
Fix an integral functional λ ∈ h∗ which is antidominant. Denote the (algebraic) lowest weight
module corresponding to λ by L(λ), and a lowest weight vector by σλ. Let Φ denote the simple
roots γ for which λ(hγ) = 0, where hγ is the coroot, P = P (Φ) the corresponding parabolic
subgroup. The Borel–Weil theorem in this context realizes L(λ) as the space of strongly regular
functions on G satisfying
f(gp) = f(g)λ(p)−1
for all g ∈ G and p ∈ P , where we have implicitly identified λ with the character of P given by
λ(u1w exp(x)u2) = expλ(x)
32 D. Pickrell
for x ∈ h, u1, u2 ∈ N+, w ∈ W (Φ). Thus we can view L(λ) as a space of sections of the line
bundle
Lλ = G×λ C→ G/P.
If g is of finite type, then L(λ) = H0(Lλ); if g is affine (and untwisted), then L(λ) consists of
the holomorphic sections of finite energy, as in [11].
Normalize σλ by σλ(1) = 1.
Proposition 5. Let w ∈W/W (Φ), and let w = rn · · · r1 be a representative of minimal length n.
Let wj = rj · · · r1. The positive roots mapped to negative roots by w are given by
τj = w−1
j−1(γj), 1 ≤ j ≤ n;
let λj = −λ(hτj ), where hτ is the coroot corresponding to τ . Then
σwλ (rniγn(gn) · · · r1iγ1(g1)) = σλ
(
w−1
n−1iγn(gn)wn−1 · · ·w−1
1 iγ2(g2)w1iγ1(g1)
)
=
n∏
1
a
λj
j ,
where g =
(
a b
c d
)
∈ SL2.
Proof of Proposition 5. The claim about the τj follows from Lemma 5. None of these roots
lie in ∆+(Φ), by the same argument as follows (5). Thus each λj > 0. It follows that Πaλj
j is
nonzero precisely on the set SL′2 × · · · × SL′2.
Now σwλ , viewed as a section of Lλ → G/P , is nonzero precisely on the w-translate of the
largest stratum,
wΣ1 = wP−P = (P−)wwP.
We claim the intersection of this with C̄w is Cw. In one direction
Cw =
(
N+ ∩ (P−)w
)
wP ⊆ (P−)wwP
by (A.4). On the other hand (N+ ∩ (P−)w) is a closed finite dimensional subgroup of (P−)w.
Since (P−)w is topologically equivalent to wΣ1, the limit points of Cw must be in the complement
of wΣ1. This establishes the other direction.
It now follows from Proposition 4 that σwλ is also nonzero precisely on SL′2×· · ·×SL′2, viewed
as a function of (gn, . . . , g1).
Write
σλ
(
w−1
n−1iγn(gn)wn−1w
−1
n−2iγn−1(gn−1)wn−2 · · · iγ1(g1)
)
= σλ
(
iτn(gn)iτn−1(gn−1) · · · iτ1(g1)
)
, (A.7)
where iτi(·) = w−1
i−1iγi(·)wi−1. Because
w−1
i−1(γi) > 0,
iτj : SL2 → G is a homomorphism onto the root subgroup corresponding to τj which is compa-
tible with the canonical triangular decompositions.
For g =
(
a b
c d
)
∈ SL′2, write g = LDU , where
L =
(
1 0
ca−1 1
)
, D =
(
a 0
0 a−1
)
, U =
(
1 a−1b
0 1
)
.
Homogeneous Poisson Structures on Loop Spaces of Symmetric Spaces 33
Then for (gj) ∈ SL′2 × · · · × SL′2, (A.7) equals
σλ(iτn(LnDnUn) · · · iτ1(L1D1U1))
= σλ(iτn(LnU ′
n)iτn−1(L
′
n−1U
′
n−1) · · · iτ1(L′1U ′
1)iτn(Dn) · · · iτ1(D1))
= σλ(iτn(LnU ′
n) · · · iτ1(L′1U ′
1))Πa
λj
j ,
where each L′j (U ′
j) has the same form as Lj (Uj , respectively). This follows from the fact that H
normalizes each exp(g±r).
Now each L′jU
′
j ∈ SL′2, so that iτn(LnU ′
n) · · · iτ1(L′1U ′
1) is in Σ1. We now conclude that
σλ(iτn(LnU ′
n) · · · iτ1(L′1U ′
1)) = 1,
by the fundamental theorem of algebra, since this is polynomial and never vanishes. �
References
[1] Brezis H., New questions related to topological degree, in The Unity of Mathematics, Prog. Math., Vol. 244,
Birkhäuser, Boston, MA, 2006, 137–154.
[2] Caine A., Compact symmetric spaces, triangular factorization, and Poisson geometry, J. Lie Theory 18
(2008), 273–294, math.SG/0608454.
[3] Caine A., Pickrell D., Homogeneous Poisson structures on symmetric spaces, Int. Math. Res. Not., to appear,
arXiv:0710.4484.
[4] Evens S., Lu J.-H., On the variety of Lagrangian subalgebras. I, Ann. Sci. École Norm. Sup. (4) 34 (2001),
631–668, math.DG/9909005.
[5] Kac V., Infinite-dimensional Lie algebras. An introduction, Birkhäuser, Boston, MA, 1983.
[6] Kac V., Constructing groups from infinite-dimensional Lie algebras, in Infinite-Dimensional Groups with
Applications (Berkeley, Calif., 1984), Editor V. Kac, Math. Sci. Res. Inst. Publ., Vol. 4, Springer, New York,
1985, 167–216.
[7] Lu J.-H., Coordinates on Schubert cells, Kostant’s harmonic forms, and the Bruhat–Poisson structure on
G/B, Transform. Groups 4 (1999), 355–374, dg-ga/9610009.
[8] Odzijewicz A., Ratiu T., Banach Lie–Poisson spaces and reduction, Comm. Math. Phys. 243 (2003), 1–54,
math.SG/0210207.
[9] Pickrell D., An invariant measure for the loop space of a simply connected compact symmetric space,
J. Funct. Anal. 234 (2006), 321–363, math-ph/0409013.
[10] Pickrell D., A survey of conformally invariant measures on Hm(∆), math.PR/0702672.
[11] Pressley A., Segal G., Loop groups, Oxford Mathematical Monographs, Oxford Science Publications, The
Clarendon Press, Oxford University Press, New York, 1986.
[12] Widom H., Asymptotic behavior of block Toeplitz matrices and determinants. II, Adv. Math. 21 (1976),
1–29.
http://arxiv.org/abs/math.SG/0608454
http://arxiv.org/abs/0710.4484
http://arxiv.org/abs/math.DG/9909005
http://arxiv.org/abs/dg-ga/9610009
http://arxiv.org/abs/math.SG/0210207
http://arxiv.org/abs/math-ph/0409013
http://arxiv.org/abs/math.PR/0702672
1 Introduction
2 Loop groups
3 Type I case
4 The S2 case
5 Type II case
6 The SU(2) case. I
7 The SU(2) case. II
A Appendix
References
|