PACIFIC JOURNAL OF MATHEMATICS

Vol. 212, No. 1, 2003

SYMMETRIC SPACE VALUED MOMENT MAPS
Matthew Leingang
For a compact Lie group G, three examples of G-spaces
which can serve as the target of a moment map are discussed.
Abstracting the work of Alekseev, Meinrenken, and Malkin,
we cast these theories into a unied framework.
Let G be a compact, connected Lie group, and M a manifold on which
G acts. There are several natural G-spaces which can be considered as the
target of a moment map originating from G. The rst is the dual g

to the
Lie algebra of G; we say that M is a Hamiltonian G-space if M has a G-
invariant symplectic form and there exists an equivariant map : M g

such that
(
M
) = d , ) (1)
for all g. is called the moment map [5], [4].
In [3], Alekseev, Meinrenken, and Malkin dene a Hamiltonian theory in
which the moment map has the group itself as target. Given an invariant
inner product B on g, M is called q-Hamiltonian if there is an invariant
two-form and an equivariant map : M G (again called a moment
map) such that
d =
1
12

B(, [, ]), (2a)

and for all g,
(
M
) =
1
2

B( +, ); (2b)
ker
x
=
_

M
(x) [ ker
_
Ad
(x)
+1
__
. (2c)
Here and are the left- and right-invariant Maurer-Cartan forms on
G. This theory is more complicated, especially when G is nonabelian. For
(2a) requires that may not be closed and (2b) requires that may not be
nondegenerate.
If G is given the structure of a Poisson-Lie group, one can also consider
the class of Poisson-Lie G-spaces [11], [10]. These have a symplectic form
and equivariant map : M G

such that
(
M
) =

G
,
_
(3)
for all . Here
G
, the right-invariant Maurer-Cartan form on G

, takes
values in g

, and hence pairs with g. The complications here are that the
103
104 MATTHEW LEINGANG
individual maps g : M M do not preserve the full structure of M, i.e., the
Poisson structure, as they do in the two cases above. Rather, the action map
itself G M M is a Poisson map. However, in [1], Alekseev introduces
another target: The canonical noncompact symmetric space Y within G
C
transverse to G. Let i : Y G
C
be the inclusion. The dierential equations
a moment map : M Y must satisfy are
d =
1
2

ImB
_

G
C
, [
G
C
,
G
C
]
_
; (4a)
and for all ,
(
M
) =
1
2

B(
G
C
+
G
C
, ); (4b)
ker
x
= 0. (4c)
They are similar to (2a)-(2c). The spaces G

and Y are equivariantly dif-

feomorphic, and Alekseev uses Y to construct a correspondence between
ordinary Hamiltonian G-spaces and Poisson-Lie G-spaces. Both G

and Y
are equivariantly dieomorphic to a slightly more natural space, G
C
/G.
Thus we have three moment map theories, or at least three natural G-
spaces which serve as targets for moment maps: g

for the classical Hamil-

tonian theory; G for the q-Hamiltonian theory, and G
C
/G representing the
Poisson or Y -valued theory. In this note we bring these theories into a uni-
ed framework. Abstracting from [3], we show that given a symmetric pair
(H, G), with a special pairing on the Lie algebra of H, we may construct
an equivariantly-closed three-form on P = H/G and a moment map the-
ory. The three most obvious P which arise this way are G g

/G = g

,
(G G)/G = G, and G
C
/G

= Y , and the moment map theories we will
construct coincide with those which have already arisen in the literature.
Furthermore, if (H, G) is one of these special symmetric pairs with H con-
nected and G simply connected, H/G must decompose into a product of
smaller such symmetric spaces each one of which is isomorphic to k

, K, or
K
C
/K for a subgroup K of G.
These results were announced in [8]. Shortly thereafter, similar results
were related to the author by Yvette Kosmann-Schwarzbach [2]. The au-
thors preprint eventually developed into [9]. He would like to thank his
advisors, Victor Guillemin and Shlomo Sternberg, as well as Eckhard Mein-
renken and Chris Woodward for many useful discussions. The reviewer also
deserves thanks for thoughtful and detailed feedback.
1. Denitions.
1.1. Moment space and moment map. There is a basic notion of a
dierential equation a moment map should obey, regardless of the target, as
well as certain conditions of minimal degeneracy. Here we generalize these
requirements.
SYMMETRIC SPACE VALUED MOMENT MAPS 105
Denition 1. A (possibly degenerate) moment space for G is a pair (P, ),
where P is a G-manifold and is an equivariantly closed three-form on P.
The form may be called the moment form. Since

3
G
(P) =
3
(P)
G
(
1
(P) g

)
G
,
we can write as + , where
3
(P)
G
is the invariant piece and
: g
1
(P) is the equivariant piece. For any G-manifold Q, the vector
eld generated by g will be denoted
Q
. The condition that d
G
= 0
can be written as three equations:
d = 0, (5a)
and for all g,
(
P
) = d(); (5b)
(
P
)() = 0. (5c)
Example 1. Let : g g

(g

) be the map ()() = , ). Then

an equivariantly closed three-form on g

is
g
= d
G
, which has no invariant
part. Written as a map g
1
(g

), it takes the form

g
()

() = , ) , (6)
for each g, g

, and T

= g

.
Example 2. As explained in the introduction, the second example of a
moment space is G itself. Let g have an invariant, positive-denite inner
product B. Then the form

G
() =
G
+
G
()
def
=
1
12
B(
G
, [
G
,
G
]) +
1
2
B(,
G
+
G
) (7)
is equivariantly closed. This is a consequence of the Cartan structure equa-
tions
d =
1
2
[, ] (8a)
d =
1
2
[, ]. (8b)
Example 3. There are two perspectives on the last example of moment
space. The rst connects with the Poisson-Lie G-spaces of Lu and Weinstein.
Let T be a maximal torus for G, t its Lie algebra, and a =

1t g
C
. a is
the Lie algebra of a subgroup A G
C
. Let n be the sum of a set of positive
root spaces. Then there is the Iwasawa decomposition of g
C
:
g
C
= g a n;
G
C
= GAN.
Then the imaginary part of B
C
restricts to a nondegenerate pairing between
g and an, and thus an

= g

. Call the group AN by G

; then G
C
= G

G.
106 MATTHEW LEINGANG
The groups G and G

are both Poisson-Lie groups, and G

the dual Poisson-

Lie group to G. Left multiplication of G on G
C
descends to an action of G
on G

, called the left dressing action.

We do not seek an equivariantly closed three-form on G

, the main dif-

culty arising from the fact that the dressing action does not preserve the
Poisson structure of G

. However, one may also consider the space G

C
/G
as a subspace of G
C
. Set
Y =
_
h G
C

h = h
1
_
.
Y is invariant under the adjoint action of G and equivariantly dieomorphic
to G

, and T
e
Y = p. Let
Y
,
Y

1
(Y, p) be the restrictions of the
Maurer-Cartan forms from G
C
to Y . Then the form

Y
() =
Y
+
Y
()
def
=
1
12
ImB
C
(
Y
, [
Y
,
Y
]) +
1
2

1
B
C
(,
Y
+
Y
) (9)
is real and equivariantly closed.
The equivariantly closed three-form allows us to dene a moment map.
Denition 2. Let M be a G-manifold and P a moment space for G. M is
called a P-Hamiltonian G-space if there exists an invariant two-form

2
(M)
G
and an equivariant map : M P such that
d
G
=

. (10)
The P-Hamiltonian G-space M will further be called nondegenerate if in
addition
ker
x
=
_

M
(x)

ker
(x)
: g T

(x)
P
_
(11)
for all x M.
We may write (10) in terms of its components
d =

; (12a)
(
M
) =

(), (12b)
for all g. For p P,
p
is dened to be the linear map g T

p
P
which takes g to the evaluation of the one-form () at the point p. In
light of (12b), we have that for p P, the fundamental vector elds of all
Lie algebra vectors in the kernel of
p
must annihilate . Thus (11) is a
condition of minimal degeneracy.
Example 4. To revisit Example 1, the condition (12b) applied to
g
=
g

is precisely (1). Equations (12a) and (11) state that must be closed and
nondegenerate, respectively.
Example 5. The conditions on G-valued moment maps are also clearly
generalized by Denition 2 (in fact, one could say this example motivates
the abstract theory). Alekseev-Meinrenken-Malkin show that the moduli
SYMMETRIC SPACE VALUED MOMENT MAPS 107
space of at G-connections on a Riemann surface with r > 0 boundary
components (divided by the action of the restricted gauge group) has the
structure of a Hamiltonian G
r+2
-space with G
r+2
-valued moment map.
Example 6. A G-space M with Poisson action is called a Poisson-Lie G-
space if there is a G-invariant symplectic form and an equivariant map
such that for all g,
(
M
) = 2

Im(,
G
). (13)
Choosing instead to work with Y -valued moment maps, we may apply
Denition 2 to
Y
and we get (4a)-(4c). Alekseev exhibits equivariant dif-
feomorphisms G

= Y , and shows that the corresponding moment map
theories are isomorphic.
For any moment space P, the most immediate candidates for P-Hamil-
tonian G-spaces are the orbits O of G. These have a natural inclusion map
i : O P. Indeed, induces a two-form on each orbit O. If p O, T
p
O is
spanned by
P
(p)[ g, and we dene

O
(
P
(p),
P
(p)) = ()
p
(
P
(p)) . (14)
By (5c), this form is well-dened and alternating, and we immediately see
that it satises (12b). In fact,
O
is characterized by the property that
(
M
)
O
= i

() for all g. We claim

O
is G-invariant, and this is a
consequence of the equivariance of . For, given , g and g G,
g

O
[
p
(
P
(p),
P
(p)) =
O
[
g
p (g

P
(p), g

P
(p))
=
O
[
g
p ((Ad
g
)
P
(gp), g

P
(p))
= (Ad
g
)
gp
(g

P
(p))
= ()
p
_
(g
1
)

P
(p)
_
=
O
[
p
(
P
(p),
P
(p)) .
Therefore, by the relation
0 = L

O
= d(
O
) +(
O
)d
O
,
we must have that
(
O
)d
O
= d(
O
)
O
= di

()
= i

d()
= i

(
O
),
which veries the moment condition (12a). We have proved the following:
Proposition 1. Let P be a moment space for G. Consider a G-orbit O P
with two-form
O
given by (14) and moment map given by inclusion. Then
(O,
O
, i : O P) is a P-Hamiltonian G-space.
108 MATTHEW LEINGANG
We incorporate the minimal degeneracy along orbits into our denition.
Denition 3. A moment space P is called nondegenerate if all orbits O
P are nondegenerate P-Hamiltonian G-spaces with two-form given by (14)
and moment map inclusion. This means that for all p P,
ker
p
g
p
= 0 , (15a)
and for all g,
()
p
_

p
(p)
_
= 0 g = ()
p
0. (15b)
Lemma 1. Let M be a nondegenerate P-Hamiltonian G-space with two-
form and moment map : M P. Let x M. Then:
(a) The map
M
(x), restricted to ker
x
ker
x
, is an isomorphism.
(b) We have ker d[
x
ker [
x
= 0.
Proof. The rst claim is obvious given (15a). For the second, let v
ker d[
x
ker [
x
. Then v =
M
(x) for some ker
(x)
. However, since
0 = d(v) = d(
M
(x)) =
P
((x)),
(the last equality is by the equivariance of ), we must have that g
(p)
.
Hence again by (15b), we have = 0.
Proposition 2. Let M
1
and M
2
be nondegenerate P-Hamiltonian G-spaces
and F : M
1
M
2
an equivariant map such that F

2
=
1
and F

2
=
1
.
Then F is an immersion.
Proof. Since F

2
=
1
, we have that ker dF[
x
= ker
1
[
x
. Also, since
F

2
=
1
, we have
ker
1
[
x

= ker
2
[
F(x)

= ker

1
(x)
.
Thus ker dF[
x
ker
1
[
x
= 0, and therefore dF[
x
is injective.
From this we can prove the P-Hamiltonian Kostant Theorem.
Theorem 1. Let M a transitive nondegenerate P-Hamiltonian G-space.
Then the moment map : M P is a covering map onto an orbit.
Proof. For x
0
M, let O be the orbit of (x
0
) P. Then since M is
transitive, the image of consists of O alone. That is a submersion onto
its image is clear since T
p
O = g
P
(p) and
P
(p) = d(
M
(x)) if p = (x).
Finally, we have that

O
= by (12b) and (14), and applying the previous
proposition, is an immersion as well.
Denition 4. Let M be a P-Hamiltonian G-space. Suppose that o M
is a G-xed point. We dene the reduced space of M at o to be M
o
=

1
(o)/G.
SYMMETRIC SPACE VALUED MOMENT MAPS 109
M
o
has a special two-form on it arising from that on M. To see this, put
Z =
1
(o), and let i : Z M, : Z M
o
be the inclusion and projection.
Denote the P-Hamiltonian two-form on M is by . It is G-invariant, and
therefore so it is restriction to Z. Furthermore, for g,
(
Z
)i

= i

(
M
)
= i

()
= 0,
since i is the constant map o. Thus i

is G-basic; there exists

o

2
(M
o
) such that

o
= i

. Notice also that since

d
o
= d

o
= di

= i

d = i

= 0,
we must have that d
o
= 0.
Theorem 2. Let M be a nondegenerate P-Hamiltonian G-space. Suppose
that dimP = dimG. Then
o
is a symplectic form if and only o is a regular
value of the moment map .
Proof. Let z Z. The map
o
is injective by (15a) and therefore an iso-
morphism since dimP = dimG. Thus [
z
is nondegenerate. Therefore, we
have a commutative diagram
g
o
T

o
P

_
T
z
M

T

z
M
where the horizontal maps are both isomorphisms. It follows from the basic
(ker T

)
0
= imT theorem of linear algebra that
(ker d[
z
)

= g
M
(z).
Now T
z
Z ker d[
z
, by the denition of Z as the inverse image of o. Thus
we have
T
z
(G z) = g
M
(z) = (ker d[
z
)

(T
z
Z)
i

.
The left-hand side of the above is the kernel of

: T
z
Z T
(z)
M
o
, and
the right-hand side is the kernel of i

o
at z. The two are equal (and
o
is
therefore nondegenerate) if and only equality holds in the last step. This is
true if and only if T
z
Z = ker d[
z
, i.e., if z is regular.
2. Manin structure.
A symmetric pair over G consists of a Lie group H G and an involution
of H such that H

= G. Let h be the Lie algebra of H and s the derivative

of at the identity. Then (h, s) is a symmetric Lie algebra, and h

= g. h
110 MATTHEW LEINGANG
has a canonical decomposition h = g p, where p is the 1 eigenspace of s.
We have the commutation relations
[g, g] g; [g, p] p; [p, p] g. (16)
Let P = H/G where, and o = eG. Then P is a symmetric space and p is
canonically identied with T
o
P. See [6].
We will often take the involution to be understood and refer to the sym-
metric pair as (H, G). There are three important example of symmetric
pairs over G.
Example 7. On H
0
= Gg

, the involution
0
is the map (g, ) (g, ).
The corresponding symmetric Lie algebra is h
0
= g g

with Lie bracket

and involution
[(, ), (, )] =
_
[, ] , ad

ad

_
;
s
0
(, ) = (, ).
Example 8. On H
+
= G G, G is embedded as the diagonal. This sub-
group is xed by the involution
+
(g
1
, g
2
) = (g
2
, g
1
). The corresponding
symmetric Lie algebra is h
+
= g g with involution

+
(
1
,
2
) = (
2
,
1
)
xing the diagonal subalgebra.
Example 9. Let G be simply connected as well, so that there is a complex,
simply connected group G
C
with G as its real form. There are the conjuga-
tion automorphisms of G
C
and h

= g
C
= g C singling out the real forms
as their xed point sets.
s

( +

1) =

1.
In order to consider the quotient spaces H/G as moment spaces for G,
we need to pair elements of g with elements of p. The following structure
makes this possible:
Denition 5. Let (h, s) be a symmetric Lie algebra. h will be called a
Manin symmetric Lie algebra if it admits a nondegenerate symmetric bilin-
ear form q with respect to which s is skew-symmetric: I.e., for all
1
,
2
h:
q(s
1
,
2
) = q(
1
, s
2
). (17)
The pairing q will be called a Manin form or Manin pairing. It is also
assumed to be invariant with respect to the adjoint action of h on itself: For
all
1
,
2
,
3
h,
q(ad

1

2
,
3
) = q(
2
, ad

1

3
), or
q([
1
,
2
],
3
) = q(
1
, [
2
,
3
]).
(18)
Let (H, G) be a symmetric pair. H will be called a Manin symmetric
pair if the associated symmetric Lie algebra h admits a Manin form which is
SYMMETRIC SPACE VALUED MOMENT MAPS 111
invariant with respect to the adjoint action of H on h. That is, in addition
to (17) and (18), we must have, for all
1
,
2
h and h H,
q(Ad
h

1
,
2
) = q(
1
, Ad
h
1
2
). (19)
Proposition 3. Let G be a compact, connected Lie group and g its Lie
algebra.
(a) h
0
= g g

has a Manin pairing given by

q
0
((
1
,
1
), (
2
,
2
)) =
1
,
2
) +
2
,
1
) . (20)
(H
0
= Gg

, G) is a Manin symmetric pair. The resulting symmetric

space is isomorphic to g

.
(b) Let g have an invariant inner product B. h
+
= g g has a Manin
pairing given by
q
+
((
1
,
1
), (
2
,
2
)) =
1
2
(B(
1
,
2
) B(
1
,
2
)) . (21)
Since G is connected, (H
+
= GG, (G)) is a Manin symmetric pair.
The resulting symmetric space is isomorphic to G.
(c) Again assume g has an inner product B. Then B extends to a C-
bilinear inner product on h

= g C. h

has a a Manin pairing given

by
q

(
1
,
2
) = ImB(
1
,
2
). (22)
Proof. Clear.
3. Construction of the moment form.
The purpose of this section is to show that given a Manin symmetric pair,
we can construct a moment space. This space will in fact be the space of
right cosets.
For this section (H, G) will be a Manin symmetric pair with involution
and a Manin pairing q. The corresponding involution of h will be denoted
s.
3.1. The equivariant form. Let be the left-invariant Maurer-Cartan
form on H taking values in h. Using s, we can decompose into its g-part
and its p-part, dening:
=
1 +s
2
; =
1 s
2
,
so
1
(H, g) and
1
(H, p). Let j : H P = H/G be the quotient
map.
Proposition 4. Dene for g a one-form
()
h
= q(, Ad
h
)
1
(H). (23)
Then:
112 MATTHEW LEINGANG
(a) () is basic with respect to the right action of G on H, so there is a
unique one-form ()
1
(P) such that j

() = ().
(b) The map () is equivariant with respect to the left action of G
on H, so is an equivariant three-form on P.
(c) We have, for all g,
(
P
)() = 0,
where
P
is the vector eld on P generated by the left action of G in
the direction .
Proof. For h H, let R
h
and L
h
denote left and right multiplication by h
as dieomorphisms of H. Since R

g
= Ad
g
1 and (g) = g, it follows that
R

g
= Ad
g
1 . Then
_
R

g
()
_
h
= q
_
, R

g
Ad
h

_
= q
_
, Ad
hg
Ad
g
1
_
= q (, Ad
h
) = ()
h
,
so () is right-invariant. Moreover, if
R
(h) = (L
h
)

is the fundamental
vector eld associated to the right action corresponding to , then (
R
) = .
Hence (
R
) = 0 and
()
h
(
R
) = 0.
Thus () is also right-horizontal, hence right-basic. This proves the rst
claim of the proposition.
For the second, note that and hence are left H-invariant, so
_
L

g
1
()
_
h
= q(, Ad
g
1 )
= q(Ad
g
, Ad
h
) = (Ad
g
)
h
.
Finally, to prove the third claim, we will show that for g,
(
L
)() = 0, (24)
where
L
is the fundamental vector eld on H associated to the left action.
Indeed,
(
h
)(
L
) = q
_
, Ad
h
Ad
h
1 Ad
(h
1
)
2

_
=
1
2
q(, )
1
2
_
Ad
h1
, Ad
(h
1
)

_
= 0.
(25)

Remark. It is only in (25) that we used the full Ad

H
-invariance of the
pairing q. In fact, the rst two claims of Proposition 4 can be proven with
only a pairing between g and p which is Ad
G
-invariant (note Ad
G
preserves
SYMMETRIC SPACE VALUED MOMENT MAPS 113
the decomposition h = g p). There is a unique extension of such a pairing
to an s-skew pairing of the full Lie algebra, and to force the associated
one-form to obey (24) is to require that
q(Ad
h
, ) = q(, Ad
h
1 ),
for all , g, and for all h of the form (k)k
1
. Such h lie in a submanifold
V =
_
h H

(h) = h
1
_
, which is transverse to G at the identity of H. In
fact T
e
V = p.
3.2. The invariant form. Here we will extend
3
G
(H/G) to an equiv-
ariantly closed three-form.
Proposition 5. Dene
3
(H) by
=
1
3
q(, [, ]). (26)
Then:
(a) is right G-basic and left G-invariant. Hence there exists a unique

3
(P)
G
such that = j

.
(b)
d = 0. (27)
(c) For g,
(
P
) = d(). (28)
Proof. Writing = + as the decomposition of relative to that of h, we
have
d =
1
2
([, ] + [, ]) ;
d = [, ].
This is an immediate consequence of the bracket identities for a symmetric
Lie algebra (16) and the Cartan structure equation (8a). The proposition
reduces to a formal calculation.
(a) This is proved similarly to the analogous claim in Proposition 4.
(b) By the Jacobi identity
_
, [, ]

=
_
, [, ]

=
_
, [, ]

= 0.
Thus,
d =
1
3
dq(, [, ])
= q(d, [, ])
= q([, ], [, ])
= q
_
,
_
, [, ]
_
= 0.
So (27) is proved.
114 MATTHEW LEINGANG
(c) Let g. Then
(
L
) =
1
3
(
L
)q(, [, ])
= q((
L
), [, ])
= q(Ad
h
1 , [, ]).
On the other hand,
d() = dq(, Ad
h
)
= q(, Ad
h
ad

) q(, Ad
h
[, ])
= q(, Ad
h
[ +, ]) q(, Ad
h
[, ])
= q(, Ad
h
[, ]).
Thus (28) is true as well.

As an immediate consequence, we have:

Theorem 3. If (H, G, , q) is a Manin symmetric pair, the equivariant
three-form = + is equivariantly closed, thus giving H/G the struc-
ture of a moment space for G.
3.3. Nondegeneracy. Along with the equivariant condition (Denition 1),
which we have just satised for an arbitrary Manin symmetric pair, there is
the nondegeneracy (actually, minimal degeneracy) condition of Denition 3.
Here we will use the nondegeneracy of the pairing to satisfy nondegeneracy
of .
Proposition 6. Let (H, G) be a Manin symmetric pair, and O an orbit of
G in P = H/G. Then P with two-form given by (14) and moment map
i : O P satises
ker
p
=
_

P
(p)

ker
p
: g T

p
P
_
. (29)
Hence O is a nondegenerate P-Hamiltonian G-space.
Thus:
Theorem 4. Let (H, G, , q) be a Manin symmetric pair. Then P = H/G
is a nondegenerate moment space for G.
Proof. What we are attempting to prove is
(
P
)
p
= 0
_

P
(p) = 0 or
ker
p
.
(30)
Suppose that ker
p
, where p = hG. This means that
0 = q
_
Ad
h
1 ,
1 s
2
Ad
h
1
_
=
1
2
q
_
Ad
h
1 , Ad
(h
1
)

_
(31)
SYMMETRIC SPACE VALUED MOMENT MAPS 115
for all g. Since the q-orthogonal space to g within h is g itself, we have
that
Ad
(h)h
1

g. (32)
Write k = (h)h
1
and note that (k) = k
1
. Then by (32) we must have
that
Ad
k
1 = (Ad
k
) = Ad
k

and therefore = Ad
2
k
or
ker(1 Ad
2
k
).
Now we have a direct sum decomposition
ker(1 Ad
2
k
) = ker(1 Ad
k
) ker(1 + Ad
k
). (33)
If is in the rst summand, we have Ad
h
1 = Ad
(h
1
)
g and therefore

P
(p) = 0. On the other hand, if is in the second summand we have
Ad
h
1 = Ad
(h
1
)
p
and thus ()
h
= 0. Therefore (30) is true.
Proof of Theorem 4. The rst summand in the right-hand side of (33) in-
tersected with g is ker
p
, and the second summand intersected with g is g
p
.
Hence ker
p
g
p
= 0. Since q is nondegenerate, g and p

are isomorphic
as vector spaces. Hence all nondegeneracy conditions are satised.
4. Recovery of the original moment spaces.
Propositions 3, 4, and 5 give equivariantly closed three-forms on each of the
symmetric spaces g

, G, and G
C
/G. In this section we will how the forms
we have constructed here coincide with those developed independently.
Consider rst H
0
= Gg

. The map j
0
: Gg

, given by projection
onto the g

factor, is right G-invariant. Thus it gives a left G-equivariant

dieomorphism between H
0
/G and g

.
Proposition 7. We have j

g
=
0
, where
0
is the form given by applying
Proposition 5 to the Manin form q
0
on h
0
= g g

.
Proof. Let g and h = (g, ) Gg

. Then a tangent vector to h H

0
can be written as L
(g,)
(, ) for some (, ) g g

. We have
(j

g
())
(g,)
_
L
(g,)
(, )
_
=
g
()

_
_
j
0
L
(g,)
_

(, )
_
=
g
()

(Ad

g
) =

, Ad

g

_
.
116 MATTHEW LEINGANG
On the other hand,

0
()
(g,)
_
L
(g,)
(, )
_
= q
0
_
, Ad
(g,)

_
L
(g,)
(, )
__
= q
0
_
, Ad
(g,)
(0, )
_
= q
0
_
(, 0), (0, Ad

g
)
_
=

, Ad

g

_
.

Let H
+
= G G and dene the map j
+
: H
+
G, (g
1
, g
2
) g
1
g
1
2
.
Embed G into H
+
as the diagonal; it acts on H
+
on the left and the right.
j
+
is then seen to be right G-invariant and thus a left G-equivariant dieo-
morphism between H
+
/G and G.
Proposition 8. The map j
+
pulls back the Alekseev-Meinrenken-Malkin
moment form (7) to
+
+
+
, the form constructed on H
+
from the Manin
pairing q
+
.
Proof. We may write
H
+
=
1
G
+
2
G
, etc. Then for each (g
1
, g
2
) H
+
,
j

(g
1
,g
2
)
= Ad
g
2
(
1
G

2
G
);
j

(g
1
,g
2
)
= Ad
g
1
(
1
G

2
G
).
Therefore
j

+
= j

+
1
12
B(
G
, [
G
,
G
]) =
1
12
B(
1
G

2
G
, [
1
G

2
G
,
1
G

2
G
]).
Now =
_

1
G

2
G
2
,

1
G

2
G
2
_
, so

+
=
1
3
q
+
(, [, ])
=
1
24
q
+
_
(
1
G

2
G
,
1
G

2
G
), [(
1
G

2
G
,
1
G

2
G
), (
1
G

2
G
,
1
G

2
G
)]
_
=
1
24
q
+
_
(
1
G

2
G
,
1
G

2
G
),
_
[
1
G

2
G
,
1
G

2
G
], [
1
G

2
G
,
1
G

2
G
]
__
=
1
12
B(
1
G

2
G
, [
1
G

2
G
,
1
G

2
G
]).
Similarly,
J

+
() =
1
2
B(, j

G
+
G
)
=
1
2
B
_
, (Ad
g
2
+Ad
g
1
)(
1
G

2
G
)
_
,
while

+
() = q
+
_
(, ), Ad
(g
1
,g
2
)
_

1
G

2
G
2
,

1
G

2
G
2
__
= q
+
_
(, ),
_
Ad
g
1

1
G

2
G
2
, Ad
g
2

1
G

2
G
2
__
=
1
2
B(, (Ad
g
1
+Ad
g
2
)(
1
G

2
G
)).

SYMMETRIC SPACE VALUED MOMENT MAPS 117

Let j

: G
C
G
C
be the map h hh

. Then j

takes values in Y , is
right G-invariant, and descends to a left G-equivariant dieomorphism of
G
C
/G with Y .
The factor of 2 appearing in (13) is not in its original denition; it is
introduced in [3] to make the theorem connecting Y to G

more clear. Up
to that same factor of 2, we can connect our moment form on G
C
/G to that
on Y .
Proposition 9. The map j

pulls back the moment form

Y
to

,
the equivariantly closed three-form on G
C
arising from applying Propositions
5 and 6 to 2q

.
Proof. Let
G
C
be the left Maurer-Cartan form, and

G
C
its complex conju-
gate. Then for all h G
C
,
j

h
= Ad
h
(
G
C

G
C
) = 2 Ad
h
;
j

h
= Ad
h
(
G
C

G
C
) = 2 Ad
h
.
So
j

Y
=
1
2
ImB
C
_
j

Y
, [j

Y
, j

Y
]
_
=
2
3
ImB
C
(, [, ]) =

.
Likewise, we compute
j

Y
() =
1
2

1
B
C
(, j

Y
+j

Y
)
=
1
2

1
B
C
_
, (Ad
h
+Ad
h
)(
G
C

G
C
)
_
.
Note that Ad
h
+Ad
h
is real and
G
C

G
C
is imaginary, so the above is in
fact real. One the other hand

()
h
= q(, Ad
h
) = 2 ImB
C
(, Ad
h
)
=
1

1
_
B
C
_
,
1
2
Ad
h
(
G
C

G
C
)
_
B
C
_
,
1
2
Ad
h
(
G
C

G
C
)
_
_
=
1
2

1
B
C
_
, (Ad
h
+Ad
h
)(
G
C

G
C
)
_
.

5. Decompositions.
We have shown how Manin symmetric pairs can give rise to moment spaces.
We now show to extent to which the known examples of Manin symmetric
pairs are the only ones.
If G = G
1
G
2
is a direct product of Lie groups, and P
1
and P
2
are
moment spaces for G
1
and G
2
, respectively, then P
1
P
2
with the equivariant
form
1
+
2
is a moment space for G. Thus we have a way of building up
118 MATTHEW LEINGANG
moment spaces. It is natural to try to go the other wayi.e., to decompose.
We start this process at the linear level, and integrate from there.
5.1. Structure of Manin symmetric pairs. We say that a symmetric
pair (H, G) is Riemannian if G acts by isometries on H. We say that a
symmetric Lie algebra (h, s) is orthogonal if g = h
s
is compactly embedded
in h and eective if g h = 0. These last two conditions are satised
whenever (h, s) is the symmetric Lie algebra associated to a Riemannian
symmetric pair (H, G).
Theorem 5. let (h, s) be an eective, orthogonal symmetric Lie algebra
with Manin pairing q. Then there exists a unique canonical decomposition
h = h
0
h
+
h

; (direct sum of ideals) (34a)

g = g
0
g
+
g

; (direct sum of ideals) (34b)

p = p
0
p
+
p

; (direct sum of subspaces) (34c)

such that, with the induced symmetric and Manin pairings given by restric-
tion, we have
h
0
= g
0
p
0

= g
0
g

0
;
h
+
= g
+
p
+

= g
+
g
+
;
h

= g

= g

C.
These isomorphisms are in fact isometries with respect to q.
We will prove this in a series of lemmas. To begin, assume that g is simple.
Let be the negative of the Killing form on h. Then is positive-denite
on g, ad
h
-invariant, and
(s
1
,
2
) = (
1
,
2
),
or,
(s
1
,
2
) = (
1
, s
2
),
for all
1
,
2
h. Dene J : h h by
q(J
1
,
2
) = (
1
,
2
).
Lemma 2. (a) The map J commutes with the adjoint action of h on itself.
That is, for all h,
J ad

= ad

J;
or, for all
1
and
2
,
J[
1
,
2
] = [J
1
,
2
].
(b) The map J anticommutes with s: J s = s J. So J takes g into p
and vice versa.
(c) The map J is self-adjoint with respect to q.
SYMMETRIC SPACE VALUED MOMENT MAPS 119
(d) The restriction J[
g
is a vector space isomorphism g

= p.
Proof. The rst two parts are straightforward. The third is a simple conse-
quence of the symmetry of and of q. The last follows from the fact that
is positive denite on g.
It follows that J
2
is an endomorphism of g as a g-module. By Lemma 2,
Part (c), J
2
is self-adjoint. Therefore, g has an orthonormal basis of eigen-
vectors with real eigenvalues. Since each eigenspace is an ideal of J
2
, it
follows by simplicity that g = g

is a single eigenspace. Thus for all , g,

[J, J] = J
2
[, ] = [, ]. (35)
If = 0, then p is an abelian ideal of h dual by q to g, and hence h

= gg

.
Otherwise, the endomorphism
1

||
J enjoys all the properties of Lemma 2,
so we may assume that [[ = 1. If = 1, the map
T
+
: h g g;
(, J)
1
2
( +, )
is an isomorphism of (h, s, q) onto (h
+
, s
+
, q
+
). On the other hand if = 1
the map
T

: h g C;
(, J) +

1
is an isomorphism onto (h

, s

, q

). This concludes the proof of Theorem 5

in the case that g is simple.
Now if h is eective, then g is at least semisimple. Therefore we have a
decomposition
g =

where is the set of eigenvalues of J

2
. The eigenspaces g

are ideals of g.
For each , let p

= Jg

.
120 MATTHEW LEINGANG
Lemma 3. For each pair of eigenvalues (, ), the following commutation
relations hold:
(a)
[g

, g

]
_
0 if ,= ;
g

if = .
(b)
[g

, p

]
_
0 if ,= ;
p

if = .
(c)
[p

, p

]
_
0 if ,= ;
g

if = .
Proof.
(a) Let g

and g

. Then since J
2
is a g-module homomorphism,
we have that [, ] g

.
(b) Given and as above, notice
[, J] = J[, ] J[g

, g

].
(c) Finally,
[J, J] = J
2
[, ] [g

, g

].

This shows that each h

= g

is an ideal of h. Each h

is isomorphic
to one of the three canonical types, and we can collect them by type. This
proves Theorem 5.
Theorem 6. Let (H, G) be a Manin symmetric pair, with G semisimple
and H connected and simply connected. Then the moment space P = H/G
has a decomposition
P = P
0
P
+
P

and G has a decomposition

G = G
0
G
+
G

such that P
0
is a moment space for G
0
isomorphic to g

0
, P
+
is a moment
space for G
+
isomorphic to G
+
, and P

is a moment space for G

isomor-
phic to (G

)
C
/G

.
SYMMETRIC SPACE VALUED MOMENT MAPS 121
Proof. It follows from the homotopy exact sequence for the bration G
H H/G that if H is simply connected and G is connected, then H/G is
simply connected.
Since G is semisimple, h is an eective, orthogonal, Manin symmetric
Lie algebra. Therefore, we can decompose h as in Theorem 5 into h =
h
0
h
+
h

. Let H
0
H
+
H

be the corresponding decomposition of H.

Likewise g decomposes and we can write G = G
0
G
+
G

. Then
P = H/G =
H
0
H
+
H

G
0
G
+
G

= H
0
/G
0
H
+
/G
+
H

/G

5.2. Relaxing H-invariance. This shows that we have exhausted all pos-
sibilities of creating moment spaces from Manin symmetric pairs, once we
allow suitable assumptions about semisimplicity and connectedness. In fact,
we can relax one of the conditions of a Manin form, weakening a hypothesis
in Theorem 5, and thus arriving at a stronger Theorem 6.
Let (H, G) be any symmetric pair such that H/G is a moment space. Let
= + be the equivariantly closed three-form. Then pulls back to a
linear map : g
1
(H)
G
op
, where we use the op-superscript to denote
the right action of G on H. Evaluating at the identity of H gives a map
b: g h

. Notice that for g G, g, and h,

b(Ad
g
, , ) = (Ad
g
)
e
().
By left-equivariance of , we have
= ()
g
1(L
g
1

).
Because is right-invariant, this is
= ()
e
(R
g
L
g
1

)
= b(, Ad
g
1 ).
Furthermore, again by right-invariance,
b(, ) = ()
e
()
= (),
R
)
e
= 0.
Since h is symmetric and has a canonical decomposition, we can uniquely
extend b to an inner product q on h with respect to which g and p are dual
isotropic subspaces and s is skew-symmetric. This form is not necessarily
completely h-invariant, however, only g-invariant. Nevertheless, this suces.
Theorem 7. Let (h, s, q) be an eective orthogonal symmetric Lie algebra,
with g semisimple and q a Manin pairing assumed to be only g-invariant.
Then there is a canonical decomposition of h as in Theorem 5.
122 MATTHEW LEINGANG
Then we can immediately, using techniques similar to Theorem 6, prove:
Theorem 8. let (H, G) be a Riemannian symmetric pair, with G connected
and H simply connected. Suppose that H/G is a moment space for G. Then
there is a decomposition of H, H, and H/G as in Theorem 6.
Proof of Theorem 7. Assume that g is simple, and complexify h, g, s, and
q. Then J : g p can still by constructed by q(J, ) = (, ). Dene for
, g,
,
def
= [J, J].
Then clearly
[
1
,
2
,
3
] = [
1
,
2
],
3
+
2
, [
1
,
3
] .
Hence , is a homomorphism of g-modules. For simple Lie algebras,
however, all such homomorphisms are scalar multiples of the Lie bracket
(see below). Thus there exists a complex number such that , = [, ]
for all and . But since J is real, must be real, too, and we are in the
same situation as in Theorem 5.
It remains to prove that
Hom
g
_

2
g, g
_
= C[, ].
Since g is simple, it is enough to show that
_
_
2
g
_
ad
= g, where for any
g module M, M
ad
denotes the ad-primary component of M. Though we
may identify g

= g; thus the algebra
_
g (on which g acts preserving the
grading) has a dierential d which is also a g-module homomorphism. Then
as shown by Kostant [7, Theorems D and E], (
_
g)
ad
= A
ad
(
_
g)
g
, where
A is the exterior subalgebra generated by the image of d
1
g
_
2
g. By
restricting to the degree two subspace, we see that
_
_
2
g
_
ad
is the image
of d
1
. But since H
1
(g, C) = 0, d
1
is injective, and so the image of d
1
is
isomorphic to g.
References
[1] A.Yu. Alekseev, On Poisson actions of compact Lie groups on symplectic manifolds,
J. Dierential Geom., 45(2) (1997), 241-256, MR 99b:58086, Zbl 0912.53018.
[2] A. Alekseev and Y. Kosmann-Schwarzbach, Manin pairs and moment maps, J. Dif-
ferential Geom., 56(1) (2000), 133-165, MR 1 863 024.
[3] A. Alekseev, A. Malkin and E. Meinrenken, Lie group valued moment maps, J. Dif-
ferential Geom., 48(3) (1998), 445-495, MR 99k:58062, Zbl 0948.53045.
[4] M.F. Atiyah and R. Bott, The moment map and equivariant cohomology, Topology,
23(1) (1984), 1-28, MR 85e:58041, Zbl 0521.58025.
SYMMETRIC SPACE VALUED MOMENT MAPS 123
[5] V. Guillemin and S. Sternberg, Symplectic Techniques in Physics, second ed., Cam-
bridge University Press, Cambridge, 1990, MR 91d:58073, Zbl 0734.58005.
[6] S. Helgason, Dierential Geometry, Lie Groups and Symmetric Spaces, Academic
Press Inc. [Harcourt Brace Jovanovich Publishers], New York, 1978, MR 80k:53081,
Zbl 0451.53038.
[7] B. Kostant, Cliord algebra analogue of the Hopf-Koszul-Samelson theorem, the -
decomposition C(g) = End V C(P), and the g-module structure of
_
g, Adv. Math.,
125(2) (1997), 275-350, MR 98k:17009, Zbl 0882.17002.
[8] M. Leingang, Symmetric pairs and moment spaces, math.SG/9810064.
[9] , Symmetric Space Valued Moment Maps, Ph.D. Thesis, Harvard, 2000.
[10] J.-H. Lu, Momentum mappings and reduction of Poisson actions, in Symplectic ge-
ometry, groupoids, and integrable systems (Berkeley, CA, 1989), 209-226, Math. Sci.
Res. Inst. Publ., 20, Springer, New York, 1991, MR 92j:58036, Zbl 0735.58004.
[11] J.-H. Lu and A. Weinstein, Poisson Lie groups, dressing transformations, and
Bruhat decompositions, J. Dierential Geom., 31(2) (1990), 501-526, MR 91c:22012,
Zbl 0673.58018.
Received April 27, 2001 and revised October 21, 2002.
Department of Mathematics
Harvard University
Science Center 325
1 Oxford Street
Cambridge, MA 02138
E-mail address: leingang@math.harvard.edu
This paper is available via http://www.pacjmath.org/2003/212-1-8.html.