A NNALI DELLA
S CUOLA N ORMALE S UPERIORE DI P ISA Classe di Scienze
P ETER H EINZNER
L UCA M IGLIORINI
M ARZIA P OLITO
Semistable quotients
Annali della Scuola Normale Superiore di Pisa, Classe di Scienze 4
esérie, tome 26, n
o2 (1998), p. 233-248
<http://www.numdam.org/item?id=ASNSP_1998_4_26_2_233_0>
© Scuola Normale Superiore, Pisa, 1998, tous droits réservés.
L’accès aux archives de la revue « Annali della Scuola Normale Superiore di Pisa, Classe di Scienze » (http://www.sns.it/it/edizioni/riviste/annaliscienze/) implique l’accord avec les conditions générales d’utilisation (http://www.numdam.org/conditions). Toute utilisa- tion commerciale ou impression systématique est constitutive d’une infraction pénale.
Toute copie ou impression de ce fichier doit contenir la présente mention de copyright.
Article numérisé dans le cadre du programme Numérisation de documents anciens mathématiques
http://www.numdam.org/
233
Semistable Quotients
PETER HEINZNER - LUCA MIGLIORINI - MARZIA POLITO (1998),
Let G be a
complex
reductive group, and let X be a(reduced) complex
space with a
holomorphic
action of G. Acomplex
space Ytogether
with aholomorphic
map 7r : X - Y is said to be a semistablequotient
of X withrespect
to the G-action if:_
(i)
7r is a G-invariantlocally
Stein map,and.
(ii) Oy
=If a semistable
quotient exists,
then it isunique
up tobiholomorphism,
andwill be denoted
by XII
G.In the
algebraic category quotients
of this type are often calledgood
quo- tients and have been studiedintensively. They
also arisenaturally
in the contextof Hamiltonian group
actions;
moreprecisely,
let Z be acomplex
space with aholomorphic G-action,
let I~ be a maximal compactsubgroup
ofG,
and assume that there is a moment map M : Z - with respect to a K-invariant Kahlerian structure cv on Z. Then the set X :={z
E Z; G ~0}
of semistable
points
of Z with respect to A is an open G-stable subset ofZ,
and thequotient XllG
exists([H-L], [S]). Moreover,
in the case where Z is aprojective manifold,
it can be shown that X coincides with a subset of semistablepoints
in the sense ofgeometric
invarianttheory, i.e.,
there is anample
G-line bundle L on Z such that X is the set of semistablepoints
withrespect to the linearization induced
by
L(see [H-M]).
The
goal
of this paper is to reduce thequestion
of existence of a semistablequotient
to the case where G is an abelian connected Lie group; moreprecisely
we prove the
following:
THEOREM. The semistable
quotient XII G
existsif
andonly if XII T
existsfor
some maximal
algebraic
torus T in G.The theorem solves a
problem
ofBialynicki-Birula
which heposed during
his stay at the Ruhr-Universitat Bochum and the
University
of Florence.The result is well known in the
algebraic category (see [BB-S l], [BB-S2]),
under the weaker
assumption
that the semistablequotient XIlTo
exists for all The research for this work was partially supported by SFB-237 of the DFG. The second author is also supported by MURST funds.Pervenuto alla Redazione il 5 novembre 1996 e in forma definitiva il 3 luglio 1997.
one-dimensional
algebraic subgroups To
of T. Since it is easy to constructcounterexamples
in theholomorphic setting
to this moregeneral
statement(see
Section
6),
the above theorem is the bestpossible
in theholomorphic
framework.One of the facts which is used in the
proof
of the above mentioned result foralgebraic
actions is that the closure of a G-orbit in X contains a closed orbit. The mainproblem
which arise in theanalytic category
is to showthat,
if the existence ofXIIT
isassumed,
then the G-orbits do not behave toowildly,
e.g. that a G-orbit is open in its
analytic Zariski-closure;
we show thisby using properties
ofsubanalytic
sets.1. - Generalities on semistable
quotients
Let K be a Lie group and X a
complex K-space, i.e.,
X is a reduced Hausdorffcomplex
space with countabletopology
and K acts on Xby
holo-morphic
transformations such that the action K x X 2013~ X,(k, x)
-~ k ~ x, is realanalytic.
Let denote thealgebra
of K-invariantholomorphic
func-tions on
X;
associated with is theequivalence
relation ^~:-X x X ; =
f(x2)
for allf
E Let Jr : X -~X/ -
be thequotient
map. In the case where X is assumed to be a Stein space and K is a compact Lie group one has the
following
result([H]).
The
quotient XI
is a Stein space such thatO(Q)
=for
any open subsetQ of X /
^~.Moreover, XI -
is thecategorical quotient
of X in the category ofcomplex
spaces, and will be denoted
by XIIK.
In order to define a natural extension of this concept, let I~ be a compact Lie group and X a
complex K-space.
We say that acomplex
space Ytogether
with a
surjective holomorphic
map : X ~ Y is a semistablequotient
of X ifevery y E Y has an open Stein
neighborhood Q
such thatn -1 ( Q )
is an open Stein subset of X, and the restriction Jr :n -1 ( Q) -~ Q
induces anisomorphism 7r-’(Q)IIK - Q.
A semistable
quotient
Y of X isunique
up toisomorphism
and will be denotedby
Let K be a
compact
Lie group, and X acomplex K-space
such thatexists. The
following properties
of the semistablequotient
n : X -~follow from the
corresponding properties
in the Steinsetting.
(i) If Aj, j = 1, 2,
are closed K-stablecomplex subspaces of
X, thenn n(A2) _ 7r(Al
nA2).
(ii)
For a closed K -stablecomplex subspace
Zof
X, theimage
is a closedanalytic subspace of XIIK,
the semistablequotient
exists and the em-bedding
Z - X induces anisomorphism Jr(Z).
(iii) If
Y is alocally
closedanalytic subspace of XIIK,
then theembedding
Jr-l (Y)
~ X induces anisomorphism
Y -7r - (Y)//K.
Let K be a compact Lie group, and let G :=
K~
be thecomplexification
of
K, i.e.,
G is acomplex
reductive group with maximalcompact subgroup
K. Since the Lie
algebra
,Cie G of G is thecomplexification
of the Liealgebra
Lie K of
K,
and since K intersects every connected component ofG,
we have= for every
holomorphic G-space X, i.e.,
for everycomplex G-space
X such that the action G x X X isholomorphic. Moreover,
for thesame
reasoning,
every K-stable closedanalytic
subset of X is G-stable.A
holomorphic G-space
X may be considered as acomplex K-space.
If inthis case a semistable
quotient exists,
then we setX // G
:=XIIK;
thismakes sense since
XIIG
does notdepend
on the choice of a maximal compactsubgroup
K of G and n : X -XII G
satisfies the conditions(i)
and(ii)
in thedefinition of a semistable
quotient.
REMARK 1. If
X // G exists,
then we claim thatn (A)
is closed inXIIG
forany G-stable closed subset A of X. Here A is not assumed to be
analytic
as in(ii).
In order to seethis,
one may assume that X is a Stein space; furthermore there is a moment map it on X such that theembedding
~ X inducesa
homeomorphism
Since the claimfollows.
Moreover,
this alsoimplies
thatXIIG
is thecategorical quotient
of Xwith respect to G in the category of
topological
Hausdorff spaces. For moredetails see e.g.
[H-H-K].
Now let G be a
complex
reductive group, and X aholomorphic G-space
such that the semistable
quotient
exists. In the definition of a semistablequotient
the map n : X -~
XII G
isonly required
to be alocally
Stein map; thefollowing
result shows that this is also
globally
the case.THEOREM.
If XII G
is a Stein space, then X is a Stein space.For the
proof
of the Theorem we need thefollowing:
LEMMA. Let n : X - be a semistable
quotient.
For every q EXIIG
there exist an open
neighborhood Q of q,
aG-representation W,
and aG-equivariant holomorphic map 0 :
X ~ W such embeds as a closedanalytic
subset of nyyl ( P ),
where P is a suitable opensubset of Wll
G and Jrw : W --~W //
Gdenotes the
quotient
map.PROOF. If X is a Stein space, then the lemma is
proved
in[H]
Section 6(see
also[Sn]). Thus,
ingeneral,
there exist an open Steinneighborhood Q of q
and aG-equivariant holomorphic
mapOu
from U : := into a G-representation
space W such thatq5u :
U - is a closedembedding,
where P is an open subset of
W// G.
On Y :=
XIIG
we have the sheaf of germs ofG-equivariant
holomor-phic
maps into W,i.e., 7~(6) == { f ; f :
- W is aG-equivariant holomorphic map}
forQ
C Y open; it is a coherent sheaf ofOy-modules ([R]),
andOU
LetIq
denote the ideal sheaf of thepoint
q ; thenOu
definesa
global
sectionOq
of thequotient
sheaf Since Y is assumed to be aStein space, the natural map r :
~-l (Y) -~ (H/12H)(y)
issurjective.
Thereforethere exists a
G-equivariant holomorphic map 0 :
X - W such thatr(q§) = Oq.
The is a
holomorphic
map from U into W withvanishing
ordertwo on
thus
X -~ W is aG-equivariant holomorphic
map suchthat:
is a closed
embedding,
and(ii) 0
is an immersion at everypoint
inThis
implies that,
aftershrinking Q,
themap 0
has the desiredproperties ([H],
Section
6,
see also Section5, Proposition
1 for a moregeneral statement).
DPROOF OF THE THEOREM. It follows
directly
from the lemma that X isholomorphically separable.
We have to prove that X isholomorphically
convex,i.e.,
for a sequence(xn )
in X such that oo, we have to show thatlim f(xn)
= oo for asubsequence
of(xn )
and someholomorphic
functionf :
X C. SinceXII G
is a Stein space, we may assume that converges to q EXII G;
therefore we may assume that xnE ~ -1 ( Q)
for all n, whereQ
is an openneighborhood of q
with theproperties
as stated in the lemma:it follows that
(~ (xn ) )
is a discrete sequence in W.If g :
W - C is aholomorphic
function on W such that = oo, thenf :
:== g hasthe desired
properties.
DREMARK 2. If X is a
holomorphic principal
bundle withcomplex
structuregroup G over a Stein space Y =
XI G,
then X is a Stein space if andonly
ifthe
complex
manifold G is Stein([M-M]).
COROLLARY. Let X be a
holomorphic G-space
such thatXIIG
exists. Thenthe
quotient
map 7r : X -~XII
G is a Stein map,i.e.,
the inverseimage of
a Steinsubspace of XIIK
is Stein.PROOF. If Y is a Stein
subspace
of then the restriction7r (Y)
~ Y is thequotient
map. Thus7r (Y)
is a Stein space. D2. - Saturation
Let G be a
complex
reductive group and X aholomorphic G-space.
For a subset A of X let
SG (A) = {x
E X ; G -x n A =1= 0}
be the saturation of A withrespect
to G. If the ambient space X is relevant for ourconsiderations,
we use the notation A subset A is said to be saturated if
SG (A)
= A.A G-stable subset U of X is said to be
G-complete
if G . x n U = G . x holds for all x E U,i.e.,
if the closure of a G-orbit in U coincides with its closure in X.Note that an open G-stable subset of X is
saturated,
and that a closed G-subset isG-complete. Moreover,
a G-stable subset A of X is saturated if andonly
if U :=X B
A isG-complete.
REMARK. If the semistable
quotient
7r : X -exists,
then a G-stableopen subset U of X is
G-complete
if andonly
if U =i.e.,
if andonly
if it is saturated withrespect
to n(Section
1 Remark1).
Moregenerally,
it follows from Remark 1 in Section 1 that
SG (A) =
for a closedG-stable subset A of X.
If a
holomorphic G-space
X can be covered withG-complete
open G-stable subsets{ Ua }
such thatexist,
then the semistablequotients
can beglued together (Remark)
to apossibly
not Hausdorffcomplex
spaceXIIG,
which hasall
properties
of a semistablequotient
except that the Hausdorff property may fail. This observation is sufficient to show thefollowing
PROPOSITION 1.
If
there exists a G-invariantlocally
Steinmap 0 from
X into acomplex
space Y, then a semistablequotient XII
G exists.PROOF. There exist a
covering
of I’by
open Stein subsets whichgives
us acovering
of Xby G-complete
open Steinsubsets,
because of the G-invariance of the map; we have to prove thatXIIG
isHausdorff, i.e.,
that twodisjoint
closed orbits G . x and G . y have
disjoint G-complete
openneighborhoods.
It suffices to consider the case
where 0 (x) = 0 (y).
LetQ
be an open Steinneighborhood
ofq5(x)
such that U : :=0-’(Q)
isStein;
since aG-complete
subset of U is
already G-complete
in X andUIIG
isHausdorff,
the existence ofdisjoint
openG-complete neighborhoods
follows from the same result in theStein
setting.
C1COROLLARY. Let G be a
complex
reductive group, and let X be aholomorphic G-space. If
a semistablequotient XIIG
exists, thenXIIH
existsfor
anycomplex
reductive
subgroup
Hof
G. 0For an
algebraic analog
of thiscorollary
seeProposition
2.1 in[BB-S 1 ] .
Now let T be a maximal
algebraic
torus inG,
and assume thatXII T
exists. The
following result,
which underlines the close connection between the G-action and the action ofT,
will be used later on. It is based on anobservation of Richardson. For an
algebraic analog
see also[BB-S2].
PROPOSITION 2.
If A is a
closed G-stable subsetof
X, thenPROOF. Given x e X, since G =
K T K,
we have to show thatimplies
for some k E K.
____
Assume that foralIkE K. Then to every y there exist an open
neighborhood Uy
and a T-invariant continuous functionfy
on X such that
fy
> 0 onUy
andfy =
0 on A(Section 1,
Remark1);
butK . x is
compact,
and therefore the sum offinitely
manyfy gives
a continuousT -invariant function
f
on X such that > 0 andf B A
= 0. Thus0
3. - Closures of orbits
If a semistable
quotient XII G exists,
then the closure of a G-orbit containsa closed G-orbit in its
closure;
in this section we show that this remains true if weonly
assume thatXIIT exists,
where T is a maximalalgebraic
torus inG. We show this
by considering special subanalytic
sets in X.Let X be a
(reduced)
realanalytic
space. A subset A of X is said to besubanalytic
in X if for anypoint x
E X there are an openneighborhood
U,finitely
many realanalytic
spacesYi, Zl ,
and properanalytic maps fi : Yi
-~ U, gi :Zi
- U such that A n U =gi ( Zi ) ) .
Note that a closed
analytic
subset of X is asubanalytic
set in X, and thatthe set of
subanalytic
sets in X is closed withrespect
to the finite set theoreticaloperations
oftaking unions,
intersections andcomplements.
Moreover, the in-verse
image
of asubanalytic
set with respect to ananalytic
map issubanalytic.
We also need the
following property
ofsubanalytic
sets([Hi], 3.8.2; [B-M],
Theorem
0.1 ).
*
Let
X - Y be ananalytic
map and A asubanalytic
set in X.If 0 1 A
A -~ Y is proper,
then q5 (A)
issubanalytic
in Y.Here A denotes the
topological
closure of A in X. In theproof
of *Hironaka uses the
Desingularization Theorem;
in ourapplication
we use *only
in the case where X and Y are closedanalytic
subsets in a realanalytic
manifold: in this case the
Desingularization
Theorem can bereplaced by
themore
elementary
Theorem o.1 of Bierstone and Millman in[B-M].
LEMMA. Let X be an irreducible
complex
space, and Z a proper closedanalytic subspace in
X.Let 0
be ameromorphic
mapfrom
X into acomplex
space Y such that:_ ~ ~ S2,
where Q :=X B
Z, isholomorphic. If
D is arelatively
compactsubanalytic
set in X, then00 (D
nS2)
issubanalytic
in Y andOQ
f1S2))
issubanalytic
in X.PROOF.
Since q5
ismeromorphic,
there exist a proper modificationr ~
X such that the restriction PQ :p-l (Q)
-Q,
po := isbiholomorphic
and a
holomorphic
r - Y such thatq§sz = 4$
oPo. 1;
since D n S2 issubanalytic
inX,
the inverseimage
S := issubanalytic
in r. Nowthe
topological closure S
C iscompact
and Y isproper; from * it follows
that (D (S) 00 (D
nS2)
issubanalytic
in Y.In order to show that n
S2))
issubanalytic
in X, set E :=p-1 (Z).
Since p is proper and~-1 (~(S))
issubanalytic
inr,
it follows thatp(~-1 (c~ (S)) B E) =
nS2))
issubanalytic
in X. DWe will now
apply
this lemma in thesetting
where analgebraic
torus Tis
acting holomorphically
on X.PROPOSITION. Let X be a
holomorphic
T -space such that the semistablequotient
7r : X -
XII T
exists, and let A be asubanalytic
set in X such that A -~XII T is
proper Then T - A issubanalytic
in X.PROOF. There are
coverings
and{C~ }
ofXIIT
which have the fol-lowing properties.
(i) Va
is open inXllT,
and isT-equivariantly biholomorphic
to aclosed
analytic
subset of an open semi-stable set in aT-representation W«
for each a,
(ii)
for everyf3
there is some a such thatCp G Va,
(iii) Cp
is a compactsubanalytic
set inXII T
for every~8,
and(iv) (Cp)
is alocally
finitecovering
ofX// T .
For := and
Ap
:= A we have A and r A =U T -
Ap. Since 7r I A :
A -XII T
is proper and islocally finite, A
is acompact subanalytic
set in X andIT - Ap)
islocally finite;
thus it is sufficientto show that T .
A~
issubanalytic
in X. This follows from thefollowing
CLAIM. Let X be an affine
T-variety
and let D be acompact subanalytic
set in X. Then T . D is
subanalytic
in X.We prove the claim
by
induction over the dimension of X. We may assume that X is irreducible.By
a theorem of Rosenlicht([Ro]),
there exist a T-stableZariski-open
subset Q ofX,
aprojective variety
Y and a rationalmap q5
fromX into Y such that
00
isregular,
U := isZariski-open
in Y andU is the
geometric quotient
of Q withrespect
to the T-action. ThusT .
(D
nQ)
= nS2 ) )
issubanalytic
in X(lemma). By induction,
T .
(D
n(X B Q))
issubanalytic
in Z :=X B
Q. Since Z isclosed,
the claimfollows. a
REMARK. In
general
T - A is notsubanalytic
for asubanalytic
set A. Forexample,
let C* act on C x C*by multiplication
on the secondfactor;
thenA :=
I(!, n); n
EN}
isanalytic
in C xC*,
but C* . A is notsubanalytic
inC x C*.
Let G be a
complex
reductive group and T an maximalalgebraic
torus in G.COROLLARY 1. Let X be a
holomorphic G-space
such that the semistablequotient XII
T exists. Then G - A issubanalytic
in Xfor
every compactsubanalytic
set A in X.
PROOF.. Let K be a maximal compact
subgroup
of G such that G = K T K.Then,
since K . A is acompact subanalytic
set, T . K . A is asubanalytic
setin
X;
thus G. A = K . T . K . A issubanalytic
in X. DCOROLLARY 2. Let X be a
holomorphic G-space
such thatXII
T exists. Thenfor
every x E X and y EG ~ x B G ~ x.
PROOF. Since G. x is
subanalytic, dimy (G - x B
G -x) dimx
G - x for everyy E G -
x B
G - x([Hi], 4.8.1).
Thus dim G - y dim G - x follows. DCOROLLARY 3..
Let N
be aholomorphic G-space
such thatXII
T exists. Thenevery G-orbit contains a closed G-orbit in its closure.
°
PROOF. An orbit of minimal dimension in G - x is closed. El
4. -
Proper
actionsIf X is a
holomorphic G-space
such that the semistablequotient exists,
then every closed G-orbit isaffine, i.e.,
theisotropy
group is reductive. In this section we show that this remains true if oneonly
assumes that the semistablequotient
with respect to a maximalalgebraic
torus in G exists.Let G be a Lie group and X a
complex G-space.
The G-action on X is said to be proper if the map G x X ~ X xX, (g, x)
-(g -
x,x),
is proper.This is the case if and
only
if([P]) (i)
the orbit spaceX /
G isHausdorff,
(ii)
every x E X has a compactisotropy
groupGx,
and(iii)
every x E X has a sliceneighborhood
U,i.e.,
a G-stable openneighborhood
U such
that, for
some closedGx-stable
subset Sof
U, the natural map G S -U, [g, s]
- g . s, is ahomeomorphism.
Here G denotes the fiber bundle associated with the
Gx -principal
bundleG ~
G/Gx.
In the
holomorphic
framework the correctanalog
of a compact Lie group is acomplex
reductive group. In order to construct aholomorphic
slice at someorbit,
it is often useful to first consider orbits with a reductiveisotropy
group.Let G be a
complex
Lie group, and let X be aholomorphic G-space.
Letxo E X be a
point
with a reductiveisotropy
group H :=Gxo
and let L be amaximal compact
subgroup
of H.LEMMA 1. There exist a
holomorphic
SteinH-space S,
So E S and an H-equivariant holomorphic
map is S - X such thatis(so) =
xo and the inducedG-equivariant holomorphic map c : G x H S - X, c[g, v]
=locally biholomorphically
onto an openneighborhood of xo.
PROOF. Since H fixes xo, the
tangent
space is anH-representation.
Let
TXOX = Txo (G ~ xo) fl3 V
be anH-equivariant splitting. Using
Cartan’sLinearization
lemma,
one sees that an L-stable openneighborhood
U of xocan be
L-equivariantly
identified with an L-stable closedanalytic
subset A inan L-stable ball B C
TXOX. Let cA :
A -~ U denote such anisomorphism
with xo. Then D :- A n V is an L-stable
analytic
subset of theball
Bv
:= B n V. It follows that S := H ~ D is a closedanalytic
subset ofthe open Stein subset H ~
Bv
of V and that the map :=tAID
extends toan
H-equivariant holomorphic
map is := 5’ -~ X([H],
Section 1.5 Extension lemma and Section 6.6Complexification theorem).
Thus there is an inducedG-equivariant holomorphic
map t : G x H S - X,[g, v]
- g . CLAIM. If U issufficiently small,
then i islocally biholomorphic.
Note that the claim is obvious if xo is a smooth
point.
In thesingular
caseone can argue as follows.
We fix an
L-equivariant isomorphism
cAl from an L-stablelocally analytic
subset
A
I inTxo X
onto an openneighborhood Ul of xo.
Then there are an open L-stableneighborhood
N of 1 EG,
where L-acts on Gby conjugation,
and anopen L-stable
neighborhood
U of xo, such that N ~ U CUl
and A := is closed in an L-stable ball B cTXOX.
The G-action on X induces alocal
actionon
U,
whichgives
a local action on A. In aneighborhood
of 0 E A the localaction is determined
by
aholomorphic map q5 :
N x A -Ai , (g, v) -~ ~ (g , v).
Now let
N
be an open L-stable Steinneighborhood
of eo : = 1 . H Eand i :
N
-~ G aholomorphic
section such that:(i) i(eo)
=1, i (N)
CN,
and(ii)
r isL-equivariant, i.e., i(h. t)
= for all h E L and t EN.
Then $ : N
x A --~Txo X, ~ (t, v) = q5 (-r (t), v)
is anL-equivariant
holo-morphic
map; since A is a closedanalytic
subset ofB,
themap ~
extends toan
L-equivariant holomorphic map li$ : N
x B -Txo X
suchthat ~ (eo, v)
= vfor all V E B. Now the map
GIH G ~ xo
CX,
geo -* g.xo, isan
injective
immersion and V is transversal toxo); therefore,
aftershrinking lil
and B, the xBv
isbiholomorphic
onto itsimage
SZ.Thus
ØD :
1N
x D - xD,
isbiholomorphic
onto its im-age
 := ØD(N x D)
In order to show thatA= QnA,,
wehave to assume that
N
is connected and that every irreduciblecomponent Aa
of Q nA
1 contains 0 E thenDa .
:=Aa
n V is not empty andx
Da )
CAa .
From dimN
xDa -
codim V + dim dim dim(Aa
nV )
+ codim V = dimDa
-~- codim V follows thatØD(N
x =Aa .
This
implies
thatA
= Q nA 1.
Finally
note thatN
x D can be viewed as an open subset of G x HS,
wherebiholomorphic
onN
x D.Equivariance implies
that i islocaly biholomorphic. 0
REMARK 1. For the
proof
of thelemma,
one needs that theimage
ofH =
Gxo
in isreductive;
inparticular,
the statement of the lemma also holds if H is a compactcomplex
Lie group, since in this case theimage
of H in
GL(TxoX)
is compact and therefore finite. In this context theproof
ofLemma 1 is
essentially
due to Holmann([Hol]).
COROLLARY 1.
If the
G-action on X is proper, then every x E X has a G-stable sliceneighborhood, i.e.,
there exists alocally
closedGx
-stablesubspace
Sof X
withx E S such that G . S is open in X and
is
biholomorphic.
PROOF. Since
Gx
acts as a finite group onS,
aftershrinking S,
thereexists an open
neighborhood
N of 1 EG,
stableby Gx
withrespect
toright multiplication,
such thatis an open
embedding. Propemess
of the G-actionimplies
the existence of anopen
Gx-stable neighborhood
V of x in X such thatf g
EG; ~}
cN;
after
replacing
S with S n V we claim that i : G X is an openembedding.
We have to show
injectivity:
for this it is sufficient to provethat g
SI = s2implies g
EGx for g
E G, S.Now, from g
sj 1 = s2 follows that g - V nV #
0 andtherefore g E N . Thus g
=-r(u) -
h for some u EN and
h E
Gx (we
use the notation of theproof
of Lemma1).
Since h EGx,
we have h - s 1 E S and s = s2 =Gx ) ~
s2 . Theinjectivity of BÎ1 implies
-r (u)
= 1 andtherefore g
EGx
follows. 0For the
following
consequence see also[Hol].
COROLLARY 2.
If G
actsproperly
on X, thenXIG
is acomplex
space. Inparticular, if
G is assumed to be acomplex
reductive group which actsproperly
onX, then
XII G
=X / G
exists. 0REMARK 2. Let G be a
complex
reductive group, and let X be a holomor-phic G-space
such that the semistablequotient
exists. Then thecomplex analytic
version of Luna’s slice theorem
(see [H]
or[Sn]) implies
that the G-action onX is proper if and
only
if dim G - x = dim G for all x E X .Now let G be a Lie group, T a Lie
subgroup
ofG,
and assume that there exist compactsubgroups K2
of G such that G =KIT K2.
Let G acttopologically
on X and assume that the T-action on X is proper; then we have thefollowing:
LEMMA 2.
If T
actsproperly
on X, then the G-action on X is proper.PROOF. We have to show that any sequence
(gn , xn )
in G x X such that(gn · xn , xn )
converges to x X has a convergentsubsequence
with~
x0 )
·For
this,
we write gn =kntnhn with kn
E ET, hn
EK2.
We mayassume that
ko - lim kn
andho - lim hn exist;
sincelim(kntnhn)
· xn - Yo, it follows thatlim hn
=ho
xo andlim tn · (hn ~ xn)
= yo. The properness of the T-actionimplies
that asubsequence
of(tn )
converges to to E T ; thus(gn)
converges tokotoho.
DREMARK. If G is a reductive group and T is its maximal torus, the same
statement is
proved
in[Mu], chap.II, Proposition
2.4. foralgebraic
actions.There the use of the
decomposition
G =KIT K 2
isreplaced by
the use of atheorem of Iwahori.
For a
complex
reductive group G with a maximalalgebraic
torus T, thisimplies
the
following (see [BB-S1]
for a similar statement foralgebraic actions):
COROLLARY 3.
If X is a holomorphic G-space
such that thegeometric quotient X /
T exists, then thegeometric quotient X /
G exists. ’ vLet H be a closed
complex subgroup
of the reductive group G. If for X : = the semistablequotient XIIT exists,
then our main result in thisspecial
casesimply
states that H is reductive. For this we need thefollowing
decomposition
theorem forcomplex
linear groups.PROPOSITION 1. Let G be a connected
complex
Liesubgroup of a complex
lineargroup
GL(W).
Then G is a semidirectproduct
G = H - U, where H is acomplex
reductive group and U is a normal solvable
simply
connectedcomplex subgroup of G.
This result is well known. It is the
complex analytic analog
of the samestatement for real Lie groups
[Ho] (p. 223);
for the convenience of the readerwe
give
herethe’ proof
in thecomplex setting.
We need the
following
remark.LEMMA 3. Let G be a connected
complex
Lie group and A a connected closed normalcomplex subgroup of G,
which does not contain a non-trivial compact sub- group.If G/A
isreductive,
then A is a semidirectfactor of
G.PROOF. Let K be a maximal compact
subgroup
of G.Then,
since A isconnected, n ( K )
is a maximal compactsubgroup
ofG / A ([I]),
wheren : G -~
G/A
denotes thequotient
map;thus, by
theassumption
onA,
the restrictionisomorphism.
SinceG/A
is the uni- versalcomplexification
of thecompact
the inversehomomorphism
a :
G,
or := extends to aholomorphic homomorphism
aC : G/A - G.
DPROOF OF PROPOSITION 1. Assume that the commutator
subgroup
G’ of Gis reductive. Then the radical R of G is
abelian,
and therefore R =Lc x V,
where L -=
(Sl)l
is the maximalcompact subgroup
of R, and V is a vector group. Note that theautomorphism
group ofLc- is
finite and thereforeLc is central; thus,
if we write G = SR, where S is asemisimple subgroup
ofG,
the group H =S Leis
reductive and G = H ~ V is a semidirectproduct.
Note that G’ =
G’,
where G denotes the Zariski-closure of G inGL (W).
If G’ is not
reductive,
it has a non-trivialunipotent
radicalRu,
which is closed and normal in G and inG’;
the groupG
:=G/RU
is a linear group, because it is contained in the linearalgebraic
groupApply
induction to thequotient,
i.e.G
=fi . U
is a semidirectproduct,
whereH
is reductive andU
is a
simply
connected solvable normalsubgroup
ofG.
If 7r : G -G
is the natural map, define U : := then is reductive and anapplication
of Lemma 3 shows that G = H ~ U is a semidirect
product.
DPROPOSITION 2.
If for
X : :=G / H
the semistablequotient X ll T
exists, then His a
complex
reductive group. Inparticular,
X isaffine.
PROOF. Let
H °
denote the connected component of theidentity
of H.Since
H°
is a connectedcomplex
linear group, there is a closedcomplex
normal
simply
connected solvablesubgroup
U ofH°,
and acomplex
reductivesubgroup
L ofH°
such thatH°
= L ~ U is a semidirectproduct.
Thus thefibration
G/H°
has L =HOIU
astypical
fiber and therefore it is aStein map; moreover, since the fibration
G /H° - G / H
is acovering,
it is alsoa Stein map: thus
G / H
is a Stein map.Hence,
for Y : := the semistablequotient YIIT exists; but,
since U is solvable andsimply
connectedand the
T -isotropy
groups arereductive,
the T-action on Y is free. Therefore G actsproperly
onGI U,
and thisimplies
U ={e}, i.e., H°
is a reductive group.We have to show that
H/H°
is finite: forthis,
setNo
:=N/H°,
whereN :=
NG(Ho)
denotes the normalizer ofHo
in G. The groups N andNo
are reductive and we may assume that the maximal torusTN
of N is contained in T and thatTNO :
:=TN / TN n H°
is a maximalalgebraic
torus ofN°;
note that the fiberX N : - N/ H
of thefibering
is closed in X =G/H
and thereforeXNIITN
exists. SinceXN = NIH
=(NIHO)I(HIHO) =
Noll’,
where r :=H/H°
is a discretesubgroup
ofNo,
the semistablequotient XNII TNO
exists. ButTNO
is a maximal torus of the reductive groupNo
and ris a discrete
subgroup,
thusTNO
actsproperly
onXN
=N°/ r (Remark 2),
andtherefore
No
also actsproperly:
thisimplies
that r =H/H°
is finite. 0 REMARK. For analgebraic proof
seeProposition
2.3 in[BB-S l].
COROLLARY 4.
If X is a holomorphic G-space
such thatXII
T exists, then everyclosed G-orbit in X is
affine.
El5. - Stein
neighborhoods
For the existence of a semistable
quotient,
it is necessary that every closed G-orbit has a G-stable open Steinneighborhood.
In this section we show thatthis
already
follows if one assumes the existence of a semistablequotient
withrespect to a maximal
algebraic
torus in G.Let K be a
compact
Lie group, and letXj
becomplex K-spaces
such thatthe semistable
quotients exist, j = 1, 2.
Let 7rj :Xj
-~ denotethe
quotient
map. Thefollowing
is a consequence of theholomorphic analog
of Luna’s Slice Theorem
([H],
Section6.3).
PROPOSITION 1.
Ifq5 :
X 1-~ X 2 is a locally biholomorphic K -equivariant map
which maps a K-stable closed
analytic
subsetA 1 biholomorphically
onto a closedanalytic
subsetA2 of X2,
maps aTCl-saturated
openneighborhood of A
Ibiholomorphically
onto a7r2-saturated
openneighborhood of A2.
PROOF. The
map q5
induces aholomorphic
suchthat:
(i) q5
mapsbiholomorphically
onto7r2(A2),
and(ii) 0
islocally biholomorphic along
Since
7rj(Aj)
are closedanalytic
subsets ofXj /1 K,
thisimplies that 0
maps an open
neighborhood Q
1 ofbiholomorphically
onto an openneighborhood Q 2
of7~2(~2).
Thus we may assumethat 0 : X211K
is
biholomorphic; by applying
Luna’s Slice Theoremagain ([H],
Section6.3),
it follows