(Quiver varieties and quantum ane algebras)
中島 啓(HirakuNakajima)
Abstract. この論文は,math.QA/0009231(t-analogueoftheq -characters ofnitedimensionalrepresentationsofquantumane algebras )に
App endixとして箙多様体の定義を付け加えたものである.
もともとのabstract: Frenkel-Reshetikhinintroducedq -characterofnite dimensionalrepresentationsofquantumanealgebras[ 6]. Wegiveacom-
binatorialalgorithm to compute them for all simple mo dules. Our to ol is
t-analogueoftheq -characters,whichissimilartoKazhdan-Lusztigp olynomi-
als,andouralgorithmhasaresemblancewiththeirdenition.
Weneed thetheoryofquivervarietiesforthedenition of t-analogues
andthepro of. Butitapp earonlyinthelastsection.Therestofthepap eris
devotedtoanexplanationofthealgorithm,whichonecanreadwithoutthe
knowledgeab outquivervarieties. Apro ofisgivenonlyinpart. Afullpro of
willapp earelsewhere.
1. The quantumloop algebra
Letgb easimpleLiealgebraoftypeADEoverC,Lg=gC[z;z 01
]b eitslo op
algebra,andU
q
(Lg)b e itsquantumuniversalenvelopingalgebra,orthequantum
lo opalgebraforshort. Itisasub quotientofthequantumanealgebraU
q (b
g),i.e.,
withoutcentralextensionanddegreeop erator. LetI b ethesetofsimplero ots,P
b e theweightlattice, andP 3
b e its dual lattice (all for g). The algebra has the
so-called Drinfeld's newrealization: It is a C(q )-algebra with generators q h
, e
k ;r ,
f
k ;r , h
k ;n
(h2 P 3
, k2I, r2Z, n 2Znf0g) withcertain relations(see e.g.,[ 1,
12.2]).
The algebra U
q
(Lg) is a Hopf algebra, where the coproduct is dened using
the Drinfeld-Jimbo realization of U
q
(Lg). So a tensor pro duct M
C(q ) M
0
of
U
q
(Lg)-mo dulesM,M 0
hasa structureofaU
q
(Lg)-module.
Let U
"
(Lg) b e its sp ecialization at q = " 2 C 3
. For precise denition of
the sp ecialization,we rstintroduce anintegral form U Z
q
(Lg) ofU
q
(Lg) and set
U
"
(Lg)=U Z
q (Lg)
Z[q ;q 01
]
C,where Z[q ;q 01
]!Cisgivenbyq 61
7!"
61
. See[3]
fordetail. Butweassume"isnot a ro otofunityinthispap er. Sowejustreplace
qby"inthedenitionof U
q (Lg).
SupportedbytheGrant-in-aidforScienticResearch(No.11740011),theMinistry ofEdu-
cation,Japan.
2 中島 啓(HirakuNakajima)
The quantumlo op algebra U
q
(Lg) contains the quantumenveloping algebra
U
q
(g)for thenitedimensional Liealgebra gas asubalgebra. Thesp ecialization
U
"
(Lg)containsthesp ecializationU
"
(g)ofU
q (g).
1.1. Finite dimensionalrepresentationsofU
"
(Lg). ThealgebraU
"
(Lg)
contains a commutative subalgebra generated by q h
, h
k;n
(h 2 P 3
, k 2 I, n 2
Znf0g). Letusintroducegenerating functions 6
k
(z)(k2I)by
6
k (z)
def:
= q 6hk
exp 6(q0q 01
) 1
X
m=1 h
k ;6m z
7m
!
:
AU
"
(Lg)-mo duleM iscalledoftype1ifM hasaweightspacedecomposition
as aU
"
(g)-mo dule:
M= M
2P
M(); M()=
n
m2M
q
h
3m="
hh;i
m o
:
Wewillonlyconsidertype1modulesinthispap er.
A type 1 mo dule M is an l-highest weight module ('l' stands for the lo op)if
thereexistsa vectorm
0
2M suchthat
e
k ;r 3m
0
=0; U
"
(Lg) 0
3m
0
=M;
6
k
(z)3m
0
=9 6
k (z )m
0
fork2I
for some 9 6
k
(z) 2 C[[z 7
]]. The pair of the I-tuple (9 +
(z);9 0
(z)) = (9 +
k (z);
9 0
k (z))
k2I
2(C[[z 7
]]
I
) 2
is called the l-highest weight of M, and m
0
is called the
l-highestweightvector.
Theorem 1.1.1(Chari-Pressley[2]). (1) Every nite-dimensional simple
U
"
(Lg)-module of type 1 is an l-highest weight module, and its l-highest weight
isgivenby
9 6
k
(z)="
degP
k
P
k ("
01
=z)
P
k ("=z )
6
(1.1.2)
for some polynomials P
k
(u) 2 C[u] with P
k
(0) =1. Here( ) 6
2 C[[z 7
]] denotes
the expansionatz=1and0 respectively.
(2)Conversely,forgivenP
k
(u)asabove,thereexistsanite-dimensional sim-
ple l-highest weight U
"
(Lg)-module M of type 1 such that the l-highest weight is
givenbythe above formula.
AssigningtoM theI-tupleP =(P
k )
k2I 2C[u]
I
(P
k
(0)=1)denesabijection
betweenthe setof allP'sandthe set ofisomorphism classesof nite-dimensional
simple U
"
(Lg)-modulesoftype1.
Wedenote by L
P
the simpleU
"
(Lg)-module asso ciated to P. Wecall P the
Drinfeld polynomial . Fortheabuseof terminology,we alsosay `P is thel-highest
weightofL
P '.
SinceChq h
,h
k ;n
iisacommutativesubalgebraofU
"
(Lg),anyU
"
(Lg)-mo dule
M decomp oses intoa direct sumM = L
M(9 +
;9 0
)of generalized eigenspaces,
where
M(9 +
;9 0
)
def:
= 8
m2M
( 6
(z)09 6
(z )Id) N
3m=0 fork2I andsucientlylargeN 9
;
for9 6
k
(z)2C[[z 7
]]. ThepairoftheI-tuple (9 +
;9 0
)=(9 +
k
;9 0
k )
k 2I
is calledan
l-weight,andM(9 +
;9 0
)iscalledanl-weight space ofM ifM(9 +
;9 0
)6=0.
Theorem 1.1.3(Frenkel-Reshetikhin[6]). Any l-weight of any nite dimen-
sional U
"
(Lg)-module M of type1 has thefollowing form
9 6
k (z)="
degQ
k 0degR
k
Q
k ("
01
=z )R
k ("=z )
Q
k ("=z )R
k ("
01
=z)
6
(1.1.4)
forsomepolynomials
Q
k (u)=
sk
Y
i=1 (10a
ki u); R
k (u)=
rk
Y
j=1 (10b
k j u):
Againfortheabuseofterminology,wealsosay`Q=Risanl-weightofM'. We
denotethel-weightspaceM(9 +
;9 0
)byM(Q=R).
Frenkel-Reshetikhin[ 6]denedtheq-characterofM by
q (M)
def:
= X
Q=R
dimM(Q=R) Y
k 2I sk
Y
i=1 rk
Y
j =1 Y
k ;a
k i Y
01
k ;b
kj :
Theorem 1.1.5(Frenkel-Reshetikhin[6]). (1)
q
denesaninjectiveringho-
momorphismfromtheGrothendieckringRepU
"
(Lg)ofnitedimensionalU
"
(Lg)-
modules of type 1 to Z[Y 6
k ;a ]
k 2I;a2C 3
(a ring of Laurent polynomials in innitely
many variables).
(2) Ifwe composeamap Y 6
k ;a 7!y
6
k
(forgetting`spectralparameters'), itgives
the usualcharacterof the restrictionof M toaU
"
(g)-module.
Definition1.1.6. Amonomial Y
k 2I sk
Y
i=1 rk
Y
j =1 Y
k ;ak i Y
01
k;bkj
app earingintheq-char-
acter
q
is called l-dominant ifr
k
=0 forallk , i.e., a pro duct of p ositivepowers
ofY
k ;c 'sor1.
If L
P
is the simple U
"
(Lg)-mo dule with l-highest weight P, its q -character
containsanl-dominantmonomialcorrespondingtothel-highestweight. Wedenote
itbym
P
. Itscoecientin
q (L
P )is1.
SincefL
P g
P
formsa basisofRepU
"
(Lg),wehavethefollowingusefulcondi-
tionforthesimplicityof anitedimensionalU
"
(Lg)-moduleM of typ e1:
If
q
(M)containsonlyonel-dominantterm,thenM issimple.
(1.1.7)
1.2. Example. Wegiveexamples ofq -characters.
Ifg=A
n
,we haveanevaluationhomomorphismev
a :U
"
(Lg)!U
"
(g) corre-
sp ondingtoLg!g;z7!a(Jimbo). HencepullbacksofsimpleU
"
(g)-mo dulesare
simpleU
"
(Lg)-mo dules.
Example1.2.1. Letg=A
1
=sl
2
andV b ethe2-dimensionalsimpleU
"
(Lg)-
mo dule. Thentheq-characterofM
a
=ev
a
(M)isgivenby 1
q (M
a )=Y
1;a +Y
01
1;a"
2 :
1
4 中島 啓(HirakuNakajima)
Since
q
isaringhomomorphism,wehave
q (M
a
M
b )=
Y
1;a +Y
01
1;a"
2
Y
1;b +Y
01
1;b"
2
=Y
1;a Y
1;b +Y
01
1;a"
2 Y
1;b +Y
1;a Y
01
1;b"
2 +Y
01
1;a"
2 Y
01
1;b"
2 :
Ifb6=a"
2
;a"
02
,thenM
a
M
b
issimplebythecriterion(1.1.7).
If b = a"
2
or a"
02
, then the second or third term b ecomes 1. In fact, it is
knownthatM
a
M
a"
2 decomp oses(inRepU
"
(Lg))toasumM 0
a 8M
00
,whereM 0
is the3-dimensional simpleU
"
(Lg)-mo dule, and M 00
is the trivial module. Thus
wehave
q (M
a
M
a"
2)
=
q (M
0
a )+
q (M
00
)
=Y
1;a Y
1;a"
2
+Y
1;a Y
01
1;a"
4 +Y
01
1;a"
2 Y
01
1;a"
4 +1:
SeealsoExamples4.1.6,6.1.3,7.2.1.
2. Standardmo dules
2.1. In[15]wedenedafamilyofnitedimensionalU
"
(Lg)-mo dulesoftyp e
1 andcalled them standardmodules. Theyare parametrizedbytheI-tuples P =
(P
k )
k 2I
2 C[u]
I
exactly as simplemo dules. Wedenote by M
P
asso ciated to P.
Thedenitionwillberecalledin x8,but wegiveheretheiralgebraicidentication
duetoVaragnolo-Vasserot[16].
Definition2.1.1. WesayL
P
anl-fundamentalrepresentation if
P
k (u)=
(
10su ifk=k
0 ,
1 otherwise;
forsomes2C 3
andk
0
2I. WedenoteL
P
byL(3
k
0 )
s . (3
k
isthek-thfundamental
weightofg.)
For s 2 C 3
and a nite sequence (k
)
= (k
1
;k
2
;:::) in I and a sequence
(n
)
=(n
1 n
2
:::)ofintegers,weset
M(s;(k
)
;(n
)
)
def:
= L(3
k1 )
"
n
1s L(3
k2 )
"
n
2
s 111:
Note that U
"
(Lg) is not co commutativeHopf algebra,so thetensor pro duct de-
p endsontheorderingoffactors.
Theorem 2.1.2(Varagnolo-Vasserot[16]). (1) A standard module M is iso-
morphictoamodule ofthe form
O
i M(s
i
;(k i
i )
i
;(n i
i )
i )
=M(s 1
;(k 1
1 )
1
;(n 1
1 )
1
)M(s 2
;(k 2
2 )
2
;(n 2
2 )
2
)111 (nite tensor product)
suchthats i
=s j
= 2"
Z
fori6=j andn i
1 n
i
2
::: foreachi.
(2) The above tensor product is independent of the ordering of the factors
M(s i
;(k i
i )
i
;(n i
i )
i ).
(3)TheI-tupleofpolynomialsP correspondingtoM istheproductofDrinfeld
NotethatifP isgiven,wecandeneamo duleMoftheab oveformbydecom-
p osingP intoaproductofDrinfeldp olynomials ofl-fundamentalrepresentations.
Thuswemaydenote theab ovemo dulebyM
P .
Thefollowingprop ertiesofM
P
wereshownin [15]:
(1) fM
P
g isa basisofRepU
"
(Lg).
(2) M
P
isan l-highest weightmo dule with l-highestweightP (i.e.,given by
(1.1.2)).
(3) L
P
istheuniquesimplequotientofM
P .
(4) M
P
dep ends`continuously'onP ina certainsense. For example,dimM
P
isindep endentofP.
(5) fora genericP,M
P
= L
P .
Conjecturally M
P
isisomorphic to thesp ecializationof themo dule V max
(),
introducedbyKashiwara[8],andfurther studiedbyChari-Pressley[4].
3. t-analoguesof q -characters
Amainto ol inthispap eris at-analogueoftheq-character:
q ;t
: RepU
"
(Lg)!Z[t;t 01
][Y 6
k;a ]
k 2I;a2C 3:
This is a homomorphism of additive groups, not of rings, and has the prop erty
q ;t=1
=
q
. Wedene
q ;t
for allstandardmo dulesM
P
. SincefM
P g
P
isabasis
ofRepU
"
(Lg),wecanextenditlinearlytoanynitedimensionalU
"
(Lg)-mo dules.
Forthedenitionweneedgeometricconstructionsofstandardmo dules,sowe
will p ostp one it to x8.3. Wegive an alternativedenition,which is conjecturally
thesameasthegeometricdenition.
3.1. Aconjecturaldenition. LetM=M
P
b eastandardmo dule,Q=Rb e
anl-weightofM,M(Q=R)b ethecorresp ondingl-weightspace. Denealtration
onM(Q=R)by
0=M 01
(Q=R)M 0
(Q=R)M 1
(Q=R)111
M n
(Q=R ) def:
=
\
k Ker(
6
k
(z)09 6
k (z)id)
n+1
:
Conjecture 3.1.1. Thet-analogue
q ;t (M
P
),denedgeometrical lyinx8.3,is
equalto
q ;t (M
P )=
X
Q=R X
n t
2n0d(Q=R ;P)
dim 0
M n
(Q=R)=M n01
(Q=R) 1
m
Q=R
;
where d(Q=R ;P) is an integer (determinedexplicitly fromQ=R , P by (5.1.2) be-
low),andm
Q=R
isamonomialinY 6
k ;a
correspondingtothel-weightspaceM(Q=R ).
Thisdenitionmakessenseforanynitedimensionalmodules,butisnot well-
denedontheGrothendiek group RepU
"
(Lg). Thusthe ab ovedo es notholdfor
simplemo dules.
3.2. Amain resultofthis pap erisa combinatorialalgorithmforcomputing
q ;t (M
P
)and[M
P :L
Q
]. It isdividedintothreesteps:
Step 1: Compute
q ;t
foralll-fundamentalrepsentations.
Step 2: Compute
q ;t (M
P
)forallstandard mo dulesM
P .
Step 3: Expressthemultiplicity[ M
P :L
Q
]intermsof
q ;t (M
R
)forvarious
6 中島 啓(HirakuNakajima)
Step1isamo dicationofFrenkel-Mukhin'salgorithm[5]forcomputing
q ofl-
fundamentalrepresentations. Step2isnothingbutastudyof
q ;t
oftensorproducts
ofl-fundamentalrepresentations. Although
q ;t
isnotaringhomomorphism,
q ;t
of tensorpro ducts isgiven bya simplymo died multiplication. Forthe pro ofwe
useanideain [13]. Step3wasessentiallydonein [15].
4. Step 3
Westart withStep 3. The algorithm issimilar to thedenition of Kazhdan-
Lusztig p olynomials [9]. It is also similar to the algorithm for computing the
transition matrix between the canonical basis and the PBW basis of type ADE
[10].
4.1. Let
A
k;a def:
= Y
k;a"
Y
k;a"
01 Y
l:l 6=k Y
ckl
l;a
;
wherec
kl
isthe(k;l)-entryoftheCartanmatrix.
Definition4.1.1. (1) Let m, m 0
b e monomialsin Y 6
k ;a
(k2I, a2 C 3
). We
deneanorderingamongmonomialsby
mm 0
() m
0
m
isa monimialinA 01
k ;a
(k2I,a2C 3
):
Here amonomial in A 01
k ;a
meansa pro ductof nonnegativep owersofA 01
k;a
. Itdo es
notcontainanyfactorsA
k ;a .
(2)If9 6
;9 06
arel-weightsofnitedimensionalU
"
(Lg)-mo dules,orQ=R ;Q 0
=R 0
arerelatedtol-weightsby(1.1.4),wewrite 9 6
9
06
,Q=RQ 0
=R 0
ifthecorre-
sp ondingmonomialsm,m 0
satisfymm 0
.
Recall that
q (L
P
) contains an l-dominant monomial m
P
corresp onding to
the highest weight vector. It is known that any monomialm app earing
q (L
P ),
q (M
P
)satisesmm
P
([5,4.1],[15, 13.5.2]).
Let
c
QP (t)
def:
= thecoecientofm
Q in
q ;t (M
P ):
Then (c
QP (t))
P ;Q
is upp er-triangular and c
PP
(t) = 1 by the ab ove mentioned
result.
Let( c QP
(t))b e theinversematrix( c
QP (t))
01
. Let
u
RP (t)
def:
= X
Q c
RQ
(t 01
)c
QP (t):
Let b etheinvolutiononZ[t;t 01
]givenbyt 61
7!t 71
.
Lemma 4.1.2(Lusztig[10, 7.10]). There exists a unique solution Z
QP (t) 2
Z[t 01
](QP)of
Z
RP (t)=
X
Q:RQP Z
RQ (t)u
QP (t);
(4.1.3)
Z
PP
(t)=1; Z
QP (t)2t
01
Z[t 01
]forQ<P: (4.1.4)
Thislemmaisprovedbyinduction,andholdsina generalsetting. Lusztighas
Theorem 4.1.5. Themultiplicity[M
P :L
Q
]ofasimple moduleL
Q
inastan-
dardmodule M
P
isequaltoZ
QP (1).
Theproofwillb egivenin x8.4.
Example4.1.6. Let g = A
1 , M
a
= ev 3
a
(M) where M is the 2-dimensional
simpleU
"
(g)-mo duleasbefore. By steps1,2explainedbelow 2
wehave
q ;t (M
a"
2M
a )=Y
1;a Y
1;a"
2+Y
1;a Y
01
1;a"
4 +Y
01
1;a"
2 Y
01
1;a"
4 +t
01
1:
(4.1.7)
LetP(u)=(10au)(10a"
2
u)(i.e.,M
P
=M
a"
2M
a
),Q(u)=1(i.e. M
Q
=trivial
mo dule). Thentheab ovealgorithmgivesusZ
QP (t)=t
01
:
5. Step 1
5.1. Somedenitions. LetM
P
b eastandardmo dule. Letm
P
bethemono-
mial corresp onding to the l-highest weightvector. Let M
P
(Q=R) b e anl-weight
space as b efore. We denote by m
Q=R
the corresp onding monomial. We dene
w
k;a (P);v
k ;a
(Q=R ;P)2Z
0
;u
k ;a
(Q=R )2Zby
m
P
= Y
k2I;a2C 3
Y w
k;a (P)
k ;a
;
m
Q=R
=m
P Y
k 2I;a2C 3
A
0vk;a(Q=R ;P)
k ;a
= Y
k2I;a2C 3
Y
uk;a(Q=R )
k ;a :
Supp osetwostandardmo dulesM
P 1
,M
P 2
andl-weightspacesM
P 1(Q
1
=R 1
)
M
P 1,
M
P 2
(Q 2
=R 2
)M
P 2
aregiven. Wedene
d(Q 1
=R 1
;P 1
;Q 2
=R 2
;P 2
)
def:
= X
k;a 0
v
k;a (Q
1
=R 1
;P 1
)u
k ;a"
01(Q 2
=R 2
)+w
k ;a"
(P 1
)v
k;a (Q
2
=R 2
;P 2
) 1
: (5.1.1)
Wealsodene
d(Q=R ;P) def:
= d(Q=R;P;Q=R;P):
(5.1.2)
Wedenoted(Q 1
=R 1
;P 1
;Q 2
=R 2
;P 2
)alsobyd(m
Q 1
=R 1
;m
P 1;
m
Q 2
=R 2;
m
P 2
).
We needthe followingmodicationof
q ;t
. Write
q ;t (M
P )=
P
m a
m (t) m,
wheremisa monomialanda
m
(t)isitsco ecient. Let
g
q ;t (M
P )
def:
= X
m t
d(m;m
P )
a
m (t)m;
(5.1.3)
whered(m;m
P
)isdenedin(5.1.2).
3
5.2. Frenkel-Mukhin[5,5.1,5.2]provedthattheimageof theq -character
q
iscontainedin
\
k2I
Z[Y 6
l ;a ]
l6=k;a2C 3Z[Y
k;b (1+A
01
k ;b"
)]
b2C 3
:
2
ordirectcalculationforthedenition(8.3.1)
3
Infact,d(m;m
P
)isdeterminedfromam(t)sothatt d(m;m
P )
am(t)isap olynomialintwith
8 中島 啓(HirakuNakajima)
Wehavethet-analogueof thisresult,replacing(1+A 01
k ;b"
) n
by
1+A 01
k ;b"
n
t def:
= n
X
r =0 t
r (n0r)
n
r
t A
0r
k ;b"
;
where[ n
r ]
t
isthet-binomialco ecient. Moreprecisely,wehave
Theorem 5.2.1. (1)Foreachk2I,g
q ;t (M
P
)isexpressedasalinearcombi-
nationof
Y
i Y
ni
k ;bi
1+A 01
k;bi"
ni
t
=Y n1
k ;b1
1+A 01
k ;b1"
n1
t Y
n2
k ;b2
1+A 01
k;b2"
n2
t 111
with coecients in Z[t][Y 6
l;a ]
l 6=k;a2C 3,
where b
i 2 C
3
, n
i 2 Z
>0 with b
i 6= b
j for
i6=j.
(2) If L
P
is an l-fundamental representation (and hence M
P
= L
P ), then
q ;t (M
P
)containsnol-dominantmonomialsotherthanm
P
andtheconditionabove
uniquelydetermines
q ;t (M
P ).
Remark 5.2.2. Thestatement(1)fort=1wasprovedbyFrenkel-Mukhin[5].
And thepro of of (2) is thesame for t =1 andthe generalcase, as illustrated in
thefollowingexamples. Inthissense,(2)should alsob ecredittedtothem.
5.3. Graph. Wegive few examples of
q ;t
of l-fundamentalrepresentations
determinedbytheab ovetheorem.
We attach to each standard mo dule M
P
, an oriented colored graph 0
P . (It
is a slight mo dication of the graph in [6, 5.3].) The vertices are monomials in
q ;t (M
P
). Wedraw ancolorededge k ;a
00!fromm
1 tom
2 ifm
2
=m
1 A
01
k ;a
. Wealso
writethemultiplicityofthemonomialsin
q ;t (M
P ).
Example5.3.1. Letg=A
3
=sl
4 andM
P
=L(3
2 )
1
. Thenthecorresponding
graph 0
P is
Y
2;1 2;"
0000! Y
1;"
Y 01
2;"
2 Y
3;"
1;"
2
0000! Y 01
1;"
3 Y
3;"
3;"
2
?
?
y
?
?
y3;"
2
Y
1;"
Y 01
3;"
3 1;"
2
0000! Y 01
1;"
3 Y
2;"
2Y 01
3;"
3 2;"
3
0000! Y 01
2;"
4 :
Let us explain how we determine this graph inductively. We start with the l-
highestweightY
2;1
. Weknowthat itsco ecientis1. ApplyingTheorem5.2.1(1)
withk=2,wegetY
1;"
Y 01
2;"
2 Y
3;"
withco ecient1. ThenweapplyTheorem5.2.1(1)
withk=1 togetY 01
1;"
3 Y
3;"
. Andso on. All multiplicitiesare1 inthiscase.
Forg=A
n
, itisknownthat theco ecientsof
q ;t (L(3
k )
a
)areall 1.
4
Thus
q ;t (L(3
k )
a )=
q ;t=1 (L(3
k )
a ).
Example5.3.2. Letg=D
4 andM
P
=L(3
2 )
1
. Thegraph0
P
isFigure1. Itis
knownthattherestrictionofM
P toaU
"
(g)-mo duleisadirectsumoftheadjoint
representation and the trivial representation. This fact is reected in
q ;t (M
P )
whereY
2;"
2Y 01
2;"
4
hastheco ecient[2]
t
andallothershas1. Notethatthenumber
4
Moregenerally,iftheco ecientsof
k
inthehighestro otis1,thenthesameholds. This
resulteasilyfollowsfromthetheoryofquivervarieties. Exercise: Checkthisusingtheab ove
Y
2;1
Y
1;"
Y 01
2;"
2 Y
3;"
Y
4;"
Y
1;"
Y 01
3;"
3 Y
4;"
Y 01
1;"
3 Y
3;"
Y
4;"
Y
1;"
Y
3;"
Y 01
4;"
3
Y 01
1;"
3 Y
2;"
2Y 01
3;"
3 Y
4;"
Y
1;"
Y
2;"
2Y 01
3;"
3 Y
01
4;"
3 Y
01
1;"
3 Y
2;"
2Y
3;"
Y 01
4;"
3
Y 01
2;"
4 Y
4;"
Y
4;"
3 Y
1;"
Y
1;"
3Y 01
2;"
4 Y
01
1;"
3 Y
2
2;"
2Y 01
3;"
3 Y
01
4;"
3 Y
01
2;"
4 Y
3;"
Y
3;"
3
Y
4;"
Y 01
4;"
5
Y
1;"
Y 01
1;"
5
[2]
t Y
2;"
2Y 01
2;"
4
Y
3;"
Y 01
3;"
5
Y
2;"
2Y 01
4;"
3 Y
01
4;"
5
Y 01
1;"
3 Y
01
1;"
5 Y
2;"
2 Y
1;"
3Y 02
2;"
4 Y
3;"
3Y
4;"
3 Y
2;"
2Y 01
3;"
3 Y
01
3;"
5
Y
1;"
3Y 01
2;"
4 Y
3;"
3Y 01
4;"
5 Y
01
1;"
5 Y
01
2;"
4 Y
3;"
3Y
4;"
3 Y
1;"
3Y 01
2;"
4 Y
01
3;"
5 Y
4;"
3
Y 01
1;"
5 Y
3;"
3Y 01
4;"
5
Y
1;"
3Y 01
3;"
5 Y
01
4;"
5
Y 01
1;"
5 Y
01
3;"
5 Y
4;"
3
Y 01
1;"
5 Y
2;"
4Y 01
3;"
5 Y
01
4;"
5
Y 01
2;"
6
? 2;"
3;"
2
? 1;"
2
j 4;"
2
? 1;"
2
j 4;"
2
3;"
2
j 4;"
2
3;"
2
? 1;"
2
2;"
3
j 4;"
2
2;"
3
? 1;"
2
3;"
2
? 2;"
3
? 4;"
4
z 4;"
2
? 1;"
4
j 1;"
2
? 2;"
3
3;"
2
? 3;"
4
? 4;"
2
? 1;"
2
? 2;"
3
? 3;"
2
j 2;"
3
j 2;"
3
4;"
4
? 1;"
4
j 3;"
4
? 2;"
3
? 1;"
4
j 3;"
4
4;"
4
j 3;"
4
4;"
4
? 1;"
4
j 3;"
4
? 1;"
4
4;"
4
? 2;"
5
Figure 1. ThegraphforL(3
2 )
1
10 中島 啓(HirakuNaka jima)
ofmonomialsis 28,whichis thedimensionoftheadjointrepresentation. Seealso
Example7.2.3below.
Letusgiveamore complicatedexample.
Example5.3.3. Let g=A
2
and M
P
=L(3
2 )
2
"
L(3
1 )
1
. Although this is
notanl-fundamentalrepresentation,
q ;t (M
P
)hasnol-dominanttermsotherthan
m
P
,so theconditionTheorem5.2.1(1)givesus
q ;t
. Thegraphis Figure2.
Y
1;1 Y
2
2;"
Y 01
1;"
2 Y
3
2;"
[2]
t Y
1;1 Y
1;"
2Y
2;"
Y 01
2;"
3
[2]
t Y
1;1 Y
01
1;"
4 Y
2;"
[3]
t Y
2
2;"
Y 01
2;"
3
Y
1;1 Y
2
1;"
2 Y
02
2;"
3
[2]
t Y
01
1;"
2 Y
01
1;"
4 Y
2
2;"
[3]
t Y
1;"
2
Y
2;"
Y 02
2;"
3
[2]
t Y
1;1 Y
1;"
2
Y 01
1;"
4 Y
01
2;"
3
Y 2
1;"
2Y 03
2;"
3
([3]
t +1)Y
01
1;"
4 Y
2;"
Y 01
2;"
3
Y
1;1 Y
02
1;"
4
[2]
t Y
1;"
2Y 01
1;"
4 Y
02
2;"
3
Y 01
1;"
2 Y
02
1;"
4 Y
2;"
Y 02
1;"
4 Y
01
2;"
3
1;"
j 2;"
2
j 2;"
2
9
1;"
3
1;"
? 2;"
2
? 2;"
2
? 2;"
2
1;"
? 1;"
3
j 2;"
2
2;"
2
? 1;"
3
1;"
? 1;"
3
? 2;"
2
2;"
2
? 1;"
j 1;"
3
2;"
2
Figure 2. ThegraphforL(3
2 )
2
"
L(3
1 )
1
Remark 5.3.4. Aswe cansee inab oveexamples,thecrystal graphsare sub-
graphs of0
P
. Thesetof verticesisthesame,but thesetof arrowsissmaller. We
wouldliketodiscussthis furtherelsewhere.
6. Step 2
6.1. LetM
P
=M(s 1
;(k 1
1 );(n
1
1
))M(s 2
;(k 2
2 );(n
2
2
))111 b eastandard
i j Z
Proposition6.1.1. Wehave
q ;t (M
P )=
q ;t (M(s
1
;(k 1
1 )
1
;(n 1
1 )
1 ))
q ;t (M(s
2
;(k 2
2 )
2
;(n 2
2 )
2 ))111
if s i
=s j
= 2"
Z
fori6=j.
Thusitisenoughto study
q ;t
(M(s;(k
)
;(n
)
))=
q ;t (L(3
k
1 )
"
n
1s L(3
k
2 )
"
n
2
s 111):
Let
q ;t (L(3
k
)
"
n
s ))=
X
r a
m
;r
(t)m
;r
;
wherem
;r
is amonomialinY 6
k ;a anda
m;r
(t)2Z[t;t 01
]isitsco ecient.
Ift=1,
q ;1
isa ringhomomorphism,hencewehave
q ;1
(M(a;(k
)
;(n
)
))=
X
r
1
;r
2
;:::
Y
a
m
r
;r
(1)m
;r
:
Theorem 6.1.2. LetP
be the Drinfeld polynomial of L(3
k )
"
n
s
. Then we
have
q ;t
(M(a;(k
);(n
)))=
X
r
1
;r
2
;:::
t P
; 6d(m;r
;m
P
;m ;r
;m
P
) Y
a
m
r
;r
(t)m
;r
;
wherethe signford(m
;r
;m
P
;
m
;r
;m
P
)is0if and+otherwise.
Example6.1.3. Forg=A
1
, wehave
d(Y 01
1;a"
2
;Y
1;a
;Y
1;a
;Y
1;a
)=1; d(Y
1;a
;Y
1;a
;Y 01
1;a
;Y
1;a"
02)=1
andallothersare0. Thenweget(4.1.7).
IfP =(10au) n
, weget
q ;t (M
P )=
n
X
r =0
n
r
t Y
n0r
1;a Y
0r
1;a"
2
from
q ;t (L(3
1 )
a ) = Y
1;a +Y
01
1;a"
2
. This also follows directly from the deni-
tion (8.3.1) b elow. The t-binomial co ecients app ear as Poincare p olynomials
ofGrassmannmanifolds.
7. Restrition to U
"
(g)
FinitedimensionalsimpleU
"
(g)-mo dulesareclassiedbyhighestweights. Let
ResM
P
b e the restriction of a standard mo dule M
P
to a U
"
(g)-mo dule. It de-
comp oses into a sumof various simplemo dules. Once
q (M
P
) is computed, the
characterofResM
P
isgivenbyreplacingY 6
k ;a byy
6
k
(Theorem1.1.5(2)). Combin-
ing with theknowledgeof charactersofsimplenite dimensional U
"
(g)-mo dules,
wecandeterminethemultiplicityofsimplemo dulesin ResM
P .
CharactersofsimplenitedimensionalU
"
(g)-mo dulesarethesame asthatof
simpleg-mo dules,henceareknown. However,weexpress themintermsof
q ;t in
12 中島 啓(HirakuNaka jima)
7.1. Foradominantweightw= P
w
k 3
k
wedenotebyL
w
thesimplehighest
weightU
"
(g)-mo dulewiththehighestweightw.
Weconsidera standard mo duleM
P
withdegP
k
=w
k
. By the`continuity'of
M
P
onP,ResM
P
dep endsonlyonw
k
=degP
k
,andnotonP itself. Letusdenote
themultiplicityofL
w
0 inResM
P byZ
w 0
;w ,i.e.,
ResM
P
= M
w 0
L 8Z
w 0
;w
w 0
:
We will give a formula expressing Z
w 0
;w
in termsof
q ;t (M
P
). Although we
can give algorithm forarbitary P in principle, the followingchoicewill make the
formulasimple.
Cho oseandxorientationsofedgesin theDynkindiagram. Wedeneinteger
m(k )foreachvertexksothatm(k )0m(l )=1ifwehaveanorientededgefromk
tol,i.e.,k!l . Thenwedene P by
P
k
(u)=(10u"
m(k)
) w
k
:
Let g
q ;t (M
P
)as in (5.1.3). Let e
t (M
P
)2 Z[t]Z[t 6
k ]
k 2I
bea t-analogue of
theordinarycharacterwhichisobtained fromg
q ;t (M
P
)bysendingY 6
k ;a toy
6
k .
Foranotherdominantweightw 0
= P
w 0
k 3
k ,let
c
w 0
;w (t)
def:
= theco ecientof Y
y w
0
k
k in e
t (M
P ):
The matrix(c
w 0
;w (t))
w 0
;w
is upper-triangular with resp ect to the usual order on
weights,anddiagonalentriesareall1.
Theorem 7.1.1. c
w 0
;w
(0)isthe weightmultiplicity ofw 0
inthehighestweight
moduleL
w
withthe highestweightw .
Thisisjustasimplerephrasingofa mainresultin [12,14]. Theproofwillb e
giveninx8.5.
Notethat c
w 0
;w
(1)givestheweightmultiplicityof w 0
in ResM
P
sinceg
t=1 is
theordinarycharacter. Thuswehave
c
w 00
;w (1)=
X
w 0
c
w 00
;w 0
(0)Z
w 0
;w :
ThisequationdeterminesthemultiplicityZ
w 0
;w
onlyfromtheknowledgeof
q ;t .
AccordingtoaconjectureofLusztig[11]togetherwithaformula(8.5.1)b elow,
c
w 0
;w
(t)shouldb ewrittenbyferminonicformofHatayamaelal.[7]. Moreprecisely,
we should have P
w 0
c w
00
;w 0
(0)c
w 0
;w
(t) = M(w ;w 00
;t 2
); where (c w
00
;w 0
(0)) is the
inversematrixof(c
w 00
;w
0(0)). See[11]for thedentionofM(w ;w 0
;q). Although
thisformulacanb echeckedinmanyexamples,thecomplexityofthecombinatorics
preventus from provingit in full generality. Conjecturally M(w ;w 0
;q=1)gives
us the multiplicities of the restriction of M
P
(Kirillov-Reshetikhin 5
). Thus the
conjectureiscompatiblewithourresultinthissection.
7.2.
Example7.2.1. Letg=A
1
andw=23
1
. WetakeP =(10u) 2
bytheab ove
choice. By Example6.1.3,wehave
e
t (M
P )=y
2
1
+(1+t 2
)+y 02
1 :
5
Thus Z
0;w
= 1. Since Res(M
P ) = L
31 L
31
= L
231 8L
0
, this is the correct
answer!
Example7.2.2. Letg=A
3
andw=3
2
. ByExample5.3.1alltheco ecients
of
q ;t (L(3
2 )
1
)are1. HenceResM
P
=ResL(3
2 )
1
issimpleas aU
"
(g)-module.
Example7.2.3. Let g = D
4
, w = 3
2 , w
0
= 0. By Example 5.3.2 wehave
c
w 0
;w
(t)=4+t 2
. ThusZ
w 0
;w
=1,i.e. Res(L(3
2 )
1 )=L
32 8L
0 .
8. Quiver varieties
Inthissection, wegivethe denitionof
q ;t
and proveTheorems 4.1.5,7.1.1.
Aswementioned,thosepro ofsareessentiallygivenin[12,14]and[15]resp ectively.
Theonlythingswedoherearetranslationofresultsintothelanguageof
q ;t . We
b elievethat thissectiongivesgo o d introductionsto [12, 14,15].
8.1. Let w = P
w
k 3
k (w
k 2 Z
0
) b e a dominant weight of the nite di-
mensional Lie algebra g. In [12, 14, 15], we have attached to each w , a map
:M(w)!M
0
(1;w)withthefollowingproperties:
(1) M(w ) isa nitedisjointunionof nonsingularquasi-projectivevarietiesof
variousdimensions.
(2) M
0
(1;w )isananealgebraicvariety.
(3) isa projectivemorphism.
(4) There exist actionsof G
w 2C
3
on M(w ) and M
0
(1;w ) such that is
equivariant.
(5) M
0
(1;w )isacone,andthevertex(denotedby0)istheuniquexedp oint
oftheC 3
-action(restrictionofG
w 2C
3
-actiontothesecondfactor).
HereG
w
= Q
k2I GL(w
k
;C).
Weconsidertheb erpro duct
Z(w ) def:
= M(w)2
M0(1;w ) M(w ):
The convolutionpro duct makes the (Borel-Mo ore) homology group H
3
(Z(w);C)
into an associative (noncommutative) algebra. One of main results in [14] is a
constructionofasurjectivealgebra homomorphism
U(g)!H
top
(Z(w );C);
whereU(g)istheuniversalenvelopingalgebraofg(NB:nota`quantum'version).
Here H
top
( ) meansthe degree =dim
R
Z(w) part ofthe homology group. More
precisely,wetakedegree=dimensionpartoneachconnectedcomp onentsofZ(w),
and then make the direct sum. Note that the the dimension diers on various
comp onents.
LetL(w )= 01
(0). ItisknownthatM(w )hasaholomorphicsymplecticform
suchthat L(w ) is alagrangian subvariety. TheconvolutionmakesH
top
(L(w );C)
(thetopdegreepartoftheBorel-Mo orehomologygroup,inthesamesenseasabove)
intoanH
top
(Z(w);C)-mo dule. Itis aU(g)-mo dulebytheab ovehomomorphism.
By [14, 10.2] it is the simple nite dimensional U(g)-module L
w
with highest
weight w . And connected components M(v;w ) of M(w) are parametrized by
vectorsv= P
v
k
k (
k
is thek th simplero otofg)so that
H
top
(L(w );C)= M
H
top
(M(v ;w)\L(w );C)
14 中島 啓(HirakuNaka jima)
istheweightspacedecomp osition ofthesimplehighestweightmo duleL
w , where
H
top
(M(v ;w)\L(w );C) has weightw0v . In particular, v =0 corresp onds to
thehighestweightvector. Infact,M(0;w )isconsistingofasinglep oint.
ThespaceM
0
(1;w )hasa stratication
M
0
(1;w)= [
M reg
0
(v ;w);
where v runs over theset of vectors such that w0v is a weightof L
w
which is
dominant [12,x3].
8.2. LetusgivetheU
q
(Lg)-versionof theconstructionoftheprevioussub-
section.
Weuse the followingnotation: Let R(G) denote therepresentation ring ofa
linearalgebraic group G. If Gactsa quasi-projectivevarietyX, K G
(X) denotes
theGrothendieckgroupofG-equivariantcoherentsheavesonX.
The representation ring R(G
w 2C
3
) of G
w 2C
3
is isomorphicto the tensor
pro duct R(G
w )
Z R (C
3
). Moreover,R(C 3
)is isomorphicto Z[q;q 01
], whereq is
thecanonical1-dimensionalrepresentationofC 3
.
TheconvolutionmakestheGrothendieckgroupK G
w 2C
3
(Z(w ))intoaR(G
w 2
C 3
)=R (G
w )[q;q
01
]-algebra. Oneof main resultsin [15] is a constructionof an
algebra homomorphism
U Z
q (Lg)
Z R(G
w )!K
G
w 2C
3
(Z(w ))=torsion:
By the equivariance of , L(w) = 01
(0) is invariant under G
w 2C
3
. The
convolutionmakesK G
w 2C
3
(L(w))intoaK G
w 2C
3
(Z(w ))-mo dule. Moreover,itis
freeofniterankoverR(G
w 2C
3
)[15, x7]. Itis aU Z
q (Lg)
Z R (G
w
)-moduleby
theab ovehomomorphism. By[15,x13],itcontainsavectorm
0
suchthat
e
k ;r 3m
0
=0;
0
U Z
q (Lg)
0
Z R(G
w )
1
3m
0
=K Gw2C
3
(L(w ));
6
k
(z)3m
0
=q wk
V
01=q z q
01
W
k
V
0q =z q
01
W
k
!
6
3m
0
fork2I:
(8.2.1)
The right hand side ofthe third equation needsan explanation: First W
k is the
vectorrepresentationofGL(w
k
;C),consideredasaG
w 2C
3
-module. Then V
u V =
P
u i
V
i
V. Since V
0q =z q
01
W
k
is 10(1=z)W
k
+::: (1 is the trivial mo dule), we
can dene
V
0q =z q
01
W
k
01
as a formal p ower series in 1=z . This gives us the
case ( ) +
of the above formula. In the case ( ) 0
, we expand as V
0q =z q
01
W
k
=
(01=z) wk
V
wk
W
k 0z
V
wk01
W
k +111
. Then V
wk
W
k
is an invertible element,
wecanalsodene
V
0q =z q
01
W
k
01
. Thevectorm
0
isthecanonicalgeneratorof
K Gw2C
3
(M(0;w )). (RecallM(0;w )isa p oint.)
The mo dule K G
w 2C
3
(L(w )) should b e considered as a `universal' standard
mo dulesincestandardmo dulesareobtainedfromitbysp ecializationsasweexplain
now.
Leta=(s;")2G
w 2C
3
beasemisimpleelement. Itdenesahomomorphism
a :R (G
w 2C
3
)!Cbysendinga representationtothevalueofthecharacterat
a. Then
K Gw2C
3
(L(w))
R (G 2C 3
) C (8.2.2)
isa mo duleoverU
"
(Lg)=U Z
q (Lg)
Z[q ;q 01
]
C. By(8.2.1)itisa nite-dimensional
l-highestweightmo dule. Thisisthestandardmodule M
P
,whereP
k (u)=
a (
V
0u
q 01
W
k
). Notethatthesetofconjugacyclassesofa=(s;")bijectivelycorresp onds
tothesetofI-tupleofp olynomialsP withdegP
k
=w
k .
8.3. LetAb etheZariskiclosureofa Z
inG
w 2C
3
. Itisanab elianreductive
group. WehaveK G
w 2C
3
(L(w ))
R(G
w 2C
3
) R (A)
= K
A
(L(w ))[15,x7]. Since
a
factorsthroughR (A),thestandardmoduleM
P
isisomorphictoK A
(L(w ))
R(A) C.
By Thomason's lo calization theorem, it is isomorphic to K(L(w) A
)
Z
C, where
L(w) A
is thexed p oint set. Furthermore, it is isomorphicto H
3 (L(w )
A
;C)via
theCherncharacterhomomorphism[15,x7].
Let A
: M(w) A
! M
0 (1;w)
A
be the restriction of the map : M(w ) !
M
0
(1;w ) to the xed point set. Let M(w ) A
= F
M() be thedecomp osition
intoconnected comp onents. EachM() is a nonsingular quasi-projective variety.
Thenwehavethedirect sumdecomposition
M
P
= H
3 (L(w )
A
;C)
= M
H
3
(M()\L(w );C):
In [15, x13, x14]we haveshownthat this is the l-weight spacedecomposition of
M
P
. Inparticular, theindex can b econsidered as anl-weightofM
P
. Thuswe
havearrivedat ageometricinterpretationof
q :
q (M
P )=
X
dimH
3
(M()\L(w );C)m
;
wherem
isthemonomialcorresp ondingtothel-weight.
Nowwedenethet-analogue
q ;t by
q ;t (M
P )
def:
= X
X
k dimH
k
(M()\L(w );C)t
k0dimCM()
m
: (8.3.1)
By[15, x14]wehaveastratication
M
0 (1;w )
A
= [
M
reg
0 ();
consisting ofnonsingularlocally closedsubvarieties. Heretheindex set fgisthe
subsetoftheab oveindexset consistingofl-dominant l-weights.
8.4. Proof of Theorem 4.1.5. Thel-highest weightP is xed throughout
the pro of. Thus thedominant weightvector w and the element a=(s;")2 G
w
arexed.
Wechangethenotationnow. Ifcorresp ondstoanl-weightspaceM
P (Q=R),
we denote ab oveM() by M(Q=R;P). Wealso denote by M reg
0
(Q;P) for above
M reg
0
() ifcorresp onds toanl-dominantl-weightQ. Thuswehave
M(w ) A
= G
Q=R
M(Q=R;P); M
0 (1;w)
A
= [
Q M
reg
0
(Q;P):
InthisnotationH
3
(M(P ;P) \L(w );C)isthel-highestweightspace. SinceM(0;w )
is a single p oint as we explained, we have M(P ;P) = M(0;w ). We also have
M reg
(P ;P)=f0g.
16 中島 啓(HirakuNaka jima)
Lemma 8.4.1. (1)dim
C
M(Q=R ;P)=d(Q=R ;P). dim
C M
reg
0
(Q;P)=d(Q;P).
(2)If M reg
0
(Q;P)M reg
0
(R;P), thenRQ.
(3) Choose x 2 M reg
0
(Q;P). Then ( A
) 01
(x)\M(S=T;P) is isomorphic to
M(S=T;Q)\L(w).
Proof. (1)Therstequationisthedimensionformula[15,4.1.6]. Thesecond
equation followsfrom dim
C M
reg
0
(Q;P) =dim
C
M(Q;P), whichis clear from the
denition [15,x4].
(2),(3) The resultsare known or trivial for Q =P. Now usethe transversal
slice atx2M reg
0
(Q;P)[15,x3] toreducea generalcasetothiscase.
LetD b
(M
0 (1;w )
A
)b etheb oundedderivedcategory ofcomplexesofsheaves
such that cohomology sheaves are constantalong each stratumM reg
0
(Q;P). Let
IC(M reg
0
(Q;P)) b e the intersection homology complex asso ciated with the con-
stant lo cal systemC
M reg
0 (Q;P)
onM reg
0
(Q;P). By using thetransversal slice [15,
x3], one can check that it is an object in D b
(M
0 (1;w )
A
). Let C
M(Q=R;P) be
the constant lo cal system on M(Q=R ;P). Then A
3 (C
M(Q=R;P)
) is an object of
D b
(M
0 (1;w )
A
) again by the transversal slice argument. Using the decomp osi-
tion theorem of Beilinson-Bernstein-Deligne, we have shownthat there exists an
isomorphism inD b
(M
0 (1;w )
A
):
A
3 (C
M(R;P) [dim
C
M(R;P)])
= M
Q;k L
Q;k
(R;P)IC(M reg
0
(Q;P))[k ] (8.4.2)
for some vector space L
Q;k
(R ;P) [15, 14.3.2]. Since A
(M(R ;P)) M reg
0
(R;P)
bydenition[15, x4],thesummationruns overQRbyLemma 8.4.1. Let
L
RQ (t)
def:
= X
k dimL
Q;k (R;P)t
0k
:
Applying theVerdierduality to the b oth hand side of (8.4.2) and using the self-
duality of A
3 (C
M(R;P) [dim
C
M(R;P)]) and IC(M reg
0
(Q;P)), we nd L
RQ (t) =
L
RQ (t).
Cho ose a p oint x
Q
from M reg
0
(Q;P) for each stratum. Let i
xQ : fx
Q g !
M
0 (1;w )
A
denotetheinclusion. Consider
H k
(i
!
x
Q
A
3 C
M(R;P)
[dimM(R;P)])=H
dim
C
M(R;P)0k ((
A
) 01
(x
Q
)\M(R;P);C):
ByLemma8.4.1(3)thisisisomorphictoH
dimCM(R;P)0k
(M(R;Q)\L(w);C).There-
forewehave
X
k dimH
k
(i
!
xQ
A
3 C
M(R;P)
[dimM(R ;P)])t dim
C
M(Q;P)0k
= X
d dimH
d
(M(R;Q)\L(w );C)t d+dim
C
M(Q;P)0dim
C M(R;P)
=c
R Q (t);
(8.4.3)
whereweuseddim
C
M(R;P)0dim
C
M(Q;P)=dim
C
M(R ;Q)inthelastequality.
By[15, 14.3.10],wehave
[M
Q :L
R
]=dimH 3
(i
!
xQ IC(M
reg
0
(R;P))):
(Infact, we denedthestandard mo duleM
Q as H
3 ((
A
) 01
(x
Q
);C)in [15, x13],
However,byusingthetransversalslice,wecanshowthattherighthand sideisthe
same forbothdenitions. cf.Lemma 8.4.1.)
Let
Z
RQ (t)
def:
= X
k dimH
k
(i
!
x
Q IC(M
reg
0
(R ;P)))t dimM
reg
0
(Q;P)0k
:
Wehave[M
Q :L
R ]=Z
R Q
(1). By thedeningpropertyoftheintersectionhomol-
ogy,Z
R Q
(t)satises(4.1.4).
Substituting(8.4.2)into(8.4.3), weget
c
SQ (t)=
X
R L
SR (t)Z
RQ (t):
NowL
S R (t)=L
S R
(t)implies(4.1.3). Thiscompletesthepro ofofTheorem 4.1.5.
8.5. Proof of Theorem 7.1.1. By theresult explainedin x8.1, the weight
multiplicityofw 0
inL
w
isequal to
dimH
top
(M(w0w 0
;w )\L(w );C):
Theassertionfollowsfrom moregeneralformula
e
t (M
P )=
X
w 0
X
d H
d
(M(w0w 0
;w)\L(w);C)t dim
C M(w0w
0
;w)0d Y
k y
w 0
k
k : (8.5.1)
NotethatM(w0w 0
;w )\L(w )isalagrangiansubvarietyinM(w0w 0
;w),sowe
havetop=dim
C
M(w0w 0
;w ).
Inorder toprove(8.5.1), we use[12, 5.7],where theBetti numbersare given
in termsof those of xed p oint comp onents. It lo oks almost thesame as ab ove.
However,there is one signicantdierence. TheC 3
-actionused there is dierent
fromourC 3
-actionusedhere,denedin[15,x2]. Thisisthereasonwhywechoose
P and corresp ondinga=(s;")as explainedin x7. Then A=a Z
is isomorphicto
C 3
andtheactionisthesame astheC 3
-actionconsideredin [12,x5].
We decomp ose M(w ) A
= F
M() intoconnected components as b efore. By
[12, 5.7]wehave 6
dimH
d
(M(w0w 0
;w )\L(w );C)= X
dimH
dim
C M(w 0w
0
;w )0d
(M();C);
wherethesummationruns overthesetofsuchthatthecorresp ondingmonomial
m
issentto Q
k y
w 0
k
k
after Y
k ;a
! y
k
. TheC 3
-action makesM
0 (1;w)
C 3
=f0g,
soM()=M()\L(w ). Hencetheaboveexpressioncoincideswiththedenition
ofthecoecientse
t .
Acknowledgement. Wewould liketo thank E. Frenkel and E. Mukhin for
explanationsoftheiralgorithmcomputingq -charactersoffundamentalrepresenta-
tions.
6
Infact, thisformulaevenholdsforgeneralP if wereplacedim
C
M(w0w 0
;w)0dbya
suitabledegree. However,thisdegreeshiftisgivenbyacomplicatedexpressionin.Soourchoice
18 中島 啓(HirakuNaka jima)
App endix A. 箙多様体
A.1. 箙(quiver). quiverQは,頂点の有限集合Iと(向きのついた)辺の集合
であって,辺の始点と終点を対応させる写像out:!I,in:!Iが与えられて いるもののことである. すなわち,有限グラフの辺に向きを入れたものに他ならない.
(図3参照)
h
in(h) out(h)
Figure 3. quiverの例 このとき道(path)とは, 辺の有限列(h
1
;h
2
;:::;h
N
)でin(h
1
) = out(h
2 ), :::,
in(h
N01
)=out(h
N
)が成り立つもののことを言う. Nを道の長さという. さらに長 さNが0の道,すなわちひとつの頂点k2Iだけからなるものも道であると約束する.
Fを体とするとき,道代数(pathalgebra) FQを道を基底とするベクトル空間に 道の合成によって積を入れたものとする. もしも二つの道が合成できないとき,すな わちそれぞれの終点と始点とが一致しないときは,積は0と定める. 例えば下の図4 の定める道代数は,一変数多項式環F[x]に他ならない. 一般には,道代数は非可換の 環になる.
Figure4. F[x]に対応するquiver
またIをFQの両側イデアルとするとき, 道代数を関係式Iで割った環FQ=Iも よく考える. 例えば,上の例でm回まわるループhmを0とする関係式を考えれば, 対応する環はF[x]=xmに他ならない.
quiverの表現とは,道代数FQの左加群のことを言う. 言い直せば,各頂点k2Iに 対してF-ベクトル空間Vkが与えられ,各辺h2に対して線型写像Bh
:V
out(h)
!
V
in(h)
が与えられているもののことである. またFQ=Iの表現は, 上のようなBhに
Iが定める条件を課したものに他ならない.
したがって, quiverの表現の同型類(ただし次元が同じものだけを考える)の集合
(すなわち, 表現のモジュライ空間)は,
M
h2 Hom(V
out(h)
;V
in(h) )
,
Y
k2I GL(V
k ) (A.1.1)
に他ならない. ベクトル空間をE (V),
Q
k 2I GL(V
k
)をGV で表わす.
表現のモジュライ空間(A.1.1)は,Riemann面上の正則ベクトル束のモジュライ空間 と類似点がある. 実際後者は,(無限次元の)アファイン空間を(やはり無限次元の)Lie 群で割った空間になっている.
relationの例であとの箙多様体と深く関係するものがあるので説明する. Q#を
Q=(I;)のダブルとする. すなわち,Iは変えず, の辺hにその向きを逆にした ものhを付け加えたものである. このとき,Q#の辺 hに対し,"(h)をh2のとき に1,h2のときに01と定める. そこで,
X
in(h)=k
"(h)hh=0; (k2I)
で生成される両側イデアルを考えIとおく. FQ#=Iがpreprojectivealgebraと呼ば れるものである. その表現の同型類は,
8
<
: (B
h )2E
[ (V)
X
in(h)=k
"(h)B
h B
h
=0 9
=
; ,
G
V (A.1.2)
で与えられる. E
[
(V)=E
(V)8E
(V)は シンプレクティックベクトル空間7で,
P
"(h)B
h B
h
は GV の作用に関するモーメント写像である. シンプレ クティック形 式やモーメント写像については, 詳しい一般論がこの論説で必要になるわけではな いが,あとに出てくる幾何学的不変式論やhyper-Kahler構造との関連も含めて,[19, 第3章]を参照されればありがたい. よって上の空間は, もしも商空間が多様体の構 造を持てばいわゆるシンプレ クティック商と捉えることが出来る. またシンプレ ク ティック商の一般論によれば,(A.1.1)のE
(V)=G
V が多様体になれば,その余接束
T 3
(E
(V)=G
V
)と同型になる. ところが, GV のような代数群に関する商空間は,一 般には位相空間としてHausdorにならない. この困難を解消するためには幾何学的 不変式論が必要で,のちに述べる箙多様体もその様に定義される. その様な技術的な 相違点があるにせよ, preprojectivealgebraの幾何学的な意味は`余接束'である.
ALE空間(より正確にはその下部構造の複素曲面)や箙多様体が(A.1.2)もしくは そのmo dicationとして定義される. その意味で,(A.1.2)は有益なシンプレクティッ ク多様体を構成するレシピと思うことが出来る.
A.2. 箙多様体(quivervariety). 箙多様体の定義をする. まず(A.1.1)のE (V)
とその双対空間の直和を考える:
E
(V)8E
(V)
3
= M
h2 Hom(V
out(h)
;V
in(h)
)8Hom(V
in(h)
;V
out(h) )
これは,Eのダブル,すなわちEの辺に逆向きの辺も付け加えてできたquiver(これ をQ#で表わす)に対応するベクトル空間と思える. EのダブルをHで表わす. さら に, 各頂点ごとにベクトル空間Wkを置いて,
M
h2H Hom(V
out(h)
;V
in(h) )
!
8 M
k2I
Hom(W
k
;V
k
)8Hom(V
k
;W
k )
!
(A.2.1)
というベクトル空間を考える. このベクトル空間をM(V;W)と書こう. M(V;W)に は自然にG
V
= Q
k2I GL(V
k
)が作用する.
さて,(A.2.1)の点の意味するところは何であろうか? 第一の成分は,quiverHの 表現であるから意味は明らかであろう. 一方,第二成分は[17]で現れたものの,quiver の表現論で扱われたことはなかったようである. 前節のように表現論との関連から,第 二成分の最高ウェイトベクトルとしての意味がはっきりしてくるのではあるが,ここ で[17]での意味を説明しよう. 箙多様体をALE空間上の正則ベクトル束のモジュラ イ空間として見るとき,(W
k )
k 2I
は,framing, すなわち無限遠でのベクトル束の自明
7この論説では,シンプレクティック形式はすべて複素数体上で考える.他の文献では,実のシンプレ クティック形式と区別するために正則シンプレクティック形式と呼ぶ場合も多いので注意.
20 中島 啓(HirakuNaka jima)
化に対応する. これをquiverの言葉に無理矢理に翻訳すると次のようになる. quiver
Q
#から一切の辺を取り除いたquiverをQ0とする. このとき,Q0の表現とは,頂点の 上にベクトル空間を置いたもの,そしてベクトル空間の間の線型写像は一切考えない ものに他ならない. Q#の表現(V
k )
k 2I , (B
h )
h2H
に対し,B
h
を忘れると, Q0の表現 が与えられる. これは,表現の間の関手である. (より一般に,Q#の部分グラフを考え れば,同様の関手が定義される.) この関手を制限と呼ぼう. さて,上の式で,(W
k )
k 2I
は, Q0の表現と考えることができる. すると
M
k 2I
Hom(W
k
;V
k )
は, (Wk )
k 2Iから(Vk )
k2i
;(B
h )
h2Iの制限へのquiverQ0での準同型の全体である.
M
k 2I Hom(V
k
;W
k )
はもちろん,(Vk )
k 2I
;(B
h )
h2Hの制限から(Wk )
k 2iへのquiverQ0での準同型の全体 である.
前に述べたような,quiverの表現とベクトル束の類似のもとでは,次の対応がある ことは納得が行くであろう.
(1) 上の関手は,ベクトル束を部分多様体へ制限することに対応する.
(2) (A.2.1)の第二成分は,部分多様体に与えられたベクトル束との間の準同型に
対応する.
さて,(A.2.1)の各成分をBh ,i
k ,j
k と書こう. このとき, 次の写像を考える.
:M! M
k 2I End(V
k ); (B
h
;i
k
;j
k )
h2H;k 2I 70!
0
@ X
h2H:in(h)=k
"(h)B
h B
h +i
k j
k 1
A
k2I :
M(V;W)は,その`半分'
M
h2 Hom(V
out(h)
;V
in(h) )8
M
k2I
Hom(W
k
;V
k ) (A.2.2)
の余接束と見ることができ, 特に自然なシンプレクティック形式を持つ. 上のベクト ル空間をM
(V;W)と書こう. GV の作用は,M(V;W)のシンプレクティック形式を 保つ. 上の写像は,運動量写像である. このあたりは,preprojectivealgebraの説明 をしたときと同様である.
さて,シンプレクティック商
01
(0)=G
V
を考えたい. もしも商空間M
(V;W)=G
V が多様体であれば, これはその余接束と 同型である. しかし,集合論的な商空間は一般には代数多様体になることが期待でき ない. 幾何学的不変式論[18]の教えは,商空間を考える代わりに商空間上の関数が何 かを考えよ,と言うことである. 今の場合は, 二つの候補がある. まず第一のものは,
01
(0)をアファイン代数多様体と思って, 座標環のG
V
不変な部分環に対応するア ファイン代数多様体を考えるものである. これは01(0)==G
V
と書かれるもので,そ の点は01(0)の閉軌道に他ならない.
もう一つは,01(0)上の直線束(今の場合,自明なものを取る)に,G
V
の作用を持 ち上げ,直線束のテンソル積のGV 不変な切断のなす次数つき環に対応する準射影多 様体を考えるものである. 自明な直線束への作用の持ち上げを, GL(Vk
)の行列式の 積と取ると,以下のように定義される`安定(stable)'な点の全体の商空間になる.