HAL Id: hal-01818302
https://hal.archives-ouvertes.fr/hal-01818302
Preprint submitted on 18 Jun 2018
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires
Symmetric pairs and branching laws
Paul-Emile Paradan
To cite this version:
Paul-Emile Paradan. Symmetric pairs and branching laws. 2018. �hal-01818302�
Symmetric pairs and branching laws
Paul-Emile PARADAN∗ June 18, 2018
Abstract
LetGbe a compact connected Lie group and letH be a subgroup fixed by an involution. A classical result assures that theHC-action on the flag varietyFofGadmits a finite number of orbits. In this article we propose a formula for the branching coefficients of the symmetric pairpG, Hqthat is parametrized byHCzF.
Contents
1 Introduction 2
2 Non abelian localization 5
2.1 Matsuki duality . . . . 6
2.2 Borel-Weil-Bott theorem . . . . 7
2.3 Localization of the Riemann-Roch character . . . . 8
2.4 Local model nearHxĂZθ . . . . 9
3 Proof of the main theorem 12 3.1 Computation ofQHxpλq . . . . 12
3.1.1 Step 1: holomorphic induction . . . . 12
3.1.2 Step 2: cotangent induction . . . . 13
3.1.3 Step 3: linear case . . . . 13
3.1.4 Conclusion . . . . 14
3.2 Another expression forQHxpλq . . . . 15
3.3 Computation of the virtual moduleMxpλq . . . . 17
∗Univ. Montpellier, CNRS, Montpellier, France,paul-emile.paradan@umontpellier.fr
4 Examples 21
4.1 KĂKˆK . . . . 22
4.2 Uppq ˆUpqq ĂUpp`qq . . . . 23
4.2.1 The critical set . . . . 23
4.2.2 Localized indices . . . . 24
4.2.3 The extreme cases : j“0 orj“q . . . . 28
4.2.4 Upn´1q ˆUp1q ĂUpnq . . . . 28
1 Introduction
Let G be a compact connected Lie group equipped with an involution θ.
Let Gθ :“ tg P G, θpgq “ gu be the subgroup fixed by the involution. We consider a subgroup H Ă G such that pGθq0 Ă H Ă Gθ. The purpose of this paper is the study of the branching laws betweenG andH.
Let T be a maximal torus of G that we choose θ-invariant. Let t be the Lie algebra of T. Let Λ Ă t˚ be the lattice of weights, and let t˚` be a Weyl chamber. The irreducible representations ofG are parametrized by the semi-group Λ`G:“ΛXt˚` of dominant weights.
Let λ P Λ`G. In order to study the restriction VλG|H of the irreducible G-representationVλG, we consider theH-action on the flag varietyF “G{T ofG. An important object is theH-invariant subset
Zθ ĂF
formed of the elements xPF for which the stabilizer subgroup Gx :“ tgP G, gx“xuis stable underθ. In orther words,gT PZθif and only ifg´1θpgq belongs to the normalizer subgroupNpTq. A well-known result tells us that the group H has finitely many orbits in Zθ, and that the finite set HzZθ
parametrizes theHC-orbits in F [7, 13, 11, 8].
LetxPZθ. The stabilizer subgroupGx is a maximal torus inGwith Lie algebragx. We will also consider the abelian subgroupHx:“GxXH (that is not necessarily connected). Any weightµPΛ determines a characterCµ
x
of the torus Gx by takingµx“g¨µif x“gT PF.
We denote byRxĂg˚xthe set of roots relative to the action of the Cartan subalgebragxon gbC. The mapµPRÞѵxPRx is an isomorphism, and we takeR`x ĂRx as the image ofR`ĂRthrough this isomorphism.
The involutionθleaves the setRx invariant, andαPRx is animaginary root if θpαq “ α. If α is imaginary, the subspace pg bCqα is θ-stable.
There are two cases. If the action of θ on pgbCqα is trivial then α is compact imaginary. If the action of´θonpgbCqα is trivial, thenαisnon- compact imaginary. We denote respectively by Rcix and byRncix the subsets
of compact imaginary and non-compact imaginary roots, and we introduce the followingGx-modules
Ecix :“ ÿ
αPRci
xXR`x
pgbCqα, Encix :“ ÿ
αPRnci
x XR`x
pgbCqα.
The weight
δpxq:“ 1 2
ÿ
αPR` xXθpR`
xq
θpαq‰α
α
defines a character Cδpxq of the abelian groupHx. Let mx“ 1
2|R`x XθpR`xq X tθpαq ‰αu| `dimEncix .
We denote byRpHqand by RpHxqthe representations rings of the com- pact Lie groups H and Hx. An element E P RpHq can be represented as a finite sum E “ ř
VPHpmVV, with mV P Z. We denote by RppHq (resp.
RpHp xq) the space of Z-valued functions on Hp (resp. Hxx). An element EPRppHq can be represented as aninfinitesumř
VPHpmVV, withmV PZ. The induction map IndKH : RpHp xq Ñ RpHqp is the dual of the restriction morphismRpHq ÑRpHxq.
The main result of this paper is the following theorem.
Theorem 1.1 Let λPΛ`G. We have the decomposition
(1.1) VλG|H “ ÿ
HxPHzZθ
QHxpλq
where the terms QHxpλq PRppHq are defined by the following relation : QHxpλq “ p´1qmxIndHHx´
Cλ
x`δpxqbdetpEncix q bSympEncix q bľ Ecix
¯ . Here SympEncix q, which is the symmetric algebra of Encix , is an admissible representation ofHxandŹEcix “Ź`EcixaŹ´Ecix is a virtual representation of Hx.
We give now another formulation for decomposition (1.1) using the (right) action of the Weyl group W “ NpTq{T on the flag variety F. If x “gT P F and w PW we take xw :“ gwT. We notice immediately that Zθ is stable under the action of W.
We associate to an elementx“gT PZθthe subgroup WxH ĂW defined by the relation w P WxH ðñ Hxw “ Hx. We denote by HzZθ{W the quotient of Zθ by the action of HˆW, and by ¯x PHzZθ{W the image of xPZθ through the quotient map. We associate to ¯xPHzZθ{W the element Q¯xpλq PRpHqp defined as follows
Qx¯pλq “ ÿ
w¯PWxHzW
QHxwpλq.
Theorem 1.1 says then that VλG|H “ ř
¯xPHzZθ{WQx¯pλq. Here is a new formulation of Theorem 1.1.
Theorem 1.2 We have VλG|H “ ř
x¯PHzZθ{WQ¯xpλq where Qx¯pλq P RpHqp has the following description
Q¯xpλq “IndHHx´
Mxpλq bCδ
pxqbľ Ecix
¯ ,
for some1 Mxpλq PRpHp xq.
We finish this section by giving two basic examples associated to the group SUp2q. Here the flag variety of SUp2q is the 2-dimensional sphere S2. For n ě0, we denote by Vn the irreducible representation of SUp2q of dimensionn`1.
Example 1. G“SUp2q and the involution θis the conjugaison by the matrix
ˆ 1 0 0 ´1
˙
. The subgroup fixed byθ is the torusT »Up1qand the critical set Zθ Ă S2 is composed by the poles S, N and the equator E, so thatTzZθ has three terms. We takeλ“ninSU{p2q »N.
ForHx“E, we haveEncix “Ecix “ t0u,Hx»Z2, and Cλ
x`δpxq“Cn|Z2. The contribution ofE is then IndUZ2p1qpCn|Z2q “Cnbř
kPZC2k. ForHx“N, we haveHx “T,Encix “C2,Ecix “ t0u, andCλ
x`δpxq “Cn. The contribution ofN is then´Cn`2bSympC2q.
For Hx “S, we have Hx “T, Encix “ C´2, Ecix “ t0u, and Cλ
x`δpxq “ C´n. The contribution ofS is then´C´n´2bSympC´2q.
Finally, Relations (1.1) become Vn|T “ Cnbÿ
kPZ
C2k´C´n´2bSympC´2q ´Cn`2bSympC2q
“ ÿ0 k“´n
C2k
`n.
1The precise expression ofMxpλqis given in Proposition 3.8.
Example 2. G “ SUp2q ˆ SUp2q and the involution θ is the map pa, bq ÞÑ pb, aq. The subgroup fixed byθ isSUp2q embedded diagonally and the critical setZθ ĂS2ˆS2 is equal to the union of the orbitsSUp2q¨pN, Nq and SUp2q ¨ pS, Nq. Letλ“ pn, mq PG.p
For x “ pN, Nq or x “ pS, Nq we have Encix “ Ecix “ t0u and Hx »T. Forx“ pN, Nq we have λx`δpxq “m`n`2, and forx“ pS, Nqwe have λx`δpxq “m´n. Relations 1.1 give then
VnbVm“IndSUT p2qpCm´nq ´IndSUT p2qpCm`n`2q.
It is not difficult to see that the previous identities correspond to the classical Clebsch-Gordan relations (see Example 4.2).
Notations Throughout the paper :
• Gdenotes a compact connected Lie group with Lie algebra g.
• T is a maximal torus in Gwith Lie algebrat.
• ΛĂt˚ is the weight lattice ofT : everyµPΛ defines a 1-dimensional T-representation, denoted by Cµ, where t “ exppXq acts by tµ :“
eixµ,Xy.
• The coadjoint action ofgPGon ξPg˚ is denoted by g¨ξ.
• When a Lie groupK acts on set X, the stabilizer subgroup of xPX is denoted by Kx :“ tkPK |k¨x “xu and the Lie algebra of Kx is denoted bykx.
• When a Lie group K acts on a manifold M, we denote by X¨m :“
d
dtetX¨m|t“0,mPM, the vector field generated byXPk.
2 Non abelian localization
Our main result is obtained by means of a non-abelian localization of the Riemann-Roch character on the flag variety F of G. For that we will use the familypΩrqr of symplectic structure parametrized by the interior of the Weyl chambert˚`. The symplectic structure Ωr comes from the identication gT Ñg¨r ofF with the coadjoint orbit Gr. The moment map Φr :F Ñg˚ associated to the action ofG on pF,Ωrq is the map gT ÞÑg¨r.
At the level of Lie algebras we haveg“h‘qwhereh“gθ andq“g´θ. For anyξ Pg“h‘q, we denote by ξ`his h-part and byξ´ hisq-part. We use a G-invariant scalar product p´,´q on g such that the involution θ is an orthogonal map. It induces identificationsg˚ »g,h˚»hand q˚ »q.
The moment map ΦHr :F Ñh˚ associated to the action ofH onpF,Ωrq is the mapgT ÞÑ pg¨rq`.
2.1 Matsuki duality
Consider the complex reductive groupsGC and HC associated to the com- pact Lie groupsGandH. LetLĂGC be the real form such thatHĂL is a maximal compact subgroup ofL.
Matsuki duality is the statement that a one-to-one correspondence exists between theHC-orbits and theL-orbits inF; two orbits are in duality when their intersection is a single orbit ofH.
Uzawa, and Mirkovic-Uzawa-Vilonen [14, 8] proved the Matsuki corre- spondence by showing that bothHC-orbits andL-orbits inFare parametrized by theH-orbits in the set of critical points of the function}ΦHr }2 :F ÑR.
First we recall the elementary but fundamental fact that the subsetZθ
is equal to the set of critical points of the function}ΦHr }2 [8, 3].
Lemma 2.1 Letx“gT PF andr PInteriorpt˚`q. The following statements are equivalent:
i) the subalgebragx is invariant under θ (i.e. xPZθ), ii) g´1θpgq PNpTq,
iii) x is a critical point of the function }ΦHr }2, iv) pg¨rq` and pg¨rq´ commutes.
Proof. Letng “g´1θpgq and letr be a regular element oft˚ »t. Since gx“Adpgqt we see that
θpgxq “gx ðñ ng PNGpTq ðñ rng¨θprq, rs “0 ðñ rθpg¨rq, g¨rs “0 ðñ rpg¨rq`,pg¨rq´s “0.
A small computation shows that for any X P g the derivative of the function t ÞÑ }ΦHr petXxq}2 at t “ 0 is equal to pX,rg¨r, θpg¨rqsq. Hence
x“gT is a critical point of the function}ΦHr }2if and only ifrg¨r, θpg¨rqs “0.
Finally we have proved that the statementsiq, iiq, iiiqandivqare equivalent.
l
Let us check the other easy fact.
Lemma 2.2 The set HzZθ is finite.
Proof. Let x “gT P Zθ. A neighborhood of x is defined by elements of the formeXeYx where XPhand Y Pq. Now we see thateXeYgT PZθ if and only ife´2g´1Y PNpTq. IfY is sufficiently small the former relation is equivalent to g´1Y P t, and in this case eXeYx “eXx. We have proved that any element inHzZθ is isolated. AsHzZθ is compact, we can conclude thatHzZθ is finite. l
2.2 Borel-Weil-Bott theorem
We first recall the Borel-Weil-Bott theorem. The flag manifoldFis equipped with theG-invariant complex structure such that
TeTF » ÿ
αPR`
pgbCqα
is an identity of T-modules. Let us consider the tangent bundle TF as a complex vector bundle on F with the invariant Hermitian structure hF
induced by the invariant scalar product ong.
Any weight λPΛ defines a line bundleLλ »GˆT Cλ on F. Definition 2.3 We associated to a weight λPΛ
‚ the spin-c bundle on F
Sλ:“ľ
C
TFbLλ,
‚ the Riemann-Roch character RRGpF,Lλq P RpGq which is the equiv- ariant index of the Dirac operator Dλ associated to the spin-c structure Sλ. The Borel-Weil-Bott theorem asserts that VλG“RRGpF,Lλq when λis dominant. Now we consider the restriction VλG|H “ RRHpF,Lλq. In the next section we will explain how we can localize theH-equivariant Riemann- Roch character RRHpF,Lλq on the critical set of the function }ΦHr }2 [9].
2.3 Localization of the Riemann-Roch character
In this section we explain how we perform the “Witten non-abelian local- ization” of the Riemann-Roch character with the help of the moment map ΦHr :F Ñh˚attached to a regular elementrof the Weyl chamber [9, 5, 10].
Let us denote byX ÞÑ rXsg{tthe projectiongÑg{t. The Kirwan vector fieldκr on F is defined as follows:
κrpxq “ ´ΦHr pxq ¨xPTxF.
Through the identificationg{t»TxF, X ÞÑ dtd|t“0getXT, the vectorκrpxq P TxF is equal to rg´1θpgq ¨rsg{t. Hence the set Zθ Ă F is exactly the set whereκr vanishes.
Lat D0 be the Dirac operator associated to the spin-c structure S0 “ Ź
CTF. The principal symbol of the elliptic operatorD0 is the bundle map σpFq PΓpT˚F,hompŹ`
CTF,Ź´
CTFqqdefined by the Clifford action σpFqpx, νq “cxp˜νq:ľ`
CTxF Ñľ´
CTxF.
whereνPT˚xO»ν˜PTxO is the one to one map associated to the identifi- cation g˚»g(see [2]).
Now we will deform the elliptic symbolσpFqby means of the vector field κr [9, 10].
Definition 2.4 The symbolσpFq shifted by the vector fieldκr is the symbol on F defined by
σrpFqpx, νq “cxp˜ν´κrpxqq for anypx, νq PT˚F.
Consider anH-invariant open subsetU ĂFsuch thatUXZθis compact inF. Then the restrictionσrpFq|U is aH-transversally elliptic symbol onU, and so its equivariant index is a well defined element inRppHq(see [1, 9, 10]).
Thus we can define the following localized equivariant indices.
Definition 2.5 Let HxĂZθ. We denote by
RRHpF,Lλ,ΦHr , Hxq P RpHqp
the equivariant index ofσrpFqbLλ|U whereU is an invariant neighbourhood of Hx so that U XZθ“Hx.
We proved in [9] that we have the decomposition RRHpF,Lλq “ ÿ
HxPHzZθ
RRHpF,Lλ,ΦHr , Hxq P RpHq.p
The computation of the characters RRHpF,Lλ,ΦHr , Hxq will be handle in Section 3.1. To undertake these calculations we need to describe geomet- rically a neighborhood ofHx inF. This is the goal of the next section.
2.4 Local model near HxĂZθ
Letx“gT PZθ. We need to compute a symplectic model of a neighborhood ofHx inpF,Ωrq. Here we use the identification g»g˚ given by the choice of an invariant scalar product. Let µ “ g¨r that we write µ “ µ``µ´ whereµ`Phand µ´Pq.
The tangent spaceTxF is equipped with the symplectic two form Ωr|x: Ωr|xpX¨x, Y ¨xq “ pµ,rX, Ysq, X, Y Pg.
We need to understand the structure of the symplectic vector space pTxF,Ωr|xq. IfaĂgis a vector subspace we denote bya¨x:“ tX¨x, X Pau the corresponding subspace of TxF. The symplectic orthogonal of a¨x is denoted bypa¨xqK,Ω.
Ifa,b are two subspaces, a small computation gives that (2.2) pa¨xqK,ΩXb¨x»aKX rb, µs,
whereaK Ăg is the orthogonal ofa relatively to the scalar product.
We denote by gµ` “ hµ` ‘qµ` the subspaces fixed by adpµ`q. Notice thatgµ“gx is an abelian subalgebra containing µ` sincerµ`, µ´s “ 0. It follows thatgx Ăgµ`.
Lemma 2.6 gµ` ¨x and rh, µ`s ¨x are symplectic subspaces of TxF. Proof. It is a direct consequence of (2.2). l
We consider now the symplectic subspace Vx Ă TxF defined by the relation
(2.3) Vx“ prh, µ`s ¨xqK,ΩX rg, µ`s ¨x.
A small computation shows thatX¨xPVx if and only if rX, µs Ă rq, µ`s.
We have the following important Lemma.
Lemma 2.7 • We have the following decomposition (2.4) TxF “gµ`¨x‘K “
h, µ`‰
¨x‘K Vx where K stands for the orthogonal relative to Ωr|x.
• gµ`¨x is symplectomorphic to hµ`{hx‘ phµ`{hxq˚.
• rh, µ`s ¨x is symplectomorphic to h{hµ` equipped with the symplectic structure Ωµ`p¯u,vq “ pµ¯ `,ru, vsq.
• Vx is symplectomorphic toph¨xqK,Ω{ph¨xqK,ΩXh¨x.
Proof. If we use the decomposition g“gµ`‘ rg, µ`sand the fact that the abelian subalgebragx is contained in gµ` we obtain
TxF “gµ`¨x‘ rg, µ`s ¨x.
It is obvious to check that the subspacesrg, µ`s¨xandgµ`¨xare orthogonal relatively to the symplectic form Ωr|x. Since rh, µ`s ¨x is a symplectic subspace we haverg, µ`s ¨x“ rh, µ`s ¨x‘K Vx whereVx is defined by (2.3).
The first point is proved.
The identitiesgx“θpgxq “gθpxqimply the decompositionsgx“hx‘qx andrgµ`, xs “ rqµ`, xs‘rhµ`, xs. The vector subspacerhµ`, xsis isomorphic to hµ`{hx, and the map v ÞÑ Ωr|xpv,´q defines an isomorphism between rqµ`, xsand the dual of rhµ`, xs. The second point is proved.
For the third point we use the isomophism j :rh, µ`s Ñh{hµ` induces by the projectionhÑh{hµ`. Then the map ¯uÞÑjpuq ¨¯ x defines a symplec- tomorphism betweenph{hµ`,Ωµ`q and rh, µ`s ¨x.
Now we see that (2.4) together with the decomposition h ¨ x “ rh, µ`s ¨x`h¨x leads to
ph¨xqK,Ω “ prh, µ`s ¨xqK,ΩX phµ` ¨xqK,Ω
“ prh, µ`s ¨xqK,ΩX pgµ` ¨xq ‘Vx
“ ph¨xqK,ΩXh¨x‘Vx. The last point follows. l
We denote by ΩVx the restriction of Ωr|x on the symplectic vector sub- spaceVx. The action ofHx onpVx,ΩVxq is Hamiltonian, with moment map ΦVx :VxÑh˚x defined by the relation
xΦVxpvq, Ay “ 1
2ΩVxpv, Avq, v PVx, APhx.
Thanks to Lemma 2.7, we know that the Hx-symplectic vector space pTxF,Ωr|xq admits the following decomposition
TxF »hµ`{hx‘ phµ`{hxq˚ K‘h{hµ`
‘K Vx
Thanks to the normal form Theorem of Marle [6] and Guillemin-Sternberg [4], we get the following result.
Corollary 2.8 An H-equivariant symplectic model of a neighborhoood of Hx in F is Fx :“HˆHµ` Yx where
Yx“Hµ`ˆHx
`phµ`{hxq˚ˆVx˘ .
The corresponding moment map on Fx is
ΦFxprh;η, vsq “hpη`µ``ΦVxpvqq for rh;η, vs PHˆHx
`phµ`{hxq˚ˆVx˘ .
We finish this section by computing a compatible complex structure on Vx.
By definition, the map that sendsX¨xtorX, µsdefines an isomorphism i: Vx Ñ rq, µ`s. The adjoint map adpµq defines also an automorphism of rg, µ`s: for any X P rg, µ`s we denote by ˜X P rg, µ`s the unique element such thatadpµqX˜ “X.
The symplectic structure Ωµ:“ pi´1q˚ΩVx satisfies the relations ΩµpX, Yq “ pµ,rX,˜ Y˜sq “ pX,Y˜q “ ´pX, Y˜ q, @X, Y P rq, µ`s.
We consider the one to one map ´adpµqadpθpµqq : rg, µ`s Ñ rg, µ`s and the Hx-invariant complex structure Jµ` “ adpµ`qp´adpµ`q2q´1{2 on rg, µ`s. It restricts to a one to one map Tx : rq, µ`s Ñ rq, µ`s and a complex structure on rq, µ`s(still denoted by Jµ`).
Let Sx :“ pTx2q´1{2Tx. The mapJVx :“Jµ` ˝Sx defines a Hx-invariant complex structure on rq, µ`s.
Lemma 2.9 The Hx-symplectic space pVx,ΩVxq is isomorphic to rq, µ`s equipped with the symplectic formΩ1µpv, wq “ pJVxv, wq.
Proof. We know already that pVx,ΩVxq » prq, µ`s,Ωµq. If one takes L“Tx˝ p´adpµ`q2q´1{4˝ pTx2q´1{4, we check easily that ΩµpLpvq, Lpwqq “ pJVxv, wq. l
3 Proof of the main theorem
We start with the following lemma.
Lemma 3.1 The quantity RRHpF,Lλ,ΦHr , Hxq does not depend on the choice of the regular element r in the Weyl chamber. In the following we will denote it by QHxpλq PRppHq.
Proof. Let r0, r1 be two regular elements of the Weyl chamber. For tP r0,1s, we consider the regular elementrptq “tr1` p1´tqr0: the Kirwan vector fieldκrptq vanishes exactly onZθ for anytP r0,1s. IfU is an invariant neighbourhood of Hx so that U XZθ “ Hx, then t P r0,1s ÞÑ σrptqpFq b Lλ|U defines an homotopy of transversally elliptic symbols. Accordingly, the equivariant index ofσr0pFq bLλ|U and σr1pFq bLλ|U are equal. l
3.1 Computation of QHxpλq
The computation ofQHxpλq is done in three steps.
3.1.1 Step 1: holomorphic induction
Let Hµ` Ă H be the stabilizer subgroup of µ` :“ ΦHr pxq. We know that an H-equivariant symplectic model of a neighborhoood of Hx in F is the manifoldHˆHµ` Yx where
Yx“Hµ`ˆHx
`phµ`{hxq˚ˆVx˘ .
The symplectic two form onYx is built from the canonical symplectic struc- ture onHµ`ˆHxphµ`{hxq˚ »T˚pHµ`{Hxqand the symplectic structure on Vx. The moment map relative to the action ofHµ` on Yx is
ΦYxprh;η, vsq “hpη`µ``ΦVxpvqq Ph˚µ`, forrh;η, vs PHµ` ˆHx
`phµ`{hxq˚ˆVx˘ .
Let κYx the Kirwan vector field on Yx. It is immediate to check that rh;η, vs P tκYx “0uif and only ifη“0 andpµ``ΦVxpvqq ¨v“0. The map vÞѵ`¨v is bijective andvÞÑΦVxpvq ¨v is homogeneous of degree equal to 3. Then there existsǫą0 such that
pµ``ΦVxpvqq ¨v “0 and }v} ďǫ ùñ v“0.
InYx, we still denote byxthe pointre,0,0s. We equipYxwith an invariant almost complex structure that is compatible with the symplectic structure,
and we denote by RRHµ`pYx,Lλ|Yx,ΦYx, Hµ`xq the Riemann-Roch charac- ter onYx localized on the componentHµ`xĂ tκYx “0u.
The quotienth{hµ`, which is equipped with the invariant complex struc- tureJµ` :“adpµ`qp´adpµ`q2q´1{2, is a complex Hµ`-module.
In [9][Theorem 7.5], we proved that QHxpλq “ RRHpF,Lλ,ΦHr , Hxq is equal to
(3.5) IndHH
µ`
´RRH
µ`pYx,Lλ|Yx,ΦYx, Hµ`xq bľ h{hµ`
¯ .
3.1.2 Step 2: cotangent induction
The map Φxpvq:“µ``ΦVxpvqis a moment map for the Hamiltonian action ofHx on Vx. The moment map on theHµ`-manifold
Yx“Hµ`ˆHx
`phµ`{hxq˚ˆVx˘ is ΦYxprh;η, vsq “hpη`Φxpvqq Ph˚µ.
Let κVxpvq “ ´Φxpvq ¨ v be the Kirwan vector field on Vx. We are interested in the connected component t0u of tκVx “ 0u. We choose a compatible almost complex structure on the symplectic vector space and we denote by RRHxpVx,Φx,t0uq PRppHxqthe Riemann-Roch character localized ont0u Ă tκVx “0u.
In Section 3.3 of [10] we have proved that (3.6)
RRHµ`pYx,Lλ|Yx,ΦYx, Hµ`xq “IndHHµ`
x pRRHxpVx,Φx,t0uq bLλ|xq. 3.1.3 Step 3: linear case
We write q{qµ` for the vector space rq, µ`s equipped with the complex structureJµ`. Soq{qµ` is aHµ`-module and we denote by Sympq{qµ`qthe corresponding symmetric algebra.
We need to compare the virtual Hx-modules Ź
JVxVx and Ź
´Jµ` Vx. The weight
δpxq:“ 1 2
ÿ
αPR` xXθpR`
xq
θpαq‰α
α
defines a character Cδ
pxq of the abelian group Hx. Recall that mx P N corresponds to the quantity 12|R`x XθpR`xq X tθpαq ‰αu| `dimEncix .
The following lemma will be proved in Section 3.2.
Lemma 3.2 The following identity holds : ľ
JVx
Vx» p´1qmxCδ
pxqbdetpEncix q b ľ
´Jµ`
Vx.
On the vector space Vx, we can work with two localized Riemann-Roch characters:
• RRHxpVx,Φx,t0uqis defined with the complex structure JVx,
• ĄRRHxpVx,Φx,t0uqis defined with the complex structure ´Jµ`. The previous Lemma gives that RRHxpVx,Φx,t0uqis equal top´1qmxCδ
pxqb ĄRRHxpVx,Φx,t0uq.
Proposition 3.3 We have
(3.7) RRHxpVx,Φx,t0uq “ p´1qmxCδ
pxqbdetpEncix q bSympq{qµ`q.
Proof. ForsP r0,1s, we consider the Hx-equivariant map Φs:VxÑh˚x defined by the relations Φspvq “µ``sΦVxpvq. The corresponding Kirwan vector field onVx is κspvq “ ´Φspvq ¨v. It is not difficult to see that there existsǫą0 such thattκs“0u X t}v} ďǫu “ t0ufor any sP r0,1s. Then a simple deformation argument gives thatRRĄHxpVx,Φs,t0uqdoes not depend onsP r0,1s. We have proved that
ĄRRHxpVx,Φs,t0uq “ĄRRHxpVx, µ`,t0uq
where µ` denotes the constant map Φ0. Standard computations gives ĄRRHxpVx, µ`,t0uq “ Sympq{qµ`q (see [9][Proposition 5.4]). Our proof is completed. l
3.1.4 Conclusion
If we use the formulas (3.5), (3.6) and (3.7) we obtain the following expres- sion
QHxpλq “ p´1qmxIndHHx
˜ Cλ
x`δpxqbdetpEncix q bSympq{qµ`q bľ
C
h{hµ`
¸
inRpHq. Herep Cλ
x is the character ofGx associated to the weightλx “gλ.
The previous formula depends on a choice of a regular element r in the Weyl chamber. In the next section we will propose another expression for QHxpλq that does not depend on this choice.
3.2 Another expression for QHxpλq
Let Rx Ă g˚x be the roots for the action of the torus Gx on gbC. The involution θ : t˚ Ñ t˚ leaves the set Rx invariant and a root α P Rx is called imaginary if θpαq “ α. We denote respectively by Rcix and by Rncix the subsets ofcompact imaginary andnon-compact imaginary roots.
We choose a generic element r Pt˚` such thatµ`“ pg¨rq` satisfies the following relation : for anyαPRx, we have
pα, µ`q “0ðñθpαq “ ´α.
Notice that an imaginary roots α is positive if and only if pα, µ`q ą0.
Definition 3.4 We consider the subset Ax ĂRx
αPAx ðñ αpµ`q ą0, θpαq ‰α
The involutionθ defines a free action ofZ2 on the setAx. We denote by Ax{Z2 its quotient. For anyα PRx, we denote by Cα the corresponding 1- dimensional representation ofGx, andCα|Hx its restriction to the subgroup Hx. We have a natural maprαs PAx{Z2 ÞÝÑCα|Hx PRpHxq.
For anyαPRx we define
˜ α“ ˘α where˘is the sign ofαpµqαpθpµqq.
We consider theHx-modulesh{hµ` :“ prh, µ`s, Jµ`q,q{qµ` :“ pVx, Jµ`q andpVx, JVxq.
Lemma 3.5 We have the following isomorphisms of Hx-modules h{hµ` » à
rαsPAx{Z2
Cα|Hx‘ à
αPRci
xXR`x
Cα|Hx rAs, q{qµ` » à
rαsPAx{Z2
Cα|Hx‘ à
αPRnci
x XR`x
Cα|Hx rBs, pVx, JVxq » à
rαsPAx{Z2
Cα˜|Hx‘ à
αPRnci
x XR`x
Cα˜|Hx rCs. Proof. Thanks to Lemma 2.9, we know that theHx-modulepVx, JVxqis isomorphic to the vector spacerq, µ`sequipped with the complex structure JVx :“Jµ`˝Sx. We consider the vector spacesrq, µ`sandrg, µ`sequipped with the complex structure Jµ`. The projection (taking the real part) r : gbCÑg induces an isomorphism ofGx-modules
r: à
αpµ`qą0
pgbCqα ÝÑ rg, µ`s.
The orthogonal projections p1 : rg, µ`s Ñ rq, µ`s and p2 : rg, µ`s Ñ rh, µ`scommutes with theHx-action, so the maps
p1˝r: à
αpµ`qą0
pgbCqα ÝÑ rq, µ`s, p2˝r: à
αpµ`qą0
pgbCqα ÝÑ rh, µ`s
are surjective morphisms ofHx-modules.
Let Vx1pαq “ p1 ˝rppgbCqαq. We notice that dimCVx1pαq P t0,1u:
Vx1pαq “ t0u only ifα is a non-compact imaginary root andVx1pαq »Cα|Hx
whenVx1pαq ‰ t0u. We notice also that Vx1pαq “Vx1pθpαqq, hence q{qµ` “ prq, µ`s, Jµ`q » à
rαsPAx{Z2
Vx1pαq ‘ à
αPRnci
x XR`x
Vx1pαq.
The identity rBsis proved.
Similarly we considerVx2pαq “p2˝rppgbCqαq. We notice that dimCVx2pαq P t0,1u: Vx2pαq “ t0u only if α is a compact imaginary root and Vx2pαq » Cα|Hx whenVx2pαq ‰ t0u. We notice also that Vx2pαq “Vx2pθpαqq, hence
h{hµ` “ prh, µ`s, Jµ`q » à
rαsPAx{Z2
Vx2pαq ‘ à
αPRci
xXR`x
Vx2pαq.
The identity rAsis proved.
Finally we check that the complex structuresJµ` and JVx preserve each Vx1pαq and that pVxpαq, JVxq »Cα˜|Hx when pα, µ`q ą 0. The identity rCs follows. l
We consider theHx-moduleVx:“ř
rαsPAx{Z2Cα|Hx, and theGx-modules Enci
x :“ ř
αPRnci
x XR`x
Cα and Eci
x :“ ř
αPRci
xXR`x
Cα. In the previous lemma we have proved thatHx-modulesh{hµ` and q{qµ` are respectively isomor- phic toVx‘Ecix andVx‘Encix . If we use the fact that SympVxq bŹVx “1, we get the following corollary.
Corollary 3.6 We have the following identity of virtual Hx-modules:
Sympq{qµ`q bľ
h{hµ` »SympEncix q bľ Ecix.
Let us now prove Lemma 3.2.
Let B :“ Ax{Z2Ť
pRncix XR`xq. We see that Ź
JVxVx “ ś
αPBp1 ´ tα˜q whereas Ź
´Jµ`Vx “ ś
αPBp1´t´αq. Accordingly we get Ź
JVxVx »
p´1q|B1|Cη bŹ
´Jµ` Vx where B1 “ tα PB,α˜ “αu and η “ř
αPB1α. Now it is easy to check that an element αPB belongs to B1 if and only if α and θpαq both belong toR`x. In other words
B1“ αPR`x XθpR`xq, θpαq ‰α( {Z2ď
Rncix XR`x. We have proved that
ľ
JVx
Vx» p´1qmxCδ
pxqbdetpEncix q b ľ
´Jµ`
Vx.
l
Finally, thanks to Lemma 3.2 and Corollary 3.6, we obtain the final formula forQHxpλq (that does not depend on the choice ofr):
QHxpλq “ p´1qmxIndHHx´ Cλ
x`δpxqbdetpEncix q bSympEncix q bľ Ecix
¯ .
3.3 Computation of the virtual module Mxpλq
According to Theorem 1.1, we have the decomposition VλG|H “ ř
x¯Q¯xpλq where Q¯xpλq “ IndHHxpAxpλqq, and Axpλq P RppHxq has the following de- scription
Axpλq “ 1
|WxH| ÿ
wPW
p´1qmxwCλ
xw`δpxwqbdetpEncixwq bSympEncixwq bľ Ecixw.
The aim of this section is to simplify the expression of the virtual Hx- module Axpλq. We start by comparing the Gx-modules Encixw and Encix . We use the decompositionEncix “`
Encix ˘`
w‘` Encix ˘´
w where
`Encix ˘`
w :“ ÿ
αPRncix XR`xXR`xw
Cα, and ` Encix ˘´
w “ ÿ
αPRncix XR`xX´R`xw
Cα.
We have the following basic lemma (see Lemma 3.10).
Lemma 3.7 TheGx-module |Encix |w :“` Encix ˘`
w ‘ pEncix q´w is isomorphic to Encixw.
Letρ“ 12
ř
αPR`α. We denote byw‚λ“wpλ`ρq ´ρthe affine action of the Weyl group on the lattice Λ.
The main result of this section is the following proposition.