Arxiv:2011.02270V1 [Math.RT] 4 Nov 2020 Orsodn Otepartitions the to Corresponding N Hoe A

Home , Equidimensionality

arXiv:2011.02270v1 [math.RT] 4 Nov 2020 orsodn otepartitions the to corresponding n hoe A. Theorem r the following representation, induced an of theorem. subrepresentations main our of part second the as result fanloetorbit nilpotent a of GL scle the called is integer iteodplnmas hsvreyi lsl eae otegeometry the to related closely is variety This polynomials. Littlewood aaerzsalteprblcsubgroups parabolic the all parametrizes iptn risadcranitgrpriin.I atclr for particular, In partitions. integer certain and orbits nilpotent lmnsi h oe hc ossso l the all of consists which poset the in elements h olcino l uhcnuaeprblcsbrusfrsaclo a forms subgroups parabolic conjugate G/P such all of collection The radicals euto hspprdsrbsti oe ftenloetobt con orbits nilpotent the of poset this describes paper this of result L sgvnby given is riso h min ru.Floig[M3 hpe ] o n nilpo any for 7], Chapter [CM93, Following group. ambient the of orbits h iptn cone nilpotent the order 2 ⊂ osdrtenloetorbit nilpotent the consider , nSco ,w ilas ics h plcto fti emti result geometric paper: this this of of theorem application main the the state discuss we also Now will we 5, Secion In nte ereo reo o h set the for freedom of degree Another -aladdress E-mail h oinof notion The iptn ri fardciegroup reductive a of orbit nilpotent A ( n n oeifraino t oooyi noe nteLittlewood-Ric the in encoded is topology its of information some and , G 1 O hr sa nqenloetobtof orbit nilpotent unique an is there , , n C 1 Abstract. ntefis ato u anterm euetegoercSaka geometric the use we theorem, main our of part first the In GL ecie h emtyo h space not the do of coefficients geometry Littlewood-Richardson the some describes if only and if n niiao aite frpeettosof representations of varieties annihilator and N ) < orsod oanloetorbit nilpotent a to corresponds ALLTLWO OYOIL,SRNE IESADTHE AND FIBERS SPRINGER POLYNOMIALS, HALL-LITTLEWOOD × ⊂ ( n O nue orbit induced 1 GL NIIAO AITE FIDCDREPRESENTATIONS INDUCED OF VARIETIES ANNIHILATOR P 2 ) × o ahpi forbits of pair each For . For : endb h lsr order closure the by defined ( [email protected] GL n nue orbits induced ⊂ N For 2 ( G , O n C 2 G ) = ,tespace the ), of g ⊂ = GL l or ind G GL sthe as g ( n, l g ∗ meddbokdaoal,let block-diagonally, embedded ( n, ( epciey h e fnloetorbits nilpotent of set The respectively. , ,β α, β < α O C O rvdsacneto ewe h iptn riso ugopa subgroup a of orbits nilpotent the between connection a provides C G in ) ) G n aaoi subgroup parabolic a and ) α,β · n aaoi subgroup parabolic a and ( -saturation EZANISIUEO SCIENCE OF INSTITUTE WEIZMANN in O G fadol if only and if = + ∩ O GL tteohred ecndfiethe define can we end, other the At . O n G O ,where ), G P γ ( p O α · n hc sa motn pc osuyfrteWitkrsuppo Whittaker the for study to space important an is which , of ( = over O . α O O × 1 1. O G HOU ZHANG ZHUOHUI ) h onepr ftecoueorder closure the of counterpart The . α l 1 LN G + and and Introduction G G ⊂ hc intersects which ⊂ obt otie in contained -orbits O β C . n O · G stecoc fprblcsubgroups parabolic of choice the is ) O uhthat such sanloetobtof orbit nilpotent a is ftesubgroup the of sa ri fteajito h odon cinof action coadjoint the or adjoint the of orbit an is GL O X 2 i · =1 o lsia rus hr sabjcinbtenthe between bijection a is there groups, classical For . k 1 α ( in O ( of α n γ g i 2 hs w risaetemxmladminimal and maximal the are orbits two These . + g ) P ≤ esciey then: repsectively, , ewl lodsrb nagbacvreywhich variety algebraic an describe also will we , n aih h eodpr formi theorem main our of part second The vanish. α = X i ) P =1 eeuvlnet rv that prove to equivalence te k . eapriinof partition a be LN sl n[GS20]. in esult = O β GL i andin tained e uvreyo h ata a manifold flag partial the of subvariety sed LN L + ihatobokLv subgroup Levi two-block a with o all for ( G l n n where sauino iptn risof orbits nilpotent of union a is , and ) · naZrsioe ust hsorbit This subset. Zariski-open a in ihasadr eisubgroup Levi standard a with N ⊂ O ( of O k. ado offiinsadteHall- the and coefficients hardson etorbit tent O + G O γ SL Bala-Carter n α · ∩ .Tefis ato h main the of part first The ). otewvfotsto the of set wavefront the to , ( om oe ihapartial a with poset a forms ( O O n p n n ewl rvd this provide will we and , 1 β ,ec partition each ), + and r h iptn orbits nilpotent the are O n O O ). α ′ P faLv subgroup Levi a of ⊂ β ⊂ n h nilpotent the and nlso inc inclusion G O eapriinof partition a be · β ( O npartitions on L + rts n = g ) . α dthe nd fthe of G l g L ( O on = ) (1) A G-orbit Oγ is contained in G · (Oα,β + n) if and only if the Littlewood-Richardson coefficient γ cα,β 6=0. (2) The Littlewood-Richardson coefficient is the number of irreducible components of the subvariety γ Pα,β = {x ∈ Oγ ∩ p | pl(x) ∈ Oα × Oβ} ⊂ Oγ ∩ p

where the map pl is the projection pl : p 7→ p/n =∼ l onto the Levi subgroup. The first part of this theorem is a direct consequence of the theory of Hall polynomials which was discussed for abelian p-groups in [Mac98, Appendix Chapter II, Theorem AZ.1], and can be proven as a corollary of the Hall’s theorem in [Kle69, Theorem 2.2]. The proof of Hall’s theorem is combinatorial and makes use of Schubert calculus. However, in order to find an interpretion of the Littlewood-Richardson coefficients in our context, we would like to give a geometric proof to the first part of the theorem as a corollary of the geometric Satake isomorphism. The geometric constructions used in the proof of the first part will shed light on the proof of the second part of Theorem A in Section 4. It is worth mentioning that since we can induce the orbits by stages, the first part of Theorem A can be generalized to arbitrary standard parabolic subgroups:

Corollary B. For any partition n = (n1,...,nr) of n and partitions αi of the integers ni, we take a standard parabolic subgroup Pn = LnNn with a Levi subgroup isomorphic to GL(n1) × ... × GL(nr). A nilpotent orbit

Oγ of GL(n) is contained in G · ( Oαi + nn) if and only if the number

N γ Q = cβ2 cβ3 cβ4 ...cγ α1,...,αr α1,α2 β2,α3 β3,α4 βr−1,αr β2,...,βXr−1 is nonzero. γ s In terms of Schur functions, the number Nα1,...,αr is the coeﬃcient of γ in the product of the Schur polynomials sn1 ... snr .

Acknowledgements. I thank Dmitry Gourevitch for his support, enlightening discussions and valuable comments about this project. This research is supported by the Israel Science Foundation (ISF) 249/17 and ERC Starting Grants 637912.

2. Preliminaries and Notation In this section, we specify the notations which are used throughout this paper. Let G be a semisimple algebraic group over C, we denote by • T ⊂ B ⊂ G a choice of maximum torus T and a Borel subgroup B, ∗ • X∗(T ) the cocharacter lattice, and X (T ) the character lattice of T , • ∆G(T ) the set of roots, ∆G,+(T ) the set of positive roots and ∆G,s(T ) the set of simple roots, and ∆G(T ) the set of coroots. We can identify the character lattice as the integer lattice Zn where n is the rank of G. Each partition λ = (λ1,...,λn) corresponds to the character

λ1 λn (t1,...,tn) 7→ t1 ...tn . In particular, under this correspondence, the half sum of positive roots 1 ρ = α 2 α∈X∆G,+ ∗ G corresponds to the partition (n−1,n−2,..., 1, 0). Fixing a maximal torus T , the quadruple (X , ∆ ,X∗, ∆G) ∗ G is called the root datum of G. Its dual root datum (X∗, ∆G,X , ∆ ) is obtained by switching the character lattice/roots and the cocharacter lattice/coroots. The algebraic group corresponding to the dual root datum is called the Langlands dual group and is denoted by Gˇ. 2 2.1. Young Diagrams. In this section we will summarize the basic concepts related to Young diagrams and define the Littlewood-Richardson coefficients. A partition α = (α1,...,αr) of an integer n corresponds to a Young diagram (whose i-th row has length αi) of the shape α. Ina Young tableau of the shape α, integers, starting with 1, are filled into the boxes of the Young diagram α. We denote the set of all Young tableaux of the shape α by Tα. A Young tableau is called standard if the entries of each box are strictly increasing from the left to right in each row, and increasing from top to bottom in each column. If the entries are weakly increasing along the rows, the tableau is called a semistandard tableau. It is worth noting that each standard Young tableau T of the shape α corresponds to a sequence of Young diagrams

0= α0 ⊂ α1 ⊂ ... ⊂ αr = α with each αi formed by the boxes in T with contents ≤ i. For example, the Young tableau

T = 1 1 2 1 3 2 corresponds to the following sequence of Young diagrams: ∅⊂ ⊂ ⊂ .

Similar to the usual Young diagrams discussed above, for each pair of Young diagrams (α, γ) such that α ⊂ γ, a skew Young diagram is the complement of α in γ, with the two diagrams aligned along their top and left edges. For example, for α = (2, 1) and γ = (4, 3, 2), the skew Young diagram γ/α looks like .

We denote the set of all semistandard skew tableaux of the shape γ/α by Tγ/α. Similar to the correspondence between Young tableaux and sequences of Young diagrams described above, any skew tableau T ∈ Tγ/α corresponds to a sequence of Young diagrams from α to γ, whose i-th stage αi consists of the boxes in T such that the entries in each of the selected boxes are ≤ i. Assuming the maximal content of the tableaux is r, there is a sequence α = α0 ⊂ α1 ⊂ α2 ⊂ ... ⊂ αr = γ. The entries of such a skew tableau also gives us a new partition

β = (|α1/α0|,..., |αr/αr−1|). For example, the skew tableau 1 1 2 1 2 1 2 yields the partition β = (4, 3). We say a tableau T , or a corresponding sequence of Young diagrams from α to γ has type (α, β; γ) if this new partition is β. 2.2. The Littlewood-Richardson Rule. For each pairs of Young diagrams (α, γ) such that α ⊂ γ, a semistandard skew Young tableau is Littlewood-Richardson if its reverse lattice words, i.e. the string of content of each box, read from top to bottom and from right to left like in Hebrew, satisfies the Yamanouchi condition, i.e. for each content i, the number of i’s in the length k prefix of its reverse lattice word is great or equal to the number of i + 1’s in the length k prefix. For example, the reverse lattice word of the tableau 1 1 1 2 2 3 is 112132. It satisfies the Yamanouchi condition. γ For each triple of Young diagrams (α, β; γ) such that α, β ⊂ γ, the Littlewood-Richardson coefficient cα,β is defined as the number of Littlewood-Richardson tableaux of type (α, β; γ). The usual combinatorial description of the Littlewood-Richardson coefficient is given by the Hall-Littlewood γ polynomials Gα,β(q). For a pair (O, p) consisting of a discrete valuation ring O and its maximal ideal p with γ a residue field of size q, and a triple of partitions (α, β; γ), the Hall-Littlewood polynomial Gα,β(q) is defined 3 ∼ ∼ γ as the number of submodules N ⊂ Mγ such that N = Mα and Mγ/N = Mβ. The Hall polynomial Gα,β(q) γ has degree hρ, α + β − γi, and its top-degree coefficient is equal to the Littlewood-Richardson coefficient cα,β. A detailed proof of this fact is in the appendix of [Mac98, Chapter 2]. However, it’s important to have a geometric model of the abelian group extensions so that we can get more information related to nilpotent orbits. In the following sections, we will introduce the theory of affine Grassmannian and a geometric interpretation of the Littlewood-Richardson coefficient.

3. Lattices and Affine Grassmannian In this paper, we will introduce four major pictures for the Littlewood-Richardson coefficient: the first picture is the extensions of torsion C[t]-modules and the top-degree coefficient of the Hall-Littlewood polynomial, which we have already introduced in the previous section. The second picture is the extension of sublattices in On. This picture is closely related to the first picture. The third picture uses the geometric Satake isomorphism and the interpretation of Littlewood-Richardson coefficients as decomposition multiplic- ities for tensor products of finite dimensional representation of GLn, which are directly related to the affine Grassmannian. The fourth picture is the number of irreducible components of certain subvarieties of the Springer fibers in a generalized flag manifold. The whole Section 3 is devoted to the discussion of the second and the third picture. The fourth picture will be introduced in Section 4. In this section we would like to discuss the link between module extensions and geometry. Throughout this paper, we fix a ring O = C [[t]] with a maximal ideal p = tO. We will establish a geometric model for the extensions of finite dimensional modules of the form

λ1 λn Mλ = C[t]/(t ) ⊕ ... C[t]/(t ) where λ = (λ1 ...,λn) is a partition. We can add zeros to the end of λ without causing any change to Mλ. The geometric model of module extensions is given by the following lemma, which we will prove in the later sections of this paper:

Lemma 3.0.1. For three partitions α,β,γ and three modules Mα,Mβ,Mγ as deﬁned above, there exists an exact sequence

0 → Mα → Mγ → Mβ → 0 γ if and only if the partitions α,β,γ makes the Littlewood-Richardson coefficient cα,β non-vanishing. The γ Littlewood-Richardson coefficient cα,β is equal to the number of isomorphism classes of all such possible extensions. The first part of Theorem A is a direct consequence of this lemma.

3.1. Preliminaries on Affine Grassmannian. The affine Grassmannian is a geometric object parametriz- ing the full-rank O-sublattices in On. Definition 3.1.1. Setting O = C [[t]] and K = C ((t)), which are the rings of formal power series and formal Laurent series, respectively, let G be a semisimple algebraic group, the affine grassmannian of G is defined as the quotient

GrG = G (K) /G (O) . The affine grassmannian is an ind-scheme, which means that it is the direct limit of finite-dimensional closed subschemes. To see this, we need a stratification on the affine Grassmannian based on the Cartan decomposition (See [MV07, Section 2] and [BR18, Proposition 1.3.2]), in which the cells are parametrized by the dominant coweights

(1) G(K)= G (O) · Lλ · G (O) . + λ∈Xa∗(T ) λ1 λn In the case that G = GLn, Lλ is the diagonal matrix Lλ = diag t ,...,t where λ = (λ1,...,λn) is a partition of length n. We define an affine Schubert cell as the quotient λ GrG = G (O) · Lλ · G (O) /G (O) . 4 λ µ By [BR18, Proposition 1.3.2], the closure GrG of this cell is the disjoint union of affine Schubert cells GrG with µ ≤ λ: λ µ GrG = GrG. + µ∈X∗(T ) µa≤λ In the disjoint union above, the order µ ≤ λ on the partitions are given by the closure order described in the beginning of Section 1. 3.2. Affine Schubert Varieties and Lattices in On. We will use the affine Schubert cycles to parametrize full-rank O-sublattices in On. The correspondence is established by the following lemma: + λ Lemma 3.2.1. For any dominant coweight λ ∈ X∗(T ) , there is a bijection between the elements x ∈ GrG n and the full-rank O-sublattices Λx ⊂ O such that

n λi O /Λx =∼ C[t]/(t ). i M λ Proof. Since each element x ∈ GrG represents an element in G(O)LλG(O) of the form

x = αLλ, α ∈ G(O),

the column vectors of Lλ generates a sublattice

−1 λ1 λn α Λx = t O⊕ ... ⊕ t O −1 n n −1 n n of α O =∼ O . The quotient α (O /Λx) =∼ O /Λx is a module which satisfies the requirement of the lemma. Conversely, if there is a lattice Λx = i Ofi which satisfies the requirement of the lemma, the isomorphism will give an isomorphism between lattices p : Otλj → Λ = Of such that f = r tλj L j x i i j i ji where the matrix (r ) is an invertible matrix in G(O). ji L L P As in the second picture mentioned in the beginning of Section 3.1, in order to for us to describe the set of extensions of lattices, we need to introduce the twisted product GrG×GrG of two copies of the affine −1 Grassmannians. Letting G(O) act on G(O)×G(O) by k·(a,b) = (ka,k b), the scheme GrG×GrG is defined as the quotient of G(K) × GrG by the action of G(O) on G(O) × G(O), fittinge into the diagram (2) G(K) × Gr e ❖G ♦♦♦ ❖❖❖ p♦♦♦ ❖❖q❖ ♦♦♦ ❖❖❖ ♦w ♦♦ ❖' GrG × GrG GrG×GrG where the morphism p is the quotient by G(O) from the first component. There is a semi-small, ind-proper e multiplication map m : GrG×˜ GrG → GrG defined by m : (g, x) 7→ gx. The reason why this map m is called the multiplication map will be explained in Lemma 3.2.4. The twisted product × can be defined similarly for the Schubert cells and Schubert varieties by simply taking the quotient of the direct product of the corresponding cells by G(O). When restricted to the twisted e α β α+β product of the Schubert varieties, the multiplication map m maps GrG×GrG surjectively onto GrG . We can also take the twisted product of multiple affine Grassmannians and Schubert varieties. The variety λ1,...,λn GrG is defined as the multiple twisted product e λ1,...,λn λ1 λn GrG = GrG × ... ×GrG .

1 n We can stratify the space Grλ ,...,λ by the twisted product of Schubert cells corresponding to smaller G e e partitions: λ1,...,λn µ1,...,µn GrG = GrG . i i µ[≤λ To count the dimension of the twisted product of these Schubert cells, we need the following lemma from [Zhu16] and [BR18]: 5 1 n λ ,...,λ n i Lemma 3.2.2. The dimension of the twisted product of cells GrG is equal to h2ρ, i=1 λ i. The 1 n −1 λ ,...,λ λ dimension of the ﬁber m (x) ∩ GrG of m over any generic point x ∈ GrG ⊂ GrG is less than or equal n i P to hρ, i=1 λ − λi, hence m is a semi-small map. λ −1 λ Proof.PSince the stablizer of the action of G(O) at L is G(O) ∩ LλG(O)(Lλ) , thus the tangent space GrG at Lλ is ∼ hα,λi g(O)/g(O) ∩ AdLλ g(O) = gα(O)/t g(O). hα,λMi≥0 The dimension of this tangent space above h2ρ, λi. The dimension of the twisted product follows from a similar stabilizer argument. The inequality for the 1 n −1 λ ,...,λ dimension of the ﬁber m (x)∩GrG follows from [MV07, Lemma 4.4], also see [BR18, Lemma 6.4].

1 t 2 1 t 2 Remark 3.2.3. For any partition λ, the number hρ, λi is closely related to 2 kλ k = 2 (λi) . Since we know t P λi − λi+1 = # j | λj = i , 2 we can express 1 kλtk in terms of λ: 2 1 2 λt = i2(λ − λ ). 2 i i+1 i=1 X Through simple calculation, we see that n 1 2 1 λt + hρ, λi = n − λ . 2 2 k Xk=1 1 n −1 λ ,...,λ Therefore, we can express the dimension of the ﬁber m (x) ∩ GrG diﬀerently in terms of the conjugate partitions:

1 n 1 2 2 (3) dim m−1(x) ∩ Grλ ,...,λ = λt − λt . G 2 i i ! X α β We can use the variety GrG×GrG to parametrize the extensions of lattices (as promised in our second picture). This fact is shown in the following lemma: e n α ˜ β Lemma 3.2.4. For any lattice Λ ⊂ O , denote by Λ = HomO(Λ, O) its dual lattice. The variety GrG×GrG n parametrizes the collection of sublattices Λγ ⊂ Λα ⊂ O , such that (1) They both span the same vector space Kn overb K = C((t)), n n (2) Λγ/O =∼ Mγ, and Λα/O =∼ Mα, (3) Λα/Λγ =∼ Mβ. c c α ˜ β α+β The multiplication map m : GrG×GrG → GrG sends any pair of such lattices (Λγ, Λα) to Λγ. α β Proof. The scheme GrG×GrG is a G(O)-quotient of the product G(O)LαG(O)×G(O)Lβ , which parametrizes pairs of elements (l1,l2) with l2 ∈ G(O)Lβ and l1 ∈ G(O)LαG(O). Each equivalent class under the quotient by G(O) has a uniquee representative of the form (xLα,yLβ), and can be moved to a representative −1 −1 n (Lα, x yLβ). In the moved lattice x O , one can consider the lattice generated by the column vectors of −1 the matrix Lαx yLβ. We would like to prove that this lattice is the correct xΛγ ,

Lα = (e1,...,en),

αi t −1 where each ei = (0,...,t ,..., 0) is only nonzero at the i-th entry. We represent by a matrix x yLβ = −1 (gi,j ), and column vectors of the matrix Lαx yLβ are

αi vj = (t gi,j)= gi,j ei i X ∼ which generates a sublattice xΛγ of i Oei which satisﬁes Λα/Λγ = Mβ. 6 P By Lemma 3.2.4, the two projections π1, π2 can be described as maps from the twisted product GrG×GrG onto the ﬁrst and the second factor, respectively. The projection π1 sends each pair of (Λγ, Λα) toΛα, while −1 −1 n the projection π2 sends the pair (Λγ , Λα) to the sublattice Λβ = Lα x Λγ of O . These two projectionse can also be used to set up an isomorphism between the twisted product GrG×˜ GrG and the direct product GrG × GrG, which sends any element y in the twisted product to (π1(y), π2(y)). In terms of group element representatives, the isomorphism can be described by the map

(4) GrG × GrG −→ GrG×˜ GrG −1 (x1, x2) 7−→ (x1, x1 x2).

3.3. Equivariant Perverse Sheaf and the Geometric Satake Equivalence. According to [BL06, Theo- rem 2.6.2], for any G(O)×G(O)-equivariant constructible sheaf F on G(K)×GrG, there is an unique G(O)- ∗ equivariant constructible sheaf F on GrG×˜ GrG such that q F = F. In particular, for two constructible ∗ ∗ ⊠ ∗ sheaves A, B on GrG, consider the pullback p (π1 A π2 B) as the sheaf F, where π1, π2 are the projection to the first and second factors definede in the previous section,e respectively. We denote the corresponding ∗ ⊠ ∗ unique constructible sheaf F by π1 A π2 B. We can thus define a convolution product in the derived category of G(O)-equivariant constructible sheaves on GrG as e e ∗ ⊠ ∗ A ⋆ B = Rm! π1 A π2 B . Equipped with the convolution product, the category PervGe(O)(GrG) of G(O)-equivariant perverse sheaves is a monoidal category with irreducible objects the intersection cohomology sheaves λ λ C λ C λ IC := IC(GrG, )= ι!∗ [dim GrG]

λ supported on each Schubert variety GrG, which are the intermediate extensions of the locally trivial bundle λ λ λ of GrG along the inclusion ι : GrG ֒→ GrG. The main takeaway in the proof of our main Lemma 3.0.1 is the calculation of the convolution product ICλ1 ⋆ ICλ2 of the two intersection cohomology sheaves. The main reference of the equivariant perverse sheaves on the aﬃne Grassmannian in geneal is [BR18, Section 7.5].

The geometric Satake equivalence (see [Gin95],[MV07] or [BR18]) is an equivalence of monoidal categories between the category of G(O)-equivariant perverse sheaves PervG(O) (GrG) and the category of ﬁnite dimensional representations RepC Gˇ of the Langlands dual group Gˇ: ∼ ˇ PervG(O) (GrG) = RepC G . The actual construction of the categorical equivalence is not trivial, and requires the Tannakaian reconstruc- tion theorems for an actual group scheme Gˇ which makes the following diagram of functors commute: S / ˇ PervG(O) (GrG) RepC G . ◆◆ ∗ s ◆◆ H ss ◆◆◆ ss ◆◆ ssforget ◆◆' sy s VectC This commutative diagram of functors respects the tensor products and the units of the three tensor categories. The total cohomology functor H∗ is the graded cohomology which decomposes into weight components: k h2ρ,µi ! H (GrG, A) =∼ H (Tµ, ι A) µ∈X∗(T ) h2ρ,µMi=k where Tµ is the locally closed subvariety of GrG consisting of the points whose limit at ∞ is Lµ under the adjoint action of a regular dominant coroot.

If we identify the cocharacter lattice X∗(T ), which parametrizes the irreducible objects of the category ∗ ˇ PervG(O) (GrG), with the character lattice X (T ) of the Cartan subgroup in the Langlands dual group, the 7 λ functor S sends the intersection cohomology sheaf supported on GrG for each dominant coweight λ to the irreducible representation V (λ) of Gˇ with highest weight λ:

ICλ /o S/o /o / V (λ) , and the convolution product of the intersection cohomology sheaves λ1 λn ∗ λ1 ⊠˜ ⊠˜ ∗ λn IC ⋆...⋆ IC = Rm! π1 IC ... πnIC

1 n is sent to the tensor product V λ ⊗ ... ⊗ V (λ ). The functor Rm!, which can be identiﬁed as Rm∗ n ∗ λ1 ⊠˜ ⊠˜ ∗ λ λ1,...,λn due to the properness of m, sends a complex of constructible sheaves π1 IC ... πnIC on GrG λ1+...+λn to a complex of constructible sheaves on GrG . As a derived direct image, the constructible sheaf 1 n ICλ ⋆...⋆ ICλ is perverse due to the semi-smallness of the morphism m. We need a lemma for use in the next section: ∗ λ1 ⊠˜ ∗ λ2 Lemma 3.3.1. The twisted product π1 IC π2 IC is isomorphic to the intersection cohomology sheaf λ1,λ2 IC(GrG ). Proof. Consider the following pull-back square

λ1 λ2 h / λ1,λ2 GrG × GrG GrG

π2 m λ1+...+λn λ1+...+λn GrG GrG −1 such that the top row isomorphism h sends (x1, x2) to (x1, x1 x2) as described in the final remarks of Section λ1 λ2 λ1,λ2 3.2. The morphism h is an isomorphism which preserves the strata of GrG × GrG and GrG , and hence sends the external tensor product of IC-sheaves ∗ λ1 ⊠ ∗ λ2 ∼ λ1 λ2 π1 IC π2 IC = IC(GrG × GrG ) ∗ λ1 ⊠ ∗ λ2 to the twisted tensor product π1 IC π2 IC . Therefore, the twisted tensor product is also the IC-sheaf λ1,λ2 IC(GrG ). e The fact that S is an equivalence of tensor categories identifies the hom-sets between the intersection cohomology sheaves and those between the finite dimensional representations: γ α β Hom(IC , IC ⋆ IC ) =∼ HomG(V (γ), V (α) ⊗ V (β)). γ The dimension of this space is equal to the Littlewood-Richardson coefficient cα,β. 3.4. The Decomposition Theorem and the Proof of the Lemma 3.0.1. The derived direct image λ1,...,λn λ1,...,λn Rm∗IC GrG of the intersection cohomology sheaf on the variety GrG along the semi-small map

m : GrG×GrG →GrG can be decomposed non-canonically as a direct sum of irreducible objects. See [dCM09, Theorem 1.6.1] for the following well-known decomposition theorem: e Theorem 3.4.1 (Be˘ilinson-Bernstein-Deligne-Gabber). Let f : X → Y be a proper map of algebraic varieties, the derived push-forward Rf∗ of an intersection cohomology sheaf ICX decomposes (noncanonically) into a direct sum of semisimple perverse sheaves p i Rf∗ICX =∼ H (Rf∗ICX )[−i]. Z Mi∈ p i There is a canonical decomposition of H (Rf∗ICX ) as a direct sum of intersection cohomology sheaves on p i p p ≥i p ≥i p Y , where H is the perverse cohomology functor τ≤i ◦ τ = τ ◦ τ≤i. We will make use of the decomposition theorem in the form of its corollary (see also [Hai03, Lemma 3.2]): 8 Corollary 3.4.2. The convolution product of IC-sheaves decomposes into a direct sum

α β γ γ IC ⋆ IC = Lα,β ⊗ IC γ≤Mα+β γ h2ρ,α+β−γi −1 α,β γ where Lα,β = Hc (m (x) ∩ GrG ) for a generic x ∈ GrG.

α,β α,β α,β α,β Proof. We consider embeddings of the big cell GrG ⊂ GrG and its complement Z in GrG :

j α,β i / α,β o ? _ α,β Z GrG GrG and the corresponding embeddings of their intersections with the ﬁber of m:

iγ jγ −1 α,β / −1 α,β o ? _ −1 α,β m (x) ∩ Z m (x) ∩ GrG m (x) ∩ GrG .

p i α,β γ The stalk of the cohomology sheaf H Rm∗IC(GrG ) at x ∈ GrG can be identiﬁed as the cohomology α,β i −1 − group H (m (x), IC(GrG ) |m 1(x)). To compute this cohomology group, we need to apply the cohomology functor to the distinguished triangle

∗ α,β α,β ∗ α,β j!j IC(GrG ) → IC(GrG ) → i!i IC(GrG ) → which gives rise to a long exact sequence of cohomologies (5) i −1 α,β α,β i −1 α,β i −1 α,β α,β ... → Hc(m (x) ∩ GrG , IC(GrG )) → H (m (x), IC(GrG )) → Hc(m (x) ∩ Z , IC(GrG )) → ....

α,β i ∗ α,β Since IC(GrG ) is a perverse sheaf, the cohomology sheaf H (iγ IC(GrG )) = 0 if i ≥ −h2ρ, α + βi. Also since the hypercohomology spectral sequence converges:

pq p −1 α,β q α,β p+q −1 α,β ∗ α,β E2 : H (m (x) ∩ Z , H IC(GrG ) =⇒ H (m (x) ∩ Z ,iγIC(GrG )), all the nonzero terms in this spectral sequence appears when q< −h2ρ, α + βi, p ≤ dim(m−1(x) ∩ Zα,β) ≤ h2ρ, α + β − γi− 2.

i −1 α,β ∗ α,β Therefore, Hc(m (x) ∩ Z ,iγIC(GrG )) can be possibly nonzero only when i ≤ −h2ρ,γi− 2. Again by the long exact sequence (5), the stalk of the cohomology sheaf

−h2ρ,γi −1 α,β ∼ −h2ρ,γi −1 α,β ∗ α,β H (m (x), IC(GrG )) = Hc (m (x) ∩ GrG , jγ IC(GrG )). α,β C When restricted to the open cell, IC(GrG ) is the constant local system [h2ρ, α + βi]. Therefore, the stalk of the cohomology sheaf at x is isomorphic to

−h2ρ,γi −1 α,β ∼ h2ρ,α+β−γi −1 α,β H (m (x), IC(GrG )) = Hc (m (x) ∩ GrG ). γ We denote this space by Lα,β. This ﬁnishes the proof of the corollary. Now we are ready to prove Lemma 3.0.1.

γ Proof of Lemma 3.0.1. By Lemma 3.2.1, any element x ∈ GrG represents a lattice Λx such that the quotient n O /Λx has Jordan type γ. If the multiplication map

α β α+β m : GrG×GrG → GrG

sends a pair y = (Λγ, Λα) to Λx, by Lemma 3.2.4 we must have Λγ = Λx. Therefore, in order to prove e γ −1 α,β Lemma 3.0.1, we will have to show that for any x ∈ GrG, the ﬁber m (x) ∩ GrG is nonempty if and only γ γ if cα,β = dim Lα,β 6= 0. 9 By the decomposition theorem and the semi-simplicity of the category of equivariant perverse sheaves, Hom(ICγ , ICα ⋆ ICβ) can be decomposed noncanonically as a direct sum:

γ α β ∼ γ α,β Hom(IC , IC ⋆ IC ) = Hom(IC ,Rm∗IC GrG )

 γ σ σ = Hom IC , Lα,β ⊗ IC  σ α+β  GrG⊂Mx∈GrG  −   m 1(x)∩Grα,β 6=∅   G  γ   = Lα,β. By the geometric Satake equivalence, the dimension of the space is equal to the multiplicity of V (γ) in γ V (α) ⊗ V (β), which is equal to the Littlewood–Richardson coeﬃcient cα,β. γ Since on each Schubert cycle GrG, the group G(K) acts transitively, the preimage of the Schubert cycle −1 γ α,β −1 α,β m (GrG) intersects GrG if and only if the ﬁber m (x) intersects GrG . By [Zhu16, Corollary 5.1.5], the dimension of the space γ h2ρ,α+β−γi −1 α,β Lα,β = Hc (m (x) ∩ GrG ).

−1 α,β is the number of irreducible components of m (x) ∩ GrG of complex dimension hρ, α + β − γi. All the i i n−i dominant coweights of GLn are sums of coweights of the form ǫ = 1 , 0 , which are all minuscule, i.e. −1 α,β hǫi, αi∈{−1, 0, 1} for any root α. By [Hai06, Theorem 1.3], the irreducible components of m (x)∩GrG , if not empty, are equidimensional and of dimension hρ, α + β − γi. Since the Littlewood-Richardson coeﬃcient γ −1 α,β −1 cα,β is equal to the number of irreducible components of m (x) ∩ GrG , we have now proven that m (x) ∩ α,β γ GrG 6= ∅ if and only if cα,β 6= 0.

4. Interpretations of the Littlewood-Richardson Coefficient γ In this section, we consider the Young diagrams with shapes α, β ⊂ γ such that cα,β 6= 0. We will ﬁx our notations to denote the length of γ by m, the length of α by n. The goal of this section is to prove the second part of the Theorem A, which follows from the geometric interpretation of the Littlewood-Richardson coeﬃcients studied by Springer in [Spr78] and Marc van Leeuwen in [vL00]. This geometric interpretation γ of cα,β is our fourth picture mentioned in the beginning of Section 3.1.

γ −1 α,β 4.1. The Satake Fiber. The fiber Sα,β = m (Lγ) ∩ GrG of the multiplication map m above the point Lγ ∈ Grγ is sometimes referred to as the Satake fiber. The following lemma identifies the Satake fiber as a subvariety of the partial flag variety G/P[n,m−n]:

γ Lemma 4.1.1. The Satake ﬁber Sα,β parametrizes the subspaces V of dimension |α| inside a vector space W of dimension |α| + |β|, such that there is a nilpotent element x which stablizes V ⊂ W , and

(1) x |V acts with Jordan type α, (2) x acts on W/V with Jordan type β, (3) x acts on W with Jordan type γ.

−1 α,β Proof. The Satake ﬁber m (Lγ)∩GrG parametrizes the pairs of lattices (Λγ , Λα) satisfying the conditions n in Lemma 3.2.4. The quotients Mα,Mβ,Mγ of O by the three lattices Λα, Λβ and Λγ, respectively, ﬁt into the exact sequence

0 → Mα → Mγ → Mβ → 0. The multiplication by the variable t of the ring O[[t]] is a nilpotent linear operator with Jordan type α,β,γ on the three spaces, respectively. Conversely, if we have a pair of spaces V ⊂ W satisfying the conditions of this lemma, V and W can be viewed as the O-modules Mα,Mγ, respectively. 10 4.2. The Springer-type Fibers. The tangent bundle of the ﬂag variety G/B can be realized as a subset of the product of the nilpotent cone N and B = G/B: T ∗B = {(x,gB) | x ∈ Ad(g)b}⊂N×B.

∗ For each nilpotent orbit Oγ ⊂N ⊂ g, the projection p from the cotangent bundle T B to B is a resolution of singularities called the Springer resolution for the nilpotent cone N : T ∗B⊂N×B q ▲▲ qq ▲▲ p πqqq ▲▲ qq ▲▲▲ qx qq ▲▲& NB

where π,p are the projections onto the ﬁrst and the second factor, respectively. The Springer ﬁber Bγ(x) above x ∈ Oγ of p is the collection of Borel subalgebras b ⊂ g containing the nilpotent element x

Bγ(x)= {(x,gB) | x ∈ Ad(g)b} . It is proven by Spaltenstein in [Spa76] and Steinberg in [Ste76] and [Ste88] that the irreducible components of the Springer fiber Bγ(x) are equidimensional and are parametrized by the standard Young tableaux of shape γ. The number of standard Young tableaux can be calculated from the hook-length formula. For G = GLm, the Springer fiber Bγ(x) can be interpreted as the collection of x-stable full flags

{0} = V0 ⊂ V1 ⊂ V2 ⊂ ... ⊂ Vm = V in a vector space Vm of dimension m, on which x acts as a nilpotent linear transformation. Since the Springer fibers are isomorphic for different choices of x in the same orbit Oµ, we can denote the Springer fiber by Bγ if there is no ambiguity caused by different choices of x.

Apart from the case of the full flag variety G/B, we can also consider the Springer-type partial resolution from the cotangent bundle of a partial flag variety G/P corresponding to the parabolic subgroup P = LN. Similar to the Springer resolution of the nilpotent cone by the cotangent bundles of a full flag variety, given a parabolic subgroup P , we can define a partial resolution NP of the nilpotent cone N

πP NP = {(x, gP ) | x ∈ N ∩ Ad(g)pg}⊂N× G/P −−→N which factorizes the Springer resolution π. The generalized Springer fiber π−1(x) above any element x g P is sometimes referred to as the Steinberg variety (cf. [BM, Chapter 3], some references also call it the Spaltenstein variety). When P = B, the partial resolution πP becomes the usual Springer resolution, and −1 the fiber πP (x) becomes the usual Springer fiber above x. Equivalently, if we consider the parabolic subgroup P as the stabilizer of a partial flag

Vk1 ⊂ Vk2 ⊂ ... ⊂ Vkr = Vm −1 where each Vki is a subspace of dimension ki. The fiber πP (x) can be identified as the collection of the x-stable flags which are stabilized by a conjugate of P .

We are particularly interested in the standard parabolic subgroups P[n,1m−n] with a Levi subgroup ∼ m−n ∼ L[n,1m−n] = GL(n) × GL(1) , and P[n,m−n] with Levi subgroup L[n,m−n] = GL(n) × GL(m − n). For the parabolic subgroup P[n,1m−n], the generalized Springer fiber above x ∈ Oγ , denoted by P[n,1m−n](x), is the set of x-stable partial flags Vn ⊂ ... ⊂ Vm−1 ⊂ Vm in an m-dimensional vector space Vm whose lowest piece Vn has dimension n, and for every two consecutive spaces, we have dim(Vi+1/Vi) = 1. The fiber P[n,1m−n](x) can be considered as −1 P[n,1m−n](x)= {(x, gP[n,1m−n]) | Ad(g) x ∈ p[n,1m−n]}.

Similarly, for the parabolic subgroup P[n,m−n], the generalized Springer ﬁber above x ∈ Oγ , denoted similarly by P[n,m−n](x), is a subvariety of G/P[n,m−n] given by

−1 P[n,m−n](x)= {(x, gPn,m−n) | Ad(g) x ∈ p[n,m−n]}. 11 4.3. Springer Fibers, Jordan Forms, and the Proof of Part 2 of Theorem A. First, let’s consider the case of P[n,1m−n](x) and interprete this ﬁber as the collection of ﬂags between Vn and Vm. Recalling the notations in Section 2.1, for any nilpotent element x, take any skew Young tableau T of the shape γ/α with the correponding sequence of Young diagrams

α = α0 ⊂ α1 ⊂ ... ⊂ αm−n = γ T such that β = (|α1/α0|,..., |αm−n/αm−n−1|), we can deﬁne a closed subvariety P (x) as the collections of x-stable ﬂags on which the nilpotent element x acts on each stage Vn+i by a Jordan form λi, i.e. the restriction of x on Vn+i has Jordan form given by a partition αi.

m For any nilpotent element x ∈ Oγ, there is a subvariety Gn (x) of the Grassmannian variety G(n,m) which parametrizes the x-stable subspaces of Vm of dimension n. By the arguments in the Section 4 of [vL00], this subvariety of the Grassmannian variety G(n,m) can be decomposed into cells Gα,β(x) such that the Jordan ′ type of x |Vn on Vn is α, and the Jordan type of x , the nilpotent linear operator acting on the quotient space Vm/Vn induced from x, is β: m Gn (x)= Gα,β(x). |α|=n |γ[|=m α,β⊂γ γ By Lemma 4.1.1, each cell Gα,β(x) is isomorphic to the Satake fiber Sα,β. Moreover, the action of the nilpotent element x with Jordan type β on the space Vm/Vn corresponds to an x-stable filtration Vn ⊂ Vn+1 ⊂ ... ⊂ Vm such that the action of x on each quotient Vn+i/Vn has Jordan form αi. The collection of all the flags which satisfy this property can be identified with the subspace Pα,β(x) of the partial flag variety G/P[n,m−n] defined below: −1 −1 Pα,β(x)= gP | Ad(g) x ∈ p[n,m−n], and pl(Ad(g) x) ∈ Oα,β ∼ where pl : p[n,m−n] → p[n,m−n]/u[n,m−n] = l[n,m−n] is the projection map. Letting (α, β) go over all the possible pairs of Young diagrams satisfying |α| + |β| = |γ|, the union of the spaces Pα,β(x) is exactly the generalized Springer fiber P[n,m−n](x) over a point x ∈ Oγ of the partial flag variety G/P[n,m−n]. m The correspondence between flags and parabolic subgroups establishes isomorphisms between Gn (x) and P[n,m−n](x), and between Gα,β(x) and Pα,β(x). Now we prove the second part of the Theorem 3.0.1:

−1 Proof. If x ∈ Oγ and gP[n,m−n] ∈Pα,β(x), each Ad(g) x ∈ p[n,m−n] ∩Oγ , and the space Pα,β(x) is a closed subvariety of G/P[n,m−n] which consists of those parabolic subalgebras q = m + u conjugate to p[n,m−n] such that x ∈ q and pm (x) ∈ Oα,β. In [Spr78], [vL00] and the proof of the Lemma 3.4, it is stated that the dimensions of the irreducible γ components of Pα,β(x), Gα,β(x) and the Satake ﬁber Sα,β are equal to hρ, α + β − γi, and the number of γ their components is equal to the Littlewood-Richardson coeﬃcient cα,β. γ Now we consider the subset Qα,β of G/P[n,m−n] consisting of parabolic subalgebras q = m + u such that

Oα,β ⊂ pm(Oγ ∩ q), γ the fiber Pα,β(x) is the intersection of Qα,β with the generalized Springer fiber P[n,m−n](x): γ Pα,β(x)= Qα,β ∩P[n,m−n](x). There is map γ πα,β : G ⊃Pα,β(x) · P −→ Oγ ∩ p[n,m−n] g 7−→ Ad(g)−1x, γ which is the quotient by the centralizer CG(x) of x from the left. We denote the image of this map πα,β by γ t 2 Pα,β. In the GL(n) case, the centralizer CG(x) has only one connected component with dimension i(γ )i . 12 P γ The number of connected components of Pα,β(x) · P is still cα,β, since multiplying by an element p ∈ P simply fixes a basis of the flag. By (3) stated in Remark 3.2.3, the dimension of the connected components of Pα,β(x) · P is equal to

1 2 2 dim P (x) · P = (γt)2 − (αt)2 − (βt)2 + αt + βt + αt βt . α,β 2 i i i i i i i i i i ! X X X X X X X γ Therefore, the dimension of Pα,β satisﬁes γ dim Pα,β = dim Pα,β(x) · P − dim CG(x) > dim n − dim Pα,β(x) > 0.

The union of the closed orbits of CG(x) will not be Zariski-dense. Therefore, the number of irreducible γ γ components of Pα,β is also cα,β. Since we can interprete the image of the map as the elements in Oγ ∩p[n,m−n] whose projections along pl lies in Oα,β, we have compeleted the proof of the second part of the Theorem A.

5. Applications: Wave Front Sets and Associated Varieties In this section, let G be a semisimple algebraic group over an archimedean ﬁeld F . We would like to introduce invariants to describe of the “growth” of irreducible representations π of the group G. We denote the Lie algebra of G by g0, and its complexiﬁed Lie algebra by g.

5.1. Associated Varieties and Annihilator Varieties. For any irreducible representations π of G, the “growth” of π is recorded in the associated variety As(π) and its annihilator variety An(π). The algebra U(g) and its ideals are ﬁltered by the degrees, and the annihilator variety An(M) and the associated variety As(M) of a module M are deﬁned as the zero sets of the ideal Gr (Ann(M)) and Ann (Gr(M)), respectively. For groups over R and C, it turns out that the annihilator variety An(π) of any irreducible representation π is the closure of an unique orbit O(π) (cf. [Jos85, 3.10]).

5.2. Induction, Jacquet Functors and Associated Varieties. In [GS20], Gourevitch and Sayag discussed a family of Lagrangian submanifolds of the annihilator variaties of any irreducible U(g)-module. According to [GS20, Corollary 6.6]: Theorem 5.2.1. For any irreducible representation π of G and a parabolic subgroup P = LN, denoting by rP the Jacquet functor along the nilpotent radical N, then for any irreducible quotient τ of rP (π), ⊥ O(τ) ⊂ pl(O(π) ∩ n ). Under the duality given by the Killing form, g∗ and g are isomorphic, and the following spaces correspond under such isomorphism: g g∗ =∼ g n⊥ p ⊥ ⊥ ⊥ projection pl : p → p/n projection ql : n → n /p O(π) ∩ n⊥ O(π) ∩ p ⊥ −1 n ∩ pl O(τ) O(τ)+ n. Therefore, under the Killing form duality, the following two inclusions are equivalent: ⊥ O(τ) ⊂ pl O(π) ∩ n ⇌ O(π) ⊂ G · (O(τ)+ n).

5.3. Examples. In this section we will give some examples and applications to the Theorem A.

5.3.1. n =3. Consider the standard parabolic subgroup P[21] = L[21]N[21] where the Levi subgroup L[21] = GL(2) × GL(1), and the unipotent radical 1 0 ∗ N[21] = 0 1 ∗ . 0 0 1 13  2 The generalized ﬂag variety G/P[21] is isomorphic to P , and the isomorphism can be written explicitly down with the Pl¨ucker coordinates

2 3 G/P[21] −→ P ∧ C [g] 7→ (g(e1 ∧ e2),g(e2 ∧ e3),g(e3 ∧ e1)). 0 0 0 For an element x = 0 0 1 ∈ O[21], the generalized Springer fiber P[21](x) above x is isomorphic to the 0 0 0 1 projective line P , and is given by the equation X1 = 0 in the Plücker coordinate [X0 : X1 : X2]. The 0 1 0 generalized Springer fiber above an element 0 0 1 is the point [1 : 0 : 0]. 0 0 0 For the case α = [11] and β = [1], the two possible choices of γ are [21] and [111] with Littlewood- Richardson coefficients c[11],[1][21] = c[11],[1][111] = 1. The orbit O[111] is the zero orbit, hence for all parabolic subalgebras q = m + u conjugate to p[21], the variety pm O[111] ∩ q contains the zero orbit O[11] × O[1]. The orbit O[21] can be explicitly described with matrices:

2 O[21] = A ∈ g | A = 0 and A 6=0 . A v We can represent any element in p as a matrix of the shape where v ∈ C2. The subvariety [21] 0 b O[21] ∩ p[21] is the following subset A v O ∩ p = 6=0 | A2 =0, Av =0 [21] [21] 0 0 −1 of p[21]. The elements [g] in the generalized Springer ﬁber P[21](x) are those who satisfy Ad(g) x ∈ O[21] ∩ p[21]. The point [0 : 0 : 1] ∈ P[21](x) projects to the zero orbit O[11] and the aﬃne chart [1 : 0 : x] projects to O[2].

The orbit O[3] is the set of matrices satisfying

3 2 O[3] = A ∈ g | A =0, A 6= 0 and A 6=0 .

Therefore, the subvariety O[3] ∩ p[21] is the following set of matrices: A v O ∩ p = 6=0 | A3 =0, A2v =0, A2 6=0 or Ab 6=0 . [3] [21] 0 0 The generalized Springer ﬁber over a point of O[3] is the point [1 : 0 : 0], and the whole set O[3] ∩ p[21] projects to the orbit O[2].

5.3.2. n =6. Consider the Levi blocks GL(4)×GL(2) ⊂ GL(6), and consider the orbit O[22] ×O[11]. Between the induced orbit [33] and [2211], the poset of G-orbits contained in G · (O[22],[11] + n) is displayed as below: [33]

[321]

[3111] [222]

[2211]

3111 222 The orbits [3111] and [222] are excluded from the poset, since c33,2211 = c33,2211 = 0. It is important to mention that the n = 6 case for P[33] is the first case in which we can expect a Littlewood- Richardson coefficient greater than 1. The partial flag variety G/P[33] is isomorphic to the Grassmannian G(3, 6) and can be embedded into the projective space P( 3 C6) =∼ P19 with the Plücker embeddings by choosing the basis of 3 C6 as e ∧ e ∧ e for each triple i

L σ : Pα,β(x, y) · P → Spec(C[p] ) which takes every g to Ad(g)−1x, and the ring C[p]L is the ring of AdL-invariant regular functions on p under the adjoint action of L. Any element of Spec(C[p]L) lying in the image of σ can be represented by a matrix

0 1 0 u1,4 u1,5 u1,6 0 0 0 u2,4 u2,5 u2,6  0 0 0 u u u  3,4 3,5 3,6 .  0000 1 0     0000 0 0     0000 0 0     

The images of the map σ for diﬀerent choices of γ are listed in the following table:

γ γ dim Pα,β(x) c[21],[21] Ideal [2211] 5 1 (u3,6,u2,6,u2,4,u3,4,u1,4 + u2,5) [222] 4 1 (u1,4 + u3,5,u2,4,u3,4,u2,6) [3111] 4 1 (u3,6,u2,6,u2,4,u3,4) . [3, 2, 1] 2 2 (u3,4,u2,4), (u2,6,u2,4) [3, 3] 1 1 (u2,4) [4, 1, 1] 1 1 (u2,6u3,4 − u2,4u3,6) [4, 2] 0 1 (0)

5.3.3. n = 8. We look at the Levi blocks GL(4) × GL(4) ⊂ GL(8), and consider the orbit O[22] × O[22]. Between the induced orbit [44] and [2222], the poset of G-orbit G · (O[22],[22] + n) is displayed as below:

[44]

[431]

[422]

[4211] [332]

[3311]

[3221]

[2222]

4211 332 The orbits [4211] and [332] are excluded from the poset, since c22,22 = c22,22 = 0. 15 5.3.4. n = 12. Consider the Levi blocks GL(6)×GL(6) ⊂ GL(12) and the orbit O[321]×O[321], if we represent any element of Spec(C[p]L) by the matrix

010000 u1,7 u1,8 u1,9 u1,10 u1,11 u1,12 001000 u u u u u u  2,7 2,8 2,9 2,10 2,11 2,12  000000 u3,7 u3,8 u3,9 u3,10 u3,11 u3,12  000010 u u u u u u   4,7 4,8 4,9 4,10 4,11 4,12   000000 u5,7 u5,8 u5,9 u5,10 u5,11 u5,12     000000 u6,7 u6,8 u6,9 u6,10 u6,11 u6,12     0000000 1 0 0 0 0     0000000 0 1 0 0 0     0000000 0 0 0 0 0     0000000 0 0 0 1 0     0000000 0 0 0 0 0     0000000 0 0 0 0 0    If we set γ = [53211], the ideals for the image of the projection map from the 6 dim ensional Satake ﬁber with 4 irreducible components are the following four ideals:

I1 = (u6,10,u6,7,u5,10,u5,7,u3,10,u3,7)

I2 = (u6,7,u5,7,u3,7,

u5,12u6,10 − u5,10u6,12,u3,12u6,10 − u3,10u6,12,u3,12u5,10 − u3,10u5,12

u3,12u4,7 + u3,12u5,8 − u2,7u5,12 − u3,8u5,12,

u3,10u4,7 + u3,10u5,8 − u2,7u5,10 − u3,8u5,10)

I3 = (u5,12,u5,10,u5,7,u3,12,u3,10,u3,7)

I4 = (u3,12,u3,10,u3,7,

u5,12u6,10 − u5,10u6,12,u5,12u6,7 − u5,7u6,12,u5,10u6,7 − u5,7u6,10

u2,10u6,7 + u3,11u6,7 − u2,7u6,10 − u3,8u6,10,

u2,10u5,7 + u3,11u5,7 − u2,7u5,10 − u3,8u5,10).

[53211] The Littlewood-Richardson coeﬃcient c[321],[321] of this case is equal to 4. References

[BL06] Joseph Bernstein and Valery Lunts. Equivariant sheaves and functors. Springer, 2006. [BM] W Borho and R MacPherson. Partial resolutions of nilpotent varieties, analysis and topology on singular spaces, ii, iii (luminy, 1981), 23–74. Astérisque, 101:102. [BR18] Pierre Baumann and Simon Riche. Notes on the geometric satake equivalence. In Relative aspects in representation theory, Langlands functoriality and automorphic forms, pages 1–134. Springer, 2018. [CM93] David H. Collingwood and William M. McGovern. Nilpotent orbits in semisimple Lie algebra: an introduction. CRC Press, 1993. [dCM09] Mark de Cataldo and Luca Migliorini. The decomposition theorem, perverse sheaves and the topology of algebraic maps. Bulletin of the American Mathematical Society, 46(4):535–633, 2009. [Gin95] Victor Ginzburg. Perverse sheaves on a loop group and Langlands’ duality. arXiv preprint alg-geom/9511007, 1995. [GS20] Dmitry Gourevitch and Eitan Sayag. Annihilator varieties of distinguished modules of reductive Lie algebras. arXiv preprint arXiv:2001.11746, 2020. [Hai03] Thomas J. Haines. Structure constants for Hecke and representation rings. International Mathematics Research No- tices, 2003(39):2103–2119, 2003. [Hai06] Thomas J. Haines. Equidimensionality of convolution morphisms and applications to saturation problems. Advances in Mathematics, 207(1):297–327, 2006. [Jos85] Anthony Joseph. On the associated variety of a primitive ideal. Journal of algebra, 93(2):509–523, 1985. [Kle69] Theresa Klein. The Hall polynomial. Journal of Algebra, 12(1):61–78, 1969. [Mac98] Ian Grant Macdonald. Symmetric functions and Hall polynomials. Oxford university press, 1998. [MV07] Ivan Mirkovićand Kari Vilonen. Geometric Langlands duality and representations of algebraic groups over commutative rings. Annals of mathematics, pages 95–143, 2007. [Spa76] Nicolas Spaltenstein. The fixed point set of a unipotent transformation on the flag manifold. In Indagationes Mathe- maticae (Proceedings), volume 79, pages 452–456. North-Holland, 1976. 16 [Spr78] Tonny Albert Springer. A construction of representations of Weyl groups. Inventiones mathematicae, 44:279–293, 1978. [Ste76] Robert Steinberg. On the desingularization of the unipotent variety. Inventiones mathematicae, 36(1):209–224, 1976. [Ste88] Robert Steinberg. An occurrence of the Robinson-Schensted correspondence. Journal of Algebra, 113(2):523–528, 1988. [vL00] Marc A.A. van Leeuwen. Flag varieties and interpretations of Young tableau algorithms. Journal of Algebra, 224(2):397–426, 2000. [Zhu16] Xinwen Zhu. An introduction to affine Grassmannians and the geometric Satake equivalence. arXiv preprint arXiv:1603.05593, 2016.