Introduction
Ringel [27 ] realized the negative part U v − U_v^- U v − of a quantum group using Hall algebras of Dynkin quivers, and Green [14 ] extended this realization to arbitrary finite acyclic quivers by identifying U v − U_v^- U v − with the corresponding generic twisted composition algebra. For quivers of Dynkin type, the orbits in the representation varieties correspond to elements of a Poincaré–Birkhoff–Witt (PBW) basis. Lusztig [18 , 19 ] constructed the canonical basis by algebraic and geometric methods. In the geometric construction, its elements are represented by simple perverse sheaves on the representation varieties.
For a connected quiver Q Q Q of tame type, Lusztig [20 ] used the representation theory of Q Q Q to describe the simple perverse sheaves that realize the canonical basis of the corresponding quantum affine algebra. These sheaves have the form IC ( X ( c , m ) , L χ ) \operatorname{IC}(X(c,m),\mathcal{L}_\chi) IC ( X ( c , m ) , L χ ) , with support X ( c , m ) X(c,m) X ( c , m ) , where the local systems L χ \mathcal{L}_\chi L χ are associated with irreducible characters of symmetric groups. Further details, including the description for arbitrary orientations, can be found in [16 ] .
For quantum affine algebras, canonical bases can also be described through algebraic constructions of PBW bases; see [3 , 4 , 25 , 30 ] . Constructions using representations of tame quivers were developed in [17 , 34 ] , with the cyclic-quiver results in [10 ] as an ingredient. In particular, Xiao, Xu, and Zhao [34 ] proved that the bar-invariant bases obtained from their PBW basis coincides with Lusztig’s canonical basis and identified the corresponding indices.
Lusztig [18 ] ’s work and Kato [15 ] ’s work show how a categorical description of PBW bases can explain the positivity and unitriangularity of the transition matrix from the canonical basis to a PBW basis. We seek such a description using constructible complexes on quiver representation varieties and their relation to Lusztig sheaves. In this paper, we establish this relation for connected quivers of tame type by studying restrictions of Lusztig sheaves and their Frobenius weights.
We describe the results more precisely. Let E ν E_\nu E ν be the representation variety of Q Q Q over F q \mathbb{F}_q F q , and let δ \delta δ be the minimal positive imaginary root. Write c = c P ⊕ c n h ⊕ c I c=c_P\oplus c_{nh}\oplus c_I c = c P ⊕ c nh ⊕ c I for a representation with 2000 Mathematics Subject Classification. 16G20, 17B37.
Key words and phrases. tame quivers, Lusztig sheaf complexes, pointwise purity, canonical basis, PBW basis. no homogeneous regular summand. For dim c + m δ = ν \dim c + m\delta= \nu dim c + m δ = ν , let Y ( c , m ) Y(c,m) Y ( c , m ) consist of the representations c ⊕ R c\oplus R c ⊕ R with R R R homogeneous regular of dimension m δ m\delta m δ . These subsets form a finite stratification of E ν E_\nu E ν . Inside Y ( c , m ) Y(c,m) Y ( c , m ) , let X ( c , m ) X(c,m) X ( c , m ) be the open dense locus where R R R is a sum of m m m pairwise non-isomorphic homogeneous regular simples. Ordering these simples defines a symmetric-group covering and local systems L χ \mathcal{L}_\chi L χ , indexed by irreducible characters of the symmetric group S m S_m S m .
Our first result controls the constituents of restricted flag sheaves. The restriction of any flag sheaf complex to Y ( c , m ) Y(c,m) Y ( c , m ) belongs to the triangulated category generated by the restrictions of IC ( X ( c , m ) , L χ ) \operatorname{IC}(X(c,m),\mathcal{L}_\chi) IC ( X ( c , m ) , L χ ) , together with constant rank-one Weil factors (Theorem 3.3 ). Our main purity theorem states that every normalized flag sheaf complex is pointwise pure of weight zero on X ( c , m ) X(c,m) X ( c , m ) (Theorem 4.1). Combined with the description of constituents, this implies that its restriction to Y ( c , m ) Y(c,m) Y ( c , m ) is pure and geometrically semisimple, with summands given by shifts and twists of these restricted intersection cohomology complexes (Proposition 4.2).
These results determine the transition coefficients geometrically. For a canonical-basis complex IC β \operatorname{IC}_\beta IC β and j c , m : Y ( c , m ) ↪ E ν j^{c,m}:Y(c,m)\hookrightarrow E_\nu j c , m : Y ( c , m ) ↪ E ν , write its geometric restriction as
( j c , m ) ∗ IC β ≃ ⨁ χ , r ( ( j c , m ) ∗ IC ( X ( c , m ) , L χ ) [ r ] ( r / 2 ) ) ⊕ b c , χ , r β (j^{c,m})^*\operatorname{IC}_\beta\simeq\bigoplus_{\chi,r}\left((j^{c,m})^*\operatorname{IC}(X(c,m),\mathcal{L}_\chi)[r](r/2)\right)^{\oplus b^\beta_{c,\chi,r}} ( j c , m ) ∗ IC β ≃ χ , r ⨁ ( ( j c , m ) ∗ IC ( X ( c , m ) , L χ ) [ r ] ( r /2 ) ) ⊕ b c , χ , r β
The nonnegative integers b c , χ , r β b^\beta_{c,\chi,r} b c , χ , r β are equivalently graded local-system multiplicities on X ( c , m ) X(c,m) X ( c , m ) . Theorem 4.8 identifies
f c , χ β ( v ) = ∑ r b c , χ , r β v − r ∈ N [ v , v − 1 ] f^\beta_{c,\chi}(v)=\sum_r b^\beta_{c,\chi,r}v^{-r}\in\mathbb{N}[v,v^{-1}] f c , χ β ( v ) = r ∑ b c , χ , r β v − r ∈ N [ v , v − 1 ]
with the coefficients in the expansion of the canonical trace function in the stratum-supported family N ( c , t χ ) N(c,t_\chi) N ( c , t χ ) , under the convention v = − q v=-\sqrt{q} v = − q . Using the relation between this family and the PBW basis E ( c , t χ ) E(c,t_\chi) E ( c , t χ ) from [34 Lemma 9.13] , the coefficients with aperiodic indices are precisely the canonical-to-PBW transition coefficients. Thus the restriction formula gives a geometric explanation of positivity and unitriangularity, with respect to the support order, for the representation-theoretic PBW basis.
In his lecture notes, Schiffmann [29 §3.7] suggests that Lusztig’s result on stalkwise Frobenius eigenvalues in finite type should extend to tame quivers. Our Theorem 4.1 gives a partial answer concerning their absolute values: for a normalized flag sheaf complex, every Frobenius eigenvalue λ \lambda λ on its degree-i i i stalk cohomology at an F q \mathbb{F}_q F q -rational point of X ( c , m ) X(c,m) X ( c , m ) satisfies ∣ λ ∣ = q i / 2 |\lambda|=q^{i/2} ∣ λ ∣ = q i /2 . Thus we establish pointwise purity on these loci; we do not assert that the eigenvalues themselves are all equal to q i / 2 q^{i/2} q i /2 .
We note that, independently of the present work, Xu and Zhao [35 ] have established pointwise purity for Lusztig sheaves of tame quivers using a different approach.
Analogous positivity and unitriangularity results for the PBW bases constructed using Khovanov–Lauda–Rouquier (KLR) algebras were obtained in finite ADE type [7 , 15 ] and symmetric affine type [24 ] .
The main remaining step is to prove purity in the presence of non-homogeneous regular tubes. The proof uses contracting transverse slices and induction over a finite stratification in regular tubes. The required tube filtrations mix prescribed semisimple regular quotients with unrestricted quotients of dimension δ \delta δ ; their pointwise purity is established in Lemma 5.2. The restriction and slice arguments then give the purity theorem for general tame quivers, including the periodic tube representations occurring in the intermediate strata (Theorem 4.1). Related results on purity and affine pavings of flag varieties can be found in [8 , 12 , 28 ] .
Our results extend the geometric comparison for the Kronecker quiver in [23 , 33 ] to all connected quivers of tame type.
In Section 1, we recall the representation theory and the strata X ( c , m ) X(c,m) X ( c , m ) and Y ( c , m ) Y(c,m) Y ( c , m ) . In Section 2, we review Lusztig sheaves and the relevant bases. In Section 3, we prove the restriction theorem. In Section 4, we derive the decomposition and transition formulas from pointwise purity, whose proof occupies Section 5.
Throughout this paper, we follow the notation and conventions of [1 , 21 , 31 ] for quiver representation theory, quantum groups, and algebraic geometry.
Preliminaries on representations of tame quivers
This section reviews basic properties of representations of tame quivers and the stratification of their representation varieties. Throughout, let k = F ‾ q k=\overline{\mathbb{F}}_q k = F q be an algebraic closure of a finite field.
Conventions and the Euler form
All representations are finite-dimensional over k k k . We identify rep k Q \operatorname{rep}_k Q rep k Q with the category of finite-dimensional left k Q kQ k Q -modules, and write ind C \operatorname{ind}\mathcal{C} ind C for the isomorphism classes of indecomposable objects of a category C \mathcal{C} C .
Definition 1.1. In this paper, a tame quiver is a finite connected acyclic quiver whose underlying unoriented graph is an extended Dynkin diagram
A ~ n ( n ≥ 1 ) , D ~ n ( n ≥ 4 ) , E ~ 6 , E ~ 7 , E ~ 8 . \widetilde{A}_n\ (n\ge1),\qquad\widetilde{D}_n\ (n\ge4),\qquad\widetilde{E}_6,\qquad\widetilde{E}_7,\qquad\widetilde{E}_8. A n ( n ≥ 1 ) , D n ( n ≥ 4 ) , E 6 , E 7 , E 8 . Here A ~ 1 \widetilde{A}_1 A 1 denotes the Kronecker quiver, with two vertices and two parallel arrows. Thus tame is used in the representation-infinite sense; Dynkin quivers are excluded. This is the Euclidean case in the classification of hereditary algebras; see [31 Chapter XIII] .
Write Q = ( Q 0 , H , s , t ) Q=(Q_0,H,s,t) Q = ( Q 0 , H , s , t ) , where Q 0 Q_0 Q 0 is the vertex set, H H H is the arrow set, and s , t : H → Q 0 s,t:H\to Q_0 s , t : H → Q 0 are the source and target maps. For a representation M = ( ( M i ) i ∈ Q 0 , ( M h ) h ∈ H ) M=((M_i)_{i\in Q_0},(M_h)_{h\in H}) M = (( M i ) i ∈ Q 0 , ( M h ) h ∈ H ) , its dimension vector is
dim M = ( dim k M i ) i ∈ Q 0 ∈ N Q 0 . \dim M=(\dim_k M_i)_{i\in Q_0}\in\mathbb{N}^{Q_0}. dim M = ( dim k M i ) i ∈ Q 0 ∈ N Q 0 .
For an Q 0 Q_0 Q 0 -graded vector space V = ⨁ i ∈ Q 0 V i V=\bigoplus_{i\in Q_0}V_i V = ⨁ i ∈ Q 0 V i , we similarly write dim V = ( dim k V i ) i ∈ Q 0 \dim V=(\dim_k V_i)_{i\in Q_0} dim V = ( dim k V i ) i ∈ Q 0 . For α , β ∈ Z Q 0 \alpha,\beta\in\mathbb{Z}^{Q_0} α , β ∈ Z Q 0 , set
⟨ α , β ⟩ = ∑ i ∈ Q 0 α i β i − ∑ h ∈ H α s ( h ) β t ( h ) . \langle\alpha,\beta\rangle=\sum_{i\in Q_0}\alpha_i\beta_i-\sum_{h\in H}\alpha_{s(h)}\beta_{t(h)}. ⟨ α , β ⟩ = i ∈ Q 0 ∑ α i β i − h ∈ H ∑ α s ( h ) β t ( h ) .
Since k Q kQ k Q is hereditary, Ext k Q j ( M , N ) = 0 \operatorname{Ext}^j_{kQ}(M,N)=0 Ext k Q j ( M , N ) = 0 for M , N ∈ rep k Q M,N\in\operatorname{rep}_k Q M , N ∈ rep k Q and j ≥ 2 j\ge2 j ≥ 2 . Moreover,
⟨ dim M , dim N ⟩ = dim k Hom k Q ( M , N ) − dim k Ext k Q 1 ( M , N ) . (1) \langle\dim M,\dim N\rangle=\dim_k\operatorname{Hom}_{kQ}(M,N)-\dim_k\operatorname{Ext}^1_{kQ}(M,N). \tag*{(1)} ⟨ dim M , dim N ⟩ = dim k Hom k Q ( M , N ) − dim k Ext k Q 1 ( M , N ) . (1)
See [17 §1.2] . Let δ \delta δ be the primitive positive imaginary root. We use the defect convention ∂ ( M ) = ⟨ δ , dim M ⟩ \partial(M)=\langle\delta,\dim M\rangle ∂ ( M ) = ⟨ δ , dim M ⟩ . Unless indicated otherwise, Hom \operatorname{Hom} Hom , Ext 1 \operatorname{Ext}^1 Ext 1 , and End \operatorname{End} End are taken over k Q kQ k Q .
Representation spaces and base change
Fix a Q 0 Q_0 Q 0 -graded vector space V = ⨁ i ∈ Q 0 V i V=\bigoplus_{i\in Q_0}V_i V = ⨁ i ∈ Q 0 V i with dim V = ν \dim V=\nu dim V = ν . Define the affine representation space and the base-change group by
E ν = E V = ⨁ h ∈ H Hom k ( V s ( h ) , V t ( h ) ) , G ν = ∏ i ∈ Q 0 GL ( V i ) . (2) E_\nu=E_V=\bigoplus_{h\in H}\operatorname{Hom}_k(V_{s(h)},V_{t(h)}),\qquad G_\nu=\prod_{i\in Q_0}\operatorname{GL}(V_i). \tag*{(2)} E ν = E V = h ∈ H ⨁ Hom k ( V s ( h ) , V t ( h ) ) , G ν = i ∈ Q 0 ∏ GL ( V i ) . (2)
For g = ( g i ) i ∈ Q 0 ∈ G ν g=(g_i)_{i\in Q_0}\in G_\nu g = ( g i ) i ∈ Q 0 ∈ G ν and x = ( x h ) h ∈ H ∈ E ν x=(x_h)_{h\in H}\in E_\nu x = ( x h ) h ∈ H ∈ E ν , the action is
( g ⋅ x ) h = g t ( h ) x h g s ( h ) − 1 ( h ∈ H ) . (3) (g\cdot x)_h=g_{t(h)}x_hg_{s(h)}^{-1}\qquad(h\in H). \tag*{(3)} ( g ⋅ x ) h = g t ( h ) x h g s ( h ) − 1 ( h ∈ H ) . (3)
The point x x x determines the representation M x = ( ( V i ) i ∈ Q 0 , ( x h ) h ∈ H ) M_x=((V_i)_{i\in Q_0},(x_h)_{h\in H}) M x = (( V i ) i ∈ Q 0 , ( x h ) h ∈ H ) . Two points lie in the G V G_V G V -orbit if and only if their representations are isomorphic. The stabilizer of x x x is Aut k Q ( M x ) \operatorname{Aut}_{kQ}(M_x) Aut k Q ( M x ) , and hence
dim ( G V ⋅ x ) = dim G V − dim k End ( M x ) , dim G V − dim E ν = ⟨ ν , ν ⟩ . (4) \dim(G_V\cdot x)=\dim G_V-\dim_k\operatorname{End}(M_x),\qquad\dim G_V-\dim E_\nu=\langle\nu,\nu\rangle. \tag*{(4)} dim ( G V ⋅ x ) = dim G V − dim k End ( M x ) , dim G V − dim E ν = ⟨ ν , ν ⟩ . (4)
See [17 §2.1] and [34 §7.3] . All closures and density statements below refer to the Zariski topology.
Indecomposables and tubes
Let τ \tau τ be the Auslander–Reiten translation. An indecomposable representation M M M is preprojective if M ≃ τ − r P ( i ) M \simeq\tau^{-r}P(i) M ≃ τ − r P ( i ) , and preinjective if M ≃ τ r I ( i ) M \simeq\tau^{r}I(i) M ≃ τ r I ( i ) , for some r ≥ 0 r \ge0 r ≥ 0 and i ∈ Q 0 i \in Q_0 i ∈ Q 0 ; here P ( i ) P(i) P ( i ) and I ( i ) I(i) I ( i ) are the indecomposable projective and injective modules associated with i i i . The remaining indecomposables are regular . Let P , R , I \mathbf{P},\mathbf{R},\mathbf{I} P , R , I denote their additive subcategories. For an indecomposable representation M M M , this classification can also be expressed as
M ∈ P ⟺ ∂ ( M ) < 0 , M ∈ R ⟺ ∂ ( M ) = 0 , M ∈ I ⟺ ∂ ( M ) > 0. (5) M \in\mathbf{P} \Longleftrightarrow\partial(M) < 0,\qquad M \in\mathbf{R} \Longleftrightarrow\partial(M) = 0,\qquad M \in\mathbf{I} \Longleftrightarrow\partial(M) > 0. \tag*{(5)} M ∈ P ⟺ ∂ ( M ) < 0 , M ∈ R ⟺ ∂ ( M ) = 0 , M ∈ I ⟺ ∂ ( M ) > 0. (5)
These assertions are recalled in [34 §5.1, Lemma 5.1] .
The regular category decomposes into tubes T z \mathcal{T}_z T z , indexed by z ∈ P 1 ( k ) z \in\mathbb{P}^1(k) z ∈ P 1 ( k ) after a choice of parametrization. Each T z \mathcal{T}_z T z is an extension-closed abelian subcategory and admits an exact equivalence
T z ≃ nilrep k C r z , \mathcal{T}_z \simeq\operatorname{nilrep}_k C_{r_z}, T z ≃ nilrep k C r z ,
where C r C_r C r is the cyclic quiver with vertices Z / r Z \mathbb{Z}/r\mathbb{Z} Z / r Z and arrows i ⟶ i + 1 i \longrightarrow i+1 i ⟶ i + 1 . The integer r z r_z r z is the rank of the tube. The tubes of rank one are homogeneous ; the others are non-homogeneous . There are finitely many non-homogeneous tubes, at parameters z 1 , … , z e z_1,\ldots,z_e z 1 , … , z e ; put T j = T z j \mathcal{T}_j=\mathcal{T}_{z_j} T j = T z j and r j = r z j > 1 r_j=r_{z_j}>1 r j = r z j > 1 . Thus e e e denotes the number of non-homogeneous tubes. We write
Z h = { z ∈ P 1 ( k ) : r z = 1 } , R n h = ⨁ j = 1 e T j , R h = ⨁ z ∈ Z h T z . Z_h=\{z\in\mathbb{P}^1(k):r_z=1\},\qquad\mathbf{R}_{nh}=\bigoplus_{j=1}^{e}\mathcal{T}_j,\qquad\mathbf{R}_h=\bigoplus_{z\in Z_h}\mathcal{T}_z. Z h = { z ∈ P 1 ( k ) : r z = 1 } , R nh = j = 1 ⨁ e T j , R h = z ∈ Z h ⨁ T z .
Only finite direct sums of objects are meant here. The ranks are as follows; entries equal to one are omitted:
Table 1.
In the first row, a , b > 0 a,b>0 a , b > 0 count the arrows in the two directions around the underlying cycle; the Kronecker case has a = b = 1 a=b=1 a = b = 1 . For the tube classification and ranks, see [11 ] and [31 Chapter XIII] ; for the cyclic-quiver equivalences, see [34 §5.3] .
Fix such an equivalence for each tube. Denote the images of the vertex simples by S z , i S_{z,i} S z , i . Let S z , i [ l ] S_{z,i}[l] S z , i [ l ] be the indecomposable of regular length l ≥ 1 l\ge1 l ≥ 1 and regular top S z , i S_{z,i} S z , i , with i ∈ Z / r z Z i\in\mathbb{Z}/r_z\mathbb{Z} i ∈ Z / r z Z . These are all the indecomposable objects of T z \mathcal{T}_z T z . For z ∈ Z h z\in Z_h z ∈ Z h write R z [ l ] = S z , 0 [ l ] R_z[l]=S_{z,0}[l] R z [ l ] = S z , 0 [ l ] ; then dim R z [ l ] = l δ \dim R_z[l]=l\delta dim R z [ l ] = l δ . Consequently, every representation has a decomposition, unique up to isomorphism of its summands,
M ≃ M P ⊕ M n h ⊕ M h r ⊕ M I . (6) M \simeq M_P \oplus M_{nh} \oplus M_{hr} \oplus M_I. \tag*{(6)} M ≃ M P ⊕ M nh ⊕ M h r ⊕ M I . (6)
Here, M P M_P M P , M n h M_{nh} M nh , M h r M_{hr} M h r , and M I M_I M I denote the preprojective, nonhomogeneous regular, homogeneous regular, and preinjective direct summands of M M M , respectively.
Hom and Ext
For P ∈ P P\in\mathbf{P} P ∈ P , R ∈ R R\in\mathbf{R} R ∈ R , and I ∈ I I\in\mathbf{I} I ∈ I ,
Hom ( R , P ) = Hom ( I , P ) = Hom ( I , R ) = 0 , (7) \operatorname{Hom}(R,P)=\operatorname{Hom}(I,P)=\operatorname{Hom}(I,R)=0, \tag*{(7)} Hom ( R , P ) = Hom ( I , P ) = Hom ( I , R ) = 0 , (7)
Ext 1 ( P , R ) = Ext 1 ( P , I ) = Ext 1 ( R , I ) = 0. (8) \operatorname{Ext}^{1}(P,R)=\operatorname{Ext}^{1}(P,I)=\operatorname{Ext}^{1}(R,I)=0. \tag*{(8)} Ext 1 ( P , R ) = Ext 1 ( P , I ) = Ext 1 ( R , I ) = 0. (8)
See the separation theory in [31 Chapters XI–XIII] . If X ∈ T z X\in\mathcal{T}_z X ∈ T z and Y ∈ T w Y\in\mathcal{T}_w Y ∈ T w with z ≠ w z\ne w z = w , then
Hom ( X , Y ) = Hom ( Y , X ) = Ext 1 ( X , Y ) = Ext 1 ( Y , X ) = 0. (9) \operatorname{Hom}(X,Y)=\operatorname{Hom}(Y,X)=\operatorname{Ext}^{1}(X,Y)=\operatorname{Ext}^{1}(Y,X)=0. \tag*{(9)} Hom ( X , Y ) = Hom ( Y , X ) = Ext 1 ( X , Y ) = Ext 1 ( Y , X ) = 0. (9)
Indeed, the tubes have disjoint sets of regular composition factors, and τ \tau τ preserves each tube. Apply the hereditary Auslander–Reiten formula
D Ext 1 ( M , N ) ≃ Hom ( N , τ M ) ≃ Hom ( τ − 1 N , M ) , D = Hom k ( − , k ) . (10) D\operatorname{Ext}^{1}(M,N)\simeq\operatorname{Hom}(N,\tau M)\simeq\operatorname{Hom}(\tau^{-1}N,M),\qquad D=\operatorname{Hom}_k(-,k). \tag*{(10)} D Ext 1 ( M , N ) ≃ Hom ( N , τ M ) ≃ Hom ( τ − 1 N , M ) , D = Hom k ( − , k ) . (10)
with τ P = 0 \tau P=0 τ P = 0 for projective P P P and τ − 1 I = 0 \tau^{-1}I=0 τ − 1 I = 0 for injective I I I ; see [2 Chapter IV] . Every preprojective or preinjective indecomposable X X X satisfies End ( X ) ≃ k \operatorname{End}(X)\simeq k End ( X ) ≃ k and Ext 1 ( X , X ) = 0 \operatorname{Ext}^{1}(X,X)=0 Ext 1 ( X , X ) = 0 .
An order compatible with the vanishing properties
Put n = ∣ Q 0 ∣ n=|Q_{0}| n = ∣ Q 0 ∣ and number the vertices 1 , … , n 1,\ldots,n 1 , … , n so that every arrow goes from a vertex with a smaller label to one with a larger label. We use the periodic reflection-functor enumerations
P r n + j − 1 = τ − r P ( n + 1 − j ) , I r n + j = τ r I ( j ) ( r ≥ 0 , 1 ≤ j ≤ n ) . (11) P_{rn+j-1}=\tau^{-r}P(n+1-j),\qquad I_{rn+j}=\tau^{r}I(j)\quad(r\geq0,\ 1\leq j\leq n). \tag*{(11)} P r n + j − 1 = τ − r P ( n + 1 − j ) , I r n + j = τ r I ( j ) ( r ≥ 0 , 1 ≤ j ≤ n ) . (11)
Thus the standard orders on the two components are
P 0 < P 1 < P 2 < ⋯ , ⋯ < I 3 < I 2 < I 1 . P_{0}<P_{1}<P_{2}<\cdots,\qquad\cdots<I_{3}<I_{2}<I_{1}. P 0 < P 1 < P 2 < ⋯ , ⋯ < I 3 < I 2 < I 1 .
They are the orders associated with the periodic admissible sequence in [34 §5.2] . Their Hom and Ext properties are also recorded in [17 §4.3, p. 844] .
Definition 1.2. Choose a total order ⊲ \lhd ⊲ on the tube index set. Define a strict partial order < < < on ind ( rep k Q ) \operatorname{ind}(\operatorname{rep}_{k}Q) ind ( rep k Q ) by
P a < P b P_{a}<P_{b} P a < P b if a < b a<b a < b , and I b < I a I_{b}<I_{a} I b < I a if a < b a<b a < b ;
P < R < I P<R<I P < R < I for every indecomposable P ∈ P P\in\mathcal{P} P ∈ P , R ∈ R R\in\mathcal{R} R ∈ R , and I ∈ I I\in\mathcal{I} I ∈ I , and P < I P<I P < I ;
X < Y X<Y X < Y for X ∈ ind T z X\in\operatorname{ind}\mathcal{T}_{z} X ∈ ind T z and Y ∈ ind T w Y\in\operatorname{ind}\mathcal{T}_{w} Y ∈ ind T w whenever z ⊲ w z\lhd w z ⊲ w .
Distinct indecomposables in a single tube are left incomparable.
For any dimension vector ν \nu ν , there exist an indecomposable preprojective module P P P and an indecomposable preinjective module I I I , with the following property. For every P ′ > P P'>P P ′ > P , where P ′ P' P ′ is indecomposable preprojective, and every I ′ < I I'<I I ′ < I , where I ′ I' I ′ is indecomposable preinjective, we have dim P ′ > ν \dim P'>\nu dim P ′ > ν and dim I ′ > ν \dim I'>\nu dim I ′ > ν .
Proposition 1.3. For indecomposables M , N M,N M , N , this order satisfies
M < N ⟹ Hom ( N , M ) = 0 = Ext 1 ( M , N ) . (12) M<N\ \Longrightarrow\ \operatorname{Hom}(N,M)=0=\operatorname{Ext}^{1}(M,N). \tag*{(12)} M < N ⟹ Hom ( N , M ) = 0 = Ext 1 ( M , N ) . (12)
Proof. For two preprojectives or two preinjectives, use the reflection-functor orders in (11) and [17 §4.3] . The remaining comparisons follow from (7)–(9). Transitivity follows from the displayed component orders and the order ⊲ \lhd ⊲ on the tubes. □
Lemma 1.4. For a regular representation M M M , if N N N is a sub-representation of M M M and R R R is a quotient representation of M M M , with ∂ ( N ) = 0 \partial(N)=0 ∂ ( N ) = 0 and ∂ ( R ) = 0 \partial(R)=0 ∂ ( R ) = 0 , then N N N and R R R are regular modules.
Proof. Since N N N is a sub-representation of a regular representation, applying Proposition 1.3, N N N has no preinjective summands. Similarly, R R R has no direct preprojective summand. Since ∂ ( N ) = ∂ ( R ) = 0 \partial(N)=\partial(R)=0 ∂ ( N ) = ∂ ( R ) = 0 , they are both regular. □
Multisegments and indexing data
For r ≥ 1 r\geq1 r ≥ 1 , let Π ( r ) \Pi(r) Π ( r ) consist of the finite multisegments
π = ∑ i ∈ Z / r Z l ≥ 1 m i , l [ i ; l ) , m i , l ∈ N , \pi=\sum_{\substack{i\in\mathbb{Z}/r\mathbb{Z}\\ l\geq1}}m_{i,l}[i;l),\qquad m_{i,l}\in\mathbb{N}, π = i ∈ Z / r Z l ≥ 1 ∑ m i , l [ i ; l ) , m i , l ∈ N ,
where only finitely many coefficients are nonzero. For a non-homogeneous tube, abbreviate S j , i [ l ] = S z j , i [ l ] S_{j,i}[l]=S_{z_{j},i}[l] S j , i [ l ] = S z j , i [ l ] . Under the chosen equivalence for T j \mathcal{T}_{j} T j , set
M j ( π ) = ⨁ i ∈ Z / r j Z l ≥ 1 S j , i [ l ] ⊕ m i , l . M_{j}(\pi)=\bigoplus_{\substack{i\in\mathbb{Z}/r_{j}\mathbb{Z}\\ l\geq1}}S_{j,i}[l]^{\oplus m_{i,l}}. M j ( π ) = i ∈ Z / r j Z l ≥ 1 ⨁ S j , i [ l ] ⊕ m i , l .
The multisegment is aperiodic precisely when
for each l ≥ 1 there exists i ∈ Z / r Z with m i , l = 0. (13) \text{for each }l\geq1\text{ there exists }i\in\mathbb{Z}/r\mathbb{Z}\text{ with }m_{i,l}=0. \tag*{(13)} for each l ≥ 1 there exists i ∈ Z / r Z with m i , l = 0. (13)
This is the convention of [17 §§3.1–3.2] .
Let N ( J ) \mathbb{N}^{(J)} N ( J ) denote the finitely supported functions J → N J \to\mathbb{N} J → N , and put
G − = N ( Z ≤ 0 ) , G + = N ( Z > 0 ) , G 0 = ∏ j = 1 e Π ( r j ) , G ′ = G − × G 0 × G + . \mathcal{G}_{-}=\mathbb{N}^{(\mathbb{Z}_{\leq0})},\qquad\mathcal{G}_{+}=\mathbb{N}^{(\mathbb{Z}_{>0})},\qquad\mathcal{G}_{0}=\prod_{j=1}^{e}\Pi(r_j),\qquad\mathcal{G}'=\mathcal{G}_{-}\times\mathcal{G}_{0}\times\mathcal{G}_{+}. G − = N ( Z ≤ 0 ) , G + = N ( Z > 0 ) , G 0 = j = 1 ∏ e Π ( r j ) , G ′ = G − × G 0 × G + .
An empty product is a singleton. For c = ( c − , c 0 , c + ) \mathbf{c}=(\mathbf{c}_{-},\mathbf{c}_{0},\mathbf{c}_{+}) c = ( c − , c 0 , c + ) with c 0 = ( π 1 , … , π e ) \mathbf{c}_{0}=(\pi_1,\ldots,\pi_e) c 0 = ( π 1 , … , π e ) , define
c P = M ( c − ) = ⨁ a ≥ 0 P a ⊕ c − ( − a ) , c I = M ( c + ) = ⨁ b ≥ 1 I b ⊕ c + ( b ) , c_P=M(\mathbf{c}_{-})=\bigoplus_{a\geq0}P_a^{\oplus\mathbf{c}_{-}(-a)},\qquad c_I=M(\mathbf{c}_{+})=\bigoplus_{b\geq1}I_b^{\oplus\mathbf{c}_{+}(b)}, c P = M ( c − ) = a ≥ 0 ⨁ P a ⊕ c − ( − a ) , c I = M ( c + ) = b ≥ 1 ⨁ I b ⊕ c + ( b ) ,
c n h = ⨁ j = 1 e M j ( π j ) , c n h , j = M j ( π j ) . c_{nh}=\bigoplus_{j=1}^{e}M_j(\pi_j),\qquad c_{nh,j}=M_j(\pi_j). c nh = j = 1 ⨁ e M j ( π j ) , c nh , j = M j ( π j ) .
We write c = M ( c ) \mathbf{c}=M(\mathbf{c}) c = M ( c ) for a representative of the isomorphism class of c P ⊕ c I ⊕ c n h c_P\oplus c_I\oplus c_{nh} c P ⊕ c I ⊕ c nh . The set of representatives c \mathbf{c} c , indexed by c ∈ G ′ \mathbf{c}\in\mathcal{G}' c ∈ G ′ , is denoted by G \mathcal{G} G .
The subsets X ( c , m ) X(c,m) X ( c , m ) and Y ( c , m ) Y(c,m) Y ( c , m )
Fix a representation c c c with no homogeneous regular direct summand. In the notation of (6), its decomposition is
c ≃ c P ⊕ c n h ⊕ c I , c n h ≃ ⨁ j = 1 e c n h , j , c n h , j ∈ T j . (14) c\simeq c_P\oplus c_{nh}\oplus c_I,\qquad c_{nh}\simeq\bigoplus_{j=1}^{e}c_{nh,j},\qquad c_{nh,j}\in\mathcal{T}_j. \tag*{(14)} c ≃ c P ⊕ c nh ⊕ c I , c nh ≃ j = 1 ⨁ e c nh , j , c nh , j ∈ T j . (14)
There is a unique c ∈ G ′ \mathbf{c}\in\mathcal{G}' c ∈ G ′ such that c P ≃ M ( c − ) c_P\simeq M(\mathbf{c}_{-}) c P ≃ M ( c − ) , c I ≃ M ( c + ) c_I\simeq M(\mathbf{c}_{+}) c I ≃ M ( c + ) , and c n h , j ≃ M j ( π j ) c_{nh,j}\simeq M_j(\pi_j) c nh , j ≃ M j ( π j ) .
Set G ~ ν = { ( c , m ) ∣ c ∈ G , dim c + m δ = ν } \widetilde{\mathcal{G}}_{\nu}=\{(c,m)\mid c\in\mathcal{G},\ \dim c+m\delta=\nu\} G ν = {( c , m ) ∣ c ∈ G , dim c + m δ = ν } .
Let m ∈ N m\in\mathbb{N} m ∈ N and choose V V V with dim V = dim c + m δ = ν \dim V=\dim c+m\delta=\nu dim V = dim c + m δ = ν . Following [34 §7.3] , define
Y ( c , m ) = { x ∈ E ν : M x ≃ c ⊕ R , R ∈ R h , dim R = m δ } , (15) Y(c,m)=\{x\in E_{\nu}:M_x\simeq c\oplus R,\ R\in\mathcal{R}_h,\ \dim R=m\delta\}, \tag*{(15)} Y ( c , m ) = { x ∈ E ν : M x ≃ c ⊕ R , R ∈ R h , dim R = m δ } , (15)
X ( c , m ) = { x ∈ E ν : M x ≃ c ⊕ R u 1 [ 1 ] ⊕ ⋯ ⊕ R u m [ 1 ] , u 1 , … , u m ∈ Z h pairwise distinct } . (16) X(c,m)=\{x\in E_{\nu}:M_x\simeq c\oplus R_{u_1}[1]\oplus\cdots\oplus R_{u_m}[1],\ u_1,\ldots,u_m\in\mathbb{Z}_h\text{ pairwise distinct}\}. \tag*{(16)} X ( c , m ) = { x ∈ E ν : M x ≃ c ⊕ R u 1 [ 1 ] ⊕ ⋯ ⊕ R u m [ 1 ] , u 1 , … , u m ∈ Z h pairwise distinct } . (16)
Thus X ( c , m ) X(c,m) X ( c , m ) consists of the points whose homogeneous regular part is a direct sum of m m m pairwise non-isomorphic regular simples. Here regular simple means simple in its tube, not necessarily simple as a k Q kQ k Q -module. Each such summand has dimension vector δ \delta δ . Both subsets are G ν G_{\nu} G ν -stable, and X ( c , m ) ⊆ Y ( c , m ) X(c,m)\subseteq Y(c,m) X ( c , m ) ⊆ Y ( c , m ) . For m = 0 m=0 m = 0 , the empty direct sum is understood, and both sets are the orbit of c c c .
The Krull–Schmidt theorem gives the finite disjoint decomposition
E ν = ⨆ ( c , m ) ∈ G ~ ν Y ( c , m ) ; (17) E_{\nu}=\bigsqcup_{(c,m)\in\widetilde{\mathcal{G}}_{\nu}}Y(c,m); \tag*{(17)} E ν = ( c , m ) ∈ G ν ⨆ Y ( c , m ) ; (17)
see [34 §7.3] .
Proposition 1.5. The subset X ( c , m ) X(c,m) X ( c , m ) is a smooth variety, open and dense in Y ( c , m ) Y(c,m) Y ( c , m ) . In particular,
X ( c , m ) ‾ = Y ( c , m ) ‾ . (18) \overline{X(c,m)}=\overline{Y(c,m)}. \tag*{(18)} X ( c , m ) = Y ( c , m ) . (18) Proof. By [20 6.10] , X ( 0 , m ) X(0,m) X ( 0 , m ) is open and dense in E m δ E_{m\delta} E m δ .
X ( c , m ) = { ( x , F ) | x ∈ E V , F = ( 0 = V 0 ⊂ V 1 ⊂ V 2 ⊂ V 3 ⊂ V 4 = V ) , x ∣ V 1 / V 0 ∈ Y ( c I , 0 ) , x ∣ V 2 / V 1 ∈ Y ( c n h , 0 ) , x ∣ V 3 / V 2 ∈ X ( 0 , m ) , x ∣ V 4 / V 3 ∈ Y ( c P , 0 ) } X(c,m)=\left\{(x,F)\ \middle|\ \begin{aligned}
&x\in E_V,\ F=(0=V_0\subset V_1\subset V_2\subset V_3\subset V_4=V),\\
&x|_{V_1/V_0}\in Y(c_I,0),\ x|_{V_2/V_1}\in Y(c_{nh},0),\\
&x|_{V_3/V_2}\in X(0,m),\ x|_{V_4/V_3}\in Y(c_P,0)
\end{aligned}\right\} X ( c , m ) = ⎩ ⎨ ⎧ ( x , F ) x ∈ E V , F = ( 0 = V 0 ⊂ V 1 ⊂ V 2 ⊂ V 3 ⊂ V 4 = V ) , x ∣ V 1 / V 0 ∈ Y ( c I , 0 ) , x ∣ V 2 / V 1 ∈ Y ( c nh , 0 ) , x ∣ V 3 / V 2 ∈ X ( 0 , m ) , x ∣ V 4 / V 3 ∈ Y ( c P , 0 ) ⎭ ⎬ ⎫ and
Y ( c , m ) = { ( x , F ) | x ∈ E V , F = ( 0 = V 0 ⊂ V 1 ⊂ V 2 ⊂ V 3 ⊂ V 4 = V ) , x ∣ V 1 / V 0 ∈ Y ( c I , 0 ) , x ∣ V 2 / V 1 ∈ Y ( c n h , 0 ) , x ∣ V 3 / V 2 ∈ Y ( 0 , m ) , x ∣ V 4 / V 3 ∈ Y ( c P , 0 ) } . Y(c,m)=\left\{(x,F)\ \middle|\ \begin{aligned}
&x\in E_V,\ F=(0=V_0\subset V_1\subset V_2\subset V_3\subset V_4=V),\\
&x|_{V_1/V_0}\in Y(c_I,0),\ x|_{V_2/V_1}\in Y(c_{nh},0),\\
&x|_{V_3/V_2}\in Y(0,m),\ x|_{V_4/V_3}\in Y(c_P,0)
\end{aligned}\right\}. Y ( c , m ) = ⎩ ⎨ ⎧ ( x , F ) x ∈ E V , F = ( 0 = V 0 ⊂ V 1 ⊂ V 2 ⊂ V 3 ⊂ V 4 = V ) , x ∣ V 1 / V 0 ∈ Y ( c I , 0 ) , x ∣ V 2 / V 1 ∈ Y ( c nh , 0 ) , x ∣ V 3 / V 2 ∈ Y ( 0 , m ) , x ∣ V 4 / V 3 ∈ Y ( c P , 0 ) ⎭ ⎬ ⎫ .
Thus Y ( c , m ) ‾ ⊂ X ( c , m ) ‾ \overline{Y(c,m)} \subset\overline{X(c,m)} Y ( c , m ) ⊂ X ( c , m ) . And X ( c , m ) X(c,m) X ( c , m ) is open in Y ( c , m ) ‾ \overline{Y(c,m)} Y ( c , m ) . Fix a corresponding flag 0 ⊂ V 1 ⊂ V 2 ⊂ V 3 ⊂ V 0 \subset V_1 \subset V_2 \subset V_3 \subset V 0 ⊂ V 1 ⊂ V 2 ⊂ V 3 ⊂ V and denote the stabilizer of it as P ⊂ G ν P \subset G_\nu P ⊂ G ν . Let
X ′ = { x ∈ E V ∣ x ∣ V 1 / V 0 ∈ Y ( c I , 0 ) , x ∣ V 2 / V 1 ∈ Y ( c n h , 0 ) , x ∣ V 3 / V 2 ∈ X ( 0 , m ) , x ∣ V 4 / V 3 ∈ Y ( c P , 0 ) } . X'=\left\{x\in E_V\mathrel{\left|\,\begin{aligned}
x|_{V_1/V_0}&\in Y(c_I,0),\quad x|_{V_2/V_1}\in Y(c_{nh},0),\\
x|_{V_3/V_2}&\in X(0,m),\quad x|_{V_4/V_3}\in Y(c_P,0)
\end{aligned}\right.}\right\}. X ′ = { x ∈ E V x ∣ V 1 / V 0 x ∣ V 3 / V 2 ∈ Y ( c I , 0 ) , x ∣ V 2 / V 1 ∈ Y ( c nh , 0 ) , ∈ X ( 0 , m ) , x ∣ V 4 / V 3 ∈ Y ( c P , 0 ) } .
Then it is clearly that X ′ X' X ′ is smooth. Thus X ( c , m ) = X ′ × P G ν X(c,m)=X'\times_PG_\nu X ( c , m ) = X ′ × P G ν is also smooth. □
Throughout, we use the following notation for the inclusions:
Y ( c , m ) → j c , m E ν , X ( c , m ) → ȷ ˉ c , m E ν , X ( c , m ) → ι c , m Y ( c , m ) . Y(c,m)\xrightarrow{j^{c,m}}E_\nu,\quad X(c,m)\xrightarrow{\bar{\jmath}^{c,m}}E_\nu,\quad X(c,m)\xrightarrow{\iota^{c,m}}Y(c,m). Y ( c , m ) j c , m E ν , X ( c , m ) ˉ c , m E ν , X ( c , m ) ι c , m Y ( c , m ) .
The partial order of Xiao–Xu–Zhao
We now recall [34 Definition 7.1] , making all auxiliary orders explicit. For distinct functions, the two lexicographic orders are
c − > L d − ⟺ c − ( t ) > d − ( t ) , t = max { u ≤ 0 : c − ( u ) ≠ d − ( u ) } , c_- >_L d_- \Longleftrightarrow c_-(t)>d_-(t),\quad t=\max\{u\le0:c_-(u)\ne d_-(u)\}, c − > L d − ⟺ c − ( t ) > d − ( t ) , t = max { u ≤ 0 : c − ( u ) = d − ( u )} ,
c + > L d + ⟺ c + ( t ) > d + ( t ) , t = min { u > 0 : c + ( u ) ≠ d + ( u ) } . (19) c_+ >_L d_+ \Longleftrightarrow c_+(t)>d_+(t),\quad t=\min\{u>0:c_+(u)\ne d_+(u)\}. \tag*{(19)} c + > L d + ⟺ c + ( t ) > d + ( t ) , t = min { u > 0 : c + ( u ) = d + ( u )} . (19)
Thus the coordinates are read in the orders 0 , − 1 , − 2 , … 0,-1,-2,\ldots 0 , − 1 , − 2 , … and 1 , 2 , 3 , … 1,2,3,\ldots 1 , 2 , 3 , … , respectively. In particular, the latter starts at the I 1 I_1 I 1 end of the preinjective order.
For multisegments of the same cyclic-quiver dimension vector, use
π ′ ≤ G π ⟺ dim k Hom ( U , M ( π ′ ) ) ≥ dim k Hom ( U , M ( π ) ) . (20) \pi'\le_G\pi\Longleftrightarrow\dim_k\operatorname{Hom}(U,M(\pi'))\ge\dim_k\operatorname{Hom}(U,M(\pi)). \tag*{(20)} π ′ ≤ G π ⟺ dim k Hom ( U , M ( π ′ )) ≥ dim k Hom ( U , M ( π )) . (20)
The inequality is required for every U ∈ nilrep k C r U\in\operatorname{nilrep}_k\mathbb{C}r U ∈ nilrep k C r ; M ( π ) M(\pi) M ( π ) denotes the corresponding cyclic-quiver representation. Equivalently, it is enough to test indecomposable U U U . Multisegments of different cyclic-quiver dimension vectors are not comparable under ≤ G \le_G ≤ G . This is the order used in [34 §4.3] .
For c 0 = ( π 1 , … , π e ) c_0=(\pi_1,\ldots,\pi_e) c 0 = ( π 1 , … , π e ) and c 0 ′ = ( π 1 ′ , … , π e ′ ) c'_0=(\pi'_1,\ldots,\pi'_e) c 0 ′ = ( π 1 ′ , … , π e ′ ) , write
c 0 ′ < G c 0 ⟺ π j ′ ≤ G π j for all j , c 0 ′ ≠ c 0 . c'_0<_Gc_0\Longleftrightarrow\pi'_j\le_G\pi_j\text{ for all }j,\quad c'_0\ne c_0. c 0 ′ < G c 0 ⟺ π j ′ ≤ G π j for all j , c 0 ′ = c 0 .
Definition 1.6 ([34 Definition 7.1] ). Fix ν ∈ N \nu\in\mathbb{N} ν ∈ N . For ( c ′ , m ′ ) , ( c , m ) ∈ G ~ ν (c',m'),(c,m)\in\widetilde{\mathcal{G}}_\nu ( c ′ , m ′ ) , ( c , m ) ∈ G ν , with corresponding elements c ′ , c ∈ G ′ c',c\in\mathcal{G}' c ′ , c ∈ G ′ , declare ( c ′ , m ′ ) ≺ ( c , m ) (c',m')\prec(c,m) ( c ′ , m ′ ) ≺ ( c , m ) precisely in the following cases:
(a) c − ′ ≥ L c − c'_-\ge_Lc_- c − ′ ≥ L c − and c + ′ ≥ L c + c'_+\ge_Lc_+ c + ′ ≥ L c + , with at least one of these inequalities strict;
(b) c − ′ = c − c'_-=c_- c − ′ = c − , c + ′ = c + c'_+=c_+ c + ′ = c + , and m ′ < m m'<m m ′ < m ;
(c) c − ′ = c − c'_-=c_- c − ′ = c − , c + ′ = c + c'_+=c_+ c + ′ = c + , m ′ = m m'=m m ′ = m , and c 0 ′ < G c 0 c'_0<_Gc_0 c 0 ′ < G c 0 .
Adjoining equality defines ⪯ \preceq ⪯ on G ~ ν \widetilde{\mathcal{G}}_\nu G ν .
Proposition 1.7. [34 Proposition 7.3] In E ν E_\nu E ν ,
X ( c , m ) ‾ ⊂ ⋃ ( c ′ , m ′ ) ⪯ ( c , m ) Y ( c ′ , m ′ ) . \overline{X(c,m)}\subset\bigcup_{(c',m')\preceq(c,m)}Y(c',m'). X ( c , m ) ⊂ ( c ′ , m ′ ) ⪯ ( c , m ) ⋃ Y ( c ′ , m ′ ) .
Lusztig sheaf complexes, the canonical basis, and the PBW basis
The variety E ν E_\nu E ν has a natural F p \mathbb{F}_p F p -structure with Frobenius map Fr \operatorname{Fr} Fr . Lusztig defines D m , G ν b ( E ν ) \mathcal{D}^b_{m,G_\nu}(E_\nu) D m , G ν b ( E ν ) as the derived category of mixed G ν G_\nu G ν -equivariant sheaf complexes ( D m , r , G ν b ( E ν ) (\mathcal{D}^b_{m,r,G_\nu}(E_\nu) ( D m , r , G ν b ( E ν ) with Frobenius map Fr r \operatorname{Fr}^r Fr r if q = p r q=p^r q = p r ). The mixed structure is taken with respect to the F q \mathbb{F}_q F q -structure of the variety, which gives rise to a Weil structure on D G ν b ( E ν ) \mathcal{D}^b_{G_\nu}(E_\nu) D G ν b ( E ν ) , the derived category of G ν G_\nu G ν -equivariant sheaf complexes. We now recall the induction functor defined by Lusztig:
Ind ν ′ , ν ′ ′ ν : D m , G ν ′ b ( E ν ′ ) × D m , G ν ′ ′ b ( E ν ′ ′ ) ⟶ D m , G ν b ( E ν ) , \operatorname{Ind}_{\nu',\nu''}^{\nu}:D^b_{m,G_{\nu'}}(E_{\nu'})\times D^b_{m,G_{\nu''}}(E_{\nu''})\longrightarrow D^b_{m,G_{\nu}}(E_{\nu}), Ind ν ′ , ν ′′ ν : D m , G ν ′ b ( E ν ′ ) × D m , G ν ′′ b ( E ν ′′ ) ⟶ D m , G ν b ( E ν ) ,
where the dimension vectors of the Q 0 Q_0 Q 0 -graded vector spaces V \mathbf{V} V , V ′ \mathbf{V}' V ′ , and V ′ ′ \mathbf{V}'' V ′′ are ν \nu ν , ν ′ \nu' ν ′ , and ν ′ ′ \nu'' ν ′′ , respectively. Let
E ν ′ = { ( x , W , ρ 1 , ρ 2 ) | x ∈ E ν , W ≅ V ′ ′ as an Q 0 -graded space , x ( W ) ⊂ W , ρ 1 : V / W ≅ V ′ , ρ 2 : W ≅ V ′ ′ } , E'_{\nu}=\left\{(x,\mathbf{W},\rho_1,\rho_2)\ \middle|\ \begin{array}{l}x\in E_{\nu},\ \mathbf{W}\cong\mathbf{V}''\text{ as an }Q_0\text{-graded space},\\ x(\mathbf{W})\subset\mathbf{W},\ \rho_1:\mathbf{V}/\mathbf{W}\cong\mathbf{V}',\ \rho_2:\mathbf{W}\cong\mathbf{V}''\end{array}\right\}, E ν ′ = { ( x , W , ρ 1 , ρ 2 ) x ∈ E ν , W ≅ V ′′ as an Q 0 -graded space , x ( W ) ⊂ W , ρ 1 : V / W ≅ V ′ , ρ 2 : W ≅ V ′′ } ,
E ν ′ ′ = { ( x , W ) ∣ x ∈ E ν , x ( W ) ⊂ W , W ≅ V ′ ′ as an Q 0 -graded space } , E''_{\nu}=\{(x,\mathbf{W})\mid x\in E_{\nu},\ x(\mathbf{W})\subset\mathbf{W},\ \mathbf{W}\cong\mathbf{V}''\text{ as an }Q_0\text{-graded space}\}, E ν ′′ = {( x , W ) ∣ x ∈ E ν , x ( W ) ⊂ W , W ≅ V ′′ as an Q 0 -graded space } ,
with maps defined by
p 1 ( x , W , ρ 1 , ρ 2 ) = ( ρ 1 ( x ∣ V / W ) ρ 1 − 1 , ρ 2 ( x ∣ W ) ρ 2 − 1 ) , p 2 ( x , W , ρ 1 , ρ 2 ) = ( x , W ) , p 3 ( x , W ) = x . p_1(x,\mathbf{W},\rho_1,\rho_2)=\left(\rho_1(x|_{\mathbf{V}/\mathbf{W}})\rho_1^{-1},\rho_2(x|_{\mathbf{W}})\rho_2^{-1}\right),\qquad p_2(x,\mathbf{W},\rho_1,\rho_2)=(x,\mathbf{W}),\qquad p_3(x,\mathbf{W})=x. p 1 ( x , W , ρ 1 , ρ 2 ) = ( ρ 1 ( x ∣ V / W ) ρ 1 − 1 , ρ 2 ( x ∣ W ) ρ 2 − 1 ) , p 2 ( x , W , ρ 1 , ρ 2 ) = ( x , W ) , p 3 ( x , W ) = x .
(21) \qquad E_{\nu'}\times E_{\nu''}\xleftarrow{\ p_1\ }E'_{\nu}\xrightarrow{\ p_2\ }E''_{\nu}\xrightarrow{\ p_3\ }E_{\nu}
We denote the dimensions of the fibers of the smooth morphisms p 1 p_1 p 1 and p 2 p_2 p 2 by d 1 d_1 d 1 and d 2 d_2 d 2 , respectively. The induction functor is defined as
Ind ν ′ , ν ′ ′ ν : = p 3 ! p 2 b p 1 ∗ [ d 1 − d 2 ] ( d 1 − d 2 2 ) . \operatorname{Ind}_{\nu',\nu''}^{\nu}:=p_{3!}p_{2b}p_1^*[d_1-d_2]\left(\frac{d_1-d_2}{2}\right). Ind ν ′ , ν ′′ ν := p 3 ! p 2 b p 1 ∗ [ d 1 − d 2 ] ( 2 d 1 − d 2 ) .
Here p 2 b p_{2b} p 2 b is the inverse functor to p 2 ∗ p_2^* p 2 ∗ . Lusztig then defines flag sheaf complexes as follows. Throughout, we write Ind ν ′ , ν ′ ′ ν ( F ⊠ G ) \operatorname{Ind}_{\nu',\nu''}^{\nu}(\mathcal{F}\boxtimes\mathcal{G}) Ind ν ′ , ν ′′ ν ( F ⊠ G ) as F ⋄ G \mathcal{F}\diamond\mathcal{G} F ⋄ G .
For ν ∈ N Q 0 \nu\in\mathbb{N}Q_0 ν ∈ N Q 0 with dim V = ν \dim\mathbf{V}=\nu dim V = ν , define the set of flags S ν S_{\nu} S ν by
S ν = { ( ν 1 , ν 2 ⋯ ν n ) ∣ ν j = l j i j , i j ∈ Q 0 , ∑ j = 1 n l j i j = ν } . S_{\nu}=\{(\nu_1,\nu_2\cdots\nu_n)\mid\nu_j=l_j i_j,\ i_j\in Q_0,\ \sum_{j=1}^{n}l_j i_j=\nu\}. S ν = {( ν 1 , ν 2 ⋯ ν n ) ∣ ν j = l j i j , i j ∈ Q 0 , j = 1 ∑ n l j i j = ν } .
For ν = ( ν j ) ∈ S ν \boldsymbol{\nu}=(\nu_j)\in S_{\nu} ν = ( ν j ) ∈ S ν , we consider the flag variety F ~ ν \widetilde{\mathcal{F}}_{\boldsymbol{\nu}} F ν defined by
F ~ ν = { ( x , f ) | x ∈ E ν , f = ( 0 = V n ⊂ V n − 1 ⊂ ⋯ ⊂ V 1 ⊂ V 0 = V ) is a filtration of N Q 0 -graded spaces , x ( V i ) ⊂ V i , dim ( V i − 1 / V i ) = ν i } . \widetilde{\mathcal{F}}_{\boldsymbol{\nu}}=
\left\{(x,f)\ \middle|\ \begin{array}{l}
x\in E_{\nu},\ f=(0=V_n\subset V_{n-1}\subset\cdots\subset V_1\subset V_0=\mathbf{V})\text{ is a filtration of }\mathbb{N}Q_0\text{-graded spaces},\\
x(V_i)\subset V_i,\ \dim(V_{i-1}/V_i)=\nu_i
\end{array}\right\}. F ν = { ( x , f ) x ∈ E ν , f = ( 0 = V n ⊂ V n − 1 ⊂ ⋯ ⊂ V 1 ⊂ V 0 = V ) is a filtration of N Q 0 -graded spaces , x ( V i ) ⊂ V i , dim ( V i − 1 / V i ) = ν i } .
As shown in [21 Chapter 9] , this is a smooth irreducible variety; denote its dimension by f ( ν ) f(\boldsymbol{\nu}) f ( ν ) .
For F = ( 0 = V n ⊂ V n − 1 ⊂ ⋯ ⊂ V 1 ⊂ V 0 = V ) \mathcal{F}=(0=V_n\subset V_{n-1}\subset\cdots\subset V_1\subset V_0=\mathbf{V}) F = ( 0 = V n ⊂ V n − 1 ⊂ ⋯ ⊂ V 1 ⊂ V 0 = V ) , we abbreviate the condition x ( V i ) ⊂ V i x(V_i)\subset V_i x ( V i ) ⊂ V i , for every i i i , as x ( F ) ⊂ F x(\mathcal{F})\subset\mathcal{F} x ( F ) ⊂ F .
There is a natural projective morphism π ν : F ~ ν → E ν \pi_{\boldsymbol{\nu}}:\widetilde{\mathcal{F}}_{\boldsymbol{\nu}}\to E_{\nu} π ν : F ν → E ν . The flag sheaf complex corresponding to ν \boldsymbol{\nu} ν is
L ν : = π ν ! ( Q ‾ ℓ ) [ f ( ν ) ] ( f ( ν ) 2 ) . \mathcal{L}_{\boldsymbol{\nu}}:=\pi_{\boldsymbol{\nu}!}(\overline{\mathbb{Q}}_\ell)[f(\boldsymbol{\nu})]\left(\frac{f(\boldsymbol{\nu})}{2}\right). L ν := π ν ! ( Q ℓ ) [ f ( ν )] ( 2 f ( ν ) ) .
By the BBD decomposition theorem [5 ] , it is semisimple. This sheaf complex carries a natural Weil structure. For a flag sheaf complex L \mathcal{L} L equipped with its natural Weil structure, we retain the notation L \mathcal{L} L .
For any ( c , m ) (c,m) ( c , m ) and λ = ( λ 1 , ⋯ , λ l ) \lambda=(\lambda_1,\cdots,\lambda_l) λ = ( λ 1 , ⋯ , λ l ) , where ∑ i = 1 l λ i = m \sum_{i=1}^{l}\lambda_i=m ∑ i = 1 l λ i = m , consider the following morphism:
{ ( x , F ) | F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , e = V 3 , 0 ⊂ ⋯ ⊂ V 3 , l ⊂ V 4 = V ) , x ∈ X ( c , m ) , M x ∣ V 3 , i / V 3 , i − 1 is homogeneous regular of dimension λ i δ , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ c I , M x ∣ V 2 , i / V 2 , i − 1 ≅ c n h , i , M x ∣ V 4 / V 3 , m ≅ c P } → p c , λ X ( c , m ) . \left\{(x,F)\ \middle|\ \begin{array}{l}
F=(0=V_0\subset V_1=V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,e}=V_{3,0}\subset\cdots\subset V_{3,l}\subset V_4=\mathbf{V}),\\
x\in X(c,m),\quad M_x|_{V_{3,i}/V_{3,i-1}}\text{ is homogeneous regular of dimension }\lambda_i\delta,\\
x(F)\subset F,\quad M_x|_{V_1/V_0}\cong c_I,\quad M_x|_{V_{2,i}/V_{2,i-1}}\cong c_{n_h,i},\quad M_x|_{V_4/V_{3,m}}\cong c_P
\end{array}\right\}
\xrightarrow{p_{c,\lambda}}X(c,m). ⎩ ⎨ ⎧ ( x , F ) F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , e = V 3 , 0 ⊂ ⋯ ⊂ V 3 , l ⊂ V 4 = V ) , x ∈ X ( c , m ) , M x ∣ V 3 , i / V 3 , i − 1 is homogeneous regular of dimension λ i δ , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ c I , M x ∣ V 2 , i / V 2 , i − 1 ≅ c n h , i , M x ∣ V 4 / V 3 , m ≅ c P ⎭ ⎬ ⎫ p c , λ X ( c , m ) .
It is an etale covering map with coset space S m / ∏ i = 1 l S λ i S_m/\prod_{i=1}^{l}S_{\lambda_i} S m / ∏ i = 1 l S λ i which is the same as in [20 6.6] , since this property is independent with the choice of c c c . Thus p c , λ ! Q ‾ ℓ p_{c,\lambda!}\overline{\mathbb{Q}}_\ell p c , λ ! Q ℓ is independent of the order of ( λ 1 , ⋯ , λ l ) (\lambda_1,\cdots,\lambda_l) ( λ 1 , ⋯ , λ l ) . We may therefore choose λ = ( λ 1 , ⋯ , λ l ) \lambda=(\lambda_1,\cdots,\lambda_l) λ = ( λ 1 , ⋯ , λ l ) with λ 1 ≥ λ 2 ≥ ⋯ ≥ λ l \lambda_1\geq\lambda_2\geq\cdots\geq\lambda_l λ 1 ≥ λ 2 ≥ ⋯ ≥ λ l . Moreover,
p c , λ ! Q ‾ ℓ ≅ ⨁ χ ∈ X ( S m ) L χ ⊕ K λ , χ p_{c,\lambda!}\overline{\mathbb{Q}}_\ell\cong\bigoplus_{\chi\in X(S_m)}\mathcal{L}^{\chi}\oplus K_{\lambda,\chi} p c , λ ! Q ℓ ≅ χ ∈ X ( S m ) ⨁ L χ ⊕ K λ , χ
with the corresponding Weil structure, where K λ , χ K_{\lambda,\chi} K λ , χ is a Kostka number. As in [20 ] , we write
I C ( X ( c , m ) , L χ ) = j α ! ∗ ( L χ [ dim X ( c , m ) ] ( dim X ( c , m ) 2 ) ) . IC(X(c,m),L_{\chi})=j_{\alpha!*}\left(L_{\chi}[\dim X(c,m)]\left(\frac{\dim X(c,m)}{2}\right)\right). I C ( X ( c , m ) , L χ ) = j α ! ∗ ( L χ [ dim X ( c , m )] ( 2 dim X ( c , m ) ) ) .
Trace map
Let X Fr r X^{\operatorname{Fr}^{r}} X Fr r and G Fr r G^{\operatorname{Fr}^{r}} G Fr r denote the fixed point sets of X X X and G G G under their respective Frobenius maps Fr r \operatorname{Fr}^{r} Fr r . We denote by H ~ G Fr r ( X Fr r ) \widetilde{\mathcal{H}}_{G^{\operatorname{Fr}^{r}}}(X^{\operatorname{Fr}^{r}}) H G Fr r ( X Fr r ) the C \mathbb{C} C -vector space of G Fr r G^{\operatorname{Fr}^{r}} G Fr r -invariant functions on X Fr r X^{\operatorname{Fr}^{r}} X Fr r . For any G G G -equivariant morphism f : X → Y f:X\to Y f : X → Y that respects the F q \mathbb{F}_{q} F q -structures, we denote its restriction to X Fr r X^{\operatorname{Fr}^{r}} X Fr r by the same letter f : X Fr r → Y Fr r f:X^{\operatorname{Fr}^{r}}\to Y^{\operatorname{Fr}^{r}} f : X Fr r → Y Fr r . This gives C \mathbb{C} C -linear maps
f ∗ : H ~ G Fr r ( Y Fr r ) → H ~ G Fr r ( X Fr r ) f^{*}:\widetilde{\mathcal{H}}_{G^{\operatorname{Fr}^{r}}}(Y^{\operatorname{Fr}^{r}})\to\widetilde{\mathcal{H}}_{G^{\operatorname{Fr}^{r}}}(X^{\operatorname{Fr}^{r}}) f ∗ : H G Fr r ( Y Fr r ) → H G Fr r ( X Fr r )
f ! : H ~ G Fr r ( X Fr r ) → H ~ G Fr r ( Y Fr r ) f_{!}:\widetilde{\mathcal{H}}_{G^{\operatorname{Fr}^{r}}}(X^{\operatorname{Fr}^{r}})\to\widetilde{\mathcal{H}}_{G^{\operatorname{Fr}^{r}}}(Y^{\operatorname{Fr}^{r}}) f ! : H G Fr r ( X Fr r ) → H G Fr r ( Y Fr r )
φ ↦ ( x ↦ φ ( f ( x ) ) ) , ψ ↦ ( y ↦ ∑ x ∈ f − 1 ( y ) ψ ( x ) ) . \varphi\mapsto(x\mapsto\varphi(f(x))),\qquad
\psi\mapsto\left(y\mapsto\sum_{x\in f^{-1}(y)}\psi(x)\right). φ ↦ ( x ↦ φ ( f ( x ))) , ψ ↦ y ↦ x ∈ f − 1 ( y ) ∑ ψ ( x ) .
For any mixed Weil complex A ∈ D G , r , m b ( X ) A\in D^{b}_{G,r,m}(X) A ∈ D G , r , m b ( X ) with Weil structure ξ : Fr r ∗ ( A ) → A \xi:\operatorname{Fr}^{r*}(A)\to A ξ : Fr r ∗ ( A ) → A and any x ∈ X Fr r x\in X^{\operatorname{Fr}^{r}} x ∈ X Fr r , s ∈ Z s\in\mathbb{Z} s ∈ Z , there is an isomorphism H s ( ξ ) x : H s ( A ) x → H s ( A ) x H^{s}(\xi)_{x}:H^{s}(A)_{x}\to H^{s}(A)_{x} H s ( ξ ) x : H s ( A ) x → H s ( A ) x of the stalk at x x x of the s s s -th cohomology sheaf. Taking the alternating sum of their traces gives
χ ( A ) ( x ) = ∑ s ∈ Z ( − 1 ) s tr ( H s ( ξ ) x ) ∈ Q ‾ ℓ ≅ C . \chi(A)(x)=\sum_{s\in\mathbb{Z}}(-1)^{s}\operatorname{tr}(H^{s}(\xi)_{x})\in\overline{\mathbb{Q}}_{\ell}\cong\mathbb{C}. χ ( A ) ( x ) = s ∈ Z ∑ ( − 1 ) s tr ( H s ( ξ ) x ) ∈ Q ℓ ≅ C .
and hence a function χ ( A ) ∈ H ~ G Fr r ( X Fr r ) \chi(A)\in\widetilde{\mathcal{H}}_{G^{\operatorname{Fr}^{r}}}(X^{\operatorname{Fr}^{r}}) χ ( A ) ∈ H G Fr r ( X Fr r ) , defining a map from D G , r , m b ( X ) D^{b}_{G,r,m}(X) D G , r , m b ( X ) to H ~ G Fr r ( X Fr r ) \widetilde{\mathcal{H}}_{G^{\operatorname{Fr}^{r}}}(X^{\operatorname{Fr}^{r}}) H G Fr r ( X Fr r ) , called the trace map.
The geometric multiplication on ⨁ ν H G Fr r ( E ν Fr r ) \bigoplus_{\nu}\mathcal{H}_{G^{\operatorname{Fr}^{r}}}(E_{\nu}^{\operatorname{Fr}^{r}}) ⨁ ν H G Fr r ( E ν Fr r ) is defined as follows.
Definition 2.1. The geometric multiplication H G Fr r ( E ν ′ Fr r ) ⊗ H G Fr r ( E ν ′ ′ Fr r ) → H G Fr r ( E ν Fr r ) \mathcal{H}_{G^{\operatorname{Fr}^{r}}}(E_{\nu'}^{\operatorname{Fr}^{r}})\otimes\mathcal{H}_{G^{\operatorname{Fr}^{r}}}(E_{\nu''}^{\operatorname{Fr}^{r}})\to\mathcal{H}_{G^{\operatorname{Fr}^{r}}}(E_{\nu}^{\operatorname{Fr}^{r}}) H G Fr r ( E ν ′ Fr r ) ⊗ H G Fr r ( E ν ′′ Fr r ) → H G Fr r ( E ν Fr r ) is defined by
f ⋅ g = ( − q ) − ∑ i ∈ Q 0 ν i ′ ν i ′ ′ − ∑ h ∈ H ν s ( h ) ′ ν t ( h ) ′ ′ ∣ G ν ′ Fr r × G ν ′ ′ Fr r ∣ p 3 ! p 2 ! p 1 ∗ ( f ⊗ g ) , f\cdot g=\frac{(-\sqrt{q})^{-\sum_{i\in Q_{0}}\nu'_{i}\nu''_{i}-\sum_{h\in H}\nu'_{s(h)}\nu''_{t(h)}}}{\lvert G_{\nu'}^{\operatorname{Fr}^{r}}\times G_{\nu''}^{\operatorname{Fr}^{r}}\rvert}p_{3!}p_{2!}p_{1}^{*}(f\otimes g), f ⋅ g = ∣ G ν ′ Fr r × G ν ′′ Fr r ∣ ( − q ) − ∑ i ∈ Q 0 ν i ′ ν i ′′ − ∑ h ∈ H ν s ( h ) ′ ν t ( h ) ′′ p 3 ! p 2 ! p 1 ∗ ( f ⊗ g ) , where p 1 p_{1} p 1 , p 2 p_{2} p 2 , and p 3 p_{3} p 3 are the maps in Equation 21.
Theorem 2.2 ([1 Theorem 5.3.13] ). For any G G G -equivariant morphism f : X → Y f:X\to Y f : X → Y that respects the F q \mathbb{F}_{q} F q -structures and any A ∈ D G , r , m b ( X ) A\in D^{b}_{G,r,m}(X) A ∈ D G , r , m b ( X ) , B ∈ D G , r , m b ( Y ) B\in D^{b}_{G,r,m}(Y) B ∈ D G , r , m b ( Y ) , we have
χ ( A [ n ] ) = ( − 1 ) n χ ( A ) , χ ( A ( n 2 ) ) = q − n χ ( A ) , χ ( A ⊠ B ) = χ ( A ) ⊗ χ ( B ) , \chi(A[n])=(-1)^{n}\chi(A),\qquad\chi\left(A\left(\frac{n}{2}\right)\right)=\sqrt{q}^{-n}\chi(A),\qquad\chi(A\boxtimes B)=\chi(A)\otimes\chi(B), χ ( A [ n ]) = ( − 1 ) n χ ( A ) , χ ( A ( 2 n ) ) = q − n χ ( A ) , χ ( A ⊠ B ) = χ ( A ) ⊗ χ ( B ) , χ ( f ∗ ( B ) ) = f ∗ ( χ ( B ) ) , χ ( f ! ( A ) ) = f ! ( χ ( A ) ) . \chi(f^{*}(B))=f^{*}(\chi(B)),\qquad\chi(f_{!}(A))=f_{!}(\chi(A)). χ ( f ∗ ( B )) = f ∗ ( χ ( B )) , χ ( f ! ( A )) = f ! ( χ ( A )) . Moreover, χ ( F ⋄ G ) = χ ( F ) ⋅ χ ( G ) \chi(F\diamond G)=\chi(F)\cdot\chi(G) χ ( F ⋄ G ) = χ ( F ) ⋅ χ ( G ) , and Lusztig proves in [22 ] that the Z [ q , q − 1 ] \mathbb{Z}[\sqrt{q},\sqrt{q}^{-1}] Z [ q , q − 1 ] -span of { χ ( L ν ) ∣ ν ∈ S ν } \{\chi(\mathcal{L}_{\nu})\mid\nu\in S_{\nu}\} { χ ( L ν ) ∣ ν ∈ S ν } is isomorphic to the Z \mathbb{Z} Z -form of U ν − U^{-}_{\nu} U ν − , v = − q v=-\sqrt{q} v = − q , with respect to the multiplication ⋅ \cdot ⋅ , as explained in detail in [21 ] .
Canonical basis
In [20 ] , Lusztig shows that the simple perverse sheaves occurring, up to shift, as direct summands of flag sheaf complexes constitute the sheaf realization of the canonical basis. We denote this set by P = ⋃ ν P ν \mathcal{P}=\bigcup_{\nu}\mathcal{P}_{\nu} P = ⋃ ν P ν , and its elements have the form I C ( X ( c , m ) , L χ ) IC(X(c,m),L_{\chi}) I C ( X ( c , m ) , L χ ) , where c n h , j c_{nh,j} c nh , j is aperiodic for every j j j and χ ∈ X ( S m ) \chi\in X(S_{m}) χ ∈ X ( S m ) is a character of the symmetric group S m S_{m} S m .
In [22 ] , Lusztig defines a Weil structure ϕ c , χ \phi_{c,\chi} ϕ c , χ on each simple perverse sheaf I C ( X ( c , m ) , L χ ) ∈ P IC(X(c,m),L_{\chi})\in\mathcal{P} I C ( X ( c , m ) , L χ ) ∈ P , and shows that the resulting trace functions χ ( I C ( X ( c , m ) , L χ ) , ϕ c , χ ) \chi(IC(X(c,m),L_{\chi}),\phi_{c,\chi}) χ ( I C ( X ( c , m ) , L χ ) , ϕ c , χ ) form the canonical basis of U ν = − q − U^{-}_{\nu=-\sqrt{q}} U ν = − q − . Actually there is a canonical Weil structure ϕ c , χ \phi_{c,\chi} ϕ c , χ on any I C ( X ( c , m ) , L χ ) IC(X(c,m),L_{\chi}) I C ( X ( c , m ) , L χ ) induced by the Weil structure on L χ ∣ X ( c , m ) L_{\chi}|_{X(c,m)} L χ ∣ X ( c , m ) , for ( c , m ) ∈ G ~ ν (c,m)\in\widetilde{\mathcal{G}}_{\nu} ( c , m ) ∈ G ν .
This Weil structure on I C ( X ( c , m ) , L χ ) IC(X(c,m),L_{\chi}) I C ( X ( c , m ) , L χ ) is induced by the following complex:
L c , ( 1 , … , 1 ) : = I C ( O c p , Q ‾ ℓ ) ⋆ I C ( O c n h , 1 , Q ‾ ℓ ) ⋆ ⋯ ⋆ I C ( O c n h , l , Q ‾ ℓ ) ⋆ ∏ i = 1 m I C ( X ( 0 , 1 ) , Q ‾ ℓ ) ⋆ I C ( O c l , Q ‾ ℓ ) , \mathcal{L}_{c,(1,\ldots,1)}:=IC(\mathcal{O}_{c_{p}},\overline{\mathbb{Q}}_{\ell})\star IC(\mathcal{O}_{c_{nh,1}},\overline{\mathbb{Q}}_{\ell})\star\cdots\star IC(\mathcal{O}_{c_{nh,l}},\overline{\mathbb{Q}}_{\ell})\star\prod_{i=1}^{m}IC(X(0,1),\overline{\mathbb{Q}}_{\ell})\star IC(\mathcal{O}_{c_{l}},\overline{\mathbb{Q}}_{\ell}), L c , ( 1 , … , 1 ) := I C ( O c p , Q ℓ ) ⋆ I C ( O c nh , 1 , Q ℓ ) ⋆ ⋯ ⋆ I C ( O c nh , l , Q ℓ ) ⋆ i = 1 ∏ m I C ( X ( 0 , 1 ) , Q ℓ ) ⋆ I C ( O c l , Q ℓ ) ,
of which I C ( X ( c , m ) , Q ‾ ℓ ) IC(X(c,m),\overline{\mathbb{Q}}_{\ell}) I C ( X ( c , m ) , Q ℓ ) is a direct summand.
PBW basis
Consider j c , m ∗ L c , ( 1 , … , 1 ) j^{c,m}* \mathcal{L}_{c,(1,\ldots,1)} j c , m ∗ L c , ( 1 , … , 1 ) , where c n h , j c_{nh,j} c nh , j need not be aperiodic. This complex carries a natural Weil structure ψ c , m \psi_{c,m} ψ c , m , whose trace function we compute using the following morphism:
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , e = V 3 , 0 ⊂ ⋯ ⊂ V 3 , m ⊂ V 4 = V ) , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ c I , M x ∣ V 2 , i / V 2 , i − 1 ≅ c n h , i , M x ∣ V 3 , i / V 3 , i − 1 is homogeneous regular , M x ∣ V 4 / V 3 , m ≅ c P } → π c , ( 1 , … , 1 ) Y ( c , m ) . \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in E_{\nu},\quad F=(0=V_{0}\subset V_{1}=V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,e}=V_{3,0}\subset\cdots\subset V_{3,m}\subset V_{4}=V),\ x(F)\subset F,\\
&M_{x|V_{1}/V_{0}}\cong c_{I},\ M_{x|V_{2,i}/V_{2,i-1}}\cong c_{nh,i},\ M_{x|V_{3,i}/V_{3,i-1}}\text{ is homogeneous regular},\ M_{x|V_{4}/V_{3,m}}\cong c_{P}
\end{aligned}\right\}
\xrightarrow{\ \pi_{c,(1,\ldots,1)}\ }Y(c,m). { ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , e = V 3 , 0 ⊂ ⋯ ⊂ V 3 , m ⊂ V 4 = V ) , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ c I , M x ∣ V 2 , i / V 2 , i − 1 ≅ c nh , i , M x ∣ V 3 , i / V 3 , i − 1 is homogeneous regular , M x ∣ V 4 / V 3 , m ≅ c P } π c , ( 1 , … , 1 ) Y ( c , m ) .
Lemma 2.3. π c , ( 1 , … , 1 ) \pi_{c,(1,\ldots,1)} π c , ( 1 , … , 1 ) is a small morphism respect to X ( c , m ) X(c,m) X ( c , m ) .
Proof. The proof is the same as the argument of [33 Lemma 2.5] . π c , ( 1 , … , 1 ) \pi_{c,(1,\ldots,1)} π c , ( 1 , … , 1 ) is proper since it could be regarded as a proper morphism
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , e = V 3 , 0 ⊂ ⋯ ⊂ V 3 , m ⊂ V 4 = V ) , x ( F ) ⊂ F , x ∣ V 1 / V 0 ∈ O ‾ c I , x ∣ V 2 , i / V 2 , i − 1 ∈ O ‾ c n h , i , x ∣ V 4 / V 3 , m ∈ O ‾ c P } ⟶ E ν \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in E_{\nu},\quad F=(0=V_{0}\subset V_{1}=V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,e}=V_{3,0}\subset\cdots\subset V_{3,m}\subset V_{4}=V),\ x(F)\subset F,\\
&x|_{V_{1}/V_{0}}\in\overline{\mathcal{O}}_{c_{I}},\quad x|_{V_{2,i}/V_{2,i-1}}\in\overline{\mathcal{O}}_{c_{nh,i}},\quad x|_{V_{4}/V_{3,m}}\in\overline{\mathcal{O}}_{c_{P}}
\end{aligned}\right\}
\longrightarrow E_{\nu} { ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , e = V 3 , 0 ⊂ ⋯ ⊂ V 3 , m ⊂ V 4 = V ) , x ( F ) ⊂ F , x ∣ V 1 / V 0 ∈ O c I , x ∣ V 2 , i / V 2 , i − 1 ∈ O c nh , i , x ∣ V 4 / V 3 , m ∈ O c P } ⟶ E ν
restricting on Y ( c , m ) Y(c,m) Y ( c , m ) .
First, X ( c , m ) X(c,m) X ( c , m ) is open and dense in Y ( c , m ) Y(c,m) Y ( c , m ) and for any x ∈ X ( c , m ) x\in X(c,m) x ∈ X ( c , m ) , ( π c , ( 1 , … , 1 ) ) − 1 ( x ) (\pi_{c,(1,\ldots,1)})^{-1}(x) ( π c , ( 1 , … , 1 ) ) − 1 ( x ) is a set of finite points. Moreover,
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , e = V 3 , 0 ⊂ ⋯ ⊂ V 3 , m ⊂ V 4 = V ) , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ c I , M x ∣ V 2 , i / V 2 , i − 1 ≅ c n h , i , M x ∣ V 3 , i / V 3 , i − 1 is homogeneous regular , M x ∣ V 4 / V 3 , m ≅ c P } \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in E_{\nu},\quad F=(0=V_{0}\subset V_{1}=V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,e}=V_{3,0}\subset\cdots\subset V_{3,m}\subset V_{4}=V),\ x(F)\subset F,\\
&M_{x|V_{1}/V_{0}}\cong c_{I},\ M_{x|V_{2,i}/V_{2,i-1}}\cong c_{nh,i},\ M_{x|V_{3,i}/V_{3,i-1}}\text{ is homogeneous regular},\ M_{x|V_{4}/V_{3,m}}\cong c_{P}
\end{aligned}\right\} { ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , e = V 3 , 0 ⊂ ⋯ ⊂ V 3 , m ⊂ V 4 = V ) , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ c I , M x ∣ V 2 , i / V 2 , i − 1 ≅ c nh , i , M x ∣ V 3 , i / V 3 , i − 1 is homogeneous regular , M x ∣ V 4 / V 3 , m ≅ c P }
is smooth with the same dimension as dim X ( c , m ) \dim X(c,m) dim X ( c , m ) by the same method as in Proposition 1.5.
Denote
Z ( c , ( t 1 , … , t r ) ) = { x ∈ E ν : M x ≅ c ⊕ ( R u 1 [ t 1 1 ] ⊕ ⋯ ⊕ R u 1 [ t 1 j 1 ] ) ⊕ ⋯ ⊕ ( R u r [ t r 1 ] ⊕ ⋯ ⊕ R u r [ t r j r ] ) , Z(c,(t_{1},\ldots,t_{r}))=\left\{x\in E_{\nu}:M_{x}\cong c\oplus(R_{u_{1}}[t_{1}^{1}]\oplus\cdots\oplus R_{u_{1}}[t_{1}^{j_{1}}])\oplus\cdots\oplus(R_{u_{r}}[t_{r}^{1}]\oplus\cdots\oplus R_{u_{r}}[t_{r}^{j_{r}}]),\right. Z ( c , ( t 1 , … , t r )) = { x ∈ E ν : M x ≅ c ⊕ ( R u 1 [ t 1 1 ] ⊕ ⋯ ⊕ R u 1 [ t 1 j 1 ]) ⊕ ⋯ ⊕ ( R u r [ t r 1 ] ⊕ ⋯ ⊕ R u r [ t r j r ]) ,
u 1 , … , u r ∈ Z h pairwise distinct } , (21) \left.u_{1},\ldots,u_{r}\in\mathbb{Z}_{h}\text{ pairwise distinct}\right\}, \tag*{(21)} u 1 , … , u r ∈ Z h pairwise distinct } , (21)
where t i = ( t i 1 , … , t i j i ) t_{i}=(t_{i}^{1},\ldots,t_{i}^{j_{i}}) t i = ( t i 1 , … , t i j i ) , t i 1 ≤ ⋯ ≤ t i j i ∈ N > 0 t_{i}^{1}\leq\cdots\leq t_{i}^{j_{i}}\in\mathbb{N}_{>0} t i 1 ≤ ⋯ ≤ t i j i ∈ N > 0 and ∑ i = 1 r ∑ s = 1 j i t i s = m \sum_{i=1}^{r}\sum_{s=1}^{j_{i}}t_{i}^{s}=m ∑ i = 1 r ∑ s = 1 j i t i s = m . It is a locally closed subset by [6 ] .
Denote l i = ∑ s = 1 j i t i s l_{i}=\sum_{s=1}^{j_{i}}t_{i}^{s} l i = ∑ s = 1 j i t i s and for x ∈ Z ( c , ( t 1 , … , t r ) ) x\in Z(c,(t_{1},\ldots,t_{r})) x ∈ Z ( c , ( t 1 , … , t r )) , the part R u i [ t i 1 ] ⊕ ⋯ ⊕ R u i [ t i j i ] R_{u_{i}}[t_{i}^{1}]\oplus\cdots\oplus R_{u_{i}}[t_{i}^{j_{i}}] R u i [ t i 1 ] ⊕ ⋯ ⊕ R u i [ t i j i ] of M x M_{x} M x is denoted by M x u i M_{x}^{u_{i}} M x u i .
Y ( c , m ) = ⨆ Z ( c , ( t 1 , … , t r ) ) Y(c,m)=\bigsqcup Z(c,(t_{1},\ldots,t_{r})) Y ( c , m ) = ⨆ Z ( c , ( t 1 , … , t r )) and Z ( c , ( ( 1 ) , … , ( 1 ) ) ) = X ( c , m ) Z(c,((1),\ldots,(1)))=X(c,m) Z ( c , (( 1 ) , … , ( 1 ))) = X ( c , m ) . Fix x 0 ∈ X ( c , m ) x_{0}\in X(c,m) x 0 ∈ X ( c , m ) .
For x ∈ Z ( c , ( t 1 , … , t r ) ) x\in Z(c,(t_{1},\ldots,t_{r})) x ∈ Z ( c , ( t 1 , … , t r )) other that X ( c , m ) X(c,m) X ( c , m ) , as in [33 Lemma 2.5] , Springer resolution is semismall, dim f − 1 ( x ) ≤ ∑ i = 1 r 1 2 ( l i ( l i − 1 ) − ( l i 2 − dim Hom k Q ( M x u i , M x u i ) ) ) \dim f^{-1}(x)\leq\sum_{i=1}^{r}\frac{1}{2}(l_{i}(l_{i}-1)-(l_{i}^{2}-\dim\operatorname{Hom}_{kQ}(M_{x}^{u_{i}},M_{x}^{u_{i}}))) dim f − 1 ( x ) ≤ ∑ i = 1 r 2 1 ( l i ( l i − 1 ) − ( l i 2 − dim Hom k Q ( M x u i , M x u i ))) .
Then
dim f − 1 ( x ) ≤ ∑ i = 1 r 1 2 ( l i ( l i − 1 ) − ( l i 2 − dim Hom k Q ( M x u i , M x u i ) ) ) = ∑ i = 1 r 1 2 ( dim Hom k Q ( M x u i , M x u i ) − l i ) = 1 2 ( ( ∑ i = 1 r dim Hom k Q ( M x u i , M x u i ) ) − m ) = 1 2 ( dim O x 0 − dim O x ) = 1 2 ( dim X ( c , m ) − dim Z ( c , ( t 1 , … , t r ) ) + r − m ) . \begin{aligned}
\dim f^{-1}(x)&\leq\sum_{i=1}^{r}\frac{1}{2}\left(l_{i}(l_{i}-1)-(l_{i}^{2}-\dim\operatorname{Hom}_{kQ}(M_{x}^{u_{i}},M_{x}^{u_{i}}))\right)\\
&=\sum_{i=1}^{r}\frac{1}{2}\left(\dim\operatorname{Hom}_{kQ}(M_{x}^{u_{i}},M_{x}^{u_{i}})-l_{i}\right)\\
&=\frac{1}{2}\left(\left(\sum_{i=1}^{r}\dim\operatorname{Hom}_{kQ}(M_{x}^{u_{i}},M_{x}^{u_{i}})\right)-m\right)\\
&=\frac{1}{2}\left(\dim\mathcal{O}_{x_{0}}-\dim\mathcal{O}_{x}\right)\\
&=\frac{1}{2}\left(\dim X(c,m)-\dim Z(c,(t_{1},\ldots,t_{r}))+r-m\right).
\end{aligned} dim f − 1 ( x ) ≤ i = 1 ∑ r 2 1 ( l i ( l i − 1 ) − ( l i 2 − dim Hom k Q ( M x u i , M x u i )) ) = i = 1 ∑ r 2 1 ( dim Hom k Q ( M x u i , M x u i ) − l i ) = 2 1 ( ( i = 1 ∑ r dim Hom k Q ( M x u i , M x u i ) ) − m ) = 2 1 ( dim O x 0 − dim O x ) = 2 1 ( dim X ( c , m ) − dim Z ( c , ( t 1 , … , t r )) + r − m ) .
For the reason that Z ( c , ( t 1 , … , t r ) ) ≠ X ( c , m ) Z(c,(t_{1},\ldots,t_{r}))\neq X(c,m) Z ( c , ( t 1 , … , t r )) = X ( c , m ) , r < m r<m r < m . Thus
dim f − 1 ( x ) < 1 2 ( dim X ( c , m ) − dim Z ( c , ( t 1 , … , t r ) ) ) . \dim f^{-1}(x)<\frac{1}{2}\left(\dim X(c,m)-\dim Z(c,(t_{1},\ldots,t_{r}))\right). dim f − 1 ( x ) < 2 1 ( dim X ( c , m ) − dim Z ( c , ( t 1 , … , t r )) ) .
We have the proof. □
Thus
π c , ( 1 , ⋯ , 1 ) ! Q ‾ ℓ [ dim X ( c , m ) ] ( dim ( X ( c , m ) ) 2 ) ≃ ι c , m ! ∗ ( ⨁ χ ∈ X ( S m ) L χ ⊕ K ( 1 , ⋯ , 1 ) , χ [ dim X ( c , m ) ] ( dim ( X ( c , m ) ) 2 ) ) \pi_{c,(1,\cdots,1)!}\overline{\mathbb{Q}}_\ell[\dim X(c,m)]\left(\frac{\dim(X(c,m))}{2}\right)\simeq\iota_{c,m!*}\left(\bigoplus_{\chi\in X(S_m)}L_\chi^{\oplus K_{(1,\cdots,1),\chi}}[\dim X(c,m)]\left(\frac{\dim(X(c,m))}{2}\right)\right) π c , ( 1 , ⋯ , 1 )! Q ℓ [ dim X ( c , m )] ( 2 dim ( X ( c , m )) ) ≃ ι c , m ! ∗ χ ∈ X ( S m ) ⨁ L χ ⊕ K ( 1 , ⋯ , 1 ) , χ [ dim X ( c , m )] ( 2 dim ( X ( c , m )) )
with the same Weil structure. Similarly, for λ = ( λ 1 , ⋯ , λ l ) \lambda=(\lambda_1,\cdots,\lambda_l) λ = ( λ 1 , ⋯ , λ l ) , where ∑ i = 1 l λ i = m \sum_{i=1}^l\lambda_i=m ∑ i = 1 l λ i = m , the morphism
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , e = V 3 , 0 ⊂ ⋯ ⊂ V 3 , l ⊂ V 4 = V ) , M x ∣ V 3 , i / V 3 , i − 1 is homogeneous regular of dimension λ i δ , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≃ c I , M x ∣ V 2 , i / V 2 , i − 1 ≃ c n h , i , M x ∣ V 4 / V 3 , m ≃ c P } → π c , λ Y ( c , m ) , \left\{(x,F)\middle|\begin{array}{l}
x\in E_\nu,\quad F=(0=V_0\subset V_1=V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,e}=V_{3,0}\subset\cdots\subset V_{3,l}\subset V_4=V),\\
M_x|_{V_{3,i}/V_{3,i-1}}\text{ is homogeneous regular of dimension }\lambda_i\delta,\\
x(F)\subset F,\quad M_x|_{V_1/V_0}\simeq c_I,\quad M_x|_{V_{2,i}/V_{2,i-1}}\simeq c_{nh,i},\quad M_x|_{V_4/V_{3,m}}\simeq c_P
\end{array}\right\}
\xrightarrow{\pi_{c,\lambda}}Y(c,m), ⎩ ⎨ ⎧ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , e = V 3 , 0 ⊂ ⋯ ⊂ V 3 , l ⊂ V 4 = V ) , M x ∣ V 3 , i / V 3 , i − 1 is homogeneous regular of dimension λ i δ , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≃ c I , M x ∣ V 2 , i / V 2 , i − 1 ≃ c nh , i , M x ∣ V 4 / V 3 , m ≃ c P ⎭ ⎬ ⎫ π c , λ Y ( c , m ) ,
is also small respect to X ( c , m ) X(c,m) X ( c , m ) and thus π c , λ ! Q ‾ ℓ \pi_{c,\lambda!}\overline{\mathbb{Q}}_\ell π c , λ ! Q ℓ is independent of the order of ( λ 1 , ⋯ , λ l ) (\lambda_1,\cdots,\lambda_l) ( λ 1 , ⋯ , λ l ) . We choose λ = ( λ 1 , ⋯ , λ l ) \lambda=(\lambda_1,\cdots,\lambda_l) λ = ( λ 1 , ⋯ , λ l ) , with λ 1 ≥ ⋯ ≥ λ l \lambda_1\geq\cdots\geq\lambda_l λ 1 ≥ ⋯ ≥ λ l .
π c , λ ! Q ‾ ℓ [ dim X ( c , m ) ] ( dim ( X ( c , m ) ) 2 ) ≃ ι c , m ! ∗ ( ⨁ χ ∈ X ( S m ) L χ [ dim X ( c , m ) ] ( dim ( X ( c , m ) ) 2 ) ⊕ K λ , χ ) (22) \pi_{c,\lambda!}\overline{\mathbb{Q}}_\ell[\dim X(c,m)]\left(\frac{\dim(X(c,m))}{2}\right)\simeq\iota_{c,m!*}\left(\bigoplus_{\chi\in X(S_m)}L_\chi[\dim X(c,m)]\left(\frac{\dim(X(c,m))}{2}\right)^{\oplus K_{\lambda,\chi}}\right) \tag*{(22)} π c , λ ! Q ℓ [ dim X ( c , m )] ( 2 dim ( X ( c , m )) ) ≃ ι c , m ! ∗ χ ∈ X ( S m ) ⨁ L χ [ dim X ( c , m )] ( 2 dim ( X ( c , m )) ) ⊕ K λ , χ (22)
with the corresponding Weil structure.
Moreover, ι c , m ! ∗ L χ [ dim X ( c , m ) ] ( dim ( X ( c , m ) ) 2 ) ≃ j ∗ c , m I C ( X ( c , m ) , L χ ) \iota_{c,m!*}L_\chi[\dim X(c,m)]\left(\frac{\dim(X(c,m))}{2}\right)\simeq j^{c,m}_*IC(X(c,m),L_\chi) ι c , m ! ∗ L χ [ dim X ( c , m )] ( 2 d i m ( X ( c , m )) ) ≃ j ∗ c , m I C ( X ( c , m ) , L χ ) with the corresponding Weil structure. We denote the Weil structure on ι c , m ! ∗ L χ [ dim X ( c , m ) ] ( dim ( X ( c , m ) ) 2 ) ≃ j ∗ c , m I C ( X ( c , m ) , L χ ) \iota_{c,m!*}L_\chi[\dim X(c,m)]\left(\frac{\dim(X(c,m))}{2}\right)\simeq j^{c,m}_*IC(X(c,m),L_\chi) ι c , m ! ∗ L χ [ dim X ( c , m )] ( 2 d i m ( X ( c , m )) ) ≃ j ∗ c , m I C ( X ( c , m ) , L χ ) by ψ c , χ = j ∗ c , m ( φ c , χ ) \psi_{c,\chi}=j^{c,m}_*(\varphi_{c,\chi}) ψ c , χ = j ∗ c , m ( φ c , χ ) .
We now recall the Hall elements of [34 Sections 6 and 8] in our notation and compare them with the above trace functions. Put q = p r q=p^r q = p r , v = − q v=-\sqrt q v = − q , and ∣ β ∣ = ∑ i ∈ Q 0 β i |\beta|=\sum_{i\in Q_0}\beta_i ∣ β ∣ = ∑ i ∈ Q 0 β i . The Hall-algebra formulas below are specialized at this value of v v v .
Definition 2.4. Let H q ( Q ) \mathcal{H}_q(Q) H q ( Q ) be the twisted Hall algebra with basis { u [ M ] } \{u_{[M]}\} { u [ M ] } , indexed by the isomorphism classes of finite-dimensional F q Q \mathbb{F}_qQ F q Q -modules, and multiplication
u [ M ] ∗ u [ N ] = v ⟨ dim M , dim N ⟩ ∑ [ L ] g M N L u [ L ] . u_{[M]}*u_{[N]}=v^{\langle\dim M,\dim N\rangle}\sum_{[L]}g^L_{MN}u_{[L]}. u [ M ] ∗ u [ N ] = v ⟨ d i m M , d i m N ⟩ [ L ] ∑ g M N L u [ L ] . where
g M N L = # { U ⊆ L : U ≃ N , L / U ≃ M } , ⟨ μ , η ⟩ = ∑ i ∈ Q 0 μ i η i − ∑ h ∈ H μ s ( h ) η t ( h ) . g^L_{MN}=\#\{U\subseteq L:U\simeq N,\ L/U\simeq M\},\qquad\langle\mu,\eta\rangle=\sum_{i\in Q_0}\mu_i\eta_i-\sum_{h\in H}\mu_{s(h)}\eta_{t(h)}. g M N L = # { U ⊆ L : U ≃ N , L / U ≃ M } , ⟨ μ , η ⟩ = i ∈ Q 0 ∑ μ i η i − h ∈ H ∑ μ s ( h ) η t ( h ) . For such a module M M M , set
⟨ M ⟩ = v − ∣ dim M ∣ + dim F q End F q Q ( M ) u [ M ] . \langle M\rangle=v^{-|\dim M|+\dim_{\mathbb{F}_q}\operatorname{End}_{\mathbb{F}_qQ}(M)}u_{[M]}. ⟨ M ⟩ = v − ∣ d i m M ∣ + d i m F q End F q Q ( M ) u [ M ] . For n ≥ 1 n\geq1 n ≥ 1 , define
H n = v − n ∣ δ ∣ ∑ [ R ] , R homogeneous regular dim R = n δ u [ R ] , H 0 = 1 , H n = 0 ( n < 0 ) . H_n=v^{-n|\delta|}\sum_{\substack{[R],\ R\text{ homogeneous regular}\\ \dim R=n\delta}}u_{[R]},\qquad H_0=1,\qquad H_n=0\quad(n<0). H n = v − n ∣ δ ∣ [ R ] , R homogeneous regular d i m R = n δ ∑ u [ R ] , H 0 = 1 , H n = 0 ( n < 0 ) . The sum is over all homogeneous regular F q Q \mathbb{F}_qQ F q Q -modules of the indicated dimension vector. For a partition λ = ( λ 1 , … , λ l ) \lambda=(\lambda_1,\ldots,\lambda_l) λ = ( λ 1 , … , λ l ) of m m m , let t λ ∈ X ( S m ) t_\lambda\in X(S_m) t λ ∈ X ( S m ) be the corresponding irreducible character, and set
S λ = det ( H λ a − a + b ) 1 ≤ a , b ≤ l . S_\lambda=\det(H_{\lambda_a-a+b})_{1\leq a,b\leq l}. S λ = det ( H λ a − a + b ) 1 ≤ a , b ≤ l . The determinant is taken in the commutative subalgebra generated by the H n H_n H n , with multiplication ∗ * ∗ . We set $S_\varnothing=1.
Let c ≃ c P ⊕ c n h ⊕ c I c \simeq c_{P} \oplus c_{nh} \oplus c_{I} c ≃ c P ⊕ c nh ⊕ c I , with c n h ≃ ⨁ j = 1 e c n h , j c_{nh} \simeq\bigoplus_{j=1}^{e} c_{nh,j} c nh ≃ ⨁ j = 1 e c nh , j . Using the corresponding F q \mathbb{F}_{q} F q -forms, define
N ( c , t λ ) : = ⟨ c P ⟩ ∗ ⟨ c n h , 1 ⟩ ∗ ⋯ ∗ ⟨ c n h , e ⟩ ∗ S λ ∗ ⟨ c I ⟩ = ⟨ c P ⟩ ∗ ⟨ c n h ⟩ ∗ S λ ∗ ⟨ c I ⟩ . \begin{aligned}
N(c,t_{\lambda}) := \langle c_{P}\rangle* \langle c_{nh,1}\rangle* \cdots* \langle c_{nh,e}\rangle* S_{\lambda} * \langle c_{I}\rangle\\
&= \langle c_{P}\rangle* \langle c_{nh}\rangle* S_{\lambda} * \langle c_{I}\rangle.
\end{aligned} N ( c , t λ ) := ⟨ c P ⟩ ∗ ⟨ c nh , 1 ⟩ ∗ ⋯ ∗ ⟨ c nh , e ⟩ ∗ S λ ∗ ⟨ c I ⟩ = ⟨ c P ⟩ ∗ ⟨ c nh ⟩ ∗ S λ ∗ ⟨ c I ⟩ .
The second equality follows from the vanishing of Hom \operatorname{Hom} Hom and Ext 1 \operatorname{Ext}^{1} Ext 1 between distinct non-homogeneous tubes. No aperiodicity assumption on c n h , j c_{nh,j} c nh , j is imposed.
We identify H G ν F r r ( E ν F r r ) \mathcal{H}_{\mathbf{G}_{\nu}}^{\mathrm{Fr}^{r}}(\mathbb{E}_{\nu}^{\mathrm{Fr}^{r}}) H G ν Fr r ( E ν Fr r ) with the degree-ν \nu ν part of H q ( Q ) \mathcal{H}_{q}(Q) H q ( Q ) by sending the characteristic function 1 O M F r r \mathbf{1}_{\mathcal{O}_{M}^{\mathrm{Fr}^{r}}} 1 O M Fr r to u [ M ] u_{[M]} u [ M ] . The geometric multiplication ⋅ \cdot ⋅ and the Hall multiplication ∗ * ∗ are compared in the following lemma.
Lemma 2.5. Let λ \lambda λ be a partition of m m m , put χ = t λ \chi=t_{\lambda} χ = t λ , and let ν = dim c + m δ \nu=\dim c+m\delta ν = dim c + m δ . With the Weil structure ψ c , χ = ( j c , m ) ∗ φ c , χ \psi_{c,\chi}=(j^{c,m})_{*}\varphi_{c,\chi} ψ c , χ = ( j c , m ) ∗ φ c , χ , set
f c , χ : = χ ( ( j c , m ) ! ( j c , m ) ∗ I C ( X ( c , m ) , L χ ) , ( j c , m ) ! ψ c , χ ) . f_{c,\chi}:=\chi\left((j^{c,m})_{!}(j^{c,m})_{*}IC(X(c,m),\mathcal{L}_{\chi}),(j^{c,m})_{!}\psi_{c,\chi}\right). f c , χ := χ ( ( j c , m ) ! ( j c , m ) ∗ I C ( X ( c , m ) , L χ ) , ( j c , m ) ! ψ c , χ ) . Under the above vector-space identification, we have
f c , χ = v − dim G ν + ∣ ν ∣ N ( c , t λ ) . f_{c,\chi}=v^{-\dim\mathbf{G}_{\nu}+|\nu|}N(c,t_{\lambda}). f c , χ = v − d i m G ν + ∣ ν ∣ N ( c , t λ ) .
Proof. For dimension vectors μ , η \mu,\eta μ , η , write
b ( μ , η ) = ∑ i ∈ Q 0 μ i η i + ∑ h ∈ H μ s ( h ) η t ( h ) , a β = dim G β − ∣ β ∣ . b(\mu,\eta)=\sum_{i\in Q_{0}}\mu_{i}\eta_{i}+\sum_{h\in H}\mu_{s(h)}\eta_{t(h)},\qquad a_{\beta}=\dim\mathbf{G}_{\beta}-|\beta|. b ( μ , η ) = i ∈ Q 0 ∑ μ i η i + h ∈ H ∑ μ s ( h ) η t ( h ) , a β = dim G β − ∣ β ∣.
The definition of the multiplication on ⨁ ν H G ν F r r ( E ν F r r ) \bigoplus_{\nu}\mathcal{H}_{\mathbf{G}_{\nu}}^{\mathrm{Fr}^{r}}(\mathbb{E}_{\nu}^{\mathrm{Fr}^{r}}) ⨁ ν H G ν Fr r ( E ν Fr r ) gives
1 O M F r r ⋅ 1 O N F r r = v − b ( μ , η ) ∑ [ L ] g M N L 1 O L F r r , μ = dim M , η = dim N . \mathbf{1}_{\mathcal{O}_{M}^{\mathrm{Fr}^{r}}}\cdot\mathbf{1}_{\mathcal{O}_{N}^{\mathrm{Fr}^{r}}}
=v^{-b(\mu,\eta)}\sum_{[L]}g_{MN}^{L}\mathbf{1}_{\mathcal{O}_{L}^{\mathrm{Fr}^{r}}},
\qquad\mu=\dim M,\quad\eta=\dim N. 1 O M Fr r ⋅ 1 O N Fr r = v − b ( μ , η ) [ L ] ∑ g M N L 1 O L Fr r , μ = dim M , η = dim N .
Since dim G β = ∑ i β i 2 \dim\mathbf{G}_{\beta}=\sum_{i}\beta_{i}^{2} dim G β = ∑ i β i 2 , we have
a μ + η − b ( μ , η ) = a μ + a η + ⟨ μ , η ⟩ . a_{\mu+\eta}-b(\mu,\eta)=a_{\mu}+a_{\eta}+\langle\mu,\eta\rangle. a μ + η − b ( μ , η ) = a μ + a η + ⟨ μ , η ⟩ .
Consequently, the maps
Φ β ( 1 O M F r r ) = v a β u [ M ] \Phi_{\beta}(\mathbf{1}_{\mathcal{O}_{M}^{\mathrm{Fr}^{r}}})=v^{a_{\beta}}u_{[M]} Φ β ( 1 O M Fr r ) = v a β u [ M ]
satisfying
Φ β ( 1 O M F r r ) ∗ Φ α ( 1 O N F r r ) = Φ β + α ( 1 O M F r r ⋅ 1 O N F r r ) . \Phi_{\beta}(\mathbf{1}_{\mathcal{O}_{M}^{\mathrm{Fr}^{r}}})*\Phi_{\alpha}(\mathbf{1}_{\mathcal{O}_{N}^{\mathrm{Fr}^{r}}})
=\Phi_{\beta+\alpha}(\mathbf{1}_{\mathcal{O}_{M}^{\mathrm{Fr}^{r}}}\cdot\mathbf{1}_{\mathcal{O}_{N}^{\mathrm{Fr}^{r}}}). Φ β ( 1 O M Fr r ) ∗ Φ α ( 1 O N Fr r ) = Φ β + α ( 1 O M Fr r ⋅ 1 O N Fr r ) .
The equality dim O M = dim G dim M − dim F q End F q Q ( M ) \dim\mathcal{O}_{M}=\dim\mathbf{G}_{\dim M}-\dim_{\mathbb{F}_{q}}\operatorname{End}_{\mathbb{F}_{q}Q}(M) dim O M = dim G d i m M − dim F q End F q Q ( M ) therefore gives
Φ dim M ( v − dim O M 1 O M F r r ) = ⟨ M ⟩ . \Phi_{\dim M}\left(v^{-\dim\mathcal{O}_{M}}\mathbf{1}_{\mathcal{O}_{M}^{\mathrm{Fr}^{r}}}\right)=\langle M\rangle. Φ d i m M ( v − d i m O M 1 O M Fr r ) = ⟨ M ⟩ .
Moreover, Y ( 0 , n ) Y(0,n) Y ( 0 , n ) is open in E n δ E_{n\delta} E n δ , and dim E n δ = dim G n δ \dim E_{n\delta}=\dim\mathbf{G}_{n\delta} dim E n δ = dim G n δ , since ⟨ δ , δ ⟩ = 0 \langle\delta,\delta\rangle=0 ⟨ δ , δ ⟩ = 0 . Hence
Φ n δ ( v − dim E n δ 1 Y ( 0 , n ) F r r ) = H n . \Phi_{n\delta}\left(v^{-\dim E_{n\delta}}\mathbf{1}_{Y(0,n)}^{\mathrm{Fr}^{r}}\right)=H_{n}. Φ n δ ( v − d i m E n δ 1 Y ( 0 , n ) Fr r ) = H n .
Put d = dim X ( c , m ) = dim Y ( c , m ) d=\dim X(c,m)=\dim Y(c,m) d = dim X ( c , m ) = dim Y ( c , m ) . For a partition μ = ( μ 1 , … , μ s ) \mu=(\mu_{1},\ldots,\mu_{s}) μ = ( μ 1 , … , μ s ) of m m m , let R c , μ R_{c,\mu} R c , μ be the trace function of
( j c , m ) ! π c , μ ! Q ‾ ℓ [ d ] ( d / 2 ) (j^{c,m})_{!}\pi_{c,\mu!}\overline{\mathbb{Q}}_{\ell}[d](d/2) ( j c , m ) ! π c , μ ! Q ℓ [ d ] ( d /2 )
with its natural Weil structure.
Applying Φ \Phi Φ , gives
Φ ν ( R c , μ ) = ⟨ c P ⟩ ∗ ⟨ c n h ⟩ ∗ H μ ∗ ⟨ c I ⟩ , H μ : = H μ 1 ∗ ⋯ ∗ H μ s . \Phi_{\nu}(R_{c,\mu})=\langle c_{P}\rangle* \langle c_{nh}\rangle* H_{\mu} * \langle c_{I}\rangle,\qquad H_{\mu}:=H_{\mu_{1}}*\cdots*H_{\mu_{s}}. Φ ν ( R c , μ ) = ⟨ c P ⟩ ∗ ⟨ c nh ⟩ ∗ H μ ∗ ⟨ c I ⟩ , H μ := H μ 1 ∗ ⋯ ∗ H μ s .
By (22) and the compatibility of the Weil structures,
R c , μ = ∑ λ ⊢ m K μ , t λ f c , t λ . R_{c,\mu}=\sum_{\lambda\vdash m}K_{\mu,t_{\lambda}}f_{c,t_{\lambda}}. R c , μ = λ ⊢ m ∑ K μ , t λ f c , t λ .
On the other hand, the complete and Schur functions satisfy
H μ = ∑ λ ⊢ m K μ , t λ S λ , H_{\mu}=\sum_{\lambda\vdash m}K_{\mu,t\lambda}S_{\lambda}, H μ = λ ⊢ m ∑ K μ , t λ S λ ,
and hence
Φ ν ( R c , μ ) = ∑ λ ⊢ m K μ , t λ N ( c , t λ ) . \Phi_{\nu}(R_{c,\mu})=\sum_{\lambda\vdash m}K_{\mu,t\lambda}N(c,t_{\lambda}). Φ ν ( R c , μ ) = λ ⊢ m ∑ K μ , t λ N ( c , t λ ) .
The Kostka matrix is invertible. Comparing the last two expansions yields Φ ν ( f c , t λ ) = N ( c , t λ ) \Phi_{\nu}(f_{c,t_{\lambda}})=N(c,t_{\lambda}) Φ ν ( f c , t λ ) = N ( c , t λ ) . Since Φ ν \Phi_{\nu} Φ ν is multiplication by v dim G ν − ∣ ν ∣ v^{\dim G_{\nu}-|\nu|} v d i m G ν − ∣ ν ∣ under the stated vector-space identification, the assertion follows.
Flag sheaf complexes over Y ( c , m ) Y(c,m) Y ( c , m )
To describe flag sheaf complexes over Y ( c , m ) Y(c,m) Y ( c , m ) more explicitly, we first prove the following lemma.
Lemma 3.1. Let k k k be a field, let ℓ \ell ℓ be a prime invertible in k k k , and let 1 ≤ n ≤ m 1\leq n\leq m 1 ≤ n ≤ m . Set
St m , n : = { A ∈ Mat m × n , k ∣ rank A = n } , \operatorname{St}_{m,n}:=\{A\in\operatorname{Mat}_{m\times n,k}\mid\operatorname{rank} A=n\}, St m , n := { A ∈ Mat m × n , k ∣ rank A = n } , with the right action of G = Gl n , k G=\operatorname{Gl}_{n,k} G = Gl n , k given by A ⋅ g = A g A\cdot g=Ag A ⋅ g = A g . Let π : V → W \pi:V\to W π : V → W be a G G G -invariant morphism of k k k -varieties. Assume that there is an étale covering { U i → W } \{U_i\to W\} { U i → W } and G G G -equivariant isomorphisms
V × W U i ≃ U i × St m , n V\times_W U_i\simeq U_i\times\operatorname{St}_{m,n} V × W U i ≃ U i × St m , n over U i U_i U i , where G G G acts trivially on U i U_i U i . Assume that the quotient X : = V / G X:=V/G X := V / G is represented by a variety, and denote the induced morphism by f : X → W f:X\to W f : X → W . Define
P n , m − n : = { λ = ( λ 1 , … , λ n ) ∣ m − n ≥ λ 1 ≥ ⋯ ≥ λ n ≥ 0 } , ∣ λ ∣ : = ∑ a = 1 n λ a . P_{n,m-n}:=\{\lambda=(\lambda_1,\ldots,\lambda_n)\mid m-n\geq\lambda_1\geq\cdots\geq\lambda_n\geq0\},\qquad
|\lambda|:=\sum_{a=1}^{n}\lambda_a. P n , m − n := { λ = ( λ 1 , … , λ n ) ∣ m − n ≥ λ 1 ≥ ⋯ ≥ λ n ≥ 0 } , ∣ λ ∣ := a = 1 ∑ n λ a . Then there is an isomorphism in D c b ( W , Q ‾ ℓ ) D_c^b(W,\overline{\mathbb Q}_{\ell}) D c b ( W , Q ℓ )
R f ! Q ‾ ℓ , X ≃ ⨁ λ ∈ P n , m − n Q ‾ ℓ , W ( − ∣ λ ∣ ) [ − 2 ∣ λ ∣ ] . (23) Rf_!\overline{\mathbb Q}_{\ell,X}\simeq\bigoplus_{\lambda\in P_{n,m-n}}\overline{\mathbb Q}_{\ell,W}(-|\lambda|)[-2|\lambda|]. \tag*{(23)} R f ! Q ℓ , X ≃ λ ∈ P n , m − n ⨁ Q ℓ , W ( − ∣ λ ∣ ) [ − 2∣ λ ∣ ] . (23) In particular, if b j : = # { λ ∈ P n , m − n : ∣ λ ∣ = j } b_j:=\#\{\lambda\in P_{n,m-n}:|\lambda|=j\} b j := # { λ ∈ P n , m − n : ∣ λ ∣ = j } , then
R 2 j f ! Q ‾ ℓ , X ≃ Q ‾ ℓ , W ( − j ) ⊕ b j , R 2 j + 1 f ! Q ‾ ℓ , X = 0. R^{2j}f_!\overline{\mathbb Q}_{\ell,X}\simeq\overline{\mathbb Q}_{\ell,W}(-j)^{\oplus b_j},\qquad R^{2j+1}f_!\overline{\mathbb Q}_{\ell,X}=0. R 2 j f ! Q ℓ , X ≃ Q ℓ , W ( − j ) ⊕ b j , R 2 j + 1 f ! Q ℓ , X = 0. If k = F q k=\mathbb F_q k = F q , the isomorphism is compatible with the natural Weil structures.
Proof. The map A ↦ im A A\mapsto\operatorname{im}A A ↦ im A identifies St m , n → Gr ( n , m ) \operatorname{St}_{m,n}\to\operatorname{Gr}(n,m) St m , n → Gr ( n , m ) with the frame bundle of the tautological rank-n n n subbundle. Consequently, f f f is étale locally isomorphic to the projection U i × Gr ( n , m ) → U i U_i\times\operatorname{Gr}(n,m)\to U_i U i × Gr ( n , m ) → U i , and p : V → X p:V\to X p : V → X is a principal G G G -bundle. In particular, f f f is proper, so that R f ! = R f ∗ Rf_!=Rf_* R f ! = R f ∗ . The associated vector bundle
S : = V × G k n S:=V\times^G k^n S := V × G k n on X X X , with the convention ( v g , z ) ∼ ( v , g z ) (vg,z)\sim(v,gz) ( v g , z ) ∼ ( v , g z ) , restricts on each geometric fiber of f f f to the tautological subbundle on Gr ( n , m ) \operatorname{Gr}(n,m) Gr ( n , m ) . For λ ∈ P n , m − n \lambda\in P_{n,m-n} λ ∈ P n , m − n , let
α λ : = s λ ( S ∨ ) ∈ H e ˊ t 2 ∣ λ ∣ ( X , Q ‾ ℓ ( ∣ λ ∣ ) ) , \alpha_\lambda:=s_\lambda(S^\vee)\in H_{\mathrm{\acute et}}^{2|\lambda|}(X,\overline{\mathbb Q}_{\ell}(|\lambda|)), α λ := s λ ( S ∨ ) ∈ H e ˊ t 2∣ λ ∣ ( X , Q ℓ ( ∣ λ ∣ )) , where s λ ( S ∨ ) s_\lambda(S^\vee) s λ ( S ∨ ) denotes the Schur polynomial in the Chern roots of S ∨ S^\vee S ∨ . Equivalently, writing
c t ( S ) − 1 = ∑ r ≥ 0 h r t r , h 0 = 1 , h r = 0 for r < 0 , c_t(S)^{-1}=\sum_{r\geq0}h_rt^r,\qquad h_0=1,\qquad h_r=0\ \text{for }r<0, c t ( S ) − 1 = r ≥ 0 ∑ h r t r , h 0 = 1 , h r = 0 for r < 0 , one has α λ = det ( h λ a − a + b ) 1 ≤ a , b ≤ n \alpha_\lambda=\det(h_{\lambda_a-a+b})_{1\leq a,b\leq n} α λ = det ( h λ a − a + b ) 1 ≤ a , b ≤ n . These are globally defined étale Chern classes; their construction does not require X X X to be smooth.
On a geometric fiber, their restrictions are the Schubert classes, which form a basis of its étale cohomology. Here one uses the Grassmannian basis theorem and Giambelli formula [13 Proposition 14.6.5 and §14.7] , together with the Schubert affine paving: the cycle classes give all even cohomology, and odd cohomology vanishes. The convention using det ( h λ a − a + b ) \det(h_{\lambda_a-a+b}) det ( h λ a − a + b ) is also recorded explicitly in [26 Theorem 5.1(1)] .
By adjunction, each α λ \alpha_\lambda α λ determines a morphism
Q ‾ ℓ W ( − ∣ λ ∣ ) [ − 2 ∣ λ ∣ ] ⟶ R f ∗ Q ‾ ℓ X . \overline{\mathbb{Q}}_\ell{}_W(-|\lambda|)[-2|\lambda|] \longrightarrow Rf_*\overline{\mathbb{Q}}_\ell{}_X. Q ℓ W ( − ∣ λ ∣ ) [ − 2∣ λ ∣ ] ⟶ R f ∗ Q ℓ X .
Taking their direct sum gives
Φ : ⨁ λ ∈ P n , m − n Q ‾ ℓ W ( − ∣ λ ∣ ) [ − 2 ∣ λ ∣ ] ⟶ R f ∗ Q ‾ ℓ X . \Phi:\bigoplus_{\lambda\in\mathcal{P}_{n,m-n}}\overline{\mathbb{Q}}_\ell{}_W(-|\lambda|)[-2|\lambda|]\longrightarrow Rf_*\overline{\mathbb{Q}}_\ell{}_X. Φ : λ ∈ P n , m − n ⨁ Q ℓ W ( − ∣ λ ∣ ) [ − 2∣ λ ∣ ] ⟶ R f ∗ Q ℓ X .
For every geometric point w ‾ → W \overline{w}\to W w → W , proper base change identifies the target of Φ w ‾ \Phi_{\overline{w}} Φ w with R Γ ( X w ‾ , Q ‾ ℓ ) R\Gamma(X_{\overline{w}},\overline{\mathbb{Q}}_\ell) R Γ ( X w , Q ℓ ) . Here we apply proper base change first with finite étale coefficients and then pass to ℓ \ell ℓ -adic coefficients. The induced map on cohomology sends the generators of the source to the Schubert basis described above. Thus Φ \Phi Φ is an isomorphism on every geometric stalk, and hence an isomorphism in D c b ( W , Q ‾ ℓ ) D^b_c(W,\overline{\mathbb{Q}}_\ell) D c b ( W , Q ℓ ) .
When k = F q k=\mathbb{F}_q k = F q , the bundle S \mathcal{S} S and its Chern classes are defined over F q \mathbb{F}_q F q . The construction of Φ \Phi Φ is therefore compatible with Frobenius, including the indicated Tate twists. □
For s ∈ 1 2 Z s\in\frac{1}{2}\mathbb{Z} s ∈ 2 1 Z and ω ∈ Q ‾ ℓ × \omega\in\overline{\mathbb{Q}}_\ell^\times ω ∈ Q ℓ × with ∣ ι ℓ ( ω ) ∣ = 1 |\iota_\ell(\omega)|=1 ∣ ι ℓ ( ω ) ∣ = 1 , let E ω , s \mathcal{E}_{\omega,s} E ω , s be the rank-one Weil sheaf on Spec F q \operatorname{Spec}\mathbb{F}_q Spec F q on which Frobenius acts by ω q s \omega q^s ω q s . Thus E ω , s = E ω , 0 ( − s ) \mathcal{E}_{\omega,s}=\mathcal{E}_{\omega,0}(-s) E ω , s = E ω , 0 ( − s ) in the usual Tate-twist convention, and it has weight 2 s 2s 2 s . In this notation, Q ‾ ℓ ( s ) \overline{\mathbb{Q}}_\ell(s) Q ℓ ( s ) is written as E 1 , − s \mathcal{E}_{1,-s} E 1 , − s . Here and below, triangulated generation means membership in the full triangulated subcategory generated by the indicated objects.
Theorem 3.3. For any flag sheaf complex L \mathcal{L} L , any ( c , m ) ∈ G ~ ν (c,m)\in\widetilde{G}_\nu ( c , m ) ∈ G ν , the restriction L ∣ Y ( c , m ) \mathcal{L}|_{Y^{(c,m)}} L ∣ Y ( c , m ) is generated, as a triangulated object, by I C ( X ( c , m ) , L χ ) ∣ Y ( c , m ) ⊠ E ω , s IC(X^{(c,m)},\mathcal{L}_\chi)|_{Y^{(c,m)}}\boxtimes\mathcal{E}_{\omega,s} I C ( X ( c , m ) , L χ ) ∣ Y ( c , m ) ⊠ E ω , s , with ∣ ω ∣ = 1 |\omega|=1 ∣ ω ∣ = 1 , where χ ∈ X ( S m ) \chi\in X(S_m) χ ∈ X ( S m ) .
Proof. We claim that for any dimension vector ν ~ \widetilde{\nu} ν , any fixed flag type ( ν 1 , ⋯ , ν d ) (\nu_1,\cdots,\nu_d) ( ν 1 , ⋯ , ν d ) , and any i < j i<j i < j , if ∑ s = i + 1 j ν s ≤ ν ~ \sum_{s=i+1}^{j}\nu_s\leq\widetilde{\nu} ∑ s = i + 1 j ν s ≤ ν , then
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 ⊂ ⋯ ⊂ V d = V ) , x ( F ) ⊂ F , dim ( V s / V s − 1 ) = ν s , x ∣ V j / V i ∈ Y ( c , m ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ d } \left\{(x,F)\ \middle|\ x\in E_{\nu},\quad F=(0=V_0\subset V_1\subset\cdots\subset V_d=V),\quad x(F)\subset F,\quad\dim(V_s/V_{s-1})=\nu_s,\quad x|_{V_j/V_i}\in Y^{(c,m)},\quad x|_{V_s/V_{s-1}}\in Y^{(c_s,m_s)},\quad1\leq s\leq d\right\} { ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 ⊂ ⋯ ⊂ V d = V ) , x ( F ) ⊂ F , dim ( V s / V s − 1 ) = ν s , x ∣ V j / V i ∈ Y ( c , m ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ d }
→ π { ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 ⊂ ⋯ ⊂ V i ⊂ V j ⊂ V d = V ) , x ( F ) ⊂ F , dim ( V s / V s − 1 ) = ν s , 1 ≤ s ≤ i , j + 1 ≤ s ≤ d , x ∣ V j / V i ∈ Y ( c , m ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , j + 1 ≤ s ≤ d } . \xrightarrow{\ \pi\ }
\left\{(x,F)\ \middle|\ \begin{aligned}
&x\in E_{\nu},\quad F=(0=V_0\subset V_1\subset\cdots\subset V_i\subset V_j\subset V_d=V),\quad x(F)\subset F,\\
&\dim(V_s/V_{s-1})=\nu_s,\ 1\leq s\leq i,\ j+1\leq s\leq d,\\
&x|_{V_j/V_i}\in Y^{(c,m)},\quad x|_{V_s/V_{s-1}}\in Y^{(c_s,m_s)},\ 1\leq s\leq i,\ j+1\leq s\leq d
\end{aligned}\right\}. π ⎩ ⎨ ⎧ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 ⊂ ⋯ ⊂ V i ⊂ V j ⊂ V d = V ) , x ( F ) ⊂ F , dim ( V s / V s − 1 ) = ν s , 1 ≤ s ≤ i , j + 1 ≤ s ≤ d , x ∣ V j / V i ∈ Y ( c , m ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , j + 1 ≤ s ≤ d ⎭ ⎬ ⎫ .
π Q ‾ l \pi\overline{\mathcal{Q}}_l π Q l is generated by π λ Q l ⊗ E ω , s \pi_{\lambda}\mathcal{Q}_l\otimes\mathcal{E}_{\omega,s} π λ Q l ⊗ E ω , s , for suitable ( ω , s ) (\omega,s) ( ω , s ) , and λ = ( λ 1 , ⋯ , λ t ) \lambda=(\lambda_1,\cdots,\lambda_t) λ = ( λ 1 , ⋯ , λ t ) , where π λ \pi_{\lambda} π λ is the morphism
{ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V ′ = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , l = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , t ⊂ V j ⊂ ⋯ ⊂ V d = V ) , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V ′ / V i ≃ M C 1 , x ∣ V 2 , s / V 2 , s − 1 ≃ M C n h , s , x ∣ V 3 , s / V 3 , s − 1 is homogeneous regular of dimension λ s δ , x ∣ V j / V 3 ≃ M C P , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , j + 1 ≤ s ≤ d . \left\{
\begin{array}{c|l}
(x,F) & \begin{aligned}
x&\in E_{\nu},\\
F&=(0=V_0\subset\cdots\subset V_i\subset V'=V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,l}=V_{3,0}\subset V_{3,1}\subset\cdots\subset V_{3,t}\subset V_j\subset\cdots\subset V_d=V),\\
\dim(V_s/V_{s-1})&=v_s,\ x(F)\subset F,\ x|_{V'/V_i}\simeq M_{C_1},\ x|_{V_{2,s}/V_{2,s-1}}\simeq M_{C_{n_h,s}},\\
x|_{V_{3,s}/V_{3,s-1}}&\text{ is homogeneous regular of dimension }\lambda_s\delta,\\
x|_{V_j/V_3}&\simeq M_{C_P},\\
x|_{V_s/V_{s-1}}&\in Y(c_s,m_s),\quad1\leq s\leq i,\ j+1\leq s\leq d.
\end{aligned}
\end{array}
\right. ⎩ ⎨ ⎧ ( x , F ) x F dim ( V s / V s − 1 ) x ∣ V 3 , s / V 3 , s − 1 x ∣ V j / V 3 x ∣ V s / V s − 1 ∈ E ν , = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V ′ = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , l = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , t ⊂ V j ⊂ ⋯ ⊂ V d = V ) , = v s , x ( F ) ⊂ F , x ∣ V ′ / V i ≃ M C 1 , x ∣ V 2 , s / V 2 , s − 1 ≃ M C n h , s , is homogeneous regular of dimension λ s δ , ≃ M C P , ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , j + 1 ≤ s ≤ d .
⟶ { ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ V 1 ⊂ ⋯ ⊂ V d = V ) , x ( F ) ⊂ F , dim ( V s / V s − 1 ) = v s , x ∣ V j / V i ∈ Y ( c , m ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , j + 1 ≤ s ≤ d . \longrightarrow
\left\{
\begin{array}{c|l}
(x,F) & \begin{aligned}
x&\in E_{\nu},\quad F=(0=V_0\subset V_1\subset\cdots\subset V_d=V),\ x(F)\subset F,\ \dim(V_s/V_{s-1})=v_s,\\
x|_{V_j/V_i}&\in Y(c,m),\quad x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\leq s\leq i,\ j+1\leq s\leq d.
\end{aligned}
\end{array}
\right. ⟶ { ( x , F ) x x ∣ V j / V i ∈ E ν , F = ( 0 = V 0 ⊂ V 1 ⊂ ⋯ ⊂ V d = V ) , x ( F ) ⊂ F , dim ( V s / V s − 1 ) = v s , ∈ Y ( c , m ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , j + 1 ≤ s ≤ d .
To prove the claim, we will using the induction on ν ~ \tilde{\nu} ν ~ . It is clearly that when ν ~ = 0 \tilde{\nu}=0 ν ~ = 0 , the proof is trivial.
To prove for general ν ~ \tilde{\nu} ν ~ , first, we consider the following case.
Let P P P be an indecomposable preinjective module and I I I an indecomposable preinjective module, and the dimension of the representation vector space of P ⊕ m ⊕ I ⊕ n P^{\oplus m}\oplus I^{\oplus n} P ⊕ m ⊕ I ⊕ n is ν ~ \tilde{\nu} ν ~ .
{ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 1 / V i ≃ P ⊕ m , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . = { ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . \begin{aligned}
\left\{
\begin{array}{c|l}
(x,F) & \begin{aligned}
x&\in E_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i+1}\subset V_{i+2}\subset\cdots V_d=V),\ \dim(V_s/V_{s-1})=v_s,\\
x(F)&\subset F,\ x|_{V_{i+1}/V_i}\simeq P^{\oplus m},\quad x|_{V_{i+2}/V_{i+1}}\simeq I^{\oplus n},\\
x|_{V_s/V_{s-1}}&\in Y(c_s,m_s),\quad1\leq s\leq i,\ i+3\leq s\leq d.
\end{aligned}
\end{array}
\right. \\
&=
\left\{
\begin{array}{c|l}
(x,F) & \begin{aligned}
x&\in E_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i+1}\subset V_{i+2}\subset\cdots V_d=V),\ \dim(V_s/V_{s-1})=v_s,\\
x(F)&\subset F,\quad x|_{V_{i+2}/V_{i+1}}\simeq I^{\oplus n},\\
x|_{V_s/V_{s-1}}&\in Y(c_s,m_s),\quad1\leq s\leq i,\ i+3\leq s\leq d.
\end{aligned}
\end{array}
\right.
\end{aligned} ⎩ ⎨ ⎧ ( x , F ) x x ( F ) x ∣ V s / V s − 1 ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , dim ( V s / V s − 1 ) = v s , ⊂ F , x ∣ V i + 1 / V i ≃ P ⊕ m , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n , ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . = ⎩ ⎨ ⎧ ( x , F ) x x ( F ) x ∣ V s / V s − 1 ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , dim ( V s / V s − 1 ) = v s , ⊂ F , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n , ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d .
− ∑ Y ( c ′ , m ′ ) ≠ Y ( P ⊕ m , 0 ) { ( x ∈ E ν , F ) F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 1 / V i ∈ Y ( c ′ , m ′ ) , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . -
\sum_{\substack{Y(c',m')\ne Y(P^{\oplus m},0)}}
\left\{
\begin{array}{c|l}
(x\in E_{\nu},F) & \begin{aligned}
F&=(0=V_0\subset\cdots\subset V_i\subset V_{i+1}\subset V_{i+2}\subset\cdots V_d=V),\ \dim(V_s/V_{s-1})=v_s,\\
x(F)&\subset F,\quad x|_{V_{i+1}/V_i}\in Y(c',m'),\\
x|_{V_{i+2}/V_{i+1}}&\simeq I^{\oplus n},\\
x|_{V_s/V_{s-1}}&\in Y(c_s,m_s),\quad1\leq s\leq i,\ i+3\leq s\leq d.
\end{aligned}
\end{array}
\right. − Y ( c ′ , m ′ ) = Y ( P ⊕ m , 0 ) ∑ ⎩ ⎨ ⎧ ( x ∈ E ν , F ) F x ( F ) x ∣ V i + 2 / V i + 1 x ∣ V s / V s − 1 = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , dim ( V s / V s − 1 ) = v s , ⊂ F , x ∣ V i + 1 / V i ∈ Y ( c ′ , m ′ ) , ≃ I ⊕ n , ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d .
Moreover, for any M M M fitting into the exact sequence
0 ⟶ P ⊕ m ⟶ M ⟶ I ⊕ n ⟶ 0 , 0\longrightarrow P^{\oplus m}\longrightarrow M\longrightarrow I^{\oplus n}\longrightarrow0, 0 ⟶ P ⊕ m ⟶ M ⟶ I ⊕ n ⟶ 0 ,
M M M has no preinjective direct summand > I >I > I , and M M M cannot be of the form I ⊕ l I^{\oplus l} I ⊕ l .
Thus, we only need to consider
{ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . \left\{
\begin{array}{c|l}
(x,F) & \begin{aligned}
x&\in E_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i+1}\subset V_{i+2}\subset\cdots V_d=V),\ \dim(V_s/V_{s-1})=v_s,\\
x(F)&\subset F,\\
x|_{V_{i+2}/V_i}&\in Y(c_r,m_r),\quad x|_{V_{i+2}/V_{i+1}}\simeq I^{\oplus n},\\
x|_{V_s/V_{s-1}}&\in Y(c_s,m_s),\quad1\leq s\leq i,\ i+3\leq s\leq d.
\end{aligned}
\end{array}
\right. ⎩ ⎨ ⎧ ( x , F ) x x ( F ) x ∣ V i + 2 / V i x ∣ V s / V s − 1 ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , dim ( V s / V s − 1 ) = v s , ⊂ F , ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n , ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d .
and
{ ( x ∈ E ν , F ) F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 1 / V i ∈ Y ( c r ′ , m r ′ ) , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . (24) \left\{
\begin{array}{c|l}
(x\in E_{\nu},F) & \begin{aligned}
F&=(0=V_0\subset\cdots\subset V_i\subset V_{i+1}\subset V_{i+2}\subset\cdots V_d=V),\ \dim(V_s/V_{s-1})=v_s,\\
x(F)&\subset F,\\
x|_{V_{i+2}/V_i}&\in Y(c_r,m_r),\quad x|_{V_{i+1}/V_i}\in Y(c'_r,m'_r),\\
x|_{V_{i+2}/V_{i+1}}&\simeq I^{\oplus n},\\
x|_{V_s/V_{s-1}}&\in Y(c_s,m_s),\quad1\leq s\leq i,\ i+3\leq s\leq d.
\end{aligned}
\end{array}
\right. \tag*{(24)} ⎩ ⎨ ⎧ ( x ∈ E ν , F ) F x ( F ) x ∣ V i + 2 / V i x ∣ V i + 2 / V i + 1 x ∣ V s / V s − 1 = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , dim ( V s / V s − 1 ) = v s , ⊂ F , ∈ Y ( c r , m r ) , x ∣ V i + 1 / V i ∈ Y ( c r ′ , m r ′ ) , ≃ I ⊕ n , ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . (24)
where neither c i , l c_{i,l} c i , l nor c i , l ′ c'_{i,l} c i , l ′ has a preinjective summand > I >I > I . Moreover, for each ( c , m ) ∈ G ~ ν (c,m)\in\widetilde{\mathcal{G}}_{\nu} ( c , m ) ∈ G ν , we have the following: ∀ x ∈ Y ( c , m ) \forall x\in Y(c,m) ∀ x ∈ Y ( c , m ) , the dimension dim Hom k Q ( M x , I ) \dim\operatorname{Hom}_{kQ}(M_x,I) dim Hom k Q ( M x , I ) is independent of the choice of x ∈ Y ( c , m ) x\in Y(c,m) x ∈ Y ( c , m ) , and we denote this constant by s c , m s_{c,m} s c , m . Here the mapping rank of an element f ∈ Hom k Q ( M x , I ⊕ n ) f\in\operatorname{Hom}_{kQ}(M_x,I^{\oplus n}) f ∈ Hom k Q ( M x , I ⊕ n ) is the dimension of the vector space spanned by { f 1 , ⋯ , f n } \{f_1,\cdots,f_n\} { f 1 , ⋯ , f n } , where f = ( f 1 , ⋯ , f n ) f=(f_1,\cdots,f_n) f = ( f 1 , ⋯ , f n ) , f s ∈ Hom k Q ( M x , I ) f_s\in\operatorname{Hom}_{kQ}(M_x,I) f s ∈ Hom k Q ( M x , I ) , s = 1 , ⋯ , n s=1,\cdots,n s = 1 , ⋯ , n . And there is a natural action Aut k Q ( I ⊕ n ) \operatorname{Aut}_{kQ}(I^{\oplus n}) Aut k Q ( I ⊕ n ) acting on Hom k Q ( M x , I ⊕ n ) \operatorname{Hom}_{kQ}(M_x,I^{\oplus n}) Hom k Q ( M x , I ⊕ n ) .
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , dim ( V s / V s − 1 ) = ν s x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d } = { ( x , F , f ) | x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , f ∈ Hom ( M x ∣ V i + 2 / V i , I ⊕ n ) is surjective x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d } / G L n = { ( x , F , f ) | x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , f ∈ Hom ( M x ∣ V i + 2 / V i , I ⊕ n ) has mapping rank n x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d } / G L n \begin{aligned}
\left\{(x,F)\ \middle|\
\begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i+1}\subset V_{i+2}\subset\cdots\subset V_d=V),\ \dim(V_s/V_{s-1})=\nu_s\\
&x(F)\subset F,\\
&x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\quad x|_{V_{i+2}/V_{i+1}}\simeq I^{\oplus n}\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\leq s\leq i,\ i+3\leq s\leq d
\end{aligned}
\right\} \\
&=
\left\{(x,F,f)\ \middle|\
\begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i+2}\subset\cdots\subset V_d=V),\\
&1\leq s\leq i,\ i+3\leq s\leq d,\ \dim(V_s/V_{s-1})=\nu_s,\ x(F)\subset F,\\
&x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\\
&f\in\operatorname{Hom}(M_{x|_{V_{i+2}/V_i}},I^{\oplus n})\text{ is surjective}\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\leq s\leq i,\ i+3\leq s\leq d
\end{aligned}
\right\}/\mathrm{GL}_n \\
&=
\left\{(x,F,f)\ \middle|\
\begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i+2}\subset\cdots\subset V_d=V),\\
&1\leq s\leq i,\ i+3\leq s\leq d,\ \dim(V_s/V_{s-1})=\nu_s,\ x(F)\subset F,\\
&x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\\
&f\in\operatorname{Hom}(M_{x|_{V_{i+2}/V_i}},I^{\oplus n})\text{ has mapping rank }n\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\leq s\leq i,\ i+3\leq s\leq d
\end{aligned}
\right\}/\mathrm{GL}_n
\end{aligned} ⎩ ⎨ ⎧ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , dim ( V s / V s − 1 ) = ν s x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d ⎭ ⎬ ⎫ = ⎩ ⎨ ⎧ ( x , F , f ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , f ∈ Hom ( M x ∣ V i + 2 / V i , I ⊕ n ) is surjective x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d ⎭ ⎬ ⎫ / GL n = ⎩ ⎨ ⎧ ( x , F , f ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , f ∈ Hom ( M x ∣ V i + 2 / V i , I ⊕ n ) has mapping rank n x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d ⎭ ⎬ ⎫ / GL n
− { ( x , F , f ) | x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , f ∈ Hom ( M x ∣ V i + 2 / V i , I ⊕ n ) has mapping rank n , ker f = V i , 1 , dim ( V i + 2 / V i , 1 ) < dim ( I ⊕ n ) x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d } / G L n . -
\left\{(x,F,f)\ \middle|\
\begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i,1}\subset V_{i+2}\subset\cdots\subset V_d=V),\\
&1\leq s\leq i,\ i+3\leq s\leq d,\ \dim(V_s/V_{s-1})=\nu_s,\ x(F)\subset F,\\
&x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\\
&f\in\operatorname{Hom}(M_{x|_{V_{i+2}/V_i}},I^{\oplus n})\text{ has mapping rank }n,\\
&\ker f=V_{i,1},\quad\dim(V_{i+2}/V_{i,1})<\dim(I^{\oplus n})\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\leq s\leq i,\ i+3\leq s\leq d
\end{aligned}
\right\}/\mathrm{GL}_n. − ⎩ ⎨ ⎧ ( x , F , f ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , f ∈ Hom ( M x ∣ V i + 2 / V i , I ⊕ n ) has mapping rank n , ker f = V i , 1 , dim ( V i + 2 / V i , 1 ) < dim ( I ⊕ n ) x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d ⎭ ⎬ ⎫ / GL n .
Fixing the dimension of ker f \ker f ker f , we obtain
∐ Y ( c r 1 , m r 1 ) { ( x , F , f ) | x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , f ∈ Hom ( M x ∣ V i + 2 / V i , I ⊕ n ) has mapping rank n , ker f = V i , 1 , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d } / G L n , \coprod_{Y(c_r^1,m_r^1)}
\left\{(x,F,f)\ \middle|\
\begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i,1}\subset V_{i+2}\subset\cdots\subset V_d=V),\\
&1\leq s\leq i,\ i+3\leq s\leq d,\ \dim(V_s/V_{s-1})=\nu_s,\ x(F)\subset F,\\
&f\in\operatorname{Hom}(M_{x|_{V_{i+2}/V_i}},I^{\oplus n})\text{ has mapping rank }n,\\
&\ker f=V_{i,1},\quad x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\quad x|_{V_{i+2}/V_{i,1}}\in Y(c_r^1,m_r^1)\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\leq s\leq i,\ i+3\leq s\leq d
\end{aligned}
\right\}/\mathrm{GL}_n, Y ( c r 1 , m r 1 ) ∐ ⎩ ⎨ ⎧ ( x , F , f ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , f ∈ Hom ( M x ∣ V i + 2 / V i , I ⊕ n ) has mapping rank n , ker f = V i , 1 , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d ⎭ ⎬ ⎫ / GL n ,
where c r 1 c_r^1 c r 1 has no preinjective direct summand > I >I > I , since M x ∣ V i + 2 / V i , 1 M_{x|_{V_{i+2}/V_{i,1}}} M x ∣ V i + 2 / V i , 1 is a submodule of I ⊕ n I^{\oplus n} I ⊕ n .
Thus, we only need to consider
{ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , ( x , F , f ) f ∈ Hom ( M x ∣ V i + 2 / V i , 1 , I ⊕ n ) has mapping rank n and is injective , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . } / G L n , \left\{
\begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i,1}\subset V_{i+2}\subset\cdots V_d=V),\\
&1\le s\le i,\ i+3\le s\le d,\ \dim(V_s/V_{s-1})=v_s,\ x(F)\subset F,\\
&(x,F,f)\quad f\in\operatorname{Hom}(M_{x|_{V_{i+2}/V_{i,1}}},I^{\oplus n})\text{ has mapping rank }n\text{ and is injective},\\
&x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\quad x|_{V_{i+2}/V_{i,1}}\in Y(c_r^1,m_r^1)\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\le s\le i,\ i+3\le s\le d.
\end{aligned}
\right\}/\mathrm{GL}_n, ⎩ ⎨ ⎧ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , ( x , F , f ) f ∈ Hom ( M x ∣ V i + 2 / V i , 1 , I ⊕ n ) has mapping rank n and is injective , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . ⎭ ⎬ ⎫ / GL n ,
which is isomorphic to
{ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , ( x , F , f ) f ∈ Hom ( M x ∣ V i + 2 / V i , 1 , I ⊕ n ) has mapping rank n , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . } / G L n \left\{
\begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i,1}\subset V_{i+2}\subset\cdots V_d=V),\\
&1\le s\le i,\ i+3\le s\le d,\ \dim(V_s/V_{s-1})=v_s,\ x(F)\subset F,\\
&(x,F,f)\quad f\in\operatorname{Hom}(M_{x|_{V_{i+2}/V_{i,1}}},I^{\oplus n})\text{ has mapping rank }n,\\
&x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\quad x|_{V_{i+2}/V_{i,1}}\in Y(c_r^1,m_r^1)\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\le s\le i,\ i+3\le s\le d.
\end{aligned}
\right\}/\mathrm{GL}_n ⎩ ⎨ ⎧ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i + 2 ⊂ ⋯ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , ( x , F , f ) f ∈ Hom ( M x ∣ V i + 2 / V i , 1 , I ⊕ n ) has mapping rank n , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . ⎭ ⎬ ⎫ / GL n
− ⊔ Y ( c r 2 , m r 2 ) { x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i , 2 ⊂ V i + 2 ⊂ ⋯ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , ( x , F , f ) f ∈ Hom ( M x ∣ V i + 2 / V i , 2 , I ⊕ n ) is injective and has mapping rank n , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) , x ∣ V i + 2 / V i , 2 ∈ Y ( c r 2 , m r 2 ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . } / G L n . -\underset{Y(c_r^2,m_r^2)}{\sqcup}
\left\{
\begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i,1}\subset V_{i,2}\subset V_{i+2}\subset\cdots V_d=V),\\
&1\le s\le i,\ i+3\le s\le d,\ \dim(V_s/V_{s-1})=v_s,\ x(F)\subset F,\\
&(x,F,f)\quad f\in\operatorname{Hom}(M_{x|_{V_{i+2}/V_{i,2}}},I^{\oplus n})\text{ is injective and has mapping rank }n,\\
&x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\quad x|_{V_{i+2}/V_{i,1}}\in Y(c_r^1,m_r^1),\\
&x|_{V_{i+2}/V_{i,2}}\in Y(c_r^2,m_r^2),\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\le s\le i,\ i+3\le s\le d.
\end{aligned}
\right\}/\mathrm{GL}_n. − Y ( c r 2 , m r 2 ) ⊔ ⎩ ⎨ ⎧ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i , 2 ⊂ V i + 2 ⊂ ⋯ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , ( x , F , f ) f ∈ Hom ( M x ∣ V i + 2 / V i , 2 , I ⊕ n ) is injective and has mapping rank n , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) , x ∣ V i + 2 / V i , 2 ∈ Y ( c r 2 , m r 2 ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . ⎭ ⎬ ⎫ / GL n .
Again, c r 2 c_r^2 c r 2 has no preinjective direct summand > I >I > I .
By induction, we only need to consider
{ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i , 2 ⊂ ⋯ V i , t ⊂ V i + 2 ⊂ ⋯ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , ( x , F , f ) x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) , … , x ∣ V i + 2 / V i , t ∈ Y ( c r t , m r t ) , f ∈ Hom ( M x ∣ V i + 2 / V i , t , I ⊕ n ) has mapping rank n , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . } / G L n ⟶ E ν , \left\{
\begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i,1}\subset V_{i,2}\subset\cdots V_{i,t}\subset V_{i+2}\subset\cdots V_d=V),\\
&1\le s\le i,\ i+3\le s\le d,\ \dim(V_s/V_{s-1})=v_s,\ x(F)\subset F,\\
&(x,F,f)\quad x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\quad x|_{V_{i+2}/V_{i,1}}\in Y(c_r^1,m_r^1),\quad\ldots,\quad x|_{V_{i+2}/V_{i,t}}\in Y(c_r^t,m_r^t),\\
&f\in\operatorname{Hom}(M_{x|_{V_{i+2}/V_{i,t}}},I^{\oplus n})\text{ has mapping rank }n,\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\le s\le i,\ i+3\le s\le d.
\end{aligned}
\right\}/\mathrm{GL}_n\longrightarrow\mathbf{E}_{\nu}, ⎩ ⎨ ⎧ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i , 2 ⊂ ⋯ V i , t ⊂ V i + 2 ⊂ ⋯ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , ( x , F , f ) x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) , … , x ∣ V i + 2 / V i , t ∈ Y ( c r t , m r t ) , f ∈ Hom ( M x ∣ V i + 2 / V i , t , I ⊕ n ) has mapping rank n , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . ⎭ ⎬ ⎫ / GL n ⟶ E ν ,
where the representations in Y ( c r u , m r u ) Y(c_r^u,m_r^u) Y ( c r u , m r u ) have no preinjective direct summand > I >I > I . Moreover, Y ( c r u , m r u ) Y(c_r^u,m_r^u) Y ( c r u , m r u ) is not of type Y ( c I ⊕ k , 0 ) Y(c_{I\oplus k},0) Y ( c I ⊕ k , 0 ) . Indeed, f f f has mapping rank n n n , and the mapping rank associated with Y ( c I ⊕ k , 0 ) Y(c_{I\oplus k},0) Y ( c I ⊕ k , 0 ) can not be larger than k < n k<n k < n when u ≥ 1 u\ge1 u ≥ 1 .
By the lemma 3.1, we form the morphism
{ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i , 2 ⊂ ⋯ V i , t ⊂ V i + 2 ⊂ ⋯ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , ( x , F , f ) x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) , … , x ∣ V i + 2 / V i , t ∈ Y ( c r t , m r t ) , f ∈ Hom ( M x ∣ V i + 2 / V i , t , I ⊕ n ) has mapping rank n , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . } / G L n \left\{
\begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i,1}\subset V_{i,2}\subset\cdots V_{i,t}\subset V_{i+2}\subset\cdots V_d=V),\\
&1\le s\le i,\ i+3\le s\le d,\ \dim(V_s/V_{s-1})=v_s,\ x(F)\subset F,\\
&(x,F,f)\quad x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\quad x|_{V_{i+2}/V_{i,1}}\in Y(c_r^1,m_r^1),\quad\ldots,\quad x|_{V_{i+2}/V_{i,t}}\in Y(c_r^t,m_r^t),\\
&f\in\operatorname{Hom}(M_{x|_{V_{i+2}/V_{i,t}}},I^{\oplus n})\text{ has mapping rank }n,\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\le s\le i,\ i+3\le s\le d.
\end{aligned}
\right\}/\mathrm{GL}_n ⎩ ⎨ ⎧ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i , 2 ⊂ ⋯ V i , t ⊂ V i + 2 ⊂ ⋯ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , ( x , F , f ) x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) , … , x ∣ V i + 2 / V i , t ∈ Y ( c r t , m r t ) , f ∈ Hom ( M x ∣ V i + 2 / V i , t , I ⊕ n ) has mapping rank n , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . ⎭ ⎬ ⎫ / GL n
→ ϕ { x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i , 2 ⊂ ⋯ V i , t ⊂ V i + 2 ⊂ ⋯ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , ( x , F ) x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) , … , x ∣ V i + 2 / V i , t ∈ Y ( c r t , m r t ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . } . \xrightarrow{\ \phi\ }
\left\{
\begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_0\subset\cdots\subset V_i\subset V_{i,1}\subset V_{i,2}\subset\cdots V_{i,t}\subset V_{i+2}\subset\cdots V_d=V),\\
&1\le s\le i,\ i+3\le s\le d,\ \dim(V_s/V_{s-1})=v_s,\ x(F)\subset F,\\
&(x,F)\quad x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\quad x|_{V_{i+2}/V_{i,1}}\in Y(c_r^1,m_r^1),\quad\ldots,\quad x|_{V_{i+2}/V_{i,t}}\in Y(c_r^t,m_r^t),\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\le s\le i,\ i+3\le s\le d.
\end{aligned}
\right\}. ϕ ⎩ ⎨ ⎧ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i , 2 ⊂ ⋯ V i , t ⊂ V i + 2 ⊂ ⋯ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , ( x , F ) x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) , … , x ∣ V i + 2 / V i , t ∈ Y ( c r t , m r t ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . ⎭ ⎬ ⎫ .
we have
ϕ i Q ‾ l = ⨁ for some k Q ‾ l [ 2 k ] ( k ) . \phi_i\overline{Q}_l=\bigoplus_{\text{for some }k}\overline{Q}_l[2k](k). ϕ i Q l = for some k ⨁ Q l [ 2 k ] ( k ) .
Thus, we only need to consider
{ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i , 2 ⊂ ⋯ ⊂ V i , t ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , ( x , F ) x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) , … , x ∣ V i + 2 / V i , t ∈ Y ( c r t , m r t ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . \left\{
\begin{array}{ll}
x\in E_\nu,&F=(0=V_0\subset\cdots\subset V_i\subset V_{i,1}\subset V_{i,2}\subset\cdots\subset V_{i,t}\subset V_{i+2}\subset\cdots\subset V_d=V),\\
&1\le s\le i,\ i+3\le s\le d,\ \dim(V_s/V_{s-1})=\nu_s,\ x(F)\subset F,\\
(x,F)&x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\quad x|_{V_{i+2}/V_{i,1}}\in Y(c_r^1,m_r^1),\ \ldots,\ x|_{V_{i+2}/V_{i,t}}\in Y(c_r^t,m_r^t),\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\le s\le i,\ i+3\le s\le d.
\end{array}
\right. ⎩ ⎨ ⎧ x ∈ E ν , ( x , F ) F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i , 2 ⊂ ⋯ ⊂ V i , t ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i , 1 ∈ Y ( c r 1 , m r 1 ) , … , x ∣ V i + 2 / V i , t ∈ Y ( c r t , m r t ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d .
Actually, in this case, any M x ∣ V i , 1 / V i − 1 M_{x|_{V_{i,1}/V_{i-1}}} M x ∣ V i , 1 / V i − 1 including M x ∣ V i , 1 / V i M_{x|_{V_{i,1}/V_i}} M x ∣ V i , 1 / V i has no projective indecomposable direct summand > I >I > I , since M x ∣ V i , 1 / V i − 1 M_{x|_{V_{i,1}/V_{i-1}}} M x ∣ V i , 1 / V i − 1 is a submodule of M x ∣ V i + 2 / V i − 1 M_{x|_{V_{i+2}/V_{i-1}}} M x ∣ V i + 2 / V i − 1 .
When x ∣ V i + 1 / V i ∈ Y ( 0 , m ) x|_{V_{i+1}/V_i}\in Y(0,m) x ∣ V i + 1 / V i ∈ Y ( 0 , m ) , the same argument applies. In the other cases, only finitely many modules, up to isomorphism, can occur as extensions of M x ∣ V i + 1 / V i M_{x|_{V_{i+1}/V_i}} M x ∣ V i + 1 / V i and I ⊕ n I^{\oplus n} I ⊕ n . Moreover, for any G ν G_\nu G ν -orbit O ⊂ E ν \mathcal{O}\subset E_\nu O ⊂ E ν , and any F ∈ D m , G ν b ( O ) \mathcal{F}\in D^b_{m,G_\nu}(\mathcal{O}) F ∈ D m , G ν b ( O ) , the sheaf H i ( F ) \mathcal{H}^i(\mathcal{F}) H i ( F ) is generated by sheaves of the form E ω , s \mathcal{E}_{\omega,s} E ω , s , for some ω , s \omega,s ω , s . And if M x ∣ V i + 1 / V i ≮ I M_{x|_{V_{i+1}/V_i}}\not< I M x ∣ V i + 1 / V i < I then the extension of them has no projective direct summands > I >I > I . All decompositions used here come from filtrations by closed subvarieties. These filtrations are induced by those associated with ⋃ Y ( c , m ) \bigcup Y(c,m) ⋃ Y ( c , m ) .
By induction on the dimension ν ~ \widetilde{\nu} ν , we only need to consider
{ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i , 1 I ⊂ V i , 1 n h ⊂ V i , 1 h r ⊂ V i , 1 h r , 1 ⊂ ⋯ ⊂ V i , 1 h r , t − 1 ⊂ V i P ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , ( x , F ) x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i P ∈ Y ( ( c r 1 ) P , 0 ) , x ∣ V i , 1 I / V i , 1 ∈ Y ( ( c r 1 ) I , 0 ) , denote V i , 1 h r , 0 = V i , 1 n h , V i , 1 h r , t = V i , 1 P , x ∣ V i , 1 n h / V i , 1 I ∈ Y ( ( c r 1 ) n h , 0 ) , x ∣ V i , 1 h r , l / V i , 1 h r , l − 1 ∈ Y ( 0 , λ l ) , l = 1 , … , t , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . \left\{
\begin{array}{ll}
x\in E_\nu,&\\
F=(0=V_0\subset\cdots\subset V_i\subset V_{i,1}\subset V_{i,1}^{I}\subset V_{i,1}^{nh}\subset V_{i,1}^{hr}\subset V_{i,1}^{hr,1}\subset\cdots\subset V_{i,1}^{hr,t-1}\subset V_i^P\subset V_{i+2}\subset\cdots\subset V_d=V),&\\
&1\le s\le i,\ i+3\le s\le d,\ \dim(V_s/V_{s-1})=\nu_s,\ x(F)\subset F,\\
(x,F)&x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\quad x|_{V_{i+2}/V_i^P}\in Y((c_r^1)_P,0),\quad x|_{V_{i,1}^{I}/V_{i,1}}\in Y((c_r^1)_I,0),\\
&\text{denote }V_{i,1}^{hr,0}=V_{i,1}^{nh},\ V_{i,1}^{hr,t}=V_{i,1}^{P},\\
&x|_{V_{i,1}^{nh}/V_{i,1}^{I}}\in Y((c_r^1)_{nh},0),\ x|_{V_{i,1}^{hr,l}/V_{i,1}^{hr,l-1}}\in Y(0,\lambda_l),\ l=1,\ldots,t,\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\le s\le i,\ i+3\le s\le d.
\end{array}
\right. ⎩ ⎨ ⎧ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i , 1 ⊂ V i , 1 I ⊂ V i , 1 nh ⊂ V i , 1 h r ⊂ V i , 1 h r , 1 ⊂ ⋯ ⊂ V i , 1 h r , t − 1 ⊂ V i P ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , ( x , F ) 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V i + 2 / V i P ∈ Y (( c r 1 ) P , 0 ) , x ∣ V i , 1 I / V i , 1 ∈ Y (( c r 1 ) I , 0 ) , denote V i , 1 h r , 0 = V i , 1 nh , V i , 1 h r , t = V i , 1 P , x ∣ V i , 1 nh / V i , 1 I ∈ Y (( c r 1 ) nh , 0 ) , x ∣ V i , 1 h r , l / V i , 1 h r , l − 1 ∈ Y ( 0 , λ l ) , l = 1 , … , t , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d .
since the case in equation (24)
{ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , ( x , F ) x ∣ V i + 1 / V i ∈ Y ( c r ′ , m r ′ ) , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n . x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . ≃ \begin{aligned}
\left\{
\begin{array}{ll}
x\in E_\nu,&\\
F=(0=V_0\subset\cdots\subset V_i\subset V_{i+1}\subset V_{i+2}\subset\cdots\subset V_d=V),\ \dim(V_s/V_{s-1})=\nu_s,\ x(F)\subset F,&\\
(x,F)&x|_{V_{i+1}/V_i}\in Y(c_r',m_r'),\\
&x|_{V_{i+2}/V_{i+1}}\simeq I^{\oplus n}.x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\le s\le i,\ i+3\le s\le d.
\end{array}
\right. \\
&\simeq
\end{aligned} ⎩ ⎨ ⎧ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , ( x , F ) x ∣ V i + 1 / V i ∈ Y ( c r ′ , m r ′ ) , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n . x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . ≃
{ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i I ⊂ V i n h ⊂ V i h r ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , ( x , F ) x ∣ V i + 1 / V i h r ∈ Y ( ( c r ′ ) P , 0 ) , x ∣ V i h r / V i n h ∈ Y ( 0 , m r ′ ) / , x ∣ V i n h / V i I ∈ Y ( ( c r ′ ) n h , 0 ) , x ∣ V i I / V i ∈ Y ( ( c r ′ ) I , 0 ) , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n . x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . \left\{
\begin{array}{ll}
x\in E_\nu,&\\
F=(0=V_0\subset\cdots\subset V_i\subset V_i^I\subset V_i^{nh}\subset V_i^{hr}\subset V_{i+1}\subset V_{i+2}\subset\cdots\subset V_d=V),&\\
&\dim(V_s/V_{s-1})=\nu_s,\ x(F)\subset F,\\
(x,F)&x|_{V_{i+1}/V_i^{hr}}\in Y((c_r')_P,0),\ x|_{V_i^{hr}/V_i^{nh}}\in Y(0,m_r')/,\\
&x|_{V_i^{nh}/V_i^I}\in Y((c_r')_{nh},0),\ x|_{V_i^I/V_i}\in Y((c_r')_I,0),\\
&x|_{V_{i+2}/V_{i+1}}\simeq I^{\oplus n}.x|_{V_{i+2}/V_i}\in Y(c_r,m_r),\\
&x|_{V_s/V_{s-1}}\in Y(c_s,m_s),\quad1\le s\le i,\ i+3\le s\le d.
\end{array}
\right. ⎩ ⎨ ⎧ x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i I ⊂ V i nh ⊂ V i h r ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , ( x , F ) dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , x ∣ V i + 1 / V i h r ∈ Y (( c r ′ ) P , 0 ) , x ∣ V i h r / V i nh ∈ Y ( 0 , m r ′ ) / , x ∣ V i nh / V i I ∈ Y (( c r ′ ) nh , 0 ) , x ∣ V i I / V i ∈ Y (( c r ′ ) I , 0 ) , x ∣ V i + 2 / V i + 1 ≃ I ⊕ n . x ∣ V i + 2 / V i ∈ Y ( c r , m r ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d .
can also be reduced to the form like above when applying the claim.
Neither c r 1 c_r^1 c r 1 nor c r ′ c_r' c r ′ has a projective direct summand > I >I > I , and neither Y ( c r 1 , m r 1 ) Y(c_r^1,m_r^1) Y ( c r 1 , m r 1 ) nor Y ( c r ′ , m r ′ ) Y(c_r',m_r') Y ( c r ′ , m r ′ ) is of type Y ( c I ⊕ k , 0 ) Y(c_{I\oplus k},0) Y ( c I ⊕ k , 0 ) . After finitely many steps, the rightmost term is therefore replaced by a representation strictly smaller than I I I , while no injective representation strictly larger than I I I occurs in the process. There are only finitely many nonisomorphic indecomposable preinjective modules I ′ < I I' < I I ′ < I , with dimension ≤ ν ~ \leq\tilde{\nu} ≤ ν ~ . Thus, after finitely many such steps, all preinjective parts are on the left.
Moreover,
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 1 / V i ∈ Y ( c n h , a , 0 ) , x ∣ V i + 2 / V i + 1 ∈ Y ( 0 , m ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d } \left\{(x,F)\ \middle|\ \begin{aligned}
&x \in E_{\nu},\quad F=(0=V_{0}\subset\cdots\subset V_{i}\subset V_{i+1}\subset V_{i+2}\subset\cdots\subset V_{d}=V),\\
&1\leq s\leq i,\ i+3\leq s\leq d,\ \dim(V_{s}/V_{s-1})=v_{s},\ x(F)\subset F,\\
&x|_{V_{i+1}/V_{i}}\in Y(c_{nh,a},0),\quad x|_{V_{i+2}/V_{i+1}}\in Y(0,m),\\
&x|_{V_{s}/V_{s-1}}\in Y(c_{s},m_{s}),\quad1\leq s\leq i,\ i+3\leq s\leq d
\end{aligned}\right\} ⎩ ⎨ ⎧ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 1 / V i ∈ Y ( c nh , a , 0 ) , x ∣ V i + 2 / V i + 1 ∈ Y ( 0 , m ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d ⎭ ⎬ ⎫
is isomorphic to
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 1 / V i ∈ Y ( 0 , m ) , x ∣ V i + 2 / V i + 1 ∈ Y ( c n h , a , 0 ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d } . \left\{(x,F)\ \middle|\ \begin{aligned}
&x \in E_{\nu},\quad F=(0=V_{0}\subset\cdots\subset V_{i}\subset V_{i+1}\subset V_{i+2}\subset\cdots\subset V_{d}=V),\\
&1\leq s\leq i,\ i+3\leq s\leq d,\ \dim(V_{s}/V_{s-1})=v_{s},\ x(F)\subset F,\\
&x|_{V_{i+1}/V_{i}}\in Y(0,m),\quad x|_{V_{i+2}/V_{i+1}}\in Y(c_{nh,a},0),\\
&x|_{V_{s}/V_{s-1}}\in Y(c_{s},m_{s}),\quad1\leq s\leq i,\ i+3\leq s\leq d
\end{aligned}\right\}. ⎩ ⎨ ⎧ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 1 / V i ∈ Y ( 0 , m ) , x ∣ V i + 2 / V i + 1 ∈ Y ( c nh , a , 0 ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d ⎭ ⎬ ⎫ .
and for i ≠ j i\neq j i = j ,
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 1 / V i ∈ Y ( c n h , u , 0 ) , x ∣ V i + 2 / V i + 1 ∈ Y ( c n h , r , 0 ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d } \left\{(x,F)\ \middle|\ \begin{aligned}
&x \in E_{\nu},\quad F=(0=V_{0}\subset\cdots\subset V_{i}\subset V_{i+1}\subset V_{i+2}\subset\cdots\subset V_{d}=V),\\
&1\leq s\leq i,\ i+3\leq s\leq d,\ \dim(V_{s}/V_{s-1})=v_{s},\ x(F)\subset F,\\
&x|_{V_{i+1}/V_{i}}\in Y(c_{nh,u},0),\quad x|_{V_{i+2}/V_{i+1}}\in Y(c_{nh,r},0),\\
&x|_{V_{s}/V_{s-1}}\in Y(c_{s},m_{s}),\quad1\leq s\leq i,\ i+3\leq s\leq d
\end{aligned}\right\} ⎩ ⎨ ⎧ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 1 / V i ∈ Y ( c nh , u , 0 ) , x ∣ V i + 2 / V i + 1 ∈ Y ( c nh , r , 0 ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d ⎭ ⎬ ⎫
is isomorphic to
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 1 / V i ∈ Y ( c n h , r , 0 ) , x ∣ V i + 2 / V i + 1 ∈ Y ( c n h , u , 0 ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d } . \left\{(x,F)\ \middle|\ \begin{aligned}
&x \in E_{\nu},\quad F=(0=V_{0}\subset\cdots\subset V_{i}\subset V_{i+1}\subset V_{i+2}\subset\cdots\subset V_{d}=V),\\
&1\leq s\leq i,\ i+3\leq s\leq d,\ \dim(V_{s}/V_{s-1})=v_{s},\ x(F)\subset F,\\
&x|_{V_{i+1}/V_{i}}\in Y(c_{nh,r},0),\quad x|_{V_{i+2}/V_{i+1}}\in Y(c_{nh,u},0),\\
&x|_{V_{s}/V_{s-1}}\in Y(c_{s},m_{s}),\quad1\leq s\leq i,\ i+3\leq s\leq d
\end{aligned}\right\}. ⎩ ⎨ ⎧ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 1 / V i ∈ Y ( c nh , r , 0 ) , x ∣ V i + 2 / V i + 1 ∈ Y ( c nh , u , 0 ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d ⎭ ⎬ ⎫ .
The argument for the preprojective part is analogous.
Hence after finite step, we only need to consider the case, for some ( λ 1 , … , λ t ) (\lambda_{1},\ldots,\lambda_{t}) ( λ 1 , … , λ t ) , ∑ s = 1 t λ s = m r \sum_{s=1}^{t}\lambda_{s}=m_{r} ∑ s = 1 t λ s = m r ,
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V ′ = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , l = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , t ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i / V i − 1 ≃ M c r 1 , x ∣ V 2 , s / V 2 , s − 1 ≃ M c r n h , s , x ∣ V 3 , s / V 3 , s − 1 is homogeneous regular of dimension λ s δ 0 , x ∣ V j / V 3 , t ≃ M c r p , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , j + 1 ≤ s ≤ d } , \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in E_{\nu},\\
&F=(0=V_{0}\subset\cdots\subset V_{i}\subset V'=V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,l}=V_{3,0}\subset V_{3,1}\subset\cdots\subset V_{3,t}\subset V_{i+2}\subset\cdots\subset V_{d}=V),\\
&\dim(V_{s}/V_{s-1})=v_{s},\ x(F)\subset F,\ x|_{V_{i}/V_{i-1}}\simeq M_{cr_{1}},\ x|_{V_{2,s}/V_{2,s-1}}\simeq M_{cr_{nh,s}},\\
&x|_{V_{3,s}/V_{3,s-1}}\text{ is homogeneous regular of dimension }\lambda_{s}\delta_{0},\\
&x|_{V_{j}/V_{3,t}}\simeq M_{cr_{p}},\\
&x|_{V_{s}/V_{s-1}}\in Y(c_{s},m_{s}),\quad1\leq s\leq i,\ j+1\leq s\leq d
\end{aligned}\right\}, ⎩ ⎨ ⎧ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V ′ = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , l = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , t ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i / V i − 1 ≃ M c r 1 , x ∣ V 2 , s / V 2 , s − 1 ≃ M c r nh , s , x ∣ V 3 , s / V 3 , s − 1 is homogeneous regular of dimension λ s δ 0 , x ∣ V j / V 3 , t ≃ M c r p , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , j + 1 ≤ s ≤ d ⎭ ⎬ ⎫ ,
which is what we need.
And for general case
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c , m ) , x ∣ V i + 1 / V i ∈ Y ( c ′ , m ′ ) , x ∣ V i + 2 / V i + 1 ∈ Y ( c ′ ′ , m ′ ′ ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d } \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in E_{\nu},\quad F=(0=V_{0}\subset\cdots\subset V_{i}\subset V_{i+1}\subset V_{i+2}\subset\cdots\subset V_{d}=V),\\
&1\leq s\leq i,\ i+3\leq s\leq d,\ \dim(V_{s}/V_{s-1})=v_{s},\ x(F)\subset F,\\
&x|_{V_{i+2}/V_{i}}\in Y(c,m),\quad x|_{V_{i+1}/V_{i}}\in Y(c',m'),\quad x|_{V_{i+2}/V_{i+1}}\in Y(c'',m''),\\
&x|_{V_{s}/V_{s-1}}\in Y(c_{s},m_{s}),\quad1\leq s\leq i,\ i+3\leq s\leq d
\end{aligned}\right\} ⎩ ⎨ ⎧ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i + 1 ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = v s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c , m ) , x ∣ V i + 1 / V i ∈ Y ( c ′ , m ′ ) , x ∣ V i + 2 / V i + 1 ∈ Y ( c ′′ , m ′′ ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d ⎭ ⎬ ⎫
which is isomorphic to
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i I ⊂ V i ′ ⊂ V i n h ⊂ V i h r ⊂ V i + 1 ⊂ V i + 1 I ⊂ V i + 1 n h ⊂ V i + 1 h r ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c , m ) , x ∣ V i + 1 / V i h r ∈ Y ( c P ′ , 0 ) , x ∣ V i h r / V i n h ∈ Y ( 0 , m ′ ) , x ∣ V i n h / V i I ∈ Y ( c n h ′ , 0 ) , x ∣ V i I / V i ∈ Y ( c I ′ , 0 ) , x ∣ V i + 2 / V i + 1 h r ∈ Y ( c P ′ ′ , 0 ) , x ∣ V i + 1 h r / V i + 1 n h ∈ Y ( 0 , m ′ ′ ) , x ∣ V i + 1 n h / V i + 1 I ∈ Y ( c n h ′ ′ , 0 ) , x ∣ V i + 1 I / V i + 1 ∈ Y ( c I ′ ′ , 0 ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . } . \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_{0}\subset\cdots\subset V_{i}\subset V_{i}^{I}\subset V_{i}^{\prime}\subset V_{i}^{nh}\subset V_{i}^{hr}\subset V_{i+1}\subset V_{i+1}^{I}\subset V_{i+1}^{nh}\subset V_{i+1}^{hr}\subset V_{i+2}\subset\cdots\subset V_{d}=\mathbf{V}),\\
&1\leq s\leq i,\ i+3\leq s\leq d,\ \dim(V_{s}/V_{s-1})=\nu_{s},\ x(F)\subset F,\\
&x|_{V_{i+2}/V_{i}}\in Y(c,m),\\
&x|_{V_{i+1}/V_{i}^{hr}}\in Y(c_{P}^{\prime},0),\quad x|_{V_{i}^{hr}/V_{i}^{nh}}\in Y(0,m^{\prime}),\quad x|_{V_{i}^{nh}/V_{i}^{I}}\in Y(c_{nh}^{\prime},0),\quad x|_{V_{i}^{I}/V_{i}}\in Y(c_{I}^{\prime},0),\\
&x|_{V_{i+2}/V_{i+1}^{hr}}\in Y(c_{P}^{\prime\prime},0),\quad x|_{V_{i+1}^{hr}/V_{i+1}^{nh}}\in Y(0,m^{\prime\prime}),\quad x|_{V_{i+1}^{nh}/V_{i+1}^{I}}\in Y(c_{nh}^{\prime\prime},0),\quad x|_{V_{i+1}^{I}/V_{i+1}}\in Y(c_{I}^{\prime\prime},0),\\
&x|_{V_{s}/V_{s-1}}\in Y(c_{s},m_{s}),\quad1\leq s\leq i,\ i+3\leq s\leq d.
\end{aligned}\right\}. ⎩ ⎨ ⎧ ( x , F ) x ∈ E ν , F = ( 0 = V 0 ⊂ ⋯ ⊂ V i ⊂ V i I ⊂ V i ′ ⊂ V i nh ⊂ V i h r ⊂ V i + 1 ⊂ V i + 1 I ⊂ V i + 1 nh ⊂ V i + 1 h r ⊂ V i + 2 ⊂ ⋯ ⊂ V d = V ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d , dim ( V s / V s − 1 ) = ν s , x ( F ) ⊂ F , x ∣ V i + 2 / V i ∈ Y ( c , m ) , x ∣ V i + 1 / V i h r ∈ Y ( c P ′ , 0 ) , x ∣ V i h r / V i nh ∈ Y ( 0 , m ′ ) , x ∣ V i nh / V i I ∈ Y ( c nh ′ , 0 ) , x ∣ V i I / V i ∈ Y ( c I ′ , 0 ) , x ∣ V i + 2 / V i + 1 h r ∈ Y ( c P ′′ , 0 ) , x ∣ V i + 1 h r / V i + 1 nh ∈ Y ( 0 , m ′′ ) , x ∣ V i + 1 nh / V i + 1 I ∈ Y ( c nh ′′ , 0 ) , x ∣ V i + 1 I / V i + 1 ∈ Y ( c I ′′ , 0 ) , x ∣ V s / V s − 1 ∈ Y ( c s , m s ) , 1 ≤ s ≤ i , i + 3 ≤ s ≤ d . ⎭ ⎬ ⎫ .
Then by repeatedly applying the claim for lower ν ~ \widetilde{\nu} ν , it will reduced to the case we consider above or the conclusion we need. The proof is similar as before, by considering the order of preprojective and preinjective modules.
The claim is proved.
Together with (22) , this completes the proof.
Remark 3.4. Theorem 3.3 applies in more general settings. For example, it can be used to compute q ! Q ‾ l q! \overline{\mathbb{Q}}_{l} q ! Q l , where
{ ( x , F ) | x ∈ E ν , F = ( 0 = V 1 ⊂ ⋯ ⊂ V n = V ) , x ( F ) ⊂ F , x ∣ V n / V 1 ∈ Y ( c , m ) , x ∣ V i / V i − 1 ∈ Y ( c i , m i ) , 2 ≤ i ≤ n } → q Y ( c , m ) . \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in\mathbf{E}_{\nu},\quad F=(0=V_{1}\subset\cdots\subset V_{n}=\mathbf{V}),\ x(F)\subset F,\\
&x|_{V_{n}/V_{1}}\in Y(c,m),\quad x|_{V_{i}/V_{i-1}}\in Y(c_{i},m_{i}),\quad2\leq i\leq n
\end{aligned}\right\}
\xrightarrow{q}Y(c,m). { ( x , F ) x ∈ E ν , F = ( 0 = V 1 ⊂ ⋯ ⊂ V n = V ) , x ( F ) ⊂ F , x ∣ V n / V 1 ∈ Y ( c , m ) , x ∣ V i / V i − 1 ∈ Y ( c i , m i ) , 2 ≤ i ≤ n } q Y ( c , m ) .
Remark 3.5. By Theorem 3.3 , for any flag sheaf complex L \mathcal{L} L , the restriction L ∣ X ( c , m ) \mathcal{L}|_{X(c,m)} L ∣ X ( c , m ) is generated, as a triangulated object, by the sheaves L χ L_{\chi} L χ , where χ ∈ X ( S m ) \chi\in X(\mathcal{S}_{m}) χ ∈ X ( S m ) . Thus H i ( L ∣ X ( c , m ) ) H^{i}(\mathcal{L}|_{X(c,m)}) H i ( L ∣ X ( c , m ) ) is lisse for every i i i .
The transition matrix between the canonical basis and the PBW basis
For any flag sheaf complex L \mathcal{L} L , any ( c , m ) ∈ G ~ ν (c,m)\in\widetilde{\mathcal{G}}_{\nu} ( c , m ) ∈ G ν need not be aperiodic, we establish the following purity statement: ∀ x ∈ X ( c , m ) \forall x\in X(c,m) ∀ x ∈ X ( c , m ) , the stalk L ∣ x \mathcal{L}|_{x} L ∣ x is pure.
Theorem 4.1. For any flag sheaf complex L \mathcal{L} L , any ( c , m ) ∈ G ~ ν (c,m)\in\widetilde{\mathcal{G}}_{\nu} ( c , m ) ∈ G ν , the restriction L ∣ X ( c , m ) \mathcal{L}|_{X(c,m)} L ∣ X ( c , m ) is pointwise pure of weight 0.
This theorem will be proved in Section 5.
Proposition 4.2. Fix a flag sheaf complex L \mathcal{L} L , and consider the morphism Y ( c , m ) → j c , m E ν Y(c,m)\xrightarrow{j^{c,m}}\mathbf{E}_{\nu} Y ( c , m ) j c , m E ν together with the complex j ∗ c , m L j^{c,m}_{*}\mathcal{L} j ∗ c , m L . This complex is semisimple and has the following form in D G ν b ( Y ( c , m ) ) D^{b}_{\mathcal{G}_{\nu}}(Y(c,m)) D G ν b ( Y ( c , m )) :
⨁ χ , l j ∗ c , m I C ( X ( c , m ) , L χ ) ⊕ a χ , l [ l ] ( l 2 ) \bigoplus_{\chi,l}j^{c,m}_{*}IC(X(c,m),L_{\chi})^{\oplus a_{\chi,l}[l]\left(\frac{l}{2}\right)} χ , l ⨁ j ∗ c , m I C ( X ( c , m ) , L χ ) ⊕ a χ , l [ l ] ( 2 l )
Choose a basis of the direct sum, for which the Weil structure matrix is upper triangular. The induced Weil structure on each component j ∗ c , m I C ( X ( c , m ) , L χ ) [ l ] ( l 2 ) j^{c,m}_{*}IC(X(c,m),L_{\chi})[l]\left(\frac{l}{2}\right) j ∗ c , m I C ( X ( c , m ) , L χ ) [ l ] ( 2 l ) has the form ψ χ , c [ l ] ( l 2 ) \psi_{\chi,c}[l]\left(\frac{l}{2}\right) ψ χ , c [ l ] ( 2 l ) for some l ω l = 1 l\omega l=1 l ω l = 1 .
Proof. Consider the morphism
Y ( c , m ) → j c , m E ν Y(c,m)\xrightarrow{j^{c,m}}\mathbf{E}_{\nu} Y ( c , m ) j c , m E ν
and a flag sheaf complex L \mathcal{L} L . We first prove that j ∗ c , m L j^{c,m}_{*}\mathcal{L} j ∗ c , m L is pure.
Consider
X ( c , m ) → j ~ c , m E ν . X(c,m)\xrightarrow{\widetilde{j}^{c,m}}\mathbf{E}_{\nu}. X ( c , m ) j c , m E ν .
Since j ~ ∗ c , m L \widetilde{j}^{c,m}_{*}\mathcal{L} j ∗ c , m L is pointwise pure of weight 0 and is a lisse sheaf complex on a smooth variety, it is pure by [9 6.2.5] .
Moreover, the morphism
{ x ∈ Y ( c , m ) , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ V 2 , 2 ⊂ ⋯ ⊂ V 2 , s = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , t ⊂ V ) , x ( F ) ⊂ F , x ∣ V 1 / V 0 ≅ C I , x ∣ V 2 , i / V 2 , i − 1 ≅ C n h , i , x ∣ V 3 , j / V 3 , j − 1 is homogeneous regular of dimension λ i δ , x ∣ V / V 3 , t ≅ C P → π c , Δ Y ( c , m ) \left\{
\begin{array}{ll}
x\in Y(c,m),\\
F=(0=V_0\subset V_1=V_{2,0}\subset V_{2,1}\subset V_{2,2}\subset\cdots\subset V_{2,s}=V_{3,0}\\
\qquad\subset V_{3,1}\subset\cdots\subset V_{3,t}\subset V),\ x(F)\subset F,\\
x|_{V_1/V_0}\cong C_I,\qquad x|_{V_{2,i}/V_{2,i-1}}\cong C_{n_h,i},\\
x|_{V_{3,j}/V_{3,j-1}}\text{ is homogeneous regular of dimension }\lambda_i\delta,\\
x|_{V/V_{3,t}}\cong C_P
\end{array}
\right.
\xrightarrow{\ \pi_{c,\Delta}\ }Y(c,m) ⎩ ⎨ ⎧ x ∈ Y ( c , m ) , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ V 2 , 2 ⊂ ⋯ ⊂ V 2 , s = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , t ⊂ V ) , x ( F ) ⊂ F , x ∣ V 1 / V 0 ≅ C I , x ∣ V 2 , i / V 2 , i − 1 ≅ C n h , i , x ∣ V 3 , j / V 3 , j − 1 is homogeneous regular of dimension λ i δ , x ∣ V / V 3 , t ≅ C P π c , Δ Y ( c , m )
yields π c , Δ ! Q ‾ l [ dim X ( c , m ) ] ( dim ( X ( c , m ) ) 2 ) ≅ ν c , m ! c , m ( ⨁ x ∈ X ( S m ) L X [ dim X ( c , m ) ] ( dim ( X ( c , m ) ) 2 ) ⊕ K λ , x ) ∼ ⨁ x ∈ X ( S m ) j c , m ∗ I C ( X ( c , m ) , L X ) ⊕ K λ , x \pi_{c,\Delta}!\overline{\mathbb{Q}}_l[\dim X(c,m)]\left(\frac{\dim(X(c,m))}{2}\right)\cong\nu_{c,m!}^{\phantom{c,m}}\left(\bigoplus_{x\in X(S_m)}L_X[\dim X(c,m)]\left(\frac{\dim(X(c,m))}{2}\right)^{\oplus K_{\lambda,x}}\right)\sim\bigoplus_{x\in X(S_m)}j^{c,m*}\mathcal{IC}(X(c,m),L_X)^{\oplus K_{\lambda,x}} π c , Δ ! Q l [ dim X ( c , m )] ( 2 d i m ( X ( c , m )) ) ≅ ν c , m ! c , m ( ⨁ x ∈ X ( S m ) L X [ dim X ( c , m )] ( 2 d i m ( X ( c , m )) ) ⊕ K λ , x ) ∼ ⨁ x ∈ X ( S m ) j c , m ∗ I C ( X ( c , m ) , L X ) ⊕ K λ , x as in Equation (22) . And ν c , m ! c , m ( L X [ dim X ( c , m ) ] ( dim ( X ( c , m ) ) 2 ) ) \nu_{c,m!}^{\phantom{c,m}}\left(L_X[\dim X(c,m)]\left(\frac{\dim(X(c,m))}{2}\right)\right) ν c , m ! c , m ( L X [ dim X ( c , m )] ( 2 d i m ( X ( c , m )) ) ) is a simple perverse sheaf.
Thus every simple generator of j c , m ∗ L j^{c,m*}\mathcal{L} j c , m ∗ L has the form ν c , m ! P \nu_{c,m!}P ν c , m ! P , up to a shift, where P P P is a simple perverse sheaf, up to a shift, occurring in j c , m ∗ L j^{c,m*}\mathcal{L} j c , m ∗ L , and ν c , m : X ( c , m ) → Y ( c , m ) \nu_{c,m}:X(c,m)\to Y(c,m) ν c , m : X ( c , m ) → Y ( c , m ) .
Moreover, P P P is a lisse sheaf up to a shift.
As in the proof of Theorem 4.1, we consider the weight filtration of P H i ( j c , m ∗ L ) \mathbf{P}\mathbf{H}^{i}(j^{c,m*}\mathcal{L}) P H i ( j c , m ∗ L ) .
0 ⊂ P 1 ⊂ P 2 ⊂ ⋯ ⊂ P l − 1 ⊂ P l = p H i ( j c , m ∗ L ) , 0\subset P_1\subset P_2\subset\cdots\subset P_{l-1}\subset P_l={}^{p}\mathbf{H}^{i}(j^{c,m*}\mathcal{L}), 0 ⊂ P 1 ⊂ P 2 ⊂ ⋯ ⊂ P l − 1 ⊂ P l = p H i ( j c , m ∗ L ) ,
where each P i / P i − 1 P_i/P_{i-1} P i / P i − 1 is pure of weight w t ( P i / P i − 1 ) wt(P_i/P_{i-1}) w t ( P i / P i − 1 ) and w t ( P i / P i − 1 ) < w t ( P i + 1 / P i ) wt(P_i/P_{i-1})<wt(P_{i+1}/P_i) w t ( P i / P i − 1 ) < w t ( P i + 1 / P i ) .
Applying l c , m ∗ l^{*}_{c,m} l c , m ∗ , we preserve both the weights and the filtration.
Moreover, p H i ( j c , m ∗ L ) ≅ H i − dim X ( c , m ) ( j c , m ∗ L ) [ dim X ( c , m ) ] {}^{p}\mathbf{H}^{i}(j^{c,m*}\mathcal{L})\cong\mathbf{H}^{i-\dim X(c,m)}(j^{c,m*}\mathcal{L})[\dim X(c,m)] p H i ( j c , m ∗ L ) ≅ H i − d i m X ( c , m ) ( j c , m ∗ L ) [ dim X ( c , m )] , since H k ( j c , m ∗ L ) \mathbf{H}^{k}(j^{c,m*}\mathcal{L}) H k ( j c , m ∗ L ) is a lisse sheaf complex for any k k k . Thus p H i ( j c , m ∗ L ) ≅ H i − dim X ( c , m ) ( j c , m ∗ L ) [ dim X ( c , m ) ] {}^{p}\mathbf{H}^{i}(j^{c,m*}\mathcal{L})\cong\mathbf{H}^{i-\dim X(c,m)}(j^{c,m*}\mathcal{L})[\dim X(c,m)] p H i ( j c , m ∗ L ) ≅ H i − d i m X ( c , m ) ( j c , m ∗ L ) [ dim X ( c , m )] , and this complex is pure of weight i i i . If l c , m ∗ ( P i / P i − 1 ) = 0 l^{*}_{c,m}(P_i/P_{i-1})=0 l c , m ∗ ( P i / P i − 1 ) = 0 , then P i / P i − 1 = 0 P_i/P_{i-1}=0 P i / P i − 1 = 0 , by the description of the simple generators of p H i ( j c , m ∗ L ) {}^{p}\mathbf{H}^{i}(j^{c,m*}\mathcal{L}) p H i ( j c , m ∗ L ) and the faithfulness in [1 Proposition 5.3.9] .
We find that p H i ( j c , m ∗ L ) {}^{p}\mathbf{H}^{i}(j^{c,m*}\mathcal{L}) p H i ( j c , m ∗ L ) is pure of the same weight as H i ( j c , m ∗ L ) \mathbf{H}^{i}(j^{c,m*}\mathcal{L}) H i ( j c , m ∗ L ) .
Thus j c , m ∗ L j^{c,m*}\mathcal{L} j c , m ∗ L is pure of weight 0 0 0 and hence is semisimple. The Weil structure can be computed by restricting it to X ( c , m ) X(c,m) X ( c , m ) . □
Remark 4.3. By the proposition above, we have
χ ( j c , m ∗ L ) = ∑ χ , l ∑ i = 1 a χ , l χ ( j c , m ∗ I C ( X ( c , m ) , L X ) , ω χ , l i ψ χ , l ) ( − q ) − l . \chi(j^{c,m*}\mathcal{L})=\sum_{\chi,l}\sum_{i=1}^{a_{\chi,l}}\chi(j^{c,m*}\mathcal{IC}(X(c,m),L_X),\omega_{\chi,l}^{i}\psi_{\chi,l})(-\sqrt{q})^{-l}. χ ( j c , m ∗ L ) = χ , l ∑ i = 1 ∑ a χ , l χ ( j c , m ∗ I C ( X ( c , m ) , L X ) , ω χ , l i ψ χ , l ) ( − q ) − l .
Proposition 4.4. [34 Proposition 8.3] There exists a Laurent polynomial ϕ D c , χ ( v ) ∈ Q [ v , v − 1 ] \phi^{c,\chi}_{D}(v)\in\mathbb{Q}[v,v^{-1}] ϕ D c , χ ( v ) ∈ Q [ v , v − 1 ] , for each ( c , m ) (c,m) ( c , m ) , such that, for q = p r q=p^r q = p r , with r r r sufficiently large, we have
χ ( L ) = ∑ c , m ∑ χ ∈ X ( S m ) ϕ D c , χ ( q ) χ ( j c , m ! j c , m ∗ I C ( X ( c , m ) , L X ) , j c , m ! ψ c , χ ) . \chi(\mathcal{L})=\sum_{c,m}\sum_{\chi\in X(S_m)}\phi^{c,\chi}_{D}(\sqrt{q})\chi(j^{c,m!}j^{c,m*}\mathcal{IC}(X(c,m),L_X),j^{c,m!}\psi_{c,\chi}). χ ( L ) = c , m ∑ χ ∈ X ( S m ) ∑ ϕ D c , χ ( q ) χ ( j c , m ! j c , m ∗ I C ( X ( c , m ) , L X ) , j c , m ! ψ c , χ ) .
Proof. This proposition follows from χ ( j c , m ! j c , m ∗ I C ( X ( c , m ) , L X ) , j c , m ! ψ c , χ ) = ( − q ) − dim G ν + ∑ i ∈ Q 0 ν i N ( c , t χ ) \chi(j^{c,m!}j^{c,m*}\mathcal{IC}(X(c,m),L_X),j^{c,m!}\psi_{c,\chi})=(-\sqrt{q})^{-\dim G_{\nu}+\sum_{i\in Q_0}\nu_iN(c,t_{\chi})} χ ( j c , m ! j c , m ∗ I C ( X ( c , m ) , L X ) , j c , m ! ψ c , χ ) = ( − q ) − d i m G ν + ∑ i ∈ Q 0 ν i N ( c , t χ ) as defined in [34 Section 8] , and the fact that χ ( L ) \chi(\mathcal{L}) χ ( L ) is a product of elements of the form N ( c ′ , t χ ′ ) N(c',t'_{\chi}) N ( c ′ , t χ ′ ) , up to multiplication by v s v^s v s , for some s ∈ Z s\in\mathbb{Z} s ∈ Z . The result then follows from [34 Proposition 8.3] . □
Lemma 4.5. [34 Lemma 8.2] { χ ( j c , m ! j c , m ∗ I C ( X ( c , m ) , L X ) , j c , m ! ψ c , χ ) ∣ ( c , m ) ∈ G ~ ν , χ ∈ X ( S m ) } \left\{\chi(j^{c,m!}j^{c,m*}\mathcal{IC}(X(c,m),L_X),j^{c,m!}\psi_{c,\chi})\mid(c,m)\in\widetilde{G}_{\nu},\chi\in X(S_m)\right\} { χ ( j c , m ! j c , m ∗ I C ( X ( c , m ) , L X ) , j c , m ! ψ c , χ ) ∣ ( c , m ) ∈ G ν , χ ∈ X ( S m ) } is C \mathbb{C} C -linearly independent when q = p r q=p^r q = p r , with r r r sufficiently large.
Lemma 4.6. In Remark 4.3, ω χ , l i = 1 \omega_{\chi,l}^{i}=1 ω χ , l i = 1 for all χ , l , i \chi,l,i χ , l , i .
Proof. Using the trace map χ ( − ) \chi(-) χ ( − ) and applying j c , m ∗ j^{c,m*} j c , m ∗ to L \mathcal{L} L , we obtain
χ ( j c , m ∗ L ) = ∑ χ , l ∑ i = 1 a χ , l χ ( j c , m ∗ I C ( X ( c , m ) , L χ ) , ω χ , l i ψ c , χ ) ( − q ) − l . \chi(j^{c,m*}\mathcal{L})=\sum_{\chi,l}\sum_{i=1}^{a_{\chi,l}}\chi(j^{c,m*}IC(X(c,m),L_{\chi}),\omega_{\chi,l}^{i}\psi_{c,\chi})(-\sqrt{q})^{-l}. χ ( j c , m ∗ L ) = χ , l ∑ i = 1 ∑ a χ , l χ ( j c , m ∗ I C ( X ( c , m ) , L χ ) , ω χ , l i ψ c , χ ) ( − q ) − l .
and
χ ( j c , m ∗ L ) = ∑ χ ∈ X ( S m ) ϕ c c , χ ( q ) χ ( j c , m ∗ I C ( X ( c , m ) , L χ ) , ψ c , χ ) . \chi(j^{c,m*}\mathcal{L})=\sum_{\chi\in X(S_m)}\phi_{c}^{c,\chi}(\sqrt{q})\chi(j^{c,m*}IC(X(c,m),L_{\chi}),\psi_{c,\chi}). χ ( j c , m ∗ L ) = χ ∈ X ( S m ) ∑ ϕ c c , χ ( q ) χ ( j c , m ∗ I C ( X ( c , m ) , L χ ) , ψ c , χ ) .
By choosing F q t \mathbb{F}_{q^t} F q t -points for different t > 0 t>0 t > 0 , we find that ∑ i ( ω χ , l i ) t \sum_i(\omega_{\chi,l}^{i})^t ∑ i ( ω χ , l i ) t is constant for all sufficiently large t t t . Thus ω χ , l i = 1 \omega_{\chi,l}^{i}=1 ω χ , l i = 1 , for all χ , l , i \chi,l,i χ , l , i .
For ( c ′ , m ′ ) (c',m') ( c ′ , m ′ ) which is aperiodic, and any ( c , m ) ∈ G ~ ν (c,m)\in\widetilde{\mathcal{G}}_{\nu} ( c , m ) ∈ G ν , χ ∈ X ( S m ) \chi\in X(S_m) χ ∈ X ( S m ) , χ ′ ∈ X ( S m ′ ) \chi'\in X(S_{m'}) χ ′ ∈ X ( S m ′ ) , define f c , χ c ′ , χ ′ ( v ) ∈ N [ v , v − 1 ] f_{c,\chi}^{c',\chi'}(v)\in\mathbb{N}[v,v^{-1}] f c , χ c ′ , χ ′ ( v ) ∈ N [ v , v − 1 ] by
f c , χ c ′ , χ ′ ( v ) = ∑ i dim ( Hom D G ~ ν b ( X ( c , m ) ) ( H i − dim X ( c , m ) ( j c , m ∗ I C ( X ( c ′ , m ′ ) , L χ ′ ) ) , L χ ) ) v i . f_{c,\chi}^{c',\chi'}(v)=\sum_i\dim\left(\operatorname{Hom}_{D^b_{\widetilde{\mathcal{G}}_{\nu}}(X(c,m))}\left(\mathrm{H}^{i-\dim X(c,m)}(j^{c,m*}IC(X(c',m'),L_{\chi'})),L_{\chi}\right)\right)v^i. f c , χ c ′ , χ ′ ( v ) = i ∑ dim ( Hom D G ν b ( X ( c , m )) ( H i − d i m X ( c , m ) ( j c , m ∗ I C ( X ( c ′ , m ′ ) , L χ ′ )) , L χ ) ) v i .
Lemma 4.7. For ( c ′ , m ′ ) (c',m') ( c ′ , m ′ ) which is aperiodic, and any ( c , m ) ∈ G ~ ν (c,m)\in\widetilde{\mathcal{G}}_{\nu} ( c , m ) ∈ G ν , χ ∈ X ( S m ) \chi\in X(S_m) χ ∈ X ( S m ) , χ ′ ∈ X ( S m ′ ) \chi'\in X(S_{m'}) χ ′ ∈ X ( S m ′ ) , f c , χ c ′ , χ ′ ( v ) f_{c,\chi}^{c',\chi'}(v) f c , χ c ′ , χ ′ ( v ) has the following properties,
f c , χ c ′ , χ ′ ( v ) = 1 f_{c,\chi}^{c',\chi'}(v)=1 f c , χ c ′ , χ ′ ( v ) = 1 ,
f c , χ c ′ , χ ′ ( v ) ∈ v − 1 N [ v − 1 ] f_{c,\chi}^{c',\chi'}(v)\in v^{-1}\mathbb{N}[v^{-1}] f c , χ c ′ , χ ′ ( v ) ∈ v − 1 N [ v − 1 ] for ( c , m ) ≺ ( c ′ , m ′ ) (c,m)\prec(c',m') ( c , m ) ≺ ( c ′ , m ′ ) and f c , χ c ′ , χ ′ ( v ) = 0 f_{c,\chi}^{c',\chi'}(v)=0 f c , χ c ′ , χ ′ ( v ) = 0 for ( c , m ) ⋠ ( c ′ , m ′ ) (c,m)\npreceq(c',m') ( c , m ) ⋠ ( c ′ , m ′ ) .
Proof.
dim ( Hom D G ~ ν b ( X ( c , m ) ) ( H i − dim X ( c , m ) ( j c , m ∗ I C ( X ( c ′ , m ′ ) , L χ ′ ) ) , L χ ) ) = dim ( Hom D G ~ ν b ( Y ( c , m ) ) ( p H i ( j c , m ∗ I C ( X ( c ′ , m ′ ) , L χ ′ ) ) , ι c , m ! ∗ ( L χ [ dim X ( c , m ) ] ) ) ) . \begin{aligned}
&\dim\left(\operatorname{Hom}_{D^b_{\widetilde{\mathcal{G}}_{\nu}}(X(c,m))}\left(\mathrm{H}^{i-\dim X(c,m)}(j^{c,m*}IC(X(c',m'),L_{\chi'})),L_{\chi}\right)\right)\\
&=\dim\left(\operatorname{Hom}_{D^b_{\widetilde{\mathcal{G}}_{\nu}}(Y(c,m))}\left({}^{p}\mathrm{H}^{i}(j^{c,m*}IC(X(c',m'),L_{\chi'})),\iota_{c,m!*}(L_{\chi}[\dim X(c,m)])\right)\right).
\end{aligned} dim ( Hom D G ν b ( X ( c , m )) ( H i − d i m X ( c , m ) ( j c , m ∗ I C ( X ( c ′ , m ′ ) , L χ ′ )) , L χ ) ) = dim ( Hom D G ν b ( Y ( c , m )) ( p H i ( j c , m ∗ I C ( X ( c ′ , m ′ ) , L χ ′ )) , ι c , m ! ∗ ( L χ [ dim X ( c , m )]) ) ) .
And supp ( I C ( X ( c ′ , m ′ ) , L χ ′ ) ) = X ( c ′ , m ′ ) ‾ ⊂ ⋃ ( c , m ) ⪯ ( c ′ , m ′ ) Y ( c , m ) \operatorname{supp}(IC(X(c',m'),L_{\chi'}))=\overline{X(c',m')}\subset\bigcup_{(c,m)\preceq(c',m')}Y(c,m) supp ( I C ( X ( c ′ , m ′ ) , L χ ′ )) = X ( c ′ , m ′ ) ⊂ ⋃ ( c , m ) ⪯ ( c ′ , m ′ ) Y ( c , m ) by Proposition 1.7. Thus f c , χ c ′ , χ ′ ( v ) = 0 f_{c,\chi}^{c',\chi'}(v)=0 f c , χ c ′ , χ ′ ( v ) = 0 for ( c , m ) ⋠ ( c ′ , m ′ ) (c,m)\npreceq(c',m') ( c , m ) ⋠ ( c ′ , m ′ ) .
By [1 Lemma 3.1.4, Lemma 3.3.4] , for ( c , m ) ≺ ( c ′ , m ′ ) (c,m)\prec(c',m') ( c , m ) ≺ ( c ′ , m ′ ) , j c , m ∗ I C ( X ( c ′ , m ′ ) , L χ ′ ) ∈ p D G ~ ν ≤ − 1 ( Y ( c , m ) ) j^{c,m*}IC(X(c',m'),L_{\chi'})\in{}^{p}D^{\leq-1}_{\widetilde{\mathcal{G}}_{\nu}}(Y(c,m)) j c , m ∗ I C ( X ( c ′ , m ′ ) , L χ ′ ) ∈ p D G ν ≤ − 1 ( Y ( c , m )) . Then we have the proof.
Theorem 4.8. For ( c ′ , m ′ ) (c',m') ( c ′ , m ′ ) which is aperiodic, and any ( c , m ) ∈ G ~ ν (c,m)\in\widetilde{\mathcal{G}}_{\nu} ( c , m ) ∈ G ν , χ ∈ X ( S m ) \chi\in X(S_m) χ ∈ X ( S m ) , χ ′ ∈ X ( S m ′ ) \chi'\in X(S_{m'}) χ ′ ∈ X ( S m ′ ) , we have the identity
χ ( I C ( X ( c ′ , m ′ ) , L χ ′ ) , ϕ c ′ , χ ′ ) = ∑ c , χ f c , χ c ′ , χ ′ ( − q ) χ ( j c , m ∗ j c , m ! I C ( X ( c , m ) , L χ ) , j ! c , m ψ c , χ ) . \chi(IC(X(c',m'),L_{\chi'}),\phi_{c',\chi'})=\sum_{c,\chi}f_{c,\chi}^{c',\chi'}(-\sqrt{q})\chi(j^{c,m*}j_{c,m!}IC(X(c,m),L_{\chi}),j^{c,m}_!\psi_{c,\chi}). χ ( I C ( X ( c ′ , m ′ ) , L χ ′ ) , ϕ c ′ , χ ′ ) = c , χ ∑ f c , χ c ′ , χ ′ ( − q ) χ ( j c , m ∗ j c , m ! I C ( X ( c , m ) , L χ ) , j ! c , m ψ c , χ ) . Proof. The object ( I C ( X ( c ′ , m ′ ) , L χ ′ ) , ϕ c ′ , χ ′ ) [ n ] ( n 2 ) (IC(X(c',m'),L_{\chi'}),\phi_{c',\chi'})[n]\left(\frac{n}{2}\right) ( I C ( X ( c ′ , m ′ ) , L χ ′ ) , ϕ c ′ , χ ′ ) [ n ] ( 2 n ) is a direct summand of some flag sheaf complex L \mathcal{L} L . Choose a direct sum decomposition of L \mathcal{L} L , for which the Weil structure matrix is upper triangular. The diagonal block corresponding to this summand is ϕ c ′ , χ ′ [ n ] ( n 2 ) \phi_{c',\chi'}[n]\left(\frac{n}{2}\right) ϕ c ′ , χ ′ [ n ] ( 2 n ) .
As proved in Lemma 4.6, we have
χ ( j c , m ∗ L ) = ∑ χ , l ∑ i = 1 a χ , l χ ( j c , m ∗ I C ( X ( c , m ) , L χ ) , ψ c , χ ) ( − q ) − l . \chi(j^{c,m*}\mathcal{L})=\sum_{\chi,l}\sum_{i=1}^{a_{\chi,l}}\chi(j^{c,m*}IC(X(c,m),L_{\chi}),\psi_{c,\chi})(-\sqrt{q})^{-l}. χ ( j c , m ∗ L ) = χ , l ∑ i = 1 ∑ a χ , l χ ( j c , m ∗ I C ( X ( c , m ) , L χ ) , ψ c , χ ) ( − q ) − l . Since ( I C ( X ( c ′ , m ′ ) , L χ ′ ) , ϕ c ′ , χ ′ ) [ n ] ( n 2 ) (IC(X(c',m'),L_{\chi'}),\phi_{c',\chi'})[n]\left(\frac{n}{2}\right) ( I C ( X ( c ′ , m ′ ) , L χ ′ ) , ϕ c ′ , χ ′ ) [ n ] ( 2 n ) is a direct summand of L \mathcal{L} L , we have
j c , m ∗ I C ( X ( c ′ , m ′ ) , L χ ′ ) ≅ ⨁ χ , l j c , m ∗ I C ( X ( c , m ) , L χ ) ⊕ b χ , l [ l ] ( l 2 ) . j^{c,m*}IC(X(c',m'),L_{\chi'})\cong\bigoplus_{\chi,l}j^{c,m*}IC(X(c,m),L_{\chi})^{\oplus b_{\chi,l}}[l]\left(\frac{l}{2}\right). j c , m ∗ I C ( X ( c ′ , m ′ ) , L χ ′ ) ≅ χ , l ⨁ j c , m ∗ I C ( X ( c , m ) , L χ ) ⊕ b χ , l [ l ] ( 2 l ) .
The induced Weil structure gives
χ ( j c , m ∗ I C ( X ( c ′ , m ′ ) , L χ ′ ) , j c , m ∗ ϕ c ′ , χ ′ ) = ∑ χ , l b χ , l χ ( j c , m ∗ I C ( X ( c , m ) , L χ ) [ l ] ( l 2 ) , ψ c , χ [ l ] ( l 2 ) ) = ∑ χ , l b χ , l χ ( j c , m ∗ I C ( X ( c , m ) , L χ ) , ψ c , χ ) ( − q − 1 ) l \chi(j^{c,m*}IC(X(c',m'),L_{\chi'}),j^{c,m*}\phi_{c',\chi'})=\sum_{\chi,l}b_{\chi,l}\chi(j^{c,m*}IC(X(c,m),L_\chi)[l](\frac{l}{2}),\psi_{c,\chi}[l](\frac{l}{2}))
=\sum_{\chi,l}b_{\chi,l}\chi(j^{c,m*}IC(X(c,m),L_\chi),\psi_{c,\chi})(-\sqrt{q}^{-1})^l χ ( j c , m ∗ I C ( X ( c ′ , m ′ ) , L χ ′ ) , j c , m ∗ ϕ c ′ , χ ′ ) = χ , l ∑ b χ , l χ ( j c , m ∗ I C ( X ( c , m ) , L χ ) [ l ] ( 2 l ) , ψ c , χ [ l ] ( 2 l )) = χ , l ∑ b χ , l χ ( j c , m ∗ I C ( X ( c , m ) , L χ ) , ψ c , χ ) ( − q − 1 ) l
Combining this with
χ ( I C ( X ( c ′ , m ′ ) , L χ ′ ) , ϕ c ′ , χ ′ ) = ∑ c , m j ! c , m j c , m ∗ χ ( I C ( X ( c ′ , m ′ ) , L χ ′ ) , ϕ c ′ , χ ′ ) \chi(IC(X(c',m'),L_{\chi'}),\phi_{c',\chi'})=\sum_{c,m}j_!^{c,m}j^{c,m*}\chi(IC(X(c',m'),L_{\chi'}),\phi_{c',\chi'}) χ ( I C ( X ( c ′ , m ′ ) , L χ ′ ) , ϕ c ′ , χ ′ ) = c , m ∑ j ! c , m j c , m ∗ χ ( I C ( X ( c ′ , m ′ ) , L χ ′ ) , ϕ c ′ , χ ′ )
completes the proof. □
Proof of Theorem 4.1
To prove purity, we first establish the following lemmas.
Lemma 5.1. Let P \mathcal{P} P be a simple perverse sheaf on a variety X X X . For a locally closed subvariety U → i X U\xrightarrow{i}X U i X , if U ∩ supp ( P ) U\cap\operatorname{supp}(\mathcal{P}) U ∩ supp ( P ) is open and nonempty in supp ( P ) \operatorname{supp}(\mathcal{P}) supp ( P ) , then P ≃ j ! ∗ j ∗ P \mathcal{P}\simeq j_{!*}j^*\mathcal{P} P ≃ j ! ∗ j ∗ P .
Proof. Write the support of P \mathcal{P} P as the closed subvariety V → i X V\xrightarrow{i}X V i X . Then i ! i ∗ P ≃ P i^!i_*\mathcal{P}\simeq\mathcal{P} i ! i ∗ P ≃ P and i ! i ! P ≃ P i^!i_!\mathcal{P}\simeq\mathcal{P} i ! i ! P ≃ P . It follows that i ∗ P ≃ i ! P i_*\mathcal{P}\simeq i_!\mathcal{P} i ∗ P ≃ i ! P . Thus, by [1 Lemma 3.1.4] , i ∗ P ≃ i ! P i_*\mathcal{P}\simeq i_!\mathcal{P} i ∗ P ≃ i ! P is perverse. Given an exact sequence of perverse sheaves
0 → P 1 → i ∗ P → P 2 → 0 , 0\to\mathcal{P}_1\to i_*\mathcal{P}\to\mathcal{P}_2\to0, 0 → P 1 → i ∗ P → P 2 → 0 ,
applying i ! i^! i ! , we obtain the exact sequence of perverse sheaves
0 → i ! P 1 → P → i ! P 2 → 0. 0\to i^!\mathcal{P}_1\to\mathcal{P}\to i^!\mathcal{P}_2\to0. 0 → i ! P 1 → P → i ! P 2 → 0.
Since P \mathcal{P} P is a simple perverse sheaf, either i ! P 1 = 0 i^!\mathcal{P}_1=0 i ! P 1 = 0 or i ! P 2 = 0 i^!\mathcal{P}_2=0 i ! P 2 = 0 . Applying i ∗ i^* i ∗ , we obtain P 1 = 0 \mathcal{P}_1=0 P 1 = 0 or P 2 = 0 \mathcal{P}_2=0 P 2 = 0 . Thus i ∗ P i_*\mathcal{P} i ∗ P is a simple perverse sheaf.
Set U ′ = U ∩ V U'=U\cap V U ′ = U ∩ V , and let j ′ : U ′ → V j':U'\to V j ′ : U ′ → V be the open embedding. Since the support of i ∗ P i_*\mathcal{P} i ∗ P is V V V , we have j ′ ∗ i ∗ P ≠ 0 j'^*i_*\mathcal{P}\ne0 j ′ ∗ i ∗ P = 0 . Moreover, j ′ ∗ i ∗ P j'^*i_*\mathcal{P} j ′ ∗ i ∗ P is perverse because j ′ j' j ′ is an open embedding. Then by [1 Lemma 3.3.8] , i ∗ P ≃ j ! ∗ ′ j ′ ∗ i ∗ P i_*\mathcal{P}\simeq j'_{!*}j'^*i_*\mathcal{P} i ∗ P ≃ j ! ∗ ′ j ′ ∗ i ∗ P . This gives
P ≃ i ! j ! ∗ ′ j ′ ∗ i ∗ P . \mathcal{P}\simeq i^!j'_{!*}j'^*i_*\mathcal{P}. P ≃ i ! j ! ∗ ′ j ′ ∗ i ∗ P .
Denote the closed embedding U ∩ V → U U\cap V\to U U ∩ V → U by i ′ i' i ′ . By the definition of the functor j ! ∗ j_{!*} j ! ∗ ,
i ! j ! ∗ ′ j ′ ∗ i ∗ P ≃ j ! ∗ i ′ ! j ′ ∗ i ∗ P . i^!j'_{!*}j'^*i_*\mathcal{P}\simeq j_{!*}i'^!j'^*i_*\mathcal{P}. i ! j ! ∗ ′ j ′ ∗ i ∗ P ≃ j ! ∗ i ′ ! j ′ ∗ i ∗ P .
Finally, supp ( j ∗ P ) ⊂ V ∩ U \operatorname{supp}(j^*\mathcal{P})\subset V\cap U supp ( j ∗ P ) ⊂ V ∩ U implies that i ′ ! j ′ ∗ i ∗ P ≃ j ∗ P i'^!j'^*i_*\mathcal{P}\simeq j^*\mathcal{P} i ′ ! j ′ ∗ i ∗ P ≃ j ∗ P . This completes the proof. □
Lemma 5.2 (Pointwise purity for filtrations in a regular tube). Let k = F q k = \mathbb{F}_q k = F q , let Q Q Q be a connected acyclic quiver of affine type with vertex set Q 0 Q_0 Q 0 , and let δ \delta δ be its minimal positive imaginary root. Fix an Q 0 Q_0 Q 0 -graded vector space V V V and x 0 ∈ E V x_0 \in E_V x 0 ∈ E V such that M x 0 M_{x_0} M x 0 belongs to a non-homogeneous regular tube T \mathcal{T} T . Let m ≥ 1 m \ge1 m ≥ 1 , let J ⊂ { 1 , … , m } J \subset\{1,\ldots,m\} J ⊂ { 1 , … , m } , and let γ = ( γ 1 , … , γ m ) \gamma= (\gamma_1,\ldots,\gamma_m) γ = ( γ 1 , … , γ m ) satisfy
∑ a = 1 m γ a = dim V , γ a = δ ( a ∉ J ) . \sum_{a=1}^{m}\gamma_a = \dim V,\qquad\gamma_a = \delta\quad(a\notin J). a = 1 ∑ m γ a = dim V , γ a = δ ( a ∈ / J ) . Let F x 0 , γ , J \mathcal{F}_{x_0,\gamma,J} F x 0 , γ , J be the reduced variety of Q 0 Q_0 Q 0 -graded flags
F = ( 0 = V 0 ⊆ V 1 ⊆ ⋯ ⊆ V m = V ) F=(0=V^0\subseteq V^1\subseteq\cdots\subseteq V^m=V) F = ( 0 = V 0 ⊆ V 1 ⊆ ⋯ ⊆ V m = V ) such that x 0 V a ⊆ V a x_0V^a\subseteq V^a x 0 V a ⊆ V a , dim ( V a / V a − 1 ) = γ a \dim(V^a/V^{a-1})=\gamma_a dim ( V a / V a − 1 ) = γ a for all a a a , and M x 0 ∣ V a / V a − 1 M_{x_0}|_{V^a/V^{a-1}} M x 0 ∣ V a / V a − 1 is semisimple regular for a ∈ J a\in J a ∈ J . Here semisimple regular means a direct sum of quasi-simple regular representations. Then, for ℓ ≠ char k \ell\ne\operatorname{char} k ℓ = char k , the complex
R Γ ( F x 0 , γ , J , Q ‾ ℓ ) R\Gamma(\mathcal{F}_{x_0,\gamma,J},\overline{\mathbb{Q}}_\ell) R Γ ( F x 0 , γ , J , Q ℓ ) on a point is pure of weight 0. The empty variety is allowed.
Proof. We may assume that F x 0 , γ , J \mathcal{F}_{x_0,\gamma,J} F x 0 , γ , J is nonempty. We may also enlarge the finite field of definition so that all the finite-dimensional data and morphisms constructed below are defined over it. This replaces Frobenius by a positive power and does not change the assertion about weights.
Every γ a \gamma_a γ a has defect zero. If U ⊆ M x 0 U\subseteq M_{x_0} U ⊆ M x 0 has defect zero, then U U U has no preinjective direct summand, by the Hom-vanishing between preinjective and regular representations. Additivity of the defect therefore excludes preprojective summands, so U U U is regular. Similarly, M x 0 / U M_{x_0}/U M x 0 / U has no preprojective summand and has defect zero, hence is regular. Orthogonality of distinct tubes implies that both objects belong to T \mathcal{T} T . Consequently, all terms and successive quotients of the flags under consideration belong to T \mathcal{T} T .
Let r r r be the rank of T \mathcal{T} T , and let Q ′ Q' Q ′ be the cyclic quiver with vertex set Q 0 ′ = Z / r Z Q'_0=\mathbb{Z}/r\mathbb{Z} Q 0 ′ = Z / r Z and arrows u ⟶ u + 1 u\longrightarrow u+1 u ⟶ u + 1 . Fix an exact k k k -linear equivalence
E : rep n i l ( k Q ′ ) → ∼ T , \mathcal{E}:\operatorname{rep}^{\mathrm{nil}}(kQ')\xrightarrow{\sim}\mathcal{T}, E : rep nil ( k Q ′ ) ∼ T ,
as in [17 Section 6.3] , and write E ( M x 0 ′ ) ≃ M x 0 \mathcal{E}(M_{x'_0})\simeq M_{x_0} E ( M x 0 ′ ) ≃ M x 0 , where x 0 ′ x'_0 x 0 ′ is nilpotent on the I ′ I' I ′ -graded space
W = ⨁ u ∈ Q 0 ′ W u . W=\bigoplus_{u\in Q'_0}W_u. W = u ∈ Q 0 ′ ⨁ W u .
For u ∈ Q 0 ′ u\in Q'_0 u ∈ Q 0 ′ , let S u S_u S u be the image of the vertex simple and put d u = dim S u d_u=\dim S_u d u = dim S u . These vectors satisfy
∑ u ∈ Q 0 ′ d u = δ , ⟨ d u , d v ⟩ = δ u v − δ v , u + 1 . \sum_{u\in Q'_0}d_u=\delta,\qquad\langle d_u,d_v\rangle=\delta_{uv}-\delta_{v,u+1}. u ∈ Q 0 ′ ∑ d u = δ , ⟨ d u , d v ⟩ = δ uv − δ v , u + 1 .
They are linearly independent over Q \mathbb{Q} Q : a relation ∑ u c u d u = 0 \sum_u c_ud_u=0 ∑ u c u d u = 0 forces c v = c v − 1 c_v=c_{v-1} c v = c v − 1 for every v v v , and their common value is zero since δ ≠ 0 \delta\ne0 δ = 0 . Thus the prescribed dimensions determine unique vectors η a ∈ N I ′ \eta_a\in\mathbb{N}^{I'} η a ∈ N I ′ such that
γ a = ∑ u ∈ Q 0 ′ η a , u d u . \gamma_a=\sum_{u\in Q'_0}\eta_{a,u}d_u. γ a = u ∈ Q 0 ′ ∑ η a , u d u .
Their existence follows from nonemptiness of the filtration variety. In particular, η a = ( 1 , … , 1 ) \eta_a=(1,\ldots,1) η a = ( 1 , … , 1 ) for a ∉ J a\notin J a ∈ / J . Put η = ( η 1 , … , η m ) \eta=(\eta_1,\ldots,\eta_m) η = ( η 1 , … , η m ) .
The equivalence identifies F x 0 , γ , J \mathcal{F}_{x_0,\gamma,J} F x 0 , γ , J with the reduced variety
{ F = ( 0 = W 0 ⊆ ⋯ ⊆ W m = W ) dim ( W a / W a − 1 ) = η a , x 0 ′ W a ⊆ W a ( 1 ≤ a ≤ m ) , x 0 ′ W a ⊆ W a − 1 ( a ∈ J ) } (25) \{
\begin{aligned}
F&=(0=W^0\subseteq\cdots\subseteq W^m=W)\\
&\dim(W^a/W^{a-1})=\eta_a,\quad x'_0W^a\subseteq W^a\quad(1\le a\le m),\\
&x'_0W^a\subseteq W^{a-1}\quad(a\in J)
\end{aligned}
\} \tag*{(25)} { F = ( 0 = W 0 ⊆ ⋯ ⊆ W m = W ) dim ( W a / W a − 1 ) = η a , x 0 ′ W a ⊆ W a ( 1 ≤ a ≤ m ) , x 0 ′ W a ⊆ W a − 1 ( a ∈ J ) } (25)
Set
E W = ⨁ u ∈ Q 0 ′ Hom k ( W u , W u + 1 ) , G W = ∏ u ∈ Q 0 ′ GL ( W u ) , g W = ⨁ u ∈ Q 0 ′ End k ( W u ) . E_W=\bigoplus_{u\in Q'_0}\operatorname{Hom}_k(W_u,W_{u+1}),\qquad
G_W=\prod_{u\in Q'_0}\operatorname{GL}(W_u),\qquad
\mathfrak{g}_W=\bigoplus_{u\in Q'_0}\operatorname{End}_k(W_u). E W = u ∈ Q 0 ′ ⨁ Hom k ( W u , W u + 1 ) , G W = u ∈ Q 0 ′ ∏ GL ( W u ) , g W = u ∈ Q 0 ′ ⨁ End k ( W u ) .
The case W = 0 \mathbf{W}=0 W = 0 is immediate. Choose a decomposition M x 0 ′ = ⨁ i = 1 s M i M_{x'_0}=\bigoplus_{i=1}^{s} M_i M x 0 ′ = ⨁ i = 1 s M i into indecomposable representations, with homogeneous Jordan bases
e i , 1 , … , e i , l i , x 0 ′ e i , j = e i , j − 1 , e i , 0 = 0. e_{i,1},\ldots,e_{i,l_i},\qquad x'_0e_{i,j}=e_{i,j-1},\qquad e_{i,0}=0. e i , 1 , … , e i , l i , x 0 ′ e i , j = e i , j − 1 , e i , 0 = 0.
Write d i = deg e i , 1 ∈ Q 0 ′ d_i=\deg e_{i,1}\in Q'_0 d i = deg e i , 1 ∈ Q 0 ′ , so that deg e i , j = d i + 1 − j \deg e_{i,j}=d_i+1-j deg e i , j = d i + 1 − j . Define
w i , j = ( s + 1 ) ( l i + 1 − 2 j ) + i − 1 , λ ( t ) e i , j = t w i , j e i , j . (26) w_{i,j}=(s+1)(l_i+1-2j)+i-1,\qquad\lambda(t)e_{i,j}=t^{w_{i,j}}e_{i,j}. \tag*{(26)} w i , j = ( s + 1 ) ( l i + 1 − 2 j ) + i − 1 , λ ( t ) e i , j = t w i , j e i , j . (26)
Each e i , j e_{i,j} e i , j is homogeneous, hence λ : G m → G W \lambda:G_m\to G_W λ : G m → G W is a homomorphism of algebraic groups. For 2 ≤ j ≤ l i 2\leq j\leq l_i 2 ≤ j ≤ l i we have w i , j − 1 − w i , j = 2 ( s + 1 ) w_{i,j-1}-w_{i,j}=2(s+1) w i , j − 1 − w i , j = 2 ( s + 1 ) , while x 0 ′ e i , 1 = 0 x'_{0}e_{i,1}=0 x 0 ′ e i , 1 = 0 . Hence
λ ( t ) x 0 ′ λ ( t ) − 1 = t 2 ( s + 1 ) x 0 ′ . (27) \lambda(t)x'_0\lambda(t)^{-1}=t^{2(s+1)}x'_0. \tag*{(27)} λ ( t ) x 0 ′ λ ( t ) − 1 = t 2 ( s + 1 ) x 0 ′ . (27)
We next construct a complement to the orbit tangent space. Consider
ψ : g W ⟶ E W , ψ ( h ) = h x 0 ′ − x 0 ′ h . \psi:\mathfrak{g}_W\longrightarrow E_W,\qquad\psi(h)=hx'_0-x'_0h. ψ : g W ⟶ E W , ψ ( h ) = h x 0 ′ − x 0 ′ h .
The stabilizer Aut k Q ′ ( M x 0 ′ ) \operatorname{Aut}_{kQ'}(M_{x'_0}) Aut k Q ′ ( M x 0 ′ ) is the open set of invertible elements in End k Q ′ ( M x 0 ′ ) \operatorname{End}_{kQ'}(M_{x'_0}) End k Q ′ ( M x 0 ′ ) , and is therefore smooth. It follows that T x 0 ′ ( G W x 0 ′ ) = Im ψ T_{x'_0}(G_Wx'_0)=\operatorname{Im}\psi T x 0 ′ ( G W x 0 ′ ) = Im ψ . Let ψ ~ \widetilde{\psi} ψ denote the same commutator map on End k ( W ) \operatorname{End}_k(W) End k ( W ) . For α = ( i , j ) \alpha=(i,j) α = ( i , j ) and β = ( i ′ , j ′ ) \beta=(i',j') β = ( i ′ , j ′ ) , define the matrix unit E α , β E_{\alpha,\beta} E α , β by
E α , β ( e i ′ , j ′ ) = e i , j , E α , β ( e h , b ) = 0 if ( h , b ) ≠ ( i ′ , j ′ ) . E_{\alpha,\beta}(e_{i',j'})=e_{i,j},\qquad E_{\alpha,\beta}(e_{h,b})=0\quad\text{if }(h,b)\ne(i',j'). E α , β ( e i ′ , j ′ ) = e i , j , E α , β ( e h , b ) = 0 if ( h , b ) = ( i ′ , j ′ ) .
Thus the first index specifies the output vector and the second the input vector; equivalently, we use column vectors throughout. With α − 1 = ( i , j − 1 ) \alpha-1=(i,j-1) α − 1 = ( i , j − 1 ) and β + 1 = ( i ′ , j ′ + 1 ) \beta+1=(i',j'+1) β + 1 = ( i ′ , j ′ + 1 ) , direct computation gives
ψ ~ ( E α , β ) = E α , β + 1 − E α − 1 , β , (28) \widetilde{\psi}(E_{\alpha,\beta})=E_{\alpha,\beta+1}-E_{\alpha-1,\beta}, \tag*{(28)} ψ ( E α , β ) = E α , β + 1 − E α − 1 , β , (28)
where a matrix unit with an out-of-range position is zero.
For a fixed ordered pair ( i , i ′ ) (i,i') ( i , i ′ ) , put A = l i A=l_i A = l i and B = l i ′ B=l_{i'} B = l i ′ . The commutator preserves the rectangular block of maps from the i ′ i' i ′ th chain to the i i i th chain. In its cokernel, equation (29) gives
[ E ( i , j ) , ( i ′ , j ′ ) ] = [ E ( i , j + 1 ) , ( i ′ , j ′ + 1 ) ] ( j < A , j ′ < B ) . [E_{(i,j),(i',j')}]=[E_{(i,j+1),(i',j'+1)}]\qquad(j<A,\ j'<B). [ E ( i , j ) , ( i ′ , j ′ ) ] = [ E ( i , j + 1 ) , ( i ′ , j ′ + 1 ) ] ( j < A , j ′ < B ) .
The surviving classes are therefore spanned by E ( i , A ) , ( i ′ , b ) E_{(i,A),(i',b)} E ( i , A ) , ( i ′ , b ) , 1 ≤ b ≤ min ( A , B ) 1\leq b\leq\min(A,B) 1 ≤ b ≤ min ( A , B ) . They are linearly independent. Taking all ordered pairs proves
End k ( W ) = Im ψ ~ ⊕ C ~ , C ~ = Span { E ( i , l i ) , ( i ′ , j ′ ) : 1 ≤ i , i ′ ≤ s , 1 ≤ j ′ ≤ min ( l i , l i ′ ) } . \operatorname{End}_k(\mathbf{W})=\operatorname{Im}\widetilde{\psi}\oplus\widetilde{C},\qquad
\widetilde{C}=\operatorname{Span}\{E_{(i,l_i),(i',j')}:1\leq i,i'\leq s,\ 1\leq j'\leq\min(l_i,l_{i'})\}. End k ( W ) = Im ψ ⊕ C , C = Span { E ( i , l i ) , ( i ′ , j ′ ) : 1 ≤ i , i ′ ≤ s , 1 ≤ j ′ ≤ min ( l i , l i ′ )} .
Give End k ( W ) \operatorname{End}_k(\mathbf{W}) End k ( W ) its I ′ I' I ′ -grading, where a map of degree d d d sends W u W_u W u into W u + d W_{u+d} W u + d . Every chosen matrix unit is homogeneous, and ψ ~ \widetilde{\psi} ψ increases degree by one. In particular, if h = ∑ d h d h=\sum_d h_d h = ∑ d h d is its homogeneous decomposition, then the degree-one part of ψ ~ ( h ) \widetilde{\psi}(h) ψ ( h ) is ψ ( h 0 ) \psi(h_0) ψ ( h 0 ) . Taking the degree-one part of the preceding direct sum therefore gives
E W = Im ψ ⊕ C , C = C ~ ∩ E W . (29) E_W=\operatorname{Im}\psi\oplus C,\qquad C=\widetilde{C}\cap E_W. \tag*{(29)} E W = Im ψ ⊕ C , C = C ∩ E W . (29)
Explicitly,
C = Span { E ( i , l i ) , ( i ′ , j ′ ) : 1 ≤ i , i ′ ≤ s , 1 ≤ j ′ ≤ min ( l i , l i ′ ) , deg e i , l i = deg e i ′ , j ′ + 1 in I ′ } . C=\operatorname{Span}\left\{E_{(i,l_i),(i',j')}:
\begin{array}{l}
1\leq i,i'\leq s,\quad1\leq j'\leq\min(l_i,l_{i'}),\\
\deg e_{i,l_i}=\deg e_{i',j'}+1\text{ in }I'
\end{array}
\right\}. C = Span { E ( i , l i ) , ( i ′ , j ′ ) : 1 ≤ i , i ′ ≤ s , 1 ≤ j ′ ≤ min ( l i , l i ′ ) , deg e i , l i = deg e i ′ , j ′ + 1 in I ′ } .
In particular, pairs with i = i ′ i=i' i = i ′ are included. Put C ′ = x 0 ′ + C C'=x'_0+C C ′ = x 0 ′ + C . The notation C C C will always denote the linear complement, and C ′ C' C ′ the affine slice.
Define an action on the full representation space by
ρ ( t ) y = t 2 ( s + 1 ) λ ( t ) − 1 y λ ( t ) . (30) \rho(t)y=t^{2(s+1)}\lambda(t)^{-1}y\lambda(t). \tag*{(30)} ρ ( t ) y = t 2 ( s + 1 ) λ ( t ) − 1 y λ ( t ) . (30)
(27) gives ρ ( t ) x 0 ′ = x 0 ′ \rho(t)x'_0=x'_0 ρ ( t ) x 0 ′ = x 0 ′ . For a retained basis vector of C C C , its weight is
2 ( s + 1 ) + w i ′ , j ′ − w i , l i = ( s + 1 ) ( l i + l i ′ + 2 − 2 j ′ ) + i ′ − i 2(s+1)+w_{i',j'}-w_{i,l_i}=(s+1)(l_i+l_{i'}+2-2j')+i'-i 2 ( s + 1 ) + w i ′ , j ′ − w i , l i = ( s + 1 ) ( l i + l i ′ + 2 − 2 j ′ ) + i ′ − i
≥ 2 ( s + 1 ) − ( s − 1 ) = s + 3 > 0. \geq2(s+1)-(s-1)=s+3>0. ≥ 2 ( s + 1 ) − ( s − 1 ) = s + 3 > 0.
Thus C C C is ρ \rho ρ -stable, and the action on C ′ C' C ′ extends to A 1 × C ′ → C ′ \mathbb{A}^{1} \times C' \to C' A 1 × C ′ → C ′ with ρ ( 0 ) y = x 0 ′ \rho(0)y=x'_0 ρ ( 0 ) y = x 0 ′ for all y ∈ C ′ y \in C' y ∈ C ′ . In particular, it contracts C ′ C' C ′ to x 0 ′ x'_0 x 0 ′ .
Consider the action map
Φ : G W × C ′ ⟶ E W , ( g , y ) ⟼ g y g − 1 . (31) \Phi:G_{\mathrm{W}} \times C' \longrightarrow E_{\mathrm{W}},\qquad(g,y)\longmapsto gyg^{-1}. \tag*{(31)} Φ : G W × C ′ ⟶ E W , ( g , y ) ⟼ g y g − 1 . (31)
Its differential at ( 1 , x 0 ′ ) (1,x'_0) ( 1 , x 0 ′ ) is ( h , c ) ↦ ψ ( h ) + c (h,c)\mapsto\psi(h)+c ( h , c ) ↦ ψ ( h ) + c , which is surjective by (29) . Since source and target are smooth, Φ \Phi Φ is smooth at ( 1 , x 0 ′ ) (1,x'_0) ( 1 , x 0 ′ ) . For the action
t ⋅ ( g , y ) = ( λ ( t ) − 1 g λ ( t ) , ρ ( t ) y ) , t\cdot(g,y)=\left(\lambda(t)^{-1}g\lambda(t),\rho(t)y\right), t ⋅ ( g , y ) = ( λ ( t ) − 1 g λ ( t ) , ρ ( t ) y ) ,
Φ \Phi Φ is equivariant with respect to ρ \rho ρ on E W E_{\mathrm{W}} E W . Consequently,
B = { y ∈ C ′ : Φ is not smooth at ( 1 , y ) } B=\{y\in C':\Phi\text{ is not smooth at }(1,y)\} B = { y ∈ C ′ : Φ is not smooth at ( 1 , y )}
is closed, being the inverse image of the non-smooth locus under y ↦ ( 1 , y ) y\mapsto(1,y) y ↦ ( 1 , y ) , and is ρ \rho ρ -stable by equivariance. If B B B were nonempty, contraction would imply x 0 ′ ∈ B x'_0\in B x 0 ′ ∈ B , a contradiction. Left translation on G W G_{\mathrm{W}} G W , together with conjugation on E W E_{\mathrm{W}} E W , then shows that Φ \Phi Φ is smooth everywhere. Moreover, dim C ′ = dim E W − dim ( O x 0 ′ ) \dim C'=\dim E_{\mathrm{W}}-\dim(\mathcal{O}_{x'_0}) dim C ′ = dim E W − dim ( O x 0 ′ ) by (29) , so C ′ C' C ′ is a transverse slice at x 0 ′ x'_0 x 0 ′ .
We now consider the incidence projection with no prescribed flag type. Let
A m ( W ) = { ξ = ( ξ 1 , … , ξ m ) ∈ ( N I ′ ) m : ∑ a ξ a = dim W } . \mathcal{A}_m(\mathbf{W})=\left\{\xi=(\xi_1,\ldots,\xi_m)\in(\mathbb{N}^{I'})^m:\sum_a\xi_a=\dim\mathbf{W}\right\}. A m ( W ) = { ξ = ( ξ 1 , … , ξ m ) ∈ ( N I ′ ) m : a ∑ ξ a = dim W } .
and let F ξ ( W ) \mathcal{F}_{\xi}(\mathbf{W}) F ξ ( W ) be the variety of I ′ I' I ′ -graded flags F = ( 0 = W 0 ⊆ ⋯ ⊆ W m = W ) F=(0=\mathbf{W}^0\subseteq\cdots\subseteq\mathbf{W}^m=\mathbf{W}) F = ( 0 = W 0 ⊆ ⋯ ⊆ W m = W ) with successive dimension vectors ξ \xi ξ . F ξ ( W ) \mathcal{F}_{\xi}(\mathbf{W}) F ξ ( W ) is a smooth projective variety. Define
F ~ ξ , J = { ( y , F ) ∈ E W × F ξ ( W ) : y W a ⊆ W a ( 1 ≤ a ≤ m ) , y W a ⊆ W a − 1 ( a ∈ J ) } , π ( y , F ) = y . (32) \widetilde{\mathcal{F}}_{\xi,J}=\left\{(y,F)\in E_{\mathrm{W}}\times\mathcal{F}_{\xi}(\mathbf{W}):\begin{array}{ll}
y\mathbf{W}^a\subseteq\mathbf{W}^a & (1\leq a\leq m),\\
y\mathbf{W}^a\subseteq\mathbf{W}^{a-1} & (a\in J)
\end{array}\right\},\qquad\pi(y,F)=y. \tag*{(32)} F ξ , J = { ( y , F ) ∈ E W × F ξ ( W ) : y W a ⊆ W a y W a ⊆ W a − 1 ( 1 ≤ a ≤ m ) , ( a ∈ J ) } , π ( y , F ) = y . (32)
The stated containments are closed conditions on the universal subbundles. Hence F ~ ξ , J \widetilde{\mathcal{F}}_{\xi,J} F ξ , J is closed in E W × F ξ ( W ) E_{\mathrm{W}}\times\mathcal{F}_{\xi}(\mathbf{W}) E W × F ξ ( W ) , and π : F ~ ξ , J → E W \pi:\widetilde{\mathcal{F}}_{\xi,J}\to E_{\mathrm{W}} π : F ξ , J → E W is proper. Fix a filtration 0 = W 0 ⊂ ⋯ ⊂ W m = W 0=\mathbf{W}^0\subset\cdots\subset\mathbf{W}^m=\mathbf{W} 0 = W 0 ⊂ ⋯ ⊂ W m = W and let P ⊂ G W P\subset G_{\mathrm{W}} P ⊂ G W be the stabilizer of this filtration. It is easy to see that F ~ ξ , J = G W × P F ~ ξ , J \widetilde{\mathcal{F}}_{\xi,J}=G_{\mathrm{W}}\times_P\widetilde{\mathcal{F}}_{\xi,J} F ξ , J = G W × P F ξ , J , where
F ~ ξ , J = { y ∈ E W : y W a ⊆ W a ( 1 ≤ a ≤ m ) , y W a ⊆ W a − 1 ( a ∈ J ) } . \widetilde{\mathcal{F}}_{\xi,J}=\left\{y\in E_{\mathrm{W}}:\begin{array}{ll}
y\mathbf{W}^a\subseteq\mathbf{W}^a & (1\leq a\leq m),\\
y\mathbf{W}^a\subseteq\mathbf{W}^{a-1} & (a\in J)
\end{array}\right\}. F ξ , J = { y ∈ E W : y W a ⊆ W a y W a ⊆ W a − 1 ( 1 ≤ a ≤ m ) , ( a ∈ J ) } .
F ~ ξ , J \widetilde{\mathcal{F}}_{\xi,J} F ξ , J is smooth thus F ~ ξ , J \widetilde{\mathcal{F}}_{\xi,J} F ξ , J is smooth. Set
K = R π ! Q ‾ ℓ , F ~ ξ , J = R π ∗ Q ‾ ℓ , F ~ ξ , J . \mathcal{K}=R\pi_!\overline{\mathbb{Q}}_{\ell,\widetilde{\mathcal{F}}_{\xi,J}}=R\pi_*\overline{\mathbb{Q}}_{\ell,\widetilde{\mathcal{F}}_{\xi,J}}. K = R π ! Q ℓ , F ξ , J = R π ∗ Q ℓ , F ξ , J .
The constant complex with its natural Weil structure is pure of weight 0 on smooth variety. Proper pushforward therefore makes K \mathcal{K} K pure of weight 0 in the mixed derived category [9 ] ; see also [1 Example 5.4.5 and Corollary 5.4.9] . The incidence projection is G W G_{\mathrm{W}} G W -equivariant. It is also equivariant for ρ \rho ρ , using the action
t ⋅ ( y , F ) = ( ρ ( t ) y , λ ( t ) − 1 F ) t\cdot(y,F)=(\rho(t)y,\lambda(t)^{-1}F) t ⋅ ( y , F ) = ( ρ ( t ) y , λ ( t ) − 1 F )
on F ~ ξ , J \widetilde{\mathcal{F}}_{\xi,J} F ξ , J . Indeed,
( ρ ( t ) y ) ( λ ( t ) − 1 W a ) = t 2 ( s + 1 ) λ ( t ) − 1 ( y W a ) , (\rho(t)y)(\lambda(t)^{-1}\mathbf{W}^a)=t^{2(s+1)}\lambda(t)^{-1}(y\mathbf{W}^a), ( ρ ( t ) y ) ( λ ( t ) − 1 W a ) = t 2 ( s + 1 ) λ ( t ) − 1 ( y W a ) ,
which preserves both containment conditions in (33). The constant complex is equivariant, and proper base change therefore gives the G W G_{\mathrm{W}} G W -equivariance of K \mathcal{K} K and its weak G m \mathbb{G}_m G m -equivariance for ρ \rho ρ ; cf. [1 Exercise 2.10.2] . These equivariance isomorphisms are induced from the natural Weil structure on the constant complex and are Frobenius-compatible.
Let ι : C ′ ↪ E W \iota:C'\hookrightarrow E_{\mathrm{W}} ι : C ′ ↪ E W be the inclusion, and put L = ι ∗ K \mathcal{L}=\iota^*\mathcal{K} L = ι ∗ K . We verify that L \mathcal{L} L is pure of weight 0, using the argument in the proof of [1 Proposition 5.7.3] for the pure complex K \mathcal{K} K . Let p 2 : G W × C ′ → C ′ p_2:G_{\mathrm{W}}\times C'\to C' p 2 : G W × C ′ → C ′ be the projection. The G W G_{\mathrm{W}} G W -equivariance of K \mathcal{K} K gives
p 2 ∗ L ≃ Φ ∗ K . (33) p_2^*\mathcal{L}\simeq\Phi^*\mathcal{K}. \tag*{(33)} p 2 ∗ L ≃ Φ ∗ K . (33)
The right-hand side is pure of weight 0, since Φ \Phi Φ is smooth. For completeness, purity descends here without assuming that an arbitrary closed pullback preserves purity. Write d G = dim G W d_G=\dim G_W d G = dim G W . Smooth duality gives
D ( p 2 ∗ L ) ≃ p 2 ∗ ( D L ) [ 2 d G ] ( d G ) . \mathbb{D}(p_2^*\mathcal{L})\simeq p_2^*(\mathbb{D}\mathcal{L})[2d_G](d_G). D ( p 2 ∗ L ) ≃ p 2 ∗ ( D L ) [ 2 d G ] ( d G ) .
Both p 2 ∗ L p_2^*\mathcal{L} p 2 ∗ L and its Verdier dual have weights at most 0. The shift and twist [ 2 d G ] ( d G ) [2d_G](d_G) [ 2 d G ] ( d G ) preserve weights, so the same upper bound holds for p 2 ∗ ( D L ) p_2^*(\mathbb{D}\mathcal{L}) p 2 ∗ ( D L ) . Pulling these two upper bounds back along C ′ → G W × C ′ C'\to G_W\times C' C ′ → G W × C ′ , y ↦ ( 1 , y ) y\mapsto(1,y) y ↦ ( 1 , y ) , shows that L \mathcal{L} L and D L \mathbb{D}\mathcal{L} D L both have weights at most 0. Thus L \mathcal{L} L is pure of weight 0.
The inclusion ι \iota ι is G m \mathbb{G}_m G m -equivariant, hence L \mathcal{L} L is weakly equivariant for the contracting action on C ′ C' C ′ . Let j 0 : { x 0 ′ } ↪ C ′ j_0:\{x'_0\}\hookrightarrow C' j 0 : { x 0 ′ } ↪ C ′ be the inclusion and let a 0 : C ′ → { x 0 ′ } a_0:C'\to\{x'_0\} a 0 : C ′ → { x 0 ′ } be the structure map. The contraction isomorphism gives
R ( a 0 ) ∗ L ≃ j 0 ∗ L . (34) R(a_0)_*\mathcal{L}\simeq j_0^*\mathcal{L}. \tag*{(34)} R ( a 0 ) ∗ L ≃ j 0 ∗ L . (34)
see [1 Theorem 2.10.3] , for the mixed ℓ \ell ℓ -adic formulation. All maps are defined over the chosen finite field, so this isomorphism is compatible with Frobenius. Since L \mathcal{L} L is pure of weight 0, the left-hand side of (35) has weights at least 0, whereas the right-hand side has weights at most 0. Both sides are therefore pure of weight 0.
For i x 0 ′ : { x 0 ′ } ↪ E W i_{x'_0}:\{x'_0\}\hookrightarrow E_W i x 0 ′ : { x 0 ′ } ↪ E W , proper base change now gives
i x 0 ′ ∗ K = j 0 ∗ L ≃ R Γ ( π − 1 ( x 0 ) , Q ‾ ℓ ) , i_{x'_0}^*\mathcal{K}=j_0^*\mathcal{L}\simeq R\Gamma(\pi^{-1}(x_0),\overline{\mathbb{Q}}_\ell), i x 0 ′ ∗ K = j 0 ∗ L ≃ R Γ ( π − 1 ( x 0 ) , Q ℓ ) ,
which is pure of weight 0. Thus R π ! Q ‾ ℓ R\pi_!\overline{\mathbb{Q}}_\ell R π ! Q ℓ is pointwise pure of weight 0 at every nilpotent point. The identification with F x 0 , γ , J F_{x_0,\gamma,J} F x 0 , γ , J proves the assertion. □ \square □
We now prove Theorem 4.1: for any x ∈ X ( c , m ) x\in X(c,m) x ∈ X ( c , m ) and any flag sheaf complex L \mathcal{L} L , the stalk L x \mathcal{L}_x L x is pure. To prove purity, we may extend the field of definition. Given an F q ℓ \mathbb{F}_{q^\ell} F q ℓ -point x x x , we may regard it as an F q ℓ ′ \mathbb{F}_{q^{\ell'}} F q ℓ ′ -point, with ℓ ′ ≥ ℓ \ell'\geq\ell ℓ ′ ≥ ℓ . We may therefore choose x x x as an F q ℓ ′ \mathbb{F}_{q^{\ell'}} F q ℓ ′ -point so that x x x decomposes as x P ⊕ x I ⊕ ⨁ j x n h , j ⊕ ⨁ i x h r , i x_P\oplus x_I\oplus\bigoplus_jx_{nh,j}\oplus\bigoplus_ix_{hr,i} x P ⊕ x I ⊕ ⨁ j x nh , j ⊕ ⨁ i x h r , i over the field F q ℓ ′ \mathbb{F}_{q^{\ell'}} F q ℓ ′ .
Proof. For any ( c , m ) (c,m) ( c , m ) , fix x ∈ X ( c , m ) x\in X(c,m) x ∈ X ( c , m ) . We can choose V I , V P , V n h , i , V h r , j V_I,V_P,V_{nh,i},V_{hr,j} V I , V P , V nh , i , V h r , j such that
x = x ∣ V I ⊕ x ∣ V P ⊕ ⨁ i x ∣ V n h , i ⊕ ⨁ j x ∣ V h r , j , x=x|_{V_I}\oplus x|_{V_P}\oplus\bigoplus_i x|_{V_{nh,i}}\oplus\bigoplus_j x|_{V_{hr,j}}, x = x ∣ V I ⊕ x ∣ V P ⊕ i ⨁ x ∣ V nh , i ⊕ j ⨁ x ∣ V h r , j ,
and M x ∣ V I ≅ c I M_{x|_{V_I}}\cong c_I M x ∣ V I ≅ c I , M x ∣ V P ≅ c P M_{x|_{V_P}}\cong c_P M x ∣ V P ≅ c P , M x ∣ V n h , i ≅ c n h , i M_{x|_{V_{nh,i}}}\cong c_{nh,i} M x ∣ V nh , i ≅ c nh , i , and M x ∣ V h r , j M_{x|_{V_{hr,j}}} M x ∣ V h r , j is a simple homogeneous regular module. Denote x I x_I x I as x ∣ V I x|_{V_I} x ∣ V I and x P x_P x P as x ∣ V P x|_{V_P} x ∣ V P .
We make the contracting action explicit. Refine the preprojective and preinjective parts into their isotypic summands:
V P = ⨁ a ∈ A P V P , a , V I = ⨁ b ∈ A I V I , b , V_P=\bigoplus_{a\in A_P}V_{P,a},\qquad V_I=\bigoplus_{b\in A_I}V_{I,b}, V P = a ∈ A P ⨁ V P , a , V I = b ∈ A I ⨁ V I , b ,
where
M x ∣ V P , a ≃ P a ⊕ m a , M x ∣ V I , b ≃ I b ⊕ n b . M_{x|_{V_{P,a}}}\simeq P_a^{\oplus m_a},\qquad M_{x|_{V_{I,b}}}\simeq I_b^{\oplus n_b}. M x ∣ V P , a ≃ P a ⊕ m a , M x ∣ V I , b ≃ I b ⊕ n b .
Here A P A_P A P and A I A_I A I are finite sets of indices, with the conventions P 0 < P 1 < ⋯ P_0<P_1<\cdots P 0 < P 1 < ⋯ and ⋯ < I 2 < I 1 \cdots<I_2<I_1 ⋯ < I 2 < I 1 of Section 1. Choose an integer
N > max ( A P ∪ A I ∪ { 0 } ) . N>\max(A_P\cup A_I\cup\{0\}). N > max ( A P ∪ A I ∪ { 0 }) .
Define a cocharacter λ : G m ⟶ G ν \lambda:\mathbb{G}_m\longrightarrow G_\nu λ : G m ⟶ G ν by
λ ( t ) ∣ V P , a = t N − a id V P , a , λ ( t ) ∣ V I , b = t b − N id V I , b , \lambda(t)|_{V_{P,a}}=t^{N-a}\operatorname{id}_{V_{P,a}},\qquad
\lambda(t)|_{V_{I,b}}=t^{b-N}\operatorname{id}_{V_{I,b}}, λ ( t ) ∣ V P , a = t N − a id V P , a , λ ( t ) ∣ V I , b = t b − N id V I , b ,
λ ( t ) ∣ V n h , i = id V n h , i , λ ( t ) ∣ V h r , j = id V h r , j . \lambda(t)|_{V_{nh,i}}=\operatorname{id}_{V_{nh,i}},\qquad
\lambda(t)|_{V_{hr,j}}=\operatorname{id}_{V_{hr,j}}. λ ( t ) ∣ V nh , i = id V nh , i , λ ( t ) ∣ V h r , j = id V h r , j .
Since x x x is block diagonal, λ ( t ) ⋅ x = x \lambda(t)\cdot x=x λ ( t ) ⋅ x = x .
We construct transversal slices in E V n h , i , E V h r , j E_{V_{nh,i}},E_{V_{hr,j}} E V nh , i , E V h r , j .
We construct complements to the tangent spaces of O n h , j \mathcal{O}_{nh,j} O nh , j and O h r , j \mathcal{O}_{hr,j} O h r , j , denoted by T n h , j , T h r , j \mathcal{T}_{nh,j},\mathcal{T}_{hr,j} T nh , j , T h r , j , respectively.
Using the notation of Section 1, for any non-homogeneous regular module M i ( π ) M_i(\pi) M i ( π ) in the tube T i \mathcal{T}_i T i , the condition Hom k Q ( M x , M i ( π ) ) > 0 \operatorname{Hom}_{kQ}(M_x,M_i(\pi))>0 Hom k Q ( M x , M i ( π )) > 0 is closed in E ν E_\nu E ν . It follows that the condition that M x M_x M x is regular and has no non-homogeneous direct summands outside T j \mathcal{T}_j T j is open.
Choose a corresponding open subset with x ∣ V n h , i ∈ U n h , i x|_{V_{nh,i}}\in U_{nh,i} x ∣ V nh , i ∈ U nh , i , such that, for every x ∈ U n h , i x\in U_{nh,i} x ∈ U nh , i , the representation x x x is regular and has no non-homogeneous direct summands in other tubes, and
G V n h , i × U n h , i ⟶ E V n h , i G_{V_{nh,i}}\times U_{nh,i}\longrightarrow E_{V_{nh,i}} G V nh , i × U nh , i ⟶ E V nh , i
is smooth.
Clearly U n h , i U_{nh,i} U nh , i is smooth.
We can also choose transversal slices T P \mathcal{T}_P T P and T I \mathcal{T}_I T I at the points x P x_P x P and x I x_I x I in E V P E_{V_P} E V P and E V I E_{V_I} E V I . By [1 Proposition 5.7.3] , G m G_m G m defined above actions on T P \mathcal{T}_P T P and T I \mathcal{T}_I T I with fixed points x P x_P x P and x I x_I x I , respectively. These G m G_m G m actions extend to A \mathbb{A} A -actions.
Set
E V , V ′ : = ⨁ h ∈ H Hom k ( V s ( h ) , V t ( h ) ′ ) . E_{V,V'}:=\bigoplus_{h\in H}\operatorname{Hom}_k(V_{s(h)},V'_{t(h)}). E V , V ′ := h ∈ H ⨁ Hom k ( V s ( h ) , V t ( h ) ′ ) .
We can choose a complement to the tangent space of O x \mathcal{O}_x O x at x x x , stable under G m ⊂ G ν G_m\subset G_\nu G m ⊂ G ν , as a subvariety T \mathcal{T} T of
∏ i T n h , i × ∏ j T h r , j × ∏ j E V I , V h r , j × ∏ i E V I , V n h , i \prod_i\mathcal{T}_{nh,i}\times\prod_j\mathcal{T}_{hr,j}\times\prod_j E_{V_I,V_{hr,j}}\times\prod_i E_{V_I,V_{nh,i}} i ∏ T nh , i × j ∏ T h r , j × j ∏ E V I , V h r , j × i ∏ E V I , V nh , i
× ∏ j E V h r , j , V P × ∏ i E V n h , i , V P × E V I , V P × T P × T I . {}\times\prod_j E_{V_{hr,j},V_P}\times\prod_i E_{V_{nh,i},V_P}\times E_{V_I,V_P}\times\mathcal{T}_P\times\mathcal{T}_I. × j ∏ E V h r , j , V P × i ∏ E V nh , i , V P × E V I , V P × T P × T I .
A ⊃ G m \mathbb{A}\supset G_m A ⊃ G m acts trivially on ∏ i T n h , i × ∏ j T h r , j × x P × x I \prod_i\mathcal{T}_{nh,i}\times\prod_j\mathcal{T}_{hr,j}\times x_P\times x_I ∏ i T nh , i × ∏ j T h r , j × x P × x I and λ ( t ) \lambda(t) λ ( t ) acts on
∏ j E V I , V h r , j × ∏ i E V I , V n h , i × ∏ j E V h r , j , V P × ∏ i E V n h , i , V P × E V I , V P × T P × T I , \prod_j E_{V_I,V_{hr,j}}\times\prod_i E_{V_I,V_{nh,i}}\times\prod_j E_{V_{hr,j},V_P}\times\prod_i E_{V_{nh,i},V_P}\times E_{V_I,V_P}\times\mathcal{T}_P\times\mathcal{T}_I, j ∏ E V I , V h r , j × i ∏ E V I , V nh , i × j ∏ E V h r , j , V P × i ∏ E V nh , i , V P × E V I , V P × T P × T I ,
satisfying λ ( 0 ) x = lim t → 0 ( λ ( t ) ( x ′ ) ) = 0 × x P × x I \lambda(0)x=\lim_{t\to0}(\lambda(t)(x'))=0\times x_P\times x_I λ ( 0 ) x = lim t → 0 ( λ ( t ) ( x ′ )) = 0 × x P × x I , for any x ′ x' x ′ , and G m ⊂ G ν G_m\subset G_\nu G m ⊂ G ν .
We can choose an open subvariety U h r ⊂ ∏ j T h r , j U_{hr}\subset\prod_j\mathcal{T}_{hr,j} U h r ⊂ ∏ j T h r , j such that U h r ⊂ X ( 0 , m ) U_{hr}\subset X(0,m) U h r ⊂ X ( 0 , m ) and denote T h r , j ∩ X ( 0 , 1 ) \mathcal{T}_{hr,j}\cap X(0,1) T h r , j ∩ X ( 0 , 1 ) as U h r , j U_{hr,j} U h r , j . The subvariety
∏ i U n h , i × U h r × ∏ j E V I , V h r , j × ∏ i E V I , V n h , i × T P × T I \prod_i U_{nh,i}\times U_{hr}\times\prod_j E_{V_I,V_{hr,j}}\times\prod_i E_{V_I,V_{nh,i}}\times\mathcal{T}_P\times\mathcal{T}_I i ∏ U nh , i × U h r × j ∏ E V I , V h r , j × i ∏ E V I , V nh , i × T P × T I
× ∏ j E V h r , j , V P × ∏ i E V n h , i , V P × E V I , V P ∩ T {}\times\prod_j E_{V_{hr,j},V_P}\times\prod_i E_{V_{nh,i},V_P}\times E_{V_I,V_P}\cap\mathcal{T} × j ∏ E V h r , j , V P × i ∏ E V nh , i , V P × E V I , V P ∩ T
is open in T \mathcal{T} T and stable under G m ⊂ G ν G_m\subset G_\nu G m ⊂ G ν , where the G m G_m G m action extends to an A \mathbb{A} A -action.
We can also choose an open subvariety U ⊂ T U\subset\mathcal{T} U ⊂ T such that, for any x ∈ U x\in U x ∈ U , Hom ( M x ∣ V n h , i , M x ∣ V n h , j ) = 0 \operatorname{Hom}(M_x|_{V_{nh,i}},M_x|_{V_{nh,j}})=0 Hom ( M x ∣ V nh , i , M x ∣ V nh , j ) = 0 for i ≠ j i\ne j i = j and Hom ( M x ∣ V n h , i , M x ∣ V h r , k ) = 0 \operatorname{Hom}(M_x|_{V_{nh,i}},M_x|_{V_{hr,k}})=0 Hom ( M x ∣ V nh , i , M x ∣ V h r , k ) = 0 , and such that this subvariety is stable under G m G_m G m and
G ν × U ⟶ E ν G_\nu\times U\longrightarrow E_\nu G ν × U ⟶ E ν
is a smooth morphism.
Finally, we choose
U ~ = U ∩ ( ( U ∩ ( x P × x I × ∏ i U n h , i × U h r × { 0 } ) ) × ∏ j E V I , V h r , j × ∏ i E V I , V n h , i \widetilde{U}=U\cap((U\cap(x_P\times x_I\times\prod_i U_{nh,i}\times U_{hr}\times\{0\}))\times\prod_j E_{V_I,V_{hr,j}}\times\prod_i E_{V_I,V_{nh,i}} U = U ∩ (( U ∩ ( x P × x I × i ∏ U nh , i × U h r × { 0 })) × j ∏ E V I , V h r , j × i ∏ E V I , V nh , i
× ∏ j E V h r , j , V P × ∏ i E V n h , i , V P × E V I , V P × T P × T I ) {}\times\prod_j E_{V_{hr,j},V_P}\times\prod_i E_{V_{nh,i},V_P}\times E_{V_I,V_P}\times\mathcal{T}_P\times\mathcal{T}_I) × j ∏ E V h r , j , V P × i ∏ E V nh , i , V P × E V I , V P × T P × T I )
which has the G m G_m G m action that extends to an A \mathbb{A} A -action and is stable under Frobenius morphism Fr q l \operatorname{Fr}_{q^l} Fr q l . Moreover, U ∩ ( x P × x I × ∏ i U n h , i × U h r × { 0 } ) U\cap(x_P\times x_I\times\prod_i U_{nh,i}\times U_{hr}\times\{0\}) U ∩ ( x P × x I × ∏ i U nh , i × U h r × { 0 }) is open in x P × x I × ∏ i U n h , i × U h r × { 0 } x_P\times x_I\times\prod_i U_{nh,i}\times U_{hr}\times\{0\} x P × x I × ∏ i U nh , i × U h r × { 0 } . U ~ \widetilde{U} U is open in U U U . Using the morphisms
U ~ → ι G ν × U ~ → ϕ E ν , \widetilde{U}\xrightarrow{\iota}G_\nu\times\widetilde{U}\xrightarrow{\phi}E_\nu, U ι G ν × U ϕ E ν ,
where ϕ \phi ϕ is smooth, together with the isomorphism
i ∗ For G ν [ − dim G ν ] ≅ i ! For G ν [ dim G ν ] ( dim G ν ) i^{*}\operatorname{For}_{G_{\nu}}[-\dim G_{\nu}] \cong i^{!}\operatorname{For}_{G_{\nu}}[\dim G_{\nu}](\dim G_{\nu}) i ∗ For G ν [ − dim G ν ] ≅ i ! For G ν [ dim G ν ] ( dim G ν )
we obtain
⇒ For G ν i ∗ ϕ ∗ L is still pure. \Rightarrow\quad\operatorname{For}_{G_{\nu}}i^{*}\phi^{*}\mathcal{L}\text{ is still pure.} ⇒ For G ν i ∗ ϕ ∗ L is still pure.
Let
U 0 = U ∩ ( x P × x I × ∏ U n h , j × U h r × { 0 } ) → u 0 E ν U_{0}=U\cap\left(x_{P}\times x_{I}\times\prod U_{nh,j}\times U_{hr}\times\{0\}\right)\xrightarrow{u_{0}}E_{\nu} U 0 = U ∩ ( x P × x I × ∏ U nh , j × U h r × { 0 } ) u 0 E ν
be the G m G_{m} G m fixed point set. Clearly, it is smooth. By [1 Theorem 2.10.3] , u 0 ∗ For G ν L u_{0}^{*}\operatorname{For}_{G^{\nu}}\mathcal{L} u 0 ∗ For G ν L is pure. In fact, for any pure G ν G_{\nu} G ν -equivariant sheaf complex F F F , u 0 ∗ For G ν F u_{0}^{*}\operatorname{For}_{G^{\nu}}F u 0 ∗ For G ν F is pure. The complex
u 0 ∗ For G ν L = L 1 ⊕ L 2 ⊕ ⋯ ⊕ L n , u_{0}^{*}\operatorname{For}_{G^{\nu}}\mathcal{L}=\mathcal{L}_{1}\oplus\mathcal{L}_{2}\oplus\cdots\oplus\mathcal{L}_{n}, u 0 ∗ For G ν L = L 1 ⊕ L 2 ⊕ ⋯ ⊕ L n ,
is semisimple, with its summands ordered by support. We will show the order and what L i L_{i} L i is in the followings.
Equip U 0 U_{0} U 0 with the stratification ⨆ ( c , m ) U 0 ∩ Y ( c , m ) = U 0 \bigsqcup_{(c,m)}U_{0}\cap Y(c,m)=U_{0} ⨆ ( c , m ) U 0 ∩ Y ( c , m ) = U 0 , and write U 0 ∩ Y ( c , m ) = U 0 c , m U_{0}\cap Y(c,m)=U_{0}^{c,m} U 0 ∩ Y ( c , m ) = U 0 c , m , U 0 ∩ X ( c , m ) = U ~ 0 c , m U_{0}\cap X(c,m)=\widetilde{U}_{0}^{c,m} U 0 ∩ X ( c , m ) = U 0 c , m . It could be equiped with an order ⨆ i = 1 n U 0 c i , m i \bigsqcup_{i=1}^{n}U_{0}^{c_{i},m_{i}} ⨆ i = 1 n U 0 c i , m i , such that U 0 c s , m s U_{0}^{c_{s},m_{s}} U 0 c s , m s is open in ⨆ i = s n U 0 c i , m i \bigsqcup_{i=s}^{n}U_{0}^{c_{i},m_{i}} ⨆ i = s n U 0 c i , m i , for any s = 1 , ⋯ , n s=1,\cdots,n s = 1 , ⋯ , n .
For any ( c , m ) ∈ G ~ ν (c,m)\in\widetilde{G}_{\nu} ( c , m ) ∈ G ν , if Y ( c , m ) ∩ U 0 ≠ ∅ Y(c,m)\cap U_{0}\ne\emptyset Y ( c , m ) ∩ U 0 = ∅ , then c P ≅ x P c_{P}\cong x_{P} c P ≅ x P , and c I ≅ x I c_{I}\cong x_{I} c I ≅ x I .
u 0 ∗ For G ν L u_{0}^{*}\operatorname{For}_{G^{\nu}}\mathcal{L} u 0 ∗ For G ν L is semisimple thus could be written as ⊕ l S l \oplus_{l}S_{l} ⊕ l S l , where S l S_{l} S l is a shift of simple perverse sheaf. Denote C i = { l ∣ supp ( S l ) ∩ ( ∪ s = 1 i − 1 U 0 c s , m s ) = ∅ , supp ( S l ) ∩ U 0 c i , m i ≠ ∅ } C_{i}=\{l\mid\operatorname{supp}(S_{l})\cap(\cup_{s=1}^{i-1}U_{0}^{c_{s},m_{s}})=\emptyset,\operatorname{supp}(S_{l})\cap U_{0}^{c_{i},m_{i}}\ne\emptyset\} C i = { l ∣ supp ( S l ) ∩ ( ∪ s = 1 i − 1 U 0 c s , m s ) = ∅ , supp ( S l ) ∩ U 0 c i , m i = ∅ } . Let L i = ⊕ l ∈ C i S l L_{i}=\oplus_{l\in C_{i}}S_{l} L i = ⊕ l ∈ C i S l . It is clearly that U 0 c i , m i ∩ supp ( L i ) U_{0}^{c_{i},m_{i}}\cap\operatorname{supp}(L_{i}) U 0 c i , m i ∩ supp ( L i ) is open in supp ( L i ) \operatorname{supp}(L_{i}) supp ( L i ) .
Use the notation
U 0 c , m → j 0 c , m E ν , U ~ 0 c , m → j ~ 0 c , m E ν , U_{0}^{c,m}\xrightarrow{j_{0}^{c,m}}E_{\nu},\qquad\widetilde{U}_{0}^{c,m}\xrightarrow{\widetilde{j}_{0}^{c,m}}E_{\nu}, U 0 c , m j 0 c , m E ν , U 0 c , m j 0 c , m E ν ,
Y ( c , m ) → j c , m E ν , X ( c , m ) → j ~ c , m E ν . Y(c,m)\xrightarrow{j^{c,m}}E_{\nu},\qquad X(c,m)\xrightarrow{\widetilde{j}^{c,m}}E_{\nu}. Y ( c , m ) j c , m E ν , X ( c , m ) j c , m E ν .
Write a c , m : U 0 c , m ↪ U 0 a^{c,m}:U_{0}^{c,m}\hookrightarrow U_{0} a c , m : U 0 c , m ↪ U 0 for the internal inclusion.
Each restriction functor j 0 c , m ∗ j_{0}^{c,m*} j 0 c , m ∗ can be expressed using the composition
U 0 c , m → i c , m Y ( c , m ) → j c , m E ν . U_{0}^{c,m}\xrightarrow{i^{c,m}}Y(c,m)\xrightarrow{j^{c,m}}E_{\nu}. U 0 c , m i c , m Y ( c , m ) j c , m E ν .
For F F F satisfying Theorem 3.3 , the restriction j 0 c , m ∗ F j_{0}^{c,m*}F j 0 c , m ∗ F is generated by i c , m ∗ ( I C ( X ( c , m ) , L χ ) ∣ Y ( c , m ) ) ≃ j 0 c , m ∗ I C ( X ( c , m ) , L χ ) i^{c,m*}(IC(X(c,m),L_{\chi})|_{Y(c,m)})\simeq j_{0}^{c,m*}IC(X(c,m),L_{\chi}) i c , m ∗ ( I C ( X ( c , m ) , L χ ) ∣ Y ( c , m ) ) ≃ j 0 c , m ∗ I C ( X ( c , m ) , L χ ) in D b ( U 0 c , m ) D^{b}(U_{0}^{c,m}) D b ( U 0 c , m ) . Moreover, u 0 ∗ I C ( X ( c , m ) , L χ ) u_{0}^{*}IC(X(c,m),L_{\chi}) u 0 ∗ I C ( X ( c , m ) , L χ ) is semisimple in D b ( U 0 ) D^{b}(U_{0}) D b ( U 0 ) , since u 0 ∗ u_{0}^{*} u 0 ∗ preserves purity. We have supp ( u 0 ∗ I C ( X ( c , m ) , L χ ) ) ⊂ Y ( c , m ) ∩ U 0 \operatorname{supp}(u_{0}^{*}IC(X(c,m),L_{\chi}))\subset Y(c,m)\cap U_{0} supp ( u 0 ∗ I C ( X ( c , m ) , L χ )) ⊂ Y ( c , m ) ∩ U 0 . Since U 0 c , m U_{0}^{c,m} U 0 c , m is open in Y ( c , m ) ∩ U 0 Y(c,m)\cap U_{0} Y ( c , m ) ∩ U 0 , the restriction j 0 c , m ∗ I C ( X ( c , m ) , L χ ) j_{0}^{c,m*}IC(X(c,m),L_{\chi}) j 0 c , m ∗ I C ( X ( c , m ) , L χ ) is semisimple in D b ( U 0 c , m ) D^{b}(U_{0}^{c,m}) D b ( U 0 c , m ) . If j 0 c , m ∗ F j_{0}^{c,m*}F j 0 c , m ∗ F is semisimple in D b ( U 0 c , m ) D^{b}(U_{0}^{c,m}) D b ( U 0 c , m ) , each of its direct summands occurs as a direct summand of j 0 c , m ∗ I C ( X ( c , m ) , L χ ) j_{0}^{c,m*}IC(X(c,m),L_{\chi}) j 0 c , m ∗ I C ( X ( c , m ) , L χ ) for some χ \chi χ .
Moreover,
a c , m ∗ u 0 ∗ Ind ( I C ( O x P , Q ‾ l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ l ) ⊠ ∏ I C ( O c n h , j , Q ‾ l ) ⊠ I C ( O x I , Q ‾ l ) ) a^{c,m*}u_{0}^{*}\operatorname{Ind}\left(IC(O_{x_{P}},\overline{\mathbb{Q}}_{l})\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_{l})\boxtimes\prod IC(O_{c_{nh,j}},\overline{\mathbb{Q}}_{l})\boxtimes IC(O_{x_{I}},\overline{\mathbb{Q}}_{l})\right) a c , m ∗ u 0 ∗ Ind ( I C ( O x P , Q l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q l ) ⊠ ∏ I C ( O c nh , j , Q l ) ⊠ I C ( O x I , Q l ) )
is still pure of weight 0 0 0 , since U 0 c , m U_{0}^{c,m} U 0 c , m is open in U 0 ∩ supp ( Ind ( I C ( O x P , Q ‾ l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ l ) ⊠ ∏ I C ( O c n h , j , Q ‾ l ) ⊠ I C ( O x I , Q ‾ l ) ) ) U_{0}\cap\operatorname{supp}\left(\operatorname{Ind}\left(IC(O_{x_{P}},\overline{\mathbb{Q}}_{l})\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_{l})\boxtimes\prod IC(O_{c_{nh,j}},\overline{\mathbb{Q}}_{l})\boxtimes IC(O_{x_{I}},\overline{\mathbb{Q}}_{l})\right)\right) U 0 ∩ supp ( Ind ( I C ( O x P , Q l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q l ) ⊠ ∏ I C ( O c nh , j , Q l ) ⊠ I C ( O x I , Q l ) ) ) and is therefore semisimple in D b ( U 0 c , m ) D^{b}(U_{0}^{c,m}) D b ( U 0 c , m ) . Each simple perverse sheaf occurring above, up to a shift, has the form a c , m ∗ P a^{c,m*}P a c , m ∗ P , where P P P is a simple perverse direct summand, up to a shift, of
u 0 ∗ Ind ( I C ( O x P , Q ‾ l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ l ) ⊠ ∏ I C ( O c n h , j , Q ‾ l ) ⊠ I C ( O x I , Q ‾ l ) ) . u_{0}^{*}\operatorname{Ind}\left(IC(O_{x_{P}},\overline{\mathbb{Q}}_{l})\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_{l})\boxtimes\prod IC(O_{c_{nh,j}},\overline{\mathbb{Q}}_{l})\boxtimes IC(O_{x_{I}},\overline{\mathbb{Q}}_{l})\right). u 0 ∗ Ind ( I C ( O x P , Q l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q l ) ⊠ ∏ I C ( O c nh , j , Q l ) ⊠ I C ( O x I , Q l ) ) .
We now consider the case ( c 1 , m 1 ) (c_{1},m_{1}) ( c 1 , m 1 ) . The map a c 1 , m 1 a^{c_{1},m_{1}} a c 1 , m 1 is an open embedding. It follows that j 0 c 1 , m 1 ∗ L j_{0}^{c_{1},m_{1}*}\mathcal{L} j 0 c 1 , m 1 ∗ L is semisimple and that its simple summands occur in
a c 1 , m 1 ∗ u 0 ∗ Ind ( I C ( O x P , Q ‾ l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ l ) ⊠ ∏ I C ( O c n h , j , Q ‾ l ) ⊠ I C ( O x I , Q ‾ l ) ) . a^{c_{1},m_{1}*}u_{0}^{*}\operatorname{Ind}\left(IC(O_{x_{P}},\overline{\mathbb{Q}}_{l})\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_{l})\boxtimes\prod IC(O_{c_{nh,j}},\overline{\mathbb{Q}}_{l})\boxtimes IC(O_{x_{I}},\overline{\mathbb{Q}}_{l})\right). a c 1 , m 1 ∗ u 0 ∗ Ind ( I C ( O x P , Q l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q l ) ⊠ ∏ I C ( O c nh , j , Q l ) ⊠ I C ( O x I , Q l ) ) .
We can therefore choose a basis consisting of shifts of simple perverse sheaves in L 1 L_1 L 1 , with respect to which the matrix of the Weil structure is upper triangular. Moreover, let P P P be a simple perverse direct summand, up to a shift, of L 1 L_1 L 1 , such that a c 1 , m 1 ∗ P a^{c_1,m_1} * P a c 1 , m 1 ∗ P is not 0. This restriction occurs in
a c 1 , m 1 ∗ u 0 ∗ Ind ( I C ( O x P , Q ‾ ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ ℓ ) ⊠ ∏ I C ( O n h , j c 1 , Q ‾ ℓ ) ⊠ I C ( O x I , Q ‾ ℓ ) ) a^{c_1,m_1} * u_0^* \operatorname{Ind}\left(IC(O_{x_P},\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(O^{c_1}_{n_h,j},\overline{\mathbb{Q}}_\ell)\boxtimes IC(O_{x_I},\overline{\mathbb{Q}}_\ell)\right) a c 1 , m 1 ∗ u 0 ∗ Ind ( I C ( O x P , Q ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ℓ ) ⊠ ∏ I C ( O n h , j c 1 , Q ℓ ) ⊠ I C ( O x I , Q ℓ ) )
up to shift.
By Lemma 5.1, P P P is a direct summand(up to shift) of
u 0 ∗ Ind ( I C ( O x P , Q ‾ ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ ℓ ) ⊠ ∏ I C ( O n h , j c 1 , Q ‾ ℓ ) ⊠ I C ( O x I , Q ‾ ℓ ) ) , u_0^*\operatorname{Ind}\left(IC(O_{x_P},\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(O^{c_1}_{n_h,j},\overline{\mathbb{Q}}_\ell)\boxtimes IC(O_{x_I},\overline{\mathbb{Q}}_\ell)\right), u 0 ∗ Ind ( I C ( O x P , Q ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ℓ ) ⊠ ∏ I C ( O n h , j c 1 , Q ℓ ) ⊠ I C ( O x I , Q ℓ ) ) ,
which is a direct summand, up to a shift and twist [ s ] ( s / 2 ) [s](s/2) [ s ] ( s /2 ) for some s s s , of
u 0 ∗ Ind ( I C ( O x P , Q ‾ ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ ℓ ) ⊠ ∏ I C ( O n h , s e m i , Q ‾ ℓ ) ⊠ I C ( O x I , Q ‾ ℓ ) ) u_0^*\operatorname{Ind}\left(IC(O_{x_P},\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(O_{nh,\mathrm{semi}},\overline{\mathbb{Q}}_\ell)\boxtimes IC(O_{x_I},\overline{\mathbb{Q}}_\ell)\right) u 0 ∗ Ind ( I C ( O x P , Q ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ℓ ) ⊠ ∏ I C ( O nh , semi , Q ℓ ) ⊠ I C ( O x I , Q ℓ ) )
where O n h , s e m i O_{nh,\mathrm{semi}} O nh , semi is the orbit of a semisimple non-homogeneous module in a single tube. The argument uses the assertion that I C ( O n h , j c 1 , Q ‾ ℓ ) [ s ] ( s / 2 ) IC(O^{c_1}_{n_h,j},\overline{\mathbb{Q}}_\ell)[s](s/2) I C ( O n h , j c 1 , Q ℓ ) [ s ] ( s /2 ) occurs as a direct summand of an iterated induction of such semisimple-orbit IC complexes by [20 Section 6.9] , [32 proof of Proposition 2.7] and [10 Corollary 3.2] . The operation [ s ] ( s / 2 ) [s](s/2) [ s ] ( s /2 ) preserves purity and weight. On a Frobenius-stable simple constituent, two pure Weil normalizations of the same weight differ by a scalar of absolute value 1. By the semisimplicity of u 0 ∗ L u_0^*\mathcal{L} u 0 ∗ L ,
u 0 ∗ Ind ( I C ( O x P , Q ‾ ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ ℓ ) ⊠ ∏ I C ( O n h , j c 1 , Q ‾ ℓ ) ⊠ I C ( O x I , Q ‾ ℓ ) ) u_0^*\operatorname{Ind}\left(IC(O_{x_P},\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(O^{c_1}_{n_h,j},\overline{\mathbb{Q}}_\ell)\boxtimes IC(O_{x_I},\overline{\mathbb{Q}}_\ell)\right) u 0 ∗ Ind ( I C ( O x P , Q ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ℓ ) ⊠ ∏ I C ( O n h , j c 1 , Q ℓ ) ⊠ I C ( O x I , Q ℓ ) )
and
u 0 ∗ Ind ( I C ( O x P , Q ‾ ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ ℓ ) ⊠ ∏ I C ( O n h , s e m i , Q ‾ ℓ ) ⊠ I C ( O x I , Q ‾ ℓ ) ) , u_0^*\operatorname{Ind}\left(IC(O_{x_P},\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(O_{nh,\mathrm{semi}},\overline{\mathbb{Q}}_\ell)\boxtimes IC(O_{x_I},\overline{\mathbb{Q}}_\ell)\right), u 0 ∗ Ind ( I C ( O x P , Q ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ℓ ) ⊠ ∏ I C ( O nh , semi , Q ℓ ) ⊠ I C ( O x I , Q ℓ ) ) ,
we can choose a basis consisting of shifts of simple perverse sheaves, ordered so that the matrix of the Weil structure is upper triangular. All the complexes have weight 0, so the Weil structures on the same shifted simple perverse sheaf in these complexes differ only by a scalar of absolute value 1. Thus, if for any ( c , m ) (c,m) ( c , m ) ,
Ind ( I C ( O x P , Q ‾ ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ ℓ ) ⊠ ∏ I C ( O n h , s e m i , Q ‾ ℓ ) ⊠ I C ( O x I , Q ‾ ℓ ) ) \operatorname{Ind}\left(IC(O_{x_P},\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(O_{nh,\mathrm{semi}},\overline{\mathbb{Q}}_\ell)\boxtimes IC(O_{x_I},\overline{\mathbb{Q}}_\ell)\right) Ind ( I C ( O x P , Q ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ℓ ) ⊠ ∏ I C ( O nh , semi , Q ℓ ) ⊠ I C ( O x I , Q ℓ ) )
is pointwise pure on U 0 ∩ X ( c , m ) U_0\cap X(c,m) U 0 ∩ X ( c , m ) , then L 1 L_1 L 1 with the induced Weil structure is pointwise pure on U 0 ∩ X ( c , m ) U_0\cap X(c,m) U 0 ∩ X ( c , m ) .
More precisely, since L 1 L_1 L 1 is semisimple, we can choose a basis consisting of shifts of simple perverse sheaves with respect to which the matrix of its Weil structure is upper triangular as a restriction on u 0 ∗ L u_0^*\mathcal{L} u 0 ∗ L . It suffices to prove pointwise purity for the Weil structure restricted to each such shifts of simple perverse sheaf, which occuring in
u 0 ∗ Ind ( I C ( O x P , Q ‾ ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ ℓ ) ⊠ ∏ I C ( O n h , s e m i , Q ‾ ℓ ) ⊠ I C ( O x I , Q ‾ ℓ ) ) , u_0^*\operatorname{Ind}\left(IC(O_{x_P},\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(O_{nh,\mathrm{semi}},\overline{\mathbb{Q}}_\ell)\boxtimes IC(O_{x_I},\overline{\mathbb{Q}}_\ell)\right), u 0 ∗ Ind ( I C ( O x P , Q ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ℓ ) ⊠ ∏ I C ( O nh , semi , Q ℓ ) ⊠ I C ( O x I , Q ℓ ) ) ,
up to shift and tate twist [ l ] ( l / 2 ) [l](l/2) [ l ] ( l /2 ) . This structure differs from the corresponding Weil structure only by a scalar of absolute value 1.
Indeed, for any x ′ ∈ U 0 ∩ X ( c ′ , m ′ ) x'\in U_0\cap X(c',m') x ′ ∈ U 0 ∩ X ( c ′ , m ′ ) for any ( c ′ , m ′ ) (c',m') ( c ′ , m ′ ) , let i x ′ : x ′ ↪ E ν i_{x'}:x'\hookrightarrow E_\nu i x ′ : x ′ ↪ E ν be the inclusion. Then
i x ′ ∗ Ind ( I C ( O x P , Q ‾ ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ ℓ ) ⊠ ∏ I C ( O n h , s e m i , Q ‾ ℓ ) ⊠ I C ( O x I , Q ‾ ℓ ) ) i_{x'}^*\operatorname{Ind}\left(IC(O_{x_P},\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(O_{nh,\mathrm{semi}},\overline{\mathbb{Q}}_\ell)\boxtimes IC(O_{x_I},\overline{\mathbb{Q}}_\ell)\right) i x ′ ∗ Ind ( I C ( O x P , Q ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ℓ ) ⊠ ∏ I C ( O nh , semi , Q ℓ ) ⊠ I C ( O x I , Q ℓ ) )
is pure of weight 0. The proof is following.
By Lemma 1.4, in this case we only need to consider the regular part. Any regular module with dimension δ \delta δ satisfies all direct summands of it in the same tube and O n h , s e m i ∩ O M = ∅ O_{nh,\mathrm{semi}}\cap O_M=\varnothing O nh , semi ∩ O M = ∅ , if O M O_M O M is the orbit of any regular module other than this O n h , s e m i O_{nh,\mathrm{semi}} O nh , semi . Then apply Lemma 5.2 and deal with multiple nonhomogeneous tubes by [1 Corollary 5.4.8] , that if F \mathcal{F} F and G \mathcal{G} G are pure of weight 0 0 0 , F ⊠ G \mathcal{F}\boxtimes\mathcal{G} F ⊠ G is pure of weight 0 0 0 . And H \mathcal{H} H is pure of weight 0 0 0 , if F \mathcal{F} F and G \mathcal{G} G are pure of weight 0 0 0 and F → H → G \mathcal{F}\to\mathcal{H}\to\mathcal{G} F → H → G is + 1 +1 + 1 .
We have the proof.
Thus L 1 \mathcal{L}_1 L 1 can be written as a sum of summands of
u 0 ∗ Ind ( I C ( O x P , Q ‾ l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ l ) ⊠ ∏ I C ( O n h , s e m i , Q ‾ l ) ⊠ I C ( O x 1 , Q ‾ l ) ) u_0^*\operatorname{Ind}\left(IC(O_{x_P},\overline{\mathbb{Q}}_l)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_l)\boxtimes\prod IC(O_{n h,\mathrm{semi}},\overline{\mathbb{Q}}_l)\boxtimes IC(O_{x_1},\overline{\mathbb{Q}}_l)\right) u 0 ∗ Ind ( I C ( O x P , Q l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q l ) ⊠ ∏ I C ( O nh , semi , Q l ) ⊠ I C ( O x 1 , Q l ) )
and L 1 \mathcal{L}_1 L 1 is pointwise pure on U 0 ∩ X ( c , m ) U_0\cap X(c,m) U 0 ∩ X ( c , m ) for any ( c , m ) (c,m) ( c , m ) .
Moreover, to study Ind ( I C ( O x P , Q ‾ l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ l ) ⊠ ∏ I C ( O n h , s e m i , Q ‾ l ) ⊠ I C ( O x 1 , Q ‾ l ) ) ∣ Y ( c , m ) \operatorname{Ind}\left(IC(O_{x_P},\overline{\mathbb{Q}}_l)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_l)\boxtimes\prod IC(O_{n h,\mathrm{semi}},\overline{\mathbb{Q}}_l)\boxtimes IC(O_{x_1},\overline{\mathbb{Q}}_l)\right)|_{Y(c,m)} Ind ( I C ( O x P , Q l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q l ) ⊠ ∏ I C ( O nh , semi , Q l ) ⊠ I C ( O x 1 , Q l ) ) ∣ Y ( c , m ) , for ( c , m ) (c,m) ( c , m ) such that Y ( c , m ) ∩ U 0 ≠ ∅ Y(c,m)\cap U_0\ne\varnothing Y ( c , m ) ∩ U 0 = ∅ , by Proposition 1.3 and 1.7, it suffices to consider
{ ( x , F ) | x ∈ Y ( c , m ) , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , t = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , m 1 ⊂ V ) , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ x 1 , dim ( V 3 , j / V 3 , j − 1 ) = δ , M x ∣ V 2 , i / V 2 , i − 1 ≅ M n h , s e m i , M x ∣ V / V 3 , r ≅ x P } ⟶ Y ( c , m ) . \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in Y(c,m),\\
&F=(0=V_0\subset V_1=V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,t}=V_{3,0}\\
&\qquad\subset V_{3,1}\subset\cdots\subset V_{3,m_1}\subset V),\\
&x(F)\subset F,\qquad M_x|_{V_1/V_0}\cong x_1,\qquad\dim(V_{3,j}/V_{3,j-1})=\delta,\\
&M_x|_{V_{2,i}/V_{2,i-1}}\cong M_{n h,\mathrm{semi}},\\
&M_x|_{V/V_{3,r}}\cong x_P
\end{aligned}\right\}\longrightarrow Y(c,m). ⎩ ⎨ ⎧ ( x , F ) x ∈ Y ( c , m ) , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , t = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , m 1 ⊂ V ) , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ x 1 , dim ( V 3 , j / V 3 , j − 1 ) = δ , M x ∣ V 2 , i / V 2 , i − 1 ≅ M nh , semi , M x ∣ V / V 3 , r ≅ x P ⎭ ⎬ ⎫ ⟶ Y ( c , m ) .
By Remark 3.4, Theorem 3.3 can apply to this morphism.
We can regard U 0 ∩ X ( c , m ) U_0\cap X(c,m) U 0 ∩ X ( c , m ) as
U 0 ∩ ( x P × x 1 × ∏ ( U n h , j ∩ X ( c n h , j , m j ) ) × ( U h r ∩ X ( 0 , m ) ) ) , U_0\cap\left(x_P\times x_1\times\prod(U_{n h,j}\cap X(c_{n h,j},m_j))\times(U_{h r}\cap X(0,m))\right), U 0 ∩ ( x P × x 1 × ∏ ( U nh , j ∩ X ( c nh , j , m j )) × ( U h r ∩ X ( 0 , m )) ) ,
which is open in x P × x 1 × ∏ ( U n h , j ∩ X ( c n h , j , m j ) ) × ( U h r ∩ X ( 0 , m ) ) x_P\times x_1\times\prod(U_{n h,j}\cap X(c_{n h,j},m_j))\times(U_{h r}\cap X(0,m)) x P × x 1 × ∏ ( U nh , j ∩ X ( c nh , j , m j )) × ( U h r ∩ X ( 0 , m )) .
U h r ∩ X ( 0 , m ) U_{h r}\cap X(0,m) U h r ∩ X ( 0 , m ) is smooth.
And regard U 0 ∩ Y ( c , m ) U_0\cap Y(c,m) U 0 ∩ Y ( c , m ) as
U 0 ∩ ( x P × x 1 × ∏ ( U n h , j ∩ Y ( c n h , j , m j ) ) × ( U h r ∩ X ( 0 , m ) ) ) , U_0\cap\left(x_P\times x_1\times\prod(U_{n h,j}\cap Y(c_{n h,j},m_j))\times(U_{h r}\cap X(0,m))\right), U 0 ∩ ( x P × x 1 × ∏ ( U nh , j ∩ Y ( c nh , j , m j )) × ( U h r ∩ X ( 0 , m )) ) ,
which is open in x P × x 1 × ∏ ( U n h , j ∩ Y ( c n h , j , m j ) ) × ( U h r ∩ X ( 0 , m ) ) x_P\times x_1\times\prod(U_{n h,j}\cap Y(c_{n h,j},m_j))\times(U_{h r}\cap X(0,m)) x P × x 1 × ∏ ( U nh , j ∩ Y ( c nh , j , m j )) × ( U h r ∩ X ( 0 , m )) .
U h r ⊂ X ( 0 , m ) U_{h r}\subset X(0,m) U h r ⊂ X ( 0 , m ) is smooth.
Using the smooth morphism G V n h , i × U n h , i → E V n h , i G_{V_{n h,i}}\times U_{n h,i}\to E_{V_{n h,i}} G V nh , i × U nh , i → E V nh , i , we see that, for every x ∈ U n h , i x\in U_{n h,i} x ∈ U nh , i , we have T x U n h , i + T x O x = T x E V n h , i T_xU_{n h,i}+T_x\mathcal{O}_x=T_xE_{V_{n h,i}} T x U nh , i + T x O x = T x E V nh , i .
U n h , j ∩ X ( c n h , j , m j ) U_{n h,j}\cap X(c_{n h,j},m_j) U nh , j ∩ X ( c nh , j , m j ) and U n h , j ∩ Y ( c n h , j , m j ) U_{n h,j}\cap Y(c_{n h,j},m_j) U nh , j ∩ Y ( c nh , j , m j ) are smooth with dimension dim U n h , j + dim X ( c n h , j , m j ) − dim E V n h , j \dim U_{n h,j}+\dim X(c_{n h,j},m_j)-\dim E_{V_{n h,j}} dim U nh , j + dim X ( c nh , j , m j ) − dim E V nh , j because the intersections are transverse and Remark 1.8. Thus U 0 ∩ X ( c , m ) U_0\cap X(c,m) U 0 ∩ X ( c , m ) and U 0 ∩ Y ( c , m ) U_0\cap Y(c,m) U 0 ∩ Y ( c , m ) are smooth with the same dimension.
For λ = ( λ 1 , ⋯ , λ t ) \lambda=(\lambda_1,\cdots,\lambda_t) λ = ( λ 1 , ⋯ , λ t ) , consider the proper morphism
W c , λ = { ( x , F ) | x ∈ U 0 ∩ Y ( c , m ) , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , s = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , t ⊂ V ) , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ x 1 , M x ∣ V 2 , i / V 2 , i − 1 ≅ c n h , i , M x ∣ V 3 , j / V 3 , j − 1 is homogeneous regular of dimension λ j δ , M x ∣ V / V 3 , t ≅ x P } → t c , λ U 0 ∩ Y ( c , m ) . W_{c,\lambda}=\left\{(x,F)\ \middle|\ \begin{aligned}
&x\in U_0\cap Y(c,m),\\
&F=(0=V_0\subset V_1=V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,s}=V_{3,0}\\
&\qquad\subset V_{3,1}\subset\cdots\subset V_{3,t}\subset V),\\
&x(F)\subset F,\ M_x|_{V_1/V_0}\cong x_1,\qquad M_x|_{V_{2,i}/V_{2,i-1}}\cong c_{n h,i},\\
&M_x|_{V_{3,j}/V_{3,j-1}}\text{ is homogeneous regular of dimension }\lambda_j\delta,\\
&M_x|_{V/V_{3,t}}\cong x_P
\end{aligned}\right\}\xrightarrow{t_{c,\lambda}}U_0\cap Y(c,m). W c , λ = ⎩ ⎨ ⎧ ( x , F ) x ∈ U 0 ∩ Y ( c , m ) , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , s = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , t ⊂ V ) , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ x 1 , M x ∣ V 2 , i / V 2 , i − 1 ≅ c nh , i , M x ∣ V 3 , j / V 3 , j − 1 is homogeneous regular of dimension λ j δ , M x ∣ V / V 3 , t ≅ x P ⎭ ⎬ ⎫ t c , λ U 0 ∩ Y ( c , m ) .
Applying Theorem 3.3 to
Ind ( I C ( O x P , Q ‾ l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ l ) ⊠ ∏ I C ( O n h , s e m i , Q ‾ l ) ⊠ I C ( O x 1 , Q ‾ l ) ) ∣ Y ( c , m ) , \operatorname{Ind}\left(IC(O_{x_P},\overline{\mathbb{Q}}_l)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_l)\boxtimes\prod IC(O_{n h,\mathrm{semi}},\overline{\mathbb{Q}}_l)\boxtimes IC(O_{x_1},\overline{\mathbb{Q}}_l)\right)|_{Y(c,m)}, Ind ( I C ( O x P , Q l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q l ) ⊠ ∏ I C ( O nh , semi , Q l ) ⊠ I C ( O x 1 , Q l ) ) ∣ Y ( c , m ) ,
we find that
j 0 c , m ∗ Ind ( I C ( O x P , Q ‾ l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ l ) ⊠ ∏ I C ( O n h , s e m i , Q ‾ l ) ⊠ I C ( O x 1 , Q ‾ l ) ) j_0^{c,m*}\operatorname{Ind}\left(IC(O_{x_P},\overline{\mathbb{Q}}_l)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_l)\boxtimes\prod IC(O_{n h,\mathrm{semi}},\overline{\mathbb{Q}}_l)\boxtimes IC(O_{x_1},\overline{\mathbb{Q}}_l)\right) j 0 c , m ∗ Ind ( I C ( O x P , Q l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q l ) ⊠ ∏ I C ( O nh , semi , Q l ) ⊠ I C ( O x 1 , Q l ) )
is generated by the objects t c , λ Q ‾ l t_{c,\lambda}\overline{\mathbb{Q}}_l t c , λ Q l in D b ( U 0 c , m ) D^b(U_0^c,m) D b ( U 0 c , m ) . We claim that W c , λ W_{c,\lambda} W c , λ is smooth. To prove the claim, we regard it as an open subset of
{ ( x , F ) | x ∈ x P × x I × ∏ ( U n h , j ∩ Y ( c n h , j , m j ) ) × U h r , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , s = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , t ⊂ V ) , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ x I , M x ∣ V 2 , i / V 2 , i − 1 ≅ c n h , i , M x ∣ V 3 , j / V 3 , j − 1 is homogeneous regular of dimension λ j δ , M x ∣ V / V 3 , t ≅ x P , Hom ( M x ∣ V n h , j , M x ∣ V n h , j ′ ) = 0 , j ≠ j ′ , Hom ( M x ∣ V n h , j , M x ∣ ⨁ V h r , i ) = 0 } . \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in x_P\times x_I\times\prod(U_{nh,j}\cap Y(c_{nh,j},m_j))\times U_{hr},\\
&F=(0=V_0\subset V_1=V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,s}=V_{3,0}\\
&\qquad\qquad\subset V_{3,1}\subset\cdots\subset V_{3,t}\subset V),\\
&x(F)\subset F,\ M_x|_{V_1/V_0}\cong x_I,\quad M_x|_{V_{2,i}/V_{2,i-1}}\cong c_{nh,i},\\
&M_x|_{V_{3,j}/V_{3,j-1}}\text{ is homogeneous regular of dimension }\lambda_j\delta,\quad M_x|_{V/V_{3,t}}\cong x_P,\\
&\operatorname{Hom}(M_x|_{V_{nh,j}},M_x|_{V_{nh,j'}})=0,\quad j\ne j',\\
&\operatorname{Hom}(M_x|_{V_{nh,j}},M_x|_{\bigoplus V_{hr,i}})=0
\end{aligned}\right\}. ⎩ ⎨ ⎧ ( x , F ) x ∈ x P × x I × ∏ ( U nh , j ∩ Y ( c nh , j , m j )) × U h r , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , s = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , t ⊂ V ) , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ x I , M x ∣ V 2 , i / V 2 , i − 1 ≅ c nh , i , M x ∣ V 3 , j / V 3 , j − 1 is homogeneous regular of dimension λ j δ , M x ∣ V / V 3 , t ≅ x P , Hom ( M x ∣ V nh , j , M x ∣ V nh , j ′ ) = 0 , j = j ′ , Hom ( M x ∣ V nh , j , M x ∣ ⨁ V h r , i ) = 0 ⎭ ⎬ ⎫ .
On this variety, for j ≠ j ′ , i ≠ i ′ j\ne j',i\ne i' j = j ′ , i = i ′ , the representations M x ∣ V n h , j , M x ∣ V n h , j ′ , M x ∣ V h r , i M_x|_{V_{nh,j}}, M_x|_{V_{nh,j'}}, M_x|_{V_{hr,i}} M x ∣ V nh , j , M x ∣ V nh , j ′ , M x ∣ V h r , i and M x ∣ V h r , i ′ M_x|_{V_{hr,i'}} M x ∣ V h r , i ′ have no direct summands in the same tube.
When they have no direct summands in the same tube, stratifying by dim ( V n h , j ∩ V 3 , i ) \dim(V_{nh,j}\cap V_{3,i}) dim ( V nh , j ∩ V 3 , i ) and dim ( V h r , j ∩ V 3 , i ) \dim(V_{hr,j}\cap V_{3,i}) dim ( V h r , j ∩ V 3 , i ) , we obtain a decomposition in which each stratum is an open and closed subvariety. This stratum is isomorphic to some form
{ ( x , F ) | x ∈ U 0 ∩ ( x P × x I × ∏ U n h , j ∩ Y ( c n h , j , m j ) × ∏ U h r , j ) , F = ( 0 = V 0 ⊂ V 1 ⊂ V 2 , 0 1 ⊂ ⋯ ⊂ V 2 , m 1 ′ 1 ⊂ V 2 , 0 2 ⊂ ⋯ ⊂ V 2 , m 2 ′ 2 ⊂ ⋯ ⊂ V 2 , 0 s ⊂ ⋯ ⊂ V 2 , m s ′ s = V 3 , 0 ⊂ ⋯ ⊂ V 3 , m = V ) , x ( F ) ⊂ F , V 1 = V I , V 2 , m i ′ i = V I ⊕ ( ⨁ j ≤ i V n h , j ) , V 3 , i = V I ⊕ ( ⨁ j ≤ i V n h , j ) ⊕ ( ⨁ s ≤ i V h r , s ) , M x ∣ V 1 / V 0 ≅ x I , M x ∣ V / V 3 , m ≅ x P , M x ∣ V 2 , 0 1 / V 1 ≅ c n h , 1 , M x ∣ V 2 , i s / V 2 , m i − 1 ′ s − 1 ≅ c n h , i , s ≥ i > 1 , M x ∣ V 2 , i j / V 2 , i − 1 j is homogeneous regular of dimension μ i j δ , m i ′ ≥ i ≥ 1 , x ∣ V 3 , j / V 3 , j − 1 ∈ U h r , j , 0 < j ≤ m . } . \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in U_0\cap(x_P\times x_I\times\prod U_{nh,j}\cap Y(c_{nh,j},m_j)\times\prod U_{hr,j}),\\
&F=(0=V_0\subset V_1\subset V_{2,0}^{1}\subset\cdots\subset V_{2,m'_1}^{1}\subset V_{2,0}^{2}\subset\cdots\subset V_{2,m'_2}^{2}\subset\cdots\\
&\qquad\subset V_{2,0}^{s}\subset\cdots\subset V_{2,m'_s}^{s}=V_{3,0}\subset\cdots\subset V_{3,m}=V),\\
&x(F)\subset F,\ V_1=V_I,\ V_{2,m'_i}^{i}=V_I\oplus\left(\bigoplus_{j\leq i}V_{nh,j}\right),\
V_{3,i}=V_I\oplus\left(\bigoplus_{j\leq i}V_{nh,j}\right)\oplus\left(\bigoplus_{s\leq i}V_{hr,s}\right),\\
&M_x|_{V_1/V_0}\cong x_I,\ M_x|_{V/V_{3,m}}\cong x_P,\ M_x|_{V_{2,0}^{1}/V_1}\cong c_{nh,1},\
M_x|_{V_{2,i}^{s}/V_{2,m'_{i-1}}^{s-1}}\cong c_{nh,i},\ s\geq i>1,\\
&M_x|_{V_{2,i}^{j}/V_{2,i-1}^{j}}\text{ is homogeneous regular of dimension }\mu_i^j\delta,\ m'_i\geq i\geq1,\ x|_{V_{3,j}/V_{3,j-1}}\in U_{hr,j},\ 0<j\leq m.
\end{aligned}\right\}. ⎩ ⎨ ⎧ ( x , F ) x ∈ U 0 ∩ ( x P × x I × ∏ U nh , j ∩ Y ( c nh , j , m j ) × ∏ U h r , j ) , F = ( 0 = V 0 ⊂ V 1 ⊂ V 2 , 0 1 ⊂ ⋯ ⊂ V 2 , m 1 ′ 1 ⊂ V 2 , 0 2 ⊂ ⋯ ⊂ V 2 , m 2 ′ 2 ⊂ ⋯ ⊂ V 2 , 0 s ⊂ ⋯ ⊂ V 2 , m s ′ s = V 3 , 0 ⊂ ⋯ ⊂ V 3 , m = V ) , x ( F ) ⊂ F , V 1 = V I , V 2 , m i ′ i = V I ⊕ ( j ≤ i ⨁ V nh , j ) , V 3 , i = V I ⊕ ( j ≤ i ⨁ V nh , j ) ⊕ ( s ≤ i ⨁ V h r , s ) , M x ∣ V 1 / V 0 ≅ x I , M x ∣ V / V 3 , m ≅ x P , M x ∣ V 2 , 0 1 / V 1 ≅ c nh , 1 , M x ∣ V 2 , i s / V 2 , m i − 1 ′ s − 1 ≅ c nh , i , s ≥ i > 1 , M x ∣ V 2 , i j / V 2 , i − 1 j is homogeneous regular of dimension μ i j δ , m i ′ ≥ i ≥ 1 , x ∣ V 3 , j / V 3 , j − 1 ∈ U h r , j , 0 < j ≤ m . ⎭ ⎬ ⎫ .
Now we only need to consider
∏ j { ( x , F ) | x ∈ U n h , j ∩ Y ( c n h , j , m j ) , F = ( 0 = V 1 ⊂ V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , m j ′ = V n h , j ) , x ( F ) ⊂ F , M x ∣ V 2 , 0 ≅ c n h , j , M x ∣ V 2 , i / V 2 , i − 1 is homogeneous regular of dimension μ i j δ } × ∏ j U h r , j \prod_{j}\left\{(x,F)\ \middle|\ \begin{aligned}
&x\in U_{nh,j}\cap Y(c_{nh,j},m_j),\\
&F=(0=V_1\subset V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,m'_j}=V_{nh,j}),\ x(F)\subset F,\\
&M_x|_{V_{2,0}}\cong c_{nh,j},\quad M_x|_{V_{2,i}/V_{2,i-1}}\text{ is homogeneous regular of dimension }\mu_i^j\delta
\end{aligned}\right\}\times\prod_j U_{hr,j} j ∏ ⎩ ⎨ ⎧ ( x , F ) x ∈ U nh , j ∩ Y ( c nh , j , m j ) , F = ( 0 = V 1 ⊂ V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , m j ′ = V nh , j ) , x ( F ) ⊂ F , M x ∣ V 2 , 0 ≅ c nh , j , M x ∣ V 2 , i / V 2 , i − 1 is homogeneous regular of dimension μ i j δ ⎭ ⎬ ⎫ × j ∏ U h r , j
since the stratum we construct is isomorphic to an open subset in it. It is clearly that U h r , j = X ( 0 , 1 ) ∩ T h r , j U_{hr,j}=X(0,1)\cap\mathcal{T}_{hr,j} U h r , j = X ( 0 , 1 ) ∩ T h r , j is smooth. And
{ ( x , F ) | x ∈ Y ( c n h , j , m j ) , F = ( 0 = V 1 ⊂ V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , m j ′ = V n h , j ) , x ( F ) ⊂ F , M x ∣ V 2 , 0 ≅ c n h , j , M x ∣ V 2 , i / V 2 , i − 1 is homogeneous regular of dimension μ i j δ } \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in Y(c_{nh,j},m_j),\\
&F=(0=V_1\subset V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,m'_j}=V_{nh,j}),\ x(F)\subset F,\\
&M_x|_{V_{2,0}}\cong c_{nh,j},\quad M_x|_{V_{2,i}/V_{2,i-1}}\text{ is homogeneous regular of dimension }\mu_i^j\delta
\end{aligned}\right\} ⎩ ⎨ ⎧ ( x , F ) x ∈ Y ( c nh , j , m j ) , F = ( 0 = V 1 ⊂ V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , m j ′ = V nh , j ) , x ( F ) ⊂ F , M x ∣ V 2 , 0 ≅ c nh , j , M x ∣ V 2 , i / V 2 , i − 1 is homogeneous regular of dimension μ i j δ ⎭ ⎬ ⎫
is smooth by the same way in Proposition 1.5.
The intersection of smooth varieties
{ ( x , F ) | x ∈ Y ( c n h , j , m j ) , F = ( 0 = V 1 ⊂ V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , m j ′ = V n h , j ) , x ( F ) ⊂ F , M x ∣ V 2 , 0 ≅ c n h , j , M x ∣ V 2 , i / V 2 , i − 1 is homogeneous regular of dimension μ i j δ } ∩ U n h , j × F c n h , j , μ j , \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in Y(c_{nh,j},m_j),\\
&F=(0=V_1\subset V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,m'_j}=V_{nh,j}),\ x(F)\subset F,\\
&M_x|_{V_{2,0}}\cong c_{nh,j},\qquad M_x|_{V_{2,i}/V_{2,i-1}}\text{ is homogeneous regular of dimension }\mu_i^j\delta
\end{aligned}\right\}
\cap U_{nh,j}\times\mathcal{F}_{c_{nh,j},\mu^j}, ⎩ ⎨ ⎧ ( x , F ) x ∈ Y ( c nh , j , m j ) , F = ( 0 = V 1 ⊂ V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , m j ′ = V nh , j ) , x ( F ) ⊂ F , M x ∣ V 2 , 0 ≅ c nh , j , M x ∣ V 2 , i / V 2 , i − 1 is homogeneous regular of dimension μ i j δ ⎭ ⎬ ⎫ ∩ U nh , j × F c nh , j , μ j ,
where F c n h , j , μ j \mathcal{F}_{c_{nh,j},\mu^j} F c nh , j , μ j is the smooth variety
{ F = ( 0 = V − 1 ⊂ V 0 ⊂ V 1 ⊂ ⋯ ⊂ V m j ′ = V n h , j ) | dim V i / V i − 1 = μ i j δ , 0 < i ≤ m j ′ , dim V 0 / V − 1 = dim c n h , j } , \left\{F=(0=V_{-1}\subset V_0\subset V_1\subset\cdots\subset V_{m'_j}=V_{nh,j})\ \middle|\
\begin{aligned}
&\dim V_i/V_{i-1}=\mu_i^j\delta,\ 0<i\leq m'_j,\\
&\dim V_0/V_{-1}=\dim c_{nh,j}
\end{aligned}\right\}, { F = ( 0 = V − 1 ⊂ V 0 ⊂ V 1 ⊂ ⋯ ⊂ V m j ′ = V nh , j ) dim V i / V i − 1 = μ i j δ , 0 < i ≤ m j ′ , dim V 0 / V − 1 = dim c nh , j } ,
is transverse in E V n h , j × F c n h , j , μ j \mathbf{E}_{V_{nh,j}}\times\mathcal{F}_{c_{nh,j},\mu^j} E V nh , j × F c nh , j , μ j . The proof is similar to the proof that U n h , j ∩ Y ( c n h , j , m j ) U_{nh,j}\cap Y(c_{nh,j},m_j) U nh , j ∩ Y ( c nh , j , m j ) is a transversal intersection. The intersection is therefore smooth, which proves the claim.
More precisely, we can compute the dimension of
∏ { ( x , F ) | x ∈ U n h , j ∩ Y ( c n h , j , m j ) , F = ( 0 = V 1 ⊂ V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , m j ′ = V n h , j ) , x ( F ) ⊂ F , M x ∣ V 2 , 0 ≅ c n h , j , M x ∣ V 2 , i / V 2 , i − 1 is homogeneous regular of dimension μ i j δ } \prod\left\{(x,F)\ \middle|\ \begin{aligned}
&x\in U_{nh,j}\cap Y(c_{nh,j},m_j),\\
&F=(0=V_1\subset V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,m'_j}=V_{nh,j}),\ x(F)\subset F,\\
&M_x|_{V_{2,0}}\cong c_{nh,j},\qquad M_x|_{V_{2,i}/V_{2,i-1}}\text{ is homogeneous regular of dimension }\mu_i^j\delta
\end{aligned}\right\} ∏ ⎩ ⎨ ⎧ ( x , F ) x ∈ U nh , j ∩ Y ( c nh , j , m j ) , F = ( 0 = V 1 ⊂ V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , m j ′ = V nh , j ) , x ( F ) ⊂ F , M x ∣ V 2 , 0 ≅ c nh , j , M x ∣ V 2 , i / V 2 , i − 1 is homogeneous regular of dimension μ i j δ ⎭ ⎬ ⎫
× ∏ U h r , j \times\prod U_{hr,j} × ∏ U h r , j , which is ∑ j dim ( U n h , j ∩ X ( c n h , j , m j ) ) + dim U h r \sum_j\dim(U_{nh,j}\cap X(c_{nh,j},m_j))+\dim U_{hr} ∑ j dim ( U nh , j ∩ X ( c nh , j , m j )) + dim U h r , the same as the dimension of U 0 ∩ X ( c , m ) U_0\cap X(c,m) U 0 ∩ X ( c , m ) . Thus dim W c . λ = dim U 0 ∩ X ( c , m ) \dim W_{c.\lambda}=\dim U_0\cap X(c,m) dim W c . λ = dim U 0 ∩ X ( c , m ) .
By Remark 3.5,
j ~ 0 c , m ∗ Ind ( I C ( O x p , Q ‾ l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ l ) ⊠ ∏ I C ( O n h , s e m i , Q ‾ l ) ⊠ I C ( O x 1 , Q ‾ l ) ) \widetilde{j}^{c,m*}_0\operatorname{Ind}\left(IC(\mathcal{O}_{x_p},\overline{\mathbb{Q}}_l)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_l)\boxtimes\prod IC(\mathcal{O}_{nh,semi},\overline{\mathbb{Q}}_l)\boxtimes IC(\mathcal{O}_{x_1},\overline{\mathbb{Q}}_l)\right) j 0 c , m ∗ Ind ( I C ( O x p , Q l ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q l ) ⊠ ∏ I C ( O nh , se mi , Q l ) ⊠ I C ( O x 1 , Q l ) )
is a lisse sheaf complex. Since the variety is smooth, pointwise purity implies purity by [9 6.2.5] .
We now only need to consider the connected component W W W of
{ ( x , F ) | x ∈ U 0 ∩ Y ( c , m ) , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , s = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , t ⊂ V ) , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ c 1 , M x ∣ V 2 , i / V 2 , i − 1 ≅ c n h , j , M x ∣ V 3 , j / V 3 , j − 1 is homogeneous regular of dimension λ j δ , M x ∣ V / V 3 , t ≅ c P } . \left\{(x,F)\ \middle|\ \begin{aligned}
&x\in U_0\cap Y(c,m),\\
&F=(0=V_0\subset V_1=V_{2,0}\subset V_{2,1}\subset\cdots\subset V_{2,s}=V_{3,0}\\
&\hspace{5em}\subset V_{3,1}\subset\cdots\subset V_{3,t}\subset V),\\
&x(F)\subset F,\ M_x|_{V_1/V_0}\cong c_1,\quad M_x|_{V_{2,i}/V_{2,i-1}}\cong c_{nh,j},\\
&M_x|_{V_{3,j}/V_{3,j-1}}\text{ is homogeneous regular of dimension }\lambda_j\delta,\quad M_x|_{V/V_{3,t}}\cong c_P
\end{aligned}\right\}. ⎩ ⎨ ⎧ ( x , F ) x ∈ U 0 ∩ Y ( c , m ) , F = ( 0 = V 0 ⊂ V 1 = V 2 , 0 ⊂ V 2 , 1 ⊂ ⋯ ⊂ V 2 , s = V 3 , 0 ⊂ V 3 , 1 ⊂ ⋯ ⊂ V 3 , t ⊂ V ) , x ( F ) ⊂ F , M x ∣ V 1 / V 0 ≅ c 1 , M x ∣ V 2 , i / V 2 , i − 1 ≅ c nh , j , M x ∣ V 3 , j / V 3 , j − 1 is homogeneous regular of dimension λ j δ , M x ∣ V / V 3 , t ≅ c P ⎭ ⎬ ⎫ .
It has the same dimension as U 0 ∩ X ( c , m ) U_0\cap X(c,m) U 0 ∩ X ( c , m ) .
The morphism W → ϕ U 0 ∩ Y ( c , m ) W\xrightarrow{\phi}U_0\cap Y(c,m) W ϕ U 0 ∩ Y ( c , m ) is small with respect to U 0 ∩ X ( c , m ) U_0\cap X(c,m) U 0 ∩ X ( c , m ) for the following reasons.
Properness follows from the construction. And for any x ∈ U 0 ∩ Y ( c , m ) x\in U_0\cap Y(c,m) x ∈ U 0 ∩ Y ( c , m ) , j ≠ j ′ j\ne j' j = j ′ , i ≠ i ′ i\ne i' i = i ′ , M x ∣ V n h , j M_x|_{V_{nh,j}} M x ∣ V nh , j , M x ∣ V n h , j ′ M_x|_{V_{nh,j'}} M x ∣ V nh , j ′ , M x ∣ V h r , i M_x|_{V_{hr,i}} M x ∣ V h r , i and M x ∣ V h r , i ′ M_x|_{V_{hr,i'}} M x ∣ V h r , i ′ have no direct summands in the same tube. Then U 0 ∩ X ( c , m ) = U 0 ∩ ( x p × x I × ∏ j X ( c n h , j , m j ) × U h r ) U_0\cap X(c,m)=U_0\cap(x_p\times x_I\times\prod_jX(c_{nh,j},m_j)\times U_{hr}) U 0 ∩ X ( c , m ) = U 0 ∩ ( x p × x I × ∏ j X ( c nh , j , m j ) × U h r ) .
∀ y ∈ U 0 ∩ X ( c , m ) \forall y\in U_0\cap X(c,m) ∀ y ∈ U 0 ∩ X ( c , m ) , ϕ − 1 ( y ) \phi^{-1}(y) ϕ − 1 ( y ) is a finite set.
∃ \exists ∃ a stratification X ( c n h , j , m j ) ⊔ ( ⨆ Y n h , j , t ) X(c_{nh,j},m_j)\sqcup(\bigsqcup Y_{nh,j,t}) X ( c nh , j , m j ) ⊔ ( ⨆ Y nh , j , t ) of Y ( c n h , j , m j ) Y(c_{nh,j},m_j) Y ( c nh , j , m j ) as in Lemma 2.3 and [33 Lemma 2.5] , such that
dim ( X ( c n h , j , m j ) ∩ U n h , j ) − dim ( Y n h , j , t ∩ U n h , j ) = dim ( X ( c n h , j , m j ) ) − dim ( Y n h , j , t ) , \dim(X(c_{nh,j},m_j)\cap U_{nh,j})-\dim(Y_{nh,j,t}\cap U_{nh,j})=\dim(X(c_{nh,j},m_j))-\dim(Y_{nh,j,t}), dim ( X ( c nh , j , m j ) ∩ U nh , j ) − dim ( Y nh , j , t ∩ U nh , j ) = dim ( X ( c nh , j , m j )) − dim ( Y nh , j , t ) ,
because U n h , j U_{nh,j} U nh , j is transverse to both Y n h , j , t Y_{nh,j,t} Y nh , j , t and X ( c n h , j , m j ) X(c_{nh,j},m_j) X ( c nh , j , m j ) . And for any Y n h , j , t Y_{nh,j,t} Y nh , j , t other than X ( c n h , j , m j ) X(c_{nh,j},m_j) X ( c nh , j , m j ) , dim ( Y n h , j , t ∩ U n h , j ) < dim ( X ( c n h , j , m j ) ∩ U n h , j ) \dim(Y_{nh,j,t}\cap U_{nh,j})<\dim(X(c_{nh,j},m_j)\cap U_{nh,j}) dim ( Y nh , j , t ∩ U nh , j ) < dim ( X ( c nh , j , m j ) ∩ U nh , j ) . Moreover,
U 0 ∩ Y ( c , m ) = ( U 0 ∩ X ( c , m ) ) ∪ ⋃ f i n i t e ( U 0 ∩ ( x p × x I × ∏ j ( Y n h , j , t j ∩ U n h , j ) × U h r ) ) . U_0\cap Y(c,m)=(U_0\cap X(c,m))\cup\bigcup_{\mathit{finite}}\left(U_0\cap(x_p\times x_I\times\prod_j(Y_{nh,j,t_j}\cap U_{nh,j})\times U_{hr})\right). U 0 ∩ Y ( c , m ) = ( U 0 ∩ X ( c , m )) ∪ finite ⋃ ( U 0 ∩ ( x p × x I × j ∏ ( Y nh , j , t j ∩ U nh , j ) × U h r ) ) .
By calculating the dimension, U 0 ∩ X ( c , m ) U_0 \cap X(c,m) U 0 ∩ X ( c , m ) is dense in U 0 ∩ Y ( c , m ) U_0 \cap Y(c,m) U 0 ∩ Y ( c , m ) .
Applying Lemma 2.3 and [33 Lemma 2.5] , for y ∈ U 0 ∩ ( x p × x I × ∏ j ( Y n h , j , t j ∩ U n h , j ) × U h r ) y \in U_0 \cap(x_p \times x_I \times\prod_j (Y_{n_h,j,t_j} \cap U_{n_h,j}) \times U_{h_r}) y ∈ U 0 ∩ ( x p × x I × ∏ j ( Y n h , j , t j ∩ U n h , j ) × U h r ) , when there exist j j j such that Y n h , j , t j ≠ X ( c n h , j , m j ) Y_{n_h,j,t_j} \ne X(c_{n_h,j},m_j) Y n h , j , t j = X ( c n h , j , m j ) , dim ϕ − 1 ( y ) < 1 2 ∑ j ( dim X ( c n h , j , m j ) − dim Y n h , j , t j ) \dim\phi^{-1}(y) < \frac{1}{2}\sum_j(\dim X(c_{n_h,j},m_j)-\dim Y_{n_h,j,t_j}) dim ϕ − 1 ( y ) < 2 1 ∑ j ( dim X ( c n h , j , m j ) − dim Y n h , j , t j ) , since M X ∣ V n h , j M_X|_{V_{n_h,j}} M X ∣ V n h , j and M X ∣ V n h , j ′ M_X|_{V_{n_h,j'}} M X ∣ V n h , j ′ have no direct summands in the same tube for different j j j and j ′ j' j ′ .
Thus the morphism W → ϕ U 0 ∩ Y ( c , m ) W \xrightarrow{\phi} U_0 \cap Y(c,m) W ϕ U 0 ∩ Y ( c , m ) is small respect to U 0 ∩ X ( c , m ) U_0 \cap X(c,m) U 0 ∩ X ( c , m ) .
Let U ~ 0 c , m → j 0 c , m U 0 c , m \widetilde{U}_0^{c,m} \xrightarrow{j_0^{c,m}} U_0^{c,m} U 0 c , m j 0 c , m U 0 c , m denote the open embedding. And L χ ∣ U 0 ∩ X ( c , m ) \mathcal{L}_{\chi}|_{U_0\cap X(c,m)} L χ ∣ U 0 ∩ X ( c , m ) is lisse sheaf and pointwise pure. With the smoothness of U 0 ∩ X ( c , m ) U_0\cap X(c,m) U 0 ∩ X ( c , m ) , L χ ∣ U 0 ∩ X ( c , m ) \mathcal{L}_{\chi}|_{U_0\cap X(c,m)} L χ ∣ U 0 ∩ X ( c , m ) is pure and thus semisimple in D b ( U ~ 0 c , m ) D^b(\widetilde{U}_0^{c,m}) D b ( U 0 c , m ) .
Moreover, we know that L χ ∣ U 0 ∩ X ( c , m ) \mathcal{L}_{\chi}|_{U_0\cap X(c,m)} L χ ∣ U 0 ∩ X ( c , m ) is semisimple local system in D b ( U ~ 0 c , m ) D^b(\widetilde{U}_0^{c,m}) D b ( U 0 c , m ) .
t c , λ ! Q ‾ ℓ t_{c,\lambda!}\overline{\mathbb{Q}}_\ell t c , λ ! Q ℓ is generated by simple perverse sheaf j c , m ! ∗ ( L ′ [ dim ( U 0 ∩ X ( c , m ) ) ] ) j_{c,m!*}(L'[\dim(U_0\cap X(c,m))]) j c , m ! ∗ ( L ′ [ dim ( U 0 ∩ X ( c , m ))]) , where L ′ L' L ′ is a simple local system summand of L χ ∣ U 0 ∩ X ( c , m ) \mathcal{L}_{\chi}|_{U_0\cap X(c,m)} L χ ∣ U 0 ∩ X ( c , m ) , for some χ ∈ X ( S m ) \chi\in X(S_m) χ ∈ X ( S m ) , in D b ( U 0 c , m ) D^b(U_0^{c,m}) D b ( U 0 c , m ) .
Now set
K = Ind ( I C ( O ‾ x P , Q ‾ ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ ℓ ) ⊠ ∏ I C ( O ‾ n h , s e m i , Q ‾ ℓ ) ⊠ I C ( O ‾ x I , Q ‾ ℓ ) ) . K=\operatorname{Ind}\left(IC(\overline{O}_{x_P},\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(\overline{O}_{n_h,\mathrm{semi}},\overline{\mathbb{Q}}_\ell)\boxtimes IC(\overline{O}_{x_I},\overline{\mathbb{Q}}_\ell)\right). K = Ind ( I C ( O x P , Q ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ℓ ) ⊠ ∏ I C ( O n h , semi , Q ℓ ) ⊠ I C ( O x I , Q ℓ ) ) .
The purity of j ~ 0 c , m ∗ K \widetilde{j}_0^{c,m*}K j 0 c , m ∗ K implies that j 0 c , m ∗ K j_0^{c,m*}K j 0 c , m ∗ K is pure, as we now show.
j 0 c , m ∗ K j_0^{c,m*}K j 0 c , m ∗ K is generated by elements t c , λ ! Q ‾ ℓ t_{c,\lambda!}\overline{\mathbb{Q}}_\ell t c , λ ! Q ℓ . By the smallness of the morphism ϕ \phi ϕ , every simple perverse generator of j 0 c , m ∗ K j_0^{c,m*}K j 0 c , m ∗ K has the form j c , m ! ∗ P j_{c,m!*}P j c , m ! ∗ P , where P P P is a simple perverse sheaf on U ~ 0 c , m \widetilde{U}_0^{c,m} U 0 c , m . Moreover, P P P is a local system up to a shift.
Now consider p H i ( j 0 c , m ∗ K ) {}^pH^i(j_0^{c,m*}K) p H i ( j 0 c , m ∗ K ) and its weight filtration:
0 ⊂ P 1 ⊂ P 2 ⊂ ⋯ ⊂ p H i ( j 0 c , m ∗ K ) . 0\subset P_1\subset P_2\subset\cdots\subset{}^pH^i(j_0^{c,m*}K). 0 ⊂ P 1 ⊂ P 2 ⊂ ⋯ ⊂ p H i ( j 0 c , m ∗ K ) .
The functor j 0 c , m ∗ j_0^{c,m*} j 0 c , m ∗ preserves this weight filtration. Moreover, p H i ( j 0 c , m ∗ K ) ≃ H i − dim U 0 c , m ( j ~ 0 c , m ∗ K ) [ dim U 0 c , m ] {}^pH^i(j_0^{c,m*}K)\simeq H^{i-\dim U_0^{c,m}}(\widetilde{j}_0^{c,m*}K)[\dim U_0^{c,m}] p H i ( j 0 c , m ∗ K ) ≃ H i − d i m U 0 c , m ( j 0 c , m ∗ K ) [ dim U 0 c , m ] , since H k ( j ~ 0 c , m ∗ K ) H^k(\widetilde{j}_0^{c,m*}K) H k ( j 0 c , m ∗ K ) is a lisse sheaf complex for any k k k .
This complex is pure of weight i i i .
Moreover, j 0 c , m ∗ ( P i / P j ) = 0 j_0^{c,m*}(P_i/P_j)=0 j 0 c , m ∗ ( P i / P j ) = 0 implies P i / P j = 0 P_i/P_j=0 P i / P j = 0 , since every generator has the form j c , m ! ∗ P j_{c,m!*}P j c , m ! ∗ P and the faithfulness in [1 Proposition 5.3.9] .
It follows that p H i ( j 0 c , m ∗ K ) {}^pH^i(j_0^{c,m*}K) p H i ( j 0 c , m ∗ K ) is pure of the same weight as p H i ( j ~ 0 c , m ∗ K ) {}^pH^i(\widetilde{j}_0^{c,m*}K) p H i ( j 0 c , m ∗ K ) . Thus j 0 c , m ∗ K j_0^{c,m*}K j 0 c , m ∗ K is also pure and hence semisimple in D b ( U 0 c , m ) D^b(U_0^{c,m}) D b ( U 0 c , m ) . Consequently, j 0 c , m ∗ L 1 j_0^{c,m*}\mathcal{L}_1 j 0 c , m ∗ L 1 is semisimple in D b ( U 0 c , m ) D^b(U_0^{c,m}) D b ( U 0 c , m ) .
It follows that j 0 c 2 , m 2 ∗ L j_0^{c_2,m_2*}\mathcal{L} j 0 c 2 , m 2 ∗ L is semisimple and by Lemma 5.1, L 2 \mathcal{L}_2 L 2 can be written as a sum of summands (up to shift) of
u 0 ∗ Ind ( I C ( O ‾ x P , Q ‾ ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ‾ ℓ ) ⊠ ∏ I C ( O ‾ c n h , j , Q ‾ ℓ ) ⊠ I C ( O ‾ x I , Q ‾ ℓ ) ) . u_0^*\operatorname{Ind}\left(IC(\overline{O}_{x_P},\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(X(0,1),\overline{\mathbb{Q}}_\ell)\boxtimes\prod IC(\overline{O}_{c_{n_h,j}},\overline{\mathbb{Q}}_\ell)\boxtimes IC(\overline{O}_{x_I},\overline{\mathbb{Q}}_\ell)\right). u 0 ∗ Ind ( I C ( O x P , Q ℓ ) ⊠ ∏ I C ( X ( 0 , 1 ) , Q ℓ ) ⊠ ∏ I C ( O c n h , j , Q ℓ ) ⊠ I C ( O x I , Q ℓ ) ) .
By descending induction on the support of L i \mathcal{L}_i L i , we conclude that j ~ 0 c , m ∗ For G ν L \widetilde{j}_0^{c,m*}\operatorname{For}^{G^\nu}\mathcal{L} j 0 c , m ∗ For G ν L is pointwise pure. Thus L \mathcal{L} L is pointwise pure on X ( c , m ) X(c,m) X ( c , m ) . □
Acknowledgements
Y. Wu and J. Xiao are partially supported by National Natural Science Foundation of China [Grant No. 12471030]. An initial version of the proof of Lemma 3.1 was developed with AI assistance, and its correctness was subsequently verified by the authors. All other mathematical content was developed independently of AI. AI tools were also used to correct grammar and spelling.
BEIJING INTERNATIONAL CENTER FOR MATHEMATICAL RESEARCH, BEIJING 100871, P. R. CHINA
Email address: 2506397175@pku.edu.cn (Y. Wu)
School of Mathematical Sciences, Beijing Normal University, Beijing 100875, P. R. China
Email address: jxiao@bnu.edu.cn (J. Xiao)