The purity conjecture
Let Σ g \Sigma_g Σ g be a smooth connected and compact topological surface of genus g ≥ 2 g \ge2 g ≥ 2 . For r ≥ 1 r \ge1 r ≥ 1 and d ∈ ( Z / r Z ) × d \in(\mathbb{Z}/r\mathbb{Z})^\times d ∈ ( Z / r Z ) × , the twisted character variety is defined as
M B ( r , d ) = { ( A 1 , B 1 , … , A g , B g ) ∈ G L r 2 g ( C ) ; ∏ i [ A i , B i ] = e 2 i π d / r } / / G L r . (1) M^B(r,d)=\left\{(A_1,B_1,\ldots,A_g,B_g)\in GL_r^{2g}(\mathbb{C});\ \prod_i[A_i,B_i]=e^{2i\pi d/r}\right\}//GL_r. \tag*{(1)} M B ( r , d ) = { ( A 1 , B 1 , … , A g , B g ) ∈ G L r 2 g ( C ) ; i ∏ [ A i , B i ] = e 2 iπ d / r } // G L r . (1)
It may be interpreted as the moduli space of G L r ( C ) GL_r(\mathbb{C}) G L r ( C ) -representations of the once-punctured Σ g \Sigma_g Σ g , with monodromy around the puncture given by e 2 i π d / r e^{2i\pi d/r} e 2 iπ d / r . The study of the cohomology of this affine algebraic variety was initiated by T. Hausel and F. Rodriguez-Villegas in the seminal paper [14 ] , who showed in particular that M B ( r , d ) M^B(r,d) M B ( r , d ) is smooth, connected, of dimension 2 N = 2 ( g − 1 ) r 2 + 2 2N=2(g-1)r^2+2 2 N = 2 ( g − 1 ) r 2 + 2 and carries a natural symplectic form.
In particular, the mixed Hodge structure on H m ( M B ( r , d ) , Q ) H^{m}(M^{B}(r,d),\mathbb{Q}) H m ( M B ( r , d ) , Q ) has weights in the interval [ m , 2 m ] [m,2m] [ m , 2 m ] . It is of Hodge-Tate type by [40 ] . The pure part of H ∗ ( M B ( r , d ) ) H^{*}(M^{B}(r,d)) H ∗ ( M B ( r , d )) is
H p u r e ∗ ( M B ( r , d ) , Q ) = ⨁ m W m H m ( M B ( r , d ) , Q ) H^{*}_{\mathrm{pure}}(M^{B}(r,d),\mathbb{Q})=\bigoplus_{m}W_{m}H^{m}(M^{B}(r,d),\mathbb{Q}) H pure ∗ ( M B ( r , d ) , Q ) = m ⨁ W m H m ( M B ( r , d ) , Q )
It is concentrated in even degrees, since W 2 k + 1 = W 2 k W_{2k+1}=W_{2k} W 2 k + 1 = W 2 k and W 2 k H 2 k + 1 = 0 W_{2k}H^{2k+1}=0 W 2 k H 2 k + 1 = 0 .
Let S g S_{g} S g stand for the quiver with one vertex and g g g loops, and let A S g , r ( q ) ∈ N [ q ] A_{S_{g},r}(q)\in\mathbb{N}[q] A S g , r ( q ) ∈ N [ q ] be the Kac polynomial of S g S_{g} S g counting absolutely indecomposable S g S_{g} S g -representations over F q \mathbb{F}_{q} F q of dimension r r r . It is of degree N = ( g − 1 ) r 2 + 1 N=(g-1)r^{2}+1 N = ( g − 1 ) r 2 + 1 .
As a special case of their conjectural expression for the full mixed Hodge polynomial of M B ( r , d ) M^{B}(r,d) M B ( r , d ) , Hausel and Rodriguez-Villegas formulated the following purity conjecture.
Conjecture 1.1 ([14 ] ). For all r ≥ 1 r\geq1 r ≥ 1 and d d d coprime to r r r we have
∑ k dim W 2 k H 2 k ( M B ( r , d ) , Q ) q k = q ( g − 1 ) r 2 + 1 A S g , r ( q − 1 ) . \sum_{k}\dim W_{2k}H^{2k}(M^{B}(r,d),\mathbb{Q})q^{k}=q^{(g-1)r^{2}+1}A_{S_{g},r}(q^{-1}). k ∑ dim W 2 k H 2 k ( M B ( r , d ) , Q ) q k = q ( g − 1 ) r 2 + 1 A S g , r ( q − 1 ) .
Equivalently, by Poincaré duality, the graded dimension of the pure part of compactly supported cohomology is q ( g − 1 ) r 2 + 1 A S g , r ( q ) q^{(g-1)r^{2}+1}A_{S_{g},r}(q) q ( g − 1 ) r 2 + 1 A S g , r ( q ) .
Example 1.2.
(a) For r = 1 r=1 r = 1 : M B ( 1 , 0 ) = ( C ∗ ) 2 g M^{B}(1,0)=(\mathbb{C}^{*})^{2g} M B ( 1 , 0 ) = ( C ∗ ) 2 g and the pure part is H 0 H^{0} H 0 . Indeed q g A 1 S g ( q − 1 ) = q g q − g = 1 q^{g}A_{1}^{S_{g}}(q^{-1})=q^{g}q^{-g}=1 q g A 1 S g ( q − 1 ) = q g q − g = 1 .
(b) For r = 2 r=2 r = 2 : A 2 S g ( q ) = q 2 g − 1 ( 1 + q 2 + ⋯ + q 2 g − 2 ) A_{2}^{S_{g}}(q)=q^{2g-1}(1+q^{2}+\cdots+q^{2g-2}) A 2 S g ( q ) = q 2 g − 1 ( 1 + q 2 + ⋯ + q 2 g − 2 ) , so the conjecture predicts one pure class in each of the degrees 0 , 4 , … , 4 g − 4 0,4,\ldots,4g-4 0 , 4 , … , 4 g − 4 (see [17 ] and [19 ] respectively).
The main result of this paper is a proof of the above purity conjecture.
Theorem 1.3. Conjecture 1.1 is true.
Main ideas of the proof
Let us say a few words about the proof, which combines several recent results about the so-called BPS Lie algebras associated to the stacks of representations of quivers or to the stack of Higgs bundles on smooth complex projective curves. First of all, we use the P=W theorem proved in [27 ] and [13 ] to translate the problem into the study of the perverse filtration, induced by the Hitchin morphism, of the cohomology of the moduli space M D ( r , d ) M^{D}(r,d) M D ( r , d ) of semistable Higgs bundles of rank r r r and degree d d d on a smooth, connected, complex projective curve C C C of genus g g g . More precisely, by Poincaré duality the purity conjecture is equivalent to the equality
∑ ℓ dim gr N − ℓ P H 2 N + 2 ℓ ( M D ( r , d ) , Q ) q ℓ = A S g , r ( q ) , (2) \sum_{\ell}\dim\operatorname{gr}^{P}_{N-\ell}H_{2N+2\ell}(M^{D}(r,d),\mathbb{Q})q^{\ell}=A_{S_{g},r}(q), \tag*{(2)} ℓ ∑ dim gr N − ℓ P H 2 N + 2 ℓ ( M D ( r , d ) , Q ) q ℓ = A S g , r ( q ) , (2)
where P k P_{k} P k is the shifted perverse filtration and H ∗ ( − , Q ) H_{*}(-,\mathbb{Q}) H ∗ ( − , Q ) stands for Borel-Moore homology. Next, we recast this in terms of the affinized BPS cohomology g C ( r , d ) \mathfrak{g}^{C}(r,d) g C ( r , d ) of M D ( r , d ) M^{D}(r,d) M D ( r , d ) . Note that, for a coprime pair ( r , d ) (r,d) ( r , d ) , the BPS cohomology coincides up to a shift with the (co)homology of the stack M s s ( r , d ) \mathcal{M}^{\mathrm{ss}}(r,d) M ss ( r , d ) of semistable Higgs bundles. Then we use the χ \chi χ -independence theorem of Kinjo-Koseki [23 ] to identify g C ( r , d ) \mathfrak{g}^{C}(r,d) g C ( r , d ) with g C ( r , 0 ) \mathfrak{g}^{C}(r,0) g C ( r , 0 ) , compatibly with the perverse filtration. This reduces the proof of the purity conjecture to the following equality, equivalent to (1.2):
∑ ℓ dim gr N − ℓ P g 2 ℓ C ( r , 0 ) q ℓ = A S g , r ( q ) . (3) \sum_{\ell}\dim\operatorname{gr}^{P}_{N-\ell}\mathfrak{g}^{C}_{2\ell}(r,0)q^{\ell}=A_{S_{g},r}(q). \tag*{(3)} ℓ ∑ dim gr N − ℓ P g 2 ℓ C ( r , 0 ) q ℓ = A S g , r ( q ) . (3)
The advantage of replacing M s s ( r , d ) \mathcal{M}^{\mathrm{ss}}(r,d) M ss ( r , d ) with M s s ( r , 0 ) \mathcal{M}^{\mathrm{ss}}(r,0) M ss ( r , 0 ) is that, by Davison’s neighborhood theorem, there is a local analytic isomorphism between a neighbourhood of the trivial Higgs bundle ( O C r , ϕ = 0 ) (\mathcal{O}_{C}^{r},\phi=0) ( O C r , ϕ = 0 ) in M s s ( r , 0 ) \mathcal{M}^{\mathrm{ss}}(r,0) M ss ( r , 0 ) and a neighborhood of 0 0 0 in the stack R e p r Π S g \mathfrak{Rep}_{r}\Pi_{S_{g}} Rep r Π S g of dimension r r r representations of the preprojective algebra of S g S_{g} S g . This is the bridge between the Higgs moduli space and the quiver variety. Note that the same bridge exists for the untwisted character variety, however the χ \chi χ -independence phenomenon has not been established on the Betti side. The remainder of the proof consists of the following three steps:
Step 1. A recent general stacky form of the P=C theorem due to R. Picovschi [35 ] , valid for arbitrary pairs ( r , d ) (r,d) ( r , d ) , implies that ⨁ ℓ g N − ℓ P g ^ 2 ℓ C ( r , 0 ) \bigoplus_{\ell} \mathfrak{g}^{\mathrm{P}}_{N-\ell}\widehat{\mathfrak{g}}^{\mathbb{C}}_{2\ell}(r,0) ⨁ ℓ g N − ℓ P g 2 ℓ C ( r , 0 ) coincides with the graded of the subspace of g ^ C ( r , 0 ) \widehat{\mathfrak{g}}^{\mathbb{C}}(r,0) g C ( r , 0 ) generated by the action of a family of tautological classes c ‾ i ( [ p t ] ) , i ≥ 2 \overline{c}_i([\mathrm{pt}]), i \ge2 c i ([ pt ]) , i ≥ 2 , on the reduced virtual fundamental class [ M s s , r d ( r , 0 ) ] [\mathcal{M}^{\mathrm{ss},\mathrm{rd}}(r,0)] [ M ss , rd ( r , 0 )] . This subspace restricts, under the local isomorphism
( M s s ( r , 0 ) , O C C ) ≅ ( R e p Π S g , 0 ) (\mathcal{M}^{\mathrm{ss}}(r,0),\mathcal{O}^{\mathbb{C}}_C) \cong(\mathop{\mathrm{Rep}}\nolimits_{\Pi_{S_g}},0) ( M ss ( r , 0 ) , O C C ) ≅ ( Rep Π S g , 0 )
to the subspace generated by the action of the tautological classes c i ( V ) , i ≥ 2 c_i(V), i \ge2 c i ( V ) , i ≥ 2 , on the reduced virtual fundamental class [ R e p r r d Π S g ] [\mathop{\mathrm{Rep}}\nolimits^{\mathrm{rd}}_r\Pi_{S_g}] [ Rep r rd Π S g ] . Note that, the stack R e p Π S g \mathop{\mathrm{Rep}}\nolimits_{\Pi_{S_g}} Rep Π S g being conical, its Borel-Moore homology is the same as that of any analytic neighborhood of its apex 0.
Step 2. We use the following quiver analog of the Markman theorem for the BPS Lie algebras of symplectic surfaces proven in [12 ] , which holds for any finite quiver. The same theorem is proved in [9 ] using an independent method.
Theorem 1.4. Let Q = ( Q 0 , Q 1 ) Q=(Q_0,Q_1) Q = ( Q 0 , Q 1 ) be any quiver and d d d a dimension vector for Q Q Q . The subspace of tautological classes
H ∗ t a u t ( R e p d Π Q , Q ) = Q [ c i ( V j ) ; i ≥ 1 , j ∈ Q 0 ] ∩ [ R e p d r d Π Q ] H_*^{\mathrm{taut}}(\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q,\mathbb{Q})=\mathbb{Q}[c_i(V_j);i\ge1,j\in Q_0]\cap[\mathop{\mathrm{Rep}}\nolimits_d^{\mathrm{rd}}\Pi_Q] H ∗ taut ( Rep d Π Q , Q ) = Q [ c i ( V j ) ; i ≥ 1 , j ∈ Q 0 ] ∩ [ Rep d rd Π Q ] in H ∗ ( R e p d Π Q , Q ) H_*(\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q,\mathbb{Q}) H ∗ ( Rep d Π Q , Q ) coincides with the BPS Lie algebra g ^ Π Q ( d ) \widehat{\mathfrak{g}}^{\Pi_Q}(d) g Π Q ( d ) .
By Theorem 3.1 , the restriction to the neighborhood of O C C \mathcal{O}^{\mathbb{C}}_C O C C establishes a graded, surjective morphism
⨁ ℓ g N − ℓ P g ^ 2 ℓ C ( r , 0 ) ↠ g ^ Π S g ( r ) / c 1 ( V ) ⋅ g ^ Π S g ( r ) . \bigoplus_{\ell}\mathfrak{g}^{\mathrm{P}}_{N-\ell}\widehat{\mathfrak{g}}^{\mathbb{C}}_{2\ell}(r,0)\twoheadrightarrow\widehat{\mathfrak{g}}^{\Pi_{S_g}}(r)/c_1(V)\cdot\widehat{\mathfrak{g}}^{\Pi_{S_g}}(r). ℓ ⨁ g N − ℓ P g 2 ℓ C ( r , 0 ) ↠ g Π S g ( r ) / c 1 ( V ) ⋅ g Π S g ( r ) .
The graded dimension of the right hand side is equal to the right hand side of (3) . This establishes the ≥ \ge ≥ form of (3) .
Step 3. We use the Hard Lefschetz theorem to identify, in the coprime case,
⨁ ℓ g N − ℓ P H 2 N + 2 ℓ ( M D ( r , d ) , Q ) ≅ ⨁ ℓ g N + ℓ P H 2 N ( M D ( r , d ) , Q ) . \bigoplus_{\ell}\mathfrak{g}^{\mathrm{P}}_{N-\ell}H_{2N+2\ell}(M^D(r,d),\mathbb{Q})\cong\bigoplus_{\ell}\mathfrak{g}^{\mathrm{P}}_{N+\ell}H_{2N}(M^D(r,d),\mathbb{Q}). ℓ ⨁ g N − ℓ P H 2 N + 2 ℓ ( M D ( r , d ) , Q ) ≅ ℓ ⨁ g N + ℓ P H 2 N ( M D ( r , d ) , Q ) .
Although we do not know how to compute the graded dimension of the latter, its ungraded dimension is equal to the constant term of the Kac polynomial A g , r ( σ ) A_{g,r}(\sigma) A g , r ( σ ) counting absolutely indecomposable vector bundles of rank r r r and degree d d d by [37 Thm. 1.1, Cor. 1.7] . Further, we have
A g , r ( 0 ) = A S g , r ( 1 ) . A_{g,r}(0)=A_{S_g,r}(1). A g , r ( 0 ) = A S g , r ( 1 ) .
This provides the desired ungraded ≤ \le ≤ form of (3) and allows us to conclude that (3) holds. The proof of Theorem 1.3 follows.
Some corollaries
Several consequences are worth pointing out:
Corollary 1.5. For any pair ( r , d ) (r,d) ( r , d ) there exists a canonical isomorphism
⨁ ℓ g N − ℓ P g ^ 2 ℓ C ( r , d ) ≅ g ^ Π S g ( r ) . \bigoplus_{\ell}\mathfrak{g}^{\mathrm{P}}_{N-\ell}\widehat{\mathfrak{g}}^{\mathbb{C}}_{2\ell}(r,d)\cong\widehat{\mathfrak{g}}^{\Pi_{S_g}}(r). ℓ ⨁ g N − ℓ P g 2 ℓ C ( r , d ) ≅ g Π S g ( r ) . Proof. This follows from the χ \chi χ -independence of ⨁ ℓ P N − ℓ g ^ 2 ℓ C ( r , d ) \bigoplus_{\ell}\mathsf{P}_{N-\ell}\widehat{\mathfrak{g}}^{\mathbb{C}}_{2\ell}(r,d) ⨁ ℓ P N − ℓ g 2 ℓ C ( r , d ) together with the relations
g N − ℓ P g ^ 2 ℓ C ( r , d ) = g N − ℓ P g ^ 2 ℓ C ( r , d ) , \mathfrak{g}^{\mathrm{P}}_{N-\ell}\widehat{\mathfrak{g}}^{\mathbb{C}}_{2\ell}(r,d)=\mathfrak{g}^{\mathrm{P}}_{N-\ell}\widehat{\mathfrak{g}}^{\mathbb{C}}_{2\ell}(r,d), g N − ℓ P g 2 ℓ C ( r , d ) = g N − ℓ P g 2 ℓ C ( r , d ) , see (46) , and
g ^ Π S g ( r ) / c 1 ( V ) ⋅ g ^ Π S g ( r ) ≅ g ^ Π S g ( r ) . \widehat{\mathfrak{g}}^{\Pi_{S_g}}(r)/c_1(V)\cdot\widehat{\mathfrak{g}}^{\Pi_{S_g}}(r)\cong\widehat{\mathfrak{g}}^{\Pi_{S_g}}(r). g Π S g ( r ) / c 1 ( V ) ⋅ g Π S g ( r ) ≅ g Π S g ( r ) . Note that, contrary to the case of a general symplectic surface, the χ \chi χ -independence isomorphism for T ∗ C T^*\mathbb{C} T ∗ C is canonical as H 2 ( T ∗ C , Q ) H^2(T^*\mathbb{C},\mathbb{Q}) H 2 ( T ∗ C , Q ) is one-dimensional. □
Corollary 1.6. There exists a surjective Lie algebra homomorphism
g C , ( 0 ) = ⨁ r g C ( r , 0 ) → ⨁ r g Π S g ( r ) = g Π S g . \mathfrak{g}^{C,(0)}=\bigoplus_{r}\mathfrak{g}^{C}(r,0)\to\bigoplus_{r}\mathfrak{g}^{\Pi S_g}(r)=\mathfrak{g}^{\Pi S_g}. g C , ( 0 ) = r ⨁ g C ( r , 0 ) → r ⨁ g Π S g ( r ) = g Π S g .
Corollary 1.7. For coprime ( r , d ) (r,d) ( r , d ) , the perverse filtration on H 2 N ( M D ( r , d ) , Q ) H^{2N}(M^D(r,d),\mathbb{Q}) H 2 N ( M D ( r , d ) , Q ) is given by
∑ ℓ dim gr N + ℓ P H 2 N ( M D ( r , d ) , Q ) q ℓ = A S g , r ( q ) . \sum_{\ell}\dim\operatorname{gr}^{P}_{N+\ell}H^{2N}(M^D(r,d),\mathbb{Q})q^\ell=A_{S_g,r}(q). ℓ ∑ dim gr N + ℓ P H 2 N ( M D ( r , d ) , Q ) q ℓ = A S g , r ( q ) .
Note that the χ \chi χ -independence allows us to extend this last corollary to arbitrary pair ( r , d ) (r,d) ( r , d ) , by replacing H 2 N ( M D ( r , d ) , Q ) H^{2N}(M^D(r,d),\mathbb{Q}) H 2 N ( M D ( r , d ) , Q ) by g 2 N C ( r , d ) \mathfrak{g}^{C}_{2N}(r,d) g 2 N C ( r , d ) , where g C ( r , d ) \mathfrak{g}^{C}(r,d) g C ( r , d ) stands for the (non-affinized) BPS Lie algebra.
Parabolic case and the Hausel-Letellier-Rodriguez-Villegas purity conjecture
The purity conjecture 1.1 has an extension to the case of a curve with punctures and prescribed generic monodromies around the punctures, see [16 ] . This parabolic purity conjecture involves a comet-shaped quiver. The statement is very similar to the unpunctured case. We expect that our proof may be adapted to this setting: both the P=W and curious Hard Lefschetz theorems are proved in the parabolic setting, see [13 ] . The Poincaré polynomial of the parabolic character variety is computed, see [30 ] . What remains to be extended to the parabolic case are the χ \chi χ -independence and P=C statements. We plan to return to this in the future.
Plan of the paper
Section 2 serves as a reminder about COHAs and BPS Lie algebras associated to quivers. The quiver analog of Markman’s theorem is proved in §3. In §§4,5 we introduce the necessary notions regarding the COHA and BPS Lie algebra associated to Higgs moduli, and recall the main results of [40 ] , [23 ] , [12 ] and [35 ] . After recalling Davison’s neighborhood theorem, the associated restriction map is defined and studied in §6. Finally, the proof of Theorem 1.3 is given in §7. The appendix provides some detail about the COHA coproduct and the proofs of various results around the primitivity of tautological classes, adapted from [12 ] .
Quivers, COHAs and the BPS Lie algebra of a quiver.
In this section, we recall the main properties of cohomological Hall algebras associated to quivers and their BPS Lie algebras, postponing some proofs and details to Appendix A. In addition to the usual COHA of the preprojective algebra Π Q \Pi_Q Π Q of a quiver Q Q Q , we also consider the COHA of the endomorphism algebra C Q [ θ ] \mathbb{C}_Q[\theta] C Q [ θ ] , see below. The two COHAs are related by dimensional reduction. The latter is better suited to our proof of Markman’s tautological generation theorem for the BPS Lie algebra, see Theorem 3.1. Throughout the entire paper, we work over C \mathbb{C} C .
Stacks of quiver representations
Let Q = ( Q 0 , Q 1 ) Q=(Q_0,Q_1) Q = ( Q 0 , Q 1 ) be an arbitrary finite quiver, with possible multiple arrows and edge loops. The Euler form ⟨ − , − ⟩ \langle-,-\rangle ⟨ − , − ⟩ and the symmetrized Euler form ( − , − ) (-,-) ( − , − ) are given by
⟨ d , e ⟩ = ∑ i d i e i − ∑ a : i → j d i e j , ( d , e ) = 2 ∑ i d i e i − ∑ a : i → j ( d i e j + d j e i ) . (4) \langle d,e\rangle=\sum_i d_i e_i-\sum_{a:i\to j}d_i e_j,\qquad(d,e)=2\sum_i d_i e_i-\sum_{a:i\to j}(d_i e_j+d_j e_i). \tag*{(4)} ⟨ d , e ⟩ = i ∑ d i e i − a : i → j ∑ d i e j , ( d , e ) = 2 i ∑ d i e i − a : i → j ∑ ( d i e j + d j e i ) . (4)
Let Q ‾ = ( Q 0 , Q 1 ⊔ Q 1 ∗ ) \overline{Q}=(Q_0,Q_1\sqcup Q_1^*) Q = ( Q 0 , Q 1 ⊔ Q 1 ∗ ) be the double quiver. For a dimension vector d ∈ N Q 0 d\in\mathbb{N}^{Q_0} d ∈ N Q 0 we put
E d = ⨁ a ∈ Q 1 Hom ( C d s ( a ) , C d t ( a ) ) , E ‾ d = T ∗ E d , G L d = ∏ i ∈ Q 0 G L d i , g l d = Lie ( G L d ) . E_d=\bigoplus_{a\in Q_1}\operatorname{Hom}(\mathbb{C}^{d_{s(a)}},\mathbb{C}^{d_{t(a)}}),\qquad\overline{E}_d=T^*E_d,\qquad GL_d=\prod_{i\in Q_0}GL_{d_i},\qquad\mathfrak{gl}_d=\operatorname{Lie}(GL_d). E d = a ∈ Q 1 ⨁ Hom ( C d s ( a ) , C d t ( a ) ) , E d = T ∗ E d , G L d = i ∈ Q 0 ∏ G L d i , gl d = Lie ( G L d ) .
For d = ( d i ) i ∈ Q 0 d=(d_i)_{i\in Q_0} d = ( d i ) i ∈ Q 0 we set
∣ d ∣ = ∑ i d i , tr : g l d → C , tr ( θ i ) = ∑ i tr ( θ i ) . |d|=\sum_i d_i,\qquad\operatorname{tr}:\mathfrak{gl}_d\to\mathbb{C},\qquad\operatorname{tr}(\theta_i)=\sum_i\operatorname{tr}(\theta_i). ∣ d ∣ = i ∑ d i , tr : gl d → C , tr ( θ i ) = i ∑ tr ( θ i ) .
The stacks of d d d -dimensional representations of Q Q Q and Q ‾ \overline{Q} Q are
R e p d Q = E d / G L d , R e p d Q ‾ = E ‾ d / G L d . \mathop{\mathrm{Rep}}\nolimits_d Q = E_d/GL_d,\qquad\mathop{\mathrm{Rep}}\nolimits_d\overline{Q} = \overline{E}_d/GL_d. Rep d Q = E d / G L d , Rep d Q = E d / G L d .
The stack R e p d Q \mathop{\mathrm{Rep}}\nolimits_d Q Rep d Q is smooth of dimension − ⟨ d , d ⟩ -\langle d,d\rangle − ⟨ d , d ⟩ . The preprojective algebra Π Q \Pi_Q Π Q of Q Q Q is the quotient of C Q ‾ \mathbb{C}\overline{Q} C Q by the two-sided ideal generated by ∑ a ∈ Q 1 [ x a , x a ∗ ] \sum_{a\in Q_1}[x_a,x_a^*] ∑ a ∈ Q 1 [ x a , x a ∗ ] . The moment map for the Hamiltonian action of G L d GL_d G L d on T ∗ E d T^*E_d T ∗ E d is
μ d : T ∗ E d ⟶ g l d , ( x , x ∗ ) ⟼ ∑ a ∈ Q 1 [ x a , x a ∗ ] . (5) \mu_d:T^*E_d\longrightarrow\mathfrak{gl}_d,\qquad(x,x^*)\longmapsto\sum_{a\in Q_1}[x_a,x_a^*]. \tag*{(5)} μ d : T ∗ E d ⟶ gl d , ( x , x ∗ ) ⟼ a ∈ Q 1 ∑ [ x a , x a ∗ ] . (5)
The derived stack of d d d -dimensional Π Q \Pi_Q Π Q -modules is thus
R e p d Π Q = T ∗ ( R e p d Q ) = μ d − 1 ( 0 ) / G L d , \mathop{\mathrm{Rep}}\nolimits_d\Pi_Q=T^*(\mathop{\mathrm{Rep}}\nolimits_dQ)=\mu_d^{-1}(0)/GL_d, Rep d Π Q = T ∗ ( Rep d Q ) = μ d − 1 ( 0 ) / G L d ,
where μ d − 1 ( 0 ) \mu_d^{-1}(0) μ d − 1 ( 0 ) is the derived zero locus. It is 0 0 0 -shifted symplectic, quasi-smooth, of virtual dimension
vdim R e p d Π Q = 2 dim R e p d Q = − ( d , d ) . \operatorname{vdim}\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q=2\dim\mathop{\mathrm{Rep}}\nolimits_dQ=-(d,d). vdim Rep d Π Q = 2 dim Rep d Q = − ( d , d ) .
We also consider the additive version of the inertia stack of R e p Q \mathop{\mathrm{Rep}}\nolimits_Q Rep Q , i.e., the stack of C Q [ θ ] \mathbb{C}Q[\theta] C Q [ θ ] -representations. It is the quotient by G L d GL_d G L d of a derived zero locus in an affine space
R e p d I Q = T [ − 1 ] R e p d Q = { ( x , θ ) ; x ∈ R e p d Q , θ ∈ g l d , [ θ , x ] = 0 } / G L d . \mathop{\mathrm{Rep}}\nolimits_dIQ=T[-1]\mathop{\mathrm{Rep}}\nolimits_dQ=\{(x,\theta);\,x\in\mathop{\mathrm{Rep}}\nolimits_dQ,\ \theta\in\mathfrak{gl}_d,\ [\theta,x]=0\}/GL_d. Rep d I Q = T [ − 1 ] Rep d Q = {( x , θ ) ; x ∈ Rep d Q , θ ∈ gl d , [ θ , x ] = 0 } / G L d .
It is quasi-smooth of virtual dimension 0 0 0 . Finally, the critical locus of the tripled quiver Q ~ \widetilde{Q} Q with potential
W = ∑ a ∈ Q 1 x a ∗ [ θ , x a ] (6) W=\sum_{a\in Q_1}x_a^*[\theta,x_a] \tag*{(6)} W = a ∈ Q 1 ∑ x a ∗ [ θ , x a ] (6)
is the stack of modules over the algebra Π Q [ θ ] \Pi_Q[\theta] Π Q [ θ ]
R e p d Π Q = T ∗ [ − 1 ] R e p d Π Q = { ( x , x ∗ , θ ) ; μ d ( x , x ∗ ) = 0 , [ θ , x ] = [ θ , x ∗ ] = 0 } / G L d . \mathop{\mathrm{Rep}}\nolimits_d\Pi_Q=T^*[-1]\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q=\{(x,x^*,\theta);\,\mu_d(x,x^*)=0,\ [\theta,x]=[\theta,x^*]=0\}/GL_d. Rep d Π Q = T ∗ [ − 1 ] Rep d Π Q = {( x , x ∗ , θ ) ; μ d ( x , x ∗ ) = 0 , [ θ , x ] = [ θ , x ∗ ] = 0 } / G L d .
It is a ( − 1 ) (-1) ( − 1 ) -shifted symplectic stack of virtual dimension 0 0 0 , carrying a canonical orientation, see [24 ex. 3.20] . Forgetting θ \theta θ or x ∗ x^* x ∗ gives the two dimensional reductions morphisms
R e p d Π Q ~ → κ ′ R e p d I Q κ ↓ R e p d Π Q \begin{CD}
\widetilde{\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q} @>{\kappa'}>> \mathop{\mathrm{Rep}}\nolimits_dIQ \\
@V{\kappa}VV @. \\
\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q @.
\end{CD} Rep d Π Q κ ↓ ⏐ Rep d Π Q κ ′ Rep d I Q
All three stacks map to B G L d BGL_d B G L d compatibly.
Conical substacks
Inside R e p d Π Q ~ \widetilde{\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q} Rep d Π Q and R e p d I Q \mathop{\mathrm{Rep}}\nolimits_dIQ Rep d I Q we consider the closed substacks defined by the following conditions on θ \theta θ
Λ d 0 , R e p d 0 I Q : θ = 0 , \Lambda_d^0,\quad\mathop{\mathrm{Rep}}\nolimits_d^0IQ:\ \theta=0, Λ d 0 , Rep d 0 I Q : θ = 0 ,
Λ d n i l , R e p d n i l I Q : θ nilpotent , \Lambda_d^{\mathrm{nil}},\quad\mathop{\mathrm{Rep}}\nolimits_d^{\mathrm{nil}}IQ:\ \theta\ \text{nilpotent}, Λ d nil , Rep d nil I Q : θ nilpotent ,
Λ d s c a l , R e p d s c a l I Q : ∃ c ∈ C , θ = c ⋅ id , \Lambda_d^{\mathrm{scal}},\quad\mathop{\mathrm{Rep}}\nolimits_d^{\mathrm{scal}}IQ:\ \exists c\in\mathbb{C},\ \theta=c\cdot\operatorname{id}, Λ d scal , Rep d scal I Q : ∃ c ∈ C , θ = c ⋅ id ,
Λ d s n i l , R e p d s n i l I Q : ∃ c ∈ C , θ − c nilpotent . \Lambda_d^{\mathrm{snil}},\quad\mathop{\mathrm{Rep}}\nolimits_d^{\mathrm{snil}}IQ:\ \exists c\in\mathbb{C},\ \theta-c\ \text{nilpotent}. Λ d snil , Rep d snil I Q : ∃ c ∈ C , θ − c nilpotent .
For each ⋆ ∈ { 0 , n i l , s c a l , s n i l } \star\in\{0,\mathrm{nil},\mathrm{scal},\mathrm{snil}\} ⋆ ∈ { 0 , nil , scal , snil } we have
Λ d ⋆ = ( κ ′ ) − 1 ( R e p d ⋆ I Q ) . \Lambda_d^\star=(\kappa')^{-1}(\mathop{\mathrm{Rep}}\nolimits_d^\star IQ). Λ d ⋆ = ( κ ′ ) − 1 ( Rep d ⋆ I Q ) .
There are closed immersions
Λ d 0 → Λ d n i l ↓ ↓ Λ d s c a l → Λ d s n i l \begin{CD}
\Lambda_d^0 @>>> \Lambda_d^{\mathrm{nil}} \\
@VVV @VVV \\
\Lambda_d^{\mathrm{scal}} @>>> \Lambda_d^{\mathrm{snil}}
\end{CD} Λ d 0 ↓ ⏐ Λ d scal Λ d nil ↓ ⏐ Λ d snil
and similarly for R e p d ⋆ I Q \mathop{\mathrm{Rep}}\nolimits^{\star}_{d}IQ Rep d ⋆ I Q . The stacks Λ d ⋆ \Lambda^{\star}_{d} Λ d ⋆ and R e p d ⋆ I Q \mathop{\mathrm{Rep}}\nolimits^{\star}_{d}IQ Rep d ⋆ I Q are taken with their classical, reduced structure. Observe that R e p d 0 I Q = R e p d Q \mathop{\mathrm{Rep}}\nolimits^{0}_{d}IQ=\mathop{\mathrm{Rep}}\nolimits_{d}Q Rep d 0 I Q = Rep d Q . The following lemma is obvious.
Lemma 2.1. For d ≠ 0 d\ne0 d = 0 , the map
( x , θ ) ↦ ( ( x , θ − c id ) , c ) , c = tr ( θ ) ∣ d ∣ (x,\theta)\mapsto((x,\theta-c\operatorname{id}),c),\qquad c=\frac{\operatorname{tr}(\theta)}{|d|} ( x , θ ) ↦ (( x , θ − c id ) , c ) , c = ∣ d ∣ tr ( θ ) gives G L d GL_d G L d -equivariant isomorphisms
R e p d s n i l I Q ≅ R e p d n i l I Q × A 1 , R e p d s c a l I Q ≅ R e p d 0 I Q × A 1 . \mathop{\mathrm{Rep}}\nolimits^{\mathrm{snil}}_{d}IQ\cong\mathop{\mathrm{Rep}}\nolimits^{\mathrm{nil}}_{d}IQ\times\mathbb{A}^{1},\qquad\mathop{\mathrm{Rep}}\nolimits^{\mathrm{scal}}_{d}IQ\cong\mathop{\mathrm{Rep}}\nolimits^{0}_{d}IQ\times\mathbb{A}^{1}. Rep d snil I Q ≅ Rep d nil I Q × A 1 , Rep d scal I Q ≅ Rep d 0 I Q × A 1 . They are compatible with the inclusions and with the maps to B G L d BGL_d B G L d .
COHAs of quivers
For l ∈ Z l\in\mathbb{Z} l ∈ Z and d ∈ N Q 0 d\in\mathbb{N}^{Q_0} d ∈ N Q 0 we put
A Π Q , l ( d ) = H l − ( d , d ) ( R e p d Π Q , Q ) , A Π Q ( d ) = ⨁ l A Π Q , l ( d ) , A Π Q = ⨁ d A Π Q ( d ) , \mathcal{A}_{\Pi_Q,l}(d)=H_{l-(d,d)}(\mathop{\mathrm{Rep}}\nolimits_{d}\Pi_Q,\mathbb{Q}),\qquad\mathcal{A}_{\Pi_Q}(d)=\bigoplus_l\mathcal{A}_{\Pi_Q,l}(d),\qquad\mathcal{A}_{\Pi_Q}=\bigoplus_d\mathcal{A}_{\Pi_Q}(d), A Π Q , l ( d ) = H l − ( d , d ) ( Rep d Π Q , Q ) , A Π Q ( d ) = l ⨁ A Π Q , l ( d ) , A Π Q = d ⨁ A Π Q ( d ) ,
A I Q , l ( d ) = H l ( R e p d I Q , Q ) , A I Q ( d ) = ⨁ l A I Q , l ( d ) , A I Q = ⨁ d A I Q ( d ) \mathcal{A}_{IQ,l}(d)=H_l(\mathop{\mathrm{Rep}}\nolimits_{d}IQ,\mathbb{Q}),\qquad\mathcal{A}_{IQ}(d)=\bigoplus_l\mathcal{A}_{IQ,l}(d),\qquad\mathcal{A}_{IQ}=\bigoplus_d\mathcal{A}_{IQ}(d) A I Q , l ( d ) = H l ( Rep d I Q , Q ) , A I Q ( d ) = l ⨁ A I Q , l ( d ) , A I Q = d ⨁ A I Q ( d )
and likewise for A I Q n i l \mathcal{A}^{\mathrm{nil}}_{IQ} A I Q nil . All three of A Π Q \mathcal{A}_{\Pi_Q} A Π Q , A I Q \mathcal{A}_{IQ} A I Q and A I Q n i l \mathcal{A}^{\mathrm{nil}}_{IQ} A I Q nil are equipped with N Q 0 × Z \mathbb{N}^{Q_0}\times\mathbb{Z} N Q 0 × Z -graded cohomological Hall algebra (COHA) structures. The multiplication in A Π Q \mathcal{A}_{\Pi_Q} A Π Q comes from the correspondence
R e p d Π Q × R e p d ′ Π Q ← g r d , d ′ F i l t d , d ′ Π Q → e v d , d ′ R e p d + d ′ Π Q (7) \mathop{\mathrm{Rep}}\nolimits_d\Pi_Q\times\mathop{\mathrm{Rep}}\nolimits_{d'}\Pi_Q\ \xleftarrow{\ \mathrm{gr}_{d,d'}\ }\ \mathrm{Filt}_{d,d'}\Pi_Q\ \xrightarrow{\ \mathrm{ev}_{d,d'}\ }\ \mathop{\mathrm{Rep}}\nolimits_{d+d'}\Pi_Q \tag*{(7)} Rep d Π Q × Rep d ′ Π Q gr d , d ′ Filt d , d ′ Π Q ev d , d ′ Rep d + d ′ Π Q (7)
where the stack F i l t d , d ′ Π Q \mathrm{Filt}_{d,d'}\Pi_Q Filt d , d ′ Π Q classifies short exact sequences
0 ⟶ M d ′ ⟶ M ⟶ M d ⟶ 0 0\longrightarrow M_{d'}\longrightarrow M\longrightarrow M_d\longrightarrow0 0 ⟶ M d ′ ⟶ M ⟶ M d ⟶ 0
of Π Q \Pi_Q Π Q -modules of respective dimensions d d d , d ′ d' d ′ and d + d ′ d+d' d + d ′ , and where the maps g r d , d ′ \mathrm{gr}_{d,d'} gr d , d ′ and e v d , d ′ \mathrm{ev}_{d,d'} ev d , d ′ associate to such a sequence its extreme, resp. middle terms. The map g r d , d ′ \mathrm{gr}_{d,d'} gr d , d ′ is quasi-smooth while e v d , d ′ \mathrm{ev}_{d,d'} ev d , d ′ is proper. This allows one to define an associative product
A Π Q ( d ) ⊗ A Π Q ( d ′ ) ⟶ A Π Q ( d + d ′ ) , c ⊗ c ′ ↦ ( e v d , d ′ ) ∗ ( g r d , d ′ ) ! ( c ⊠ c ′ ) \mathcal{A}_{\Pi_Q}(d)\otimes\mathcal{A}_{\Pi_Q}(d')\longrightarrow\mathcal{A}_{\Pi_Q}(d+d'),\qquad c\otimes c'\mapsto(\mathrm{ev}_{d,d'})_*(\mathrm{gr}_{d,d'})!(c\boxtimes c') A Π Q ( d ) ⊗ A Π Q ( d ′ ) ⟶ A Π Q ( d + d ′ ) , c ⊗ c ′ ↦ ( ev d , d ′ ) ∗ ( gr d , d ′ )! ( c ⊠ c ′ )
where g r d , d ′ ! \mathrm{gr}_{d,d'}! gr d , d ′ ! is the virtual pullback. With the shift by − ( d , d ) -(d,d) − ( d , d ) in the definition of A Π Q , l ( d ) \mathcal{A}_{\Pi_Q,l}(d) A Π Q , l ( d ) , this multiplication is of cohomological degree zero. The construction of the algebra structure on A I Q \mathcal{A}_{IQ} A I Q and A I Q n i l \mathcal{A}^{\mathrm{nil}}_{IQ} A I Q nil is entirely similar.
Moreover, denoting by
ι n i l : R e p n i l I Q ⟶ R e p I Q , R e p n i l I Q = ⋃ d R e p d n i l I Q , R e p I Q = ⋃ d R e p d I Q (8) \iota^{\mathrm{nil}}:\mathop{\mathrm{Rep}}\nolimits^{\mathrm{nil}}IQ\longrightarrow\mathop{\mathrm{Rep}}\nolimits IQ,\qquad\mathop{\mathrm{Rep}}\nolimits^{\mathrm{nil}}IQ=\bigcup_d\mathop{\mathrm{Rep}}\nolimits^{\mathrm{nil}}_{d}IQ,\qquad\mathop{\mathrm{Rep}}\nolimits IQ=\bigcup_d\mathop{\mathrm{Rep}}\nolimits_{d}IQ \tag*{(8)} ι nil : Rep nil I Q ⟶ Rep I Q , Rep nil I Q = d ⋃ Rep d nil I Q , Rep I Q = d ⋃ Rep d I Q (8)
the closed embedding, the pushforward
ι ∗ n i l : A I Q n i l ⟶ A I Q \iota^{\mathrm{nil}}_*:\mathcal{A}^{\mathrm{nil}}_{IQ}\longrightarrow\mathcal{A}_{IQ} ι ∗ nil : A I Q nil ⟶ A I Q
is an algebra homomorphism. We refer to [38 ] , [20 ] and [36 ] for details. We also define
A I Q , l s n i l ( d ) = H l ( R e p d s n i l I Q , Q ) , A I Q , l s c a l ( d ) = H l ( R e p d s c a l I Q , Q ) , A I Q , l 0 ( d ) = H l ( R e p d 0 I Q , Q ) = H l ( R e p d Q , Q ) \mathcal{A}^{\mathrm{snil}}_{IQ,l}(d)=H_l(\mathop{\mathrm{Rep}}\nolimits^{\mathrm{snil}}_{d}IQ,\mathbb{Q}),\qquad\mathcal{A}^{\mathrm{scal}}_{IQ,l}(d)=H_l(\mathop{\mathrm{Rep}}\nolimits^{\mathrm{scal}}_{d}IQ,\mathbb{Q}),\qquad\mathcal{A}^{0}_{IQ,l}(d)=H_l(\mathop{\mathrm{Rep}}\nolimits^{0}_{d}IQ,\mathbb{Q})=H_l(\mathop{\mathrm{Rep}}\nolimits_dQ,\mathbb{Q}) A I Q , l snil ( d ) = H l ( Rep d snil I Q , Q ) , A I Q , l scal ( d ) = H l ( Rep d scal I Q , Q ) , A I Q , l 0 ( d ) = H l ( Rep d 0 I Q , Q ) = H l ( Rep d Q , Q )
Note that
A I Q s n i l = ⨁ l , d A I Q , l s n i l ( d ) , A I Q s c a l = ⨁ l , d A I Q , l s c a l ( d ) , A I Q 0 = ⨁ l , d A I Q , l 0 ( d ) \mathcal{A}^{\mathrm{snil}}_{IQ}=\bigoplus_{l,d}\mathcal{A}^{\mathrm{snil}}_{IQ,l}(d),\qquad\mathcal{A}^{\mathrm{scal}}_{IQ}=\bigoplus_{l,d}\mathcal{A}^{\mathrm{scal}}_{IQ,l}(d),\qquad\mathcal{A}^{0}_{IQ}=\bigoplus_{l,d}\mathcal{A}^{0}_{IQ,l}(d) A I Q snil = l , d ⨁ A I Q , l snil ( d ) , A I Q scal = l , d ⨁ A I Q , l scal ( d ) , A I Q 0 = l , d ⨁ A I Q , l 0 ( d )
do not carry natural algebra structures as they correspond to substacks that are not stable under extensions. There is a 1 1 1 d-COHA structure on A I Q 0 \mathcal{A}^{0}_{IQ} A I Q 0 . We do not consider it here.
Both A Π Q \mathcal{A}_{\Pi_Q} A Π Q and A I Q \mathcal{A}_{IQ} A I Q come from the same critical COHA associated to R e p Π Q \mathop{\mathrm{Rep}}\nolimits\Pi_Q Rep Π Q . By [2 ] , the stack R e p d Π Q \mathop{\mathrm{Rep}}\nolimits_d\Pi_Q Rep d Π Q carries a Joyce perverse sheaf φ R e p d Π Q \varphi_{\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q} φ Rep d Π Q and
A ~ = ⨁ l , d H − l ( R e p d Π Q , φ R e p d Π Q ) \widetilde{\mathcal{A}}=\bigoplus_{l,d}H^{-l}\left(\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q,\varphi_{\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q}\right) A = l , d ⨁ H − l ( Rep d Π Q , φ Rep d Π Q )
carries an associative algebra structure defined via an induction diagram very similar to (7) . Moreover, there are dimensional reduction isomorphisms
κ ∗ φ R e p d Π Q ≅ D Q R e p d Π Q ⊗ L − 1 2 ( d , d ) and κ ′ ∗ φ R e p d Π Q ≅ D Q R e p d I Q \kappa^*\varphi_{\mathfrak{Rep}_{d}\Pi_Q}\cong\mathbb{D}\mathbb{Q}_{\mathfrak{Rep}_{d}\Pi_Q}\otimes\mathbb{L}^{-\frac{1}{2}(d,d)}
\quad\text{and}\quad
\kappa'^*\varphi_{\mathfrak{Rep}_{d}\Pi_Q}\cong\mathbb{D}\mathbb{Q}_{\mathfrak{Rep}_{d}I_Q} κ ∗ φ Rep d Π Q ≅ D Q Rep d Π Q ⊗ L − 2 1 ( d , d ) and κ ′ ∗ φ Rep d Π Q ≅ D Q Rep d I Q
see [4 ] or [12 §2] . Taking global sections gives the isomorphisms
A Π Q , l ( d ) ≅ A ~ l ( d ) ≅ A I Q , l ( d ) . (9) \mathcal{A}_{\Pi_Q,l}(d)\cong\widetilde{\mathcal{A}}_l(d)\cong\mathcal{A}_{I_Q,l}(d). \tag*{(9)} A Π Q , l ( d ) ≅ A l ( d ) ≅ A I Q , l ( d ) . (9)
These isomorphisms are H G L d ∗ H^*_{G_{L_d}} H G L d ∗ -linear and intertwine the three COHA products. For the closed conical substacks R e p d ⋆ I Q \mathfrak{Rep}_{d}^{\star}I_Q Rep d ⋆ I Q with ⋆ ∈ { n i l , s n i l , s c a l , 0 } \star\in\{\mathrm{nil},\mathrm{snil},\mathrm{scal},0\} ⋆ ∈ { nil , snil , scal , 0 } , a base change with respect to the map κ ′ \kappa' κ ′ yields in the same fashion isomorphisms
H ∗ Λ ⋆ ( R e p d Π Q , Q ) ≅ H ∗ ( R e p d ⋆ I Q , Q ) , (10) H^{\Lambda^\star}_{*}(\mathfrak{Rep}_{d}\Pi_Q,\mathbb{Q})\cong H_{*}(\mathfrak{Rep}_{d}^{\star}I_Q,\mathbb{Q}), \tag*{(10)} H ∗ Λ ⋆ ( Rep d Π Q , Q ) ≅ H ∗ ( Rep d ⋆ I Q , Q ) , (10)
which are compatible with (9) via the pushforward along the closed immersions
R e p d ⋆ I Q ↪ R e p d I Q . \mathfrak{Rep}_{d}^{\star}I_Q\hookrightarrow\mathfrak{Rep}_{d}I_Q. Rep d ⋆ I Q ↪ Rep d I Q .
The left hand side of (10) is the microlocal homology with support. It is defined by
H l Λ ⋆ ( R e p d Π Q , Q ) = H − l ( R e p d Π Q , ι ~ ! ( ι ⋆ ) ∗ φ R e p d Π Q ) , H^{\Lambda^\star}_{l}(\mathfrak{Rep}_{d}\Pi_Q,\mathbb{Q})=H^{-l}(\mathfrak{Rep}_{d}\Pi_Q,\widetilde{\iota}^{!}( \iota^{\star})^*\varphi_{\mathfrak{Rep}_{d}\Pi_Q}), H l Λ ⋆ ( Rep d Π Q , Q ) = H − l ( Rep d Π Q , ι ! ( ι ⋆ ) ∗ φ Rep d Π Q ) ,
where ι ~ ⋆ : Λ d ⋆ → R e p d Π Q \widetilde{\iota}^{\star}:\Lambda^\star_d\to\mathfrak{Rep}_{d}\Pi_Q ι ⋆ : Λ d ⋆ → Rep d Π Q is the closed immersion, see [12 §2.3] . In particular, there is an algebra isomorphism
⨁ d H ∗ Λ n i l ( R e p d Π Q , Q ) ≅ A I Q n i l . \bigoplus_d H^{\Lambda^{\mathrm{nil}}}_{*}(\mathfrak{Rep}_{d}\Pi_Q,\mathbb{Q})\cong\mathcal{A}^{\mathrm{nil}}_{I_Q}. d ⨁ H ∗ Λ nil ( Rep d Π Q , Q ) ≅ A I Q nil .
Purity
For future use, we mention the following result.
Proposition 2.2. The mixed Hodge structures on the spaces
A Π Q ( d ) , A I Q ( d ) , A I Q ⋆ ( d ) , ⋆ ∈ { 0 , s c a l , n i l , s n i l } \mathcal{A}_{\Pi_Q}(d),\ \mathcal{A}_{I_Q}(d),\ \mathcal{A}^{\star}_{I_Q}(d),\quad\star\in\{0,\mathrm{scal},\mathrm{nil},\mathrm{snil}\} A Π Q ( d ) , A I Q ( d ) , A I Q ⋆ ( d ) , ⋆ ∈ { 0 , scal , nil , snil } are all pure, for all d ∈ N Q 0 d\in\mathbb{N}^{Q_0} d ∈ N Q 0 .
Proof. Purity of A Π Q ( d ) \mathcal{A}_{\Pi_Q}(d) A Π Q ( d ) is shown in [5 ] , [6 ] . The same thus holds for A I Q \mathcal{A}_{I_Q} A I Q . Purity of the conical substacks R e p d ⋆ I Q \mathfrak{Rep}^{\star}_{d}I_Q Rep d ⋆ I Q is much easier: it is obvious for ⋆ ∈ { 0 , s c a l } \star\in\{0,\mathrm{scal}\} ⋆ ∈ { 0 , scal } and the case of ⋆ = s n i l \star=\mathrm{snil} ⋆ = snil follows from the case ⋆ = n i l \star=\mathrm{nil} ⋆ = nil . In the latter case, it is a consequence of the Jordan stratification, see, e.g., [33 ] or Corollary 3.7 below. □
The BPS sheaves and the BPS Lie algebras g Π Q \mathfrak{g}^{\Pi_Q} g Π Q
We next briefly recall the main features of the theory of BPS sheaves and Lie algebras, as developed in the work of Davison and his collaborators in [8 ] , [11 ] , [12 ] . We refer to Appendix A for some details. Denote by
p : ( R e p d Π Q ) r e d ⟶ Rep d Π Q = μ d − 1 ( 0 ) / / G L d p:(\mathfrak{Rep}_{d}\Pi_Q)^{\mathrm{red}}\longrightarrow\operatorname{Rep}_{d}\Pi_Q=\mu_d^{-1}(0)//GL_d p : ( Rep d Π Q ) red ⟶ Rep d Π Q = μ d − 1 ( 0 ) // G L d
the good moduli spaces and by
⊕ d , d ′ : Rep d Π Q × Rep d ′ Π Q ⟶ Rep d + d ′ Π Q \oplus_{d,d'}:\operatorname{Rep}_{d}\Pi_Q\times\operatorname{Rep}_{d'}\Pi_Q\longrightarrow\operatorname{Rep}_{d+d'}\Pi_Q ⊕ d , d ′ : Rep d Π Q × Rep d ′ Π Q ⟶ Rep d + d ′ Π Q
the addition maps. There are pure mixed Hodge modules B P S d Π Q \mathcal{BPS}^{\Pi_Q}_{d} B P S d Π Q over Rep d Π Q \operatorname{Rep}_{d}\Pi_Q Rep d Π Q such that
Sym ⊕ ( ⨁ d B P S Π Q ( d ) ⊗ H G m ∗ ) ≅ ⨁ d p ∗ ( D Q R e p d Π Q ⊗ L − 1 2 ( d , d ) ) . (11) \operatorname{Sym}_{\oplus}\left(\bigoplus_d\mathcal{BPS}^{\Pi_Q}(d)\otimes H^*_{G_m}\right)\cong\bigoplus_d p_*\left(\mathbb{D}\mathbb{Q}_{\mathfrak{Rep}_{d}\Pi_Q}\otimes\mathbb{L}^{-\frac{1}{2}(d,d)}\right). \tag*{(11)} Sym ⊕ ( d ⨁ B P S Π Q ( d ) ⊗ H G m ∗ ) ≅ d ⨁ p ∗ ( D Q Rep d Π Q ⊗ L − 2 1 ( d , d ) ) . (11)
The BPS Lie algebras and affinized BPS Lie algebras of the quiver Q Q Q are defined as
g Π Q = ⨁ d g Π Q ( d ) = ⨁ d H − ( Rep d Π Q , B P S Π Q ( d ) ) , \mathfrak{g}^{\Pi_Q}=\bigoplus_d\mathfrak{g}^{\Pi_Q}(d)=\bigoplus_d H^-\left(\operatorname{Rep}_{d}\Pi_Q,\mathcal{BPS}^{\Pi_Q}(d)\right), g Π Q = d ⨁ g Π Q ( d ) = d ⨁ H − ( Rep d Π Q , B P S Π Q ( d ) ) ,
g ^ Π Q = ⨁ d g ^ Π Q ( d ) = ⨁ d H − ( Rep d Π Q , B P S Π Q ( d ) ⊗ H G m ∗ ) . \widehat{\mathfrak{g}}^{\Pi_Q}=\bigoplus_d\widehat{\mathfrak{g}}^{\Pi_Q}(d)=\bigoplus_d H^-\left(\operatorname{Rep}_{d}\Pi_Q,\mathcal{BPS}^{\Pi_Q}(d)\otimes H^*_{G_m}\right). g Π Q = d ⨁ g Π Q ( d ) = d ⨁ H − ( Rep d Π Q , B P S Π Q ( d ) ⊗ H G m ∗ ) .
They are N Q 0 × Z \mathbb{N}^{Q_0} \times\mathbb{Z} N Q 0 × Z -graded subspaces in A Π Q A_{\Pi_Q} A Π Q . The generator u ∈ H G m 2 u \in H^2_{\mathbb{G}_m} u ∈ H G m 2 is in degree − 2 -2 − 2 . Both are Lie subalgebras of the COHA. Moreover, we have
U ( g ^ Π Q ) = A Π Q (12) U(\widehat{\mathfrak{g}}^{\Pi_Q}) = A_{\Pi_Q} \tag*{(12)} U ( g Π Q ) = A Π Q (12)
Passing to global sections in (11) recovers the PBW isomorphism
Sym ( ⨁ d g ^ Π Q ( d ) ) → ∼ A Π Q (13) \operatorname{Sym}\left(\bigoplus_d \widehat{\mathfrak{g}}^{\Pi_Q}(d)\right) \xrightarrow{\sim} A_{\Pi_Q} \tag*{(13)} Sym ( d ⨁ g Π Q ( d ) ) ∼ A Π Q (13)
where the map is given by the COHA multiplication. Finally, there is a comultiplication [12 ]
Δ : A Π Q → A Π Q ⊗ A Π Q (14) \Delta: A_{\Pi_Q} \to A_{\Pi_Q} \otimes A_{\Pi_Q} \tag*{(14)} Δ : A Π Q → A Π Q ⊗ A Π Q (14)
which is compatible with the COHA product and whose space of primitive elements is g ^ Π Q \widehat{\mathfrak{g}}^{\Pi_Q} g Π Q . By [5 ] the graded dimensions of g Π Q ( d ) \mathfrak{g}^{\Pi_Q}(d) g Π Q ( d ) and g ^ Π Q ( d ) \widehat{\mathfrak{g}}^{\Pi_Q}(d) g Π Q ( d ) are given by
∑ l dim g l Π Q ( d ) t l = A Q , d ( t 2 ) , ∑ l dim g ^ l Π Q ( d ) t l = 1 1 − t − 2 A Q , d ( t 2 ) , (15) \sum_l \dim\mathfrak{g}^{\Pi_Q}_l(d)t^l = A_{Q,d}(t^2), \qquad\sum_l \dim\widehat{\mathfrak{g}}^{\Pi_Q}_l(d)t^l = \frac{1}{1-t^{-2}}A_{Q,d}(t^2), \tag*{(15)} l ∑ dim g l Π Q ( d ) t l = A Q , d ( t 2 ) , l ∑ dim g l Π Q ( d ) t l = 1 − t − 2 1 A Q , d ( t 2 ) , (15)
where A Q , d ( q ) A_{Q,d}(q) A Q , d ( q ) is Kac’s polynomial counting absolutely indecomposable representations of Q Q Q of dimension d d d over the finite field F q \mathbb{F}_q F q . We define the Lie subalgebras
g I Q , g ^ I Q ⊂ A I Q \mathfrak{g}^{IQ},\ \widehat{\mathfrak{g}}^{IQ} \subset A_{IQ} g I Q , g I Q ⊂ A I Q
through the identification (9) , so that (12) , (13) hold with I Q IQ I Q in place of Q Q Q .
Remark 2.3. The above BPS Lie algebra formalism also applies to
H Λ n i l ( Rep Π Q , Q ) ≅ A I Q n i l H^{\Lambda^{\mathrm{nil}}}(\operatorname{Rep}_{\Pi_Q},\mathbb{Q}) \cong A^{\mathrm{nil}}_{IQ} H Λ nil ( Rep Π Q , Q ) ≅ A I Q nil
There is a Lie subalgebra
g ^ I Q , n i l ⊂ A I Q n i l \widehat{\mathfrak{g}}^{IQ,\mathrm{nil}} \subset A^{\mathrm{nil}}_{IQ} g I Q , nil ⊂ A I Q nil
such that
U ( g ^ I Q , n i l ) = A I Q n i l , g ^ I Q , n i l ≅ g ^ I Q [ − 2 ] (as vector spaces) U(\widehat{\mathfrak{g}}^{IQ,\mathrm{nil}}) = A^{\mathrm{nil}}_{IQ}, \qquad\widehat{\mathfrak{g}}^{IQ,\mathrm{nil}} \cong\widehat{\mathfrak{g}}^{IQ}[-2]\ \text{(as vector spaces)} U ( g I Q , nil ) = A I Q nil , g I Q , nil ≅ g I Q [ − 2 ] (as vector spaces)
where [ n ] [n] [ n ] indicates a shift by n n n in the cohomological grading. The Lie algebra structures on g ^ I Q , n i l \widehat{\mathfrak{g}}^{IQ,\mathrm{nil}} g I Q , nil and g ^ I Q \widehat{\mathfrak{g}}^{IQ} g I Q are very different: the latter is the positive half of a generalized Kac-Moody algebra by [11 ] (see also [10 ] ), while the former can be shown to be abelian by [34 ] . We will not need this.
The following characterizations of g ^ Π Q \widehat{\mathfrak{g}}^{\Pi_Q} g Π Q and g ^ I Q \widehat{\mathfrak{g}}^{IQ} g I Q play a key role in this paper. Recall that
ι ~ s n i l : Λ d s n i l → Rep d Π Q , ι s n i l : Rep d s n i l I Q → Rep d I Q (16) \widetilde{\iota}^{\mathrm{snil}} : \Lambda_d^{\mathrm{snil}} \to\operatorname{Rep}_{d}\Pi_Q,\qquad\iota^{\mathrm{snil}} : \operatorname{Rep}^{\mathrm{snil}}_{d}IQ \to\operatorname{Rep}_{d}IQ \tag*{(16)} ι snil : Λ d snil → Rep d Π Q , ι snil : Rep d snil I Q → Rep d I Q (16)
denote the closed immersions.
Proposition 2.4. Under the dimensional reduction isomorphisms (9) and (10) we have
Im ( ι ~ ∗ s n i l : H Λ s n i l ( Rep d Π Q , Q ) → A Π Q ( d ) ) = g ^ Π Q ( d ) , (17) \operatorname{Im}\left(\widetilde{\iota}^{\mathrm{snil}}_* : H^{\Lambda^{\mathrm{snil}}}(\operatorname{Rep}_{d}\Pi_Q,\mathbb{Q}) \to A_{\Pi_Q}(d)\right) = \widehat{\mathfrak{g}}^{\Pi_Q}(d), \tag*{(17)} Im ( ι ∗ snil : H Λ snil ( Rep d Π Q , Q ) → A Π Q ( d ) ) = g Π Q ( d ) , (17) Im ( ι ∗ s n i l : A I Q s n i l ( d ) → A I Q ( d ) ) = g ^ I Q ( d ) . (18) \operatorname{Im}\left(\iota^{\mathrm{snil}}_* : A^{\mathrm{snil}}_{IQ}(d) \to A_{IQ}(d)\right) = \widehat{\mathfrak{g}}^{IQ}(d). \tag*{(18)} Im ( ι ∗ snil : A I Q snil ( d ) → A I Q ( d ) ) = g I Q ( d ) . (18)
Proof. Equality (18) follows from (17) by base change and dimensional reduction. The proof of (17) is given in [12 ] in a more general setting. We provide some details in Appendix A. □ \square □
Tautological classes
For each vertex i ∈ Q 0 i \in Q_0 i ∈ Q 0 let V i V_i V i be the tautological rank d i d_i d i bundle on R e p d Π Q \mathop{\mathrm{Rep}}\nolimits_d\Pi_Q Rep d Π Q . We will use the same notation for R e p d I Q \mathop{\mathrm{Rep}}\nolimits_d I_Q Rep d I Q .
Lemma 2.5. (a) The pullbacks H G L d ∗ → H ∗ ( R e p d Π Q , Q ) H^*_{GL_d} \to H^*(\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q,\mathbb{Q}) H G L d ∗ → H ∗ ( Rep d Π Q , Q ) and H G L d ∗ → H ∗ ( R e p d I Q , Q ) H^*_{GL_d} \to H^*(\mathop{\mathrm{Rep}}\nolimits_d I_Q,\mathbb{Q}) H G L d ∗ → H ∗ ( Rep d I Q , Q ) are isomorphisms.
(b) H ∗ ( R e p d Π Q , Q ) H^*(\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q,\mathbb{Q}) H ∗ ( Rep d Π Q , Q ) and H ∗ ( R e p d I Q , Q ) H^*(\mathop{\mathrm{Rep}}\nolimits_d I_Q,\mathbb{Q}) H ∗ ( Rep d I Q , Q ) are generated by the Chern classes of the V i V_i V i .
Proof. The cohomology of a derived stack being equal to that of its classical truncation,
H ∗ ( R e p d Π Q , Q ) = H G d L ∗ ( μ d − 1 ( 0 ) c l , Q ) . H^*(\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q,\mathbb{Q}) = H^*_{G_dL}(\mu_d^{-1}(0)^{\mathrm{cl}},\mathbb{Q}). H ∗ ( Rep d Π Q , Q ) = H G d L ∗ ( μ d − 1 ( 0 ) cl , Q ) .
The scaling action t ⋅ ( x , x ∗ ) = ( t x , t x ∗ ) t \cdot(x,x^*)=(tx,tx^*) t ⋅ ( x , x ∗ ) = ( t x , t x ∗ ) on R e p d Q ‾ \mathop{\mathrm{Rep}}\nolimits_d\overline{Q} Rep d Q commutes with G L d GL_d G L d and preserves the conical subvariety μ d − 1 ( 0 ) c l \mu_d^{-1}(0)^{\mathrm{cl}} μ d − 1 ( 0 ) cl . Hence μ d − 1 ( 0 ) c l \mu_d^{-1}(0)^{\mathrm{cl}} μ d − 1 ( 0 ) cl is G L d GL_d G L d -equivariantly contractible to 0 0 0 , and we get
H G d L ∗ ( μ d − 1 ( 0 ) c l , Q ) ≅ H G d L ∗ . (19) H^*_{G_dL}(\mu_d^{-1}(0)^{\mathrm{cl}},\mathbb{Q}) \cong H^*_{G_dL}. \tag*{(19)} H G d L ∗ ( μ d − 1 ( 0 ) cl , Q ) ≅ H G d L ∗ . (19)
The proof is the same for R e p d I Q \mathop{\mathrm{Rep}}\nolimits_d I_Q Rep d I Q . The second statement follows obviously from the first. □
Consider the graded ring
S Q = Q [ c h i , j ; j ∈ Q 0 , i ≥ 1 ] , deg ( c h i , j ) = 2 i . S_Q=\mathbb{Q}[\mathrm{ch}_{i,j};\ j\in Q_0,\ i\geq1],\qquad\deg(\mathrm{ch}_{i,j})=2i. S Q = Q [ ch i , j ; j ∈ Q 0 , i ≥ 1 ] , deg ( ch i , j ) = 2 i .
For any d ∈ N Q 0 d\in\mathbb{N}^{Q_0} d ∈ N Q 0 the assignment c h i , j ↦ c h i ( V j ) \mathrm{ch}_{i,j}\mapsto\mathrm{ch}_i(V_j) ch i , j ↦ ch i ( V j ) defines surjective algebra morphisms
a d : S Q → H ∗ ( R e p d Π Q , Q ) , a d : S Q → H ∗ ( R e p d I Q , Q ) . a_d:S_Q\to H^*(\mathop{\mathrm{Rep}}\nolimits_d\Pi_Q,\mathbb{Q}),\qquad a_d:S_Q\to H^*(\mathop{\mathrm{Rep}}\nolimits_d I_Q,\mathbb{Q}). a d : S Q → H ∗ ( Rep d Π Q , Q ) , a d : S Q → H ∗ ( Rep d I Q , Q ) .
The action of cohomology on Borel-Moore homology gives rise to actions of S Q S_Q S Q on A Π Q ( d ) A_{\Pi_Q}(d) A Π Q ( d ) and A I Q ( d ) A_{I_Q}(d) A I Q ( d ) by
c ∙ x = a d ( c ) ∩ x . c\bullet x=a_d(c)\cap x. c ∙ x = a d ( c ) ∩ x .
To unburden the notation, we will often suppress the notation a d a_d a d .
Lemma 2.6. The isomorphism (9) intertwines the actions of S Q S_Q S Q on A Π Q A_{\Pi_Q} A Π Q and A I Q A_{I_Q} A I Q . □
The main compatibility properties between the action of S Q S_Q S Q and the COHA operations are summarized in the following lemma. We equip S Q S_Q S Q with the unique comultiplication Δ \Delta Δ satisfying
Δ ( c h i , j ) = c h i , j ⊗ 1 + 1 ⊗ c h i , j , ∀ j ∈ Q 0 , i ≥ 1. \Delta(\mathrm{ch}_{i,j})=\mathrm{ch}_{i,j}\otimes1+1\otimes\mathrm{ch}_{i,j},\qquad\forall j\in Q_0,\ i\geq1. Δ ( ch i , j ) = ch i , j ⊗ 1 + 1 ⊗ ch i , j , ∀ j ∈ Q 0 , i ≥ 1.
Lemma 2.7. For c ∈ S Q c\in S_Q c ∈ S Q and homogeneous x , y ∈ A Π Q x,y\in A_{\Pi_Q} x , y ∈ A Π Q , resp. x , y ∈ A I Q x,y\in A_{I_Q} x , y ∈ A I Q , we have
(a) c ∙ ( x y ) = ∑ ( c ( 1 ) ∙ x ) ( c ( 2 ) ∙ y ) c\bullet(xy)=\sum(c_{(1)}\bullet x)(c_{(2)}\bullet y) c ∙ ( x y ) = ∑ ( c ( 1 ) ∙ x ) ( c ( 2 ) ∙ y ) ,
(b) Δ ( c ∙ x ) = Δ ( c ) ∙ Δ ( x ) \Delta(c\bullet x)=\Delta(c)\bullet\Delta(x) Δ ( c ∙ x ) = Δ ( c ) ∙ Δ ( x ) ,
(c) S Q ∙ g ^ Π Q ⊆ g ^ Π Q S_Q\bullet\widehat{\mathfrak{g}}_{\Pi_Q}\subseteq\widehat{\mathfrak{g}}_{\Pi_Q} S Q ∙ g Π Q ⊆ g Π Q and likewise for g ^ I Q \widehat{\mathfrak{g}}_{I_Q} g I Q .
Proof. The proof is the same as in [12 §6] (see Appendix A. for a brief sketch of the argument). □
Although the Lie subalgebra g Π Q \mathfrak{g}_{\Pi_Q} g Π Q is defined canonically by geometric means, it may also be recovered from g ^ Π Q \widehat{\mathfrak{g}}_{\Pi_Q} g Π Q as follows. Assuming d ≠ 0 d\ne0 d = 0 , the tautological line bundle
L d = ⨂ i det V i \mathcal{L}_d=\bigotimes_i\det V_i L d = i ⨂ det V i
has a nontrivial weight under the standard B G m B\mathbb{G}_m B G m -action by scalar multiplication. It is thus a global equivariant parameter in the sense of [12 §2.5.2] . Then the inclusion g Π Q ⊂ g ^ Π Q \mathfrak{g}_{\Pi_Q}\subset\widehat{\mathfrak{g}}_{\Pi_Q} g Π Q ⊂ g Π Q induces an isomorphism
g Π Q ≅ g ^ Π Q / ( ⨁ d c 1 ( L d ) ∙ g ^ Π Q ( d ) ) . (20) \mathfrak{g}_{\Pi_Q}\cong\widehat{\mathfrak{g}}_{\Pi_Q}\bigg/\left(\bigoplus_d c_1(\mathcal{L}_d)\bullet\widehat{\mathfrak{g}}_{\Pi_Q}(d)\right). \tag*{(20)} g Π Q ≅ g Π Q / ( d ⨁ c 1 ( L d ) ∙ g Π Q ( d ) ) . (20)
Using the relation L d = gr d ′ , d ′ ′ ∗ ( L d ′ ⊗ L d ′ ′ ) \mathcal{L}_d=\operatorname{gr}^*_{d',d''}(\mathcal{L}_{d'}\otimes\mathcal{L}_{d''}) L d = gr d ′ , d ′′ ∗ ( L d ′ ⊗ L d ′′ ) whenever d = d ′ + d ′ ′ d=d'+d'' d = d ′ + d ′′ , one can show that ⨁ d c 1 ( L d ) ∙ g ^ Π Q ( d ) \bigoplus_d c_1(\mathcal{L}_d)\bullet\widehat{\mathfrak{g}}_{\Pi_Q}(d) ⨁ d c 1 ( L d ) ∙ g Π Q ( d ) is a Lie ideal and (2.17) is a Lie algebra isomorphism.
The reduced virtual fundamental class
The moment map μ d \mu_d μ d of (5) factors through
g l d 0 = ker ( tr : g l d → C ) . \mathfrak{gl}_d^0=\ker\left(\operatorname{tr}:\mathfrak{gl}_d\to\mathbb{C}\right). gl d 0 = ker ( tr : gl d → C ) .
Let μ d 0 : Rep d Q ‾ → g l d 0 \mu_d^0:\operatorname{Rep}_d\overline{Q}\to\mathfrak{gl}_d^0 μ d 0 : Rep d Q → gl d 0 be the factorization. Set
Rep d r d Π Q = ( μ d 0 ) − 1 ( 0 ) / G L d . \operatorname{Rep}_d^{\mathrm{rd}}\Pi_Q=(\mu_d^0)^{-1}(0)/GL_d. Rep d rd Π Q = ( μ d 0 ) − 1 ( 0 ) / G L d .
Then, we have
( Rep d r d Π Q ) c l = ( Rep d Π Q ) c l , vdim ( Rep d r d Π Q ) = vdim Rep d Π Q + 1 = 1 − ( d , d ) \left(\operatorname{Rep}_d^{\mathrm{rd}}\Pi_Q\right)^{\mathrm{cl}}=\left(\operatorname{Rep}_d\Pi_Q\right)^{\mathrm{cl}},\qquad\operatorname{vdim}\left(\operatorname{Rep}_d^{\mathrm{rd}}\Pi_Q\right)=\operatorname{vdim}\operatorname{Rep}_d\Pi_Q+1=1-(d,d) ( Rep d rd Π Q ) cl = ( Rep d Π Q ) cl , vdim ( Rep d rd Π Q ) = vdim Rep d Π Q + 1 = 1 − ( d , d )
and the isomorphism of derived stacks
Rep d Π Q ≃ Rep d r d Π Q × Spec Sym ( C [ 1 ] ) . \operatorname{Rep}_d\Pi_Q\simeq\operatorname{Rep}_d^{\mathrm{rd}}\Pi_Q\times\operatorname{Spec}\operatorname{Sym}(\mathbb{C}[1]). Rep d Π Q ≃ Rep d rd Π Q × Spec Sym ( C [ 1 ]) .
The reduced virtual fundamental class of Rep d Π Q \operatorname{Rep}_d\Pi_Q Rep d Π Q is
[ Rep d Π Q ] r d = [ Rep d r d Π Q ] ∈ H − 2 ( d , d ) + 2 ( Rep d Π Q , Q ) . [\operatorname{Rep}_d\Pi_Q]^{\mathrm{rd}}=[\operatorname{Rep}_d^{\mathrm{rd}}\Pi_Q]\in H_{-2(d,d)+2}(\operatorname{Rep}_d\Pi_Q,\mathbb{Q}). [ Rep d Π Q ] rd = [ Rep d rd Π Q ] ∈ H − 2 ( d , d ) + 2 ( Rep d Π Q , Q ) .
Note that [ Rep d Π Q ] = 0 [\operatorname{Rep}_d\Pi_Q]=0 [ Rep d Π Q ] = 0 for d ≠ 0 d\ne0 d = 0 , because the obstruction theory contains a summand C \mathbb{C} C .
We need the following microlocal characterization of [ Rep d r d Π Q ] [\operatorname{Rep}_d^{\mathrm{rd}}\Pi_Q] [ Rep d rd Π Q ] . Namely, by [12 §7.5] , the reduced virtual class [ Rep d r d Π Q ] [\operatorname{Rep}_d^{\mathrm{rd}}\Pi_Q] [ Rep d rd Π Q ] is the image of 1 ∈ Q 1\in\mathbb{Q} 1 ∈ Q under the chain of maps
Q ≃ H ( d , d ) − 2 ( Rep d Π Q , ( ι ~ s c a l ) ! ( ι ~ s c a l ) ! φ Rep d Π Q ) → H ( d , d ) − 2 ( Rep d Π Q , φ Rep d Π Q ) ≃ H 2 − 2 ( d , d ) ( Rep d Π Q , Q ) . \mathbb{Q}\simeq H^{(d,d)-2}\left(\operatorname{Rep}_d\Pi_Q,(\widetilde{\iota}^{\mathrm{scal}})!(\widetilde{\iota}^{\mathrm{scal}})!\varphi_{\operatorname{Rep}_d\Pi_Q}\right)\to H^{(d,d)-2}\left(\operatorname{Rep}_d\Pi_Q,\varphi_{\operatorname{Rep}_d\Pi_Q}\right)\simeq H_{2-2(d,d)}(\operatorname{Rep}_d\Pi_Q,\mathbb{Q}). Q ≃ H ( d , d ) − 2 ( Rep d Π Q , ( ι scal )! ( ι scal )! φ Rep d Π Q ) → H ( d , d ) − 2 ( Rep d Π Q , φ Rep d Π Q ) ≃ H 2 − 2 ( d , d ) ( Rep d Π Q , Q ) .
Proposition 2.8. For any d ∈ N Q 0 d\in\mathbb{N}^{Q_0} d ∈ N Q 0 we have [ Rep d r d Π Q ] ∈ g Π Q ( d ) [\operatorname{Rep}_d^{\mathrm{rd}}\Pi_Q]\in\mathfrak{g}^{\Pi_Q}(d) [ Rep d rd Π Q ] ∈ g Π Q ( d ) .
Proof. This follows from the microlocal characterization of g Π Q \mathfrak{g}^{\Pi_Q} g Π Q and the chain of inclusions
Λ d s c a l ⊂ Λ d s n i l ⊂ Rep ~ d Π Q . \Lambda_d^{\mathrm{scal}}\subset\Lambda_d^{\mathrm{snil}}\subset\widetilde{\operatorname{Rep}}_d\Pi_Q. Λ d scal ⊂ Λ d snil ⊂ Rep d Π Q .
Proposition 2.9. Under the isomorphism (9) , the reduced virtual fundamental class [ Rep d r d Π Q ] [\operatorname{Rep}_d^{\mathrm{rd}}\Pi_Q] [ Rep d rd Π Q ] is the pushforward ( ι s c a l ) ∗ [ Rep d s c a l I Q ] (\iota^{\mathrm{scal}})_*[\operatorname{Rep}_d^{\mathrm{scal}}IQ] ( ι scal ) ∗ [ Rep d scal I Q ] of the fundamental class of the smooth stack Rep d s c a l I Q ≃ Rep d Q × A 1 \operatorname{Rep}_d^{\mathrm{scal}}IQ\simeq\operatorname{Rep}_dQ\times\mathbb{A}^1 Rep d scal I Q ≃ Rep d Q × A 1 .
Proof. The square
Λ d s c a l → ι ~ s c a l Rep ~ d Π Q κ ′ ↓ ↓ κ ′ Rep d s c a l I Q → ι s c a l Rep d I Q \begin{CD}
\Lambda_d^{\mathrm{scal}} @>{\widetilde{\iota}^{\mathrm{scal}}}>> \widetilde{\operatorname{Rep}}_d\Pi_Q \\
@V{\kappa'}VV @VV{\kappa'}V \\
\operatorname{Rep}_d^{\mathrm{scal}}IQ @>{\iota^{\mathrm{scal}}}>> \operatorname{Rep}_dIQ
\end{CD} Λ d scal κ ′ ↓ ⏐ Rep d scal I Q ι scal ι scal Rep d Π Q ↓ ⏐ κ ′ Rep d I Q
is Cartesian. Since κ ′ \kappa' κ ′ is proper on the relevant supports, base change gives a commutative diagram
κ ∗ ′ ( ι ~ s c a l ) ! ( ι ~ s c a l ) ! φ Rep ~ d Π Q → κ ∗ ′ φ Rep ~ d Π Q ≃ ↓ ↓ ≃ ( ι s c a l ) ! ( ι s c a l ) ! ( D Q Rep d I Q ) = ( ι s c a l ) ! D Q Rep d s c a l I Q → D Q Rep d I Q , \begin{CD}
\kappa'_*(\widetilde{\iota}^{\mathrm{scal}})!(\widetilde{\iota}^{\mathrm{scal}})!\varphi_{\widetilde{\operatorname{Rep}}_d\Pi_Q} @>>> \kappa'_*\varphi_{\widetilde{\operatorname{Rep}}_d\Pi_Q} \\
@V{\simeq}VV @VV{\simeq}V \\
(\iota^{\mathrm{scal}})!(\iota^{\mathrm{scal}})!(\mathbb{D}\mathbb{Q}_{\operatorname{Rep}_dIQ})=(\iota^{\mathrm{scal}})!\mathbb{D}\mathbb{Q}_{\operatorname{Rep}_d^{\mathrm{scal}}IQ} @>>> \mathbb{D}\mathbb{Q}_{\operatorname{Rep}_dIQ},
\end{CD} κ ∗ ′ ( ι scal )! ( ι scal )! φ Rep d Π Q ≃ ↓ ⏐ ( ι scal )! ( ι scal )! ( D Q Rep d I Q ) = ( ι scal )! D Q Rep d scal I Q κ ∗ ′ φ Rep d Π Q ↓ ⏐ ≃ D Q Rep d I Q ,
in which the horizontal arrows are the counit maps. On global sections the bottom arrow is the map
( ι s c a l ) ∗ : H ∗ B M ( Rep d s c a l I Q , Q ) → H ∗ B M ( Rep d I Q , Q ) . (\iota^{\mathrm{scal}})_*:H_*^{\mathrm{BM}}(\operatorname{Rep}_d^{\mathrm{scal}}IQ,\mathbb{Q})\to H_*^{\mathrm{BM}}(\operatorname{Rep}_dIQ,\mathbb{Q}). ( ι scal ) ∗ : H ∗ BM ( Rep d scal I Q , Q ) → H ∗ BM ( Rep d I Q , Q ) .
The top degree of H ∗ B M ( Rep d s c a l I Q , Q ) H_*^{\mathrm{BM}}(\operatorname{Rep}_d^{\mathrm{scal}}IQ,\mathbb{Q}) H ∗ BM ( Rep d scal I Q , Q ) is spanned by the fundamental class, which corresponds to the generator of the top left-hand side. □ \square □
Thanks to Lemma 2.6 and Proposition 2.9, the isomorphism (2.6) identifies
c ∩ [ Rep d r d Π Q ] ← ( 2.6 ) ( ι s c a l ) ∗ ( c ∩ [ Rep d s c a l I Q ] ) , ∀ c ∈ S Q . \mathfrak{c} \cap[\operatorname{Rep}^{\mathrm{rd}}_{\mathbf{d}}\Pi_Q] \xleftarrow{\ (2.6)\ } (\iota^{\mathrm{scal}})_*\bigl(\mathfrak{c} \cap[\operatorname{Rep}^{\mathrm{scal}}_{\mathbf{d}} IQ]\bigr), \qquad\forall\mathfrak{c} \in S_Q. c ∩ [ Rep d rd Π Q ] ( 2.6 ) ( ι scal ) ∗ ( c ∩ [ Rep d scal I Q ] ) , ∀ c ∈ S Q .
We define
g ^ t a u t Π Q ( d ) = S Q ∙ [ Rep d r d Π Q ] , g ^ t a u t Π Q = ⨁ d g ^ t a u t Π Q ( d ) \widehat{\mathfrak{g}}^{\Pi_Q}_{\mathrm{taut}}(\mathbf{d}) = S_Q \bullet[\operatorname{Rep}^{\mathrm{rd}}_{\mathbf{d}}\Pi_Q], \qquad\widehat{\mathfrak{g}}^{\Pi_Q}_{\mathrm{taut}} = \bigoplus_{\mathbf{d}} \widehat{\mathfrak{g}}^{\Pi_Q}_{\mathrm{taut}}(\mathbf{d}) g taut Π Q ( d ) = S Q ∙ [ Rep d rd Π Q ] , g taut Π Q = d ⨁ g taut Π Q ( d )
and likewise for g ^ t a u t I Q \widehat{\mathfrak{g}}^{IQ}_{\mathrm{taut}} g taut I Q . By Lemma 2.7 (c), we have
g ^ t a u t Π Q ⊆ g ^ Π Q , g ^ t a u t I Q ⊆ g ^ I Q . (21) \widehat{\mathfrak{g}}^{\Pi_Q}_{\mathrm{taut}} \subseteq\widehat{\mathfrak{g}}^{\Pi_Q}, \qquad\widehat{\mathfrak{g}}^{IQ}_{\mathrm{taut}} \subseteq\widehat{\mathfrak{g}}^{IQ}. \tag*{(21)} g taut Π Q ⊆ g Π Q , g taut I Q ⊆ g I Q . (21)
A quiver analog of Markman’s theorem.
The aim of this section is to prove the following quiver analog of [12 Thm. 7.5] .
Theorem 3.1. We have
g ^ t a u t Π Q = g ^ Π Q , g ^ t a u t I Q = g ^ I Q . (22) \widehat{\mathfrak{g}}^{\Pi_Q}_{\mathrm{taut}} = \widehat{\mathfrak{g}}^{\Pi_Q}, \qquad\widehat{\mathfrak{g}}^{IQ}_{\mathrm{taut}} = \widehat{\mathfrak{g}}^{IQ}. \tag*{(22)} g taut Π Q = g Π Q , g taut I Q = g I Q . (22) Equivalently, the inclusions of (2.18) are isomorphisms.
The two equalities in (3.1) are equivalent to each other. By the PBW isomorphism (2.10), it is enough to prove that the COHA multiplication induces either of the equivalent isomorphisms
Sym ( ⨁ d g ^ t a u t Π Q ( d ) ) → ≅ ⨁ d A Π Q ( d ) , Sym ( ⨁ d g ^ t a u t I Q ( d ) ) → ≅ ⨁ d A I Q ( d ) . (23) \operatorname{Sym}\left(\bigoplus_{\mathbf{d}} \widehat{\mathfrak{g}}^{\Pi_Q}_{\mathrm{taut}}(\mathbf{d})\right) \xrightarrow{\ \cong\ } \bigoplus_{\mathbf{d}} A^{\Pi_Q}(\mathbf{d}), \qquad\operatorname{Sym}\left(\bigoplus_{\mathbf{d}} \widehat{\mathfrak{g}}^{IQ}_{\mathrm{taut}}(\mathbf{d})\right) \xrightarrow{\ \cong\ } \bigoplus_{\mathbf{d}} A^{IQ}(\mathbf{d}). \tag*{(23)} Sym ( d ⨁ g taut Π Q ( d ) ) ≅ d ⨁ A Π Q ( d ) , Sym ( d ⨁ g taut I Q ( d ) ) ≅ d ⨁ A I Q ( d ) . (23)
Our strategy is to prove the second isomorphism in (3.2). To this end, we consider the nilpotent COHA A I Q n i l A^{\mathrm{nil}}_{IQ} A I Q nil , the seminilpotent vector space A I Q s n i l A^{\mathrm{snil}}_{IQ} A I Q snil and we use the microlocal characterisation of g ^ I Q \widehat{\mathfrak{g}}^{IQ} g I Q in Proposition 2.4.
Remark 3.2. Theorem 3.1 should be compared to the Kirwan surjectivity theorem for Nakajima quiver varieties which is proved in [28 ] . Here, we are considering the entire stack of representations of the preprojective algebra and not an open substack of semistable points for some stability condition, which accounts for the fact the entire H ∗ ( Rep d Π Q , Q ) H_*(\operatorname{Rep}_{\mathbf{d}}\Pi_Q,\mathbb{Q}) H ∗ ( Rep d Π Q , Q ) is not tautologically generated.
The map Φ \Phi Φ .
Let us put
g ^ 0 ( d ) = S Q ∙ [ Rep d 0 I Q ] ⊆ A I Q n i l ( d ) . \widehat{\mathfrak{g}}^0(d) = S_Q \bullet[\operatorname{Rep}^0_d IQ] \subseteq A^{\mathrm{nil}}_{IQ}(d). g 0 ( d ) = S Q ∙ [ Rep d 0 I Q ] ⊆ A I Q nil ( d ) .
For d ≠ 0 d \ne0 d = 0 we consider the composition of maps
Φ : A I Q n i l ( d ) → ≅ A I Q s n i l ( d ) → s ∗ s n i l A I Q ( d ) \Phi: A^{\mathrm{nil}}_{IQ}(d) \xrightarrow{\ \cong\ } A^{\mathrm{snil}}_{IQ}(d) \xrightarrow{\ s^{\mathrm{snil}}_*\ } A_{IQ}(d) Φ : A I Q nil ( d ) ≅ A I Q snil ( d ) s ∗ snil A I Q ( d )
The first map is the smooth pullback along the projection of Lemma 2.1. We also set Φ ( 1 ) = 1 \Phi(1)=1 Φ ( 1 ) = 1 in degree 0. Note that Φ \Phi Φ is of (co)homological degree 2.
Proposition 3.3. The following hold:
(a) Φ ( A I Q n i l ( d ) ) = g ^ I Q ( d ) \Phi(A^{\mathrm{nil}}_{IQ}(d)) = \widehat{\mathfrak{g}}^{IQ}(d) Φ ( A I Q nil ( d )) = g I Q ( d ) .
(b) Φ ( g ^ 0 ( d ) ) = g ^ t a u t I Q ( d ) \Phi(\widehat{\mathfrak{g}}^0(d)) = \widehat{\mathfrak{g}}^{IQ}_{\mathrm{taut}}(d) Φ ( g 0 ( d )) = g taut I Q ( d ) .
(c) Φ ( x ∗ x ′ ) = 0 \Phi(x * x') = 0 Φ ( x ∗ x ′ ) = 0 for all x ∈ A I Q n i l ( d ) x \in A^{\mathrm{nil}}_{IQ}(d) x ∈ A I Q nil ( d ) , x ′ ∈ A I Q n i l ( d ′ ) x' \in A^{\mathrm{nil}}_{IQ}(d') x ′ ∈ A I Q nil ( d ′ ) with d , d ′ ≠ 0 d,d' \ne0 d , d ′ = 0 , where ∗ * ∗ is the COHA product on A I Q n i l A^{\mathrm{nil}}_{IQ} A I Q nil .
Proof. The first statement is a direct consequence of Proposition 2.4 and the fact that
R e p d s n i l I Q ( d ) ≅ R e p d n i l I Q ( d ) × A 1 . \mathfrak{Rep}^{\mathrm{snil}}_{d}IQ(d) \cong\mathfrak{Rep}^{\mathrm{nil}}_{d}IQ(d) \times\mathbb{A}^{1}. Rep d snil I Q ( d ) ≅ Rep d nil I Q ( d ) × A 1 .
To prove (b), note that
Φ ( [ R e p d 0 I Q ] ) = ι ∗ s n i l ( [ R e p d 0 I Q × A 1 ] ) = ι ∗ s c a l ( [ R e p d s c a l I Q ] ) \Phi\left([\mathfrak{Rep}^{0}_{d}IQ]\right)=\iota^{\mathrm{snil}}_{*}\left([\mathfrak{Rep}^{0}_{d}IQ\times\mathbb{A}^{1}]\right)=\iota^{\mathrm{scal}}_{*}\left([\mathfrak{Rep}^{\mathrm{scal}}_{d}IQ]\right) Φ ( [ Rep d 0 I Q ] ) = ι ∗ snil ( [ Rep d 0 I Q × A 1 ] ) = ι ∗ scal ( [ Rep d scal I Q ] )
and that Φ \Phi Φ is S Q S_{Q} S Q -equivariant. By Proposition 2.9 we have
Φ ( c ∩ [ R e p d 0 I Q ] ) = ι ∗ s n i l ( c ∩ [ R e p d s c a l I Q ] ) = c ∩ ( ι s c a l ) ∗ [ R e p d s c a l I Q ] = c ∩ [ R e p d Π Q ] r d . \begin{aligned}
\Phi\left(c\cap[\mathfrak{Rep}^{0}_{d}IQ]\right)=\iota^{\mathrm{snil}}_{*}\left(c\cap[\mathfrak{Rep}^{\mathrm{scal}}_{d}IQ]\right) \\
&=c\cap(\iota^{\mathrm{scal}})_{*}[\mathfrak{Rep}^{\mathrm{scal}}_{d}IQ] \\
&=c\cap[\mathfrak{Rep}_{d}\Pi_{Q}]^{\mathrm{rd}}.
\end{aligned} Φ ( c ∩ [ Rep d 0 I Q ] ) = ι ∗ snil ( c ∩ [ Rep d scal I Q ] ) = c ∩ ( ι scal ) ∗ [ Rep d scal I Q ] = c ∩ [ Rep d Π Q ] rd .
The proof of (c) is more delicate. Consider the action on R e p d Π Q \mathfrak{Rep}_{d}\Pi_{Q} Rep d Π Q of the torus T = G m \mathbb{T}=\mathbb{G}_{m} T = G m which acts with weight 1 on θ \theta θ , weight − 1 -1 − 1 on x ∗ x^{*} x ∗ and weight 0 on x x x . This action preserves the potential W W W in (6) . Thus it commutes with the dimensional reduction isomorphisms (9) . All of the above spaces, algebras and maps admit T \mathbb{T} T -equivariant versions A I Q n i l , T A^{\mathrm{nil},\mathbb{T}}_{IQ} A I Q nil , T , etc., replacing everywhere H ∗ ( − , Q ) H_{*}(-,\mathbb{Q}) H ∗ ( − , Q ) by H ∗ T ( − , Q ) H^{\mathbb{T}}_{*}(-,\mathbb{Q}) H ∗ T ( − , Q ) . These are modules over H T ∗ = Q [ t ] H^{*}_{\mathbb{T}}=\mathbb{Q}[t] H T ∗ = Q [ t ] . The isomorphism of Lemma 2.1 is T \mathbb{T} T -equivariant, with T \mathbb{T} T acting on c ∈ A 1 c\in A^{1} c ∈ A 1 with weight 1. The self-intersection formula for the zero section s : R e p d n i l I Q → R e p d s n i l I Q s:\mathfrak{Rep}^{\mathrm{nil}}_{d}IQ\to\mathfrak{Rep}^{\mathrm{snil}}_{d}IQ s : Rep d nil I Q → Rep d snil I Q of the T \mathbb{T} T -equivariant line bundle
pr : R e p d s n i l I Q ≅ R e p d n i l I Q × A 1 → R e p d n i l I Q , \operatorname{pr}:\mathfrak{Rep}^{\mathrm{snil}}_{d}IQ\cong\mathfrak{Rep}^{\mathrm{nil}}_{d}IQ\times\mathbb{A}^{1}\to\mathfrak{Rep}^{\mathrm{nil}}_{d}IQ, pr : Rep d snil I Q ≅ Rep d nil I Q × A 1 → Rep d nil I Q ,
yields
( s T ) ∗ ( x ) = t ⋅ ( pr T ) ∗ ( x ) . (s^{\mathbb{T}})_{*}(x)=t\cdot(\operatorname{pr}^{\mathbb{T}})^{*}(x). ( s T ) ∗ ( x ) = t ⋅ ( pr T ) ∗ ( x ) .
Hence
ι ∗ n i l , T = t ⋅ Φ T : A I Q n i l , T ( d ) → A I Q T ( d ) , ( d ≠ 0 ) . \iota^{\mathrm{nil},\mathbb{T}}_{*}=t\cdot\Phi^{\mathbb{T}}:A^{\mathrm{nil},\mathbb{T}}_{IQ}(d)\to A^{\mathbb{T}}_{IQ}(d),\qquad(d\ne0). ι ∗ nil , T = t ⋅ Φ T : A I Q nil , T ( d ) → A I Q T ( d ) , ( d = 0 ) .
For x ∈ A I Q n i l , T ( d ) x\in A^{\mathrm{nil},\mathbb{T}}_{IQ}(d) x ∈ A I Q nil , T ( d ) and x ′ ∈ A I Q n i l , T ( d ′ ) x'\in A^{\mathrm{nil},\mathbb{T}}_{IQ}(d') x ′ ∈ A I Q nil , T ( d ′ ) with d , d ′ ≠ 0 d,d'\ne0 d , d ′ = 0 , we have
t Φ T ( x ∗ x ′ ) = ι ∗ n i l , T ( x ∗ x ′ ) = ι ∗ n i l , T ( x ) ι ∗ n i l , T ( x ′ ) = t 2 Φ T ( x ) ∗ Φ T ( x ′ ) . t\Phi^{\mathbb{T}}(x*x')=\iota^{\mathrm{nil},\mathbb{T}}_{*}(x*x')=\iota^{\mathrm{nil},\mathbb{T}}_{*}(x)\iota^{\mathrm{nil},\mathbb{T}}_{*}(x')=t^{2}\Phi^{\mathbb{T}}(x)*\Phi^{\mathbb{T}}(x'). t Φ T ( x ∗ x ′ ) = ι ∗ nil , T ( x ∗ x ′ ) = ι ∗ nil , T ( x ) ι ∗ nil , T ( x ′ ) = t 2 Φ T ( x ) ∗ Φ T ( x ′ ) .
From the purity in Proposition 2.2 we deduce that A I Q T ≅ A Π Q T A^{\mathbb{T}}_{IQ}\cong A^{\mathbb{T}}_{\Pi Q} A I Q T ≅ A Π Q T is a free Q [ t ] \mathbb{Q}[t] Q [ t ] -module with A I Q T / t A I Q T = A I Q A^{\mathbb{T}}_{IQ}/tA^{\mathbb{T}}_{IQ}=A_{IQ} A I Q T / t A I Q T = A I Q . Hence
Φ T ( x ∗ x ′ ) = t Φ T ( x ) ∗ Φ T ( x ′ ) . \Phi^{\mathbb{T}}(x*x')=t\Phi^{\mathbb{T}}(x)*\Phi^{\mathbb{T}}(x'). Φ T ( x ∗ x ′ ) = t Φ T ( x ) ∗ Φ T ( x ′ ) .
Reducing modulo t t t , i.e., applying the forgetful maps, which are compatible with pushforward, smooth pullback and products, gives
Φ ( x ˉ ∗ x ˉ ′ ) = 0. \Phi(\bar{x}*\bar{x}')=0. Φ ( x ˉ ∗ x ˉ ′ ) = 0.
To conclude, we use the fact that each class of A I Q n i l ( d ) A^{\mathrm{nil}}_{IQ}(d) A I Q nil ( d ) lifts to A I Q n i l , T ( d ) A^{\mathrm{nil},\mathbb{T}}_{IQ}(d) A I Q nil , T ( d ) since A I Q n i l A^{\mathrm{nil}}_{IQ} A I Q nil is pure. □
The generation theorem for A Π Q A_{\Pi Q} A Π Q
The following proposition is proved in the next section.
Proposition 3.4. The nilpotent COHA A I Q n i l A^{\mathrm{nil}}_{IQ} A I Q nil is generated by ⨁ d g ^ 0 ( d ) \bigoplus_{d}\widehat{\mathfrak{g}}^{0}(d) ⨁ d g 0 ( d ) as an algebra.
Using this result, we can prove the generation theorem Theorem 3.1 for A Π Q A_{\Pi Q} A Π Q .
Proof of Theorem 3.1. The theorem follows by combining Propositions 2.4 , 3.3 and 3.4 . Indeed, we have
A I Q n i l ( d ) = g ^ 0 ( d ) + ∑ d ′ + d ′ ′ = d , d ′ , d ′ ′ ≠ 0 A I Q n i l ( d ′ ) ∗ A I Q n i l ( d ′ ′ ) . A^{\mathrm{nil}}_{IQ}(d)=\widehat{\mathfrak{g}}^{0}(d)+\sum_{\substack{d'+d''=d,\\d',d''\ne0}}A^{\mathrm{nil}}_{IQ}(d')*A^{\mathrm{nil}}_{IQ}(d''). A I Q nil ( d ) = g 0 ( d ) + d ′ + d ′′ = d , d ′ , d ′′ = 0 ∑ A I Q nil ( d ′ ) ∗ A I Q nil ( d ′′ ) .
Applying Φ \Phi Φ we get
g ^ I Q ( d ) = Φ ( A I Q n i l ( d ) ) = Φ ( g ^ 0 ( d ) ) = g ^ t a u t I Q ( d ) . \widehat{\mathfrak{g}}^{IQ}(d)=\Phi(A^{\mathrm{nil}}_{IQ}(d))=\Phi(\widehat{\mathfrak{g}}^{0}(d))=\widehat{\mathfrak{g}}^{IQ}_{\mathrm{taut}}(d). g I Q ( d ) = Φ ( A I Q nil ( d )) = Φ ( g 0 ( d )) = g taut I Q ( d ) .
The theorem follows by (9) . □
The generation theorem for A I Q n i l \mathcal{A}^{\mathrm{nil}}_{IQ} A I Q nil
In order to prove Proposition 3.4, we first introduce a suitable stratification of M d n i l I Q \mathfrak{M}^{\mathrm{nil}}_{d}IQ M d nil I Q by smooth and cohomologically pure pieces.
Jordan types
A Jordan type of dimension d d d is a tuple of dimension vectors
d = ( d 1 , … , d s ) , d i ∈ N Q 0 , d s ≠ 0 , ∣ d ∣ = ∑ i i d i = d . d = (d^{1}, \ldots, d^{s}), \qquad d^{i} \in\mathbb{N}Q_{0}, \qquad d^{s} \ne0, \qquad|d| = \sum_{i} i d^{i} = d. d = ( d 1 , … , d s ) , d i ∈ N Q 0 , d s = 0 , ∣ d ∣ = i ∑ i d i = d .
For such d d d put
e i = d i + d i + 1 + ⋯ + d s , 1 ≤ i ≤ s , e s + 1 = 0 , e_{i} = d^{i} + d^{i+1} + \cdots+ d^{s}, \qquad1 \le i \le s, \qquad e_{s+1} = 0, e i = d i + d i + 1 + ⋯ + d s , 1 ≤ i ≤ s , e s + 1 = 0 ,
We have
e 1 ≥ e 2 ≥ ⋯ ≥ e s > 0 , ∑ i e i = d , d i = e i − e i + 1 e_{1} \ge e_{2} \ge\cdots\ge e_{s} > 0, \qquad\sum_{i} e_{i} = d, \qquad d^{i} = e_{i} - e_{i+1} e 1 ≥ e 2 ≥ ⋯ ≥ e s > 0 , i ∑ e i = d , d i = e i − e i + 1
and
e 1 + ⋯ + e i = dim ker θ i . (24) e_{1} + \cdots+ e_{i} = \dim\ker\theta^{i}. \tag*{(24)} e 1 + ⋯ + e i = dim ker θ i . (24)
Let
J ( d ) = { ( x , θ ) ∈ M d n i l I Q ; θ s = 0 , dim ( ker θ i / ker θ i − 1 ) = e i for 1 ≤ i ≤ s } . J(d) = \left\{(x,\theta) \in\mathfrak{M}^{\mathrm{nil}}_{d}IQ; \ \theta^{s}=0,\ \dim\left(\ker\theta^{i}/\ker\theta^{i-1}\right)=e_{i}\text{ for }1\le i\le s\right\}. J ( d ) = { ( x , θ ) ∈ M d nil I Q ; θ s = 0 , dim ( ker θ i / ker θ i − 1 ) = e i for 1 ≤ i ≤ s } .
Thus d i d^{i} d i is the dimension vector of the multiplicity space of Jordan blocks of size i i i . Since every ker θ i \ker\theta^{i} ker θ i is a subrepresentation
( M d n i l I Q ) c l = ∐ ∣ d ∣ = d J ( d ) \left(\mathfrak{M}^{\mathrm{nil}}_{d}IQ\right)^{\mathrm{cl}} = \coprod_{|d|=d} J(d) ( M d nil I Q ) cl = ∣ d ∣ = d ∐ J ( d )
is a finite decomposition into locally closed substacks. For s = 1 s=1 s = 1 we get
J ( ( d ) ) = M d 0 I Q ≅ M d Q . J((d))=\mathfrak{M}^{0}_{d}IQ\cong\mathfrak{M}_{d}Q. J (( d )) = M d 0 I Q ≅ M d Q .
Tautological bundles on J ( d ) J(d) J ( d )
On J ( d ) J(d) J ( d ) the subspaces ker θ i \ker\theta^{i} ker θ i form vector bundles, which are subrepresentations of the tautological quiver bundle. Multiplication by θ \theta θ induces injections
ker θ i + 1 / ker θ i ↪ ker θ i / ker θ i − 1 . \ker\theta^{i+1}/\ker\theta^{i}\hookrightarrow\ker\theta^{i}/\ker\theta^{i-1}. ker θ i + 1 / ker θ i ↪ ker θ i / ker θ i − 1 .
Let
V i = coker ( ker θ i + 1 / ker θ i → θ ker θ i / ker θ i − 1 ) . (25) \mathcal{V}^{i}=\operatorname{coker}\left(\ker\theta^{i+1}/\ker\theta^{i}\xrightarrow{\theta}\ker\theta^{i}/\ker\theta^{i-1}\right). \tag*{(25)} V i = coker ( ker θ i + 1 / ker θ i θ ker θ i / ker θ i − 1 ) . (25)
It is a C Q CQ C Q -representation of dimension e i − e i + 1 = d i e_{i}-e_{i+1}=d^{i} e i − e i + 1 = d i .
Lemma 3.5. The morphism
π d : J ( d ) ⟶ ∏ i = 1 s M d i Q , ( x , θ ) ↦ ( V i ) i , (26) \pi_{d}:J(d)\longrightarrow\prod_{i=1}^{s}\mathfrak{M}_{d^{i}}Q,\qquad(x,\theta)\mapsto(\mathcal{V}^{i})_{i}, \tag*{(26)} π d : J ( d ) ⟶ i = 1 ∏ s M d i Q , ( x , θ ) ↦ ( V i ) i , (26)
is an iterated vector bundle stack. Hence J ( d ) J(d) J ( d ) is smooth and irreducible, the Borel-Moore homology space H ∗ ( J ( d ) , Q ) H_{*}(J(d),\mathbb{Q}) H ∗ ( J ( d ) , Q ) is pure, and there is an isomorphism
⨂ i = 1 s H G L d i ∗ ≅ H ∗ ( J ( d ) , Q ) , ⨂ i c i ↦ π d ∗ ( ⨂ i c i ) ∩ [ J ( d ) ] . \bigotimes_{i=1}^{s}H^{*}_{GL_{d^{i}}}\cong H_{*}(J(d),\mathbb{Q}),\qquad\bigotimes_{i}c_{i}\mapsto\pi_{d}^{*}\left(\bigotimes_{i}c_{i}\right)\cap[J(d)]. i = 1 ⨂ s H G L d i ∗ ≅ H ∗ ( J ( d ) , Q ) , i ⨂ c i ↦ π d ∗ ( i ⨂ c i ) ∩ [ J ( d )] .
Moreover, we have
dim J ( d ) = − ∑ i , j min ( i , j ) ⟨ d i , d j ⟩ = − ∑ l = 1 s ⟨ e l , e l ⟩ . (27) \dim J(d)=-\sum_{i,j}\min(i,j)\langle d^{i},d^{j}\rangle=-\sum_{l=1}^{s}\langle e_{l},e_{l}\rangle. \tag*{(27)} dim J ( d ) = − i , j ∑ min ( i , j ) ⟨ d i , d j ⟩ = − l = 1 ∑ s ⟨ e l , e l ⟩ . (27)
Proof. This is a general result, valid for the tangent bundle to the stack of objects in a category of global dimension one and not only for M d I Q \mathfrak{M}_{d}IQ M d I Q , see, e.g., [33 ] . □
The order and closures.
For a , a ′ ∈ N Q 0 a,a' \in\mathbb{N}Q_0 a , a ′ ∈ N Q 0 we write a ′ < a a' < a a ′ < a if a − a ′ ∈ N Q 0 ∖ { 0 } a-a' \in\mathbb{N}Q_0 \setminus\{0\} a − a ′ ∈ N Q 0 ∖ { 0 } . For Jordan types d ‾ \underline{d} d and d ‾ ′ \underline{d}' d ′ of the same dimension d ‾ \underline{d} d , with associated sequences ( e i ) i (e_i)_i ( e i ) i and ( e i ′ ) i (e'_i)_i ( e i ′ ) i , we write d ‾ < d ‾ ′ \underline{d} < \underline{d}' d < d ′ if d ‾ ≠ d ‾ ′ \underline{d} \ne\underline{d}' d = d ′ and there exists l l l such that for each i < l i<l i < l we have
e 1 ′ + ⋯ + e i ′ = e 1 + ⋯ + e i , e 1 ′ + ⋯ + e l ′ < e 1 + ⋯ + e l . e'_1+\cdots+e'_i=e_1+\cdots+e_i,\qquad e'_1+\cdots+e'_l<e_1+\cdots+e_l. e 1 ′ + ⋯ + e i ′ = e 1 + ⋯ + e i , e 1 ′ + ⋯ + e l ′ < e 1 + ⋯ + e l .
By (3.3), we have d ‾ < d ‾ ′ \underline{d}<\underline{d}' d < d ′ if and only if dim ker θ i \dim\ker\theta^i dim ker θ i is the same for d ‾ \underline{d} d and for d ‾ ′ \underline{d}' d ′ if i < l i<l i < l and strictly larger for d ‾ \underline{d} d if i = l i=l i = l . Write d ‾ ≤ d ‾ ′ \underline{d}\leq\underline{d}' d ≤ d ′ if either d ‾ < d ‾ ′ \underline{d}<\underline{d}' d < d ′ or d ‾ = d ‾ ′ \underline{d}=\underline{d}' d = d ′ . We consider the substacks
Z ≤ d ‾ = ⋃ d ‾ ′ ≤ d ‾ J ( d ‾ ′ ) , Z < d ‾ = ⋃ d ‾ ′ < d ‾ J ( d ‾ ′ ) . Z_{\leq\underline{d}}=\bigcup_{\underline{d}'\leq\underline{d}}J(\underline{d}'),\qquad Z_{<\underline{d}}=\bigcup_{\underline{d}'<\underline{d}}J(\underline{d}'). Z ≤ d = d ′ ≤ d ⋃ J ( d ′ ) , Z < d = d ′ < d ⋃ J ( d ′ ) .
Lemma 3.6. For every d ‾ \underline{d} d the substacks Z ≤ d ‾ Z_{\leq\underline{d}} Z ≤ d and Z < d ‾ Z_{<\underline{d}} Z < d are closed in Rep d ‾ n i l I Q \operatorname{Rep}^{\mathrm{nil}}_{\underline{d}}IQ Rep d nil I Q .
Proof. Since there are finitely many strata, it is enough to consider the case of the substacks Z ≤ d ‾ Z_{\leq\underline{d}} Z ≤ d . The functions
( x , θ ) ↦ dim ker θ i ∈ N Q 0 (x,\theta)\mapsto\dim\ker\theta^i\in\mathbb{N}Q_0 ( x , θ ) ↦ dim ker θ i ∈ N Q 0
are upper semicontinuous. If y y y lies in the closure of the stratum of x x x , then
dim ker θ y i ≥ dim ker θ x i . \dim\ker\theta_y^i\geq\dim\ker\theta_x^i. dim ker θ y i ≥ dim ker θ x i .
Hence at the first i i i where they differ, the inequality is strict, i.e., the type of y y y is ≤ \leq ≤ the type of x x x . So Z ≤ d ‾ Z_{\leq\underline{d}} Z ≤ d is closed under specialization. It is constructible, being a finite union of locally closed strata, hence closed. □
Corollary 3.7. For every d ‾ \underline{d} d there is a short exact sequence
0 → H ∗ ( Z < d ‾ , Q ) → H ∗ ( Z ≤ d ‾ , Q ) → H ∗ ( J ( d ‾ ) , Q ) → 0. 0\to H_*(Z_{<\underline{d}},\mathbb{Q})\to H_*(Z_{\leq\underline{d}},\mathbb{Q})\to H_*(J(\underline{d}),\mathbb{Q})\to0. 0 → H ∗ ( Z < d , Q ) → H ∗ ( Z ≤ d , Q ) → H ∗ ( J ( d ) , Q ) → 0.
The mixed Hodge structure of A I Q n i l ( d ‾ ) A^{\mathrm{nil}}_{IQ}(\underline{d}) A I Q nil ( d ) is pure.
Proof. Fix a total order ⪯ \preceq ⪯ on Jordan types refining ≤ \leq ≤ . By Lemma 3.6, the substacks Z ⪯ d ‾ = ⋃ e ‾ ⪯ d ‾ J ( e ‾ ) Z_{\preceq\underline{d}}=\bigcup_{\underline{e}\preceq\underline{d}}J(\underline{e}) Z ⪯ d = ⋃ e ⪯ d J ( e ) are closed. Each step of the resulting filtration gives a localization long exact sequence, all of whose terms are pure by Lemma 3.5 and induction. Hence, the long exact sequences split. □
Ordered products.
Fix d ‾ \underline{d} d with associated sequence e ‾ = ( e 1 , … , e s ) \underline{e}=(e_1,\ldots,e_s) e = ( e 1 , … , e s ) . Set f ‾ = ( e s , … , e 1 ) \underline{f}=(e_s,\ldots,e_1) f = ( e s , … , e 1 ) . We consider the nilpotent COHA induction diagram for flags of subobjects with subquotients of dimensions e s , … , e 1 e_s,\ldots,e_1 e s , … , e 1 :
∏ i Rep e i n i l I Q ← q f ‾ Filt f ‾ n i l I Q → p f ‾ Rep d ‾ n i l I Q . (28) \prod_i\operatorname{Rep}^{\mathrm{nil}}_{e_i}IQ \xleftarrow{q_{\underline{f}}}\operatorname{Filt}^{\mathrm{nil}}_{\underline{f}}IQ\xrightarrow{p_{\underline{f}}}\operatorname{Rep}^{\mathrm{nil}}_{\underline{d}}IQ. \tag*{(28)} i ∏ Rep e i nil I Q q f Filt f nil I Q p f Rep d nil I Q . (28)
We have
a s ∗ ⋯ ∗ a 1 = ( p f ‾ ) ∗ ( q f ‾ ) ! ( a s ⊠ ⋯ ⊠ a 1 ) . a_s*\cdots*a_1=(p_{\underline{f}})_*(q_{\underline{f}})^!(a_s\boxtimes\cdots\boxtimes a_1). a s ∗ ⋯ ∗ a 1 = ( p f ) ∗ ( q f ) ! ( a s ⊠ ⋯ ⊠ a 1 ) .
We define the stack
Filt f ‾ 0 I Q = q f ‾ − 1 ( ∏ i Rep e i 0 I Q ) . \operatorname{Filt}^{0}_{\underline{f}}IQ=q_{\underline{f}}^{-1}\left(\prod_i\operatorname{Rep}^{0}_{e_i}IQ\right). Filt f 0 I Q = q f − 1 ( i ∏ Rep e i 0 I Q ) .
This stack parametrizes flags 0 = F 0 ⊂ F 1 ⊂ ⋯ ⊂ F s 0=F_0\subset F_1\subset\cdots\subset F_s 0 = F 0 ⊂ F 1 ⊂ ⋯ ⊂ F s of subrepresentations of a d ‾ \underline{d} d -dimensional representation ( x , θ ) (x,\theta) ( x , θ ) with θ ( F i ) ⊆ F i − 1 \theta(F_i)\subseteq F_{i-1} θ ( F i ) ⊆ F i − 1 and dim F i / F i − 1 = e i \dim F_i/F_{i-1}=e_i dim F i / F i − 1 = e i . Observe that if θ ( F i ) ⊆ F i − 1 \theta(F_i)\subseteq F_{i-1} θ ( F i ) ⊆ F i − 1 for all i i i , then F i ⊆ ker θ i F_i\subseteq\ker\theta^i F i ⊆ ker θ i . Hence dim ker θ i ≥ e 1 + ⋯ + e i \dim\ker\theta^i\geq e_1+\cdots+e_i dim ker θ i ≥ e 1 + ⋯ + e i for all i i i , and the Jordan type of ( x , θ ) (x,\theta) ( x , θ ) is ≤ d ‾ \leq\underline{d} ≤ d . Thus, we have
p f ‾ ( Filt f ‾ 0 I Q ) ⊆ Z ≤ d ‾ . p_{\underline{f}}(\operatorname{Filt}^{0}_{\underline{f}}IQ)\subseteq Z_{\leq\underline{d}}. p f ( Filt f 0 I Q ) ⊆ Z ≤ d .
Let q f ‾ 0 q^0_{\underline{f}} q f 0 and p f ‾ 0 p^0_{\underline{f}} p f 0 be the restrictions of the maps p f ‾ , q f ‾ p_{\underline{f}},q_{\underline{f}} p f , q f to Filt f ‾ 0 I Q \operatorname{Filt}^{0}_{\underline{f}}IQ Filt f 0 I Q . We define
( Filt f ‾ 0 I Q ) ∘ = ( p f ‾ 0 ) − 1 ( J ( d ‾ ) ) , \left(\operatorname{Filt}^{0}_{\underline{f}}IQ\right)^\circ=(p^0_{\underline{f}})^{-1}(J(\underline{d})), ( Filt f 0 I Q ) ∘ = ( p f 0 ) − 1 ( J ( d )) ,
with the open embedding
j : ( Filt f ‾ 0 I Q ) ∘ ↪ Filt f ‾ 0 I Q . j:\left(\operatorname{Filt}^{0}_{\underline{f}}IQ\right)^\circ\hookrightarrow\operatorname{Filt}^{0}_{\underline{f}}IQ. j : ( Filt f 0 I Q ) ∘ ↪ Filt f 0 I Q .
The induction diagram (3.7) fits into the following commutative diagram
\begin{CD}
\prod_i \mathop{\mathrm{Rep}}\nolimits_{e_i}^{\mathrm{nil}} IQ @<q_f<< \widetilde{\mathrm{Filt}}\nolimits_f^{\mathrm{nil}} IQ @<i<< (\widetilde{\mathrm{Filt}}\nolimits_f^{\mathrm{nil}} IQ)^{\mathrm{cl}} @>p_f>> \mathop{\mathrm{Rep}}\nolimits_d^{\mathrm{nil}} IQ \\
@A j AA @A j AA @A j^{\mathrm{cl}} AA @A j AA \\
\prod_i \mathop{\mathrm{Rep}}\nolimits_{e_i}^{0} IQ @<q_f^0<< \widetilde{\mathrm{Filt}}\nolimits_f^0 IQ @<i<< (\widetilde{\mathrm{Filt}}\nolimits_f^0 IQ)^{\mathrm{cl}} @>p_f^0>> Z_{\le d} \\
@. @A j AA @A j^{\mathrm{cl}} AA @A j AA \\
@. (\widetilde{\mathrm{Filt}}\nolimits_f^0 IQ)^\circ@<i^\circ<< (\widetilde{\mathrm{Filt}}\nolimits_f^0 IQ)^{\circ,\mathrm{cl}} @>p_f^{00}>> J(d). \tag*{(29)}
\end{CD}
The pushforward functors i ∗ i_* i ∗ and i ∗ ∘ i^\circ_* i ∗ ∘ are isomorphisms in Borel-Moore homology. Note that the closed substack Z ≤ d Z_{\le d} Z ≤ d has no obvious derived enhancement. Finally, recall the bundles V i \mathcal{V}^i V i on J ( d ) J(d) J ( d ) defined in (25) . Let E i \mathcal{E}^i E i be the tautological bundle on R e p e i 0 I Q \mathop{\mathrm{Rep}}\nolimits_{e_i}^0 IQ Rep e i 0 I Q . We set E s + 1 = 0 \mathcal{E}^{s+1}=0 E s + 1 = 0 .
Lemma 3.8. The following hold:
(a) The maps i ∘ i^\circ i ∘ and p_f^{00}: (\widetilde{\mathrm{Filt}}\nolimits_f^0 IQ)^{\circ,\mathrm{cl}} \to J(d) are isomorphisms.
(b) In K K K -theory we have
( p f 00 ) ∗ [ V i ] = ( j 0 ) ∗ ( [ E i ] − [ E i + 1 ] ) , 1 ≤ i ≤ s . (p_f^{00})^*[\mathcal{V}^i] = (j^0)^*( [\mathcal{E}^i] - [\mathcal{E}^{i+1}] ), \qquad1 \le i \le s. ( p f 00 ) ∗ [ V i ] = ( j 0 ) ∗ ([ E i ] − [ E i + 1 ]) , 1 ≤ i ≤ s . (c) In Borel-Moore homology we have
(j^0)^*(q_f^0)_*\left(\boxtimes_i[\mathop{\mathrm{Rep}}\nolimits_{e_i}^0 IQ]\right) = [(\widetilde{\mathrm{Filt}}\nolimits_f^0 IQ)^\circ].
Proof. The second statement of (a) follows from the fact that if ( x , θ ) ∈ J ( d ) (x,\theta) \in J(d) ( x , θ ) ∈ J ( d ) then F i = ker θ i F_i=\ker\theta^i F i = ker θ i is the unique flag of subrepresentations such that θ ( F i ) ⊂ F i − 1 \theta(F_i)\subset F_{i-1} θ ( F i ) ⊂ F i − 1 and dim F i / F i − 1 = e i \dim F_i/F_{i-1}=e_i dim F i / F i − 1 = e i . Next, we argue that (\widetilde{\mathrm{Filt}}\nolimits_f^0 IQ)^\circ is classical. As j j j is an open embedding and q f 0 q_f^0 q f 0 is quasi-smooth, (\widetilde{\mathrm{Filt}}\nolimits_f^0 IQ)^\circ is itself quasi-smooth. Moreover, the relative (virtual) dimension of q f 0 q_f^0 q f 0 is
− ∑ i < j ( e j , e i ) C Q [ θ ] = 0 , -\sum_{i<j}(e_j,e_i)_{CQ[\theta]}=0, − i < j ∑ ( e j , e i ) C Q [ θ ] = 0 ,
because the Euler form on C Q [ θ ] CQ[\theta] C Q [ θ ] vanishes. Thus we have
\operatorname{vdim}\widetilde{\mathrm{Filt}}\nolimits_f^0 IQ
=\operatorname{vdim}\prod_i\mathop{\mathrm{Rep}}\nolimits_{e_i}^0 IQ
=-\sum_i\langle e_i,e_i\rangle
=\operatorname{vdim}J(d)
by (27) . Thus the dimension of (\widetilde{\mathrm{Filt}}\nolimits_f^0 IQ)^\circ coincides with that of its classical truncation. An irreducible quasi-smooth derived stack whose classical truncation is smooth of dimension equal to the virtual dimension has a vanishing obstruction sheaf, hence is classical, and its virtual fundamental class is the fundamental class by Theorem B.1 . We turn to (b). If (x,\theta,\mathbf{F}_\bullet)\in(\widetilde{\mathrm{Filt}}\nolimits_f^0 IQ) then F i / F i − 1 = ker θ i / ker θ i − 1 F_i/F_{i-1}=\ker\theta^i/\ker\theta^{i-1} F i / F i − 1 = ker θ i / ker θ i − 1 , which is the pullback of E i \mathcal{E}^i E i under the map q f 0 q_f^0 q f 0 . The claim follows from the short exact sequence
0 → E i + 1 → θ E i → V i → 0. 0\to\mathcal{E}^{i+1}\xrightarrow{\theta}\mathcal{E}^i\to\mathcal{V}^i\to0. 0 → E i + 1 θ E i → V i → 0.
Finally, statement (c) is a property of quasi-smooth pullback. □
For all i i i , the algebra H ∗ ( R e p d i Q , Q ) ≅ H G L d i ∗ H^*(\mathop{\mathrm{Rep}}\nolimits_{d_i}Q,\mathbb{Q})\cong H^*_{GL_{d_i}} H ∗ ( Rep d i Q , Q ) ≅ H G L d i ∗ is a free polynomial algebra in the Chern classes ch k ( V j ) \operatorname{ch}_k(\mathcal{V}^j) ch k ( V j ) of the tautological bundles V j \mathcal{V}^j V j for j ∈ Q 0 j\in Q_0 j ∈ Q 0 and k = 1 , … , d j i k=1,\ldots,d_j^i k = 1 , … , d j i . Thus there is a well-defined algebra morphism
h d : ⨂ i = 1 s H G L d i ∗ ⟶ ⨂ i = 1 s H ∗ ( R e p e i 0 I Q , Q ) , ch k ( V i ) ⟼ ch k ( E i ) − ch k ( E i + 1 ) . h_d:\bigotimes_{i=1}^{s}H^*_{GL_{d_i}}\longrightarrow\bigotimes_{i=1}^{s}H^*(\mathop{\mathrm{Rep}}\nolimits_{e_i}^0 IQ,\mathbb{Q}),\qquad\operatorname{ch}_k(\mathcal{V}^i)\longmapsto\operatorname{ch}_k(\mathcal{E}^i)-\operatorname{ch}_k(\mathcal{E}^{i+1}). h d : i = 1 ⨂ s H G L d i ∗ ⟶ i = 1 ⨂ s H ∗ ( Rep e i 0 I Q , Q ) , ch k ( V i ) ⟼ ch k ( E i ) − ch k ( E i + 1 ) .
Recall the map π d \pi_d π d in (26) . By Lemma 3.8 (b) we have
( p f 0 ) ∗ π d ∗ ( c ) = j ∗ ( q f 0 ) ∗ h d ( c ) . (30) (p_f^0)_* \pi_d^*(c)=j^*(q_f^0)_*h_d(c). \tag*{(30)} ( p f 0 ) ∗ π d ∗ ( c ) = j ∗ ( q f 0 ) ∗ h d ( c ) . (30)
For every d d d and every c ∈ ⨂ i H G L d i ∗ c\in\bigotimes_i H^*_{GL_{d_i}} c ∈ ⨂ i H G L d i ∗ we define
μ d ( c ) = ( p f ) ∗ ( q f ) ! ( h d ( c ) ∩ ⊠ i [ Rep e i 0 I Q ] ) ∈ ⨁ g ^ 0 ( e s ) ∗ ⋯ ∗ g ^ 0 ( e 1 ) . \mu_d(c)=(p_f)_*(q_f)!\left(h_d(c)\cap\boxtimes_i[\operatorname{Rep}_{e_i}^0 IQ]\right)\in\bigoplus\widehat{\mathfrak{g}}^0(e_s)*\cdots*\widehat{\mathfrak{g}}^0(e_1). μ d ( c ) = ( p f ) ∗ ( q f )! ( h d ( c ) ∩ ⊠ i [ Rep e i 0 I Q ] ) ∈ ⨁ g 0 ( e s ) ∗ ⋯ ∗ g 0 ( e 1 ) .
Lemma 3.9.
(a) For every d d d and every c ∈ ⨂ i H G L d i ∗ c\in\bigotimes_i H^*_{GL_{d_i}} c ∈ ⨂ i H G L d i ∗ , the class μ d ( c ) \mu_d(c) μ d ( c ) is supported on Z ≤ d Z_{\leq d} Z ≤ d .
(b) Its restriction to the open stratum J ( d ) ⊂ Z ≤ d J(d)\subset Z_{\leq d} J ( d ) ⊂ Z ≤ d is π d ∗ ( c ) ∩ [ J ( d ) ] \pi_d^*(c)\cap[J(d)] π d ∗ ( c ) ∩ [ J ( d )] .
Proof. By base change and the projection formula, we have
( q f 0 ) ! ( h d ( c ) ∩ ⊠ i [ Rep e i 0 I Q ] ) = ι ∗ ( ( q f 0 ) ! ( h d ( c ) ∩ ⊠ i [ Rep e i 0 I Q ] ) ) . (q_f^0)!\left(h_d(c)\cap\boxtimes_i[\operatorname{Rep}_{e_i}^0 IQ]\right)
=\iota_*\left((q_f^0)!\left(h_d(c)\cap\boxtimes_i[\operatorname{Rep}_{e_i}^0 IQ]\right)\right). ( q f 0 )! ( h d ( c ) ∩ ⊠ i [ Rep e i 0 I Q ] ) = ι ∗ ( ( q f 0 )! ( h d ( c ) ∩ ⊠ i [ Rep e i 0 I Q ] ) ) .
Hence the left hand side is supported on Filt f 0 I Q \operatorname{Filt}_f^0 IQ Filt f 0 I Q . Its pushforward by p f p_f p f is therefore supported on
Im ( p f 0 ) ⊆ Z ≤ d . \operatorname{Im}(p_f^0)\subseteq Z_{\leq d}. Im ( p f 0 ) ⊆ Z ≤ d .
This proves part (a). For (b), note that restriction to the open substack J ( d ) J(d) J ( d ) commutes with proper pushforward via the right Cartesian square in (29) . Using Lemma 3.8 (c) and (30) , we get
j ∗ ( p f ) ∗ ( q f ) ! ( h d ( c ) ∩ ⊠ i [ Rep e i 0 I Q ] ) = ( p f 00 ) ∗ ( j ∗ ( q f 0 ) ∗ h d ( c ) ∩ [ Filt f 0 I Q ] ) = ( p f 00 ) ∗ ( ( p f 00 ) ∗ ( π d ) ∗ ( c ) ∩ [ Filt f 0 I Q ] ) = π d ∗ ( c ) ∩ [ J ( d ) ] . \begin{aligned}
j^*(p_f)_*(q_f)!\left(h_d(c)\cap\boxtimes_i[\operatorname{Rep}_{e_i}^0 IQ]\right)
=(p_f^{00})_*\left(j^*(q_f^0)_*h_d(c)\cap[\operatorname{Filt}_f^0 IQ]\right) \\
&=(p_f^{00})_*\left((p_f^{00})_*(\pi_d)^*(c)\cap[\operatorname{Filt}_f^0 IQ]\right) \\
&=\pi_d^*(c)\cap[J(d)].
\end{aligned} j ∗ ( p f ) ∗ ( q f )! ( h d ( c ) ∩ ⊠ i [ Rep e i 0 I Q ] ) = ( p f 00 ) ∗ ( j ∗ ( q f 0 ) ∗ h d ( c ) ∩ [ Filt f 0 I Q ] ) = ( p f 00 ) ∗ ( ( p f 00 ) ∗ ( π d ) ∗ ( c ) ∩ [ Filt f 0 I Q ] ) = π d ∗ ( c ) ∩ [ J ( d )] .
Finally, the class h d ( c ) h_d(c) h d ( c ) being a sum of pure tensors ⨂ i a i \bigotimes_i a_i ⨂ i a i with
a i ∈ H ∗ ( Rep e i 0 I Q , Q ) = H G L e i ∗ , a_i\in H^*(\operatorname{Rep}_{e_i}^0 IQ,\mathbb{Q})=H^*_{GL_{e_i}}, a i ∈ H ∗ ( Rep e i 0 I Q , Q ) = H G L e i ∗ ,
the class μ d ( c ) \mu_d(c) μ d ( c ) is a sum of ordered products of elements of g ^ 0 \widehat{\mathfrak{g}}^0 g 0 of the form
( a s ∩ [ Rep e s 0 I Q ] ) ∗ ⋯ ∗ ( a 1 ∩ [ Rep e 1 0 I Q ] ) . (a_s\cap[\operatorname{Rep}_{e_s}^0 IQ])*\cdots*(a_1\cap[\operatorname{Rep}_{e_1}^0 IQ]). ( a s ∩ [ Rep e s 0 I Q ]) ∗ ⋯ ∗ ( a 1 ∩ [ Rep e 1 0 I Q ]) .
Proof of Proposition 3.4. We are now ready to finish the proof of Markman’s theorem. Fix a total order ⪯ \preceq ⪯ on the set of Jordan types of dimension d d d refining ≤ \leq ≤ . Put Z ⪯ d = ⋃ e ⪯ d J ( e ) Z_{\preceq d}=\bigcup_{e\preceq d}J(e) Z ⪯ d = ⋃ e ⪯ d J ( e ) . These are closed substacks by the proof of Lemma 3.6. By Corollary 3.7, the pushforwards by the inclusion
Z ⪯ d ⟶ Rep d n i l I Q Z_{\preceq d}\longrightarrow\operatorname{Rep}^{\mathrm{nil}}_d IQ Z ⪯ d ⟶ Rep d nil I Q
is injective. This yields a filtration
H ∗ ( Z ⪯ d , Q ) ⊂ A I Q n i l ( d ) H_*(Z_{\preceq d},\mathbb{Q})\subset A^{\mathrm{nil}}_{IQ}(d) H ∗ ( Z ⪯ d , Q ) ⊂ A I Q nil ( d )
whose associated graded is given by
H ∗ ( Z ⪯ d , Q ) / H ∗ ( Z ≺ d , Q ) ≅ H ∗ ( J ( d ) , Q ) . H_*(Z_{\preceq d},\mathbb{Q})/H_*(Z_{\prec d},\mathbb{Q})\cong H_*(J(d),\mathbb{Q}). H ∗ ( Z ⪯ d , Q ) / H ∗ ( Z ≺ d , Q ) ≅ H ∗ ( J ( d ) , Q ) .
By Lemma 3.9, we have
μ d ( c ) ∈ ( π d ( c ) ∩ [ J ( d ) ] ) ⊕ H ∗ ( Z ≺ d , Q ) , ∀ c ∈ ⨂ i H G L d i ∗ . \mu_d(c)\in\left(\pi_d(c)\cap[J(d)]\right)\oplus H_*(Z_{\prec d},\mathbb{Q}),\qquad\forall c\in\bigotimes_i H^*_{GL_{d_i}}. μ d ( c ) ∈ ( π d ( c ) ∩ [ J ( d )] ) ⊕ H ∗ ( Z ≺ d , Q ) , ∀ c ∈ i ⨂ H G L d i ∗ .
By induction, this implies that the classes μ d ( c ) \mu_d(c) μ d ( c ) for all d d d and c c c running over a basis of ⨂ i H G L d i ∗ \bigotimes_i H^*_{GL_{d_i}} ⨂ i H G L d i ∗ form a basis of H ∗ ( Rep d n i l I Q , Q ) H_*(\operatorname{Rep}^{\mathrm{nil}}_d IQ,\mathbb{Q}) H ∗ ( Rep d nil I Q , Q ) . We deduce that for every d d d we have
A I Q n i l ( d ) = ∑ ∣ d ∣ = d ∑ g ^ 0 ( e s ) ∗ ⋯ ∗ g ^ 0 ( e 1 ) , A^{\mathrm{nil}}_{IQ}(d)=\sum_{|d|=d}\sum\widehat{\mathfrak{g}}^0(e_s)*\cdots*\widehat{\mathfrak{g}}^0(e_1), A I Q nil ( d ) = ∣ d ∣ = d ∑ ∑ g 0 ( e s ) ∗ ⋯ ∗ g 0 ( e 1 ) ,
where the inner sum runs over all Jordan types of weight d d d and e e e is the sequence associated with d d d . This proves Proposition 3.4 and Theorem 3.1.
Character varieties and Higgs moduli.
In this section we fix the notation and briefly collect known results on the (co)homology of twisted character varieties and the (co)homology of moduli stacks of Higgs bundles on smooth projective curves. In particular, we recall the C = W C=W C = W and P = C \mathrm{P}=\mathrm{C} P = C theorems as well as the curious Hard Lefschetz property. We refer to [14 ] , [40 ] , [31 ] , [27 ] and [13 ] for details. Throughout, we fix an integer g > 1 g>1 g > 1 , a compact topological surface Σ g \Sigma_g Σ g of genus g g g , and a smooth complex projective curve C C C of genus g g g .
Character varieties.
Let us fix a pair of coprime integers ( r , d ) (r,d) ( r , d ) with r > 0 r>0 r > 0 and define the genus g g g twisted character variety M B ( r , d ) M^B(r,d) M B ( r , d ) by (1) . It is a smooth affine algebraic variety of dimension 2 N 2N 2 N . Its cohomology H ∗ ( M B ( r , d ) , Q ) H^*(M^B(r,d),\mathbb{Q}) H ∗ ( M B ( r , d ) , Q ) carries a mixed Hodge structure which is of Hodge-Tate type. The weight filtration is
W ℓ H m ( M B ( r , d ) , Q ) , ℓ ∈ [ m , 2 m ] . W_\ell H^m(M^B(r,d),\mathbb{Q}),\qquad\ell\in[m,2m]. W ℓ H m ( M B ( r , d ) , Q ) , ℓ ∈ [ m , 2 m ] .
We next describe tautological generators for the cohomology ring H ∗ ( M B ( r , d ) , Q ) H^*(M^B(r,d),\mathbb{Q}) H ∗ ( M B ( r , d ) , Q ) . The quotient M ~ B ( r , d ) \widetilde{M}^B(r,d) M B ( r , d ) of M B ( r , d ) M^B(r,d) M B ( r , d ) by the action of ( G m ) 2 g (\mathbb{G}_m)^{2g} ( G m ) 2 g given by rescaling the generators ( A i , B i ) i = 1 , … , g (A_i,B_i)_{i=1,\ldots,g} ( A i , B i ) i = 1 , … , g is a connected component of the P G L r ( C ) PGL_r(\mathbb{C}) P G L r ( C ) (untwisted) character variety. There is a ring isomorphism
H ∗ ( M B ( r , d ) , Q ) ≅ H ∗ ( M ~ B ( r , d ) ) ⊗ H ∗ ( ( C ∗ ) 2 g , Q ) . (31) H^*(M^B(r,d),\mathbb{Q})\cong H^*(\widetilde{M}^B(r,d))\otimes H^*((\mathbb{C}^*)^{2g},\mathbb{Q}). \tag*{(31)} H ∗ ( M B ( r , d ) , Q ) ≅ H ∗ ( M B ( r , d )) ⊗ H ∗ (( C ∗ ) 2 g , Q ) . (31)
There is a tautological (non-algebraic) P G L r ( C ) PGL_r(\mathbb{C}) P G L r ( C ) -principal bundle V ~ \widetilde{\mathcal{V}} V on M ~ B ( r , d ) × Σ g \widetilde{M}^B(r,d)\times\Sigma_g M B ( r , d ) × Σ g yielding tautological classes
c ‾ k ( η ) = ∫ Σ g c ‾ k ( V ~ ) ∪ p 2 ∗ ( η ) ∈ H 2 k − 2 + ∣ η ∣ ( M ~ B ( r , d ) , Q ) , ∀ k ≥ 2 , η ∈ H ∗ ( Σ g , Q ) \overline{c}_k(\eta)=\int_{\Sigma_g}\overline{c}_k(\widetilde{\mathcal{V}})\cup p_2^*(\eta)\in H^{2k-2+|\eta|}(\widetilde{M}^B(r,d),\mathbb{Q}),\qquad\forall k\geq2,\ \eta\in H^*(\Sigma_g,\mathbb{Q}) c k ( η ) = ∫ Σ g c k ( V ) ∪ p 2 ∗ ( η ) ∈ H 2 k − 2 + ∣ η ∣ ( M B ( r , d ) , Q ) , ∀ k ≥ 2 , η ∈ H ∗ ( Σ g , Q )
where p 2 : M ~ B ( r , d ) × Σ g → Σ g p_2:\widetilde{M}^B(r,d)\times\Sigma_g\to\Sigma_g p 2 : M B ( r , d ) × Σ g → Σ g is the obvious projection and c ‾ k \overline{c}_k c k is given, in terms of Chern roots { ζ i } i \{\zeta_i\}_i { ζ i } i by the k k k th elementary symmetric function of { ζ i − ζ ‾ } i \{\zeta_i-\overline{\zeta}\}_i { ζ i − ζ } i where ζ ‾ \overline{\zeta} ζ is the average of the ζ i \zeta_i ζ i . These classes are denoted
{ α k , β k , φ i , k } , i = 1 , … , 2 g , k = 2 , … , r \{\alpha_k,\beta_k,\varphi_{i,k}\},\qquad i=1,\ldots,2g,\quad k=2,\ldots,r { α k , β k , φ i , k } , i = 1 , … , 2 g , k = 2 , … , r
and are defined as
α k = c ‾ k ( 1 ) ∈ H 2 k − 2 ( M ~ B ( r , d ) , Q ) , β k = c ‾ k ( [ Σ g ] ) ∈ H 2 k ( M ~ B ( r , d ) , Q ) , \alpha_k=\overline{c}_k(1)\in H^{2k-2}(\widetilde{M}^B(r,d),\mathbb{Q}),\qquad\beta_k=\overline{c}_k([\Sigma_g])\in H^{2k}(\widetilde{M}^B(r,d),\mathbb{Q}), α k = c k ( 1 ) ∈ H 2 k − 2 ( M B ( r , d ) , Q ) , β k = c k ([ Σ g ]) ∈ H 2 k ( M B ( r , d ) , Q ) ,
φ i , k = c ‾ k ( γ i ) ∈ H 2 k − 1 ( M ~ B ( r , d ) , Q ) \varphi_{i,k}=\overline{c}_k(\gamma_i)\in H^{2k-1}(\widetilde{M}^B(r,d),\mathbb{Q}) φ i , k = c k ( γ i ) ∈ H 2 k − 1 ( M B ( r , d ) , Q )
where { γ 1 , … , γ 2 g } \{\gamma_1,\ldots,\gamma_{2g}\} { γ 1 , … , γ 2 g } form a basis of H 1 ( Σ g , Q ) H^1(\Sigma_g,\mathbb{Q}) H 1 ( Σ g , Q ) . By Markman’s theorem [26 Cor. 2] , the classes { α k , β k , φ i , k } \{\alpha_k,\beta_k,\varphi_{i,k}\} { α k , β k , φ i , k } generate the cohomology ring H ∗ ( M ~ B ( r , d ) , Q ) H^*(\widetilde{M}^B(r,d),\mathbb{Q}) H ∗ ( M B ( r , d ) , Q ) , yielding in turn a collection of generators for H ∗ ( M B ( r , d ) , Q ) H^*(M^B(r,d),\mathbb{Q}) H ∗ ( M B ( r , d ) , Q ) by (31) . As shown by Hausel and Rodriguez-Villegas, the pure part of H ∗ ( M B ( r , d ) , Q ) H^*(M^B(r,d),\mathbb{Q}) H ∗ ( M B ( r , d ) , Q ) is the subring generated by { β k } \{\beta_k\} { β k } , see [14 Cor.4.1.10] ,
H p u r e ∗ ( M B ( r , d ) , Q ) = ⨁ m W m H m ( M B ( r , d ) , Q ) = Q ⟨ β 2 , β 3 , … , β r ⟩ . (32) H^*_{\mathrm{pure}}(M^B(r,d),\mathbb{Q})=\bigoplus_m W_mH^m(M^B(r,d),\mathbb{Q})=\mathbb{Q}\langle\beta_2,\beta_3,\ldots,\beta_r\rangle. \tag*{(32)} H pure ∗ ( M B ( r , d ) , Q ) = m ⨁ W m H m ( M B ( r , d ) , Q ) = Q ⟨ β 2 , β 3 , … , β r ⟩ . (32)
More generally, the place in the weight filtration of each of the above generators is determined by Shende [40 ] who shows that the Chern degree coincides with the weight (C = W C=W C = W ):
c ‾ k ( η ) ∈ W 2 k − 2 + ∣ η ∣ H 2 k − 2 + ∣ η ∣ ( M ~ B ( r , d ) , Q ) , k ≥ 2 , η ∈ H ∗ ( Σ g , Q ) , (33) \overline{c}_k(\eta)\in W_{2k-2+|\eta|}H^{2k-2+|\eta|}(\widetilde{M}^B(r,d),\mathbb{Q}),\qquad k\geq2,\ \eta\in H^*(\Sigma_g,\mathbb{Q}), \tag*{(33)} c k ( η ) ∈ W 2 k − 2 + ∣ η ∣ H 2 k − 2 + ∣ η ∣ ( M B ( r , d ) , Q ) , k ≥ 2 , η ∈ H ∗ ( Σ g , Q ) , (33)
and the degree 1 classes generating H ∗ ( ( G m ) 2 g , Q ) H^*((\mathbb{G}_m)^{2g},\mathbb{Q}) H ∗ (( G m ) 2 g , Q ) all have weight 2.
Moduli spaces of Higgs bundles
Fix a pair ( r , d ) (r,d) ( r , d ) of coprime integers with r > 0 r>0 r > 0 . Let M D ( r , d ) M^D(r,d) M D ( r , d ) be the moduli space of semistable Higgs bundles
( E , φ ) , φ ∈ Hom O C ( E , E ⊗ Ω C ) (\mathcal{E},\varphi),\quad\varphi\in\operatorname{Hom}_{\mathcal{O}_C}(\mathcal{E},\mathcal{E}\otimes\Omega_C) ( E , φ ) , φ ∈ Hom O C ( E , E ⊗ Ω C )
of rank r r r and degree d d d on C C C . Thanks to the coprimality assumption on ( r , d ) (r,d) ( r , d ) , semistable Higgs bundles of rank r r r and degree d d d are automatically stable, and M D ( r , d ) M^D(r,d) M D ( r , d ) is a smooth quasi-projective algebraic variety, equipped with a natural symplectic structure. It is of dimension 2 N 2N 2 N with
N = ( g − 1 ) r 2 + 1. N=(g-1)r^2+1. N = ( g − 1 ) r 2 + 1.
Scaling the Higgs field φ \varphi φ defines a G m \mathbb{G}_m G m -action, which endows M D ( r , d ) M^D(r,d) M D ( r , d ) with the structure of a semi-projective variety, see [15 ] . In particular, its cohomology carries a pure mixed Hodge structure. Let
h : M D ( r , d ) ⟶ A r = ⨁ k = 1 r H 0 ( C , Ω C ⊗ k ) ≅ A N . h:M^D(r,d)\longrightarrow A_r=\bigoplus_{k=1}^{r}H^0(C,\Omega_C^{\otimes k})\cong\mathbb{A}^N. h : M D ( r , d ) ⟶ A r = k = 1 ⨁ r H 0 ( C , Ω C ⊗ k ) ≅ A N .
denote the Hitchin map. It is a Lagrangian fibration. In particular, it is proper. The Hitchin map induces a perverse filtration on H ∗ ( M D ( r , d ) , Q ) H^*(M^D(r,d),\mathbb{Q}) H ∗ ( M D ( r , d ) , Q ) such that
P k H ∗ ( M D ( r , d ) , Q ) = H ∗ ( A r , p τ ≤ k + N h ∗ ( Q M D ( r , d ) ) ) . (34) P_k H^*(M^D(r,d),\mathbb{Q})=H^*(A_r,{}^p\tau_{\leq k+N}h_*(\mathbb{Q}_{M^D(r,d)})). \tag*{(34)} P k H ∗ ( M D ( r , d ) , Q ) = H ∗ ( A r , p τ ≤ k + N h ∗ ( Q M D ( r , d ) )) . (34)
It is normalized to be nontrivial in the interval [ 0 , 2 N ] [0,2N] [ 0 , 2 N ] .
As for the character varieties, the cohomology of M D ( r , d ) M^D(r,d) M D ( r , d ) is tautologically generated. Consider the action of T ∗ Pic 0 ( C ) T^*\operatorname{Pic}^0(C) T ∗ Pic 0 ( C ) on M D ( r , d ) M^D(r,d) M D ( r , d ) given by
( L , θ ) ⋅ ( E , φ ) = ( L ⊗ E , id ⊗ φ + θ ⊗ id ) . (\mathcal{L},\theta)\cdot(\mathcal{E},\varphi)=(\mathcal{L}\otimes\mathcal{E},\operatorname{id}\otimes\varphi+\theta\otimes\operatorname{id}). ( L , θ ) ⋅ ( E , φ ) = ( L ⊗ E , id ⊗ φ + θ ⊗ id ) .
The quotient of M D ( r , d ) M^D(r,d) M D ( r , d ) by this action is a connected component M ~ D ( r , d ) \widetilde{M}^D(r,d) M D ( r , d ) of the moduli space of stable P G L r PGL_r P G L r -Higgs bundles on C C C , and there is a canonical algebra isomorphism
H ∗ ( M D ( r , d ) , Q ) ≅ H ∗ ( M ~ D ( r , d ) , Q ) ⊗ H ∗ ( T ∗ Pic 0 ( C ) , Q ) . (35) H^*(M^D(r,d),\mathbb{Q})\cong H^*(\widetilde{M}^D(r,d),\mathbb{Q})\otimes H^*(T^*\operatorname{Pic}^0(C),\mathbb{Q}). \tag*{(35)} H ∗ ( M D ( r , d ) , Q ) ≅ H ∗ ( M D ( r , d ) , Q ) ⊗ H ∗ ( T ∗ Pic 0 ( C ) , Q ) . (35)
The scheme M ~ D ( r , d ) × C \widetilde{M}^D(r,d)\times C M D ( r , d ) × C carries a tautological algebraic P G L r ( C ) PGL_r(\mathbb{C}) P G L r ( C ) -principal bundle V ~ \widetilde{\mathcal{V}} V . We define the tautological classes to be
c ˉ k ( η ) = ∫ C c k ( V ~ ) ∪ p 2 ∗ ( η ) ∈ H 2 k − 2 + ∣ η ∣ ( M ~ D ( r , d ) , Q ) , ∀ k ≥ 2 , η ∈ H ∗ ( C , Q ) \bar{c}_k(\eta)=\int_C c_k(\widetilde{\mathcal{V}})\cup p_2^*(\eta)\in H^{2k-2+|\eta|}(\widetilde{M}^D(r,d),\mathbb{Q}),\quad\forall k\geq2,\eta\in H^*(C,\mathbb{Q}) c ˉ k ( η ) = ∫ C c k ( V ) ∪ p 2 ∗ ( η ) ∈ H 2 k − 2 + ∣ η ∣ ( M D ( r , d ) , Q ) , ∀ k ≥ 2 , η ∈ H ∗ ( C , Q )
where p 2 : M ~ D ( r , d ) × C → C p_2:\widetilde{M}^D(r,d)\times C\to C p 2 : M D ( r , d ) × C → C is the projection. By Markman’s theorem again, the cohomology algebra H ∗ ( M ~ D ( r , d ) , Q ) H^*(\widetilde{M}^D(r,d),\mathbb{Q}) H ∗ ( M D ( r , d ) , Q ) is generated by the tautological classes c ˉ k ( η ) \bar{c}_k(\eta) c ˉ k ( η ) for k = 2 , … , r k=2,\ldots,r k = 2 , … , r and η ∈ H ∗ ( C , Q ) \eta\in H^*(C,\mathbb{Q}) η ∈ H ∗ ( C , Q ) . Moreover, thanks to [27 ] , [13 ] , their perverse degree is equal to their Chern degree, i.e., we have
c ˉ k ( η ) ∈ P k H 2 k − 2 + ∣ η ∣ ( M ~ D ( r , d ) , Q ) , ∀ k ≥ 2 , η ∈ H ∗ ( C , Q ) . (36) \bar{c}_k(\eta)\in P_kH^{2k-2+|\eta|}(\widetilde{M}^D(r,d),\mathbb{Q}),\quad\forall k\geq2,\eta\in H^*(C,\mathbb{Q}). \tag*{(36)} c ˉ k ( η ) ∈ P k H 2 k − 2 + ∣ η ∣ ( M D ( r , d ) , Q ) , ∀ k ≥ 2 , η ∈ H ∗ ( C , Q ) . (36)
The perverse degrees of the degree 1 classes generating H ∗ ( T ∗ Pic 0 ( C ) , Q ) H^*(T^*\operatorname{Pic}^0(C),\mathbb{Q}) H ∗ ( T ∗ Pic 0 ( C ) , Q ) are all equal to 1. The perverse filtration on H ∗ ( M D ( r , d ) , Q ) H^*(M^D(r,d),\mathbb{Q}) H ∗ ( M D ( r , d ) , Q ) is multiplicative ( P = C ) (P=C) ( P = C ) . In particular, we have
⨁ k P k H 2 k ( M D ( r , d ) , Q ) = Q ⟨ c ˉ 2 ( [ p t ] ) , c ˉ 3 ( [ p t ] ) , … , c ˉ r ( [ p t ] ) ⟩ . (37) \bigoplus_k P_kH^{2k}(M^D(r,d),\mathbb{Q})=\mathbb{Q}\langle\bar{c}_2([{\rm pt}]),\bar{c}_3([{\rm pt}]),\ldots,\bar{c}_r([{\rm pt}])\rangle. \tag*{(37)} k ⨁ P k H 2 k ( M D ( r , d ) , Q ) = Q ⟨ c ˉ 2 ([ pt ]) , c ˉ 3 ([ pt ]) , … , c ˉ r ([ pt ])⟩ . (37)
Non-abelian Hodge correspondence and P=W
Fix an isomorphism of topological spaces Σ g ≅ C \Sigma_g \cong C Σ g ≅ C . This identifies H ∗ ( Σ g , Q ) H^*(\Sigma_g,\mathbb{Q}) H ∗ ( Σ g , Q ) with H ∗ ( C , Q ) H^*(C,\mathbb{Q}) H ∗ ( C , Q ) . The non-abelian Hodge correspondence of Simpson and Corlette provides, for any coprime pair ( r , d ) (r,d) ( r , d ) , a diffeomorphism
Θ : M B ( r , d ) → ∼ M D ( r , d ) \Theta: M^B(r,d) \xrightarrow{\sim} M^D(r,d) Θ : M B ( r , d ) ∼ M D ( r , d )
which is compatible with tautological classes in the following sense: the associated ring isomorphism
Θ ∗ : H ∗ ( M B ( r , d ) , Q ) → ∼ H ∗ ( M D ( r , d ) , Q ) \Theta_* : H^*(M^B(r,d),\mathbb{Q}) \xrightarrow{\sim} H^*(M^D(r,d),\mathbb{Q}) Θ ∗ : H ∗ ( M B ( r , d ) , Q ) ∼ H ∗ ( M D ( r , d ) , Q )
maps c ‾ k ( η ) \overline{c}_k(\eta) c k ( η ) to c ‾ k ( η ) \overline{c}_k(\eta) c k ( η ) for all k ≥ 2 k \ge2 k ≥ 2 , η ∈ H ∗ ( C , Q ) \eta\in H^*(C,\mathbb{Q}) η ∈ H ∗ ( C , Q ) and restricts to the isomorphism
H ∗ ( G m , Q ) 2 g ≅ H ∗ ( T ∗ Pic 0 ( C ) , Q ) H^*(\mathbb{G}_m,\mathbb{Q})^{2g} \cong H^*(T^*\operatorname{Pic}^0(C),\mathbb{Q}) H ∗ ( G m , Q ) 2 g ≅ H ∗ ( T ∗ Pic 0 ( C ) , Q )
induced by the homeomorphism Σ g ≅ C \Sigma_g \cong C Σ g ≅ C . The P=W identity is the coincidence between (33) and (36) .
Curious Hard Lefschetz
There is an extra symmetry on H ∗ ( M D ( r , d ) , Q ) H^*(M^D(r,d),\mathbb{Q}) H ∗ ( M D ( r , d ) , Q ) , provided on the Higgs moduli side by the relative Hard Lefschetz isomorphism. For α ∈ H 2 ( M D ( r , d ) , Q ) \alpha\in H^2(M^D(r,d),\mathbb{Q}) α ∈ H 2 ( M D ( r , d ) , Q ) a class which is relatively ample with respect to h h h , capping with α \alpha α yields isomorphisms
α j : gr N − j P H m ( M r , d D , Q ) → ∼ gr N + j P H m + 2 j ( M r , d D , Q ) \alpha^j : \operatorname{gr}^{P}_{N-j} H^m(M^D_{r,d},\mathbb{Q}) \xrightarrow{\sim} \operatorname{gr}^{P}_{N+j} H^{m+2j}(M^D_{r,d},\mathbb{Q}) α j : gr N − j P H m ( M r , d D , Q ) ∼ gr N + j P H m + 2 j ( M r , d D , Q )
for any j , m j,m j , m . Taking j = N − k j=N-k j = N − k and m = 2 k m=2k m = 2 k , and using the fact that
gr k P H 2 k ( M r , d D , Q ) = P k H 2 k ( M r , d D , Q ) , \operatorname{gr}^{P}_{k} H^{2k}(M^D_{r,d},\mathbb{Q}) = P_k H^{2k}(M^D_{r,d},\mathbb{Q}), gr k P H 2 k ( M r , d D , Q ) = P k H 2 k ( M r , d D , Q ) ,
e.g., by P=W and the bound W m H ℓ = 0 W_mH^\ell=0 W m H ℓ = 0 for m ∉ [ ℓ , 2 ℓ ] m\notin[\ell,2\ell] m ∈ / [ ℓ , 2 ℓ ] , this gives
P k H 2 k ( M D ( r , d ) , Q ) ≅ gr 2 N − k P H 2 N ( M D ( r , d ) , Q ) . (38) P_k H^{2k}(M^D(r,d),\mathbb{Q}) \cong\operatorname{gr}^{P}_{2N-k} H^{2N}(M^D(r,d),\mathbb{Q}). \tag*{(38)} P k H 2 k ( M D ( r , d ) , Q ) ≅ gr 2 N − k P H 2 N ( M D ( r , d ) , Q ) . (38)
Since M D ( r , d ) M^D(r,d) M D ( r , d ) retracts to the global nilpotent cone, we have
j > 2 N ⇒ H j ( M D ( r , d ) , Q ) = 0. j>2N \Rightarrow H^j(M^D(r,d),\mathbb{Q})=0. j > 2 N ⇒ H j ( M D ( r , d ) , Q ) = 0.
Hence, the relative Hard Lefschetz isomorphism implies that
P N − 1 H 2 N ( M D ( r , d ) , Q ) = 0. P_{N-1}H^{2N}(M^D(r,d),\mathbb{Q})=0. P N − 1 H 2 N ( M D ( r , d ) , Q ) = 0.
In other words, (38) relates the perverse pure part of H ∗ ( M D ( r , d ) , Q ) H^*(M^D(r,d),\mathbb{Q}) H ∗ ( M D ( r , d ) , Q ) with the perverse filtration of the middle cohomology of M D ( r , d ) M^D(r,d) M D ( r , d ) . Thus, using [37 Cor. 1.7] , [29 ] and [14 Rem. 4.4.6] , we deduce that
dim ( ⨁ k P k H 2 k ( M D ( r , d ) , Q ) ) = dim H 2 N ( M D ( r , d ) , Q ) = A g , r ( 0 ) = A S g , r ( 1 ) , (39) \dim\left(\bigoplus_k P_kH^{2k}(M^D(r,d),\mathbb{Q})\right)=\dim H^{2N}(M^D(r,d),\mathbb{Q})=A_{g,r}(0)=A_{S_g,r}(1), \tag*{(39)} dim ( k ⨁ P k H 2 k ( M D ( r , d ) , Q ) ) = dim H 2 N ( M D ( r , d ) , Q ) = A g , r ( 0 ) = A S g , r ( 1 ) , (39)
where A g , r ( σ ) A_{g,r}(\sigma) A g , r ( σ ) is the Kac polynomial counting indecomposable vector bundles of rank r r r . Note that this dimension is also the number of irreducible components of the semistable global nilpotent cone in M D ( r , d ) M^D(r,d) M D ( r , d ) , see [37 Cor.1.9] .
The Higgs COHAs and the BPS Lie algebra g C \mathfrak{g}^C g C
In this section, we drop the coprimality assumption on ( r , d ) (r,d) ( r , d ) and introduce the BPS Lie algebra of the stack of semistable Higgs sheaves on C C C . We recall the χ \chi χ -independence theorem of Kinjo-Koseki as well the stacky P=C conjecture of Picovschi. We refer to [23 ] , [12 ] and [35 ] for details.
The Higgs COHA
The slope of a coherent sheaf F \mathcal{F} F of class α = ( r , d ) \alpha=(r,d) α = ( r , d ) , i.e., of rank r r r and degree d d d , is
μ ( F ) = μ ( α ) = d r . \mu(\mathcal{F})=\mu(\alpha)=\frac{d}{r}. μ ( F ) = μ ( α ) = r d .
For α , α ′ ∈ Z 2 \alpha,\alpha'\in\mathbb{Z}^{2} α , α ′ ∈ Z 2 we write
( α , α ′ ) = 2 ( 1 − g ) rank ( α ) rank ( α ′ ) . (\alpha,\alpha')=2(1-g)\operatorname{rank}(\alpha)\operatorname{rank}(\alpha'). ( α , α ′ ) = 2 ( 1 − g ) rank ( α ) rank ( α ′ ) .
Note that, this coincides with the Euler form (4) for the S g S_g S g quiver, which is the only quiver that we will consider in the rest of the paper. Let M s s ( α ) \mathcal{M}^{\mathrm{ss}}(\alpha) M ss ( α ) be the stack of semistable Higgs bundles on C C C of class α \alpha α . It is a quasi-smooth, quasi-compact derived stack of virtual dimension
vdim M s s ( α ) = − ( α , α ) . \operatorname{vdim}\mathcal{M}^{\mathrm{ss}}(\alpha)=-(\alpha,\alpha). vdim M ss ( α ) = − ( α , α ) .
Fix a slope ν ∈ Q \nu\in\mathbb{Q} ν ∈ Q . The COHA of semistable Higgs bundles of slope ν \nu ν is
A C , l ( α ) = H l − ( α , α ) ( M s s ( α ) , Q ) , A C ( α ) = ⨁ l A C , l ( α ) , A C ( ν ) = ⨁ μ ( α ) = ν A C ( α ) . A_{C,l}(\alpha)=H_{l-(\alpha,\alpha)}\left(\mathcal{M}^{\mathrm{ss}}(\alpha),\mathbb{Q}\right),\qquad A_C(\alpha)=\bigoplus_l A_{C,l}(\alpha),\qquad A_C^{(\nu)}=\bigoplus_{\mu(\alpha)=\nu}A_C(\alpha). A C , l ( α ) = H l − ( α , α ) ( M ss ( α ) , Q ) , A C ( α ) = l ⨁ A C , l ( α ) , A C ( ν ) = μ ( α ) = ν ⨁ A C ( α ) .
There is an induction diagram in all ways similar to (7) giving rise to a N × Z \mathbb{N}\times\mathbb{Z} N × Z -graded associative algebra structure on A C ( ν ) A_C^{(\nu)} A C ( ν ) .
The BPS Lie algebra g C \mathfrak{g}^C g C
When α = ( r , d ) \alpha=(r,d) α = ( r , d ) is primitive, i.e., for coprime ( r , d ) (r,d) ( r , d ) , the classical truncation M s s , c l ( α ) \mathcal{M}^{\mathrm{ss,cl}}(\alpha) M ss , cl ( α ) is a G m \mathbb{G}_m G m -gerbe over the moduli space M D ( r , d ) M^D(r,d) M D ( r , d ) . In general there is a good moduli space
p : M s s , r e d ( r , d ) ⟶ M D ( r , d ) . p:\mathcal{M}^{\mathrm{ss,red}}(r,d)\longrightarrow M^D(r,d). p : M ss , red ( r , d ) ⟶ M D ( r , d ) .
For μ ( α ) = μ ( α ′ ) \mu(\alpha)=\mu(\alpha') μ ( α ) = μ ( α ′ ) there is an addition map
⊕ α , α ′ : M D ( α ) × M D ( α ′ ) ⟶ M D ( α + α ′ ) . \oplus_{\alpha,\alpha'}:M^D(\alpha)\times M^D(\alpha')\longrightarrow M^D(\alpha+\alpha'). ⊕ α , α ′ : M D ( α ) × M D ( α ′ ) ⟶ M D ( α + α ′ ) .
Let us fix a slope ν = μ ( α 0 ) \nu=\mu(\alpha_0) ν = μ ( α 0 ) with α 0 = ( r 0 , d 0 ) \alpha_0=(r_0,d_0) α 0 = ( r 0 , d 0 ) primitive. There is a collection of pure mixed Hodge modules B P S n α 0 C BPS^C_{n\alpha_0} B P S n α 0 C on M D ( n α 0 ) M^D(n\alpha_0) M D ( n α 0 ) for n ≥ 1 n\geq1 n ≥ 1 such that
Sym ⊕ ( ⨁ n ≥ 1 B P S n α 0 C ⊗ H G m ∗ ) ≅ ⨁ n ≥ 1 p ∗ ( D Q M s s ( n α 0 ) ⊗ L − n 2 2 ( α 0 , α 0 ) ) . (40) \operatorname{Sym}_{\oplus}\left(\bigoplus_{n\geq1}BPS^C_{n\alpha_0}\otimes H^*_{\mathbb{G}_m}\right)\cong\bigoplus_{n\geq1}p_*\left(\mathbb{D}\mathbb{Q}_{\mathcal{M}^{\mathrm{ss}}(n\alpha_0)}\otimes\mathbb{L}^{-\frac{n^2}{2}(\alpha_0,\alpha_0)}\right). \tag*{(40)} Sym ⊕ ( n ≥ 1 ⨁ B P S n α 0 C ⊗ H G m ∗ ) ≅ n ≥ 1 ⨁ p ∗ ( D Q M ss ( n α 0 ) ⊗ L − 2 n 2 ( α 0 , α 0 ) ) . (40)
The COHA of semistable Higgs bundles enjoys the same structural properties as its quiver cousin, and, in fact, as the COHA of any 2 C Y 2CY 2 C Y category in the sense of [11 ] . Namely, we define the BPS Lie algebras and affinized BPS Lie algebras as
g C , ( ν ) = ⨁ n ≥ 1 g C ( n α 0 ) = ⨁ n ≥ 1 H − ∗ ( M D ( n α 0 ) , B P S n α 0 C ) , \mathfrak{g}^{C,(\nu)}=\bigoplus_{n\geq1}\mathfrak{g}^C(n\alpha_0)=\bigoplus_{n\geq1}\mathbb{H}^{-*}\left(M^D(n\alpha_0),BPS^C_{n\alpha_0}\right), g C , ( ν ) = n ≥ 1 ⨁ g C ( n α 0 ) = n ≥ 1 ⨁ H −∗ ( M D ( n α 0 ) , B P S n α 0 C ) ,
g ^ C , ( ν ) = ⨁ n ≥ 1 g ^ C ( n α 0 ) = ⨁ n ≥ 1 H − ∗ ( M D ( n α 0 ) , B P S n α 0 C ⊗ H G m ∗ ) . \widehat{\mathfrak{g}}^{C,(\nu)}=\bigoplus_{n\geq1}\widehat{\mathfrak{g}}^C(n\alpha_0)=\bigoplus_{n\geq1}\mathbb{H}^{-*}\left(M^D(n\alpha_0),BPS^C_{n\alpha_0}\otimes H^*_{\mathbb{G}_m}\right). g C , ( ν ) = n ≥ 1 ⨁ g C ( n α 0 ) = n ≥ 1 ⨁ H −∗ ( M D ( n α 0 ) , B P S n α 0 C ⊗ H G m ∗ ) .
They are graded subspaces in A C ( ν ) A_C^{(\nu)} A C ( ν ) , where the generator u ∈ H G m 2 u\in H^2_{\mathbb{G}_m} u ∈ H G m 2 is in degree − 2 -2 − 2 , and both form Lie subalgebras under the COHA multiplication. Note that we only consider the semistable BPS Lie algebras, and not the global BPS Lie algebra considered in [12 ] . Morever, we have
U ( g ^ C , ( ν ) ) = A C ( ν ) . (41) U\left(\widehat{\mathfrak{g}}^{C,(\nu)}\right)=A_C^{(\nu)}. \tag*{(41)} U ( g C , ( ν ) ) = A C ( ν ) . (41)
and passing to global sections in (5.1) recovers the PBW isomorphism
Sym ( ⨁ n ≥ 1 g ^ C ( n α 0 ) ) → ≅ A C ( ν ) (42) \operatorname{Sym}\left(\bigoplus_{n\geq1}\widehat{\mathfrak{g}}^C(n\alpha_0)\right)\xrightarrow{\cong}A_C^{(\nu)} \tag*{(42)} Sym ( n ≥ 1 ⨁ g C ( n α 0 ) ) ≅ A C ( ν ) (42)
where the map is given by the COHA multiplication. Finally, there is a comultiplication [12 ]
Δ : A C ( ν ) ⟶ A C ( ν ) ⊗ A C ( ν ) (43) \Delta:A_C^{(\nu)}\longrightarrow A_C^{(\nu)}\otimes A_C^{(\nu)} \tag*{(43)} Δ : A C ( ν ) ⟶ A C ( ν ) ⊗ A C ( ν ) (43)
It is compatible with the COHA product. The space of primitive elements is g ^ C , ( ν ) \widehat{\mathfrak{g}}^{C,(\nu)} g C , ( ν ) . The graded dimension of g C , ( ν ) \mathfrak{g}^{C,(\nu)} g C , ( ν ) is given in terms of the Kac polynomial A g , r ( σ ) A_{g,r}(\sigma) A g , r ( σ ) of [37 ] . We will not need this.
Remark 5.1. The non-abelian Hodge correspondence yields an isomorphism of COHAs and BPS Lie algebras, involving A C ( 0 ) A_C^{(0)} A C ( 0 ) on the Higgs side and the COHA of untwisted local systems on Σ g \Sigma_g Σ g on the Betti side, see [18 ] , [36 ] .
The perverse filtration on the BPS Lie algebra
The Hitchin map
h : M D ( r , d ) → A r h:M^D(r,d)\to\mathcal{A}_r h : M D ( r , d ) → A r
is well-defined for every ( r , d ) (r,d) ( r , d ) . It is proper and induces a perverse filtration on g C ( r , d ) \mathfrak{g}^{C}(r,d) g C ( r , d ) defined by
P k g C ( r , d ) = Im ( H − ℓ ( A r , p τ ≤ k − N h ∗ B P S ( r , d ) C ) → H − ℓ ( A r , h ∗ B P S ( r , d ) C ) ) . (44) P_k\mathfrak{g}^{C}(r,d)=\operatorname{Im}\left(\mathrm{H}^{-\ell}\left(\mathcal{A}_r,{}^p\tau_{\leq k-N}h_*\mathcal{BPS}^{C}_{(r,d)}\right)\to\mathrm{H}^{-\ell}\left(\mathcal{A}_r,h_*\mathcal{BPS}^{C}_{(r,d)}\right)\right). \tag*{(44)} P k g C ( r , d ) = Im ( H − ℓ ( A r , p τ ≤ k − N h ∗ B P S ( r , d ) C ) → H − ℓ ( A r , h ∗ B P S ( r , d ) C ) ) . (44)
This definition ensures that the perverse filtration on g C ( r , d ) \mathfrak{g}^{C}(r,d) g C ( r , d ) is non trivial only in the interval [ 0 , 2 N ] [0,2N] [ 0 , 2 N ] , i.e., we have
P − 1 g C ( r , d ) = { 0 } , P 2 N g C ( r , d ) = g C ( r , d ) . P_{-1}\mathfrak{g}^{C}(r,d)=\{0\},\qquad P_{2N}\mathfrak{g}^{C}(r,d)=\mathfrak{g}^{C}(r,d). P − 1 g C ( r , d ) = { 0 } , P 2 N g C ( r , d ) = g C ( r , d ) .
For coprime ( r , d ) (r,d) ( r , d ) , every semistable Higgs bundle is stable and M D ( r , d ) M^D(r,d) M D ( r , d ) is smooth of complex dimension 2 N 2N 2 N . In this case the BPS sheaf is the intersection complex
B P S ( r , d ) C = IC ( M D ( r , d ) , Q ) = Q M D ( r , d ) [ 2 N ] \mathcal{BPS}^{C}_{(r,d)}=\operatorname{IC}(M^D(r,d),\mathbb{Q})=\mathbb{Q}_{M^D(r,d)}[2N] B P S ( r , d ) C = IC ( M D ( r , d ) , Q ) = Q M D ( r , d ) [ 2 N ]
Taking hypercohomology in degree − i -i − i therefore gives
g i C ( r , d ) ≅ H 2 N − i ( M D ( r , d ) , Q ) . (45) \mathfrak{g}^{C}_i(r,d)\cong\mathrm{H}^{2N-i}(M^D(r,d),\mathbb{Q}). \tag*{(45)} g i C ( r , d ) ≅ H 2 N − i ( M D ( r , d ) , Q ) . (45)
This is the form of Poincaré duality compatible with the Borel–Moore homology realization of the BPS Lie algebra: the shift [ 2 N ] [2N] [ 2 N ] changes BPS degree − i -i − i into ordinary cohomological degree 2 N − i 2N-i 2 N − i . It also explains the normalization of the perverse filtration. Indeed,
p τ ≤ k − N ( h ∗ Q M D ( r , d ) [ 2 N ] ) ≅ ( p τ ≤ k + N h ∗ Q M D ( r , d ) ) [ 2 N ] , {}^p\tau_{\leq k-N}(h_*\mathbb{Q}_{M^D(r,d)}[2N])\cong({}^p\tau_{\leq k+N}h_*\mathbb{Q}_{M^D(r,d)})[2N], p τ ≤ k − N ( h ∗ Q M D ( r , d ) [ 2 N ]) ≅ ( p τ ≤ k + N h ∗ Q M D ( r , d ) ) [ 2 N ] ,
so that the filtration (5.5), under the identification (5.6), is precisely (4.4).
Definition (5.5) extends to the affinized BPS Lie algebra g ^ C \widehat{\mathfrak{g}}^C g C , by replacing B P S ( r , d ) C \mathcal{BPS}^{C}_{(r,d)} B P S ( r , d ) C with B P S ( r , d ) C ⊗ H G m ∗ \mathcal{BPS}^{C}_{(r,d)}\otimes\mathrm{H}^*_{\mathbb{G}_m} B P S ( r , d ) C ⊗ H G m ∗ . Write H G m ∗ = Q [ u ] \mathrm{H}^*_{\mathbb{G}_m}=\mathbb{Q}[u] H G m ∗ = Q [ u ] , where u u u has homological degree − 2 -2 − 2 and perverse degree 2. The canonical inclusion of the constant term induces an isomorphism
⨁ ℓ gr N − ℓ P g 2 ℓ C ( r , d ) ≅ ⨁ ℓ gr N − ℓ P g ^ 2 ℓ C ( r , d ) . (46) \bigoplus_{\ell}\operatorname{gr}^{P}_{N-\ell}\mathfrak{g}^{C}_{2\ell}(r,d)\cong\bigoplus_{\ell}\operatorname{gr}^{P}_{N-\ell}\widehat{\mathfrak{g}}^{C}_{2\ell}(r,d). \tag*{(46)} ℓ ⨁ gr N − ℓ P g 2 ℓ C ( r , d ) ≅ ℓ ⨁ gr N − ℓ P g 2 ℓ C ( r , d ) . (46)
Here the absence of contributions from positive powers of u u u uses, in addition to their bidegrees, the following vanishing:
P N − s − 1 g 2 s C ( r , d ) = 0 ( s ∈ Z ) . P_{N-s-1}\mathfrak{g}^{C}_{2s}(r,d)=0\qquad(s\in\mathbb{Z}). P N − s − 1 g 2 s C ( r , d ) = 0 ( s ∈ Z ) .
For coprime ( r , d ) (r,d) ( r , d ) this follows from (5.6), P=W and the inequality W 2 k − 1 H 2 k ( M B ( r , d ) , Q ) = 0 W_{2k-1}\mathrm{H}^{2k}(M^B(r,d),\mathbb{Q})=0 W 2 k − 1 H 2 k ( M B ( r , d ) , Q ) = 0 ; for arbitrary d d d it follows by the filtered χ \chi χ -independence isomorphisms stated below in (5.8), after choosing d ′ d' d ′ coprime to r r r . Indeed, keeping track of both gradings gives
gr N − ℓ P g ^ 2 ℓ C ( r , d ) ≅ ⨁ m ≥ 0 u m gr N − ℓ − 2 m P g 2 ( ℓ + m ) C ( r , d ) . \operatorname{gr}^{P}_{N-\ell}\widehat{\mathfrak{g}}^{C}_{2\ell}(r,d)\cong\bigoplus_{m\geq0}u^m\operatorname{gr}^{P}_{N-\ell-2m}\mathfrak{g}^{C}_{2(\ell+m)}(r,d). gr N − ℓ P g 2 ℓ C ( r , d ) ≅ m ≥ 0 ⨁ u m gr N − ℓ − 2 m P g 2 ( ℓ + m ) C ( r , d ) .
For m ≥ 1 m\geq1 m ≥ 1 we have N − ℓ − 2 m < N − ( ℓ + m ) N-\ell-2m<N-(\ell+m) N − ℓ − 2 m < N − ( ℓ + m ) , so the corresponding summand vanishes by the preceding bound with s = ℓ + m s=\ell+m s = ℓ + m . Only m = 0 m=0 m = 0 survives, which proves (5.7).
Tautological classes
Let V \mathcal{V} V be the tautological vector bundle on M s s ( α ) × C \mathcal{M}^{\mathrm{ss}}(\alpha) \times C M ss ( α ) × C . As before, we define tautological cohomology classes by
ch k ( η ) = ∫ C ch k ( V ) ∪ p 2 ∗ ( η ) ∈ H 2 k − 2 + ∣ η ∣ ( M s s ( α ) , Q ) , ∀ k ≥ 1 , η ∈ H ∗ ( C , Q ) \operatorname{ch}_{k}(\eta)=\int_{C}\operatorname{ch}_{k}(\mathcal{V})\cup p_{2}^{*}(\eta)\in H^{2k-2+|\eta|}(\mathcal{M}^{\mathrm{ss}}(\alpha),\mathbb{Q}),\qquad\forall k\geq1,\ \eta\in H^{*}(C,\mathbb{Q}) ch k ( η ) = ∫ C ch k ( V ) ∪ p 2 ∗ ( η ) ∈ H 2 k − 2 + ∣ η ∣ ( M ss ( α ) , Q ) , ∀ k ≥ 1 , η ∈ H ∗ ( C , Q )
and we define an algebra
S C = Q [ ch k ( [ p t ] ) , ch k ( 1 ) , ch k ( γ i ) ; k ≥ 1 , i = 1 , … , 2 g ] S_{C}=\mathbb{Q}[\operatorname{ch}_{k}([\mathrm{pt}]),\operatorname{ch}_{k}(1),\operatorname{ch}_{k}(\gamma_{i});\ k\geq1,i=1,\ldots,2g] S C = Q [ ch k ([ pt ]) , ch k ( 1 ) , ch k ( γ i ) ; k ≥ 1 , i = 1 , … , 2 g ]
acting on A C ( ν ) A_{C}^{(\nu)} A C ( ν ) for any slope ν \nu ν . Lemma 2.7 holds in the present setting as well, and thus the action of S C S_{C} S C on A C ( ν ) A_{C}^{(\nu)} A C ( ν ) preserves g ^ C , ( ν ) \widehat{\mathfrak{g}}^{C,(\nu)} g C , ( ν ) . We put
u = c 1 ( [ p t ] ) ∈ H 2 ( M s s ( α ) , Q ) . u=c_{1}([\mathrm{pt}])\in H^{2}(\mathcal{M}^{\mathrm{ss}}(\alpha),\mathbb{Q}). u = c 1 ([ pt ]) ∈ H 2 ( M ss ( α ) , Q ) .
Note that by definition, u = c 1 ( V x ) u=c_{1}(\mathcal{V}_{x}) u = c 1 ( V x ) for any x ∈ C x\in C x ∈ C , where V x \mathcal{V}_{x} V x is the restriction of V \mathcal{V} V to M s s ( r , d ) × { x } \mathcal{M}^{\mathrm{ss}}(r,d)\times\{x\} M ss ( r , d ) × { x } .
If ( r , d ) (r,d) ( r , d ) is primitive, then the map
p : M s s , r e d ( r , d ) ⟶ M D ( r , d ) p:\mathcal{M}^{\mathrm{ss,red}}(r,d)\longrightarrow M^{D}(r,d) p : M ss , red ( r , d ) ⟶ M D ( r , d )
is a G m \mathbb{G}_{m} G m -gerbe and we may consider a P G L r PGL_{r} P G L r -principal bundle V ~ \widetilde{\mathcal{V}} V on the moduli space M D ( r , d ) M^{D}(r,d) M D ( r , d ) . It is related to the tautological bundle V \mathcal{V} V by the relation
( p × id ) ∗ ( V ~ ) = V ⊗ ( det V ) − 1 / r (p\times\operatorname{id})^{*}(\widetilde{\mathcal{V}})=\mathcal{V}\otimes(\det\mathcal{V})^{-1/r} ( p × id ) ∗ ( V ) = V ⊗ ( det V ) − 1/ r
where det V \det\mathcal{V} det V is the universal line bundle pulled back from the map
det : M s s , r e d ( r , d ) ⟶ T ∗ Pic d ( C ) , ( E , φ ) ⟼ ( det E , tr ( φ ) ) . \det:\mathcal{M}^{\mathrm{ss,red}}(r,d)\longrightarrow T^{*}\operatorname{Pic}^{d}(C),\qquad(\mathcal{E},\varphi)\longmapsto(\det\mathcal{E},\operatorname{tr}(\varphi)). det : M ss , red ( r , d ) ⟶ T ∗ Pic d ( C ) , ( E , φ ) ⟼ ( det E , tr ( φ )) .
In particular, the tautological classes on M s s ( r , d ) \mathcal{M}^{\mathrm{ss}}(r,d) M ss ( r , d ) and M D ( r , d ) M^{D}(r,d) M D ( r , d ) are related via the equality
ch ( V ) = ( p × id ) ∗ ch ( V ~ ) e c 1 ( V ) . \operatorname{ch}(\mathcal{V})=(p\times\operatorname{id})^{*}\operatorname{ch}(\widetilde{\mathcal{V}})e^{c_{1}(\mathcal{V})}. ch ( V ) = ( p × id ) ∗ ch ( V ) e c 1 ( V ) .
Note that c 1 ( 1 ) = d c_{1}(1)=d c 1 ( 1 ) = d , the classes ϕ i , 1 = c 1 ( γ i ) \phi_{i,1}=c_{1}(\gamma_{i}) ϕ i , 1 = c 1 ( γ i ) generate a copy of H ∗ ( T ∗ Pic d , Q ) H^{*}(T^{*}\operatorname{Pic}^{d},\mathbb{Q}) H ∗ ( T ∗ Pic d , Q ) , and u u u has a non-zero contribution in H G m 2 H^{2}_{\mathbb{G}_{m}} H G m 2 for any trivialization of the G m \mathbb{G}_{m} G m -gerbe p p p . We deduce that
H ∗ ( M s s ( r , d ) , Q ) = p ∗ ( H ∗ ( M D ( r , d ) ) ) [ u ] . H^{*}(\mathcal{M}^{\mathrm{ss}}(r,d),\mathbb{Q})=p^{*}(H^{*}(M^{D}(r,d)))[u]. H ∗ ( M ss ( r , d ) , Q ) = p ∗ ( H ∗ ( M D ( r , d ))) [ u ] .
The reduced virtual fundamental class and Markman’s theorem for g ^ C \widehat{\mathfrak{g}}^{C} g C
We refer to [12 Sec.7.2] for the definition of the reduced derived stack M s s , r d ( α ) \mathcal{M}^{\mathrm{ss,rd}}(\alpha) M ss , rd ( α ) . It is a closed substack of M s s ( α ) \mathcal{M}^{\mathrm{ss}}(\alpha) M ss ( α ) . It should not to be confused with the reduced stack M s s , r e d ( α ) \mathcal{M}^{\mathrm{ss,red}}(\alpha) M ss , red ( α ) , which is a classical stack, and is a closed substack of M s s , r d ( α ) \mathcal{M}^{\mathrm{ss,rd}}(\alpha) M ss , rd ( α ) . The reduced derived stack is quasi-smooth and satisfies
( M s s , r d ( α ) ) c l = M s s , c l ( α ) , vdim ( M s s , r d ) ( α ) = vdim M s s ( α ) + 1 = 1 − ( α , α ) . (\mathcal{M}^{\mathrm{ss,rd}}(\alpha))^{\mathrm{cl}}=\mathcal{M}^{\mathrm{ss,cl}}(\alpha),\qquad\operatorname{vdim}(\mathcal{M}^{\mathrm{ss,rd}})(\alpha)=\operatorname{vdim}\mathcal{M}^{\mathrm{ss}}(\alpha)+1=1-(\alpha,\alpha). ( M ss , rd ( α ) ) cl = M ss , cl ( α ) , vdim ( M ss , rd ) ( α ) = vdim M ss ( α ) + 1 = 1 − ( α , α ) .
It is obtained from M s s ( α ) \mathcal{M}^{\mathrm{ss}}(\alpha) M ss ( α ) by removing the trivial one-dimensional summand in the obstruction space via the trace map
tr F : Ext 2 ( F , F ) ≅ Hom ( F , F ) ∨ ⟶ C . \operatorname{tr}_{\mathcal{F}}:\operatorname{Ext}^{2}(\mathcal{F},\mathcal{F})\cong\operatorname{Hom}(\mathcal{F},\mathcal{F})^{\vee}\longrightarrow\mathbb{C}. tr F : Ext 2 ( F , F ) ≅ Hom ( F , F ) ∨ ⟶ C .
We define
g ^ t a u t C ( α ) = S C ∙ [ M s s , r d ( α ) ] . \widehat{\mathfrak{g}}_{\mathrm{taut}}^{C}(\alpha)=S_{C}\bullet[\mathcal{M}^{\mathrm{ss,rd}}(\alpha)]. g taut C ( α ) = S C ∙ [ M ss , rd ( α )] .
The following may be viewed as an analog of Markman’s generation theorem for the cohomology of M D ( r , d ) M^{D}(r,d) M D ( r , d ) for nonprimitive pairs ( r , d ) (r,d) ( r , d ) .
Theorem 5.2 ([12 Thm.7.5] ). For all classes α \alpha α of positive rank we have
g ^ t a u t C ( α ) = g ^ C ( α ) . \widehat{\mathfrak{g}}_{\mathrm{taut}}^{C}(\alpha)=\widehat{\mathfrak{g}}^{C}(\alpha). g taut C ( α ) = g C ( α ) .
χ \chi χ -independence
The Hitchin base A r \mathcal{A}_r A r is independent of d d d . Fixing r r r , it thus makes sense to compare the semisimple constructible complexes h ∗ ( B P S C ( r , d ) ) ∈ D b ( A r ) h_*(\mathcal{BPS}^{C}(r,d)) \in D^b(\mathcal{A}_r) h ∗ ( B P S C ( r , d )) ∈ D b ( A r ) for various values of d d d .
Theorem 5.3 ([23 Thm. 1.2, Cor. 5.15] ). For any r r r and any two d , d ′ d,d' d , d ′ there is a canonical isomorphism
h ∗ ( B P S C ( r , d ) ) ≅ h ∗ ( B P S C ( r , d ′ ) ) . h_*(\mathcal{BPS}^{C}(r,d)) \cong h_*(\mathcal{BPS}^{C}(r,d')). h ∗ ( B P S C ( r , d )) ≅ h ∗ ( B P S C ( r , d ′ )) .
Passing to global sections, this yields isomorphisms of graded vector spaces
g ^ C ( r , d ) ≅ g ^ C ( r , d ′ ) , g C ( r , d ) ≅ g C ( r , d ′ ) (47) \widehat{\mathfrak{g}}^{C}(r,d) \cong\widehat{\mathfrak{g}}^{C}(r,d'), \qquad\mathfrak{g}^{C}(r,d) \cong\mathfrak{g}^{C}(r,d') \tag*{(47)} g C ( r , d ) ≅ g C ( r , d ′ ) , g C ( r , d ) ≅ g C ( r , d ′ ) (47)
which are compatible with the perverse filtration.
Stacky P=C
The P=C equality admits an extension to the case of an arbitrary pair ( r , d ) (r,d) ( r , d ) due to R. Picovschi. It is best understood in terms of the (affinized) BPS Lie algebra. Recall the ring
S C = Q [ c k ( α ) ; k ≥ 1 , α ] = Q [ ch k ( α ) ; k ≥ 1 , α ] , S_C = \mathbb{Q}[c_k(\alpha); k \ge1,\alpha] = \mathbb{Q}[\operatorname{ch}_k(\alpha); k \ge1,\alpha], S C = Q [ c k ( α ) ; k ≥ 1 , α ] = Q [ ch k ( α ) ; k ≥ 1 , α ] ,
where α \alpha α runs over a basis of H ∗ ( C , Q ) H^*(C,\mathbb{Q}) H ∗ ( C , Q ) . We equip S C S_C S C with the Chern grading defined by
deg ( ch k ( α ) ) = k ∀ k ≥ 1 , α ∈ H ∗ ( C , Q ) . \deg(\operatorname{ch}_k(\alpha)) = k \quad\forall\ k \ge1,\alpha\in H^*(C,\mathbb{Q}). deg ( ch k ( α )) = k ∀ k ≥ 1 , α ∈ H ∗ ( C , Q ) .
Notice that deg ( ch k ( α ) ) \deg(\operatorname{ch}_k(\alpha)) deg ( ch k ( α )) coincides with the cohomological degree (up to a factor of 2) if and only if α ∈ Q [ p t ] \alpha\in\mathbb{Q}[\mathrm{pt}] α ∈ Q [ pt ] . Let us write
S C = ⨁ n ≥ 0 S C , n , S C , ≤ n = ⨁ ℓ ≤ n S C , n . S_C = \bigoplus_{n \ge0} S_{C,n}, \qquad S_{C,\le n} = \bigoplus_{\ell\le n} S_{C,n}. S C = n ≥ 0 ⨁ S C , n , S C , ≤ n = ℓ ≤ n ⨁ S C , n .
For every ( r , d ) (r,d) ( r , d ) we define an automorphism of S C S_C S C given on generators by
φ r , d : ch n ( α ) ↦ ch n ( α ) + d n r ch n − 1 ( [ p t ] ∪ α ) , ( n ≥ 1 , α ) . \varphi_{r,d}: \operatorname{ch}_n(\alpha) \mapsto\operatorname{ch}_n(\alpha) + \frac{dn}{r}\operatorname{ch}_{n-1}([\mathrm{pt}] \cup\alpha), \qquad(n \ge1,\alpha). φ r , d : ch n ( α ) ↦ ch n ( α ) + r d n ch n − 1 ([ pt ] ∪ α ) , ( n ≥ 1 , α ) .
Note that φ r , n ∘ φ r , n ′ = φ r , n + n ′ \varphi_{r,n} \circ\varphi_{r,n'} = \varphi_{r,n+n'} φ r , n ∘ φ r , n ′ = φ r , n + n ′ , φ r , 0 = I d \varphi_{r,0} = \mathrm{Id} φ r , 0 = Id , φ r , d ( S C , ≤ n ) = S C , ≤ n \varphi_{r,d}(S_{C,\le n}) = S_{C,\le n} φ r , d ( S C , ≤ n ) = S C , ≤ n . Note also that φ r , d ( ch k ( α ) ) = ch k ( α ) \varphi_{r,d}(\operatorname{ch}_k(\alpha)) = \operatorname{ch}_k(\alpha) φ r , d ( ch k ( α )) = ch k ( α ) unless ∣ α ∣ = 0 |\alpha| = 0 ∣ α ∣ = 0 .
Theorem 5.4 ([35 ] ). There exists a splitting of the perverse filtration
P k g ^ C ( r , d ) = ⨁ ℓ = 0 k L ℓ g ^ C ( r , d ) P_k\widehat{\mathfrak{g}}^{C}(r,d) = \bigoplus_{\ell=0}^{k} L_\ell\widehat{\mathfrak{g}}^{C}(r,d) P k g C ( r , d ) = ℓ = 0 ⨁ k L ℓ g C ( r , d ) such that
φ r , d ( c s ( η ) ‾ ) ∙ L ℓ g ^ k C ( r , d ) ⊆ L ℓ + s g ^ k − 2 s + 2 − ∣ η ∣ C ( r , d ) , \varphi_{r,d}(\overline{c_s(\eta)}) \bullet L_\ell\widehat{\mathfrak{g}}^{C}_k(r,d) \subseteq L_{\ell+s}\widehat{\mathfrak{g}}^{C}_{k-2s+2-|\eta|}(r,d), φ r , d ( c s ( η ) ) ∙ L ℓ g k C ( r , d ) ⊆ L ℓ + s g k − 2 s + 2 − ∣ η ∣ C ( r , d ) , u ∙ L ℓ g ^ k C ( r , d ) ⊆ L ℓ + 2 g ^ k − 2 C ( r , d ) u \bullet L_\ell\widehat{\mathfrak{g}}^{C}_k(r,d) \subseteq L_{\ell+2}\widehat{\mathfrak{g}}^{C}_{k-2}(r,d) u ∙ L ℓ g k C ( r , d ) ⊆ L ℓ + 2 g k − 2 C ( r , d ) for any s ≥ 1 s \ge1 s ≥ 1 and ( s , η ) ≠ ( 1 , [ p t ] ) (s,\eta) \ne(1,[\mathrm{pt}]) ( s , η ) = ( 1 , [ pt ]) .
As a consequence, we have
⨁ ℓ L N − ℓ g ^ ℓ C ( r , d ) = Q [ c ‾ 2 ( [ p t ] ) , c ‾ 3 ( [ p t ] ) , … ] ∙ [ M s s , r d ( r , d ) ] (48) \bigoplus_{\ell} L_{N-\ell}\widehat{\mathfrak{g}}^{C}_\ell(r,d) = \mathbb{Q}[\overline{c}_2([\mathrm{pt}]),\overline{c}_3([\mathrm{pt}]),\ldots] \bullet[\mathcal{M}^{\mathrm{ss,rd}}(r,d)] \tag*{(48)} ℓ ⨁ L N − ℓ g ℓ C ( r , d ) = Q [ c 2 ([ pt ]) , c 3 ([ pt ]) , … ] ∙ [ M ss , rd ( r , d )] (48) and for any sequence ( s 1 , … , s p ) (s_1,\ldots,s_p) ( s 1 , … , s p ) we have
∏ i c ‾ s i ( [ p t ] ) ∩ [ M s s , r d ( r , d ) ] ∈ L ∑ i s i g ^ C ( r , d ) . (49) \prod_i \overline{c}_{s_i}([\mathrm{pt}]) \cap[\mathcal{M}^{\mathrm{ss,rd}}(r,d)] \in L_{\sum_i s_i}\widehat{\mathfrak{g}}^{C}(r,d). \tag*{(49)} i ∏ c s i ([ pt ]) ∩ [ M ss , rd ( r , d )] ∈ L ∑ i s i g C ( r , d ) . (49)
Remark 5.5. For the purpose of this paper, only the pure part of Theorem 5.4, i.e. (5.9) and (5.10) are needed. This statement can be given a simpler proof, based on [13 ] and [12 ] . For the reader’s comfort, we provide a proof of these in Appendix C. We are greatful to Jiahao Bi for pointing this out.
Restriction around the trivial Higgs bundle.
We keep the notation of the previous section: C C C is a smooth connected projective complex curve of genus g g g , M s s ( r , 0 ) \mathcal{M}^{\mathrm{ss}}(r,0) M ss ( r , 0 ) and M D ( r , 0 ) M^D(r,0) M D ( r , 0 ) are the derived moduli stack and coarse moduli space of semistable Higgs bundles on C C C of rank r r r and degree 0 0 0 , and p : M s s , c l ( r , 0 ) → M D ( r , 0 ) p:\mathcal{M}^{\mathrm{ss},\mathrm{cl}}(r,0)\to M^D(r,0) p : M ss , cl ( r , 0 ) → M D ( r , 0 ) is the good moduli space.
6.1. Davison’s neighborhood theorem. The Higgs bundle ( O r , φ = 0 ) (\mathcal{O}^r,\varphi=0) ( O r , φ = 0 ) is polystable. Thus it corresponds to a point of M D ( r , 0 ) M^D(r,0) M D ( r , 0 ) . Let O r \mathcal{O}^r O r denote this point. We have
Ext 0 ( O , O ) ≅ Ext 2 ( O , O ) ∨ ≅ C , Ext 1 ( O , O ) ≅ T ∗ H 1 ( C , O r ) = C 2 g \operatorname{Ext}^{0}(\mathcal{O},\mathcal{O})\cong\operatorname{Ext}^{2}(\mathcal{O},\mathcal{O})^{\vee}\cong\mathbb{C},\qquad\operatorname{Ext}^{1}(\mathcal{O},\mathcal{O})\cong T^{*}H^{1}(C,\mathcal{O}^{r})=\mathbb{C}^{2g} Ext 0 ( O , O ) ≅ Ext 2 ( O , O ) ∨ ≅ C , Ext 1 ( O , O ) ≅ T ∗ H 1 ( C , O r ) = C 2 g
The associated Ext-quiver is the S g S_g S g quiver. The relation between the COHA and the BPS Lie algebra A C , ( 0 ) A^{C,(0)} A C , ( 0 ) and g C , ( 0 ) \mathfrak{g}^{C,(0)} g C , ( 0 ) attached to M s s ( r , 0 ) \mathcal{M}^{\mathrm{ss}}(r,0) M ss ( r , 0 ) , and the COHA and BPS Lie algebra A S g A^{S_g} A S g and g S g \mathfrak{g}^{S_g} g S g attached to S g S_g S g is given by the following theorem due to Davison.
Theorem 6.1 ([6 Thm. 5.11] ). There are open analytic neighborhoods U U U of O r \mathcal{O}^{r} O r in M D ( r , 0 ) M^D(r,0) M D ( r , 0 ) and U ′ U' U ′ of 0 0 0 in Rep Π S g \operatorname{Rep}\Pi_{S_g} Rep Π S g together with analytic isomorphisms ϕ : U ≅ U ′ \phi:U\cong U' ϕ : U ≅ U ′ mapping O r \mathcal{O}^{r} O r to 0 0 0 and ϕ : U = p − 1 ( U ) ≅ p − 1 ( U ′ ) = : U ′ \phi:\mathcal{U}=p^{-1}(U)\cong p^{-1}(U')=:\mathcal{U}' ϕ : U = p − 1 ( U ) ≅ p − 1 ( U ′ ) =: U ′ making the following diagram commute
M s s , c l ( r , 0 ) ← j U U → ϕ U ′ → j U ′ Rep r c l Π S g ↓ p ↓ p ↓ p ↓ p M D ( r , 0 ) ← j U U → ϕ U ′ → j U ′ Rep r Π S g . \begin{aligned}
\mathcal{M}^{\mathrm{ss},\mathrm{cl}}(r,0)&\xleftarrow{\ j_U\ }\mathcal{U}\xrightarrow{\ \phi\ }\mathcal{U}'\xrightarrow{\ j_{U'}\ }\operatorname{Rep}^{\mathrm{cl}}_r\Pi_{S_g}\\
\downarrow p&\qquad\downarrow p\qquad\downarrow p\qquad\downarrow p\\
M^D(r,0)&\xleftarrow{\ j_U\ }U\xrightarrow{\ \phi\ }U'\xrightarrow{\ j_{U'}\ }\operatorname{Rep}_r\Pi_{S_g}.
\end{aligned} M ss , cl ( r , 0 ) ↓ p M D ( r , 0 ) j U U ϕ U ′ j U ′ Rep r cl Π S g ↓ p ↓ p ↓ p j U U ϕ U ′ j U ′ Rep r Π S g . Here the leftmost and rightmost horizontal arrows are open embeddings.
Note that, since Rep Π S g \operatorname{Rep}\Pi_{S_g} Rep Π S g is a conical stack with apex 0 0 0 , the restriction to any star-shaped open neighborhood U ′ U' U ′ of 0 0 0 induces an isomorphism in Borel-Moore homology
H ∗ ( U ′ , Q ) ≅ H ∗ ( Rep Π S g , Q ) . H_{*}(U',\mathbb{Q})\cong H_{*}(\operatorname{Rep}\Pi_{S_g},\mathbb{Q}). H ∗ ( U ′ , Q ) ≅ H ∗ ( Rep Π S g , Q ) .
Composing with ϕ ∗ \phi_{*} ϕ ∗ , we obtain a restriction morphism, which is independent of the choices of U U U , U ′ U' U ′ :
res U : H ∗ ( M s s ( r , 0 ) , Q ) ⟶ H ∗ ( Rep Π S g , Q ) . \operatorname{res}_{U}:H_{*}(\mathcal{M}^{\mathrm{ss}}(r,0),\mathbb{Q})\longrightarrow H_{*}(\operatorname{Rep}\Pi_{S_g},\mathbb{Q}). res U : H ∗ ( M ss ( r , 0 ) , Q ) ⟶ H ∗ ( Rep Π S g , Q ) .
Considering all values of r ≥ 0 r\geq0 r ≥ 0 , we can form a system of open neighborhoods U r U_r U r in M D ( r , 0 ) M^D(r,0) M D ( r , 0 ) such that
⨁ r , d 1 ( U r + s ) ⊂ U r × U s , ∀ r , s ≥ 0. \bigoplus_{r,d}^{1}(U_{r+s})\subset U_r\times U_s,\qquad\forall r,s\geq0. r , d ⨁ 1 ( U r + s ) ⊂ U r × U s , ∀ r , s ≥ 0.
The same thus holds for { U r ′ } r \{U'_r\}_r { U r ′ } r in Rep Π S g \operatorname{Rep}\Pi_{S_g} Rep Π S g .
Corollary 6.2. The collection of restriction maps res U r \operatorname{res}_{U_r} res U r induces an algebra morphism A C ( 0 ) → A Π S g A_C^{(0)}\to A_{\Pi_{S_g}} A C ( 0 ) → A Π S g which restricts to Lie algebra morphisms g C , ( 0 ) → g Π S g \mathfrak{g}^{C,(0)}\to\mathfrak{g}^{\Pi_{S_g}} g C , ( 0 ) → g Π S g and g ^ C , ( 0 ) → g ^ Π S g \widehat{\mathfrak{g}}^{C,(0)}\to\widehat{\mathfrak{g}}^{\Pi_{S_g}} g C , ( 0 ) → g Π S g .
Proof. The COHA multiplication A C ( r , 0 ) ⊗ A C ( s , 0 ) → A C ( r + s , 0 ) A_C(r,0)\otimes A_C(s,0)\to A_C(r+s,0) A C ( r , 0 ) ⊗ A C ( s , 0 ) → A C ( r + s , 0 ) is obtained by taking derived global sections of a morphism of complexes over M D ( r + s , 0 ) M^D(r+s,0) M D ( r + s , 0 )
⊕ ∗ ( p ∗ ( D Q M s s ( r , 0 ) ⊗ L ( g − 1 ) r 2 ) ⊠ p ∗ ( D Q M s s ( s , 0 ) ⊗ L ( g − 1 ) s 2 ) ) ⟶ p ∗ ( D Q M s s ( r + s , 0 ) ⊗ L ( g − 1 ) ( r + s ) 2 ) \oplus_{*}\left(p_{*}\left(\mathbb{D}\mathcal{Q}_{\mathcal{M}^{\mathrm{ss}}(r,0)}\otimes\mathcal{L}^{(g-1)r^{2}}\right)\boxtimes p_{*}\left(\mathbb{D}\mathcal{Q}_{\mathcal{M}^{\mathrm{ss}}(s,0)}\otimes\mathcal{L}^{(g-1)s^{2}}\right)\right)\longrightarrow p_{*}\left(\mathbb{D}\mathcal{Q}_{\mathcal{M}^{\mathrm{ss}}(r+s,0)}\otimes\mathcal{L}^{(g-1)(r+s)^{2}}\right) ⊕ ∗ ( p ∗ ( D Q M ss ( r , 0 ) ⊗ L ( g − 1 ) r 2 ) ⊠ p ∗ ( D Q M ss ( s , 0 ) ⊗ L ( g − 1 ) s 2 ) ) ⟶ p ∗ ( D Q M ss ( r + s , 0 ) ⊗ L ( g − 1 ) ( r + s ) 2 )
and likewise for A Π S g A_{\Pi_{S_g}} A Π S g . By open base change and further restricting to U r + s U_{r+s} U r + s we get a similar morphism
⊕ ∗ ( p ∗ ( D Q U r ⊗ L ( g − 1 ) r 2 ) ⊠ p ∗ ( D Q U s ⊗ L ( g − 1 ) s 2 ) ) ⟶ p ∗ ( D Q U r + s ⊗ L ( g − 1 ) ( r + s ) 2 ) \oplus_{*}\left(p_{*}\left(\mathbb{D}\mathcal{Q}_{U_r}\otimes\mathcal{L}^{(g-1)r^{2}}\right)\boxtimes p_{*}\left(\mathbb{D}\mathcal{Q}_{U_s}\otimes\mathcal{L}^{(g-1)s^{2}}\right)\right)\longrightarrow p_{*}\left(\mathbb{D}\mathcal{Q}_{U_{r+s}}\otimes\mathcal{L}^{(g-1)(r+s)^{2}}\right) ⊕ ∗ ( p ∗ ( D Q U r ⊗ L ( g − 1 ) r 2 ) ⊠ p ∗ ( D Q U s ⊗ L ( g − 1 ) s 2 ) ) ⟶ p ∗ ( D Q U r + s ⊗ L ( g − 1 ) ( r + s ) 2 )
and likewise on the quiver side. The restriction map is thus an algebra morphism on the Hitchin side and an isomorphism on the quiver side. This proves the first statement. The second statement follows from the fact that the construction of the BPS sheaf is local over the coarse moduli space. ∎
Restriction of tautological classes
Davison’s neighborhood isomorphism is compatible with the reduced virtual fundamental classes in the sense that
res U ( [ M s s , r d ( r , 0 ) ] ) = [ R e p r r d Π S g ] . (50) \operatorname{res}_{U}\left([\mathcal{M}^{\mathrm{ss},\mathrm{rd}}(r,0)]\right)=[\mathfrak{Rep}^{\mathrm{rd}}_{r}\Pi_{S_g}]. \tag*{(50)} res U ( [ M ss , rd ( r , 0 )] ) = [ Rep r rd Π S g ] . (50)
As we could not locate this statement in the literature, we provide an argument when g > 1 g>1 g > 1 , which is enough for our purpose. In that case, we claim that [ M s s , r d ( r , 0 ) ] [\mathcal{M}^{\mathrm{ss},\mathrm{rd}}(r,0)] [ M ss , rd ( r , 0 )] and [ R e p r r d Π S g ] [\mathfrak{Rep}^{\mathrm{rd}}_{r}\Pi_{S_g}] [ Rep r rd Π S g ] are both equal to the corresponding classical fundamental classes, and these are obviously compatible with open restriction. Indeed, by [32 Prop. 8.1] the stack M s s ( r , 0 ) \mathcal{M}^{\mathrm{ss}}(r,0) M ss ( r , 0 ) is irreducible and of classical dimension 2 N − 1 2N-1 2 N − 1 , which is the same as the virtual dimension of M s s , r d ( r , 0 ) \mathcal{M}^{\mathrm{ss},\mathrm{rd}}(r,0) M ss , rd ( r , 0 ) . Likewise, by [3 Cor. 1.4.] , the stack R e p r c l Π S g = μ − 1 ( 0 ) / G L r \mathfrak{Rep}^{\mathrm{cl}}_{r}\Pi_{S_g}=\mu^{-1}(0)/GL_r Rep r cl Π S g = μ − 1 ( 0 ) / G L r is of dimension 2 N − 1 2N-1 2 N − 1 , which is the same as the virtual dimension of R e p r r d Π S g \mathfrak{Rep}^{\mathrm{rd}}_{r}\Pi_{S_g} Rep r rd Π S g . In addition, both stacks have dense open substacks which are smooth: it suffices to take the respective substacks parametrizing simple sheaves or representations. Hence, by Corollary B.6, it follows that
[ M s s , r d ( r , 0 ) ] = [ M s s , c l ( r , 0 ) ] , [ R e p r c l Π S g ] = [ R e p r r d Π S g ] [\mathcal{M}^{\mathrm{ss},\mathrm{rd}}(r,0)]=[\mathcal{M}^{\mathrm{ss},\mathrm{cl}}(r,0)],\qquad[\mathfrak{Rep}^{\mathrm{cl}}_{r}\Pi_{S_g}]=[\mathfrak{Rep}^{\mathrm{rd}}_{r}\Pi_{S_g}] [ M ss , rd ( r , 0 )] = [ M ss , cl ( r , 0 )] , [ Rep r cl Π S g ] = [ Rep r rd Π S g ]
as wanted.
We next examine the behavior of res U \operatorname{res}_{U} res U with respect to tautological classes. By construction, the tautological classes on R e p r Π S g \mathfrak{Rep}_{r}\Pi_{S_g} Rep r Π S g are all pulled back via the morphism R e p r Π S g → B G L r ( C ) \mathfrak{Rep}_{r}\Pi_{S_g}\to BGL_r(\mathbb{C}) Rep r Π S g → B G L r ( C ) . We denote by
ch k = ch k ( V ) ∈ H 2 k ( R e p r Π S g , Q ) \operatorname{ch}_{k}=\operatorname{ch}_{k}(V)\in H^{2k}(\mathfrak{Rep}_{r}\Pi_{S_g},\mathbb{Q}) ch k = ch k ( V ) ∈ H 2 k ( Rep r Π S g , Q )
the k k k th Chern character of the tautological bundle V V V . On the Higgs moduli side, we fix a point x ∈ C x\in C x ∈ C and let V x V_x V x be the restriction of the tautological bundle V V V to M s s ( r , 0 ) × { x } \mathcal{M}^{\mathrm{ss}}(r,0)\times\{x\} M ss ( r , 0 ) × { x } . It induces a morphism
ϵ x : M s s ( r , 0 ) → B G L r ( C ) . \epsilon_x:\mathcal{M}^{\mathrm{ss}}(r,0)\to BGL_r(\mathbb{C}). ϵ x : M ss ( r , 0 ) → B G L r ( C ) .
We consider the closed immersions
i O r : B G L r ( C ) = B A u t ( O r ) → U , i 0 : B G L r ( C ) = B A u t ( 0 ) → U ′ i_{\mathcal{O}^r}:BGL_r(\mathbb{C})=BAut(\mathcal{O}^r)\to\mathcal{U},\qquad i_0:BGL_r(\mathbb{C})=BAut(0)\to\mathcal{U}' i O r : B G L r ( C ) = B A u t ( O r ) → U , i 0 : B G L r ( C ) = B A u t ( 0 ) → U ′
Finally, let us denote by
u : U → M s s ( r , 0 ) , u ′ : U ′ → R e p r Π S g u:\mathcal{U}\to\mathcal{M}^{\mathrm{ss}}(r,0),\qquad u':\mathcal{U}'\to\mathfrak{Rep}_{r}\Pi_{S_g} u : U → M ss ( r , 0 ) , u ′ : U ′ → Rep r Π S g
the open analytic embeddings.
Lemma 6.3. The following hold:
(a) The maps i 0 ∗ : H ∗ ( U ′ , Q ) → H G L r ∗ i_0^*:H^*(\mathcal{U}',\mathbb{Q})\to H^*_{GL_r} i 0 ∗ : H ∗ ( U ′ , Q ) → H G L r ∗ and ( u ′ ) ∗ : H ∗ ( R e p r Π S g , Q ) → H ∗ ( U ′ , Q ) (u')^*:H^*(\mathfrak{Rep}_{r}\Pi_{S_g},\mathbb{Q})\to H^*(\mathcal{U}',\mathbb{Q}) ( u ′ ) ∗ : H ∗ ( Rep r Π S g , Q ) → H ∗ ( U ′ , Q ) are algebra isomorphisms.
(b) The maps ϵ x ∗ : H G L r ∗ → H ∗ ( U , Q ) \epsilon_x^*:H^*_{GL_r}\to H^*(\mathcal{U},\mathbb{Q}) ϵ x ∗ : H G L r ∗ → H ∗ ( U , Q ) and i O r ∗ : H ∗ ( U , Q ) → H G L r ∗ i_{\mathcal{O}^r}^*:H^*(\mathcal{U},\mathbb{Q})\to H^*_{GL_r} i O r ∗ : H ∗ ( U , Q ) → H G L r ∗ are inverse algebra isomorphisms. Composing these maps yields an algebra morphism res U : H ∗ ( M s s ( r , 0 ) , Q ) → H ∗ ( R e p r Π S g , Q ) \operatorname{res}_{U}:H^*(\mathcal{M}^{\mathrm{ss}}(r,0),\mathbb{Q})\to H^*(\mathfrak{Rep}_{r}\Pi_{S_g},\mathbb{Q}) res U : H ∗ ( M ss ( r , 0 ) , Q ) → H ∗ ( Rep r Π S g , Q ) .
(c) We have res U ( ch k ( [ p t ] ) ) = ch k \operatorname{res}_{U}(\operatorname{ch}_{k}([pt]))=\operatorname{ch}_{k} res U ( ch k ([ pt ])) = ch k for any k ≥ 1 k\geq1 k ≥ 1 .
(d) The map res U \operatorname{res}_{U} res U restricts to a graded algebra isomorphism Q [ ch 1 ( [ p t ] ) , ch 2 ( [ p t ] ) , … ] ≅ H ∗ ( R e p r Π S g , Q ) \mathbb{Q}[\operatorname{ch}_{1}([pt]),\operatorname{ch}_{2}([pt]),\ldots]\cong H^*(\mathfrak{Rep}_{r}\Pi_{S_g},\mathbb{Q}) Q [ ch 1 ([ pt ]) , ch 2 ([ pt ]) , … ] ≅ H ∗ ( Rep r Π S g , Q ) .
Proof. Part (a) comes from the fact that R e p r Π S g \mathfrak{Rep}_{r}\Pi_{S_g} Rep r Π S g is conical, 0 0 0 is its apex and U ′ \mathcal{U}' U ′ is a star-shaped open neighborhood. Next, the composition
B G L r ( C ) = B A u t ( O r ) → i O r U → ϵ x B G L r ( C ) (51) BGL_r(\mathbb{C})=BAut(\mathcal{O}^r)\xrightarrow{i_{\mathcal{O}^r}}\mathcal{U}\xrightarrow{\epsilon_x}BGL_r(\mathbb{C}) \tag*{(51)} B G L r ( C ) = B A u t ( O r ) i O r U ϵ x B G L r ( C ) (51)
is the identity morphism, hence i O r ∗ ∘ ϵ x ∗ = Id i_{\mathcal{O}^r}^*\circ\epsilon_x^*=\operatorname{Id} i O r ∗ ∘ ϵ x ∗ = Id . By (a), the map i O r ∗ i_{\mathcal{O}^r}^* i O r ∗ is an isomorphism and (b) follows. As [ p t ] ∈ H 2 ( C , Q ) [pt]\in H^2(C,\mathbb{Q}) [ pt ] ∈ H 2 ( C , Q ) is Poincaré dual to a point, we have
ch k ( [ p t ] ) = ∫ C ch k ( V ) ∪ p 2 ∗ ( [ p t ] ) = i x ∗ ch k ( V ) = ch k ( V x ) = ϵ x ∗ ( ch k ) , \operatorname{ch}_{k}([pt])=\int_C\operatorname{ch}_{k}(V)\cup p_2^*([pt])=i_x^*\operatorname{ch}_{k}(V)=\operatorname{ch}_{k}(V_x)=\epsilon_x^*(\operatorname{ch}_{k}), ch k ([ pt ]) = ∫ C ch k ( V ) ∪ p 2 ∗ ([ pt ]) = i x ∗ ch k ( V ) = ch k ( V x ) = ϵ x ∗ ( ch k ) ,
where i x : { x } → C i_x:\{x\}\to C i x : { x } → C is the inclusion. Statement (c) easily follows. Finally (d) is a consequence of (c) and of the fact that H ∗ ( R e p r Π S g , Q ) ≅ H G L r ∗ H^*(\mathfrak{Rep}_{r}\Pi_{S_g},\mathbb{Q})\cong H^*_{GL_r} H ∗ ( Rep r Π S g , Q ) ≅ H G L r ∗ .
Note that the restriction maps res U \operatorname{res}_{U} res U in Borel–Moore homology and in cohomology are compatible with the ∙ \bullet ∙ action, i.e., for any c ∈ H ∗ ( M s s ( r , 0 ) , Q ) c \in H^{*}(\mathcal{M}^{\mathrm{ss}}(r,0),\mathbb{Q}) c ∈ H ∗ ( M ss ( r , 0 ) , Q ) and a ∈ H ∗ ( M s s ( r , 0 ) , Q ) a \in H_{*}(\mathcal{M}^{\mathrm{ss}}(r,0),\mathbb{Q}) a ∈ H ∗ ( M ss ( r , 0 ) , Q ) , we have
res U ( c ∙ a ) = res U ( c ) ∙ res U ( a ) . (52) \operatorname{res}_{U}(c \bullet a)=\operatorname{res}_{U}(c)\bullet\operatorname{res}_{U}(a). \tag*{(52)} res U ( c ∙ a ) = res U ( c ) ∙ res U ( a ) . (52)
Proof of the Hausel-Rodriguez-Villegas purity conjecture.
We now have all the ingredients needed to prove Theorem 1.3 . Fix a coprime pair ( r , d ) (r,d) ( r , d ) . Recall that N = ( g − 1 ) r 2 + 1 N=(g-1)r^{2}+1 N = ( g − 1 ) r 2 + 1 . The P=W theorem yields
∑ k dim W 2 k H 2 k ( M B ( r , d ) , Q ) q k = ∑ k dim gr k P H 2 k ( M D ( r , d ) , Q ) q k . \sum_{k}\dim W_{2k}H^{2k}(M^{B}(r,d),\mathbb{Q})q^{k}=\sum_{k}\dim\operatorname{gr}^{P}_{k}H^{2k}(M^{D}(r,d),\mathbb{Q})q^{k}. k ∑ dim W 2 k H 2 k ( M B ( r , d ) , Q ) q k = k ∑ dim gr k P H 2 k ( M D ( r , d ) , Q ) q k .
and by (39) we have
∑ k dim gr k P H 2 k ( M D ( r , d ) , Q ) = A S g , r ( 1 ) . (53) \sum_{k}\dim\operatorname{gr}^{P}_{k}H^{2k}(M^{D}(r,d),\mathbb{Q})=A_{S_{g,r}}(1). \tag*{(53)} k ∑ dim gr k P H 2 k ( M D ( r , d ) , Q ) = A S g , r ( 1 ) . (53)
On the other hand, by Poincaré duality, see (45) , (46) and the χ \chi χ -independence (47) we have
∑ k dim gr k P H 2 k ( M D ( r , d ) , Q ) q k = ∑ ℓ dim gr N − ℓ P G ^ 2 ℓ C ( r , d ) q N − ℓ = ∑ ℓ dim gr N − ℓ P G ^ 2 ℓ C ( r , d ) q N − ℓ = ∑ ℓ dim gr N − ℓ P G ^ 2 ℓ C ( r , 0 ) q N − ℓ = ∑ ℓ dim L N − ℓ G ^ 2 ℓ C ( r , 0 ) q N − ℓ . (54) \begin{aligned}
\sum_{k}\dim\operatorname{gr}^{P}_{k}H^{2k}(M^{D}(r,d),\mathbb{Q})q^{k}
=\sum_{\ell}\dim\operatorname{gr}^{P}_{N-\ell}\widehat{\mathcal{G}}^{C}_{2\ell}(r,d)q^{N-\ell} \\
&=\sum_{\ell}\dim\operatorname{gr}^{P}_{N-\ell}\widehat{\mathcal{G}}^{C}_{2\ell}(r,d)q^{N-\ell} \\
&=\sum_{\ell}\dim\operatorname{gr}^{P}_{N-\ell}\widehat{\mathcal{G}}^{C}_{2\ell}(r,0)q^{N-\ell} \\
&=\sum_{\ell}\dim L_{N-\ell}\widehat{\mathcal{G}}^{C}_{2\ell}(r,0)q^{N-\ell}. \tag*{(54)}
\end{aligned} k ∑ dim gr k P H 2 k ( M D ( r , d ) , Q ) q k = ℓ ∑ dim gr N − ℓ P G 2 ℓ C ( r , d ) q N − ℓ = ℓ ∑ dim gr N − ℓ P G 2 ℓ C ( r , d ) q N − ℓ = ℓ ∑ dim gr N − ℓ P G 2 ℓ C ( r , 0 ) q N − ℓ = ℓ ∑ dim L N − ℓ G 2 ℓ C ( r , 0 ) q N − ℓ . (54)
Next, by the stacky P=C theorem, see (48) , we have
dim L N − ℓ G ^ 2 ℓ C ( r , 0 ) = dim S ‾ C , N − ℓ ∙ [ M s s , r d ( r , 0 ) ] (55) \dim L_{N-\ell}\widehat{\mathcal{G}}^{C}_{2\ell}(r,0)=\dim\overline{S}_{C,N-\ell}\bullet[\mathcal{M}^{\mathrm{ss},\mathrm{rd}}(r,0)] \tag*{(55)} dim L N − ℓ G 2 ℓ C ( r , 0 ) = dim S C , N − ℓ ∙ [ M ss , rd ( r , 0 )] (55)
where
S ‾ C = ⨁ k S ‾ C , k = Q [ c ‾ 2 ( [ p t ] ) , c ‾ 3 ( [ p t ] ) , … , c ‾ r ( [ p t ] ) ] ⊂ S C \overline{S}_{C}=\bigoplus_{k}\overline{S}_{C,k}=\mathbb{Q}[\overline{c}_{2}([\mathrm{pt}]),\overline{c}_{3}([\mathrm{pt}]),\ldots,\overline{c}_{r}([\mathrm{pt}])]\subset S_{C} S C = k ⨁ S C , k = Q [ c 2 ([ pt ]) , c 3 ([ pt ]) , … , c r ([ pt ])] ⊂ S C
and deg ( c ‾ k ( [ p t ] ) ) = k \deg(\overline{c}_{k}([\mathrm{pt}]))=k deg ( c k ([ pt ])) = k for k = 2 , … , r k=2,\ldots,r k = 2 , … , r . Applying the restriction map res U \operatorname{res}_{U} res U of §6.1 and using (50) together with Lemma 6.3 (c), we deduce that for any ℓ \ell ℓ we have
dim S ‾ C , N − ℓ ∙ [ M s s , r d ( r , 0 ) ] ≥ dim S ‾ S g , N − ℓ ∙ [ Rep r r d Π S g ] . (56) \dim\overline{S}_{C,N-\ell}\bullet[\mathcal{M}^{\mathrm{ss},\mathrm{rd}}(r,0)]\geq\dim\overline{S}_{S_{g},N-\ell}\bullet[\operatorname{Rep}^{\mathrm{rd}}_{r}\Pi_{S_{g}}]. \tag*{(56)} dim S C , N − ℓ ∙ [ M ss , rd ( r , 0 )] ≥ dim S S g , N − ℓ ∙ [ Rep r rd Π S g ] . (56)
where
S ‾ S g = ⨁ k S ‾ S g , k = Q [ c ‾ 2 , c ‾ 3 , … , c ‾ r ] ⊂ S S g \overline{S}_{S_{g}}=\bigoplus_{k}\overline{S}_{S_{g},k}=\mathbb{Q}[\overline{c}_{2},\overline{c}_{3},\ldots,\overline{c}_{r}]\subset S_{S_{g}} S S g = k ⨁ S S g , k = Q [ c 2 , c 3 , … , c r ] ⊂ S S g
and deg ( c ‾ k ) = k \deg(\overline{c}_{k})=k deg ( c k ) = k for k = 2 , … , r k=2,\ldots,r k = 2 , … , r . By the quiver version of the Markman Theorem 3.1 we have
S ‾ S g [ c 1 ] ∙ [ Rep r r d Π S g ] = G ^ Π S g ( r ) . \overline{S}_{S_{g}}[c_{1}]\bullet[\operatorname{Rep}^{\mathrm{rd}}_{r}\Pi_{S_{g}}]=\widehat{\mathcal{G}}^{\Pi_{S_{g}}}(r). S S g [ c 1 ] ∙ [ Rep r rd Π S g ] = G Π S g ( r ) .
Because Q [ c 1 ] \mathbb{Q}[c_{1}] Q [ c 1 ] acts freely on G ^ S g ( r ) \widehat{\mathcal{G}}^{S_{g}}(r) G S g ( r ) and G ^ S g ( r ) / c 1 ∙ G ^ S g ( r ) ≅ G ^ Π S g ( r ) \widehat{\mathcal{G}}^{S_{g}}(r)/c_{1}\bullet\widehat{\mathcal{G}}^{S_{g}}(r)\cong\widehat{\mathcal{G}}^{\Pi_{S_{g}}}(r) G S g ( r ) / c 1 ∙ G S g ( r ) ≅ G Π S g ( r ) , see (20) , we deduce that
dim S ‾ S g , N − ℓ ∙ [ Rep r r d Π S g ] ≥ dim G 2 ℓ Π S g ( r ) \dim\overline{S}_{S_{g},N-\ell}\bullet[\operatorname{Rep}^{\mathrm{rd}}_{r}\Pi_{S_{g}}]\geq\dim\mathcal{G}^{\Pi_{S_{g}}}_{2\ell}(r) dim S S g , N − ℓ ∙ [ Rep r rd Π S g ] ≥ dim G 2 ℓ Π S g ( r )
and hence from (15) that
∑ ℓ dim S ‾ S g , N − ℓ ∙ [ Rep r r d Π S g ] q N − ℓ ≥ q N A S g , r ( q − 1 ) , (57) \sum_{\ell}\dim\overline{S}_{S_{g},N-\ell}\bullet[\operatorname{Rep}^{\mathrm{rd}}_{r}\Pi_{S_{g}}]q^{N-\ell}\geq q^{N}A_{S_{g},r}(q^{-1}), \tag*{(57)} ℓ ∑ dim S S g , N − ℓ ∙ [ Rep r rd Π S g ] q N − ℓ ≥ q N A S g , r ( q − 1 ) , (57)
where ≥ \geq ≥ means coefficient-wise inequality. Combining (54) , (55) , (56) and (57) we see that
∑ k dim gr k P H 2 k ( M D ( r , d ) , Q ) q k ≥ q N A S g , r ( q − 1 ) . \sum_{k}\dim\operatorname{gr}^{P}_{k}H^{2k}(M^{D}(r,d),\mathbb{Q})q^{k}\geq q^{N}A_{S_{g},r}(q^{-1}). k ∑ dim gr k P H 2 k ( M D ( r , d ) , Q ) q k ≥ q N A S g , r ( q − 1 ) .
Because of (53) , it must be an equality: indeed, two polynomials P , P ′ ∈ N [ q ] P,P' \in\mathbb{N}[q] P , P ′ ∈ N [ q ] for which P ≥ P ′ P \ge P' P ≥ P ′ and P ( 1 ) = P ′ ( 1 ) P(1)=P'(1) P ( 1 ) = P ′ ( 1 ) must be equal. This finishes the proof of Theorem 1.3 .
AI disclosure
The paper was fully written by us and all the proofs and ideas are ours. AI (Claude and ChatGPT) was used for proofreading purposes and to improve the presentation.
Acknowledgements
We are grateful to J. Bi, B. Davison, L. Hennecart, T. Kinjo, R. Picovschi and N. Seroux for fruitful discussions and correspondence. The quiver version of Markman’s theorem was first announced by Hennecart in the workshop ’Algebraic Geometry Days’ 2026. The proofs were worked out independently and communicated between us and the authors of [9 ] before submission to the arXiv.
Appendix A. Complements on the coproduct on A Π Q A_{\Pi_Q} A Π Q
The aim of this Appendix is to collect and adapt from [12 §2, §6] to the present quiver context some known results and constructions concerning the BPS Lie algebra, its coproduct and the action of tautological classes.
Transfer of the construction of the coproduct to quivers
A.1. Transfer of the construction of the coproduct [12 ] to quivers. We let Q = ( Q 0 , Q 1 ) Q=(Q_0,Q_1) Q = ( Q 0 , Q 1 ) be an arbitrary finite quiver and use the notations from § 2 regarding the stacks of representations of Q Q Q and Π Q \Pi_Q Π Q . We keep the notations from the main body of the article.
Hypotheses of [12, §2.4.1]
A.1.1. Hypotheses of [12 §2.4.1] . The following hold, for any dimension vector d ∈ N Q 0 d \in\mathbb{N}^{Q_0} d ∈ N Q 0 .
(a) R e p d Π Q \mathcal{R}ep_d\Pi_Q R e p d Π Q is a quasi-smooth quasi-compact Artin 1-stack, being the quotient of a derived affine scheme by G L d GL_d G L d .
(b) ⨆ d R e p d Π Q \bigsqcup_d \mathcal{R}ep_d\Pi_Q ⨆ d R e p d Π Q is closed under direct sums and summands, and R e p 0 Π Q = { 0 } \mathcal{R}ep_0\Pi_Q=\{0\} R e p 0 Π Q = { 0 } .
(c) R e p d Π Q \mathcal{R}ep_d\Pi_Q R e p d Π Q has affine diagonal (it is a global quotient of an affine scheme by an affine group).
(d) R e p d Π Q \mathcal{R}ep_d\Pi_Q R e p d Π Q is Θ \Theta Θ -reductive: the stack F i l t d , d ′ Π Q \mathcal{F}ilt_{d,d'}\Pi_Q F i l t d , d ′ Π Q is a closed substack of a bundle of quiver flag varieties, hence the morphism F i l t d , d ′ Π Q → R e p d + d ′ Π Q \mathcal{F}ilt_{d,d'}\Pi_Q \to\mathcal{R}ep_{d+d'}\Pi_Q F i l t d , d ′ Π Q → R e p d + d ′ Π Q is proper.
(e) Each R e p d Π Q \mathcal{R}ep_d\Pi_Q R e p d Π Q is quasi-compact. The graded points are quasi-compact for the same reason.
(f) Each R e p d Π Q \mathcal{R}ep_d\Pi_Q R e p d Π Q has a quasi-compact good moduli space p : R e p d c l Π Q → R e p d Π Q p:\mathcal{R}ep_d^{\mathrm{cl}}\Pi_Q \to\mathcal{R}ep_d\Pi_Q p : R e p d cl Π Q → R e p d Π Q .
The 2CY property of Π Q \Pi_Q Π Q -mod, which fails for Dynkin Q Q Q , is used in [12 ] only to produce the 0-shifted symplectic structure on R e p d Π Q \mathcal{R}ep_d\Pi_Q R e p d Π Q and the orientation data on R e p d Π Q \mathcal{R}ep_d\Pi_Q R e p d Π Q . Here both come for free from the cotangent description. Note that, unlike [12 ] , each R e p d Π Q \mathcal{R}ep_d\Pi_Q R e p d Π Q being quasi-compact, no pro-objects are needed. Further, we do not consider categories of semistable objects, or, rather, we only consider the zero stability condition, i.e., for which every object is semistable.
The above hypotheses (a)-(e) imply the existence of a COHA structure on A Π Q A_{\Pi_Q} A Π Q . Condition (f) implies the existence of a local COHA, BPS sheaves B P S d ∈ M H M ( R e p d Π Q ) \mathcal{BPS}_d \in\mathrm{MHM}(\mathcal{R}ep_d\Pi_Q) B P S d ∈ MHM ( R e p d Π Q ) and a BPS (resp. affinized BPS) Lie algebra g Π Q \mathfrak{g}^{\Pi_Q} g Π Q (resp. g ^ Π Q \widehat{\mathfrak{g}}^{\Pi_Q} g Π Q ) satisfying the absolute integrality (PBW) theorem. More precisely, the inclusion
B P S Π Q ( d ) → p ∗ ( D Q R e p d Π Q ) \mathcal{BPS}^{\Pi_Q}(d) \to p_*(\mathbb{D}\mathbb{Q}_{\mathcal{R}ep_d\Pi_Q}) B P S Π Q ( d ) → p ∗ ( D Q R e p d Π Q )
is defined by dimensional reduction from the inclusion of the lowest perverse cohomology sheaf
p H 1 ( p ~ d ∗ Q R e p d Π Q ) {}^p\mathcal{H}^1\left(\widetilde{p}_d{}_*\mathcal{Q}_{\mathcal{R}ep_d\Pi_Q}\right) p H 1 ( p d ∗ Q R e p d Π Q )
in ( p ~ d ) ∗ Q R e p d Π Q (\widetilde{p}_d)_*\mathcal{Q}_{\mathcal{R}ep_d\Pi_Q} ( p d ) ∗ Q R e p d Π Q , the BPS Lie algebra is then defined as
g Π Q ( d ) = H − ∗ ( R e p d Π Q , B P S Π Q ( d ) ) . \mathfrak{g}^{\Pi_Q}(d)=\mathrm{H}^{-*}\left(\mathcal{R}ep_d\Pi_Q,\mathcal{BPS}^{\Pi_Q}(d)\right). g Π Q ( d ) = H −∗ ( R e p d Π Q , B P S Π Q ( d ) ) .
To define the affinized BPS Lie algebra, we consider the determinant line bundle
L d = ⨂ i ∈ Q 0 det V i \mathcal{L}_d = \bigotimes_{i\in Q_0} \det V_i L d = i ∈ Q 0 ⨂ det V i
where V i ∈ Coh ( Rep d Π Q ) V_i \in\operatorname{Coh}(\operatorname{Rep}_d\Pi_Q) V i ∈ Coh ( Rep d Π Q ) is the tautological rank d i d_i d i vector bundle corresponding to the i i i th vertex. Under the standard G m \mathbb{G}_m G m -action, the weight of det V \det V det V is non-zero, so that det V \det V det V is an equivariant parameter in the sense of [12 §2.5.2] . Slightly abusing notation, we put
u = c 1 ( L d ) ∈ H G L d 2 . u=c_1(\mathcal{L}_d)\in H^2_{G_{\mathcal{L}_d}}. u = c 1 ( L d ) ∈ H G L d 2 .
which we consider as an operator via the cap product on A Π Q A_{\Pi_Q} A Π Q . We now define g ^ Π Q ( d ) \widehat{\mathfrak{g}}^{\Pi_Q}(d) g Π Q ( d ) as the image of the subspace g Π Q ( d ) ⊗ H G m ∗ \mathfrak{g}^{\Pi_Q}(d)\otimes H^*_{\mathbb{G}_m} g Π Q ( d ) ⊗ H G m ∗ under the map
A_{\Pi_Q}(d)\otimes H^*_{\mathbb{G}_m}^{\ell}\longrightarrow A_{\Pi_Q}(d),\qquad x\otimes z^\ell\mapsto u^\ell\mathbin{\bullet}x.
The following microlocal characterization of the affinized BPS Lie algebra is proved as in [12 Cor. 2.25] :
Lemma A.1. We have
g ^ Π Q ( d ) = Im ( H ∗ B M ( Rep d Π Q , Q ) ⟶ A Π Q ( d ) ) . (58) \widehat{\mathfrak{g}}^{\Pi_Q}(d)=\operatorname{Im}\left(H^{\mathrm{BM}}_*(\operatorname{Rep}_d\Pi_Q,\mathbb{Q})\longrightarrow A_{\Pi_Q}(d)\right). \tag*{(58)} g Π Q ( d ) = Im ( H ∗ BM ( Rep d Π Q , Q ) ⟶ A Π Q ( d ) ) . (58)
Coproduct and Milnor-Moore theorem
For an analytic open U ⊂ C U\subset\mathbb{C} U ⊂ C , let
Rep U Π Q ⊂ ∐ d Rep d Π Q \operatorname{Rep}_U\Pi_Q\subset\coprod_d\operatorname{Rep}_d\Pi_Q Rep U Π Q ⊂ d ∐ Rep d Π Q
be the open substack of pairs ( M , θ ) (M,\theta) ( M , θ ) with all eigenvalues of θ \theta θ in U U U . The openness proof of [12 §6.2] applies verbatim. From this, the construction of [12 §6.2] produces a coassociative, cocommutative coproduct Δ \Delta Δ on A Π Q A_{\Pi_Q} A Π Q which is compatible with the product, making A Π Q A_{\Pi_Q} A Π Q into a connected, N Q 0 × Z \mathbb{N}Q_0\times\mathbb{Z} N Q 0 × Z -graded cocommutative bialgebra with finite-dimensional graded pieces. The Euler form ( − , − ) (-,-) ( − , − ) is symmetric, so the sign-twisted braiding of [12 §6.1] is defined. By the Milnor-Moore theorem, we have
A Π Q ≅ U ( A Π Q p r i m ) , A Π Q p r i m = { x ∈ A Π Q ; Δ ( x ) = x ⊗ 1 + 1 ⊗ x } . (59) A_{\Pi_Q}\cong U(A_{\Pi_Q}^{\mathrm{prim}}),\qquad A_{\Pi_Q}^{\mathrm{prim}}=\{x\in A_{\Pi_Q};\ \Delta(x)=x\otimes1+1\otimes x\}. \tag*{(59)} A Π Q ≅ U ( A Π Q prim ) , A Π Q prim = { x ∈ A Π Q ; Δ ( x ) = x ⊗ 1 + 1 ⊗ x } . (59)
The proof of [12 Lem. 6.12] applies verbatim to give the following identification of g ^ Π Q \widehat{\mathfrak{g}}^{\Pi_Q} g Π Q with the Lie algebra of primitive elements.
Lemma A.2. We have g ^ Π Q ( d ) = A Π Q p r i m ( d ) \widehat{\mathfrak{g}}^{\Pi_Q}(d)=A_{\Pi_Q}^{\mathrm{prim}}(d) g Π Q ( d ) = A Π Q prim ( d ) for any d ∈ N Q 0 d\in\mathbb{N}Q_0 d ∈ N Q 0 .
Sketch of the proof of Lemma 2.7
For simplicity, we drop the mention of the dimension vector d d d in the indices. We begin with (a). Recall the induction diagram (7) defining the COHA multiplication. By additivity of the Chern character
ev ∗ ( c ) = gr ∗ ( Δ c ) , c ∈ S Q . \operatorname{ev}^*(c)=\operatorname{gr}^*(\Delta c),\qquad c\in S_Q. ev ∗ ( c ) = gr ∗ ( Δ c ) , c ∈ S Q .
The projection formula then gives
c ∙ ev ∗ gr ! ( x ⊗ y ) = ev ∗ ( gr ∗ ( Δ ( c ) ) ∩ gr ! ( x ⊗ y ) ) = ev ∗ gr ! ( Δ ( c ) ∙ ( x ⊗ y ) ) . c\mathbin{\bullet}\operatorname{ev}_*\operatorname{gr}!(x\otimes y)=\operatorname{ev}_*\left(\operatorname{gr}^*(\Delta(c))\cap\operatorname{gr}!(x\otimes y)\right)=\operatorname{ev}_*\operatorname{gr}!\left(\Delta(c)\mathbin{\bullet}(x\otimes y)\right). c ∙ ev ∗ gr ! ( x ⊗ y ) = ev ∗ ( gr ∗ ( Δ ( c )) ∩ gr ! ( x ⊗ y ) ) = ev ∗ gr ! ( Δ ( c ) ∙ ( x ⊗ y ) ) .
Now, we turn to (b). For U ⊂ A 1 U\subset\mathbb{A}^1 U ⊂ A 1 an analytic open subset, put
( Rep Π Q ) U = { ( x , x ∗ , θ ) ∈ Rep Π Q ; Spec ( θ ) ⊂ U } , DT ( U ) = H ∗ ( ( Rep Π Q ) U , φ Rep Π Q ∣ ( Rep Π Q ) U ) . (\operatorname{Rep}\Pi_Q)_U=\{(x,x^*,\theta)\in\operatorname{Rep}\Pi_Q;\ \operatorname{Spec}(\theta)\subset U\},\qquad\operatorname{DT}(U)=H^*((\operatorname{Rep}\Pi_Q)_U,\varphi_{\operatorname{Rep}\Pi_Q}|_{(\operatorname{Rep}\Pi_Q)_U}). ( Rep Π Q ) U = {( x , x ∗ , θ ) ∈ Rep Π Q ; Spec ( θ ) ⊂ U } , DT ( U ) = H ∗ (( Rep Π Q ) U , φ Rep Π Q ∣ ( Rep Π Q ) U ) .
The coproduct is constructed via a factorization process and is given by the restriction DT ( U ) → DT ( U 1 ⊔ U 2 ) \operatorname{DT}(U)\to\operatorname{DT}(U_1\sqcup U_2) DT ( U ) → DT ( U 1 ⊔ U 2 ) of a the so-called Donaldson-Thomas sheaf and by the identification
( Rep Π Q ) U 1 ⊔ U 2 ≅ ( Rep Π Q ) U 1 × ( Rep Π Q ) U 2 . (60) (\operatorname{Rep}\Pi_Q)_{U_1\sqcup U_2}\cong(\operatorname{Rep}\Pi_Q)_{U_1}\times(\operatorname{Rep}\Pi_Q)_{U_2}. \tag*{(60)} ( Rep Π Q ) U 1 ⊔ U 2 ≅ ( Rep Π Q ) U 1 × ( Rep Π Q ) U 2 . (60)
Under the dimensional reduction, cap product with c c c on A Π Q A_{\Pi Q} A Π Q corresponds to the action of
κ ∗ ( c ) ∈ H ∗ ( R e p Π Q , Q ) \kappa^*(c) \in H^*(\mathop{\mathrm{Rep}}\nolimits_{\Pi Q},\mathbb{Q}) κ ∗ ( c ) ∈ H ∗ ( Rep Π Q , Q )
on H ∗ ( R e p Π Q , φ R e p Π Q ) H^*(\mathop{\mathrm{Rep}}\nolimits_{\Pi Q},\varphi_{\mathop{\mathrm{Rep}}\nolimits_{\Pi Q}}) H ∗ ( Rep Π Q , φ Rep Π Q ) . Restriction commutes with the action of cohomology classes, and under the identification (A.3), κ ∗ c \kappa^*c κ ∗ c restricts to ∑ κ ∗ c ( 1 ) ⊠ κ ∗ c ( 2 ) \sum\kappa^*c_{(1)}\boxtimes\kappa^*c_{(2)} ∑ κ ∗ c ( 1 ) ⊠ κ ∗ c ( 2 ) , where Δ ( c ) = ∑ c ( 1 ) ⊗ c ( 2 ) \Delta(c)=\sum c_{(1)}\otimes c_{(2)} Δ ( c ) = ∑ c ( 1 ) ⊗ c ( 2 ) . We have again used additivity of the Chern character, this time applied to V = V U 1 ⊕ V U 2 V=V_{U_1}\oplus V_{U_2} V = V U 1 ⊕ V U 2 .
Finally, we deal with (c). If x x x is primitive, then by (b) we have
Δ ( c ∙ x ) = ∑ ( c ( 1 ) ∙ x ) ⊗ ( c ( 2 ) ∙ 1 ) + ∑ ( c ( 1 ) ∙ 1 ) ⊗ ( c ( 2 ) ∙ x ) . \Delta(c\mathbin{\bullet}x)=\sum(c_{(1)}\mathbin{\bullet}x)\otimes(c_{(2)}\mathbin{\bullet}1)+\sum(c_{(1)}\mathbin{\bullet}1)\otimes(c_{(2)}\mathbin{\bullet}x). Δ ( c ∙ x ) = ∑ ( c ( 1 ) ∙ x ) ⊗ ( c ( 2 ) ∙ 1 ) + ∑ ( c ( 1 ) ∙ 1 ) ⊗ ( c ( 2 ) ∙ x ) .
Since A Π Q , 0 = Q A_{\Pi Q,0}=\mathbb{Q} A Π Q , 0 = Q and S Q S_Q S Q acts on it through the counit, the rhs is equal to
( c ∙ x ) ⊗ 1 + 1 ⊗ ( c ∙ x ) (c\mathbin{\bullet}x)\otimes1+1\otimes(c\mathbin{\bullet}x) ( c ∙ x ) ⊗ 1 + 1 ⊗ ( c ∙ x )
as wanted. □
Quasi-smooth stacks with smooth truncation
Conventions
We work over a field k k k of characteristic 0. Derived stacks are derived Artin stacks locally almost of finite presentation over k k k , modelled on commutative simplicial algebras or dg-algebras. Complexes are graded cohomologically, so h j ( − ) = π − j ( − ) h^j(-)=\pi_{-j}(-) h j ( − ) = π − j ( − ) . A complex F F F is n n n -connective if h j ( F ) = 0 h^j(F)=0 h j ( F ) = 0 for j > − n j>-n j > − n . It is connective if it is 0-connective. Let A A A be a connective simplicial commutative ring. An A A A -module N N N is discrete if its homotopy is concentrated in degree 0. Likewise, a quasi-coherent sheaf on a derived stack Y Y Y is discrete if its cohomology is concentrated in degree 0. If N N N is a discrete A A A -module, the A A A -action factors through π 0 A \pi_0A π 0 A : the higher homotopy groups π n A \pi_nA π n A , n ≥ 1 n\geq1 n ≥ 1 , act through maps π n A ⊗ π 0 N → π n N = 0 \pi_nA\otimes\pi_0N\to\pi_nN=0 π n A ⊗ π 0 N → π n N = 0 . The functor N ↦ π 0 N N\mapsto\pi_0N N ↦ π 0 N gives an equivalence
{ discrete A -modules } ≃ { ordinary π 0 A -modules } \{\text{discrete }A\text{-modules}\}\simeq\{\text{ordinary }\pi_0A\text{-modules}\} { discrete A -modules } ≃ { ordinary π 0 A -modules }
Similarly, discrete quasi-coherent sheaves on Y Y Y are the same as ordinary quasi-coherent sheaves on the classical truncation t 0 Y t_0Y t 0 Y .
A perfect complex E E E on a derived stack Y Y Y has Tor-amplitude in [ a , b ] [a,b] [ a , b ] if for every discrete quasi-coherent sheaf N N N on t 0 Y t_0Y t 0 Y (or, smooth locally, every discrete module N N N over π 0 A \pi_0A π 0 A for an affine chart Spec A → Y \operatorname{Spec}A\to Y Spec A → Y ) the complex E ⊗ L N E\otimes^{\mathbf L}N E ⊗ L N has cohomology concentrated in degrees [ a , b ] [a,b] [ a , b ] . If E E E has Tor-amplitude in [ a , b ] [a,b] [ a , b ] , then h j ( E ) = 0 h^j(E)=0 h j ( E ) = 0 for j > b j>b j > b . If moreover Y Y Y is classical, then (taking N = O Y N=\mathcal{O}_Y N = O Y ) the cohomology sheaves of E E E are concentrated in [ a , b ] [a,b] [ a , b ] . This fails for derived Y Y Y : for Y = Spec A Y=\operatorname{Spec}A Y = Spec A with A = k [ ϵ ] A=k[\epsilon] A = k [ ϵ ] and ϵ \epsilon ϵ of homological degree 1, the complex O Y \mathcal{O}_Y O Y has Tor-amplitude in [ 0 , 0 ] [0,0] [ 0 , 0 ] but h − 1 ( O Y ) = π 1 A = k ≠ 0 h^{-1}(\mathcal{O}_Y)=\pi_1A=k\ne0 h − 1 ( O Y ) = π 1 A = k = 0 . Tor-amplitude in [ 0 , 0 ] [0,0] [ 0 , 0 ] means Tor j ( E , N ) = 0 \operatorname{Tor}_j(E,N)=0 Tor j ( E , N ) = 0 for j > 0 j>0 j > 0 , i.e., E E E is flat, hence, since E E E is perfect it is a vector bundle in degree 0. Tor-amplitude in [ − 1 , 0 ] [-1,0] [ − 1 , 0 ] means that E E E is locally quasi-isomorphic to a two-term complex of vector bundles E − 1 → E 0 E^{-1}\to E^0 E − 1 → E 0 .
A perfect complex E E E on a connected Y Y Y let χ ( E ) ∈ Z \chi(E)\in\mathbb{Z} χ ( E ) ∈ Z be its rank (=Euler characteristic). A derived stack X X X is quasi-smooth if its cotangent complex L X \mathbb{L}_X L X is perfect of Tor-amplitude in [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] . Its virtual dimension is the locally constant function vdim X = χ ( L X ) \operatorname{vdim}X=\chi(\mathbb{L}_X) vdim X = χ ( L X ) . Let i : X c l ↪ X i:X^{\mathrm{cl}}\hookrightarrow X i : X cl ↪ X be the inclusion of the classical truncation. The complex E = i ∗ L X E=i^*\mathbb{L}_X E = i ∗ L X is perfect of Tor-amplitude in [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] , and we let
ϕ : E ⟶ L X c l \phi:E\longrightarrow\mathbb{L}_{X^{\mathrm{cl}}} ϕ : E ⟶ L X cl
be the natural map. It is the (perfect) obstruction theory induced on X c l X^{\mathrm{cl}} X cl [39 ] .
Theorem B.1. Let X X X be a quasi-smooth derived stack. Suppose that M = X c l M=X^{\mathrm{cl}} M = X cl is smooth over k k k with dim M = vdim X \dim M=\operatorname{vdim}X dim M = vdim X on every connected component. Then
(a) ϕ : E → L M \phi:E\to\mathbb{L}_M ϕ : E → L M is an equivalence,
(b) i : M → X i:M\to X i : M → X is an equivalence,
(c) the virtual fundamental class of X X X is the fundamental class, i.e., [ X ] v i r = [ M ] [X]^{\mathrm{vir}}=[M] [ X ] vir = [ M ] .
Throughout we assume M M M is connected, so that dim M \dim M dim M and vdim X \operatorname{vdim}X vdim X are integers.
Lemmas
Lemma B.2. Let f : A → B f:A\to B f : A → B be a map of connective commutative dg-algebras over k k k . Suppose the mapping cone cofib ( f ) \operatorname{cofib}(f) cofib ( f ) is m m m -connective for some m ≥ 1 m\ge1 m ≥ 1 . Then L B / A \mathbb{L}_{B/A} L B / A is m m m -connective, and the natural map
B ⊗ A cofib ( f ) ⟶ L B / A B\otimes_A\operatorname{cofib}(f)\longrightarrow\mathbb{L}_{B/A} B ⊗ A cofib ( f ) ⟶ L B / A
has ( m + 1 ) (m+1) ( m + 1 ) -connective mapping cone. In particular
π m L B / A ≅ π m ( B ⊗ A cofib ( f ) ) ≅ π 0 B ⊗ π 0 A π m cofib ( f ) . \pi_m\mathbb{L}_{B/A}\cong\pi_m(B\otimes_A\operatorname{cofib}(f))\cong\pi_0B\otimes_{\pi_0A}\pi_m\operatorname{cofib}(f). π m L B / A ≅ π m ( B ⊗ A cofib ( f )) ≅ π 0 B ⊗ π 0 A π m cofib ( f ) .
Proof. This is [25 Thm. 7.4.3.1] and, in the simplicial setting, [41 §2.2.2] . □
Lemma B.3. For any derived stack X X X , the relative cotangent complex L X c l / X \mathbb{L}_{X^{\mathrm{cl}}/X} L X cl / X is 2 2 2 -connective. Consequently ϕ : E → L X c l \phi:E\to\mathbb{L}_{X^{\mathrm{cl}}} ϕ : E → L X cl induces isomorphisms on h j h^j h j for j ≥ 0 j\ge0 j ≥ 0 and a surjection on h − 1 h^{-1} h − 1 .
Proof. The question is smooth local on X X X because for a smooth chart U → X U\to X U → X we have U c l = U × X X c l U^{\mathrm{cl}}=U\times_X X^{\mathrm{cl}} U cl = U × X X cl and L U c l / U \mathbb{L}_{U^{\mathrm{cl}}/U} L U cl / U is the pullback of L X c l / X \mathbb{L}_{X^{\mathrm{cl}}/X} L X cl / X . So we may assume
X = Spec A , X c l = Spec π 0 A , f : A → π 0 A . X=\operatorname{Spec}A,\qquad X^{\mathrm{cl}}=\operatorname{Spec}\pi_0A,\qquad f:A\to\pi_0A. X = Spec A , X cl = Spec π 0 A , f : A → π 0 A .
The mapping cone of f f f is cofib ( f ) = τ ≥ 1 A [ 1 ] \operatorname{cofib}(f)=\tau_{\ge1}A[1] cofib ( f ) = τ ≥ 1 A [ 1 ] . It is 2 2 2 -connective. Lemma B.2 with m = 2 m=2 m = 2 gives the first claim. The second follows from the long exact sequence of the sequence E → L X c l → L X c l / X E\to\mathbb{L}_{X^{\mathrm{cl}}}\to\mathbb{L}_{X^{\mathrm{cl}}/X} E → L X cl → L X cl / X . □
Lemma B.4. Let X X X be a derived stack with L X c l / X ≅ 0 \mathbb{L}_{X^{\mathrm{cl}}/X}\cong0 L X cl / X ≅ 0 . Then i : X c l → X i:X^{\mathrm{cl}}\to X i : X cl → X is an equivalence.
Proof. Being an equivalence is smooth local on X X X and passing to the classical truncation commutes with smooth base change, so we may assume X = Spec A X=\operatorname{Spec}A X = Spec A , and we must show π n A = 0 \pi_nA=0 π n A = 0 for all n ≥ 1 n\ge1 n ≥ 1 . Suppose inductively that π k A = 0 \pi_kA=0 π k A = 0 for 1 ≤ k < n 1\le k<n 1 ≤ k < n . Then τ ≥ 1 A = τ ≥ n A \tau_{\ge1}A=\tau_{\ge n}A τ ≥ 1 A = τ ≥ n A , so cofib ( A → π 0 A ) = τ ≥ n A [ 1 ] \operatorname{cofib}(A\to\pi_0A)=\tau_{\ge n}A[1] cofib ( A → π 0 A ) = τ ≥ n A [ 1 ] is ( n + 1 ) (n+1) ( n + 1 ) -connective with π n + 1 = π n A \pi_{n+1}=\pi_nA π n + 1 = π n A . By Lemma B.2 with m = n + 1 m=n+1 m = n + 1 ,
0 = π n + 1 L π 0 A / A ≅ π 0 A ⊗ π 0 A π n A = π n A . 0=\pi_{n+1}\mathbb{L}_{\pi_0A/A}\cong\pi_0A\otimes_{\pi_0A}\pi_nA=\pi_nA. 0 = π n + 1 L π 0 A / A ≅ π 0 A ⊗ π 0 A π n A = π n A .
Hence π n A = 0 \pi_nA=0 π n A = 0 for all n ≥ 1 n\ge1 n ≥ 1 , i.e., the map A → π 0 A A\to\pi_0A A → π 0 A is an equivalence. □
Lemma B.5. Let M M M be a classical stack. Let C C C be a perfect complex on M M M of Tor-amplitude in [ a , b ] [a,b] [ a , b ] whose cohomology vanishes in degrees > a >a > a . We have C ≅ E [ − a ] C\cong\mathcal{E}[-a] C ≅ E [ − a ] for a locally free sheaf E \mathcal{E} E of finite rank, and χ ( C ) = ( − 1 ) a rk E \chi(C)=(-1)^a\operatorname{rk}\mathcal{E} χ ( C ) = ( − 1 ) a rk E .
Proof. Since cohomology is concentrated in [ a , b ] [a,b] [ a , b ] and vanishes above a a a , the complex C C C has a single cohomology sheaf E = h a ( C ) \mathcal{E}=h^a(C) E = h a ( C ) . It is coherent because C C C is perfect. Thus C ≅ E [ − a ] C\cong\mathcal{E}[-a] C ≅ E [ − a ] . For any discrete N N N we have
h a − j ( C ⊗ L N ) = Tor j ( E , N ) , h^{a-j}(C\otimes^{\mathbb{L}}N)=\operatorname{Tor}_j(\mathcal{E},N), h a − j ( C ⊗ L N ) = Tor j ( E , N ) ,
and the Tor-amplitude bound ≥ a \ge a ≥ a forces Tor j ( E , N ) = 0 \operatorname{Tor}_j(\mathcal{E},N)=0 Tor j ( E , N ) = 0 for j > 0 j>0 j > 0 . So E \mathcal{E} E is flat and a flat coherent sheaf on a scheme is locally free of finite rank. Finally χ ( E [ − a ] ) = ( − 1 ) a rk E \chi(\mathcal{E}[-a])=(-1)^a\operatorname{rk}\mathcal{E} χ ( E [ − a ]) = ( − 1 ) a rk E . □
Proof of Theorem B.1
Let d = dim M = vdim X d=\dim M=\operatorname{vdim}X d = dim M = vdim X . Since M M M is a smooth Artin stack of dimension d d d , the cotangent complex L M \mathbb{L}_M L M is perfect of Tor-amplitude in [ 0 , 1 ] [0,1] [ 0 , 1 ] of rank χ ( L M ) = d \chi(\mathbb{L}_M)=d χ ( L M ) = d . Consider the sequence of perfect complexes on M M M
E → ϕ L M ⟶ L M / X . (61) \mathcal{E}\xrightarrow{\phi}\mathbb{L}_M\longrightarrow\mathbb{L}_{M/X}. \tag*{(61)} E ϕ L M ⟶ L M / X . (61)
From (B.1) we get a sequence
L M ⟶ L M / X ⟶ E [ 1 ] . \mathbb{L}_M\longrightarrow\mathbb{L}_{M/X}\longrightarrow\mathcal{E}[1]. L M ⟶ L M / X ⟶ E [ 1 ] .
Here L M \mathbb{L}_M L M has Tor-amplitude in [ 0 , 1 ] [0,1] [ 0 , 1 ] and E [ 1 ] \mathcal{E}[1] E [ 1 ] has Tor-amplitude in [ − 2 , 0 ] [-2,0] [ − 2 , 0 ] . The Tor-amplitude is preserved by extensions, so L M / X \mathbb{L}_{M/X} L M / X is perfect of Tor-amplitude in [ − 2 , 1 ] [-2,1] [ − 2 , 1 ] . By Lemma B.3, we have
h j ( L M / X ) = 0 ∀ j > − 2. h^j(\mathbb{L}_{M/X})=0\qquad\forall j>-2. h j ( L M / X ) = 0 ∀ j > − 2.
By Lemma B.5 with a = − 2 a=-2 a = − 2 , we have L M / X ≅ E [ 2 ] \mathbb{L}_{M/X}\cong\mathcal{E}[2] L M / X ≅ E [ 2 ] with E \mathcal{E} E locally free, and the additivity of χ \chi χ in (B.1) yields
rk E = χ ( L M / X ) = χ ( L M ) − χ ( E ) = d − vdim X = 0. \operatorname{rk}\mathcal{E}=\chi(\mathbb{L}_{M/X})=\chi(\mathbb{L}_M)-\chi(\mathcal{E})=d-\operatorname{vdim}X=0. rk E = χ ( L M / X ) = χ ( L M ) − χ ( E ) = d − vdim X = 0.
A locally free sheaf of rank 0 is zero, so L M / X ≅ 0 \mathbb{L}_{M/X}\cong0 L M / X ≅ 0 and ϕ \phi ϕ is an equivalence. This proves the part (a).
Since L M / X ≅ 0 \mathbb{L}_{M/X}\cong0 L M / X ≅ 0 , Lemma B.4 shows that i : M = X c l → X i\colon M=X^{\mathrm{cl}}\to X i : M = X cl → X is an equivalence. This proves the part (b).
The virtual structure sheaf of a quasi-smooth derived stack is
[ O X v i r ] = ∑ n ≥ 0 ( − 1 ) n [ π n O X ] ∈ K 0 ( X c l ) [\mathcal{O}^{\mathrm{vir}}_X]=\sum_{n\geq0}(-1)^n[\pi_n\mathcal{O}_X]\in K_0(X^{\mathrm{cl}}) [ O X vir ] = n ≥ 0 ∑ ( − 1 ) n [ π n O X ] ∈ K 0 ( X cl )
By (b) we have π n O X = 0 \pi_n\mathcal{O}_X=0 π n O X = 0 for n ≥ 1 n\geq1 n ≥ 1 and π 0 O X = O M \pi_0\mathcal{O}_X=\mathcal{O}_M π 0 O X = O M . We deduce that [ O X v i r ] = [ O M ] [\mathcal{O}^{\mathrm{vir}}_X]=[\mathcal{O}_M] [ O X vir ] = [ O M ] in K 0 ( M ) K_0(M) K 0 ( M ) . We now prove part (c) in Chow groups. First let M M M be Deligne-Mumford. By [1 ] , the virtual class attached to the perfect obstruction theory ϕ : E → L M \phi\colon\mathcal{E}\to\mathbb{L}_M ϕ : E → L M is
[ X ] v i r = 0 E ! [ C M ] , E = h 1 / h 0 ( E ∨ ) , [X]^{\mathrm{vir}}=0^!_{\mathcal{E}}[\mathcal{C}_M],\qquad\mathcal{E}=h^1/h^0(\mathcal{E}^{\vee}), [ X ] vir = 0 E ! [ C M ] , E = h 1 / h 0 ( E ∨ ) ,
where
C M ⊂ N M = h 1 / h 0 ( T M ) \mathcal{C}_M\subset\mathcal{N}_M=h^1/h^0(\mathbb{T}_M) C M ⊂ N M = h 1 / h 0 ( T M )
is the intrinsic normal cone of M M M and C M ↪ E \mathcal{C}_M\hookrightarrow\mathcal{E} C M ↪ E is the closed embedding induced by ϕ \phi ϕ . Since M M M is smooth, we have
C M = N M = h 1 / h 0 ( T M ) ≅ B T M , \mathcal{C}_M=\mathcal{N}_M=h^1/h^0(\mathbb{T}_M)\cong BT_M, C M = N M = h 1 / h 0 ( T M ) ≅ B T M ,
and by (a) the embedding C M → E \mathcal{C}_M\to\mathcal{E} C M → E is the identity of B T M BT_M B T M . Concretely, take the global resolution
E ∨ ≅ [ E 0 → E 1 ] \mathcal{E}^{\vee}\cong[E_0\to E_1] E ∨ ≅ [ E 0 → E 1 ]
with E 0 = T M E_0=T_M E 0 = T M and E 1 = 0 E_1=0 E 1 = 0 . Hence, the lifted cone C 1 ⊂ E 1 = 0 C_1\subset E_1=0 C 1 ⊂ E 1 = 0 vanishes, i.e., we have C 1 = M C_1=M C 1 = M , and
[ X ] v i r = 0 E 1 ! [ C 1 ] = 0 M ! [ M ] = [ M ] . [X]^{\mathrm{vir}}=0^!_{E_1}[C_1]=0^!_M[M]=[M]. [ X ] vir = 0 E 1 ! [ C 1 ] = 0 M ! [ M ] = [ M ] .
Equivalently, when M M M is smooth and the obstruction sheaf O b = h 1 ( E ∨ ) \mathrm{Ob}=h^1(\mathcal{E}^{\vee}) Ob = h 1 ( E ∨ ) is locally free one has
[ X ] v i r = e ( O b ) ∩ [ M ] [X]^{\mathrm{vir}}=e(\mathrm{Ob})\cap[M] [ X ] vir = e ( Ob ) ∩ [ M ]
by [1 §5] . Here E ∨ ≅ T M \mathcal{E}^{\vee}\cong T_M E ∨ ≅ T M by (a), so O b = 0 \mathrm{Ob}=0 Ob = 0 and e ( O b ) = 1 e(\mathrm{Ob})=1 e ( Ob ) = 1 .
For a general Artin stack, the virtual fundamental class of a quasi-smooth derived stack is constructed in [22 ] by deformation to the normal bundle and Gysin pullback along the 1-shifted cotangent bundle. By (b), the stack X ≅ M X\cong M X ≅ M is classical and smooth. For smooth X X X this construction returns the fundamental class [ X ] [X] [ X ] , see [22 ] . Hence [ X ] v i r = [ M ] [X]^{\mathrm{vir}}=[M] [ X ] vir = [ M ] . □
Corollary B.6. Let X X X be a quasi-smooth derived stack whose classical truncation M = X c l M=X^{\mathrm{cl}} M = X cl is quasicompact. Assume that
(a) M M M is of pure dimension vdim X \operatorname{vdim}X vdim X
(b) there is a dense open substack M ∘ ⊂ M M^\circ\subset M M ∘ ⊂ M which is smooth.
Then [ X ] v i r = [ M r e d ] [X]^{\mathrm{vir}}=[M_{\mathrm{red}}] [ X ] vir = [ M red ] in H 2 vdim X ( M , Q H_{2\operatorname{vdim}X}(M,\mathbb{Q} H 2 vdim X ( M , Q ).
Proof. Set d = vdim X d=\operatorname{vdim}X d = vdim X and let X ∘ ⊂ X X^\circ\subset X X ∘ ⊂ X be the open substack with ( X c l ) ∘ = M ∘ (X^{\mathrm{cl}})^\circ=M^\circ ( X cl ) ∘ = M ∘ . Then X ∘ X^\circ X ∘ is quasi-smooth of virtual dimension d d d , and by (i) and (ii) its truncation M ∘ M^\circ M ∘ is smooth of dimension d d d . By Theorem B.1 (c), [ X ∘ ] v i r = [ M ∘ ] [X^\circ]^{\mathrm{vir}}=[M^\circ] [ X ∘ ] vir = [ M ∘ ] . Since virtual fundamental classes are compatible with restriction to open substacks [22 ] , we get
[ X ] v i r ∣ M ∘ = [ M ∘ ] = [ M r e d ] ∣ M ∘ . [X]^{\mathrm{vir}}|_{M^\circ}=[M^\circ]=[M_{\mathrm{red}}]|_{M^\circ}. [ X ] vir ∣ M ∘ = [ M ∘ ] = [ M red ] ∣ M ∘ .
Let Z = M ∖ M ∘ Z=M\setminus M^\circ Z = M ∖ M ∘ with its reduced structure. By (i) and the density of M ∘ M^\circ M ∘ , we have dim Z ≤ d − 1 \dim Z\leq d-1 dim Z ≤ d − 1 , hence H j ( Z , Q ) = 0 H_j(Z,\mathbb{Q})=0 H j ( Z , Q ) = 0 for j > 2 d − 2 j>2d-2 j > 2 d − 2 . The localization sequence
H 2 d ( Z , Q ) ⟶ H 2 d ( M , Q ) ⟶ H 2 d ( M ∘ , Q ) ⟶ H 2 d − 1 ( Z , Q ) H_{2d}(Z,\mathbb{Q})\longrightarrow H_{2d}(M,\mathbb{Q})\longrightarrow H_{2d}(M^\circ,\mathbb{Q})\longrightarrow H_{2d-1}(Z,\mathbb{Q}) H 2 d ( Z , Q ) ⟶ H 2 d ( M , Q ) ⟶ H 2 d ( M ∘ , Q ) ⟶ H 2 d − 1 ( Z , Q )
shows that restriction H 2 d ( M , Q ) → H 2 d ( M ∘ , Q ) H_{2d}(M,\mathbb{Q})\to H_{2d}(M^\circ,\mathbb{Q}) H 2 d ( M , Q ) → H 2 d ( M ∘ , Q ) is an isomorphism, and the claim follows. □
Tautological classes and the χ \chi χ -independence operator
We sketch a direct proof of (49) based on [32 ] and [12 ] . We begin with some recollection about the global BPS Lie algebra defined in [12 ] , to which we refer for details. Consider the stack M ( r , d ) ⊇ M s s ( r , d ) \mathcal{M}(r,d) \supseteq\mathcal{M}^{\mathrm{ss}}(r,d) M ( r , d ) ⊇ M ss ( r , d ) parametrizing all Higgs sheaves on C C C of rank r r r and degree d d d . One may consider its Borel-Moore homology spaces H i ( M ( r , d ) , Q ) H_i(\mathcal{M}(r,d),\mathbb{Q}) H i ( M ( r , d ) , Q ) and equip the direct sum
A C : = ⨁ r , d H ∗ + 2 ( g − 1 ) r 2 ( M ( r , d ) , Q ) \mathcal{A}_C := \bigoplus_{r,d} H_{\ast+2(g-1)r^2}(\mathcal{M}(r,d),\mathbb{Q}) A C := r , d ⨁ H ∗+ 2 ( g − 1 ) r 2 ( M ( r , d ) , Q )
with the structure of an associative algebra structure. One also defines a compatible comultiplication
Δ : A C ⟶ A C ⊗ ^ A C \Delta:\mathcal{A}_C\longrightarrow\mathcal{A}_C\widehat{\otimes}\mathcal{A}_C Δ : A C ⟶ A C ⊗ A C
and considers the Lie algebra of primitive elements1 ^{1} 1
h C : = ⨁ r , d h C ( r , d ) , h C ( r , d ) = A C ( r , d ) p r i m . \mathfrak{h}^C:=\bigoplus_{r,d}\mathfrak{h}^C(r,d),\qquad\mathfrak{h}^C(r,d)=\mathcal{A}_C(r,d)^{\mathrm{prim}}. h C := r , d ⨁ h C ( r , d ) , h C ( r , d ) = A C ( r , d ) prim .
The restriction map induces for every pair ( r , d ) (r,d) ( r , d ) an isomorphism of vector spaces
h C ( r , d ) ≃ g ^ C ( r , d ) . \mathfrak{h}^C(r,d)\simeq\widehat{\mathfrak{g}}^C(r,d). h C ( r , d ) ≃ g C ( r , d ) .
The χ \chi χ -independence operator is defined by
T 0 ( [ p t ] ) : = ad D 1 , 0 ( [ p t ] ) , D 1 , 0 ( [ p t ] ) = [ p t ] ∩ [ M c l ( 0 , 1 ) ] ∈ h C ( 0 , 1 ) . T_0([\mathrm{pt}]):=\operatorname{ad}D_{1,0}([\mathrm{pt}]),\qquad D_{1,0}([\mathrm{pt}])=[\mathrm{pt}]\cap[\mathcal{M}^{\mathrm{cl}}(0,1)]\in\mathfrak{h}_C(0,1). T 0 ([ pt ]) := ad D 1 , 0 ([ pt ]) , D 1 , 0 ([ pt ]) = [ pt ] ∩ [ M cl ( 0 , 1 )] ∈ h C ( 0 , 1 ) .
It induces an isomorphism of vector spaces
T 0 ( [ p t ] ) : g ^ C ( r , d ) ≃ h C ( r , d ) → ∼ h C ( r , d + 1 ) ≃ g ^ C ( r , d + 1 ) . T_0([\mathrm{pt}]):\widehat{\mathfrak{g}}^C(r,d)\simeq\mathfrak{h}^C(r,d)\xrightarrow{\sim}\mathfrak{h}^C(r,d+1)\simeq\widehat{\mathfrak{g}}^C(r,d+1). T 0 ([ pt ]) : g C ( r , d ) ≃ h C ( r , d ) ∼ h C ( r , d + 1 ) ≃ g C ( r , d + 1 ) .
Furthermore, from the explicit description of the structure of the COHA of zero-dimensional sheaves (or a direct computation) it follows that for any i ≥ 1 i\geq1 i ≥ 1
[ T 0 ( [ p t ] ) , c i ( [ p t ] ) ] = 0 [T_0([\mathrm{pt}]),c_i([\mathrm{pt}])]=0 [ T 0 ([ pt ]) , c i ([ pt ])] = 0
from which it also follows that [ T 0 ( [ p t ] ) , c ‾ i ( [ p t ] ) ] = 0 [T_0([\mathrm{pt}]),\overline{c}_i([\mathrm{pt}])]=0 [ T 0 ([ pt ]) , c i ([ pt ])] = 0 (note that this is false for general classes c i ( γ ) c_i(\gamma) c i ( γ ) ).
We now explain the proof of (49) . Fix ( r , d ) (r,d) ( r , d ) and first assume that gcd ( r , d ) = 1 \gcd(r,d)=1 g cd( r , d ) = 1 . In that case we have g ^ C ( r , d ) = H ∗ ( M s s ( r , d ) , Q ) \widehat{\mathfrak{g}}^C(r,d)=H_\ast(\mathcal{M}^{\mathrm{ss}}(r,d),\mathbb{Q}) g C ( r , d ) = H ∗ ( M ss ( r , d ) , Q ) and g C ( r , d ) = H ∗ ( M D ( r , d ) , Q ) \mathfrak{g}^C(r,d)=H_\ast(\mathcal{M}^{D}(r,d),\mathbb{Q}) g C ( r , d ) = H ∗ ( M D ( r , d ) , Q ) and (5.10) follows from Poincaré duality together with the fact2 ^{2} 2 proven in [13 ] that the perverse degree of c ‾ k ( [ p t ] ) \overline{c}_k([\mathrm{pt}]) c k ([ pt ]) is equal to k k k . The case of a general pair ( r , d ) (r,d) ( r , d ) is now a consequence of the facts that the χ \chi χ -independence operator commutes with c ‾ k ( [ p t ] ) \overline{c}_k([\mathrm{pt}]) c k ([ pt ]) and that (for degree reasons) it maps [ M s s , r d ( r , d ) ] [\mathcal{M}^{\mathrm{ss},rd}(r,d)] [ M ss , r d ( r , d )] to a nonzero multiple of [ M s s , r d ( r , d + 1 ) ] [\mathcal{M}^{\mathrm{ss},rd}(r,d+1)] [ M ss , r d ( r , d + 1 )] .
References
O. Schiffmann: Laboratoire de Mathématiques d’Orsay, Bât 307, Faculté des Sciences d’Orsay, Université Paris-Saclay, 91405 Orsay, France and Simon Stoilow Institute of Mathematics, P.O. Box 1-764, RO-014700 Bucharest, Romania Email address: olivier.schiffmann@universite-paris-saclay.fr
E. Vasserot: IMJ-PRG, UMR7586, Université Paris-Cité, 75013 Paris, France Email address: vasserot@imj-prg.fr