ARITHMETIC MIRROR SYMMETRY FOR GENUS 1 CURVES WITH n MARKED POINTS YANKI LEKILI ALEXANDER POLISHCHUK 6 1 Abstract. Weestablisha Z[[t1,...,tn]]-linearderivedequivalencebetweentherelative 0 Fukaya category of the 2-torus with n distinct marked points and the derived category 2 of perfect complexes on the n-Tate curve. Specialising to t = ... = t = 0 gives a 1 n t Z-linear derived equivalence between the Fukaya category of the n-punctured torus and c thederivedcategoryofperfectcomplexesonthestandard(N´eron) n-gon. Weprovethat O this equivalence extends to a Z-linear derived equivalence between the wrapped Fukaya 3 category of the n-punctured torus and the derived category of coherent sheaves on the 1 standard n-gon. The corresponding results for n=1 were established in [31]. ] G Introduction S . h Over the past few decades, Kontsevich’s Homological Mirror Symmetry (HMS) conjecture t a ([28]), which predicts a remarkable equivalence between an A-model category associated m to a symplectic manifold and a B-model category associated to a complex manifold, has [ beenstudiedextensivelyinmanyinstances: abelianvarieties,toricvarieties,hypersurfaces 4 in projective spaces, etc. From the beginning, the case of the symplectic 2-torus played v a special role as this is the simplest instance of a Calabi-Yau manifold where concrete 1 4 computations in the Fukaya category can be made, and the non-trivial nature of the HMS 1 conjecture becomes manifest. Calculations provided in [38], [39], [40] and [41] lead to a 6 proof of the HMS for the 2-torus over a Novikov field (see [4] for a streamlined exposition). 0 . Recently, various versions of the HMS have been verified when the A-model category is 1 0 a flavour of the Fukaya category of a, possibly non-compact, symplectic 2-manifold (see 6 for example [46], [15], [1], [7], [30]). 1 : In [31], the authors have explored an arithmetic refinement of the homological mirror v i symmetry which relates exact symplectic topology to arithmetic algebraic geometry. Con- X cretely, they proved a derived equivalence of the Fukaya category of the 2-torus, relative r a to a base-point D = {z}, with the category of perfect complexes of coherent sheaves on the Tate curve over the formal disc Spec Z[[t]]. Setting t = 0, one obtains a derived equivalence over Z of the Fukaya category of the punctured torus with perfect complexes on the nodal projective cubic y2z +xyz = x3 in P2. Z The main result of this paper is a generalization of the result of [31], where we work with the Fukaya category of the 2-torus relative to a divisor D = {z ,...,z } for n > 1. Thus, 1 n on the symplectic side we consider the relative Fukaya category F(T,D) of a symplectic 2-torus T with n distinct marked points z ,...,z (see Sec. 3). On the B-side we define a 1 n Y.L. is supported in part by the Royal Society and the NSF grant DMS-1509141. A.P. is supported in part by the NSF grant DMS-1400390. 1 certain family of curves T over Z[[t ,...,t ]], which we call the n-Tate curve (for n = 1 n 1 n we get the usual Tate curve; specializing to t = ... = t we get the n-gon Tate curve of 1 n [12, VII]). The generic fiber of T is a smooth elliptic curve, while changing the base to Z n (by setting t = 0) we get the curve G over Z, which we call following [12] the standard i n n-gon (or N´eron n-gon) over Z. For n > 1 this is the wheel of n projective lines over Z, while for n = 1 this is the projective line with 0 and ∞ identified. Theorem A. There is a Z[[t ,...,t ]]-linear equivalence of triangulated categories be- 1 n tween the split-closed derived Fukaya category DπF(T,D) of the 2-torus with n marked points and the derived category of perfect complexes Perf(T ) on the n-Tate curve T . n n The key ingredient in the proof is the isomorphism between the moduli space of minimal A -structures on a certain finite-dimensional graded algebra E with a certain moduli ∞ 1,n space of curves, established in our previous paper [33]. Namely, the algebra E arises 1,n as the self-Ext-algebra of the coherent sheaf n (cid:77) O ⊕ O , C pi i=1 where C is a projective curve of arithmetic genus one, and p ,...,p are distinct smooth 1 n points of C, such that H1(C,O(p )) = 0 for each i. Taking into account higher products i coming from the dg-enhancement of the derived category of coherent sheaves on C, we see that each such n-pointed curve (C,p ,...,p ) gives rise to an equivalence class of 1 n A -structures on E . Our result in [33] implies that every A -structure on E arises ∞ 1,n ∞ 1,n in such a way (in the case n = 2 there is a problem with the characteristic 2, so we give a special argument in Section 1.2 below). To prove Theorem A we combine this with some direct computations in the relative Fukaya category and with the product formula for theta functions over the n-Tate curve. We denote by T the n-punctured torus T\{z ,...,z }, and we denote by F(T ) and 0 1 n 0 W(T ) its exact Fukaya and wrapped Fukaya categories (see Sec. 3). For every com- 0 mutative ring R we consider the R-linear A -categories F(T ) ⊗ R and W(T ) ⊗ R ∞ 0 0 obtained from F(T ) and W(T ) by the extension of scalars from Z to R. We also set 0 0 G = G ×Spec(R). n,R n Theorem B. Let R be a commutative Noetherian ring. (i) There is an R-linear triangulated equivalence Dπ(F(T )⊗R) (cid:39) Perf(G ) 0 n,R between the split-closed derived exact Fukaya category of T and the derived category of 0 perfect complexes on the standard n-gon. (ii)(=Theorem 3.7.2) Assume that R is regular. Then the equivalence of (i) extends to an R-linear triangulated equivalence of derived categories Dπ(W(T )⊗R) (cid:39) Db(CohG ) 0 n,R between the wrapped Fukaya category of T and the derived category of coherent sheaves 0 on the standard n-gon. If R = Z or R is a field then Dπ(W(T )⊗R) = Db(W(T )⊗R). 0 0 2 Theorem B(i) is a direct consequence of the proof of Theorem A. However, we give also another argument that avoids computations, but uses instead a certain characterization of the A -structure associated with the standard n-gon, established in Theorem 1.3.1. To ∞ prove Theorem B(ii) we first check that the natural Yoneda functor from Db(W(T )⊗R) 0 to the derived category of modules over F(T )⊗R is fully faithful (see Theorem 3.6.2), 0 and then identify the image with Db(CohG ) using the equivalence of Theorem B(i). n,R In addition (working over a field) we prove that the derived categories of the wrapped Fukaya category W(T ) and of the Fukaya category F(T ) are Koszul dual, in the fol- 0 0 lowing sense. In Sec. 3.1 we find certain natural generators n n (cid:77) (cid:77) ˆ ˆ L = L , L = L i i i=0 i=0 for the derived categories of F(T ) and W(T ) respectively. Denote A = CF∗(L,L) 0 0 0 and B = CF∗(Lˆ,Lˆ). The A −B-bimodule CF∗(Lˆ,L) yields augmentations of A and 0 0 B over the semisimple ring K = (cid:76)n k (in the generalized sense considered in [25, Sec. i=0 10]), with the property that RHom (K,K) (cid:39) B , RHom (K,K) (cid:39) A . A0 B 0 ˆ Note that there is a choice in the definition of L : one has to choose one of the punctures. 0 These n choices lead to different augmentations on A and are permuted by a natural 0 Z/n-action (see Proposition 4.1.3). Relation to previous work. As alluded to before, the main results in this paper are gen- eralizations of the main results of [31] to n > 1. However, we should emphasize that classification of the relevant moduli of A -structures given in [33] and the computations ∞ of the homogeneous coordinate rings are rather more complicated in the case n > 1. We would like to note that yet another proof of Theorem B(i) can be given more directly via the corresponding result for n = 1 proven in [31] by using the fact that there are cyclic covering maps from an n-punctured torus to a once-punctured torus, and the standard n-gon to the standard 1-gon. These coverings yield models of F(T ) and Perf(G ) as a 0 n semi-direct product of Z/nZ with a subcategory of the corresponding category associated to the base of the covering (such covering arguments are well known and go back to [45]). A sketch based on this approach to prove Theorem B(i) appeared in the recent [24] (for n = 3) during the writing of this article. However, to complete the proof one needs to show that the two actions of Z/n arising from the covering picture, the one on the Fukaya category of the 1-punctured torus and the one on Perf(G ) (both given by tensoring with 1 order n line bundles), are identified by the equivalence constructed in [31]. We have not pursued this approach as it does not directly give a way of proving Theorem A where the weights of the marked points differ. In[52],theauthorsconstructedanequivalencebetween Perf(G ) andadg-category,called n the “constructible plumbing model”. The construction of this dg-category is inspired by a suggestion of Kontsevich ([29]) that the Fukaya category of T can be calculated via 0 a category of constructible sheaves associated to the Lagrangian skeleton (cf. [35]). The authors of [52] conjecture that their model is quasi-equivalent to the DπF(T ). Clause 0 (i) of Theorem B implies this conjecture. 3 In [1], a Landau-Ginzburg (B-model) mirror to W(T ) was constructed (over C) for some 0 n. In [7] certain non-commutative mirrors to W(T ) for n ≥ 3 were given. It may be 0 interesting to compare these more directly to the derived category of coherent sheaves on the standard n-gon. Outline of Sections. In Section 1, we recall and extend our results from [33] about classifi- cation of A -structures on the associative algebra E via the moduli of certain pointed ∞ 1,n curves of arithmetic genus 1. Also, in Section 1.3 we give a characterization of the A - ∞ structure associated with the standard n-gon. In Section 2, we give a construction of the n-Tate curve and compute its homogeneous coordinate ring. In Section 3, we first give generators for various flavours of Fukaya categories associated with the pair (T,D). Then we relate these to generators of various derived categories associated to the n-Tate curve, establishing the homological mirror symmetry results as stated in Theorems A and B by using classification results from Section 2 and computing the homogeneous coordinate ring of the n-Tate curve inside the Fukaya category of (T,D). Finally, in Section 4, we prove the Koszul duality result between F(T ) and W(T ) by using the equivalences from 0 0 Section 3. Conventions. We use the terminology for generation of triangulated subcategories which may be non-standard for algebraic geometers (but is standard in symplectic geometry): we say that a set of objects S split-generates (resp., generates) a triangulated category T if the smallest thick subcategory (resp., triangulated subcategory) of T containing S is the entire T . For a set of objects S in a triangulated category we denote by (cid:104)S(cid:105) the thick subcategory split-generated by S. By the elliptic n-fold curve we mean the projective curve of arithmetic genus 1, which is the union of n generic projective lines passing through one point in Pn−1, for n ≥ 3, the union of two P1’s intersecting at one tacnode point, for n = 2, and the cuspidal plane cubic, for n = 1. 1. Curves of arithmetic genus 1 with n marked points and A -structures ∞ 1.1. Extension of some results from [33]. Below we always assume that n ≥ 2. Recall that in the previous work [33] we studied the moduli stacks Usns of curves C of 1,n arithmetic genus 1 together with n distinct smooth marked points p ,...,p satisfying 1 n • h0(O (p )) = 1 for all i, and C i • O (p +...+p ) is ample. C 1 n We denote by U(cid:101)sns → Usns the G -torsor corresponding to a choice of a nonzero element 1,n 1,n m ω ∈ H0(C,ω ). C In the case n ≥ 3 we identified U(cid:101)sns with an explicit affine scheme of finite type over Z, 1,n while in the case n = 2 we proved that over Z[1/2] one has U(cid:101)sns (cid:39) A3 (see [33, Thm. 1,2 1.4.2]). More precisely, in the case n ≥ 3 we showed that the ring O(U(cid:101)sns) is generated over Z 1,n by the functions defined as follows. Let (C,p ,...,p ,ω) be the universal family, and let 1 n 4 h , for i (cid:54)= j, be elements of H0(C,O(p +p )) such that Res (h ω) = 1. We normalize ij i j pi ij the elements h by the condition h (p ) = 0 for i ≥ 3 and h (p ) = 0. Then there is a 1i 1i 2 12 3 relation of the form h h2 −h2 h = ah h +bh +ch +d, 12 13 12 13 12 13 12 13 and the functions a,b,c,d together with (1.1.1) c := h (p ), ij 1i j where i,j ≥ 2, i (cid:54)= j, generate the algebra of functions on U(cid:101)sns over Z (see [33, Sec. 1.1]). 1,n Note that these functions have the following weights with respect to the G -action: m (1.1.2) wt(c ) = wt(a) = 1, wt(b) = wt(c) = 2, wt(d) = 3. ij In the case n = 2 we showed (see [33, Sec. 1.2]) that U(cid:101)sns ⊗Z[1/2] is isomorphic to the 1,2 affine 3-space over Z[1/2] with the coordinates α, β and γ of weights 2, 3 and 4, so that the affine part C \{p ,p } of the universal curve is given by the equation 1 2 y2 −yx2 = α(y −x2)+βx+γ, which is simply the unfolding of the tacnode. Furthermore, for each (C,p ,...,p ) as above we considered the generator 1 n G = O ⊕O ⊕...⊕O C p1 pn of the perfect derived category of C. The standard dg-enhancement of this category allows to construct a minimal A -algebra, equivalent to REnd∗(G). ∞ The choice of a nonzero element in H0(C,ω ) gives rise to a canonical identification of C the underlying associative algebra, Ext∗(G,G), with the algebra E ⊗k, associated with 1,n the following quiver Q = Q with relations. n 3 n 2 1 Figure 1. The quiver Q n The path algebra Z[Q] is generated by (A ,B ), where A is the arrow from the central i i i vertex to the vertex i and B is the arrow in the opposite direction. We define a grading i on Z[Q] by letting |A | = 0 and |B | = 1. The ideal of relations J is generated by the i i relations B A = B A , A B = 0 for i (cid:54)= j. i i j j i j 5 We set (1.1.3) E := Z[Q]/J. 1,n Let M = M (E ) be the functor on the category of commutative rings, associating ∞ ∞ 1,n to R the set of gauge equivalence classes of minimal A -structures on E ⊗R (strictly ∞ 1,n unital with respect to the idempotents e ∈ E corresponding to the vertices) extending i 1,n the given m on E ⊗R. 2 1,n The map associating with a point (C,p ,...,p ,ω) ∈ U(cid:101)sns(R) the corresponding A - 1 n 1,n ∞ structure on E ⊗ R (defined up to a gauge equivalence), extends to a morphism of 1,n functors (1.1.4) U(cid:101)sns → M . 1,n ∞ Namely, we can use the homological perturbation construction associated with a dg- model for the subcategory generated by O ⊕O ⊕...⊕O , described in [42, Sec. 3]. C p1 pn This requires a choice of relative formal parameters t at p (compatible with ω so that i i Res (ω/t ) = 1). However, this affects only the choice of homotopies in the homological pi i perturbation, and hence, a different choice leads to a gauge equivalent A -structure. ∞ Furthermore, the morphism (1.1.4) is compatible with the G -action, where the action m of λ ∈ G on M is given by the rescalings m (cid:55)→ λ2−nm . m ∞ n n We proved in [33] that the map (1.1.4) becomes an isomorphism once we change the base to any field (of characteristic (cid:54)= 2, if n = 2). Thus, over a field k, every minimal A - ∞ structure on E ⊗k can be realised, up to gauge equivalence, in the derived category of 1,n some curve (C,p ,...,p ) as above. 1 n The reason we restricted to working over a field in [33] was the dependence on the results of [42] about moduli of A -structures. Using the new work [43] on the relative moduli ∞ of A -structures we can now extend our results, so that everything works over Z, for ∞ n ≥ 3. For n = 2 the similar arguments work over Z[1/2], but we can also use the more ad hoc construction as in [31] to establish a partial result we need over Z (see Section 1.2 below). Theorem 1.1.1. Assume that n ≥ 3. Then the functor M is represented by an affine ∞ scheme of finite type over Z. Furthermore, the morphism (1.1.4) is an isomorphism. For n = 2 the same assertions hold over Z[1/2]. Proof. Assume first that n ≥ 3. By [43, Thm. 2.2.6], the representability of our functor by an affine scheme follows from the vanishing (1.1.5) HHi(E ⊗k) = 0 1,n <0 for i ≤ 1 and any field k. For i = 1 this is [33, Eq. (2.2.1)], while for i = 0 this is clear. Furthermore, to check that M is a closed subscheme of the scheme M of finite type, ∞ n parametrizing A -structures, it is enough to have the vanishing n HH2(E ⊗k) = 0 1,n <−d for some d and all fields k. This is indeed the case for d = 3 by [33, Cor. 2.2.6]. 6 Recall that the morphism (1.1.4) is also compatible with the G -action, where the action m on the global functions λ (cid:55)→ (λ−1)∗ has positive weights on generators over Z: for U(cid:101)sns 1,n the weights are given by (1.1.2), while for M the weight of the coordinates of m , for ∞ n n > 2, is equal to n−2. Thus, the morphism (1.1.4) corresponds to a homomorphism of non-negatively finitely generated graded algebras over Z f : A → B, such that A = B = Z, f⊗Q and f⊗Z/p are isomorphisms. In addition, we know that 0 0 each B is a free Z-module. Indeed, for n ≥ 5 this follows from [33, Cor. 1.1.7], while n for n = 3 (resp., n = 4) this follows from the identification of the moduli space with the affine space A4 (resp., n = 5) given in [33, Prop. 1.1.5]. Thus, C = coker(f : A → B ) n n n n is a finitely generated abelian group such that C ⊗ Q = 0 and C ⊗ Z/p = 0. Hence, n n C = 0 and so f is surjective. Since B is flat over Z, for each p we have an exact n n n sequence 0 → (kerf )⊗Z/p → A ⊗Z/p → B ⊗Z/p → 0 n n n which shows that kerf ⊗Z/p = 0. Since kerf ⊗Q = 0, we derive that kerf = 0. n n n The argument in the case n = 2 is similar. The vanishing of (1.1.5) in the case when char(k) (cid:54)= 2 follows from [33, Eq. (2.1.4), Cor. 2.2.2]. In the proof of the second assertion we use the identification of U(cid:101)sns ⊗Z[1/2] with A3, with coordinates of weights 2, 3 and 1,2 4. (cid:3) Remark 1.1.2. An alternative way to prove the second assertion of Theorem 1.1.1 is to mimic the proof of [43, Thm. B] by first showing that the deformations over Z of the tacnode point of U(cid:101)sns(k), for any field k of characteristic (cid:54)= 2, match the deformations 1,2 of the A -structures. ∞ 1.2. Case n = 2. In the cases n = 1 and n = 2 the moduli stack U(cid:101)sns over Spec(Z) is 1,2 not an affine scheme. The case n = 1 is considered in detail in [31], so here we discuss the case n = 2. Given a curve C of arithmetic genus 1 with smooth marked points p , p , such that 1 2 H1(C,O(p )) = 0, a choice of a nonzero element in H0(C,ω ) is equivalent to a choice i C of a nonzero tangent vector at p (see [33, Lem. 1.1.1]). Let t be a formal parameter at 1 1 p compatible with this tangent vector. Then there exist elements x ∈ H0(C,O(p +p ) 1 1 2 and y ∈ H0(C,O(2p )) such that 1 1 1 x ≡ +k[[t ]], y ≡ +t−1k[[t ]] t 1 t2 1 1 1 1 at p , defined uniquely up to adding a constant. It is easy to see that then the elements 1 1,x,x2,y,xy form a basis of H0(C,O(3p +2p )). Note that y2 −yx2 ∈ H0(C,O(3p +2p )). Hence, 1 2 1 2 we should have a relation of the form y2 = yx2 +axy +by +b(cid:48)x2 +cx+d. 7 Adding a constant to y we can make the term with x2 to disappear, so there is a unique choice of y, so that the above relation takes form (1.2.1) y2 = yx2 +axy +by +cx+d. There remains ambiguity in a choice of x: changing x to x−α leads to the transformation (1.2.2) (a,b,c,d) (cid:55)→ (a+2α,b+αa+α2,c,d+αc). Let us consider the quotient stack A4/G , where the action of the additive group on A4 a is given by (1.2.2). Note that the action (1.2.2) is compatible with the G -action on A4 m such that wt(a) = 1, wt(b) = 2, wt(c) = 3, wt(d) = 4 and the standard G -action on m G (so wt(α) = 1). a Proposition 1.2.1. One has a natural isomorphism U(cid:101)sns (cid:39) A4/G , compatible with the 1,2 a G -actions. m Proof. Consider the G -torsor over U(cid:101)sns corresponding to a choice of a function x ∈ a 1,2 H0(C,O(p +p )) such that the polar part of x at p is 1 . Then as we have seen above, 1 2 1 t1 we can uniquely find y ∈ H0(C,O(2p )), such that the defining equation is of the form 1 (1.2.1). Conversely, starting from the affine curve Spec(A) given by such an equation we construct the projective curve C by taking ProjR(A), where R(A) is the Rees algebra associated with the filtration by degree, where deg(x) = 1, deg(y) = 2. As in [42, Thm. 1.2.4], one easily checks that this gives an isomorphism between our G -torsor over U(cid:101)sns a 1,2 and A4. (cid:3) Proposition 1.2.2. The map A4(R) → M (R) associating with (a,b,c,d) ∈ R4 the ∞ A -structure coming from the corresponding curve in U(cid:101)sns, is surjective in the following ∞ 1,2 cases: (i) R is any field; (ii) R is an integral domain with the quotient field of charac- teristic zero. Proof. We mimic the proof of [31, Thm. C] (see [31, Sec. 5.3]). Let W denote the tangent space to A4 at 0. The derivative of the action (1.2.2) at 0 gives a map d : Lie(G ) → W, a which we can easily compute: d(∂ ) = 2∂ . x a As in [31, Sec. 5.3], the map from A4/G to the functor of A -structures, induces at a ∞ the infinitesimal level a chain map κ from the dg Lie algebra [Lie(G ) → W] (living in a degrees 0 and 1) to the shifted Hochschild cochain complex CH∗(E ) [1] (truncated 1,2 <0 in negative internal degrees). Similarly to [31, Thm. 5.5] we claim that this map induces an isomorphism of cohomology in degrees 0 and 1, when tensored with any field k, or over Z. Indeed, first let us check that H1(κ⊗k) : coker(d⊗k) → HH2(E ⊗k) 1,2 <0 8 is an isomorphism. Note that the source can be viewed as the tangent space to the defor- mationsofthetacnodecurve (C ,p ,p ,ω) in U(cid:101)sns, andthemapitselfasthetangentmap tn 1 2 1,2 to the morphism of deformation functors associating to a deformation of (C ,p ,p ,ω) tn 1 2 the corresponding family of A -structures on E . Now [42, Prop. 4.3.1], applied to ∞ 1,2 families over k[(cid:15)]/((cid:15)2) implies that the map H1(κ⊗k) is injective. On the other hand, we claim that the source and the target have the same dimension, which is equal to 3 when char(k) (cid:54)= 2 and is equal to 4, when char(k) = 2. Indeed, for the source this is easy to see, while for the target this follows from [33, Cor. 2.2.6] in the case char(k) (cid:54)= 2. In the case char(k) = 2 it suffices to show that dimHH2(E ⊗k) ≤ 4, which follows from 1,2 <0 Lemma 1.2.3 below. Thus, we conclude that H1(κ⊗k) is an isomorphism. Now we turn to showing that H0(κ⊗k) : ker(d⊗k) → HH1(E ⊗k) 1,2 <0 is an isomorphism. If char(k) (cid:54)= 2 then ker(d⊗k) = 0 and HH1(E ⊗k) = 0 (see [33, 1,2 <0 Cor. 2.2.2]). The interesting case is when k has characteristic 2. Then ker(d⊗k) = (cid:104)∂ (cid:105) x and ∂ maps under κ to the nonzero element of the one-dimensional space x HH1(E ) (cid:39) HH1(C ) (cid:39) H0(C ,T ) 1,2 <0 tn <0 tn <0 corresponding to the global vector field ∂ of weight −1 on the tacnode C (see [33, x tn Prop. 2.1.3, Lem. 1.5.2] and the proof of [33, Cor. 2.2.2]). The rest of the proof goes as in [31, Sec. 5.3]. (cid:3) We have used the following result. Lemma 1.2.3. Let k be a field of characteristic 2. Then dimHH2(E ⊗k) = 4. 1,2 Proof. Let C = C be the (projective) tacnode curve over k, equipped with a pair of tn smoothpoints p ,p oneachcomponent. Thenwehaveanisomorphism HH∗(E ⊗k) (cid:39) 1 2 1,2 ∗ HH∗(C), so that the second grading is induced by the G -action on C. Indeed, this m follows from the homotopical triviality of the A -structure on E ⊗k associated with ∞ 1,2 C (see [33, Lem. 2.1.2] and [42, Prop. 4.4.1]). We have an exact sequence 0 → H1(C,T ) → HH2(C) → HH2(U) → 0, where U = C \{p ,p } (see [31, Sec. 4.1.3]). We claim that H1(C,T ) = 0. Indeed, let 1 2 V = C\q, where q isthesingularpoint. Then (U,V) isanaffinecoveringof C. Let x ,x 1 2 be the natural coordinates on the components of U (both vanishing at q). Derivations of O(U ∩V) are just pairs (P (x ,x−1)∂ ,P (x ,x−1)∂ ). Derivations of O(U) are those 1 1 1 x1 2 2 2 x2 pairs, for which there exist constants a,b such that P ≡ a+bx modx2k[x ], for i = 1,2 i i i 1 (see [33, Lem. 1.5.2]). On the other hand, when P are linear combinations of xn for i i n ≤ 2 then the corresponding derivation extends to V . This immediately implies that every derivation of O(U ∩ V) is a sum of those extending either to U or to V , hence, H1(C,T ) = 0. Finally, U is an affine plane curve k[x,y]/(y2−yx2), so HH2(U) is given by the corresponding Tjurina algebra k[x,y]/(x2,y2), which is 4-dimensional. (cid:3) 9 1.3. A -characterization of the wheel of projective lines. For a commutative ring ∞ R we consider G := G ×Spec(R), n,R n the standard n-gon over R. Note that it has natural smooth R-points p ,...,p (corre- 1 n sponding to the point 1 ∈ P1 on each component, where the points 0 and ∞ are used for gluing). Furthermore, there is a natural choice of a section ω of the dualizing sheaf of G over R (see [33, Ex. 1.1.9]), so we can view (G ,p ,...,p ,ω) as a family in n,R n,R 1 n U(cid:101)sns(R). By abuse of notation we will sometimes refer to this family simply as G . 1,n n,R Nowlet k beafield. Wearegoingtogiveseveralcharacterizationsoftheequivalenceclass of minimal A -structures on E ⊗k associated with G (via the morphism (1.1.4)). ∞ 1,n n,k Note that for every subset S ⊂ {1,...,n} we have a natural subquiver in Q such that n the corresponding subalgebra is isomorphic to E . In particular, we have n subquivers 1,|S| Q (i) ⊂ Q (where i = 1,...,n) that give embeddings of E into E . Now given a 1 n 1,1 1,n minimal A -structure m on E , for each i we have a well defined restriction m | ∞ • 1,n • Q1(i) which is a minimal A -structure on E (recall that we consider A -structures unital ∞ 1,1 ∞ with respect to the idempotents in E ). On the other hand, every such m gives a 1,n • structure of right A -module on P = e E , i = 0,...,n (where e ,...,e are the ∞ i i 1,n 0 n idempotents in E corresponding to the vertices in Q ). 1,n n Theorem 1.3.1. Let k be a field, and let mwh be the minimal A -structure on E ⊗k • ∞ 1,n associated with G , where n ≥ 2. Then mwh is characterized uniquely (among the A - n,k • ∞ structures we consider in Theorem 1.1.1) up to gauge equivalence and up to G -action, m by the following conditions (i) and either (ii) or (ii’): (i) for every i = 1,...,n, the restriction mwh| is not homotopically trivial; • Q1(i) (ii) dimHH2(E ,mwh) = n; 1,n • (ii’) the subcategories (cid:104)P ,P (cid:105) split-generated by the right A -modules P and P (where 0 i ∞ 0 i the A -structure comes from mwh) are all distinct for i = 1,...,n. ∞ • Furthermore, for all minimal A -structures m on E satisfying (i), one has ∞ • 1,n dimHH2(E ,m ) ≤ n. 1,n • Lemma 1.3.2. One has dimHH2(G ) = n. n,k Proof. Let us write G instead of G for brevity. We have an isomorphism n n,k HH2(G ) (cid:39) Ext1(L ,O ) (cid:39) Ext1(Ω ,O ), n Gn/k Gn Gn/k Gn where L is the cotangent complex, which in this case is isomorphic to Ω (since Gn/k Gn/k G is a locally complete intersection). Let us pick n smooth points p ,...,p ∈ G , one n 1 n n on each component, and let D = p +...+p . Then the exact sequence 1 n 0 → O (−D) → O → O → 0 Gn Gn D induces a long exact sequence 0 → Hom(Ω ,O (−D)) → Hom(Ω ,O ) → Hom(Ω ,O ) → Gn/k Gn Gn/k Gn Gn/k D Ext1(Ω ,O (−D)) → Ext1(Ω ,O ) → Ext1(Ω ,O ) → ... Gn/k Gn Gn/k Gn Gn/k D 10