# Doubling a Path Algebra, or: How to extend Indecomposable

код для вставкиDoubling a Path Algebra, or: How to extend Indecomposable Modules to Simple Modules Wolfgang Rump Introduction. For a field k of characteristic 0 and a representation-finite quiver в€†, it has been observed [8] that the indecomposable k-linear в€†-representations are independent of the chosen orientation. In fact, they can be associated [6] to the positive roots of the corresponding semisimple Lie algebra g = g(в€†), where в€† в†” denotes the unoriented graph underlying в€†. Thus if we consider the graph в€† with pairs of arrows in opposite direction for every edge of в€†, we may ask whether there в†” exists a в€†-representation which induces the indecomposables for each orientation в€† of в€†. Of course, for a reasonable solution, one should expect that the maps representing the arrows of в€† are connected with the opposite maps by a suitable в†” relation, and the resulting в€†-representation should be more or less unique. Our first aim in this paper is to show that this problem is indeed solvable (Theorem 1) and admits a unique solution depending on a fixed element О± in the root space kв€†0 which does not lie on a wall of the Weyl chambers. (For arbitrary О± в€€ kв€†0 , в†” the в€†-representation is still unique, but for some orientations в€†, the induced representation might decompose.) The mentioned relation (see (1) below) between the maps representing в€† and the opposite maps is well-known in the particular case О± = 0: Then it turns into the defining relation of the preprojective algebra [10, 7] of the graph в€†. Our construction in В§2 therefore leads to a semisimple deformation A of the preprojective algebra, which, as we shall see, can be regarded as a вЂќdoubleвЂќ of the path algebra kв€†. We shall prove that the indecomposable kв€†-representations correspond to the simple A-modules. в†” There are quite different contexts where quivers of the form в€† and relations similar to (1) have occurred. We thank the referee for directing our attention to some of them. McKay [15] observed that for a binary polyhedral group G, if the twodimensional irreducible representation R (over C) is tensored with all irreducibles (a ) R1 , . . . , Rm , say R вЉ— Ri = Rj ij , the oriented graph with adjacency matrix (aij ) в†” is of the form в€†, with an extended Dynkin diagram в€†. An explanation is provided, among others, by M. Auslander [3] (see also [4]). He establishes a one-to-one correspondence between R1 , . . . , Rm and the indecomposable projective modules over the twisted group ring S[G], where S = C[[x, y]], and shows that the McKay quiver в†” в†” в€† coincides with the Gabriel quiver of S[G]. Now if в€† is regarded as a translation в†” quiver with the identity as translation, the relations in в€† given by the ring structure of S[G] are just the mesh relations, that is, the relations (1) with О± = 0. On the other hand, the inhomogenous equations (1) are connected with the resolution of singularities, and with minimum action solutions of SU2 Yang-Mills fields, also called instantons or pseudo-particles. Atiyah and Ward [2] have shown that self-dual instantons in R4 correspond to certain two-dimensional algebraic vector bundles over P3 (C), and in [1] these bundles have been constructed in terms of pure linear algebra. A quadratic equation between matrices, similar to (1), results as a defining relation. Every binary polyhedral group G, viewed as a subgroup of SU2 (C), gives rise to a quotient singularity C2 /G with intersection matrix of Dynkin type. If XG denotes the 4-manifold underlying the minimal resolution of C2 /G, Peter Kronheimer has shown [11] that XG admits the structure of an ALE hyper-KВЁahler 4-manifold, that is, a self-dual gravitational instanton. Here, ALE (= asymptotically locally Euclidean) signifies that at infinity, XG resembles R4 /G, and the Riemannian metric is Euclidean up to O(r в€’4 ). вЂњHyper-KВЁahlerвЂќ means that XG is equipped with three covariant-constant complex structures I, J, K connected by the relations of quaternion units. Moreover, he proved that every ALE hyper-KВЁahler 4-manifold is diffeomorphic to some XG . In his construction of XG , Kronheimer starts with the manifold M = (R вЉ— EndCG)G , on which the group F of G-invariant unitary transformations of CG operates. Then XG is obtained as a hyper-KВЁahler quotient M /F , where M вЉ‚ M is defined by a quadratic relation similar to (1). The left-hand side of this relation plays the rЛ†ole of a (hyper-KВЁahler) moment map. In a similar way, relation (1) enters into the definition of certain varieties of quivers considered by Lusztig ([12], В§8; [13], В§12) and NakajimaвЂ™s quiver varieties [16, 17]. Our subsequent article will be self-contained. No use is made of the above mentioned connections. 1. Extensions of indecomposables. Let k be a field of characteristic 0 and в€† a finite oriented graph with vertex set в€†0 = {1, . . . , n}. A representation (V, f ) of в€† over k is defined as a family V = (V1 , . . . , Vn ) of finite dimensional k-vector spaces and a family f of k-linear maps fji : Vi в†’ Vj for each arrow i в†’ j in в€†. Let kв€† denote the path algebra of в€†, i.e. the k-algebra with paths p : i0 в†’ i1 в†’ . . . в†’ ir in в€† as basis such that, for another path q : j0 в†’ . . . в†’ js in в€†, the product qp is given by the composition i0 в†’ . . . в†’ ir в†’ j1 в†’ . . . в†’ js if ir = j0 , and qp = 0 otherwise. Then in an obvious way, each в€†-representation (V, f ) can be regarded as a finitely generated kв€†-module. The vector dim(V, f ) = (d1 , . . . , dn ) в€€ Nn with di = dimk Vi is called the dimension vector of (V, f ). GabrielвЂ™s theorem [8] states that the number of (isomorphism classes of) indecomposable в€†-representations (i.e. kв€†-modules) is finite if and only if the unoriented graph в€† underlying в€† is a disjoint union of Dynkin diagrams. Moreover, the set of dimension vectors of indecomposable в€†-representations coincides with the set О¦+ 2 of positive roots of the semisimple Lie algebra g of type в€†. Thus for a given root d в€€ О¦+ , there is an indecomposable representation with dimension vector d for each orientation в€† of в€†. в†” Now let в€† be a Dynkin diagram. If в€† denotes the oriented graph with arrows iв†’j and jв†’i for each edge iвЂ”j in в€†, the question arises for a given dimension vector в†” d в€€ О¦+ , whether there exists a в€†-representation which induces the indecomposable в†” в€†-representations of all orientations via the natural embedding kв€† в†’ kв€†. Our first result (Theorem 1) will give a solution to this problem, including an в†” explicit construction of the k в€†-modules in question. Furthermore, we shall prove в†” that these kв€†-modules can be regarded as simple modules over a semisimple algebra associated with в€†. Let us define a в€†-representation of type О± = (О±1 , . . . , О±n ) в€€ kn в†” as a в€†-representation (V, f ) which satisfies for each i в€€ в€†0 the relation fij fji = О±i В· 1Vi (1) j where j runs over the vicinity V(i) of i, that is, the set of vertices j adjacent to i in в€†. If d is the dimension vector of (V, f ), the following relation necessarily holds for the О±i : В±di О±i = 0. (2) в€’ Here, the + and в€’ signs are chosen according to any fixed partition в€†0 = в€†+ 0 в€Є в€†0 such that adjacent vertices belong to different signs. (Since в€† is a tree, there are only two such partitions!) In fact, if we take the trace on both sides of (1), we get dij = О±i di jв€€V(i) where dij := tr(fij fji ) = dji . Thus (2) immediately follows. For a fixed partition в€’ в€†0 = в€† + 0 в€Є в€†0 , let us define О¦d в€€ k[x1 , . . . , xn ] by О¦d = В±di xi = d i xi в€’ iв€€в€†+ 0 d i xi . (3) iв€€в€†в€’ 0 Theorem 1 Let в€† be a Dynkin diagram and О± в€€ kn with О¦d (О±) = 0. Then every indecomposable в€†-representation (V, f ) with dimension vector d has a unique extension to a в€†-representation (VЛњ , fЛњ) of type О±. Moreover, the fЛњij depend polynomially on О±. As a consequence, we get the solution to the above question: Corollary. Let (V, f ) be a в€†-representation of dimension vector d в€€ О¦+ and type О± such that В±di О±i = 0 for all d = d in О¦+ . Then the induced в€†-representations of all orientations are indecomposable. 3 Note that this choice of О± is possible since k is infinite. The corollary follows immediately by a proposition which we shall prove together with Theorem 1: Лњ a в€†-representation which Proposition 1 Let в€† be an orientation of в€† and M Лњ extends some в€†-representation M . Then M has a subrepresentation with underlying в€†-representation S such that S is an indecomposable direct summand of M . 2. The double of a path algebra. Our second purpose of this paper is to show that by Theorem 1, the decomposition g = g+ вЉ• g0 вЉ• gв€’ of a semisimple Lie algebra g has an analogue for associative algebras. Namely, the nilpotent Lie algebra g+ corresponds to our path algebra A+ = kв€†, the negative part gв€’ to Aв€’ = (kв€†)op , and the abelian part g0 is replaced by some commutative algebra A0 generated by n elements О±1 , . . . , О±n . The semisimple Lie algebra g corresponds to a semisimple algebra A containing an order О›0 generated by A+ , Aв€’ , and A0 . Firstly, the в€†-representations over some extension field of k can be interpreted as modules over the k-algebra в†” (4) О› = Rв€—в€† /I where Rв€— is the polynomial ring k[x1 , . . . , xn ] and I the principal ideal generated by eij eji в€’ xi e i . (5) Here eij denotes the path j в†’ i, and ei is the primitive idempotent (empty path) corresponding to the vertex i в€€ в€†0 . The first sum in (5) is to be extended over the edges iвЂ”j, the second over the vertices i in в€†. The residue class of xi modulo I will be denoted by О±i . Thus О±1 , . . . , О±n в€€ О› generate a subring R = k[О±1 , . . . , О±n ] in the center of О›, and О› becomes an R-algebra. Recall that an ideal P in a ring О“ is said to be prime if for any two ideals I, J in О“, the inclusion P вЉѓ I В· J implies P вЉѓ I or P вЉѓ J. An element r в€€ О“ is called Лњ is said regular if the left and right multiplication by r is injective. A subring О“ of О“ Лњ Лњ to be an order in О“ if each regular r в€€ О“ is invertible in О“, and every element of Лњ is of the form r в€’1 a, and also of the form ar в€’1 , with r, a в€€ О“ and r regular. The О“ intersection N(О“) of all prime ideals in О“ is called the prime radical ([19], chap. XV), and О“ is said to be semiprime if N(О“) = 0. Now we define О›0 := О›/N(О›). The natural homomorphism О› О›0 maps R onto a central subring R0 = k[О± ВЇ1, . . . , О± ВЇ n ] of О›0 , generated by the residue classes of the О±i modulo N(О›). Let О¦ в€€ Rв€— denote the product of all О¦d , d в€€ О¦+ . By virtue of (5), the defining relations (e2i = ei , ei eij = eij , etc.) of the eij which encode the graph structure of в€†, are turned into a single relation О¦(О± ВЇ1, . . . , О± ВЇ n ) = 0 in R0 encoding the root system of в€†: Theorem 2 The kernel of the natural map ПЃ : Rв€— R0 is the principal ideal (О¦). If M is any в€†-representation such that for some orientation в€† of в€†, the underlying kв€†-module is indecomposable, then N(О›)M = 0, i.e. M can be regarded as a О› 0 module. 4 For each orientation в€† of в€†, the kв€†-module kв€† extends to a в€†-representation M of type (0, . . . , 0). Thus Theorem 2 implies N(О›)M = 0, whence kв€† в€© N(О›) = 0 in О›. Therefore, the natural ring homomorphism kв€† в†’ О› О›0 (6) is injective, i.e. kв€† and (kв€†)op can be regarded as subrings of О›0 , and О›0 is generated by these subrings together with R0 . Since for each d в€€ О¦+ , there is an integral domain Rd := Rв€— /(О¦d ) isomorphic to k[x1 , . . . , xnв€’1 ], with quotient field Kd , say, Theorem 2 yields a natural embedding Kd =: K, R0 в†’ (7) dв€€О¦+ and K is the classical quotient ring [19] of R0 . Consequently, there is a natural homomorphism О› в€’в†’ K вЉ•R О› =: A. (8) Theorem 3 The ring A is semisimple, the kernel of (8) is N(О›), and thus О› 0 is an order in A. The blocks of A correspond to the positive roots d в€€ О¦ + , and are matrix algebras Md (Kd ) with d = d1 + . . . + dn . This theorem leads to a fairly precise description of the structure of О›0 . In particular, each indecomposable в€†-representation M (for some orientation в€† of в€†) extends to a О›0 -representation E via Theorem 1, and for fixed d = dimM and all orientations в€†, the О›0 -lattices E belong to a unique simple A-module Sd . (Such lattices over an order О›0 are said to be irreducible.) The reason that M extends to a О›0 -lattice E depends on the fact that the extension problem in Theorem 1 admits an вЂњintegralвЂќ solution. For example, let в€† be the following orientation of E6 : 4 вќ„ 1 f12 вњІ 2 вњІ 3вњ› 5вњ› 6 and (V, f ) the indecomposable пЈ« пЈ¶в€†-representation пЈ« пЈ¶ with пЈ« пЈ¶ 1 0 1 0 0 1 1 f21 = ; f32 = пЈ 0 0 пЈё ; f34 = пЈ 1 пЈё ; f35 = пЈ 1 0 пЈё ; f56 = . 0 0 0 1 1 0 1 Then the extension to a в€†-representation of type О± is given by Оґ в€’ОІ в€’ОІ 1 6 ; f65 = (О±6 ОІ) ; f43 = (в€’Оі в€’Оґ ОІ); f53 = ОіОі О±5 в€’О± = (О±1 ОІ); f23 = О±2 в€’О± 0 0 О±5 Оґ О±2 where ОІ = О±3 в€’ О±2 в€’ О±5 , Оі = О±2 в€’ О±1 в€’ О±3 , and Оґ = О±5 в€’ О±3 в€’ О±6 . The entries of all matrices fij are integral in the О±1 , . . . , О±6 . In general, this follows by means of 3. Reflection functors for в€†-representations, which, in virtue of Theorem 1, can be viewed as extensions of the functors FiВ± of Bernstein, Gelfand, and Ponomarev [6] to в€†-representations. 5 Firstly, for each vertex i в€€ в€†0 , define a linear map Пѓi в€€ Aut kn by Пѓi (x1 , . . . , xn ) = (x1 , . . . , xn ), where пЈ± for j = i пЈІ в€’xi xj в€’ xi for j в€€ V(i) xj = (9) пЈі xj otherwise. For a given в€†-representation (V, f ) of type О±, and i в€€ в€†0 , let us define a в€†representation Пѓi+ (V, f ) of type Пѓi (О±) as follows. For j в€€ V(i), the maps fij : Vj в†’ Vi and fji : Vi в†’ Vj give rise to k-linear maps fi0 : Vj в†’ Vi and f0i : Vi в†’ Vj such that fi0 f0i = О±i В· 1Vi . (10) If f0i : Vi в†’ Vj is the kernel of fi0 , then fi0 (f0i fi0 в€’ О±i В· 1L Vj ) = 0, whence there exists a unique fi0 : Vj в†’ Vi such that f0i fi0 в€’ О±i В· 1L Vj = f0i fi0 . (11) Now the f0i and fi0 decompose into k-linear maps fji : Vi в†’ Vj and fij : Vj в†’ Vi , which replace the fji and fij in order to obtain Пѓi+ (V, f ). It is easily verified that (1) is satisfied for each vertex j в€€ в€†0 if О± is replaced by Пѓi (О±). The dimension vector d is changed into Пѓi (d) = (d1 , . . . , dn ), where di = в€’di + jв€€V(i) dj and dj = dj for j = i. If Пѓi+ is restricted to в€†-representations, we obtain a functor which is applicable without any condition on в€†. If i is a sink, i.e. no arrows start at i, this functor coincides with the вЂњimage functorвЂќ Fi+ in [6]. Dually to Пѓi+ , we define Пѓiв€’ by taking the cokernel fi0 : Vj Vi of f0i , etc. Remark. Since these functors are well defined even if k is replaced by an integral domain, they can be applied to the irreducible representations E of О›0 considered above. However, since ПѓiВ± is universally applicable, even if i is neither a sink nor a source, there exist вЂњnon-orientedвЂќ irreducibles of О›0 , i.e. those irreducible О›0 -lattices which do not arise by extension of some indecomposable в€†-representation via Theorem 1. Now let (V, f ) be a в€†-representation of type О± and i в€€ в€†0 . If О±i = 0, then (10) and (11) state that the maps f0i , f0i and О±1i fi0 , в€’ О±1i fi0 form a biproduct diagram [14] Vi в€’в†’ в†ђв€’ Vj в†ђв€’ в€’в†’ Vi , (12) jв€€V(i) that is, we have a natural isomorphism Vj в€ј = Vi вЉ• Vi . Hence Пѓi+ and Пѓiв€’ coincide on (V, f ). In this case, we simply write Пѓi instead of ПѓiВ± . If О±i = 0, we put Пѓi = 1, the identity functor. Via these Пѓi , the Weyl group of в€† operates on the category of в€†-representations. 4. Proofs. For the proof of the existence part of Theorem 1, these reflection functors Пѓi suffice. For the uniqueness part, however, we have to make use of Пѓi+ and Пѓiв€’ : 6 Proof of Theorem 1. If (V, f ) is simple, then О¦d (О±) = В±О±i for some i в€€ в€†0 , and the assertion is trivial. Otherwise, let i be a sink of в€†. Then Fiв€’ Fi+ (V, f ) = (V, f ), where FiВ± denote the classical reflection functors [6] for в€†-representations. As О¦d (О±) = 0 implies О¦d (О± ) = 0 for d = Пѓi (d) and О± = Пѓi (О±), we may assume by induction that Fi+ (V, f ) is extendable to a в€†-representation (W, g) of type Пѓi (О±). Then (V, f ) extends to the в€†-representation Пѓiв€’ (W, g) of type Пѓi Пѓi (О±) = О±. To prove the uniqueness of this в€†-representation, let (VЛњ , fЛњ) be any such extension. Then Пѓi+ (VЛњ , fЛњ) extends Fi+ (V, f ), whence Пѓi+ (VЛњ , fЛњ) = (W, g). Therefore, we get (VЛњ , fЛњ) = Пѓiв€’ Пѓi+ (VЛњ , fЛњ) = Пѓiв€’ (W, g). By the construction of ПѓiВ± , an inductive argument also proves the integrality property of fЛњ. Лњ which Proof of Proposition 1. If i в€€ в€†0 is a source of в€†, we may apply Пѓiв€’ to M в€’ + в€’ Лњ Лњ , whence by induction, extends the application of Fi to M . Then Пѓi Пѓi M = M we may assume that M has a simple direct summand S concentrated at a source i. Лњ. Thus О±i = 0, and S extends to a subrepresentation SЛњ of M Next we shall focus our attention to Theorems 2 and 3. If О± is specialized to (0, . . . , 0), then О› turns into the preprojective algebra О (в€†) of в€† [10, 7, 5]. For k в€€ N, define О›k as the R-submodule of О› generated by the paths of length в‰¤ k. This gives a filtration R вЉ‚ О› 0 вЉ‚ О›1 вЉ‚ О›2 вЉ‚ . . . вЉ‚ О› (13) of О›, i.e. О›i О›j вЉ‚ О›i+j for i, j в€€ N. With О›в€’1 := 0, we can form the associated graded R-algebra (О›i /О›iв€’1 ), wherein the defining relation (5) simplifies to eВЇij eВЇji = 0, if eВЇij denotes the residue class of eij in О›1 /О›0 . Hence, we have a natural epimorphism of R-algebras: R вЉ—k О (в€†) (О›i /О›iв€’1 ). (14) iв€€N Lemma 1 О› is finitely generated as an R-module. Proof. By virtue of (14), this follows from the well-known fact that О (в€†) is finite dimensional. For the convenience of the reader, let us give a proof (cf. [9, 18]) which also sheds some light upon the structure of О› via (14). It suffices to show that for large m в€€ N, any path of length в‰Ґ m becomes zero in О (в€†). Let i в€€ в€†0 be a fixed vertex. We choose the unique orientation в€† of в€† such that for each j в€€ в€†0 , there is a path from i to j. Define Zв€† as the oriented graph with vertex set в€†0 Г— Z and arrows (i, l) в†’ (j, l) в†’ (i, l + 1) for each arrow i в†’ j in в€† and l в€€ Z. (Note that, as an abstract oriented graph, Zв€† does not depend on the orientation of в€†.) The numbering of Zв€† is given by the embedding в€† в†’ Zв€† with i в†’ (i, 0) for i в€€ в€†0 . Now let (V, f ) be i-th projective в€†-representation, i.e. Vj = k for all j в€€ в€†0 and fjk = 1 for each arrow k в†’ j in в€†. There is a natural way to extend (V, f ) to a Zв€†-representation. By induction, suppose that (V, f ) is already extended to the full subgraph О“ of Zв€†, and let (i, l) be a source in О“ such that (j, m) в€€ О“ if there is an arrow (i, l) в†’ (j, m) in Zв€†, and (i, l + 1) в€€ О“. We apply the classical reflection в€’ functor F(i,l) to (V, f ). This gives a representation (V , f ) where (i, l) is a sink. 7 Define V(i,l+1) := V(i,l) , f(i,l+1)(k,l) := f(i,l)(k,l) and f(i,l+1)(j,l+1) := f(i,l)(j,l+1) for arrows j в†’ i в†’ k in в€†. Thus we obtain a Zв€†-representation (V, f ). If (W, g) is a nonsimple indecomposable representation of some orientation of в€†, and j is a source for this orientation, then Wj в†’ kв€€V(j) Wk is always injective. Therefore, it is easily seen that a path in О (в€†) starting in i vanishes if and only if the corresponding map V(i,0) в†’ V(j,l) in (V, f ) is zero. But since в€† is representation-finite, V(j,m) = 0 for sufficiently large m. Lemma 2 Every prime ideal P of О› contains some О¦d (О±) with d в€€ О¦+ . Proof. Since О› is noetherian, GoldieвЂ™s first theorem ([19], chap. II, Prop. 2.6) implies that О›/P is an order in a simple ring B. The ideal p = P в€© R of R is prime, whence B contains the quotient field F of R+P/P в€ј = R/p. By Lemma 1, О›/P and F generate a finite dimensional F -algebra in B which therefore coincides with B. Consequently, B gives rise to a в€†-representation over F . By Proposition 1, at least one relation О¦d (О±) = 0 must hold in F , whence О¦d (О±) в€€ p вЉ‚ P . Corollary. There is a positive integer s with О¦(О±)s = 0 in О›. Proof. By Lemma 2, О¦(О±) lies in the prime radical N(О›) of О› which is a nil ideal ([19], chap. XV, Prop. 1.2). Next we consider the natural epimorphism ПЃ : k[x1 , . . . , xn ] k[О±1 , . . . , О±n ] = R (15) with ПЃ(xi ) = О±i . Lemma 3 The kernel of ПЃ is contained in the principal ideal (О¦). Proof. Let d в€€ О¦+ be given. Take any orientation в€† of в€†. By Theorem 1, the indecomposable kв€†-module with dimension vector d admits a unique extension to a в€†-representation M of type (ВЇ x1 , . . . , xВЇn ) over the quotient field Kd of Rd = в€ј k[x1 , . . . , xn ]/(О¦d ) = k[ВЇ x1 , . . . , xВЇn ] = k[x1 , . . . , xnв€’1 ], where xВЇi is the residue class of xi modulo О¦d . Hence, if f в€€ KerПЃ, then fВЇM = 0 for the corresponding fВЇ в€€ Rd . Thus fВЇ = 0, i.e. f в€€ (О¦d ). Since this holds for each d в€€ О¦+ , we obtain Ker ПЃ вЉ‚ (О¦d ) = (О¦). For any d в€€ О¦+ , the preceding lemma implies that p := О¦d (О±) is a prime element of R. For Ad := R(p) вЉ—R О›, we have: Lemma 4 p В· Ad = 0. 8 Proof. Without loss of generality, we may suppose that |V(n)| в‰¤ 1, and dn = 1. Let ПЂ : R в†’ R(p) be the natural homomorphism, and О±i = ПЂ(О±i ) for i в€€ {1, ..., nв€’ 1}. The residue class field R(p) /pR(p) is isomorphic to the quotient field Kd of Rd = R/pR which is isomorphic to the function field k(x1 , . . . , xnв€’1 ). Hence, Kd is isomorphic to the subfield F := k(О±1 , . . . , О±nв€’1 ) of R(p) , and R(p) = F вЉ• pR(p) . Moreover, there is a unique О±n в€€ F such that ПЂ(О±n ) = О±n + ПЂ(p) ; О¦d (О±1 , . . . , О±n ) = 0 (16) holds in R(p) . Next we consider the exact sequence pAd /p2 Ad в†’ Ad /p2 Ad Ad /pAd . (17) For an arbitrary orientation в€† of в€†, let X be the indecomposable F в€†-module. Since pAd /p2 Ad and Ad /pAd are в€†-representations of type (О±1 , . . . , О±n ) over F , the corresponding F в€†-modules are isomorphic to powers of X. Hence, Ext(X, X) = 0 implies that the sequence (17) of F в€†-modules splits. Regarding Ad /p2 Ad as a в€†representation (V, f ) over the ring R(p) /p2 R(p) , the F в€†-module (V, f ) is thus of the form (C, f ) вЉ• (pC, f ), that is, for each vertex i в€€ в€†0 , we have Vi = Ci вЉ• pCi , and for each arrow i в†’ j in в€†, fji = f0ji f0 ; fij = hfij f0 . ij ji ij Furthermore, the multiplication by p gives rise to an epimorphism (C, f ) (pC, f ) of F в€†-modules which splits by virtue of Ext(X, X) = 0. Therefore, p can be regarded as the natural projection in a decomposition (C, f ) = (pC, f )вЉ•(D, g). Thus p induces an endomorphism p of (C, f ), and the hij induce F -linear maps hij : Cj в†’ pCi в†’ Ci which extend the F в€†-module (C, f ) to a в€†-representation (C, fЛњ) of type (0, . . . , 0, p ). For any edge iвЂ”j in в€†, let dij be the trace tr(fЛњij fЛњji ). Then dij = dji , and for each vertex i в€€ в€†0 , 0 if i = n dij = tr p |Cn if i = n. jв€€V(i) But this implies tr p |Cn = 0, whence pC = 0 and thus pAd /p2 Ad = 0. By NakayamaвЂ™s lemma, we conclude pAd = 0. Using the notation of (7), we thus obtain that Ad is a finite dimensional Kd algebra. Since the relation О¦d (О±) = 0 in Kd holds for no positive root other than d, Proposition 1 and a similar argument as in the preceding proof, together with the uniqueness part of Theorem 1 imply that the в€†-representation Ad decomposes into simple в€†-representations of a single type О±. Hence, the Kd -algebra Ad is simple. Consider the natural map ПЂd : О› в€’в†’ Ad . (18) The image of ПЂd is an order in Ad . Thus, by GoldieвЂ™s theorem ([19], chap. II, Prop. 2.6), the kernel Pd of ПЂd is a prime ideal of О›: Pd = {a в€€ О› | в€ѓr в€€ R \ RО¦d (О±) : ra = 0}. 9 (19) Lemma 5 Pd в€© R = RО¦d (О±). Proof. Since О¦d (О±) в€€ R is prime, the inclusion вЂњвЉ‚вЂќ holds. Conversely, suppose О¦d (О±) в€€ Pd . By Lemma 2, there would exist some d в€€ О¦+ with d = d and О¦d (О±) в€€ Pd . Consider the epimorphism (15). By (19), we infer KerПЃ вЉ‚ (О¦d ), which contradicts Lemma 3. Lemma 6 Let M be a в€†-representation of type О± в€€ F n over an extension field F of k such that, for some orientation в€† of в€†, the underlying kв€†-module is indecomposable. If N := dв€€О¦+ Pd , then N M = 0. Proof. There is a unique k-algebra-homomorphism П„ : k[x1 , . . . , xn ] в†’ F with П„ (xi ) = О±i . The kernel p of П„ is a prime ideal of k[x1 , . . . , xn ], and О¦d в€€ p holds for the dimension vector d в€€ О¦+ of M . By Theorem 1, there is a unique в€†-representation Md of type (ВЇ x1 , . . . , xВЇn ) over the quotient field Kd of Rd = k[x1 , . . . , xn ]/(О¦d ), where xВЇi = xi + (О¦d ), and there is an Rd -lattice E in Md which is a в€†-representation over the ring Rd such that M в€ј = F вЉ—Rd E. Since Md is an Ad -module, we have N Md вЉ‚ Pd Md = 0 and thus N M = 0. Pd and R в€© N(О›) = R В· О¦(О±). Lemma 7 N(О›) = dв€€О¦+ Proof. Let N be as in Lemma 6 and P a prime ideal in О›. As in the proof of Lemma 2, О›/P is an order in a simple F -algebra B = S d with a simple в€†representation S over F . Choose any orientation в€† of в€†. By Proposition 1, S is indecomposable as a kв€†-module. Hence, Lemma 6 implies N (О›/P ) вЉ‚ N B = 0. Consequently, N вЉ‚ P and thus N(О›) = N . The second equation follows by Lemma 5 and Lemma 3. в€ј Proof of Theorem 2 and 3. Theorem 2 follows by Lemma 6 and 7. Since Ad = Kd вЉ—R О› by Lemma 4, Theorem 3 follows immediately by (18), (19), and Lemma 7. The endomorphism ring of the simple Ad -modules is Kd since this holds for the underlying indecomposable Kd в€†-modules. 5. An open question. We have shown that the element О¦(О±) = О¦d (О±) of О› is nilpotent. On the other hand, Lemma 2 and (19) imply that there is a polynomial r в€€ k[x1 , . . . , xn ] \ (О¦) with r(О±)О¦(О±) = 0. Since k[x1 , . . . , xn ] is not a principal ideal domain for n в‰Ґ 2, we cannot conclude О¦(О±) = 0. A direct calculation in О›, however, shows that О¦(О±) = 0 at least for в€† = An and D4 . If О¦(О±) = 0 holds, then the distinction between R and R0 can be dropped. A further simplification would arise if О› is semiprime: then О›0 would coincide with О›. By Lemma 7, this would also imply О¦(О±) = 0. Remark. The above results have been presented on a conference in Constantza, in 1995. The first question was anwered by W. Crawley-Boevey around 1998 (cf. his paper: Geometry of the moment map for representations of quivers, Compos. Math.126 (2001), 257-293). His method can be extended to obtain a positive answer also to the more general question, i. e. the above ring О› is in fact semiprime. 10 References [1] M. F. Atiyah, N. J. Hitchin, V. G. Drinfeld, Yu. I. Manin: Construction of instantons, Phys. Lett. 65A (1978), 185-187 [2] M. F. Atiyah, R. S. Ward: Instantons and Algebraic Geometry, Commun. Math. Phys. 55 (1977), 117-124 [3] M. Auslander: Rational singularities and almost split sequences, Trans. AMS 293 (1986), 511-531 [4] M. Auslander, I. Reiten: McKay Quivers and Extended Dynkin Diagrams, Trans. AMS 293 (1986), 293-301 [5] D. Baer, W. Geigle, H. Lenzing: The preprojective algebra of a tame hereditary artin algebra, Comm. in Alg. 15 (1987), 425-457 [6] I. N. Bernstein, I. M. Gelfand, V. A. Ponomarev: Coxeter Functors and GabrielвЂ™s Theorem, Uspechi Mat. Nauk 28 (1973), 19-33 = Russ. Math. Surveys 28 (1973), 17-32 [7] V. Dlab, C. M. Ringel: The preprojective algebra of a modulated graph, in: Springer Lecture Notes in Math. 832 (1980), 216-231 [8] P. Gabriel: Unzerlegbare Darstellungen I, Manuscripta math. 6 (1972), 71-103 [9] P. Gabriel: Christine Riedtmann and the selfinjective algebras of finite representation-type, Proc. Conf. on Ring Theory, Antwerp 1978. New York - Basel 1979 [10] I. M. Gelfand, V. A. Ponomarev: Model algebras and representations of graphs, Funkc. anal. i priloЛ‡z. 13 (1979), 1-12 = Funct. Anal. Appl. 13 (1979), 157-166 [11] P. B. Kronheimer: The construction of ALE spaces as hyper-KВЁahler quotients, J. Diff. Geom. 29 (1989), 665-683 [12] G. Lusztig: Canonical Bases Arising from Quantized Enveloping Algebras II, Progress of Theor. Physics Suppl. No. 102 (1990), 175-201 [13] G. Lusztig: Quivers, Perverse Sheaves, and Quantized Enveloping Algebras, J. AMS 4 (1991), 365-421 [14] S. Mac Lane: Categories for the Working Mathematician, New York - Heidelberg - Berlin 1971 [15] J. McKay: Graphs, singularities, and finite groups, Proc. Sympos. Pure Math. 37, AMS Providence, R. I. (1980), 183-186 11 [16] H. Nakajima: Instantons on ALE spaces, quiver varieties, and Kac-Moody algebras, Duke Math. J. 76 (1994), 365-416 [17] H. Nakajima: Varieties Associated with Quivers, Preprint ВЁ [18] Chr. Riedtmann: Algebren, DarstellungskВЁocher, Uberlagerungen und zurВЁ uck, Comm. Math. Helv. 55 (1980), 199-224 [19] Bo StenstrВЁom: Rings of Quotients, New York - Heidelberg - Berlin 1975 12

1/--страниц