Geometric characterizations of the representation type of hereditary algebras and of canonical algebras
aa r X i v : . [ m a t h . R T ] N ov GEOMETRIC CHARACTERIZATIONS OF THE REPRESENTATION TYPE OFHEREDITARY ALGEBRAS AND OF CANONICAL ALGEBRAS
CALIN CHINDRISA
BSTRACT . We show that a finite connected quiver Q with no oriented cycles is tame ifand only if for each dimension vector d and each integral weight θ of Q , the moduli space M ( Q, d ) ssθ of θ -semi-stable d -dimensional representations of Q is just a projective space. Inorder to prove this, we show that the tame quivers are precisely those whose weight spacesof semi-invariants satisfy a certain log-concavity property. Furthermore, we characterizethe tame quivers as being those quivers Q with the property that for each Schur root d of Q , the field of rational invariants k ( rep ( Q, d )) GL ( d ) is isomorphic to k or k ( t ) . Next, we ex-tend this latter description to canonical algebras. More precisely, we show that a canonicalalgebra Λ is tame if and only if for each generic root d of Λ and each indecomposable irre-ducible component C of rep ( Λ, d ) , the field of rational invariants k ( C ) GL ( d ) is isomorphicto k or k ( t ) . Along the way, we establish a general reduction technique for studying fieldsof rational invariants on Schur irreducible components of representation varieties. Introduction
Quiver invariant theory
Proof of Theorem 1.1
Quivers with relations
Exceptional sequences and rational invariants
Canonical algebras
References
NTRODUCTION
Throughout this paper, we work over an algebraically closed field k of characteristiczero. One of the fundamental problems in the representation theory of algebras is thatof classifying the indecomposable representations. The representation type of a finite-dimensional algebra reflects the complexity of its indecomposable representations. Analgebra is of tame representation type if, for each dimension d , all but a finite number of d -dimensional indecomposable representations belong to a finite number of -parameterfamilies. Within the class of tame algebras, we distinguish the subclass of algebras of finiterepresentation type ; these are the algebras with only finitely many indecomposable repre-sentations up to isomorphism. An algebra is of wild representation type if its representationtheory is at least as complicated as that of a free algebra in two variables. The remark-able Tame-Wild Dichotomy Theorem of Y. Drozd [24] says that every finite-dimensionalalgebra is of tame representation type or wild representation type and these types aremutually exclusive. Since the representation theory of a free algebra in two variables is Date : September 26, 2010; Revised: November 19, 2018.2000
Mathematics Subject Classification.
Primary 16G20; Secondary 16G10, 16G60, 16R30.
Key words and phrases.
Canonical algebras, exceptional sequences, moduli spaces, rational invariants,representation type, semi-invariants, tame algebras. nown to be undecidable, one can hope to meaningfully classify the indecomposable rep-resentations only for tame algebras. For more precise definitions, see [4, Section 4.4] andthe reference therein.The tame quivers are well understood. P. Gabriel’s famous result [27] identifies the con-nected quivers of finite representation type as being those whose underlying graphs arethe Dynkin diagrams of types A , D , or E . Later on, L. A. Nazarova [43], and P. Donovanand M. R. Freislich [23] found the representation-infinite tame connected quivers. Theirunderlying graphs are the Euclidean diagrams of types e A , e D , or e E .In this paper, we seek for an interpretation of the representation type of an algebra interms of its (birational) invariant theory. A first result in this direction was obtained by A.Skowro ´nski and J. Weyman in [57] where they showed that a finite-dimensional algebraof global dimension one is tame if and only if all of its algebras of semi-invariants are com-plete intersections. Unfortunately, this result does not extend to algebras of higher globaldimension. In fact, W. Kra´skiewicz [39] found examples of algebras of global dimensiontwo for which [57, Theorem 1] does not hold. As it was suggested by Weyman [59], in or-der to detect the tameness of an algebra, one should impose geometric conditions on thevarious moduli spaces of semi-stable representations rather than on the entire algebras ofsemi-invariants.We begin with the following characterization of the tameness of finite-dimensional pathalgebras. Theorem 1.1.
Let Q be a finite, connected quiver without oriented cycles. The following condi-tions are equivalent: (1) the path algebra kQ is tame; (2) for each dimension vector d and each integral weight θ of Q such that d is θ -semi-stable, M ( Q, d ) ssθ is a projective space; (3) for each dimension vector d and each integral weight θ of Q , the sequence { dim k SI ( Q, d ) Nθ } N ≥ is log-concave, i.e., dim k SI ( Q, d ) ( N + ) θ · dim k SI ( Q, d ) ( N − ) θ ≤ ( dim k SI ( Q, d ) Nθ ) , ∀ N ≥ ; (4) for each Schur root d of Q , the field of rational invariants k ( rep ( Q, d )) GL ( d ) is isomorphicto k or k ( t ) . We point out that the implication ( ) = ⇒ ( ) was proved by M. Domokos and H. Lenz-ing by first studying moduli spaces of regular representations for concealed-canonical al-gebras (see [22]). The other implication ( ) = ⇒ ( ) has been recently proved by Domokosin [20] using the local quiver technique of J. Adriaenssens and L. Le Bruyn (see [1]). Ourproof of ( ) ⇐⇒ ( ) is different from the one in [20, 22]. More specifically, we workentirely within the category of representations of the quiver in question and use in afundamental way: ( i ) the study of the log-concavity property for weight spaces of semi-invariants which, in turn, was motivated by A. Okounkov’s log-concavity ex-conjecture(see [14]); ( ii ) the H. Derksen and Weyman’s notion of θ -stable decomposition for dimen-sion vectors (see [19]). Regarding the implication ( ) = ⇒ ( ) , we want to point out that aproof can also be obtained from the work of C. Ringel [48] on rational invariants for tamequivers, or from the work of A. Schofield [53] on the birational classification of modulispaces of representations for quivers. Our proof of ( ) = ⇒ ( ) follows from the generalreduction result described in Theorem 1.2 below (see also Corollary 5.5). ur next goal in this paper is to extend the equivalence ( ) ⇐⇒ ( ) of Theorem 1.1 toother classes of algebras. A fundamental role in achieving this goal is played by the fol-lowing reduction technique. Let Λ be the bound quiver algebra of a bound quiver ( Q, R ) and let E = ( E , . . . , E t ) be an orthogonal exceptional sequence of finite-dimensional rep-resentations of Λ . Using the A ∞ -formalism, one can show that E gives rise to a triangularalgebra Λ E and an equivalence F E of categories from rep ( Λ E ) to the subcategory filt E of rep ( Λ ) . (The details of our notations can be found in Section 4 and Section 5.) Denote by Q E the Gabriel quiver of the (smaller) algebra Λ E . Consider a dimension vector d ′ of Q E and set d = P ≤ i ≤ t d ′ ( i ) dim E i . Now, we can state our next result. Theorem 1.2.
Keep the same notations as above. Assume that rep ( Λ E , d ′ ) is an irreducible rep-resentation variety containing a Schur representation and let C be an irreducible component of rep ( Λ, d ) such that C ∩ filt E ( d ) = ∅ . Then, k ( rep ( Λ E , d ′ )) GL ( d ′ ) and k ( C ) GL ( d ) are isomorphic(as k -algebras). Next, we focus on canonical algebras which were discovered and studied by Ringel[47]. They form a distinguished class of algebras of global dimension two and play animportant role in the representation theory of algebras. Moreover, W. Geigle and Lenzingfound in [28] a beautiful interpretation of canonical algebras and their representationsin terms of coherent sheaves over weighted projective lines. The invariant theory forcanonical algebras in the regular case has been investigated in a number of papers, see[6], [5], [21], [22], [56]. By applying Theorem 1.2 to tame canonical algebras, we are ableto describe the fields of rational invariants when the dimension vector in question is notnecessarily regular. More precisely, we have:
Theorem 1.3.
Let Λ be a canonical algebra. The following conditions are equivalent: (1) Λ is tame; (2) for each generic root d of Λ and each indecomposable irreducible component C of rep ( Λ, d ) , k ( C ) GL ( d ) ≃ k or k ( t ) . Note that the condition on the fields of rational invariants in Theorem 1.1(4) and The-orem 1.3(2) simply says that the rational quotients rep ( Q, d ) / GL ( d ) and C/ GL ( d ) are(birationally equivalent to) a point or P whenever d is a generic root, and this is verymuch in sync with the philosophy behind the tameness of an algebra.In [33], V. Kac showed that the problem of computing fields of rational invariants forquivers can be reduced to the case where the dimension vectors involved are Schur roots.In Proposition 4.7, we explain how to extend this result to fields of rational invariants forarbitrary finite-dimensional algebras. As a direct consequence of Theorem 1.1, Theorem1.3, and Proposition 4.7, we have: Proposition 1.4.
Let Λ be either a tame path algebra or a tame canonical algebra. If d is adimension vector of Λ and C is an irreducible component of rep ( Λ, d ) then k ( C ) GL ( d ) ≃ k ( t , . . . , t N ) , where N is the sum of the multiplicities of the isotropic imaginary roots that occur in the genericdecomposition of d in C . Let us mention that our approach to proving Proposition 1.4 when Λ is a tame pathalgebra gives a short and conceptual proof of Ringel’s result in [48]. he layout of the paper is as follows. In Section 2, we recall some fundamental resultsfrom quiver invariant theory. This includes A. King’s construction of moduli spaces ofquiver representations, and Derksen-Weyman’s results on the θ -stable decomposition fordimension vectors of quivers. The proof of Theorem 1.1 can be found in Section 3. InSection 4, we review some important results about representation varieties and their irre-ducible components. We prove Theorem 1.2 in Section 5 where we also review importantproperties of categories of the form filt E which are due to B. Keller, and W. Crawley-Boevey and J. Schr ¨oer. In Section 6, we first review some fundamental results aboutcanonical algebras, including a description of the indecomposable irreducible compo-nents for tame canonical algebras due to G. Bobi ´nski and Skowro ´nski, and Ch. Geissand Schr ¨oer. Furthermore, we present a systematic approach to finding short orthogo-nal exceptional sequences of representations via the study of facets of cones of effectiveweights for quivers with relations; in particular, this requires an extension of the Derksen-Weyman’s notion of θ -stable decomposition to quivers with relations. We prove Theorem1.3 and Proposition 1.4 at the end of this final section.2. Q UIVER INVARIANT THEORY
Let Q = ( Q , Q , t, h ) be a finite quiver with vertex set Q and arrow set Q . The twofunctions t, h : Q → Q assign to each arrow a ∈ Q its tail ta and head ha , respectively.A representation V of Q over k is a collection ( V ( i ) , V ( a )) i ∈ Q ,a ∈ Q of finite-dimensional k -vector spaces V ( i ) , i ∈ Q , and k -linear maps V ( a ) ∈ Hom k ( V ( ta ) , V ( ha )) , a ∈ Q . Thedimension vector of a representation V of Q is the function dim V : Q → Z defined by ( dim V )( i ) = dim k V ( i ) for i ∈ Q . Let S i be the one-dimensional representation of Q atvertex i ∈ Q and let us denote by e i its dimension vector. By a dimension vector of Q ,we simply mean a function d ∈ Z Q ≥ .Given two representations V and W of Q , we define a morphism ϕ : V → W to bea collection ( ϕ ( i )) i ∈ Q of k -linear maps with ϕ ( i ) ∈ Hom k ( V ( i ) , W ( i )) for each i ∈ Q ,and such that ϕ ( ha ) V ( a ) = W ( a ) ϕ ( ta ) for each a ∈ Q . We denote by Hom Q ( V, W ) the k -vector space of all morphisms from V to W . Let V and W be two representations of Q .We say that V is a subrepresentation of W if V ( i ) is a subspace of W ( i ) for each i ∈ Q and V ( a ) is the restriction of W ( a ) to V ( ta ) for each a ∈ Q . In this way, we obtain theabelian category rep ( Q ) of all quiver representations of Q .Given two quiver representations V and W , we have the Ringel’s [49] canonical exactsequence:(1) → Hom Q ( V, W ) → M i ∈ Q Hom k ( V ( i ) , W ( i )) d VW − → M a ∈ Q Hom k ( V ( ta ) , W ( ha )) , where d VW (( ϕ ( i ) i ∈ Q ) = ( ϕ ( ha ) V ( a ) − W ( a ) ϕ ( ta )) a ∈ Q and coker ( d VW ) = Ext ( V, W ) . The Ringel form of Q is the bilinear form h· , ·i Q : Z Q × Z Q → Z defined by(2) h d , e i Q = X i ∈ Q d ( i ) e ( i ) − X a ∈ Q d ( ta ) e ( ha ) . (When no confusion arises, we drop the subscript Q .) It follows from (1) and (2) that(3) h dim V, dim W i = dim k Hom Q ( V, W ) − dim k Ext ( V, W ) . he Tits form of Q is the integral quadratic form q Q : Z Q → Z defined by q Q ( d ) = h d , d i for d ∈ Z Q .2.1. Semi-invariants of quivers.
Let d be a dimension vector of Q . The representationspace of d -dimensional representations of Q is the affine space rep ( Q, d ) = Y a ∈ Q Mat d ( ha ) × d ( ta ) ( k ) . The group GL ( d ) = Q i ∈ Q GL ( d ( i ) , k ) acts on rep ( Q, d ) by simultaneous conjugation, i.e.,for g = ( g ( i )) i ∈ Q ∈ GL ( d ) and V = ( V ( a )) a ∈ Q ∈ rep ( Q, d ) , we define g · V by ( g · V )( a ) = g ( ha ) V ( a ) g ( ta ) − , ∀ a ∈ Q . In this way, rep ( Q, d ) becomes a rational representation of the linearly reductive group GL ( d ) and the GL ( d )− orbits in rep ( Q, d ) are in one-to-one correspondence with the iso-morphism classes of the d -dimensional representations of Q . From now on, we assume that Q is a quiver without oriented cycles. Under thisassumption, one can show that there is only one closed GL ( d )− orbit in rep ( Q, d ) , andhence the invariant ring I ( Q, d ) := k [ rep ( Q, d )] GL ( d ) is exactly the base field k .Now, consider the subgroup SL ( d ) ⊆ GL ( d ) defined by SL ( d ) = Y i ∈ Q SL ( d ( i ) , k ) . Although there are only constant GL ( d )− invariant polynomial functions on rep ( Q, d ) ,the action of SL ( d ) on rep ( Q, d ) provides us with a highly non-trivial ring of semi-invariants.Note that any θ ∈ Z Q defines a rational character χ θ : GL ( d ) → k ∗ by χ θ (( g ( i )) i ∈ Q ) = Y i ∈ Q ( det g ( i )) θ ( i ) . In this way, we can identify Γ = Z Q with the group X ⋆ ( GL ( d )) of rational charactersof GL ( d ) , assuming that d is a sincere dimension vector. In general, we have only thenatural epimorphism Γ → X ∗ ( GL ( d )) . We also refer to the rational characters of GL ( d ) as(integral) weights of Q .Let us now consider the ring of semi-invariants SI ( Q, d ) := k [ rep ( Q, d )] SL ( d ) . As SL ( d ) is the commutator subgroup of GL ( d ) and GL ( d ) is linearly reductive, we have SI ( Q, d ) = M θ ∈ X ⋆ ( GL ( d )) SI ( Q, d ) θ , where SI ( Q, d ) θ = { f ∈ k [ rep ( Q, d )] | gf = θ ( g ) f for all g ∈ GL ( d ) } is called the space of semi-invariants of weight θ .If d ∈ Γ , we define θ = h d , ·i by θ ( i ) = h d , e i i , ∀ i ∈ Q . Similarly, one can define the weight τ = h· , d i . .2. Reciprocity and polynomiality properties.
The following remarkable properties ofweight spaces of semi-invariants, due to Derksen and Weyman [17, 18], play a crucial rolein the proof of Theorem 1.1.
Proposition 2.1 (Reciprocity Property) . [17, Corollary 1] Let Q be a quiver and let d and e betwo dimension vectors of Q . Then dim k SI ( Q, e ) h d , ·i = dim k SI ( Q, d ) − h· , e i . For two dimension vectors d and e , we define d ◦ e = dim k SI ( Q, e ) h d , ·i = dim k SI ( Q, d ) − h· , e i . The next result tells us how the dimensions ( N d ) ◦ e and d ◦ ( N e ) vary as N ∈ Z ≥ varies. Proposition 2.2. [18, Corollary 1]
Let Q be a quiver and let d and e be two dimension vectorsof Q such that d ◦ e = . Then there exist polynomials P, Q ∈ Q [ X ] (both depending on d and e )with P ( ) = Q ( ) = , and ( N d ) ◦ e = P ( N ) , ∀ N ≥ and d ◦ ( N e ) = Q ( N ) , ∀ N ≥ Remark 2.3.
Note that Proposition 2.2 immediately implies the fact that weight spacesof semi-invariants of quivers are asymptotically log-concave in both arguments (see [14]).However, they are not log-concave in general.2.3.
Moduli spaces of quiver representations.
In [37], King constructed, via GIT, modulispaces of representations for finite-dimensional algebras. Let d be a dimension vector of Q . Then the one-dimensional torus T = { ( t Id d ( i ) ) i ∈ Q | t ∈ k ∗ } ⊆ GL ( d ) acts trivially on rep ( Q, d ) , and so there is a well-defined action of PGL ( d ) := GL ( d ) /T on rep ( Q, d ) . Definition 2.4. [37, Definition 2.1] Let θ ∈ Z Q be an integral weight of Q . A representa-tion V ∈ rep ( Q, d ) is said to be:(1) θ -semi-stable if there exists a semi-invariant f ∈ SI ( Q, d ) nθ with n ≥ , such that f ( V ) = ;(2) θ -stable if there exists a semi-invariant f ∈ SI ( Q, d ) nθ with n ≥ , such that f ( V ) = and, furthermore, the GL ( d ) -action on the principal open subset defined by f isclosed and dim GL ( d ) V = dim PGL ( d ) .Now, consider the (possibly empty) open subsets rep ( Q, d ) ssθ = { V ∈ rep ( Q, d ) | V is θ -semi-stable } and rep ( Q, d ) sθ = { V ∈ rep ( Q, d ) | V is θ -stable } of d -dimensional θ (-semi)-stable representations.We say that a dimension vector d is θ (-semi)-stable if there exists θ (-semi)-stable rep-resentation V ∈ rep ( Q, d ) . A dimension vector d is called a Schur root if there exists a epresentation V ∈ rep ( Q, d ) such that End Q ( V ) = k ; we call such a representation a Schur representation . Note that if d is θ -stable for some integral weight θ then d is a Schurroot.The GIT-quotient of rep ( Q, d ) ssθ by PGL ( d ) is M ( Q, d ) ssθ := Proj ( M n ≥ SI ( Q, d ) nθ ) . This is an irreducible projective variety whose closed points parameterize the closed GL ( d ) -orbits in rep ( Q, d ) ssθ .From geometric invariant theory we also know that M ( Q, d ) ssθ contains a (possiblyempty) open subset M ( Q, d ) sθ which is a geometric quotient of rep ( Q, d ) sθ by PGL ( d ) . Remark 2.5.
Let d be a θ -stable dimension vector where θ ∈ Z Q . It follows from Rosen-licht’s theorem [51] that k ( M ( Q, d ) ssθ ) ≃ k ( rep ( Q, d )) GL ( d ) . Also, a simple dimensioncount shows that dim M ( Q, d ) ssθ = − h d , d i . Theorem 2.6. [52]
Let d be a dimension vector of Q . Then d is a Schur root if and only if d is θ d -stable where θ d = h d , ·i − h· , d i . The θ -stable decomposition for dimension vectors. In this section, the θ -stable de-composition for dimension vectors, due to Derksen and Weyman [19], is reviewed.Let Q be a quiver and let d be a θ -semi-stable dimension vector of Q where θ ∈ Z Q .(Note that in particular this implies θ ( d ) := P i ∈ Q θ ( i ) d ( i ) = .) One of the funda-mental results about semi-stable representations is the King’s [37] numerical criterion for(semi-)stability that says that a representation V ∈ rep ( Q, d ) is θ -semi-stable if and only if θ ( dim V ′ ) ≤ for all subrepresentations V ′ of V . Furthermore, V is θ -stable if and only if θ ( dim V ′ ) < 0 for all proper subrepresentations V ′ of V .We define rep ( Q ) ssθ to be the full subcategory of rep ( Q ) consisting of all θ -semi-stablerepresentations. Similarly, we denote by rep ( Q ) sθ the full subcategory of rep ( Q ) consistingof all θ -stable representations. (Of course, the zero representation is always semi-stablebut not stable.)It is easy to see that rep ( Q ) ssθ is a full exact abelian subcategory of rep ( Q ) which isclosed under extensions and whose simple objects are precisely the θ -stable represen-tations. Moreover, rep ( Q ) ssθ is Artinian and Noetherian, and hence every θ -semi-stablerepresentation has a Jordan-H ¨older filtration in rep ( Q ) ssθ .Following [19], we say that d = d ∔ d ∔ . . . ∔ d l is the θ -stable decomposition of d if a general representation in rep ( Q, d ) has a Jordan-H ¨older filtration in rep ( Q ) ssθ with factors of dimensions d , . . . , d l in some order.Recall that a root of Q is just the dimension vector of an indecomposable representationof Q . We say that a root d is real if h d , d i = . If h d , d i = , d is said to be an imaginaryisotropic root. Finally, we say that d is an imaginary but non-isotropic root if h d , d i < 0 .In what follows, we write m · d instead of d ∔ d ∔ . . . ∔ d | {z } m . The projective scheme Proj ( L n ≥ S m ( SI ( Q, d ) nθ )) is denoted by S m ( M ( Q, d ) ssθ ) . (Here, S m ( SI ( Q, d ) nθ ) is the m th ymmetric power of SI ( Q, d ) nθ .) The following theorem of Derksen and Weyman, playsa crucial role in our study. Theorem 2.7. [19]
Let d = d ∔ d ∔ . . . ∔ d l be the θ -stable decomposition of d and let m be apositive integer. (1) The θ -stable decomposition of m d is m d = [ m d ] ∔ . . . ∔ [ m d l ] , where [ m d i ] = (cid:14) m · d i if d i is real or isotropic ; m d i otherwise. (2) Suppose that d = m · d ∔ . . . ∔ m n · d n with m i positive integers and d i = d j for ≤ i = j ≤ n . Then SI ( Q, d ) mθ ≃ n O i = S m i ( SI ( Q, d i ) mθ ) and M ( Q, d ) ssθ ≃ S m ( M ( Q, d ) ssθ ) × . . . × S m n ( M ( Q, d n ) ssθ ) . Remark 2.8. ( ) Note that in particular Theorem 2.7(1) says that if d is θ -stable and h d , d i < 0 then m d is still θ -stable for all integers m ≥ . ( ) Although the above isomorphism between moduli spaces is not explicitly mentionedin [19, Theorem 3.20], it follows immediately from its proof.3. P
ROOF OF T HEOREM Q be a quiver without oriented cycles and let d be a θ -semi-stable dimension vector of Q where θ ∈ Z Q . Let π : rep ( Q, d ) ssθ → M ( Q, d ) ssθ be the quotient map and let ξ be aclosed point of M ( Q, d ) ssθ . Then, the fiber π − ( ξ ) contains a unique orbit GL ( d ) M whichis closed in rep ( Q, d ) ssθ . As shown by King in [37], this is equivalent to saying that M = L li = M m i i with M , . . . , M l pairwise non-isomorphic θ -stable representations (call sucha representation M θ -polystable). Next, consider the local quiver setup ( Q ξ , d ξ ) where Q ξ has vertex set {
1, . . . , l } and dim k Ext ( M i , M j ) arrows from vertex i to vertex j ; thedimension vector d ξ of Q ξ is defined to be the vector ( m , . . . , m l ) . It was proved in [1]that M ( Q, d ) ssθ is smooth at ξ if and only if the ring of invariants k [ rep ( Q ξ , d ξ )] GL ( d ξ ) is apolynomial algebra. If this is the case, we call ( Q ξ , d ξ ) a coregular quiver setup. Lemma 3.1.
Let Q be a Euclidean quiver and let δ be the unique isotropic Schur root of Q . If θ ∈ Z Q is a weight such that δ is θ -stable then M ( Q, δ ) ssθ ≃ P . Remark 3.2.
This result is well-known. For example, it follows from the work of Domokosand Lenzing on moduli spaces of regular representations for concealed-canonical alge-bras (see [22, Corollary 7.3]). Here, we give a direct proof working entirely within thecategory of representations of the tame quiver Q . More precisely, we make essential useof: (i) the rationality of k ( rep ( Q, δ )) GL ( δ ) ; (ii) the Derksen-Weyman θ -stable decompositionfor dimension vectors; (iii) Adriaenssens-Le Bruyn local quiver technique. Proof.
First, note that the moduli space M ( Q, δ ) ssθ is a projective curve since h δ, δ i = .Moreover, if follows from the work of Ringel [50] or Schofield [53] that M ( Q, δ ) ssθ is arational variety. (For another proof of this rationality property, see our Corollary 5.5.) To rove that M ( Q, δ ) ssθ is precisely P , it remains to show that M ( Q, δ ) ssθ is smooth (see forexample [31, Exercise I.6.1]). For this, let M = L li = M m i i be a δ -dimensional θ -polystablerepresentation with M , . . . , M l pairwise non-isomorphic θ -stable representations, andlet ( Q M , d M ) be the corresponding local quiver setup. Of course, if l = then the ring ofinvariants k [ rep ( Q M , d M )] GL ( d M ) is just k [ t ] . So, let us assume that l > 1 . Note that sinceeach dim M i is a Schur root smaller (coordinatewise) than δ , dim M i has to be a real Schurroot, and hence M i is an exceptional representation. In particular, Q M has no loops.If at least one of the m i , say m , is bigger than one then we claim that Q M has nooriented cycles. To see why this is so, consider the θ -polystable representation M ′ = M m − ⊕ L li = M m i i and denote its dimension vector by d ′ . Since d ′ < δ , we know thatall the Schur roots that occur in the canonical decomposition of d ′ are real, and hence GL ( d ′ ) acts with a dense orbit on rep ( Q, d ′ ) (see [34, Corollary 1]). In particular, M ′ isthe unique d ′ -dimensional θ -polystable representation, up to isomorphism. Hence, d ′ =( m − ) · dim M ∔ . . . ∔ m l · dim M l is the θ -stable decomposition of d ′ . It now follows from[19, Proposition 3.18(d)] that Q M has no oriented cycles, and so k [ rep ( Q M , d M )] GL ( d M ) = k .Next, let I ⊂ {
1, . . . , l } be a proper subset. Denote by d I = P i ∈ I m i dim M i and note that GL ( d I ) acts with a dense orbit on rep ( Q, d I ) as d I < δ . Arguing as before we deduce thatthe (full) subquiver of Q M with vertex set I has no oriented cycles.From the discussion above, it remains to look into the case when Q M has oriented cy-cles, m i = for all ≤ i ≤ l , and any oriented cycle in Q M uses all the vertices. So, we canreorder the vertices of Q M , if needed, so that if k i,j denotes the number of arrows from i to j , ≤ i, j ≤ l , then k · k · . . . · k l − · k l,1 = , and the rest of the k i,j are zero. Further-more, we have that k + . . . + k l,1 = l as h δ, δ i = ; in other words, Q M is just the oriented l -cycle. But for this quiver and the thin sincere dimension vector d M , the correspondingring of invariants is known to be a polynomial algebra (see for example [10]).Finally, using the local quiver technique and the fact that all local quiver setups associ-ated to M ( Q, δ ) ssθ are coregular, we conclude that M ( Q, δ ) ssθ is smooth. This finishes theproof. (cid:3) Now, we are ready to prove Theorem 1.1.
Proof of Theorem 1.1.
First, let us prove the implication ( ) = ⇒ ( ) . If Q is a Dynkin quiver, M ( Q, d ) ssθ is just a point as GL ( d ) acts with a dense orbit on rep ( Q, d ) . Next, let assumethat Q is a Euclidean quiver and let d be a θ -semi-stable dimension vector where θ ∈ Z Q .If the isotropic Schur root δ of Q does not occur in the θ -stable decomposition of d thenTheorem 2.7(2) tells us that M ( Q, d ) ssθ is just a point. Otherwise, let m be the multiplicityof δ in the θ -stable decomposition of d . It follows again from Theorem 2.7(2) that M ( Q, d ) ssθ ≃ S m ( M ( Q, δ ) ssθ ) . Furthermore, M ( Q, δ ) ssθ ≃ P by Lemma 3.1, and so we deduce that M ( Q, d ) ssθ ≃ P m .Next, we prove the implication ( ) = ⇒ ( ) . Let d be a θ -semi-stable dimension vectorwhere θ is an integral weight. Then, we know that M ( Q, d ) ssθ ≃ P m ֒ → P r . Choosean integer l ≥ for which the graded algebra L n ≥ SI ( Q, d ) n ( lθ ) is generated by thedegree one component SI ( Q, d ) lθ . Pulling back the line bundle O ( ) over P r , we get aline bundle over M ( Q, d ) ssθ which has to be of the form O ( d ) . The image of the map ( P r , O ( N )) → Γ ( P m , O ( Nd )) is precisely SI ( Q, d ) N ( lθ ) , and furthermore this map is sur-jective for sufficiently large values of N (see for example [31, Exercise II.5.9]). Hence, dim k SI ( Q, d ) N ( lθ ) = (cid:18) Nd + mm (cid:19) for sufficiently large values of N . From Proposition 2.2, we deduce that dim k SI ( Q, d ) nθ = (cid:18) qn + mm (cid:19) , for all integers n ≥ where q = dl . This clearly shows that the weight spaces of semi-invariants of Q are log-concave.To prove that ( ) = ⇒ ( ) , we follow the arguments in [14, Sec. 3.4]. Let us assumeto the contrary that Q is a wild quiver. Under this assumption, we can always find anon-isotropic imaginary Schur root d ′ . Then d ′ is stable with respect to the weight θ d ′ = h d ′ , ·i − h· , d ′ i by Theorem 2.6. We can clearly assume that d ′ is sincere since otherwisewe can just simply work with the full subquiver of Q whose vertex set consists of thosevertices i ∈ Q for which d ′ ( i ) > 0 . So, we can write θ d ′ = h d ′′ , ·i for a unique dimensionvector d ′′ of Q due to [32, Lemma 6.5.7] (see also [17, Theorem 1]).By Remark 2.8, we know that m d ′ is still h d ′′ , ·i -stable for any integer m ≥ . Con-sequently, the dimension of the moduli space M ( Q, m d ′ ) ss h d ′′ , ·i is − m h d ′ , d ′ i . So, forany integer m ≥ , the Hilbert function ( N d ′′ ) ◦ ( m d ′ ) is a polynomial in N of degree − m h d ′ , d ′ i . As − h d ′ , d ′ i ≥ , we have that for sufficiently large N , ( N d ′′ ) ◦ ( d ′ ) > (( N d ′′ ) ◦ d ′ ) , which is equivalent to dim k SI ( Q, d ) > ( dim k SI ( Q, d ) θ ) , where d = N d ′′ and θ = − h· , d ′ i . But this is a contradiction. So, Q must be a tame quiver.Finally, it remains to prove the equivalence ( ) ⇐⇒ ( ) . First, let us assume that foreach Schur root d , the field of rational invariants k ( rep ( Q, d )) GL ( d ) is isomorphic to k or k ( t ) . For any dimension vector d of Q , we have tr . deg k k ( rep ( Q, d )) GL ( d ) = dim rep ( Q, d )− dim GL ( d )+ min { dim k End Q ( V ) | V ∈ rep ( Q, d ) } . (This formula follows immediately from Rosenlicht’s theorem [51, Theorem 2] and thefiber dimension theorem [55].) So, if d is a Schur root of Q then it is easy to see that q Q ( d ) ≥ . Now, let d be a dimension vector of Q and consider its canonical decomposi-tion d = d ⊕ . . . ⊕ d m , where d i , ≤ i ≤ m , are Schur roots and ext ( d i , d j ) = , ∀ ≤ i = j ≤ m . In particular,we have q Q ( d ) ≥ P ≤ i ≤ m q Q ( d i ) ≥ . This implies that Q is a Dynkin or Euclidean quiver.The implication ( ) = ⇒ ( ) follows from Corollary 5.5 or [48]. (cid:3)
4. Q
UIVERS WITH RELATIONS
Given a quiver Q , its path algebra kQ has a k -basis consisting of all paths and themultiplication in kQ is given by concatenation of paths. It is easy to see that any finite-dimensional left kQ -module defines a representation of Q , and vice-versa. Furthermore, he category mod ( kQ ) of finite-dimensional left kQ -modules is equivalent to the category rep ( Q ) . In what follows, we identify mod ( kQ ) and rep ( Q ) , and use the same notation fora module and the corresponding representation.A relation in Q with coefficients in k is an element r ∈ kQ of the form r = l X i = λ i p i , where λ , . . . , λ l ∈ k are non-zero scalars and p , . . . , p l are paths in kQ of length at leasttwo with tp = · · · = tp l and hp = · · · = hp l .A set R of relations is said to be minimal if for every r ∈ R , r does not belong to thetwo-sided ideal h R \ { r } i of kQ generated by R \ { r } . A bound quiver consists of a quiver Q and a minimal finite set R of relations such that there exists a positive integer L withthe property that any path in Q of length at least L belongs to the two sided ideal h R i of kQ generated by R . We call kQ/ h R i the bound quiver algebra of the bound quiver ( Q, R ) . Arepresentation M of kQ/ h R i (or ( Q, R ) ) is just a representation M of Q such that M ( r ) = for all r ∈ R .It is well-known that any finite-dimensional basic algebra Λ is isomorphic to the boundquiver algebra of a bound quiver ( Q Λ , R ) , where Q Λ is the Gabriel quiver of Λ . Note thatthe set of relations R is not uniquely determined by Λ . We say that Λ is a triangular algebraif its Gabriel quiver has no oriented cycles.Fix a bound quiver ( Q, R ) and let Λ = kQ/ h R i be its bound quiver algebra. The cat-egory mod ( Λ ) of finite-dimensional left Λ -modules is equivalent to the category rep ( Λ ) of representations of Λ . As before, we identify mod ( Λ ) and rep ( Λ ) , and make no distinc-tion between Λ -modules and representations of Λ . By a Λ -module, we always mean afinite-dimensional left Λ -module. For each vertex v ∈ Q , we denote by e v the primitiveidempotent corresponding to v .4.1. Representation varieties and the Tits form.
Let d be a dimension vector of Λ (orequivalently, of Q ). The variety of d -dimensional representations of Λ is the affine variety rep ( Λ, d ) = { M ∈ rep ( Q, d ) | M ( r ) = ∀ r ∈ R } . It is clear that rep ( Λ, d ) is a GL ( d ) -invariant closed subset of rep ( Q, d ) . Note that rep ( Λ, d ) does not have to be irreducible. We call rep ( Λ, d ) the representation/module variety of d -dimensional representations/modules of Λ .In what follows, we list a series of important results describing the representation typeof a (triangular) algebra Λ in terms of the so-called Tits form of Λ . Proposition 4.1. [16] If Λ is a tame algebra then dim GL ( d ) ≥ dim rep ( Λ, d ) , for each dimension vector d of Λ . Assume form now on that Λ has finite global dimension; this happens, for example,when Q has no oriented cycles. The Ringel form of Λ is the bilinear form h· , ·i Λ : Z Q × Z Q → Z defined by h d , e i Λ = X l ≥ (− ) l X i,j ∈ Q dim k Ext lΛ ( S i , S j ) d ( i ) e ( j ) . ote that if M is a d -dimensional representation of Λ and N is an e -dimensional repre-sentation of Λ then h d , e i Λ = X l ≥ (− ) l dim k Ext lΛ ( M, N ) . The quadratic form induced by h· , ·i Λ is denoted by χ Λ .The Tits form of Λ is the integral quadratic form q Λ : Z Q → Z defined by q Λ ( d ) := X i ∈ Q d ( i ) − X i,j ∈ Q dim k Ext ( S i , S j ) d ( i ) d ( j ) + X i,j ∈ Q dim k Ext ( S i , S j ) d ( i ) d ( j ) . Next, let us assume that Λ is triangular. Under this assumption, the Tits form q Λ isrelated to the geometry of the varieties of representations of Λ in the following way. First, r ( i, j ) := | R ∩ e j h R i e i | is precisely dim k Ext ( S i , S j ) , ∀ i, j ∈ Q , as shown by K. Bongartz in[11]. So, we can write q Λ ( d ) = X i ∈ Q d ( i ) − X a ∈ Q d ( ta ) d ( ha ) + X i,j ∈ Q r ( i, j ) d ( i ) d ( j ) . Now, let d be a dimension vector of Λ and M ∈ rep ( Λ, d ) . By Krull’s Principal IdealTheorem (see for example [25]), we have dim M rep ( Λ, d ) ≥ X a ∈ Q d ( ta ) d ( ha ) − X i,j ∈ Q r ( i, j ) d ( i ) d ( j ) . In particular, we have that(4) q Λ ( d ) ≥ dim GL ( d ) − dim rep ( Λ, d ) . This inequality together with Proposition 4.1 proves:
Theorem 4.2. [16] If Λ is a tame triangular algebra then q Λ ( d ) ≥ for each dimension vector d of Λ . We should point out the the converse of this theorem is false in general (see [7, Example2.1]). However, there are important classes of finite-dimensional algebras for which theconverse holds true.
Theorem 4.3. [8, 12]
Let Λ be either a quasi-tilted algebra or a strongly simply-connected algebra.Then, Λ is tame if and only if the Tits form q Λ is weakly positive semi-definite. The generic decomposition for irreducible components.
Let Λ be the bound quiveralgebra of a bound quiver ( Q, R ) , d a dimension vector of Λ , and C an irreducible compo-nent of rep ( Λ, d ) .We say that C is an indecomposable irreducible component if C has a non-empty opensubset of indecomposable representations. We call C a Schur irreducible component if C contains a Schur representation. Note that a Schur irreducible component is always inde-composable. The converse is always true for finite-dimensional path algebras. Finally, wesay that d is a generic root of Λ if rep ( Λ, d ) has an indecomposable irreducible component.Now, let us consider a decomposition d = d + . . . + d t where d i ∈ Z Q ≥ , 1 ≤ i ≤ t . If C i is a GL ( d i ) -invariant subset of rep ( Λ, d i ) , ≤ i ≤ t , we denote by C ⊕ . . . ⊕ C t theconstructible subset of rep ( Λ, d ) consisting of all modules isomorphic to direct sums ofthe form L ti = X i with X i ∈ C i , ∀ ≤ i ≤ t . he following fundamental result, which defines the generic decomposition for irre-ducible components in representation varieties, is due to J. A. de la Pe ˜na [16, Section 1.3]and Crawley-Boevey-Schr ¨oer [15, Theorem 1.1]. Theorem 4.4. If C is an irreducible component of rep ( Λ, d ) then there are unique generic roots d , . . . , d t of Λ such that d = d + . . . + d t and C = C ⊕ . . . ⊕ C t for some indecomposable irreducible components C i of rep ( Λ, d i ) , 1 ≤ i ≤ t . Moreover, theindecomposable irreducible components C i , 1 ≤ i ≤ t, are uniquely determined by this property. With the notations of the theorem above, we call d = d ⊕ . . . ⊕ d t the generic decom-position of d in C , and C = C ⊕ . . . ⊕ C t the generic decomposition of C .Let us record the following useful lemma: Lemma 4.5. [16, Lemma 1.3]
Let C be an irreducible component in rep ( Λ, d ) and let C = C ⊕ . . . ⊕ C t be its generic decomposition where C i ⊆ rep ( Λ, d i ) , ≤ i ≤ t , are indecomposableirreducible components. Then, dim GL ( d ) − dim C ≥ t X i = ( dim GL ( d i ) − C i ) . Now, we are ready to prove:
Proposition 4.6.
Assume that the field of rational invariants k ( C ) GL ( d ) ≃ k or k ( t ) for eachgeneric root d of Λ and each indecomposable irreducible component C of rep ( Λ, d ) . Then, thefollowing statements hold true. (1) If Λ is a triangular algebra then q Λ ( d ) ≥ for each dimension vector d of Λ . (2) If Λ is a quasi-tilted algebra or a strongly simply-connected algebra then Λ is tame.Proof. (1) Let d be a generic root of Λ and let C be an indecomposable irreducible compo-nent of rep ( Λ, d ) . Choose M in C so that dim k End Λ ( M ) = min { dim k End Λ ( M ′ ) | M ′ ∈ C } .Then, we have that tr.deg k k ( C ) GL ( d ) = dim C − dim GL ( d ) + dim k End Λ ( M ) , and so, dim GL ( d ) − dim C ≥ . But this remains true for any dimension vector d of Λ andany irreducible component C of rep ( Λ, d ) by Lemma 4.5. From this and inequality (4), wededuce that q Λ ( d ) ≥ for each dimension vector d .(2) This follows now from the first part and Theorem 4.3. (cid:3) In what follows, we explain how to reduce the problem of describing fields of rationalinvariants on irreducible components of representation varieties to the case where theirreducible components involved are indecomposable. This was already done by Kac in[34] in the context of quivers with no relations. It turns out that Kac’s proof in loc. cit. can be extended to arbitrary finite-dimensional algebras. Indeed, let Λ = kQ/ h R i be thebound quiver algebra of a bound quiver ( Q, R ) , d a dimension vector, and C an irreduciblecomponent of rep ( Λ, d ) . et d = d ⊕ m ⊕ . . . ⊕ d ⊕ m n n be the generic decomposition of d in C where d , . . . , d n aredistinct generic roots of Λ , and m , . . . , m n are positive integers. Next, we assume thatthe generic decomposition of C is of the form C = C ⊕ m ⊕ . . . ⊕ C ⊕ m n n , where C i ⊆ rep ( Λ, d i ) , ≤ i ≤ n , are indecomposable irreducible components. Fix adecomposition k d ( v ) = k d ( v ) ⊕ . . . ⊕ k d ( v ) | {z } m ⊕ . . . ⊕ k d n ( v ) ⊕ . . . ⊕ k d n ( v ) | {z } m n for each vertex v ∈ Q , and then embed e C := C m × . . . × C m n n diagonally in C and G := GL ( d ) m × . . . × GL ( d n ) m n diagonally in GL ( d ) .Denote by T ( i ) the one-dimensional torus in GL ( d i ) , set T := ( T ( ) ) m × . . . × ( T ( n ) ) m n ,and note that T ⊆ C GL ( d ) ( M ) for any M ∈ e C . Next, we choose an open and dense subset C i,0 of C i consisting of indecomposable representations such that GL ( d ) C m × . . . × C m n n,0 = C. Furthermore, for any representation M ∈ C m × . . . × C m n n,0 , it is not difficult to see that T is a maximal torus of C GL ( d ) ( M ) (see for example [38, Section 2.2]). Also, note that thecentralizer of T in GL ( d ) is precisely G , and the normalizer N of T in GL ( d ) is N = ( GL ( d ) m ⋊ S m ) × . . . × ( GL ( d n ) m n ⋊ S m n ) . (Here, S m denotes the symmetric group on m elements.)Let us summarize what we have obtained so far:(1) GL ( d ) e C = C ;(2) the generic point M in e C has the property that T is a maximal torus of C GL ( d ) ( M ) ;(3) e C is an N -invariant closed subvariety of C .In what follows, if R is an integral domain, we denote its field of fractions by Quot ( R ) .Moreover, if K/k is a field extension and m is a positive integer, we define S m ( K/k ) to bethe field ( Quot ( K ⊗ m )) S m which is, in fact, the same as Quot (( K ⊗ m ) S m ) since S m is a finitegroup. Proposition 4.7.
Let C = C ⊕ m ⊕ . . . ⊕ C ⊕ m n n be the generic decomposition of C where C i ⊆ rep ( Λ, d i ) , ≤ i ≤ n , are indecomposable irreducible components, m , . . . , m n are positiveintegers, and d i = d j , ∀ ≤ i = j ≤ n . Then, k ( C ) GL ( d ) ≃ Quot ( n O i = S m i ( k ( C i ) GL ( d i ) /k )) . Proof.
Let π : C C/ GL ( d ) be the rational quotient map for the action of GL ( d ) on C . Now, property ( ) above tells us that the restriction φ = π | e C : e C C/ GL ( d ) is awell-defined dominant rational map.Let M be a generic point in e C , M ∈ φ − ( φ ( M )) , and g ∈ GL ( d ) such that M = gM .Note that T and g − Tg are maximal tori of C GL ( d ) ( M ) , and so T = ( gg ′ ) − T ( gg ′ ) for some g ′ ∈ C GL ( d ) ( M ) . Hence, g := gg ′ ∈ N and M = g M , i.e., NM = φ − ( φ ( M )) . It nowfollows from the universal property of rational quotients that φ is the rational quotientmap for the action of N on e C and hence k ( e C ) N ≃ k ( C ) GL ( d ) . The proof now follows. (cid:3) emark 4.8. Let Λ be a tame finite-dimensional k -algebra, d a Schur root of Λ , and C a Schur irreducible component of rep ( Λ, d ) . Using Proposition 4.1 it is easy to see that tr . deg k k ( C ) GL ( d ) ∈ {
0, 1 } , and hence k ( C ) GL ( d ) ≃ k or k ( t ) ⇐⇒ k ( C ) GL ( d ) is a rational field over k. We refer to the problem that asks to prove that k ( C ) GL ( d ) is rational over k for eachSchur root d of Λ and each Schur irreducible component C of rep ( Λ, d ) as the rationalityproblem for Λ . We have seen that the rationality problem for tame quivers is already set-tled. Moreover, Schofield has obtained in [53] a birational classification of moduli spacesof representations for quivers. In particular, he solved the rationality problem for quiverswhen the dimension vectors involved are indivisible Schur roots. However, the rational-ity problem for wild quivers in the non-indivisible case is a long-standing open problem(see for example [40] and [26]).To tackle the rationality problem for finite-dimensional algebras, we are going to usehomological algebra. This strategy is explained in the next section.5. E XCEPTIONAL SEQUENCES AND RATIONAL INVARIANTS
In this section we explain how exceptional sequences can be used in the study of thefields of rational invariants for finite-dimensional algebras.Let Λ be the bound quiver algebra of a bound quiver ( Q, R ) . A sequence E = ( E , . . . , E t ) of finite-dimensional Λ -modules is called an orthogonal exceptional sequence if the followingconditions are satisfied:(1) E i is an exceptional module, i.e, End Λ ( E i ) = k and Ext lΛ ( E i , E i ) = for all l ≥ and ≤ i ≤ t ;(2) Ext lΛ ( E i , E j ) = for all l ≥ and ≤ i < j ≤ t ;(3) Hom Λ ( E j , E i ) = for all ≤ i < j ≤ t .(If we drop condition ( ) , we simply call E an exceptional sequence .)Given an orthogonal exceptional sequence E , consider the full subcategory filt E of rep ( Λ ) whose objects M have a finite filtration = M ⊆ M ⊆ · · · ⊆ M s = M of submodulessuch that each factor M j /M j − is isomorphic to one the E i . It is clear that filt E is a fullexact subcategory of rep ( Λ ) which is closed under extensions. Moreover, Ringel [50] (seealso [19]) showed that filt E is an abelian subcategory whose simple objects are precisely E , . . . , E t .The category filt E is determined by the Yoneda algebra Ext • Q ( L ti = E i , L ti = E i ) equippedwith its (minimal) A ∞ -algebra structure as shown by Keller [36, 35]. More precisely, letus write Ext lΛ ( L ti = E i , L ti = E i ) = L i,j Ext lΛ ( E j , E i ) and consider the induced R -bimodulestructure on Ext lΛ ( L ti = E i , L ti = E i ) where R is the commutative k -algebra k t . It is clearthat each multiplication map m n of the A ∞ -algebra Ext • Λ ( L ti = E i , L ti = E i ) defines an R -bimodule map m n : Ext ( t M i = E i , t M i = E i ) N nR → Ext ( t M i = E i , t M i = E i ) . ow, let Q E be the quiver with vertices
1, . . . , t and dim k Ext ( E i , E j ) ∗ arrows from i to j . The Maurer-Cartan map is, by definition, the map m = L n ≥ m n , and hence its dualis m ∗ : Ext ( L ti = E i , L ti = E i ) ∗ → kQ E . Note that Q E has no oriented cycles and that iswhy for the dual of the Maurer-Cartan map we can just work with the path algebra of Q E instead of its completed path algebra. Also, the two-sided ideal of kQ E generated by Im ( m ∗ ) is an admissible ideal and, hence, is generated by finitely many relations. Finally,we define Λ E = kQ E / ( Im ( m ∗ )) .Now, we are ready to state the following important result: Theorem 5.1.
Keeping the same notations as above, the following statements are true. (1)
There exists an equivalence of categories F E from rep ( Λ E ) to filt E sending the simple Λ E -module S i at vertex i to E i for all ≤ i ≤ t . (2) Given a dimension vector d ∈ Z Q ≥ , the set filt E ( d ) := { M ∈ rep ( Λ, d ) | M is in filt E } is open in rep ( Λ, d ) . The first part of the theorem above is due to Keller [36, 35] and uses the A ∞ -formalism.The second part was proved by Crawley-Boevey and Schr ¨oer in [15, Corollary 1.5].Let E = ( E , . . . , E t ) be an orthogonal exceptional sequence of Λ -modules and let F E be a functor as in Theorem 5.1(1). Now, consider a dimension vector d ′ of Q E and set d = P ≤ i ≤ t d ′ ( i ) dim E i . Next, we explain how the functor F E gives rise to well-behavedmorphisms at the level of representation varieties: Proposition 5.2.
Keep the same notations as above. Then, there exist a morphism of algebraicgroups ϕ : GL ( d ′ ) → GL ( d ) and a regular morphism f E : rep ( Λ E , d ′ ) → rep ( Λ, d ) such that: (1) f E ( M ′ ) ≃ F E ( M ′ ) for all M ′ ∈ rep ( Λ E , d ′ ) ; (2) f E ( g ′ · M ′ ) = ϕ ( g ′ ) · f E ( M ′ ) for all M ′ ∈ rep ( Λ E , d ′ ) and g ′ ∈ GL ( d ′ ) . Remark 5.3.
Note that this proposition already appears in the context of quivers withno relations in [46]. Furthermore, the proof of Proposition 2.3 in [46] works for arbitraryfinite-dimensional algebras, as well. Nonetheless, for completeness, we include below anexplicit proof.
Proof.
In what follows, we denote by e ′ i the primitive idempotent in Λ E correspondingto vertex i ∈ {
1, 2, . . . , l } . Furthermore, if i and j are two vertices of Q E , p ′ is a linearcombination of paths in kQ E from vertex i to vertex j , and M ′ ∈ rep ( Λ E ) , we define M ′ ( p ′ ) to be the corresponding linear combination of products of matrices. We also denote theresidue class of p ′ in Λ E by p ′ .According to Theorem 2 in [58], we can assume that F = P ⊗ Λ E where P is a finite-dimensional Λ − Λ E -bimodule which is projective as a right Λ E -module. In fact, we canwrite P = L v ∈ Q e v P where e v P = L li = ( e ′ i Λ E ) d Ei ( v ) , ∀ v ∈ Q .Since P is a Λ − Λ E -bimodule, we have that for each arrow a ∈ Q , P ( a ) ∈ Hom Λ E ( e ta P, e ha P ) which, after fixing bases, can be viewed as an l × l block-matrix whose ( i, j ) -block entryis a matrix of size d E i ( ha ) × d E j ( ta ) whose entries are of the form p ′ ∈ Λ E with p ′ a linearcombination of paths from vertex j to vertex i in kQ E . ow, given a representation M ′ ∈ rep ( Λ E , d ′ ) and an arrow a ∈ Q , we define f E ( M ′ )( a ) to be the l × l block-matrix obtained from P ( a ) by replacing each entry p ′ ∈ e ′ i Λ E e ′ j by the d ′ ( i ) × d ′ ( j ) matrix M ′ ( p ′ ) . As for the morphism ϕ , we simply take the natural diagonalembedding of GL ( d ′ ) into GL ( d ) . The proof now follows. (cid:3) For the convenience of the reader, we now recall some fundamental facts from bira-tional invariant theory that will be used in the proof of Theorem 1.2. Let G be a linear al-gebraic group acting regularly on an irreducible variety X . The field k ( X ) G of G -invariantrational functions on X is always finitely generated over k since it is a subfield of k ( X ) which is finitely generated over k . A rational quotient of X by (the action of) G is anirreducible variety Y such that k ( Y ) = k ( X ) G together with the dominant rational map π : X Y induced by the inclusion k ( X ) G ⊂ k ( X ) . Of course, Y is uniquely defined up tobirational isomorphism.Now, a theorem of Rosenlicht [51] tells us that there is a G -invariant open and densesubset X of X such that the restriction of π to X is a dominant regular morphism and π − ( π ( x )) = Gx for all x ∈ X . Furthermore, one can show that a rational quotient π : X Y satisfies the following universal property (see [54, Section 2.4], [44, Remark 2.5]):Let ρ : X Y ′ be a rational map such that ρ − ( ρ ( x )) = Gx for x ∈ X in general position.Then there exists a rational map ρ : Y Y ′ such that ρ = ρ ◦ π . If in addition ρ is dominant then ρ becomes a birational isomorphism.One usually writes X/G in place of Y and call it the rational quotient of X by G .We also need some facts about homogeneous fiber spaces. Let ϕ : H → G be a homo-morphism of algebraic groups and let Z be an H -variety. Consider the action of G × H on G × Z defined by ( g, h ) · ( g ′ , z ) = ( gg ′ ϕ ( h ) − , hz ) and denote by G ∗ H Z the rationalquotient of G × Z by the above action of { } × H . We call X := G ∗ H Z a homogeneous fiberspace . Note that G has a naturally defined rational action on X which, in general, is notregular. However, it is always possible (see [44, Definition 2.12]) to choose a model Y for X such that the G -action on G × Z descends to a regular action on Y , and thus the rationalquotient map π : G × Z Y is G -equivariant.Let us record the following very useful result: Lemma 5.4. [45, Lemma 6.1]
Keep the same notation as above. Then, k ( X ) G and k ( Z ) H areisomorphic as k -algebras. Now, we are ready to prove Theorem 1.2.
Proof of Theorem 1.2.
Denote by T ′ the -dimensional torus in GL ( d ′ ) and by T the -dimensional torus in GL ( d ) . Since filt E ( d ) contains a Schur Λ -module by assumption, weimmediately deduce that ϕ ( T ′ ) ≤ T . Hence, we have a well-defined action of PGL ( d ′ ) on PGL ( d ) induced by the action g ′ · g = g · ϕ ( g ′ ) − .Let us now consider the PGL ( d ′ ) -invariant morphism µ : PGL ( d ) × rep ( Λ E , d ′ ) → C induced by f E . By Theorem 5.1(2) and Proposition 5.2, we know that C = filt E ( d ) = GL ( d ) · Im f E , i.e., µ is a dominant morphism.Next, let ( g, M ) ∈ PGL ( d ) × rep ( Λ E , d ′ ) be a generic point with M a Schur Λ E -module.(Here, g denotes the image of g ∈ GL ( d ) in PGL ( d ) .) Then we claim that µ − ( µ ( g, M )) = GL ( d ′ ) · ( g, M ) . Indeed, let ( h, M ′ ) ∈ GL ( d ) × rep ( Λ E , d ′ ) be so that hf E ( M ′ ) = gf E ( M ) .In particular, M ≃ M ′ and so there exists an g ′ ∈ GL ( d ′ ) such that M ′ = g ′ M . Since f E ( M ) is a Schur module, we get that h = g ′ · g , and so g ′ ( g, M ) = ( h, M ′ ) . The claimnow follows.From the uniqueness of rational quotients, we know that µ is actually the rational quo-tient map for the PGL ( d ′ ) -action on PGL ( d ) × rep ( Λ E , d ′ ) , i.e., µ descends to a birationalisomorphism µ : PGL ( d ) ∗ PGL ( d ′ ) rep ( Λ E , d ′ ) C such that µ = µ ◦ π , where π : PGL ( d ) × rep ( Λ E , d ′ ) PGL ( d ) ∗ PGL ( d ′ ) rep ( Λ E , d ′ ) isthe rational quotient map. Note that as π is PGL ( d ) -equivariant, so is µ . The proof nowfollows from Lemma 5.4. (cid:3) As an immediate consequence of Theorem 1.2, we have:
Corollary 5.5. If Q is a Euclidean quiver then k ( rep ( Q, d )) GL ( d ) ≃ k or k ( t ) for each Schur root d of Q .Proof. If d is a real Schur root then k ( rep ( Q, d )) GL ( d ) ≃ k . Next, we denote by δ Q , theunique isotropic Schur root of Q . Choose a vertex i ∈ Q such that Q \ { i } is a Dynkinquiver. Without loss of generality, let us assume that i is a source. In this case, wecan choose two exceptional representations E and E of Q with dim E = δ Q − e i and dim E = e i . Then, E := ( E , E ) is an orthogonal exceptional sequence with dim k Ext ( E , E ) = Hence, Q ( E ) is the Kronecker quiver K : · · k s But, for K and dimension vector (
1, 1 ) , the corresponding field of rational invariants isclearly k ( t ) . The proof now follows from Theorem 1.2. (cid:3) Remark 5.6.
Note that the sequence E above corresponds to a facet of the cone Eff ( Q, δ ) ofeffective weights associated to ( Q, δ ) . In fact, the facets of C ( Q, δ ) give rise to all orthog-onal exceptional sequences E of Q of length two for which filt E ( δ ) = ∅ (see [19, Theorem5.1]). We will come back to this approach in Section 6.2.6. C ANONICAL ALGEBRAS
Let m = ( m , . . . , m n ) , n ≥ , be a sequence of positive integers greater than one, andlet λ = ( λ , . . . , λ n ) be a sequence of pairwise distinct non-zero scalars in k with λ = .The canonical algebra Λ = Λ ( m, λ ) is, by definition, the bound quiver algebra of thebound quiver ( ∆ ( m ) , R ( m, λ )) where ∆ ( m ) is the quiver: ∞ • (
1, m − ) ••• (
1, 1 ) • • (
2, m − ) ••• (
2, 1 ) • ( n, m n − ) ••• ( n, 1 ) a \ \ ::::::::::::::::::::: a − o o a o o a i i RRRRRRRRRRRRRRR a − o o a o o a n,mn | | yyyyyyyyyyyyyyyyy a n,mn − o o a n,2 o o a (cid:2) (cid:2) (cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4)(cid:4) a u u lllllllllllllll a n,1 b b EEEEEEEEEEEEEEEEE and R ( m, λ ) consists of the following relations: a a . . . a + λ i a a . . . a − a i,1 a i,2 . . . a i,m i , 3 ≤ i ≤ n. The virtual genus of Λ , denoted by g Λ , is g Λ = +
12 m ( n − − − · · · − n ) , where m = lcm { m , . . . , m n } . Note that g Λ is always non-negative. Moreover, the virtualgenus g Λ controls the representation type of Λ in the following way (see [47] or [7, Section7]).(1) If g Λ = then n = and m is one of the following triples ( l −
2, 2, 2 ) with l ≥ , (
3, 3, 2 ) , (
4, 3, 2 ) or (
5, 3, 2 ) , i.e., ∆ \ { ∞ } is a Dynkin quiver of type D or E . In thiscase, Λ is a concealed algebra of extended Dynkin type e D l , e E , e E , or e E (see forexample the D. Happel-D. Vossieck’s list in [30]).(2) If g Λ = (or equivalently, Λ ≤ ) then m is one of the following four tuples (
2, 2, 2, 2 ) , (
3, 3, 3 ) , (
4, 4, 2 ) , and (
6, 3, 2 ) , i.e., ∆ \{ ∞ } is an extended Dynkin diagramof type e D or e E . In this case, we call Λ a tubular canonical algebra . The classificationof the indecomposable modules over a tubular canonical algebra turns out to beclosely related to Atiyah’s [3] classification of indecomposable bundles over anelliptic curve.(3) Λ is wild if and only if g Λ > 1 .In what follows, we briefly recall some of the key features of canonical algebras (see forexample [47] or [42]). First of all, Λ has global dimension two. In particular, the Tits form q Λ coincides with χ Λ . The rank and degree of a dimension vector d of Λ , denoted by rk Λ ( d ) and deg Λ ( d ) respectively, are rk Λ ( d ) = d ( ) − d ( ∞ ) and deg Λ ( d ) = n X i = mm i m i − X j = d ( i, j ) ! − ( n − ) m − n X i = mm i ! d ( ∞ ) . We denote by h the dimension vector of Λ that takes value at all vertices of ∆ . It turnsout that rk Λ ( d ) = h d , h i Λ = − h h , d i Λ for any dimension vector d of Λ . et P ( R , Q , respectively) be the full subcategory of mod ( Λ ) consisting of all Λ -modulesthat are direct sums of indecomposable Λ -modules X such that rk Λ ( dim X ) > 0 ( rk Λ ( dim X ) = rk Λ ( dim X ) < 0 , respectively). The following properties hold true.(i) mod ( Λ ) = P W R W Q .(ii) Hom Λ ( N, M ) =
Ext ( M, N ) = if either N ∈ R W Q , M ∈ P or N ∈ Q , M ∈P W R .(iii) pd Λ M ≤ for all M ∈ P W R and id Λ N ≤ for all N ∈ R W Q .We end this subsection with the Riemann-Roch formula for canonical algebras due toGeigle-Lenzing [28] (see also [42] and [41]). Denote by Φ Λ the Coxeter matrix of Λ . TheRiemann-Roch formula tells us that for any two dimension vectors d and e of Λ :(5) m − X i = h Φ iΛ d , e i Λ = m ( − g Λ ) rk Λ ( d ) rk Λ ( e ) + det (cid:18) rk Λ ( d ) rk Λ ( e ) deg Λ ( d ) deg Λ ( e ) (cid:19) . Irreducible components for tame canonical algebras.
In this section, we reviewsome results of Bobi ´nski and Skowro ´nski [8], and of Geiss and Schr ¨oer [29] on the in-decomposable irreducible components for a tame canonical algebra Λ . Theorem 6.1.
Let Λ be a tame canonical algebra and let d be a generic root of Λ . Then d is anindivisible Schur root of Λ , q Λ ( d ) ∈ {
0, 1 } , and rep ( Λ, d ) has a unique indecomposable irreduciblecomponent. More precisely, if q Λ ( d ) = then there exists a unique d -dimensional exceptional Λ -module M and GL ( d ) M is the unique indecomposable irreducible component of rep ( Λ, d ) . If q Λ ( d ) = then rep ( Λ, d ) is irreducible. Remark 6.2.
Note that when Λ is tame concealed the only Schur root d for which q Λ ( d ) = is precisely h (see for example [30]). Moreover, h generates the radical of χ Λ in this case.The dimension vector of an indecomposable Λ -module is called a root of Λ . A root d of Λ is said to be real if q Λ ( d ) = . We call the root d of Λ isotropic q Λ ( d ) = .Using the Riemann-Roch formula ( ) and the fact that the Schur roots of tame canonicalalgebras are indivisible, one can show: Lemma 6.3.
Let Λ be a tubular canonical algebra. If d is an isotropic Schur root of Λ then Φ Λ d = d . In particular, h d , e i Λ = det (cid:18) rk Λ ( d ) rk Λ ( e ) deg Λ ( d ) deg Λ ( e ) (cid:19) , for all dimension vectors e of Λ . Remark 6.4.
Note that the condition Φ Λ d = d is equivalent to h d , e i Λ + h e , d i Λ = for all e ∈ Z ∆ , i.e., d is in the radical of χ Λ . Proof.
Since Λ is tubular, we know that Φ mΛ d ′ = d ′ for any dimension vectors d ′ of Λ (seefor example [41]). Given a dimension vector d ′ of Λ , set r ( d ′ ) = min { i ≥ | Φ iΛ d ′ = d ′ } and l ( d ′ ) = g.c.d { (cid:16) P r ( d ′ )− = Φ iΛ d ′ (cid:17) ( v ) | v ∈ ∆ } . Then, iso ( d ′ ) := r ( d ′ ) m · l ( d ′ ) P m − = Φ jΛ d ′ is anindivisible isotropic Schur root of Λ such that Φ Λ iso ( d ′ ) = iso ( d ′ ) . sing the Riemann-Roch formula ( ) , we obtain that h iso ( d ) , d i Λ = . It now followsfrom the general theory of tubular algebras that the two isotropic roots iso ( d ) and d aremultiple of each other, and so they must be equal as they are both indivisible. The proofnow follows. (cid:3) Exceptional sequences from cones of effective weights.
In what follows we pro-vide a systematic approach to finding “convenient” orthogonal exceptional sequences ofmodules. This approach is based on the notion of θ -stable decomposition of dimensionvectors in irreducible components of module varieties. From this point on until Lemma6.6 below, Λ = kQ/ h R i is the bound quiver algebra of an arbitrary bound quiver ( Q, R ) , d is a dimension vector of Λ , and θ ∈ R Q . Recall that a module M ∈ rep ( Λ, d ) is saidto be θ -semi-stable if θ ( dim M ) = and θ ( dim M ′ ) ≤ for all submodules M ′ ⊆ M .We say that M is θ -stable if θ ( dim M ) = and θ ( dim M ′ ) < 0 for all proper submod-ules { } ⊂ M ′ ⊂ M . Denote by rep ( Λ ) s ( s ) θ the full subcategory of rep ( Λ ) consisting of all θ -(semi-)stable Λ -modules. Then, rep ( Λ ) ssθ is an abelian subcategory of rep ( Λ ) which isclosed under extensions, and whose simple objects are precisely the θ -stable Λ -modules.Moreover, rep ( Λ ) ssθ is Artinian and Noetherian, and hence, every θ -semi-stable finite-dimensional Λ -module has a Jordan-H ¨older filtration in rep ( Q ) ssθ .Now, let C be an irreducible component of rep ( Λ, d ) . For a real weight θ ∈ R Q , wedefine C s ( s ) θ = { M ∈ C | M is θ -(semi-)stable } . Next, the cone of effective weights of C is,by definition, the set Eff ( C ) = { θ ∈ R Q | C ssθ = ∅ } . We know that there are only finitely many GIT-classes in the cone Eff ( Q, d ) of effectiveweights associated to ( Q, d ) (see for example [2] or [13]). Among a set of representativesfor these GIT-classes, we denote by θ , . . . , θ l the integral weights for which the corre-sponding semi-stable loci in C are non-empty. Note that for any θ ∈ Eff ( C ) , C ssθ is open in C . Moreover, for any representation M ∈ T li = C ssθ i , we have Eff ( C ) = { θ ∈ R Q | θ ( d ) = and θ ( d M ′ ) ≤ ∀ M ′ ⊆ M } , and so Eff ( C ) is a rational convex polyhedral cone.Let θ be a lattice point in Eff ( C ) . For each sequence D = ( d , . . . , d t ) of dimensionvectors of θ -stable Λ -modules, consider the subset C D of C consisting of all Λ -modules M ∈ C ssθ that have a Jordan-H ¨older filtration { } = M ⊂ M ⊂ . . . ⊂ M t = M in rep ( Λ ) ssθ such that the sequence ( dim M , dim M /M , . . . , dim M/M t − ) is the same as D up to permutation. It is not difficult to see that C D is a constructible subset of C (seefor example [15]). Since C ssθ is irreducible, we deduce that there exists a unique, up topermutation, such sequence D = ( d , . . . , d t ) for which C D contains an open and densesubset of C . We write d = d ∔ . . . ∔ d t , and call it the θ -stable decomposition of d in C .In what follows, for a given d ∈ R Q , we denote by H ( d ) the hyperplane in R Q orthog-onal to d , i.e., H ( d ) = { θ ∈ R Q | θ ( d ) = } . Lemma 6.5.
Let F be a face of Eff ( C ) of positive dimension and let θ ∈ relint Eff ( C ) ∩ Z Q . If d = m · d ∔ . . . m t · d t is the θ -stable decomposition of d in C with d i = d j , ∀ ≤ i < j ≤ t , hen F = Eff ( C ) ∩ t \ i = H ( d i ) . Proof.
It follows from the discussion above that we can always choose a module M ∈ C ssθ such that • Eff ( C ) = { θ ∈ R Q | θ ( d ) = and θ ( d M ′ ) ≤ ∀ M ′ ⊆ M } , and • M has a Jordan-H ¨older filtration { } = M ⊂ M ⊂ . . . ⊂ M N = M in rep ( Λ ) ssθ such that the sequence ( dim M , dim M /M , . . . , dim M/M N − ) is the same as ( d m , . . . , d m t t ) up to permutation. (Here, N = m + . . . + m t .)In particular, we get that F = Eff ( C ) ∩ \ { θ ∈ R Q | θ ( dim M ′ ) = } , where the union is over all submodules M ′ of M for which θ ( dim M ′ ) = .Now, we clearly have that θ ( dim M i ) = ∀ ≤ i ≤ t , and so F ⊆ Eff ( C ) ∩ T ti = H ( d i ) .To show the other inclusion, first note that if M ′ ⊆ M is a submodule such that θ ( dim M ′ ) = then M ′ is θ -semi-stable, and using the uniqueness of the factors of a Jordan-H ¨olderfiltration in rep ( Λ ) ssθ , we deduce that dim M ′ is a linear combination of some of d , . . . , d t .The other inclusion now follows. (cid:3) We also have the following useful lemma:
Lemma 6.6.
Let Λ be a tame canonical algebra and let d be an isotropic Schur root of Λ . Then, rep ( Λ, d ) sθ d = ∅ where θ d denotes the weight h d , ·i Λ .Proof. From the general theory of tame concealed algebras and of tubular algebras (seefor example [9, Section 2]), we know that d is the dimension vector of an indecompos-able Λ -module lying at the mouth of a homogeneous tube which is part of a family, callit T , of pairwise orthogonal tubes. Specifically, T is the full subcategory of mod ( Λ ) con-sisting of all Λ -modules that are direct sums of indecomposable Λ -modules X such that θ d ( dim X ) = . Moreover, let e P ( e Q , respectively) be the full subcategory of mod ( Λ ) con-sisting of all Λ -modules that are direct sums of indecomposable Λ -modules X such that θ d ( dim X ) < 0 ( θ d ( dim X ) > 0 , respectively). Then, we have:(1) mod ( Λ ) = e P W T W e Q ;(2) Hom Λ ( X, Y ) = if either X ∈ e Q , Y ∈ T or X ∈ T , Y ∈ e P .It is now clear that rep ( Λ, d ) ssθ d = ∅ since any d -dimensional Λ -module from T is θ d -semi-stable. Let M ∈ rep ( Λ, d ) be an indecomposable module that lies in a homogeneoustube of T . We are going to show that M is θ d -stable. Assume to the contrary that M isnot θ d -stable and consider a Jordan-H ¨older filtration of M in rep ( Λ ) ssθ d . This way, we cansee that M must have a proper θ d -stable submodule M ′ . Then, M ′ must belong to thehomogeneous tube of M , and from this we deduce that dim M ′ is an integer multiple of d . But this is a contradiction. (cid:3) Now, we are ready to prove: roposition 6.7. Let Λ be a tame canonical algebra and let d be an isotropic Schur root of Λ .Then, there exists an orthogonal exceptional sequence E = ( E , E ) of Λ -modules such that filt E ( d ) contains a Schur Λ -module and Λ E is the path algebra of the Kronecker quiver K .Proof. We know that rep ( Λ, d ) is irreducible by Theorem 6.1 and let us denote by Eff ( Λ, d ) the cone of effective weights of rep ( Λ, d ) . It follows from Lemma 6.6 that dim Eff ( Λ, d ) = | ∆ | − . Next, choose a facet F of the cone Eff ( Λ, d ) and a weight θ ∈ relint F ∩ Z ∆ .Now, consider the θ -stable decomposition of d in rep ( Λ, d ) : d = m · d ∔ . . . ∔ m t · d t , with m , . . . , m t positive integers and d i = d j , ∀ ≤ i = j ≤ t . Note that d , . . . , d t areindivisible Schur roots by Theorem 6.1.For each ≤ i ≤ t , let E i be a d i -dimensional θ -stable module that arises as a factor ofa Jordan-H ¨older filtration of a generic module M in rep ( Λ, d ) . Note that we can choose M to be θ d -stable by Lemma 6.6. Furthermore, we have that Hom Λ ( E i , E j ) = ∀ ≤ i = j ≤ t since E , . . . , E t are pairwise non-isomorphic ( θ -)stable modules.Claim: F = Eff ( Λ, d ) ∩ H ( d ) ∩ H ( d ) and d = n d + n d for unique numbers n and n . Proof of Claim:
Note that F has dimension | ∆ | − , and so t ≥ . Moreover, the dimensionof the subspace of R ∆ orthogonal to the subspace spanned by { d , d , d } is at least ∆ − since it contains F . In particular, the set { d , d , d } is linearly dependent. Since d and d are distinct indivisible vectors, we deduce that d = n d + n d for unique numbers n and n .When t = , the proof follows from Lemma 6.5. Now, let us assume that t ≥ . Arguingas before, we deduce that d is a linear combination of d i and d , and d is also a linearcombination of d i and d for all ≤ i ≤ t . So, d i is a linear combination of d and d for all i , and this implies that H ( d ) ∩ H ( d ) = T ti = H ( d i ) . The proof of the claim nowfollows again from Lemma 6.5. (cid:3) There are three possible cases that we need to distinguish:Case 1: q Λ ( d ) = q Λ ( d ) = . First note that h d , d i Λ + h d , d i Λ = since d is in theradical of χ Λ by Lemma 6.3.If rk Λ ( d ) · rk Λ ( d ) ≥ then h d i , d j i Λ = − dim k Ext ( E i , E j ) , ∀ ≤ i = j ≤ . Conse-quently, h d , d i Λ = h d , d i Λ = . But then the two isotropic roots d and d would haveto be multiple of each other. So, d = d which is a contradiction.If rk Λ ( d ) > 0 > rk Λ ( d ) or rk Λ ( d ) > 0 > rk Λ ( d ) then either h d , d i Λ = or h d , d i Λ = . Since we are in the isotropic case, this would again imply that d = d .But this is a contradiction.Case 2: q Λ ( d ) =
1, q Λ ( d ) = (or the other way around). Using the claim above andthe fact that q Λ ( d ) = , we deduce that n = n n (− h d , d i Λ − h d , d i Λ ) . This relationcombined with the fact that d is in the radical of χ Λ implies that n = , which is acontradiction.Case 3: q Λ ( d ) =
1, q Λ ( d ) = . In this case, both E and E are exceptional Λ -modules.To simplify notation, set l = − h d , d i Λ − h d , d i Λ . Then, using the fact that d is anisotropic root in the radical of χ Λ , we deduce that = n l, 2n = n l , and n + n = n n . It is now easy to see that n = n = and l = . Without loss of general-ity, we can assume that E is a submodule of M and E = M/E . Then, we have that dim k Ext ( E , E ) > 0 .We also note that h d , d i Λ = − and h d , d i Λ = − since otherwise both of these twoinner products would have to be − , and this would imply that θ d ( dim E ) = . But thiswould contradict the fact that M is θ d -stable.Case 3.1: rk Λ ( d ) · rk Λ ( d ) > 0 . In this case, we have that h d , d i Λ = − dim k Ext ( E , E ) and h d , d i Λ = − dim k Ext ( E , E ) . It now follows that ( E , E ) is an orthogonal excep-tional sequence with the desired properties.Case 3.2: rk Λ ( d ) · rk Λ ( d ) ≤ . First, note that rk Λ ( d ) and rk Λ ( d ) can not be bothzero since otherwise rk Λ ( d ) would be zero, and this would imply that d = h . But, then θ d ( dim E ) = which is not possible.It is now easy to see that properties (ii)-(iii) mentioned at the beginning of this sectionimply that ( E , E ) is an orthogonal exceptional sequence with the desired properties. (cid:3) Now, we are ready to prove Theorem 1.3.
Proof of Theorem 1.3.
The implication ( ) = ⇒ ( ) has been proved in Proposition 4.6.Now, let us assume that Λ is a tame canonical algebra and let d be a generic root of Λ .We know from Theorem 6.1 that d is a Schur root and rep ( Λ, d ) has a unique indecom-posable irreducible component, call it C .If q Λ ( d ) = then k ( C ) GL ( d ) ≃ k since C is an orbit closure in this case.It remains to look into the case when d is an isotropic Schur root of Λ . It follows fromProposition 6.7 that there exists an orthogonal exceptional sequence E = ( E , E ) such that C ∩ filt d ( E ) = ∅ and Λ E is the path algebra of K . Now, applying the reduction Theorem1.2, we conclude that k ( C ) GL ( d ) ≃ k ( rep ( K , (
1, 1 ))) GL (( )) ≃ k ( t ) . (cid:3) Finally, let us prove Proposition 1.4:
Proof of Proposition 1.4.
We know from Theorem 1.1 and Theorem 1.3 that if C is an inde-composable irreducible component of rep ( Λ, d ) then S m ( k ( C ) GL ( d ) ) is isomorphic to either k in case d is a real Schur root or to k ( t , . . . , t m ) in case d is isotropic. The proof now fol-lows from Proposition 4.7. (cid:3) Remark 6.8.
In a sequel to the current work, we plan to use a similar strategy to provethe analogous of Theorem 1.3 for other classes of algebras, including the class of quasi-tilted algebras and of string algebras. Of course, the ultimate goal is to prove Theorem1.3 for arbitrary tame algebras. Since it is believed that the representation theory of tamealgebras can be reduced to that of tame strongly simply-connected algebras via coveringfunctors, the next natural steps are: (1) to solve the rationality problem for tame stronglysimply-connected algebras; (2) to show that the rationality of fields of rational invariantsis preserved under covering functors (in the relevant cases). We plan to address all theseproblems in future work in which the reduction Theorem 1.2 combined with the system-atic approach to finding short orthogonal exceptional sequences from Section 6.2 will playa fundamental role. CKNOWLEDGMENT
I would like to thank Frauke Bleher, Grzegorz Bobi ´nski, Harm Derksen, Christof Geiss,Dirk Kusin, Hagen Meltzer, Steven Sam, and Jerzy Weyman for helpful conversations onthe subject of the paper. R
EFERENCES [1] J. Adriaenssens and L. Le Bruyn. Local quivers and stable representations.
Comm. Algebra , 31(4):1777–1797, 2003.[2] I.V. Arzhantsev and J. Hausen. Geometric invariant theory via cox rings. Preprint, ar X iv:0706.4353v1,2007.[3] M. F. Atiyah. Vector bundles over an elliptic curve. Proc. London Math. Soc. (3) , 7:414–452, 1957.[4] D. J. Benson.
Representations and cohomology. I: Basic representation theory of finite groups and associativealgebras , volume 30 of
Cambridge Studies in Advanced Mathematics . Cambridge University Press, Cam-bridge, second edition, 1998.[5] G. Bobi ´nski. Geometry and the zero sets of semi-invariants for homogeneous modules over canonicalalgebras.
J. Algebra , 319(3):1320–1335, 2008.[6] G. Bobi ´nski. On the zero set of semi-invariants for regular modules over tame canonical algebras.
J.Pure Appl. Algebra , 212(6):1457–1471, 2008.[7] G. Bobi ´nski, Ch. Riedtmann, and A. Skowro ´nski. Semi-invariants of quivers and their zero sets. In
Trends in representation theory of algebras and related topics , EMS Ser. Congr. Rep., pages 49–99. Eur. Math.Soc., Z ¨urich, 2008.[8] G. Bobi ´nski and A. Skowro ´nski. Geometry of modules over tame quasi-tilted algebras.
Colloq. Math. ,79(1):85–118, 1999.[9] G. Bobi ´nski and A. Skowro ´nski. Geometry of periodic modules over tame concealed and tubular alge-bras.
Algebr. Represent. Theory , 5(2):187–200, 2002.[10] R. Bocklandt. Smooth quiver representation spaces.
J. Algebra , 253(2):296–313, 2002.[11] K. Bongartz. Algebras and quadratic forms.
J. London Math. Soc. (2) , 28(3):461–469, 1983.[12] Th. Br ¨ustle, J. A. de la Pe ˜na, and A. Skowronski. Tame algebras and Tits quadratic forms. To appear.[13] C. Chindris. On
GIT -fans for quivers. Preprint avilable at ar X iv:0805.1440v1 [math.RT].[14] C. Chindris, H. Derksen, and J. Weyman. Counterexamples to Okounkov’s log-concavity conjecture. Compositio Mathematica , 143(6):1545–1557, 2007.[15] W. Crawley-Boevey and J. Schr ¨oer. Irreducible components of varieties of modules.
J. Reine Angew.Math. , 553:201–220, 2002.[16] J. A. de la Pe ˜na. On the dimension of the module-varieties of tame and wild algebras.
Comm. Algebra ,19(6):1795–1807, 1991.[17] H. Derksen and J. Weyman. Semi-invariants of quivers and saturation for L ittlewood- R ichardson co-efficients. J. Amer. Math. Soc. , 13(3):467–479, 2000.[18] H. Derksen and J. Weyman. On the Littlewood-Richardson polynomials.
J. Algebra , 255(2):247–257,2002.[19] H. Derksen and J. Weyman. The combinatorics of quiver representations. Preprint available atarXiv.math.RT/0608288, 2006.[20] M. Domokos. On singularities of quiver moduli. Preprint, arXiv:0903.4139v2 [math.RT], 2009.[21] M. Domokos and H. Lenzing. Invariant theory of canonical algebras.
J. Algebra , 228(2):738–762, 2000.[22] M. Domokos and H. Lenzing. Moduli spaces for representations of concealed-canonical algebras.
J.Algebra , 251(1):371–394, 2002.[23] P. Donovan and M. R. Freislich.
The representation theory of finite graphs and associated algebras . Number 5in Carleton Mathematical Lecture Notes. Carleton University, Ottawa, Ont., 1973.[24] Yu. A. Drozd. Tame and wild matrix problems. In
Representations and quadratic forms (Russian) , pages39–74, 154. Akad. Nauk Ukrain. SSR Inst. Mat., Kiev, 1979.[25] D. Eisenbud.
Commutative algebra , volume 150 of
Graduate Texts in Mathematics . Springer-Verlag, NewYork, 1995. With a view toward algebraic geometry.
26] E. Formanek. The ring of generic matrices.
J. Algebra , 258(1):310–320, 2002. Special issue in celebrationof Claudio Procesi’s 60th birthday.[27] P. Gabriel. Unzerlegbare D arstellungen. I . ( G erman. E nglish summary). Manuscripta Math. , 6(2):71–103; correction, ibid. 6 (1972), 309., 1972.[28] W. Geigle and H. Lenzing. A class of weighted projective curves arising in representation theoryof finite-dimensional algebras. In
Singularities, representation of algebras, and vector bundles (Lambrecht,1985) , volume 1273 of
Lecture Notes in Math. , pages 265–297. Springer, Berlin, 1987.[29] Ch. Geiss and J. Schr ¨oer. Varieties of modules over tubular algebras.
Colloq. Math. , 95(2):163–183, 2003.[30] D. Happel and D. Vossieck. Minimal algebras of infinite representation type with preprojective com-ponent.
Manuscripta Math. , 42(2-3):221–243, 1983.[31] R. Hartshorne.
Algebraic geometry . Springer-Verlag, New York, 1977. Graduate Texts in Mathematics,No. 52.[32] K. Igusa, K. Orr, G. Todorov, and J. Weyman. Cluster complexes via semi-invariants.
Compos. Math. ,145(4):1001–1034, 2009.[33] V.G. Kac. Infinite root systems, representations of graphs and invariant theory.
Invent. Math , 56(1):57–92, 1980.[34] V.G. Kac. Infinite root systems, representations of graphs and invariant theory II . J. Algebra , 78(1):141–162, 1982.[35] B. Keller. A -infinity algebras in representation theory. In Representations of algebra. Vol. I, II , pages 74–86.Beijing Norm. Univ. Press, Beijing, 2002.[36] B. Keller. A -infinity algebras, modules and functor categories. In Trends in representation theory of alge-bras and related topics , volume 406 of
Contemp. Math. , pages 67–93. Amer. Math. Soc., Providence, RI,2006.[37] A.D. King. Moduli of representations of finite-dimensional algebras.
Quart. J. Math. Oxford Ser.(2) ,45(180):515–530, 1994.[38] H. Kraft and Ch. Riedtmann. Geometry of representations of quivers. In
Representations of algebras(Durham, 1985) , volume 116 of
London Math. Soc. Lecture Note Ser. , pages 109–145. Cambridge Univ.Press, Cambridge, 1986.[39] W. Kra´skiewicz. On semi-invariants of tilted algebras of type A n . Colloq. Math. , 90(2):253–267, 2001.[40] L. Le Bruyn. Centers of generic division algebras, the rationality problem 1965–1990.
Israel J. Math. ,76(1-2):97–111, 1991.[41] H. Lenzing. A K -theoretic study of canonical algebras. In Representation theory of algebras (Cocoyoc, 1994) ,volume 18 of
CMS Conf. Proc. , pages 433–454. Amer. Math. Soc., Providence, RI, 1996.[42] H. Lenzing and H. Meltzer. Sheaves on a weighted projective line of genus one, and representations ofa tubular algebra [ MR1206953 (94d:16019)]. In
Representations of algebras (Ottawa, ON, 1992) , volume 14of
CMS Conf. Proc. , pages 313–337. Amer. Math. Soc., Providence, RI, 1993.[43] L.A. Nazarova. Representations of quivers of infinite type. ( R ussian). Izv. Akad. Nauk SSSR Ser. Mat. ,37:752–791, 1973.[44] Z. Reichstein. On the notion of essential dimension for algebraic groups.
Transform. Groups , 5(3):265–304, 2000.[45] Z. Reichstein and N. Vonessen. Tame group actions on central simple algebras.
J. Algebra , 318(2):1039–1056, 2007.[46] Ch. Riedtmann and A. Schofield. On open orbits and their complements.
J. Algebra , 130(2):388–411,1990.[47] C. Ringel.
Tame algebras and integral quadratic forms , volume 1099 of
Lecture Notes in Mathematics .Springer-Verlag, Berlin, 1984.[48] C. M. Ringel. The rational invariants of the tame quivers.
Invent. Math. , 58(3):217–239, 1980.[49] C.M. Ringel. Representations of K -species and bimodules. J. Algebra , 41(2):269–302, 1976.[50] C.M. Ringel. The braid group action on the set of exceptional sequences of a hereditary Artin algebra.In
Abelian group theory and related topics (Oberwolfach, 1993) , volume 171 of
Contemp. Math. , pages 339–352. Amer. Math. Soc., Providence, RI, 1994.[51] M. Rosenlicht. Some basic theorems on algebraic groups.
Amer. J. Math. , 78:401–443, 1956.[52] A. Schofield. General representations of quivers.
Proc. London Math. Soc. (3) , 65(1):46–64, 1992.[53] A. Schofield. Birational classification of moduli spaces of representations of quivers.
Indag. Math. (N.S.) ,12(3):407–432, 2001.
54] I. R. Shafarevich, editor.
Algebraic geometry. IV , volume 55 of
Encyclopaedia of Mathematical Sciences .Springer-Verlag, Berlin, 1994. Linear algebraic groups. Invariant theory, A translation of
Algebraicgeometry. 4 (Russian), Akad. Nauk SSSR Vsesoyuz. Inst. Nauchn. i Tekhn. Inform., Moscow, 1989 [MR1100483 (91k:14001)], Translation edited by A. N. Parshin and I. R. Shafarevich.[55] I. R. Shafarevich.
Basic algebraic geometry. 1 . Springer-Verlag, Berlin, second edition, 1994. Varieties inprojective space, Translated from the 1988 Russian edition and with notes by Miles Reid.[56] A. Skowro ´nski and J. Weyman. Semi-invariants of canonical algebras.
Manuscripta Math. , 100(3):391–403, 1999.[57] A. Skowro ´nski and J. Weyman. The algebras of semi-invariants of quivers.
Transform. Groups , 5(4):361–402, 2000.[58] C E. Watts. Intrinsic characterizations of some additive functors.
Proc. Amer. Math. Soc. , 11:5–8, 1960.[59] J. Weyman. Personal communication.U
NIVERSITY OF M ISSOURI , D
EPARTMENT OF M ATHEMATICS , C
OLUMBIA , MO 65211, USA
E-mail address , Calin Chindris: [email protected]@missouri.edu