Introduction
The composition factors of a finite-length module, counted with multiplicities, are among its simplest invariants. In general, nonisomorphic indecomposable modules may have the same composition factors. Let Λ \Lambda Λ be an Artin algebra. For a finitely generated Λ \Lambda Λ -module M M M and a simple Λ \Lambda Λ -module S S S , we write [ M : S ] [M:S] [ M : S ] for the multiplicity of S S S in a composition series of M M M . We call Λ \Lambda Λ a composition factor algebra (see Definition 5.1) if
[ M : S ] = [ N : S ] for every simple S ⟹ M ≅ N [M:S] = [N:S]\text{ for every simple }S \quad\Longrightarrow\quad M \cong N [ M : S ] = [ N : S ] for every simple S ⟹ M ≅ N
for any two indecomposable finitely generated Λ \Lambda Λ -modules M , N M,N M , N .
The works of Happel [8 ] and Auslander and Reiten [2 ] give conditions under which composition factors determine indecomposable modules; see also the historical notes in [3, Chapter IX, p. 333]. Auslander, Reiten and Smalø conjectured that every composition factor algebra is representation-finite [3 Conjecture (4), p. 409] . Here representation-finite means that there are only finitely many isomorphism classes of indecomposable finitely generated modules. Blasco proved the conjecture for commutative, hereditary or radical-square-zero Artin algebras [4 ] . Our main result establishes it for arbitrary Artin algebras.
Theorem 1.1. Every composition factor algebra is representation-finite.
The proof of Theorem 1.1 uses the following finiteness criteria for comma categories (cf. [12 II, §6] ) and factorization categories (cf. [13 §1] ). Here and throughout, R R R is a commutative Artin ring.
Definition (see Definition 3.1 ). Let D D D be a division Artin R R R -algebra, and let X , Y \mathcal{X}, \mathcal{Y} X , Y be Hom-finite Krull–Schmidt R R R -linear categories. Fix additive R R R -linear functors G : X → mod D G:\mathcal{X}\to\operatorname{mod} D G : X → mod D and F : Y → mod D F:\mathcal{Y}\to\operatorname{mod} D F : Y → mod D . The comma category M ( G , F ) = ( G ↓ F ) \mathcal{M}(G,F)=(G\downarrow F) M ( G , F ) = ( G ↓ F ) is defined as follows.
Objects are triples ( X , Y , u ) (X,Y,u) ( X , Y , u ) with X ∈ X X\in\mathcal{X} X ∈ X , Y ∈ Y Y\in\mathcal{Y} Y ∈ Y and a homomorphism of D D D -modules u : G ( X ) → F ( Y ) u:G(X)\to F(Y) u : G ( X ) → F ( Y ) .
A morphism ( X , Y , u ) → ( X ′ , Y ′ , u ′ ) (X,Y,u)\to(X',Y',u') ( X , Y , u ) → ( X ′ , Y ′ , u ′ ) is a pair ( f , g ) (f,g) ( f , g ) with f : X → X ′ f:X\to X' f : X → X ′ in X \mathcal{X} X and g : Y → Y ′ g:Y\to Y' g : Y → Y ′ in Y \mathcal{Y} Y satisfying F ( g ) ∘ u = u ′ ∘ G ( f ) F(g)\circ u=u'\circ G(f) F ( g ) ∘ u = u ′ ∘ G ( f ) .
Identities and composition are defined componentwise.
The following theorem gives a finiteness criterion for comma categories.
Theorem A (Theorem 3.3 ). In the setting above, assume that X \mathcal{X} X and Y \mathcal{Y} Y are representation-finite. If any two indecomposable objects of M ( G , F ) \mathcal{M}(G,F) M ( G , F ) with isomorphic components are isomorphic, then M ( G , F ) \mathcal{M}(G,F) M ( G , F ) is representation-finite.
Let Λ \Lambda Λ be a basic composition factor algebra. Choose a primitive idempotent e e e and write e ′ = 1 − e e'=1-e e ′ = 1 − e , A = e Λ e A=e\Lambda e A = e Λ e , B = e ′ Λ e ′ B=e'\Lambda e' B = e ′ Λ e ′ , U = e Λ e ′ U=e\Lambda e' U = e Λ e ′ and V = e ′ Λ e V=e'\Lambda e V = e ′ Λ e . For M ∈ mod Λ M\in\operatorname{mod}\Lambda M ∈ mod Λ , we write X = e M X=eM X = e M and N = e ′ M N=e'M N = e ′ M . The Λ \Lambda Λ -action on M M M induces homomorphisms of A A A -modules
U ⊗ B N ⟶ X ⟶ Hom B ( V , N ) , N ∈ mod B , X ∈ mod A . U\otimes_B N\longrightarrow X\longrightarrow\operatorname{Hom}_B(V,N),\qquad N\in\operatorname{mod} B,\quad X\in\operatorname{mod} A. U ⊗ B N ⟶ X ⟶ Hom B ( V , N ) , N ∈ mod B , X ∈ mod A .
By Lemma 5.6 , J ( A ) U = 0 = V J ( A ) J(A)U=0=VJ(A) J ( A ) U = 0 = V J ( A ) . Hence both U ⊗ B N U\otimes_B N U ⊗ B N and Hom B ( V , N ) \operatorname{Hom}_B(V,N) Hom B ( V , N ) are annihilated by J ( A ) J(A) J ( A ) and are naturally modules over the division ring D = A / J ( A ) D=A/J(A) D = A / J ( A ) . The functors U ⊗ B − U\otimes_B- U ⊗ B − and Hom B ( V , − ) \operatorname{Hom}_B(V,-) Hom B ( V , − ) , together with the natural transformation induced by multiplication in Λ \Lambda Λ , define a factorization category equivalent to mod Λ \operatorname{mod}\Lambda mod Λ (see Proposition 5.7 ). We give the general definition below.
Definition (see Definition 4.1 ). Let A A A be a local Artin R R R -algebra, write D = A / J ( A ) D=A/J(A) D = A / J ( A ) , and let C \mathcal{C} C be a Hom-finite Krull–Schmidt R R R -linear category. Fix additive R R R -linear functors G , F : C → mod D G,F:\mathcal{C}\to\operatorname{mod} D G , F : C → mod D and a natural transformation μ : G → F \mu:G\to F μ : G → F . We view D D D -modules as A A A -modules through the quotient A → D A\to D A → D . The factorization category F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) is defined as follows.
Objects are tuples ( N , X , β , α ) (N,X,\beta,\alpha) ( N , X , β , α ) with N ∈ C N\in\mathcal{C} N ∈ C , X ∈ mod A X\in\operatorname{mod} A X ∈ mod A and homomorphisms of A A A -modules
G ( N ) → β X → α F ( N ) , α ∘ β = μ N . G(N)\xrightarrow{\beta}X\xrightarrow{\alpha}F(N),\qquad\alpha\circ\beta=\mu_N. G ( N ) β X α F ( N ) , α ∘ β = μ N . A morphism ( N , X , β , α ) → ( N ′ , X ′ , β ′ , α ′ ) (N,X,\beta,\alpha)\to(N',X',\beta',\alpha') ( N , X , β , α ) → ( N ′ , X ′ , β ′ , α ′ ) is a pair ( f , g ) (f,g) ( f , g ) with f : N → N ′ f:N\to N' f : N → N ′ in C \mathcal{C} C and g : X → X ′ g:X\to X' g : X → X ′ in mod A \operatorname{mod} A mod A satisfying g ∘ β = β ′ ∘ G ( f ) g\circ\beta=\beta'\circ G(f) g ∘ β = β ′ ∘ G ( f ) and F ( f ) ∘ α = α ′ ∘ g F(f)\circ\alpha=\alpha'\circ g F ( f ) ∘ α = α ′ ∘ g .
Identities and composition are defined componentwise.
For factorization categories, the following analogue of Theorem A holds.
Theorem B. (Theorem 4.8). In the setting above, assume that C C C and mod A \operatorname{mod} A mod A are representation-finite. If any two indecomposable objects of F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) with isomorphic C C C -components and isomorphic A A A -components are isomorphic, then F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) is representation-finite.
We first pass to the reduced category of Definition 4.4, whose objects are factorizations with zero composite. By Lemma 4.5, this construction induces a bijection between the isomorphism classes of indecomposable objects. Two applications of Theorem A, using the socle and a semisimple quotient of the middle term, then prove Theorem B.
We apply these criteria to prove Theorem 1.1 by induction on the R R R -length of the algebra. After passage to a basic algebra, Lemma 5.6 applied to the block decomposition in (9) gives
U V = 0 , J ( A ) U = 0 = V J ( A ) , U ⊗ B V = 0. UV = 0,\qquad J(A)U = 0 = VJ(A),\qquad U \otimes_B V = 0. U V = 0 , J ( A ) U = 0 = V J ( A ) , U ⊗ B V = 0.
These relations yield the factorization description above. The local corner A A A is a quotient of Λ \Lambda Λ , hence a composition factor algebra. It is Nakayama by Proposition 5.4. A suitable proper quotient of Λ \Lambda Λ gives representation-finiteness of B B B by induction. Under the factorization description, indecomposable objects with isomorphic components correspond to indecomposable Λ \Lambda Λ -modules with the same composition factors. Since Λ \Lambda Λ is a composition factor algebra, the corresponding modules, and hence the factorization objects, are isomorphic. Theorem B therefore implies that Λ \Lambda Λ is representation-finite.
All the results in this paper were obtained in early October 2026. During the final preparation of this manuscript, we became aware of the work [5 ] , which has just appeared on arXiv and proves the same conjecture.
The paper is organized as follows. Section 2 recalls the required facts about Artin algebras and Krull–Schmidt categories. Section 3 proves weak positivity of the quadratic form associated with a comma category and applies Simson’s criterion for finite prinjective type [16 Theorem 1.6] to establish Theorem A. Section 4 introduces factorization categories, establishes their basic properties, and proves Theorem B using Theorem A. In Section 5, we study composition factor algebras and describe their module categories by factorizations. We then prove Theorem 1.1 and give examples.
Conventions
Throughout the paper, R R R is a commutative Artin ring. The conventions below apply unless stated otherwise. Algebras are associative unital R R R -algebras that are finitely generated as R R R -modules. Modules are finitely generated left modules. We write mod Λ \operatorname{mod}\Lambda mod Λ for the category of such Λ \Lambda Λ -modules and J ( Λ ) J(\Lambda) J ( Λ ) for the Jacobson radical of Λ \Lambda Λ .
All additive categories are essentially small and R R R -linear. A category is Hom \operatorname{Hom} Hom -finite if its morphism spaces have finite length over the base ring R R R . An additive category is Krull–Schmidt if every object is a finite direct sum of objects with local endomorphism rings. Functors between R R R -linear categories are assumed to be R R R -linear.
Preliminaries
Artin algebras and modules
We recall some basic notions and facts about Artin algebras and their modules from [3 ] . For M ∈ mod Λ M \in\operatorname{mod}\Lambda M ∈ mod Λ , write rad M \operatorname{rad} M rad M and soc M \operatorname{soc} M soc M for its radical and socle, respectively, and set top M = M / rad M \operatorname{top} M = M/\operatorname{rad} M top M = M / rad M . Thus rad M = J ( Λ ) M \operatorname{rad} M = J(\Lambda)M rad M = J ( Λ ) M and top M = M / J ( Λ ) M \operatorname{top} M = M/J(\Lambda)M top M = M / J ( Λ ) M . We denote composition length over Λ \Lambda Λ by ℓ Λ ( M ) \ell_\Lambda(M) ℓ Λ ( M ) and length over R R R by ℓ R ( M ) \ell_R(M) ℓ R ( M ) .
Definition 2.1. (1) An algebra Λ \Lambda Λ is local if Λ / J ( Λ ) \Lambda/J(\Lambda) Λ/ J ( Λ ) is a division ring, and basic if Λ / J ( Λ ) \Lambda/J(\Lambda) Λ/ J ( Λ ) is a product of division rings.
(2) A module is uniserial if it has a unique composition series.
(3) An algebra is Nakayama if its indecomposable projective and indecomposable injective left modules are uniserial.
Fact 2.2. Let M M M be a nonzero Λ \Lambda Λ -module.
(1) The socle soc M \operatorname{soc} M soc M is essential in M M M . In particular, soc M ∩ N ≠ 0 \operatorname{soc} M \cap N \ne0 soc M ∩ N = 0 for every nonzero submodule N N N of $M.
(2) If top M \operatorname{top} M top M is simple, then rad M \operatorname{rad} M rad M is the unique maximal submodule of M M M .
Consequently, a module with simple top or simple socle is indecomposable.
Let E R E_R E R be the direct sum of the injective envelopes of a complete set of simple R R R -modules, and set
D = Hom R ( − , E R ) . \mathbb{D}=\operatorname{Hom}_R(-,E_R). D = Hom R ( − , E R ) .
Matlis duality D : mod Λ ⟶ mod Λ op \mathbb{D}:\operatorname{mod}\Lambda\longrightarrow\operatorname{mod}\Lambda^{\operatorname{op}} D : mod Λ ⟶ mod Λ op is an exact contravariant duality. It preserves R R R -length and interchanges projective and injective modules. In particular, it preserves indecomposability and induces a bijection between the isomorphism classes of simple modules on the two sides. Consequently, ℓ Λ op ( D ( M ) ) = ℓ Λ ( M ) \ell_{\Lambda^{\operatorname{op}}}(\mathbb{D}(M))=\ell_\Lambda(M) ℓ Λ op ( D ( M )) = ℓ Λ ( M ) for every M ∈ mod Λ M\in\operatorname{mod}\Lambda M ∈ mod Λ .
Let A A A be a local algebra and put D = A / J ( A ) D=A/J(A) D = A / J ( A ) . We regard D D D -modules as A A A -modules through the canonical quotient map. This identifies mod D \operatorname{mod}D mod D with the full subcategory of semisimple A A A -modules. Since D D D is a division ring, every V ∈ mod D V\in\operatorname{mod}D V ∈ mod D is a finite direct sum of copies of D D _D D D D . In particular, ℓ A ( V ) = dim D V \ell_A(V)=\dim_D V ℓ A ( V ) = dim D V .
Direct sums and endomorphisms
We collect some basic facts about direct sums and endomorphism rings for later use.
Definition 2.3. A Krull–Schmidt category is representation-finite if it has only finitely many isomorphism classes of indecomposable objects.
Lemma 2.4 (cf. [10 Proposition 2.9 and the paragraph following Corollary 4.4] ). Let C C C be a Hom-finite Krull–Schmidt category. Let X 1 , … , X r X_1,\ldots,X_r X 1 , … , X r be pairwise nonisomorphic indecomposable objects of C C C , and set
X = ⨁ i = 1 r X i n i , E i = End C ( X i ) , n i ≥ 1. X=\bigoplus_{i=1}^{r}X_i^{n_i},\qquad E_i=\operatorname{End}_C(X_i),\qquad n_i\ge1. X = i = 1 ⨁ r X i n i , E i = End C ( X i ) , n i ≥ 1. The block-diagonal projection induces an isomorphism
End C ( X ) / J ( End C ( X ) ) ≅ ∏ i = 1 r M n i ( E i / J ( E i ) ) . \operatorname{End}_C(X)/J(\operatorname{End}_C(X))\cong\prod_{i=1}^{r}M_{n_i}(E_i/J(E_i)). End C ( X ) / J ( End C ( X )) ≅ i = 1 ∏ r M n i ( E i / J ( E i )) . More precisely , an endomorphism f = ( f i j ) f=(f_{ij}) f = ( f ij ) belongs to J ( End C ( X ) ) J(\operatorname{End}_C(X)) J ( End C ( X )) if and only if f i i ∈ M n i ( J ( E i ) ) f_{ii}\in M_{n_i}(J(E_i)) f ii ∈ M n i ( J ( E i )) for every i i i .
Lemma 2.5. Let k \mathbb{k} k be a field and C C C a Hom-finite Krull–Schmidt k \mathbb{k} k -linear category. Let F : C → mod k F:C\to\operatorname{mod}\mathbb{k} F : C → mod k be a faithful k \mathbb{k} k -linear functor. Suppose that f ∈ End C ( M ) f\in\operatorname{End}_C(M) f ∈ End C ( M ) satisfies ϕ ( f ) = 0 \phi(f)=0 ϕ ( f ) = 0 for some monic irreducible polynomial ϕ ∈ k [ T ] \phi\in\mathbb{k}[T] ϕ ∈ k [ T ] of degree h h h . Then M M M has at most dim k F ( M ) / h \dim_{\mathbb{k}}F(M)/h dim k F ( M ) / h pairwise nonisomorphic indecomposable direct summands.
Proof. Write
M = ⨁ i = 1 t M i m i , M=\bigoplus_{i=1}^{t}M_i^{m_i}, M = i = 1 ⨁ t M i m i ,
where the M i M_i M i are pairwise nonisomorphic indecomposable objects. Set E = End C ( M ) E=\operatorname{End}_C(M) E = End C ( M ) and V = F ( M ) V=F(M) V = F ( M ) . The action a ⋅ v = F ( a ) ( v ) a\cdot v=F(a)(v) a ⋅ v = F ( a ) ( v ) makes V V V a left E E E -module, which is faithful since F F F is faithful. By Lemma 2.4 , E E E has exactly t t t isomorphism classes of simple left modules, represented by S 1 , … , S t S_1,\ldots,S_t S 1 , … , S t . Since V V V is a faithful module over the finite-dimensional k k k -algebra E E E , each S i S_i S i occurs as a composition factor of V V V . Indeed, choosing a k k k -basis v 1 , … , v d v_1,\ldots,v_d v 1 , … , v d of V V V gives an injective E E E -linear map E ⟶ V ⊕ d E\longrightarrow V^{\oplus d} E ⟶ V ⊕ d , a ↦ ( a v 1 , … , a v d ) a\mapsto(av_1,\ldots,av_d) a ↦ ( a v 1 , … , a v d ) .
Since ϕ ( f ) = 0 \phi(f)=0 ϕ ( f ) = 0 and ϕ \phi ϕ is irreducible, the action of f f f on S i S_i S i has minimal polynomial ϕ \phi ϕ . Thus dim k S i ≥ h \dim_k S_i\ge h dim k S i ≥ h , and hence t h ≤ ∑ i = 1 t dim k S i ≤ dim k V th\le\sum_{i=1}^{t}\dim_k S_i\le\dim_k V t h ≤ ∑ i = 1 t dim k S i ≤ dim k V . This proves the assertion. □
For a ring E E E , let E × E^\times E × denote its group of units. We shall use the following elementary lemma.
Lemma 2.6. Let E 1 , E 2 E_1,E_2 E 1 , E 2 be local Artin rings and W W W a nonzero E 1 E_1 E 1 –E 2 E_2 E 2 -bimodule. Suppose that E 1 × × E 2 × E_1^\times\times E_2^\times E 1 × × E 2 × acts transitively on W ∖ { 0 } W\setminus\{0\} W ∖ { 0 } by ( u , v ) ⋅ w = u w v − 1 (u,v)\cdot w=uwv^{-1} ( u , v ) ⋅ w = u w v − 1 . Then J ( E 1 ) W = 0 = W J ( E 2 ) J(E_1)W=0=WJ(E_2) J ( E 1 ) W = 0 = W J ( E 2 ) .
Proof. For convenience, denote J ( E 1 ) J(E_1) J ( E 1 ) by J J J and set Z = J W Z=JW Z = J W . We have
u Z v − 1 = ( u J ) ( W v − 1 ) = J W = Z uZv^{-1}=(uJ)(Wv^{-1})=JW=Z u Z v − 1 = ( u J ) ( W v − 1 ) = J W = Z
for all u ∈ E 1 × u\in E_1^\times u ∈ E 1 × and v ∈ E 2 × v\in E_2^\times v ∈ E 2 × . Suppose that Z ≠ 0 Z\ne0 Z = 0 and choose 0 ≠ z ∈ Z ⊆ W 0\ne z\in Z\subseteq W 0 = z ∈ Z ⊆ W . For every 0 ≠ w ∈ W 0\ne w\in W 0 = w ∈ W , transitivity gives units u , v u,v u , v such that w = u z v − 1 w=uzv^{-1} w = u z v − 1 . Hence w ∈ Z w\in Z w ∈ Z , so J W = W JW=W J W = W . Since J J J is nilpotent, J W = W JW=W J W = W implies W = 0 W=0 W = 0 , a contradiction. Thus J ( E 1 ) W = 0 J(E_1)W=0 J ( E 1 ) W = 0 . Similarly, we have W J ( E 2 ) = 0 WJ(E_2)=0 W J ( E 2 ) = 0 . □
The following lemma is well known. We include a proof for completeness.
Lemma 2.7. Let A \mathcal{A} A be a Hom-finite Krull–Schmidt category and B \mathcal{B} B an additive category. Let F : A → B F:\mathcal{A}\to\mathcal{B} F : A → B be a full, essentially surjective additive R R R -linear functor. Suppose that, for every X ∈ A X\in\mathcal{A} X ∈ A , the ideal
I X = ker ( End A ( X ) ⟶ End B ( F ( X ) ) ) I_X=\ker\bigl(\operatorname{End}_{\mathcal{A}}(X)\longrightarrow\operatorname{End}_{\mathcal{B}}(F(X))\bigr) I X = ker ( End A ( X ) ⟶ End B ( F ( X )) ) is nilpotent. Then B \mathcal{B} B is also Hom-finite and Krull–Schmidt. The functor F F F reflects isomorphisms and induces a bijection on isomorphism classes preserving finite direct sums and indecomposability.
Proof. By fullness and R R R -linearity, Hom B ( F ( X ) , F ( Y ) ) \operatorname{Hom}_{\mathcal{B}}(F(X),F(Y)) Hom B ( F ( X ) , F ( Y )) is a quotient of Hom A ( X , Y ) \operatorname{Hom}_{\mathcal{A}}(X,Y) Hom A ( X , Y ) as an R R R -module for all X , Y ∈ A X,Y\in\mathcal{A} X , Y ∈ A . Essential surjectivity therefore implies that B \mathcal{B} B is Hom-finite.
For each indecomposable object X ∈ A X\in\mathcal{A} X ∈ A , we have End B ( F ( X ) ) ≅ End A ( X ) / I X \operatorname{End}_{\mathcal{B}}(F(X))\cong\operatorname{End}_{\mathcal{A}}(X)/I_X End B ( F ( X )) ≅ End A ( X ) / I X . Since End A ( X ) \operatorname{End}_{\mathcal{A}}(X) End A ( X ) is local and I X I_X I X is nilpotent, this quotient is a nonzero local ring. Applying F F F to indecomposable decompositions in A \mathcal{A} A and using essential surjectivity shows that B \mathcal{B} B is Krull–Schmidt.
We next show that F F F reflects isomorphisms. Let u : X → Y u:X\to Y u : X → Y be a morphism such that F ( u ) F(u) F ( u ) is invertible. By fullness, choose v : Y → X v:Y\to X v : Y → X with F ( v ) = F ( u ) − 1 F(v)=F(u)^{-1} F ( v ) = F ( u ) − 1 . Then
v ∘ u = 1 X + ε X , u ∘ v = 1 Y + ε Y , v\circ u=1_X+\varepsilon_X,\qquad u\circ v=1_Y+\varepsilon_Y, v ∘ u = 1 X + ε X , u ∘ v = 1 Y + ε Y ,
where ε X ∈ I X \varepsilon_X\in I_X ε X ∈ I X and ε Y ∈ I Y \varepsilon_Y\in I_Y ε Y ∈ I Y . Since ε X \varepsilon_X ε X and ε Y \varepsilon_Y ε Y are nilpotent, both v ∘ u v\circ u v ∘ u and u ∘ v u\circ v u ∘ v are invertible. Hence u u u is invertible. Essential surjectivity gives surjectivity on isomorphism classes. By fullness, any isomorphism F ( X ) ≅ F ( Y ) F(X)\cong F(Y) F ( X ) ≅ F ( Y ) lifts to a morphism X → Y X\to Y X → Y , which is invertible by the preceding argument. Thus F F F induces a bijection on isomorphism classes. Preservation of finite direct sums follows from additivity.
The isomorphism End B ( F ( X ) ) ≅ End A ( X ) / I X \operatorname{End}_{\mathcal{B}}(F(X))\cong\operatorname{End}_{\mathcal{A}}(X)/I_X End B ( F ( X )) ≅ End A ( X ) / I X shows that F F F preserves indecomposability. Since F F F is additive and F ( X ) ≅ 0 F(X)\cong0 F ( X ) ≅ 0 implies X ≅ 0 X\cong0 X ≅ 0 , it also reflects indecomposability. □
A finiteness criterion for comma categories
Throughout this section, D D D denotes a division Artin R R R -algebra.
Definition 3.1 (cf. [12 II, §6] ). For Hom-finite Krull–Schmidt categories X \mathcal{X} X and Y \mathcal{Y} Y , fix additive functors
G : X ⟶ mod D , F : Y ⟶ mod D . G:\mathcal{X}\longrightarrow\operatorname{mod} D,\qquad F:\mathcal{Y}\longrightarrow\operatorname{mod} D. G : X ⟶ mod D , F : Y ⟶ mod D . The comma category M ( G , F ) = ( G ↓ F ) \mathcal{M}(G,F)=(G\downarrow F) M ( G , F ) = ( G ↓ F ) is defined as follows.
Objects are triples ( X , Y , u ) (X,Y,u) ( X , Y , u ) with X ∈ X X\in\mathcal{X} X ∈ X , Y ∈ Y Y\in\mathcal{Y} Y ∈ Y and a homomorphism of D D D -modules u : G ( X ) ⟶ F ( Y ) u:G(X)\longrightarrow F(Y) u : G ( X ) ⟶ F ( Y ) .
A morphism ( X , Y , u ) ⟶ ( X ′ , Y ′ , u ′ ) (X,Y,u)\longrightarrow(X',Y',u') ( X , Y , u ) ⟶ ( X ′ , Y ′ , u ′ ) is a pair ( f , g ) (f,g) ( f , g ) with f : X ⟶ X ′ f:X\longrightarrow X' f : X ⟶ X ′ in X \mathcal{X} X and g : Y ⟶ Y ′ g:Y\longrightarrow Y' g : Y ⟶ Y ′ in Y \mathcal{Y} Y satisfying
F ( g ) ∘ u = u ′ ∘ G ( f ) . F(g)\circ u=u'\circ G(f). F ( g ) ∘ u = u ′ ∘ G ( f ) . Identities and composition are defined componentwise.
Remark 3.2. A morphism ( f , g ) (f,g) ( f , g ) in M ( G , F ) \mathcal{M}(G,F) M ( G , F ) is an isomorphism if and only if both f f f and g g g are isomorphisms. The category M ( G , F ) \mathcal{M}(G,F) M ( G , F ) is additive and idempotent complete by [7 Proposition 3.5] , with direct sums formed componentwise. Each morphism space is an R R R -submodule of the direct sum of the two component morphism spaces, so M ( G , F ) \mathcal{M}(G,F) M ( G , F ) is Hom-finite. It is therefore Krull–Schmidt by [10 Corollary 4.4] . If X , Y X,Y X , Y are indecomposable and u ≠ 0 u\ne0 u = 0 , then ( X , Y , u ) (X,Y,u) ( X , Y , u ) is indecomposable because a nontrivial decomposition would place X X X and Y Y Y in different summands and force u = 0 u=0 u = 0 .
We now state the main theorem of this section.
Theorem 3.3. Let M ( G , F ) \mathcal{M}(G,F) M ( G , F ) be as in Definition 3.1 , and assume that X \mathcal{X} X and Y \mathcal{Y} Y are representation-finite. Suppose that any two indecomposable objects ( X , Y , u ) (X,Y,u) ( X , Y , u ) and ( X ′ , Y ′ , u ′ ) (X',Y',u') ( X ′ , Y ′ , u ′ ) of M ( G , F ) \mathcal{M}(G,F) M ( G , F ) are isomorphic whenever X ≅ X ′ X\cong X' X ≅ X ′ and Y ≅ Y ′ Y\cong Y' Y ≅ Y ′ . Then M ( G , F ) \mathcal{M}(G,F) M ( G , F ) is representation-finite.
Quadratic forms and weak positivity
Consider an additive functor G : A ⟶ mod D G:\mathcal{A}\longrightarrow\operatorname{mod} D G : A ⟶ mod D , where A \mathcal{A} A is a Hom-finite Krull–Schmidt category. We say that G G G is faithful on objects if G ( X ) = 0 G(X)=0 G ( X ) = 0 implies X ≅ 0 X\cong0 X ≅ 0 . The kernel ideal of G G G is given by
( ker G ) ( X , Y ) = { f ∈ Hom A ( X , Y ) : G ( f ) = 0 } . (\ker G)(X,Y)=\{f\in\operatorname{Hom}_{\mathcal{A}}(X,Y):G(f)=0\}. ( ker G ) ( X , Y ) = { f ∈ Hom A ( X , Y ) : G ( f ) = 0 } .
Set A ‾ = A / ker G \overline{\mathcal{A}}=\mathcal{A}/\ker G A = A / ker G , and let π : A ⟶ A ‾ \pi:\mathcal{A}\longrightarrow\overline{\mathcal{A}} π : A ⟶ A be the canonical quotient functor. Then G G G induces a faithful additive functor G ‾ : A ‾ ⟶ mod D \overline{G}:\overline{\mathcal{A}}\longrightarrow\operatorname{mod} D G : A ⟶ mod D with G = G ‾ ∘ π G=\overline{G}\circ\pi G = G ∘ π .
For a Hom-finite Krull–Schmidt category B \mathcal{B} B and an additive functor F : B ⟶ mod D F:\mathcal{B}\longrightarrow\operatorname{mod} D F : B ⟶ mod D , the quotient functor induces
Π : M ( G , F ) ⟶ M ( G ‾ , F ) , ( X , Y , u ) ⟼ ( π ( X ) , Y , u ) , \Pi:\mathcal{M}(G,F)\longrightarrow\mathcal{M}(\overline{G},F),\qquad(X,Y,u)\longmapsto(\pi(X),Y,u), Π : M ( G , F ) ⟶ M ( G , F ) , ( X , Y , u ) ⟼ ( π ( X ) , Y , u ) ,
with Π ( f , g ) = ( π ( f ) , g ) \Pi(f,g)=(\pi(f),g) Π ( f , g ) = ( π ( f ) , g ) on morphisms.
Lemma 3.4. Assume that G G G is faithful on objects. Then A ‾ \overline{\mathcal{A}} A is Hom-finite and Krull–Schmidt. Both π \pi π and Π \Pi Π induce bijections on isomorphism classes preserving direct sums and indecomposability. Similarly, the induced functor M ( F , G ) ⟶ M ( F , G ‾ ) \mathcal{M}(F,G)\longrightarrow\mathcal{M}(F,\overline{G}) M ( F , G ) ⟶ M ( F , G ) induces a bijection on isomorphism classes preserving direct sums and indecomposability.
Proof. For X ∈ A X\in\mathcal{A} X ∈ A , set
I X = ker ( End A ( X ) ⟶ End D ( G ( X ) ) ) . I_X=\ker\bigl(\operatorname{End}_{\mathcal{A}}(X)\longrightarrow\operatorname{End}_{D}(G(X))\bigr). I X = ker ( End A ( X ) ⟶ End D ( G ( X )) ) .
We first show that I X ⊆ J ( End A ( X ) ) I_X\subseteq J(\operatorname{End}_{\mathcal{A}}(X)) I X ⊆ J ( End A ( X )) . If X X X is indecomposable, I X I_X I X is a proper ideal because G ( X ) ≠ 0 G(X)\ne0 G ( X ) = 0 , and hence is contained in the radical of the local ring End A ( X ) \operatorname{End}_{\mathcal{A}}(X) End A ( X ) . In general, write X X X as a direct sum of indecomposables. If G ( f ) = 0 G(f)=0 G ( f ) = 0 , the preceding observation on each indecomposable summand and Lemma 2.4 give f ∈ J ( End A ( X ) ) f\in J(\operatorname{End}_{\mathcal{A}}(X)) f ∈ J ( End A ( X )) . Thus I X I_X I X is nilpotent. Since π \pi π is full and surjective on objects, Lemma 2.7 proves the assertions about A ‾ \overline{\mathcal{A}} A and π \pi π . Since the quotient category has the same objects as A \mathcal{A} A and G ‾ ∘ π = G \overline{G} \circ\pi= G G ∘ π = G , the functor Π \Pi Π is surjective on objects. To prove fullness, take a morphism ( f ‾ , g ) : Π ( X , Y , u ) → Π ( X ′ , Y ′ , u ′ ) (\overline{f},g):\Pi(X,Y,u)\to\Pi(X',Y',u') ( f , g ) : Π ( X , Y , u ) → Π ( X ′ , Y ′ , u ′ ) and choose f : X → X ′ f:X\to X' f : X → X ′ with π ( f ) = f ‾ \pi(f)=\overline{f} π ( f ) = f . Since ( f ‾ , g ) (\overline{f},g) ( f , g ) is a morphism in M ( G ‾ , F ) \mathcal{M}(\overline{G},F) M ( G , F ) , we have
F ( g ) ∘ u = u ′ ∘ G ‾ ( f ‾ ) = u ′ ∘ G ( f ) . F(g)\circ u=u'\circ\overline{G}(\overline{f})=u'\circ G(f). F ( g ) ∘ u = u ′ ∘ G ( f ) = u ′ ∘ G ( f ) .
Thus ( f , g ) (f,g) ( f , g ) is a morphism in M ( G , F ) \mathcal{M}(G,F) M ( G , F ) and Π ( f , g ) = ( f ‾ , g ) \Pi(f,g)=(\overline{f},g) Π ( f , g ) = ( f , g ) , so Π \Pi Π is full. For Z = ( X , Y , u ) Z=(X,Y,u) Z = ( X , Y , u ) , the kernel of the induced map on endomorphism rings is
{ ( f , 0 ) : f ∈ I X } . \{(f,0):f\in I_X\}. {( f , 0 ) : f ∈ I X } .
Since I X I_X I X is nilpotent, this kernel is nilpotent. By Lemma 2.7 , Π \Pi Π induces a bijection on isomorphism classes preserving direct sums and indecomposability. □
Let m = ker ( R → D ) \mathfrak{m}=\ker(R\to D) m = ker ( R → D ) and k = R / m \mathbb{k}=R/\mathfrak{m} k = R / m . Since D D D is a division ring, m \mathfrak{m} m is prime. Since R R R is a commutative Artinian ring, the prime ideal m \mathfrak{m} m is maximal by [1 Proposition 8.1] . Hence k = R / m \mathbb{k}=R/\mathfrak{m} k = R / m is a field. Since m D = 0 \mathfrak{m}D=0 m D = 0 , any finite set of R R R -module generators of D D D spans D D D over k \mathbb{k} k . Thus D D D is finite-dimensional over k \mathbb{k} k .
Let H : E → mod D H:\mathcal{E}\to\operatorname{mod}D H : E → mod D be a faithful additive R R R -linear functor, where E \mathcal{E} E is Hom-finite. For r ∈ m r\in\mathfrak{m} r ∈ m and a morphism f f f in E \mathcal{E} E , we have H ( r f ) = r H ( f ) = 0 H(rf)=rH(f)=0 H ( r f ) = r H ( f ) = 0 . Faithfulness gives r f = 0 rf=0 r f = 0 . Thus each morphism space of E \mathcal{E} E is annihilated by m \mathfrak{m} m and, since it has finite R R R -length, is finite-dimensional over k \mathbb{k} k . Consequently, E \mathcal{E} E is a Hom-finite k \mathbb{k} k -linear category, and H H H is a k \mathbb{k} k -linear functor.
Definition 3.5. Let X , Y \mathcal{X},\mathcal{Y} X , Y be nonzero representation-finite Hom-finite Krull–Schmidt k \mathbb{k} k -linear categories, and let G : X → mod D G:\mathcal{X}\to\operatorname{mod}D G : X → mod D and F : Y → mod D F:\mathcal{Y}\to\operatorname{mod}D F : Y → mod D be faithful additive k \mathbb{k} k -linear functors.
(1) For X ∈ X X\in\mathcal{X} X ∈ X and Y ∈ Y Y\in\mathcal{Y} Y ∈ Y , we define q ( X , Y ) q(X,Y) q ( X , Y ) by
q ( X , Y ) = dim k End X ( X ) + dim k End Y ( Y ) − dim k Hom D ( G ( X ) , F ( Y ) ) . (1) q(X,Y)=\dim_{\mathbb{k}}\operatorname{End}_{\mathcal{X}}(X)+\dim_{\mathbb{k}}\operatorname{End}_{\mathcal{Y}}(Y)-\dim_{\mathbb{k}}\operatorname{Hom}_{D}(G(X),F(Y)). \tag*{(1)} q ( X , Y ) = dim k End X ( X ) + dim k End Y ( Y ) − dim k Hom D ( G ( X ) , F ( Y )) . (1) (2) The form q q q is weakly positive if q ( X , Y ) > 0 q(X,Y)>0 q ( X , Y ) > 0 whenever X X X and Y Y Y are not both zero.
Choose representatives X 1 , … , X r X_1,\ldots,X_r X 1 , … , X r and Y 1 , … , Y s Y_1,\ldots,Y_s Y 1 , … , Y s of the isomorphism classes of indecomposable objects in X \mathcal{X} X and Y \mathcal{Y} Y , respectively. We write
X = ⨁ i = 1 r X i n i , Y = ⨁ j = 1 s Y j m j , n i , m j ≥ 0. X=\bigoplus_{i=1}^{r}X_i^{n_i},\qquad Y=\bigoplus_{j=1}^{s}Y_j^{m_j},\qquad n_i,m_j\geq0. X = i = 1 ⨁ r X i n i , Y = j = 1 ⨁ s Y j m j , n i , m j ≥ 0. Then q q q is an integral quadratic form in ( n , m ) (\mathbf{n},\mathbf{m}) ( n , m ) .
Proposition 3.6. Let X , Y , G , F , q \mathcal{X},\mathcal{Y},G,F,q X , Y , G , F , q be as in Definition 3.5. Suppose that any two indecomposable objects of M ( G , F ) \mathcal{M}(G,F) M ( G , F ) with isomorphic components are isomorphic. Then the following hold.
(1) For any indecomposable objects X 1 ∈ X X_1\in\mathcal{X} X 1 ∈ X and Y 1 ∈ Y Y_1\in\mathcal{Y} Y 1 ∈ Y , the rings End X ( X 1 ) \operatorname{End}_{\mathcal{X}}(X_1) End X ( X 1 ) and End Y ( Y 1 ) \operatorname{End}_{\mathcal{Y}}(Y_1) End Y ( Y 1 ) are division rings.
(2) The quadratic form q q q is weakly positive.
Proof. (1) Fix indecomposable objects X 1 ∈ X X_1\in\mathcal{X} X 1 ∈ X and Y 1 ∈ Y Y_1\in\mathcal{Y} Y 1 ∈ Y , and set
E X = End X ( X 1 ) , E Y = End Y ( Y 1 ) , E_X=\operatorname{End}_{\mathcal{X}}(X_1),\qquad E_Y=\operatorname{End}_{\mathcal{Y}}(Y_1), E X = End X ( X 1 ) , E Y = End Y ( Y 1 ) , U = G ( X 1 ) , V = F ( Y 1 ) , W = Hom D ( U , V ) . U=G(X_1),\qquad V=F(Y_1),\qquad W=\operatorname{Hom}_{D}(U,V). U = G ( X 1 ) , V = F ( Y 1 ) , W = Hom D ( U , V ) . Since G G G and F F F are faithful, U , V U,V U , V are nonzero, so W ≠ 0 W\ne0 W = 0 . For nonzero u , v ∈ W u,v\in W u , v ∈ W , the objects ( X 1 , Y 1 , u ) (X_1,Y_1,u) ( X 1 , Y 1 , u ) and ( X 1 , Y 1 , v ) (X_1,Y_1,v) ( X 1 , Y 1 , v ) are indecomposable by Remark 3.2. They have the same components, so the hypothesis gives an isomorphism ( f , g ) : ( X 1 , Y 1 , u ) → ( X 1 , Y 1 , v ) (f,g):(X_1,Y_1,u)\to(X_1,Y_1,v) ( f , g ) : ( X 1 , Y 1 , u ) → ( X 1 , Y 1 , v ) , with f ∈ E X × f\in E_X^{\times} f ∈ E X × and g ∈ E Y × g\in E_Y^{\times} g ∈ E Y × . Then we have
F ( g ) ∘ u = v ∘ G ( f ) , v = F ( g ) ∘ u ∘ G ( f ) − 1 . (2) F(g)\circ u=v\circ G(f),\qquad v=F(g)\circ u\circ G(f)^{-1}. \tag*{(2)} F ( g ) ∘ u = v ∘ G ( f ) , v = F ( g ) ∘ u ∘ G ( f ) − 1 . (2)
Therefore E X × × E Y × E_X^{\times} \times E_Y^{\times} E X × × E Y × acts transitively on W ∖ { 0 } W\setminus\{0\} W ∖ { 0 } by postcomposition and inverse precomposition. Since E X , E Y E_X, E_Y E X , E Y are finite-dimensional local k k k -algebras, Lemma 2.6 gives
J ( E Y ) W = 0 = W J ( E X ) . J(E_Y)W = 0 = WJ(E_X). J ( E Y ) W = 0 = W J ( E X ) .
Let f ∈ J ( E X ) f \in J(E_X) f ∈ J ( E X ) . Then w ∘ G ( f ) = 0 w \circ G(f) = 0 w ∘ G ( f ) = 0 for every w ∈ W w \in W w ∈ W . If G ( f ) ≠ 0 G(f) \ne0 G ( f ) = 0 , choose x ∈ U x \in U x ∈ U with z = G ( f ) ( x ) ≠ 0 z = G(f)(x) \ne0 z = G ( f ) ( x ) = 0 . Extend z z z to a D D D -basis z , z 2 , … , z t z,z_2,\ldots,z_t z , z 2 , … , z t of U U U . Choose 0 ≠ y ∈ V 0 \ne y \in V 0 = y ∈ V and define w : U → V w: U \to V w : U → V D D D -linearly by w ( z ) = y w(z)=y w ( z ) = y and w ( z i ) = 0 w(z_i)=0 w ( z i ) = 0 for i ≥ 2 i \ge2 i ≥ 2 . Then ( w ∘ G ( f ) ) ( x ) = y ≠ 0 (w \circ G(f))(x)=y \ne0 ( w ∘ G ( f )) ( x ) = y = 0 , a contradiction. Thus G ( f ) = 0 G(f)=0 G ( f ) = 0 , and faithfulness of G G G gives f = 0 f=0 f = 0 . Hence J ( E X ) = 0 J(E_X)=0 J ( E X ) = 0 . Similarly, J ( E Y ) = 0 J(E_Y)=0 J ( E Y ) = 0 . Since End X ( X 1 ) \operatorname{End}_{\mathcal{X}}(X_1) End X ( X 1 ) and End Y ( Y 1 ) \operatorname{End}_{\mathcal{Y}}(Y_1) End Y ( Y 1 ) are local rings with zero radical, they are division rings.
(2) The inequality q ( X , Y ) > 0 q(X,Y)>0 q ( X , Y ) > 0 is immediate when exactly one of X X X and Y Y Y is zero. We may therefore assume that X X X and Y Y Y are nonzero and write
X = ⨁ i = 1 r X i n i , Y = ⨁ j = 1 s Y j m j . X=\bigoplus_{i=1}^{r}X_i^{n_i}, \qquad Y=\bigoplus_{j=1}^{s}Y_j^{m_j}. X = i = 1 ⨁ r X i n i , Y = j = 1 ⨁ s Y j m j .
Set
E 1 = End Y ( Y ) , E 2 = End X ( X ) , W = Hom D ( G ( X ) , F ( Y ) ) , E_1=\operatorname{End}_{\mathcal{Y}}(Y), \qquad E_2=\operatorname{End}_{\mathcal{X}}(X), \qquad W=\operatorname{Hom}_D(G(X),F(Y)), E 1 = End Y ( Y ) , E 2 = End X ( X ) , W = Hom D ( G ( X ) , F ( Y )) ,
c = dim k E 1 + dim k E 2 , w = dim k W . c=\dim_k E_1+\dim_k E_2, \qquad w=\dim_k W. c = dim k E 1 + dim k E 2 , w = dim k W .
The space W W W is an E 1 E_1 E 1 –E 2 E_2 E 2 -bimodule by postcomposition and precomposition, and q ( X , Y ) = c − w q(X,Y)=c-w q ( X , Y ) = c − w . Suppose that c ≤ w c\le w c ≤ w .
Set M = M ( G , F ) \mathcal{M}=\mathcal{M}(G,F) M = M ( G , F ) and
δ = ∑ i = 1 r n i + ∑ j = 1 s m j , N = dim k G ( X ) + dim k F ( Y ) . \delta=\sum_{i=1}^{r}n_i+\sum_{j=1}^{s}m_j, \qquad N=\dim_k G(X)+\dim_k F(Y). δ = i = 1 ∑ r n i + j = 1 ∑ s m j , N = dim k G ( X ) + dim k F ( Y ) .
For each positive integer h h h , the group Γ h = GL h ( E 1 ) × GL h ( E 2 ) \Gamma_h=\operatorname{GL}_h(E_1)\times\operatorname{GL}_h(E_2) Γ h = GL h ( E 1 ) × GL h ( E 2 ) acts on the space M h ( W ) M_h(W) M h ( W ) of h × h h\times h h × h matrices with entries in W W W by ( g 1 , g 2 ) ⋅ u = g 1 u g 2 − 1 (g_1,g_2)\cdot u=g_1ug_2^{-1} ( g 1 , g 2 ) ⋅ u = g 1 u g 2 − 1 . For u ∈ M h ( W ) u\in M_h(W) u ∈ M h ( W ) , write Z u = ( X ⊕ h , Y ⊕ h , u ) ∈ M Z_u=(X^{\oplus h},Y^{\oplus h},u)\in\mathcal{M} Z u = ( X ⊕ h , Y ⊕ h , u ) ∈ M . By the definition of morphisms in M \mathcal{M} M , we have Z u ≅ Z v Z_u\cong Z_v Z u ≅ Z v if and only if u u u and v v v belong to the same Γ h \Gamma_h Γ h -orbit.
Let Ω h \Omega_h Ω h be the union of those Γ h \Gamma_h Γ h -orbits corresponding to objects Z u Z_u Z u admitting an endomorphism annihilated by some monic irreducible polynomial ϕ ∈ k [ T ] \phi\in k[T] ϕ ∈ k [ T ] of degree h h h . Let β h \beta_h β h be the number of Γ h \Gamma_h Γ h -orbits in Ω h \Omega_h Ω h , equivalently, the number of isomorphism classes of objects Z u Z_u Z u with u ∈ Ω h u\in\Omega_h u ∈ Ω h .
For u ∈ Ω h u\in\Omega_h u ∈ Ω h , let Z ′ = ( X ′ , Y ′ , v ) Z'=(X',Y',v) Z ′ = ( X ′ , Y ′ , v ) be an indecomposable direct summand of Z u Z_u Z u . Then
X ′ ≅ ⨁ i = 1 r X i n i ′ , Y ′ ≅ ⨁ j = 1 s Y j m j ′ , 0 ≤ n i ′ ≤ h n i , 0 ≤ m j ′ ≤ h m j . X'\cong\bigoplus_{i=1}^{r}X_i^{n'_i}, \qquad Y'\cong\bigoplus_{j=1}^{s}Y_j^{m'_j}, \qquad0\le n'_i\le hn_i,\qquad0\le m'_j\le hm_j. X ′ ≅ i = 1 ⨁ r X i n i ′ , Y ′ ≅ j = 1 ⨁ s Y j m j ′ , 0 ≤ n i ′ ≤ h n i , 0 ≤ m j ′ ≤ h m j .
By the hypothesis, each pair of multiplicity vectors gives at most one isomorphism class of indecomposable objects. Let U 1 , … , U t U_1,\ldots,U_t U 1 , … , U t be pairwise nonisomorphic representatives of the indecomposable direct summands of Z u Z_u Z u , as u u u ranges over Ω h \Omega_h Ω h . Then
t ≤ ∏ i = 1 r ( h n i + 1 ) ∏ j = 1 s ( h m j + 1 ) ≤ ( h δ + 1 ) r + s . t\le\prod_{i=1}^{r}(hn_i+1)\prod_{j=1}^{s}(hm_j+1)\le(h\delta+1)^{r+s}. t ≤ i = 1 ∏ r ( h n i + 1 ) j = 1 ∏ s ( h m j + 1 ) ≤ ( h δ + 1 ) r + s .
By Krull–Schmidt, for each u ∈ Ω h u\in\Omega_h u ∈ Ω h we have
Z u ≅ ⨁ ℓ = 1 t U ℓ a ℓ Z_u\cong\bigoplus_{\ell=1}^{t}U_\ell^{a_\ell} Z u ≅ ℓ = 1 ⨁ t U ℓ a ℓ
with 0 ≤ a ℓ ≤ h δ 0\le a_\ell\le h\delta 0 ≤ a ℓ ≤ h δ .
Consider the functor H : M → mod k H:\mathcal{M}\to\operatorname{mod} k H : M → mod k defined by H ( X ′ , Y ′ , v ) = G ( X ′ ) ⊕ F ( Y ′ ) H(X',Y',v)=G(X')\oplus F(Y') H ( X ′ , Y ′ , v ) = G ( X ′ ) ⊕ F ( Y ′ ) and H ( f , g ) = G ( f ) ⊕ F ( g ) H(f,g)=G(f)\oplus F(g) H ( f , g ) = G ( f ) ⊕ F ( g ) . This functor is faithful and k k k -linear. For u ∈ Ω h u\in\Omega_h u ∈ Ω h , we have dim k H ( Z u ) = h N \dim_k H(Z_u)=hN dim k H ( Z u ) = h N . By Lemma 2.5 , Z u Z_u Z u has at most N N N pairwise nonisomorphic indecomposable direct summands. Hence at most N N N of a 1 , … , a t a_1,\ldots,a_t a 1 , … , a t are nonzero. Thus the isomorphism class of Z u Z_u Z u is determined by at most N N N pairs ( U ℓ , a ℓ ) (U_\ell,a_\ell) ( U ℓ , a ℓ ) , where 1 ≤ ℓ ≤ t 1\leq\ell\leq t 1 ≤ ℓ ≤ t and 1 ≤ a ℓ ≤ h δ 1\leq a_\ell\leq h\delta 1 ≤ a ℓ ≤ h δ . Consequently,
β h ≤ ( 1 + t h δ ) N ≤ ( h δ + 1 ) N ( r + s + 1 ) . (3) \beta_h\leq(1+th\delta)^N\leq(h\delta+1)^{N(r+s+1)}. \tag*{(3)} β h ≤ ( 1 + t h δ ) N ≤ ( h δ + 1 ) N ( r + s + 1 ) . (3)
Case (1). k k k is an infinite field. We now take h = 1 h=1 h = 1 . Since the zero endomorphism of every Z u Z_u Z u is annihilated by the monic irreducible polynomial T T T , we have Ω 1 = W \Omega_1=W Ω 1 = W . Thus the action of E 1 × × E 2 × E_1^\times\times E_2^\times E 1 × × E 2 × on W W W has finitely many orbits. We now show that each orbit is contained in the zero set of a nonzero polynomial on W W W .
Let a = dim k E 1 a=\dim_k E_1 a = dim k E 1 and b = dim k E 2 b=\dim_k E_2 b = dim k E 2 . Then a + b = c a+b=c a + b = c . Choose k k k -bases e 1 , … , e a e_1,\ldots,e_a e 1 , … , e a of E 1 E_1 E 1 , d 1 , … , d b d_1,\ldots,d_b d 1 , … , d b of E 2 E_2 E 2 , and f 1 , … , f w f_1,\ldots,f_w f 1 , … , f w of W W W . Fix u ∈ W u\in W u ∈ W , and denote its orbit by O ( u ) \mathcal{O}(u) O ( u ) . We have
O ( u ) = { F ( g ) ∘ u ∘ G ( f ) : g ∈ E 1 × , f ∈ E 2 × } . \mathcal{O}(u)=\{F(g)\circ u\circ G(f):g\in E_1^\times,\ f\in E_2^\times\}. O ( u ) = { F ( g ) ∘ u ∘ G ( f ) : g ∈ E 1 × , f ∈ E 2 × } .
For g = ∑ i = 1 a s i e i g=\sum_{i=1}^{a}s_i e_i g = ∑ i = 1 a s i e i and f = ∑ j = 1 b t j d j f=\sum_{j=1}^{b}t_jd_j f = ∑ j = 1 b t j d j , k k k -linearity gives
F ( g ) ∘ u ∘ G ( f ) = ∑ i = 1 a ∑ j = 1 b s i t j F ( e i ) ∘ u ∘ G ( d j ) . F(g)\circ u\circ G(f)=\sum_{i=1}^{a}\sum_{j=1}^{b}s_it_jF(e_i)\circ u\circ G(d_j). F ( g ) ∘ u ∘ G ( f ) = i = 1 ∑ a j = 1 ∑ b s i t j F ( e i ) ∘ u ∘ G ( d j ) .
Here u u u remains fixed, so the vectors F ( e i ) ∘ u ∘ G ( d j ) F(e_i)\circ u\circ G(d_j) F ( e i ) ∘ u ∘ G ( d j ) are fixed as g , f g,f g , f vary. Since g g g is invertible, some coordinate s p s_p s p is nonzero. Replace g , f g,f g , f by
g ′ = s p − 1 g = e p + ∑ i = 1 i ≠ p a s i s p e i , f ′ = s p f = ∑ j = 1 b s p t j d j . g'=s_p^{-1}g=e_p+\sum_{\substack{i=1\\i\ne p}}^{a}\frac{s_i}{s_p}e_i,\qquad f'=s_pf=\sum_{j=1}^{b}s_pt_jd_j. g ′ = s p − 1 g = e p + i = 1 i = p ∑ a s p s i e i , f ′ = s p f = j = 1 ∑ b s p t j d j .
These are again units, and
F ( g ′ ) ∘ u ∘ G ( f ′ ) = s p − 1 s p F ( g ) ∘ u ∘ G ( f ) = F ( g ) ∘ u ∘ G ( f ) . F(g')\circ u\circ G(f')=s_p^{-1}s_pF(g)\circ u\circ G(f)=F(g)\circ u\circ G(f). F ( g ′ ) ∘ u ∘ G ( f ′ ) = s p − 1 s p F ( g ) ∘ u ∘ G ( f ) = F ( g ) ∘ u ∘ G ( f ) .
Thus the same orbit point has a representation in which the coefficient of e p e_p e p in g ′ g' g ′ is 1 1 1 . Once p p p is chosen, the remaining coefficients x i = s i / s p x_i=s_i/s_p x i = s i / s p for i ≠ p i\ne p i = p and y j = s p t j y_j=s_pt_j y j = s p t j still vary. There are ( a − 1 ) + b = c − 1 (a-1)+b=c-1 ( a − 1 ) + b = c − 1 such variables.
For each p ∈ { 1 , … , a } p\in\{1,\ldots,a\} p ∈ { 1 , … , a } , define
Φ u , p : k a − 1 × k b ⟶ W , \Phi_{u,p}:k^{a-1}\times k^b\longrightarrow W, Φ u , p : k a − 1 × k b ⟶ W ,
( x , y ) ⟼ F ( e p + ∑ i = 1 i ≠ p a x i e i ) ∘ u ∘ G ( ∑ j = 1 b y j d j ) , (x,y)\longmapsto F\left(e_p+\sum_{\substack{i=1\\i\ne p}}^{a}x_ie_i\right)\circ u\circ G\left(\sum_{j=1}^{b}y_jd_j\right), ( x , y ) ⟼ F e p + i = 1 i = p ∑ a x i e i ∘ u ∘ G ( j = 1 ∑ b y j d j ) ,
where x = ( x i ) i ≠ p x=(x_i)_{i\ne p} x = ( x i ) i = p and y = ( y 1 , … , y b ) y=(y_1,\ldots,y_b) y = ( y 1 , … , y b ) . The index p p p specifies which coordinate is set to 1 1 1 . Since every invertible g g g has a nonzero coordinate, the preceding normalization shows that
O ( u ) ⊆ ⋃ p = 1 a im Φ u , p . \mathcal{O}(u)\subseteq\bigcup_{p=1}^{a}\operatorname{im}\Phi_{u,p}. O ( u ) ⊆ p = 1 ⋃ a im Φ u , p .
Expanding in the basis f 1 , … , f w f_1,\ldots,f_w f 1 , … , f w of W W W gives
F ( e i ) ∘ u ∘ G ( d j ) = ∑ ℓ = 1 w γ ℓ i j f ℓ , γ ℓ i j ∈ k . F(e_i)\circ u\circ G(d_j)=\sum_{\ell=1}^{w}\gamma_{\ell ij}f_\ell,\qquad\gamma_{\ell ij}\in k. F ( e i ) ∘ u ∘ G ( d j ) = ℓ = 1 ∑ w γ ℓ ij f ℓ , γ ℓ ij ∈ k .
For a fixed p p p , the coordinates of Φ u , p \Phi_{u,p} Φ u , p in the chosen basis of W W W are therefore given by
Φ u , p ( x , y ) = ∑ ℓ = 1 w r ℓ ( x , y ) f ℓ , \Phi_{u,p}(x,y)=\sum_{\ell=1}^{w}r_\ell(x,y)f_\ell, Φ u , p ( x , y ) = ℓ = 1 ∑ w r ℓ ( x , y ) f ℓ ,
r ℓ ( x , y ) = ∑ j = 1 b γ ℓ p j y j + ∑ i = 1 i ≠ p a ∑ j = 1 b γ ℓ i j x i y j . r_\ell(x,y)=\sum_{j=1}^{b}\gamma_{\ell pj}y_j+\sum_{\substack{i=1\\i\ne p}}^{a}\sum_{j=1}^{b}\gamma_{\ell ij}x_iy_j. r ℓ ( x , y ) = j = 1 ∑ b γ ℓ p j y j + i = 1 i = p ∑ a j = 1 ∑ b γ ℓ ij x i y j .
Each r ℓ r_\ell r ℓ is a polynomial of total degree at most 2 in the c − 1 c-1 c − 1 remaining variables. Since c − 1 < w c-1<w c − 1 < w , these w w w polynomials are algebraically dependent by [11 VIII, Theorem 1.1] . Hence there is a nonzero polynomial P u , p ∈ k [ T 1 , … , T w ] P_{u,p}\in k[T_1,\ldots,T_w] P u , p ∈ k [ T 1 , … , T w ] such that
P u , p ( r 1 ( x , y ) , … , r w ( x , y ) ) = 0. P_{u,p}\bigl(r_1(\mathbf{x},y),\ldots,r_w(\mathbf{x},y)\bigr)=0. P u , p ( r 1 ( x , y ) , … , r w ( x , y ) ) = 0.
Using the chosen basis of W W W , we regard P u , p P_{u,p} P u , p as a polynomial function on W W W . Thus it vanishes on im Φ u , p \operatorname{im}\Phi_{u,p} im Φ u , p . Every point of O ( u ) O(u) O ( u ) lies in at least one of these images, so P u = ∏ p = 1 β 1 P u , p P_u=\prod_{p=1}^{\beta_1}P_{u,p} P u = ∏ p = 1 β 1 P u , p vanishes on O ( u ) O(u) O ( u ) . Since the polynomial ring is a domain, P u ≠ 0 P_u\ne0 P u = 0 .
Choose representatives u 1 , … , u β 1 u_1,\ldots,u_{\beta_1} u 1 , … , u β 1 of the finitely many orbits. The product P u 1 ⋯ P u β 1 P_{u_1}\cdots P_{u_{\beta_1}} P u 1 ⋯ P u β 1 is nonzero and vanishes at every point of W ≅ k w W\cong k^w W ≅ k w . This contradicts [11 IV, Corollary 1.7] .
Case (2). k k k is a finite field. Write k = F p n k=\mathbb{F}_{p^n} k = F p n , where p p p is prime and n ≥ 1 n\ge1 n ≥ 1 .
We now assume that h h h is prime. Let I h \mathcal{I}_h I h be the set of monic irreducible polynomials of degree h h h over k k k . By [11 Chapter V, Exercise 22] , we have ∣ I h ∣ = ( p n h − p n ) / h |\mathcal{I}_h|=(p^{nh}-p^n)/h ∣ I h ∣ = ( p nh − p n ) / h .
For ϕ ∈ I h \phi\in\mathcal{I}_h ϕ ∈ I h , let M ϕ ∈ M h ( k ) M_\phi\in M_h(k) M ϕ ∈ M h ( k ) be its companion matrix. Since h > 1 h>1 h > 1 , we have ϕ ( 0 ) ≠ 0 \phi(0)\ne0 ϕ ( 0 ) = 0 , so M ϕ M_\phi M ϕ is invertible. Via the central embeddings of k k k in E 1 , E 2 E_1,E_2 E 1 , E 2 , set
g ϕ = ( M ϕ , M ϕ ) ∈ Γ h . g_\phi=(M_\phi,M_\phi)\in\Gamma_h. g ϕ = ( M ϕ , M ϕ ) ∈ Γ h .
The minimal polynomial of M ϕ M_\phi M ϕ over k k k is ϕ \phi ϕ .
If u ∈ M h ( W ) u\in M_h(W) u ∈ M h ( W ) is fixed by g ϕ g_\phi g ϕ , then M ϕ u = u M ϕ M_\phi u=uM_\phi M ϕ u = u M ϕ , so ( M ϕ , M ϕ ) (M_\phi,M_\phi) ( M ϕ , M ϕ ) is an endomorphism of Z u Z_u Z u . Since ϕ ( M ϕ ) = 0 \phi(M_\phi)=0 ϕ ( M ϕ ) = 0 , this endomorphism is annihilated by ϕ \phi ϕ . Thus, by the definition of Ω h \Omega_h Ω h , every fixed point of g ϕ g_\phi g ϕ on M h ( W ) M_h(W) M h ( W ) belongs to Ω h \Omega_h Ω h . For g ∈ Γ h g\in\Gamma_h g ∈ Γ h , write Ω h g \Omega_h^g Ω h g for the fixed-point set of g g g in Ω h \Omega_h Ω h and C Γ h ( g ) C_{\Gamma_h}(g) C Γ h ( g ) for its centralizer in Γ h \Gamma_h Γ h .
We now compute ∣ Ω h g ϕ ∣ |\Omega_h^{g_\phi}| ∣ Ω h g ϕ ∣ . Fix ϕ ∈ I h \phi\in\mathcal{I}_h ϕ ∈ I h and set
C ϕ = { A ∈ M h ( k ) : M ϕ A = A M ϕ } , d ϕ = dim k C ϕ . C_\phi=\{A\in M_h(k):M_\phi A=AM_\phi\},\qquad d_\phi=\dim_k C_\phi. C ϕ = { A ∈ M h ( k ) : M ϕ A = A M ϕ } , d ϕ = dim k C ϕ .
Choose a k k k -basis f 1 , … , f w f_1,\ldots,f_w f 1 , … , f w of W W W . Expanding each entry of u = ( u i j ) ∈ M h ( W ) u=(u_{ij})\in M_h(W) u = ( u ij ) ∈ M h ( W ) in this basis gives the unique expression
u = ∑ ℓ = 1 w A ℓ f ℓ , A ℓ = ( a i j ( ℓ ) ) ∈ M h ( k ) , u i j = ∑ ℓ = 1 w a i j ( ℓ ) f ℓ , u=\sum_{\ell=1}^{w}A_\ell f_\ell,\qquad A_\ell=(a_{ij}^{(\ell)})\in M_h(k),\qquad u_{ij}=\sum_{\ell=1}^{w}a_{ij}^{(\ell)}f_\ell, u = ℓ = 1 ∑ w A ℓ f ℓ , A ℓ = ( a ij ( ℓ ) ) ∈ M h ( k ) , u ij = ℓ = 1 ∑ w a ij ( ℓ ) f ℓ ,
where A ℓ f ℓ A_\ell f_\ell A ℓ f ℓ denotes the matrix ( a i j ( ℓ ) f ℓ ) i j (a_{ij}^{(\ell)}f_\ell)_{ij} ( a ij ( ℓ ) f ℓ ) ij . Since the two scalar actions of k k k on W W W agree,
g ϕ ⋅ u = u ⟺ M ϕ u = u M ϕ ⟺ A ℓ ∈ C ϕ ( 1 ≤ ℓ ≤ w ) . g_\phi\cdot u=u\quad\Longleftrightarrow\quad M_\phi u=uM_\phi\quad\Longleftrightarrow\quad A_\ell\in C_\phi\quad(1\le\ell\le w). g ϕ ⋅ u = u ⟺ M ϕ u = u M ϕ ⟺ A ℓ ∈ C ϕ ( 1 ≤ ℓ ≤ w ) .
Since every fixed point of g ϕ g_\phi g ϕ on M h ( W ) M_h(W) M h ( W ) belongs to Ω h \Omega_h Ω h , the map
C ϕ ⊕ w ⟶ Ω h g ϕ , ( A 1 , … , A w ) ⟼ ∑ ℓ = 1 w A ℓ f ℓ C_\phi^{\oplus w}\longrightarrow\Omega_h^{g_\phi},\qquad(A_1,\ldots,A_w)\longmapsto\sum_{\ell=1}^{w}A_\ell f_\ell C ϕ ⊕ w ⟶ Ω h g ϕ , ( A 1 , … , A w ) ⟼ ℓ = 1 ∑ w A ℓ f ℓ
is a k k k -linear isomorphism. Hence dim k Ω h g ϕ = d ϕ w \dim_k\Omega_h^{g_\phi}=d_\phi w dim k Ω h g ϕ = d ϕ w .
For i = 1 , 2 i=1,2 i = 1 , 2 , let
C ϕ , i = { B ∈ M h ( E i ) : M ϕ B = B M ϕ } . C_{\phi,i}=\{B\in M_h(E_i):M_\phi B=BM_\phi\}. C ϕ , i = { B ∈ M h ( E i ) : M ϕ B = B M ϕ } .
Similarly, dim k C ϕ , i = d ϕ dim k E i \dim_k C_{\phi,i}=d_\phi\dim_k E_i dim k C ϕ , i = d ϕ dim k E i . A pair ( g 1 , g 2 ) ∈ Γ h (g_1,g_2)\in\Gamma_h ( g 1 , g 2 ) ∈ Γ h commutes with g ϕ g_\phi g ϕ if and only if g i M ϕ = M ϕ g i g_iM_\phi=M_\phi g_i g i M ϕ = M ϕ g i for i = 1 , 2 i=1,2 i = 1 , 2 . Hence
C Γ h ( g ϕ ) = ( C ϕ , 1 ∩ GL h ( E 1 ) ) × ( C ϕ , 2 ∩ GL h ( E 2 ) ) . C_{\Gamma_h}(g_\phi)=(C_{\phi,1}\cap\operatorname{GL}_h(E_1))\times(C_{\phi,2}\cap\operatorname{GL}_h(E_2)). C Γ h ( g ϕ ) = ( C ϕ , 1 ∩ GL h ( E 1 )) × ( C ϕ , 2 ∩ GL h ( E 2 )) .
Since ∣ k ∣ = p n |k|=p^n ∣ k ∣ = p n and dim k E 1 + dim k E 2 = c \dim_k E_1+\dim_k E_2=c dim k E 1 + dim k E 2 = c , we obtain
∣ Ω h g ϕ ∣ = p n d ϕ w , ∣ C Γ h ( g ϕ ) ∣ ≤ ∣ C ϕ , 1 ∣ ∣ C ϕ , 2 ∣ = p n d ϕ c . |\Omega_h^{g_\phi}|=p^{nd_\phi w},\qquad|C_{\Gamma_h}(g_\phi)|\le|C_{\phi,1}||C_{\phi,2}|=p^{nd_\phi c}. ∣ Ω h g ϕ ∣ = p n d ϕ w , ∣ C Γ h ( g ϕ ) ∣ ≤ ∣ C ϕ , 1 ∣∣ C ϕ , 2 ∣ = p n d ϕ c .
Since w ≥ c w\ge c w ≥ c , we have
∣ Ω h g ϕ ∣ ∣ C Γ h ( g ϕ ) ∣ ≥ p n d ϕ ( w − c ) ≥ 1. \frac{|\Omega_h^{g_\phi}|}{|C_{\Gamma_h}(g_\phi)|}\ge p^{nd_\phi(w-c)}\ge1. ∣ C Γ h ( g ϕ ) ∣ ∣ Ω h g ϕ ∣ ≥ p n d ϕ ( w − c ) ≥ 1.
If g ϕ g_{\phi} g ϕ and g ψ g_{\psi} g ψ are conjugate in Γ h \Gamma_h Γ h , then M ϕ M_{\phi} M ϕ and M ψ M_{\psi} M ψ are conjugate in GL h ( E 1 ) \operatorname{GL}_h(E_1) GL h ( E 1 ) . Since conjugation preserves their minimal polynomials over k k k , we have ϕ = ψ \phi=\psi ϕ = ψ .
Consider the action of Γ h \Gamma_h Γ h on Ω h \Omega_h Ω h . By the Cauchy–Frobenius formula [15 Theorem 3.22] , we have
β h = ∑ [ g ] ∣ Ω h g ∣ ∣ C Γ h ( g ) ∣ ≥ ∑ ϕ ∈ I h ∣ Ω h g ϕ ∣ ∣ C Γ h ( g ϕ ) ∣ ≥ ∣ I h ∣ = p n h − p n h . \beta_h=\sum_{[g]}\frac{|\Omega_h^g|}{|C_{\Gamma_h}(g)|}\geq\sum_{\phi\in I_h}\frac{|\Omega_h^{g_\phi}|}{|C_{\Gamma_h}(g_\phi)|}\geq|I_h|=\frac{p^{nh}-p^n}{h}. β h = [ g ] ∑ ∣ C Γ h ( g ) ∣ ∣ Ω h g ∣ ≥ ϕ ∈ I h ∑ ∣ C Γ h ( g ϕ ) ∣ ∣ Ω h g ϕ ∣ ≥ ∣ I h ∣ = h p nh − p n .
where the first sum is over the conjugacy classes of Γ h \Gamma_h Γ h . Together with (3.2), this yields
p n h − p n h ≤ ( h δ + 1 ) N ( r + s + 1 ) , \frac{p^{nh}-p^n}{h}\leq(h\delta+1)^{N(r+s+1)}, h p nh − p n ≤ ( h δ + 1 ) N ( r + s + 1 ) ,
which is impossible as h h h tends to infinity through the primes.
This contradiction proves that c > w c>w c > w in both cases. □
From positivity to finite type
We now use Simson’s criterion for finite prinjective type. We first recall it for triangular matrix algebras and then realize M ( G , F ) \mathcal{M}(G,F) M ( G , F ) as a prinjective category.
For division k k k -algebras D 1 D_1 D 1 , D 2 D_2 D 2 and a finite-dimensional D 1 D_1 D 1 –D 2 D_2 D 2 -bimodule V V V , we write dim D 1 V \dim_{D_1}V dim D 1 V for its dimension as a left D 1 D_1 D 1 -vector space and dim V D 2 \dim_{V D_2} dim V D 2 for its dimension as a right D 2 D_2 D 2 -vector space.
Let Λ 1 , Λ 2 \Lambda_1,\Lambda_2 Λ 1 , Λ 2 be nonzero basic finite-dimensional k k k -algebras, and let N N N be a finite-dimensional Λ 1 \Lambda_1 Λ 1 –Λ 2 \Lambda_2 Λ 2 -bimodule. Consider the triangular matrix algebra
Γ = ( Λ 1 N 0 Λ 2 ) . \Gamma=\begin{pmatrix}\Lambda_1&N\\0&\Lambda_2\end{pmatrix}. Γ = ( Λ 1 0 N Λ 2 ) .
Choose complete sets of primitive orthogonal idempotents e 1 , … , e r e_1,\ldots,e_r e 1 , … , e r of Λ 1 \Lambda_1 Λ 1 and η 1 , … , η s \eta_1,\ldots,\eta_s η 1 , … , η s of Λ 2 \Lambda_2 Λ 2 . In this subsection, unless explicitly stated otherwise, modules over Λ 1 \Lambda_1 Λ 1 , Λ 2 \Lambda_2 Λ 2 and Γ \Gamma Γ are right modules.
Following [16 p. 96] , we write prin ( Γ ) \operatorname{prin}(\Gamma) prin ( Γ ) for Simson’s category prin ( Γ ) Λ 2 Λ 1 \operatorname{prin}(\Gamma)_{\Lambda_2}^{\Lambda_1} prin ( Γ ) Λ 2 Λ 1 . This is the full subcategory of finitely generated right Γ \Gamma Γ -modules represented by triples ( P , I , φ ) (P,I,\varphi) ( P , I , φ ) , where P P P is a projective right Λ 1 \Lambda_1 Λ 1 -module, I I I is an injective right Λ 2 \Lambda_2 Λ 2 -module, and
φ : P ⊗ Λ 1 N ⟶ I \varphi:P\otimes_{\Lambda_1}N\longrightarrow I φ : P ⊗ Λ 1 N ⟶ I
is a homomorphism of right Λ 2 \Lambda_2 Λ 2 -modules.
The corresponding right action on P ⊕ I P\oplus I P ⊕ I is
( p , i ) ( a n 0 b ) = ( p a , φ ( p ⊗ n ) + i b ) . (p,i)\begin{pmatrix}a&n\\0&b\end{pmatrix}=(pa,\varphi(p\otimes n)+ib). ( p , i ) ( a 0 n b ) = ( p a , φ ( p ⊗ n ) + ib ) .
A morphism ( P , I , φ ) → ( P ′ , I ′ , φ ′ ) (P,I,\varphi)\to(P',I',\varphi') ( P , I , φ ) → ( P ′ , I ′ , φ ′ ) is a pair ( s , t ) (s,t) ( s , t ) consisting of a homomorphism s : P → P ′ s:P\to P' s : P → P ′ of right Λ 1 \Lambda_1 Λ 1 -modules and a homomorphism t : I → I ′ t:I\to I' t : I → I ′ of right Λ 2 \Lambda_2 Λ 2 -modules such that t ∘ φ = φ ′ ∘ ( s ⊗ 1 N ) t\circ\varphi=\varphi'\circ(s\otimes1_N) t ∘ φ = φ ′ ∘ ( s ⊗ 1 N ) . For n = ( n 1 , … , n r ) ∈ Z r \mathbf{n}=(n_1,\ldots,n_r)\in\mathbb{Z}^r n = ( n 1 , … , n r ) ∈ Z r and m = ( m 1 , … , m s ) ∈ Z s \mathbf{m}=(m_1,\ldots,m_s)\in\mathbb{Z}^s m = ( m 1 , … , m s ) ∈ Z s , define
q Γ ( n , m ) = ∑ i = 1 r ∑ i ′ = 1 r dim k ( e i Λ 1 e i ′ ) n i n i ′ + ∑ j = 1 s ∑ j ′ = 1 s dim k ( η j Λ 2 η j ′ ) m j m j ′ − ∑ i = 1 r ∑ j = 1 s dim k ( e i N η j ) n i m j . (4) q_{\Gamma}(\mathbf{n},\mathbf{m})=
\sum_{i=1}^{r}\sum_{i'=1}^{r}\dim_k(e_i\Lambda_1e_{i'})n_i n_{i'}
+\sum_{j=1}^{s}\sum_{j'=1}^{s}\dim_k(\eta_j\Lambda_2\eta_{j'})m_jm_{j'}
-\sum_{i=1}^{r}\sum_{j=1}^{s}\dim_k(e_iN\eta_j)n_i m_j. \tag*{(4)} q Γ ( n , m ) = i = 1 ∑ r i ′ = 1 ∑ r dim k ( e i Λ 1 e i ′ ) n i n i ′ + j = 1 ∑ s j ′ = 1 ∑ s dim k ( η j Λ 2 η j ′ ) m j m j ′ − i = 1 ∑ r j = 1 ∑ s dim k ( e i N η j ) n i m j . (4)
The form q Γ q_{\Gamma} q Γ is weakly positive if q Γ ( n , m ) > 0 q_{\Gamma}(\mathbf{n},\mathbf{m})>0 q Γ ( n , m ) > 0 for every nonzero ( n , m ) ∈ Z ≥ 0 r + s (\mathbf{n},\mathbf{m})\in\mathbb{Z}_{\geq0}^{r+s} ( n , m ) ∈ Z ≥ 0 r + s .
Proposition 3.7 (cf. [16 Theorem 1.6] ). Assume that (i) e i Λ 1 e i e_i\Lambda_1e_i e i Λ 1 e i and η j Λ 2 η j \eta_j\Lambda_2\eta_j η j Λ 2 η j are division rings for every i , j i,j i , j .
(ii) The left Λ 1 \Lambda_1 Λ 1 -module N η j N\eta_j N η j and the right Λ 2 \Lambda_2 Λ 2 -module e i N e_iN e i N are faithful for every i , j i,j i , j .
If q Γ q_\Gamma q Γ is weakly positive, then prin ( Γ ) \operatorname{prin}(\Gamma) prin ( Γ ) is representation-finite.
Proof. Conditions (i) and (ii) say precisely that Γ \Gamma Γ is piecewise peak in the sense of [16 p. 97] . The algebra Γ \Gamma Γ is Artinian because it is finite-dimensional over k k k . It also satisfies a polynomial identity: if d = dim k Γ d=\dim_k\Gamma d = dim k Γ , the nonzero noncommutative polynomial
s d + 1 ( x 1 , … , x d + 1 ) = ∑ σ ∈ S d + 1 sgn ( σ ) x σ ( 1 ) ⋯ x σ ( d + 1 ) s_{d+1}(x_1,\ldots,x_{d+1})=\sum_{\sigma\in S_{d+1}}\operatorname{sgn}(\sigma)x_{\sigma(1)}\cdots x_{\sigma(d+1)} s d + 1 ( x 1 , … , x d + 1 ) = σ ∈ S d + 1 ∑ sgn ( σ ) x σ ( 1 ) ⋯ x σ ( d + 1 )
vanishes on Γ \Gamma Γ because its evaluation is an alternating ( d + 1 ) (d+1) ( d + 1 ) -linear map on a d d d -dimensional vector space. Thus Γ \Gamma Γ is a basic Artinian bipartite piecewise peak PI-ring, as required in [16 Theorem 1.6] .
Write ( ε 1 , … , ε r + s ) = ( e 1 , … , e r , η 1 , … , η s ) (\varepsilon_1,\ldots,\varepsilon_{r+s})=(e_1,\ldots,e_r,\eta_1,\ldots,\eta_s) ( ε 1 , … , ε r + s ) = ( e 1 , … , e r , η 1 , … , η s ) and set f ℓ = dim k ( ε ℓ Γ ε ℓ ) f_\ell=\dim_k(\varepsilon_\ell\Gamma\varepsilon_\ell) f ℓ = dim k ( ε ℓ Γ ε ℓ ) . For ℓ ≠ m \ell\ne m ℓ = m , let
d ℓ m = dim ( ε m Γ ε m ) ( ε ℓ Γ ε m ) , d ℓ m ′ = dim ( ε ℓ Γ ε ℓ ) ( ε ℓ Γ ε m ) . d_{\ell m}=\dim_{(\varepsilon_m\Gamma\varepsilon_m)}(\varepsilon_\ell\Gamma\varepsilon_m),\qquad d'_{\ell m}=\dim_{(\varepsilon_\ell\Gamma\varepsilon_\ell)}(\varepsilon_\ell\Gamma\varepsilon_m). d ℓ m = dim ( ε m Γ ε m ) ( ε ℓ Γ ε m ) , d ℓ m ′ = dim ( ε ℓ Γ ε ℓ ) ( ε ℓ Γ ε m ) .
Then
dim k ( ε ℓ Γ ε m ) = f m d ℓ m = f ℓ d ℓ m ′ . \dim_k(\varepsilon_\ell\Gamma\varepsilon_m)=f_md_{\ell m}=f_\ell d'_{\ell m}. dim k ( ε ℓ Γ ε m ) = f m d ℓ m = f ℓ d ℓ m ′ .
In particular, f m d ℓ m = f ℓ d ℓ m ′ f_md_{\ell m}=f_\ell d'_{\ell m} f m d ℓ m = f ℓ d ℓ m ′ is the balancing relation for the valuation used by Simson. In his notation, ℓ ≺ m \ell\prec m ℓ ≺ m means that ℓ ≠ m \ell\ne m ℓ = m and ε ℓ Γ ε m ≠ 0 \varepsilon_\ell\Gamma\varepsilon_m\ne0 ε ℓ Γ ε m = 0 . Consequently, (4) has diagonal weights f ℓ f_\ell f ℓ and off-diagonal weights f m d ℓ m f_md_{\ell m} f m d ℓ m , with positive signs for blocks of Λ 1 \Lambda_1 Λ 1 and Λ 2 \Lambda_2 Λ 2 and negative signs for blocks of N N N . Thus (4) is precisely Simson’s weighted quadratic form (1.7), with the two positive coefficients corrected as explained in Remark 3.8. The implication ( c ) ⇒ ( a ) (c)\Rightarrow(a) ( c ) ⇒ ( a ) in [16 Theorem 1.6] proves the assertion. □
Remark 3.8. Formula (1.7) in [16 p. 99] contains a typographical error in its two positive off-diagonal sums. In Simson’s notation [16 (1.5), p. 98] , d i j d_{ij} d ij denotes the right dimension of ε i Γ ε j \varepsilon_i\Gamma\varepsilon_j ε i Γ ε j over ε j Γ ε j \varepsilon_j\Gamma\varepsilon_j ε j Γ ε j , while d i j ′ d'_{ij} d ij ′ is its left dimension over ε i Γ ε i \varepsilon_i\Gamma\varepsilon_i ε i Γ ε i . Hence
dim k ( ε i Γ ε j ) = f j d i j = f i d i j ′ . \dim_k(\varepsilon_i\Gamma\varepsilon_j)=f_jd_{ij}=f_id'_{ij}. dim k ( ε i Γ ε j ) = f j d ij = f i d ij ′ .
The coefficients printed as f i d i j f_id_{ij} f i d ij and f p d p q f_pd_{pq} f p d pq in the two positive sums of (1.7) must therefore be f j d i j f_jd_{ij} f j d ij and f q d p q f_qd_{pq} f q d pq , respectively. These coefficients agree with the Euler-form comparison in [16 p. 129] .
Lemma 3.9. Let X , Y , G , F , q \mathcal{X},\mathcal{Y},G,F,q X , Y , G , F , q be as in Definition 3.5, with indecomposable representatives X 1 , … , X r X_1,\ldots,X_r X 1 , … , X r and Y 1 , … , Y s Y_1,\ldots,Y_s Y 1 , … , Y s . Assume that
(1) End X ( X i ) \operatorname{End}_{\mathcal{X}}(X_i) End X ( X i ) and End Y ( Y j ) \operatorname{End}_{\mathcal{Y}}(Y_j) End Y ( Y j ) are division rings for every i , j i,j i , j .
(2) The quadratic form q q q is weakly positive.
Then M ( G , F ) \mathcal{M}(G,F) M ( G , F ) is representation-finite.
Proof. We write D k = Hom k ( − , k ) D_k=\operatorname{Hom}_k(-,k) D k = Hom k ( − , k ) and define
X 0 = ⨁ i = 1 r X i , Y 0 = ⨁ j = 1 s Y j , Λ 1 = End X ( X 0 ) , Λ 2 = End Y ( Y 0 ) , X_0=\bigoplus_{i=1}^{r}X_i,\qquad Y_0=\bigoplus_{j=1}^{s}Y_j,\qquad\Lambda_1=\operatorname{End}_{\mathcal{X}}(X_0),\qquad\Lambda_2=\operatorname{End}_{\mathcal{Y}}(Y_0), X 0 = i = 1 ⨁ r X i , Y 0 = j = 1 ⨁ s Y j , Λ 1 = End X ( X 0 ) , Λ 2 = End Y ( Y 0 ) ,
M = Hom D ( G ( X 0 ) , F ( Y 0 ) ) , N = D k ( M ) , Γ = ( Λ 1 N 0 Λ 2 ) . M=\operatorname{Hom}_{\mathcal{D}}(G(X_0),F(Y_0)),\qquad N=D_k(M),\qquad\Gamma=\begin{pmatrix}\Lambda_1 & N\\ 0 & \Lambda_2\end{pmatrix}. M = Hom D ( G ( X 0 ) , F ( Y 0 )) , N = D k ( M ) , Γ = ( Λ 1 0 N Λ 2 ) .
The Λ 2 \Lambda_2 Λ 2 –Λ 1 \Lambda_1 Λ 1 -bimodule structure on M M M is given by
b ⋅ u ⋅ a = F ( b ) ∘ u ∘ G ( a ) ( a ∈ Λ 1 , b ∈ Λ 2 , u ∈ M ) . b\cdot u\cdot a=F(b)\circ u\circ G(a)\qquad(a\in\Lambda_1,\ b\in\Lambda_2,\ u\in M). b ⋅ u ⋅ a = F ( b ) ∘ u ∘ G ( a ) ( a ∈ Λ 1 , b ∈ Λ 2 , u ∈ M ) .
Thus N N N is a Λ 1 \Lambda_1 Λ 1 –Λ 2 \Lambda_2 Λ 2 -bimodule. Let e i ∈ Λ 1 e_i\in\Lambda_1 e i ∈ Λ 1 and η j ∈ Λ 2 \eta_j\in\Lambda_2 η j ∈ Λ 2 be the primitive idempotents corresponding to X i X_i X i and Y j Y_j Y j .
We now construct an equivalence
Φ : M ( G , F ) ⟶ ∼ prin ( Γ ) . (5) \Phi:\mathcal{M}(G,F)\overset{\sim}{\longrightarrow}\operatorname{prin}(\Gamma). \tag*{(5)} Φ : M ( G , F ) ⟶ ∼ prin ( Γ ) . (5)
For X ∈ X X\in\mathcal{X} X ∈ X and Y ∈ Y Y\in\mathcal{Y} Y ∈ Y , we define
P X = Hom X ( X 0 , X ) , I Y = D k ( Hom Y ( Y , Y 0 ) ) . P_X=\operatorname{Hom}_{\mathcal{X}}(X_0,X),\qquad I_Y=D_k(\operatorname{Hom}_{\mathcal{Y}}(Y,Y_0)). P X = Hom X ( X 0 , X ) , I Y = D k ( Hom Y ( Y , Y 0 )) .
For morphisms f : X → X ′ f:X\to X' f : X → X ′ and g : Y → Y ′ g:Y\to Y' g : Y → Y ′ , we define
P ( f ) ( p ) = f ∘ p , I ( g ) ( ξ ) ( h ′ ) = ξ ( h ′ ∘ g ) P(f)(p)=f\circ p,\qquad I(g)(\xi)(h')=\xi(h'\circ g) P ( f ) ( p ) = f ∘ p , I ( g ) ( ξ ) ( h ′ ) = ξ ( h ′ ∘ g )
for p ∈ P X p\in P_X p ∈ P X , ξ ∈ I Y \xi\in I_Y ξ ∈ I Y and h ′ ∈ Hom Y ( Y ′ , Y 0 ) h'\in\operatorname{Hom}_{\mathcal{Y}}(Y',Y_0) h ′ ∈ Hom Y ( Y ′ , Y 0 ) . These functors give equivalences from X \mathcal{X} X to the finitely generated projective right Λ 1 \Lambda_1 Λ 1 -modules and from Y \mathcal{Y} Y to the finitely generated injective right Λ 2 \Lambda_2 Λ 2 -modules.
For u : G ( X ) → F ( Y ) u:G(X)\to F(Y) u : G ( X ) → F ( Y ) , we define φ u : P X ⊗ Λ 1 N → I Y \varphi_u:P_X\otimes_{\Lambda_1}N\to I_Y φ u : P X ⊗ Λ 1 N → I Y by
φ u ( p ⊗ λ ) ( h ) = λ ( F ( h ) ∘ u ∘ G ( p ) ) (6) \varphi_u(p\otimes\lambda)(h)=\lambda(F(h)\circ u\circ G(p)) \tag*{(6)} φ u ( p ⊗ λ ) ( h ) = λ ( F ( h ) ∘ u ∘ G ( p )) (6)
for p ∈ P X p\in P_X p ∈ P X , λ ∈ N \lambda\in N λ ∈ N and h ∈ Hom Y ( Y , Y 0 ) h\in\operatorname{Hom}_{\mathcal{Y}}(Y,Y_0) h ∈ Hom Y ( Y , Y 0 ) . This formula is balanced over Λ 1 \Lambda_1 Λ 1 and defines a homomorphism of right Λ 2 \Lambda_2 Λ 2 -modules. We define Φ \Phi Φ on objects and morphisms by
Φ ( X , Y , u ) = ( P X , I Y , φ u ) , Φ ( f , g ) = ( P ( f ) , I ( g ) ) . \Phi(X,Y,u)=(P_X,I_Y,\varphi_u),\qquad\Phi(f,g)=(P(f),I(g)). Φ ( X , Y , u ) = ( P X , I Y , φ u ) , Φ ( f , g ) = ( P ( f ) , I ( g )) .
By (6) , for f : X ′ → X f:X'\to X f : X ′ → X and g : Y → Y ′ g:Y\to Y' g : Y → Y ′ , we have
φ F ( g ) ∘ u ∘ G ( f ) = I ( g ) ∘ φ u ∘ ( P ( f ) ⊗ 1 N ) . \varphi_{F(g)\circ u\circ G(f)}=I(g)\circ\varphi_u\circ(P(f)\otimes1_N). φ F ( g ) ∘ u ∘ G ( f ) = I ( g ) ∘ φ u ∘ ( P ( f ) ⊗ 1 N ) .
The displayed identity gives naturality and shows that Φ \Phi Φ is well defined on morphisms. Functoriality follows from that of P P P and I I I .
For X = X 0 X=X_0 X = X 0 and Y = Y 0 Y=Y_0 Y = Y 0 , we have P X 0 = Λ 1 P_{X_0}=\Lambda_1 P X 0 = Λ 1 and I Y 0 = D k ( Λ 2 ) I_{Y_0}=D_k(\Lambda_2) I Y 0 = D k ( Λ 2 ) . Under Λ 1 ⊗ Λ 1 N ≅ N \Lambda_1\otimes_{\Lambda_1}N\cong N Λ 1 ⊗ Λ 1 N ≅ N , the standard tensor–dual identifications are
Hom Λ 2 ( N , D k ( Λ 2 ) ) ≅ D k ( N ⊗ Λ 2 Λ 2 ) ≅ D k ( N ) ≅ M . \operatorname{Hom}_{\Lambda_2}(N,D_k(\Lambda_2))\cong D_k(N\otimes_{\Lambda_2}\Lambda_2)\cong D_k(N)\cong M. Hom Λ 2 ( N , D k ( Λ 2 )) ≅ D k ( N ⊗ Λ 2 Λ 2 ) ≅ D k ( N ) ≅ M .
They send ψ \psi ψ to the element of M M M corresponding to the functional λ ↦ ψ ( λ ) ( 1 Λ 2 ) \lambda\mapsto\psi(\lambda)(1_{\Lambda_2}) λ ↦ ψ ( λ ) ( 1 Λ 2 ) . Since φ u ( λ ) ( 1 Λ 2 ) = λ ( u ) \varphi_u(\lambda)(1_{\Lambda_2})=\lambda(u) φ u ( λ ) ( 1 Λ 2 ) = λ ( u ) , they send φ u \varphi_u φ u to u u u . Every object of X \mathcal{X} X is a direct summand of a finite direct sum of copies of X 0 X_0 X 0 , and similarly for Y \mathcal{Y} Y and Y 0 Y_0 Y 0 . Additivity and naturality therefore give a natural isomorphism
Hom Λ 2 ( P X ⊗ Λ 1 N , I Y ) ≅ Hom D ( G ( X ) , F ( Y ) ) , φ u ⟼ u . (7) \operatorname{Hom}_{\Lambda_2}(P_X\otimes_{\Lambda_1}N,I_Y)\cong\operatorname{Hom}_{D}(G(X),F(Y)),\qquad\varphi_u\longmapsto u. \tag*{(7)} Hom Λ 2 ( P X ⊗ Λ 1 N , I Y ) ≅ Hom D ( G ( X ) , F ( Y )) , φ u ⟼ u . (7)
For objects ( X , Y , u ) (X,Y,u) ( X , Y , u ) and ( X ′ , Y ′ , u ′ ) (X',Y',u') ( X ′ , Y ′ , u ′ ) and morphisms f : X → X ′ f:X\to X' f : X → X ′ and g : Y → Y ′ g:Y\to Y' g : Y → Y ′ , naturality and (7) give
F ( g ) ∘ u = u ′ ∘ G ( f ) ⟺ I ( g ) ∘ φ u = φ u ′ ∘ ( P ( f ) ⊗ 1 N ) . F(g)\circ u=u'\circ G(f)\Longleftrightarrow I(g)\circ\varphi_u=\varphi_{u'}\circ(P(f)\otimes1_N). F ( g ) ∘ u = u ′ ∘ G ( f ) ⟺ I ( g ) ∘ φ u = φ u ′ ∘ ( P ( f ) ⊗ 1 N ) .
Since P P P and I I I are fully faithful, Φ \Phi Φ is fully faithful. Every object of prin ( Γ ) \operatorname{prin}(\Gamma) prin ( Γ ) is isomorphic to one of the form ( P X , I Y , ψ ) (P_X,I_Y,\psi) ( P X , I Y , ψ ) , and (7) gives a unique u u u with ψ = φ u \psi=\varphi_u ψ = φ u . Thus Φ \Phi Φ is essentially surjective, and hence an equivalence.
The algebras Λ 1 \Lambda_1 Λ 1 and Λ 2 \Lambda_2 Λ 2 are basic and finite-dimensional over k k k . Their primitive diagonal corner rings are End X ( X i ) \operatorname{End}_{\mathcal{X}}(X_i) End X ( X i ) and End Y ( Y j ) \operatorname{End}_{\mathcal{Y}}(Y_j) End Y ( Y j ) , which are division rings by hypothesis.
For each j j j , the right Λ 1 \Lambda_1 Λ 1 -module
η j M = Hom D ( G ( X 0 ) , F ( Y j ) ) \eta_jM=\operatorname{Hom}_{D}(G(X_0),F(Y_j)) η j M = Hom D ( G ( X 0 ) , F ( Y j ))
is faithful. In fact, if a ∈ Λ 1 a\in\Lambda_1 a ∈ Λ 1 annihilates all these maps, then u ∘ G ( a ) = 0 u\circ G(a)=0 u ∘ G ( a ) = 0 for every D D D -linear map u : G ( X 0 ) → F ( Y j ) u:G(X_0)\to F(Y_j) u : G ( X 0 ) → F ( Y j ) . If G ( a ) ( x ) ≠ 0 G(a)(x)\ne0 G ( a ) ( x ) = 0 , extend this vector to a D D D -basis of G ( X 0 ) G(X_0) G ( X 0 ) and choose u u u sending it to a nonzero vector of F ( Y j ) F(Y_j) F ( Y j ) . This contradicts u ∘ G ( a ) = 0 u\circ G(a)=0 u ∘ G ( a ) = 0 . Hence G ( a ) = 0 G(a)=0 G ( a ) = 0 , and faithfulness of G G G gives a = 0 a=0 a = 0 . Similarly, each left Λ 2 \Lambda_2 Λ 2 -module
M η i = Hom D ( G ( X i ) , F ( Y 0 ) ) M\eta_i=\operatorname{Hom}_{D}(G(X_i),F(Y_0)) M η i = Hom D ( G ( X i ) , F ( Y 0 ))
is faithful. Duality therefore makes the left Λ 1 \Lambda_1 Λ 1 -module N η j N\eta_j N η j and the right Λ 2 \Lambda_2 Λ 2 -module e i N e_iN e i N faithful. Thus Γ \Gamma Γ satisfies conditions (i) and (ii) of Proposition 3.7.
By the choice of e i e_i e i and η j \eta_j η j as the projections onto X i X_i X i and Y j Y_j Y j , respectively, we have
e i Λ 1 e i ′ ′ ≅ Hom X ( X i ′ ′ , X i ) , η j Λ 2 η j ′ ′ ≅ Hom Y ( Y j ′ ′ , Y j ) . e_i\Lambda_1e_{i'}' \cong\operatorname{Hom}_{\mathcal{X}}(X_{i'}',X_i), \qquad\eta_j\Lambda_2\eta_{j'}' \cong\operatorname{Hom}_{\mathcal{Y}}(Y_{j'}',Y_j). e i Λ 1 e i ′ ′ ≅ Hom X ( X i ′ ′ , X i ) , η j Λ 2 η j ′ ′ ≅ Hom Y ( Y j ′ ′ , Y j ) .
For X = ⨁ i = 1 r X i n i X=\bigoplus_{i=1}^{r}X_i^{n_i} X = ⨁ i = 1 r X i n i and Y = ⨁ j = 1 s Y j m j Y=\bigoplus_{j=1}^{s}Y_j^{m_j} Y = ⨁ j = 1 s Y j m j , we have
dim k End X ( X ) = ∑ i = 1 r ∑ i ′ = 1 r dim k ( e i Λ 1 e i ′ ′ ) n i n i ′ , \dim_k\operatorname{End}_{\mathcal{X}}(X)=\sum_{i=1}^{r}\sum_{i'=1}^{r}\dim_k(e_i\Lambda_1e_{i'}')n_in_{i'}, dim k End X ( X ) = i = 1 ∑ r i ′ = 1 ∑ r dim k ( e i Λ 1 e i ′ ′ ) n i n i ′ ,
dim k End Y ( Y ) = ∑ j = 1 s ∑ j ′ = 1 s dim k ( η j Λ 2 η j ′ ′ ) m j m j ′ . \dim_k\operatorname{End}_{\mathcal{Y}}(Y)=\sum_{j=1}^{s}\sum_{j'=1}^{s}\dim_k(\eta_j\Lambda_2\eta_{j'}')m_jm_{j'}. dim k End Y ( Y ) = j = 1 ∑ s j ′ = 1 ∑ s dim k ( η j Λ 2 η j ′ ′ ) m j m j ′ .
Moreover, η j M e i ≅ Hom D ( G ( X i ) , F ( Y j ) ) \eta_jMe_i\cong\operatorname{Hom}_{D}(G(X_i),F(Y_j)) η j M e i ≅ Hom D ( G ( X i ) , F ( Y j )) and e i N η j ≅ D k ( η j M e i ) e_iN\eta_j\cong D_k(\eta_jMe_i) e i N η j ≅ D k ( η j M e i ) , so we have
dim k Hom D ( G ( X ) , F ( Y ) ) = ∑ i = 1 r ∑ j = 1 s dim k Hom D ( G ( X i ) , F ( Y j ) ) n i m j = ∑ i = 1 r ∑ j = 1 s dim k ( e i N η j ) n i m j . \begin{aligned}
\dim_k\operatorname{Hom}_{D}(G(X),F(Y))=\sum_{i=1}^{r}\sum_{j=1}^{s}\dim_k\operatorname{Hom}_{D}(G(X_i),F(Y_j))n_im_j \\
&=\sum_{i=1}^{r}\sum_{j=1}^{s}\dim_k(e_iN\eta_j)n_im_j.
\end{aligned} dim k Hom D ( G ( X ) , F ( Y )) = i = 1 ∑ r j = 1 ∑ s dim k Hom D ( G ( X i ) , F ( Y j )) n i m j = i = 1 ∑ r j = 1 ∑ s dim k ( e i N η j ) n i m j .
By (1) and (4) , we have q ( X , Y ) = q Γ ( n , m ) q(X,Y)=q_{\Gamma}(\mathbf{n},\mathbf{m}) q ( X , Y ) = q Γ ( n , m ) . Hence q Γ q_{\Gamma} q Γ is weakly positive. By Proposition 3.7, prin ( Γ ) \operatorname{prin}(\Gamma) prin ( Γ ) is representation-finite, and the equivalence (5) proves the assertion. □
We can now prove Theorem 3.3 .
Proof of Theorem 3.3 . Let X ′ \mathcal{X}' X ′ and Y ′ \mathcal{Y}' Y ′ be the full additive subcategories generated by indecomposable X X X with G ( X ) ≠ 0 G(X)\ne0 G ( X ) = 0 and Y Y Y with F ( Y ) ≠ 0 F(Y)\ne0 F ( Y ) = 0 , respectively. For Z = ( X , Y , u ) ∈ M ( G , F ) Z=(X,Y,u)\in\mathcal{M}(G,F) Z = ( X , Y , u ) ∈ M ( G , F ) , write X ≅ X 0 ⊕ X 1 X\cong X_0\oplus X_1 X ≅ X 0 ⊕ X 1 and Y ≅ Y 0 ⊕ Y 1 Y\cong Y_0\oplus Y_1 Y ≅ Y 0 ⊕ Y 1 , where G ( X 0 ) = F ( Y 0 ) = 0 G(X_0)=F(Y_0)=0 G ( X 0 ) = F ( Y 0 ) = 0 , X 1 ∈ X ′ X_1\in\mathcal{X}' X 1 ∈ X ′ and Y 1 ∈ Y ′ Y_1\in\mathcal{Y}' Y 1 ∈ Y ′ . So we have
Z ≅ ( X 0 , 0 , 0 ) ⊕ ( 0 , Y 0 , 0 ) ⊕ ( X 1 , Y 1 , u 1 ) , Z\cong(X_0,0,0)\oplus(0,Y_0,0)\oplus(X_1,Y_1,u_1), Z ≅ ( X 0 , 0 , 0 ) ⊕ ( 0 , Y 0 , 0 ) ⊕ ( X 1 , Y 1 , u 1 ) ,
where u 1 : G ( X 1 ) → F ( Y 1 ) u_1:G(X_1)\to F(Y_1) u 1 : G ( X 1 ) → F ( Y 1 ) is induced by u u u . Since X \mathcal{X} X and Y \mathcal{Y} Y are representation-finite, there are only finitely many indecomposables of the forms ( X , 0 , 0 ) (X,0,0) ( X , 0 , 0 ) and ( 0 , Y , 0 ) (0,Y,0) ( 0 , Y , 0 ) up to isomorphism. So it suffices to prove that the full subcategory M ( G ∣ X ′ , F ∣ Y ′ ) \mathcal{M}(G|_{\mathcal{X}'},F|_{\mathcal{Y}'}) M ( G ∣ X ′ , F ∣ Y ′ ) is representation-finite. Since this subcategory satisfies the same hypothesis, we may replace X , Y \mathcal{X},\mathcal{Y} X , Y by X ′ , Y ′ \mathcal{X}',\mathcal{Y}' X ′ , Y ′ and assume that G , F G,F G , F are faithful on objects. The conclusion is immediate if X = 0 \mathcal{X}=0 X = 0 or Y = 0 \mathcal{Y}=0 Y = 0 , so we assume that both are nonzero.
Apply Lemma 3.4 to both functors and set
X ‾ = X / ker G , Y ‾ = Y / ker F . \overline{\mathcal{X}}=\mathcal{X}/\ker G,\qquad\overline{\mathcal{Y}}=\mathcal{Y}/\ker F. X = X / ker G , Y = Y / ker F .
The induced functors G ‾ , F ‾ \overline{G},\overline{F} G , F are faithful, and both quotient categories remain nonzero, Hom-finite, Krull–Schmidt and representation-finite. We now show that any two indecomposable objects of M ( G ‾ , F ‾ ) \mathcal{M}(\overline{G},\overline{F}) M ( G , F ) with isomorphic components are isomorphic. Choose objects in M ( G , F ) \mathcal{M}(G,F) M ( G , F ) lifting the two given indecomposable objects. By Lemma 3.4 , these lifts are indecomposable, and their components are isomorphic because the component quotient functors induce bijections on isomorphism classes. The hypothesis makes the lifts isomorphic, so their images are isomorphic as well.
Set k = R / ker ( R → D ) k=R/\ker(R\to D) k = R / ker ( R → D ) . By the observation preceding Definition 3.5 , D D D is finite-dimensional over k k k , and the categories X ‾ , Y ‾ \overline{\mathcal{X}},\overline{\mathcal{Y}} X , Y and functors G ‾ , F ‾ \overline{G},\overline{F} G , F satisfy the assumptions of that definition. By Proposition 3.6 and Lemma 3.9 , M ( G ‾ , F ‾ ) \mathcal{M}(\overline{G},\overline{F}) M ( G , F ) is representation-finite. By Lemma 3.4 , M ( G , F ) \mathcal{M}(G,F) M ( G , F ) is also representation-finite. □
Factorization categories
Factorizations of a natural transformation
We use the factorization construction of MacPherson and Vilonen [13 §1] in the following additive setting.
Definition 4.1. Let A A A be a local Artin R R R -algebra and put D = A / J ( A ) D=A/J(A) D = A / J ( A ) . For a Hom-finite Krull–Schmidt category C \mathcal{C} C , fix additive functors
G , F : C ⟶ mod D G,F:\mathcal{C}\longrightarrow\operatorname{mod}D G , F : C ⟶ mod D and a natural transformation μ : G → F \mu:G\to F μ : G → F . We regard D D D -modules as semisimple A A A -modules through A → D A\to D A → D . The factorization category F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) is defined as follows.
Objects are tuples ( N , X , β , α ) (N,X,\beta,\alpha) ( N , X , β , α ) with N ∈ C N\in\mathcal{C} N ∈ C , X ∈ mod A X\in\operatorname{mod}A X ∈ mod A and homomorphisms of A A A -modules
G ( N ) → β X → α F ( N ) G(N)\xrightarrow{\ \beta\ }X\xrightarrow{\ \alpha\ }F(N) G ( N ) β X α F ( N ) such that α ∘ β = μ N \alpha\circ\beta=\mu_N α ∘ β = μ N .
A morphism ( N , X , β , α ) → ( N ′ , X ′ , β ′ , α ′ ) (N,X,\beta,\alpha)\to(N',X',\beta',\alpha') ( N , X , β , α ) → ( N ′ , X ′ , β ′ , α ′ ) is a pair ( f , g ) (f,g) ( f , g ) with f : N → N ′ f:N\to N' f : N → N ′ in C \mathcal{C} C and g : X → X ′ g:X\to X' g : X → X ′ in mod A \operatorname{mod}A mod A for which the following diagram commutes:
G ( N ) → β X → α F ( N ) G ( f ) ↓ g ↓ F ( f ) ↓ G ( N ′ ) → β ′ X ′ → α ′ F ( N ′ ) \begin{CD}
G(N) @>{\beta}>> X @>{\alpha}>> F(N) \\
@V{G(f)}VV @V{g}VV @V{F(f)}VV \\
G(N') @>{\beta'}>> X' @>{\alpha'}>> F(N')
\end{CD} G ( N ) G ( f ) ↓ ⏐ G ( N ′ ) β β ′ X g ↓ ⏐ X ′ α α ′ F ( N ) F ( f ) ↓ ⏐ F ( N ′ ) Identities and composition are defined componentwise.
Lemma 4.3. The category F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) is a Hom-finite, idempotent-complete R R R -linear additive category, with finite direct sums formed componentwise. In particular, it is Krull–Schmidt.
Proof. Under the canonical identifications supplied by additivity of G , F G,F G , F , naturality gives μ N 1 ⊕ N 2 = μ N 1 ⊕ μ N 2 \mu_{N_1\oplus N_2}=\mu_{N_1}\oplus\mu_{N_2} μ N 1 ⊕ N 2 = μ N 1 ⊕ μ N 2 . Hence componentwise direct sums, injections and projections make F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) an additive category.
Let ( f , g ) (f,g) ( f , g ) be an idempotent endomorphism of Z = ( N , X , β , α ) Z=(N,X,\beta,\alpha) Z = ( N , X , β , α ) . Split f f f and g g g in C \mathcal{C} C and mod A \operatorname{mod}A mod A , respectively, and write
N ≅ N 1 ⊕ N 2 , X ≅ X 1 ⊕ X 2 , f = 1 N 1 ⊕ 0 , g = 1 X 1 ⊕ 0. N\cong N_1\oplus N_2,\qquad X\cong X_1\oplus X_2,\qquad f=1_{N_1}\oplus0,\qquad g=1_{X_1}\oplus0. N ≅ N 1 ⊕ N 2 , X ≅ X 1 ⊕ X 2 , f = 1 N 1 ⊕ 0 , g = 1 X 1 ⊕ 0. Since ( f , g ) (f,g) ( f , g ) is an endomorphism of Z Z Z , we have g ∘ β = β ∘ G ( f ) g\circ\beta=\beta\circ G(f) g ∘ β = β ∘ G ( f ) and F ( f ) ∘ α = α ∘ g F(f)\circ\alpha=\alpha\circ g F ( f ) ∘ α = α ∘ g . These relations make β \beta β and α \alpha α block diagonal in the chosen decompositions. Their diagonal blocks satisfy
α i ∘ β i = μ N i ( i = 1 , 2 ) \alpha_i\circ\beta_i=\mu_{N_i}\qquad(i=1,2) α i ∘ β i = μ N i ( i = 1 , 2 ) by naturality of μ \mu μ . They therefore define objects Z i = ( N i , X i , β i , α i ) Z_i=(N_i,X_i,\beta_i,\alpha_i) Z i = ( N i , X i , β i , α i ) with Z ≅ Z 1 ⊕ Z 2 Z\cong Z_1\oplus Z_2 Z ≅ Z 1 ⊕ Z 2 , and ( f , g ) (f,g) ( f , g ) is the projection onto Z 1 Z_1 Z 1 . Thus F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) is idempotent complete.
For Z = ( N , X , β , α ) Z=(N,X,\beta,\alpha) Z = ( N , X , β , α ) and Z ′ = ( N ′ , X ′ , β ′ , α ′ ) Z'=(N',X',\beta',\alpha') Z ′ = ( N ′ , X ′ , β ′ , α ′ ) , the space Hom F A ( μ ; G , F ) ( Z , Z ′ ) \operatorname{Hom}_{\mathcal{F}_A(\mu;G,F)}(Z,Z') Hom F A ( μ ; G , F ) ( Z , Z ′ ) is an R R R -submodule of Hom C ( N , N ′ ) ⊕ Hom A ( X , X ′ ) \operatorname{Hom}_{\mathcal{C}}(N,N')\oplus\operatorname{Hom}_{A}(X,X') Hom C ( N , N ′ ) ⊕ Hom A ( X , X ′ ) , so F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) is Hom-finite. Since it is also idempotent complete, it is Krull–Schmidt by [10 Corollary 4.4] . □
Reduction and finiteness
Throughout this subsection, let A A A be a local Artin R R R -algebra and write D = A / J ( A ) D=A/J(A) D = A / J ( A ) . Let μ : G → F \mu:G\to F μ : G → F be a natural transformation between additive functors G , F : C → mod D G,F:\mathcal{C}\to\operatorname{mod}D G , F : C → mod D , where C \mathcal{C} C is Hom-finite and Krull–Schmidt. We work with the category F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) .
We consider the kernel, image and cokernel functors
K , I , C : C ⟶ mod D , K,I,C:\mathcal{C}\longrightarrow\operatorname{mod}D, K , I , C : C ⟶ mod D ,
given by
K ( N ) = ker μ N , I ( N ) = im μ N , C ( N ) = coker μ N . K(N)=\ker\mu_N,\qquad I(N)=\operatorname{im}\mu_N,\qquad C(N)=\operatorname{coker}\mu_N. K ( N ) = ker μ N , I ( N ) = im μ N , C ( N ) = coker μ N .
Naturality of μ \mu μ induces the maps K ( f ) K(f) K ( f ) , I ( f ) I(f) I ( f ) and C ( f ) C(f) C ( f ) for every f : N → N ′ f:N\to N' f : N → N ′ . Since every D D D -module is semisimple, we choose decompositions of G ( N ) G(N) G ( N ) and F ( N ) F(N) F ( N ) for each N N N in which μ N \mu_N μ N has the form
G ( N ) = K ( N ) ⊕ I ( N ) → μ N = ( 0 1 0 0 ) I ( N ) ⊕ C ( N ) = F ( N ) . G(N)=K(N)\oplus I(N)\xrightarrow{\ \mu_N=\begin{pmatrix}0&1\\0&0\end{pmatrix}\ }I(N)\oplus C(N)=F(N). G ( N ) = K ( N ) ⊕ I ( N ) μ N = ( 0 0 1 0 ) I ( N ) ⊕ C ( N ) = F ( N ) .
For an object Z = ( N , X , β , α ) Z=(N,X,\beta,\alpha) Z = ( N , X , β , α ) of F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) , set
ι = β ∣ I ( N ) , p = pr I ( N ) ∘ α : X ⟶ I ( N ) , \iota=\beta|_{I(N)},\qquad p=\operatorname{pr}_{I(N)}\circ\alpha:X\longrightarrow I(N), ι = β ∣ I ( N ) , p = pr I ( N ) ∘ α : X ⟶ I ( N ) ,
where pr I ( N ) : I ( N ) ⊕ C ( N ) → I ( N ) \operatorname{pr}_{I(N)}:I(N)\oplus C(N)\to I(N) pr I ( N ) : I ( N ) ⊕ C ( N ) → I ( N ) is the projection onto the first summand. The relation α ∘ β = μ N \alpha\circ\beta=\mu_N α ∘ β = μ N gives p ∘ ι = 1 I ( N ) p\circ\iota=1_{I(N)} p ∘ ι = 1 I ( N ) . Taking Y = ker p Y=\ker p Y = ker p , we obtain a decomposition
X ≅ I ( N ) ⊕ Y . X\cong I(N)\oplus Y. X ≅ I ( N ) ⊕ Y .
In these coordinates, the maps take the form
β = ( 0 1 b 0 ) : K ( N ) ⊕ I ( N ) ⟶ I ( N ) ⊕ Y , \beta=\begin{pmatrix}0&1\\b&0\end{pmatrix}:K(N)\oplus I(N)\longrightarrow I(N)\oplus Y, β = ( 0 b 1 0 ) : K ( N ) ⊕ I ( N ) ⟶ I ( N ) ⊕ Y ,
α = ( 1 0 0 a ) : I ( N ) ⊕ Y ⟶ I ( N ) ⊕ C ( N ) , (8) \alpha=\begin{pmatrix}1&0\\0&a\end{pmatrix}:I(N)\oplus Y\longrightarrow I(N)\oplus C(N), \tag*{(8)} α = ( 1 0 0 a ) : I ( N ) ⊕ Y ⟶ I ( N ) ⊕ C ( N ) , (8)
with a ∘ b = 0 a\circ b=0 a ∘ b = 0 .
Definition 4.4. The category F A ( 0 ; K , C ) \mathcal{F}_A(0;K,C) F A ( 0 ; K , C ) is called the reduced category of F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) . The object
( N , Y , b , a ) ∈ F A ( 0 ; K , C ) (N,Y,b,a)\in\mathcal{F}_A(0;K,C) ( N , Y , b , a ) ∈ F A ( 0 ; K , C )
constructed above is called a reduced object associated with Z = ( N , X , β , α ) Z=(N,X,\beta,\alpha) Z = ( N , X , β , α ) .
Conversely, for ( N , Y , b , a ) ∈ F A ( 0 ; K , C ) (N,Y,b,a)\in\mathcal{F}_A(0;K,C) ( N , Y , b , a ) ∈ F A ( 0 ; K , C ) , set X = I ( N ) ⊕ Y X=I(N)\oplus Y X = I ( N ) ⊕ Y and define β , α \beta,\alpha β , α by (8) . Their composite is μ N \mu_N μ N because a ∘ b = 0 a\circ b=0 a ∘ b = 0 .
Lemma 4.5. The construction above induces a bijection between the isomorphism classes of objects of F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) and those of F A ( 0 ; K , C ) \mathcal{F}_A(0;K,C) F A ( 0 ; K , C ) , given by
[ ( N , X , β , α ) ] ⟼ [ ( N , Y , b , a ) ] . [(N,X,\beta,\alpha)]\longmapsto[(N,Y,b,a)]. [( N , X , β , α )] ⟼ [( N , Y , b , a )] .
This bijection preserves direct sums and reflects zero objects. In particular, it restricts to a bijection between the isomorphism classes of indecomposable objects. Corresponding objects have the same C C C -component and satisfy
X ≅ I ( N ) ⊕ Y . X\cong I(N)\oplus Y. X ≅ I ( N ) ⊕ Y .
Proof. Fix any choices of splittings for two objects and write them in the coordinates of (8) . Let ( f , g ) (f,g) ( f , g ) be a morphism between them. In the chosen decompositions the naturality equation F ( f ) ∘ μ N = μ N ′ ∘ G ( f ) F(f)\circ\mu_N=\mu_{N'}\circ G(f) F ( f ) ∘ μ N = μ N ′ ∘ G ( f ) forces the lower-left blocks of G ( f ) G(f) G ( f ) and F ( f ) F(f) F ( f ) to vanish and their blocks on I ( N ) I(N) I ( N ) to agree. The conditions g ∘ β = β ′ ∘ G ( f ) g\circ\beta=\beta'\circ G(f) g ∘ β = β ′ ∘ G ( f ) and F ( f ) ∘ α = α ′ ∘ g F(f)\circ\alpha=\alpha'\circ g F ( f ) ∘ α = α ′ ∘ g are expressed by the following commutative diagram:
K ( N ) ⊕ I ( N ) → ( 0 1 b 0 ) I ( N ) ⊕ Y → ( 1 0 0 a ) I ( N ) ⊕ C ( N ) G ( f ) = ( k u 0 i ) ↓ g ↓ F ( f ) = ( i v 0 c ) ↓ K ( N ′ ) ⊕ I ( N ′ ) → ( 0 1 b ′ 0 ) I ( N ′ ) ⊕ Y ′ → ( 1 0 0 a ′ ) I ( N ′ ) ⊕ C ( N ′ ) \begin{CD}
K(N)\oplus I(N) @>{\left(\begin{smallmatrix}0&1\\b&0\end{smallmatrix}\right)}>> I(N)\oplus Y @>{\left(\begin{smallmatrix}1&0\\0&a\end{smallmatrix}\right)}>> I(N)\oplus C(N)\\
@V{G(f)=\left(\begin{smallmatrix}k&u\\0&i\end{smallmatrix}\right)}VV @V{g}VV @V{F(f)=\left(\begin{smallmatrix}i&v\\0&c\end{smallmatrix}\right)}VV\\
K(N')\oplus I(N') @>{\left(\begin{smallmatrix}0&1\\b'&0\end{smallmatrix}\right)}>> I(N')\oplus Y' @>{\left(\begin{smallmatrix}1&0\\0&a'\end{smallmatrix}\right)}>> I(N')\oplus C(N')
\end{CD} K ( N ) ⊕ I ( N ) G ( f ) = ( k 0 u i ) ↓ ⏐ K ( N ′ ) ⊕ I ( N ′ ) ( 0 b 1 0 ) ( 0 b ′ 1 0 ) I ( N ) ⊕ Y g ↓ ⏐ I ( N ′ ) ⊕ Y ′ ( 1 0 0 a ) ( 1 0 0 a ′ ) I ( N ) ⊕ C ( N ) F ( f ) = ( i 0 v c ) ↓ ⏐ I ( N ′ ) ⊕ C ( N ′ )
Here k = K ( f ) k=K(f) k = K ( f ) , i = I ( f ) i=I(f) i = I ( f ) and c = C ( f ) c=C(f) c = C ( f ) . The diagram commutes if and only if
g = ( i v ∘ a b ′ ∘ u h ) , h ∘ b = b ′ ∘ k , a ′ ∘ h = c ∘ a . g=\begin{pmatrix}i&v\circ a\\b'\circ u&h\end{pmatrix},\qquad h\circ b=b'\circ k,\qquad a'\circ h=c\circ a. g = ( i b ′ ∘ u v ∘ a h ) , h ∘ b = b ′ ∘ k , a ′ ∘ h = c ∘ a .
The last two equations say exactly that ( f , h ) (f,h) ( f , h ) is a morphism between the corresponding objects of F A ( 0 ; K , C ) \mathcal F_A(0;K,C) F A ( 0 ; K , C ) . If f f f is an isomorphism, then k , i , c k,i,c k , i , c are invertible. Using a ∘ b = 0 a\circ b=0 a ∘ b = 0 and h ∘ b = b ′ ∘ k h\circ b=b'\circ k h ∘ b = b ′ ∘ k , we obtain
g ( 1 0 − b ∘ k − 1 ∘ u 1 ) = ( i v ∘ a 0 h ) . g\begin{pmatrix}1&0\\-b\circ k^{-1}\circ u&1\end{pmatrix}
=
\begin{pmatrix}i&v\circ a\\0&h\end{pmatrix}. g ( 1 − b ∘ k − 1 ∘ u 0 1 ) = ( i 0 v ∘ a h ) .
Thus ( f , g ) (f,g) ( f , g ) is an isomorphism if and only if ( f , h ) (f,h) ( f , h ) is.
We next check that the construction is well defined on isomorphism classes. Let two choices of splittings for the same object Z = ( N , X , β , α ) Z=(N,X,\beta,\alpha) Z = ( N , X , β , α ) give reduced objects ( N , Y 1 , b 1 , a 1 ) (N,Y_1,b_1,a_1) ( N , Y 1 , b 1 , a 1 ) and ( N , Y 2 , b 2 , a 2 ) (N,Y_2,b_2,a_2) ( N , Y 2 , b 2 , a 2 ) . Express ( 1 N , 1 X ) (1_N,1_X) ( 1 N , 1 X ) in the corresponding normal-form coordinates. Since its middle component is invertible, the calculation above gives an isomorphism h : Y 1 → Y 2 h:Y_1\to Y_2 h : Y 1 → Y 2 with h ∘ b 1 = b 2 h\circ b_1=b_2 h ∘ b 1 = b 2 and a 2 ∘ h = a 1 a_2\circ h=a_1 a 2 ∘ h = a 1 . Thus ( 1 N , h ) (1_N,h) ( 1 N , h ) is an isomorphism between the reduced objects. If two original objects are isomorphic, the same calculation applied to an isomorphism between them shows that their reduced objects are isomorphic.
The isomorphism criterion above also gives injectivity on isomorphism classes. Every reduced object is obtained by reducing its reconstruction, which gives surjectivity.
Since K , I , C K,I,C K , I , C are additive, taking direct sums of the chosen splittings shows that the bijection preserves direct sums. A reduced object ( N , Y , b , a ) (N,Y,b,a) ( N , Y , b , a ) is zero precisely when N ≅ 0 N\cong0 N ≅ 0 and Y = 0 Y=0 Y = 0 . By X ≅ I ( N ) ⊕ Y X\cong I(N)\oplus Y X ≅ I ( N ) ⊕ Y , this is equivalent to Z = ( N , X , β , α ) Z=(N,X,\beta,\alpha) Z = ( N , X , β , α ) being zero. If a reduced object decomposes into two nonzero summands, their inverse images give a decomposition of the original object, by additivity and injectivity of the bijection. The reverse implication follows from preservation of direct sums. Thus indecomposable objects correspond.
Definition 4.6. We say that F A ( μ ; G , F ) \mathcal F_A(\mu;G,F) F A ( μ ; G , F ) satisfies property (P) if any two indecomposable objects Z = ( N , X , β , α ) Z=(N,X,\beta,\alpha) Z = ( N , X , β , α ) and Z ′ = ( N ′ , X ′ , β ′ , α ′ ) Z'=(N',X',\beta',\alpha') Z ′ = ( N ′ , X ′ , β ′ , α ′ ) in F A ( μ ; G , F ) \mathcal F_A(\mu;G,F) F A ( μ ; G , F ) are isomorphic whenever N ≅ N ′ N\cong N' N ≅ N ′ in C \mathcal C C and X ≅ X ′ X\cong X' X ≅ X ′ in mod A \operatorname{mod} A mod A .
We first relate the reduced category to two comma categories.
Lemma 4.7. Suppose that K ( T ) = 0 K(T)=0 K ( T ) = 0 or C ( T ) = 0 C(T)=0 C ( T ) = 0 for every indecomposable T ∈ C T\in\mathcal C T ∈ C . Let C 1 \mathcal C_1 C 1 and C 2 \mathcal C_2 C 2 be the full additive subcategories generated by the indecomposable objects T T T with K ( T ) ≠ 0 K(T)\ne0 K ( T ) = 0 and K ( T ) = 0 K(T)=0 K ( T ) = 0 , respectively. Set
A = M ( K ∣ C 1 , soc ) , soc : mod A ⟶ mod D . \mathcal A=\mathcal M(K|_{\mathcal C_1},\operatorname{soc}),\qquad\operatorname{soc}:\operatorname{mod} A\longrightarrow\operatorname{mod} D. A = M ( K ∣ C 1 , soc ) , soc : mod A ⟶ mod D .
There are additive functors
Q : A ⟶ mod D , Q:\mathcal{A}\longrightarrow\operatorname{mod}D, Q : A ⟶ mod D ,
Θ : M ( Q , C ∣ C 2 ) ⟶ F A ( 0 ; K , C ) \Theta:\mathcal{M}(Q,\mathcal{C}|_{\mathcal{C}_2})\longrightarrow\mathcal{F}_{\mathcal{A}}(0;K,C) Θ : M ( Q , C ∣ C 2 ) ⟶ F A ( 0 ; K , C )
such that Θ \Theta Θ induces a bijection on isomorphism classes of indecomposable objects.
Proof. By assumption, C ∣ C 1 = 0 = K ∣ C 2 C|_{\mathcal{C}_1}=0=K|_{\mathcal{C}_2} C ∣ C 1 = 0 = K ∣ C 2 . For T i ∈ C i T_i\in\mathcal{C}_i T i ∈ C i , we use the canonical identifications
K ( T 1 ⊕ T 2 ) ≅ K ( T 1 ) , C ( T 1 ⊕ T 2 ) ≅ C ( T 2 ) . K(T_1\oplus T_2)\cong K(T_1),\qquad C(T_1\oplus T_2)\cong C(T_2). K ( T 1 ⊕ T 2 ) ≅ K ( T 1 ) , C ( T 1 ⊕ T 2 ) ≅ C ( T 2 ) .
For Z = ( T 1 , Y , b ) ∈ A Z=(T_1,Y,b)\in\mathcal{A} Z = ( T 1 , Y , b ) ∈ A , regard b : K ( T 1 ) ⟶ soc Y b:K(T_1)\longrightarrow\operatorname{soc}Y b : K ( T 1 ) ⟶ soc Y as a map into Y Y Y and set
Q ( Z ) = Y / ( J ( A ) Y + im b ) . Q(Z)=Y/(J(\mathcal{A})Y+\operatorname{im}b). Q ( Z ) = Y / ( J ( A ) Y + im b ) .
Let ρ Z : Y ⟶ Q ( Z ) \rho_Z:Y\longrightarrow Q(Z) ρ Z : Y ⟶ Q ( Z ) be the quotient map. For a morphism ( f 1 , g ) : Z ⟶ Z ′ (f_1,g):Z\longrightarrow Z' ( f 1 , g ) : Z ⟶ Z ′ in A \mathcal{A} A , where Z ′ = ( T 1 ′ , Y ′ , b ′ ) Z'=(T'_1,Y',b') Z ′ = ( T 1 ′ , Y ′ , b ′ ) , we have g ∘ b = b ′ ∘ K ( f 1 ) g\circ b=b'\circ K(f_1) g ∘ b = b ′ ∘ K ( f 1 ) , and so g ( im b ) ⊆ im b ′ g(\operatorname{im}b)\subseteq\operatorname{im}b' g ( im b ) ⊆ im b ′ . Since g g g is A \mathcal{A} A -linear, we have g ( J ( A ) Y ) ⊆ J ( A ) Y ′ g(J(\mathcal{A})Y)\subseteq J(\mathcal{A})Y' g ( J ( A ) Y ) ⊆ J ( A ) Y ′ . Hence
g ( J ( A ) Y + im b ) ⊆ J ( A ) Y ′ + im b ′ , g(J(\mathcal{A})Y+\operatorname{im}b)\subseteq J(\mathcal{A})Y'+\operatorname{im}b', g ( J ( A ) Y + im b ) ⊆ J ( A ) Y ′ + im b ′ ,
so ( f 1 , g ) (f_1,g) ( f 1 , g ) induces a D D D -linear map Q ( f 1 , g ) : Q ( Z ) ⟶ Q ( Z ′ ) Q(f_1,g):Q(Z)\longrightarrow Q(Z') Q ( f 1 , g ) : Q ( Z ) ⟶ Q ( Z ′ ) satisfying
Q ( f 1 , g ) ∘ ρ Z = ρ Z ′ ∘ g . Q(f_1,g)\circ\rho_Z=\rho_{Z'}\circ g. Q ( f 1 , g ) ∘ ρ Z = ρ Z ′ ∘ g .
These assignments define an additive functor Q : A ⟶ mod D Q:\mathcal{A}\longrightarrow\operatorname{mod}D Q : A ⟶ mod D .
For an object ( Z , T 2 , a ˉ ) (Z,T_2,\bar a) ( Z , T 2 , a ˉ ) of M ( Q , C ∣ C 2 ) \mathcal{M}(Q,\mathcal{C}|_{\mathcal{C}_2}) M ( Q , C ∣ C 2 ) , with Z = ( T 1 , Y , b ) Z=(T_1,Y,b) Z = ( T 1 , Y , b ) , set
Θ ( Z , T 2 , a ˉ ) = ( T 1 ⊕ T 2 , Y , b , a ˉ ∘ ρ Z ) . \Theta(Z,T_2,\bar a)=(T_1\oplus T_2,Y,b,\bar a\circ\rho_Z). Θ ( Z , T 2 , a ˉ ) = ( T 1 ⊕ T 2 , Y , b , a ˉ ∘ ρ Z ) .
On morphisms, set
Θ ( ( f 1 , g ) , f 2 ) = ( f 1 ⊕ f 2 , g ) . \Theta((f_1,g),f_2)=(f_1\oplus f_2,g). Θ (( f 1 , g ) , f 2 ) = ( f 1 ⊕ f 2 , g ) .
The relation ρ Z ∘ b = 0 \rho_Z\circ b=0 ρ Z ∘ b = 0 and the naturality identity for ρ Z \rho_Z ρ Z show that these formulas define an additive functor
Θ : M ( Q , C ∣ C 2 ) ⟶ F A ( 0 ; K , C ) . \Theta:\mathcal{M}(Q,\mathcal{C}|_{\mathcal{C}_2})\longrightarrow\mathcal{F}_{\mathcal{A}}(0;K,C). Θ : M ( Q , C ∣ C 2 ) ⟶ F A ( 0 ; K , C ) .
We first show that Θ \Theta Θ is essentially surjective. Let W = ( N , Y , b , a ) ∈ F A ( 0 ; K , C ) W=(N,Y,b,a)\in\mathcal{F}_{\mathcal{A}}(0;K,C) W = ( N , Y , b , a ) ∈ F A ( 0 ; K , C ) . By Krull–Schmidt, write N ≅ T 1 ⊕ T 2 N\cong T_1\oplus T_2 N ≅ T 1 ⊕ T 2 with T i ∈ C i T_i\in\mathcal{C}_i T i ∈ C i . Since J ( A ) K ( T 1 ) = 0 J(\mathcal{A})K(T_1)=0 J ( A ) K ( T 1 ) = 0 , the A \mathcal{A} A -linear map b : K ( T 1 ) ⟶ Y b:K(T_1)\longrightarrow Y b : K ( T 1 ) ⟶ Y has image in soc Y \operatorname{soc}Y soc Y . Thus Z = ( T 1 , Y , b ) Z=(T_1,Y,b) Z = ( T 1 , Y , b ) is an object of A \mathcal{A} A . The map a : Y ⟶ C ( T 2 ) a:Y\longrightarrow C(T_2) a : Y ⟶ C ( T 2 ) vanishes on J ( A ) Y + im b J(\mathcal{A})Y+\operatorname{im}b J ( A ) Y + im b because J ( A ) C ( T 2 ) = 0 J(\mathcal{A})C(T_2)=0 J ( A ) C ( T 2 ) = 0 and a ∘ b = 0 a\circ b=0 a ∘ b = 0 . Hence there is a unique D D D -linear map a ˉ : Q ( Z ) ⟶ C ( T 2 ) \bar a:Q(Z)\longrightarrow C(T_2) a ˉ : Q ( Z ) ⟶ C ( T 2 ) such that
a = a ˉ ∘ ρ Z . a=\bar a\circ\rho_Z. a = a ˉ ∘ ρ Z .
Thus Θ ( Z , T 2 , a ˉ ) = ( T 1 ⊕ T 2 , Y , b , a ) ≅ W \Theta(Z,T_2,\bar a)=(T_1\oplus T_2,Y,b,a)\cong W Θ ( Z , T 2 , a ˉ ) = ( T 1 ⊕ T 2 , Y , b , a ) ≅ W .
Now let U = ( Z , T 2 , a ˉ ) U=(Z,T_2,\bar a) U = ( Z , T 2 , a ˉ ) and U ′ = ( Z ′ , T 2 ′ , a ˉ ′ ) U'=(Z',T'_2,\bar a') U ′ = ( Z ′ , T 2 ′ , a ˉ ′ ) , where Z = ( T 1 , Y , b ) Z=(T_1,Y,b) Z = ( T 1 , Y , b ) and Z ′ = ( T 1 ′ , Y ′ , b ′ ) Z'=(T'_1,Y',b') Z ′ = ( T 1 ′ , Y ′ , b ′ ) . Suppose that ( f , g ) : Θ ( U ) ⟶ Θ ( U ′ ) (f,g):\Theta(U)\longrightarrow\Theta(U') ( f , g ) : Θ ( U ) ⟶ Θ ( U ′ ) is an isomorphism, and write
f = ( f 1 u v f 2 ) : T 1 ⊕ T 2 ⟶ T 1 ′ ⊕ T 2 ′ . f=\begin{pmatrix}f_1&u\\v&f_2\end{pmatrix}:T_1\oplus T_2\longrightarrow T'_1\oplus T'_2. f = ( f 1 v u f 2 ) : T 1 ⊕ T 2 ⟶ T 1 ′ ⊕ T 2 ′ .
Since C 1 \mathcal{C}_1 C 1 and C 2 \mathcal{C}_2 C 2 have no common indecomposable objects up to isomorphism, Krull–Schmidt gives T i ≅ T i ′ T_i\cong T'_i T i ≅ T i ′ . By Lemma 2.4, f 1 f_1 f 1 and f 2 f_2 f 2 are isomorphisms. The morphism relations give
g ∘ b = b ′ ∘ K ( f 1 ) , C ( f 2 ) ∘ a ˉ ∘ ρ Z = a ˉ ′ ∘ ρ Z ′ ∘ g . g\circ b=b'\circ K(f_1),\qquad C(f_2)\circ\bar a\circ\rho_Z=\bar a'\circ\rho_{Z'}\circ g. g ∘ b = b ′ ∘ K ( f 1 ) , C ( f 2 ) ∘ a ˉ ∘ ρ Z = a ˉ ′ ∘ ρ Z ′ ∘ g .
The first identity shows that ( f 1 , g ) (f_1,g) ( f 1 , g ) is an isomorphism in A \mathcal{A} A . Since Q ( f 1 , g ) ∘ ρ Z = ρ Z ′ ∘ g Q(f_1,g)\circ\rho_Z=\rho_{Z'}\circ g Q ( f 1 , g ) ∘ ρ Z = ρ Z ′ ∘ g and ρ Z \rho_Z ρ Z is surjective, the second identity gives
C ( f 2 ) ∘ a ˉ = a ˉ ′ ∘ Q ( f 1 , g ) . C(f_2)\circ\bar a=\bar a'\circ Q(f_1,g). C ( f 2 ) ∘ a ˉ = a ˉ ′ ∘ Q ( f 1 , g ) .
Therefore ( ( f 1 , g ) , f 2 ) : U ⟶ U ′ ((f_1,g),f_2):U\longrightarrow U' (( f 1 , g ) , f 2 ) : U ⟶ U ′ is an isomorphism in M ( Q , C ∣ C 2 ) \mathcal{M}(Q,\mathcal{C}|_{\mathcal{C}_2}) M ( Q , C ∣ C 2 ) . This proves that Θ \Theta Θ induces a bijection on isomorphism classes.
Finally, Θ \Theta Θ is additive and reflects zero objects. If Θ ( U ) ≅ W 1 ⊕ W 2 \Theta(U) \cong W_1 \oplus W_2 Θ ( U ) ≅ W 1 ⊕ W 2 , choose U i U_i U i with Θ ( U i ) ≅ W i \Theta(U_i) \cong W_i Θ ( U i ) ≅ W i . Additivity and injectivity on isomorphism classes give U ≅ U 1 ⊕ U 2 U \cong U_1 \oplus U_2 U ≅ U 1 ⊕ U 2 . Thus U U U is indecomposable if and only if Θ ( U ) \Theta(U) Θ ( U ) is, and the bijection restricts to a bijection on isomorphism classes of indecomposable objects. □
We now apply Theorem 3.3 to factorization categories.
Theorem 4.8. Assume that C \mathcal{C} C and mod A \operatorname{mod} A mod A are representation-finite. If F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) satisfies property ( P ) (P) ( P ) , then F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) is representation-finite.
Proof. By Lemma 4.5, it suffices to show that F = F A ( 0 ; K , C ) \mathcal{F}=\mathcal{F}_A(0;K,\mathcal{C}) F = F A ( 0 ; K , C ) is representation-finite. In fact, the category F \mathcal{F} F also satisfies property ( P ) (P) ( P ) . Suppose that two indecomposable objects of F \mathcal{F} F have isomorphic components N ≅ N ′ N\cong N' N ≅ N ′ and Y ≅ Y ′ Y\cong Y' Y ≅ Y ′ . Since I ( N ) ≅ I ( N ′ ) I(N)\cong I(N') I ( N ) ≅ I ( N ′ ) , their reconstructions have isomorphic middle terms I ( N ) ⊕ Y ≅ I ( N ′ ) ⊕ Y ′ I(N)\oplus Y\cong I(N')\oplus Y' I ( N ) ⊕ Y ≅ I ( N ′ ) ⊕ Y ′ . The reconstructions are isomorphic by property ( P ) (P) ( P ) for F A ( μ ; G , F ) \mathcal{F}_A(\mu;G,F) F A ( μ ; G , F ) , so Lemma 4.5 shows that the two reduced objects are isomorphic. We work in F \mathcal{F} F for the rest of the proof.
We claim that K ( T ) = 0 K(T)=0 K ( T ) = 0 or C ( T ) = 0 C(T)=0 C ( T ) = 0 for every indecomposable T ∈ C T\in\mathcal{C} T ∈ C . Otherwise, choose nonzero D D D -linear maps b : K ( T ) → D b:K(T)\to D b : K ( T ) → D and a : D → C ( T ) a:D\to C(T) a : D → C ( T ) . Since T T T and D D D are indecomposable in C \mathcal{C} C and mod A \operatorname{mod} A mod A , respectively, and b , a b,a b , a are nonzero, the objects
( T , D , b , 0 ) , ( T , D , 0 , a ) (T,D,b,0), \qquad(T,D,0,a) ( T , D , b , 0 ) , ( T , D , 0 , a )
are indecomposable by Remark 4.2. Property ( P ) (P) ( P ) gives an isomorphism between them, whose middle component g : D → D g:D\to D g : D → D is invertible and satisfies g ∘ b = 0 g\circ b=0 g ∘ b = 0 . This contradicts b ≠ 0 b\ne0 b = 0 .
Let C 1 , C 2 , A , Q \mathcal{C}_1,\mathcal{C}_2,A,Q C 1 , C 2 , A , Q and Θ \Theta Θ be as in Lemma 4.7 . Both C 1 \mathcal{C}_1 C 1 and C 2 \mathcal{C}_2 C 2 are representation-finite. For an indecomposable Z = ( T 1 , Y , b ) ∈ A Z=(T_1,Y,b)\in\mathcal{A} Z = ( T 1 , Y , b ) ∈ A , the object ( Z , 0 , 0 ) (Z,0,0) ( Z , 0 , 0 ) is indecomposable in M ( Q , C ∣ C 2 ) \mathcal{M}(Q,\mathcal{C}|_{\mathcal{C}_2}) M ( Q , C ∣ C 2 ) . By the lemma, Θ ( Z , 0 , 0 ) \Theta(Z,0,0) Θ ( Z , 0 , 0 ) is therefore indecomposable in F \mathcal{F} F , with components T 1 T_1 T 1 and Y Y Y . If two such objects Z , Z ′ Z,Z' Z , Z ′ have isomorphic components, property ( P ) (P) ( P ) gives Θ ( Z , 0 , 0 ) ≅ Θ ( Z ′ , 0 , 0 ) \Theta(Z,0,0)\cong\Theta(Z',0,0) Θ ( Z , 0 , 0 ) ≅ Θ ( Z ′ , 0 , 0 ) , so Z ≅ Z ′ Z\cong Z' Z ≅ Z ′ by Lemma 4.7 . Since C 1 \mathcal{C}_1 C 1 and mod A \operatorname{mod} A mod A are representation-finite, Theorem 3.3 shows that A \mathcal{A} A is representation-finite.
Now let U = ( Z , T 2 , a ‾ ) U=(Z,T_2,\overline{a}) U = ( Z , T 2 , a ) and U ′ = ( Z ′ , T 2 ′ , a ‾ ′ ) U'=(Z',T'_2,\overline{a}') U ′ = ( Z ′ , T 2 ′ , a ′ ) be indecomposable objects of M ( Q , C ∣ C 2 ) \mathcal{M}(Q,\mathcal{C}|_{\mathcal{C}_2}) M ( Q , C ∣ C 2 ) with Z ≅ Z ′ Z\cong Z' Z ≅ Z ′ and T 2 ≅ T 2 ′ T_2\cong T'_2 T 2 ≅ T 2 ′ . Writing Z = ( T 1 , Y , b ) Z=(T_1,Y,b) Z = ( T 1 , Y , b ) and Z ′ = ( T 1 ′ , Y ′ , b ′ ) Z'=(T'_1,Y',b') Z ′ = ( T 1 ′ , Y ′ , b ′ ) , we have T 1 ≅ T 1 ′ T_1\cong T'_1 T 1 ≅ T 1 ′ and Y ≅ Y ′ Y\cong Y' Y ≅ Y ′ . Thus Θ ( U ) \Theta(U) Θ ( U ) and Θ ( U ′ ) \Theta(U') Θ ( U ′ ) are indecomposable objects of F \mathcal{F} F with isomorphic components. By property ( P ) (P) ( P ) , Θ ( U ) ≅ Θ ( U ′ ) \Theta(U)\cong\Theta(U') Θ ( U ) ≅ Θ ( U ′ ) . Hence U ≅ U ′ U\cong U' U ≅ U ′ by Lemma 4.7 . Since A \mathcal{A} A and C 2 \mathcal{C}_2 C 2 are representation-finite, Theorem 3.3 shows that M ( Q , C ∣ C 2 ) \mathcal{M}(Q,\mathcal{C}|_{\mathcal{C}_2}) M ( Q , C ∣ C 2 ) is representation-finite. Lemma 4.7 now implies that F \mathcal{F} F is representation-finite. □
Composition factor algebras and finite representation type
Local composition factor algebras and matrix decompositions
Definition 5.1. An Artin algebra Λ \Lambda Λ is called a composition factor algebra if, for any two indecomposable modules M , N ∈ mod Λ M,N\in\operatorname{mod}\Lambda M , N ∈ mod Λ , the equalities
[ M : S ] = [ N : S ] for every simple Λ -module S [M:S]=[N:S]\qquad\text{for every simple }\Lambda\text{-module }S [ M : S ] = [ N : S ] for every simple Λ -module S
imply M ≅ N M\cong N M ≅ N . Here [ M : S ] [M:S] [ M : S ] denotes the multiplicity of S S S in a composition series of M M M .
Remark 5.2. The class of composition factor algebras is closed under Morita equivalence and passage to opposite algebras. Quotients of composition factor algebras are composition factor algebras [4 Proposition 2.8] . For a two-sided ideal I I I of Λ \Lambda Λ , restriction of scalars along Λ → Λ / I \Lambda\to\Lambda/I Λ → Λ/ I identifies mod ( Λ / I ) \operatorname{mod}(\Lambda/I) mod ( Λ/ I ) with the full subcategory of mod Λ \operatorname{mod}\Lambda mod Λ consisting of modules annihilated by I I I . This identification preserves submodules, composition series and direct-sum decompositions.
We begin with an elementary consequence of the definition.
Lemma 5.3. Let Λ \Lambda Λ be a composition factor algebra. If M ∈ mod Λ M \in\operatorname{mod}\Lambda M ∈ mod Λ has simple top and simple socle, both isomorphic to S S S , then M M M is uniserial and every composition factor of M M M is isomorphic to S S S .
Proof. We argue by induction on ℓ Λ ( M ) \ell_{\Lambda}(M) ℓ Λ ( M ) . The assertion is clear when M M M is simple. Otherwise, the simple top makes rad M \operatorname{rad} M rad M the unique maximal submodule, so every proper submodule of M M M , including soc M \operatorname{soc} M soc M , lies in rad M \operatorname{rad} M rad M . Consequently,
soc ( rad M ) = soc M ≅ S , top ( M / soc M ) ≅ M / rad M ≅ S . \operatorname{soc}(\operatorname{rad} M)=\operatorname{soc} M\cong S,\qquad\operatorname{top}(M/\operatorname{soc} M)\cong M/\operatorname{rad} M\cong S. soc ( rad M ) = soc M ≅ S , top ( M / soc M ) ≅ M / rad M ≅ S .
Thus rad M \operatorname{rad} M rad M and M / soc M M/\operatorname{soc} M M / soc M are indecomposable by Fact 2.2 . Both have the composition factors of M M M with one copy of S S S removed. Since Λ \Lambda Λ is a composition factor algebra, rad M ≅ M / soc M \operatorname{rad} M\cong M/\operatorname{soc} M rad M ≅ M / soc M . In particular, rad M \operatorname{rad} M rad M has simple top and socle, both isomorphic to S S S . By induction, rad M \operatorname{rad} M rad M is uniserial with all composition factors isomorphic to S S S . The same is true of M M M : its top is S S S , and all its proper submodules lie in the uniserial module rad M \operatorname{rad} M rad M .
We next consider the local case.
Proposition 5.4. Let Λ \Lambda Λ be a local composition factor algebra. Then Λ \Lambda Λ is Nakayama and hence representation-finite.
Proof. We write S = Λ / J ( Λ ) S=\Lambda/J(\Lambda) S = Λ/ J ( Λ ) . Since S S S is the unique simple left Λ \Lambda Λ -module up to isomorphism, additivity of R R R -length gives
ℓ R ( M ) = ℓ Λ ( M ) ℓ R ( S ) ( M ∈ mod Λ ) . \ell_R(M)=\ell_{\Lambda}(M)\ell_R(S)\qquad(M\in\operatorname{mod}\Lambda). ℓ R ( M ) = ℓ Λ ( M ) ℓ R ( S ) ( M ∈ mod Λ ) .
The regular modules Λ Λ {}_{\Lambda}\Lambda Λ Λ and Λ Λ \Lambda_{\Lambda} Λ Λ are indecomposable because Λ \Lambda Λ is local. Matlis duality preserves indecomposability and R R R -length, so Λ Λ {}_{\Lambda}\Lambda Λ Λ and D ( Λ Λ ) D(\Lambda_{\Lambda}) D ( Λ Λ ) are indecomposable left modules of the same composition length. Since Λ \Lambda Λ is a composition factor algebra, we obtain
Λ Λ ≅ D ( Λ Λ ) . {}_{\Lambda}\Lambda\cong D(\Lambda_{\Lambda}). Λ Λ ≅ D ( Λ Λ ) .
Thus Λ \Lambda Λ is self-injective, with top ( Λ Λ ) ≅ S ≅ soc ( Λ Λ ) \operatorname{top}({}_{\Lambda}\Lambda)\cong S\cong\operatorname{soc}({}_{\Lambda}\Lambda) top ( Λ Λ ) ≅ S ≅ soc ( Λ Λ ) . By Lemma 5.3, the module Λ Λ {}_{\Lambda}\Lambda Λ Λ is uniserial. Since Λ \Lambda Λ is local and self-injective, it is Nakayama, and hence representation-finite [3 VI, Theorem 2.1(b)] .
Notation 5.5. For a basic Artin algebra Λ \Lambda Λ and a primitive idempotent e e e , set e ′ = 1 − e e'=1-e e ′ = 1 − e . The decomposition by these idempotents gives the natural algebra isomorphism
Λ ≅ ( A U V B ) , A = e Λ e , B = e ′ Λ e ′ , U = e Λ e ′ , V = e ′ Λ e . (9) \Lambda\cong
\begin{pmatrix}
A & U\\
V & B
\end{pmatrix},
\qquad A=e\Lambda e,\quad B=e'\Lambda e',\quad U=e\Lambda e',\quad V=e'\Lambda e. \tag*{(9)} Λ ≅ ( A V U B ) , A = e Λ e , B = e ′ Λ e ′ , U = e Λ e ′ , V = e ′ Λ e . (9)
The modules U U U and V V V are respectively an A A A –B B B -bimodule and a B B B –A A A -bimodule.
Since Λ e \Lambda e Λ e is indecomposable, its endomorphism ring End Λ ( Λ e ) ≅ A op \operatorname{End}_{\Lambda}(\Lambda e)\cong A^{\operatorname{op}} End Λ ( Λ e ) ≅ A op is local [3 II, §2] . Thus the corner algebra A = e Λ e A=e\Lambda e A = e Λ e is local.
Write S e = Λ e / J ( Λ ) e S_e=\Lambda e/J(\Lambda)e S e = Λ e / J ( Λ ) e for the simple module corresponding to e e e . Since Λ \Lambda Λ is basic, e S e = S e eS_e=S_e e S e = S e and e ′ S e = 0 e'S_e=0 e ′ S e = 0 . As a left A A A -module, S e S_e S e is naturally isomorphic to A / J ( A ) A/J(A) A / J ( A ) . The other simple Λ \Lambda Λ -modules are annihilated by e e e and restrict to pairwise nonisomorphic simple B B B -modules. Restriction by an idempotent is exact. Hence, for every M ∈ mod Λ M\in\operatorname{mod}\Lambda M ∈ mod Λ ,
[ M : S e ] = ℓ A ( e M ) , [ M : S ] = [ e ′ M : e ′ S ] ( S simple , S ≇ S e ) . (10) [M:S_e]=\ell_A(eM),\qquad[M:S]=[e'M:e'S]\quad(S\text{ simple},\ S\not\cong S_e). \tag*{(10)} [ M : S e ] = ℓ A ( e M ) , [ M : S ] = [ e ′ M : e ′ S ] ( S simple , S ≅ S e ) . (10)
The following lemma provides the relations needed for the factorization description in the next subsection.
Lemma 5.6. Let Λ \Lambda Λ be a basic composition factor algebra, and let e e e be a primitive idempotent. Use the notation in (9) . Then we have
U V = 0 UV=0 U V = 0 .
J ( A ) U = 0 = V J ( A ) J(A)U=0=VJ(A) J ( A ) U = 0 = V J ( A ) .
(3) U ⊗ B V = 0 U \otimes_B V = 0 U ⊗ B V = 0 .
Proof. (1) Suppose that $0 \ne c \in UV \subseteq \Lambda e$. Choose a left Λ \Lambda Λ -submodule N ⊆ Λ e N \subseteq\Lambda e N ⊆ Λ e maximal among those not containing c c c , and let c ˉ \bar{c} c ˉ be the image of c c c in M = Λ e / N M = \Lambda e/N M = Λ e / N . By maximality, every nonzero submodule of M M M contains c ˉ \bar{c} c ˉ , and hence contains Λ c ˉ \Lambda\bar{c} Λ c ˉ . Choose a simple submodule T ⊆ M T \subseteq M T ⊆ M . Then $0 \ne \Lambda\bar{c} \subseteq T$, so T = Λ c ˉ T = \Lambda\bar{c} T = Λ c ˉ . Thus soc M = Λ c ˉ \operatorname{soc} M = \Lambda\bar{c} soc M = Λ c ˉ is simple, and e c ˉ = c ˉ ≠ 0 e\bar{c} = \bar{c} \ne0 e c ˉ = c ˉ = 0 gives soc M ≅ S e \operatorname{soc} M \cong S_e soc M ≅ S e . The module M M M also has top S e S_e S e , being a nonzero quotient of the projective module Λ e \Lambda e Λ e with simple top. By Lemma 5.3, all its composition factors are isomorphic to S e S_e S e .
Exactness of e ′ ( − ) e'(-) e ′ ( − ) now gives e ′ M = 0 e'M = 0 e ′ M = 0 . Thus V = e ′ Λ e V = e'\Lambda e V = e ′ Λ e is contained in N N N , and the Λ \Lambda Λ -submodule N N N also contains U V UV U V . This contradicts c ∉ N c \notin N c ∈ / N , proving U V = 0 UV = 0 U V = 0 .
(2) By (1), U V = 0 UV = 0 U V = 0 , so V V V is a left Λ \Lambda Λ -submodule of Λ e \Lambda e Λ e . The canonical R R R -module isomorphism Λ e / V ≅ A \Lambda e/V \cong A Λ e / V ≅ A gives A A A a left Λ \Lambda Λ -module structure. For a = ( r u v b ) ∈ Λ a = \begin{pmatrix} r & u \\ v & b \end{pmatrix} \in\Lambda a = ( r v u b ) ∈ Λ , these actions are
a ⋅ w = b w ( w ∈ V ) , a ⋅ x = r x ( x ∈ A ) . a \mathbin{\cdot} w = bw \quad(w \in V), \qquad a \mathbin{\cdot} x = rx \quad(x \in A). a ⋅ w = b w ( w ∈ V ) , a ⋅ x = r x ( x ∈ A ) .
In particular, every left B B B -submodule of V V V inherits a left Λ \Lambda Λ -module structure.
For convenience, we denote J ( A ) J(A) J ( A ) by J J J . Let Y Y Y be an indecomposable direct summand of the left B B B -module V V V . For 0 ≠ q ∈ Hom B ( V , Y ) 0 \ne q \in\operatorname{Hom}_B(V,Y) 0 = q ∈ Hom B ( V , Y ) , define a Λ \Lambda Λ -action on the R R R -module M q = A ⊕ Y M_q = A \oplus Y M q = A ⊕ Y by
( r u v b ) ( x y ) = ( r x q ( v x ) + b y ) . \begin{pmatrix} r & u \\ v & b \end{pmatrix}
\begin{pmatrix} x \\ y \end{pmatrix}
=
\begin{pmatrix} rx \\ q(vx) + by \end{pmatrix}. ( r v u b ) ( x y ) = ( r x q ( v x ) + b y ) .
This defines a Λ \Lambda Λ -module structure, since q q q is a homomorphism of B B B -modules and, by (1),
U V = 0 , ( V U ) Y ⊆ V ( U V ) = 0. UV = 0, \qquad(VU)Y \subseteq V(UV) = 0. U V = 0 , ( V U ) Y ⊆ V ( U V ) = 0.
We claim that M q M_q M q is indecomposable. Suppose that M q = M 1 ⊕ M 2 M_q = M_1 \oplus M_2 M q = M 1 ⊕ M 2 with both summands nonzero. Since e M q ≅ A A eM_q \cong{}_A A e M q ≅ A A is indecomposable, we may assume that e M 2 = 0 eM_2 = 0 e M 2 = 0 . Then e ′ M 2 ≠ 0 e'M_2 \ne0 e ′ M 2 = 0 , so the indecomposability of e ′ M q ≅ Y e'M_q \cong Y e ′ M q ≅ Y gives e ′ M 1 = 0 e'M_1 = 0 e ′ M 1 = 0 . It follows that
V M 1 ⊆ e ′ M 1 = 0 , V M 2 = V e M 2 = 0. VM_1 \subseteq e'M_1 = 0, \qquad VM_2 = VeM_2 = 0. V M 1 ⊆ e ′ M 1 = 0 , V M 2 = V e M 2 = 0.
Since q ≠ 0 q \ne0 q = 0 , choose v ∈ V v \in V v ∈ V with q ( v ) ≠ 0 q(v) \ne0 q ( v ) = 0 . Then v ( e , 0 ) = ( 0 , q ( v ) ) ≠ 0 v(e,0) = (0,q(v)) \ne0 v ( e , 0 ) = ( 0 , q ( v )) = 0 , contradicting V M q = 0 VM_q = 0 V M q = 0 .
Moreover, the inclusion of Y Y Y and the projection onto A A A give a short exact sequence of Λ \Lambda Λ -modules
0 ⟶ Y ⟶ M q ⟶ A ⟶ 0. 0 \longrightarrow Y \longrightarrow M_q \longrightarrow A \longrightarrow0. 0 ⟶ Y ⟶ M q ⟶ A ⟶ 0.
Thus the modules M q M_q M q have the same composition factors for fixed Y Y Y .
Suppose that V J ≠ 0 VJ \ne0 V J = 0 , and choose n ≥ 1 n \ge1 n ≥ 1 maximal with V J n ≠ 0 VJ^n \ne0 V J n = 0 . Such an n n n exists because J J J is nilpotent. Choose an indecomposable direct summand Y Y Y of the left B B B -module V V V whose projection p : V → Y p : V \to Y p : V → Y satisfies p ( V J n ) ≠ 0 p(VJ^n) \ne0 p ( V J n ) = 0 . There exists t ∈ J n t \in J^n t ∈ J n such that the homomorphism
h : V ⟶ Y , h ( v ) = p ( v t ) h : V \longrightarrow Y, \qquad h(v) = p(vt) h : V ⟶ Y , h ( v ) = p ( v t )
is nonzero. Since V J n + 1 = 0 VJ^{n+1} = 0 V J n + 1 = 0 , we have h ( V J ) = 0 h(VJ) = 0 h ( V J ) = 0 , whereas p ( V J ) ≠ 0 p(VJ) \ne0 p ( V J ) = 0 . The action formula therefore gives
( V J ) M h = 0 , ( V J ) M p ≠ 0. (VJ)M_h = 0, \qquad(VJ)M_p \ne0. ( V J ) M h = 0 , ( V J ) M p = 0.
Consequently, M h M_h M h and M p M_p M p are nonisomorphic indecomposable modules with the same composition factors. This contradicts the assumption that Λ \Lambda Λ is a composition factor algebra. Hence V J ( A ) = 0 VJ(A) = 0 V J ( A ) = 0 . By Remark 5.2, Λ op \Lambda^{\operatorname{op}} Λ op is also a composition factor algebra. Applying the same argument to Λ op \Lambda^{\operatorname{op}} Λ op gives J ( A ) U = 0 J(A)U = 0 J ( A ) U = 0 .
(3) Suppose that U ⊗ B V ≠ 0 U \otimes_B V \ne0 U ⊗ B V = 0 . By additivity of tensor product, there is an indecomposable direct summand Y Y Y of the left B B B -module V V V with U ⊗ B Y ≠ 0 U \otimes_B Y \ne0 U ⊗ B Y = 0 . Let p : V → Y p : V \to Y p : V → Y be its projection. Since A A A is local, a simple quotient of U ⊗ B Y U \otimes_B Y U ⊗ B Y is isomorphic to a simple submodule of A A A A AA . Composing the quotient map with such an embedding gives
0 ≠ λ ∈ Hom A ( U ⊗ B Y , A ) . 0 \ne\lambda\in\operatorname{Hom}_A(U \otimes_B Y,A). 0 = λ ∈ Hom A ( U ⊗ B Y , A ) .
Define a Λ \Lambda Λ -module N λ N_\lambda N λ on A ⊕ Y A \oplus Y A ⊕ Y by
( r u v b ) ⋅ ( x y ) = ( r x + λ ( u ⊗ y ) b y ) . \begin{pmatrix} r & u \\ v & b \end{pmatrix} \cdot\begin{pmatrix} x \\ y \end{pmatrix}
=
\begin{pmatrix} rx+\lambda(u\otimes y) \\ by \end{pmatrix}. ( r v u b ) ⋅ ( x y ) = ( r x + λ ( u ⊗ y ) b y ) .
The relations U V = 0 UV=0 U V = 0 and ( V U ) Y = 0 (VU)Y=0 ( V U ) Y = 0 ensure that this is a Λ \Lambda Λ -module structure. The same decomposition argument as in (2) shows that N λ N_\lambda N λ is indecomposable: a nontrivial decomposition would place its two indecomposable components A A A and Y Y Y in different summands, forcing the action of U U U to be zero, contrary to λ ≠ 0 \lambda\ne0 λ = 0 .
The module M p M_p M p from (2) is also indecomposable. The exact sequence there and the exact sequence
0 ⟶ A ⟶ N λ ⟶ Y ⟶ 0 0 \longrightarrow A \longrightarrow N_\lambda\longrightarrow Y \longrightarrow0 0 ⟶ A ⟶ N λ ⟶ Y ⟶ 0
show that M p M_p M p and N λ N_\lambda N λ have the same composition factors. The action of U U U is zero on M p M_p M p and nonzero on N λ N_\lambda N λ , since λ ≠ 0 \lambda\ne0 λ = 0 . Hence these modules are not isomorphic. This contradicts the assumption that Λ \Lambda Λ is a composition factor algebra and proves U ⊗ B V = 0 U \otimes_B V=0 U ⊗ B V = 0 . □ \square □
Modules as factorizations
Let Λ \Lambda Λ be a basic composition factor algebra and e e e a primitive idempotent. Set e ′ = 1 − e e'=1-e e ′ = 1 − e . As in (9) , write
A = e Λ e , B = e ′ Λ e ′ , U = e Λ e ′ , V = e ′ Λ e , A=e\Lambda e,\qquad B=e'\Lambda e',\qquad U=e\Lambda e',\qquad V=e'\Lambda e, A = e Λ e , B = e ′ Λ e ′ , U = e Λ e ′ , V = e ′ Λ e ,
and set D = A / J ( A ) D=A/J(A) D = A / J ( A ) .
For N ∈ mod B N\in\operatorname{mod} B N ∈ mod B , set
L ( N ) = U ⊗ B N , H ( N ) = Hom B ( V , N ) . (11) L(N)=U\otimes_B N,\qquad H(N)=\operatorname{Hom}_B(V,N). \tag*{(11)} L ( N ) = U ⊗ B N , H ( N ) = Hom B ( V , N ) . (11)
The bimodule structures give L ( N ) L(N) L ( N ) and H ( N ) H(N) H ( N ) the left A A A -actions
r ( u ⊗ n ) = ( r u ) ⊗ n , ( r h ) ( v ) = h ( v r ) . r(u\otimes n)=(ru)\otimes n,\qquad(rh)(v)=h(vr). r ( u ⊗ n ) = ( r u ) ⊗ n , ( r h ) ( v ) = h ( v r ) .
Since J ( A ) U = 0 = V J ( A ) J(A)U=0=VJ(A) J ( A ) U = 0 = V J ( A ) by Lemma 5.6 (2), L ( N ) L(N) L ( N ) and H ( N ) H(N) H ( N ) are annihilated by J ( A ) J(A) J ( A ) and hence are naturally D D D -modules. For a B B B -module homomorphism f : N → N ′ f:N\to N' f : N → N ′ , define
L ( f ) = 1 U ⊗ f , H ( f ) ( h ) = f ∘ h . L(f)=1_U\otimes f,\qquad H(f)(h)=f\circ h. L ( f ) = 1 U ⊗ f , H ( f ) ( h ) = f ∘ h .
These formulas define additive functors
L , H : mod B ⟶ mod D . L,H:\operatorname{mod} B\longrightarrow\operatorname{mod} D. L , H : mod B ⟶ mod D .
For each N ∈ mod B N\in\operatorname{mod} B N ∈ mod B , multiplication in Λ \Lambda Λ induces a map
μ N : L ( N ) ⟶ H ( N ) , μ N ( u ⊗ n ) ( v ) = ( v u ) n . \mu_N:L(N)\longrightarrow H(N),\qquad\mu_N(u\otimes n)(v)=(vu)n. μ N : L ( N ) ⟶ H ( N ) , μ N ( u ⊗ n ) ( v ) = ( v u ) n .
Associativity in Λ \Lambda Λ makes this a well-defined homomorphism of D D D -modules. For a B B B -module homomorphism f : N → N ′ f:N\to N' f : N → N ′ , B B B -linearity gives H ( f ) ∘ μ N = μ N ′ ∘ L ( f ) H(f)\circ\mu_N=\mu_{N'}\circ L(f) H ( f ) ∘ μ N = μ N ′ ∘ L ( f ) . Thus the maps μ N \mu_N μ N define a natural transformation μ : L → H \mu:L\to H μ : L → H . We consider the factorization category F A ( μ ; L , H ) \mathcal{F}_A(\mu;L,H) F A ( μ ; L , H ) .
We now define a functor Φ : mod Λ → F A ( μ ; L , H ) \Phi:\operatorname{mod}\Lambda\to\mathcal{F}_A(\mu;L,H) Φ : mod Λ → F A ( μ ; L , H ) . For M ∈ mod Λ M\in\operatorname{mod}\Lambda M ∈ mod Λ , the Λ \Lambda Λ -action induces homomorphisms of A A A -modules
U ⊗ B e ′ M → β M e M → α M Hom B ( V , e ′ M ) , U\otimes_B e'M \xrightarrow{\beta_M} eM \xrightarrow{\alpha_M} \operatorname{Hom}_B(V,e'M), U ⊗ B e ′ M β M e M α M Hom B ( V , e ′ M ) ,
where β M ( u ⊗ e ′ m ) = u m \beta_M(u\otimes e'm)=um β M ( u ⊗ e ′ m ) = u m and α M ( e m ) ( v ) = v m \alpha_M(em)(v)=vm α M ( e m ) ( v ) = v m . These maps satisfy α M ∘ β M = μ e ′ M \alpha_M\circ\beta_M=\mu_{e'M} α M ∘ β M = μ e ′ M . On objects and Λ \Lambda Λ -module homomorphisms φ : M → M ′ \varphi:M\to M' φ : M → M ′ , we set
Φ ( M ) = ( e ′ M , e M , β M , α M ) , Φ ( φ ) = ( φ ∣ e ′ M , φ ∣ e M ) . \Phi(M)=(e'M,eM,\beta_M,\alpha_M),\qquad\Phi(\varphi)=(\varphi|_{e'M},\varphi|_{eM}). Φ ( M ) = ( e ′ M , e M , β M , α M ) , Φ ( φ ) = ( φ ∣ e ′ M , φ ∣ e M ) .
The Λ \Lambda Λ -linearity of φ \varphi φ makes the two squares in Definition 4.1 commute. Thus these assignments define a functor.
The following proposition is a special case of the construction of Dlab and Ringel [6 §1, Proposition 1 and the following remark] . We include a proof for completeness.
Proposition 5.7. The functor Φ : mod Λ → F A ( μ ; L , H ) \Phi: \operatorname{mod} \Lambda\to\mathcal{F}_A(\mu; L, H) Φ : mod Λ → F A ( μ ; L , H ) is an equivalence of categories.
Proof. We construct a quasi-inverse Ψ \Psi Ψ . For an object T = ( N , X , β , α ) T=(N,X,\beta,\alpha) T = ( N , X , β , α ) of F A ( μ ; L , H ) \mathcal{F}_A(\mu;L,H) F A ( μ ; L , H ) , define an action on the R R R -module X ⊕ N X\oplus N X ⊕ N by
( r u v b ) ⋅ ( x n ) = ( r x + β ( u ⊗ n ) α ( x ) ( v ) + b n ) . \begin{pmatrix} r & u \\ v & b \end{pmatrix}\cdot\begin{pmatrix} x \\ n \end{pmatrix}
=
\begin{pmatrix} rx+\beta(u\otimes n) \\ \alpha(x)(v)+bn \end{pmatrix}. ( r v u b ) ⋅ ( x n ) = ( r x + β ( u ⊗ n ) α ( x ) ( v ) + bn ) .
The element 1 Λ = e + e ′ 1_\Lambda=e+e' 1 Λ = e + e ′ acts as the identity. To verify associativity, it suffices to check compatibility with the products V U VU V U and U V UV U V :
α ( β ( u ⊗ n ) ) ( v ) = ( v u ) n , β ( u ⊗ α ( x ) ( v ) ) = ( u v ) x . \alpha(\beta(u\otimes n))(v)=(vu)n,\qquad\beta(u\otimes\alpha(x)(v))=(uv)x. α ( β ( u ⊗ n )) ( v ) = ( v u ) n , β ( u ⊗ α ( x ) ( v )) = ( uv ) x .
The first identity is exactly α ∘ β = μ N \alpha\circ\beta=\mu_N α ∘ β = μ N . For the second, ( u v ) x = 0 (uv)x=0 ( uv ) x = 0 by Lemma 5.6 (1), so it remains to show that β ( u ⊗ α ( x ) ( v ) ) = 0 \beta(u\otimes\alpha(x)(v))=0 β ( u ⊗ α ( x ) ( v )) = 0 . The A A A -linearity of α \alpha α and β \beta β and the B B B -linearity of each α ( x ) \alpha(x) α ( x ) give a homomorphism of left A A A -modules
( U ⊗ B V ) ⊗ A X ⟶ X , ( u ⊗ v ) ⊗ x ⟼ β ( u ⊗ α ( x ) ( v ) ) . (U\otimes_B V)\otimes_A X\longrightarrow X,\qquad(u\otimes v)\otimes x\longmapsto\beta(u\otimes\alpha(x)(v)). ( U ⊗ B V ) ⊗ A X ⟶ X , ( u ⊗ v ) ⊗ x ⟼ β ( u ⊗ α ( x ) ( v )) .
By Lemma 5.6 (3), we have ( U ⊗ B V ) ⊗ A X = 0 (U\otimes_B V)\otimes_A X=0 ( U ⊗ B V ) ⊗ A X = 0 , and hence β ( u ⊗ α ( x ) ( v ) ) = 0 \beta(u\otimes\alpha(x)(v))=0 β ( u ⊗ α ( x ) ( v )) = 0 . The action formula therefore defines a Λ \Lambda Λ -module Ψ ( T ) \Psi(T) Ψ ( T ) . For a morphism ( f , g ) : T → T ′ (f,g):T\to T' ( f , g ) : T → T ′ , set
Ψ ( f , g ) = g ⊕ f : X ⊕ N ⟶ X ′ ⊕ N ′ . \Psi(f,g)=g\oplus f:X\oplus N\longrightarrow X'\oplus N'. Ψ ( f , g ) = g ⊕ f : X ⊕ N ⟶ X ′ ⊕ N ′ .
The relations g ∘ β = β ′ ∘ L ( f ) g\circ\beta=\beta'\circ L(f) g ∘ β = β ′ ∘ L ( f ) and H ( f ) ∘ α = α ′ ∘ g H(f)\circ\alpha=\alpha'\circ g H ( f ) ∘ α = α ′ ∘ g show that g ⊕ f g\oplus f g ⊕ f is Λ \Lambda Λ -linear. Hence these assignments define a functor Ψ \Psi Ψ .
For M ∈ mod Λ M\in\operatorname{mod}\Lambda M ∈ mod Λ , the natural map
Ψ ( Φ ( M ) ) = e M ⊕ e ′ M ⟶ M , ( x , n ) ⟼ x + n \Psi(\Phi(M))=eM\oplus e'M\longrightarrow M,\qquad(x,n)\longmapsto x+n Ψ ( Φ ( M )) = e M ⊕ e ′ M ⟶ M , ( x , n ) ⟼ x + n
is a Λ \Lambda Λ -module isomorphism. Conversely, under the canonical identifications e Ψ ( T ) ≅ X e\Psi(T)\cong X e Ψ ( T ) ≅ X and e ′ Ψ ( T ) ≅ N e'\Psi(T)\cong N e ′ Ψ ( T ) ≅ N , the action maps satisfy β Ψ ( T ) = β \beta_{\Psi(T)}=\beta β Ψ ( T ) = β and α Ψ ( T ) = α \alpha_{\Psi(T)}=\alpha α Ψ ( T ) = α . These identifications are natural in T T T and give Φ ∘ Ψ ≅ 1 F A ( μ ; L , H ) \Phi\circ\Psi\cong1_{\mathcal{F}_A(\mu;L,H)} Φ ∘ Ψ ≅ 1 F A ( μ ; L , H ) . Thus Ψ \Psi Ψ is a quasi-inverse of Φ \Phi Φ . □
Proof of Theorem 1.1
Proof of Theorem 1.1. By Remark 5.2, Morita equivalence allows us to assume that Λ \Lambda Λ is basic. We induct on ℓ R ( Λ ) \ell_R(\Lambda) ℓ R ( Λ ) among basic composition factor R R R -algebras. Every quotient by a nonzero proper two-sided ideal has smaller R R R -length and is again a basic composition factor algebra.
If J ( Λ ) = 0 J(\Lambda)=0 J ( Λ ) = 0 , the algebra is semisimple and representation-finite. Otherwise J ( Λ ) J(\Lambda) J ( Λ ) contains a simple left ideal S S S . Since Λ \Lambda Λ is basic, Λ / J ( Λ ) \Lambda/J(\Lambda) Λ/ J ( Λ ) is a product of division rings, so there is a primitive idempotent e e e with e S = S eS=S e S = S . We use the notation in (5.1) for this idempotent and set I = S Λ I=S\Lambda I = S Λ . The two-sided ideal I I I satisfies
0 ≠ I ⊆ J ( Λ ) , e I = I . 0\ne I\subseteq J(\Lambda),\qquad eI=I. 0 = I ⊆ J ( Λ ) , e I = I .
For e ′ = 1 − e e'=1-e e ′ = 1 − e , we have e ′ I = 0 e'I=0 e ′ I = 0 , so the quotient map induces an isomorphism
B = e ′ Λ e ′ ≅ e ′ ( Λ / I ) e ′ . B=e'\Lambda e'\cong e'(\Lambda/I)e'. B = e ′ Λ e ′ ≅ e ′ ( Λ/ I ) e ′ .
We also write e e e and e ′ e' e ′ for their images in Λ / I \Lambda/I Λ/ I . By induction, Λ / I \Lambda/I Λ/ I is representation-finite. The functor
( Λ / I ) e ′ ⊗ B − : mod B ⟶ mod ( Λ / I ) (\Lambda/I)e'\otimes_B-:\operatorname{mod}B\longrightarrow\operatorname{mod}(\Lambda/I) ( Λ/ I ) e ′ ⊗ B − : mod B ⟶ mod ( Λ/ I )
is fully faithful: it is left adjoint to Hom Λ / I ( ( Λ / I ) e ′ , − ) ≅ e ′ ( − ) \operatorname{Hom}_{\Lambda/I}((\Lambda/I)e',-)\cong e'(-) Hom Λ/ I (( Λ/ I ) e ′ , − ) ≅ e ′ ( − ) , and its unit is the canonical isomorphism N ≅ e ′ ( Λ / I ) e ′ ⊗ B N = B ⊗ B N N\cong e'(\Lambda/I)e'\otimes_B N=B\otimes_B N N ≅ e ′ ( Λ/ I ) e ′ ⊗ B N = B ⊗ B N . It therefore preserves indecomposability and reflects isomorphisms, so B B B is representation-finite.
By Lemma 5.6[1 ] , we have U V = 0 UV=0 U V = 0 , and so the projection
Λ ⟶ A , ( a u v b ) ⟼ a \Lambda\longrightarrow A,\qquad
\begin{pmatrix} a & u \\ v & b \end{pmatrix}\longmapsto a Λ ⟶ A , ( a v u b ) ⟼ a
is a surjective algebra homomorphism. Thus A A A is a composition factor algebra by Remark 5.2. Since A A A is local, Proposition 5.4 shows that mod A \operatorname{mod} A mod A is representation-finite.
By Proposition 5.7, we have an equivalence
mod Λ ≅ F A ( μ ; L , H ) \operatorname{mod}\Lambda\cong\mathcal{F}_{A}(\mu;L,H) mod Λ ≅ F A ( μ ; L , H )
for C = mod B \mathcal{C}=\operatorname{mod} B C = mod B and the functors L , H L,H L , H in (11) . Let Z = ( N , X , β , α ) Z=(N,X,\beta,\alpha) Z = ( N , X , β , α ) and Z ′ = ( N ′ , X ′ , β ′ , α ′ ) Z'=(N',X',\beta',\alpha') Z ′ = ( N ′ , X ′ , β ′ , α ′ ) be indecomposable objects of F A ( μ ; L , H ) \mathcal{F}_{A}(\mu;L,H) F A ( μ ; L , H ) with N ≅ N ′ N\cong N' N ≅ N ′ and X ≅ X ′ X\cong X' X ≅ X ′ . The corresponding indecomposable Λ \Lambda Λ -modules M , M ′ M,M' M , M ′ have e M ≅ X eM\cong X e M ≅ X and e ′ M ≅ N e'M\cong N e ′ M ≅ N , and similarly for M ′ M' M ′ . By (10) , they have the same composition factors. Since Λ \Lambda Λ is a composition factor algebra, M ≅ M ′ M\cong M' M ≅ M ′ , and the equivalence gives Z ≅ Z ′ Z\cong Z' Z ≅ Z ′ . Thus F A ( μ ; L , H ) \mathcal{F}_{A}(\mu;L,H) F A ( μ ; L , H ) satisfies property ( P ) (P) ( P ) . Theorem 4.8 now shows that Λ \Lambda Λ is representation-finite, completing the induction. □
Examples of composition factor algebras
We collect some classes of composition factor algebras. By Remark 5.2, their quotient algebras are still composition factor algebras.
Example 5.8.
A local Artin algebra is a composition factor algebra if and only if it is Nakayama. One direction follows from Proposition 5.4. For the converse, every indecomposable module over a Nakayama algebra is uniserial and is determined by its top and length [3 VI, Theorem 2.1(a)] . Since the algebra is local, each indecomposable module is determined by its length. Thus the algebra is a composition factor algebra.
Let Q Q Q be a quiver of Dynkin type and k k k a field. The indecomposable k Q kQ k Q -modules are determined by their dimension vectors. Thus k Q kQ k Q is a composition factor algebra; see [3 VIII, Corollary 2.4] .
A cycle of length t ≥ 1 t\geq1 t ≥ 1 in mod Λ \operatorname{mod}\Lambda mod Λ is a sequence M 0 ⟶ M 1 ⟶ ⋯ ⟶ M t = M 0 M_0\longrightarrow M_1\longrightarrow\cdots\longrightarrow M_t=M_0 M 0 ⟶ M 1 ⟶ ⋯ ⟶ M t = M 0 of nonzero nonisomorphisms between indecomposable modules. An Artin algebra Λ \Lambda Λ is representation-directed if mod Λ \operatorname{mod}\Lambda mod Λ has no cycles. Such algebras are composition factor algebras [3 IX, Corollary 2.2] . In particular, representation-finite piecewise hereditary algebras over algebraically closed fields are representation-directed [9 Theorem 1.1(iii)] . For more examples, see [14 Chapter 6] .
More generally, if mod Λ \operatorname{mod}\Lambda mod Λ has no short cycles, then Λ \Lambda Λ is a composition factor algebra [3 IX, Corollary 2.3] . A short cycle is a cycle of length at most two.
Let k k k be a field and Q n Q_n Q n the oriented cycle with n ≥ 3 n\geq3 n ≥ 3 vertices. We write J J J for the arrow ideal of k Q n kQ_n k Q n . For 2 ≤ m < n 2\leq m<n 2 ≤ m < n , the Nakayama algebra k Q n / J m kQ_n/J^m k Q n / J m is a composition factor algebra. Indeed, every indecomposable module has length at most m < n m<n m < n . Its composition factors correspond to a proper segment of the cycle, so they determine its top and length and hence its isomorphism class [3 VI, Theorem 2.1(a)] .
Acknowledgements. The work is supported by National Natural Science Foundation of China Grant No. 12371034.
Declaration of AI use. The main ideas and overall approach of this paper are due to the authors. ChatGPT (OpenAI) was used to assist with the proof of Proposition 3.6 (2). The tool was also used to search for relevant literature and check arguments and references. The authors take full responsibility for the content of this paper.
Bo Le
Department of Mathematical Sciences, Tsinghua University, Beijing, 100084, People’s Republic of China.
E-mail: bole_math@163.com
Bin Zhu
Department of Mathematical Sciences, Tsinghua University, Beijing, 100084, People’s Republic of China.
E-mail: zhu-b@mail.tsinghua.edu.cn