betaFeedback

WorksMathematics

Derived tame monomial algebras

Abstract

We prove that a finite-dimensional connected monomial algebra over an algebraically closed field is derived tame if and only if every finite connected convex subcategory of its universal cover has non-negative Euler form. Derived tameness is also equivalent to tameness of its repetitive category and trivial extension. For one-cycle algebras, this condition is equivalent to the non-negativity on R\mathbb R of ∣Q0∣|Q_0| explicitly computable polynomials in Q[t]\mathbb Q[t]. This gives a finite algorithm for deciding derived tameness. We also construct an infinite family of derived-tame one-cycle string algebras with minimal cubic relations, no two of which are derived equivalent.

Introduction

Brüstle and Geiss independently proved that a finite tree algebra is derived tame if and only if its Euler form is non-negative [6, 12]. For triangular monomial algebras, Novoa and de la Peña proved that derived tameness is equivalent to the tameness of the trivial extension [11 Theorem 2.1], and that these conditions hold precisely when every finite convex subcategory of the universal cover has a non-negative Euler form [11 Corollary 3.2]. Derived-tame quadratic string algebras and Nakayama algebras were classified in [9] and [3], respectively. Bekkert and Merklen described the indecomposable objects in the derived categories of gentle algebras [4].

Theorem A extends the local Euler criterion and the equivalences in [11 Theorem 2.1 and Corollary 3.2] to all finite-dimensional connected monomial algebras, accommodating arbitrary oriented cycles and algebras with infinite global dimension. When the global dimension is infinite, the usual Ext formula need not define an Euler form on K0(A)K_0(A). To overcome this limitation, our proof operates locally on finite connected convex full subcategories B⊆A~B \subseteq\widetilde{A} of the universal cover. As finite tree subcategories, each BB has finite global dimension, ensuring that the local Euler form χB\chi_B is well-defined. Under the local Euler condition (χB≥0\chi_B \ge0), the repetitive categories B^\widehat{B} are tame by the theorem of Geiss and Krause [10]. Lemma 2.1 then extends this tameness to the repetitive category A~\widetilde{A} of the cover, and the covering theorem establishes tameness of A~\widetilde{A}, yielding derived tameness of AA.

The local Euler criterion involves infinitely many finite convex subcategories. For a one-cycle algebra, Theorem 3.1 replaces these checks by a finite computation from the quiver and its monomial relations. The deck group is Z\mathbb{Z}, so nonzero paths between vertex lifts define a Hermitian Laurent polynomial matrix. The relation between the Euler and Cartan forms, together with a Fourier calculation, reduces the local Euler condition to positive semidefiniteness of this matrix on the unit circle. A rational parametrization of the circle then gives ∣Q0∣|Q_0| polynomials in Q[t]\mathbb{Q}[t] whose non-negativity on R\mathbb{R} decides derived tameness. Their degrees are bounded in terms of the longest nonzero path. The criterion applies to one-cycle monomial algebras with relations of arbitrary length, including non-quadratic string algebras. When the unique cycle is not oriented, Corollary 3.3 gives test polynomials of degree at most 2j2j for 1≤j≤∣Q0∣1 \le j \le|Q_0|.

The two main results are as follows. Let k\mathbf{k} be an algebraically closed field and let A=kQ/IA = \mathbf{k}Q/I be finite-dimensional, basic and connected. Write A~\widetilde{A} for its universal bound-quiver cover, A^\widehat{A} for its repetitive category, and T(A)=A⋉DAT(A) = A \ltimes DA for its trivial extension. We call AA one-cycle if ∣Q1∣−∣Q0∣+1=1|Q_1| - |Q_0| + 1 = 1. The underlying undirected graph retains any loops and multiple edges of QQ; in the one-cycle case, its cycle may have any orientation. For a finite tree algebra BB, its Euler form is non-negative if χB(x)≥0\chi_B(x) \ge0 for every x∈K0(B)x \in K_0(B).

Theorem A. For a finite-dimensional connected monomial algebra AA, the following conditions are equivalent:

  1. AA is derived tame;

  2. every finite connected convex full subcategory B⊆A~B \subseteq\widetilde{A} has non-negative Euler form;

  3. A~\widetilde{A} is tame;

  4. A^\widehat{A} is tame;

  5. T(A)T(A) is tame.

When the underlying graph has a cycle, these conditions are also equivalent to the condition that every finite connected convex full subcategory of A~\widetilde{A} is derived equivalent to a semichain algebra (Proposition 2.6).

Theorem B. Let A=kQ/IA = \mathbf{k}Q/I be a finite-dimensional connected one-cycle monomial algebra, and form H(z)H(z) from the path counts in (1) using spanning-tree lifts. Let RQR_Q be the arrow ideal, and set q=∣Q0∣q = |Q_0|, L=min⁡{ℓ≥2:RQℓ⊆I}L = \min\{\ell\ge2 : R_Q^\ell\subseteq I\} and M=L−1M = L - 1. Now, define

K(t)=(1+t2)MH(1+it1−it),det⁡(sIq+K(t))=sq+∑j=1qcj(t)sq−j.K(t) = (1 + t^2)^M H\left(\frac{1 + it}{1 - it}\right), \qquad\det(sI_q + K(t)) = s^q + \sum_{j=1}^{q} c_j(t)s^{q-j}.

The following conditions are equivalent:

  1. AA is derived tame;

  2. H(z)H(z) is positive semidefinite for every ∣z∣=1|z| = 1;

  3. cj(t)≥0c_j(t) \ge0 for every t∈Rt \in\mathbb{R} and 1≤j≤q1 \le j \le q.

Each cjc_j lies in Q[t]\mathbb{Q}[t] and has degree at most 2j(L−1)2j(L - 1) when nonzero. These polynomials are computable from QQ and a finite list of monomial generators of II, and their non-negativity gives a finite test for derived tameness.

Acknowledgments

Zheng Xin is grateful to his supervisor, Prof. Wen Chang, for his academic guidance and support. Zheng Xin also thanks Marlon Pimenta Fonseca for patiently and thoroughly answering all his questions during his study of [9].

Derived tameness and tree algebras

Throughout this paper, kk denotes an algebraically closed field and D=Hom⁡k(−,k)D = \operatorname{Hom}_{k}(-, k) is the standard duality. Unless otherwise stated, all algebras are assumed to be finite-dimensional, basic, and connected. We work with finitely generated right modules; for an algebra AA, we denote by mod⁡A\operatorname{mod} A the category of finitely generated right AA-modules, by Db(A)=Db(mod⁡A)D^{b}(A) = D^{b}(\operatorname{mod} A) its bounded derived category, and write K−,b(proj⁡A)K^{-,b}(\operatorname{proj} A) for the homotopy category of right-bounded complexes of finitely generated projective AA-modules with bounded cohomology. Composition of paths in quivers is written from left to right. The sets of vertices and arrows of QQ are denoted by Q0Q_{0} and Q1Q_{1}, respectively. Its underlying undirected graph retains loops and multiple edges.

Let QQ be a finite quiver, let RQ⊆kQR_{Q} \subseteq kQ be the arrow ideal, and let I⊆kQI \subseteq kQ be an admissible ideal (RQL⊆I⊆RQ2R_{Q}^{L} \subseteq I \subseteq R_{Q}^{2} for some L≥2L \ge2) generated by paths. The algebra A=kQ/IA = kQ/I is called a monomial algebra.

The classes of paths outside II, including the trivial paths eve_{v}, form a basis of AA. We call these classes the nonzero paths of AA. Every nonzero path has length less than LL. This basis will be used in the covering arguments and in the matrices of § 5.

Definition 1.1. Let A=kQ/IA = kQ/I be a monomial algebra.

(1) We call AA a string algebra if at most two arrows start and at most two arrows end at each vertex, and every arrow α\alpha has at most one successor β\beta with αβ∉I\alpha\beta\notin I and at most one predecessor γ\gamma with γα∉I\gamma\alpha\notin I. Only composable pairs of arrows are considered; see [5].

(2) We call AA gentle if it is a string algebra, II is generated by paths of length two, and every arrow has at most one successor and at most one predecessor whose composition belongs to II.

(3) We call AA triangular if QQ has no oriented cycles.

(4) We call AA a tree algebra if the underlying undirected graph of QQ is a tree.

Definition 1.2. [2 Definition 1.2] Let A=kQ/IA = kQ/I be a finite-dimensional algebra.

(1) The vector rank of a bounded complex of projective AA-modules P∙P_{\bullet} is the sequence r∙(P∙)=(r(Pn))n∈Z\mathbf{r}_{\bullet}(P_{\bullet}) = (\mathbf{r}(P_{n}))_{n\in\mathbb{Z}}, where r(Pn)=(rv)v∈Q0\mathbf{r}(P_{n}) = (r_{v})_{v\in Q_{0}} records the multiplicity of each indecomposable projective evAe_{v}A in PnP_{n}.

(2) A rational family of minimal complexes over AA is a bounded complex (P∙,d∙)(P_{\bullet},d_{\bullet}) of projective (A⊗kΛ)(A\otimes_{k}\Lambda)-modules, where Λ=k[t,f(t)−1]\Lambda= k[t,f(t)^{-1}] (f≠0f\ne0), such that Im⁡dn⊆(rad⁡A)Pn−1\operatorname{Im}d_{n}\subseteq(\operatorname{rad}A)P_{n-1}. Its specialization at (m,λ)(m,\lambda) (for m≥1m\ge1 and f(λ)≠0f(\lambda)\ne0) is the complex of AA-modules

P∙(m,λ)=P∙⊗Λk[t](t−λ)m.P_{\bullet}(m,\lambda)=P_{\bullet}\otimes_{\Lambda}\frac{k[t]}{(t-\lambda)^{m}}.

(3) The algebra AA is derived tame if, for each vector rank r∙\mathbf{r}_{\bullet}, all indecomposable minimal complexes of rank r∙\mathbf{r}_{\bullet}, except for finitely many isomorphism classes, are isomorphic to P∙(m,λ)P_{\bullet}(m,\lambda) for some m,λm,\lambda and some family P∙P_{\bullet} belonging to a finite set P(r∙)\mathfrak{P}(\mathbf{r}_{\bullet}) of rational families.

(4) The algebra AA is derived wild if there exists a bounded complex P∙P_\bullet of projective (A⊗kΣ)(A \otimes_k \Sigma)-modules, where Σ=k⟨x,y⟩\Sigma= k\langle x,y\rangle, such that the functor

−⊗ΣP∙:mod⁡Σ⟶K−,b(proj⁡A)-\otimes_{\Sigma} P_\bullet:\operatorname{mod}\Sigma\longrightarrow K^{-,b}(\operatorname{proj} A)

preserves indecomposability and reflects isomorphism classes.

The definition above is equivalent to the formulation in terms of cohomology dimensions in [10], see the discussion preceding [2 Definition 1.2]. We also use the tame–wild dichotomy [2 Theorem 1.3] and invariance of derived tameness under derived equivalence [10 Theorem 5.1].

Definition 1.3. A kk-linear category RR is called locally bounded if

(1) distinct objects of RR are non-isomorphic;

(2) every endomorphism ring of an object is local;

(3) every morphism space is finite-dimensional; and

(4) each object has nonzero morphisms to or from only finitely many objects.

Write R0R_0 for the set of objects. A finite-dimensional RR-module has a dimension vector of finite support.

Definition 1.4. Let RR be a locally bounded kk-linear category. A one-parameter family of RR-modules is a representation M\mathcal{M} over a ring Λ=k[t,f(t)−1]\Lambda= k[t,f(t)^{-1}], where f≠0f \ne0, such that M\mathcal{M} has finite support, every M(x)\mathcal{M}(x) is a finite free Λ\Lambda-module, and all structure maps are Λ\Lambda-linear. For λ∈k\lambda\in k with f(λ)≠0f(\lambda) \ne0, its fibre at λ\lambda is

Mλ=M⊗ΛΛ/(t−λ).\mathcal{M}_{\lambda}=\mathcal{M}\otimes_{\Lambda}\Lambda/(t-\lambda).

The dimension vector of the family is the vector of the ranks of its components. We call RR tame if, for each fixed dimension vector, all but finitely many indecomposable RR-modules are fibres of finitely many one-parameter families. Representation-finite categories are included in this definition.

Definition 1.5. A full subquiver Q′Q' of QQ is called convex if every oriented path in QQ whose endpoints lie in Q′Q' is entirely contained in Q′Q', including paths that belong to II (equivalently, paths that are zero modulo II). The associated convex full subcategory has the induced quiver and the induced relations.

Lemma 1.6. For a finite-dimensional connected monomial algebra A=kQ/IA=kQ/I, the universal covering quiver is a locally finite tree, and its deck group is free of rank ∣Q1∣−∣Q0∣+1|Q_1|-|Q_0|+1. The covering category A~\widetilde{A} is locally bounded. Every finite set of its vertices lies in a finite connected convex full subcategory of finite global dimension.

Proof. A monomial relation does not identify distinct paths in the homotopy relation defining π1(Q,I)\pi_1(Q,I). Thus π1(Q,I)\pi_1(Q,I) is the fundamental group of the underlying graph. The universal graph cover, with the lifted relations, gives the bound-quiver cover A~\widetilde{A} [9 Remark 2.5]. Its deck group GG acts freely on vertices, and A~/G≅A\widetilde{A}/G \cong A.

Each vertex of the covering tree is incident with finitely many arrows, and every nonzero path has length less than LL. Only finitely many nonzero paths can therefore start or end at a fixed vertex. Thus A~\widetilde{A} is locally bounded.

Given a nonempty finite set of vertices, take the union of the undirected paths joining them. The result is a finite connected subtree. Every oriented path between its vertices lies in this subtree, so the corresponding full subcategory BB is convex. Order the vertices so that every arrow goes from a smaller index to a larger one. The composition factors of rad⁡(eiB)\operatorname{rad}(e_iB) then lie at vertices with larger indices. Induction, starting at the last vertex, shows that each simple BB-module has finite projective dimension. Hence BB has finite global dimension. □\square

For any finite-dimensional basic algebra AA, put K0(A)=K0(mod⁡A)K_0(A)=K_0(\operatorname{mod} A). This group is free on the classes of the simple modules, so rank⁡K0(A)=∣Q0∣\operatorname{rank} K_0(A)=|Q_0|. The class of a bounded complex is the alternating sum of the classes of its cohomology modules. This identifies K0(A)K_0(A) with K0(Db(A))K_0(D^b(A)), so derived equivalence preserves its rank even when the global dimension is infinite.

For a finite tree algebra BB, the global dimension is finite. Its Euler pairing and quadratic form are therefore given by

⟨[X],[Y]⟩B=∑j≥0(−1)jdim⁡kExt⁡Bj(X,Y),χB(x)=⟨x,x⟩B.\langle[X],[Y]\rangle_B=\sum_{j\geq0}(-1)^j\dim_k\operatorname{Ext}^j_B(X,Y),\qquad\chi_B(x)=\langle x,x\rangle_B.

The sum is finite and defines a bilinear form on K0(B)K_0(B). We write χB≥0\chi_B\geq0 if it is non-negative on every element of K0(B)K_0(B), including vectors with negative coordinates in the basis of simple classes. This is equivalent to non-negativity on K0(B)⊗ZRK_0(B)\otimes_{\mathbb{Z}}\mathbb{R}, since homogeneity gives non-negativity on rational vectors and continuity extends it to real vectors.

For n≥1n\geq1 and 0≤m≤n0\leq m\leq n, the semichain algebra S(n,m)S(n,m) is the incidence algebra of a poset with nn levels. The first mm levels contain two incomparable elements each; the remaining levels are singletons. Every element of an earlier level is smaller than every element of a later level. Recall that the incidence algebra has basis exye_{xy} for x≤yx\leq y, with exyeyz=exze_{xy}e_{yz}=e_{xz} and all other basis products zero. We use the connected members of this family [6 p. 302], excluding S(1,1)≅k×kS(1,1)\cong k\times k.

Theorem 1.7. A finite connected tree algebra BB is derived tame if and only if χB≥0\chi_B\geq0. In that case BB is derived equivalent to a semichain algebra, a hereditary algebra of type EpE_p or E~p\widetilde{E}_p with p=6,7,8p=6,7,8, or a tubular algebra. Semichain algebras are derived tame.

If χB≥0\chi_B\geq0 and BB contains a convex subcategory of one of these exceptional or tubular derived types, then BB itself has one of these types; in particular, rank⁡K0(B)≤10\operatorname{rank} K_0(B)\leq10.

The classification is due to Brüstle [6 Theorems 1.1 and 1.2]. The assertion about convex subcategories follows from [1 Proposition 2.1]. The hereditary types have at most nine vertices, and the tubular types have Grothendieck group ranks 6, 8, 9, 10. These ranks give the bound used in Proposition 2.6.

A covering criterion for derived tameness

For a locally bounded category RR, the repetitive category R^\widehat{R} has objects (x,i)∈R0×Z(x,i)\in R_0\times\mathbb{Z} and morphism spaces

R^((x,i),(y,j))={R(x,y),j=i,DR(y,x),j=i+1,0,otherwise.\widehat{R}\bigl((x,i),(y,j)\bigr)= \begin{cases} R(x,y), & j=i,\\ DR(y,x), & j=i+1,\\ 0, & \text{otherwise}. \end{cases}

For a bound-quiver category, R(x,y)R(x,y) is the space of paths from xx to yy modulo the relations. Each layer is a copy of RR. Composition within a layer is that of RR, and composition between adjacent layers uses the natural bimodule structure on DRDR. The product of two morphisms in the dual spaces is zero.

The trivial extension of a finite-dimensional algebra AA is T(A)=A⋉DAT(A)=A\ltimes DA. Its underlying vector space is A⊕DAA\oplus DA, with multiplication

(a,f)(b,g)=(ab,ag+fb).(a,f)(b,g)=(ab,ag+fb).

The layer translation ν(x,i)=(x,i+1)\nu(x,i)=(x,i+1) gives A^/⟨ν⟩≅T(A)\widehat{A}/\langle\nu\rangle\cong T(A).

The next lemma uses finite quotient categories to study modules with fixed support in a repetitive category.

Lemma 2.1. Let RR be locally bounded. Suppose that every finite subset of R0R_0 is contained in a finite full subcategory BB for which B^\widehat{B} is tame. Then R^\widehat{R} is locally bounded and tame.

Proof. The displayed morphism spaces show that R^\widehat{R} is locally bounded. Fix a finite set S⊆R^0S\subseteq\widehat{R}_0 and choose BB containing the projection of SS to R0R_0. Since BB is full, B^\widehat{B} is the full subcategory of R^\widehat{R} on B0×ZB_0\times\mathbb{Z}. Both categories have the same full subcategory CC on SS.

Let JBJ_B be the ideal of CC generated by morphisms that factor through an object of B^\widehat{B} outside SS. Define JRJ_R in the same way using R^\widehat{R}. Every factorization in B^\widehat{B} is also a factorization in R^\widehat{R}, so

JB⊆JR,C/JB↠C/JR.J_B\subseteq J_R,\qquad C/J_B\twoheadrightarrow C/J_R.

The ideal JBJ_B acts as zero on every B^\widehat{B}-module supported on SS. Conversely, a C/JBC/J_B-module extends by zero to B^\widehat{B}, since compositions through objects outside SS act as zero by the definition of JBJ_B. Restriction and extension by zero therefore give an equivalence

mod⁡(C/JB)≃{M∈mod⁡B^:supp⁡M⊆S}.\operatorname{mod}(C/J_B)\simeq\{M\in\operatorname{mod}\widehat{B}:\operatorname{supp}M\subseteq S\}.

It preserves dimensions, isomorphism classes and indecomposability. The same statement holds for JRJ_R and R^\widehat{R}.

For a one-parameter family with dimension vector supported on SS, each component outside SS has free rank zero and hence vanishes. The action therefore factors through C/JBC/J_B. Restriction and extension by zero commute with tensoring over Λ\Lambda, so the equivalence also applies to families and their specializations.

For a fixed total dimension, there are only finitely many dimension vectors on SS. Since B^\widehat{B} is tame, the equivalence gives finitely many one-parameter families, with finitely many exceptions, for the indecomposable C/JBC/J_B-modules of that dimension. Thus C/JBC/J_B is tame. To pass to C/JRC/J_R, choose Λ\Lambda-bases for each family. If infinitely many fibres are annihilated by JR/JBJ_R/J_B, every matrix entry describing its action vanishes at infinitely many allowed parameter values. These entries lie in Λ=k[t,f(t)−1]\Lambda=\mathbf{k}[t,f(t)^{-1}], so they vanish identically and the family is a C/JRC/J_R-family. Each remaining family contributes only finitely many C/JRC/J_R-modules, which are added to the finite exceptional set. Hence C/JRC/J_R is tame. Applying this argument to the support of each dimension vector of R^\widehat{R} proves that R^\widehat{R} is tame. □

We use two comparison theorems. Let R→R/GR\to R/G be a Galois covering of locally bounded categories, with GG torsion-free. Then

R is tame  ⟺  R/G is tameR\text{ is tame}\iff R/G\text{ is tame}

by [8 Corollary 10.7(1)]; see also [DO00, Theorem 9.1(1)].

Geiss and Krause [10 Theorem 5.3] proved that

A^ tame⟹A derived tame.\widehat{A}\text{ tame}\Longrightarrow A\text{ derived tame}.

The converse holds when gldim⁡A<∞\operatorname{gldim}A<\infty.

Lemma 2.2. The universal covering F:A~→AF:\widetilde{A}\to A of a finite-dimensional monomial algebra induces a Galois covering

F~:A~^→A^,A~^/G≅A^,\widetilde{F}:\widehat{\widetilde{A}}\to\widehat{A},\qquad\widehat{\widetilde{A}}/G\cong\widehat{A},

where g(x,i)=(gx,i)g(x,i)=(gx,i) for g∈Gg\in G.

Proof. Nonzero paths form bases of the morphism spaces of AA and A~\widetilde{A}. On each layer define F^\widehat{F} by FF. On adjacent layers, send the functional dual to a lifted path pp to the functional dual to F(p)F(p).

Since the algebras are monomial, a product of two nonzero paths is either another nonzero path or zero. Write p∨p^{\vee} for the functional dual to a nonzero path pp. With our convention on path multiplication, the dual bimodule actions are

q⋅p∨={(p′)∨,p=p′q,0,otherwise,p∨⋅r={(p′′)∨,p=rp′′,0,otherwise.q \cdot p^{\vee} = \begin{cases} (p')^{\vee}, & p=p'q,\\ 0, & \text{otherwise}, \end{cases} \qquad p^{\vee}\cdot r = \begin{cases} (p'')^{\vee}, & p=rp'',\\ 0, & \text{otherwise}. \end{cases}

The remaining paths p′p' and p′′p'' may be trivial. Each path has a unique lift once its starting vertex is fixed, and its subpaths lift with it, so these products agree under F^\widehat{F}. Products of two dual morphisms are zero in both repetitive categories. Hence F^\widehat{F} preserves composition.

For lifts x,yx,y of a,ba,b, respectively, path lifting gives

⨁g∈GA~(x,gy)≅A(a,b),⨁g∈GA~(gx,y)≅A(a,b).\bigoplus_{g\in G}\widetilde{A}(x,gy)\cong A(a,b),\qquad \bigoplus_{g\in G}\widetilde{A}(gx,y)\cong A(a,b).

Only finitely many summands are nonzero, by local boundedness. Dualizing these isomorphisms with aa and bb interchanged gives the covering isomorphisms between adjacent layers. For other pairs of layers both sides are zero. The group acts freely on objects, and its orbits are indexed by (a,i)∈Q0×Z(a,i)\in Q_0\times\mathbb{Z}. Hence F^\widehat{F} is a Galois covering and A~/G≅A^\widetilde{A}/G\cong\widehat{A}. □

Proof of Theorem A. (i)⇒\Rightarrow(ii). Suppose that (ii) fails for some BB. By Theorem 1.7 and the tame–wild dichotomy, BB is derived wild. Derived wildness passes from a finite full subcategory of the monomial universal cover to AA [9 Lemma 2.8(a)]. Thus AA is derived wild, contrary to (i).

(ii)⇒\Rightarrow(iii). Let SS be a finite set of vertices of A~\widetilde{A}. By Lemma 1.6, it lies in a finite connected convex tree subcategory BB of finite global dimension. The assumed Euler condition and Theorem 1.7 show that BB is derived tame. Since BB has finite global dimension, the Geiss–Krause theorem shows that B^\widehat{B} is tame. Lemma 2.1 gives tameness of A~\widetilde{A}.

(iii)⇔\Leftrightarrow(iv). The deck group GG is free, hence torsion-free. The equivalence follows from the covering theorem and Lemma 2.2.

(iv)⇒\Rightarrow(i). This is the Geiss–Krause implication stated above.

(iv)⇔\Leftrightarrow(v). The layer translation acts freely, and its group is isomorphic to Z\mathbb{Z}. Apply the same covering theorem to A^/⟨ν⟩≅T(A)\widehat{A}/\langle\nu\rangle\cong T(A). □

Remark 2.3. When ∣Q1∣−∣Q0∣+1≥1|Q_1|-|Q_0|+1\geq1, the covering quiver is an infinite tree. Theorem A then involves infinitely many finite connected convex full subcategories. A single such subcategory whose Euler form takes a negative value proves derived wildness, as the next corollary shows. For one-cycle algebras, § 5 gives a finite test for the local Euler condition.

Corollary 2.4. Let A=kQ/IA=\mathsf{k}Q/I be a finite-dimensional connected monomial algebra. The following conditions are equivalent:

  1. AA is derived wild;

  2. there is a finite connected convex full subcategory B⊆A~B\subseteq\widetilde{A} whose Euler form takes a negative value;

(iii) there is a finite connected convex full subcategory B⊆A~B \subseteq\widetilde{A} that is derived wild and whose proper connected convex full subcategories are all derived tame.

Proof. Theorem Theorem A and the derived tame–wild dichotomy give (i)⇔(ii). If (ii) holds, then Theorem 1.7 and the dichotomy give a finite connected convex full subcategory that is derived wild. Choose one with the least number of vertices. Each of its proper connected convex full subcategories is convex in A~\widetilde{A} and has fewer vertices, so it is derived tame. This proves (iii). Conversely, Theorem 1.7 shows that the Euler form of the subcategory in (iii) takes a negative value, giving (ii).

Example 2.5. Let A=kQ/⟨αβα⟩A = \mathbb{k}Q/\langle\alpha\beta\alpha\rangle, where QQ is the quiver

Quiver with vertices 0, 1, 2, 3, 4, and 5 and arrows labelled $\alpha$, $\beta$, and $\gamma_i$

Figure 1.

Every path of length five contains αβα\alpha\beta\alpha as a subpath, so AA is finite-dimensional. Its underlying graph has six vertices and six edges, hence AA is one-cycle.

Choose a lift 0~\widetilde{0} of vertex 0. The lift of β\beta ending at 0~\widetilde{0} and the four lifts of γi\gamma_i starting there form the following subtree of the universal cover:

Subtree of the universal cover with lifted vertices and arrows

Figure 2.

It defines a finite connected convex full subcategory BB. No lifted relation is contained in this subtree, so BB is the path algebra of the displayed five-armed star. Let x∈K0(B)x \in K_0(B) have coordinate 2 at the centre and 1 at each leaf. Since BB is hereditary,

χB(x)=22+5⋅12−5⋅2⋅1=−1.\chi_B(x) = 2^2 + 5 \cdot1^2 - 5 \cdot2 \cdot1 = -1.

Thus AA is derived wild by Corollary 2.4.

When the underlying graph has a cycle, the next proposition adds a semichain condition to the equivalent conditions of Theorem A.

Proposition 2.6. Let A=kQ/IA = \mathbb{k}Q/I be a finite-dimensional connected monomial algebra with ∣Q1∣−∣Q0∣+1≥1|Q_1| - |Q_0| + 1 \ge1. Every finite connected convex full subcategory of A~\widetilde{A} has non-negative Euler form if and only if every such subcategory is derived equivalent to a semichain algebra.

Proof. Assume the local Euler condition. The covering group is infinite, since it is free of positive rank. Its free action on vertices implies that the covering tree is infinite.

Suppose that a finite connected convex subcategory CC has one of the exceptional or tubular derived types in Theorem 1.7. Choose a finite set T⊇C0T \supseteq C_0 of covering vertices with ∣T∣>10|T| > 10 and take the union of the undirected paths joining them. By Lemma 1.6, the resulting full subcategory BB is finite, connected and convex, contains CC, and has ∣B0∣>10|B_0| > 10. Since CC remains convex in BB and χB≥0\chi_B \ge0, the extension property in Theorem 1.7 gives rank K0(B)≤10K_0(B) \le10. This contradicts rank K0(B)=∣B0∣>10K_0(B) = |B_0| > 10. The

classification in Theorem 1.7 now shows that every finite connected convex subcategory is derived equivalent to a semichain algebra.

Conversely, suppose every finite connected convex subcategory is derived equivalent to a semichain algebra. Semichain algebras are derived tame, and derived equivalence preserves derived tameness. Theorem 1.7 therefore gives the local Euler condition. □

Remark 2.7. The rank bound explains the conclusion of Proposition 2.6. The exceptional hereditary and tubular derived types in Theorem 1.7 have Grothendieck group rank at most ten. Under the local Euler condition, the same bound holds for every finite connected convex full extension of a subcategory of one of these types. In the infinite covering tree, such extensions have arbitrarily many vertices. The finite-tree classification therefore leaves only the semichain derived-equivalence classes.

A polynomial criterion for one-cycle monomial algebras

Throughout this section, A=kQ/IA = kQ/I is a finite-dimensional connected one-cycle monomial algebra, with covering group G=⟨τ⟩≅ZG = \langle\tau\rangle\cong\mathbb{Z}. We use its nonzero paths to form a matrix that tests the Euler forms of the finite convex subcategories of A~\widetilde{A}.

Set q=∣Q0∣q = \lvert Q_0\rvert and choose a lift xix_i of each vertex i∈Q0i \in Q_0. Every vertex of the cover has a unique expression τkxi\tau^k x_i. For d∈Zd \in\mathbb{Z}, define

(Dd)ij=#{nonzero paths from xi to τdxj}.(D_d)_{ij} = \#\{\text{nonzero paths from }x_i\text{ to }\tau^d x_j\}.

These counts include the trivial paths, so each DdD_d has integer entries. Only finitely many DdD_d are nonzero, since every nonzero path has length less than LL and the covering quiver is locally finite. Put

P(z)=∑d∈ZDdzd,H(z)=12(P(z)+P(z−1)T).(1)P(z) = \sum_{d\in\mathbb{Z}} D_d z^d,\qquad H(z) = \frac{1}{2}\left(P(z) + P(z^{-1})^{\mathsf T}\right). \tag*{(1)}

For ∣z∣=1\lvert z\rvert= 1, the matrix H(z)H(z) is Hermitian, and its positive semidefiniteness is tested over C\mathbb{C}. Its coefficients are determined by integer path counts, independently of the characteristic of kk.

The matrices can be computed directly from QQ and a finite set R\mathcal{R} of monomial generators of II. A path is nonzero in AA exactly when it contains no member of R\mathcal{R} as a subpath. Starting from the trivial paths, we form paths of length ℓ\ell by appending arrows to the nonzero paths of length ℓ−1\ell- 1 and discarding those that contain a member of R\mathcal{R}. This process lists all nonzero paths by length and terminates at the first empty list, since AA is finite-dimensional. A one-cycle quiver has an arrow, and I⊆RQ2I \subseteq R_Q^2, so this first occurs at some length ℓ≥2\ell\ge2. At that point RQℓ⊆IR_Q^\ell\subseteq I, while the preceding list is nonempty. Hence ℓ=L\ell= L, so the procedure also computes LL.

Spanning-tree lifts. Choose a spanning tree TT of the underlying graph and a lift T~\widetilde{T} to the covering tree. From now on, take its vertices as the lifts xix_i. The unique arrow outside TT, together with the unique undirected path in TT joining its endpoints, gives a generator of the fundamental group. Consequently, its lift starting in T~\widetilde{T} ends in τT~\tau\widetilde{T} for a generator τ\tau of the deck group. Give this arrow weight 11 and every tree arrow weight 00. A path from ii to jj of total weight dd then lifts from xix_i to τdxj\tau^d x_j. Thus (Dd)ij(D_d)_{ij} counts the nonzero paths from ii to jj of weight dd. Every path has weight 00 or 11. Since every nonzero path has length less than LL, these spanning-tree lifts give Dd=0D_d = 0 unless 0≤d<L0 \le d < L.

Theorem 3.1. Let A=kQ/IA = kQ/I be a finite-dimensional connected one-cycle monomial algebra, with q=∣Q0∣q = \lvert Q_0\rvert and L=min⁡{ℓ≥2:RQℓ⊆I}L = \min\{\ell\ge2 : R_Q^\ell\subseteq I\}. Form H(z)H(z) as in (1), using the spanning-tree lifts above. Set M=L−1M = L - 1 and define

z(t)=1+it1−it,K(t)=(1+t2)MH(z(t)),(2)z(t) = \frac{1+it}{1-it}, \qquad K(t) = (1+t^2)^M H(z(t)), \tag*{(2)}

and

det⁡(sIq+K(t))=sq+c1(t)sq−1+⋯+cq(t).(3)\det(sI_q + K(t)) = s^q + c_1(t)s^{q-1} + \cdots+ c_q(t). \tag*{(3)}

The following conditions are equivalent:

(i) every finite connected convex full subcategory of A~\widetilde{A} has non-negative Euler form;

(ii) H(z)⪰0H(z) \succeq0 for every ∣z∣=1\lvert z\rvert= 1;

(iii) cj(t)≥0c_j(t) \ge0 for every t∈Rt \in\mathbb{R} and 1≤j≤q1 \le j \le q.

Each cjc_j lies in Q[t]\mathbb{Q}[t], and deg⁡cj≤2Mj=2j(L−1)\deg c_j \le2Mj = 2j(L - 1) when cj≠0c_j \ne0. These polynomials can be computed from the quiver and its monomial relations, giving a finite test for the three conditions.

Proof. 1. Euler forms and path counts. Let BB be a finite connected convex subcategory of A~\widetilde{A}. Since its quiver is acyclic, we can order its vertices as u1,…,umu_1,\ldots,u_m so that every arrow goes from a smaller index to a larger one. Let (PB)ab(P_B)_{ab} count the nonzero paths from uau_a to ubu_b, and write SaS_a for the simple at uau_a. For the right projective Pa=euaBP_a = e_{u_a}B, the multiplicity of SbS_b is (PB)ab(P_B)_{ab}. Thus the Cartan matrix, whose columns are the projective classes in the basis of simple classes, is CB=PBTC_B = P_B^T. If EBE_B is the Euler matrix in this basis, then ⟨[Pa],[Sb]⟩B=δab\langle[P_a], [S_b]\rangle_B = \delta_{ab} gives CBTEB=IC_B^T E_B = I, hence EB=PB−1E_B = P_B^{-1}.

The matrix PBP_B is upper triangular with diagonal entries 11, so PBTP_B^T is invertible. For any y∈Rmy \in\mathbb{R}^m, set x=PBTyx = P_B^T y. Then

χB(x)=χB(PBTy)=(PBTy)TPB−1PBTy=yTPBPB−1PBTy=yTPBTy=yTPBy,(4)\chi_B(x) = \chi_B(P_B^T y) = (P_B^T y)^T P_B^{-1}P_B^T y = y^T P_B P_B^{-1}P_B^T y = y^T P_B^T y = y^T P_B y, \tag*{(4)}

where yTPBTy=yTPByy^T P_B^T y = y^T P_B y holds since yTPBTyy^T P_B^T y is a scalar. Since y↦PBTyy \mapsto P_B^T y is an automorphism of Rm\mathbb{R}^m, χB(x)≥0\chi_B(x) \ge0 for all xx if and only if yTPBy≥0y^T P_B y \ge0 for all y∈Rmy \in\mathbb{R}^m.

For the relation between the Euler and Cartan forms, see [1 §1.6, proof of Lemma 1].

2. Vectors of finite support. Let y=(yk)k∈Zy = (y_k)_{k\in\mathbb{Z}} be a finitely supported family with yk∈Rqy_k \in\mathbb{R}^q, and set

Φ(y)=∑k,l∈ZykTDl−kyl.\Phi(y) = \sum_{k,l\in\mathbb{Z}} y_k^T D_{l-k}y_l.

Assign (yk)i(y_k)_i to the vertex τkxi\tau^k x_i. Applying τ−k\tau^{-k} gives a bijection from paths between τkxi\tau^k x_i and τlxj\tau^l x_j to paths between xix_i and τl−kxj\tau^{l-k}x_j. By Lemma 1.6, the support of yy lies in a finite connected convex subtree, which contains every path between its vertices. Its path-count form on yy is therefore Φ(y)\Phi(y).

Conversely, every vector on a finite convex subcategory gives such a family. By 3.4, the local Euler condition is equivalent to Φ(y)≥0\Phi(y) \ge0 for all finitely supported real families.

3. The unit-circle condition. Define Hd=12(Dd+D−dT)H_d = \frac{1}{2}(D_d + D_{-d}^T) so that H(z)=∑d∈ZHdzdH(z) = \sum_{d\in\mathbb{Z}} H_d z^d. For any complex sequence y=u+ivy = u + iv with finite support, H−d=HdTH_{-d} = H_d^T yields Ψ(y):=∑k,lyk∗Hl−kyl=Φ(u)+Φ(v)\Psi(y) := \sum_{k,l} y_k^* H_{l-k}y_l = \Phi(u) + \Phi(v). Thus, non-negativity over real sequences is equivalent to non-negativity over complex sequences.

For fy(z)=∑kykz−kf_y(z) = \sum_k y_k z^{-k}, orthogonality of characters on S1S^1 gives

Ψ(y)=12π∫02πfy(eiθ)∗H(eiθ)fy(eiθ) dθ.(5)\Psi(y) = \frac{1}{2\pi}\int_0^{2\pi} f_y(e^{i\theta})^* H(e^{i\theta}) f_y(e^{i\theta})\,d\theta. \tag*{(5)}

If H(z)≥0H(z) \ge0 for all z∈S1z \in S^1, then Ψ(y)≥0\Psi(y) \ge0.

Conversely, assume Ψ(y)≥0\Psi(y) \ge0 for all finitely supported sequences. Given any z0∈S1z_0 \in S^1 and v∈Cqv \in\mathbb{C}^q, define yk=z0kvy_k=z_0^k v for 0≤k<N0\le k<N and yk=0y_k=0 otherwise. Counting pairs with l−k=dl-k=d yields:

0≤Ψ(y)N=∑∣d∣<N(1−∣d∣N)z0dv∗Hdv.0\le\frac{\Psi(y)}{N}=\sum_{\lvert d\rvert<N}\left(1-\frac{\lvert d\rvert}{N}\right)z_0^d v^*H_dv.

Since only finitely many HdH_d are non-zero, taking N→∞N\to\infty gives v∗H(z0)v≥0v^*H(z_0)v\ge0. As z0∈S1z_0\in S^1 and v∈Cqv\in\mathbb{C}^q were arbitrary, H(z)≥0H(z)\ge0 on the entire unit circle S1S^1.

4. A finite polynomial test. For the spanning-tree lifts, Hd=0H_d=0 when ∣d∣>L−1\lvert d\rvert>L-1. With M=L−1M=L-1 and K(t)K(t) as in (2), the terms with exponents dd and −d-d, for 0≤d≤M0\le d\le M, become

(1+t2)Mz(t)±d=(1+t2)M−d(1±it)2d.(1+t^2)^M z(t)^{\pm d}=(1+t^2)^{M-d}(1\pm it)^{2d}.

It follows that the entries of K(t)K(t) lie in Q(i)[t]\mathbb{Q}(i)[t] and have degree at most 2M2M. For all real tt, the factor (1+t2)M(1+t^2)^M is positive and K(t)K(t) is Hermitian.

In (3), the coefficient cj(t)c_j(t) is the sum of the principal minors of order jj of K(t)K(t); equivalently, it is the jjth elementary symmetric function of the eigenvalues of K(t)K(t). Each entry has degree at most 2M2M, so cj(t)∈Q(i)[t]c_j(t)\in\mathbb{Q}(i)[t] and deg⁡cj≤2Mj\deg c_j\le2Mj when cjc_j is nonzero. Since K(t)K(t) is Hermitian for all real tt, its eigenvalues and each cj(t)c_j(t) are real on R\mathbb{R}. The imaginary part of cjc_j is therefore a polynomial vanishing on R\mathbb{R}, so cj∈Q[t]c_j\in\mathbb{Q}[t].

For a fixed real tt, let λ1,…,λq\lambda_1,\ldots,\lambda_q be the eigenvalues of K(t)K(t). If they are all non-negative, then the coefficients of ∏a=1q(s+λa)\prod_{a=1}^q(s+\lambda_a) are non-negative. Conversely, if all cj(t)c_j(t) are non-negative, the right side of (3.3) is strictly positive for s>0s>0. A negative eigenvalue λa\lambda_a would give the positive root s=−λas=-\lambda_a, a contradiction. Thus

K(t)⪰0⟺cj(t)≥0(1≤j≤q).K(t)\succeq0\Longleftrightarrow c_j(t)\ge0\quad(1\le j\le q).

As tt runs through R\mathbb{R}, z(t)z(t) covers the unit circle except for −1-1, and z(t)→−1z(t)\to-1 as t→+∞t\to+\infty. If H(z(t))⪰0H(z(t))\succeq0 for all real tt, then for every v∈Cqv\in\mathbb{C}^q, continuity gives

v∗H(−1)v=lim⁡t→+∞v∗H(z(t))v≥0.v^*H(-1)v=\lim_{t\to+\infty}v^*H(z(t))v\ge0.

Consequently,

H(z)⪰0 for all ∣z∣=1⟺cj(t)≥0 for all t∈R and 1≤j≤q.H(z)\succeq0\text{ for all }\lvert z\rvert=1\Longleftrightarrow c_j(t)\ge0\text{ for all }t\in\mathbb{R}\text{ and }1\le j\le q.

A nonzero real polynomial is non-negative on R\mathbb{R} exactly when its leading coefficient is positive and every real root has even multiplicity. Indeed, an odd-multiplicity real root changes the sign, while an even-multiplicity root does not. The zero polynomial is non-negative as well. For a polynomial in Q[t]\mathbb{Q}[t], these conditions can be checked by square-free factorization and exact real-root isolation. The criterion thus reduces to qq finite polynomial tests with the stated degree bounds. □

Remark 3.2. The choice of lifts does not affect the criterion. If xi′=τsixix_i'=\tau^{s_i}x_i and D(z)=diag⁡(zs1,…,zsq)D(z)=\operatorname{diag}(z^{s_1},\ldots,z^{s_q}), then

P′(z)=D(z)P(z)D(z)−1,H′(z)=D(z)H(z)D(z)∗(∣z∣=1).P'(z)=D(z)P(z)D(z)^{-1},\qquad H'(z)=D(z)H(z)D(z)^*\quad(\lvert z\rvert=1).

Since D(z)D(z) is unitary on the unit circle, H′H' and HH are positive semidefinite at the same points. With M=L−1M=L-1 fixed, similarity also gives

det⁡(sIq+(1+t2)MH′(z(t)))=det⁡(sIq+K(t)).\det\left(sI_q+(1+t^2)^M H'(z(t))\right)=\det\left(sI_q+K(t)\right).

This is an identity of rational functions in tt, so the determinant coefficients are the same polynomials cj(t)c_j(t). The spanning-tree lifts ensure that the entries of K(t)K(t) themselves are polynomials of degree at most 2M2M. Replacing τ\tau by τ−1\tau^{-1} replaces zz by z−1z^{-1} and leaves the unit-circle condition unchanged.

Theorems A and 3.1 together prove Theorem B.

Corollary 3.3. Let A=kQ/IA = \mathbb{k}Q/I be a finite-dimensional connected one-cycle monomial algebra whose unique cycle is not oriented. Using spanning-tree lifts, define

K1(t)=(1+t2)H(1+it1−it),det⁡(sIq+K1(t))=sq+∑j=1qbj(t)sq−j.K_1(t) = (1+t^2)H\left(\frac{1+it}{1-it}\right), \qquad\det(sI_q+K_1(t))=s^q+\sum_{j=1}^{q}b_j(t)s^{q-j}.

Then bj∈Q[t]b_j\in\mathbb{Q}[t] and deg⁡bj≤2j\deg b_j\leq2j whenever bj≠0b_j\ne0. Moreover, AA is derived tame if and only if bj(t)≥0b_j(t)\geq0 for all t∈Rt\in\mathbb{R} and 1≤j≤q1\leq j\leq q.

Proof. The quiver QQ has no oriented cycles, so a directed path visits each vertex at most once. It therefore uses the unique arrow outside the spanning tree at most once. Hence Dd=0D_d=0 for d∉{0,1}d\notin\{0,1\} and P(z)=D0+zD1P(z)=D_0+zD_1. It follows that

K1(t)=12((1+t2)(D0+D0T)+(1+it)2D1+(1−it)2D1T).K_1(t)=\frac{1}{2}\left((1+t^2)(D_0+D_0^{T})+(1+it)^2D_1+(1-it)^2D_1^{T}\right).

Its entries lie in Q(i)[t]\mathbb{Q}(i)[t] and have degree at most two. The coefficient bjb_j is the sum of its principal minors of order jj, so it has degree at most 2j2j when nonzero. Since K1(t)K_1(t) is Hermitian for real tt, these coefficients are real on R\mathbb{R} and therefore belong to Q[t]\mathbb{Q}[t].

For the matrix K(t)K(t) and coefficients cj(t)c_j(t) in Theorem 3.1, we have

K(t)=(1+t2)L−2K1(t),cj(t)=(1+t2)j(L−2)bj(t).K(t)=(1+t^2)^{L-2}K_1(t), \qquad c_j(t)=(1+t^2)^{j(L-2)}b_j(t).

The factors are positive for real tt, so the two sets of polynomial inequalities are equivalent. Theorem B now gives the result. □

Applications to one-cycle monomial algebras

We first apply the criterion to truncated cycles and to path algebras of acyclic orientations of a cycle. We then consider a string algebra with a minimal cubic relation, give a second proof of its derived tameness by flat deformation, and construct its finite cyclic covers.

Proposition 4.1. [3 Lemma 3.2] Let QQ be an oriented cycle on n≥1n\geq1 vertices and let An,r=kQ/RQrA_{n,r}=\mathbb{k}Q/R_Q^r, where r≥2r\geq2. Then An,rA_{n,r} is derived tame if and only if r=2r=2.

Proof. The universal cover is the oriented infinite line with all paths of length rr zero, independently of nn. By Theorem A, it suffices to take n=1n=1. In that case the matrix has one entry:

Hr(eiθ)=∑j=0r−1cos⁡(jθ).H_r(e^{i\theta})=\sum_{j=0}^{r-1}\cos(j\theta).

For r=2r=2 this equals 1+cos⁡θ≥01+\cos\theta\geq0. For r>2r>2, choose π/(r−1)<θ<2π/r\pi/(r-1)<\theta<2\pi/r; then

Hr(eiθ)=sin⁡(rθ/2)cos⁡((r−1)θ/2)sin⁡(θ/2)<0.H_r(e^{i\theta})=\frac{\sin(r\theta/2)\cos((r-1)\theta/2)}{\sin(\theta/2)}<0.

The result follows from Theorem B. □

For r=3r=3, the polynomial test in Theorem 3.1 with M=2M=2 gives

(1+t2)2H3(1+it1−it)=t4−4t2+3=(t2−1)(t2−3).(1+t^2)^2H_3\left(\frac{1+it}{1-it}\right)=t^4-4t^2+3=(t^2-1)(t^2-3).

This polynomial is negative for 1<t2<31 < t^2 < 3, although H3(1)=3>0H_3(1) = 3 > 0. Thus a test at z=1z = 1 alone does not decide positive semidefiniteness on the unit circle.

We next consider a cycle whose arrows do not form an oriented cycle, with no relations.

Proposition 4.2. [11 §1.4(a)] Let QQ be an acyclic orientation of a cycle. Then kQkQ is derived tame.

Proof. Put A=kQA = kQ and q=∣Q0∣q = |Q_0|. Choose a spanning tree of the underlying graph. Give the unique arrow outside this tree weight 1, and give every tree arrow weight 0, as in § 5. Define

N(z)ij=∑a:i→jzw(a).N(z)_{ij} = \sum_{a:i\to j} z^{w(a)}.

The entries of N(z)rN(z)^r count paths of length rr, with each path contributing zz raised to its total weight.

Since QQ has no oriented cycles, every path has length less than qq. Thus AA is finite-dimensional and N(z)q=0N(z)^q = 0. There are no relations, so the path-count matrix is

P(z)=Iq+N(z)+⋯+N(z)q−1=(Iq−N(z))−1.P(z) = I_q + N(z) + \cdots+ N(z)^{q-1} = (I_q - N(z))^{-1}.

Fix zz with ∣z∣=1|z| = 1, and write R(z)=Iq−N(z)R(z) = I_q - N(z). Then R(z)R(z) is invertible and

H(z)=12(P(z)+P(z)∗),H(z) = \frac{1}{2}(P(z) + P(z)^*),

where ∗* denotes conjugate transpose. Hence

2R(z)∗H(z)R(z)=R(z)∗P(z)R(z)+R(z)∗P(z)∗R(z)=R(z)∗+R(z)=2Iq−N(z)−N(z)∗.\begin{aligned} 2R(z)^*H(z)R(z) &= R(z)^*P(z)R(z) + R(z)^*P(z)^*R(z) \\ &= R(z)^* + R(z) \\ &= 2I_q - N(z) - N(z)^*. \end{aligned}

Every vertex of the underlying graph has degree two. Consequently, for x∈Cqx \in\mathbb{C}^q,

x∗(2Iq−N(z)−N(z)∗)x=∑a:i→j∣xi−zw(a)xj∣2.x^*(2I_q - N(z) - N(z)^*)x = \sum_{a:i\to j} \left|x_i - z^{w(a)}x_j\right|^2.

This proves that 2Iq−N(z)−N(z)∗2I_q - N(z) - N(z)^* is positive semidefinite. Since R(z)R(z) is invertible, H(z)H(z) is positive semidefinite for every ∣z∣=1|z| = 1. It shows that AA is derived tame by Theorem B. □

Remark 4.3. In the setting of Proposition 4.2, the matrix H(z)H(z) in (3.1) is positive definite for ∣z∣=1|z| = 1 and z≠1z \ne1, while H(1)H(1) has corank one.

Indeed, suppose that the sum of squares in the proof vanishes. Every tree arrow has weight zero, so its endpoints have equal coordinates. Since the spanning tree is connected, all coordinates of xx have a common value cc. The remaining arrow gives c=zcc = zc. If z≠1z \ne1, then c=0c = 0, so 2Iq−N(z)−N(z)∗2I_q - N(z) - N(z)^* is positive definite. If z=1z = 1, its kernel consists of the constant vectors and has dimension one. The invertible congruence used in the proof preserves positive definiteness and the dimension of the kernel, so both conclusions hold for H(z)H(z).

In contrast to Proposition 4.1, where cubic relations on an oriented cycle yield derived wildness, a cubic relation located along a tree branch outside the central cycle does not destroy derived tameness. The following example illustrates this phenomenon.

Example 4.4. Let QQ be the quiver

Quiver with vertices 1 through 13 and arrows labeled $\alpha_i$, $\beta_i$, $\gamma_i$, and $\delta_i$

Figure 3.

With α\alpha-subscripts read modulo 6, put

A=kQ/⟨αiαi+1 (1≤i≤6), β1β2β3⟩.A = kQ/\langle\alpha_i\alpha_{i+1}\ (1\leq i\leq6),\ \beta_1\beta_2\beta_3\rangle.

Then AA is a derived-tame one-cycle string algebra with a minimal cubic relation.

Proof. The graph has 13 vertices and 13 edges, with at most two arrows entering and at most two arrows leaving each vertex. Its nonzero length-two paths are

L2∪{β2β3},L2={α6β1, β1β2, α2γ1, γ1γ2, δ2δ1, δ1α5}.\mathcal{L}_2\cup\{\beta_2\beta_3\},\qquad\mathcal{L}_2=\{\alpha_6\beta_1,\ \beta_1\beta_2,\ \alpha_2\gamma_1,\ \gamma_1\gamma_2,\ \delta_2\delta_1,\ \delta_1\alpha_5\}.

In this list, each arrow has at most one nonzero successor and at most one nonzero predecessor, so AA is a string algebra. Every path of length four is zero. Both length-two subpaths of β1β2β3\beta_1\beta_2\beta_3 are nonzero, so this is a minimal cubic relation. Thus the ideal is admissible and the algebra is not quadratic.

Proof by the matrix criterion. We compute the matrix in Theorem 3.1. Give α6\alpha_6 weight 1 and all other arrows weight 0. In the vertex order (1,…,6∣7,…,13)(1,\ldots,6\mid7,\ldots,13), write

2H(z)=(W(z)U(z)U(z)∗V),∣z∣=1.2H(z)= \begin{pmatrix} W(z) & U(z)\\ U(z)^* & V \end{pmatrix}, \qquad|z|=1.

The only nonzero paths between cycle vertices are the trivial paths and the arrows αi\alpha_i. Thus WW has diagonal 2, entries Wi,i+1=Wi+1,i=1W_{i,i+1}=W_{i+1,i}=1 for 1≤i≤51\leq i\leq5, W6,1=zW_{6,1}=z, W1,6=z−1W_{1,6}=z^{-1}, and all other entries zero. Write JmJ_m for the all-ones matrix and uiu_i for the standard basis vectors of C6\mathbb{C}^6. The remaining blocks are

V=diag⁡(I3+J3,I2+J2,I2+J2),b=(1,1,0),c=(1,1),V=\operatorname{diag}(I_3+J_3,I_2+J_2,I_2+J_2),\qquad b=(1,1,0),\qquad c=(1,1),
U(z)=((u1+zu6)b∣(u2+u3)c∣(u5+u6)c).U(z)=\bigl((u_1+zu_6)b\mid(u_2+u_3)c\mid(u_5+u_6)c\bigr).

The blocks of VV come from the paths within the three branches. For example, the branch on 7, 8, 9 has the paths β2\beta_2, β3\beta_3 and β2β3\beta_2\beta_3, so its block is I3+J3I_3+J_3. The first block of UU counts the paths from 1 and 6 to 7 and 8; the relation β1β2β3=0\beta_1\beta_2\beta_3=0 excludes paths from either vertex to 9. The second block counts the paths from 2 and 3 to 10 and 11. The third block comes from paths in the opposite direction, from 12 and 13 to 5 and 6, through the term P(z−1)TP(z^{-1})^T in 2H(z)2H(z). All paths contributing to the second and third blocks have weight zero.

The matrix Im+JmI_m+J_m has eigenvalue m+1m+1 once and eigenvalue 1 with multiplicity m−1m-1, and

(Im+Jm)−1=Im−1m+1Jm.(I_m+J_m)^{-1}=I_m-\frac{1}{m+1}J_m.

Thus VV is positive definite. The Schur complement of VV is S=W−UV−1U∗S=W-UV^{-1}U^*. For x∈C6x\in\mathbb{C}^6 and y∈C7y\in\mathbb{C}^7, we have

(xy)∗(WUU∗V)(xy)=x∗Sx+(y+V−1U∗x)∗V(y+V−1U∗x).\begin{pmatrix}x\\y\end{pmatrix}^{*} \begin{pmatrix}W&U\\U^*&V\end{pmatrix} \begin{pmatrix}x\\y\end{pmatrix} =x^*Sx+(y+V^{-1}U^*x)^*V(y+V^{-1}U^*x).

Thus it remains to check that SS is positive semidefinite. The identities

b(I3+J3)−1bT=1,c(I2+J2)−1cT=23b(I_3+J_3)^{-1}b^T=1,\qquad c(I_2+J_2)^{-1}c^T=\frac{2}{3}

give the following contributions to UV−1U∗UV^{-1}U^*: (1z−1z1)\left(\begin{smallmatrix}1&z^{-1}\\ z&1\end{smallmatrix}\right) on indices (1,6)(1,6), and 23J2\frac{2}{3}J_2 on each of (2,3)(2,3) and (5,6)(5,6). After subtraction from WW, all entries containing zz cancel. Thus

x∗Sx=∣x1+x2∣2+13∣x2+x3∣2+∣x3+x4∣2+∣x4+x5∣2+13∣x5+x6∣2≥0,x∈C6.x^*Sx=|x_1+x_2|^2+\frac{1}{3}|x_2+x_3|^2+|x_3+x_4|^2+|x_4+x_5|^2+\frac{1}{3}|x_5+x_6|^2\geq0,\quad x\in\mathbb{C}^6.

Thus H(z)⪰0H(z)\succeq0 on the entire unit circle, and AA is derived tame by Theorem B.

Proof by a flat deformation. We adapt the local construction in [11 Lemma 2.4] and verify flatness directly. Add an arrow θ:7→9\theta:7\to9 to obtain Q+Q^+, and set

A=k[t]Q+/⟨αiαi+1, β2β3−tθ, β1θ⟩.(6)\mathcal{A}=k[t]Q^+/\langle\alpha_i\alpha_{i+1},\ \beta_2\beta_3-t\theta,\ \beta_1\theta\rangle. \tag*{(6)}

A k[t]k[t]-basis is

{e1,…,e13}∪Q1+∪L2∪{α6β1β2, α2γ1γ2, δ2δ1α5}.\{e_1,\ldots,e_{13}\}\cup Q_1^+\cup L_2\cup\{\alpha_6\beta_1\beta_2,\ \alpha_2\gamma_1\gamma_2,\ \delta_2\delta_1\alpha_5\}.

To verify this basis, use the rules αiαi+1↦0\alpha_i\alpha_{i+1}\mapsto0, β2β3↦tθ\beta_2\beta_3\mapsto t\theta and β1θ↦0\beta_1\theta\mapsto0. The only arrow entering vertex 77 is β1\beta_1, and no arrow leaves vertex 99. Thus every path of length at least two that contains θ\theta has β1θ\beta_1\theta as a subpath and reduces to zero. Every reduction either gives zero or shortens the path, so the reduction process terminates.

The only overlapping left sides occur in αiαi+1αi+2\alpha_i\alpha_{i+1}\alpha_{i+2}, and reducing either pair gives zero. Reductions on disjoint subpaths commute. Also, β1β2β3\beta_1\beta_2\beta_3 reduces first to tβ1θt\beta_1\theta and then to zero. Hence the displayed paths are exactly the irreducible paths, and every element has a unique normal form as a k[t]k[t]-linear combination of them. Since there are 13+14+6+3=3613+14+6+3=36 such paths, A\mathcal{A} is free of rank 3636 over k[t]k[t] and the family is flat.

For λ≠0\lambda\neq0, eliminating θ=λ−1β2β3\theta=\lambda^{-1}\beta_2\beta_3 identifies A/(t−λ)A\mathcal{A}/(t-\lambda)\mathcal{A} with AA. The zero fibre is

A0=kQ+/⟨αiαi+1, β2β3, β1θ⟩.A_0=kQ^+/\langle\alpha_i\alpha_{i+1},\ \beta_2\beta_3,\ \beta_1\theta\rangle.

The zero fibre A0A_0 is gentle: at most two arrows start and at most two end at each vertex, and its nonzero paths of length two are precisely those in L2L_2. This list and the displayed relations give each arrow at most one nonzero and one zero successor, with the same bounds for predecessors. Since A0A_0 is derived tame [4] and every nonzero fibre is isomorphic to AA, [2 Corollary 1.5] implies that AA is derived tame.

Remark 4.5

Let AA be a one-cycle monomial algebra. Write C0C_0 for the vertices on its unique cycle and put F0=Q0∖C0F_0=Q_0\setminus C_0. Assume F0≠∅F_0\neq\varnothing, and choose the arrow of weight 11 on the cycle. Suppose that no nonzero path with both endpoints in F0F_0 passes through this arrow. With the vertices ordered as (C0,F0)(C_0,F_0), write

2H(z)=(W(z)U(z)U(z)∗V).2H(z)=\begin{pmatrix}W(z)&U(z)\\ U(z)^*&V\end{pmatrix}.

Every path contributing to VV has weight zero, so VV is a constant real symmetric matrix. If VV is positive definite, the completion of the square used in Example 4.4 gives

H(z)⪰0⟺W(z)−U(z)V−1U(z)∗⪰0(∣z∣=1).H(z)\succeq0\quad\Longleftrightarrow\quad W(z)-U(z)V^{-1}U(z)^*\succeq0\qquad(|z|=1).

This reduces the test to a matrix of order ∣C0∣|C_0|.

To check positive definiteness of VV, suppose further that every nonzero path with both endpoints in F0F_0 stays in F0F_0, and let BB be the algebra of the induced bound subquiver on F0F_0. Then BB is a product of finite tree algebras and V=PB+PBTV=P_B+P_B^T. By (4), applied to each component, VV is positive definite if and only if the Euler form of BB is positive definite. In particular, this holds if each component is a directed path with no relations: then VV is a direct sum of matrices Im+JmI_m+J_m. These conditions hold in Example 4.4.

Finite cyclic covers preserve derived tameness. Applying the following proposition to Example 4.4 gives an infinite family.

Proposition 4.6. Let AA be a finite-dimensional connected one-cycle monomial algebra, with universal cover A~\widetilde{A} and deck group ⟨τ⟩\langle\tau\rangle. For n≥1n \ge1, let A(n)=A~/⟨τn⟩A^{(n)} = \widetilde{A}/\langle\tau^n \rangle. Then A(n)A^{(n)} is a finite-dimensional connected one-cycle monomial algebra, and

A(n) is derived tame⟺A is derived tame.A^{(n)}\text{ is derived tame}\quad\Longleftrightarrow\quad A\text{ is derived tame}.

Proof. The quotient quiver is connected and has n∣Q0∣n|Q_0| vertices and n∣Q1∣n|Q_1| arrows. Its relations are the lifts of the monomial relations of AA, so every nonzero path has length less than LL and A(n)A^{(n)} is finite-dimensional and monomial. The identity n∣Q1∣−n∣Q0∣+1=1n|Q_1|-n|Q_0|+1=1 shows that it is one-cycle.

The projection A~→A(n)\widetilde{A}\to A^{(n)} is a bound-quiver covering with deck group ⟨τn⟩\langle\tau^n\rangle. Its source has a tree as its underlying graph and is therefore simply connected. Hence it is the universal cover of A(n)A^{(n)}. Thus AA and A(n)A^{(n)} have the same local Euler condition, and Theorem A gives the equivalence. □

Corollary 4.7. For each n≥1n\ge1, there is a derived-tame one-cycle string algebra with 13n13n vertices, dimension 36n36n, and nn minimal cubic relations. No two algebras in this family are derived equivalent.

Proof. Let AA be the algebra in Example 4.4, and take A(n)A^{(n)} as in Proposition 4.6. An explicit presentation has vertices

(j,i),j∈Z/nZ,1≤i≤13.(j,i),\qquad j\in\mathbb{Z}/n\mathbb{Z},\quad1\le i\le13.

For each jj, copy all arrows of the example except α6\alpha_6, and replace that arrow by

αj,6:(j,6)⟶(j+1,1).\alpha_{j,6}:(j,6)\longrightarrow(j+1,1).

Write αj,i\alpha_{j,i}, βj,i\beta_{j,i}, γj,i\gamma_{j,i} and δj,i\delta_{j,i} for the resulting arrows. The relations are

αj,iαj,i+1=0(1≤i≤5),αj,6αj+1,1=0,βj,1βj,2βj,3=0(j∈Z/nZ).\alpha_{j,i}\alpha_{j,i+1}=0\quad(1\le i\le5),\qquad \alpha_{j,6}\alpha_{j+1,1}=0,\qquad \beta_{j,1}\beta_{j,2}\beta_{j,3}=0\quad(j\in\mathbb{Z}/n\mathbb{Z}).

The covering preserves the number of arrows at each vertex and the allowed continuations of each arrow, so A(n)A^{(n)} is a string algebra. Both length-two subpaths of each cubic relation are nonzero, giving nn distinct minimal cubic relations. Every path of length four is zero.

Each nonzero path of AA, including a trivial path, has exactly nn lifts to this quiver. Thus dim⁡kA(n)=ndim⁡kA=36n\dim_k A^{(n)}=n\dim_k A=36n. Derived tameness follows from Proposition 4.6. Since derived equivalence preserves the rank of the Grothendieck group, the equality rank⁡K0(A(n))=13n\operatorname{rank}K_0(A^{(n)})=13n distinguishes the derived equivalence classes for different values of nn. □

Remark 4.8. The construction in Corollary 4.7 is entirely general. If AA is any derived tame one-cycle monomial algebra with dd vertices, the sequence of finite cyclic coverings A(n)=A~/⟨τn⟩A^{(n)}=\widetilde{A}/\langle\tau^n\rangle yields an infinite family of pairwise non-derived equivalent derived tame algebras, distinguished by rank⁡K0(A(n))=nd\operatorname{rank}K_0(A^{(n)})=nd. Consequently, every derived tame one-cycle monomial algebra gives rise to an infinite family of this type.

References

Department of Mathematics and Applied Mathematics, Federal University of Lavras (UFLA) – Brazil

Email address: [email protected]

School of Mathematics and Statistics, Shaanxi Normal University, Xi’an 710062, China

Email address: [email protected]

References

  1. [1][BBD00] M. Barot, T. Brüstle and J. A. de la Peña, Derived-tame tree algebras of type E. Forum Math. 12 (2000), no. 6, 713–721. doi:10.1515/form.2000.024.DOI
  2. [2][BD03] V. Bekkert and Y. Drozd, Tame–wild dichotomy for derived categories, preprint (2003), arXiv:math/0310352.DOI
  3. [3][BGV21] V. Bekkert, H. Giraldo and J. A. Vélez-Marulanda, Derived tame Nakayama algebras. Proc. Amer. Math. Soc. 149 (2021), no. 11, 4555–4567. doi:10.1090/proc/15612.DOI
  4. [4][BM03] V. Bekkert and H. A. Merklen, Incomposables in derived categories of gentle algebras. Algebra. Represent. Theory 6 (2003), 285–302. doi:10.1023/A:1025142023594.DOI
  5. [5][BR87] M. C. R. Butler and C. M. Ringel, Auslander–Reiten sequences with few middle terms and applications to string algebras. Comm. Algebra 15 (1987), no. 1–2, 145–179. doi:10.1080/009278780823416.DOI
  6. [6][Bru01] T. Brüstle, Derived-tame tree algebras. Compositio Math. 129 (2001), no. 3, 301–323. doi:10.1023/A:1012591326777.DOI
  7. [8][Dro01] Y. A. Drozd, Reduction algorithm and representations of boxes and algebras. C. R. Math. Acad. Sci. Soc. R. Can. 23 (2001), no. 4, 97–125.
  8. [9][Fon22] M. P. Fonseca, Derived tame quadratic string algebras. Comm. Algebra 50 (2022), no. 4, 1662–1684. doi:10.1080/00927872.2021.1986832.DOI
  9. [10][GK02] C. Geiss and H. Krause, On the notion of derived tameness. J. Algebra Appl. 1 (2002), no. 2, 133–157. doi:10.1142/S0219498802000112.DOI
  10. [11][NdP01] C. Novoa and J. A. de la Peña, On the tameness of trivial extensions of monomial algebras. Ann. Sci. Math. Québec 25 (2001), no. 2, 199–209.
  11. [12][Gei02] C. Geiss, Derived tame algebras (with an appendix by the author and B. Keller). Mathematische Zeitschrift 239 (2002), no. 4, 829–862. doi:10.1007/s002090100349.

Paper details

vlnc:W_dcwjpxj8
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents