Homological dimensions over almost gentle algebras Thanks: ∗ Corresponding author. Thanks: MSC2020: 16G10; 16E10; 16E65; 18G20; Thanks: Keywords: Almost gentle algebras, global dimension; self-injective dimension; projective dimension, Auslander–Reiten conjecture. Demin Wang, Zhaoyong Huang, Yu-Zhe Liu ∗ Abstract . We provide a method for computing the global dimension and self-injective dimension of almost gentle algebras, and prove that an almost gentle algebra is Gorenstein if it satisfies the Auslander condition. 1. Introduction In [ 2 ] , Assem and Skowroński introduced gentle algebras and use them to study tilted algebras and derived equivalence. Afterwards, (skew-)gentle algebras play an important role in representation theory; and theirs module categories and derived categories were described by using string algebras, Auslander–Reiten theory, combinatorial methods and geometric models, see [ 9 , 7 , 1 , 8 , 20 , 23 , 21 ] and so on. In particular, the global dimension and self-injective dimension of (locally) gentle algebras have been studied, and (locally) gentle algebras were shown to be Gorenstein by using different methods [ 13 , 17 , 12 ] . As a generalization of gentle algebras, almost gentle algebras are important finite-dimensional algebras introduced by Green and Schroll [ 15 ] , which are monomial special multiserial algebras, and are closely related to some hypergraphs and Brauer configuration algebras. In [ 11 ] and [ 16 ] , complexes on almost gentle algebras and the trivial extension of almost gentle algebras were studied, respectively. Based on the above, we have the following questions. Question 1.1 . (1) How to characterize the global dimension and self-injective dimension of almost gentle algebras? (2) Whether or when are almost gentle algebras Gorenstein? Let \mathds{k} be an algebraically closed field. In this paper, we provide a method for computing the global dimension and self-injective dimension of almost gentle algebras by using forbidden paths introduced by Avella-Alaminos and Geiß [ 5 ] (see Section 3.1 ). We define claw, written as \kappa , and anti-claw, written as \xi (Definition 3.1 ), to describe indecomposable projective modules and injective modules, and introduce tow sets \mathcal{F}(v) and \mathcal{F}(\xi) whose elements are some special forbidden paths on the bound quiver (\mathcal{Q},\mathcal{I}) of an almost gentle algebra A=\mathds{k}\mathcal{Q}/\mathcal{I} , which are decided by some claws, where v is any vertex of the quiver \mathcal{Q} , see Notations 3.11 , 4.15 , and Definition 4.18 . By [ 14 , Corollary 2.4] , we have that the finitistic dimension conjecture holds true for monomial algebras. Observe that an algebra is monomial if and only if so is its opposite algebra. It then follows from [ 3 , Proposition 6.10] that the left and right self-injective dimensions of a monomial algebra, and hence, of an almost gentle algebra A , are identical. For simplicity, we call this common quantity the self-injective dimension of A . The global dimension \mathrm{gl.dim}A and self-injective dimension \mathrm{inj.dim}A of an almost gentle algebra A can be described by the following results. Theorem 1.2 . (1) (Theorem 3.13 ) \mathrm{gl.dim}A=\sup_{F\in\mathcal{F}}\ell(F), where \mathcal{F}:=\bigcup\limits_{v\in\mathcal{Q}_{0}}\mathcal{F}(v) and \ell(F) is the length of F . (2) (Theorem 4.20 ) \mathrm{inj.dim}A=\sup_{F\in\mathcal{F}_{\mathrm{a}}}\ell(F), where \mathcal{F}_{\mathrm{a}}:=\bigcup_{v\in\mathcal{Q}_{0}}\mathcal{F}(\xi_{v}) with \xi_{v} the anti-claw corresponding to the indecomposable injective module E(v) . Theorem 1.2 (2) provides a method for judging the Gorensteiness of almost gentle algebras as follows. Theorem 1.3 . For an almost gentle algebra A=\mathds{k}\mathcal{Q}/\mathcal{I} , the following statements are equivalent. (1) \mathrm{inj.dim}A=\infty . (2) (A direct consequence of Theorem 4.20 ) There exists an anti-claw \xi corresponding to some indecomposable injective module such that \mathcal{F}(\xi) contains a forbidden cycle whose length is infinite. (3) (Theorem 5.4 ) (\mathcal{Q},\mathcal{I}) contains an oriented cycle \mathscr{C}=a_{0}a_{1}\cdots a_{\ell-1}= \ell-2 \ell-1 1 2 3 a_{\ell-2} a_{\ell-1} a_{1} a_{2} such that a_{\overline{i}}a_{\overline{i+1}}\in\mathcal{I} ( for any x\in\mathbb{N} , \overline{x} is defined as x modulo \ell ) , and there is a vertex v on \mathscr{C} such that one of the following condition holds. (A) there is an arrow \alpha ( \neq a_{\overline{v-1}} ) ending at v satisfying \alpha a_{v}\in\mathcal{I} ; (B) there is an arrow \beta ( \neq a_{v} ) starting at v satisfying a_{\overline{v-1}}\beta\in\mathcal{I} . The oriented cycle as above is said to be a forbidden cycle. (4) (Theorem 5.1 (1)) There is a vertex v on some forbidden cycle \mathscr{C} such that v is not invalid ( see Definition 4.11 for the definition of invalid vertices ) . Recall that a left and right Noetherian ring R is called Gorenstein if its left and right self-injective dimensions are finite, and R is said to satisfy the Auslander condition if the flat dimension of the i -th term in a minimal injective coresolution of R as a left R -module is at most i-1 for any i\geqslant 1 . The Auslander condition is left and right symmetric [ 10 , Theorem 3.7] . Bass [ 6 ] proved that a commutative Noetherian ring R is Gorenstein if and only if it satisfied the Auslander condition. Based on it, Auslander and Reiten [ 4 ] conjectured that an Artin algebra is Gorenstein if it satisfies the Auslander condition. We call this conjecture ARC for short. It is situated between the well known Nakayama conjecture and the generalized Nakayama conjecture [ 4 , p.2] . All these conjectures remain still open. As an application of the above results, we obtain the following result. Theorem 1.4 . (1) (Corollary 5.6 ) For an almost gentle algebra A=\mathds{k}\mathcal{Q}/\mathcal{I} , the projective dimension of the injective envelope of A_{A} is infinite. (2) (Theorem 5.7 ) ARC holds true for almost gentle algebras. 2. Preliminaries Throughout this paper, assume that \mathcal{Q}=(\mathcal{Q}_{0},\mathcal{Q}_{1},\mathfrak{s},\mathfrak{t}) is a finite connected quiver. Here, \mathcal{Q}_{0} and \mathcal{Q}_{1} respectively are the vertex and arrow sets of \mathcal{Q} , and \mathfrak{s} and \mathfrak{t} are functions from \mathcal{Q}_{1} to \mathcal{Q}_{0} respectively sending each arrow to its source and sink. For arbitrary two arrows \alpha and \beta of the quiver \mathcal{Q} , if \mathfrak{t}(\alpha)=\mathfrak{s}(\beta) , then the composition of \alpha and \beta is denoted by \alpha\beta . For any v\in\mathcal{Q}_{0} , we use e_{v} to denote the path of length zero corresponding to v . For a path on \mathcal{Q} , we use \ell(p)=n to denote the length of p . Let A be a finite-dimensional algebra. We use \mathsf{mod}(A) to denote the category of finitely generated right A -modules. For arbitrary two modules M and N , if M is a direct summand of N , then we write M\leq_{\oplus}N . Moreover, we use S(v_{0}),P(v_{0}) and E(v_{0}) to denote the simple, indecomposable projective and indecomposable injective modules corresponding to the vertex v_{0}\in\mathcal{Q}_{0} , respectively. 2.1. Almost gentle algebras Recall from [ 15 ] that a bound quiver (\mathcal{Q},\mathcal{I}) is called an almost gentle pair if it satisfies the following conditions: (1) I , the admissible ideal of the path algebra \mathds{k}\mathcal{Q}:=\mathrm{span}_{\mathds{k}}(\mathcal{Q}_{l}\mid l\in\mathbb{N}) , is a subspace of the \mathds{k} -linear space \mathds{k}\mathcal{Q} which is generated by some paths on \mathcal{Q} of length two; (2) for any arrow a\in\mathcal{Q}_{1} , there is at most one arrow b\in\mathcal{Q}_{1} such that ab\notin\mathcal{I} and at most one arrow c\in\mathcal{Q}_{1} such that ca\notin\mathcal{I} . Definition 2.1 . [ 15 ] A finite-dimensional algebra A is called an almost gentle algebra if its bound quiver is an almost gentle pair. Example 2.2 . Let \mathcal{Q} be the quiver shown in F IGURE 2.1 , and let \mathcal{I} be the admissible ideal of \mathds{k}\mathcal{Q} given by the paths a_{1,2_{\mathrm{L}}}a_{2_{\mathrm{L}},3_{\mathrm{L}}} , a_{2,3}a_{3,4} , a_{1,2}a_{2,3} , a_{3,4}a_{4,5} , a_{2,3^{\prime}}a_{3^{\prime},4} , a_{2_{\mathrm{R}},3_{\mathrm{R}}}a_{3_{\mathrm{R}},5} , a_{1,2}a_{2,3^{\prime}} , a_{3^{\prime},4}a_{4,5} , a_{1,2}a_{2,4} , a_{3_{\mathrm{L}},4_{\mathrm{L}}}b_{4_{\mathrm{L}},5} , b_{1,2_{\mathrm{R}}}a_{2_{\mathrm{R}},3_{\mathrm{R}}} . 1 2 3 3^{\prime} 4 5 2_{\mathrm{L}} 3_{\mathrm{L}} 4_{\mathrm{L}} 2_{\mathrm{R}} 3_{\mathrm{R}} 4_{\mathrm{R}} a_{4,5} a_{2,4} a_{1,2} a_{2,3} a_{2,3^{\prime}} a_{3,4} a_{3^{\prime},4} a_{1,2_{\mathrm{R}}} b_{1,2_{\mathrm{R}}} a_{1,2_{\mathrm{L}}} a_{2_{\mathrm{L}},3_{\mathrm{L}}} a_{3_{\mathrm{L}},4_{\mathrm{L}}} a_{4_{\mathrm{L}},5} a_{2_{\mathrm{R}},3_{\mathrm{R}}} a_{3_{\mathrm{R}},4_{\mathrm{R}}} a_{4_{\mathrm{R}},5} b_{4_{\mathrm{L}},5} Figure 2.1. An almost gentle pair Then (\mathcal{Q},\mathcal{I}) is an almost gentle pair and A=\mathds{k}\mathcal{Q}/\mathcal{I} is an almost gentle algebra. 2.2. Strings and string modules For any arrow a\in\mathcal{Q}_{1} , Butler and Ringel [ 9 ] introduced the formal inverse a , written as a^{-1} , and, naturally, define \mathfrak{t}(a^{-1})=\mathfrak{s}(a) and \mathfrak{s}(a^{-1})=\mathfrak{t}(a) . We denote by \mathcal{Q}_{1}^{-1}=\{a^{-1}\mid a\in\mathcal{Q}_{1}\} the set of all formal inverses of arrows. Then any path p=a_{1}a_{2}\cdots a_{n} in (\mathcal{Q},I) naturally provides a formal inverse path p^{-1}=a_{n}^{-1}a_{n-1}^{-1}\cdots a_{1}^{-1} of p . For any path p , we define (p^{-1})^{-1}=p , and for any path e_{v} of length zero corresponding to v\in\mathcal{Q}_{0} , we define e_{v}^{-1}=e_{v} . In a bound quiver (\mathcal{Q},\mathcal{I}) of a finite-dimensional algebra A , a string (of length m ) on (\mathcal{Q},\mathcal{I}) is a sequence s=(a_{1},\ldots,a_{m}) such that • each a_{i} ( 1\leqslant i\leqslant m ) is either an arrow or a formal inverse of the arrow; • if a_{i}\in\mathcal{Q}_{1} and a_{i+1}\in\mathcal{Q}_{1}^{-1} , then a_{i}\notin a_{i+1}^{-1} ; • \mathfrak{t}(a_{i})=\mathfrak{s}(a_{i+1}) holds for all 1\leqslant i\leqslant n-1 . In particular, a string of length zero is said to be a simple string ; and a string which is one of the forms \bullet\longrightarrow\bullet\longrightarrow\cdots\longrightarrow\bullet and \bullet\longleftarrow\bullet\longleftarrow\cdots\longleftarrow\bullet (length \geqslant 0 ) is said to be a directed string ; and we define 0 , the zero vector in \mathds{k}\mathcal{Q} , is also a string which is said to be a trivial string . Two strings s_{1} and s_{2} are called equivalent if s_{1}=s_{2}^{-1} or s_{1}=s_{2} . We denote by \mathrm{Str}(A) the set of all equivalent classes of strings. Now, similar to [ 9 ] , we define a string module with respect to a given string. Definition 2.3 (String modules) . Suppose that s=a_{1}a_{2}\cdots a_{n} is a string on \mathcal{Q} , where a_{1},\ldots,a_{n} are arrows such that \mathfrak{s}(a_{i})=v_{i} ( 1\leqslant i\leqslant n ) and \mathfrak{t}(a_{n})=v_{n+1} . It induces a string module , denote it by \mathds{M}(s) , which satisfies the following conditions: • for v\in\mathcal{Q}_{0} , we have that \dim_{k}(\mathds{M}(s)e_{v}) equals to the multiplicities of s transverses v ; • any a_{i}\in\mathcal{Q}_{1} on s provides an identity from \mathds{k}_{i} to \mathds{k}_{i+1} , where \mathds{k}_{i} and \mathds{k}_{i+1} are copies of \mathds{k} which, as two \mathds{k} -vector spaces, are direct summands of \mathds{M}(s)e_{v_{i}} and \mathds{M}(s)e_{v_{i+1}} , respectively; • any arrow a\in\mathcal{Q}_{1} which does not on s provides a zero action \mathds{M}(s)a=0 . For a directed string s=a_{1}a_{2}\cdots a_{n} with a_{1},a_{2},\ldots,a_{n}\in\mathcal{Q}_{1} , if either \mathfrak{t}(s) is a sink or for each arrow \alpha with \mathfrak{s}(\alpha)=\mathfrak{t}(s) , and the concatenation s\alpha=0 , then s is said to be a right maximal directed string . Dually, a left maximal directed string is defined. Furthermore, the string module induced by a (right maximal) directed string s is said to be a ( right maximal ) directed string module . Remark 2.4 . (1) A band (of length m ) is a string b=(b_{1},\ldots,b_{m}) with \mathfrak{t}(b_{m})=\mathfrak{s}(b_{1}) such that • b^{2} is a string of length 2m ; • b is not a non-trivial power of some string, that is, there is no string s such that b=s^{n} for some n\geqslant 2 . Two bands b=b_{1}\cdots b_{m} and b^{\prime}=b_{1}^{\prime}\cdots b_{m}^{\prime} are called equivalent if either b[t]=b^{\prime} or b[t]^{-1}=b^{\prime} , where b[t]=b_{1+t}b_{2+t}\cdots b_{n-1}b_{n}b_{1}\cdots b_{t} . We denote by \mathrm{Ban}(A) the set of all equivalent classes of bands on the bound quiver of A . Suppose that b=b_{1}b_{2}\cdots b_{n} is a band on \mathcal{Q} , where \mathfrak{s}(a_{i})=v_{i} ( 1\leqslant i\leqslant n ). For any u\in\mathbb{N}^{+} and 0\neq\lambda\in\mathds{k} , it induces a band module , denote it by \mathds{M}(s,J_{n}(\lambda)) , which satisfies the following conditions: • for v\in\mathcal{Q}_{0} , we have \dim_{k}(\mathds{M}(s)e_{v})=tu , where u is the multiplicities of s transverses v ; • any b_{i}\in\mathcal{Q}_{1} with 1t , then there is a path \alpha_{k}\beta_{l}\notin\mathcal{I} crossing v such that \alpha_{k}\beta_{l}\neq\alpha_{i}\beta_{i} for any 1\leqslant i\leqslant t . In this case, by the second condition of ( 4.1 ), we have k\leqslant t . Then \alpha_{k}\beta_{j}\notin\mathcal{I} for some j with j\neq k and 1\leqslant j\leqslant d . It follows that both \beta_{k} and \beta_{j} are arrows starting at v such that \alpha_{k}\beta_{k} and \alpha_{k}\beta_{j} are paths not lying in \mathcal{I} , this contradicts the definition of almost gentle pairs. The claim is proved. ∎ If there are t non-zero paths of length two on an almost gentle pair (\mathcal{Q},\mathcal{I}) crossing a (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex v , then, using the notation from Lemma 4.1 and F IGURE 4.1 , the anti-claw corresponding to E(v) is of the form s_{1}{\ \shortmid\mkern-13.0mu\vee\ }s_{2}{\ \shortmid\mkern-13.0mu\vee\ }\cdots{\ \shortmid\mkern-13.0mu\vee\ }s_{c}, where each s_{i} is a left maximal directed string of the form \cdots\alpha_{i} , and there are d right maximal directed string s_{1}^{\prime} , \ldots , s_{d}^{\prime} starting at v , where each s_{j}^{\prime} is a path of the form \beta_{j}\cdots , see F IGURE 4.2 . E(v) \textstyle{x_{1}\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces} \scriptstyle{s_{1}} \textstyle{x_{2}\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces} \scriptstyle{s_{2}} \textstyle{\cdots} \textstyle{x_{c-1}\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces} \scriptstyle{s_{c-1}} \textstyle{x_{c}\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces} \scriptstyle{s_{c}} \textstyle{v\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces} \scriptstyle{s_{1}^{\prime}} \scriptstyle{s_{2}^{\prime}} \scriptstyle{s_{d-1}^{\prime}} \scriptstyle{s_{d}^{\prime}} \textstyle{y_{1}} \textstyle{y_{2}} \textstyle{\cdots} \textstyle{y_{d-1}} \textstyle{y_{d}} s_{i}=\cdots\alpha_{i} ( 1\leqslant i\leqslant c ) s_{j}^{\prime}=\beta_{j}\cdots ( 1\leqslant j\leqslant d ) Figure 4.2. (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex v on an almost gentle pair (\mathcal{Q},\mathcal{I}) ( s_{1}^{\prime},\ldots,s_{d}^{\prime} are right maximal directed strings) Lemma 4.2 . Let v be a (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex on an almost gentle pair (\mathcal{Q},\mathcal{I}) is shown in F IGURE 4.2 . Then \Omega_{1}(E(v)) has a decomposition as follows: \Omega_{1}(E(v))\cong\mathit{\Psi}\oplus\bigoplus_{\jmath\in J}S_{\jmath}\oplus\bigoplus_{\imath\in I}M_{\imath}, where I and J are index sets, \mathit{\Psi} is either zero or a module whose top is a direct sum of some copies of S(v) , all S_{\jmath} are copies of S(v) , and all M_{\imath} are either simple modules or right maximal directed string modules. (Note that \mathit{\Psi} may be not indecomposable, and, in this case, \mathit{\Psi} may contain a direct summand which is indecomposable projective.) Proof. By Lemma 4.1 , we may assume \alpha_{1}\beta_{1},\ldots,\alpha_{t}\beta_{t}\notin\mathcal{I} . Then we have s_{1}s_{1}^{\prime},\ldots,s_{t}s_{t}^{\prime}\notin\mathcal{I} since \mathcal{I} is an admissible ideal generated by some paths of length two, and so, for any 1\leqslant i\leqslant c , the claw corresponding to the indecomposable projective module P(x_{i}) is of the form \kappa=s_{i1}{\ \shortmid\mkern-13.0mu\wedge\ }s_{i2}{\ \shortmid\mkern-13.0mu\wedge\ }\cdots{\ \shortmid\mkern-13.0mu\wedge\ }s_{im_{i}}, where • s_{i1}=\begin{cases}s_{i}s_{i}^{\prime},&i\leqslant t;\\ s_{i},&i\geqslant t+1,\end{cases} here, the path s_{i} is of the form (4.2) \displaystyle v_{i,1,1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,1,1}}v_{i,1,2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,1,2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,1,\ell_{i1}}}v_{i,1,\ell_{i1}+1}\ \ (v_{i,1,1}=x_{i},v_{i,1,\ell_{i1}}+1=v), and the path s_{i}^{\prime} is of the form (4.3) \displaystyle v_{i,1,\ell_{i1}+1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,1,\ell_{i1}+1}}v_{i,1,\ell_{i1}+2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,1,l_{i1}+2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,1,l_{i1}}}v_{i,1,l_{i1}+1}\ \ (v_{i,1,\ell_{i1}}=v,v_{i,1,l_{i1}}=y_{i}); • s_{ij} is a path v_{i,j,1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,j,1}}v_{i,j,2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,j,2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,j,l_{ij}}}v_{i,j,l_{ij}+1}\ \ (x_{i}=v_{i,j,1}) of length l_{ij} ( 12 . Then the direct summand \mathit{\Psi}_{0}\leq_{\oplus}\Omega_{1}(E(v)) is projective if and only if d=0 ( where t is the integer with 0\leqslant t\leqslant\min\{c,d\} given in the proof of Lemma 4.2 ) . Proof. By Lemma 4.2 and Corollary 4.3 , if d>0 , then \mathit{\Psi}_{0}\leq_{\oplus}\Omega_{1}(E(v)) is a module such that the following conditions hold: • \mathrm{top}(\mathit{\Psi}_{0})\cong S(v)^{\oplus(c-1)} ; • \displaystyle\mathrm{rad}(\mathit{\Psi}_{0})\cong\bigoplus_{1\leqslant i\leqslant t}\mathds{M}(s_{i}^{\prime\prime}) , where s_{i}^{\prime\prime} is the path v_{i,1,\ell_{i1}+2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,1,l_{i1}+2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,1,l_{i1}}}v_{i,1,l_{i1}+1} satisfying s^{\prime}=a_{i,1,l_{i1}+1}s_{i}^{\prime\prime} , see ( 4.3 ). Then the projective cover of \mathit{\Psi}_{0} is P(v)^{\oplus(c-1)}\to\mathit{\Psi}_{0} satisfying \displaystyle\dim_{\mathds{k}}(P(v)^{\oplus(c-1)})-\dim_{\mathds{k}}\mathit{\Psi}_{0} \displaystyle=\ \displaystyle(c-1)\bigg(1+\sum_{i=1}^{d}\ell(s_{i}^{\prime})\bigg)-\bigg((c-1)+\sum_{i=1}^{t}\ell(s_{i}^{\prime})\bigg) \displaystyle=\ \displaystyle(c-2)\sum_{i=1}^{t}\ell(s_{i}^{\prime})+(c-1)\sum_{i=t+1}^{d}\ell(s_{i}^{\prime})>0. Here, t=0 admits \sum_{i=1}^{t}\ell(s_{i}^{\prime})=0 , and d=t=1 admits \sum_{i=t+1}^{d}\ell(s_{i}^{\prime})=0 . Thus, if d>0 , then \mathit{\Psi}_{0} is non-projective. If d=0 , then v is a sink which admits that \mathit{\Psi} is isomorphic to a direct sum of some copies of S(v)\cong P(v) . ∎ Example 4.8 . Consider the 1 -st syzygy \Omega_{1}(E(4)) of the indecomposable injective module E(4) given in Example 4.4 . We have that \mathit{\Psi}_{0}\cong({{}^{4}_{5}})\oplus(4) is non-projective. Here, we have c=3>2 , d=1\neq 0 , \mathds{M}(s_{1}^{\prime})\cong P(4) , \ell(s_{1}^{\prime})=1 , and \dim_{\mathds{k}}(P(4)^{\oplus 2})-\dim_{\mathds{k}}\mathit{\Psi}_{0}=(3-1)(1+\ell(s_{1}^{\prime}))-3=1>0. Lemma 4.9 . Keep the notation in Lemma 4.2 and Corollary 4.3 , and let d>2 . Then, when c\geqslant 1 , the direct summand \mathit{\Psi}_{0}\leq_{\oplus}\Omega_{1}(E(v)) is projective if and only if one of the following conditions holds. (1) c=1 ( in this case, the path s_{1} , see F IGURE 4.2 or F IGURE 4.4 , is a unique path ending at v ) , and there is a unique right maximal path s_{j}^{\prime} (1\leqslant j\leqslant d) such that s_{1}s_{j}^{\prime}\notin\mathcal{I} , \ell(s_{j}^{\prime})=1 , and \mathfrak{t}(s_{j}^{\prime}) is a sink of the quiver \mathcal{Q} . (2) c=1 , and there is a unique right maximal path s_{j}^{\prime}= v=v_{1,1,\ell_{11}+1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,\ell_{11}+1}}v_{1,1,\ell_{11}+2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,\ell_{11}+2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,l_{11}}}v_{{1,1,l_{11}+1}}\ \ \text{{\rm(}cf. {\rm(}\ref{formula:1st-syzygy of E II}{\rm)}{\rm)}} (1\leqslant j\leqslant d) such that s_{1}s_{j}^{\prime}\notin\mathcal{I} , \ell(s_{j}^{\prime})\geqslant 2 , and the vertex v_{1,1,\ell_{11}+2} (\in v^{\mathscr{\downarrow}}) on the path s_{j}^{\prime} is not an (a_{1,1,\ell_{11}+1},\boldsymbol{\downarrow}) -relational vertex. (3) c=1 , and s_{1}s_{j}^{\prime}\in\mathcal{I} holds for all 1\leqslant j\leqslant d . Proof. First of all, if c>2 , then \mathit{\Psi}_{0} is non-projective by Lemma 4.7 . Thus, we only need to consider the case for 1\leqslant c\leqslant 2 . If c=2 , we have two subcases as follows: (a) t=1 . In this subcase, cf. F IGURE 4.2 , there is a unique path lying in \{s_{1}^{\prime},s_{2}^{\prime},\ldots,s_{d}^{\prime}\} , assuming s_{1}^{\prime} without loss of generality, such that s_{1}s_{1}^{\prime}\notin\mathcal{I} holds. Then, by Lemma 4.2 , \mathit{\Psi}_{0} is a string module corresponding to the string s_{1}^{\prime} , then the projective cover is of the following form P(v)\to\mathit{\Psi}_{0} which admits \displaystyle\dim_{\mathds{k}}P(v)-\dim_{\mathds{k}}\mathit{\Psi}_{0} \displaystyle=\ \displaystyle\bigg(1+\sum_{i=1}^{d}\ell(s_{i}^{\prime})\bigg)-\bigg(1+\ell(s_{1}^{\prime})\bigg) \displaystyle=\ \displaystyle\sum_{i=2}^{d}\ell(s_{i}^{\prime})>0.\ \ \text{(for $d>2$)} It follows that the kernel of P(v)\to\mathit{\Psi}_{0} is not zero. Thus, \mathit{\Psi}_{0} is non-projective as required. (b) t=2 . In this subcase, cf. F IGURE 4.2 , there are two paths lying in \{s_{1}^{\prime},s_{2}^{\prime},\ldots,s_{d}^{\prime}\} , assuming s_{1}^{\prime} and s_{2}^{\prime} without loss of generality, such that s_{1}s_{1}^{\prime}\notin\mathcal{I} and s_{2}s_{2}^{\prime}\notin\mathcal{I} hold. Then, by Lemma 4.2 , \mathit{\Psi}_{0} is a string module corresponding to the string (s_{1}^{\prime})^{-1}s_{2}^{\prime} , and we obtain that the projective cover is of the following form P(v)\to\mathit{\Psi}_{0} which admits \displaystyle\dim_{\mathds{k}}P(v)-\dim_{\mathds{k}}\mathit{\Psi}_{0} \displaystyle=\ \displaystyle\bigg(1+\sum_{i=1}^{d}\ell(s_{i}^{\prime})\bigg)-\bigg(1+\ell(s_{1}^{\prime})+\ell(s_{2}^{\prime})\bigg) \displaystyle=\ \displaystyle\sum_{i=3}^{d}\ell(s_{i}^{\prime})>0.\ \ \text{(by using $d>2$)} It follows that \mathit{\Psi}_{0} is non-projective by an argument similar to that in (a). Thus, \mathit{\Psi}_{0} is non-projective in the case of c=2 . If c=1 , then E(v) is both an injective module and a directed string module. We have the following two cases. (c) there is a unique path lying in \{s_{1}^{\prime},s_{2}^{\prime},\ldots,s_{d}^{\prime}\} , assuming s_{1}^{\prime} without loss of generality, such that s_{1}s_{1}^{\prime}\notin\mathcal{I} holds. In this case, \mathit{\Psi}_{0} is a string module corresponding to the string s_{1}^{\prime\prime} which is obtained by deleting the first arrow a_{1,1,\ell_{11}+1} of s^{\prime} where \ell_{11}+1\geqslant 1 , v_{1,1,\ell_{11}+1}=v , see ( 4.3 ). Then the following two subcases need to be considered. (c-1) \ell(s_{1}^{\prime})=1 . In this subcase, s_{1}^{\prime\prime} is a path of length zero given by the vertex v_{1,1,\ell_{11}+2} . It follows that \mathit{\Psi}_{0}=S(v_{1,1,\ell_{11}+2}) is projective if and only if v_{1,1,\ell_{11}+2}=\mathfrak{t}(s_{1}^{\prime\prime})=\mathfrak{t}(s_{1}^{\prime}) is a sink. We obtain (1). (c-2) \ell(s_{1}^{\prime})\geqslant 2 . In this subcase, s_{1}^{\prime\prime}=\ \ v_{1,1,\ell_{11}+2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,\ell_{11}+2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,l_{11}}}v_{1,1,l_{11}+1} is a right maximal directed string of length \geqslant 1 . We have \mathit{\Psi}_{0}\cong\mathds{M}(s_{1}^{\prime\prime}) . Thus, \mathit{\Psi}_{0} is non-projective if and only if there is at least one arrow \alpha starting at v_{1,1,\ell_{i1+2}} , and by the definition of almost gentle algebra, we have a_{1,1,\ell_{11}+1}\alpha\in\mathcal{I} . That is, \mathit{\Psi}_{0} is non-projective if and only if v_{1,1,\ell_{i1+2}} is (a_{1,1,\ell_{11}+1},\boldsymbol{\downarrow}) -relational. We obtain (2). (d) For all path s_{j}^{\prime}\in\{s_{1}^{\prime},s_{2}^{\prime},\ldots,s_{d}^{\prime}\} , we have s_{1}s_{j}^{\prime}\in\mathcal{I} . In this case, \mathit{\Psi}_{0}=0 is projective. We obtain (3). Therefore, \mathit{\Psi}_{0} is projective if and only if the (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex v lies in one of the cases (c-1), (c-2) and (d). ∎ Example 4.10 . (1) Consider the almost gentle pair (\mathcal{Q},\mathcal{I}^{\prime}) given by the quiver \mathcal{Q} provided in Example 2.2 and the admissible ideal \mathcal{I}^{\prime}=\mathcal{I}\backslash\{a_{1,2}a_{2,4}\}\cap\{a_{2,4}a_{2,5}\} see F IGURE 4.7 . 1 2 3 3^{\prime} 4 5 2_{\mathrm{L}} 3_{\mathrm{L}} 4_{\mathrm{L}} 2_{\mathrm{R}} 3_{\mathrm{R}} 4_{\mathrm{R}} a_{4,5} a_{2,4} a_{1,2} a_{2,3} a_{2,3^{\prime}} a_{3,4} a_{3^{\prime},4} a_{1,2_{\mathrm{R}}} b_{1,2_{\mathrm{R}}} a_{1,2_{\mathrm{L}}} a_{2_{\mathrm{L}},3_{\mathrm{L}}} a_{3_{\mathrm{L}},4_{\mathrm{L}}} a_{4_{\mathrm{L}},5} a_{2_{\mathrm{R}},3_{\mathrm{R}}} a_{3_{\mathrm{R}},4_{\mathrm{R}}} a_{4_{\mathrm{R}},5} b_{4_{\mathrm{L}},5} Figure 4.7. An almost gentle pair (\mathcal{Q},\mathcal{I}^{\prime}) ( \mathcal{I}^{\prime}=\mathcal{I}\backslash\{a_{1,2}a_{2,4}\}\cap\{a_{2,4}a_{2,5}\} ) Then the vertex 2 is a (1^{\mathrm{in}},3^{\mathrm{out}}) -type vertex such that the conditions given in Lemma 4.9 (1) are satisfied, except the condition \mathfrak{t}(s_{j}^{\prime}) to be a sink. That is, c=1 , d=3(>2) , s_{1}=a_{1,2} , s_{1}^{\prime}=a_{2,3} , s_{2}^{\prime}=a_{2,4}(=s_{j}^{\prime}) , s_{3}^{\prime}=a_{2,3^{\prime}} , \ell(s_{2}^{\prime})=1 , and s_{1}s_{2}^{\prime}\notin\mathcal{I} hold. In this instance, \mathit{\Psi}_{0} is non-projective. Indeed, one can check that \mathit{\Psi}_{0}\cong S(4) is non-projective simple. (2) Consider the almost gentle pair (\mathcal{Q},\mathcal{I}^{\prime\prime}) given by the quiver \mathcal{Q} provided in Example 2.2 and the admissible ideal \mathcal{I}^{\prime\prime}=\mathcal{I}\backslash\{a_{1,2}a_{2,4}\} , Then the vertex 2 is a (1^{\mathrm{in}},3^{\mathrm{out}}) -type vertex such that the conditions given in Lemma 4.9 (2) are satisfied. Thus, for E(2)=(^{1}_{2}) , we have that \mathit{\Psi}_{0} is projective. Indeed, one can check that \mathit{\Psi}_{0}\cong({{}^{4}_{5}})\cong P(4) since P(1) is the indecomposable projective module corresponding to the claw a_{1,2_{\mathrm{L}}}{\ \shortmid\mkern-13.0mu\wedge\ }a_{1,2}a_{2,4}a_{4,5}{\ \shortmid\mkern-13.0mu\wedge\ }a_{1,2_{\mathrm{R}}}a_{2_{\mathrm{R}},3_{\mathrm{R}}}a_{3_{\mathrm{R}},4_{\mathrm{R}}}{\ \shortmid\mkern-13.0mu\wedge\ }b_{1,2_{\mathrm{R}}}. (3) The indecomposable injective module E(2) over the almost gentle algebra A=\mathds{k}\mathcal{Q}/\mathcal{I} shown in Example 2.2 is the one corresponding to the (1^{\mathrm{in}},3^{\mathrm{out}}) -type vertex 2 . We have c=1 , d=3 (\geqslant 2) , s_{1}=a_{1,2} , s_{1}^{\prime}=a_{2,3} , s_{2}^{\prime}=a_{2,4}a_{4,5} , and s_{3}^{\prime}=a_{2,3^{\prime}} . Obviously, 4\in 2^{\mathscr{\downarrow}}=\{3,4,3^{\prime}\} is a vertex on the path s_{2}^{\prime} , and the length of s_{2}^{\prime} is two. Notice that s_{1}s_{j}^{\prime}\in\mathcal{I} holds for all j\in\{1,2,3\} , then \mathit{\Psi}_{0} is projective by Lemma 4.9 (3). In fact, we can check that \displaystyle\Omega_{1}(E(2))\cong(2_{\mathrm{L}})\oplus(2_{\mathrm{R}})\oplus\left({\begin{smallmatrix}2_{\mathrm{R}}\\ 3_{\mathrm{R}}\\ 4_{\mathrm{R}}\end{smallmatrix}}\right)=\bigoplus_{\imath\in I=\{1,2,3\}}M_{\imath} \displaystyle\mathit{\Psi}_{0}=0,\ \text{ and }\bigoplus_{\jmath\in J=\emptyset}S_{\jmath}=0 by using Lemma 4.2 . Thus \mathit{\Psi}_{0}=0 is projective. Definition 4.11 . For a vertex v of almost gentle pair (\mathcal{Q},\mathcal{I}) , assume that v is a (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex. We call it an invalid vertex if one of the following conditions holds. (1) v is a gentle vertex with c=2 ; (2) c is arbitrary and d=0 ; (3) c=1 , there is a unique right maximal path s_{j}^{\prime} (1\leqslant j\leqslant d) such that s_{1}s_{j}^{\prime}\notin\mathcal{I} , \ell(s_{j}^{\prime})=1 , and \mathfrak{t}(s_{j}^{\prime}) is a sink of the quiver \mathcal{Q} ; (4) c=1 , there is a unique right maximal path s_{j}^{\prime}= v=v_{1,1,\ell_{11}+1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,\ell_{11}+1}}v_{1,1,\ell_{11}+2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,\ell_{11}+2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,l_{11}}}v_{{1,1,l_{11}+1}} (1\leqslant j\leqslant d) such that s_{1}s_{j}^{\prime}\notin\mathcal{I} , \ell(s_{j}^{\prime})\geqslant 2 , and the vertex v_{1,1,\ell_{11}+2} is not an (a_{1,1,\ell_{11}+1},\boldsymbol{\downarrow}) -relational vertex; (5) c=1 and s_{1}s_{j}^{\prime}\in\mathcal{I} for any 1\leqslant j\leqslant d ; Proposition 4.12 . Keep the notation in Lemma 4.2 and Corollary 4.3 . For the indecomposable injective module E(v) corresponding to the (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex v , if c\geqslant 1 , then the direct summand \mathit{\Psi}_{0} (\leq_{\oplus}\Omega_{1}(E(v))) is projective if and only if v is invalid . Proof. In the case for 1\leqslant c\leqslant 2 and d=0 , we have that \mathit{\Psi}_{0}\cong S(v)^{\oplus(c-1)} is projective since v is a sink. Thus, in the condition (2), we assume c\geqslant 3 . The conditions (1)–(5) given in Definition 4.11 are given by Lemmata 4.5 , 4.7 and 4.9 , respectively. To be more precisely, if c=2 , d\leqslant 2 and v is a gentle vertex, then Lemma 4.5 provides the condition (1); if c\geqslant 3 , then Lemma 4.7 provides the condition (2); if c=1 and d\geqslant 3 , then v satisfying the conditions given in Lemma 4.9 is a vertex such that the conditions (3)–(5) hold. Thus, we only need to consider the following two cases: (A) 1\leqslant c\leqslant 2 , d\leqslant 2 and v is not a gentle vertex; (B) c=1 , d\leqslant 2 , and v is a gentle vertex. For the case (A), by the definition of gentle vertices, we have that one of the following cases occurs: (a) c=1 , d=2 , s_{1}s_{1}^{\prime}\in\mathcal{I} , and s_{1}s_{2}^{\prime}\in\mathcal{I} ; (b) c=2 , d=1 , s_{1}s_{1}^{\prime}\in\mathcal{I} , and s_{2}s_{1}^{\prime}\in\mathcal{I} ; (c) c=d=2 such that s_{1}s_{2}^{\prime} and s_{1}s_{2}^{\prime} lie in \mathcal{I} , and at least one of s_{1}s_{1}^{\prime} and s_{2}s_{2}^{\prime} lies in \mathcal{I} . In the case (a), we have \mathit{\Psi}_{0}=0 which is a trivial projective module, we obtain the condition (5). In the case (b), we have \mathit{\Psi}_{0}\cong S(v) which is non-projective since v does not be a sink of \mathcal{Q} . In the case (c), if s_{1}s_{1}^{\prime}\in\mathcal{I} and s_{2}s_{2}^{\prime}\notin\mathcal{I} , then \mathit{\Psi}_{0} is a directed string module corresponding to the string s_{2}^{\prime} . In this case, the kernel of the projective cover of \mathit{\Psi}_{0} , which is of the form P_{0}\to\mathit{\Psi}_{0} , is non-zero since the following formula \dim_{\mathds{k}}P_{0}-\dim_{\mathds{k}}\mathit{\Psi}_{0}=(\ell(s_{1}^{\prime})+\ell(s_{2}^{\prime})+1)-(\ell(s_{2}^{\prime})+1)=\ell(s_{1}^{\prime})>0 holds. Thus, \mathit{\Psi}_{0} is non-projective. Similarly, We have that \mathit{\Psi}_{0} is non-projective in other cases. For the case (B), if d=0 then it is trivial that \mathit{\Psi}_{0}=0 . Thus, we have 1\leqslant d\leqslant 2 , and obtain the following two cases. (d) c=1 and d=1 . (e) c=1 and d=2 . In the case (d), we have that v is a (1^{\mathrm{in}},1^{\mathrm{out}}) -type vertex and s_{1}s_{1}^{\prime} equals to \overbrace{v_{1,1,1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,1}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,\ell_{11}}}v}^{s_{1}}=\overbrace{v_{1,1,\ell_{11}+1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,\ell_{11}+1}}\underbrace{v_{1,1,\ell_{11}+2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,\ell_{11}+2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,l_{11}}}v_{{1,1,l_{11}+1}}}_{s_{1}^{\prime\prime}}}^{s_{1}^{\prime}}. Then \mathit{\Psi}_{0}\cong\mathds{M}(s^{\prime\prime}) is projective if and only if one of the following conditions holds: (d-1) s_{1}s_{1}^{\prime}\notin\mathcal{I} , \ell(s_{1}^{\prime})=1 , and v_{1,1,\ell_{11}+2}=\mathfrak{t}(s_{1}^{\prime}) is a sink, that is, v_{1,1,\ell_{11}+2} is a vertex lying in v^{\mathscr{\downarrow}} satisfying the condition (3); (d-2) s_{1}s_{1}^{\prime}\notin\mathcal{I} , \ell(s_{1}^{\prime})\geqslant 2 , and v_{1,1,\ell_{11}+2} is not a (a_{1,1,\ell_{11}+2},\boldsymbol{\downarrow}) -relational vertex, that is, the condition (4) holds; (d-3) s_{1}s_{1}^{\prime}\in\mathcal{I} . In this case, the vertex v satisfies the condition (5). If v is a (1^{\mathrm{in}},2^{\mathrm{out}}) -type vertex, then at least one of s_{1}s_{1}^{\prime} and s_{1}s_{2}^{\prime} lies in \mathcal{I} . If s_{1}s_{1}^{\prime}\notin\mathcal{I} , then s_{1}s_{2}^{\prime}\in\mathcal{I} . Thus, similar to the case (d), \mathit{\Psi}_{0} is projective if and only if one of the following conditions holds: (e-1) s_{1}s_{1}^{\prime}\notin\mathcal{I} , \ell(s_{1}^{\prime})=1 , and v_{1,1,\ell_{11}+2}=\mathfrak{t}(s_{1}^{\prime}) is a sink, this satisfies the condition (3); (e-2) s_{1}s_{1}^{\prime}\notin\mathcal{I} , \ell(s_{1}^{\prime})\geqslant 2 , and v_{1,1,\ell_{11}+2} is not a (a_{1,1,\ell_{11}+2},\boldsymbol{\downarrow}) -relational vertex, this satisfies the condition (4); The cases for s_{1}s_{1}^{\prime}\in\mathcal{I} and s_{1}s_{2}^{\prime}\notin\mathcal{I} are dual. (e-3) s_{1}s_{1}^{\prime} and s_{1}s_{2}^{\prime} lie in \mathcal{I} . Then we have \mathit{\Psi}_{0}=0 is projective, this satisfies the condition (5). ∎ 4.3. Self-injective dimension We divide all vertices of an almost gentle pair (\mathcal{Q},\mathcal{I}) to three parts: sources; (1^{\mathrm{in}},d^{\mathrm{out}}) -vertices, and ((\geqslant 2)^{\mathrm{in}},d^{\mathrm{out}}) -vertices. If a vertex v is a source, then E(v) is a simple module corresponding to v , and in this situation we have computed the projective dimension \mathrm{proj.dim}E(v) in Proposition 3.10 , i.e., \mathrm{proj.dim}E(v)=\mathrm{proj.dim}S(v)=\ell(F_{v}), where F_{v} is the right maximal forbidden path starting at the sink v . Next, we compute \mathrm{proj.dim}E(v) in the case for v being a vertex lies in the second and third situations. By Lemma 4.2 , we have that \Omega_{1}(E(v)) is a direct sum of some right maximal directed string modules and the module \mathit{\Psi}_{0} . The projective resolution of any right maximal directed string module can be computed by using Lemma 3.3 . Therefore, we need to construct the projective resolution of \mathit{\Psi}_{0} . Lemma 4.13 . For any n\geqslant 1 , the n -th syzygy \Omega_{n}(\mathit{\Psi}_{0}) of \mathit{\Psi}_{0} is a direct sum of some right maximal directed string modules and some simple module. Proof. Let v be a (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex shown in F IGURE 4.2 , and assume that s_{i}s_{j}^{\prime}\notin\mathcal{I} if and only if 0\leqslant i=j\leqslant t (<\min\{c,d\}) . Recall that any A -module M over the almost gentle algebra A=\mathds{k}\mathcal{Q}/\mathcal{I} can be described by the bound quiver (\mathcal{Q},\mathcal{I}) as the quiver representation (V_{w},\varphi_{a})_{w\in\mathcal{Q}_{0},a\in\mathcal{Q}_{1}} , where V_{w} is the \mathds{k} -linear space which is \mathds{k} -linear isomorphic to Me_{w} and \varphi_{a} is a \mathds{k} -linear space V_{\mathfrak{s}(a)}\to V_{\mathfrak{t}(a)} such that \varphi_{a}\varphi_{b}=0 ( \mathfrak{t}(a)=\mathfrak{s}(b) ) if and only if ab\in\mathcal{I} . Computing the kernel of the projective cover of E(v) , we obtain that the quiver representation of \mathit{\Psi} is of the form (\mathit{\Psi}e_{w},\varphi_{a})_{w\in\mathcal{Q}_{0},a\in\mathcal{Q}_{1}}, where for any vertex w on the path s_{j}^{\prime\prime} except v ( s_{j}^{\prime\prime} a string obtained by deleting the first arrow of s_{j}^{\prime} ), and \mathit{\Psi}e_{w}\cong\mathds{M}(s_{j}^{\prime\prime})e_{w}, to be precise, we have \mathrm{rad}(\mathit{\Psi})\cong\bigoplus_{j=1}^{d}\mathds{M}(s_{j}^{\prime\prime}) . F IGURE 4.8 provides an example for the above, and for simplicity we assume that each path s_{j}^{\prime} has no self-injection in this figure. \textstyle{\mathds{k}^{\oplus(t-1)}\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces\lx@inpgf@ignorespaces} \scriptstyle{s_{1}^{\prime}} \scriptstyle{s_{2}^{\prime}} \scriptstyle{s_{d-1}^{\prime}} \scriptstyle{s_{d}^{\prime}} \textstyle{\mathds{k}_{1}} \textstyle{\mathds{k}_{2}} \textstyle{\cdots} \textstyle{\mathds{k}_{d-1}} \textstyle{\mathds{k}_{d}} \mathds{k}_{j}\cong\mathds{k} , 1\leqslant j\leqslant d Figure 4.8. The quiver representation of \mathit{\Psi} (all \mathds{k} -linear spaces corresponding to vertices on each s_{j}^{\prime} is one-dimensional, except the \mathds{k} -linear space \mathds{k}^{\oplus(t-1)} corresponding to the vertex v ) Furthermore, \mathrm{top}(\mathit{\Psi})\cong S(v)^{\oplus(t-1)} , and then the projective cover of \mathit{\Psi} is of the form P(v)^{\oplus(t-1)}\to\mathit{\Psi} . Now, we have \Omega_{1}(\mathit{\Psi})\cong\bigoplus_{j=1}^{d}\mathds{M}(s_{j}^{\prime\prime}) , then it is clear that each direct summand of \Omega_{1}(\mathit{\Psi}) is either a right maximal directed string module or a simple module, and so is \Omega_{n}(\mathit{\Psi}_{1}) by using Proposition 3.5 and induction. On the other hand, by Lemma 4.2 , we have \mathit{\Psi}_{0}=\mathit{\Psi}\oplus\bigoplus_{\jmath\in J}S_{\jmath}, it follows that \displaystyle\Omega_{n}(\mathit{\Psi}_{0})=\Omega_{n}(\mathit{\Psi})\oplus\bigoplus_{\jmath\in J}\Omega_{n}(S_{\jmath}) . Notice that S_{\jmath} is a direct string module corresponding to a simple string, so each direct summand of \Omega_{n}(S_{\jmath}) is either a right maximal directed string module or a simple module by using Proposition 3.5 and induction. ∎ For a directed string \delta=a_{1}a_{2}\cdots a_{l} , a forbidden path F=b_{1}b_{2}\cdots b_{l} is said to be a \delta -forbidden if one of the following conditions holds: • \mathfrak{s}(F)=\mathfrak{t}(\delta) and \delta b_{1}\notin\mathcal{I} ; • \mathfrak{s}(F)=\mathfrak{s}(\delta) and a_{1}\neq b_{1} . Example 4.14 . The indecomposable injective module E(2)=({{}^{1}_{2}}) over the almost gentle algebra A=\mathds{k}\mathcal{Q}/\mathcal{I} given in Example 2.2 is a directed string module corresponding to the directed string a_{1,2} . Then the forbidden paths b_{1,2_{\mathrm{R}}}a_{2_{\mathrm{R}},3_{\mathrm{R}}} , a_{1,2_{\mathrm{R}}} and b_{1,2_{\mathrm{L}}}a_{2_{\mathrm{L}},3_{\mathrm{L}}} are a_{1,2} -forbidden paths. But a_{2,4} is not an a_{1,2} -forbidden path because a_{1,2}a_{2,4}\in\mathcal{I} . Notation 4.15 . Let \mathcal{F}(\delta) be the set of all \delta -forbidden paths. If \ell(\delta)=0 , then \mathcal{F}(\delta)=\mathcal{F}(v) , where v=\mathfrak{s}(\delta)=\mathfrak{t}(\delta) . Now we show the following proposition. Proposition 4.16 . Let \delta be either a right maximal directed string or a simple string. Then \Omega_{n-1}(\mathds{M}(\delta)) is non-projective if and only if there is a forbidden path F\in\mathcal{F}(\delta) whose length is n . Proof. The case for \delta to be simple is shown in Proposition 3.10 . Now, we assume that \delta is the right maximal directed string v_{1,1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1}}v_{1,2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,l_{1}}}v_{1,l_{1}+1} with length l_{1}\geqslant 1 and v_{1,1} a (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex ( d\geqslant 1 ), and the claw corresponding to P(v_{1,1}) is \kappa=\delta_{1}{\ \shortmid\mkern-13.0mu\wedge\ }\delta_{2}{\ \shortmid\mkern-13.0mu\wedge\ }\cdots{\ \shortmid\mkern-13.0mu\wedge\ }\delta_{d}, where \delta=\delta_{1} , \delta_{i}=\ v_{i,1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,1}}v_{i,2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,l_{i}}}v_{i,l_{i}+1} ( l_{i}\geqslant 1 , 1\leqslant i\leqslant d ) is right maximal, and v_{1,1}=v_{2,1}=\cdots=v_{d,1} . Then we have \Omega_{1}(\mathds{M}(\delta))\cong\bigoplus_{i=1}^{d}\mathds{M}(\delta_{i}^{*}), where \delta_{i}^{*}=\begin{cases}0,&\text{ if }i=1;\\ v_{i,2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{i,l_{i}}}v_{i,l_{i}+1},&\text{ if }i\geqslant 2.\end{cases} In the case for d=1 , \mathds{M}(\delta) is projective, and so \Omega_{1}(\mathds{M}(\delta))=0 is projective. In the case for d\geqslant 2 , for any i\geqslant 2 , we have \mathds{M}(\delta_{i}^{*})\neq 0 . Furthermore, \mathds{M}(\delta_{i}^{*}) is non-projective if and only if v_{i,2} is (a_{i,1},\boldsymbol{\downarrow}) -relational. It follows that • there is an arrow a_{i,2}^{(1)} such that a_{i,1}a_{i,2}^{(1)} is a forbidden path lying in \mathcal{F}(\delta) ; • the 1 -st syzygy \Omega_{1}(\mathds{M}(\delta)) is non-projective; • there is a directed string \delta_{i}^{(1)} which is of the form a_{i,2}^{(1)}\cdots such that \mathds{M}(\delta_{i}^{(1)*})\leq_{\oplus}\Omega_{1}(\mathds{M}(\delta_{i}^{*}))\leq_{\oplus}\Omega_{2}(\mathds{M}(\delta)), where \delta_{i}^{(1)*} is the string obtained by deleting the first arrow a_{i,2}^{(1)} of \delta_{i}^{(1)} , see F IGURE 4.9 . \delta_{i}^{(1)*} \delta_{i}^{(1)} v_{i,1} v_{i,2} v_{i,3} v_{i,4} a_{i,1} a_{i,2} a_{i,3} a_{i,4} a_{i,2}^{(1)} v_{i,3}^{(1)} v_{i,4}^{(1)} a_{i,3}^{(1)} a_{i,4}^{(1)} Figure 4.9. \Omega_{1}(\mathds{M}(\delta)) is non-projective Therefore, if \mathds{M}(\delta_{i}) is non-projective, then so is \Omega_{1}(\mathds{M}(\delta)) , and in this case, we get a forbidden path a_{i,1}a_{i,2}^{(1)} lying in \mathcal{F}(\delta) ; if \mathds{M}(\delta_{i}^{(1)*}) is non-projective, then so is \Omega_{2}(\mathds{M}(\delta)) , and in this case, by the method similar to above, there is a directed string \delta_{i}^{(2)} which is of the form a_{i,3}^{(2)}\cdots such that a_{i,2}^{(1)}a_{i,3}^{(2)} is a forbidden path, then we get a forbidden path a_{i,1}a_{i,2}^{(1)}a_{i,3}^{(2)} lying in \mathcal{F}(\delta) , see F IGURE 4.10 . \delta_{i}^{(2)*} \delta_{i}^{(2)} \delta_{i}^{(1)*} \delta_{i}^{(1)} v_{i,1} v_{i,2} v_{i,3} v_{i,4} a_{i,1} a_{i,2} a_{i,3} a_{i,4} a_{i,2}^{(1)} v_{i,3}^{(1)} v_{i,4}^{(1)} a_{i,3}^{(1)} a_{i,4}^{(1)} a_{i,3}^{(2)} v_{i,4}^{(2)} a_{i,4}^{(2)} Figure 4.10. \Omega_{2}(\mathds{M}(\delta)) is non-projective (note that a_{i,3}^{(1)} and a_{i,3}^{(2)} may be the same) Repeating the above step, we can find a forbidden path of length n which lies in \mathcal{F}(\delta) if \Omega_{n-1}(\mathds{M}(\delta)) is non-projective. Conversely, if \delta is a simple string corresponding to v_{1} , and by assumption there is a forbidden path of length n lying in \mathcal{F}(\delta)=\mathcal{F}(v_{1}) . Then \Omega_{n-1}(\mathds{M}(\delta)) is non-projective by Proposition 3.10 . Next, consider the case for \delta to be a right maximal directed string v_{1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1}}v_{2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{l}}v_{l+1} with length l\geqslant 1 . Since \mathcal{F}(\delta) contains a forbidden path of length n and \delta is a right maximal directed string, all forbidden paths lying in \mathcal{F}(\delta) share the same source v_{1,1} by the definition of \mathcal{F}(\delta) . Suppose that there is a forbidden path F=b_{1}b_{2}\cdots b_{n} which is of the form \displaystyle w_{1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{b_{1}}w_{2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{b_{2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{b_{n}}w_{n+1} such that w_{1}=v_{1} and a_{1}\neq b_{1} . The 1 -st syzygy \Omega_{1}(\mathds{M}(\delta)) contains a direct summand M_{1} which is a directed string module (see Lemma 3.2 ) such that M_{1}b_{2}=0 and \mathrm{top}(M_{1})\cong S(w_{2}) by computing the kernel of the projective cover of \mathds{M}(\delta) . Note that M_{1} is non-projective. Otherwise, M_{1}\cong e_{w_{2}}A . It follows that M_{1}b_{2}=e_{w_{2}}Ab_{2}\neq 0 since e_{w_{2}}Ab_{2} contains at least one element b_{2}\neq 0 . This is a contradiction. Then \Omega_{1}(\mathds{M}(\delta)) is non-projective. Since w_{2} is (b_{1},\boldsymbol{\downarrow}) -relational, the projective cover of M_{1} is P(w_{2})\to M_{1} , then \Omega_{1}(M_{1}) has a direct summand M_{2} which is a directed string module (see Lemma 3.2 ) such that M_{2}b_{3}=0 and \mathrm{top}(M_{2})\cong S(w_{3}) by computing the kernel of the projective cover of M_{1} . Similarly, we have M_{2}\leq_{\oplus}\Omega_{1}(M_{1}) ( \leq_{\oplus}\Omega_{2}(\mathds{M}(\delta)) ) is non-projective, and so \Omega_{2}(\mathds{M}(\delta)) is non-projective. Therefore \Omega_{n-1}(\mathds{M}(\delta)) is non-projective by induction. ∎ Example 4.17 . Recall that the minimal projective resolution of the indecomposable injective module E(2) over the almost gentle algebra A=\mathds{k}\mathcal{Q}/\mathcal{I} given in Example 2.2 is \displaystyle 0 \displaystyle\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits P_{2}=P(3_{\mathrm{L}})\oplus P(3_{\mathrm{R}})\lx@xy@svg{\hbox{\raise 2.5pt\hbox{\kern 3.0pt\hbox{\ignorespaces\ignorespaces\ignorespaces\hbox{\vtop{\halign{\entry@#!@&&\entry@@#!@\cr&\crcr}}}\ignorespaces{\hbox{\kern-3.0pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise-2.5pt\hbox{$\textstyle{\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 1.67036pt\raise 17.25555pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-2.5pt\hbox{$\scriptstyle{\left(\begin{smallmatrix}a_{2_{\mathrm{L}},3_{\mathrm{L}}}&0\\ 0&a_{2_{\mathrm{R}},3_{\mathrm{R}}}\\ 0&0\end{smallmatrix}\right)}$}}}\kern 3.0pt}}}}}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 24.44016pt\raise-5.18748pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-0.8264pt\hbox{$\scriptstyle{p_{2}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 59.9055pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 59.9055pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise-2.5pt\hbox{$\textstyle{}$}}}}}}}\ignorespaces}}}}\ignorespaces P_{1}=P(2_{\mathrm{L}})\oplus P(2_{\mathrm{R}})\oplus P(2_{\mathrm{R}}) \displaystyle\lx@xy@svg{\hbox{\raise 2.5pt\hbox{\kern 6.51857pt\hbox{\ignorespaces\ignorespaces\ignorespaces\hbox{\vtop{\halign{\entry@#!@&&\entry@@#!@\cr&\crcr}}}\ignorespaces{\hbox{\kern-3.0pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise-2.5pt\hbox{$\textstyle{\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern-6.51857pt\raise 6.96329pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-1.28671pt\hbox{$\scriptstyle{(a_{1,2_{\mathrm{L}}}\ b_{1,2_{\mathrm{R}}}\ a_{1,2_{\mathrm{R}}})}$}}}\kern 3.0pt}}}}}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 24.44016pt\raise-5.18748pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-0.8264pt\hbox{$\scriptstyle{p_{1}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 59.9055pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 59.9055pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise-2.5pt\hbox{$\textstyle{}$}}}}}}}\ignorespaces}}}}\ignorespaces P_{0}=P(1)\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits_{p_{0}}\mathds{M}(a_{1,2})\cong E(2)\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits 0, see Example 3.6 . Then we have \Omega_{1}(E(2))\cong(2_{\mathrm{L}})\oplus(2_{\mathrm{R}})\oplus\left(\begin{smallmatrix}2_{\mathrm{R}}\\ 3_{\mathrm{R}}\\ 4_{\mathrm{R}}\end{smallmatrix}\right)=S(2_{\mathrm{L}})\oplus S(2_{\mathrm{R}})\oplus P(2_{\mathrm{R}}) is non-projective since S(2_{\mathrm{L}}) and S(2_{\mathrm{R}}) are non-projective. Here, E(2) is also a right maximal directed string module, and the directed string corresponding to it is a_{1,2} . We have three a_{1,2} -forbidden paths of length \geqslant 1 as follows: a_{1,2_{\mathrm{L}}}a_{2_{\mathrm{L}},3_{\mathrm{L}}} , b_{1,2_{\mathrm{R}}}a_{2_{\mathrm{R}},3_{\mathrm{R}}} , and a_{1,2_{\mathrm{R}}} . We have that S(2_{\mathrm{L}}) and S(2_{\mathrm{R}}) correspond to the forbidden paths a_{1,2_{\mathrm{L}}}a_{2_{\mathrm{L}},3_{\mathrm{L}}} and b_{1,2_{\mathrm{R}}}a_{2_{\mathrm{R}},3_{\mathrm{R}}} of length two, respectively. The fact that P(2_{\mathrm{R}})\leq_{\oplus}\Omega_{1}(E(2)) yields \Omega_{1}(P(2_{\mathrm{R}}))=0 . Proposition 4.16 shows that the forbidden paths a_{1,2_{\mathrm{L}}}a_{2_{\mathrm{L}},3_{\mathrm{L}}} and b_{1,2_{\mathrm{R}}}a_{2_{\mathrm{R}},3_{\mathrm{R}}} , as two elements in \mathcal{F}(a_{1,2}) , exist. In Lemma 4.2 , we show that the 1 -st syzygy of an indecomposable injective E(v) ( v is a (c^{\mathrm{in}},d^{\mathrm{out}}) )-type vertex module over an almost gentle algebra can be described by three parts: • a module \mathit{\Psi} whose top is a direct sum of some copies of S(v) ; • a semi-simple module which is of the form S(v)^{\oplus(c-t)} , where t is an integer satisfying 0\leqslant t\leqslant c ; • a direct sum whose direct summand is either a right maximal directed module or a simple module; Lemma 4.13 shows that any direct summand of the n -th ( n\geqslant 1 ) syzygy \Omega_{n}(\mathit{\Psi}_{0}) of \mathit{\Psi}_{0}=\mathit{\Psi}\oplus S(v)^{\oplus(c-d)} is either a right maximal directed string module or a simple module. Propositions 3.10 and 4.16 provide a correspondence from any right maximal directed string module or simple module to a forbidden path on the bound quiver of an almost gentle algebra. Thus, for any n\geqslant 1 , each non-projective direct summand of \Omega_{n}(E(v)) can be described by using a forbidden path of length n+1 , except the direct summand \mathit{\Psi}_{0} of \Omega_{1}(E(v)) . The projectivity of \mathit{\Psi}_{0} is described in Proposition 4.12 , thus we can compute the projective resolution of any indecomposable injective module over almost gentle algebra. Definition 4.18 . For an indecomposable injective module corresponding a (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex v , assume that the anti-claw of E(v) , written as \xi in this section, is of the form shown in F IGURE 4.2 . A forbidden path F is said to be a \xi -forbidden path if it satisfies one of the following conditions: (1) there is an integer 1\leqslant i\leqslant c such that F\in\mathcal{F}(\mathfrak{s}(s_{i})) ; (2) F is a forbidden path starting at v such that v is not an invalid vertex. The set of all \xi -forbidden paths is written as \mathcal{F}(\xi) . In particular, if c=0 , then E(v)=S(v) is simple, and we have \mathcal{F}(\xi)=\mathcal{F}(v) . Theorem 4.19 . Let A=(\mathcal{Q},\mathcal{I}) be an almost gentle algebra. Then for any (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex v ( keep the notation in Lemma 4.2 and F IGURE 4.2 ) , we have \mathrm{proj.dim}E(v)=\sup\limits_{F\in\mathcal{F}(\xi)}\ell(F), where \xi is the anti-claw corresponding to E(v) . (Note that if \mathcal{F}(\xi)=\emptyset , then take \sup\limits_{F\in\mathcal{F}(\xi)}\ell(F)=0 .) Proof. If c=0 , then E(v)=S(v) . Thus, we have \mathrm{proj.dim}E(v)=\mathrm{proj.dim}S(v)=\sup\limits_{F\in\mathcal{F}(v)}\ell(F). When c\geqslant 1 , we have two cases as follows: (1) v is invalid ​​; (2) v is not invalid ​​. Assume \xi=s_{1}{\ \shortmid\mkern-13.0mu\vee\ }s_{2}{\ \shortmid\mkern-13.0mu\vee\ }\cdots{\ \shortmid\mkern-13.0mu\vee\ }s_{c} . In the case (1), \mathcal{F}(\xi) contains c ( \geqslant 1 ) right maximal forbidden paths F_{1},\ldots,F_{c} which respectively correspond to s_{1},\ldots,s_{c} such that s_{i}b_{i1}\notin\mathcal{I} , where b_{i1} is the first arrow of F_{i}=b_{i1}b_{i2}\cdots b_{il_{i}} ( l_{i}\geqslant 1 ), see Definition 4.18 (1). By Proposition 4.16 , we have that \Omega_{t}(E(v)) is non-projective if and only if there is an integer i with 1\leqslant i\leqslant c such that \ell(F_{i})=t+1 . Without loss of generality, suppose (4.6) \displaystyle\ell(F_{1})=\sup\limits_{1\leqslant i\leqslant c}\ell(F_{i}). Since \Omega_{\ell(F_{1})-1}(E(v)) is non-projective, we have \mathrm{proj.dim}E(v)\geqslant\ell(F_{1}) . If \mathrm{proj.dim}E(v)>\ell(F_{1}) , then \Omega_{\ell(F_{1})}(E(v)) is non-projective. In this case, by Proposition 4.16 , there exists a forbidden path F_{j}\in\{F_{1},\ldots,F_{c}\} such that \ell(F_{j})\geqslant\ell(F_{1})+1 . It contradicts the maximality of \ell(F_{1}) . Thus, we have (4.7) \displaystyle\mathrm{proj.dim}E(v)=\ell(F_{1}), and therefore (4.8) \displaystyle\mathrm{proj.dim}E(v)=\sup\limits_{1\leqslant i\leqslant c}\ell(F_{i}), as required. In the case (2), we obtain that \mathcal{F}(\xi) contains c ( \geqslant 1 ) right maximal forbidden paths F_{1},\ldots,F_{c} and F_{c+1} , where F_{1},\ldots,F_{c} respectively correspond to s_{1},\ldots,s_{c} such that s_{i}b_{i1}\notin\mathcal{I} ( b_{i1} is the first arrow of F_{i} ); and F_{c+1} corresponds to v , that is, F_{c+1} is a forbidden path starting at v (see Definition 4.18 (2)). Next, we compute \mathrm{proj.dim}\mathit{\Psi}_{0} . By Proposition 4.12 , v is not a sink since d=0 contradicts Definition 4.11 (2). We have the following subcases: (2.1) c=1 and d\geqslant 1 ; (2.2) c=2 , then v is not a gentle vertex; (2.3) c\geqslant 3 , then \mathit{\Psi}_{0} is non-projective by Lemma 4.7 . We only prove the case (2.1). The proofs of (2.2) and (2.3) are similar. In the subcase (2.1), there is a unique s_{j}^{\prime} ( 1\leqslant j\leqslant d ) such that s_{1}s_{j}^{\prime}\notin\mathcal{I} by Definition 4.11 (5). Without loss of generality, let s_{1}= v_{1,1,1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,1}}v_{1,1,2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,\ell_{11}}}v_{1,1,\ell_{11}+1}=v and assume j=1 and s_{j}^{\prime}=s_{1}^{\prime}= v=v_{1,1,\ell_{11}+1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,\ell_{11}+1}}v_{1,1,\ell_{11}+2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,\ell_{11}+2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{a_{1,1,l_{11}}}v_{{1,1,l_{11}+1}}. By Definition 4.11 (3)(4), we have that v_{1,1,\ell_{11}+2} is (a_{1,1,\ell_{11}+1},\boldsymbol{\downarrow}) -relational. Otherwise, (2.1.1) v_{1,1,\ell_{11}+2} is a sink, and so \ell(s_{1}^{\prime})=1 , it contradicts Definition 4.11 (3); (2.1.2) v_{1,1,\ell_{11}+2} is not a sink, and so \ell(s_{1}^{\prime})\geqslant 1 , it contradicts Definition 4.11 (2). Thus, there is an arrow \alpha_{1} such that \alpha_{1}\neq a_{1,1,\ell_{11}+1} , \mathfrak{s}(\alpha_{1})=v , and a_{1,1,\ell_{11}}\alpha_{1}\in\mathcal{I} is a forbidden path lying in \mathcal{F}(\xi) . Thus, if \mathfrak{t}(\alpha_{1}) is (\alpha_{1},\boldsymbol{\downarrow}) -relational, that is, there is an arrow \alpha_{2} such that \mathfrak{t}(\alpha_{1})=\mathfrak{s}(\alpha_{2}) and \alpha_{1}\alpha_{2}\in\mathcal{I} . then \Omega_{1}(\mathit{\Psi}) ( \leq_{\oplus}\Omega_{2}(E(v) ) has a direct summand M_{1} such that M_{1}\alpha_{2}=0 and \mathrm{top}(M_{1})\cong S(\mathfrak{t}(\alpha_{1})) , The direct summand M_{1} is non-projective. Otherwise, M_{1}\cong e_{\mathfrak{t}(\alpha_{1})}A , and so M_{1}\alpha_{2}\cong e_{\mathfrak{t}(\alpha_{1})}A\alpha_{2}\neq 0 . We find a forbidden path a_{1,1,\ell_{11}}\alpha_{1}\alpha_{2}\in\mathcal{F}(\xi) whose length equals to 3 . Next, any direct summand of \Omega_{\geqslant 2}(\mathit{\Psi}) is right maximal directed string module. We get (4.9) \displaystyle\mathrm{proj.dim}E(v)=\sup\limits_{1\leqslant i\leqslant c+1}\ell(F_{i}) by using an argument similar to the proof of ( 4.8 ). ∎ Now, we can prove the following result. Theorem 4.20 . Let A=\mathds{k}\mathcal{Q}/\mathcal{I} be an almost gentle algebra. Then \mathrm{inj.dim}A=\mathrm{proj.dim}D(A)=\sup_{F\in\mathcal{F}}\ell(F), where \mathcal{F}:=\bigcup_{v\in\mathcal{Q}_{0}}\mathcal{F}(\xi_{v}) and \xi_{v} is the anti-claw corresponding to the indecomposable injective module E(v) . Proof. By Theorem 4.19 , we have \mathrm{proj.dim}E(v)=\sup\limits_{F\in\mathcal{F}(\xi_{v})}\ell(F) for each v . Thus, \mathrm{inj.dim}A=\mathrm{proj.dim}D(A)=\mathrm{proj.dim}\bigoplus_{v\in\mathcal{Q}_{0}}E(v)=\sup_{F\in\mathcal{F}}\ell(F). ∎ 5. Auslander–Reiten Conjecture In this section, we prove that the Auslander–Reiten conjecture ( ARC for short), mentioned in Introduction, holds true for almost gentle algebras. 5.1. The second description of self-injective dimension Recall that an oriented cycle \mathscr{C}=a_{1}a_{2}\cdots a_{n} is said to be a forbidden cycle if a_{1}a_{2},\ldots,a_{n-1}a_{n} and a_{n}a_{1} are generators of \mathcal{I} . For simplicity, assume that a forbidden cycle \mathscr{C} of length \ell is of the following form: \textstyle{1\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{a_{1}} \textstyle{2\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{a_{2}} \textstyle{0\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{a_{0}} \textstyle{\ell-1\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{a_{\ell-1}} Theorem 5.1 . For an almost gentle algebra A=\mathds{k}\mathcal{Q}/\mathcal{I} whose bound quiver (\mathcal{Q},\mathcal{I}) has at least one forbidden cycle, then \mathrm{inj.dim}A=\infty if and only if there is a vertex v on some forbidden cycle such that v is not invalid ​​. Proof. If \mathrm{inj.dim}A=\infty , then, by Theorem 4.20 , there is an anti-claw \xi such that \mathcal{F}(\xi) contains a forbidden path F with \ell(F)=\infty and F being of the form: F=(\alpha_{1}\cdots\alpha_{n-1})(\overbrace{\alpha_{n}\alpha_{n+1}\cdots\alpha_{n+\ell-1}}^{\mathscr{C}=a_{0}a_{1}\cdots a_{\ell-1}})(\overbrace{\alpha_{n}\alpha_{n+1}\cdots\alpha_{n+\ell-1}}^{\mathscr{C}=a_{0}a_{1}\cdots a_{\ell-1}})\cdots,(n\geqslant 2). We get \mathfrak{t}(\alpha_{n-1})=\mathfrak{t}(\alpha_{n+\ell-1})=\mathfrak{s}(\alpha_{n}) ( =\mathfrak{s}(a_{v})=v for some 0\leqslant v\leqslant\ell-1 ). Assume that v is a (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex. Obviously, c\geqslant 2 and d\geqslant 1 . (a) When c>2 , v is invalid if and only if d=0 , see Definition 4.11 . Thus, v is not invalid ​​, a contradiction. (b) When c=2 , we have \mathrm{top}(\Omega_{1}(E(v)))\geq_{\oplus}\mathrm{top}(\mathit{\Psi}_{0})\cong S(v)^{\oplus c-1}=S(v). Since v is a vertex on a forbidden cycle, one can check that \mathrm{proj.dim}(\mathrm{top}(\mathit{\Psi}_{0}))=\infty . It follows that \mathit{\Psi}_{0} is not projective, then v is not invalid by Proposition 4.12 . Conversely, assume that every vertex on forbidden cycle is invalid ​​. Then for any vertex v on each forbidden cycle, the direct summand \mathit{\Psi}_{0}=\mathit{\Psi}\oplus\bigoplus_{\jmath\in J}S_{\jmath} of \Omega_{1}(E(v)) is projective by Lemma 4.2 and Proposition 4.12 . In this case, if \mathrm{inj.dim}A=\infty=\mathrm{proj.dim}D(A) , then there is a right maximal directed string module M\cong\mathds{M}(\delta) such that \mathrm{proj.dim}M=\infty , where \delta=b_{1}b_{2}\cdots b_{m} is a right maximal directed string. It follows that \mathcal{F}(\delta) contains a forbidden path F such that \ell(F)=\infty . If \ell(\delta)\geqslant 1 , then, by the definition of \mathcal{F}(\delta) , we have that \mathfrak{s}(F) is a vertex on the claw \kappa=r_{1}{\ \shortmid\mkern-13.0mu\wedge\ }r_{2}{\ \shortmid\mkern-13.0mu\wedge\ }\cdots{\ \shortmid\mkern-13.0mu\wedge\ }r_{n} (where r_{j}=\ w_{j,1}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{\beta_{j,1}}w_{j,2}\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{\beta_{j,2}}\cdots\mathop{-\!\!\!-\!\!\!\longrightarrow}\limits^{\beta_{j,l_{j}}}w_{j,l_{j}+1} , 1\leqslant j\leqslant n , r_{1}=\delta , and \mathfrak{s}(b_{1})=w_{1,1}=w_{2,1}=\ldots=w_{n,1} ) corresponding to P(\mathfrak{s}(\delta)) , the indecomposable projective module given by the projective cover of \mathds{M}(\delta) , such that \mathfrak{s}(F) is a \kappa -relational vertex lying in \mathfrak{s}(\delta)^{\mathscr{\downarrow}}\backslash\{\mathfrak{t}(b_{1})\}=\{w_{j,2}\mid 2\leqslant j\leqslant n\} . Without loss of generality, assume \mathfrak{s}(F)=w_{2,2} . Then l_{2}=1 (i.e., r_{2}=\beta_{2,1} ) and F is of the following form: (5.1) \displaystyle F=(\alpha_{1}\cdots\alpha_{n-1})(\overbrace{\alpha_{n}\alpha_{n+1}\cdots\alpha_{n+\ell-1}}^{\mathscr{C}=a_{0}a_{1}\cdots a_{\ell-1}})(\overbrace{\alpha_{n}\alpha_{n+1}\cdots\alpha_{n+\ell-1}}^{\mathscr{C}=a_{0}a_{1}\cdots a_{\ell-1}})\cdots (we take \alpha_{1}\cdots\alpha_{n-1}=e_{\mathfrak{s}(\alpha_{n})} in the case for n=0 ) such that \beta_{2,1}\alpha_{1}\in\mathcal{I} (we take \beta_{2,1}\alpha_{0}\in\mathcal{I} in the case for n=0 ). By using an argument similar to the proof of the necessity, we get that (A) If n=0 , then \beta_{2,1}F is a forbidden path which admits the vertex \mathfrak{s}(\alpha_{0})=\mathfrak{t}(\beta_{2,1})=\mathfrak{t}(\alpha_{\ell-1}) on the forbidden cycle \mathscr{C} which is not invalid (B) If n\geqslant 1 , then \mathfrak{s}(\alpha_{n})=\mathfrak{t}(\alpha_{n-1})=\mathfrak{t}(\alpha_{n+\ell-1}) is a vertex on the forbidden cycle \mathscr{C} which is not invalid If \ell(\delta)=0 , then \delta=e_{v} is a path of length zero corresponding to some vertex v , and \mathcal{F}(\delta) contains a forbidden path F (which is of the form as shown in ( 5.1 )) with \ell(F)=\infty such that there is an arrow \alpha starting at v and ending at \mathfrak{s}(F) such that \alpha F is a forbidden path. It follows the vertex \mathfrak{s}(\alpha_{n}) is not invalid by using an argument similar to the proof of the necessity. ∎ Corollary 5.2 . Let A=\mathds{k}\mathcal{Q}/\mathcal{I} be an almost gentle algebra. If \mathrm{inj.dim}A=\infty , then the following statements hold. (1) The bound quiver (\mathcal{Q},\mathcal{I}) has at least one forbidden cycle. (2) There is a vertex v on some forbidden cycle \mathscr{C}=a_{0}a_{1}\cdots a_{\ell-1} such that one of the following statements holds: (A) there is an arrow \alpha ending at v satisfying \alpha a_{v}\in\mathcal{I} ; (B) there is an arrow \beta starting at v satisfying a_{\overline{v-1}}\beta\in\mathcal{I} , where \overline{v-1} is v-1 modulo \ell . Proof. The statement (1) holds by Theorem 5.1 . Now we prove (2). By Theorem 5.1 , we can find a vertex v on some forbidden cycle \mathscr{C}=a_{0}a_{1}\cdots a_{\ell-1} such that v is not invalid ​​. Assume that v is a (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex ( c\geqslant 1 , d\geqslant 1 ), and for all arrows \beta_{1},\ldots,\beta_{d} starting at v (see F IGURE 4.1 ), we have a_{\overline{v-1}}\beta_{j}\notin\mathcal{I} holds for any 1\leqslant j\leqslant d . Then, by the definition of almost gentle algebras, we have d=1 . (a) When c\geqslant 2 , if there are two integers i_{1} and i_{2} with 1\leqslant i_{1},i_{2}\leqslant c such that \alpha_{i_{1}}a_{v} and \alpha_{i_{2}}a_{v} are paths of length two lying in \mathcal{I} , then it contradicts the definition of almost gentle pairs. (b) When c=1 , there is no arrow \alpha starting at v such that \alpha a_{v} lies in \mathcal{I} . Then v is invalid by Definition 4.11 (5), a contradiction. (Notice that if all vertices on forbidden cycle satisfy the case (b), then (\mathcal{Q},\mathcal{I}) is a forbidden cycle in the case for \mathcal{Q} to be connected. Furthermore, A is self-injective in this case.) ∎ Lemma 5.3 . Let A=\mathds{k}\mathcal{Q}/\mathcal{I} be an almost gentle algebra such that the bound quiver (\mathcal{Q},\mathcal{I}) contains at least one forbidden cycle. Assume that there is a vertex v on some forbidden cycle \mathscr{C}=a_{0}a_{1}\cdots a_{\ell-1} such that one of the following conditions holds: (A) there is an arrow \alpha ending at v satisfying \alpha a_{v}\in\mathcal{I} ; (B) there is an arrow \beta starting at v satisfying a_{\overline{v-1}}\beta\in\mathcal{I} . Then \mathrm{inj.dim}A=\infty . Proof. Assume that v is a (c^{\mathrm{in}},d^{\mathrm{out}}) -type vertex as shown in F IGURE 4.2 . In Case (A), we have c\geqslant 2 and d\geqslant 1 , see F IGURE 5.1 . \overline{v-1} v \overline{v+1} a_{\overline{v-1}} a_{v} \alpha w Figure 5.1. There exists an arrow \alpha starting with v such that \alpha a_{v}\in\mathcal{I} If c>2 , then v is invalid if and only if d=0 , see Definition 4.11 (2). If c=2 , then v is invalid if and only if v is gentle, see Definition 4.11 (1). Thus, v is not invalid in Case (A). Let \xi be the anti-claw corresponding to E(v) . Then \mathcal{F}(\xi) contains the forbidden path F=(a_{v}a_{\overline{v+1}}\cdots a_{\ell-1})(\overbrace{a_{0}a_{1}\cdots a_{\ell-1}}^{\mathscr{C}})(\overbrace{a_{0}a_{1}\cdots a_{\ell-1}}^{\mathscr{C}})\cdots whose length is infinite, and thus \mathrm{proj.dim}E(v)=\infty by Theorem 4.19 . In Case (B), we have c\geqslant 1 and d\geqslant 2 . If c\geqslant 2 , then one can check that v is not invalid by Definition 4.11 (1)(2). In this case, we obtain \mathrm{proj.dim}E(v) is infinite by an argument similar to the proof of Case (A), and so is \mathrm{proj.dim}D(A) ( =\mathrm{inj.dim}A ), as required. Now, consider the case for c=1 in Case (B). Assume \mathfrak{t}(\beta)=w . Then we have S(v)\leq_{\oplus}\mathrm{top}(E(w)) . It follows that \Omega_{1}(E(w)) has a direct summand, written as M , such that • M is a right maximal directed string module; • \mathrm{top}M=S(\overline{v+1}) . Then \mathrm{proj.dim}M=\infty since \overline{v+1} is a vertex on \mathscr{C} , and so is \mathrm{proj.dim}E(w) . Thus \mathrm{proj.dim}D(A)=\infty , as required. ∎ The following result provides a description of the infiniteness of \mathrm{inj.dim}A . Theorem 5.4 . Let A=\mathds{k}\mathcal{Q}/\mathcal{I} be an almost gentle algebra such that the bound quiver (\mathcal{Q},\mathcal{I}) contains at least one forbidden cycle. \mathscr{C}=a_{0}a_{1}\cdots a_{\ell-1} . Then \mathrm{inj.dim}A=\infty if and only if there is a vertex v on \mathscr{C} such that one of the conditions (A) and (B) given in Lemma 5.3 holds. Proof. It follows from Lemma 5.3 and Corollary 5.2 . ∎ 5.2. Auslander–Reiten Conjecture Lemma 5.5 . Let A=\mathds{k}\mathcal{Q}/\mathcal{I} be an almost gentle algebra whose bound quiver (\mathcal{Q},\mathcal{I}) has at least one forbidden cycle. If \mathrm{inj.dim}A=\infty , then the injective envelope E^{0} of A has an infinite projective dimension. Proof. By Theorem 5.4 , there exists a vertex v on some forbidden cycle \mathscr{C}=a_{0}a_{1}\cdots a_{\ell-1} such that one of the conditions (A) and (B) given in Lemma 5.3 holds. If A satisfies the condition (A), that is, there is an arrow \alpha ending at v satisfying \alpha a_{v}\in\mathcal{I} , then there are two cases as follows: (1) For any arrow \beta starting at v , we have \alpha\beta\in\mathcal{I} . (2) There exists a unique arrow \beta with \mathfrak{s}(\beta)=v , we have \alpha\beta\notin\mathcal{I} . In Case (1), assume \mathfrak{s}(\alpha)=w (cf. F IGURE 5.1 ). Then the injective envelope E^{0}_{P(w)} of P(w) has a direct summand which is isomorphic to E(v) . We have that the projective dimension \mathrm{proj.dim}E(v) of E(v) is infinite, see the proof of Lemma 5.3 (A), thus \mathrm{proj.dim}E^{0}_{P(w)}=\infty , it follows that \mathrm{proj.dim}E^{0}=\infty . In Case (2), there is a right maximal directed string p=\beta_{1}\beta_{2}\cdots\beta_{l} ( \beta=\beta_{1} ) such that \alpha p\notin\mathcal{I} , cf. F IGURE 5.2 . \overline{v-1} v \overline{v+1} a_{\overline{v-1}} a_{v} \alpha w \beta_{1}=\beta \beta_{l} u_{1} u_{l-1} u_{1} Figure 5.2. There exists an path p=\beta_{1}\cdots\beta_{l} starting with v such that \alpha p\notin\mathcal{I} (i.e., such that \alpha\beta_{1}\notin\mathcal{I} ) In this case, the injective envelope E^{0}_{P(v)} of P(v) has a direct summand E(\mathfrak{t}(\beta)) (i.e., E(u_{1}) , where u_{1} is shown in F IGURE 5.2 ). By the definition of almost gentle algebras, we have a_{\overline{v-1}}\beta\in\mathcal{I} , and then the anti-claw \xi corresponding to E(u_{1}) is of the form \xi=\ r_{1}{\ \shortmid\mkern-13.0mu\vee\ }r_{2}{\ \shortmid\mkern-13.0mu\vee\ }\cdots{\ \shortmid\mkern-13.0mu\vee\ }r_{m} such that r_{1}=p . Then F=(a_{\overline{v+1}}\cdots a_{\ell-1})(\overbrace{a_{0}a_{1}\cdots a_{\ell-1}}^{\mathscr{C}})(\overbrace{a_{0}a_{1}\cdots a_{\ell-1}}^{\mathscr{C}})\cdots is a forbidden path lying in \mathcal{F}(\xi) whose length is infinite. By Theorem 4.19 , we have \mathrm{proj.dim}E(u_{1})=\infty , and so \mathrm{proj.dim}E^{0}=\infty . ∎ Corollary 5.6 . Let A be an almost gentle algebra with \mathrm{inj.dim}A=\infty , then \mathrm{proj.dim}E^{0}=\infty , where E^{0} is the injective envelope of A_{A} . Proof. Since \mathrm{inj.dim}A=\infty , there is an indecomposable injective module E(v) such that \mathrm{proj.dim}E(v)=\infty . Then by Theorem 4.19 , the set \mathcal{F}(\xi) given by the anti-claw \xi corresponding to E(v) contains a forbidden path F whose length is infinite. Thus F provides a forbidden cycle \mathscr{C} , that is, (\mathcal{Q},\mathcal{I}) contains a forbidden cycle. Then \mathrm{proj.dim}E^{0}=\infty by Lemma 5.5 , ∎ As a consequence, we obtain the following result. Theorem 5.7 . ARC holds true for almost gentle algebras. Proof. Let A be an almost gentle algebra. If \mathrm{inj.dim}A<\infty , then A is Gorenstein. If \mathrm{inj.dim}A=\infty , then A does not satisfy the Auslander condition by Corollary 5.6 . The proof is finished. ∎