On momoid graded semihereditary rings Thanks: 2020 Mathematics Subject Classification: 16W50, 16D40, 16D50, 16E60 Thanks: Key Words and Phrases: Graded left hereditary ring, graded left semihereditary ring, graded-Prüfer domain, graded injective module, graded flat module Haneen Falah Ghalib Al-Kharsan, Parviz Sahandi and Nematollah Shirmohammadi Abstract . In order to study graded left hereditary and left semihereditary rings graded by a cancelation monoid in terms of their modules, we need to revisit graded free, projective, injective, and flat modules and provide graded versions of specific results concerning these modules, like Baer’s criterion on injectivity and Lazard’s theorem on flatness. Then, among other things, we can give some characterization of graded left hereditary and left semihereditary rings, in particular, of graded-Prüfer and graded-Dedekind domains. 1. Introduction After Kaplansky’s work on Dedekind and Prüfer domains [ 11 ] , Cartan and Eilenberg defined the notion of hereditary and semihereditary rings in [ 7 ] . Let R be an associative ring with nonzero identity. Recall that R is left hereditary if every left ideal is a projective R -module. It is known that R is hereditary if and only if each submodule of a projective left R -module is projective if and only if each quotient of an injective left R -module is injective [ 7 , Page 14] . A hereditary domain is a Dedekind domain. It was proved that R is a Dedekind domain if and only if every divisible R -module is injective [ 18 , Theorem 4.24] . Recall also that R is left semihereditary if every finitely generated left ideal is a projective R -module. It is known that R is left semihereditary if and only if R is left coherent, and each submodule of a flat left R -module is flat if and only if each torsion-free right R -module is flat [ 18 , Theorem 4.32] . A semihereditary domain is a Prüfer domain. It is shown that R is a Prüfer domain if and only if every finitely generated torsion-free R -module is projective [ 18 , Theorem 4.34] . The graded analogue of Dedekind and Prüfer domains was defined and studied by several authors (see [ 14 , 2 , 4 , 19 ] ). This paper aims to define and study the graded analogues of hereditary and semihereditary rings. Let \Gamma be a cancelation monoid, and R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} be a \Gamma -graded ring. To this purpose, one needs to study \Gamma -graded free, projective, injective, and flat modules. These modules were extensively studied in case where \Gamma=\mathbb{Z} in [ 9 , 6 ] , and in case where \Gamma is a group in [ 15 ] . The first two of them were investigated in case where \Gamma is a cancelation monoid in [ 13 ] . Then, we examine \Gamma -graded injective and flat modules where \Gamma is a cancelation monoid. The organization of this paper is as follows. In the remainder of the introduction and Section 2, we recall some preliminaries on graded rings and modules. In Section 3, after reviewing some results on the graded free and projective modules, we investigate the graded divisible, injective, and flat modules. We provide a graded analogue of Baer’s Theorem and show that every graded injective module is graded divisible and every R -module admits a graded injective envelope. Then, we define and study graded flat modules and extend Lazard’s Theorem to the monoid-graded case. More precisely, it is shown that any graded flat module is a direct limit of finitely generated graded free modules. In section 4, we study graded hereditary rings. A \Gamma -graded ring R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is called graded left hereditary if every left homogeneous ideal is projective. Then, among other things, we prove the graded version of the Cartan-Eilenberg Theorem: R is graded left hereditary if and only if every graded submodule of a graded projective R -module is graded projective if and only if every quotient of a graded injective R -module is graded injective. Then we show that a graded (commutative) domain is graded-Dedekind if and only if every graded-divisible left R -module is graded-injective. Finally, in Section 5, we study graded semihereditary rings. A \Gamma -graded ring R is called graded left semihereditary if every finitely generated left homogeneous ideal is projective. Among other things, it is shown that R is graded left semihereditary if and only if every finitely generated graded submodule of a graded projective left R -module is graded projective if and only if R is graded left coherent and every homogeneous left ideal is graded flat. It is also shown that a graded integral domain R is a graded-Prüfer domain if and only if every finitely generated torsion-free graded R -module is graded projective. 1.1. Graded Rings and modules Let \Gamma be a cancelation monoid, that is, \Gamma is a semigroup with a neutral element \varepsilon , not necessarily commutative, and \Gamma satisfies the (left and right) cancelation law: \alpha\alpha_{1}=\alpha\alpha_{2} implies \alpha_{1}=\alpha_{2} , and \alpha_{1}\alpha=\alpha_{2}\alpha implies \alpha_{1}=\alpha_{2} for all \alpha,\alpha_{1},\alpha_{2}\in\Gamma . By a \Gamma -graded ring R , we mean an associative ring with the multiplicative identity 1, which has a direct sum decomposition R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} , where each R_{\alpha} is an additive subgroup of R such that R_{\alpha}R_{\beta}\subseteq R_{\alpha\beta} for all \alpha,\beta\in\Gamma . For each \alpha\in\Gamma , we call 0\neq a\in R_{\alpha} a homogeneous element of degree \alpha and write \deg(a)=\alpha . For the basics about graded rings and graded modules we used in this paper, one may refer to [ 15 ] , though it mainly deals with group G -graded rings. Let R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} be a \Gamma -graded ring as above. By the definition, it is clear that the degree- \varepsilon part R_{\varepsilon} is a subring of R , and by using the cancelation law on \Gamma , it is easy to check that 1\in R_{\varepsilon} . Let I be a left (resp. right) ideal of R . If I is generated by homogeneous elements, then we say that I is a homogeneous left (resp. right) ideal of R . By using the cancelation law on \Gamma , it is straightforward to check that I is a homogeneous left (resp. right) ideal of R if and only if I=\bigoplus_{\alpha\in\Gamma}(I\cap R_{\alpha}) if and only if a=\Sigma_{i}a_{\alpha_{i}}\in I with a_{\alpha_{i}}\in R_{\alpha_{i}} implies a_{\alpha_{i}}\in I for all i if and only if R/I=\bigoplus_{\alpha\in\Gamma}(R_{\alpha}+I)/I . If each graded left ideal of R is finitely generated, then R is called a left gr-Noetherian ring . Let R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} be a \Gamma -graded ring. By a \Gamma -graded R -module M , we mean a left R -module which has a direct sum decomposition M=\bigoplus_{\alpha\in\Gamma}M_{\alpha} , where each M_{\alpha} is an additive subgroup of M , such that R_{\alpha}M_{\beta}\subseteq M_{\alpha\beta} for all \alpha,\beta\in\Gamma . Similarly, one can define a \Gamma -graded right R -module. For each \alpha\in\Gamma , we call 0\neq x\in M_{\alpha} a homogeneous element of degree \alpha and write \deg(x)=\alpha . By [ 13 , Lemma 1.1] , if \{x_{i}\}_{i\in J} is a homogeneous generating set of M with \deg(x_{i})=\alpha_{i} , i\in J , then M_{\alpha}=\sum_{i\in J\text{, }\alpha^{\prime}_{i}\alpha_{i}=\alpha}R_{\alpha^{\prime}_{i}}x_{i}. Let N be a submodule of M . If N is generated by some homogeneous elements, then we say that N is a graded submodule of M . By using the cancelation law on \Gamma , it is straightforward to check that N is a graded submodule of M if and only if N=\bigoplus_{\alpha\in\Gamma}(N\cap M_{\alpha}) if and only if x=\Sigma_{i}x_{\alpha_{i}}\in N with x_{\alpha_{i}}\in M_{\alpha_{i}} implies x_{\alpha_{i}}\in N for all i if and only if M/N=\bigoplus_{\alpha\in\Gamma}(M_{\alpha}+N)/N . We let H(M)=\bigcup_{\alpha\in\Gamma}M_{\alpha} be the set of homogeneous elements of M . Let M=\bigoplus_{\alpha\in\Gamma}M_{\alpha} and N=\bigoplus_{\alpha\in\Gamma}N_{\alpha} be \Gamma -graded left R -modules. If \varphi:M\to N is an R -module homomorphism such that \varphi(M_{\alpha})\subseteq N_{\alpha} for all \alpha\in\Gamma , then we call \varphi a homogeneous homomorphism . In the case that \varphi:M\to N is a homogeneous homomorphism, it is easy to see that \ker(\varphi) and \operatorname{im}(\varphi) are graded submodules of M and N , respectively. 2. Preliminaries Let \Gamma be a cancelation monoid with neutral element \varepsilon , and R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} a \Gamma -graded ring. In this section, we recall some preliminaries of \Gamma -graded R -modules. Let \{M_{i},\varphi_{ij}\}_{i,j\in I} be a direct system of \Gamma -graded left R -modules, i.e., I is a directed partially ordered set and, for i\leq j , \varphi_{ij}:M_{i}\to M_{j} is a homogeneous homomorphism which is compatible with the ordering. Then M=\varinjlim_{i\in I}M_{i} is a \Gamma -graded left R -module with homogeneous components M_{\alpha}=\varinjlim_{i\in I}M_{i\alpha} . For a detailed construction of such direct limits see [ 5 , II, Section 11.3, Remark 3] . In particular, let \{M_{i}\mid i\in I\} be \Gamma -graded left R -modules. Then \bigoplus_{i\in I}M_{i} has a natural \Gamma -graded left R -module given by (\bigoplus_{i\in I}M_{i})_{\alpha}=\bigoplus_{i\in I}M_{i\alpha} , for all \alpha\in\Gamma . Now, let \{N_{i},\psi_{ij}\}_{i,j\in I} be another direct system of \Gamma -graded left R -modules and set N=\varinjlim_{i\in I}N_{i} . We say that \{\tau_{i}\}:\{M_{i},\varphi_{ij}\}\to\{N_{i},\psi_{ij}\} is a morphism of direct systems if \tau_{i}:M_{i}\to N_{i} , i\in I , is a homogeneous homomorphism, such that the following diagram is commutative for all i\leq j : \textstyle{M_{i}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\tau_{i}} \scriptstyle{\varphi_{ij}} \textstyle{N_{i}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\psi_{ij}} \textstyle{M_{j}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\tau_{j}} \textstyle{N_{j}.} In this case, there exists a homomorphism \tau:M\to N such that the diagram \textstyle{M_{i}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\tau_{i}} \scriptstyle{\varphi_{i}} \textstyle{N_{i}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\psi_{i}} \textstyle{M\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\tau} \textstyle{N,} commutes for all i\in I , where \varphi_{i},\psi_{i} are natural homogeneous homomorphisms. It can be seen that \tau is homogeneous. Let M , N be \Gamma -graded left R -modules and \alpha\in\Gamma . A homogeneous R -module homomorphism of degree \alpha is an R -module homomorphism f:M\to N , such that f(M_{\beta})\subseteq N_{\beta\alpha} for any \beta\in\Gamma . Let \operatorname{Hom}_{R}(M,N) be the group of all R -module homomorphism from M to N , and denote by \operatorname{Hom}_{\alpha}(M,N) the subgroup of \operatorname{Hom}_{R}(M,N) consisting of all homogeneous R -module homomorphisms of degree \alpha . By using the cancelation law on \Gamma , it is easy to see that \operatorname{Hom}_{\alpha}(M,N) , \alpha\in\Gamma , constitutes a direct sum. We put \sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(M,N):=\bigoplus_{\alpha\in\Gamma}\operatorname{Hom}_{\alpha}(M,N). Again, by using the cancelation law on \Gamma , one notices that the functors \sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(-,N) and \sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(M,-) are left exact on the category of \Gamma -graded left R -modules. In general, \sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(M,N) may not equal to \operatorname{Hom}_{R}(M,N) , but under certain conditions we have \sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(M,N)=\operatorname{Hom}_{R}(M,N) , for instance, for the case that \Gamma=G is a group, if M is finitely generated ( [ 15 , Corollary 2.4.4] ), or if both M and N have finite support ( [ 15 , Corollary 2.4.5] ), in particular in the case where the grading group G is finite. In a similar manner, for \Gamma -graded rings R and S , we can consider the \Gamma -graded R-S -bimodule N . That is, N is a R-S -bimodule and additionally N=\bigoplus_{\alpha\in\Gamma}N_{\alpha} is a \Gamma -graded left R -module and a \Gamma -graded right S -module, i.e., R_{\alpha}N_{\gamma}S_{\beta}\subseteq N_{\alpha\gamma\beta} , where \alpha,\gamma,\beta\in\Gamma . Assume now that N is a graded R-S -bimodule, then it is clear that \sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(M,N) has a natural structure of a \Gamma -graded right S -module, with homogeneous components \operatorname{Hom}_{\alpha}(M,N) , \alpha\in\Gamma . Let M be a \Gamma -graded right R -module and N be a \Gamma -graded left R -module. We will observe that the tensor product M\otimes_{R}N has a natural \Gamma -graded \mathbb{Z} -module structure (here \mathbb{Z} has trivial \Gamma -graded structure). Since each of M_{\alpha} , \alpha\in\Gamma , is a right R_{\varepsilon} -module and similarly N_{\alpha} , \alpha\in\Gamma , is a left R_{\varepsilon} -module, then M\otimes_{R_{\varepsilon}}N can be decomposed as a direct sum M\otimes_{R_{\varepsilon}}N=\bigoplus_{\alpha\in\Gamma}(M\otimes N)_{\alpha}, where (M\otimes N)_{\alpha}=\{\sum_{i}m_{i}\otimes n_{i}\mid m_{i}\in H(M),n_{i}\in H(N),\deg(m_{i})+\deg(n_{i})=\alpha\}. It now follows that M\otimes_{R}N=(M\otimes_{R_{\varepsilon}}N)/J , where J is a subgroup of M\otimes_{R_{\varepsilon}}N generated by the homogeneous elements \{mr\otimes n-m\otimes rn\mid m\in H(M),n\in H(N),r\in H(R)\}. This shows that M\otimes_{R}N is also a \Gamma -graded \mathbb{Z} -module. Lemma 2.1 . Let I be a directed partially ordered set, and \{M_{i},\varphi_{ij}\} , \{N_{i},\psi_{ij}\} , and \{L_{i},\theta_{ij}\} be direct systems of \Gamma -graded left R -modules. (1) Assume that there are morphisms of direct systems over I \{M_{i},\varphi_{ij}\}\to\{N_{i},\psi_{ij}\}\to\{L_{i},\theta_{ij}\} such that the sequence M_{i}\to N_{i}\to L_{i} is exact for each i\in I . Then there is an exact sequence of \Gamma -graded left R -modules with homogeneous homomorphisms \underset{i\in I}{\varinjlim}M_{i}\to\underset{i\in I}{\varinjlim}N_{i}\to\underset{i\in I}{\varinjlim}L_{i}. (2) Assume that A is a \Gamma -graded right R -module. Then there is a homogeneous isomorphism A\otimes_{R}(\underset{i\in I}{\varinjlim}M_{i})\cong\underset{i\in I}{\varinjlim}(A\otimes_{R}M_{i}). (3) Assume that J is a cofinal subset in I . Then J is directed and there is a homogeneous isomorphism \underset{j\in J}{\varinjlim}M_{j}\cong\underset{i\in I}{\varinjlim}M_{i} . Proof. For (1), it remains to prove the exactness, which in turn follows by [ 17 , Proposition 8.9] , and (2) follows as in the ungraded case [ 17 , Corollary 8.8] , while (3) is an easy exercise. ∎ 3. Graded projective, injective, and flat modules Let \Gamma be a cancelation monoid with neutral element \varepsilon and R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} a \Gamma -graded ring. In this section, we investigate graded projective, injective, and flat modules, and we compare each of them with the usual ungraded one. 3.1. Graded-projective modules Recall that a \Gamma -graded left R -module F=\bigoplus_{\alpha\in\Gamma}F_{\alpha} is called graded free (for short, gr-free ) if F is a free R -module (in the usual sense) but with a homogeneous free R -basis; namely, F has a free R -basis \{e_{i}\}_{i\in J} consisting of homogeneous elements. Given any free R -module F=\bigoplus_{i\in J}Re_{i} and an arbitrarily chosen family \{\alpha_{i}\}_{i\in J} of elements of \Gamma , one can give a \Gamma -graduation on F such that \{e_{i}\}_{i\in J} will be a homogeneous basis with \deg(e_{i})=\alpha_{i} . This shows that if M is a \Gamma -graded left R -module with homogeneous generating set \{x_{i}\}_{i\in J} , and the gr-free R -module F=\bigoplus_{i\in J}Re_{i} , with \deg(e_{i})=\deg(x_{i}) , then the map \varphi:F\to M defined by \varphi(e_{i})=x_{i} is a homogeneous epimorphism. Hence, every nonzero \Gamma -graded left R -module is a graded homomorphic image of some gr-free R -module (see [ 13 , Page 2698] for details). As in [ 13 ] , a graded left R -module P is called a graded projective module (for short a gr-projective module ) if for any diagram of \Gamma -graded left R -modules and homogeneous homomorphisms \textstyle{P\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{f} \scriptstyle{h} \textstyle{M\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{g} \textstyle{N\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{0.} there is a homogeneous homomorphism h:P\to M with gh=f . Proposition 3.1 . Assume that \Gamma is a cancelation monoid, R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a \Gamma -graded ring and P is a \Gamma -graded left R -module. Then the following are equivalent: (1) P is \Gamma -graded and projective. (2) P is gr-projective. (3) \sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(P,-) is an exact functor in the category of \Gamma -graded left R -modules. (4) Every short exact sequence of \Gamma -graded left R -module and homogeneous homomorphisms 0\to L\stackrel{{\scriptstyle f}}{{\to}}M\stackrel{{\scriptstyle g}}{{\to}}P\to 0 splits via a homogeneous homomorphism. (5) P is a graded direct summand of a gr-free \Gamma -graded left R -module. Proof. For (1)\Leftrightarrow(2)\Leftrightarrow(5) see [ 13 , Proposition 1.2] . (2)\Rightarrow(3) As we have already seen \sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(P,-) is left exact. The right exactness follows immediately from the definition of gr-projective modules. (3)\Rightarrow(4) Applying the exact functor \sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(P,-) to the short exact sequence 0\to L\stackrel{{\scriptstyle f}}{{\to}}M\stackrel{{\scriptstyle g}}{{\to}}P\to 0 gives rise to a homogeneous homomorphism h:P\to M such that gh=1_{P} . (4)\Rightarrow(5) As we recalled above, there exists a homogeneous homomorphism \varphi:F\to P , where F is a gr-free \Gamma -graded left R -module. By the assumption, there is a homogeneous homomorphism h:P\to F such that gh=1_{P} . Then \theta:P\oplus\ker(g)\to F given by \theta(x,y)=h(x)+y is a homogeneous isomorphism. ∎ Assume that M is a finitely presented \Gamma -graded left R -module. Let F_{1}\to F_{0}\stackrel{{\scriptstyle f}}{{\to}}M\to 0 be a finite presentation of M (in the usual sense); so that 0\to\ker(f)\to F_{0}\stackrel{{\scriptstyle f}}{{\to}}M\to 0 is exact. Note that M is finitely generated; so we can choose a finite set of generators x_{1},\ldots,x_{n} , such that each x_{i} is a homogeneous element of degree \alpha_{i} . Let F^{\prime}_{0}:=\oplus_{i=1}^{n}Re_{i} be the gr-free \Gamma -graded left R -module with homogeneous basis \{e_{1},\ldots,e_{n}\} , \deg(e_{i})=\alpha_{i} . Define the homogeneous homomorphism f^{\prime}:F^{\prime}_{0}\to M by e_{i}\mapsto x_{i} . Hence, there is an exact sequence 0\to\ker(f^{\prime})\to F^{\prime}_{0}\stackrel{{\scriptstyle f^{\prime}}}{{\to}}M\to 0 . It follows from Schanuel’s Lemma [ 18 , Proposition 3.12] that \ker(f)\oplus F^{\prime}_{0}\cong\ker(f^{\prime})\oplus F_{0} . Since \ker(f) is finitely generated, we see that \ker(f^{\prime}) is finitely generated, too. Hence \ker(f^{\prime}) is an image of a finitely generated gr-free left R -module F^{\prime}_{1} . Therefore, M is finitely presented in the category of \Gamma -graded left R -modules. 3.2. Graded-injective modules The notion of graded injective modules was investigated in [ 9 , 6 ] in the case where \Gamma=\mathbb{Z} and in [ 15 ] in the case where \Gamma=G is a group. It is known that the notions of graded injective and (usual) injective are not the same. For example, over an arbitrary field k , consider the Laurent polynomial ring R=k[X,X^{-1}] with \mathbb{Z} -gradation. Then by [ 15 , Remark 2.3.3] , the left R -module R is graded injective but not injective. A \Gamma -graded left R -module E is said to be a graded injective module (for short, gr-injective module ) if for every homogeneous homomorphisms f:M\to E and g:M\to N with g is a homogeneous injection, there exists a homogeneous homomorphism h:N\to E such that the diagram \textstyle{0\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{M\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{g} \scriptstyle{f} \textstyle{N\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{h} \textstyle{E} commutes. We now provide some useful characterization of gr-injective modules, which will be used in the sequel (see [ 15 , Corollary 2.4.8] when \Gamma=G is a group). Theorem 3.2 . (The graded version of Baer’s Theorem) Let \Gamma be a cancelation monoid and M be a \Gamma -graded left R -module. Then M is gr-injective if and only if for any homogeneous left ideal I of R , any \alpha\in\Gamma and any homogeneous homomorphism f:I\to M of degree \alpha , there exists m_{\alpha}\in M_{\alpha} such that f(r)=rm_{\alpha} for any r\in I . Proof. Assume that M is gr-injective, I is a homogeneous left ideal of R , and f:I\to M is a homogeneous homomorphism of degree \alpha\in\Gamma . There is a gr-free left R -module F=Re with homogeneous basis \{e\} of degree \alpha . Consider the following diagram \textstyle{0\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{Ie\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\overline{f}} \textstyle{F} \textstyle{M,} where \overline{f}(re)=f(r) , which is a homogeneous homomorphism. By assumption there is a homogeneous homomorphism \overline{g}:F\to M such that \overline{g}\mid_{Ie}=\overline{f} . Putting m_{\alpha}:=\overline{g}(e) , we have f(r)=\overline{f}(re)=\overline{g}(re)=rm_{\alpha} . Conversely, assume we have the following diagram with exact row of \Gamma -graded left R -modules and homogeneous homomorphisms: \textstyle{0\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{A\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\varphi} \scriptstyle{f} \textstyle{B} \textstyle{M.} Let \Sigma be the set of all ordered pairs (B^{\prime},g^{\prime}) where \varphi(A)\subseteq B^{\prime}\subseteq B , B^{\prime} is a graded submodule of B and \textstyle{0\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{A\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\varphi} \scriptstyle{f} \textstyle{B^{\prime}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{g^{\prime}} \textstyle{M} commutes. Partially order \Sigma by (B^{\prime\prime},g^{\prime\prime})\leq(B^{\prime},g^{\prime}) when B^{\prime\prime}\subseteq B^{\prime} and g^{\prime}\mid_{B^{\prime\prime}}=g^{\prime\prime} . Then \Sigma is nonempty and inductive. Let (B^{\prime},g^{\prime}) be a maximal element of \Sigma . We claim that B^{\prime}=B . Otherwise, choose a homogeneous element x\in B\setminus B^{\prime} of degree \alpha . Let I=\{r\in R\mid rx\in B^{\prime}\} which is a homogeneous left ideal of R . Define \overline{f}:I\to M by \overline{f}(r)=g^{\prime}(rx) for r\in I . It is clear that \overline{f} is a homogeneous homomorphism of degree \alpha . By assumption, there is m_{\alpha}\in M_{\alpha} such that \overline{f}(r)=rm_{\alpha} for r\in I . Set B^{\prime\prime}:=B^{\prime}+Rx and define g^{\prime\prime}:B^{\prime\prime}\to M by g^{\prime\prime}(b+rx)=g^{\prime}(b)+rm_{\alpha} . It can be see that g^{\prime\prime} is a well defined homomorphism. We show that g^{\prime\prime} is homogeneous. Indeed, assume that \gamma\in\Gamma and b^{\prime\prime}\in B^{\prime\prime}_{\gamma} . Then there are b\in B^{\prime}_{\gamma} and r\in R_{\beta} with \beta\alpha=\gamma such that b^{\prime\prime}=b+rx . Hence rm_{\alpha} is homogeneous of degree \beta\alpha=\gamma ; so g^{\prime\prime}(b^{\prime\prime})=g^{\prime}(b)+rm_{\alpha} is homogeneous of degree \gamma . This means that (B^{\prime\prime},g^{\prime\prime}) properly extends (B^{\prime},g^{\prime}) , contradicting the maximality of (B^{\prime},g^{\prime}) . ∎ Proposition 3.3 . Assume that \Gamma is a cancelation monoid, R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a \Gamma -graded ring and E is a \Gamma -graded left R -module. Then the following are equivalent: (1) E is gr-injective. (2) \sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(-,E) is an exact functor in the category of \Gamma -graded left R -modules. (3) Every short exact sequence of \Gamma -graded left R -module and homogeneous homomorphisms 0\to E\to M\to N\to 0 splits via a homogeneous homomorphism. Proof. (1)\Rightarrow(2) and (2)\Rightarrow(3) are easy and (3)\Rightarrow(1) is straightforward by pushout construction. ∎ It is well known that every (not necessarily graded) injective R -module is divisible, and the converse is true if R is a Dedekind domain [ 18 , Theorem 4.24] . The notion of a graded divisible module was introduced in [ 10 , Page 47] for group-graded modules. This notion can be generalized to monoid-graded modules as in the following. Recall that a nonzero element a\in R is a right zero divisor if there exists a nonzero element b\in R such that ba=0 . The element a\in R is called to be a right nonzero divisor if it is not a right zero divisor. Definition 3.4 . Assume that \Gamma is a cancelation monoid. A \Gamma -graded left R -module M is called graded divisible (for short, gr-divisible) if for \alpha,\beta,\gamma\in\Gamma with \alpha\gamma=\beta and every right nonzero divisor r\in R_{\alpha} and every y\in M_{\beta} , there exists x\in M_{\gamma} such that rx=y . It is clear that every graded and divisible R -module is gr-divisible. But the converse is no longer true. For, over an arbitrary field k , consider the Laurent polynomial ring R=k[X,X^{-1}] with \mathbb{Z} -gradation. Then by [ 15 , Remark 2.3.3] , the left R -module R is gr-injective but not injective. Hence, it is gr-divisible by the following proposition but not divisible since R is a Dedekind domain. Proposition 3.5 . Assume that \Gamma is a cancelation monoid. Any gr-injective left R -module is gr-divisible. Proof. Assume that E is a gr-injective left R -module, and choose \alpha,\beta,\gamma\in\Gamma with \alpha\gamma=\beta , a right nonzero divisor r\in R_{\alpha} and y\in M_{\beta} . It follows that the map \varphi:Ra\to E , defined by \varphi(ra)=ry is well defined, and using the cancelation law on \Gamma , it is a homogeneous homomorphism of degree \gamma . Then by Theorem 3.2 , there exists x\in E_{\gamma} such that \varphi(r)=rx for r\in Ra . Hence, y=\varphi(a)=ax . ∎ In the following, we show that every \Gamma -graded left R -module has a graded injective envelope. For this, we follow [ 6 , Sections 3.2 and 3.6] . Let M be a \Gamma -graded left R -module, and let N be a graded submodule of M . The extension N\subseteq M is called a graded-essential extension (for short, gr-essential extension ) if U\cap N\neq 0 for any graded submodule 0\neq U\subseteq M . If in addition N\neq M , the extension is called a proper gr-essential extension . It can be seen as in the case where \Gamma=\mathbb{Z} (see [ 9 , Lemma 1.1] ) or \Gamma=G a group (see [ 15 , Proposition 2.3.5] ), one can show that N\subseteq M is a gr-essential extension if and only if it is an essential extension. Proposition 3.6 . A \Gamma -graded left R -module N is gr-injective if and only if it has no proper gr-essential extension. Proof. Assume that M is a \Gamma -graded left R -module such that N is a graded submodule of M . If N is gr-injective, then there exists a graded submodule W of M such that M=N\oplus W by Proposition 3.3 . Then N\cap W=0 and so, if the extension is gr-essential, W=0 . It follows that N=M . Conversely, suppose that N has no proper gr-essential extension. Given a homogeneous monomorphism \varphi:U\to V and a homogeneous homomorphism f:U\to M , we want to construct a homogeneous homomorphism h:V\to N such that f=h\phi . We consider the pushout diagram \textstyle{U\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\varphi} \scriptstyle{f} \textstyle{V\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{g} \textstyle{N\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\psi} \textstyle{W.} Indeed, we may choose W:=(V\oplus N)/C with C:=\{(\varphi(x),-f(x))\mid x\in U\} which is a graded submodule of V\oplus N ; g and \psi are the natural homogeneous projections. Here \psi is a homogeneous monomorphism, since \varphi is a monomorphism. Thus we may consider N as a graded submodule of W . It now follows from Zorn’s lemma that there exists a maximal graded submodule D\subseteq W such that N\cap D=0 , and so N may even be considered as a graded submodule of W/D ; obviously, W/D is a gr-essential extension of N . It follows that N=W/D , since N has no proper gr-essential extension, and so W=N\oplus D . Let \pi:W\to N be the natural projection of W onto the first summand. The composition \pi g:V\to N is an extension of f . ∎ We now prove that any \Gamma -graded left R -module has a graded injective envelope. In analogy to the definition in the ungraded case, a \Gamma -graded left R -module E is called a gr-injective envelope of M if it is gr-injective and a gr-essential extension of M . Theorem 3.7 . Any \Gamma -graded left R -module M admits a gr-injective envelope, and any two gr-injective envelopes are isomorphic as \Gamma -graded left R -module. Proof. We embed M into a (not necessarily graded) injective R -module T . We consider the set \mathcal{S}:=\{N\mid M\subseteq N\text{ is gr-essential and }N\text{ is a submodule of }T\} . We define a partial order \leq on \mathcal{S} by setting N_{1}\leq N_{2} if N_{1} is a graded submodule of N_{2} . By Zorn’s lemma, \mathcal{S} has a maximal element E ; hence M\subseteq E is gr-essential and E is a submodule of T . Suppose E is not gr-injective; then E has a proper gr-essential extension E\subsetneq E^{\prime} by Proposition 3.6 . As T is injective, there exists an R -module homomorphism \varphi:E^{\prime}\to T (not necessarily homogeneous), extending the inclusion E\subseteq T . We claim that \varphi is injective. Indeed, assume that there is a nonzero element x\in\ker\varphi , say x=x_{\alpha_{1}}+\cdots+x_{\alpha_{n}} is the decomposition of x into nonzero homogeneous components. We show by induction on n that there exists a homogeneous element a\in R such that 0\neq ax\in E . Since \varphi\mid_{E} is injective, this gives a contradiction. If n=1 , x is homogeneous, and the assertion follows since the extension E\subseteq E^{\prime} is gr-essential. Now suppose that n>1 . We choose a homogeneous element a\in R such that 0\neq ax_{\alpha_{1}}\in E . Let x^{\prime}:=x-x_{\alpha_{1}}=x_{\alpha_{2}}+\cdots+x_{\alpha_{n}} . If ax^{\prime}=0 then 0\neq ax=ax_{\alpha_{1}}\in E and we are done. Otherwise ax^{\prime}\neq 0 , and by our induction hypothesis we may choose a homogeneous element b\in R such that 0\neq bax^{\prime}\in E . Then 0\neq bax=bax^{\prime}+bax_{\alpha_{1}}\in E . Next let \widetilde{E}:=\operatorname{im}\varphi . As \varphi is injective, we may give \widetilde{E} a natural graded structure \widetilde{E}_{\alpha}=\varphi(E^{\prime}_{\alpha}) for all \alpha\in\Gamma . Then E\subsetneq\widetilde{E} is a proper gr-essential extension with \widetilde{E}\subseteq T , contradicting the maximality of E . The uniqueness of the gr-injective envelope is easily obtained. ∎ 3.3. Graded-flat modules Lazard’s theorem says that any flat module is a direct limit of finitely generated free modules. In this subsection, we deal with graded flat modules and extend Lazard’s theorem to monoid-graded modules. Let \Gamma be a cancelation monoid and R be a \Gamma -graded ring. A \Gamma -graded left R -module F is said to be a graded-flat module (for short, gr-flat module ) if f\otimes_{R}1_{F}:M\otimes_{R}F\to N\otimes_{R}F is a homogeneous monomorphism for every homogeneous monomorphism f:M\to N of \Gamma -graded right R -modules M,N . We begin with the following simple lemma. Lemma 3.8 . Every direct limit of a directed system over a directed set I of gr-flat left R -modules is gr-flat. Proof. Assume that \{F_{i},\varphi_{ij}\} is a direct system over I of gr-flat left R -modules and that f:M\to N is a homogeneous monomorphism of graded right R -modules. Then, we obtain a commutative diagram with homogeneous homomorphisms \textstyle{M\otimes_{R}(\underset{i\in I}{\varinjlim}F_{i})\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{f\otimes 1} \textstyle{N\otimes_{R}(\underset{i\in I}{\varinjlim}F_{i})\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \textstyle{\underset{i\in I}{\varinjlim}(M\otimes_{R}F_{i})\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\vartheta} \textstyle{\underset{i\in I}{\varinjlim}(N\otimes_{R}F_{i}),} where the vertical maps are the homogeneous isomorphisms, and \vartheta is the homogeneous monomorphism by Lemma 2.1 . Hence f\otimes 1 is a monomorphism. Therefore \underset{i\in I}{\varinjlim}F_{i} is gr-flat. ∎ Assume that C is a \Gamma -graded left R -module and D is a graded submodule of C . Let \mathcal{A} and \mathcal{B} be nonempty families of graded submodules of C . As in [ 17 , Section 8.4] , by these data one can define a graded (C,D) -subquotient system subject to the following conditions: (1) \mathcal{A} and \mathcal{B} are directed under inclusions. (2) C=\bigcup_{A\in\mathcal{A}}A and D=\bigcup_{B\in\mathcal{B}}B . (3) For each B\in\mathcal{B} , there exists A\in\mathcal{A} with A\supseteq B . Given a graded (C,D) -subquotient system, set \mathcal{I}=\{(A,B)\in\mathcal{A}\times\mathcal{B}\mid A\supseteq B\}. Partially order \mathcal{I} by (A,B)\leq(A^{\prime},B^{\prime}) if and only if A\subseteq A^{\prime} and B\subseteq B^{\prime} . For i=(A,B)\in\mathcal{I} we set A=A_{i} and B=B_{i} . For i,j\in\mathcal{I} with i\leq j , define homomorphisms \varphi_{ij}:A_{i}/B_{i}\to A_{j}/B_{j} by \varphi_{ij}(x+B_{i})=x+B_{j} and \psi_{i}:A_{i}/B_{i}\to C/D by \psi_{i}(x+B_{i})=x+D . It is clear that \varphi_{ij} and \psi_{i} are homogeneous and \{A_{i}/B_{i}\}_{i\in\mathcal{I}} together with \varphi_{ij} is a direct system of graded modules. The proof of the following proposition is the same as in the ungraded case [ 17 , Proposition 8.12] , so we omit it. Proposition 3.9 . Assume \mathcal{A} and \mathcal{B} give a graded (C,D) -subquotient system. Then, as in the above notation, \mathcal{I} is directed and there is a homogeneous isomorphism of graded modules \underset{i\in\mathcal{I}}{\varinjlim}A_{i}/B_{i}\cong C/D . Lemma 3.10 . Suppose that A is a \Gamma -graded left R -module, with A=A_{1}\oplus A_{2} , where A_{1} and A_{2} are graded submodules of A . Suppose B_{1} is a graded submodule of A_{1} , and B_{2} is a graded submodule of A_{2} . Suppose \theta:A_{1}/B_{1}\to A_{2}/B_{2} is a homogeneous homomorphism, and suppose A_{1} is generated by homogeneous elements x_{1},\ldots,x_{n} . For each k=1,\ldots,n , choose the homogeneous element y_{k}\in A_{2} such that \theta(x_{k}+B_{1})=y_{k}+B_{2} . Let B be the (graded) submodule of A generated by B_{1},B_{2} , and the homogeneous elements x_{1}-y_{1},\ldots,x_{n}-y_{n} . Let \psi_{i}:A_{i}/B_{i}\to A/B be the natural homogeneous homomorphism \psi_{i}(x+B_{i})=x+B . Then \psi_{2} is a homogeneous isomorphism, and \psi_{2}\theta=\psi_{1} . Proof. Since \psi_{2} is homogeneous, it remains to show that it is an isomorphism, which in turn follows from [ 17 , Lemma 8.13] that \psi_{2} is an isomorphism and \psi_{2}\theta=\psi_{1} . ∎ Proposition 3.11 . Assume that \Gamma is a cancelation monoid, R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a \Gamma -graded ring and E is a \Gamma -graded left R -module. With the above notation, there exists a graded (C,D) -subquotient system with the following properties: (1) C/D\cong E in the category of \Gamma -graded left R -modules. (2) A_{i}/B_{i} is \Gamma -graded finitely presented for all i\in\mathcal{I} . (3) If F is \Gamma -graded finitely presented, and if \eta:F\to C/D is a homogeneous homomorphism, then there exists an i\in\mathcal{I} and a homogeneous isomorphism \eta^{\prime}:F\to A_{i}/B_{i} such that the triangle \textstyle{A_{i}/B_{i}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\psi_{i}} \textstyle{C/D} \textstyle{F\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\eta^{\prime}} \scriptstyle{\eta} commutes. (4) If i\in\mathcal{I} and \psi_{i}=\rho\sigma , where \sigma:A_{i}/B_{i}\to F and \rho:F\to C/D are homogeneous homomorphisms, with F a \Gamma -graded finitely presented module, then there exists j\in\mathcal{I} , j\geq i , and a homogeneous isomorphism \tau:F\to A_{j}/B_{j} such that the diagram \textstyle{A_{j}/B_{j}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\psi_{j}} \textstyle{A_{i}/B_{i}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\varphi_{ij}} \scriptstyle{\sigma} \textstyle{C/D} \textstyle{F\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\tau} \scriptstyle{\eta} commutes. Proof. Let C be the gr-free \Gamma -graded left R -module with homogeneous basis \{c_{(e,n)}\mid e\in H(E),n\in\mathbb{N}\} such that \deg(c_{(e,n)})=\deg(e) for each e\in H(E),n\in\mathbb{N} . Let \pi:C\to E be the homogeneous epimorphism defined by \pi(c_{(e,n)})=e . Hence D:=\ker\pi is a graded submodule of C . Let \mathcal{A} be the family of graded submodules of C generated by finite subsets of \{c_{(e,n)}\mid e\in H(E),n\in\mathbb{N}\} (hence finitely generated gr-free), and let \mathcal{B} be the family of finitely generated graded submodules of D . With this setup we have a graded (C,D) -subquotient system and C/D\cong E as \Gamma -graded R -modules, while each A_{i}/B_{i} is graded finitely presented. Hence (1) and (2) are true. Part (3) is a special case of (4), and the proof of (4) follows as in the ungraded case; see the proof of part (d) in [ 17 , Proposition 8.14] using Lemma 3.10 instead of [ 17 , Lemma 8.13] . ∎ Assume that F and E are \Gamma -graded left R -modules. It is easy to see that there exists a natural group homomorphism \Phi_{F,E}:\sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(F,R)\otimes_{R}E\to\sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(F,E), defined by \Phi_{F,E}(f\otimes e)=(\Phi_{F,E})_{f\otimes e} in which (\Phi_{F,E})_{f\otimes e}(x)=f(x)\cdot e , for f\in\sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(F,R) and e\in E , such that \Phi_{F,E}((\sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(F,R)\otimes_{R}E)_{\alpha})\subseteq\operatorname{Hom}_{\alpha}(F,E), for each \alpha\in\Gamma . It is easy to see that \Phi_{-,E}:\sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(-,R)\otimes_{R}E\to\sideset{{}^{*}}{{}_{R}}{\operatorname{Hom}}(-,E) is a natural transformation. Theorem 3.12 . (Graded version of Lazard’s Theorem) Assume that \Gamma is a cancelation monoid, R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a \Gamma -graded ring and E is a \Gamma -graded left R -module. The following are equivalent: (1) E is gr-flat. (2) For all graded finitely presented left R -module F , \Phi_{F,E} is isomorphism. (3) For all graded finitely presented left R -module F , \Phi_{F,E} is epimorphism. (4) E is a direct limit of finitely generated gr-free R -modules. Proof. (4) \Rightarrow (1) follows from Lemma 3.8 . (1) \Rightarrow (2) is obtained using the naturality of \Phi_{-,E} as the ungraded case (see [ 17 , Proposition 4.18] ), and (2) \Rightarrow (3) is trivial. (3) \Rightarrow (4) Let \mathcal{I} , A_{i} , B_{i} , \psi_{i} , \varphi_{ij} be as in Proposition 3.11 , and set \mathcal{J}:=\{j\in\mathcal{I}\mid A_{j}/B_{j}\text{ is gr-free}\}. Assume that i\in\mathcal{I} . Then \psi_{i}\in\operatorname{Hom}_{\varepsilon}(A_{i}/B_{i},C/D) lies in the image of \Phi_{A_{i}/B_{i},C/D} . Hence there are \sigma_{k}\in\operatorname{Hom}_{\alpha_{k}}(A_{i}/B_{i},R) , k=1,\ldots,n , and homogeneous elements \overline{y}_{1},\ldots,\overline{y}_{n}\in C/D , with \deg(\overline{y}_{k})=\beta_{k} , \alpha_{k}\beta_{k}=\varepsilon , such that \Phi_{A_{i}/B_{i},C/D}(\Sigma_{k=1}^{n}\sigma_{k}\otimes\overline{y}_{k})=\psi_{i} . Then \psi_{i}(\overline{x})=\Sigma_{k=1}^{n}\sigma_{k}(\overline{x})\overline{y}_{k} . Define homogeneous homomorphisms \sigma=(\sigma_{1},\ldots,\sigma_{n}):A_{i}/B_{i}\to L , where L:=\bigoplus_{k=1}^{n}Re_{k} is gr-free with \deg(e_{k})=\beta_{k} , and \rho:L\to C/D by \rho(r_{1},\ldots,r_{n})=\Sigma_{k=1}^{n}r_{k}\overline{y}_{k} . Thus \psi_{i}=\rho\sigma , so by Proposition 3.11 , there exist j\in\mathcal{I} , j\geq i , and a homogeneous isomorphism \tau:L\to A_{j}/B_{j} , for which the diagram \textstyle{A_{j}/B_{j}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\psi_{j}} \textstyle{A_{i}/B_{i}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\varphi_{ij}} \scriptstyle{\sigma} \textstyle{C/D} \textstyle{L\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\tau} \scriptstyle{\cong} \scriptstyle{\eta} is commutative. So, A_{j}/B_{j}\cong L is a gr-free \Gamma -graded R -module, hence j\in\mathcal{J} . This means that \mathcal{J} is cofinal in \mathcal{I} . Hence by Lemma 2.1 (3) and Proposition 3.9 , \mathcal{J} is directed, and E\cong C/D\cong\underset{i\in\mathcal{I}}{\varinjlim}A_{i}/B_{i}\cong\underset{j\in\mathcal{J}}{\varinjlim}A_{j}/B_{j}. This completes the proof. ∎ The following corollary generalizes [ 9 , Proposition 3.2] in the case where \Gamma=\mathbb{Z} and [ 15 , Exercise 11, Page 72] in the case where \Gamma is a Group. Corollary 3.13 . Assume that \Gamma is a cancelation monoid, and that R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a \Gamma -graded ring. A \Gamma -graded left R -module F is gr-flat if and only if it is a flat R -module. Proof. It can be seen that f\otimes_{R}1_{F}:M\otimes_{R}F\to N\otimes_{R}F is a homogeneous homomorphism of \Gamma -graded modules for every homogeneous homomorphism f:M\to N of \Gamma -graded right R -modules M,N . This proves that if F is a flat R -module, then it is a gr-flat module. Conversely, suppose F is gr-flat. Then, Theorem 3.12 shows that the gr-flat module F is the direct limit of a direct system of finitely generated gr-free modules (in the category of \Gamma -graded left R -modules) and hence, forgetting the grading, F is flat by [ 17 , Corollary 8.11] . ∎ Remark 3.14 . (1) As in the ungraded case, it can be seen that every graded R -module is a direct limit of all finitely generated graded submodules. Hence, using Lemma 3.8 , a graded module is (gr-)flat if all of its finitely generated graded submodules are gr-flat. (2) It follows from [ 18 , Theorem 3.56] , Proposition 3.1 and Corollary 3.13 , that a finitely presented graded left R -module is gr-flat if and only if it is gr-projective. In particular, if R is graded left Noetherian, then a finitely generated graded left R -module is gr-flat if and only if it is gr-projective. 4. Graded hereditary rings Let \Gamma be a cancelation monoid with neutral element \varepsilon and R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} be a \Gamma -graded ring. Graded-Dedekind domains were introduced in [ 14 ] , in the case where \Gamma=\mathbb{Z} and in [ 3 ] , in the case where \Gamma is a torsionless commutative cancelation monoid. Recall that a graded (commutative) integral domain R is called a graded-Dedekind domain if each nonzero homogeneous ideal of R is invertible. We remind that an ideal I in a commutative integral domain D with quotient field K is called invertible if II^{-1}=D , where I^{-1}=\{x\in K\mid xI\subseteq D\} . Notice also that a nonzero ideal of an integral domain is invertible if and only if it is projective [ 18 , Proposition 4.21] . Hence, R is a graded-Dedekind domain if and only if each nonzero homogeneous ideal of R is (gr-)projective. In this section, we consider this property in a general \Gamma -graded ring. Definition 4.1 . Assume that \Gamma is a cancelation monoid. A \Gamma -graded ring R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is graded left (resp. right) hereditary if every left (resp. right) homogeneous ideal is gr-projective. It is clear from the definition that the graded hereditary rings among graded commutative domains are precisely the graded-Dedekind rings. Example 4.2 . (1) We can give a graduation to Small’s example, which provides us a graded right hereditary, but not a graded left hereditary ring. Indeed, the triangular ring T:=\left(\begin{array}[]{cc}\mathbb{Z}&\mathbb{Q}\\ 0&\mathbb{Q}\end{array}\right) is an \mathbb{N}_{0} -graded ring, where \mathbb{N}_{0} is the additive monoid of nonnegative integers, with T_{0}=\left(\begin{array}[]{cc}\mathbb{Z}&0\\ 0&\mathbb{Q}\end{array}\right) , T_{1}=\left(\begin{array}[]{cc}0&\mathbb{Q}\\ 0&0\end{array}\right) and T_{i}=0 for i>1 . According to [ 12 , Page 46] , all right ideals of T are projective; it follows that T is graded right hereditary. But the left ideal \left(\begin{array}[]{cc}0&\mathbb{Q}\\ 0&0\end{array}\right) , which is homogeneous, is not projective. This means that T is not a graded left hereditary ring. (2) Let k be a field and R be a free algebra generated over k by \{x_{i}\mid i\in I\} . By [ 12 , Page 45] , R is a (left and right) hereditary ring. Let now \{n_{i}\mid i\in I\} be a family of elements of \mathbb{Z} . Assigning the degree n_{i} to x_{i} , R becomes a \mathbb{Z} -graded ring. Then R is a graded (left and right) hereditary ring. (3) By (2), k\langle x,y\rangle , the free algebra generated over k by x,y , is a \mathbb{Z} -graded ring, where \deg(x)=1 and \deg(y)=-1 . Assume further k is of characteristic zero. Then the first Weyl algebra A_{1}(k):=k\langle x,y\rangle/(xy-yx-1) is a \mathbb{Z} -graded ring. By [ 12 , Page 45] , A_{1}(k) is a (left and right) hereditary ring. Hence A_{1}(k) is a graded (left and right) hereditary ring. (4) Let D be a Dedekind domain and \{X_{\alpha}\} be a nonempty set of indeterminates over D . If we set \Gamma:=\bigoplus_{\alpha}\mathbb{Z}_{\alpha} , where each \mathbb{Z}_{\alpha} is the additive group of integers, then R=D[\{X_{\alpha},X_{\alpha}^{-1}\}] is a \Gamma -graded graded-Dedekind domain. However, if |\{X_{\alpha}\}|\geq 2 , then R is not a Dedekind domain (see [ 3 ] ). Therefore a graded left (resp. right) hereditary ring need not be a left (resp. right) hereditary ring. Let us first prove the following major result on graded submodules of gr-free left modules over graded left hereditary rings. Proposition 4.3 . (Graded version of Kaplansky’s Theorem) Assume that \Gamma is a cancelation monoid, and R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a graded left hereditary ring. Then every graded submodule M of a gr-free left R -module F with the set of homogeneous basis X is isomorphic to \bigoplus_{e\in X}I_{e}e for some homogeneous left ideals I_{e} of R . Proof. Let X=\{e_{i}:i\in J\} be a basis of homogeneous elements of F , and suppose the index set J is well ordered. Set F_{0}=0 , where 0 is the smallest index in J . For each j\in J , define F_{j}:=\bigoplus_{i1 . According to [ 12 , Page 47] , T is left semihereditary; it follows that T is graded left semihereditary. But the right ideal \left(\begin{array}[]{cc}0&R\\ 0&0\end{array}\right) , which is homogeneous, is not projective. This means that T is not a graded right semihereditary ring. (4) Let D be a Prüfer domain which is not a field and set R:=D[X,X^{-1}] , for an indeterminate X over D . Then by [ 2 , Example 3.6] , R is a \mathbb{Z} -graded-Prüfer domain which is not Prüfer. Therefore a graded (left, right) semihereditary ring need not be a (left, right) semihereditary ring. Proposition 5.3 . Assume that \Gamma is a cancelation monoid, and R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a graded left semihereditary ring. Then every finitely generated graded submodule M of a gr-free left R -module F with the set of homogeneous basis X is isomorphic to I_{1}e_{1}+\cdots+I_{n}e_{n} for some finitely generated homogeneous left ideals I_{1},\ldots,I_{n} of R and e_{1},\ldots,e_{n}\in X . Proof. Let X=\{e_{i}:i\in I\} be a basis of homogeneous elements of F , and let M=\langle x_{1},\ldots,x_{m}\rangle be a graded finitely generated submodule of F . Each x_{k} is a finite linear combination of e_{i} ’s. Collect all of these e_{i} ’s as a set Y which is finite and M\subseteq\langle Y\rangle . Hence \langle Y\rangle is a gr-free submodule of F , and so we may assume that F is finitely generated with homogeneous basis \{e_{1},\ldots,e_{n}\} , and \alpha_{i}:=\deg(e_{i}) . We prove, by induction on n\geq 1 , that M is isomorphic to I_{1}e_{1}+\cdots+I_{n}e_{n} for some finitely generated homogeneous left ideals I_{1},\ldots,I_{n} of R . If n=1 , then there exists a finitely generated homogeneous left ideal I of R such that M=Ie_{1} . Assume that n>1 and that the assertion holds for n-1 . Now by the notation of the proof of Theorem 4.3 , M=(M\cap F_{n})\oplus N_{n} , where F_{n}=\oplus_{i=1}^{n-1}Re_{i} and N_{n}\cong I_{n}e_{n} for some homogeneous left ideal I_{n} of R . Being a direct summand of M , M\cap F_{n} is a finitely generated graded submodule of F_{n} . Hence, the induction hypothesis completes the proof. ∎ Theorem 5.4 . Assume that \Gamma is a cancelation monoid. Then a graded ring R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is graded left semihereditary if and only if every finitely generated graded submodule M of a gr-projective left R -module P is gr-projective. Proof. Assume that R is graded left semihereditary and that M a finitely generated graded submodule of a gr-projective left R -module P . Since P is gr-projective, it is a graded submodule, even a summand, of a gr-free left R -module F . Hence M is a finitely generated graded submodule of F . By Proposition 5.3 , M is isomorphic to I_{1}e_{1}+\cdots+I_{n}e_{n} for some finitely generated homogeneous left ideals I_{1},\ldots,I_{n} of R and e_{1},\ldots,e_{n} elements of a homogeneous basis of F . As each of these I_{i} s is gr-projective, using Proposition 3.1 , M is gr-projective too. Conversely, every finitely generated homogeneous left ideal is a finitely generated graded submodule of the gr-free R -module R . Hence, such ideals are gr-projective, and R is graded left semihereditary. ∎ A graded ring R is called a graded left coherent ring if every finitely generated graded left ideal is finitely presented. Using [ 18 , Proposition 3.11] , every left semihereditary ring is left coherent. Proposition 5.5 . Assume that \Gamma is a cancelation monoid and that R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a \Gamma -graded ring. Then R is graded left semihereditary if and only if R is graded left coherent and every homogeneous left ideal is gr-flat. Proof. Assume first that R is graded left semihereditary. Then every finitely generated homogeneous left ideal I is (gr-)projective; hence, I is finitely presented, by [ 18 , Proposition 3.11] . Therefore R is graded left coherent. Since every finitely generated homogeneous left ideal is (gr-)projective, it is (gr-)flat. It follows from Remark 3.14 (1) that every graded left ideal is gr-flat. Conversely, if I is a finitely generated homogeneous left ideal, then I is a graded submodule of the gr-flat module R , and so I is gr-flat; since R is left coherent, I is also finitely presented. Hence, I is projective, by Remark 3.14 (2), and so R is graded left semihereditary. ∎ Recall that the monoid \Gamma is torsionless if from \gamma^{n}=\gamma^{\prime n} , where n\geq 1 is an integer and \gamma,\gamma^{\prime}\in\Gamma , follows \gamma=\gamma^{\prime} . It is well known that the cancellation monoid \Gamma is torsionless if and only if it can be given a total order \leq compatible with the monoid operation [ 16 , page 123] . From now on we assume that \Gamma is a torsionless commutative cancellation monoid. Let R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} be a \Gamma -graded integral domain and M be a \Gamma -graded left R -module. The torsion submodule of M is defined as T(M):=\{x\in M\mid rx=0,\text{ for some }0\neq r\in R\}. Since R is a domain, T(M) is a submodule of M . The R -module M is called torsion-free if T(M)=0 . In the following lemma we show that T(M) is a graded submodule. Lemma 5.6 . Assume that \Gamma is a torsionless commutative cancelation monoid, R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a \Gamma -graded integral domain and M is a \Gamma -graded left R -module. Then T(M) is a graded submodule of M . Proof. Assume that x is a nonzero element of T(M) , and that x=x_{\beta_{1}}+\cdots+x_{\beta_{m}} is the decomposition of x into nonzero homogeneous components with \beta_{1}<\cdots<\beta_{m} . There exists a nonzero element r\in R such that rx=0 . Assume also that r=r_{\alpha_{1}}+\cdots+r_{\alpha_{n}} is the decomposition of r into nonzero homogeneous components with \alpha_{1}<\cdots<\alpha_{n} . It follows from rx=0 that r_{\alpha_{1}}x_{\beta_{1}}=0 . Hence, x_{\beta_{1}}\in T(M) and x-x_{\beta_{1}}=x_{\beta_{2}}+\cdots+x_{\beta_{m}}\in T(M) . By the same way, we see that x_{\beta_{2}},\ldots,x_{\beta_{m}}\in T(M) . This completes the proof. ∎ Let \Gamma be a (nontrivial) commutative cancelation monoid and \langle\Gamma\rangle=\{\alpha/\beta\mid\alpha,\beta\in\Gamma\} be the quotient group of \Gamma . Let R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} be a \Gamma -graded commutative integral domain and H=\bigcup_{\alpha\in\Gamma}(R_{\alpha}\setminus\{0\}) . We observed that the localization R_{H} is a \langle\Gamma\rangle -graded integral domain. Notice that each nonzero homogeneous elements of R_{H} is invertible. Lemma 5.7 . Assume that \Gamma is a torsionless commutative cancelation monoid, R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a graded integral domain, and M is a graded torsion-free R -module. Then M can be embedded in a gr-free R_{H} -module. Moreover, if M is finitely generated, then M can be embedded in a finitely generated gr-free R -module. Proof. Imbed M in a gr-injective module E by Theorem 3.7 . Since M is torsion-free, M is also imbedded in E/T(E) , which is torsion-free and gr-divisible. It is easy to see that E/T(E) is a graded module over R_{H} , hence, is a gr-free R_{H} -module by [ 8 , Lemma 4] . Assume M is finitely generated and choose a homogeneous basis for the gr-free R_{H} -module E/T(E) . Then, each of the homogeneous generators of M is a linear combination of only finitely many elements of the basis. Thus, we may assume M is imbedded in a finitely generated gr-free R_{H} -module V . Let \{v_{1},\ldots,v_{m}\} be a homogeneous basis of V and \{x_{1},\ldots,x_{n}\} be a generating set of homogeneous elements of M . Then each x_{i}=\sum_{j}r_{ij}/s_{ij}v_{j} , where r_{ij}\in R and s_{ij}\in H . If s is the product of all s_{ij} , then \{s^{-1}v_{1},\ldots,s^{-1}v_{m}\} , is independent in V , and the graded R -submodule N of V generated by this set is gr-free. Clearly M is imbedded in N . ∎ In the following result, we give a new characterization of graded-Prüfer domains in terms of modules. Theorem 5.8 . Assume that \Gamma is a torsionless commutative cancelation monoid. A graded integral domain R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a graded-Prüfer domain if and only if every finitely generated torsion-free graded R -module is gr-projective. Proof. Assume that R is a graded-Prüfer domain and M is a finitely generated torsion-free graded R -module. Then by Lemma 5.7 , M can be imbedded in a finitely generated gr-free R -module. It follows from Theorem 5.4 that M is gr-projective. Conversely, every homogeneous ideal of R is torsion-free, so every finitely generated homogeneous ideal of R is gr-projective. ∎ Corollary 5.9 . Assume that \Gamma is a torsionless commutative cancelation monoid, and that R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a graded-Prüfer domain. Then every graded R -module is gr-flat if and only if it is torsion-free. Proof. Suppose that F is gr-flat. Thus F is flat by Corollary 3.13 , and so is torsion-free by [ 18 , Proposition 3.49] . Conversely, assume that F is a torsion-free graded R -module. By Remark 3.14 , it is enough to show that every finitely generated graded submodule of F is gr-flat. Since R is a graded-Prüfer domain, by Theorem 5.8 , every such finitely generated graded submodule is gr-projective, hence is gr-flat. ∎ Let \{M_{j}\}_{j\in J} be a family of \Gamma -graded left R -modules. The direct product {}^{*}\Pi_{j}M_{j} exists in the category of \Gamma -graded left R -modules with (^{*}\Pi_{j}M_{j})_{\alpha}=\Pi_{j}(M_{j})_{\alpha} for all \alpha\in\Gamma . It is clear that {}^{*}\Pi_{j}M_{j}=\Pi_{j}M_{j} whenever \Gamma is a finite group. In the following proposition, we do not assume that \Gamma is torsionless. Proposition 5.10 . Assume that \Gamma is a finite group, R=\bigoplus_{\alpha\in\Gamma}R_{\alpha} is a \Gamma -graded integral domain and that every torsion-free graded R -module is gr-flat. Then R is a graded-Prüfer domain. Proof. Let J be an arbitrary set. Since \Gamma is finite, the direct product R^{J}:=\Pi_{j\in J}R is a \Gamma -graded R -module by the paragraph above. By the hypothesis, R^{J} is gr-flat, hence is flat. It follows from [ 12 , Theorem 4.47] that R is coherent. Let now I be a homogeneous ideal of R . Then I is torsion-free and hence is gr-flat by the hypothesis. Hence R is a graded-Prüfer domain by Proposition 5.5 . ∎