Let \(S\) be a graded ring, generated by \(S_1\) as an \(S_0\)-algebra, let \(M\) be a graded \(S\)-module, and let \(X = \Proj S\).
Show that there is a natural homomorphism \(\alpha: M \to \Gamma_*(\widetilde M)\).
Assume now that \(S_0 = A\) is a finitely generated \(k\)-algebra for some field \(k\), that \(S_1\) is a finitely generated \(A\)-module, and that \(M\) is a finitely generated \(S\)-module. Show that \(\alpha\) is an isomorphism in all large enough degrees: there is a \(d_0 \in \ZZ\) such that for all \(d \geq d_0\), the map \(\alpha_d: M_d \to \Gamma(X,\widetilde M(d))\) is an isomorphism.
With the same hypotheses, define an equivalence relation \(\approx\) on graded \(S\)-modules by \(M \approx M'\) if there is an integer \(d\) with \(M_{\geq d} \cong M'_{\geq d}\), where \(M_{\geq d} = \bigoplus_{n \geq d} M_n\). Call a graded \(S\)-module quasi-finitely generated if it is equivalent to a finitely generated module.
Show that the functors \(\sim\) and \(\Gamma_*\) induce an equivalence of categories between the category of quasi-finitely generated graded \(S\)-modules modulo \(\approx\), and the category of coherent \(\OO_X\)-modules.
Hint
For part (b), use the methods of the proof of [@Har10a, Theorem II.5.19].
Solution
Use the associated sheaf and graded section module conventions \[ M(d)_n=M_{n+d},\qquad \Gamma_*(\mcf)=\bigoplus_{d\in\ZZ}\Gamma(X,\mcf(d)). \] All graded-module morphisms in this solution have degree zero.
1. The maps \(\alpha_d\) form a natural graded \(S\)-linear homomorphism, proving part (a).
For \(m\in M_d\), the element \(m/1\) lies in the degree-zero part of \(M(d)_f\) for every homogeneous \(f\in S\) of positive degree. These elements agree after further localization, so they define a global section of \(\widetilde{M(d)}\). Since \(S\) is generated in degree one, the opens \(D_+(f)\) with \(f\in S_1\) cover \(X\). On each of them, powers of \(f\) trivialize \(\OO_X(d)\) and give the canonical isomorphism \[ \widetilde{M(d)}\cong\widetilde M\otimes_{\OO_X}\OO_X(d)=\widetilde M(d) \] [@Har10a, Proposition II.5.12]. Composing with this isomorphism defines \(\alpha_d\). Multiplication of the local fractions by a homogeneous \(s\in S_e\) sends the section defined by \(m\) to that defined by \(sm\). Thus the direct sum \(\alpha_M=\bigoplus_d\alpha_d\) is graded \(S\)-linear. The construction also commutes with every graded homomorphism \(M\to N\), proving naturality.
2. Under the hypotheses of part (b), \(\alpha_d\) is an isomorphism for every sufficiently large \(d\).
The rings \(A\) and \(S\) are noetherian by the Hilbert basis theorem. Choose a finite set of degree-one generators of \(S\) over \(A\), giving a graded surjection \[ P=A[x_0,\ldots,x_r]\longrightarrow S. \] A zero generator may be included if \(S_1=0\). This gives a closed immersion \(i:X\hookrightarrow\PP_A^r\). Regard \(M\) as a finite graded \(P\)-module. The sheaf associated to this \(P\)-module is \(i_*\widetilde M\), as follows by comparing the degree-zero localizations on the standard affine charts. The same identification holds after twisting.
Choose finite graded free \(P\)-modules \(F_0,F_1\) and exact sequences \[ 0\longrightarrow K\longrightarrow F_0\longrightarrow M\longrightarrow0, \qquad 0\longrightarrow L\longrightarrow F_1\longrightarrow K\longrightarrow0. \] The kernels are finite because \(P\) is noetherian. Sheafification on \(\PP_A^r\) is exact, and Serre vanishing gives \[ H^1(\PP_A^r,\widetilde K(d))=H^1(\PP_A^r,\widetilde L(d))=0 \] for all sufficiently large \(d\) [@Har10a, Theorem III.5.2]. The associated long exact sequences then give a right-exact sequence \[ \Gamma(\PP_A^r,\widetilde F_1(d)) \longrightarrow\Gamma(\PP_A^r,\widetilde F_0(d)) \longrightarrow\Gamma(X,\widetilde M(d))\longrightarrow0. \] For \(j=0,1\), the comparison map from \((F_j)_d\) to the corresponding section module is an isomorphism for large \(d\): each \(F_j\) is a finite sum of shifts of \(P\), and the canonical map \[ P_q\longrightarrow\Gamma(\PP_A^r,\OO(q)) \] is an isomorphism for \(q\ge0\) [@Har10a, Theorem III.5.1]. For \(r=0\) this is the identification \(Ax_0^q\cong A\) on \(\Spec A\). Consequently \(M_d\) and \(\Gamma(X,\widetilde M(d))\) are cokernels of the same map \((F_1)_d\to(F_0)_d\). Naturality from step 1 identifies the induced isomorphism with \(\alpha_d\).
3. The category in part (c) is realized by morphisms defined on sufficiently high truncations.
Let \(\mathcal Q\) have quasi-finitely generated graded \(S\)-modules as objects and set \[ \Hom_{\mathcal Q}(M,N) \coloneqq\varinjlim_{d\to+\infty} \Hom_{\mathrm{gr},S}(M_{\ge d},N_{\ge d}). \] The transition maps restrict a degree-zero homomorphism to a higher truncation. Two representatives are equal when their restrictions agree in all sufficiently large degrees. Composition is defined by restricting both representatives to a common truncation and composing there; further restriction preserves this composite. Identities are represented by identity homomorphisms.
A graded isomorphism between truncations gives an isomorphism in \(\mathcal Q\). Conversely, representatives of inverse morphisms in \(\mathcal Q\) become mutually inverse after one sufficiently large common truncation. Thus the isomorphism relation on its objects is exactly \(\approx\). This also proves transitivity of \(\approx\); reflexivity and symmetry follow from identity and inverse graded isomorphisms.
The inclusion \(M_{\ge d}\to M\) induces an isomorphism of associated sheaves. Indeed, on \(D_+(f)\) with \(f\in S_1\), every homogeneous fraction of degree zero has the form \(m/f^n\) with \(m\in M_n\). Multiplying its numerator and denominator by a sufficiently large power of \(f\) makes the numerator lie in \(M_{\ge d}\). Injectivity follows from exactness of localization. Hence sheafification defines a functor \(T:\mathcal Q\to\operatorname{Coh}(X)\). It takes values in coherent sheaves because an object of \(\mathcal Q\) has a truncation isomorphic to a truncation of a finite module, and \(S\) is noetherian.
4. For every coherent sheaf \(\mcf\) on \(X\), the graded module \(\Gamma_*(\mcf)\) is quasi-finitely generated.
By Serre generation, choose a surjection \[ \mathcal E=\bigoplus_{j=1}^t\OO_X(-a_j)\longrightarrow\mcf \] with coherent kernel \(\mathcal K\) [@Har10a, Theorem II.5.17]. By Serre vanishing, \(H^1(X,\mathcal K(d))=0\) for all sufficiently large \(d\). Thus \(\Gamma_*(\mathcal E)_{\ge d_0}\to\Gamma_*(\mcf)_{\ge d_0}\) is surjective for a suitable \(d_0\). Set \(F=\bigoplus_j S(-a_j)\). Step 2 identifies \(\Gamma_*(\mathcal E)_{\ge d_0}\) with \(F_{\ge d_0}\) after increasing \(d_0\). The latter is a finite \(S\)-module because it is a submodule of the finite module \(F\) over the noetherian ring \(S\). Its quotient \(\Gamma_*(\mcf)_{\ge d_0}\) is finite as well. This is precisely quasi-finite generation.
5. The canonical evaluation morphism \(\beta_{\mcf}:\widetilde{\Gamma_*(\mcf)}\to\mcf\) is a natural isomorphism for every coherent \(\mcf\).
5.1. Evaluation defines a natural morphism, and it is an isomorphism for finite sums of twists of \(\OO_X\).
On a standard affine open \(D_+(f)\) with \(f\in S_1\), define the evaluation map by \[ s/f^n\longmapsto s|_{D_+(f)}\otimes f^{-n}, \qquad s\in\Gamma(X,\mcf(n)). \] The factor \(f^{-n}\) trivializes \(\OO_X(-n)\) on this open set. The rule respects equality of fractions, commutes with restriction to intersections, and is linear. It therefore gives \(\beta_{\mcf}\), naturally in \(\mcf\).
3.Hence \(\beta_{\widetilde N}\) is its inverse.
3.2. Evaluation is surjective for every coherent sheaf.
4.Naturality identifies the composite\[ \widetilde{\Gamma_*(\mathcal E)}\longrightarrow \widetilde{\Gamma_*(\mcf)}\xrightarrow{\beta_{\mcf}}\mcf \]with \(\widetilde{\Gamma_*(\mathcal E)}\xrightarrow{\beta_{\mathcal E}}\mathcal E\to\mcf\).The latter is surjective because \(\beta_{\mathcal E}\) is an isomorphism by step
4.1.Thus \(\beta_{\mcf}\) is surjective.
4.3. Evaluation is injective for every coherent sheaf.
4.The vanishing used there makes the sequence of graded section modules exact after sufficiently high truncation.Sheafification is exact and ignores those truncations, giving\[ 0\longrightarrow\widetilde{\Gamma_*(\mathcal K)} \longrightarrow\widetilde{\Gamma_*(\mathcal E)} \longrightarrow\widetilde{\Gamma_*(\mcf)}\longrightarrow0. \]At a stalk, lift an element of \(\ker\beta_{\mcf}\) to \(e\) in the middle term.Its image under \(\beta_{\mathcal E}\) belongs to \(\mathcal K\).By step 2, this element of \(\mathcal K\) lifts under \(\beta_{\mathcal K}\).Subtract its image from \(e\).The difference maps to zero under the isomorphism \(\beta_{\mathcal E}\), so it is zero.Thus the original element of \(\widetilde{\Gamma_*(\mcf)}\) was zero.Evaluation is injective on every stalk, hence injective.
4.4. Q.E.D.
Step 1 constructs the natural morphism; steps 2 and 3 prove that it is an isomorphism.
6. The functors \(T\) and \(G:\mcf\mapsto\Gamma_*(\mcf)\) are inverse equivalences, proving part (c).
Step 4 makes \(G\) a functor from \(\operatorname{Coh}(X)\) to \(\mathcal Q\), and step 5 gives a natural isomorphism \(TG\cong\operatorname{id}\). For \(M\in\mathcal Q\), choose \(e\) with \(M_{\ge e}\) finite. Step 2 applied to \(M_{\ge e}\), together with the naturality of \(\alpha\) for \(M_{\ge e}\to M\), shows that \(\alpha_M\) is an isomorphism in all sufficiently large degrees. Thus \(\alpha\) gives a natural isomorphism \(\operatorname{id}_{\mathcal Q}\cong GT\).
5.Conversely, naturality of \(\alpha\) recovers any morphism between truncations from its sheafification by this formula after increasing the truncation degree.Therefore the correspondence also identifies all morphisms, not only objects.
7. Q.E.D.
Step 1 proves part (a), step 2 proves part (b), and steps <1>3–<1>6 construct and prove the equivalence in part (c).