Quasi-Coherent Sheaves

⚡ Key Takeaways

Quasi-coherent sheaves are the geometrically natural sheaves of modules on a scheme — those locally presented by an exact sequence of free modules. On an affine scheme $\operatorname{Spec} A$ they correspond exactly to $A$ -modules via $M \mapsto \widetilde{M}$ . They form the abelian category in which all of cohomological algebraic geometry takes place, and they were systematized by Serre (1955) and Grothendieck (EGA, 1960s).

💡 The Core Insight

The functor $M \mapsto \widetilde{M}$ glues local module data into a sheaf, and quasi-coherence is precisely the condition that a sheaf of modules be of this form locally. The deep payoff is that quasi-coherence is intrinsic — it does not depend on the chosen affine cover — because localization commutes with the formation of $\widetilde{M}$ , i.e. $\widetilde{M}|_{D(f)} = \widetilde{M_f}$ . This single compatibility lets us pass freely between local algebra and global geometry.

DefinitionThe sheaf associated to a module

Let $A$ be a ring, $X = \operatorname{Spec} A$ , and $M$ an $A$ -module. The associated sheaf $\widetilde{M}$ is the $\mathcal{O}_X$ -module whose sections over a basic open $D(f) = \{\mathfrak{p} : f \notin \mathfrak{p}\}$ are the localization $M_f = M \otimes_A A_f$ :

1

\quad \widetilde{M}(D(f)) = M_f, \qquad \widetilde{M}_{\mathfrak{p}} = M_{\mathfrak{p}}.

Derivation hint: define a presheaf on the basis $\{D(f)\}$ by $D(f)\mapsto M_f$ with restrictions the localization maps $M_f \to M_{fg}$ , then sheafify; exactness of localization shows the presheaf is already a sheaf on the basis. Here $A_f = S^{-1}A$ for $S = \{1,f,f^2,\dots\}$ , and $M_{\mathfrak{p}} = M_{A\setminus\mathfrak{p}}$ is the stalk. Symbols: $\mathfrak{p}$ ranges over prime ideals; $\mathcal{O}_X = \widetilde{A}$ is the structure sheaf.

DefinitionQuasi-coherent sheaf

An $\mathcal{O}_X$ -module $\mathcal{F}$ on a scheme $X$ is quasi-coherent if there exists an open affine cover $X = \bigcup_i U_i$ , $U_i = \operatorname{Spec} A_i$ , and $A_i$ -modules $M_i$ with

2

\quad \mathcal{F}|_{U_i} \;\cong\; \widetilde{M_i} \quad \text{as } \mathcal{O}_{U_i}\text{-modules.}

Derivation hint: equivalently, $\mathcal{F}$ admits, locally, a presentation $\mathcal{O}_X^{(J)} \to \mathcal{O}_X^{(I)} \to \mathcal{F} \to 0$ with $I,J$ possibly infinite index sets — take a free resolution of $M_i$ and apply $\widetilde{(-)}$ , which is exact. $\mathcal{F}$ is coherent if in addition $X$ is noetherian and each $M_i$ is finitely generated (equivalently, finitely presented). $\mathcal{F}$ is locally free of rank $n$ if each $M_i$ may be taken to be $A_i^{\oplus n}$ .

TheoremEquivalence of categories on affines

Let $A$ be a ring, $X=\operatorname{Spec} A$ . The functor $M \mapsto \widetilde M$ is an exact, fully faithful embedding inducing an equivalence of abelian categories

3

\quad A\text{-}\mathbf{Mod} \;\xrightarrow{\ \sim\ }\; \mathbf{QCoh}(X), \qquad M \mapsto \widetilde{M},

with quasi-inverse $\mathcal{F} \mapsto \Gamma(X,\mathcal{F})$ . Moreover $\widetilde{(-)}$ commutes with arbitrary direct sums, colimits, tensor products, and on a noetherian scheme with finite limits. Derivation hint: full faithfulness reduces to $\operatorname{Hom}_A(M,N) \cong \operatorname{Hom}_{\mathcal{O}_X}(\widetilde M,\widetilde N)$ via global sections; essential surjectivity is the definition; exactness follows from flatness of localization.

ProofThat quasi-coherence is intrinsic and

\widetilde{(-)}

is a sheaf

We prove the foundational claims in numbered steps.

(1) $\widetilde M$ is a sheaf on the basis of distinguished opens. Cover $D(f) = \bigcup_j D(g_j)$ . We must show

4

\quad 0 \to M_f \to \prod_j M_{g_j} \to \prod_{j,k} M_{g_jg_k}

is exact. Since $D(f)$ is quasi-compact we may take finitely many $g_j$ ; the $D(g_j)$ cover $D(f)$ iff $f \in \sqrt{(g_1,\dots,g_n)}$ in $A_f$ , i.e. $1 \in (g_1,\dots,g_n)A_f$ . Then exactness is a partition-of-unity argument: a compatible family $(m_j/g_j^{N})$ glues because $\sum a_j g_j = 1$ in $A_f$ . Derivation hint: this is the module version of the standard structure-sheaf gluing; write $1 = \sum a_j g_j^N$ and set $m = \sum a_j g_j^{N-1} m_j$ .

(2) Localization commutes with $\widetilde{(-)}$ . For $f \in A$ ,

5

\quad \widetilde{M}\big|_{D(f)} \;\cong\; \widetilde{M_f} \quad \text{as sheaves on } \operatorname{Spec} A_f.

Derivation hint: $D(f) \cong \operatorname{Spec} A_f$ and $(M_f)_g = M_{fg}$ , so the section rings on basic opens agree.

(3) Intrinsic character. Quasi-coherence can be tested on any affine open and need not refer to a chosen cover. If $\mathcal F|_{U} \cong \widetilde M$ for one affine $U=\operatorname{Spec}A$ , then for every $f$ , $\mathcal F|_{D(f)} \cong \widetilde{M_f}$ by (5). A theorem (EGA I, 1.4) then shows: an $\mathcal O_X$ -module is quasi-coherent iff for every affine open $\operatorname{Spec}A$ it restricts to some $\widetilde M$ . The two definitions agree because the gluing data is forced by the stalks. Derivation hint: reduce to the case of two affine opens whose intersection is again affine; compare the modules $M_f$ and $N_g$ over $A_f \cong B_g$ on the overlap.

(4) Closure under kernels and cokernels. Given $\phi:\widetilde M \to \widetilde N$ , by full faithfulness $\phi = \widetilde\psi$ for $\psi:M\to N$ . Then $\ker\phi = \widetilde{\ker\psi}$ and $\operatorname{coker}\phi = \widetilde{\operatorname{coker}\psi}$ because $\widetilde{(-)}$ is exact. Hence $\mathbf{QCoh}(X)$ is an abelian subcategory of $\mathcal O_X$ -modules, closed under kernels, cokernels, extensions, and arbitrary direct sums. $\qquad\blacksquare$

∎

⚠️ The exactness subtlety

The inclusion $\mathbf{QCoh}(X) \hookrightarrow \mathbf{Mod}(\mathcal{O}_X)$ is exact and closed under kernels/cokernels/extensions, but it is generally not closed under arbitrary products or under taking the sheaf-Hom $\mathcal{Hom}$ unless the source is coherent. On a noetherian scheme coherence is stable under $\mathcal{Hom}$ when both arguments are coherent; on non-noetherian schemes even "coherent" must be defined with care (finite type is not enough).

mathematician · Alexander Grothendieck | 1928–2014

Though the sheaf-theoretic reformulation of algebraic geometry began with Jean-Pierre Serre's 1955 paper Faisceaux Algébriques Cohérents (FAC), it was Grothendieck who, in Éléments de Géométrie Algébrique (EGA, with Dieudonné, 1960–67), placed quasi-coherent sheaves at the center of the theory. Grothendieck recognized that allowing non-finitely-generated modules — quasi-coherence rather than mere coherence — was essential to obtain a category with arbitrary colimits, enough injectives, and good base-change behavior. His insistence on relative geometry (working over an arbitrary base scheme $S$ ) made the functoriality of $f_*$ and $f^*$ on quasi-coherent sheaves the technical heart of EGA III and IV. The slogan "geometry = the category of quasi-coherent sheaves" later matured into the Tannakian and derived-categorical reconstructions of schemes.

🧠 Intuition

Think of a quasi-coherent sheaf as a family of modules varying algebraically over the points of $X$ . Over each point $x$ you have a fiber $\mathcal{F}_x \otimes \kappa(x)$ (a vector space over the residue field), and quasi-coherence guarantees these fibers are knit together by genuine algebraic relations — the local presentations by free modules. A locally free sheaf is the case where the fiber dimension is locally constant: a vector bundle. General quasi-coherent sheaves allow the fiber dimension to jump — for instance a skyscraper sheaf concentrated at a point, or the structure sheaf of a closed subscheme pushed forward to $X$ . The condition " $\widetilde{M}|_{D(f)} = \widetilde{M_f}$ " is the statement that zooming in on an open subset corresponds to inverting functions: geometry restricts, algebra localizes.

Figure

Quasi-coherent sheaf as algebraically varying fibers over

\operatorname{Spec} A

, with restriction = localization

ExampleComputing

\widetilde{M}

for a module over

\mathbb{Z}

Given: $A = \mathbb{Z}$ , $X = \operatorname{Spec}\mathbb{Z}$ , and the module $M = \mathbb{Z}/6\mathbb{Z}$ . Describe the quasi-coherent sheaf $\widetilde{M}$ and its stalks.

Steps: 1. The points of $\operatorname{Spec}\mathbb{Z}$ are $(0)$ and the primes $(p)$ . 2. Stalk at $(p)$ : $\widetilde{M}_{(p)} = M_{(p)} = (\mathbb{Z}/6\mathbb{Z})_{(p)}$ .

For $p = 2$ : invert all odd numbers in $\mathbb{Z}/6$ . The class $3$ becomes a unit's multiple, and $(\mathbb{Z}/6)_{(2)} \cong \mathbb{Z}/2$ .
For $p = 3$ : $(\mathbb{Z}/6)_{(3)} \cong \mathbb{Z}/3$ .
For $p \neq 2,3$ : $6$ is invertible, so $M_{(p)} = 0$ .

3. Stalk at the generic point $(0)$ : $M_{(0)} = M \otimes \mathbb{Q} = 0$ (since $M$ is torsion). 4. Hence $\widetilde{M}$ is supported on $\{(2),(3)\}$ — a "double skyscraper."

Answer:

\boxed{\;\widetilde{M}_{(p)} = \begin{cases}\mathbb{Z}/2 & p=2\\ \mathbb{Z}/3 & p=3\\ 0 & \text{else}\end{cases}\qquad \Gamma(X,\widetilde M)=\mathbb{Z}/6\;}

This illustrates a quasi-coherent (here even coherent) sheaf with jumping, non-constant fiber dimension and finite support.

ExampleCohomology of

\mathcal{O}(d)

on

\mathbb{P}^1

via global sections

Given: $X = \mathbb{P}^1_k$ with homogeneous coordinates $x_0,x_1$ , and the invertible (locally free rank 1) quasi-coherent sheaf $\mathcal{O}(d)$ , $d \in \mathbb{Z}$ . Compute $\dim_k H^0$ and $H^1$ .

Steps: 1. $\mathcal{O}(d)$ corresponds to the graded module $S(d)$ where $S = k[x_0,x_1]$ ; its degree- $0$ part of global sections is the space of homogeneous polynomials of degree $d$ . 2. For $d \ge 0$ : $H^0(\mathbb{P}^1,\mathcal O(d)) = S_d$ , with basis $x_0^d, x_0^{d-1}x_1,\dots,x_1^d$ , so $\dim = d+1$ . 3. By Serre duality $H^1(\mathcal O(d)) \cong H^0(\mathcal O(-2-d))^\vee$ (since $\omega_{\mathbb P^1}=\mathcal O(-2)$ ). 4. For $d \ge -1$ : $H^1 = 0$ . For $d \le -2$ : $\dim H^1 = (-d-2)+1 = -d-1$ .

Answer:

\boxed{\;h^0(\mathcal O(d)) = \max(d+1,0),\qquad h^1(\mathcal O(d)) = \max(-d-1,0)\;}

Euler characteristic check: $\chi(\mathcal O(d)) = h^0 - h^1 = d+1$ , matching Riemann–Roch on $\mathbb{P}^1$ .

⚠️ Quasi-coherent ≠ coherent

"Quasi-coherent" does not mean "almost coherent." Coherent sheaves require finite generation (and a noetherian or coherent base); quasi-coherent sheaves explicitly allow infinitely generated modules, infinite direct sums, and presentations $\mathcal{O}^{(J)} \to \mathcal{O}^{(I)} \to \mathcal F \to 0$ with infinite $I,J$ . The class $\mathbf{QCoh}$ is closed under arbitrary direct sums and filtered colimits; the class $\mathbf{Coh}$ is not. A common error is to assume every quasi-coherent sheaf has finite-dimensional cohomology — false in general; finiteness needs properness plus coherence (Serre's finiteness theorem).

CounterexampleA sheaf of modules that is NOT quasi-coherent

Let $X = \operatorname{Spec}\mathbb{Z}_{(p)}$ (local ring at $p$ ), a two-point space $\{(0),(p)\}$ with the generic point $(0)$ open. Define an $\mathcal{O}_X$ -module $\mathcal{F}$ by hand: put $\mathcal{F}((0)) = 0$ and $\mathcal{F}(X) = \mathbb{Q}$ , with the restriction map $\mathbb{Q}\to 0$ . For a quasi-coherent sheaf $\widetilde M$ on this affine, the restriction $\Gamma(X,\widetilde M) = M \to M_{(0)} = \widetilde M((0))$ must be the localization map. But $M = \mathbb{Q}$ localizes to $\mathbb{Q}_{(0)} = \mathbb{Q} \neq 0$ , so no $M$ produces our hand-built $\mathcal{F}$ . Hence $\mathcal{F}$ is an $\mathcal{O}_X$ -module that fails the localization compatibility (5) and is not quasi-coherent. The lesson: quasi-coherence is exactly the obstruction-free compatibility of sections with localization.

connection

Quasi-coherent sheaves sit at the crossroads of several theories. They are the natural input for Sheaf Cohomology, where the derived functors $R^i\Gamma$ of global sections produce the cohomology groups $H^i(X,\mathcal F)$ central to all of algebraic geometry. Their finite-type, locally free instances are precisely Vector Bundles, and the rank-1 case organizes into the Picard Group and the theory of Divisors. The pullback/pushforward adjunction $f^* \dashv f_*$ along a morphism $f:X\to Y$ makes $\mathbf{QCoh}$ a fibered category — the foundation of Descent Theory and the Functor of Points. Passing to the derived category $D_{\mathrm{qc}}(X)$ links them to Derived Categories and Grothendieck Duality. Finally, the equivalence $\mathbf{QCoh}(\operatorname{Spec}A)\simeq A\text{-Mod}$ ties the subject directly to Commutative Algebra and Tensor and Tor.

🔬 Where this appears

⏳ History

- 1850s–1860s. Riemann and Roch study spaces of meromorphic functions with prescribed poles — implicitly

H^0

of line bundles on curves.

- 1930s. Oka and Cartan develop coherent analytic sheaves in several complex variables; "Theorems A and B" foreshadow the cohomological framework.

- 1945–1950. Leray introduces sheaves and sheaf cohomology; Cartan's seminar systematizes coherent analytic sheaves.

- 1955. Serre's Faisceaux Algébriques Cohérents (FAC) transports coherent sheaves into the Zariski topology of varieties, founding modern sheaf-theoretic algebraic geometry.

- 1957. Serre's GAGA theorem equates coherent algebraic and analytic sheaves on projective varieties.

- 1960–1967. Grothendieck and Dieudonné's EGA develops the full theory of quasi-coherent sheaves on arbitrary schemes, relative cohomology, base change, and duality.

- 1966. Hartshorne's Residues and Duality (after Grothendieck's notes) establishes Grothendieck–Serre duality in the derived category.

- 1977. Hartshorne's Algebraic Geometry makes quasi-coherent sheaves and their cohomology the standard graduate-level subject.

- 1990s–present. Derived/∞-categorical refinements (Bondal–Van den Bergh, Neeman, Lurie) reconstruct schemes from

D_{\mathrm{qc}}

and connect to noncommutative and derived algebraic geometry.

open-problem

Several foundational and research questions surround quasi-coherent sheaves. (1) Non-noetherian pathology: on non-noetherian schemes the relation between "coherent," "finite type," and "finitely presented" sheaves fractures, and many finiteness theorems require extra hypotheses (e.g. coherent rings); a clean general theory remains delicate. (2) Compact generation: $D_{\mathrm{qc}}(X)$ is compactly generated for quasi-compact quasi-separated $X$ (Neeman, Bondal–Van den Bergh), but the precise relationship between compact objects and perfect complexes, and the structure of $D_{\mathrm{qc}}$ for general stacks, is an active area. (3) Tannakian reconstruction: to what extent does the symmetric monoidal category $\mathbf{QCoh}(X)$ determine $X$ ? For nice schemes/stacks this holds (Lurie's Tannaka duality), but the boundary cases — non-quasi-compact, non-separated, or genuinely stacky targets — are subtle. (4) Vanishing and ampleness in positive characteristic: Kodaira-type vanishing for cohomology of quasi-coherent sheaves can fail in characteristic $p$ (Raynaud), and characterizing when it holds drives current research in birational geometry and the minimal model program.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also