Sheaves, Presheaves, and Sheaf Cohomology

⚡ Key Takeaways

One-line statement: A sheaf is a presheaf of local data that can be uniquely glued from compatible local pieces; sheaf cohomology measures the obstruction to passing from local to global solutions.

Key result in words: The global sections functor $\Gamma(X,-)$ is left exact but generally not exact; its right derived functors $H^i(X,\mathcal{F})$ quantify the failure, and they reconcile the local-to-global principle that underlies modern geometry.

Where it applies: Algebraic geometry, complex/differential geometry, algebraic topology, number theory, mathematical physics (gauge theory, string theory), and any setting requiring local-global comparison.

Who/when: Jean Leray (1945, prison camp), refined by Henri Cartan, Jean-Pierre Serre, and Alexander Grothendieck (1950s–1960s).

💡 The Core Insight

Local data + a consistency condition = a global object, up to a measurable obstruction. Presheaves record local data; the sheaf axioms enforce that this data behaves like functions (uniquely determined and freely gluable). Whenever you try to lift a local solution to a global one — a logarithm, a section of a line bundle, a primitive of a closed form — the failure is organized into a graded abelian group $H^\bullet(X,\mathcal{F})$ . Cohomology is the bookkeeping device for everything that is locally trivial but globally non-trivial.

DefinitionPresheaf

Let $X$ be a topological space and $\mathbf{Open}(X)$ the category whose objects are open sets and whose morphisms are inclusions. A presheaf of abelian groups is a contravariant functor

1

\quad \mathcal{F}:\mathbf{Open}(X)^{\mathrm{op}}\longrightarrow \mathbf{Ab}.

Derivation hint: A contravariant functor reverses arrows; an inclusion $V\hookrightarrow U$ yields the restriction $\rho^U_V:\mathcal{F}(U)\to\mathcal{F}(V)$ . Concretely $\mathcal{F}$ assigns to each open $U$ an abelian group $\mathcal{F}(U)$ (the sections over $U$ , written $\Gamma(U,\mathcal{F})$ ) and to each inclusion $V\subseteq U$ a homomorphism $\rho^U_V$ such that

2

\quad \rho^U_U=\mathrm{id}_{\mathcal{F}(U)},\qquad \rho^V_W\circ\rho^U_V=\rho^U_W \ \ (W\subseteq V\subseteq U).

Derivation hint: These are precisely the functor identity and composition axioms transcribed for inclusions. We write $s|_V:=\rho^U_V(s)$ for $s\in\mathcal{F}(U)$ .

DefinitionSheaf

A presheaf $\mathcal{F}$ is a sheaf if for every open $U$ and every open cover $U=\bigcup_{i\in I}U_i$ :

(Separation / Identity) If $s,t\in\mathcal{F}(U)$ satisfy $s|_{U_i}=t|_{U_i}$ for all $i$ , then $s=t$ .

(Gluing / Locality) Given $s_i\in\mathcal{F}(U_i)$ with $s_i|_{U_i\cap U_j}=s_j|_{U_i\cap U_j}$ for all $i,j$ , there exists $s\in\mathcal{F}(U)$ with $s|_{U_i}=s_i$ . Equivalently, the following sequence is exact (an equalizer) for every open cover:

3

\quad 0\to\mathcal{F}(U)\xrightarrow{\ \mathrm{res}\ }\prod_{i}\mathcal{F}(U_i)\xrightrightarrows[\ \beta\ ]{\ \alpha\ }\prod_{i,j}\mathcal{F}(U_i\cap U_j),

where $\alpha(s_i)_{ij}=s_i|_{U_{ij}}$ and $\beta(s_i)_{ij}=s_j|_{U_{ij}}$ , with $U_{ij}:=U_i\cap U_j$ .

DefinitionStalk and Sheafification

The stalk of $\mathcal{F}$ at $x\in X$ is the filtered colimit over open neighborhoods $U\ni x$ :

4

\quad \mathcal{F}_x:=\varinjlim_{U\ni x}\mathcal{F}(U),

whose elements are germs $[s]_x$ . The sheafification $\mathcal{F}^+$ is the unique sheaf equipped with $\theta:\mathcal{F}\to\mathcal{F}^+$ inducing isomorphisms on all stalks and universal among morphisms from $\mathcal{F}$ to sheaves.

TheoremExistence of Sheaf Cohomology

The category $\mathbf{Sh}(X,\mathbf{Ab})$ of sheaves of abelian groups is an abelian category with enough injectives. The global sections functor $\Gamma(X,-):\mathbf{Sh}(X,\mathbf{Ab})\to\mathbf{Ab}$ is left exact. Its right derived functors

5

\quad H^i(X,\mathcal{F}):=R^i\Gamma(X,\mathcal{F})=H^i\big(\Gamma(X,\mathcal{I}^\bullet)\big),

computed from any injective resolution $0\to\mathcal{F}\to\mathcal{I}^0\to\mathcal{I}^1\to\cdots$ , are well-defined up to canonical isomorphism, with $H^0(X,\mathcal{F})=\Gamma(X,\mathcal{F})$ .

TheoremLong Exact Sequence

A short exact sequence of sheaves $0\to\mathcal{F}'\to\mathcal{F}\to\mathcal{F}''\to 0$ induces a natural long exact sequence

6

\quad 0\to H^0(X,\mathcal{F}')\to H^0(X,\mathcal{F})\to H^0(X,\mathcal{F}'')\xrightarrow{\delta} H^1(X,\mathcal{F}')\to\cdots

7

\quad \cdots\to H^i(X,\mathcal{F}')\to H^i(X,\mathcal{F})\to H^i(X,\mathcal{F}'')\xrightarrow{\delta} H^{i+1}(X,\mathcal{F}')\to\cdots

Derivation hint: A short exact sequence of complexes (via injective resolutions and the horseshoe lemma) yields the snake-lemma connecting maps $\delta$ . The connecting map $\delta$ is precisely the local-to-global obstruction: an element of $H^0(X,\mathcal{F}'')$ is a global section that lifts locally to $\mathcal{F}$ ; $\delta$ measures whether it lifts globally.

Why It Holds / Derivation

mathematician · Jean Leray | 1906–1998

Leray invented sheaves, the spectral sequence, and the notion of cohomology with supports while a prisoner of war (1940–1945) at Oflag XVII-A in Austria. To avoid being put to work on military applications of his fluid-dynamics expertise, he reinvented himself as a topologist. His lectures, smuggled out and published in 1945–1946, introduced faisceaux (sheaves) as a tool to track how local cohomological data varies. Henri Cartan's Paris seminar (1948–1951) reorganized the theory into its modern stalk-and-section form; Serre's FAC (1955) and Grothendieck's Tôhoku paper (1957) recast everything in the language of derived functors and abelian categories.

🧠 Intuition

Think of a presheaf as data attached to regions, and a sheaf as data that behaves like functions: you can chop a function up over a cover and reassemble it uniquely. The classic failure of gluing — and hence the birth of cohomology — is the logarithm. On any small disk in $\mathbb{C}^\times$ you can define $\log z$ ; these local logs agree on overlaps up to a constant $2\pi i k$ , but no global continuous $\log$ exists on the punctured plane. That obstruction is a class in $H^1$ . The hierarchy of cohomology degrees has a memorable reading:

$H^0$ : global solutions that actually exist (e.g., global sections, locally constant functions).
$H^1$ : first obstruction to gluing — torsors, line bundles, extensions, the failure of "locally solvable $\Rightarrow$ globally solvable."
$H^{\ge 2}$ : higher coherence obstructions — gerbes, deformation obstructions, secondary glueing data.

A sheaf is fine or soft when it has partitions of unity (true for smooth functions): then any local data can be averaged into global data, and all higher cohomology vanishes. Cohomology is thus a measure of the rigidity of the sheaf against being smoothed away.

Figure

Local sections over a cover failing to glue: the genesis of

H^1

. Local logarithms (gold) agree on overlaps up to a period, but no global branch exists on the annulus.

ExampleČech cohomology of the circle

S^1

with the constant sheaf

\underline{\mathbb{Z}}

Given: $X=S^1$ , sheaf $\underline{\mathbb{Z}}$ of locally constant integer-valued functions. Cover $S^1$ by two connected arcs $U_0,U_1$ whose intersection $U_0\cap U_1$ is two disjoint arcs $A\sqcup B$ .

Step 1 — Sections. $\underline{\mathbb{Z}}(U_0)=\mathbb{Z}$ , $\underline{\mathbb{Z}}(U_1)=\mathbb{Z}$ (each arc connected), and $\underline{\mathbb{Z}}(U_0\cap U_1)=\mathbb{Z}\oplus\mathbb{Z}$ (two components).

Step 2 — Čech complex.

0\to \mathbb{Z}\oplus\mathbb{Z}\xrightarrow{\ d^0\ }\mathbb{Z}\oplus\mathbb{Z}\to 0,\qquad d^0(a,b)=(a-b,\ a-b).

Step 3 — $H^0$ . $\ker d^0=\{(a,b):a=b\}\cong\mathbb{Z}$ . ✓ (Connected space: one global constant.)

Step 4 — $H^1$ . $\mathrm{coker}\,d^0=(\mathbb{Z}\oplus\mathbb{Z})/\{(t,t)\}\cong\mathbb{Z}$ via $(x,y)\mapsto x-y$ .

Answer:

\boxed{\ H^0(S^1,\underline{\mathbb{Z}})=\mathbb{Z},\quad H^1(S^1,\underline{\mathbb{Z}})=\mathbb{Z},\quad H^{\ge2}=0\ }

matching singular cohomology $H^\bullet(S^1)$ . The generator of $H^1$ is the "winding" cocycle.

ExampleCohomology of

\mathcal{O}(d)

on

\mathbb{P}^1

over a field

k

Given: Line bundles $\mathcal{O}(d)$ on $\mathbb{P}^1_k=U_0\cup U_1$ with affine charts $U_0=\mathrm{Spec}\,k[t]$ , $U_1=\mathrm{Spec}\,k[t^{-1}]$ , $t=x_1/x_0$ .

Step 1 — Sections. Global sections of $\mathcal{O}(d)$ are degree- $d$ homogeneous polynomials in $x_0,x_1$ :

H^0(\mathbb{P}^1,\mathcal{O}(d))=\begin{cases}k^{\,d+1} & d\ge 0\\ 0 & d<0.\end{cases}

Step 2 — Čech on two charts. With $\mathcal{O}(d)(U_0)=k[t]$ , $\mathcal{O}(d)(U_1)=k[t^{-1}]$ , $\mathcal{O}(d)(U_{01})=k[t,t^{-1}]$ ,

H^1=\frac{k[t,t^{-1}]}{\,t^{d}k[t]+k[t^{-1}]\,}\quad(\text{after twisting by }t^{d}).

Step 3 — Count. The quotient is spanned by Laurent monomials $t^{-1},t^{-2},\dots,t^{d+1}$ , giving dimension $-d-1$ when $d\le -2$ .

Answer:

\boxed{\ \dim H^0=\max(d+1,0),\qquad \dim H^1=\max(-d-1,0)\ }

This satisfies Serre duality $H^1(\mathcal{O}(d))\cong H^0(\mathcal{O}(-2-d))^\vee$ (note $K_{\mathbb{P}^1}=\mathcal{O}(-2)$ ), and Euler characteristic $\chi(\mathcal{O}(d))=d+1$ , consistent with Riemann–Roch on genus $0$ .

⚠️ Common mistake

A presheaf cokernel is not a sheaf. If $\phi:\mathcal{F}\to\mathcal{G}$ is a sheaf morphism, the presheaf $U\mapsto\mathrm{coker}(\mathcal{F}(U)\to\mathcal{G}(U))$ generally violates gluing; the sheaf cokernel is its sheafification. Consequently, $\phi$ can be surjective as a sheaf map (surjective on all stalks) while $\Gamma(X,\phi)$ is not surjective — this is exactly what $H^1$ detects.

Sheaf cohomology is not always the topology of $X$ . $H^i(X,\mathcal{F})$ depends on both $X$ and $\mathcal{F}$ . The constant sheaf recovers singular cohomology only for sufficiently nice (e.g. locally contractible) spaces; coherent sheaves on a scheme see algebraic, not topological, structure.

Čech $\ne$ derived functor cohomology in general. They agree on paracompact spaces (Leray's theorem with acyclic covers) and for quasi-coherent sheaves on separated schemes (affine covers), but Čech cohomology can differ on pathological spaces. Always check the cover is acyclic (a Leray cover) before trusting Čech computations.

CounterexampleConstant presheaf is not a sheaf

Let $X$ be a space with two points (or any disconnected open set) and define the constant presheaf $\mathcal{F}(U)=\mathbb{Z}$ for all nonempty $U$ , with identity restriction maps. Take $U=U_1\sqcup U_2$ disconnected. Sections $1\in\mathcal{F}(U_1)$ and $2\in\mathcal{F}(U_2)$ trivially agree on the empty overlap, but there is no element of $\mathcal{F}(U)=\mathbb{Z}$ restricting to both. Gluing fails. The sheafification is the locally constant sheaf $\underline{\mathbb{Z}}$ with $\underline{\mathbb{Z}}(U)=\mathbb{Z}^{\#\{\text{components of }U\}}$ . This single example explains why $H^0$ counts connected components.

connection

Sheaf cohomology is the keystone linking many theories:

<a class="wiki-link" href="/wiki/de-rham-cohomology">De Rham Cohomology</a>: The Poincaré lemma makes the de Rham complex $0\to\underline{\mathbb{R}}\to\Omega^0\to\Omega^1\to\cdots$ a fine (hence acyclic) resolution of the constant sheaf, so $H^i_{\mathrm{dR}}(M)\cong H^i(M,\underline{\mathbb{R}})$ — the de Rham theorem.
Dolbeault Cohomology: For a complex manifold, the $\bar\partial$ -Poincaré lemma gives $H^q(X,\Omega^p)\cong H^{p,q}_{\bar\partial}(X)$ .
Čech Cohomology: A combinatorial model agreeing with derived cohomology on Leray covers; the bridge is the Čech-to-derived spectral sequence.
Singular Cohomology: Equals $H^\bullet(X,\underline{\mathbb{Z}})$ on locally contractible spaces.
Grothendieck Topology: Replaces open covers by abstract covering families, yielding sheaves on a site and étale cohomology.
Spectral Sequences: The Leray spectral sequence $H^p(Y,R^q f_*\mathcal{F})\Rightarrow H^{p+q}(X,\mathcal{F})$ computes cohomology through a continuous map $f:X\to Y$ .
Derived Categories: The natural home of $R\Gamma$ , where sheaf cohomology becomes hyper-derived functor cohomology.

🔬 Where this appears

⏳ History

- 1945–1946 — Jean Leray, recently freed from an Oflag POW camp, publishes the first definition of faisceaux and spectral sequences in the Comptes Rendus and his Collège de France course.

- 1948–1951 — Henri Cartan's Paris seminar (Séminaire Cartan) systematizes sheaves via stalks and sections; the étale-space definition and the gluing axioms take modern form.

- 1953 — Cartan & Eilenberg's Homological Algebra provides the derived-functor language soon applied to sheaves.

- 1955 — Jean-Pierre Serre's FAC (Faisceaux Algébriques Cohérents) introduces coherent sheaves and their cohomology into algebraic geometry, proving finiteness and vanishing theorems.

- 1956 — Hirzebruch uses sheaf cohomology to prove the Riemann–Roch–Hirzebruch theorem.

- 1957 — Alexander Grothendieck's Tôhoku paper recasts sheaf cohomology as derived functors in abelian categories (axiom AB5, enough injectives), removing all reliance on covers.

- 1958–1964 — Grothendieck and the SGA seminars introduce Grothendieck topologies, sites, topoi, and étale cohomology, aiming at the Weil conjectures.

- 1973 — Pierre Deligne completes the proof of the Weil conjectures using

\ell

-adic étale cohomology.

- 1990s–present — Derived/triangulated categories, perverse sheaves (BBD), and persistent sheaf cohomology in applied topology extend the framework.

open-problem

Hodge conjecture (Clay Millennium Problem). On a smooth projective complex variety, are all rational $(p,p)$ -Hodge classes in $H^{2p}(X,\mathbb{Q})$ algebraic (i.e., combinations of cohomology classes of subvarieties)? This is fundamentally a question about which sheaf-cohomological classes arise geometrically. Still open in general.

Standard conjectures (Grothendieck). A web of conjectures about algebraic cycles and motives that would unify the various Weil cohomology theories (de Rham, $\ell$ -adic, crystalline) into a universal motivic cohomology; largely open.

Computational limits. While $H^i$ of coherent sheaves on projective space is algorithmically computable (Gröbner bases, Macaulay2), cohomology of arbitrary sheaves on general spaces is uncomputable; even for nice cases, vanishing theorems beyond Kodaira (e.g. effective bounds, characteristic- $p$ failures of Kodaira vanishing à la Raynaud) remain delicate.

Non-abelian cohomology. $H^1(X,\underline{G})$ for non-abelian $G$ classifies torsors but is only a pointed set; a satisfactory general $H^{\ge 2}$ (higher gerbes, $\infty$ -stacks) is an active area where the "right" definitions are still being negotiated in derived/higher categorical frameworks.

Sheaves, Presheaves, and Sheaf Cohomology

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also