Čech cohomology of sheaves on schemes

Intuition [Beginner]

Čech cohomology is the bookkeeping that turns an open cover of a space into a number. Cover a scheme by open pieces, record sheaf data on each piece and on each overlap, and the cohomology counts the data that lives consistently on overlaps but does not glue to a global section. The cover is the input; the cohomology is the output.

The name to keep in mind is cocycle. A cocycle is a piece of data on each pairwise overlap, with a matching condition on triple overlaps. A cocycle is a coboundary when it comes from data on the single pieces by subtracting one restriction from another. Čech cohomology in degree one is cocycles modulo coboundaries — the data that lives on overlaps and refuses to come from data on the pieces. In higher degree the same pattern repeats with longer chains of overlaps.

The reason this construction earns its place in algebraic geometry: schemes have a preferred kind of open piece, namely an affine open. On an affine open with a quasi-coherent sheaf, all higher cohomology vanishes (Serre's theorem). So an affine open cover of a scheme records all the cohomological information of the sheaf, and Čech cohomology of an affine cover computes the answer. This is the reason every cohomology calculation in projective algebraic geometry runs through Čech.

Visual [Beginner]

Three open pieces of a scheme, two pairwise overlaps between adjacent pieces, and a triple overlap where all three meet. Sheaf data lives on each piece. The Čech differential records how data on a piece restricts to its overlaps with neighbours.

Worked example [Beginner]

Compute the Čech cohomology of the structure sheaf of an affine line ${\mathbb{A}}^1$ minus the origin, covered by two pieces. Call the two open pieces $U_0=\{x>0\}$ and $U_1=\{x<0\}$ in the model picture (the algebraic version uses two affine opens whose union is the punctured line). The space is $X=U_0\cup U_1$, and the overlap $U_0\cap U_1$ is empty in this model picture, so we cannot use it directly. Switch to a less degenerate cover.

Cover the projective line ${\mathbb{P}}^1$ instead by two pieces $U_0={\mathbb{P}}^1\setminus \{\infty \}$ and $U_1={\mathbb{P}}^1\setminus \{0\}$. Each piece is a copy of the affine line. The overlap $U_0\cap U_1$ is the affine line minus the origin.

Step 1. The Čech zero-cochains record a polynomial on $U_0$ and a polynomial on $U_1$. Each piece supplies a copy of the polynomial ring in one variable.

Step 2. The Čech one-cochains record a Laurent polynomial on the overlap. Polynomials in $x$ and in $1/x$ together with mixed monomials.

Step 3. The Čech differential subtracts the restriction from $U_0$ and from $U_1$ on the overlap. A pair of polynomials $(f_0,f_1)$ maps to the difference $f_1-f_0$ on the overlap.

Step 4. The cocycles are the Laurent polynomials. The coboundaries are the differences of polynomials in $x$ and polynomials in $1/x$, which together fill the entire Laurent polynomial ring.

What this tells us. The first Čech cohomology of the structure sheaf of ${\mathbb{P}}^1$ on this cover is zero. The structure sheaf of the projective line has no nonzero first cohomology — every overlap datum comes from data on the two pieces. The same calculation with a twisting sheaf in place of the structure sheaf gives a nonzero answer in some degrees, as the next chapter records.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $X$ be a scheme 04.02.01, $\mathcal{F}$ a sheaf of abelian groups on $X$ (often taken to be a quasi-coherent ${\mathcal{O}}_X$-module), and $\mathcal{U}=\{U_i{\}}_{i\in I}$ an open cover of $X$ indexed by a totally ordered set $I$. For an ordered tuple $i_0<i_1<\cdots <i_p$ of indices write $U_{i_0\cdots i_p}=U_{i_0}\cap U_{i_1}\cap \cdots \cap U_{i_p}$ for the $(p+1)$-fold intersection.

The Čech cochain group in degree $p$ is

$${\check{C}}^p(\mathcal{U},\mathcal{F}):=\prod _{i_0<i_1<\cdots <i_p}\mathcal{F}(U_{i_0{i}_1\cdots i_p}).$$

A $p$-cochain $\alpha$ assigns a section ${\alpha }_{i_0\cdots i_p}\in \mathcal{F}(U_{i_0\cdots i_p})$ to each ordered tuple. The Čech differential ${\delta }^p:{\check{C}}^p(\mathcal{U},\mathcal{F})\to {\check{C}}^{p+1}(\mathcal{U},\mathcal{F})$ is the alternating-sum-of-restriction map

$$({\delta }^p\alpha {)}_{i_0\cdots i_{p+1}}=\sum _{k=0}^{p+1}(-1{)}^k{\alpha }_{i_0\cdots \widehat{i_k}\cdots i_{p+1}}{|}_{U_{i_0\cdots i_{p+1}}},$$

where the hat denotes omission of the index. Direct computation on faces gives ${\delta }^{p+1}\circ {\delta }^p=0$, so ${\check{C}}^{\bullet }(\mathcal{U},\mathcal{F})$ is a cochain complex of abelian groups.

The Čech cohomology of $\mathcal{F}$ on the cover $\mathcal{U}$ is

$${\check{H}}^p(\mathcal{U},\mathcal{F}):=\frac{\ker ({\delta }^p)}{\mathrm{im}({\delta }^{p-1})}.$$

A refinement of a cover $\mathcal{U}=\{U_i{\}}_{i\in I}$ is a cover $\mathcal{V}=\{V_j{\}}_{j\in J}$ together with a refinement map $\tau :J\to I$ such that $V_j\subseteq U_{\tau (j)}$ for every $j$. A refinement induces a chain map ${\check{C}}^{\bullet }(\mathcal{U},\mathcal{F})\to {\check{C}}^{\bullet }(\mathcal{V},\mathcal{F})$ whose effect on cohomology depends only on $\mathcal{V}$ refining $\mathcal{U}$ (not on the choice of $\tau$). The Čech cohomology of $\mathcal{F}$ on $X$ is the colimit over all open covers, ordered by refinement:

$${\check{H}}^p(X,\mathcal{F}):={\mathrm{colim}}_{\mathcal{U}}{\check{H}}^p(\mathcal{U},\mathcal{F}).$$

In degree zero, ${\check{H}}^0(\mathcal{U},\mathcal{F})=\mathcal{F}(X)$ for every cover, since a zero-cocycle is exactly a compatible family of local sections that glue to a global section by the sheaf axiom.

Comparison with derived-functor cohomology. There is a canonical map ${\check{H}}^p(X,\mathcal{F})\to H^p(X,\mathcal{F})$ from Čech cohomology to the derived-functor sheaf cohomology of 04.03.01. For paracompact Hausdorff $X$ and an arbitrary sheaf $\mathcal{F}$, this map is an isomorphism in every degree. For general topological spaces the comparison fails in degrees $\ge 2$, and one passes to hypercovers to recover the agreement (Verdier).

Cartan's comparison theorem (scheme version). Let $X$ be a separated scheme, $\mathcal{U}=\{U_i\}$ an affine open cover, and $\mathcal{F}$ a quasi-coherent ${\mathcal{O}}_X$-module. Then for every $p\ge 0$ the canonical map

$${\check{H}}^p(\mathcal{U},\mathcal{F})\to H^p(X,\mathcal{F})$$

is an isomorphism. Separatedness is what makes the intersections $U_{i_0\cdots i_p}$ themselves affine (the diagonal is a closed embedding, and intersections of affine opens of a separated scheme are affine), and Serre's affine vanishing theorem then forces the higher derived-presheaf summands of the Čech-derived spectral sequence to vanish, collapsing the spectral sequence onto its bottom row.

Counterexamples to common slips

• The cocycle condition $\delta \alpha =0$ on a one-cochain $\alpha =({\alpha }_{ij})$ on triple overlaps reads ${\alpha }_{jk}-{\alpha }_{ik}+{\alpha }_{ij}=0$, with the orientation convention determined by the sign $(-1{)}^k$ in the alternating sum. Reversing the convention reverses signs.
• Without separatedness of $X$, intersections of affine opens need not be affine, so Cartan's theorem fails on a general scheme. Hartshorne III.4.5 gives a non-separated example where Čech cohomology disagrees with derived-functor cohomology.
• Without quasi-coherence of $\mathcal{F}$, Serre's affine vanishing fails: a generic sheaf of abelian groups on $\mathrm{Spec}A$ can have nonzero higher cohomology, and the comparison theorem leaves the quasi-coherent setting.
• The colimit definition ${\check{H}}^p(X,\mathcal{F})={\mathrm{colim}}_{\mathcal{U}}{\check{H}}^p(\mathcal{U},\mathcal{F})$ is necessary even on schemes: a single affine cover already computes the answer for quasi-coherent sheaves (by Cartan), but the colimit version is what makes Čech a functor on $\mathbf{Sh}$ rather than a cover-dependent gadget.

Key theorem with proof [Intermediate+]

Theorem (Čech computes derived-functor cohomology on affine covers of separated schemes). Let $X$ be a separated scheme, $\mathcal{U}=\{U_i{\}}_{i\in I}$ an affine open cover indexed by a totally ordered set, and $\mathcal{F}$ a quasi-coherent ${\mathcal{O}}_X$-module. Then the canonical map

$${\check{H}}^p(\mathcal{U},\mathcal{F})\stackrel{\sim }{\to }{H}^p(X,\mathcal{F})$$

is an isomorphism for every $p\ge 0$.

Proof. The argument runs through the Čech-derived spectral sequence. For a sheaf $\mathcal{F}$ and a cover $\mathcal{U}$ of any topological space, define the presheaf

$${\mathcal{H}}^q(\mathcal{F}):V\mapsto H^q(V,\mathcal{F})$$

assigning to an open $V$ the $q$-th derived-functor cohomology of $\mathcal{F}$ restricted to $V$. The Čech construction applied to the presheaf ${\mathcal{H}}^q(\mathcal{F})$ produces a Čech cohomology ${\check{H}}^p(\mathcal{U},{\mathcal{H}}^q(\mathcal{F}))$ on the cover. The spectral sequence

$$E_2^{p,q}={\check{H}}^p(\mathcal{U},{\mathcal{H}}^q(\mathcal{F}))⟹H^{p+q}(X,\mathcal{F})$$

converges to the derived-functor cohomology of $\mathcal{F}$. Construction: filter the Cartan-Eilenberg double complex ${\check{C}}^{\bullet }(\mathcal{U},{\mathcal{I}}^{\bullet })$ obtained by applying the Čech construction to an injective resolution $\mathcal{F}\to {\mathcal{I}}^{\bullet }$. The two filtrations of the double complex give two spectral sequences with the same abutment; one of them collapses to $H^{\ast }(X,\mathcal{F})$ on the column $p=0$ (by injectivity of ${\mathcal{I}}^q$), while the other has $E_2$-page exactly ${\check{H}}^p(\mathcal{U},{\mathcal{H}}^q(\mathcal{F}))$.

To prove the theorem it suffices to show that ${\mathcal{H}}^q(\mathcal{F})=0$ on every finite intersection $U_{i_0\cdots i_p}$ for every $q\ge 1$ — for then the $E_2$-page is concentrated on the $q=0$ row, the spectral sequence degenerates at $E_2$, and the abutment $H^{p+q}(X,\mathcal{F})$ in total degree $p$ reads off as ${\check{H}}^p(\mathcal{U},{\mathcal{H}}^0(\mathcal{F}))={\check{H}}^p(\mathcal{U},\mathcal{F})$ (since ${\mathcal{H}}^0(\mathcal{F})=\mathcal{F}$ as a sheaf; here we use that the underlying space of $X$ is separated enough for the presheaf ${\mathcal{H}}^0$ to coincide with $\mathcal{F}$ on opens, which holds without further hypothesis).

The vanishing of ${\mathcal{H}}^q(\mathcal{F})$ on finite intersections has two ingredients.

Intersections are affine. Separatedness of $X$ means the diagonal $\Delta :X\to X\times X$ is a closed embedding. For affine opens $U_i,U_j\subseteq X$, the intersection $U_i\cap U_j$ is the preimage of $\Delta (X)$ under $U_i\times U_j\to X\times X$, and since $\Delta (X)$ is closed and $U_i\times U_j$ is affine (a product of affines is affine), $U_i\cap U_j$ is closed in an affine and hence affine. Iterating: every finite intersection $U_{i_0\cdots i_p}$ is affine.

Higher cohomology of quasi-coherent sheaves on affines vanishes. This is Serre's affine vanishing theorem: for an affine scheme $\mathrm{Spec}A$ and a quasi-coherent sheaf $\mathcal{G}=\tilde{M}$ corresponding to an $A$-module $M$, the derived-functor cohomology $H^q(\mathrm{Spec}A,\tilde{M})=0$ for all $q\ge 1$. The proof of the affine-vanishing theorem itself is by induction via Čech: cover $\mathrm{Spec}A$ by the principal opens $D(f_i)$ for a finite set of generators $f_1,\dots ,f_n$ of the unit ideal; the Čech complex of $\tilde{M}$ on this cover is the Koszul complex of $f_1,\dots ,f_n$ acting on $M$, which is acyclic since the $f_i$ generate the unit ideal. So the Čech cohomology of $\tilde{M}$ on the principal-open cover vanishes in higher degree, and one then promotes this to the derived-functor cohomology by an injective-resolution argument.

Combining the two: every finite intersection $U_{i_0\cdots i_p}$ is affine, and the restriction of $\mathcal{F}$ to it is quasi-coherent, so $H^q(U_{i_0\cdots i_p},\mathcal{F}{|}_{U_{i_0\cdots i_p}})=0$ for $q\ge 1$. The presheaf ${\mathcal{H}}^q(\mathcal{F})$ is thus zero on every finite intersection, the spectral sequence collapses at $E_2$, and the canonical map ${\check{H}}^p(\mathcal{U},\mathcal{F})\to H^p(X,\mathcal{F})$ is an isomorphism. $\square$

The theorem is the engine of every cohomology calculation in projective algebraic geometry: a cover by standard affines $D_{+}(x_i)$ on ${\mathbb{P}}_k^n$ is affine (each piece is ${\mathbb{A}}_k^n$) and finite, so Čech cohomology computes the derived-functor cohomology directly, with no spectral-sequence machinery required at the level of the calculation.

Bridge. The construction here builds toward 04.03.04 (cohomology of projective space — $H^i({\mathbb{P}}^n,\mathcal{O}(d))$), where the same machinery — affine cover by $\{D_{+}(x_i)\}$, quasi-coherent twisting sheaf $\mathcal{O}(d)$, alternating-sum Čech complex — produces the foundational dimension formulas of projective algebraic geometry. The defining pattern appears again in 04.03.05 (Serre vanishing and finiteness) in a sharpened form, where the Čech complex on a finite affine cover is what forces $H^i(X,\mathcal{F})$ to be a finite-dimensional $k$-vector space for $X$ projective and $\mathcal{F}$ coherent. Putting these together, the foundational insight is that the choice of cover is the choice of presentation: a separated scheme with an affine open cover and a quasi-coherent sheaf on it is exactly the data the Čech complex translates into a finite-dimensional linear-algebra problem, and that translation drives every concrete cohomology calculation downstream — Riemann-Roch, Serre duality, cohomology of curves and surfaces. The unit also couples to 03.04.11 (Čech-de Rham double complex), where the same Čech bookkeeping appears on the differential-form side, with the de Rham complex in place of a quasi-coherent sheaf.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

lean_status: none — Mathlib has the structure-sheaf API and abstract derived-functor cohomology in CategoryTheory.Sites.Cohomology, but no scheme-specific Čech cohomology functor. The intended definition would read schematically:

[object Promise]

The proof gap is substantive on multiple fronts. The alternating-sum differential requires the indexed-product API on a totally ordered cover. The ${\delta }^2=0$ identity follows from the standard simplicial-face-relation calculation. The Čech-derived spectral sequence and its degeneration on affine covers of separated schemes is the heaviest piece, requiring the Cartan-Eilenberg double-complex machinery on injective resolutions and the affine-vanishing theorem for quasi-coherent sheaves. Each component is formalisable from existing Mathlib categorical infrastructure but has not yet been packaged as a named theorem in the algebraic-geometry namespace.

Advanced results [Master]

Theorem (Čech-derived spectral sequence). Let $X$ be a topological space, $\mathcal{U}$ an open cover, and $\mathcal{F}$ a sheaf of abelian groups. There is a convergent first-quadrant spectral sequence

$$E_2^{p,q}={\check{H}}^p(\mathcal{U},{\mathcal{H}}^q(\mathcal{F}))⟹H^{p+q}(X,\mathcal{F}),$$

where ${\mathcal{H}}^q(\mathcal{F})$ is the presheaf $V\mapsto H^q(V,\mathcal{F})$. The spectral sequence degenerates at $E_2$ when ${\mathcal{H}}^q(\mathcal{F})=0$ on every finite intersection $U_{i_0\cdots i_p}$ for $q\ge 1$, in which case ${\check{H}}^p(\mathcal{U},\mathcal{F})\cong H^p(X,\mathcal{F})$.

The construction filters the Cartan-Eilenberg double complex ${\check{C}}^p(\mathcal{U},{\mathcal{I}}^q)$ obtained by applying the Čech construction degree-wise to an injective resolution $\mathcal{F}\to {\mathcal{I}}^{\bullet }$. The two filtrations of the double complex give two spectral sequences with the same abutment $H^{\ast }(X,\mathcal{F})$. One filtration collapses on the column $p=0$ (using injectivity of ${\mathcal{I}}^q$, which forces ${\check{H}}^p(\mathcal{U},{\mathcal{I}}^q)=0$ for $p\ge 1$) and identifies the abutment with $H^{\ast }(X,\mathcal{F})$. The other has $E_2$-page exactly ${\check{H}}^p(\mathcal{U},{\mathcal{H}}^q(\mathcal{F}))$.

Cohomology of projective space (Serre's calculation, statement). For projective space ${\mathbb{P}}_k^n$ over a field $k$ and the twisting sheaf $\mathcal{O}(d)$:

$$H^i({\mathbb{P}}_k^n,\mathcal{O}(d))=\{\begin{array}{ll}k\text{-vector space of degree-}d\text{ homogeneous polynomials in }{x}_0,\dots ,x_n& i=0,d\ge 0,\\ k\text{-vector space of degree-}(-d-n-1)\text{ polynomials, Serre dual}& i=n,d\le -n-1,\\ 0& \text{otherwise.}\end{array}$$

The proof runs entirely through Čech cohomology on the standard affine cover $\mathcal{U}=\{D_{+}(x_i){\}}_{i=0}^n$. Each $D_{+}(x_i)$ is affine; intersections $D_{+}(x_{i_0}\cdots x_{i_p})$ are affine; Čech-of-quasi-coherent on this cover computes derived-functor cohomology by Cartan; and the alternating-sum Čech complex of $\mathcal{O}(d)$ on this cover is a direct calculation in graded polynomial rings localised at products of the $x_i$. The dimension count of the surviving cocycles in degree zero gives $\left(\genfrac{}{}{0ex}{}{n+d}{n}\right)$; the dimension count of surviving cocycles in degree $n$ gives $\left(\genfrac{}{}{0ex}{}{-d-1}{-d-n-1}\right)$, which is the Serre-dual formula. The full calculation belongs to 04.03.04.

Affine vanishing theorem. Let $X=\mathrm{Spec}A$ be an affine scheme with $A$ Noetherian, and let $\mathcal{F}=\tilde{M}$ be a quasi-coherent sheaf on $X$. Then $H^p(X,\mathcal{F})=0$ for every $p\ge 1$.

This is Serre's theorem from FAC (1955). The proof for Noetherian $A$ proceeds via Čech cohomology on the principal-open cover $\{D(f_i)\}$ for a finite generating set of the unit ideal, identifying the Čech complex with the Koszul complex of $f_1,\dots ,f_n$ on $M$ and exploiting the contracting homotopy provided by the unit-ideal partition. Without Noetherian hypothesis the statement requires extra care (Grothendieck's quasi-compact-quasi-separated framework); see Liu §5.2 for the general case.

Leray's theorem on acyclic covers (scheme version). Let $X$ be a scheme, $\mathcal{F}$ a sheaf of abelian groups, and $\mathcal{U}$ an open cover such that $H^q(U_{i_0\cdots i_p},\mathcal{F})=0$ for every $q\ge 1$ and every finite intersection. Then ${\check{H}}^p(\mathcal{U},\mathcal{F})\cong H^p(X,\mathcal{F})$ for every $p$. Cartan's theorem is the special case where the cover is affine and the sheaf is quasi-coherent. Leray's theorem is the abstract statement for any acyclic-on-intersections cover.

Refinement and the colimit structure. The Čech functor $\mathcal{U}\mapsto {\check{C}}^{\bullet }(\mathcal{U},\mathcal{F})$ is functorial in covers under refinement, with refinement maps inducing chain maps that are well-defined up to chain homotopy. Passing to cohomology and taking the filtered colimit over all open covers gives ${\check{H}}^p(X,\mathcal{F})$, a functor on the category of sheaves on $X$ (no choice of cover required at the limit). The colimit is a directed system: any two covers admit a common refinement (intersect them), so the colimit is well-defined and computes a single answer per sheaf.

Synthesis. The Čech construction on a scheme is the bridge between two pictures of cohomology: the combinatorial picture of cocycles on overlapping pieces of an open cover, and the homological picture of derived functors on the abelian category of sheaves. The same data — a sheaf and a cover — appears on both sides, with Čech cohomology recording the combinatorial bookkeeping and derived-functor cohomology recording the abstract derived-category bookkeeping. Cartan's comparison theorem identifies the two pictures whenever the scheme is separated, the cover affine, and the sheaf quasi-coherent. The bridge is the Čech-derived spectral sequence: a single double-complex construction whose two filtrations recover the two pictures, with the agreement of their abutments expressing the comparison. Putting these together, the foundational reason cohomology of coherent sheaves on projective varieties is computable is exactly the combinatorial finiteness of the standard affine cover on ${\mathbb{P}}_k^n$: an $(n+1)$-piece cover whose intersections are affine and whose alternating-sum complex is a finite-dimensional linear-algebra problem in graded polynomial rings. This computational tractability is what makes Riemann-Roch, Serre duality, and the entire cohomology theory of curves and surfaces explicit calculations rather than abstract existence statements. The same combinatorial-versus-derived dichotomy appears again in 03.04.11 (Čech-de Rham double complex), where the Čech bookkeeping is paired with the de Rham complex of differential forms in place of a quasi-coherent sheaf, and the dual-proof discipline of computing the same total cohomology via two filtrations recovers de Rham's theorem.

Full proof set [Master]

Theorem (Čech-derived spectral sequence), proof. Let $\mathcal{F}\to {\mathcal{I}}^{\bullet }$ be an injective resolution in ${\mathbf{Sh}}_{\mathbf{Ab}}(X)$. Apply the Čech construction degree-wise on the cover $\mathcal{U}$:

$$K^{p,q}={\check{C}}^p(\mathcal{U},{\mathcal{I}}^q)=\prod _{i_0<\cdots <i_p}{\mathcal{I}}^q(U_{i_0\cdots i_p}).$$

This is a first-quadrant double complex with horizontal differential the Čech $\delta$ and vertical differential the resolution $d$. Form the total complex ${\mathrm{Tot}}^n(K)={⨁}_{p+q=n}{K}^{p,q}$ with total differential $D=\delta +(-1{)}^{p}d$ on bidegree $(p,q)$, satisfying $D^2=0$.

The double complex carries two filtrations (by rows and by columns) and hence two spectral sequences, both converging to the cohomology of the total complex.

First spectral sequence (filter by columns, take horizontal cohomology first). The $E_1$-page is the horizontal Čech cohomology of the rows:

$${}_I{E}_1^{p,q}={\check{H}}^p(\mathcal{U},{\mathcal{I}}^q).$$

Since each ${\mathcal{I}}^q$ is injective, the Čech complex of ${\mathcal{I}}^q$ on any cover is acyclic in positive degree (Hartshorne III.4.3): an injective sheaf has ${\check{H}}^p(\mathcal{U},{\mathcal{I}}^q)=0$ for $p\ge 1$, with ${\check{H}}^0(\mathcal{U},{\mathcal{I}}^q)={\mathcal{I}}^q(X)=\Gamma (X,{\mathcal{I}}^q)$. The $E_1$-page is therefore concentrated in the column $p=0$, with entries $\Gamma (X,{\mathcal{I}}^q)$. The vertical differential on $E_1$ is the resolution differential applied to global sections, so

$${}_I{E}_2^{0,q}=H^q(\Gamma (X,{\mathcal{I}}^{\bullet }))=H^q(X,\mathcal{F}),$$

with zero elsewhere. The spectral sequence degenerates at $E_2$, and the total cohomology in degree $n$ equals $H^n(X,\mathcal{F})$.

Second spectral sequence (filter by rows, take vertical cohomology first). The $E_1$-page is the vertical cohomology of the columns:

$${}_{II}{E}_1^{p,q}=H_d^q(K^{p,\bullet })=\prod _{i_0<\cdots <i_p}{H}^q(U_{i_0\cdots i_p},\mathcal{F}).$$

The key observation: applying ${\mathcal{I}}^{\bullet }$ to the open $V$ gives an injective resolution of $\mathcal{F}{|}_V$ in $\mathbf{Sh}(V)$, so the vertical cohomology of ${\mathcal{I}}^{\bullet }(V)$ computes $H^{\ast }(V,\mathcal{F})$. This is the presheaf ${\mathcal{H}}^q(\mathcal{F})$ of 04.03.01 evaluated on the cover. Hence

$${}_{II}{E}_1^{p,q}={\check{C}}^p(\mathcal{U},{\mathcal{H}}^q(\mathcal{F})),$$

and applying the horizontal Čech differential gives

$${}_{II}{E}_2^{p,q}={\check{H}}^p(\mathcal{U},{\mathcal{H}}^q(\mathcal{F})).$$

This is the Čech-derived $E_2$-page.

Identification of abutments. Both spectral sequences converge to $H^{\ast }(\mathrm{Tot}K)$. The first identifies the abutment with $H^{\ast }(X,\mathcal{F})$. The second identifies the $E_2$-page as ${\check{H}}^p(\mathcal{U},{\mathcal{H}}^q(\mathcal{F}))$. So

$${\check{H}}^p(\mathcal{U},{\mathcal{H}}^q(\mathcal{F}))⟹H^{p+q}(X,\mathcal{F}),$$

which is the Čech-derived spectral sequence in the form claimed.

Degeneration on acyclic covers. When ${\mathcal{H}}^q(\mathcal{F})=0$ on every finite intersection for $q\ge 1$, the $E_2$-page collapses to the row $q=0$, giving ${\check{H}}^p(\mathcal{U},\mathcal{F})\cong H^p(X,\mathcal{F})$. Hartshorne III.4.4 and III.4.5 carry the construction in this form; Liu §5.2 gives an alternative presentation via flasque resolutions. $\square$

Theorem (Cartan's comparison on separated schemes), proof. Apply the spectral sequence of the previous theorem with $X$ separated, $\mathcal{U}$ an affine open cover, and $\mathcal{F}$ quasi-coherent. The presheaf ${\mathcal{H}}^q(\mathcal{F}):V\mapsto H^q(V,\mathcal{F})$ vanishes on every finite intersection $U_{i_0\cdots i_p}$ for $q\ge 1$: separatedness forces the intersections to be affine (intersection of affines in a separated scheme is affine), and Serre's affine vanishing theorem then forces $H^q$ of the quasi-coherent restriction to be zero. The $E_2$-page collapses to the row $q=0$, with $E_2^{p,0}={\check{H}}^p(\mathcal{U},\mathcal{F})$, and the abutment identifies ${\check{H}}^p(\mathcal{U},\mathcal{F})\cong H^p(X,\mathcal{F})$. $\square$

Theorem (affine vanishing), proof for Noetherian $A$. Let $A$ be Noetherian and $X=\mathrm{Spec}A$. Let $\mathcal{F}=\tilde{M}$ for an $A$-module $M$. Choose a finite generating set $f_1,\dots ,f_n$ of the unit ideal in $A$ (possible by quasi-compactness; finite by Noetherianity). The principal opens $D(f_i)$ form an affine cover of $X$. The Čech complex of $\tilde{M}$ on this cover is

$${\check{C}}^p(\mathcal{U},\tilde{M})=\prod _{i_0<\cdots <i_p}{M}_{f_{i_0}\cdots f_{i_p}},$$

with alternating-sum differential. This is the localised Koszul complex of $f_1,\dots ,f_n$ on $M$. Since the $f_i$ generate the unit ideal, there exist $g_i\in A$ with $\sum g_i{f}_i=1$; the multiplication-by-$g_i$ maps assemble into a contracting chain homotopy showing the Koszul complex is acyclic in positive degree. Hence ${\check{H}}^p(\mathcal{U},\tilde{M})=0$ for $p\ge 1$.

Apply Cartan's comparison: $X$ is separated (every affine scheme is separated, since the diagonal $\mathrm{Spec}A\to \mathrm{Spec}A{\otimes }_{\mathbb{Z}}A$ corresponds to the multiplication map $A{\otimes }_{\mathbb{Z}}A\to A$, which is surjective and hence the diagonal is a closed embedding), the cover is affine, the sheaf is quasi-coherent, so ${\check{H}}^p(\mathcal{U},\tilde{M})\cong H^p(X,\tilde{M})$. Combining: $H^p(X,\tilde{M})=0$ for $p\ge 1$. $\square$

Theorem (Leray on acyclic covers, scheme version), proof. Apply the Čech-derived spectral sequence with the acyclicity hypothesis ${\mathcal{H}}^q(\mathcal{F})=0$ on intersections for $q\ge 1$. The $E_2$-page collapses to the row $q=0$, giving ${\check{H}}^p(\mathcal{U},\mathcal{F})=H^p(X,\mathcal{F})$. $\square$

Connections [Master]

• Sheaf cohomology 04.03.01. Čech cohomology on schemes is the concrete computational counterpart to derived-functor sheaf cohomology, with the comparison theorem (Cartan) identifying the two when the scheme is separated, the cover affine, and the sheaf quasi-coherent. The same long exact sequence in cohomology associated to a short exact sequence of sheaves is computed by either picture, with the Čech picture giving an explicit cocycle representative.

• Affine scheme 04.02.02. The vanishing of higher cohomology on an affine scheme for a quasi-coherent sheaf is the load-bearing input to Cartan's theorem — without it, the Čech-derived spectral sequence does not degenerate at $E_2$. Affine-vanishing is the reason the standard affine cover of a separated scheme records all cohomological data.

• Cohomology of projective space 04.03.04. The foundational computation $H^i({\mathbb{P}}_k^n,\mathcal{O}(d))$ runs entirely through Čech on the standard affine cover $\{D_{+}(x_i)\}$. The dimension formulas $\left(\genfrac{}{}{0ex}{}{n+d}{n}\right)$ and $\left(\genfrac{}{}{0ex}{}{-d-1}{-d-n-1}\right)$ are direct counts of surviving cocycles in the alternating-sum complex, and Serre duality is the symmetry between the two formulas.

• Serre vanishing and finiteness 04.03.05. Serre's vanishing $H^i(X,\mathcal{F}\otimes \mathcal{O}(m))=0$ for $m\gg 0$, $i\ge 1$, $\mathcal{F}$ coherent on a projective scheme, runs through Čech on a finite affine cover; the same finite affine cover gives the finite-dimensionality ${\dim }_k{H}^i(X,\mathcal{F})<\infty$ for $\mathcal{F}$ coherent and $X$ projective.

• Čech-de Rham double complex 03.04.11. The differential-form analogue: the Čech construction is paired with the de Rham complex ${\Omega }^{\bullet }$ in place of a quasi-coherent sheaf, and the same dual-proof discipline (two filtrations of the same total complex) recovers de Rham's theorem. Connection type: parallel-construction.

• Picard group 04.05.02. The Picard group $\mathrm{Pic}(X)=H^1(X,{\mathcal{O}}_X^{\times })$ is computed via Čech one-cocycles on a trivialising cover: a line bundle is recorded by its transition functions $g_{ij}\in {\mathcal{O}}_X^{\times }(U_i\cap U_j)$ satisfying the cocycle condition $g_{ij}{g}_{jk}=g_{ik}$ on triple overlaps, modulo coboundaries. The Čech-on-schemes machinery is exactly what makes this description rigorous.

• Riemann-Roch theorem for curves 04.04.01. The dimension of the cohomology groups in Riemann-Roch are computed via Čech on a finite affine cover of the curve. The skyscraper-sheaf short exact sequences used in the inductive proof of Riemann-Roch translate into long exact sequences in Čech cohomology, with each degree computed by a finite-dimensional alternating-sum complex.

• Mayer-Vietoris. The two-set Čech complex on a cover $\mathcal{U}=\{U,V\}$ specialises to the Mayer-Vietoris sequence; the general theory generalises Mayer-Vietoris to arbitrary covers via the Čech-derived spectral sequence.

Historical & philosophical context [Master]

Eduard Čech introduced the construction now bearing his name in Théorie générale de l'homologie dans un espace quelconque (Fundamenta Mathematicae 19 (1932) 149-183) as a homology theory for general topological spaces, defined combinatorially via nerves of open covers. The original 1932 construction was formulated in homology rather than cohomology and used the nerve of the cover as a simplicial complex; the cohomological formulation with the alternating-sum differential on $\prod \mathcal{F}(U_{i_0\cdots i_p})$ is the modern dressing.

The scheme-theoretic version was introduced by Jean-Pierre Serre in Faisceaux Algébriques Cohérents (Annals of Mathematics 61 (1955) 197-278), the founding paper of sheaf-cohomological algebraic geometry. Serre's framing — Čech cohomology of coherent sheaves on a complex algebraic variety, computed on a finite affine cover and matched with derived-functor cohomology — became the template for every subsequent cohomology computation in algebraic geometry. The 1955 paper proved affine vanishing, computed $H^i({\mathbb{P}}^n,\mathcal{O}(d))$, and established Serre duality, all through Čech on standard covers. Hartshorne III, two decades later, codified the framework in the form learned by every contemporary algebraic geometer.

Henri Cartan's comparison theorem (mid-1950s, in the Cartan seminar) placed the Čech-versus-derived-functor agreement on a precise abstract footing: paracompactness on the topological side, separatedness with affine cover and quasi-coherent sheaf on the scheme side. The Čech-derived spectral sequence — the technical mechanism behind the comparison — is the prototype of the more general comparison spectral sequences (Verdier hypercoverings, Brown representability) developed in the 1960s and 1970s.

The combinatorial-versus-derived dichotomy that Čech cohomology embodies recurs throughout twentieth-century mathematics: Galois cohomology computed via cocycles on profinite groups versus derived functors of $G$-invariants; étale cohomology computed via Čech on étale covers versus derived functors on the étale site; sheaf cohomology computed via Čech on a good cover versus derived functors on the small site. In every case the combinatorial picture provides the calculation and the derived picture provides the abstract framework, with a comparison theorem identifying the two on the appropriate hypothesis.

Bibliography [Master]

[object Promise]