Hypercohomology of a complex of sheaves

Intuition [Beginner]

Ordinary sheaf cohomology measures the failure of global sections to assemble nicely from local data on a single sheaf. Hypercohomology answers the same question, but for a chain of sheaves linked together by maps. You have a sequence of sheaves $K^0\to K^1\to K^2\to \cdots$ where each map composed with the next vanishes, and you want one invariant that records the combined cohomological content of every sheaf in the chain and the relations between them.

The everyday analogy is a relay race. A single sheaf is one runner; sheaf cohomology asks how that runner does on their own. A complex of sheaves is the whole team passing the baton; hypercohomology asks how the relay performs as a single unit. Different runners might be strong or weak, but what matters is whether the baton makes it through.

The reason this matters is that many of the most important geometric invariants are not the cohomology of a single sheaf but of a chain. De Rham cohomology of a smooth manifold is the hypercohomology of the chain ${\Omega }^0\to {\Omega }^1\to {\Omega }^2\to \cdots$ of smooth-form sheaves. Deligne cohomology, the home of higher line bundles and gerbes, is the hypercohomology of a similar chain truncated at some degree.

Visual [Beginner]

A schematic showing a row of sheaves $K^0,K^1,K^2$ arranged horizontally, with arrows between them labelled $d$ and one column for each sheaf showing its layers of cohomology. A second row above repeats the same picture with the sheaves replaced by injective resolutions, with vertical arrows linking the two. A caption explains that hypercohomology is computed by taking global sections of the total complex of the upper double row.

The picture captures the essential idea. To compute hypercohomology, you replace every sheaf in the chain by a column of well-behaved (injective) sheaves whose cohomology vanishes in positive degrees, then assemble all the columns into one big total complex and take global sections at the end.

Worked example [Beginner]

Compute the hypercohomology of the constant sheaf $\underline{\mathbb{R}}$ on the circle $S^1$, viewed both as a one-term complex and as the de Rham complex ${\Omega }^0\to {\Omega }^1$ on $S^1$, and verify the two answers agree.

Step 1. View $\underline{\mathbb{R}}$ as a complex concentrated in degree zero. The hypercohomology in this case is just ordinary sheaf cohomology with constant coefficients: ${\mathbb{H}}^n(S^1,\underline{\mathbb{R}})=H^n(S^1,\underline{\mathbb{R}})=H^n(S^1,\mathbb{R})$. The circle has $H^0=\mathbb{R}$ (it is path-connected, one constant per component), $H^1=\mathbb{R}$ (one loop generator), $H^n=0$ for $n\ge 2$.

Step 2. View the de Rham complex ${\Omega }^{\bullet }:{\Omega }^0\to {\Omega }^1$ on $S^1$ as a two-term complex of sheaves. The Poincaré lemma says this chain is a resolution of $\underline{\mathbb{R}}$ by sheaves of smooth forms.

Step 3. The sheaves ${\Omega }^0$ and ${\Omega }^1$ of smooth forms are acyclic for global sections: every smooth bump function shows they admit partitions of unity. So the hypercohomology of ${\Omega }^{\bullet }$ is computed by taking global sections term by term, giving $H^{\ast }({\Omega }^0(S^1)\to {\Omega }^1(S^1))$.

Step 4. Compute. The smooth functions $C^{\infty }(S^1)$ form the degree-zero forms; the multiples $f(\theta )d\theta$ form the degree-one forms. The exterior derivative sends $f$ to $f^{\prime }(\theta )d\theta$. The kernel is the constants, so $H^0=\mathbb{R}$. The cokernel is one-dimensional because integrating a one-form around the circle gives a real number, and the integration map has kernel exactly the exact forms. So $H^1=\mathbb{R}$.

Step 5. Both routes give ${\mathbb{H}}^0=\mathbb{R}$ and ${\mathbb{H}}^1=\mathbb{R}$. They agree, as the de Rham theorem demands.

What this tells us: hypercohomology of the de Rham resolution recovers the cohomology of the constant sheaf. Changing how a sheaf is presented as a complex does not change the answer.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $X$ be a topological space and let $\mathrm{Sh}(X)$ denote the category of sheaves of abelian groups on $X$. This category has enough injectives ^{[Grothendieck 1957]}. Write $\Gamma :\mathrm{Sh}(X)\to \mathrm{Ab}$ for the global-sections functor; it is left-exact.

A bounded-below complex of sheaves is a sequence $$ K^\bullet : \cdots \to 0 \to K^{n_0} \xrightarrow{d^{n_0}} K^{n_0 + 1} \xrightarrow{d^{n_0 + 1}} K^{n_0 + 2} \to \cdots $$ of sheaves on $X$ with $K^n=0$ for $n<n_0$ and $d^{n+1}\circ d^n=0$. The cohomology sheaves are ${\mathcal{H}}^q(K^{\bullet }):=\ker (d^q)/\mathrm{im}(d^{q-1})$ in $\mathrm{Sh}(X)$.

Definition (Cartan-Eilenberg resolution). A Cartan-Eilenberg resolution of $K^{\bullet }$ is a double complex $I^{\bullet ,\bullet }$ of injective sheaves, with horizontal differential $d_h$ and vertical differential $d_v$ satisfying $d_h^2=d_v^2=d_h{d}_v+d_v{d}_h=0$, together with augmentations $K^p\to I^{p,0}$ such that

• for each $p$, the augmented column $0\to K^p\to I^{p,0}\to I^{p,1}\to I^{p,2}\to \cdots$ is an injective resolution of $K^p$;
• for each $p$, the induced complex of cocycles $Z_v(I^{p,\bullet })$, coboundaries $B_v(I^{p,\bullet })$, and cohomology ${\mathcal{H}}_v^q(I^{\bullet ,\bullet })$ form injective resolutions of $Z^p(K^{\bullet })$, $B^p(K^{\bullet })$, and ${\mathcal{H}}^p(K^{\bullet })$ respectively.

Such resolutions exist for every bounded-below $K^{\bullet }$ by a standard construction (resolve the short exact sequences $0\to Z^p\to K^p\to B^{p+1}\to 0$ and $0\to B^p\to Z^p\to {\mathcal{H}}^p\to 0$ separately by injectives, then assemble) ^{[Cartan-Eilenberg 1956]}.

Definition (hypercohomology). Choose a Cartan-Eilenberg resolution $K^{\bullet }\to I^{\bullet ,\bullet }$. The total complex $\mathrm{Tot}(I^{\bullet ,\bullet })$ has ${\mathrm{Tot}}^n={⨁}_{p+q=n}{I}^{p,q}$ with differential $d=d_h+(-1{)}^p{d}_v$. The hypercohomology of $K^{\bullet }$ is $$ \mathbb{H}^n(X, K^\bullet) := H^n(\Gamma(X, \mathrm{Tot}(I^{\bullet, \bullet}))). $$

Equivalently, ${\mathbb{H}}^{\ast }(X,-)$ is the right-derived functor of $\Gamma :\mathrm{Sh}(X)\to \mathrm{Ab}$ extended to the bounded-below derived category, and ${\mathbb{H}}^n(X,K^{\bullet })=H^n(R\Gamma (K^{\bullet }))$ ^{[Hartshorne 1966]}.

Counterexamples to common slips

• The cohomology sheaves ${\mathcal{H}}^q(K^{\bullet })$ are not the same as the hypercohomology ${\mathbb{H}}^q(X,K^{\bullet })$. The former are sheaves; the latter are abelian groups. They are linked by the spectral sequence $H^p(X,{\mathcal{H}}^q(K^{\bullet }))\Rightarrow {\mathbb{H}}^{p+q}(X,K^{\bullet })$, not by an isomorphism.
• A complex with all cohomology sheaves vanishing is called acyclic as a complex but need not have vanishing hypercohomology — the spectral sequence shows the converse holds when the complex is quasi-isomorphic to zero, which is stronger.
• Quasi-isomorphic complexes have the same hypercohomology: if $f:K^{\bullet }\to L^{\bullet }$ induces isomorphisms on every cohomology sheaf ${\mathcal{H}}^q$, then ${\mathbb{H}}^{\ast }(X,f):{\mathbb{H}}^{\ast }(X,K^{\bullet })\to {\mathbb{H}}^{\ast }(X,L^{\bullet })$ is an isomorphism. This is the operational reason hypercohomology is a derived-category invariant.
• Sign conventions matter. The differential on the total complex is $d_h+(-1{)}^p{d}_v$ with the sign chosen so that $d^2=0$. Other conventions exist; consistency across a calculation is what matters.

Key theorem with proof [Intermediate+]

Theorem (Independence of resolution). Let $X$ be a topological space and let $K^{\bullet }$ be a bounded-below complex of sheaves on $X$. The hypercohomology ${\mathbb{H}}^n(X,K^{\bullet })$ is independent (up to canonical isomorphism) of the choice of Cartan-Eilenberg resolution.

Proof. Suppose $K^{\bullet }\to I^{\bullet ,\bullet }$ and $K^{\bullet }\to J^{\bullet ,\bullet }$ are two Cartan-Eilenberg resolutions. By the standard injective-resolution lift, the identity $\mathrm{id}:K^{\bullet }\to K^{\bullet }$ extends to a morphism of double complexes $\phi :I^{\bullet ,\bullet }\to J^{\bullet ,\bullet }$ over ${\mathrm{id}}_{K^{\bullet }}$, unique up to chain homotopy. Symmetrically there is $\psi :J^{\bullet ,\bullet }\to I^{\bullet ,\bullet }$. The composites $\psi \circ \phi$ and $\phi \circ \psi$ are each chain-homotopic to the identity over ${\mathrm{id}}_{K^{\bullet }}$.

Passing to total complexes preserves chain homotopy: if $h^{p,q}:I^{p,q}\to J^{p,q-1}$ is a homotopy on the double complex, then $h={\sum }_{p+q=n}{h}^{p,q}$ is a homotopy on the total complex. Applying $\Gamma$ preserves chain homotopy (it is an additive functor). Taking cohomology of $\Gamma (\mathrm{Tot}(I^{\bullet ,\bullet }))$ versus $\Gamma (\mathrm{Tot}(J^{\bullet ,\bullet }))$ then gives isomorphisms induced by $\phi$ and $\psi$, mutually inverse. To get canonicity, observe that any two choices of $\phi$ are themselves chain-homotopic, so the induced isomorphism on cohomology is the same map regardless of $\phi$.

For the comparison of two arbitrary resolutions to be a canonical isomorphism, the standard third-resolution argument applies. Take a third Cartan-Eilenberg resolution $L^{\bullet ,\bullet }$ together with quasi-isomorphisms $L^{\bullet ,\bullet }\to I^{\bullet ,\bullet }$ and $L^{\bullet ,\bullet }\to J^{\bullet ,\bullet }$ (build $L$ component-wise by combining $I$ and $J$). The two composite isomorphisms ${\mathbb{H}}^{\ast }(I)\leftarrow {\mathbb{H}}^{\ast }(L)\to {\mathbb{H}}^{\ast }(J)$ recover the comparison and exhibit it as independent of intermediate choices. $\square$

Theorem (Functoriality). A morphism of bounded-below complexes $f^{\bullet }:K^{\bullet }\to L^{\bullet }$ induces a map $\mathbb{H}^(X, f^\bullet) : \mathbb{H}^(X, K^\bullet) \to \mathbb{H}^(X, L^\bullet)$,andacontinuousmap$g : X \to Y$inducesapullback$g^* : \mathbb{H}^(Y, K^\bullet) \to \mathbb{H}^(X, g^{-1} K^\bullet)$.If$f^\bullet$isaquasi-isomorphism,$\mathbb{H}^(X, f^\bullet)$ is an isomorphism.

Proof sketch. Lift $f^{\bullet }$ to a morphism of Cartan-Eilenberg resolutions, take total complexes, apply $\Gamma$, take cohomology. The quasi-isomorphism statement is the operational form of the fact that the derived category inverts quasi-isomorphisms; it is also visible from the first spectral sequence below. The pullback for $g$ uses that $g^{-1}$ is exact and sends injectives to injectives in the appropriate categorical sense for derived functors ^{[Iversen 1986]}. $\square$

Bridge. The construction of hypercohomology builds toward every modern derived-category invariant in algebraic geometry and differential topology. The foundational reason it works is exactly that injective resolutions exist for sheaves and propagate cleanly to complexes through the Cartan-Eilenberg construction. This is the central insight: a complex of sheaves is no harder to derive than a single sheaf once the resolution is staged as a double complex. The same pattern appears again in 04.03.06 (derived functors and Ext), where two-variable derived functors are computed by resolving in either argument. Putting these together, the hypercohomology of a complex on a smooth manifold identifies with its de Rham cohomology when the complex is the de Rham resolution, and the bridge is the recognition that integration of forms is exactly the global-sections functor applied to a free injective resolution at the smooth-function level. The same pattern generalises to the Hodge filtration through the holomorphic de Rham complex on a complex manifold, and is dual to the homological framework of Tor in spectral-sequence language; this also appears again in 03.13.01 (spectral sequences of a filtered or double complex) where the two-step filtration of $\mathrm{Tot}(I^{\bullet ,\bullet })$ produces the two convergent spectral sequences.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

Mathlib has the categorical infrastructure for right-derived functors and double complexes, but the hypercohomology functor and its two spectral sequences are not yet packaged as named theorems. A schematic Lean statement of the construction looks like:

import Mathlib.Algebra.Homology.DerivedCategory.Basic
import Mathlib.Algebra.Homology.HomologicalBicomplex
import Mathlib.Algebra.Homology.TotalComplex
import Mathlib.CategoryTheory.Abelian.DerivedFunctor

open CategoryTheory CategoryTheory.Limits

variable {X : TopCat} (K : CochainComplex (Sheaf (TopCat.presheaf X) AddCommGrp) ℤ)

/-- Hypercohomology of a bounded-below complex of sheaves: defined as
    the cohomology of the global-sections functor applied to the total
    complex of a Cartan-Eilenberg resolution. Mathlib has the derived
    category framework but not the named functor. -/
def hyperCohomology (n : ℤ) : AddCommGrp :=
  sorry  -- (n-th cohomology of) R Γ K

/-- The first hypercohomology spectral sequence: E_2^{p,q} = H^p(X, H^q K)
    converges to H^{p+q}(X, K). Filter Tot(I) by the q-direction. -/
theorem hyperCohomology_spectralSequence_I
    (n : ℤ) : SpectralSequence (fun p q => sheafCohomology (K.cohomologySheaf q) p)
      (hyperCohomology K) (n) :=
  sorry  -- filtration argument on the total complex

/-- The second hypercohomology spectral sequence: E_2^{p,q} = H^q(X, K^p)
    converges to H^{p+q}(X, K). Filter Tot(I) by the p-direction. -/
theorem hyperCohomology_spectralSequence_II
    (n : ℤ) : SpectralSequence (fun p q => sheafCohomology (K.X p) q)
      (hyperCohomology K) (n) :=
  sorry  -- filtration argument on the total complex

The proof gap is substantive. Mathlib needs: a packaged Cartan-Eilenberg resolution lemma for bounded-below complexes in an abelian category with enough injectives (existence plus the up-to-homotopy universal property), the total-complex construction with the standard sign convention, the spectral sequence of a filtered complex applied in both filtrations, and the convergence statement for bounded-below double complexes. Each piece is formalisable but the consolidation is open.

Advanced results [Master]

Theorem (first hypercohomology spectral sequence). Let $X$ be a topological space and let $K^{\bullet }$ be a bounded-below complex of sheaves on $X$. There is a convergent spectral sequence $$ ,_{I}E_2^{p, q} = H^p(X, \mathcal{H}^q(K^\bullet)) \Rightarrow \mathbb{H}^{p + q}(X, K^\bullet), $$ natural in $K^{\bullet }$, with differentials $d_r:E_r^{p,q}\to E_r^{p+r,q-r+1}$.

The $E_2$ page identifies the cohomology of the cohomology sheaves of the complex with the abutment. This is the spectral sequence used in practice to relate hypercohomology of $K^{\bullet }$ to ordinary sheaf cohomology of its cohomology sheaves; it is the foundational reason hypercohomology depends only on the quasi-isomorphism class of $K^{\bullet }$.

Theorem (second hypercohomology spectral sequence). Same setup. There is a second convergent spectral sequence $$ ,_{II}E_2^{p, q} = H^q(X, K^p) \Rightarrow \mathbb{H}^{p + q}(X, K^\bullet), $$ natural in $K^{\bullet }$, with differentials $d_r:E_r^{p,q}\to E_r^{p+r,q-r+1}$. Equivalently, the $E_1$ page is $E_1^{p,q}=H^q(X,K^p)$ with $d_1$ induced by the differential $d^p:K^p\to K^{p+1}$.

This spectral sequence is the operational tool: when each $K^p$ is $\Gamma$-acyclic (e.g. fine, soft, flasque, or injective), $E_2$ is concentrated in the row $q=0$, collapsing to ${\mathbb{H}}^n(X,K^{\bullet })=H^n(\Gamma (X,K^{\bullet }))$. This is the spectral-sequence proof of the de Rham theorem and of every analogous "compute hypercohomology by global sections of an acyclic resolution" identification.

Theorem (de Rham theorem, hypercohomology form). Let $X$ be a smooth manifold and let ${\Omega }_X^{\bullet }$ be the sheaf complex of smooth differential forms. Then ${\mathbb{H}}^n(X,{\Omega }_X^{\bullet })=H_{dR}^n(X)=H^n(X,\mathbb{R})$.

The Poincaré lemma identifies the cohomology sheaves as ${\mathcal{H}}^0({\Omega }_X^{\bullet })=\underline{\mathbb{R}}$ and ${\mathcal{H}}^q({\Omega }_X^{\bullet })=0$ for $q\ge 1$. The first spectral sequence collapses, giving ${\mathbb{H}}^n(X,{\Omega }_X^{\bullet })=H^n(X,\underline{\mathbb{R}})$. The smooth-form sheaves ${\Omega }_X^p$ are fine (admit partitions of unity), hence $\Gamma$-acyclic; the second spectral sequence collapses, giving ${\mathbb{H}}^n(X,{\Omega }_X^{\bullet })=H^n(\Gamma (X,{\Omega }_X^{\bullet }))=H_{dR}^n(X)$. The two collapses produce the de Rham theorem.

Theorem (Dolbeault and the Hodge-to-de-Rham spectral sequence). Let $X$ be a complex manifold with holomorphic de Rham complex ${\Omega }_X^{\bullet }$ (holomorphic forms with the holomorphic exterior derivative). There is a convergent spectral sequence $$ E_1^{p, q} = H^q(X, \Omega^p_X) \Rightarrow \mathbb{H}^{p + q}(X, \Omega^\bullet_X) = H^{p + q}(X, \mathbb{C}). $$ On a compact Kähler manifold the spectral sequence degenerates at $E_1$, producing the Hodge decomposition $H^n(X,\mathbb{C})={⨁}_{p+q=n}{H}^q(X,{\Omega }_X^p)$.

The $E_1$ page identification with Dolbeault cohomology $H_{\bar{\partial }}^{p,q}(X)=H^q(X,{\Omega }_X^p)$ is the Dolbeault theorem applied fibrewise. The $E_1$ degeneration on a Kähler manifold is the Hodge identity ${\Delta }_d=2{\Delta }_{\bar{\partial }}=2{\Delta }_{\partial }$ in Kähler geometry. On a non-Kähler complex manifold the spectral sequence need not degenerate; Hopf surfaces are the standard counterexample. Originator: Deligne ^{[Deligne 1971]} gave the hypercohomology framing; the underlying analysis is Hodge 1941 and Kodaira 1949 for the Kähler case.

Theorem (Deligne complex and Deligne cohomology). Let $X$ be a complex manifold and let $p\ge 0$. The Deligne complex is $$ \mathbb{Z}(p)\mathcal{D} : \underbrace{\mathbb{Z}}{\text{deg } 0} \xrightarrow{(2\pi i)^p} \underbrace{\mathcal{O}X}{\text{deg } 1} \xrightarrow{d} \underbrace{\Omega^1_X}{\text{deg } 2} \xrightarrow{d} \cdots \xrightarrow{d} \underbrace{\Omega^{p - 1}X}{\text{deg } p}. $$ The Deligne cohomology of $X$ with weight $p$ is $$ H^k\mathcal{D}(X, \mathbb{Z}(p)) := \mathbb{H}^k(X, \mathbb{Z}(p)\mathcal{D}). $$ *For $p=1$, $H^2\mathcal{D}(X, \mathbb{Z}(1)) = H^1(X, \mathcal{O}X^\times)$viatheexponentialsheafsequence—theholomorphicPicardgroup,fittinglinebundleswithconnection.For$p = 2$,$H^3\mathcal{D}(X, \mathbb{Z}(2))$classifiesgerbeswithconnectivestructureandcurvingon$X$—theentry-pointofBrylinsk{i}^{\prime }sgeometric-quantisationprogramme[ref:TOD{O}_{R}EFBrylinski1993].ThesmoothDelignecomplex(replacingholomorphic$\Omega^p_X$bysmoothcomplex-valued$\mathcal{A}^p_X$) is the variant relevant for differential geometry; the full smooth-Deligne development is the subject of 03.04.15 pending.*

Theorem (compact support and Verdier duality). Let $X$ be a locally compact Hausdorff space of finite cohomological dimension and let $K^{\bullet }$ be a bounded complex of sheaves on $X$. There are two distinct hypercohomology theories: $\mathbb{H}^(X, K^\bullet)$via$\Gamma$and$\mathbb{H}^_c(X, K^\bullet)$via$\Gamma_c$(sectionswithcompactsupport).Verdierdualityidentifies$\mathbb{H}^_c(X, K^\bullet)$withthedualof$\mathbb{H}^(X, K^\bullet \otimes^L \omega_X)$viathedualisingcomplex$\omega_X$.

This is the natural compact-support refinement; the dualising complex ${\omega }_X$ extends Poincaré duality to arbitrary locally compact spaces and is the home for hypercohomology phenomena like Borel-Moore homology and the residue of the topological trace formula. Originator: Verdier ^{[Verdier 1967]}; canonical reference Iversen Cohomology of Sheaves.

Synthesis. Hypercohomology is the central insight that turns "cohomology of a single sheaf" into "cohomology of a complex of sheaves" with the same construction. The foundational reason this generalisation is clean is exactly that injective resolutions exist in $\mathrm{Sh}(X)$ and propagate to complexes through the Cartan-Eilenberg double-complex construction; the rest is bookkeeping in the total complex. The two spectral sequences are dual filtrations of the total complex by columns and by rows. The first identifies hypercohomology with sheaf cohomology of the cohomology sheaves and is dual to the second, which identifies hypercohomology with cohomology of the chain of global sections when each entry is acyclic. Putting these together, hypercohomology is the quasi-isomorphism-invariant invariant: any two complexes with the same cohomology sheaves and compatible filtrations have the same hypercohomology.

The de Rham theorem, the Dolbeault theorem, the Hodge decomposition, Deligne cohomology, the Picard group, gerbes, and Verdier duality are all corollaries of this one construction applied to specific complexes. The bridge to the spectral-sequence framework is the recognition that hypercohomology is the simplest non-degenerate two-direction spectral sequence: a single complex of sheaves resolved by injectives in the second direction collapses to ordinary cohomology of one sheaf when the complex is concentrated in a single degree, and otherwise records the obstructions to that collapse in $d_2,d_3,\dots$. This pattern appears again in 03.13.01 (spectral sequences), where every convergent spectral sequence in cohomology factors through some filtered or double complex; hypercohomology identifies the relevant double complex for sheaf-theoretic invariants. This identifies derived-functor invariants with double-complex spectral sequences and is the conceptual bridge from sheaf theory to derived-category geometry ^{[Hartshorne 1966]}.

Full proof set [Master]

Proposition (existence of Cartan-Eilenberg resolution). Let $\mathcal{A}$ be an abelian category with enough injectives and let $K^{\bullet }$ be a bounded-below complex in $\mathcal{A}$. There exists a Cartan-Eilenberg resolution $K^{\bullet }\to I^{\bullet ,\bullet }$.

Proof. Write $Z^p$, $B^p$, ${\mathcal{H}}^p$ for the cocycles, coboundaries, and cohomology of $K^{\bullet }$ in degree $p$. The short exact sequences $$ 0 \to B^p \to Z^p \to \mathcal{H}^p \to 0, \qquad 0 \to Z^p \to K^p \to B^{p+1} \to 0 $$ exist in $\mathcal{A}$. Choose injective resolutions $B^p\to I_B^{p,\bullet }$ and ${\mathcal{H}}^p\to I_H^{p,\bullet }$ for each $p$. By the horseshoe lemma applied to the first sequence, an injective resolution $Z^p\to I_Z^{p,\bullet }$ exists with $I_Z^{p,q}=I_B^{p,q}\oplus I_H^{p,q}$ and vertical differential compatible with the SES. Apply the horseshoe lemma a second time to the second SES to produce $K^p\to I^{p,\bullet }$ with $I^{p,q}=I_Z^{p,q}\oplus I_B^{p+1,q}$.

Define the horizontal differential $d_h^{p,q}:I^{p,q}\to I^{p+1,q}$ by the composite $I_Z^{p,q}\oplus I_B^{p+1,q}\twoheadrightarrow I_B^{p+1,q}\hookrightarrow I_Z^{p+1,q}\hookrightarrow I^{p+1,q}$. This satisfies $d_h^2=0$ by construction. The vertical differential $d_v$ is the differential of the resolutions. By the horseshoe-lemma compatibility, $d_h{d}_v+d_v{d}_h=0$ on the components — choose signs to make the standard double-complex sign convention work. The cocycles, coboundaries, and cohomology in the vertical direction of $I^{\bullet ,\bullet }$ are by construction $I_Z^{p,\bullet }$, $I_B^{p+1,\bullet }$, and $I_H^{p,\bullet }$ — injective resolutions of $Z^p$, $B^p$, ${\mathcal{H}}^p$. This is the Cartan-Eilenberg property. $\square$

Proposition (independence of resolution). The hypercohomology ${\mathbb{H}}^n(X,K^{\bullet })$ does not depend on the chosen Cartan-Eilenberg resolution up to canonical isomorphism.

Proof. Let $I^{\bullet ,\bullet }$ and $J^{\bullet ,\bullet }$ be two Cartan-Eilenberg resolutions of $K^{\bullet }$. Build a third resolution $L^{\bullet ,\bullet }$ together with morphisms $\phi :L^{\bullet ,\bullet }\to I^{\bullet ,\bullet }$ and $\psi :L^{\bullet ,\bullet }\to J^{\bullet ,\bullet }$ over ${\mathrm{id}}_{K^{\bullet }}$ by applying the lifting property of injective resolutions degree-by-degree to the cocycles/cohomology/coboundary subobjects.

Each of $\phi$ and $\psi$ is a quasi-isomorphism of double complexes (it induces isomorphisms on every ${\mathcal{H}}_v^q$ resolution of ${\mathcal{H}}^p(K^{\bullet })$, hence on every page of the second spectral sequence, hence on the abutment). Passing to total complexes, applying $\Gamma$, and taking cohomology, we obtain isomorphisms ${\mathbb{H}}^{\ast }(L)\to {\mathbb{H}}^{\ast }(I)$ and ${\mathbb{H}}^{\ast }(L)\to {\mathbb{H}}^{\ast }(J)$. The composite ${\mathbb{H}}^{\ast }(I)\stackrel{\sim }{\leftarrow }{\mathbb{H}}^{\ast }(L)\stackrel{\sim }{\to }{\mathbb{H}}^{\ast }(J)$ is independent of the choice of $L$ because any two such third resolutions are themselves connected through a fourth, and the diagram chase produces the canonical isomorphism. $\square$

Proposition (first spectral sequence). Let $K^{\bullet }\to I^{\bullet ,\bullet }$ be a Cartan-Eilenberg resolution and let $T^{\bullet }=\mathrm{Tot}(I^{\bullet ,\bullet })$. The filtration $F^p{T}^n={⨁}_{p^{\prime }\ge p}{I}^{p^{\prime },n-p^{\prime }}$ gives a convergent spectral sequence with $E_2$ page $E_2^{p,q}=H^p(X,{\mathcal{H}}^q(K^{\bullet }))$ converging to ${\mathbb{H}}^{p+q}(X,K^{\bullet })$.

Proof. Apply $\Gamma$ to $T^{\bullet }$ to obtain a filtered complex $F^p\Gamma (T^n)\subset \Gamma (T^n)$. The standard spectral sequence of a bounded-below filtered complex has $E_0^{p,q}=F^p\Gamma (T^{p+q})/F^{p+1}\Gamma (T^{p+q})=\Gamma (I^{p,q})$ and $d_0$ the vertical differential. So $E_1^{p,q}=H_v^q(\Gamma (I^{p,\bullet }))$. By the Cartan-Eilenberg property, $I^{p,\bullet }$ is an injective resolution of $K^p$, and similarly $Z_v(I^{p,\bullet })$, $B_v(I^{p,\bullet })$, ${\mathcal{H}}_v^q(I^{p,\bullet })$ are injective resolutions of $Z^p,B^p,{\mathcal{H}}^p$ respectively.

Applying $\Gamma$ and taking cohomology in the $q$-direction first: $H_v^q(\Gamma (I^{p,\bullet }))=H^q(\Gamma (I^{p,\bullet }))=R^q\Gamma (K^p)=H^q(X,K^p)$. Wait, this is the second spectral sequence's $E_1$. Re-set: filter instead by the column index $F^q{T}^n={⨁}_{q^{\prime }\ge q}{I}^{n-q^{\prime },q^{\prime }}$, so $E_0^{p,q}=\Gamma (I^{p,q})$ with $d_0$ the horizontal differential. Then $E_1^{p,q}=H_h^p(\Gamma (I^{\bullet ,q}))$. In the horizontal direction, $I^{\bullet ,q}$ together with the Cartan-Eilenberg structure has cohomology sheaves ${\mathcal{H}}^p(K^{\bullet })$ tensored with the $q$-th degree of an injective resolution; more precisely, ${\mathcal{H}}_h^p(I^{\bullet ,q})$ is the $q$-th term of an injective resolution of ${\mathcal{H}}^p(K^{\bullet })$. So $E_2^{p,q}=H^q(X,{\mathcal{H}}^p(K^{\bullet }))$. Convergence is by bounded-below-ness of the filtration. Re-indexing $p\leftrightarrow q$ (different sources permute differently; we follow the Iversen convention $E_2^{p,q}=H^p(X,{\mathcal{H}}^q)$) gives the asserted statement. $\square$

Proposition (second spectral sequence). With the same setup, the column filtration $F^p{T}^n={⨁}_{p^{\prime }\ge p}{I}^{p^{\prime },n-p^{\prime }}$ gives a convergent spectral sequence with $E_1$ page $E_1^{p,q}=H^q(X,K^p)$ converging to ${\mathbb{H}}^{p+q}(X,K^{\bullet })$, with $d_1$ induced by $d:K^p\to K^{p+1}$.

Proof. Apply $\Gamma$ to the column-filtered total complex. $E_0^{p,q}=\Gamma (I^{p,q})$ with $d_0$ the vertical differential. So $E_1^{p,q}=H_v^q(\Gamma (I^{p,\bullet }))=H^q(X,K^p)$ since $I^{p,\bullet }$ is an injective resolution of $K^p$. The $d_1$ differential is induced by the horizontal differential of $I^{\bullet ,\bullet }$, which on cohomology in the $q$-direction is the differential $d:K^p\to K^{p+1}$ functorially derived to $H^q(X,K^p)\to H^q(X,K^{p+1})$. The $E_2$ page is $E_2^{p,q}=H^p(H^q(X,K^{\bullet }),d_1)=H^p$ of the chain of cohomology groups. The spectral sequence converges by bounded-below filtration. $\square$

Proposition (de Rham via hypercohomology). On a smooth manifold $X$, ${\mathbb{H}}^n(X,{\Omega }_X^{\bullet })=H^n(X,\mathbb{R})=H_{dR}^n(X)$.

Proof. Two routes via the two spectral sequences.

Route 1 (first spectral sequence). By the Poincaré lemma, ${\mathcal{H}}^0({\Omega }_X^{\bullet })=\underline{\mathbb{R}}$ and ${\mathcal{H}}^q({\Omega }_X^{\bullet })=0$ for $q\ge 1$. So $E_2^{p,q}=H^p(X,{\mathcal{H}}^q({\Omega }_X^{\bullet }))=H^p(X,\underline{\mathbb{R}})$ if $q=0$, else zero. The spectral sequence is concentrated in a single row, degenerates at $E_2$, and abuts to ${\mathbb{H}}^n(X,{\Omega }_X^{\bullet })=H^n(X,\underline{\mathbb{R}})=H^n(X,\mathbb{R})$.

Route 2 (second spectral sequence). The smooth-form sheaves ${\Omega }_X^p$ admit partitions of unity, so they are fine (the multiplication-by-a-smooth-bump operation provides the needed acyclicity). Hence $H^q(X,{\Omega }_X^p)=0$ for $q\ge 1$ and $H^0(X,{\Omega }_X^p)={\Omega }_X^p(X)$. The $E_1$ page is concentrated in the row $q=0$ with $E_1^{p,0}={\Omega }_X^p(X)$ and $d_1$ the exterior derivative. So $E_2^{p,0}=H_{dR}^p(X)$, and the spectral sequence degenerates. The abutment is ${\mathbb{H}}^n(X,{\Omega }_X^{\bullet })=H_{dR}^n(X)$.

Both routes converge to the same hypercohomology, so $H^n(X,\mathbb{R})=H_{dR}^n(X)$. $\square$

Proposition (worked Master computation: truncated de Rham on $S^1$). Let $X=S^1$ and let ${\Omega }_X^{\le 1}:{\Omega }^0\to {\Omega }^1$ be the de Rham complex truncated at degree one. Then ${\mathbb{H}}^0(S^1,{\Omega }_X^{\le 1})=\mathbb{R}$ and ${\mathbb{H}}^1(S^1,{\Omega }_X^{\le 1})=\mathbb{R}$.

Proof. On the one-dimensional manifold $S^1$, the full de Rham complex is itself already ${\Omega }_X^{\le 1}$ — there are no forms of degree $\ge 2$, so the truncation ${\Omega }^{\le n}$ equals ${\Omega }^{\bullet }$ for $n\ge 1$. By the de Rham theorem from the previous proposition, ${\mathbb{H}}^k(S^1,{\Omega }_X^{\le 1})={\mathbb{H}}^k(S^1,{\Omega }_X^{\bullet })=H^k(S^1,\mathbb{R})$. Compute the right-hand side: $H^0(S^1,\mathbb{R})=\mathbb{R}$ (one path component); $H^1(S^1,\mathbb{R})=\mathbb{R}$ (one loop generator); $H^k(S^1,\mathbb{R})=0$ for $k\ge 2$.

To verify this through the spectral-sequence calculation directly: the second spectral sequence has $E_1^{0,0}={\Omega }^0(S^1)=C^{\infty }(S^1)$, $E_1^{1,0}={\Omega }^1(S^1)=C^{\infty }(S^1)d\theta$, and $E_1^{p,q}=0$ otherwise (acyclicity of fine sheaves). The $d_1$ differential is the exterior derivative $d:C^{\infty }(S^1)\to C^{\infty }(S^1)d\theta$, $f\mapsto f^{\prime }d\theta$. The kernel is constants, giving $E_2^{0,0}=H_{dR}^0(S^1)=\mathbb{R}$. The cokernel is $C^{\infty }(S^1)d\theta /d{C}^{\infty }(S^1)$. The integration map ${\int }_{S^1}:C^{\infty }(S^1)d\theta \to \mathbb{R}$ has kernel exactly $d{C}^{\infty }(S^1)$ (Stokes on $S^1$: ${\int }_{S^1}dg=0$ for $g$ a smooth function; converse: a $1$-form $\omega$ with ${\int }_{S^1}\omega =0$ has a primitive). So the cokernel is one-dimensional and $E_2^{1,0}=\mathbb{R}$. The spectral sequence degenerates at $E_2$, abutting to ${\mathbb{H}}^0=\mathbb{R}$, ${\mathbb{H}}^1=\mathbb{R}$.

On a higher-dimensional $X$ the truncation ${\Omega }^{\le 1}$ differs from ${\Omega }^{\bullet }$ — the truncated complex computes a refined invariant. The cohomology sheaves are ${\mathcal{H}}^0({\Omega }_X^{\le 1})=\underline{\mathbb{R}}$ and ${\mathcal{H}}^1({\Omega }_X^{\le 1})={\Omega }_X^1/d{\Omega }_X^0$ (the sheaf of locally closed-modulo-exact one-forms, locally ${\Omega }_X^1/d{C}^{\infty }$ — a substantive sheaf in dimensions $\ge 2$). The first spectral sequence captures this: $E_2^{p,q}=H^p(X,{\mathcal{H}}^q({\Omega }_X^{\le 1}))$ with contributions in $q=0$ and $q=1$, with possible $d_2$ differentials between rows. On $S^1$ this collapses because dimension one is too small for $q=1$ to contribute substantively; on higher $X$ it does not, and the truncation records strictly more than full de Rham. The truncation matters above dimension one. $\square$

Proposition (Dolbeault and Hodge filtration). Let $X$ be a complex manifold with holomorphic de Rham complex ${\Omega }_X^{\bullet }$. The second hypercohomology spectral sequence has $E_1^{p,q}=H^q(X,{\Omega }_X^p)$ and abuts to ${\mathbb{H}}^{p+q}(X,{\Omega }_X^{\bullet })=H^{p+q}(X,\mathbb{C})$. On a compact Kähler manifold the spectral sequence degenerates at $E_1$, producing the Hodge decomposition.

Proof. The $E_1$ identification is the second hypercohomology spectral sequence. The abutment is $H^{p+q}(X,\mathbb{C})$ by the holomorphic Poincaré lemma identifying ${\Omega }_X^{\bullet }$ as a resolution of the constant sheaf $\underline{\mathbb{C}}$ on $X$. The $E_1$ degeneration on a compact Kähler manifold is the Hodge identity from Kähler geometry: the harmonic-form decomposition of ${\Delta }_d=2{\Delta }_{\bar{\partial }}=2{\Delta }_{\partial }$ produces a canonical splitting of the filtration on $H^n(X,\mathbb{C})$ by $H^q(X,{\Omega }_X^p)$, forcing every $d_r$ to vanish for $r\ge 1$. The non-Kähler complex case admits genuine $d_1,d_2$ differentials; Hopf surfaces provide the standard counterexample, where $E_1\ne E_{\infty }$. $\square$

Connections [Master]

• Čech-de Rham double complex 03.04.11. The Čech-de Rham double complex is a special case of the Cartan-Eilenberg construction applied to the de Rham complex with the Čech resolution by an open cover playing the role of the injective resolution. The two spectral sequences of the Čech-de Rham double complex are exactly the two hypercohomology spectral sequences, specialised: one shows that Čech-de Rham collapses to de Rham cohomology, the other shows it collapses to Čech cohomology of the constant sheaf. The unit 03.04.11 develops the explicit Čech model; this unit packages the same construction as a general derived-functor framework.

• De Rham cohomology 03.04.06. The de Rham theorem $H_{dR}^n(X)=H^n(X,\mathbb{R})$ is the hypercohomology statement ${\mathbb{H}}^n(X,{\Omega }_X^{\bullet })=H^n(X,\underline{\mathbb{R}})$ together with the acyclicity of smooth-form sheaves. The hypercohomology framing is the modern packaging of de Rham's 1931 isomorphism; it generalises to the holomorphic and Dolbeault settings and produces the Hodge filtration on a Kähler manifold by the same spectral-sequence machinery.

• Singular cohomology and the de Rham theorem 03.04.13. Hypercohomology provides the third proof route for the de Rham theorem (after Mayer-Vietoris induction and the Čech-de Rham double complex). The unit 03.04.13 surveys all three routes; hypercohomology is the route that generalises most cleanly to non-smooth and to algebraic settings, and is the one that opens the door to Deligne cohomology and beyond.

• Spectral sequences of a filtered or double complex 03.13.01. The two hypercohomology spectral sequences are the prototypical example of the spectral sequence of a filtered complex. Every other spectral sequence used in sheaf-theoretic cohomology — Leray, Cartan-Leray, Hodge-to-de-Rham, Frölicher — is a hypercohomology spectral sequence for some specific complex of sheaves on some specific space. The unit 03.13.01 develops the abstract spectral-sequence machinery; this unit applies it to the canonical sheaf-theoretic setting.

• Derived functors and Ext 04.03.06. Hypercohomology is the derived functor $R^n\Gamma :D^{+}(\mathrm{Sh}(X))\to \mathrm{Ab}$ of global sections, extended from a single sheaf to the bounded-below derived category. The unit 04.03.06 develops the abstract derived-functor framework with the Tor and Ext functors as paradigm cases. Hypercohomology is the same construction with $\Gamma$ in place of $\mathrm{Hom}$ or $\otimes$, and with the input upgraded from a single object to a complex.

• Deligne complex (forthcoming 03.04.15 pending). The smooth Deligne complex $\mathbb{Z}(p{)}_{\mathcal{D}}^{\infty }$ is a specific complex of sheaves whose hypercohomology defines the smooth Deligne cohomology groups ${\hat{H}}^k(X,\mathbb{Z}(p))$. These classify line bundles with connection ($p=1$), gerbes with connective structure and curving ($p=2$), and higher gerbes ($p\ge 3$) — the geometric content of Brylinski's loop-space programme. The hypercohomology framework of this unit is the technical input for 03.04.15 pending; the truncation ${\Omega }_X^{\le p-1}$ in the Deligne complex is what distinguishes Deligne cohomology from ordinary cohomology, and the difference is detected by the first hypercohomology spectral sequence on the cohomology sheaves of the truncation.

Historical & philosophical context [Master]

Hypercohomology originated in two convergent threads in the 1950s. The first thread was Cartan and Eilenberg's Homological Algebra (Princeton University Press 1956) ^{[Cartan-Eilenberg 1956]}, which introduced the systematic study of derived functors and double complexes. Chapter XVII of Cartan-Eilenberg defined the "hyperhomology" of a complex with respect to a functor, established its independence of resolution, and proved the two convergent spectral sequences via the standard filtration of the total complex. The terminology hypercohomology came into wide use shortly afterwards as the cohomological dual.

The second thread was Grothendieck's Sur quelques points d'algèbre homologique (Tôhoku Math. J. 9 (1957), 119-221) ^{[Grothendieck 1957]}, which provided the abelian-category foundation: enough injectives, right-derived functors as a universal $\delta$-functor, and the Yoneda-Ext characterisation. Grothendieck's framework allowed hypercohomology to be defined for any left-exact functor between abelian categories and applied not only to sheaves on a topological space but to sheaves in any Grothendieck topology (étale, Nisnevich, fppf, fpqc), which became the foundation of his arithmetic geometry programme. By the late 1960s, Hartshorne's Residues and Duality (Lecture Notes in Math. 20, Springer 1966) ^{[Hartshorne 1966]} had packaged hypercohomology in the language of the bounded-below derived category $D^{+}(\mathcal{A})$, with ${\mathbb{H}}^{\ast }(X,K^{\bullet })=H^{\ast }(R\Gamma (K^{\bullet }))$ as the modern definition.

The connection to differential geometry and complex algebraic geometry was made by Deligne in Théorie de Hodge II (Publ. Math. IHÉS 40 (1971), 5-57) ^{[Deligne 1971]}. Deligne identified the Hodge filtration on the cohomology of a smooth projective variety as the spectral-sequence filtration of the hypercohomology of the truncated holomorphic de Rham complex, and in subsequent work introduced the Deligne complex $\mathbb{Z}(p{)}_{\mathcal{D}}$ whose hypercohomology defines the Deligne cohomology groups. Deligne cohomology is the natural setting for mixed Hodge structures, regulators in algebraic K-theory (Beilinson 1984), and the cycle class map of Bloch-Beilinson conjectures.

Brylinski's Loop Spaces, Characteristic Classes and Geometric Quantization (Birkhäuser Progress in Mathematics 107, 1993) ^{[Brylinski 1993]} developed the smooth Deligne complex as the cohomological home for geometric quantisation, gerbes, and higher line bundles. Brylinski identified $H_{\mathcal{D}}^3(X,\mathbb{Z}(2))$ (smooth Deligne) as the classifying group for gerbes with connective structure and curving on a smooth manifold $X$, generalising the classification $H^2(X,\mathbb{Z})=$ Picard group for ordinary line bundles. Iversen's Cohomology of Sheaves (Springer Universitext 1986) ^{[Iversen 1986]} is the canonical modern reference for the technical apparatus.

Bibliography [Master]

@book{CartanEilenberg1956,
  author    = {Cartan, Henri and Eilenberg, Samuel},
  title     = {Homological Algebra},
  publisher = {Princeton University Press},
  year      = {1956},
  note      = {Ch. XVII introduces hyperhomology via Cartan-Eilenberg resolutions and the two convergent spectral sequences}
}

@article{Grothendieck1957Tohoku,
  author  = {Grothendieck, Alexander},
  title   = {Sur quelques points d'alg\`ebre homologique},
  journal = {T\^ohoku Math. J. (2)},
  volume  = {9},
  year    = {1957},
  pages   = {119--221}
}

@book{Hartshorne1966ResiduesDuality,
  author    = {Hartshorne, Robin},
  title     = {Residues and Duality},
  series    = {Lecture Notes in Mathematics},
  volume    = {20},
  publisher = {Springer-Verlag},
  year      = {1966},
  note      = {Bounded-below derived category framework; $\mathbb{H}^*(X, K^\bullet) = H^* R\Gamma(K^\bullet)$}
}

@article{Deligne1971HodgeII,
  author  = {Deligne, Pierre},
  title   = {Th\'eorie de Hodge II},
  journal = {Publ. Math. Inst. Hautes \'Etudes Sci.},
  volume  = {40},
  year    = {1971},
  pages   = {5--57}
}

@book{Brylinski1993LoopSpaces,
  author    = {Brylinski, Jean-Luc},
  title     = {Loop Spaces, Characteristic Classes and Geometric Quantization},
  series    = {Progress in Mathematics},
  volume    = {107},
  publisher = {Birkh\"auser},
  year      = {1993}
}

@book{Iversen1986CohomologyOfSheaves,
  author    = {Iversen, Birger},
  title     = {Cohomology of Sheaves},
  series    = {Universitext},
  publisher = {Springer-Verlag},
  year      = {1986}
}

@book{Voisin2002HodgeI,
  author    = {Voisin, Claire},
  title     = {Hodge Theory and Complex Algebraic Geometry I},
  series    = {Cambridge Studies in Advanced Mathematics},
  volume    = {76},
  publisher = {Cambridge University Press},
  year      = {2002}
}

@book{GriffithsHarris1978,
  author    = {Griffiths, Phillip and Harris, Joseph},
  title     = {Principles of Algebraic Geometry},
  publisher = {John Wiley \& Sons},
  year      = {1978}
}

@book{Godement1958,
  author    = {Godement, Roger},
  title     = {Topologie alg\'ebrique et th\'eorie des faisceaux},
  publisher = {Hermann},
  address   = {Paris},
  year      = {1958}
}

@incollection{Verdier1967DualityCohomology,
  author    = {Verdier, Jean-Louis},
  title     = {A duality theorem in the \'etale cohomology of schemes},
  booktitle = {Proceedings of a Conference on Local Fields},
  publisher = {Springer-Verlag},
  year      = {1967},
  pages     = {184--198}
}

@article{Beilinson1984HigherRegulators,
  author  = {Be\u\i linson, Alexander A.},
  title   = {Higher regulators and values of $L$-functions},
  journal = {J. Soviet Math.},
  volume  = {30},
  year    = {1985},
  pages   = {2036--2070},
  note    = {Russian original 1984}
}