Six-functor formalism — adjunctions and base change

Intuition Beginner

A morphism $f$ between two geometric spaces comes with a natural family of operations that move sheaves back and forth between the spaces. The simplest two are the inverse image $f^{\ast }$, which pulls a sheaf on the target back to the source by composition, and the direct image $f_{\ast }$, which pushes a sheaf on the source forward to the target by taking sections over preimages. These two operations are linked: $f^{\ast }$ is left adjoint to $f_{\ast }$, meaning that a map of sheaves out of $f^{\ast }G$ into $F$ is the same data as a map of $G$ into $f_{\ast }F$. The link is symmetric and universal — every continuous map of spaces produces this pair.

The six-functor formalism adds four more operations to the picture and organises everything into one coherent package. Two of the new operations are proper direct image $f_{!}$ (the direct image with compact-support condition, which agrees with $f_{\ast }$ when $f$ is proper) and exceptional inverse image $f^{!}$ (a right adjoint to $f_{!}$ that records duality data). The remaining two are the derived tensor product and the derived internal Hom, the standard linear operations on sheaves promoted to derived categories. Together the six operations form a calculus.

The key theorems are the adjunctions $(f^{\ast }⊣f_{\ast })$ and $(f_{!}⊣f^{!})$, the projection formula relating $f_{!}$ to tensor product, and base change: when the morphism $f$ is changed by pulling back along another morphism, the four functors transform in a controlled way that produces an isomorphism of pushforward-then-pullback with pullback-then-pushforward. This package is what makes cohomology of families of varieties a computable theory.

Visual Beginner

A diagram with two squares side by side. On the left, the Cartesian square — a pullback diagram with four corners $X^{\prime }$, $X$, $S^{\prime }$, $S$ and four morphisms forming a commuting square. On the right, the same square but labelled with the four operations $f^{\ast }$, $f_{\ast }$, $f_{!}$, $f^{!}$ assigned to the morphism $f:X\to S$, with an isomorphism sign in the middle indicating the base-change isomorphism between the pullback-pushforward and pushforward-pullback compositions.

The picture captures the structural shape: the six operations attach to morphisms, the adjunctions pair them into two adjoint pairs $(f^{\ast }⊣f_{\ast })$ and $(f_{!}⊣f^{!})$, and the base-change isomorphism connects the operations across a pullback square. A reader who internalises this picture will recognise the six-functor diagram everywhere — in derived algebraic geometry, in étale cohomology, and in the analytic and topological theory of sheaves on locally compact spaces.

Worked example Beginner

Compute the pullback and pushforward operations for the simplest morphism: the constant map $f:X\to \mathrm{pt}$ from a topological space $X$ to a single point. Then verify the adjunction relation in a small example.

Step 1. Pullback. Given a vector space $V$ on the point (a sheaf on a single point is just a vector space), the pullback $f^{\ast }V$ is the constant sheaf on $X$ with stalks all equal to $V$. So the operation $f^{\ast }$ takes a vector space and spreads it uniformly over the source space.

Step 2. Direct image. Given a sheaf $F$ on $X$, the direct image $f_{\ast }F$ on the point is the vector space of global sections $\Gamma (X,F)=F(X)$. So the operation $f_{\ast }$ takes a sheaf on $X$ and extracts its global sections as a single vector space.

Step 3. Adjunction. The adjunction $f^{\ast }⊣f_{\ast }$ becomes the identity $\mathrm{Hom}(f^{\ast }V,F)=\mathrm{Hom}(V,f_{\ast }F)$, which spells out as $\mathrm{Hom}(\underline{V},F)=\mathrm{Hom}(V,\Gamma (X,F))$. The left side is maps from the constant $V$-sheaf into $F$, which is the same data as maps from $V$ to the global sections of $F$, which is the right side. The two sides are tautologically the same data.

Step 4. Proper direct image. If $X$ is compact, the proper direct image $f_{!}F$ agrees with $f_{\ast }F=\Gamma (X,F)$. If $X$ is not compact, $f_{!}F$ is the compactly-supported global sections ${\Gamma }_c(X,F)$ — global sections that vanish outside a compact subset. For $X=\mathbb{R}$ and $F$ the constant sheaf with stalk $V$, the compactly-supported sections are zero (a constant non-zero function does not vanish outside a compact set), so $f_{!}\underline{V}=0$ even though $f_{\ast }\underline{V}=V$. The two pushforwards diverge on non-compact spaces, and the divergence is exactly what makes $f_{!}$ a distinct operation worth tracking.

What this tells us. Even in the simplest possible example — constant map to a point — the four functors $f^{\ast }$, $f_{\ast }$, $f_{!}$, $f^{!}$ carry distinct geometric content: $f^{\ast }$ spreads data, $f_{\ast }$ collects sections, $f_{!}$ collects compactly-supported sections, and the adjunction connects them. The same picture generalises to any morphism of geometric spaces, and the universality of this package is the reason it organises so much of modern geometry.

Check your understanding Beginner

Formal definition Intermediate+

Let $f:X\to Y$ be a morphism in one of two settings: (a) a morphism of locally compact Hausdorff spaces of finite cohomological dimension, with sheaves of $k$-modules for a commutative noetherian ring $k$; or (b) a morphism of finite-type schemes over a noetherian base, with derived categories of quasi-coherent or constructible sheaves. The six-functor formalism consists of six derived functors attached to $f$ together with two adjunction pairs and several compatibility relations.

Definition (the six operations). The six functors associated to a morphism $f:X\to Y$ and the symmetric monoidal structure on the derived category of $X$ are:

1. Derived inverse image $L{f}^{\ast }:D(Y)\to D(X)$, the left derived functor of the sheaf-theoretic pullback $f^{-1}$ (or its module-theoretic refinement $f^{\ast }$ in the coherent setting).
2. Derived direct image $R{f}_{\ast }:D(X)\to D(Y)$, the right derived functor of the pushforward $f_{\ast }$.
3. Derived direct image with proper support $R{f}_{!}:D(X)\to D(Y)$, defined by $R{f}_{!}F=R{f}_{\ast }(j_{!}\tilde{F})$ where $f=\overline{f}\circ j$ is a Nagata compactification (a factorisation through a proper morphism $\overline{f}$ via an open immersion $j$); independent of the compactification up to canonical isomorphism. In the topological setting $f_{!}F$ is the subsheaf of $f_{\ast }F$ of compactly-supported sections.
4. Twisted inverse image $f^{!}:D(Y)\to D(X)$, the right adjoint to $R{f}_{!}$ (existence: Brown representability for the compactly-generated triangulated category $D(X)$, or explicit Hartshorne-Verdier construction via dualising complexes for finite-type schemes).
5. Derived tensor product $-{\otimes }_{{\mathcal{O}}_X}^L-:D(X)\times D(X)\to D(X)$.
6. Derived internal Hom $R\mathcal{H}o{m}_{{\mathcal{O}}_X}(-,-):D(X{)}^{\mathrm{op}}\times D(X)\to D(X)$.

Definition (adjunctions). The six functors satisfy two adjunction pairs across $f$:

$$L{f}^{\ast }⊣R{f}_{\ast },R{f}_{!}⊣f^{!},$$

and one adjunction internal to each space:

$$-{\otimes }^{L}F⊣R\mathcal{H}om(F,-).$$

The first adjunction $(L{f}^{\ast }⊣R{f}_{\ast })$ exists for every morphism; the second $(R{f}_{!}⊣f^{!})$ requires $f$ to satisfy a finiteness condition (locally compact Hausdorff with finite cohomological dimension, or separated of finite type for schemes) for $f^{!}$ to exist as a genuine right adjoint.

Definition (Cartesian square). A Cartesian square is a commutative square of morphisms $$ \begin{array}{ccc} X' & \xrightarrow{\tilde g} & X \ \downarrow{\tilde f} & & \downarrow{f} \ S' & \xrightarrow{g} & S \end{array} $$ in which $X^{\prime }$ is the fibre product $X{\times }_S{S}^{\prime }$ and $\tilde{f},\tilde{g}$ are the canonical projections.

Definition (base change). Given a Cartesian square as above, the base-change isomorphism is a canonical isomorphism of functors $D(X)\to D(S^{\prime })$, $$ g^* Rf_! \xrightarrow{\sim} R\tilde f_! \tilde g^. $$ By adjunction, this is equivalent to the dual isomorphism $\tilde g_ f^! \xrightarrow{\sim} \tilde f^! g_*$ on the right adjoints (with appropriate finiteness hypotheses).

Definition (projection formula). For a morphism $f:X\to Y$ in the six-functor setting and objects $F\in D(X)$, $G\in D(Y)$, the projection formula is the canonical isomorphism $$ Rf_!(F \otimes^L Lf^* G) \xrightarrow{\sim} Rf_! F \otimes^L G \in D(Y). $$ Its dual form, using the adjunction $(R{f}_{!}⊣f^{!})$ and the internal Hom, reads $$ R\mathcal{H}om(Lf^* G, f^! H) \cong f^! R\mathcal{H}om(G, H). $$

Definition (relative dualising complex). For a morphism $f:X\to Y$ in the six-functor setting, the relative dualising complex is $$ \omega^\bullet_{X/Y} = f^!(\mathcal{O}Y) \in D(X). $$ When $f$ is a smooth morphism of relative dimension $d$, $\omega^\bullet{X/Y} = \Omega^d_{X/Y}[d]$,thecanonicalsheafindegree$-d$.When$f$isproper,thedualityformula$f^! \cong D_X \circ Lf^* \circ D_Y$relatesthetwistedinverseimagetothedualisingfunctors$D_X(-) = R\mathcal{H}om(-, \omega^\bullet_X)$.

Counterexamples to common slips

• $R{f}_{!}$ is not the right derived functor of a single underived functor on the abelian category of sheaves — its construction goes through Nagata compactification or, in the topological setting, the underived $f_{!}$ of compactly-supported sections followed by derivation. Treating $R{f}_{!}$ as $R(f_{!})$ in the naïve sense fails when $f$ is not proper.
• The base-change isomorphism for $R{f}_{\ast }$ in general fails without proper-or-flat hypotheses; the analogous statement for $R{f}_{!}$ holds for arbitrary Cartesian squares with $f$ separated of finite type. Confusing the two versions of base change is the most common error in computations.
• The right adjoint $f^{!}$ exists in general only when the appropriate finiteness conditions are met. In the unbounded derived category of all schemes the right adjoint to $R{f}_{!}$ may fail to exist; the standard setting is finite-type morphisms of noetherian schemes (Hartshorne-Verdier) or locally compact Hausdorff spaces of finite cohomological dimension (Kashiwara-Schapira).
• For a smooth morphism of relative dimension $d$, $f^{!}=L{f}^{\ast }[2d]$ in the étale or analytic setting and $f^{!}=L{f}^{\ast }\otimes {\omega }_{X/Y}[d]$ in the coherent setting; the difference reflects whether one works with the relative orientation in degree-$2d$ (étale Tate twist) or in degree-$d$ (coherent canonical-sheaf twist).

Key theorem with proof Intermediate+

Theorem (base change for $R{f}_{!}$; SGA 4 XVII Théorème 5.2.6 / Kashiwara-Schapira III.3.1.9). Let

$$\begin{array}{ccc}{X}^{\prime }& \stackrel{\tilde{g}}{\to }& X\\ \downarrow \tilde{f}& & \downarrow f\\ S^{\prime }& \stackrel{g}{\to }& S\end{array}$$

be a Cartesian square of locally compact Hausdorff spaces of finite cohomological dimension (or separated finite-type morphisms of noetherian schemes), with $X^{\prime }=X{\times }_S{S}^{\prime }$. For any $F\in D^{+}(X)$ there is a canonical isomorphism in $D^{+}(S^{\prime })$,

$$g^{\ast }R{f}_{!}F\stackrel{\sim }{\to }R{\tilde{f}}_{!}{\tilde{g}}^{\ast }F.$$

Proof. The argument has four steps. First, construct the base-change comparison morphism using the adjunctions. Second, reduce to the case where $f$ is either proper or an open immersion by Nagata compactification. Third, prove the proper case by direct stalk-and-fibre analysis. Fourth, prove the open-immersion case.

Step 1: construction of the comparison map. The unit of the adjunction $(g^{\ast }⊣g_{\ast })$ on the bottom of the square gives a morphism $\eta :\mathrm{id}\to g_{\ast }{g}^{\ast }$. Applied to $R{f}_{!}F$, this is $R{f}_{!}F\to g_{\ast }{g}^{\ast }R{f}_{!}F$. By adjunction $(g^{\ast }⊣g_{\ast })$, the comparison map we seek is equivalent to a map $R{f}_{!}F\to g_{\ast }R{\tilde{f}}_{!}{\tilde{g}}^{\ast }F=R{f}_{!}{\tilde{g}}_{\ast }{\tilde{g}}^{\ast }F$ (using that the diagram commutes: $g\circ \tilde{f}=f\circ \tilde{g}$, so $g_{\ast }R{\tilde{f}}_{!}=R{f}_{\ast }{\tilde{g}}_{\ast }$ at the underived level extends via $R{f}_{!}$ derivation). Applying $R{f}_{!}$ to the unit $F\to {\tilde{g}}_{\ast }{\tilde{g}}^{\ast }F$ produces this map. Composing the two gives the canonical base-change comparison morphism

$$g^{\ast }R{f}_{!}F\to R{\tilde{f}}_{!}{\tilde{g}}^{\ast }F,$$

functorial in $F$ and in the square. The theorem asserts that this canonical map is an isomorphism.

Step 2: reduction to proper and open-immersion cases. By Nagata's compactification theorem (for schemes) or by direct argument (for topological spaces), every separated finite-type morphism $f:X\to S$ factors as $f=\overline{f}\circ j$ where $\overline{f}:\overline{X}\to S$ is proper and $j:X\hookrightarrow \overline{X}$ is an open immersion. Pull back the entire square along $g:S^{\prime }\to S$: this gives a Cartesian square for $\overline{f}$ and one for $j$ separately, with $X^{\prime }={\overline{X}}^{\prime }{\times }_{\overline{X}}X$ via the open immersion $\tilde{j}:X^{\prime }\hookrightarrow {\overline{X}}^{\prime }$. Since $R{f}_{!}=R{\overline{f}}_{\ast }\circ j_{!}$ by the Nagata definition (where $j_{!}$ is extension by zero — exact on sheaves, so no derivation is needed for $j_{!}$ on sheaves alone), the base-change isomorphism for $f$ follows from base change for $R{\overline{f}}_{\ast }$ on a proper morphism and base change for $j_{!}$ on an open immersion.

Step 3: proper case. Suppose $f$ is proper, so $R{f}_{!}=R{f}_{\ast }$. The proper-base-change theorem (the central theorem) asserts that the comparison map $g^{\ast }R{f}_{\ast }F\to R{\tilde{f}}_{\ast }{\tilde{g}}^{\ast }F$ is an isomorphism. The standard proof goes by checking stalks at every point $s\in S^{\prime }$ and reducing to the fibre. At the stalk over $s\in S^{\prime }$, the left side is $(g^{\ast }R{f}_{\ast }F{)}_s=(R{f}_{\ast }F{)}_{g(s)}=H^{\ast }(X_{g(s)},F{|}_{X_{g(s)}})$ where $X_{g(s)}=f^{-1}(g(s))$ is the fibre, using the standard identification of the stalk of $R{f}_{\ast }F$ on a proper morphism with the cohomology of the fibre (proper base change in its stalk form). The right side is $(R{\tilde{f}}_{\ast }{\tilde{g}}^{\ast }F{)}_s=H^{\ast }(X_s^{\prime },({\tilde{g}}^{\ast }F){|}_{X_s^{\prime }})$ where $X_s^{\prime }={\tilde{f}}^{-1}(s)$ is the fibre on the pulled-back side. But $X_s^{\prime }=X{\times }_S{S}^{\prime }{\times }_{S^{\prime }}\{s\}=X{\times }_S\{s\}\cdot (S^{\prime }{\times }_{S^{\prime }}\{s\})=X_{g(s)}$ — the fibre of the pullback over $s$ equals the fibre of the original over $g(s)$, since both compute the same fibre product. Under this fibre identification, ${\tilde{g}}^{\ast }F$ restricts to the same sheaf as $F$ on the fibre, and the two stalks agree.

The fibre identification uses only the universal property of the pullback, the stalk identification uses only properness (the proper-base-change theorem for ordinary direct image on the abelian category of sheaves, which lifts to the derived functor by injective-resolution arguments). Together they give the proper case of base change.

Step 4: open-immersion case. Suppose $f=j$ is an open immersion. Then $j_{!}F$ is the extension-by-zero sheaf, with stalk $F_x$ at $x\in X$ and zero elsewhere. The pulled-back morphism $\tilde{j}:X^{\prime }\hookrightarrow S^{\prime }$ is also an open immersion. The comparison map $g^{\ast }{j}_{!}F\to {\tilde{j}}_{!}{\tilde{g}}^{\ast }F$ on stalks reads: at $s\in S^{\prime }$ with $s\notin X^{\prime }$, both sides are zero; at $s\in X^{\prime }\subset S^{\prime }$, the left side is $(j_{!}F{)}_{g(s)}=F_{g(s)}$ (since $g(s)\in X$ when $s\in X^{\prime }$, by the Cartesian-square fact $X^{\prime }=X{\times }_S{S}^{\prime }$), and the right side is $({\tilde{g}}^{\ast }F{)}_s=F_{\tilde{g}(s)}=F_{g(s)}$ (the projection $\tilde{g}$ on $X^{\prime }\subset X{\times }_S{S}^{\prime }$ sends $s$ to its $X$-component, which is the unique point of $X$ mapping to $g(s)$ via the Cartesian property). The two stalks agree, so the comparison is an isomorphism. $\square$

Bridge. The base-change theorem for $R{f}_{!}$ is the central compatibility of the six-functor formalism, and the foundational reason the formalism works is that it builds toward a unified calculus of derived pullback and pushforward across pullback squares — the comparison $g^{\ast }R{f}_{!}\cong R{\tilde{f}}_{!}{\tilde{g}}^{\ast }$ identifies pull-then-push with push-then-pull, and this is exactly what makes cohomology of families a functorial theory. The bridge is the projection formula $R{f}_{!}(F{\otimes }^{L}L{f}^{\ast }G)\cong R{f}_{!}F{\otimes }^{L}G$, which packages the same compatibility for the symmetric-monoidal structure: pulling back a tensor product through $f^{\ast }$ and then pushing through $f_{!}$ identifies with pushing the first factor and tensoring on the target. Putting these together, the six-functor formalism generalises the projection formula of classical algebraic topology to the derived category, and is dual to the Grothendieck-duality identification $f^{!}({\mathcal{O}}_Y)={\omega }_{X/Y}^{\bullet }$ that locates the relative dualising complex inside the formalism.

This pattern appears again in 04.08.03 (Serre duality), where the duality pairing $H^p(X,F)\otimes H^{n-p}(X,{\omega }_X\otimes F^{\vee })\to k$ on a smooth projective variety is the global-sections case of the six-functor adjunction $(R{f}_{!}⊣f^{!})$ for the structure map $f:X\to \mathrm{pt}$, and in 04.03.15 (Leray spectral sequence), where the spectral sequence $H^p(S,R^q{f}_{\ast }F)\Rightarrow H^{p+q}(X,F)$ is the cohomology spectral sequence of the functor $R{f}_{\ast }$ derived from the adjunction $(f^{\ast }⊣R{f}_{\ast })$. The central insight is that every classical duality theorem and every classical base-change isomorphism is a corollary of the universal six-functor package applied to a specific morphism with specific finiteness data.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — Mathlib has the derived-category framework for an abelian category and the higher-direct-image functor $R^q{f}_{\ast }$ packaged from the prereq unit on higher direct images, but the named six-functor formalism on a morphism of schemes is not assembled. The intended formalisation reads schematically:

import Mathlib.AlgebraicGeometry.Scheme
import Mathlib.AlgebraicGeometry.Morphisms.Proper
import Mathlib.AlgebraicGeometry.Morphisms.Separated
import Mathlib.CategoryTheory.Localization.DerivedCategory

namespace Codex.AlgebraicGeometry.SixFunctor

variable {X Y : Scheme} (f : X ⟶ Y)

/-- The four functors of the six-functor formalism attached to a separated
    finite-type morphism of noetherian schemes. -/
noncomputable def Lpullback : DerivedCategory Y ⥤ DerivedCategory X := sorry
noncomputable def Rpushforward : DerivedCategory X ⥤ DerivedCategory Y := sorry
noncomputable def Rproper_pushforward [Separated f] [LocallyOfFiniteType f] :
    DerivedCategory X ⥤ DerivedCategory Y := sorry
noncomputable def twisted_pullback [Separated f] [LocallyOfFiniteType f] :
    DerivedCategory Y ⥤ DerivedCategory X := sorry

/-- The first adjunction of the six-functor formalism. -/
theorem adj_pullback_pushforward :
    Adjunction (Lpullback f) (Rpushforward f) :=
  sorry

/-- The second adjunction of the six-functor formalism. -/
theorem adj_proper_twisted [Separated f] [LocallyOfFiniteType f] :
    Adjunction (Rproper_pushforward f) (twisted_pullback f) :=
  sorry

/-- Base change for $Rf_!$ along a Cartesian square. -/
theorem base_change_proper_pushforward
    {S S' X' : Scheme} (f : X ⟶ S) (g : S' ⟶ S)
    (hsq : IsPullback (X' := X ×ₛ[f] S') ...) :
    g.Lpullback ⋙ Rproper_pushforward f ≅
      Rproper_pushforward (pullback_morphism f g) ⋙ (pullback_proj g).Lpullback :=
  sorry

/-- Projection formula. -/
theorem projection_formula
    (F : DerivedCategory X) (G : DerivedCategory Y) :
    Rproper_pushforward f (F ⊗ Lpullback f G) ≅
      Rproper_pushforward f F ⊗ G :=
  sorry

end Codex.AlgebraicGeometry.SixFunctor

The proof gap is substantive on many fronts. The construction of $R{f}_{!}$ requires either Nagata's compactification theorem (a major Mathlib gap on its own) or a direct construction via the topology of compact supports. The right adjoint $f^{!}$ requires Brown representability for the compactly-generated derived category, plus the dualising-complex API. Base change requires verifying the comparison map's naturality and proving its invertibility under the appropriate hypotheses. The projection formula requires the symmetric monoidal structure on the derived category to be compatible with $R{f}_{!}$, which is a tightly coupled theorem about derived tensor products of complexes. Each component is formalisable in principle but requires substantial coordinated infrastructure.

Advanced results Master

Theorem (Grothendieck duality for proper morphisms; Hartshorne Residues and Duality Ch. VI). Let $f:X\to Y$ be a proper morphism of noetherian schemes. There is a canonical isomorphism of functors $D(X)\to D(Y)$, $$ Rf_* R\mathcal{H}om_X(F, f^! G) \xrightarrow{\sim} R\mathcal{H}om_Y(Rf_* F, G), $$ natural in $F\in D(X)$ and $G\in D(Y)$. Equivalently, $f^! \cong D_X \circ Lf^ \circ D_Y$where$D_X(-) = R\mathcal{H}om(-, \omega^\bullet_X)$isthedualisingfunctoron$X$and$\omega^\bullet_X = f^!(\mathcal{O}_Y)$ is the relative dualising complex.*

Grothendieck duality is the master theorem of the six-functor package on the coherent-sheaf side, and it identifies the right adjoint $f^{!}$ in the proper case as a composite of pullback with two duality reflections. For $f$ a smooth proper morphism of relative dimension $d$, this reduces to the classical Serre duality on a smooth proper variety; for $f$ a closed immersion, it reduces to the local-cohomology adjunction $i_{\ast }\circ i^{!}⊣i^{\ast }$ on a closed subscheme. The universal nature of the duality formula is what makes Grothendieck duality the foundational tool for trace maps, residues, and the Riemann-Roch theorem in the derived-category form.

Theorem (Verdier duality on locally compact spaces; Kashiwara-Schapira III.3.1). Let $X$ be a locally compact Hausdorff space of finite cohomological dimension. Verdier duality is the natural isomorphism $$ R\Gamma_c(X, F)^\vee \xrightarrow{\sim} R\Gamma(X, D_X F), $$ where $D_{X}F=R\mathcal{H}om(F,{\omega }_X^{\bullet })$ and ${\omega }_X^{\bullet }=a^{!}(k)$ for the structure map $a:X\to \mathrm{pt}$.

Verdier duality is the topological counterpart of Grothendieck duality, and it identifies compactly-supported cohomology of a sheaf with ordinary cohomology of its Verdier dual. For $X$ an oriented manifold of dimension $n$, ${\omega }_X^{\bullet }=k_X[n]$ and Verdier duality recovers Poincaré-Lefschetz duality: $H_c^p(X,k{)}^{\vee }\cong H^{n-p}(X,k)$. The general statement extends Poincaré duality to non-orientable spaces, singular spaces, and constructible sheaves; Goresky-MacPherson's intersection homology theory is precisely the Verdier duality for intersection complexes on stratified spaces.

Theorem (proper-base-change theorem in étale cohomology; SGA 4 XII / Deligne Cohomologie étale §1). Let $f:X\to S$ be a proper morphism of noetherian schemes and let $\ell$ be a prime invertible on $S$. The proper-base-change isomorphism $$ g^* Rf_* F \xrightarrow{\sim} R\tilde f_* \tilde g^* F $$ holds for constructible $\ell$-adic sheaves $F$ on $X$ and any base-change $g:S^{\prime }\to S$.

Proper-base-change in étale cohomology is the central technical theorem of SGA 4, and it is the input to Deligne's proof of the Weil conjectures, the Lefschetz trace formula for the Frobenius, and the existence of cohomology of varieties over arithmetic bases. The proof in étale cohomology is substantially harder than the proof for ordinary sheaves on a topological space — it requires the construction of the proper-direct-image functor $R{f}_{!}$ on étale sites via Nagata compactification, plus a delicate analysis of stalk cohomology in terms of the henselisation of the local ring at the base point.

Theorem (six-functor formalism for stable $\infty$-categories; Lurie Higher Algebra §6.4 and SAG Ch. 3). The six-functor formalism on a geometric category $\mathcal{C}$ (such as schemes, spaces, or analytic spaces) lifts to a presheaf of symmetric monoidal stable $\infty$-categories $D:{\mathcal{C}}^{\mathrm{op}}\to {\mathrm{Cat}}_{\infty }^{\otimes }$ together with six operations satisfying the adjunctions, projection formula, and base-change isomorphisms in the strict $\infty$-categorical sense.

The $\infty$-categorical lift of the six-functor formalism is Lurie's framework for formalising base-change and projection-formula compatibilities as coherent higher-categorical structures. The $\infty$-categorical version automatically captures all higher coherences — the fact that base change is natural not only as a map of derived functors but as a map of presentable $\infty$-categories with all higher associativities — and is the technical foundation for the modern motivic, étale, and derived algebraic geometry programmes. Bhatt-Scholze's prismatic cohomology, Clausen-Scholze's condensed mathematics, and the Fargues-Scholze geometrisation of the local Langlands correspondence all use the $\infty$-categorical six-functor formalism as their working environment.

Theorem (Beilinson-Bernstein-Deligne decomposition theorem; BBD Faisceaux pervers 1982). Let $f:X\to Y$ be a proper morphism of finite-type schemes over a field, with $X$ smooth. Then there is a (non-canonical) direct-sum decomposition $$ Rf_* \overline{\mathbb{Q}}{\ell, X}[\dim X] \cong \bigoplus_i \mathrm{IC}{Z_i}(\mathcal{L}i)[n_i] $$ *in $D^b_c(Y, \overline{\mathbb{Q}}\ell)$,wherethe$\mathrm{IC}_{Z_i}(\mathcal{L}_i)$aresimpleperversesheavesonclosedsubvarieties$Z_i \subseteq Y$withlocal-systemcoefficients$\mathcal{L}_i$,andthe$n_i$ are integer shifts.*

The BBD decomposition theorem is the deepest application of the six-functor formalism in algebraic geometry, and it asserts that the derived pushforward of a constant sheaf along a proper map decomposes (over a base field) into a sum of irreducible "atomic" pieces — the intersection complexes of closed subvarieties. The theorem requires the full six-functor apparatus plus the theory of perverse sheaves (the heart of the perverse $t$-structure on the derived category of constructible sheaves), plus Deligne's purity theorem from his proof of the Weil conjectures. It is the technical underpinning of the Kazhdan-Lusztig theory of Hecke algebras, the Beilinson-Bernstein localisation of representation theory, and the geometric Langlands programme.

Theorem (proper smooth base change as cohomology with constant fibre dimension). Let $f:X\to S$ be a proper smooth morphism of noetherian schemes. Then the higher direct images $R^q f_ \mathbb{Z}\ell$arelocalsystemson$S$,withmonodromyin$\mathrm{GL}(H^q(X_s, \mathbb{Z}\ell))$.*

The proper-smooth case of base change asserts that not only do the dimensions $\dim H^q(X_s,{\mathbb{Z}}_{\ell })$ stay constant along $S$, but the cohomology groups assemble into a local system on $S$ — the family is locally trivialisable. This is the basis for Deligne's proof of the Weil conjectures: the cohomology of a smooth projective variety over ${\mathbb{F}}_q$ is the fibre of a local system on a moduli space, and the Frobenius monodromy of the local system controls the zeta function. The general (non-smooth) case admits a similar statement via constructible coefficients but requires the technical input of perverse sheaves.

Synthesis. The six-functor formalism builds toward a unified calculus that organises every derived-category construction in algebraic geometry and topology into one universal package, and the foundational reason it works is that every morphism of geometric spaces carries six naturally attached functors — $f^{\ast },f_{\ast },f_{!},f^{!},{\otimes }^L,R\mathcal{H}om$ — that satisfy a tight network of adjunctions, base-change isomorphisms, and a projection formula. The package is dual to itself: the duality formula $f^{!}\cong D_X\circ L{f}^{\ast }\circ D_Y$ for proper $f$ identifies the twisted inverse image as a composite of pullback with two duality reflections, and the base-change isomorphism $g^{\ast }R{f}_{!}\cong R{\tilde{f}}_{!}{\tilde{g}}^{\ast }$ for a Cartesian square is exactly the universal compatibility that makes the cohomology of a family of varieties a functorial theory. Putting these together, the central insight is that classical theorems — Poincaré duality, Serre duality, the projection formula in algebraic topology, proper-base-change in étale cohomology, the Lefschetz trace formula, and Grothendieck duality for proper morphisms — are all corollaries of the six-functor package applied to specific morphisms with specific finiteness data. This is exactly the unification that Grothendieck envisaged in 1958 in his correspondence with Serre and that Verdier, Deligne, and the SGA 4 / 4½ collaborators formalised between 1962 and 1977.

The formalism appears again in 04.08.03 (Serre duality), where the global-sections duality on a smooth projective variety is the structure-map specialisation of the second six-functor adjunction; in 04.03.07 (higher direct images and base change), where the cohomology-and-base-change theorem is the underived shadow of the six-functor base-change isomorphism; and in 04.03.15 (Leray spectral sequence), where the spectral sequence for the composition of two pushforwards is derived from the six-functor adjunction $(f^{\ast }⊣R{f}_{\ast })$. The bridge to motivic and arithmetic cohomology is the extension of the formalism to constructible $\ell$-adic sheaves, $p$-adic sheaves, and motivic sheaves; the bridge to higher algebra is the $\infty$-categorical lift of Lurie, which makes the formalism a presheaf of stable symmetric monoidal $\infty$-categories with all higher coherences built in. The recursion stabilises: the six-functor formalism is the universal package for derived-category-era geometry, and every subsequent generalisation (perverse sheaves, prismatic cohomology, condensed sheaves, motivic cohomology) is the six-functor formalism applied to a more general site or geometric category.

The synthesis is structural: every classical duality theorem in algebraic geometry and algebraic topology is a corollary of the second six-functor adjunction $(R{f}_{!}⊣f^{!})$ applied to a specific morphism, and every classical base-change theorem is a corollary of the six-functor base-change isomorphism on a Cartesian square. The six-functor formalism is the universal duality oracle, and the input is the morphism together with its finiteness data.

Full proof set Master

Proposition (uniqueness of $R{f}_{!}$ from a Nagata compactification). Let $f:X\to Y$ be a separated morphism of finite type between noetherian schemes, and let $f={\overline{f}}_1\circ j_1={\overline{f}}_2\circ j_2$ be two Nagata compactifications, with ${\overline{f}}_i:{\overline{X}}_i\to Y$ proper and $j_i:X\hookrightarrow {\overline{X}}_i$ open immersions. Then the two definitions $R\overline{f}_{1,} \circ j_{1,!}$and$R\overline{f}{2,*} \circ j{2,!}$arecanonicallyisomorphicasfunctors$D^+(X) \to D^+(Y)$.*

Proof. Form the fibre product $\overline{X}={\overline{X}}_1{\times }_Y{\overline{X}}_2$ over $Y$, which is also proper over $Y$ via the structure morphism $\overline{f}:\overline{X}\to Y$. The two open immersions $j_1,j_2$ induce a diagonal open immersion $j:X\hookrightarrow \overline{X}$ (using that $X\to {\overline{X}}_1{\times }_Y{\overline{X}}_2$ via the universal property of the fibre product is an open immersion when both $j_1,j_2$ are open immersions and $X$ maps separatedly to $Y$). The projections $p_i:\overline{X}\to {\overline{X}}_i$ are proper, and $p_i\circ j=j_i$. We compute, for $F\in D^{+}(X)$:

$$R{\overline{f}}_{1,\ast }\circ j_{1,!}F=R{\overline{f}}_{1,\ast }\circ R{p}_{1,\ast }\circ j_{!}F=R{\overline{f}}_{\ast }\circ j_{!}F,$$

using $j_{1,!}=R{p}_{1,\ast }\circ j_{!}$ (extension-by-zero pushed forward along the proper $p_1$ remains extension-by-zero on the larger compactification — a standard consequence of open-immersion exactness). Similarly $R{\overline{f}}_{2,\ast }\circ j_{2,!}F=R{\overline{f}}_{\ast }\circ j_{!}F$. The two expressions agree with the common middle expression. $\square$

Proposition (projection formula on the abelian level for an open immersion). Let $j:U\hookrightarrow X$ be an open immersion of locally compact Hausdorff spaces. For any sheaves $F$ on $U$ and $G$ on $X$, the projection formula $j_!(F \otimes j^ G) \cong j_! F \otimes G$holdsasanisomorphismofsheaveson$X$.*

Proof. Both sides are sheaves on $X$. At a stalk $x\in X$:

• If $x\notin U$: $(j_{!}F{)}_x=0$ by extension-by-zero, so the right side stalk is $0\otimes G_x=0$; the left side stalk is $(j_{!}(F\otimes j^{\ast }G){)}_x=0$ also by extension-by-zero. Both sides are zero.
• If $x\in U$: $(j_{!}F{)}_x=F_x$ and $(j^{\ast }G{)}_x=G_x$ (the stalk of pullback equals the stalk of the original at points of the open). The right-side stalk is $F_x\otimes G_x$; the left-side stalk is $(F\otimes j^{\ast }G{)}_x=F_x\otimes (j^{\ast }G{)}_x=F_x\otimes G_x$. The two stalks agree.

The stalk-by-stalk agreement together with the naturality of the comparison map produces the desired isomorphism. The derived version follows by passage through flat or projective resolutions in the symmetric monoidal derived category. $\square$

Proposition (base change for an open immersion). Let $g:S^{\prime }\to S$ be a morphism of schemes and let $j:U\hookrightarrow S$ be an open immersion, with Cartesian square $$ \begin{array}{ccc} U' & \xrightarrow{\tilde g} & U \ \downarrow{\tilde j} & & \downarrow{j} \ S' & \xrightarrow{g} & S, \end{array} $$ so $U^{\prime }=g^{-1}(U)\hookrightarrow S^{\prime }$ is an open immersion and $\tilde{g}=g{|}_{U^{\prime }}$. The base-change isomorphism $g^ j_! F \cong \tilde j_! \tilde g^* F$holdsforanysheaf$F$on$U$.*

Proof. Both sides are sheaves on $S^{\prime }$. Compute stalks at $s\in S^{\prime }$:

• If $s\notin U^{\prime }$: $({\tilde{j}}_{!}{\tilde{g}}^{\ast }F{)}_s=0$ by extension-by-zero, and $(g^{\ast }{j}_{!}F{)}_s=(j_{!}F{)}_{g(s)}=0$ since $g(s)\notin U$ when $s\notin U^{\prime }=g^{-1}(U)$.
• If $s\in U^{\prime }$: $({\tilde{j}}_{!}{\tilde{g}}^{\ast }F{)}_s=({\tilde{g}}^{\ast }F{)}_s=F_{\tilde{g}(s)}$ where $\tilde{g}(s)\in U$; and $(g^{\ast }{j}_{!}F{)}_s=(j_{!}F{)}_{g(s)}=F_{g(s)}=F_{\tilde{g}(s)}$ since $\tilde{g}=g{|}_{U^{\prime }}$.

The two stalks agree at every point, and the comparison map is the natural one, so the base-change isomorphism holds on sheaves. The derived version $g^{\ast }R{j}_{!}\cong R{\tilde{j}}_{!}{\tilde{g}}^{\ast }$ follows from exactness of $j_{!}$ and ${\tilde{j}}_{!}$ — extension-by-zero is exact, so $R{j}_{!}=j_{!}$ and $R{\tilde{j}}_{!}={\tilde{j}}_{!}$ as derived functors. $\square$

Proposition (the adjunction $(f_{!}⊣f^{!})$ for $f$ a closed immersion). Let $i:Z\hookrightarrow X$ be a closed immersion of schemes. The functor $i_ = i_!$(closed-immersiondirectimage,whichagreeswithproper-direct-imagesince$i$isproper)isleftadjointtothelocal-cohomologyfunctor$i^! F = R\underline{\Gamma}_Z(F)|_Z$.*

Proof. The adjunction $\mathrm{Hom}(i_{\ast }G,F)=\mathrm{Hom}(G,i^{!}F)$ for $G\in D(Z)$ and $F\in D(X)$ unwinds as follows. A morphism $\varphi :i_{\ast }G\to F$ in $D(X)$ is a morphism of complexes of sheaves on $X$ from the closed-support sheaf $i_{\ast }G$ to $F$. By the support condition on $i_{\ast }G$, the image of $\varphi$ lies in the local-cohomology subsheaf $R{\underline{\Gamma }}_Z(F)$ — every section of $i_{\ast }G$ has support in $Z$, so the morphism into $F$ factors through the largest subsheaf of $F$ supported on $Z$, which is $R{\underline{\Gamma }}_Z(F)$. The morphism $\varphi$ thus produces a morphism $i_{\ast }G\to R{\underline{\Gamma }}_Z(F)$, equivalently (by restriction to $Z$ via the equivalence between sheaves on $X$ supported on $Z$ and sheaves on $Z$) a morphism $G\to (R{\underline{\Gamma }}_Z(F)){|}_Z=i^{!}F$.

The reverse direction extends a morphism $\psi :G\to i^{!}F=R{\underline{\Gamma }}_Z(F){|}_Z$ to a morphism $i_{\ast }G\to R{\underline{\Gamma }}_Z(F)\to F$ in $D(X)$. The two passages are inverse, producing the adjunction isomorphism $\mathrm{Hom}(i_{\ast }G,F)=\mathrm{Hom}(G,i^{!}F)$. The naturality in $G$ and $F$ is built into the construction. $\square$

Proposition (proper-base-change isomorphism on stalks, fibre form). Let $f:X\to S$ be a proper morphism of noetherian schemes and let $s\in S$ be a geometric point. Then the stalk of $Rf_ F$at$s$iscanonicallyisomorphictothecohomologyofthefibre$X_s = X \times_S {s}$,* $$ (R^q f_* F)s \xrightarrow{\sim} H^q(X_s, F|{X_s}). $$

Proof. This is the central content of proper base change in its stalk form, and is proved by reduction to the case $S=\mathrm{Spec}(R)$ for $R$ a strictly henselian local ring (the strict henselisation at $s$). Over such an $R$, the morphism $f:X_R\to \mathrm{Spec}(R)$ is proper, and the strict henselisation property means that the residue field $k(s)$ is separably closed and the local rings of $S$ at $s$ have the right "small neighbourhood" topology so that the natural map from sections over a neighbourhood to sections over the fibre is an isomorphism after derivation.

Specifically: $(R{f}_{\ast }F{)}_s={\mathrm{colim}}_{s\in U}R\Gamma (f^{-1}(U),F)$ over small étale neighbourhoods $U$ of $s$. By proper-base-change for an étale neighbourhood (which is itself the simplest case), this colimit equals $R\Gamma (X_s,F{|}_{X_s})=H^{\ast }(X_s,F{|}_{X_s})$. The proper hypothesis enters in commuting the colimit with the cohomology — without properness, the limit could fail to commute with the higher direct image, and the stalk would diverge from the fibre cohomology. The full proof for noetherian schemes is recorded in SGA 4 XII Théorème 5.1 ^{[source pending]}, and the corresponding theorem on topological spaces is in Kashiwara-Schapira II.6.7 ^{[source pending]}. $\square$

Theorem (Grothendieck duality), stated without proof here — full proof in Hartshorne Residues and Duality Ch. VI ^{[source pending]}. The proof constructs the dualising complex ${\omega }_X^{\bullet }$ in stages — first for projective space, then for any projective scheme over a field, then for a finite morphism of projective schemes, then for a proper morphism of noetherian schemes — and proves the duality formula by reducing each stage to the previous one via the projection formula and the adjunction $(R{f}_{\ast }⊣f^{!})$ for proper $f$. The combinatorial heart of the proof is the trace map ${\int }_f:R{f}_{\ast }{\omega }_X^{\bullet }\to {\omega }_Y^{\bullet }$ that produces the global duality pairing in the proper case.

Connections Master

• Higher direct images and base change 04.03.07. The underived theorem on cohomology and base change is the abelian-category shadow of the six-functor base-change isomorphism. The semicontinuity statement on the jump locus of $\dim H^q(X_s,{\mathcal{F}}_s)$ in the underived setting becomes, in the derived setting, the constructibility of $R{f}_{!}F$ for a constructible $F$, and the underived base change at flat points becomes the unconstrained derived base change for $R{f}_{!}$. The six-functor formalism is the natural derived-category home for the cohomology-and-base-change theorem.

• Serre duality 04.08.03. The classical Serre duality on a smooth projective variety $X$ over a field $k$, $H^p(X,F{)}^{\vee }\cong H^{n-p}(X,{\omega }_X\otimes F^{\vee })$, is the structure-map specialisation of the second six-functor adjunction $(R{f}_{!}⊣f^{!})$ applied to $f:X\to \mathrm{pt}$. The relative dualising complex ${\omega }_{X/\mathrm{pt}}^{\bullet }=f^{!}(k)$ is the canonical sheaf ${\omega }_X$ shifted by the dimension $n$ of $X$, and the duality pairing on global sections is the unit of the adjunction evaluated at ${\omega }_X^{\bullet }$.

• Derived functors $RF$ and $LF$ 04.03.12. The four functors $f^{\ast },f_{\ast },f_{!},f^{!}$ of the six-functor formalism are all instances of the general construction of total derived functors via injective, projective, flat, or acyclic resolutions on a derived category. The technical statements about adjunctions on derived categories — that adjoint pairs of underived functors lift to adjoint pairs of derived functors under appropriate exactness hypotheses — are the foundational input that makes the six-functor adjunctions $(L{f}^{\ast }⊣R{f}_{\ast })$ and $(R{f}_{!}⊣f^{!})$ exist as honest derived-functor adjunctions.

• Leray spectral sequence 04.03.15. The Leray spectral sequence $E_2^{p,q}=H^p(S,R^q{f}_{\ast }F)\Rightarrow H^{p+q}(X,F)$ for a continuous map $f:X\to S$ is the cohomology spectral sequence of the composition $R{\Gamma }_S\circ R{f}_{\ast }$, which is the structure-map specialisation of the first six-functor adjunction $(L{f}^{\ast }⊣R{f}_{\ast })$. The general six-functor adjunction packages the Leray spectral sequence as a corollary at the level of total derived functors.

• Derived category $D(\mathcal{A})$ 04.03.11. The six-functor formalism is a calculus on the derived category $D(X)$ of sheaves on $X$, and the universal property of the derived category as the localisation of the homotopy category at quasi-isomorphisms is what allows the four functors to be defined as functors between derived categories rather than between abelian categories. The triangulated structure on $D(X)$ is preserved by all six operations, and the long exact sequences attached to distinguished triangles in $D(X)$ are the technical input that makes the formalism compatible with short exact sequences of sheaves.

• Sheaf cohomology 04.03.01. The structure-map specialisation $R{f}_{\ast }F=R\Gamma (X,F)$ recovers ordinary sheaf cohomology as the six-functor pushforward to a point. The cohomology-and-base-change interaction between sheaf cohomology and the relative cohomology of a family is the foundational structural feature of the six-functor formalism on the topological side.

• Higher direct images and base change 04.03.07 (repeat). The closed/open decomposition of a sheaf into local-cohomology along a closed subscheme and pushforward from the complement is the standard application of the four functors $i_{\ast },i^{!},j_{\ast },j^{!}$ for a closed immersion $i$ and its open complement $j$. The distinguished triangle $i_{\ast }{i}^{!}F\to F\to R{j}_{\ast }{j}^{\ast }F\to i_{\ast }{i}^{!}F[1]$ records this decomposition and underlies the theory of perverse sheaves.

Historical & philosophical context Master

The six-functor formalism originated in Alexander Grothendieck's 1958 letter to Jean-Pierre Serre ^{[source pending]} dated 13 November 1958, where Grothendieck sketched the idea that every reasonable morphism of geometric spaces should carry a quadruple of derived functors $(f^{\ast },f_{\ast },f_{!},f^{!})$ with two adjunctions and a base-change isomorphism on Cartesian squares. The formal development took place in two phases. The étale phase was carried out in SGA 4 (1963–1964 seminar; Séminaire de Géométrie Algébrique du Bois-Marie 4, Théorie des topos et cohomologie étale des schémas, Springer LNM 269, 270, 305, 1972–1973) ^{[source pending]}, with Grothendieck, Michael Artin, and Jean-Louis Verdier as principal authors; SGA 4 Exposés XVII and XVIII contain the construction of $R{f}_{!}$ via Nagata compactification and the right adjoint $f^{!}$ via Brown representability, plus the proper-base-change theorem (XII) for étale cohomology of proper morphisms.

The coherent-sheaf phase was carried out by Robin Hartshorne in Residues and Duality (Springer LNM 20, 1966) ^{[source pending]}, based on notes from Grothendieck's seminar at Harvard in 1963–1964; Hartshorne's book is the canonical reference for the twisted inverse image $f^{!}$ on noetherian schemes and the dualising complex ${\omega }_X^{\bullet }$. The streamlined exposition appeared in Verdier's Catégories dérivées, état 0 (SGA 4½, Springer LNM 569, 1977) ^{[source pending]}, with Pierre Deligne's Cohomologie étale (SGA 4½, same volume) ^{[source pending]} providing the simplified account of the proper-base-change theorem.

The topological version of the six-functor formalism on locally compact Hausdorff spaces was developed by Birger Iversen in Cohomology of Sheaves (Springer Universitext 1986) ^{[source pending]} and by Masaki Kashiwara and Pierre Schapira in Sheaves on Manifolds (Springer Grundlehren 292, 1990) ^{[source pending]}, with the latter providing the comprehensive modern reference for the six operations on sheaves of $k$-modules on a manifold. The $\infty$-categorical lift was developed by Jacob Lurie in Higher Algebra (book draft, 2017) ^{[source pending]} §6.4 and in Spectral Algebraic Geometry (book draft) Ch. 3, where the six-functor formalism becomes a presheaf of symmetric monoidal stable $\infty$-categories with all higher coherences automatically captured. The motivic version was developed by Voevodsky, Cisinski-Déglise, Ayoub, and others, culminating in the Cisinski-Déglise monograph Triangulated Categories of Mixed Motives (Springer 2019). The six-functor formalism is the technical foundation of Deligne's proof of the Weil conjectures (Annals of Math. 1974, 1980), the Beilinson-Bernstein-Deligne decomposition theorem (Astérisque 100, 1982) ^{[source pending]}, Drinfeld-Laumon's geometric Langlands programme (1980s onward), Bhatt-Scholze's prismatic cohomology (2018), Clausen-Scholze's condensed mathematics (2019), and the Fargues-Scholze geometrisation of the local Langlands correspondence (2021).

Bibliography Master

@book{SGA4,
  editor    = {Artin, Michael and Grothendieck, Alexander and Verdier, Jean-Louis},
  title     = {S{\'e}minaire de G{\'e}om{\'e}trie Alg{\'e}brique du Bois-Marie 4: Th{\'e}orie des topos et cohomologie {\'e}tale des sch{\'e}mas},
  publisher = {Springer-Verlag},
  series    = {Lecture Notes in Mathematics},
  volume    = {269, 270, 305},
  year      = {1972--1973}
}

@book{SGA4.5,
  author    = {Deligne, Pierre and Verdier, Jean-Louis},
  title     = {Cohomologie {\'e}tale (SGA 4{$\frac{1}{2}$})},
  publisher = {Springer-Verlag},
  series    = {Lecture Notes in Mathematics},
  volume    = {569},
  year      = {1977}
}

@book{HartshorneRD,
  author    = {Hartshorne, Robin},
  title     = {Residues and Duality},
  publisher = {Springer-Verlag},
  series    = {Lecture Notes in Mathematics},
  volume    = {20},
  year      = {1966}
}

@book{KashiwaraSchapira,
  author    = {Kashiwara, Masaki and Schapira, Pierre},
  title     = {Sheaves on Manifolds},
  publisher = {Springer-Verlag},
  series    = {Grundlehren der mathematischen Wissenschaften},
  volume    = {292},
  year      = {1990}
}

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

@book{GelfandManin,
  author    = {Gelfand, Sergei I. and Manin, Yuri I.},
  title     = {Methods of Homological Algebra},
  publisher = {Springer-Verlag},
  edition   = {2},
  year      = {2003}
}

@book{LurieHA,
  author    = {Lurie, Jacob},
  title     = {Higher Algebra},
  note      = {Book draft, available at the author's website},
  year      = {2017}
}

@book{LurieSAG,
  author    = {Lurie, Jacob},
  title     = {Spectral Algebraic Geometry},
  note      = {Book draft, available at the author's website},
  year      = {2018}
}

@article{BBD,
  author    = {Beilinson, Alexander and Bernstein, Joseph and Deligne, Pierre},
  title     = {Faisceaux pervers},
  journal   = {Ast{\'e}risque},
  volume    = {100},
  year      = {1982},
  pages     = {1--171}
}

@article{Deligne1974,
  author    = {Deligne, Pierre},
  title     = {La conjecture de Weil. I},
  journal   = {Publications Math{\'e}matiques de l'IH{\'E}S},
  volume    = {43},
  year      = {1974},
  pages     = {273--307}
}

@article{Deligne1980,
  author    = {Deligne, Pierre},
  title     = {La conjecture de Weil. II},
  journal   = {Publications Math{\'e}matiques de l'IH{\'E}S},
  volume    = {52},
  year      = {1980},
  pages     = {137--252}
}

@incollection{GrothendieckLetter1958,
  author    = {Grothendieck, Alexander and Serre, Jean-Pierre},
  title     = {Correspondance Grothendieck-Serre},
  publisher = {Soci{\'e}t{\'e} Math{\'e}matique de France and American Mathematical Society},
  series    = {Documents Math{\'e}matiques},
  volume    = {2},
  year      = {2003},
  note      = {Letter dated 13 November 1958 sketching the duality formalism}
}

@book{CisinskiDeglise,
  author    = {Cisinski, Denis-Charles and D{\'e}glise, Fr{\'e}d{\'e}ric},
  title     = {Triangulated Categories of Mixed Motives},
  publisher = {Springer},
  series    = {Springer Monographs in Mathematics},
  year      = {2019}
}