TheoremComplete

Faithfully Flat Descent

Faithfully flat descent is one of the most important theorems in modern algebraic geometry. It provides the theoretical foundation for gluing constructions along flat covers, which is essential for defining algebraic stacks, for proving that various moduli functors are representable, and for extending the scope of algebraic geometry beyond what the Zariski topology allows.


Statement of the Theorem

Theorem1.1Faithfully Flat Descent for Modules

Let φ:AB\varphi: A \to B be a faithfully flat ring homomorphism. Then the base change functor

AB:ModADD(B/A)- \otimes_A B: \operatorname{Mod}_A \to \operatorname{DD}(B/A)

from the category of AA-modules to the category of descent data for BB-modules relative to AA is an equivalence of categories.

Concretely, this means:

  1. (Fully faithful / Descent for morphisms) For any AA-modules M,NM, N, the natural map

HomA(M,N)HomBdesc(MAB,NAB)\operatorname{Hom}_A(M, N) \to \operatorname{Hom}_{B}^{\mathrm{desc}}(M \otimes_A B, N \otimes_A B)

is a bijection, where Homdesc\operatorname{Hom}^{\mathrm{desc}} denotes BB-module maps compatible with the descent data.

  1. (Essential surjectivity / Effectiveness) Every descent datum (MB,φ)(M_B, \varphi) (where MBM_B is a BB-module and φ:MBABBAMB\varphi: M_B \otimes_A B \xrightarrow{\sim} B \otimes_A M_B is an isomorphism satisfying the cocycle condition) is effective: there exists an AA-module MM with MABMBM \otimes_A B \cong M_B compatibly with φ\varphi.

Moreover, MM is recovered as the equalizer:

M=ker(MBBAMB)M = \ker\left(M_B \rightrightarrows B \otimes_A M_B\right)

where the two maps are m1mm \mapsto 1 \otimes m and mφ(m1)m \mapsto \varphi(m \otimes 1).

Theorem1.2Faithfully Flat Descent for Quasi-coherent Sheaves

Let f:YXf: Y \to X be a faithfully flat quasi-compact (fpqc) morphism of schemes. Then the pullback functor

f:QCoh(X)DD(Y/X,QCoh)f^*: \operatorname{QCoh}(X) \to \operatorname{DD}(Y/X, \operatorname{QCoh})

is an equivalence of categories, where DD(Y/X,QCoh)\operatorname{DD}(Y/X, \operatorname{QCoh}) is the category of quasi-coherent sheaves on YY equipped with descent data relative to ff.


Significance

RemarkWhy faithfully flat descent matters

Faithfully flat descent has far-reaching consequences:

  1. Foundation for stacks: The definition of an algebraic stack requires that descent be effective for certain types of objects (schemes, algebraic spaces) along smooth or etale covers. Faithfully flat descent for modules and algebras is the technical engine behind this.

  2. Representability results: To show a functor is representable (or representable by an algebraic space), one often constructs the representing object locally in some flat topology and then descends.

  3. Descent of properties: Many properties of morphisms (flat, smooth, etale, proper, finite, etc.) can be checked after a faithfully flat base change. The theorem ensures that these local checks determine global properties.

  4. Galois descent: For a Galois extension L/KL/K, faithfully flat descent specializes to classical Galois descent, relating KK-structures to LL-structures with Galois action.

  5. Torsors and cohomology: The classification of torsors via Cech cohomology relies on descent. A GG-torsor that is trivial on a cover YXY \to X is determined by descent data, which is a 1-cocycle.


The Amitsur Complex

Definition1.3Amitsur complex

For a ring map ABA \to B, the Amitsur complex is the cosimplicial ring

BBAB    BABAB  B \rightrightarrows B \otimes_A B \; \substack{\to \\ \to \\ \to} \; B \otimes_A B \otimes_A B \; \cdots

More precisely, let Bn=BAABB^{\otimes n} = B \otimes_A \cdots \otimes_A B (nn copies). The coface maps di:BnB(n+1)d^i: B^{\otimes n} \to B^{\otimes (n+1)} insert a 11 in the ii-th position:

di(b1bn)=b1bi1bi+1bn.d^i(b_1 \otimes \cdots \otimes b_n) = b_1 \otimes \cdots \otimes b_i \otimes 1 \otimes b_{i+1} \otimes \cdots \otimes b_n.

For a BB-module MM, the Amitsur complex of MM is

MMABMABABM \to M \otimes_A B \to M \otimes_A B \otimes_A B \to \cdots

with differentials d=(1)idid = \sum (-1)^i d^i. The descent datum is an isomorphism making this complex exact at the first few terms.

ExampleExactness of the Amitsur complex

When ABA \to B is faithfully flat, the Amitsur complex

0ABBABBABAB0 \to A \to B \to B \otimes_A B \to B \otimes_A B \otimes_A B \to \cdots

is exact. This is the key technical fact underlying faithfully flat descent.

For the proof: if ABA \to B has a section s:BAs: B \to A (e.g., AA×CA \to A \times C with projection), then the complex has a contracting homotopy hn:B(n+1)Bnh^n: B^{\otimes(n+1)} \to B^{\otimes n} given by hn(b0bn)=s(b0)b1bnh^n(b_0 \otimes \cdots \otimes b_n) = s(b_0) \cdot b_1 \otimes \cdots \otimes b_n. In the general faithfully flat case, we reduce to this case by the faithful flatness of ABA \to B (the complex is exact after tensoring with BB, and faithful flatness detects exactness).


Applications

ExampleDescent of ring-theoretic properties

Let ABA \to B be faithfully flat. Then many properties of AA can be detected from BB:

  • AA is Noetherian     \iff BB is Noetherian (and ABA \to B is of finite type).
  • AA is reduced     \iff BB is reduced.
  • AA is normal     \iff BB is normal.
  • AA is regular     \iff BB is regular (when both are Noetherian).

More precisely, these properties "descend" along faithfully flat morphisms: if BB has the property, then AA has the property. The converse ("ascent") also holds for most of these.

This is stronger than the Affine Communication Lemma (which handles only Zariski covers): faithfully flat descent works for arbitrary faithfully flat covers, including etale and fppf covers.

ExampleDescent for vector bundles

A vector bundle on XX is a locally free OX\mathcal{O}_X-module of finite rank. By faithfully flat descent:

If f:YXf: Y \to X is an fpqc cover and E\mathcal{E} is a locally free sheaf on YY with descent data, then E\mathcal{E} descends to a unique locally free sheaf on XX.

In particular, a rank-nn vector bundle on XX that is trivial on YY is classified by a 1-cocycle gGLn(O(Y×XY))g \in GL_n(\mathcal{O}(Y \times_X Y)) satisfying p23(g)p12(g)=p13(g)p_{23}^*(g) \cdot p_{12}^*(g) = p_{13}^*(g). This gives

{rank-n vector bundles on X, trivial on Y}/  =  Hˇ1({YX},GLn).\{\text{rank-}n \text{ vector bundles on } X, \text{ trivial on } Y\} / \cong \; = \; \check{H}^1(\{Y \to X\}, GL_n).

ExampleGalois descent for vector spaces

Let L/KL/K be a Galois extension with group GG. By faithfully flat descent (since KLK \to L is faithfully flat), the category of KK-vector spaces is equivalent to the category of LL-vector spaces with GG-semilinear action.

Concretely, given an LL-vector space VV with semilinear GG-action (i.e., σ(λv)=σ(λ)σ(v)\sigma(\lambda v) = \sigma(\lambda) \sigma(v) for σG,λL\sigma \in G, \lambda \in L), the KK-vector space W=VGW = V^G (the invariants) satisfies WKLVW \otimes_K L \cong V.

Example: Let L=CL = \mathbb{C}, K=RK = \mathbb{R}, G={1,σ}G = \{1, \sigma\} with σ\sigma = complex conjugation. A C\mathbb{C}-vector space VV with a semilinear involution σ\sigma descends to W=VσW = V^\sigma. If V=C2V = \mathbb{C}^2 and σ(z1,z2)=(zˉ2,zˉ1)\sigma(z_1, z_2) = (\bar{z}_2, \bar{z}_1), then W={(z,zˉ):zC}R2W = \{(z, \bar{z}) : z \in \mathbb{C}\} \cong \mathbb{R}^2.

ExampleWeil restriction via descent

Let L/KL/K be a finite field extension and XX a scheme over LL. The Weil restriction (or restriction of scalars) ResL/K(X)\operatorname{Res}_{L/K}(X) is a scheme over KK with the property:

ResL/K(X)(T)=X(T×KL)\operatorname{Res}_{L/K}(X)(T) = X(T \times_K L)

for any KK-scheme TT. When L/KL/K is separable, ResL/K(X)\operatorname{Res}_{L/K}(X) exists and can be constructed using etale descent.

For X=Gm,LX = \mathbb{G}_{m,L} (the multiplicative group over LL), ResL/K(Gm)(T)=(TKL)×\operatorname{Res}_{L/K}(\mathbb{G}_m)(T) = (T \otimes_K L)^\times. When L/KL/K is Galois with group GG, we have ResL/K(Gm)×KLgGGm,L\operatorname{Res}_{L/K}(\mathbb{G}_m) \times_K L \cong \prod_{g \in G} \mathbb{G}_{m,L}, and the descent datum encodes the Galois action permuting the factors.

ExampleTwisted forms and inner forms

Let X0X_0 be a scheme over a field kk. A twisted form (or kk-form) of X0X_0 is a scheme XX over kk such that X×kksepX0×kksepX \times_k k^{\text{sep}} \cong X_0 \times_k k^{\text{sep}} (they become isomorphic over a separable closure).

By etale descent, twisted forms of X0X_0 are classified by

Het1(Spec(k),Aut(X0))H^1_{\text{et}}(\operatorname{Spec}(k), \operatorname{Aut}(X_0))

where Aut(X0)\operatorname{Aut}(X_0) is the sheaf of automorphisms of X0X_0.

Example: Twisted forms of Pn1\mathbb{P}^{n-1} are Brauer-Severi varieties, classified by H1(k,PGLn)Br(k)[n]H^1(k, PGL_n) \cong \operatorname{Br}(k)[n] (the nn-torsion of the Brauer group). A Brauer-Severi variety VV is trivial iff it has a kk-rational point.

Example: Twisted forms of Mn(k)M_n(k) (the matrix algebra) are central simple algebras of degree nn. Their isomorphism classes form Br(k)[n]\operatorname{Br}(k)[n].

ExampleDescent of morphism properties

Let f:XYf: X \to Y be a morphism and g:YYg: Y' \to Y a faithfully flat quasi-compact morphism. Let f:XYf': X' \to Y' be the base change (X=X×YYX' = X \times_Y Y'). Then many properties of ff can be checked on ff':

f has property P    f has property Pf \text{ has property } \mathcal{P} \iff f' \text{ has property } \mathcal{P}

for P{flat, smooth, eˊtale, unramified, proper, finite, affine, quasi-finite, surjective, open immersion, closed immersion, isomorphism, ...}\mathcal{P} \in \{\text{flat, smooth, étale, unramified, proper, finite, affine, quasi-finite, surjective, open immersion, closed immersion, isomorphism, ...}\}.

This is immensely powerful: to check a morphism is smooth, it suffices to check it after any faithfully flat base change.


Descent for Schemes

Theorem1.4Descent for affine schemes

Let f:YXf: Y \to X be an fpqc morphism. The category of affine XX-schemes is equivalent (via base change ff^*) to the category of affine YY-schemes with descent data. That is, descent is effective for affine schemes along fpqc morphisms.

RemarkDescent for general schemes

For arbitrary schemes (not necessarily affine), descent along fpqc covers is generally NOT effective. However:

  1. Descent is effective for separated schemes along etale covers (Artin).
  2. Descent is effective for quasi-affine schemes along fpqc covers.
  3. Descent is effective for closed subschemes along fpqc covers.
  4. Descent is effective in the category of algebraic spaces along etale covers (by definition, essentially).

The failure of descent for general schemes is precisely what motivates the introduction of algebraic spaces and algebraic stacks: they are defined so that descent becomes effective in the appropriate sense.

ExampleDescent for closed subschemes

Let f:YXf: Y \to X be an fpqc cover. Then closed subschemes of XX correspond bijectively to closed subschemes of YY with descent data. The descent datum for a closed subscheme ZYZ \hookrightarrow Y is an equality p11(Z)=p21(Z)p_1^{-1}(Z) = p_2^{-1}(Z) as closed subschemes of Y×XYY \times_X Y.

In terms of ideal sheaves: descent for closed subschemes corresponds to descent for quasi-coherent ideal sheaves IOY\mathcal{I} \subseteq \mathcal{O}_Y, which is a special case of descent for quasi-coherent modules.


Connection to Stacks

RemarkFrom descent to stacks

The theory of descent naturally leads to the concept of a stack:

A prestack (fibered category) F\mathcal{F} over Sch/S\mathbf{Sch}/S satisfies descent for morphisms in a topology τ\tau if for every τ\tau-cover {UiU}\{U_i \to U\}, the natural functor F(U)DD({Ui}/U,F)\mathcal{F}(U) \to \operatorname{DD}(\{U_i\}/U, \mathcal{F}) is fully faithful.

A stack additionally satisfies effectiveness of descent: the functor F(U)DD({Ui}/U,F)\mathcal{F}(U) \to \operatorname{DD}(\{U_i\}/U, \mathcal{F}) is an equivalence of categories.

Faithfully flat descent tells us that:

  • The fibered category of quasi-coherent sheaves is a stack for the fpqc topology.
  • The fibered category of affine schemes is a stack for the fpqc topology.
  • The fibered category of all schemes is a stack for the Zariski topology but NOT for the etale or fpqc topology.

To remedy this, we enlarge our category: the fibered category of algebraic spaces is a stack for the etale topology, and algebraic stacks are stacks for the fppf (or smooth/etale) topology by construction.


Historical Notes

RemarkHistory

Faithfully flat descent was developed by Grothendieck in the late 1950s and early 1960s (SGA 1, FGA). The key ideas appear in:

  • FGA (Fondements de la Geometrie Algebrique), especially the Bourbaki seminars on descent (1959-1960).
  • SGA 1 (Revetements etales et groupe fondamental), Expose VIII on faithfully flat descent.
  • EGA IV, Section 2, on descent of properties of morphisms.

The formalism was later refined by Giraud (for non-abelian cohomology), Deligne and Mumford (for stacks), and Artin (for algebraic spaces). The modern treatment can be found in the Stacks Project.