Faithfully flat descent is a technique from algebraic geometry, allowing one to draw conclusions about objects on the target of a faithfully flat morphism. Such morphisms, that are flat and surjective, are common, one example coming from an open cover.
In practice, from an affine point of view, this technique allows one to prove some statement about a ring or scheme after faithfully flat base change.
"Vanilla" faithfully flat descent is generally false; instead, faithfully flat descent is valid under some finiteness conditions (e.g., quasi-compact or locally of finite presentation).
A faithfully flat descent is a special case of Beck's monadicity theorem.
A\toB
B
B
A
For example, given some elements
f1,...,fr
B=\prodi
-1 | |
A[f | |
i |
]
A
\operatorname{Spec}(B)=
r | |
cup | |
i=1 |
-1 | |
\operatorname{Spec}(A[f | |
i |
])
\operatorname{Spec}(A)
B
A
Mi
-1 | |
A[f | |
i |
]
Mi,Mj
-1 | |
\operatorname{Spec}(A[f | |
i |
,
-1 | |
f | |
j |
])
Let
A\toB
A
M
B
N=M ⊗ AB
A\toB
M\hookrightarrowM ⊗ AB
\varphi:N ⊗ B\overset{\sim}\toN ⊗ B
B ⊗
B ⊗ \simeqB ⊗ ,x ⊗ y\mapstoy ⊗ x
\varphi1=\varphi0\circ\varphi2
\varphii:N ⊗ B ⊗ \overset{\sim}\toN ⊗ B ⊗
\varphi0(n ⊗ b ⊗ c)=\rho1(b)\varphi(n ⊗ c)
\varphi1(n ⊗ b ⊗ c)=\rho2(b)\varphi(n ⊗ c)
\varphi2(n ⊗ b ⊗ c)=\varphi(n ⊗ b) ⊗ c
i(x)(y | |
\rho | |
0 |
⊗ … ⊗ yr)=y0 … yi-1 ⊗ x ⊗ yi … yr
\varphii:N ⊗ B ⊗ \overset{\sim}\toN ⊗ B ⊗
\varphi
M.
Now, the most basic form of faithfully flat descent says that the above construction can be reversed; i.e., given a
B
N
B ⊗
\varphi:N ⊗ B\overset{\sim}\toN ⊗ B
\varphi1=\varphi0\circ\varphi2
M=\{n\inN|\varphi(n ⊗ 1)=n ⊗ 1\}\subsetN
M ⊗ B=N
Here is the precise definition of descent datum. Given a ring homomorphism
A\toB
di:B ⊗ \toB ⊗
A\toB
d0
B ⊗ \simeqA ⊗ AB ⊗ \toB ⊗ AB ⊗ =B ⊗
d1
B ⊗ \simeqB ⊗ A ⊗ B ⊗ \toB ⊗
-
⊗ | |
di |
B ⊗
B ⊗
B ⊗
di
Now, given a
B
N
\varphi
M
d0-\varphi\circd1:N\toN
⊗ | |
d0 |
B ⊗
M ⊗ B\toN,x ⊗ a\mapstoxa
A\toB
\begin{array}{lccclcl} 0&\to&M ⊗ AB&\to& N ⊗ AB&\xrightarrow{d0-\varphi\circd1}&N
⊗ | |
d0 |
B ⊗ ⊗ AB\\ &&\downarrow&&\varphi\circd1\downarrow&& \downarrow\varphi
⊗ | |
d0,d1 |
B ⊗ \circd2\\ 0&\to&N&\to& N
⊗ | |
d0 |
B ⊗ &\xrightarrow{d0-d1}&N
⊗ | |
d0,d1 |
B ⊗ \\ \end{array}
The forgoing can be summarized simply as follows:
The Zariski descent refers simply to the fact that a quasi-coherent sheaf can be obtained by gluing those on a (Zariski-)open cover. It is a special case of a faithfully flat descent but is frequently used to reduce the descent problem to the affine case.
In details, let
l{Q}coh(X)
Fi
Ui\subsetX
X=cupUi
\varphiij:Fi
| | |
Ui\capUj |
\overset{\sim}\toFj
| | |
Ui\capUj |
\varphiii=\operatorname{id}
\varphiik=\varphijk\circ\varphiij
Ui\capUj\capUk
F
F| | |
Ui |
\simeqFi
F| | |
Uj |
\simeqFj
F| | |
Ui\capUj |
\simeqFi|
Ui\capUj |
\overset{\varphiij
In a fancy language, the Zariski descent states that, with respect to the Zariski topology,
l{Q}coh
l{C}
p:l{C}\to
l{Q}coh
(U,F)
p
(U,F)\mapstoU
There is a succinct statement for the major result in this area: (the prestack of quasi-coherent sheaves over a scheme S means that, for any S-scheme X, each X-point of the prestack is a quasi-coherent sheaf on X.)
The proof uses Zariski descent and the faithfully flat descent in the affine case.
Here "quasi-compact" cannot be eliminated.
Let F be a finite Galois field extension of a field k. Then, for each vector space V over F,
V ⊗ kF\simeq\prod\sigmaV,v ⊗ a\mapsto\sigma(a)v
F/k
An étale descent is a consequence of a faithfully descent.