Processing math: 100%

The Stacks project

10.118 Generic flatness

Basically this says that a finite type algebra over a domain becomes flat after inverting a single element of the domain. There are several versions of this result (in increasing order of strength).

Lemma 10.118.1. Let R \to S be a ring map. Let M be an S-module. Assume

  1. R is Noetherian,

  2. R is a domain,

  3. R \to S is of finite type, and

  4. M is a finite type S-module.

Then there exists a nonzero f \in R such that M_ f is a free R_ f-module.

Proof. Let K be the fraction field of R. Set S_ K = K \otimes _ R S. This is an algebra of finite type over K. We will argue by induction on d = \dim (S_ K) (which is finite for example by Noether normalization, see Section 10.115). Fix d \geq 0. Assume we know that the lemma holds in all cases where \dim (S_ K) < d.

Suppose given R \to S and M as in the lemma with \dim (S_ K) = d. By Lemma 10.62.1 there exists a filtration 0 \subset M_1 \subset M_2 \subset \ldots \subset M_ n = M so that M_ i/M_{i - 1} is isomorphic to S/\mathfrak q for some prime \mathfrak q of S. Note that \dim ((S/\mathfrak q)_ K) \leq \dim (S_ K). Also, note that an extension of free modules is free (see basic notion 50). Thus we may assume M = S and that S is a domain of finite type over R.

If R \to S has a nontrivial kernel, then take a nonzero f \in R in this kernel. In this case S_ f = 0 and the lemma holds. (This is really the case d = -1 and the start of the induction.) Hence we may assume that R \to S is a finite type extension of Noetherian domains.

Apply Lemma 10.115.7 and replace R by R_ f (with f as in the lemma) to get a factorization

R \subset R[y_1, \ldots , y_ d] \subset S

where the second extension is finite. Choose z_1, \ldots , z_ r \in S which form a basis for the fraction field of S over the fraction field of R[y_1, \ldots , y_ d]. This gives a short exact sequence

0 \to R[y_1, \ldots , y_ d]^{\oplus r} \xrightarrow {(z_1, \ldots , z_ r)} S \to N \to 0

By construction N is a finite R[y_1, \ldots , y_ d]-module whose support does not contain the generic point (0) of \mathop{\mathrm{Spec}}(R[y_1, \ldots , y_ d]). By Lemma 10.40.5 there exists a nonzero g \in R[y_1, \ldots , y_ d] such that g annihilates N, so we may view N as a finite module over S' = R[y_1, \ldots , y_ d]/(g). Since \dim (S'_ K) < d by induction there exists a nonzero f \in R such that N_ f is a free R_ f-module. Since (R[y_1, \ldots , y_ d])_ f \cong R_ f[y_1, \ldots , y_ d] is free also we conclude by the already mentioned fact that an extension of free modules is free. \square

Lemma 10.118.2.slogan Let R \to S be a ring map. Let M be an S-module. Assume

  1. R is a domain,

  2. R \to S is of finite presentation, and

  3. M is an S-module of finite presentation.

Then there exists a nonzero f \in R such that M_ f is a free R_ f-module.

Proof. Write S = R[x_1, \ldots , x_ n]/(g_1, \ldots , g_ m). For g \in R[x_1, \ldots , x_ n] denote \overline{g} its image in S. We may write M = S^{\oplus t}/\sum Sn_ i for some n_ i \in S^{\oplus t}. Write n_ i = (\overline{g}_{i1}, \ldots , \overline{g}_{it}) for some g_{ij} \in R[x_1, \ldots , x_ n]. Let R_0 \subset R be the subring generated by all the coefficients of all the elements g_ i, g_{ij} \in R[x_1, \ldots , x_ n]. Define S_0 = R_0[x_1, \ldots , x_ n]/(g_1, \ldots , g_ m). Define M_0 = S_0^{\oplus t}/\sum S_0n_ i. Then R_0 is a domain of finite type over \mathbf{Z} and hence Noetherian (see Lemma 10.31.1). Moreover via the injection R_0 \to R we have S \cong R \otimes _{R_0} S_0 and M \cong R \otimes _{R_0} M_0. Applying Lemma 10.118.1 we obtain a nonzero f \in R_0 such that (M_0)_ f is a free (R_0)_ f-module. Hence M_ f = R_ f \otimes _{(R_0)_ f} (M_0)_ f is a free R_ f-module. \square

Lemma 10.118.3. Let R \to S be a ring map. Let M be an S-module. Assume

  1. R is a domain,

  2. R \to S is of finite type, and

  3. M is a finite type S-module.

Then there exists a nonzero f \in R such that

  1. M_ f and S_ f are free as R_ f-modules, and

  2. S_ f is a finitely presented R_ f-algebra and M_ f is a finitely presented S_ f-module.

Proof. We first prove the lemma for S = R[x_1, \ldots , x_ n], and then we deduce the result in general.

Assume S = R[x_1, \ldots , x_ n]. Choose elements m_1, \ldots , m_ t which generate M. This gives a short exact sequence

0 \to N \to S^{\oplus t} \xrightarrow {(m_1, \ldots , m_ t)} M \to 0.

Denote K the fraction field of R. Denote S_ K = K \otimes _ R S = K[x_1, \ldots , x_ n], and similarly N_ K = K \otimes _ R N, M_ K = K \otimes _ R M. As R \to K is flat the sequence remains exact after tensoring with K. As S_ K = K[x_1, \ldots , x_ n] is a Noetherian ring (see Lemma 10.31.1) we can find finitely many elements n'_1, \ldots , n'_ s \in N_ K which generate it. Choose n_1, \ldots , n_ r \in N such that n'_ i = \sum a_{ij}n_ j for some a_{ij} \in K. Set

M' = S^{\oplus t}/\sum \nolimits _{i = 1, \ldots , r} Sn_ i

By construction M' is a finitely presented S-module, and there is a surjection M' \to M which induces an isomorphism M'_ K \cong M_ K. We may apply Lemma 10.118.2 to R \to S and M' and we find an f \in R such that M'_ f is a free R_ f-module. Thus M'_ f \to M_ f is a surjection of modules over the domain R_ f where the source is a free module and which becomes an isomorphism upon tensoring with K. Thus it is injective as M'_ f \subset M'_ K as it is free over the domain R_ f. Hence M'_ f \to M_ f is an isomorphism and the result is proved.

For the general case, choose a surjection R[x_1, \ldots , x_ n] \to S. Think of both S and M as finite modules over R[x_1, \ldots , x_ n]. By the special case proved above there exists a nonzero f \in R such that both S_ f and M_ f are free as R_ f-modules and finitely presented as R_ f[x_1, \ldots , x_ n]-modules. Clearly this implies that S_ f is a finitely presented R_ f-algebra and that M_ f is a finitely presented S_ f-module. \square

Let R \to S be a ring map. Let M be an S-module. Consider the following condition on an element f \in R:

10.118.3.1
\begin{equation} \label{algebra-equation-flat-and-finitely-presented} \left\{ \begin{matrix} S_ f & \text{is of finite presentation over }R_ f \\ M_ f & \text{is of finite presentation as }S_ f\text{-module} \\ S_ f, M_ f & \text{are free as }R_ f\text{-modules} \end{matrix} \right. \end{equation}

We define

10.118.3.2
\begin{equation} \label{algebra-equation-good-locus} U(R \to S, M) = \bigcup \nolimits _{f \in R\text{ with }(051U)} D(f) \end{equation}

which is an open subset of \mathop{\mathrm{Spec}}(R).

Lemma 10.118.4. Let R \to S be a ring map. Let 0 \to M_1 \to M_2 \to M_3 \to 0 be a short exact sequence of S-modules. Then

U(R \to S, M_1) \cap U(R \to S, M_3) \subset U(R \to S, M_2).

Proof. Let u \in U(R \to S, M_1) \cap U(R \to S, M_3). Choose f_1, f_3 \in R such that u \in D(f_1), u \in D(f_3) and such that (10.118.3.1) holds for f_1 and M_1 and for f_3 and M_3. Then set f = f_1f_3. Then u \in D(f) and (10.118.3.1) holds for f and both M_1 and M_3. An extension of free modules is free, and an extension of finitely presented modules is finitely presented (Lemma 10.5.3). Hence we see that (10.118.3.1) holds for f and M_2. Thus u \in U(R \to S, M_2) and we win. \square

Lemma 10.118.5. Let R \to S be a ring map. Let M be an S-module. Let f \in R. Using the identification \mathop{\mathrm{Spec}}(R_ f) = D(f) we have U(R_ f \to S_ f, M_ f) = D(f) \cap U(R \to S, M).

Proof. Suppose that u \in U(R_ f \to S_ f, M_ f). Then there exists an element g \in R_ f such that u \in D(g) and such that (10.118.3.1) holds for the pair ((R_ f)_ g \to (S_ f)_ g, (M_ f)_ g). Write g = a/f^ n for some a \in R. Set h = af. Then R_ h = (R_ f)_ g, S_ h = (S_ f)_ g, and M_ h = (M_ f)_ g. Moreover u \in D(h). Hence u \in U(R \to S, M). Conversely, suppose that u \in D(f) \cap U(R \to S, M). Then there exists an element g \in R such that u \in D(g) and such that (10.118.3.1) holds for the pair (R_ g \to S_ g, M_ g). Then it is clear that (10.118.3.1) also holds for the pair (R_{fg} \to S_{fg}, M_{fg}) = ((R_ f)_ g \to (S_ f)_ g, (M_ f)_ g). Hence u \in U(R_ f \to S_ f, M_ f) and we win. \square

Lemma 10.118.6. Let R \to S be a ring map. Let M be an S-module. Let U \subset \mathop{\mathrm{Spec}}(R) be a dense open. Assume there is a covering U = \bigcup _{i \in I} D(f_ i) of opens such that U(R_{f_ i} \to S_{f_ i}, M_{f_ i}) is dense in D(f_ i) for each i \in I. Then U(R \to S, M) is dense in \mathop{\mathrm{Spec}}(R).

Proof. In view of Lemma 10.118.5 this is a purely topological statement. Namely, by that lemma we see that U(R \to S, M) \cap D(f_ i) is dense in D(f_ i) for each i \in I. By Topology, Lemma 5.21.4 we see that U(R \to S, M) \cap U is dense in U. Since U is dense in \mathop{\mathrm{Spec}}(R) we conclude that U(R \to S, M) is dense in \mathop{\mathrm{Spec}}(R). \square

Lemma 10.118.7. Let R \to S be a ring map. Let M be an S-module. Assume

  1. R \to S is of finite type,

  2. M is a finite S-module, and

  3. R is reduced.

Then there exists a subset U \subset \mathop{\mathrm{Spec}}(R) such that

  1. U is open and dense in \mathop{\mathrm{Spec}}(R),

  2. for every u \in U there exists an f \in R such that u \in D(f) \subset U and such that we have

    1. M_ f and S_ f are free over R_ f,

    2. S_ f is a finitely presented R_ f-algebra, and

    3. M_ f is a finitely presented S_ f-module.

Proof. Note that the lemma is equivalent to the statement that the open U(R \to S, M), see Equation (10.118.3.2), is dense in \mathop{\mathrm{Spec}}(R). We first prove the lemma for S = R[x_1, \ldots , x_ n], and then we deduce the result in general.

Proof of the case S = R[x_1, \ldots , x_ n] and M any finite module over S. Note that in this case S_ f = R_ f[x_1, \ldots , x_ n] is free and of finite presentation over R_ f, so we do not have to worry about the conditions regarding S, only those that concern M. We will use induction on n.

There exists a finite filtration

0 \subset M_1 \subset M_2 \subset \ldots \subset M_ t = M

such that M_ i/M_{i - 1} \cong S/J_ i for some ideal J_ i \subset S, see Lemma 10.5.4. Since a finite intersection of dense opens is dense open, we see from Lemma 10.118.4 that it suffices to prove the lemma for each of the modules R/J_ i. Hence we may assume that M = S/J for some ideal J of S = R[x_1, \ldots , x_ n].

Let I \subset R be the ideal generated by the coefficients of elements of J. Let U_1 = \mathop{\mathrm{Spec}}(R) \setminus V(I) and let

U_2 = \mathop{\mathrm{Spec}}(R) \setminus \overline{U_1}.

Then it is clear that U = U_1 \cup U_2 is dense in \mathop{\mathrm{Spec}}(R). Let f \in R be an element such that either (a) D(f) \subset U_1 or (b) D(f) \subset U_2. If for any such f the lemma holds for the pair (R_ f \to R_ f[x_1, \ldots , x_ n], M_ f) then by Lemma 10.118.6 we see that U(R \to S, M) is dense in \mathop{\mathrm{Spec}}(R). Hence we may assume either (a) I = R, or (b) V(I) = \mathop{\mathrm{Spec}}(R).

In case (b) we actually have I = 0 as R is reduced! Hence J = 0 and M = S and the lemma holds in this case.

In case (a) we have to do a little bit more work. Note that every element of I is actually the coefficient of a monomial of an element of J, because the set of coefficients of elements of J forms an ideal (details omitted). Hence we find an element

g = \sum \nolimits _{K \in E} a_ K x^ K \in J

where E is a finite set of multi-indices K = (k_1, \ldots , k_ n) with at least one coefficient a_{K_0} a unit in R. Actually we can find one which has a coefficient equal to 1 as 1 \in I in case (a). Let m = \# \{ K \in E \mid a_ K \text{ is not a unit}\} . Note that 0 \leq m \leq \# E - 1. We will argue by induction on m.

The case m = 0. In this case all the coefficients a_ K, K \in E of g are units and E \not= \emptyset . If E = \{ K_0\} is a singleton and K_0 = (0, \ldots , 0), then g is a unit and J = S so the result holds for sure. (This happens in particular when n = 0 and it provides the base case of the induction on n.) If not E = \{ (0, \ldots , 0)\} , then at least one K is not equal to (0, \ldots , 0), i.e., g \not\in R. At this point we employ the usual trick of Noether normalization. Namely, we consider

G(y_1, \ldots , y_ n) = g(y_1 + y_ n^{e_1}, y_2 + y_ n^{e_2}, \ldots , y_{n - 1} + y_ n^{e_{n - 1}}, y_ n)

with 0 \ll e_{n -1} \ll e_{n - 2} \ll \ldots \ll e_1. By Lemma 10.115.2 it follows that G(y_1, \ldots , y_ n) as a polynomial in y_ n looks like

a_ K y_ n^{k_ n + \sum _{i = 1, \ldots , n - 1} e_ i k_ i} + \text{lower order terms in }y_ n

As a_ K is a unit we conclude that M = R[x_1, \ldots , x_ n]/J is finite over R[y_1, \ldots , y_{n - 1}]. Hence U(R \to R[x_1, \ldots , x_ n], M) = U(R \to R[y_1, \ldots , y_{n - 1}], M) and we win by induction on n.

The case m > 0. Pick a multi-index K \in E such that a_ K is not a unit. As before set U_1 = \mathop{\mathrm{Spec}}(R_{a_ K}) = \mathop{\mathrm{Spec}}(R) \setminus V(a_ K) and set

U_2 = \mathop{\mathrm{Spec}}(R) \setminus \overline{U_1}.

Then it is clear that U = U_1 \cup U_2 is dense in \mathop{\mathrm{Spec}}(R). Let f \in R be an element such that either (a) D(f) \subset U_1 or (b) D(f) \subset U_2. If for any such f the lemma holds for the pair (R_ f \to R_ f[x_1, \ldots , x_ n], M_ f) then by Lemma 10.118.6 we see that U(R \to S, M) is dense in \mathop{\mathrm{Spec}}(R). Hence we may assume either (a) a_ KR = R, or (b) V(a_ K) = \mathop{\mathrm{Spec}}(R). In case (a) the number m drops, as a_ K has turned into a unit. In case (b), since R is reduced, we conclude that a_ K = 0. Hence the set E decreases so the number m drops as well. In both cases we win by induction on m.

At this point we have proven the lemma in case S = R[x_1, \ldots , x_ n]. Assume that (R \to S, M) is an arbitrary pair satisfying the conditions of the lemma. Choose a surjection R[x_1, \ldots , x_ n] \to S. Observe that, with the notation introduced in (10.118.3.2), we have

U(R \to S, M) = U(R \to R[x_1, \ldots , x_ n], S) \cap U(R \to R[x_1, \ldots , x_ n], M)

Hence as we've just finished proving the right two opens are dense also the open on the left is dense. \square


Comments (0)


Your email address will not be published. Required fields are marked.

In your comment you can use Markdown and LaTeX style mathematics (enclose it like $\pi$). A preview option is available if you wish to see how it works out (just click on the eye in the toolbar).

Unfortunately JavaScript is disabled in your browser, so the comment preview function will not work.

All contributions are licensed under the GNU Free Documentation License.