## 10.108 Pure ideals

The material in this section is discussed in many papers, see for example [Lazard], [Bkouche], and [DeMarco].

Definition 10.108.1. Let $R$ be a ring. We say that $I \subset R$ is pure if the quotient ring $R/I$ is flat over $R$.

Lemma 10.108.2. Let $R$ be a ring. Let $I \subset R$ be an ideal. The following are equivalent:

1. $I$ is pure,

2. for every ideal $J \subset R$ we have $J \cap I = IJ$,

3. for every finitely generated ideal $J \subset R$ we have $J \cap I = JI$,

4. for every $x \in R$ we have $(x) \cap I = xI$,

5. for every $x \in I$ we have $x = yx$ for some $y \in I$,

6. for every $x_1, \ldots , x_ n \in I$ there exists a $y \in I$ such that $x_ i = yx_ i$ for all $i = 1, \ldots , n$,

7. for every prime $\mathfrak p$ of $R$ we have $IR_{\mathfrak p} = 0$ or $IR_{\mathfrak p} = R_{\mathfrak p}$,

8. $\text{Supp}(I) = \mathop{\mathrm{Spec}}(R) \setminus V(I)$,

9. $I$ is the kernel of the map $R \to (1 + I)^{-1}R$,

10. $R/I \cong S^{-1}R$ as $R$-algebras for some multiplicative subset $S$ of $R$, and

11. $R/I \cong (1 + I)^{-1}R$ as $R$-algebras.

Proof. For any ideal $J$ of $R$ we have the short exact sequence $0 \to J \to R \to R/J \to 0$. Tensoring with $R/I$ we get an exact sequence $J \otimes _ R R/I \to R/I \to R/I + J \to 0$ and $J \otimes _ R R/I = J/JI$. Thus the equivalence of (1), (2), and (3) follows from Lemma 10.39.5. Moreover, these imply (4).

The implication (4) $\Rightarrow$ (5) is trivial. Assume (5) and let $x_1, \ldots , x_ n \in I$. Choose $y_ i \in I$ such that $x_ i = y_ ix_ i$. Let $y \in I$ be the element such that $1 - y = \prod _{i = 1, \ldots , n} (1 - y_ i)$. Then $x_ i = yx_ i$ for all $i = 1, \ldots , n$. Hence (6) holds, and it follows that (5) $\Leftrightarrow$ (6).

Assume (5). Let $x \in I$. Then $x = yx$ for some $y \in I$. Hence $x(1 - y) = 0$, which shows that $x$ maps to zero in $(1 + I)^{-1}R$. Of course the kernel of the map $R \to (1 + I)^{-1}R$ is always contained in $I$. Hence we see that (5) implies (9). Assume (9). Then for any $x \in I$ we see that $x(1 - y) = 0$ for some $y \in I$. In other words, $x = yx$. We conclude that (5) is equivalent to (9).

Assume (5). Let $\mathfrak p$ be a prime of $R$. If $\mathfrak p \not\in V(I)$, then $IR_{\mathfrak p} = R_{\mathfrak p}$. If $\mathfrak p \in V(I)$, in other words, if $I \subset \mathfrak p$, then $x \in I$ implies $x(1 - y) = 0$ for some $y \in I$, implies $x$ maps to zero in $R_{\mathfrak p}$, i.e., $IR_{\mathfrak p} = 0$. Thus we see that (7) holds.

Assume (7). Then $(R/I)_{\mathfrak p}$ is either $0$ or $R_{\mathfrak p}$ for any prime $\mathfrak p$ of $R$. Hence by Lemma 10.39.18 we see that (1) holds. At this point we see that all of (1) – (7) and (9) are equivalent.

As $IR_{\mathfrak p} = I_{\mathfrak p}$ we see that (7) implies (8). Finally, if (8) holds, then this means exactly that $I_{\mathfrak p}$ is the zero module if and only if $\mathfrak p \in V(I)$, which is clearly saying that (7) holds. Now (1) – (9) are equivalent.

Assume (1) – (9) hold. Then $R/I \subset (1 + I)^{-1}R$ by (9) and the map $R/I \to (1 + I)^{-1}R$ is also surjective by the description of localizations at primes afforded by (7). Hence (11) holds.

The implication (11) $\Rightarrow$ (10) is trivial. And (10) implies that (1) holds because a localization of $R$ is flat over $R$, see Lemma 10.39.18. $\square$

Lemma 10.108.3. Let $R$ be a ring. If $I, J \subset R$ are pure ideals, then $V(I) = V(J)$ implies $I = J$.

Proof. For example, by property (7) of Lemma 10.108.2 we see that $I = \mathop{\mathrm{Ker}}(R \to \prod _{\mathfrak p \in V(I)} R_{\mathfrak p})$ can be recovered from the closed subset associated to it. $\square$

Lemma 10.108.4. Let $R$ be a ring. The rule $I \mapsto V(I)$ determines a bijection

$\{ I \subset R \text{ pure}\} \leftrightarrow \{ Z \subset \mathop{\mathrm{Spec}}(R)\text{ closed and closed under generalizations}\}$

Proof. Let $I$ be a pure ideal. Then since $R \to R/I$ is flat, by going up generalizations lift along the map $\mathop{\mathrm{Spec}}(R/I) \to \mathop{\mathrm{Spec}}(R)$. Hence $V(I)$ is closed under generalizations. This shows that the map is well defined. By Lemma 10.108.3 the map is injective. Suppose that $Z \subset \mathop{\mathrm{Spec}}(R)$ is closed and closed under generalizations. Let $J \subset R$ be the radical ideal such that $Z = V(J)$. Let $I = \{ x \in R : x \in xJ\}$. Note that $I$ is an ideal: if $x, y \in I$ then there exist $f, g \in J$ such that $x = xf$ and $y = yg$. Then

$x + y = (x + y)(f + g - fg)$

Verification left to the reader. We claim that $I$ is pure and that $V(I) = V(J)$. If the claim is true then the map of the lemma is surjective and the lemma holds.

Note that $I \subset J$, so that $V(J) \subset V(I)$. Let $I \subset \mathfrak p$ be a prime. Consider the multiplicative subset $S = (R \setminus \mathfrak p)(1 + J)$. By definition of $I$ and $I \subset \mathfrak p$ we see that $0 \not\in S$. Hence we can find a prime $\mathfrak q$ of $R$ which is disjoint from $S$, see Lemmas 10.9.4 and 10.17.5. Hence $\mathfrak q \subset \mathfrak p$ and $\mathfrak q \cap (1 + J) = \emptyset$. This implies that $\mathfrak q + J$ is a proper ideal of $R$. Let $\mathfrak m$ be a maximal ideal containing $\mathfrak q + J$. Then we get $\mathfrak m \in V(J)$ and hence $\mathfrak q \in V(J) = Z$ as $Z$ was assumed to be closed under generalization. This in turn implies $\mathfrak p \in V(J)$ as $\mathfrak q \subset \mathfrak p$. Thus we see that $V(I) = V(J)$.

Finally, since $V(I) = V(J)$ (and $J$ radical) we see that $J = \sqrt{I}$. Pick $x \in I$, so that $x = xy$ for some $y \in J$ by definition. Then $x = xy = xy^2 = \ldots = xy^ n$. Since $y^ n \in I$ for some $n > 0$ we conclude that property (5) of Lemma 10.108.2 holds and we see that $I$ is indeed pure. $\square$

Lemma 10.108.5. Let $R$ be a ring. Let $I \subset R$ be an ideal. The following are equivalent

1. $I$ is pure and finitely generated,

2. $I$ is generated by an idempotent,

3. $I$ is pure and $V(I)$ is open, and

4. $R/I$ is a projective $R$-module.

Proof. If (1) holds, then $I = I \cap I = I^2$ by Lemma 10.108.2. Hence $I$ is generated by an idempotent by Lemma 10.21.5. Thus (1) $\Rightarrow$ (2). If (2) holds, then $I = (e)$ and $R = (1 - e) \oplus (e)$ as an $R$-module hence $R/I$ is flat and $I$ is pure and $V(I) = D(1 - e)$ is open. Thus (2) $\Rightarrow$ (1) $+$ (3). Finally, assume (3). Then $V(I)$ is open and closed, hence $V(I) = D(1 - e)$ for some idempotent $e$ of $R$, see Lemma 10.21.3. The ideal $J = (e)$ is a pure ideal such that $V(J) = V(I)$ hence $I = J$ by Lemma 10.108.3. In this way we see that (3) $\Rightarrow$ (2). By Lemma 10.78.2 we see that (4) is equivalent to the assertion that $I$ is pure and $R/I$ finitely presented. Moreover, $R/I$ is finitely presented if and only if $I$ is finitely generated, see Lemma 10.5.3. Hence (4) is equivalent to (1). $\square$

We can use the above to characterize those rings for which every finite flat module is finitely presented.

Lemma 10.108.6. Let $R$ be a ring. The following are equivalent:

1. every $Z \subset \mathop{\mathrm{Spec}}(R)$ which is closed and closed under generalizations is also open, and

2. any finite flat $R$-module is finite locally free.

Proof. If any finite flat $R$-module is finite locally free then the support of $R/I$ where $I$ is a pure ideal is open. Hence the implication (2) $\Rightarrow$ (1) follows from Lemma 10.108.3.

For the converse assume that $R$ satisfies (1). Let $M$ be a finite flat $R$-module. The support $Z = \text{Supp}(M)$ of $M$ is closed, see Lemma 10.40.5. On the other hand, if $\mathfrak p \subset \mathfrak p'$, then by Lemma 10.78.5 the module $M_{\mathfrak p'}$ is free, and $M_{\mathfrak p} = M_{\mathfrak p'} \otimes _{R_{\mathfrak p'}} R_{\mathfrak p}$ Hence $\mathfrak p' \in \text{Supp}(M) \Rightarrow \mathfrak p \in \text{Supp}(M)$, in other words, the support is closed under generalization. As $R$ satisfies (1) we see that the support of $M$ is open and closed. Suppose that $M$ is generated by $r$ elements $m_1, \ldots , m_ r$. The modules $\wedge ^ i(M)$, $i = 1, \ldots , r$ are finite flat $R$-modules also, because $\wedge ^ i(M)_{\mathfrak p} = \wedge ^ i(M_{\mathfrak p})$ is free over $R_{\mathfrak p}$. Note that $\text{Supp}(\wedge ^{i + 1}(M)) \subset \text{Supp}(\wedge ^ i(M))$. Thus we see that there exists a decomposition

$\mathop{\mathrm{Spec}}(R) = U_0 \amalg U_1 \amalg \ldots \amalg U_ r$

by open and closed subsets such that the support of $\wedge ^ i(M)$ is $U_ r \cup \ldots \cup U_ i$ for all $i = 0, \ldots , r$. Let $\mathfrak p$ be a prime of $R$, and say $\mathfrak p \in U_ i$. Note that $\wedge ^ i(M) \otimes _ R \kappa (\mathfrak p) = \wedge ^ i(M \otimes _ R \kappa (\mathfrak p))$. Hence, after possibly renumbering $m_1, \ldots , m_ r$ we may assume that $m_1, \ldots , m_ i$ generate $M \otimes _ R \kappa (\mathfrak p)$. By Nakayama's Lemma 10.20.1 we get a surjection

$R_ f^{\oplus i} \longrightarrow M_ f, \quad (a_1, \ldots , a_ i) \longmapsto \sum a_ im_ i$

for some $f \in R$, $f \not\in \mathfrak p$. We may also assume that $D(f) \subset U_ i$. This means that $\wedge ^ i(M_ f) = \wedge ^ i(M)_ f$ is a flat $R_ f$ module whose support is all of $\mathop{\mathrm{Spec}}(R_ f)$. By the above it is generated by a single element, namely $m_1 \wedge \ldots \wedge m_ i$. Hence $\wedge ^ i(M)_ f \cong R_ f/J$ for some pure ideal $J \subset R_ f$ with $V(J) = \mathop{\mathrm{Spec}}(R_ f)$. Clearly this means that $J = (0)$, see Lemma 10.108.3. Thus $m_1 \wedge \ldots \wedge m_ i$ is a basis for $\wedge ^ i(M_ f)$ and it follows that the displayed map is injective as well as surjective. This proves that $M$ is finite locally free as desired. $\square$

Comment #6651 by 七海辣辣米 on

In lemma 04PU, the first line, 'by going up' should be 'by going down'.

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).