Let $A$ be a commutative noetherian ring, and let $M$ be a finitely generated projective $A$-module. It is well known and easy to prove that $A$ is locally free in the sense that for every $p \in\operatorname{Spec} A$, the module $M_p$ is a free $A_p$-module.

Is it true that projectives are also locally free in the following (more geometric?) sense:

There are elements $f_1,\dots,f_n \in A$ such that $(f_1,\dots,f_n) = 1$, and such that $M_{f_i}$ is a free $A_{f_i}$-module for all $1\le i \le n$.

Is this true? if so, can you provide a reference or explain how to prove it?



Yes, this is true. See this Math Overflow question for a precise statement and a reference to its proof in Bourbaki's Commutative Algebra.

This result is also stated in my commutative algebra notes, but the proof is not unfortunately not yet written up there. I certainly hope that this will be remedied soon though, as I will be teaching a course out of these notes starting on Monday. When the proof gets written, I will update this answer with a page number.

Added: Here is something in the MO answer that I decided was worth a comment here. For finitely generated modules, this stronger version of local freeness is actually equivalent to projectivity, whereas the weaker "pointwise local freeness" is subtly weaker in general.

  • 4
    $\begingroup$ Thank you for the reference. For the intrested reader, I should mention that Bourbaki proves the following much stronger result: If $M$ is a finitely presented module and if for some prime $p$, the module $M_p$ is free, then there exists a neighborhood $D(f)$ of $p$ such that $M_f$ is free over $A_f$, so that "being free at a point" implies "being free at a neigborhood of a point". $\endgroup$ – the L Jan 8 '11 at 16:51
  • 4
    $\begingroup$ @anonymous: Dear anonymous, more generally, if $M,N$ are finitely presented modules and $M_p, N_p$ are isomorphic, then $M_f \simeq N_f$ for some $f \in A-p$. This is explained in more detail (and generality) in EGA IV-8, where it is proved that (in a certain sense) if $\{R_\alpha\}$ is an inductive system of rings, then the category of f.p. modules over $\varinjlim R_\alpha$ is the "colimit of the categories of modules over $R_\alpha$," which has many useful applications in reducing questions about arbitrary f.p. modules to questions about finitely generated modules over noetherian rings. $\endgroup$ – Akhil Mathew Jan 15 '11 at 18:40
  • 2
    $\begingroup$ (contd.) For instance, Grothendieck's generic flatness lemma (if $A$ noetherian integral domain, $B$ finite $A$-algebra, $M$ finite $B$-module, then there is $f \in A$ such that $M_f$ is flat over $A$) works for finitely presented modules because any such module must "descend" to a finitely presented module over a noetherian (e.g. finitely generated) subring. $\endgroup$ – Akhil Mathew Jan 15 '11 at 18:41

For future reference, I have written up a constructive and reasonably self-contained, if somewhat dense, proof (one page). The basic idea is to first verify that idempotent matrices over local rings are equivalent to diagonal matrices with entries $1$ and $0$, thus showing that finitely generated projective modules over local rings are free.

  • $\begingroup$ Well, I think there are simpler proofs for finitely generated projective modules over local rings are free. Never mind, +1. $\endgroup$ – user26857 May 8 '15 at 16:40
  • 1
    $\begingroup$ @user26857: I would be interested to hear them! Are they constructive in the technical sense that they avoid the use of the law of excluded middle? I only know of my write-up and the one in Mulvey's gem Intuitionistic algebra and representations of rings. $\endgroup$ – Ingo Blechschmidt May 8 '15 at 16:54
  • $\begingroup$ math.stackexchange.com/questions/147754/… $\endgroup$ – user26857 May 8 '15 at 17:15
  • 1
    $\begingroup$ @user26857: Thanks! The problem is that a proof beginning with "Consider a minimal system of generators" is not constructive, since constructively one cannot show that any inhabited subset of the natural numbers contains a minimal element. In some cases though, the problem can be avoided by a clever rewording; I didn't try that in this case hard enough yet. $\endgroup$ – Ingo Blechschmidt May 8 '15 at 18:03

This answer shows that if its stalk $M_\mathfrak{p}$ at $\mathfrak{p}$ is free, then there is an open neighbourhood $\mathfrak{p}\in D(f)$ on which its value $M_f$ is also free.

Proof: Reduce to the case that the natural map $M\to M_\mathfrak{p}$ is an injection by localising some $f_0\in A$. Now use the basis of $M_\mathfrak{p}$ to give a surjection $\alpha:A^n\to M$ whose localisation at $\mathfrak{p}$ is the isomorphism $A^n_\mathfrak{p}\to M_\mathfrak{p}$. Thus $\ker \alpha_\mathfrak{p}=0$, so $\ker \alpha_f=0$ for some $f\in A$. Thus $M_f\simeq A^n_f$ is free, completing the proof.

Thus the quasicompact space $\text{Spec}A$ admits a finite cover by $D(f_i)$'s, on each of which $M_{f_i}$ is free. $\text{Spec}A=\cup D(f_i)$ means precisely that $(f_1,...,f_n)=A$.


I have a somehow naïve idea, that $M_{p}=\mbox{colim}_{f\notin p}M_{f}$. Now choose a basis $x_{i}$ for $M_{p}$ and we may assume $x_{i}\in M$, then they are linear independent over $A$. Then we "consider the distance of $M_{f}$ from being free as $f$ goes to infinity", which is the exact sequence

$$0\rightarrow \bigoplus A_{f}x_{i} \rightarrow M_{f} \rightarrow M_{f}/\bigoplus A_{f}x_{i}\rightarrow 0$$

Taking colimit with $f$ and note that localization and taking colimit are exact,

$$\mbox{colim} M_{f}/\bigoplus A_{f}x_{i}=M_{p}/\bigoplus A_{p}x_{i}=0$$

Hence there exist some $f$, such that $M_{f}/\bigoplus A_{f}x_{i}=0$, provided $M$ is finitely generated. Now for every $p$ there is a distinguished open neighborhood, hence the ideal generated by these $f$'s must be trivial.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.