19
$\begingroup$

I'm trying to prove that for two finitely generated $A$-modules $M,N$ ($A$ being any cmmutative ring), the tensor product $M\otimes_A N$ is zero iff $\operatorname{Ann}(M)+\operatorname{Ann}(N)=A$.

The if direction is of course easy- just show $1$ as a sum of two annihilating elements, $r\in Ann(M)$ and $s\in Ann(N)$, and for any $m\otimes n\in M\otimes_A N$ we have that $$m\otimes n=(r+s)(m\otimes n)=rm\otimes n+m\otimes sn=0$$

The only if directions is what got me baffled. By now I know that it suffices to show that $$M\otimes_A N=0\Rightarrow N=Ann(M)N\text{ or } M=Ann(N)M\tag{$**$}$$ since both modules are fin.gen, the claim will follow (either trivially, if one of the annihilators is zero, or by Nakayama's Lemma, if both are non-zero).

But I'm stuck on showing that, and I'm not even sure if it is always the case that $(**)$ holds... Does anybody have any idea? Maybe a hint in the case where it is true? (It could be that the ring $A$ should be noetherian, I'm not sure about that...)

In any case- I would very much appreciate if someone can suggest some intuitions on how to prove when a tensor product is non-zero, or equivalently, what can be entailed from a zero tensor product.

Thanks in advance

.................................................

ADDED: In response to @Dylan Moreland's question on how I intend on using Nakayama's Lemma:

Once we've seen that (for example) $N=Ann(M)N$, since $N$ is a finitely generated module (and after seeing that $Ann(M)\subsetneq A$) we have by NL that there exists some $\alpha\equiv 1\mod Ann(M)$ such that $\alpha N=0$. In paticular $\alpha\in Ann(N)$, and since $\alpha\equiv 1\mod Ann(M)$ we have some $\beta\in Ann(M)$ such that $\alpha+\beta=1$. This implies that $1\in Ann(N)+Ann(M)$ and so the only if direction is proved (at least I think this proof is sound, if someone sees a flaw, I'd be happy to hear about it)

  • 2
    $M \otimes N$ being zero means that every $A$-bilinear map emanating from $M \times N$ is identically zero. Does this help as a hint?2011-12-21
  • 0
    @Dactyl: Thanks, It might- any chance of fishing for a little more? I'm thinking of finding a surjective bilinear map $M\times N\to N/Ann(M)N$, am I in the right direction?2011-12-21
  • 1
    I asked a similar question, about what can we say about a zero tensor product, here: http://math.stackexchange.com/questions/81640/exact-sequence-induced-on-the-tensor-product2011-12-21
  • 0
    @Klaus: We can look at the tensor produce $A\beta(b)\otimes_AAm$. If my assertion is correct, then $1=s+r\in Ann(\beta(b))+Ann(m)$, and $(s+r)(b\otimes m)=sb\otimes m+0$. since we know that $s\beta(b)=\beta(sb)=0$ this imples that $sb\in\ker(\beta)=Im(\alpha)$ and $b\otimes m=\alpha(x)\otimes m$ for some $x\in A$2011-12-21
  • 2
    Dear kneidell, Try the contrapositive. If $Ann(M) + Ann(N)$ is a proper ideal, what interesting kind of ideal can you choose that contains it? Try using that ideal to help. (As a general rule, to show that $M\otimes N$ is non-zero, try to find a map to some quotient that is simpler to understand.) Regards,2011-12-21
  • 0
    @Matt E: Thank you!!! I think I got it now. You might want to write your comment as an answer so that I can accept :)2011-12-21
  • 0
    Dear kneidell, I'm glad it helped. I reposted my comment as an answer. Regards,2011-12-21
  • 0
    @kneidell That looks good to me! Very cool.2011-12-21
  • 0
    Nice generalization of the Nakayama lemma.2013-09-14
  • 0
    I don't understand the accepted answer.2016-02-20

3 Answers 3

12

Try the contrapositive. If $Ann(M) + Ann(N)$ is a proper ideal, what interesting kind of ideal can you choose that contains it? Try using that ideal to help.

Also, as a general rule, to show that $M\otimes N$ is non-zero, try to find a map to some quotient that is simpler to understand (and so simpler to show is non-zero).

  • 0
    How do we proceed? The natural candidate for an ideal containing $J=Ann(M)+Ann(N)$ is taking a maximal ideal $I$ containing $J$. Then do we claim that the bilinear map $M \times N \to M \otimes N / I (M \otimes N)$ given by $f(m,n)=m\otimes n + I M \otimes N$ is nonzero? And why?2016-02-15
  • 1
    What I did (about $\aleph_0$ years ago, when I posted this question), was to show that $M/IM$ and $N/IN$ are non-zero vector spaces over $A/I$ (you'll have to use Nakayama for this one), and that there's a surjective map $M\otimes_A N\to (M/IM)\otimes_(A/I) (N/I N)$. In particular, we get that the domain of the map cannot be zero.2016-02-22
7

Here is an alternative proof, using the facts

  1. $\mathrm{supp}(M \otimes N) = \mathrm{supp}(M) \cap \mathrm{supp}(N)$ (Hint: Nakayama and Linear algebra)
  2. $\mathrm{supp}(M) = V(\mathrm{Ann}(M))$

This easily implies $V(\mathrm{Ann}(M \otimes N)) = V(\mathrm{Ann}(M) + \mathrm{Ann}(N))$. Hence, $M \otimes N=0$ iff the set is empty iff $\mathrm{Ann}(M) + \mathrm{Ann}(N)=A$.

1

Just a summary, let $M,N$ fin. gen. $A$-modules. TFAE:

i) $M\otimes N=0$

ii) $Ann(M) + Ann(N)=A$.

iii)$Ann(M) N = N$

Proof: $i)\Leftrightarrow ii)$ Done above.

$ii)\Rightarrow iii)$ Multiply $ii)$ by $N$.

$iii) \Rightarrow i)$ $M\otimes N = M\otimes Ann(M) N =Ann(M) M\otimes N =0$.

In particular $(**)$ holds.

RH & eduard