6

We know that

Given a prime ideal $P$ of a commutative ring, there is a one-to-one correspondence between $\lbrace\text{prime ideals }Q\subset P\rbrace$ and $\lbrace\text{prime ideals of } S^{-1}R \rbrace$ by $I \longrightarrow S^{-1}I$, where $S^{-1}R$ is the localization at $P$ and $S^{-1}I=\lbrace \frac{a}{s}|a\in I, s\in R-P \rbrace$;

If $I$ is principal, say $I=(a)$ , then $S^{-1}I=(\frac{a}{1})$.

For $R=\mathbb{Z}/6\mathbb{Z}$ and $S=R-2\mathbb{Z}/6\mathbb{Z}=\lbrace 1, 3 ,5 \rbrace$, $S^{-1}R=\lbrace \frac{0}{1},\frac{1}{1} \rbrace$, which is a finite field of order 2.

But combining these two facts above $(0)\longrightarrow (\frac{0}{1})$ and $(2)=2\mathbb{Z}/6\mathbb{Z} \longrightarrow (\frac{2}{1})=(\frac{0}{1})$ since $\frac{2}{1}=\frac{0}{1}$ and get a contradiction!

Could someone point out anything wrong?

gogosk8
  • 396

1 Answers1

6

Your problem, as we resolved in the comments, is that $(0)$ is not prime. In fact, after localization $(0)=(2)$. Let's generalize: we shall determine when two ideals of a certain class of rings (including all quotients $\mathbb Z/n\mathbb Z$ for $n\ne 0$) are equal after localization.

Given a multiplicatively closed set $S\subset R$, the set $\{r\in R:rs=0\text{ for some } s\in S\}$ is an ideal, which we will call $S^{\bot}$.

Our general theorem is,
Theorem. If $R$ is a commutative ring with unity such that every prime ideal is maximal, then $S^{-1}R\cong R/S^{\bot}$ for any multiplicatively closed set $S$.

In particular, if $I$ and $J$ are ideals of $R$, then $S^{-1}I=S^{-1}J$ if and only if $I+S^{\bot}=J+S^{\bot}$.

In the case $R=\mathbb Z/(n)$, we can calculate the generator of pullback of the ideal $S^{\bot}$ to $\mathbb Z$. We shall call it $d$. Firstly, $S^{\bot}=\sum_{s\in S}\operatorname{ann}(s)$, since $S^{\bot}$ is simply the union of those annihilators. Second, if $rs=0$ in $\mathbb Z/(n)$, and $s\ne 0$, then $r's'=kn$ and $s'\ne 0$ for representatives $r',s'$ in $\mathbb Z$. So, $r'=kn/s'\in (n/\gcd(n,s'))$. Moreover, the converse, that $rs=0$ when $r'\in (n/\gcd(n,s'))$ for a pair of representatives $r',s'$ is clearly true. This means $\operatorname{ann}(s)$ is the image of $(n/\gcd(n,s'))$ in $\mathbb Z/(n)$. Thus, $S^{\bot}$ is the image of $$\sum_{s\in S}(n/\gcd(n,s'))=(\gcd_{s\in S}\{n/\gcd(n,s')\}).\tag{1}$$ Finally, the number $\gcd(n,s')$ is uniquely determined by $s$, so the generator of the ideal in $(1)$ is uniquely determined by the set $S$.

Let $I$ and $J$ be ideals of $\mathbb Z/(n)$. Let $(a)$ and $(b)$ be their respective pullbacks in $\mathbb Z$. Then, $S^{-1}I=S^{-1}J$ if and only if $(a)+(d)=(b)+(d)$ if and only if $\gcd(a,d)=\gcd(b,d)$.


Proof (Theorem). Let $R$ be such a ring. Since the minimal prime ideals of a commutative ring consist of zero divisors, the set of all units of $R$ is precisely the set of its non zero-divisors.

First, consider the case $S^{\bot}=0$. If $s\in S$ were a nonunit, then it would be a member of some ideal of $R$, so it would be a zero-divisor. If $rs=0$, though, then $r\in S^{\bot}$ implying $r=0$ which is a contradiction. Therefore, $S$ consists of units in this case, so $R/S^{\bot}=R=S^{-1}R$.

Now, return to the general case. Set $A=R/S^{\bot}$. Then, the canonical homomorphism $R\to S^{-1}R$ factors as $$R\xrightarrow{\phi} A\to S^{-1}R.$$ Define $\bar S=\phi(S)$. If $f:R\to R'$ is a ring homomorphism such that $f(s)$ is invertible for all $s\in S$, then there exists $\rho: S^{-1}R\to R'$ such that $f$ is the same as the composite $R\to S^{-1}R\to R'$. This determines a homomorphism $g: A\to R'$ such that $g\circ\phi=f$ and $g(\bar s)$ is invertible when $\bar s\in\bar S$. Use the universal property of localization to determine a map $h: \bar S^{-1}A\to R'$, where $R\to A\to \bar S^{-1}A\to R'$ is $f$. Note that $h$ is uniquely determined by $f$, since it is determined uniquely by $g$, which in turn is determined uniquely by $f$. Therefore, $\bar S^{-1}A\cong S^{-1}R$.

Finally, let's calculate $\bar S^{\bot}$. In $R$, if $rs\in S^{\bot}$ where $r\in R$ and $s\in S$, then $rss'=0$ for some $s'\in S$. Since $S$ is multiplicatively closed, it follows that $r\in S^{\bot}$. Hence, if $\phi(r)\phi(s)\in\bar S^{\bot}$, then $rs\in S^{\bot}$, so $\phi(r)\in\phi(S^{\bot})=0$. Hence, $\bar S^{\bot}=0$. We already resolved this case above, $\bar S^{\bot}A=A$. Therefore, $S^{-1}R\cong A$ q.e.d.

  • It is not true that $\bigcup_{s \in S} \text{ann}(s) = \bigoplus_{s \in S} \text{ann}(s)$. For example if $s \in S$, then $s^2 \in S$, but $\text{ann}(s) \subseteq \text{ann}(s^2)$, so $\text{ann}(s)$ and $\text{ann}(s^2)$ need not intersect trivially. – zcn Jan 27 '14 at 02:29
  • @user115654 Thanks. Not sure why I wrote $\oplus$ there. – Karl Kroningfeld Jan 27 '14 at 02:33
  • 1
    What's $\frac{n}{(s, n)}$ and how is GCD defined between two such numbers? Is it $n+(n,s')$ for some $s':s=s'+(n)$? Is it short for $\frac{(n,1)}{(s, n)}$? Or are you talking about $GCD(n,s')$ for $n$ and $s$ coprime? – Loki Clock Jan 27 '14 at 03:36
  • 1
    @LokiClock Pick any representative $s'$ as you have, then $n/(s',n)$--where $(s',n)$ denotes the gcd of the two--is a well-defined integer, as you can check. That's what I mean by those fractions. Then take the GCD of those integers. Algebraically speaking, $n/(s,n)$ would best refer to the colon ideal $(0:s)\subset\mathbb Z/(n)$ and the gcd of them is the sum of these ideals – Karl Kroningfeld Jan 27 '14 at 03:51
  • 1
    @LokiClock At any rate, my edit should make all of this clear. Thank you for your feedback. – Karl Kroningfeld Jan 27 '14 at 04:48
  • Your theorem is correct, but I can't see in the proof where you've used that $\dim R=0$. (I think at some moment you have to use this, but where exactly you did it?) – user26857 Feb 23 '14 at 17:05
  • @user121097 I used it exactly once, I think. It is in the paragraph starting with "First, consider". The implication, nonunit implies zero-divisor, requires all minimal prime ideals to be maximal. – Karl Kroningfeld Feb 23 '14 at 22:09
  • @user121097 Btw, your comment seems to indicate you have seen the theorem before, do you have a reference? – Karl Kroningfeld Feb 23 '14 at 22:39
  • 1
    I've "invented" (long time ago!) such a result for finite commutative rings in order to find their localizations. Your result is sightly more general, but the proof is the same. – user26857 Feb 23 '14 at 23:02