1
$\begingroup$

I wanted to find an integral domain in which we do not have the property that any two elements have a greatest common divisor (gcd).

Somewhere on this forum someone claimed that the following quotient ring of the polynomial ring in four variables over a field $k$ is such a ring: $A:=k[X,Y,Z,W]/(XW−YZ)$. I would appreciate some help in proving this. We denote the classes of $X,Y,Z,W$ in $A$ by $x,y,z,w$.

1) $A$ is an integral domain
It suffices to prove that $XW−YZ$ is irreducible in $k[X,Y,Z,W]$. But this holds with the Eisenstein criterion if we consider $XW−YZ$ as an element in $k[Y,Z,W][X]$.

2) The elements $xw$ and $xy$ have no gcd in $A$.
Obviously the idea is to use that (because of the relation $xw=yz$), both $x$ and $y$ are common divisors of $xw$ and $xy$. But I'm having a hard time proving rigorously that there cannot exist a gcd.

$\endgroup$
3
  • 2
    $\begingroup$ If $d=\gcd(xw,xy)$ then $d=xd'$ (since $x\mid d$) and therefore $d'=\gcd(w,y)$. But in this answer I showed that $x,y,z,w$ are irreducible, so $d'=1$. We get $d=x$. On the other side, $y\mid xw(=yz)$ and $y\mid xy$, so $y\mid x$, a contradiction. $\endgroup$
    – user26857
    Jun 22 '17 at 21:26
  • 2
    $\begingroup$ At OP: the argument sketched in the comment of @user26857 essentially applies a general form of Euclid's Lemma, viz $\ (y,w)=1,\ y\mid xw\,\Rightarrow\, y\mid x\ \ {\bf if}\ \ (xw,xy)\,$ exists, see this answer. As I explain there, failure of Euclid's Lemma yields such nonexistant gcds (this is a generic example of such failure). $\endgroup$ Jun 23 '17 at 20:34
  • $\begingroup$ Many thanks for the arguments above! I will now try to put all the pieces together and then post it here $\endgroup$
    – Layer Cake
    Jun 24 '17 at 14:04
1
$\begingroup$

Here is a proof I put together using the above arguments of @user26857.

First I state all the lemmas we will use in the proof. We use the notation $(a,b)$ for $\gcd(a,b)$.

Lemma $1$: Let $R$ be an integral domain and $a,b\in R$ two irreducible elements that aren't associates. Then $(a,b)=1$.
Euclid's Lemma: Let $R$ be an integral domain and $a,b,c\in R$. If $(ac,bc)$ exists, then $(a,b) = (ac,bc)/c$.
Claim $2$: $y\nmid x$ in our Ring $A$. In particular, $y$ and $x$ aren't associates.
Claim $3$: $x,y,z,w$ are irreducible in our ring $A$.

Proofs for Euclid's Lemma and Claim $3$ were given by @Bill Dubuque and @user26857 in the comments to my question. Lemma $1$ is trivial.
Proof of Claim $2$:
Suppose $y\mid x$, i.e. that there exists a polynomial $f$ such that $y\cdot f(x,y,z,w)=x$. Then $Y\cdot f(X,Y,Z,W)-X\in (XW-YZ)$. By plugging in $0$ for $Y,Z,W$ we get $0\cdot f(X,0,0,0)-X=0 $, a contradiction.

Finally, we prove that $d=(xw,xy)$ doesn't exist. If $d=(xw,xy)$ then $d=xd′$ (since $x \mid d$) and therefore by Euclid's Lemma $d′=(w,y)$. Now, $y$ and $w$ are both irreducible by Claim $3$ and not associates by Claim $2$ (just replace $x$ with $w$ in proof of Claim $2$), so by Lemma $1$ we get $d′=1$. This means $d=x$. On the other side, $y∣xw(=yz)$ and $y∣xy$, so $y∣x$, a contradiction to Claim $2$ and we are done.

$\endgroup$
1
  • $\begingroup$ 1) I'm not sure if that result is called "Euclid's Lemma". 2) Irreducibility of $x,y,z,w$ is given in a linked thread, not in the comment. $\endgroup$
    – user26857
    Jun 24 '17 at 22:20
-1
$\begingroup$

I would suggest you denote the classes of $X,Y,Z,W$ by $x,y,z,w.$ Otherwise, it's difficult to know in which ring you perform computations.

I didn't check the details, but it seems that $A$ may be identified to a subring of $B=k(Y)[X,W]$ (namely $k[X,Y,XW/Y,W]$), by sending $Z$ to $XW/Y$. The latter ring is a UFD. If $xw$ and $xy$ have a gcd $d=P(x,y,z,w)$, then $D=P(X,Y,XW/Y,W)$ would be a gcd of $XW$ and $XY$ in $B$. Hence, $D$ would be of the form $F(Y)X$, and I'm sure you can derive a contradiction from here.

Otherwise, I can provide you another kind of example: take $A=\mathbb{Z}[i\sqrt{5}]$. This is an integral domain.

One can show that the common divisors of $9$ in $A$ are $\pm 1,\pm 3,\pm 9,\pm (2+i\sqrt{5}), \pm (2-i\sqrt{5})$.

(Hint: solve the equations $\vert z\vert^2=1,3$ or $9$ , for $z\in A.$)

I let you deduce from the latter that $9$ and $3(2+i\sqrt{5})$ have no gcd in $A.$

(HInt: find the possible common divisors among the list above, and check that none of them satisfy the definition of a gcd)

$\endgroup$
3
  • $\begingroup$ I will try to work this out now and then report my progress. $\endgroup$
    – Layer Cake
    Jun 21 '17 at 13:41
  • $\begingroup$ Ok. I'll try ot have a closer look to your example. If I find a simpler method, I'll get back to you. $\endgroup$
    – GreginGre
    Jun 21 '17 at 13:48
  • 2
    $\begingroup$ 1) The OP didn't ask for an alternative example. 2) I have serious doubts that your argument is correct. Basically you say that if $A\subset B$ is a ring extension, and $B$ a UFD then $\gcd(a_1,a_2)$ in $A$ is also a $\gcd$ in $B$. One can easily find counterexamples to this claim. $\endgroup$
    – user26857
    Jun 22 '17 at 22:14

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.