8
$\begingroup$

Let $k$ be a field and $f,g \in k[x,y]$ be two non-constant polynomials in two variables. Is it true that the following conditions are equivalent:

  1. $f$ and $g$ are relatively prime in $k[x,y]$, in the sense that they have no non-constant common divisors, so if we have the identity $\alpha f + \beta g = 0$ in $k[x,y]$ for some polynomials $\alpha,\beta\in k[x,y]$, then $\alpha = -g \gamma$ and $\beta = f \gamma$ for some $\gamma \in k[x,y]$.

  2. The ideal $(f,g)$ generated by $f,g$ is of finite codimension in $k[x,y]$ over $k$, i.e. $\dim_{k} k[x,y] / (f,g) < \infty$.

The following simple arguments prove that 2) implies 1).

Suppose $f = a q$ and $g = b q$ for some $a,b,q\in k[x,y]$ and $q$ is non-constant, so 1) fail. Exchanging if necessary $x$ and $y$, one can assume that $q(x,y) \not= x^k$ for all $k$. Then neither of the following infinite and linearly independent over $k$ family of polynomials $\{x^i\}_{i\geq0}$ belongs to $(q) \supset (f,g)$. Whence $\dim_{k} k[x,y] / (f,g) \geq \dim_{k} k[x,y] / (q) = \infty$, and thus 2) fails as well.

Thus the question is whether 1) implies 2)?

Perhaps one should assume that $k$ has characteristic $0$.

$\endgroup$
9
  • 3
    $\begingroup$ This is also equivalent to: "$V(f)\cap V(g)$ is finite", where $V(f)$ is the set of zeros of $f$ in $\bar{k}^2$. $\endgroup$ – YCor Jun 27 at 23:43
  • 3
    $\begingroup$ If I remember correctly you can find a proof in Fulton, Algebraic Curves. Notice that k is algebraically closed wlog. $\endgroup$ – Martin Brandenburg Jun 28 at 0:54
  • 2
    $\begingroup$ Condition (1) means that the irreducible components of $V(f)$ and $V(g)$ are distinct. Now the intersection of two distinct irreducible closed subsets of dimension $1$ of $A^2_k$ consists of a finite set of closed points. This implies (2). $\endgroup$ – Damian Rössler Jun 28 at 12:41
  • 3
    $\begingroup$ the usual way to show this would be to compute the resultants en.m.wikipedia.org/wiki/Resultant $\endgroup$ – Dima Pasechnik Jun 28 at 14:51
  • 5
    $\begingroup$ @MarkSapir Sure, but there is an easy reduction (linear algebra) to the case k algebraically closed, which is presumably what Martin meant. $\endgroup$ – Will Sawin Jun 28 at 21:44
9
$\begingroup$

A standard undergraduate maths approach is via resultants.

I am not going to survey resultants here (but see below), I'll just say that an immediate consequence of 1) is that there exists a nonzero monic $u:=Res_y(f,g)\in k[x]\cap (f,g)$ and a nonzero monic $v:=Res_x(f,g)\in k[y]\cap (f,g)$. Therefore, any monomial $x^s y^t$ in $h\in R:=k[x,y]/(f,g)$ satisfies $s<\deg u$, $t<\deg v$, as you can reduce $x$-degree below $\deg u$ by replacing $x^{\deg u}$ with $u(x)-x^{\deg u}$, and analogously for $y$-degree (using $v$). Hence $R$ is finite-dimensional.


EDIT: Remarks, definitions and explanations.

Degree $d$ polynomials form $d+1$-dimensional vectorspace $k_d[t]\subset k[t]$ over $k$. The resultant $Res_t(p,q)$ of $p\in k_d[t]$, $q\in k_e[t]$ is the determinant $\det M$ of the linear map $M:k_d[t]\times k_e[t]\to k_{d+e}[t]$ defined by $M(w,z)=wq+zp$ (here $w\in k_d[t]$, $z\in k_e[t]$) - I am cheating here a bit with dimensions, as $\dim (k_d[t]\times k_e[t])=\dim(k_{d+e}[t])+1$, but we don't want this extra 1, so we assume w.l.o.g. that $z$ is monic, i.e. $z(t)=t^e+z_{e-1}t^{e-1}+\dots +z_0$. Assuming that $p,q$ have a common root $t^*$ in the algebraic closure of $k$, we see that $M$ cannot be 1-1 in this case, as $M$ would be divisible by $t-t^*$ for any $w,z$, i.e. $\det M=Res_t(p,q)=0$. Usually $M$ is written using monomial bases of $k_d[t]$ and $k_e[t]$, as Sylvester_matrix.

We see that it readily generalises to rings, e.g. if we replace $k$ with $k[x]$ then $\det M\in k[x]$, and we can think of $Res_y(f,g)\in k[y]$, vanishing at common roots of $f$ and $g$ in the algebraic closure of $k$.

$\endgroup$
7
  • 1
    $\begingroup$ It looks like you've typed $(x,y)$ for $(f,g)$ twice? $\endgroup$ – Mark Wildon Jun 28 at 22:14
  • $\begingroup$ I did, thanks for catching this. Fixed. $\endgroup$ – Dima Pasechnik Jun 28 at 22:37
  • $\begingroup$ The answer is nice but one can add two short phrases: that $u=Resultant(f,g,y)$ and $v=Resultant(f,g,x)$. The fact that $u,v\ne 0$ is immediate from 1) and the fact that $u.v\in (f,g)$ follows from a definition of resultant. I am not sure about undergraduate math, though. $\endgroup$ – Mark Sapir Jun 29 at 1:46
  • $\begingroup$ I might be spoilt by Oxford; here is what they teach in 3rd year undergraduate: courses.maths.ox.ac.uk/node/48825 $\endgroup$ – Dima Pasechnik Jun 29 at 9:00
  • 1
    $\begingroup$ For resultants, you can't assume that the polynomials in question are monic when the coefficients are only assumed to be elements of a ring. And it does not help to pass to the algebraic closures of the fraction fields $k(x)$ and $k(y)$. Instead, one needs to exploit the fact that $k[x,y]$ is a UFD, i.e. the first point in the question, to see that the resultant map $(u,v)\mapsto uf+vg$ in question is injective. $\endgroup$ – Z. M Jun 29 at 15:32
7
$\begingroup$

There's perhaps a more elementary answer than this, but here's what I have so far. Define $A := k[x,y]$ and $I := (f,g)$. I'll be using the following fact:

Fact: $A$ is a Cohen-Macaulay UFD.

To prove that $\dim_k A/I < \infty$, it suffices to show that the Krull dimension of $A/I$ (denoted simply by $\dim A/I$) is zero. For then $A/I$ is an Artinian $k$-algebra of finite type, a fortiori finite-dimensional over $k$.

Now recall in a Cohen-Macaulay ring $A$ that we have the inequality $\dim A \geq \dim A/I + \operatorname{ht}(I)$, where $\operatorname{ht}(I)$ is the height of $I$, defined as $$\operatorname{ht}(I) = \min \{\operatorname{ht}(\mathfrak{p}) | I \subseteq \mathfrak{p} \text{ is prime} \}.$$ Therefore, it is enough to show that there is no prime ideal $\mathfrak{p}$ containing $I$ of height $0$ or $1$.

Suppose $I$ is contained in some $\mathfrak{p}$ of height zero, i.e. a minimal prime. Since $A$ is a domain this means that $\mathfrak{p} = 0$, i.e. that $(f,g) = 0$ which is ridiculous. If $I \subset \mathfrak{q}$ where $\operatorname{ht}(\mathfrak{q}) = 1$, then since $A$ is a UFD, we must have $\mathfrak{q} = (h)$ for some polynomial $h$. Then we have $(f,g) \subset (h)$, contradicting the fact that $f,g$ are relatively prime.

Edit: As Bogdan Zavyalov pointed out to me, it's not necessarily true for an ideal $I$ in a Cohen-Macaulay ring $A$ that $\operatorname{ht}(I) + \dim A/I = \dim A$. Consider $A = \mathbf{Z}_p[X]$ and $I = (1 - pX)$. By the Krull Hauptidealsatz, $\operatorname{ht}(I) = 1$. On the other hand, $A/I = \mathbf{Q}_p$ which is zero-dimensional. Hence $\operatorname{ht}(I) + \dim A/I = 1$ which is strictly less than $2$ (the Krull dimension of $A$).

$\endgroup$
3
  • $\begingroup$ Just out of curiosity, what does "f.t." stand for? Is it finite type? $\endgroup$ – Malkoun Jun 28 at 13:43
  • $\begingroup$ @Malkoun Yes that's right. $\endgroup$ – David Benjamin Lim Jun 28 at 14:42
  • $\begingroup$ Is it true that in a 2-dim UFD, any coprime pair $(f,g)$ generates an ideal of finite colength? or is CM necessary? $\endgroup$ – YCor Jul 22 at 6:36

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.