2

While reading about the Hilbert's Nullstellensatz, the following question naturally posed itself:

Question. Let $k$ denote a field, not necessarily algebraically closed.

For which ideals $I$ is the quotient ring $k[x_1,\ldots,x_n]/I$ a finite dimensional vector space over $k$?

goblin GONE
  • 67,744
  • 2
    What do you think? Have you looked at $n=1$ or $2$ and seen what you can get from there? – Arthur Jan 05 '18 at 06:35
  • 1
    The answer in the "duplicate" is one correct equivalence but there are many other possible equivalences, as witnessed by the answers below, which give results of an entirely different flavour. One could argue that checking whether the $n$ vector spaces $I\cap k[x_i]$ are zero or not is algorithmically easier (via Gröbner bases) than checking whether $k[x_1,...,x_n]/I$ has Krull dimension zero, which is the answer in the preceding thread. Anyway the more criteria, the better. – Georges Elencwajg Jan 05 '18 at 16:36
  • @GeorgesElencwajg As you can notice I closed the question after two extended answers were posted. Anyway, all these are well known and can be found in Kreuzer and Robbiano, Computational Commutative Algebra 1, 2000, Section 3.7.A. – user26857 Jan 05 '18 at 18:50
  • @user26857. Everything on this site is "well known" to those who know the right references (I didn't know the one you mention and I'll certainly check it. Thanks for providing it). For original results I recommend the Annals of Mathematics and other research journals :-) – Georges Elencwajg Jan 05 '18 at 19:10

2 Answers2

3

One way to describe such ideals is that they are exactly the ideals $I$ such that $\sqrt{I}$ is a finite intersection of maximal ideals. Indeed, suppose $\sqrt{I}$ is a finite intersection of maximal ideals, say $M_1,\dots,M_k$. Let $A=k[x_1,\dots,x_n]$, let $B=A/I$, and let $C=A/\sqrt{I}$. By the Chinese remainder theorem, $C$ is isomorphic to $\prod_i A/M_i$. Each $A/M_i$ is a field and is finite-dimensional over $k$ (by Zariski's lemma), so $C$ is finite-dimensional over $k$. Moreover, $C$ is the quotient of $B$ by its nilradical $N$. For some $d$, $N^d=0$, and $N^r/N^{r+1}$ is a finitely generated $C$-module for each $r$ and hence a finite dimensional $k$-vector space. It follows that $B$ is a finite-dimensional $k$-vector space (since by reverse induction on $r$ starting from $r=d$, $N^r$ is finite-dimensional for all $r$, with the final case being $r=0$).

Conversely, suppose $I$ is such that $A/I$ is finite-dimensional. Then $C=A/\sqrt{I}$ is finite-dimensional as well, and in particular is artinian. So $C$ is a reduced artinian ring, and is therefore isomorphic to a finite product of fields $\prod L_i$. The kernels of the quotient maps $A\to L_i$ are then maximal ideals $M_i$ of $A$, and $\sqrt{I}$ is their intersection (the kernel of the map $A\to \prod L_i$).

Eric Wofsey
  • 330,363
3

The following are equivalent:

i) $k[x_1,\ldots,x_n]/I$ is a finite dimensional vector space over $k$
ii) For all $i=1\cdots n$ we have $I\cap k[x_i]\neq 0$
iii) For every field extension $k\subset k'$ the algebraic subset $V_I(k')=\{a\in k'^n\vert \forall P\in I, P(a)=0\}\subset k'^n$ is finite.
iv) For $\Omega$ an algebraic closure of $k$, the subset $V_I(\Omega)\subset \Omega^n$ is finite

Beware that it is not sufficient to suppose that $V_I(k)$ is finite to deduce i):
For example if $k=\mathbb R, n\gt 1$ and $I=\langle x^2_1+\cdots+x_n^2 \rangle $, the subset $V_I(\mathbb R)=\{(0,\cdots,0)\}$ has just one element but the $\mathbb R$-algebra $\mathbb R[x_1,\ldots,x_n]/\langle x^2_1+\cdots+x_n^2 \rangle $ is infinite dimensional over $\mathbb R$.

Edit: proof of equivalence

$\:i)\Rightarrow ii):$

Since the canonical linear map $\; k[x_i]/I\cap k[x_i]\hookrightarrow k[x_1,\ldots,x_n]/I$ is injective, we see that $\; k[x_i]/I\cap k[x_i]$ is finite dimensional, which implies $I\cap k[x_i]\neq 0$

$\:iv)\Rightarrow ii):$

Since the i-th projection $F_i:=pr_i(V_I(\Omega))\subset \Omega$ is a finite set, there is a polynomial $0\neq P_i(x_i)\in k[x_i]$ which vanishes on that set. [For example take the product of the minimal polynomials over $k$ of the elements of $F_i$]
A fortiori the polynomial $P_i(x_i)$ seen as a polynomial in $k[x_1,\ldots,x_n]$ vanishes on $V_I(\Omega)$.
But then the Nullstellensatz tells us that for some power $r$ we have $(P_i(x_i))^r\in I$ and we have found our polynomial $0\neq(P_i(x_i))^r\in I\cap k[x_i]$, as required by ii).

$ii)\Rightarrow i):$

Since $k[x_i]$ is a principal ring, $I\cap k[x_i]=\langle f(x_i)\rangle$ for a unique monic polynomial of degree $d_i$.
But then it is clear that every monomial $x_1^{s_1}\cdots x_n^{s_n}$ is congruent modulo $I$ to a product $Q_1(x_1)\cdots Q_n(x_n)$ where $Q_i(x_i)$ is a polynomial in $x_i$ with $\operatorname {degree}Q_i(x_i) \lt d_i$.
Hence the obvious $k$-linear map $\bigotimes_{i=1}^n k[x_i]/\langle f_i(x_i)\rangle \to k[x_1,\ldots,x_n]/I$ is surjective and $k[x_1,\ldots,x_n]/I$ is finite-dimensional over $k$, with dimension bound $\operatorname {dim}_k k[x_1,\ldots,x_n]/I \leq d_1\cdots d_n$.

$\:ii)\Rightarrow iii):$

Let $0\neq f(x_i)\in I\cap k[x_i]$ and let $Z_i\subset k'$ be the finite set of zeros of $f(x_i)$ in $k'$.
Then $V_I(k')\subset Z_1\times \cdots Z_n$ and thus $V_I(k')$ is also finite.

$\:iii)\Rightarrow iv):$
Trivial.