94
$\begingroup$

For a commutative ring $R$ we clearly have $\dim(R[T]) \geq \dim(R)+1$. If $R$ is noetherian, we have equality. Every proof I'm aware of uses quite a bit of commutative algebra and non-trivial theorems such as Krull's intersection theorem.

Recently T. Coquand and H. Lombardi have found a surprisingly elementary "almost" first-order characterization of the Krull dimension, see here:

For $n \in \mathbb{N}$ we have $\dim(R) \leq n$ if and only if for all $x_0,\dotsc,x_n \in R$ there are $a_0,\dotsc,a_n \in R$ and $m_0,\dotsc,m_n \in \mathbb{N}$ such that $$x_0^{m_0} (\dotsc ( x_n^{m_n} (1+a_n x_n)+\dotsc)+a_0 x_0)=0.$$

This relies on the observation that $\dim(R) \leq n$ if and only if $\forall x \in R : \dim(R_{\{x\}}) \leq n-1$, where $R_{\{x\}}$ denotes the localization of $R$ at the subset $x^{\mathbb{N}} (1+xR) \subseteq R$.

A consequence of this is a new short proof of $\dim\bigl(K[x_1,\dotsc,x_n]\bigr)=n$, where $K$ is a field. Using Noether normalization and the fact that integral extensions don't change the dimension, it follows that $\dim(R \otimes_K S)=\dim(R)+\dim(S)$ if $R,S$ are finitely generated $K$-algebras; in particular $\dim(R[T])=\dim(R)+1$. This could be useful for introductory courses on algebraic geometry which don't want to waste too much time with dimension theory.

Can we use the elementary characterization of the Krull dimension by T. Coquand and H. Lombardi to give a new short proof of $\dim(R[T])=\dim(R)+1$ for noetherian commutative rings $R$?

$\endgroup$
  • 5
    $\begingroup$ Note: I've asked this on stackexchange math.stackexchange.com/questions/358423 and was encouraged to ask this here. $\endgroup$ – Martin Brandenburg Jun 21 '14 at 8:11
  • 4
    $\begingroup$ I would like to know more generally what happens if you take the Coquand-Lombardi characterization as a definition and try to develop the basics of dimension theory from there. $\endgroup$ – Neil Strickland Jun 26 '14 at 14:48
  • 5
    $\begingroup$ @NeilStrickland: The following example can be found in Hutchins, Examples of Commutative Rings, Example 27: Let $k$ be a field and let $R = k(y)[[x]] \times_{k(y)} k$ the ring of those power series in $x$ with coefficients in $k(y)$ resp. $k$ for the constant term. Then $\dim(R)=1$ and $\dim(R[T])=3$. I've also read that actually every number between $\dim(R)+1$ and $2 \dim(R)+1$ may appear as $\dim(R[T])$. $\endgroup$ – Martin Brandenburg Aug 4 '14 at 14:13
  • 6
    $\begingroup$ @Neil: What Martin read (every number between the bound appers) you can also read, namely in A. Seidenberg, On the dimension theory of rings (II), Pacific J. Math. 4 (1954), 603-614. $\endgroup$ – Fred Rohrer Aug 25 '14 at 21:21
  • 6
    $\begingroup$ The elementary characterization is also very useful in a topos-internal context. For example, a scheme $X$ is of dimension $\leq n$ if and only if, from the internal perspective of the little Zariski topos $\mathrm{Sh}(X)$, the (then plain old) ring $\mathcal{O}_X$ is of Krull dimension $\leq n$. See Proposition 3.13 of these sketchy notes. $\endgroup$ – Ingo Blechschmidt Feb 6 '15 at 15:51
16
$\begingroup$

There are two kinds of primes $\mathfrak q \subset R[T]$. The first possibility is that $\mathfrak q = \mathfrak p[T]$ with $\mathfrak p = R \cap \mathfrak q$. Let us call such a prime "small". The second possibility is that the inclusion $\mathfrak p[T] \subset \mathfrak q$ is strict. Let us call such a prime "big".

Suppose we have a sequence of primes $\mathfrak q_0 \subset \mathfrak q_1 \subset \ldots \subset q_r$ in $R[T]$. Then we get a sequence of big/small. If the sequence has only one switch from small to big, then of course $r \leq \dim(R) + 1$. The problem comes from a sequence with multiple switches. But thinking about it for a moment we see that it suffices to prove the following.

Scholium: If $\mathfrak q_0 \subset \mathfrak q_1$ in $R[T]$ lies over $\mathfrak p_0 \subset \mathfrak p_1$ in $R$ and if $\mathfrak q_0$ is big and $\mathfrak q_1$ is small, then there is a prime strictly in between $\mathfrak p_0$ and $\mathfrak p_1$.

To prove this we argue by contradiction and assume there is no prime strictly in between. Observe that in any case $\mathfrak p_0 \not = \mathfrak p_1$ by our definition of big and small. After replacing $R$ by $(R/\mathfrak p_0)_{\mathfrak p_1}$ we reach the situation where $R$ is a local Noetherian domain of dimension $1$. Then $\mathfrak p_0 = (0)$ and $\mathfrak p_1 = \mathfrak m$ is the maximal ideal.

Translating we have to derive a contradiction from the following: we have a nonzero prime $\mathfrak q \subset \mathfrak m[T]$ with $\mathfrak q \not = \mathfrak m[T]$.

Let $K$ be the fraction field of $R$. Let $\mathfrak q_K \subset K[T]$ be the ideal generated by $\mathfrak q$ in $K[T]$. Then $\mathfrak q = \mathfrak q_K \cap R[T]$.

For every $n \geq 0$ let $R[T]_{\leq n}$ be the polynomials of degree $\leq n$. Let $M_n = \mathfrak q \cap R[T]_{\leq n}$ and $Q_n = R[T]_{\leq n}/M_n$ so that we have a short exact sequence $$ 0 \to M_n \to R[T]_{\leq n} \to Q_n \to 0 $$ Now observe that $Q_n$ is a finite $R$-module, is torsion free, and has rank bounded independently of $n$. Namely, over $K$ we know that $\mathfrak q_K$ is generated by a polynomial of degree $d$ and we see that $Q_n \otimes_R K$ has dimension over $K$ at most $d$.

Pick $a \in \mathfrak m$ nonzero. Then (1) $R/aR$ has finite length $c$, (2) for any finite torsion free module $Q$ of rank $r$ the length of $Q/aQ$ is $rc$, and (3) a module $Q$ with length $Q/aQ$ bounded by $rc$ is generated by $\leq rc$ elements. [Hints for elementary proofs: To prove (1) you show for any $b \in \mathfrak m$ some power of $b$ is in $aR$ otherwise $R/aR$ would have a second prime. To prove (2) you choose $R^{\oplus r} \subset Q$ and you use the snake lemma for multiplication by a on the corresponding ses. To prove (3) use Nakayama and that a finite length module is generated by at most its length number of elements.]

Take $n > dc$ where $d$ is the upper bound for the ranks of all $Q_n$ found above. Then we conclude that there exists an element in $M_n$ which is not in $\mathfrak m(R[T]_{\leq n})$ because we have seen above that $Q_n$ can be generated by $\leq dc$ elements. Small standard argument omitted.

This is the desired contradiction because we assumed $\mathfrak q \subset \mathfrak m[T]$. QED

This answer shows that with usual commutative algebra there is a very short proof. Enjoy!

$\endgroup$
  • 3
    $\begingroup$ Where do you use the characterization of the Krull dimension by T. Coquand and H. Lombardi? $\endgroup$ – Martin Brandenburg Feb 29 '16 at 16:15
  • 2
    $\begingroup$ @MartinBrandenburg: exactly! $\endgroup$ – darx Feb 29 '16 at 18:20
  • 3
    $\begingroup$ If you don't use it, this is not an answer to my question. $\endgroup$ – Martin Brandenburg Mar 1 '16 at 9:54
  • 2
    $\begingroup$ @MartinBrandenburg: Of course I agree with you. I could "use" the characterization by just quoting it. But that would make this proof longer! Anyway, my purpose with this answer was twofold: by spelling out what the usual commutative algebra proof amounts to we see that it is quite easy/short and having this here may help the next visitor to do what you asked for. PS: The statement about dimensions of polynomial algebras over fields can be proven very quickly in a course if you do not want to explain more commutative algebra to the students along the way. Best wishes, Darx $\endgroup$ – darx Mar 1 '16 at 15:59
  • 10
    $\begingroup$ You did answer the question "what is a short proof of the dimension formula for polynomial rings", but my question is "what is a proof of the dimension formula using the characterization of the Krull dimension by Coquand and Lombardi". It doesn't matter if you think that this characterization is necessary or not. $\endgroup$ – Martin Brandenburg Mar 1 '16 at 17:08

protected by Community Feb 2 '16 at 12:33

Thank you for your interest in this question. Because it has attracted low-quality or spam answers that had to be removed, posting an answer now requires 10 reputation on this site (the association bonus does not count).

Would you like to answer one of these unanswered questions instead?

Not the answer you're looking for? Browse other questions tagged or ask your own question.