5
$\begingroup$

I am only considering commutative rings with $1$. Dimension refers to Krull dimension.

In the paper "Products of commutative rings and zero-dimensionality", Gilmer and Heinzer give necessary and sufficient conditions for an arbitrary direct product of rings to be at least countably infinite-dimensional. As there are uncountably many prime ideals in an arbitrary product of rings, I wonder whether there is an uncountable chain of prime ideals in the infinite-dimensional case.

An analogous problem has been studied for power series rings. See "How to construct huge chains of prime ideals in power series rings" by Kang and Toan.

$\endgroup$
1

1 Answer 1

6
$\begingroup$

Yes, whenever the product is not zero-dimensional there is an uncountable chain of primes (in fact, a chain of cardinality $\mathfrak{c}$ and uncountable cofinality). Let $(A_\alpha)_{\alpha\in I}$ be an infinite collection of rings such that $\prod A_\alpha$ is not zero-dimensional. It follows from Gilmer and Heinzer's Theorem 3.4 that we can partition $I$ into infinitely many sets $I_n$ such that $\prod_{I_n} A_\alpha$ is infinite-dimensional for each $n$. It thus suffices to prove the following: Let $A_n$ be an infinite sequence of rings such that $\dim A_n\geq n$ for each $n$. Then there is an uncountable chain of primes in $A=\prod A_n$.

First, for each $n$ choose a chain of primes $P^n_0\subset P^n_1\subset\dots\subset P^n_n\subset A_n$ for each $n$ and fix a nonprincipal ultrafilter $U$ on $\mathbb{N}$. For any function $f:\mathbb{N}\to\mathbb{N}$ such that $f(n)\leq n$ for all $n$, let $P_f$ be the ideal of sequences $(y_n)\in A$ such that $\{n:y_n\in P^n_{f(n)}\}\in U$. Since $U$ is an ultrafilter, $P_f$ is prime. Write $f\sim g$ if $\{n:f(n)=g(n)\}\in U$ and $f<g$ if $\{n:f(n)<g(n)\}\in U$. Since $U$ is an ultrafilter, $<$ is a total ordering on the set of $\sim$-equivalence classes of functions. Furthermore, whenever $f<g$, $P_f\subset P_g$. The collection of all primes of the form $P_f$ is thus a chain, and is in bijection with the set of $\sim$-equivalence classes of functions $f$ such that $f(n)\leq n$. This latter set, also known as the ultraproduct $\prod_U [n]$ of the sets $[n]=\{0,1\dots,n\}$, is well-known to be countably saturated and have cardinality $\mathfrak{c}$.

$\endgroup$
7
  • $\begingroup$ I know that the prime ideals of an infinite product of fields are in bijection with the ultrafilters on the index set. Is there a more general theorem for an arbitrary direct product of rings? $\endgroup$ Commented Mar 24, 2015 at 13:25
  • $\begingroup$ Assume the continuum hypothesis is true. Given a answer to my previous comment, I hope that the Krull dimension of an arbitrary non-zero dimensional product of rings exists. At least for a countable index set, would the dimension equal $\mathfrak{c}$? $\endgroup$ Commented Mar 24, 2015 at 13:31
  • $\begingroup$ For your first question, there is always a surjection onto the set of ultrafilters of the index set, given by looking at elements which are $1$ or $0$ on every coordinate. This is a bijection iff each factor is local and the product is zero-dimensional (iff there is an integer $n$ such that every non-unit $x$ in a factor satisfies $x^n=0$). $\endgroup$ Commented Mar 24, 2015 at 14:57
  • $\begingroup$ For your second question, there are multiple ways you might define an infinite Krull dimension (do you just care about cardinality of chains, or do you want the length of well-ordered chains, or something else?). I don't know of an upper bound for the dimension of a countable product of finite-dimensional rings under any of these definitions, though. For instance, suppose $A_n$ is of the form $k[x_i]/(x_i^n)$, where $\{x_i\}$ is a very large infinite set of variables. It seems plausible to me that the product might have at least as many prime ideals as there are variables. $\endgroup$ Commented Mar 24, 2015 at 15:13
  • $\begingroup$ Actually, I believe I can say affirmatively that in the latter example, there are chains of primes in the product that are isomorphic to any ordering of the set of variables. Once you've passed to an ultraproduct and modded out the nilpotents, the ideal generated by any subset of the constant sequences corresponding to the variables will be prime. So a countable product of zero-dimensional rings can have arbitrarily large dimension, no matter how you define it. $\endgroup$ Commented Mar 24, 2015 at 16:27

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .