6
$\begingroup$

Suppose one has a filter (a collection of subsets closed under increasing the size of the set and under finite intersection) $F$ on a ring $R$. Say that $F$ is (ring) ideal-like if for every set $U \subset R$ in $F$, there exists a $U'\in F$ such that $U' + U' \subset U$, and there exists a $U'' \in F$ such that $U'' R \subset U$. (This corresponds to the filtered intersection of elements of $F$ giving rise to a pro-set which satisfies the ideal conditions.) Does this imply that $F$ is generated as a filter by actual ring ideals of $R$? In other words, for any $U \in F$, does there exist an ideal $I$ such that $I \in F$ and $I \subset U$?

If such a situation is possible, is it still possible if $R$ is Noetherian? If $R = \mathbb{Z}$?

$\endgroup$
1
  • $\begingroup$ Surely, you assume $R$ to be commutative? $\endgroup$ Commented Aug 23, 2023 at 9:19

1 Answer 1

7
$\begingroup$

This does not hold for all commutative rings, but it does hold for Noetherian rings and for valuation rings (assuming the convention that filters don't contain $\emptyset,$ or else $\mathcal{P}(R)$ is a trivial counterexample for any ring).

Suppose $R$ is a commutative ring, $F$ is an ideal-like filter on $R,$ and $U_0 \in F$ is such that there is no ideal $I \in F \cap \mathcal{P}(U_0).$ Recursively choose $U_n \in F$ such that $U_{n+1}+U_{n+1} \subset U_n$ and $U_{n+1}=U_{n+1}R.$ Since no $U_n$ is empty or an ideal, we immediately have $U_{n+1} \subsetneq U_n.$

For $n \ge 1,$ we recursively choose $u_n \in U_n$ for $n \ge 1$ as follows: first, let $u_1$ be any element of $U_1 \setminus U_2.$ Suppose $u_1, \ldots, u_n$ have been chosen. An inductive argument shows that $I_n:=\langle u_1, \ldots, u_n \rangle \subset U_0,$ so $I_n \not \in F.$ Let $u_{n+1}$ be any element of $U_{n+1} \setminus I_n.$ Then $\{I_n\}$ is a strictly increasing sequence of ideals, so $R$ is non-Noetherian. Furthermore, $\langle u_1 \rangle$ and $\langle u_2 \rangle$ are incomparable, so $R$ is not a valuation ring.

Now we'll construct a commutative ring $R$ with an ideal-like filter $F$ not generated by ring ideals. Let $R = C^{\infty}(S^1)$ be the ring of smooth real-valued functions on the circle. Every maximal ideal in $R$ is of the form $\mathfrak{m}_x=\{f \in R: f(x)=0\}$ for some $x \in S^1$ (see https://math.stackexchange.com/a/183213/210610). Let $A_n=\{f \in R: \lambda(f^{-1}(0))> 1- \frac{1}{n}\},$ and let $F$ be the filter on $R$ generated by the $A_n$'s, i.e. $$F=\bigcup_{n=1}^{\infty} \{X \subset R: A_n \subset X\}.$$

It's clear that every $X \in F$ is infinite (in fact, $|X|=|\mathbb{R}|$) and that for every finite $S \subset R \setminus \{0\},$ $R \setminus S \in F.$ We will show that there is no proper ideal in $F.$ Suppose $I \in F$ is a proper ideal on $R.$ There is $x$ such that $I \subset \mathfrak{m}_x \in F.$ But for every $n,$ there is a bump function in $A_n \setminus \mathfrak{m}_x,$ contradiction. Thus, $U=R \setminus \{1\} \in F$ does not contain any ideal in $F.$

Finally, we verify that $F$ is ideal-like. Fix $U \in F.$ Let $n$ be such that $A_n \subset U.$ Then $V=A_{2n}$ satisfies $V=VR$ and $V+V \subset A_n \subset U.$


Original answer:

This does not hold for all commutative rings, but if $R$ is a Dedekind domain, and we include the filter convention that $\emptyset \not \in F,$ then an ideal-like filter $F$ is generated by ring ideals.

Suppose $F$ is a counterexample. Let $U \in F$ be such that there is no ideal $I \in F \cap \mathcal{P}(U).$ Let $V, W \in F$ be such that $V+V+V \subset U$ and $WR \subset V.$ We may assume $W=WR.$ Since $W$ is nonempty and not an ideal, we can fix a nonzero non-unit $w \in W,$ and prime ideals $P_i$ such that $\langle w \rangle = \prod_{i=1}^n P_i.$

Let $A \subset \{1,\ldots, n\}$ be maximal such that $\prod_{i \in A} P_i \in F.$ Let $X \in F$ be such that $X=XR$ and $\sum_{i=1}^n X \subset W.$ For $j \in A,$ set $x_j=0,$ and otherwise let $x_j$ be an arbitrary element of $X \cap \prod_{i \in A} P_i \setminus \prod_{i \in A \cup \{j\}} P_i,$ which is nonempty since $X \cap \prod_{i \in A} P_i \in F.$ Let $x, y$ be such that $\langle x, y \rangle = \langle x_i: i \le n \rangle \subset W.$ Consider $$I:=\langle w, x, y \rangle \subset (W + W + W) \cap \prod_{i \in A} P_i \subset U \cap \prod_{i \in A} P_i.$$ Since $\prod_{i=1}^n P_i \subset I \subsetneq \prod_{i \in A} P_i \in F,$ there is $B \supsetneq A$ such that $I = \prod_{i \in B} P_i.$

Fix $j \in B \setminus A.$ Then $A \cup \{j\} \subset B$ and $x_j \in \langle x, y \rangle,$ so $x_j \in \prod_{i \in B} P_i \setminus \prod_{i \in A \cup \{j\}} P_i = \emptyset,$ contradiction.

$\endgroup$
7
  • $\begingroup$ Sorry, I do not understand why $X \cap \langle \prod_{i \in A} p_i \rangle \setminus \langle \prod_{i \in A \cup \{j\}} p_i \rangle $ is in $F$. But it seems that you only need it to be non-empty, and this is easy: otherwise $F\ni X \cap \langle \prod_{i \in A} p_i \rangle = X \cap \langle \prod_{i \in A \cup \{j\}} p_i \rangle \subset \langle \prod_{i \in A \cup \{j\}} p_i \rangle $. Nice proof! $\endgroup$ Commented Aug 23, 2023 at 10:30
  • 2
    $\begingroup$ You're right, I'll correct that part (I spend too much time working with ultrafilters, where the stronger claim I made would hold...). I used the fact that $R$ is a PID in defining $x$ and $B.$ $\endgroup$ Commented Aug 23, 2023 at 10:41
  • 1
    $\begingroup$ I've generalized the claim to Dedekind domains. I don't think this argument can be pushed much further, but it's plausible the claim holds for arbitrary Noetherian domains. $\endgroup$ Commented Aug 23, 2023 at 11:26
  • 1
    $\begingroup$ I added a counterexample for a general commutative ring. $\endgroup$ Commented Aug 24, 2023 at 0:26
  • 1
    $\begingroup$ @IlyaBogdanov Sorry to repeatedly ping you, but I've now proven the claim for Noetherian rings. $\endgroup$ Commented Aug 25, 2023 at 9:04

You must log in to answer this question.

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