Let $X$ be a scheme. Let $U$ be an open subset of $X$. It is clear that if $U$ contains all the generic points of $X$ (by which I mean the generic points of irreducible components of $X$) then $U$ is dense in $X$.

Is the converse of this statement true in general? That is, if $U$ is dense in $X$ does it contain all the generic points of $X$?

I know the converse is true for an affine scheme with finitely many minimal primes, hence for a scheme $X$ whose set of irreducible components is locally finite. So the statement holds for any locally Noetherian scheme.

$\textbf{Proof for the affine case when the ring has finitely many minimal primes:}$ Let $X = Spec(A)$. Let $p_1, \dots, p_n$ be the minimal primes of $A$. Let $U$ be a dense open of $Spec(A)$. Suppose $U = \{p_a : a \in A\}$. Then $$\overline{U} = \mathbb{V}(\cap_{a \in A} p_a) = Spec(A)$$

So, $\cap_{a \in A} p_a = nil(A)$. Let $p_{b_1}, \dots, p_{b_k}$ be the minimal primes of $A$ in $U$. Then $\cap_{a \in A} p_a = \cap p_{b_j}$. So for any minimal prime $p_i$, $nil(A) = \cap p_{b_j} \subset p_i$. By minimality and the fact that $p_i$ is prime, $p_i = p_{b_j}$ for some $j$ ($\textbf{here I need finiteness of the set of minimal primes}$), and so $p_i \in U$.


$\DeclareMathOperator{\Spec}{\operatorname{Spec}}$This is not true, even if $X = \Spec R$ is affine. Consider the ring

$$R := k[x_1, x_2, \ldots]/(x_1x_2, x_3x_4, \ldots) = k[x_i]/(x_ix_{i+1} \mid i \text{ odd})$$

This ring has infinitely many minimal primes (corresponding to binary sequences of even/odd choices for each pair of annihilating variables $\{x_i, x_{i+1}\}$). Set $p := (x_1, x_3, x_5, \ldots)$ be the minimal prime consisting of odd variables, and $U := \Spec(R) \setminus V(p)$, so that $p \not \in U$. Notice that for $f \in R$, $D(f) \subseteq U \iff V(f) \supseteq V(p) \iff f \in p$. Also, since $R$ is reduced, $D(f) = \emptyset$ iff $f = 0$.

To show $U$ is dense in $R$, it suffices to show that for all $g \ne 0 \in R$, $U \cap D(g) \ne \emptyset$, or equivalently, there exists $f \in R$ with $D(f) \subseteq U$ and $D(f) \cap D(g) = D(fg) \ne \emptyset$. In other words, for all $0 \ne g \in R$, there exists $f \in p$ with $fg \ne 0$. But such a choice of $f$ always exists for any $g$, by taking $f = x_k$, for some odd $k$ greater than the largest index appearing in $g$.

(To show this last claim, note that $x_kg = 0$ in $R$ involves only finitely many variables in $k[x_i]$, hence would hold in a sub-polynomial ring in finitely many variables, but in the finite case $k[x_1, \ldots, x_{n+1}]/(x_1x_2, \ldots, x_nx_{n+1}) = \Big(k[\ldots]/(\ldots)\Big)[x_k,x_{k+1}]/(x_kx_{k+1})$, $\text{ann}(x_k) = (x_{k+1})$).

  • $\begingroup$ This is great! I suspected that there was a counterexample. $\endgroup$ – Rankeya Oct 12 '14 at 22:12

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.