5
$\begingroup$

If we have $Z\subset X$ a closed irreducible subscheme of an integral scheme $X$ (which you can take to have various further niceness properties if you want), one can take its generic point $\eta_Z$ and localize at it, to get a local ring with quotient field the quotient field of $Z$.

However, intuitively, I'd like to be able to "take a small neighborhood of $Z$" and not just of $\eta_Z$, i.e. I'd like a canonically constructed subscheme of $X$ whose "special fiber" is $Z$ and doesn't contain any points of $X$ besides the ones which specialize to points of $Z$. Certainly in general, I wouldn't expect this to be possible, so I don't expect a general functorial construction like usual localizations. But are there known cases when something like this exists?

I'm particularly interested in when $\pi:X\to S$ is a flat family or something similarly nice, and $Z$ is a section of $\pi$, so that what I'm asking for is basically like the "relative local ring of the point $Z$ over the base $S$". Then I'd also like to ask that the pullback of my "localization at $Z$" to any fiber of $\pi$ over a closed point of $S$ yields the localization of that fiber at the point where it intersects $Z$.

In this case, if $X$ is a trivial bundle over $S$, like if $X=S\times_k Y$ for some $Y$, then obviously this is possible; I'm wondering if it's possible more generally.

(I'd also appreciate an explicit counterexample to it being possible in general flat families, since I don't expect it to be possible in that generality, but find it difficult to prove it's impossible in any particular case.)

$\endgroup$
5
  • 1
    $\begingroup$ Just as a local ring is not by definition a subscheme, I don't think you could expect a subscheme (not finitary enough), but is the thing you're seeking something like inverting all functions whose zero locus is disjoint from $Z$? It seems like it would be something like the coproduct in the category of $X$-schemes of the local ring at $\eta_Z$ and the formal completion of $X$ at $Z$. When $Z$ is a section, I would be thinking about relative effective Cartier divisors instead of arbitrary functions. In any case, I myself am very much interested in seeing a good answer. $\endgroup$ Aug 22, 2022 at 2:47
  • $\begingroup$ I imagine that you would like your "small neighborhood of $Z$ in $X$" to be the same as the corresponding "small neighborhood of $Z$ in $U$" if $U$ is an open subset of $X$ that contains $Z$. But this seems to mean that you do want to cut out every closed subset of $X$ that is disjoint from $Z$. This does not coincide with what you say you expect in the case when $X=Z\times Y$. $\endgroup$ Aug 22, 2022 at 11:16
  • $\begingroup$ A naive thing to do is take the limit of all open neighbourhoods $U \subseteq X$ containing $Z$. This exists as a locally ringed space (and agrees with the same limit as ringed space), but is not always a scheme (e.g. take $Z$ to be a line in $\mathbf P^2$). When $X$ is affine and $Z$ is closed, this is just a usual localisation of rings, but this procedure does not even commute with open immersions $U \hookrightarrow X$ (let alone with taking fibres). $\endgroup$ Aug 22, 2022 at 12:33
  • $\begingroup$ @R.vanDobbendeBruyn A native question. Why in the affine case it is just a localization? The process your are describing is just removing all closed subsets disjoint from $Z$. These closed subsets can have higher codimensions. Is it true that any higher codimension closed subset that is disjoint from $Z$ can be covered by codimension $1$ divisors that are disjoint from $Z$? $\endgroup$
    – user127776
    Aug 23, 2022 at 18:31
  • 1
    $\begingroup$ @user127776 In the affine case, yes. If $I + J = R$ for ideals $I, J \subseteq R$ (i.e. $V(I) \cap V(J) = \varnothing$), then $I + (f) = R$ for some $f \in J$ (i.e. $V(I) \cap V(f) = \varnothing$). By the way, my construction agrees with Johan's, which is most easily seen if you already know it's also a limit of ringed spaces (as opposed to locally ringed spaces). $\endgroup$ Aug 23, 2022 at 22:37

1 Answer 1

7
$\begingroup$

Let $W \subset X$ denote the set of points specializing to $Z$ with the induced topology. Denote $\mathcal{O}_W$ the pullback of $\mathcal{O}_X$. Then $W = (W, \mathcal{O}_W)$ is a locally ringed space and we can ask:

Question: is $W$ a scheme?

Usually not: for example if $X = \mathbf{P}^2_k$ and $Z$ is a line, then this is false. (Hint: show that there aren't any nonempty affine opens.)

Sometimes yes: if $Z = \{x\}$ consists of a closed point, then $W$ is the spectrum of the local ring $\mathcal{O}_{X, x}$.

Here is a criterion: suppose that we can write $W = \bigcap_{i \in I} U_i$ as the intersection of quasi-compact opens with the following properties: (a) $\forall i, j \in I, \exists k \in I$ such that $U_k \subset U_i \cap U_j$ and (b) $\forall i, j \in I$ if $U_i \subset U_j$ then $U_i \to U_j$ is an affine morphism. Then $W = \lim U_i$ is a scheme.

If $X$ is a smooth projective surface then (a) and (b) hold if $Z$ is the fibre of a nonconstant morphism from $X$ to a curve or if $Z$ can be blown down (to a point on another algebraic surface -- does not work if the contraction gives an algebraic space).

If $X$ is a smooth projective surface over the algebraic closure of a finite field and $f : X \to S$ is a nonconstant morphism to a smooth projective curve and $Z$ is the image of a section of $f$, then it is often the case that the normal bundle of $Z$ in $X$ is negative (think about elliptic surfaces for example). Then by Artin you can contract $Z$ and hence $W$ is a scheme.

Anyway, it is fun to play around with these ideas, but I do not know a good way to work with $W$ like this. Many people would, I think, instead consider the formal completion of $X$ along $Z$ and work with that instead.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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