(The word "geometric theory" below is used in the sense of logic / classifying topos.)

We know that the (big) Zariski topos over $\text{Spec}\mathbb{Z}$ classifies the theory of local rings. My question is, what geometric theory does the (big) Zariski topos over $\mathbb{P}^1$ classify?

From algebraic geometry point of view it should classify something like a local rings together with a map to $\mathbb{P}^1$, i.e. something like isomorphism classes of $(A, L, e_0, e_1)$ where $A$ is a local ring, $L$ is an locally free $A$-module of rank 1, and $e_0, e_1\in L$ spans $L$. Is there a good way to describe a geometric / coherent theory of it?

I saw some discussions here but it's mostly about the affine case. I'm curious on what role properness may play here. If possible I'd like to see some description on the geometric theory for a general scheme (or algebraic curve).

(On "locally free $A$-module of rank 1": in this case L must be free, but that uses another result, merely asking for being locally free of rank 1 seems more natural to me.)

  • 3
    $\begingroup$ Because it's a local construction, the difference between the affine case and the general case shouldn't matter much. I think in general, the big Zariski topos of $X$ classifies pairs of a local ring $R$ and an $R$-point of $X$. What is bad about your description of $\mathbb P^1$? $\endgroup$
    – Will Sawin
    Jan 1 '17 at 6:43
  • $\begingroup$ @WillSawin Here by "classify" I mean "how does it serve as a classifying topos of some particular logic theory" so I'd like to see a few axioms for it. (e.g. local rings can be described by the axioms of rings + that for all x, either $x$ or $1-x$ is invertible.) The axioms would look like some extra data on $R$. Here I guess $\mathbb{P}^1$ is easy, as it's homogeneous, but for general X it won't be so easy, for reasons similar to that people usually don't have a good way to describe a general scheme as a representable functor. $\endgroup$
    – h__
    Jan 1 '17 at 8:16
  • $\begingroup$ It seems to me that one can do a general projective variety by the same trick with a locally free module of rank $1$ - one has an invertible module and $n$ elements such that at least one of the elements invertible and such that the $n$ elements satisfy certain homogenous relations in some tensor powers of the model. This handles algebraic curves, at least. $\endgroup$
    – Will Sawin
    Jan 1 '17 at 8:37
  • $\begingroup$ @WillSawin Yes of course. One can use this for functors represented by projective schemes. But that's very ad hoc. For example it's hard to see if two functors are isomorphic this way. $\endgroup$
    – h__
    Jan 1 '17 at 8:52
  • $\begingroup$ I don't think there's any particular reason to expect a description in terms of a logical theory to be the best tool to see if the two functors are isomorphic. $\endgroup$
    – Will Sawin
    Jan 1 '17 at 9:56

First note that a morphism $\operatorname{Spec}(A) \to \mathbb{P}^1$ is just given by an element of the "classical projective space" $\mathbb{P}^1(A) = \{ [a:b] \,|\, \text{$a$ is invertible or $b$ is invertible} \}$, if $A$ is a local ring. The description you gave is valid as well, but since (as you remark) "locally free" is equivalent to "free" over a local ring, it can be simplified further.

The big Zariski topos of $\mathbb{P}^1$ classifies the theory of a "local ring together with a point $[a:b]$". This theory can be explicitly described as follows:

  • A sort $R$ together with function symbols, constants, and axioms expressing that $R$ is a local ring.
  • A sort $P$ (to be thought of as the set of $[a:b]$ with $a,b:R$ where at least one coordinate is invertible) together with a relation $\langle\cdot,\cdot,\cdot\rangle$ on $R \times R \times P$ and the following axioms:
    • $\text{$a$ is invertible} \vee \text{$b$ is invertible} \dashv\vdash_{a,b:R} \exists p:P.\ \langle a,b,p \rangle$
    • $\langle a,b,p \rangle \wedge \langle a,b,p' \rangle \vdash_{a,b:R,\, p,p':P} p = p'$
    • $\top \vdash_{p:P} \exists a,b:R.\ \langle a,b,p \rangle$
    • $\langle a,b,p \rangle \wedge \langle a',b',p \rangle \dashv\vdash_{a,a',b,b':R,\, p:P} \exists s:R.\ \text{$s$ is invertible} \wedge a' = s a \wedge b' = s b$
  • A constant of sort $P$.

In a topos $\mathcal{E}$, a model of this theory is given by a local ring $A$ in $\mathcal{E}$ together with a point of the "classical projective space of $A$" as defined above. (Equivalently, by a rank-1 quotient of $A^2$ up to isomorphism of quotients.)

A different theory which the big Zariski topos of $\mathbb{P}^1$ classifies is the theory of a "homogeneous filter $F$ of $\mathbb{Z}[X,Y]$ meeting the irrelevant ideal together with a local ring over the degree-zero part $A := \mathbb{Z}[X,Y][F^{-1}]_0$ which is local over $A$".

For an arbitrary $\mathbb{N}$-graded ring $S$, the big Zariski topos of $\operatorname{Proj}(S)$ classifies the theory of a "homogeneous filter $F$ of $S$ meeting the irrelevant ideal together with a local ring over the degree-zero part $A := S[F^{-1}]_0$ which is local over $A$". (This in turn can be rewritten using the universal property of the degree-zero part of graded localization.) This description follows by combining the description of the little Zariski topos of $\operatorname{Proj}(S)$ (which classifies the theory of a "homogeneous filter $F$ of $S$ meeting the irrelevant ideal") and the description of the big Zariski topos as a topos over the little Zariski topos (which classifies the theory of a "local ring over $\mathcal{O}_S$ which is local over $\mathcal{O}_S$").

Proofs can be found in these note of mine (Section 12.8 and Section 16.1 at the time of writing).

If you care about issues of size, note that all the statements made here are only valid if we define the big Zariski topos of a scheme $X$ using the site consisting of only the locally finitely presented $X$-schemes. (It doesn't matter whether we restrict to affine schemes or not.) If we employ the site consisting of all $X$-schemes, the construction will not be well-defined in ordinary set theory, and if we restrict to those $X$-schemes contained in some universe (either Grothendieck or partial as in the Stacks project), then the resulting topos will have much more points.


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.