Is there a geometric realization of $\mathbf{C}((t))$-varieties?

Let $MV_F$ be the $\mathbf{A}^1$-homotopy category over the field $F$. Let $H$ be the homotopy category of spaces, and let $H_{/S^1}$ be the homotopy category of spaces over the circle.

When $F = \mathbf{C}$, there's a Betti realization functor $MV_{\mathbf{C}} \to H$. Is there a similar realization functor $MV_{\mathbf{C}((t))} \to H_{/S^1}$?

Etale homotopy theory gives a realization in profinite spaces over a profinite circle. I'm looking for a refinement of that.

If a variety over $\mathbf{C}((t))$ is defined using only convergent power series, I think it is clear which bundle over $S^1$ I want. If the variety is defined using finitely many nonconvergent power series, I can imagine some tricks for defining the bundle over S^1 that I want, even up to homeomorphism. But if a map between varieties has some nonconvergent power series in it, I bet there is no way to write down a map between those bundles, not functorially. I hope that it can be done up to homotopy.

• Did you consider (1) finding a model over $\mathbb{C}[[t]]$, (2) applying Artin approximation? At least if your variety is smooth and proper, the topology of the general fiber should depend only on the model mod $t^N$ for some big $N$ (e.g. s.t. $t^N$ kills the torsion in $\Omega_{X/\mathbb{C}[[t]]}$). Apply Artin approximation to get a variety over $\mathbb{C}\{t\}$ (henselization), and hence over convergent power series. – Piotr Achinger Dec 25 '16 at 10:01
• A minor point, but do you mean a) the (homotopy category of spaces) over the circle, or b) the homotopy category of (spaces over the circle)? They are potentially a little different. – David Roberts Dec 25 '16 at 23:57
• Hi David, I want (b) or something even stronger, but getting to (a) is already hard for me. – David Treumann Dec 26 '16 at 4:32

I know nothing about $\mathbf{A}^1$-homotopy. I will describe how one can get a functor from smooth varieties over $\mathbf{C}((t))$ to ${\rm Top}_{/\mathbf{S}^1}$ using log smooth proper models and Kato-Nakayama spaces of their special fibers.

Review of Kato–Nakayama spaces

If $X$ is an fs (finite and saturated) log scheme such that the underlying scheme $\underline X$ is of finite type over $\mathbf{C}$, one can associate to it in a functorial way a topological space $X^{\rm log}$, called the Kato–Nakayama space of $X$, together with a proper map $\tau:X^{\rm log}\to X$. The points of $X^{\rm log}$ correspond to pairs $(x, h)$ of a point $x\in\underline X$ and a map $h:\mathscr{M}_{X, x}\to \mathbf{S}^1$ such that for $f\in \mathscr{O}^\times_{X, x}$, $h(f) = f(x)/|f(x)|$.

In the special case when $X=\mathbf{A}^1$ with the usual log structure given by the open immersion $j:\mathbf{G}_m\hookrightarrow\mathbf{A}^1$, $X^{\rm log}$ is the real blow-up' $$X^{\rm log} = \mathbf{R}_{\geq 0}\times \mathbf{S}^1$$ and the map $\tau$ sends $(r, \theta)$ to $r\cdot\theta\in \mathbf{C}=\mathbf{A}^1(\mathbf{C})$. The open immersion $j$ factors through $\tau$, and the fiber $\tau^{-1}(0)\cong \mathbf{S}^1$.

In general, the Kato-Nakayama space has the following nice properties:

1. If $f:X\to Y$ is strict (i.e., the induced $f^*\mathscr{M}_Y\to \mathscr{M}_X$ is an isomorphism), then the square $\require{AMScd}$ \begin{CD} X^{\rm log} @>>> Y^{\rm log} \\ @VVV @VVV \\ X @>>> Y \\ \end{CD} is cartesian.

1. If $X=\mathbf{A}_P$ for an fs monoid $P$ (by definition, $\mathbf{A}_P={\rm Spec}(P\to \mathbf{C}[P])$), then $$X^{\rm log} = {\rm Hom}(P, \mathbf{R}_{\geq 0}\times \mathbf{S}^1 )$$ and the map $\tau$ is induced by the multiplication $\mathbf{R}_{\geq 0}\times \mathbf{S}^1 \to \mathbf{C}$. In particular, $X^{\rm log}$ is a manifold with boundary. Since an fs log scheme is by definition one which locally admits a strict map to some $\mathbf{A}_P$, this together with 1. gives a complete local description of $X^{\rm log}$.

2. The fiber over $x\in \underline X$ is a torsor under ${\rm Hom}(\mathscr{M}_{X, x}, \mathbf{S}^1)$ in a natural way.

3. If $X$ is log regular (say, log smooth over $\mathbf{C}$ with the trivial log structure), then $X^{\rm log}$ is a manifold with boundary. If $X_{\rm tr}$ denotes the biggest open of $\underline X$ where the log structure is trivial, then the inclusion $j:X_{\rm tr}\hookrightarrow \underline X$ lifts to an open immersion $\bar j : X_{\rm tr}\hookrightarrow X^{\rm log}$ along $\tau$. Moreover, $\bar j$ is a homotopy equivalence.

4. (Nakayama–Ogus) If $f:X\to Y$ is log smooth and exact, then the induced $f^{\rm log}:X^{\rm log}\to Y^{\rm log}$ is a topological submersion (i.e., locally on the source looks like a projection), and its fibers are manifolds with boundary. If $\underline f$ is proper, then $f^{\rm log}$ is a locally trivial fibration.

Notation. $V=\mathbf{C}[[t]]$, $K = {\rm Frac}(V)= \mathbf{C}((t))$, $\bar K = \bigcup_n \mathbf{C}((t^{1/n}))$ the algebraic closure of $K$, $k=V/m = \mathbf{C}$, $S={\rm Spec}\, V$, $\eta = {\rm Spec}\, K$, $\bar\eta = {\rm Spec}\, \bar K$, $s={\rm Spec}\, k$. We give $S$ the natural log structure coming from the open immersion $\eta\hookrightarrow S$.

Let $C$ be the category of smooth schemes over $\eta$, and let $\bar C$ be the category of proper log smooth fs log schemes over $S$. We have two crucial functors: $$X\mapsto X_{\rm tr} : \bar C \to C$$ $$X\mapsto X_s^{\rm log} : \bar C \to {\rm Top}_{/\mathbf{S}^1}$$ The first one associates to a log scheme $X/S$ the locus $X_{\rm tr}$ where the log structure is trivial. This maps to $S_{\rm tr} = \eta$, so is an $\eta$-scheme, and it is smooth when $X$ is log smooth. The second functor is the Kato–Nakayama space of the special fiber $X_s$ with the induced log structure, endowed with the natural map $X_s^{\rm log} \to s^{\rm log} \cong \mathbf{S}^1$. It is a locally trivial fibration whose fibers are manifolds with boundary (by Nakayama-Ogus).

The functors have the following nice properties:

1. Suppose that an object $X$ of $\bar C$ arises as a base change of a proper log smooth analytic space $X$ over an open disc $\Delta$ with log structure induced by $\Delta\setminus\{0\}$. Then $Y^{\rm log}\to U^{\rm log}$ is a locally trivial fibration (by Nakayama–Ogus) and $X^{\rm log}_s\to s^{\rm log}$ is its base change to the base point $s=0$. So in this convergent' case the construction recovers what you want.

2. The functor $X\mapsto X_{\rm tr}$ is essentially surjective. This is easily proved using Hironaka: embed a given smooth $Y/\eta$ into a proper $\bar Y_0/S$ by Nagata, then resolve the singularities to get a regular $\bar Y/S$ such that $\bar Y\setminus Y$ is a divisor with normal crossings. Finally, endow $\bar Y$ with the log structure associated to the inclusion $Y\hookrightarrow \bar Y$. Then $\bar Y/S$ is log smooth (this fails if $k$ has positive characteristic). It may not be semistable, and the morphism to $S$ might not be saturated, but I guess we don't care.

Target theorem. The functor $X\mapsto X_s^{\rm log}$ factors through $X\mapsto X_{\rm tr}$ (at least if we change the target category into the homotopy category).

I will sketch how I would prove it. Let $Y$ be an object of $C$. The second functor is essentially surjective; let $X$ be an object of $\bar C$ with $X_{\rm tr} \cong Y$. We want to check that $Y\mapsto X_s^{\rm log}$ is independent of $X$. Given two such $X, X'$, we can find a third one related to both by log blow-ups $X''\to X$, $X''\to X'$. Log blow-ups are log etale and should not change the Kato-Nakayama space of the special fiber, so the induced $(X'')^{\rm log}_s\to X^{\rm log}$ are isomorphisms. For functoriality, given a map $f:Y'\to Y$ in $C$ one can using Hironaka find a `model' $X'\to X$ in $\bar C$ and look at the associated map $(X')^{\rm log}_s\to X^{\rm log}_s$.

End note. We could even work with proper formal schemes over $S$ and the theory should be the same.

Below is my initial answer. Let $X/\mathbb{C}((t))$ be a smooth proper variety. I will define the bundle over $\mathbb{S}^1$ you want up to a finite covering $\mathbb{S}^1\to\mathbb{S}^1$.

After a finite ramified base change, $X$ has a log smooth (semistable) model $\mathscr{X}/\mathbb{C}[[t]]$ where $\mathbb{C}[[t]]$ is given the natural log structure. Let $X_0/S$ be the log special fiber over the log point $S={\rm Spec}(\mathbb{N}\to\mathbb{C})$. Consider the associated map on Kato–Nakayama spaces $$X_0^{\rm log} \to S^{\log}\cong \mathbb{S}^1.$$ This is a locally trivial bundle over $\mathbb{S}^1$ (because $X_0/S$ is log smooth and proper), and if the model $\mathscr{X}/\mathbb{C}[[t]]$ arises as the base change of an algebraic family, it agrees with what happens topologically on the generic fiber.

If $X$ is non-smooth or non-proper, you can probably use simplicial methods to reduce to the smooth proper case, adding some horizontal log structure.

P.S. If I find the time, I will try to add some more details later today.

• After the degree $n$ base change, there will be a natural order $n$ automorphism of $X$, and thus I think of the semistable model (if chosen correctly), lifting a generating automorphism of the ramified cover. Does this act on the Kato-Nakayama space? Can you quotient by it? – Will Sawin Dec 25 '16 at 13:16
• @WillSawin I hope so. I'm not sure how to get an action on a semistable model, but I guess this should be possible. – Piotr Achinger Dec 25 '16 at 13:35
• I think it's now known how to do resolution of singularities canonically. That should make it possible to do semistable reduction canonically as well. – Will Sawin Dec 25 '16 at 15:58
• Thanks Piotr and Will. What are the prospects for making this functorial for C((t))-algebraic maps? It probably doesn't make it any easier, but getting it to work for open smooth varieties is more important to me than getting it to work for singular varieties. – David Treumann Dec 25 '16 at 19:33
• How strongly is $X^{log}_s$ invariant under blowups? In other words, if I choose a different $X''$, will I get the same map? I am suspecting based on Will's answer below that I should not expect this in general, but that one can probably build "a contractible space of choices." – Mohammed Abouzaid Dec 26 '16 at 1:50

I don't know about $\mathbb A^1$-homotopy categories. Let me try to do something that I think will work just for the category of quasiprojective varieties.

The key lemma is this:

There is a canonical way to associate to a variety $X$ over $\mathbb C$ and a $\mathbb C((t))$-point of $X$ a homotopy class of maps from the circle to $X$ that is functorial for maps of varieties over $\mathbb C$.

To do this, by a stratification of $X$ reduce to the case when $X$ is smooth, then by resolution of singularities we may assume that $X$ has a smooth normal crossings compactification. Then extend the $\mathbb C((t))$-point to a $\mathbb C[[t]]$-point of the compactification. Locally near that point, $X$ looks like $\mathbb R^n$ times a torus, so a homotopy class of circle maps is a tuple of integers - you just take the $t$-adic valuations of the coordinates in the obvious way.

Could this depend on the compactification? No. It only depends on the coordinates up to some higher power of $t$, i.e. it is a locally constant function in the $t$-adic topology on $X(\mathbb C((t)))$. Because we may approximate any solution by a solution with algebraic (and hence convergent) power series using Hensel's lemma for the ring of algebraic power series, it is sufficient to check this independence for convergent power series, where it is obvious, because our definition recovers the usual definition given by evaluating the power series on a small circle.

An identical reduction argument proves functorially.

We view a quasiprojective variety $X - Z \subseteq \mathbb P^n$ over $\mathbb C((t))$ as a $\mathbb C((t))$-point of the flag Hilbert scheme parameterizing pairs of closed subscheme $Z \subseteq X \subseteq \mathbb P^n$. This Hilbert scheme admits a stratification into strata where the fibers have constant topological type. For each stratum $M$ in the Hilbert scheme, we have a fibration of manifolds (or more general spaces) over $M(\mathbb C)$, so $\mathbb C(t))$-point of $M$ defines a homotopy class of circle embeddings into $M$ and hence a homeomorphism class of manifolds over the circle.

For functoriality, we use a more intricate moduli space parameterizing ordered triples of two quasiprojective varieties and a map between them. Again there will be a stratification of constant topological types, and a homotopy class of circle embeddings into a stratum, giving a class of pairs of manifolds w/ a map between them, up to, I believe, homotopy.

You should be able to check that this agrees with composition of maps bby passing to the moduli space of triples of varieties and pairs of morphisms. To check that the homotopies defined their satisfy the appropriate coherence condition, you would have pass to the moduli space of quadruples of varieties with triples of morphisms, and so on.

• Will that is a very appealing approach. – David Treumann Dec 26 '16 at 4:16
• For this to work (that is to provide a functor from the $\mathbb{A}^1$-homotopy category) you need to deal with iterated homotopies too (that is for any sequence of composable maps of varieties $X_0→X_1→\cdots→X_n$ you need to provide higher homotopies $|X_i|×(\Delta^1)^{j-i}→|X_j|$ (where to the vertices of the cube correspond to the composition of partial maps $|X_{i_0}|→\cdots→|X_{i_k}|$). I think your strategy, using more complicated Hilbert schemes, will work, but it is not completely trivial. – Denis Nardin Dec 26 '16 at 11:51
• @DenisNardin I will think about how to justify it precisely... – Will Sawin Dec 26 '16 at 14:11