ScalingStacks

3 Integral affine structures [04P9]

Original official author HTML, exact retained edition. Historical TeX conversion verdicts remain unchanged. Cited-edition alignment and mathematical self-containment are not assessed.

Complete original source context · Original author HTML

3 Integral affine structures

Let X/KX/K be a smooth nn-dimensional maximally degenerate Calabi–Yau variety. In this chapter we compute the transition functions between the charts of the integral affine structure on Sk⁡(X)\Sk(X) associated with a minimal model of XX, or obtained by combining several minimal models. This relies on and generalizes the construction in [NXY19].
We then focus on certain degenerations of quartic K​3K3 surfaces (Section 3.3), and later of quintic 33-folds (Section 4): we apply Theorem B to reconstruct integral affine structures on the essential skeleton, and provide explicit formulas for the monodromy transformations around the singularities.

3.1 Integral affine structure induced by a model

Let 𝒳/R\mathscr{X}/R be a minimal model of XX; we assume that the special fiber 𝒳k=∑i∈IDi\mathscr{X}_{k}=\sum_{i\in I}D_{i} is reduced. We consider a one-dimensional stratum C=D1∩…∩DnC=D_{1}\cap\ldots\cap D_{n} of 𝒳k\mathscr{X}_{k}, which is therefore a smooth rational curve, and is such that (𝒳,𝒳k)(\mathscr{X},\mathscr{X}_{k}) is an snc pair in a formal neighbourhood of CC by [NXY19, Corollary 4.6]. Since (C,ΔC)(C,\Delta_{C}) is log Calabi–Yau, we may write its boundary as ΔC=p0+p∞\Delta_{C}=p_{0}+p_{\infty}, where p0=C∩D0p_{0}=C\cap D_{0} and p∞=C∩D∞p_{\infty}=C\cap D_{\infty} for two irreducible components D0,D∞D_{0},D_{\infty} of 𝒳k\mathscr{X}_{k} meeting CC transversally.

Following [NXY19], we write bi=−(C⋅Di)b_{i}=-(C\cdot D_{i}) for i=1,…,ni=1,\ldots,n; from C⋅𝒳k=0C\cdot\mathscr{X}_{k}=0 we infer ∑i=1nbi=2\sum_{i=1}^{n}b_{i}=2. The Star⁡(τC)\Star(\tau_{C}) consists on the union of two maximal faces corresponding to the zero-dimensional strata p0,p∞p_{0},p_{\infty}, meeting along τC\tau_{C}. The goal of this section is to describe the integral affine structure on Star⁡(τC)\Star(\tau_{C}) in terms of the intersection numbers bib_{i}’s, with no assumption on their positivity.

Proposition 3.1.1.

Let ρ𝒳\rho_{\mathscr{X}} be the retraction associated with the model 𝒳\mathscr{X}, and endow ​S​k​(X)\emph{Sk}(X) with the ℤ\mathbb{Z}-affine structure induced by ρ𝒳\rho_{\mathscr{X}} away from the codimension 2 faces of Sk⁡(𝒳)\Sk(\mathscr{X}). Then Star⁡(τC)\Star(\tau_{C}) is ℤ\mathbb{Z}-affine isomorphic to the union of the simplices <v0,v1,…,vn><v_{0},v_{1},\ldots,v_{n}> and <v1,…,vn,v∞><v_{1},\ldots,v_{n},v_{\infty}> in ℝn\mathbb{R}^{n} where

v0=(1,0,…,0)v_{0}=(1,0,\ldots,0), v1=(0,1,…,0)v_{1}=(0,1,\ldots,0),…, vn=0v_{n}=0 and v∞=(−1,b1,…,bn−1)v_{\infty}=(-1,b_{1},\ldots,b_{n-1}).

Proof.

We write b=mini⩽n⁡bib=\min_{i\leqslant n}b_{i}; we assume bb to be negative or zero by the condition b1+…+bn=2b_{1}+\ldots+b_{n}=2, as the case n=2n=2 and b1=b2=1b_{1}=b_{2}=1 is already treated in the proof of [NXY19, prop. 5.4].

The blow-up 𝒳1\mathscr{X}_{1} of the point p∞p_{\infty} in 𝒳\mathscr{X} yields a new irreducible component D∞,1D_{\infty,1} (we denote the strict transforms by the same letters for notational simplicity) with multiplicity N∞,1=n+1N_{\infty,1}=n+1, the point p∞,1=C∩D∞,1p_{\infty,1}=C\cap D_{\infty,1} and the intersection numbers bi,1≔−(C⋅Di)𝒳1=bi+1b_{i,1}\coloneqq-(C\cdot D_{i})_{\mathscr{X}_{1}}=b_{i}+1. If we repeat the process ss times, we obtain the models 𝒳s\mathscr{X}_{s}, the exceptional divisors D∞,sD_{\infty,s} with multiplicity N∞,s=n​s+1N_{\infty,s}=ns+1, the points p∞,s=C∩D∞,sp_{\infty,s}=C\cap D_{\infty,s} and the intersection numbers bi,s≔−(C⋅Di)𝒳s=bi+sb_{i,s}\coloneqq-(C\cdot D_{i})_{\mathscr{X}_{s}}=b_{i}+s.

For s=1−bs=1-b, we have mini⩽n⁡{bi,1−b}>0\min_{i\leqslant n}\{b_{i,1-b}\}>0, and by [NXY19] the integral affine structure induced by 𝒳1−b\mathscr{X}_{1-b} on Star⁡(τC)\Star(\tau_{C}) is given by v0,…,vnv_{0},\ldots,v_{n} and

(3.1.2) v∞,1−b=1n⁡(1−b)+1​(−1,b1+1−b,…,bn−1+1−b).v_{\infty,1-b}=\frac{1}{n(1-b)+1}(-1,b_{1}+1-b,\ldots,b_{n-1}+1-b).

The sequence of blow-ups 𝒳s+1→𝒳s\mathscr{X}_{s+1}\rightarrow\mathscr{X}_{s} induces (weighted) barycentric subdivisions of the faces τp∞,s\tau_{p_{\infty,s}} with vertices such that

(3.1.3) N∞,s+1​v∞,s+1=N∞,s​v∞,s+∑i=1nvi.N_{\infty,s+1}v_{\infty,s+1}=N_{\infty,s}v_{\infty,s}+\sum_{i=1}^{n}v_{i}.

Combining Eq. 3.1.2 and Eq. 3.1.3, at each step we obtain that

v∞,s=1n​s+1​(−1,b1+s,…,bn−1+s),v_{\infty,s}=\frac{1}{ns+1}(-1,b_{1}+s,\ldots,b_{n-1}+s),

and in particular v∞=(−1,b1,…,bn−1)v_{\infty}=(-1,b_{1},\ldots,b_{n-1}). The proposition follows from the following lemma. ∎

Lemma 3.1.4.

Let B=τ1∪τ2B=\tau_{1}\cup\tau_{2} be the union of two nn-dimensional simplices along a face of codimension one. Assume we are given a ℤ\mathbb{Z}-affine structure on BB, compatible with those on the τi\tau_{i}’s.

Suppose there exists a sequence of (weighted) star subdivisions of τ1\tau_{1} such that B′≔Star⁡(τ1∩τ2)B^{\prime}\coloneqq\Star(\tau_{1}\cap\tau_{2}) (with respect to this subdivision) can be embedded in ℝn\mathbb{R}^{n} compatibly with the ℤ\mathbb{Z}-affine structure. Then this embedding extends to BB, and the ℤ\mathbb{Z}-affine structure on BB is uniquely recovered by this embedding.

Proof.

The assumptions yield two charts for the ℤ\mathbb{Z}-affine structure on BB: the ℤ\mathbb{Z}-affine subsets B′B^{\prime} and τ1\tau_{1}. These two charts are glued along B′∩τ1B^{\prime}\cap\tau_{1} which is a simplex and thus has no non-trivial ℤ\mathbb{Z}-automorphisms preserving the vertices, hence the affine structure on BB is uniquely determined. The set B′⊂ℝnB^{\prime}\subset\mathbb{R}^{n} can be obtained as the result of the same star subdivisions of a subset B~⊂ℝn\tilde{B}\subset\mathbb{R}^{n}, and uniqueness of the affine structure ensures a ℤ\mathbb{Z}-affine isomorphism B≃B~B\simeq\tilde{B}. ∎

Remark 3.1.5.

Consider an irreducible component DiD_{i} of 𝒳k\mathscr{X}_{k} and write ΔDi=∑j≠iDj∩Di\Delta_{D_{i}}=\sum_{j\neq i}D_{j}\cap D_{i}. By adjunction, the pair (Di,ΔDi)(D_{i},\Delta_{D_{i}}) is log Calabi–Yau, i.e. DiD_{i} is a smooth projective variety over kk and ΔDi\Delta_{D_{i}} is a divisor such that KDi+ΔDiK_{D_{i}}+\Delta_{D_{i}} is trivial. By [EM21, Theorem 6.14] there exists a Lagrangian torus fibration

ϕ:𝒰→B⊆Star⁡(vDi)∖W\phi:\mathcal{U}\rightarrow B\subseteq\Star(v_{D_{i}})\setminus W

where 𝒰\mathcal{U} is a symplectic tubular neighborhood of the 11-dimensional strata of ΔDi\Delta_{D_{i}}, BB is a retract of Star⁡(vDi)∖W\Star(v_{D_{i}})\setminus W, and WW is the union of cells of codimension ⩾2\geqslant 2 in Sk⁡(𝒳)\Sk(\mathscr{X}). The fibration ϕ\phi is constructed gluing toric moment maps defined in the neighborhood of each stratum curve of ΔDi\Delta_{D_{i}}. Evans and Mauri compare the monodromy TϕT_{\phi} induced by ϕ\phi on BB to the monodromy Tρ𝒳T_{\rho_{\mathscr{X}}} induced by the affinoid torus fibration

ρ𝒳:ρ𝒳−1​(Star⁡(vDi)∖W)→Star⁡(vDi)∖W{\rho_{\mathscr{X}}}:\rho_{\mathscr{X}}^{-1}(\Star(v_{D_{i}})\setminus W)\rightarrow\Star(v_{D_{i}})\setminus W

and conclude that they are dual. This means that given a loop γ∈π1​(B)≃π1​(Star⁡(vDi)∖W)\gamma\in\pi_{1}(B)\simeq\pi_{1}(\Star(v_{D_{i}})\setminus W), we have Tρ𝒳​(γ)=(Tϕ​(γ)−1)TT_{\rho_{\mathscr{X}}}(\gamma)=(T_{\phi}(\gamma)^{-1})^{T}. Thus the affine structure constructed in [NXY19] has a symplectic topological analog. The duality is due to the fact that the image of the moment maps is MℝM_{\mathbb{R}}, while the image of the tropicalization map val\val is in NℝN_{\mathbb{R}}.

3.1.1 Case of K3 surfaces

Let X/KX/K be a maximally degenerate K​3K3 surface and let 𝒳/R\mathscr{X}/R be a minimal model of XX with reduced special fiber 𝒳k=∑i∈IDi\mathscr{X}_{k}=\sum_{i\in I}D_{i}. The dual complex 𝒟⁡(𝒳k)\mathcal{D}(\mathscr{X}_{k}) is well-known to be a triangulated sphere, whose vertices correspond to the irreducible components of 𝒳k\mathscr{X}_{k}.
We focus our attention to such a vertex vDv_{D}, and hence to the corresponding irreducible component DD of 𝒳k\mathscr{X}_{k}, which has boundary ΔD≔∑i=1r(Di∩D)=∑i=1rCi\Delta_{D}\coloneqq\sum_{i=1}^{r}(D_{i}\cap D)=\sum_{i=1}^{r}C_{i}. Since the simple normal crossing curve ΔD∈|−KD|\Delta_{D}\in\lvert-K_{D}\rvert is an anticanonical curve by adjunction, it follows from general surface theory that ΔD\Delta_{D} is a cycle of rational curves (Ci)i≤r(C_{i})_{i\leq r}, whose geometry is encoded by the bi=−(Ci⋅D)=−(Ci2)Db_{i}=-(C_{i}\cdot D)=-(C_{i}^{2})_{D}. We label the curves so that for i≤ri\leq r, Ci∩Ci+1≠∅C_{i}\cap C_{i+1}\neq\varnothing, with convention Cr+1=C1C_{r+1}=C_{1}.

One can associate to the pair (D,ΔD)(D,\Delta_{D}) a pseudo-fan, which is a singular affine structure on ℝ2\mathbb{R}^{2}, singular at most at 00. The singularity at 00 is a way to measure the defect of (D,ΔD)(D,\Delta_{D}) of being toric: the affine structure affine extends smoothly at 00 if and only (D,ΔD)(D,\Delta_{D}) is a toric pair [Eng18, Proposition 3.9].
The construction, as explained in [GHK15, §1.2], is the following. For each node pi=Ci∩Ci+1p_{i}=C_{i}\cap C_{i+1}, consider a cone σi≔ℝ≥0​vi+ℝ≥0​vi+1⊂ℝ2\sigma_{i}\coloneqq\mathbb{R}_{\geq 0}v_{i}+\mathbb{R}_{\geq 0}v_{i+1}\subset\mathbb{R}^{2}, (vi,vi+1)(v_{i},v_{i+1}) being a basis of the lattice ℤ2\mathbb{Z}^{2}. The cones σi\sigma_{i} and σi+1\sigma_{i+1} are then glued to each other along ℝ≥0​vi+1\mathbb{R}_{\geq 0}v_{i+1}, and the affine structure is extended through the edge by pretending that the pair (D,ΔD)(D,\Delta_{D}) is toric. If the pair was toric, the σi\sigma_{i}’s would be the maximal cones of its fan, and the relation

vi+2+vi=bi+1​vi+1v_{i+2}+v_{i}=b_{i+1}v_{i+1}

would hold by Eq. 1.2.4, so that the chart ψi:σi∪σi+1\psi_{i}:\sigma_{i}\cup\sigma_{i+1} that defines the ℤ\mathbb{Z}-affine structure satisfies ψi​(0)=0\psi_{i}(0)=0, ψi​(vi)=(1,0)\psi_{i}(v_{i})=(1,0), ψi​(vi+1)=(0,1)\psi_{i}(v_{i+1})=(0,1) and ψi​(vi+2)=(−1,bi+1)\psi_{i}(v_{i+2})=(-1,b_{i+1}), and is extended by dilatation. The unions of the σi\sigma_{i}’s glued along the successive edge is homeomorphic to ℝ2\mathbb{R}^{2}, and we obtain this way an ℤ\mathbb{Z}-affine structure away from the origin, extending to 00 if and only the pair is toric.

It follows from Proposition 3.1.1 that the singular ℤ\mathbb{Z}-affine structure induced by the Berkovich retraction ρ𝒳\rho_{\mathscr{X}} coincides with the one described above. We now determine the monodromy around the singularities.

Corollary 3.1.6.

Let DD be a component of 𝒳k\mathscr{X}_{k}, with boundary ΔD=∑i=1rCi\Delta_{D}=\sum_{i=1}^{r}C_{i}. Writing bi=−(Ci2)Db_{i}=-(C_{i}^{2})_{D}, the monodromy Tρ𝒳T_{\rho_{\mathscr{X}}} of the ℤ\mathbb{Z}-affine structure induced by ρ𝒳\rho_{\mathscr{X}} around vDv_{D} is given by

Tρ𝒳=(br1−10)⋅…⋅(b21−10)⋅(b11−10)T_{\rho_{\mathscr{X}}}=\left(\begin{matrix}b_{r}&1\\ -1&0\end{matrix}\right)\cdot\ldots\cdot\left(\begin{matrix}b_{2}&1\\ -1&0\end{matrix}\right)\cdot\left(\begin{matrix}b_{1}&1\\ -1&0\end{matrix}\right)

with respect to the basis (vDr,vD1)(v_{D_{r}},v_{D_{1}}) and origin vDv_{D}.

Proof.

By Proposition 3.1.1 the integral affine structure on Star⁡(τCi)\Star(\tau_{C_{i}}) identifies (vDi−1,vDi,vD,vDi+1)(v_{D_{i-1}},v_{D_{i}},v_{D},v_{D_{i+1}}) with

(v0=(1,0),v1=(0,1),v2=(0,0),v∞=(−1,bi)),(v_{0}=(1,0),v_{1}=(0,1),v_{2}=(0,0),v_{\infty}=(-1,b_{i})),

while on Star⁡(τCi+1)\Star(\tau_{C_{i+1}}) identifies (vDi,vDi+1,vD,vDi+2)(v_{D_{i}},v_{D_{i}+1},v_{D},v_{D_{i+2}}) with

(v0=(1,0),v1=(0,1),v2=(0,0),v∞=(−1,bi+1)).(v_{0}=(1,0),v_{1}=(0,1),v_{2}=(0,0),v_{\infty}=(-1,b_{i+1})).

It follows that the transition map from the chart Star⁡(τCi)\Star(\tau_{C_{i}}) to Star⁡(τCi+1)\Star(\tau_{C_{i+1}}) of the integral affine structure on Star⁡(τCi)∩Star⁡(τCi+1)\Star(\tau_{C_{i}})\cap\Star(\tau_{C_{i+1}}) is given by the matrix (bi1−10)\left(\begin{matrix}b_{i}&1\\ -1&0\end{matrix}\right). Thus, the composition of such matrices gives the monodromy around vDv_{D}, along a loop oriented as the path connecting vD1,vD2,…,vDr,vD1v_{D_{1}},v_{D_{2}},\ldots,v_{D_{r}},v_{D_{1}}. ∎

Remark 3.1.7.

It is well-known (see for instance [GHK15]) that Tρ𝒳=IdT_{\rho_{\mathscr{X}}}=\Id if and only the pair (D,ΔD=∑i=1rCi)(D,\Delta_{D}=\sum_{i=1}^{r}C_{i}) is toric, or if and only if the charge QQ vanishes, where

Q=χtop​(D∖ΔD)=12+∑i=1r(bi−3).Q=\chi_{\text{top}}(D\setminus\Delta_{D})=12+\sum_{i=1}^{r}(b_{i}-3).

3.2 Integral affine structure induced by combining several models

We start with a general definition.

Definition 3.2.1.

Let τ\tau be a simplex of dimension mm and consider the first barycentric subdivision τ′\tau^{\prime} of τ\tau. For each vertex vv of τ\tau, we denote the star of vv in τ′\tau^{\prime} by Star⁡(v)′\Star(v)^{\prime} and define Γm−1\Gamma_{m-1} to be the polyhedral complex of dimension m−1m-1 given by

Γm−1≔τ∖⋃v∈τStar⁡(v)′⊂τ.\Gamma_{m-1}\coloneqq\tau\setminus\bigcup_{v\in\tau}\Star(v)^{\prime}\subset\tau.

For instance, if m=2m=2, Γ1\Gamma_{1} is the union of the three line segments joining the barycenter of the triangle to the barycenters of the edges.

We return to the setting of Section 3, that is, let X/KX/K be a smooth nn-dimensional maximally degenerate Calabi–Yau variety. Assume we are given two minimal models 𝒳\mathscr{X}, 𝒳′\mathscr{X}^{\prime} of XX such that 𝒟⁡(𝒳k)=𝒟⁡(𝒳k′)\mathcal{D}(\mathscr{X}_{k})=\mathcal{D}(\mathscr{X}^{\prime}_{k}), so that Sk⁡(𝒳)=Sk⁡(𝒳′)=Sk⁡(X)\Sk(\mathscr{X})=\Sk(\mathscr{X}^{\prime})=\Sk(X), not only as sets but also with the same triangulation. We fix an ordered labelling (1,…,s)(1,\ldots,s) of the vertices of Sk⁡(X)\Sk(X), equivalently of the irreducible components of the special fiber of 𝒳\mathscr{X} (resp. 𝒳′\mathscr{X}^{\prime}).
Fix a codimension 1 face τ\tau of 𝒟⁡(𝒳k)=𝒟⁡(𝒳k′)\mathcal{D}(\mathscr{X}_{k})=\mathcal{D}(\mathscr{X}^{\prime}_{k}), with vertices vi1,…,vinv_{i_{1}},\ldots,v_{i_{n}}. We write CC (resp. C′C^{\prime}) for the corresponding strata curves of 𝒳\mathscr{X} (resp. 𝒳′\mathscr{X}^{\prime}), and DilD_{i_{l}} (resp. Dil′D^{\prime}_{i_{l}}), l=1,…,nl=1,\ldots,n the corresponding components. We then have C=Di1∩…∩DinC=D_{i_{1}}\cap\ldots\cap D_{i_{n}} , and similarly for C′C^{\prime}. We write

bil=−(C⋅Dil)𝒳b_{i_{l}}=-(C\cdot D_{i_{l}})_{\mathscr{X}}

the intersection number computed inside 𝒳\mathscr{X}, and similarly:

bil′=−(C′⋅Dil′)𝒳′b^{\prime}_{i_{l}}=-(C^{\prime}\cdot D^{\prime}_{i_{l}})_{\mathscr{X}^{\prime}}

for l=1,…,nl=1,\ldots,n. The (n−1)(n-1)-dimensional face τ\tau is contained in two maximal faces τp0\tau_{p_{0}} and τp∞\tau_{p_{\infty}} of 𝒟⁡(𝒳k)\mathcal{D}(\mathscr{X}_{k}), since the boundary of CC in (𝒳,𝒳k)(\mathscr{X},\mathscr{X}_{k}) consists of two strata points p0=C∩Di0p_{0}=C\cap D_{i_{0}} and p∞=C∩Di∞p_{\infty}=C\cap D_{i_{\infty}}; we assume i0<i∞i_{0}<i_{\infty}. We set 𝒮≔τp0∪τp∞⊂Sk⁡(X)\mathcal{S}\coloneqq\tau_{p_{0}}\cup\tau_{p_{\infty}}\subset\Sk(X), and Γ≔Γn−2⊂τC\Gamma\coloneqq\Gamma_{n-2}\subset\tau_{C} as in Definition 3.2.1.
Given two vertices vim,vim′v_{i_{m}},v_{i_{m^{\prime}}} of τC\tau_{C}, with corresponding components DimD_{i_{m}} and Dim′D_{i_{m^{\prime}}} of 𝒳k\mathscr{X}_{k} containing CC, we assume im<im′i_{m}<i_{m^{\prime}} and construct a loop γ\gamma as follows:

  • -

    γ\gamma is contained in 𝒮∖Γ\mathcal{S}\setminus\Gamma;

  • -

    γ\gamma goes around the segment joining the barycenter of τC\tau_{C} with the barycenter of the edge between vimv_{i_{m}} and vim′v_{i_{m^{\prime}}};

  • -

    γ\gamma has an orientation induced by the fixed ordered labelling on the vertices of Sk⁡(X)\Sk(X) in the following way: in 𝒮∖Γ\mathcal{S}\setminus\Gamma, γ\gamma is homotopy equivalent to the closed path given by the edges which connect in order

    vDi0,vDim,vDi∞,vDim′.v_{D_{i_{0}}},v_{D_{i_{m}}},v_{D_{i_{\infty}}},v_{D_{i_{m^{\prime}}}}.
vim=v2v_{i_{m}}=v_{2}v3=vim′v_{3}=v_{i_{m^{\prime}}}v4v_{4}v5=vi∞v_{5}=v_{i_{\infty}}vi0=v1v_{i_{0}}=v_{1}
vim=v2v_{i_{m}}=v_{2}v3v_{3}vim′=v4v_{i_{m^{\prime}}}=v_{4}v5=vi∞v_{5}=v_{i_{\infty}}vi0=v1v_{i_{0}}=v_{1}
Figure 1: *

Two examples of loops γ\gamma in the case n=3n=3 and C=D2∩D3∩D4C=D_{2}\cap D_{3}\cap D_{4}

Suppose we are given a retraction ρ:Xan⟶Sk⁡(X)\rho:X^{\an}\longrightarrow\Sk(X) such that

ρ={ρ𝒳 over ​Uim≔Int​(τp0)∪Int​(τp∞)∪Star⁡(vim)′ρ𝒳′ over ​Uim′≔Int​(τp0)∪Int​(τp∞)∪Star⁡(vim′)′.\rho=\begin{cases}\rho_{\mathscr{X}}&\textrm{ over }U_{i_{m}}\coloneqq\textrm{Int}(\tau_{p_{0}})\cup\textrm{Int}(\tau_{p_{\infty}})\cup\Star(v_{i_{m}})^{\prime}\\ \rho_{\mathscr{X}^{\prime}}&\textrm{ over }U_{i_{m^{\prime}}}\coloneqq\textrm{Int}(\tau_{p_{0}})\cup\textrm{Int}(\tau_{p_{\infty}})\cup\Star(v_{i_{m^{\prime}}})^{\prime}.\end{cases}
Proposition 3.2.2.

The monodromy along the loop γ\gamma, of the ℤ\mathbb{Z}-affine structure induced by ρ\rho on Uim∪Uim′U_{i_{m}}\cup U_{i_{m^{\prime}}} is

(3.2.3) Tρ​(γ)=(100…0bi1−bi1′10…0bi2−bi2′01…0⋱bin−1−bin−1′00…1)T_{\rho}(\gamma)=\left(\begin{matrix}1&0&0&\ldots&0\\ b_{i_{1}}-b^{\prime}_{i_{1}}&1&0&\ldots&0\\ b_{i_{2}}-b^{\prime}_{i_{2}}&0&1&\ldots&0\\ \vdots&\vdots&&\ddots&\\ b_{i_{n-1}}-b^{\prime}_{i_{n-1}}&0&0&\ldots&1\end{matrix}\right)

with respect to the basis (vDi0,vDi1,…,vDin−1)(v_{D_{i_{0}}},v_{D_{i_{1}}},\ldots,v_{D_{i_{n-1}}}) and origin vDinv_{D_{i_{n}}}.

Proof.

We need to compute the parallel transport of the vectors

(u0,…,un−1)≔(vDi0−vDin,vDi1−vDin,…,vDin−1−vDin)(u_{0},\ldots,u_{n-1})\coloneqq(v_{D_{i_{0}}}-v_{D_{i_{n}}},v_{D_{i_{1}}}-v_{D_{i_{n}}},\ldots,v_{D_{i_{n-1}}}-v_{D_{i_{n}}})

along the loop γ\gamma. By Proposition 3.1.1 the ℤ\mathbb{Z}-affine structure on Star⁡(τC)\Star(\tau_{C}) induced by ρ𝒳\rho_{\mathscr{X}} is described by the chart which has the following vertices:

v0=(1,0,…,0)v_{0}=(1,0,\ldots,0), v1=(0,1,…,0),…,vn=0v_{1}=(0,1,\ldots,0),\,\ldots,\,v_{n}=0 and v∞=(−1,bi1,…,bin−1);v_{\infty}=(-1,b_{i_{1}},\ldots,b_{i_{n-1}});

while the ℤ\mathbb{Z}-affine structure induced by ρ𝒳′\rho_{\mathscr{X}^{\prime}} is given by:

v0′=(1,0,…,0)v^{\prime}_{0}=(1,0,\ldots,0), v1′=(0,1,…,0),…,vn′=0v^{\prime}_{1}=(0,1,\ldots,0),\,\ldots,\,v^{\prime}_{n}=0 and v∞′=(−1,bi1′,…,bin−1′).v^{\prime}_{\infty}=(-1,b^{\prime}_{i_{1}},\ldots,b^{\prime}_{i_{n-1}}).

Moreover, the vectors ulu_{l} correspond to the vectors vlv_{l} (resp. vl′v^{\prime}_{l}) in the chart for ρ𝒳\rho_{\mathscr{X}} (resp. ρ𝒳′\rho_{\mathscr{X}^{\prime}}). We now have v0=−v∞+∑l=1n−1bil​vlv_{0}=-v_{\infty}+\sum_{l=1}^{n-1}b_{i_{l}}v_{l}, so that the vectors we are transporting are written on τp∞\tau_{p_{\infty}}

(−v∞+∑l=1n−1bil​vl,v1,…,vn−1)(-v_{\infty}+\sum_{l=1}^{n-1}b_{i_{l}}v_{l}\,,v_{1},\ldots,v_{n-1})

in the chart for ρ𝒳\rho_{\mathscr{X}}. These are thus mapped to the tuple (−v∞′+∑l=1n−1bil​vl′,v1′,…,vn−1′)(-v^{\prime}_{\infty}+\sum_{l=1}^{n-1}b_{i_{l}}v^{\prime}_{l},v^{\prime}_{1},\ldots,v^{\prime}_{n-1}) by the chart for ρ𝒳′\rho_{\mathscr{X}^{\prime}}. We now transport back across τC\tau_{C} in the chart for ρ𝒳′\rho_{\mathscr{X}^{\prime}}, to get the tuple of vectors

(−v0′−∑l=1n−1bil′​vl′+∑l=1n−1bil​vl′,v1′,…,vn−1′)(-v^{\prime}_{0}-\sum_{l=1}^{n-1}b^{\prime}_{i_{l}}v^{\prime}_{l}+\sum_{l=1}^{n-1}b_{i_{l}}v^{\prime}_{l}\,,v^{\prime}_{1},\ldots,v^{\prime}_{n-1})

according to the relation −v∞′=−v0′−∑l=1n−1bil′​vl′-v^{\prime}_{\infty}=-v^{\prime}_{0}-\sum_{l=1}^{n-1}b^{\prime}_{i_{l}}v^{\prime}_{l}. We now see that after parallel transport the vectors (u0,…,un−1)(u_{0},\ldots,u_{n-1}) have changed to

(u0+∑l=1n−1(bil−bil′)​ul,u1,…,un−1),(u_{0}+\sum_{l=1}^{n-1}(b_{i_{l}}-b^{\prime}_{i_{l}})u_{l}\,,u_{1},\ldots,u_{n-1}),

hence the formula Eq. 3.2.3 for the monodromy matrix. ∎

3.2.1 Case of K3 surfaces

We focus on the case of a maximally degenerate K​3K3 surface X/KX/K. We have C=Dim∩Dim′C=D_{i_{m}}\cap D_{i_{m^{\prime}}}, Γ={a}\Gamma=\{a\} is a point in the interior of τC\tau_{C} and γ\gamma is a loop around aa, oriented as the path joining in order vDi0,vDim,vDi∞,vDim′v_{D_{i_{0}}},v_{D_{i_{m}}},v_{D_{i_{\infty}}},v_{D_{i_{m^{\prime}}}}. We assume we have a retraction ρ:Xan⟶Sk⁡(X)\rho:X^{\an}\longrightarrow\Sk(X) such that

ρ={ρ𝒳 over Int​(τp0)∪Int​(τp∞)∪[vim,a)ρ𝒳′ over Int​(τp0)∪Int​(τp∞)∪[vim′,a)\rho=\begin{cases}\rho_{\mathscr{X}}&\textrm{ over }\textrm{Int}(\tau_{p_{0}})\cup\textrm{Int}(\tau_{p_{\infty}})\cup[v_{i_{m}},a)\\ \rho_{\mathscr{X}^{\prime}}&\textrm{ over }\textrm{Int}(\tau_{p_{0}})\cup\textrm{Int}(\tau_{p_{\infty}})\cup[v_{i_{m^{\prime}}},a)\end{cases}

where [vi⋅,a)[v_{i_{\cdot}},a) is the part of the edge τC\tau_{C} joining the vertex to aa, but not including aa. Then Proposition 3.2.2 may be rewritten as follows.

Corollary 3.2.4.

The monodromy along the loop γ\gamma, of the ℤ\mathbb{Z}-affine structure induced by ρ\rho on Star⁡(τC)∖{a}\Star(\tau_{C})\setminus\{a\}, is

(3.2.5) Tρ​(γ)=(10bim−bim′1)T_{\rho}(\gamma)=\left(\begin{matrix}1&0\\ b_{i_{m}}-b^{\prime}_{i_{m}}&1\end{matrix}\right)

with respect to the basis (vDi0,vDim)(v_{D_{i_{0}}},v_{D_{i_{m}}}) and origin vDim′v_{D_{i_{m^{\prime}}}}.

3.3 Degeneration of quartic K3 surfaces

We consider 𝒳={z1z2z3z4+tF4(z1,z2,z3,z4)=0}⊂ℙR3\mathscr{X}=\{z_{1}z_{2}z_{3}z_{4}+tF_{4}(z_{1},z_{2},z_{3},z_{4})=0\}\subset\mathbb{P}^{3}_{R}, where F4F_{4} is a generic homogeneous polynomial of degree 4. The degeneration 𝒳\mathscr{X} has the following properties:

  • 1.

    the special fiber 𝒳k\mathscr{X}_{k} is reduced, consisting of four Weil divisors, i.e. Di={zi=0,t=0}D_{i}=\{z_{i}=0,t=0\};

  • 2.

    𝒳\mathscr{X} has 2424 singular points, given by {zi=0,zj=0,t=0,F4=0}\{z_{i}=0,z_{j}=0,t=0,F_{4}=0\}, hence 4 on each Di∩DjD_{i}\cap D_{j}; the local model around a singular point is given by 𝒰={xy=zt}⊂𝔸R3\mathscr{U}=\{xy=zt\}\subset\mathbb{A}^{3}_{R};

  • 3.

    the pair (𝒳,𝒳k)(\mathscr{X},\mathscr{X}_{k}) is dlt. Indeed, it is snc away from the singularities, and around a singular point (𝒰,𝒰k)(\mathscr{U},\mathscr{U}_{k}) is log canonical by [CLS11, Proposition 11.4.24];

  • 4.

    the dual complex 𝒟⁡(𝒳k)\mathcal{D}(\mathscr{X}_{k}) is a PL-isomorphic to a tetrahedron, and hence homeomorphic to 𝕊2\mathbb{S}^{2};

  • 5.

    by adjunction, the canonical bundle K𝒳K_{\mathscr{X}} is trivial.

We conclude that 𝒳\mathscr{X} is a minimal dlt model of the K3 surface X≔𝒳KX\coloneqq\mathscr{X}_{K}, but it is not good in the sense of Section 1.1 since the prime components of the special fiber are not ℚ\mathbb{Q}-Cartier.

Our goal is to construct some explicit minimal models of XX starting from 𝒳\mathscr{X}, and then to study the integral affine structure on Sk⁡(X)\Sk(X) induced by these models or by combining several of them. To this purpose, we will apply Corollary 3.1.6 and Corollary 3.2.4.

Some good minimal models of XX are obtained by performing the following small resolutions of 𝒳\mathscr{X}. For any triple of elements i,j,ki,j,k in {1,…,4}\{1,\ldots,4\} and any fixed order (i,j,k)(i,j,k) on them, we blow-up in order the divisors DiD_{i}, DjD_{j} and DkD_{k}, and denote the resulting model by 𝒳i​j​k\mathscr{X}_{ijk} and the morphism by

gi​j​k:𝒳i​j​k→𝒳.g_{ijk}:\mathscr{X}_{ijk}\rightarrow\mathscr{X}.

The exceptional locus of gi​j​kg_{ijk} consists of 2424 smooth rational curves whose images via gi​j​kg_{ijk} are the singular points of 𝒳\mathscr{X}. In particular, the strict transform of DiD_{i} is isomorphic to the blow-up of DiD_{i} along the 1212 singular points in DiD_{i}; similarly for DjD_{j} at 88 points, and for DkD_{k} at the 44 remaining singular points. Instead, for h≠i,j,kh\neq i,j,k, DhD_{h} is isomorphic to its strict transform. These facts follow from local computations on 𝒰\mathscr{U}. Blowing-up Dx≔{x=t=0}D_{x}\coloneqq\{x=t=0\} induces an exceptional curve inside the strict transform D~x\tilde{D}_{x}, which is isomorphic to the blow-up of DxD_{x} along qq. The above claims now follow, since the singularities of 𝒳\mathscr{X} are isolated.
If we denote by D~m\tilde{D}_{m} for m∈{1,…,4}m\in\{1,\ldots,4\} the irreducible components of the special fiber of 𝒳i​j​k\mathscr{X}_{ijk}, and by Cm​m′=D~m∩D~m′C_{mm^{\prime}}=\tilde{D}_{m}\cap\tilde{D}_{m^{\prime}} the strata curves, then the intersection numbers in 𝒳i​j​k\mathscr{X}_{ijk} are

(3.3.1)
D~i\tilde{D}_{i} D~j\tilde{D}_{j} D~k\tilde{D}_{k} D~h\tilde{D}_{h} h≠i,j,kh\neq i,j,k
Ci​jC_{ij} 1 -3 1 1
Ci​kC_{ik} 1 1 -3 1
Ci​hC_{ih} 1 1 1 -3
Cj​kC_{jk} 1 1 -3 1
Cj​hC_{jh} 1 1 1 -3
Ck​hC_{kh} 1 1 1 -3

3.3.1 Integral affine structure induced by the model 𝒳i​j​k\mathscr{X}_{ijk}

By [NXY19] the non-archimedean SYZ fibration ρ𝒳i​j​k:San→Sk⁡(𝒳i​j​k)=Sk⁡(X)≃𝕊2\rho_{\mathscr{X}_{ijk}}:S^{\an}\rightarrow\Sk(\mathscr{X}_{ijk})=\Sk(X)\simeq\mathbb{S}^{2} is an affinoid torus fibration (at least) away from the vertices of the triangulation of Sk⁡(X)\Sk(X) induced by the special fiber of 𝒳i​j​k\mathscr{X}_{ijk}, i.e. away from the vDmv_{D_{m}}’s.
By Theorem B, ρ𝒳i​j​k\rho_{\mathscr{X}_{ijk}} is an affinoid torus fibration over Star⁡(τDh)\Star(\tau_{D_{h}}) for h≠i,j,kh\neq i,j,k, as Dh≃ℙ2D_{h}\simeq\mathbb{P}^{2} and gi​j​kg_{ijk} is an isomorphism on the strict transform of DhD_{h}. Moreover, by Remark 3.1.7 and Eq. 3.3.1 the integral affine structure induced by ρ𝒳i​j​k\rho_{\mathscr{X}_{ijk}} does not extend to vDiv_{D_{i}}, vDjv_{D_{j}} and vDkv_{D_{k}}.

We conclude that the singular points of the affine structure on Sk⁡(X)\Sk(X) induced by ρ𝒳i​j​k\rho_{\mathscr{X}_{ijk}} are precisely vDiv_{D_{i}}, vDjv_{D_{j}} and vDkv_{D_{k}}. Corollary 3.1.6 establishes that the monodromies around these vertices are

Tρ𝒳i​j​k\displaystyle T_{\rho_{\mathscr{X}_{ijk}}} (γi)=(218−8−3)\displaystyle(\gamma_{i})=\left(\begin{matrix}21&8\\ -8&-3\end{matrix}\right)
in the basis (vDj,vDk) and origin vDi,\displaystyle\text{in the basis $(v_{D_{j}},v_{D_{k}})$ and origin $v_{D_{i}}$},
Tρ𝒳i​j​k\displaystyle T_{\rho_{\mathscr{X}_{ijk}}} (γj)=(−15−441)\displaystyle(\gamma_{j})=\left(\begin{matrix}-15&-4\\ 4&1\end{matrix}\right)
in the basis (vDk,vDh) and origin vDj,\displaystyle\text{in the basis $(v_{D_{k}},v_{D_{h}})$ and origin $v_{D_{j}}$},
Tρ𝒳i​j​k\displaystyle T_{\rho_{\mathscr{X}_{ijk}}} (γk)=(1041)\displaystyle(\gamma_{k})=\left(\begin{matrix}1&0\\ 4&1\end{matrix}\right)
in the basis (vDh,vDi) and origin vDk.\displaystyle\text{in the basis $(v_{D_{h}},v_{D_{i}})$ and origin $v_{D_{k}}$}.
v2=vjv_{2}=v_{j}vh=v4v_{h}=v_{4}v3=vkv_{3}=v_{k}v1=viv_{1}=v_{i}γj\gamma_{j}γk\gamma_{k}γi\gamma_{i}

3.3.2 Integral affine structure induced combining more models

We recall a construction from [KS06, §4.2.5]. Consider the resolution h:𝒵→𝒳h:\mathscr{Z}\rightarrow\mathscr{X} obtained by blowing-up the 2424 singular points of 𝒳\mathscr{X}, which in particular dominates any model 𝒳i​j​k\mathscr{X}_{ijk}. Then the special fiber is 𝒵k=∑i=14Di+∑q=124Eq\mathscr{Z}_{k}=\sum_{i=1}^{4}D_{i}+\sum_{q=1}^{24}E_{q} and the associated dual complex is the boundary of a tetrahedron with four additional 22-cells glued along each edge of the tetrahedron; following Kontsevich–Soibelman we call such 22-cells wings.

We parametrize each edge ee of 𝒟⁡(𝒳k)\mathcal{D}(\mathscr{X}_{k}) by the interval [−1,1][-1,1], and each wing WqW_{q} glued to ee by the 22-simplex in ℝ(x,y)2\mathbb{R}^{2}_{(x,y)} bounded by ee and 0⩽y⩽1−|x|0\leqslant y\leqslant 1-|x|.

Lemma 3.3.2.

Let WqW_{q} be a wing over the edge el​he_{lh}, for l∈{i,j,k}l\in\{i,j,k\} and assume i,j,k,hi,j,k,h all distinct. Then the retraction ρ𝒳i​j​k:Wq→el​h\rho_{\mathscr{X}_{ijk}}:W_{q}\rightarrow e_{lh} is the contraction of WqW_{q} to the edge el​he_{lh} parallel to the edge el​qe_{lq}:

ρ𝒳i​j​k:Wq\displaystyle\rho_{\mathscr{X}_{ijk}}:W_{q} →el​h\displaystyle\rightarrow e_{lh}
(x,y)\displaystyle(x,y) ↦(x−y,0)\displaystyle\mapsto(x-y,0)
vhv_{h}vlv_{l}vqv_{q}xxyy
Proof.

The morphism 𝒵→𝒳i​j​k\mathscr{Z}\rightarrow\mathscr{X}_{ijk} is the blow-up of the 24 exceptional curves of gi​j​kg_{ijk}. In particular, the exceptional divisor EqE_{q} is the preimage in 𝒵\mathscr{Z} of a curve contained in D~l\tilde{D}_{l}; it follows that vEq​(z~l)=1v_{E_{q}}(\tilde{z}_{l})=1 and vEq​(z~h)=0v_{E_{q}}(\tilde{z}_{h})=0, where z~l,z~h\tilde{z}_{l},\tilde{z}_{h} are local equations for D~l,D~h\tilde{D}_{l},\tilde{D}_{h} on 𝒳i​j​k\mathscr{X}_{ijk}. The Berkovich retraction ρ𝒳i​j​k\rho_{\mathscr{X}_{ijk}} is linear on WqW_{q} and hence depends only on the image of vqv_{q}, which is determined by vEq​(z~l)v_{E_{q}}(\tilde{z}_{l}) and vEq​(z~h)v_{E_{q}}(\tilde{z}_{h}). Thus we conclude that ρ𝒳i​j​k​(vq)=vl\rho_{\mathscr{X}_{ijk}}(v_{q})=v_{l} and we have the result. ∎

Kontsevich and Soibelman define a retraction

ρ:Xan→ρ𝒵Sk⁡(𝒵)→ρ′𝕊2≃Sk⁡(X)=Sk⁡(𝒳)\rho:X^{\textrm{an}}\xrightarrow{\rho_{\mathscr{Z}}}\Sk(\mathscr{Z})\xrightarrow{\rho^{\prime}}\mathbb{S}^{2}\simeq\Sk(X)=\Sk(\mathscr{X})

where ρ𝒵\rho_{\mathscr{Z}} is the Berkovich retraction onto the skeleton Sk⁡(𝒵)\Sk(\mathscr{Z}), and ρ′\rho^{\prime} is a retraction of the 24 wings of Sk⁡(𝒵)\Sk(\mathscr{Z}) to the sphere given as follows. For each edge ee of 𝒟⁡(𝒳k)\mathcal{D}(\mathscr{X}_{k}) we choose a point ae=(ae,0)a_{e}=(a_{e},0) in the interior of ee, and define the retraction of WqW_{q} onto ee by

(x+y,0)(x+y,0) if x+y⩽aex+y\leqslant a_{e}
ρ′:(x,y)↦\rho^{\prime}:\,(x,y)\mapsto (x−y,0)(x-y,0) if x−y⩾aex-y\geqslant a_{e}
(ae,0)(a_{e},0) otherwise.
Picture for ae=0a_{e}=0aea_{e}vqv_{q}xxyy

We note that

  • -

    over the interior of any 22-dimensional face τ⊂Sk⁡(𝒳)\tau\subset\Sk(\mathscr{X}), ρ\rho is equal to ρ𝒵\rho_{\mathscr{Z}}, thus it is an affinoid torus fibration (see Example 1.6.2).

  • -

    Around any vertex vDv_{D}, ρ\rho is equal to ρ𝒳i​j​k\rho_{\mathscr{X}_{ijk}} for any triple such that D≠Di,Dj,DkD\neq D_{i},D_{j},D_{k}, as follows from the previous lemma. Thus, from Section 3.3.1, ρ\rho is an affinoid torus fibration around vDv_{D}, and the affine structure induced there is the fan structure induced by DD, by Corollary 2.6.1.

  • -

    For any edge ee corresponding to Ce=Dim∩Dim′C_{e}=D_{i_{m}}\cap D_{i_{m^{\prime}}}, adopting the notation of Section 3.2,

    ρ={ρ𝒳i​j​k for im≠i,j,k, over Int​(τp0)∪Int​(τp∞)∪[vim,ae)ρ𝒳i′​j′​k′ for im′≠i′,j′,k′, over Int​(τp0)∪Int​(τp∞)∪[vim′,ae),\rho=\begin{cases}\rho_{\mathscr{X}_{ijk}}&\textrm{ for $i_{m}\neq i,j,k$, over }\textrm{Int}(\tau_{p_{0}})\cup\textrm{Int}(\tau_{p_{\infty}})\cup[v_{i_{m}},a_{e})\\ \rho_{\mathscr{X}_{i^{\prime}j^{\prime}k^{\prime}}}&\textrm{ for $i_{m^{\prime}}\neq i^{\prime},j^{\prime},k^{\prime}$, over }\textrm{Int}(\tau_{p_{0}})\cup\textrm{Int}(\tau_{p_{\infty}})\cup[v_{i_{m^{\prime}}},a_{e}),\end{cases}

    and thus is an affinoid torus fibration over the union of these two open sets.

We conclude that ρ\rho induces an integral affine structure on Sk⁡(X)\Sk(X) away from the points aea_{e}. By Corollary 3.2.4, we can compute the monodromy around the singularities. As all these computations are analogous, we exhibit the case Ce=D1∩D2C_{e}=D_{1}\cap D_{2}:

Tρ​(γae)\displaystyle T_{\rho}(\gamma_{a_{e}}) =(10b1,𝒳234−b1,𝒳1341)\displaystyle=\left(\begin{matrix}1&0\\ b_{1,\mathscr{X}_{234}}-b_{1,\mathscr{X}_{134}}&1\end{matrix}\right)
=(103−(−1)1)\displaystyle=\left(\begin{matrix}1&0\\ 3-(-1)&1\end{matrix}\right)
=(1041)\displaystyle=\left(\begin{matrix}1&0\\ 4&1\end{matrix}\right)
γae\gamma_{a_{e}}aea_{e}v2=vim′v_{2}=v_{i_{m^{\prime}}}vi∞=v4v_{i_{\infty}}=v_{4}v3=vi0v_{3}=v_{i_{0}}v1=vimv_{1}=v_{i_{m}}

with respect to the basis (v3,v1)(v_{3},v_{1}) and origin v2v_{2}. This formula was already stated in [KS06, §4.2.5].

3.3.3 Dispersion of singularities

We construct a third singular integral affine structure on Sk⁡(X)\Sk(X) pushing forward the techniques developed so far. This can be viewed as a dispersion of singularieties with respect to the integral affine structure studied in Section 3.3.2: on each edge we pass from one singular point around which the monodromy is (1041)\begin{pmatrix}1&0\\ 4&1\end{pmatrix}, to 44 singular points around each of which the monodromy is (1011)\begin{pmatrix}1&0\\ 1&1\end{pmatrix}. in the literature Such singularities are called focus-focus and are the most standard examples of singularities for ℤ\mathbb{Z}-affine structures in dimension 2. Those arise for instance when considering the hyperkähler rotation of a generic elliptic K3 surface f:S⟶ℂ​ℙ1f:S\longrightarrow\mathbb{C}\mathbb{P}^{1}, with ff an elliptic fibration: the hyperkähler rotation ShkS^{\HK} is a complex surface with same underlying topological space as SS, and hence comes with a map fhk:Shk⟶𝕊2f^{\HK}:S^{\HK}\longrightarrow\mathbb{S}^{2}, induced by ff at the level of topological spaces. The map fhkf^{\HK} is no longer holomorphic, but is a symplectic torus fibration inducing a ℤ\mathbb{Z}-affine structure with 24 focus-focus singularities on 𝕊2\mathbb{S}^{2} and acting as an SYZ fibration for ShkS^{\HK}. We refer the reader to [GW00] for more details.

Let ee be an edge of Sk⁡(𝒳)\Sk(\mathscr{X}), let Ce=De1∩De2C_{e}=D_{e_{1}}\cap D_{e_{2}} be the corresponding stratum curve in 𝒳k\mathscr{X}_{k}. We recall that as the degree four polynomial F4F_{4} is generic, CeC_{e} contains four singular points p1,…,p4p_{1},\ldots,p_{4} of 𝒳\mathscr{X}, which are ordinary double points. Around each pip_{i}, 𝒳\mathscr{X} is étale locally of the form {xy=wt}⊂𝔸R3\{xy=wt\}\subset\mathbb{A}^{3}_{R}, with xx and yy being local equations for De1D_{e_{1}} and De2D_{e_{2}} away from pip_{i}. Blowing-up the singular point pip_{i} yields an exceptional divisor E≃ℙ1×ℙ1E\simeq\mathbb{P}^{1}\times\mathbb{P}^{1}. Contracting one or the other ruling of EE, we obtain two distinct small resolutions of 𝒳/Ce^\widehat{\mathscr{X}_{/C_{e}}} around pip_{i}, respectively with an exceptional curve inside De1D_{e_{1}} or De2D_{e_{2}}.

For j∈{0,…,4}j\in\{0,\ldots,4\}, we denote by 𝒳e,j\mathscr{X}_{e,j} the following small resolution of 𝒳/Ce^\widehat{\mathscr{X}_{/C_{e}}}: around pip_{i} for i⩽ji\leqslant j we consider the small resolution such that the exceptional curve over pip_{i} lies in De1D_{e_{1}}, while for i>ji>j the small resolution such that the exceptional curves lie in De2D_{e_{2}}. The gluing of these local small resolutions is done in the étale topology, so that in general the obtained models are no longer schemes but only algebraic spaces. Nevertheless, 𝒳e,j\mathscr{X}_{e,j} is dominated by 𝒵\mathscr{Z} (defined in Section 3.3.2) and still induces a Berkovich retraction ρ𝒳e,j\rho_{\mathscr{X}_{e,j}}, as described in Section 5. In particular, by Proposition 5.0.4, ρ𝒳e,j\rho_{\mathscr{X}_{e,j}} is an affinoid torus fibration over Star⁡(τCe)\Star(\tau_{C_{e}}).

We construct the following continuous retraction

ρ¯:Xan→ρ𝒵Sk⁡(𝒵)→ρ¯′Sk⁡(X)=Sk⁡(𝒳),\overline{\rho}:X^{\an}\xrightarrow{\rho_{\mathscr{Z}}}\Sk(\mathscr{Z})\xrightarrow{\overline{\rho}^{\prime}}\Sk(X)=\Sk(\mathscr{X}),

where ρ𝒵\rho_{\mathscr{Z}} is the Berkovich retraction onto the skeleton Sk⁡(𝒵)\Sk(\mathscr{Z}), and ρ¯′\overline{\rho}^{\prime} is a retraction of the 24 wings of Sk⁡(𝒵)\Sk(\mathscr{Z}) to the sphere given as follows. We fix four distinct, ordered, interior points ae,1,…,ae,4a_{e,1},\ldots,a_{e,4} of each edge ee. Then the map ρ¯′\overline{\rho}^{\prime} on the wing WiW_{i} attached to ee is defined as the map ρ′\rho^{\prime} of Section 3.3.2, setting ae=ae,ia_{e}=a_{e,i}, for each i∈{1,…,4}.i\in\{1,\ldots,4\}.

Proposition 3.3.3.

The map ρ¯\overline{\rho} is an affinoid torus fibration away from the 2424 points ae,ia_{e,i}. Furthermore, the monodromy of the ℤ\mathbb{Z}-affine structure induced by ρ¯\overline{\rho}, around each singular point, is SL2⁡(ℤ)\SL_{2}(\mathbb{Z})-conjugate to

Tρ¯=(1011).T_{\overline{\rho}}=\left(\begin{matrix}1&0\\ 1&1\end{matrix}\right).
Proof.

Over Int​(τ)\textrm{Int}(\tau) of any 22-dimensional face τ⊂Sk⁡(𝒳)\tau\subset\Sk(\mathscr{X}), ρ¯\overline{\rho} is equal to ρ𝒵\rho_{\mathscr{Z}}, hence is an affinoid torus fibration. Around any vertex vDv_{D}, ρ¯\overline{\rho} is equal to ρ𝒳i​j​k\rho_{\mathscr{X}_{ijk}} for any triple such that D≠Di,Dj,DkD\neq D_{i},D_{j},D_{k}. It follows from Section 3.3.1 that ρ¯\overline{\rho} is an affinoid torus fibration around vDv_{D}. We denote by p0+p∞p_{0}+p_{\infty} the boundary of CeC_{e}, with p0=Ce∩Di0p_{0}=C_{e}\cap D_{i_{0}} and p∞=Ce∩Di∞p_{\infty}=C_{e}\cap D_{i_{\infty}}; we write ae,0=vDe1a_{e,0}=v_{D_{e_{1}}} and ae,5=vDe2a_{e,5}=v_{D_{e_{2}}}, and denote by (⋅,⋅)(\cdot,\cdot) the open segment joining two points. Then, for j∈{0,…,4}j\in\{0,\ldots,4\}, ρ¯\overline{\rho} is equal to ρ𝒳e,j\rho_{\mathscr{X}_{e,j}} over Int​(τp0)∪Int​(τp∞)∪(ae,j,ae,j+1)\textrm{Int}(\tau_{p_{0}})\cup\textrm{Int}(\tau_{p_{\infty}})\cup(a_{e,j},a_{e,j+1}), thus is an affinoid torus fibration. We conclude that ρ¯\overline{\rho} is an affinoid torus fibration away from the points ae,ia_{e,i} for i∈{1,…,4}i\in\{1,\ldots,4\} .

For a singular point ae,ia_{e,i}, we consider a loop γ\gamma around it and contained in Int​(τp0)∪Int​(τp∞)∪(ae,i−1,ae,i)∪(ae,i,ae,i+1)\textrm{Int}(\tau_{p_{0}})\cup\textrm{Int}(\tau_{p_{\infty}})\cup(a_{e,i-1},a_{e,i})\cup(a_{e,i},a_{e,i+1}). We apply Corollary 3.2.4 to compute the monodromy along γ\gamma: the numbers be1,𝒳e,i−1b_{e_{1},\mathscr{X}_{e,i-1}} and be1,𝒳e,ib_{e_{1},\mathscr{X}_{e,i}} differ by 11, as the model 𝒳e,i\mathscr{X}_{e,i} has an additional exceptional curves in De1D_{e_{1}} with respect to 𝒳e,i−1\mathscr{X}_{e,i-1}. Therefore, we obtain

Tρ¯​(γae,i)\displaystyle T_{\overline{\rho}}(\gamma_{a_{e,i}}) =(10be1,𝒳e,i−1−be1,𝒳e,i1)\displaystyle=\left(\begin{matrix}1&0\\ b_{e_{1},\mathscr{X}_{e,i-1}}-b_{e_{1},\mathscr{X}_{e,i}}&1\end{matrix}\right)
=(103−(i−1)−(3−i)1)\displaystyle=\left(\begin{matrix}1&0\\ 3-(i-1)-(3-i)&1\end{matrix}\right)
=(1011)\displaystyle=\left(\begin{matrix}1&0\\ 1&1\end{matrix}\right)
γae,2\gamma_{a_{e,2}}ae,1a_{e,1}ae,2a_{e,2}ae,3a_{e,3}ae,4a_{e,4}v2=ve2=ae,5v_{2}=v_{e_{2}}=a_{e,5}vi∞=v4v_{i_{\infty}}=v_{4}v3=vi0v_{3}=v_{i_{0}}v1=ve1=ae,0v_{1}=v_{e_{1}}=a_{e,0}

with respect to the basis (vDi0,vDe1)(v_{D_{i_{0}}},v_{D_{e_{1}}}) and origin vDe2v_{D_{e_{2}}}. ∎

Note that for a generic family of quartic surfaces X/KX/K, the metric aspects of the Kontsevich-Soibelman conjecture suggest that there should exist a distinguished singular affine structure on Sk⁡(X)\Sk(X) (coming from the Gromov-Hausdorff limit of the family), and hence a canonical choice of interior points ae,ia_{e,i} for each edge. To the authors’ knowledge there does not exist a way to produce such a canonical set of ae,ia_{e,i}’s using non-archimedean techniques.

3.3.4 Collision of singularities

In Section 3.3.2, the retraction ρ′\rho^{\prime} depends on the choice of the points aea_{e}; the same holds for the induced integral affine structure, whose singular locus consists indeed of the points aea_{e}. Moving a point aea_{e} in the interior of the edge ee affects the location of the singular points, but it does not change the monodromy around the point (see Section 3.3.3). We observe now, in two examples, what happens if we let a point aea_{e} move to a vertex of Sk⁡(𝒳)\Sk(\mathscr{X}). We write ρ=ρae\rho=\rho_{a_{e}} to emphasize the dependency on the choice of singular points.

When all the points aea_{e} lie in the interior of the respective edges as in Section 3.3.2, ρae=ρ𝒳i​k​h\rho_{a_{e}}=\rho_{\mathscr{X}_{ikh}} on Star⁡(vj)′\Star(v_{j})^{\prime}, the induced integral affine structure is smooth at vjv_{j} and such that

Tρ​(γae24)\displaystyle T_{\rho}(\gamma_{a_{e_{24}}}) =(10b4,𝒳123−b4,𝒳1341)=(1041)\displaystyle=\left(\begin{matrix}1&0\\ b_{4,\mathscr{X}_{123}}-b_{4,\mathscr{X}_{134}}&1\end{matrix}\right)=\left(\begin{matrix}1&0\\ 4&1\end{matrix}\right)
in the basis (vDk,vDh) and origin vDj,\displaystyle\text{in the basis $(v_{D_{k}},v_{D_{h}})$ and origin $v_{D_{j}}$},
Tρ​(γae23)\displaystyle T_{\rho}(\gamma_{a_{e_{23}}}) =(10b3,𝒳124−b3,𝒳1341)−1=(1041)−1=(10−41)\displaystyle=\left(\begin{matrix}1&0\\ b_{3,\mathscr{X}_{124}}-b_{3,\mathscr{X}_{134}}&1\end{matrix}\right)^{-1}=\left(\begin{matrix}1&0\\ 4&1\end{matrix}\right)^{-1}=\left(\begin{matrix}1&0\\ -4&1\end{matrix}\right)
in the basis (vDh,vDk) and origin vDj.\displaystyle\text{in the basis $(v_{D_{h}},v_{D_{k}})$ and origin $v_{D_{j}}$}.
v2=vjv_{2}=v_{j}vh=v4v_{h}=v_{4}v3=vkv_{3}=v_{k}v1=viv_{1}=v_{i}γae24\gamma_{a_{e_{24}}}γae23\gamma_{a_{e_{23}}}

When ae24a_{e_{24}} collides with the vertex vjv_{j}, ρae\rho_{a_{e}} on Star⁡(vj)′\Star(v_{j})^{\prime} is equal to ρ𝒳i​k​j\rho_{\mathscr{X}_{ikj}}, the integral affine structure is singular at vjv_{j} with

Tρ𝒳i​k​j​(γj)\displaystyle T_{\rho_{\mathscr{X}_{ikj}}}(\gamma_{j}) =(1041)=Tρ​(γae24)\displaystyle=\left(\begin{matrix}1&0\\ 4&1\end{matrix}\right)=T_{\rho}(\gamma_{a_{e_{24}}})
in the basis (vDk,vDh) and origin vDj.\displaystyle\text{in the basis $(v_{D_{k}},v_{D_{h}})$ and origin $v_{D_{j}}$}.
v2=vjv_{2}=v_{j}vh=v4v_{h}=v_{4}v3=vkv_{3}=v_{k}v1=viv_{1}=v_{i}γj\gamma_{j}γae23\gamma_{a_{e_{23}}}

When both ae24a_{e_{24}} and ae23a_{e_{23}} collide with vjv_{j}, ρae\rho_{a_{e}} on Star⁡(vj)′\Star(v_{j})^{\prime} is equal to ρ𝒳i​j​k\rho_{\mathscr{X}_{ijk}}, the integral affine structure is singular at vjv_{j} with

Tρ𝒳i​j​k​(γj)\displaystyle T_{\rho_{\mathscr{X}_{ijk}}}(\gamma_{j}) =(−15−441)\displaystyle=\left(\begin{matrix}-15&-4\\ 4&1\end{matrix}\right)
=(1−401)​(1041)=Tρ​(γae23)​Tρ​(γae24)\displaystyle=\left(\begin{matrix}1&-4\\ 0&1\end{matrix}\right)\left(\begin{matrix}1&0\\ 4&1\end{matrix}\right)=T_{\rho}(\gamma_{a_{e_{23}}})T_{\rho}(\gamma_{a_{e_{24}}})
in the basis (vDk,vDh) and origin vDj.\displaystyle\text{in the basis $(v_{D_{k}},v_{D_{h}})$ and origin $v_{D_{j}}$}.
v2=vjv_{2}=v_{j}vh=v4v_{h}=v_{4}v3=vkv_{3}=v_{k}v1=viv_{1}=v_{i}γj\gamma_{j}

The computations above suggest that the singularities and the monodromy representation induced by the non-archimedean SYZ fibration ρ𝒳i​j​k\rho_{\mathscr{X}_{ijk}} can be viewed respectively as a collision of singular points and a product of monodromies induced by the ρae\rho_{a_{e}} when the aea_{e}’s collide. The affine structure induced by ρ\rho turns out to be more symmetric and simpler, as all the singular points have the same monodromy and are of focus-focus type.

Original mathematics by the credited authors. Source-backed reader collection; mathematical self-containment is not assessed.