ScalingStacks

3. Some examples [04SF]

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. Some examples

3.1. Riemann surfaces

Let SS be a closed Riemann surface of genus g>1g>1. It is well-known that SS admits a decomposition into pairs-of-pants. Namely, there exist 3​g−33g-3 disjoint embedded circles Cj⊂SC_{j}\subset S such that S∖⋃j=13​g−3CjS\smallsetminus\bigcup\limits_{j=1}^{3g-3}C_{j} is a disjoint union of 2​g−22g-2 copies of the pair-of-pants PP. The pair-of-pants surface PP is homeomorphic to the Riemann sphere ℂ​ℙ1{\mathbb{C}}{\mathbb{P}}^{1} punctured in three points.

To such a decomposition we associate a graph Γ\Gamma. The vertices of Γ\Gamma correspond to the pairs-of-pants while the edges correspond to the circles CjC_{j}. Each edge joins the vertices corresponding to the adjacent pairs-of-pants.

There exists a fibration π:S→Γ\pi:S\to\Gamma such that the circles CjC_{j} are inverse images of the midpoints of the edges of Γ\Gamma. Such fibration is canonically associated to our decomposition into pairs-of-pants. To construct it we fiber each individual pair-of-pants over a tripod graph as pictured on the left-hand-side of Figure 5.

Refer to caption       Refer to caption

Figure 5. Circle fibrations on a pair-of-pants and on a surface with a pair-of-pants decomposition.

3.2. The elliptic curve and the K3-surface

Here we consider the well-known fibrations of the elliptic curve and the K3-surface.

Let ℂ​E\mathbb{C}E be an elliptic curve, i.e. a Riemann surface of genus 1. Since ℂ​E\mathbb{C}E is topologically a torus, there is a trivial S1S^{1}-fibration λE:ℂ​E→S1\lambda_{E}:\mathbb{C}E\to S^{1}.

Suppose that the elliptic curve ℂ​E⊂ℂ​TΔ\mathbb{C}E\subset\mathbb{C}T_{\Delta} is presented as a curve in a toric surface ℂ​TΔ\mathbb{C}T_{\Delta}, where Δ\Delta is the Newton polygon of a polynomial defining ℂ​E\mathbb{C}E. By the genus formula (see [8]), Int⁡Δ\operatorname{Int}\Delta contains a unique lattice point. By Proposition 1.10 a dual Δ\Delta-complex is homotopy equivalent to a circle. It is easy to see that the fibration from Theorem 1’ coincides up to homotopy with the trivial S1S^{1}-fibration ℂ​E≈S1×S1→S1\mathbb{C}E\approx S^{1}\times S^{1}\to S^{1}.

Another famous fibration λK:ℂ​K→S2\lambda_{K}:\mathbb{C}K\to S^{2} has the K3-surface ℂ​K\mathbb{C}K as its total space. All its fibers, except for 24 of them are Lagrangian tori.

Suppose that the polygon Δ\Delta has exactly one interior lattice point. Then, by Khovanskii’s formula [8], the zero locus ℂ​K\mathbb{C}K of a generic polynomial with the Newton polygon Δ\Delta is a K3-surface. A dual Δ\Delta-complex is homotopy equivalent to a sphere S2S^{2} by Proposition 1.10.

Again, the fibration λ\lambda can be deformed to a fibration like λK\lambda_{K} by so-called shelling of Π¯\bar{\Pi} 11 1 A higher-dimensional version of such deformation will be the subject for a future paper..

In higher dimensions, if Δ\Delta is a non-singular polyhedron with a unique interior lattice point, then the corresponding hypersurface V⊂ℂ​TΔV\subset\mathbb{C}T_{\Delta} is a smooth Calabi-Yau manifold. Singular torus fibrations V→SnV\to S^{n} were constructed by Zharkov [17]. Ruan [13] noted that such fibrations can be made Lagrangian.

Theorem 1’ constructs in this case a stratified torus fibration over a polyhedral complex homotopy equivalent to SnS^{n}.

3.3. Hyperplanes in the projective space

This is a fundamental example for the main theorems. Let H={z1+⋯+zn+1+1=0}⊂ℂℙn+1H=\{z_{1}+\dots+z_{n+1}+1=0\}\subset{\mathbb{C}}{\mathbb{P}}^{n+1} be a hyperplane. Its toric part H∘=H∩(ℂ∗)n+1H^{\circ}=H\cap(\mathbb{C}^{*})^{n+1} is an open pair-of-pants.

Let Log\operatorname{Log} be the moment map for (ℂ∗)n+1(\mathbb{C}^{*})^{n+1} (see (3)).

Lemma 3.1.

Σn⊂Log⁡(H∘)\Sigma_{n}\subset\operatorname{Log}(H^{\circ}).

Proof.

By [12] Σn\Sigma_{n} is a spine of the amoeba Log⁡(H∘)\operatorname{Log}(H^{\circ}) and, therefore, its subset. The lemma can alternatively be verified by writing explicit inequalities defining Log⁡(H∘)\operatorname{Log}(H^{\circ}). ∎

The complement ℝn+1∖Σn\mathbb{R}^{n+1}\smallsetminus\Sigma_{n} consists of n+2n+2 components. Each component is the region where one of the functions 0,x1,…,xn+10,x_{1},\dots,x_{n+1} is maximal. In the component corresponding to xjx_{j} we consider the foliation into straight lines parallel to the gradient of xjx_{j} (the jjth basis vector). In the component corresponding to 00 we consider the foliation into straight lines parallel to (1,…,1)(1,\dots,1). These foliations glue to a singular foliation ℱ′\mathcal{F}^{\prime} which has singularities at Σn\Sigma_{n}.

Refer to caption

Figure 6. The amoeba Log⁡(H∘)\operatorname{Log}(H^{\circ}) together with the foliation ℱ′\mathcal{F}^{\prime} and its deformation ℱ\mathcal{F}.

It is easy to smooth out ℱ′\mathcal{F}^{\prime} (in a symmetric way with respect to the homogeneous coordinates permutations) at the open nn-cells of Σn\Sigma_{n} (see Figure 6). However, the singularities at the smaller-dimensional cells are essential. The leaves passing through an open (n−k)(n-k)-cell are homeomorphic to the cone over k+2k+2 points.

We denote the resulting foliation with ℱ\mathcal{F}. The foliation ℱ\mathcal{F} is a singular fibration and defines the projection πℱ:ℝn+1→Σn\pi_{\mathcal{F}}:\mathbb{R}^{n+1}\to\Sigma_{n}.

The following statement is a key lemma in the proof of the main theorems of this paper.

Lemma 3.2.

The composition

λH=πℱ∘Log:H∘→Σn\lambda_{H}=\pi_{\mathcal{F}}\circ\operatorname{Log}:H^{\circ}\to\Sigma_{n}

is a stratified TnT^{n}-fibration in the sense of Definition 7. It satisfies to all conclusions of Theorem 3 except for the third one. The fibration λH\lambda_{H} can be deformed so that the third condition will also hold.

The proof of this lemma occupies the rest of this subsection.

To figure out the fibers of λH\lambda_{H} we need to understand the critical points of Log|H∘\operatorname{Log}|_{H^{\circ}}. Following [6] and [11] for a hypersurface V∘⊂(ℂ∗)n+1V^{\circ}\subset(\mathbb{C}^{*})^{n+1} we define the logarithmic Gauss map

γ:V∘→ℂ​ℙn\gamma:V^{\circ}\to{\mathbb{C}}{\mathbb{P}}^{n}

by taking the composition of a branch of a holomorphic logarithm of each coordinate with the conventional Gauss map. This produces the following formula

γ(z1,…,zn+1)=[z1∂f∂z1:…:zn+1∂f∂zn+1],\gamma(z_{1},\dots,z_{n+1})=[z_{1}\frac{\partial f}{\partial z_{1}}:\dots:z_{n+1}\frac{\partial f}{\partial z_{n+1}}],

where ff is the polynomial defining V∘V^{\circ}.

Note that the Newton polyhedron of zj​∂f∂zjz_{j}\frac{\partial f}{\partial z_{j}} coincides with the Newton polyhedron Δ\Delta of ff. Therefore, by Kouchnirenko’s formula [9], deg⁡γ=(n+1)!​Vol⁡Δ\deg\gamma=(n+1)!\operatorname{Vol}\Delta. In particular, if V∘=H∘V^{\circ}=H^{\circ} then deg⁡γ=1\deg\gamma=1.

Lemma 3.3 (cf. Lemma 3 of [11]).

The set of critical points of Log|V∘\operatorname{Log}|_{V^{\circ}} coincides with γ−1​(ℝ​Pn)\gamma^{-1}(\mathbb{R}P^{n}).

Proof.

Let z∈V∘z\in V^{\circ} and let ℒ​o​g\mathcal{L}og be a branch of a holomorphic logarithm (z1,…,zn+1)↦(log⁡(z1),…,log⁡(zn+1))(z_{1},\dots,z_{n+1})\mapsto(\log(z_{1}),\dots,\log(z_{n+1})) defined in a neighborhood of zz. The point zz is critical for Log|V∘\operatorname{Log}|_{V^{\circ}} iff V∘V^{\circ} and the orbit of the real torus TnT^{n} are not transversal at zz. But ℒ​o​g\mathcal{L}og takes the tangent space to an orbit of TnT^{n} to a translate of i​ℝn+1i\mathbb{R}^{n+1} in ℂn+1\mathbb{C}^{n+1}.

Therefore, zz is critical iff ℒ​o​g​(Tz​V∘)\mathcal{L}og(T_{z}V^{\circ}) contains at least nn purely imaginary vectors which is, in turn, equivalent to γ⁡(z)∈ℝ​Pn\gamma(z)\in\mathbb{R}P^{n}. ∎

Corollary 3.4.

The set of critical points of Log|H∘\operatorname{Log}|_{H^{\circ}} coincides with the real locus ℝ​H∘\mathbb{R}H^{\circ} of H∘H^{\circ} (i.e. with the set of real solutions of z1+⋯+zn+1+1=0z_{1}+\dots+z_{n+1}+1=0).

Proof.

Note that, since H∘H^{\circ} is defined over ℝ\mathbb{R}, we have γ⁡(ℝ​H∘)⊂ℝ​ℙn\gamma(\mathbb{R}H^{\circ})\subset{\mathbb{R}}{\mathbb{P}}^{n}. Note that γ\gamma extends to a map H→ℂ​ℙnH\to{\mathbb{C}}{\mathbb{P}}^{n} which is an isomorphism, since deg⁡γ=1\deg\gamma=1. ∎

Corollary 3.5.

The locus 𝒟⊂Log⁡(H∘)\mathcal{D}\subset\operatorname{Log}(H^{\circ}) of critical values of Log|H∘\operatorname{Log}|_{H^{\circ}} is an immersed manifold transverse to the foliation ℱ\mathcal{F}.

Proof.

The map Log|ℝ​H∘:ℝH∘→𝒟⊂ℝn+1\operatorname{Log}|_{\mathbb{R}H^{\circ}}:\mathbb{R}H^{\circ}\to\mathcal{D}\subset\mathbb{R}^{n+1} is an immersion since the map Log|(ℝ∗)n+1:(ℝ∗)n+1→ℝn+1\operatorname{Log}|_{(\mathbb{R}^{*})^{n+1}}:(\mathbb{R}^{*})^{n+1}\to\mathbb{R}^{n+1} is an immersion (it is a trivial 2n+12^{n+1}-covering of ℝn+1\mathbb{R}^{n+1}).

To see the transversality we recall the definition of the foliation ℱ′\mathcal{F}^{\prime}. For each component of ℝn+1∖Σn\mathbb{R}^{n+1}\smallsetminus\Sigma_{n} the foliation ℱ′\mathcal{F}^{\prime} is parallel to a vector v→\stackrel{{\scriptstyle\to}}{{v}} normal to a facet of the Newton polyhedron of H∘H^{\circ}. Therefore, any hyperplane in the image γ⁡(ℝ​H∘)\gamma(\mathbb{R}H^{\circ}) is transverse to v→\stackrel{{\scriptstyle\to}}{{v}}. Furthermore, hyperplanes close to being parallel to v→\stackrel{{\scriptstyle\to}}{{v}} are close to the hyperplane in ℂ​ℙn+1{\mathbb{C}}{\mathbb{P}}^{n+1} corresponding to this facet and therefore are far from the given component of ℝn+1∖Σn\mathbb{R}^{n+1}\smallsetminus\Sigma_{n}. Thus the result ℱ\mathcal{F} of smoothing is also transverse to 𝒟\mathcal{D} and the angle between them in ℝn+1\mathbb{R}^{n+1} is separated from 0. ∎

Note that πℱ\pi_{\mathcal{F}} is a stratified [−1,1][-1,1]-fibration. Thus, the transversality of 𝒟\mathcal{D} and ℱ\mathcal{F} implies that λH\lambda_{H} is a stratified fibration for Σn\Sigma_{n}. We need to show that the restriction of πF\pi_{F} to open nn-cells of Σ\Sigma is a torus fibration.

Consider a point x=(−t,…,−t,0)x=(-t,\dots,-t,0) for a large t>0t>0. Note that 𝒟\mathcal{D} is almost horizontal near xx. Thus the fiber of λH\lambda_{H} over xx is diffeomorphic to the fiber FF of a composition of Log|H∘\operatorname{Log}|_{H^{\circ}} and the linear projection onto the first nn coordinates. Note that the map F→TnF\to T^{n} obtained by taking the arguments of the first nn coordinates is a diffeomorphism. Recall that H∘H^{\circ} is given by the equation z1+⋯+zn+1+1=0z_{1}+\dots+z_{n+1}+1=0. The absolute values of the coordinates z1,…,znz_{1},\dots,z_{n} are fixed. For any value of their argument we take zn+1=1−z1−⋯−znz_{n+1}=1-z_{1}-\dots-z_{n} to get the unique point from FF corresponding to this choice of the arguments. Since |z1|,…,|zn||z_{1}|,\dots,|z_{n}| are small zn+1≠0z_{n+1}\neq 0.

We verify the conclusions of Theorem 3 item-by-item. The first and the last conclusions are vacuous in this case, since Σn\Sigma_{n} (and, therefore, Σ¯n\bar{\Sigma}_{n} as well) is contractible. The second one holds since H∘H^{\circ} is itself an open pair-of-pants.

To make the third conclusion true we have to modify λH\lambda_{H} a little. The fiber FF is not Lagrangian, but it is close to a Lagrangian torus Λ={|zj|=const,j=1,…,n,zn+1=−1}\Lambda=\{|z_{j}|=\operatorname{const},\ j=1,\dots,n,\ z_{n+1}=-1\}. We can deform H∘H^{\circ} a little in a neighborhood of Log−1⁡(x)\operatorname{Log}^{-1}(x) to make it intersect the fiber of πℱ∘Log\pi_{\mathcal{F}}\circ\operatorname{Log} along Λ\Lambda. Therefore, FF is Lagrangian for a nearby symplectic structure. By Moser’s trick (see e.g. [2]) there exists a self-diffeomorphism hh of H∘H^{\circ} constant outside of a neighborhood of Log−1⁡(x)\operatorname{Log}^{-1}(x) and taking one symplectic structure to another. We redefine λ\lambda as λ∘h\lambda\circ h. This ensures a Lagrangian fiber over one of the (n+22)\begin{pmatrix}n+2\\ 2\end{pmatrix} open nn-cells of Σn\Sigma_{n}. We do the same for all other nn-cells.

3.4. A localization Qn⊂(ℂ∗)n+1Q^{n}\subset(\mathbb{C}^{*})^{n+1} of the standard hyperplane

The toric part H∘⊂(ℂ∗)n+1H^{\circ}\subset(\mathbb{C}^{*})^{n+1} of a hyperplane from 3.3 is a nice embedding of 𝒫n\mathcal{P}_{n} to (ℂ∗)n+1(\mathbb{C}^{*})^{n+1}. However for our purposes it is convenient to modify it in a neighborhood of infinity to get a different submanifold QnQ^{n} which is better suited for gluing.

Note that the symmetric group Sn+2S_{n+2} acts on (ℂ∗)n+1(\mathbb{C}^{*})^{n+1} by interchanging the functions z1,…,zn+1,1z1​…​zn+1z_{1},\dots,z_{n+1},\frac{1}{z_{1}\dots z_{n+1}}. This action is inherited from the action of Sn+2S_{n+2} on ℂ​ℙn+1{\mathbb{C}}{\mathbb{P}}^{n+1} interchanging the homogeneous coordinates since (ℂ∗)n+1⊂ℂ​ℙn+1(\mathbb{C}^{*})^{n+1}\subset{\mathbb{C}}{\mathbb{P}}^{n+1} is invariant. The hyperplane H∘H^{\circ} is invariant with respect to this action.

Proposition 3.6.

There exists a proper submanifold Qn⊂(ℂ∗)n+1Q^{n}\subset(\mathbb{C}^{*})^{n+1} such that

  • •

    QnQ^{n} is embedded to (ℂ∗)n+1(\mathbb{C}^{*})^{n+1} symplectically, i.e. so that the restriction of the form (2) to QnQ^{n} is a symplectic form.

  • •

    QnQ^{n} is isotopic to H∘H^{\circ} in (ℂ∗)n+1(\mathbb{C}^{*})^{n+1}.

  • •

    The composition πℱ∘Logt|Qn\pi_{\mathcal{F}}\circ\operatorname{Log}_{t}|_{Q^{n}} is a stratified TnT^{n}-fibration that satisfies to all hypotheses of Theorem 3.

  • •

    the closure Qn¯\bar{Q^{n}} of QnQ^{n} in ℂ​ℙn+1⊃(ℂ∗)n+1{\mathbb{C}}{\mathbb{P}}^{n+1}\supset(\mathbb{C}^{*})^{n+1} is a smooth manifold isotopic to HH.

  • •

    QnQ^{n} is invariant with respect to the action of the symmetric group Sn+2S_{n+2} on (ℂ∗)n+1(\mathbb{C}^{*})^{n+1} (see above).

  • •

    For a sufficiently large M>0M>0

    Qn∩(ℂ∗)−Mn+1=Qn−1×C−M∗,Q^{n}\cap(\mathbb{C}^{*})^{n+1}_{-M}=Q^{n-1}\times\\ C^{*}_{-M},

    where (ℂ∗)−Mn+1={(z1,…,zn+1)∈(ℂ∗)n+1​|log|​zn+1|<−M}(\mathbb{C}^{*})^{n+1}_{-M}=\{(z_{1},\dots,z_{n+1})\in(\mathbb{C}^{*})^{n+1}\ |\ \log|z_{n+1}|<-M\} and ℂ−M∗={z∈ℂ∗​|log|​z|<−M}\mathbb{C}^{*}_{-M}=\{z\in\mathbb{C}^{*}\ |\ \log|z|<-M\}. In particular, the intersection Qn∩(ℂ∗)−Mn+1Q^{n}\cap(\mathbb{C}^{*})^{n+1}_{-M} is invariant under a translation zn+1↦c​zn+1z_{n+1}\mapsto cz_{n+1}, 0<c<10<c<1.

Refer to caption

Figure 7. The amoeba of the localization QnQ^{n} of a hyperplane.
Proof.

We construct QnQ^{n} inductively by dimension nn. If n=0n=0 then H∘H^{\circ} is a point and Q0=H∘Q^{0}=H^{\circ}. Assume that QkQ^{k}, k<nk<n is already constructed. Consider the simplex

Δn(R)={x∈ℝn+1|−xj≤R,∑jxj≤R}.\Delta_{n}(R)=\{x\in\mathbb{R}^{n+1}\ |\ -x_{j}\leq R,\sum\limits_{j}x_{j}\leq R\}.

Each its kk-dimensional face is dual to a (n+1−k)(n+1-k)-cell of Σn\Sigma_{n}. Fix a sufficiently large number Rn>0R_{n}>0.

First we define Qn∩Log−1⁡(∂Δ⁡(Rn))Q^{n}\cap\operatorname{Log}^{-1}(\partial\Delta(R_{n})). Each kk-face of Δ⁡(Rn)\Delta(R_{n}) is contained in a unique affine kk-space AA in ℝn+1\mathbb{R}^{n+1}. Furthermore, the adjoint faces cut the polyhedron Δk−1​(Rn)⊂A\Delta_{k-1}(R_{n})\subset A. Thus we may identify AA with ℝk\mathbb{R}^{k} and, therefore, Log−1⁡(A)\operatorname{Log}^{-1}(A) with (ℂ∗)k(\mathbb{C}^{*})^{k}. By the induction assumption we already have Qk−1⊂(ℂ∗)k→ℝkQ^{k-1}\subset(\mathbb{C}^{*})^{k}\to\mathbb{R}^{k}. We define Qn∩Log−1⁡(∂Δ⁡(Rn))Q^{n}\cap\operatorname{Log}^{-1}(\partial\Delta(R_{n})) to be equal to the union of these QkQ^{k} for all faces of ∂Δ⁡(Rn)\partial\Delta(R_{n}). By the induction hypothesis (and since RnR_{n} was large enough) the choices over different faces agree.

Our next step is to extend QnQ^{n} to the complement of Log−1⁡(Δ⁡(Rn))\operatorname{Log}^{-1}(\Delta(R_{n})). For each face Δ′\Delta^{\prime} of ∂Δ⁡(Rn)\partial\Delta(R_{n}) consider its outer normal cone CΔ′⊂ℝn+1C_{\Delta^{\prime}}\subset\mathbb{R}^{n+1} (e.g. if Δ′\Delta^{\prime} is a facet then CΔ′C_{\Delta^{\prime}} is a ray). We define

Qn∩Log−1(Δ′+CΔ′)=⋃v→∈CΔ′ev→Qn∩Log−1(Δ′).Q^{n}\cap\operatorname{Log}^{-1}(\Delta^{\prime}+C_{\Delta^{\prime}})=\bigcup\limits_{\stackrel{{\scriptstyle\to}}{{v}}\in C_{\Delta^{\prime}}}e^{\stackrel{{\scriptstyle\to}}{{v}}}Q^{n}\cap\operatorname{Log}^{-1}(\Delta^{\prime}).

In other words, we span the region above the normal cone of a kk-face Δ′\Delta^{\prime} by the translates of the manifold QkQ^{k}.

We set Qn∩Log−1⁡(Δ⁡(Rn−1))=H∘∩Log−1⁡(Δ⁡(Rn−1))Q^{n}\cap\operatorname{Log}^{-1}(\Delta(R_{n}-1))=H^{\circ}\cap\operatorname{Log}^{-1}(\Delta(R_{n}-1)). By now we have defined QnQ^{n} everywhere, but Log−1⁡(Δ⁡(Rn)∖Δ⁡(Rn−1))\operatorname{Log}^{-1}(\Delta(R_{n})\smallsetminus\Delta(R_{n}-1)).

Consider a facet Δ′\Delta^{\prime} of ∂Δ⁡(Rn−1)\partial\Delta(R_{n}-1), e.g. the one sitting in the hyperplane A={xn+1=Rn−1}A=\{x_{n+1}=R_{n}-1\}. Since RnR_{n} is large enough, zn+1Rn−1z_{n+1}^{R_{n}-1} is small enough and the intersection H∘∩Log−1⁡(A)H^{\circ}\cap\operatorname{Log}^{-1}(A) is close enough to the zero set of z1+⋯+zn+1=0z_{1}+\dots+z_{n}+1=0. By the induction hypothesis this zero set can be deformed to Qn−1Q^{n-1}. We define Qn∩{Log|zn+1|=t}Q^{n}\cap\{\operatorname{Log}|z_{n+1}|=t\}, −Rn≤t≤−Rn+1-R_{n}\leq t\leq-R_{n}+1 using this deformation. We repeat the same procedure for all other facets of Δ⁡(Rn−1)\Delta(R_{n}-1). ∎

Denote Q¯n=Qn∩Log−1⁡(Δ⁡(Rn+1))\bar{Q}^{n}=Q^{n}\cap\operatorname{Log}^{-1}(\Delta(R_{n}+1)). This is “the kernel” of QnQ^{n} and is diffeomorphic to a closed pair-of-pants 𝒫n¯\bar{\mathcal{P}_{n}} (as a manifold with boundary and corners).

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