ScalingStacks

5 Piecewise smooth fibrations [04JN]

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

5 Piecewise smooth fibrations

It is now commonly accepted that to produce Lagrangian fibrations of the type described in §2 one should also allow piecewise smooth fibrations (cf. [6], [21], [29]). Here we present a simple way to produce local models of piecewise smooth Lagrangian fibrations. We suspect that models of the sort presented here are also implicit in Ruan’s fibrations but we have been unable to verify this. Our method is inspired by ideas of Gross [6], Goldstein [5] and Joyce [21].

Fibrations with torus symmetry.

Let (X,ω)(X,\omega) be a symplectic 2​n2n-manifold and let μ:(X,ω)→𝔱∗\mu:(X,\omega)\rightarrow\mathfrak{t}^{\ast} be the moment map of a Hamiltonian TkT^{k}-action. Let t∈μ⁡(X)t\in\mu(X) and let πt:μ−1​(t)→Xt\pi_{t}:\mu^{-1}(t)\rightarrow X_{t} be the projection modulo the TkT^{k} action. When tt is a regular value of μ\mu, XtX_{t} is a smooth manifold and the symplectic form ω\omega descends to a symplectic form ωt\omega_{t} on XtX_{t}. When tt is a critical value of μ\mu, XtX_{t} may be a singular space and ωt\omega_{t} will be only defined on the smooth part of XtX_{t}. The space (Xt,ωt)(X_{t},\omega_{t}) is the Marsden-Weinstein reduced space at tt.

Remark 5.1.

We shall denote by

ωℂm=i2​∑kd​zk∧d​z¯k\omega_{\mathbb{C}^{m}}=\frac{i}{2}\sum_{k}dz_{k}\wedge d\overline{z}_{k}

the standard symplectic structure on ℂm\mathbb{C}^{m} and ω0\omega_{0} will denote the reduced symplectic form of the reduced space (Xt,ωt)(X_{t},\omega_{t}) at time t=0t=0.

Goldstein [5] and Gross [6] used reduced spaces to construct TkT^{k}-invariant (special) Lagrangian fibrations. The following is a particular case of [6]Thm. 1.2:

Proposition 5.2.

Let TkT^{k} act effectively on XX, k≤n−1k\leq n-1. Suppose that there is a continuous map G:X→MG:X\rightarrow M to an (n−k)(n-k)-dimensional manifold MM such that G⁡(T⋅x)=G⁡(x)G(T\cdot x)=G(x) for all T∈TkT\in T^{k}. Suppose that for tt in a dense subset of μ⁡(X)\mu(X) the induced maps Gt:Xt→MG_{t}:X_{t}\rightarrow M have fibres that are Lagrangian with respect to ωt\omega_{t}. Then f:X→μ⁡(X)×Mf:X\rightarrow\mu(X)\times M given by:

f=(μ,G)f=(\mu,G) (23)

defines a TkT^{k}-invariant Lagrangian fibration.

When the TkT^{k}-action has fixed points, the construction of Proposition 5.2 will produce fibrations with interesting singular fibres. We will give some explicit examples shortly.

Remark 5.3.

In the extremal case when k=n−1k=n-1, constructing Lagrangian fibrations using Proposition 5.2 is very easy. In this situation, the reduced spaces XtX_{t} are two dimensional and every map Gt:Xt→ℝG_{t}:X_{t}\rightarrow\mathbb{R} with 11-dimensional level sets defines a Lagrangian fibration on XtX_{t}. In particular, any Tn−1T^{n-1}-invariant continuous map G:X→ℝG:X\rightarrow\mathbb{R} which, on each XtX_{t}, descends to a map GtG_{t} with 11-dimensional level sets can be used to construct Lagrangian fibrations. We will make much use of this fact later on.

The reduced geometry.

Consider the following S1S^{1} action on ℂn\mathbb{C}^{n}, with n≥2n\geq 2:

ei​θ​(z1,z2,z3,…,zn)=(ei​θ​z1,e−i​θ​z2,z3,…,zn).e^{i\theta}(z_{1},z_{2},z_{3},\ldots,z_{n})=(e^{i\theta}z_{1},e^{-i\theta}z_{2},z_{3},\ldots,z_{n}). (24)

This action is Hamiltonian with respect to ωℂn\omega_{\mathbb{C}^{n}}. Clearly it is singular along the 2​(n−2)2(n-2) dimensional symplectic submanifold Crit(μ)={z1=z2=0}\Crit(\mu)=\{z_{1}=z_{2}=0\}. The moment map is:

μ⁡(z1,…,zn)=|z1|2−|z2|22.\mu(z_{1},\ldots,z_{n})=\frac{|z_{1}|^{2}-|z_{2}|^{2}}{2}. (25)

The only critical value of μ\mu is t=0t=0 and Crit⁡(μ)⊂μ−1​(0)\Crit(\mu)\subset\mu^{-1}(0).

Now consider the map π¯\bar{\pi} as in Remark 2.5. Recall that π¯\bar{\pi} is given by

π¯:ℂn→ℝ×ℂn−1(z1,…,zn)↦(μ,z1​z2,z3,…,zn).\begin{array}[]{rll}\bar{\pi}:\mathbb{C}^{n}&\rightarrow&\mathbb{R}\times\mathbb{C}^{n-1}\\ (z_{1},\ldots,z_{n})&\mapsto&(\mu,z_{1}z_{2},z_{3},\ldots,z_{n}).\end{array} (26)

When restricted to ℂn−Crit⁡(μ)\mathbb{C}^{n}-\Crit(\mu), the above is an S1S^{1}-bundle onto (ℝ×ℂn−1)−π¯​(Crit⁡(μ))(\mathbb{R}\times\mathbb{C}^{n-1})-\bar{\pi}(\Crit(\mu)) with Chern class c1=1c_{1}=1. Let πt\pi_{t} be the restriction to μ−1​(t)\mu^{-1}(t) of the map

(z1,…,zn)↦(z1​z2,z3,…,zn).(z_{1},\ldots,z_{n})\mapsto(z_{1}z_{2},z_{3},\ldots,z_{n}). (27)

Then πt\pi_{t} can be used to identify the reduced space μ−1​(t)/S1\mu^{-1}(t)/S^{1} with ℂn−1\mathbb{C}^{n-1}. Under this identification, i.e. letting the coordinates u1=z1​z2u_{1}=z_{1}z_{2} and uj=zj+1u_{j}=z_{j+1} when 2≤j≤n−12\leq j\leq n-1, the reduced symplectic form ωt\omega_{t} can be written as:

ωt=i2​(12​t2+|u1|2​d​u1∧d​u¯1+∑j=2n−1d​uj∧d​u¯j).\omega_{t}=\frac{i}{2}\left(\frac{1}{2\sqrt{t^{2}+|u_{1}|^{2}}}\,du_{1}\wedge d\overline{u}_{1}+\sum_{j=2}^{n-1}\,du_{j}\wedge d\overline{u}_{j}\right). (28)

Clearly, away from t=0t=0, the reduced spaces are smooth manifolds.

On the other hand, at t=0t=0 the reduced form ω0\omega_{0} blows up along the hyperplane

Σ:=π0(Crit(μ))={u1=0},\Sigma:=\pi_{0}(\Crit(\mu))=\{u_{1}=0\},

so the reduced space (ℂn−1,ω0)(\mathbb{C}^{n-1},\omega_{0}) is singular. However, it was observed by Guillemin and Sternberg in [15], that it can be smoothed out, i.e. it can be identified with (ℂn−1,ωℂn−1)(\mathbb{C}^{n-1},\omega_{\mathbb{C}^{n-1}}). Indeed, the identification is given by the following

Γ0:(u1,u2,…,un−1)↦(u1|u1|,u2,…,un−1).\Gamma_{0}:(u_{1},u_{2},\ldots,u_{n-1})\mapsto\left(\frac{u_{1}}{\sqrt{|u_{1}|}},u_{2},\ldots,u_{n-1}\right). (29)

The map Γ0\Gamma_{0} is continuous, smooth away from u1=0u_{1}=0 and such that Γ0∗​ωℂn−1=ω0\Gamma_{0}^{\ast}\omega_{\mathbb{C}^{n-1}}=\omega_{0}. One can do more: one can identify all the reduced spaces with (ℂn−1,ωℂn−1)(\mathbb{C}^{n-1},\omega_{\mathbb{C}^{n-1}}) at once. Consider the map

Γt:(u1,u2,…,un−1)↦(u1|t|+t2+|u1|2,u2,…,un−1).\Gamma_{t}:(u_{1},u_{2},\ldots,u_{n-1})\mapsto\left(\frac{u_{1}}{\sqrt{|t|+\sqrt{t^{2}+|u_{1}|^{2}}}},u_{2},\ldots,u_{n-1}\right). (30)

One can verify that Γt\Gamma_{t} is a symplectomorphism between (ℂn−1,ωt)(\mathbb{C}^{n-1},\omega_{t}) and the standard symplectic space ℂn−1\mathbb{C}^{n-1}. However, this identification has the problem that, although continuous and smooth for fixed t∈ℝt\in\mathbb{R}, it is not smooth in tt when t=0t=0. In fact one can show that it cannot be otherwise.

A construction

We now illustrate a general method to construct piecewise smooth Lagrangian fibrations using Proposition 5.2 and the observations about the reduced geometry with respect to the S1S^{1} action as in (24).

Let Log:(ℂ∗)n−1→ℝn−1\Log:(\mathbb{C}^{\ast})^{n-1}\rightarrow\mathbb{R}^{n-1} be the map defined by

Log⁡(v1,…,vn−1)=(log⁡|v1|,…,log⁡|vn−1|).\Log(v_{1},\ldots,v_{n-1})=(\log|v_{1}|,\ldots,\log|v_{n-1}|). (31)

Clearly, the above map is a Lagrangian fibration with respect to the restriction of ωℂn−1\omega_{\mathbb{C}^{n-1}} to (ℂ∗)n−1(\mathbb{C}^{\ast})^{n-1}. Moreover, it defines a trivial Tn−1T^{n-1}-bundle over ℝn−1\mathbb{R}^{n-1}. Let the map

Φ:ℂn−1→ℂn−1\Phi:\mathbb{C}^{n-1}\rightarrow\mathbb{C}^{n-1}

be a smooth symplectomorphism of the standard ℂn−1\mathbb{C}^{n-1}. Let XtX_{t} be the open and dense subsets of (ℂn−1,ωt)(\mathbb{C}^{n-1},\omega_{t}) defined by

Xt=Γt−1∘Φ−1​((ℂ∗)n−1).X_{t}=\Gamma_{t}^{-1}\circ\Phi^{-1}\left((\mathbb{C}^{\ast})^{n-1}\right).

Denote, with slight abuse of notation,

Σ:={u1=0}∩X0.\Sigma:=\{u_{1}=0\}\cap X_{0}.

Then examples of maps Gt:Xt→ℝn−1G_{t}:X_{t}\rightarrow\mathbb{R}^{n-1} as in Proposition 5.2 can be defined by

Gt=Log∘Φ∘Γt.G_{t}=\Log\circ\Phi\circ\Gamma_{t}.

This clearly makes sense also when t=0t=0. It is also clear that, for all fixed t∈ℝt\in\mathbb{R}, GtG_{t} is a Lagrangian fibration with respect to the reduced symplectic form (28). We summarize this in the following:

Proposition 5.4.

Let Φ\Phi, XtX_{t} and GtG_{t} be as defined above. Let QQ be the map given by

Q⁡(t,u1,…,un−1)=(t,Gt​(u1,…,un−1)).Q(t,u_{1},\ldots,u_{n-1})=(t,G_{t}(u_{1},\ldots,u_{n-1})). (32)

Then QQ is defined on the dense open subset Y⊆ℝ×ℂn−1Y\subseteq\mathbb{R}\times\mathbb{C}^{n-1} defined by

Y={(t,u1,…,un−1)∈ℝ×ℂn−1|(u1,…,un−1)∈Xt}.Y=\{(t,u_{1},\ldots,u_{n-1})\in\mathbb{R}\times\mathbb{C}^{n-1}\ |\ (u_{1},\ldots,u_{n-1})\in X_{t}\}.

Letting π¯\bar{\pi} be as in (26) and

X=(π¯)−1​(Y)X=(\bar{\pi})^{-1}(Y)

with the standard symplectic form induced from ℂn\mathbb{C}^{n}, the map f:X→ℝnf:X\rightarrow\mathbb{R}^{n} given by

f=Q∘π¯f=Q\circ\bar{\pi}

is a piecewise smooth Lagrangian fibration of XX which fails to be smooth on the (2​n−1)(2n-1)-dimensional subspace μ−1​(0)∩X\mu^{-1}(0)\cap X.

It is clear that all the singular fibres of ff must lie in μ−1​(0)∩X\mu^{-1}(0)\cap X. In fact, the singular fibres are all the lifts of fibres of G0G_{0} in X0X_{0} which intersect Σ\Sigma. The topology of the singularity depends on the topology of this intersection. The discriminant locus of the fibration is therefore the set Δ⊂ℝn\Delta\subset\mathbb{R}^{n} given by

Δ={0}×(Log∘Φ∘Γ0​(Σ)).\Delta=\{0\}\times\left(\Log\circ\Phi\circ\Gamma_{0}(\Sigma)\right).

Given a point b=(0,b1,…,bn−1)∈Δb=(0,b_{1},\ldots,b_{n-1})\in\Delta, the fibre f−1​(b)f^{-1}(b) looks like S1×G0−1​(b1,…,bn−1)S^{1}\times G_{0}^{-1}(b_{1},\ldots,b_{n-1}) after the circles over all points in G0−1​(b1,…,bn−1)∩ΣG_{0}^{-1}(b_{1},\ldots,b_{n-1})\cap\Sigma have been collapsed to points (cf. Figure 6).

Examples

In the following examples we use the above construction with n=2n=2 or 33. Define the piecewise smooth map γ:ℂ2→ℂ\gamma:\mathbb{C}^{2}\rightarrow\mathbb{C} by

γ⁡(z1,z2)={z1​z2|z1|,when​μ​(z1,z2)≥0z1​z2|z2|,when​μ​(z1,z2)<0.\gamma(z_{1},z_{2})=\begin{cases}\frac{z_{1}z_{2}}{|z_{1}|},\quad\text{when}\ \mu(z_{1},z_{2})\geq 0\\ \\ \frac{z_{1}z_{2}}{|z_{2}|},\quad\text{when}\ \mu(z_{1},z_{2})<0.\end{cases} (33)

If πt\pi_{t} is the restriction of the map (27) to μ−1​(t)\mu^{-1}(t), then one can easily see that for all (z1,z2,z3)∈μ−1​(t)(z_{1},z_{2},z_{3})\in\mu^{-1}(t), the map Γt∘πt\Gamma_{t}\circ\pi_{t} is given by

Γt∘πt:(z1,z2,z3)↦(γ⁡(z1​z2),z3).\Gamma_{t}\circ\pi_{t}:(z_{1},z_{2},z_{3})\mapsto(\gamma(z_{1}z_{2}),z_{3}).

From Proposition 5.4, we see that Γt∘πt\Gamma_{t}\circ\pi_{t} can be twisted by a symplectomorphism Φ\Phi. The topology of the resulting fibration depends on how we choose Φ\Phi.

Example 5.5 (The amoeba).

Take as a symplectomorphism Φ\Phi the linear map

Φ⁡(u1,u2)=12​(u1−u2,u1+u2−2).\Phi(u_{1},u_{2})=\frac{1}{\sqrt{2}}\left(u_{1}-u_{2},u_{1}+u_{2}-\sqrt{2}\right). (34)

Then the fibration resulting from Proposition 5.4 can be written explicitly in the coordinates of the total space. We obtain:

f⁡(z1,z2,z3)=(12​(|z1|2−|z2|2),log⁡12​|γ−z3|,log⁡12​|γ+z3−2|),f(z_{1},z_{2},z_{3})=\left(\frac{1}{2}\left(|z_{1}|^{2}-|z_{2}|^{2}\right),\log\frac{1}{\sqrt{2}}\left|\gamma-z_{3}\right|,\log\frac{1}{\sqrt{2}}\left|\gamma+z_{3}-\sqrt{2}\right|\right), (35)

where γ\gamma is as in (33). It is not difficult to see that Φ∘Γ0\Phi\circ\Gamma_{0} sends Σ\Sigma to the surface in (ℂ∗)2(\mathbb{C}^{\ast})^{2} given by

Σ′={v1+v2+1=0},\Sigma^{\prime}=\{v_{1}+v_{2}+1=0\},

which is, topologically, a pair of pants. Then the discriminant locus is

Δ={0}×Log⁡(Σ′),\Delta=\{0\}\times\Log(\Sigma^{\prime}),

which has the shape in Figure 4. This example is topologically conjugate to the one in Example 2.9, before the surface Σ′\Sigma^{\prime} has been twisted. For the discussion of the topology of the fibres in this example we refer to Example 2.9.

In dimension n=2n=2 we have the following:

Example 5.6 (Stitched focus-focus).

Using Proposition 5.4 we can obtain the following piecewise smooth fibration:

f⁡(z1,z2)=(|z1|2−|z2|22,log⁡|γ⁡(z1,z2)+1|).f(z_{1},z_{2})=\left(\frac{|z_{1}|^{2}-|z_{2}|^{2}}{2},\,\log|\gamma(z_{1},z_{2})+1|\right). (36)

It is clearly well defined on X={(z1,z2)∈ℂ2|γ⁡(z1,z2)+1≠0}.X=\{(z_{1},z_{2})\in\mathbb{C}^{2}\ |\ \gamma(z_{1},z_{2})+1\neq 0\}. Observe that ff is topologically conjugate to a focus-focus fibration, hence to Example 2.6. The only singular fibre is f−1​(0)f^{-1}(0) and it is a pinched torus. The fibration fails to be smooth on μ−1​(0)\mu^{-1}(0). This example consists of the union of two smooth Lagrangian fibrations meeting along the “stitch”, μ−1​(0)\mu^{-1}(0). We study this kind of piecewise smoothness in detail in §6.

Notice that in this example we are in the extremal case of Proposition 5.2, i.e. the reduced spaces are 2-dimensional and Remark 5.3 applies. In particular, the second component of ff in (36) could be replaced by any T2T^{2} invariant function GG, i.e. depending on t=12​(|z1|2−|z2|2)t=\frac{1}{2}\left(|z_{1}|^{2}-|z_{2}|^{2}\right) and u1=z1​z2u_{1}=z_{1}z_{2}, subject to the condition that all maps GtG_{t} have 11-dimensional level sets. Using this idea it is easy to construct everywhere smooth fibrations, such as the one in Example 3.20 where G⁡(t,u1)=log⁡|u1+1|G(t,u_{1})=\log|u_{1}+1|. Of course the topology of the resulting fibration depends on the topology of the maps GtG_{t}.

We have an analogous model in dimension n=3n=3:

Example 5.7 (The leg).

Consider the following affine symplectomorphism of (ℂ2,ωℂ2)(\mathbb{C}^{2},\omega_{\mathbb{C}^{2}})

Φ:(u1,u2)↦(−u2,u1−1).\Phi:(u_{1},u_{2})\mapsto(-u_{2},u_{1}-1). (37)

The surface Σ\Sigma is sent by Φ∘Γ0\Phi\circ\Gamma_{0} to Σ′={v2+1=0}\Sigma^{\prime}=\{v_{2}+1=0\}. The amoeba of Σ′\Sigma^{\prime} is just a straight line. The resulting fibration ff is

f⁡(z1,z2,z3)=(|z1|2−|z2|22,log⁡|z3|,log⁡|γ⁡(z1,z2)−1|).f(z_{1},z_{2},z_{3})=\left(\frac{|z_{1}|^{2}-|z_{2}|^{2}}{2},\,\log|z_{3}|,\,\log|\gamma(z_{1},z_{2})-1|\right). (38)

The discriminant locus is {0}×ℝ×{0}⊂ℝ3\{0\}\times\mathbb{R}\times\{0\}\subset\mathbb{R}^{3}, a horizontal line in the plane {0}×ℝ2\{0\}\times\mathbb{R}^{2}. The fibration is a piecewise smooth version of the generic singular fibration in Example 4.5. Notice that this fibration is invariant under the Hamiltonian T2T^{2}-action

(ei​θ1,ei​θ2)⋅(z1,z2,z3)=(ei​θ1​z1,e−i​θ1​z2,e2​i​θ2​z3),(e^{i\theta_{1}},e^{i\theta_{2}})\cdot(z_{1},z_{2},z_{3})=(e^{i\theta_{1}}z_{1},\,e^{-i\theta_{1}}z_{2},\,e^{2i\theta_{2}}z_{3}), (39)

whose moment map is

(z1,z2,z3)↦(|z1|2−|z2|22,|z3|2).(z_{1},z_{2},z_{3})\mapsto\left(\frac{|z_{1}|^{2}-|z_{2}|^{2}}{2},\,|z_{3}|^{2}\right).

There are other choices of symplectomorphisms Φ\Phi giving piecewise smooth generic fibrations. Although not very different from the previous one, we will write other two for convenience, since we will need them in the next example. The first one is

Φ:(u1,u2)↦(u1−1,u2−2).\Phi:(u_{1},u_{2})\mapsto(u_{1}-1,u_{2}-\sqrt{2}). (40)

It gives the fibration

f⁡(z1,z2,z3)=(|z1|2−|z2|22,log⁡|γ⁡(z1,z2)−1|,log⁡|z3−2|),f(z_{1},z_{2},z_{3})=\left(\frac{|z_{1}|^{2}-|z_{2}|^{2}}{2},\,\log|\gamma(z_{1},z_{2})-1|,\,\log\left|z_{3}-\sqrt{2}\right|\right), (41)

whose discriminant locus is the vertical line {0}×{0}×ℝ⊂ℝ3\{0\}\times\{0\}\times\mathbb{R}\subset\mathbb{R}^{3}. Also in this case it is clearly invariant under a T2T^{2} action. The last choice of Φ\Phi is

Φ:(u1,u2)↦12​(u1−u2,u1+u2),\Phi:(u_{1},u_{2})\mapsto\frac{1}{\sqrt{2}}(u_{1}-u_{2},\,u_{1}+u_{2}), (42)

giving

f⁡(z1,z2,z3)=(|z1|2−|z2|22,log⁡|γ⁡(z1,z2)−z3|,log⁡|γ⁡(z1,z2)+z3|),f(z_{1},z_{2},z_{3})=\left(\frac{|z_{1}|^{2}-|z_{2}|^{2}}{2},\,\log|\gamma(z_{1},z_{2})-z_{3}|,\,\log|\gamma(z_{1},z_{2})+z_{3}|\right), (43)

whose discriminant is the slope +1 diagonal through zero in {0}×ℝ2\{0\}\times\mathbb{R}^{2}. The T2T^{2} action in this case is given by

(ei​θ1,ei​θ2)⋅(z1,z2,z3)=(ei⁡(θ2+θ1)​z1,ei⁡(θ2−θ1)​z2,e2​i​θ2​z3),(e^{i\theta_{1}},e^{i\theta_{2}})\cdot(z_{1},z_{2},z_{3})=(e^{i(\theta_{2}+\theta_{1})}z_{1},\,e^{i(\theta_{2}-\theta_{1})}z_{2},\,e^{2i\theta_{2}}z_{3}), (44)

whose moment map is

(z1,z2,z3)↦(|z1|2−|z2|22,|z1|2+|z2|22+|z3|2).(z_{1},z_{2},z_{3})\mapsto\left(\frac{|z_{1}|^{2}-|z_{2}|^{2}}{2},\,\frac{|z_{1}|^{2}+|z_{2}|^{2}}{2}+|z_{3}|^{2}\right).

In the above examples, the reduced spaces are all 2-dimensional. Using Remark 5.3 we can construct variations of (38) by replacing the last component of (38) with any function depending on t=|z1|2−|z2|22t=\frac{|z_{1}|^{2}-|z_{2}|^{2}}{2}, s=|z3|2s=|z_{3}|^{2} and u1=z1​z2u_{1}=z_{1}z_{2}, subject to the condition that all the maps GtG_{t} have 11-dimensional level sets. A choice providing an example of a smooth fibration is given by G=log⁡|u1−1|G=\log|u_{1}-1|, which gives us Example 4.5. One can do more. In fact, one can take a function GG which gives an interpolation between the piecewise smooth fibration in (38) and the smooth one in Example 4.5. This can be done by taking GG depending also on ss, such that GG is equal to log⁡|γ⁡(z1,z2)−1|\log|\gamma(z_{1},z_{2})-1| when ss is big and equal to log⁡|u1−1|\log|u_{1}-1| when ss is small. We will say more about this later on, as this idea is useful in an important step of the main construction of the paper.

Example 5.8 (The amoeba with thin legs).

We now construct an example which interpolates Example 5.5 and 5.7. Consider the smooth function:

H0=π4​Im⁡(u1​u¯2)H_{0}=\frac{\pi}{4}\im(u_{1}\overline{u}_{2})

and let ηH0\eta_{H_{0}} be the Hamiltonian vector field associated to H0H_{0}. If Φs\Phi_{s} is the flow generated by ηH0\eta_{H_{0}}, then the Hamiltonian symplectomorphism associated to H0H_{0} is defined to be ΦH0=Φ1\Phi_{H_{0}}=\Phi_{1}. One computes that in our case

ΦH0:(u1,u2)↦12​(u1−u2,u1+u2).\Phi_{H_{0}}:(u_{1},u_{2})\mapsto\frac{1}{\sqrt{2}}(u_{1}-u_{2},u_{1}+u_{2}).

It maps {u1=0}\{u_{1}=0\} to {v1+v2=0}\{v_{1}+v_{2}=0\}. We now want a symplectomorphism which acts like ΦH0\Phi_{H_{0}} in a small ball centered at the origin and like the identity outside a slightly bigger ball. So choose a cut-off function k:ℝ≥0→[0,1]k:\mathbb{R}_{\geq 0}\rightarrow[0,1] such that, for some ϵ>0\epsilon>0,

k⁡(t)={1when​ 0<t≤ϵ;0when​t≥2​ϵk(t)=\left\{\begin{array}[]{l}1\ \quad\text{when}\ 0<t\leq\epsilon;\\ 0\ \quad\text{when}\ t\geq 2\epsilon\end{array}\right. (45)

and define the Hamiltonian

H=k⁡(|u1|2+|u2|2)​H0.H=k(|u_{1}|^{2}+|u_{2}|^{2})H_{0}.

The Hamiltonian symplectomorphism ΦH\Phi_{H} associated to HH satisfies

ΦH​(u1,u2)={Idℂ2,when​|u1|2+|u2|2≥2​ϵ;12​(u1−u2,u1+u2),when​|u1|2+|u2|2≤ϵ.\Phi_{H}(u_{1},u_{2})=\begin{cases}\id_{\mathbb{C}^{2}},&\quad\text{when}\ |u_{1}|^{2}+|u_{2}|^{2}\geq 2\epsilon;\\ \\ \frac{1}{\sqrt{2}}(u_{1}-u_{2},u_{1}+u_{2}),&\quad\text{when}\ |u_{1}|^{2}+|u_{2}|^{2}\leq\epsilon.\end{cases}

Now let Ψ\Psi be the affine symplectomorphism

Ψ:(v1,v2)↦12​(v1−v2,v1+v2−2).\Psi:(v_{1},v_{2})\mapsto\frac{1}{\sqrt{2}}(v_{1}-v_{2},v_{1}+v_{2}-\sqrt{2}).

and finally, define Φ=Ψ∘ΦH\Phi=\Psi\circ\Phi_{H}. It is clear that

Φ⁡(u1,u2)={Ψ,when​|u1|2+|u2|2≥2​ϵ;(−u2,u1−1),when​|u1|2+|u2|2≤ϵ.\Phi(u_{1},u_{2})=\begin{cases}\Psi,&\quad\text{when}\ |u_{1}|^{2}+|u_{2}|^{2}\geq 2\epsilon;\\ \\ (-u_{2},u_{1}-1),&\quad\text{when}\ |u_{1}|^{2}+|u_{2}|^{2}\leq\epsilon.\end{cases}

Notice that Φ\Phi acts like in (37) on the ball of radius ϵ\sqrt{\epsilon} around the origin, i.e. in a neighborhood of the surface {u1=0}\{u_{1}=0\}, and like in (34) outside a larger ball. We use this Φ\Phi to construct a fibration ff using Proposition 5.4. One can then see that Φ∘Γ0\Phi\circ\Gamma_{0} sends Σ\Sigma to a surface Σ′\Sigma^{\prime} such that Log⁡(Σ′)⊂ℝ2\Log(\Sigma^{\prime})\subset\mathbb{R}^{2} is a 3-legged amoeba with the end of the horizontal leg pinched down to a straight line. The discriminant locus of ff is then Δ={0}×Log⁡(Σ′)⊂ℝ3\Delta=\{0\}\times\Log(\Sigma^{\prime})\subset\mathbb{R}^{3}. Of course, ff fails to be smooth on the slice μ−1​(0)\mu^{-1}(0). Using the same method we can twist Σ\Sigma suitably and obtain a fibrations having discriminant locus an amoeba with three thin legs (cf. Figure 5). For example, to pinch the diagonal leg to a thin line, choose a smooth function H0H_{0} generating the Hamiltonian symplectomorphism

(u1,u2)→(u1+12,u2+12).(u_{1},u_{2})\rightarrow\left(u_{1}+\frac{1}{\sqrt{2}},u_{2}+\frac{1}{\sqrt{2}}\right).

Cut H0H_{0} off with a function ρ\rho which vanishes when |u2|2≤M/2|u_{2}|^{2}\leq M/2, for some big MM, and is equal to 11 when |u2|2≥M|u_{2}|^{2}\geq M. This produces a Hamiltonian HH. Now one proceeds as before. With an almost identical procedure one pinches down the vertical leg. The final choice of symplectomorphism Φ\Phi pinching down all three legs simultaneously may look like:

Φ⁡(u1,u2)={(−u2,u1−1),when​|u1|2+|u2|2≤ϵ;(u1−1,u2−2),when​|u1|2+|u2−2|2≤ϵ;12​(u1−u2,u1+u2),when​|u2|2≥M;Ψ,everywhere else.\Phi(u_{1},u_{2})=\begin{cases}(-u_{2},u_{1}-1),&\quad\text{when}\ |u_{1}|^{2}+|u_{2}|^{2}\leq\epsilon;\\ \\ (u_{1}-1,u_{2}-\sqrt{2}),&\quad\text{when}\ |u_{1}|^{2}+|u_{2}-\sqrt{2}|^{2}\leq\epsilon;\\ \\ \frac{1}{\sqrt{2}}(u_{1}-u_{2},\,u_{1}+u_{2}),&\quad\text{when}\ |u_{2}|^{2}\geq M;\\ \\ \Psi,&\quad\text{everywhere else}.\end{cases} (46)

It is clear that this piecewise smooth example is topologically conjugate to the one in Example 2.9. Here we have made explicit the twistings described there. In §7 we will show that this fibration can be modified so that it is actually smooth towards the ends of the three legs. For this we will develop further the smoothing method sketched at the end of Example 5.7. Also in §7, we will show that this fibration can be modified so that it is smooth away from a neighborhood homeomorphic to a 2-disk containing the codimension 1 part of its discriminant.

The next result states existence of Lagrangian sections of the fibrations in the previous examples.

Proposition 5.9.

The fibrations in Example 5.5 and 5.8 have smooth Lagrangian sections which do not intersect the critical surface Crit⁡(f)\Crit(f).

Proof.

Consider the symplectomorphism Φ\Phi from Example 5.5. The reduced fibration at time t=0t=0, i.e. the map G0=Log∘Φ∘Γ0G_{0}=\Log\circ\Phi\circ\Gamma_{0}, has many Lagrangian sections, since the Log\Log fibration has many. In particular we can choose one which does not intersect Σ=Crit⁡(f)\Sigma=\Crit(f), this follows for example by observing that the following Lagrangian section of the Log\Log fibration

(x1,x2)↦(i​ex1,ex2)(x_{1},x_{2})\mapsto(ie^{x_{1}},e^{x_{2}}) (47)

does not intersect the surface Σ′={v1+v2+1}\Sigma^{\prime}=\{v_{1}+v_{2}+1\}. It is easy to see that a section which does not intersect Σ\Sigma can be lifted to μ−1​(0)\mu^{-1}(0). The image of this lift is a coisotropic 22 dimensional submanifold of XX. Applying the coisotropic embedding theorem, we can extend this submanifold to a Lagrangian submanifold along a direction which is transversal to μ−1​(0)\mu^{-1}(0), e.g. along i​ηi\eta, where η\eta is the Hamiltonian vector field of the S1S^{1} action. This submanifold is then the image of a section of the fibration in Example 5.5.

In the case of Φ\Phi from Example 5.8, Φ⁡(Σ)\Phi(\Sigma) is a small perturbation of Σ′\Sigma^{\prime} as above. One can see that the section in (47) also avoids Φ⁡(Σ)\Phi(\Sigma). Then the argument follows as before. ∎

We notice that “smooth section” in the above statement means a section whose image is a smooth, manifold. In fact there is no obvious notion of what a smooth map from the base is, since there is no notion of smooth coordinates.

In view of Proposition 5.4, the fibrations of Examples 5.5 and 5.8 are all given by piecewise C∞C^{\infty} maps. More precisely, away from Σ\Sigma, they are the union of two honest C∞C^{\infty} fibrations meeting and coinciding along μ−1​(0)\mu^{-1}(0). A similar phenomenon occurs in special Lagrangian geometry [21]. This kind of piecewise smoothness deserves careful attention and we study it in the next Section.

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