ScalingStacks

Verified tagged author-source HTML · 2006.16961v1 · cited publication edition alignment unverified.

2.1 Quantitative stratification and good test functions

There is a quantitative stratification on any smooth fibre XtX_{t} induced by the intersection pattern of EiE_{i}: for J⊂IJ\subset I such that EJ=∩i∈JEi≠∅E_{J}=\cap_{i\in J}E_{i}\neq\emptyset, the corresponding statum is

EJ0={x∈Xt|dω𝒳(x,EJ)≲ϵ}∖{x∈Xt|dω𝒳(x,EJ′)≲ϵ, some J′⊋J},E_{J}^{0}=\{x\in X_{t}|d_{\omega_{\mathcal{X}}}(x,E_{J})\lesssim\epsilon\}\setminus\{x\in X_{t}|d_{\omega_{\mathcal{X}}}(x,E_{J^{\prime}})\lesssim\epsilon,\text{ some }J^{\prime}\supsetneq J\},

namely a small ‘ϵ\epsilon-tubular neighbourhood’ of EJE_{J} minus the deeper strata. For J={i}J=\{i\} we write Ei0=E{i}0E_{i}^{0}=E_{\{i\}}^{0}. Here the disc 𝔻t\mathbb{D}_{t} and the small parameter ϵ≪1\epsilon\ll 1 can be shrinked for convenience; the essential thing is that all parameters should be independent of the coordinate tt.

It is useful to introduce local coordinates {zi}0n\{z_{i}\}_{0}^{n} around EJ⊂𝒳E_{J}\subset\mathcal{X}, such that z0,…,zpz_{0},\ldots,z_{p} with p=|J|−1p=|J|-1 are the local defining equations of EjE_{j} for j∈Jj\in J, and locally the fibration map is t=z0​…​zpt=z_{0}\ldots z_{p}. Then up to uniform equivalence, locally

ω𝒳∼∑0n−1​d​zi∧d​z¯i.\omega_{\mathcal{X}}\sim\sum_{0}^{n}\sqrt{-1}dz_{i}\wedge d\bar{z}_{i}.

The rest of this section is devoted to the construction of good test functions. Given any of these divisors E0E_{0}, we can find a nonnegative function h=hE0h=h_{E_{0}} on 𝒳\mathcal{X}, such that

  • •

    In the local charts near E0E_{0} with z0z_{0} being the defining function for E0E_{0},

    h=|z0|2​h~​(z0,…​zn)h=|z_{0}|^{2}\tilde{h}(z_{0},\ldots z_{n})

    for some positive smooth function h~\tilde{h};

  • •

    Away from E0E_{0} the function hh is comparable to 1.

We observe

  • •

    The form ∂∂¯​log⁡h=∂∂¯​log⁡h~\partial\bar{\partial}\log h=\partial\bar{\partial}\log\tilde{h} extends smoothly;

  • •

    For |t|2≪h≲δ≪1|t|^{2}\ll h\lesssim\delta\ll 1 inside XtX_{t}, so that |z0|≫|t||z_{0}|\gg|t|, by a local calculation near EJE_{J} with 0∈J0\in J,

    −1​∂log⁡h∧∂¯​log⁡h∧ω𝒳|Xtn−1≥−12​|z0|2​d​z0∧d​z¯0∧ω𝒳|Xtn−1≳min⁡{1|z0|2,max1≤i≤p⁡|zi|−2}​ω𝒳|Xtn≳min{1h,h1/p|t|−2/p}ω𝒳|Xtn≳min{1h,h1/n|t|−2/n}ω𝒳|Xtn.\begin{split}&\sqrt{-1}\partial\log h\wedge\bar{\partial}\log h\wedge\omega_{\mathcal{X}}|_{X_{t}}^{n-1}\geq\frac{\sqrt{-1}}{2|z_{0}|^{2}}dz_{0}\wedge d\bar{z}_{0}\wedge\omega_{\mathcal{X}}|_{X_{t}}^{n-1}\\ &\gtrsim\min\{\frac{1}{|z_{0}|^{2}},\max_{1\leq i\leq p}|z_{i}|^{-2}\}\omega_{\mathcal{X}}|_{X_{t}}^{n}\\ &\gtrsim\min\{\frac{1}{h},h^{1/p}|t|^{-2/p}\}\omega_{\mathcal{X}}|_{X_{t}}^{n}\\ &\gtrsim\min\{\frac{1}{h},h^{1/n}|t|^{-2/n}\}\omega_{\mathcal{X}}|_{X_{t}}^{n}.\end{split}

    Here in the first line we need to fix δ≪1\delta\ll 1 so that the effect of ∂log⁡h\partial\log h is dominated by d​log⁡z0d\log z_{0}. The second line uses that for 1≤k≤p1\leq k\leq p, the volume forms on XtX_{t}

    1|z0|2​d​z0∧d​z¯0∧∏j≠k,1≤j≤n−1​d​zj∧d​z¯j∼1|zk|2​∏1≤j≤n−1​d​zj∧d​z¯j,\frac{1}{|z_{0}|^{2}}dz_{0}\wedge d\bar{z}_{0}\wedge\prod_{j\neq k,1\leq j\leq n}\sqrt{-1}dz_{j}\wedge d\bar{z}_{j}\sim\frac{1}{|z_{k}|^{2}}\prod_{1\leq j\leq n}\sqrt{-1}dz_{j}\wedge d\bar{z}_{j},

    and the third line uses |t|=|z0​…​zp|∼h1/2​|z1​…​zp||t|=|z_{0}\ldots z_{p}|\sim h^{1/2}|z_{1}\ldots z_{p}|.

  • •

    On XtX_{t} the function h≳|t|2h\gtrsim|t|^{2}. The region {|t|2∼h}⊂Xt\{|t|^{2}\sim h\}\subset X_{t} can be identified as E00E_{0}^{0}, namely the vicinity of E0E_{0} away from deeper strata. Here

    −1​∂log⁡h∧∂¯​log⁡h∧ω𝒳|Xtn−1≥0,−1​∂∂¯​log⁡h∧ω𝒳|Xtn−1≳−ω𝒳|Xtn.\begin{split}&\sqrt{-1}\partial\log h\wedge\bar{\partial}\log h\wedge\omega_{\mathcal{X}}|_{X_{t}}^{n-1}\geq 0,\\ &\sqrt{-1}\partial\bar{\partial}\log h\wedge\omega_{\mathcal{X}}|_{X_{t}}^{n-1}\gtrsim-\omega_{\mathcal{X}}|_{X_{t}}^{n}.\end{split}
007S

Lemma 2.1. (Good test function) Given the divisor E0E_{0}, we can choose a C2C^{2} test function vv on XtX_{t} such that the following hold uniformly for small t≠0t\neq 0:

  • •

    vv is zero for h≥δh\geq\delta.

  • •

    Globally 0≤v≤−log⁡|t|0\leq v\leq-\log|t|.

  • •

    For any divisor EjE_{j} intersecting E0E_{0}, there is a subset of Ej0E_{j}^{0} with measure at least C2C_{2} on which −1​∂∂¯​v∧ω𝒳|Xtn−1≥C3​ω𝒳|Xtn\sqrt{-1}\partial\bar{\partial}v\wedge\omega_{\mathcal{X}}|_{X_{t}}^{n-1}\geq C_{3}\omega_{\mathcal{X}}|_{X_{t}}^{n}.

  • •

    For C4​|t|2≤h≤δC_{4}|t|^{2}\leq h\leq\delta, the form −1​∂∂¯​v∧ω𝒳|Xtn−1≥0\sqrt{-1}\partial\bar{\partial}v\wedge\omega_{\mathcal{X}}|_{X_{t}}^{n-1}\geq 0.

  • •

    For h≤C4​|t|2h\leq C_{4}|t|^{2}, the form −1​∂∂¯​v∧ω𝒳|Xtn−1≥−C5​ω𝒳|Xtn\sqrt{-1}\partial\bar{\partial}v\wedge\omega_{\mathcal{X}}|_{X_{t}}^{n-1}\geq-C_{5}\omega_{\mathcal{X}}|_{X_{t}}^{n}.

007T

Proof. We seek the test function in the form v=Φ∘log⁡hv=\Phi\circ\log h for some convex, non-increasing, non-negative C2C^{2}-function Φ\Phi. Compute

∂∂¯​v=Φ′′​∂log⁡h∧∂¯​log⁡h+Φ′​(∂∂¯​log⁡h~),\partial\bar{\partial}v=\Phi^{\prime\prime}\partial\log h\wedge\bar{\partial}\log h+\Phi^{\prime}(\partial\bar{\partial}\log\tilde{h}),

so using the properties of hh above,

−1​∂∂¯​v∧ω𝒳|Xtn−1≥{(Φ′′C1′min{1h,h1/n|t|−2/n}+Φ′C2′)ω𝒳|Xtn,|t|2≲h≤δ,C3′Φ′ω𝒳|Xtn,h≲|t|2.\sqrt{-1}\partial\bar{\partial}v\wedge\omega_{\mathcal{X}}|_{X_{t}}^{n-1}\geq\begin{cases}\left(\Phi^{\prime\prime}C_{1}^{\prime}\min\{\frac{1}{h},h^{1/n}|t|^{-2/n}\}+\Phi^{\prime}C_{2}^{\prime}\right)\omega_{\mathcal{X}}|_{X_{t}}^{n},\quad&|t|^{2}\lesssim h\leq\delta,\\ C_{3}^{\prime}\Phi^{\prime}\omega_{\mathcal{X}}|_{X_{t}}^{n},\quad&h\lesssim|t|^{2}.\end{cases}

To satisfy our conditions on vv, it is enough to have

  • •

    Φ⁡(x)=0\Phi(x)=0 for x≥log⁡δx\geq\log\delta.

  • •

    |Φ′​(x)|≲1|\Phi^{\prime}(x)|\lesssim 1 for 2​log⁡|t|≲x≤log⁡δ2\log|t|\lesssim x\leq\log\delta.

  • •

    −dd​xlog|Φ′|=Φ′′|Φ′|≥C4′max{h,h−1/n|t|2/n}-\frac{d}{dx}\log|\Phi^{\prime}|=\frac{\Phi^{\prime\prime}}{|\Phi^{\prime}|}\geq C_{4}^{\prime}\max\{h,h^{-1/n}|t|^{2/n}\} for h=ex≤δ,h=e^{x}\leq\delta, where C4′>C2′/C1′C_{4}^{\prime}>C_{2}^{\prime}/C_{1}^{\prime}. Morever, for x<δx<\delta, we need Φ′<0\Phi^{\prime}<0 so that −1​∂∂¯​v∧ω𝒳|Xtn−1\sqrt{-1}\partial\bar{\partial}v\wedge\omega_{\mathcal{X}}|_{X_{t}}^{n-1} has some strict positivity for δ/2<h<δ\delta/2<h<\delta. Notice convexity of Φ\Phi is a consequence of these conditions.

To construct such Φ\Phi, we can prescribe the behaviour near x=log⁡δx=\log\delta by Φ′​(x)=−e1/(x−log⁡δ)\Phi^{\prime}(x)=-e^{1/(x-\log\delta)} for x<log⁡δx<\log\delta, and match this with a solution to

−dd​xlog|Φ′|=C4′max{ex,e−x/n|t|2/n},x<logδ-\frac{d}{dx}\log|\Phi^{\prime}|=C_{4}^{\prime}\max\{e^{x},e^{-x/n}|t|^{2/n}\},\quad x<\log\delta

for some large enough C4′C_{4}^{\prime}, such that Φ′\Phi^{\prime} remains C1C^{1} at the matching point. Integration shows that |Φ′||\Phi^{\prime}| remains uniformly bounded at h∼|t|2h\sim|t|^{2}, or equivalently x∼2​log⁡|t|x\sim 2\log|t|. ∎

Original mathematics by the credited authors. Source collection and HTML conversion remain in progress.