ScalingStacks

Proof of Theorem 6.6 . [02WA]

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

Proof of Theorem 6.6.

For short, we set Δ=ΔΨ\Delta=\Delta_{\Psi} and ψ=ψ∥⋅∥\psi=\psi_{\|\cdot\|}. Let ΣΔ\Sigma_{\Delta} be the fan associated to Δ\Delta as in Remark 4.43. There is a toric morphism φ:XΣ→XΣΔ\varphi\colon X_{\Sigma}\to X_{\Sigma_{\Delta}}. The function ψ∨\psi^{\vee} defines an approachable metric ∥⋅∥′\|\cdot\|^{\prime} on 𝒪​(DΨΔ)an\mathcal{O}(D_{\Psi_{\Delta}})^{{\text{\rm an}}}. We denote L¯′=(𝒪(DΨΔ),∥⋅∥′){\overline{L}}^{\prime}=(\mathcal{O}(D_{\Psi_{\Delta}}),\|\cdot\|^{\prime}). Then there is an isometry φ∗​(L¯′)=L¯\varphi^{\ast}({\overline{L}}^{\prime})={\overline{L}}. By Corollary 5.25 there is an isometry φ∗(L¯′)can=L¯can\varphi^{\ast}({\overline{L}}^{\prime}{}^{{\operatorname{can}}})={\overline{L}}^{{\operatorname{can}}}.

If the dimension of Δ\Delta is less than nn, then the right-hand side of equation (6.7) is zero. Moreover, n=dim(XΣ)>dim(XΣΔ)n=\dim(X_{\Sigma})>\dim(X_{\Sigma_{\Delta}}) and the metrized line bundles L¯{\overline{L}} and L¯can{\overline{L}}^{{\operatorname{can}}} come from a variety of smaller dimension. Therefore, by Theorem 2.46(2), the left-hand side of equation (6.7) is also zero, because φ∗​XΣ\varphi_{*}X_{\Sigma} is the cycle zero. If Δ\Delta has dimension nn then φ\varphi is a birational morphism, so, by Theorem 2.46(2),

hL¯tor⁡(XΣ)=hL¯′tor⁡(XΣΔ).\operatorname{h}^{\operatorname{tor}}_{{\overline{L}}}(X_{\Sigma})=\operatorname{h}^{\operatorname{tor}}_{{\overline{L}}^{\prime}}(X_{\Sigma_{\Delta}}).

Therefore it is enough to prove the theorem for XΣΔX_{\Sigma_{\Delta}}. By construction, the fan ΣΔ\Sigma_{\Delta} is regular; hence the variety XΣΔX_{\Sigma_{\Delta}} is projective and L′L^{\prime} is ample. Thus we are reduced to prove the theorem in the case when Σ\Sigma is regular and LL is ample.

Now the proof is done by induction on nn, the dimension of XΣX_{\Sigma}. If n=0n=0 then XΣ=ℙ0X_{\Sigma}=\mathbb{P}^{0}, Ψ=0\Psi=0, Δ={0}\Delta=\{0\} and L=𝒪⁡(D0)=𝒪ℙ0L=\mathcal{O}(D_{0})=\mathcal{O}_{\mathbb{P}^{0}}. By equation (5.15), log⁡‖s‖=λK​ψ​(0)\log\|s\|=\lambda_{K}\psi(0) and log⁡‖s‖can=λK​Ψ​(0)=0\log\|s\|_{{\operatorname{can}}}=\lambda_{K}\Psi(0)=0. The Legendre-Fenchel dual of ψ\psi satisfies ψ∨​(0)=−ψ​(0)\psi^{\vee}(0)=-\psi(0). By equation (2.40), hL¯⁡(XΣ;s)=−λK​ψ​(0)\operatorname{h}_{{\overline{L}}}(X_{\Sigma};s)=-\lambda_{K}\psi(0) and hL¯can⁡(XΣ;s)=0\operatorname{h}_{{\overline{L}}^{{\operatorname{can}}}}(X_{\Sigma};s)=0. Therefore

hL¯tor⁡(XΣ)=−λK​ψ​(0)=λK​ψ∨​(0)=1!​λK​∫Δψ∨​d​volM.\operatorname{h}^{\operatorname{tor}}_{{\overline{L}}}(X_{\Sigma})=-\lambda_{K}\psi(0)=\lambda_{K}\psi^{\vee}(0)=1!\lambda_{K}\int_{\Delta}\psi^{\vee}\,\text{\rm d}\operatorname{vol}_{M}.

Let n≥1n\geq 1 and let s0,…,sn−1s_{0},\dots,s_{n-1} be rational sections of 𝒪⁡(DΨ){\mathcal{O}}(D_{\Psi}) such that s0,…,sn−1,ss_{0},\dots,s_{n-1},s intersect XΣX_{\Sigma} properly. By the construction of local heights (Definition 2.39),

(6.10) hL¯⁡(XΣ,s0,…,sn−1,s)=hL¯⁡(div⁡(s)CLOSE;\displaystyle\operatorname{h}_{{\overline{L}}}(X_{\Sigma};s_{0},\dots,s_{n-1},s)=\operatorname{h}_{{\overline{L}}}(\operatorname{div}(s); OPENs0,…,sn−1)\displaystyle s_{0},\dots,s_{n-1})
−∫XΣanlog∥s∥c1(L¯)n∧δXΣ\displaystyle-\int_{X_{\Sigma}^{{\text{\rm an}}}}\log\|s\|c_{1}({\overline{L}})^{n}\wedge\delta_{X_{\Sigma}}

and a similar formula holds for the canonical metric.

For each facet FF of Δ\Delta let vFv_{F} be as in Notation 3.103. Since LL is ample, Proposition 4.46 implies

(6.11) hL¯(div(s);s0,…,sn−1)=∑F−⟨vF,F⟩hL¯(V(τF);s0,…,sn−1),\operatorname{h}_{{\overline{L}}}(\operatorname{div}(s);s_{0},\dots,s_{n-1})=\sum_{F}-\langle v_{F},F\rangle\operatorname{h}_{{\overline{L}}}(V(\tau_{F});s_{0},\dots,s_{n-1}),

where the sum is over the facets FF of Δ\Delta. Observe that the local height of V⁡(τF)V(\tau_{F}) with respect to the metrized line bundle L¯{\overline{L}} coincides with the local height associated to the restriction of L¯{\overline{L}} to this subvariety. Moreover by Corollary 5.23, the restriction of the canonical metric of LanL^{{\text{\rm an}}} to this subvariety agrees with the canonical metric of Lan|V⁡(τF)L^{{\text{\rm an}}}|_{V(\tau_{F})}. Hence, by substracting from equation (6.11) the analogous formula for the canonical metric, we obtain

(6.12) ∑F−⟨mF,vF⟩hL¯|V⁡(τF)tor(V(τF))=hL¯(div(s);\displaystyle\sum_{F}-\langle m_{F},v_{F}\rangle\operatorname{h}^{\operatorname{tor}}_{{\overline{L}}|_{V(\tau_{F})}}(V(\tau_{F}))=\operatorname{h}_{{\overline{L}}}(\operatorname{div}(s); OPENs0,…,sn−1)\displaystyle s_{0},\dots,s_{n-1})
−hL¯can⁡(div⁡(s),s0,…,sn−1).\displaystyle-\operatorname{h}_{{\overline{L}}^{{\operatorname{can}}}}(\operatorname{div}(s);s_{0},\dots,s_{n-1}).

Moreover, Proposition 2.37 implies that

∫XΣanlog⁡‖s‖​c1​(L¯)n∧δXΣ=∫XΣ,0anlog⁡‖s‖​c1​(L¯)n∧δXΣ.\int_{X_{\Sigma}^{{\text{\rm an}}}}\log\|s\|c_{1}({\overline{L}})^{n}\wedge\delta_{X_{\Sigma}}=\int_{X_{\Sigma,0}^{{\text{\rm an}}}}\log\|s\|c_{1}({\overline{L}})^{n}\wedge\delta_{X_{\Sigma}}.

By equation (5.15), log⁡‖s‖=(valK)∗​(λK​ψ)\log\|s\|=({\operatorname{val}}_{K})^{*}(\lambda_{K}\psi). Moreover

∫XΣ,0an(valK)∗​(λK​ψ)​c1​(L¯)n∧δXΣ=\displaystyle\int_{X_{\Sigma,0}^{{\text{\rm an}}}}({\operatorname{val}}_{K})^{*}(\lambda_{K}\psi)c_{1}({\overline{L}})^{n}\wedge\delta_{X_{\Sigma}}= ∫NℝλK​ψ​(valK)∗​(c1​(L¯)n∧δXΣ)\displaystyle\int_{N_{\mathbb{R}}}\lambda_{K}\psi({\operatorname{val}}_{K})_{*}(c_{1}({\overline{L}})^{n}\wedge\delta_{X_{\Sigma}})

and by Theorem 5.81, (valK)∗​(c1​(L¯)n∧δXΣ)=n!​ℳM​(ψ)({\operatorname{val}}_{K})_{*}(c_{1}({\overline{L}})^{n}\wedge\delta_{X_{\Sigma}})=n!{\mathcal{M}}_{M}(\psi). Hence

(6.13) ∫XΣanlog⁡‖s‖​c1​(L¯)n∧δXΣ=n!​λK​∫Nℝψ​ℳM​(ψ).\int_{X_{\Sigma}^{{\text{\rm an}}}}\log\|s\|c_{1}({\overline{L}})^{n}\wedge\delta_{X_{\Sigma}}=n!\lambda_{K}\int_{N_{\mathbb{R}}}\psi{\mathcal{M}}_{M}(\psi).

By Example 3.96, ℳM​(Ψ)=volM⁡(Δ)​δ0{\mathcal{M}}_{M}(\Psi)=\operatorname{vol}_{M}(\Delta)\delta_{0}. Therefore, in the case of the canonical metric, equation (6.13) reads as

(6.14) ∫XΣanlog⁡‖s‖can​c1​(L¯can)n∧δXΣ=n!​λK​volM⁡(Δ)​Ψ​(0)=0.\int_{X_{\Sigma}^{{\text{\rm an}}}}\log\|s\|_{{\operatorname{can}}}c_{1}({\overline{L}}^{{\operatorname{can}}})^{n}\wedge\delta_{X_{\Sigma}}=n!\lambda_{K}\operatorname{vol}_{M}(\Delta)\Psi(0)=0.

Thus, substracting from equation (6.10) the analogous formula for the canonical metric and using equations (6.12), (6.13) and (6.14), we obtain

(6.15) hL¯tor(XΣ)=∑F−⟨vF,F⟩hL¯|V⁡(τF)tor(V(τF))−n!λK∫NℝψℳM(ψ).\operatorname{h}^{\operatorname{tor}}_{{\overline{L}}}(X_{\Sigma})=\sum_{F}-\langle v_{F},F\rangle\operatorname{h}^{\operatorname{tor}}_{{\overline{L}}|_{V(\tau_{F})}}(V(\tau_{F}))-n!\lambda_{K}\int_{N_{\mathbb{R}}}\psi{\mathcal{M}}_{M}(\psi).

By the inductive hypothesis and equation (5.77)

hL¯|V⁡(τF)tor⁡(V⁡(τF))=n!​λK​∫Fψ∨​d​volM⁡(F).\operatorname{h}^{\operatorname{tor}}_{{\overline{L}}|_{V(\tau_{F})}}(V(\tau_{F}))=n!\lambda_{K}\int_{F}\psi^{\vee}\,\text{\rm d}\operatorname{vol}_{M({F})}.

Hence, by Corollary 3.104,

hL¯tor⁡(XΣ)\displaystyle\operatorname{h}^{\operatorname{tor}}_{{\overline{L}}}(X_{\Sigma}) =−n!λK∑F⟨vF,F⟩∫Fψ∨dvolM⁡(F)−n!λK∫NℝψℳM(ψ)\displaystyle=-n!\lambda_{K}\sum_{F}\langle v_{F},F\rangle\int_{F}\psi^{\vee}\,\text{\rm d}\operatorname{vol}_{M({F})}-n!\lambda_{K}\int_{N_{\mathbb{R}}}\psi{\mathcal{M}}_{M}(\psi)
=(n+1)!​λK​∫Δψ∨​d​volM,\displaystyle=(n+1)!\lambda_{K}\int_{\Delta}\psi^{\vee}\,\text{\rm d}\operatorname{vol}_{M},

proving the theorem ∎

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