ScalingStacks

5.7. Approachable and integrable metrics [02VK]

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

5.7. Approachable and integrable metrics

We are now in position to characterize the approachable metrics. In this section KK is either ℝ\mathbb{R}, ℂ\mathbb{C} or a complete field with respect to an absolute value associated to a nontrivial discrete valuation. We fix a complete fan Σ\Sigma of NℝN_{\mathbb{R}}, so that XΣX_{\Sigma} is proper. Let Ψ\Psi be a support function on Σ\Sigma, ΔΨ\Delta_{\Psi} the corresponding polytope, and (LΨ,sΨ)(L_{\Psi},s_{\Psi}) the corresponding toric line bundle and section. For short, write X=XΣX=X_{\Sigma}, L=LΨL=L_{\Psi} and s=sΨs=s_{\Psi}.

Theorem 5.73.

Assume the previous hypothesis.

  1. (1)

    The assignment ∥⋅∥↦ψ∥⋅∥\|\cdot\|\mapsto\psi_{\|\cdot\|} is a bijection between the space of approachable toric metrics on LanL^{{\text{\rm an}}} and the space of continuous concave functions ψ\psi on NℝN_{\mathbb{R}} such that |ψ−Ψ||\psi-\Psi| is bounded.

  2. (2)

    The assignment ∥⋅∥↦ψ∥⋅∥∨\|\cdot\|\mapsto\psi_{\|\cdot\|}^{\vee} is a bijection between the space of approachable toric metrics on LanL^{{\text{\rm an}}} and the space of continuous concave functions on ΔΨ\Delta_{\Psi}.

Proof.

By Proposition 3.77(2) and Proposition 3.80, the statements (1) and (2) are equivalent.

Let ∥⋅∥\|\cdot\| be an approachable toric metric. By Corollary 5.17 the function |ψ∥⋅∥−Ψ||\psi_{\|\cdot\|}-\Psi| is bounded. By approachability there is a sequence ∥⋅∥l\|\cdot\|_{l} of smooth (resp. algebraic) semipositive metrics that converges to ∥⋅∥\|\cdot\|. Since ∥⋅∥\|\cdot\| is toric, ∥⋅∥𝕊=∥⋅∥\|\cdot\|_{\mathbb{S}}=\|\cdot\|. Hence, the sequence of toric metrics (∥⋅∥l)𝕊(\|\cdot\|_{l})_{\mathbb{S}} also converges to ∥⋅∥\|\cdot\|. We denote ψl=ψ(∥⋅∥l)𝕊\psi_{l}=\psi_{(\|\cdot\|_{l})_{\mathbb{S}}}. By Proposition 5.38 and Proposition 5.67 the functions ψl\psi_{l} are concave. Since the sequence (ψl)l(\psi_{l})_{l} converge uniformly to ψ∥⋅∥\psi_{\|\cdot\|}, the latter is concave.

Let now ψ\psi be a concave function on NℝN_{\mathbb{R}} such that |Ψ−ψ||\Psi-\psi| is bounded. Then ψ\psi determines a metric ∥⋅∥\|\cdot\| on the restriction of LanL^{{\text{\rm an}}} to X0anX_{0}^{{\text{\rm an}}}. Since stab⁡(ψ)=stab⁡(Ψ)=ΔΨ\operatorname{stab}(\psi)=\operatorname{stab}(\Psi)=\Delta_{\Psi}, by Proposition 3.81 there is a sequence of rational piecewise affine concave functions ψl\psi_{l} that converge uniformly to ψ\psi and with rec⁡(ψl)=Ψ\operatorname{rec}(\psi_{l})=\Psi. By Remark 5.46, the functions Ψ−ψl\Psi-\psi_{l} can be extended to continuous functions on NΣN_{\Sigma}. Therefore, Ψ−ψ\Psi-\psi can be extended to a continuous function on NΣN_{\Sigma}. Consequently the metric ∥⋅∥\|\cdot\| can be extended to XanX^{{\text{\rm an}}}. Let ∥⋅∥l\|\cdot\|_{l} be the metric associated to ψl\psi_{l}. Then the sequence of metrics ∥⋅∥l\|\cdot\|_{l} converges to ∥⋅∥\|\cdot\|. By Corollary 5.28, the metrics ∥⋅∥l\|\cdot\|_{l} are approachable. We deduce that ∥⋅∥\|\cdot\| is approachable. ∎

Remark 5.74.

For the case K=ℂK=\mathbb{C}, statement (2) in the above result is related to the Guillemin-Abreu classification of Kähler structures on symplectic toric varieties as explained in [Abr03]. By definition, a symplectic toric variety is a compact symplectic manifold of dimension 2​n2n together with a Hamiltonian action of the compact torus 𝕊an≃(S1)n\mathbb{S}^{{\text{\rm an}}}\simeq(S^{1})^{n}. These spaces are classified by Delzant polytopes of MℝM_{\mathbb{R}}, see for instance [Gui95]. For a given Delzant polytope Δ⊂Mℝ\Delta\subset M_{\mathbb{R}}, the possible (S1)n(S^{1})^{n}-invariant Kähler forms on the symplectic toric variety corresponding to Δ\Delta are classified by smooth convex functions on Δ∘\Delta^{\circ} satisfying some conditions near the border of Δ\Delta. Several differential geometric invariants of a Kähler toric variety can be translated and studied in terms of this convex function, also called the ‘‘symplectic potential’’.

For a smooth positive toric metric ∥⋅∥\|\cdot\| on LΨΔ​(ℂ)L_{\Psi_{\Delta}}(\mathbb{C}), the Chern form defines a Kähler structure on the complex toric variety XΣΔ​(ℂ)X_{\Sigma_{\Delta}}(\mathbb{C}). It turns out that the corresponding symplectic potential coincides with minus the function ψ∨∥⋅∥\psi^{\vee}_{\|\cdot\|}. It would be most interesting to explore further this connection.

We now study the compatibility of the restriction of approachable toric metrics to toric orbits and its inverse image by equivariant maps with direct and inverse image of concave functions. This is an extension of propositions 4.99 and 4.108. We start with the case of orbits, and we state a variant of Proposition 5.22 for approachable metrics.

Proposition 5.75.

Let ∥⋅∥\|\cdot\| be an approachable toric metric on LanL^{{\text{\rm an}}}, and denote L¯=(L,∥⋅∥){\overline{L}}=(L,\|\cdot\|) and ψ=ψL¯,s\psi=\psi_{{\overline{L}},s} the associated concave function on NℝN_{\mathbb{R}}. Let σ∈Σ\sigma\in\Sigma and mσ∈Mm_{\sigma}\in M such that Ψ|σ=mσ|σ\Psi|_{\sigma}=m_{\sigma}|_{\sigma}. Let πσ:Nℝ→N​(σ)ℝ\pi_{\sigma}\colon N_{\mathbb{R}}\to N(\sigma)_{\mathbb{R}} be the projection, πσ∨:M​(σ)ℝ→Mℝ\pi^{\vee}_{\sigma}\colon M(\sigma)_{\mathbb{R}}\to M_{\mathbb{R}} the dual inclusion and ι:V⁡(σ)→X\iota\colon V(\sigma)\to X the closed immersion. Set s′=χmσ​ss^{\prime}=\chi^{m_{\sigma}}s. Then

(5.76) ψι∗​L¯,ι∗​s′=(πσ)∗​(ψ−mσ).\psi_{\iota^{\ast}{\overline{L}},\iota^{\ast}s^{\prime}}=(\pi_{\sigma})_{\ast}(\psi-m_{\sigma}).

Dually, we have that

(5.77) ψι∗​L¯,ι∗​s′∨=(πσ∨+mσ)∗​ψ∨.\psi_{\iota^{\ast}{\overline{L}},\iota^{\ast}s^{\prime}}^{\vee}=(\pi^{\vee}_{\sigma}+m_{\sigma})^{\ast}\psi^{\vee}.

In other words, the Legendre-Fenchel dual of ψι∗​L¯,ι∗​s′\psi_{\iota^{\ast}{\overline{L}},\iota^{\ast}s^{\prime}} is the restriction of ψ∨\psi^{\vee} to the face FσF_{\sigma} translated by −mσ-m_{\sigma}.

Proof.

As in the proof of Proposition 4.99, it is enough to prove equation (5.76). By replacing ψ\psi by ψ−mσ\psi-m_{\sigma}, we can assume without loss of generality that mσ=0m_{\sigma}=0. By the continuity of the metric, the function ψ\psi can be extended to a continuous function ψ¯σ{\overline{\psi}}_{\sigma} on NσN_{\sigma}. Fix u0∈N​(σ)ℝu_{0}\in N(\sigma)_{\mathbb{R}}, write s=ψ¯σ​(u0)s={\overline{\psi}}_{\sigma}(u_{0}) and let u∈Nℝu\in N_{\mathbb{R}} such that πσ​(u)=u0\pi_{\sigma}(u)=u_{0}. By definition

(πσ)∗​(ψ)​(u0)=supp∈ℝ​σψ⁡(u+p).(\pi_{\sigma})_{\ast}(\psi)(u_{0})=\sup_{p\in\mathbb{R}\sigma}\psi(u+p).

It is clear that supp∈ℝ​σψ⁡(u+p)≥s\sup_{p\in\mathbb{R}\sigma}\psi(u+p)\geq s. Suppose that supp∈ℝ​σψ⁡(u+p)>s\sup_{p\in\mathbb{R}\sigma}\psi(u+p)>s. Let q∈ℝ​σq\in\mathbb{R}\sigma such that ψ⁡(u+q)>s\psi(u+q)>s and let ε=(ψ⁡(u+q)−s)/2\varepsilon=(\psi(u+q)-s)/2. By the definition of the topology of NσN_{\sigma}, there exists a p∈ℝ​σp\in\mathbb{R}\sigma such that

(5.78) s−ε<ψ⁡(u+p+σ)<s+ε.s-\varepsilon<\psi(u+p+\sigma)<s+\varepsilon.

Since σ\sigma is a cone of maximal dimension in ℝ​σ\mathbb{R}\sigma, there exists a point r∈(q+σ)∩(p+σ)r\in(q+\sigma)\cap(p+\sigma). By the right inequality of equation (5.78) ψ⁡(u+r)<ψ⁡(u+q)\psi(u+r)<\psi(u+q). By concavity of ψ\psi this implies that

(5.79) limλ→∞ψ⁡(u+r+λ⁡(r−q))=−∞.\lim_{\lambda\to\infty}\psi(u+r+\lambda(r-q))=-\infty.

Since, by construction u+r+ℝ≥0​(r−q)u+r+\mathbb{R}_{\geq 0}(r-q) is contained in u+p+σu+p+\sigma, equation (5.79) contradicts the left inequality of equation (5.78). Hence supp∈ℝ​σψ⁡(u+p)=s\sup_{p\in\mathbb{R}\sigma}\psi(u+p)=s, which proves equation (5.76). ∎

We now interpret the inverse image of an approachable toric metric by an equivariant map whose image intersects the principal open subset in terms of direct and inverse images of concave functions.

Proposition 5.80.

Let N1N_{1} and N2N_{2} be lattices and Σi\Sigma_{i} a complete fan in Ni,ℝN_{i,\mathbb{R}}, i=1,2i=1,2. Let H:N1→N2H\colon N_{1}\to N_{2} be a linear map such that, for each σ1∈Σ1\sigma_{1}\in\Sigma_{1}, there exists σ2∈Σ2\sigma_{2}\in\Sigma_{2} with H⁡(σ1)⊂σ2H(\sigma_{1})\subset\sigma_{2}. Let p∈XΣ2,0​(K)p\in X_{\Sigma_{2},0}(K) and write A:N1,ℝ→N2,ℝA\colon N_{1,\mathbb{R}}\to N_{2,\mathbb{R}} for the affine map A=H+val⁡(p)A=H+{\operatorname{val}}(p). Let ∥⋅∥\|\cdot\| be an approachable toric metric on 𝒪​(DΨ2)an{\mathcal{O}}(D_{\Psi_{2}})^{{\text{\rm an}}}. Then

ψφp.H∗∥⋅∥=A∗ψ∥⋅∥.\psi_{\varphi_{p.H}^{\ast}\|\cdot\|}=A^{*}\psi_{\|\cdot\|}.

Moreover, the Legendre-Fenchel dual of this function is given by

ψφp.H∗∥⋅∥∨=(H∨)∗(ψ∥⋅∥∨−val(p)).\psi_{\varphi_{p.H}^{\ast}\|\cdot\|}^{\vee}=(H^{\vee})_{*}\big(\psi^{\vee}_{\|\cdot\|}-{\operatorname{val}}(p)\big).
Proof.

The first statement is a direct consequence of Proposition 5.24 while the second one follows from Proposition 3.78(1). ∎

We next characterize the measures associated to an approachable metric.

Theorem 5.81.

Let Σ\Sigma be a complete fan of NℝN_{\mathbb{R}}, let Ψ\Psi be a support function on Σ\Sigma and let L=𝒪⁡(DΨ)L=\mathcal{O}(D_{\Psi}). Let ∥⋅∥\|\cdot\| be an approachable metric on LanL^{{\text{\rm an}}} and let ψ=ψ∥⋅∥\psi=\psi_{\|\cdot\|} be the corresponding concave function. Then

(5.82) (valK)∗​(c1​(L¯)n∧δXΣ)=n!​ℳ¯M​(ψ).({\operatorname{val}}_{K})_{\ast}(c_{1}(\overline{L})^{n}\land\delta_{X_{\Sigma}})=n!{\overline{\mathcal{M}}}_{M}(\psi).

Moreover, the measure c1​(L¯)n∧δXΣc_{1}(\overline{L})^{n}\land\delta_{X_{\Sigma}} is characterized, in the Archimedean case, by equation (5.82) and the fact of being toric, while in the non-Archimedean case it is given by

c1​(L¯)n∧δXΣ=(θΣ)∗​(𝐞K)∗​n!​ℳ¯M​(ψ).c_{1}(\overline{L})^{n}\land\delta_{X_{\Sigma}}=(\theta_{\Sigma})_{\ast}({\operatorname{\mathbf{e}}}_{K})_{\ast}n!{\overline{\mathcal{M}}}_{M}(\psi).
Proof.

For short, denote μ=(valK)∗​(c1​(L¯)n∧δXΣ)\mu=({\operatorname{val}}_{K})_{\ast}(c_{1}(\overline{L})^{n}\land\delta_{X_{\Sigma}}). Let ∥⋅∥l\|\cdot\|_{l} be a sequence of semipositive smooth (respectively algebraic) metrics converging to ∥⋅∥\|\cdot\|. By Proposition 2.33, the measures c1(L,∥⋅∥l)n∧δXΣc_{1}(L,\|\cdot\|_{l})^{n}\land\delta_{X_{\Sigma}} converge to c1​(L¯)n∧δXΣc_{1}({\overline{L}})^{n}\land\delta_{X_{\Sigma}}. Therefore, the measures (valK)∗(c1(L,∥⋅∥l)n∧δXΣ)({\operatorname{val}}_{K})_{\ast}(c_{1}(L,\|\cdot\|_{l})^{n}\land\delta_{X_{\Sigma}}) converge to the measure μ\mu on NΣN_{\Sigma}. Proposition 2.37 implies that the measure of XΣan∖X0anX_{\Sigma}^{{\text{\rm an}}}\setminus X_{0}^{{\text{\rm an}}} with respect to c1​(L¯)n∧δXΣc_{1}({\overline{L}})^{n}\land\delta_{X_{\Sigma}} is zero. Therefore NΣ∖NℝN_{\Sigma}\setminus N_{\mathbb{R}} has μ\mu-measure zero. Denote ψl=ψ(∥⋅∥l)𝕊\psi_{l}=\psi_{(\|\cdot\|_{l})_{\mathbb{S}}}. By Proposition 3.108, the measures ℳM​(ψl)\mathcal{M}_{M}(\psi_{l}) converge to the measure ℳM​(ψ)\mathcal{M}_{M}(\psi). Thus μ|Nℝ=n!​ℳM​(ψ)\mu|_{N_{\mathbb{R}}}=n!\mathcal{M}_{M}(\psi). If we add to this that the measure of NΣ∖NℝN_{\Sigma}\setminus N_{\mathbb{R}} is zero, we deduce equation (5.82). The last statement of the theorem is clear from Theorem 5.33 and Theorem 5.70. ∎

We end this section by characterizing integrable metrics.

Corollary 5.83.

Let Σ\Sigma be a complete fan. Then the map ∥⋅∥↦ψ∥⋅∥\|\cdot\|\mapsto\psi_{\|\cdot\|} is a bijection between the space of integrable toric metrics on 𝒪​(DΨ)an\mathcal{O}(D_{\Psi})^{{\text{\rm an}}} and the space of functions ψ∈𝒟¯​(Nℝ)\psi\in{\overline{\mathscr{D}}}(N_{\mathbb{R}}) such that rec⁡(ψ)=Ψ\operatorname{rec}(\psi)=\Psi, were 𝒟¯​(Nℝ){\overline{\mathscr{D}}}(N_{\mathbb{R}}) is the space of functions of Definition 3.82.

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