ScalingStacks

2. Metrized line bundles and their associated heights [02IF]

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

2. Metrized line bundles and their associated heights

In this section we will recall the adelic theory of heights as introduced by Zhang [Zha95b] and developed by Gubler [Gub02, Gub03] and Chambert-Loir [Cha06]. These heights generalize the ones that can be obtained from the arithmetic intersection theory of Gillet and Soulé [GS90, BGS94].

To explain the difference between both points of view, consider a smooth variety XX over ℚ\mathbb{Q}. In Gillet-Soulé’s theory, we choose a regular proper model 𝒳{\mathcal{X}} over ℤ\mathbb{Z} of XX, and we also consider the real analytic space XanX^{{\text{\rm an}}} given by the set of complex points X⁡(ℂ)X(\mathbb{C}) and the anti-linear involution induced by the complex conjugation. By contrast, in the adelic point of view we consider the whole family of analytic spaces XvanX^{{\text{\rm an}}}_{v}, v∈𝔐ℚv\in{\mathfrak{M}}_{\mathbb{Q}}. For the Archimedean place, XvanX^{{\text{\rm an}}}_{v} is the real analytic space considered before, while for the non-Archimedean places, this is the associated Berkovich space [Ber90]. Both points of view have advantages and disadvantages. In the former point of view, there exists a complete formalism of intersection theory and characteristic classes, with powerful theorems like the arithmetic Riemann-Roch theorem and the Lefschetz fixed point theorem, but one is restricted to smooth varieties and needs an explicit integral model of XX. In the latter point of view, one can define heights, but does not dispose yet of a complete formalism of intersection theory. Its main advantages are that it can be easily extended to non-smooth varieties and that there is no need of an integral model of XX. Moreover, all places, Archimedean and non-Archimedean, are set on a similar footing.

2.1. Smooth metrics in the Archimedean case

Let XX be an algebraic variety over ℂ\mathbb{C} and XanX^{{\text{\rm an}}} its associated complex analytic space. We recall the definition of differential forms on XanX^{{\text{\rm an}}} introduced by Bloom and Herrera [BH69]. The space XanX^{{\text{\rm an}}} can be covered by a family of open subsets {Ui}i\{U_{i}\}_{i} such that each UiU_{i} can be identified with a closed analytic subset of an open ball in ℂr\mathbb{C}^{r} for some rr. On each UiU_{i}, the differential forms are defined as the restriction to this subset of smooth complex-valued differential forms defined on an open neighbourhood of UiU_{i} in ℂr\mathbb{C}^{r}. Two differential forms on UiU_{i} are identified if they coincide on the non-singular locus of UiU_{i}. We denote by 𝒜∗​(Ui)\mathscr{A^{\ast}}(U_{i}) the complex of differential forms of UiU_{i}, which is independent of the chosen embedding. In particular, if UiU_{i} is non-singular, we recover the usual complex of differential forms. These complexes glue together to define a sheaf 𝒜Xan∗\mathscr{A}^{\ast}_{X^{{\text{\rm an}}}}. This sheaf is equipped with differential operators  d, dc\operatorname{d^{c}}, ∂\partial, ∂¯\bar{\partial}, an external product and inverse images with respect to analytic morphisms: these operations are defined locally on each 𝒜∗​(Ui)\mathscr{A^{\ast}}(U_{i}) by extending the differential forms to a neighbourhood of UiU_{i} in ℂr\mathbb{C}^{r} as above and applying the corresponding operations for ℂr\mathbb{C}^{r}. We write 𝒪Xan\mathcal{O}_{X^{{\text{\rm an}}}} and CXan∞=𝒜Xan0C^{\infty}_{X^{{\text{\rm an}}}}=\mathscr{A}^{0}_{X^{{\text{\rm an}}}} for the sheaves of analytic functions and of smooth functions of XanX^{{\text{\rm an}}}, respectively.

Let LL be an algebraic line bundle on XX and LanL^{{\text{\rm an}}} its analytification.

Definition 2.1.

A metric on LanL^{{\text{\rm an}}} is an assignment that, to each local section ss of LanL^{{\text{\rm an}}} on an open subset U⊂XanU\subset X^{{\text{\rm an}}}, associates a continuous function

‖s⁡(⋅)‖:U⟶ℝ≥0\|s(\cdot)\|\colon U\longrightarrow\mathbb{R}_{\geq 0}

such that, for all p∈Up\in U,

  1. (1)

    ‖s⁡(p)‖=0\|s(p)\|=0 if and only if s⁡(p)=0s(p)=0;

  2. (2)

    for any λ∈𝒪Xan​(U)\lambda\in\mathcal{O}_{X^{{\text{\rm an}}}}(U), it holds ‖(λ​s)​(p)‖=|λ⁡(p)|​‖s⁡(p)‖.\|(\lambda s)(p)\|=|\lambda(p)|\,\|s(p)\|.

The pair L¯:=(L,∥⋅∥){\overline{L}}:=(L,\|\cdot\|) is called a metrized line bundle.The metric ∥⋅∥\|\cdot\| is smooth if for every local section ss of LanL^{{\text{\rm an}}}, the function ‖s⁡(⋅)‖2\|s(\cdot)\|^{2} is smooth.

We remark that what we call “metric” in this text is called “continuous metric” in other contexts.

Let L¯=(L,∥⋅∥){\overline{L}}=(L,\|\cdot\|) be a smooth metrized line bundle. Given a local section ss of LanL^{{\text{\rm an}}} on an open subset UU, the first Chern form of L¯{\overline{L}} is the (1,1)(1,1)-form defined on UU as

c1⁡(L¯)=∂∂¯​log⁡‖s‖2∈𝒜1,1​(U).\operatorname{c}_{1}({\overline{L}})=\partial\bar{\partial}\log\|s\|^{2}\in\mathscr{A}^{1,1}(U).

It does not depend on the choice of local section and can be extended to a global closed (1,1)(1,1)-form. Observe that we are using the algebro-geometric convention, and so c1⁡(L¯)\operatorname{c}_{1}({\overline{L}}) determines a class in H2​(Xan,2​π​i​ℤ)H^{2}(X^{{\text{\rm an}}},2\pi i\,\mathbb{Z}).

Example 2.2.

Let X=ℙℂnX=\mathbb{P}^{n}_{\mathbb{C}} and L=𝒪⁡(1)L={\mathcal{O}}(1), the universal line bundle of ℙℂn\mathbb{P}^{n}_{\mathbb{C}}. A rational section ss of 𝒪⁡(1){\mathcal{O}}(1) can be identified with a homogeneous rational function ρs∈ℂ⁡(x0,…,xn)\rho_{s}\in\mathbb{C}(x_{0},\dots,x_{n}) of degree 1. The poles of this section coincide which those of ρs\rho_{s}. For a point p=(p0:…:pn)∈ℙn(ℂ)p=(p_{0}:\dots:p_{n})\in\mathbb{P}^{n}(\mathbb{C}) outside this set of poles, the Fubini-Study metric of 𝒪​(1)an{\mathcal{O}}(1)^{{\text{\rm an}}} is defined as

‖s⁡(p)‖FS=|ρs​(p0,…,pn)|(∑i|pi|2)1/2.\|s(p)\|_{\operatorname{FS}}=\frac{|\rho_{s}(p_{0},\dots,p_{n})|}{(\sum_{i}|p_{i}|^{2})^{1/2}}.

Clearly, this definition does not depend on the choice of a representative of pp. The pair (𝒪(1),∥⋅∥FS)({\mathcal{O}}(1),\|\cdot\|_{{\operatorname{FS}}}) is a metrized line bundle.

Many smooth metrics can be obtained as the inverse image of the Fubini-Study metric. Let XX be a variety over ℂ\mathbb{C} and LL a line bundle on XX, and assume that there is an integer e≥1e\geq 1 such that L⊗eL^{\otimes e} is generated by global sections. Choose a basis of the space of global sections Γ⁡(X,L⊗e)\Gamma(X,L^{\otimes e}) and let φ:X→ℙℂM\varphi\colon X\to\mathbb{P}^{M}_{\mathbb{C}} be the induced morphism. Given a local section ss of LL, let s′s^{\prime} be a local section of 𝒪⁡(1){\mathcal{O}}(1) such that s⊗e=φ∗​s′s^{\otimes e}=\varphi^{\ast}s^{\prime}. Then, the smooth metric on LanL^{{\text{\rm an}}} obtained from the Fubini-Study metric by inverse image is given by

‖s⁡(p)‖=‖s′​(φ⁡(p))‖FS1/e\|s(p)\|=\|s^{\prime}(\varphi(p))\|^{1/e}_{{\operatorname{FS}}}

for any p∈Xanp\in X^{{\text{\rm an}}} which is not a pole of ss.

Definition 2.3.

Let L¯{\overline{L}} be a smooth metrized line bundle and 𝔻={z∈ℂ||z|≤1}\mathbb{D}=\{z\in\mathbb{C}|\,|z|\leq 1\}, the unit disk of ℂ\mathbb{C}. We say that L¯{\overline{L}} is semipositive if, for every holomorphic map φ:𝔻⟶Xan,\varphi\colon\mathbb{D}\longrightarrow X^{{\text{\rm an}}},

12​π​i​∫𝔻φ∗​c1⁡(L¯)≥0.\frac{1}{2\pi i}\int_{\mathbb{D}}\varphi^{\ast}\operatorname{c}_{1}({\overline{L}})\geq 0.

We say that L¯{\overline{L}} is positive if this integral is strictly positive for all non-constant holomorphic maps as before.

Example 2.4.

The Fubini-Study metric (Example 2.2) is positive because its first Chern form defines a smooth metric on the holomorphic tangent bundle of ℙn​(ℂ)\mathbb{P}^{n}(\mathbb{C}) [GH94, Chapter 0, §2]. All metrics obtained as inverse image of the Fubini-Study metric are semipositive.

A family of smooth metrized line bundles L¯0,…,L¯d−1{\overline{L}}_{0},\dots,{\overline{L}}_{d-1} on XX and a dd-dimensional cycle YY of XX define a signed measure on XanX^{{\text{\rm an}}} as follows. First suppose that YY is a subvariety of XX and let δY\delta_{Y} denote the current of integration along the analytic subvariety YanY^{{\text{\rm an}}}, defined as δY​(ω)=1(2​π​i)d​∫Yanω\delta_{Y}(\omega)=\frac{1}{(2\pi i)^{d}}\int_{Y^{{\text{\rm an}}}}\omega for ω∈𝒜Xan2​d\omega\in\mathscr{A}^{2d}_{X^{{\text{\rm an}}}}. Then the current

c1⁡(L¯0)∧⋯∧c1⁡(L¯d−1)∧δY\operatorname{c}_{1}({\overline{L}}_{0})\wedge\cdots\wedge\operatorname{c}_{1}({\overline{L}}_{d-1})\wedge\delta_{Y}

is a signed measure on XanX^{{\text{\rm an}}}. This notion extends by linearity to Y∈Zd​(X)Y\in Z_{d}(X). If L¯i{\overline{L}}_{i}, i=0,…,d−1i=0,\dots,d-1, are semipositive and YY is effective, this signed measure is a measure.

Remark 2.5.

We can reduce the study of algebraic varieties and line bundles over the field of real numbers to the complex case by using the following standard technique. A variety XX over ℝ\mathbb{R} induces a variety XℂX_{\mathbb{C}} over ℂ\mathbb{C} together with an anti-linear involution σ:Xℂ→Xℂ\sigma\colon X_{\mathbb{C}}\to X_{\mathbb{C}} such that the diagram

Xℂ\textstyle{X_{\mathbb{C}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}σ\scriptstyle{\sigma}Xℂ\textstyle{X_{\mathbb{C}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}Spec⁡(ℂ)\textstyle{\operatorname{Spec}(\mathbb{C})\ignorespaces\ignorespaces\ignorespaces\ignorespaces}Spec⁡(ℂ)\textstyle{\operatorname{Spec}(\mathbb{C})}

commutes, where the arrow below denotes the map induced by complex conjugation. A line bundle LL on XX determines a line bundle LℂL_{\mathbb{C}} on XℂX_{\mathbb{C}} and an isomorphism α:σ∗​Lℂ→Lℂ\alpha\colon\sigma^{*}L_{\mathbb{C}}\to L_{\mathbb{C}} such that a section ss of LℂL_{\mathbb{C}} is real if and only if α⁡(σ∗​s)=s\alpha(\sigma^{*}s)=s. By a metric on LanL^{{\text{\rm an}}} we will mean a metric ∥⋅∥\|\cdot\| on LℂanL_{\mathbb{C}}^{{\text{\rm an}}} such that the induced map σ∗(Lℂ,∥⋅∥)→(Lℂ,∥⋅∥)\sigma^{*}(L_{\mathbb{C}},\|\cdot\|)\to(L_{\mathbb{C}},\|\cdot\|) is an isometry.

In this way, the above definitions can be extended to metrized line bundles on varieties over ℝ\mathbb{R}. For instance, a real smooth metrized line bundle is semipositive if and only if its associated complex smooth metrized line bundle is semipositive. The corresponding signed measure is a measure over XℂanX_{\mathbb{C}}^{{\text{\rm an}}} which is invariant under σ\sigma.

In the sequel, every time we have a real variety, we will work with the associated complex variety and quietly ignore the anti-linear involution σ\sigma, because it will play no role in our results.

2.2. Berkovich spaces of schemes

In this section we recall Berkovich’s theory of analytic spaces. We will not present the most general theory developed in [Ber90] but we will content ourselves with the analytic spaces associated to algebraic varieties, that are simpler to define and enough for our purposes.

Let KK be a field complete with respect to a nontrivial non-Archimedean absolute value |⋅||\cdot|. Such fields will be called non-Archimedean fields. Let K∘={α∈K∣|α|≤1}K^{\circ}=\{\alpha\in K\mid|\alpha|\leq 1\} be the valuation ring, K∘⁣∘={α∈K∣|α|<1}K^{\circ\circ}=\{\alpha\in K\mid|\alpha|<1\} the maximal ideal and k=K∘/K∘⁣∘k=K^{\circ}/K^{\circ\circ} the residue field.

Let XX be a scheme of finite type over KK. Following [Ber90, §1 and Remark 3.4.2], we can associate an analytic space XanX^{{\text{\rm an}}} to the scheme XX as follows. First assume that X=Spec⁡(A)X=\operatorname{Spec}(A), where AA is a finitely generated KK-algebra. Then, the points of XanX^{{\text{\rm an}}} are the multiplicative seminorms of AA that extend the absolute value of KK, see [Ber90, §1.1]. Every element aa of AA defines a function |a⁡(⋅)|:Xan→ℝ≥0|a(\cdot)|\colon X^{{\text{\rm an}}}\to\mathbb{R}_{\geq 0} given by evaluation of the seminorm. The topology of XanX^{{\text{\rm an}}} is the coarsest topology that makes the functions |a⁡(⋅)||a(\cdot)| continuous for all a∈Aa\in A.

To each point p∈Xanp\in X^{{\text{\rm an}}} we attach a prime ideal

𝔭p={a∈A∣|a⁡(p)|=0}.\mathfrak{p}_{p}=\{a\in A\mid|a(p)|=0\}.

This induces a map π:Xan→X\pi\colon X^{{\text{\rm an}}}\to X defined as π⁡(p)=𝔭p\pi(p)=\mathfrak{p}_{p}. The point pp is a multiplicative seminorm on AA and so it induces a non-Archimedean absolute value on the field of fractions of A/𝔭pA/\mathfrak{p}_{p}. We denote by ℋ⁡(p)\mathscr{H}(p) the completion of this field with respect to that absolute value.

Let UU be an open subset of XanX^{{\text{\rm an}}}. An analytic function on UU is a function

f:U⟶∐p∈Uℋ⁡(p)f\colon U\longrightarrow\coprod_{p\in{U}}\mathscr{H}(p)

such that, for each p∈Up\in U, f⁡(p)∈ℋ⁡(p)f(p)\in\mathscr{H}(p) and there is an open neigborhood U′⊂UU^{\prime}\subset U of pp with the property that, for all ε>0\varepsilon>0, there are elements a,b∈Aa,b\in A with b∉𝔭qb\not\in\mathfrak{p}_{q} and |f⁡(q)−a⁡(q)/b⁡(q)|<ε|f(q)-a(q)/b(q)|<\varepsilon for all q∈U′q\in U^{\prime}. The analytic functions form a sheaf, denoted 𝒪Xan\mathcal{O}_{X^{{\text{\rm an}}}}, and (Xan,𝒪Xan)(X^{{\text{\rm an}}},\mathcal{O}_{X^{{\text{\rm an}}}}) is a locally ringed space [Ber90, §1.5 and Remark 3.4.2]. In particular, every element a∈Aa\in A determines an analytic function on XanX^{{\text{\rm an}}}, also denoted aa. The function |a⁡(⋅)||a(\cdot)| can then be obtained by composing aa with the absolute value map

|⋅|:∐p∈Xanℋ(p)⟶ℝ≥0,|\cdot|\colon\coprod_{p\in X^{{\text{\rm an}}}}\mathscr{H}(p)\longrightarrow\mathbb{R}_{\geq 0},

which justifies its notation.

Now, if XX is a scheme of finite type over KK, the analytic space XanX^{{\text{\rm an}}} is defined by gluing together the affine analytic spaces obtained from an affine open cover of XX. If we want to stress the base field we will denote XanX^{{\text{\rm an}}} by XKanX_{K}^{{\text{\rm an}}}.

Let K′K^{\prime} be a complete extension of KK and XK′anX^{\text{\rm an}}_{K^{\prime}} the analytic space associated to the scheme XK′X_{K^{\prime}}. There is a natural map XK′an→XKanX_{K^{\prime}}^{{\text{\rm an}}}\to X_{K}^{{\text{\rm an}}} defined locally by restricting seminorms.

Definition 2.6.

A rational point of XKanX_{K}^{\text{\rm an}} is a point p∈Xanp\in X^{{\text{\rm an}}} satisfying ℋ⁡(p)=K\mathscr{H}(p)=K. We denote by Xan​(K)X^{\text{\rm an}}(K) the set of rational points of XanX^{{\text{\rm an}}}. More generally, for a complete extension K′K^{\prime} of KK, the set of K′K^{\prime}-rational points of XanX^{{\text{\rm an}}} is defined as Xan​(K′)=XK′an​(K′)X^{\text{\rm an}}(K^{\prime})=X^{\text{\rm an}}_{K^{\prime}}(K^{\prime}). There is a map Xan​(K′)→XanX^{{\text{\rm an}}}(K^{\prime})\to X^{{\text{\rm an}}}, defined by the composing the inclusion Xan​(K′)↪XK′anX^{{\text{\rm an}}}(K^{\prime})\hookrightarrow X_{K^{\prime}}^{{\text{\rm an}}} with the map XK′an→XKanX_{K^{\prime}}^{{\text{\rm an}}}\to X_{K}^{{\text{\rm an}}} as above. The set of algebraic points of XanX^{{\text{\rm an}}} is the union of Xan​(K′)X^{{\text{\rm an}}}(K^{\prime}) for all finite extensions K′K^{\prime} of KK. Its image in XanX^{{\text{\rm an}}} is denoted XalganX^{{\text{\rm an}}}_{{\text{\rm alg}}}. We have that Xalgan={p∈X|[ℋ(p):K]<∞}X^{{\text{\rm an}}}_{{\text{\rm alg}}}=\{p\in X|\,[\mathscr{H}(p):K]<\infty\}.

The basic properties of XanX^{{\text{\rm an}}} are summarized in the following theorem.

Theorem 2.7.

Let XX be a scheme of finite type over KK and XanX^{{\text{\rm an}}} the associated analytic space.

  1. (1)

    XanX^{{\text{\rm an}}} is a locally compact and locally arc-connected topological space.

  2. (2)

    XanX^{{\text{\rm an}}} is Hausdorff (respectively compact and Hausdorff, arc-connected) if and only if XX is separated (respectively proper, connected).

  3. (3)

    The map π:Xan→X\pi\colon X^{{\text{\rm an}}}\to X is continuous. A locally constructible subset T⊂XT\subset X is open (respectively closed, dense) if and only if π−1​(T)\pi^{-1}(T) is open (respectively closed, dense).

  4. (4)

    Let ψ:X⟶Y\psi\colon X\longrightarrow Y be a morphism of schemes of finite type over KK and ψan:Xan⟶Yan\psi^{{\text{\rm an}}}\colon X^{{\text{\rm an}}}\longrightarrow Y^{{\text{\rm an}}} its analytification. Then ψ\psi is flat (respectively unramified, étale, smooth, separated, injective, surjective, open immersion, isomorphism) if and only if ψan\psi^{{\text{\rm an}}} has the same property.

  5. (5)

    Let K′K^{\prime} be a complete extension of KK. Then the map πK′:XK′an→XK′\pi_{K^{\prime}}:X^{{\text{\rm an}}}_{K^{\prime}}\to X_{K^{\prime}} induces a bijection between Xan​(K′)X^{{\text{\rm an}}}(K^{\prime}) and X⁡(K′)X(K^{\prime}).

  6. (6)

    Set Xalg={p∈X|[K(p):K]<∞}X_{{\text{\rm alg}}}=\{p\in X|\,[K(p):K]<\infty\}. Then π\pi induces a bijection between XalganX^{{\text{\rm an}}}_{{\text{\rm alg}}} and XalgX_{{\text{\rm alg}}}. The subset Xalgan⊂XanX^{{\text{\rm an}}}_{{\text{\rm alg}}}\subset X^{{\text{\rm an}}} is dense.

Proof.

The proofs can be found in [Ber90] and the next pointers are with respect to the numeration in this reference: (1) follows from Theorem 1.2.1, Corollary 2.2.8 and Theorem 3.2.1, (2) is Theorem 3.4.8, (3) is Corollary 3.4.5, (4) is Proposition 3.4.6, (5) is Theorem 3.4.1(i), while (6) follows from Theorem 3.4.1(i) and Proposition 2.1.15. ∎

Example 2.8.

Let MM be a finitely generated free ℤ\mathbb{Z}-module of rank nn. Consider the associated group algebra K⁡[M]K[M] and the algebraic torus 𝕋M=Spec⁡(K⁡[M])\mathbb{T}_{M}=\operatorname{Spec}(K[M]). The corresponding analytic space 𝕋Man\mathbb{T}_{M}^{{\text{\rm an}}} is the set of multiplicative seminorms of K⁡[M]K[M] that extend the absolute value of KK. This is an analytic group. We warn the reader that the set of points of an analytic group is not an abstract group, hence some care has to be taken when speaking of actions and orbits. The precise definitions and basic properties can be found in [Ber90, §5.1].

Its analytification 𝕋Man\mathbb{T}_{M}^{{\text{\rm an}}} is an analytic torus as in [Ber90, §6.3]. The subset

𝕊an={p∈𝕋Man||χm​(p)|=1​ for all ​m∈M}.\mathbb{S}^{{\text{\rm an}}}=\{p\in\mathbb{T}_{M}^{{\text{\rm an}}}|\,|\chi^{m}(p)|=1\text{ for all }m\in M\}.

is a compact subgroup, called the compact torus of 𝕋Man\mathbb{T}_{M}^{{\text{\rm an}}}.

Remark 2.9.

Not every analytic space in the sense of Berkovich can be obtained by the above procedure. The general theory is based on spectra of affinoid KK-algebras, that provide compact analytic spaces that are the building blocks of the more general analytic spaces.

2.3. Algebraic metrics in the non-Archimedean case

Let KK be a field complete with respect to a nontrivial non-Archimedean absolute value, as in the previous section. For simplicity, we will assume from now on that K∘K^{\circ} is a discrete valuation ring (DVR), and we will fix a generator ϖ\varpi of its maximal ideal K∘⁣∘K^{\circ\circ}. This is the only case we will need in the sequel and it allows us to use a more elementary definition of measures and local heights. Nevertheless, the reader can consult [Gub03, Gub07] for the general case.

Let XX be an algebraic variety over KK and LL a line bundle on XX. Let XanX^{{\text{\rm an}}} and LanL^{{\text{\rm an}}} be their respective analytifications.

Definition 2.10.

A metric on LanL^{{\text{\rm an}}} is an assignment that, to each local section ss of LanL^{{\text{\rm an}}} on an open subset U⊂XanU\subset X^{{\text{\rm an}}}, associates a continuous function

‖s⁡(⋅)‖:U⟶ℝ≥0,\|s(\cdot)\|\colon U\longrightarrow\mathbb{R}_{\geq 0},

such that, for all p∈Up\in U,

  1. (1)

    ‖s⁡(p)‖=0\|s(p)\|=0 if and only if s⁡(p)=0s(p)=0;

  2. (2)

    for any λ∈𝒪Xan​(U)\lambda\in\mathcal{O}_{X^{{\text{\rm an}}}}(U), it holds ‖(λ​s)​(p)‖=|λ⁡(p)|​‖s⁡(p)‖.\|(\lambda s)(p)\|=|\lambda(p)|\,\|s(p)\|.

The pair L¯:=(L,∥⋅∥){\overline{L}}:=(L,\|\cdot\|) is called a metrized line bundle.

Models of varieties and line bundles give rise to an important class of metrics. To introduce and study these metrics, we first consider the notion of model of varieties. Write S=Spec⁡(K∘)S=\operatorname{Spec}(K^{\circ}). The scheme SS has two points: the special point oo and the generic point η\eta. Given a scheme 𝒳{\mathcal{X}} over SS, we set 𝒳o=𝒳×Spec⁡(k)\mathcal{X}_{o}=\mathcal{X}\times\operatorname{Spec}(k) and 𝒳η=𝒳×Spec⁡(K)\mathcal{X}_{\eta}=\mathcal{X}\times\operatorname{Spec}(K) for its special fibre and its generic fibre, respectively.

Definition 2.11.

A model over SS of XX is a flat scheme 𝒳{\mathcal{X}} of finite type over SS together with a fixed isomorphism X≃𝒳ηX\simeq\mathcal{X}_{\eta}. This isomorphism is part of the model, and so we can identify 𝒳η{\mathcal{X}}_{\eta} with XX. When XX is proper, we say that the model is proper whenever the scheme 𝒳{\mathcal{X}} is proper over SS.

Given a model 𝒳{\mathcal{X}} of XX, there is a reduction map defined on a closed subset of XanX^{{\text{\rm an}}} with values in 𝒳o\mathcal{X}_{o} [Ber90, §2.4]. This map can be described as follows. Let {𝒰i}i∈I\{\mathcal{U}_{i}\}_{i\in I} be a finite open affine cover of 𝒳\mathcal{X} by schemes over SS of finite type and, for each ii, let 𝒜i{\mathcal{A}}_{i} be a K∘K^{\circ}-algebra such that 𝒰i=Spec⁡(𝒜i){\mathcal{U}}_{i}=\operatorname{Spec}({\mathcal{A}}_{i}). Set Ui=𝒰i∩XU_{i}=\mathcal{U}_{i}\cap X and let CiC_{i} be the closed subset of UianU_{i}^{{\text{\rm an}}} defined as

(2.12) Ci={p∈Uian∣|a(p)|≤1,∀a∈𝒜i}C_{i}=\{p\in U_{i}^{{\text{\rm an}}}\mid|a(p)|\leq 1,\forall a\in{\mathcal{A}}_{i}\}

For each p∈Cip\in C_{i}, the prime ideal 𝔮p:={a∈𝒜i∣|a⁡(p)|<1}⊂𝒜i\mathfrak{q}_{p}:=\{a\in{\mathcal{A}}_{i}\mid|a(p)|<1\}\subset{\mathcal{A}}_{i} contains K∘⁣∘​𝒜iK^{\circ\circ}{\mathcal{A}}_{i} and so it determines a point red⁡(p):=𝔮p/K∘⁣∘​𝒜i∈𝒰i,o⊂𝒳o{\operatorname{red}}(p):=\mathfrak{q}_{p}/K^{\circ\circ}{\mathcal{A}}_{i}\in\mathcal{U}_{i,o}\subset\mathcal{X}_{o}. Consider the closed subset C=⋃iCi⊂XanC=\bigcup_{i}C_{i}\subset X^{{\text{\rm an}}}. The above maps glue together to define a map

(2.13) red:C⟶𝒳o.{\operatorname{red}}\colon C\longrightarrow\mathcal{X}_{o}.

This map is surjective and anti-continuous, in the sense that the preimages of the open subsets are closed [Ber90, §2.4]. For each irreducible component VV of 𝒳o{\mathcal{X}}_{o}, there is a unique point ξV∈C\xi_{V}\in C such that

(2.14) red⁡(ξV)=ηV,{\operatorname{red}}(\xi_{V})=\eta_{V},

where ηV\eta_{V} denotes the generic point of VV [Ber90, Proposition 2.4.4]. The finite subset {ξV}V⊂Xan\{\xi_{V}\}_{V}\subset X^{{\text{\rm an}}} is called the Shilov boundary of XanX^{{\text{\rm an}}}. Observe that it depends on the choice of 𝒳{\mathcal{X}}.

If both XX and 𝒳{\mathcal{X}} are proper, then C=XanC=X^{{\text{\rm an}}} and the reduction map is defined on the whole of XanX^{{\text{\rm an}}}. If both XX and 𝒳{\mathcal{X}} are normal, we can compute the Shilov boundary. Let VV be an irreducible component of 𝒳o\mathcal{X}_{o} and choose a finite type affine open subset 𝒰=Spec⁡(𝒜)⊂𝒳\mathcal{U}=\operatorname{Spec}({\mathcal{A}})\subset{\mathcal{X}} containing ηV\eta_{V}. Put A=𝒜⊗K∘K{A}={\mathcal{A}}\otimes_{K^{\circ}}K and U=𝒰∩XU={\mathcal{U}}\cap X. Then the point ξV∈U⊂Xan\xi_{V}\in U\subset X^{{\text{\rm an}}} is the multiplicative seminorm on AA given by

(2.15) |a⁡(ξV)|=|ϖ|ordV⁡(a)/ordV⁡(ϖ),|a(\xi_{V})|=|\varpi|^{{\operatorname{ord}}_{V}(a)/{\operatorname{ord}}_{V}(\varpi)},

for each a∈Aa\in A, where ordV⁡(f){\operatorname{ord}}_{V}(f) is the order of ff at the generic point of VV.

Next we recall the definition of models of line bundles. Let LL be a line bundle on XX.

Definition 2.16.

A model over SS of (X,L)(X,L) is a triple (𝒳,ℒ,e)(\mathcal{X},\mathcal{L},e), where 𝒳\mathcal{X} is a model over SS of XX, ℒ\mathcal{L} is a line bundle on 𝒳\mathcal{X} and e≥1e\geq 1 is an integer, together with a fixed isomorphism ℒ|X≃L⊗e\mathcal{L}|_{X}\simeq L^{\otimes e}. When e=1e=1, the model (𝒳,ℒ,1)(\mathcal{X},\mathcal{L},1) will be denoted (𝒳,ℒ)(\mathcal{X},\mathcal{L}) for short. A model of (𝒳,ℒ,e)(\mathcal{X},\mathcal{L},e) is called proper whenever 𝒳{\mathcal{X}} is proper.

We assume that the variety XX is proper for the rest of this section. To a proper model of a line bundle we can associate a metric.

Definition 2.17.

Let (𝒳,ℒ,e)({\mathcal{X}},{\mathcal{L}},e) be a proper model of (X,L)(X,L). Let ss be a local section of LanL^{{\text{\rm an}}} defined at a point p∈Xanp\in X^{{\text{\rm an}}}. Let 𝒰⊂𝒳\mathcal{U}\subset{\mathcal{X}} be a trivializing open neighbourhood of red⁡(p){\operatorname{red}}(p) and σ\sigma a generator of ℒ|𝒰{\mathcal{L}}|_{{\mathcal{U}}}. Let U=𝒰∩XU={\mathcal{U}}\cap X and λ∈𝒪Uan\lambda\in{\mathcal{O}}_{U^{{\text{\rm an}}}} such that s⊗e=λ​σs^{\otimes e}=\lambda\sigma on UanU^{{\text{\rm an}}}. Then, the metric induced by the proper model (𝒳,ℒ,e)({\mathcal{X}},{\mathcal{L}},e) on LanL^{{\text{\rm an}}},, denoted ∥⋅∥𝒳,ℒ,e\|\cdot\|_{{\mathcal{X}},{\mathcal{L}},e}, is given by

‖s⁡(p)‖𝒳,ℒ,e=|λ⁡(p)|1/e.\|s(p)\|_{{\mathcal{X}},{\mathcal{L}},e}=|\lambda(p)|^{1/e}.

This definition does neither depend on the choice of the open set 𝒰{\mathcal{U}} nor of the section σ\sigma, and it gives a metric on LanL^{{\text{\rm an}}}. The metrics on LanL^{{\text{\rm an}}} obtained in this way are called algebraic, and a pair L¯:=(L,∥⋅∥𝒳,ℒ,e){\overline{L}}:=(L,\|\cdot\|_{{\mathcal{X}},{\mathcal{L}},e}) is called an algebraic metrized line bundle.

Different models may give rise to the same metric.

Proposition 2.18.

Let (𝒳,ℒ,e)(\mathcal{X},\mathcal{L},e) and (𝒳′,ℒ′,e′)(\mathcal{X}^{\prime},{\mathcal{L}}^{\prime},e^{\prime}) be proper models of (X,L)(X,L), and f:𝒳′→𝒳f\colon\mathcal{X}^{\prime}\to\mathcal{X} a morphism of models such that (ℒ′)⊗e≃f∗​ℒ⊗e′({\mathcal{L}}^{\prime})^{\otimes e}\simeq f^{\ast}{\mathcal{L}}^{\otimes e^{\prime}}. Then the metrics on LanL^{{\text{\rm an}}} induced by both models agree.

Proof.

Let ss be a local section of LanL^{{\text{\rm an}}} defined on a point p∈Xanp\in X^{{\text{\rm an}}}. Let 𝒰⊂𝒳{\mathcal{U}}\subset{\mathcal{X}} be a trivializing open neighbourhood of red𝒳⁡(p){\operatorname{red}}_{{\mathcal{X}}}(p), the reduction of pp with respect to the model 𝒳{\mathcal{X}}, and σ\sigma a generator of ℒ|𝒰{\mathcal{L}}|_{{\mathcal{U}}}. Let λ\lambda be an analytic function on (𝒰∩X)an({\mathcal{U}}\cap X)^{{\text{\rm an}}} such that s⊗e=λ​σs^{\otimes e}=\lambda\sigma.

We have that red𝒳′⁡(p)=f−1​(red⁡(p)){\operatorname{red}}_{{\mathcal{X}}^{\prime}}(p)=f^{-1}({\operatorname{red}}(p)) and 𝒰′:=f−1​(𝒰){\mathcal{U}}^{\prime}:=f^{-1}({\mathcal{U}}) is a trivializing open set of ℒ′⊗e{\mathcal{L}}^{\prime\otimes e} with generator f∗​σ⊗e′f^{*}\sigma^{\otimes e^{\prime}}. Then s⊗e​e′=λe′​f∗​σ⊗e′s^{\otimes ee^{\prime}}=\lambda^{e^{\prime}}f^{*}\sigma^{\otimes e^{\prime}} on (𝒰′∩X)an=(𝒰∩X)an({\mathcal{U}}^{\prime}\cap X)^{{\text{\rm an}}}=({\mathcal{U}}\cap X)^{{\text{\rm an}}}. Now the proposition follows directly from Definition 2.17. ∎

The inverse image of an algebraic metric is algebraic.

Proposition 2.19.

Let φ:X1→X2\varphi\colon X_{1}\to X_{2} be a morphism of proper algebraic varieties over KK and L¯2{\overline{L}}_{2} a line bundle on X2X_{2} equipped with an algebraic metric. Assume that X1X_{1} admits a proper model. Then φ∗​L¯2\varphi^{*}{\overline{L}}_{2}, the inverse image under φ\varphi of L¯2{\overline{L}}_{2}, is a line bundle on X1X_{1} equipped with an algebraic metric.

Proof.

Let (𝒳2,ℒ2,e)({\mathcal{X}}_{2},{\mathcal{L}}_{2},e) be a proper model of (X2,L2)(X_{2},L_{2}) which induces the metric in L¯2{\overline{L}}_{2}, and 𝒳1′{\mathcal{X}}^{\prime}_{1} be a proper model of X1X_{1}. Let 𝒳1{\mathcal{X}}_{1} be the Zariski closure of the graph of φ\varphi in 𝒳1′×S𝒳2{\mathcal{X}}_{1}^{\prime}\times_{S}{\mathcal{X}}_{2}. This is a proper model of X1X_{1} equipped with a morphism φS:𝒳1→𝒳2\varphi_{S}\colon{\mathcal{X}}_{1}\to{\mathcal{X}}_{2}. Then (𝒳1,φS∗​ℒ2,e)({\mathcal{X}}_{1},\varphi_{S}^{*}{\mathcal{L}}_{2},e) is a proper model of (X1,φ∗​L2)(X_{1},\varphi^{*}L_{2}) which induces the metric of φ∗​L¯2\varphi^{*}{\overline{L}}_{2}. ∎

Next we give a second description of an algebraic metric. As before, let XX be a proper variety over KK and LL a line bundle on XX, and ∥⋅∥𝒳,ℒ,e\|\cdot\|_{\mathcal{X},\mathcal{L},e} an algebraic metric on LanL^{{\text{\rm an}}}. Let p∈Xanp\in X^{{\text{\rm an}}} and put H=ℋ⁡(p)H=\mathscr{H}(p), which is a complete extension of KK. Let H∘H^{\circ} be its valuation ring, and oo and η\eta the special and the generic point of Spec⁡(H∘)\operatorname{Spec}(H^{\circ}), respectively. The point pp induces a morphism of schemes Spec⁡(H)→X\operatorname{Spec}(H)\to X. By the valuative criterion of properness, there is a unique extension

(2.20) p~:Spec⁡(H∘)⟶𝒳.\widetilde{p}\colon\operatorname{Spec}(H^{\circ})\longrightarrow\mathcal{X}.

It satisfies p~​(η)=π​(p)\widetilde{p}(\eta)=\pi(p), where π:Xan→X\pi\colon X^{{\text{\rm an}}}\rightarrow X is the natural map introduced at the beginning of §2.2, and p~​(o)=red⁡(p)\widetilde{p}(o)={\operatorname{red}}(p).

Proposition 2.21.

With notation as above, let ss be a local section of LL in a neighbourhood of π⁡(p)\pi(p). Then

(2.22) ∥s(p)∥𝒳,ℒ,e=inf{|a|1/e|a∈H×,a−1p~∗s⊗e∈p~∗ℒ}.\|s(p)\|_{\mathcal{X},\mathcal{L},e}=\inf\big\{|a|^{1/e}\big|a\in H^{\times},a^{-1}{\widetilde{p}}^{\ast}s^{\otimes e}\in\widetilde{p}^{\ast}\mathcal{L}\big\}.
Proof.

Write ∥⋅∥=∥⋅∥𝒳,ℒ,e\|\cdot\|=\|\cdot\|_{\mathcal{X},\mathcal{L},e} for short. Let 𝒰=Spec⁡(𝒜)∋red⁡(p)\mathcal{U}=\operatorname{Spec}({\mathcal{A}})\ni{\operatorname{red}}(p) be an open affine trivializing set of ℒ\mathcal{L} and σ\sigma be a generator of ℒ|𝒰\mathcal{L}|_{{\mathcal{U}}}. Then s⊗e=λ​σs^{\otimes e}=\lambda\sigma with λ\lambda in the fraction field of 𝒜{\mathcal{A}}. We have that λ⁡(p)∈H\lambda(p)\in H and, by definition, ‖s⁡(p)‖=|λ⁡(p)|1/e\|s(p)\|=|\lambda(p)|^{1/e}. If λ⁡(p)=0\lambda(p)=0, the equation is clearly satisfied. Denote temporarily by CC the right-hand side of (2.22). If λ⁡(p)≠0\lambda(p)\not=0,

λ​(p)−1​p~∗​s⊗e=p~∗​σ∈p~∗​ℒ.\lambda(p)^{-1}\widetilde{p}^{\ast}s^{\otimes e}=\widetilde{p}^{\ast}\sigma\in{\widetilde{p}}^{*}\mathcal{L}.

Hence ‖s⁡(p)‖≥C\|s(p)\|\geq C. Moreover, if a∈H×a\in H^{\times} is such that a−1​p~∗​s⊗e∈p~∗​ℒa^{-1}\widetilde{p}^{\ast}s^{\otimes e}\in{\widetilde{p}}^{*}\mathcal{L}, then there is an element α∈H∘∖{0}\alpha\in H^{\circ}\setminus\{0\} with a−1​p~∗​s⊗e=α​p~∗​σa^{-1}\widetilde{p}^{\ast}s^{\otimes e}=\alpha\widetilde{p}^{\ast}\sigma. Therefore, a=λ⁡(p)/αa=\lambda(p)/\alpha and |a|1/e=|λ⁡(p)|1/e/|α|1/e≥|λ⁡(p)|1/e|a|^{1/e}=|\lambda(p)|^{1/e}/|\alpha|^{1/e}\geq|\lambda(p)|^{1/e}. Thus, ‖s⁡(p)‖≤C\|s(p)\|\leq C. ∎

We give a third description of an algebraic metric in terms of intersection theory that makes evident the relationship with higher dimensional Arakelov theory. Let (𝒳,ℒ,e)({\mathcal{X}},{\mathcal{L}},e) be a proper model of (X,L)(X,L) and ι:𝒴→𝒳\iota\colon{\mathcal{Y}}\to{\mathcal{X}} a closed algebraic curve. Let 𝒴~{\widetilde{{\mathcal{Y}}}} be the normalization of 𝒴{\mathcal{Y}} and ι~:𝒴~→𝒳{\widetilde{\iota}}\colon{\widetilde{{\mathcal{Y}}}}\to{\mathcal{X}} and ρ:𝒴~→Spec⁡(K∘)\rho\colon{\widetilde{{\mathcal{Y}}}}\to\operatorname{Spec}(K^{\circ}) the induced morphisms. Let ss be a rational section of ℒ{\mathcal{L}} such that div⁡(s)\operatorname{div}(s) intersects properly 𝒴{\mathcal{Y}}. Then the intersection number (ι⋅div⁡(s))(\iota\cdot\operatorname{div}(s)) is defined as

(ι⋅div⁡(s))=deg⁡(ρ∗​(div⁡(ι~∗​s))).(\iota\cdot\operatorname{div}(s))=\deg(\rho_{\ast}(\operatorname{div}({\widetilde{\iota}}^{\ast}s))).
Proposition 2.23.

With the above notation, let p∈Xalganp\in X^{{\text{\rm an}}}_{{\text{\rm alg}}}. Let p~\widetilde{p} as in (2.20). This is a closed algebraic curve. Let ss be a local section of LL defined at pp and such that s⁡(p)≠0s(p)\not=0. Then

log⁡‖s⁡(p)‖log⁡|ϖ|=(p~⋅div⁡(s⊗e))e[ℋ(p):K].\frac{\log\|s(p)\|}{\log|\varpi|}=\frac{(\widetilde{p}\cdot\operatorname{div}(s^{\otimes e}))}{e[\mathscr{H}(p):K]}.
Proof.

We keep the notation in the proof of Proposition 2.21. In particular, s⊗e=λ​σs^{\otimes e}=\lambda\sigma with λ\lambda in the fraction field of 𝒜{\mathcal{A}}, and ℋ⁡(p)=H\mathscr{H}(p)=H. We verify that

log⁡‖s⁡(p)‖log⁡|ϖ|=log⁡|λ⁡(p)|e​log⁡|ϖ|=log⁡|NH/K⁡(λ⁡(p))|e[H:K]log|ϖ|=ordϖ⁡(NH/K⁡(λ⁡(p)))e[H:K]\frac{\log\|s(p)\|}{\log|\varpi|}=\frac{\log|\lambda(p)|}{e\log|\varpi|}=\frac{\log|\operatorname{N}_{H/K}(\lambda(p))|}{e[H:K]\log|\varpi|}=\frac{{\operatorname{ord}}_{\varpi}(\operatorname{N}_{H/K}(\lambda(p)))}{e[H:K]}

and

(p~⋅div⁡(s⊗e))=deg⁡(ρ∗​(div⁡(p~∗​s⊗e)))=deg⁡(ρ∗​(div⁡(λ⁡(p))))=deg⁡(div⁡(NH/K⁡(λ⁡(p))))=ordϖ⁡(NH/K⁡(λ⁡(p))),({\widetilde{p}}\cdot\operatorname{div}(s^{\otimes e}))=\deg(\rho_{\ast}(\operatorname{div}({\widetilde{p}}^{\ast}s^{\otimes e})))=\deg(\rho_{\ast}(\operatorname{div}(\lambda(p))))\\ =\deg(\operatorname{div}(\operatorname{N}_{H/K}(\lambda(p))))={\operatorname{ord}}_{\varpi}(\operatorname{N}_{H/K}(\lambda(p))),

which proves the statement. ∎

Example 2.24.

Let X=ℙK0=Spec⁡(K)X=\mathbb{P}^{0}_{K}=\operatorname{Spec}(K). A line bundle LL on XX is necessarily trivial, that is, L≃KL\simeq K. Consider the model (𝒳,ℒ,e)({\mathcal{X}},{\mathcal{L}},e) of (X,L)(X,L) given by 𝒳=Spec⁡(K∘){\mathcal{X}}=\operatorname{Spec}(K^{\circ}), e≥1e\geq 1, and ℒ{\mathcal{L}} a free K∘K^{\circ}-submodule of L⊗eL^{\otimes e} of rank one. Let v∈L⊗ev\in L^{\otimes e} be a basis of ℒ{\mathcal{L}}. For a section ss of LL we can write s⊗e=α​vs^{\otimes e}=\alpha v with α∈K\alpha\in K. Hence,

‖s‖=|α|1/e.\|s\|=|\alpha|^{1/e}.

All algebraic metrics on LanL^{{\text{\rm an}}} can be obtained in this way.

Example 2.25.

Let X=ℙKnX=\mathbb{P}_{K}^{n} and L=𝒪⁡(1)L=\mathcal{O}(1), the universal line bundle of ℙKn{\mathbb{P}_{K}^{n}}. As a model for (X,L)(X,L) we consider 𝒳=ℙK∘n\mathcal{X}=\mathbb{P}_{K^{\circ}}^{n}, the projective space over Spec⁡(K∘)\operatorname{Spec}(K^{\circ}), ℒ=𝒪ℙK∘n​(1)\mathcal{L}=\mathcal{O}_{\mathbb{P}_{K^{\circ}}^{n}}(1), and e=1e=1. A rational section ss of LL can be identified with a homogeneous rational function ρs∈K⁡(x0,…,xn)\rho_{s}\in K(x_{0},\dots,x_{n}) of degree 1.

Let p=(p0:…:pn)∈(ℙKn)an∖div(s)p=(p_{0}:\dots:p_{n})\in(\mathbb{P}_{K}^{n})^{\text{\rm an}}\setminus\operatorname{div}(s) and set H=ℋ⁡(p)H=\mathscr{H}(p). Let i0i_{0} be such that |pi0|=maxi⁡{|pi|}|p_{i_{0}}|=\max_{i}\{|p_{i}|\}. Take U≃𝔸KnU\simeq\mathbb{A}_{K}^{n} (respectively 𝒰≃𝔸K∘n\mathcal{U}\simeq\mathbb{A}_{K^{\circ}}^{n}) as the affine set xi0≠0x_{i_{0}}\not=0 over HH (respectively H∘H^{\circ}). The point pp corresponds to the algebraic morphism

p∗:K⁡[X0,…,Xi0−1,Xi0+1,…,Xn]⟶Hp^{\ast}\colon K[X_{0},\dots,X_{i_{0}-1},X_{i_{0}+1},\dots,X_{n}]\longrightarrow H

that sends XiX_{i} to pi/pi0p_{i}/p_{i_{0}}. The extension p~{\widetilde{p}} factors through the algebraic morphism

p~∗:K∘​[X1,…,Xi0−1,Xi0+1,…,Xn]⟶H∘,{\widetilde{p}}^{\ast}\colon K^{\circ}[X_{1},\dots,X_{i_{0}-1},X_{i_{0}+1},\dots,X_{n}]\longrightarrow H^{\circ},

with the same definition. Then

‖s⁡(p)‖\displaystyle||s(p)|| =inf{|z||z∈H×,z−1p~∗s∈p~∗ℒ}\displaystyle=\inf\big\{|z|\ \big|z\in H^{\times},z^{-1}{\widetilde{p}}^{\ast}s\in{\widetilde{p}}^{*}{\mathcal{L}}\big\}
=inf{|z||z∈H×,z−1ρs(p0/pi0,…,1,…,pn/pi0)∈H∘}\displaystyle=\inf\big\{|z|\ \big|z\in H^{\times},z^{-1}\rho_{s}(p_{0}/p_{i_{0}},\dots,1,\dots,p_{n}/p_{i_{0}})\in H^{\circ}\big\}
=|ρr​(p0,…,pn)pi0|\displaystyle=\left|\frac{\rho_{r}(p_{0},\dots,p_{n})}{p_{i_{0}}}\right|
=|ρr​(p0,…,pn)|maxi⁡{|pi|}.\displaystyle=\frac{|\rho_{r}(p_{0},\dots,p_{n})|}{\max_{i}\{|p_{i}|\}}.

We call this the canonical metric of 𝒪​(1)an{\mathcal{O}}(1)^{{\text{\rm an}}} and we denote it by ∥⋅∥can\|\cdot\|_{{\operatorname{can}}}.

Many other algebraic metrics can be obtained from Example 2.25, by considering maps of varieties to projective spaces. Let XX be a proper variety over KK equipped with a line bundle LL such that L⊗eL^{\otimes e} is generated by global sections for an integer e≥1e\geq 1. A set of global sections in Γ⁡(X,L⊗e)\Gamma(X,L^{\otimes e}) that generates L⊗eL^{\otimes e} induces a morphism φ:X→ℙKn\varphi\colon X\to\mathbb{P}_{K}^{n} and, by inverse image, a metric φ∗∥⋅∥can\varphi^{*}\|\cdot\|_{{\operatorname{can}}} on LL. If XX admits a a proper model, Proposition 2.19 shows that this metric is algebraic.

Now we recall the notion of semipositivity for algebraic metrics. A curve CC in 𝒳{\mathcal{X}} is vertical if it is contained in 𝒳o{\mathcal{X}}_{o}.

Definition 2.26.

Let ∥⋅∥\|\cdot\| be an algebraic metric on LL and set L¯=(L,∥⋅∥){\overline{L}}=(L,\|\cdot\|). We say that L¯{\overline{L}} is semipositive if there is a model (𝒳,ℒ,e)(\mathcal{X},\mathcal{L},e) of (X,L)(X,L) that induces the metric such that, for every vertical curve CC in 𝒳{\mathcal{X}},

degℒ⁡(C)≥0.\deg_{\mathcal{L}}(C)\geq 0.

With the hypothesis in Proposition 2.19, the inverse image of a semipositive algebraic metric is also a semipositive algebraic metric.

Example 2.27.

The canonical metric in Example 2.25 is semipositive: for a vertical curve CC, its degree with respect to 𝒪ℙK∘n​(1){\mathcal{O}}_{\mathbb{P}^{n}_{K^{\circ}}}(1) equals its degree with respect to the restriction of this model to the special fibre. This restriction identifies with 𝒪ℙkn​(1){\mathcal{O}}_{\mathbb{P}^{n}_{k}}(1), the universal line bundle of ℙkn\mathbb{P}^{n}_{k}, which is ample. Hence all the metrics obtained by inverse image of the canonical metric of 𝒪​(1)an{\mathcal{O}}(1)^{{\text{\rm an}}} are also semipositive.

Finally, we recall the definition of the signed measures associated with algebraic metrics.

Definition 2.28.

Let L¯i{\overline{L}}_{i}, i=0,…,d−1i=0,\dots,d-1, be line bundles on XX equipped with algebraic metrics. For each ii, choose a model (𝒳i,ℒi,ei)(\mathcal{X}_{i},\mathcal{L}_{i},e_{i}) that realizes the metric of L¯i{\overline{L}}_{i}. We can assume without loss of generality that the models 𝒳i\mathcal{X}_{i} agree with a common model 𝒳\mathcal{X}. Let YY be a dd-dimensional subvariety of XX and YanY^{{\text{\rm an}}} its analytification. Let 𝒴⊂𝒳\mathcal{Y}\subset\mathcal{X} be the closure of YY, 𝒴~{\widetilde{\mathcal{Y}}} be its normalization, 𝒴~o{\widetilde{{\mathcal{Y}}}}_{o} its special fibre, and 𝒴~o(0){\widetilde{{\mathcal{Y}}}}_{o}^{(0)} the set of irreducible components of this special fibre. For each V∈𝒴~o(0)V\in{\widetilde{{\mathcal{Y}}}}_{o}^{(0)}, consider the point ξV∈Yan\xi_{V}\in Y^{{\text{\rm an}}} defined by (2.15). Let δξV\delta_{\xi_{V}} be the Dirac delta measure on XanX^{{\text{\rm an}}} supported on ξV\xi_{V}. We define a discrete signed measure on XanX^{{\text{\rm an}}} by

(2.29) c1⁡(L¯0)∧⋯∧c1⁡(L¯d−1)∧δY=∑V∈𝒴~o(0)ordV⁡(ϖ)​degℒ0,…,ℒd−1⁡(V)e0​…​ed−1​δξV.\operatorname{c}_{1}({\overline{L}}_{0})\land\dots\land\operatorname{c}_{1}({\overline{L}}_{d-1})\land\delta_{Y}=\sum_{V\in{\widetilde{{\mathcal{Y}}}}_{o}^{(0)}}{\operatorname{ord}}_{V}(\varpi)\frac{\deg_{\mathcal{L}_{0},\dots,\mathcal{L}_{d-1}}(V)}{e_{0}\dots e_{d-1}}\delta_{\xi_{V}}.

This notion extends by linearity to the group of dd-dimensional cycles of XX.

This signed measure only depends on the metrics and not on the particular choice of models [Cha06, Proposition 2.7]. Observe that ordV⁡(ϖ){\operatorname{ord}}_{V}(\varpi) is the multiplicity of the component VV in 𝒴~o{\widetilde{{\mathcal{Y}}}}_{o} and that the total mass of this measure equals degL0,…,Ld−1⁡(Y)\deg_{L_{0},\dots,L_{d-1}}(Y). If L¯i{\overline{L}}_{i} is semipositive for all ii and YY is effective, this signed measure is a measure.

Remark 2.30.

The above measure was introduced by Chambert-Loir [Cha06]. For the subvarieties of a projective space equipped with the canonical metric, it is also possible to define similar measures through the theory of Chow forms, see [Phi94].

2.4. Approachable and integrable metrics, measures and local heights

Let KK be either ℝ\mathbb{R} or ℂ\mathbb{C} (the Archimedean case) as in §2.1, or a complete field with respect to a nontrivial non-Archimedean absolute value (the non-Archimedean case) as in §2.3. Let XX be a proper variety over KK. Its analytification XanX^{{\text{\rm an}}} will be a complex analytic space in the Archimedean case (equipped with an anti-linear involution when K=ℝK=\mathbb{R}), or an analytic space in the sense of Berkovich, in the non-Archimedean case. A metrized line bundle on XX is a pair L¯=(L,∥⋅∥){\overline{L}}=(L,\|\cdot\|), where LL is a line bundle on XX and ∥⋅∥\|\cdot\| is a metric on LanL^{{\text{\rm an}}}. Recall that the operations on line bundles of tensor product, dual and inverse image under a morphism extend to metrized line bundles.

Given two metrics ∥⋅∥\|\cdot\| and ∥⋅∥′\|\cdot\|^{\prime} on LanL^{{\text{\rm an}}}, their quotient defines a continuous function Xan→ℝ>0X^{{\text{\rm an}}}\to\mathbb{R}_{>0} given by ‖s⁡(p)‖/‖s⁡(p)‖′\|s(p)\|/\|s(p)\|^{\prime} for any local section ss of LL not vanishing at pp. The distance between ∥⋅∥\|\cdot\| and ∥⋅∥′\|\cdot\|^{\prime} is defined as the supremum of the absolute value of the logarithm of this function. In other words,

dist(∥⋅∥,∥⋅∥′)=supp∈Xan∖div⁡(s)|log(∥s(p)∥/∥s(p)∥′)|,\operatorname{dist}(\|\cdot\|,\|\cdot\|^{\prime})=\sup_{p\in X^{\text{\rm an}}\setminus\operatorname{div}(s)}|\log(\|s(p)\|/\|s(p)\|^{\prime})|,

for any non-zero rational section ss of LL.

Definition 2.31.

Let L¯=(L,∥⋅∥){\overline{L}}=(L,\|\cdot\|) be a metrized line bundle on XX. The metric ∥⋅∥\|\cdot\| is approachable if there exists a sequence of semipositive smooth (in the Archimedean case) or semipositive algebraic (in the non-Archimedean case) metrics (∥⋅∥l)l≥0(\|\cdot\|_{l})_{l\geq 0} on LanL^{{\text{\rm an}}} such that

liml→∞dist(∥⋅∥,∥⋅∥l)=0.\lim_{l\to\infty}\operatorname{dist}(\|\cdot\|,\|\cdot\|_{l})=0.

If this is the case, we say that L¯{\overline{L}} is approachable. This metrized line bundle is integrable if there are approachable line bundles M¯{\overline{M}}, N¯{\overline{N}} such that L¯=M¯⊗N¯−1{\overline{L}}={\overline{M}}\otimes{\overline{N}}^{-1}.

The tensor product and the inverse image of approachable line bundles are also approachable. The tensor product, the dual and the inverse image of integrable line bundles are also integrable.

Example 2.32.

Let X=ℙnX=\mathbb{P}^{n} be the projective space over ℂ\mathbb{C} and L=𝒪⁡(1)L=\mathcal{O}(1). The canonical metric of 𝒪​(1)an\mathcal{O}(1)^{{\text{\rm an}}} is the metric given, for p=(p0:…:pn)∈ℙn(ℂ)p=(p_{0}:\dots:p_{n})\in\mathbb{P}^{n}(\mathbb{C}), by

‖s⁡(p)‖can=|ρs​(p0,…,pn)|maxi⁡{|pi|},\|s(p)\|_{\operatorname{can}}=\frac{|\rho_{s}(p_{0},\dots,p_{n})|}{\max_{i}\{|p_{i}|\}},

for any rational section ss of LL defined at pp and the homogeneous rational function ρs∈ℂ⁡(x0,…,xn)\rho_{s}\in\mathbb{C}(x_{0},\dots,x_{n}) associated to ss.

This is an approachable metric. Indeed, consider the mm-power map [m]:ℙn→ℙn[m]:\mathbb{P}^{n}\to\mathbb{P}^{n} defined as [m](p0:…:pn)=(p0m:…:pnm)[m](p_{0}:\dots:p_{n})=(p^{m}_{0}:\dots:p^{m}_{n}). The mm-th root of the inverse image by [m][m] of the Fubini-Study metric of 𝒪​(1)an\mathcal{O}(1)^{{\text{\rm an}}} is the semipositive smooth metric on LanL^{{\text{\rm an}}} given by

‖s⁡(p)‖m=|s⁡(p0,…,pn)|(∑i|pi|2​m)1/2​m.\|s(p)\|_{m}=\frac{|s(p_{0},\dots,p_{n})|}{(\sum_{i}|p_{i}|^{2m})^{1/2m}}.

The family of metrics obtained varying mm converges uniformly to the canonical metric.

Proposition 2.33.

Let YY be a dd-dimensional subvariety of XX and L¯i=(Li,∥⋅∥i){\overline{L}}_{i}=(L_{i},\|\cdot\|_{i}), i=0,…,d−1i=0,\dots,d-1, a collection of approachable metrized line bundles on XX. For each ii, let (∥⋅∥i,l)l≥0(\|\cdot\|_{i,l})_{l\geq 0} be a sequence of semipositive smooth (in the Archimedean case) or algebraic (in the non-Archimedean case) metrics on LianL_{i}^{{\text{\rm an}}} that converge to ∥⋅∥i\|\cdot\|_{i}. Then the measures c1(L0,∥⋅∥0,l)∧⋯∧c1(Ld−1,∥⋅∥d−1,l)∧δY\operatorname{c}_{1}(L_{0},\|\cdot\|_{0,l})\land\dots\land\operatorname{c}_{1}(L_{d-1},\|\cdot\|_{d-1,l})\wedge\delta_{Y} converge weakly to a measure on XanX^{\text{\rm an}}.

Proof.

The non-Archimedean case is proven in [Cha06, Proposition 2.7(b)] and in [Gub07, Proposition 3.12]. The Archimedean case can be proved similarly. ∎

Definition 2.34.

Let L¯i=(Li,∥⋅∥i){\overline{L}}_{i}=(L_{i},\|\cdot\|_{i}), i=0,…,d−1i=0,\dots,d-1, be a collection of approachable metrized line bundles on XX. For a dd-dimensional subvariety Y⊂XY\subset X, we denote by c1⁡(L¯0)∧⋯∧c1⁡(L¯d−1)∧δY\operatorname{c}_{1}({\overline{L}}_{0})\land\dots\land\operatorname{c}_{1}({\overline{L}}_{d-1})\wedge\delta_{Y} the limit measure in Proposition 2.33. For integrable bundles L¯i{\overline{L}}_{i} and a dd-dimensional cycle YY of XX, we can associate a signed measure c1⁡(L¯0)∧⋯∧c1⁡(L¯d−1)∧δY\operatorname{c}_{1}({\overline{L}}_{0})\land\dots\land\operatorname{c}_{1}({\overline{L}}_{d-1})\wedge\delta_{Y} on XanX^{{\text{\rm an}}} by multilinearity.

This signed measure behaves well under field extensions.

Proposition 2.35.

With the previous notation, let K′K^{\prime} be a finite extension of KK. Set (X′,Y′)=(X,Y)×Spec⁡(K′)(X^{\prime},Y^{\prime})=(X,Y)\times\operatorname{Spec}(K^{\prime}) and let φ:X′an→Xan\varphi\colon{X^{\prime}}^{{\text{\rm an}}}\to X^{{\text{\rm an}}} be the induced map. Let φ∗​L¯i\varphi^{\ast}{\overline{L}}_{i}, i=0,…,d−1i=0,\dots,d-1, be the line bundles with algebraic metrics on X′X^{\prime} obtained by base change. Then

φ∗​(c1⁡(φ∗​L¯0)∧⋯∧c1⁡(φ∗​L¯d−1)∧δY′)=c1⁡(L¯0)∧⋯∧c1⁡(L¯d−1)∧δY.\varphi_{\ast}\left(\operatorname{c}_{1}(\varphi^{\ast}{\overline{L}}_{0})\land\dots\land\operatorname{c}_{1}(\varphi^{\ast}{\overline{L}}_{d-1})\land\delta_{{Y^{\prime}}}\right)=\operatorname{c}_{1}({\overline{L}}_{0})\land\dots\land\operatorname{c}_{1}({\overline{L}}_{d-1})\land\delta_{Y}.
Proof.

This follows from [Gub07, Remark 3.10]. ∎

We also have the following functorial property.

Proposition 2.36.

Let φ:X′→X\varphi\colon X^{\prime}\to X be a morphism of proper varieties over KK, Y′Y^{\prime} a dd-dimensional cycle of X′X^{\prime}, and L¯i=(Li,∥⋅∥i){\overline{L}}_{i}=(L_{i},\|\cdot\|_{i}), i=0,…,d−1i=0,\dots,d-1, a collection of integrable metrized line bundles on XX. Then

φ∗​(c1⁡(φ∗​L¯0)∧⋯∧c1⁡(φ∗​L¯d−1)∧δY′)=c1⁡(L¯0)∧⋯∧c1⁡(L¯d−1)∧δφ∗​Y.\varphi_{\ast}\left(\operatorname{c}_{1}(\varphi^{\ast}{\overline{L}}_{0})\land\dots\land\operatorname{c}_{1}(\varphi^{\ast}{\overline{L}}_{d-1})\land\delta_{{Y^{\prime}}}\right)=\operatorname{c}_{1}({\overline{L}}_{0})\land\dots\land\operatorname{c}_{1}({\overline{L}}_{d-1})\land\delta_{\varphi_{*}Y}.
Proof.

In the non-Archimedean, this follows from [Gub07, Corollary 3.9(2)]. In the Archimedean case, this follows from the functoriality of Chern classes, the projection formula, and the continuity of direct image of measures. ∎

These signed measures allow us to integrate continuous functions on XanX^{{\text{\rm an}}}. Indeed, it is also possible to integrate certain functions with logarithmic singularities that play an important role in the definition of local heights.

Proposition 2.37.

Let YY be a dd-dimensional cycle of XX, L¯i=(Li,∥⋅∥i){\overline{L}}_{i}=(L_{i},\|\cdot\|_{i}), i=0,…,di=0,\dots,d, a collection of integrable metrized line bundles, and sds_{d} a rational section of LdL_{d} such that div⁡(sd)\operatorname{div}(s_{d}) intersects YY properly. Then log⁡‖sd‖\log\|s_{d}\| is integrable with respect to the measure c1⁡(L¯0)∧⋯∧c1⁡(L¯d−1)∧δY\operatorname{c}_{1}({\overline{L}}_{0})\land\dots\land\operatorname{c}_{1}({\overline{L}}_{d-1})\wedge\delta_{Y}.

Proof.

This is proved in [CT09, Theorem 4.1] for completions of number fields. The argument can be easily extended to cover the general case. ∎

Definition 2.38.

Let YY be a dd-dimensional cycle of XX and LiL_{i} a line bundle on XX and sis_{i} a rational section of LiL_{i}, i=0,…,di=0,\dots,d. We say that s0,…,sds_{0},\dots,s_{d} meet properly YY if, for all I⊂{0,…,d}I\subset\{0,\dots,d\},

dim(Y∩⋂i∈I|div⁡si|)=d−#​I.\dim\left(Y\cap\bigcap_{i\in I}|\operatorname{div}s_{i}|\right)=d-\#I.
Definition 2.39.

The local height on XX is the function that, to each dd-dimensional cycle YY and each family of integrable metrized line bundles with sections (L¯i,si)({\overline{L}}_{i},s_{i}), i=0,…,di=0,\dots,d, such that the sections meet YY properly, associates a real number hL¯0,…,L¯d⁡(Y,s0,…,sd)\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d}}(Y;s_{0},\dots,s_{d}) determined inductively by the properties:

  1. (1)

    h⁡(∅)=0\operatorname{h}(\emptyset)=0;

  2. (2)

    if YY is a cycle of dimension d≥0d\geq 0, then

    hL¯0,…,L¯d⁡(Y,s0,…,sd)=hL¯0,…,L¯d−1⁡(Y⋅div⁡sd,s0,…,sd−1)−∫Xanlog∥sd∥c1(L¯0)∧⋯∧c1(L¯d−1)∧δY.\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d}}(Y;s_{0},\dots,s_{d})=\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d-1}}(Y\cdot\operatorname{div}s_{d};s_{0},\dots,s_{d-1})\\ -\int_{X^{{\text{\rm an}}}}\log\|s_{d}\|\operatorname{c}_{1}(\overline{L}_{0})\land\dots\wedge\operatorname{c}_{1}(\overline{L}_{d-1})\land\delta_{Y}.

In particular, for p∈X⁡(K)∖|div⁡(s0)|p\in X(K)\setminus|\operatorname{div}(s_{0})|,

(2.40) hL¯0⁡(p;s0)=−log⁡‖s0​(p)‖.\operatorname{h}_{{\overline{L}}_{0}}(p;s_{0})=-\log\|s_{0}(p)\|.
Remark 2.41.

Definition 2.39 works better when the variety XX is projective. In this case, for every cycle YY there exist sections that meet YY properly, thanks to the moving lemma. This does not necessarily occur for arbitrary proper varieties. Nevertheless, we will be able to define the global height (Definition 2.56) of any cycle of a proper variety by using Chow’s lemma. Similarly we will be able to define the toric local height (Definition 6.1) of any cycle of a proper toric variety.

Remark 2.42.

When XX is regular and the metrics are smooth (in the Archimedean case) or algebraic (in the non-Archimedean case), the local heights of Definition 2.39 agree with the local heights that can be derived using the Gillet-Soulé arithmetic intersection product. In particular, in the Archimedean case, this local height agrees with the Archimedean contribution of the Arakelov global height introduced by Bost, Gillet and Soulé in [BGS94]. In the non-Archimedean case, the local height can be interpreted in terms of an intersection product. Assume that YY is prime and choose models (𝒳i,ℒi,ei)(\mathcal{X}_{i},\mathcal{L}_{i},e_{i}) of (X,Li)(X,L_{i}) that realize the algebraic metrics of L¯i{\overline{L}}_{i}. Without loss of generality, we may assume that all the models 𝒳i\mathcal{X}_{i} agree with a common model 𝒳\mathcal{X}. The sections si⊗eis^{\otimes e_{i}}_{i} can be seen as rational sections of ℒi\mathcal{L}_{i} over 𝒳\mathcal{X}. With the notations in Definition 2.28, the equation (2.15) implies that

log⁡‖sd​(ξV)‖=log⁡|ϖ|​ordV⁡(sd⊗ed)ed​ordv​(ϖ).\log\|s_{d}(\xi_{V})\|=\frac{\log|\varpi|{\operatorname{ord}}_{V}(s_{d}^{\otimes e_{d}})}{e_{d}{\operatorname{ord}}_{v}(\varpi)}.

Therefore, in this case the equation in Definition 2.39(2) can be written as

(2.43) hL¯0,…,L¯d⁡(Y,s0,…,sd)=hL¯0,…,L¯d−1⁡(Y⋅div⁡(sd),s0,…,sd−1)−log⁡|ϖ|e0​…​ed∑V∈𝒴~0(0)ordV(sd⊗ed)degℒ0,…,ℒd−1(V).\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d}}(Y;s_{0},\dots,s_{d})=\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d-1}}(Y\cdot\operatorname{div}(s_{d});s_{0},\dots,s_{d-1})\\ -\frac{\log|\varpi|}{e_{0}\dots e_{d}}\sum_{V\in{\widetilde{\mathcal{Y}}}_{0}^{(0)}}{\operatorname{ord}}_{V}(s_{d}^{\otimes e_{d}})\deg_{\mathcal{L}_{0},\dots,\mathcal{L}_{d-1}}(V).
Remark 2.44.

It is a fundamental observation by Zhang [Zha95b] that the non-Archimedean contribution of the Arakelov global height of a variety can be expressed in terms of a family of metrics. In particular, this global height only depends on the metrics and not on a particular choice of models, exhibiting the analogy between the Archimedean and non-Archimedean settings. The local heights were extended by Gubler [Gub02, Gub03] to non-necessarily discrete valuations and he also weakened the hypothesis of proper intersection.

Remark 2.45.

The local heights of Definition 2.39 agree with the local heights introduced by Gubler, see [Gub03, Proposition 3.5] for the Archimedean case and [Gub03, Remark 9.4] for the non-Archimedean case. In the Archimedean case, the local height in [Gub03] is defined in terms of a refined star product of Green currents based on [Bur94]. The hypothesis needed in Gubler’s definition of local heights are weaker than the ones we use. We have chosen the current definition because it is more elementary and suffices for our purposes.

Theorem 2.46.

The local height function satisfies the following properties.

  1. (1)

    It is symmetric and multilinear with respect to ⊗\otimes in the pairs (L¯i,si)({\overline{L}}_{i},s_{i}), i=0,…,di=0,\dots,d, provided that all terms are defined.

  2. (2)

    Let φ:X′→X\varphi\colon X^{\prime}\to X be a morphism of proper varieties over KK, YY a dd-dimensional cycle of X′X^{\prime}, and (L¯i,si)({\overline{L}}_{i},s_{i}) an integrable metrized line bundle on XX and a section, i=0,…,di=0,\dots,d. Then

    hφ∗​L¯0,…,φ∗​L¯d⁡(Y,φ∗​s0,…,φ∗​sd)=hL¯0,…,L¯d⁡(φ∗​Y,s0,…,sd),\operatorname{h}_{\varphi^{\ast}{\overline{L}}_{0},\dots,\varphi^{\ast}{\overline{L}}_{d}}(Y;\varphi^{\ast}s_{0},\dots,\varphi^{\ast}s_{d})=\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d}}(\varphi_{\ast}Y;s_{0},\dots,s_{d}),

    provided that both terms are defined.

  3. (3)

    Let ZZ be the zero-cycle Y⋅div(s0)⋯div(sd−1)Y\cdot\operatorname{div}(s_{0})\cdots\operatorname{div}(s_{d-1}) and ff a rational function such that the section f​sdfs_{d} meets ZZ properly. Then

    hL¯0,…,L¯d⁡(Y,s0,…,sd)−hL¯0,…,L¯d⁡(Y,s0,…,f​sd)=log⁡|f⁡(Z)|,\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d}}(Y;s_{0},\dots,s_{d})-\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d}}(Y;s_{0},\dots,fs_{d})=\log|f(Z)|,

    where, if Z=∑lml​plZ=\sum_{l}m_{l}p_{l}, then f⁡(Z)=∏lf​(pl)mlf(Z)=\prod_{l}f(p_{l})^{m_{l}}.

  4. (4)

    Let L′¯d=(Ld,∥⋅∥′){\overline{L^{\prime}}}_{d}=(L_{d},\|\cdot\|^{\prime}) be another choice of metric. Then

    hL¯0,…,L¯d⁡(Y,s0,…,sd)−hL¯0,…,L¯d′⁡(Y,s0,…,sd)=−∫Ylog(∥sd(p)∥/∥sd(p)∥′)c1(L¯0)∧⋯∧c1(L¯d−1)∧δY\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d}}(Y;s_{0},\dots,s_{d})-\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}^{\prime}_{d}}(Y;s_{0},\dots,s_{d})=\\ -\int_{Y}\log(\|s_{d}(p)\|/\|s_{d}(p)\|^{\prime})\operatorname{c}_{1}(\overline{L}_{0})\land\dots\wedge\operatorname{c}_{1}(\overline{L}_{d-1})\land\delta_{Y}

    is independent of the choice of sections.

Proof.

In the Archimedean case, statement (1) is [Gub03, Proposition 3.4], statement (2) is [Gub03, Proposition 3.6]. In the non-Archimedean case, statement (1) and (2) are [Gub03, Remark 9.3]. The other two statements follow easily from the definition. ∎

2.5. Adelic metrics and global heights

To define global heights, we first introduce the notion of adelic field, which is a generalization of the notion of global field. In [Gub03] one can find a more general theory of global heights based on the concept of MM-fields.

Definition 2.47.

Let 𝕂\mathbb{K} be a field and 𝔐𝕂\mathfrak{M}_{\mathbb{K}} a family of absolute values on 𝕂\mathbb{K} with real weights. For each v∈𝔐𝕂v\in\mathfrak{M}_{\mathbb{K}} we denote by |⋅|v|\cdot|_{v} the corresponding absolute value, by nv∈ℝn_{v}\in\mathbb{R} the weight, and by 𝕂v\mathbb{K}_{v} the completion of 𝕂\mathbb{K} with respect to |⋅|v|\cdot|_{v}. We say that (𝕂,𝔐𝕂)(\mathbb{K},\mathfrak{M}_{\mathbb{K}}) is an adelic field if

  1. (1)

    for each v∈𝔐𝕂v\in\mathfrak{M}_{\mathbb{K}}, the absolute value |⋅|v|\cdot|_{v} is Archimedean or associated to a nontrivial discrete valuation;

  2. (2)

    for each α∈𝕂×\alpha\in\mathbb{K}^{\times}, |α|v=1|\alpha|_{v}=1 except a for a finite number of vv.

Observe that the complete fields 𝕂v\mathbb{K}_{v} are either ℝ\mathbb{R}, ℂ\mathbb{C} or of the kind of fields considered in §2.3.

Definition 2.48.

Let (𝕂,𝔐𝕂)(\mathbb{K},\mathfrak{M}_{\mathbb{K}}) be an adelic field. For α∈𝕂×\alpha\in\mathbb{K}^{\times}, the defect of α\alpha is

def⁡(α)=∑v∈𝔐𝕂nv​log⁡|α|v.\operatorname{def}(\alpha)=\sum_{v\in\mathfrak{M}_{\mathbb{K}}}n_{v}\log|\alpha|_{v}.

Since def:𝕂×→ℝ\operatorname{def}\colon\mathbb{K}^{\times}\to\mathbb{R} is a group homomorphism, we have that def⁡(𝕂×)\operatorname{def}(\mathbb{K}^{\times}) is a subgroup of ℝ\mathbb{R}. If def⁡(𝕂×)=0\operatorname{def}(\mathbb{K}^{\times})=0, then 𝕂\mathbb{K} is said to satisfy the product formula. The group of global heights of 𝕂\mathbb{K} is ℝ/def⁡(𝕂×)\mathbb{R}/\!\operatorname{def}(\mathbb{K}^{\times}).

Let (𝕂,𝔐𝕂)(\mathbb{K},\mathfrak{M}_{\mathbb{K}}) be an adelic field and 𝔽\mathbb{F} a finite extension of 𝕂\mathbb{K}. For each v∈𝔐𝕂v\in\mathfrak{M}_{\mathbb{K}}, put 𝔐v\mathfrak{M}_{v} for the set of absolute values |⋅|w|\cdot|_{w} of 𝔽\mathbb{F} that extend |⋅|v|\cdot|_{v}, with weight

nw=[𝔽w:𝕂v][𝔽:𝕂]nv.n_{w}=\frac{[\mathbb{F}_{w}:\mathbb{K}_{v}]}{[\mathbb{F}:\mathbb{K}]}n_{v}.

Set 𝔐𝔽=∐v𝔐v\mathfrak{M}_{\mathbb{F}}=\coprod_{v}\mathfrak{M}_{v}. Then (𝔽,𝔐𝔽)(\mathbb{F},\mathfrak{M}_{\mathbb{F}}) is an adelic field and def(𝔽×)⊂1[𝔽:𝕂]def(𝕂×)\operatorname{def}(\mathbb{F}^{\times})\subset\frac{1}{[\mathbb{F}:\mathbb{K}]}\operatorname{def}(\mathbb{K}^{\times}). In particular, if 𝕂\mathbb{K} satisfies the product formula so does 𝔽\mathbb{F}.

Example 2.49.

Let 𝔐ℚ\mathfrak{M}_{\mathbb{Q}} be the set of places of ℚ\mathbb{Q}, where the corresponding absolute values are normalized in the standard way. Then (ℚ,𝔐ℚ)(\mathbb{Q},\mathfrak{M}_{\mathbb{Q}}) is an adelic field that satisfies the product formula. If 𝕂\mathbb{K} is a number field, by the construction above, we obtain an adelic field (𝕂,𝔐𝕂)(\mathbb{K},\mathfrak{M}_{\mathbb{K}}) which satisfies the product formula too.

Example 2.50.

Let BB be a irreducible projective variety over a field kk, which is regular in codimension 1, and LL an ample line bundle on BB. Set 𝕂=k⁡(B)\mathbb{K}=k(B). For a prime divisor vv on BB and α∈𝕂×\alpha\in\mathbb{K}^{\times}, we denote by ordv⁡(α){\operatorname{ord}}_{v}(\alpha) the order of α\alpha at vv. Fix a constant c>1c>1 and denote by 𝔐𝕂\mathfrak{M}_{\mathbb{K}} the set of prime divisors on BB. For each v∈𝔐𝕂v\in\mathfrak{M}_{\mathbb{K}}, the corresponding absolute value and weight are defined as

|α|v=c−ordv⁡(α),nv=degL⁡(v).|\alpha|_{v}=c^{-{\operatorname{ord}}_{v}(\alpha)},\quad n_{v}=\deg_{L}(v).

Then (𝕂,𝔐𝕂)(\mathbb{K},\mathfrak{M}_{\mathbb{K}}) is an adelic field. Moreover, 𝕂\mathbb{K} satisfies the product formula, since the degree of a principal divisor is zero,

Definition 2.51.

The adelic fields in examples 2.49 and 2.50 will be called global fields. For a finite subset S⊂𝔐𝕂S\subset\mathfrak{M}_{\mathbb{K}} containing the Archimedean places, we consider the Noetherian ring 𝕂S∘={α∈𝕂||α|v≤1,∀v∉S}\mathbb{K}^{\circ}_{S}=\{\alpha\in\mathbb{K}\,|\,|\alpha|_{v}\leq 1,\forall v\notin S\}.

Definition 2.52.

Let (𝕂,𝔐𝕂)(\mathbb{K},\mathfrak{M}_{\mathbb{K}}) be an adelic field. Let XX be a proper variety over 𝕂\mathbb{K} and LL a line bundle on XX. For each v∈𝔐𝕂v\in\mathfrak{M}_{\mathbb{K}} set Xv=X×Spec⁡(Kv)X_{v}=X\times\operatorname{Spec}(K_{v}) and Lv=L×Spec⁡(Kv)L_{v}=L\times\operatorname{Spec}(K_{v}).

  1. (1)

    A metric on LL is a family of metrics ∥⋅∥v\|\cdot\|_{v}, v∈𝔐𝕂v\in\mathfrak{M}_{\mathbb{K}}, where ∥⋅∥v\|\cdot\|_{v} is a metric on LvanL_{v}^{{\text{\rm an}}}. We will denote by L¯=(L,(∥⋅∥v)v){\overline{L}}=(L,(\|\cdot\|_{v})_{v}) the corresponding metrized line bundle. The metric is said to be approachable (respectively integrable) if the metrics ∥⋅∥v\|\cdot\|_{v} are approachable (respectively integrable) for all v∈𝔐𝕂v\in\mathfrak{M}_{\mathbb{K}}.

  2. (2)

    Suppose that (𝕂,𝔐𝕂)(\mathbb{K},\mathfrak{M}_{\mathbb{K}}) is a global field. A metric on LL is called quasi-algebraic if there exists a finite subset S⊂𝔐𝕂S\subset\mathfrak{M}_{\mathbb{K}} containing the Archimedean places, an integer e≥1e\geq 1 and a proper model (𝒳,ℒ,e)({\mathcal{X}},{\mathcal{L}},e) over 𝕂S∘\mathbb{K}^{\circ}_{S} of (X,L)(X,L) such that, for each v∉Sv\notin S, the metric ∥⋅∥v\|\cdot\|_{v} is induced by the localization of this model at vv.

Definition 2.53.

Let (𝕂,𝔐𝕂)(\mathbb{K},\mathfrak{M}_{\mathbb{K}}) be an adelic field, XX a proper variety over 𝕂\mathbb{K} and L¯i{\overline{L}}_{i}, i=0,…,di=0,\dots,d, a family of integrable metrized line bundles on XX. Let YY be a dd-dimensional cycle of XX. We say that YY is integrable with respect to L¯0,…,L¯d{\overline{L}}_{0},\dots,{\overline{L}}_{d} if there is a proper map φ:X′→X\varphi\colon X^{\prime}\to X, a cycle Y′Y^{\prime} of X′X^{\prime} such that φ∗​Y′=Y\varphi_{\ast}Y^{\prime}=Y, and rational sections sis_{i} of φ∗​Li\varphi^{\ast}L_{i}, i=0,…,di=0,\dots,d, that intersect Y′Y^{\prime} properly and such that for all but a finite number of v∈𝔐𝕂v\in\mathfrak{M}_{\mathbb{K}},

(2.54) hv,φ¯∗​L0,…,φ¯∗​Ld⁡(Y′,s0,…,sd)=0,\operatorname{h}_{v,{\overline{\varphi}}^{\ast}L_{0},\dots,{\overline{\varphi}}^{\ast}L_{d}}(Y^{\prime};s_{0},\dots,s_{d})=0,

where hv\operatorname{h}_{v} denotes the local height function on XvX_{v}.

The notion of integrability of cycles is stable under tensor product and inverse image of integrable metrized line bundles, thanks to Theorem 2.46(1,2). For an integrable cycle YY, the condition (2.54) is satisfied for any choice of morphism φ\varphi, cycle Y′Y^{\prime} and sections that intersect Y′Y^{\prime} properly, thanks to the definition of adelic field and Theorem 2.46(3).

We are mainly interested in global fields and quasi-algebraic metrics. In this case, all cycles are integrable.

Proposition 2.55.

Let (𝕂,𝔐𝕂)(\mathbb{K},\mathfrak{M}_{\mathbb{K}}) be a global field and XX a proper variety over 𝕂\mathbb{K} of dimension nn. Let d≤nd\leq n and let L¯i{\overline{L}}_{i}, i=0,…,di=0,\dots,d, be a family of line bundles with quasi-algebraic integrable metrics. Then every dd-dimensional cycle of XX is integrable with respect to L¯0,…,L¯d{\overline{L}}_{0},\dots,{\overline{L}}_{d}.

Proof.

It is enough to prove that every prime cycle is integrable. Applying the Chow Lemma to the support of the cycle and using that the inverse image of a quasi-algebraic metric is quasi-algebraic, we are reduced to the case when XX is projective.

We proceed by induction on dd. For d=−1d=-1, the statement is clear, and so we consider the case when d≥0d\geq 0. Let YY be a dd-dimensional cycle of XX and sis_{i}, i=0,…,di=0,\dots,d, rational sections of LiL_{i} that intersect YY properly. Let (𝒳,ℒd)({\mathcal{X}},{\mathcal{L}}_{d}) be a proper model over 𝕂S∘\mathbb{K}^{\circ}_{S} of (X,Ld⊗ed)(X,L_{d}^{\otimes e_{d}}). Then sd⊗eds_{d}^{\otimes e_{d}} is a non-zero rational section of ℒd{\mathcal{L}}_{d} and so it defines a finite number of vertical components. Hence, for all places v∉Sv\notin S which are not below any of these vertical components,

hv,L¯0,…,L¯d⁡(Y,s0,…,sd)=hv,L¯0,…,L¯d−1⁡(Y⋅div⁡(sd),s0,…,sd−1),\operatorname{h}_{v,{\overline{L}}_{0},\dots,{\overline{L}}_{d}}(Y;s_{0},\dots,s_{d})=\operatorname{h}_{v,{\overline{L}}_{0},\dots,{\overline{L}}_{d-1}}(Y\cdot\operatorname{div}(s_{d});s_{0},\dots,s_{d-1}),

thanks to the equation (2.43). The statement follows then from the inductive hypothesis. ∎

Definition 2.56.

Let XX be a proper variety over 𝕂\mathbb{K}, L¯0,…,L¯d{\overline{L}}_{0},\dots,{\overline{L}}_{d} integrable metrized line bundles on XX, and YY an integrable dd-dimensional cycle of XX. Let X′X^{\prime}, Y′Y^{\prime} and s0,…,sds_{0},\dots,s_{d} be as in Definition 2.53. The global height of YY with respect to s0,…,sds_{0},\dots,s_{d} is defined as

hL¯0,…,L¯d⁡(Y,s0,…,sd)=∑v∈𝔐𝕂nv​hv,φ∗​L¯0,…,φ∗​L¯d​(Y′,s0,…,sd)∈ℝ.\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d}}(Y;s_{0},\dots,s_{d})=\sum_{v\in\mathfrak{M}_{\mathbb{K}}}n_{v}\operatorname{h}_{v,\varphi^{\ast}{\overline{L}}_{0},\dots,\varphi^{\ast}{\overline{L}}_{d}}(Y^{\prime};s_{0},\dots,s_{d})\in\mathbb{R}.

The global height of YY, denoted hL¯0,…,L¯d⁡(Y)\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d}}(Y), is the class of hL¯0,…,L¯d⁡(Y,s0,…,sd)\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d}}(Y;s_{0},\dots,s_{d}) in the quotient group ℝ/def⁡(𝕂×)\mathbb{R}/\!\operatorname{def}(\mathbb{K}^{\times}).

The global height is well-defined as an element of ℝ/def⁡(𝕂×)\mathbb{R}/\!\operatorname{def}(\mathbb{K}^{\times}) because of Theorem 2.46(3). In particular, if 𝕂\mathbb{K} satisfies the product formula, the global height is a well-defined real number.

Theorem 2.57.

The global height of integrable cycles satisfies the following properties.

  1. (1)

    It is symmetric and multilinear with respect to tensor products of integrable metrized line bundles.

  2. (2)

    Let φ:X′→X\varphi\colon X^{\prime}\to X be a morphism of proper varieties over KK, L¯i{\overline{L}}_{i}, i=0,…,di=0,\dots,d, integrable metrized line bundles on XX, and YY an integrable dd-dimensional cycle of X′X^{\prime}. Then

    hφ∗​L¯0,…,φ∗​L¯d⁡(Y)=hL¯0,…,L¯d⁡(φ∗​Y).\operatorname{h}_{\varphi^{\ast}{\overline{L}}_{0},\dots,\varphi^{\ast}{\overline{L}}_{d}}(Y)=\operatorname{h}_{{\overline{L}}_{0},\dots,{\overline{L}}_{d}}(\varphi_{\ast}Y).
Proof.

This follows readily from Theorem 2.46(1,2). ∎

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