ScalingStacks

5. Algebraic approximation [00NC]

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. Algebraic approximation

In this section, we approximate the Banach algebra norm ⦀⋅⦀ϕX|Y\vvvert\mathord{\cdot}\vvvert_{\phi_{X|Y}} by affinoid norms. This algebraic approximation exploits subtle consequences of existence of a ultra-metric orthogonal basis in (Vn​(L),∥⋅∥n​ϕ)(V_{n}(L),\lVert\mathord{\cdot}\rVert_{n\phi}) for some n∈ℕn\in\mathbb{N}.

In this section (k,|⋅|)(k,\lvert\mathord{\cdot}\rvert) is assumed to be discretely valued. With this assumption, recall that d+1d+1 classes of real numbers {α⁡(p0),…,α⁡(pd)}\{\alpha(p_{0}),\dots,\alpha(p_{d})\} in ℝ/H⁡(k,|⋅|)\mathbb{R}/H(k,\lvert\mathord{\cdot}\rvert) are said to be ℚ\mathbb{Q}-independent if there exists no (a0,…,ad)∈ℚd+1(a_{0},\dots,a_{d})\in\mathbb{Q}^{d+1} and no p∈H⁡(k,|⋅|)p\in H(k,\lvert\mathord{\cdot}\rvert) such that ∑i∈{0,…,d}ai⋅pi=p\sum_{i\in\{0,\dots,d\}}a_{i}\cdot p_{i}=p. Recall that thanks to the discreteness of |⋅|\lvert\mathord{\cdot}\rvert, by Proposition 2.15, any finite dimensional ultrametric normed kk-vector space has an orthogonal basis.

5.1. Algebra norm induced by Fubini-Study metric

5.1.1. Case for (ℙd,𝒪⁡(1))(\mathbb{P}^{d},\mathscr{O}(1))

Let ϕ\phi be a metric on 𝒪⁡(1)\mathscr{O}(1), one studies the algebra norm ⦀⋅⦀ϕ\vvvert\mathord{\cdot}\vvvert_{\phi} on V∙​(𝒪​(1))V_{{\scriptscriptstyle\bullet}}(\mathscr{O}(1)). One would like to show that with various assumptions, it is a Gauss algebra norm, namely the standard affinoid algebra norm on the polynomial algebra. Then the normed section algebra (V∙(𝒪(1)),⦀⋅⦀ϕ)(V_{{\scriptscriptstyle\bullet}}(\mathscr{O}(1)),\vvvert\mathord{\cdot}\vvvert_{\phi}) will be a Tate affinoid algebra. (see Definition 2.38)

By Proposition 2.15, there exist an orthogonal basis {Ti}i∈{0,…,d}\{T_{i}\}_{i\in\{0,\dots,d\}} for the normed vector space (V1​(𝒪⁡(1)),∥⋅∥ϕ)(V_{1}(\mathscr{O}(1)),\lVert\mathord{\cdot}\rVert_{\phi}). For any i∈{0,…,d}i\in\{0,\dots,d\}, one denotes by rir_{i} the value ∥Ti∥ϕ\lVert T_{i}\rVert_{\phi}, and by 𝒓∈(ℝ+)d+1\boldsymbol{r}\in(\mathbb{R}_{+})^{d+1} the multi-radius (r0,…,rd)(r_{0},\dots,r_{d}). One fixes such an orthogonal basis, and identify the graded kk-algebra V∙​(𝒪​(1))V_{{\scriptscriptstyle\bullet}}(\mathscr{O}(1)) with k⁡[T0,…,Td]k[T_{0},\dots,T_{d}]. For any multi-index J=(j0,…,jd)∈ℕd+1J=(j_{0},\dots,j_{d})\in\mathbb{N}^{d+1}, one denotes by 𝑻J\boldsymbol{T}^{J} the monomial element ∏i∈{0,…,d}(Ti)ji∈V|J|​(𝒪⁡(1))\prod_{i\in\{0,\dots,d\}}(T_{i})^{j_{i}}\in V_{\lvert J\rvert}(\mathscr{O}(1)).

Note that in general, the sub-spaces Vn​(𝒪​(1))V_{n}(\mathscr{O}(1)) are orthogonal with respect to ⦀⋅⦀ϕ\vvvert\mathord{\cdot}\vvvert_{\phi} for different n∈ℕn\in\mathbb{N}, while a Gauss algebra norm exhibits a much finer orthogonality: the sub-spaces generated by each mononial 𝑻J\boldsymbol{T}^{J} should be orthogonal for different J∈ℕd+1J\in\mathbb{N}^{d+1}.

First, on monomial elements, the algebra norm ⦀⋅⦀ϕ\vvvert\mathord{\cdot}\vvvert_{\phi} resembles a Gauss norm.

Proposition 5.1.

For any J∈ℕd+1J\in\mathbb{N}^{d+1}, one has

∥𝑻J∥|J|​ϕ=∏i∈{0,…,d}∥Ti∥ϕji.\lVert\boldsymbol{T}^{J}\rVert_{|J|\phi}=\prod_{i\in\{0,\dots,d\}}\lVert T_{i}\rVert_{\phi}^{j_{i}}.
Proof.

Take a complete non-Archimedean valued field extension (K,|⋅|K)(K,\lvert\mathord{\cdot}\rvert_{K}) of (k,|⋅|k)(k,\lvert\mathord{\cdot}\rvert_{k}) such that

∀i∈{0,…,d},{ri}i∈{0,…,d}⊆|k×|K,\forall i\in\{0,\dots,d\},\quad\{r_{i}\}_{i\in\{0,\dots,d\}}\subseteq\lvert k^{\times}\rvert_{K},

hence for any i∈{0,…,d}i\in\{0,\dots,d\}, there exist elements κi∈K\kappa_{i}\in K such that |κi|K=ri\lvert\kappa_{i}\rvert_{K}=r_{i}. One denotes by x⁡(𝒓)∈(ℙkd)anx(\boldsymbol{r})\in(\mathbb{P}^{d}_{k})^{\mathrm{an}} the point given by coordinates [κ0:…:κd][\kappa_{0}:\dots:\kappa_{d}].

Claim 5.2.

For any i∈{0,…,d}i\in\{0,\dots,d\}, one has

∥Ti∥ϕ=ri=|Ti|ϕ​(x⁡(𝒓)).\lVert T_{i}\rVert_{\phi}=r_{i}=\lvert T_{i}\rvert_{\phi}(x(\boldsymbol{r})).

In other words, the maximum of the function |Ti|ϕ​(x)\lvert T_{i}\rvert_{\phi}(x) on (ℙkd)an(\mathbb{P}^{d}_{k})^{\mathrm{an}} is rir_{i}, and the maximum values of these d+1d+1 functions can be attained at the same point x⁡(𝒓)x(\boldsymbol{r}).

Proof.

By the orthogonality of the basis {Ti}i∈{0,…,d}\{T_{i}\}_{i\in\{0,\dots,d\}}, we can compute

|Ti|ϕ​(x⁡(𝒓))=inf(∑m∈{0,…,d}fm⋅Tm)​(x⁡(𝒓))=(Ti)​(x⁡(𝒓))(f0,…,fd)∈kd+1∥∑m∈{0,…,d}fm⋅Tm∥ϕ=inf∑m∈{0,…,d}fm​κm=κimaxm∈{0,…,d}⁡{∥fm⋅Tm∥ϕ}=inf∑m∈{0,…,d}fm​κm=κimaxm∈{0,…,d}⁡{|fm|​|κm|}=inf∑m∈{0,…,d}fm​(κm/κi)=1maxm∈{0,…,d}⁡{|fm|​|κm/κi|⋅|κi|}=|κi|=ri.\begin{split}\lvert T_{i}\rvert_{\phi}(x(\boldsymbol{r}))&=\inf_{\begin{subarray}{c}(\sum_{m\in\{0,\dots,d\}}f_{m}\cdot T_{m})(x(\boldsymbol{r}))=(T_{i})(x(\boldsymbol{r}))\\ (f_{0},\dots,f_{d})\in k^{d+1}\end{subarray}}\Big\lVert\sum_{m\in\{0,\dots,d\}}f_{m}\cdot T_{m}\Big\rVert_{\phi}\\ &=\inf\limits_{\sum_{m\in\{0,\dots,d\}}f_{m}\kappa_{m}=\kappa_{i}}\max_{m\in\{0,\dots,d\}}\Big\{\lVert f_{m}\cdot T_{m}\rVert_{\phi}\Big\}\\ &=\inf\limits_{\sum_{m\in\{0,\dots,d\}}f_{m}\kappa_{m}=\kappa_{i}}\max_{m\in\{0,\dots,d\}}\Big\{\lvert f_{m}\rvert\lvert\kappa_{m}\rvert\Big\}\\ &=\inf\limits_{\sum_{m\in\{0,\dots,d\}}f_{m}(\kappa_{m}/\kappa_{i})=1}\max_{m\in\{0,\dots,d\}}\Big\{\lvert f_{m}\rvert\lvert\kappa_{m}/\kappa_{i}\rvert\cdot\lvert\kappa_{i}\rvert\Big\}\\ &=\lvert\kappa_{i}\rvert=r_{i}.\end{split}

The last equality is obtained by Lemma 3.13. ∎

By this Claim, for any multi-index JJ, the function |𝑻J|ϕ​(x)\lvert\boldsymbol{T}^{J}\rvert_{\phi}(x) can attain its maximum value ∏i∈{0,…,d}riji\prod_{i\in\{0,\dots,d\}}r_{i}^{j_{i}} at the point x⁡(𝒓)∈(ℙkd)a​nx(\boldsymbol{r})\in(\mathbb{P}_{k}^{d})^{an} as the product of maximum of factors of the monomial. By definition,

∥𝑻J∥|J|​ϕ=supx∈(ℙd)a​n|𝑻J|ϕ​(x)=∏i∈{0,…,d}riji=∏i∈{0,…,d}∥Ti∥ϕji.\lVert\boldsymbol{T}^{J}\rVert_{|J|\phi}=\sup_{x\in(\mathbb{P}^{d})^{an}}\lvert\boldsymbol{T}^{J}\rvert_{\phi}(x)=\prod_{i\in\{0,\dots,d\}}r_{i}^{j_{i}}=\prod_{i\in\{0,\dots,d\}}\lVert T_{i}\rVert_{\phi}^{j_{i}}.

∎

Second, one calculates the algebra norm ⦀⋅⦀ϕ\vvvert\mathord{\cdot}\vvvert_{\phi} on any (homogeneous) combination of monomials. For a general metric, one needs a ℚ\mathbb{Q}-independence assumption to gain finer orthogonality.

Proposition 5.3.

Assume that {α⁡(∥Ti∥ϕ)}i∈{0,…,d}\{\alpha(\lVert T_{i}\rVert_{\phi})\}_{i\in\{0,\dots,d\}} are ℚ\mathbb{Q}-independent in ℝ/H⁡(k,|⋅|)\mathbb{R}/H(k,\lvert\mathord{\cdot}\rvert). Let S⊆ℕd+1S\subseteq\mathbb{N}^{d+1} be a finite set of multi-indices, then for any J∈SJ\in S any fJ∈kf_{J}\in k, one has

⦀∑J∈SfJ⋅𝑻J⦀=supJ∈S∥fJ⋅𝑻J∥|J|​ϕ.\Big\vvvert\sum_{\begin{subarray}{c}J\in S\end{subarray}}f_{J}\cdot\boldsymbol{T}^{J}\Big\vvvert=\sup_{\begin{subarray}{c}J\in S\end{subarray}}\ \lVert f_{J}\cdot\boldsymbol{T}^{J}\rVert_{|J|\phi}.

In other words, the algebra norm ⦀⋅⦀ϕ\vvvert\mathord{\cdot}\vvvert_{\phi} on V∙​(𝒪​(1))V_{{\scriptscriptstyle\bullet}}(\mathscr{O}(1)) is a Gauss norm on k⁡[T0,…,Td]k[T_{0},\dots,T_{d}] of multi-radius 𝒓\boldsymbol{r}. The Banach kk-algebra V^∙​(𝒪​(1),ϕ)\widehat{V}_{{\scriptscriptstyle\bullet}}(\mathscr{O}(1),\phi) is an affinoid algebra.

Proof.

By the ℚ\mathbb{Q}-independence assumption and Proposition 5.1, for any two distinct multi-index JJ and J′J^{\prime}, and any two non-zero coefficients fJf_{J} and fJ′f_{J^{\prime}} in kk, we have

∥fJ⋅𝑻J∥|J|​ϕ=|fJ|⋅∏i∈{0,…,d}riji≠|fJ′|⋅∏i∈{0,…,d}riji′=∥fJ′⋅𝑻J′∥|J′|​ϕ.\lVert f_{J}\cdot\boldsymbol{T}^{J}\rVert_{|J|\phi}=\lvert f_{J}\rvert\cdot\prod_{i\in\{0,\dots,d\}}r_{i}^{j_{i}}\neq\lvert f_{J^{\prime}}\rvert\cdot\prod_{i\in\{0,\dots,d\}}r_{i}^{j^{\prime}_{i}}=\lVert f_{J^{\prime}}\cdot\boldsymbol{T}^{J^{\prime}}\rVert_{|J^{\prime}|\phi}.

By Lemma 2.13, the elements {𝑻J}J∈S\{\boldsymbol{T}^{J}\}_{J\in S} form an orthogonal basis for the normed vector space (⨁J∈Sk⋅𝑻J,⦀⋅⦀ϕ)(\bigoplus_{J\in S}k\cdot\boldsymbol{T}^{J},\vvvert\mathord{\cdot}\vvvert_{\phi}). So the equality in the conclusion holds. ∎

Corollary 5.4.

With the same assumptions as above, the envelop metric 𝒫⁡(ϕ)\mathcal{P}(\phi) is a Fubini-Study metric induced by ∥⋅∥ϕ\lVert\mathord{\cdot}\rVert_{\phi}, and is continuous.

For a Fubini-Study metric, one does not need the ℚ\mathbb{Q}-independence. For any 𝜹=(δ0,…,δd)∈ℝd+1\boldsymbol{\delta}=(\delta_{0},\dots,\delta_{d})\in\mathbb{R}^{d+1}, one constructs a perturbed metric ϕ⁡(𝜹)\phi(\boldsymbol{\delta}) as follows. Let ∥⋅∥ϕ⁡(𝜹)\lVert\mathord{\cdot}\rVert_{\phi(\boldsymbol{\delta})} be the norm on V1​(𝒪​(1))V_{1}(\mathscr{O}(1)) such that {Ti}i∈{1,…,d}\{T_{i}\}_{i\in\{1,\dots,d\}} is an orthogonal basis with new norms

∀i∈{0,…,d},∥Ti∥ϕ​(𝜹)=eδi​∥Ti∥ϕ.\forall i\in\{0,\dots,d\},\quad\lVert T_{i}\rVert_{\phi}(\boldsymbol{\delta})=\mathrm{e}^{\delta_{i}}\lVert T_{i}\rVert_{\phi}.

Let ϕ⁡(𝜹)\phi(\boldsymbol{\delta}) be the metric FS⁡(∥⋅∥ϕ⁡(𝜹))\mathrm{FS}(\lVert\mathord{\cdot}\rVert_{\phi(\boldsymbol{\delta})}) on 𝒪⁡(1)\mathscr{O}(1). Let |𝜹|\lvert\boldsymbol{\delta}\rvert denote the number maxi∈{0,…,d}⁡|δi|∈ℝ+\max_{i\in\{0,\dots,d\}}\lvert\delta_{i}\rvert\in\mathbb{R}_{+}.

Lemma 5.5.

Assume that ϕ\phi is a Fubini-Study metric. For any ϵ>0\epsilon>0, there exists 𝜹∈ℝd+1\boldsymbol{\delta}\in\mathbb{R}^{d+1} with |𝜹|≤ϵ|\boldsymbol{\delta}|\leq\epsilon such that

∀n∈ℕ,dist⁡(∥⋅∥n​ϕ,∥⋅∥n​ϕ​(𝜹))≤n​ϵ.\forall n\in\mathbb{N},\quad\dist(\lVert\mathord{\cdot}\rVert_{n\phi},\lVert\mathord{\cdot}\rVert_{n\phi(\boldsymbol{\delta})})\leq n\epsilon.
Proof.

Choose an arbitrary 𝜹\boldsymbol{\delta} with |𝜹|≤ϵ|\boldsymbol{\delta}|\leq\epsilon. Then

dist⁡(∥⋅∥ϕ,∥⋅∥ϕ⁡(𝜹))≤ϵ.\dist(\lVert\mathord{\cdot}\rVert_{\phi},\lVert\mathord{\cdot}\rVert_{\phi(\boldsymbol{\delta})})\leq\epsilon.

By Proposition 3.12, we have

dist⁡(FS⁡(∥⋅∥ϕ),FS⁡(∥⋅∥ϕ⁡(𝜹)))=dist⁡(FS⁡(∥⋅∥ϕ),ϕ⁡(𝜹))≤ϵ,\dist(\mathrm{FS}(\lVert\mathord{\cdot}\rVert_{\phi}),\mathrm{FS}(\lVert\mathord{\cdot}\rVert_{\phi(\boldsymbol{\delta})}))=\dist(\mathrm{FS}(\lVert\mathord{\cdot}\rVert_{\phi}),\phi(\boldsymbol{\delta}))\leq\epsilon,

by the assumption and Proposition 3.11,

FS⁡(∥⋅∥ϕ)=ϕ,\mathrm{FS}(\lVert\mathord{\cdot}\rVert_{\phi})=\phi,

so the conlusion holds. ∎

Proposition 5.6.

Assume that ϕ\phi is a Fubini-Study metric. The conclusion of Proposition 5.3 holds without the assumption of ℚ\mathbb{Q}-independence of {α⁡(∥Ti∥1)}i∈{0,…,d}\{\alpha(\lVert T_{i}\rVert_{1})\}_{i\in\{0,\dots,d\}}.

Proof.

Since |⋅|k\lvert\mathord{\cdot}\rvert_{k} is discrete, for any ϵ>0\epsilon>0, there exists 𝜹\boldsymbol{\delta} with |𝜹|≤ϵ|\boldsymbol{\delta}|\leq\epsilon such that the elements {α⁡(∥Ti∥ϕ⁡(𝜹))}i∈{0,…,d}\{\alpha(\lVert T_{i}\rVert_{\phi(\boldsymbol{\delta})})\}_{i\in\{0,\dots,d\}} are ℚ\mathbb{Q}-independent in ℝ/H⁡(k,|⋅|)\mathbb{R}/H(k,\lvert\mathord{\cdot}\rvert). By Proposition 5.5, for any n∈ℕn\in\mathbb{N} and any sn=∑|J|=nfJ⋅𝑻J∈Vn​(𝒪⁡(1))s_{n}=\sum_{\lvert J\rvert=n}f_{J}\cdot\boldsymbol{T}^{J}\in V_{n}(\mathscr{O}(1)),

e−n​ϵ​∥∑|J|=nfJ⋅𝑻J∥n​ϕ​(𝜹)≤∥∑|J|=nfJ⋅𝑻J∥n​ϕ≤en​ϵ​∥∑|J|=nfJ⋅𝑻J∥n​ϕ​(𝜹).\mathrm{e}^{-n\epsilon}\Big\lVert\sum_{\lvert J\rvert=n}f_{J}\cdot\boldsymbol{T}^{J}\Big\rVert_{n\phi(\boldsymbol{\delta})}\leq\Big\lVert\sum_{\lvert J\rvert=n}f_{J}\cdot\boldsymbol{T}^{J}\Big\rVert_{n\phi}\leq\mathrm{e}^{n\epsilon}\Big\lVert\sum_{\lvert J\rvert=n}f_{J}\cdot\boldsymbol{T}^{J}\Big\rVert_{n\phi(\boldsymbol{\delta})}.

By Proposition 5.3, one has

max|J|=n⁡{e−n​ϵ​|fJ|​∏i∈{0,…,d}(eδi​ri)ji}≤∥∑|J|=nfJ⋅𝑻J∥n​ϕ≤max|J|=n⁡{en​ϵ​|fJ|​∏i∈{0,…,d}(eδi​ri)ji}.\max_{\lvert J\rvert=n}\Big\{\mathrm{e}^{-n\epsilon}\lvert f_{J}\rvert\prod_{i\in\{0,\dots,d\}}(\mathrm{e}^{\delta_{i}}r_{i})^{j_{i}}\Big\}\leq\Big\lVert\sum_{\lvert J\rvert=n}f_{J}\cdot\boldsymbol{T}^{J}\Big\rVert_{n\phi}\leq\max_{\lvert J\rvert=n}\Big\{\mathrm{e}^{n\epsilon}\lvert f_{J}\rvert\prod_{i\in\{0,\dots,d\}}(\mathrm{e}^{\delta_{i}}r_{i})^{j_{i}}\Big\}.

Fix nn and let ϵ→0\epsilon\to 0, one gets

∥∑|J|=nfJ⋅𝑻J∥n​ϕ=max|J|=n⁡{|fJ|​∏i∈{0,…,d}riji}.\Big\lVert\sum_{\lvert J\rvert=n}f_{J}\cdot\boldsymbol{T}^{J}\Big\rVert_{n\phi}=\max_{\lvert J\rvert=n}\Big\{\lvert f_{J}\rvert\prod_{i\in\{0,\dots,d\}}r_{i}^{j_{i}}\Big\}.

So one has

⦀∑|J|<∞fJ⋅𝑻J⦀n​ϕ=supn∈ℕmax|J|=n{|fJ|∏i∈{0,…,d}riji}=max|J|<∞{|fJ|∏i∈{0,…,d}riji}.\Big\vvvert\sum_{\lvert J\rvert<\infty}f_{J}\cdot\boldsymbol{T}^{J}\Big\vvvert_{n\phi}=\sup_{n\in\mathbb{N}}\max_{\lvert J\rvert=n}\Big\{\lvert f_{J}\rvert\prod_{i\in\{0,\dots,d\}}r_{i}^{j_{i}}\Big\}=\max_{\lvert J\rvert<\infty}\Big\{\lvert f_{J}\rvert\prod_{i\in\{0,\dots,d\}}r_{i}^{j_{i}}\Big\}.

Hence ⦀⋅⦀ϕ\vvvert\mathord{\cdot}\vvvert_{\phi} is a Gauss norm of multi-radius 𝒓\boldsymbol{r} on V∙​(𝒪​(1))V_{{\scriptscriptstyle\bullet}}(\mathscr{O}(1)). ∎

5.1.2. Case for general (X,L)(X,L)

Proposition 5.7.

Assume that ϕ\phi is a Fubini-Study metric. If LL is very ample, then V^∙​(L,ϕ)\widehat{V}_{{\scriptscriptstyle\bullet}}(L,\phi), V^​(LX|Y,ϕX|Y)\widehat{V}(L_{X|Y},\phi_{X|Y}) and V^∙​(L|Y,ϕ|Y)\widehat{V}_{{\scriptscriptstyle\bullet}}(L|_{Y},\phi|_{Y}) are affinoid algebras.

Proof.

By the assumption, the elements of V1​(L)V_{1}(L) induces an embedding

ι1:X→ℙkd1\iota_{1}:X\rightarrow\mathbb{P}^{d_{1}}_{k}

such that ι1∗​𝒪​(1)=L\iota_{1}^{*}\mathscr{O}(1)=L with dimk​V1=d1+1\mathrm{dim}_{k}V_{1}=d_{1}+1. Moreover there exists a norm ∥⋅∥1\lVert\mathord{\cdot}\rVert_{1} on V1​(L)V_{1}(L) such that ϕ=FS⁡(∥⋅∥1)\phi=\mathrm{FS}(\lVert\mathord{\cdot}\rVert_{1}). View ∥⋅∥1\lVert\mathord{\cdot}\rVert_{1} as a norm on V1​(𝒪​(1))V_{1}(\mathscr{O}(1)), we get a metric ψ=FS⁡(∥⋅∥1)\psi=\mathrm{FS}(\lVert\mathord{\cdot}\rVert_{1}) on 𝒪⁡(1)\mathscr{O}(1). By construction ψ|X=ϕ\psi|_{X}=\phi.

By Proposition 5.6, the Banach algebra V^∙(𝒪(1),⦀⋅⦀ψ)\widehat{V}_{{\scriptscriptstyle\bullet}}(\mathscr{O}(1),\vvvert\mathord{\cdot}\vvvert_{\psi}) is an affinoid algebra. Hence the quotient Banach algebra V^∙(L,⦀⋅⦀ψ,ℙd1|X)\widehat{V}_{{\scriptscriptstyle\bullet}}(L,\vvvert\mathord{\cdot}\vvvert_{\psi,\mathbb{P}^{d_{1}}|X}) is an affinoid algebra.

By Proposition 3.27, the algebra norm ⦀⋅⦀ϕ\vvvert\mathord{\cdot}\vvvert_{\phi} is the spectral norm of ⦀⋅⦀ψ,ℙd1|X\vvvert\mathord{\cdot}\vvvert_{\psi,\mathbb{P}^{d_{1}}|X}. The later is an affinoid norm, hence is equivalent to its spectral norm by Proposition 2.58. So ⦀⋅⦀ϕ\vvvert\mathord{\cdot}\vvvert_{\phi} is also an affinoid algebra norm. Thus V^∙​(L,ϕ)\widehat{V}_{{\scriptscriptstyle\bullet}}(L,\phi) is an affinoid algebra.

Similarly, by Proposition 3.27, on V∙​(LX|Y)V_{{\scriptscriptstyle\bullet}}(L_{X|Y}), the algebra norm ⦀⋅⦀ϕ|Y\vvvert\mathord{\cdot}\vvvert_{\phi|_{Y}} is the spectral norm of ⦀⋅⦀ϕ,X|Y\vvvert\mathord{\cdot}\vvvert_{\phi,X|Y}, hence is itself an affinoid algebra norm. ∎

Corollary 5.8.

Assume that ϕ\phi is a Fubini-Study metric. If LL is just ample, then V^∙​(L,ϕ)\widehat{V}_{{\scriptscriptstyle\bullet}}(L,\phi), V^​(LX|Y,ϕX|Y)\widehat{V}(L_{X|Y},\phi_{X|Y}) and V^∙​(L|Y,ϕ|Y)\widehat{V}_{{\scriptscriptstyle\bullet}}(L|_{Y},\phi|_{Y}) are affinoid algebras.

Proof.

By assumption, L⊗ML^{\otimes M} is very ample. So V^∙(M)​(L,ϕ)\widehat{V}_{{\scriptscriptstyle\bullet}}^{(M)}(L,\phi), V^∙(M)​(LX|Y,ϕX|Y)\widehat{V}_{{\scriptscriptstyle\bullet}}^{(M)}(L_{X|Y},\phi_{X|Y}) and V^∙(M)​(L|Y,ϕ|Y)\widehat{V}_{{\scriptscriptstyle\bullet}}^{(M)}(L|_{Y},\phi|_{Y}) are affinoid algebras. Since V∙​(L)V_{{\scriptscriptstyle\bullet}}(L) is integral and is finite over V∙(M)​(L)V_{{\scriptscriptstyle\bullet}}^{(M)}(L), by Proposition 2.45, the Banach algebras V^∙​(L,ϕ)\widehat{V}_{{\scriptscriptstyle\bullet}}(L,\phi), V^​(LX|Y,ϕX|Y)\widehat{V}(L_{X|Y},\phi_{X|Y}) and V^∙​(L|Y,ϕ|Y)\widehat{V}_{{\scriptscriptstyle\bullet}}(L|_{Y},\phi|_{Y}) are Banach finite over V^∙(M)​(L,ϕ)\widehat{V}_{{\scriptscriptstyle\bullet}}^{(M)}(L,\phi), V^∙(M)​(LX|Y,ϕX|Y)\widehat{V}_{{\scriptscriptstyle\bullet}}^{(M)}(L_{X|Y},\phi_{X|Y}) and V^∙(M)​(L|Y,ϕ|Y)\widehat{V}_{{\scriptscriptstyle\bullet}}^{(M)}(L|_{Y},\phi|_{Y}) respectively. Hence they are affinoid algebras. ∎

Proposition 5.9.

Assume that ϕ\phi is a Fubini-Study metric. Then there exist C⁡(ϕ,Y)>0C(\phi,Y)>0 such that for any t¯∈V∙​(LX|Y)\underline{t}\in V_{{\scriptscriptstyle\bullet}}(L_{X|Y}), there exists s¯∈V∙​(L)\underline{s}\in V_{{\scriptscriptstyle\bullet}}(L) with

⦀s¯⦀ϕ≤C(ϕ,Y,X)⋅⦀t¯⦀ϕ|Y.\vvvert\underline{s}\vvvert_{\phi}\leq C(\phi,Y,X)\cdot\vvvert\underline{t}\vvvert_{\phi|_{Y}}.

In particular, for any n∈ℕn\in\mathbb{N} and tn∈V∙​(LX|Y)t_{n}\in V_{{\scriptscriptstyle\bullet}}(L_{X|Y}), there exists sn∈V∙​(L)s_{n}\in V_{{\scriptscriptstyle\bullet}}(L) with

∥sn∥n​ϕ≤C⁡(ϕ,Y,X)⋅∥tn∥n​ϕ|Y.\lVert s_{n}\rVert_{n\phi}\leq C(\phi,Y,X)\cdot\lVert t_{n}\rVert_{n\phi|_{Y}}.
Proof.

By Corollary 5.8, the Banach algebra norm ⦀⋅⦀ϕX|Y\vvvert\mathord{\cdot}\vvvert_{\phi_{X|Y}} is an affinoid algebra norm. By Proposition 2.58, there exists C⁡(ϕ,Y)>0C(\phi,Y)>0 such that

⦀⋅⦀ϕX|Y≤C(ϕ,Y,X)⋅⦀⋅⦀ϕX|Y;sp.\vvvert\mathord{\cdot}\vvvert_{\phi_{X|Y}}\leq C(\phi,Y,X)\cdot\vvvert\mathord{\cdot}\vvvert_{\phi_{X|Y};\mathrm{sp}}.

Since ⦀⋅⦀ϕX|Y;sp=⦀⋅⦀ϕ|Y\vvvert\mathord{\cdot}\vvvert_{\phi_{X|Y};\mathrm{sp}}=\vvvert\mathord{\cdot}\vvvert_{\phi|_{Y}} by Corollary 3.27, one gets the bounds. ∎

Remark 5.10.

With the metric finiteness properties of affinoid algebra norm, here the upper bound for metric extension of a Fubini-Study metric is much better than what was expected, compared to (3) or even to (2), for its (in)depence on n∈ℕn\in\mathbb{N}. This independence suggest that it would be reasonable to compare this affinoid algebra technique in this non-Archimedean setting with the use of Ohsawa-Takegoshi L2L^{2} extension technique in the complex analytic setting.

5.2. Algebra norm induced by asymptotic Fubini-Study metric

With the extra assumption of discreteness for the base valued field, one can give another proof of Theorem 4.5.

Theorem 5.11.

Suppose that (k,|⋅|)(k,\lvert\mathord{\cdot}\rvert) is discretely valued. Let ϕ\phi be an asymptotic Fubini-Study metric on LL. Then for any ϵ>0\epsilon>0, there exists nY∈ℕn_{Y}\in\mathbb{N} such that for any n≥nYn\geq n_{Y} and any tn∈Vn​(L|Y)t_{n}\in V_{n}(L|_{Y}), there exits sn∈Vn​(L)s_{n}\in V_{n}(L) such that sn|Y=tns_{n}|_{Y}=t_{n} and

∥sn∥n​ϕ≤en​ϵ⋅∥tn∥n​ϕ|Y.\lVert s_{n}\rVert_{n\phi}\leq\mathrm{e}^{n\epsilon}\cdot\lVert t_{n}\rVert_{n\phi|_{Y}}.
Proof.

Recall that one can find M∈ℕM\in\mathbb{N} such that L⊗ML^{\otimes M} is very ample and for any n≥Mn\geq M, the restriction map from Vn​(L)V_{n}(L) to Vn​(L|Y)V_{n}(L|_{Y}) is surjective.

By the asymptotic Fubini-Study assumption, there exist norms {∥⋅∥n}n∈ℕ\{\lVert\mathord{\cdot}\rVert_{n}\}_{n\in\mathbb{N}} on Vn​(L)V_{n}(L) such that ϕ\phi is given by 𝒫⁡({∥⋅∥n}n∈ℕ)\mathcal{P}(\{\lVert\mathord{\cdot}\rVert_{n}\}_{n\in\mathbb{N}}). Let ψn\psi_{n} denote the metric 1n​FS​(∥⋅∥)n\frac{1}{n}\mathrm{FS}(\lVert\mathord{\cdot}\rVert)_{n} on LL. By the continuity assumption, the convergence to envelop metric is uniform (see §3.1 12.). So for any ϵ>0\epsilon>0, there exists M′≥MM^{\prime}\geq M such that

dist⁡(ψM′,ϕ)≤13​ϵ,\dist(\psi_{M^{\prime}},\phi)\leq\frac{1}{3}\epsilon,

hence for any n∈ℕn\in\mathbb{N}, one has

dist⁡(∥⋅∥n​ψM′,∥⋅∥n​ϕ)≤13​n​ϵ,dist⁡(∥⋅∥n​ψM′,X|Y,∥⋅∥n​ϕ,X|Y)≤13​n​ϵ.\dist(\lVert\mathord{\cdot}\rVert_{n\psi_{M^{\prime}}},\lVert\mathord{\cdot}\rVert_{n\phi})\leq\frac{1}{3}n\epsilon,\quad\dist(\lVert\mathord{\cdot}\rVert_{n\psi_{M^{\prime}},X|Y},\lVert\mathord{\cdot}\rVert_{n\phi,X|Y})\leq\frac{1}{3}n\epsilon.

By Corollary 5.8, the Banach algebra V^∙​(LX|Y,(ψM′)X|Y)\widehat{V}_{{\scriptscriptstyle\bullet}}(L_{X|Y},(\psi_{M^{\prime}})_{X|Y}) is an affinoid algebra. By Proposition 5.9, there exists CM′​(X,Y,ϕ)>0C_{M^{\prime}}(X,Y,\phi)>0 such that

∀t¯∈V∙(LX|Y),⦀t¯⦀ψM′,X|Y≤CM′⋅⦀t¯⦀ψM′,X|Y;sp=⦀t¯⦀ψM′,Y.\forall\underline{t}\in V_{{\scriptscriptstyle\bullet}}(L_{X|Y}),\quad\vvvert\underline{t}\vvvert_{\psi_{M^{\prime}},X|Y}\leq C_{M^{\prime}}\cdot\vvvert\underline{t}\vvvert_{\psi_{M^{\prime}},X|Y;\mathrm{sp}}=\vvvert\underline{t}\vvvert_{\psi_{M^{\prime}},Y}.

Combining this comparison with above estimates, one has that for every n≥nY:=⌈ln⁡(CM′)/(ϵ/3)⌉n\geq n_{Y}:=\lceil\ln(C_{M^{\prime}})/(\epsilon/3)\rceil

∥tn∥n​ϕ,X|Y≤e13​n​ϵ⋅∥tn∥n​ψM′,X|Y≤e13​n​ϵ⋅CM′⋅∥tn∥n​ψM′|Y≤e23​n​ϵ⋅CM′⋅∥tn∥n​ϕ|Y≤en​ϵ⋅∥tn∥n​ϕ|Y.\begin{split}\lVert t_{n}\rVert_{n\phi,X|Y}&\leq\mathrm{e}^{\frac{1}{3}n\epsilon}\cdot\lVert t_{n}\rVert_{n\psi_{M^{\prime}},X|Y}\\ &\leq\mathrm{e}^{\frac{1}{3}n\epsilon}\cdot C_{M^{\prime}}\cdot\lVert t_{n}\rVert_{n\psi_{M^{\prime}}|_{Y}}\\ &\leq\mathrm{e}^{\frac{2}{3}n\epsilon}\cdot C_{M^{\prime}}\cdot\lVert t_{n}\rVert_{n\phi|_{Y}}\\ &\leq\mathrm{e}^{n\epsilon}\cdot\lVert t_{n}\rVert_{n\phi|_{Y}}.\end{split}

Hence there exists sn∈Vn​(L)s_{n}\in V_{n}(L) with ∥sn∥n​ϕ≤en​ϵ⋅∥tn∥n​ϕ|Y\lVert s_{n}\rVert_{n\phi}\leq\mathrm{e}^{n\epsilon}\cdot\lVert t_{n}\rVert_{n\phi|_{Y}}. ∎

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