3 · Reduction, formal models and residual geometry

134 NAG Tags

An analytic point can encode a variety over a residue field. Reduction makes this information visible and connects analytic functions with models. We distinguish two constructions from the start: canonical reduction uses the intrinsic spectral seminorm, whereas a formal model chooses an integral algebra. Their special fibres need not coincide. Graded reduction will then retain the sizes that ordinary reduction forgets.

3.1 Power-bounded functions and canonical reduction

Definition 3.1.1. For a Banach affinoid algebra AA, an element is power-bounded if the sequence of its powers is bounded, and topologically nilpotent if that sequence tends to zero. Denote these subsets by A∘A^\circ and A∘∘A^{\circ\circ} respectively.

Proposition 3.1.2. NAG-000207 For an affinoid algebra one has A∘={a:ρA(a)≤1},A∘∘={a:ρA(a)<1}.A^\circ=\{a:\rho_A(a)\leq1\},\qquad A^{\circ\circ}=\{a:\rho_A(a)<1\}. The first is a subring, the second an ideal in it, and A~=A∘/A∘∘\widetilde A=A^\circ/A^{\circ\circ} is reduced.

Proof. If ρA(a)<1\rho_A(a)<1, Corollary 1.20.3 implies ∥am∥→0\|a^m\|\to 0. Conversely, if the powers tend to zero, some power has norm less than one and hence ρA(a)<1\rho_A(a)<1. A power-bounded element has spectral radius at most one. For the reverse inclusion use Theorem 1.20.2, which gives ∥am∥≤CaρA(a)m\|a^m\|\leq C_a\rho_A(a)^m when ρA(a)>0\rho_A(a)>0. If ρA(a)=0\rho_A(a)=0, Proposition 3.21.4 makes aa nilpotent. The ring and ideal assertions follow from the spectral seminorm inequalities. If a~m=0\widetilde a^m=0, then ρA(a)m<1\rho_A(a)^m<1, hence ρA(a)<1\rho_A(a)<1 and a~=0\widetilde a=0. ◻

3.2 Reduction and the residual directions of the disc

Let A∘A^\circ be the ring of power-bounded elements and A∘∘A^{\circ\circ} the ideal of topologically nilpotent elements. If a∈A∘a\in A^\circ, the boundedness of the sequence ama^m and Lemma 1.6.1 imply ∣a(x)∣≤1|a(x)|\leq 1. If a∈A∘∘a\in A^{\circ\circ}, then am→0a^m\to 0, which implies ∣a(x)∣<1|a(x)|<1. Thus evaluation induces a homomorphism A~=A∘/A∘∘⟶H(x)~.\widetilde A=A^\circ/A^{\circ\circ} \longrightarrow\widetilde{\mathcal H(x)}. Its prime kernel defines red⁡(x)∈Spec⁡A~\operatorname{red}(x)\in\operatorname{Spec}\widetilde A. When a formal model with ring of definition A0⊂A∘A_0\subset A^\circ is chosen, the same construction on A0A_0 gives specialization to its special fibre. The model reduction and the canonical reduction need not have identical coordinate rings.

Proposition 3.2.1. NAG-000635 For f~1,…,f~m∈A~\widetilde f_1,\ldots,\widetilde f_m\in\widetilde A represented by fi∈A∘f_i\in A^\circ, one has red⁡−1V(f~1,…,f~m)={x:∣fi(x)∣<1 for every i}.\operatorname{red}^{-1}V(\widetilde f_1,\ldots,\widetilde f_m) =\{x:|f_i(x)|<1\text{ for every }i\}. This set is ordinary open. In particular, when A~\widetilde A is noetherian, reduction is anticontinuous: inverse images of Zariski closed subsets are open.

Proof. Membership in V(f~1,…,f~m)V(\widetilde f_1,\ldots,\widetilde f_m) says exactly that the reductions of all fi(x)f_i(x) vanish in the residual field. This is equivalent to the displayed strict inequalities. Each inequality defines an open set, and only finitely many occur. Noetherianity permits every closed subset to be expressed by finitely many equations. ◻

Example 3.2.2. For A=K⟨T⟩A=K\langle T\rangle with its Gauss norm, multiplicativity gives A∘={f:∥f∥≤1}A^\circ=\{f:\|f\|\leq 1\} and A∘∘={f:∥f∥<1}A^{\circ\circ}=\{f:\|f\|<1\}. A convergent series has only finitely many coefficients of size one, so reduction gives A~=K~[T]\widetilde A=\widetilde K[T]. At the Gauss point ζG\zeta_G, every nonzero reduced polynomial has value one. Its reduction ideal is therefore zero, and red⁡(ζG)\operatorname{red}(\zeta_G) is the generic point of the residual affine line. Moreover, the residue field of the completed Gauss field is K~(T)\widetilde K(T): rational functions with nonzero reduction supply this field, and completion does not change residues because sufficiently close elements have the same residue.

If a∈K∘a\in K^\circ lifts a~∈K~\widetilde a\in\widetilde K, the tube of the residual point T=a~T=\widetilde a is ∣T−a∣<1|T-a|<1. Thus different residual points specify different open directions adjacent to the Gauss point. Their union is not a replacement for the Gauss point itself. This is the simplest example of residual algebraic geometry carried by a single analytic point.

The same mechanism appears in a model metric. A formal line bundle has norm-one local frames, while its restriction to the special fibre may be nontrivial. Its degrees on residual curves enter the later comparison between PL positivity and residual nefness.

References. For the surrounding affinoid, reduction and graded-reduction theory, see [Gub13] (Sections 3–6) and [CLD25] (Chapters 10–12).

3.3 Shilov boundary and maximum values

Definition 3.3.1. A closed subset S⊂M(A)S\subset\mathcal M(A) is a boundary for the algebra if every ∣f∣|f|, f∈Af\in A, attains its maximum on SS. The Shilov boundary is the smallest such closed subset. Its existence is a theorem.

Theorem 3.3.2. NAG-000225 An affinoid space has a finite Shilov boundary. If AA is strictly affinoid over a nontrivially valued field, A~\widetilde A is a finite-type K~\widetilde K-algebra, the canonical reduction map is surjective, and each generic point of Spec⁡A~\operatorname{Spec}\widetilde A has a unique analytic preimage. Those preimages are exactly the Shilov boundary.

Proof. Corollary 3.12.5 proves finite Shilov boundaries for all radii. In the strict case, Theorem 3.16.3 proves finite generation of canonical reduction, and Theorem 3.16.16 realizes every prime ideal by an analytic point. Theorem 3.16.7 identifies the minimal primes with the Shilov points and proves uniqueness of each preimage. ◻

References. [TemIntro] (3.4.1.1–3.4.1.7).

3.4 Norms and analytic points

Let V=M(A)V=\mathcal M(A) be a nonempty affinoid space over a complete non-Archimedean field. Its spectral seminorm is the supremum of the evaluations at analytic points:

∥f∥sp=sup⁡x∈V∣f(x)∣.\|f\|_{\mathrm{sp}}=\sup_{x\in V}|f(x)|. The spectral seminorm is power-multiplicative. Multiplicativity on two unrelated elements is a stronger assertion, and is precisely what allows this seminorm to represent one analytic point.

Proposition 3.4.1. NAG-000657 The Shilov boundary of VV consists of one point if and only if its spectral seminorm is multiplicative. When these conditions hold, the unique boundary point is the spectral seminorm.

Proof. If the Shilov boundary is {ξ}\{\xi\}, its maximum property gives the following identity for every element of AA:

∥f∥sp=∣f(ξ)∣.\|f\|_{\mathrm{sp}}=|f(\xi)|. The right side is multiplicative. Conversely, a multiplicative spectral seminorm is a bounded multiplicative seminorm extending the absolute value of the ground field, hence a point ξ\xi of VV. It attains the supremum of every function. Thus {ξ}\{\xi\} is a boundary. Corollary 3.12.5 gives the smallest closed boundary, which must therefore be {ξ}\{\xi\}. ◻

Example 3.4.2. (Disc, circle and annulus). The closed disc has one Shilov point, its Gauss point. A strict closed annulus ∣a∣≤∣T∣≤1|a|\leq|T|\leq 1, 0<∣a∣<10<|a|<1, has the model A=Spf⁡K∘⟨U,V⟩/(UV−a),U=T,V=a/T.\mathfrak A=\operatorname{Spf}K^\circ\langle U,V\rangle/(UV-a), \quad U=T,\quad V=a/T. Its special fibre has equation uv=0uv=0. In this example it agrees with canonical reduction. The two irreducible components give the endpoint Gauss points of the annulus. For the radius-one circle, the algebra is K⟨T,T−1⟩K\langle T,T^{-1}\rangle and the reduction is K~[t,t−1]\widetilde K[t,t^{-1}], which is integral; the two endpoints have coalesced into one Gauss point.

These assertions can also be checked directly from Laurent expansions: the spectral norm on the annulus is the maximum of the two endpoint Gauss norms, and neither endpoint can be omitted, since TT and a/Ta/T attain their maxima at opposite ends.

Proposition 3.4.3. NAG-000154 (A higher-dimensional boundary distinction). Let KK be nontrivially valued and X=M(K⟨T,S⟩)X=\mathcal M(K\langle T,S\rangle). Its Shilov boundary consists of the Gauss point ζ1,1\zeta_{1,1}, but its analytic boundary contains every weighted Gauss point ζ1,ρ\zeta_{1,\rho} with ρ∈∣K×∣\rho\in|K^\times|, 0<ρ<10<\rho<1.

Proof. The unit Gauss norm is multiplicative, so Proposition 3.4.1 gives the Shilov assertion. Canonical reduction is K~[t,s]\widetilde K[t,s] by Proposition 3.9.2. At ζ1,ρ\zeta_{1,\rho} its kernel in this ring is (s)(s): positive powers of SS become small, while nonzero polynomials in TT retain norm one. The prime (s)(s) is not maximal. Theorem 3.25.1 identifies the absolute interior with the inverse image of the closed points of canonical reduction, so this point lies in the analytic boundary. ◻

For a pure one-dimensional strict affinoid, Theorem 3.25.4 proves that the analytic and Shilov boundaries coincide: the nonclosed reduction points are precisely its generic points. The bidisc illustrates the distinction in higher dimension.

3.5 Formal models and their two fibres

In this section the valuation is nontrivial. Fix π∈K∘∘∖{0}\pi\in K^{\circ\circ}\setminus\{0\}; powers of π\pi define the topology of K∘K^\circ. We use admissible formal schemes locally of finite presentation over K∘K^\circ. An affine chart has the form X=Spf⁡A0,A0=K∘⟨T1,…,Tm⟩/I,\mathfrak X=\operatorname{Spf}A_0, \qquad A_0=K^\circ\langle T_1,\ldots,T_m\rangle/I, with the completed topology, finitely generated defining ideal, and no π\pi-torsion. The last condition excludes vertical torsion; over a valuation ring it expresses the required flatness. Its generic and special fibres are Xη=M(A0⊗K∘K),Xs=Spec⁡(A0⊗K∘K~).\mathfrak X_\eta=\mathcal M(A_0\otimes_{K^\circ}K), \qquad \mathfrak X_s=\operatorname{Spec}(A_0\otimes_{K^\circ}\widetilde K). The equivalence with flatness follows directly from the valuation-ring structure. For every nonzero b∈K∘b\in K^\circ, some πN\pi^N belongs to bK∘bK^\circ, so absence of π\pi-torsion implies absence of all K∘K^\circ-torsion. Every finitely generated ideal of K∘K^\circ is principal, and its tensor map to a torsion-free module is injective; the finite-ideal criterion [StacksAlg] (Tag 00HD) gives flatness, while flatness preserves injectivity of multiplication by each nonzero scalar and therefore implies torsion-freeness.

The generic-fibre algebra is the corresponding affinoid algebra, with its complete quotient topology. Formal open charts glue along domains on the generic fibre, giving the generic-fibre functor. A formal model of an analytic space includes an identification of this generic fibre with the specified space.

References. [TemIntro] (§5.2); [GRW16] (§§2–3).

Every a∈A0a\in A_0 has absolute value at most one on the generic fibre. Evaluation at xx therefore gives A0→H(x)∘A_0\to\mathcal H(x)^\circ, and reduction gives its specialization red⁡X(x)∈Xs\operatorname{red}_{\mathfrak X}(x)\in\mathfrak X_s. If a locally closed stratum ZZ is specified in the special fibre, its tube is the subset ] ⁣Z[X=red⁡X−1(Z)]\!Z[_{\mathfrak X}=\operatorname{red}_{\mathfrak X}^{-1}(Z) with its natural analytic-domain structure. On each formal affine chart, finite equations and inequations express it by strict inequalities and equalities of absolute values.

Example 3.5.1. For Spf⁡K∘⟨T⟩\operatorname{Spf}K^\circ\langle T\rangle, the generic fibre is the closed unit disc and the special fibre is the residual affine line. The tube of a~∈K~\widetilde a\in\widetilde K is ∣T−a∣<1|T-a|<1. For UV=aUV=a as in Example 3.4.2, the tube of the node u=v=0u=v=0 is ∣U∣<1,∣V∣<1⟺∣a∣<∣T∣<1.|U|<1,\quad |V|<1 \quad\Longleftrightarrow\quad |a|<|T|<1. It is an open annulus, not a single analytic point. The generic points of the two components have endpoint Gauss points above them. Thus different sorts of strata have very different analytic fibres.

Remark 3.5.2. The special fibre of a chosen model can be nonreduced; canonical reduction is reduced by Proposition 3.1.2. Nor is the ring A∘A^\circ automatically an admissible algebra of finite presentation in every category under consideration. We therefore distinguish the canonical reduction Spec⁡A~\operatorname{Spec}\widetilde A from an arbitrary model Spf⁡A0\operatorname{Spf}A_0. Whenever a canonical formal model is used in a later source theorem, its existence in the stated category is part of that theorem’s hypotheses or construction.

Example 3.5.3. (A reduced algebra with two different norms). Let KK be discretely valued, with uniformizer π\pi, and put A=K⟨T⟩/(T2−π)A=K\langle T\rangle/(T^2-\pi). Normalize its additive valuation by v(π)=1v(\pi)=1. A root b∈Kb\in K would satisfy 2v(b)=12v(b)=1, impossible since v(K×)=Zv(K^\times)=\mathbb Z. Thus the quadratic polynomial is irreducible, and Weierstrass division identifies AA with a quadratic field LL. The quotient of the unit Gauss norm gives ∥T∥quot=1\|T\|_{\mathrm{quot}}=1, whereas its spectral norm is ρA(T)=∣π∣1/2<1\rho_A(T)=|\pi|^{1/2}<1.

To verify the quotient norm, apply Theorem 1.3.1 to the polynomial T2−πT^2-\pi. Every lift of the class of TT has remainder TT, and the remainder operator is contractive. Its norm is therefore at least one; the lift TT attains one. On the other hand T2=πT^2=\pi forces every multiplicative seminorm to take the value ∣π∣1/2|\pi|^{1/2} on TT. This proves the spectral assertion.

The chosen model K∘[T]/(T2−π)K^\circ[T]/(T^2-\pi) has special fibre K~[t]/(t2)\widetilde K[t]/(t^2), while canonical reduction of the field is its residue field, which is reduced. Hence reducedness of the generic algebra does not identify a quotient Gauss norm with its spectral norm, or identify a chosen special fibre with canonical reduction. Nilpotent multiplicity in a model can coexist with a reduced generic fibre.

3.6 Changing a model without changing its generic fibre

An admissible formal blow-up is obtained by blowing up an open coherent ideal and taking the corresponding formal completion, with the admissible torsion-free charts. Locally on the formal scheme, an open ideal contains a power of π\pi. After passing to the generic fibre, that local power is a unit, so the centre disappears algebraically. The resulting map on analytic generic fibres is an isomorphism. This invariance is the basic reason that different models can describe the same analytic space.

Example 3.6.1. Blow up the ideal (T,a)(T,a) on Spf⁡K∘⟨T⟩\operatorname{Spf}K^\circ\langle T\rangle, where 0<∣a∣<10<|a|<1. The chart in which aa generates the ideal has coordinate U=T/aU=T/a and generic domain ∣T∣≤∣a∣|T|\leq|a|. The chart in which TT generates it has coordinate V=a/TV=a/T and generic domain ∣a∣≤∣T∣≤1|a|\leq|T|\leq 1. These two domains cover the original disc and meet in ∣T∣=∣a∣|T|=|a|. Their gluing is the identity on the original generic fibre, while the special fibre has acquired a new component.

Lemma 3.17.3 proves generic-fibre invariance for admissible formal blowups. For two quasi-paracompact admissible formal models with the same generic fibre, Lemma 3.17.7 constructs a common formal modification. Such a common model permits the comparison of integral data defining model metrics.

References. [TemIntro] (§§5.2–5.3).

3.7 Abhyankar dimension and Gauss fields

Definition 3.7.1. For an analytic point xx, put sK(x)=trdeg⁡K~H(x)~,tK(x)=dim⁡Q((v(H(x)×)/Γ)⊗ZQ),dK(x)=sK(x)+tK(x).\begin{aligned} s_K(x)&=\operatorname{trdeg}_{\widetilde K}\widetilde{\mathcal H(x)},\\ t_K(x)&=\dim_\mathbb Q\bigl((v(\mathcal H(x)^\times)/\Gamma)\otimes_\mathbb Z\mathbb Q\bigr),\\ d_K(x)&=s_K(x)+t_K(x). \end{aligned} The analytic Abhyankar inequality says dK(x)≤dim⁡xXd_K(x)\leq\dim_xX. A point attaining equality is called Abhyankar in this local-dimensional sense. On a pure nn-dimensional space, the full-dimensional Abhyankar points have dK(x)=nd_K(x)=n.

Theorem 3.22.9 proves the inequality for arbitrary analytic spaces, using the affinoid dimension calculation and its compatibility with analytic domains.

References. For the general dimension theorem, see [CLD25] (11.2.3, especially (11.2.3.1)) and [Duc18] (§1.4).

Neither sK(x)s_K(x) nor tK(x)t_K(x) is the dimension of the singleton {x}\{x\} as a topological space. They measure independent residual parameters and independent values. The following calculation proves the equality for the Gauss fields used throughout this book. The subsequent semistable chart is first treated as an explicit affinoid equation; its global skeletal interpretation is developed in Chapter 4.

3.8 Value-group rank and residual dimension at a Gauss point

A skeleton has a real dimension, but each of its points also has a completed residue field. These dimensions are related by the Abhyankar equality, not by identifying the residue field with the residue field of one formal stratum. A direct Gauss-norm calculation makes the difference visible.

Let KK be complete with additive value group Γ⊂R\Gamma\subset\mathbb R. On K(Z1,…,Zn)K(Z_1,\ldots,Z_n) use the weighted Gauss valuation with v(Zi)=wiv(Z_i)=w_i. Write HwH_w for its completion and set Γw=Γ+∑iZwi,t=dim⁡Q((Γw/Γ)⊗ZQ),Lw={m∈Zn:⟨m,w⟩∈Γ}.\begin{gathered}\Gamma_w=\Gamma+\sum_i\mathbb Zw_i,\qquad t=\dim_\mathbb Q\bigl((\Gamma_w/\Gamma)\otimes_\mathbb Z\mathbb Q\bigr), \\ L_w=\{m\in\mathbb Z^n:\langle m,w\rangle\in\Gamma\}.\end{gathered}

Proposition 3.8.1. NAG-000464 The value group of HwH_w is Γw\Gamma_w. The group LwL_w has rank n−tn-t, and the residue field of HwH_w is a purely transcendental extension of K~\widetilde K of degree n−tn-t. In particular, trdeg⁡K~H~w+dim⁡Q((v(Hw×)/Γ)⊗Q)=n.\operatorname{trdeg}_{\widetilde K}\widetilde H_w +\dim_\mathbb Q\bigl((v(H_w^\times)/\Gamma)\otimes\mathbb Q\bigr)=n.

Proof. The value of a nonzero Laurent polynomial is the minimum of its finitely many term values, so it belongs to Γw\Gamma_w. Quotients have the same property, and the constants and variables generate the whole group Γw\Gamma_w. Completion adds no values: for a nonzero limit aa, a sufficiently close approximation aja_j has ∣a−aj∣<∣a∣|a-a_j|<|a|, hence ∣aj∣=∣a∣|a_j|=|a|. It adds no residue classes either, because an approximation with difference of norm less than one has the same residue.

The homomorphism Zn→Γw/Γ\mathbb Z^n\to\Gamma_w/\Gamma, m↦⟨m,w⟩m\mapsto\langle m,w\rangle, is surjective with kernel LwL_w. Tensoring with Q\mathbb Q gives rank⁡Lw=n−t\operatorname{rank}L_w=n-t. Choose a basis m1,…,mn−tm_1,\ldots,m_{n-t} of this free abelian group and constants cj∈K×c_j\in K^\times with v(cj)=−⟨mj,w⟩v(c_j)=-\langle m_j,w\rangle. The monomials Uj=cjZmjU_j=c_jZ^{m_j} have norm one.

Their residues are algebraically independent. A nonzero polynomial relation in their residues could be lifted to a Laurent polynomial in distinct exponent vectors ∑jajmj\sum_j a_jm_j, with all nonzero leading coefficients of norm one. The Gauss norm of this polynomial is one, so its residue cannot be zero. Independence of the mjm_j ensures that the exponent vectors are distinct.

They generate the residue field as well. Let F/GF/G be a rational function of norm one. Divide numerator and denominator by one term of GG attaining its Gauss norm. Both resulting Laurent polynomials have norm one. Each surviving term in their reductions has exponent difference m∈Lwm\in L_w and is a residue-field scalar times a Laurent monomial in the UjU_j. Terms of smaller norm disappear. The denominator reduction is nonzero by algebraic independence, so the residue of F/GF/G belongs to K~(U~1,…,U~n−t)\widetilde K(\widetilde U_1,\ldots,\widetilde U_{n-t}). The completion observation proves the same assertion for HwH_w. ◻

The semistable simplex and its residue field

Consider the standard formal chart X=Spf⁡K∘⟨X0,…,Xr,T1,…,Td−r⟩/(X0⋯Xr−a),0<∣a∣<1.\mathfrak X=\operatorname{Spf} K^\circ\langle X_0,\ldots,X_r,T_1,\ldots,T_{d-r}\rangle /(X_0\cdots X_r-a),\qquad 0<|a|<1. For ui≥0u_i\geq 0 with ∑i=0rui=v(a)\sum_{i=0}^ru_i=v(a), eliminate X0=a/(X1⋯Xr)X_0=a/(X_1\cdots X_r) and take the Gauss valuation with weights u1,…,uru_1,\ldots,u_r on the Laurent variables and zero on the TjT_j. All original coordinates have norm at most one, so the valuation defines an analytic point of the generic fibre. Multiplicativity follows from the Gauss-norm argument; boundedness lets it extend from polynomials to the affinoid algebra.

These points form the canonical simplex. Their completed residue fields are the completions of the displayed Gauss rational fields: polynomials are dense in the affinoid algebra for a norm dominating the point seminorm, and rational quotients are then dense in the residue-field completion. Proposition 3.8.1 gives s(x)=d−t(x),s(x)+t(x)=d.s(x)=d-t(x),\qquad s(x)+t(x)=d. Thus every point of the standard simplex is Abhyankar of the full analytic dimension dd, although the simplex itself has real dimension only rr. The remaining dimensions are carried by residue fields.

At a point in the relative interior of this real simplex, all XiX_i reduce to zero in this formal chart. Its reduction lies at the generic point of the stratum with free coordinates T1,…,Td−rT_1,\ldots,T_{d-r}, of dimension d−rd-r. The completed residue field of the analytic point may have larger residual transcendence degree d−t(x)d-t(x), because it also contains residues of norm-one Laurent quotients in the XiX_i. Those quotients need not be regular functions on the original formal chart. They can become visible after a suitable formal modification.

Proposition 3.26.1 supplies the integral normal form of this chart, its rational-domain presentation and the topology of the displayed simplex. It also proves the local analytic dimension used above and realizes any finite family of the indicated Laurent residues after an admissible formal modification.

Example 3.8.2. (Two points on the same annular skeleton). Let the chart be XY=aXY=a, so d=r=1d=r=1, and take an interior skeletal point with v(X)=uv(X)=u. If uu is rational over Γ\Gamma, choose m>0m>0 and c∈K×c\in K^\times with v(c)=muv(c)=mu. Then Xm/cX^m/c has norm one and transcendental residue, while the new value-group quotient has rational rank zero. Hence (s,t)=(1,0)(s,t)=(1,0). If uu is rationally independent modulo Γ\Gamma, no such monomial exists and (s,t)=(0,1)(s,t)=(0,1). Both points reduce to the same node in the original semistable model.

The calculation also clarifies why “generic point of a formal stratum”, “Shilov boundary point of a tropical fibre”, and “Abhyankar point” are related but different descriptions. The first depends on a model, the second on a tropicalization, and the third on an equality of valued-field dimensions. In regular-immersion arguments, it is the last equality that gives the local analytic dimension needed for the finite-flat comparison.

References. The standard semistable charts and skeletal valuations are those of [GRW16] (Sections 3–4). The Gauss-field calculation is proved here; general analytic dimension and invariance statements remain the geometric theorems cited in Chapter 2.

3.9 Graded reduction

Ordinary reduction records elements of size at most one modulo those of size less than one. To compare radii not supplied by KK, it is useful to retain the same construction at every positive size.

Definition 3.9.1. For a valued field FF, its graded reduction is F~gr=⨁r>0F≤r/F<r.\widetilde F^{\mathrm{gr}} =\bigoplus_{r>0}F_{\leq r}/F_{<r}. Multiplication sends degree rr times degree ss to degree rsrs. For an affinoid algebra, use the filtration defined by its spectral seminorm. A graded field means a graded ring in which every nonzero homogeneous element is invertible; its underlying ungraded ring need not be a field.

Proposition 3.9.2. NAG-000189 The graded reduction of the weighted Gauss polynomial algebra is the graded polynomial algebra over K~gr\widetilde K^{\mathrm{gr}} with deg⁡(T~i)=ri\deg(\widetilde T_i)=r_i. The same assertion holds for the generalized Tate algebra with its Gauss norm.

Proof. A homogeneous leading class of size ss is represented by the finitely many monomials aνTνa_\nu T^\nu having ∣aν∣rν=s|a_\nu|r^\nu=s. The remaining terms have strictly smaller norm. For a convergent series there are still only finitely many terms at the maximal size, and its tail is strictly smaller. This proves surjectivity. A nonzero homogeneous polynomial cannot map to zero, because its Gauss norm is exactly its specified size; this proves injectivity. Multiplication is compatible by Proposition 1.2.2. ◻

For the one-radius field KrK_r, where r∉∣K×∣r\notin\sqrt{|K^\times|}, a nonzero Laurent sum has a single leading monomial. Consequently its graded reduction is K~gr[t,t−1]\widetilde K^{\mathrm{gr}}[t,t^{-1}] with deg⁡t=r\deg t=r. Its degree-one part is only K~\widetilde K, although the full graded ring records a new value. This is the algebraic counterpart of (s,t)=(0,1)(s,t)=(0,1) in the annulus calculation.

The general theory uses homogeneous prime ideals, graded fraction fields and graded valuation rings. Many elementary arguments retain their form after all generators and localizations are required to be homogeneous. Their application to analytic germs, however, is a geometric theorem. Temkin’s reduction of a germ organizes compatible residual data over modifications. In CLD25 these constructions support residual line bundles and the degree computations entering calibrations. We use the specified germ-reduction theorems where they are needed, rather than identifying a germ with the reduction of one chosen affinoid model.

References. [Tem04] (§§1–3); [TemIntro] (§5.7); [CLD25] (Chapter 10).

3.10 Reduction of a germ

Ordinary reduction belongs to an affinoid algebra, or to a chosen formal model. A local construction at an analytic point must survive changing both. The appropriate residual object is a birational space. We describe it first in the strict, nontrivially valued setting and then indicate the graded extension.

Definition 3.10.1. For a field extension F/kF/k, let RZ⁡k(F)\operatorname{RZ}_k(F) be the set of valuation rings VV with k⊂V⊂Fk\subset V\subset F and fraction field FF. Its basic open subsets have the form U(f1,…,fm)={V:f1,…,fm∈V},fi∈F.U(f_1,\ldots,f_m)=\{V:f_1,\ldots,f_m\in V\},\qquad f_i\in F. If ZZ is an integral kk-scheme with function field contained in FF, a centre of VV on ZZ is a point zz such that OZ,z⊂V\mathcal O_{Z,z}\subset V and the maximal ideal of VV contracts to that of OZ,z\mathcal O_{Z,z}.

Proposition 3.10.2. NAG-000216 The space RZ⁡k(F)\operatorname{RZ}_k(F) is quasi-compact and quasi-separated. Each displayed basic open is quasi-compact.

Proof. We use product compactness in [StacksAlg] (Tag 08ZU). Encode a subset V⊂FV\subset F by its characteristic function in {0,1}F\{0,1\}^F, with the product of discrete topologies. The conditions that VV contain kk, be closed under addition, multiplication and negatives, and satisfy a∈Va\in V or a−1∈Va^{-1}\in V for every a≠0a\ne 0, are closed conditions on finitely many coordinates at a time. They characterize valuation subrings with fraction field FF. Thus these subrings form a closed subset of the compact product.

Each U(f1,…,fm)U(f_1,\ldots,f_m) is clopen in this finer topology and therefore compact there. The Riemann–Zariski topology is coarser, so the same set is quasi-compact in it. These sets form a basis closed under finite intersections; hence the space is also quasi-separated. This proof does not assume that an arbitrary inverse limit of quasi-compact topological spaces is quasi-compact. ◻

References. Compare [TemIntro] (5.7.2.1–5.7.2.4). The argument is the usual finite-condition compactness proof, included to make the topological input explicit.

A proper modification of ZZ does not change the valuations that have a centre: the valuative criterion lifts a centre to the modification, and separatedness makes the lift unique when the generic map is fixed. This explains why compatible centres on models, rather than one preferred scheme, are natural residual data.

Let xx lie in a compact strictly analytic neighborhood UU, with an admissible formal model U\mathfrak U. Evaluation gives a pointed residual scheme Spec⁡H(x)~⟶Us.\operatorname{Spec}\widetilde{\mathcal H(x)}\longrightarrow\mathfrak U_s. Its image has a closure ZZ in Us\mathfrak U_s. Replacing the model by a common admissible modification replaces this pointed scheme by a proper comparison inducing the identity on H(x)~\widetilde{\mathcal H(x)}. Formally one localizes the category of pointed residual schemes at such proper maps. The corresponding object, or its realization as a birational space, is the reduction of the germ (X,x)(X,x). Its independence of UU and U\mathfrak U is Temkin’s theorem.

The category in this ordinary-reduction construction is the strictly analytic category over a nontrivially valued field, as fixed at the start of the section. Its subdomain morphisms are strict subdomains. The graded construction in Theorems 3.19.9 and 3.19.10 treats arbitrary analytic domains and arbitrary complete ground fields.

NAG-000186 · Open at full scope

Theorem 3.10.3. NAG-000186 (Reduction of strict analytic germs). For strictly analytic spaces, germ reduction is functorial and independent of the model and neighborhood. Germ subdomains correspond to birational subspaces, compatibly with inclusions, intersections and finite unions. A germ is good precisely when its reduction is affine in the birational category. A morphism is separated at a point precisely when the corresponding reduction morphism is separated; the point lies in relative interior precisely when that reduction morphism is proper.

Proof. The ordinary reduction, its functoriality, and its independence of strict charts and neighbourhoods are constructed in Theorem 3.19.22. Theorem 3.19.25 identifies it with the pointed residual scheme construction above. Its comparisons under common formal modification are proper by Proposition 3.19.24; Lemma 3.19.23 identifies the corresponding birational spaces.

The strict subdomain classification is part of Theorem 3.19.22. It preserves inclusions; intersections and finite unions are then identified by their respective greatest-lower-bound and least-upper-bound properties. The same theorem proves the three geometric criteria: affineness follows from Theorem 3.19.19 and degree-one valuation restriction; separatedness follows from Proposition 3.19.16; and relative interior follows from Theorem 2.46.1. These are internal results, and their strict/graded comparison is the pullback identity proved in Theorem 3.19.22. ◻

References. [TemIntro] (5.7.1.3, 5.7.2.10 and 5.7.4.1); [Tem04].

The word affine here refers to a birational space admitting an affine scheme model in the stated category. It does not assert that the entire projective residual curve at a type II point is affine. This distinction is visible in the next example.

In particular, a pointed proper curve Spec⁡F→C\operatorname{Spec}F\to C has the proper comparison C→Spec⁡kC\to\operatorname{Spec}k. The latter point, still pointed by FF, is an affine scheme model of the same birational space. Scheme models in this category need not have function field equal to FF.

Example 3.10.4. Let xx be the unit Gauss point. For the closed unit disc, a residual model of its germ is Spec⁡K~[t]\operatorname{Spec}\widetilde K[t], pointed by the inclusion into H(x)~=K~(t)\widetilde{\mathcal H(x)}=\widetilde K(t). For P1,an\mathbb P^{1,\mathrm{an}} at the same point, use the standard proper model; the residual model is PK~1\mathbb P^1_{\widetilde K} with that function field. The map from the disc germ to the projective-line germ is represented by AK~1↪PK~1\mathbb A^1_{\widetilde K}\hookrightarrow\mathbb P^1_{\widetilde K}. It is not proper, reflecting that the Gauss point lies on the relative boundary of the closed disc. The omitted point at infinity is the missing outward direction.

For arbitrary radii, replace the fields and reductions in this construction by graded ones. On an affinoid chart the starting map is A~gr→H(x)~gr\widetilde A^{\mathrm{gr}}\to\widetilde{\mathcal H(x)}^{\mathrm{gr}}. The reductions on good subdomains glue. The germ-subdomain, goodness and relative-interior statements have graded versions; see [TemIntro] (5.7.5.1). This extension is necessary for the radius-rr field of Section 1.8, whose ordinary residual field alone loses one analytic dimension.

Where a residual line bundle lives

For a formal line bundle L\mathfrak L on U\mathfrak U, restrict Ls\mathfrak L_s to the closure ZZ above. After a common formal refinement, pull it back to the corresponding residual model. These restrictions are compatible under further pullback. They define line-bundle data on models of the germ reduction, rather than an arbitrary line bundle on the residue field itself. Intersection degrees are computed on a model where the relevant bundles and subvarieties are represented, and independence is then a projection-formula question.

This describes the formal-model origin of residual data. The extension to the PL metrics of CLD, and the comparison between their residual degrees and analytic curvature, require the constructions and theorems of [CLD25] (Chapters 10, 17–18). Chapter 20 retains these as separate inputs. In particular, identifying the residual object does not already prove a positivity theorem.

3.11 Formal line bundles and model metrics

Let X\mathfrak X be an admissible formal model of XX and let L\mathfrak L be an invertible sheaf on it. Its generic fibre is an analytic line bundle LL. On a formal open chart where L\mathfrak L has a frame ee, prescribe ∥e(x)∥=1\|e(x)\|=1 on the associated generic domain. For a local section s=fes=fe, set ∥s(x)∥=∣f(x)∣\|s(x)\|=|f(x)|.

NAG-000176 · Open at full scope

Proposition 3.11.1. NAG-000176 These prescriptions define a continuous metric on LL, independent of the formal frames. They commute with pullback of formal line bundles and remain unchanged when the model and line bundle are pulled back along an admissible formal blow-up.

Proof. Two formal frames differ by a formal unit uu. Both uu and u−1u^{-1} have absolute value at most one on the generic fibre, so ∣u(x)∣=1|u(x)|=1. Thus the prescribed norms agree on overlaps. Proposition 3.17.2 constructs the generic fibre from the affinoid domains of formal affine charts. These domains form its analytic net. On every such chart the norm of fefe is the continuous function ∣f∣|f|, so the chartwise continuity criterion in Lemma 2.40.2 proves continuity of the glued metric. A pulled-back frame again has prescribed norm one, proving pullback compatibility. The generic-fibre isomorphism in Lemma 3.17.3 then proves invariance under admissible formal blowup. ◻

Model metrics on supplied compatible atlases. Let KK be complete and nontrivially valued. Let X\mathfrak X be an admissible formal model with its identified analytic generic fibre XX and supplied compatible formal-affine generic atlas. Let L\mathfrak L be an invertible sheaf with generic fibre LL. Prescribing norm one for formal frames defines a continuous metric on LL. For a formal morphism with a supplied compatible generic map this metric commutes with pullback. In particular it is unchanged under an admissible blowup with the compatible source and target atlases of generic invariance on supplied atlases.

Proof. On each formal affine the evaluation of an integral formal function lies in the valuation ring of the completed residue field, as in the affine specialization construction of Theorem 3.16.15. On each supplied formal affine, refine the trivializing formal opens by a finite cover of principal formal opens. The formal-topology clause of NAG-001944 gives the Noetherian topology on that affine; quasi-compactness and the principal basis give the finite refinement. Proposition  3.17.1 identifies the generic trace of each principal restriction with its actual closed rational domain on the given affine chart. The open traces on supplied atlases identify these domains with the full inverse images of the formal opens under specialization. They therefore give a finite covering of the actual generic affine chart. The transition maps on these domains are the supplied bounded maps. If another formal frame differs by a unit uu, the evaluations of both uu and u−1u^{-1} have norm at most one there. Thus ∣u∣=1|u|=1 on the overlap and the frame norms agree.

These trivializing domains refine the given analytic net by Lemma 2.39.2. On each one the norm of a section fefe in the formal frame ee is ∣f∣|f|. This is continuous, and the quasi-net topology test of Lemma 2.40.2 gives continuity of the glued metric. Equivalently, in any analytic trivialization its local norm function is tested on the intersections with this net.

The compatible generic pullback of a formal frame is the generic frame of the pulled-back formal sheaf. Its prescribed norm is again one, which proves the pullback equality on the trivializing domains and hence on the whole generic fibre. For the indicated blowup the supplied-atlas invariance identifies its actual generic map with an isomorphism. Under this identification the pullback equality gives the claimed invariance. ◻

For a model of L⊗mL^{\otimes m}, taking the positive mmth root gives the associated rational model metric on LL. If LL is the trivial bundle with chosen section 11, its weight φ=−log⁡∥1∥\varphi=-\log\|1\| is a model function. Changing the chosen trivialization of LL changes this weight by the logarithm of an analytic unit; the metric itself remains an object on the line bundle.

Example 3.11.2. On the blow-up of (T,a)(T,a) from Example 3.6.1, the pulled-back ideal becomes an invertible sheaf. Its generic fibre is canonically trivial because aa is a unit over KK. On the chart with generator aa, the trivial section 1=a−1a1=a^{-1}a has norm ∣a∣−1|a|^{-1}; on the chart with generator TT, it has norm ∣T∣−1|T|^{-1}. Its weight is therefore log⁡max⁡(∣T∣,∣a∣)\log\max(|T|,|a|). With t=−log⁡∣T∣t=-\log|T| and c=−log⁡∣a∣c=-\log|a|, this is −min⁡(t,c)-\min(t,c). The dual line bundle gives min⁡(t,c)\min(t,c).

This example connects an integral modification with a corner of a PL function. To turn the corner into curvature, one still needs a differential calculus and its normalization. To turn a model line bundle into a positive metric, one needs numerical conditions on residual curves. Those are the additional assertions proved and compared in Chapters 15, 19 and 20. The construction here supplies their geometric input without assuming those conclusions.

3.12 Finite boundaries and the generic points of graded reduction

Finite normalization provides a direct route from the affinoid algebra theorems to the existence of a Shilov boundary. A finite algebra over a Tate algebra has finitely many points above its Gauss point. Minimal polynomials show that these points already detect every maximum. We first prove this assertion, then identify the minimal such boundary through graded reduction. All ground fields retain their given absolute values; trivial valuations and arbitrary positive affinoid radii are allowed.

Lemma 3.12.1. NAG-001901 (Roots of a monic polynomial). Let HH be a complete non-Archimedean field and let P(Z)=Zd+c1Zd−1+⋯+cdP(Z)=Z^d+c_1Z^{d-1}+\cdots+c_d be monic of positive degree. In a finite splitting field, endowed with the extending absolute value, max⁡P(α)=0∣α∣=max⁡1≤i≤d∣ci∣1/i.\max_{P(\alpha)=0}|\alpha| =\max_{1\leq i\leq d}|c_i|^{1/i}.

Proof. A finite splitting field exists by [StacksAlg] (Tag 09HU); Lemma 1.12.1 gives its extending absolute value. Write RR for the maximum on the right. If ∣α∣>R|\alpha|>R, the leading term αd\alpha^d has strictly larger absolute value than every other term of P(α)P(\alpha), so P(α)≠0P(\alpha)\ne 0. Thus every root has size at most RR. Conversely, if SS is the maximum size of the roots, the elementary symmetric expressions for the coefficients give ∣ci∣≤Si|c_i|\leq S^i. This proves R≤SR\leq S. The argument includes repeated roots and the polynomial ZdZ^d. ◻

Proposition 3.12.2. NAG-001902 (A boundary above a normalization Gauss point). Let BB be a nonzero strict KK-affinoid domain, and let R=K⟨T1,…,Td⟩↪BR=K\langle T_1,\ldots,T_d\rangle\hookrightarrow B be a bounded finite injection. Set F=Frac⁡(R)F=\operatorname{Frac}(R) and let HH be the completion of FF for the extending Gauss absolute value. The finite-dimensional HH-algebra D=B⊗RHD=B\otimes_R H is nonzero. Its finitely many maximal ideals, with the absolute values on their residue fields, give a finite set S⊂M(B)S\subset\mathcal M(B), and max⁡x∈M(B)∣b(x)∣=max⁡s∈S∣b(s)∣(b∈B).\max_{x\in\mathcal M(B)}|b(x)|=\max_{s\in S}|b(s)| \qquad(b\in B).

Proof. The multiplicative Gauss norm extends to the fraction field by ∣u/v∣=∥u∥/∥v∥|u/v|=\|u\|/\|v\|. Its metric completion is again a valued field: if a nonzero Cauchy class is represented by unu_n, then ∣un∣|u_n| is eventually constant and positive, and ∣un−1−um−1∣=∣un−um∣/(∣un∣∣um∣)|u_n^{-1}-u_m^{-1}|=|u_n-u_m|/(|u_n||u_m|) proves convergence of the inverses. Put E=B⊗RFE=B\otimes_R F. This is a finite-dimensional FF-domain, hence a field. Consequently D=E⊗FHD=E\otimes_F H has positive finite dimension over HH.

A finite-dimensional algebra has only finitely many maximal ideals. Indeed, any rr distinct maximal ideals give, by the Chinese remainder theorem, a quotient which is a product of rr nonzero fields, forcing r≤dim⁡HDr\leq\dim_H D. Each residue field LL is finite over HH and has the absolute value of Lemma 1.12.1. The resulting maps B→LB\to L are bounded. To check this last assertion, choose finite RR-module generators b1,…,bmb_1,\ldots,b_m of BB. Proposition 1.14.4 identifies the given norm of BB with its canonical finite-module norm. Thus every bb has an expression b=∑aibib=\sum a_i b_i with max⁡i∥ai∥≤C∥b∥\max_i\|a_i\|\leq C\|b\|, where CC is independent of bb. Evaluation in LL then gives ∣b∣≤C∥b∥max⁡i∣bi∣|b|\leq C\|b\|\max_i|b_i|. These characters therefore define the asserted points.

Fix b∈Bb\in B. By Theorem 1.13.1 and the minimal-polynomial argument in Lemma 1.13.3, the monic minimal polynomial of bb over FF belongs to R[Z]R[Z]; write it as Ze+a1Ze−1+⋯+aeZ^e+a_1Z^{e-1}+\cdots+a_e. Put r=max⁡i∥ai∥1/ir=\max_i\|a_i\|^{1/i}. Every analytic point xx of BB restricts to a bounded point of RR, so ∣ai(x)∣≤∥ai∥|a_i(x)|\leq\|a_i\|. The dominant-term argument from the preceding lemma gives ∣b(x)∣≤r|b(x)|\leq r.

At the Gauss point the coefficients have sizes exactly ∥ai∥\|a_i\|, so the same lemma supplies a root α\alpha of size rr in a finite splitting field over HH. The algebra F[b]⊗FH=H[Z]/(Ze+a1Ze−1+⋯+ae)F[b]\otimes_F H=H[Z]/(Z^e+a_1Z^{e-1}+\cdots+a_e) has the maximal ideal defined by this root. The extension F[b]⊗FH⟶E⊗FH=DF[b]\otimes_F H\longrightarrow E\otimes_F H=D is finite free of positive rank, since EE is a finite-dimensional vector space over the field F[b]F[b]. It is in particular integral and faithfully flat. Lying over, [StacksAlg] (Tag 00GQ), gives a maximal ideal of DD above the chosen one. Its residue field contains H(α)H(\alpha); uniqueness of the extending absolute value implies that the corresponding point s∈Ss\in S satisfies ∣b(s)∣=r|b(s)|=r. This proves the maximum identity. ◻

Theorem 3.12.3. NAG-001903 (Existence of a finite boundary). Every nonempty KK-affinoid spectrum has a finite boundary.

Proof. First suppose that AA is strict. Its finitely many minimal primes pj\mathfrak p_j exist by the Noetherian theorem (Theorem 1.11.5) and [StacksAlg] (Tag 00FR). Each quotient Bj=A/pjB_j=A/\mathfrak p_j is a strict affinoid domain. Theorem 1.12.2 and Proposition 3.12.2 give a finite boundary of each M(Bj)\mathcal M(B_j). Every point of M(A)\mathcal M(A) kills some pj\mathfrak p_j: its prime kernel contains a minimal prime by [StacksAlg] (Tag 00E0). The induced seminorm on BjB_j is bounded, as is seen by taking the infimum over the representatives in the quotient norm. The union of the finite boundaries therefore detects every maximum on M(A)\mathcal M(A).

For arbitrary radii, choose a free extension L=KrL=K_r for which ArA_r is strict, as in Theorem 1.17.6. The strict case gives a finite boundary SrS_r upstairs. The spectral projection is surjective by Theorem 1.19.1. For every a∈Aa\in A, its values upstairs are exactly the values of aa at the projected points. It follows that the finite image of SrS_r is a boundary of M(A)\mathcal M(A). ◻

Lemma 3.12.4. NAG-001904 (A minimal finite boundary is the smallest closed boundary). Suppose that a compact space XX of multiplicative seminorms on an algebra has continuous evaluation functions and has a finite boundary. Every inclusion-minimal finite boundary is contained in every closed boundary. In particular it is unique.

Proof. Choose an inclusion-minimal finite boundary SS and a point ξ∈S\xi\in S. Minimality supplies aa with m=∣a(ξ)∣>max⁡s∈S∖{ξ}∣a(s)∣=q,m>0,m=|a(\xi)|>\max_{s\in S\setminus\{\xi\}}|a(s)|=q, \qquad m>0, where the maximum over the empty set is zero. Let CC be any closed boundary and put C0={y∈C:∣a(y)∣=m}C_0=\{y\in C:|a(y)|=m\}, a nonempty compact set. For bb in the algebra and y∈C0y\in C_0, the boundary property of SS gives ∣b(y)∣≤max⁡s∈S(∣a(s)∣/m)n∣b(s)∣→n→∞∣b(ξ)∣.|b(y)|\leq \max_{s\in S}\bigl(|a(s)|/m\bigr)^n|b(s)| \xrightarrow[n\to\infty]{}|b(\xi)|. If ∣b(ξ)∣>0|b(\xi)|>0, choose for each sufficiently large nn a point yn∈Cy_n\in C attaining the maximum of ∣anb∣|a^n b|. The maximum equals mn∣b(ξ)∣m^n|b(\xi)| for large nn. Since ∣a∣≤m|a|\leq m and ∣b∣|b| is bounded, we obtain ∣a(yn)∣→m|a(y_n)|\to m and ∣b(yn)∣≥∣b(ξ)∣|b(y_n)|\geq|b(\xi)|. A convergent subnet has limit in C0C_0 and attains ∣b(ξ)∣|b(\xi)| there. For finitely many b1,…,btb_1,\ldots,b_t with nonzero values at ξ\xi, apply this conclusion to their product. The preceding upper bound for each factor forces all their values to equal the respective values at ξ\xi at one common point of C0C_0. Factors vanishing at ξ\xi vanish throughout C0C_0 by that same upper bound.

Thus the closed sets {y∈C0:∣b(y)∣=∣b(ξ)∣}\{y\in C_0:|b(y)|=|b(\xi)|\}, indexed by all bb, have the finite intersection property. Compactness produces a point with all evaluations equal to those of ξ\xi. Points are the seminorms themselves, so this point is ξ\xi. Hence ξ∈C\xi\in C. This holds for every ξ∈S\xi\in S and every closed boundary CC, proving the assertion. ◻

Corollary 3.12.5. NAG-001905 (The affinoid Shilov boundary). Every nonempty affinoid spectrum has a unique finite Shilov boundary. The Shilov boundary of the empty spectrum is empty.

Proof. Apply Theorem 3.12.3 and Lemma 3.12.4, using the compact Hausdorff spectrum of Proposition 2.24.1. ◻

Proposition 3.12.6. NAG-001906 (Power-bounded elements and spectral reduction). For every affinoid algebra AA, power-bounded elements are exactly those with spectral radius at most one; topologically nilpotent elements are exactly those with spectral radius less than one. The ring A∘/A∘∘A^\circ/A^{\circ\circ} is reduced.

Proof. The first equivalence is Theorem 1.20.2 with s=1s=1, together with Corollary 1.20.3. If ρ(a)<1\rho(a)<1, choose ss strictly between ρ(a)\rho(a) and one; the same power bound gives ∥an∥≤Csn→0\|a^n\|\leq Cs^n\to 0. Conversely, if an→0a^n\to 0, some power has norm less than one, whence ρ(a)<1\rho(a)<1 by the spectral formula. The ultrametric and submultiplicative inequalities for ρ\rho show that the first set is a ring and the second an ideal in it. Finally, a~n=0\widetilde a^n=0 implies ρ(a)n<1\rho(a)^n<1, hence ρ(a)<1\rho(a)<1 and a~=0\widetilde a=0. ◻

Theorem 3.12.7. NAG-001907 (Generic points of graded reduction). Let AA be a nonzero affinoid algebra, let ΓA\Gamma_A be its Shilov boundary, and let A~gr\widetilde A^{\mathrm{gr}} be its graded spectral reduction. Evaluation at ξ∈ΓA\xi\in\Gamma_A induces a homogeneous prime Pξ=ker⁡(A~gr⟶H(ξ)~gr).P_\xi=\ker\bigl(\widetilde A^{\mathrm{gr}} \longrightarrow\widetilde{\mathcal H(\xi)}^{\mathrm{gr}}\bigr). The ideals PξP_\xi are precisely the minimal homogeneous prime ideals of A~gr\widetilde A^{\mathrm{gr}}. They are distinct, their intersection is zero, and each has exactly one analytic preimage under reduction, namely ξ\xi.

Proof. Evaluation preserves each filtered piece and its strictly smaller subspace, so it defines the displayed map. A graded valued-field reduction has no homogeneous zero divisors. Since the positive real grading group is totally ordered, the largest nonzero components of two finite graded sums have nonzero product; thus the graded ring is also an ordinary domain, and the kernel is a homogeneous prime. If a homogeneous class of degree rr belongs to all PξP_\xi, its lift aa satisfies ∣a(ξ)∣<r|a(\xi)|<r for all ξ∈ΓA\xi\in\Gamma_A. The boundary is finite, so ρ(a)=max⁡ΓA∣a∣<r\rho(a)=\max_{\Gamma_A}|a|<r and the class vanishes. Thus the intersection is zero.

For each ξ\xi, minimality of the finite boundary (Corollary 3.12.5) supplies aξa_\xi whose maximum on ΓA\Gamma_A occurs only at ξ\xi. Its initial class lies in every PηP_\eta for η≠ξ\eta\ne\xi, but not in PξP_\xi. Thus these primes are pairwise incomparable. Any homogeneous prime containing zero contains one of them: otherwise choose a homogeneous element of each PξP_\xi outside that prime and multiply the finitely many choices, contradicting their zero intersection. Pairwise incomparability then shows that these are exactly the minimal homogeneous primes.

It remains to identify the fibre. Suppose that an analytic point yy has reduction kernel PξP_\xi, and choose a=aξa=a_\xi as above with m=∣a(ξ)∣>q=max⁡η≠ξ∣a(η)∣m=|a(\xi)|>q=\max_{\eta\ne\xi}|a(\eta)|. Since the initial class of aa is not in PξP_\xi, we have ∣a(y)∣=m|a(y)|=m. If ∣b(ξ)∣>0|b(\xi)|>0, then, for sufficiently large nn, ρ(anb)=mn∣b(ξ)∣\rho(a^n b)=m^n|b(\xi)|. The initial class of anba^n b does not lie in PξP_\xi, so its value at yy equals this radius. Dividing by mnm^n gives ∣b(y)∣=∣b(ξ)∣|b(y)|=|b(\xi)|. If b(ξ)=0b(\xi)=0, the estimate mn∣b(y)∣≤qnmax⁡η≠ξ∣b(η)∣m^n|b(y)|\leq q^n\max_{\eta\ne\xi}|b(\eta)| for all nn forces b(y)=0b(y)=0. Thus all evaluations agree and y=ξy=\xi. ◻

References. The generic-point description is Temkin’s [Tem04] (Proposition 3.3(ii)–(iii)). The proof here starts from the finite-normalization and minimal-polynomial results already established in the book; its finite-boundary argument supplies the existence input before applying graded reduction.

3.13 Graded reduction and coefficient extensions

Proposition 3.13.1. NAG-001909 (The spectral norm of a coefficient expansion). Let AA be an affinoid algebra and rr a finite tuple of positive radii. For a series in A{r−1T}A\{r^{-1}T\}, ρ(∑ν∈NmaνTν)=max⁡νρA(aν)rν.\rho\left(\sum_{\nu\in\mathbf N^m}a_\nu T^\nu\right) =\max_\nu\rho_A(a_\nu)r^\nu. If rr is free over KK, the same formula, with indices in Zm\mathbf Z^m, holds in Ar=A⊗^KKrA_r=A\widehat\otimes_K K_r.

Proof. At every spectral point of either coefficient algebra, restriction to AA and the coordinate bounds give the upper bound on the right. The series tails tend uniformly to zero in the coefficient norm, so this estimate passes from polynomials to series. Conversely, for every x∈M(A)x\in\mathcal M(A) the Gauss seminorm on the polynomial algebra over H(x)\mathcal H(x) extends by that same tail estimate and takes the value max⁡ν∣aν(x)∣rν\max_\nu|a_\nu(x)|r^\nu. In the Laurent case this is precisely the section of Theorem 1.19.1. Taking the supremum over xx and interchanging two suprema proves the formula, by Corollary 1.20.3. The coefficient weights tend to zero, so every positive supremum is a maximum; when all weights vanish, both sides are zero. ◻

Theorem 3.13.2. NAG-001910 (Graded polynomial and free scalar extension). There are canonical graded algebra isomorphisms A~gr[t1,…,tm]≃A{r−1T}~gr,deg⁡ti=ri.\widetilde A^{\mathrm{gr}}[t_1,\ldots,t_m] \simeq\widetilde{A\{r^{-1}T\}}^{\mathrm{gr}}, \qquad \deg t_i=r_i. When rr is free over KK, there are also canonical isomorphisms A~gr⊗K~grKr~gr≃A~gr[t1±1,…,tm±1]≃Ar~gr.\widetilde A^{\mathrm{gr}}\otimes_{\widetilde K^{\mathrm{gr}}} \widetilde {K_r}^{\mathrm{gr}} \simeq \widetilde A^{\mathrm{gr}}[t_1^{\pm1},\ldots,t_m^{\pm1}] \simeq\widetilde{A_r}^{\mathrm{gr}}.

Proof. In a homogeneous class of degree ss, only finitely many terms of a convergent series have spectral size ss; all remaining terms together have spectral size strictly less than ss. This follows because their coefficient norms tend to zero. Taking the initial classes of the maximizing coefficients therefore proves surjectivity of the displayed polynomial or Laurent map. Conversely, a nonzero homogeneous polynomial of degree ss has a coefficient with nonzero class at its prescribed degree. The formula in Proposition 3.13.1 makes the spectral radius of any corresponding lift exactly ss. Its image cannot vanish. This proves injectivity, and multiplication respects the initial classes by the filtered-algebra construction.

For KrK_r itself, the same coefficient calculation gives Kr~gr=K~gr[t1±1,…,tm±1].\widetilde {K_r}^{\mathrm{gr}} =\widetilde K^{\mathrm{gr}}[t_1^{\pm1},\ldots,t_m^{\pm1}].

The freeness condition is the one used to construct the complete field KrK_r. Tensoring this polynomial description with A~gr\widetilde A^{\mathrm{gr}} gives the first isomorphism in the second line. ◻

References. These coefficient comparisons are the free-radius part of [Tem04] (Proposition 3.1(i)). The spectral coefficient formula and the actual free-extension presentations make the comparisons explicit.

3.14 Exact unit-ball generators and admissible formal algebras

For the formal constructions we need more than a bounded choice of generators: generators of norm one must give coefficient bounds with constant one. The following coefficient-ring argument supplies this strengthening of Theorem 1.11.4.

Lemma 3.14.1. NAG-001912 (A coefficient ring with a uniform gap). Let KK be a complete non-Archimedean field. For every sequence (ai)(a_i) in K∘K^\circ tending to zero, there is a closed subring S⊂K∘S\subset K^\circ containing all aia_i, such that every element of SS of absolute value one is a unit in SS and there is a real θ<1\theta<1 with ∣s∣≤θ|s|\leq\theta for every s∈Ss\in S satisfying ∣s∣<1|s|<1.

Proof. Call a subring with the latter gap property a gap ring. The prime subring is a gap ring: in positive characteristic its induced absolute value is trivial, while in characteristic zero its elements of size less than one form a principal ideal of Z\mathbf Z. Localizing a gap ring at its elements of size one preserves the gap and makes its residue ring a field.

Suppose S0S_0 is such a localized gap ring and ∣a∣=1|a|=1. If the residue of aa is transcendental over S~0\widetilde S_0, a polynomial f∈S0[Z]f\in S_0[Z] can have ∣f(a)∣<1|f(a)|<1 only when all its coefficients have size less than one, so the same gap works. If the residue is algebraic, lift its monic minimal polynomial to g∈S0[Z]g\in S_0[Z]. Choose θ<1\theta<1 bounding both ∣g(a)∣|g(a)| and the old gap. For f∈S0[Z]f\in S_0[Z] divide by the monic gg, obtaining f=qg+rf=qg+r with deg⁡r<deg⁡g\deg r<\deg g. If ∣f(a)∣<1|f(a)|<1, then ∣r(a)∣<1|r(a)|<1; the minimality of the residual polynomial forces every coefficient of rr to have size less than one. Hence ∣f(a)∣≤θ|f(a)|\leq\theta. Thus adjoining one element of size one preserves the gap.

Only finitely many of the aia_i have size one; adjoin these first. The remaining sequence is bounded by some ε<1\varepsilon<1. In any polynomial in those remaining elements, every nonconstant monomial has size at most ε\varepsilon. Its constant term lies in the already constructed gap ring. Consequently the enlarged ring has gap bounded by the maximum of the old gap and ε\varepsilon. Localize once more at the size-one elements and take its closure in KK. The gap survives closure. A size-one element of the closure differs by size less than one from a size-one element of the uncompleted ring; its inverse is obtained by a convergent geometric series. The resulting SS has all the stated properties. ◻

Lemma 3.14.2. NAG-001913 (Lifting a residual basis with a uniform gap). Let V=c0(I,K)V=c_0(I,K) with its maximum norm, where II is finite or countable, and let eie_i be its coordinate basis. Let S⊂K∘S\subset K^\circ have the properties of Lemma 3.14.1. Suppose yj∈c0(I,S)y_j\in c_0(I,S) and that their residues form a basis of V~=K~(I)\widetilde V=\widetilde K^{(I)}. Then the yjy_j are an orthonormal basis of VV.

Proof. For a finite linear combination of the yjy_j, divide the coefficients by one of maximal absolute value. Residual linear independence then proves that its norm is the maximum coefficient norm. Passing to convergent series gives an isometric map c0(J,K)→Vc_0(J,K)\to V with closed image. The same argument identifies the closed SS-span WSW_S of the yjy_j isometrically with c0(J,S)c_0(J,S).

The residue field kSk_S of SS embeds in K~\widetilde K. The residual vectors of the yjy_j have finite support and coordinates in kSk_S. They form a basis already over kSk_S: linear independence descends immediately, and spanning descends by row reduction of the finite matrix involved in any expression for a coordinate vector. Thus, for every ii, there is a finite SS-linear combination ziz_i of the yjy_j whose residue equals that of eie_i. The gap gives ∥ei−zi∥≤θ\|e_i-z_i\|\leq\theta and ∥zi∥=1\|z_i\|=1.

For x=∑iaiei∈c0(I,S)x=\sum_i a_i e_i\in c_0(I,S), put P(x)=∑iaizi∈WSP(x)=\sum_i a_i z_i\in W_S. The series converges and satisfies ∥x−P(x)∥≤θ∥x∥\|x-P(x)\|\leq\theta\|x\|. Repeating this construction on the error gives a convergent sum in WSW_S equal to xx. Hence c0(I,S)=WSc_0(I,S)=W_S. In particular every eie_i belongs to the closed KK-span of the yjy_j. Since these coordinate vectors span a dense subspace of VV, that closed span is all of VV. ◻

Theorem 3.14.3. NAG-001914 (Exact generators for Tate submodules). Put T=K⟨T1,…,Tn⟩T=K\langle T_1,\ldots,T_n\rangle with its Gauss norm. For every TT-submodule N⊂TsN\subset T^s, there are finitely many v1,…,vq∈Nv_1,\ldots,v_q\in N of norm one such that every v∈Nv\in N has an expression v=∑j=1qajvj,aj∈T,max⁡j∥aj∥≤∥v∥.v=\sum_{j=1}^q a_jv_j,\qquad a_j\in T, \qquad \max_j\|a_j\|\leq\|v\|. Consequently N∩(K∘⟨T⟩)sN\cap(K^\circ\langle T\rangle)^s is generated by the same vjv_j over K∘⟨T⟩K^\circ\langle T\rangle.

Proof. The image N‾\overline N of the unit ball of NN in K~[T1,…,Tn]s\widetilde K[T_1,\ldots,T_n]^s is a submodule. The latter polynomial ring is Noetherian by [StacksAlg] (Tag 00FN), so choose generators of N‾\overline N and lift them to norm-one vectors v1,…,vq∈Nv_1,\ldots,v_q\in N. From the monomial multiples TνvjT^\nu v_j select a residual K~\widetilde K-basis of N‾\overline N, and complete it to a basis of K~[T]s\widetilde K[T]^s by adding some coordinate monomials TνeiT^\nu e_i. Denote these selected lifts by (yβ)β∈J(y_\beta)_{\beta\in J} and the first part by J0⊂JJ_0\subset J.

The coefficients of the finitely many vjv_j form a sequence tending to zero. Lemma 3.14.1 places them in a single gap ring SS. All coefficients of every yβy_\beta belong to this same ring. Lemma 3.14.2 therefore shows that (yβ)β∈J(y_\beta)_{\beta\in J} is an orthonormal basis of TsT^s. Expand v=∑βcβyβv=\sum_\beta c_\beta y_\beta and let v0=∑β∈J0cβyβv_0=\sum_{\beta\in J_0}c_\beta y_\beta. Grouping the latter series according to the chosen vjv_j gives v0=∑jajvjv_0=\sum_j a_jv_j with aj∈Ta_j\in T and ∥aj∥≤∥v∥\|a_j\|\leq\|v\|. The coefficient sequences of each aja_j tend to zero because (cβ)∈c0(J,K)(c_\beta)\in c_0(J,K). Hence v0∈Nv_0\in N.

If v−v0v-v_0 were nonzero, divide it by one of its largest coefficients in the orthonormal expansion. Its nonzero reduction would lie both in N‾\overline N and in the span of the complementary residual basis vectors. This is impossible. Thus v=v0v=v_0, which proves the assertion and the unit-ball consequence. ◻

References. The uniform-gap and orthonormal-basis method reconstructs [Bosch14] (§2.3, Proposition 3, Theorem 6 and Corollary 10, pp. 25–29). The explicit operator PP records the uniform contraction in the basis lifting.

For the remainder of this section KK is nontrivially valued, O=K∘O=K^\circ, and π∈O\pi\in O satisfies 0<∣π∣<10<|\pi|<1. An algebra of topologically finite type means a quotient of O⟨T1,…,Tn⟩O\langle T_1,\ldots,T_n\rangle with its quotient π\pi-adic topology.

Theorem 3.14.4. NAG-001915 (Torsion-free finite modules are finitely presented). Let A0A_0 be a topologically finite type OO-algebra and let MM be a finite A0A_0-module with no OO-torsion. Then MM is finitely presented over A0A_0. In particular, a topologically finite type OO-algebra with no π\pi-torsion is topologically finitely presented.

Proof. Put P=O⟨T1,…,Tn⟩P=O\langle T_1,\ldots,T_n\rangle and choose a surjection P↠A0P\twoheadrightarrow A_0. Since MM is finite over PP, take a surjection Ps↠MP^s\twoheadrightarrow M with kernel NN. Inside K⟨T⟩s=Ps[1/π]K\langle T\rangle^s=P^s[1/\pi], torsion-freeness gives N=N[1/π]∩Ps.N=N[1/\pi]\cap P^s. The equality P[1/π]=K⟨T⟩P[1/\pi]=K\langle T\rangle follows by bounding the coefficients of a restricted series. Apply Theorem 3.14.3 to N[1/π]N[1/\pi]. Its norm-one generators lie in NN, and its coefficient bound shows that they generate NN over PP. Thus MM is finitely presented over PP. Their images also generate the kernel of A0s↠MA_0^s\twoheadrightarrow M, so MM is finitely presented over A0A_0. Finally, in a real-valued rank-one field, π\pi-torsion-freeness implies OO-torsion-freeness: for 0≠a∈O0\ne a\in O choose mm with ∣π∣m≤∣a∣|\pi|^m\leq|a|, so that πm∈aO\pi^m\in aO. Apply the result to M=A0M=A_0 as a quotient of PP; its defining ideal is then finitely generated. ◻

Proposition 3.14.5. NAG-001916 (Adic topology on finite formal modules). Let A0A_0 be topologically finite type over OO, let MM be a finite A0A_0-module, and let N⊂MN\subset M be any submodule. Its π\pi-saturation Nsat={m∈M:πem∈N for some e≥0}N^{\mathrm{sat}}=\{m\in M:\pi^e m\in N\text{ for some }e\geq0\} is finitely generated. There is an e≥0e\geq 0 such that, for all n≥0n\geq 0, πn+eM∩N ⊂ πnN ⊂ πnM∩N.\pi^{n+e}M\cap N\ \subset\ \pi^nN\ \subset\ \pi^nM\cap N. Every finite A0A_0-module is π\pi-adically complete and separated. In particular every submodule of a finite module is closed.

Proof. Choose P=O⟨T⟩↠A0P=O\langle T\rangle\twoheadrightarrow A_0 and a finite free PP-module mapping onto MM. Its composite map to M/NsatM/N^{\mathrm{sat}} has finitely generated kernel by the proof of Theorem 3.14.4, since the target is OO-torsion-free. The image of this kernel is NsatN^{\mathrm{sat}}, which is therefore finite. A common exponent ee sends its finite generators into NN. If x=πn+em∈Nx=\pi^{n+e}m\in N, then m∈Nsatm\in N^{\mathrm{sat}} and πem∈N\pi^em\in N, giving the first inclusion.

To prove completeness of a finite module, present it as a quotient of PsP^s. A compatible sequence of classes modulo πn\pi^n has lifts xn∈Psx_n\in P^s with xn+1−xn∈πnPsx_{n+1}-x_n\in\pi^nP^s: recursively adjust a lift by an element of the presentation kernel. The xnx_n converge in the complete module PsP^s, yielding a lift of all the classes. For separatedness take x∈⋂nπnMx\in\bigcap_n\pi^nM and apply the first inclusion to the cyclic submodule A0xA_0x. It gives x∈πA0xx\in\pi A_0x, so x=πaxx=\pi ax for some a∈A0a\in A_0. The element 1−πa1-\pi a is a unit: lift aa to PP and use the convergent geometric series there. Hence x=0x=0. Finally M/NM/N is finite and separated, which says exactly that NN is closed in MM. ◻

Corollary 3.14.6. NAG-001917 (Coherence of formal coefficient rings). Every topologically finitely presented OO-algebra is coherent.

Proof. For P=O⟨T⟩P=O\langle T\rangle, the image of a homomorphism Pr→PsP^r\to P^s is finite and OO-torsion-free, so its presentation kernel is finite by Theorem 3.14.4. Thus all such kernels are finite. Now write A0=P/(g1,…,gt)A_0=P/(g_1,\ldots,g_t). Lift a matrix A0r→A0sA_0^r\to A_0^s to a matrix over PP, and augment it by the stst columns gjeig_j e_i. The kernel of this augmented matrix is finite over PP. Projection to the first rr coordinates, followed by reduction modulo (g1,…,gt)(g_1,\ldots,g_t), maps it onto the kernel of the original matrix. Hence that kernel is finite over A0A_0, which is the finite-matrix criterion for coherence. ◻

References. The formal-module arguments reconstruct the rank-one valuation-ring proof in [Bosch14] (§7.3, Theorem 4, Corollaries 5–6, Lemma 7 and Proposition 8, pp. 163–166). Their analytic input is the exact generator theorem proved above.

Proposition 3.14.7. NAG-001918 (Affinoid generic fibres and integral presentations). Let A0A_0 be a topologically finite type OO-algebra without π\pi-torsion. Then A=A0[1/π]A=A_0[1/\pi] is naturally a strict KK-affinoid algebra. A presentation O⟨T⟩↠A0O\langle T\rangle\twoheadrightarrow A_0 gives a complete quotient norm on AA whose closed unit ball is precisely A0A_0. Conversely every strict KK-affinoid algebra has such an admissible integral presentation. Moreover, A0⟨U1,…,Um⟩[1/π]≃A⟨U1,…,Um⟩.A_0\langle U_1,\ldots,U_m\rangle[1/\pi] \simeq A\langle U_1,\ldots,U_m\rangle. A homomorphism of admissible integral algebras induces a bounded homomorphism of these generic affinoid algebras.

Proof. Write P=O⟨T⟩P=O\langle T\rangle and let JJ be the presentation kernel. Torsion-freeness gives J=J[1/π]∩PJ=J[1/\pi]\cap P. The orthonormal basis constructed in the proof of Theorem 3.14.3, applied to the submodule J[1/π]⊂K⟨T⟩J[1/\pi]\subset K\langle T\rangle, splits that Banach vector space as the orthogonal sum of J[1/π]J[1/\pi] and a closed complement. Every quotient class therefore has a lift attaining its quotient norm. In particular the quotient unit ball is P/J=A0P/J=A_0. The quotient is the affinoid algebra A=A0[1/π]A=A_0[1/\pi]. This also proves the converse: starting with K⟨T⟩/IK\langle T\rangle/I, use P/(I∩P)P/(I\cap P), which is torsion-free and hence topologically finitely presented by Theorem 3.14.4.

The π\pi-adic topology on A0A_0 is the topology induced by this quotient norm. A restricted series with coefficients in AA has bounded coefficients, so multiplication by a single power of π\pi puts all of them in A0A_0; their convergence to zero is then equivalent to π\pi-adic convergence. This proves the displayed identity. Norm-attaining coefficient lifts also show that O⟨T,U⟩↠A0⟨U⟩O\langle T,U\rangle\twoheadrightarrow A_0\langle U\rangle is surjective. The latter ring is torsion-free and topologically finite type, hence admissible. Finally a map sending A0A_0 into B0B_0 is bounded after inverting π\pi: scaling by an appropriate power of π\pi gives a bound with constant ∣π∣−1|\pi|^{-1} in the two quotient norms. ◻

Lemma 3.14.8. NAG-001919 (Completion of an algebraic quotient). For a ring RR with its π\pi-adic topology and an ideal J⊂RJ\subset R, the natural map R^→R/J^\widehat R\to\widehat{R/J} is surjective and its kernel is the closure of the image of JJ in R^\widehat R. If multiplication by π\pi is injective on R/JR/J, it is injective on R/J^\widehat{R/J} as well, and (R/J)/πn(R/J)≃R/J^/πnR/J^(n≥1).(R/J)/\pi^n(R/J)\simeq \widehat{R/J}/\pi^n\widehat{R/J}\quad(n\geq1).

Proof. Lift a compatible system in R/(J+πnR)R/(J+\pi^nR) recursively to rn∈Rr_n\in R. Compatibility allows one to alter rn+1r_{n+1} by an element of JJ so that rn+1−rn∈πnRr_{n+1}-r_n\in\pi^nR. The resulting Cauchy system defines the required element of R^\widehat R. An element in the kernel can, modulo every πn\pi^n, be represented by an element of JJ, which is precisely the asserted closure condition. Put C=R/JC=R/J. If πx=0\pi x=0 in C^\widehat C, choose a representative c∈Cc\in C of its component modulo πn+1\pi^{n+1}. Then πc∈πn+1C\pi c\in\pi^{n+1}C; cancellation in CC yields c∈πnCc\in\pi^nC. Thus every component of xx is zero. Finally a compatible system vanishing modulo πn\pi^n can be divided by πn\pi^n componentwise, using the same cancellation to verify compatibility. The kernel of C^→C/πnC\widehat C\to C/\pi^nC is therefore πnC^\pi^n\widehat C, proving the last assertion. ◻

3.15 Admissible blowups on affine formal models

Theorem 3.15.1. NAG-001921 (The charts of an admissible formal blowup). Let A0A_0 be an admissible OO-algebra and let I=(f0,…,fm)⊂A0I=(f_0,\ldots,f_m)\subset A_0 contain πN\pi^N for some NN. For each ii, set Ci=A0[Uj:j≠i]/(fiUj−fj:j≠i),Bi=Ci/(π-torsion)^.C_i=A_0[U_j:j\ne i]/(f_iU_j-f_j:j\ne i),\qquad B_i=\widehat{C_i/(\pi\text{-torsion})}. The BiB_i are admissible OO-algebras and are the affine charts of the π\pi-adic completion of Bl⁡I(Spec⁡A0)\operatorname{Bl}_I(\operatorname{Spec}A_0). The ideal IBiIB_i is generated by the nonzerodivisor fif_i. Equivalently, Bi≃(A0⟨Uj:j≠i⟩/(fiUj−fj:j≠i))/(π-torsion),B_i\simeq \left(A_0\langle U_j:j\ne i\rangle/(f_iU_j-f_j:j\ne i)\right) /(\pi\text{-torsion}), and in the parenthesized ring its π\pi-torsion equals its fif_i-torsion.

Proof. Write Ri=A0[Uj:j≠i]R_i=A_0[U_j:j\ne i] and let LiL_i be the ideal of the displayed relations. Evaluation Uj↦fj/fiU_j\mapsto f_j/f_i maps RiR_i onto the subring Di=A0[I/fi]⊂A0[1/fi]D_i=A_0[I/f_i]\subset A_0[1/f_i]. Its kernel is Li:fi∞L_i:f_i^\infty: after localizing at fif_i, the relations eliminate all the variables, and clearing denominators gives the assertion. Choose aj∈A0a_j\in A_0 with πN=∑jajfj\pi^N=\sum_j a_jf_j. In CiC_i this becomes πN=fi(ai+∑j≠iajUj).(∗)\pi^N=f_i\left(a_i+\sum_{j\ne i}a_jU_j\right). \tag{\(*\)} Consequently fif_i-torsion in CiC_i is π\pi-torsion. Conversely DiD_i, being a subring of A0[1/fi]A_0[1/f_i], is π\pi-torsion-free, so all π\pi-torsion in CiC_i lies in that kernel. Thus Di=Ci/(π-torsion)D_i=C_i/(\pi\text{-torsion}). By Lemma 3.14.8, Bi=D^iB_i=\widehat D_i is π\pi-torsion-free and is a quotient of A0⟨Uj:j≠i⟩A_0\langle U_j:j\ne i\rangle. It is topologically finite type, and Theorem 3.14.4 makes it topologically finitely presented. Equation (∗)(*) shows that fif_i is a nonzerodivisor on BiB_i and that IBi=(fi)IB_i=(f_i).

The degree-zero part of the localization of the Rees algebra ⨁d≥0Idtd\bigoplus_{d\geq 0}I^dt^d at fitf_it is precisely DiD_i: its elements are g/fidg/f_i^d with g∈Idg\in I^d. Equality of two such fractions in A0[1/fi]A_0[1/f_i] is the equality obtained in the homogeneous localization, because both are checked after multiplication by a power of fif_i. These standard affine opens cover the Proj of the Rees algebra. Completing them gives the asserted blowup charts.

It remains to compare the two orders of completion and torsion removal. Put Ci′=A0⟨U⟩/LiA0⟨U⟩C_i'=A_0\langle U\rangle/L_iA_0\langle U\rangle. The polynomial π\pi-torsion in CiC_i maps into the π\pi-torsion τ\tau of Ci′C_i'. The latter is finitely generated by Proposition 3.14.5, hence killed by one power of π\pi; its induced topology is therefore discrete. Its image in BiB_i vanishes. Conversely the kernel of Ci′→BiC_i'\to B_i is the closure of the image of the polynomial torsion, by Lemma 3.14.8. The submodule τ\tau is closed by Proposition 3.14.5, so this closure is contained in τ\tau. The reverse inclusion holds because BiB_i is π\pi-torsion-free. Thus this kernel equals τ\tau. Its dense polynomial-torsion submodule equals τ\tau because τ\tau is discrete. Every element of τ\tau is consequently killed by a power of fif_i, as is already true polynomially. Equation (∗)(*) gives the reverse torsion inclusion, completing the comparison. ◻

Theorem 3.15.2. NAG-001922 (Generic-fibre invariance of an affine admissible blowup). In Theorem 3.15.1, let A=A0[1/π]A=A_0[1/\pi]. There are canonical bounded isomorphisms Bi[1/π]≃A⟨Uj:j≠i⟩/(fiUj−fj:j≠i).B_i[1/\pi]\simeq A\langle U_j:j\ne i\rangle/(f_iU_j-f_j:j\ne i). The associated spectra are the rational domains Vi={x∈M(A):∣fj(x)∣≤∣fi(x)∣ for all j}.V_i=\{x\in\mathcal M(A): |f_j(x)|\leq|f_i(x)| \text{ for all }j\}. They cover M(A)\mathcal M(A), and the formal chart transitions induce exactly their rational-domain intersections. Thus the generic fibre of the completed blowup, glued from these charts, is canonically M(A)\mathcal M(A).

Proof. The final presentation in Theorem 3.15.1 and Proposition 3.14.7 give the displayed algebra isomorphism. It is bounded in both directions: all coordinate images lie in the respective integral unit balls, and the finite affinoid quotient topologies agree. Equation (∗)(*) in the preceding proof makes fif_i invertible in the generic chart. Its inverse is π−N(ai+∑j≠iajUj)\pi^{-N}(a_i+\sum_{j\ne i}a_jU_j). Thus the bounded universal property of this quotient is exactly that of the rational domain ViV_i, with Uj=fj/fiU_j=f_j/f_i.

At every point of M(A)\mathcal M(A) at least one fif_i is nonzero, since πN∈I\pi^N\in I. An index of maximal absolute value therefore supplies a chart containing that point. On the overlap of the iith and ℓ\ellth algebraic blowup charts, fℓ/fif_\ell/f_i is invertible. Its formal completion has generic algebra obtained by adjoining its inverse with norm at most one. Since ∣fℓ/fi∣≤1|f_\ell/f_i|\leq 1 already on ViV_i, this cuts out ∣fℓ/fi∣=1|f_\ell/f_i|=1, precisely Vi∩VℓV_i\cap V_\ell. The inverse-coordinate substitutions agree on triple intersections. The rational-domain structure sheaf consequently identifies the gluing of all generic charts with M(A)\mathcal M(A). ◻

Proposition 3.15.3. NAG-001923 (The formal universal property). Let X=Spf⁡(A0)\mathfrak X=\operatorname{Spf}(A_0) and let X′→X\mathfrak X'\to\mathfrak X be the completed blowup above. Every morphism of adic formal OO-schemes h:Y→Xh:\mathfrak Y\to\mathfrak X for which IOYI\mathcal O_{\mathfrak Y} is an invertible ideal factors uniquely through X′\mathfrak X'.

Proof. Locally on Y\mathfrak Y, one of the fif_i generates the invertible ideal: after choosing a local generator, the coefficients of the finitely many fif_i generate the unit ideal, so one is a unit near the given point. On such an affine formal open Spf⁡(B)\operatorname{Spf}(B), the generator fif_i is a nonzerodivisor. There are unique uj∈Bu_j\in B with fiuj=fjf_iu_j=f_j. The corresponding polynomial map Ci→BC_i\to B kills its fif_i-torsion, hence factors through DiD_i. It extends uniquely to the adic completion BiB_i, since BB is complete. The uniqueness of the fractions makes these maps agree on overlaps. Conversely any factorization must map this open into the chart where fif_i generates the pulled-back ideal and must have the specified coordinates. This proves global uniqueness. ◻

References. The chart and universal-property statements are those of [Bosch14] (§8.2, Propositions 6–9, pp. 186–189). The proof above obtains the completion and torsion comparisons from exact unit-ball generators and the adic module topology, and gives the rational-domain identification explicitly.

3.16 Canonical reduction and finite morphisms

We now return to arbitrary complete ground fields, including trivially valued ones. Throughout this section the affinoid algebras are strict. The finite Gauss fibre used above also bounds the residue fields in which the canonical reduction embeds. The allowed algebraic input, finiteness of normalization for polynomial rings over a field, then supplies its finite generation.

Lemma 3.16.1. NAG-001925 (Residue fields of the normalization Gauss fibre). Let R=K⟨T1,…,Td⟩R=K\langle T_1,\ldots,T_d\rangle, and let HH be the completion of Frac⁡(R)\operatorname{Frac}(R) for the Gauss absolute value. Then H~=K~(T1,…,Td)\widetilde H=\widetilde K(T_1,\ldots,T_d). If L/HL/H is finite with its extending absolute value, then [L~:H~]≤[L:H][\widetilde L:\widetilde H]\leq[L:H].

Proof. An element of H∘H^\circ can be approximated to error less than one by a fraction f/gf/g with f,g∈Rf,g\in R and ∥f∥≤∥g∥\|f\|\leq\|g\|. The Gauss norm of a nonzero series belongs to ∣K×∣|K^\times|; divide both ff and gg by a scalar of size ∥g∥\|g\|. Their reductions are polynomials, and the denominator has nonzero reduction. Thus every residue of HH lies in K~(T)\widetilde K(T). Conversely, lift the numerator and denominator of a rational function over K~\widetilde K to polynomials of Gauss norm at most one, with the denominator of norm one. Their quotient in HH has the prescribed residue.

Lift a finite linearly independent family of elements of L~\widetilde L to norm-one elements u1,…,uq∈Lu_1,\ldots,u_q\in L. If they satisfied a nonzero HH-linear relation, divide its coefficients by a coefficient of maximal absolute value. Reduction would give a nonzero H~\widetilde H-linear relation among the chosen residues. Hence the uiu_i are linearly independent over HH, so q≤[L:H]q\leq[L:H]. This bounds the dimension of the entire residue extension and proves the claim. ◻

Proposition 3.16.2. NAG-001926 (Finite canonical reduction over a normalization). Let BB be a strict affinoid domain with a bounded finite injection R=K⟨T1,…,Td⟩↪BR=K\langle T_1,\ldots,T_d\rangle\hookrightarrow B. Then the induced map K~[T1,…,Td]=R~→B~\widetilde K[T_1,\ldots,T_d]=\widetilde R\to\widetilde B is finite.

Proof. Use the finite extensions L1,…,LsL_1,\ldots,L_s of the Gauss field HH constructed in Proposition 3.12.2. They give an injective map B~⟶∏i=1sL~i.\widetilde B\longrightarrow\prod_{i=1}^s\widetilde L_i. Indeed, a power-bounded element maps to zero in every residue field exactly when all its finitely many boundary values are less than one, which is exactly ρB(b)<1\rho_B(b)<1. Put k=K~k=\widetilde K and S=k[T1,…,Td]S=k[T_1,\ldots,T_d]. For b∈B∘b\in B^\circ, its minimal polynomial over Frac⁡(R)\operatorname{Frac}(R) has coefficients in RR, as in Lemma 1.13.3. The root calculation in Proposition 3.12.2 gives ∥cj∥1/j≤ρB(b)≤1\|c_j\|^{1/j}\leq\rho_B(b)\leq 1 for each coefficient cjc_j. Reducing that monic equation shows that b~\widetilde b is integral over SS.

By Lemma 3.16.1, each L~i\widetilde L_i is a finite extension of Frac⁡(S)\operatorname{Frac}(S). The polynomial ring SS is Nagata by [StacksAlg] (Tag 0335, Proposition 10.162.16); hence its integral closure SiS_i in L~i\widetilde L_i is finite over SS. The preceding injection is SS-linear and every coordinate is integral over SS. Its image is therefore an SS-submodule of the finite SS-module ∏i=1sSi\prod_{i=1}^s S_i. The Nagata property already gives Noetherianity of SS by definition [StacksAlg] (Tag 032R(2)). The submodule assertion of [StacksAlg] (Tag 00IK, Lemma 10.51.1) now makes B~\widetilde B finite over SS. ◻

Theorem 3.16.3. NAG-001927 (Finite generation of strict canonical reduction). For every admissible strict presentation K⟨X1,…,Xn⟩↠AK\langle X_1,\ldots,X_n\rangle\twoheadrightarrow A, the map K~[X1,…,Xn]⟶A~\widetilde K[X_1,\ldots,X_n]\longrightarrow\widetilde A is finite. In particular A~\widetilde A is a reduced finite type K~\widetilde K-algebra.

Proof. The zero algebra is immediate. Otherwise take the finitely many minimal primes pj\mathfrak p_j of AA. The canonical map A~⟶∏jA/pj~\widetilde A\longrightarrow\prod_j\widetilde{A/\mathfrak p_j} is injective: if an element has spectral radius less than one on every component, the finite maximum of those radii is less than one on AA. For each quotient, apply the presentation-compatible normalization theorem, Theorem 1.12.2. It gives a bounded finite injection from a Tate algebra RjR_j and a contractive lift Rj→K⟨X⟩R_j\to K\langle X\rangle. Proposition 3.16.2 makes A/pj~\widetilde{A/\mathfrak p_j} finite over R~j\widetilde R_j, and therefore finite over K~[X]\widetilde K[X] via that lift. Their finite product is a finite module over the Noetherian ring K~[X]\widetilde K[X]. The displayed injection makes A~\widetilde A a submodule, so it too is finite. Reducedness was proved in Proposition 3.12.6. ◻

Lemma 3.16.4. NAG-001928 (Simultaneous monic division with a small perturbation). Let RR be a complete non-Archimedean Banach algebra. For 1≤i≤m1\leq i\leq m, let Pi∈R[Yi]P_i\in R[Y_i] be monic of degree di>0d_i>0 with all lower coefficients of norm at most one, and let Gi∈R⟨Y1,…,Ym⟩G_i\in R\langle Y_1,\ldots,Y_m\rangle satisfy ∥Gi∥<1\|G_i\|<1. Then the quotient by the Pi−GiP_i-G_i is generated over RR by the monomials YνY^\nu with 0≤νi<di0\leq\nu_i<d_i. These generators admit coefficient lifts bounded by the norm of a representative.

Proof. Successively divide by the univariate monic polynomials PiP_i, using Theorem 1.10.3. Since the coefficients of PiP_i lie in RR, division in YiY_i preserves the degree bounds already obtained in the earlier variables. This gives bounded RR-linear maps DiD_i and rr, all of norm at most one, with u=r(u)+∑iPiDi(u),r(u)∈∑0≤νi<diRYν.u=r(u)+\sum_iP_iD_i(u),\qquad r(u)\in\sum_{0\leq\nu_i<d_i}RY^\nu. Put E(u)=∑iGiDi(u)E(u)=\sum_iG_iD_i(u) and θ=max⁡i∥Gi∥<1\theta=\max_i\|G_i\|<1. Then ∥E(u)∥≤θ∥u∥\|E(u)\|\leq\theta\|u\|. Substituting Pi=(Pi−Gi)+GiP_i=(P_i-G_i)+G_i and iterating yields the convergent identity u=∑n≥0r(Enu)+∑i(Pi−Gi)∑n≥0Di(Enu).u=\sum_{n\geq0}r(E^nu) +\sum_i(P_i-G_i)\sum_{n\geq0}D_i(E^nu). The first sum remains in the finite free monomial module and has coefficient norm at most ∥u∥\|u\|. This proves both assertions. ◻

Theorem 3.16.5. NAG-001929 (The strict reduction criterion for finiteness). A bounded homomorphism A→BA\to B of strict KK-affinoid algebras is finite if and only if the induced homomorphism A~→B~\widetilde A\to\widetilde B is finite.

Proof. Suppose first that BB is finite over AA. For each minimal prime q\mathfrak q of BB, the domain B/qB/\mathfrak q is finite over A/(A∩q)A/(A\cap\mathfrak q). Normalize the latter using a fixed strict presentation of AA. The resulting normalizing Tate algebra RqR_{\mathfrak q} lifts contractively to that presentation, and B/qB/\mathfrak q is finite over RqR_{\mathfrak q}. Proposition 3.16.2 shows that B/q~\widetilde{B/\mathfrak q} is finite over R~q\widetilde R_{\mathfrak q}, hence finite over A~\widetilde A. The injection B~↪∏qB/q~\widetilde B\hookrightarrow\prod_{\mathfrak q} \widetilde{B/\mathfrak q} and the Noetherian property of A~\widetilde A (Theorem 3.16.3) prove this direction.

Conversely assume B~\widetilde B is finite over A~\widetilde A. Choose the images b1,…,bmb_1,\ldots,b_m of the coordinates in a strict presentation of BB; they belong to B∘B^\circ. Each residue b~i\widetilde b_i satisfies a monic equation over A~\widetilde A: one may obtain it by applying the determinant trick to multiplication by b~i\widetilde b_i on a finite generating module. Lift its coefficients to cij∈A∘c_{ij}\in A^\circ. Then wi=bidi+∑j<dicijbij,ρB(wi)<1.w_i=b_i^{d_i}+\sum_{j<d_i}c_{ij}b_i^j, \qquad \rho_B(w_i)<1. Append variables mapping to the finitely many power-bounded cijc_{ij} to a strict presentation of AA. Bounded substitution gives an admissible surjection R↠AR\twoheadrightarrow A from a Tate algebra for which these coefficients have lifts of norm at most one. The induced map q:R⟨Y1,…,Ym⟩↠Bq:R\langle Y_1,\ldots,Y_m\rangle\twoheadrightarrow B is admissible: the chosen presentation of BB, using the YiY_i alone, already supplies uniformly bounded lifts. Let Wi∈R[Yi]W_i\in R[Y_i] be the lifted monic polynomial. Since wiw_i is topologically nilpotent, choose ei>0e_i>0 and GiG_i of norm less than one with q(Gi)=wieiq(G_i)=w_i^{e_i}. Thus Wiei−Gi∈ker⁡qW_i^{e_i}-G_i\in\ker q.

Apply Lemma 3.16.4 to the univariate polynomials WieiW_i^{e_i}. It shows that their perturbed quotient, and hence BB, is finite over RR. The action of RR on BB factors through AA, so the same finite list of generators generates BB as an AA-module. ◻

References. This is the strict canonical-reduction criterion used in the proof of [Tem04] (Proposition 3.1(iii)). Its reconstruction here uses finite normalization, the finite Gauss fibre, the Stacks normalization theorem, and the explicit simultaneous division argument.

Lemma 3.16.6. NAG-001930 (Spectral values in the strict case). For a strict KK-affinoid algebra AA, every nonzero spectral radius ρA(a)\rho_A(a) belongs to ∣K×∣\sqrt{|K^\times|}. The same holds for every nonzero value of aa at a Shilov point.

Proof. In the finite-boundary construction, each point is obtained from a finite extension L/HL/H of a completed Gauss field. The value group of HH is ∣K×∣|K^\times|: a nonzero limit has the absolute value of any sufficiently accurate approximant, and fractions of Gauss series have values in ∣K×∣|K^\times|. For 0≠u∈L0\ne u\in L, a polynomial equation over HH has at least two terms of maximal absolute value. Equality of their sizes gives ∣u∣e∈∣H×∣|u|^e\in|H^\times| for some nonzero integer ee. Thus all nonzero values at the constructed finite boundary lie in ∣K×∣\sqrt{|K^\times|}. Their finite maximum is ρA(a)\rho_A(a). The minimal Shilov boundary is a subset of this boundary by Lemma 3.12.4, proving both assertions. ◻

Theorem 3.16.7. NAG-001931 (Generic points of strict canonical reduction). For a strict affinoid algebra AA, the reduction kernels Pξ=ker⁡(A~⟶H(ξ)~),ξ∈ΓA,P_\xi=\ker\bigl(\widetilde A\longrightarrow \widetilde{\mathcal H(\xi)}\bigr), \qquad \xi\in\Gamma_A, are exactly the minimal prime ideals of A~\widetilde A. They are distinct, and each has the unique analytic preimage ξ\xi.

Proof. These are prime ideals and their intersection is zero, by the finite boundary maximum formula. For each ξ\xi, choose aξa_\xi peaking at ξ\xi among ΓA\Gamma_A. If its maximum is mm, Lemma 3.16.6 gives e>0e>0 and c∈K×c\in K^\times with ∣c∣=me|c|=m^e. Replacing aξa_\xi by aξe/ca_\xi^e/c produces a power-bounded function with value one at ξ\xi and value strictly less than one at every other Shilov point. Its reduction is in every PηP_\eta except PξP_\xi. Thus these primes are pairwise incomparable. Their zero intersection implies that every prime contains at least one of them, by taking a product of elements outside that prime if none were contained. They are therefore exactly the minimal primes.

Suppose an analytic point yy has reduction kernel PξP_\xi, and use the normalized peaking function a=aξa=a_\xi. Then ∣a(y)∣=1|a(y)|=1. If ∣b(ξ)∣>0|b(\xi)|>0, choose nn large enough that ρ(anb)=∣b(ξ)∣\rho(a^nb)=|b(\xi)|, then choose e>0e>0 and c∈K×c\in K^\times with ∣c∣=∣b(ξ)∣e|c|=|b(\xi)|^e. The power-bounded element anebe/ca^{ne}b^e/c has value one at ξ\xi; its reduction is outside PξP_\xi and hence has value one at yy. It follows that ∣b(y)∣=∣b(ξ)∣|b(y)|=|b(\xi)|. If b(ξ)=0b(\xi)=0, put q=max⁡η≠ξ∣a(η)∣<1q=\max_{\eta\ne\xi}|a(\eta)|<1, with empty maxima equal to zero. The estimates ∣b(y)∣≤qnmax⁡η≠ξ∣b(η)∣|b(y)|\leq q^n\max_{\eta\ne\xi}|b(\eta)| for all nn give b(y)=0b(y)=0. Thus yy and ξ\xi agree on every element of AA, and are the same analytic point. ◻

Finite changes of affine formal models

Theorem 3.16.8. NAG-001932 (Adjoining power-bounded functions). Let A0A_0 be an admissible K∘K^\circ-algebra and A=A0[1/π]A=A_0[1/\pi]. If f1,…,fs∈A∘f_1,\ldots,f_s\in A^\circ, then B0=A0[f1,…,fs]⊂AB_0=A_0[f_1,\ldots,f_s]\subset A is a finite admissible A0A_0-algebra, with B0[1/π]=AB_0[1/\pi]=A. There is a nonzero c∈K∘∘c\in K^{\circ\circ} with cfi∈A0cf_i\in A_0, and Spf⁡(B0)→Spf⁡(A0)\operatorname{Spf}(B_0)\to\operatorname{Spf}(A_0) is the admissible blowup of the open ideal (c,cf1,…,cfs)(c,cf_1,\ldots,cf_s).

Proof. Choose an admissible presentation of A0A_0. By Proposition 3.14.7, A0A_0 is the closed unit ball for the induced quotient norm on AA. The monic equations in Theorem 1.13.4 therefore show that each fif_i is integral over A0A_0. Successively reducing powers by those equations makes B0B_0 a finite A0A_0-module. It is complete by Proposition 3.14.5. Sending restricted-series variables to the fif_i is well defined, since all monomials in them belong to the bounded set B0B_0; this gives a topologically finite type presentation over A0A_0, and hence over K∘K^\circ. As a subring of AA, B0B_0 is π\pi-torsion free, and Theorem 3.14.4 makes it admissible.

Clear the finitely many denominators to choose cc, and put I=(c,cf1,…,cfs)I=(c,cf_1,\ldots,cf_s). A power of π\pi belongs to cA0cA_0, so II is open. Its extension to B0B_0 is (c)(c). Consider an admissible formal scheme Y\mathfrak Y over Spf⁡(A0)\operatorname{Spf}(A_0) on which II becomes invertible. Locally one of the displayed generators generates IOYI\mathcal O_{\mathfrak Y}. If cc does, every quotient cfj/c=fjcf_j/c=f_j is a section. If cficf_i does, the equation c=(cfi)dc=(cf_i)d gives d=fi−1d=f_i^{-1} on the generic algebra. The monic integral equation for fif_i, multiplied by fi1−rf_i^{1-r} when its degree is rr, expresses fif_i as a polynomial in dd with coefficients in A0A_0. Thus fif_i is a unit in OY\mathcal O_{\mathfrak Y} and cc also generates IOYI\mathcal O_{\mathfrak Y}. In either case all fjf_j extend uniquely. The extensions glue, since admissible formal rings inject into their generic algebras. They give the unique factorization through Spf⁡(B0)\operatorname{Spf}(B_0).

The actual blowup is admissible by Theorem 3.15.1. Its universal property and the property just proved give inverse morphisms between it and Spf⁡(B0)\operatorname{Spf}(B_0). This proves the last assertion. ◻

Corollary 3.16.9. NAG-001933 (Lifting an affine analytic morphism). Let A0,B0A_0,B_0 be admissible K∘K^\circ-algebras. Every bounded KK-homomorphism φ:B0[1/π]→A0[1/π]\varphi:B_0[1/\pi]\to A_0[1/\pi] extends to B0→A0′B_0\to A_0' after a finite admissible blowup Spf⁡(A0′)→Spf⁡(A0)\operatorname{Spf}(A_0')\to\operatorname{Spf}(A_0). Two homomorphisms between admissible formal algebras that agree after inverting π\pi are equal.

Proof. Choose finitely many topological generators bib_i of B0B_0. They are power bounded in its generic algebra. A bounded map takes power-bounded elements to power-bounded elements, since it takes the bounded set of their powers to a bounded set. Apply Theorem 3.16.8 to fi=φ(bi)f_i=\varphi(b_i). Evaluation of restricted series now gives B0→A0′B_0\to A_0'. All defining relations vanish because they vanish in A0′[1/π]A_0'[1/\pi] and A0′A_0' has no π\pi-torsion. The same injectivity proves the last assertion. ◻

Corollary 3.16.10. NAG-001934 (A common affine formal model). Two admissible affine formal models of the same strict KK-affinoid algebra are dominated by a common admissible affine model whose morphisms to both are finite admissible blowups.

Proof. Regard the two models A0,B0A_0,B_0 as subrings of their common generic algebra AA. If b1,…,bsb_1,\ldots,b_s topologically generate B0B_0, put C0=A0[b1,…,bs]C_0=A_0[b_1,\ldots,b_s]. Theorem 3.16.8 makes this a complete finite model containing B0B_0: any restricted series in the bib_i converges in C0C_0 to its value in AA. Thus C0=A0[B0]C_0=A_0[B_0]. Reversing the roles gives D0=B0[A0]D_0=B_0[A_0] by the same argument. Both are the subring of AA generated by A0A_0 and B0B_0, so C0=D0C_0=D_0. The cited theorem gives both required blowups. ◻

References. The finite-model modification is the construction in [Bosch14] (8.4, Lemma 6, pp. 208–209). Here integrality, completeness and finite presentation are supplied by the proved unit-ball and formal-module statements, and the blowup identification follows from its explicit universal property.

Specialization to arbitrary reduction points

Lemma 3.16.11. NAG-001935 (Integral finite normalization of a formal model). Let A0≠0A_0\ne 0 be an admissible K∘K^\circ-algebra. There is a finite injective continuous homomorphism R0=K∘⟨Y1,…,Yd⟩↪A0R_0=K^\circ\langle Y_1,\ldots,Y_d\rangle\hookrightarrow A_0. The ring R0R_0 is an integrally closed domain.

Proof. Use the presentation K∘⟨X⟩↠A0K^\circ\langle X\rangle\twoheadrightarrow A_0 and its generic-fibre presentation. The inductive proof of Theorem 1.12.2 retains integral coefficients. Here are the norm bounds that verify this assertion. The coordinate substitutions used there are isometries and therefore preserve the unit balls. For the normalized distinguished element, Weierstrass division sends an input of norm at most one to a quotient and a polynomial remainder whose coefficients have norm at most one. Consequently the images of 1,Xn,…,Xns−11,X_n,\ldots,X_n^{s-1} generate the unit ball of the quotient over the image of K∘⟨X1,…,Xn−1⟩K^\circ\langle X_1,\ldots,X_{n-1}\rangle. Every element of A0A_0 admits a lift of norm at most one by Proposition 3.14.7, so this assertion applies to every element, not merely to a dense subring.

The intermediate quotient in the induction is again an admissible formal algebra: it is the quotient of a restricted-series algebra by the contraction of the generic kernel, hence is torsion free and topologically finite type; its finite presentation follows from Theorem 3.14.4. Apply the same argument to it and multiply the resulting finite generating lists. The base case is the full restricted-series ring. Thus the normalization injection from the cited theorem restricts to a finite injection R0↪A0R_0\hookrightarrow A_0.

The fraction field of R0R_0 is that of R=K⟨Y⟩R=K\langle Y\rangle, since inverting π\pi gives RR. An element integral over R0R_0 is integral over RR and hence lies in RR by Theorem 1.13.1. In its monic integral equation over R0R_0, a Gauss norm greater than one would make the leading term strictly dominant. Thus its Gauss norm is at most one, and it belongs to R0R_0. ◻

Lemma 3.16.12. NAG-001936 (Lifting finitely generated residue fields). Let KK be complete and let l/K~l/\widetilde K be a finitely generated field extension. There is a complete valued extension L/KL/K with a K~\widetilde K-embedding l↪L~l\hookrightarrow\widetilde L.

Proof. Choose a transcendence basis t1,…,tet_1,\ldots,t_e such that l/K~(t1,…,te)l/\widetilde K(t_1,\ldots,t_e) is finite. The completion of Frac⁡(K⟨T1,…,Te⟩)\operatorname{Frac}(K\langle T_1,\ldots,T_e\rangle) for the Gauss norm has residue field K~(t1,…,te)\widetilde K(t_1,\ldots,t_e) by Lemma 3.16.1. Express the remaining finite extension as a tower of simple extensions. At one step, lift the coefficients of the monic irreducible residue polynomial to the valuation ring of the current complete field. A root in a finite extension has absolute value at most one, by Lemma 3.12.1. Its residue satisfies the chosen irreducible polynomial, so the resulting residue field contains the required simple extension. The extending absolute value and completeness are supplied by Lemma 1.12.1. Repeating the finitely many steps proves the assertion, including inseparable residue extensions. ◻

Lemma 3.16.13. NAG-001937 (A zero-kernel point above a polydisc reduction point). Assume KK is nontrivially valued. For every prime p⊂K~[Y1,…,Yd]\mathfrak p\subset\widetilde K[Y_1,\ldots,Y_d], there is a bounded multiplicative norm on K⟨Y⟩K\langle Y\rangle whose reduction kernel is p\mathfrak p.

Proof. Put l=Frac⁡(K~[Y]/p)l=\operatorname{Frac}(\widetilde K[Y]/\mathfrak p) and realize it in L~\widetilde L by Lemma 3.16.12. Choose ai∈L∘a_i\in L^\circ lifting the images of YiY_i. In L⟨U1,…,Ud⟩L\langle U_1,\ldots,U_d\rangle substitute Yi=ai+πUiY_i=a_i+\pi U_i and take the Gauss norm. The resulting homomorphism is injective: scalar extension K⟨Y⟩↪L⟨Y⟩K\langle Y\rangle\hookrightarrow L\langle Y\rangle is coefficientwise injective, translation by the aia_i is an isometric automorphism of the latter Tate algebra, and Yi↦πUiY_i\mapsto\pi U_i is coefficientwise injective. The map is contractive and the Gauss norm multiplicative. Modulo elements of norm less than one, the substitution is Yi↦a~iY_i\mapsto\widetilde a_i, whose kernel is exactly p\mathfrak p. This gives the required norm with zero kernel. ◻

Lemma 3.16.14. NAG-001938 (Prescribing the center in an integral extension). Let RR be an integrally closed domain with fraction field FF, and let BB be an integral domain finite over RR. Suppose a real-valued non-Archimedean absolute value on FF is at most one on RR, with center p={r∈R:∣r∣<1}\mathfrak p=\{r\in R:|r|<1\}. For every prime q⊂B\mathfrak q\subset B above p\mathfrak p, the absolute value extends to Frac⁡(B)\operatorname{Frac}(B) with center exactly q\mathfrak q.

Proof. Let E=Frac⁡(B)E=\operatorname{Frac}(B) and choose a finite normal extension N/FN/F containing EE, by taking a splitting field of finitely many minimal polynomials; see [StacksAlg] (Tag 09HU). Let CC be the integral closure of RR in NN. There is a prime QQ of CC above q\mathfrak q, since B⊂CB\subset C is integral and lying-over applies [StacksAlg] (Tag 00GQ). An extension of the absolute value to NN exists without a completeness assumption on FF: complete FF to HH, choose a maximal ideal of the nonzero finite HH-algebra N⊗FHN\otimes_F H, and use Lemma 1.12.1 on its residue field. The resulting map from the field NN is injective. Denote the extended absolute value by ∣⋅∣0|\cdot|_0. Every element of CC has norm at most one, by its monic equation over RR. Thus it has a center Q0Q_0 in CC, also above p\mathfrak p.

Put G=Aut⁡F(N)G=\operatorname{Aut}_F(N). The fixed field NG/FN^G/F is purely inseparable by [StacksAlg] (Tag 030M). Every element z∈CGz\in C^G consequently has a power zpe∈Rz^{p^e}\in R in positive characteristic, and already lies in RR in characteristic zero: the power lies in FF and is integral over the normal ring RR. It follows that QQ and Q0Q_0 have the same contraction to CGC^G, because membership of zz is determined by membership of that power. By [StacksAlg] (Tag 0BRI), some σ∈G\sigma\in G satisfies σ(Q)=Q0\sigma(Q)=Q_0. The absolute value ∣z∣=∣σ(z)∣0|z|=|\sigma(z)|_0 then has center QQ on CC. Its restriction to EE has center q\mathfrak q on BB, as required. ◻

Theorem 3.16.15. NAG-001939 (Surjectivity of formal specialization). Let A0A_0 be an admissible K∘K^\circ-algebra and A=A0[1/π]A=A_0[1/\pi]. The map sp⁡:M(A)⟶Spec⁡(A0/K∘∘A0),x⟼{a∈A0:∣a(x)∣<1}/K∘∘A0\operatorname{sp}:\mathcal M(A)\longrightarrow \operatorname{Spec}(A_0/K^{\circ\circ}A_0),\qquad x\longmapsto\{a\in A_0:|a(x)|<1\}/K^{\circ\circ}A_0 is well defined and surjective onto all prime ideals.

Proof. Every analytic point has value at most one on A0A_0, by Proposition 3.14.7 and Lemma 1.6.1. Its elements of value less than one form a prime ideal containing K∘∘A0K^{\circ\circ}A_0, which proves well-definedness.

First suppose A0A_0 is a domain, and let q⊂A0\mathfrak q\subset A_0 represent a given special-fibre point. By Lemma 3.16.11, choose a finite injection R0=K∘⟨Y⟩↪A0R_0=K^\circ\langle Y\rangle\hookrightarrow A_0. The contraction of q\mathfrak q defines a prime of K~[Y]\widetilde K[Y]. Lemma 3.16.13 gives a multiplicative norm on K⟨Y⟩K\langle Y\rangle centered there. Extend it to the fraction field of R0R_0 by quotients. Lemma 3.16.14 extends this absolute value to Frac⁡(A0)\operatorname{Frac}(A_0) with center q\mathfrak q. It is bounded on AA: every element of A0A_0 is integral over R0R_0, hence has value at most one, and the quotient-unit-ball description gives contractivity on AA. It therefore defines an analytic point specializing to q\mathfrak q.

For general A0A_0, let P1,…,PtP_1,\ldots,P_t be the finitely many minimal primes of the Noetherian algebra AA, and put Qi=Pi∩A0Q_i=P_i\cap A_0. Their intersection is the nilradical of A0A_0, since A0↪AA_0\hookrightarrow A. A prime q\mathfrak q of A0A_0 therefore contains some QiQ_i: otherwise a product of elements of the QiQ_i outside q\mathfrak q gives a contradiction. The quotient A0/QiA_0/Q_i is complete by Proposition 3.14.5, topologically finite type, and torsion free as a subring of A/PiA/P_i. It is admissible by Theorem 3.14.4. Apply the domain case to q/Qi\mathfrak q/Q_i, then compose the resulting point with A→A/PiA\to A/P_i. The zero algebra has empty source and target. ◻

Theorem 3.16.16. NAG-001940 (Surjectivity of strict canonical reduction). For every strict KK-affinoid algebra, including a trivially valued ground field, the canonical reduction map M(A)→Spec⁡(A~)\mathcal M(A)\to\operatorname{Spec}(\widetilde A) is surjective.

Proof. Suppose first that KK is nontrivially valued. Choose an admissible model A0⊂AA_0\subset A by Proposition 3.14.7. The algebra A~\widetilde A is finitely generated by Theorem 3.16.3. Lift a finite generating list to A∘A^\circ and adjoin it to A0A_0. Theorem 3.16.8 gives a new admissible model B0B_0 for which B0→A~B_0\to\widetilde A is surjective. A prime p⊂A~\mathfrak p\subset\widetilde A pulls back to a special-fibre prime q\mathfrak q of B0B_0. Choose xx with center q\mathfrak q by Theorem 3.16.15. For any b∈A∘b\in A^\circ, choose a∈B0a\in B_0 with b−a∈A∘∘b-a\in A^{\circ\circ}. Then ∣b(x)∣<1|b(x)|<1 if and only if ∣a(x)∣<1|a(x)|<1, because ∣b(x)−a(x)∣<1|b(x)-a(x)|<1. The reduction kernel of xx is therefore exactly p\mathfrak p.

If KK is trivially valued, a strict Tate algebra is a polynomial algebra with the trivial norm, and its quotient norm is trivial. Thus A∘=AA^\circ=A, and a sequence of powers tends to zero exactly when it is eventually zero. Consequently A∘∘=(0)A^{\circ\circ}=\sqrt{(0)}. For a prime of A~=A/(0)\widetilde A=A/\sqrt{(0)}, give the corresponding prime quotient of AA the trivial multiplicative norm. It is bounded for the quotient norm on AA and has precisely the requested reduction kernel. ◻

Corollary 3.16.17. NAG-001941 (Surjectivity of homogeneous graded reduction). For an arbitrary generalized KK-affinoid algebra AA, every homogeneous prime ideal of gr⁡A\operatorname{gr}A is the reduction kernel of a point of M(A)\mathcal M(A).

Proof. First let AA be strict and let P⊂gr⁡AP\subset\operatorname{gr}A be homogeneous prime. Its degree-one part is a prime P1⊂A~P_1\subset\widetilde A; realize it by xx using Theorem 3.16.16. If the initial class of aa is nonzero of degree r=ρA(a)r=\rho_A(a), Lemma 3.16.6 supplies e>0e>0 and c∈K×c\in K^\times with ∣c∣=re|c|=r^e. The class of ae/ca^e/c has degree one. Primality shows that the initial class of aa belongs to PP exactly when this degree-one class belongs to P1P_1, equivalently ∣a(x)∣<r|a(x)|<r. This proves that the full graded kernel of xx is PP.

In general choose a free polyradius r\boldsymbol r making ArA_{\boldsymbol r} strict. By Theorem 3.13.2, its graded reduction is the Laurent polynomial algebra gr⁡A[t1±1,…,tm±1]\operatorname{gr}A[t_1^{\pm 1},\ldots,t_m^{\pm 1}] with the stated degrees. Extend PP to this Laurent polynomial algebra; the resulting homogeneous ideal is prime because (gr⁡A)/P(\operatorname{gr}A)/P is a domain. The strict case gives a point of M(Ar)\mathcal M(A_{\boldsymbol r}) with that kernel. Its restriction to AA has graded kernel PP, because the Gauss extension is spectrally isometric on AA. ◻

References. The norm-one normalization follows the proof of [Bosch14] (8.3, Proposition 8, pp. 200–202). The additional independent Gauss perturbation and the prescribed-center argument above realize arbitrary prime ideals, as needed for the Berkovich and graded reduction maps; they also avoid a restriction to closed points of the special fibre.

3.17 Formal gluing and the Raynaud equivalence

In this section KK is any complete nontrivially real-valued non-Archimedean field. Its value group need not be discrete. The analytic category is the category of strict analytic GG-spaces: its affinoid charts are strict and its admissible open coverings are the strict affinoid coverings and their composites. A covering is of finite type if each member meets only finitely many others. Quasi-paracompact means admitting such a covering by quasi-compact admissible opens. The same definition, with ordinary open subsets, is used for formal schemes.

Proposition 3.17.1. NAG-001943 (Principal formal opens and flat restriction). Let A0A_0 be admissible and f∈A0f\in A_0. Write A0{f−1}A_0\{f^{-1}\} for the π\pi-adic completion of A0[1/f]A_0[1/f]. It is an admissible algebra, A0{f−1}=A0⟨U⟩/(fU−1),A0{f−1}/πn=(A0/πn)[1/f].A_0\{f^{-1}\}=A_0\langle U\rangle/(fU-1),\qquad A_0\{f^{-1}\}/\pi^n=(A_0/\pi^n)[1/f]. The homomorphism A0→A0{f−1}A_0\to A_0\{f^{-1}\} is flat. On Spf⁡(A0)\operatorname{Spf}(A_0), it is the restriction ring of the principal open D(f)D(f). Its generic fibre is the strict rational domain {x:∣f(x)∣=1}\{x:|f(x)|=1\}.

Proof. The polynomial quotient A0[U]/(fU−1)A_0[U]/(fU-1) is A0[1/f]A_0[1/f] and is π\pi-torsion free. Lemma 3.14.8 gives the completion and reduction formulas. The ideal generated by fU−1fU-1 in A0⟨U⟩A_0\langle U\rangle is closed by Proposition 3.14.5, so its completion has the displayed quotient presentation. It is torsion free by the same completion lemma and hence admissible.

Here is a direct flatness proof. Put R=A0[1/f]R=A_0[1/f] and B0=R^B_0=\widehat R. Let I⊂A0I\subset A_0 be a finitely generated ideal, choose a surjection α:A0s→I\alpha:A_0^s\to I, and let N=ker⁡αN=\ker\alpha. The module NN is finite by Corollary 3.14.6. Proposition 3.14.5 gives e≥0e\geq 0 with I∩πn+eA0⊂πnI(n≥0).I\cap\pi^{n+e}A_0\subset\pi^n I\qquad(n\geq0). Algebraic localization preserves this containment. For v∈ker⁡(B0s→αB0)v\in\ker(B_0^s\xrightarrow{\alpha}B_0) choose wn∈Rsw_n\in R^s with v−wn∈πn+eB0sv-w_n\in\pi^{n+e}B_0^s. The reduction formula shows α(wn)∈IR∩πn+eR⊂πnIR\alpha(w_n)\in I R\cap\pi^{n+e}R\subset\pi^n I R. Thus α(wn)=πnα(un)\alpha(w_n)=\pi^n\alpha(u_n) for some un∈Rsu_n\in R^s, and wn−πnun∈NRw_n-\pi^n u_n\in NR. These vectors approach vv. The image NB0⊂B0sNB_0\subset B_0^s is closed by Proposition 3.14.5, so v∈NB0v\in NB_0. Hence I⊗A0B0→B0I\otimes_{A_0}B_0\to B_0 is injective. The finite-ideal flatness criterion [StacksAlg] (Tag 00HD) proves flatness.

The formal structure sheaf is the inverse limit of the structure sheaves on Spec⁡(A0/πn)\operatorname{Spec}(A_0/\pi^n). Taking sections on D(f)D(f) gives precisely the displayed inverse limit. After inverting π\pi, the quotient presentation is A⟨U⟩/(fU−1)A\langle U\rangle/(fU-1), the domain ∣f∣≥1|f|\geq 1. Since f∈A0f\in A_0 has value at most one everywhere, this is the asserted equality domain. ◻

NAG-001944 · Open at full scope

Proposition 3.17.2. NAG-001944 (The global generic fibre and its faithfulness). The affine generic-fibre construction glues to a functor from admissible formal K∘K^\circ-schemes to strict analytic GG-spaces. For a formal open U⊂X\mathfrak U\subset\mathfrak X, Uη=sp⁡−1(U).\mathfrak U_\eta=\operatorname{sp}^{-1}(\mathfrak U). The specialization map is surjective and functorial. The generic-fibre functor is faithful; it detects equality and covering of formal open subspaces. A quasi-paracompact formal scheme has a quasi-separated quasi-paracompact generic fibre.

Proof. On a principal formal open, the formula follows from Proposition 3.17.1 and f∉sp⁡(x)f\notin\operatorname{sp}(x) if and only if ∣f(x)∣=1|f(x)|=1. Principal opens form a basis, so it holds for every open. Finite intersections and refinements of these principal covers give the same rational restriction maps; hence the affinoid pieces glue, with the analytic structure sheaf of the affinoid GG-site, and the resulting construction is independent of the chosen formal affine cover. Morphisms glue as well because their restrictions are induced by the given continuous algebra homomorphisms. Specialization is compatible with these maps by contraction of the ideals of elements of value less than one. Its surjectivity is local and follows from Theorem 3.16.15.

Two formal morphisms with equal generic fibres have equal underlying point maps, by specialization surjectivity and functoriality. We may therefore compare them on a common formal affine source and target. Their ring homomorphisms agree after inverting π\pi, and hence agree before inverting it, as the target ring is torsion free. This proves faithfulness. The same surjectivity and the displayed formula detect formal open subsets and their covers.

For every c∈K∘∘c\in K^{\circ\circ} some power cnc^n belongs to πK∘\pi K^\circ, while π∈K∘∘\pi\in K^{\circ\circ}; consequently πA0=K∘∘A0\sqrt{\pi A_0}=\sqrt{K^{\circ\circ}A_0}. Thus Spec⁡(A0/πA0)\operatorname{Spec}(A_0/\pi A_0) and Spec⁡(A0/K∘∘A0)\operatorname{Spec}(A_0/K^{\circ\circ}A_0) have the same underlying topological space, so the affine formal space is topologically Noetherian because its special fibre is a finite-type algebra over K~\widetilde K; no Noetherianity of A0/πA0A_0/\pi A_0 is needed. Intersections of quasi-compact formal opens are quasi-compact, and have finite principal-affine covers. Their generic fibres consequently have finite strict affinoid covers. A finite-type quasi-compact formal cover therefore induces a finite-type admissible cover of the same kind on the generic fibre; the intersection observation proves quasi-separatedness. ◻

NAG-001945 · Open at full scope

Lemma 3.17.3. NAG-001945 (Products of open ideals and global blowups). A coherent open ideal on an admissible formal scheme has an admissible formal blowup, obtained by gluing the affine constructions. It is compatible with restriction to formal opens, has the invertible-ideal universal property, and induces an isomorphism on generic fibres. Blowing up a finite product of coherent open ideals makes each factor invertible, so dominates the blowup of each factor. For a formal morphism X→Y\mathfrak X\to\mathfrak Y, every blowup of Y\mathfrak Y is dominated over Y\mathfrak Y by the blowup of its pulled-back ideal on X\mathfrak X.

Proof. The affine universal property in Proposition 3.15.3 identifies the constructions on overlapping formal open subsets: each is the universal space making the same restricted ideal invertible. The unique identifications satisfy the cocycle condition. Gluing locally ringed spaces [StacksAlg] (Tag 01JB) gives the formal scheme, since being an admissible formal affine chart is a local property. Its generic-fibre invariance follows from Theorem 3.15.2 on each chart.

It remains to verify the elementary assertion about products. In a local ring let IJ=(g)IJ=(g) with gg a nonzerodivisor. Write g=∑iaibig=\sum_i a_i b_i with ai∈Ia_i\in I, bi∈Jb_i\in J, and aibi=gcia_i b_i=gc_i. Since ∑ici=1\sum_i c_i=1, some cic_i is a unit. For that index, ai,bia_i,b_i are nonzerodivisors, and for every a∈Ia\in I the relation abi∈(g)ab_i\in(g) gives a∈(ai)a\in(a_i) after cancelling bib_i. Thus I=(ai)I=(a_i), and likewise J=(bi)J=(b_i). Apply this argument successively. An open ideal contains a power of π\pi, so a generator of its invertible extension is a nonzerodivisor. The universal property now gives every stated domination. The pulled-back ideal is again open and finitely generated; the same universal property proves the base-change assertion. ◻

Generic invariance on supplied atlases. Let b:X′→Xb:\mathfrak X'\to\mathfrak X be the admissible blowup of a coherent open ideal. Use actual compatible formal-affine generic atlases, including the affine blowup charts and their principal refinements. Then the chartwise generic-fibre isomorphisms identify bηb_\eta with an isomorphism of these analytic GG-spaces.

Indeed, the formal restriction and universal property in the preceding lemma identify the formal blowup on a formal open with the blowup of the restricted ideal. Refine overlaps of the supplied formal charts by principal formal opens. The generic trace of such an open D(f)D(f) is the actual rational equality domain ∣f∣=1|f|=1, by Proposition 3.17.1 and the equation f∉sp⁡(x)f\notin\operatorname{sp}(x) if and only if ∣f(x)∣=1|f(x)|=1. Specialization contracts along the given chart maps, so the inverse image of that trace under bηb_\eta is the trace of b−1(D(f))b^{-1}(D(f)). On an affine chart, Theorem 3.15.2 identifies the generic blowup charts and transitions with the actual rational domains and intersections of the original affinoid. On the common principal refinements the chart maps are consequently restrictions of the same chart isomorphism for the restricted formal blowup. Their inverses agree there as well. The two families glue as morphisms on the supplied admissible covers by Theorem 2.40.8. The two composites are identities on those covers, so are identities globally. This argument also transports the existing affine atlas to these compatible blowup charts, with the indicated restriction maps. It applies at each stage of an actual sequence of such blowups.

Open traces on supplied atlases. For an actual supplied formal-affine generic atlas, the preceding principal trace calculation gives the generic trace of a formal open by union on its charts. This is the specialization inverse image in the analytic GG-site. Specialization is surjective on these charts by Theorem 3.16.15, hence surjective on the model. Inverse images therefore detect equality, coverage and disjointness of the formal opens and preserve their intersections.

Faithfulness for supplied generic arrows. For two given formal arrows with actual compatible generic charts, equality of their generic maps gives equality of their formal point maps by specialization surjectivity and contraction. At a formal source point, choose a common target affine neighborhood of its image and a source affine neighborhood mapping there under both arrows. The two pullback homomorphisms agree after inverting π\pi. The coefficient ring of the source is π\pi-torsion-free, so the homomorphisms agree there integrally. The arrows agree on this affine cover and hence on the source.

Lemma 3.17.4. NAG-001946 (Extending an open ideal with controlled support). Let X\mathfrak X be a quasi-paracompact admissible formal scheme and U⊂X\mathfrak U\subset\mathfrak X a quasi-compact open. Every coherent open ideal IU\mathcal I_{\mathfrak U} extends to a coherent open ideal I\mathcal I on X\mathfrak X that is the unit ideal on every open disjoint from U\mathfrak U. Consequently a finite composition of admissible blowups on U\mathfrak U extends to such a composition on X\mathfrak X, unchanged over opens disjoint from U\mathfrak U.

Proof. A finite-type quasi-compact open cover of X\mathfrak X contains a finite subfamily covering U\mathfrak U. The union W\mathfrak W of these members and all members meeting them is quasi-compact and contains U‾\overline{\mathfrak U}. Indeed, any member containing a point of this closure meets U\mathfrak U and hence one of the chosen finitely many members. The space W\mathfrak W is Noetherian topologically, by Proposition 3.17.2. In particular W∖U‾\mathfrak W\setminus\overline{\mathfrak U} and U\mathfrak U are quasi-compact opens in it.

Choose NN with πNOU⊂IU\pi^N\mathcal O_{\mathfrak U} \subset\mathcal I_{\mathfrak U}. On the disjoint open union U∪(W∖U‾)\mathfrak U\cup(\mathfrak W\setminus\overline{\mathfrak U}) of the scheme WN=(W,OW/πN)\mathfrak W_N=(\mathfrak W, \mathcal O_{\mathfrak W}/\pi^N), prescribe the finite-type ideal IU/πN\mathcal I_{\mathfrak U}/\pi^N on the first part and the unit ideal on the second. Apply [StacksAlg] (Tag 01PF) to extend this as a finite-type quasi-coherent ideal on WN\mathfrak W_N. Its inverse image in OW\mathcal O_{\mathfrak W} is locally generated by lifts of its finite generators and πN\pi^N. Flatness of principal formal restrictions, proved in Proposition 3.17.1, identifies these ideals on overlaps; coherence follows from Corollary 3.14.6. Glue it with the unit ideal on X∖U‾\mathfrak X\setminus\overline{\mathfrak U}. The two prescriptions agree on their intersection and give the required I\mathcal I.

For a composition of blowups, extend the first center, then the second center on the inverse image of U\mathfrak U, and so on. At every stage the modification is an isomorphism on opens disjoint from that inverse image. Blowups of quasi-paracompact formal schemes remain quasi-paracompact: over each member of a finite-type cover their finitely many affine blowup charts form a finite refinement of the pulled-back cover, and nonintersecting original members remain disjoint. This justifies the induction. ◻

NAG-001947 · Open at full scope

Proposition 3.17.5. NAG-001947 (Formalizing a finite-type analytic covering). Use an actual compatible formal-affine generic atlas on X\mathfrak X and its principal traces on formal opens. On the blowups in the construction use its compatible affine blowup charts as in the supplied-atlas generic-invariance result above. Let X\mathfrak X be quasi-paracompact and admissible, and let (Uα)(U_\alpha) be a finite-type admissible covering of Xη\mathfrak X_\eta by quasi-compact strict analytic opens. There is an admissible blowup X′→X\mathfrak X'\to\mathfrak X and a finite-type covering by quasi-compact formal opens (Uα′)(\mathfrak U_\alpha') with generic fibres exactly UαU_\alpha.

Proof. Use compatible supplied generic atlases throughout. Suppose first X=Spf⁡(A0)\mathfrak X=\operatorname{Spf}(A_0). The cover is finite: quasi-compactness gives a finite subcover, and its members meet only finitely many members of the original finite-type cover. Every UαU_\alpha has a finite strict affinoid cover; Theorem 2.27.5 then refines these affinoid domains by rational domains of M(A)\mathcal M(A). Consider one such domain V={x:∣fj(x)∣≤∣f0(x)∣, 1≤j≤s},(f0,…,fs)A=A.V=\{x:|f_j(x)|\leq|f_0(x)|,\ 1\leq j\leq s\}, \qquad(f_0,\ldots,f_s)A=A. Multiply all fjf_j by the same power of π\pi to put them in A0A_0. Clearing the coefficients in a unit-ideal relation shows that I=(f0,…,fs)A0I=(f_0,\ldots,f_s)A_0 contains a power of π\pi. In its blowup, the formal open chart where f0f_0 generates the transformed ideal has generic fibre VV, by Theorem 3.15.2. Blow up the product of the finitely many ideals obtained for the rational refinements. By the formal product property of Lemma 3.17.3, this dominates all their blowups. By generic invariance on the supplied charts, the inverse images of the selected charts have the prescribed rational generic fibres. Their union for each α\alpha gives Uα′\mathfrak U_\alpha'. These opens cover by detection of formal covers on the supplied atlas.

In general choose a finite-type formal affine cover (Vi)(\mathfrak V_i) and put Vi=(Vi)ηV_i=(\mathfrak V_i)_\eta. Only finitely many UαU_\alpha meet a fixed ViV_i: cover ViV_i by finitely many UαU_\alpha and use finite type. The intersections Uα∩ViU_\alpha\cap V_i are quasi-compact by quasi-separatedness. The affine case gives an open coherent ideal IiI_i on Vi\mathfrak V_i realizing all these intersections. Extend it by Lemma 3.17.4, as the unit ideal away from Vi‾\overline{\mathfrak V_i}. The product of the extended ideals is locally finite: on Vj\mathfrak V_j only those ii with Vi∩Vj≠∅\mathfrak V_i\cap\mathfrak V_j\ne\varnothing contribute. It is a coherent open ideal; blow it up.

Over Vi\mathfrak V_i the formal product property gives domination of the blowup of IiI_i. Use generic invariance on the supplied atlases for these blowups. Pull back the opens representing Uα∩ViU_\alpha\cap V_i and take their union over ii to define Uα′\mathfrak U_\alpha'. Its generic fibre is UαU_\alpha. Only finitely many ViV_i meet the quasi-compact UαU_\alpha, by the same finite-type argument. Each local formal open is quasi-compact, since it lies in a quasi-compact blowup with Noetherian underlying space. Thus Uα′\mathfrak U_\alpha' is quasi-compact. Detection of coverage and disjointness on the supplied atlas proves that these opens give the required finite-type cover. ◻

NAG-001948 · Open at full scope

Theorem 3.17.6. NAG-001948 (Lifting a global analytic morphism). Use actual compatible formal-affine generic atlases on X\mathfrak X and Y\mathfrak Y. On formal opens use their principal traces, and on successive blowups use the compatible affine blowup charts of the supplied-atlas generic-invariance result. Let X,Y\mathfrak X,\mathfrak Y be quasi-paracompact admissible formal schemes. Every analytic morphism f:Xη→Yηf:\mathfrak X_\eta\to\mathfrak Y_\eta lifts to a formal morphism f′:X′→Yf':\mathfrak X'\to\mathfrak Y after a finite composition s:X′→Xs:\mathfrak X'\to\mathfrak X of admissible blowups, with fη′=f∘sηf'_\eta=f\circ s_\eta. Two stages suffice.

Proof. Use finite-type formal affine covers of source and target with their supplied compatible generic atlases. Pull back the target admissible cover and refine it on each affinoid source chart by finitely many strict affinoids. The resulting source cover is of finite type, and each member maps into a target affine generic chart. By cover realization on supplied atlases, one blowup makes it a formal open cover. Refine each quasi-compact formal member by finitely many formal affine opens. The resulting cover (Wi)(\mathfrak W_i) is still of finite type, and each (Wi)η(\mathfrak W_i)_\eta maps into the generic fibre of a formal affine target chart Vi\mathfrak V_i.

Corollary 3.16.9 gives on each Wi\mathfrak W_i a blowup and a lifting to Vi\mathfrak V_i. Extend its center by Lemma 3.17.4, and blow up the locally finite product of the extended centers. This second stage dominates all the local blowups by the formal product property of Lemma 3.17.3. By generic invariance on the supplied atlases, the resulting morphisms to Y\mathfrak Y have generic maps that are restrictions of ff under the blowup identifications. On each overlap they agree by faithfulness for the supplied generic arrows. They glue to f′f'. ◻

NAG-001949 · Open at full scope

Lemma 3.17.7. NAG-001949 (A common modification of isomorphic generic fibres). Use actual compatible formal-affine generic atlases on the two models, with their principal traces on formal opens and the compatible affine blowup charts along the successive modifications. Let X,Y\mathfrak X,\mathfrak Y be quasi-paracompact admissible formal schemes with a specified generic-fibre isomorphism. There is a quasi-paracompact admissible formal scheme Z\mathfrak Z and morphisms to both, each a finite composition of admissible blowups, inducing that isomorphism.

Proof. With the supplied compatible generic atlases, lift the isomorphism and its inverse by lifting on supplied atlases. Obtain modifications s:X1→Xs:\mathfrak X_1\to\mathfrak X, t:Y1→Yt:\mathfrak Y_1\to\mathfrak Y and maps u:X1→Yu:\mathfrak X_1\to\mathfrak Y, v:Y1→Xv:\mathfrak Y_1\to\mathfrak X. Pull back the successive centers of tt along uu and blow them up in order. This gives p:X2→X1p:\mathfrak X_2\to\mathfrak X_1 and a:X2→Y1a:\mathfrak X_2\to\mathfrak Y_1 with ta=upta=up. Likewise pull back the centers of ss along vv, obtaining q:Y2→Y1q:\mathfrak Y_2\to\mathfrak Y_1 and b:Y2→X1b:\mathfrak Y_2\to\mathfrak X_1 with sb=vqsb=vq. All s,t,p,qs,t,p,q are finite composites of admissible blowups.

Identify the generic fibres at each stage by generic invariance on the supplied atlases. The respective composites have equal generic maps, so faithfulness for the supplied generic arrows gives va=spva=sp and ub=tqub=tq. The first equality shows that the successive centers of ss pulled back along vava are already invertible on X2\mathfrak X_2, since the map factors through ss. The universal properties defining qq therefore factor aa through α:X2→Y2\alpha:\mathfrak X_2\to\mathfrak Y_2. Similarly bb factors through β:Y2→X2\beta:\mathfrak Y_2\to\mathfrak X_2. The pullbacks of invertible ideals are invertible here: each contains a power of π\pi, a nonzerodivisor on the admissible source. The generic maps of α,β\alpha,\beta are inverse, so the same faithfulness gives αβ=1\alpha\beta=1 and βα=1\beta\alpha=1. Identify the spaces by this isomorphism. Their maps sp,tqsp,tq are the required modifications. ◻

Generic fibres of quasi-paracompact formal schemes. Retain the standing complete nontrivially real-valued field KK, its valuation ring OO, and π∈O\pi\in O with 0<∣π∣<10<|\pi|<1. Let X\mathfrak X be an admissible formal OO-scheme that admits a finite-type ordinary open cover by quasi-compact opens. Here finite type means that each member meets only finitely many others. The affine generic fibres glue canonically to a quasi-separated quasi-paracompact strict analytic GG-space Xη\mathfrak X_\eta, functorially for formal arrows between such schemes. Specialization is natural and surjective. For every formal open U⊆X\mathfrak U\subseteq\mathfrak X, the generic fibre of U\mathfrak U identifies with the analytic domain sp⁡X−1(U)\operatorname{sp}_{\mathfrak X}^{-1}(\mathfrak U) with its restricted strict charts and bounded restriction maps.

Proof. Refine each member of the finite-type ordinary open cover by finitely many formal affines Pi=Spf⁡(Ai)P_i=\operatorname{Spf}(A_i). The resulting family still has finite incidence. Each PiP_i is topologically Noetherian by the formal-topology clause of NAG-001944. In particular, every open subset of PiP_i has a finite principal refinement. We use the strict generic affinoid Xi=M(Ai[1/π])X_i=\mathcal M(A_i[1/\pi]) from Proposition 3.14.7, with the affine specialization map sis_i.

We first produce the actual overlaps and their transitions. For a point of Pi∩PjP_i\cap P_j, take a principal open Di(f)D_i(f) contained in that overlap and containing the point. Inside it take a principal open Dj(g)D_j(g) of PjP_j containing the point. On Di(f)D_i(f), the reduction modulo π\pi of the restriction of gg is a/fna/f^n for some a∈Ai/πAia\in A_i/\pi A_i, by Proposition 3.17.1. Consequently Dj(g)=Di(fa^)D_j(g)=D_i(f\widehat a), where a^∈Ai\widehat a\in A_i lifts aa. This is an equality of open subspaces of the same formal scheme with its restricted structure sheaf. Iterating the argument works on a finite intersection of formal affines. A principal open of a shared principal piece is principal in each ambient formal affine, by the same localization formula. Since each formal overlap is quasi-compact, finitely many common principal pieces cover it.

Set Xij=si−1(Pi∩Pj)X_{ij}=s_i^{-1}(P_i\cap P_j) in XiX_i. The traces of the common principal pieces are the actual strict rational domains given by Proposition 3.17.1. They cover XijX_{ij} by the affine equality si−1(D(f))={∣f∣=1}s_i^{-1}(D(f))=\{|f|=1\}. Thus XijX_{ij} is a compact closed analytic domain of the Hausdorff affinoid XiX_i, by Proposition 2.39.3. The finite rational traces are an admissible cover by Lemma 2.39.2. A shared affine formal open has the same coefficient ring viewed from either side. Its two generic identifications are bounded with bounded inverses by Proposition 3.14.7 and the principal restriction proposition. The local identifications agree on the common principal refinements because both restrict the same formal structure sheaf maps. Theorem 2.40.8 glues them and their inverses to vij:Xij→Xjiv_{ij}:X_{ij}\to X_{ji}.

The maps vijv_{ij} preserve the formal centre by affine contraction. In particular vij(Xij∩Xik)=Xji∩Xjkv_{ij}(X_{ij}\cap X_{ik})=X_{ji}\cap X_{jk}. On the common principal pieces of the actual triple overlap, vjk∘vij=vikv_{jk}\circ v_{ij}=v_{ik} because all three maps are induced by the same restricted formal rings. This equality holds on the entire triple overlap by uniqueness of morphism gluing. If Pi∩PjP_i\cap P_j is empty, its trace is empty. The spaces XiX_i, their actual closed overlap domains, and the transitions therefore satisfy the second case of Theorem 2.40.10. Its gluing is Hausdorff, and the strict chart images form a locally finite admissible cover with finite incidence. This gives quasi-paracompactness. Intersections of quasi-compact analytic domains are analytic domains by Proposition 2.40.5 and are quasi-compact in this Hausdorff space. This gives quasi-separatedness.

For a second finite-type formal affine cover, its members and the PiP_i together have finite incidence. Indeed, a fixed quasi-compact formal affine is covered by finitely many members of the other cover. Every member of that other cover meeting the affine must meet one of these finitely many covering members. Construct the gluing for the union of the two covers. Both old chart families cover it, since a formal centre belongs to a chart in either family. Each of these subcovers is still locally finite and closed and therefore admissible. The bounded chart identities and their inverses glue by Theorem 2.40.8, giving the canonical comparisons. For three covers the comparisons compose by their identities on the common charts.

For any formal open U\mathfrak U, the opens U∩Pi\mathfrak U\cap P_i are quasi-compact and retain finite incidence. Refine each by finitely many principal pieces. Their generic affinoids form the actual restricted charts on the specialization inverse image. This identifies the generic fibre of U\mathfrak U with that domain, by the same gluing uniqueness. For a formal arrow, take a finite principal refinement on each source affine subordinate to the inverse images of the formal target affines. The bounded affine generic maps agree on overlaps, as can be checked after the actual principal restrictions of the source and target. They glue by Theorem 2.40.8. Identity and composition hold on the same refinements, hence globally. The maps sis_i agree by contraction and glue as maps of sets. The contraction formula also proves naturality for the induced arrows. Each sis_i is surjective by Theorem 3.16.15, so specialization is surjective. The empty cover gives the empty generic fibre. ◻

Theorem 3.17.8. NAG-001950 (Existence of quasi-paracompact formal models). Every quasi-separated quasi-paracompact strict analytic GG-space over KK has a quasi-paracompact admissible formal K∘K^\circ-model.

Proof companion for NAG-001950. Use the standing complete nontrivially real-valued field and the original quasi-separated quasi-paracompact strict analytic GG-space XX. The following verifies the gluing and stabilization steps in the original proof.

First refine each quasi-compact admissible member of the original finite-type cover by finitely many strict affinoid domains. The local strict affinoids form a quasi-net on the member, and compactness gives a finite point cover by Lemma 2.39.2. Each retained domain is closed on the member. Indeed its intersection with every affinoid test chart is quasi-compact by quasi-separatedness and hence closed in that Hausdorff test chart. Lemma 2.40.2 detects closedness on the quasi-net of test charts. The finite closed cover is thus admissible. Transitivity gives a finite-type strict affinoid cover of XX. It is locally finite: a neighborhood contained in finitely many members can meet only members meeting at least one of that finite list. The cover has actual closed overlap domains with their inherited identifications. The closed-overlap case of Theorem 2.40.10, together with the quasi-net topology and the given chart maps, identifies its gluing with XX and makes XX Hausdorff.

Consequently a finite union of the compact domains used below is again an actual compact analytic domain. The members are closed in the union, so their finite cover there is a quasi-net. Refining by their affinoid charts and using transitivity gives a quasi-net of actual affinoid domains of XX on the union, which is the defining domain condition. The same test on each compact domain certifies its finite closed refinements as admissible. All the overlaps below carry these induced domain structures.

For the two-piece step, let V1,V2V_1,V_2 be compact domains already equipped with quasi-paracompact formal models with the specified generic identifications. Their intersection WW is a compact strict analytic domain, by the domain intersection rule and Hausdorffness. If it is empty, take the disjoint union of the models. Otherwise the cover (W,Vi)(W,V_i) of ViV_i is admissible because it contains the entire domain. Apply the supplied-atlas scope of NAG-001947 on the two actual model atlases. The indicated blowups make WW the generic fibre of one quasi-compact formal open in each modified model. The finite-incidence generic-fibre construction supplies their compatible generic atlases and the restricted identifications, so the supplied-atlas scope of NAG-001949 applies to the identity on WW. It gives a common formal open model after the two actual finite sequences of modifications. Extend each entire sequence to its ambient model by Lemma 3.17.4. The restrictions to the successive inverse images of the original opens are the given sequences. The common modified models of WW are therefore actual formal opens in the resulting ambient models.

Glue the two formal schemes along these ordinary opens. The underlying space is the topological gluing and the structure sheaf consists of sections with equal restrictions on the identified opens. Each point therefore retains an admissible formal affine neighborhood. The maps on the actual principal generic restrictions reconstruct precisely the given analytic identification of WW. The finite-incidence construction and uniqueness of analytic gluing give the original analytic union V1∪V2V_1\cup V_2 as the generic fibre. Starting with Proposition 3.14.7 on strict affinoids, induction over a finite admissible affinoid cover gives the compact case.

For the full space, take the locally finite intersection graph of the given finite-type compact cover. The unions for different components are disjoint and open-and-closed: each such union and its complement are locally finite unions of closed subsets. Work in one component and let UnU_n be the finite union of members at graph distance nn from a chosen member. Each layer is a compact analytic domain. Nonneighboring layers are disjoint, and the family of layers is a locally finite closed admissible cover. At stage nn keep a model of Vn=⋃i≤nUiV_n=\bigcup_{i\le n}U_i with quasi-compact formal opens representing each layer. The intersection of Un+1U_{n+1} with VnV_n is exactly its intersection with UnU_n. Apply the two-piece construction to the terminal formal open for UnU_n and a compact model of Un+1U_{n+1}. Extend the sequence on the terminal side to the current ambient model via NAG-001946 and perform the prescribed ordinary formal open gluing. Take the inverse images of the earlier opens and the new layer’s open as the new covering. The earlier opens remain quasi-compact since the inverse image of a quasi-compact formal open under each blowup has finite affine blowup covers over its finite affine charts.

If i<n−1i<n-1, the open for UiU_i is disjoint from the terminal formal open. This follows from the disjointness detection on the actual generic atlases. Every extended center is the unit ideal there, so the sequence is the canonical identity over that open. For each fixed ii the open consequently stabilizes after finitely many stages. The intersections and transition identifications of stabilized opens are read in one stage after both have stabilized. For a triple take a stage after all three have stabilized. The identifications are compatible under passage to later stages, and the triple cocycle holds there. Gluing the stabilized ordinary formal opens gives an admissible formal scheme. Their finite incidence follows from reflected disjointness for nonneighboring layers, and they are quasi-compact, so the result is quasi-paracompact. Its generic charts and restrictions are those of XX on the actual closed admissible layer cover. Take disjoint unions over the components. The empty space has the empty formal model.

Proof. An affinoid space has a model by Proposition 3.14.7. Suppose two quasi-compact analytic opens U1,U2U_1,U_2 have models U1,U2\mathfrak U_1,\mathfrak U_2. Their intersection WW is quasi-compact by quasi-separatedness. Apply cover realization on supplied atlases to the cover (W,Ui)(W,U_i) of UiU_i: after a blowup Ui′→Ui\mathfrak U_i'\to\mathfrak U_i, a quasi-compact formal open Wi′⊂Ui′\mathfrak W_i'\subset\mathfrak U_i' represents WW. By common modification on supplied atlases, the two models of WW have a common modification W′′\mathfrak W''. Extend its two sequences of centers to Ui′\mathfrak U_i' by Lemma 3.17.4. The resulting models Ui′′\mathfrak U_i'' have the same formal open W′′\mathfrak W'' and glue along it. Their generic fibres glue along exactly WW, giving U1∪U2U_1\cup U_2. Induction on a finite affinoid cover proves existence for every quasi-compact space.

In general choose a finite-type cover by quasi-compact analytic opens. Its intersection graph is locally finite. Each graph component is countable, and distinct components give disjoint admissible open pieces, which can be treated separately. In one component fix a vertex, and let UnU_n be the union of the finitely many covering members at graph distance nn from it. Each UnU_n is quasi-compact, the UnU_n cover that piece, and Un∩Um=∅U_n\cap U_m=\varnothing whenever ∣n−m∣>1|n-m|>1.

Inductively construct a model Vn\mathfrak V_n of Vn=U0∪⋯∪UnV_n=U_0\cup\cdots\cup U_n, containing formal opens representing each UiU_i, i≤ni\leq n. To add Un+1U_{n+1}, apply the two-piece gluing construction above to the current formal open representing UnU_n and a model of Un+1U_{n+1}. Their analytic intersection equals Vn∩Un+1V_n\cap U_{n+1}. Extend the modifications on the UnU_n side to Vn\mathfrak V_n, with support in the closure of that formal open, by Lemma 3.17.4. This leaves each formal open representing UiU_i, i<n−1i<n-1, unchanged, because it is disjoint from the UnU_n open. Glue in the modified model of Un+1U_{n+1}. The preimages of the existing formal opens and the new open give the required cover of Vn+1\mathfrak V_{n+1}.

For fixed ii, its formal open is unchanged after finitely many stages. Write Ui∞\mathfrak U_i^\infty for this stabilized open. Overlap identifications are read at any sufficiently late common stage; stabilization makes them independent of the stage, and a common stage for three indices proves the cocycle identity. Glue these opens. Each Ui∞\mathfrak U_i^\infty is quasi-compact, having undergone only finitely many quasi-compact blowups, and meets only its two neighbors, since generic fibres detect disjointness. The glued model is therefore quasi-paracompact and has the desired generic fibre. Taking disjoint unions over graph components finishes the proof. ◻

Theorem 3.17.9. NAG-001951 (Raynaud equivalence). Let F\mathcal F be the category of quasi-paracompact admissible formal K∘K^\circ-schemes, and let S\mathcal S be the class of admissible blowups. The generic-fibre functor induces an equivalence F[S−1]≃{quasi-separated quasi-paracompactstrict analytic G-spaces over K}.\mathcal F[\mathcal S^{-1}] \simeq \left\{\begin{gathered} \text{quasi-separated quasi-paracompact}\\ \text{strict analytic \(G\)-spaces over }K \end{gathered}\right\}.

Proof companion for NAG-001951. Retain exactly the category F\mathcal F of quasi-paracompact admissible formal K∘K^\circ-schemes in the standing field setting, and the category A\mathcal A of quasi-separated quasi-paracompact strict analytic GG-spaces. The finite-incidence construction gives the generic-fibre functor on every object and arrow of F\mathcal F. It supplies the actual atlases needed by the supplied-atlas NAG-001945 scope to invert each admissible blowup. Inverting single blowups is equivalent to inverting their finite composites.

For a roof X←sP→fY\mathfrak X\xleftarrow{s}\mathfrak P\xrightarrow{f}\mathfrak Y with ss a finite composite, the generic arrow is fηsη−1f_\eta s_\eta^{-1}. To compose with a roof Y←tQ→gZ\mathfrak Y\xleftarrow{t}\mathfrak Q\xrightarrow{g}\mathfrak Z, pull back the centers of the successive stages of tt along the current map from the current blowup of P\mathfrak P. At each stage the formal-only NAG-001945 clause and the affine universal property give the blowup and the arrow to the corresponding stage of Q\mathfrak Q. Thus there are an actual finite composite p:P′→Pp:\mathfrak P'\to\mathfrak P and an arrow a:P′→Qa:\mathfrak P'\to\mathfrak Q with ta=fpta=fp. The composed roof has arrows spsp and gaga. Its generic arrow is the composite, since the generic denominator maps are isomorphisms.

For two roofs with the same generic arrow, identify their modified sources over the common generic source. Apply the supplied-atlas NAG-001949 scope to that specified identification. It gives a common formal source with finite composites to the two modified sources. On this refinement the arrows to X\mathfrak X agree by supplied-arrow faithfulness, since their generic arrows agree. The arrows to Y\mathfrak Y also agree by the same faithfulness and the premise on the two roofs. Conversely a common refinement with both arrows equal gives equal generic arrows. The supplied-atlas NAG-001948 scope represents every generic arrow by a roof. Its proof uses the exact supplied-atlas NAG-001947 scope for the first covering blowup. These applications all receive the actual atlases from the finite-incidence construction.

Thus the roof classes are in bijection with the analytic arrows between the corresponding generic fibres. Composition calculated above is independent of choices and associative by this bijection, and the identity roof gives the identity. For a functor HH inverting the admissible blowups, assign the roof the arrow H(f)H(s)−1H(f)H(s)^{-1}. The equality on common refinements makes this assignment well defined. The equality ta=fpta=fp makes the assignment compatible with composition. Each roof is the composite of the image of its forward arrow with the inverse of the image of its denominator. The extension of HH is therefore unique, and the same factorization proves the corresponding assertion on natural transformations. This is the localization universal property. The generic arrow bijection proves full faithfulness, and the exact NAG-001950 model construction proves essential surjectivity.

Proof. Every blowup is sent to an isomorphism by generic invariance on supplied atlases. Replacing S\mathcal S by its finite composites does not change the localization: a functor inverts one class if and only if it inverts the other. Use roofs X←sX′→fY\mathfrak X\xleftarrow{s}\mathfrak X'\xrightarrow{f}\mathfrak Y with ss such a finite composite. They can be composed by successively blowing up the pullbacks of the second roof’s denominator centers on X′\mathfrak X'. The universal properties give the middle map, and the new source map remains a finite composite.

Two choices have a common refinement. Their modified sources have identified generic fibres, so common modification on supplied atlases gives modifications from a common source to both. The maps to the original source agree by generic-fibre faithfulness. If the roofs have the same analytic map, their target maps also agree on this refinement. Conversely, equality on a common refinement implies equality of analytic maps. lifting on supplied atlases ensures that every analytic map is represented by a roof. Thus roofs modulo common refinement are in bijection with the analytic morphisms between generic fibres. Taking common refinements of the two iterated pullback constructions proves associativity; the identity roof is the identity morphism.

This roof category has the localization universal property. If H:F→DH:\mathcal F\to\mathcal D inverts blowups, send the roof to H(f)H(s)−1H(f)H(s)^{-1}. Common refinements leave this expression unchanged, and the pullback construction proves compatibility with composition. This extension is unique, since every roof is the composite of its forward map and its denominator’s inverse. Hence the category is F[S−1]\mathcal F[\mathcal S^{-1}]. The preceding bijection of morphisms proves full faithfulness, and Theorem 3.17.8 proves essential surjectivity. ◻

References. The gluing route is reconstructed from [Bosch14] (8.4, Theorem 3 and Lemmas 4–6, pp. 204–213). All rank-one valuation rings are retained. Its geometric inputs are the proved rational refinement, affine blowup charts, specialization, and ideal extension. Finite composites suffice for the localization, so no replacement of a composite by a single blowup is used.

3.18 General graded finiteness and inner homomorphisms

Proposition 3.18.1. NAG-001953 (Localization at the spectral maximum). Let AA be affinoid, let f∈Af\in A, and suppose r=ρA(f)>0r=\rho_A(f)>0. Set B=A{rf−1}B=A\{rf^{-1}\} and write fˉ\bar f for the initial form of ff. For a∈Aa\in A, ρB(a)=lim⁡n⟶∞r−nρA(fna).\rho_B(a)=\lim_{n\longrightarrow\infty}r^{-n}\rho_A(f^na). If this value is positive, the sequence is eventually constant. The canonical homomorphism is an isomorphism (A~gr)fˉ→ ∼ B~gr.(\widetilde A^{\rm gr})_{\bar f} \xrightarrow{\ \sim\ }\widetilde B^{\rm gr}.

Proof. Let Γ\Gamma be the finite Shilov boundary from Corollary 3.12.5, and set Γf={x∈Γ:∣f(x)∣=r}\Gamma_f=\{x\in\Gamma:|f(x)|=r\}. This set is nonempty. The rational point theorem identifies M(B)\mathcal M(B) with {∣f∣=r}⊂M(A)\{|f|=r\}\subset\mathcal M(A). For every nn its points satisfy ∣a(x)∣≤r−nρA(fna)=max⁡y∈Γ(∣f(y)∣r)n∣a(y)∣.|a(x)|\leq r^{-n}\rho_A(f^na) =\max_{y\in\Gamma}\left(\frac{|f(y)|}{r}\right)^n|a(y)|. The finite maximum on the right decreases to m=max⁡Γf∣a∣m=\max_{\Gamma_f}|a|. Conversely Γf\Gamma_f consists of points of BB, so ρB(a)≥m\rho_B(a)\geq m. This proves the formula. If m>0m>0, every term with ∣f(y)∣<r|f(y)|<r is smaller than mm for all sufficiently large nn, giving eventual equality.

On BB one has ∣f∣=r|f|=r, hence ρB(fjb)=rjρB(b)\rho_B(f^jb)=r^j\rho_B(b) for all integers jj. If a homogeneous class represented by aa dies in B~gr\widetilde B^{\rm gr}, the limit formula gives ρA(fna)<rnρA(a)\rho_A(f^na)<r^n\rho_A(a) for some nn. Thus it is already killed by a power of fˉ\bar f, proving injectivity after localization. Every nonzero homogeneous class in B~gr\widetilde B^{\rm gr} has a representative of the form a/fja/f^j: the polynomial algebra A[f−1]A[f^{-1}] is dense in BB, so approximate a representative with error strictly below its positive spectral size. Increase jj if necessary. Eventual equality gives ρA(fna)=rnρB(a)\rho_A(f^na)=r^n\rho_B(a), and the initial class of fna/fj+nf^na/f^{j+n} then comes from the displayed localization. This proves surjectivity. ◻

Lemma 3.18.2. NAG-001954 (Orthogonal coordinates on one Gauss field). Let K(t)rK(t)_r be K(t)K(t) with the Gauss absolute value of radius r>0r>0, and let K(r)K(r) be its completion. It has an orthogonal KK-basis (ei)(e_i), in the Banach sense, containing 11. The basis remains orthogonal for Gauss evaluation over every complete extension of KK. The constant-coordinate projection is a bounded KK-linear retraction K(r)→KK(r)\to K.

Proof. First take r=1r=1. For each monic irreducible polynomial qˉ∈K~[t]\bar q\in\widetilde K[t], choose a monic lift q∈K∘[t]q\in K^\circ[t]. Use the family 1,t,t2,…,tjq(t)m(m≥1,0≤j<deg⁡q).1,t,t^2,\ldots,\qquad \frac{t^j}{q(t)^m}\quad (m\geq1,\quad0\leq j<\deg q). All its members have norm one. Their reductions are the partial-fraction basis of K~(t)\widetilde K(t) over K~\widetilde K. For a finite linear combination, divide by a coefficient of largest norm. A nonzero reduced combination remains, so the norm is exactly the largest coefficient norm. The same argument works after any field extension: these rational functions remain linearly independent over the extended residue field, as is seen by their distinct principal parts and their polynomial parts.

Their span is dense in K(t)1K(t)_1. To check this, normalize the denominator of a rational function to have Gauss norm one. Factor its nonzero reduction into irreducible factors and lift the product, including its leading scalar, to a polynomial DD. The denominator differs from DD by a polynomial of norm strictly less than one. Its inverse is therefore a convergent geometric series with terms in the algebra generated by tt and the chosen q−1q^{-1}’s. Ordinary polynomial division and partial fractions put each term in the displayed span. Completion gives the asserted orthonormal basis.

If rr is free over ∣K×∣\sqrt{|K^\times|}, use the Laurent basis (tn)n∈Z(t^n)_{n\in\mathbf Z} from Lemma 1.16.1; every nonzero polynomial has one dominant monomial, so its inverse is a convergent Laurent geometric series. Otherwise choose the least d>0d>0 with rd=∣c∣r^d=|c|, c∈K×c\in K^\times, and put s=td/cs=t^d/c. Apply the radius-one construction to K(s)K(s) and multiply its basis by 1,t,…,td−11,t,\ldots,t^{d-1}. The decomposition into powers of tt modulo dd shows both that this spans K(t)K(t) over K(s)K(s) and that its weighted Gauss norm is the maximum of the coordinate norms. After changing the coefficient field, the same assertion follows by clearing a common polynomial denominator in ss: the remaining polynomials in tt have disjoint exponent classes modulo dd. This also proves the claimed extension-stability. In every case the coefficient of the basis vector 11 gives the stated retraction. ◻

Theorem 3.18.3. NAG-001955 (Graded extension by arbitrary Gauss fields). Let AA be affinoid and let LL be an iterated Gauss-field extension of KK, with finitely many arbitrary positive radii. There are canonical isomorphisms A~gr⊗K~grL~gr≃A⊗^KL~gr.\widetilde A^{\rm gr}\otimes_{\widetilde K^{\rm gr}} \widetilde L^{\rm gr} \simeq\widetilde{A\widehat\otimes_KL}^{\rm gr}. If (ei)(e_i) is the resulting orthogonal basis of L/KL/K, then for every coefficient expansion, ρAL(∑iaiei)=max⁡iρA(ai)∣ei∣.\rho_{A_L}\left(\sum_i a_i e_i\right) =\max_i\rho_A(a_i)|e_i|.

Proof. At each stage of the iterated Gauss extension, the scalar algebra is the actual affinoid scalar extension of Proposition 2.33.5.

Let KK be complete, r>0r>0, and L=K(t)^rL=\widehat{K(t)}_r. Use the orthogonal Banach basis (ei)i∈I(e_i)_{i\in I} from Lemma 3.18.2 and put wi=∣ei∣w_i=|e_i|. Here c0(I,w;A)c_0(I,w;A) consists of all families for which only finitely many ∥ai∥wi\|a_i\|w_i exceed any fixed positive threshold. Let AA carry its given admissible Banach algebra norm. At this step the scalar algebra is the actual separated completed projective product AL=A⊗^KLA_L=A\widehat\otimes_K L.

The coefficient functionals of the basis are bounded maps ℓi:L→K\ell_i:L\to K of norm at most wi−1w_i^{-1}. On A⊗KLA\otimes_K L they give wi∥(1⊗ℓi)(z)∥≤∥z∥π.w_i\|(1\otimes\ell_i)(z)\|\leq \|z\|_\pi . Indeed for every finite representation z=∑jaj⊗λjz=\sum_j a_j\otimes\lambda_j the ultrametric inequality bounds the left side by max⁡j∥aj∥∣λj∣\max_j\|a_j\||\lambda_j|. Take the infimum over all such representations. For a finite coefficient family, the representation ∑iai⊗ei\sum_i a_i\otimes e_i gives the reverse bound for ∥z∥π\|z\|_\pi, hence equality with the weighted maximum. Finite basis expansions approximate all scalar factors of every finite tensor in the projective seminorm. Thus finite coefficient families are dense in the separated completion. They are also dense in the complete weighted c0c_0-space, so that AL≃c0(I,w;A)A_L \simeq c_0(I,w;A) isometrically as Banach spaces, via the actual coefficient maps. This argument uses finite truncations for each family and places no countability condition on II.

For x∈MK(A)x\in\mathcal M_K(A) set F=H(x)F=\mathcal H(x) and Lx=F(t)^rL_x=\widehat{F(t)}_r. The Gauss absolute value is multiplicative on polynomials by Proposition 1.2.2, hence extends by ratios to the rational function field. The inclusion K(t)↪F(t)K(t)\hookrightarrow F(t) is isometric for these Gauss norms and extends to L↪LxL\hookrightarrow L_x. The map a⊗l↦a(x)la\otimes l\mapsto a(x)l is contractive for each finite tensor representation, by Lemma 1.6.1 and the isometric scalar inclusion. It therefore extends to a bounded LL-algebra map AL→LxA_L\to L_x and gives a bounded point xrx_r of ALA_L. By the extension-stability in Lemma 3.18.2, the value of a finite coefficient family at this point is max⁡i∣ai(x)∣wi\max_i|a_i(x)|w_i. The same equality holds for every weighted c0c_0-family by truncation and contractivity.

Every bounded point of ALA_L restricts along the constant inclusion to a bounded point of AA and has scalar values ∣ei∣=wi|e_i|=w_i. The ultrametric inequality and truncation bound its evaluation above by max⁡iρA(ai)wi\max_i \rho_A(a_i)w_i. Conversely, for each ii use the points xrx_r as xx ranges through MK(A)\mathcal M_K(A) to bound the upper spectral seminorm from below by ρA(ai)wi\rho_A(a_i)w_i. Consequently ρAL(∑iaiei)=max⁡iρA(ai)wi.\rho_{A_L}\Bigl(\sum_i a_i e_i\Bigr)= \max_i \rho_A(a_i)w_i . Here ρA(ai)≤∥ai∥\rho_A(a_i)\leq\|a_i\|, so a positive supremum is a maximum. If all weights of the evaluated coefficients vanish the same inequalities give zero. The zero algebra gives both sides zero with the usual zero convention.

Assume the one-step coefficient and spectral formulas for the chosen positive Gauss radius, including the actual coefficient expansion on AL=A⊗^KLA_L=A\widehat\otimes_K L. Write LL in the orthogonal Banach basis (ei)(e_i) of Lemma 3.18.2 and put wi=∣ei∣w_i=|e_i|. For a submultiplicative spectral seminorm ρ\rho, use the positive degree qq quotient {ρ≤q}/{ρ<q}\{\rho\leq q\}/\{\rho<q\}.

The bounded structural maps from AA and LL into ALA_L are filtered for these seminorms. For AA this follows by restricting points and for LL by their scalar absolute values. The same inequalities apply to the zero algebra. Hence multiplication of homogeneous representatives induces the canonical graded algebra homomorphism A~gr⊗K~grL~gr⟶AL~gr.\widetilde A^{\rm gr}\otimes_{\widetilde K^{\rm gr}} \widetilde L^{\rm gr}\longrightarrow\widetilde{A_L}^{\rm gr}. For z=∑iaieiz=\sum_i a_i e_i with ρAL(z)≤q\rho_{A_L}(z)\leq q, only finitely many indices satisfy ρA(ai)wi=q>0\rho_A(a_i)w_i=q>0. Indeed ρA(ai)≤∥ai∥\rho_A(a_i)\leq\|a_i\|, and the coefficient weights tend to zero in the weighted c0c_0 sense. Among the remaining indices the supremum of the spectral weights is strictly below qq: all but finitely many are smaller than q/2q/2, and every retained nonleading weight is strictly below qq. The spectral formula therefore identifies the degree qq class of zz with the finite sum of the classes in⁡q/wi(ai)in⁡wi(ei)\operatorname{in}_{q/w_i}(a_i)\operatorname{in}_{w_i}(e_i) at equality indices. For a finite sum of homogeneous coefficients in degree qq, choose representatives of size at most q/wiq/w_i. If some coefficient class is nonzero, the formula gives spectral size exactly qq, so its image class is nonzero.

Applying the same argument to LL with its orthogonal coefficient norm shows that the in⁡wi(ei)\operatorname{in}_{w_i}(e_i) freely generate L~gr\widetilde L^{\rm gr} over K~gr\widetilde K^{\rm gr} as a graded module. The preceding surjectivity and injectivity in each degree prove that the displayed map is an isomorphism. Its multiplication is the one induced by multiplying actual lifts.

For iteration, let K⊂L⊂L′K\subset L\subset L' be successive stages, and let (fj)(f_j) be the next orthogonal basis, of weights vj=∣fj∣v_j=|f_j|. In the actual scalar algebra write z=∑jbjfjz=\sum_j b_j f_j, bj=∑iaijeib_j=\sum_i a_{ij}e_i. Each column is a weighted c0c_0-family and the numbers (sup⁡i∥aij∥wi)vj(\sup_i\|a_{ij}\|w_i)v_j tend to zero. This is equivalent to the single weighted c0c_0 condition on (∥aij∥wivj)i,j(\|a_{ij}\|w_iv_j)_{i,j}: above any positive threshold only finitely many columns can occur and each contributes finitely many entries. Conversely the single condition implies both column conditions by testing at half the given threshold. The same norm calculation with A=KA=K gives the product orthogonal basis (eifj)(e_i f_j) of L′/KL'/K.

The scalar-tower comparison of Corollary 2.33.9 identifies (AL)L′(A_L)_{L'} with AL′A_{L'}, compatibly with structural maps. On finite coefficient sums it sends the two-stage expressions to ∑i,jaij(eifj)\sum_{i,j}a_{ij}(e_i f_j). The weighted c0c_0 description and continuity give the same equality for limits. The two spectral formulas give ρAL′(z)=max⁡jvjρAL(bj)=max⁡i,jvjwi ρA(aij).\rho_{A_{L'}}(z) =\max_j v_j\rho_{A_L}(b_j) =\max_{i,j}v_jw_i\,\rho_A(a_{ij}). The graded maps compose by the usual algebraic tensor associator, since each is induced by multiplication of representatives. Iteration proves the claim for any finite tower, including the empty tower, whose comparison is the identity. ◻

Lemma 3.18.4. NAG-001956 (Graded scalar descent and degree-one reduction). Every graded module over a graded field has a homogeneous basis. Scalar extension by a nonzero graded algebra over that field is faithful and exact. Finite generation of an algebra or a module descends under this scalar extension. If a graded kk-algebra CC is supported in the degree group of kk, multiplication induces C1⊗k1k≃C.C_1\otimes_{k_1}k\simeq C. For algebras with this support condition, a graded map is finite precisely when its degree-one map is finite.

Proof. Choose a maximal linearly independent family of homogeneous elements. If its span is not the whole module, a nonzero homogeneous class in the quotient generates a copy of a shifted kk; lifting it enlarges the family, a contradiction. This proves freeness and hence faithful exactness. If finitely many tensors generate a scalar extension, take the finitely many original coefficients appearing in them. Their cokernel becomes zero after faithful extension, so it was already zero. For algebra generation apply this argument to the quotient by the subalgebra generated by those coefficients, viewed as a graded vector space. Finally, in each nonempty degree gg choose ug∈kg×u_g\in k_g^\times. Multiplication by ugu_g identifies C1C_1 with CgC_g and proves the displayed isomorphism. A finite homogeneous set of module generators can be divided by the corresponding ugu_g’s, reducing the last assertion to degree one. ◻

Lemma 3.18.5. NAG-001957 (Finite admissible maps descend from Gauss fields). Let A→BA\to B be a bounded affinoid homomorphism and L/KL/K a finite iterated Gauss-field extension. Then A→BA\to B is finite and admissible if and only if AL→BLA_L\to B_L is finite and admissible.

Proof. In an orthogonal scalar basis, all completed tensor products are weighted coefficient c0c_0-spaces. A bounded admissible surjection Am→BA^m\to B therefore remains such a surjection: lift each coefficient with a fixed common norm bound, which preserves coefficient decay. This proves the forward direction.

Conversely let b1′,…,bm′b_1',\ldots,b_m' generate BLB_L over ALA_L with a bounded lifting constant CC. Approximate these generators by finite scalar sums bi′′b_i'' so closely that Cmax⁡i∥bi′−bi′′∥<1C\max_i\|b_i'-b_i''\|<1. Let MM be a bound for the action AL×BL→BLA_L\times B_L\to B_L. In choosing the approximations require MCmax⁡i∥bi′−bi′′∥<1MC\max_i\|b_i'-b_i''\|<1. They still generate with a bounded lifting constant. Indeed lift an element using the original generators, replace them by the approximations, and repeat on the resulting error. The errors decrease geometrically, and the coefficient corrections converge in ALmA_L^m. Collect the finitely many elements bj∈Bb_j\in B occurring in the finite scalar expansions of the bi′′b_i''. They generate BLB_L, again with a bounded lifting constant. For b∈Bb\in B write b=∑jcjbjb=\sum_j c_jb_j, cj∈ALc_j\in A_L, with max⁡∥cj∥≤C′∥b∥\max\|c_j\|\leq C'\|b\|. Let p0=pA,0:AL→Ap_0=p_{A,0}:A_L\to A be the constant-coordinate projection from Lemma 3.18.2. It is AA-linear, has norm at most one, and fixes AA. The corresponding projection pB,0:BL→Bp_{B,0}:B_L\to B fixes BB. In the final equality b=∑jφL(cj)(bj⊗1)b=\sum_j\varphi_L(c_j)(b_j\otimes 1), apply the BB-valued constant-coordinate projection pB,0p_{B,0}. On each product it equals φ(pA,0(cj))bj\varphi(p_{A,0}(c_j))b_j by the coefficient formula, first on finite sums and then by continuity. Hence b=∑jp0(cj)bj,max⁡j∥p0(cj)∥≤C′∥b∥.b=\sum_jp_0(c_j)b_j,\qquad \max_j\|p_0(c_j)\|\leq C'\|b\|. This proves finite generation and admissibility over AA. Only linearity of p0p_0 is used. ◻

Lemma 3.18.6. NAG-001958 (A Gauss extension containing all spectral values). For finitely many KK-affinoid algebras AjA_j, one can choose a finite iterated Gauss extension L/KL/K such that every (Aj)L(A_j)_L is strict and all its nonzero spectral values belong to ∣L×∣|L^\times|. One may also require LL to be nontrivially valued.

Proof. First make the algebras strict by a common free-radius extension Ks/KK_s/K. For each strict algebra, the proof of Theorem 3.12.3 gives a finite boundary consisting of points whose residue fields are finite extensions of Gauss fields of strict polydiscs over KsK_s. The latter Gauss fields have value group ∣Ks×∣|K_s^\times|. In a finite valued-field extension the value-group quotient is finite: vectors with values in distinct cosets are linearly independent over the base field, since a sum with a uniquely largest term cannot vanish. Its cardinality is therefore bounded by the field degree. Choose finitely many positive numbers representing generators of all these finite quotient groups. The spectral values of every original AjA_j lie in the group generated by ∣K×∣|K^\times|, the entries of ss, and these finitely many numbers, because they are attained on the specified finite boundaries.

Adjoin Gauss variables with precisely these radii, and, if necessary, one radius different from one. The resulting LL has this generated value group and still makes all AjA_j strict. By Theorem 3.18.3, every positive spectral value on (Aj)L(A_j)_L is the maximum of finitely many products of an original spectral value and a basis-vector norm. Both factors belong to ∣L×∣|L^\times|, proving the assertion. ◻

NAG-001959 · Open at full scope

Theorem 3.18.7. NAG-001959 (The general graded finite-map criterion). A bounded homomorphism A→BA\to B of generalized affinoid algebras is finite and admissible if and only if A~gr⟶B~gr\widetilde A^{\rm gr}\longrightarrow\widetilde B^{\rm gr} is finite. In particular A~gr\widetilde A^{\rm gr} is a finitely generated graded K~gr\widetilde K^{\rm gr}-algebra.

Proof. Choose LL by Lemma 3.18.6. By Lemma 3.18.5, the analytic condition is equivalent to finite admissibility of AL→BLA_L\to B_L. These algebras are strict over a nontrivially valued field, so Theorem 3.16.5 identifies this with finiteness of their canonical degree-one reductions. The support condition and Lemma 3.18.4 identify the latter with finiteness of the full graded map. By Theorem 3.18.3 that map is the scalar extension of the displayed map by L~gr/K~gr\widetilde L^{\rm gr}/\widetilde K^{\rm gr}, and finite generation descends by Lemma 3.18.4. This proves both implications.

Apply the criterion to an admissible quotient presentation K{r−1T}↠AK\{r^{-1}T\}\twoheadrightarrow A. Its source reduction is the weighted polynomial algebra, by the first assertion of Theorem 3.13.2, proved using the coefficient formula of Proposition 3.13.1. The reduction of AA is finite over the image of this finitely generated algebra, so is itself finitely generated. ◻

Finite graded reduction for a fixed absolute presentation. Let KK be complete and let P=K{r−1T}↠BP=K\{r^{-1}T\}\twoheadrightarrow B be an admissible absolute weighted Tate presentation. Then B~gr\widetilde B^{\rm gr} is finite as a module over the image of P~gr=K~gr[t1,…,tm]\widetilde P^{\rm gr}=\widetilde K^{\rm gr}[t_1,\ldots,t_m], where deg⁡(ti)=ri\deg(t_i)=r_i and tit_i maps to the class of the image of TiT_i in degree rir_i.

Proof. Apply Lemma 3.18.6 to the two actual presentations of PP and BB, giving the common nontrivially valued field LL. The weighted Gauss norm of PP is multiplicative by Lemma 1.2.2. It is itself a bounded point and bounds all bounded evaluations by Lemma 1.6.1, so it is the spectral norm. This argument applies to the actual weighted source PLP_L of Proposition 2.33.5. Consequently each ri=ρPL(Ti)r_i=\rho_{P_L}(T_i) belongs to ∣L×∣\lvert L^\times\rvert. Choose ci∈L×c_i\in L^\times of absolute value rir_i. The variables Ti/ciT_i/c_i identify PLP_L isometrically with a unit Tate algebra, and the map to BLB_L is the same admissible quotient in those coordinates.

For an actual upper affinoid CC, its ordinary reduction is its degree-one spectral quotient. Bounded evaluation of all powers gives ρC(b)n≤∥bn∥\rho_C(b)^n\leq\|b^n\|. The positive-radius bound of Theorem 1.20.2 at radius one gives power boundedness when ρC(b)≤1\rho_C(b)\leq1. If ρC(b)<1\rho_C(b)<1, choose a positive ss with ρC(b)<s<1\rho_C(b)<s<1. The bound at ss gives topological nilpotence, including when ρC(b)=0\rho_C(b)=0. Conversely, a power of norm less than one bounds every evaluation strictly below one. These identifications use the original lifts.

Suppose BL≠0B_L\ne0. Use the actual finite family of minimal-prime quotients BjB_j in the strict case of Theorem 3.12.3. The product map of ordinary reductions BL~→∏jB~j\widetilde{B_L}\to\prod_j\widetilde B_j is injective. Indeed, for a power-bounded lift with zero reduction in every component, the maximum of its finitely many component spectral sizes is strictly less than one. The degree-one comparison gives zero reduction upstairs. For the admissible strict quotient BjB_j of the same unit Tate source PLP_L, Theorem 1.12.2 gives the finite unit Tate normalizer of BjB_j and its contractive lift to PLP_L. Proposition 3.16.2 gives finiteness of B~j\widetilde B_j over the ordinary reduction of this normalizer. Its action factors through the ordinary residue polynomial ring S=PL~S=\widetilde{P_L} by the lift. Thus ∏jB~j\prod_j\widetilde B_j is finite over SS. Equip L~\widetilde L with the trivial valuation. Its unit Tate algebra is the polynomial ring SS, so Theorem 1.11.5 gives finiteness of the actual SS-submodule BL~\widetilde{B_L} of this product. The zero quotient has the same finiteness conclusion.

Both graded algebras over LL have support in ∣L×∣\lvert L^\times\rvert. The degree-one assertion of Lemma 3.18.4 gives module finiteness for the full graded map over LL. The natural scalar comparison of Theorem 3.18.3 identifies that map with scalar extension of the given graded map over the graded base field. Finiteness descends by Lemma 3.18.4. Finally the absolute weighted Gauss formula identifies the graded reduction of PP with the displayed polynomial ring. For each nonzero series, the terms of maximal weighted size form a finite sum and the remaining terms have strictly smaller norm. The Gauss maximum detects every nonzero homogeneous polynomial and the leading terms represent every homogeneous class. The induced map sends each variable to the stated initial class, proving the assertion.

Corollary 3.18.8. NAG-001960 (Finite comparison maps for graded reductions). Let A→B,A→CA\to B,A\to C be bounded affinoid maps, and let L/KL/K be any complete valued extension. The natural maps B~gr⊗A~grC~gr⟶B⊗^AC~gr,A~gr⊗K~grL~gr⟶AL~gr\widetilde B^{\rm gr}\otimes_{\widetilde A^{\rm gr}} \widetilde C^{\rm gr} \longrightarrow\widetilde{B\widehat\otimes_A C}^{\rm gr}, \qquad \widetilde A^{\rm gr}\otimes_{\widetilde K^{\rm gr}} \widetilde L^{\rm gr} \longrightarrow\widetilde{A_L}^{\rm gr} are finite.

Proof. Present BB and CC by weighted Tate algebras over KK. Their combined variables give an admissible quotient presentation of D=B⊗^ACD=B\widehat\otimes_A C, by Proposition 2.33.8. The fixed absolute presentation result makes D~gr\widetilde D^{\rm gr} finite over the image of the corresponding graded polynomial algebra. The image of each polynomial variable either vanishes or is the image of the initial form of its image in BB or CC. Indeed nonvanishing means that its spectral value has remained equal to its prescribed radius, so this equality already held in the appropriate factor. Consequently the image of the first comparison map contains the image of that polynomial algebra. The same finite module generators therefore work over this larger image.

For the second map, extend an admissible presentation K{r−1T}↠AK\{r^{-1}T\}\twoheadrightarrow A to L{r−1T}↠ALL\{r^{-1}T\}\twoheadrightarrow A_L. The same argument applies, with L~gr\widetilde L^{\rm gr} as coefficient ring. Any nonzero initial coordinate after extension comes from its original initial coordinate in A~gr\widetilde A^{\rm gr}. ◻

Lemma 3.18.9. NAG-001961 (Shrinking generator radii to spectral bounds). Let A{r−1T}↠BA\{r^{-1}T\}\twoheadrightarrow B be an admissible affinoid presentation with Ti↦biT_i\mapsto b_i. If 0<si≤ri0<s_i\leq r_i and ρB(bi)≤si\rho_B(b_i)\leq s_i, substitution gives an admissible surjection A{s−1T}↠BA\{s^{-1}T\}\twoheadrightarrow B.

Proof. Theorem 1.20.2 gives ∥bin∥≤Cisin\|b_i^n\|\leq C_is_i^n for all nn. Multiplying these finitely many estimates proves bounded convergence of the substituted series. The original series space maps contractively into the smaller-radius series space, and its composite with substitution is the original surjection. Thus surjectivity and the original bounded lifting estimate both persist. ◻

NAG-001962 · Open at full scope

Theorem 3.18.10. NAG-001962 (The graded innerness criterion). Let AA be affinoid, let BB be an AA-affinoid algebra, and let φ:B→D\varphi:B\to D be a bounded AA-homomorphism to a commutative Banach AA-algebra. Call φ\varphi inner over AA if some admissible surjection A{r−1T}↠BA\{r^{-1}T\}\twoheadrightarrow B has ρD(φ(Ti))<ri\rho_D(\varphi(T_i))<r_i for every coordinate. Then φ\varphi is inner over AA if and only if im⁡(B~gr→D~gr)is finite overim⁡(A~gr→D~gr).\operatorname{im}(\widetilde B^{\rm gr}\to\widetilde D^{\rm gr}) \quad\hbox{is finite over}\quad \operatorname{im}(\widetilde A^{\rm gr}\to\widetilde D^{\rm gr}).

Proof. For an inner presentation, Theorem 3.18.7 and the relative polynomial identity in the first assertion of Theorem 3.13.2 make B~gr\widetilde B^{\rm gr} finite over A~gr[T1,…,Tn]\widetilde A^{\rm gr}[T_1,\ldots,T_n]. All these coordinates vanish after mapping to D~gr\widetilde D^{\rm gr}, so the resulting image is finite over the image of A~gr\widetilde A^{\rm gr}.

Conversely take a relative presentation with generators bjb_j. By Lemma 3.18.9, reduce each radius to sj=ρB(bj)s_j=\rho_B(b_j) when this is positive; when it is zero, choose any smaller positive radius. Write βj\beta_j for the corresponding initial coordinate, which is zero in the latter case. Finiteness of the image makes φ(βj)\varphi(\beta_j) integral over the image of A~gr\widetilde A^{\rm gr}. Taking the homogeneous part of a monic equation gives one of degree sjdjs_j^{d_j} with coefficient of Tdj−iT^{d_j-i} of degree sjis_j^i. Lift these coefficients to elements aji∈Aa_{ji}\in A with ρA(aji)≤sji\rho_A(a_{ji})\leq s_j^i, and put Pj(T)=Tdj+∑i=1djajiTdj−i,wj=Pj(bj),tj=sjdj.P_j(T)=T^{d_j}+\sum_{i=1}^{d_j}a_{ji}T^{d_j-i}, \qquad w_j=P_j(b_j),\qquad t_j=s_j^{d_j}. Then ρB(wj)≤tj\rho_B(w_j)\leq t_j and ρD(φ(wj))<tj\rho_D(\varphi(w_j))<t_j. For a zero initial coordinate one can simply take Pj(T)=TP_j(T)=T. Weighted power bounds define a bounded map C=A{t−1U}→BC=A\{t^{-1}U\}\to B, Uj↦wjU_j\mapsto w_j. In the graded reduction each βj\beta_j satisfies the monic equation P‾j(βj)−wˉj=0\overline P_j(\beta_j)-\bar w_j=0 over the image of C~gr\widetilde C^{\rm gr}, with wˉj=0\bar w_j=0 if its degree has dropped. Moreover B~gr\widetilde B^{\rm gr} is finite over A~gr[β1,…,βn]\widetilde A^{\rm gr}[\beta_1,\ldots,\beta_n] by the general graded criterion applied to the shrunk presentation. Successively adjoining the integral βj\beta_j’s proves that C~gr→B~gr\widetilde C^{\rm gr}\to\widetilde B^{\rm gr} is finite. The same criterion shows that C→BC\to B is finite and admissible.

Choose finite CC-module generators v1,…,vmv_1,\ldots,v_m of BB and radii Ri>∥vi∥BR_i>\|v_i\|_B. The map A{t−1U,R−1V}⟶B,Uj↦wj,Vi↦vi,A\{t^{-1}U,R^{-1}V\}\longrightarrow B, \qquad U_j\mapsto w_j,\quad V_i\mapsto v_i, is bounded and surjective. It is admissible because a bounded CC-module lift b=∑icivib=\sum_i c_iv_i gives the lift ∑iciVi\sum_i c_iV_i with a fixed norm bound. The UU-coordinates have strictly smaller spectral size in DD by construction, and the VV-coordinates do as well, since bounded homomorphisms decrease spectral radii. This is the required inner presentation. ◻

Forward implication for absolute evaluation. Let X=M(B)X=\mathcal M(B) and x∈Int⁡(X/M(K))x\in\operatorname{Int}(X/\mathcal M(K)). Then the image of B~gr⟶H(x)~gr\widetilde B^{\rm gr}\longrightarrow \widetilde{\mathcal H(x)}^{\rm gr} is finite over the embedded graded field K~gr\widetilde K^{\rm gr}.

Proof. Definition 2.7.1 supplies an admissible absolute presentation with ∣Ti(x)∣<ri\lvert T_i(x)\rvert<r_i for every coordinate. The isometric structure map K→H(x)K\to\mathcal H(x) induces the embedded graded coefficient field. Evaluation in this actual complete valued field is filtered, since ∣b(x)∣≤ρB(b)\lvert b(x)\rvert\leq\rho_B(b). In degree rir_i the class of the image of TiT_i maps to zero. By the fixed absolute presentation result, B~gr\widetilde B^{\rm gr} has a finite list of module generators over the image of the initial-coordinate polynomial ring. Their images generate the evaluated image over K~gr\widetilde K^{\rm gr}, since all the initial coordinates vanish there.

Reverse finite admissibility for an absolute Tate algebra. Let Q=K{t−1U}Q=K\{t^{-1}U\} be an absolute weighted Tate algebra and ψ:Q→B\psi:Q\to B a bounded map to a generalized KK-affinoid algebra. If B~gr\widetilde B^{\rm gr} is finite over the image of Q~gr\widetilde Q^{\rm gr}, then ψ\psi is finite and admissible.

Proof. Choose the common scalar tower of Lemma 3.18.6 for the actual map ψ\psi. The two extended algebras are strict over the nontrivially valued field LL, and their graded supports lie in its value group. Theorem 3.18.3 identifies the extended graded map with the scalar extension of the original one, so it remains finite. Lemma 3.18.4 gives finiteness in degree one, which is the map on the strict canonical reductions. The reverse implication of Theorem 3.16.5 makes the actual map over LL finite and admissible. Finally Lemma 3.18.5 descends finite admissibility to ψ\psi.

Reverse implication for absolute evaluation. Take A=KA=K, BB a generalized KK-affinoid algebra, x∈M(B)x\in\mathcal M(B), and D=H(x)D=\mathcal H(x), with the evaluation map B→DB\to D. If Sx=im⁡(B~gr⟶H(x)~gr)S_x=\operatorname{im}\bigl( \widetilde B^{\rm gr}\longrightarrow\widetilde{\mathcal H(x)}^{\rm gr} \bigr) is finite over K~gr\widetilde K^{\rm gr}, then x∈Int⁡(M(B)/M(K))x\in\operatorname{Int}(\mathcal M(B)/\mathcal M(K)).

Proof. Start with an admissible absolute presentation with generators bjb_j. By Lemma 3.18.9, use radii sj=ρB(bj)s_j=\rho_B(b_j) when the spectral radius is positive and choose smaller positive radii when it is zero. Write βj\beta_j for the class of bjb_j in degree sjs_j; in the zero spectral radius case the class is zero. Finiteness of SxS_x gives for the image of each βj\beta_j a monic homogeneous equation of degree sjdjs_j^{d_j} over K~gr\widetilde K^{\rm gr}. Lift the coefficient of degree sjis_j^i to aji∈Ka_{ji}\in K with ∣aji∣≤sji\lvert a_{ji}\rvert\le s_j^i, choosing zero for a zero coefficient. Put Pj(T)=Tdj+∑i=1djajiTdj−i,wj=Pj(bj),tj=sjdj.P_j(T)=T^{d_j}+\sum_{i=1}^{d_j}a_{ji}T^{d_j-i}, \qquad w_j=P_j(b_j),\qquad t_j=s_j^{d_j}. For βj=0\beta_j=0, take Pj(T)=TP_j(T)=T. Then ρB(wj)≤tj\rho_B(w_j)\le t_j and ∣wj(x)∣<tj\lvert w_j(x)\rvert<t_j. Bounded weighted substitution gives an actual map Q=K{t−1U}→BQ=K\{t^{-1}U\}\to B, Uj↦wjU_j\mapsto w_j.

By the fixed absolute presentation result, B~gr\widetilde B^{\rm gr} is finite over the graded algebra generated by the βj\beta_j and K~gr\widetilde K^{\rm gr}. Each βj\beta_j is integral over the image of Q~gr\widetilde Q^{\rm gr}: its monic relation there is P‾j(βj)−w‾j=0\overline P_j(\beta_j)-\overline w_j=0, with the class w‾j\overline w_j taken in degree tjt_j. Adjoining all the βj\beta_j and using these finite modules shows that B~gr\widetilde B^{\rm gr} is finite over the image of Q~gr\widetilde Q^{\rm gr}. By reverse finite admissibility for an absolute Tate algebra, the actual map Q→BQ\to B is finite and admissible.

Choose module generators viv_i giving an admissible finite QQ-module presentation of BB, and put a radius uiu_i on each new variable with ui>ρB(vi)u_i>\rho_B(v_i). The map Q{u−1V}→BQ\{u^{-1}V\}\to B, Vi↦viV_i\mapsto v_i, is admissible: its linear polynomials in the ViV_i already give the admissible module presentation, after the harmless rescaling of the finite product norm by the positive uiu_i. The bounds at xx for the UjU_j are ∣wj(x)∣<tj\lvert w_j(x)\rvert<t_j, and those for the ViV_i follow from ∣vi(x)∣≤ρB(vi)<ui\lvert v_i(x)\rvert\le\rho_B(v_i)<u_i. This is an inner absolute presentation of BB at xx.

Relative innerness at an affinoid point. Let A→BA\to B be a bounded map of norm-admissible generalized affinoid algebras over the same complete non-Archimedean field KK. For y∈M(B)y\in\mathcal M(B), let E=H(y)E=\mathcal H(y), and for D=A,BD=A,B let RDR_D be the image of D~gr\widetilde D^{\rm gr} in E~gr\widetilde E^{\rm gr} under evaluation at yy. Then y∈Int⁡(M(B)/M(A))y\in\operatorname{Int}(\mathcal M(B)/\mathcal M(A)) if and only if RBR_B is finite over RAR_A.

Proof. Fix an admissible absolute weighted Tate presentation P=K{r−1T}↠AP=K\{r^{-1}T\}\twoheadrightarrow A. In any admissible relative presentation A{s−1X}↠BA\{s^{-1}X\}\twoheadrightarrow B, adjoining XX coefficientwise to P↠AP\twoheadrightarrow A and composing gives an admissible absolute presentation P{s−1X}↠BP\{s^{-1}X\}\twoheadrightarrow B, by Proposition 1.17.3. If this relative presentation witnesses innerness at yy, the initial classes of the XjX_j at the assigned degrees sjs_j evaluate to zero in E~gr\widetilde E^{\rm gr}. The fixed absolute presentation result gives a finite module over the initial-coordinate polynomial image. Evaluating the finite generators, only the images of the coefficient field and the TiT_i are needed, and these belong to RAR_A. Thus RBR_B is finite over RAR_A.

For the converse, choose the coordinates bjb_j of a relative presentation and shrink to radii sj=ρB(bj)s_j=\rho_B(b_j) when these are positive. For zero spectral radius choose any positive sjs_j at most the original radius. The presentation stays admissible by Lemma 3.18.9. Let βj\beta_j denote the class of bjb_j at degree sjs_j, allowing the zero class. If RBR_B is finite over RAR_A, the image of βj\beta_j satisfies a monic equation over RAR_A. Take its homogeneous component of degree sjdjs_j^{d_j}, where djd_j is the monic degree. Lift its coefficient of degree sjis_j^i through A~gr\widetilde A^{\rm gr} to aji∈Aa_{ji}\in A with ρA(aji)≤sji\rho_A(a_{ji})\leq s_j^i, using zero for a zero coefficient. Put pj(Z)=Zdj+∑i=1djajiZdj−i,tj=sjdj,wj=pj(bj).p_j(Z)=Z^{d_j}+\sum_{i=1}^{d_j}a_{ji}Z^{d_j-i}, \qquad t_j=s_j^{d_j},\qquad w_j=p_j(b_j). For zero evaluated class one may take pj(Z)=Zp_j(Z)=Z. Then ρB(wj)≤tj\rho_B(w_j)\leq t_j and ∣wj(y)∣<tj|w_j(y)|<t_j. Positive-radius power bounds 1.20.2 give the bounded substitution C=A{t−1U}→BC=A\{t^{-1}U\}\to B, with Uj↦wjU_j\mapsto w_j.

In B~gr\widetilde B^{\rm gr}, each βj\beta_j is integral over the image of C~gr\widetilde C^{\rm gr}: its monic equation is p‾j(βj)−w‾j=0\overline p_j(\beta_j)-\overline w_j=0 in degree tjt_j, with w‾j\overline w_j the image of the class of UjU_j. The fixed absolute presentation of BB obtained above gives finiteness over the algebra generated by the image of A~gr\widetilde A^{\rm gr} and the βj\beta_j. Indeed, if an old coefficient coordinate survives at its degree in BB, spectral contraction shows it survives at that degree in AA. Its class in BB therefore comes from AA. It follows by the monic equations that B~gr\widetilde B^{\rm gr} is finite over the image of C~gr\widetilde C^{\rm gr}.

Apply coefficientwise presentation once more to obtain the admissible absolute presentation Q=P{t−1U}↠CQ=P\{t^{-1}U\}\twoheadrightarrow C. The fixed absolute result makes C~gr\widetilde C^{\rm gr} finite over the image of Q~gr\widetilde Q^{\rm gr}. Transitivity, with both rings mapped to B~gr\widetilde B^{\rm gr}, makes B~gr\widetilde B^{\rm gr} finite over the image of Q~gr\widetilde Q^{\rm gr}. The selected reverse finite admissibility result applies to the actual absolute map Q→BQ\to B. Choose its finite module generators viv_i with bounded coefficient lifts. Sending the coefficients through Q↠CQ\twoheadrightarrow C yields lifts with a uniform bound for the same generators over CC. Choose radii ui>ρB(vi)u_i>\rho_B(v_i), so weighted substitution defines C{u−1V}→BC\{u^{-1}V\}\to B, Vi↦viV_i\mapsto v_i. Linear polynomials in ViV_i give bounded lifts: their norms are bounded by the common coefficient-lift bound times max⁡iui\max_i u_i. This is therefore an admissible relative presentation over AA with coordinates wj,viw_j,v_i. Their values at yy are strictly below tj,uit_j,u_i respectively, proving relative innerness. ◻

References. The graded comparison and innerness statements are [Tem04] (Propositions 3.1 and 3.4). The reconstruction uses the finite Shilov boundary for maximum localization, the partial-fraction Gauss basis for scalar descent, and the internally proved canonical finite-map criterion.

3.19 Graded valuation spaces and analytic germs

Gradings in this section are by the ordered group R>0\mathbf R_{>0}. A graded valuation subring VV of a graded field FF is a graded subring such that, for every nonzero homogeneous a∈Fa\in F, either aa or a−1a^{-1} belongs to VV. Its homogeneous nonunits form its unique maximal homogeneous ideal. For a graded subfield k⊂Fk\subset F, write Pk(F)\mathscr P_k(F) for these valuation subrings containing kk. For homogeneous elements aia_i, the sets Pk(F){a1,…,an}={V:a1,…,an∈V}\mathscr P_k(F)\{a_1,\ldots,a_n\} =\{V:a_1,\ldots,a_n\in V\} form a basis of its topology. For a graded subalgebra R⊂FR\subset F, use Pk(F){R}\mathscr P_k(F)\{R\} for the valuations containing RR. A birational space with field FF is a connected quasi-compact quasi-separated space with a local homeomorphism to Pk(F)\mathscr P_k(F). It is affine if this map identifies it with Pk(F){R}\mathscr P_k(F)\{R\} for a finitely generated graded kk-subalgebra RR.

Lemma 3.19.1. NAG-001964 (Graded domination and the integral criterion). Every local graded subring of FF is dominated by a graded valuation subring of FF. For graded kk-subalgebras R⊂S⊂FR\subset S\subset F, Pk(F){R}=Pk(F){S}⟺S is integral over R.\mathscr P_k(F)\{R\}=\mathscr P_k(F)\{S\} \quad\Longleftrightarrow\quad S\text{ is integral over }R. If SS is finitely generated over k⊂Rk\subset R, these conditions are also equivalent to SS being finite over RR.

Proof. A homogeneous prime of a ring graded by an ordered group is an ordinary prime: the extreme nonzero homogeneous terms in a product cannot cancel. Let CC be a local graded subring with homogeneous maximal ideal m\mathfrak m. Apply [StacksAlg] (Tag 00IA) to the ordinary local domain Cm⊂Frac⁡(F)C_{\mathfrak m}\subset\operatorname{Frac}(F). Take the sum of the homogeneous parts of the resulting ordinary valuation ring. This is a graded valuation subring of FF; its maximal homogeneous ideal contracts to m\mathfrak m. This proves domination.

A graded valuation ring is integrally closed on homogeneous elements. If bb were integral but outside it, b−1b^{-1} would lie in its maximal homogeneous ideal. Dividing a homogeneous monic equation by its leading power would then put 11 in that ideal. Conversely, if a homogeneous bb is not integral over RR, then b−1b^{-1} is not a unit of R[b−1]R[b^{-1}]: an expression for its inverse would give a monic equation for bb. By [StacksAlg] (Tag 00E0(2)), choose an ordinary maximal ideal containing b−1b^{-1}. Its homogeneous part is a homogeneous prime: a product of two homogeneous elements in that part has one factor in it. Localize at the homogeneous elements outside this prime and apply domination to the resulting graded local ring. The resulting valuation contains RR and excludes bb. Thus equality of the valuation sets detects integrality of all homogeneous elements of SS. These elements generate SS, so this is ordinary integrality as well. Finally, a finite set of algebra generators, each integral over RR, gives a finite module by reducing their powers with their monic equations. ◻

Proposition 3.19.2. NAG-001965 (Compactness of graded valuation spaces). For every graded subalgebra R⊂FR\subset F, Pk(F){R}\mathscr P_k(F)\{R\} is quasi-compact. In particular Pk(F)\mathscr P_k(F) is quasi-compact and quasi-separated. Every nonempty open subset contains the point FF, so these spaces are irreducible.

Proof. Encode a graded subring by the membership functions of the homogeneous elements of FF in the product of copies of {0,1}\{0,1\}. The conditions of containing RR and kk, being closed under products and sums in one degree, and containing aa or a−1a^{-1} are closed finite-coordinate conditions. The product is compact by [StacksAlg] (Tag 08ZU); it is Hausdorff since distinct points are separated by a coordinate cylinder. Hence its closed solution set is compact: add the open complement to any cover of that set and use the finite-subcover property. Its valuation topology is coarser, proving quasi-compactness. Finite intersections of the basic opens are of the same kind, hence quasi-separatedness. Finally FF satisfies every membership condition, proving irreducibility. ◻

Lemma 3.19.3. NAG-001966 (Laurent refinement in valuation spaces). Every finite covering of a subset W⊂Pk(F)W\subset\mathscr P_k(F) by sets W{a1,…,an}W\{a_1,\ldots,a_n\} admits a refinement by a Laurent covering W{b1ϵ1,…,bmϵm}W\{b_1^{\epsilon_1},\ldots,b_m^{\epsilon_m}\}, ϵi∈{1,−1}\epsilon_i\in\{1,-1\}, with homogeneous nonzero bib_i.

Proof. Write the given cover as Wi=W{ai1,…,ain}W_i=W\{a_{i1},\ldots,a_{in}\}, removing zero entries and appending entries 11 so that ain=1a_{in}=1 for every ii. Let JJ consist of tuples j=(j1,…,js)j=(j_1,\ldots,j_s) with at least one ji=nj_i=n, and set cj=∏iaijic_j=\prod_i a_{ij_i}. The rational sets Vj=W{cj′/cj:j′∈J}V_j=W\{c_{j'}/c_j:j'\in J\} cover WW, since finitely many nonzero homogeneous elements in a valuation field have a largest one for divisibility. Fix ii with ji=nj_i=n and a valuation in VjV_j. It belongs to some WℓW_\ell. If ℓ=i\ell=i there is nothing to prove. Otherwise replacing jℓj_\ell by nn shows that aℓjℓ−1a_{\ell j_\ell}^{-1} belongs to the valuation ring. It also contains aℓjℓa_{\ell j_\ell}, so this is a unit. Replacing jij_i by hh and jℓj_\ell by nn now gives aih/aℓjℓa_{ih}/a_{\ell j_\ell} in the ring for every hh. Thus Vj⊂WiV_j\subset W_i.

A rational covering defined by finitely many cjc_j has a Laurent refinement using all ratios cj/cj′c_j/c_{j'}. On a chosen Laurent cell these ratio conditions orient the complete finite graph. Contract the strongly connected components and choose a terminal component. Products along directed paths show that the ratio of every cjc_j to any chosen member of this terminal component belongs to all valuation rings in the cell; ratios inside a component are units. The cell is therefore contained in its rational chart. This proves the desired refinement. ◻

For an affinoid X=M(A)X=\mathcal M(A) and x∈Xx\in X, put Fx=H(x)~gr,RA,x=im⁡(A~gr→Fx),R(X,x)=PK~gr(Fx){RA,x}.F_x=\widetilde{\mathcal H(x)}^{\rm gr},\qquad R_{A,x}=\operatorname{im}(\widetilde A^{\rm gr}\to F_x), \qquad \mathscr R(X,x)=\mathscr P_{\widetilde K^{\rm gr}}(F_x)\{R_{A,x}\}. Choose an admissible absolute weighted Tate presentation of AA. By the fixed absolute presentation result, A~gr\widetilde A^{\rm gr} is finite over the image of the graded polynomial algebra of that presentation. Evaluation in FxF_x passes this finiteness to the images of both algebras. Hence RA,xR_{A,x} is a finitely generated graded K~gr\widetilde K^{\rm gr}-algebra, and R(X,x)\mathscr R(X,x) is an affine birational space.

Proposition 3.19.4. NAG-001967 (Affinoid neighbourhood invariance). If Y=M(B)⊂XY=\mathcal M(B)\subset X is an affinoid domain containing xx, its reduction is an open subspace of R(X,x)\mathscr R(X,x). If YY is a neighbourhood of xx in XX, this open subspace equals R(X,x)\mathscr R(X,x). Consequently R(X,x)\mathscr R(X,x) depends only on the good analytic germ (X,x)(X,x), with its actual completed residue field.

Proof. Corollary 2.33.15 identifies the completed residue fields. Under this identification RA,x⊂RB,x⊂FxR_{A,x}\subset R_{B,x}\subset F_x. Choose an admissible absolute weighted Tate presentation of BB. Apply the fixed absolute presentation result and evaluate in FxF_x, as for AA. Finiteness over the evaluated polynomial image makes RB,xR_{B,x} a finitely generated graded K~gr\widetilde K^{\rm gr}-algebra. Containment of homogeneous generators is an open condition in R(X,x)\mathscr R(X,x), and Proposition 3.19.2 makes this open quasi-compact.

If YY is a neighbourhood, the topology defined by function values gives a Laurent affinoid neighbourhood W⊂YW\subset Y of xx cut out by finitely many inequalities ∣fi∣≤ri|f_i|\leq r_i and ∣gj∣≥sj|g_j|\geq s_j that are all strict at xx. Put D=O(W)D=\mathcal O(W). Its presentation over AA adjoins fif_i at radius rir_i and gj−1g_j^{-1} at radius sj−1s_j^{-1}. Choose an admissible absolute weighted Tate presentation of AA. Proposition 1.17.3 combines it with the displayed relative presentation into an admissible absolute weighted Tate presentation of DD. Apply the fixed absolute presentation result and evaluate in the common field FxF_x. The resulting finite module generators for RD,xR_{D,x} are taken over the evaluated polynomial image. Each old variable has the same evaluated class in its assigned degree as the class from AA. Each new variable has evaluated norm strictly less than its assigned radius, so its class evaluates to zero. Thus the coefficient image lies in RA,xR_{A,x}, and the same generators make RD,xR_{D,x} finite over RA,xR_{A,x}. The intermediate RB,xR_{B,x} is integral over RA,xR_{A,x} and finitely generated as an algebra, hence finite. Lemma 3.19.1 gives equality of the valuation spaces. Any two affinoid neighbourhoods of the good germ contain such a common Laurent affinoid neighbourhood, with its domain structures identified by Corollary 2.40.9. The identifications are canonical and satisfy the cocycle condition. ◻

Lemma 3.19.5. NAG-001968 (The reduction of an active Weierstrass inequality). For a good germ (X,x)(X,x), let f1,…,fmf_1,\ldots,f_m be analytic near xx, with ri=∣fi(x)∣>0r_i=|f_i(x)|>0, and write λi∈Fx\lambda_i\in F_x for their initial values. Then R(X{r−1f},x)=R(X,x){λ1,…,λm}.\mathscr R\bigl(X\{r^{-1}f\},x\bigr) =\mathscr R(X,x)\{\lambda_1,\ldots,\lambda_m\}. Every subdomain germ of a good germ has a finite cover by germs of this form. Every nonzero homogeneous element of FxF_x is the initial value of an invertible analytic function on some neighbourhood of xx.

Proof. Take an affinoid representative AA on which the functions are defined and put D=A{r−1f}D=A\{r^{-1}f\}. Choose an admissible absolute weighted Tate presentation of AA. Combine it with the actual quotient A{r−1T}↠DA\{r^{-1}T\}\twoheadrightarrow D by Proposition 1.17.3 to obtain an absolute presentation of DD. In the free-polyradius comparisons used for this presentation, Proposition 2.32.1 identifies the extended quotients naturally for the given bounded maps. Apply the fixed absolute presentation result and evaluate at xx in the original field FxF_x. The old variables give evaluated classes from RA,xR_{A,x} in their assigned degrees. The new variable for fif_i has assigned radius ri=∣fi(x)∣r_i=|f_i(x)| and evaluates in that degree to λi\lambda_i. The resulting evaluated polynomial image is contained in RA,x[λ1,…,λm]⊂RD,xR_{A,x}[\lambda_1,\ldots,\lambda_m]\subset R_{D,x}. The finite module generators obtained by evaluation therefore make RD,xR_{D,x} finite over RA,x[λ1,…,λm]R_{A,x}[\lambda_1,\ldots,\lambda_m]. Lemma 3.19.1 gives the formula.

Theorem 2.36.6 gives a finite rational cover of any affinoid subdomain, and a subdomain germ is covered by finitely many affinoid subdomain germs. Discard rational members not containing xx after shrinking the ambient neighbourhood. Every remaining denominator is nonzero at xx, so becomes invertible after another shrinking. The rational inequalities then become Weierstrass inequalities. Those strict at xx can be omitted on a neighbourhood, leaving precisely the displayed active inequalities. Finally Frac⁡(A/ker⁡x)\operatorname{Frac}(A/\ker x) is dense in H(x)\mathcal H(x). Approximate a lift of a nonzero homogeneous element with relative error less than one. It has the same initial value as a fraction a/ba/b, with b(x)≠0b(x)\ne 0 and a(x)≠0a(x)\ne 0; on a smaller neighbourhood this fraction is analytic and invertible. ◻

Proposition 3.19.6. NAG-001969 (Finite coverings and inverse images of germ reductions). Let (X,x)(X,x) be a good germ and let (Xi,x)(X_i,x) be finitely many good subdomain germs covering it. Then R(X,x)=⋃iR(Xi,x).\mathscr R(X,x)=\bigcup_i\mathscr R(X_i,x). For a morphism of good germs (Y,y)→(X,x)(Y,y)\to(X,x) and a subdomain germ X′⊂XX'\subset X, reduction of the inverse image is the inverse image of R(X′,x)\mathscr R(X',x) under restriction of graded valuations. Intersections and finite unions of subdomain germs are consequently compatible with reduction. Here the reduction of a general subdomain germ of a good germ is the union of the reductions in any finite affinoid cover; this union is independent of that cover.

Proof. Use Lemma 3.19.5 to refine the cover and shrink an affinoid representative of XX so that the members have the form Xi={z:∣fij(z)∣≤rij, 1≤j≤n},rij=∣fij(x)∣,fin=1.X_i=\{z:|f_{ij}(z)|\leq r_{ij},\ 1\leq j\leq n\}, \qquad r_{ij}=|f_{ij}(x)|,\qquad f_{in}=1. All the functions may be made invertible near xx. Let JJ be the tuples a=(a1,…,am)a=(a_1,\ldots,a_m) with at least one ai=na_i=n, and put ga=∏ifiaig_a=\prod_i f_{ia_i}, sa=∏iriais_a=\prod_i r_{ia_i}. The rational domains Va={z:∣ga′(z)/ga(z)∣≤sa′/sa for every a′∈J}V_a=\{z:|g_{a'}(z)/g_a(z)|\leq s_{a'}/s_a \text{ for every }a'\in J\} cover XX, by choosing a largest ∣ga(z)∣/sa|g_a(z)|/s_a. They refine the XiX_i. Indeed fix ii with ai=na_i=n and z∈Vaz\in V_a, and choose ℓ\ell with z∈Xℓz\in X_\ell. If ℓ≠i\ell\ne i, the tuples replacing aℓa_\ell by nn first show ∣fℓaℓ(z)∣=rℓaℓ|f_{\ell a_\ell}(z)|=r_{\ell a_\ell}. Replacing also aia_i by jj then gives ∣fij(z)∣≤rij|f_{ij}(z)|\leq r_{ij} for every jj. If ℓ=i\ell=i this already holds. Thus Va⊂XiV_a\subset X_i. Write γa∈Fx\gamma_a\in F_x for the initial value of gag_a. Lemma 3.19.5 identifies the reductions of the VaV_a with R(X,x){γa′/γa:a′∈J}.\mathscr R(X,x)\{\gamma_{a'}/\gamma_a:a'\in J\}. These rational valuation sets cover R(X,x)\mathscr R(X,x): any finite list of nonzero homogeneous elements admits a largest member in the divisibility preorder of a valuation ring. Since each lies in the reduction of the corresponding XiX_i, the covering assertion follows.

Define the reduction of a subdomain germ by the stated finite union. Each affinoid piece gives an open containment locus by Proposition 3.19.4, quasi-compact by Proposition 3.19.2. The finite union is thus a quasi-compact open. Two such covers have a common finite refinement by affinoid subdomain germs: refine their pairwise intersections. The proved covering assertion on each original affinoid piece identifies all three unions. Thus the definition is independent of the cover. For an active Weierstrass inequality, inverse image simply replaces its defining function by its pullback. The isometric embedding H(x)↪H(y)\mathcal H(x)\hookrightarrow\mathcal H(y) preserves its initial value and degree. The displayed formula of Lemma 3.19.5 therefore proves the inverse-image assertion for each such inequality, and hence for finite intersections of them. Refine a general subdomain by these pieces and use the proved finite-cover assertion. Intersection is the particular case of inverse image by a subdomain inclusion; finite union follows from finite covers. ◻

Lemma 3.19.7. NAG-001970 (A valuative inclusion gives a germ inclusion). Let VV be an affinoid subdomain germ of a good germ (X,x)(X,x), and let W=X{∣g1∣≤s1,…,∣gm∣≤sm}W=X\{|g_1|\leq s_1,\ldots,|g_m|\leq s_m\}, where sj=∣gj(x)∣>0s_j=|g_j(x)|>0. If R(V,x)⊆R(W,x)\mathscr R(V,x)\subseteq\mathscr R(W,x), then V⊆WV\subseteq W as subdomain germs.

Proof. Write V=M(B)V=\mathcal M(B) after choosing sufficiently small representatives. By Lemmas 3.19.1 and 3.19.5, each initial value μj\mu_j of gj(x)g_j(x) is integral over RB,xR_{B,x}. Take a homogeneous monic equation and lift its coefficients to aji∈Ba_{ji}\in B with ρB(aji)≤sji\rho_B(a_{ji})\leq s_j^i. The lifted polynomial satisfies ∣gjdj+∑i=1djajigjdj−i∣(x)<sjdj.\left|g_j^{d_j}+\sum_{i=1}^{d_j}a_{ji}g_j^{d_j-i}\right|(x) <s_j^{d_j}. This strict inequality persists in an open neighbourhood of xx in VV. There ∣gj∣>sj|g_j|>s_j is impossible: its leading power would strictly dominate every other term, so the displayed absolute value would equal ∣gj∣dj>sjdj|g_j|^{d_j}>s_j^{d_j}. Intersect the finitely many neighbourhoods obtained this way. It lies in WW, which is exactly the required germ inclusion. ◻

Theorem 3.19.8. NAG-001971 (Subdomains of a good germ). Reduction gives an order-preserving bijection between subdomain germs of a good germ (X,x)(X,x) and nonempty quasi-compact open subspaces of R(X,x)\mathscr R(X,x). It preserves intersections, finite unions and inverse images.

Proof. Every quasi-compact open is a finite union of basic opens R(X,x){λi1,…,λini}\mathscr R(X,x)\{\lambda_{i1},\ldots,\lambda_{in_i}\}. Lift the nonzero homogeneous λij\lambda_{ij} to invertible analytic functions near xx, by Lemma 3.19.5, and impose the corresponding inequalities with radii ∣fij(x)∣|f_{ij}(x)|. These finitely many closed domains in a common affinoid representative give a finite quasi-net on their union by Lemma 2.39.2. The reduction of that union is the prescribed one by Proposition 3.19.6. This proves surjectivity.

First suppose a subdomain germ Y⊂XY\subset X has full reduction. Refine YY by finitely many active Weierstrass germs YiY_i, using Lemma 3.19.5. Their reductions cover R(X,x)\mathscr R(X,x) and therefore admit a Laurent refinement by Lemma 3.19.3. Lift the finitely many homogeneous Laurent functions to invertible analytic functions near xx. The resulting Laurent domains VϵV_\epsilon are actual joint domains by Theorem 2.31.1 and Proposition 2.31.3. Each opposite inequality is active at xx, and Lemma 3.19.5 computes its reduction. The Laurent domains cover a neighbourhood of xx in XX. Each reduction is contained in the reduction of some YiY_i, so Lemma 3.19.7 gives Vϵ⊂YiV_\epsilon\subset Y_i as germs. There are only finitely many cells, so one common shrinking therefore makes their whole union lie in YY. Thus Y=XY=X as germs.

Now suppose R(Y,x)⊆R(Z,x)\mathscr R(Y,x)\subseteq\mathscr R(Z,x). Cover YY and ZZ by finitely many affinoid subdomain germs YiY_i and ZjZ_j. Proposition 3.19.6 shows that the reductions of the Yi∩ZjY_i\cap Z_j, for fixed ii, cover the reduction of YiY_i. The preceding full-reduction assertion, applied inside the good germ YiY_i, implies Yi⊂ZY_i\subset Z. Thus Y⊂ZY\subset Z. This proves injectivity and order reflection. The compatibility assertions are already Proposition 3.19.6. ◻

Theorem 3.19.9. NAG-001972 (The reduction functor on all analytic germs). There is a canonical functor from analytic germs over KK to graded birational spaces over K~gr\widetilde K^{\rm gr}, extending (X,x)↦R(X,x)(X,x)\mapsto\mathscr R(X,x) for affinoid germs. It carries subdomain inclusions to open embeddings, finite subdomain coverings to open coverings, and commutes with intersections, finite unions and inverse images. Its field at (X,x)(X,x) is H(x)~gr\widetilde{\mathcal H(x)}^{\rm gr}.

Proof. Use the domain chart quasi-nets of Proposition 2.40.5 and their restriction and refinement properties in Lemma 2.39.2. Choose finitely many affinoid subdomain germs (Xi,x)(X_i,x) covering a neighbourhood of xx. Refine each pairwise intersection by finitely many affinoid subdomain germs. By Proposition 3.19.4 each reduced piece of the overlap is open in the corresponding reduced chart, with the canonical identification of the completed residue field. Proposition 3.19.6 identifies the overlap with their finite union. Compute in either chart using the common finite refinement and those same field identifications. On triple intersections the maps agree, since each is the identity on the same field and is checked on the common affinoid refinement. Glue these open subspaces.

Proposition 3.19.2 makes each affine piece and each basic overlap piece quasi-compact, so the finite overlap unions are quasi-compact. The gluing is quasi-compact and quasi-separated and has a local homeomorphism to PK~gr(Fx)\mathscr P_{\widetilde K^{\rm gr}}(F_x). Each affine piece is irreducible by the same proposition, and their overlap identifications identify the common trivial graded valuation FxF_x. Thus the gluing is connected. A refinement of the affinoid cover gives the same gluing by the finite-cover assertion of Proposition 3.19.6. Any two choices have a common refinement, proving independence of the atlas and of neighbourhood representatives.

By Theorem 2.40.8, a morphism of germs is represented after finite affinoid refinements by maps of good germs. On each such map use the isometric embedding of completed residue fields and restriction of valuations. These local maps agree on intersections by the inverse-image assertion of Proposition 3.19.6, hence glue. The same assertion proves all stated compatibilities; identities and compositions are checked on the common affinoid refinements. This also proves uniqueness of the extension. ◻

Theorem 3.19.10. NAG-001973 (Subdomains of an arbitrary analytic germ). For every analytic germ (X,x)(X,x), reduction is an order-preserving bijection between its subdomain germs and the nonempty quasi-compact open subspaces of its graded birational reduction. This includes arbitrary positive radii and trivially valued ground fields.

Proof. Take a finite affinoid subdomain-germ cover (Xi,x)(X_i,x). For a quasi-compact open WW in the reduction, the intersections W∩R(Xi,x)W\cap\mathscr R(X_i,x) are quasi-compact by quasi-separatedness. Theorem 3.19.8 lifts them to subdomain germs Yi⊂XiY_i\subset X_i. On overlaps the two lifts coincide, because their reductions coincide and the good-germ theorem reflects equality on every affinoid piece of the overlap. Hence their union is a subdomain germ of XX with reduction WW. This proves surjectivity. If two subdomain germs have equal reductions, intersect with each XiX_i and use the good-germ theorem to obtain equality there. The pieces cover the original germ, so the subdomains are equal. The same argument with inclusions proves order reflection. ◻

References. For the historical construction of graded valuation spaces and germ reductions, see [Tem04] (Lemmas 1.4 and 2.1–2.4; Proposition 4.1; Theorem 4.5). The argument here uses strict Laurent inequalities, fixed absolute weighted Tate presentations and evaluation in the graded residue field. The rational-product refinement makes the covering step explicit.

Products and separation of germ reductions

A morphism of birational spaces with graded field embedding E↪FE\hookrightarrow F is a continuous map compatible with restriction Pk(F)→Pk(E)\mathscr P_k(F)\to\mathscr P_k(E). It is separated, respectively proper, if its canonical map Q⟶P×Pk(E)Pk(F)Q\longrightarrow P\times_{\mathscr P_k(E)}\mathscr P_k(F) is injective, respectively bijective. This canonical map is always a local homeomorphism: use charts on which both structural maps are open embeddings into the same valuation space. Thus the two conditions mean an open embedding and a homeomorphism, respectively. A proper inverse image of an affine birational space is affine.

Proposition 3.19.11. NAG-001974 (Reduction of a fibre-product germ). Let T=Y×XZT=Y\times_X Z and let tt have images y,z,xy,z,x. Put F=H(t)~grF=\widetilde{\mathcal H(t)}^{\rm gr}. Pull the three reductions back to PK~gr(F)\mathscr P_{\widetilde K^{\rm gr}}(F), and denote them by Y0,Z0,X0Y_0,Z_0,X_0. Then R(T,t)≃Y0×X0Z0.\mathscr R(T,t)\simeq Y_0\times_{X_0}Z_0 .

Proof. First take X=M(A)X=\mathcal M(A), Y=M(B)Y=\mathcal M(B) and Z=M(C)Z=\mathcal M(C). Write RA,RB,RC,RDR_A,R_B,R_C,R_D for the images in FF of the graded reductions, where D=B⊗^ACD=B\widehat\otimes_A C. Corollary 3.18.8 says that RDR_D is finite over the graded subalgebra generated by RB,RCR_B,R_C. Lemma 3.19.1 consequently gives P(F){RD}=P(F){RB,RC}=P(F){RB}×P(F){RA}P(F){RC}.\mathscr P(F)\{R_D\} =\mathscr P(F)\{R_B,R_C\} =\mathscr P(F)\{R_B\} \times_{\mathscr P(F)\{R_A\}}\mathscr P(F)\{R_C\}. These are identities of open subspaces with their structural maps, not just identities of points. In general choose finite affinoid subdomain-germ covers of the three spaces, refining the inverse image of the target cover. Their products cover the product germ by the analytic fibre-product construction. The displayed identifications agree on overlaps by the inverse-image compatibility in Theorem 3.19.9. They therefore glue to the asserted isomorphism. ◻

Proposition 3.19.12. NAG-001975 (Complete scalar extension of a germ reduction). Let L/KL/K be a complete valued extension and let y∈XLy\in X_L map to x∈Xx\in X. Put F=H(y)~grF=\widetilde{\mathcal H(y)}^{\rm gr}. Then the reduction of (XL,y)(X_L,y) is the pullback of R(X,x)\mathscr R(X,x) along PL~gr(F)⟶PK~gr(H(x)~gr).\mathscr P_{\widetilde L^{\rm gr}}(F) \longrightarrow \mathscr P_{\widetilde K^{\rm gr}} (\widetilde{\mathcal H(x)}^{\rm gr}).

Proof. On an affinoid piece, the second finite comparison map of Corollary 3.18.8 identifies, up to a finite integral extension, the image of AL~gr\widetilde{A_L}^{\rm gr} in FF with the subalgebra generated by the images of A~gr\widetilde A^{\rm gr} and L~gr\widetilde L^{\rm gr}. Every valuation under consideration already contains the latter field. The integral criterion therefore gives exactly the displayed inverse image. Affinoid subdomain-germ covers and their overlaps commute with scalar extension, so Theorem 3.19.9 glues the identities. ◻

Corollary 3.19.13. NAG-001976 (Proper reductions and base change). If the reduction of a morphism of analytic germs is proper, then its reduction after an analytic base change over any complete valued extension is proper. Properness of birational morphisms is local on an open covering of the target.

Proof. A proper birational map is a homeomorphism to the specified pullback of its target. A further pullback is again a homeomorphism. For analytic products this is Proposition 3.19.11, and for change of ground field it is Proposition 3.19.12. Factor a general change into these two changes. For target locality, a local homeomorphism is bijective if its inverse images over every member of an open cover are bijective. This proves the final assertion. ◻

Proposition 3.19.14. NAG-001977 (The interior criterion for good germs). For a morphism of good germs (Y,y)→(X,x)(Y,y)\to(X,x), the point yy is in the relative analytic interior if and only if R(Y,y)→R(X,x)\mathscr R(Y,y)\to\mathscr R(X,x) is proper.

Proof. Choose affinoid neighbourhood representatives M(B)→M(A)\mathcal M(B)\to\mathcal M(A). The ordinary local test identifies interiority here with interiority for the given good germs. In F=H(y)~grF=\widetilde{\mathcal H(y)}^{\rm gr}, let RBR_B and RAR_A be the images of the two graded algebras. The canonical birational map is the inclusion P(F){RB} ⊆ P(F){RA}.\mathscr P(F)\{R_B\}\ \subseteq\ \mathscr P(F)\{R_A\}. It is bijective exactly when RBR_B is finite over RAR_A, by Lemma 3.19.1 and the finite generation from the fixed absolute presentation. The relative point criterion, applied to A→BA\to B at yy, identifies this with innerness at yy. Shrinking the good representatives does not change their reductions by Proposition 3.19.4, so the criterion is a criterion on the original germs. ◻

Lemma 3.19.15. NAG-001978 (Closed immersions and reductions). A closed immersion of analytic germs induces an isomorphism on reductions. Suppose a germ morphism factors as a closed immersion into a subdomain germ of its target. It is a closed immersion near its distinguished point if and only if its reduction is an isomorphism.

Proof. For an admissible quotient A↠BA\twoheadrightarrow B and a point of M(B)\mathcal M(B), the completed residue fields on the two sides are identical. Compose an admissible absolute weighted Tate presentation of AA with A↠BA\twoheadrightarrow B. The fixed absolute presentation result gives a finite module over the evaluated image of the graded polynomial algebra. By spectral contraction each coordinate surviving at its prescribed degree in BB already survives at that degree in AA. Thus the image of B~gr\widetilde B^{\rm gr} is finite over the image of A~gr\widetilde A^{\rm gr} in the same field. The valuation criterion then identifies the reductions. This proves the assertion on affinoid pieces; finite subdomain covers prove it for closed immersions of arbitrary germs.

For the converse write the factorization as (Y,y)↪(X′,x)↪(X,x)(Y,y)\hookrightarrow(X',x)\hookrightarrow(X,x). The first part identifies R(Y,y)\mathscr R(Y,y) with R(X′,x)\mathscr R(X',x). Hence an isomorphism to R(X,x)\mathscr R(X,x) means that the subdomain germ X′X' has full reduction. Theorem 3.19.10 gives (X′,x)=(X,x)(X',x)=(X,x), so the factorization is a closed immersion on an ordinary neighbourhood. The forward direction follows from the first part. ◻

Proposition 3.19.16. NAG-001979 (Separation of analytic germs). A morphism of analytic germs is separated if and only if its reduction is separated. In particular, an analytic germ whose reduction is affine admits a compact separated neighbourhood representative.

Proof. The diagonal is locally a closed immersion into a subdomain. Choose finite affinoid subdomain-germ covers of source and target, with each source piece YiY_i mapping to one target piece XiX_i. In Z=Y×XYZ=Y\times_XY the spaces Zi=Yi×XiYiZ_i=Y_i\times_{X_i}Y_i are analytic domains. On ZiZ_i the diagonal is given by the admissible quotient Bi⊗^AiBi↠BiB_i\widehat\otimes_{A_i}B_i\twoheadrightarrow B_i; the bounded section b↦b⊗1b\mapsto b\otimes 1 proves admissibility. The finitely generated kernel ideals agree on overlaps: after a common affinoid refinement they are the kernels of the same restricted quotient. The finite-module equivalence of Theorem 2.39.6 glues these ideals to a closed subspace of ⋃iZi\bigcup_i Z_i, whose charts are the YiY_i. Thus the diagonal factors, as a germ, through this subdomain by a closed immersion.

Let z=Δ(y)z=\Delta(y) and F=H(y)~grF=\widetilde{\mathcal H(y)}^{\rm gr}. Proposition 3.19.11 identifies R(Z,z)\mathscr R(Z,z) with the fibre product of two copies of R(Y,y)\mathscr R(Y,y) over R(X,x)×P(Fx)P(F)\mathscr R(X,x)\times_{\mathscr P(F_x)}\mathscr P(F). Its diagonal is an isomorphism precisely when the canonical map from R(Y,y)\mathscr R(Y,y) to that last space is injective. Lemma 3.19.15 therefore identifies separatedness of the reduction with the diagonal being a closed immersion on a neighbourhood of zz.

We spell out the neighbourhood step. First restrict the target to a compact Hausdorff neighbourhood of xx and the source to a compact Hausdorff neighbourhood of yy mapping into it; these exist by Proposition 2.40.5. Their fibre product is compact Hausdorff by the construction in Theorem 2.40.11. Suppose the diagonal is a closed immersion over an open neighbourhood OO of zz. Its inverse image is an open neighbourhood of yy. Choose a compact analytic neighbourhood Y′Y' of yy inside that inverse image. The diagonal of Y′Y' is a closed immersion over the inverse image of OO in Y′×XY′Y'\times_XY'. Its full image is closed, because Y′Y' is compact and this product is Hausdorff. The open complement of that image, together with the inverse image of OO, covers the product; on the complement the closed subspace is empty. Thus the diagonal of Y′Y' is a closed immersion everywhere. This proves separatedness of a representative, rather than merely a condition at one point of its diagonal. Conversely a separated representative has a closed diagonal, giving the reduction condition above. The last assertion follows since an affine birational space embeds into its valuation space. ◻

Lemma 3.19.17. NAG-001980 (Affinoid intersections in a separated space). The intersection of two affinoid domains in a separated analytic space is affinoid.

Proof. For domains U,V⊂XU,V\subset X, their intersection represents U×XVU\times_XV by the domain universal property. It is the inverse image of the closed diagonal of XX in the affinoid space U×KVU\times_KV. A closed analytic subspace of an affinoid is affinoid: its coherent ideal is I~\widetilde I for a finite ideal II by Theorem 2.39.6, and the space is represented by the admissible quotient A/IA/I. Completed tensor products carry that quotient to its base-changed quotient. These observations prove the assertion. ◻

NAG-001981 · Open at full scope

Lemma 3.19.18. NAG-001981 (Preparing a two-sided Laurent neighbourhood). Suppose a separated germ (X,x)(X,x) is covered by good subdomain germs Y,ZY,Z, and its reduction is affine. Suppose, for a nonzero homogeneous λ∈Fx\lambda\in F_x, that R(Y,x)=R(X,x){λ},R(Z,x)=R(X,x){λ−1}.\mathscr R(Y,x)=\mathscr R(X,x)\{\lambda\},\qquad \mathscr R(Z,x)=\mathscr R(X,x)\{\lambda^{-1}\}. Then one can choose affinoid neighbourhood representatives Y=M(B)Y=\mathcal M(B), Z=M(C)Z=\mathcal M(C), whose union is a neighbourhood of xx, and elements and radii satisfying every hypothesis of Theorem 2.45.2.

Proof. Set t=deg⁡(λ)t=\deg(\lambda) and write R(X,x)=P(Fx){α1,…,αm}\mathscr R(X,x)=\mathscr P(F_x)\{\alpha_1,\ldots,\alpha_m\}, with homogeneous nonzero αi\alpha_i of degree rir_i. All shrinkings below are neighbourhood shrinkings inside YY and ZZ. Their union remains a neighbourhood in XX: the two domains form a finite closed cover of a compact neighbourhood, and this is the quasi-net topology criterion of Lemma 2.40.2.

Step 1: an exact common annular overlap. Lift λ\lambda to invertible functions ff on a neighbourhood in YY and gg on a neighbourhood in ZZ. The cut ∣f∣≤t|f|\le t in YY has the same reduction as YY, and ∣g−1∣≤t−1|g^{-1}|\le t^{-1} in ZZ has the same reduction as ZZ. By Theorem 3.19.8 these cuts are neighbourhoods of xx. Choose affinoid neighbourhoods inside them, so ρB(f)≤t,ρC(g−1)≤t−1.\rho_B(f)\le t,\qquad \rho_C(g^{-1})\le t^{-1}. Their intersection WW is affinoid by Lemma 3.19.17. At xx we have ∣(f−g)(x)∣<t|(f-g)(x)|<t. Shrink the two neighbourhoods so that ρW(f−g)<t\rho_W(f-g)<t; continuity on the old intersection, and the relative-open-set definition, permit one common shrinking. Thus on WW one has ∣f∣=∣g∣=t|f|=|g|=t, and W⊂Y{∣f∣≥t},W⊂Z{∣g∣≤t}.W\subset Y\{|f|\ge t\},\qquad W\subset Z\{|g|\le t\}. The three domain germs in this display have reduction R(X,x){λ,λ−1}\mathscr R(X,x)\{\lambda,\lambda^{-1}\}. The domain classification gives affinoid neighbourhoods Y′⊂YY'\subset Y, Z′⊂ZZ'\subset Z with Y′∩Z=Y′{∣f∣≥t},Y∩Z′=Z′{∣g∣≤t}.Y'\cap Z=Y'\{|f|\ge t\},\qquad Y\cap Z'=Z'\{|g|\le t\}. Choose a strict Laurent neighbourhood W′W' of xx inside Y′∩Z′Y'\cap Z' in the old affinoid WW. Write it using finitely many conditions ∣ui∣≤ai|u_i|\le a_i and ∣vj∣≥bj|v_j|\ge b_j, strict at xx. The first displayed overlap is a Weierstrass domain of Y′Y' (its defining function is the analytic function f−1f^{-1}), and the second is a Weierstrass domain of Z′Z'. Restriction to a Weierstrass domain has dense image: truncate its Tate series in the adjoined functions, which are already functions on the source. Approximate ui,vju_i,v_j from each side, with errors respectively less than ai,bja_i,b_j on that overlap. The non-Archimedean inequality makes the resulting cuts have exactly W′W' as their intersections with the old overlap. They are neighbourhoods of xx because the inequalities there are strict. Replacing Y,Z,WY,Z,W by these two cuts and W′W' gives the exact identities W=Y{∣f∣≥t}=Z{∣g∣≤t},A=B{tf−1}=C{t−1g}.W=Y\{|f|\ge t\}=Z\{|g|\le t\},\qquad A=B\{t f^{-1}\}=C\{t^{-1}g\}.

Step 2: presentations with small extra coordinates. Lift each αi\alpha_i to an invertible analytic function fif_i near xx in YY, and impose ∣fi∣≤ri|f_i|\le r_i. Every such cut has full reduction in YY, since every valuation of R(Y,x)\mathscr R(Y,x) contains αi\alpha_i. It is therefore a neighbourhood. Perform the first step again if needed to retain the exact overlap identities. Choose p<tp<t with ρB(f−1)<p−1\rho_B(f^{-1})<p^{-1}, and set DB=K{r1−1T1,…,rm−1Tm,t−1S,pS−1}.D_B=K\{r_1^{-1}T_1,\ldots,r_m^{-1}T_m, t^{-1}S,pS^{-1}\}. Map TiT_i to fif_i and SS to ff. Use the quotient presentations of the preceding Weierstrass cuts: the powers of fif_i and ff are bounded by constant multiples of rinr_i^n and tnt^n, respectively. The strict inequality for f−1f^{-1} gives its corresponding power bound with radius p−1p^{-1}, either by the spectral-radius formula or Theorem 1.20.2. Thus this Tate substitution is a bounded homomorphism. The graded image of DBD_B in FxF_x contains K~gr[α1,…,αm,λ]\widetilde K^{\rm gr}[\alpha_1,\ldots,\alpha_m,\lambda]: the displayed coordinates have exactly the prescribed initial values, and the inverse coordinate has zero initial value at its allowed radius p−1>t−1p^{-1}>t^{-1}. Equality of the valuation sets for YY and this finitely generated subalgebra makes the image of B~gr\widetilde B^{\rm gr} finite over it, by Lemma 3.19.1. It is therefore finite over the possibly larger graded image of DBD_B as well. Theorem 3.18.10, applied relatively to DBD_B, gives an admissible epimorphism DB{a1−1U1,…,al−1Ul}↠B,∣uj(x)∣<aj.D_B\{a_1^{-1}U_1,\ldots,a_l^{-1}U_l\} \twoheadrightarrow B,\qquad |u_j(x)|<a_j. Give A=B{tf−1}A=B\{t f^{-1}\} the quotient norm from the corresponding boundary presentation in which ∣S∣=t|S|=t. By density of C→AC\to A, choose gi,g′∈Cg_i,g'\in C with ∥fi−gi∥A<ri,∥f−g′∥A<t.\|f_i-g_i\|_A<r_i,\qquad \|f-g'\|_A<t. On WW one has ∣g′−g∣<t|g'-g|<t. Restrict ZZ to ∣g′−g∣≤d<t|g'-g|\le d<t, where dd is larger than the supremum on WW; this cut contains WW and is a neighbourhood of it. Then ∣g′∣=∣g∣|g'|=|g| throughout the new ZZ, so g′g' is invertible, has inverse bounded by t−1t^{-1}, and defines the same overlap. Replace gg by g′g'. Impose also ∣gi∣≤ri|g_i|\le r_i. These cuts contain WW and are neighbourhoods of xx, since their reductions are full: the initial value of gi(x)g_i(x) is αi\alpha_i. Choose q>tq>t and q>ρC(g)q>\rho_C(g). Repeating the relative innerness argument with DC=K{r1−1T1,…,rm−1Tm,q−1S,tS−1}.D_C=K\{r_1^{-1}T_1,\ldots,r_m^{-1}T_m, q^{-1}S,tS^{-1}\}. Its graded image contains K~gr[α1,…,αm,λ−1]\widetilde K^{\rm gr}[\alpha_1,\ldots,\alpha_m,\lambda^{-1}]. The same integral criterion gives an admissible epimorphism DC{b1−1V1,…,bh−1Vh}↠C,∣vj(x)∣<bj.D_C\{b_1^{-1}V_1,\ldots,b_h^{-1}V_h\} \twoheadrightarrow C,\qquad |v_j(x)|<b_j.

Step 3: match all coordinates. Density of C→AC\to A and B→AB\to A gives elements ui′∈Cu_i'\in C, vj′∈Bv_j'\in B with ∥ui−ui′∥A<ai,∥vj−vj′∥A<bj.\|u_i-u_i'\|_A<a_i,\qquad \|v_j-v_j'\|_A<b_j . Cut YY by ∣vj′∣≤bj|v_j'|\le b_j and ZZ by ∣ui′∣≤ai|u_i'|\le a_i. These cuts contain all of WW, because the original vj,uiv_j,u_i have the stated spectral bounds there; they are neighbourhoods of xx because the values at xx are strictly smaller than the bounds. Thus the overlap and its exact Weierstrass descriptions are unchanged. The two presentations now have coordinate lists (f1,…,fm,u1,…,ul,v1′,…,vh′),(g1,…,gm,u1′,…,ul′,v1,…,vh).(f_1,\ldots,f_m,u_1,\ldots,u_l,v'_1,\ldots,v'_h),\qquad (g_1,\ldots,g_m,u'_1,\ldots,u'_l,v_1,\ldots,v_h). Use radii (ri,ai,bj)(r_i,a_i,b_j) for these lists and p<t<qp<t<q for the annulus coordinate. Appending the cut coordinates can only decrease the induced quotient norm on AA, since every old lift is still a permitted lift of the same norm. Hence all strict difference estimates above remain valid in the new, actual overlap quotient norm. These are exactly the hypotheses of Theorem 2.45.2. ◻

NAG-001982 · Open at full scope

Theorem 3.19.19. NAG-001982 (Goodness is detected by graded reduction). An analytic germ over an arbitrary complete non-Archimedean field is good if and only if its graded birational reduction is affine.

Proof. For a good germ choose an affinoid neighbourhood. The definition of its reduction and finite generation in Theorem 3.18.7 show that the reduction is affine.

Conversely suppose R=R(X,x)R=\mathscr R(X,x) is affine. Proposition 3.19.16 permits a compact separated neighbourhood representative. Cover its germ by finitely many affinoid subdomain germs. Their reductions are affine opens in the affine space RR. Express each as R{λ1,…,λe}R\{\lambda_1,\ldots,\lambda_e\}: each is affine in the common valuation space, so adjoining its finite generators to generators for RR has this effect. Lemma 3.19.3 refines this cover by the Laurent cells R{μ1ϵ1,…,μnϵn},ϵi∈{1,−1}.R\{\mu_1^{\epsilon_1},\ldots,\mu_n^{\epsilon_n}\}, \qquad\epsilon_i\in\{1,-1\}. Each cell has an affinoid lift inside one original chart: lift the finitely many μi\mu_i to invertible functions near xx in that chart and impose the indicated Weierstrass inequalities, shrinking to an affinoid neighbourhood first. The reduction formula of Lemma 3.19.5 identifies its reduction with that cell. Theorem 3.19.10 says these lifts cover the germ.

Induct on nn. For n=0n=0 a single affinoid lift has full reduction and is a neighbourhood by the domain classification. For n>0n>0, group the cells according to the sign of μ1\mu_1. Each group has affine reduction R{μ1±1}R\{\mu_1^{\pm 1}\} and a Laurent cover with n−1n-1 remaining functions. The induction hypothesis makes its lifted germ good. Thus the original germ is the union of two good subdomain germs whose reductions are R{μ1}R\{\mu_1\} and R{μ1−1}R\{\mu_1^{-1}\}. Lemma 3.19.18 supplies affinoid neighbourhood representatives with matched Laurent presentations. Theorem 2.45.2 makes their union affinoid. This union is a neighbourhood of xx, proving goodness. ◻

References. The diagonal criterion and the goodness theorem reconstruct [Tem04] (Proposition 4.8 and Theorem 5.1) and the detailed neighbourhood preparation in [Tem00] (Theorem 3.1). The graded finite criterion, domain-germ classification and double Laurent gluing used in this proof are proved internally above.

Strict germs and ordinary formal-model reduction

Lemma 3.19.20. NAG-001983 (Restriction to degree one). For an extension of graded fields k⊂Fk\subset F, the map π:Pk(F)⟶Pk1(F1),V⟼V∩F1\pi:\mathscr P_k(F)\longrightarrow\mathscr P_{k_1}(F_1), \qquad V\longmapsto V\cap F_1 is surjective. If Q0Q_0 is an ordinary birational space with field F1F_1, its pullback QQ along π\pi is affine if and only if Q0Q_0 is affine.

Proof. Let V0⊂F1V_0\subset F_1 be a valuation ring containing k1k_1. In the graded subring C=kV0⊂FC=kV_0\subset F, its degree-rr part, for rr in the support of kk, is crV0c_rV_0 for any nonzero cr∈krc_r\in k_r. Distinct degrees are independent. Consequently mV0C\mathfrak m_{V_0}C is proper and its contraction to V0V_0 is mV0\mathfrak m_{V_0}. Choose an ordinary maximal ideal above it by [StacksAlg] (Tag 00E0(2)), take its homogeneous part, and localize homogeneously. Lemma 3.19.1 gives a graded valuation ring VV dominating this local ring. It contains kk and contracts to V0V_0: if a∈F1∖V0a\in F_1\setminus V_0, then a−1∈mV0a^{-1}\in\mathfrak m_{V_0} lies in the maximal ideal of VV, so a∉Va\notin V. This proves surjectivity.

If Q0=Pk1(F1){b1,…,bm}Q_0=\mathscr P_{k_1}(F_1)\{b_1,\ldots,b_m\}, then Q=Pk(F){b1,…,bm}Q=\mathscr P_k(F)\{b_1,\ldots,b_m\}, proving one direction. Conversely suppose QQ is affine. Surjectivity of π\pi shows that the structural local homeomorphism of Q0Q_0 is injective; hence Q0Q_0 is an open subspace of Pk1(F1)\mathscr P_{k_1}(F_1). Write Q=Pk(F){a1,…,an}Q=\mathscr P_k(F)\{a_1,\ldots,a_n\} with nonzero homogeneous aia_i. Let HH be the subgroup of degrees whose positive powers belong to the support of kk. If deg⁡(ai)∉H\deg(a_i)\notin H, then FH[ai−1]F_H[a_i^{-1}] is a polynomial graded algebra over the graded subfield FHF_H: distinct powers have distinct degree classes modulo HH. Localize at the homogeneous prime (ai−1)(a_i^{-1}) and use graded domination. The resulting valuation contains FHF_H, hence F1F_1 and kk, but excludes aia_i. Its restriction is the trivial valuation F1F_1, which belongs to every nonempty open Q0Q_0. This contradicts Q=π−1(Q0)Q=\pi^{-1}(Q_0). Thus deg⁡(ai)di=deg⁡(ci)\deg(a_i)^{d_i}=\deg(c_i) for some di>0d_i>0 and nonzero homogeneous ci∈kc_i\in k. Put bi=aidi/ci∈F1b_i=a_i^{d_i}/c_i\in F_1. A valuation containing kk contains aia_i exactly when it contains bib_i. Hence Q=π−1(Pk1(F1){b1,…,bn}).Q=\pi^{-1}\bigl(\mathscr P_{k_1}(F_1)\{b_1,\ldots,b_n\}\bigr). Surjectivity gives the same expression for Q0Q_0, proving that it is affine. ◻

Proposition 3.19.21. NAG-001984 (Strict affinoid germs and model rings). Assume KK is nontrivially valued. Let AA be strict affinoid, A0A_0 an admissible affine model, and x∈M(A)x\in\mathcal M(A). Put F=H(x)~grF=\widetilde{\mathcal H(x)}^{\rm gr} and let R0⊂F1R_0\subset F_1 be the image of A0/K∘∘A0A_0/K^{\circ\circ}A_0 under evaluation at xx. Then R(M(A),x)=π−1(PK~(F1){R0}).\mathscr R(\mathcal M(A),x) =\pi^{-1}\bigl(\mathscr P_{\widetilde K}(F_1)\{R_0\}\bigr). The ordinary birational space on the right is independent of A0A_0. It can also be computed using the image of the canonical reduction A~\widetilde A in F1F_1.

Proof. The ring R0R_0 is a finitely generated K~\widetilde K-algebra: reduction of a restricted power series is a polynomial, and a finite topological presentation of A0A_0 supplies finitely many generators. Its image lies in the degree-one part of the image of A~gr\widetilde A^{\rm gr}, giving one inclusion of valuation sets.

Conversely let a graded valuation contain K~gr\widetilde K^{\rm gr} and R0R_0. A nonzero homogeneous element of the image of A~gr\widetilde A^{\rm gr} is the initial value of some a∈Aa\in A with ∣a(x)∣=ρA(a)=r>0|a(x)|=\rho_A(a)=r>0. By Lemma 3.16.6, choose d>0d>0, c∈K×c\in K^\times with rd=∣c∣r^d=|c|. The element b=ad/cb=a^d/c is power-bounded. The integral adjunction proved in Theorem 3.16.8 makes bb integral over A0A_0. Reduce a monic equation at xx: the initial value of bb is integral over R0R_0 and therefore belongs to the valuation ring. It equals the ddth power of the initial value of aa, divided by the homogeneous unit given by cc. The valuation ring thus also contains that initial value of aa. This proves equality of the two sets.

The identical argument with all power-bounded elements in place of the model generators gives the canonical reduction formula. Finally Lemma 3.19.20 makes π\pi surjective, so the ordinary valuation sets are identical for any two models. Their topologies are the induced valuation topology, giving the asserted independence. ◻

NAG-001985 · Open at full scope

Theorem 3.19.22. NAG-001985 (Ordinary reduction on the strict category). For a strictly KK-analytic germ, with KK nontrivially valued, there is a canonical ordinary birational reduction R0(X,x)\mathscr R_0(X,x) with field H(x)~\widetilde{\mathcal H(x)}. It is functorial and independent of strict affinoid charts and neighbourhoods. Its nonempty quasi-compact open subspaces classify strict subdomain germs. It is related to the graded reduction by R(X,x)=R0(X,x)×PK~(H(x)~)PK~gr(H(x)~gr).\mathscr R(X,x)= \mathscr R_0(X,x) \times_{\mathscr P_{\widetilde K} (\widetilde{\mathcal H(x)})} \mathscr P_{\widetilde K^{\rm gr}} (\widetilde{\mathcal H(x)}^{\rm gr}). Goodness, separation of germ morphisms, and relative interiority are detected by affineness, separatedness, and properness of this ordinary reduction, respectively.

Proof. On each strict affinoid chart use the ordinary space in Proposition 3.19.21. Strict affinoid atlas intersections have common finite strict affinoid subdomain-germ refinements. The formula in that proposition identifies the pullbacks of the ordinary overlap spaces with the graded overlap spaces. Surjectivity of π\pi in Lemma 3.19.20 reflects equality, inclusion and a finite covering of these open subspaces. Consequently the ordinary charts glue with exactly the same cocycle argument as Theorem 3.19.9, and their pullback is the displayed graded reduction. Common strict affinoid refinements give independence; valuation restriction gives functoriality. A neighbourhood shrinking has unchanged graded reduction, and hence unchanged ordinary reduction by the same surjectivity.

For the subdomain assertion work first in one strict affinoid chart. A quasi-compact ordinary open is a finite union of basic opens defined by elements of F1=H(x)~F_1=\widetilde{\mathcal H(x)}. Lift each nonzero element to an invertible analytic function of absolute value one at xx, using the fraction approximation in Lemma 3.19.5. One may take strict rational neighbourhoods on which the numerator and denominator are invertible: strict rational neighbourhoods form a basis, since ∣K×∣\sqrt{|K^\times|} is dense in R>0\mathbf R_{>0}. The cuts ∣f∣≤1|f|\le 1 are strict Weierstrass domains. Their finite unions lift the given ordinary open. Equality and inclusion of lifts follow by pulling back with π\pi and applying Theorem 3.19.10. On a general strict germ perform the construction in finitely many strict charts; equality on overlaps glues the subdomain. This proves the stated classification.

The displayed pullback and Lemma 3.19.20 identify affineness of the two reductions. Apply Theorem 3.19.19. For a germ morphism the canonical map which defines ordinary birational separation or properness becomes, after pullback by the surjective degree-one restriction on the source field, the canonical map for the graded reductions. Injectivity and bijectivity are reflected by a surjective base change of sets. Hence ordinary and graded separation or properness are equivalent. Now use Proposition 3.19.16 and Theorem 2.46.1. This proves all three criteria. ◻

Strict ordinary reduction and neighborhood invariance. Let (X,x)(X,x) be a germ in the strict analytic GG-category over the standing complete nontrivially valued field. Put F=H(x)~grF=\widetilde{\mathcal H(x)}^{\rm gr} and F0=H(x)~F_0=\widetilde{\mathcal H(x)}. There is a canonical ordinary birational reduction R0(X,x)\mathscr R_0(X,x) over the valuation space with field F0F_0. It is independent of the strict charts and of ordinary neighborhood representatives. Its pullback along degree-one restriction is the given graded reduction. On an actual strict affinoid chart D=M(A)D=\mathcal M(A) containing xx, with admissible affine model A0A_0, its chart is OD=PK~(F0){im⁡(A0/K∘∘A0→F0)}.O_D=\mathscr P_{\widetilde K}(F_0) \{\operatorname{im}(A_0/K^{\circ\circ}A_0\to F_0)\}.

Use the strict chart net of the germ. Finitely many charts through xx cover a neighborhood, and in the strict site each pairwise or triple intersection has such finite common strict refinements relative to that intersection. For example, choose the finitely many members through xx by the quasi-net condition on the intersection. For a triple, apply the same condition a second time to each pairwise refining member and the third chart, then use transitivity of the quasi-net condition. In a prescribed ordinary open neighborhood, shrink the finitely many chart members through xx by Lemma 2.35.25 inside each member. These are strict affinoid domains containing relative neighborhoods of xx, so their union still contains a neighborhood of xx. We use strict-site refinements throughout.

The completed residue fields of these nested chart domains are canonically the same by Corollary 2.33.15. Use an affine model of each actual strict algebra, which exists by Proposition 3.14.7. Put P0=PK~(F0)P_0=\mathscr P_{\widetilde K}(F_0), P=PK~gr(F)P=\mathscr P_{\widetilde K^{\rm gr}}(F), and let π1:P→P0\pi_1:P\to P_0 be restriction of graded valuations. It is surjective by Lemma 3.19.20 and continuous by the defining containment opens. Proposition  3.19.21 gives the displayed ordinary chart, independently of the model, and identifies its inverse image with the actual graded chart.

For charts Di,DjD_i,D_j, let OijO_{ij} be the finite union of the ordinary valuation opens of a common strict refinement of their intersection germ. Its inverse image under π1\pi_1 is exactly the graded overlap by Theorem 3.19.9. Surjectivity therefore puts OijO_{ij} in both ordinary charts and makes it independent of the chosen overlap refinement. Identify the two copies of OijO_{ij} by the identity on valuations in P0P_0. The graded triple intersection identities imply Oij∩Oik=Oij∩OjkO_{ij}\cap O_{ik}=O_{ij}\cap O_{jk} as subsets of P0P_0, again by surjectivity. The transition maps hence satisfy the cocycle. Glue these opens, with the topology tested on the open charts. The resulting space maps by a local homeomorphism to P0P_0. Pullback recovers the graded gluing: the pulled-back charts cover the pullback, and their identifications are the actual graded overlap identifications. Distinct chart sheets are kept in this gluing.

Each ordinary chart is a basic quasi-compact valuation open, since its residue algebra image is finitely generated as in Proposition 3.19.21. The quasi-compactness and irreducibility argument of Proposition 3.19.2 applies to the valuation spaces concentrated in degree one. The ordinary space is quasi-compact by its finite chart cover. Two basic opens in different charts intersect in the corresponding overlap, cut by their two containment conditions. This is a finite union of basic quasi-compact opens, since each overlap is such a union. Every quasi-compact open has a finite cover by chart basic opens, so intersections of quasi-compact opens are quasi-compact. The trivial valuation F0F_0 lies in every chart and every overlap, so the irreducible chart spaces glue to a connected space.

A common strict refinement of chart covers induces maps of the ordinary glued spaces over P0P_0. On a refining chart, the map is the inclusion of its ordinary valuation locus in the original chart. This inclusion and its agreement on overlaps follow from the corresponding graded maps and the surjectivity of π1\pi_1. After pullback the comparison is the canonical graded comparison of Theorem 3.19.9. A continuous map over P0P_0 between spaces locally homeomorphic to P0P_0 is itself a local homeomorphism: restrict around the source point and its image to two such charts, and use their common open image in P0P_0. If the pullback of this map along π1\pi_1 is an isomorphism, its map on each fibre over P0P_0 is bijective, since that fibre can be tested at a lift of its base point. It is thus a bijective local homeomorphism and hence an isomorphism. This proves independence of the strict covers. For ordinary neighborhood restriction, use the finite strict shrinkings above. The graded reductions are those of the same germ by Theorem 3.19.9, so the same argument gives the canonical ordinary comparison. The comparisons compose because they agree on common strict refinements.

Lemma 3.19.23. NAG-001986 (Valuations and a proper pointed comparison). Let ZZ be an integral finite-type kk-scheme equipped with a dominant map Spec⁡F→Z\operatorname{Spec}F\to Z. The spaces Pk(F){R}\mathscr P_k(F)\{R\} for affine opens Spec⁡R⊂Z\operatorname{Spec}R\subset Z glue to a birational space Val⁡(Z,F)\operatorname{Val}(Z,F). Its points are valuations of F/kF/k together with a centre on ZZ. If Z′→ZZ'\to Z is proper and the specified generic point lifts to Spec⁡F→Z′\operatorname{Spec}F\to Z', then the induced map Val⁡(Z′,F)⟶Val⁡(Z,F)\operatorname{Val}(Z',F)\longrightarrow \operatorname{Val}(Z,F) is an isomorphism, with Z′Z' replaced by the reduced closure of that generic point when necessary.

Proof. A valuation ring V⊂FV\subset F containing RR has the unique centre mV∩R\mathfrak m_V\cap R on Spec⁡R\operatorname{Spec}R. Indeed the elements outside that prime are units of VV, so the map R→VR\to V factors through the required local homomorphism. Conversely a centre supplies this containment. For a principal open D(h)D(h), the centre lies in D(h)D(h) exactly when h−1∈Vh^{-1}\in V. Thus restriction to scheme opens gives the usual basic open subspaces of the valuation space. Affine coverings of intersections therefore provide compatible gluing maps. On a nonseparated scheme the same valuation may have distinct centres; the gluing retains these as distinct points unless an overlap identifies them.

For a point (V,z)(V,z) of Val⁡(Z,F)\operatorname{Val}(Z,F), apply the valuative criterion [StacksAlg] (Tag 0BX5) to the specified generic map and Spec⁡V→Z\operatorname{Spec}V\to Z. It gives exactly one lift to Z′Z'. Its image is contained in the closure of the specified generic point. The morphism factors through the reduced closure: a section of its defining radical ideal vanishes at the generic point of the integral scheme Spec⁡V\operatorname{Spec}V, and hence vanishes identically. This proves bijectivity on centred valuations.

The map is a local homeomorphism. Around any lifted centre choose affine opens Spec⁡R′⊂Z′\operatorname{Spec}R'\subset Z' and Spec⁡R⊂Z\operatorname{Spec}R\subset Z with the first mapping into the second. The induced valuation map is the inclusion of the open Pk(F){R′}\mathscr P_k(F)\{R'\} into Pk(F){R}\mathscr P_k(F)\{R\}. Here the images of the coordinate rings in FF are understood; they are finitely generated, so these are basic quasi-compact opens. A bijective local homeomorphism is a homeomorphism, and the field map is the identity. This proves the birational-space isomorphism. ◻

Proposition 3.19.24. NAG-001987 (Special fibres of admissible modifications). Let KK be nontrivially valued. An admissible formal blowup induces a locally projective morphism of special fibres over K~\widetilde K, and hence a proper morphism. A finite composite of admissible formal blowups induces a proper morphism as well. For a fixed point of the common generic fibre, the induced map between the reduced closures of its specializations is proper.

Proof. Work on an affine formal open Spf⁡(A0)\operatorname{Spf}(A_0), and write the open ideal as I=(f0,…,fm)I=(f_0,\ldots,f_m). The algebraic blowup of Spec⁡A0\operatorname{Spec}A_0 in II is projective by [StacksAlg] (Tag 02NS). In the notation of Theorem 3.15.1, its affine chart is Di=A0[I/fi]D_i=A_0[I/f_i] and the formal chart is Bi=D^iB_i=\widehat D_i. Both are π\pi-torsion-free. Lemma 3.14.8 identifies Di/πDiD_i/\pi D_i with Bi/πBiB_i/\pi B_i. Since π∈K∘∘\pi\in K^{\circ\circ}, taking the further quotient by K∘∘/(π)K^{\circ\circ}/(\pi) gives Di/K∘∘Di≃Bi/K∘∘Bi.D_i/K^{\circ\circ}D_i\simeq B_i/K^{\circ\circ}B_i. These identifications respect the fractional chart transitions. Thus the special fibre of the formal blowup is the special fibre of the algebraic blowup. A local closed embedding of the latter into a projective space remains a closed embedding after reduction. The special-fibre map is consequently locally projective and is proper by [StacksAlg] (Tag 01WC). Finite composites are proper by [StacksAlg] (Tag 01W3).

Let Z′Z' and ZZ be the reduced closures of the two specializations. The map Z′→XsZ'\to\mathfrak X_s factors through ZZ: a function vanishing on ZZ vanishes at the specified generic point of Z′Z', and therefore vanishes on the reduced integral Z′Z'. Locally on ZZ, the scheme Z′Z' is a closed subscheme of the base change of the preceding projective morphism. It is therefore locally projective over ZZ, hence proper by the same criterion. Applying this at each blowup gives the assertion for any finite admissible modification. ◻

Actual generic atlases for the compact common-model comparison. Let UU be the quasi-compact quasi-separated strict neighborhood in the formal-model realization of a strict germ. Write the two admissible models as X,Y\mathfrak X,\mathfrak Y, with their actual formal-affine generic charts carrying the identified generic fibres to UU by iX,iYi_X,i_Y. Each model has a finite compatible formal affine atlas. Every formal open and every stage of an actual finite sequence of admissible blowups of either compact model has such an atlas, with actual generic maps for the formal arrows. The generic arrow of each admissible blowup is the canonical isomorphism on its rational charts.

Proof. For either given model, the actual affine generic charts cover UU admissibly. Choose a finite point-cover by Lemma 2.39.2, retaining a finite germ-cover witness at xx if one is specified. The given principal traces and affine specialization identify these charts with the full inverse images of their formal affines. Each affine specialization is surjective by Theorem 3.16.15. The same contraction on the overlaps glues these specializations, so the corresponding formal affines cover the model. This gives the required finite-incidence formal cover. The finite family of analytic charts is still admissible: each chart meets every test affinoid in a compact subset by quasi-separatedness, hence in a closed subset of that Hausdorff test chart. The quasi-net topology test of Lemma 2.40.2 gives closedness in UU, and a finite closed cover is a quasi-net. The finite-incidence generic-fibre construction on this cover recovers the given generic fibre. On the actual common principal refinements the bounded chart identities agree with the supplied ones; the identities and their inverses glue. These comparisons also preserve the marked point and its completed residue field.

For any compact model in the construction, a formal open is quasi-compact. Intersect it with each member of a finite affine cover and use the topological Noetherian clause of NAG-001944 on that affine. Each resulting intersection has a finite principal cover there. An admissible blowup of the compact model has a finite formal affine cover: restrict to each original affine and use the finitely many affine blowup charts for a finite generating list of the actual open ideal. The actual restrictions come from the formal existence and restriction clause of NAG-001945 and Theorem 3.15.1. This proves the compactness assertion by induction for finite composites. It includes the blowups of the successively pulled-back centres, since the actual pulled-back open ideals are finitely generated and contain a power of π\pi. A finite product of the finitely many extended centres at a compact stage is again an open ideal. All of these statements precede any assertion about generic invariance.

Apply the finite-incidence construction to each resulting finite cover. On formal intersections it gives actual bounded principal identifications with their triple cocycle, and on formal arrows it gives actual analytic maps, checked on principal pieces subordinate to target affines. Thus for a blowup arrow the atlas on its source and the atlas on its target are compatible with the actual formal map. On an original affine, the actual blowup charts are the rational domains prescribed by Theorem 3.15.2. Their transition maps are the bounded rational restrictions. These are the same restrictions on the common principal pieces used in the finite-incidence construction. The supplied-atlas invariance of NAG-001945 therefore applies to this pair of atlases. It identifies the actual generic blowup arrow, an isomorphism, with the maps on these rational charts. This step supplies the isomorphism only after the underlying compact formal scheme and its compatible generic maps have been constructed. ◻

Application to common modification. In cover realization use these atlases on the affine local blowups for the rational domains, on the product blowups and on the model obtained by extending the local centres. The formal opens produced have the actual restricted principal atlases. In morphism lifting use those same atlases at the cover-realization stage, on the affine lifting blowups and on the second product blowup. The comparison of the formal lifts is on the actual generic restrictions of the given analytic arrow. In common modification the same construction applies to the lifts in each direction, the successive pulled-centre blowups and all the comparison arrows. Thus the hypotheses of the supplied-atlas NAG-001947–001949 proofs are supplied at each use. No application of model existence or localization is needed. Take the specified isomorphism in NAG-001949 to be iY−1iXi_Y^{-1}i_X. Its conclusion, with the generic blowup isomorphisms just verified, gives iX(bX)η=iY(bY)ηi_X(b_X)_\eta=i_Y(b_Y)_\eta as isomorphisms from a single actual generic fibre to UU.

Pointed reduction of a supplied compact formal model. Let KK be the standing complete nontrivially valued field. Let UU be a quasi-compact quasi-separated strict analytic neighborhood of xx, with a supplied admissible formal model U\mathfrak U and its compatible formal-affine generic atlas on UU. Here compatibility includes the actual bounded restriction maps on the common principal traces of formal affine opens. Put k=K~k=\widetilde K, F=H(x)~grF=\widetilde{\mathcal H(x)}^{\rm gr}, F0=H(x)~F_0=\widetilde{\mathcal H(x)}, and P=PK~gr(F),P0=Pk(F0).P=\mathscr P_{\widetilde K^{\rm gr}}(F),\qquad P_0=\mathscr P_k(F_0). Let ZZ be the reduced closure in Us\mathfrak U_s of the image of the specified residual evaluation Spec⁡F0→Us\operatorname{Spec}F_0 \to\mathfrak U_s. Then there is a canonical isomorphism over PP Val⁡(Z,F0)×P0P≃R(U,x).\operatorname{Val}(Z,F_0)\times_{P_0}P \simeq \mathscr R(U,x). With the strict ordinary reduction defined by the ordinary chart gluing, this comparison gives Val⁡(Z,F0)≃R0(U,x)\operatorname{Val}(Z,F_0)\simeq\mathscr R_0(U,x). If a common model with finite admissible blowup maps to two such models is supplied, assume its actual generic maps induce the same identification with UU. Then the two proper pointed maps induce the ordinary chart comparisons over P0P_0.

Proof. On each formal affine Qi=Spf⁡(Ai)Q_i=\operatorname{Spf}(A_i), take the affine specialization from Theorem 3.16.15. These maps agree on the supplied principal trace identifications: a formal principal restriction inverts ff, its generic trace is ∣f∣=1\lvert f\rvert=1, and its residual point map is the restriction of the same evaluation. Thus the specialization sp⁡\operatorname{sp} on this supplied model is surjective and the generic chart Di=M(Ai[1/π])D_i=\mathcal M(A_i[1/\pi]) is precisely the inverse image of QiQ_i. The latter assertion follows on every other chart by the principal trace description of the formal overlaps. Choose finitely many DiD_i covering UU by Lemma 2.39.2, and include the finitely many chart members through xx whose union contains a neighborhood of xx. Surjectivity now shows that the corresponding finite formal affines cover U\mathfrak U. The selected chart members through xx cover its germ. In particular U\mathfrak U is quasi-compact, with a finite-incidence affine cover.

The image of Spec⁡F0\operatorname{Spec}F_0 is s=sp⁡(x)s=\operatorname{sp}(x), the generic point of ZZ. A formal chart meets ZZ on the special fibre exactly when it contains ss, which is exactly when DiD_i contains xx. On such a chart, the reduced closure is Spec⁡Ri\operatorname{Spec}R_i, where Ri=im⁡(Ai/K∘∘Ai→F0)R_i=\operatorname{im}(A_i/K^{\circ\circ}A_i\to F_0): the kernel of evaluation in the field is prime and defines the reduced closure of the image point. These finitely many charts cover ZZ; hence ZZ is integral and of finite type over kk.

For each pair of formal affines, their intersection has a finite cover by opens principal in both. In fact, around any point of the intersection take a principal open in the first affine and then a principal open of the second inside it. In the first principal open, the equation of the second has reduction modulo π\pi of the form a/fna/f^n, by Proposition 3.17.1. Clearing the denominator gives a principal open in the first affine as well. Each formal affine is topologically Noetherian by the formal-topology clause of NAG-001944, so finitely many such opens cover the intersection. Repeat the same refinement for a third affine. The formal restriction maps on these actual principal opens satisfy the triple cocycle.

Keep the pieces meeting ZZ, or equivalently containing ss. On a piece principal in QiQ_i, with equation ff, one has fˉ≠0\bar f\ne 0 in RiR_i and the residual algebra image is Ri[1/fˉ]R_i[1/\bar f] in F0F_0. This again follows from the principal localization formula, first modulo π\pi and then over kk. Its analytic trace contains xx, and the principal traces retained this way cover the corresponding analytic overlap germ. Indeed the entire overlap is covered by finitely many compact rational traces which are closed in the affinoid DiD_i. Their finite cover is admissible, and the members through xx cover a relative neighborhood. The same argument applies to the triple overlap.

By Proposition 3.19.21, the actual graded reduction of the chart DiD_i at xx is the inverse image of Pk(F0){Ri}⊆P0\mathscr P_k(F_0)\{R_i\}\subseteq P_0. The identical assertion for the retained principal trace uses Ri[1/fˉ]R_i[1/\bar f], with the same completed residue field by Corollary 2.33.15. Thus these are also the exact transition opens after pulling back to PP. On the scheme side they are the valuation loci with centre in the actual principal pieces, by Lemma 3.19.23. On the analytic side the actual graded overlap is the union of the indicated graded opens by Theorem 3.19.9. The two open gluings and their triple transitions agree over PP. A valuation with distinct centres is identified only when the common open identifies the centres.

For the ordinary comparison, use the strict ordinary chart construction. Use these same charts and principal overlaps to compute the strict ordinary gluing. On every chart both maps are the identity of Pk(F0){Ri}\mathscr P_k(F_0)\{R_i\}, and on overlaps both identify the identical principal valuation loci. Thus this ordinary comparison is compatible with further common refinements. It preserves the specified field F0F_0.

Finally, let bX,bYb_X,b_Y be the supplied finite composites with common formal source W\mathfrak W. Give W\mathfrak W its actual compatible formal-affine generic atlas, with both generic arrows induced by the given formal maps. On the compact models these atlases are supplied by the finite-incidence construction and the compact-model atlas. Apply the supplied-atlas invariance of NAG-001945 at each blowup stage. By the required equality of the generic identifications, j=iX(bX)η=iY(bY)η ⁣:Wη→∼U.j=i_X(b_X)_\eta=i_Y(b_Y)_\eta \colon \mathfrak W_\eta\xrightarrow{\sim} U . In particular w=j−1(x)w=j^{-1}(x) is a single point above xx on both legs, with the same canonically identified F0F_0.

The pointed maps on special fibres commute with evaluation in F0F_0: on compatible affine charts they are induced by the formal coefficient homomorphisms, followed by reduction of the same evaluation at ww. Each maps the reduced pointed closure into the corresponding reduced pointed closure downstairs and is proper by Proposition 3.19.24. Apply Lemma 3.19.23 to each leg. Each resulting isomorphism keeps the valuation and sends its centre to its image.

To compare these isomorphisms with the ordinary chart gluing, refine the chart net on Wη\mathfrak W_\eta by the actual principal traces of a formal affine refinement subordinate simultaneously to the two inverse images of target affine covers. Such a refinement is obtained by taking principal opens inside the pairwise inverse images on each source affine. The formal topology clause gives finite principal refinements there, and the supplied trace construction gives their admissible generic covers. Keep the pieces through ww; they cover a neighborhood in that net. Under the single isomorphism jj they are a common strict affinoid refinement at xx for the two generic atlases, with their actual bounded restriction maps. On every retained piece, both induced maps of valuation charts are the inclusions over the identity of the represented valuations of F0F_0. Their centres contract along the corresponding formal maps. These are precisely the maps in the ordinary chart gluing. The comparison on overlaps and its compatibility with any further such supplied modification follow by composing these same maps on the common principal refinements. ◻

Theorem 3.19.25. NAG-001988 (Formal-model realization of a strict germ). Let UU be a quasi-compact quasi-separated strictly analytic neighbourhood of xx over a nontrivially valued field KK, and let U\mathfrak U be any admissible formal model of UU. Write F0=H(x)~F_0=\widetilde{\mathcal H(x)}, and let ZZ be the reduced closure in Us\mathfrak U_s of the image of Spec⁡F0→Us\operatorname{Spec}F_0\to\mathfrak U_s. Then there is a canonical isomorphism Val⁡(Z,F0)≃R0(U,x).\operatorname{Val}(Z,F_0)\simeq\mathscr R_0(U,x). It is independent of the formal model and the neighbourhood. Common admissible modifications realize the comparison through proper pointed residual schemes, inducing the identity on F0F_0.

Proof. A formal model of UU, in the standing convention of Formal models and their two fibres, is given with an identification of its analytic generic fibre with UU. The generic fibre in that convention is the gluing of the actual formal-affine generic charts along their analytic domains. Transport this gluing to UU by the specified identification. Its formal-affine generic charts thus give the compatible admissible chart family on UU. On the formal principal opens the restriction maps are the bounded maps of Proposition 3.17.1; on intersections they are checked on common principal refinements. The underlying specialization maps on the affine charts are those of Theorem 3.16.15 and agree there by contraction. This uses the generic fibre of the given model as data. Its finite chart cover and the atlases for its modifications are obtained by the compact common-model atlas argument. The principal traces and contraction identify the full inverse images of the formal affine opens and the specialization maps on their overlaps.

Apply the pointed reduction of a compact formal model to this actual atlas and evaluation at xx. It identifies the ordinary strict chart gluing with the valuation space of the reduced pointed closure ZZ, over the specified field F0F_0. The ordinary neighbourhood invariance identifies the ordinary chart gluings after shrinking representatives of the analytic germ at xx. The identifications use the same completed residue field at xx and compose on further common refinements.

For two models of the same UU, use the compact-model atlas and common modification with the analytic isomorphism specified by their identifications with UU. It supplies the compatible atlases on both legs and their common source, and the two actual generic maps from that source agree as isomorphisms to UU. Thus the two legs lift xx to the same point with the same F0F_0. The supplied-leg part of the pointed reduction gives proper maps of the reduced pointed closures. On valuations they keep the field and the valuation and contract the centre. By the common principal refinements they are the ordinary chart comparisons already used above, including after further such modifications. ◻

References. The ordinary realization reconstructs the formal-model description of germ reduction in [Tem00] (Section 1.5). The proof uses the internal strict/graded comparison, the internal Raynaud reconstruction, and only the displayed Stacks algebraic properness inputs.

3.20 Tubes, line bundles and elementary formal models

In this section KK is complete and nontrivially valued, O=K∘O=K^\circ, and k=K~k=\widetilde K. The following arguments make explicit the local constructions used in the examples of this chapter. Their formal and analytic gluing inputs have already been established.

Proposition 3.20.1. NAG-001798 (Tubes of locally closed strata). Let X\mathfrak X be an admissible formal OO-scheme, X=XηX=\mathfrak X_\eta, and Z⊂XsZ\subset\mathfrak X_s a locally closed subset. Then ] ⁣Z[X=sp⁡−1(Z)]\!Z[_{\mathfrak X}=\operatorname{sp}^{-1}(Z) is an analytic domain of XX, canonically with its induced analytic structure. On an affine formal chart it has a finite cover by sets of the form {x:∣fi(x)∣<1 (1≤i≤m),∣g(x)∣=1},fi,g∈OX(U).\{x:|f_i(x)|<1\ (1\leq i\leq m),\quad |g(x)|=1\}, \qquad f_i,g\in\mathcal O_{\mathfrak X}(\mathfrak U). This construction commutes with intersections and inverse images under formal morphisms.

Proof. On U=Spf⁡(A0)\mathfrak U=\operatorname{Spf}(A_0), every integral element has absolute value at most one by Proposition 3.14.7. Thus evaluation induces A0/K∘∘A0→H(x)~A_0/K^{\circ\circ}A_0\to\widetilde{\mathcal H(x)}; equivalently the source is A0⊗OkA_0\otimes_O k. The special fibre is of finite type over kk, hence Noetherian by [StacksAlg] (Tag 00FN). Write Z∩UsZ\cap\mathfrak U_s as a finite union of V(fˉ1,…,fˉm)∩D(gˉ)V(\bar f_1,\ldots,\bar f_m)\cap D(\bar g), with lifts in A0A_0. The definition of specialization gives exactly the displayed absolute-value conditions.

The inequalities ∣fi∣<1|f_i|<1 define an ordinary open subset WW of Uη\mathfrak U_\eta. The locus ∣g∣=1|g|=1 is the Laurent affinoid domain ∣g∣≥1|g|\geq 1, because ∣g∣≤1|g|\leq 1 everywhere. Their intersection is an analytic domain by Proposition 2.40.5. To verify the finite-union assertion, first take the common open subset WW cut out by the equations for the closed part of ZZ, and write the open part as a finite union of principal opens. Each of the resulting domains is relatively closed in their union. At a point, omit the finitely many closed pieces not containing it, and take small rational neighborhoods inside the remaining pieces. They give the quasi-net required by Definition 2.40.4. Hence the union is an analytic domain with its induced topology.

The formal affine generic fibres form the analytic net of Proposition 3.17.2. The local domains just constructed therefore glue: their intersections agree as subsets, and analytic-domain structures are unique by Corollary 2.40.9 and chartwise domain calculus. The same uniqueness proves compatibility with intersections. Functoriality of specialization gives the asserted inverse-image identity, and domain base change supplies its analytic structure. The equations and the finite covers used above are local on formal affine charts; no global quasi-compactness of ZZ is assumed. ◻

Proposition 3.20.2. NAG-001799 (The generic fibre of a formal line bundle). An invertible sheaf L\mathfrak L on an admissible formal scheme X\mathfrak X has a canonical invertible OXG\mathcal O_{X_G}-module Lη\mathfrak L_\eta on X=XηX=\mathfrak X_\eta. This construction commutes with tensor products, duals and formal pullback. Its formal transition functions have absolute value one.

Proof. Choose a formal affine open cover Ui\mathfrak U_i with frames eie_i of L\mathfrak L. On overlaps write ej=uijeie_j=u_{ij}e_i, where uiju_{ij} is a formal unit. By Proposition 3.17.2, the domains Ui=(Ui)ηU_i=(\mathfrak U_i)_\eta form an admissible analytic cover, and the functions uiju_{ij} restrict to analytic units on Ui∩UjU_i\cap U_j. They satisfy uik=uijujku_{ik}=u_{ij}u_{jk} on triple overlaps. For an analytic domain V⊂XV\subset X, let L(V)\mathcal L(V) consist of families of analytic coefficients aia_i on V∩UiV\cap U_i satisfying ai=uijaja_i=u_{ij}a_j on V∩Ui∩UjV\cap U_i\cap U_j. The structure-sheaf gluing theorem 2.40.6 makes this a sheaf of OXG\mathcal O_{X_G}-modules. On UiU_i it is freely generated by eie_i: a coefficient there determines every other local coefficient by the displayed transition relation. Thus it is an invertible sheaf.

A change of formal frames multiplies these coefficients by the corresponding formal units and gives a canonical isomorphism. Two covers have their common formal refinement; the isomorphisms just described agree there and hence globally. This proves independence. Products, inverses and pullbacks of the cocycles give respectively tensor products, duals and pullback, with the usual compatibility identities holding coefficientwise. Finally uiju_{ij} and uij−1u_{ij}^{-1} are integral on every formal affine subchart of the overlap, so both have absolute value at most one. It follows that ∣uij∣=1|u_{ij}|=1 everywhere on the analytic overlap. ◻

Lemma 3.20.3. NAG-001800 (Canonical reduction of the standard annulus). Let a∈Ka\in K satisfy 0<∣a∣<10<|a|<1, and put A0=O⟨U,V⟩/(UV−a),A=K⟨U,V⟩/(UV−a).A_0=O\langle U,V\rangle/(UV-a),\qquad A=K\langle U,V\rangle/(UV-a). Then A0A_0 is admissible, A0=A∘A_0=A^\circ, and A~=A0⊗Ok=k[u,v]/(uv).\widetilde A=A_0\otimes_O k=k[u,v]/(uv). The Shilov boundary consists of the Gauss points with ∣U∣=1|U|=1 and ∣U∣=∣a∣|U|=|a|. Likewise the unit-circle algebra K⟨T,T−1⟩K\langle T,T^{-1}\rangle has canonical reduction k[t,t−1]k[t,t^{-1}] and a single Shilov point.

Proof. Apply Theorem 2.13.2 to the interval [∣a∣,1][|a|,1], rescaling its inverse coordinate by aa. Every element of AA has a unique expression f=∑n≥0cnUn+∑n≥1dnVn,cn⟶0,dn⟶0.f=\sum_{n\geq0}c_nU^n+\sum_{n\geq1}d_nV^n, \qquad c_n\longrightarrow0,\quad d_n\longrightarrow0. Its quotient norm is N(f)=max⁡{max⁡n∣cn∣,max⁡n≥1∣dn∣}N(f)=\max\{\max_n|c_n|,\max_{n\geq 1}|d_n|\}. The same normal-form operators preserve integral coefficients: on a monomial one replaces UiVjU^iV^j by amin⁡(i,j)Ui−ja^{\min(i,j)}U^{i-j} or amin⁡(i,j)Vj−ia^{\min(i,j)}V^{j-i}; the corresponding quotient by UV−aUV-a has integral coefficients. The operators are contractive after this rescaling, so the polynomial identities pass to restricted series. Hence A0A_0 embeds in AA and is exactly the subset N(f)≤1N(f)\leq 1. In particular it has no OO-torsion and is admissible by Theorem 3.14.4.

At any analytic point, ∣U∣,∣V∣≤1|U|,|V|\leq 1, so ∣f∣≤N(f)|f|\leq N(f). At the two endpoint Gauss points, the respective values are max⁡{max⁡n∣cn∣,max⁡n≥1∣dn∣∣a∣n},max⁡{max⁡n∣cn∣∣a∣n,max⁡n≥1∣dn∣}.\max\{\max_n|c_n|,\max_{n\geq1}|d_n||a|^n\},\qquad \max\{\max_n|c_n||a|^n,\max_{n\geq1}|d_n|\}. Their maximum is N(f)N(f). Thus N=ρAN=\rho_A and A0=A∘A_0=A^\circ. A nonzero element with N(f)<1N(f)<1 has a coefficient attaining this maximum, so it is an element of K∘∘A0K^{\circ\circ}A_0; the converse is immediate. Therefore A∘∘=K∘∘A0A^{\circ\circ}=K^{\circ\circ}A_0. Reducing the unique normal forms gives k[u,v]/(uv)k[u,v]/(uv). The endpoint pair is a boundary, and neither point can be omitted from this pair: UU attains its maximum only at the outer endpoint of the pair, and VV only at the inner one. Lemma 3.12.4 therefore makes the pair the Shilov boundary. For the unit circle the same normal form, now with interval [1,1][1,1], gives the multiplicative Laurent Gauss norm and reduction k[t,t−1]k[t,t^{-1}]. Proposition 3.4.1 gives its single Shilov point. ◻

Corollary 3.20.4. NAG-001801 (The elementary formal blowup and its weight). For 0<∣a∣<10<|a|<1, the admissible blowup of (T,a)(T,a) on Spf⁡O⟨T⟩\operatorname{Spf}O\langle T\rangle has generic domains ∣T∣≤∣a∣|T|\leq|a| and ∣a∣≤∣T∣≤1|a|\leq|T|\leq 1. Its special fibre consists of the original affine line and one projective line meeting at a point. The invertible pulled-back ideal has canonically trivial generic fibre, and its model weight for the section 11 is log⁡max⁡(∣T∣,∣a∣)\log\max(|T|,|a|).

Proof. The ideal is open since aa divides a sufficiently large power of any fixed topologically nilpotent π\pi. Theorem 3.15.1 gives the charts O⟨U⟩O\langle U\rangle, with T=aUT=aU, and O⟨T,V⟩/(TV−a)O\langle T,V\rangle/(TV-a), with V=a/TV=a/T. Their formal overlap has UU and VV inverse units. The first special chart is Ak1\mathbb A^1_k with coordinate uu; the second is tv=0tv=0, as in Lemma 3.20.3. Gluing the t=0t=0 component to the first chart by u=v−1u=v^{-1} gives Pk1\mathbb P^1_k. The v=0v=0 component remains the original affine line, meeting it at t=v=0t=v=0. Theorem 3.15.2 identifies the two generic domains and their gluing with the original disc. Proposition 3.20.2 constructs the generic line bundle of the invertible ideal, and Proposition 3.11.1 gives its model metric. On the first chart the ideal is generated by aa, and on the second by TT. Its generic ideal is the unit ideal because a∈K×a\in K^\times. Thus 1=a−1a1=a^{-1}a has norm ∣a∣−1|a|^{-1} on the first chart and 1=T−1T1=T^{-1}T has norm ∣T∣−1|T|^{-1} on the second. Taking minus the logarithm gives the stated weight, and the formulas agree on the overlap. ◻

Corollary 3.20.5. NAG-001802 (Disc and projective-line reductions at the Gauss point). At the unit Gauss point xx, the ordinary germ reduction of the closed unit disc is represented by Ak1\mathbb A^1_k pointed by k(t)k(t); that of P1,an\mathbb P^{1,\mathrm{an}} is represented by Pk1\mathbb P^1_k pointed by the same field. The disc inclusion induces the usual open immersion of these models, which is not proper in the pointed birational category.

Proof. The two formal affine lines Spf⁡O⟨T⟩\operatorname{Spf}O\langle T\rangle and Spf⁡O⟨S⟩\operatorname{Spf}O\langle S\rangle glue along the principal formal opens TS=1TS=1. Their special fibre is Pk1\mathbb P^1_k. By Proposition 3.17.1, their generic fibres glue along ∣T∣=1|T|=1; the result is P1,an\mathbb P^{1,\mathrm{an}}, since its two closed unit discs cover it, as proved in Proposition 2.33.30, and the coordinate transition is S=T−1S=T^{-1}. The unit Gauss completed residue field has residue k(t)k(t) by Proposition 3.8.1. Its specialization is the generic point of each of these residual models. The closure is therefore Ak1\mathbb A^1_k for the disc and Pk1\mathbb P^1_k for the projective line. Apply the fixed-model comparison to the whole disc and the whole projective line, each as a neighbourhood of xx in its own space. The disc is the very same pointed TT-chart in the projective-line model. The ordinary chart gluing therefore gives the inclusion via Ak1↪Pk1\mathbb A^1_k\hookrightarrow\mathbb P^1_k. Finally the valuation ring k[t−1](t−1)k[t^{-1}]_{(t^{-1})} in k(t)k(t) has a centre on Pk1\mathbb P^1_k but none on Ak1\mathbb A^1_k, since it does not contain tt. Thus the corresponding valuation-space map is not surjective, and the inclusion is not proper in the pointed birational category. By Proposition 2.44.5, the relative interior of this analytic-domain inclusion is its ordinary interior in P1,an\mathbb P^{1,\mathrm{an}}. The Gauss seminorms of radii r>1r>1 on the affine TT-line (Proposition 1.2.2) lie outside the disc and converge to xx as r↓1r\downarrow 1, since the Gauss maximum of each polynomial is continuous in rr. Thus xx lies in the relative boundary of the inclusion. ◻

3.21 Dimension and the complete spectral norm

We complete two parts of the affinoid algebra theorem: the dimension in a finite normalization, and the equivalence of the spectral norm with the affinoid norm on a reduced algebra. For the second assertion the positive-characteristic argument passes through the completion of the Gauss fraction field.

Proposition 3.21.1. NAG-001657 (Dimension of a strict normalization). For every complete non-Archimedean field KK, dim⁡K⟨T1,…,Tn⟩=n\dim K\langle T_1,\ldots,T_n\rangle=n. If K⟨T1,…,Td⟩↪AK\langle T_1,\ldots,T_d\rangle\hookrightarrow A is a finite injection into a nonzero strict affinoid algebra, then d=dim⁡Ad=\dim A.

Proof. Put Rn=K⟨T1,…,Tn⟩R_n=K\langle T_1,\ldots,T_n\rangle. The coordinate ideals (0)(0), (T1)(T_1), …, (T1,…,Tn)(T_1,\ldots,T_n) form a strictly increasing chain of primes: each quotient is a Tate algebra and its Gauss norm is multiplicative. Thus dim⁡Rn≥n\dim R_n\geq n. For the reverse inequality induct on nn, starting with the field R0=KR_0=K. Given a nonzero prime p⊂Rn\mathfrak p\subset R_n, choose 0≠f∈p0\ne f\in\mathfrak p. Lemma 1.11.3 and Theorem 1.10.3 show, after an isometric coordinate change, that Rn/(f)R_n/(f) is finite over Rn−1R_{n-1}. The same holds for Rn/pR_n/\mathfrak p. The dimension inequality for an integral map, [StacksAlg] (Tag 00OJ), gives dim⁡(Rn/p)≤n−1\dim(R_n/\mathfrak p)\leq n-1. Any prime chain in the domain RnR_n can be preceded by (0)(0); after its first nonzero member its length is at most n−1n-1. Its total length is therefore at most nn. Finally, an integral injection preserves dimension by [StacksAlg] (Tag 00OK). Apply this to the displayed finite injection. ◻

Lemma 3.21.2. NAG-001658 (Coordinates on a Banach space of countable type). Let VV be a Banach KK-space with a countable set whose linear span is dense. For every 0<c<10<c<1 there is a finite or countable family (ei)(e_i) such that every element has a unique convergent expansion ∑iaiei\sum_i a_i e_i and cmax⁡i∣ai∣∥ei∥ ≤ ∥∑iaiei∥ ≤ max⁡i∣ai∣∥ei∥.c\max_i |a_i|\|e_i\| \ \leq\ \left\|\sum_i a_i e_i\right\| \ \leq\ \max_i |a_i|\|e_i\|. In particular its coordinate functionals are bounded and separate points.

Proof. Enumerate a dense spanning set as v1,v2,…v_1,v_2,\ldots, and choose 0<cj<10<c_j<1 with ∏j≥1cj>c\prod_{j\geq 1}c_j>c. Process the vjv_j in order. If vjv_j lies in the span WW of the vectors already chosen, add nothing. Otherwise WW is finite dimensional, hence complete and closed by Lemma 1.12.1. Its distance δ=inf⁡w∈W∥vj−w∥\delta=\inf_{w\in W}\|v_j-w\| is positive. Choose wj∈Ww_j\in W with ∥vj−wj∥<δ/cj\|v_j-w_j\|<\delta/c_j, and adjoin e=vj−wje=v_j-w_j. For w∈Ww\in W and a∈Ka\in K, the distance estimate and the ultrametric inequality give ∥w+ae∥≥cjmax⁡{∥w∥,∣a∣∥e∥}.\|w+ae\|\geq c_j\max\{\|w\|,|a|\|e\|\}. Indeed, if ∥w∥>∣a∣∥e∥\|w\|>|a|\|e\| the left side equals ∥w∥\|w\|; otherwise it is at least ∣a∣δ|a|\delta. Induction proves the stated two-sided estimate for finite sums. The weighted coefficient space of Lemma 2.13.1 is complete, so the sum map extends to a bounded isomorphism onto a closed subspace of VV. Its image contains every vjv_j and is therefore all of VV. The lower estimate proves uniqueness and the bound ∣λi(v)∣≤c−1∥ei∥−1∥v∥|\lambda_i(v)|\leq c^{-1}\|e_i\|^{-1}\|v\| for each coordinate. The zero space has the empty family. ◻

Theorem 3.21.3. NAG-001659 (Separable completion of the Gauss fraction field). Let R=K⟨T1,…,Tn⟩R=K\langle T_1,\ldots,T_n\rangle, let F=Frac⁡(R)F=\operatorname{Frac}(R) carry the Gauss absolute value, and let HH be its completion. Then H/FH/F is geometrically reduced. In particular, for every finite field extension E/FE/F, the finite-dimensional HH-algebra E⊗FHE\otimes_F H is reduced.

Proof. In characteristic zero every field extension is separable, hence geometrically reduced; this is the characteristic-zero case of [StacksAlg] (Tags 030Z and 00I4). Assume char⁡K=p>0\operatorname{char}K=p>0. Endow K1/pK^{1/p} with ∣a∣=∣ap∣1/p|a|=|a^p|^{1/p}; it is complete, since taking ppth powers identifies Cauchy sequences and their limits with those in KK. Put Si=Ti1/pS_i=T_i^{1/p}. Coefficientwise Frobenius identifies R1/p=K1/p⟨S1,…,Sn⟩,F1/p=Frac⁡(R1/p).R^{1/p}=K^{1/p}\langle S_1,\ldots,S_n\rangle,\qquad F^{1/p}=\operatorname{Frac}(R^{1/p}). We prove that every finite FF-linearly independent family u1,…,um∈F1/pu_1,\ldots,u_m\in F^{1/p} remains linearly independent over HH in H1/pH^{1/p}. Multiplying the family by a common nonzero element of RR puts all its members in R1/pR^{1/p}: write uip=ai/biu_i^p=a_i/b_i with ai,bi∈Ra_i,b_i\in R and use ∏ibi\prod_i b_i. This multiplication preserves both kinds of independence.

Let K′K' be the closure, in K1/pK^{1/p}, of the KK-algebra generated by the coefficients of the finitely many uiu_i. This algebra is a field: any nonzero element aa has inverse ap−1/apa^{p-1}/a^p, with ap∈Ka^p\in K. Its closure is again a field, and the countably many monomials in its generators have dense KK-linear span. Apply Lemma 3.21.2 to K′K'. Its bounded coordinate functionals λj:K′→K\lambda_j:K'\to K, together with the unique decomposition of exponents modulo pp, define bounded RR-linear maps Λj,μ:K′⟨S⟩⟶R,∑νaνSν⟼∑η∈Nnλj(aμ+pη)Tη,0≤μi<p.\Lambda_{j,\mu}:K'\langle S\rangle\longrightarrow R,\qquad \sum_\nu a_\nu S^\nu\longmapsto \sum_{\eta\in\mathbf N^n}\lambda_j(a_{\mu+p\eta})T^\eta, \quad 0\leq\mu_i<p. The series on the right converges, its Gauss norm is at most ∥λj∥op\|\lambda_j\|_{\rm op} times the norm of the original series, and the family of all Λj,μ\Lambda_{j,\mu} separates points. RR-linearity follows first for polynomial coefficients and then by continuity. The matrix whose columns are all the coordinates Λj,μ(ui)\Lambda_{j,\mu}(u_i) has independent columns over FF: a relation, after clearing its denominators in RR, would give an element of K′⟨S⟩K'\langle S\rangle with all coordinates zero. Finite-dimensional linear algebra therefore supplies mm rows with invertible m×mm\times m matrix over FF. Its inverse and the bounds just obtained give a constant C>0C>0 such that max⁡i∣bi∣F≤C∣∑ibiui∣F1/p(b1,…,bm∈F).\max_i |b_i|_F \leq C\left|\sum_i b_i u_i\right|_{F^{1/p}} \qquad (b_1,\ldots,b_m\in F). For coefficients in RR this follows directly from the selected rows; for coefficients in FF clear one common denominator and use multiplicativity of the Gauss norm. The field H1/pH^{1/p} is complete by the same Frobenius argument used for K1/pK^{1/p}. Approximate any hi∈Hh_i\in H by elements of FF and pass to the limit in the last inequality. It proves the asserted HH-linear independence. Hence H⊗FF1/p⟶H1/pH\otimes_F F^{1/p}\longrightarrow H^{1/p} is injective and its source is reduced. [StacksAlg] (Tag 030W, Lemma 10.44.2) now implies that H/FH/F is geometrically reduced. Applying its definition to E/FE/F proves the last assertion. ◻

References. The countable-coefficient argument follows the weak-stability proof for the Gauss fraction field in [BGR84] (§5.3.1, pp. 212–213). The formulation above records the needed geometrically reduced completion directly.

Proposition 3.21.4. NAG-001660 (The spectral kernel and free scalar extension). For every generalized KK-affinoid algebra AA, the kernel of its spectral seminorm ρA\rho_A is its nilradical. If AA is reduced, every free-radius scalar extension ArA_r is reduced as well.

Proof. For a strict affinoid domain BB, use the notation of Proposition 3.12.2. Any maximal ideal of D=E⊗FHD=E\otimes_F H gives a field map from the field EE to its residue field, hence an injection. The resulting bounded point of BB has zero kernel. For a general strict AA, apply this observation to each quotient by a minimal prime. Thus any element vanishing at all analytic points belongs to every minimal prime and is nilpotent by [StacksAlg] (Tag 00E0(4),(7)). Conversely every multiplicative seminorm kills nilpotents.

For arbitrary radii choose a free strictification ArA_r. The Gauss section of Theorem 1.19.1 shows ρAr(a⊗1)=ρA(a)\rho_{A_r}(a\otimes 1)=\rho_A(a). If this value is zero, the strict case makes a⊗1a\otimes 1 nilpotent. The coefficient inclusion A↪ArA\hookrightarrow A_r is injective, so aa is nilpotent. Finally, if AA is reduced and z=∑νaνTν∈Arz=\sum_\nu a_\nu T^\nu\in A_r is nilpotent, evaluation at every Gauss-section point gives max⁡ν∣aν(x)∣rν=0\max_\nu|a_\nu(x)|r^\nu=0. Each aνa_\nu therefore has spectral seminorm zero and is zero by the result just proved. Thus z=0z=0. ◻

Theorem 3.21.5. NAG-001661 (Completeness of the reduced spectral norm). Let AA be a reduced generalized KK-affinoid algebra with its given norm-admissible affinoid norm. Then ρA\rho_A is a complete norm, uniformly equivalent to the given norm. The statement holds for arbitrary positive radii and for a trivially valued ground field.

Proof. First let AA be a nonzero strict domain. Choose a finite normalization R↪AR\hookrightarrow A, put F=Frac⁡(R)F=\operatorname{Frac}(R), and let HH be its Gauss completion. Write E=A⊗RFE=A\otimes_R F. By Theorem 3.21.3, the finite-dimensional algebra D=E⊗FHD=E\otimes_F H is reduced. Every prime of a finite-dimensional algebra over a field is maximal, and there are finitely many maximal ideals; the Chinese remainder theorem and the nilradical formula therefore give D=∏s=1tLsD=\prod_{s=1}^t L_s with finite field extensions Ls/HL_s/H. Its maximum absolute-value norm is complete and equivalent to every HH-coordinate norm by Lemma 1.12.1.

Choose an FF-basis e1,…,eme_1,\ldots,e_m of EE. Since AA is finite over RR, one common nonzero denominator b∈Rb\in R makes all coordinates of bAbA belong to RR. We obtain an injective RR-linear map j:A→Rmj:A\to R^m. The canonical finite-module theorem, and its compatibility with the affinoid norm in Proposition 1.14.4, imply ∥a∥A≤C∥j(a)∥Rm≤C′max⁡s∣a∣Ls(a∈A)\|a\|_A\leq C\|j(a)\|_{R^m} \leq C'\max_s|a|_{L_s}\qquad(a\in A) for constants independent of aa. The second inequality is the finite-dimensional coordinate bound on DD, including the fixed factor ∣b∣|b|. Proposition 3.12.2 identifies the last maximum with ρA(a)\rho_A(a). The reverse inequality ρA(a)≤∥a∥A\rho_A(a)\leq\|a\|_A is the spectral-radius formula. This proves norm equivalence for a strict domain.

For a reduced strict AA, the diagonal map into the finite product of its minimal-prime quotients is an injection of finite AA-modules. It is norm-strict by Theorem 1.11.5. The domain case on the factors and the fact that every point kills a minimal prime give the same two-sided estimate for AA. For arbitrary radii choose a free strictification ArA_r. It is reduced by Proposition 3.21.4. Apply the strict estimate upstairs and restrict to the isometric coefficient inclusion A↪ArA\hookrightarrow A_r. The Gauss section again gives ρAr(a⊗1)=ρA(a)\rho_{A_r}(a\otimes 1)=\rho_A(a), proving the desired estimate. Completeness and definiteness follow from equivalence with the given Banach norm. The zero algebra has the zero Banach space and causes no exception. ◻

References. This is the reduced affinoid norm theorem of [BGR84] (§6.2.4, Theorem 1). The proof above retains finite normalization and the Gauss fraction field, expressing the norm comparison through an explicit finite-module coordinate injection.

Dimension after complete scalar extension

The Krull dimension of a non-strict affinoid algebra need not be its analytic dimension. The following comparison makes the latter independent of the extension used to obtain a strict presentation.

Lemma 3.21.6. NAG-001662 (Scalar invariance of strict affinoid dimension). Let AA be a nonzero strict KK-affinoid algebra and let L/KL/K be any complete valued extension. Then ALA_L is strict and dim⁡AL=dim⁡A\dim A_L=\dim A.

Proof. By Theorem 1.12.2, choose a finite injection R=K⟨T1,…,Td⟩↪AR=K\langle T_1,\ldots,T_d\rangle\hookrightarrow A. Proposition 3.21.1 gives d=dim⁡Ad=\dim A. The finite RR-module topology on AA is its affinoid topology. Indeed finite generators give a bounded surjection Rm→AR^m\to A; when KK is nontrivially valued, Lemma 1.14.1 makes it norm-strict. When KK is trivially valued, the defining strict quotient norms on RR and AA, and the finite-module quotient norm, take only the values zero and one, so they agree. The injection is now norm-strict: the image of the finite RR-submodule RR is closed with its canonical topology by Theorem 1.11.5. The strict exactness of completed scalar tensoring in Theorem 2.33.3 therefore gives an injection RL↪ALR_L\hookrightarrow A_L. The finite-module base-change theorem 2.33.19 shows that this is a finite map. The scalar presentation theorem 2.33.5 identifies RLR_L with L⟨T1,…,Td⟩L\langle T_1,\ldots,T_d\rangle and makes ALA_L strict. Applying Proposition 3.21.1 over LL gives dim⁡AL=d\dim A_L=d. ◻

Lemma 3.21.7. NAG-001663 (Comparing a free strictification with any extension). Let ss be a finite KK-free polyradius and let KsK_s be its Gauss field. For every complete extension L/KL/K, there is a complete field MM with compatible isometric embeddings L↪ML\hookrightarrow M and Ks↪MK_s\hookrightarrow M. One can take the completion of L(T1,…,Tm)L(T_1,\ldots,T_m) for the Gauss absolute value of polyradius ss.

Proof. The weighted Gauss absolute value on a polynomial ring over a valued field is multiplicative by Proposition 1.2.2. It extends to the rational function field by taking quotients. The inclusion K(T)↪L(T)K(T)\hookrightarrow L(T) is isometric because the coefficientwise maximum defining the Gauss value is unchanged. It therefore extends to an isometric map from the completion of K(T)K(T) into MM. This completion is KsK_s: the Laurent monomials are an orthogonal basis in the free-radius case, and a nonzero polynomial has a unique dominant monomial, so its reciprocal is a convergent Laurent geometric series, as in Lemma 1.16.1. The constant-field inclusion L↪ML\hookrightarrow M is also isometric. Both maps restrict to the given inclusion of KK, which proves compatibility. No independence of ss over the larger value group of LL is required. ◻

Theorem 3.21.8. NAG-001664 (Intrinsic analytic dimension and scalar extension). Let AA be a nonzero generalized KK-affinoid algebra. For any complete extension L/KL/K making ALA_L strict, the integer dim⁡AL\dim A_L is independent of LL. Denote it by dim⁡KM(A)\dim_K\mathcal M(A). For every complete extension E/KE/K, dim⁡EM(AE)=dim⁡KM(A).\dim_E\mathcal M(A_E)=\dim_K\mathcal M(A).

Proof. Choose once a free strictification F=KsF=K_s using Theorem 1.17.6, and set d=dim⁡AFd=\dim A_F. Given another strictifying LL, choose the common complete extension MM of FF and LL in Lemma 3.21.7. Associativity of completed affinoid scalar extension 2.33.9 identifies (AF)M(A_F)_M and (AL)M(A_L)_M with AMA_M. Both AFA_F and ALA_L are strict, so Lemma 3.21.6 gives d=dim⁡AF=dim⁡AM=dim⁡AL.d=\dim A_F=\dim A_M=\dim A_L. This proves independence, including strictifying extensions which are not themselves affinoid over KK. For arbitrary E/KE/K, choose a complete extension N/EN/E making (AE)N(A_E)_N strict. It is ANA_N by the same associativity, so its dimension is dd by the independence just proved over KK. The definition over EE now gives the displayed equality. Nonvanishing under all scalar extensions follows from Theorem 2.33.6. ◻

References. This comparison implements the strictification definition in [TemIntro] (§3.5) using the internal finite-normalization and completed-tensor theorems.

3.22 Abhyankar dimension and the dimension of analytic domains

The dimension of an affinoid can be recovered from its completed residue fields. Finite normalization proves this first in the strict case. A free Gauss extension then preserves the residual dimension of its canonical section, while no other point over the same point has larger relative residual dimension. This gives the general statement and, at the same time, the dimension comparison for domains and coverings.

Definition 3.22.1. For a valued extension E/FE/F, put s(E/F)=trdeg⁡F~E~,t(E/F)=dim⁡Q((v(E×)/v(F×))⊗ZQ),d(E/F)=s(E/F)+t(E/F).\begin{aligned} s(E/F)&=\operatorname{trdeg}_{\widetilde F}\widetilde E,\\ t(E/F)&=\dim_{\mathbf Q} ((v(E^\times)/v(F^\times))\otimes_{\mathbf Z}\mathbf Q),\\ d(E/F)&=s(E/F)+t(E/F). \end{aligned} The values are nonnegative integers or infinity. The invariant of an analytic point is dK(x)=d(H(x)/K)d_K(x)=d(\mathcal H(x)/K).

Lemma 3.22.2. NAG-001586 (Towers, algebraic extensions, and completion). For a tower K⊂F⊂EK\subset F\subset E of valued fields, d(E/K)=d(E/F)+d(F/K).d(E/K)=d(E/F)+d(F/K). Completion does not change the value group or the residue field. An algebraic valued extension has s=t=0s=t=0. Consequently algebraic extensions and completion do not change d(−/K)d(-/K). No completeness assumption is needed for the tower or algebraic assertions.

Proof. The residue fields form a tower. Their transcendence degrees add: when both degrees are finite, choose a finite transcendence basis in each step. Their union is algebraically independent over the bottom field, by collecting coefficients of a polynomial in the second basis, and the whole extension is algebraic over that union. If one degree is infinite, arbitrarily long finite independent families already force the degree at the top to be infinite. For value groups the sequence 0⟶(v(F×)/v(K×))⊗Q⟶(v(E×)/v(K×))⊗Q⟶(v(E×)/v(F×))⊗Q⟶0\begin{aligned} 0\longrightarrow{}& (v(F^\times)/v(K^\times))\otimes\mathbf Q \longrightarrow (v(E^\times)/v(K^\times))\otimes\mathbf Q\\ \longrightarrow{}& (v(E^\times)/v(F^\times))\otimes\mathbf Q \longrightarrow0 \end{aligned} is exact: clearing integer denominators verifies both injectivity and the middle kernel. Vector-space dimensions therefore add.

If a≠0a\ne 0 is in the completion, approximate it by aja_j in the original field with ∣a−aj∣<∣a∣|a-a_j|<|a|. Then ∣aj∣=∣a∣|a_j|=|a|, so no value is added. A norm-one element has the same residue as an approximation whose error is less than one; hence no residue is added either.

Let aa be nonzero and algebraic over FF. In a polynomial relation ∑ciai=0\sum c_i a^i=0, the largest nonzero term size is attained at least twice. Thus (i−j)v(a)∈v(F×)(i-j)v(a)\in v(F^\times) for two distinct indices, proving that the value-group quotient is torsion. For a norm-one algebraic element aa, divide a nonzero polynomial relation by a coefficient of largest size. Its coefficients now have size at most one and not all their residues vanish. Reduction gives a nonzero polynomial relation for a~\widetilde a over F~\widetilde F. Elements of size less than one reduce to zero. The residue extension is therefore algebraic. This proves the last assertion and completes the proof. ◻

Lemma 3.22.3. NAG-001587 (The valued-field Abhyankar inequality). Let E/FE/F be a valued extension with finite transcendence degree nn. Then d(E/F)≤nd(E/F)\le n. The same bound holds after completion and after any subsequent algebraic valued extension.

Proof. Choose norm-one elements u1,…,ua∈Eu_1,\ldots,u_a\in E with residues algebraically independent over F~\widetilde F, and choose z1,…,zb∈E×z_1,\ldots,z_b\in E^\times whose values are rationally independent modulo v(F×)v(F^\times). We show that the a+ba+b elements ui,zju_i,z_j are algebraically independent over FF.

For every nonzero polynomial QQ over FF in the uiu_i, ∣Q(u1,…,ua)∣=max⁡{ ∣c∣:c is a coefficient of Q }.|Q(u_1,\ldots,u_a)|=\max\{\,|c|:c\text{ is a coefficient of }Q\,\}. Indeed, divide by a coefficient attaining the maximum and reduce; algebraic independence makes the resulting residue nonzero. Now write a nonzero polynomial in all variables as ∑νQν(u)zν\sum_\nu Q_\nu(u)z^\nu, omitting zero coefficients. The sizes of its summands lie in pairwise distinct cosets of ∣F×∣|F^\times|: equality for two exponents would give a nontrivial integer relation between the v(zj)v(z_j) modulo v(F×)v(F^\times). There is consequently a unique largest summand, and the sum is nonzero. Thus a+b≤na+b\le n. Taking all finite choices proves the inequality, including finiteness of both invariants. Lemma 3.22.2 proves the assertions about completion and algebraic extension. ◻

Lemma 3.22.4. NAG-001588 (Residual dimension of a Gauss field). Let FF be any complete valued field and let EE be the completion of F(T1,…,Tm)F(T_1,\ldots,T_m) with its weighted Gauss absolute value, for arbitrary positive radii. Then d(E/F)=md(E/F)=m.

Proof. Consider one variable TT, with w=v(T)w=v(T). Its Gauss fraction field has value group v(F×)+Zwv(F^\times)+\mathbf Z w, since the value of a polynomial is the minimum of its term values and quotients subtract these values. If no positive multiple of ww lies in v(F×)v(F^\times), every nonzero polynomial has a unique term of largest size. In a rational function of size one the quotient of the two leading terms has exponent zero; it differs from a scalar of norm one by an element of norm less than one. The residue field is therefore F~\widetilde F, and the value-group quotient has rational rank one.

Otherwise choose the least q>0q>0 with qw=v(c)qw=v(c) for some c∈F×c\in F^\times, and put U=Tq/cU=T^q/c. Its residue is transcendental over F~\widetilde F: a nonzero polynomial in UU with coefficients of size at most one and at least one unit coefficient has Gauss norm one. Every residue of a rational function of norm one belongs to F~(U~)\widetilde F(\widetilde U). To see this, divide its numerator and denominator by a maximal term of the denominator. The terms that survive reduction have exponent differences divisible by qq, by its minimality, and become scalar multiples of Laurent monomials in U~\widetilde U. The reduced denominator is nonzero by the just-proved transcendence. Thus the residue field is F~(U~)\widetilde F(\widetilde U), and the value-group quotient is torsion. In both cases d=1d=1. Completion changes neither invariant.

Adjoin the variables one at a time. The resulting completed field agrees with the completion for the simultaneous weighted Gauss norm. Indeed the norms agree on rational functions over FF, and a rational function with finitely many coefficients in an intermediate completed field is approximated by replacing those coefficients by elements of its dense rational subfield. A sufficiently close denominator has the same positive norm, so its inverse also converges. This proves density at each step. The tower formula now gives d(E/F)=md(E/F)=m, including m=0m=0. ◻

Proposition 3.22.5. NAG-001589 (Residual dimension under free scalar extension). Let AA be a generalized KK-affinoid algebra, let rr be a finite tuple of mm radii free over KK, and put L=KrL=K_r. For x∈M(A)x\in\mathcal M(A) and its Gauss lift σr(x)\sigma_r(x), dL(σr(x))=dK(x).d_L(\sigma_r(x))=d_K(x). For every y∈ML(AL)y\in\mathcal M_L(A_L) above xx, dL(y)≤dK(x).d_L(y)\le d_K(x).

Proof. Write F=HA(x)F=\mathcal H_A(x) and E=HAL(y)E=\mathcal H_{A_L}(y). The residue-field universal property embeds FF and LL isometrically in EE. Denote the images of the mm Gauss parameters by t1,…,tmt_1,\ldots,t_m. The field EE is the completion of F(t1,…,tm)F(t_1,\ldots,t_m) with the induced absolute value. Here the density assertion is essential: finite Laurent polynomials with coefficients in AA are dense in ALA_L by Theorem 1.16.2; bounded evaluation therefore puts the entire image of ALA_L in the closure of F(t1,…,tm)F(t_1,\ldots,t_m). Conversely the completion defining EE contains FF, the parameters, and the closure of their fraction field. Inverting nonzero limits is continuous, so these two closures are the same complete valued field.

Lemma 3.22.3 gives d(E/F)≤md(E/F)\le m. The free-polyradius field LL is the completion of the Gauss rational field over KK: the unique dominant term expands every rational inverse as a convergent Laurent geometric series. Thus Lemma 3.22.4 gives d(L/K)=md(L/K)=m. The two tower formulas yield m+d(E/L)=d(E/K)=d(F/K)+d(E/F).m+d(E/L)=d(E/K)=d(F/K)+d(E/F). If d(F/K)d(F/K) is infinite, the asserted inequality is automatic; otherwise all terms are finite and cancellation of mm proves it.

At the canonical point σr(x)\sigma_r(x) the norm of a finite Laurent polynomial is precisely its weighted Gauss norm over FF, by Theorem 1.19.1. The same density argument identifies its completed residue field with the completion of F(T1,…,Tm)F(T_1,\ldots,T_m) for that norm. In particular the parameters are algebraically independent here, even if their radii are no longer free over FF. Lemma 3.22.4 gives d(E/F)=md(E/F)=m, and the displayed tower equality proves d(E/L)=d(F/K)d(E/L)=d(F/K), with infinity allowed. ◻

Lemma 3.22.6. NAG-001590 (Completed residue fields of a finite morphism). Let A→BA\to B be a bounded homomorphism of affinoid algebras with BB finite as an AA-module. If y∈M(B)y\in\mathcal M(B) lies above x∈M(A)x\in\mathcal M(A), then H(y)/H(x)\mathcal H(y)/\mathcal H(x) is finite. Hence dK(y)=dK(x)d_K(y)=d_K(x).

Proof. Put F=H(x)F=\mathcal H(x) and view it as a closed subfield of H(y)\mathcal H(y). If b1,…,bqb_1,\ldots,b_q generate BB over AA, the FF-span of the evaluated BB is generated by their images. This span is a finite-dimensional FF-subalgebra and a domain. Multiplication by a nonzero element is an injective endomorphism of a finite-dimensional space, hence surjective, so this subalgebra is a field. It is complete and closed, by the finite-dimensional norm comparison in Lemma 1.12.1. It contains the evaluated fraction field of B/ker⁡yB/\ker y, which is dense in H(y)\mathcal H(y). It is therefore all of H(y)\mathcal H(y). The invariant equality follows from Lemma 3.22.2. ◻

Theorem 3.22.7. NAG-001591 (Dimension as a maximum over analytic points). For every nonempty generalized affinoid space X=M(A)X=\mathcal M(A) over any complete non-Archimedean field with its specified absolute value, dim⁡KX=max⁡x∈XdK(x).\dim_K X=\max_{x\in X}d_K(x). In particular every dK(x)d_K(x) is finite. This includes nonreduced algebras, trivial valuations, and arbitrary real radii.

Proof. Step 1: the strict upper bound. Suppose AA is strict and fix xx. Its kernel p\mathfrak p is prime, and B=A/pB=A/\mathfrak p is a strict integral affinoid algebra. The evaluated fraction fields of AA and BB at this point coincide, so their completed residue fields coincide. Choose a finite normalization R=K⟨T1,…,Te⟩↪B,e=dim⁡B≤dim⁡AR=K\langle T_1,\ldots,T_e\rangle\hookrightarrow B,\qquad e=\dim B\le\dim A using Theorem 1.12.2 and Proposition 3.21.1. Let zz be the image of xx in M(R)\mathcal M(R). Lemma 3.22.6 gives dK(x)=dK(z)d_K(x)=d_K(z).

The field HR(z)\mathcal H_R(z) is the completion of Frac⁡(K[T1,…,Te]/q)\operatorname{Frac}(K[T_1,\ldots,T_e]/\mathfrak q), where q\mathfrak q is the kernel on the polynomial subalgebra. Indeed polynomials are dense in RR for its Gauss norm and evaluation at zz is contractive. After approximating numerator and denominator, with denominator error smaller than its nonzero value, the same assertion holds for fractions. This proves the claimed density in the completed residue field. The polynomial fraction field has transcendence degree at most ee. Lemma 3.22.3 therefore gives dK(z)≤e≤dim⁡Ad_K(z)\le e\le\dim A. This argument uses the polynomial fraction field, not the generally much larger fraction field of the Tate algebra.

Step 2: a strict point attaining the bound. Choose a prime quotient B=A/pB=A/\mathfrak p with dim⁡B=dim⁡A=d\dim B=\dim A=d. Such a quotient exists: a prime chain of maximal length starts at such a prime. The dimension is finite: a strict presentation Rn↠AR_n\twoheadrightarrow A identifies every prime chain of AA with a prime chain of RnR_n, whose dimension is nn by Proposition 3.21.1. Normalize BB finitely over RdR_d. Proposition 3.12.2 gives a point xx above the unit Gauss point η\eta of RdR_d. Polynomial density identifies H(η)\mathcal H(\eta) with the completion of the unit Gauss field K(T1,…,Td)K(T_1,\ldots,T_d). Lemma 3.22.4 gives dK(η)=dd_K(\eta)=d, and Lemma 3.22.6 gives dK(x)=dd_K(x)=d. Its image in M(A)\mathcal M(A) has the same residue field, proving attainment.

Step 3: arbitrary radii. Choose L=KrL=K_r making ALA_L strict by Theorem 1.17.6. Its dimension is d=dim⁡KXd=\dim_KX by Theorem 3.21.8. For every x∈Xx\in X, the Gauss-section equality in Proposition 3.22.5 and Step 1 give dK(x)=dL(σr(x))≤d.d_K(x)=d_L(\sigma_r(x))\le d. By Step 2 choose y∈ML(AL)y\in\mathcal M_L(A_L) with dL(y)=dd_L(y)=d, and let xx be its image. The other inequality in that same proposition gives d≤dK(x)d\le d_K(x). Combining both bounds gives dK(x)=dd_K(x)=d. Every construction used the original absolute values. ◻

Corollary 3.22.8. NAG-001592 (Dimension of domains and affinoid covers). If VV is a nonempty affinoid domain of an affinoid space XX, then dim⁡KV≤dim⁡KX\dim_KV\le\dim_KX. If (Vi)(V_i) is any cover of a nonempty affinoid XX by nonempty affinoid domains, then dim⁡KX=sup⁡idim⁡KVi.\dim_KX=\sup_i\dim_KV_i. A finite cover has a maximum in place of the supremum.

Proof. Corollary 2.33.15 identifies the completed residue fields of a point on VV and on XX. The preceding theorem therefore gives dim⁡KV=max⁡x∈VdK(x)≤max⁡x∈XdK(x)=dim⁡KX.\dim_KV=\max_{x\in V}d_K(x)\le\max_{x\in X}d_K(x)=\dim_KX. The union of the ViV_i is XX, so taking the supremum of these equalities proves the covering formula. No flat-ring dimension inequality or boundary hypothesis is used. ◻

We give the empty analytic space dimension −∞-\infty.

Theorem 3.22.9. NAG-001593 (Local analytic dimension and the Abhyankar inequality). For any nonempty analytic space XX, the supremum of the dimensions of its affinoid charts is independent of the chosen analytic atlas and equals dim⁡KX=sup⁡x∈XdK(x).\dim_KX=\sup_{x\in X}d_K(x). It agrees with the affinoid dimension already defined. An analytic domain has dimension at most that of its ambient space. If dim⁡xX\dim_xX is the infimum of the dimensions of the analytic neighborhoods of xx, then dK(x)≤dim⁡xX.d_K(x)\le\dim_xX. These conclusions require neither goodness nor Hausdorffness. At a good point the infimum can be taken over affinoid neighborhoods. At every point dim⁡xX\dim_xX is finite.

Proof. The completed residue field attached to a point is independent of its affinoid chart: on a common refinement this is Corollary 2.33.15, and the analytic-net construction identifies the maps on successive refinements. Thus dK(x)d_K(x) is an intrinsic point invariant. For any analytic atlas (Xi)(X_i), Theorem 3.22.7 gives sup⁡idim⁡KXi=sup⁡isup⁡x∈XidK(x)=sup⁡x∈XdK(x).\sup_i\dim_KX_i =\sup_i\sup_{x\in X_i}d_K(x) =\sup_{x\in X}d_K(x). This proves atlas independence rather than assuming it in the definition. It also agrees with the old dimension on an affinoid chart, by the same theorem. For an analytic domain U⊂XU\subset X, the residue-field invariance is checked on affinoid refinements of its domain atlas; it gives dim⁡KU=sup⁡x∈UdK(x)≤dim⁡KX\dim_KU=\sup_{x\in U}d_K(x)\le\dim_KX.

Every analytic neighborhood UU of xx contains that point, so dK(x)≤dim⁡KUd_K(x)\le\dim_KU. Taking their infimum proves the desired inequality. If xx is good, affinoid neighborhoods are cofinal among analytic neighborhoods: take one inside the ordinary open neighborhood contained in a given analytic neighborhood. Finally the net axiom gives finitely many affinoid charts XiX_i containing xx whose union contains an ordinary open neighborhood WW of xx. Give WW its open analytic structure. For every w∈Ww\in W one of these charts contains ww, so the already proved formula gives dim⁡KW=sup⁡w∈WdK(w)≤max⁡idim⁡KXi<∞.\dim_KW=\sup_{w\in W}d_K(w)\le\max_i\dim_KX_i<\infty. Thus WW is an analytic neighborhood of finite dimension. This proves finiteness of dim⁡xX\dim_xX without a good-space assumption. ◻

References. The valued-field argument is the classical one formulated in [TemIntro] (Exercise 2.3.3.2). The strictification convention is [TemIntro] (Definition 3.5.1.2). All arguments needed here, including the Gauss-lift comparison and the analytic domain/atlas dimension comparison, have been given internally.

3.23 A compact analytic domain without affinoid neighbourhoods

The distinction between an analytic domain and a good analytic space already occurs in a polydisc. The obstruction is visible in the valuation space of punctured affine space.

Lemma 3.23.1. NAG-001885 (The valuation space of punctured affine space). Let kk be a field, let n≥2n\ge 2, and put R=k[t1,…,tn]R=k[t_1,\ldots,t_n], F=Frac⁡(R)F=\operatorname{Frac}(R). Inside the space of valuation rings of FF containing kk, set Q=⋃i=1n{V:R[ti−1]⊂V}.Q=\bigcup_{i=1}^n \{V:R[t_i^{-1}]\subset V\}. Then QQ is not affine as a birational space. In fact, ⋂V∈QV=R\bigcap_{V\in Q}V=R, but QQ does not contain every valuation ring of FF containing RR.

Proof. Regard kk as a complete trivially valued field. Its strict Tate algebra in nn variables is exactly RR: a convergent series has only finitely many nonzero coefficients. Theorem 1.13.1 therefore proves that RR is a unique factorization domain. If h∈F∖Rh\in F\setminus R, write it as a reduced fraction and choose an irreducible factor PP of its denominator. The exponent of PP in a nonzero fraction defines a discrete valuation vPv_P of FF; its valuation ring VPV_P contains RR and does not contain hh. The irreducible PP cannot divide both t1t_1 and t2t_2. Hence some tit_i is a unit of VPV_P, and VP∈QV_P\in Q. Every member of QQ contains RR, so this proves ⋂V∈QV=R.\bigcap_{V\in Q}V=R.

For a nonzero polynomial ff, let w(f)w(f) be the least total degree of a nonzero homogeneous component of ff. The product of the lowest homogeneous components is nonzero, so w(fg)=w(f)+w(g)w(fg)=w(f)+w(g), and w(f+g)≥min⁡(w(f),w(g))w(f+g)\ge\min(w(f),w(g)) whenever the sum is nonzero. Thus ww extends by subtraction to a valuation of FF. Its valuation ring V0V_0 contains RR, while w(ti)=1w(t_i)=1 for every ii. Consequently V0∉QV_0\notin Q.

If QQ were affine, its inclusion in the fixed valuation space would identify it with {V:S⊂V}\{V:S\subset V\} for a finitely generated kk-subalgebra S⊂FS\subset F. Every element of SS would belong to ⋂V∈QV=R\bigcap_{V\in Q}V=R. But then S⊂R⊂V0S\subset R\subset V_0, which would force V0∈QV_0\in Q, a contradiction. ◻

Theorem 3.23.2. NAG-001886 (A punctured closed polydisc is not good). Let KK be any complete non-Archimedean field, possibly trivially valued, let n≥2n\ge 2, and let r1,…,rn>0r_1,\ldots,r_n>0. In X=M(K{r1−1T1,…,rn−1Tn})X=\mathcal M(K\{r_1^{-1}T_1,\ldots,r_n^{-1}T_n\}) consider the subset U=X∖{z:∣Ti(z)∣<ri for every i}=⋃i=1nUi,Ui={z∈X:∣Ti(z)∣=ri}.\begin{aligned} U&=X\setminus\{z:|T_i(z)|<r_i\text{ for every }i\} =\bigcup_{i=1}^n U_i,\\ U_i&=\{z\in X:|T_i(z)|=r_i\}. \end{aligned} Then UU is a compact Hausdorff analytic domain of XX. At the weighted Gauss point x(∑νaνTν)=max⁡ν∣aν∣rνx(\sum_\nu a_\nu T^\nu)=\max_\nu|a_\nu|r^\nu, the space UU has no affinoid neighbourhood.

Proof. Step 1: the domain structure. Each UiU_i is the Laurent domain ∣Ti∣≥ri|T_i|\ge r_i in XX; the opposite inequality already holds throughout XX. All finite intersections are Laurent affinoid domains. Their finite cover of UU is an analytic net: at a point z∈Uz\in U, remove the finitely many closed members which do not contain zz. The remaining relative open neighbourhood is covered by the members containing zz. The domain and net constructions in Proposition 2.39.3 and Proposition 2.40.5 therefore give the analytic-domain structure on UU. The displayed complement is open in XX, so UU is closed in the compact Hausdorff space XX, hence compact Hausdorff.

Step 2: unit radii. Assume first that every ri=1r_i=1, put k=K~k=\widetilde K, and let xx be the unit Gauss point. Polynomials are dense in the Tate algebra. Approximating the numerator and denominator of a fraction, with error smaller than the nonzero denominator, shows that H(x)\mathcal H(x) is the completion of K(T1,…,Tn)K(T_1,\ldots,T_n) for the Gauss absolute value. Every nonzero Gauss norm of a polynomial belongs to ∣K×∣|K^\times|. For a norm-one fraction, divide its numerator and denominator by coefficients of their maximal norms. Reduction then identifies the residue field with k(t1,…,tn)k(t_1,\ldots,t_n); completion has the same residue field by Lemma 3.22.2. The graded image of the Tate algebra in Fx=H(x)~ grF_x=\widetilde{\mathcal H(x)}^{\,\mathrm{gr}} is K~ gr[t1,…,tn]\widetilde K^{\,\mathrm{gr}}[t_1,\ldots,t_n]: only finitely many terms of a convergent series can have its maximal norm, and these give precisely its initial form.

All the UiU_i contain xx. Near xx, TiT_i is invertible; the additional active inequality defining UiU_i is ∣Ti−1∣≤1|T_i^{-1}|\le 1. By Lemma 3.19.5, its reduction adds ti−1t_i^{-1} to the graded image just computed. Gluing the domain germs by Theorem 3.19.9 and Theorem 3.19.10, the reduction of (U,x)(U,x) is therefore the union of the graded valuation subspaces containing K~ gr[t1,…,tn,ti−1],1≤i≤n.\widetilde K^{\,\mathrm{gr}} [t_1,\ldots,t_n,t_i^{-1}],\qquad 1\le i\le n. It is exactly the pullback, by restriction to degree one, of the ordinary birational space QQ in Lemma 3.23.1. That lemma says QQ is not affine. Lemma 3.19.20 shows that its graded pullback is not affine either, including when the valuation of KK is trivial. Theorem C.1.114 now shows that xx has no affinoid neighbourhood in UU.

Step 3: arbitrary radii. Equip the rational function field K(S1,…,Sn)K(S_1,\ldots,S_n) with the Gauss absolute value of polyradius rr and let LL be its completion. Multiplicativity and completion are the Gauss-field construction used in Lemma 3.18.2; no independence of the radii is required. The images cic_i of SiS_i satisfy ∣ci∣=ri|c_i|=r_i. Scalar extension identifies XL=M(L{r−1T})≃M(L{Z1,…,Zn}),Ti=ciZi.X_L=\mathcal M(L\{r^{-1}T\}) \simeq\mathcal M(L\{Z_1,\ldots,Z_n\}), \qquad T_i=c_iZ_i . The exact scalar presentation and domain base-change statements (2.33.5, 2.36.9, 2.40.12) identify (Ui)L(U_i)_L with ∣Ti∣=ri|T_i|=r_i, hence with ∣Zi∣=1|Z_i|=1. Since the finite domain cover and its intersections base change, UL=⋃i(Ui)LU_L=\bigcup_i(U_i)_L is exactly the unit-radius punctured polydisc. Equivalently its underlying set is the inverse image of UU: the projection preserves every number ∣Ti∣|T_i|, so it preserves the displayed complement condition as well.

Let yy be its unit Gauss point in the ZZ-coordinates. For a polynomial with coefficients in KK, evaluation at yy is max⁡ν∣aνcν∣=max⁡ν∣aν∣rν\max_\nu|a_\nu c^\nu| =\max_\nu|a_\nu|r^\nu; by continuity the same holds for every element of K{r−1T}K\{r^{-1}T\}. Thus yy lies over xx. If xx had an affinoid neighbourhood WW in UU, then WLW_L would be an affinoid domain of ULU_L. Moreover, an ordinary open neighbourhood of xx contained in WW has an open inverse image containing yy and contained in WLW_L. Hence WLW_L would be an affinoid neighbourhood of yy, contradicting Step 2 over LL. This proves the assertion for every positive polyradius. ◻

References. The example is the classical one in [TemIntro] (Example 4.2.1.4(i)). Its proof here uses the internally proved germ-reduction criterion; the valuation obstruction and the arbitrary-radius base change are spelled out above.

3.24 Discrete residue fields and free Gauss coordinates

The elementary valued-field arguments in this section provide the field inputs for relative annulus–polydisc neighbourhoods. All absolute values retain their original normalization.

Lemma 3.24.1. NAG-001595 (Uniform approximation implies density). Let EE be a real-valued field and let V⊂EV\subset E be an additive subgroup. If some 0<q<10<q<1 has the property that every a∈Ea\in E admits b∈Vb\in V with ∣a−b∣≤q∣a∣|a-b|\le q|a|, then VV is dense in EE.

Proof. Fix aa. Start with b0=0b_0=0. Having chosen bj∈Vb_j\in V, choose cj∈Vc_j\in V with ∣a−bj−cj∣≤q∣a−bj∣|a-b_j-c_j|\le q|a-b_j| and put bj+1=bj+cjb_{j+1}=b_j+c_j. Induction gives ∣a−bj∣≤qj∣a∣|a-b_j|\le q^j|a|, hence bj→ab_j\to a. ◻

Lemma 3.24.2. NAG-001596 (Value groups and residues of a finite extension). For a finite valued extension E/FE/F, both e=[∣E×∣:∣F×∣]e=[|E^\times|:|F^\times|] and f=[E~:F~]f=[\widetilde E:\widetilde F] are finite, and ef≤[E:F]ef\le[E:F]. If ∣F×∣|F^\times| is nontrivial cyclic, then ∣E×∣|E^\times| is nontrivial cyclic as well.

Proof. Choose finitely many ai∈E×a_i\in E^\times with distinct value classes modulo ∣F×∣|F^\times|, and norm-one uj∈Eu_j\in E with residues linearly independent over F~\widetilde F. The elements aiuja_i u_j are FF-linearly independent. Indeed, in a purported relation first collect terms with fixed ii. A nonzero sum ∑jcjuj\sum_j c_j u_j has norm max⁡j∣cj∣\max_j|c_j|: divide by a coefficient attaining the maximum and use residue independence. The nonzero collected sums for different ii have distinct norms, so a largest term cannot cancel. This proves independence. Arbitrarily large finite choices are impossible in a finite dimensional FF-space; hence e,fe,f are finite and ef≤[E:F]ef\le[E:F].

Write ∣F×∣=qZ|F^\times|=q^{\mathbf Z} with 0<q<10<q<1. The finite quotient of value groups is killed by some integer N>0N>0, so ∣E×∣⊂q(1/N)Z|E^\times|\subset q^{(1/N)\mathbf Z}. A subgroup of this cyclic group is cyclic, and it contains qZq^{\mathbf Z}, so is nontrivial. ◻

Lemma 3.24.3. NAG-001597 (Finiteness over a complete discrete subfield). Let EE be a valued field whose value group is nontrivial cyclic. Let F⊂EF\subset E be complete and nontrivially valued. If E~/F~\widetilde E/\widetilde F is finite, then E/FE/F is finite. In particular EE is complete. No completeness of EE is assumed in the hypotheses.

Proof. The nontrivial subgroup ∣F×∣|F^\times| has finite index in the cyclic group ∣E×∣|E^\times|. Choose representatives a1,…,ae∈E×a_1,\ldots,a_e\in E^\times of these value classes, and lift a residue-field basis to norm-one elements u1,…,uf∈Eu_1,\ldots,u_f\in E. Let VV be the FF-span of the finitely many aiuja_i u_j. For z≠0z\ne 0, choose ii and c∈F×c\in F^\times with ∣z∣=∣cai∣|z|=|ca_i|. Expand the residue of z/(cai)z/(ca_i) in the chosen residue basis and lift its coefficients to F∘F^\circ. This gives v∈Vv\in V with ∣z−v∣<∣z∣|z-v|<|z|.

Write ∣E×∣=qZ|E^\times|=q^{\mathbf Z} with 0<q<10<q<1. The last strict inequality implies ∣z−v∣≤q∣z∣|z-v|\le q|z|, uniformly in zz. The preceding density lemma makes VV dense in EE. The restricted absolute value is a norm on the finite dimensional FF-space VV; Lemma 1.12.1 makes VV complete, hence closed in the metric space EE. Thus V=EV=E. ◻

Lemma 3.24.4. NAG-001598 (Coordinate changes in a free Gauss field). Let r=(r1,…,rm)r=(r_1,\ldots,r_m) be free over a complete field KK, and let Si∈KrS_i\in K_r satisfy ∣Si∣=ri|S_i|=r_i. The isometric substitution Ti↦SiT_i\mapsto S_i is an automorphism of KrK_r. Thus the complete subfield generated by K,S1,…,SmK,S_1,\ldots,S_m is all of KrK_r.

Proof. The unique dominant term in the Laurent expansion of SiS_i has norm rir_i. Freeness implies that it is αiTi\alpha_iT_i with αi∈K\alpha_i\in K and ∣αi∣=1|\alpha_i|=1. Rescaling SiS_i by αi−1\alpha_i^{-1} does not change the generated complete subfield and is itself an invertible coordinate scaling. We may therefore assume ∣Ti−Si∣≤qri|T_i-S_i|\le q r_i for one q<1q<1 and every ii. The identity for inverse differences also gives ∣Ti−1−Si−1∣≤qri−1|T_i^{-1}-S_i^{-1}|\le q r_i^{-1}.

For every Laurent monomial, telescoping products gives ∣Tν−Sν∣≤qrν(ν∈Zm).|T^\nu-S^\nu|\le q r^\nu \qquad(\nu\in\mathbf Z^m). The weighted Laurent description in Lemma 1.16.1 defines the substitution ϕ(∑aνTν)=∑aνSν\phi(\sum a_\nu T^\nu)=\sum a_\nu S^\nu. It is an isometry: each nonzero finite sum has a unique largest term, and completion preserves the equality of norms. It is a field homomorphism by density of Laurent polynomials. The displayed estimate gives ∣a−ϕ(a)∣≤q∣a∣(a∈Kr).|a-\phi(a)|\le q|a| \qquad(a\in K_r). Hence its image is dense by Lemma 3.24.1. The image is also complete, since ϕ\phi is an isometry of a complete field, and therefore closed. It is all of KrK_r. ◻

Proposition 3.24.5. NAG-001599 (Independent differentials of free coordinates). Let YY be a KK-analytic space and let y∈Yy\in Y with an isometric KK-identification H(y)=Kr\mathcal H(y)=K_r, where r=(r1,…,rm)r=(r_1,\ldots,r_m) is free over KK. If analytic functions gig_i near yy satisfy ∣gi(y)∣=ri|g_i(y)|=r_i, then (dg1)(y),…,(dgm)(y)inΩY/Kan⊗H(y)(dg_1)(y),\ldots,(dg_m)(y) \quad\hbox{in}\quad \Omega^{\rm an}_{Y/K}\otimes\mathcal H(y) are linearly independent. In particular this applies to quasi-smooth spaces.

Proof. Choose an affinoid domain containing yy on which the functions are represented, and write its algebra as AA. By Theorem 2.41.5, its differential fibre is the asserted one. Set Si=gi(y)S_i=g_i(y). The preceding lemma makes SiS_i coordinates of KrK_r. On a Laurent series in these coordinates the operator Dj(∑ν∈ZmaνSν)=∑ννjaνSν−ejD_j\left(\sum_{\nu\in\mathbf Z^m} a_\nu S^\nu\right) =\sum_\nu \nu_j a_\nu S^{\nu-e_j} is bounded with norm at most rj−1r_j^{-1}, since every integer has absolute value at most one. The weighted coefficients still tend to zero, so the formula defines a continuous operator. The Leibniz identity on Laurent polynomials extends by continuity; hence DjD_j is a bounded KK-derivation and Dj(Si)=δijD_j(S_i)=\delta_{ij}.

Compose DjD_j with the bounded evaluation A→H(y)=KrA\to\mathcal H(y)=K_r. The resulting bounded derivation from AA to the complete Banach AA-module KrK_r is represented by Theorem 2.41.3. After tensoring with KrK_r, it gives a KrK_r-linear functional on the differential fibre taking (dgi)(y)(dg_i)(y) to δij\delta_{ij}. These mm functionals form a left inverse to the map with columns (dgi)(y)(dg_i)(y), proving independence. ◻

Lemma 3.24.6. NAG-001600 (A discrete-residue point over a trivially valued field). Let FF be a trivially valued field and let XX be a nonempty boundaryless FF-analytic space. There is x∈Xx\in X such that H(x)\mathcal H(x) is either finite over FF, or is finite over an isometrically embedded FrF_r for some 0<r<10<r<1.

Proof. Fix 0<s<10<s<1. The field FsF_s is the Laurent-series field F((T))F((T)), with ∣T∣=s|T|=s: the weighted coefficient condition allows only finitely many negative powers and arbitrary nonnegative powers. In particular its value group is sZs^{\mathbf Z} and its residue field is FF. Theorem 2.40.12 gives surjectivity of complete scalar extension on analytic spaces, so XFsX_{F_s} is nonempty. By Definition 2.44.4, each point of the boundaryless space XX has an ordinary affinoid neighbourhood VV on which it is interior over M(F)\mathcal M(F). Indeed the only affinoid target domain containing its image is M(F)\mathcal M(F) itself. For a point of XFsX_{F_s} over that point, the scalar extension VFsV_{F_s} is an ordinary affinoid neighbourhood, by the analytic-domain preimage assertion of Theorem 2.40.12. On the affinoid map V→M(F)V\to\mathcal M(F), the forward base-change inclusion of Proposition 2.33.21 makes that lift interior in VFsV_{F_s} over M(Fs)\mathcal M(F_s). Since the new target is also a one-point spectrum, the same definition makes XFsX_{F_s} boundaryless. The neighbourhoods just exhibited also make it good.

At any point of XFsX_{F_s} apply Proposition 2.51.3 to its structure map. This gives a nonempty strict affinoid domain. Theorem 1.12.3 supplies a point yy whose completed residue field LL is finite over FsF_s. Let xx be its image in XX and identify the complete field E=H(x)E=\mathcal H(x) with its isometric image in LL. Lemma 3.24.2 makes ∣L×∣|L^\times| nontrivial cyclic and L~/F\widetilde L/F finite. Therefore E~/F\widetilde E/F is finite and ∣E×∣|E^\times| is a subgroup of a cyclic group.

If EE is trivially valued, it equals its residue field and is finite over FF. Otherwise choose λ∈E\lambda\in E with 0<r=∣λ∣<10<r=|\lambda|<1. Since FF is trivially valued, distinct powers of λ\lambda have distinct sizes; evaluation on F[T,T−1]F[T,T^{-1}] is therefore isometric for the radius-rr Gauss norm. Its completion is an isometric copy of FrF_r inside EE, with residue field FF. The group ∣E×∣|E^\times| is nontrivial cyclic and E~/F\widetilde E/F is finite. Apply Lemma 3.24.3 to Fr⊂EF_r\subset E. It gives the required finiteness. ◻

Lemma 3.24.7. NAG-001601 (A discrete complete field with its coefficient field). Let EE be a complete valued field with nontrivial cyclic value group. Suppose that K⊂EK\subset E is trivially valued and that reduction identifies KK with E~\widetilde E. For a uniformizer τ∈E\tau\in E, put r=∣τ∣∈(0,1)r=|\tau|\in(0,1). Then substitution defines an isometric isomorphism Kr=K((T))→∼EK_r=K((T))\xrightarrow{\sim}E, sending TT to τ\tau.

Proof. The distinct powers of τ\tau have distinct norms, so substitution is an isometry on Laurent polynomials and extends isometrically to KrK_r. To prove surjectivity, take a∈E∘a\in E^\circ. Choose c0∈Kc_0\in K with the same residue as aa. Since ∣τ∣|\tau| generates the value group below one, a1=(a−c0)/τa_1=(a-c_0)/\tau still belongs to E∘E^\circ. Repeat, obtaining cj∈Kc_j\in K and aj+1∈E∘a_{j+1}\in E^\circ with aj=cj+τaj+1a_j=c_j+\tau a_{j+1}. Thus ∣a−∑j=0N−1cjτj∣≤rN.\left|a-\sum_{j=0}^{N-1}c_j\tau^j\right|\le r^N. Completeness gives a=∑j≥0cjτja=\sum_{j\ge 0}c_j\tau^j. Multiplication by a suitable power of τ\tau handles every a∈Ea\in E. Uniqueness follows from the norm of the first nonzero coefficient. ◻

Lemma 3.24.8. NAG-001602 (Uniformization after a finite extension of constants). Let FF be any field, endowed with the trivial absolute value, let 0<s<10<s<1, and let L/Fs=F((t))L/F_s=F((t)) be finite separable. There are a finite extension K/FK/F, a number 0<r<10<r<1, and a surjective FF-algebra homomorphism L⊗FK↠KrL\otimes_F K\twoheadrightarrow K_r whose restriction to LL is isometric for its given absolute value. The restriction to KK is the constant-field embedding. Here the tensor product is the ordinary algebraic tensor product.

Proof. Step 1: algebraize the discrete valuation. The dense-coefficient lemma 2.54.2 supplies Q∈F(t)[U]Q\in F(t)[U] such that L=F((t))[U]/(Q)L=F((t))[U]/(Q). Let E=F(t)[U]/(Q)E=F(t)[U]/(Q), identified with its image in LL. It is a field, since QQ is irreducible even over F((t))F((t)). Moreover EE is dense in LL: approximate the finitely many coefficients of an expression in the basis 1,U,…,Udeg⁡Q−11,U,\ldots,U^{\deg Q-1}.

Let BB be the integral closure of F[t]F[t] in EE. It is finite over F[t]F[t], by [StacksAlg] (Tag 0335) and the definition of a Japanese domain, [StacksAlg] (Tag 032F). Thus Y=Spec⁡(B)Y=\operatorname{Spec}(B) is a normal affine curve; its dimension is one by [StacksAlg] (Tag 00OK). Every b∈Bb\in B has ∣b∣≤1|b|\le 1, since it satisfies a monic equation whose coefficients have norm at most one. Hence P={b∈B:∣b∣<1}P=\{b\in B:|b|<1\} is a prime over (t)(t) and is maximal, with residue field finite over FF. By [StacksAlg] (Tag 0B8Y), BPB_P is a discrete valuation ring. Its units have norm one and its uniformizer has norm less than one. Consequently its valuation, scaled by the prescribed value ∣t∣=s|t|=s, is exactly the restriction of the given absolute value on LL. In particular LL is the completion of EE at this closed point PP.

Step 2: obtain a smooth rational center. Apply [StacksAlg] (Tag 0GK5) to YY. After a finite purely inseparable extension F′/FF'/F, the normalization of YF′Y_{F'} is a finite disjoint union of curves smooth and geometrically irreducible over finite extensions of F′F'. The normalization is finite and surjective, by [StacksAlg] (Tag 0BXR); choose a closed point Q0Q_0 above PP on one component ZZ. Write F0F_0 for its field of definition in that statement and put K=κ(Q0)K=\kappa(Q_0), which is finite over F0F_0 by [StacksAlg] (Tag 00FV), and hence finite over FF. The base change ZKZ_K is a smooth integral KK-curve, and Q0Q_0 gives a KK-rational point QQ on it.

The function field E′E' of ZKZ_K is generated as a field by the images of EE and KK. Indeed normalization does not change the function field of the chosen irreducible component of YF′Y_{F'}; this field is a quotient of E⊗FF′E\otimes_F F'. After the indicated base change its function field is the corresponding quotient of E⊗FKE\otimes_F K. Since this tensor product is finite over EE, its domain quotient is already a field. Thus E′E' is precisely the compositum of those two images.

Step 3: complete and check algebraic surjectivity. The local ring at the smooth rational point QQ is a discrete valuation ring by [StacksAlg] (Tag 0B8Y), with residue field KK. Choose a uniformizer τ\tau and normalize its real absolute value so that the image of tt still has value ss. Let MM be the completion of E′E' for this valuation. Completion changes neither the value group nor the residue field, as proved in Lemma 3.22.2. Therefore Lemma 3.24.7 identifies MM with KrK_r, where r=∣τ∣r=|\tau|.

The valuation on E′E' restricts to the one centered at PP on EE: the local map BP→OZK,QB_P\to\mathcal O_{Z_K,Q} sends units to units and the maximal ideal into the maximal ideal, and the normalization of ∣t∣|t| fixes the scale. It consequently extends to an isometric embedding L↪ML\hookrightarrow M. Together with the constant inclusion K↪MK\hookrightarrow M it gives L⊗FK→ML\otimes_F K\to M. Its image is a finite-dimensional LL-algebra and a domain, hence a field: multiplication by a nonzero element is an injective endomorphism of a finite-dimensional vector space and is therefore surjective. By Lemma 1.12.1, this image is complete for its restricted absolute value and hence closed in MM. It contains E′E', which is dense in MM. The image is therefore all of M=KrM=K_r, proving the assertion for the algebraic tensor product itself. ◻

Proposition 3.24.9. NAG-001603 (Gauss or rational points after finite constants). Let FF be trivially valued, and let XX be a nonempty smooth FF-analytic space. There exist a finite extension K/FK/F and z∈XKz\in X_K such that H(z)\mathcal H(z) is isometrically KK or KrK_r for some 0<r<10<r<1. The image of zz can be required to lie in any prescribed nonempty ordinary open subset of XX.

Proof. Replace XX by the prescribed open subset. Fix 0<s<10<s<1. Scalar extension is surjective, and smoothness is preserved by base change; hence XFsX_{F_s} is nonempty and smooth. By Lemma 2.55.3, it has a point yy for which L=H(y)L=\mathcal H(y) is finite separable over FsF_s. Write xx for its image in XX and E=H(x)⊂LE=\mathcal H(x)\subset L. If EE is trivially valued, it is finite over FF: its residue field is a subfield of L~\widetilde L, which is finite over FF by Lemma 3.24.2. Take K=EK=E; the diagonal evaluation gives a KK-rational point over xx.

Suppose now that EE is nontrivially valued. Lemma 3.24.8 gives a finite K/FK/F and a quotient L⊗FK↠M=Kr0L\otimes_F K\twoheadrightarrow M=K_{r_0} whose map on LL is isometric. Let EKE_K be the image of E⊗FKE\otimes_F K in MM. This image is a finite-dimensional EE-algebra and a domain, so is a field and is complete by Lemma 1.12.1. Its value group is a nontrivial subgroup of the cyclic group ∣M×∣|M^\times|. Its residue field is exactly KK, since it is a subfield of M~=K\widetilde M=K and contains the constant field KK. Lemma 3.24.7 therefore gives EK=KrE_K=K_r isometrically, for some 0<r<10<r<1.

It remains to check that this field is an actual completed residue field on XKX_K. On an affinoid domain M(A)\mathcal M(A) through xx, compose evaluation with A⊗FK→EKA\otimes_F K\to E_K. The map is bounded and extends to the completed tensor product: if e1,…,ede_1,\ldots,e_d is an FF-basis of KK, then the norm of ∑aiei\sum a_i e_i in the target is at most max⁡i∣ai(x)∣\max_i|a_i(x)|, since the nonzero constants have norm one; this is bounded by a finite-module tensor norm. It defines a point zz over xx. The fraction field of its evaluation image contains the image of Frac⁡(A/ker⁡x)\operatorname{Frac}(A/\ker x) and KK. The first is dense in EE; finite sums then show that the generated field is dense in EKE_K. Conversely EKE_K is already complete. Thus H(z)=EK=Kr\mathcal H(z)=E_K=K_r as asserted. ◻

References. The valued-field and point constructions reconstruct [Duc18] (Lemmas 6.1.1–6.1.6). The differential-independence argument uses bounded coordinate derivations directly, and thus avoids an extra pure-dimension comparison.

3.25 Absolute boundary and canonical reduction

Theorem 3.25.1. NAG-001635 (Closed reduction points and absolute interior). Let AA be a strict affinoid algebra over a complete non-Archimedean field KK, with its specified absolute value. For X=M(A)X=\mathcal M(A), the canonical reduction map satisfies Int⁡(X/K)=red⁡−1((Spec⁡A~)closed).\operatorname{Int}(X/K)= \operatorname{red}^{-1}\bigl((\operatorname{Spec}\widetilde A)_{ \mathrm{closed}}\bigr).

Proof. Fix x∈Xx\in X, put k=K~k=\widetilde K and R=A~R=\widetilde A, and let SS be the image of A~gr\widetilde A^{\rm gr} in H(x)~gr\widetilde{\mathcal H(x)}^{\rm gr}. Its degree-one part is S1=R/pS_1=R/\mathfrak p, where p=red⁡(x)\mathfrak p=\operatorname{red}(x). The forward and reverse implications for absolute evaluation show that A→H(x)A\to\mathcal H(x) is inner over KK precisely when SS is finite over K~gr\widetilde K^{\rm gr}. We show this is equivalent to S1S_1 being finite-dimensional over kk.

A finite graded module has finitely many homogeneous generators. Every homogeneous component of the graded field K~gr\widetilde K^{\rm gr} has dimension at most one over kk. Thus finiteness of SS implies finite dimension of S1S_1. Conversely, choose an admissible absolute weighted Tate presentation of AA. The fixed absolute presentation result gives finitely many homogeneous module generators over the image of the initial-coordinate polynomial algebra. Taking their images in SS together with the initial coordinates gives finitely many homogeneous algebra generators uju_j over K~gr\widetilde K^{\rm gr}. Discard the zero generators. The degree rjr_j of each remaining generator is the spectral radius of a representative in AA, since its initial form has nonzero image. Lemma 3.16.6 gives ej>0e_j>0 and cj∈K×c_j\in K^\times with rjej=∣cj∣r_j^{e_j}=|c_j|. Hence ujej/c~j∈S1u_j^{e_j}/\widetilde c_j\in S_1. If S1S_1 is finite-dimensional over kk, the algebra K~grS1\widetilde K^{\rm gr}S_1 is finite over K~gr\widetilde K^{\rm gr}, and every uju_j is integral over it. Adjoining these finitely many integral generators proves that SS is finite over K~gr\widetilde K^{\rm gr}.

Theorem 3.16.3 makes RR a finite-type kk-algebra. By the Nullstellensatz [StacksAlg] (Tag 00FV), R/pR/\mathfrak p is finite over kk when p\mathfrak p is maximal. Conversely, a finite-dimensional domain over a field is a field: multiplication by any nonzero element is an injective linear endomorphism, hence surjective. Thus S1S_1 is finite-dimensional exactly when p\mathfrak p is a closed point. ◻

Lemma 3.25.2. NAG-001636 (Dimension of strict canonical reduction). For every nonempty strict affinoid space X=M(A)X=\mathcal M(A), dim⁡A~≤dim⁡KX.\dim\widetilde A\leq\dim_KX.

Proof. The ring R=A~R=\widetilde A is a finite-type k=K~k=\widetilde K-algebra. For every prime p\mathfrak p of RR, Theorem 3.16.16 gives a point xx reducing to p\mathfrak p. The inclusion R/p↪H(x)~R/\mathfrak p\hookrightarrow\widetilde{\mathcal H(x)} extends to its fraction field. Thus dim⁡(R/p)=trdeg⁡kFrac⁡(R/p)≤trdeg⁡kH(x)~≤dK(x)≤dim⁡KX.\dim(R/\mathfrak p) =\operatorname{trdeg}_{k}\operatorname{Frac}(R/\mathfrak p) \leq\operatorname{trdeg}_{k}\widetilde{\mathcal H(x)} \leq d_K(x)\leq\dim_KX. The first equality is [StacksAlg] (Tag 00P0), and the last inequality is Theorem 3.22.7. Every finite chain of prime ideals of the Noetherian ring RR lies over a minimal prime. Taking the maximum of the dimensions of these finitely many components gives the assertion. ◻

Lemma 3.25.3. NAG-001637 (Isolated points of a strict affinoid space). An isolated point of a strict affinoid space is rigid and has local analytic dimension zero. In particular a pure one-dimensional strict affinoid space has no isolated points.

Proof. Let X=M(A)X=\mathcal M(A) and suppose {x}\{x\} is open. It is also closed because XX is Hausdorff. Glue the constant function one on {x}\{x\} and zero on its complement using Theorem 2.40.6. The resulting global function is an idempotent e∈Ae\in A. The direct factor B=A/(1−e)B=A/(1-e) is a strict affinoid algebra whose spectrum is {x}\{x\}. Theorem 1.12.3 supplies a rigid point of its nonempty spectrum, necessarily xx. Consequently H(x)/K\mathcal H(x)/K is finite and dK(x)=0d_K(x)=0 by Lemma 3.24.2. Theorem 3.22.7 now gives dim⁡KM(B)=0\dim_K\mathcal M(B)=0. Since this is an open neighborhood of xx, its local dimension in XX is zero. ◻

Theorem 3.25.4. NAG-001638 (The boundary of a strict affinoid curve). Let XX be a nonempty strict affinoid space of dimension at most one with no isolated points. Then its absolute analytic boundary equals its Shilov boundary. In particular this holds for every pure one-dimensional strict affinoid space.

Proof. Write X=M(A)X=\mathcal M(A) and R=A~R=\widetilde A. By Lemma 3.25.2, RR has dimension at most one. Theorem 3.16.7 identifies its minimal primes with the Shilov points, each having a unique analytic preimage. A minimal prime p\mathfrak p cannot also be maximal. Indeed, choose finitely many generators fˉi\bar f_i of p\mathfrak p and lifts fi∈A∘f_i\in A^\circ. If p\mathfrak p were maximal, then V(p)={p}V(\mathfrak p)=\{\mathfrak p\} and red⁡−1(p)={y∈X:∣fi(y)∣<1 for every i}\operatorname{red}^{-1}(\mathfrak p) =\{y\in X:|f_i(y)|<1\text{ for every }i\} would be an open singleton, contrary to the hypothesis. In a ring of dimension at most one, a nonmaximal prime is minimal: otherwise a strictly smaller minimal prime and a strictly larger maximal prime would form a chain of length two. Thus the nonclosed points of Spec⁡R\operatorname{Spec}R are exactly its generic points. Theorem 3.25.1 says that their inverse image is the analytic boundary, while Theorem 3.16.7 says that the same inverse image is the Shilov boundary. The last assertion follows from Lemma 3.25.3. ◻

Corollary 3.25.5. NAG-001639 (Weighted Gauss points in the bidisc boundary). For X=M(K⟨T,S⟩)X=\mathcal M(K\langle T,S\rangle), the Shilov boundary is {ζ1,1}\{\zeta_{1,1}\}. For every real 0<ρ<10<\rho<1, the weighted Gauss point ζ1,ρ\zeta_{1,\rho} lies in the absolute analytic boundary of XX.

Proof. The unit Gauss norm is multiplicative and is the spectral norm, so its point alone is a boundary. The minimality assertion of Lemma 3.12.4 gives the Shilov statement. Reduction of the coefficient series gives A~=K~[t,s]\widetilde A=\widetilde K[t,s]: only finitely many coefficients of a restricted series can have norm one, and the others disappear modulo A∘∘A^{\circ\circ}. At ζ1,ρ\zeta_{1,\rho} the reduction kernel is (s)(s). Indeed, a power-bounded series has value one there exactly when its part of SS-degree zero has a coefficient of norm one. All positive SS-degrees have norm at most ρ<1\rho<1. The prime (s)(s) is not maximal, since its quotient is K~[t]\widetilde K[t]. Apply Theorem 3.25.1. ◻

References. The classical strict interior criterion and the curve/bidisc distinction are explained in [TemIntro] (3.4.2.3(vi), 3.4.2.5). The proofs here use the internal graded innerness, reduction and dimension theorems.

3.26 The standard monomial chart

Proposition 3.26.1. NAG-002016 (The monomial simplex and its residual data). Let KK be complete and nontrivially valued, let 0<∣a∣<10<|a|<1, and let 0≤r≤d0\leq r\leq d. Put A0=K∘⟨X0,…,Xr,T1,…,Td−r⟩/(X0⋯Xr−a),A=A0[1/a].A_0=K^\circ\langle X_0,\ldots,X_r,T_1,\ldots,T_{d-r}\rangle /(X_0\cdots X_r-a),\qquad A=A_0[1/a]. Then A0A_0 is admissible and M(A)\mathcal M(A) is the rational domain ∣a∣≤∣X1⋯Xr∣|a|\leq|X_1\cdots X_r| in the unit dd-dimensional polydisc with coordinates X1,…,Xr,TX_1,\ldots,X_r,T. The monomial points parametrized by Δa={(u0,…,ur)∈R≥0r+1:u0+⋯+ur=v(a)}\Delta_a=\{(u_0,\ldots,u_r)\in\mathbb R_{\geq0}^{r+1}: u_0+\cdots+u_r=v(a)\} form a subspace homeomorphic to Δa\Delta_a. At every such point xux_u, the completed residue field is the completion of K(X1,…,Xr,T)K(X_1,\ldots,X_r,T) for its Gauss valuation of weights u1,…,ur,0,…,0u_1,\ldots,u_r,0,\ldots,0, and sK(xu)+tK(xu)=dim⁡xuM(A)=d.s_K(x_u)+t_K(x_u)=\dim_{x_u}\mathcal M(A)=d. If all ui>0u_i>0, specialization is the generic point of the stratum X0=⋯=Xr=0X_0=\cdots=X_r=0, of dimension d−rd-r. Any finite family of Laurent monomials in the XiX_i with coefficients in KK and value one at xux_u becomes integral on a formal affine neighborhood of its specialization after an admissible formal modification. Its nonzero residues there are the original residues in H(xu)~\widetilde{\mathcal H(x_u)}.

Proof. Integral normal form. Write M=X0⋯XrM=X_0\cdots X_r. In a restricted series replace a monomial XνTμX^\nu T^\mu by aqXν−q1Tμa^qX^{\nu-q\mathbf 1}T^\mu, where q=min⁡iνiq=\min_i\nu_i. There is a unique resulting series supported on exponent vectors whose minimum XX-exponent is zero. Its coefficient at such a vector (λ,μ)(\lambda,\mu) is ∑q≥0aqcλ+q1,μ.\sum_{q\geq0}a^q c_{\lambda+q\mathbf1,\mu}. These sums converge and preserve integral coefficients. The identity Mq−aq=(M−a)∑j=0q−1ajMq−1−jM^q-a^q=(M-a)\sum_{j=0}^{q-1}a^jM^{q-1-j} gives a contractive division operator on polynomials. The coefficients of its value on a restricted series are ∑q≥1aq−1cν+q1,μ\sum_{q\geq 1}a^{q-1}c_{\nu+q\mathbf 1,\mu}; the finite-tail condition makes this a restricted series again. Thus the polynomial division identities extend continuously, and their kernel is exactly (M−a)(M-a), not just its closure. The quotient is the torsion-free module of these normal forms. Hence A0A_0 is admissible by Theorem 3.14.4. Inverting aa gives its complete strict affinoid algebra by Proposition 3.14.7. Eliminating X0=a/(X1⋯Xr)X_0=a/(X_1\cdots X_r) is exactly the rational-localization presentation asserted in the statement. For r=0r=0 the empty product is one and the domain is the whole dd-polydisc.

Points and topology. At u∈Δau\in\Delta_a, give XiX_i value uiu_i and each TjT_j value zero. After eliminating X0X_0, the Gauss norm is multiplicative by Proposition 1.2.2, and all original coordinates have norm at most one. It therefore extends to a bounded multiplicative seminorm on AA. A polynomial becomes a finite Laurent polynomial after elimination; its value is a maximum of finitely many continuous functions of uu. Restricted-series tails tend to zero uniformly on Δa\Delta_a, because all original coordinates have norm at most one. Thus the same continuity holds for every element of AA. The inverse on the image is x↦(v(X0(x)),…,v(Xr(x)))x\mapsto(v(X_0(x)),\ldots,v(X_r(x))); the coordinates are invertible in AA, so these functions are continuous. This proves the claimed homeomorphism.

Residue fields and dimension. The Laurent polynomial algebra K[X1±1,…,Xr±1,T]K[X_1^{\pm 1},\ldots,X_r^{\pm 1},T] is dense in AA and its evaluation at xux_u is injective. Indeed distinct normal-form monomials give distinct Laurent exponents, so their Gauss values cannot cancel. Passing to fractions and then completion identifies H(xu)\mathcal H(x_u) with the stated completed Gauss field; density is preserved on taking fractions because inversion is continuous at every nonzero element. Proposition 3.8.1 gives dK(xu)=dd_K(x_u)=d. The rational-domain description and Corollary 3.22.8 give dim⁡M(A)≤d\dim\mathcal M(A)\leq d. The local Abhyankar inequality in Theorem 3.22.9 now gives d=dK(xu)≤dim⁡xuM(A)≤dim⁡M(A)≤d.d=d_K(x_u)\leq\dim_{x_u}\mathcal M(A) \leq\dim\mathcal M(A)\leq d.

If all ui>0u_i>0, every XiX_i reduces to zero, whereas the reductions of the TjT_j are algebraically independent by the same Gauss calculation. The kernel of special-fibre evaluation is consequently (x0,…,xr)(x_0,\ldots,x_r). The stratum is affine (d−r)(d-r)-space, and its dimension is d−rd-r by [StacksAlg] (Tag 00P0). This proves the specialization assertion.

Making a residue visible on a model. Write one of the specified monomials as h=f/gh=f/g with f,g∈A0f,g\in A_0 monomials having integral coefficients. This is possible after multiplication of numerator and denominator by a scalar. The denominator g=cXνg=cX^\nu divides a nonzero element of K∘K^\circ: choose N≥max⁡iνiN\geq\max_i\nu_i and use gXN1−ν=caNgX^{N\mathbf 1-\nu}=ca^N. Hence (f,g)(f,g) is an open ideal. The gg-chart of its admissible blowup has f/gf/g integral by Theorem 3.15.1, and contains the unique generic lift of xux_u since ∣f(xu)∣=∣g(xu)∣|f(x_u)|=|g(x_u)|. For a finite family of monomials, blow up the product of these open ideals. Lemma 3.17.3 makes the resulting modification dominate all the individual blowups. The intersection of the inverse images of their gg-charts is a formal open containing the specialization of the lifted point, by functoriality of specialization and Proposition 3.17.2. Take a formal affine neighborhood inside that intersection. All the chosen monomials are integral there, and their value one at the point makes their reductions nonzero. Generic-fibre invariance and evaluation identify them with the original residues. ◻

Lemma 3.26.2. NAG-002017 (A spectral unit ball need not be an admissible model). Assume KK is nontrivially valued and r∉∣K×∣r\notin\sqrt{|K^\times|}. For the one-radius field KrK_r, its valuation ring Kr∘K_r^\circ is not topologically of finite type over K∘K^\circ.

Proof. Choose π∈K\pi\in K with 0<∣π∣<10<|\pi|<1. Every element of KrK_r becomes power-bounded after multiplication by a power of π\pi, so Kr∘[1/π]=KrK_r^\circ[1/\pi]=K_r. If the torsion-free ring Kr∘K_r^\circ were topologically of finite type, Theorem 3.14.4 and Proposition 3.14.7 would make KrK_r a strict KK-affinoid algebra. But its Laurent coordinate has spectral norm rr, contrary to Lemma 3.16.6, which puts every nonzero spectral value of a strict affinoid algebra in ∣K×∣\sqrt{|K^\times|}. ◻

3.27 Flatness over the ground field

Throughout this section KK is a complete non-Archimedean field, with arbitrary characteristic and possibly trivial absolute value. Completed tensor products use the projective norm.

Lemma 3.27.1. NAG-002018 (Banach coefficients over an affinoid algebra). Let CC be a generalized KK-affinoid algebra and let EE be a Banach KK-space. Put T=E⊗^KCT=E\widehat\otimes_K C. For every finite canonical CC-module NN, there is a natural algebraic CC-linear isomorphism T⊗CN → ∼  E⊗^KN,(e⊗c)⊗n⟼e⊗cn.T\otimes_C N\ \xrightarrow{\ \sim\ }\ E\widehat\otimes_K N, \qquad (e\otimes c)\otimes n\longmapsto e\otimes cn. The algebraic CC-module TT is flat.

Proof. The algebra CC is Noetherian by Theorem 1.18.2. Choose a finite presentation Ca→DCb→N→0C^a\xrightarrow{D}C^b\to N\to 0, and put R=im⁡DR=\operatorname{im}D. The image RR is closed; the maps Ca→RC^a\to R and Cb→NC^b\to N are norm-strict for the induced and canonical norms, by Theorem 1.18.3. Apply Theorem 2.33.3 to their kernel sequences. The finite coordinate injections and projections identify E⊗^KCjE\widehat\otimes_K C^j with TjT^j. We obtain an exact sequence Ta→DTb⟶E⊗^KN⟶0.T^a\xrightarrow{D}T^b\longrightarrow E\widehat\otimes_K N \longrightarrow0. By [StacksAlg] (Tag 00DF), ordinary tensoring gives the same cokernel presentation for T⊗CNT\otimes_C N. The resulting isomorphism is the displayed map: the bounded action on NN extends its pure-tensor formula in the first factor. This also proves naturality.

For a finitely generated ideal I⊂CI\subset C, its induced norm is a canonical finite-module norm. Under the comparison just proved, the map I⊗CT→TI\otimes_C T\to T is E⊗^KI→E⊗^KCE\widehat\otimes_K I\to E\widehat\otimes_K C. It is injective by Theorem 2.33.2. The finite-ideal criterion [StacksAlg] (Tag 00HD) proves flatness. ◻

Corollary 3.27.2. NAG-002019 (Finite modules on an affinoid product). Let B,CB,C be generalized KK-affinoid algebras, let MM be a finite canonical BB-module, and put D=B⊗^KCD=B\widehat\otimes_K C. There is a natural bounded DD-linear isomorphism with bounded inverse D⊗BM → ∼  M⊗^KC,(b⊗c)⊗m⟼bm⊗c.D\otimes_B M\ \xrightarrow{\ \sim\ }\ M\widehat\otimes_K C, \qquad (b\otimes c)\otimes m\longmapsto bm\otimes c. The left side has its canonical finite-DD-module norm. This module is flat over CC.

Proof. Proposition 2.33.8 makes DD a generalized affinoid algebra. Apply Theorem 2.33.3 to a finite BB-module presentation of MM, split into its two norm-strict kernel sequences as in the preceding proof. It identifies M⊗^KCM\widehat\otimes_K C with the quotient of DbD^b by the image of the same presentation matrix on DaD^a. This is D⊗BMD\otimes_B M as an algebraic module. The quotient is norm-strict, so its norm is equivalent to the canonical finite-DD-module norm by Theorem 1.18.3. The formula on pure tensors identifies the DD-actions and gives naturality. Lemma 3.27.1, applied to the Banach KK-space MM, gives CC-flatness. ◻

Proposition 3.27.3. NAG-002020 (Product flatness on ordinary stalks). Let Y,XY,X be good KK-analytic spaces, let F\mathcal F be coherent on YGY_G, and put P=Y×KXP=Y\times_KX. For w∈Pw\in P with image x∈Xx\in X, the module (pY∗F)w(p_Y^*\mathcal F)_w is flat over OX,x\mathcal O_{X,x}.

Proof. Write y=pY(w)y=p_Y(w). Choose ordinary affinoid neighborhoods U=M(B)U=\mathcal M(B) of yy and V0=M(C0)V_0=\mathcal M(C_0) of xx. Their product is affinoid by Theorem 2.40.11 and contains the intersection of the inverse images of their ordinary interiors. Thus it is a neighborhood of ww, and these products show that PP is good. By Theorem 2.39.6, F∣UG\mathcal F|_{U_G} is represented by a finite canonical BB-module MM. Put Q=(pY∗F)wQ=(p_Y^*\mathcal F)_w.

First fix a rational domain V⊂V0V\subset V_0 containing an ordinary neighborhood of xx. Write CV=O(V)C_V=\mathcal O(V) and DV=B⊗^KCVD_V=B\widehat\otimes_K C_V. On U×KVU\times_KV the pullback sheaf is represented by QV=DV⊗BM≃M⊗^KCV.Q_V=D_V\otimes_B M\simeq M\widehat\otimes_K C_V. This follows from the finite presentation defining coherent pullback; Corollary 3.27.2 shows that QVQ_V is CVC_V-flat. Affinoid neighborhoods WW of ww contained in U×KVU\times_KV are cofinal among its ordinary neighborhoods in PP. Each WW is an affinoid domain of this product, so its algebra AWA_W is DVD_V-flat by Proposition 2.38.2. The restricted module QV⊗DVAWQ_V\otimes_{D_V}A_W is therefore CVC_V-flat: tensoring an injection first with QVQ_V and then with AWA_W preserves injectivity.

The module version of Proposition 2.33.16, obtained from the same finite presentation, gives Q=lim→⁡W(QV⊗DVAW).Q=\varinjlim_W\bigl(Q_V\otimes_{D_V}A_W\bigr). For a finitely generated ideal J⊂CVJ\subset C_V, an element of J⊗CVQJ\otimes_{C_V}Q is represented at one stage WW. If its image in QQ vanishes, that image vanishes after one further restriction. Flatness at the new stage makes the tensor zero there. The finite-ideal criterion proves that QQ is flat over each fixed CVC_V.

Now put R=OX,x=lim→⁡VCVR=\mathcal O_{X,x}=\varinjlim_V C_V. Let I=(a1,…,am)⊂RI=(a_1,\ldots,a_m)\subset R, and let z=∑iai⊗qi∈I⊗RQz=\sum_i a_i\otimes q_i\in I\otimes_RQ have zero image in QQ. Lift the aia_i to elements a~i\widetilde a_i of a single CVC_V, and put J=(a~1,…,a~m)⊂CVJ=(\widetilde a_1,\ldots,\widetilde a_m)\subset C_V. The coefficients qiq_i remain in QQ. Then zV=∑ia~i⊗qi∈J⊗CVQz_V=\sum_i\widetilde a_i\otimes q_i\in J\otimes_{C_V}Q maps to zz and has zero image in QQ. The CVC_V-flatness just proved gives zV=0z_V=0, hence z=0z=0. A final application of [StacksAlg] (Tag 00HD) proves RR-flatness. ◻

Theorem 3.27.4. NAG-002021 (Coherent sheaves are flat over the ground field). Let YY be a KK-analytic space and let F\mathcal F be coherent on YGY_G. Then F\mathcal F is analytically flat over M(K)\mathcal M(K). More generally, for every KK-analytic space XX, the sheaf pY∗Fp_Y^*\mathcal F on Y×KXY\times_KX is analytically XX-flat. In particular the projection Y×KX→XY\times_KX\to X is analytically flat.

Proof. First let Y,XY,X be good. Let L/KL/K be any complete valued extension, let X′X' be a good LL-analytic space with a morphism to XX, and consider any point of the base-changed product. By Theorem 2.40.12 and Corollary 2.33.9, the base change and its coherent pullback identify with (Y×KX)×XX′≃YL×LX′,pYL∗FL.(Y\times_KX)\times_XX'\simeq Y_L\times_LX', \qquad p_{Y_L}^*\mathcal F_L. The space YLY_L is good: the inverse image of an ordinary affinoid neighborhood is affinoid and contains the inverse image of its ordinary interior. The sheaf FL\mathcal F_L is coherent by its local finite presentations. Proposition 3.27.3, applied over LL, proves naive flatness at every point of this product. These are all the good base-change tests, over all complete extensions, in Definition 2.9.1.

For general Y,XY,X, choose affinoid analytic nets on both spaces. Their affinoid products form an analytic atlas of Y×KXY\times_KX by Theorem 2.40.11. On every such product, the restricted sheaf is the pullback of the finite-module sheaf on the source chart. The good-space case supplies all the required flatness tests there. The GG-local definition therefore gives the assertion on the general product. Taking X=M(K)X=\mathcal M(K) proves the first statement; taking F=OY\mathcal F=\mathcal O_Y proves the projection statement. ◻