2 · Analytic spaces, morphisms and fibres

370 NAG Tags

Affinoid algebras provide compact analytic pieces. To work locally on them, inequalities must define analytic domains and functions must glue on overlaps. We first construct this local language. We then describe products and fibres, where the completed residue field of a point becomes indispensable. The final sections explain the geometric hypotheses used later in Stokes, finite-flat degree calculations and model comparisons.

2.1 Domains and bounded localization

Definition 2.1.1. Let X=M(A)X=\mathcal M(A). An affinoid domain is a subset V⊂XV\subset X represented by a bounded map A→AVA\to A_V such that M(AV)→X\mathcal M(A_V)\to X identifies its points with VV, and every bounded morphism M(B)→X\mathcal M(B)\to X whose image lies in VV factors uniquely through M(AV)\mathcal M(A_V). The uniqueness is in the affinoid category, not merely among maps of point sets.

The algebra A[f−1]A[f^{-1}] alone is not an analytic localization: it has no prescribed bound on f−1f^{-1}. By adjoining an inverse with a bound, we obtain an analytic domain. The same principle gives three useful presentations: ∣fi∣≤riA{r−1T}/(Ti−fi)(Weierstrass)∣f∣≥rA{rS}/(fS−1)(Laurent)∣fi∣≤ri∣g∣A{r−1T}/(gTi−fi)(rational).\begin{array}{c|c} |f_i|\leq r_i&A\{r^{-1}T\}/(T_i-f_i)\quad\text{(Weierstrass)}\\ |f|\geq r&A\{rS\}/(fS-1)\quad\text{(Laurent)}\\ |f_i|\leq r_i|g|&A\{r^{-1}T\}/(gT_i-f_i)\quad\text{(rational)}. \end{array} In the rational case g,f1,…,fmg,f_1,\ldots,f_m must generate the unit ideal. Their common zero locus would otherwise introduce points not described by ratios. The universal properties of these presentations are proved in Theorem 2.31.2. Theorem 2.36.6 then gives a finite original-algebra rational cover of every affinoid domain. We begin with the point calculation.

References. [TemIntro] (§3.2, especially 3.2.3.4); [Ber93] (§1.3).

2.2 Rational domains and logarithmic coordinates

A rational-domain presentation converts finitely many inequalities into a Banach algebra. Let g,f1,…,fm∈Ag,f_1,\ldots,f_m\in A generate the unit ideal and let ri>0r_i>0. Put D={x∈M(A):∣fi(x)∣≤ri∣g(x)∣ (1≤i≤m)},D=\{x\in\mathcal M(A):|f_i(x)|\leq r_i|g(x)|\ (1\leq i\leq m)\}, B=A{r1−1T1,…,rm−1Tm}/(gTi−fi)i=1m.B=A\{r_1^{-1}T_1,\ldots,r_m^{-1}T_m\}/(gT_i-f_i)_{i=1}^m. The quotient is the affinoid quotient, with its complete quotient topology. Theorem 2.31.2 proves its rational-domain universal property. The following proof identifies its points and the role of the unit-ideal hypothesis.

Proposition 2.2.1. NAG-000626 The map M(B)→M(A)\mathcal M(B)\to\mathcal M(A) is a homeomorphism onto DD. For x∈Dx\in D, the corresponding evaluation sends TiT_i to fi(x)/g(x)f_i(x)/g(x) in H(x)\mathcal H(x). In particular gg is a unit in BB.

Proof. Write a0g+∑iaifi=1a_0g+\sum_i a_if_i=1 in AA. In BB this becomes g(a0+∑iaiTi)=1g(a_0+\sum_i a_iT_i)=1. Thus gg is invertible. A point of BB satisfies ∣Ti∣≤ri|T_i|\leq r_i, and hence its image belongs to DD. Conversely, if x∈Dx\in D and g(x)=0g(x)=0, then every fi(x)f_i(x) is zero, contradicting the displayed unit-ideal relation. Therefore g(x)≠0g(x)\ne 0 and the quotients fi(x)/g(x)f_i(x)/g(x) have sizes at most rir_i. Substitution into a convergent power series defines a bounded homomorphism to H(x)\mathcal H(x), which kills the defining relations of BB. This gives a point over xx.

For any point over xx, the relations force those same values of the variables in the completed residue field. Values on polynomial expressions are therefore fixed by the seminorm of xx on fractions with denominator a power of gg. Such polynomial expressions are dense in BB, and bounded seminorms are continuous, so the point is unique. Finally, the map is continuous, its source is compact and its target is Hausdorff. A continuous bijection onto DD is consequently a homeomorphism. ◻

Example 2.2.2. The domain ∣f∣≥r>0|f|\geq r>0 is represented by adjoining f−1f^{-1} with bound r−1r^{-1}. On its interior, −log⁡∣f∣-\log|f| is an original smooth function: ff is now an analytic unit and defines a one-dimensional moment map. If ff has zeros elsewhere, the same logarithm is not a finite smooth function across those zeros. This is the local distinction used when regularized maxima discard inactive logarithmic branches.

Strict inequalities give ordinary open subsets because evaluation is continuous. This elementary fact explains why rational-domain arguments often begin with two sets of bounds: weaker non-strict bounds define a compact affinoid domain, while stronger bounds place the point or support in its ordinary interior. The existence of a suitable presentation around every point is a local geometric input; it should be distinguished from the explicit pointwise construction above.

Example 2.2.3. (Domain versus closed subspace). The closed disc ∣T∣≤r|T|\leq r in the analytic affine line is a domain: it retains the analytic dimension of the ambient curve. The zero locus T=0T=0 is a closed analytic subspace, defined by a quotient of the function algebra. The thickening Tm=0T^m=0 has the same point but a different structure sheaf. A closed subset in the ordinary topology therefore need not be a closed analytic subspace.

Proposition 2.2.4. NAG-000197 If f(x)≠0f(x)\ne 0 on an affinoid space, there is an ordinary neighborhood of xx on which ff is an analytic unit. On any analytic domain where ff is a unit, the function −log⁡∣f∣-\log|f| is finite and continuous.

Proof. Choose 0<r<∣f(x)∣0<r<|f(x)|. The Laurent domain ∣f∣≥r|f|\geq r contains the ordinary open neighborhood ∣f∣>r|f|>r of xx, and its defining algebra contains f−1f^{-1}. Restrict that inverse to the open neighborhood. Continuity follows from continuity of evaluation and of the logarithm on positive real numbers. ◻

2.3 G-topology and the structure sheaf

A finite affinoid-domain cover can contain a point only on interfaces; it need not contain that point in the ordinary interior of one member. This is why analytic domains form a site rather than just a basis of ordinary open sets. On an affinoid XX, first use compact analytic domains, finite unions of affinoid domains, with finite domain covers. The assignment V↦AVV\mapsto A_V on affinoid domains extends to the analytic structure sheaf OXG\mathcal O_{X_G}.

Theorem 2.3.1. NAG-000234 (Tate acyclicity). Let X=M(A)X=\mathcal M(A) and let X=⋃i=1mViX=\bigcup_{i=1}^m V_i be a finite affinoid-domain cover. For a finite AA-module MM, the augmented Čech complex of the corresponding localized modules is exact. In particular, for M=AM=A, 0⟶A⟶∏iAVi⟶∏i,jAVi∩Vj⟶⋯0\longrightarrow A\longrightarrow\prod_i A_{V_i} \longrightarrow\prod_{i,j}A_{V_i\cap V_j}\longrightarrow\cdots is exact, with the restriction maps and alternating differentials. The maps are admissible for the natural Banach-module topologies.

Proof. Apply Theorem 2.38.1 to AA, MM and the stated cover. Its all-tuples version gives exactly the displayed differentials. Proposition 1.18.7 identifies the localized ordinary modules with their completed tensors and canonical Banach topologies. Norm-strict exactness is therefore precisely the asserted admissibility and exactness. ◻

References. [TemIntro] (3.3.2.1); the original proof routes are recorded in Section 2.11.2.

Proposition 2.3.2. NAG-000194 (Gluing across a circle). Let 0<s<10<s<1. Cover the closed unit disc by V−={∣T∣≤s}V_- =\{|T|\leq s\} and V+={s≤∣T∣≤1}V_+=\{s\leq|T|\leq 1\}. With A−=K{s−1T},A+=K{T,sT−1},C=K{s−1T,sT−1}.A_-=K\{s^{-1}T\},\quad A_+=K\{T,sT^{-1}\},\quad C=K\{s^{-1}T,sT^{-1}\}. The sequence 0→K⟨T⟩→A−⊕A+→C→00\to K\langle T\rangle\to A_-\oplus A_+\to C\to 0, with last map (f−,f+)↦f−−f+(f_-,f_+)\mapsto f_--f_+, is exact.

Proof. A Laurent series in CC splits into its nonnegative and negative powers. The nonnegative part converges on V−V_-; the negative part converges on V+V_+ because its worst bound is attained at radius ss. This proves surjectivity of the difference map. If two restrictions agree on the circle, uniqueness of Laurent coefficients shows that the annular series has no negative powers. Its nonnegative coefficients tend to zero at radius one, since it belongs to A+A_+. They therefore define an element of K⟨T⟩K\langle T\rangle restricting to both functions. Injectivity follows from uniqueness of the coefficients. ◻

For a finite AA-module MM, the sheaf associated with it has sections M⊗AAVM\otimes_A A_V on an affinoid domain VV. Finite-module completeness makes this ordinary tensor product sufficient. The converse is Kiehl’s coherence theorem: coherent sheaves on XX are equivalent to finite AA-modules. Thus finite presentations of functions and their relations glue, just as the functions themselves do. This is used later for ideals of closed subspaces, analytic line bundles and finite-flat modules.

References. [TemIntro] (3.3.4.1).

2.4 From affinoid atlases to analytic spaces

An analytic space is obtained by gluing affinoid spaces along analytic domains with compatible bounded restriction maps. In the general definition, an atlas is a net of compact affinoid pieces: locally, finitely many pieces containing a point cover a neighborhood of it, and the overlap data admit compatible affinoid refinements. Morphisms are locally represented, after such refinements, by bounded maps of affinoid algebras. Section 2.40 constructs the atlas, proves independence under refinement and glues its structure sheaf and morphisms, following [Ber93] (§§1.1–1.3).

Definition 2.4.1. An analytic space is good if every point has an affinoid neighborhood: an affinoid domain containing an ordinary open neighborhood of the point. An analytic domain of a general space is a locally represented subset with the corresponding universal factorization property. Its structure is unique when it exists.

For good spaces, ordinary local arguments can be performed inside one affinoid neighborhood. Open subspaces remain good by the rational-neighborhood basis, and the construction in Theorem 2.33.23 makes analytifications of varieties good. In contrast, removing the open bidisc from the closed bidisc of the same radius gives a compact analytic domain which is not good. Theorem 3.23.2 proves this example, recalled from [TemIntro] (4.2.1.4), in every dimension at least two.

The ordinary structure sheaf is obtained from the analytic sheaf on ordinary opens. For a good space XX, set OX,x=lim→⁡x∈UOX(U)\mathcal O_{X,x}=\varinjlim_{x\in U}\mathcal O_X(U) over ordinary open neighborhoods. A germ is invertible precisely when its value at xx is nonzero, by Proposition 2.2.4. Consequently this is a local ring with maximal ideal of germs vanishing at xx.

2.5 Local residue fields and completed residue fields

Definition 2.5.1. For a point of a good analytic space, distinguish κX(x)=OX,x/mx,H(x)=κX(x)^,H(x)~=H(x)∘/H(x)∘∘.\kappa_X(x)=\mathcal O_{X,x}/\mathfrak m_x, \qquad \mathcal H(x)=\widehat{\kappa_X(x)}, \qquad \widetilde{\mathcal H(x)} =\mathcal H(x)^\circ/\mathcal H(x)^{\circ\circ}. The first is the residue field of an ordinary local ring, the second its completion for the induced absolute value, and the third the residue field of a valuation ring.

The identification of the middle field with the affinoid construction of Chapter 1 follows from the evaluation maps on neighborhoods. For X=M(A)X=\mathcal M(A), the fraction field of A/pxA/\mathfrak p_x embeds in κX(x)\kappa_X(x): a denominator nonzero at xx is invertible on a neighborhood. Both fields embed in H(x)\mathcal H(x), and the former is dense by its construction. Thus the completion of κX(x)\kappa_X(x) is exactly H(x)\mathcal H(x). This also explains why the completed field is the appropriate invariant for fibres.

If V⊂XV\subset X is an analytic domain containing xx, its completed residue field is canonically the same H(x)\mathcal H(x), by Corollary 2.33.15. For good VV and XX, the ordinary local ring and its uncompleted residue field can change when VV is not an ordinary neighborhood. If it is a neighborhood, the stalks agree. The example in Section 2.9 uses precisely this change of local ring.

References. [Ber93] (§2); [TemIntro] (4.2.1).

2.6 Completed tensor products and analytic fibres

A product must allow evaluation at points taking values in compatible complete extensions. The algebraic tensor product has the required algebraic universal property, but generally lacks completeness. On B⊗ACB\otimes_A C, use the quotient projective seminorm obtained from ∥z∥π=inf⁡z=∑ibi⊗cimax⁡i∥bi∥∥ci∥.\|z\|_\pi=\inf_{z=\sum_i b_i\otimes c_i} \max_i\|b_i\|\|c_i\|. After quotienting by its zero seminorm and completing, one obtains B⊗^ACB\widehat\otimes_A C. Its universal property concerns compatible bounded maps. Affinoid completed tensor products are affinoid, and the constructions glue.

Theorem 2.6.1. NAG-000170 Fibre products exist in the analytic category. For affinoid morphisms A→B,A→CA\to B,A\to C, and for a complete extension L/KL/K, M(B)×M(A)M(C)=M(B⊗^AC),M(A)L=M(A⊗^KL).\mathcal M(B)\times_{\mathcal M(A)}\mathcal M(C) =\mathcal M(B\widehat\otimes_A C),\qquad \mathcal M(A)_L=\mathcal M(A\widehat\otimes_K L). If f:Y→Xf:Y\to X and x∈Xx\in X, its analytic fibre is the H(x)\mathcal H(x)-analytic space Yx=Y×XM(H(x))Y_x=Y\times_X\mathcal M(\mathcal H(x)). For affinoid X,YX,Y, its algebra is B⊗^AH(x)B\widehat\otimes_A\mathcal H(x). Its underlying topological space identifies with f−1(x)f^{-1}(x).

Proof. Proposition 2.33.8 constructs the affinoid completed tensor product and proves its bounded universal property. Theorem 2.40.11 glues these products on arbitrary analytic atlases. Theorem 2.40.12 supplies complete scalar extension. The actual fibre identification, including its topology and its field of definition, is Theorem 2.33.11 in affinoid charts and Theorem 2.40.13 after gluing. ◻

References. [Ber93] (§1.4); [TemIntro] (3.1.4.1, 4.1.5–4.1.7).

The fibre and scalar-extension formulas use the analytic category allowing complete extensions of the ground field, as in [TemIntro] (4.1.6). In particular M(H(x))\mathcal M(\mathcal H(x)) need not be a KK-affinoid space. Treating it as an analytic space over its own field is what makes the fibre formula valid for nonclassical points.

Example 2.6.2. For the projection M(K⟨S,T⟩)→M(K⟨S⟩)\mathcal M(K\langle S,T\rangle)\to\mathcal M(K\langle S\rangle), the fibre at xx is M(H(x)⟨T⟩)\mathcal M(\mathcal H(x)\langle T\rangle). At a nonclassical xx, replacing H(x)\mathcal H(x) by KK would discard the value of SS and its residual geometry. In the fibre, SS is a scalar S(x)∈H(x)S(x)\in\mathcal H(x), while TT remains a variable.

Example 2.6.3. Let L/KL/K be a nontrivial finite Galois extension. Both M(L)\mathcal M(L) and M(K)\mathcal M(K) have one point, but M(L)×M(K)M(L)≃∐σ∈Gal⁡(L/K)M(L).\mathcal M(L)\times_{\mathcal M(K)}\mathcal M(L) \simeq\coprod_{\sigma\in\operatorname{Gal}(L/K)}\mathcal M(L). Indeed L⊗^KL=L⊗KL≃∏σLL\widehat\otimes_K L=L\otimes_K L\simeq\prod_\sigma L, the last map sending a⊗ba\otimes b to (aσ(b))σ(a\sigma(b))_\sigma. Completion adds nothing in finite dimension. Thus the underlying set of an analytic product need not be the Cartesian product of the underlying sets.

Proposition 2.6.4. NAG-000173 Suppose a finite affinoid morphism is given by a finite locally free AA-algebra BB of constant rank dd. For x∈M(A)x\in\mathcal M(A) put F=H(x)F=\mathcal H(x) and E=B⊗AFE=B\otimes_A F. Then EE is an Artinian FF-algebra of dimension dd. Writing it as the product of its local factors EjE_j with residue fields FjF_j, d=∑jlength⁡Ej(Ej)[Fj:F].d=\sum_j\operatorname{length}_{E_j}(E_j)[F_j:F].

Proof. Finite projectivity commutes with scalar extension, giving dim⁡FE=d\dim_FE=d; finite-dimensional algebras are complete by Lemma 1.12.1. The Artinian product decomposition and finite length follow from [StacksAlg] (Tag 00JB). A composition series of EjE_j as an EjE_j-module has length⁡Ej(Ej)\operatorname{length}_{E_j}(E_j) factors, each isomorphic to FjF_j. Taking FF-dimensions gives the formula. ◻

This is the finite-fibre degree formula used in Chapter 22. It retains nilpotent length and residue-field degree separately. It does not determine the normalization of integrals after replacing the ground field; that is the separate comparison in Chapter 23.

2.7 Boundary, finite maps and properness

Definition 2.7.1. For an affinoid morphism Y=M(B)→X=M(A)Y=\mathcal M(B)\to X=\mathcal M(A), a point yy lies in Int⁡(Y/X)\operatorname{Int}(Y/X) if there exists an admissible surjection A{r1−1T1,…,rm−1Tm}↠Bwith∣Ti(y)∣<rifor all i,A\{r_1^{-1}T_1,\ldots,r_m^{-1}T_m\}\twoheadrightarrow B \quad\text{with}\quad |T_i(y)|<r_i\quad\text{for all }i, Its complement is the relative analytic boundary ∂(Y/X)\partial(Y/X). Absolute interior and boundary are relative to M(K)\mathcal M(K). For general morphisms these notions are defined locally and are independent of the chosen affinoid descriptions.

For general spaces the precise definition tests all target affinoid domains, as in Definition 2.44.4; Theorem 2.46.3 proves its locality. The strict inequalities above are existential in the relative presentation. The graded criterion in Theorem 3.18.10 gives an intrinsic description.

NAG-000157 · Open at full scope

Theorem 2.7.2. NAG-000157 Relative interior is open and behaves locally on the base under analytic-domain covers. If V⊂XV\subset X is an analytic domain, Int⁡(V/X)\operatorname{Int}(V/X) equals its ordinary topological interior in XX. An affinoid morphism M(B)→M(A)\mathcal M(B)\to\mathcal M(A) is boundaryless if and only if A→BA\to B is finite and admissible. Finite morphisms are stable under base change.

Proof. Theorem 2.46.2 proves ordinary openness for arbitrary analytic spaces, and Theorem 2.46.3 gives the exact boundary identity on analytic-domain covers of the base. Proposition 2.44.5 identifies the relative interior of every analytic-domain inclusion with its ordinary topological interior. Theorem 2.43.8 proves the affinoid finite criterion, including the canonical finite-module topology and hence admissibility over trivially valued fields. Finally, Proposition 2.33.19 identifies a finite affinoid base change with its finite tensor algebra equipped with that topology. Applying this on affinoid charts of an arbitrary base change and using the analytic-domain gluing theorem proves stability of finite morphisms. ◻

References. [TemIntro] (3.4.2.3 and 4.2.2).

A closed immersion is locally represented by an admissible quotient of affinoid algebras. A morphism is finite if it is locally represented on the base by finite admissible affinoid-algebra maps. Admissibility matters when the valuation is trivial. A morphism is separated if its diagonal is a closed immersion. A morphism is proper if it is compact and has no relative analytic boundary. Here a compact space is quasi-compact and Hausdorff, and compactness of a morphism means that inverse images of compact analytic domains are compact in this sense, as in Definition 2.53.1. Theorem 2.53.4 proves that a compact boundaryless morphism is separated. For good Hausdorff spaces this agrees with the usual compact-inverse-image formulation.

Example 2.7.3. The closed disc is compact but not proper over KK: its algebra is not finite over KK, so Theorem 2.7.2 gives nonempty analytic boundary. The analytification of AK1\mathbb A^1_K has empty analytic boundary but is not compact. The analytification of PK1\mathbb P^1_K is both compact and boundaryless, hence proper. The point at infinity removed from P1\mathbb P^1 is an algebraic boundary divisor; it is not an analytic boundary of A1,an\mathbb A^{1,\mathrm{an}}.

These distinctions enter integration concretely. Compact support controls convergence of an integral; support away from analytic boundary removes the boundary contribution in Stokes. A compact affinoid domain meets the first condition automatically for its whole underlying space, but not the second.

References. [TemIntro] (4.2.3–4.2.4 and 5.1.2).

2.8 Dimension and the information carried by a point

Definition 2.8.1. For a strictly affinoid space M(A)\mathcal M(A), its analytic dimension is the Krull dimension of AA. For a general affinoid space, choose a complete extension L/KL/K for which ALA_L is strictly affinoid and define dim⁡M(A)=dim⁡AL\dim\mathcal M(A)=\dim A_L. This is independent of the strictifying extension. Local dimension is the infimum of the dimensions of analytic neighborhoods; a space is pure of dimension nn if all its irreducible components have dimension nn.

Theorem 3.21.8 proves independence and scalar-extension invariance. The distinction from ordinary Krull dimension is visible in KrK_r from Proposition 1.8.1: its ring is a field, whereas scalar extension to KrK_r turns the radius-rr circle into a strict one-dimensional affinoid annulus. Thus its analytic dimension over KK is one.

The residual transcendence degree and rational value-group rank of H(x)\mathcal H(x) give a pointwise dimension invariant. Their sum is bounded by local analytic dimension in Theorem 3.22.9. A skeleton can have lower real dimension, as the monomial calculation in Chapter 3 illustrates.

2.9 Flatness and quasi-smoothness

Definition 2.9.1. For f:Y→Xf:Y\to X between good spaces, a coherent sheaf F\mathcal F is naively XX-flat at yy if Fy\mathcal F_y is flat over OX,f(y)\mathcal O_{X,f(y)}. It is analytically XX-flat at yy if naive flatness holds at every point over yy after every good analytic base change, including base changes over arbitrary complete extensions of the ground field. The morphism is flat when this applies to OY\mathcal O_Y. The general definition is G-local.

References. [Duc18] (4.1.2–4.1.3 and 4.1.8).

Example 2.9.2. (A boundary obstruction to naive flatness). Take a nontrivially valued KK, choose π\pi with 0<∣π∣<10<|\pi|<1, and put f(T)=∑n≥0π⌊n⌋Tnf(T)=\sum_{n\geq 0}\pi^{\lfloor\sqrt n\rfloor}T^n. This series converges on the closed unit disc but has radius of convergence exactly one. Let YY be that disc and map it to A2,an\mathbb A^{2,\mathrm{an}} by ϕ(t)=(t,f(t))\phi(t)=(t,f(t)). At the image xx of its Gauss point η\eta, Theorem 2.60.4 shows that OA2,an,x\mathcal O_{\mathbb A^{2,\mathrm{an}},x} is a field. Consequently ϕ\phi is naively flat at η\eta.

Now base change by the analytic domain X={∣T1∣≤1}⊂A2,anX=\{|T_1|\leq 1\}\subset\mathbb A^{2,\mathrm{an}}. In XX, the series f(T1)f(T_1) is an analytic function and the graph is the closed subspace T2−f(T1)=0T_2-f(T_1)=0. The local ring OX,x\mathcal O_{X,x} is a domain and g=T2−f(T1)g=T_2-f(T_1) is a nonzero element of its maximal ideal. The quotient by gg is not flat: tensoring the injection given by multiplication by gg with that quotient produces a zero map on a nonzero module. Thus naive flatness is lost under this domain base change.

The convergence radius follows from ∣π∣⌊n⌋→0|\pi|^{\lfloor\sqrt n\rfloor}\to 0 and lim⁡n∣π∣⌊n⌋/n=1\lim_n|\pi|^{\lfloor\sqrt n\rfloor/n}=1. Theorem 2.60.5 proves all the local-ring and base-change assertions used here. This is Ducros’s example from [Duc18] (4.4.2–4.4.6).

Relative analytic differentials form a coherent sheaf ΩY/X\Omega_{Y/X} representing analytic derivations. Write rk⁡yΩY/X\operatorname{rk}_y\Omega_{Y/X} for its fibre dimension over H(y)\mathcal H(y). A Jacobian presentation at yy is a factorization W↪V⊂AXn,anW\hookrightarrow V\subset\mathbb A_X^{n,\mathrm{an}}, where WW is an affinoid domain of YY containing yy, VV is an affinoid domain, and the closed immersion has an ideal generated by r=n−rk⁡yΩY/Xr=n-\operatorname{rk}_y\Omega_{Y/X} elements. A morphism admitting such a presentation is called quasi-smooth at yy. These are [Duc18] (Definitions 5.2.2 and 5.2.4); they allow boundary points.

An LL-analytic space ZZ is geometrically regular at zz if, for every complete valued extension E/LE/L, every point z′∈ZEz'\in Z_E above zz, and every affinoid domain V⊂ZEV\subset Z_E containing z′z', the ordinary local ring OV,z′\mathcal O_{V,z'} is regular. Thus the condition is formulated on good charts even when ZZ itself is not good. For good ZZ this is equivalent to requiring that OZE,z′\mathcal O_{Z_E,z'} be regular for all such E,z′E,z'. One implication follows by choosing an ordinary affinoid neighborhood of z′z'. For the converse, apply the Jacobian-recovery argument in Theorem 2.35.14 to Z→M(L)Z\to\mathcal M(L): at its rational lift the base local ring is a field, so the flatness needed in that argument is automatic. The resulting Jacobian presentation is preserved under scalar extension and analytic-domain restriction by Theorem 2.41.8; the forward regularity argument of Theorem 2.35.14 then proves regularity on every required chart.

NAG-000209 · Open at full scope

Theorem 2.9.3. NAG-000209 A morphism f:Y→Xf:Y\to X is quasi-smooth at yy if and only if it is analytically flat there and its H(f(y))\mathcal H(f(y))-analytic fibre is geometrically regular at yy. For morphisms between good spaces, it is smooth at yy if and only if it is quasi-smooth and boundaryless there. Analytic-domain inclusions are flat. For a morphism between good spaces, naive flatness of a coherent sheaf at an interior point implies analytic flatness there.

Proof. The quasi-smoothness criterion, with all complete-extension tests on the analytic fibre and all good analytic base changes for flatness, is Theorem 2.35.14. Its last paragraph passes from good charts to general analytic spaces. The smoothness comparison for good spaces is Theorem 2.50.3, proved with the ordinary-open étale local definition of smoothness. Analytic-domain flatness is Theorem 2.47.1. Finally, for the ordinary-stalk notion of naive flatness in Definition 2.9.1, Theorem 2.62.3 proves the asserted implication at every interior point, over every complete ground field. ◻

References. The classical statements are [Duc18] (Theorem 5.3.4, Corollary 5.4.8, 4.1.11 and Theorem 8.3.4). The internal proofs organize them through Jacobian charts, ordinary local-ring comparison, smooth envelopes and coherent inner-flatness.

For example, a closed disc is quasi-smooth over KK, but its boundary point is not a smooth point of its structure morphism. Its open disc is smooth. Likewise an affinoid neighborhood in a smooth space may acquire boundary when considered as a space over KK. This does not prevent its use for local equations or integration on supports contained in its interior.

2.10 Analytification and algebraic comparison

For an affine finite-type KK-scheme Z=Spec⁡BZ=\operatorname{Spec}B, define ZanZ^{\mathrm{an}} using multiplicative seminorms on BB extending ∣⋅∣K|\cdot|_K. There is no single global boundedness condition on this finitely generated algebra. Choose generators bib_i; the loci ∣bi∣≤ri|b_i|\leq r_i are affinoid pieces, obtained from a presentation of BB by replacing polynomials with convergent power series of those radii. As the radii grow, these pieces exhaust ZanZ^{\mathrm{an}}. The construction is independent of the generators and glues on Zariski opens.

Proposition 2.10.1. NAG-000136 Points of ZanZ^{\mathrm{an}} are equivalently pairs consisting of a scheme point ξ∈Z\xi\in Z and an absolute value on κ(ξ)\kappa(\xi) extending that of KK. A scheme morphism induces the corresponding analytic morphism by restriction of seminorms.

Proof. The kernel of a multiplicative seminorm on BB is a prime p\mathfrak p, and its norm on B/pB/\mathfrak p extends uniquely to κ(p)\kappa(\mathfrak p). Conversely such a valued field gives a seminorm on BB. Every chosen finite set of generators has finite sizes, so the point lies in one of the affinoid pieces above. The two constructions are inverse, and composition of algebra homomorphisms gives functoriality. ◻

Theorem 2.10.2. NAG-000179 (Algebraic comparison used in this book). Analytification of finite-type KK-schemes commutes with fibre products and complete scalar extension. It preserves closed immersions, finite morphisms, smooth morphisms and proper morphisms. Analytifications have no analytic boundary over KK. For a proper KK-scheme ZZ, analytification gives an equivalence between algebraic and analytic coherent sheaves, compatible with coherent cohomology. In particular algebraic line bundles and their homomorphisms are recovered from the analytic ones.

Proof. Theorem 2.33.25 proves compatibility with fibre products and every complete scalar extension. Theorem 2.33.26 proves preservation of closed immersions and finite morphisms. Corollary 2.35.3 proves preservation of smooth morphisms, using the ordinary-open étale local definition of analytic smoothness. Theorem 2.33.31 proves preservation of proper morphisms. Applying Proposition 2.33.27 to the structure morphism gives the assertion about the boundary. Finally Theorem 2.34.8 is the asserted equivalence for a proper scheme, including its restriction to line bundles and full faithfulness on their homomorphisms; its cohomology comparison is Theorem 2.34.6. ◻

References. The classical comparison statements are described in [TemIntro] (5.1.1–5.1.2) and [Ber93] (§2.6). Their internal proofs here use the explicit affinoid construction, projective cohomology, and the Chow support argument.

GAGA concerns coherent algebraic geometry. It does not identify an arbitrary real-valued smooth function with an algebraic function, nor does it provide positivity or a product of currents. Its role here is narrower and concrete: it lets algebraic line bundles, divisors and finite maps be used as analytic geometric inputs. The additional metric and differential constructions occupy the later parts of the book.

2.11 Tate acyclicity and its proof dependencies

The proof of Theorem 2.3.1 proceeds from bounded Laurent calculations to rational covers, then to affinoid-domain covers. The structure sheaf and Kiehl coherence are constructed afterwards. This section describes the complex, records its classical sources, and gives a guide from the original proof tasks to their internal solutions.

The exact complex and its scope

Let KK be complete for its fixed real-valued non-Archimedean absolute value, possibly trivial. Let AA be a generalized KK-affinoid algebra in the norm-admissible convention, X=M(A)X=\mathcal M(A), and V1,…,VmV_1,\ldots,V_m affinoid domains whose union is XX. The theorem allows nonreduced AA, arbitrary positive radii and boundary points. It assumes neither smoothness nor algebraic closedness. Before forming the complex one needs representing algebras AIA_I for all finite intersections VIV_I and compatible bounded restriction maps. Empty intersections contribute the zero module.

For a finite AA-module MM with its canonical norm, the normalized complex has C−1=M,Cq=0(q≥m),Cq=∏i0<⋯<iqM⊗AAi0⋯iq(0≤q<m).\begin{gathered} C^{-1}=M,\qquad C^q=0\quad(q\geq m),\\ C^q=\prod_{i_0<\cdots<i_q}M\otimes_A A_{i_0\cdots i_q} \quad(0\leq q<m). \end{gathered} The differential is the alternating sum of restrictions, and finite products use the maximum norm. Exactness in every degree includes a bounded lifting estimate for each map onto its image; its constant may depend on the cover and norms. The ordered-tuple version, including repeated indices, is compared with this normalized complex by the bounded homotopies in Lemma 2.21.2. Theorem 2.38.1 gives both versions for arbitrary affinoid-domain covers.

Each tensor factor is finite over its own algebra AIA_I, and usually is not finite over AA. Thus Theorem 1.11.5 alone does not prove that the Čech differentials are admissible. Nor does finiteness of MM justify tensoring an exact complex: localization flatness or a direct finite-module argument is needed. This is a Čech statement; a derived-sheaf-cohomology assertion requires a separate comparison theorem.

Lemma 2.11.1. NAG-001383 (Admissibility of a finite closed cover). Let F1,…,FmF_1,\ldots,F_m be closed subsets covering a topological space XX. For each x∈Xx\in X, the union of those FiF_i containing xx contains an ordinary open neighborhood of xx.

Proof. The finite union of those FjF_j not containing xx is closed and misses xx. Its complement is the required open neighborhood, since the entire family covers XX. ◻

An affinoid domain in an affinoid space is compact and the ambient spectrum is Hausdorff, so the lemma explains the local finite-neighborhood condition for this G-cover. Compactness and the domain’s topological representation are separate analytic inputs. A single covering member need not contain an ordinary neighborhood, and the argument does not apply to arbitrary infinite set-theoretic covers.

Verified source locators

All Temkin introductory-note locators refer to arXiv:1010.2235v2, dated 16 December 2011. Fact 3.3.2.1, printed p. 17, states exactness and admissibility for finite modules. It does not supply a complete proof there; Section 1.2 explicitly describes the use of unproved Facts. Section 3.3 assumes H≠1H\ne 1. The all-radii category uses H=R>0H=\mathbb R_{>0} and includes trivially valued KK; it must not be confused with the strict H=1H=1 site over such a field.

Berkovich’s 1993 paper, Section 1.2, printed p. 15, states both Tate and Kiehl in the general analytic setting, with trivial valuation permitted. For the nontrivial strict case it cites BGR 8.2.1/5 and 9.4.3/3; the passage to the general setting points to [Ber90] (2.2.5 and 2.1.11). Remark 1.2.1(i) on the same page explicitly corrects Ber90 2.2.6(iii): in addition to affinoidness of AVA_V, one needs bijectivity of V→M(AV)V\to\mathcal M(A_V). This is an author-issued correction, unlike the three independently checked literal-scope examples in Section 1.9. The original proofs at these two Ber90 locators were not obtained in this round. Their numbers are verified as citations in Ber93, not as independently read Ber90 proof pages.

The original BGR scan was inspected at the following proof passages. The book treats the strict rigid setting over a nontrivially valued complete field.

  • Theorem 8.2.1/1, p. 327, gives the structure-presheaf acyclicity statement. Its proof is in Sections 8.2.2–8.2.3, pp. 331–335. The finite-cover reduction uses 8.2.2/2–5, including Gerritzen–Grauert 7.3.5/3 and the small-scalar bound 7.3.4/7. The binary Laurent calculation uses the separate injectivity input 7.3.2/5.

  • Corollary 8.2.1/5, p. 330, treats every algebraic AA-module. Its free-resolution argument invokes localization flatness 7.3.2/6 and the finite-cover cohomological bound 8.1.3/2. Its text is an algebraic exactness assertion; the required Banach norm estimates must still be tracked for finite modules.

  • Theorem 9.4.3/3, p. 378, is proved using Lemmas 9.4.3/5–6, pp. 378–382. These supply quantitative Laurent gluing and global finite generation, using complete module norms, coherent kernels, residue comparisons and Nakayama’s lemma. None of these is silently included in the elementary foundation of this project.

References. [Ber93] (§1.2, p. 15); [BGR84] (§§8.2, 9.4.3).

Domains, the structure sheaf and coherence

Theorem 2.11.2. NAG-001399 (Rational-domain input). Let AA be a generalized KK-affinoid algebra and let g,f1,…,fsg,f_1,\ldots,f_s generate the unit ideal. For positive radii rir_i, the norm-admissible quotient A{r−1T}/(gTi−fi)iA\{r^{-1}T\}/(gT_i-f_i)_i represents the domain ∣fi(x)∣≤ri∣g(x)∣|f_i(x)|\leq r_i|g(x)|. Every bounded map A→BA\to B to a generalized affinoid algebra whose spectrum maps into this domain factors uniquely through that quotient as a bounded map.

Proof. Theorem 2.31.1 constructs the displayed actual quotient as a norm-admissible affinoid algebra and identifies its spectrum homeomorphically with the stated subset. Apply Theorem 2.31.2 to the given generalized affinoid target BB. Its geometric criterion is exactly the image condition in the statement, and yields the unique bounded factorization. Both results allow all positive radii and a trivially valued ground field; zero targets have the unique zero factorization. No strictness or rational-cover acyclicity hypothesis is added. ◻

References. [TemIntro] (3.1.2.5 and 3.2.2.1).

Theorem 2.11.3. NAG-001384 (Gerritzen–Grauert input). Every affinoid domain VV of a generalized KK-affinoid space XX is a finite union of rational domains of XX. The field may be trivially valued.

Proof. Theorem 2.36.6 supplies a finite covering by rational domains defined over the original algebra AA. Its proof allows arbitrary positive radii and both trivial and nontrivial absolute values. The domain convention in the statement agrees with that theorem by Theorem 2.36.7. ◻

References. [Tem05] (Theorems 1.1 and 3.1, Corollary 3.2).

Theorem 2.11.4. NAG-001372 (Affinoid structure-sheaf input). On the site of compact analytic domains of X=M(A)X=\mathcal M(A) with finite domain coverings, the affinoid assignment V↦AVV\mapsto A_V extends to a sheaf of Banach algebras with bounded restrictions. For a compact domain covered by affinoid domains, its sections are the compatible tuples of affinoid sections, independently of the chosen finite cover up to canonical bounded isomorphism and uniformly equivalent norms.

Proof. Restrict the ring sheaf of Theorem 2.39.4 to compact analytic domains. Theorem 2.39.5 realizes its sections as the closed matching kernel for every finite affinoid cover, proves boundedness of restrictions, and identifies the kernels obtained from two covers by mutually bounded maps. Those maps are induced by the unique restriction maps on common affinoid refinements, so the comparison is canonical. Their two boundedness estimates give uniform equivalence of the Banach norms. ◻

References. [TemIntro] (Exercise 3.3.2.2(i)); [Ber93] (§1.2, pp. 15–16).

Theorem 2.11.5. NAG-001385 (Kiehl coherence input). With the preceding affinoid structure sheaf, finite AA-modules are equivalent to coherent OXG\mathcal O_{X_G}-modules. The functor sends MM to the sheaf with sections M⊗AAVM\otimes_A A_V on affinoid domains; a quasi-inverse is global sections. Here a coherent module is locally finitely generated, and for every analytic domain UU, the kernel of every map OUGn→F∣UG\mathcal O_{U_G}^n\to\mathcal F|_{U_G} is locally finitely generated.

Proof. Theorem 2.38.5 proves the equivalence on the finite affinoid-domain site, including effective gluing of arbitrary finite local modules. Theorem 2.39.6 transports it to the full analytic GG-site through the ringed-topos comparison. Its coherence condition is local finite generation together with local finite generation of the stated kernels, so it is exactly the condition here. Evaluating on an affinoid domain gives M⊗AAVM\otimes_A A_V; evaluating on XX recovers MM. Restriction to the compact-domain basis preserves both the sheaf and these local properties. ◻

References. [BGR84] (9.4.3/3 and its Lemmas 5–6); [Ber93] (§1.2, p. 15); [TemIntro] (3.3.4.1).

The original proof tasks and their solutions

The following task statements retain their original labels. The proof guide following the list identifies their internal solutions, including the complete replacement routes used for the Laurent and Gerritzen–Grauert tasks.

The general finite-module and spectral tasks are proved in Theorems 1.18.3, 1.20.1 and 1.20.2. Intersections and localization flatness are established in Corollary 2.36.10 and Proposition 2.32.3, together with Proposition 2.38.2.

For Laurent exactness, we use the strict coefficient estimate and then descend bounded primitives through one simultaneous free-radius extension. The finite comparison and refinement are Lemma 2.21.2 and Proposition 2.22.8; the resulting all-radius theorem is Theorem 2.32.2. Its proof uses the coefficient projection only as a bounded linear map.

Theorem 2.36.6 combines strict Gerritzen–Grauert with the Gauss-section finite truncation of Lemma 2.36.4. The tensor and graded-reduction results are proved on independent branches in Theorems 2.33.3 and 3.18.7. Finally, Theorems 2.28.6, 2.37.3 and 2.39.6 give quantitative gluing, finite-module descent and coherence on the full analytic site.

2.12 A quantitative criterion for an elementary Laurent pair

For an elementary Laurent cover, the coefficient splitting can be separated from the geometric localization theorem. The next result makes this separation precise. Its input is a numerical lower bound for one operator on two-sided Laurent series. It supplies strict exactness once that bound is available; it does not assert the bound for every affinoid module.

Let AA be a Banach KK-algebra and let MM be a Banach AA-module with ∥am∥≤∥a∥∥m∥\|am\|\leq\|a\|\|m\|, as in Section 1.10. Fix f∈Af\in A and r>0r>0. Define E+={∑n≥0mnTn:∥mn∥rn⟶0},E−={∑n≤0mnTn:∥mn∥rn⟶0 as n→−∞},E={∑n∈ZmnTn:∥mn∥rn⟶0 as ∣n∣→∞}.\begin{split} E_+&=\left\{\sum_{n\geq0}m_nT^n:\|m_n\|r^n\longrightarrow0\right\},\\ E_-&=\left\{\sum_{n\leq0}m_nT^n:\|m_n\|r^n\longrightarrow0\text{ as }n\to-\infty\right\},\\ E&=\left\{\sum_{n\in\mathbb Z}m_nT^n:\|m_n\|r^n\longrightarrow0\text{ as }|n|\to\infty\right\}. \end{split} Each has norm max⁡n∥mn∥rn\max_n\|m_n\|r^n. The coefficientwise argument of Lemma 1.10.1, with the indicated index sets, proves their completeness. We regard E+,E−E_+,E_- as closed submodules of EE. Multiplication by TT is the shift of coefficients and rescales the norm by rr. The following operators preserve their indicated modules: Q+=T−f:E+⟶E+,Q−=1−fT−1:E−⟶E−,L=1−fT−1:E⟶E.\begin{gathered} Q_+=T-f:E_+\longrightarrow E_+,\qquad Q_-=1-fT^{-1}:E_-\longrightarrow E_-, \\ L=1-fT^{-1}:E\longrightarrow E. \end{gathered}

Theorem 2.12.1. NAG-001373 (Quantitative Laurent criterion). Suppose that there is a real number c>0c>0 such that ∥Lw∥≥c∥w∥(w∈E).\|Lw\|\geq c\|w\|\qquad(w\in E). Endow P+=E+/Q+E+P_+=E_+/Q_+E_+, P−=E−/Q−E−P_-=E_-/Q_-E_- and P0=E/LEP_0=E/LE with their quotient norms. They are Banach AA-modules, and the sequence 0⟶M→ ι P+⊕P−→ δ P0⟶0,ι(m)=([m],[m]),δ([u],[v])=[u−v].\begin{gathered} 0\longrightarrow M\xrightarrow{\ \iota\ }P_+\oplus P_- \xrightarrow{\ \delta\ }P_0\longrightarrow0,\\ \iota(m)=([m],[m]),\quad \delta([u],[v])=[u-v]. \end{gathered} is strict exact. The direct sum has its maximum norm. More precisely, with C=max⁡(1,max⁡(r,∥f∥)cr),C=\max\left(1,\frac{\max(r,\|f\|)}{cr}\right), one has C−1∥m∥≤∥ι(m)∥≤∥m∥C^{-1}\|m\|\leq\|\iota(m)\|\leq\|m\|. The induced map (P+⊕P−)/ι(M)→P0(P_+\oplus P_-)/\iota(M)\to P_0 is isometric for the quotient norms.

Proof. Step 1: the coefficient identity. Define d0(u,v)=u−v,d1(a,b)=Ta−b,E+⊕E−⟶E.d_0(u,v)=u-v,\qquad d_1(a,b)=Ta-b, \qquad E_+\oplus E_-\longrightarrow E. The kernel of d0d_0 consists of diagonal constants, since E+∩E−=ME_+\cap E_-=M. Splitting into nonnegative and strictly negative powers gives a right inverse for d0d_0 of norm at most one. The supports of TaTa and bb in d1(a,b)d_1(a,b) are disjoint: the former has exponents at least one, the latter exponents at most zero. Thus d1d_1 is bijective and ∥d1(a,b)∥=max⁡(r∥a∥,∥b∥),d0(Q+a,Q−b)=Ld1(a,b).\|d_1(a,b)\|=\max(r\|a\|,\|b\|), \qquad d_0(Q_+a,Q_-b)=Ld_1(a,b). Taking b=0b=0 or a=0a=0 gives ∥Q+a∥≥cr∥a∥\|Q_+a\|\geq cr\|a\| and ∥Q−b∥≥c∥b∥\|Q_-b\|\geq c\|b\|. Each of Q+,Q−,LQ_+,Q_-,L therefore has closed image: convergence of the images makes the original sequence Cauchy by its lower bound, and continuity identifies its limit. Lemma 1.11.1 now proves completeness and definiteness of the three quotient norms. The identity also makes δ\delta well defined.

Step 2: exactness. Surjectivity of δ\delta follows from the coefficient splitting for d0d_0. Suppose δ([u],[v])=0\delta([u],[v])=0, so that u−v=Lwu-v=Lw for some w∈Ew\in E. Write w=Ta−bw=Ta-b uniquely. Then u−Q+a=v−Q−b.u-Q_+a=v-Q_-b. The left side belongs to E+E_+ and the right side to E−E_-; hence they equal a constant m∈Mm\in M. This proves exactness in the middle. If ι(m)=0\iota(m)=0, write m=Q+a=Q−bm=Q_+a=Q_-b. Then L(Ta−b)=0L(Ta-b)=0. The lower bound on LL gives Ta=bTa=b, and their disjoint supports force a=b=0a=b=0, so m=0m=0.

Step 3: the augmentation estimate. Representatives of the two classes of a constant mm have the form u=m+Q+au=m+Q_+a and v=m+Q−bv=m+Q_-b. Step 1 gives max⁡(r∥a∥,∥b∥)≤c−1∥u−v∥≤c−1max⁡(∥u∥,∥v∥).\max(r\|a\|,\|b\|) \leq c^{-1}\|u-v\| \leq c^{-1}\max(\|u\|,\|v\|). Since ∥Q+a∥≤max⁡(r,∥f∥)∥a∥\|Q_+a\|\leq\max(r,\|f\|)\|a\|, the identity m=u−Q+am=u-Q_+a gives ∥m∥≤Cmax⁡(∥u∥,∥v∥)\|m\|\leq C\max(\|u\|,\|v\|). Taking infima over both representatives proves the lower bound for ι\iota; using the constant representatives proves its upper bound. Thus ι\iota is strict. Its image is also the kernel of the continuous map δ\delta and is closed.

Step 4: the final quotient norm. The difference map is contractive. Conversely, each representative z∈Ez\in E of a class in P0P_0 has a splitting z=u−vz=u-v with max⁡(∥u∥,∥v∥)≤∥z∥\max(\|u\|,\|v\|)\leq\|z\|. The class ([u],[v])([u],[v]) is a lift with no larger norm. Taking infima over representatives zz proves that the quotient norm induced from P+⊕P−P_+\oplus P_- is at most the norm of the given class in P0P_0. Contractivity gives the reverse inequality. The induced map is therefore isometric, which proves strictness of δ\delta. ◻

For finite modules over a strict affinoid algebra and value-group radii, Theorem 2.16.4 supplies the required lower bound. Theorem 2.18.3 identifies these coefficient quotients with their analytic section modules, giving the elementary Laurent step in the finite-cover argument. Simultaneous strictification then yields all positive radii in Theorem 2.32.2; affinoid-domain coverings are treated in Theorem 2.38.1.

2.13 Laurent coefficient normal forms

The Laurent coefficient module in Theorem 2.12.1 was defined directly. We now identify it with a quotient of a two-variable coefficient module, including the quotient norm and the actual image of the relation map. The same construction treats two boundary radii 0<a≤b0<a\leq b; setting a=b=ra=b=r gives the equal-radius module used in that theorem.

Let KK be complete for its specified real-valued non-Archimedean absolute value, which may be trivial. Let AA be a commutative unital Banach KK-algebra, and let MM be a unital Banach AA-module: its norm is definite, non-Archimedean, complete and homogeneous over KK, and ∥am∥≤∥a∥∥m∥\|am\|\leq\|a\|\|m\|. No finiteness hypothesis is imposed on AA or MM. All coefficient spaces below use these specified norms.

Lemma 2.13.1. NAG-001414 (Weighted coefficient completeness). Let II be a countable set and let (wi)i∈I(w_i)_{i\in I} be positive real numbers. Put c0(I,M;w)={(mi)i∈I:for every ε>0, #{i:∥mi∥wi≥ε}<∞},∥(mi)∥w=sup⁡i∈I∥mi∥wi.\begin{split} c_0(I,M;w) &=\{(m_i)_{i\in I}: \text{for every }\varepsilon>0,\ \#\{i:\|m_i\|w_i\geq\varepsilon\}<\infty\},\\ \|(m_i)\|_w&=\sup_{i\in I}\|m_i\|w_i . \end{split} We put sup⁡∅=0\sup\varnothing=0 if II is empty. This is a Banach AA-module, and its finitely supported families are dense. Each nonzero family attains its norm at a coordinate. The coordinate map at ii has bound wi−1w_i^{-1}.

Proof. The finite-tail condition makes the supremum finite. The ultrametric inequality, homogeneity, definiteness and the contractive AA-action follow coordinatewise. To see that a nonzero family attains its norm, choose a coordinate of positive weighted norm tt. Only finitely many coordinates have weighted norm at least tt, and the maximum among them is the supremum of all coordinates.

Let x(j)=(mi(j))x^{(j)}=(m_i^{(j)}) be Cauchy. For each ii, the inequality ∥mi(j)−mi(k)∥≤wi−1∥x(j)−x(k)∥w\|m_i^{(j)}-m_i^{(k)}\| \leq w_i^{-1}\|x^{(j)}-x^{(k)}\|_w gives a limit mi∈Mm_i\in M. Given ε>0\varepsilon>0, choose jj so that the norm on the right before multiplication by wi−1w_i^{-1} is less than ε/2\varepsilon/2 for all sufficiently large kk. Passing to the limit in each coordinate gives ∥mi−mi(j)∥wi≤ε/2(i∈I).\|m_i-m_i^{(j)}\|w_i\leq\varepsilon/2 \qquad(i\in I). Outside a finite set, ∥mi(j)∥wi<ε/2\|m_i^{(j)}\|w_i<\varepsilon/2, so ∥mi∥wi≤ε/2\|m_i\|w_i\leq\varepsilon/2 there. Thus x=(mi)x=(m_i) belongs to the displayed space, and the same estimate proves convergence to xx. Finally, retaining all coordinates with weighted norm at least ε/2\varepsilon/2 gives a finite truncation with error at most ε/2\varepsilon/2. This proves density and completeness. The coordinate bound is part of the definition of the supremum norm. ◻

Fix 0<a≤b0<a\leq b and write λ=b/a\lambda=b/a. We use T,UT,U for two independent variables, and ZZ for the Laurent variable. Define Da,b(M)=M{b−1T,aU}={∑i,j≥0mijTiUj:∥mij∥bia−j⟶0},∥g∥D=sup⁡i,j≥0∥mij∥bia−j,wn={bnn≥0,ann<0,La,b(M)={∑n∈ZcnZn:∥cn∥wn⟶0 as ∣n∣→∞},∥c∥L=sup⁡n∈Z∥cn∥wn.\begin{split} D_{a,b}(M) &=M\{b^{-1}T,aU\}\\ &=\left\{\sum_{i,j\geq0}m_{ij}T^iU^j: \|m_{ij}\|b^ia^{-j}\longrightarrow0\right\},\\ \|g\|_D&=\sup_{i,j\geq0}\|m_{ij}\|b^ia^{-j},\\ w_n&=\begin{cases}b^n&n\geq0,\\a^n&n<0,\end{cases}\\ L_{a,b}(M) &=\left\{\sum_{n\in\mathbb Z}c_nZ^n: \|c_n\|w_n\longrightarrow0\text{ as }|n|\to\infty\right\},\\ \|c\|_L&=\sup_{n\in\mathbb Z}\|c_n\|w_n . \end{split} In the definition of Da,b(M)D_{a,b}(M), convergence to zero means convergence outside finite subsets of N2\mathbb N^2. This is a definition by coefficients, not a tensor-product assertion. The preceding lemma proves completeness and polynomial density for both spaces. Multiplication by TUTU shifts both indices up by one and multiplies the norm by λ\lambda. Consequently P=TU−1:Da,b(M)⟶Da,b(M),∥Pg∥D≤λ∥g∥DP=TU-1:D_{a,b}(M)\longrightarrow D_{a,b}(M), \qquad \|Pg\|_D\leq\lambda\|g\|_D is a bounded AA-linear operator.

Theorem 2.13.2. NAG-001415 (The Laurent coefficient normal form). For the data just specified, there are AA-linear maps N:Da,b(M)⟶La,b(M),S:La,b(M)⟶Da,b(M),H:Da,b(M)⟶Da,b(M).\begin{aligned} \mathcal N&:D_{a,b}(M)\longrightarrow L_{a,b}(M),\\ \mathcal S&:L_{a,b}(M)\longrightarrow D_{a,b}(M),\\ \mathcal H&:D_{a,b}(M)\longrightarrow D_{a,b}(M). \end{aligned} given by the convergent coefficient formulas (Ng)n=∑k≥0mn++k,n−+k,n+=max⁡(n,0),n−=max⁡(−n,0),Sc=∑n≥0cnTn+∑n<0cnU−n,(Hg)ij=∑k≥1mi+k,j+k.\begin{split} (\mathcal Ng)_n &=\sum_{k\geq0}m_{n_++k,n_-+k}, \qquad n_+=\max(n,0),\quad n_-=\max(-n,0),\\ \mathcal Sc &=\sum_{n\geq0}c_nT^n+\sum_{n<0}c_nU^{-n},\\ (\mathcal Hg)_{ij} &=\sum_{k\geq1}m_{i+k,j+k}. \end{split} They satisfy ∥Ng∥L≤∥g∥D,∥Sc∥D=∥c∥L,∥Hg∥D≤λ−1∥g∥D,PH=1−SN,HP=1,NP=0,NS=1,HS=0.\begin{gathered} \|\mathcal Ng\|_L\leq\|g\|_D,\qquad \|\mathcal Sc\|_D=\|c\|_L,\qquad \|\mathcal Hg\|_D\leq\lambda^{-1}\|g\|_D,\\ P\mathcal H=1-\mathcal S\mathcal N,\qquad \mathcal HP=1,\qquad \mathcal NP=0,\qquad \mathcal N\mathcal S=1,\qquad \mathcal H\mathcal S=0 . \end{gathered} In particular, every gg has a unique decomposition g=Pq+Sc,q=Hg,c=Ng,∥g∥D=max⁡{λ∥q∥D,∥c∥L}.\begin{gathered} g=Pq+\mathcal Sc,\qquad q=\mathcal Hg,\quad c=\mathcal Ng,\\ \|g\|_D=\max\{\lambda\|q\|_D,\|c\|_L\}. \end{gathered} The relation map satisfies ∥Pq∥D=λ∥q∥D\|Pq\|_D=\lambda\|q\|_D and has closed image. Its image is the actual kernel of N\mathcal N, so ker⁡N=(TU−1)Da,b(M),Da,b(M)/(TU−1)Da,b(M)→ N‾ La,b(M)\begin{gathered} \ker\mathcal N=(TU-1)D_{a,b}(M),\\ D_{a,b}(M)/(TU-1)D_{a,b}(M) \xrightarrow{\ \overline{\mathcal N}\ }L_{a,b}(M) \end{gathered} is an isometric isomorphism of Banach AA-modules. The quotient norm of [g][g] is ∥Ng∥L\|\mathcal Ng\|_L, attained at its unique representative supported on the two coordinate axes, with the origin counted once. The associated short exact sequence is split as a sequence of Banach AA-modules.

Proof. Step 1: polynomial reduction and its bounds. On a monomial mTiUjmT^iU^j, put d=min⁡(i,j)d=\min(i,j) and define N(mTiUj)=mZi−j,H(mTiUj)=m∑ℓ=0d−1Ti−1−ℓUj−1−ℓ,\mathcal N(mT^iU^j)=mZ^{i-j},\qquad \mathcal H(mT^iU^j) =m\sum_{\ell=0}^{d-1}T^{i-1-\ell}U^{j-1-\ell}, where the sum is zero if d=0d=0. Extend these formulas AA-linearly to polynomials, and define S\mathcal S by the displayed axis formula. The weights satisfy bia−j=λdwi−j.b^ia^{-j}=\lambda^d w_{i-j}. Thus N\mathcal N does not increase the norm of a monomial. Every monomial in its H\mathcal H-image has weight at most λ−1bia−j\lambda^{-1}b^ia^{-j}, because its weight relative to the original one is λ−1−ℓ\lambda^{-1-\ell}. Applying the ultrametric inequality to finite sums proves ∥Ng∥L≤∥g∥D,∥Hg∥D≤λ−1∥g∥D(g∈M[T,U]).\|\mathcal Ng\|_L\leq\|g\|_D,\qquad \|\mathcal Hg\|_D\leq\lambda^{-1}\|g\|_D \qquad(g\in M[T,U]). The axis formula preserves each nonzero coefficient and its weight, so S\mathcal S is isometric, on Laurent polynomials and on the full coefficient space.

Step 2: completion and the explicit sums. Choose finite polynomial truncations gνg_\nu converging to gg. The preceding bounds make Ngν\mathcal Ng_\nu and Hgν\mathcal Hg_\nu Cauchy in their respective complete coefficient spaces. Their limits define Ng\mathcal Ng and Hg\mathcal Hg. Applying the same bounds to differences of two approximations proves independence of the truncations, linearity and the asserted bounds. This constructs the extensions directly.

For a fixed nn, the terms in the claimed formula for (Ng)n(\mathcal Ng)_n satisfy ∥mn++k,n−+k∥wn=λ−k∥mn++k,n−+k∥bn++ka−(n−+k)⟶0.\|m_{n_++k,n_-+k}\|w_n =\lambda^{-k} \|m_{n_++k,n_-+k}\|b^{n_++k}a^{-(n_-+k)} \longrightarrow0 . For fixed i,ji,j, the terms in the formula for (Hg)ij(\mathcal Hg)_{ij} satisfy ∥mi+k,j+k∥bia−j=λ−k∥mi+k,j+k∥bi+ka−(j+k)⟶0.\|m_{i+k,j+k}\|b^ia^{-j} =\lambda^{-k}\|m_{i+k,j+k}\|b^{i+k}a^{-(j+k)} \longrightarrow0 . Both series therefore converge in MM by Lemma 1.1.2. The bounded coordinate maps of Lemma 2.13.1 identify their sums with the limits constructed from polynomial truncations. In particular, the formulas define families with the required uniform finite-tail condition; this also follows by approximating their images by the corresponding finite polynomial images.

Step 3: the identities. For a polynomial monomial, telescoping gives (TU−1)m∑ℓ=0d−1Ti−1−ℓUj−1−ℓ=mTiUj−mTi−dUj−d,H(mTi+1Uj+1)−H(mTiUj)=mTiUj.\begin{split} (TU-1)m\sum_{\ell=0}^{d-1}T^{i-1-\ell}U^{j-1-\ell} &=mT^iU^j-mT^{i-d}U^{j-d},\\ \mathcal H(mT^{i+1}U^{j+1})-\mathcal H(mT^iU^j) &=mT^iU^j . \end{split} Summing over monomials proves PH=1−SNP\mathcal H=1-\mathcal S\mathcal N and HP=1\mathcal HP=1 on polynomials. Substitution of Z,Z−1Z,Z^{-1} gives NP=0\mathcal NP=0. The axis formula gives NS=1\mathcal N\mathcal S=1 and HS=0\mathcal H\mathcal S=0. Every map in these identities is bounded. Taking limits of polynomial truncations proves all five identities on the complete spaces.

Step 4: the relation image and the quotient norm. The identities give the displayed decomposition. Applying H\mathcal H and N\mathcal N to any such decomposition proves uniqueness. Their bounds, together with the bound for PP, imply ∥Pq+Sc∥D≤max⁡{λ∥q∥D,∥c∥L},λ∥q∥D=λ∥H(Pq+Sc)∥D≤∥Pq+Sc∥D,∥c∥L=∥N(Pq+Sc)∥L≤∥Pq+Sc∥D.\begin{split} \|Pq+\mathcal Sc\|_D &\leq\max\{\lambda\|q\|_D,\|c\|_L\},\\ \lambda\|q\|_D &=\lambda\|\mathcal H(Pq+\mathcal Sc)\|_D \leq\|Pq+\mathcal Sc\|_D,\\ \|c\|_L &=\|\mathcal N(Pq+\mathcal Sc)\|_L \leq\|Pq+\mathcal Sc\|_D. \end{split} These are the asserted norm equality. Taking c=0c=0 gives ∥Pq∥D=λ∥q∥D\|Pq\|_D=\lambda\|q\|_D, and hence injectivity of PP. If PqνPq_\nu converges, this equality makes qνq_\nu Cauchy. Completeness gives a limit qq, and boundedness of PP identifies the original limit as PqPq. Thus PDa,b(M)PD_{a,b}(M) is closed.

We have NP=0\mathcal NP=0, while Ng=0\mathcal Ng=0 implies g=PHgg=P\mathcal Hg. This proves the kernel equality with the algebraic image, without first taking its closure. For every h∈Da,b(M)h\in D_{a,b}(M), ∥Ng∥L=∥N(g+Ph)∥L≤∥g+Ph∥D.\|\mathcal Ng\|_L=\|\mathcal N(g+Ph)\|_L\leq\|g+Ph\|_D. The choice h=−Hgh=-\mathcal Hg gives g+Ph=SNgg+Ph=\mathcal S\mathcal Ng and attains equality. Hence the quotient norm is exactly ∥Ng∥L\|\mathcal Ng\|_L. The target is complete, so the quotient is a Banach module. The right inverse S\mathcal S, the left inverse H\mathcal H and the decomposition prove the splitting and norm-strictness. A representative supported on the two axes equals Sc\mathcal Sc for a unique cc; applying N\mathcal N identifies cc with the class, proving the final uniqueness assertion. ◻

In particular, if a=b=ra=b=r, then wn=rnw_n=r^n, λ=1\lambda=1, and the theorem gives M{r−1T,rU}/(TU−1)M{r−1T,rU}≃Lr,r(M),Lr,r(M)={∑n∈ZcnZn:∥cn∥rn⟶0 as ∣n∣→∞},∥(TU−1)q∥=∥q∥.\begin{gathered} M\{r^{-1}T,rU\}/(TU-1)M\{r^{-1}T,rU\} \simeq L_{r,r}(M),\\ L_{r,r}(M)=\left\{\sum_{n\in\mathbb Z}c_nZ^n: \|c_n\|r^n\longrightarrow0\text{ as }|n|\to\infty\right\},\\ \|(TU-1)q\|=\|q\|. \end{gathered} Here the isomorphism is isometric. Neither the equal-radius assertion nor the two-radius assertion requires a radius in ∣K×∣|K^\times|.

Proposition 2.13.3. NAG-001416 (Multiplication, module actions and naturality).

Coefficient convolution makes Da,b(A)D_{a,b}(A) and La,b(A)L_{a,b}(A) Banach KK-algebras, and makes Da,b(M)D_{a,b}(M) and La,b(M)L_{a,b}(M) contractive modules over them. The map NA\mathcal N_A is an algebra homomorphism, and NM(fg)=NA(f)NM(g)(f∈Da,b(A), g∈Da,b(M)).\mathcal N_M(fg)=\mathcal N_A(f)\mathcal N_M(g) \qquad(f\in D_{a,b}(A),\ g\in D_{a,b}(M)). Consequently, the preceding quotient isomorphism is compatible with the isometric algebra isomorphism Da,b(A)/(TU−1)Da,b(A) ≃ La,b(A).D_{a,b}(A)/(TU-1)D_{a,b}(A)\ \simeq\ L_{a,b}(A). All the maps and the quotient identification are natural for bounded coefficient maps. More precisely, let ϕ:A→A′\phi:A\to A' be a bounded homomorphism of Banach KK-algebras, let M′M' be a contractive Banach A′A'-module, and let v:M→M′v:M\to M' be a KK-linear map satisfying v(am)=ϕ(a)v(m),∥v(m)∥≤C∥m∥v(am)=\phi(a)v(m),\qquad \|v(m)\|\leq C\|m\| for some fixed C>0C>0. Applying vv to coefficients gives maps D(v)D(v) and L(v)L(v) bounded by CC, commuting with P,N,S,HP,\mathcal N,\mathcal S,\mathcal H in their appropriate source and target spaces. The induced map on the relation quotients also has bound CC.

Proof. For DD, the weight of a product monomial is the product of its weights. For LL, write wn=max⁡(an,bn)w_n=\max(a^n,b^n); then wn+m≤wnwm(n,m∈Z).w_{n+m}\leq w_nw_m\qquad(n,m\in\mathbb Z). Finite coefficient convolution and the ultrametric inequality therefore give ∥fg∥≤∥f∥∥g∥\|fg\|\leq\|f\|\|g\| for the polynomial products and actions in both cases. To construct a product of general elements, take polynomial approximations fν,gνf_\nu,g_\nu. They are bounded and Cauchy. The identity fνgν−fμgμ=(fν−fμ)gν+fμ(gν−gμ)f_\nu g_\nu-f_\mu g_\mu =(f_\nu-f_\mu)g_\nu+f_\mu(g_\nu-g_\mu) and the product bound make the products Cauchy. Completeness defines their limit; applying the same estimate to two choices of approximations proves independence. The bound persists under limits, and associativity, the identity and the module law follow from the polynomial identities by continuity.

For DD, each coefficient is a finite convolution sum. For LL, the coefficient of ZnZ^n is the convergent sum ∑k∈Zfkgn−k.\sum_{k\in\mathbb Z} f_k g_{n-k}. Indeed, ∥fkgn−k∥wn≤(∥fk∥wk)(∥gn−k∥wn−k)≤(∥fk∥wk)∥g∥,\|f_k g_{n-k}\|w_n \leq(\|f_k\|w_k)(\|g_{n-k}\|w_{n-k}) \leq(\|f_k\|w_k)\|g\|, so the terms tend to zero as ∣k∣→∞|k|\to\infty. Lemma 1.1.2 gives convergence independent of enumeration. Bounded coordinate evaluation identifies this sum with the coefficient of the product constructed by truncation.

On polynomials, NA\mathcal N_A is substitution T↦ZT\mapsto Z, U↦Z−1U\mapsto Z^{-1}; it preserves multiplication, and the corresponding module identity holds. Its boundedness and the product bounds extend both identities to the complete spaces. Theorem 2.13.2 identifies its kernel and quotient norm, proving the algebra and module assertions. The axis section S\mathcal S is only asserted to be AA-linear; it need not preserve multiplication.

Finally, the coefficient bound for vv preserves the finite-tail conditions and gives ∥D(v)∥,∥L(v)∥≤C\|D(v)\|,\|L(v)\|\leq C. On polynomials, commutation with the four displayed operators follows from their finite coefficient formulas; continuity extends it to all series. The relation image is therefore preserved, so there is an induced map on quotients. For every representative gg its image has a representative of norm at most C∥g∥C\|g\|. Taking the infimum over representatives gives the asserted quotient bound. Semilinearity of the coefficient maps gives the corresponding compatibility with algebra maps and module actions. ◻

The proof uses completeness, coefficient estimates and finite telescoping identities. It applies at trivial valuation and in every residue characteristic. It makes no identification with sections on an analytic domain, assumes no finite-module tensor comparison, and proves no lower bound for 1−fZ−11-fZ^{-1} with a general f∈Af\in A. Those are separate inputs to the geometric application of Theorem 2.12.1.

A coefficient gluing sequence

The normal form gives a complete algebraic justification for the coefficient splitting used in Proposition 2.3.2. We state this consequence for Banach module coefficients and arbitrary positive endpoint radii. The objects in the statement are coefficient spaces; their identification with sections on analytic domains is a separate assertion.

Proposition 2.13.4. NAG-001417 (Gluing coefficient series at an intermediate radius). Let AA and MM satisfy the standing Banach hypotheses, and let 0<a≤b0<a\leq b. Write Ec(M)=M{c−1X}E_c(M)=M\{c^{-1}X\} for c=a,bc=a,b, and write La,b(M)L_{a,b}(M) for the annular Laurent coefficient space, with weight bnb^n for n≥0n\geq 0 and ana^n for n<0n<0. The coefficient-preserving maps give a strict exact sequence 0⟶Eb(M)→jEa(M)⊕La,b(M)→δLa,a(M)⟶0,j(h)=(h,h),δ(u,v)=u−v.\begin{gathered} 0\longrightarrow E_b(M)\xrightarrow{j} E_a(M)\oplus L_{a,b}(M)\xrightarrow{\delta} L_{a,a}(M)\longrightarrow0,\\ j(h)=(h,h),\qquad\delta(u,v)=u-v. \end{gathered} Here the middle term has its maximum norm. The map jj is isometric, and δ\delta has an AA-linear right inverse of norm at most one. Consequently, the norm on La,a(M)L_{a,a}(M) equals the quotient norm induced by δ\delta. For M=AM=A, Proposition 2.13.3 identifies the Laurent coefficient spaces with the corresponding quotient algebras.

Proof. For n≥0n\geq 0, the inequality an≤bna^n\leq b^n shows that both restriction maps to the smaller coefficient space are contractive. The map from Eb(M)E_b(M) to the nonnegative part of La,b(M)L_{a,b}(M) is isometric, so jj is isometric. The difference map δ\delta is contractive.

If w=∑n∈ZmnXn∈La,a(M)w=\sum_{n\in\mathbb Z}m_nX^n\in L_{a,a}(M), set s(w)=(∑n≥0mnXn,−∑n<0mnXn).s(w)=\left(\sum_{n\geq0}m_nX^n,-\sum_{n<0}m_nX^n\right). The first series belongs to Ea(M)E_a(M). The second belongs to La,b(M)L_{a,b}(M) because its negative-degree weights are still ana^n. These two series have disjoint coefficient supports and their maximum norm is ∥w∥\|w\|. Thus ss is an AA-linear right inverse of δ\delta with norm at most one.

Suppose u−v=0u-v=0 in La,a(M)L_{a,a}(M). Equality means equality of every coefficient. Since uu has no negative powers, neither does vv. The nonnegative coefficients of vv tend to zero with weights bnb^n, so they define a unique h∈Eb(M)h\in E_b(M) satisfying (u,v)=j(h)(u,v)=j(h). This proves exactness in the middle; injectivity of jj is already part of its isometry. Finally, contractivity of δ\delta makes the target norm no larger than its quotient norm, while ss gives the reverse inequality. All asserted norm statements follow. ◻

Taking A=M=KA=M=K, b=1b=1 and a=s∈(0,1)a=s\in(0,1) supplies the coefficient normal forms and exactness used in Proposition 2.3.2. This calculation does not prove the Laurent estimate for an arbitrary element f∈Af\in A, nor does it identify these quotients with all analytic localization modules. Those are the additional inputs required by Theorem 2.12.1.

2.14 Weighted coefficient extensions of finite modules

Uniformly bounded lifts of its relations allow a finite presentation to extend to convergent coefficient families. The strict affinoid finite-module theorem provides precisely this bound. We first prove that taking weighted coefficient families preserves the bound, and then identify the resulting coefficient module with an ordinary tensor product. The weights may be arbitrary positive real numbers; no change of the ground field is made.

Throughout, KK is complete for its fixed real-valued non-Archimedean absolute value, which may be trivial. A Banach KK-space has a definite, complete, non-Archimedean, KK-homogeneous norm. Banach algebras are commutative and unital, with a submultiplicative norm and ∥1∥=1\|1\|=1; Banach module actions satisfy ∥am∥≤∥a∥∥m∥\|am\|\leq\|a\|\|m\|. All finite powers have the maximum norm; the maximum over an empty family is zero. Norm-strictness has the quantitative meaning of Definition 1.9.1.

For a countable set II and positive weights w=(wi)i∈Iw=(w_i)_{i\in I}, write Cw(I,E)=c0(I,E;w)\mathcal C_w(I,E)=c_0(I,E;w) as in Lemma 2.13.1; thus Cw(I,E)={(ei)i∈I:for every ε>0,#{i:wi∥ei∥≥ε}<∞},∥(ei)∥w=sup⁡i∈Iwi∥ei∥.\begin{gathered} \mathcal C_w(I,E)= \left\{(e_i)_{i\in I}: \text{for every }\varepsilon>0, \#\{i:w_i\|e_i\|\geq\varepsilon\}<\infty\right\},\\ \|(e_i)\|_w=\sup_{i\in I}w_i\|e_i\|. \end{gathered} We put sup⁡∅=0\sup\varnothing=0. The cited lemma, with coefficient algebra KK if necessary, proves completeness and density of finite-support families. A bounded linear map u:E→Fu:E\to F with bound UU induces a coefficient map with the same bound, since wi∥u(ei)∥≤Uwi∥ei∥w_i\|u(e_i)\|\leq U w_i\|e_i\|. There is also a canonical isometry Cw(I,Em)≃Cw(I,E)m(m<∞).\mathcal C_w(I,E^m)\simeq\mathcal C_w(I,E)^m \qquad(m<\infty). It merely rearranges coefficients: a supremum commutes with a finite maximum, and a finite union of finite sets is finite, so both the norms and the finite-tail conditions agree.

Proposition 2.14.1. NAG-001419 (Strict exactness with unchanged lifting bounds). Let 0⟶E′→ιE→pE′′⟶00\longrightarrow E'\xrightarrow{\iota}E \xrightarrow{p}E''\longrightarrow0 be an exact sequence of Banach KK-spaces with bounded maps. Suppose there are D,C>0D,C>0 such that ∥x∥≤D∥ι(x)∥(x∈E′),∀z∈E′′∃y∈E:p(y)=z,∥y∥≤C∥z∥.\|x\|\leq D\|\iota(x)\|\quad(x\in E'), \qquad \forall z\in E''\quad\exists y\in E: p(y)=z,\quad\|y\|\leq C\|z\|. For every countable II and positive ww, the coefficient sequence is exact and norm-strict: 0⟶Cw(I,E′)→ιwCw(I,E)→pwCw(I,E′′)⟶0.0\longrightarrow\mathcal C_w(I,E') \xrightarrow{\iota_w}\mathcal C_w(I,E) \xrightarrow{p_w}\mathcal C_w(I,E'')\longrightarrow0. The inverse bound DD, the lifting bound CC, and the upper bounds of the original maps remain valid. In particular, these constants do not depend on II or ww. If the norm on E′′E'' is the quotient norm induced by pp, the induced map Cw(I,E)/ιwCw(I,E′)⟶Cw(I,E′′)\mathcal C_w(I,E)/\iota_w\mathcal C_w(I,E') \longrightarrow\mathcal C_w(I,E'') is isometric for the quotient norm. The same assertions hold for bounded maps of Banach modules over a fixed Banach algebra.

Proof. Injectivity and the inclusion im⁡ιw⊆ker⁡pw\operatorname{im}\iota_w\subseteq\ker p_w hold coefficientwise. If y=(yi)∈ker⁡pwy=(y_i)\in\ker p_w, exactness gives unique xi∈E′x_i\in E' with ι(xi)=yi\iota(x_i)=y_i. The bound wi∥xi∥≤Dwi∥yi∥w_i\|x_i\|\leq D w_i\|y_i\| proves that (xi)(x_i) lies in the required coefficient space and gives the inverse bound for ιw\iota_w.

For z=(zi)∈Cw(I,E′′)z=(z_i)\in\mathcal C_w(I,E''), choose a lift yiy_i satisfying ∥yi∥≤C∥zi∥\|y_i\|\leq C\|z_i\|, taking yi=0y_i=0 when zi=0z_i=0. Then (yi)(y_i) has a vanishing weighted tail and ∥(yi)∥w≤C∥z∥w\|(y_i)\|_w\leq C\|z\|_w. This proves surjectivity and its lifting bound. Only countably many coordinate choices are involved. No linear section of pp is asserted or needed. The coefficientwise estimate preceding the proposition gives the upper map bounds. Thus the maps are norm-strict; the middle kernel is closed because pwp_w is continuous.

Now assume that E′′E'' has the quotient norm. The map pwp_w is contractive, so its quotient norm is at least ∥z∥w\|z\|_w. For each fixed λ>1\lambda>1, choose lifts coefficientwise with ∥yi∥≤λ∥zi∥\|y_i\|\leq\lambda\|z_i\|. They give a lift of zz with norm at most λ∥z∥w\lambda\|z\|_w. Taking the infimum over lifts and then letting λ↓1\lambda\downarrow 1 proves equality of the norms. For module maps, the coefficient maps remain AA-linear, and the same argument proves exactness and all the estimates. The selected lifts only establish these bounds; they are not asserted to depend linearly on the data. ◻

We apply this elementary lifting result to a finite presentation over a strict affinoid algebra. The coefficient weights need not be strict: the closed-image theorem is used only over the original coefficient algebra.

Theorem 2.14.2. NAG-001420 (Finite presentations after coefficient extension). Let AA be a strict KK-affinoid algebra with a submultiplicative Banach norm uniformly equivalent to its defining quotient norm. Let MM be a finite AA-module. Choose a surjection q:Am→Mq:A^m\to M and give MM its quotient norm. Choose a finite presentation As→DAm→qM⟶0.A^s\xrightarrow{D}A^m\xrightarrow{q}M\longrightarrow0. For every countable set II and every family of positive weights wiw_i, the coefficient sequence Cw(I,A)s→DwCw(I,A)m→qwCw(I,M)⟶0\mathcal C_w(I,A)^s\xrightarrow{D_w} \mathcal C_w(I,A)^m\xrightarrow{q_w} \mathcal C_w(I,M)\longrightarrow0 is exact. The relation image is closed; DwD_w is norm-strict onto that image, and qwq_w induces an isometry from its normed quotient onto Cw(I,M)\mathcal C_w(I,M). More precisely, there is CD>0C_D>0, depending on DD and the norms but not on I,wI,w, such that every v∈ker⁡qwv\in\ker q_w has a lift aa with Dwa=v,∥a∥w≤CD∥v∥w.D_w a=v,\qquad \|a\|_w\leq C_D\|v\|_w. For every fixed λ>1\lambda>1, each z∈Cw(I,M)z\in\mathcal C_w(I,M) has a lift bb with qwb=zq_wb=z and ∥b∥w≤λ∥z∥w\|b\|_w\leq\lambda\|z\|_w.

Proof. Theorem 1.11.5 makes R=ker⁡qR=\ker q a finite closed submodule of AmA^m. In particular, the quotient norm on MM is complete and is a canonical finite-module norm. Its algebra action is contractive: if vv lifts mm, then avav lifts amam, and taking the infimum of ∥av∥≤∥a∥∥v∥\|av\|\leq\|a\|\|v\| proves the assertion. A finite set of generators of RR provides the stated matrix DD.

Give RR its induced norm from AmA^m. The same theorem, applied to the map D:As→AmD:A^s\to A^m, supplies CDC_D such that every r∈Rr\in R has a preimage of norm at most CD∥r∥C_D\|r\|. This is a bound for the actual image norm, not merely for a quotient topology. Coefficientwise lifting with this constant gives im⁡Dw=Cw(I,R)=ker⁡qw.\operatorname{im}D_w=\mathcal C_w(I,R) =\ker q_w. The second equality and the lifting and quotient-norm assertions for qwq_w follow from Proposition 2.14.1 applied to 0→R→Am→M→00\to R\to A^m\to M\to 0. The finite-power identification was verified above. The kernel is closed by continuity of qwq_w. Finally the matrix DwD_w has the same coefficient bound as DD, and the displayed controlled lifts prove its norm-strictness. No theorem about arbitrary submodules over the extended coefficient algebra has been used. ◻

Power-series and Laurent coefficient algebras

To interpret the preceding presentation as scalar extension, the coefficient space must carry multiplication. Let Γ=Np×Zq\Gamma=\mathbb N^p\times\mathbb Z^q, where p,q≥0p,q\geq 0 are finite. Choose a weight w:Γ→R>0w:\Gamma\to\mathbb R_{>0} satisfying w0=1,wγ+δ≤wγwδ(γ,δ∈Γ).w_0=1,\qquad w_{\gamma+\delta}\leq w_\gamma w_\delta \quad(\gamma,\delta\in\Gamma). For a Banach AA-module MM, abbreviate Cw(M)=Cw(Γ,M)\mathcal C_w(M)=\mathcal C_w(\Gamma,M) and write its elements as ∑γmγXγ\sum_\gamma m_\gamma X^\gamma. The symbols denote coefficient families; convergence refers to the weighted norm just defined.

Lemma 2.14.3. NAG-001421 (Convolution with two-sided exponents). Convolution makes B=Cw(A)B=\mathcal C_w(A) a Banach AA-algebra and Cw(M)\mathcal C_w(M) a Banach BB-module. The constant embedding A→BA\to B is isometric, and ∥bm∥w≤∥b∥w∥m∥w,(bm)η=∑γ+δ=ηbγmδ.\|bm\|_w\leq\|b\|_w\|m\|_w, \qquad (bm)_\eta=\sum_{\gamma+\delta=\eta}b_\gamma m_\delta. The coefficient sum converges even when it is infinite, and its value is independent of its enumeration.

Proof. For finitely supported families, convolution is a finite sum. The module inequality and the weight inequality give wη∥∑γ+δ=ηbγmδ∥≤max⁡γ+δ=η(wγ∥bγ∥)(wδ∥mδ∥)≤∥b∥w∥m∥w.w_\eta\left\|\sum_{\gamma+\delta=\eta}b_\gamma m_\delta\right\| \leq\max_{\gamma+\delta=\eta} (w_\gamma\|b_\gamma\|)(w_\delta\|m_\delta\|) \leq\|b\|_w\|m\|_w. Choose finite-support approximations b(n)→bb^{(n)}\to b and m(n)→mm^{(n)}\to m. These sequences are bounded, and b(n)m(n)−b(k)m(k)=(b(n)−b(k))m(n)+b(k)(m(n)−m(k)).b^{(n)}m^{(n)}-b^{(k)}m^{(k)} = (b^{(n)}-b^{(k)})m^{(n)} +b^{(k)}(m^{(n)}-m^{(k)}). The finite-product estimate makes their products Cauchy. Lemma 2.13.1 gives a limit independent of the approximations, with the same norm bound. Associativity, commutativity for BB, the unit identity and the module law pass from finite supports to limits by this estimate. The constant embedding is isometric because w0=1w_0=1.

To identify the coefficient formula, fix η\eta. The two nonnegative real families wγ∥bγ∥w_\gamma\|b_\gamma\| and wδ∥mδ∥w_\delta\|m_\delta\| are bounded and tend to zero outside finite subsets. For every ε>0\varepsilon>0, only finitely many pairs have product at least ε\varepsilon: if both supremums are nonzero, each factor must be at least ε\varepsilon divided by the other supremum. For pairs with γ+δ=η\gamma+\delta=\eta, wη∥bγmδ∥≤(wγ∥bγ∥)(wδ∥mδ∥).w_\eta\|b_\gamma m_\delta\| \leq(w_\gamma\|b_\gamma\|)(w_\delta\|m_\delta\|). The summability criterion of Lemma 1.1.2 therefore applies. Coordinate evaluation is continuous, so the sum equals the coefficient of the product constructed by finite-support approximation. If either family is zero, this assertion is immediate. ◻

Two choices cover the coefficient spaces needed below. For arbitrary positive radii r1,…,rpr_1,\ldots,r_p and t1,…,tqt_1,\ldots,t_q, the weight wα,β=rαtβ,(α,β)∈Np×Zq,w_{\alpha,\beta}=r^\alpha t^\beta, \qquad(\alpha,\beta)\in\mathbb N^p\times\mathbb Z^q, is multiplicative. It gives multivariable power series and two-sided Laurent series at fixed radii. More generally, for 0<aj≤bj0<a_j\leq b_j use wα,β=rα∏j=1qmax⁡(ajβj,bjβj).w_{\alpha,\beta}=r^\alpha \prod_{j=1}^q\max(a_j^{\beta_j},b_j^{\beta_j}). This weight is submultiplicative: for each Laurent coordinate it is the maximum of the two multiplicative weights at the endpoints. The one-variable case is the coefficient algebra of Proposition 2.13.3. These are coefficient-space definitions; their identification with particular analytic-domain quotients is a separate matter.

Theorem 2.14.4. NAG-001422 (Ordinary and completed coefficient tensor products). Let A,M,q,DA,M,q,D be as in Theorem 2.14.2, and let B=Cw(A)B=\mathcal C_w(A) have the convolution structure above. The natural map Φ:B⊗AM⟶Cw(M),b⊗m⟼(bγm)γ∈Γ,\Phi:B\otimes_A M\longrightarrow\mathcal C_w(M), \qquad b\otimes m\longmapsto(b_\gamma m)_{\gamma\in\Gamma}, is a BB-linear isomorphism. Give the algebraic tensor product its projective seminorm ∥z∥π=inf⁡z=∑j=1lbj⊗mjmax⁡1≤j≤l(∥bj∥w∥mj∥),\|z\|_\pi= \inf_{z=\sum_{j=1}^{l}b_j\otimes m_j} \max_{1\leq j\leq l}\bigl(\|b_j\|_w\|m_j\|\bigr), where only finite tensor expressions occur. Then ∥z∥π=∥Φ(z)∥w.\|z\|_\pi=\|\Phi(z)\|_w. In particular this seminorm is definite and complete, and the canonical map to its separated completion is an isometric isomorphism: B⊗AM≃B⊗^AM≃Cw(M).B\otimes_A M\simeq B\widehat\otimes_A M \simeq\mathcal C_w(M). These identifications are natural in finite AA-modules. No strictness or Noetherianity assumption on BB is required.

Proof. Write e1,…,eme_1,\ldots,e_m for the images of the standard generators under qq. To see the algebraic presentation without assuming exactness of any completed tensor functor, put Q=Bm/DBBsQ=B^m/D_BB^s, where DBD_B is the same matrix with its entries regarded as constants in BB. Sending a vector (bi)(b_i) to ∑ibi⊗ei\sum_i b_i\otimes e_i defines a map Q⟶B⊗AM.Q\longrightarrow B\otimes_A M. Conversely, given b∈Bb\in B and m∈Mm\in M, choose a∈Ama\in A^m with q(a)=mq(a)=m and send (b,m)(b,m) to the class of baba in QQ. Two such lifts differ by DcDc for some c∈Asc\in A^s, so their classes differ by DB(bc)D_B(bc) and agree. Sums of lifts and scalar multiples of lifts show additivity and AA-balancedness. The defining property of the algebraic tensor product therefore gives an inverse map. Thus B⊗AM=QB\otimes_A M=Q by an explicit generators-and-relations argument.

Under the coefficient identifications, DBD_B and the generator map Bm→Cw(M)B^m\to\mathcal C_w(M) are exactly DwD_w and qwq_w. This is first checked on finitely supported families; continuity and Lemma 2.14.3 give BB-linearity on all families. Theorem 2.14.2 identifies QQ with Cw(M)\mathcal C_w(M), with its coefficient norm equal to the quotient norm from BmB^m. The resulting map is precisely Φ\Phi. This proves its algebraic bijectivity and shows why its relation image is closed.

For every tensor expression, contractivity of the AA-action on MM gives ∥Φ(∑jbj⊗mj)∥w≤max⁡j∥bj∥w∥mj∥.\left\|\Phi\left(\sum_jb_j\otimes m_j\right)\right\|_w \leq\max_j\|b_j\|_w\|m_j\|. Taking infima proves ∥Φ(z)∥w≤∥z∥π\|\Phi(z)\|_w\leq\|z\|_\pi. For the reverse inequality, ∥ei∥≤1\|e_i\|\leq 1 because MM has the quotient norm. Every lift (bi)∈Bm(b_i)\in B^m of Φ(z)\Phi(z) gives the tensor expression z=∑ibi⊗eiz=\sum_i b_i\otimes e_i, and hence ∥z∥π≤max⁡i∥bi∥w.\|z\|_\pi\leq\max_i\|b_i\|_w. The infimum of the right side is ∥Φ(z)∥w\|\Phi(z)\|_w by the quotient-norm assertion already proved. This establishes isometry. Completeness and definiteness follow from Lemma 2.13.1, so separated completion adds no elements and removes no nonzero elements.

For an AA-linear map u:M→Nu:M\to N, coefficientwise application of uu sends (bγm)γ(b_\gamma m)_\gamma to (bγu(m))γ(b_\gamma u(m))_\gamma. Thus the identifications commute with 1⊗u1\otimes u on pure tensors, and hence on the whole algebraic tensor product. The maps are bounded by Theorem 1.11.5, so the same assertion holds after completion. ◻

The isometry uses the specified quotient norm on MM. With another uniformly equivalent norm ∥⋅∥∗\|\cdot\|_*, choose 0<a≤b0<a\leq b such that a∥m∥≤∥m∥∗≤b∥m∥a\|m\|\leq\|m\|_*\leq b\|m\|. The corresponding coefficient and projective norms each satisfy the same comparisons with the original ones. Consequently ab∥Φ(z)∥w,∗≤∥z∥π,∗≤ba∥Φ(z)∥w,∗.\frac{a}{b}\|\Phi(z)\|_{w,*} \leq\|z\|_{\pi,*} \leq\frac{b}{a}\|\Phi(z)\|_{w,*}. Thus the bounded isomorphism and completeness are independent of the chosen canonical norm. Uniform equivalence, rather than equality of topologies, is the relevant hypothesis.

Corollary 2.14.5. NAG-001423 (Exact scalar extension on finite modules). With A,BA,B as above, let 0→M′→M→M′′→00\to M'\to M\to M''\to 0 be an exact sequence of finite AA-modules with canonical norms. Then both its ordinary tensor sequence and its separated completed tensor sequence with BB are exact. Under their coefficient identifications they are norm-strict. If the original injection has inverse bound DD on its image and the original surjection has lifting bound CC, these same constants work for the coefficient sequence.

Proof. The original maps are bounded and norm-strict by Theorem 1.11.5. Proposition 2.14.1 gives exactness and the stated constants for coefficient families. Apply the natural identifications of Theorem 2.14.4. For quotient norms these are isometric. For other canonical norms they are uniformly bounded isomorphisms by the preceding comparison, which preserves norm-strictness; the stated constants continue to refer to the coefficient norms themselves. ◻

Remark 2.14.6 (The Laurent interface and its remaining input). Let AA be strict affinoid and MM a finite AA-module with its canonical norm. For one variable and r>0r>0, write B+={∑n≥0anTn:∥an∥rn→0},B−={∑n≤0anTn:∥an∥rn→0},B_+=\left\{\sum_{n\geq0}a_nT^n:\|a_n\|r^n\to0\right\}, \qquad B_-=\left\{\sum_{n\leq0}a_nT^n:\|a_n\|r^n\to0\right\}, B0={∑n∈ZanTn:∥an∥rn→0 as ∣n∣→∞}.B_0=\left\{\sum_{n\in\mathbb Z}a_nT^n: \|a_n\|r^n\to0\text{ as }|n|\to\infty\right\}. The three coefficient modules E+,E−,EE_+,E_-,E used in Theorem 2.12.1 are now canonically B+⊗AMB_+\otimes_A M, B−⊗AMB_-\otimes_A M, B0⊗AMB_0\otimes_A M, respectively, and agree with their separated completed tensors. For B−B_-, use the variable S=T−1S=T^{-1} and radius r−1r^{-1}. The identifications preserve the inclusions into bilateral series and multiplication by every element of AA. All assertions allow arbitrary r>0r>0 and a trivially valued KK, while AA remains strict.

The closed relation images proved here are those of matrices whose entries lie in the original algebra AA. The operator 1−fT−11-fT^{-1} has a nonconstant coefficient in B0B_0, so its injectivity, closed image and a uniform lower bound have not been deduced from this result. Those are the next Laurent inputs. Likewise no arbitrary finite B0B_0-module theorem or analytic-domain universal property is asserted here. The ordinary/completed tensor comparison is proved for this coefficient extension; it is not the false assertion for a finite module tensored with an arbitrary Banach module.

2.15 Algebraic separation from the Stacks foundations

We use the stated parts of two Stacks Project results as declared foundations. Only the short application needed by the Laurent argument is proved below. For an RR-module NN and g∈Rg\in R, the localization NgN_g consists of fractions x/gax/g^a, a≥0a\geq 0, with xga=ygb⟺there is e≥0 with ge(gbx−gay)=0.\frac{x}{g^a}=\frac{y}{g^b} \quad\Longleftrightarrow\quad \text{there is }e\geq0\text{ with }g^e(g^bx-g^ay)=0. In particular, x/1=0x/1=0 means that gex=0g^ex=0 for some e≥0e\geq 0. This is the defining equivalence relation, not an additional exactness theorem.

Corollary 2.15.1. NAG-001426 (Detection by maximal-ideal powers). Let RR be a commutative Noetherian ring with identity and let MM be a finite unital RR-module. Then ⋂m∈Max⁡(R) ⋂k≥1mkM=0.\bigcap_{\mathfrak m\in\operatorname{Max}(R)} \ \bigcap_{k\geq1}\mathfrak m^kM=0.

Proof. If M=0M=0 there is nothing to prove; this includes the zero ring. We interpret an empty intersection as MM. Suppose that a nonzero xx belongs to the displayed intersection. Its annihilator J=Ann⁡R(x)J=\operatorname{Ann}_R(x) is proper. Apply [StacksAlg] (Tag 00E0, Lemma 10.17.2(2)) to the nonzero ring R/JR/J and take the inverse image of a maximal ideal. This gives a maximal ideal m\mathfrak m of RR containing JJ. Put N=⋂k≥1mkMN=\bigcap_{k\geq 1}\mathfrak m^kM.

Apply [StacksAlg] (Tag 00IQ, Lemma 10.51.5(1)) with I=p=mI=\mathfrak p=\mathfrak m. It gives g∉mg\notin\mathfrak m such that Ng=0N_g=0. Since x∈Nx\in N, the defining equivalence relation for localization gives gex=0g^e x=0 for some integer e≥0e\geq 0. Hence ge∈J⊆mg^e\in J\subseteq\mathfrak m. This is impossible, since no nonnegative power of g∉mg\notin\mathfrak m belongs to the prime ideal m\mathfrak m. ◻

References. The localization convention is the fraction construction in [StacksAlg] (Tag 00CM, immediately before Definition 10.9.6 (Tag 07JZ)). The two cited theorem parts are used as the declared Stacks foundation; the argument above is only their short application to the displayed detection statement.

The conclusion concerns the intersection over all maximal ideals. An individual maximal-adic intersection need not vanish over a nonlocal ring: for a field FF, take R=F×FR=F\times F, M=FM=F with action through the first projection, and m=F×0\mathfrak m=F\times 0. Then m\mathfrak m is maximal and mkM=M\mathfrak m^kM=M for every k≥1k\geq 1.

2.16 Injectivity and lower bounds for Laurent operators

The coefficient criterion in Theorem 2.12.1 reduces an elementary Laurent pair to a lower bound for 1−fT−11-fT^{-1}. We now prove injectivity for every positive radius over a strict affinoid algebra. When the radius lies in the value group of the ground field, the coefficient algebra is again strict affinoid, and the finite-module theorem supplies the lower bound. The algebraic separation input is the Stacks-based Corollary 2.15.1; the arguments involving norms are given here.

Keep the fixed complete non-Archimedean field KK, with trivial valuation permitted. For a Banach KK-space VV and r>0r>0, put Er(V)={(vn)n∈Z:∥vn∥rn⟶0 as ∣n∣⟶∞},∥v∥r=sup⁡n∈Z∥vn∥rn.E_r(V)= \left\{(v_n)_{n\in\mathbb Z}: \|v_n\|r^n\longrightarrow0\ \text{as }|n|\longrightarrow\infty\right\}, \qquad \|v\|_r=\sup_{n\in\mathbb Z}\|v_n\|r^n . We also write its elements as Laurent series ∑nvnTn\sum_n v_nT^n. Lemma 2.13.1 gives completeness. For a bounded operator ff on VV, the operator Lf=1−fT−1L_f=1-fT^{-1} is defined by (Lfv)n=vn−fvn+1.(L_fv)_n=v_n-fv_{n+1}.

Lemma 2.16.1. NAG-001428 (Scalar recurrence over a normed residue field). Let FF be a field which is a Banach KK-algebra with a definite submultiplicative norm. Let VV be a finite-dimensional FF-vector space with a Banach KK-norm uniformly equivalent to the maximum coordinate norm for an FF-basis. For every λ∈F\lambda\in F and every r>0r>0, the operator LλL_\lambda on Er(V)E_r(V) is bounded and injective. No multiplicativity of the norm on FF, and no finiteness of F/KF/K, is assumed.

Proof. Choose a basis and comparison constants C,D>0C,D>0 such that, for v=∑jajejv=\sum_j a_je_j, max⁡j∥aj∥F≤C∥v∥,∥v∥≤Dmax⁡j∥aj∥F.\max_j\|a_j\|_F\leq C\|v\|, \qquad \|v\|\leq D\max_j\|a_j\|_F . These inequalities identify Er(V)E_r(V), by a bounded isomorphism with bounded inverse, with a finite power of Er(F)E_r(F). Multiplication by λ\lambda has bound DC∥λ∥FDC\|\lambda\|_F, so LλL_\lambda is bounded. If λ=0\lambda=0, it is the identity. Suppose λ≠0\lambda\ne 0 and Lλv=0L_\lambda v=0. Each coordinate sequence then satisfies an=λan+1,an=λ−na0(n∈Z).a_n=\lambda a_{n+1}, \qquad a_n=\lambda^{-n}a_0 \quad(n\in\mathbb Z). If a0≠0a_0\ne 0, then ana−n=a02≠0a_na_{-n}=a_0^2\ne 0, whereas submultiplicativity gives 0<∥a02∥F≤(∥an∥Frn)(∥a−n∥Fr−n)⟶0(n⟶+∞).0<\|a_0^2\|_F \leq \bigl(\|a_n\|_Fr^n\bigr) \bigl(\|a_{-n}\|_Fr^{-n}\bigr) \longrightarrow0 \qquad(n\longrightarrow+\infty). This is impossible. Thus every coordinate has a0=0a_0=0, and the recurrence makes every coefficient zero. The argument also covers V=0V=0. ◻

Theorem 2.16.2. NAG-001429 (Laurent injectivity for finite strict affinoid modules). Let AA be a strict KK-affinoid algebra with a submultiplicative Banach norm uniformly equivalent to its defining quotient norm. Let MM be a finite AA-module with a canonical Banach norm and contractive AA-action. For every f∈Af\in A and every real r>0r>0, Lf=1−fT−1:Er(M)⟶Er(M)L_f=1-fT^{-1}:E_r(M)\longrightarrow E_r(M) is injective. This assertion concerns injectivity alone; it does not assert a lower norm bound for every rr.

Proof. Residue-field quotients. By Theorem 1.11.5, AA is Noetherian, its ideals are closed, and finite modules and their subquotients have canonical complete norms. Fix a maximal ideal m\mathfrak m. The field F=A/mF=A/\mathfrak m, with the quotient norm, is a Banach KK-algebra. Indeed, definiteness and completeness follow from Lemma 1.11.1; approximating representatives in a product proves submultiplicativity. Also ∥1F∥≤1\|1_F\|\leq 1 and 0<∥1F∥≤∥1F∥20<\|1_F\|\leq\|1_F\|^2, so ∥1F∥=1\|1_F\|=1. The scalar homogeneity follows from the quotient definition.

Every finite AA-module VV annihilated by m\mathfrak m is a finite-dimensional FF-vector space. An FF-basis gives an AA-linear isomorphism Fd→VF^d\to V. Applying Theorem 1.11.5 to this isomorphism and its inverse proves uniform equivalence of the canonical norm on VV and the maximum coordinate norm. Lemma 2.16.1 therefore proves injectivity on Er(V)E_r(V) for the action of f mod mf\bmod\mathfrak m.

Powers of a maximal ideal. We prove injectivity on Er(M/mkM)E_r(M/\mathfrak m^kM) by induction on k≥1k\geq 1. The case k=1k=1 is the preceding case. For k>1k>1, use 0⟶mk−1M/mkM⟶M/mkM⟶M/mk−1M⟶0.0\longrightarrow \mathfrak m^{k-1}M/\mathfrak m^kM \longrightarrow M/\mathfrak m^kM \longrightarrow M/\mathfrak m^{k-1}M \longrightarrow0 . All three modules are finite; their maps are bounded and norm-strict by Theorem 1.11.5. Proposition 2.14.1, with weights rnr^n, makes the corresponding coefficient sequence exact. The left term is annihilated by m\mathfrak m, so the residue-field argument applies to it. If a coefficient family in the middle is killed by LfL_f, its image on the right is zero by induction. It comes from the left, and injectivity of the coefficient inclusion shows that this preimage is also killed by LfL_f. Injectivity on the left makes the preimage zero. This proves the induction.

Separation of coefficients. Let v∈Er(M)v\in E_r(M) satisfy Lfv=0L_fv=0. Its image in every Er(M/mkM)E_r(M/\mathfrak m^kM) is zero by the induction. Each coefficient vnv_n thus belongs to mkM\mathfrak m^kM for every maximal m\mathfrak m and every k≥1k\geq 1. Corollary 2.15.1 gives vn=0v_n=0. Hence v=0v=0. At no point did we use finite-dimensionality of FF over KK, a multiplicative quotient norm, or a point of the analytic spectrum. ◻

Injectivity does not by itself provide the numerical input of the Laurent criterion. For the unit radius, the coefficient module is finite over another strict affinoid algebra. This permits a second application of the quantitative finite-module theorem.

Lemma 2.16.3. NAG-001430 (The unit Laurent coefficient algebra is strict). For AA as in Theorem 2.16.2, the convolution algebra B=E1(A)B=E_1(A) is strict KK-affinoid with its coefficient norm. More precisely, the normal-form map gives an isometric algebra isomorphism A⟨T,U⟩/(TU−1) → ∼  B,T⟼Z,U⟼Z−1.A\langle T,U\rangle/(TU-1)\ \xrightarrow{\ \sim\ }\ B, \qquad T\longmapsto Z,\quad U\longmapsto Z^{-1}. For every finite AA-module MM with a canonical norm, E1(M)E_1(M) is a finite BB-module and its coefficient norm is uniformly equivalent to its canonical finite-BB-module norm. If MM has the quotient norm from a chosen Am→MA^m\to M, then E1(M)E_1(M) has exactly the quotient norm from the induced Bm→E1(M)B^m\to E_1(M).

Proof. Theorem 2.13.2 and Proposition 2.13.3, with a=b=1a=b=1, give the displayed algebra isomorphism, the actual kernel (TU−1)A⟨T,U⟩(TU-1)A\langle T,U\rangle, and equality of the quotient and coefficient norms. To verify strictness rather than assume it, choose an admissible bounded surjection π:Rd=K⟨X1,…,Xd⟩⟶A.\pi:R_d=K\langle X_1,\ldots,X_d\rangle\longrightarrow A . Its kernel is closed, and admissibility supplies a uniform bound for lifts of elements of AA. Applying Proposition 2.14.1 to 0→ker⁡π→Rd→A→00\to\ker\pi\to R_d\to A\to 0, with index set N2\mathbb N^2 and unit weights, extends π\pi to a bounded surjection with the same lifting bound Rd⟨T,U⟩⟶A⟨T,U⟩.R_d\langle T,U\rangle\longrightarrow A\langle T,U\rangle . Regrouping coefficients gives an isometric algebra identification Rd⟨T,U⟩=K⟨X1,…,Xd,T,U⟩.R_d\langle T,U\rangle=K\langle X_1,\ldots,X_d,T,U\rangle. To check the convergence condition, joint coefficients tending to zero outside finite subsets imply that each (T,U)(T,U)-row has the same property and that the row norms tend to zero. Conversely, only finitely many rows can have norm at least a fixed ε>0\varepsilon>0, and in each of those rows only finitely many coefficients can have norm at least ε\varepsilon. The norms are equal because the two suprema commute. Multiplication agrees first for polynomials and then by continuity.

The normal-form quotient onto BB has an isometric AA-linear section. Composing it with the preceding surjection gives a bounded surjection from this unit-radius Tate algebra to BB, with uniform lifts. Thus the coefficient norm on BB is uniformly equivalent to its defining quotient norm, as required by the definition of strict affinoid. Finally Theorem 2.14.4, for Z\mathbb Z and unit weights, identifies E1(M)E_1(M) with the finite BB-module B⊗AMB\otimes_A M. The coefficient presentation in Theorem 2.14.2 gives the asserted equality with the quotient norm from BmB^m. Replacing the chosen canonical norm on MM by an equivalent one preserves uniform equivalence of coefficient norms. ◻

Theorem 2.16.4. NAG-001431 (A lower Laurent bound at value-group radii). Under the hypotheses on A,MA,M in Theorem 2.16.2, let f∈Af\in A and r∈∣K×∣r\in|K^\times|. There exists a real constant c=c(A,M,f,r,∥⋅∥)>0c=c(A,M,f,r,\|\cdot\|)>0 such that ∥Lfw∥r≥c∥w∥r(w∈Er(M)).\|L_fw\|_r\geq c\|w\|_r\qquad(w\in E_r(M)). The constant is independent of ww. Consequently LfL_f has closed image and is norm-strict onto its image. For a trivially valued ground field this theorem includes the radius r=1r=1.

Proof. First let r=1r=1. Lemma 2.16.3 makes E1(M)E_1(M) a finite module with canonical norm over the strict affinoid algebra B=E1(A)B=E_1(A). The operator LfL_f is multiplication by 1−fZ−1∈B1-fZ^{-1}\in B, hence is BB-linear. By Theorem 1.11.5, it has closed image and is norm-strict. Thus some C>0C>0 has the property that every yy in the image has a preimage of norm at most C∥y∥1C\|y\|_1. Theorem 2.16.2 makes that preimage unique. Substituting y=Lfwy=L_fw gives ∥w∥1≤C∥Lfw∥1\|w\|_1\leq C\|L_fw\|_1 and hence the required bound with c=C−1c=C^{-1}.

For general r∈∣K×∣r\in|K^\times|, choose s∈K×s\in K^\times with ∣s∣=r|s|=r. The substitution T=sZT=sZ gives an isometry Φs:Er(M)⟶E1(M),(mn)n⟼(snmn)n.\Phi_s:E_r(M)\longrightarrow E_1(M), \qquad (m_n)_n\longmapsto(s^nm_n)_n . The norms and vanishing tails agree term by term. A coefficient calculation gives ΦsLf=Lf/sΦs.\Phi_s L_f=L_{f/s}\Phi_s . Applying the unit-radius bound to f/sf/s proves the assertion. The image and norm-strictness statements follow either by this isometry or directly from the lower bound and completeness: convergence of images makes preimages Cauchy, and continuity identifies their limit. No scalar extension or change in the absolute value is involved. ◻

Corollary 2.16.5. NAG-001432 (Strict exactness for a coefficient Laurent pair). Let A,M,f,rA,M,f,r satisfy Theorem 2.16.4. Define E+,E−,EE_+,E_-,E, the relation operators, and P+,P−,P0P_+,P_-,P_0 exactly as in Theorem 2.12.1, with these M,f,rM,f,r. All three quotients are Banach AA-modules for their quotient norms, and 0⟶M→m↦([m],[m])P+⊕P−→([u],[v])↦[u−v]P0⟶00\longrightarrow M \xrightarrow{m\mapsto([m],[m])}P_+\oplus P_- \xrightarrow{([u],[v])\mapsto[u-v]}P_0 \longrightarrow0 is strict exact. With cc supplied by Theorem 2.16.4, the augmentation estimate and the final quotient isometry of Theorem 2.12.1 hold. This applies to every f∈Af\in A, without a condition that ff be a coordinate or a unit.

Proof. Theorem 2.16.4 verifies the sole additional lower-bound hypothesis of Theorem 2.12.1. Apply that theorem to the specified coefficient modules and their actual relation images. Its proof gives closedness of all relation images, completeness of the quotients, both exactness assertions, and the stated quantitative estimates. ◻

The coefficient argument feeds into the analytic Laurent sequence of Theorem 2.18.3. Finite refinement and simultaneous strictification extend it to the rational Čech complexes of Theorem 2.32.2. The subsequent domain and sheaf arguments culminate in Theorems 2.38.1 and 2.39.6.

2.17 Strict quotient presentations and finite modules

The coefficient construction is useful for localization only after its relation quotients are controlled. In this section the new radii lie in the value group. This allows the finite-module theorem over strict affinoids to control the actual images of the relation maps. No statement about a structure sheaf is used.

Fix a complete real-valued non-Archimedean field KK, allowing the trivial absolute value. A nonzero algebra below carries the submultiplicative norm and norm-admissibility convention of Definition 1.1.3. We also allow a relation ideal to be the whole algebra: its quotient is the zero ring with its zero norm, and every unital module over it is zero. Statements about such a quotient have this explicit degenerate interpretation; the normalization ∥1∥=1\|1\|=1 is required only for nonzero algebras.

Lemma 2.17.1. NAG-001434 (Strict weighted adjunction and relation quotients). Let AA be a nonzero strict KK-affinoid algebra with a Banach norm uniformly equivalent to a defining quotient norm. Let n≥0n\geq 0 and choose ρ1,…,ρn∈∣K×∣\rho_1,\ldots,\rho_n\in|K^\times|. Put D=A{ρ−1X},∥∑α∈NnaαXα∥ρ=sup⁡α∥aα∥ρα.D=A\{\rho^{-1}X\},\qquad \left\|\sum_{\alpha\in\mathbb N^n}a_\alpha X^\alpha\right\|_\rho =\sup_\alpha\|a_\alpha\|\rho^\alpha. Then DD is strict KK-affinoid in the norm-admissible sense. Every algebraic ideal J⊂DJ\subset D is finitely generated and closed. With the quotient norm, B=D/JB=D/J is a strict affinoid quotient, or the zero quotient under the convention above. If J=(h1,…,ht)J=(h_1,\ldots,h_t), the map Dt⟶D,(dj)⟼∑jhjdjD^t\longrightarrow D,\qquad (d_j)\longmapsto\sum_jh_jd_j has actual image JJ and is norm-strict onto that image. No unit-ideal hypothesis on other elements of AA is needed for this assertion.

Proof. Choose a bounded admissible surjection p:Rd=K⟨Z1,…,Zd⟩→Ap:R_d=K\langle Z_1,\ldots,Z_d\rangle\to A. There are constants Cp,Lp>0C_p,L_p>0 such that ∥p(z)∥≤Cp∥z∥\|p(z)\|\leq C_p\|z\| and every a∈Aa\in A has a lift of norm at most Lp∥a∥L_p\|a\|, choosing zero for a=0a=0. Applying the coefficient lifting argument of Proposition 2.14.1, with index set Nn\mathbb N^n and unit weights, gives a surjection Rd⟨Y1,…,Yn⟩⟶A⟨Y1,…,Yn⟩R_d\langle Y_1,\ldots,Y_n\rangle \longrightarrow A\langle Y_1,\ldots,Y_n\rangle with the same upper bound and lifting bound. This is an algebra homomorphism: the assertion holds on polynomial coefficient families and passes to their limits by Lemma 2.14.3. Regrouping the joint coefficients identifies its source isometrically with Rd+nR_{d+n}. Indeed, a joint family tends to zero outside finite subsets exactly when each row does so and the row norms tend to zero; finite unions of the finitely many relevant rows prove the reverse implication.

Choose ci∈K×c_i\in K^\times with ∣ci∣=ρi|c_i|=\rho_i. The substitution Xi=ciYiX_i=c_iY_i is an isometric algebra isomorphism from DD to A⟨Y1,…,Yn⟩A\langle Y_1,\ldots,Y_n\rangle: it sends aαa_\alpha to cαaαc^\alpha a_\alpha, preserving both the weighted norm and the finite-tail condition. Thus the displayed surjection supplies a bounded admissible unit-radius Tate presentation of DD. Theorem 1.11.5, now applied to DD, proves finite generation and closedness of JJ, as well as norm-strictness of the displayed relation map.

Lemma 1.11.1 makes BB complete with a definite quotient norm. Approximating lifts in DD shows that this norm is submultiplicative. If B≠0B\ne 0, its unit has norm at most one; writing u=∥1B∥>0u=\|1_B\|>0, submultiplicativity gives u≤u2u\leq u^2, hence u=1u=1. The quotient map has lifts bounded by any fixed factor λ>1\lambda>1. Composing these lifts with the preceding Tate-presentation lifts gives a bounded admissible surjection from Rd+nR_{d+n} to BB. This proves strictness when B≠0B\ne 0; when J=DJ=D, all quotient assertions reduce to the zero space. ◻

The same presentation also controls the relations on a finite module. It is useful to retain exact quotient norms here, because these give an isometry rather than only an unspecified bounded comparison.

Theorem 2.17.2. NAG-001435 (Finite modules over a strict relation quotient). Use the notation of Lemma 2.17.1. Let MM be a finite AA-module, choose a surjection q:Am→Mq:A^m\to M, and give MM its quotient norm. Write E=M{ρ−1X},JE={∑j=1thjej:ej∈E},Q=E/JE.E=M\{\rho^{-1}X\},\qquad JE=\left\{\sum_{j=1}^t h_je_j:e_j\in E\right\},\qquad Q=E/JE. Here J=(h1,…,ht)J=(h_1,\ldots,h_t), and EE has its coefficient norm. Then EE is a finite DD-module with a canonical norm; JEJE is its actual algebraic submodule, is closed, and does not depend on the chosen generators of JJ. The relation map Et→EE^t\to E is norm-strict onto JEJE. The coefficient and tensor maps induce natural isometric BB-linear isomorphisms B⊗AM ≃ B⊗^AM ≃ Q,d‾⊗m⟼[d m].B\otimes_A M \ \simeq\ B\widehat\otimes_A M \ \simeq\ Q,\qquad \overline d\otimes m\longmapsto[d\,m]. The algebraic tensor product carries the projective seminorm, the completed tensor product is its separated completion, and QQ has the quotient norm. This is also the finite-BB-module quotient norm induced by the generators q(e1),…,q(em)q(e_1),\ldots,q(e_m). For any other uniformly equivalent canonical norm on MM, the same maps are bounded isomorphisms with uniformly equivalent norms.

Proof. Step 1: the actual relation module. Choose a finite presentation As→UAm→qM⟶0.A^s\xrightarrow{U}A^m\xrightarrow{q}M\longrightarrow0. Theorems 2.14.2 and 2.14.4, for the multiplicative weights ρα\rho^\alpha, identify E=Dm/UDDs=D⊗AM,E=D^m/U_DD^s=D\otimes_A M, with its coefficient norm equal to the quotient norm from DmD^m. Thus it is a finite DD-module with a canonical norm. Lemma 2.17.1 makes DD strict. Theorem 1.11.5 applied over DD now shows that the map (ej)↦∑jhjej(e_j)\mapsto\sum_jh_je_j has closed image and is norm-strict. Its image is the algebraic submodule JEJE, not merely its closure. If one changes generators of JJ, each new generator is a finite DD-linear combination of the old ones, so this submodule is unchanged.

Step 2: the algebraic identification and the quotient norm. Apply right exactness of the ordinary tensor product, in [StacksAlg] (Tag 00DF, Lemma 10.12.10), to the displayed finite presentation. It gives B⊗AM=Bm/UBBs.B\otimes_A M=B^m/U_BB^s. Under the surjection Dm→ED^m\to E, the inverse image of JEJE is exactly UDDs+JDm.U_DD^s+JD^m. Indeed, lift the finitely many terms of an expression ∑jhjej\sum_jh_je_j to DmD^m; their difference from any other lift lies in UDDsU_DD^s. The reverse inclusion is immediate. Under Dm→BmD^m\to B^m, the inverse image of UBBsU_BB^s is the same submodule: lift a finite vector in BsB^s to DsD^s and subtract its image under UDU_D. Consequently both QQ and B⊗AMB\otimes_A M are the quotient of DmD^m by this identical actual submodule. This proves the displayed algebraic identification, including the formula d‾⊗m↦[d m]\overline d\otimes m\mapsto[d\,m].

The maximum norm on BmB^m is exactly the quotient norm from DmD^m. To check the only nontrivial inequality, lift each of the finitely many coordinates with any fixed factor λ>1\lambda>1, then let λ↓1\lambda\downarrow 1. Successive quotient norms therefore give on both modules the infimum of the norms of all their lifts in DmD^m. They agree. This also shows that the norm on QQ is its indicated finite-BB-module norm. Since JEJE is closed, Lemma 1.11.1 makes it definite and complete.

Step 3: the projective norm. Let ν\nu denote this common quotient norm on P=B⊗AMP=B\otimes_A M, and write mi=q(ei)m_i=q(e_i). Since ∥mi∥≤1\|m_i\|\leq 1, every lift (bi)∈Bm(b_i)\in B^m of zz gives z=∑ibi⊗mi,∥z∥π≤max⁡i∥bi∥.z=\sum_i b_i\otimes m_i,\qquad \|z\|_\pi\leq\max_i\|b_i\|. Taking the infimum gives ∥z∥π≤ν(z)\|z\|_\pi\leq\nu(z). Conversely, fix a finite expression z=∑jbj⊗vjz=\sum_jb_j\otimes v_j and λ>1\lambda>1. Choose lifts aj=(aij)∈Ama_j=(a_{ij})\in A^m with ∥aj∥≤λ∥vj∥\|a_j\|\leq\lambda\|v_j\|, using zero when vj=0v_j=0. The map A→BA\to B is contractive, because the constant inclusion in DD is isometric and D→BD\to B is a norm quotient. Thus z=∑i(∑jbjaij‾)⊗mi,ν(z)≤λmax⁡j(∥bj∥∥vj∥).z=\sum_i\left(\sum_jb_j\overline{a_{ij}}\right)\otimes m_i, \qquad \nu(z)\leq \lambda\max_j\bigl(\|b_j\|\|v_j\|\bigr). Letting λ↓1\lambda\downarrow 1 and taking the infimum over finite tensor expressions proves ν(z)≤∥z∥π\nu(z)\leq\|z\|_\pi. The projective seminorm therefore equals the complete definite quotient norm. Its separated completion neither adds elements nor kills nonzero elements. If B=0B=0 or M=0M=0, the same conclusions hold on the zero module.

Step 4: functoriality and other norms. For an AA-linear map u:M→M′u:M\to M', coefficientwise application is bounded by Theorem 1.11.5. It commutes with each multiplication by hjh_j, so it descends to the relation quotients. On pure tensors the resulting map is d‾⊗m↦d‾⊗u(m)\overline d\otimes m\mapsto\overline d\otimes u(m); this proves naturality for both ordinary and completed tensors. Finally, if a∥m∥≤∥m∥∗≤b∥m∥a\|m\|\leq\|m\|_*\leq b\|m\| for 0<a≤b0<a\leq b, then coefficient, quotient and projective norms each satisfy the same comparisons with their original norms. Thus the same identifications remain bounded isomorphisms for any other canonical norm. ◻

In particular, one may take J=(gX1−f1,…,gXn−fn)J=(gX_1-f_1,\ldots,gX_n-f_n) for arbitrary g,fi∈Ag,f_i\in A. The relation quotient and its finite-module comparison are already defined in this generality. A unit-ideal condition on (g,f1,…,fn)(g,f_1,\ldots,f_n) enters only when this quotient is to represent a rational analytic domain. No flatness of A→BA\to B follows from the theorem: it identifies a single tensor quotient and does not assert preservation of injections.

2.18 Iterated quotients and elementary Laurent diagrams

The preceding quotient construction permits a direct comparison of presentations. We first show that adjoining a second block of variables commutes with taking a quotient, with equality of the resulting norms. We then identify the coefficient Laurent sequence with a sequence of finite modules over explicit strict affinoid quotient algebras. These are statements about Banach algebras and modules; identifying them with sections on analytic domains requires the geometric localization theorem.

Throughout this section, KK is complete for its fixed real-valued non-Archimedean absolute value, which may be trivial, and AA is a strict KK-affinoid algebra with the norm conventions of Lemma 2.17.1. All displayed radii belong to ∣K×∣|K^\times|. A relation quotient is either a strict affinoid algebra or the zero algebra with its zero norm. The latter is allowed in this section as a separate case; it does not satisfy the normalization ∥1∥=1\|1\|=1 in Definition 1.1.3. Every unital module over it is zero. All quotients below use the actual algebraic ideals or submodules, and their indicated quotient norms.

Lemma 2.18.1. NAG-001437 (Adjoining variables after a quotient). Let ρ=(ρ1,…,ρp)\rho=(\rho_1,\ldots,\rho_p) and σ=(σ1,…,σq)\sigma=(\sigma_1,\ldots,\sigma_q) be finite tuples in ∣K×∣|K^\times|, allowing p=0p=0 or q=0q=0. Put D=A{ρ−1X},B=D/I,F=A{ρ−1X,σ−1Y},D=A\{\rho^{-1}X\},\qquad B=D/I,\qquad F=A\{\rho^{-1}X,\sigma^{-1}Y\}, where II is an ideal of DD. The coefficient quotient induces an isometric algebra isomorphism F/IF → ∼  B{σ−1Y}.F/IF\ \xrightarrow{\ \sim\ }\ B\{\sigma^{-1}Y\}. In particular, its kernel is the actual extended ideal IFIF, which is closed. If J=(q1,…,qt)J=(q_1,\ldots,q_t) is an ideal of B{σ−1Y}B\{\sigma^{-1}Y\} and Qj∈FQ_j\in F are any lifts of its displayed generators, then (B{σ−1Y})/J ≃ F/(IF+(Q1,…,Qt))\bigl(B\{\sigma^{-1}Y\}\bigr)/J \ \simeq\ F/(IF+(Q_1,\ldots,Q_t)) isometrically. The last ideal and the resulting isomorphism are independent of the chosen lifts.

Proof. Regrouping coefficients identifies FF isometrically with D{σ−1Y}D\{\sigma^{-1}Y\}. Indeed, for the coefficients aαβa_{\alpha\beta}, the joint finite-tail condition with weights ρασβ\rho^\alpha\sigma^\beta is equivalent to the following two conditions: each β\beta-row belongs to DD, and the row norms multiplied by σβ\sigma^\beta tend to zero outside finite sets. In the reverse direction, only finitely many rows can have weighted norm at least a fixed ε>0\varepsilon>0, and each of these rows contributes only finitely many coefficients of that size. Both norms equal sup⁡α,β∥aαβ∥ρασβ\sup_{\alpha,\beta}\|a_{\alpha\beta}\|\rho^\alpha\sigma^\beta. Multiplication agrees on polynomials and hence everywhere by density and continuity.

By Lemma 2.17.1 and Theorem 1.11.5, DD is strict, II is finite and closed, and D→BD\to B is a quotient map of Banach spaces. Proposition 2.14.1 therefore shows that its coefficient extension is surjective, has kernel c0(Nq,I;σβ),c_0(\mathbb N^q,I;\sigma^\beta), and induces precisely the coefficient norm on the quotient. Choose generators u1,…,usu_1,\ldots,u_s of II. The generator map Ds→ID^s\to I, where II has its induced norm, is norm-strict by Theorem 1.11.5. Thus some C>0C>0 permits every i∈Ii\in I to be written as i=∑ℓuℓdℓi=\sum_\ell u_\ell d_\ell with max⁡ℓ∥dℓ∥≤C∥i∥\max_\ell\|d_\ell\|\leq C\|i\|. Applying this bound coefficientwise to any element of the displayed kernel produces ss series in D{σ−1Y}D\{\sigma^{-1}Y\}, proving that the kernel is IFIF, without taking a closure. The reverse inclusion follows by applying the quotient map.

Write π:F→B{σ−1Y}\pi:F\to B\{\sigma^{-1}Y\} for this surjection. If h∈π−1(J)h\in\pi^{-1}(J), express π(h)=∑jqjbj\pi(h)=\sum_j q_j b_j and lift each of the finitely many bjb_j to Hj∈FH_j\in F. Then h−∑jQjHj∈IFh-\sum_jQ_jH_j\in IF. Hence π−1(J)=IF+(Q1,…,Qt)\pi^{-1}(J)=IF+(Q_1,\ldots,Q_t) as actual ideals. This ideal is closed because FF is strict. The two successive quotient norms agree with the direct quotient norm: taking the infimum over lifts through the intermediate quotient and then over representatives modulo JJ is the same as taking the infimum over all representatives modulo π−1(J)\pi^{-1}(J). Approximation of each infimum by any fixed factor greater than one proves both inequalities even when an infimum is not attained. Changing a lift QjQ_j adds an element of IFIF, so does not change the ideal. If I=DI=D, both first quotients are zero and the same statements hold directly. ◻

Theorem 2.18.2. NAG-001438 (The completed tensor product of two quotient presentations). Let B=A{ρ−1X}/I,C=A{σ−1Y}/J,H=A{ρ−1X,σ−1Y}/(I,J).\begin{aligned} B&=A\{\rho^{-1}X\}/I,\\ C&=A\{\sigma^{-1}Y\}/J,\\ H&=A\{\rho^{-1}X,\sigma^{-1}Y\}/(I,J). \end{aligned} where (I,J)(I,J) denotes the sum of the two extended ideals. Use the stated quotient norms on B,C,HB,C,H, and the projective seminorm on B⊗ACB\otimes_A C. The multiplication of the two canonical maps into HH induces an isometric algebra isomorphism B⊗^AC → ∼  H.B\widehat\otimes_A C\ \xrightarrow{\ \sim\ }\ H. This is a statement about the separated completion; it does not assert that the algebraic tensor product is already complete or has definite projective seminorm. The isomorphism commutes with the two maps from B,CB,C.

Proof. If either BB or CC is zero, HH is zero as well and the assertion follows. Assume otherwise, and put G=B⊗^ACG=B\widehat\otimes_A C. To make the analytic construction explicit, the seminorm on the algebraic tensor product is ∥z∥π=inf⁡z=∑j=1nbj⊗cjmax⁡j∥bj∥∥cj∥.\|z\|_\pi= \inf_{z=\sum_{j=1}^n b_j\otimes c_j} \max_j\|b_j\|\|c_j\|. Multiplying two finite tensor expressions and taking infima gives ∥zz′∥π≤∥z∥π∥z′∥π\|zz'\|_\pi\leq\|z\|_\pi\|z'\|_\pi. Thus its seminorm-zero subspace is an ideal, multiplication descends to the separated quotient and extends continuously to GG. The maps b↦b⊗1b\mapsto b\otimes 1 and c↦1⊗cc\mapsto 1\otimes c are contractive. The unit has norm at most one; if G≠0G\ne 0, submultiplicativity then forces its norm to be one. The construction also gives a commutative unital Banach algebra when nonzero.

The canonical maps B→HB\to H and C→HC\to H are contractive: the maps from their coefficient algebras preserve their weights and then pass to quotient norms. Their product is an AA-balanced bilinear map with bound one. On a finite tensor expression its image consequently has norm at most max⁡j∥bj∥∥cj∥\max_j\|b_j\|\|c_j\|. Taking the infimum and then completing gives a contractive algebra homomorphism μ:G→H\mu:G\to H.

In the other direction, let Xˉi∈B\bar X_i\in B and Yˉj∈C\bar Y_j\in C denote the variable classes. Their powers satisfy ∥Xˉin⊗1∥G≤ρin,∥1⊗Yˉjn∥G≤σjn(n≥0).\|\bar X_i^n\otimes1\|_G\leq\rho_i^n, \qquad \|1\otimes\bar Y_j^n\|_G\leq\sigma_j^n \qquad(n\geq0). The two maps from AA agree by the balancing relation. Proposition 1.9.3, with all bounds equal to one, therefore defines a contractive map from A{ρ−1X,σ−1Y}A\{\rho^{-1}X,\sigma^{-1}Y\} to GG, sending the variables to these tensors. If G=0G=0, the zero map has the same properties directly. Its restriction to the first coefficient algebra factors through BB, as can first be checked on polynomials and then on convergent series. Its restriction to the second factors through CC. It therefore kills II and JJ and induces a contractive map ν:H→G\nu:H\to G.

The composite μν\mu\nu fixes every coefficient and variable, hence fixes the dense polynomial image in HH and is the identity. Likewise νμ\nu\mu fixes the images of BB and CC, since each is a quotient of a coefficient algebra with dense polynomial image. It fixes every pure tensor and hence the dense algebraic tensor image in GG. It is therefore the identity. Two inverse contractive maps are isometric, proving the assertion. ◻

Theorem 2.18.3. NAG-001439 (The quotient-algebra Laurent sequence). Let MM be a finite AA-module, give it the quotient norm from a chosen surjection Am→MA^m\to M, and fix f∈Af\in A and r∈∣K×∣r\in|K^\times|. Put A+=A{r−1T}/(T−f),A−=A{rU}/(1−fU),A0=A{r−1T,rU}/(T−f,TU−1),Mτ=Aτ⊗AM(τ∈{+,−,0}).\begin{split} A_+&=A\{r^{-1}T\}/(T-f),\\ A_-&=A\{rU\}/(1-fU),\\ A_0&=A\{r^{-1}T,rU\}/(T-f,TU-1),\\ M_\tau&=A_\tau\otimes_A M\qquad(\tau\in\{+,-,0\}). \end{split} Give each MτM_\tau its quotient norm from AτmA_\tau^m. These are finite canonical modules over the corresponding strict quotient algebras, and the ordinary-to-completed tensor maps are isometries. There are canonical isometric identifications M+≃P+,M−≃P−,M0≃P0,M_+\simeq P_+,\qquad M_-\simeq P_-,\qquad M_0\simeq P_0, where P+,P−,P0P_+,P_-,P_0 are the coefficient quotients of Corollary 2.16.5. Under them the quotient-algebra maps give exactly the coefficient inclusions used there, and hence 0⟶M→m↦(1⊗m,1⊗m)M+⊕M−→(u,v)↦u∣0−v∣0M0⟶00\longrightarrow M \xrightarrow{m\mapsto(1\otimes m,1\otimes m)} M_+\oplus M_- \xrightarrow{(u,v)\mapsto u|_0-v|_0} M_0\longrightarrow0 is strict exact with the same augmentation estimate and final quotient isometry. Moreover, the two maps to A0A_0 induce an isometric isomorphism A+⊗^AA−≃A0A_+\widehat\otimes_A A_-\simeq A_0. If the norm on MM is replaced by any uniformly equivalent canonical norm, the identifications are uniformly bounded isomorphisms and the sequence remains strict exact.

Proof. Apply Lemma 2.17.1 and Theorem 2.17.2 to each displayed presentation. For the first presentation the coefficient module is E+=M{r−1T}E_+=M\{r^{-1}T\}, with relation image (T−f)E+(T-f)E_+. For the second, the isometry M{rU}⟶E−,∑j≥0mjUj⟼∑j≥0mjZ−jM\{rU\}\longrightarrow E_-,\qquad \sum_{j\geq0}m_jU^j\longmapsto\sum_{j\geq0}m_j Z^{-j} takes multiplication by 1−fU1-fU to multiplication by 1−fZ−11-fZ^{-1}. This proves the first two identifications, including their norms and actual relation images.

For the third, set DM=M{r−1T,rU}D_M=M\{r^{-1}T,rU\}. Theorem 2.13.2 with a=b=ra=b=r gives an isometric quotient map N:DM⟶Er(M),ker⁡N=(TU−1)DM.\mathcal N:D_M\longrightarrow E_r(M),\qquad \ker\mathcal N=(TU-1)D_M. By Proposition 2.13.3, this map respects the algebra actions and sends (T−f)DM(T-f)D_M onto (Z−f)Er(M)(Z-f)E_r(M). Since multiplication by ZZ is a bounded automorphism of Er(M)E_r(M) commuting with Lf=1−fZ−1L_f=1-fZ^{-1}, (Z−f)Er(M)=LfEr(M).(Z-f)E_r(M)=L_fE_r(M). Consequently the preimage of LfEr(M)L_fE_r(M) is exactly (T−f)DM+(TU−1)DM(T-f)D_M+(TU-1)D_M. The third identification follows by taking the second quotient. Its norm is unchanged, because successive quotient infima equal the direct quotient infimum, as in Lemma 2.18.1. Theorem 2.17.2 identifies that same direct quotient with M0M_0 and its stated canonical norm.

On coefficients, the maps to M0M_0 send TjmT^jm to ZjmZ^jm and UjmU^jm to Z−jmZ^{-j}m. Constant elements mm map to the same constant class. These identities hold first for polynomial families and then for all series by continuity. They verify every arrow of the displayed sequence, including the sign of the difference map. Corollary 2.16.5 now proves strict exactness and all its stated norm estimates.

Theorem 2.18.2 identifies A+⊗^AA−A_+\widehat\otimes_A A_- isometrically with A{r−1T,rU}/(T−f,1−fU).A\{r^{-1}T,rU\}/(T-f,1-fU). The ideals (T−f,1−fU)(T-f,1-fU) and (T−f,TU−1)(T-f,TU-1) are equal, because 1−fU=U(T−f)−(TU−1),TU−1=U(T−f)−(1−fU).1-fU=U(T-f)-(TU-1),\qquad TU-1=U(T-f)-(1-fU). They are quotients of the same normed coefficient algebra by the same actual ideal, so their quotient norms also agree. This gives the last algebra isometry with the asserted maps. Finally, uniform comparison constants for two norms on MM pass to their coefficient spaces, projective tensor seminorms and quotients. This proves the assertion about another canonical norm. The proof includes f=0f=0, nonunits, and zero quotient algebras; it uses no localization flatness. ◻

Proposition 2.18.4. NAG-001440 (Finite diagrams of elementary constraints). Fix a finite labelled list of data (fi,ri,ϵi)(f_i,r_i,\epsilon_i), where fi∈Af_i\in A, ri∈∣K×∣r_i\in|K^\times| and ϵi∈{+,−}\epsilon_i\in\{+,-\}. For each sublist SS, give XiX_i radius Ri=riR_i=r_i if ϵi=+\epsilon_i=+ and radius Ri=ri−1R_i=r_i^{-1} if ϵi=−\epsilon_i=-, and put hi={Xi−fiϵi=+,1−fiXiϵi=−,A(S)=A{(Ri−1Xi)i∈S}/(hi:i∈S),A(∅)=A,M(S)=A(S)⊗AM,\begin{gathered} h_i=\begin{cases}X_i-f_i&\epsilon_i=+,\\1-f_iX_i&\epsilon_i=-,\end{cases} \qquad A(S)=A\{(R_i^{-1}X_i)_{i\in S}\}/(h_i:i\in S),\\ A(\varnothing)=A,\qquad M(S)=A(S)\otimes_A M, \end{gathered} where MM and all finite-module norms are chosen as in the preceding theorem. These presentations have the following properties. For S⊆TS\subseteq T, the variable inclusions give contractive maps A(S)→A(T)A(S)\to A(T) and M(S)→M(T)M(S)\to M(T), compatible with composition and with the augmentation from A,MA,M. Successive imposition of the listed constraints gives the same quotient algebra and the same quotient norm as simultaneous imposition. For any S,TS,T, A(S)⊗^AA(T) ≃ A(S∪T)A(S)\widehat\otimes_A A(T)\ \simeq\ A(S\cup T) is isometric, with the evident maps, where a constraint appearing in both lists is imposed only once on the right. Permuting constraints or repeating an identical constraint gives canonical isometric identifications. All these maps commute. For S⊆TS\subseteq T, the corresponding finite-module identification is A(T)⊗A(S)M(S) ≃ A(T)⊗^A(S)M(S) ≃ M(T),A(T)\otimes_{A(S)}M(S) \ \simeq\ A(T)\widehat\otimes_{A(S)}M(S) \ \simeq\ M(T), isometrically. The assertions remain valid when one of the quotient algebras is zero. The algebra A(S∪T)A(S\cup T) is a simultaneous constraint algebra here; no assertion that it represents an analytic-domain intersection is included.

Proof. Each variable has powers bounded by its displayed radius to the corresponding power. The coefficient inclusions therefore extend contractively by Proposition 1.9.3, kill the indicated relations, and pass to quotient norms. The same coefficient inclusions for MM are contractive and descend through Theorem 2.17.2. Equality on coefficients and variables proves compatibility with composition; polynomial density extends this equality to the algebras and coefficient modules.

Lemma 2.18.1 identifies the algebra obtained by adjoining the variables in T∖ST\setminus S over A(S)A(S) and imposing their relations with A(T)A(T), including equality of the quotient norms. Applying the lemma repeatedly proves the assertion about any order of successive imposition. A permutation of variables preserves the weights term by term and gives the stated isometry.

To remove an identical repeated constraint, first retain separate variables X,X′X,X' with the same radius. The map sending both to XX is contractive; the map in the other direction keeps the first variable. In the repeated quotient the two variables are equal. For a positive constraint, both equal ff. For a negative constraint, fX=fX′=1fX=fX'=1, so X=XfX′=X′X=XfX'=X'. Thus the maps are inverse on every coefficient and variable, and hence everywhere by continuity. They are inverse contractions, so are isometries. Theorem 2.18.2, followed by these repetitions and permutations, now identifies the completed tensor product with A(S∪T)A(S\cup T).

For the module statement, ordinary tensor reparenthesization is supplied by [StacksAlg] (Tag 00D2, Lemma 10.12.7), with (A,A(S))(A,A(S))-bimodule A(S)A(S); we verify the required norm identification here. Choose a fixed finite presentation As→DAm→M→0A^s\xrightarrow{D}A^m\to M\to 0. By Theorem 2.17.2, M(S)M(S) has the quotient presentation A(S)s→DA(S)m→M(S)→0A(S)^s\xrightarrow{D}A(S)^m\to M(S)\to 0, with its displayed quotient norm. Apply that theorem again to the presentation of A(T)A(T) over A(S)A(S) supplied by Lemma 2.18.1. Both resulting modules have the quotient presentation A(T)s→DA(T)mA(T)^s\xrightarrow{D}A(T)^m and the same norm from A(T)mA(T)^m. This gives the asserted isometries, with a⊗(b⊗m)a\otimes(b\otimes m) corresponding to ab⊗mab\otimes m. All identities of maps are checked on these generating tensors. If A(S)=0A(S)=0, its further relation quotients and all unital modules are zero, so these assertions hold directly. ◻

The last proposition concerns a fixed finite list with all fif_i in the original algebra AA. An arbitrary tree in which new functions are chosen on different intermediate quotients still requires its own mixed-presentation comparison. These results supply the algebra and norm identifications needed to reapply the elementary Laurent sequence after a verified strict quotient construction. They do not assert exactness of an arbitrary localization functor, admissibility of a geometric covering, or higher Čech acyclicity. Such assertions require their own hypotheses and proofs. In particular, no tensoring of the displayed exact Laurent sequence by an unproved flat algebra has been used.

2.19 Bounded factorization and points of strict rational quotients

The quotient construction records relations and bounds on powers. These data first give a universal property for bounded maps. The description of points is a separate consequence, proved below directly from polynomial approximation. We retain the zero-quotient convention of Section 2.17; its spectrum is empty, since a multiplicative seminorm must take zero to zero and the unit to one.

Proposition 2.19.1. NAG-001442 (The bounded universal property). Let AA be a nonzero strict norm-admissible KK-affinoid algebra, let ρ=(ρ1,…,ρs)∈∣K×∣s\rho=(\rho_1,\ldots,\rho_s)\in|K^\times|^s, and put D=A{ρ−1X}D=A\{\rho^{-1}X\}. For an ideal J=(h1,…,ht)⊂DJ=(h_1,\ldots,h_t)\subset D, give B=D/JB=D/J its quotient norm and write xix_i for the image of XiX_i. Let CC be any Banach KK-algebra, with the zero case allowed, and let ϕ:A→C\phi:A\to C be a bounded unital KK-algebra homomorphism. Bounded AA-algebra maps ψ:B→C\psi:B\to C are in bijection with tuples b=(b1,…,bs)∈Csb=(b_1,\ldots,b_s)\in C^s satisfying both conditions:

  1. For each ii there is Ci≥1C_i\geq 1 such that ∥bin∥≤Ciρin\|b_i^n\|\leq C_i\rho_i^n for every integer n≥0n\geq 0.

  2. The convergent evaluations hj(b)h_j(b), with coefficients mapped by ϕ\phi, vanish for 1≤j≤t1\leq j\leq t.

The correspondence sends ψ\psi to (ψ(xi))i(\psi(x_i))_i. If ∥ϕ(a)∥≤Cϕ∥a∥\|\phi(a)\|\leq C_\phi\|a\|, the map obtained from such a tuple satisfies ∥ψ(v)∥≤Cϕ∏i=1sCi ∥v∥(v∈B).\|\psi(v)\|\leq C_\phi\prod_{i=1}^s C_i\,\|v\| \qquad(v\in B).

Proof. Lemma 2.17.1 gives a closed actual ideal JJ and a complete quotient BB. If ψ\psi is bounded by CC, then ∥xin∥≤ρin\|x_i^n\|\leq\rho_i^n gives ∥ψ(xi)n∥≤Cρin\|\psi(x_i)^n\|\leq C\rho_i^n; enlarge CC to at least one. Composing with D→BD\to B shows that every defining relation evaluates to zero.

Conversely, Proposition 1.9.3 extends the tuple to a bounded homomorphism Φ:D→C\Phi:D\to C with bound L=Cϕ∏iCiL=C_\phi\prod_iC_i. The second condition kills the actual ideal JJ, not just its chosen generators. Thus Φ\Phi induces an algebraic map ψ:B→C\psi:B\to C. For every lift FF of vv and every j∈Jj\in J, ∥ψ(v)∥=∥Φ(F+j)∥≤L∥F+j∥.\|\psi(v)\|=\|\Phi(F+j)\|\leq L\|F+j\|. Taking the infimum proves the stated quotient-norm bound; attainment of that infimum is unnecessary. The images of polynomials are dense in BB. Two bounded maps with the same values on AA and the xix_i therefore agree. All constructions include the zero target; a map from the zero source to a nonzero unital target cannot occur, because it would identify zero and one. ◻

Corollary 2.19.2. NAG-001443 (Power-controlled rational factorization). Keep A,ρ,C,ϕA,\rho,C,\phi as above. Let g,f1,…,fsg,f_1,\ldots,f_s generate the unit ideal of AA, and put B=A{ρ−1X}/(gXi−fi)i=1s.B=A\{\rho^{-1}X\}/(gX_i-f_i)_{i=1}^s. There is a bounded factorization A→B→CA\to B\to C if and only if ϕ(g)\phi(g) is a unit and, for each ii, the ratios bi=ϕ(fi)ϕ(g)−1b_i=\phi(f_i)\phi(g)^{-1} satisfy ∥bin∥≤Ciρin\|b_i^n\|\leq C_i\rho_i^n for all n≥0n\geq 0 and some CiC_i. Such a factorization is unique.

Proof. Choose a0g+∑iaifi=1a_0g+\sum_i a_if_i=1. In BB the equality becomes g(a0+∑iaixi)=1.g\left(a_0+\sum_i a_ix_i\right)=1. Thus every factorization makes ϕ(g)\phi(g) invertible and forces ψ(xi)=ϕ(fi)ϕ(g)−1\psi(x_i)=\phi(f_i)\phi(g)^{-1}. Proposition 2.19.1 supplies the necessary power bounds. Conversely, the stated ratios satisfy ϕ(g)bi=ϕ(fi)\phi(g)b_i=\phi(f_i), so that proposition constructs the unique factorization. In the zero ring, the unit equation is interpreted with 1=01=0; the assertion then concerns the unique map to zero. ◻

A point supplies multiplicative bounds on every power. This lets us identify the spectrum of this quotient without invoking a spectral-radius theorem or compactness. The following proof uses the already written elementary seminorm calculation in Lemma 1.6.1; its contractivity and valued-fraction-field assertions have been assessed separately.

Theorem 2.19.3. NAG-001444 (Points of a strict rational quotient). With A,g,fi,ρ,BA,g,f_i,\rho,B as in Corollary 2.19.2, the map induced by A→BA\to B is a homeomorphism M(B)→∼V={x∈M(A):∣fi∣x≤ρi∣g∣x for 1≤i≤s},\mathcal M(B)\xrightarrow{\sim} V=\{x\in\mathcal M(A): |f_i|_x\leq\rho_i|g|_x \text{ for }1\leq i\leq s\}, where both spectra carry the evaluation topology and VV the subspace topology. The subset VV is closed. This includes B=0B=0 and an empty spectrum; it does not assert that an algebra with empty spectrum must be zero.

Proof. Step 1: construct the point over xx. The unit-ideal identity in the preceding proof shows that ∣g∣x>0|g|_x>0 for every x∈Vx\in V. By Lemma 1.6.1, xx defines an absolute value on the field Fx=Frac⁡(A/ker⁡x)F_x=\operatorname{Frac}(A/\ker x), with ∣a∣x≤∥a∥|a|_x\leq\|a\| for a∈Aa\in A. Evaluate polynomials at Xi=fi/gX_i=f_i/g in FxF_x. The resulting multiplicative seminorm αx\alpha_x on A[X]A[X] satisfies αx(∑νaνXν)≤max⁡ν∣aν∣x∏i∣fi/g∣xνi≤max⁡ν∥aν∥ρν.\alpha_x\left(\sum_\nu a_\nu X^\nu\right) \leq\max_\nu |a_\nu|_x\prod_i|f_i/g|_x^{\nu_i} \leq\max_\nu\|a_\nu\|\rho^\nu. For polynomials P,QP,Q, the ultrametric inequality also gives ∣αx(P)−αx(Q)∣≤αx(P−Q)≤∥P−Q∥.|\alpha_x(P)-\alpha_x(Q)|\leq\alpha_x(P-Q)\leq\|P-Q\|. Given F∈D=A{ρ−1X}F\in D=A\{\rho^{-1}X\}, choose polynomial truncations Pn→FP_n\to F. The displayed estimate makes αx(Pn)\alpha_x(P_n) a Cauchy sequence of real numbers and makes its limit independent of the truncations. Define αx(F)\alpha_x(F) to be this limit. Polynomial density follows from Lemma 2.13.1. Addition, multiplication and the scalar rule pass to the limit: for multiplication, polynomial approximations are norm-bounded and their products converge by submultiplicativity. Thus the extension is a multiplicative seminorm with αx(F)≤∥F∥,∣αx(F)−αx(G)∣≤∥F−G∥.\alpha_x(F)\leq\|F\|,\qquad |\alpha_x(F)-\alpha_x(G)|\leq\|F-G\|. It kills every gXi−figX_i-f_i and therefore the actual ideal JJ. If F−G∈JF-G\in J, the ultrametric inequality gives αx(F)=αx(G)\alpha_x(F)=\alpha_x(G). The induced seminorm on BB is bounded by its quotient norm, by taking the infimum over lifts. It extends the fixed absolute value on KK and defines a point yxy_x over xx. In particular, when B=0B=0, this construction rules out any x∈Vx\in V. It uses no completion of FxF_x.

Step 2: image and uniqueness. A bounded multiplicative seminorm on the nonzero strict affinoid algebra BB is contractive by Lemma 1.6.1. Hence ∣xi∣y≤ρi|x_i|_y\leq\rho_i, and the relations show that its restriction belongs to VV. For a polynomial PP of total degree at most NN, write QP=∑νaνfνgN−∣ν∣∈A,gNP(x1,…,xs)=QPin B.Q_P=\sum_\nu a_\nu f^\nu g^{N-|\nu|}\in A, \qquad g^N P(x_1,\ldots,x_s)=Q_P\quad\text{in }B. For every point yy over xx, this gives ∣P∣y=∣QP∣x/∣g∣xN|P|_y=|Q_P|_x/|g|_x^N. Thus its values on polynomial images are fixed. These images are dense in BB, and contractivity gives the same continuity estimate used in Step 1. Therefore y=yxy=y_x. Restriction is continuous because each evaluation on AA is an evaluation on BB.

Step 3: continuity of the inverse. For each polynomial PP, the function x↦∣QP∣x/∣g∣xNx\mapsto|Q_P|_x/|g|_x^N is continuous on VV. Fix v∈Bv\in B and any lift F∈DF\in D. Polynomial truncations satisfy, uniformly for x∈Vx\in V, ∣∣v∣yx−∣Pn∣yx∣≤∥F−Pn∥⟶0.\bigl||v|_{y_x}-|P_n|_{y_x}\bigr|\leq\|F-P_n\|\longrightarrow0. It follows that x↦∣v∣yxx\mapsto|v|_{y_x} is continuous. Explicitly, given a point of VV and ε>0\varepsilon>0, choose a truncation with error less than ε/3\varepsilon/3 and use continuity of its evaluation to bound the difference of its values by ε/3\varepsilon/3 on a neighborhood. The two approximation errors then bound the difference for vv by ε\varepsilon. Since the evaluation topology is the weakest topology making all these functions continuous, x↦yxx\mapsto y_x is continuous. Finally, the finitely many inequalities defining VV are closed conditions on continuous real-valued functions. ◻

The bounded substitution criterion becomes a pointwise universal property in Theorem 2.31.2: Theorem 1.20.1 supplies the required unit, and Theorem 1.20.2 supplies the scaled power bounds at the prescribed radius. The same point theorem identifies an empty affinoid spectrum with the zero algebra.

References. For the classical special affinoid quotients, compare [Bosch14] (§3.3, Proposition 11, pp. 50–53); for the general affinoid formulation, see [TemIntro] (3.2.2.1) with the correction recorded in Example 1.9.5. The arguments here establish only the stated bounded and pointwise interfaces and do not invoke those domain theorems as proof inputs.

2.20 Flatness of elementary strict localizations

The preceding tensor comparisons identify the relation quotients, but they do not by themselves show that tensor products preserve injections. For an elementary localization, the missing information is available: its relation operator is injective on the coefficient extension of every finite module. This permits a direct use of the snake lemma, followed by the ideal criterion for flatness. We use both algebraic results from the Stacks Project and prove the required coefficient and norm comparisons here.

Here an algebraic RR-module NN is flat if the ordinary tensor functor −⊗RN-\otimes_R N preserves exact sequences; a ring map R→SR\to S is flat when SS is flat as an RR-module. These are the definitions in [StacksAlg] (Tag 00HB, Definition 10.39.1(1),(3)). No completion or norm is part of this definition.

Lemma 2.20.1. NAG-001446 (One-sided relation operators). Let AA be a nonzero strict norm-admissible KK-affinoid algebra, let MM be a finite AA-module with a canonical Banach norm, and let f∈Af\in A and r>0r>0. On the one-sided coefficient spaces E+(M)=M{r−1T},E−(M)=M{rU}E_+(M)=M\{r^{-1}T\},\qquad E_-(M)=M\{rU\} the bounded AA-linear operators Q+=T−f:E+(M)→E+(M),Q−=1−fU:E−(M)→E−(M)Q_+=T-f:E_+(M)\to E_+(M),\qquad Q_-=1-fU:E_-(M)\to E_-(M) are injective. No value-group restriction on rr is needed for this injectivity statement.

Proof. First give MM a quotient norm from a finite free presentation, so that its AA-action is contractive. Embed both spaces isometrically into the bilateral coefficient module Er(M)E_r(M) of Section 2.16, by sending TT to ZZ and UU to Z−1Z^{-1}. Their images consist of the nonnegative and the nonpositive powers of ZZ, respectively. Multiplication by ZZ is a bounded automorphism of Er(M)E_r(M); its norm scales by rr and its inverse scales by r−1r^{-1}. Under these embeddings the two operators become Z−f=Z(1−fZ−1),1−fZ−1.Z-f=Z(1-fZ^{-1}),\qquad 1-fZ^{-1}. Theorem 2.16.2 makes 1−fZ−11-fZ^{-1} injective for every r>0r>0, so both restrictions are injective. Their boundedness follows from the shift bounds and the bounded action of ff. Finally, replacing the quotient norm on MM by a uniformly equivalent canonical norm leaves each coefficient space unchanged as a set and gives a bounded identity map with bounded inverse. It therefore preserves boundedness and injectivity of the same algebraic operators. ◻

Theorem 2.20.2. NAG-001447 (Elementary flatness and strict finite-module exactness). Let AA be as above, fix f∈Af\in A and r∈∣K×∣r\in|K^\times|, and put B+=A{r−1T}/(T−f),B−=A{rU}/(1−fU),B_+=A\{r^{-1}T\}/(T-f),\qquad B_-=A\{rU\}/(1-fU), with their quotient norms and the explicit zero-quotient convention. For either sign ϵ∈{+,−}\epsilon\in\{+,-\} the ring map A→BϵA\to B_\epsilon is flat. For every short exact sequence of finite AA-modules with canonical norms, 0⟶M′⟶M⟶M′′⟶0,0\longrightarrow M'\longrightarrow M\longrightarrow M''\longrightarrow0, its ordinary tensor sequence and its separated completed tensor sequence with BϵB_\epsilon are exact and norm-strict. The resulting modules have their canonical finite-BϵB_\epsilon-module norms, equivalently the corresponding projective norms. Flatness here concerns ordinary algebraic tensor products on all AA-modules; the Banach assertion is restricted to finite modules with these norms.

Proof. Step 1: exactness before imposing the relation. Choose quotient norms from finite free presentations of M′,M,M′′M',M,M''. All maps remain bounded and norm-strict by Theorem 1.11.5. Corollary 2.14.5, with the one-sided multiplicative weights rnr^n or r−nr^{-n}, gives an exact coefficient sequence for either sign. Multiplication by the chosen relation commutes with its arrows, since the original maps are AA-linear. We therefore have the following diagram of underlying AA-modules, with exact rows: 0⟶Eϵ(M′)⟶Eϵ(M)⟶Eϵ(M′′)⟶0↓Qϵ↓Qϵ↓Qϵ0⟶Eϵ(M′)⟶Eϵ(M)⟶Eϵ(M′′)⟶0.\begin{array}{ccccccccc} 0&\longrightarrow&E_\epsilon(M')&\longrightarrow&E_\epsilon(M) &\longrightarrow&E_\epsilon(M'')&\longrightarrow&0\\ &&\downarrow Q_\epsilon&&\downarrow Q_\epsilon &&\downarrow Q_\epsilon\\ 0&\longrightarrow&E_\epsilon(M')&\longrightarrow&E_\epsilon(M) &\longrightarrow&E_\epsilon(M'')&\longrightarrow&0 . \end{array}

Step 2: pass to the actual cokernels. The right vertical map is injective by Lemma 2.20.1. The snake lemma, [StacksAlg] (Tag 010H, Lemma 12.5.17(2)), applied in the abelian category of algebraic AA-modules, now gives the exact sequence 0⟶Eϵ(M′)/QϵEϵ(M′)⟶Eϵ(M)/QϵEϵ(M)⟶Eϵ(M′′)/QϵEϵ(M′′)⟶0.\begin{split} 0\longrightarrow& E_\epsilon(M')/Q_\epsilon E_\epsilon(M') \longrightarrow E_\epsilon(M)/Q_\epsilon E_\epsilon(M)\\ &\longrightarrow E_\epsilon(M'')/Q_\epsilon E_\epsilon(M'') \longrightarrow0 . \end{split} The zero at the left comes from ker⁡(Qϵ∣Eϵ(M′′))=0\ker(Q_\epsilon|_{E_\epsilon(M'')})=0 in the snake sequence; the rightmost surjectivity comes from the surjectivity in the bottom row. Theorem 2.17.2 identifies each actual cokernel naturally with Bϵ⊗AB_\epsilon\otimes_A the corresponding finite module, and with its separated completed tensor. That theorem also proves that the displayed relation images are closed and that the quotient norms are canonical. This is where the condition r∈∣K×∣r\in|K^\times| is used.

The induced arrows are BϵB_\epsilon-linear by the same naturality. If Bϵ≠0B_\epsilon\ne 0, it is strict by Lemma 2.17.1, so Theorem 1.11.5 over BϵB_\epsilon makes these arrows bounded with closed images and norm-strict. If Bϵ=0B_\epsilon=0, all terms are zero. Uniformly equivalent choices of the original canonical norms preserve these conclusions. We have used the snake lemma only on algebraic modules, not on a category of Banach spaces.

Step 3: test algebraic flatness. Let I⊂AI\subset A be a finitely generated ideal. Both II and A/IA/I are finite AA-modules, with the canonical norms supplied by Theorem 1.11.5. Apply the exactness just proved to 0⟶I⟶A⟶A/I⟶0.0\longrightarrow I\longrightarrow A\longrightarrow A/I\longrightarrow0. It gives injectivity of I⊗ABϵ→A⊗ABϵ=BϵI\otimes_A B_\epsilon\to A\otimes_A B_\epsilon=B_\epsilon, with the natural multiplication map. The finitely generated ideal criterion [StacksAlg] (Tag 00HD, Lemma 10.39.5(4)) proves that BϵB_\epsilon is a flat AA-module. This criterion upgrades the verified finite-module statement to algebraic flatness on all modules; it adds no assertion about completed tensors of arbitrary Banach modules. ◻

Corollary 2.20.3. NAG-001448 (Finite compositions of elementary localizations). Start with a nonzero strict norm-admissible C0=AC_0=A. For 1≤j≤n1\leq j\leq n, choose fj∈Cj−1f_j\in C_{j-1} and rj∈∣K×∣r_j\in|K^\times|, and let CjC_j be either Cj−1{rj−1Tj}/(Tj−fj)orCj−1{rjUj}/(1−fjUj),C_{j-1}\{r_j^{-1}T_j\}/(T_j-f_j) \quad\text{or}\quad C_{j-1}\{r_jU_j\}/(1-f_jU_j), with the quotient norm; once a zero algebra occurs, every subsequent algebra is zero. Every map in the sequence and every composite is algebraically flat. For finite modules over any nonzero intermediate algebra, extension to a later algebra preserves short exact sequences, and the ordinary and separated completed tensor sequences agree by bounded isomorphisms and are norm-strict for canonical norms. In particular, the maps A(S)→A(T)A(S)\to A(T) for S⊆TS\subseteq T in Proposition 2.18.4 are flat, as are the maps from AA to A+,A−,A0A_+,A_-,A_0 in Theorem 2.18.3.

Proof. At each nonzero stage Lemma 2.17.1 shows that the next algebra is a strict quotient or zero, and Theorem 2.20.2 proves flatness of the next map. The zero module is flat because its ordinary tensor functor is zero. Flatness of every finite composite now follows directly from [StacksAlg] (Tag 00HC, Lemma 10.39.4).

Every resulting map is bounded, being a composite of contractive coefficient inclusions followed by quotient maps. For finite modules, algebraic flatness gives exactness, Proposition 1.11.6 identifies ordinary and completed tensors by bounded isomorphisms, and Theorem 1.11.5 over the final nonzero algebra gives norm-strictness. The zero target is immediate. This proves the assertion for any pair of intermediate stages.

For A(S)→A(T)A(S)\to A(T), impose the constraints in T∖ST\setminus S one at a time and use Proposition 2.18.4 to identify the result with the simultaneous presentation. Finally, after forming A+A_+, impose the negative relation 1−fˉU1-\bar fU with UU of radius r−1r^{-1}. Lemma 2.18.1 and the ideal equality in the proof of Theorem 2.18.3 identify this two-stage quotient isometrically with A{r−1T,rU}/(T−f,1−fU)=A{r−1T,rU}/(T−f,TU−1)=A0.A\{r^{-1}T,rU\}/(T-f,1-fU) =A\{r^{-1}T,rU\}/(T-f,TU-1)=A_0 . This proves flatness of the stated Laurent intersection map. ◻

The finite-composition assertion permits a new element fjf_j of the current algebra at each step. The geometric rational universal property is proved in Theorem 2.31.2; Proposition 2.38.2 gives flatness for arbitrary affinoid-domain restrictions. The full covering comparison and higher Čech exactness are established in Theorems 2.39.4 and 2.38.1.

2.21 Finite Laurent coverings and strict Čech exactness

The elementary Laurent sequence controls one binary splitting. To pass to all degrees for a finite family, one must also compare the mixed intersections. We carry out that comparison on a finite diagram of Banach modules. The homological comparison is taken from the Stacks Project; the norm estimates are proved by finite successive lifting. In particular, no exactness of a completed tensor functor is inferred from algebraic flatness.

Definition 2.21.1. (Finite intersection coefficient diagrams). Let R\mathscr R be a finite partially ordered set with a greatest element XX and a meet U∩VU\cap V for every pair of elements. A Banach coefficient diagram NN on R\mathscr R assigns a non-Archimedean Banach KK-space N(U)N(U) to each UU and a bounded restriction N(U)→N(V)N(U)\to N(V) whenever V≤UV\leq U, with identity restrictions and compatibility with composition. Additional linear actions may be retained throughout. For an ordered finite family U=(Ui)i∈I\mathcal U=(U_i)_{i\in I} and V∈RV\in\mathscr R, put Cp(U,N;V)=∏i0<⋯<ipN(V∩Ui0∩⋯∩Uip),p≥0,C^p(\mathcal U,N;V) =\prod_{i_0<\cdots<i_p}N(V\cap U_{i_0}\cap\cdots\cap U_{i_p}), \qquad p\geq0, with the maximum norm; an empty product of modules is zero. The differential is the alternating sum of restrictions, and the augmentation is N(V)→C0(U,N;V)N(V)\to C^0(\mathcal U,N;V). We say that this augmented complex is strictly acyclic if it is exact, the augmentation has a bounded inverse on its image, and each positive-degree cocycle has a primitive with norm bounded by a fixed multiple of its norm. The multiplier may depend on the diagram and the degree. Since the ordered complex is finite, one multiplier may be chosen for all degrees. The same definition applies to a family rooted at UU, using N(U∩V)N(U\cap V) for its augmentation.

For comparison with the usual Čech convention, the complex using all tuples has degree pp term ∏Ip+1N(V∩Ui0∩⋯∩Uip)\prod_{I^{p+1}}N(V\cap U_{i_0}\cap\cdots\cap U_{i_p}), including repeated indices. Repetitions and permutations refer to the same meet, so their coefficient identifications are the identity.

These definitions do not assert that a member of R\mathscr R is an analytic domain. They also do not require a sheaf. The formal Čech results for presheaves apply to the intersection terms as follows. Give R\mathscr R the topology of downward closed subsets and write OU={V:V≤U}O_U=\{V:V\leq U\}. Then OU∩OV=OU∩VO_U\cap O_V=O_{U\cap V}. The prescription F(W)=lim←⁡OU⊆WN(U),F(∅)=0.\mathcal F(W)=\varprojlim_{O_U\subseteq W}N(U),\qquad \mathcal F(\varnothing)=0. Compatibility in this finite limit defines a presheaf. On OUO_U its sections are naturally N(U)N(U): the UU-component determines every other component by restriction. These identifications are bounded with bounded inverses, and are isometric when restrictions are contractive. The family (OUi)(O_{U_i}) is an open covering of its union, with exactly the required intersection terms. Its augmentation from N(V)N(V) is the restriction from OVO_V and is preserved by the comparison maps below. Thus the formal comparisons require no unproved geometric localization statement.

Lemma 2.21.2. NAG-001460 (Quantitative comparison of a finite double complex). Let Dp,qD^{p,q} be a double complex of Banach KK-spaces in a finite rectangle 0≤p≤a0\leq p\leq a, 0≤q≤b0\leq q\leq b, with bounded commuting differentials dh,dvd_h,d_v. Use d=dh+(−1)pdvd=d_h+(-1)^p d_v on its total complex and maximum norms on finite sums. Let L∙L^\bullet be a complex of Banach KK-spaces with bounded differentials, concentrated in degrees 0,…,b0,\ldots,b, and let αq:Lq→D0,q\alpha^q:L^q\to D^{0,q} be a bounded compatible horizontal augmentation. Assume that every augmented horizontal row 0⟶Lq→αqD0,q→dhD1,q⟶⋯⟶Da,q⟶00\longrightarrow L^q\xrightarrow{\alpha^q}D^{0,q} \xrightarrow{d_h}D^{1,q}\longrightarrow\cdots \longrightarrow D^{a,q}\longrightarrow0 is strictly exact. Then α:L∙→Tot⁡(D)\alpha:L^\bullet\to\operatorname{Tot}(D) is a quasi-isomorphism of underlying algebraic complexes. Let PP be a Banach KK-space and let ϵ:P→L0\epsilon:P\to L^0 be a bounded augmentation with dLϵ=0d_L\epsilon=0, placing PP in degree −1-1. Its augmented complex is strictly acyclic if and only if P→Tot⁡(D)P\to\operatorname{Tot}(D) is strictly acyclic. All lifting and inverse bounds for either complex can be obtained from finitely many bounds for the rows, the differentials, α\alpha, and the other complex. If all these maps are contractive and every horizontal row has a contracting homotopy of norm at most one, a strict acyclicity bound Q≥1Q\geq 1 for P→L∙P\to L^\bullet also works for P→Tot⁡(D)P\to\operatorname{Tot}(D). The corresponding assertions with the two axes interchanged hold as well.

Proof. The algebraic assertion is the version with interchanged variables of [StacksAlg] (Tag 0133, Lemma 12.25.4), in the abelian category of KK-vector spaces, or of the retained algebraic modules. Its finite-diagonal hypothesis holds because the rectangle is finite. We prove the additional norm statement directly. Choose B,C≥1B,C\geq 1 so that BB bounds the differentials and α\alpha, and every horizontal cycle of positive horizontal degree has a preimage bounded by CC times its norm, while ∥u∥≤C∥αu∥\|u\|\leq C\|\alpha u\|. Such a common CC exists because there are finitely many strictly exact rows and terms. Lifts need not be linear or chosen compatibly.

Let z∈Tot⁡k(D)z\in\operatorname{Tot}^k(D) be a cocycle. Starting with the largest p>0p>0 for which its (p,k−p)(p,k-p) component is nonzero, that component is horizontally closed: the (p+1,k−p)(p+1,k-p) component of dz=0dz=0 says so, because the component one column farther to the right is zero. Lift it horizontally to y∈Dp−1,k−py\in D^{p-1,k-p} and replace zz by z−dyz-dy. This removes that component and introduces a possible new component only one column to the left. After at most aa repetitions, only z0,kz_{0,k} remains. It equals αu\alpha u for a unique u∈Lku\in L^k, and dLu=0d_Lu=0 follows from dvαu=0d_v\alpha u=0 and injectivity of α\alpha. Thus z=αu+dy,dLu=0.z=\alpha u+dy,\qquad d_Lu=0. At each elimination the current norm increases by at most BCBC. Consequently ∥u∥\|u\| and the norm of the sum of the chosen lifts are bounded by C(BC)a∥z∥C(BC)^a\|z\|. This bound is independent of kk. If u=dLvu=d_Lv, then y+αvy+\alpha v is a primitive of zz. In degree zero the same construction, followed by the inverse bound for the augmentation P→L0P\to L^0, controls the unique preimage in PP. Injectivity of that augmentation is also preserved. This proves one implication.

For the converse, let u∈Lku\in L^k be a positive-degree cocycle and suppose αu=dw\alpha u=dw. Apply the same elimination to ww. Although ww need not be a cocycle, every component of dwdw of positive horizontal degree is zero, which is precisely the condition used in each elimination. The result is w−dy=αvw-dy=\alpha v with ∥v∥≤C(BC)a∥w∥\|v\|\leq C(BC)^a\|w\|. Therefore αu=dαv=αdLv\alpha u=d\alpha v=\alpha d_Lv, and injectivity gives u=dLvu=d_Lv. A bounded primitive for αu\alpha u thus yields one for uu. In degree zero, a preimage of αu\alpha u in PP maps to uu by injectivity of α\alpha, with its norm bounded by BB times the corresponding total-complex bound. These estimates also prove the converse injectivity statement. For example, multiplying the bound for the other complex by (BC)a+2(BC)^{a+2} is sufficient in either direction.

If the row contractions and all maps have norm at most one, use the row contraction to choose each horizontal lift and its degree-zero preimage. No elimination increases the norm. Thus ∥y∥,∥u∥≤∥z∥\|y\|,\|u\|\leq\|z\|, and the asserted unchanged bound follows. Interchanging the two axes changes the total-complex convention by the isometric sign map xp,q↦(−1)pqxp,qx^{p,q}\mapsto(-1)^{pq}x^{p,q}; hence the same argument proves the symmetric version. Finally, exactness identifies each differential image with the kernel of the next bounded map. The images are therefore closed, and the primitive bounds are exactly the needed bounds for the inverse maps from image norms to quotient norms. This is norm-strictness, without an appeal to an open mapping theorem. ◻

Theorem 2.21.3. NAG-001461 (Finite binary splitting trees). Let NN be a Banach coefficient diagram as above. Take a finite rooted binary tree whose nodes are elements U∈RU\in\mathscr R, with root XX and children U+,U−≤UU_+,U_-\leq U. Assume that for every internal node UU and every V∈RV\in\mathscr R the sequence 0⟶N(U∩V)⟶N(U+∩V)⊕N(U−∩V)→(a,b)↦b∣−a∣N(U+∩U−∩V)⟶00\longrightarrow N(U\cap V) \longrightarrow N(U_+\cap V)\oplus N(U_-\cap V) \xrightarrow{(a,b)\mapsto b|-a|} N(U_+\cap U_-\cap V)\longrightarrow0 is strictly exact. Then the augmented ordered Čech complex of the leaves, restricted to any V∈RV\in\mathscr R, is strictly acyclic in every degree. Its bounds depend only on finitely many restriction bounds and the assumed binary inverse and lifting bounds. The assertion is also valid for the complex using all tuples. It concerns the specified coefficient diagram: verifying the displayed hypothesis on all node intersections is part of applying the theorem.

Proof. We prove the ordered assertion simultaneously for all VV, by induction on the number of internal nodes. With no internal node the leaf is XX, and the complex is the identity N(V)→N(V)N(V)\to N(V). For the induction step, let P=(Pi)\mathcal P=(P_i) and Q=(Qj)\mathcal Q=(Q_j) be the leaves of the two subtrees rooted at U+U_+ and U−U_-. Order all PiP_i before all QjQ_j. The induction hypothesis applies to either subtree after restriction by any finite intersection in R\mathscr R.

Write W=V∩U+∩U−W=V\cap U_+\cap U_-. The mixed terms are organized by the finite double complex Dp,q=∏i0<⋯<ip∏j0<⋯<jqN(V∩Pi0∩⋯∩Pip∩Qj0∩⋯∩Qjq).D^{p,q}= \prod_{i_0<\cdots<i_p}\prod_{j_0<\cdots<j_q} N(V\cap P_{i_0}\cap\cdots\cap P_{i_p} \cap Q_{j_0}\cap\cdots\cap Q_{j_q}). For a fixed jj-tuple, its horizontal augmentation is the Čech complex of the left subtree restricted to V∩Qj0∩⋯∩QjqV\cap Q_{j_0}\cap\cdots\cap Q_{j_q}. By induction this row is strictly exact, with augmentation term N(U+∩V∩Qj0∩⋯∩Qjq)N(U_+\cap V\cap Q_{j_0}\cap\cdots\cap Q_{j_q}). These augmentation terms form L∙=C∙(Q,N;W)L^\bullet=C^\bullet(\mathcal Q,N;W), which is strictly acyclic after augmentation by N(W)N(W), again by induction. Lemma 2.21.2 therefore proves that N(W)→Tot⁡(D)N(W)\to\operatorname{Tot}(D) is strictly acyclic. The constants are finite because there are finitely many rows and terms.

To see exactly how this total complex occurs, let C∙C^\bullet be the ordered complex of all leaves, without its augmentation, and let K∙K^\bullet be the kernel of restriction to the two pure subcomplexes. A mixed tuple has p+1p+1 left indices followed by q+1q+1 right indices and lies in degree p+q+1p+q+1. The identification Tot⁡(D)[−1] → ∼  K∙,ap,q⟼(−1)pap,q\operatorname{Tot}(D)[-1]\ \xrightarrow{\ \sim\ }\ K^\bullet, \qquad a^{p,q}\longmapsto(-1)^p a^{p,q} is isometric and respects the differential. Indeed, deleting a left index gives dhd_h, while deleting a right index contributes (−1)p+1dv(-1)^{p+1}d_v in the leaf complex. Conjugation by (−1)p(-1)^p gives −dh−(−1)pdv-d_h-(-1)^pd_v, which is the differential of the shift [−1][-1]. This also fixes the sign of the degree-one obstruction.

Let z∈Ckz\in C^k be a cocycle with k≥1k\geq 1. Its two pure restrictions have bounded primitives by induction. Extend these primitives by zero on mixed tuples and subtract their differential. The remaining cocycle z′z' belongs to KkK^k, and its norm is bounded by a fixed multiple of ∥z∥\|z\|. If k≥2k\geq 2, strict acyclicity of Tot⁡(D)\operatorname{Tot}(D) supplies a bounded primitive in Kk−1K^{k-1}. If k=1k=1, exactness at degree zero of the augmented total complex gives a unique w∈N(W)w\in N(W) whose restrictions are the mixed components of z′z', with a bound for ∥w∥\|w\|. The assumed binary surjection at the root gives a∈N(U+∩V)a\in N(U_+\cap V) and b∈N(U−∩V)b\in N(U_-\cap V) with b∣−a∣=wb|-a|=w and a bounded maximum norm. Restrict a,ba,b to the leaves of their respective subtrees. Their differential is zero on pure tuples and equals z′z' on mixed pairs. This supplies the required primitive also in degree one.

In degree zero, a compatible family on the leaves gives unique elements a∈N(U+∩V)a\in N(U_+\cap V) and b∈N(U−∩V)b\in N(U_-\cap V) by induction. Their difference in N(W)N(W) restricts to zero on every mixed pair. The augmentation N(W)→Tot⁡0(D)N(W)\to\operatorname{Tot}^0(D) is injective, so b∣−a∣=0b|-a|=0. Binary exactness gives a unique element of N(V)N(V) inducing the family, with its norm bounded by the binary inverse bound and the two subtree bounds. If an element of N(V)N(V) restricts to zero on all leaves, the same injectivity argument makes it zero. This proves strict acyclicity of the ordered complex.

Finally, [StacksAlg] (Tag 01FM, Lemma 20.23.6) identifies the ordered (equivalently alternating) complex and the complex using all tuples by homotopy equivalences. The finite-poset presheaf above verifies its scope. The sorting and repeated-index homotopies in equations (20.23.6.1)–(20.23.6.2) use only finite sums with coefficients ±1\pm 1 and the identity identifications of repeated intersections. Hence they are bounded in each degree, with norm at most one for the maximum norms here. The projection to increasing tuples and alternating extension are also contractive. If f,gf,g are the resulting comparison maps and gf−1=dh+hdgf-1=dh+hd, a positive-degree cocycle zz with fz=dyfz=dy has primitive gy−hzgy-hz. This proves the required norm bounds for the full complex. The maps preserve the augmentation and the degree-zero cycles, so exactness and the inverse bound at the augmentation are preserved as well. ◻

Theorem 2.21.4. NAG-001462 (All-sign elementary Laurent Čech exactness). Let KK be complete for its fixed real-valued non-Archimedean absolute value, allowing the trivial valuation. Let AA be a nonzero strict norm-admissible KK-affinoid algebra with its normalized submultiplicative norm, and let MM be a finite AA-module with the quotient norm from a chosen surjection Am→MA^m\to M. Fix n≥0n\geq 0, f1,…,fn∈Af_1,\ldots,f_n\in A, and r1,…,rn∈∣K×∣r_1,\ldots,r_n\in|K^\times|. For a set SS of signed constraints (i,+),(i,−)(i,+),(i,-), use the simultaneous presentation A(S)=A{(ri−1Ti)(i,+)∈S,(riUi)(i,−)∈S}(Ti−fi:(i,+)∈S,  1−fiUi:(i,−)∈S),M(S)=A(S)⊗AM,A(S)= \frac{A\{(r_i^{-1}T_i)_{(i,+)\in S},(r_iU_i)_{(i,-)\in S}\}} {(T_i-f_i:(i,+)\in S,\;1-f_iU_i:(i,-)\in S)}, \qquad M(S)=A(S)\otimes_A M, with the exact quotient norms, including the zero-algebra convention. For σ∈{+,−}n\sigma\in\{+,-\}^n put Sσ={(i,σi):1≤i≤n}S_\sigma=\{(i,\sigma_i):1\leq i\leq n\}. Then 0⟶M⟶∏σ0M(Sσ0)⟶∏σ0<σ1M(Sσ0∪Sσ1)⟶⋯⟶00\longrightarrow M\longrightarrow \prod_{\sigma_0}M(S_{\sigma_0})\longrightarrow \prod_{\sigma_0<\sigma_1}M(S_{\sigma_0}\cup S_{\sigma_1}) \longrightarrow\cdots\longrightarrow0 is strictly exact in every degree, with the usual alternating restriction differential and maximum norms. The same is true when all tuples, including repetitions, are used. For each fixed datum there is Q≥1Q\geq 1 such that the augmentation has inverse bound QQ and every positive-degree cocycle zz has a primitive of norm at most Q∥z∥Q\|z\|. The same conclusions hold after imposing any fixed finite collection of elementary constraints first, and replacing MM by its finite module over that quotient. Uniformly equivalent canonical norms preserve strict exactness.

Proof. Take R\mathscr R to be the signed-constraint subsets, ordered by reverse inclusion, with meet given by union, and put N(S)=M(S)N(S)=M(S). Proposition 2.18.4 supplies its compatible contractive restrictions, all repeated-constraint identifications, and the exact norm identifications for successive constructions. Lemma 2.17.1 and Theorem 2.17.2 show that each nonzero A(S)A(S) is strict and that M(S)M(S) is a finite canonical Banach module. A zero algebra has the zero module.

Split successively according to f1,…,fnf_1,\ldots,f_n, creating both signs at every node. Consider any such binary splitting after any additional constraint set SS has already been imposed. If A(S)=0A(S)=0, its entire binary sequence is zero. Otherwise apply Theorem 2.18.3 over the strict algebra A(S)A(S), to the finite module M(S)M(S) and the image of the relevant fif_i, with radius rir_i. The simultaneous and iterated presentations, and every arrow of the resulting sequence, agree isometrically by Proposition 2.18.4. If that constraint already occurs in SS, repetition is removed by the same proposition. Changing the sign of the last arrow gives the difference convention of Theorem 2.21.3. Thus the latter theorem applies to every required intersection. Its leaves are exactly the 2n2^n sets SσS_\sigma, and it proves the displayed strict exactness. This argument reapplies the proved binary theorem over each quotient; it does not tensor an exact sequence through an unverified completed functor.

A useful bound can be read from the proof. Split off fnf_n first and write Q+,Q−,Q0≥1Q_+,Q_-,Q_0\geq 1 for bounds for the remaining n−1n-1 functions over A+,A−,A0A_+,A_-,A_0. In the mixed double complex the two index sets are the same old all-sign set. Fixing a nonempty second tuple JJ, choose j∈Jj\in J. The first cover restricted to that intersection has its jjth member equal to the whole intersection. The insertion homotopy of [StacksAlg] (Tag 0G6T, Lemma 20.23.7, second proof), with its sign chosen so that dh+hd=1dh+hd=1, has norm at most one. In alternating notation it sends a cochain to its component with jj prepended; repeated indices give zero. The coefficient meet is unchanged because jj already belongs to JJ. The horizontal augmentation is isometric since one of its coordinates is the identity. Thus Lemma 2.21.2 gives the bound Q0Q_0 for the mixed total complex.

Let E≥1E\geq 1 be the inverse bound for the binary augmentation M→M+⊕M−M\to M_+\oplus M_-. Its final map is a quotient isometry by Theorem 2.18.3, so every element of M0M_0 has a lift of norm at most twice its norm (zero is lifted by zero). The induction proof then permits Qn=max⁡{E, 2Q0}max⁡{Q+,Q−},Qempty list=1.Q_n=\max\{E,\,2Q_0\}\max\{Q_+,Q_-\},\qquad Q_{\text{empty list}}=1. This is a bound for a fixed list and module, not a bound uniform over all data. The insertion homotopies and the ordered-to-all-tuples homotopies do not increase it. Finally, starting with a further fixed constraint quotient is precisely the same argument with that quotient as base. Uniform norm comparisons pass to the finitely presented modules and finite products, proving the assertion for other canonical norms. ◻

The comparison with two coverings is classical; see [BGR84] (§8.1.4, Lemma 1 and Corollaries 3–4, pp. 325–327). We use that double-complex arrangement, giving the coefficient maps and finite norm estimates explicitly. The cited BGR results are historical sources here; the proof inputs are the stated diagram hypotheses and the proved comparison lemma.

Proposition 2.21.5. NAG-001463 (Strict acyclic refinement and common refinement). Let NN be a finite intersection coefficient diagram, and let U=(Ui)i∈I\mathcal U=(U_i)_{i\in I} and V=(Vj)j∈J\mathcal V=(V_j)_{j\in J} be nonempty finite families. Suppose that for every nonempty tuple of U\mathcal U-indices, the augmented ordered complex of V\mathcal V restricted to Ui0∩⋯∩UipU_{i_0}\cap\cdots\cap U_{i_p} is strictly acyclic. No coefficient term is discarded on the basis of point-set emptiness.

  1. If V\mathcal V refines U\mathcal U, meaning that a map t:J→It:J\to I satisfies Vj≤Ut(j)V_j\leq U_{t(j)}, then the augmented complexes of U\mathcal U and V\mathcal V are strictly acyclic simultaneously. The usual refinement map is a quasi-isomorphism, compatible with augmentation; its cohomology map is independent of tt.

  2. Without a refinement assumption, put W=(Ui∩Vj)(i,j)∈I×J\mathcal W=(U_i\cap V_j)_{(i,j)\in I\times J}. Then the augmented complexes of U\mathcal U and W\mathcal W are strictly acyclic simultaneously. The hypotheses on every U\mathcal U-intersection remain necessary in this assertion.

All conclusions are valid for all-tuples complexes as well, with bounded comparison maps and degreewise bounded homotopies. Only finitely many of the stipulated local inverse and lifting bounds are needed.

Proof. For part (1), form the ordered double complex Dp,q=∏i0<⋯<ip∏j0<⋯<jqN(Ui0∩⋯∩Uip∩Vj0∩⋯∩Vjq).D^{p,q}=\prod_{i_0<\cdots<i_p}\prod_{j_0<\cdots<j_q} N(U_{i_0}\cap\cdots\cap U_{i_p}\cap V_{j_0}\cap\cdots\cap V_{j_q}). The assumed vertical augmented rows compare C∙(U,N)C^\bullet(\mathcal U,N) with Tot⁡(D)\operatorname{Tot}(D) by Lemma 2.21.2. In the other direction, fix a nonempty jj-tuple and choose any one of its indices j0j_0. Since Vj0∩⋯∩Vjq≤Ut(j0)V_{j_0}\cap\cdots\cap V_{j_q}\leq U_{t(j_0)}, the restricted U\mathcal U-family has a member equal to the entire intersection. Its augmented complex has the insertion contraction of [StacksAlg] (Tag 0G6T, Lemma 20.23.7). With the maximum norms, insertion has norm at most one; no local sheaf theorem is used. The same comparison lemma now compares C∙(V,N)C^\bullet(\mathcal V,N) with Tot⁡(D)\operatorname{Tot}(D). The two applications, preserving the common augmentation N(X)N(X), prove strict acyclicity in both directions.

We record the actual maps. In alternating notation the refinement map is (Ttc)j0…jk=ct(j0)…t(jk)∣Vj0∩⋯∩Vjk.(T_tc)_{j_0\ldots j_k} =c_{t(j_0)\ldots t(j_k)}\big|_{V_{j_0}\cap\cdots\cap V_{j_k}}. Repeated indices on the right give zero. Let aa and bb be the augmentations of the two complexes into the total double complex, placing them in bidegrees (k,0)(k,0) and (0,k)(0,k) respectively. The bounded homotopy (Hc)i0…ip;j0…jqp,q=(−1)pci0…ip t(j0)…t(jq)∣,p+q=k−1,(Hc)^{p,q}_{i_0\ldots i_p;j_0\ldots j_q} =(-1)^p c_{i_0\ldots i_p\,t(j_0)\ldots t(j_q)}\big|, \qquad p+q=k-1, satisfies dH+Hd=bTt−adH+Hd=bT_t-a. All restrictions are to the indicated mixed intersection. To check the sign, the terms obtained by deleting an ii-index cancel between dHdH and HdHd in positive horizontal degree; the terms obtained by deleting a t(j)t(j)-index cancel in positive vertical degree. The boundary terms are bTtbT_t at p=0p=0 and −a-a at q=0q=0. Thus the quasi-isomorphism supplied by the double complex is induced by the usual refinement map. The prism formula in [StacksAlg] (Tag 01FP, Section 20.25, paragraph “Refinements”, following (20.25.0.2)) gives a homotopy for two choices t,t′t,t'. It consists of finitely many signed restrictions and is bounded in each degree. In the finite-poset realization, one may restrict the first open family to the union of the refining family before applying that formula; pulling it back gives precisely the displayed coefficient maps. This verifies the presheaf scope without identifying N(X)N(X) with sections of an unproved sheaf.

For part (2), W\mathcal W refines U\mathcal U. Fix an intersection Y=Ui0∩⋯∩UipY=U_{i_0}\cap\cdots\cap U_{i_p} and choose i0i_0 among its indices. On YY, the family W∣Y\mathcal W|_Y refines V∣Y\mathcal V|_Y by (i,j)↦j(i,j)\mapsto j, while V∣Y\mathcal V|_Y refines W∣Y\mathcal W|_Y by j↦(i0,j)j\mapsto(i_0,j) because Y∩Vj=Y∩Ui0∩VjY\cap V_j=Y\cap U_{i_0}\cap V_j. The composites are homotopic to the identity by the same prism formula. These bounded homotopies preserve augmentation. For clarity, if f,gf,g are such maps with gf−1=dh+hdgf-1=dh+hd and fz=dyfz=dy, then gy−hzgy-hz is a primitive of zz; its norm is controlled by the norms of g,hg,h and the given primitive. The analogous degree-zero assertion follows from preservation of augmentation and injectivity. Hence strict acyclicity of V∣Y\mathcal V|_Y implies that of W∣Y\mathcal W|_Y. Apply part (1) to W→U\mathcal W\to\mathcal U. Finally apply Tag 01FM as in Theorem 2.21.3 to obtain the all-tuples versions. ◻

The finite rational-cover argument uses these comparisons after identifying every mixed intersection with its actual coefficient algebra. The required refinements lead to Theorem 2.22.9; Theorem 2.32.2 then treats arbitrary radii by simultaneous strictification.

2.22 Rational presentations and Laurent refinements

We work over a complete non-Archimedean field KK, allowing trivial valuation except in the small-scalar lemmas where nontriviality is stated. Every nonzero algebra has a strict norm-admissible Banach norm, and every relation quotient has its actual quotient norm. The zero algebra is admitted separately with its zero norm. All radii in this section are one; a radius in ∣K×∣|K^\times| can first be reduced to this convention by scaling the corresponding numerator. These conventions concern the displayed presentations, not a spectral norm.

For a tuple g,f1,…,fsg,f_1,\ldots,f_s generating the unit ideal of AA, write A⟨f/g⟩=A{T1,…,Ts}/(gTj−fj)1≤j≤s.A\langle f/g\rangle =A\{T_1,\ldots,T_s\}/(gT_j-f_j)_{1\leq j\leq s}. For a unit-ideal tuple f0,…,frf_0,\ldots,f_r, its standard rational family consists of the r+1r+1 presentations Ri=A{Tj:j≠i}/(fiTj−fj)j≠i,0≤i≤r.R_i=A\{T_j:j\ne i\}/(f_iT_j-f_j)_{j\ne i}, \qquad 0\leq i\leq r. We put Ti=1T_i=1 when using a formula involving all indices. The joint presentation of a finite family means the quotient with disjoint variable blocks and all their relations. Theorem 2.18.2 identifies it with the separated completed tensor product. Its module attached to a finite AA-module MM is its ordinary tensor with MM, equipped with the quotient norm from a fixed finite free presentation of MM. Theorem 2.17.2 gives its comparison with the completed tensor. These choices specify the arrows and norms before a geometric sheaf is available.

Lemma 2.22.1. NAG-001465 (A rational quotient as an elementary chain). Let B=A⟨f/g⟩B=A\langle f/g\rangle and choose a0g+∑j=1sajfj=1.a_0g+\sum_{j=1}^s a_jf_j=1. Choose a∈K×a\in K^\times with ∣a∣max⁡0≤j≤s∥aj∥≤1|a|\max_{0\leq j\leq s}\|a_j\|\leq 1. Such a choice is always possible for nontrivial valuation. For trivial valuation use a defining quotient norm on AA and take a=1a=1; this replaces any given admissible norm by a uniformly equivalent one. The following isometry uses the norm for which the stated scalar inequality holds. Put C0=A{U}/(gU−a),hj=fjUˉ/a∈C0,C1=C0{V1,…,Vs}/(Vj−hj)j.\begin{aligned} C_0&=A\{U\}/(gU-a),\\ h_j&=f_j\bar U/a\in C_0,\\ C_1&=C_0\{V_1,\ldots,V_s\}/(V_j-h_j)_j. \end{aligned} There is an isometric AA-algebra isomorphism B≃C1B\simeq C_1 sending TjT_j to VjV_j. Consequently A→BA\to B is algebraically flat. It preserves short exact sequences of finite canonical AA-modules under ordinary or separated completed tensor product; the resulting sequences are norm-strict. The assertion about completed tensor products is restricted to these finite modules.

Proof. For nontrivial valuation, powers of a scalar of absolute value less than one give the choice of aa. For trivial valuation, the strict Tate algebra is a polynomial ring with norm one on each nonzero polynomial, so every element of its quotient has norm at most one. Thus a=1a=1 works with the stipulated defining norm. In BB set v=a0+∑jajTˉjv=a_0+\sum_j a_j\bar T_j. The defining relations give gv=1gv=1, and ∥av∥≤1\|av\|\leq 1. In C0C_0 the equation gUˉ=ag\bar U=a makes gg a unit with inverse Uˉ/a\bar U/a. Hence in C1C_1 one has gVˉj=fjg\bar V_j=f_j, with ∥Vˉj∥≤1\|\bar V_j\|\leq 1. The bounded universal property, Proposition 2.19.1, therefore supplies a contractive map B→C1B\to C_1.

Conversely, send UU to avav and VjV_j to Tˉj\bar T_j in BB. These elements have norm at most one. The first relation vanishes because gav=agav=a, and the remaining relations vanish because fjv=Tˉjf_jv=\bar T_j. Thus the same universal property and Lemma 2.18.1 give a contractive map C1→BC_1\to B. The composites fix AA and all the Tj,VjT_j,V_j. They also fix UU, since gg is a unit in C1C_1 and both possible images equal ag−1ag^{-1}. Polynomial density proves that the maps are inverse, and two inverse contractions are isometric. This argument also covers zero quotients; every displayed map is a map of actual relation quotients.

Multiplication of gU−agU-a by the nonzero scalar −a−1-a^{-1} identifies C0C_0 with the negative elementary localization for g/ag/a at radius one. Adjoin the VjV_j one at a time; each next map is a positive elementary localization for the image of hjh_j. Corollary 2.20.3 gives flatness of the composite and the finite-module assertions, which transfer across the isometry just proved. Uniform equivalence of the original admissible norms preserves these assertions, including in the trivial case. No general relation quotient is asserted to be flat. ◻

Lemma 2.22.2. NAG-001466 (A redundant bounded rational constraint). Let B=A⟨f/g⟩B=A\langle f/g\rangle and let CC be a strict relation quotient over AA. Suppose gg is a unit in CC and the elements bj=fjg−1∈Cb_j=f_jg^{-1}\in C satisfy ∥bj∥≤1\|b_j\|\leq 1. Then the canonical map C⟶C{T1,…,Ts}/(gTj−fj)j≃C⊗^ABC\longrightarrow C\{T_1,\ldots,T_s\}/(gT_j-f_j)_j \simeq C\widehat\otimes_A B is an isometry. More generally, if each bjb_j is power bounded, with a fixed bound on ∥bjn∥\|b_j^n\| for all n≥0n\geq 0, the same map is a bounded isomorphism with bounded inverse. The isomorphisms are compatible with subsequent bounded maps and with the associated finite-module quotients.

Proof. The image of gg is a unit before adjoining the variables, so the actual ideal generated by gTj−fjgT_j-f_j equals the ideal generated by Tj−bjT_j-b_j. Substitution at bjb_j gives an inverse to the coefficient inclusion by Proposition 2.19.1. In the first case both maps are contractive; in the second their bounds are finite. The two composites fix the coefficients and variables, hence are identities by polynomial density. The completed-tensor comparison is Theorem 2.18.2. The same maps on a fixed finite free presentation of a module descend to its relation quotient by Theorem 2.17.2; they retain the stated norm bounds. The formulae show compatibility. The zero case is immediate. ◻

Lemma 2.22.3. NAG-001467 (A Laurent refinement of a unit-generated standard family). Suppose f0,…,frf_0,\ldots,f_r are units of AA. Form the all-sign Laurent family for the fixed elements qij=fi/fjq_{ij}=f_i/f_j, 0≤i<j≤r0\leq i<j\leq r, at radius one. Thus each branch DϵD_\epsilon imposes, for each pair, either Xij=qijX_{ij}=q_{ij} or qijXij=1q_{ij}X_{ij}=1, with ∥Xij∥≤1\|X_{ij}\|\leq 1. For every branch there is an index k=k(ϵ)k=k(\epsilon) for which all fi/fkf_i/f_k have norm at most one in DϵD_\epsilon. In particular, Rk→DϵR_k\to D_\epsilon is a contractive map and Dϵ⊗^ARk≃DϵD_\epsilon\widehat\otimes_A R_k\simeq D_\epsilon isometrically. The same conclusion holds after imposing any additional elementary constraints on that branch.

Proof. The assertion for the zero branch has its stated degenerate meaning. Otherwise, direct an edge i→ji\to j when the chosen relation gives ∥fi/fj∥≤1\|f_i/f_j\|\leq 1, and direct j→ij\to i for the other sign. This defines one directed edge between each pair of vertices. There is a vertex reachable from every vertex by a directed path. To prove this, begin with one vertex kk. When adding a new vertex vv, keep kk if the edge is v→kv\to k, and replace kk by vv if the edge is k→vk\to v. The previously constructed paths followed by this last edge establish the assertion inductively.

For a path i=i0→i1→⋯→im=ki=i_0\to i_1\to\cdots\to i_m=k, multiply its bounded ratios. Their product is fi/fkf_i/f_k, so submultiplicativity gives ∥fi/fk∥≤1\|f_i/f_k\|\leq 1. The bounded universal property gives the map from RkR_k, and Lemma 2.22.2 gives the isometry. Subsequent quotient maps are contractive, preserve the unit equations, and preserve the inequalities just established. Thus the conclusion persists on every further joint presentation. This is a norm calculation; it does not infer a power bound from a pointwise inequality. ◻

Lemma 2.22.4. NAG-001468 (Separating the small members of a standard family). Assume the valuation of KK is nontrivial. Let f0,…,frf_0,\ldots,f_r generate the unit ideal and fix ∑iaifi=1\sum_i a_if_i=1. Choose κ∈K×\kappa\in K^\times with θ=∣κ∣max⁡i∥ai∥<1.\theta=|\kappa|\max_i\|a_i\|<1. Form the all-sign Laurent family for fi/κf_i/\kappa. For a branch CC, let PP be the positive indices and NN the negative indices, and write its relations as fp=κXp(p∈P),fnXn=κ(n∈N),∥Xi∥≤1.f_p=\kappa X_p\quad(p\in P),\qquad f_nX_n=\kappa\quad(n\in N),\qquad \|X_i\|\leq1. If NN is empty, CC is the actual zero quotient. If i∈Pi\in P, the joint quotient C⊗^ARiC\widehat\otimes_A R_i is zero. For i∈Ni\in N, the element fif_i is a unit in CC, and there is an isometric identification C⊗^ARi≃C{Tj:j∈N∖{i}}/(Tj−fj/fi)j.C\widehat\otimes_A R_i \simeq C\{T_j:j\in N\setminus\{i\}\}/(T_j-f_j/f_i)_j. These identifications commute with all intersection maps. They persist after further elementary constraints on CC. Consequently the restricted standard rational family is, after removing actual zero terms, the standard family generated by the units (fi)i∈N(f_i)_{i\in N}.

Proof. If every index is positive, the relation ideal contains 1−κ∑iaiXi=∑iai(fi−κXi).1-\kappa\sum_i a_iX_i =\sum_i a_i(f_i-\kappa X_i). The subtracted element has norm at most θ<1\theta<1 in the coefficient algebra. Hence this displayed element is a unit by Lemma 1.1.4. The relation ideal is the whole coefficient algebra. This proves the actual zero assertion without using an empty-spectrum implication.

For i∈Pi\in P, work in the coefficient algebra for the joint presentation with RiR_i, and again put Ti=1T_i=1. Its relation ideal contains 1−κXi∑jajTj=∑jaj(fj−fiTj)+(fi−κXi)∑jajTj.1-\kappa X_i\sum_j a_jT_j =\sum_j a_j(f_j-f_iT_j) +(f_i-\kappa X_i)\sum_j a_jT_j. The norm of κXi∑jajTj\kappa X_i\sum_j a_jT_j is at most θ\theta. The same geometric inverse proves that this joint ideal is the whole algebra. In particular, every further intersection involving such a term is an actual zero quotient.

If i∈Ni\in N, then fi−1=Xi/κf_i^{-1}=X_i/\kappa in CC. Multiplying the rational relations by this unit replaces them by Tj−fj/fiT_j-f_j/f_i without changing the actual ideal. For p∈Pp\in P one has fp/fi=XpXi,∥XpXi∥≤1.f_p/f_i=X_pX_i,\qquad \|X_pX_i\|\leq1. Thus all variables indexed by PP can be eliminated by the two contractive substitution maps of Lemma 2.22.2. What remains is precisely the displayed presentation. The maps fix coefficients and send each surviving variable to the same ratio, so they commute with adding further index blocks. Their norm inequalities and the explicit unit equations survive all later contractive quotient maps. This proves the assertions about intersections and further constraints. ◻

The preceding two lemmas construct a finite two-level Laurent refinement of a standard rational family. The first level uses functions in AA. On a first-level branch the second level uses ratios of units in that branch algebra. We do not identify this construction with a single all-sign family of functions in the original algebra. Every mixed intersection needed below is instead defined by its joint presentation, using Lemma 2.18.1. The actual-zero computations are retained on those intersections.

References. The classical reduction is due to Tate. Compare [Bosch14] (§4.3, Lemmas 5–6, pp. 85–87) and [BGR84] (§8.2.2, Lemmas 3–4, pp. 333–334). The proofs here replace the smallness choice by a specified unit-ideal certificate and produce bounded refinement maps by multiplication along directed paths. For the elementary-chain construction compare [Bosch14] (§3.3, Proposition 16, p. 56); its norm proof above uses the same unit-ideal certificate rather than an unproved maximum principle.

The strict-target domain interface

The next interface uses two analytic facts separately: Theorem 1.12.3 detects a unit ideal from points, and Theorem 1.13.4 converts a pointwise bound into a bound on all powers in a strict affinoid algebra. The earlier presentation arguments did not require the second fact.

Theorem 2.22.5. NAG-001470 (Rational factorization for strict targets). Let B=A⟨f/g⟩B=A\langle f/g\rangle, and let V={x∈M(A):∣fj(x)∣≤∣g(x)∣ for every j}.V=\{x\in\mathcal M(A):|f_j(x)|\leq|g(x)| \text{ for every }j\}. For every strict norm-admissible affinoid CC and bounded unital KK-algebra map ϕ:A→C\phi:A\to C, there is a bounded factorization A→B→CA\to B\to C if and only if the induced map M(C)→M(A)\mathcal M(C)\to\mathcal M(A) has image in VV. The factorization is unique. Zero targets and zero quotients are included. Consequently two strict rational presentations of the same subset of M(A)\mathcal M(A) have a unique bounded AA-algebra isomorphism with bounded inverse; an inclusion of such subsets gives the unique bounded restriction map in the reverse direction. These maps commute with composition. No assertion about generalized non-strict targets is made.

Proof. A bounded homomorphism pulls a bounded multiplicative seminorm back to one, so the stated map on spectra is defined. If the factorization exists, Theorem 2.19.3 gives its image condition. Conversely, assume this condition and first take C≠0C\ne 0. At a point y∈M(C)y\in\mathcal M(C), the inequalities and the image of the unit-ideal identity imply ∣ϕ(g)(y)∣>0|\phi(g)(y)|>0. Theorem 1.12.3 therefore makes ϕ(g)\phi(g) a unit of CC. Put bj=ϕ(fj)ϕ(g)−1b_j=\phi(f_j)\phi(g)^{-1}. For every yy one has ∣bj(y)∣≤1|b_j(y)|\leq 1; Theorem 1.13.4 gives a finite bound for all ∥bjn∥\|b_j^n\|, n≥0n\geq 0. Corollary 2.19.2 constructs the unique bounded factorization. If C=0C=0, the unique map has the stated properties directly. In particular, a zero BB cannot factor into a nonzero CC; the preceding construction would contradict its unit equation.

For two presentations with the same subset, apply the first assertion to each presentation as a target of the other. The composites and identity maps agree on AA, so uniqueness makes the composites identities. The same uniqueness proves composition compatibility for inclusions. The isomorphisms need not preserve the chosen presentation norms; boundedness in both directions is the conclusion. ◻

Proposition 2.22.6. NAG-001471 (Rational intersections with their actual presentations). Let B=A⟨f/g⟩B=A\langle f/g\rangle and C=A⟨e/h⟩C=A\langle e/h\rangle, with both displayed tuples generating the unit ideal. The joint quotient H=A{T1,…,Ts,S1,…,St}/(gTi−fi, hSj−ej)i,jH=A\{T_1,\ldots,T_s,S_1,\ldots,S_t\}/ (gT_i-f_i,\ hS_j-e_j)_{i,j} is isometric to the rational presentation with denominator ghgh and numerator list (hfi)i,(gej)j,(fiej)i,j.(hf_i)_i,\qquad (ge_j)_j,\qquad (f_ie_j)_{i,j}. Its point set is the intersection of the point sets for BB and CC, and H≃B⊗^ACH\simeq B\widehat\otimes_A C. Finite iteration gives the analogous statement for every finite intersection, including repeated members. The intersection and restriction maps are independent of the chosen rational presentations up to the unique bounded isomorphisms of Theorem 2.22.5. For a finite AA-module, the corresponding maps on relation-quotient modules are bounded and satisfy the same composition identities.

Proof. Multiplying unit-ideal identities for (g,fi)(g,f_i) and (h,ej)(h,e_j) shows that ghgh and the listed numerators generate the unit ideal. Let PP be their rational presentation, with variables Yi,Zj,WijY_i,Z_j,W_{ij} in the listed order. In HH, both gg and hh are units by their separate unit-ideal identities. Send Yi⟼Tˉi,Zj⟼Sˉj,Wij⟼TˉiSˉj.Y_i\longmapsto\bar T_i,\qquad Z_j\longmapsto\bar S_j,\qquad W_{ij}\longmapsto\bar T_i\bar S_j. The images have norm at most one and satisfy all relations, so this gives a contractive map P→HP\to H. In PP, the product ghgh is a unit; hence gg and hh are units separately. Dividing the defining equations gives gYi=figY_i=f_i, hZj=ejhZ_j=e_j, and Wij=YiZjW_{ij}=Y_iZ_j. Thus Ti↦YˉiT_i\mapsto\bar Y_i, Sj↦ZˉjS_j\mapsto\bar Z_j gives a contractive map H→PH\to P. The two maps are inverse on coefficients and variables, hence on the complete quotients by density. They are isometric.

Theorem 2.19.3 identifies the point set of PP. On that set ghgh is nonzero. The inequalities for hfihf_i and gejge_j can therefore be divided by ∣h∣|h| and ∣g∣|g|, giving the two original sets of inequalities. Conversely the original inequalities imply all three product inequalities. The completed-tensor assertion is Theorem 2.18.2. Repeat this construction to treat finite intersections; each step uses actual ideals and the quotient norms, by Lemma 2.18.1. A repeated member is redundant by Lemma 2.22.2.

The strict-target universal property gives the asserted representation independence and makes every composition identity a consequence of uniqueness. A bounded algebra map induces a bounded map on the finite free modules in a fixed presentation of MM, hence on their actual quotient modules. Theorem 2.17.2 identifies these with the ordinary and completed extensions. Their maps coincide on the images of the fixed generators, which proves the same compatibility for modules. ◻

From all-sign families to finite rational covers

For a finite indexed rational family U=(Bi)i∈I\mathcal U=(B_i)_{i\in I} and a finite module MM, let BJB_J be its joint presentation for a nonempty subset J⊂IJ\subset I and put MJ=BJ⊗AMM_J=B_J\otimes_A M. Order II. The ordered augmented complex is 0⟶M⟶∏i∈IM{i}⟶∏i0<i1M{i0,i1}⟶⋯ .0\longrightarrow M\longrightarrow \prod_{i\in I}M_{\{i\}}\longrightarrow \prod_{i_0<i_1}M_{\{i_0,i_1\}}\longrightarrow\cdots. (dc)i0…iq+1=∑ν=0q+1(−1)νci0…iν^…iq+1∣i0…iq+1.(dc)_{i_0\ldots i_{q+1}} =\sum_{\nu=0}^{q+1}(-1)^\nu c_{i_0\ldots\widehat{i_\nu}\ldots i_{q+1}}\big|_{i_0\ldots i_{q+1}}. Use the maximum norm on each finite product. The compatibility just proved makes d2=0d^2=0. The all-tuples convention, including repeated indices, is compared with this one by the bounded normalization argument at the end of the proof of Theorem 2.21.3. Strict exactness means that each image equals the next kernel and the quotient norm on that image is uniformly equivalent to the induced norm; all constants may depend on the finite family, MM, and the degree.

Theorem 2.22.7. NAG-001473 (Strict Čech exactness for a standard rational family). Assume that KK is nontrivially valued. For every unit-ideal tuple f0,…,fr∈Af_0,\ldots,f_r\in A and every finite canonical AA-module MM, the augmented ordered Čech complex of its standard rational family is strict exact in every degree. The same is true after replacing AA by any strict relation quotient and using the image tuple and any finite module over that quotient.

Proof. First assume that all fif_i are units. Use the all-sign Laurent family W\mathcal W on the ratios fi/fjf_i/f_j from Lemma 2.22.3. Theorem 2.21.4 gives strict acyclicity of W\mathcal W. It also gives strict acyclicity of W\mathcal W restricted to every joint rational presentation RJR_J: the images of the ratios are elements of that strict algebra, and the restricted family is their all-sign family. The module at the base of this restricted complex is the finite RJR_J-module RJ⊗AMR_J\otimes_A M. No tensoring of an exact Banach complex by an arbitrary module is involved.

On every nonempty joint presentation of members of W\mathcal W, select one of its members WW. Lemma 2.22.3 supplies k(W)k(W), and Lemma 2.22.2 identifies adjoining Rk(W)R_{k(W)} with the identity on this joint presentation, isometrically. Thus the restricted rational family contains an identity member. Inserting its index contracts its augmented Čech complex; the maps are bounded, and in the matched quotient norms have bound one. The common-refinement comparison, Lemma 2.21.2, now applies in both directions: the Laurent columns are strict exact, and the rational rows on their intersections are contracted. It identifies the two augmentations through the common double complex and proves strict exactness for (Ri)i(R_i)_i. These are the two local acyclicity inputs required for the comparison. Equivalently, one can apply Proposition 2.21.5(1) to this refinement.

For a general unit-ideal tuple, take the first-level all-sign family V\mathcal V of Lemma 2.22.4. It and its restriction to every RJR_J are strict acyclic by Theorem 2.21.4, applied afresh over that strict algebra and its finite module. On any nonempty intersection of members of V\mathcal V, choose one member and its sets P,NP,N. The explicit zero identities from Lemma 2.22.4 remain valid there. The terms indexed by PP can therefore be discarded, and every term involving one of those indices is zero. The remaining rational family is the standard family on the units (fi)i∈N(f_i)_{i\in N}, by the same lemma and its intersection compatibility; some of its members may also be zero. Its complex is strict exact by the unit case just proved. If the intersection algebra is zero, all its modules and complexes are zero. The same common-refinement comparison therefore proves the theorem for the original tuple.

Every application uses finitely many complexes in each relevant degree. The comparison theorem supplies uniform bounds for their successive lifts, which is the norm-strictness assertion; algebraic exactness alone would not supply those bounds. All auxiliary algebras are verified strict actual quotients by Lemma 2.17.1. Replacing AA by such an algebra merely repeats the argument, and equivalent canonical module norms preserve strict exactness. ◻

Proposition 2.22.8. NAG-001474 (A standard refinement of a finite rational cover). Let U=(Ui)1≤i≤n\mathcal U=(U_i)_{1\leq i\leq n} be a finite cover of M(A)\mathcal M(A), where Ui={x:∣fj(i)(x)∣≤∣f0(i)(x)∣,1≤j≤ri},(f0(i),…,fri(i))=A.U_i=\{x:|f^{(i)}_j(x)|\leq|f^{(i)}_0(x)|, 1\leq j\leq r_i\}, \qquad (f^{(i)}_0,\ldots,f^{(i)}_{r_i})=A. Write BiB_i for the associated rational presentation. Put Λ=∏i=1n{0,…,ri},Λ0={λ∈Λ:λi=0 for some i},Fλ=∏i=1nfλi(i).\Lambda=\prod_{i=1}^n\{0,\ldots,r_i\},\qquad \Lambda_0=\{\lambda\in\Lambda:\lambda_i=0\text{ for some }i\}, \qquad F_\lambda=\prod_{i=1}^n f^{(i)}_{\lambda_i}. The tuple (Fλ)λ∈Λ0(F_\lambda)_{\lambda\in\Lambda_0} generates the unit ideal. Its standard rational family (Sλ)λ∈Λ0(S_\lambda)_{\lambda\in\Lambda_0} covers M(A)\mathcal M(A). For any choice i(λ)i(\lambda) with λi(λ)=0\lambda_{i(\lambda)}=0, the point set of SλS_\lambda is contained in Ui(λ)U_{i(\lambda)}, and there is a unique bounded map Bi(λ)→SλB_{i(\lambda)}\to S_\lambda over AA. These maps give a compatible refinement of all finite intersection diagrams.

Proof. Suppose all FλF_\lambda, λ∈Λ0\lambda\in\Lambda_0, vanished at xx. Choose ii with x∈Uix\in U_i. The unit-ideal identity for row ii implies f0(i)(x)≠0f^{(i)}_0(x)\ne 0. In every other row, at least one entry is nonzero at xx, again by its unit-ideal identity. Choosing such entries and taking λi=0\lambda_i=0 gives a nonzero Fλ(x)F_\lambda(x), a contradiction. Theorem 1.12.3 now implies that these FλF_\lambda generate the unit ideal. The standard family covers because at each point one of the finitely many nonzero values ∣Fλ∣|F_\lambda| is maximal; Theorem 2.19.3 identifies its quotient point sets.

Fix λ∈Λ0\lambda\in\Lambda_0, write k=i(λ)k=i(\lambda), and let xx be a point of its standard rational set. Here Fλ(x)≠0F_\lambda(x)\ne 0. Choose jj with x∈Ujx\in U_j. If j=kj=k, the desired inclusion holds. Otherwise, for 0≤m≤rk0\leq m\leq r_k, replace the jjth slot of λ\lambda by zero and the kkth slot by mm, leaving the others unchanged. The resulting index μ\mu belongs to Λ0\Lambda_0, and ∣Fμ(x)Fλ(x)∣=∣f0(j)(x)fλj(j)(x)∣∣fm(k)(x)f0(k)(x)∣≤1.\left|\frac{F_\mu(x)}{F_\lambda(x)}\right| =\left|\frac{f^{(j)}_0(x)}{f^{(j)}_{\lambda_j}(x)}\right| \left|\frac{f^{(k)}_m(x)}{f^{(k)}_0(x)}\right|\leq1. The first factor is at least one because x∈Ujx\in U_j. Thus the second factor is at most one for every mm, proving x∈Ukx\in U_k. All divisions are legitimate: the factors of Fλ(x)F_\lambda(x) are nonzero, and f0(j)(x)f^{(j)}_0(x) is nonzero on UjU_j.

Theorem 2.22.5 now gives the bounded map Bk→SλB_k\to S_\lambda and its uniqueness. In particular, this step uses the pointwise-to-power-bound theorem: a numerator with the only zero slot replaced need not occur in Λ0\Lambda_0, so its ratio is not directly one of the norm-one variables of SλS_\lambda. We have proved its pointwise bound and then used precisely the analytic interface required to obtain a bounded map. Uniqueness and Proposition 2.22.6 give compatibility on every finite intersection. ◻

Theorem 2.22.9. NAG-001475 (Strict finite rational Čech acyclicity). Let KK be complete, allowing the trivial absolute value. Let AA be a strict norm-admissible KK-affinoid algebra and let U\mathcal U be a finite cover of M(A)\mathcal M(A) by strict rational subsets with radii in ∣K×∣|K^\times|. For every finite canonical AA-module MM, the augmented ordered Čech complex of the rational quotient modules is strict exact in all degrees. The all-tuples complex, with repeated indices, has the corresponding strict exactness. This statement concerns finite rational covers and finite modules; it asserts neither acyclicity for arbitrary affinoid covers nor exactness of completed base change on arbitrary Banach modules.

Proof. Scale the finitely many numerators to reduce the radii to one, using the isometric change of weighted variables. If A=0A=0 every term is zero. Assume first that A≠0A\ne 0 and KK is nontrivially valued. Proposition 2.22.8 provides a standard rational refinement V\mathcal V. By Theorem 2.22.7, its augmented complex is strict exact. For every finite intersection of members of U\mathcal U, its joint algebra BJB_J is strict by Proposition 2.22.6. The pullback of V\mathcal V is the standard rational family generated by the images of the same unit-ideal tuple (Fλ)(F_\lambda). Theorem 2.22.7, now over BJB_J and for the finite BJB_J-module BJ⊗AMB_J\otimes_A M, proves strict exactness on that intersection.

In the other direction, every nonempty intersection of members of V\mathcal V maps into one chosen member of U\mathcal U, because one of its factors already does. Theorem 2.22.5 and Lemma 2.22.2 identify that restricted member with the identity, by bounded inverse maps. Inserting its index gives a bounded contraction of the restricted coarse complex. Thus both local inputs of Lemma 2.21.2 are verified, or equivalently all hypotheses of Proposition 2.21.5(1). It transfers strict exactness from V\mathcal V to U\mathcal U with the stated augmentation. This is the common-refinement argument in all degrees, not merely the degree-zero gluing implication.

It remains to treat trivial valuation. Choose a defining strict quotient norm on AA. Then K⟨X1,…,Xd⟩=K[X1,…,Xd]K\langle X_1,\ldots,X_d\rangle=K[X_1,\ldots,X_d] with norm one on every nonzero polynomial: the finite-tail condition permits only finitely many nonzero coefficients. The quotient norm of every nonzero class in AA, and the quotient norm of every nonzero vector in a finite canonical module, is also one. The original admissible norms are uniformly equivalent to these discrete norms. With these choices each rational algebra is the ordinary polynomial quotient Bi=A[T1,…,Tri]/(giTj−fj(i))j≃A[1/gi],gi=f0(i).B_i=A[T_1,\ldots,T_{r_i}]/(g_iT_j-f^{(i)}_j)_j \simeq A[1/g_i],\qquad g_i=f^{(i)}_0. Indeed the unit-ideal identity gives an inverse of gig_i in the displayed quotient, and the inverse map sends TjT_j to fj(i)/gif^{(i)}_j/g_i. These maps are inverse algebraically and bounded for the discrete norms. Joint quotients and their modules are therefore A[1/∏i∈Jgi]A[1/\prod_{i\in J}g_i] and M[1/∏i∈Jgi]M[1/\prod_{i\in J}g_i], with the usual localization maps.

For every maximal ideal m⊂A\mathfrak m\subset A, Theorem 1.12.3 provides the bounded point from the finite residue field A/mA/\mathfrak m. Its extending absolute value is trivial by Lemma 1.12.1. Since this point belongs to some UiU_i, the unit-ideal identity for that row forces gi∉mg_i\notin\mathfrak m. Thus (g1,…,gn)=A(g_1,\ldots,g_n)=A. This argument uses these particular maximal-residue-field points; it does not assert that every point of M(A)\mathcal M(A) is trivially valued.

Apply [StacksAlg] (Tag 0G6K, Lemma 15.30.5(4)) to the ordinary extended alternating localization complex, with f=1∈(g1,…,gn)f=1\in(g_1,\ldots,g_n). It is exactly our ordered augmented complex, with its degree numbering shifted by one, and is exact in every degree. For discrete norms every linear map is bounded, every image is closed, and the quotient norm of a nonzero image element is one. The complex is consequently norm-strict. Uniform equivalence transfers this conclusion to the original canonical norms. Finally the bounded comparison between ordered and all-tuples complexes at the end of the proof of Theorem 2.21.3 gives the all-tuples assertion in both valuation cases. ◻

References. For the classical product refinement compare [Bosch14] (§4.3, Lemma 4, pp. 84–85) and [BGR84] (§8.2.2, Lemma 2, pp. 332–333). The passage to higher degrees is governed by the two-sided refinement comparison, cf. [BGR84] (§8.1.4, Corollaries 3–4, p. 327; §8.2.2, Proposition 5, p. 334). Bosch’s Proposition 4.3/7 proves the sheaf reduction; his discussion of the full acyclicity theorem explicitly refers to BGR rather than supplying its proof. The argument above supplies the finite-module and norm comparisons within the stated strict rational scope.

2.23 The structure sheaf on the fixed rational site

The acyclicity theorem already gives a useful sheaf construction. We specify the site so that no theorem about arbitrary affinoid domains is concealed in its definition. Let AA be a strict KK-affinoid algebra over a complete field with its fixed non-Archimedean absolute value, allowing trivial valuation, and put X=M(A)X=\mathcal M(A).

Definition 2.23.1. NAG-001477 (The fixed rational pretopology). Let RX\mathcal R_X be the partially ordered category of subsets of XX admitting a rational presentation over the original algebra AA with radii in ∣K×∣|K^\times|. It contains XX and the empty set. Arrows are inclusions. A covering family of V∈RXV\in\mathcal R_X is a finite family (Vi⊆V)(V_i\subseteq V) of objects of RX\mathcal R_X whose union is VV. The empty family covers the empty set. Choose one rational presentation algebra AVA_V for each VV, with AX=AA_X=A and A∅=0A_\varnothing=0, and use the canonical bounded restriction maps of Theorem 2.22.5. For a finite AA-module MM, write OXrat(V)=AV,M~rat(V)=AV⊗AM.\mathcal O_X^{\mathrm{rat}}(V)=A_V, \qquad \widetilde M^{\mathrm{rat}}(V)=A_V\otimes_A M.

Corollary 2.23.2. NAG-001478 (Rational structure sheaves and finite modules). The covering families just defined form a pretopology. The assignments OXrat\mathcal O_X^{\mathrm{rat}} and M~rat\widetilde M^{\mathrm{rat}} are respectively a sheaf of rings and a sheaf of modules on the resulting site, independent up to canonical isomorphism of the presentation choices. One has Γ(X,OXrat)=A,Γ(X,M~rat)=M.\Gamma(X,\mathcal O_X^{\mathrm{rat}})=A, \qquad \Gamma(X,\widetilde M^{\mathrm{rat}})=M. The functor M↦M~ratM\mapsto\widetilde M^{\mathrm{rat}} from finite AA-modules to OXrat\mathcal O_X^{\mathrm{rat}}-modules is exact and fully faithful. Each finite covering in this site has a norm-strict exact augmented Čech complex for M~rat\widetilde M^{\mathrm{rat}}, with canonical finite-module norms.

Proof. Proposition 2.22.6 represents V∩WV\cap W by a rational algebra over AA. It is the categorical fiber product in this partially ordered category. Identity coverings, pullbacks of finite set-theoretic coverings, and finite composites of such coverings satisfy the pretopology axioms, since intersection commutes with finite union. This uses only objects rational over the fixed AA. It does not require a theorem saying that every rational domain over every intermediate algebra has a rational presentation over AA.

Uniqueness in Theorem 2.22.5 proves that restriction maps compose, that identity restrictions are identities, and that two choices of a presentation give compatible bounded inverse isomorphisms. If Vi⊆VV_i\subseteq V, the joint algebra AV⊗^AAViA_V\widehat\otimes_A A_{V_i} is canonically AViA_{V_i} by the same theorem. Viewed over AVA_V, its defining tuple is the image of a unit-ideal tuple from AA, hence still generates one. Thus any finite covering (Vi)(V_i) of VV becomes a finite rational covering of M(AV)\mathcal M(A_V) in the scope of Theorem 2.22.9. Apply that theorem to the finite AVA_V-module AV⊗AMA_V\otimes_A M. The ordinary tensor associativity of [StacksAlg] (Tag 00D2, Lemma 10.12.7), together with Theorem 2.17.2, identifies all its terms and arrows with those in the asserted Čech complex.

Degree-zero exactness is precisely the equalizer condition for a sheaf. The value on the empty object is zero, as required by its empty covering. Taking M=AM=A gives the ring sheaf, and multiplication on the local tensor modules gives the module sheaves. Their global sections have the displayed values by the chosen presentation of XX.

For a short exact sequence of finite AA-modules, rational flatness from Lemma 2.22.1 proves exactness on every object when the valuation is nontrivial. If the valuation is trivial, equip AA with a defining quotient norm; all nonzero elements have norm one. In the proof of that lemma the unit certificate then has coefficient norms at most one, so one may take a=1a=1. Its same elementary chain proves rational flatness in this case as well, using Corollary 2.20.3, which allows trivial valuation. Proposition 1.11.6 gives the corresponding normed finite-module identifications. Objectwise exactness implies exactness of the sequence of sheaves.

An AA-linear map u:M→Nu:M\to N produces the sheaf map 1⊗u1\otimes u. Conversely, let ψ:M~rat→N~rat\psi:\widetilde M^{\mathrm{rat}}\to\widetilde N^{\mathrm{rat}} be a morphism of OXrat\mathcal O_X^{\mathrm{rat}}-modules. Its map on XX is an AA-linear u:M→Nu:M\to N. For every VV, compatibility with restriction gives ψV(1⊗m)=1⊗u(m)\psi_V(1\otimes m)=1\otimes u(m). Since every element of AV⊗AMA_V\otimes_A M is a finite sum of elements b⊗mb\otimes m, AVA_V-linearity forces ψV=1⊗u\psi_V=1\otimes u. Thus the map on Hom groups is a bijection. The norm-strict Čech assertion was obtained from the same application of the theorem. ◻

Theorem 2.28.12 supplies essential surjectivity on this rational site by effective descent of finite modules. The corresponding equivalence for generalized affinoid domains is Theorem 2.38.5; its extension to the full analytic G-site, together with Theorem B, is Theorem 2.39.6.

2.24 Compact spectra and finite rational coverings

The algebraic unit-ideal test uses maximal residue fields. A covering theorem, however, concerns every Berkovich point. We record the topological bridge and the exact finite-cover test that will be used below.

Proposition 2.24.1. NAG-001496 (Compactness of the bounded spectrum). Let KK be a complete field with its fixed non-Archimedean absolute value, and let AA be a commutative unital Banach KK-algebra with a submultiplicative homogeneous norm. Its bounded multiplicative spectrum M(A)\mathcal M(A), with the evaluation topology, is compact and Hausdorff; it may be empty. Every bounded unital homomorphism A→BA\to B induces a continuous map M(B)→M(A)\mathcal M(B)\to\mathcal M(A). Any subset defined by finitely many inequalities ∣fi(x)∣≤ri∣gi(x)∣|f_i(x)|\leq r_i|g_i(x)|, with ri>0r_i>0, is closed and therefore compact. No nonemptiness theorem is part of this statement.

Proof. The power argument of Lemma 1.6.1 applies verbatim to a bounded multiplicative seminorm on any such AA: ∣a∣xn≤C∥a∥n|a|_x^n\leq C\|a\|^n and taking nnth roots gives ∣a∣x≤∥a∥|a|_x\leq\|a\|. Hence evaluation embeds the spectrum in P=∏a∈A[0,∥a∥].P=\prod_{a\in A}[0,\|a\|]. Each closed real interval is compact. For completeness, if an open cover of an interval had no finite subcover, repeatedly bisect and keep a half without a finite subcover. The nested intervals have lengths tending to zero and a common point, by completeness of the real numbers. An open set of the cover containing that point contains a sufficiently small retained interval, a contradiction. By the authorized Tychonov theorem [StacksAlg] (Tag 08ZU, Theorem 5.14.4), PP is compact. It is Hausdorff because two distinct points differ in a coordinate, where disjoint real neighbourhoods separate them.

The image of the spectrum is cut out by s0=0,s1=1,sc=∣c∣ (c∈K),sab=sasb,sa+b≤max⁡(sa,sb).s_0=0,\quad s_1=1,\quad s_c=|c|\ (c\in K),\quad s_{ab}=s_as_b,\quad s_{a+b}\leq\max(s_a,s_b). All these conditions are closed, since real addition, multiplication and maximum are continuous. Conversely they give exactly a multiplicative ultrametric seminorm extending the fixed absolute value and bounded by the chosen norm. Thus M(A)\mathcal M(A) is a closed subspace of PP, with exactly its evaluation topology. A closed subspace of a compact space is compact: adjoin its open complement to an open cover and take a finite subcover of the ambient space. Hausdorffness is inherited. Composition with a bounded homomorphism preserves boundedness and turns each evaluation into another evaluation, proving continuity. Finally each stated inequality is the inverse image of a closed subset of R2\mathbb R^2 under the pair of evaluations; their finite intersection is closed. ◻

Theorem 2.24.2. NAG-001497 (Testing a finite rational cover at maximal points). Let AA be a strict KK-affinoid algebra over any complete non-Archimedean field, possibly trivially valued. A finite family of value-group rational subsets of M(A)\mathcal M(A) covers the spectrum if and only if it contains the bounded point associated to every maximal ideal of AA. The assertion includes the empty family when A=0A=0.

Proof. Scale the finitely many radii by nonzero elements of KK, so that the members have presentations Ui={x:∣fij(x)∣≤∣gi(x)∣ for all j},(gi,fi1,…,fimi)=A.U_i=\{x:|f_{ij}(x)|\leq|g_i(x)|\text{ for all }j\}, \qquad (g_i,f_{i1},\ldots,f_{im_i})=A. The forward implication is immediate. For the converse first suppose the valuation is nontrivial, and fix π∈K\pi\in K with 0<∣π∣<10<|\pi|<1. If xx lies outside every UiU_i, choose for each ii an index jij_i and put hi=fijih_i=f_{ij_i}, so ∣hi(x)∣>∣gi(x)∣|h_i(x)|>|g_i(x)|. Choose an integer Ni≥1N_i\geq 1 and a nonzero scalar cic_i with ∣gi(x)∣Ni≤∣π∣ ∣hi(x)∣Ni,∣ci∣≤∣hi(x)∣Ni.|g_i(x)|^{N_i}\leq|\pi|\,|h_i(x)|^{N_i}, \qquad |c_i|\leq|h_i(x)|^{N_i}. This is possible because the ratio is strictly less than one and ∣hi(x)∣>0|h_i(x)|>0. Let ViV_i be the rational subset with denominator hiNih_i^{N_i} and numerators π−1giNi,ci\pi^{-1}g_i^{N_i},c_i. Its defining tuple generates the unit ideal since ci≠0c_i\ne 0. It contains xx, and every y∈Viy\in V_i satisfies ∣hi(y)∣>0,∣gi(y)∣≤∣π∣1/Ni∣hi(y)∣<∣hi(y)∣.|h_i(y)|>0,\qquad |g_i(y)|\leq|\pi|^{1/N_i}|h_i(y)|<|h_i(y)|. Hence Vi∩Ui=∅V_i\cap U_i=\varnothing. The finite intersection V=⋂iViV=\bigcap_iV_i has a strict rational presentation by Proposition 2.22.6. Its presentation algebra AVA_V is nonzero: the point x∈Vx\in V lifts to its spectrum by Theorem 2.19.3. Theorem 1.12.3 supplies a bounded maximal-residue evaluation AV→LA_V\to L, with L/KL/K finite. Its restriction to AA has maximal kernel: its image is a finite-dimensional KK-domain, hence a field by [StacksAlg] (Tag 00GS(1)). Uniqueness of the extending absolute value, Lemma 1.12.1, identifies the resulting point with the maximal point of that kernel. This point belongs to VV and to none of the UiU_i, contrary to the hypothesis.

If the valuation is trivial, equip AA with a strict presentation quotient norm, which is discrete. Every element has pointwise value at most one. A unit-ideal certificate for (gi,fi1,…,fimi)(g_i,f_{i1},\ldots,f_{im_i}) consequently shows that Ui={x:∣gi(x)∣=1}.U_i=\{x:|g_i(x)|=1\}. Indeed, all the numerators have value at most one, and if the denominator and numerators all had value strictly less than one, their unit-ideal relation would contradict ∣1∣=1|1|=1. At a maximal point ∣gi∣=1|g_i|=1 is equivalent to gi∉mg_i\notin\mathfrak m, since finite extensions of a trivially valued field have the trivial extending absolute value. The assumed covering of maximal points therefore gives (g1,…,gt)=A(g_1,\ldots,g_t)=A by [StacksAlg] (Tag 00E0(2)). Evaluating a unit certificate at any xx shows that some ∣gi(x)∣=1|g_i(x)|=1, so that x∈Uix\in U_i. The empty family is dealt with by the nonemptiness criterion of Theorem 1.12.3. ◻

The trivial-valuation argument is not a density argument. For example, if KK is trivially valued and A=K[T]A=K[T] with its discrete norm, the Gauss seminorm of radius r∈(0,1)r\in(0,1) is a bounded point. Every maximal residue point gives ∣T∣=0|T|=0 or 11, whereas an open band a<∣T∣<ba<|T|<b with 0<a<r<b<10<a<r<b<1 contains that Gauss point and no maximal point. The finite rational-cover test remains valid because its inequalities have value-group radii, not arbitrary real radii.

2.25 Rational domains inside rational domains

The fixed rational site can be compared with the rational site of one of its members only after a transitivity theorem has been proved. The issue is that a function on a rational domain need not come from the original algebra. We first control small changes of its defining functions, and then replace finitely many functions by fractions with one common denominator. The resulting algebra comparison will be proved on the actual relation quotients.

Throughout, KK is a complete real-valued non-Archimedean field; trivial valuation is allowed. Nonzero affinoid algebras are strict and norm-admissible, with submultiplicative norm and unit of norm one. Quotients use their actual quotient norms. The zero algebra and its zero modules have the separate zero-norm interpretation of Section 2.17. A rational presentation has a unit-ideal tuple. We use unit radii in the lemmas; radii in ∣K×∣|K^\times| are reduced to this case by scaling variables.

Lemma 2.25.1. NAG-001499 (Stability under a small change of a rational tuple). Let EE be a nonzero strict norm-admissible affinoid algebra. Fix a0,…,at,b0,…,bt∈Ea_0,\ldots,a_t,b_0,\ldots,b_t\in E with ∑j=0tbjaj=1,H=max⁡(1,∥b0∥,…,∥bt∥).\sum_{j=0}^t b_ja_j=1,\qquad H=\max(1,\|b_0\|,\ldots,\|b_t\|). Suppose a0′,…,at′∈Ea'_0,\ldots,a'_t\in E and put δ=max⁡0≤j≤t∥aj′−aj∥<H−1.\delta=\max_{0\leq j\leq t}\|a'_j-a_j\|<H^{-1}. Then (a0′,…,at′)=E(a'_0,\ldots,a'_t)=E, with a unit-ideal certificate whose coefficient norms are at most HH. The two rational subsets defined by the denominators a0a_0 and a0′a'_0 are equal. Their actual presentation algebras Q=E{S1,…,St}/(a0Sj−aj)j,Q′=E{S1′,…,St′}/(a0′Sj′−aj′)jQ=E\{S_1,\ldots,S_t\}/(a_0S_j-a_j)_j, \qquad Q'=E\{S'_1,\ldots,S'_t\}/(a'_0S'_j-a'_j)_j have a unique bounded EE-algebra isomorphism, and this isomorphism is isometric for the displayed quotient norms. For a finite EE-module with a fixed finite free presentation, the induced comparison of its two relation-quotient modules is also isometric. These assertions include zero presentation algebras.

Proof. Write ϵj=aj′−aj\epsilon_j=a'_j-a_j and u=∑jbjaj′=1+∑jbjϵju=\sum_jb_ja'_j=1+\sum_jb_j\epsilon_j. Since ∥u−1∥≤Hδ<1\|u-1\|\leq H\delta<1, Lemma 1.1.4 gives u−1u^{-1} with ∥u−1∥≤1\|u^{-1}\|\leq 1. Thus bj′=u−1bjb'_j=u^{-1}b_j satisfies ∑jbj′aj′=1\sum_jb'_ja'_j=1 and ∥bj′∥≤H\|b'_j\|\leq H. All the ideals and quotients appearing here are actual closed ideals and complete quotients by Lemma 2.17.1.

In QQ, put sj=S‾js_j=\overline S_j. The certificate gives a0−1=b0+∑j≥1bjsja_0^{-1}=b_0+\sum_{j\geq 1}b_js_j, hence ∥a0−1∥≤H\|a_0^{-1}\|\leq H. Consequently a0′=a0(1+ϵ0a0−1)a'_0=a_0(1+\epsilon_0a_0^{-1}) is a unit and aj′a0′=(sj+ϵja0−1)(1+ϵ0a0−1)−1,∥aj′a0′∥≤1.\frac{a'_j}{a'_0} =\bigl(s_j+\epsilon_ja_0^{-1}\bigr) \bigl(1+\epsilon_0a_0^{-1}\bigr)^{-1}, \qquad \left\|\frac{a'_j}{a'_0}\right\|\leq1. Proposition 2.19.1 therefore gives a contractive map Q′→QQ'\to Q sending Sj′S'_j to the displayed ratio. In Q′Q', use the certificate (bj′)(b'_j) and the differences −ϵj-\epsilon_j. The same calculation gives a contractive map Q→Q′Q\to Q'. The composites fix EE, and each variable is forced to be the corresponding numerator divided by its unit denominator. Thus the composites fix the variables and, by polynomial density, are identities. Two inverse contractions are isometric. The same argument shows uniqueness among bounded EE-algebra maps. It also applies when a quotient is zero: unit equations are interpreted with 1=01=0, and the two constructed maps force both quotients to be zero together.

For completeness, equality of the point sets can also be checked without an algebra comparison. At x∈M(E)x\in\mathcal M(E) put m(x)=max⁡j∣aj(x)∣m(x)=\max_j|a_j(x)|. The certificate and Lemma 1.6.1 give m(x)≥H−1>δm(x)\geq H^{-1}>\delta. A coordinate attaining this maximum retains its absolute value after perturbation, and every other perturbed coordinate has value at most m(x)m(x). If ∣a0(x)∣=m(x)|a_0(x)|=m(x), the same holds for a0′a'_0; if ∣a0(x)∣<m(x)|a_0(x)|<m(x), then ∣a0′(x)∣<m(x)|a'_0(x)|<m(x). This is exactly the equivalence of the two rational conditions. Finally choose Ev→FEw→N→0E^v\xrightarrow{F}E^w\to N\to 0; such a finite presentation exists by Theorem 1.11.5. The two induced modules are Qw/FQQvQ^w/F_QQ^v and (Q′)w/FQ′(Q′)v(Q')^w/F_{Q'}(Q')^v by Theorem 2.17.2. The coordinatewise isometry carries one actual relation image to the other, so it induces an isometry of their quotient norms. ◻

The strict inequality on the error is essential. Over E=KE=K, the tuples (1,0)(1,0) and (0,1)(0,1) are at distance one and both generate the unit ideal, but their rational conditions define respectively all of M(K)\mathcal M(K) and the empty set. In particular, a pointwise bound is not being silently substituted for a strict error estimate.

Lemma 2.25.2. NAG-001500 (Approximation by fractions with one denominator). Let g,f1,…,fsg,f_1,\ldots,f_s generate the unit ideal of AA, and set B=A⟨f/g⟩B=A\langle f/g\rangle. The element gg is a unit in BB, and the image of A[g−1]A[g^{-1}] is dense in BB. More explicitly, given v0,…,vt∈Bv_0,\ldots,v_t\in B and η>0\eta>0, there exist N≥0N\geq 0 and p0,…,pt∈Ap_0,\ldots,p_t\in A such that ∥vj−g−Npj∥B<η(0≤j≤t).\|v_j-g^{-N}p_j\|_B<\eta\qquad(0\leq j\leq t). If KK is trivially valued and AA carries a defining strict quotient norm, all these approximations may be chosen to be equalities. In that case every nonzero element of AA or BB has norm one, and BB is the ordinary localization A[g−1]A[g^{-1}] with that discrete norm.

Proof. Choose c0g+∑icifi=1c_0g+\sum_i c_if_i=1. In BB, with xi=X‾ix_i=\overline X_i, one has g−1=c0+∑icixig^{-1}=c_0+\sum_i c_ix_i. Polynomial images are dense in BB by Lemma 2.13.1 and the quotient construction of Lemma 2.17.1. For a polynomial P(X)=∑αeαXαP(X)=\sum_\alpha e_\alpha X^\alpha and an integer NN at least its total degree, put pP=∑αeαfαgN−∣α∣∈A.p_P=\sum_\alpha e_\alpha f^\alpha g^{N-|\alpha|}\in A. The defining relations give gNP(x)=pPg^NP(x)=p_P in the actual quotient. Choose polynomial approximations to the finitely many vjv_j, and then choose one NN at least all their degrees. This proves the claimed simultaneous approximation. No injectivity of A→BA\to B or of A[g−1]→BA[g^{-1}]\to B has been assumed.

For trivial valuation, a strict Tate algebra over KK is a polynomial ring: a coefficient family tending to zero has finite support. Its norm is one on nonzero elements. A defining quotient norm on AA therefore has the same property. Consequently A{X}=A[X]A\{X\}=A[X], and every class of BB already has a polynomial representative. The quotient norm of a nonzero class is one. The unit equation for gg and the maps Xi↦fi/gX_i\mapsto f_i/g identify this polynomial quotient with A[g−1]A[g^{-1}]. The assertions for a zero quotient have their stated degenerate meaning. ◻

Lemma 2.25.3. NAG-001501 (Flattening a rational presentation by exact fractions). Let B=A⟨f/g⟩B=A\langle f/g\rangle and fix certificates c0g+∑i=1scifi=1in A,∑j=0tbjaj=1in B.c_0g+\sum_{i=1}^s c_if_i=1\quad\text{in }A, \qquad \sum_{j=0}^t b_ja_j=1\quad\text{in }B. Suppose aj=g−Npja_j=g^{-N}p_j in BB for one integer N≥0N\geq 0 and pj∈Ap_j\in A. Put L=max⁡(1,∥c0∥,…,∥cs∥),H=max⁡(1,∥b0∥,…,∥bt∥),L=\max(1,\|c_0\|,\ldots,\|c_s\|),\qquad H=\max(1,\|b_0\|,\ldots,\|b_t\|), and choose γ∈K×\gamma\in K^\times satisfying ∣γ∣LN+1H≤1|\gamma|L^{N+1}H\leq 1. Set d=gp0d=gp_0. Then C′=B{S1,…,St}/(a0Sj−aj)j,D=A{Y1,…,Ys,Z1,…,Zt,V}/(dYi−p0fi, dZj−gpj, dV−γ)i,j\begin{aligned} C'&=B\{S_1,\ldots,S_t\}/(a_0S_j-a_j)_j,\\ D&=A\{Y_1,\ldots,Y_s,Z_1,\ldots,Z_t,V\}/ (dY_i-p_0f_i,\ dZ_j-gp_j,\ dV-\gamma)_{i,j} \end{aligned} are canonically isometric as AA-algebras. The tuple defining DD generates the unit ideal because it contains γ\gamma; thus DD is a rational presentation over AA. The isometry identifies its variables by Yi⟼X‾i,Zj⟼S‾j,V⟼γg−(N+1)a0−1in C′.Y_i\longmapsto\overline X_i,\qquad Z_j\longmapsto\overline S_j,\qquad V\longmapsto\gamma g^{-(N+1)}a_0^{-1} \quad\text{in }C'. Zero quotients are included. No unit-ideal condition on (p0,…,pt)(p_0,\ldots,p_t) in AA is required.

Proof. All displayed ideals are actual closed ideals by Lemma 2.17.1. In C′C', the two certificates give g−1=c0+∑iciX‾i,a0−1=b0+∑j=1tbjS‾j.g^{-1}=c_0+\sum_i c_i\overline X_i, \qquad a_0^{-1}=b_0+\sum_{j=1}^t b_j\overline S_j. Their norms are at most LL and HH, respectively. Since p0=gNa0p_0=g^Na_0, the element d=gN+1a0d=g^{N+1}a_0 is a unit and ∥γd−1∥≤∣γ∣LN+1H≤1\|\gamma d^{-1}\|\leq |\gamma|L^{N+1}H\leq 1. The proposed images of Yi,Zj,VY_i,Z_j,V all have norm at most one. The relations for the first two blocks follow from gX‾i=fig\overline X_i=f_i and a0S‾j=aj=g−Npja_0\overline S_j=a_j=g^{-N}p_j; the last relation follows from the definition of γd−1\gamma d^{-1}. Proposition 2.19.1 therefore gives a contractive map D→C′D\to C'.

Conversely, in DD the equation dV‾=γd\overline V=\gamma makes dd a unit. Therefore gg and p0p_0 are units separately. Dividing dY‾i=p0fid\overline Y_i=p_0f_i by p0p_0 gives gY‾i=fig\overline Y_i=f_i. The images of the YiY_i have norm at most one, so Proposition 2.19.1 gives a contractive coefficient map B→DB\to D. Under this map the equality aj=g−Npja_j=g^{-N}p_j persists. Dividing dZ‾j=gpjd\overline Z_j=gp_j by gg and multiplying by g−Ng^{-N} gives a0Z‾j=aja_0\overline Z_j=a_j. The same proposition now gives a contractive map C′→DC'\to D.

The composites fix AA, XiX_i, SjS_j, YiY_i and ZjZ_j. They also fix VV: the relation dV=γdV=\gamma and the invertibility of dd uniquely determine this element. Polynomial density then makes the composites identities, proving the isometry. If B=0B=0, the already constructed unital map B→DB\to D forces D=0D=0, and C′=0C'=0 directly; the second universal-property step is then replaced by the unique map of zero algebras. All other zero-quotient cases are covered by the same calculations. Thus the proof never identifies an empty spectrum with a zero ring without an algebraic argument. ◻

The scalar γ\gamma serves two purposes. It gives a unit-ideal certificate in the original algebra, and its chosen size makes the additional variable contractive on the iterated quotient. The existence of such a scalar is automatic over a nontrivially valued field. Over a trivially valued field it will follow from the defining discrete norms, with L=H=1L=H=1.

Theorem 2.25.4. NAG-001502 (Transitivity of strict rational presentations). Let AA be a strict norm-admissible KK-affinoid algebra. Let BB be a rational presentation over AA, and let CC be a rational presentation over BB, with all radii in ∣K×∣|K^\times|. Then there is a rational presentation DD over AA and a unique bounded AA-algebra isomorphism D≃CD\simeq C with bounded inverse. Under the point homeomorphisms of Theorem 2.19.3, the image of M(C)\mathcal M(C) in M(A)\mathcal M(A) is precisely the rational subset represented by DD.

If the valuation of KK is nontrivial, DD can be chosen so that this isomorphism is isometric for the given actual presentation norms. For trivial valuation, the same holds after replacing the norm on AA by a defining strict quotient norm and using the induced quotient norms on the presentations; for the original norms the conclusion is a bounded isomorphism in both directions. The construction and conclusion allow zero quotients.

For every finite canonical AA-module MM, the resulting identifications D⊗AM ≃ C⊗AM ≃ C⊗B(B⊗AM)D\otimes_A M\ \simeq\ C\otimes_A M \ \simeq\ C\otimes_B(B\otimes_A M) are natural bounded isomorphisms with bounded inverses. The corresponding assertions hold with separated completed tensors. With norms from one fixed finite free presentation of MM, the module comparisons are isometric whenever the algebra comparison is isometric. This completed-tensor assertion concerns these finite modules.

Proof. By the weighted change of variables in Lemma 2.17.1, reduce every radius to one. This change is isometric. Write B=A⟨f/g⟩,C=B⟨h/k⟩,h=(h1,…,ht).B=A\langle f/g\rangle, \qquad C=B\langle h/k\rangle, \qquad h=(h_1,\ldots,h_t). If A=0A=0 or B=0B=0, then C=0C=0; the rational presentation A{V}/(0V−1)A\{V\}/(0V-1) is zero and gives the conclusion, including the module assertions. We may therefore assume A,B≠0A,B\ne 0. Choose certificates c0g+∑icifi=1,b0k+∑j=1tbjhj=1,c_0g+\sum_i c_if_i=1, \qquad b_0k+\sum_{j=1}^t b_jh_j=1, and define L=max⁡(1,∥ci∥)L=\max(1,\|c_i\|) and H=max⁡(1,∥bj∥)H=\max(1,\|b_j\|). Choose 0<η<H−10<\eta<H^{-1}. Lemma 2.25.2 supplies a common NN and p0,…,pt∈Ap_0,\ldots,p_t\in A such that a0′=g−Np0,aj′=g−Npj,∥k−a0′∥<η,∥hj−aj′∥<η.a'_0=g^{-N}p_0,\qquad a'_j=g^{-N}p_j, \qquad \|k-a'_0\|<\eta,\quad \|h_j-a'_j\|<\eta. Lemma 2.25.1 gives a unit-ideal certificate for (aj′)(a'_j) with coefficient norms at most HH, and an isometric BB-algebra identification C≃C′=B⟨a1′/a0′,…,at′/a0′⟩.C\simeq C'=B\langle a'_1/a'_0,\ldots,a'_t/a'_0\rangle. If KK is nontrivially valued, powers of a scalar of absolute value less than one give γ∈K×\gamma\in K^\times with ∣γ∣LN+1H≤1|\gamma|L^{N+1}H\leq 1. Lemma 2.25.3 then gives the required DD, with denominator gp0gp_0 and numerator list (p0fi)1≤i≤s,(gpj)1≤j≤t,γ.(p_0f_i)_{1\leq i\leq s},\qquad (gp_j)_{1\leq j\leq t},\qquad \gamma. Both comparisons are isometric, so D≃CD\simeq C is isometric. Notice that adding γ\gamma is indispensable to this construction: clearing denominators only proves a unit-ideal statement in BB, and does not establish (p0,…,pt)=A(p_0,\ldots,p_t)=A.

For trivial valuation, replace the norm on AA by a defining strict quotient norm. Uniformly equivalent coefficient norms give uniformly equivalent Tate and quotient norms, with the same underlying complete algebras: the coefficient convergence condition is unchanged, and taking infima over representatives preserves both norm comparisons. By Lemma 2.25.2, every nonzero element of AA and BB now has norm one. Hence L=H=1L=H=1, the approximations may be exact, and γ=1\gamma=1 meets the required inequality. The same two lemmas give an isometric comparison for these norms. Returning to the original uniformly equivalent norms gives the asserted bounded inverse maps.

The isomorphism is over AA. Applying Theorem 2.19.3 first to A→BA\to B and then to B→CB\to C identifies M(C)\mathcal M(C) homeomorphically with its image in M(A)\mathcal M(A). The same theorem for A→DA\to D, together with the constructed isomorphism, identifies this image with the rational subset of the displayed tuple. This proves the point assertion from the algebra comparison; no spectral-radius or maximum principle is needed in this proof. Any bounded AA-algebra endomorphism of CC restricts on BB to its unique bounded rational factorization and then has its values on the second-stage variables forced by their unit denominator. Thus it is the identity. The same argument after composing with the constructed inverse proves the asserted uniqueness. Different choices of approximations and of γ\gamma are consequently compared by unique bounded AA-algebra maps.

For the module assertion choose Av→FAw→M→0A^v\xrightarrow{F}A^w\to M\to 0. Ordinary tensor associativity, [StacksAlg] (Tag 00D2, Lemma 10.12.7), gives C⊗B(B⊗AM)≃C⊗AMC\otimes_B(B\otimes_A M)\simeq C\otimes_A M, sending c⊗(b⊗m)c\otimes(b\otimes m) to cb⊗mcb\otimes m. Theorem 2.17.2 identifies the successive relation-quotient module with Cw/FCCvC^w/F_CC^v, with its quotient norm. The algebra comparison gives a bounded inverse comparison with Dw/FDDvD^w/F_DD^v, and it is isometric if the coefficient comparison is isometric. All these maps agree on the images of the chosen generators and are therefore natural in MM. Finally, Theorem 2.17.2 and Proposition 1.11.6 identify each finite ordinary tensor here with its separated completion. No passage from algebraic flatness to an assertion about arbitrary completed tensors is involved. ◻

References. The transitivity theorem is classical: see [BGR84] (§7.2.4, Theorem 2, statement and proof on pp. 285–286) and [Bosch14] (§3.3, Proposition 17, pp. 56–57). For the stability of special domains under small changes compare [BGR84] (§7.2.3, Proposition 3, p. 281). The proofs above reconstruct the rational case with explicit norm bounds, actual quotient maps, and a separate argument at trivial valuation. The classical texts provide attribution and comparison; their transitivity or maximum-principle statements are not proof inputs.

Proposition 2.25.5. NAG-001503 (The rational site of a rational member). Let U⊆X=M(A)U\subseteq X=\mathcal M(A) be represented by a strict rational algebra BB with value-group radii. Under M(B)≃U\mathcal M(B)\simeq U, the rational subsets defined over BB are exactly the subsets of UU which are rational over AA. Hence the rational site of M(B)\mathcal M(B) is identified with the slice RX/U\mathcal R_X/U of Definition 2.23.1, with finite set-theoretic covers on both sides. The structure sheaves agree under this identification, independently up to their unique bounded isomorphisms of all presentation choices. For every finite AA-module MM, the restriction of M~rat\widetilde M^{\mathrm{rat}} corresponds to the rational module sheaf associated to B⊗AMB\otimes_A M. The coefficient and finite-module comparison maps are bounded in both directions.

Proof. Every rational subset over BB is rational over AA by Theorem 2.25.4. Conversely, let V⊆UV\subseteq U have a rational presentation AVA_V over AA. Map its unit-ideal tuple into BB. This gives a rational presentation over BB, and Theorem 2.19.3 identifies its point set with the inverse image of VV, which is VV under M(B)≃U\mathcal M(B)\simeq U. Proposition 2.22.6 identifies the joint algebra with B⊗^AAVB\widehat\otimes_A A_V. Its point set is U∩V=VU\cap V=V. Theorem 2.22.5 therefore gives its unique bounded AA-algebra isomorphism with AVA_V, with bounded inverse. The coefficient map from BB agrees under this isomorphism by uniqueness of the factorization through UU.

The two categories have now been identified as partially ordered sets: arrows on both sides are the same inclusions of subsets. Their finite covering families are the same families of subsets with the same union, including the empty covering of the empty object. On every object the coefficient algebras have just been identified. Restriction maps commute with these comparisons, since each is the unique bounded factorization for the corresponding inclusion. Hence the ring presheaves, and therefore the sheaves of Corollary 2.23.2, agree. Theorem 2.25.4 supplies the compatible finite-module comparison C⊗B(B⊗AM)≃C⊗AM≃AV⊗AMC\otimes_B(B\otimes_A M)\simeq C\otimes_A M \simeq A_V\otimes_A M whenever CC is a chosen BB-rational presentation of VV. These maps agree on pure tensors, so their compatibility with restrictions and with morphisms of MM follows directly. This proves the assertion for module sheaves and their canonical norms. ◻

Corollary 2.25.6. NAG-001504 (Finite trees with functions on intermediate rational domains). Take a finite rooted tree of strict rational presentations, with root AA and with every child a rational presentation over its parent, using value-group radii. Then every vertex represents a rational subset of M(A)\mathcal M(A). All finite intersections of these subsets admit rational presentations over AA. Their coefficient algebras, restriction maps and associated finite-AA-module quotients are independent of the chosen flattenings up to unique compatible bounded isomorphisms. A presentation of an intersection obtained by a further finite rational tower has the same comparison.

Suppose in addition that the children of each nonterminal vertex cover its spectrum. Then the leaves form a finite rational cover of M(A)\mathcal M(A). For every finite canonical AA-module, its augmented ordered and all-tuples leaf Čech complexes, with these intersection identifications, are norm-strict exact in every degree. Repeated indices and actual zero quotients are retained with the conventions already stated.

Proof. Apply Theorem 2.25.4 inductively along the path from the root to each vertex. This terminates because the tree is finite. The point comparison identifies every parent-child map with the corresponding inclusion in M(A)\mathcal M(A). Proposition 2.22.6 now gives each finite intersection over AA, with its actual relation quotient. Theorem 2.22.5 gives uniqueness and compatibility of its restriction maps. The finite-module identifications follow from Theorem 2.25.4 on the paths and Theorem 2.17.2 on the intersections. An additional finite rational tower is flattened by the same induction, and its presentation of the same subset is then compared by the same uniqueness theorem.

If the children cover each nonterminal vertex, induction from the leaves towards the root shows that the union of the leaves is the root spectrum. Apply Theorem 2.22.9 to this finite rational cover of M(A)\mathcal M(A). The preceding compatible algebra and module maps identify its terms and differentials with the stated leaf complex. In every fixed degree only finitely many terms occur, so the product of the comparison maps and its inverse are bounded. Such a bounded isomorphism of complexes preserves exactness, closed images and the quotient-to-induced norm comparison on each image. The theorem therefore gives the claimed norm-strict exactness, including its all-tuples convention. ◻

Finite rational constructions can therefore be expressed over the original algebra, with compatible section modules. This is the geometric input to the effective descent argument in Theorem 2.28.12. The comparison with affinoid-domain sites is established in Theorems 2.30.6 and 2.38.4.

2.26 Extending dense image across a rational boundary

A dense-image statement on a rational domain can persist when its defining bounds are enlarged slightly. To prove this with quotient norms, the radii must be allowed to vary through a dense subgroup of R>0\mathbb R_{>0}. We first realize the required radii over the original field. No extension of scalars will be used. Put ΓKdiv={r>0:rn∈∣K×∣ for some integer n≥1}.\Gamma_K^{\mathrm{div}} =\{r>0:r^n\in|K^\times|\text{ for some integer }n\geq1\}. The field KK is complete. All preparatory statements allow trivial valuation, although the hypothesis r0<r1r_0<r_1 in ΓKdiv\Gamma_K^{\mathrm{div}} has no instance in that case. The extension theorem explicitly assumes nontrivial valuation. Nonzero strict affinoid algebras have norm-admissible Banach norms, and zero quotients use the zero-norm convention.

Lemma 2.26.1. NAG-001506 (Radii in the divisible closure of the value group).

Let AA be strict norm-admissible and let r1,…,rn∈ΓKdivr_1,\ldots,r_n\in\Gamma_K^{\mathrm{div}}. Then the weighted coefficient algebra D=A{r−1T}D=A\{r^{-1}T\} is strict norm-admissible. Every algebraic ideal of DD is finite and closed, and its actual quotient is strict, with zero allowed. Adjoining another finite block with radii in ΓKdiv\Gamma_K^{\mathrm{div}} commutes isometrically with taking such a quotient: for B=D/IB=D/I and F=D{s−1Y}F=D\{s^{-1}Y\}, F/IF≃B{s−1Y}F/IF\simeq B\{s^{-1}Y\} isometrically, with actual kernel IFIF. A further finite relation quotient is the direct quotient by the sum of the lifted actual ideals, with the same norm.

Proof. For each ii, choose an integer ni≥1n_i\geq 1 and ci∈K×c_i\in K^\times with ∣ci∣=rini|c_i|=r_i^{n_i}. Write Si=ci−1TiniS_i=c_i^{-1}T_i^{n_i} and put R=A⟨S1,…,Sn⟩R=A\langle S_1,\ldots,S_n\rangle. Grouping exponents modulo the nin_i gives a unique expansion D=⨁0≤βi<niR Tβ,∥∑βFβ(S)Tβ∥r=max⁡βrβ∥Fβ∥.D=\bigoplus_{0\leq\beta_i<n_i}R\,T^\beta, \qquad \left\|\sum_\beta F_\beta(S)T^\beta\right\|_r =\max_\beta r^\beta\|F_\beta\|. Indeed, the coefficient of SνTβS^\nu T^\beta is anν+βcνa_{n\nu+\beta}c^\nu, whose weighted norm is ∥anν+β∥rnν+β\|a_{n\nu+\beta}\|r^{n\nu+\beta}. There are only finitely many β\beta, so the finite-tail conditions on both sides are equivalent. Multiplication is determined by Tini=ciSiT_i^{n_i}=c_iS_i and agrees on polynomials, hence by continuity on the coefficient algebras. Thus DD is a finite free RR-algebra. Its displayed norm is uniformly equivalent to the maximum norm in this finite basis. Lemma 2.17.1 makes RR strict, and Proposition 1.14.4 makes DD strict with precisely this norm-admissibility. No assertion about arbitrary positive radii follows from this finite regrouping.

Theorem 1.11.5 now gives finite generation and closedness of every ideal of DD. Its quotient is complete by Lemma 1.11.1; composing bounded quotient lifts with a strict Tate presentation of DD proves that the quotient norm is admissible. This also covers the zero quotient.

For the iterated assertion, regrouping coefficients first identifies the joint coefficient algebra with F=D{s−1Y}F=D\{s^{-1}Y\}, with equality of norms. Proposition 2.14.1, applied to D→D/ID\to D/I, gives a surjective coefficient map with quotient norm equal to the coefficient norm on B{s−1Y}B\{s^{-1}Y\}. Its kernel consists of the convergent coefficient families in II. Choose generators u1,…,uqu_1,\ldots,u_q of II. By Theorem 1.11.5, every i∈Ii\in I can be written i=∑ℓuℓdℓi=\sum_\ell u_\ell d_\ell with max⁡ℓ∥dℓ∥≤C∥i∥\max_\ell\|d_\ell\|\leq C\|i\| for a fixed CC. Applying these lifts coefficientwise proves that the kernel is the actual ideal IFIF. This repeats the coefficient-lifting argument of Lemma 2.18.1, whose needed strictness is now supplied by the preceding paragraph. Finally the inverse image of a finitely generated relation ideal is IFIF plus its lifted generators: lift the finitely many coefficients of any relation expression and subtract. Taking successive infima over representatives gives the direct quotient norm. No infimum need be attained. ◻

Proposition 2.26.2. NAG-001507 (Power presentations for rational inequalities). Let (g,f1,…,fs)=A(g,f_1,\ldots,f_s)=A and let ri∈ΓKdivr_i\in\Gamma_K^{\mathrm{div}}. Choose a common integer n≥1n\geq 1 and ci∈K×c_i\in K^\times with ∣ci∣=rin|c_i|=r_i^n. Then Q=A{r−1T}/(gTi−fi)i,P=A{W1,…,Ws}/(gnWi−ci−1fin)iQ=A\{r^{-1}T\}/(gT_i-f_i)_i, \qquad P=A\{W_1,\ldots,W_s\}/(g^nW_i-c_i^{-1}f_i^n)_i are canonically isomorphic by bounded inverse AA-algebra maps. The tuple defining PP generates the unit ideal. Thus QQ is a strict affinoid, and its spectrum identifies homeomorphically with {x∈M(A):∣fi(x)∣≤ri∣g(x)∣ for all i}.\{x\in\mathcal M(A):|f_i(x)|\leq r_i|g(x)|\text{ for all }i\}. It has the strict-target domain universal property of Theorem 2.22.5. These conclusions concern the stated actual quotient norms, with bounded comparison; an isometry with the power presentation is not asserted.

Proof. Raise a unit-ideal identity for (g,f1,…,fs)(g,f_1,\ldots,f_s) to the exponent (s+1)(n−1)+1(s+1)(n-1)+1. In every term, at least one of these s+1s+1 elements occurs to an exponent at least nn. Thus (gn,f1n,…,fsn)=A(g^n,f_1^n,\ldots,f_s^n)=A, and multiplying individual entries by ci−1c_i^{-1} preserves that ideal. Lemma 2.26.1 makes QQ a strict actual quotient; the ordinary strict quotient theorem does the same for PP. In QQ the element gg is a unit by the original certificate. Sending WiW_i to Tin/ciT_i^n/c_i, which has norm at most one, gives a contractive map P→QP\to Q.

In PP, its unit-ideal certificate makes gng^n, and hence gg, a unit. Put bi=fi/gb_i=f_i/g. The relation gives bin=ciWib_i^n=c_iW_i. If m=qn+jm=qn+j with 0≤j<n0\leq j<n, then ∥bim∥≤∣ci∣q∥bij∥≤Cirim,Ci=max⁡0≤j<nri−j∥bij∥.\|b_i^m\|\leq |c_i|^q\|b_i^j\| \leq C_i r_i^m, \qquad C_i=\max_{0\leq j<n}r_i^{-j}\|b_i^j\|. Proposition 1.9.3 therefore defines a bounded map A{r−1T}→PA\{r^{-1}T\}\to P sending TiT_i to bib_i. It kills the actual relation ideal and descends boundedly to QQ by taking infima over lifts. The two composites fix the variables: in PP, fin/(cign)=Wif_i^n/(c_ig^n)=W_i, and in QQ, fi/g=Tif_i/g=T_i. Polynomial density proves that they are inverse. Zero targets and zero quotients satisfy the same conclusion directly.

The point theorem and the strict-target universal property for the unit-radius presentation PP now apply. Its inequalities are ∣fi∣n≤∣ci∣ ∣g∣n|f_i|^n\leq|c_i|\,|g|^n, exactly the displayed inequalities for QQ. Transferring across the bounded inverse isomorphism proves both assertions. In particular, every such domain also has a unit-radius rational presentation over the original field. ◻

Lemma 2.26.3. NAG-001508 (Perturbing affinoid generators). Let π0:A⟨X1,…,Xn⟩↠B\pi_0:A\langle X_1,\ldots,X_n\rangle\twoheadrightarrow B be a surjective homomorphism, with AA strict norm-admissible and BB carrying its actual quotient norm. Write hi0=π0(Xi)h_i^0=\pi_0(X_i). If hi∈Bh_i\in B satisfy δ=max⁡i∥hi−hi0∥<1,\delta=\max_i\|h_i-h_i^0\|<1, then substitution Xi↦hiX_i\mapsto h_i defines a contractive surjection π:A⟨X⟩→B\pi:A\langle X\rangle\to B. The quotient norm induced by π\pi equals the given norm on BB. The case n=0n=0 uses δ=0\delta=0.

Proof. Each hi0h_i^0 and hih_i has norm at most one. The bounded substitution criterion gives π\pi. A telescoping product expansion gives ∥hα−(h0)α∥≤δ\|h^\alpha-(h^0)^\alpha\|\leq\delta for every nonconstant monomial, and the constant monomial has difference zero. The ultrametric inequality, followed by polynomial approximation, therefore gives ∥π(F)−π0(F)∥≤δ∥F∥(F∈A⟨X⟩).\|\pi(F)-\pi_0(F)\|\leq\delta\|F\| \qquad(F\in A\langle X\rangle). Fix λ>1\lambda>1 with q=λδ<1q=\lambda\delta<1. For b∈Bb\in B, choose a π0\pi_0-lift F0F_0 with ∥F0∥≤λ∥b∥\|F_0\|\leq\lambda\|b\|, taking zero when b=0b=0. Then e1=b−π(F0)e_1=b-\pi(F_0) has norm at most q∥b∥q\|b\|. Repeat for each successive error. This produces FjF_j with ∥Fj∥≤λqj∥b∥,∥b−π(∑j=0mFj)∥≤qm+1∥b∥.\|F_j\|\leq\lambda q^j\|b\|, \qquad \left\|b-\pi\left(\sum_{j=0}^m F_j\right)\right\| \leq q^{m+1}\|b\|. The sum F=∑jFjF=\sum_jF_j converges, has norm at most λ∥b∥\lambda\|b\|, and satisfies π(F)=b\pi(F)=b. Thus π\pi is surjective. Its quotient norm is at least ∥b∥\|b\| because it is contractive, and at most λ∥b∥\lambda\|b\| for every sufficiently small λ>1\lambda>1. Letting λ↓1\lambda\downarrow 1 proves equality. This proof requires only approximate quotient lifts; it does not assume that a norm-minimizing lift exists. If B=0B=0, the statement is immediate. ◻

Proposition 2.26.4. NAG-001509 (Dense image and a Weierstrass factorization). For a bounded homomorphism ϕ:A→B\phi:A\to B of strict KK-affinoids, the following are equivalent: ϕ(A)\phi(A) is dense in BB; and there are finitely many a1,…,an∈Aa_1,\ldots,a_n\in A and a bounded surjection A⟨a1,…,an⟩:=A{X1,…,Xn}/(Xi−ai)i↠BA\langle a_1,\ldots,a_n\rangle :=A\{X_1,\ldots,X_n\}/(X_i-a_i)_i \twoheadrightarrow B whose composite with AA is ϕ\phi. Thus density supplies the algebraic factorization called a Runge immersion in the classical terminology, without using a stalk criterion.

Proof. Choose a relative presentation π0:A⟨X⟩↠B\pi_0:A\langle X\rangle\twoheadrightarrow B by Lemma 1.14.2, and use its quotient norm on BB. This replaces its original norm by a uniformly equivalent one. If ϕ(A)\phi(A) is dense, choose ai∈Aa_i\in A so that ∥π0(Xi)−ϕ(ai)∥<1\|\pi_0(X_i)-\phi(a_i)\|<1. Lemma 2.26.3 makes substitution Xi↦ϕ(ai)X_i\mapsto\phi(a_i) surjective. It kills each Xi−aiX_i-a_i and hence factors through the displayed actual quotient as a bounded surjection.

Conversely, in that quotient every polynomial in the variables is the image of an element of AA, because Xi=aiX_i=a_i. Polynomial density therefore makes AA dense in the Weierstrass quotient. A bounded surjection from it to BB sends this dense image to a dense subset of BB: approximate any preimage of a given element. Thus ϕ(A)\phi(A) is dense. The argument includes zero algebras and the empty list of variables. ◻

The next estimate concerns a fixed finite list of elements and a fixed relation ideal. It is enough for the extension theorem; it is not a claim that quotient norms vary continuously for every element of an arbitrary family.

Lemma 2.26.5. NAG-001510 (Preserving a strict quotient-norm bound).

Let RR be strict norm-admissible, let r0<r1r_0<r_1 belong to ΓKdiv\Gamma_K^{\mathrm{div}}, and let T=(T1,…,Ts)T=(T_1,\ldots,T_s). For each r∈ΓKdiv∩[r0,r1]r\in\Gamma_K^{\mathrm{div}}\cap[r_0,r_1], put Er=R{r−1T}E_r=R\{r^{-1}T\}. Fix polynomials H1,…,Hq∈R[T]H_1,\ldots,H_q\in R[T], let Jr=(H1,…,Hq)ErJ_r=(H_1,\ldots,H_q)E_r, and give Qr=Er/JrQ_r=E_r/J_r its actual quotient norm. If finitely many F1,…,Fn∈Er1F_1,\ldots,F_n\in E_{r_1} satisfy ∥[Fi]∥Qr0<1\|[F_i]\|_{Q_{r_0}}<1, then there is a real number r2r_2 with r0<r2≤r1r_0<r_2\leq r_1 such that ∥[Fi]∥Qr<1for all i and r∈ΓKdiv∩[r0,r2].\|[F_i]\|_{Q_r}<1 \quad\text{for all }i \text{ and }r\in\Gamma_K^{\mathrm{div}}\cap[r_0,r_2].

Proof. All the ideals in the statement are actual closed ideals by Lemma 2.26.1. For each ii, the quotient-norm inequality gives ji∈Jr0j_i\in J_{r_0} with ∥Fi+ji∥r0<1\|F_i+j_i\|_{r_0}<1. Write ji=∑νuiνHνj_i=\sum_\nu u_{i\nu}H_\nu, where uiν∈Er0u_{i\nu}\in E_{r_0}. Approximate the finitely many uiνu_{i\nu} by polynomials viν∈R[T]v_{i\nu}\in R[T] closely enough that Gi=Fi+∑νviνHν∈Er1,∥Gi∥r0<1.G_i=F_i+\sum_\nu v_{i\nu}H_\nu\in E_{r_1}, \qquad \|G_i\|_{r_0}<1. This step uses an expression in the actual finitely generated ideal, rather than merely membership in a closure. It does not require a norm-minimizing representative. Choose θ<1\theta<1 strictly larger than every ∥Gi∥r0\|G_i\|_{r_0}. Write Gi=∑αaiαTαG_i=\sum_\alpha a_{i\alpha}T^\alpha. Since Gi∈Er1G_i\in E_{r_1}, outside a finite set of indices one has ∥aiα∥r1∣α∣<θ\|a_{i\alpha}\|r_1^{|\alpha|}<\theta, simultaneously for all ii. Those coefficients satisfy the same bound for every r≤r1r\leq r_1. For each of the finitely many remaining indices, the real function r↦∥aiα∥r∣α∣r\mapsto\|a_{i\alpha}\|r^{|\alpha|} is continuous and is strictly less than θ\theta at r0r_0. Hence it remains less than θ\theta on some common interval [r0,r2][r_0,r_2] with r2>r0r_2>r_0. Shrink r2r_2 to at most r1r_1. Then ∥Gi∥r≤θ<1\|G_i\|_r\leq\theta<1 throughout this interval. Since Gi−Fi∈JrG_i-F_i\in J_r, the same strict bound holds in QrQ_r. An empty list of elements requires no estimate. ◻

Theorem 2.26.6. NAG-001511 (Extending dense image to a larger rational domain). Assume that KK is complete and nontrivially valued. Let ϕ:A→B\phi:A\to B be a bounded homomorphism of strict norm-admissible KK-affinoids, and let (g,f1,…,fs)=A(g,f_1,\ldots,f_s)=A. For r∈ΓKdivr\in\Gamma_K^{\mathrm{div}}, set Ar=A{r−1T}/(gTi−fi)i,Br=B{r−1T}/(ϕ(g)Ti−ϕ(fi))i.A_r=A\{r^{-1}T\}/(gT_i-f_i)_i, \qquad B_r=B\{r^{-1}T\}/(\phi(g)T_i-\phi(f_i))_i. Use their actual quotient norms and let ϕr:Ar→Br\phi_r:A_r\to B_r be the induced bounded map. If ϕr0(Ar0)\phi_{r_0}(A_{r_0}) is dense in Br0B_{r_0} for some r0∈ΓKdivr_0\in\Gamma_K^{\mathrm{div}}, then there is a real r2>r0r_2>r_0 such that ϕr(Ar)\phi_r(A_r) is dense in BrB_r for every r∈ΓKdiv∩[r0,r2]r\in\Gamma_K^{\mathrm{div}}\cap[r_0,r_2]. In particular, the conclusion holds at some strictly larger radius. Each such dense map has the Weierstrass quotient factorization of Proposition 2.26.4.

Proof. All displayed algebras are strict by Lemma 2.26.1; Proposition 2.26.2 identifies their spectra with the stated rational inequalities. The map on coefficients induced by ϕ\phi is bounded with the same bound on every weighted coefficient algebra. It descends to the actual quotients by taking infima over representatives. Since KK is nontrivially valued, ΓKdiv\Gamma_K^{\mathrm{div}} is dense in R>0\mathbb R_{>0}: for any π∈K\pi\in K with 0<∣π∣<10<|\pi|<1, it contains ∣π∣Q|\pi|^{\mathbb Q}. Choose r1∈ΓKdivr_1\in\Gamma_K^{\mathrm{div}} with r1>r0r_1>r_0.

First reduce to Weierstrass equations over a slightly larger domain. Put A‾=Ar1\overline A=A_{r_1} and B‾=Br1\overline B=B_{r_1}. The unit-ideal certificate makes gg a unit in A‾\overline A, so put ui=fi/g∈A‾u_i=f_i/g\in\overline A. For every r≤r1r\leq r_1 in the indicated group there are canonical isometric identifications Ar≃A‾{r−1S}/(Si−ui)i,Br≃B‾{r−1S}/(Si−ϕ(ui))i.A_r\simeq\overline A\{r^{-1}S\}/(S_i-u_i)_i, \qquad B_r\simeq\overline B\{r^{-1}S\}/(S_i-\phi(u_i))_i. For the first identification, the map from left to right sends TiT_i to SiS_i. The map in the other direction first sends the radius-r1r_1 variable of A‾\overline A to the radius-rr variable TiT_i, and then sends SiS_i to TiT_i. Both substitutions are contractive because r≤r1r\leq r_1. They kill the actual relations; their composites fix coefficients and variables, so polynomial density makes them inverse. The proof for BB is identical, and the squares with ϕr\phi_r commute. These arguments also cover zero quotients. If A‾=0\overline A=0, then B‾=0\overline B=0 and all the algebras at smaller radii vanish, so the conclusion already follows.

Replace A,BA,B by A‾,B‾\overline A,\overline B and use these identifications. Thus it suffices to treat the family Ar=A{r−1T}/(Ti−ui)A_r=A\{r^{-1}T\}/(T_i-u_i) and the analogous family for BB, for r0≤r≤r1r_0\leq r\leq r_1. The image of AA is dense in each ArA_r, since polynomials in the variables reduce to elements of AA. Density of Ar0→Br0A_{r_0}\to B_{r_0} therefore implies density of the image of AA in Br0B_{r_0}.

Choose a relative presentation π0:E=A⟨X1,…,Xn⟩↠B\pi_0:E=A\langle X_1,\ldots,X_n\rangle\twoheadrightarrow B by Lemma 1.14.2. Replace the norm of BB by its quotient norm. This is a uniformly equivalent norm; the resulting coefficient and quotient norms at every radius are again uniformly equivalent to the original ones, so density is unchanged. Its kernel II has finitely many generators H1,…,Hq∈EH_1,\ldots,H_q\in E by Theorem 1.11.5. The iterated quotient assertion of Lemma 2.26.1 gives isometric descriptions Br≃E{r−1T}/(H1,…,Hq,Ti−ui)i,Ar⟨X⟩↠Br,B_r\simeq E\{r^{-1}T\}/(H_1,\ldots,H_q,T_i-u_i)_i, \qquad A_r\langle X\rangle\twoheadrightarrow B_r, where the second map also induces the displayed norm as its actual quotient norm. In particular, the images hj0h_j^0 of the XjX_j have norm at most one. Choose aj∈Aa_j\in A with ∥hj0−ϕ(aj)∥Br0<1\|h_j^0-\phi(a_j)\|_{B_{r_0}}<1; this is possible by the density just proved. The elements Fj=Xj−ajF_j=X_j-a_j belong to E⊂E{r1−1T}E\subset E\{r_1^{-1}T\}. Apply Lemma 2.26.5 with coefficient algebra EE and the fixed polynomials HνH_\nu and Ti−uiT_i-u_i. It gives some r2>r0r_2>r_0, at most r1r_1, such that ∥hj0−ϕ(aj)∥Br<1(r∈ΓKdiv∩[r0,r2]).\|h_j^0-\phi(a_j)\|_{B_r}<1 \quad (r\in\Gamma_K^{\mathrm{div}}\cap[r_0,r_2]). For each such rr, Lemma 2.26.3 says that substitution Xj↦ϕ(aj)X_j\mapsto\phi(a_j) makes Ar⟨X⟩→BrA_r\langle X\rangle\to B_r surjective. Every polynomial image is already in the image of ArA_r, so polynomial density proves that ϕr(Ar)\phi_r(A_r) is dense in BrB_r. Proposition 2.26.4 supplies its stated factorization. Finally, density of ΓKdiv\Gamma_K^{\mathrm{div}} gives a radius strictly between r0r_0 and r2r_2. The proof never changes the scalar field, uses no Gerritzen–Grauert theorem, and includes empty variable lists and actual zero quotients. ◻

Source and scope. The generator perturbation and the dense-image criterion reconstruct Bosch, [Bosch14] (§4.2, Lemma 8 and Proposition 7, pp. 72–73). The extension argument reconstructs [Bosch14] (§4.2, Extension Lemma 9, pp. 73–76). The proof above replaces norm-attaining lifts by approximate quotient lifts and realizes radii in ΓKdiv\Gamma_K^{\mathrm{div}} through a finite algebra over the original field. The extension conclusion is restricted to nontrivial valuation, because the required strictly larger radii do not exist in ΓKdiv\Gamma_K^{\mathrm{div}} for a trivially valued field. The finite-algebra realization itself, the perturbation lemma, and the dense-image criterion do allow trivial valuation.

Gerritzen–Grauert in the strict category

We now use the relative Weierstrass results to obtain a finite rational description of affinoid domains. The argument first treats bounded categorical epimorphisms of strict affinoid algebras. It produces rational pieces on which the image is dense. The domain universal property then removes the closed equations from the resulting Weierstrass factorization. Except in the following definition and its immediate observations, KK is complete and its fixed absolute value is nontrivial. All algebra homomorphisms are unital and bounded.

Definition 2.27.1. NAG-001513 For any complete non-Archimedean field KK, including trivial valuation, let AA and BB be strict KK-affinoid algebras. A strict-test affinoid-domain presentation of a subset U⊂M(A)U\subset\mathcal M(A) is a bounded map φ:A→B\varphi:A\to B such that the map M(B)→M(A)\mathcal M(B)\to\mathcal M(A) identifies its points bijectively with UU, and every bounded KK-algebra map A→CA\to C, with CC a strict KK-affinoid and with spectrum image contained in UU, factors uniquely through a bounded map B→CB\to C. The zero algebra represents the empty subset.

This is the universal property of Definition 2.1.1 restricted to strict source algebras, strict representing algebras, and strict test algebras. An affinoid domain in that original definition satisfies the present condition whenever its two algebras are strict. The present definition alone makes no assertion about generalized test algebras.

The empty-set assertion follows from Theorem 1.12.3, which says that a strict algebra with empty spectrum is zero. Every such presentation is a strict-affinoid epimorphism in the sense of Definition 1.14.7: two maps from BB agreeing on AA are two factorizations of the same map into a strict test algebra. Also, the point identification is a homeomorphism onto UU, since the source spectrum is compact and the target spectrum is Hausdorff by Proposition 2.24.1.

Lemma 2.27.2. NAG-001514 (A kernel series with unit coefficient ideal). Let φ:A→B\varphi:A\to B be a strict-affinoid epimorphism, let A≠0A\ne 0, and let π:A⟨T1,…,Tn⟩↠B\pi:A\langle T_1,\ldots,T_n\rangle\twoheadrightarrow B be a bounded admissible relative presentation with n≥1n\geq 1. There is an element F∈ker⁡πF\in\ker\pi whose coefficients generate the unit ideal of AA.

Proof. Fix a maximal ideal m\mathfrak m of AA and put κ=A/m\kappa=A/\mathfrak m. Theorem 1.12.3 makes κ\kappa a complete finite extension of KK. Coefficientwise quotient exactness, Proposition 2.14.1 and Theorem 2.14.2, gives κ⟨T1,…,Tn⟩↠D=B/mB,D=B⊗^Aκ.\kappa\langle T_1,\ldots,T_n\rangle\twoheadrightarrow D=B/\mathfrak mB, \qquad D=B\widehat\otimes_A\kappa. Here the equality with the completed pushout also follows from Proposition 1.14.3; all quotients are by actual closed ideals. In particular, DD is a strict κ\kappa-affinoid. The map κ→D\kappa\to D is a strict-affinoid epimorphism over κ\kappa. Indeed, a strict κ\kappa-affinoid test algebra is strict over KK by Lemma 1.14.6; two bounded maps from DD agreeing on κ\kappa give maps from BB agreeing on AA, hence are equal. By Lemma 1.14.9, either D=0D=0 or D=κD=\kappa. In either case the displayed map from κ⟨T⟩\kappa\langle T\rangle has nonzero kernel: for n≥1n\geq 1 this Tate algebra has the two distinct maximal evaluation ideals at (0,…,0)(0,\ldots,0) and (1,0,…,0)(1,0,\ldots,0) and cannot be a field.

The map ker⁡π→ker⁡(κ⟨T⟩→D)\ker\pi\to\ker(\kappa\langle T\rangle\to D) is surjective. To see this, lift a relation to G∈A⟨T⟩G\in A\langle T\rangle. Its image under π\pi lies in mB\mathfrak mB, which is the image of mA⟨T⟩\mathfrak m A\langle T\rangle because π\pi is surjective. Subtract a lift in that ideal. Thus, at every maximal m\mathfrak m, some series in ker⁡π\ker\pi has a coefficient outside m\mathfrak m. The ideal of AA generated by all coefficients of all these series is the unit ideal, by [StacksAlg] (Tag 00E0(2), Lemma 10.17.2). A finite unit certificate uses finitely many series G1,…,Gr∈ker⁡πG_1,\ldots,G_r\in\ker\pi.

The algebra AA is Noetherian by Theorem 1.11.5. Choose d≥1d\geq 1 such that, for every ii, the coefficients of GiG_i of total degree less than dd generate the same ideal as all its coefficients. Set F=G1+TndG2+⋯+Tn(r−1)dGr∈ker⁡π.F=G_1+T_n^dG_2+\cdots+T_n^{(r-1)d}G_r\in\ker\pi. At a maximal point xx, let ii be the first index with Gi,x≠0G_{i,x}\ne 0. Such an index exists by the unit certificate. Some coefficient of Gi,xG_{i,x} of total degree less than dd is nonzero. After the indicated shift, its degree is less than idid. Earlier summands vanish at xx, while every later summand has total degree at least idid, so that coefficient cannot cancel. Hence Fx≠0F_x\ne 0 for every maximal xx. Its coefficients have no common maximal zero and therefore generate the unit ideal, as required. ◻

Theorem 2.27.3. NAG-001515 (Rational-local density for strict-affinoid epimorphisms). Let φ:A→B\varphi:A\to B be a strict-affinoid epimorphism over the complete nontrivially valued field KK. There is a finite rational cover M(A)=⋃iRi\mathcal M(A)=\bigcup_iR_i such that, writing AiA_i for the rational algebra of RiR_i and Bi=B⊗^AAiB_i=B\widehat\otimes_AA_i for its pullback, every map Ai→BiA_i\to B_i has dense image. All the algebras Ai,BiA_i,B_i are strict; zero members are permitted. The cover concerns all bounded multiplicative seminorms.

Proof. We first record the base-change facts used in the induction. A categorical epimorphism remains an epimorphism under the completed pushouts of Proposition 1.14.3: two maps from the pushout agreeing on its new base agree on BB by the old epimorphism property, and therefore agree everywhere by the pushout property. For a rational base change, this pushout is precisely the rational quotient of BB with the pulled-back defining functions, by its universal property. An admissible relative presentation A⟨T⟩↠BA\langle T\rangle\twoheadrightarrow B gives an admissible relative presentation A′⟨T⟩↠B⊗^AA′A'\langle T\rangle\twoheadrightarrow B\widehat\otimes_AA' by the same actual-ideal presentation formula. Every maximal point of a strict rational algebra over AA restricts to a maximal point of AA, and the finite residue extension preserves absolute values, as proved in Lemma 1.15.4. Consequently fibrewise distinguished orders are preserved under these base changes.

Induct on the number nn of variables in an admissible relative presentation, which exists by Lemma 1.14.2. For n=0n=0 the map is surjective, and the one-member cover suffices. If A=0A=0 or B=0B=0, the same conclusion is immediate. For n≥1n\geq 1, Lemma 2.27.2 supplies a series in the kernel with unit coefficient ideal. By Lemma 1.15.3, an isometric coordinate change makes it TnT_n-distinguished of order at most some integer ss at every maximal point. The relative presentation remains admissible.

Assume the theorem proved for presentations with fewer than nn variables. For the fixed number nn, we use a second induction on ss for presentations whose kernel contains such a series FF. If s=0s=0, Proposition 1.15.5 gives A⟨T⟩/(F)=0A\langle T\rangle/(F)=0, so B=0B=0. Suppose s>0s>0. Let Rs⊂M(A)R_s\subset\mathcal M(A) be the rational maximal-order locus of Lemma 1.15.4, with algebra AsA_s, and put Bs=B⊗^AAsB_s=B\widehat\otimes_AA_s. If As=0A_s=0, all maximal orders are at most s−1s-1, and the inner induction applies directly. Otherwise, over AsA_s the image of FF has order exactly ss. Proposition 1.15.5 makes Cs=As⟨T1,…,Tn−1⟩⟶BsC_s=A_s\langle T_1,\ldots,T_{n-1}\rangle\longrightarrow B_s finite, since BsB_s is a quotient of As⟨T⟩/(F)A_s\langle T\rangle/(F). It is also a strict-affinoid epimorphism: two maps agreeing on CsC_s agree on AsA_s. Lemma 1.14.8 makes it surjective, and Lemma 1.14.1 makes this surjection admissible. The outer induction therefore yields a finite rational cover Rs=⋃i=1rUiR_s=\bigcup_{i=1}^rU_i on which the corresponding maps have dense image. By rational transitivity, Theorem 2.25.4, each UiU_i is rational in M(A)\mathcal M(A), with the same algebra and pullback algebra up to the unique bounded comparison maps.

Choose presentations Ui={x:∣fij(x)∣≤∣gi(x)∣ (1≤j≤mi)},(gi,fi1,…,fimi)=A.U_i=\{x:|f_{ij}(x)|\leq|g_i(x)|\ (1\leq j\leq m_i)\}, \qquad (g_i,f_{i1},\ldots,f_{im_i})=A. The extension theorem, Theorem 2.26.6, supplies ϵi∈∣K×∣\epsilon_i\in\sqrt{|K^\times|} with ϵi>1\epsilon_i>1 such that the map on Ui+={x:∣fij(x)∣≤ϵi∣gi(x)∣ for all j}U_i^+=\{x:|f_{ij}(x)|\leq\epsilon_i|g_i(x)|\text{ for all }j\} still has dense image. These sets have strict rational presentations by Proposition 2.26.2. We construct a finite rational family covering what remains, and disjoint from RsR_s. For each ii and 1≤j≤mi1\leq j\leq m_i, put Vij={x:ϵi∣gi(x)∣≤∣fij(x)∣,∣fiℓ(x)∣≤∣fij(x)∣ (ℓ≠j)}.V_{ij}=\{x:\epsilon_i|g_i(x)|\leq|f_{ij}(x)|, \quad |f_{i\ell}(x)|\leq|f_{ij}(x)|\ (\ell\ne j)\}. To see that this is strict rational, choose Ni≥1N_i\geq 1 and ci∈K×c_i\in K^\times with ∣ci∣=ϵiNi|c_i|=\epsilon_i^{N_i} and raise the inequalities to the NiN_ith power. The denominator is fijNif_{ij}^{N_i} and the numerators are cigiNic_ig_i^{N_i} and the other fiℓNif_{i\ell}^{N_i}. Their ideal is the unit ideal, since any maximal ideal containing them would contain gig_i and all the fiℓf_{i\ell}. The displayed inequalities show Vij∩Ui=∅V_{ij}\cap U_i=\varnothing: otherwise (ϵi−1)∣gi(x)∣≤0(\epsilon_i-1)|g_i(x)|\leq 0, so all entries of a unit tuple vanish at xx. If x∉Ui+x\notin U_i^+, choose jj attaining the maximum of the ∣fiℓ(x)∣|f_{i\ell}(x)|. Then x∈Vijx\in V_{ij}.

Take all finite intersections V=⋂iVijiV=\bigcap_iV_{ij_i}. They are strict rational by Proposition 2.22.6, they are disjoint from every UiU_i and hence from RsR_s, and together with the Ui+U_i^+ they cover all of M(A)\mathcal M(A). If some mi=0m_i=0, then Ui=M(A)U_i=\mathcal M(A) and the already dense map on UiU_i settles the assertion; thus the intersection family is only needed when every mi>0m_i>0. On each nonzero algebra AVA_V, every maximal point restricts to a maximal point of AA outside RsR_s. The restricted kernel series consequently has distinguished order at most s−1s-1. The inner induction gives a finite rational cover of VV with dense maps. By transitivity these final members are rational in M(A)\mathcal M(A). They, together with the Ui+U_i^+, give the desired finite cover. Both induction parameters are nonnegative integers and strictly decrease at their indicated steps. ◻

Lemma 2.27.4. NAG-001516 (A closed strict-test domain is Weierstrass). Suppose a bounded surjection q:C↠Dq:C\twoheadrightarrow D of strict KK-affinoid algebras is a strict-test affinoid-domain presentation. Then its kernel is generated by an idempotent e∈Ce\in C. For any c∈Kc\in K with ∣c∣>1|c|>1, its image is the Weierstrass domain {x∈M(C):∣ce(x)∣≤1},\{x\in\mathcal M(C):|ce(x)|\leq1\}, and the natural map C⟨S⟩/(S−ce)→DC\langle S\rangle/(S-ce)\to D is a bounded isomorphism.

Proof. Put I=ker⁡qI=\ker q. The algebra CC is Noetherian and I,I2I,I^2 are closed by Theorem 1.11.5. A multiplicative seminorm on CC annihilates I2I^2 if and only if it annihilates II, since a2∈I2a^2\in I^2 for every a∈Ia\in I. A bounded seminorm annihilating an ideal descends boundedly to its quotient: take the infimum of its bound over representatives. Thus M(C/I2)\mathcal M(C/I^2) and M(C/I)\mathcal M(C/I) have the same image in M(C)\mathcal M(C). The quotient C/IC/I is boundedly isomorphic to DD by Lemma 1.14.1. Apply the strict-test domain property to the map C→C/I2C\to C/I^2. Its factorization through C/IC/I forces I⊂I2I\subset I^2, hence I=I2I=I^2. The authorized algebraic lemma [StacksAlg] (Tag 00EH, Lemma 10.21.5) gives an idempotent ee with I=(e)I=(e).

At every multiplicative point, ∣e∣|e| is zero or one. Thus ∣ce∣≤1|ce|\leq 1 holds exactly where ee vanishes, which is the image of M(D)\mathcal M(D). To identify the algebras, put E=C⟨S⟩/(S−ce)E=C\langle S\rangle/(S-ce). In EE the element ee satisfies em=ee^m=e and Sm=cmeS^m=c^me. Since the variable SS is power bounded in its actual quotient norm, these identities give ∥e∥≤∣c∣−m∥Sm∥→0\|e\|\leq |c|^{-m}\|S^m\|\to 0, so e=0e=0. Consequently C→EC\to E factors through D=C/(e)D=C/(e). Conversely C→DC\to D sends cece to zero, so bounded substitution gives E→DE\to D with S↦0S\mapsto 0. The two maps are inverse on CC and on the variable SS, hence on the completions. Their boundedness follows from quotient norms and the admissible comparison with DD. ◻

Theorem 2.27.5. NAG-001517 (Strict Gerritzen–Grauert). Let KK be complete and nontrivially valued. Let φ:A→B\varphi:A\to B be a strict-test affinoid-domain presentation of U⊂M(A)U\subset\mathcal M(A). There is a finite rational cover M(A)=⋃iRi\mathcal M(A)=\bigcup_iR_i such that U∩RiU\cap R_i is a Weierstrass domain in RiR_i, represented by Bi=B⊗^AAiB_i=B\widehat\otimes_AA_i, where AiA_i is the rational algebra of RiR_i. In particular, UU is a finite union of rational domains of M(A)\mathcal M(A). These are assertions about all Berkovich points, with bounded algebra identifications and the fixed absolute value on KK.

Proof. As observed after Definition 2.27.1, φ\varphi is a strict-affinoid epimorphism. Theorem 2.27.3 supplies a finite rational cover for which every Ai→BiA_i\to B_i has dense image. This map is again a strict-test domain presentation of U∩RiU\cap R_i. Indeed, Theorem 2.19.3 identifies M(Bi)\mathcal M(B_i) with the inverse image of RiR_i in M(B)\mathcal M(B). If a strict test map from AiA_i has image in U∩RiU\cap R_i, its restriction from AA factors uniquely through BB; the two compatible maps from BB and AiA_i then give the required unique map from the completed pushout BiB_i.

Apply Proposition 2.26.4 to factor the dense map as Ai⟶Ci=Ai⟨S1,…,St⟩/(Sj−aj)↠Bi,aj∈Ai.A_i\longrightarrow C_i =A_i\langle S_1,\ldots,S_t\rangle/(S_j-a_j) \twoheadrightarrow B_i, \qquad a_j\in A_i. The first map represents a Weierstrass domain Wi⊂RiW_i\subset R_i, and the second map is surjective. The latter is itself a strict-test domain presentation. To check its factorization property, take a bounded map Ci→EC_i\to E, where EE is a strict test algebra, whose spectrum image lies in M(Bi)⊂M(Ci)\mathcal M(B_i)\subset\mathcal M(C_i). Composing with AiA_i gives a map with image in U∩RiU\cap R_i, hence a unique map Bi→EB_i\to E. Its composite from CiC_i equals the original map because both extend the same map from AiA_i through the rational domain CiC_i, whose bounded factorization is unique by Theorem 2.22.5. The point identification follows from the already established identifications into RiR_i.

By Lemma 2.27.4, the map Ci↠BiC_i\twoheadrightarrow B_i is a Weierstrass-domain map. It remains to retain the stronger conclusion that its composite with Ai→CiA_i\to C_i is Weierstrass, not merely rational. Let e∈Cie\in C_i be the kernel idempotent. The image of AiA_i is dense in CiC_i, because polynomials in the adjoined variables reduce to elements of AiA_i. Choose a∈Aia\in A_i such that ∥a−e∥Ci<∣c∣−1\|a-e\|_{C_i}<|c|^{-1} for some c∈Kc\in K with ∣c∣>1|c|>1. Since ∣e(x)∣|e(x)| is zero or one, the ultrametric inequality gives, for every x∈M(Ci)x\in\mathcal M(C_i), ∣ca(x)∣≤1⟺e(x)=0.|ca(x)|\leq1\quad\Longleftrightarrow\quad e(x)=0. Consequently U∩RiU\cap R_i is the Weierstrass subset of RiR_i defined by the original inequalities ∣aj∣≤1|a_j|\leq 1 together with ∣ca∣≤1|ca|\leq 1. Its standard strict rational algebra and BiB_i satisfy the same strict-test universal property, so they are uniquely boundedly isomorphic. Finally, each of these Weierstrass subsets is rational in RiR_i; rational transitivity makes it rational in M(A)\mathcal M(A). Their finite union is UU because the RiR_i cover the entire spectrum. ◻

Theorem 2.36.6 extends the finite rational-cover conclusion to generalized affinoid domains over every complete field. Its proof combines free-radius strictification with a finite truncation on the Gauss section, producing rational domains over the original algebra. This gives the general statement recorded in Theorem 2.11.3.

References. The finite induction and its relative Weierstrass preparations are due to Gerritzen–Grauert; the complete classical presentations used here are [BGR84] (7.3.5, Theorem 1 and Corollaries 2–3, p. 309; proof and preparations, pp. 310–315) and [Bosch14] (4.2, Theorem 10, p. 76; proof, pp. 79–81; Corollaries 11–12, pp. 76–77). The monomorphism formulation is also discussed in [Tem05] (Proposition 1.2, pp. 3–4); we do not use its proof or the Banach tensor-injectivity input occurring there. The present strict proof replaces analytic-stalk arguments by the explicit finite categorical-epimorphism lemmas and replaces the closed-domain step by the nilpotent-thickening test and Stacks Tag 00EH.

2.28 Finite modules and Laurent descent

Tate exactness concerns the localizations of a module that is already defined over the original algebra. For coherence we must also construct a global module from finite modules given separately on a covering. The additional analytic step is a bounded lifting argument for arbitrary finite Laurent gluing data. We then use maximal residue fields to obtain generators and apply the lifting argument once more to the relations. This reconstructs the method of Kiehl’s theorem as developed in [BGR84] (§9.4.3, Lemmas 5–6, pp. 378–382), with the finite-module norms and the restricted site specified explicitly.

Unless an additional hypothesis is stated, KK is a complete real-valued non-Archimedean field, its valuation may be trivial, and AA is a strict KK-affinoid algebra with a norm-admissible Banach norm. Relation ideals are actual ideals; their closedness and the canonical norms of finite modules are supplied by Theorem 1.11.5. The zero algebra and its zero module are allowed separately. Radii belong to ∣K×∣|K^\times| and can be scaled to one. Ordinary tensor products in this section are algebraic tensor products. A completed-tensor assertion is made only after its finite-module comparison has been established.

Lemma 2.28.1. NAG-001519 (Maximal fibers of a rational localization). Let g,f1,…,fsg,f_1,\ldots,f_s generate the unit ideal of AA, put B=A{T1,…,Ts}/(gTj−fj)j,V={x∈M(A):∣fj(x)∣≤∣g(x)∣ for all j}.\begin{aligned} B&=A\{T_1,\ldots,T_s\}/(gT_j-f_j)_j,\\ V&=\{x\in\mathcal M(A): |f_j(x)|\leq|g(x)|\text{ for all }j\}. \end{aligned} For a maximal ideal m⊂A\mathfrak m\subset A, let L=A/mL=A/\mathfrak m and let xmx_{\mathfrak m} be its bounded residue-field seminorm. There is a canonical bounded isomorphism B/mB ≃ B⊗AL ≃ {L,xm∈V,0,xm∉V.B/\mathfrak m B\ \simeq\ B\otimes_A L\ \simeq\ \begin{cases} L,&x_{\mathfrak m}\in V,\\ 0,&x_{\mathfrak m}\notin V. \end{cases} The nonzero isomorphism has bounded inverse for canonical norms and is evaluation at fj(xm)/g(xm)f_j(x_{\mathfrak m})/g(x_{\mathfrak m}). Every maximal ideal n⊂B\mathfrak n\subset B is of the form mB\mathfrak mB for a unique maximal ideal m⊂A\mathfrak m\subset A with xm∈Vx_{\mathfrak m}\in V, and A/m→B/nA/\mathfrak m\to B/\mathfrak n is an isomorphism. These fiber identifications commute with rational restriction maps and with finite intersections.

Proof. Theorems 1.12.3 and 2.17.2 identify the fiber, with its canonical norm, with L{T1,…,Ts}/(gˉTj−fˉj)j.L\{T_1,\ldots,T_s\}/(\bar gT_j-\bar f_j)_j. The coefficient map is the quotient A→LA\to L, so this identification respects constants and all displayed variables. The field LL is finite over KK. By Lemma 1.12.1, its extending absolute value is equivalent to its canonical quotient norm.

Suppose first that xm∈Vx_{\mathfrak m}\in V. The unit-ideal relation forces gˉ≠0\bar g\ne 0. Set bj=fˉj/gˉb_j=\bar f_j/\bar g. Since ∣bj∣L≤1|b_j|_L\leq 1, the sequences (bjn)n≥0(b_j^n)_{n\geq 0} are bounded for the canonical norm on LL. Proposition 2.19.1 therefore permits substitution Tj↦bjT_j\mapsto b_j. Because gˉ\bar g is a unit, the relation ideal equals (Tj−bj)j(T_j-b_j)_j. Substitution and the inclusion of constants are bounded inverse maps, by polynomial density. This proves the nonzero case without identifying the canonical norm with the field absolute value.

If xm∉Vx_{\mathfrak m}\notin V and gˉ=0\bar g=0, at least one fˉj\bar f_j is nonzero, so the relation ideal contains a unit. Otherwise some ∣fˉj/gˉ∣L>1|\bar f_j/\bar g|_L>1. Put λ=gˉ/fˉj\lambda=\bar g/\bar f_j. The norm equivalence gives ∥λn∥≤C∣λ∣Ln→0\|\lambda^n\|\leq C|\lambda|_L^n\to 0, so ∑n≥0λnTjn\sum_{n\geq 0}\lambda^nT_j^n belongs to L{T1,…,Ts}L\{T_1,\ldots,T_s\} and is the inverse of 1−λTj1-\lambda T_j, as is checked by its finite partial sums. The relation gˉTj−fˉj=−fˉj(1−λTj)\bar gT_j-\bar f_j=-\bar f_j(1-\lambda T_j) is thus a unit. This proves that the actual fiber algebra is zero.

Now take a maximal ideal n\mathfrak n of BB. Its residue field EE is finite over KK by Theorem 1.12.3. The image of AA in EE is a finite-dimensional KK-domain and hence a field by [StacksAlg] (Tag 00GS, Lemma 10.36.19(1)). Thus m=n∩A\mathfrak m=\mathfrak n\cap A is maximal. The absolute value on this subfield is the unique extending one by Lemma 1.12.1. The bounded point of BB defined by EE restricts to xmx_{\mathfrak m}, which belongs to VV by Theorem 2.19.3. The already computed fiber B/mB=LB/\mathfrak mB=L has no proper nonzero ideal, and its quotient B/nB/\mathfrak n is nonzero. Consequently n=mB\mathfrak n=\mathfrak mB and E=LE=L. Conversely, for xm∈Vx_{\mathfrak m}\in V, the computed fiber shows that mB\mathfrak mB is maximal. All nonzero fiber maps fix LL and substitute the prescribed ratios; maps involving a zero fiber have their unique possible value. This proves compatibility with the restriction and joint-presentation maps of Proposition 2.22.6. ◻

Corollary 2.28.2. NAG-001520 (Algebraic faithful flatness of a finite rational covering). If V1,…,VtV_1,\ldots,V_t are rational subsets covering M(A)\mathcal M(A), with their strict value-group presentation algebras BiB_i, then A→∏i=1tBiA\to\prod_{i=1}^t B_i is a faithfully flat ring map. Faithfulness here tests arbitrary algebraic AA-modules. It does not identify Bi⊗ABjB_i\otimes_A B_j with the completed intersection algebra.

Proof. Lemma 2.22.1 gives algebraic flatness of each BiB_i. To spell out its trivial-valuation branch, choose a defining strict quotient norm on AA. Every nonzero element then has norm one, so every coefficient of a unit certificate has norm at most one; the scalar in that lemma can be taken to be a=1a=1. Its elementary chain is therefore available, and Corollary 2.20.3, which allows trivial valuation, proves the same algebraic flatness. A finite product is a finite direct sum as an AA-module, so tensoring by it preserves exactness. For every maximal m⊂A\mathfrak m\subset A, its residue point belongs to some ViV_i. Lemma 2.28.1 then gives Bi/mBi=A/m≠0B_i/\mathfrak mB_i=A/\mathfrak m\ne 0. The product therefore has nonzero fiber at every maximal ideal. Apply [StacksAlg] (Tag 00HP, Lemma 10.39.15, equivalence of (1) and (4)). If A=0A=0, all modules under consideration are zero and the assertion has its usual degenerate meaning. ◻

Definition 2.28.3. NAG-001521 (Finite Laurent gluing data). For the next three results fix f∈Af\in A and use the following actual quotient algebras with their quotient norms: A+=A{T}/(T−f),A−=A{U}/(1−fU),A0=A{T,U}/(T−f,TU−1).\begin{aligned} A_+&=A\{T\}/(T-f),& A_-&=A\{U\}/(1-fU),\\ A_0&=A\{T,U\}/(T-f,TU-1). \end{aligned} A finite Laurent gluing datum means finite modules N+N_+ over A+A_+ and N−N_- over A−A_-, together with an A0A_0-linear isomorphism between A0⊗A+N+A_0\otimes_{A_+}N_+ and A0⊗A−N−A_0\otimes_{A_-}N_-. Denote the common module by N0N_0, the two maps by ρ+,ρ−\rho_+,\rho_-, and their difference by δ\delta. Equip the finite modules with canonical norms. A morphism of such data is a pair of local module maps commuting with the given overlap isomorphisms.

The gluing isomorphism and its inverse are bounded by Theorem 1.11.5. Thus boundedness follows from the finite canonical module structure.

Lemma 2.28.4. NAG-001522 (Bounded lifting for arbitrary finite Laurent data). For every finite Laurent gluing datum, the map δ:N+⊕N−⟶N0,(v,w)⟼ρ+(v)−ρ−(w)\delta:N_+\oplus N_-\longrightarrow N_0, \qquad (v,w)\longmapsto\rho_+(v)-\rho_-(w) is surjective and norm-strict. More precisely, there is a constant C>0C>0, depending on the datum and the chosen canonical norms, such that every z∈N0z\in N_0 has a preimage (v,w)(v,w) with max⁡(∥v∥,∥w∥)≤C∥z∥\max(\|v\|,\|w\|)\leq C\|z\|. No linear choice of preimages is asserted.

Proof. First, the image of A−A_- in A0A_0 is dense. The identity TU−1=(T−f)U+(fU−1)TU-1=(T-f)U+(fU-1) shows that the ideals (T−f,TU−1)(T-f,TU-1) and (T−f,fU−1)(T-f,fU-1) coincide. Lemma 2.18.1 therefore identifies A0A_0 with A−{T}/(T−f)A_-\{T\}/(T-f). Polynomials in TT are dense before taking this quotient by Lemma 2.13.1, and their classes belong to the image of A−A_- because T=fT=f there. Second, Theorem 2.18.3 for M=AM=A gives the following scalar estimate: for each fixed β>1\beta>1 and every a∈A0a\in A_0, one can write a=b∣0−d∣0,b∈A+, d∈A−,max⁡(∥b∥,∥d∥)≤β∥a∥.a=b|_0-d|_0,\qquad b\in A_+,\ d\in A_-,\qquad \max(\|b\|,\|d\|)\leq\beta\|a\|. If N0=0N_0=0, the conclusion is immediate. Otherwise choose generators e1,…,eme_1,\ldots,e_m of N+N_+ and w1,…,wnw_1,\ldots,w_n of N−N_-. In this proof use their quotient norms, and give N0N_0 the quotient norm from the generators eˉi=ρ+(ei)\bar e_i=\rho_+(e_i). Changing back to any stipulated canonical norms only changes constants. The elements wˉj=ρ−(wj)\bar w_j=\rho_-(w_j) also generate N0N_0. Choose cji∈A0c_{ji}\in A_0 such that eˉi=∑j=1ncjiwˉj,H=max⁡(1,∥wˉ1∥,…,∥wˉn∥).\bar e_i=\sum_{j=1}^n c_{ji}\bar w_j, \qquad H=\max(1,\|\bar w_1\|,\ldots,\|\bar w_n\|). Fix 0<η<10<\eta<1. By density choose cji′∈A−c'_{ji}\in A_- whose images in A0A_0 satisfy max⁡j,i∥cji−cji′∣0∥≤ηβ2H,C0=β2max⁡(1,max⁡j,i∥cji′∥).\max_{j,i}\|c_{ji}-c'_{ji}|_0\| \leq\frac{\eta}{\beta^2H}, \qquad C_0=\beta^2\max(1,\max_{j,i}\|c'_{ji}\|). These are finitely many choices, made once for the datum. For z∈N0z\in N_0, choose ai∈A0a_i\in A_0 with z=∑iaieˉiz=\sum_i a_i\bar e_i and max⁡i∥ai∥≤β∥z∥\max_i\|a_i\|\leq\beta\|z\|; for z=0z=0 take all coefficients zero. Split each ai=bi∣0−di∣0a_i=b_i|_0-d_i|_0 by the scalar estimate and put v=∑ibiei,w=∑i,jdicji′wj.v=\sum_i b_i e_i,\qquad w=\sum_{i,j}d_i c'_{ji}w_j. The quotient module norms satisfy ∥av∥≤∥a∥∥v∥\|av\|\leq\|a\|\|v\|. Hence max⁡(∥v∥,∥w∥)≤C0∥z∥,z−δ(v,w)=−∑i,jdi∣0 (cji−cji′∣0)wˉj,\max(\|v\|,\|w\|)\leq C_0\|z\|, \qquad z-\delta(v,w) =-\sum_{i,j}d_i|_0\,(c_{ji}-c'_{ji}|_0)\bar w_j, and the last expression has norm at most β2∥z∥⋅η/(β2H)⋅H=η∥z∥\beta^2\|z\|\cdot\eta/(\beta^2H)\cdot H=\eta\|z\|. We have therefore obtained a uniformly bounded approximate lift with a fixed contraction of the error.

Apply this construction successively to z0=zz_0=z and zq+1=zq−δ(vq,wq)z_{q+1}=z_q-\delta(v_q,w_q). Then ∥zq∥≤ηq∥z∥\|z_q\|\leq\eta^q\|z\| and max⁡(∥vq∥,∥wq∥)≤C0ηq∥z∥\max(\|v_q\|,\|w_q\|)\leq C_0\eta^q\|z\|. Completeness of the finite canonical modules, from Theorem 1.11.5, implies convergence of v=∑q≥0vqv=\sum_{q\geq 0}v_q and w=∑q≥0wqw=\sum_{q\geq 0}w_q. Boundedness of δ\delta gives δ(v,w)=z\delta(v,w)=z, and the ultrametric inequality gives max⁡(∥v∥,∥w∥)≤C0∥z∥\max(\|v\|,\|w\|)\leq C_0\|z\|. The kernel of the bounded map δ\delta is closed. The lift bound and boundedness of δ\delta identify its quotient norm with the target norm up to fixed constants, proving norm-strictness. The argument also works for trivial valuation; neither small scalars nor an open mapping theorem have been used. ◻

Lemma 2.28.5. NAG-001523 (Global matching elements generate the local modules). For a finite Laurent gluing datum set N=ker⁡δN=\ker\delta. The image of NN generates N+N_+ over A+A_+ and generates N−N_- over A−A_-. Consequently a single finite list of elements of NN generates both local modules after restriction. Finiteness of NN over AA is not assumed here.

Proof. Fix a maximal ideal m\mathfrak m of AA. Put Iτ=mNτI_\tau=\mathfrak mN_\tau and Qτ=Nτ/IτQ_\tau=N_\tau/I_\tau for τ=+,−,0\tau=+,-,0. These modules are finite by Theorem 1.11.5. The maps A+→A0A_+\to A_0 and A−→A0A_-\to A_0 are elementary localizations: they adjoin respectively f−1f^{-1} and a radius-one variable equal to ff. Their algebraic flatness follows from Theorem 2.20.2 and Lemma 2.18.1. Tensoring I±↪N±I_\pm\hookrightarrow N_\pm therefore identifies its image with I0I_0, and right exactness identifies the quotient with Q0Q_0. Thus I±I_\pm and Q±Q_\pm are themselves finite Laurent gluing data.

Lemma 2.28.4 applied to II shows that N⟶ker⁡(Q+⊕Q−⟶Q0)N\longrightarrow \ker(Q_+\oplus Q_-\longrightarrow Q_0) is surjective. Indeed, lift a matching pair in Q+⊕Q−Q_+\oplus Q_- to N+⊕N−N_+\oplus N_-. Its discrepancy lies in I0I_0; subtract a preimage of this discrepancy from I+⊕I−I_+\oplus I_- to obtain a matching lift. This is an algebraic lifting argument using the proved surjectivity, not a tensor product of an unproved exact sequence.

Let x=xmx=x_{\mathfrak m} and L=A/mL=A/\mathfrak m. The two Laurent subsets are V+={∣f∣≤1}V_+=\{|f|\leq 1\} and V−={∣f∣≥1}V_-=\{|f|\geq 1\}. Lemma 2.28.1 computes the fibers of A+,A−,A0A_+,A_-,A_0 as LL or zero according as xx belongs to V+,V−V_+,V_-, or their intersection. If xx belongs to only one member, the displayed quotient equalizer is exactly its corresponding QτQ_\tau. If xx belongs to both, the two maps from Q±Q_\pm to Q0Q_0 are isomorphisms and the same conclusion holds for either projection. Consequently N⟶Nτ/mNτis surjective whenever xm∈Vτ.N\longrightarrow N_\tau/\mathfrak mN_\tau \quad\text{is surjective whenever }x_{\mathfrak m}\in V_\tau.

Let Pτ⊆NτP_\tau\subseteq N_\tau be the AτA_\tau-submodule generated by the image of NN, and set Cτ=Nτ/PτC_\tau=N_\tau/P_\tau. This quotient is finite. Every maximal ideal n\mathfrak n of AτA_\tau has the form mAτ\mathfrak mA_\tau with xm∈Vτx_{\mathfrak m}\in V_\tau, by Lemma 2.28.1. The surjectivity just proved gives Cτ/nCτ=0C_\tau/\mathfrak nC_\tau=0. After localization at n\mathfrak n, Nakayama’s lemma, [StacksAlg] (Tag 00DV, Lemma 10.20.1(2)), gives (Cτ)n=0(C_\tau)_{\mathfrak n}=0. The maximal-local zero test, [StacksAlg] (Tag 00HN, Lemma 10.23.1(2)), then gives Cτ=0C_\tau=0. Thus Pτ=NτP_\tau=N_\tau. Each of the two finite modules has a finite generating list, and each generator is a finite linear combination of images of elements of NN. The union of the finitely many elements of NN so obtained is the required common list. Zero local algebras contribute no condition. ◻

Theorem 2.28.6. NAG-001524 (Effective descent of finite modules for a Laurent pair). The functor from finite AA-modules to finite Laurent gluing data, given by ordinary base change to A+A_+ and A−A_-, is an equivalence of categories. Its inverse sends the datum to N=ker⁡δN=\ker\delta. This NN is a finite AA-module, and its norm induced from N+⊕N−N_+\oplus N_- is uniformly equivalent to its canonical finite-module norm. The canonical maps Aτ⊗AN⟶Nτ(τ=+,−,0)A_\tau\otimes_A N\longrightarrow N_\tau\qquad(\tau=+,-,0) are bounded isomorphisms with bounded inverses, and the same assertions hold for separated completed tensor products. In particular, the sequence 0→N→N+⊕N−→δN0→00\to N\to N_+\oplus N_-\xrightarrow{\delta}N_0\to 0 is norm-strict exact. All assertions extend to a radius r∈∣K×∣r\in|K^\times| by scalar rescaling.

Proof. Choose n1,…,nr∈Nn_1,\ldots,n_r\in N as in Lemma 2.28.5. They give surjections α±:A±r→N±\alpha_\pm:A_\pm^r\to N_\pm and a common surjection α0:A0r→N0\alpha_0:A_0^r\to N_0. Put Jτ=ker⁡ατJ_\tau=\ker\alpha_\tau. These kernels are finite canonical modules; give them the norms induced from the corresponding free modules, which are canonical by Theorem 1.11.5. Algebraic flatness of A±→A0A_\pm\to A_0, as verified in the preceding proof, identifies A0⊗A±J±A_0\otimes_{A_\pm}J_\pm with J0J_0. Thus the JτJ_\tau form a finite Laurent gluing datum.

We prove that the generator map α:Ar→N\alpha:A^r\to N is surjective with a uniform lift bound. For n=(n+,n−)∈Nn=(n_+,n_-)\in N, norm-strictness of α±\alpha_\pm gives lifts u±∈A±ru_\pm\in A_\pm^r with max⁡(∥u+∥,∥u−∥)≤C1∥n∥eq,∥n∥eq=max⁡(∥n+∥,∥n−∥).\max(\|u_+\|,\|u_-\|)\leq C_1\|n\|_{\rm eq}, \qquad \|n\|_{\rm eq}=\max(\|n_+\|,\|n_-\|). The difference z=u+∣0−u−∣0z=u_+|_0-u_-|_0 lies in J0J_0 and has its induced norm at most C1∥n∥eqC_1\|n\|_{\rm eq}. Apply Lemma 2.28.4 to JJ: there are j±∈J±j_\pm\in J_\pm with j+∣0−j−∣0=zj_+|_0-j_-|_0=z and max⁡(∥j+∥,∥j−∥)≤C2∥z∥\max(\|j_+\|,\|j_-\|)\leq C_2\|z\|. The corrected free vectors u±−j±u_\pm-j_\pm match. Theorem 2.18.3 for M=ArM=A^r now gives a unique a∈Ara\in A^r with those restrictions, and its augmentation estimate gives α(a)=n,∥a∥≤C3max⁡(C1,C1C2)∥n∥eq.\alpha(a)=n, \qquad \|a\|\leq C_3\max(C_1,C_1C_2)\|n\|_{\rm eq}. Therefore NN is finite over AA. Conversely, the fixed images of the standard basis of ArA^r give ∥α(a)∥eq≤C4∥a∥\|\alpha(a)\|_{\rm eq}\leq C_4\|a\|. These two inequalities show that the quotient norm from ArA^r and ∥⋅∥eq\|\cdot\|_{\rm eq} are uniformly equivalent. This verifies canonicality of the equalizer norm by explicit lifts; it has not been inferred merely from completeness of an arbitrary norm on a finite module.

Let J=ker⁡(Ar→αN)J=\ker(A^r\xrightarrow{\alpha}N). The free Laurent equalizer identifies JJ with ker⁡(J+⊕J−→J0)\ker(J_+\oplus J_-\to J_0). Lemma 2.28.5, now applied to this kernel datum, says that A±J=J±A_\pm J=J_\pm inside A±rA_\pm^r. Ordinary right exactness of tensor products, [StacksAlg] (Tag 00DF, Lemma 10.12.10), therefore identifies both A±⊗ANA_\pm\otimes_A N and N±N_\pm with A±r/J±A_\pm^r/J_\pm. The same conclusion for A0A_0 follows by tensor associativity, [StacksAlg] (Tag 00D2, Lemma 10.12.7). These canonical algebraic isomorphisms are bounded with bounded inverses by Theorem 1.11.5. Proposition 1.11.6 then gives the completed-tensor statements. No exactness claim for the completed tensor product of arbitrary Banach modules occurs.

A compatible pair of morphisms of gluing data preserves the two equalizers and hence gives an AA-linear map between the descended modules. The canonical base-change isomorphisms recover the original morphisms. Conversely, starting from a finite AA-module, Theorem 2.18.3 identifies its equalizer with that same module. These statements verify that the two functors are mutually inverse, including on morphisms. The equalizer is closed, its norm is canonical by the preceding estimates, and the last map is norm-strict by Lemma 2.28.4; thus the asserted sequence is norm-strict exact. For r=∣c∣r=|c| with c∈K×c\in K^\times, replace ff by f/cf/c, TT by T/cT/c and UU by cUcU. This is the isometric radius normalization of the displayed presentations and proves the last assertion. ◻

Keeping the refinement inside the fixed rational site

Write X=M(A)X=\mathcal M(A), and use exactly the site RX\mathcal R_X of Definition 2.23.1: its objects are rational subsets presented over the original AA, and its coverings are finite set-theoretic coverings by such objects. In particular, the algebra of an object VV is denoted by AVA_V.

Definition 2.28.7. NAG-001526 (A finite Laurent tree in the fixed site). A finite Laurent tree in this site means a finite rooted binary tree with root XX, such that each node VV is an object of RX\mathcal R_X and each nonterminal node is covered by an elementary Laurent pair for some h∈AVh\in A_V and radius in ∣K×∣|K^\times|. Its two child algebras and their intersection algebra must be the corresponding actual elementary quotients, up to the canonical bounded isomorphisms. Empty nodes may be terminal. This definition imposes the presentation requirement; it does not assert its existence.

Proposition 2.28.8. NAG-001527 (A Laurent refinement tree over the original algebra). Assume that the valuation of KK is nontrivial. Every finite covering of XX by objects of RX\mathcal R_X has a subordinate finite Laurent tree in the sense just specified: each leaf is contained in one covering member. All nodes, including those obtained by comparing ratios in intermediate algebras, remain rational over the original AA. The child and intersection maps agree with the actual elementary quotient maps.

Proof. If A=0A=0, use the single empty node. Otherwise Proposition 2.22.8 gives a standard rational refinement generated by a unit-ideal tuple f0,…,fs∈Af_0,\ldots,f_s\in A. Choose a unit certificate ∑iaifi=1\sum_i a_if_i=1 and κ∈K×\kappa\in K^\times with ∣κ∣max⁡i∥ai∥<1|\kappa|\max_i\|a_i\|<1. Split successively by the elementary Laurent pairs for f0/κ,…,fs/κf_0/\kappa,\ldots,f_s/\kappa. Each partial branch is an intersection of rational subsets over AA: its positive condition has denominator 11, and its negative condition has tuple (fi;κ)(f_i;\kappa), which generates the unit ideal because κ≠0\kappa\ne 0. Proposition 2.22.6 and Lemma 2.18.1 identify its algebra with the simultaneous relation quotient. Theorem 2.19.3 identifies each binary pair with a covering of its parent.

Consider a completed first-stage branch with algebra CC. Let PP and NN be its positive and negative index sets. Lemma 2.22.4 says that C=0C=0 when NN is empty. For i∈Ni\in N, the element fif_i is a unit in CC and ∥κ/fi∥≤1\|\kappa/f_i\|\leq 1. For a nonzero branch, split further, successively, by the finitely many ratios qij=fi/fjq_{ij}=f_i/f_j with i<ji<j in NN. We verify that this second-stage operation stays in RX\mathcal R_X, rather than assuming rational transitivity.

The two rational subsets of the original XX Wij+={∣fi∣≤∣fj∣, ∣κ∣≤∣fj∣},Wij−={∣fj∣≤∣fi∣, ∣κ∣≤∣fi∣}\begin{aligned} W_{ij}^+&=\{|f_i|\leq|f_j|,\ |\kappa|\leq|f_j|\},\\ W_{ij}^-&=\{|f_j|\leq|f_i|,\ |\kappa|\leq|f_i|\} \end{aligned} have unit-ideal tuples (fj;fi,κ)(f_j;f_i,\kappa) and (fi;fj,κ)(f_i;f_j,\kappa). Let DD be CC or any partial second-stage branch. By induction its quotient map from CC is contractive, so both fi,fjf_i,f_j remain units and the bounds ∥κ/fi∥,∥κ/fj∥≤1\|\kappa/f_i\|,\|\kappa/f_j\|\leq 1 persist. Over DD, intersection with Wij+W_{ij}^+ has the joint presentation D{Z,Y}/(fjZ−fi,fjY−κ)≃D{Z,Y}/(Z−qij,Y−κ/fj)≃D{Z}/(Z−qij).D\{Z,Y\}/(f_jZ-f_i,f_jY-\kappa) \simeq D\{Z,Y\}/(Z-q_{ij},Y-\kappa/f_j) \simeq D\{Z\}/(Z-q_{ij}). The first equality uses a unit and does not change the actual ideal. In the second, inclusion and substitution Y=κ/fjY=\kappa/f_j are inverse contractions by Lemma 2.22.2 and the iterated-quotient lemma. Thus this is the positive elementary quotient with its actual quotient norm. The negative intersection is similarly D{Z}/(Z−fj/fi)D\{Z\}/(Z-f_j/f_i), whose actual ideal equals (1−qijZ)(1-q_{ij}Z) because qijq_{ij} is a unit. Imposing both signs gives the same joint presentation as their elementary intersection. All maps fix DD and the indicated ratios, so these isometries commute with restriction. This proves the inductive presentation requirement for every partial branch. The subsets cover their parent because any two nonnegative real numbers are comparable and the scalar lower bounds already hold there.

At a second-stage leaf, Lemma 2.22.3 chooses k∈Nk\in N with ∥fi/fk∥≤1\|f_i/f_k\|\leq 1 for every i∈Ni\in N. For p∈Pp\in P, write fp=κXpf_p=\kappa X_p and κ/fk=Xk\kappa/f_k=X_k from the first-stage relations; then fp/fk=XpXkf_p/f_k=X_pX_k also has norm at most one. Hence the leaf factors contractively through the kkth member of the standard rational family. That member is contained in an original covering member. The construction has finitely many stages and finitely many branches. Empty branches use the zero algebra, as follows either from their explicit relation or from Theorem 1.12.3. This completes the required finite tree. ◻

Definition 2.28.9. NAG-001528 (A finite module model on a rational slice). For an object V∈RXV\in\mathcal R_X, denote the slice site by RX/V\mathcal R_X/V and its restricted ring sheaf by OV\mathcal O_V. If PP is a finite AVA_V-module, put P~V(W)=AW⊗AVP(W⊆V,W∈RX).\widetilde P_V(W)=A_W\otimes_{A_V}P\qquad(W\subseteq V, W\in\mathcal R_X). A finite module model for a sheaf F\mathcal F on VV means an isomorphism P~V≃F∣RX/V\widetilde P_V\simeq\mathcal F|_{\mathcal R_X/V} for such a PP. It includes all objects of the slice site, not just the value at VV.

Lemma 2.28.10. NAG-001529 (Finite-module sheaves on a rational slice). The assignment P↦P~VP\mapsto\widetilde P_V is an exact fully faithful functor from finite AVA_V-modules to OV\mathcal O_V-modules, and Γ(V,P~V)=P\Gamma(V,\widetilde P_V)=P. It commutes with restriction to W⊆VW\subseteq V by ordinary tensor associativity and canonical finite-module tensor comparison. Every finite covering of an object of RX/V\mathcal R_X/V has a norm-strict exact augmented Čech complex for P~V\widetilde P_V.

Proof. If W⊆VW\subseteq V are rational over AA, Proposition 2.22.6 and Theorem 2.22.5 identify the joint algebra AV⊗^AAWA_V\widehat\otimes_A A_W with AWA_W. Viewed over AVA_V, the original unit-ideal tuple presenting WW is still a unit-ideal tuple. Thus AWA_W is a rational quotient over AVA_V; this direction of the comparison does not require every rational subset over AVA_V to be rational over AA. Every covering in the slice is consequently a finite rational covering over the algebra of its target. Apply Theorem 2.22.9 to that target algebra and its finite module. Tensor associativity, [StacksAlg] (Tag 00D2, Lemma 10.12.7), and Theorem 2.17.2 identify all terms and arrows with the displayed presheaf. Degree-zero exactness proves the sheaf condition, including the value zero on the empty object, and the full theorem gives the asserted norm-strict complex.

Rational flatness from Lemma 2.22.1 proves objectwise exactness for short exact sequences of finite AVA_V-modules, and hence exactness as sheaves. For trivial valuation use a defining discrete quotient norm on AVA_V and a=1a=1 in that lemma’s elementary chain, exactly as in the proof of Corollary 2.23.2; Corollary 2.20.3 applies in this case as well. Evaluation at VV gives AV⊗AVP=PA_V\otimes_{A_V}P=P. A sheaf map is determined by its value at VV: its value on WW must send b⊗pb\otimes p to bb times the restriction of the image of pp. Conversely, every AVA_V-linear map has these compatible values. This proves full faithfulness. The same formula and ordinary tensor associativity prove compatibility with further restriction; Proposition 1.11.6 supplies the corresponding bounded finite-module identifications. ◻

Proposition 2.28.11. NAG-001530 (Descent along a finite Laurent tree). Let F\mathcal F be an OXrat\mathcal O_X^{\mathrm{rat}}-module sheaf, and suppose there is a finite Laurent tree in RX\mathcal R_X on whose leaves F\mathcal F has finite module models. Then M=F(X)M=\mathcal F(X) is a finite AA-module and the canonical morphism M~rat→F\widetilde M^{\mathrm{rat}}\to\mathcal F is an isomorphism. With any canonical norms on the finite leaf models, the norm of a global matching section obtained by taking the maximum of its leaf restrictions is uniformly equivalent to the canonical norm on MM.

Proof. Proceed from leaves towards the root. A zero node has only the zero sheaf and zero module, by the empty covering. Consider a parent VV whose two children V+,V−V_+,V_- already have finite models P+,P−P_+,P_-. Their common restriction to V0=V+∩V−V_0=V_+\cap V_- gives an isomorphism AV0⊗AV+P+ ≃ AV0⊗AV−P−.A_{V_0}\otimes_{A_{V_+}}P_+ \ \simeq\ A_{V_0}\otimes_{A_{V_-}}P_-. This is an isomorphism of finite modules: evaluate the two sheaf models on V0V_0 and use Lemma 2.28.10. The tree’s presentation condition therefore makes them a finite Laurent gluing datum over AVA_V. Theorem 2.28.6 produces a finite AVA_V-module PP whose two base changes recover the children. The sheaf equalizer identifies PP with F(V)\mathcal F(V).

For every W⊆VW\subseteq V in the slice, multiplication of restricted sections defines AW⊗AVP→F(W)A_W\otimes_{A_V}P\to\mathcal F(W). This gives a morphism of sheaves. On each child it is the base-change isomorphism just proved. For arbitrary WW, pull back the binary covering to W∩V+,W∩V−W\cap V_+,W\cap V_-. The morphism is an isomorphism on both members and their intersection, so the two sheaf equalizer diagrams show it is an isomorphism on WW. This proves the required model on all of RX/V\mathcal R_X/V, rather than only at the parent object.

The norm comparison at each parent is the explicit equalizer comparison in Theorem 2.28.6. Composing these finitely many two-sided inequalities gives the maximum-of-leaves norm comparison at the root. All constants are finite and may depend on the chosen tree, models and canonical norms. No assertion of a uniform constant over all coverings is made. The finite induction ends at XX, proving both conclusions. ◻

Theorem 2.28.12. NAG-001531 (Kiehl equivalence on the fixed rational site). Let KK be complete, with trivial valuation allowed, and let AA be strict norm-admissible. On the ringed site (RX,OXrat)(\mathcal R_X,\mathcal O_X^{\mathrm{rat}}), the following conditions on an OXrat\mathcal O_X^{\mathrm{rat}}-module sheaf F\mathcal F are equivalent:

  1. F\mathcal F is coherent in the ringed-site sense of [StacksAlg] (Tag 03DL, Definition 18.23.1(8)): it is of finite type, and on each slice the kernel of every map from a finite free module to it is of finite type.

  2. F\mathcal F is of finite presentation in the sense of the same definition, part (7).

  3. Some finite covering of XX in RX\mathcal R_X admits finite module models for F\mathcal F on every member.

  4. There is a finite AA-module MM with F≃M~rat\mathcal F\simeq\widetilde M^{\mathrm{rat}}.

Consequently M↦M~ratM\mapsto\widetilde M^{\mathrm{rat}} is an exact equivalence from finite AA-modules to coherent modules on this ringed site, with inverse F↦Γ(X,F)\mathcal F\mapsto\Gamma(X,\mathcal F). Every such sheaf has a global finite presentation by finite free OXrat\mathcal O_X^{\mathrm{rat}}-modules. For any finite covering in this site, its augmented Čech complex is norm-strict exact with the induced canonical finite-module norms. This is the stated fixed-rational-site result; no identification with the full analytic G-site or with all generalized affinoid domains is included.

Proof. We first compare the local conditions without using descent. If F\mathcal F is coherent, finite type gives a finite covering of XX on whose members there are surjections OVn→F∣V\mathcal O_V^n\to\mathcal F|_V. By coherence each kernel is of finite type. A further finite covering supplies finitely many generators for that kernel, hence presentations OWm→OWn→F∣W→0\mathcal O_W^m\to\mathcal O_W^n\to\mathcal F|_W\to 0. The finite composite covering proves (2). More generally, covers of any object are obtained by restriction of these covers, so this is the ringed-site definition, not only a condition at XX.

For (2)⇒\Rightarrow(3), take a finite covering with the displayed finite presentations. A morphism between finite free sheaves on a slice is a matrix with entries in OW(W)=AW\mathcal O_W(W)=A_W: its columns are the images of the standard global basis sections. Let PP be the cokernel of this matrix over AWA_W. It is finite. By exactness of Lemma 2.28.10, the sheaf cokernel is precisely P~W\widetilde P_W. Thus these are the required finite models.

Assume (3), first with nontrivial valuation. Proposition 2.28.8 gives a subordinate finite Laurent tree whose every node remains in RX\mathcal R_X. A leaf contained in a model object inherits its finite model by Lemma 2.28.10. Proposition 2.28.11 then gives F≃M~rat\mathcal F\simeq\widetilde M^{\mathrm{rat}} with M=F(X)M=\mathcal F(X) finite. This proves (4) in this valuation case.

For trivial valuation choose a defining strict quotient norm. The finite-tail condition makes each unit-radius Tate algebra an ordinary polynomial algebra, and each nonzero quotient element has norm one. For a covering member with rational tuple (gi;f1(i),…,fsi(i))(g_i;f^{(i)}_1,\ldots,f^{(i)}_{s_i}), the unit certificate makes gig_i invertible in its actual relation quotient. Substitution Tj=fj(i)/giT_j=f^{(i)}_j/g_i therefore gives Bi=AVi≃A[1/gi],AVi∩Vj≃A[1/(gigj)]≃Bi⊗ABj,B_i=A_{V_i}\simeq A[1/g_i],\qquad A_{V_i\cap V_j}\simeq A[1/(g_ig_j)]\simeq B_i\otimes_A B_j, and likewise for triple intersections, with the actual restriction maps. This is the discrete-norm computation also used in Theorem 2.22.9; here it verifies ordinary, not merely completed, tensor products. Put S=∏iBiS=\prod_i B_i. Corollary 2.28.2 makes A→SA\to S faithfully flat. The finite local model modules PiP_i give the finite SS-module P=∏iPiP=\prod_iP_i. Their sheaf identifications on pairwise intersections, evaluated there, give an ordinary descent datum for A→SA\to S: the decomposition of S⊗ASS\otimes_A S by its finitely many factor idempotents has components Bi⊗ABjB_i\otimes_A B_j, and the cocycle identity is the identity of the three restriction identifications on each triple intersection.

Apply the authorized algebraic descent theorem [StacksAlg] (Tag 023N, Proposition 35.3.9). It produces an AA-module MM with S⊗AM≃PS\otimes_A M\simeq P, whose restrictions and descent datum recover the given ones. By finite-type descent, [StacksAlg] (Tag 03C4, Lemma 10.83.2(1)), MM is finite. In particular Bi⊗AM≃PiB_i\otimes_A M\simeq P_i. These module isomorphisms give compatible sheaf isomorphisms on the covering slices by Lemma 2.28.10. To see the global isomorphism explicitly, their compatibility identifies each m∈Mm\in M with a matching family in F(Vi)\mathcal F(V_i), hence with a section of F(X)\mathcal F(X). Restriction and multiplication define M~rat→F\widetilde M^{\mathrm{rat}}\to\mathcal F. For every object WW, use the covering (W∩Vi)(W\cap V_i) and the two sheaf equalizer diagrams; the morphism is an isomorphism on all its members and intersections, hence on WW. This proves (4) also for trivial valuation. The use of algebraic faithfully flat descent is confined to the case in which the ordinary tensor intersection was just checked.

Finally suppose (4). The finite AA-module MM is finitely presented by Theorem 1.11.5; applying the exact functor of Corollary 2.23.2 gives a global finite free presentation of its sheaf. In particular the sheaf is of finite type. On any slice RX/V\mathcal R_X/V, it has the finite model MV=AV⊗AMM_V=A_V\otimes_A M. Given any finite list of its sections over VV, the associated map OVn→MV~V\mathcal O_V^n\to\widetilde{M_V}_V is induced by AVn→MVA_V^n\to M_V. Its kernel is the sheaf associated to the algebraic kernel, by Lemma 2.28.10. The algebraic kernel is finite since AVA_V is Noetherian. Its sheaf is generated by a finite list of global sections on that slice. Thus the kernel is of finite type for every such map and every object VV, which is exactly coherence in (1).

Full faithfulness, exactness and the inverse by global sections are now Corollary 2.23.2, together with the essential surjectivity just proved. On each object the section module is finite and therefore has a canonical norm up to uniform equivalence. The norm-strict Čech statement follows from Lemma 2.28.10, or directly from Theorem 2.22.9 after the global model has been obtained. None of these conclusions changes the object category of the site. ◻

References. Kiehl’s original theorem is due to R. Kiehl, Theorem A und Theorem B in der nichtarchimedischen Funktionentheorie, Invent. Math. 2 (1967), 256–273. The proof reconstructed here was checked against the later monograph proof in [BGR84] (§9.4.3, Theorem 3 and Lemmas 5–6, pp. 378–382); no theorem or proof number from the original article is asserted without inspection. The present theorem is an internal proof for the specified strict rational ringed site. Passing to a full affinoid G-site, non-strict radii or general analytic spaces remains a separate comparison problem.

2.29 Strict domains over a trivially valued field

The nontrivial-valuation hypothesis in the local Weierstrass form of Gerritzen–Grauert cannot simply be removed. Over a trivially valued field, strict affinoid algebras carry discrete defining norms. This makes an algebraic route available for finite rational coverings of a domain, but it also makes the stronger ambient local Weierstrass conclusion false. We prove the positive statement and the obstruction separately.

Except in Corollary 2.29.4, KK is complete with its fixed trivial absolute value. Algebras are commutative and unital, and a strict affinoid algebra is equipped with a norm-admissible Banach norm. The term strict-test affinoid domain has exactly the meaning of Definition 2.27.1; its targets are strict KK-affinoid algebras. In the trivial-valuation assertions, all strict rational and Weierstrass radii lie in ∣K×∣={1}|K^\times|=\{1\}. We allow the zero algebra with its usual conventions.

Lemma 2.29.1. NAG-001542 (The discrete algebraic bridge). Every strict KK-affinoid algebra is a finite-type KK-algebra, and each of its defining strict quotient norms is the discrete norm: every nonzero element has norm one. Conversely, every finite-type KK-algebra, equipped with this norm, is strict affinoid. Every KK-algebra homomorphism between such algebras is bounded for their admissible norms and is of finite presentation. If A→BA\to B and A→CA\to C are such maps, the ordinary algebra B⊗ACB\otimes_A C is strict affinoid, and B⊗AC ⟶ B⊗^ACB\otimes_A C\ \longrightarrow\ B\widehat\otimes_A C is a bounded isomorphism. For discrete defining norms, its projective tensor norm is exactly the discrete norm and the map is an isometry.

Proof. A family of coefficients in KK tending to zero has only finitely many nonzero terms. Thus K⟨X1,…,Xn⟩=K[X1,…,Xn]K\langle X_1,\ldots,X_n\rangle=K[X_1,\ldots,X_n], and the Gauss norm is discrete. Every ideal is closed for this norm. In a nonzero quotient class, every representative has norm one, so the quotient norm is one. These observations prove both statements about the algebras and their defining norms. A given norm-admissible norm is uniformly equivalent to a defining norm, so every algebra map, contractive for the discrete norms, remains bounded.

A finite set of KK-algebra generators of BB also generates it over AA. Choose a surjection P=A[X1,…,Xn]→BP=A[X_1,\ldots,X_n]\to B. The algebra PP is itself a finite-type KK-algebra and hence strict. Theorem 1.11.5 makes it Noetherian, so the kernel is finitely generated. This proves finite presentation of the map.

The tensor algebra is a quotient of B⊗KCB\otimes_K C, hence is of finite type over KK. For its projective seminorm, every pure-tensor representation has maximum factor norm at most one. If the represented element is nonzero, at least one term in every representation has both factors nonzero; its factor norm is then one. The infimum over representations is therefore exactly one on every nonzero element. The seminorm is separated, and this discrete norm is complete. Passing to the separated completion changes nothing. Uniformly equivalent norms on B,CB,C give uniformly equivalent projective tensor norms, by applying the two fixed factor bounds to every representation. This proves the assertion for all stipulated norms. ◻

Proposition 2.29.2. NAG-001543 (A strict-test domain gives an étale ring epimorphism). Let φ:A→B\varphi:A\to B present a strict-test affinoid domain U⊆M(A)U\subseteq\mathcal M(A). As a map of abstract rings, φ\varphi is an epimorphism and is formally étale: for every commutative AA-algebra CC, every ideal J⊂CJ\subset C with J2=0J^2=0, and every AA-algebra map B→C/JB\to C/J, there is a unique lift B→CB\to C. No topology or finite-type condition is imposed on CC in this assertion. Consequently A→BA\to B is an étale ring map of finite presentation, and is algebraically flat.

Proof. The strict-test domain property makes A→BA\to B an epimorphism in the strict-affinoid category. Indeed, if u,v:B→Eu,v:B\to E agree on AA, their common composite has spectrum image in UU, and the uniqueness part of the domain property gives u=vu=v. By Lemma 2.29.1, the ordinary algebra D=B⊗ABD=B\otimes_A B is an allowed strict target, and the two maps b↦b⊗1b\mapsto b\otimes 1 and b↦1⊗bb\mapsto 1\otimes b are bounded. They therefore agree. Now let u,v:B→Eu,v:B\to E be any two homomorphisms of abstract commutative rings agreeing on AA. The balanced map D⟶E,b⊗b′⟼u(b)v(b′)D\longrightarrow E,\qquad b\otimes b'\longmapsto u(b)v(b') takes their equality to u(b)=v(b)u(b)=v(b). Thus φ\varphi is an epimorphism against arbitrary ring targets, not just strict targets.

Choose the finite presentation π:P=A[X1,…,Xn]↠B\pi:P=A[X_1,\ldots,X_n]\twoheadrightarrow B supplied by the preceding lemma, put I=ker⁡πI=\ker\pi, and set Q=P/I2Q=P/I^2. The algebra QQ is finite type over KK, so it is an allowed strict test algebra. Every multiplicative seminorm on QQ annihilates I/I2I/I^2, since each element of this ideal has square zero. It therefore factors through B=P/IB=P/I. Boundedness descends by taking the infimum over quotient representatives, or directly from the discrete defining norms. Conversely, a bounded point of BB pulls back to a bounded point of QQ. Consequently the maps A→QA\to Q and A→BA\to B have the same spectrum image.

The domain property now supplies an AA-algebra map σ:B→Q\sigma:B\to Q. If q:Q→Bq:Q\to B is the quotient map, the endomorphisms qσq\sigma and idB\mathrm{id}_B agree on AA. The ring-epimorphism property just proved gives qσ=idBq\sigma=\mathrm{id}_B. This single section is obtained using only a finite-type strict test.

To pass to an arbitrary square-zero diagram, let ψ:B→C/J\psi:B\to C/J be as in the statement. Choose in CC one lift of each of the finitely many elements ψ(π(Xi))\psi(\pi(X_i)). They define an AA-algebra map α:P→C\alpha:P\to C whose reduction modulo JJ is ψπ\psi\pi. Hence α(I)⊂J\alpha(I)\subset J, and α(I2)=0\alpha(I^2)=0. There is an induced map αˉ:Q→C\bar\alpha:Q\to C, and αˉσ\bar\alpha\sigma lifts ψ\psi because qσ=idBq\sigma=\mathrm{id}_B. Any two lifts agree on AA, so the ring-epimorphism property gives uniqueness. This proves the asserted lifting property without assuming that the arbitrary target CC itself is a strict affinoid.

This is formal étaleness in [StacksAlg] (Tag 00UQ, Definition 10.150.1). Finite presentation was proved in Lemma 2.29.1. Apply [StacksAlg] (Tag 00UR, Lemma 10.150.3) to obtain étaleness, and [StacksAlg] (Tag 00U2, Lemma 10.143.3(6)) to obtain algebraic flatness. These are authorized external algebraic inputs, not additional analytic assumptions. ◻

Theorem 2.29.3. NAG-001544 (A finite rational covering in the trivial-valuation case). Let A→BA\to B present a strict-test affinoid domain U⊆M(A)U\subseteq\mathcal M(A). There exist finitely many elements g1,…,gt∈Ag_1,\ldots,g_t\in A such that, with WW the image of the ordinary prime spectrum Spec⁡(B)→Spec⁡(A)\operatorname{Spec}(B)\to\operatorname{Spec}(A), W=⋃i=1tDA(gi),A[1/gi]→ ∼ B[1/φ(gi)],U=⋃i=1tRi,Ri={x∈M(A):∣gi(x)∣=1}.\begin{aligned} W&=\bigcup_{i=1}^t D_A(g_i),& A[1/g_i]&\xrightarrow{\ \sim\ } B[1/\varphi(g_i)],\\ U&=\bigcup_{i=1}^t R_i,& R_i&=\{x\in\mathcal M(A):|g_i(x)|=1\}. \end{aligned} The elements φ(gi)\varphi(g_i) generate the unit ideal of BB. Each RiR_i is rational over the original AA, with strict algebra A⟨Z⟩/(giZ−1)=A[1/gi]A\langle Z\rangle/(g_iZ-1)=A[1/g_i]. The displayed algebra isomorphisms and B⊗^AA[1/gi] ≃ B[1/φ(gi)] ≃ A[1/gi]B\widehat\otimes_A A[1/g_i]\ \simeq\ B[1/\varphi(g_i)] \ \simeq\ A[1/g_i] are bounded with bounded inverses; they are isometric for the discrete defining norms. They commute with restriction to finite intersections, obtained by inverting products of the gig_i. Thus the domain itself has a finite original-AA rational covering with the stated algebra identifications. No local Weierstrass assertion on a covering of the whole ambient spectrum is made. The empty domain is represented by B=0B=0 and uses the empty family.

Proof. If B=0B=0, all assertions have the indicated empty meaning. Suppose B≠0B\ne 0. By Proposition 2.29.2, A→BA\to B is a flat ring map of finite presentation. The authorized openness theorem [StacksAlg] (Tag 00I1, Proposition 10.41.8) therefore makes WW open in Spec⁡(A)\operatorname{Spec}(A). By the definition of the Zariski topology, every point of WW lies in some principal open DA(g)D_A(g) contained in WW: write the closed complement as V(J)V(J) and choose g∈Jg\in J outside that prime.

Let GG be the set of all g∈Ag\in A with DA(g)⊆WD_A(g)\subseteq W. Their images generate the unit ideal in BB. Otherwise that ideal lies in a maximal ideal n\mathfrak n of BB, by [StacksAlg] (Tag 00E0, Lemma 10.17.2(2)) applied to its nonzero quotient. The contraction of n\mathfrak n belongs to WW, and a principal neighbourhood as above supplies g∈Gg\in G whose image is not in n\mathfrak n, a contradiction. Hence there is a finite certificate ∑i=1tbiφ(gi)=1,bi∈B,gi∈G.\sum_{i=1}^t b_i\varphi(g_i)=1,\qquad b_i\in B,\quad g_i\in G. This also proves W=⋃iDA(gi)W=\bigcup_iD_A(g_i). Indeed, choose a prime of BB over any point of WW; at least one term φ(gi)\varphi(g_i) lies outside that prime. The reverse inclusion holds by the choice of GG.

Fix any g∈Gg\in G. The ring map Ag→BgA_g\to B_g, where BgB_g abbreviates B[1/φ(g)]B[1/\varphi(g)], is étale by the base-change assertion [StacksAlg] (Tag 00U2, Lemma 10.143.3(3)), hence flat by part (6). It remains an epimorphism: two maps from BgB_g agreeing on AgA_g agree on BB by Proposition 2.29.2, and then agree on the inverse of φ(g)\varphi(g). Every maximal ideal m\mathfrak m of AgA_g has nonzero fiber in BgB_g. To verify this, contract it to a prime p\mathfrak p of AA. Since g∉pg\notin\mathfrak p and DA(g)⊆WD_A(g)\subseteq W, there is a prime q\mathfrak q of BB over p\mathfrak p. The fraction ideal qBg\mathfrak qB_g is prime and contracts to m\mathfrak m; the fraction description verifies both assertions directly. Its residue field consequently receives a unital map from Bg⊗Agκ(m)B_g\otimes_{A_g}\kappa(\mathfrak m), so this fiber is nonzero.

By [StacksAlg] (Tag 00HP, Lemma 10.39.15, (1)⇔\Leftrightarrow(4)), the map Ag→BgA_g\to B_g is faithfully flat. The authorized epimorphism criterion [StacksAlg] (Tag 04VU, Lemma 10.107.7) makes it an isomorphism. If Ag=0A_g=0, its unital target BgB_g is also zero and the same conclusion holds. Thus the open-immersion consequence is obtained by actual principal-localization isomorphisms; no separate analytic local-ring theorem is involved. All these maps and their inverses are bounded by Lemma 2.29.1.

We now compare the two spectra carefully. For x∈M(A)x\in\mathcal M(A), contractivity for the discrete norm, Lemma 1.6.1, gives ∣a(x)∣≤1|a(x)|\leq 1 for every a∈Aa\in A. The subset rx={a∈A:∣a(x)∣<1}\mathfrak r_x=\{a\in A:|a(x)|<1\} is a prime ideal: the ultrametric inequality and the bound by one give the ideal axioms, while ∣ab(x)∣<1|ab(x)|<1 implies ∣a(x)∣<1|a(x)|<1 or ∣b(x)∣<1|b(x)|<1. It is a proper ideal because ∣1∣=1|1|=1. Hence rx∈DA(g)⟺∣g(x)∣=1.\mathfrak r_x\in D_A(g)\quad\Longleftrightarrow\quad |g(x)|=1. This prime is not in general the kernel of xx. Thus the comparison uses the value-one condition, not merely nonvanishing of gg at xx.

The rational algebra for {∣g∣=1}\{|g|=1\} is A⟨Z⟩/(gZ−1)A\langle Z\rangle/(gZ-1). Its tuple (g;1)(g;1) generates the unit ideal, and the inequality 1≤∣g∣1\leq|g| is equivalent to equality because all values are at most one. By the discrete bridge, this actual quotient is A[1/g]A[1/g]. Theorem 2.19.3 identifies its spectrum with that rational subset. The already constructed bounded map B→Bg≃AgB\to B_g\simeq A_g shows that this subset lies in UU whenever g∈Gg\in G. Conversely, for y∈M(B)y\in\mathcal M(B), the displayed unit certificate and the bounds ∣bi(y)∣≤1|b_i(y)|\leq 1 give ∣φ(gi)(y)∣=1|\varphi(g_i)(y)|=1 for at least one ii. Thus every point of UU lies in some RiR_i. Equivalently, UU is precisely the set of xx for which rx∈W\mathfrak r_x\in W, as follows from the finite principal covering just proved.

Finally, localization gives the ordinary algebra isomorphism B⊗AAg≃BgB\otimes_A A_g\simeq B_g: the mutually inverse maps send b⊗a/gnb\otimes a/g^n to ba/φ(g)nba/\varphi(g)^n and send b/φ(g)nb/\varphi(g)^n to b⊗1/gnb\otimes 1/g^n. Lemma 2.29.1 identifies the ordinary tensor with its separated completion. Repeating this construction for products of the gig_i proves the finite-intersection assertions. Every map is fixed on coefficients and on the specified inverses, so the restriction maps commute. ◻

Corollary 2.29.4. NAG-001545 (A rational basis for strict-test domains over any complete field). Let KK now be any complete real-valued non-Archimedean field, with its fixed absolute value, possibly trivial. Let A→BA\to B be a strict-test affinoid-domain presentation of U⊆M(A)U\subseteq\mathcal M(A). Then UU has a finite covering by rational subsets R1,…,RtR_1,\ldots,R_t defined over the original AA, with unit radii or, equivalently after scaling, radii in ∣K×∣|K^\times|. For each RiR_i, and for each intersection RR indexed by a nonempty finite list of these members, there is a canonical bounded isomorphism B⊗^AAR ≃ ARB\widehat\otimes_A A_R\ \simeq\ A_R with bounded inverse. It identifies RR as a rational domain over BB with the same algebra ARA_R. All identifications commute with the restriction maps between these members and intersections. Empty resulting intersections have algebra zero; the empty domain uses the empty covering. This statement concerns finite rational coverings of the domain, and does not assert ambient local Weierstrass for trivial valuation.

Proof. If UU is empty, the point correspondence and Theorem 1.12.3 give B=0B=0, and the empty covering suffices. For nontrivial valuation, Theorem 2.27.5 supplies a finite original-AA rational covering of UU. For trivial valuation, Theorem 2.29.3 supplies one. Thus in both cases the members have the asserted scope.

Proposition 2.22.6 makes every indicated finite intersection rational over AA. It is contained in UU. Apply Lemma 2.30.1 to this inclusion: the image of its unit-ideal tuple in BB defines the same domain, and the actual completed pushout is ARA_R. This gives the displayed bounded isomorphisms for both valuation cases without using an intersection theorem for arbitrary strict-test domains. Their compatibility follows from the same lemma, or directly from uniqueness in the strict-test factorization property: after a further restriction, the two maps from BB extend the same map from AA, and the rational variables are prescribed. No arbitrary Banach completed-tensor exactness is used. ◻

Proposition 2.29.5. NAG-001546 (Failure of ambient local Weierstrass over trivial valuation). Let A=K[T]A=K[T] and B=K[T,T−1]B=K[T,T^{-1}], both with their discrete norms. The rational map A→B=A⟨Z⟩/(TZ−1)A\to B=A\langle Z\rangle/(TZ-1) presents the nonempty proper strict-test domain U={x∈M(A):∣T(x)∣=1}.U=\{x\in\mathcal M(A):|T(x)|=1\}. There is no finite strict rational covering M(A)=⋃iVi\mathcal M(A)=\bigcup_iV_i for which every U∩ViU\cap V_i is a Weierstrass domain in ViV_i with radii in ∣K×∣|K^\times|. The same obstruction applies if such radii are allowed in ∣K×∣\sqrt{|K^\times|}, which is also {1}\{1\}.

Proof. The rational point and domain universal properties, Theorems 2.19.3 and 2.22.5, prove the asserted domain presentation. The seminorm η\eta that assigns one to every nonzero polynomial and zero to zero is multiplicative and bounded, and belongs to UU. Evaluation at T=0T=0 gives another bounded point x0x_0, which does not belong to UU.

Every nonempty strict rational subset of M(A)\mathcal M(A) contains η\eta. Indeed, write it with a unit-ideal tuple (g;f1,…,fs)(g;f_1,\ldots,f_s) and inequalities ∣fj∣≤∣g∣|f_j|\leq|g|. If g=0g=0, these inequalities force every fjf_j to vanish at a point, contradicting the unit certificate, so the subset is empty. Otherwise ∣g(η)∣=1|g(\eta)|=1 and ∣fj(η)∣≤1|f_j(\eta)|\leq 1 for every jj, so η\eta belongs to it.

For any nonzero strict affinoid algebra CC over this KK, all elements have values at most one at every bounded point, by Lemma 2.29.1 and Lemma 1.6.1. Every unit-radius Weierstrass condition ∣hj∣≤1|h_j|\leq 1 is therefore automatic. Algebraically, its presentation is the polynomial quotient C[S1,…,Sm]/(Sj−hj)C[S_1,\ldots,S_m]/(S_j-h_j), canonically equal to CC. Thus a nonempty such Weierstrass domain is the entire spectrum of CC.

In any rational covering of M(A)\mathcal M(A), choose a member VV containing x0x_0. It is nonempty, so it also contains η\eta. Hence U∩VU\cap V is nonempty and proper in VV. The algebra representing VV is strict; the preceding paragraph shows that this intersection cannot be a unit-radius Weierstrass domain in VV. This proves the contradiction, even if an empty domain is admitted separately by convention. ◻

References. The positive result is an algebraic reconstruction for the stated trivial-valuation strict-test category. Its external theorem inputs are the explicitly located Stacks results used above; the finite-presentation flat-epimorphism argument is the affine algebraic form of the open-immersion criterion in [StacksAlg] (Tag 025G, Theorem 41.14.1), which is only a comparison reference here. It does not invoke a pre-existing analytic Gerritzen–Grauert theorem. Proposition 2.29.5 concerns the stronger ambient local Weierstrass statement; it does not contradict the finite-rational-union conclusion of Theorem 2.29.3. Generalized affinoids and arbitrary real radii are outside this section.

2.30 Finite affinoid coverings and coherent modules

Throughout this section KK is complete, with trivial valuation allowed. The rational calculations now extend to finite coverings by strict affinoid domains. The geometric input is the finite rational covering of Corollary 2.29.4. Its nontrivial-valuation branch comes from Gerritzen–Grauert; its trivial-valuation branch is the separate algebraic argument just proved. The stronger ambient local Weierstrass assertion is not used. We retain the strict-test domain convention of Definition 2.27.1. Thus every ordinary strict affinoid domain is included, but no assertion about non-strict test algebras is built into the arguments below.

Lemma 2.30.1. NAG-001533 (Restriction and rational subdomains). Let KK be complete, with trivial valuation allowed, and let A→BA\to B represent a strict-test domain V⊆X=M(A)V\subseteq X=\mathcal M(A). If A→CA\to C represents another such domain W⊆VW\subseteq V, there is a unique bounded AA-homomorphism B→CB\to C, and it represents WW as a strict-test domain of M(B)\mathcal M(B). These restriction maps compose. Composing two strict-test domain morphisms again gives a strict-test domain morphism.

In particular, if WW is rational over AA with radii in ∣K×∣|K^\times| and W⊆VW\subseteq V, then WW is rational over BB, its rational presentation algebra is canonically boundedly isomorphic to AWA_W, and B⊗^AAW ≃ AW.B\widehat\otimes_A A_W\ \simeq\ A_W.

Proof. The domain property for A→BA\to B, tested on the strict algebra CC, gives the unique map B→CB\to C. Its map of spectra is the inclusion W⊆VW\subseteq V under the given identifications. Suppose that a bounded B→EB\to E, with EE strict, has spectral image in WW. The domain property for A→CA\to C gives a unique C→EC\to E over AA. It is also over BB: the two resulting AA-maps from BB to EE agree by the uniqueness clause for VV. This proves the relative domain property. The same uniqueness proves composition of restrictions. To prove transitivity of domains, factor any eligible A→EA\to E first through BB and then through the second domain algebra. Both existence and uniqueness follow in this order, and the spectral identifications compose.

For the rational assertion use the original tuple defining WW over AA, now in BB. Its entries still generate the unit ideal. Its actual rational quotient is D=B⊗^AAWD=B\widehat\otimes_A A_W, by Proposition 1.14.3 and the joint-presentation calculation of Proposition 2.22.6. By Theorem 2.19.3, its spectrum is the inverse image of WW in M(B)\mathcal M(B), hence is WW. The restriction map B→AWB\to A_W just constructed, together with the identity of AWA_W, gives D→AWD\to A_W. This is inverse to the canonical AW→DA_W\to D. Indeed, the composite fixes AWA_W; on BB it agrees with the given map by domain uniqueness, since M(D)\mathcal M(D) maps into W⊆VW\subseteq V. The pushout universal property proves equality on DD. All maps are bounded, so this is the asserted Banach-algebra isomorphism. Zero algebras give the empty-domain case. ◻

Proposition 2.30.2. NAG-001534 (Intersections of strict-test domains). Let KK be complete, with trivial valuation allowed. Let A→BA\to B and A→CA\to C represent strict-test domains V,W⊆M(A)V,W\subseteq\mathcal M(A). Then D=B⊗^ACD=B\widehat\otimes_A C is strict, with zero allowed, and represents the strict-test domain V∩WV\cap W. Its spectrum maps homeomorphically onto that intersection. Iterated intersections are canonically associative, and their bounded algebra maps agree with restriction. In particular, an empty intersection has the zero algebra.

Proof. Proposition 1.14.3 makes DD strict and gives its bounded pushout property. Every point of M(D)\mathcal M(D) restricts to points of M(B)\mathcal M(B) and M(C)\mathcal M(C), so its image lies in V∩WV\cap W. To prove that this map is bijective, rather than assuming a general theorem about Berkovich fibre products, take the finite rational covering V=⋃αRαV=\bigcup_\alpha R_\alpha over AA supplied by Corollary 2.29.4. Lemma 2.30.1 identifies the algebra of RαR_\alpha as a rational algebra over BB with ARαA_{R_\alpha}. Its pullback to DD is rational and has algebra Dα=D⊗^BARα≃C⊗^AARα.D_\alpha =D\widehat\otimes_B A_{R_\alpha} \simeq C\widehat\otimes_A A_{R_\alpha}. The bounded isomorphism follows either from the two pushout properties or from the common relative presentation; it is not an assertion that an ordinary tensor product is already complete. The right-hand side is a rational quotient of CC. Its spectrum identifies with W∩RαW\cap R_\alpha, by the explicit rational point theorem. The DαD_\alpha cover M(D)\mathcal M(D), because the RαR_\alpha cover M(B)\mathcal M(B). For any x∈V∩Wx\in V\cap W, choose α\alpha with x∈Rαx\in R_\alpha; the point in M(C)\mathcal M(C) above xx gives a point of M(Dα)\mathcal M(D_\alpha) and hence of M(D)\mathcal M(D). If two points of M(D)\mathcal M(D) map to xx, both lie in this rational piece, and the injectivity of M(Dα)→M(C)\mathcal M(D_\alpha)\to\mathcal M(C) makes them equal. This proves the bijection.

The map is continuous, and compactness and Hausdorffness from Proposition 2.24.1 make it a homeomorphism onto its image. If a bounded A→EA\to E with EE strict has spectral image in V∩WV\cap W, the two domain properties supply unique compatible maps B,C→EB,C\to E, and the pushout supplies a unique D→ED\to E. This is exactly the required domain property. Iterating it gives canonical restriction and association isomorphisms by uniqueness. Finally, a nonzero strict affinoid has a point by Theorem 1.12.3; hence an empty spectrum forces D=0D=0. ◻

Theorem 2.30.3. NAG-001535 (Tate acyclicity for finite strict affinoid coverings). Let KK be complete, with trivial valuation allowed, let AA be a strict KK-affinoid algebra, and let MM be a finite AA-module. Let (Vi)i∈I(V_i)_{i\in I} be a finite covering of X=M(A)X=\mathcal M(A) by strict-test affinoid domains. Give each intersection algebra AVi0∩⋯∩ViqA_{V_{i_0}\cap\cdots\cap V_{i_q}} a norm-admissible Banach norm, and its finite modules their canonical norms. Then the augmented complex 0⟶M⟶∏iAVi⊗AM⟶∏i0<i1AVi0∩Vi1⊗AM⟶⋯0\longrightarrow M\longrightarrow \prod_i A_{V_i}\otimes_A M\longrightarrow \prod_{i_0<i_1} A_{V_{i_0}\cap V_{i_1}}\otimes_A M \longrightarrow\cdots is norm-strict exact in every degree. Equivalently the all-tuples version is norm-strict exact. Every ordinary finite-module tensor in these complexes agrees, by a bounded isomorphism, with its completed tensor. No reducedness, separability or residue-characteristic hypothesis is needed.

Proof. Choose for each ViV_i the finite covering by AA-rational domains supplied by Corollary 2.29.4, and let W\mathcal W be their finite union as a covering family of XX. It refines (Vi)(V_i) with a specified containing index for each member. Its augmented complex is norm-strict exact by Theorem 2.22.9.

Fix a nonempty tuple of coarse indices, and put Y=Vi0∩⋯∩VipY=V_{i_0}\cap\cdots\cap V_{i_p}. Proposition 2.30.2 gives its strict algebra AYA_Y and all compatible intersection maps. Each member of W\mathcal W restricts to a rational quotient of AYA_Y: take its original unit-ideal tuple in AA, map it to AYA_Y, and use the rational point theorem. These quotients cover M(AY)\mathcal M(A_Y). The module AY⊗AMA_Y\otimes_A M is finite over AYA_Y. Apply Theorem 2.22.9 over this algebra. Associativity of ordinary tensors, [StacksAlg] (Tag 00D2, Lemma 10.12.7), and Proposition 1.11.6 identify its terms, restrictions and augmentation with the mixed intersection diagram. Thus the required local complexes are norm-strict exact on every coarse intersection. Empty intersections have actual zero algebras by the proposition, so their terms are zero for a proved algebraic reason.

All hypotheses of Proposition 2.21.5(1) now hold. It transfers norm-strict exactness from W\mathcal W to (Vi)(V_i), in every degree and with the given augmentation. The same proposition supplies the all-tuples comparison. The finite-module completion statement is Proposition 1.11.6. The lifting bounds may depend on the finite cover, the module and the chosen equivalent norms; no bound uniform over all covers is asserted. If XX is empty, A=0A=0 and M=0M=0, and the conclusion also covers its empty covering. ◻

Proposition 2.30.4. NAG-001536 (Flat restriction to a strict affinoid domain). Under the same field hypothesis, every strict-test domain map A→BA\to B is flat as an ordinary ring map. A finite covering by such domains gives a faithfully flat ordinary ring map A→∏iAViA\to\prod_i A_{V_i}. These assertions concern algebraic tensor products of arbitrary modules; they do not assert exactness of completed tensor products of arbitrary Banach modules.

Proof. Choose an AA-rational covering (Rj)(R_j) of the domain by Corollary 2.29.4. Lemma 2.30.1 makes it a rational covering over BB with the same algebras. Put S=∏jARjS=\prod_j A_{R_j}. Corollary 2.28.2 makes B→SB\to S faithfully flat, whereas rational flatness makes A→SA\to S flat. If N′→NN'\to N is an injective map of AA-modules and QQ is the kernel of B⊗AN′→B⊗ANB\otimes_A N'\to B\otimes_A N, tensoring over BB with the flat algebra SS identifies S⊗BQS\otimes_B Q with the kernel of S⊗AN′→S⊗ANS\otimes_A N'\to S\otimes_A N, which is zero. Faithfulness gives Q=0Q=0. Together with right exactness of ordinary tensor products, [StacksAlg] (Tag 00DF, Lemma 10.12.10), this is flatness of BB over AA. For an empty domain, B=0B=0 is flat and the assertion is immediate.

For a finite covering, refine each member by the preceding rational covers. The resulting AA-rational cover has faithfully flat product A→SA\to S, and SS is an algebra over B′=∏iAViB'=\prod_iA_{V_i}. The already proved flatness makes B′B' flat over AA. If B′⊗AN=0B'\otimes_A N=0, then S⊗AN=0S\otimes_A N=0, so N=0N=0. Thus tensoring with B′B' is faithful as well as exact. This is the asserted faithful flatness. ◻

Definition 2.30.5. NAG-001537 (The finite strict-domain site). For KK complete, with trivial valuation allowed, and X=M(A)X=\mathcal M(A) strict, let DX\mathcal D_X be the partially ordered category of strict-test domains in XX, with inclusions as arrows. Coverings are finite set-theoretic coverings by objects, including the empty covering of the empty set. Choose representing algebras AVA_V, with AX=AA_X=A and A∅=0A_\varnothing=0. Use the unique bounded restrictions of Lemma 2.30.1, and set OXstr(V)=AV,M~str(V)=AV⊗AM(M finite over A).\mathcal O_X^{\mathrm{str}}(V)=A_V, \qquad \widetilde M^{\mathrm{str}}(V)=A_V\otimes_A M \quad(M\text{ finite over }A).

Theorem 2.30.6. NAG-001538 (The strict-domain structure sheaf and rational comparison). The coverings of DX\mathcal D_X form a pretopology. The displayed assignments are a ring sheaf and module sheaves, independent up to canonical isomorphism of the representing algebras. The functor M↦M~strM\mapsto\widetilde M^{\mathrm{str}} is exact and fully faithful, and its global sections are MM. Inclusion RX↪DX\mathcal R_X\hookrightarrow\mathcal D_X induces an equivalence of ringed topoi, carrying OXstr\mathcal O_X^{\mathrm{str}} to OXrat\mathcal O_X^{\mathrm{rat}} and M~str\widetilde M^{\mathrm{str}} to M~rat\widetilde M^{\mathrm{rat}}.

Proof. Intersections exist by Proposition 2.30.2. Pulling back and composing finite set-theoretic covers proves the pretopology axioms. For a covering of an object VV, its members and all intersections are strict-test domains over AVA_V by Lemma 2.30.1. Apply Theorem 2.30.3 over AVA_V to the finite module AV⊗AMA_V\otimes_A M. Tensor associativity identifies its terms with the displayed presheaf. Degree-zero exactness is precisely the sheaf equalizer condition. The empty value is zero. Taking M=AM=A gives the ring sheaf. Uniqueness of all restriction maps proves independence of the choices.

Proposition 2.30.4 gives objectwise exactness for finite-module short exact sequences. At XX the value is MM. A sheaf map between two such module sheaves is determined by its map uu at XX, because on each VV it must send b⊗mb\otimes m to b⊗u(m)b\otimes u(m). Conversely this formula defines a sheaf map for every AA-linear uu. Thus the functor is exact and fully faithful.

For the site comparison apply the authorized comparison lemma [StacksAlg] (Tag 03A0, Lemma 7.29.1). Its hypotheses can be checked directly here. Inclusion is fully faithful, so both local faithfulness and local fullness hold. It preserves covers and finite intersections; restriction of a sheaf is therefore a sheaf, so it is continuous. For a covering of a rational object by strict-test domains, refine each member by its finite AA-rational covering from Corollary 2.29.4. Their union is a covering in RX\mathcal R_X refining the given one, which is cocontinuity. The same corollary covers every object of DX\mathcal D_X by objects from RX\mathcal R_X. These are exactly the five hypotheses of the comparison lemma, including the empty object. It gives an equivalence of topoi whose restriction functor is the inclusion-induced restriction of sheaves. On rational objects the chosen representing algebras and restrictions agree by domain uniqueness. Thus the ring sheaves and the finite-module sheaves restrict as asserted, making this an equivalence of ringed topoi. ◻

Theorem 2.30.7. NAG-001539 (Kiehl equivalence on the finite strict-domain site). For a complete KK, with trivial valuation allowed, and a strict KK-affinoid X=M(A)X=\mathcal M(A), the functors M⟼M~str,F⟼Γ(X,F)M\longmapsto\widetilde M^{\mathrm{str}}, \qquad \mathcal F\longmapsto\Gamma(X,\mathcal F) give inverse exact equivalences between finite AA-modules and coherent OXstr\mathcal O_X^{\mathrm{str}}-modules on DX\mathcal D_X. Coherence here is Definition 18.23.1(8), Tag 03DL, of the Stacks Project. It is also equivalent to finite presentation on this site. Every coherent module has a global finite free presentation, and every finite strict-domain covering has a norm-strict exact augmented Čech complex for it, with the canonical norms on its finite section modules.

Proof. Theorem 2.30.6 identifies the ringed topoi of DX\mathcal D_X and RX\mathcal R_X. Coherence and finite presentation are intrinsic under this equivalence, by [StacksAlg] (Tag 03DM, Lemma 18.23.2); the special cocontinuous comparison required in that lemma is precisely the one just verified. Theorem 2.28.12 therefore identifies the coherent modules with the sheaves M~rat\widetilde M^{\mathrm{rat}} for finite AA-modules MM. Under the equivalence these are exactly M~str\widetilde M^{\mathrm{str}}. Full faithfulness and the global-sections inverse were already proved in Theorem 2.30.6, as was exactness. A finite presentation of MM over the Noetherian ring AA gives a global finite free presentation of its sheaf. Finally apply Theorem 2.30.3 to MM. Its terms are exactly the section modules of F≃M~str\mathcal F\simeq\widetilde M^{\mathrm{str}}, proving the last assertion. ◻

The preceding acyclicity statements concern explicit complexes. To pass to derived sheaf cohomology, one needs a comparison theorem; vanishing for a single covering would not suffice. Here vanishing has been proved on every object and for every covering of the stated site.

Corollary 2.30.8. NAG-001540 (Theorem B on the specified affinoid sites). Let KK be complete, with trivial valuation allowed, and let AA be strict. If F\mathcal F is a coherent module on the fixed rational ringed site RX\mathcal R_X, then Hq(V,F)=0(V∈RX, q>0).H^q(V,\mathcal F)=0 \qquad(V\in\mathcal R_X,\ q>0). For the same complete field KK, including trivial valuation, the same assertion holds for every coherent module on DX\mathcal D_X and every V∈DXV\in\mathcal D_X. Here HqH^q denotes the right derived functors of sections in the category of abelian sheaves of the specified site, applied to the underlying abelian sheaf.

Proof. For RX\mathcal R_X, Theorem 2.28.12 writes F=M~rat\mathcal F=\widetilde M^{\mathrm{rat}} for a finite AA-module. Lemma 2.28.10 gives vanishing of higher Čech cohomology for every covering of every object. For DX\mathcal D_X, Theorem 2.30.7 gives the same conclusion. Apply [StacksAlg] (Tag 03F9, Lemma 21.10.9), with its set B\mathcal B equal to all objects and its family Cov\mathrm{Cov} equal to all the finite coverings in the chosen pretopology. These coverings are a cofinal system by definition of the site. All iterated fibre products required by that lemma are intersections and remain in the site, by Proposition 2.22.6 or Proposition 2.30.2, respectively. Thus all three hypotheses of the cited lemma hold, and it gives the claimed derived vanishing in every positive degree. This application uses the general site lemma, not the different theorem about open covers of a topological space. ◻

The strict finite-domain theory supplies the local algebra for the general construction. Theorem 2.36.6 gives rational refinement at arbitrary radii, and Theorems 2.38.5 and 2.39.6 give the corresponding sheaf equivalences. Section 2.40 glues these affinoid sites through analytic nets, including non-good spaces.

2.31 Rational domains with arbitrary radii

Fix a complete non-Archimedean field KK with its specified real-valued absolute value, allowing trivial valuation. All affinoids in this section may have arbitrary positive presentation radii. Norms are admissible in the quantitative sense of Definition 1.9.1; quotients use actual algebraic ideals. The generalized finite-module theorem proves that these ideals are closed.

For the specified coefficient and quotient norms, nonzero Banach algebras have homogeneous submultiplicative norms normalized by ∥1∥=1\|1\|=1, as in the finite-module section. The zero algebra has its zero norm. Statements of isometry use precisely these norms. Replacing them by uniformly equivalent admissible norms preserves the bounded isomorphisms and their bounded inverses, but need not preserve literal isometry.

Theorem 2.31.1. NAG-001576 (Points of an arbitrary-radius rational presentation). Let AA be a norm-admissible KK-affinoid algebra, let ρ1,…,ρs>0\rho_1,\ldots,\rho_s>0, and suppose (g,f1,…,fs)=A(g,f_1,\ldots,f_s)=A. With the coefficient norm and its quotient norm, put D=A{ρ−1T},B=D/(gTi−fi)i=1s,V={x∈M(A):∣fi(x)∣≤ρi∣g(x)∣ ∀i}.\begin{aligned} D&=A\{\rho^{-1}T\},\qquad B=D/(gT_i-f_i)_{i=1}^s,\\ V&=\{x\in\mathcal M(A):|f_i(x)|\leq\rho_i|g(x)|\ \forall i\}. \end{aligned} Then BB is a norm-admissible KK-affinoid algebra, possibly zero, and restriction gives a homeomorphism M(B)≃V\mathcal M(B)\simeq V. Moreover B=0B=0 if and only if VV is empty. The case s=0s=0 is allowed and gives B=AB=A, since gg is then a unit.

Proof. Proposition 1.17.3 makes DD affinoid, and Corollary 1.18.4 gives the asserted complete actual quotient BB. Choose a unit-ideal identity a0g+∑iaifi=1a_0g+\sum_i a_if_i=1. In BB it becomes g(a0+∑iaiTˉi)=1.g\left(a_0+\sum_i a_i\bar T_i\right)=1. Thus gg is a unit in BB, with the usual zero-ring interpretation. At every x∈Vx\in V, the same identity and the inequalities force ∣g(x)∣>0|g(x)|>0.

For x∈Vx\in V, Lemma 1.6.1 gives the valued fraction field Fx=Frac⁡(A/ker⁡x)F_x=\operatorname{Frac}(A/\ker x). Polynomial evaluation at Ti=fi/gT_i=f_i/g in this field has multiplicative seminorm pxp_x bounded by the weighted coefficient norm: px(∑αaαTα)≤max⁡α∣aα(x)∣∏i∣fi/g∣xαi≤max⁡α∥aα∥ρα.p_x\left(\sum_\alpha a_\alpha T^\alpha\right) \leq\max_\alpha |a_\alpha(x)|\prod_i|f_i/g|_x^{\alpha_i} \leq\max_\alpha\|a_\alpha\|\rho^\alpha. For polynomials P,QP,Q, ∣px(P)−px(Q)∣≤∥P−Q∥|p_x(P)-p_x(Q)|\leq\|P-Q\|. Hence norm-convergent polynomial truncations extend pxp_x uniquely to a seminorm on DD. Addition, multiplication and its scalar values pass to the limit, because products of bounded convergent sequences converge. It kills all gTi−figT_i-f_i and their ideal. Taking the infimum over representatives gives a contractive multiplicative seminorm yxy_x on BB extending xx. No completion of FxF_x is needed here.

Conversely, a point yy of BB is contractive for its quotient norm, so ∣Tˉi(y)∣≤ρi|\bar T_i(y)|\leq\rho_i. Its restriction belongs to VV. To prove uniqueness over a given xx, take a polynomial P=∑aαTαP=\sum a_\alpha T^\alpha of total degree at most NN and form QP=∑αaαfαgN−∣α∣∈A.Q_P=\sum_\alpha a_\alpha f^\alpha g^{N-|\alpha|}\in A. In BB one has gNP(Tˉ)=QPg^NP(\bar T)=Q_P, so every point over xx must satisfy ∣P(Tˉ)(y)∣=∣QP(x)∣/∣g(x)∣N|P(\bar T)(y)|=|Q_P(x)|/|g(x)|^N. The polynomial image is dense in BB, and contractivity determines the remaining values uniquely. This proves the bijection. For a polynomial, the displayed value is a continuous function of x∈Vx\in V. For any b∈Bb\in B, choose a lift F∈DF\in D and polynomial truncations Pn→FP_n\to F. Uniformly in x∈Vx\in V, ∣∣b(yx)∣−∣Pn(Tˉ)(yx)∣∣≤∥F−Pn∥.\bigl||b(y_x)|-|P_n(\bar T)(y_x)|\bigr|\leq\|F-P_n\|. Thus all evaluations along x↦yxx\mapsto y_x are continuous. Restriction is also continuous by its definition, proving the homeomorphism. Theorem 1.20.1 applied to BB proves the assertion about emptiness. Zero algebras and an empty VV require no exceptional point construction. ◻

Theorem 2.31.2. NAG-001577 (Bounded and geometric rational factorization). Keep the rational presentation A→BA\to B and subset VV above. For every bounded unital KK-algebra homomorphism ϕ:A→C\phi:A\to C to a Banach KK-algebra, a bounded KK-algebra factorization A→B→CA\to B\to C exists if and only if ϕ(g)\phi(g) is a unit and, for each ii, there is Ci≥1C_i\geq 1 with ∥(ϕ(fi)ϕ(g)−1)m∥≤Ciρim(m≥0).\|\bigl(\phi(f_i)\phi(g)^{-1}\bigr)^m\|\leq C_i\rho_i^m \qquad(m\geq0). The factorization is unique. If CC is a norm-admissible affinoid over any specified complete extension L/KL/K, this condition is equivalent to the image of ML(C)→MK(A)\mathcal M_L(C)\to\mathcal M_K(A) being contained in VV. In particular the rational presentation has the domain universal property for all such complete-extension affinoid tests, not only for strict tests over KK. Zero targets are allowed.

Proof. In BB, the preceding unit identity makes gg invertible and forces any factorization to send Tˉi\bar T_i to bi=ϕ(fi)ϕ(g)−1b_i=\phi(f_i)\phi(g)^{-1}. Since ∥Tˉim∥≤ρim\|\bar T_i^m\|\leq\rho_i^m, boundedness of the factorization gives the required estimates. Conversely, Proposition 1.9.3 gives a bounded evaluation A{ρ−1T}→CA\{\rho^{-1}T\}\to C at the bib_i. It kills the actual ideal generated by gTi−figT_i-f_i. If its bound is L0L_0, taking the infimum over lifts of b∈Bb\in B gives ∥ψ(b)∥≤L0∥b∥\|\psi(b)\|\leq L_0\|b\|. Thus it descends boundedly. Its values on the dense polynomial image are forced, proving uniqueness. This proof uses the explicit power estimates for an arbitrary Banach target.

Now let CC be affinoid over LL. If a factorization exists, restricting its spectral points and using Theorem 2.31.1 gives the image condition. Conversely, at every y∈ML(C)y\in\mathcal M_L(C) the inequalities and the image of the unit-ideal identity imply ∣ϕ(g)(y)∣>0|\phi(g)(y)|>0. Theorem 1.20.1, over LL, makes ϕ(g)\phi(g) a unit. Its ratios have ∣bi(y)∣≤ρi|b_i(y)|\leq\rho_i at every such point. Theorem 1.20.2, again over LL and with the actual radii ρi\rho_i, gives the required scaled power estimates. The Banach factorization just proved applies because CC is also a Banach KK-algebra. The zero target has its unique factorization directly; the equivalence of spectral emptiness and zero prevents an exceptional nonzero target with vacuous hypotheses. ◻

Taking one rational algebra as the test target for another now gives a unique bounded restriction map whenever the corresponding subsets are nested. Equal subsets give bounded inverse maps, and uniqueness makes restriction compatible with composition. These comparisons preserve the algebra and its Banach structure up to equivalence, not necessarily a particular presentation norm. We next identify finite intersections by their actual presentations.

Proposition 2.31.3. NAG-001578 (Joint presentations and completed pushouts). For any two arbitrary-radius affinoid quotient presentations over AA, put B=A{ρ−1T}/I,C=A{τ−1S}/J,H=A{ρ−1T,τ−1S}/(I,J),\begin{aligned} B&=A\{\rho^{-1}T\}/I,\qquad C=A\{\tau^{-1}S\}/J,\\ H&=A\{\rho^{-1}T,\tau^{-1}S\}/(I,J), \end{aligned} where (I,J)(I,J) is the sum of the actual extended ideals. With the displayed quotient norms there is a canonical isometric algebra isomorphism B⊗^AC≃HB\widehat\otimes_A C\simeq H. This is the separated completed tensor product, not an assertion that the ordinary algebraic tensor is already complete. It has the bounded pushout universal property for commutative Banach targets.

If I=(gTi−fi)I=(gT_i-f_i) and J=(hSj−ej)J=(hS_j-e_j) are rational relation ideals, then HH is isometric to the rational presentation with denominator ghgh and the following numerator-radius pairs: (hfi,ρi),(gej,τj),(fiej,ρiτj).(hf_i,\rho_i),\qquad (ge_j,\tau_j),\qquad (f_ie_j,\rho_i\tau_j). Its spectrum is the intersection of the two original rational subsets. Finite iteration gives all intersections, including repeated members, with compatible restriction maps. Empty intersections have zero algebra.

Proof. All weighted coefficient algebras here are affinoid by Proposition 1.17.3; their actual ideals are closed and their quotients are complete by Corollary 1.18.4. On the ordinary balanced tensor product, the projective seminorm is submultiplicative by multiplying finite representations and taking their two infima. Its null space is an ideal. Separation and completion thus give a Banach algebra G=B⊗^ACG=B\widehat\otimes_A C, or zero. The structural maps are contractive. The two contractive maps into HH give a contractive map G→HG\to H by their balanced product.

Conversely the elements Tˉi⊗1\bar T_i\otimes 1 and 1⊗Sˉj1\otimes\bar S_j have all powers bounded by ρim\rho_i^m and τjm\tau_j^m. Proposition 1.9.3 gives a contractive map from the joint coefficient algebra to GG. It kills II and JJ, first on each whole coefficient algebra by bounded polynomial approximation, hence on their extended ideals. It descends to a contractive H→GH\to G. The composites fix all coefficients and variables in HH and all pure tensors in GG. Density makes them identities, proving the isometry. More generally, two bounded maps from B,CB,C to a Banach algebra agreeing on AA have a bounded balanced product; its finite-representation bound extends it uniquely to GG. This proves the bounded pushout property.

For the rational assertion, multiply the two unit-ideal identities. They show that ghgh and the listed numerators generate the unit ideal. Call their rational algebra PP, with variables Yi,Zj,WijY_i,Z_j,W_{ij} of the listed radii. In HH, both gg and hh are units. The assignments Yi⟼Tˉi,Zj⟼Sˉj,Wij⟼TˉiSˉjY_i\longmapsto\bar T_i,\qquad Z_j\longmapsto\bar S_j, \qquad W_{ij}\longmapsto\bar T_i\bar S_j respect all bounds and relations and give a contractive P→HP\to H. In PP, the product ghgh is a unit, so both factors are units. Dividing its relations gives gYi=figY_i=f_i, hZj=ejhZ_j=e_j, and Wij=YiZjW_{ij}=Y_iZ_j. Sending TiT_i to YiY_i and SjS_j to ZjZ_j therefore gives a contractive H→PH\to P, inverse on all variables. The point theorem identifies the resulting inequalities with the two original sets: divide the hfihf_i and gejge_j inequalities by the nonzero factors; conversely the two original sets imply all product inequalities.

Iteration gives finite intersections. Their comparison and restriction maps agree by the uniqueness in Theorem 2.31.2; a repeated member is consequently redundant by bounded inverse maps. Theorem 1.20.1 makes an empty intersection algebra zero. Every zero-quotient case in the pushout assertion also follows directly: if one factor ideal contains one, so does the joint ideal. ◻

The preceding statements concern pointwise rational domains and bounded maps. Comparing separate spectral maxima of fif_i and gg cannot replace their simultaneous pointwise inequalities; Example 1.9.5 already gives a counterexample even when gg is a unit. Likewise, the arbitrary-Banach-target criterion retains bounds on all powers, not only a spectral-radius bound.

2.32 Tate acyclicity for arbitrary-radius rational coverings

A finite rational covering can be made strict over one free-polyradius field. To descend its Čech complex, however, one must compare its actual quotient algebras and finite modules, not just their point sets. We first make those identifications, retaining the maps as well as the objects. All fields below are complete for the specified real-valued non-Archimedean absolute value; trivial valuation and zero affinoids are allowed.

The normalized coefficient and quotient norms from the rational-domain section may be chosen for the proof. All conclusions of bounded isomorphism and norm-strict exactness are invariant under uniformly equivalent admissible norms. Isometries below refer only to the specified coefficient and quotient norms.

Proposition 2.32.1. NAG-001580 (Actual quotient and finite-module base change).

Let AA be a norm-admissible KK-affinoid algebra, let ρ\rho be any finite positive tuple, put D=A{ρ−1T}D=A\{\rho^{-1}T\}, and let J⊆DJ\subseteq D be an algebraic ideal. Give B=D/JB=D/J its quotient norm. For every finite KK-free tuple rr, put L=KrL=K_r and Er=E⊗^KLE_r=E\widehat\otimes_K L for Banach KK-spaces EE. Coefficient and quotient maps induce isometric algebra identifications Dr≃Ar{ρ−1T},Br≃Dr/JDr.\begin{aligned} D_r&\simeq A_r\{\rho^{-1}T\},\\ B_r&\simeq D_r/JD_r. \end{aligned} The ideal JDrJD_r here is the actual algebraic extended ideal; it is closed and equals c0(Zℓ,r;J)c_0(\mathbf Z^{\ell},r;J) in the coefficient description, where ℓ\ell is the length of rr. For every finite canonical AA-module MM, the natural maps give a bounded isomorphism with bounded inverse (B⊗AM)r≃Br⊗ArMr.(B\otimes_A M)_r \simeq B_r\otimes_{A_r}M_r. Each ordinary tensor displayed is complete for its canonical finite-module norm and agrees boundedly with its separated completed tensor. These identifications commute with bounded maps of such quotient algebras and with maps of finite AA-modules. No arbitrary complete extension is substituted for the specified free extension in the isometric coefficient statements.

Proof. Theorem 1.16.2 and Lemma 1.17.1 identify both descriptions of DrD_r with the same joint coefficient families, with norm sup⁡ν,α∥aνα∥rνρα\sup_{\nu,\alpha}\|a_{\nu\alpha}\|r^\nu\rho^\alpha. They identify multiplication first on finite polynomials and then by continuity. By Theorem 1.18.2, JJ is finite and closed and any finite generator map Dt→JD^t\to J has a uniform lifting bound for the induced norm on JJ. Applying the coefficient exactness of Proposition 2.14.1 to 0→J→D→B→00\to J\to D\to B\to 0 gives a surjection Dr→BrD_r\to B_r with quotient norm equal to the coefficient norm, and with kernel c0(Zℓ,r;J)c_0(\mathbf Z^{\ell},r;J).

Write J=(h1,…,ht)J=(h_1,\ldots,h_t). For each coefficient jνj_\nu of an element of this kernel, choose dkν∈Dd_{k\nu}\in D such that jν=∑khkdkνj_\nu=\sum_kh_kd_{k\nu} and max⁡k∥dkν∥≤C∥jν∥\max_k\|d_{k\nu}\|\leq C\|j_\nu\|, choosing zero lifts for zero coefficients. The same bound makes each ∑νdkνTrν\sum_\nu d_{k\nu}T_r^\nu a member of DrD_r. Thus the kernel is contained in JDrJD_r; the reverse inclusion follows from the quotient map. This proves equality with the actual ideal, including its closedness, without replacing a dense image by its closure. It also proves the second isometry.

Let MB=B⊗AMM_B=B\otimes_A M with a canonical finite-BB-module norm. Proposition 1.18.7 gives its Banach interpretation. Apply Theorem 1.18.6 with weights rνr^\nu, first over BB and then over AA, and use ordinary tensor associativity [StacksAlg] (Tag 00D2, Lemma 10.12.7). Algebraically the required comparison is (MB)r≃Br⊗B(B⊗AM)≃Br⊗AM≃Br⊗Ar(Ar⊗AM)≃Br⊗ArMr.\begin{aligned} (M_B)_r &\simeq B_r\otimes_B(B\otimes_A M) \simeq B_r\otimes_A M\\ &\simeq B_r\otimes_{A_r}(A_r\otimes_A M) \simeq B_r\otimes_{A_r}M_r. \end{aligned} These are bounded comparisons, not an assumed associativity of arbitrary completed tensors. Indeed, a finite presentation Au→Av→M→0A^u\to A^v\to M\to 0 presents both sides by the same matrix on Bru→BrvB_r^u\to B_r^v. Coefficient presentation, Theorem 1.18.5, gives its canonical quotient norm, up to the stated equivalences. The finite-module theorem then makes the algebraic comparison and its inverse bounded. The coefficient and ordinary tensor formulas agree on finite Laurent sums and module generators. Density and boundedness therefore prove compatibility with every bounded algebra map and finite-module map in the statement. This includes all quotient restriction maps used below. ◻

Theorem 2.32.2. NAG-001581 (Finite rational Tate acyclicity). Let AA be any norm-admissible KK-affinoid algebra and let MM be any finite AA-module with a canonical norm. Let (Vi)i=1n(V_i)_{i=1}^n be any finite covering of M(A)\mathcal M(A) by rational domains with arbitrary positive radii. For each finite intersection, let Ai0⋯iqA_{i_0\cdots i_q} be its actual joint rational algebra, with any admissible norm. Then the augmented complex 0⟶M⟶∏iAi⊗AM⟶∏i0<i1Ai0i1⊗AM⟶⋯0\longrightarrow M\longrightarrow \prod_i A_i\otimes_A M\longrightarrow \prod_{i_0<i_1}A_{i_0i_1}\otimes_A M\longrightarrow\cdots is norm-strict exact in every degree. The all-tuples complex, with repeated indices and the usual alternating differential, has the same conclusion. Each finite-module ordinary tensor agrees boundedly with its separated completed tensor. For every fixed degree, cycles have primitives with a uniform norm bound depending on the finite cover, MM, that degree and the chosen norms. No bound uniform over all covers or all degrees is asserted.

Proof. Step 1: one strictification for the whole finite diagram. Choose rational presentations for the ViV_i. Take a free extension L=KrL=K_r as in Theorem 1.17.6, using the finite set consisting of the presentation radii of AA and all radii in these rational presentations. Thus ArA_r is strict and all these rational radii belong to ∣L×∣\sqrt{|L^\times|}. All products of them that arise in finite joint intersections have the same property. Proposition 2.32.1 identifies every extended intersection algebra with its actual joint presentation over ArA_r.

Every LL-spectral point of ArA_r restricts to a KK-spectral point of AA. The original covering therefore implies that the extended rational subsets cover ML(Ar)\mathcal M_L(A_r). This only uses restriction of points. By Proposition 2.26.2, each extended rational algebra has a unit-radius rational presentation over ArA_r, with bounded inverse comparison maps. This step uses powers of the inequalities: belonging to the divisible closure of the value group does not permit direct rescaling of the original variable. The comparison maps on all intersections commute with restrictions by the uniqueness of rational factorization.

The module MrM_r is finite canonical over ArA_r by Theorem 1.18.6. Theorem 2.22.9 now gives norm-strict exactness of the strict rational covering complex over LL, in both versions. This uses no generalized Gerritzen–Grauert theorem.

Step 2: identify the extended complex, including its arrows. Write C−1=MC^{-1}=M and let CqC^q be the displayed finite product in degree q≥0q\geq 0, with maximum norm. Each factor is a Banach space by Proposition 1.18.7. Coefficient families commute isometrically with finite products: a supremum commutes with a finite maximum, and the finite-tail condition is preserved by finite unions. Proposition 2.32.1 consequently identifies (Cq)r(C^q)_r, by bounded inverse maps, with the corresponding term of the upper complex. These maps identify restrictions and augmentation, not just their modules. Thus the extended differential is exactly the coefficientwise extension of dq:Cq→Cq+1d^q:C^q\to C^{q+1}, including each alternating sign. Each fixed degree has only finitely many factors, so the comparison constants can be absorbed into norm-strictness.

Step 3: descend cycles and their norm bounds. If z∈Cqz\in C^q is a cycle with q≥0q\geq 0, its constant embedding is an upper cycle. The strict exactness there supplies w∈(Cq−1)rw\in(C^{q-1})_r satisfying drw=ι(z)d_r w=\iota(z) and ∥w∥≤Lq∥z∥\|w\|\leq L_q\|z\| for a fixed LqL_q. Apply the zeroth-coefficient map of Theorem 1.16.2. Its naturality for bounded KK-linear maps gives d(p0w)=z,∥p0w∥≤Lq∥z∥.d(p_0w)=z,\qquad \|p_0w\|\leq L_q\|z\|. Thus exactness and actual bounded lifts descend without presupposing that a lower differential has closed image. Its image is now the kernel of the next bounded differential and hence is closed. At the augmentation, upper injectivity and the isometric constant embedding give injectivity of M→C0M\to C^0; applying its upper inverse bound to constant families gives the lower inverse bound. This proves norm-strict exactness in every degree. The same argument works degree by degree for all tuples, since each degree still has a finite product. If the cover is empty, its covering condition forces M(A)=∅\mathcal M(A)=\varnothing, hence A=M=0A=M=0 by Theorem 1.20.1. All assertions are then immediate. ◻

Proposition 2.32.3. NAG-001582 (Algebraic flatness of rational localization). Every arbitrary-radius rational map A→BA\to B above is flat as a homomorphism of ordinary rings. It carries a short exact sequence of finite canonical AA-modules to a norm-strict exact sequence under ordinary or separated completed tensor product. A finite rational covering gives a faithfully flat ordinary ring map A→∏iAiA\to\prod_i A_i. These statements do not assert completed-tensor exactness for arbitrary Banach modules.

Proof. Choose a single free extension strictifying AA and the radii of BB. Proposition 2.26.2 makes Ar→BrA_r\to B_r boundedly isomorphic to a unit-radius rational map. Its algebraic flatness and finite-module exactness are already proved in Lemma 2.22.1; no covering theorem is required. A short exact sequence of finite AA-modules is norm-strict by Theorem 1.18.3. Its coefficient extension is exact by Theorem 1.16.3, and it consists of finite canonical ArA_r-modules by Theorem 1.18.6. Tensoring this sequence with BrB_r is exact.

The resulting maps are norm-strict finite-BrB_r-module maps. Proposition 2.32.1 identifies this sequence with the completed free scalar extension of the original tensor sequence over BB. That original sequence consists of Banach spaces by Proposition 1.18.7. Reflection in Theorem 1.16.3 therefore proves the finite-module assertion before extension. For every finitely generated ideal I⊆AI\subseteq A, apply it to 0→I→A→A/I→00\to I\to A\to A/I\to 0. The map I⊗AB→BI\otimes_A B\to B is injective. The finite-ideal criterion [StacksAlg] (Tag 00HD, Lemma 10.39.5(4)) now proves ordinary flatness for all algebraic modules.

Write BB for the base of the given finite rational cover, BjB_j for its actual row algebras, and Q=∏jBjQ=\prod_j B_j. The preceding single-row result makes each BjB_j flat over BB, so QQ is flat by the finite direct-sum identity. If B=0B=0, the assertion is immediate. For any maximal ideal m\mathfrak m of BB, its actual nonzero quotient is affinoid by Corollary 1.18.4. Nonemptiness in Theorem 1.20.1 gives a bounded point of this quotient, hence a point of the domain that kills m\mathfrak m. Some row in the cover over BB contains that point. By Theorem 2.31.1 the point lifts to its actual row algebra, so Q/mQ≠0Q/\mathfrak m Q\ne 0. For a nonzero BB-module choose a cyclic submodule B/JB/J with JJ proper and choose m⊇J\mathfrak m\supseteq J. The quotient Q/JQQ/JQ surjects onto Q/mQQ/\mathfrak m Q and injects into the tensor of the given module by flatness. This proves that QQ is faithfully flat over BB. ◻

Corollary 2.32.4. NAG-001583 (Finite-module sheaves on the finite rational site). Let X=M(A)X=\mathcal M(A). Use as objects the rational subsets of XX with arbitrary positive radii, and as coverings their finite set-theoretic rational coverings. For a finite AA-module MM, the assignment M~(V)=AV⊗AM\widetilde M(V)=A_V\otimes_A M with the unique bounded restrictions is a sheaf on this site. The assignment M↦M~M\mapsto\widetilde M is exact, Γ(X,M~)=M\Gamma(X,\widetilde M)=M, and A~\widetilde A is a sheaf of rings. This statement specifies a finite rational site; it does not identify it with the entire analytic G-site or assert a new coherence equivalence.

Proof. The identity object is rational. Proposition 2.31.3 gives rational intersections, and finite covers are preserved by pullback through them and by finite composition. These facts define the stated pretopology. Theorem 2.31.2 defines restrictions independently of the chosen presentation, with the required composition identities. If Vi⊆VV_i\subseteq V are rational over AA, their defining tuples remain unit-generated in AVA_V. Their rational presentations over AVA_V represent ViV_i by Theorem 2.31.1. The identity and restriction maps, together with uniqueness, identify these algebras boundedly with AViA_{V_i}. Equivalently the rational pullback algebra is AV⊗^AAVi≃AViA_V\widehat\otimes_A A_{V_i}\simeq A_{V_i}.

For a finite cover of VV, apply Theorem 2.32.2 over the affinoid AVA_V to its finite module AV⊗AMA_V\otimes_A M. Ordinary tensor associativity identifies the degree-zero and degree-one terms with the claimed restrictions. Exactness at these degrees is precisely the sheaf equalizer condition. The empty object has zero algebra, so its sheaf value is zero. Proposition 2.32.3 proves exactness of the functor on each object, hence exactness as a functor to sheaves. At V=XV=X, the identity presentation gives AX=AA_X=A, so its sections are MM. Taking M=AM=A, the restrictions are algebra homomorphisms and the equalizer respects multiplication, giving the asserted ring sheaf. ◻

Generalized Gerritzen–Grauert, Theorem 2.36.6, provides the rational refinement used to pass from this result to the affinoid-cover theorem 2.38.1. Effective module descent then gives Kiehl equivalence on the affinoid-domain site, and Theorem 2.39.6 carries it to the full analytic G-site.

2.33 Completed tensors, scalar extension and fibres

The algebra of an analytic fibre is obtained by changing coefficients to a completed residue field. We first establish the norm estimates needed for this construction. Throughout this section, KK is complete for its specified real-valued non-Archimedean absolute value, which may be trivial. Vector-space norms are homogeneous for that absolute value.

Finite-dimensional projections and tensor norms

For normed KK-spaces E,FE,F, the projective tensor seminorm is ∥z∥π=inf⁡z=∑j=1mej⊗fjmax⁡1≤j≤m∥ej∥∥fj∥.\|z\|_\pi=\inf_{z=\sum_{j=1}^m e_j\otimes f_j} \max_{1\leq j\leq m}\|e_j\|\|f_j\|. Its separated completion is denoted E⊗^KFE\widehat\otimes_K F. The next argument is the finite-dimensional form of Gruson’s tensor theorem; compare [GrP74] (§5, Lemma 5.2).

Lemma 2.33.1. NAG-001680 (Almost contractive finite-dimensional projections). Let VV be a finite-dimensional normed KK-vector space and W⊂VW\subset V a subspace. For every real c>1c>1 there is a KK-linear map P:V→WP:V\to W with P∣W=1WP|_W=1_W and ∥P(v)∥≤c∥v∥\|P(v)\|\leq c\|v\| for every v∈Vv\in V.

Proof. First suppose dim⁡(V/W)=1\dim(V/W)=1. By the finite-dimensional assertion of Lemma 1.12.1, WW is complete and therefore closed in VV. Choose v0∉Wv_0\notin W and put δ=inf⁡w∈W∥v0−w∥>0\delta=\inf_{w\in W}\|v_0-w\|>0. There is w0∈Ww_0\in W such that v=v0−w0v=v_0-w_0 satisfies ∥v∥<cδ\|v\|<c\delta. Define P(w+av)=wP(w+av)=w. If a≠0a\ne 0, homogeneity and the definition of δ\delta give ∣a∣δ≤∥w+av∥|a|\delta\leq\|w+av\|. The same estimate is immediate for a=0a=0. Thus ∥P(w+av)∥≤max⁡{∥w+av∥,∣a∣∥v∥}≤c∥w+av∥.\|P(w+av)\| \leq\max\{\|w+av\|,|a|\|v\|\} \leq c\|w+av\|. For codimension d>1d>1, choose a flag W=V0⊂V1⊂⋯⊂Vd=VW=V_0\subset V_1\subset\cdots\subset V_d=V with successive codimension one. Compose the corresponding projections with bounds c1/dc^{1/d}. For W=VW=V use the identity; the zero-dimensional cases are immediate. ◻

Theorem 2.33.2. NAG-001681 (Isometric tensor embeddings). For any normed KK-spaces E,FE,F, the projective seminorm on E⊗KFE\otimes_K F is a norm and satisfies ∥e⊗f∥π=∥e∥∥f∥\|e\otimes f\|_\pi=\|e\|\|f\|. If E↪E′E\hookrightarrow E' and F↪F′F\hookrightarrow F' are isometric linear embeddings, then E⊗KF⟶E′⊗KF′,E⊗^KF⟶E′⊗^KF′E\otimes_K F\longrightarrow E'\otimes_K F', \qquad E\widehat\otimes_K F\longrightarrow E'\widehat\otimes_K F' are isometric embeddings; the image of the second map is closed.

Proof. Algebraic injectivity follows by extending vector-space bases. We first compare seminorms. Fix z∈E⊗KFz\in E\otimes_K F and finite-dimensional subspaces U⊂EU\subset E, V⊂FV\subset F containing the factors of one expression for zz. An arbitrary competing expression z=∑j=1mej′⊗fj′in E′⊗KF′.z=\sum_{j=1}^m e'_j\otimes f'_j \quad\text{in }E'\otimes_K F'. Let U′⊂E′U'\subset E' be spanned by UU and the ej′e'_j, and let V′⊂F′V'\subset F' be spanned by VV and the fj′f'_j. The equality holds already in U′⊗KV′U'\otimes_K V', by algebraic injectivity. For a fixed c>1c>1, Lemma 2.33.1 gives projections P:U′→UP:U'\to U, Q:V′→VQ:V'\to V of norm at most cc. Applying P⊗QP\otimes Q yields z=∑jP(ej′)⊗Q(fj′),∥z∥π,E,F≤c2max⁡j∥ej′∥∥fj′∥.z=\sum_jP(e'_j)\otimes Q(f'_j), \qquad \|z\|_{\pi,E,F} \leq c^2\max_j\|e'_j\|\|f'_j\|. Taking the infimum over competing expressions and then letting c↓1c\downarrow 1 gives ∥z∥π,E,F≤∥z∥π,E′,F′\|z\|_{\pi,E,F}\leq\|z\|_{\pi,E',F'}. The reverse inequality follows from the inclusion of the sets of expressions. Only projections between the displayed finite-dimensional spaces have been used.

On a finite-dimensional tensor U⊗KVU\otimes_K V, choose bases and their coordinate functionals. These functionals are bounded by Lemma 1.12.1. If the (i,j)(i,j)-coordinate of zz is nonzero, their tensor gives 0<∣(λi⊗μj)(z)∣≤∥λi∥op∥μj∥op∥z∥π.0<|(\lambda_i\otimes\mu_j)(z)| \leq\|\lambda_i\|_{\rm op}\|\mu_j\|_{\rm op}\|z\|_\pi. Thus the finite-dimensional projective seminorm is definite. Every nonzero tensor lies in such a subspace, whose inclusion is isometric by the preceding argument; this proves definiteness in general. For one-dimensional factors the norm is the product norm, and the same isometric inclusion proves the pure-tensor formula. An isometry extends to an isometry of metric completions. Its image is complete, hence closed. ◻

Theorem 2.33.3. NAG-001682 (Exactness of completed tensors over a field). Let 0→E→F→G→00\to E\to F\to G\to 0 be an exact sequence of Banach KK-spaces, with EE carrying the induced norm of a closed subspace of FF and GG the quotient norm. For every Banach KK-space HH, the sequence 0⟶E⊗^KH⟶F⊗^KH⟶G⊗^KH⟶00\longrightarrow E\widehat\otimes_K H \longrightarrow F\widehat\otimes_K H \longrightarrow G\widehat\otimes_K H \longrightarrow0 is exact, its injection is isometric, and its surjection induces an isometry on the quotient. In particular completed tensoring preserves norm-strict short exact sequences with any equivalent Banach norms.

Proof. Algebraic tensoring over KK is exact, as follows by choosing an algebraic linear splitting of F→GF\to G. On the algebraic quotient (F⊗KH)/(E⊗KH)(F\otimes_K H)/(E\otimes_K H), compare its quotient projective norm with the projective norm on G⊗KHG\otimes_K H. Contractivity of F→GF\to G gives one inequality. For the other, take an expression z=∑jgj⊗hjz=\sum_j g_j\otimes h_j and c>1c>1. Lift each nonzero gjg_j to fj∈Ff_j\in F with ∥fj∥≤c∥gj∥\|f_j\|\leq c\|g_j\|, and lift zero by zero. The tensor ∑jfj⊗hj\sum_j f_j\otimes h_j represents zz and has norm at most cmax⁡j∥gj∥∥hj∥c\max_j\|g_j\|\|h_j\|. Taking both infima and then c↓1c\downarrow 1 proves equality of the two quotient norms.

By Theorem 2.33.2, E⊗^KHE\widehat\otimes_K H embeds isometrically in F⊗^KHF\widehat\otimes_K H. Its image is exactly the closure of E⊗KHE\otimes_K H. Quotienting by this closed image gives a Banach space by Lemma 1.11.1. The image of the algebraic quotient is dense and has the quotient norm just calculated: for an algebraic tensor, taking distance to a subspace or to its closure gives the same infimum. Consequently this Banach quotient is the completion of G⊗KHG\otimes_K H for its projective norm, hence is G⊗^KHG\widehat\otimes_K H. This proves all three assertions. Uniform inequalities between equivalent norms pass to projective norms, quotient norms and completions, giving the last statement. ◻

Lemma 2.33.4. NAG-001683 (Scalar extension of weighted coefficient spaces). Let II be a countable set, let wi>0w_i>0, and let EE be a Banach KK-space. There is a natural isometric linear isomorphism Cw(I,K)⊗^KE≃Cw(I,E),(ai)i⊗e⟼(aie)i.\mathcal C_w(I,K)\widehat\otimes_K E \simeq \mathcal C_w(I,E),\qquad (a_i)_i\otimes e\longmapsto(a_ie)_i. For a complete valued extension L/KL/K, this gives the isometric algebra isomorphism K{r−1T}⊗^KL≃L{r−1T}K\{r^{-1}T\}\widehat\otimes_K L\simeq L\{r^{-1}T\}. Also E→E⊗^KLE\to E\widehat\otimes_K L, e↦e⊗1e\mapsto e\otimes 1, is isometric.

Proof. The displayed bilinear map is contractive for the weighted supremum norm. Write ϵi\epsilon_i for the coordinate vector of norm wiw_i. For a finitely supported tensor z=∑iϵi⊗eiz=\sum_i\epsilon_i\otimes e_i, its displayed expression gives ∥z∥π≤max⁡iwi∥ei∥\|z\|_\pi\leq\max_i w_i\|e_i\|. Conversely the iith coordinate functional on Cw(I,K)\mathcal C_w(I,K) has norm wi−1w_i^{-1}. Applying its tensor with 1E1_E to any representation gives wi∥ei∥≤∥z∥πw_i\|e_i\|\leq\|z\|_\pi. Hence equality holds. Finitely supported vectors are dense on the right. They are also dense on the left: truncate each of the finitely many coefficient families in an algebraic tensor, then use density of algebraic tensors in the completion. This proves the isometry and surjectivity. For Tate algebras it preserves multiplication first on polynomials and then by continuity. Finally, the pure-tensor formula of Theorem 2.33.2 gives ∥e⊗1∥=∥e∥\|e\otimes 1\|=\|e\|. ◻

Generalized affinoids over complete extensions

Proposition 2.33.5. NAG-001685 (Scalar extension and actual relation ideals). Let L/KL/K be a complete valued extension, and let A=K{r−1T}/IA=K\{r^{-1}T\}/I be a generalized KK-affinoid presentation with its quotient norm. Then AL:=A⊗^KL≃L{r−1T}/I L{r−1T}A_L:=A\widehat\otimes_K L \simeq L\{r^{-1}T\}/I\,L\{r^{-1}T\} is an isometric algebra isomorphism for the displayed quotient norm. The ideal on the right is the actual extended ideal, which is closed. In particular ALA_L is a generalized LL-affinoid algebra, and A→ALA\to A_L is isometric. With any equivalent admissible norm the same maps are bounded isomorphisms with bounded inverses.

Proof. Put D=K{r−1T}D=K\{r^{-1}T\}. Theorems 1.18.2 and 1.18.3 make II a closed finite DD-module. Choose generators f1,…,fsf_1,\ldots,f_s; the surjection Ds→ID^s\to I is norm-strict for the induced norm on II. Applying Theorem 2.33.3 to its kernel sequence and to 0→I→D→A→00\to I\to D\to A\to 0 gives (D⊗^KL)s↠I⊗^KL↪D⊗^KL⟶A⊗^KL⟶0.(D\widehat\otimes_K L)^s\twoheadrightarrow I\widehat\otimes_K L \hookrightarrow D\widehat\otimes_K L \longrightarrow A\widehat\otimes_K L\longrightarrow0 . The injection is isometric and closed; the last map is a quotient isometry. Under Lemma 2.33.4, the first map is the row (f1,…,fs)(f_1,\ldots,f_s) over L{r−1T}L\{r^{-1}T\}. Its image is therefore exactly the algebraic ideal generated by the fif_i, not just its closure. The last quotient is the asserted presentation, and the algebra operations agree on dense polynomial tensors. The scalar embedding is isometric by the same lemma. Equivalent norms give the final assertion by taking the corresponding projective and quotient infima. ◻

Theorem 2.33.6. NAG-001686 (Finite modules and faithful scalar extension). For every complete valued extension L/KL/K and generalized KK-affinoid AA, the ring map A→ALA\to A_L is faithfully flat. For every finite canonical AA-module MM there are natural bounded isomorphisms AL⊗AM≃AL⊗^AM≃M⊗^KL.A_L\otimes_A M \simeq A_L\widehat\otimes_A M \simeq M\widehat\otimes_K L . Their inverses are bounded. For the quotient norm from a fixed finite free presentation of MM, and the corresponding quotient norm after extension, the comparisons are isometric. They identify short exact sequences of finite modules with norm-strict short exact sequences after either ordinary or completed scalar extension.

Proof. Choose a finite presentation As→DAm→qM→0A^s\xrightarrow{D}A^m\xrightarrow{q}M\to 0, and give MM its quotient norm. Its relation module R=ker⁡qR=\ker q is closed and finite; both As→RA^s\to R and qq are norm-strict by Theorem 1.18.3. Apply Theorem 2.33.3 to these two kernel sequences. Finite coordinate sums commute isometrically with completed scalar extension, either by their coordinate projections or by Lemma 2.33.4 with finite II. We obtain ALs→DLALm⟶M⊗^KL⟶0,M⊗^KL≃ALm/DLALsA_L^s\xrightarrow{D_L}A_L^m \longrightarrow M\widehat\otimes_K L\longrightarrow0, \qquad M\widehat\otimes_K L\simeq A_L^m/D_LA_L^s with closed relation image and the indicated quotient norm. Ordinary tensor right exactness, [StacksAlg] (Tag 00DF), identifies this algebraic quotient with AL⊗AMA_L\otimes_A M. The map is the natural one on pure tensors. Proposition 1.18.7, with target ALA_L, identifies its ordinary and completed tensors over AA, including the isometry for the chosen norms. Naturality follows on generators and then by continuity. Theorem 2.33.3 and the finite-module theorem give the exact-sequence assertion.

For a finitely generated ideal J⊂AJ\subset A, the algebraic map J⊗AAL→ALJ\otimes_A A_L\to A_L is now identified with the injective map J⊗^KL→A⊗^KLJ\widehat\otimes_K L\to A\widehat\otimes_K L. The induced norm on JJ is canonical, so Theorem 2.33.2 applies. The finitely generated ideal criterion [StacksAlg] (Tag 00HD, Lemma 10.39.5(4)) proves flatness on all algebraic AA-modules. If JJ is proper, the isometric map A/J↪(A/J)⊗^KL≃AL/JALA/J\hookrightarrow (A/J)\widehat\otimes_K L \simeq A_L/JA_L shows that the quotient on the right is nonzero. In particular the fibre at every maximal ideal is nonzero. The faithful-flatness criterion [StacksAlg] (Tag 00HP, Lemma 10.39.15) completes the proof. For the zero algebra the module category is zero and the same assertions hold directly. ◻

Lemma 2.33.7. NAG-001687 (Generalized relative presentations). Every bounded homomorphism A→BA\to B of generalized KK-affinoid algebras extends to a bounded norm-strict surjection A{r1−1T1,…,rn−1Tn}↠BA\{r_1^{-1}T_1,\ldots,r_n^{-1}T_n\}\twoheadrightarrow B for suitable positive radii rir_i. Its kernel is an actual closed ideal, and its quotient norm is equivalent to the given admissible norm on BB.

Proof. Choose an admissible presentation K{r−1T}↠BK\{r^{-1}T\}\twoheadrightarrow B, and temporarily use its quotient norm on BB and a presentation norm on AA. The variable images bib_i satisfy ∥bim∥≤rim\|b_i^m\|\leq r_i^m. The given bounded map on coefficients therefore extends by Proposition 1.9.3 to the displayed bounded map. It is surjective because it extends the original presentation. For a fixed c>1c>1, every nonzero b∈Bb\in B has a lift in K{r−1T}K\{r^{-1}T\} of norm at most c∥b∥c\|b\|. Its image in A{r−1T}A\{r^{-1}T\} has the same norm, since the coefficient inclusion K↪AK\hookrightarrow A is isometric. These lifts and boundedness of the map compare its quotient norm with the norm of BB in both directions. Proposition 1.17.3 makes the coefficient algebra affinoid, and Theorem 1.18.2 makes its kernel ideal closed. Equivalent initial norms only change the constants. Zero algebras are immediate. ◻

Proposition 2.33.8. NAG-001688 (Pushouts with complete-extension affinoids). Let A→BA\to B be a bounded map of generalized KK-affinoid algebras. Let L/KL/K be a complete valued extension, let CC be a generalized LL-affinoid algebra, and let A→CA\to C be a bounded KK-homomorphism. Then the separated completed projective tensor algebra B⊗^ACB\widehat\otimes_A C is a generalized LL-affinoid algebra. If B=A{r−1T}/IB=A\{r^{-1}T\}/I is a norm-admissible relative presentation, there is a natural bounded algebra isomorphism, with bounded inverse, B⊗^AC≃C{r−1T}/I C{r−1T}.B\widehat\otimes_A C \simeq C\{r^{-1}T\}/I\,C\{r^{-1}T\}. The extended ideal is actual and closed. For the displayed quotient norm on BB, the coefficient norm on C{r−1T}C\{r^{-1}T\}, and a contractive map A→CA\to C, this isomorphism is isometric. The tensor algebra is the pushout for compatible bounded homomorphisms to commutative Banach KK-algebras. Zero algebras are allowed.

Proof. Lemma 2.33.7 supplies the relative presentation. Put G=B⊗^ACG=B\widehat\otimes_A C and H=C{r−1T}/I C{r−1T}H=C\{r^{-1}T\}/I\,C\{r^{-1}T\}. By Proposition 1.17.3 and Corollary 1.18.4, applied over LL, HH is a complete actual quotient and is LL-affinoid. On the algebraic tensor, multiplying finite tensor expressions proves submultiplicativity of the projective seminorm. Its zero seminorm subspace is an ideal, so separation and completion make GG a Banach algebra.

The natural maps B→HB\to H, C→HC\to H are bounded and agree on AA. Their balanced product extends to a bounded map G→HG\to H. In the other direction, the elements Tˉi⊗1\bar T_i\otimes 1 in GG have their mmth powers bounded by rimr_i^m when BB has its relative quotient norm. Substitution, with coefficients in CC, gives a bounded map C{r−1T}→GC\{r^{-1}T\}\to G. The two maps from A{r−1T}A\{r^{-1}T\} to GG agree on coefficients and variables and hence, by polynomial density, everywhere. Therefore the substitution kills II and its extended ideal. It descends boundedly to H→GH\to G by taking quotient infima. The composites fix every coefficient and variable in HH and every pure tensor in GG. Density makes them identity maps. When A→CA\to C is contractive, every map just constructed is contractive, giving the isometry. Equivalent admissible norms give the stated bounded comparisons.

Finally, compatible bounded maps B,C→DB,C\to D to a commutative Banach KK-algebra have bounded balanced product (b,c)↦bc(b,c)\mapsto b c. On any finite tensor expression its norm is bounded by the product of their operator bounds times the projective bound. Taking the infimum, separating and completing gives a unique bounded map G→DG\to D. Uniqueness follows from the dense algebraic tensor image. ◻

Corollary 2.33.9. NAG-001689 (Iterated base change). For compatible bounded affinoid maps A→BA\to B, A→C→DA\to C\to D, allowing complete coefficient-field extensions for C,DC,D, there is a natural bounded isomorphism with bounded inverse (B⊗^AC)⊗^CD≃B⊗^AD.(B\widehat\otimes_A C)\widehat\otimes_C D \simeq B\widehat\otimes_A D . For complete valued extensions K⊂L⊂L′K\subset L\subset L', this includes (AL)L′≃AL′(A_L)_{L'}\simeq A_{L'}. All comparison maps commute with the structural homomorphisms.

Proof. Both algebras represent compatible bounded homomorphisms from BB and DD agreeing on AA, by Proposition 2.33.8. Applying that universal property in both directions gives inverse maps; their composites are identities by uniqueness. The same argument applies to the scalar-extension pushouts. ◻

Residue fields and pointwise fibres

Recall that H(x)\mathcal H(x) is the completion of the valued fraction field of A/ker⁡xA/\ker x. This completion is a field: a Cauchy sequence with nonzero limit has eventually constant positive absolute value, and ∣an−1−am−1∣=∣an−am∣∣an∣ ∣am∣|a_n^{-1}-a_m^{-1}| =\frac{|a_n-a_m|}{|a_n|\,|a_m|} shows that its inverse sequence is Cauchy. The field operations and multiplicativity pass to limits. Proposition 1.6.2 therefore applies to every bounded evaluation in a complete valued target.

Proposition 2.33.10. NAG-001691 (Rational restriction preserves completed residue fields). Let U⊂M(A)U\subset\mathcal M(A) be a generalized rational domain, represented by A→AUA\to A_U. At every x∈Ux\in U, the induced isometric embedding HA(x)⟶HAU(x)\mathcal H_A(x)\longrightarrow\mathcal H_{A_U}(x) is an isomorphism.

Proof. Write AU=A{ρ−1T}/(gTi−fi)A_U=A\{\rho^{-1}T\}/(gT_i-f_i) with (g,f1,…,fs)=A(g,f_1,\ldots,f_s)=A. At x∈Ux\in U the denominator is nonzero, and ∣fi(x)/g(x)∣≤ρi|f_i(x)/g(x)|\leq\rho_i. Evaluation at these ratios gives a bounded map AU→HA(x)A_U\to\mathcal H_A(x) by Proposition 1.9.3 and quotient infima. Its seminorm is the unique rational point over xx by Theorem 2.31.1. Proposition 1.6.2 gives an isometric map HAU(x)→HA(x)\mathcal H_{A_U}(x)\to\mathcal H_A(x). Its composite with the map in the statement fixes the dense field Frac⁡(A/ker⁡x)\operatorname{Frac}(A/\ker x) and is therefore the identity. The reverse composite fixes the image of AUA_U: this follows first on the coefficients and the forced ratios fi/gf_i/g, then on their convergent power series. It consequently fixes the dense fraction field of the evaluated AUA_U, hence is the identity. ◻

Theorem 2.33.11. NAG-001692 (Affinoid fibres at all Berkovich points). Let ϕ:A→B\phi:A\to B be a bounded homomorphism of generalized KK-affinoid algebras, with spectral map f:M(B)→M(A)f:\mathcal M(B)\to\mathcal M(A). For x∈M(A)x\in\mathcal M(A) put F=H(x)F=\mathcal H(x) and Bx=B⊗^AF.B_x=B\widehat\otimes_A F . Then BxB_x is a generalized FF-affinoid algebra. Restriction induces a homeomorphism MF(Bx)→ ∼ f−1(x).\mathcal M_F(B_x)\xrightarrow{\ \sim\ }f^{-1}(x). For the point zz corresponding to y∈f−1(x)y\in f^{-1}(x), the structural map gives a canonical isometric isomorphism HB(y)≃HBx(z)\mathcal H_B(y)\simeq\mathcal H_{B_x}(z) compatible with the embedding F↪HB(y)F\hookrightarrow\mathcal H_B(y).

Proof. The map A→FA\to F is bounded, and FF is affinoid over itself. Proposition 2.33.8 proves the affinoid assertion; in a relative presentation B=A{r−1T}/IB=A\{r^{-1}T\}/I, it gives Bx=F{r−1T}/IFB_x=F\{r^{-1}T\}/I_F.

For y∈f−1(x)y\in f^{-1}(x), Proposition 1.6.2 supplies an isometric embedding j:F↪HB(y)j:F\hookrightarrow\mathcal H_B(y) agreeing with A→B→HB(y)A\to B\to\mathcal H_B(y). The compatible maps from BB and FF extend uniquely to a bounded homomorphism ρy:Bx→HB(y)\rho_y:B_x\to\mathcal H_B(y). Its absolute value defines a point zy∈MF(Bx)z_y\in\mathcal M_F(B_x) restricting to yy. Conversely, a point z∈MF(Bx)z\in\mathcal M_F(B_x) restricts to a point y∈M(B)y\in\mathcal M(B) with f(y)=xf(y)=x, since its values on the factor FF are the prescribed absolute value of FF.

To prove uniqueness, let i:HB(y)↪HBx(z)i:\mathcal H_B(y)\hookrightarrow\mathcal H_{B_x}(z) be the residue embedding induced by B→BxB\to B_x. The structural embedding F→HBx(z)F\to\mathcal H_{B_x}(z) and i∘ji\circ j agree on AA, hence on Frac⁡(A/ker⁡x)\operatorname{Frac}(A/\ker x), and then on FF by density. Consequently evaluation at zz equals i∘ρyi\circ\rho_y on B⊗AFB\otimes_A F and, by continuity, on BxB_x. Its seminorm is therefore zyz_y. Moreover the image i(HB(y))i(\mathcal H_B(y)) is a closed field containing the evaluated image of BxB_x. The closure of its fraction field is HBx(z)\mathcal H_{B_x}(z); hence ii is surjective. This proves the residue-field assertion as well as bijectivity.

Restriction is continuous, since evaluation on b∈Bb\in B is evaluation on b⊗1b\otimes 1. The source is compact and the target is a subspace of the Hausdorff space M(B)\mathcal M(B) by Proposition 2.24.1. A continuous bijection from a compact space to a Hausdorff space is a homeomorphism: closed subsets of the source are compact, and their images are compact and therefore closed. The argument includes the empty fibre, when Bx=0B_x=0. ◻

Proposition 2.33.12. NAG-001693 (Rational domains under complete-extension base change). Let U⊂M(A)U\subset\mathcal M(A) be a generalized rational domain, let CC be a generalized affinoid over a complete extension L/KL/K, and let ϕ:A→C\phi:A\to C be bounded. Then AU⊗^AC≃Cϕ−1(U)A_U\widehat\otimes_A C\simeq C_{\phi^{-1}(U)} as bounded CC-algebras, where the domain on the right is the inverse image under the spectral map. The comparison commutes with restriction to nested rational domains and finite intersections. Finite rational covers pull back to finite rational covers. In particular, for π:M(AL)→M(A)\pi:\mathcal M(A_L)\to\mathcal M(A), (AU)L≃(AL)π−1(U).(A_U)_L\simeq (A_L)_{\pi^{-1}(U)}.

Proof. For the presentation AU=A{ρ−1T}/(gTi−fi)A_U=A\{\rho^{-1}T\}/(gT_i-f_i), Proposition 2.33.8 gives the actual algebra C{ρ−1T}/(ϕ(g)Ti−ϕ(fi)).C\{\rho^{-1}T\}/\bigl(\phi(g)T_i-\phi(f_i)\bigr). The images of g,f1,…,fsg,f_1,\ldots,f_s still generate the unit ideal. By Theorem 2.31.1, its spectrum is precisely the locus ∣ϕ(fi)∣≤ρi∣ϕ(g)∣|\phi(f_i)|\leq\rho_i|\phi(g)|, namely the inverse image of UU. The uniqueness in Theorem 2.31.2 gives compatibility with restriction. The joint presentations of Proposition 2.31.3 give the same statement for intersections. The pointwise inverse images of a cover cover the target. For C=ALC=A_L, both (AU)L(A_U)_L and AU⊗^AALA_U\widehat\otimes_A A_L are the pushout of AUA_U and LL over KK; uniqueness of the bounded pushout identifies them and proves the final formula. ◻

Theorem 2.33.13. NAG-001694 (Surjectivity after an arbitrary complete extension). For every generalized KK-affinoid AA and every complete valued extension L/KL/K, the natural map π:ML(AL)⟶MK(A)\pi:\mathcal M_L(A_L)\longrightarrow\mathcal M_K(A) is a continuous closed surjection.

Proof. Continuity follows from the evaluation topology, and closedness follows from compactness of the source and the Hausdorff property of the target, Proposition 2.24.1. Fix x∈M(A)x\in\mathcal M(A). Consider all the closed rational domains U+(a,r)={v:∣a(v)∣≤r}(r>∣a(x)∣),U−(a,r)={v:∣a(v)∣≥r}(0<r<∣a(x)∣).\begin{aligned} U^+(a,r)&=\{v:|a(v)|\leq r\}&&\quad(r>|a(x)|),\\ U^-(a,r)&=\{v:|a(v)|\geq r\}&&\quad(0<r<|a(x)|). \end{aligned} Every finite intersection UU of these domains is rational by Proposition 2.31.3 and contains xx. Its algebra AUA_U is nonzero. Lemma 2.33.4 gives an isometric embedding AU↪(AU)LA_U\hookrightarrow(A_U)_L, so the latter is nonzero. It is LL-affinoid by Proposition 2.33.5, and hence has a point by Theorem 1.20.1. By Proposition 2.33.12, this supplies a point of π−1(U)\pi^{-1}(U). Thus the inverse images of the displayed domains have the finite intersection property in the compact space ML(AL)\mathcal M_L(A_L).

Choose a point yy in their intersection. For every a∈Aa\in A, all upper inequalities give ∣a(π(y))∣≤∣a(x)∣|a(\pi(y))|\leq|a(x)|. If ∣a(x)∣>0|a(x)|>0, all lower inequalities give the reverse inequality; if ∣a(x)∣=0|a(x)|=0, the upper inequalities already force equality. Hence π(y)=x\pi(y)=x. This proves surjectivity. If A=0A=0, both spectra are empty and the assertion is immediate. ◻

Proposition 2.33.14. NAG-001695 (Affinoid products and pairs of points). For bounded maps A→B,A→CA\to B,A\to C of generalized KK-affinoid algebras, M(B⊗^AC)\mathcal M(B\widehat\otimes_A C) is their fibre product in the affinoid category. The natural map to the ordinary topological fibre product M(B⊗^AC)⟶M(B)×M(A)topM(C)\mathcal M(B\widehat\otimes_A C) \longrightarrow \mathcal M(B)\times_{\mathcal M(A)}^{\mathrm{top}}\mathcal M(C) is continuous, closed and surjective.

Proof. The categorical assertion is the opposite of the bounded pushout property in Proposition 2.33.8. Restriction gives the displayed continuous map. Its target is a closed subspace of the compact Hausdorff product M(B)×M(C)\mathcal M(B)\times\mathcal M(C): it is the inverse image of the diagonal of M(A)\mathcal M(A). Compactness of the source therefore proves closedness.

Take y∈M(B)y\in\mathcal M(B) and z∈M(C)z\in\mathcal M(C) mapping to x∈M(A)x\in\mathcal M(A), and put F=H(x)F=\mathcal H(x), E=H(y)E=\mathcal H(y). Theorem 2.33.11 identifies zz with a point of the FF-affinoid Cx=C⊗^AFC_x=C\widehat\otimes_A F. The isometric embedding F↪EF\hookrightarrow E makes E/FE/F a complete valued extension. By Theorem 2.33.13, the point zz lifts to a point ww of Cx⊗^FEC_x\widehat\otimes_F E. In H(w)\mathcal H(w) the maps from CC and B→EB\to E agree on AA. Their bounded product therefore gives a point of M(B⊗^AC)\mathcal M(B\widehat\otimes_A C). Its restrictions are zz and yy: the first follows from the choice of ww, and the second because evaluation on EE is its prescribed absolute value. This proves surjectivity. ◻

Ordinary stalks and domain flatness

Corollary 2.33.15. NAG-001697 (Completed residue fields on affinoid domains). If A→BA\to B represents an affinoid domain U⊂M(A)U\subset\mathcal M(A), then at every x∈Ux\in U the canonical map HA(x)→HB(x)\mathcal H_A(x)\to\mathcal H_B(x) is an isometric isomorphism.

Proof. Theorem 2.36.6 gives a finite cover of UU by domains rational both over AA and over BB. Choose a member containing xx, with algebra CC. Both maps HA(x)→HC(x)\mathcal H_A(x)\to\mathcal H_C(x) and HB(x)→HC(x)\mathcal H_B(x)\to\mathcal H_C(x) are isomorphisms by Proposition 2.33.10. Their triangle commutes on the evaluated elements of AA and therefore on its completed fraction field, which proves the assertion. ◻

Proposition 2.33.16. NAG-001698 (Ordinary affinoid stalks). Let X=M(A)X=\mathcal M(A), and restrict OXG\mathcal O_{X_G} to ordinary open subsets to obtain OX\mathcal O_X. For every x∈Xx\in X, OX,x≃lim→⁡V rational in Xx∈int⁡top(V)AV,\mathcal O_{X,x} \simeq\varinjlim_{\substack{V\text{ rational in }X\\ x\in\operatorname{int}_{\rm top}(V)}} A_V, where the system is directed by shrinking. Evaluation defines a local ring structure: its maximal ideal consists of germs vanishing at xx. Its residue field κX(x)\kappa_X(x) embeds isometrically in HA(x)\mathcal H_A(x) and has completion HA(x)\mathcal H_A(x).

Proof. Proposition 2.39.3 constructs, inside every ordinary neighborhood of xx, a rational domain containing an ordinary neighborhood of xx. Finite intersections have the same property. Theorem 2.39.4 identifies the sections on these domains with AVA_V. A section on an ordinary open neighborhood restricts to one of them. Conversely, a section on such a VV restricts to its ordinary interior near xx. These restrictions induce mutually inverse maps on germs, proving the displayed formula.

By Proposition 2.33.10, the evaluations on all AVA_V take values compatibly in HA(x)\mathcal H_A(x), so they define a ring map OX,x→HA(x)\mathcal O_{X,x}\to\mathcal H_A(x). If a germ is represented by b∈AVb\in A_V with b(x)≠0b(x)\ne 0, choose 0<c<∣b(x)∣0<c<|b(x)|. The Laurent domain ∣b∣≥c|b|\geq c in VV contains a neighborhood of xx in XX and has an inverse for bb in its algebra. Hence the germ is a unit. A germ vanishing at xx cannot be a unit, since evaluation is a unital map into a field. The kernel of evaluation is consequently exactly the unique maximal ideal.

It follows that κX(x)\kappa_X(x) is identified with the evaluated image, a subfield of HA(x)\mathcal H_A(x). It contains Frac⁡(A/ker⁡x)\operatorname{Frac}(A/\ker x): each denominator nonzero at xx has the local inverse just constructed. This fraction field is dense in HA(x)\mathcal H_A(x) by definition. The intermediate field κX(x)\kappa_X(x) is therefore dense as well, proving the completion assertion. ◻

Theorem 2.33.17. NAG-001699 (Domain flatness on ordinary local rings). Let U⊂X=M(A)U\subset X=\mathcal M(A) be an affinoid domain. For every x∈Ux\in U, the local homomorphism OX,x⟶OU,x\mathcal O_{X,x}\longrightarrow\mathcal O_{U,x} is flat. This assertion remains true after every bounded affinoid base change over every complete valued extension of KK.

Proof. Write R=OX,xR=\mathcal O_{X,x} and S=OU,xS=\mathcal O_{U,x}. Fix a rational neighborhood VV of xx in XX. Affinoid neighborhoods WW of xx in UU contained in VV form a cofinal system: intersect an ordinary neighborhood with VV, then choose a rational neighborhood inside the affinoid space UU, as in Proposition 2.39.3. By transitivity and restriction of affinoid domains, each such WW is an affinoid domain of VV. Proposition 2.38.2 makes AV→AWA_V\to A_W flat.

We verify flatness of S=lim→⁡WAWS=\varinjlim_W A_W over this fixed AVA_V by the finite-ideal criterion [StacksAlg] (Tag 00HD). For a finitely generated ideal J⊂AVJ\subset A_V, every element of J⊗AVSJ\otimes_{A_V}S is represented by a finite sum at some stage AWA_W. If it maps to zero in SS, its image vanishes at a smaller neighborhood W′W'. Flatness of AW′A_{W'} implies that its representative in J⊗AVAW′J\otimes_{A_V}A_{W'} is zero. Thus J⊗AVS→SJ\otimes_{A_V}S\to S is injective. This also spells out the exactness of the filtered passage being used.

Now take a finitely generated ideal I⊂RI\subset R and z∈I⊗RSz\in I\otimes_R S mapping to zero. Choose generators a1,…,ama_1,\ldots,a_m of II and write z=∑iai⊗siz=\sum_i a_i\otimes s_i with si∈Ss_i\in S. By Proposition 2.33.16, the aia_i lift to elements a~i\widetilde a_i of one AVA_V. Put J=(a~1,…,a~m)J=(\widetilde a_1,\ldots,\widetilde a_m). Then zV=∑ia~i⊗si∈J⊗AVSz_V=\sum_i\widetilde a_i\otimes s_i\in J\otimes_{A_V}S maps to zz and has zero image in SS. We have just proved SS flat over AVA_V, so zV=0z_V=0, and hence z=0z=0. Again applying [StacksAlg] (Tag 00HD) proves RR-flatness. The map is local because both maximal ideals are defined by vanishing at xx, and their residue evaluations are compatible by Corollary 2.33.15.

For a bounded map A→CA\to C with CC affinoid over a complete extension, Theorem 2.36.9 identifies the inverse image of UU with an affinoid domain of M(C)\mathcal M(C). Apply the same proof over the complete coefficient field of CC. ◻

Finite morphisms and relative interior

Theorem 2.33.18. NAG-001701 (Finite algebras with their canonical norm). Let AA be a generalized KK-affinoid algebra. Every commutative finite AA-algebra DD admits a complete submultiplicative norm, with ∥1∥=1\|1\|=1 when D≠0D\ne 0, equivalent to its canonical finite-AA-module norm. With this norm it is a generalized KK-affinoid algebra and A→DA\to D is bounded. The resulting Banach structure is independent, up to bounded isomorphism, of the chosen canonical module norm. Every AA-linear map between finite algebras with these structures is bounded. Moreover Int⁡(M(D)/M(A))=M(D).\operatorname{Int}\bigl(\mathcal M(D)/\mathcal M(A)\bigr) =\mathcal M(D).

Proof. Give AA a presentation norm, and DD a quotient norm from a finite free AA-module. For d∈Dd\in D, multiplication is an AA-linear map Ld:D→DL_d:D\to D and is bounded by Theorem 1.18.3. Choose module generators e1,…,eme_1,\ldots,e_m. Controlled lifts of d=∑iaieid=\sum_i a_i e_i and contractivity of the AA-action give a constant CC independent of dd with ∥Ld∥op≤C∥d∥,∥d∥=∥Ld(1)∥≤∥Ld∥op∥1∥.\|L_d\|_{\rm op}\leq C\|d\|,\qquad \|d\|=\|L_d(1)\|\leq\|L_d\|_{\rm op}\|1\|. Thus ∥d∥∗=∥Ld∥op\|d\|_*=\|L_d\|_{\rm op} is equivalent to the complete module norm. It is homogeneous and submultiplicative, since Ldd′=LdLd′L_{dd'}=L_dL_{d'}, and ∥1∥∗=1\|1\|_*=1. The map A→DA\to D is contractive for this norm. Treat the zero algebra separately with its zero norm.

Choose real radii ri>∥ei∥∗r_i>\|e_i\|_*. Substitution gives a bounded surjection A{r−1T}→DA\{r^{-1}T\}\to D, Ti↦eiT_i\mapsto e_i. It is norm-strict: a controlled module lift d=∑iaieid=\sum_i a_i e_i is represented by the linear polynomial ∑iaiTi\sum_i a_i T_i, whose coefficient norm is at most (max⁡iri)max⁡i∥ai∥(\max_i r_i)\max_i\|a_i\|. Its kernel is closed, and the coefficient algebra is affinoid by Proposition 1.17.3. Corollary 1.18.4 now realizes DD as the required affinoid quotient with an equivalent norm.

Every point y∈M(D)y\in\mathcal M(D) satisfies ∣ei(y)∣≤∥ei∥∗<ri|e_i(y)|\leq\|e_i\|_*<r_i, by Lemma 1.6.1. The same relative presentation therefore puts every yy in the interior of Definition 2.7.1. Equivalence of canonical module norms and boundedness of all linear maps follow from Theorem 1.18.3. ◻

Proposition 2.33.19. NAG-001702 (General finite base change). Let A→BA\to B be a bounded map of generalized KK-affinoid algebras, where BB is finite over AA and its given norm is equivalent to its canonical finite-AA-module norm. For every bounded map A→CA\to C to an affinoid over a complete extension L/KL/K, the natural map B⊗AC⟶B⊗^ACB\otimes_A C\longrightarrow B\widehat\otimes_A C is a bounded algebra isomorphism with bounded inverse when the source has its canonical finite-CC-module norm. The result is finite and norm-admissible over CC.

Proof. Apply Proposition 1.18.7 to the finite canonical AA-module BB and the target CC. It identifies the projective tensor norm with the complete canonical finite-CC-module norm up to uniform equivalence. The separated completion consequently changes no elements. Multiplication agrees on the algebraic tensor and its completion by continuity. Proposition 2.33.8 gives the affinoid structure, and the norm comparison just obtained gives finite admissibility. ◻

Example 2.33.20. (Algebraic finiteness and the specified norm). Suppose the absolute value of KK is trivial and 0<s<r<10<s<r<1. As rings, Ar=K{r−1T}=K[[T]]=K{s−1T}=As.A_r=K\{r^{-1}T\}=K[[T]]=K\{s^{-1}T\}=A_s. The identity Ar→AsA_r\to A_s is a bounded algebraic isomorphism, hence algebraically finite, but it is not norm-strict. Indeed, every formal power series converges for each of these radii, and for a nonzero series of order nn its two norms are rnr^n and sns^n. The identity is contractive, whereas an inverse bound would require (r/s)n(r/s)^n to remain bounded. Thus the finite-module norm in Proposition 2.33.19 is a genuine condition when the valuation is trivial.

Proposition 2.33.21. NAG-001704 (Presentation calculus for relative interior). For affinoid morphisms, the relative interior of Definition 2.7.1 is open and is unchanged by bounded isomorphisms of the given Banach-algebra presentations. For composable affinoid morphisms Z→gY→fXZ\xrightarrow{g}Y\xrightarrow{f}X, Int⁡(Z/Y)∩g−1Int⁡(Y/X)⊆Int⁡(Z/X).\operatorname{Int}(Z/Y)\cap g^{-1}\operatorname{Int}(Y/X) \subseteq\operatorname{Int}(Z/X). If Y=M(B)Y=\mathcal M(B), X=M(A)X=\mathcal M(A) and X′=ML(C)X'=\mathcal M_L(C) for a bounded A→CA\to C over a complete extension L/KL/K, then, with Y′=ML(B⊗^AC)Y'=\mathcal M_L(B\widehat\otimes_A C) and projection p:Y′→Yp:Y'\to Y, p−1Int⁡(Y/X)⊆Int⁡(Y′/X′).p^{-1}\operatorname{Int}(Y/X) \subseteq\operatorname{Int}(Y'/X').

Proof. For each norm-admissible relative presentation, the locus defined by its finitely many strict inequalities is open in the evaluation topology. Relative interior is the union of these loci over all presentations, so it is open. A bounded algebra isomorphism with bounded inverse transports norm-admissible surjections to such surjections and preserves the pointwise inequalities. This proves the invariance assertion.

For composition, write X=M(A)X=\mathcal M(A), Y=M(B)Y=\mathcal M(B), Z=M(D)Z=\mathcal M(D). At a point in the left side choose presentations A{r−1T}↠BA\{r^{-1}T\}\twoheadrightarrow B and B{s−1S}↠DB\{s^{-1}S\}\twoheadrightarrow D whose coordinate values there satisfy the strict inequalities. Coefficientwise controlled lifts, Proposition 2.14.1, give a norm-strict surjection A{r−1T,s−1S}⟶B{s−1S}⟶D.A\{r^{-1}T,s^{-1}S\} \longrightarrow B\{s^{-1}S\}\longrightarrow D . The iterated coefficient algebra equals the displayed joint algebra by Lemma 1.17.1. Composition multiplies the lifting bounds, so the composite is norm-strict. Both lists of coordinates satisfy the required strict inequalities at the original point, proving the inclusion.

For base change, use a relative presentation witnessing that p(y′)∈Int⁡(Y/X)p(y')\in\operatorname{Int}(Y/X). Proposition 2.33.8 identifies its base change with the norm-admissible presentation C{r−1T}↠B⊗^ACC\{r^{-1}T\}\twoheadrightarrow B\widehat\otimes_A C. The values of every coordinate at y′y' equal its values at p(y′)p(y'). The same strict inequalities therefore witness that y′y' is interior over X′X'. ◻

Proposition 2.33.22. NAG-001705 (Affinoid diagonals and finite proper maps). Every affinoid morphism is separated. A finite norm-admissible affinoid morphism is proper, and remains finite, separated and proper after every affinoid base change over a complete valued extension.

Proof. For A→BA\to B, the diagonal is represented by multiplication μ:B⊗^AB→B\mu:B\widehat\otimes_A B\to B. It is bounded, and b↦b⊗1b\mapsto b\otimes 1 is a bounded linear section. Thus μ\mu is a norm-strict surjection: the section gives a controlled lift for every bb. Its kernel is a closed ideal, so the diagonal is a closed immersion. This is separatedness.

If BB is finite and norm-admissible over AA, Theorem 2.33.18 gives empty relative boundary, and its conclusion is preserved by the equivalent given norm. The map of spectra is continuous between compact Hausdorff spaces. The inverse image of any compact analytic domain in the target is closed in the compact source and hence compact. Together with boundarylessness this is properness. Proposition 2.33.19 preserves finite admissibility under each specified base change, so the same argument applies after base change. ◻

Analytification from affinoid exhaustions

The following construction expands the proof indicated in [TemIntro] (Exercise 5.1.1.1). Every algebra occurring in a bounded piece is an actual quotient of a weighted Tate algebra. Passing from those pieces to an analytic space uses the atlas gluing of Theorem 2.40.10.

Theorem 2.33.23. NAG-001707 (Affine analytification and its mapping property). Let R=K[T1,…,Tn]/IR=K[T_1,\ldots,T_n]/I be a finite-type KK-algebra. The set XRX_R of multiplicative seminorms on RR extending the absolute value of KK, with its evaluation topology, has a canonical good Hausdorff KK-analytic structure. For every r∈R>0nr\in\mathbf R_{>0}^n, its locus ∣Ti∣≤ri|T_i|\le r_i is the affinoid space XR,r=M(K{r−1T}/IK{r−1T}).X_{R,r}=\mathcal M\bigl(K\{r^{-1}T\}/I K\{r^{-1}T\}\bigr). These pieces form an exhaustion by affinoid neighbourhoods. The construction is independent of the presentation and represents the functor Hom⁡K-an(Y,XR)≃Hom⁡K-alg(R,Γ(YG,OY))\operatorname{Hom}_{K\text{-an}}(Y,X_R) \simeq \operatorname{Hom}_{K\text{-alg}} \bigl(R,\Gamma(Y_G,\mathcal O_Y)\bigr) for every KK-analytic space YY.

Proof. If xx is a seminorm with the stated coordinate bounds, then every polynomial satisfies ∣p(x)∣≤max⁡ν∣aν∣rν|p(x)|\le\max_\nu |a_\nu|r^\nu. Thus evaluation extends continuously to the weighted Tate algebra, uniquely by polynomial density, and kills the actual ideal IK{r−1T}I K\{r^{-1}T\}. Conversely, a point of the displayed affinoid quotient gives precisely such a seminorm on RR. This identifies the sets. Its topology agrees with the topology induced by polynomial evaluations: for a convergent series hh and a polynomial truncation pp, the inequality ∣∣h(x)∣−∣p(x)∣∣≤∥h−p∥||h(x)|-|p(x)||\le\|h-p\| holds uniformly on this piece. All series evaluations are therefore uniform limits of polynomial evaluations.

If ri≤sir_i\le s_i for all ii, adjoining variables UiU_i of radii rir_i with relations Ui−Ti=0U_i-T_i=0 identifies XR,rX_{R,r} with a Weierstrass domain of XR,sX_{R,s}. Both bounded maps between the resulting presentations are given by these variables; they are inverse on the dense polynomials. Arbitrary intersections are the pieces with radii min⁡(ri,si)\min(r_i,s_i). Every point has an open neighbourhood ∣Ti∣<ri|T_i|<r_i contained in one such piece, by increasing all radii past its coordinate values. Thus the pieces and their affinoid subdomains form an affinoid net. Their identifications satisfy the cocycle condition on the coordinates, giving the asserted analytic space by the atlas construction. Its evaluation topology is Hausdorff because distinct seminorms differ on an element of RR, and the affinoid neighbourhoods show goodness. Scalar radii ri=mr_i=m, for positive integers mm, already give an exhaustion.

First let Y=M(A)Y=\mathcal M(A). Any homomorphism R→AR\to A sends its finitely many generators to elements of finite norm. Choosing radii exceeding those norms gives a bounded map from the corresponding affinoid quotient to AA, and therefore an analytic map to XRX_R. Increasing the radii does not change the map. Conversely, the compact image of any analytic map M(A)→XR\mathcal M(A)\to X_R lies in some XR,mX_{R,m}: the open sets ∣Ti∣<m|T_i|<m form an increasing cover, so compactness supplies one mm. The domain mapping property of Theorem 2.40.8 factors the map through XR,mX_{R,m}. Its ring map restricts to a homomorphism R→AR\to A. Polynomial density proves that these constructions are inverse, and proves uniqueness of an analytic map with the prescribed coordinate functions.

For general YY, restrict a homomorphism R→Γ(YG,OY)R\to\Gamma(Y_G,\mathcal O_Y) to every affinoid chart. The preceding construction gives analytic maps on the charts. They agree on each affinoid subdomain of an overlap because their coordinate functions agree, and hence glue by Theorem 2.40.8. Conversely, pullback of the coordinates along an analytic map gives global analytic functions. The affinoid calculation proves both compositions are identities. This mapping property determines XRX_R up to a unique analytic isomorphism and therefore proves presentation independence. ◻

Proposition 2.33.24. NAG-001708 (Principal opens and global analytification). For g∈Rg\in R, the natural map XRg→XRX_{R_g}\to X_R identifies its source with the open analytic domain ∣g∣>0|g|>0. Consequently affine analytification glues uniquely to a functor from finite-type KK-schemes to good KK-analytic spaces. Open immersions of schemes give open analytic immersions.

Proof. Multiplicativity gives a unique extension of a seminorm with ∣g∣>0|g|>0 to RgR_g, by ∣a/gm∣=∣a∣/∣g∣m|a/g^m|=|a|/|g|^m. On the bounded piece XR,rX_{R,r}, a bound ∣g−1∣≤t|g^{-1}|\le t is represented by (K{r−1T}/I){t−1V}/(gV−1).\bigl(K\{r^{-1}T\}/I\bigr)\{t^{-1}V\}/(gV-1). This is exactly its Laurent domain ∣g∣≥t−1|g|\ge t^{-1} by the rational universal property. These domains, as rr and tt grow, form neighbourhoods of every point with ∣g∣>0|g|>0. They are also the affinoid pieces for the presentation Rg=R[V]/(gV−1)R_g=R[V]/(gV-1). Thus the pointwise bijection is an analytic isomorphism onto that open domain.

Every open subset of an affine scheme is covered by principal opens. The identifications just constructed agree on intersections, since localizing successively at gg and hh gives RghR_{gh} and the same seminorm and ring maps. An affine open cover of a finite-type scheme can therefore be analytified and glued along its open overlaps by Theorem 2.40.10. Any two covers have a common affine-open refinement, and the same local identifications give the unique comparison isomorphism. Ring maps on affine neighbourhoods similarly glue to the map attached to a scheme morphism; their composites agree on coordinates. Goodness holds locally on these open affine analytifications. ◻

Theorem 2.33.25. NAG-001709 (Products and extension of the ground field). For finite-type KK-schemes, analytification commutes with fibre products. For every complete valued extension L/KL/K, there is a natural analytic isomorphism (Z×KL)an≃Zan⊗^KL.(Z\times_K L)^{\mathrm{an}}\simeq Z^{\mathrm{an}}\widehat\otimes_K L . Both identifications respect all scheme morphisms and iterated complete extensions.

Proof. For affine schemes defined by R→SR\to S and R→TR\to T, the mapping property of Theorem 2.33.23 gives, for every analytic YY, Hom⁡(Y,XS⊗RT)≃Hom⁡(Y,XS)×Hom⁡(Y,XR)Hom⁡(Y,XT).\operatorname{Hom}(Y,X_{S\otimes_R T}) \simeq \operatorname{Hom}(Y,X_S)\mathop{\times}_{\operatorname{Hom}(Y,X_R)} \operatorname{Hom}(Y,X_T). The middle algebraic step is the defining universal property of the ordinary tensor product of commutative algebras. Thus XS⊗RTX_{S\otimes_R T} is the analytic fibre product. To pass to schemes, cover the target by affine opens and both sources over these opens by affine opens. Their pairwise fibre products cover the scheme fibre product by affine opens. Proposition 2.33.24 identifies the overlap maps, so the affine isomorphisms glue, with the universal property checked on the same cover.

If R=K[T]/IR=K[T]/I, its scalar extension is L[T]/IL[T]L[T]/IL[T]. Proposition 2.33.5 gives, on every radius piece, (K{r−1T}/IK{r−1T})⊗^KL≃L{r−1T}/IL{r−1T}.\bigl(K\{r^{-1}T\}/I K\{r^{-1}T\}\bigr) \widehat\otimes_K L \simeq L\{r^{-1}T\}/I L\{r^{-1}T\}. The transition maps agree on polynomials and hence agree everywhere by continuity. Thus these isomorphisms glue over the exhaustion; they further glue over affine Zariski covers. The formulas on polynomial coordinates also prove compatibility with morphisms. Iterated extension agrees by Corollary 2.33.9, applied piecewise. ◻

Theorem 2.33.26. NAG-001710 (Finite maps and closed immersions under analytification). Analytification preserves finite morphisms and closed immersions. More precisely, for a finite ring map R→SR\to S of finite-type KK-algebras and an affinoid domain V=M(A)⊂XRV=\mathcal M(A)\subset X_R, the inverse image of VV in XSX_S is the finite norm-admissible affinoid space M(A⊗RS),\mathcal M(A\otimes_R S), where the algebra has its canonical finite-AA-module topology. If S=R/JS=R/J, its algebra is A/JAA/JA.

Proof. Set D=A⊗RSD=A\otimes_R S. It is finite over AA, so Theorem 2.33.18 gives its canonical affinoid algebra structure. Let M(C)\mathcal M(C) be any affinoid test space. A pair of maps to VV and to XSX_S with the same composite to XRX_R is a bounded homomorphism A→CA\to C and an algebra homomorphism S→CS\to C agreeing on RR. The algebraic tensor property gives a unique AA-linear algebra map D→CD\to C. It is bounded: choose finitely many generators eje_j of DD as an AA-module and controlled lifts of its elements from AmA^m; the finite numbers ∥ej↦C∥\|e_j\mapsto C\| and the bound on A→CA\to C give a uniform operator bound. Conversely any bounded D→CD\to C supplies the pair.

The bijection is natural and glues on affinoid covers of any analytic test space. Hence M(D)\mathcal M(D) is V×XRXSV\times_{X_R}X_S. The domain mapping property identifies this fibre product with the inverse-image analytic domain. The affinoid domains VV cover XRX_R, proving finiteness of the analytified morphism. When S=R/JS=R/J, the algebraic tensor quotient is A/JAA/JA. The ideal JAJA is finitely generated and closed by the affinoid finite-module theorem, so this is an actual closed immersion. Both properties are local on the affine-open cover of a scheme target, which proves the global assertion. ◻

Proposition 2.33.27. NAG-001711 (Relative neighbourhoods for algebraic morphisms). The analytification of every morphism between finite-type KK-schemes is boundaryless. More explicitly, for every affinoid domain WW in the target and every point yy of its inverse image, there is an affinoid neighbourhood DD of yy in that inverse image with y∈Int⁡(D/W)y\in\operatorname{Int}(D/W). In particular every finite-type scheme analytification is boundaryless over KK.

Proof. Fix an affinoid target domain WW and y∈fan,−1(W)y\in f^{\mathrm{an},-1}(W), and write x=fan(y)x=f^{\mathrm{an}}(y). Choose affine scheme neighbourhoods Spec⁡(S)→Spec⁡(R)\operatorname{Spec}(S)\to\operatorname{Spec}(R) of the two underlying scheme points, with S=R[U1,…,Um]/JS=R[U_1,\ldots,U_m]/J. The intersection W∩XRW\cap X_R is an ordinary open neighbourhood of xx in WW. By Lemma 2.41.1, it contains a rational neighbourhood V=M(A)V=\mathcal M(A) of xx in WW such that x∈Int⁡(V/W)x\in\operatorname{Int}(V/W). Choose positive radii sj>∣Uj(y)∣s_j>|U_j(y)|. The affinoid algebra D=A{s−1U}/JA{s−1U}.D=A\{s^{-1}U\}/J A\{s^{-1}U\}. It represents the part of the inverse image of VV with ∣Uj∣≤sj|U_j|\le s_j. Indeed, for every affinoid test algebra CC, a map to that part gives A→CA\to C and the coordinate functions from SS. The pointwise bounds on the latter imply the required power bounds by Theorem 1.20.2; substitution gives the unique bounded map D→CD\to C. The converse follows directly from evaluation. The resulting domain identifications agree on overlaps, so this represents the actual analytic domain, not only its set of points.

The inverse image of the interior of VV in WW, intersected with XSX_S and the strict inequalities ∣Uj∣<sj|U_j|<s_j, is an open neighbourhood of yy in fan,−1(W)f^{\mathrm{an},-1}(W) contained in M(D)\mathcal M(D). Thus M(D)\mathcal M(D) is an affinoid neighbourhood in that inverse image. Its displayed norm-admissible quotient presentation witnesses y∈Int⁡(M(D)/V)y\in\operatorname{Int}(\mathcal M(D)/V). Together with x∈Int⁡(V/W)x\in\operatorname{Int}(V/W), the composition inclusion of Proposition 2.33.21 gives y∈Int⁡(M(D)/W)y\in\operatorname{Int}(\mathcal M(D)/W). This verifies the boundaryless condition for every target affinoid domain. Taking the target to be Spec⁡(K)\operatorname{Spec}(K) proves the last assertion. ◻

Proposition 2.33.28. NAG-001712 (Residue fields, algebraic fibres and surjectivity). Let x∈Zanx\in Z^{\mathrm{an}} lie above the scheme point ξ\xi. Then H(x)\mathcal H(x) is the completion of the valued field κ(ξ)\kappa(\xi) defined by xx. For a morphism Y→ZY\to Z of finite-type KK-schemes, its analytic fibre over xx is naturally the analytification, over H(x)\mathcal H(x), of Y×ZSpec⁡H(x)Y\times_Z\operatorname{Spec}\mathcal H(x). A surjective morphism of finite-type KK-schemes induces a surjective map of analytifications.

Proof. Work first on an affine neighbourhood Spec⁡(R)\operatorname{Spec}(R) of ξ\xi and a radius piece A=K{r−1T}/IA=K\{r^{-1}T\}/I containing an ordinary neighbourhood of xx. Polynomial density shows that the image of RR is dense in AA, also for the seminorm at xx. Its fraction field modulo the point kernel is therefore dense in HA(x)\mathcal H_A(x): approximate each numerator and denominator by polynomials, and keep denominators nonzero by sufficiently close approximation. Corollary 2.33.15 makes the identification independent of the affinoid neighbourhood. Since κ(ξ)=Frac⁡(R/ker⁡x)\kappa(\xi)=\operatorname{Frac}(R/\ker x), this proves the first assertion.

Put F=H(x)F=\mathcal H(x). The scheme morphism Spec⁡(F)→ZF\operatorname{Spec}(F)\to Z_F determined by xx is an FF-rational point. Analytifying its fibre square over FF, Theorems 2.33.25 and 2.33.26 identify the resulting analytic fibre with the fibre of YFan→ZFanY_F^{\mathrm{an}}\to Z_F^{\mathrm{an}} over this rational point. On an affinoid chart A→BA\to B about the original map, that fibre has algebra B⊗^AFB\widehat\otimes_A F, by Corollary 2.33.9 and Theorem 2.33.11. The identification agrees on chart intersections, proving the fibre assertion.

Now assume Y→ZY\to Z is surjective. Its scheme fibre over ξ\xi is nonempty. It stays nonempty after extension from κ(ξ)\kappa(\xi) to FF: a nonzero affine algebra in that fibre remains nonzero under a field extension, as follows by extending a vector-space basis. Thus Y×ZSpec⁡(F)Y\times_Z\operatorname{Spec}(F) has a nonempty affine open of finite type over FF. Choose a maximal ideal in its nonzero coordinate ring. Its residue field EE is finite over FF by the Hilbert Nullstellensatz, [StacksAlg] (Tag 00FV, Theorem 10.34.1(1)). Lemma 1.12.1 extends the absolute value to EE and makes it complete. This valued closed point gives a point of the analytic fibre and hence a point above xx. ◻

Lemma 2.33.29. NAG-001713 (Separated algebraic maps and the underlying topology). If Y→ZY\to Z is a separated morphism of finite-type KK-schemes, then Yan→ZanY^{\mathrm{an}}\to Z^{\mathrm{an}} is separated and is a Hausdorff map of topological spaces. In particular the inverse image of a Hausdorff subspace of ZanZ^{\mathrm{an}} is Hausdorff.

Proof. Its algebraic diagonal is a closed immersion. Theorems 2.33.25 and 2.33.26 identify its analytification with the analytic diagonal and preserve its closed-immersion property. A closed immersion restricts on any affinoid target domain to the quotient by its finite defining ideal: this follows first on defining affinoid charts by the quotient base-change presentation, and then on a finite affinoid refinement by Kiehl gluing, Theorem 2.39.6. Thus pulling the diagonal back to any affinoid product gives an affinoid closed subspace.

Consider two points y1,y2y_1,y_2 above the same point zz. Goodness permits affinoid neighbourhoods VV of zz and UiU_i of yiy_i with maps Ui→VU_i\to V. The analytic product P=U1×VU2P=U_1\times_V U_2 is compact Hausdorff. Its closed diagonal pullback DD is compact. By Proposition 2.33.14, the map from PP onto the ordinary topological fibre product is closed. The image of DD is exactly the ordinary diagonal in U1×∣V∣U2U_1\times_{|V|}U_2, and is therefore closed there. Restricting to the ordinary interiors of the chosen neighbourhoods proves that the relative ordinary diagonal is closed locally, hence closed. Distinct points in a common fibre can consequently be separated by neighbourhoods in YanY^{\mathrm{an}}. Points with distinct images in a Hausdorff subspace are separated by inverse images of disjoint neighbourhoods of their images. These two cases give the last assertion. ◻

Proposition 2.33.30. NAG-001714 (Compactness of analytified projective maps). The analytification of a projective morphism of finite-type KK-schemes is compact: the inverse image of every compact Hausdorff subset of the target is compact Hausdorff. In particular (PKn)an(\mathbf P^n_K)^{\mathrm{an}} is compact.

Proof. First consider PRn→Spec⁡(R)\mathbf P^n_R\to\operatorname{Spec}(R) and an affinoid domain V=M(A)⊂XRV=\mathcal M(A)\subset X_R. For each homogeneous coordinate XiX_i, the domain Vi={ y above V: ∣Xj/Xi∣(y)≤1 for j≠i }≃M(A{T0,…,T^i,…,Tn})V_i=\{\,y\text{ above }V:\ |X_j/X_i|(y)\le1 \text{ for }j\ne i\,\} \simeq\mathcal M\bigl(A\{T_0,\ldots,\widehat T_i,\ldots,T_n\}\bigr) is compact. These n+1n+1 domains cover the inverse image of VV. Indeed, at a scheme point choose homogeneous coordinates in its residue field, not all zero, and evaluate them using its analytic absolute value. Choosing a coordinate of maximal absolute value puts the analytic point in the corresponding domain. The identifications of the domains follow either from their displayed coordinate ratios or from Theorem 2.33.25. The union is Hausdorff by Lemma 2.33.29, because projective space has closed diagonal. It is a finite union of compact spaces and hence compact. A closed subscheme of projective space gives a closed analytic subspace, so its inverse image over VV is compact as well.

For a compact subset C⊂XRC\subset X_R, finitely many affinoid neighbourhoods cover CC. Their inverse images form a finite compact union. Since XRX_R is Hausdorff, CC is closed, and its inverse image is a closed subset of that union. This proves compactness over an affine scheme base. For a general scheme base, cover a given compact Hausdorff subset CC by finitely many analytified affine opens. A compact Hausdorff space admits a finite closed refinement of any finite open cover: choose, at each point, an open neighbourhood whose closure lies in one chosen cover member, using regularity of compact Hausdorff spaces, and take a finite subcover. Apply the affine result to these finitely many closed subsets of CC. Their inverse images are compact and cover the required inverse image. The latter is Hausdorff by the preceding lemma. Hence it is compact Hausdorff. ◻

Theorem 2.33.31. NAG-001715 (Proper algebraic maps remain proper analytically). If f:Y→Zf:Y\to Z is a proper morphism of finite-type KK-schemes, then fan:Yan→Zanf^{\mathrm{an}}:Y^{\mathrm{an}}\to Z^{\mathrm{an}} is proper. This includes nonreduced schemes and trivially valued complete ground fields.

Proof. The target is Noetherian by [StacksAlg] (Tag 01T6). Apply Chow’s lemma, [StacksAlg] (Tag 0200, Lemma 30.18.1), to ff. It gives a proper surjective π:Y′→Y\pi:Y'\to Y and an immersion j:Y′→PZnj:Y'\to\mathbf P^n_Z. Since fπf\pi is proper by [StacksAlg] (Tag 01W3, Lemma 29.42.4), and PZn→Z\mathbf P^n_Z\to Z is separated, [StacksAlg] (Tag 01W6, Lemma 29.42.7) shows that jj is proper. Its image is closed. An immersion with closed image is a closed immersion: factor it as a closed immersion into an open subspace of the target, and add the open complement of its closed image; the two local defining ideals glue on this open cover. Thus fπf\pi is projective.

By Proposition 2.33.28, πan\pi^{\mathrm{an}} is surjective. Let CC be any compact Hausdorff subset of ZanZ^{\mathrm{an}}. Proposition 2.33.30 makes (fπ)an,−1(C)(f\pi)^{\mathrm{an},-1}(C) compact. Its continuous image under πan\pi^{\mathrm{an}} is exactly fan,−1(C)f^{\mathrm{an},-1}(C), so the latter is quasi-compact. The map ff is separated; Lemma 2.33.29 therefore makes this inverse image Hausdorff. Hence fanf^{\mathrm{an}} is compact. Proposition 2.33.27 proves that it is boundaryless. These are the two defining conditions of an analytic proper morphism. ◻

Flatness of analytification

Lemma 2.33.32. NAG-001717 (A rational-point completion calculation). Let P=K[T1,…,Tn]P=K[T_1,\ldots,T_n], C=K⟨T1,…,Tn⟩C=K\langle T_1,\ldots,T_n\rangle, and a∈Kna\in K^n with ∣ai∣≤1|a_i|\le 1. For p=(T1−a1,…,Tn−an)⊂P\mathfrak p=(T_1-a_1,\ldots,T_n-a_n)\subset P and n=pC\mathfrak n=\mathfrak p C, the local map Pp→CnP_{\mathfrak p}\to C_{\mathfrak n} is faithfully flat. Both maximal-ideal completions are canonically K[[U1,…,Un]]K[[U_1,\ldots,U_n]], where Ui=Ti−aiU_i=T_i-a_i.

Proof. Translation by aa is a polynomial automorphism and a bounded Tate-algebra automorphism with bounded inverse: substitution by Ti+aiT_i+a_i and by Ti−aiT_i-a_i is contractive. We may therefore assume a=0a=0. For every integer q≥1q\ge 1, truncation in total degree gives P/(T)q≃C/(T)q≃⨁∣ν∣<qKTν.P/(T)^q \simeq C/(T)^q \simeq \bigoplus_{|\nu|<q} K T^\nu . The kernel in CC is the actual ideal (T)q(T)^q: partition all terms of total degree at least qq among the finitely many monomials of degree qq which divide them. Dividing each group by its chosen monomial leaves a convergent series. Every element with nonzero constant term is a unit in this truncated ring, by a finite geometric sum. Localization at (T)(T) therefore leaves the displayed quotients unchanged. Taking their inverse limits proves the completion identity.

Both local rings are Noetherian: for PP use [StacksAlg] (Tag 00FO, Lemma 10.31.3), and for CC use Theorem 1.11.5; localization preserves finite generation of ideals by clearing denominators. By [StacksAlg] (Tag 00MC, Lemma 10.97.3), each local ring maps faithfully flatly to the common completion D=K[[T]]D=K[[T]]. We spell out the descent step. For a finitely generated ideal I⊂P(T)I\subset P_{(T)}, let NN be the kernel of I⊗P(T)C(T)→C(T)I\otimes_{P_{(T)}}C_{(T)}\to C_{(T)}. Tensoring with the flat C(T)C_{(T)}-algebra DD identifies N⊗DN\otimes D with the kernel of I⊗D→DI\otimes D\to D, which is zero because DD is flat over P(T)P_{(T)}. Faithfulness gives N=0N=0. The ideal criterion, [StacksAlg] (Tag 00HD), proves flatness. The map is local with nonzero residue field KK, so [StacksAlg] (Tag 00HP) makes it faithfully flat. ◻

Theorem 2.33.33. NAG-001718 (Polynomial-to-Tate flatness in arbitrary radii). For every complete non-Archimedean field KK and every positive polyradius rr, the ring map K[T1,…,Tn]⟶K{r−1T}K[T_1,\ldots,T_n]\longrightarrow K\{r^{-1}T\} is flat. Consequently, if R=K[T]/IR=K[T]/I and Ar=K{r−1T}/IK{r−1T}A_r=K\{r^{-1}T\}/I K\{r^{-1}T\}, then R→ArR\to A_r is flat.

Proof. First take r=(1,…,1)r=(1,\ldots,1) and write P=K[T]P=K[T], C=K⟨T⟩C=K\langle T\rangle. For a maximal ideal m⊂C\mathfrak m\subset C, Theorem 1.12.3 gives the finite complete residue field E=C/mE=C/\mathfrak m. Let ai∈Ea_i\in E be the images of TiT_i; contractivity gives ∣ai∣≤1|a_i|\le 1. The scalar extension CE=E⟨T⟩C_E=E\langle T\rangle is C⊗KEC\otimes_K E as an ordinary finite CC-module, by Lemma 2.33.4 and a finite KK-basis of EE. It is free over CC. Evaluation at aa has maximal kernel n=(Ti−ai)\mathfrak n=(T_i-a_i) and restricts to the given quotient C→EC\to E, so n∩C=m\mathfrak n\cap C=\mathfrak m.

Localization of a flat module is flat, since fractions preserve exact sequences: a fraction maps to zero precisely when some denominator annihilates its numerator. Hence Cm→(CE)nC_{\mathfrak m}\to(C_E)_{\mathfrak n} is flat, and it is faithfully flat because it is local. Lemma 2.33.32 says that (CE)n(C_E)_{\mathfrak n} is flat over E[T](T−a)E[T]_{(T-a)}, and therefore over PP: the polynomial scalar extension is free and the intervening localization is flat. Applying the kernel descent argument of that lemma to Cm→(CE)nC_{\mathfrak m}\to(C_E)_{\mathfrak n} proves that CmC_{\mathfrak m} is flat over PP. For a finitely generated ideal J⊂PJ\subset P, the kernel of J⊗PC→CJ\otimes_P C\to C vanishes after localization at every maximal ideal of CC. It is therefore zero by [StacksAlg] (Tag 00HN, Lemma 10.23.1(2)). The finite-ideal criterion proves P→CP\to C flat.

For arbitrary rr, the multiplicative Gauss norm of Proposition 1.2.2 extends to the fraction field of K{r−1S}K\{r^{-1}S\}. Let LL be its completion. It contains elements sis_i with ∣si∣=ri|s_i|=r_i. By scalar extension and rescaling Ti=siViT_i=s_iV_i, K{r−1T}⊗^KL≃L{r−1T}≃L⟨V⟩.K\{r^{-1}T\}\widehat\otimes_K L \simeq L\{r^{-1}T\}\simeq L\langle V\rangle . The first map from K{r−1T}K\{r^{-1}T\} is faithfully flat by Theorem 2.33.6. The strict case proves that the right side is flat over L[T]L[T], hence over K[T]K[T] by the free scalar extension. The same kernel descent argument proves flatness before extension. Finally Ar=K{r−1T}⊗K[T]RA_r=K\{r^{-1}T\}\otimes_{K[T]}R as ordinary algebras. Tensoring a short exact sequence of RR-modules through this equality proves the last assertion. ◻

Theorem 2.33.34. NAG-001719 (Exact analytification of coherent modules). For a finite-type KK-scheme ZZ, the natural map of locally ringed spaces π:Zan→Z\pi:Z^{\mathrm{an}}\to Z is flat on local rings. Its stalk map at xx, over ξ=π(x)\xi=\pi(x), is faithfully flat. The functor F↦Fan=π∗F\mathcal F\mapsto\mathcal F^{\mathrm{an}}=\pi^*\mathcal F sends coherent modules to coherent modules and is exact. It commutes with tensor products and, for coherent modules F,G\mathcal F,\mathcal G, with the sheaf of homomorphisms.

Proof. On an affine scheme chart Spec⁡(R)\operatorname{Spec}(R), the map π\pi sends a seminorm to its prime kernel. It is continuous because the inverse image of D(g)D(g) is ∣g∣>0|g|>0. A fraction in RgR_g defines an analytic function there, compatibly with further restriction, giving the structural sheaf map. At xx, a fraction whose value is nonzero is invertible in an analytic neighbourhood by a Laurent localization. Thus the map on stalks is local.

Choose a radius neighbourhood XR,rX_{R,r} of xx. Its algebra ArA_r is flat over RR by Theorem 2.33.33. On each smaller rational neighbourhood VV of xx, the algebra AVA_V is flat over ArA_r by Proposition 2.38.2. The analytic stalk is the filtered colimit of these AVA_V, by Proposition 2.33.16. Filtered colimits of flat modules are flat: a tensor relation uses finitely many elements, and if it becomes zero in the colimit it becomes zero at some stage, where injectivity follows from flatness. Equivalently one applies the finite-ideal criterion to this finite-stage argument. Thus OZan,x\mathcal O_{Z^{\mathrm{an}},x} is flat over RR and over Rp=OZ,ξR_{\mathfrak p}=\mathcal O_{Z,\xi}. Its maximal ideal lies above p\mathfrak p, and its residue field is nonzero. The faithful-flat criterion for this local map proves the stalk assertion.

By [StacksAlg] (Tag 01XZ, Lemma 30.9.1), F\mathcal F is represented on Spec⁡(R)\operatorname{Spec}(R) by a finite module MM. Thus its pullback to XR,rX_{R,r} is the coherent module associated with Ar⊗RMA_r\otimes_R M. These descriptions agree on overlaps by the tensor mapping property, so they give the stated coherent analytification. Flatness of R→ArR\to A_r preserves exact sequences of finite modules on every such chart, proving exactness on the analytic GG-site as well as on the ordinary topology. Tensor compatibility is algebraic associativity. For homomorphisms, take a finite presentation Ra→Rb→M→0R^a\to R^b\to M\to 0 and apply Hom⁡R(−,N)\operatorname{Hom}_R(-,N). Its first three terms identify Hom⁡R(M,N)\operatorname{Hom}_R(M,N) with the kernel of a matrix Nb→NaN^b\to N^a. Flat scalar extension to ArA_r preserves this kernel and gives exactly Hom⁡Ar(Ar⊗M,Ar⊗N)\operatorname{Hom}_{A_r}(A_r\otimes M,A_r\otimes N). These natural chartwise identifications glue to the asserted sheaf isomorphism. ◻

Compact restriction and finite-dimensional cohomology

The following functional-analytic argument reconstructs the absolute-field case of the completely continuous method in [Bosch14] (§6.4). The countable-type hypothesis on the Banach spaces is what supplies the bounded splittings used below.

Lemma 2.33.35. NAG-001721 (Bounded splittings in countable type). Let EE be a Banach KK-space of countable type, meaning that a countable set has dense linear span. Every closed subspace N⊂EN\subset E is of countable type and admits a bounded linear projection E→NE\to N. More generally, every norm-strict surjection of Banach spaces onto a space of countable type admits a bounded linear section. Finite modules with their canonical norms over generalized KK-affinoid algebras are of countable type.

Proof. Let p:B→Qp:B\to Q be the asserted norm-strict surjection. Replace the norm on QQ by the equivalent quotient norm. Choose a family (ei)(e_i) as in Lemma 3.21.2, with lower constant 0<c<10<c<1. For each ii, choose a lift e^i\widehat e_i with ∥e^i∥≤2∥ei∥\|\widehat e_i\|\le 2\|e_i\|. If q=∑iaieiq=\sum_i a_ie_i, then ∣ai∣∥ei∥→0|a_i|\|e_i\|\to 0. Hence ∑iaie^i\sum_i a_i\widehat e_i converges in BB, and defines a linear section s:Q→Bs:Q\to B satisfying ∥s(q)∥≤2c−1∥q∥\|s(q)\|\le 2c^{-1}\|q\|. This proves the splitting assertion.

Now take Q=E/NQ=E/N. It is Banach by Lemma 1.11.1. The images of a countable dense spanning set of EE span densely in QQ, so the preceding argument gives a bounded section ss. The map P=1−sp:E→NP=1-sp:E\to N is a bounded projection. Applying PP to that same countable set gives a dense spanning set in NN. Finally, a generalized affinoid algebra is a quotient of a finite-variable Tate algebra, whose polynomial monomials form a countable dense spanning set. A finite module is a bounded quotient of a finite power of this algebra by Theorem 1.18.3. Images of the corresponding spanning sets prove the last assertion. ◻

Lemma 2.33.36. NAG-001722 (The completely continuous perturbation lemma). Assume that KK is nontrivially valued. A bounded linear map is called completely continuous if it is an operator-norm limit of bounded maps with finite-dimensional image. Let f:E→Ff:E\to F be a surjective bounded map of Banach KK-spaces, and let g:E→Fg:E\to F be completely continuous. Then (f+g)(E)(f+g)(E) is closed and has finite codimension in FF. Completely continuous maps remain such after bounded pre- or postcomposition. If their image lies in a closed subspace of a countable-type target, they are completely continuous as maps to that subspace.

Proof. The open mapping theorem, Lemma 1.14.1, gives a constant C>0C>0 such that every y∈Fy\in F has a lift xx with ∥x∥≤C∥y∥\|x\|\le C\|y\|. Choose a finite-rank map hh with C∥g−h∥<1C\|g-h\|<1, and put f′=f+(g−h)f'=f+(g-h). This map is surjective. Indeed, lift yy by ff; the error under f′f' has norm at most C∥g−h∥∥y∥C\|g-h\|\|y\|. Lift that error and repeat. The errors decrease geometrically, so the successive lifts converge and give a preimage under f′f'. The same estimates give a bounded lifting constant for f′f', hence it is open.

Write N=ker⁡hN=\ker h. This is closed, and E/NE/N is finite dimensional. The image of ker⁡f′\ker f' in E/NE/N is a finite-dimensional subspace, hence closed by Lemma 1.12.1. Consequently N+ker⁡f′N+\ker f' is closed in EE. Since f′f' is an open quotient map, f′(N)f'(N) is closed in FF. The vector space F/f′(N)F/f'(N) is finite dimensional, because it is a quotient of E/NE/N. The image of f′+hf'+h contains f′(N)f'(N) and corresponds to a subspace of this finite-dimensional Hausdorff quotient. It is therefore closed and has finite codimension. As f′+h=f+gf'+h=f+g, this proves the first assertion.

Bounded composition preserves convergence in operator norm and finite rank, proving the second assertion. For the last one, use Lemma 2.33.35 to project the target boundedly onto the closed subspace and compose each finite-rank approximant with that projection. ◻

Proposition 2.33.37. NAG-001723 (Completely continuous restriction). Let A=K{r−1T}/IA=K\{r^{-1}T\}/I and let A→BA\to B be a bounded homomorphism of generalized affinoid algebras. Suppose that the image bib_i of every TiT_i has spectral radius strictly smaller than rir_i. For a finite AA-module MM, the natural restriction map M⟶B⊗AMM\longrightarrow B\otimes_A M between the canonical Banach modules is completely continuous. This assertion holds also over a trivially valued field.

Proof. Choose numbers ρB(bi)<γi<ri\rho_B(b_i)<\gamma_i<r_i. By the spectral-radius formula, Corollary 1.20.3, there are constants CiC_i with ∥bid∥≤Ciγid\|b_i^d\|\le C_i\gamma_i^d for every d≥0d\ge 0. Choose finite generators m1,…,msm_1,\ldots,m_s of MM, and write N=B⊗AMN=B\otimes_A M and nj=1⊗mjn_j=1\otimes m_j. The composite K{r−1T} s⟶M⟶NK\{r^{-1}T\}^{\,s}\longrightarrow M\longrightarrow N sends ∑νaνjTν\sum_\nu a_{\nu j}T^\nu in component jj to ∑νaνjbνnj\sum_\nu a_{\nu j}b^\nu n_j. Let 0<θ=max⁡i(γi/ri)<10<\theta=\max_i(\gamma_i/r_i)<1, and put C=(∏iCi)max⁡j∥nj∥C=(\prod_iC_i)\max_j\|n_j\|, enlarged if necessary for the module-action bound. Terms with ∣ν∣≥d|\nu|\ge d have norm at most Cθd∣aνj∣rνC\theta^d|a_{\nu j}|r^\nu. Truncation to ∣ν∣<d|\nu|<d is a bounded finite-rank map, and its operator-norm error is at most CθdC\theta^d. Hence this composite is completely continuous. The first arrow is norm-strict by the quotient presentation and Theorem 1.18.3; its target is of countable type. Lemma 2.33.35 gives a bounded linear section. Composing with it proves the assertion for M→NM\to N. If there are no Tate variables, the source module is finite dimensional over KK and the assertion is immediate. ◻

Theorem 2.33.38. NAG-001724 (Finiteness of coherent cohomology on projective space). For every complete non-Archimedean field KK, every integer n≥0n\ge 0, and every coherent module F\mathcal F on PGP_G, where P=(PKn)anP=(\mathbf P^n_K)^{\mathrm{an}}, the groups Hq(PG,F)H^q(P_G,\mathcal F) are finite-dimensional KK-vector spaces and vanish for q>nq>n. On the standard unit-polydisc cover, the coboundary images are closed in the canonical Banach spaces.

Proof. First assume that KK is nontrivially valued. Fix a real R>1R>1 and set Ui={∣Xj∣≤R∣Xi∣ for every j},Vi={∣Xj∣≤∣Xi∣ for every j},0≤i≤n.\begin{aligned} U_i&=\{|X_j|\le R|X_i|\text{ for every }j\},\\ V_i&=\{|X_j|\le |X_i|\text{ for every }j\}, \qquad 0\le i\le n. \end{aligned} These are affinoid domains, and both families cover PP: at any point, choose a homogeneous coordinate of maximal nonzero value. Their finite intersections are affinoid. More explicitly, on Ui0U_{i_0} write zj=Xj/Xi0z_j=X_j/X_{i_0} and zi0=1z_{i_0}=1. For each additional i∈Ii\in I, the conditions defining UiU_i are ∣zj∣≤R∣zi∣|z_j|\le R|z_i| for every jj, including j=i0j=i_0. They form a rational domain because the numerator 11 is present. The rational presentation adjoins all zj/ziz_j/z_i with radius RR. Consequently AI=O(UI)A_I=\mathcal O(U_I) has a quotient presentation by a Tate algebra whose variables are the finitely many ratios Xj/XiX_j/X_i for i∈Ii\in I, all with radius RR. On VIV_I all these ratios have value at most 11. The spectral-radius formula therefore makes their spectral radii at most 1<R1<R.

Put CUq=∏∣I∣=q+1F(UI)C_U^q=\prod_{|I|=q+1}\mathcal F(U_I) and define CVqC_V^q similarly. These are Banach spaces of countable type. Kiehl’s equivalence on each affinoid identifies restrictions of F\mathcal F with finite-module base change. Proposition 2.33.37 proves that every component of the restriction rq:CUq→CVqr^q:C_U^q\to C_V^q is completely continuous, hence so is their finite product. Let ZUq,ZVqZ_U^q,Z_V^q be the kernels of the corresponding bounded differentials. They are closed, hence are Banach spaces of countable type by Lemma 2.33.35. Projecting CVqC_V^q boundedly onto ZVqZ_V^q shows that the induced map r:ZUq→ZVqr:Z_U^q\to Z_V^q is completely continuous.

By Proposition 2.42.1, both complexes compute Hq(PG,F)H^q(P_G,\mathcal F), and restriction induces the identity on this derived-section group. Therefore Fq:CVq−1⊕ZUq⟶ZVq,(a,z)⟼da+r(z)F_q:C_V^{q-1}\oplus Z_U^q\longrightarrow Z_V^q,\qquad (a,z)\longmapsto da+r(z) is surjective, with CV−1=0C_V^{-1}=0. The map Gq(a,z)=−r(z)G_q(a,z)=-r(z) is completely continuous. Lemma 2.33.36 shows that Fq+GqF_q+G_q, whose image is dCVq−1dC_V^{q-1}, has closed image and finite-dimensional cokernel. That cokernel is exactly Hq(PG,F)H^q(P_G,\mathcal F). This proves finite dimensionality and closedness, without assuming closed coboundary images in advance. Vanishing above nn follows from the length of the ordered complex.

Now suppose that KK is trivially valued. On the unit cover (Vi)(V_i), each algebra O(VI)\mathcal O(V_I) is the corresponding polynomial–Laurent algebra with the norm which is 11 on every nonzero element, by Lemma 2.42.2. Each finite module, with a quotient norm from a finite free presentation, has the same discrete 0/10/1 norm. Thus every vector subspace of each CVqC_V^q is closed, and every exact sequence among its kernels, images and cohomology spaces is norm-strict. Choose a nontrivially valued complete extension L/KL/K: for example, complete the fraction field of K{t−1S}K\{t^{-1}S\} in its Gauss norm for any 0<t<10<t<1. The scalar-extension construction and finite-module base change identify CV∙⊗^KLC_V^\bullet\widehat\otimes_K L with the unit-cover complex of the coherent pullback FL\mathcal F_L on (PLn)an(\mathbf P^n_L)^{\mathrm{an}}. By Theorem 2.33.3, its cohomology is Hq(PG,F)⊗^KL≃Hq((PLn)Gan,FL).H^q(P_G,\mathcal F)\widehat\otimes_K L \simeq H^q((\mathbf P^n_L)^{\mathrm{an}}_G,\mathcal F_L). The right side is finite dimensional by the nontrivial case. If the left-hand source had arbitrarily many linearly independent vectors over KK, their finite-dimensional spans would, by Theorem 2.33.2, inject after tensoring into the displayed left side and have the same dimension over LL. This contradicts finite dimensionality. Hence Hq(PG,F)H^q(P_G,\mathcal F) is finite dimensional over KK, as required. ◻

2.34 From projective comparison to proper GAGA

The projective calculation also provides the relative comparison needed for a Chow modification. We give the support argument explicitly, so that passing from projective to proper schemes requires no analytic proper mapping theorem as an additional input.

Proposition 2.34.1. NAG-001726 (The flat analytification morphism of ringed sites). For a finite-type KK-scheme XX, inverse image of Zariski opens defines a flat morphism of ringed topoi αX:XGan→XZar\alpha_X:X^{\mathrm{an}}_G\to X_{\mathrm{Zar}}. If W=Spec⁡(R)⊂XW=\operatorname{Spec}(R)\subset X and V=M(A)V=\mathcal M(A) is any affinoid domain in WanW^{\mathrm{an}}, then R→AR\to A is flat. On coherent modules αX∗\alpha_X^* is the analytification already constructed, and αX∗\alpha_X^* is exact on all modules.

Proof. The rule W↦WanW\mapsto W^{\mathrm{an}} preserves finite intersections, the final object and coverings, by Proposition 2.33.24. It therefore defines the stated morphism of sites, with structural ring map given by evaluation of regular functions. Choose a finite generating list of RR over KK. Its values on the compact space VV are bounded. Consequently VV lies in a sufficiently large radius affinoid XR,rX_{R,r}. It is an affinoid domain there by the domain restriction property. The factorization R⟶Ar⟶AR\longrightarrow A_r\longrightarrow A is flat by Theorem 2.33.33 and Proposition 2.38.2.

Here is a sheaf-level verification of flatness, including modules which are not coherent. A finite relation with coefficients in αX−1OX\alpha_X^{-1}\mathcal O_X and values in OXGan\mathcal O_{X^{\mathrm{an}}_G} can, locally for the analytic site, be represented by coefficients r1,…,rtr_1,\ldots,r_t in one affine ring RR and values a1,…,ata_1,\ldots,a_t in one affinoid ring AA, with ∑riai=0\sum r_i a_i=0. To obtain this representation, first use the local definition of an inverse-image sheaf for the finitely many coefficients, and then refine to affinoids. Since RR is Noetherian, the relation module N=ker⁡(Rt→R)N=\ker(R^t\to R) is finite. Flatness of R→AR\to A gives ker⁡(At→A)=A⊗RN\ker(A^t\to A)=A\otimes_R N. Thus the vector (ai)(a_i) is a finite AA-linear combination of relations already over RR. This is the finite-relation criterion for flatness, equivalently the finite-ideal criterion [StacksAlg] (Tag 00HD), applied locally to the ring sheaves. It proves that OXGan\mathcal O_{X^{\mathrm{an}}_G} is flat over αX−1OX\alpha_X^{-1}\mathcal O_X. Inverse image of sheaves is exact; tensoring with this flat ring sheaf is exact as well. On finite presentations the result is A⊗RMA\otimes_R M, agreeing with Theorem 2.33.34. ◻

Lemma 2.34.2. NAG-001727 (Every scheme point has an analytic lift). For every finite-type KK-scheme XX, the map ∣Xan∣→∣X∣|X^{\mathrm{an}}|\to |X| is surjective. Consequently coherent analytification detects zero sheaves and isomorphisms.

Proof. Let ξ∈X\xi\in X and put E=κ(ξ)E=\kappa(\xi), a finitely generated field extension of KK. Choose a transcendence basis from a finite generating list. Then EE is finite over F=K(t1,…,td)F=K(t_1,\ldots,t_d). The Gauss absolute value on K[t1,…,td]K[t_1,\ldots,t_d] is multiplicative by Proposition 1.2.2; it extends by quotients to FF. Let HH be its completion. The finite nonzero HH-algebra E⊗FHE\otimes_F H has a maximal ideal, and its quotient LL is a finite field extension of HH. The map E→LE\to L is injective because EE is a field and 11 has nonzero image. Lemma 1.12.1 gives LL its extending absolute value. Its restriction to EE defines a multiplicative seminorm on any affine coordinate ring at ξ\xi, with kernel exactly the prime of ξ\xi. The radius construction of Theorem 2.33.23 makes this an analytic point above ξ\xi. At such a point the local ring map is faithfully flat by Theorem 2.33.34. A coherent sheaf whose pullback is zero therefore has every stalk zero. Apply the same assertion to the kernel and cokernel, using exactness, to detect an isomorphism. ◻

Theorem 2.34.3. NAG-001728 (Relative projective cohomology comparison). Let f:Y→Xf:Y\to X be a projective morphism of finite-type KK-schemes, and let E\mathcal E be coherent on YY. There is a natural quasi-isomorphism αX∗Rf∗E ≃ Rf∗an(Ean).\alpha_X^* Rf_*\mathcal E \ \simeq\ Rf^{\mathrm{an}}_*(\mathcal E^{\mathrm{an}}). In particular (Rqf∗E)an≃Rqf∗anEan(R^qf_*\mathcal E)^{\mathrm{an}}\simeq R^qf^{\mathrm{an}}_*\mathcal E^{\mathrm{an}} for every qq. If X=Spec⁡(R)X=\operatorname{Spec}(R) and V=M(A)⊂XanV=\mathcal M(A)\subset X^{\mathrm{an}} is an affinoid domain, the corresponding comparison is A⊗RHq(Y,E) ≃ Hq(fan,−1(V)G,Ean).A\otimes_R H^q(Y,\mathcal E) \ \simeq\ H^q(f^{\mathrm{an},-1}(V)_G,\mathcal E^{\mathrm{an}}).

Proof. Projective space. First take Y=PRnY=\mathbf P^n_R. For O(m)\mathcal O(m) the standard unit-affinoid covering above VV has coordinate algebras consisting of convergent Laurent series with coefficients in AA. The proof of Lemma 2.42.2 works coefficient by coefficient over the complete Banach algebra AA: its reduction, isometric section, and convergent ideal multipliers have norm at most one. The exponent sets and the simplex insertion operators in Theorem 2.42.3 are unchanged. They yield a bounded contraction onto the finite free AA-modules indexed by the same surviving monomials. The algebraic computation [StacksAlg] (Tag 01XT) over RR has the same representatives. Hence the displayed comparison holds for every twist, and respects all restriction maps in VV.

For an arbitrary coherent E\mathcal E, choose, by [StacksAlg] (Tag 01YS(1)–(2)), an exact sequence 0→R→L→E→00\to\mathcal R\to\mathcal L\to\mathcal E\to 0, where L\mathcal L is a finite sum of twists and R\mathcal R is coherent. Both the algebraic standard affine cover and the analytic unit cover have length nn, so all cohomology above nn vanishes, using [StacksAlg] (Tag 01XB) and Proposition 2.42.1. Tensor the algebraic long exact sequence with AA; it stays exact by Proposition 2.34.1. Analytification and restriction to fan,−1(V)f^{\mathrm{an},-1}(V) also preserve the short exact sequence. Descend on qq, simultaneously for all coherent sheaves. The exact diagram first lifts an analytic qq-class of E\mathcal E by lifting its connecting (q+1)(q+1)-class of R\mathcal R and correcting by a class of L\mathcal L; this proves surjectivity for all E\mathcal E. Apply this surjectivity to R\mathcal R to lift the correction of a class of L\mathcal L mapping to zero analytically. Injectivity for L\mathcal L and the (q+1)(q+1)-term then proves injectivity for E\mathcal E. This is the exact diagram chase of Theorem 2.42.4, now with the flat coefficient extension R→AR\to A.

Closed projective subspaces and direct images. Factor a general projective ff as a closed immersion into PXn\mathbf P^n_X followed by projection. Closed direct image is exact and coherent on both sides, by finite quotient modules and Lemma 2.42.7; it commutes with analytification on each chart. The same affinoid and affine Čech complexes compute cohomology before and after this closed direct image. Thus the preceding argument applies to i∗Ei_*\mathcal E. The algebraic modules Hq(Y,E)H^q(Y,\mathcal E) are finite over RR by [StacksAlg] (Tag 02O5); their localizations are the sections of Rqf∗ER^qf_*\mathcal E, as also follows by localizing the finite affine Čech complex. On the analytic side, [StacksAlg] (Tag 072W) identifies the higher direct image as the sheafification of the cohomology on inverse images of objects. The displayed isomorphisms, compatible with affinoid restriction, therefore give the asserted isomorphisms of cohomology sheaves.

Finally the commutative analytification square of ringed topoi has the canonical derived comparison map of [StacksAlg] (Tag 07A7). Proposition 2.34.1 makes its derived pullbacks ordinary pullbacks. The cohomology maps just calculated are those of this map: on Čech resolutions they send a regular section to its analytic section, which is the adjunction map used in its construction. They are isomorphisms in every degree on an affinoid basis, so the comparison is a quasi-isomorphism. This also proves its compatibility with the adjunction F→Rf∗f∗F\mathcal F\to Rf_*f^*\mathcal F and with restriction. ◻

Lemma 2.34.4. NAG-001729 (Finite covers and closed supports on proper analytifications). Let XX be a proper finite-type KK-scheme. Its analytification has a finite affinoid covering all of whose finite intersections are affinoid. If i:T↪Xi:T\hookrightarrow X is a closed subscheme, closed direct image preserves coherent modules and their cohomology on both the algebraic and the analytic sites; the resulting cohomology identifications commute with analytification.

Proof. The space XanX^{\mathrm{an}} is good by the radius construction, compact by Theorem 2.33.31, and has a closed analytic diagonal by Lemma 2.33.29. Choose affinoid neighbourhoods and a finite subcover of their interiors. If V1,V2V_1,V_2 are two of these domains, their fibre product over XanX^{\mathrm{an}} is the inverse image of its closed diagonal in the affinoid space V1×KV2V_1\times_K V_2. It is therefore affinoid, by the quotient presentation of a closed analytic subspace. The universal properties of domains identify this fibre product with V1∩V2V_1\cap V_2. Iteration proves the assertion for every finite intersection.

The inverse images of this covering on TanT^{\mathrm{an}} are closed affinoid subspaces. Thus Lemma 2.42.7 proves the analytic assertions by the identical finite Čech complexes. Algebraically, a finite affine cover of the separated scheme XX also has affine finite intersections: apply its closed diagonal to products of affine opens. A finite module over a quotient R/JR/J is the same as a finite RR-module killed by JJ, so closed direct image is exact and coherent. The same affine Čech complex computes cohomology before and after that direct image, by [StacksAlg] (Tags 01XB and 03F7). On every affine–affinoid chart the comparison sends a regular section to its analytic section. The identical-complex identifications therefore commute with comparison. ◻

Lemma 2.34.5. NAG-001730 (An algebraic thickening captures analytic support). Let XX be proper over KK, let Z↪XZ\hookrightarrow X be a closed subscheme with ideal I\mathcal I, and let F\mathcal F be a coherent module on XGanX^{\mathrm{an}}_G. If F\mathcal F vanishes on Xan∖ZanX^{\mathrm{an}}\setminus Z^{\mathrm{an}}, then (Ian)NF=0(\mathcal I^{\mathrm{an}})^N\mathcal F=0 for some integer N≥1N\ge 1. Hence F\mathcal F is the closed direct image of a coherent module on the analytification of ZN=(X,OX/IN)Z_N=(X,\mathcal O_X/\mathcal I^N).

Proof. Use a finite affinoid cover from Lemma 2.34.4. On one member V=M(A)V=\mathcal M(A) write M=F(V)M=\mathcal F(V) and J=Ian(V)J=\mathcal I^{\mathrm{an}}(V); both are finite over AA. For x∈Vx\in V, with prime p=ker⁡∣⋅∣x\mathfrak p=\ker|\cdot|_x, the fibre M⊗AH(x)M\otimes_A\mathcal H(x) is zero exactly when Mp=0M_{\mathfrak p}=0. Indeed, the residue-field extension κ(p)⊂H(x)\kappa(\mathfrak p)\subset\mathcal H(x) is faithful, and Nakayama’s lemma applies to the finite module MpM_{\mathfrak p}. Moreover Mp=0M_{\mathfrak p}=0 exactly when Ann⁡A(M)⊄p\operatorname{Ann}_A(M)\not\subset\mathfrak p: if the localization vanishes, the product of finitely many denominators killing a generating list of MM lies in the annihilator and outside p\mathfrak p; the converse is immediate.

Put B=A/Ann⁡A(M)B=A/\operatorname{Ann}_A(M) with its affinoid quotient norm. Every point of M(B)\mathcal M(B) gives a nonzero fibre of MM, so the support hypothesis forces every element of JJ to vanish there. By the spectral-radius formula and Proposition 3.21.4, its image in BB is nilpotent. If a1,…,asa_1,\ldots,a_s generate JJ and aieia_i^{e_i} annihilates MM, then J1+∑i(ei−1)M=0J^{1+\sum_i(e_i-1)}M=0: each monomial of this total degree has some exponent at least eie_i. Take the maximum of these finitely many exponents over the cover. The resulting global annihilation and Lemma 2.42.7 give the stated factorization through ZNanZ_N^{\mathrm{an}}. ◻

Theorem 2.34.6. NAG-001731 (Cohomology comparison for proper schemes). For every proper finite-type KK-scheme XX and every coherent F\mathcal F on XX, the natural maps are isomorphisms Hq(X,F) ⟶ Hq(XGan,Fan)(q≥0).H^q(X,\mathcal F)\ \longrightarrow\ H^q(X^{\mathrm{an}}_G,\mathcal F^{\mathrm{an}})\qquad(q\ge0). The same comparison is a quasi-isomorphism for every bounded complex with coherent cohomology sheaves.

Proof. We first record how to use a comparison already known for the cohomology sheaves of a bounded complex CC. The canonical truncation triangles [StacksAlg] (Tag 08J5) express CC by finitely many successive extensions of Hj(C)[−j]\mathcal H^j(C)[-j]. Flatness of αX∗\alpha_X^* preserves these triangles and their cohomology sheaves. Applying derived global sections and the long exact sequences shows, by induction on the number of nonzero cohomology sheaves, that comparison for each Hj(C)\mathcal H^j(C) implies comparison for CC. The comparison maps here are the maps induced by regular sections mapping to analytic sections, or equivalently the adjunction maps for the flat morphism αX\alpha_X. They are natural in distinguished triangles.

Fix XX and perform Noetherian induction on closed subsets Z⊂∣X∣Z\subset|X|. The induction assertion is the comparison for every coherent module on XX supported in ZZ. It is clear for the empty set. Let F\mathcal F have support contained in ZZ. If its support is a proper closed subset, the induction hypothesis applies. Otherwise let T↪XT\hookrightarrow X be defined by Ann⁡(F)\operatorname{Ann}(\mathcal F); its underlying space is ZZ. The module F\mathcal F is the closed direct image of a coherent module FT\mathcal F_T on TT, and Lemma 2.34.4 identifies the two comparison problems.

Apply the algebraic Chow lemma [StacksAlg] (Tag 0200) to the proper scheme TT. It supplies a proper surjection π:Y→T\pi:Y\to T, an isomorphism above a dense open U⊂TU\subset T, and an immersion Y→PKnY\to\mathbf P^n_K. Since YY is proper over KK, the immersion is closed: its map to PKn\mathbf P^n_K is proper by [StacksAlg] (Tag 01W6), and a proper immersion is closed by its open–closed factorization. Thus YY is projective over KK. The graph of π\pi is closed because TT is separated; combining it with Y↪PKnY\hookrightarrow\mathbf P^n_K gives a closed immersion Y↪PTnY\hookrightarrow\mathbf P^n_T. Consequently π\pi is also projective. Put E=π∗FT\mathcal E=\pi^*\mathcal F_T and D=Rπ∗ED=R\pi_*\mathcal E. Its cohomology sheaves are coherent by [StacksAlg] (Tag 02O5) and vanish in sufficiently large degrees, as the finite standard relative-projective cover computes them.

Take the cone CC in the natural triangle FT⟶D⟶C⟶FT[1].\mathcal F_T\longrightarrow D\longrightarrow C \longrightarrow\mathcal F_T[1]. The first map is adjunction, followed by the map to the derived direct image. On UU it is the identity, because π\pi is an isomorphism there. Hence every Hj(C)\mathcal H^j(C) is coherent and supported in the proper closed subset T∖UT\setminus U of ZZ. Closed direct image to XX, the induction hypothesis, and the first paragraph give comparison for CC.

For DD, the derived composition of global sections and direct image [StacksAlg] (Tag 0D6E), together with Theorem 2.34.3, identifies its comparison with RΓ(Y,E)⟶RΓ(YGan,Ean).R\Gamma(Y,\mathcal E)\longrightarrow R\Gamma(Y^{\mathrm{an}}_G,\mathcal E^{\mathrm{an}}). This is a quasi-isomorphism by Corollary 2.42.9. Compatibility with adjunction in the relative comparison identifies the maps in the two triangles, not only their objects. The long exact cohomology sequences of these triangles therefore give comparison for FT\mathcal F_T and hence for F\mathcal F. Noetherian induction proves the assertion for all coherent sheaves. The first paragraph then proves the assertion for every bounded coherent complex. ◻

Proposition 2.34.7. NAG-001732 (Hom and extension comparison on proper schemes). Let XX be proper over KK, and let E,F\mathcal E,\mathcal F be coherent modules on XX. For every q≥0q\ge 0, exact analytification induces an isomorphism Ext⁡OXq(E,F) ≃ Ext⁡OXGanq(Ean,Fan).\operatorname{Ext}^q_{\mathcal O_X}(\mathcal E,\mathcal F) \ \simeq\ \operatorname{Ext}^q_{\mathcal O_{X^{\mathrm{an}}_G}} (\mathcal E^{\mathrm{an}},\mathcal F^{\mathrm{an}}). For q=1q=1 this is the map which analytifies short exact sequences; for q=0q=0 it gives full faithfulness.

Proof. We first compare sheaf Ext locally. On an affine open Spec⁡(R)\operatorname{Spec}(R) choose a resolution of the finite module representing E\mathcal E by finite free RR-modules, continuing as far as necessary. Each successive kernel is finite because RR is Noetherian. Flat analytification makes this an exact resolution by finite free analytic modules. Applying Hom into F\mathcal F and its analytification gives complexes which agree after pullback, since each free term is finite. In each fixed degree only finitely many terms are needed to compute kernel modulo image. Exact pullback thus gives (E ⁣xtXj(E,F))an≃E ⁣xtXGanj(Ean,Fan).(\mathcal E\!xt_X^j(\mathcal E,\mathcal F))^{\mathrm{an}} \simeq \mathcal E\!xt_{X^{\mathrm{an}}_G}^j (\mathcal E^{\mathrm{an}},\mathcal F^{\mathrm{an}}). The algebraic Ext sheaves are coherent by the same finite-free calculation. These local identifications are independent of the resolution: they are the cohomology maps of the canonical derived-Hom pullback morphism [StacksAlg] (Tag 08JF). They therefore glue to a quasi-isomorphism αX∗RH ⁣omX(E,F)≃RH ⁣omXGan(Ean,Fan).\alpha_X^*R\mathcal H\!om_X(\mathcal E,\mathcal F) \simeq R\mathcal H\!om_{X^{\mathrm{an}}_G} (\mathcal E^{\mathrm{an}},\mathcal F^{\mathrm{an}}). Put C=RH ⁣omX(E,F)C=R\mathcal H\!om_X(\mathcal E,\mathcal F). It has coherent cohomology in nonnegative degrees, although it need not be bounded above. Fix qq. The truncation τ≤qC\tau_{\le q}C is a bounded coherent complex. The complementary truncation belongs to D≥q+1D^{\ge q+1}; its derived global sections also belong to D≥q+1D^{\ge q+1}, as a bounded below injective resolution may be chosen to start in that degree. Thus HqRΓ(C)H^qR\Gamma(C) is computed by τ≤qC\tau_{\le q}C. The same holds after exact pullback. Theorem 2.34.6 and the truncation triangle now give equality in degree qq of the two derived global Hom complexes. By [StacksAlg] (Tag 08JA), applied after shifting the second argument by qq, these groups are the stated Ext groups. Finally [StacksAlg] (Tag 06XV) identifies derived Ext in degree one with short exact extensions. Its identification takes an extension to its connecting morphism. Exact pullback preserves that triangle and connecting morphism, proving the claimed compatibility. ◻

Theorem 2.34.8. NAG-001733 (Proper analytic GAGA). For every proper finite-type scheme XX over a complete non-Archimedean field KK, analytification is an exact tensor equivalence Coh⁡(X) → ∼  Coh⁡(XGan).\operatorname{Coh}(X)\ \xrightarrow{\ \sim\ }\ \operatorname{Coh}(X^{\mathrm{an}}_G). It preserves coherent cohomology and restricts to an equivalence on vector bundles, and hence on line bundles.

Proof. Full faithfulness is Proposition 2.34.7. Exactness and compatibility with tensor products are Theorem 2.33.34; cohomology comparison is Theorem 2.34.6. It remains to prove essential surjectivity. We do this by Noetherian induction on closed subsets Z⊂∣X∣Z\subset|X|, with the assertion that every analytic coherent module supported in ZanZ^{\mathrm{an}} is algebraizable on XX. For a closed subset, ZanZ^{\mathrm{an}} means the analytification of its reduced closed subscheme; this underlying set is unchanged by a thickening. The empty case is immediate.

Let F\mathcal F be supported in ZanZ^{\mathrm{an}}. Lemma 2.34.5 realizes it as the closed direct image of a coherent module FT\mathcal F_T on TanT^{\mathrm{an}}, for a closed subscheme T↪XT\hookrightarrow X with underlying space ZZ. Apply Chow as in the preceding proof: there is a projective surjection π:Y→T\pi:Y\to T with YY projective over KK, which is an isomorphism above a dense open U⊂TU\subset T. The coherent module (πan)∗FT(\pi^{\mathrm{an}})^*\mathcal F_T is algebraizable on YY by Corollary 2.42.9; choose an algebraization E\mathcal E and its identifying isomorphism. Relative projective comparison in degree zero identifies G:=π∗an(πan)∗FT≃(π∗E)an.\mathcal G:=\pi^{\mathrm{an}}_*(\pi^{\mathrm{an}})^*\mathcal F_T \simeq (\pi_*\mathcal E)^{\mathrm{an}}. In particular G\mathcal G is coherent. The kernel K\mathcal K and cokernel C\mathcal C of the adjunction FT→G\mathcal F_T\to\mathcal G are coherent and vanish on UanU^{\mathrm{an}}. Their closed direct images on XanX^{\mathrm{an}} are supported in the proper closed subset Z′=T∖UZ'=T\setminus U of ZZ. By the induction hypothesis they are algebraizations of coherent modules KX,CXK_X,C_X on XX. These algebraizations are annihilated by the ideal of TT: the respective multiplication maps by this ideal analytify to zero, so full faithfulness makes them zero already on XX. Thus they descend to coherent KT,CTK_T,C_T on TT, whose analytifications are K,C\mathcal K,\mathcal C.

The map G→C\mathcal G\to\mathcal C algebraizes by full faithfulness on the proper scheme TT. Let JTJ_T be the kernel of the resulting map π∗E→CT\pi_*\mathcal E\to C_T. Exactness gives an exact analytic sequence 0⟶KTan⟶FT⟶JTan⟶0.0\longrightarrow K_T^{\mathrm{an}}\longrightarrow\mathcal F_T \longrightarrow J_T^{\mathrm{an}}\longrightarrow0. Proposition 2.34.7 in degree one lifts its class to Ext⁡T1(JT,KT)\operatorname{Ext}^1_T(J_T,K_T). By [StacksAlg] (Tag 06XV) this class is represented by an algebraic short exact sequence 0→KT→FT→JT→00\to K_T\to F_T\to J_T\to 0; its middle term is coherent. Indeed, locally take a finite free surjection P→JTP\to J_T and lift its finitely many basis sections into FTF_T after shrinking. Then KT⊕P→FTK_T\oplus P\to F_T is surjective and its kernel, by projection to PP, is isomorphic to the coherent kernel of P→JTP\to J_T. Thus FTF_T is a cokernel of a map between coherent modules. The analytic extension is isomorphic to the displayed sequence with identity on its ends. Hence FTan≃FTF_T^{\mathrm{an}}\simeq\mathcal F_T. Closed direct image to XX algebraizes F\mathcal F. This completes the Noetherian induction and proves the equivalence.

Finally finite local freeness descends along a faithfully flat local map. Here is the elementary argument in the form needed. If R→SR\to S is such a map, MM is finite over the Noetherian local ring RR, and S⊗RMS\otimes_R M is free of rank rr, then the residue-field extension shows dim⁡κ(R)M/mRM=r\dim_{\kappa(R)}M/\mathfrak m_RM=r. Lift a basis; Nakayama gives a surjection Rr→MR^r\to M. After tensoring with SS this is a surjection between two rank-rr free modules over the local ring SS, and hence an isomorphism (reduce its determinant modulo the maximal ideal). Faithful flatness kills its kernel, so MM is free.

For completeness, local freeness in the analytic GG-topology also yields free ordinary stalks here. On an affinoid M(A)\mathcal M(A) representing a finite module MM, choose a trivializing affinoid domain M(B)\mathcal M(B) containing the point in question. The domain map A→BA\to B is flat; localizing at the corresponding primes gives a faithfully flat local map. The preceding argument makes MM free at that prime of AA. A finite presentation then makes it free after inverting one element not in the prime: lift the basis and clear the finitely many relations for the kernel and cokernel. A small rational neighbourhood on which that element is invertible gives the same free ordinary analytic stalk. Now every scheme point has an analytic lift by Lemma 2.34.2, and its local map is faithfully flat by Theorem 2.33.34. Applying the local descent argument shows that an algebraization of an analytic vector bundle is locally free of the same rank. The converse follows by tensoring a local basis. Rank one gives the line-bundle assertion. ◻

2.35 Finite neighbourhoods and preservation of smoothness

We use the usual local definitions: an analytic morphism is quasi-finite if, near each source point, it restricts on ordinary open neighbourhoods to a finite morphism. It is étale if these finite local morphisms are flat and their coherent relative differential sheaf vanishes. It is smooth if locally on ordinary opens it factors as an étale morphism to a relative analytic affine space followed by its projection. The next argument verifies this local form directly for algebraic smooth morphisms.

Lemma 2.35.1. NAG-001735 (Separating the branches of a finite morphism). Let f:Y→Xf:Y\to X be a norm-admissible finite morphism of good analytic spaces. Its fibre over any x∈Xx\in X is a finite set {y1,…,ym}\{y_1,\ldots,y_m\}. Given neighbourhoods NiN_i of these points, there is an ordinary open neighbourhood UU of xx such that f−1(U)=∐i=1mWi,yi∈Wi⊂Ni,f^{-1}(U)=\coprod_{i=1}^m W_i,\qquad y_i\in W_i\subset N_i, where each WiW_i is open and closed in f−1(U)f^{-1}(U), and Wi→UW_i\to U is finite. On ordinary local rings there is a canonical product decomposition (f∗OY)x≃∏i=1mOY,yi.(f_*\mathcal O_Y)_x \simeq \prod_{i=1}^m\mathcal O_{Y,y_i}. If X=M(A)X=\mathcal M(A) and Y=M(B)Y=\mathcal M(B), its left side is B⊗AOX,xB\otimes_A\mathcal O_{X,x}.

Proof. Work over an affinoid neighbourhood VV of xx. The fibre has algebra D=B⊗AH(x)D=B\otimes_A\mathcal H(x), finite dimensional over the complete field H(x)\mathcal H(x), by Proposition 2.33.19 and Theorem 2.33.11. Every prime of a finite-dimensional algebra over a field is maximal, since a finite-dimensional domain is a field: multiplication by a nonzero element is injective and hence surjective. There are only finitely many maximal ideals. Indeed, the Chinese remainder map for any finite collection of distinct maximal ideals is surjective onto their residue fields, so their number is at most dim⁡H(x)D\dim_{\mathcal H(x)}D. Each residue field is finite over H(x)\mathcal H(x) and has exactly one extending absolute value by Lemma 1.12.1; the quotient map is bounded for the canonical finite-dimensional norms. Consequently M(D)\mathcal M(D) has exactly one point for each of these finitely many maximal ideals.

The space f−1(V)f^{-1}(V) is affinoid, hence compact Hausdorff, and its map to VV is closed. Choose pairwise disjoint ordinary open neighbourhoods Oi⊂NiO_i\subset N_i of its fibre points. The complement of their union is compact and has closed image not containing xx. Remove that image and intersect with the ordinary interior of VV to obtain UU. Then Wi=Oi∩f−1(U)W_i=O_i\cap f^{-1}(U) give the stated open-and-closed partition. For any affinoid domain V′⊂UV'\subset U, f−1(V′)=M(B′)f^{-1}(V')=\mathcal M(B') is affinoid and finite over V′V'. The function equal to one on Wi∩f−1(V′)W_i\cap f^{-1}(V') and zero on its complement is analytic by sheaf gluing on this ordinary open partition. It is an idempotent ei∈B′e_i\in B', since the global sections on an affinoid are its algebra. Thus Wi∩f−1(V′)=M(eiB′)W_i\cap f^{-1}(V')=\mathcal M(e_iB'). The summand eiB′e_iB' is a finite module with its canonical norm: projection by eie_i is bounded and finite-module maps are norm-strict. This proves finiteness of Wi→UW_i\to U.

Restriction of germs defines the displayed product map on local rings. Given a germ at each yiy_i, choose representatives on NiN_i, separate the neighbourhoods, and shrink UU as just done. Their restrictions glue on the disjoint WiW_i to a section over f−1(U)f^{-1}(U). This proves surjectivity. If all germs of such a section vanish, choose neighbourhoods where they vanish and shrink once more; the section then vanishes on an entire inverse-image neighbourhood. This proves injectivity. In the affinoid case, take the filtered colimit over affinoid neighbourhoods V′V' of xx. Finite base change gives O(f−1(V′))=B⊗AAV′\mathcal O(f^{-1}(V'))=B\otimes_A A_{V'}; ordinary tensor commutes with this colimit, which is OX,x\mathcal O_{X,x} by Proposition 2.33.16. This gives the final identification. ◻

Theorem 2.35.2. NAG-001736 (Analytification preserves étale morphisms). An étale morphism of finite-type KK-schemes analytifies to an étale analytic morphism. More precisely, near every analytic source point it restricts on ordinary open neighbourhoods to a finite flat morphism whose relative analytic differentials vanish.

Proof. Fix a source point yy over xx. By [StacksAlg] (Tag 02GT), choose affine Zariski neighbourhoods containing their scheme images on which the morphism is standard étale. In the notation of [StacksAlg] (Tag 00UB), its rings have the form R⟶Sg,S=R[T]/(P),P monic,P′∈Sg×.R\longrightarrow S_g,\qquad S=R[T]/(P),\quad P\text{ monic},\quad P'\in S_g^\times. The scheme Spec⁡(Sg)\operatorname{Spec}(S_g) is the open subscheme D(g)D(g) of the finite free scheme Spec⁡(S)\operatorname{Spec}(S) over Spec⁡(R)\operatorname{Spec}(R). Analytification preserves this finite morphism and this open immersion, by Theorem 2.33.26 and Proposition 2.33.24. Take an affinoid neighbourhood V=M(A)V=\mathcal M(A) of xx in Spec⁡(R)an\operatorname{Spec}(R)^{\mathrm{an}}. The inverse image in Spec⁡(S)an\operatorname{Spec}(S)^{\mathrm{an}} is M(B)\mathcal M(B) with B=A⊗RS=A[T]/(P)B=A\otimes_R S=A[T]/(P) and its canonical finite module norm. It is a finite free AA-module.

Apply Lemma 2.35.1, requiring the neighbourhood of yy to be contained in D(g)anD(g)^{\mathrm{an}}. It gives ordinary open neighbourhoods WW of yy and UU of xx for which W→UW\to U is finite. For each affinoid V′⊂UV'\subset U, its inverse-image algebra is a summand eB′eB', where B′=A′⊗RSB'=A'\otimes_R S and A′=O(V′)A'=\mathcal O(V'). Since B′B' is finite free over A′A', this summand is projective, and in particular flat, over A′A'. The element gg is a unit in eB′eB': it vanishes at no point of its spectrum, so the affinoid unit criterion applies. Therefore P′P' is also a unit in eB′eB'.

For the finite algebra B′=A′[T]/(P)B'=A'[T]/(P), a bounded A′A'-derivation into any finite Banach B′B'-module is determined by dTdT with the sole relation P′dT=0P'dT=0. Conversely this relation defines a derivation on all polynomials modulo PP; it is bounded because the monic remainder basis is a finite module basis. Thus its analytic differential module is B′ dT/(P′dT)B'\,dT/(P'dT). Restriction to the open-and-closed summand eB′eB' tensors this module with eB′eB', by Theorem 2.41.5. It is zero because P′P' is a unit there. The local maps W→UW\to U are therefore finite flat with vanishing analytic differentials, which is the asserted étale local form. ◻

Corollary 2.35.3. NAG-001737 (Analytification preserves smooth morphisms). If f:Y→Xf:Y\to X is a smooth morphism of finite-type KK-schemes, then fan:Yan→Xanf^{\mathrm{an}}:Y^{\mathrm{an}}\to X^{\mathrm{an}} is smooth.

Proof. At the scheme image of any analytic source point, [StacksAlg] (Tag 054L) gives Zariski open neighbourhoods Y0,X0Y_0,X_0 and a factorization Y0→ eˊtale AX0d⟶X0.Y_0\xrightarrow{\ \text{\'etale}\ }\mathbf A^d_{X_0} \longrightarrow X_0. Proposition 2.33.24 makes their analytifications ordinary open neighbourhoods. By Theorem 2.33.25, the analytification of the middle space is the relative analytic affine space over X0anX_0^{\mathrm{an}}; its projection is the usual analytic projection. The first arrow is étale after analytification by Theorem 2.35.2. This is exactly the defining local smooth factorization. ◻

References. The local finite-branch argument follows the mechanism of [Ber93] (§§3.2–3.3); using the standard étale form makes the finite covering algebra free and avoids a separate analytic flatness theorem. All algebraic existence inputs in these proofs are the cited Stacks lemmas.

Scalar stalks and descent of analytic flatness

Theorem 2.35.4. NAG-001739 (Faithful flatness of scalar-extension stalks). Let XX be a good KK-analytic space, let L/KL/K be any complete valued extension, and let u∈XLu\in X_L map to x∈Xx\in X. The local homomorphism OX,x→OXL,u\mathcal O_{X,x}\to\mathcal O_{X_L,u} is faithfully flat. For every coherent F\mathcal F on XX, its analytic scalar pullback satisfies (FL)u≃Fx⊗OX,xOXL,u.(\mathcal F_L)_u \simeq \mathcal F_x\otimes_{\mathcal O_{X,x}}\mathcal O_{X_L,u}.

Proof. Put R=OX,xR=\mathcal O_{X,x} and S=OXL,uS=\mathcal O_{X_L,u}. Fix an affinoid neighborhood V=M(A)V=\mathcal M(A) of xx. Its inverse image VLV_L is an affinoid neighborhood of uu: the inverse image of the ordinary interior of VV is an ordinary neighborhood contained in VLV_L. Rational neighborhoods WW of uu inside VLV_L form a cofinal family. Each composite A⟶AL⟶AWA\longrightarrow A_L\longrightarrow A_W is flat by Theorem 2.33.6 and Proposition 2.38.2. Proposition 2.33.16 identifies SS with the filtered colimit of these AWA_W. The finite-relation proof used in Theorem 2.33.17 therefore proves that SS is flat over this fixed AA.

To pass from AA to RR, let I⊂RI\subset R be finitely generated and let z∈I⊗RSz\in I\otimes_R S have zero image in SS. Choose generators aia_i of II and write z=∑iai⊗siz=\sum_i a_i\otimes s_i with si∈Ss_i\in S. Lift the aia_i to the algebra AA of one sufficiently small affinoid neighborhood VV. If J⊂AJ\subset A is the ideal generated by the lifts, the corresponding tensor in J⊗ASJ\otimes_A S maps to zz and has zero image in SS. Since SS is AA-flat, this element is zero. The ideal criterion [StacksAlg] (Tag 00HD) gives RR-flatness. Evaluation is compatible with the isometric embedding H(x)↪H(u)\mathcal H(x)\hookrightarrow\mathcal H(u), so this map is local. It is faithfully flat by [StacksAlg] (Tag 00HP).

On VV, Theorem 2.39.6 represents F∣V\mathcal F|_V by a finite AA-module MM. Pullback sends a finite free presentation to its cokernel presentation with the same matrix over ALA_L. Thus its scalar pullback is represented by M⊗AALM\otimes_A A_L, with the finite scalar comparison of Theorem 2.33.6. Taking stalks gives M⊗ASM\otimes_A S, while Fx=M⊗AR\mathcal F_x=M\otimes_A R. The displayed identity is now the associativity of ordinary tensor products. ◻

Theorem 2.35.5. NAG-001740 (Amalgamation of complete valued fields). Let EE and FF be complete valued extensions of a complete valued field KK. There is a complete valued field Ω\Omega with isometric embeddings E↪ΩE\hookrightarrow\Omega and F↪ΩF\hookrightarrow\Omega agreeing on KK.

Proof. For every finite subset S⊂F×S\subset F^\times, consider the bounded evaluation map K{(∣s∣−1Ts)s∈S}⟶F,Ts⟼s,K\{(|s|^{-1}T_s)_{s\in S}\}\longrightarrow F,\qquad T_s\longmapsto s, and let ASA_S be its quotient by the kernel. The kernel is closed; the quotient is a nonzero generalized affinoid algebra by Corollary 1.18.4. Evaluation gives a point xS∈M(AS)x_S\in\mathcal M(A_S) and an injective map from K[S]K[S] into ASA_S. If S⊂TS\subset T, substitution gives a bounded map AS→ATA_S\to A_T: the variables have norms at most their assigned radii, and an element of the first kernel maps to zero because its evaluation in FF is zero. These maps preserve the evaluation points.

By Theorem 2.33.13, the fibre PS={w∈ME((AS)E):w∣AS=xS}P_S=\{w\in\mathcal M_E((A_S)_E):w|_{A_S}=x_S\} is nonempty and compact. The maps AS→ATA_S\to A_T define an inverse system of these compact spaces. Its inverse limit is nonempty. Indeed, the product ∏SPS\prod_S P_S is compact by [StacksAlg] (Tag 08ZU), and each compatibility equation is closed in it. A finite collection of equations involves only finitely many subsets; choose a finite TT containing their union and restrict any point of PTP_T to all the subsets involved. Thus these closed conditions have the finite intersection property.

Choose a compatible family (wS)(w_S). For S⊂TS\subset T, the completed-residue-field construction gives an isometric embedding H(wS)↪H(wT)\mathcal H(w_S)\hookrightarrow\mathcal H(w_T), by agreement of evaluations. These embeddings satisfy the cocycle identity on the dense evaluated fraction fields. Form their directed union as fields and complete it; call the resulting field Ω\Omega. Every H(wS)\mathcal H(w_S) contains EE isometrically, and the transition maps fix it. The compatible maps K[S]→H(wS)K[S]\to\mathcal H(w_S) are injective and preserve the given absolute value on FF, because wSw_S lies above xSx_S. As the finite subsets range over F×F^\times, their union is all of FF. They therefore give an isometric embedding F↪ΩF\hookrightarrow\Omega, agreeing with the embedding of EE on KK. Completion of a valued field is again a valued field: inversion of a Cauchy sequence with nonzero limit is Cauchy once its norms are bounded below. This completes the construction. ◻

Lemma 2.35.6. NAG-001741 (Flatness descent through a square of local rings). Consider a commutative square of rings R→R′R\to R', R→SR\to S, R′→S′R'\to S', S→S′S\to S'. Suppose R′R' is flat over RR and S′S' is faithfully flat over SS. If an SS-module MM has M⊗SS′M\otimes_S S' flat over R′R', then MM is flat over RR.

Proof. The module M′=M⊗SS′M'=M\otimes_S S' is flat over RR, by composition of the two exact tensor functors. For a finitely generated ideal I⊂RI\subset R, let NN be the kernel of I⊗RM→MI\otimes_R M\to M. This is an SS-module. Flatness over SS identifies N⊗SS′N\otimes_S S' with the kernel of I⊗RM′→M′I\otimes_R M'\to M', which is zero. Faithfulness gives N=0N=0. The criterion [StacksAlg] (Tag 00HD) concludes. ◻

Theorem 2.35.7. NAG-001742 (Descent from one scalar-extension point). Let f:Y→Xf:Y\to X be a morphism of good KK-analytic spaces, let F\mathcal F be coherent on YY, and let L/KL/K be a complete valued extension. For y∈Yy\in Y and any point u∈YLu\in Y_L above yy, F\mathcal F is analytically XX-flat at yy if and only if FL\mathcal F_L is analytically XLX_L-flat at uu.

Proof. The forward implication follows directly by composing the base changes quantified in Definition 2.9.1. For the converse, let F/KF/K be any complete valued extension, let X′→XX'\to X be a good FF-analytic base change, and let y′∈Y′=Y×XX′y'\in Y'=Y\times_X X' map to yy. Write x′x' for its image in X′X'. We must prove that FY′,y′\mathcal F_{Y',y'} is flat over OX′,x′\mathcal O_{X',x'}.

Apply Theorem 2.35.5 to the complete extensions H(u)\mathcal H(u) and H(y′)\mathcal H(y') of H(y)\mathcal H(y). It gives a complete field Ω\Omega containing both compatibly. In particular Ω\Omega contains LL and FF over KK. The inclusion H(y′)↪Ω\mathcal H(y')\hookrightarrow\Omega defines an Ω\Omega-rational point ω\omega of YΩ′Y'_\Omega above y′y', by the pointwise fibre description in Theorem 2.33.11, applied locally. Its image in YLY_L is uu: on any affinoid chart the evaluations of its analytic functions and of the coefficients in LL are exactly those obtained through H(u)↪Ω\mathcal H(u)\hookrightarrow\Omega. Denote its image in XΩ′X'_\Omega by vv.

Associativity of the analytic fibre products identifies YΩ′Y'_\Omega with the pullback of YLY_L along XΩ′→XLX'_\Omega\to X_L. The assumed analytic flatness at uu therefore makes (FYΩ′)ω(\mathcal F_{Y'_\Omega})_\omega flat over OXΩ′,v\mathcal O_{X'_\Omega,v}. In the commutative local square OX′,x′⟶OXΩ′,v↓↓OY′,y′⟶OYΩ′,ω\begin{matrix} \mathcal O_{X',x'}&\longrightarrow&\mathcal O_{X'_\Omega,v}\\ \big\downarrow&&\big\downarrow\\ \mathcal O_{Y',y'}&\longrightarrow&\mathcal O_{Y'_\Omega,\omega} \end{matrix} both horizontal maps are faithfully flat by Theorem 2.35.4. The same theorem identifies the pullback coherent stalk with the ordinary tensor product. Lemma 2.35.6 now proves the required naive flatness at y′y'. Since the complete extension and analytic base change were arbitrary, F\mathcal F is analytically flat at yy. ◻

References. The descent architecture follows [Duc18] (4.5.3 and 4.5.6). Scalar stalk flatness and the compatible complete-field construction are proved above; these citations identify the source of the method.

Convergent germs and ordinary local completions

Lemma 2.35.8. NAG-001744 (Noetherianity of convergent germs). Suppose KK is nontrivially valued. Let CnC_n be the subring of K[[T1,…,Tn]]K[[T_1,\ldots,T_n]] consisting of the series convergent on some positive polydisc about 00. Then CnC_n is a Noetherian local ring with maximal ideal (T1,…,Tn)(T_1,\ldots,T_n), residue field KK, and maximal-ideal completion K[[T1,…,Tn]]K[[T_1,\ldots,T_n]].

Proof. A series with nonzero constant term becomes a unit on a sufficiently small polydisc: its nonconstant part has norm smaller than the constant term, and geometric inversion applies. A series with zero constant term is a sum ∑iTigi\sum_iT_i g_i with each gig_i convergent, by partitioning its monomials according to a variable dividing them. Thus the asserted maximal ideal and residue field follow. Invertible linear changes of the variables preserve CnC_n, since either linear substitution maps a sufficiently small polydisc into any prescribed one.

We prove Noetherianity by induction on nn, with C0=KC_0=K. Let I≠0I\ne 0 be an ideal and choose 0≠f∈I0\ne f\in I. If its constant term is nonzero, I=CnI=C_n. Otherwise, let PdP_d be its first nonzero homogeneous part. The field KK is infinite because its absolute value is nontrivial. A nonzero polynomial over an infinite field cannot vanish on every point: induct on the number of variables and use the root bound in one variable. Hence, after permuting coordinates if needed, choose a1,…,an−1∈Ka_1,\ldots,a_{n-1}\in K with Pd(a1,…,an−1,1)≠0P_d(a_1,\ldots,a_{n-1},1)\ne 0. The invertible linear substitution Ti↦Ti+aiTnT_i\mapsto T_i+a_iT_n makes f(0,…,0,Tn)=cTnd+terms of higher degree,c≠0.f(0,\ldots,0,T_n)=cT_n^d+\text{terms of higher degree}, \qquad c\ne0. Write T′=(T1,…,Tn−1)T'=(T_1,\ldots,T_{n-1}) and f=∑jaj(T′)Tnjf=\sum_j a_j(T')T_n^j. For arbitrarily small s>0s>0, one can choose a positive polyradius r′r' such that ada_d is a unit in A=K{(r′)−1T′}A=K\{(r')^{-1}T'\} and the normalized series ad−1fa_d^{-1}f is distinguished of degree dd at radius ss. Here are the norm checks. First choose ss small enough that the scalar tail ∑j>daj(0)Tnj\sum_{j>d}a_j(0)T_n^j has ss-norm strictly less than ∣c∣sd|c|s^d. At this fixed ss, convergence leaves only finitely many coefficients outside any prescribed smaller bound for the weighted tail. Shrinking r′r' makes those finitely many aj(T′)−aj(0)a_j(T')-a_j(0) arbitrarily small, makes ad−ca_d-c smaller than ∣c∣|c|, and makes ∥aj∥sj≤∣c∣sd\|a_j\|s^j\le |c|s^d for j<dj<d, since aj(0)=0a_j(0)=0. The omitted tail remains uniformly small under this shrinking. Thus ∥ad−1∥=∣c∣−1\|a_d^{-1}\|=|c|^{-1} and all hypotheses of Theorem 1.10.3 hold.

Given any g∈Cng\in C_n, take these radii small enough that gg also converges. Division yields g=qf+∑j<dbj(T′)Tnjg=qf+\sum_{j<d}b_j(T')T_n^j with q∈Cnq\in C_n and bj∈Cn−1b_j\in C_{n-1}. Consequently Cn/(f)C_n/(f) is finite as a Cn−1C_{n-1}-module. By induction Cn−1C_{n-1} is Noetherian, so the submodule I/(f)I/(f) is finitely generated over Cn−1C_{n-1}. To recall the elementary module step, submodules of a finite free module are finitely generated by induction on its rank, using the first-coordinate ideal and its kernel; the statement for a finite module follows by inverse image under a finite free surjection. Lift generators of I/(f)I/(f) and adjoin ff. They generate II over CnC_n. Undoing the coordinate change proves that every ideal is finitely generated.

For every q≥1q\ge 1, total-degree truncation identifies Cn/(T)qC_n/(T)^q with K[T]/(T)qK[T]/(T)^q. Its kernel is exactly (T)q(T)^q: partition the terms of total degree at least qq among the finitely many degree-qq monomials dividing them; the quotients remain convergent. The inverse limit of these finite polynomial truncations is the formal power-series ring. ◻

Lemma 2.35.9. NAG-001745 (Rational points lie inside analytic domains). If UU is an analytic domain of a good KK-analytic space XX and x∈Ux\in U satisfies H(x)=K\mathcal H(x)=K, then UU contains an ordinary neighborhood of xx in XX. In particular the ordinary stalks on UU and XX agree at xx.

Proof. Choose an affinoid neighborhood X0X_0 of xx in XX and an affinoid domain of U∩X0U\cap X_0 containing xx. By Theorem 2.36.6, this domain has a finite cover by domains rational in X0X_0. Choose one containing xx, given by ∣fi∣≤ri∣g∣|f_i|\le r_i|g|. The unit-ideal condition implies g(x)≠0g(x)\ne 0. Write ai=fi(x)∈Ka_i=f_i(x)\in K and b=g(x)∈K×b=g(x)\in K^\times. On the ordinary neighborhood ∣g−b∣<∣b∣|g-b|<|b|, we have ∣g∣=∣b∣|g|=|b|. If ai≠0a_i\ne 0, further impose ∣fi−ai∣<∣ai∣|f_i-a_i|<|a_i|; then ∣fi∣=∣ai∣≤ri∣b∣|f_i|=|a_i|\le r_i|b|. If ai=0a_i=0, impose ∣fi∣<ri∣b∣|f_i|<r_i|b|. These finitely many open conditions hold at xx and force all the defining rational inequalities. Intersect also with the ordinary interior of X0X_0. The resulting ordinary neighborhood is contained in UU. ◻

Proposition 2.35.10. NAG-001746 (Formal presentations at rational analytic points). Let KK be nontrivially valued and X=M(K{r−1T}/I)X=\mathcal M(K\{r^{-1}T\}/I). At a KK-rational point a=(a1,…,an)a=(a_1,\ldots,a_n), put Ui=Ti−aiU_i=T_i-a_i and let IaI_a denote the ideal of CnC_n generated by the translated elements of II. Then OX,a≃Cn/Ia,OX,a^≃K[[U1,…,Un]]/IaK[[U1,…,Un]].\mathcal O_{X,a}\simeq C_n/I_a,\qquad \widehat{\mathcal O_{X,a}}\simeq K[[U_1,\ldots,U_n]]/ I_aK[[U_1,\ldots,U_n]]. If XX is replaced near aa by an analytic domain, these identifications are unchanged. If Z↪XZ\hookrightarrow X is a closed analytic subspace with coherent ideal J\mathcal J, then OZ,a=OX,a/Ja\mathcal O_{Z,a}=\mathcal O_{X,a}/\mathcal J_a.

Proof. Small polydiscs centered at aa form an ordinary neighborhood basis in the ambient polydisc. To verify cofinality, an ordinary neighborhood is specified by finitely many evaluations of analytic functions. On a sufficiently small centered polydisc, each such function differs uniformly from its value at aa by an arbitrarily small amount, by its convergent translated series. For a nonzero value this preserves its absolute value; for a zero value it gives any prescribed positive upper bound. These conditions realize every finite evaluation neighborhood. All sufficiently small centered polydiscs lie in the original polydisc, since ∣ai∣≤ri|a_i|\le r_i.

Intersecting them with XX gives a cofinal system of affinoid neighborhoods. Closed-subspace base change and finite-module exactness identify their algebras with the corresponding Tate algebras modulo the extended ideal II. The ideal II is finitely generated by Theorem 1.18.2. Passing to the filtered colimit therefore gives Cn/IaC_n/I_a: tensoring a finite ideal presentation, or lifting a finite equality to a smaller polydisc, verifies both its surjectivity and its kernel. The same argument applies to any coherent closed ideal J\mathcal J after choosing finite generators on an affinoid neighborhood.

Lemma 2.35.8 makes CnC_n Noetherian and computes its completion. Exactness of completion for finite modules, [StacksAlg] (Tag 00MA), identifies the completion of its quotient with the displayed formal quotient. Finally Lemma 2.35.9 says that any analytic domain containing aa is an ordinary neighborhood there, and hence does not change the stalk. ◻

Theorem 2.35.11. NAG-001747 (Noetherian ordinary local rings). For every good analytic space over any complete non-Archimedean field, every ordinary local ring is Noetherian.

Proof. Fix x∈Xx\in X and choose an affinoid neighborhood. Put E=H(x)E=\mathcal H(x). If EE is nontrivially valued, take L=EL=E; otherwise use Lemma 1.16.1 to adjoin one radius different from 11, obtaining a nontrivially valued complete extension L/EL/E. Evaluation A→E→LA\to E\to L, together with the identity on LL, defines an LL-rational point u∈XLu\in X_L above xx. The completed-residue-field and relative pushout properties verify this point directly.

By Proposition 2.35.10, S=OXL,uS=\mathcal O_{X_L,u} is Noetherian. The map R=OX,x→SR=\mathcal O_{X,x}\to S is faithfully flat by Theorem 2.35.4. Noetherianity descends here by an elementary ideal argument. For any ideal J⊂RJ\subset R, the ideal JSJS has finitely many generators. Their expressions use finitely many elements a1,…,am∈Ja_1,\ldots,a_m\in J, so JS=(a1,…,am)SJS=(a_1,\ldots,a_m)S. If J0=(a1,…,am)⊂RJ_0=(a_1,\ldots,a_m)\subset R, flatness identifies (J/J0)⊗RS(J/J_0)\otimes_R S with JS/J0S=0JS/J_0S=0, and faithfulness gives J=J0J=J_0. Thus every ideal of RR is finitely generated. ◻

References. This reconstructs the reduction in [Ber93] (Theorem 2.1.4): the convergent-germ argument, the rational-point completion, and faithful scalar-stalk descent are all explicit.

The Jacobian criterion by completed local rings

Lemma 2.35.12. NAG-001749 (The fibre over a rational base point). Let f:Y→Xf:Y\to X be a morphism of good spaces over a nontrivially valued field KK, and suppose x=f(y)x=f(y) is KK-rational. Put R=OX,xR=\mathcal O_{X,x}, S=OY,yS=\mathcal O_{Y,y}. Then OYx,y≃S/mRS.\mathcal O_{Y_x,y}\simeq S/\mathfrak m_R S. If also yy is KK-rational and a local relative presentation has nn variables and relation ideal II, then the cotangent-space dimension of this fibre local ring is nn minus the rank at yy of the relative Jacobian matrix of any finite generators of II.

Proof. Choose affinoid neighborhoods X0=M(A)X_0=\mathcal M(A) and Y0=M(B)Y_0=\mathcal M(B) with f(Y0)⊂X0f(Y_0)\subset X_0. Evaluation at xx is a bounded quotient A→KA\to K. Since KK is a finite AA-module, Proposition 2.33.19 identifies the fibre algebra with the ordinary quotient B/mxBB/\mathfrak m_xB. In an affinoid presentation A=K{r−1Z}/JA=K\{r^{-1}Z\}/J, translate xx to Z=0Z=0. Monomial partition, as in Proposition 2.35.10, shows both that mx\mathfrak m_x is generated by the ZiZ_i and that their germs generate mR\mathfrak m_R.

The point yy need not be rational. Take a cofinal system of ordinary affinoid neighborhoods VV of yy in Y0Y_0. Since the fibre is a closed analytic subspace with its subspace topology, the intersections V∩YxV\cap Y_x are cofinal ordinary affinoid neighborhoods of yy in YxY_x: any ordinary neighborhood in the closed subspace is the intersection with an ambient open neighborhood, inside which goodness supplies such a VV. On every VV, closed-subspace base change identifies the fibre algebra with O(V)/(Z1,…,Zd)O(V)\mathcal O(V)/(Z_1,\ldots,Z_d)\mathcal O(V). Passing to the filtered colimit gives OYx,y=S/(Z1,…,Zd)S\mathcal O_{Y_x,y}=S/(Z_1,\ldots,Z_d)S: every representative lifts on the same VV, and a representative vanishing in the quotient is, on some smaller VV, a finite combination of these generators. Because their germs generate mR\mathfrak m_R, this is the asserted quotient.

At a rational yy, its completed fibre ring is K[[T1,…,Tn]]/I^xK[[T_1,\ldots,T_n]]/\widehat I_x, after translating its relative coordinates to zero. The maximal ideal modulo its square is consequently KnK^n modulo the linear terms of the relations. Those terms are the rows of the relative Jacobian evaluated at yy. Completion does not change this vector space: its quotient modulo the square of the maximal ideal is the second finite truncation in the definition of completion. ◻

Proposition 2.35.13. NAG-001750 (The completed Jacobian calculation). Let KK be nontrivially valued, let XX be good, and let DD be an analytic domain of AXn,an\mathbb A_X^{n,\mathrm{an}}. Suppose a closed analytic subspace Z⊂DZ\subset D is defined near a KK-rational point zz by rr functions g1,…,grg_1,\ldots,g_r whose relative Jacobian has rank rr at zz. Write xx for its image and R=OX,xR=\mathcal O_{X,x}, S=OZ,zS=\mathcal O_{Z,z}. Then, as complete local R^\widehat R-algebras, S^≃R^[[V1,…,Vn−r]].\widehat S\simeq\widehat R[[V_1,\ldots,V_{n-r}]]. The local homomorphism R→SR\to S is flat. The fibre local ring at zz is regular of dimension n−rn-r.

Proof. The point xx is KK-rational, since its completed residue field embeds into H(z)=K\mathcal H(z)=K over KK. By Lemma 2.35.9, the domain DD is an ordinary neighborhood of zz in the relative affine space. Choose local affinoid coordinates X0=M(K{s−1U}/J)X_0=\mathcal M(K\{s^{-1}U\}/J) at xx and translate x,zx,z to zero. Proposition 2.35.10, applied also to the relative polydisc, gives R^=K[[U]]/J^,S^=K[[U,T]]/(J^,g^1,…,g^r).\widehat R=K[[U]]/\widehat J,\qquad \widehat S=K[[U,T]]/(\widehat J,\widehat g_1,\ldots,\widehat g_r). Here each gig_i is a germ modulo JJ; choose any convergent representative in (U,T)(U,T). The relative derivatives at the origin are unaffected by this choice, because the added ideal consists of functions of UU times other germs, and all elements of JJ vanish at U=0U=0. After reordering the TiT_i, the first rr columns give an invertible minor. The substitution in the formal ring which fixes all UiU_i, sends TiT_i to g^i\widehat g_i for i≤ri\le r, and fixes TiT_i for i>ri>r has invertible linear part. It is a formal automorphism. Explicitly, invert its linear matrix; then correct an inverse modulo degree qq by solving the degree-qq discrepancy with this same invertible matrix. The corrections have increasing degree, so their sum converges in the formal maximal-ideal topology. The same recursion gives uniqueness, which makes the resulting left and right inverses equal. This argument works in every characteristic.

This automorphism fixes K[[U]]K[[U]] and hence its extended ideal J^\widehat J. It carries the ideal generated by T1,…,TrT_1,\ldots,T_r to that generated by g^1,…,g^r\widehat g_1,\ldots,\widehat g_r. Passing to the quotient gives the asserted R^\widehat R-algebra isomorphism.

Both RR and SS are Noetherian by Theorem 2.35.11; their completions are Noetherian by [StacksAlg] (Tag 0316). The ring R^[[V]]\widehat R[[V]] is the (V)(V)-adic completion of the Noetherian polynomial ring R^[V]\widehat R[V]. It is flat over R^\widehat R by [StacksAlg] (Tag 00MB) and the freeness of a polynomial ring over its coefficient ring. Also R^\widehat R is flat over RR, and S^\widehat S is faithfully flat over SS, by [StacksAlg] (Tag 00MC). For each finitely generated ideal I⊂RI\subset R, the kernel of I⊗RS→SI\otimes_R S\to S becomes zero after tensoring with S^\widehat S, because S^\widehat S is RR-flat. Faithfulness makes the original kernel zero; [StacksAlg] (Tag 00HD) proves the required flatness.

Finally Lemma 2.35.12 and exact completion identify the completed fibre local ring with K[[V1,…,Vn−r]]K[[V_1,\ldots,V_{n-r}]]. This is a regular local ring of dimension n−rn-r, by [StacksAlg] (Tag 032C). Regularity passes to the original local ring by [StacksAlg] (Tag 07NY), and its embedding dimension remains n−rn-r because completion preserves the quotient of the maximal ideal by its square. For a regular local ring dimension equals embedding dimension, proving the last assertion. ◻

Theorem 2.35.14. NAG-001751 (Quasi-smoothness, flatness and geometric regularity). For a morphism f:Y→Xf:Y\to X of analytic spaces, quasi-smoothness at yy is equivalent to analytic flatness at yy together with geometric regularity at yy of the H(f(y))\mathcal H(f(y))-analytic fibre. For good Y,XY,X, whenever these conditions hold one may choose the Jacobian presentation with its source an ordinary affinoid neighborhood of yy. More precisely, after fixing any relative affinoid presentation near yy, one can restrict it to an ordinary rational neighborhood of yy in its ambient polydisc so that the relation ideal is generated by n−rk⁡yΩY/Xn-\operatorname{rk}_y\Omega_{Y/X} elements with independent relative differentials.

Proof. A Jacobian chart is analytically flat. Start with a presentation W↪D⊂AXn,anW\hookrightarrow D\subset\mathbb A_X^{n,\mathrm{an}} whose ideal is generated by rr functions with independent relative differentials at yy. First treat the good chart itself, without assuming that WW is an ordinary neighborhood in any larger source. After any good analytic base change over any complete extension, the same equations and an invertible Jacobian minor give a presentation at each point over yy, by Proposition 2.41.4 and the actual domain base-change theorem. At such a point vv, extend the ground field to a nontrivially valued complete field Ω\Omega containing H(v)\mathcal H(v). Evaluation gives an Ω\Omega-rational lift ww. The base point is then Ω\Omega-rational as well. Proposition 2.35.13 proves naive flatness at ww. Both scalar-stalk maps are faithfully flat by Theorem 2.35.4, so Lemma 2.35.6 descends naive flatness to vv. All the tests in Definition 2.9.1 have now been verified.

Its fibres are geometrically regular. Let FF be any complete extension of H(f(y))\mathcal H(f(y)), and take any point vv of the extended fibre above yy. The pulled-back chart is still a Jacobian presentation over FF. Choose a nontrivially valued complete Ω⊃H(v)\Omega\supset\mathcal H(v) and its rational lift. The completed local ring at that lift is Ω[[T1,…,Tn−r]]\Omega[[T_1,\ldots,T_{n-r}]] by Proposition 2.35.13. Thus its local ring is regular. Its faithfully flat local map from the original fibre stalk has Noetherian source and target, by Theorem 2.35.11. Regularity descends by [StacksAlg] (Tag 00OF). This proves geometric regularity with every complete-extension test.

If the original quasi-smooth presentation uses only an affinoid domain W⊂YW\subset Y containing yy, analytic flatness descends from WW to the good source YY by Corollary 2.47.3. On each complete-extension fibre, the inverse image of WW is a domain containing every point over yy. Its ordinary local map from the ambient fibre is faithfully flat by Theorem 2.47.1. Applying [StacksAlg] (Tag 00OF) again descends regularity from this domain to that fibre. Thus the preceding conclusions hold for the original source, not only for the chosen presentation.

Recovering a Jacobian presentation. Now suppose Y,XY,X good, ff analytically flat at yy, and its analytic fibre geometrically regular at yy. Choose ordinary affinoid neighborhoods Y0=M(B)Y_0=\mathcal M(B), X0=M(A)X_0=\mathcal M(A) with f(Y0)⊂X0f(Y_0)\subset X_0. Lemma 2.33.7 gives B=P/IB=P/I, where P=A{s−1T1,…,sn−1Tn}P=A\{s^{-1}T_1,\ldots,s_n^{-1}T_n\}. Let rr be the rank at yy of the relative differentials of II, so that r=n−rk⁡yΩB/Ar=n-\operatorname{rk}_y\Omega_{B/A} by Proposition 2.41.6. Choose g1,…,gr∈Ig_1,\ldots,g_r\in I with independent such differentials, put J=(g1,…,gr)J=(g_1,\ldots,g_r), and let Z=M(P/J)Z=\mathcal M(P/J). The image point, still denoted yy, has the same completed residue field. An invertible Jacobian minor makes Z→X0Z\to X_0 quasi-smooth at yy, so the forward part already proved applies to it.

Choose a nontrivially valued complete Ω⊃H(y)\Omega\supset\mathcal H(y), with rational lifts u∈Y0,Ω⊂ZΩu\in Y_{0,\Omega}\subset Z_\Omega and v∈X0,Ωv\in X_{0,\Omega}. Put R=OX0,Ω,v,S=OZΩ,u,T=OY0,Ω,u,N=ker⁡(S⟶T).R=\mathcal O_{X_{0,\Omega},v},\quad S=\mathcal O_{Z_\Omega,u},\quad T=\mathcal O_{Y_{0,\Omega},u},\quad N=\ker(S\longrightarrow T). These rings are Noetherian, and NN is a finite SS-module. The RR-module TT is flat, by the assumed analytic flatness. The two fibre local rings S/mRS,T/mRTS/\mathfrak m_R S,\qquad T/\mathfrak m_R T are regular. For the first this follows from Proposition 2.35.13; for the second it is the assumed geometric regularity, using the identification of this fibre with the scalar extension of Yf(y)Y_{f(y)} by associativity of completed base change. Both have embedding dimension n−rn-r, by Lemma 2.35.12 and invariance of the Jacobian rank under the field embedding. Their dimensions therefore both equal n−rn-r. The surjection between them has zero kernel by [StacksAlg] (Tag 00NR): an ideal with regular quotient of the same dimension is generated by zero members of a regular parameter system.

Let RR be a commutative ring and let 0→A→B→C→00\to A\to B\to C\to 0 be an exact sequence of RR-modules with CC flat. For every RR-module XX, tensoring the sequence with XX is exact. To check injectivity on the left, choose a free module FF mapping onto XX, with kernel GG. Lift an element of X⊗RAX\otimes_R A that dies in X⊗RBX\otimes_R B to v∈F⊗RAv\in F\otimes_R A. Its image in F⊗RBF\otimes_R B comes from some w∈G⊗RBw\in G\otimes_R B. The image of ww in G⊗RCG\otimes_R C is zero because it is zero in F⊗RCF\otimes_R C and CC is flat. Right exactness lifts ww to G⊗RAG\otimes_R A. Subtract the image of this lift from vv. The difference vanishes in F⊗RBF\otimes_R B, hence in F⊗RAF\otimes_R A since FF is free. Thus vv comes from G⊗RAG\otimes_R A, and the original element of X⊗RAX\otimes_R A is zero.

If AA is also flat, then BB is flat. Given an injection X↪YX\hookrightarrow Y, tensor the short exact sequence with XX and with YY, using the first assertion for the left injections. An element of X⊗RBX\otimes_R B that dies in Y⊗RBY\otimes_R B maps to zero in X⊗RCX\otimes_R C, by flatness of CC. It comes from X⊗RAX\otimes_R A. The image of this preimage in Y⊗RAY\otimes_R A is zero by the left injection in the row for YY; flatness of AA kills the preimage. Hence the induced map on tensoring with BB is injective for every injection, which is flatness of BB.

Apply the first assertion to 0→N→S→T→00\to N\to S\to T\to 0 and X=R/mRX=R/\mathfrak m_R. The quotient TT is flat over RR, so N/mRNN/\mathfrak m_RN injects into the zero fibre kernel just computed. Hence N=mRNN=\mathfrak m_RN. Because mRS⊂mS\mathfrak m_RS\subset\mathfrak m_S and NN is finite, N=0N=0. Explicitly, writing finite generators in terms of themselves with coefficients in mRS\mathfrak m_RS gives a matrix CC with entries in mS\mathfrak m_S; the adjugate identity for 1−C1-C and the unit det⁡(1−C)\det(1-C) annihilate every generator.

The coherent ideal I/JI/J on ZZ thus has zero scalar-pullback stalk at uu. Theorem 2.35.4 identifies that stalk with its tensor product over OZ,y\mathcal O_{Z,y} with the faithfully flat OZΩ,u\mathcal O_{Z_\Omega,u}. Consequently (I/J)y=0(I/J)_y=0. Its finite generators vanish on an ordinary neighborhood NyN_y in ZZ. Since ZZ is closed in the ambient polydisc, choose an ordinary open neighborhood of yy in that polydisc whose intersection with ZZ is contained in NyN_y, and then a rational affinoid neighborhood D′D' inside it. On D′D', the two closed subspaces Y0Y_0 and ZZ coincide, with ideal generated by g1,…,grg_1,\ldots,g_r. Their intersection with D′D' is an ordinary affinoid neighborhood of yy in YY, and the minor stays nonzero at yy. This is the asserted strengthened Jacobian presentation.

For general analytic spaces, choose affinoid domains in source and target and use the GG-local definition of analytic flatness. Quasi-smoothness is invariant under these restrictions by Theorem 2.41.8. Completed residue fields on domains agree, so the local analytic fibres are precisely analytic domains in the original fibre. In the forward direction the chart argument above proves regularity on all such domains after all complete extensions. In the reverse direction geometric regularity of the fibre supplies regularity on its affinoid charts, while analytic flatness supplies the good-space tests there. The good case therefore applies on every required chart and proves the general equivalence. ◻

References. The equivalence and the strengthened presentation are the classical assertions of [Duc18] (5.3.2–5.3.4). The proof here organizes their analytic content through rational lifts, explicit completed local rings and faithfully flat descent; the commutative-algebra leaves are the checked Stacks statements displayed above.

Finite direct images and overconvergent restriction

Proposition 2.35.15. NAG-001752 (Coherent direct image under a finite morphism). Let π:Y→T\pi:Y\to T be a norm-admissible finite analytic morphism and let F\mathcal F be coherent on YGY_G. Then π∗F\pi_*\mathcal F is coherent on TGT_G, and formation of this direct image is exact and commutes with every analytic base change, including base change over an arbitrary complete extension of the ground field.

Proof. For an affinoid domain V=M(A)⊂TV=\mathcal M(A)\subset T, write π−1(V)=M(B)\pi^{-1}(V)=\mathcal M(B), where BB is a finite AA-module with its canonical norm. By Theorem 2.39.6, the restriction of F\mathcal F is represented by a finite BB-module MM. It is also finite over AA. For every affinoid domain V′⊂VV'\subset V, finite base change and the finite-module description of coherent restriction identify its inverse-image sections with M⊗AAV′M\otimes_A A_{V'}. Thus the direct image on VGV_G is the coherent module sheaf associated to MM over AA. These identifications agree with the natural restriction maps and hence glue on the full analytic site. Exactness follows on each VV from affinoid coherent acyclicity, or equivalently from exactness of the finite-module correspondence and restriction of scalars.

Now take an affinoid piece M(C)\mathcal M(C) of any analytic base change, with CC an LL-affinoid algebra and L/KL/K complete. Proposition 2.33.19 identifies the new finite algebra with D=B⊗ACD=B\otimes_A C, with the canonical finite CC-module norm; the ordinary tensor product already gives its completed analytic pushout. The pulled-back coherent module on this finite space is M⊗BD=M⊗ACM\otimes_BD=M\otimes_A C, by the finite-module pullback construction. This is precisely the module representing the pullback of π∗F\pi_*\mathcal F. These canonical identifications commute with restrictions and with the original base-change map, so they prove that map is an isomorphism globally. ◻

Lemma 2.35.16. NAG-001753 (A flat coherent ordinary stalk is locally free). Let TT be good and G\mathcal G coherent on TGT_G. If Gt\mathcal G_t is flat over OT,t\mathcal O_{T,t}, then G\mathcal G is free of finite rank on an ordinary neighborhood of tt. In particular it is analytically TT-flat at tt.

Proof. Put R=OT,tR=\mathcal O_{T,t} and M=GtM=\mathcal G_t. The ring RR is Noetherian by Theorem 2.35.11, and MM is finite. Lift a basis of M/mRMM/\mathfrak m_RM to mm elements of MM. Nakayama’s determinant argument makes the induced map Rm→MR^m\to M surjective. Its kernel NN is finite. Tensoring with the residue field is left exact at NN, since flatness of MM kills the first Tor group; the chosen residue basis therefore gives N/mRN=0N/\mathfrak m_RN=0. Another application of the determinant argument gives N=0N=0. Hence MM is free.

Represent the finitely many basis germs by local sections, giving a morphism Om→G\mathcal O^m\to\mathcal G on an ordinary neighborhood. Its kernel and cokernel are coherent by the local finite-module theorem. Each has zero ordinary stalk at tt, and finite generators on an affinoid neighborhood therefore vanish after shrinking to an ordinary neighborhood of tt. The morphism is an isomorphism there. Any analytic pullback of this isomorphism is again a finite-rank free module, whose ordinary stalks are flat. This checks all complete-extension and good-base-change tests. ◻

Theorem 2.35.17. NAG-001754 (Finite pushforward detects flatness). Let Y→πT→XY\xrightarrow{\pi}T\to X be analytic morphisms with π\pi finite, let t∈Tt\in T, and write π−1(t)={y1,…,ym}\pi^{-1}(t)=\{y_1,\ldots,y_m\}. For a coherent sheaf F\mathcal F on YGY_G, if the spaces are good there is a canonical decomposition (π∗F)t≃⨁i=1mFyi.(\pi_*\mathcal F)_t \simeq \bigoplus_{i=1}^{m}\mathcal F_{y_i}. Consequently π∗F\pi_*\mathcal F is naively XX-flat at tt if and only if F\mathcal F is naively XX-flat at every yiy_i. The analogous equivalence for analytic XX-flatness holds for arbitrary analytic spaces.

Proof. For good spaces, use Lemma 2.35.1. Representatives of finitely many section germs at the yiy_i may be restricted to its disjoint open-and-closed branches above a smaller ordinary neighborhood of tt; they glue to a section of F\mathcal F on that full inverse image. If all germs of such a section vanish, shrinking the same way makes it vanish on a full inverse-image neighborhood. This proves the displayed isomorphism for module sheaves just as for structure sheaves. It respects the action of OX,f(t)\mathcal O_{X,f(t)}. A finite direct sum of modules is flat exactly when every summand is flat: tensor an injection and inspect its component kernels. This gives the naive equivalence.

Take any good analytic base change X′→XX'\to X over any complete extension. The map π′:Y′→T′\pi':Y'\to T' is still finite, and Proposition 2.35.15 identifies (π∗F)T′(\pi_*\mathcal F)_{T'} with π∗′(FY′)\pi'_*(\mathcal F_{Y'}). At any t′t' above tt, the naive equivalence already proved applies to its finite fibre. Every point of that fibre lies over one of the yiy_i, and every point above a yiy_i is in the fibre of its own image t′t'. Thus the complete collection of naive-flatness tests on the two sides is the same. This proves the analytic equivalence in the good case. For general spaces, restrict to affinoid domains in X,TX,T and their finite inverse-image domains in YY. The general GG-local definition and the canonical base-change isomorphism give the same equivalence. ◻

Corollary 2.35.18. NAG-001755 (Naive flatness for a finite morphism). For a finite morphism Y→XY\to X of good spaces and a coherent sheaf F\mathcal F on YGY_G, naive XX-flatness at a point yy is equivalent to analytic XX-flatness there.

Proof. The analytic assertion always implies the naive one. For the converse, separate the finite branches over x=f(y)x=f(y) by Lemma 2.35.1, and retain the branch containing yy. It is an ordinary neighborhood of yy, finite over an ordinary neighborhood UU of xx, and its fibre over xx consists only of yy. Denote its finite map by π\pi. Theorem 2.35.17 makes (π∗F)x(\pi_*\mathcal F)_x flat over OU,x\mathcal O_{U,x}. Lemma 2.35.16 makes the direct image locally free near xx, hence analytically UU-flat. The analytic equivalence in Theorem 2.35.17 gives analytic flatness at yy. Ordinary open restrictions do not change any of the pointwise tests, so this is the assertion for the original morphism. ◻

Proposition 2.35.19. NAG-001756 (The overconvergent restriction interface). Let Y→XY\to X and Y′→XY'\to X be morphisms of good spaces. Suppose an ordinary analytic neighborhood WW of yy in YY is identified over XX with a good analytic domain in Y′Y', and suppose coherent sheaves F\mathcal F on YY and G\mathcal G on Y′Y' have identified restrictions to WW. Then F\mathcal F is naively XX-flat at yy if and only if G\mathcal G is; the same equivalence holds for analytic XX-flatness. Both statements are unchanged by restricting XX to an ordinary neighborhood of the image point and the sources to its inverse images.

Proof. The ordinary restriction from YY to WW preserves the stalk at yy. After any good analytic base change, its inverse image is again an ordinary neighborhood of every point over yy, so it preserves all analytic-flatness tests too. On the other side, Corollary 2.47.3 gives the equivalence of both types of flatness for G\mathcal G and its restriction to the source domain WW. Combine these two facts with the given isomorphism. A target ordinary open restriction changes neither the target stalk nor the source stalk; every base-change test at the point factors locally through that target open set. Hence it also preserves both assertions. ◻

References. These arguments reconstruct the finite-pushforward step and the restriction mechanism used in [Duc18] (4.1.14–4.1.15 and 8.3.1–8.3.4). The boundaryless inner-flatness theorem is a separate input to the final overconvergent application; the proposition above is the exact transfer needed once that theorem is established.

Finite algebraic tests realized on analytic germs

Theorem 2.35.20. NAG-001757 (Realizing a finite ordinary-stalk algebra). Let XX be good, let x∈Xx\in X, and put R=OX,xR=\mathcal O_{X,x}. For every finite RR-algebra CC there are an ordinary affinoid neighborhood V=M(A)V=\mathcal M(A) of xx and a finite AA-algebra BB, with its canonical affinoid norm, such that C≃B⊗AR.C\simeq B\otimes_A R. Writing Z=M(B)Z=\mathcal M(B) and π−1(x)={z1,…,zm}\pi^{-1}(x)=\{z_1,\ldots,z_m\} for the resulting finite map, one has C≃∏iOZ,ziC\simeq\prod_i\mathcal O_{Z,z_i}. If CC is a nonzero domain, one can choose BB to be a domain; then m=1m=1 and OZ,z1≃C\mathcal O_{Z,z_1}\simeq C.

Proof. The ring RR is Noetherian by Theorem 2.35.11. Choose finitely many algebra generators cjc_j of CC; every cjc_j is integral over RR, by the determinant argument on a finite generating set of the RR-module CC containing 11. Thus the kernel of R[T1,…,Ts]→CR[T_1,\ldots,T_s]\to C contains a monic polynomial in each TjT_j. It is finitely generated, because the polynomial ring is Noetherian by [StacksAlg] (Tag 00FN). Adjoin these monic polynomials to a finite generating set of the kernel.

All coefficients of this finite list of relations are germs represented on a common ordinary affinoid neighborhood VV of xx, by Proposition 2.33.16. Write A=O(V)A=\mathcal O(V) and take the quotient BB of A[T1,…,Ts]A[T_1,\ldots,T_s] by these representatives. The monic relations show that bounded powers of the TjT_j span BB over AA, so BB is finite. Tensoring its presentation with RR recovers exactly the chosen presentation of CC. Theorem 2.33.18 gives BB its actual finite affinoid structure. Lemma 2.35.1 then identifies B⊗ARB\otimes_A R with the displayed product of ordinary local rings.

Suppose CC is a nonzero domain. Let II be the kernel of B→B⊗AR=CB\to B\otimes_A R=C. Since CC is a domain, B/IB/I is a domain. It is still finite over AA, and its tensor product with RR is CC: right exactness gives the quotient of CC by the image of I⊗ARI\otimes_A R, and that image is zero by the definition of II. Replace BB by B/IB/I. The resulting product of nonzero local rings is the domain CC, so exactly one factor occurs. This gives the last assertion, without normalization or an excellence assumption. ◻

Corollary 2.35.21. NAG-001758 (Henselian ordinary local rings). Every ordinary local ring RR of a good analytic space has the following properties: every finite RR-algebra is a finite product of local rings, and every simple residue-field root of a monic polynomial over RR lifts uniquely to a root in RR. In particular RR is henselian.

Proof. The first assertion is Theorem 2.35.20. To prove the second directly, let P∈R[T]P\in R[T] be monic and let a∈R/mRa\in R/\mathfrak m_R be a simple root of P‾\overline P. Apply the first assertion to the finite free algebra B=R[T]/(P)B=R[T]/(P). Every local factor is finite over RR and hence has its maximal ideal over mR\mathfrak m_R: a field integral over the image of RR forces that image to be a field. The factor CC corresponding to the maximal ideal (mR,T−a)(\mathfrak m_R,T-a) therefore satisfies C/mRC≃R/mRC/\mathfrak m_RC\simeq R/\mathfrak m_R. Here simplicity of the root ensures that the associated local factor of (R/mR)[T]/(P‾)(R/\mathfrak m_R)[T]/(\overline P) has dimension one and no nilpotents; factor P‾=(T−a)Q\overline P=(T-a)Q with Q(a)≠0Q(a)\ne 0 and use the Chinese remainder identity.

Let ϵ:R→C\epsilon:R\to C be the structure map and put k=R/mRk=R/\mathfrak m_R. Its cokernel QCQ_C is finite. The isomorphism k→C/mRCk\to C/\mathfrak m_RC gives QC/mRQC=0Q_C/\mathfrak m_RQ_C=0 by right exactness. Apply [StacksAlg] (Tag 00DV(2)) with M=QCM=Q_C and I=mRI=\mathfrak m_R to see that ϵ\epsilon is surjective.

Let pC:B→Cp_C:B\to C be the projection. Lift the images under pCp_C of the standard free RR-basis of BB along ϵ\epsilon. These lifts define an RR-linear map B→RB\to R. Its restriction to the direct summand CC splits ϵ\epsilon as a map of RR-modules. Thus N=ker⁡ϵN=\ker\epsilon is a direct summand of RR, hence finite, and reduction of this split sequence identifies N/mRNN/\mathfrak m_RN with the kernel of k→C/mRCk\to C/\mathfrak m_RC. That kernel is zero. Apply [StacksAlg] (Tag 00DV(2)) with M=NM=N and I=mRI=\mathfrak m_R to get N=0N=0. The image of TT in C=RC=R is the desired lifted root. If b,c∈Rb,c\in R were two lifts, then P(b)−P(c)=(b−c)Q1(b,c)P(b)-P(c)=(b-c)Q_1(b,c), where Q1(b,c)Q_1(b,c) reduces to P‾′(a)≠0\overline P'(a)\ne 0; hence Q1(b,c)Q_1(b,c) is a unit and b=cb=c. ◻

Proposition 2.35.22. NAG-001759 (Realization of prime and finite-field fibre tests). Let f:Y→Xf:Y\to X be a morphism of good spaces, let x=f(y)x=f(y), and put R=OX,xR=\mathcal O_{X,x}, S=OY,yS=\mathcal O_{Y,y}. For a prime p⊂R\mathfrak p\subset R and a finite extension E/Frac⁡(R/p)E/\operatorname{Frac}(R/\mathfrak p), there are, after ordinary neighborhood restrictions, an integral finite analytic space Z→XZ\to X with a unique point zz above xx, and finitely many points t1,…,tmt_1,\ldots,t_m above yy in T=Y×XZT=Y\times_X Z, such that C:=OZ,z is a domain,Frac⁡(C)=E,ker⁡(R→C)=p,C:=\mathcal O_{Z,z}\ \text{is a domain},\qquad \operatorname{Frac}(C)=E,\qquad \ker(R\to C)=\mathfrak p, S⊗RE≃(∏i=1mOT,ti)⊗CE.S\otimes_R E \simeq \left(\prod_{i=1}^{m}\mathcal O_{T,t_i}\right) \otimes_C E . Every local ring of Spec⁡(S⊗RE)\operatorname{Spec}(S\otimes_R E) is therefore a localization of some OT,ti\mathcal O_{T,t_i} at a prime whose contraction to the affinoid domain algebra of ZZ is zero.

Proof. Write R0=R/pR_0=R/\mathfrak p and F=Frac⁡(R0)F=\operatorname{Frac}(R_0). Choose finitely many generators α1,…,αs\alpha_1,\ldots,\alpha_s of the finite field extension E/FE/F. Each satisfies a monic equation over FF. For each jj, choose a nonzero dj∈R0d_j\in R_0 clearing the finitely many denominators in that equation. Then βj=djαj\beta_j=d_j\alpha_j is integral over R0R_0: in the transformed monic equation, its iith coefficient is the old iith coefficient multiplied by djid_j^i, which lies in R0R_0. Thus C=R0[β1,…,βs]C=R_0[\beta_1,\ldots,\beta_s] is a finite integral domain over R0R_0, and its fraction field is exactly EE. This construction uses no integral-closure finiteness theorem.

Apply Theorem 2.35.20 to CC, obtaining an ordinary affinoid neighborhood X0=M(A)X_0=\mathcal M(A) and a finite domain BB over AA, with Z=M(B)Z=\mathcal M(B), a unique point zz above xx, and B⊗AR=C=OZ,zB\otimes_A R=C=\mathcal O_{Z,z}. The construction there may and does arrange B↪CB\hookrightarrow C, by quotienting the kernel of this map. Choose an affinoid neighborhood Y0Y_0 of yy mapping to X0X_0; this changes neither RR nor SS. The projection T=Y0×X0Z→Y0T=Y_0\times_{X_0}Z\to Y_0 is finite. Lemma 2.35.1 and finite algebra base change give ∏iOT,ti≃B⊗AS≃(B⊗AR)⊗RS=C⊗RS.\prod_i\mathcal O_{T,t_i} \simeq B\otimes_A S \simeq (B\otimes_A R)\otimes_R S =C\otimes_R S . Localize this identity by C∖{0}C\setminus\{0\}. Since Frac⁡(C)=E\operatorname{Frac}(C)=E, it becomes the asserted isomorphism. Every prime of a finite product belongs to exactly one factor: the complementary factor idempotents have product zero and sum one. A prime in the localization avoids every nonzero element of CC, and hence every nonzero element of its subring BB. Localizing once more at that prime gives precisely the local ring of the original factor at its corresponding prime. This proves the last assertion and explicitly realizes every finite-field test in an algebraic fibre of the ordinary-stalk morphism. ◻

References. Compare [Duc18] (2.1.6 and 2.3.13–2.3.14). Clearing finitely many monic-equation denominators replaces the normalization input in the finite-field test. The realization theorem is a reduction of tests, not a proof that those local rings are regular.

Regularity of domains in affine space

Theorem 2.35.23. NAG-001760 (Regular affine-space domains). Let KK be any complete non-Archimedean field, and let V=M(A)V=\mathcal M(A) be an affinoid domain in AKn,an\mathbb A_K^{n,\mathrm{an}}. Every ordinary local ring OV,v\mathcal O_{V,v} is regular, and AA is a regular Noetherian ring. The same assertions hold after every complete valued extension of KK.

Proof. Fix v∈Vv\in V. Choose a nontrivially valued complete extension LL of H(v)\mathcal H(v); if necessary adjoin a free Gauss radius. The evaluated coordinates give an LL-rational point ww of VLV_L above vv. Indeed, the bounded map A→H(v)→LA\to\mathcal H(v)\to L and the identity of LL induce a bounded map A⊗^KL→LA\widehat\otimes_K L\to L. By analytic domain base change, VLV_L is an affinoid domain of ALn,an\mathbb A_L^{n,\mathrm{an}}. Lemma 2.35.9 makes VLV_L an ordinary neighborhood of ww in that affine space. Consequently OVL,w=OALn,an,w,OVL,w^≃L[[T1,…,Tn]].\mathcal O_{V_L,w}=\mathcal O_{\mathbb A_L^{n,\mathrm{an}},w}, \qquad \widehat{\mathcal O_{V_L,w}}\simeq L[[T_1,\ldots,T_n]]. The completion formula is Proposition 2.35.10, with the coordinates translated by their values at ww. The convergent-germ theorem makes the local ring Noetherian. Its completion is regular by [StacksAlg] (Tag 032C); the regularity/completion criterion [StacksAlg] (Tag 07NY) therefore makes OVL,w\mathcal O_{V_L,w} regular. Theorem 2.35.4 makes OV,v→OVL,w\mathcal O_{V,v}\to\mathcal O_{V_L,w} a faithfully flat local homomorphism. Both rings are Noetherian by Theorem 2.35.11, so regularity descends by [StacksAlg] (Tag 00OF).

To pass from ordinary local rings to AA, take a maximal ideal m\mathfrak m of AA. It is closed by the finite-module theorem; the nonzero affinoid quotient A/mA/\mathfrak m has a point by Theorem 1.7.1. Its point kernel is zero, since this quotient is a field. It therefore gives v∈Vv\in V with point kernel exactly m\mathfrak m. The map A→OV,vA\to\mathcal O_{V,v} is flat: the latter is the filtered colimit of the rational-neighborhood algebras of Proposition 2.33.16, each of which is flat over AA by Proposition 2.38.2; the finite-ideal criterion commutes with this filtered colimit. Evaluation now gives a local flat map Am→OV,vA_{\mathfrak m}\to\mathcal O_{V,v}. Hence AmA_{\mathfrak m} is regular by the same descent lemma. Every ApA_{\mathfrak p} is a localization of some such AmA_{\mathfrak m} and is regular by [StacksAlg] (Tag 0AFS). Thus AA is regular.

Finally, affinoid domain base change identifies VFV_F with an affinoid domain of AFn,an\mathbb A_F^{n,\mathrm{an}} for every complete F/KF/K. Apply the argument over FF. ◻

References. The proof reconstructs the rational-lift argument of A. Ducros, Les espaces de Berkovich sont excellents, Lemma 2.1, using the ordinary-stalk and completion results proved above.

NAG-001761 · Open at full scope

Proposition 2.35.24. NAG-001761 (A uniform separable regular locus). Let DD be the affinoid algebra of an affinoid domain in AKn,an\mathbb A_K^{n,\mathrm{an}}, assume that DD is a domain, and let D↪AD\hookrightarrow A be finite with AA a domain. Suppose Frac⁡(A)/Frac⁡(D)\operatorname{Frac}(A)/\operatorname{Frac}(D) is separable. There is 0≠d∈D0\ne d\in D such that, for every complete L/KL/K, writing AL=A⊗^KLA_L=A\widehat\otimes_KL and XL=M(AL)X_L=\mathcal M(A_L), the following rings are regular: AL[1/d],OXL,y[1/d],OXL,y^[1/d](y∈XL).A_L[1/d],\qquad \mathcal O_{X_L,y}[1/d],\qquad \widehat{\mathcal O_{X_L,y}}[1/d]\quad(y\in X_L). Here a hat denotes completion at the maximal ideal of the ordinary local ring. In particular this conclusion holds for every strict integral KK-affinoid algebra AA when char⁡K=0\operatorname{char}K=0, by choosing a finite Tate normalization.

Proof. Put F=Frac⁡(D)F=\operatorname{Frac}(D) and E=Frac⁡(A)E=\operatorname{Frac}(A). The finite-dimensional domain A⊗DFA\otimes_DF is a field, and is therefore EE. The primitive-element theorem [StacksAlg] (Tag 030N) gives E=F(θ)E=F(\theta). After multiplying θ\theta by a nonzero element of DD, we may take θ∈A\theta\in A. Let P∈F[T]P\in F[T] be its monic minimal polynomial. Choose finitely many DD-module generators of AA, and express each as an FF-linear combination of 1,θ,…,θm−11,\theta,\ldots,\theta^{m-1}, where m=deg⁡Pm=\deg P. Separability gives P′(θ)≠0P'(\theta)\ne 0; express its inverse in the same basis. Choose 0≠d∈D0\ne d\in D clearing the coefficients of PP and all these finitely many expressions. Then A[1/d]=D[1/d][θ]≃D[1/d][T]/(P),P′(θ)∈A[1/d]×.A[1/d]=D[1/d][\theta] \simeq D[1/d][T]/(P), \qquad P'(\theta)\in A[1/d]^\times . The displayed presentation is exact: division by the monic polynomial reduces every relation to degree less than mm, and 1,θ,…,θm−11,\theta,\ldots,\theta^{m-1} are FF-linearly independent. Thus D[1/d]→A[1/d]D[1/d]\to A[1/d] is finite standard étale. This uses the standard-étale verification after Definition 59.26.3 in [StacksAlg] (Tag 03PA). Its base changes and product factors are étale by [StacksAlg] (Tag 00U2, parts (3) and (11)).

Fix L/KL/K and put DL=D⊗^KLD_L=D\widehat\otimes_KL. Finite norm-admissible base change, Proposition 2.33.19, identifies ALA_L with A⊗DDLA\otimes_DD_L. Theorem 2.35.23 makes DLD_L regular. Hence AL[1/d]A_L[1/d] is regular: at every prime it is étale over a localization of the regular ring DL[1/d]D_L[1/d], and [StacksAlg] (Tag 025N) preserves regularity under such a local étale map.

Let t∈M(DL)t\in\mathcal M(D_L), put R=OM(DL),tR=\mathcal O_{\mathcal M(D_L),t}, and list the finitely many points yiy_i above tt. Finite branch separation gives an isomorphism of finite RR-algebras AL⊗DLR≃∏iSi,Si=OXL,yi.A_L\otimes_{D_L}R \simeq \prod_i S_i,\qquad S_i=\mathcal O_{X_L,y_i}. The ring RR is regular by Theorem 2.35.23. Localizing the displayed product at dd makes each factor finite étale over R[1/d]R[1/d]. This proves regularity of all Si[1/d]S_i[1/d] by the same local argument.

For completeness, the finite-module completion identity [StacksAlg] (Tag 00MA, part (3)) identifies the mR\mathfrak m_R-adic completion of the product with AL⊗DLR^A_L\otimes_{D_L}\widehat R. Each SiS_i is a finite local RR-algebra, so Si/mRSiS_i/\mathfrak m_RS_i is an Artinian local algebra over the residue field of RR. Its maximal ideal is nilpotent: the descending sequence of its powers stabilizes in a finite-dimensional vector space, and Nakayama makes the stable power zero. Thus mSiNi⊂mRSi⊂mSi\mathfrak m_{S_i}^{N_i}\subset\mathfrak m_RS_i\subset\mathfrak m_{S_i} for some NiN_i. The two adic filtrations are cofinal, giving AL⊗DLR^≃∏iSi^.A_L\otimes_{D_L}\widehat R \simeq\prod_i\widehat{S_i}. The ring R^\widehat R is regular by [StacksAlg] (Tag 07NY). After inverting dd, each factor on the right is finite étale over R^[1/d]\widehat R[1/d] and hence regular. This proves the completion assertion. Finally, for strict integral AA in characteristic zero, Theorem 1.12.2 supplies D=K{T1,…,Tn}D=K\{T_1,\ldots,T_n\}; the finite extension of fraction fields is separable because a nonconstant irreducible polynomial in characteristic zero has nonzero derivative and is coprime to that derivative. ◻

References. This is the separable finite-normalization part of Ducros, Les espaces de Berkovich sont excellents, Proposition 2.2, with a single denominator chosen before arbitrary scalar extension.

Lemma 2.35.25. NAG-001762 (Strict ordinary neighborhoods). Suppose the valuation of KK is nontrivial and X=M(A)X=\mathcal M(A) is strict. Every point of XX has a basis of ordinary neighborhoods consisting of strict affinoid domains. Consequently, in Theorem 2.35.20, when XX is strict, the neighborhood algebra and the finite algebra may both be chosen strict.

Proof. The topology of the spectrum is generated by evaluation inequalities. A prescribed open neighborhood of xx therefore contains a finite intersection of conditions ∣fi∣<ai|f_i|<a_i and ∣gj∣>bj|g_j|>b_j holding at xx. Since the valuation is nontrivial, ∣K×∣\sqrt{|K^\times|} is dense in R>0\mathbb R_{>0}: the rational powers of any element of absolute value different from one already have logarithms dense in R\mathbb R. Choose ri,sjr_i,s_j in this group with ∣fi(x)∣<ri<ai,bj<sj<∣gj(x)∣.|f_i(x)|<r_i<a_i,\qquad b_j<s_j<|g_j(x)|. The domain defined by ∣fi∣≤ri|f_i|\le r_i and ∣gj∣≥sj|g_j|\ge s_j is an ordinary neighborhood of xx contained in the prescribed open set. Write rimi=∣ci∣r_i^{m_i}=|c_i| and sjnj=∣dj∣s_j^{n_j}=|d_j|, with nonzero scalars ci,dj∈Kc_i,d_j\in K. The inequalities are equivalent to ∣ci−1fimi∣≤1,∣dj−1gjnj∣≥1.|c_i^{-1}f_i^{m_i}|\le1,\qquad |d_j^{-1}g_j^{n_j}|\ge1 . They give a finite succession of unit-radius Weierstrass and Laurent domains, hence a strict affinoid domain by the established presentation and intersection theorems. The finite list of germs in the proof of Theorem 2.35.20 can thus be represented on such a strict neighborhood. Its finite algebra is strict by Proposition 1.14.4; taking the quotient by the kernel used there preserves strictness. ◻

Theorem 2.35.26. NAG-001763 (Ordinary regular comparison in the strict characteristic-zero case). Let KK be nontrivially valued of characteristic zero, let AA be strict KK-affinoid, let x∈X=M(A)x\in X=\mathcal M(A), and put p=ker⁡(x)\mathfrak p=\ker(x) and R=OX,xR=\mathcal O_{X,x}. Both of the following homomorphisms are regular: Ap⟶R,R⟶R^.A_{\mathfrak p}\longrightarrow R, \qquad R\longrightarrow\widehat R . Here regular means flat with geometrically regular algebraic fibres; the fibre fields and tensor products in this assertion are those of ordinary commutative algebra.

Proof. All rings are Noetherian, by the affinoid finite-module theorem, Theorem 2.35.11, and [StacksAlg] (Tag 0316). The map Ap→RA_{\mathfrak p}\to R is flat: RR is the filtered colimit of ordinary rational-neighborhood algebras, all flat over AA, and elements of A∖pA\setminus\mathfrak p are units in RR. The completion map is faithfully flat by [StacksAlg] (Tag 00MC). It remains to check their algebraic fibres.

Let q⊂p\mathfrak q\subset\mathfrak p be any prime of AA, set B=A/qB=A/\mathfrak q, and put F=Frac⁡(B)F=\operatorname{Frac}(B). The point xx lies on Z=M(B)Z=\mathcal M(B), and finite branch separation for this closed immersion gives R/qR=OZ,x,R⊗AF=OZ,x⊗BF.R/\mathfrak qR=\mathcal O_{Z,x}, \qquad R\otimes_A F=\mathcal O_{Z,x}\otimes_B F . The ring BB is strict and integral. Proposition 2.35.24 supplies a nonzero d∈Bd\in B for which OZ,x[1/d]\mathcal O_{Z,x}[1/d] is regular. The displayed fibre is its further localization by the nonzero elements of BB, so it is a regular Noetherian ring. Its ground field FF has characteristic zero. By [StacksAlg] (Tag 0381), a Noetherian algebra over a characteristic-zero field is geometrically regular once it is regular: the only finite purely inseparable field extension to be tested is the identity extension. This proves geometric regularity of every fibre of Ap→RA_{\mathfrak p}\to R, at all primes q⊂p\mathfrak q\subset\mathfrak p.

Now let P\mathfrak P be any prime of RR and put C=R/PC=R/\mathfrak P. By Lemma 2.35.25 and Theorem 2.35.20, there is a strict integral affinoid algebra BB, a point z∈Z=M(B)z\in Z=\mathcal M(B), and an identification B↪OZ,z=C.B\hookrightarrow \mathcal O_{Z,z}=C . The injection is part of the kernel-quotient construction in that theorem. Finite-module completion, [StacksAlg] (Tag 00MA, part (3)), gives R^⊗RC=C^\widehat R\otimes_R C=\widehat C, since the maximal ideal of the quotient CC is the image of the maximal ideal of RR. Consequently the algebraic fibre of R→R^R\to\widehat R at P\mathfrak P is R^⊗Rκ(P)=C^⊗CFrac⁡(C).\widehat R\otimes_R\kappa(\mathfrak P) =\widehat C\otimes_C\operatorname{Frac}(C). Apply Proposition 2.35.24 to BB. It gives 0≠d∈B0\ne d\in B such that OZ,z^[1/d]=C^[1/d]\widehat{\mathcal O_{Z,z}}[1/d]=\widehat C[1/d] is regular. Because B↪CB\hookrightarrow C, the image of dd is nonzero in CC. The last displayed fibre is thus a further localization of this regular ring, and is regular. It is Noetherian and is an algebra over the characteristic-zero field Frac⁡(C)\operatorname{Frac}(C), so the same Tag 0381 makes it geometrically regular. This treats every prime of RR, and proves the second assertion. ◻

References. The generic-fibre organization follows Ducros, Les espaces de Berkovich sont excellents, Proposition 2.9. Here separable normalization supplies the regular locus, and finite ordinary-germ realization handles every prime of the ordinary local ring.

Proposition 2.35.27. NAG-001764 (A uniform characteristic-zero regular open set). Let KK be any complete field of characteristic zero and let AA be an integral generalized KK-affinoid algebra. There exists 0≠a∈A0\ne a\in A such that, for every complete L/KL/K, the ring AL[1/a]A_L[1/a] is regular and every ordinary local ring of M(AL)\mathcal M(A_L) at a point where aa is nonzero is regular.

Proof. Choose a finite free polyradius rr such that F=KrF=K_r is nontrivially valued and ArA_r is strict, by Theorem 1.17.6. The algebra ArA_r is integral. Indeed, Lemma 2.52.1 supplies x∈M(A)x\in\mathcal M(A) with kernel zero. Its Gauss lift has value ∣∑νbνTν∣σr(x)=max⁡ν∣bν(x)∣rν.\left|\sum_\nu b_\nu T^\nu\right|_{\sigma_r(x)} =\max_\nu |b_\nu(x)|r^\nu. This vanishes only when every coefficient is zero. The lift is a multiplicative seminorm by Theorem 1.19.1; its kernel-zero property makes ArA_r a domain. Apply Proposition 2.35.24 over FF, and obtain 0≠d∈Ar0\ne d\in A_r with its conclusions over every complete extension of FF. Write d=∑νaνTνd=\sum_\nu a_\nu T^\nu, and choose one nonzero coefficient a=aν0∈Aa=a_{\nu_0}\in A.

Fix a complete extension L/KL/K and a point y∈M(AL)y\in\mathcal M(A_L) with ∣a(y)∣>0|a(y)|>0. Let EE be the completion of H(y)(T1,…,Ts)\mathcal H(y)(T_1,\ldots,T_s) for the weighted Gauss absolute value of polyradius rr. Lemma 3.21.7 gives compatible isometric embeddings L⊂H(y)⊂EL\subset\mathcal H(y)\subset E and F⊂EF\subset E. Evaluation AL→H(y)⊂EA_L\to\mathcal H(y)\subset E defines an EE-rational lift ww in M(AE)\mathcal M(A_E) above yy. Associativity identifies AE=Ar⊗^FE.A_E=A_r\widehat\otimes_F E . The image of dd at this lift has absolute value max⁡ν∣aν(y)∣rν>0\max_\nu|a_\nu(y)|r^\nu>0. To justify the formula for the infinite expansion, its finite Laurent truncations converge in the Gauss field, since the weighted coefficient values tend to zero; the maximum formula passes to the limit. Thus dd is a unit in the ordinary local ring at ww. The separable regular-locus proposition makes that local ring regular. The local map OM(AL),y⟶OM(AE),w\mathcal O_{\mathcal M(A_L),y} \longrightarrow\mathcal O_{\mathcal M(A_E),w} is faithfully flat by Theorem 2.35.4. Regularity therefore descends by [StacksAlg] (Tag 00OF). This proves the assertion about all ordinary local rings over D(a)D(a), uniformly in LL.

For any prime q\mathfrak q of ALA_L not containing aa, Lemma 2.52.1 gives a point yy with kernel q\mathfrak q. It satisfies ∣a(y)∣>0|a(y)|>0. The map (AL)q→OM(AL),y(A_L)_{\mathfrak q}\to\mathcal O_{\mathcal M(A_L),y} is a faithfully flat local map, by the filtered rational-neighborhood description of the stalk and domain flatness. The target is regular by the preceding paragraph, so the source is regular by Tag 00OF. These are all local rings of AL[1/a]A_L[1/a], proving its regularity. ◻

Proposition 2.35.28. NAG-001765 (Generic regularity of ordinary germs and their completions). Under the hypotheses of Proposition 2.35.27, one may choose 0≠a∈A0\ne a\in A such that, for every x∈X=M(A)x\in X=\mathcal M(A), both of the following rings are regular: OX,x[1/a],OX,x^[1/a].\mathcal O_{X,x}[1/a], \qquad\widehat{\mathcal O_{X,x}}[1/a].

Proof. Take aa from Proposition 2.35.27 and a nontrivial free strictification F=KrF=K_r. Write B=ArB=A_r, take any z∈M(B)z\in\mathcal M(B) above xx, and put R=OX,xR=\mathcal O_{X,x} and S=OM(B),zS=\mathcal O_{\mathcal M(B),z}. Such a zz exists by the Gauss section. The ring B[1/a]B[1/a] is regular. Theorem 2.35.26 says that Bker⁡(z)→SB_{\ker(z)}\to S and S→S^S\to\widehat S are regular homomorphisms.

Let Q\mathfrak Q be a prime of SS avoiding aa, and let q\mathfrak q be its contraction to BB. The local ring BqB_{\mathfrak q} is regular. The map Bq→SQB_{\mathfrak q}\to S_{\mathfrak Q} is flat and has regular closed fibre, since that fibre is a localization of a geometrically regular fibre of Bker⁡(z)→SB_{\ker(z)}\to S. The ascent lemma [StacksAlg] (Tag 031E) makes SQS_{\mathfrak Q} regular. Therefore S[1/a]S[1/a] is regular. Applying exactly the same local argument to the regular map S→S^S\to\widehat S gives regularity of S^[1/a]\widehat S[1/a].

The map R→SR\to S is faithfully flat by Theorem 2.35.4. All local rings are Noetherian, so [StacksAlg] (Tag 0C4G) makes R^→S^\widehat R\to\widehat S flat; it is local, hence faithfully flat by [StacksAlg] (Tag 00HP). Inverting aa preserves faithful flatness for both maps. Every prime of either source then lifts to the corresponding target by [StacksAlg] (Tag 00HQ). Localization yields a flat local map with regular target, and Tag 00OF descends regularity. This proves both assertions. ◻

Theorem 2.35.29. NAG-001766 (Characteristic-zero ordinary regular comparison). Let KK be any complete field of characteristic zero, with possibly trivial valuation. Let AA be any generalized KK-affinoid algebra, let x∈X=M(A)x\in X=\mathcal M(A), and put p=ker⁡(x)\mathfrak p=\ker(x) and R=OX,xR=\mathcal O_{X,x}. The following maps are regular homomorphisms: Ap⟶R,R⟶R^.A_{\mathfrak p}\longrightarrow R, \qquad R\longrightarrow\widehat R.

Proof. Flatness and Noetherianity are proved exactly as at the start of Theorem 2.35.26; those arguments use no strictness or nontriviality of the valuation. Let q⊂p\mathfrak q\subset\mathfrak p, put B=A/qB=A/\mathfrak q, and let F=Frac⁡(B)F=\operatorname{Frac}(B). The finite closed immersion M(B)→X\mathcal M(B)\to X gives R/qR=OM(B),xR/\mathfrak qR=\mathcal O_{\mathcal M(B),x}. The integral generalized algebra BB satisfies Proposition 2.35.28, so this ordinary local ring becomes regular after inverting one nonzero element of BB. Its further localization by all of B∖{0}B\setminus\{0\} is the fibre R⊗AFR\otimes_A F, hence is regular. By [StacksAlg] (Tag 0381) it is geometrically regular over FF. This treats every fibre of the first map.

For a prime P⊂R\mathfrak P\subset R, put C=R/PC=R/\mathfrak P. Theorem 2.35.20 realizes CC as OZ,z\mathcal O_{Z,z} for an integral generalized affinoid Z=M(B)Z=\mathcal M(B), with B↪CB\hookrightarrow C. Proposition 2.35.28 gives 0≠a∈B0\ne a\in B for which C^[1/a]\widehat C[1/a] is regular. Since aa stays nonzero in CC, the fibre R^⊗Rκ(P)=C^⊗CFrac⁡(C)\widehat R\otimes_R\kappa(\mathfrak P) =\widehat C\otimes_C\operatorname{Frac}(C) is a localization of a regular ring. The equality follows from finite-module completion, Tag 00MA. The fibre is geometrically regular over its characteristic-zero ground field by Tag 0381. This treats all primes of RR and completes the proof. ◻

References. The coefficientwise regular-open descent and the completion square reconstruct the characteristic-zero part of Ducros, Les espaces de Berkovich sont excellents, Propositions 2.2 and 2.9. The field extensions used in the Gauss argument retain the original absolute value.

Theorem 2.35.30. NAG-001767 (Ordinary regular maps at quasi-smooth points in characteristic zero). Let KK be any complete field of characteristic zero, let f:Y→Xf:Y\to X be a morphism of good KK-analytic spaces, and let ff be quasi-smooth at yy, with x=f(y)x=f(y). Then the ordinary local homomorphism OX,x⟶OY,y\mathcal O_{X,x}\longrightarrow\mathcal O_{Y,y} is regular.

Proof. Put R=OX,xR=\mathcal O_{X,x} and S=OY,yS=\mathcal O_{Y,y}. These rings are Noetherian, and the homomorphism is flat by Theorem 2.35.14. We prove regularity of its fibre at every prime P⊂R\mathfrak P\subset R. By Theorem 2.35.20, after an ordinary affinoid restriction of XX, the domain C=R/PC=R/\mathfrak P has a realization B↪C=OZ,z,Z=M(B)⟶XB\hookrightarrow C=\mathcal O_{Z,z},\qquad Z=\mathcal M(B)\longrightarrow X with BB integral and Z→XZ\to X finite. There is a unique point zz above xx. The base change T=Y×XZ→YT=Y\times_X Z\to Y is finite. List its finitely many points tit_i above yy, and write Si=OT,tiS_i=\mathcal O_{T,t_i}. Finite branch separation gives S⊗RFrac⁡(C)≃(∏iSi)⊗CFrac⁡(C).S\otimes_R\operatorname{Frac}(C) \simeq \left(\prod_i S_i\right)\otimes_C\operatorname{Frac}(C). The map T→ZT\to Z is quasi-smooth at each tit_i, by the Jacobian base-change theorem. Choose an ordinary affinoid neighborhood Ti=M(Di)T_i=\mathcal M(D_i) of each tit_i on which this morphism is quasi-smooth everywhere. Such a neighborhood exists by the strengthened ordinary-neighborhood presentation in Theorem 2.35.14: the equations generate on that neighborhood, and a nonzero Jacobian minor stays invertible on a smaller rational neighborhood.

Apply Proposition 2.35.27 to BB, obtaining 0≠a∈B0\ne a\in B such that, after every complete scalar extension, the ordinary local rings of ZZ are regular at all points where aa is nonzero. We claim first that every ordinary local ring of TiT_i at a point vv with ∣a(v)∣>0|a(v)|>0 is regular. Choose a nontrivially valued complete field Ω⊃H(v)\Omega\supset\mathcal H(v) and the rational lift ww of vv in (Ti)Ω(T_i)_\Omega. Its image uu in ZΩZ_\Omega is rational and has ∣a(u)∣>0|a(u)|>0. Hence R′=OZΩ,uR'=\mathcal O_{Z_\Omega,u} is regular. The lifted morphism is quasi-smooth at ww, and Proposition 2.35.13 gives O(Ti)Ω,w^≃R′^[[V1,…,Vd]].\widehat{\mathcal O_{(T_i)_\Omega,w}} \simeq \widehat{R'}[[V_1,\ldots,V_d]]. The right side is regular. Indeed, R′^\widehat{R'} is regular by Tag 07NY; its power-series extension is flat by the completion argument in that proposition, with regular closed fibre κ(u)[[V1,…,Vd]]\kappa(u)[[V_1,\ldots,V_d]] by Tag 032C. Tag 031E gives regularity of the power-series ring. Tag 07NY then makes the local ring at ww regular, and the faithfully flat scalar-stalk map descends regularity to OTi,v\mathcal O_{T_i,v} by Tag 00OF. This proves the claim.

Every prime of DiD_i avoiding aa is the kernel of an analytic point, by Lemma 2.52.1. At that point the ordinary local ring is regular by the claim. Its faithfully flat local map from the corresponding localization of DiD_i descends regularity. Thus Di[1/a]D_i[1/a] is regular. Theorem 2.35.29 makes (Di)ker⁡(ti)→Si(D_i)_{\ker(t_i)}\to S_i regular. For any prime of SiS_i avoiding aa, its contracted localization in DiD_i is regular, and the local map has regular closed fibre. Tag 031E therefore gives regularity of every such localization of SiS_i. Hence Si[1/a]S_i[1/a] is regular.

Finally aa stays nonzero in CC, since B↪CB\hookrightarrow C. The displayed algebraic fibre is therefore a localization of the finite product of regular rings ∏iSi[1/a]\prod_i S_i[1/a], and is regular. Its ground field Frac⁡(C)\operatorname{Frac}(C) has characteristic zero, so Tag 0381 makes it geometrically regular. Since P\mathfrak P was arbitrary, all algebraic fibres of the flat homomorphism R→SR\to S are geometrically regular. ◻

References. This reconstructs, in characteristic zero, the ordinary-local-ring comparison appearing in [Duc18] (5.5.3). The argument uses all prime fibres of the ordinary local rings, through finite-germ realization and generic regularity.

Theorem 2.35.31. NAG-001768 (Quasi-excellence in characteristic zero). Every generalized affinoid algebra over a complete field of characteristic zero is a quasi-excellent ring and is Nagata. Its algebraic normalization is finite.

Proof. Let AA be such an algebra. It is Noetherian. We first prove that AA is a GG-ring, meaning that for every prime p⊂A\mathfrak p\subset A the completion map Ap→Ap^A_{\mathfrak p}\to\widehat{A_{\mathfrak p}} is regular. Choose x∈M(A)x\in\mathcal M(A) with kernel p\mathfrak p, using Lemma 2.52.1, and put D=ApD=A_{\mathfrak p} and R=OM(A),xR=\mathcal O_{\mathcal M(A),x}. Theorem 2.35.29 makes D→RD\to R and R→R^R\to\widehat R regular.

We verify the composite on its fibres. For q∈Spec⁡(D)\mathfrak q\in\operatorname{Spec}(D), set F=κ(q)F=\kappa(\mathfrak q), RF=R⊗DFR_F=R\otimes_D F and HF=R^⊗DFH_F=\widehat R\otimes_D F. The ring RFR_F is regular. The map RF→HFR_F\to H_F is flat. A prime of RFR_F corresponds to a prime Q\mathfrak Q of RR over q\mathfrak q, with the same residue field, because RFR_F is obtained from RR by quotient and localization. Its fibre in HFH_F is therefore R^⊗Rκ(Q)\widehat R\otimes_R\kappa(\mathfrak Q), which is regular. All these rings are Noetherian. At each prime of HFH_F, the local source is regular and the local map has regular closed fibre; Tag 031E makes the target local ring regular. Thus HFH_F is regular, and Tag 0381 makes it geometrically regular over FF. This proves directly that D→R^D\to\widehat R is regular.

By Tag 0C4G, D^→R^\widehat D\to\widehat R is flat, since D→RD\to R is a flat local map; it is local and hence faithfully flat. After tensoring with each F=κ(q)F=\kappa(\mathfrak q) over DD, the resulting map D^⊗DF⟶R^⊗DF\widehat D\otimes_D F \longrightarrow \widehat R\otimes_D F remains faithfully flat. Its target is regular by the preceding paragraph. Tags 00HQ and 00OF descend regularity to its source, prime by prime. Tag 0381 then gives geometric regularity. Together with flatness of completion, this proves regularity of D→D^D\to\widehat D at every algebraic fibre. Hence AA is a GG-ring.

To prove the JJ-2 condition, use [StacksAlg] (Tag 07PC, equivalence (4)⇒\Rightarrow(1)). For any prime p⊂A\mathfrak p\subset A, the field κ(p)\kappa(\mathfrak p) has characteristic zero, so its finite purely inseparable extension is itself. Take the finite AA-algebra A/pA/\mathfrak p. It is a domain, has fraction field κ(p)\kappa(\mathfrak p), and has a nonempty regular principal open subset by Proposition 2.35.27. These are exactly the hypotheses of that criterion. Thus AA is JJ-2. By definition, a Noetherian GG-ring which is JJ-2 is quasi-excellent. The Nagata conclusion follows from [StacksAlg] (Tag 07QV), and finiteness of normalization from [StacksAlg] (Tag 035S), applied to Spec⁡(A)\operatorname{Spec}(A). ◻

Corollary 2.35.32. NAG-001769 (Normal loci and quasi-smooth ascent in characteristic zero). Let KK be complete of characteristic zero. For an affinoid X=M(A)X=\mathcal M(A) and x∈Xx\in X, put p=ker⁡(x)\mathfrak p=\ker(x). Then ApA_{\mathfrak p} is normal if and only if OX,x\mathcal O_{X,x} is normal. The set of such points is a Zariski open subset of XX. The normal-point set of a good KK-analytic space is therefore ordinarily open. If f:Y→Xf:Y\to X is a morphism of good spaces, is quasi-smooth at yy, and XX is normal at f(y)f(y), then YY is normal at yy.

Proof. The map Ap→OX,xA_{\mathfrak p}\to\mathcal O_{X,x} is regular by Theorem 2.35.29. Regular rings are normal by [StacksAlg] (Tag 0567), so all its fibres are normal. Normality ascends by [StacksAlg] (Tag 0C22). Conversely suppose the target R=OX,xR=\mathcal O_{X,x} is normal. This faithfully flat local map is injective, so ApA_{\mathfrak p} is a domain, since RR is a domain. If u/v∈Frac⁡(Ap)u/v\in\operatorname{Frac}(A_{\mathfrak p}) is integral over ApA_{\mathfrak p}, it is integral over RR and belongs to RR. Thus u∈vRu\in vR. Faithful flatness contracts the ideal vRvR to vApvA_{\mathfrak p}: the class of uu in Ap/vApA_{\mathfrak p}/vA_{\mathfrak p} becomes zero after tensoring with RR, hence was already zero. Consequently u/v∈Apu/v\in A_{\mathfrak p}, proving normality.

For openness, let NN be the algebraic normalization of AA, namely the product of the integral closures of its finitely many minimal-prime quotients in their fraction fields. It is finite by Theorem 2.35.31. A localization ApA_{\mathfrak p} is normal exactly when the map Ap→NpA_{\mathfrak p}\to N_{\mathfrak p} is an isomorphism. Indeed normalization commutes with localization: clearing the finitely many denominators of a monic equation proves that integral closure does so, and the minimal-prime factors localize to the surviving components. If ApA_{\mathfrak p} is normal it is reduced with one integral component and already integrally closed; conversely an isomorphism to the localization of NN makes it normal.

The kernel and cokernel of A→NA\to N are finite AA-modules. Their supports are closed: for a finite module MM, its localization at p\mathfrak p vanishes exactly when some element outside p\mathfrak p annihilates all members of a fixed finite generating set, equivalently when p\mathfrak p avoids Ann⁡(M)\operatorname{Ann}(M). Hence the normal locus of Spec⁡(A)\operatorname{Spec}(A) is the complement of these two closed supports. Pulling this open set back by x↦ker⁡(x)x\mapsto\ker(x) gives a Zariski open subset of M(A)\mathcal M(A); a principal open pulls back to ∣a(x)∣>0|a(x)|>0. The normality equivalence already proved identifies it with the ordinary normal-point set. Affinoid ordinary neighborhoods then prove ordinary openness on every good space.

Finally Theorem 2.35.30 makes OX,f(y)→OY,y\mathcal O_{X,f(y)}\to\mathcal O_{Y,y} regular. Its fibres are normal by Tag 0567. Applying Tag 0C22 gives the last assertion. ◻

Topological pp-bases and finite differential criteria

Theorem 2.35.33. NAG-001770 (Countable exponent-one extensions). Let FF be a complete non-Archimedean field of characteristic p>0p>0. Let EE be a complete intermediate field F⊂E⊂F1/pF\subset E\subset F^{1/p} which has a countable dense FF-linear span. There is a finite or countable family (ai)i∈I(a_i)_{i\in I} in EE such that the monomials aν=∏i∈Iaiνi,0≤νi<p,#{i:νi≠0}<∞,a^\nu=\prod_{i\in I}a_i^{\nu_i}, \qquad 0\leq \nu_i<p,\qquad \#\{i:\nu_i\ne0\}<\infty, form a topological FF-basis of EE. More precisely, the summation map from the weighted c0c_0-space with weights ∣aν∣|a^\nu| onto EE is a bounded linear isomorphism with bounded inverse.

Proof. We first treat a single extension L=H(x)L=H(x) of degree pp, with xp∈Hx^p\in H and HH complete. The HH-derivation D=d/dxD=d/dx is well defined on LL, and is bounded by finite-dimensional norm equivalence, Lemma 1.12.1. Suppose first that ∣H×∣|H^\times| is dense in R>0\mathbb R_{>0}. Given 0<θ<10<\theta<1, choose 0<η<10<\eta<1 with η2(p−1)>θ\eta^{2(p-1)}>\theta. Put C=sup⁡y≠0∣Dy∣/∣y∣C=\sup_{y\ne 0}|Dy|/|y|, which is finite and positive. Choose yy with ∣Dy∣>ηC∣y∣|Dy|>\eta C|y|, and multiply it by an element of HH so that a=λya=\lambda y satisfies η<∣a∣≤1\eta<|a|\leq 1. Then ∣Da∣>η2C|Da|>\eta^2C. Thus the derivation ∂=D/(Da)\partial=D/(Da) satisfies ∂a=1\partial a=1 and ∥∂∥<η−2\|\partial\|<\eta^{-2}.

For z=∑i=0p−1ciaiz=\sum_{i=0}^{p-1}c_i a^i, Taylor’s finite identity gives ci=∑j=ip−1(−a)j−ii!(j−i)! ∂jz.c_i=\sum_{j=i}^{p-1}\frac{(-a)^{j-i}}{i!(j-i)!}\,\partial^jz . All factorials here are nonzero elements of the prime field and have absolute value one. Hence θmax⁡i∣ci∣≤η2(p−1)max⁡i∣ci∣≤∣z∣≤max⁡i∣ci∣.\theta\max_i|c_i|\leq \eta^{2(p-1)}\max_i|c_i| \leq |z|\leq\max_i|c_i|. In particular, aa generates L/HL/H and its power basis has a coefficient bound as close to one as desired.

If HH is discretely valued, the value group of L⊂H1/pL\subset H^{1/p} is discrete as well. The bounded set {∣Dz∣:∣z∣≤1}\{|Dz|:|z|\leq 1\} therefore has a positive maximum. Choose aa in the unit ball attaining it. The derivation ∂=D/(Da)\partial=D/(Da) preserves that unit ball. The same finite Taylor identity shows that the coefficients of every zz in the unit ball, expressed in 1,a,…,ap−11,a,\ldots,a^{p-1}, lie in H∘H^\circ. Consequently L∘=⨁i=0p−1H∘ai.L^\circ=\bigoplus_{i=0}^{p-1}H^\circ a^i . For a trivial absolute value all these assertions follow directly from the algebraic power basis.

Choose a sequence generating a dense subfield E′⊂EE'\subset E over FF, and write E′=⋃nFnE'=\bigcup_n F_n, where each nontrivial step Fn/Fn−1F_n/F_{n-1} has degree pp. Each FnF_n is complete and contained in F1/pF^{1/p}. If FF is nondiscretely and nontrivially valued, choose numbers θn∈(0,1)\theta_n\in(0,1) with θ∗=∏nθn>0\theta_*=\prod_n\theta_n>0, and use the first construction at the nnth step. Iteration of its coefficient estimate gives, for every finite linear combination of the resulting pp-monomials, θ∗max⁡ν∣cν∣≤∣∑νcνaν∣≤max⁡ν∣cν∣.\theta_*\max_\nu|c_\nu| \leq\left|\sum_\nu c_\nu a^\nu\right| \leq\max_\nu|c_\nu|. If FF is discretely valued, use the integral power bases instead. They give (E′)∘=⨁νF∘aν(E')^\circ=\bigoplus_\nu F^\circ a^\nu. For a uniformizer π∈F\pi\in F, scaling a nonzero sum into the interval (∣π∣,1](|\pi|,1] gives the same estimate with θ∗=∣π∣\theta_*=|\pi|. For trivial FF, take θ∗=1\theta_*=1. In all cases θ∗≤∣aν∣≤1\theta_*\leq|a^\nu|\leq 1, so the displayed norm is equivalent to the weighted maximum norm. Completing this finite-support direct sum gives exactly the indicated weighted c0c_0-space. Its image is closed by the lower estimate and dense because it contains E′E'. It is therefore all of EE, and both continuity assertions follow. ◻

References. This is a reconstruction of Kiehl, Ausgezeichnete Ringe in der nichtarchimedischen analytischen Geometrie, J. reine angew. Math. 234 (1969), pp. 90–91, Satz 1.2 and Satz 1.4. The derivation estimate above makes the uniform coefficient constant explicit.

Lemma 2.35.34. NAG-001771 (Cofinite subfields of a topological pp-basis). Let (ai)i∈I(a_i)_{i\in I} be a topological pp-basis of a complete exponent-one extension E/FE/F. For J⊂IJ\subset I, let EJE_J be the closure of F(aj:j∈J)F(a_j:j\in J) in EE. It consists precisely of the convergent pp-monomial expansions supported on JJ. If I∖JI\setminus J is finite, then E=⨁0≤νi<pi∈I∖JEJaν,[E:EJ]=p∣I∖J∣,ΩE/EJ=⨁i∈I∖JE dai.E=\bigoplus_{\substack{0\leq \nu_i<p\\i\in I\setminus J}} E_J a^\nu,\qquad [E:E_J]=p^{|I\setminus J|},\qquad \Omega_{E/E_J}=\bigoplus_{i\in I\setminus J}E\,da_i . Moreover ⋂I∖J finiteEJ=F\bigcap_{I\setminus J\text{ finite}}E_J=F.

Proof. The subspace of expansions supported on JJ is closed, because each coordinate functional in a topological basis is continuous. Finite such expansions form the field F(aj:j∈J)F(a_j:j\in J): the ppth power of every element lies in FF, so the inverse of a nonzero element bb is bp−1/bpb^{p-1}/b^p. Its closure is again a field, since inversion is continuous away from zero. Thus the claimed description of EJE_J follows. Regrouping the expansion by the finitely many exponents outside JJ gives the displayed direct sum; bounded coordinate maps justify regrouping and convergence. Independence follows from uniqueness of the full expansion. The presentation EJ[Ui:i∈I∖J]/(Uip−aip)i → ∼  EE_J[U_i:i\in I\setminus J]/(U_i^p-a_i^p)_i \ \xrightarrow{\ \sim\ }\ E then computes the differentials: the defining equations have zero differential and the indicated monomials form a basis on both sides. Finally, membership in every EJE_J forces all coefficients of nonconstant monomials to vanish, since each such monomial involves an index excluded by some cofinite JJ. Only the constant coefficient remains. ◻

Lemma 2.35.35. NAG-001772 (Finite rank stabilization over descending fields). Let (Fλ)(F_\lambda) be a downward directed family of subfields of a field LL, and put F∞=⋂λFλF_\infty=\bigcap_\lambda F_\lambda. For every finite family v1,…,vd∈Lv_1,\ldots,v_d\in L, its rank over F∞F_\infty equals its rank over some Fλ0F_{\lambda_0}. In particular, a family independent over F∞F_\infty is independent over one member of the family.

Proof. Choose λ0\lambda_0 at which the integer rank is maximal, and choose from the viv_i a basis w1,…,wrw_1,\ldots,w_r of their span over Fλ0F_{\lambda_0}. For every λ\lambda with Fλ⊂Fλ0F_\lambda\subset F_{\lambda_0}, these elements stay independent, and maximality makes them a basis of the span over FλF_\lambda. The coefficients expressing any viv_i in this basis are unique among coefficient tuples in Fλ0F_{\lambda_0}, hence independent of λ\lambda, and lie in every such FλF_\lambda. This subfamily is cofinal by downward directedness, so those coefficients lie in F∞F_\infty. Thus the same wjw_j span over F∞F_\infty and remain independent over it. ◻

Lemma 2.35.36. NAG-001773 (A triangular presentation on a finite generic open). Let D⊂AD\subset A be a finite extension of Noetherian integral domains. There is a nonzero d∈Dd\in D such that A[1/d]≃D[1/d][U1,…,Um]/(P1,…,Pm),A[1/d]\simeq D[1/d][U_1,\ldots,U_m]/(P_1,\ldots,P_m), where PjP_j is monic in UjU_j, has coefficients in D[1/d][U1,…,Uj−1]D[1/d][U_1,\ldots,U_{j-1}], and the reduced monomials form a free D[1/d]D[1/d]-basis.

Proof. Put F=Frac⁡(D)F=\operatorname{Frac}(D) and E=Frac⁡(A)E=\operatorname{Frac}(A). Since AA is finite and integral over DD, A⊗DF=EA\otimes_DF=E. Choose a tower F⊂F(u1)⊂⋯⊂F(u1,…,um)=EF\subset F(u_1)\subset\cdots\subset F(u_1,\ldots,u_m)=E of simple finite extensions, with the uju_j chosen among elements of AA generating EE. The minimal polynomial of uju_j over the preceding field is monic. Express its coefficients in the reduced power basis of that preceding field. Clearing the finitely many denominators in DD gives triangular monic polynomials over D[1/d]D[1/d]. Their quotient is free with the reduced monomials as basis, by successive division by monic polynomials. Its evaluation map into A[1/d]A[1/d] becomes an isomorphism over FF, so its kernel is zero because this free module is torsion-free. The cokernel is a finite D[1/d]D[1/d]-module whose localization at FF vanishes. A further nonzero multiplier in DD kills that cokernel. Replacing dd by its product with this multiplier proves the assertion. ◻

Lemma 2.35.37. NAG-001774 (Equal differential ranks and regularity). Let S→BS\to B be a ring map with BB Noetherian and regular, and suppose that ΩB/S\Omega_{B/S} is free of rank NN. Let C=B[U1,…,Um]/(P1,…,Pm)C=B[U_1,\ldots,U_m]/(P_1,\ldots,P_m) is finite over BB, with the displayed triangular monic presentation. If ΩC/S\Omega_{C/S} is free of rank NN, then CC is regular. The same conclusion holds for each direct factor of CC.

Proof. The differential presentation is Cm→(dPj)CN+m⟶ΩC/S⟶0.C^m\xrightarrow{(dP_j)} C^{N+m}\longrightarrow \Omega_{C/S}\longrightarrow0. The last module is free of rank NN, so its kernel is projective of constant rank mm. The first map surjects onto this kernel. After localization at any prime of CC, a surjection between free modules of rank mm is an isomorphism: reduce modulo the maximal ideal, then its determinant is a unit. Hence the dPjdP_j are independent modulo every such prime. Let q\mathfrak q be its inverse image in B[U]B[U]. This polynomial ring is regular by [StacksAlg] (Tag 07NF), so its localization R=B[U]qR=B[U]_{\mathfrak q} is regular local. A relation ∑cjPj∈q2R\sum c_jP_j\in\mathfrak q^2R would, on differentiating and reducing modulo q\mathfrak q, imply ∑c‾j dPj=0\sum\overline c_j\,dP_j=0. Therefore the classes of the PjP_j are independent in qR/(qR)2\mathfrak qR/(\mathfrak qR)^2. Extend them to a minimal set of generators; [StacksAlg] (Tag 00NQ) shows that R/(P1,…,Pm)R/(P_1,\ldots,P_m) is regular local. These are all the local rings of CC, proving regularity. Local rings of a direct factor are among them. ◻

Definition 2.35.38. (Analytically separable scalar extensions). In characteristic p>0p>0, a complete extension L/KL/K is called analytically separable if multiplication Lp⊗^KpK⟶LL^p\widehat\otimes_{K^p}K\longrightarrow L is injective with closed image and its inverse onto that image is bounded. In characteristic zero every complete extension is called analytically separable.

Lemma 2.35.39. NAG-001776 (Free Gauss extensions and preservation of pp-bases). The identity extension and every free Gauss extension Kr/KK_r/K are analytically separable. If L/KL/K is analytically separable and K0K_0 is a complete intermediate field Kp⊂K0⊂KK^p\subset K_0\subset K with a topological pp-basis aa over KpK^p, then aa remains topologically pp-independent over LpL^p. For every complete subfield F⊂LF\subset L containing LpK0L^pK_0 and of countable type over LpL^p, and every cofinite subset b⊂ab\subset a, there is a complete subfield F♯⊂FF^\sharp\subset F containing the closure KbK_b of Kp(b)K^p(b) such that K0⊗KbF♯→ ∼ F.K_0\otimes_{K_b}F^\sharp \xrightarrow{\ \sim\ }F.

Proof. The assertion for the identity is immediate. If L=KrL=K_r, the coefficient description gives Lp=KrppL^p=K^p_{r^p}. The map Lp⊗^KpK→LL^p\widehat\otimes_{K^p}K\to L identifies with the subspace of Laurent expansions whose exponents are divisible by pp. The coefficient formula of Theorem 1.16.2 makes this an isometric embedding with closed image. For general analytically separable L/KL/K, Theorem 2.33.2 embeds Lp⊗^KpK0L^p\widehat\otimes_{K^p}K_0 into Lp⊗^KpKL^p\widehat\otimes_{K^p}K. A bounded topological basis and its inverse remain bounded after completed tensor product. Thus the pp-monomials in aa form a topological LpL^p-basis of the closure F0F_0 of LpK0L^pK_0 in LL.

The extension F/F0F/F_0 has exponent one and countable type. By Theorem 2.35.33 it has a topological pp-basis cc. Combining the two bounded basis isomorphisms shows that a∪ca\cup c is a topological pp-basis of F/LpF/L^p. Let F♯F^\sharp be the closed subfield generated by LpL^p and b∪cb\cup c. Lemma 2.35.34 shows that a∖ba\setminus b is a finite algebraic pp-basis of F/F♯F/F^\sharp. It is also a pp-basis of K0/KbK_0/K_b. The multiplication map in the statement therefore carries the same monomial basis to the same monomial basis, and is an isomorphism. ◻

Lemma 2.35.40. NAG-001777 (Countable descent of completed ordinary germs). Let LL be complete of characteristic p>0p>0, let F0⊂LF_0\subset L be complete with Lp⊂F0L^p\subset F_0, and let X0X_0 be an affinoid F0F_0-space. Put X=(X0)LX=(X_0)_L, choose x∈Xx\in X, and denote its image in (X0)F(X_0)_F by xFx_F. If FF ranges over the complete intermediate fields F0⊂F⊂LF_0\subset F\subset L of countable type over F0F_0, then colim⁡F O(X0)F,xF^→ ∼ OX,x^.\underset{F}{\operatorname{colim}}\, \widehat{\mathcal O_{(X_0)_F,x_F}} \xrightarrow{\ \sim\ }\widehat{\mathcal O_{X,x}} . All transition maps, and all maps into the ring on the right, are faithfully flat local maps.

Proof. The ordinary local maps are faithfully flat by Theorem 2.35.4; their completions are flat by Tag 0C4G, and local, hence faithfully flat. In particular they are injective. Every individual germ descends to some FF in the stated family. Indeed represent it on a rational affinoid ordinary neighborhood, using finitely many inequalities. Lift the finitely many functions in those inequalities, and then the represented analytic function on the rational algebra, to convergent weighted series. Only countably many scalar coefficients of LL occur. Adjoining those coefficients to F0F_0 and taking the closure gives such an FF: each coefficient has its ppth power in Lp⊂F0L^p\subset F_0, and the finitely supported pp-monomials in countably many generators have countable dense linear span. The inequalities defining the ordinary neighborhood hold in a neighborhood of xFx_F, since their strict versions hold at xFx_F exactly when they hold at xx.

Write R=OX,xR=\mathcal O_{X,x}, choose finitely many generators f1,…,fdf_1,\ldots,f_d of its maximal ideal, and descend them to some F1F_1. For g∈R^g\in\widehat R, choose gn∈Rg_n\in R with g−gn∈(f1,…,fd)nR^g-g_n\in(f_1,\ldots,f_d)^n\widehat R. Descend all gng_n and the fif_i to one field FF; a countable union of the coefficient sets above is still countable. Put RF=O(X0)F,xFR_F=\mathcal O_{(X_0)_F,x_F}, and denote the descended elements by fi′f_i' and gn′g_n'. The faithfully flat map RF/(fi′)⟶R/(fi)R_F/(f_i')\longrightarrow R/(f_i) has a field as target, so its source is a field. Thus the fi′f_i' generate the maximal ideal of RFR_F. Faithful flatness contracts every power of this ideal, and therefore gn+1′−gn′∈(fi′)ng_{n+1}'-g_n'\in(f_i')^n. The compatible residues of the gn′g_n' define an element of RF^\widehat{R_F} whose image is gg. This proves surjectivity of the colimit map; injectivity was proved at the start. ◻

Theorem 2.35.41. NAG-001778 (A uniform regular open in positive characteristic). Let KK be complete of characteristic p>0p>0, let rr be a finite KK-free polyradius, and let D=Kr{T1,…,Tl}⊂AD=K_r\{T_1,\ldots,T_l\}\subset A be a finite extension of integral domains. Equip AA with its finite affinoid norm. There exists 0≠d∈D0\ne d\in D such that, for every analytically separable complete extension L/KL/K and every x∈XL=M(AL)x\in X_L=\mathcal M(A_L), OXL,x[1/d],OXL,x^[1/d],AL[1/d]\mathcal O_{X_L,x}[1/d],\qquad \widehat{\mathcal O_{X_L,x}}[1/d],\qquad A_L[1/d] are regular rings.

Proof. Step 1: descend the finite algebra. Choose finitely many integral generators for A/DA/D, their monic equations, and a finite set of generators for the ideal of relations in a polynomial presentation. Every coefficient in DD is a Laurent–Tate series with countably many coefficients in KK. Let K0K_0 be the closed subfield generated by KpK^p and all these coefficients. It is of countable type over KpK^p. The descended equations define a finite algebra A0A_0 over D0=(K0)r{T}D_0=(K_0)_r\{T\} and an isomorphism A0⊗D0D≃AA_0\otimes_{D_0}D\simeq A. Indeed the chosen monic equations make A0A_0 finite, and the full finite relation list gives that isomorphism. The map D0→DD_0\to D is faithfully flat by Theorem 2.33.6. Thus A0A_0 injects into AA and is an integral domain; also D0⊂A0D_0\subset A_0. Put F0=Frac⁡(D0)F_0=\operatorname{Frac}(D_0), E0=Frac⁡(A0)E_0=\operatorname{Frac}(A_0) and F=Frac⁡(D)F=\operatorname{Frac}(D). A basis α1,…,αe\alpha_1,\ldots,\alpha_e of E0/F0E_0/F_0 remains a basis after extension to FF, since E0⊗F0F=Frac⁡(A)E_0\otimes_{F_0}F=\operatorname{Frac}(A).

Step 2: retain finitely many pp-directions. Choose a topological pp-basis aa of K0/KpK_0/K^p by Theorem 2.35.33. For a cofinite b⊂ab\subset a, let KbK_b be the closure of Kp(b)K^p(b) and put Db=(Kb)rp{T1p,…,Tlp},Fb=Frac⁡(Db).D_b=(K_b)_{r^p}\{T_1^p,\ldots,T_l^p\},\qquad F_b=\operatorname{Frac}(D_b). Here the Laurent variables in DbD_b are the ppth powers of those in D0D_0. Grouping their exponents and the TT-exponents modulo pp, and using Lemma 2.35.34, makes D0D_0 finite free and radicial over DbD_b, of rank pNp^N, where N=∣a∖b∣+∣r∣+lN=|a\setminus b|+|r|+l. Its differentials over DbD_b are free of rank NN, with basis given by the retained coefficient directions, the Laurent variables, and the TiT_i.

We need the equality ⋂bFb=Fp\bigcap_b F_b=F^p. Set Da=(K0)rp{Tp}D_a=(K_0)_{r^p}\{T^p\}. An element of the intersection is g/hg/h with g,h∈Dag,h\in D_a, h≠0h\ne 0. Replacing numerator and denominator by ghp−1gh^{p-1} and hph^p puts the denominator in (Kp)rp{Tp}(K^p)_{r^p}\{T^p\}. For every cofinite bb, the new numerator belongs both to DaD_a and to FbF_b. Since Db→DaD_b\to D_a is faithfully flat, Da∩Fb=DbD_a\cap F_b=D_b: a relation vg=uv g=u with u,v∈Dbu,v\in D_b, v≠0v\ne 0, contracts the principal ideal vDavD_a to vDbvD_b. Hence the numerator belongs to every DbD_b. Uniqueness of Laurent–Tate coefficients and ⋂bKb=Kp\bigcap_b K_b=K^p identify their intersection with (Kp)rp{Tp}(K^p)_{r^p}\{T^p\}. Taking fractions proves the required equality.

The elements αip\alpha_i^p are independent over FpF^p, by taking ppth roots of a putative relation and using the FF-independence of the αi\alpha_i. Lemma 2.35.35 therefore supplies a cofinite bb for which they are independent over FbF_b. Since they form a basis of E0p/F0pE_0^p/F_0^p and F0p⊂FbF_0^p\subset F_b, this implies [E0pFb:Fb]=e,[E0:E0pFb]=[F0:Fb]=pN.[E_0^pF_b:F_b]=e,\qquad [E_0:E_0^pF_b]=[F_0:F_b]=p^N. The universal derivation over FbF_b annihilates E0pFbE_0^pF_b. An algebraic pp-basis of the finite exponent-one extension E0/E0pFbE_0/E_0^pF_b computes its differentials by the equations Uip−ciU_i^p-c_i; thus dim⁡E0ΩE0/Fb=N\dim_{E_0}\Omega_{E_0/F_b}=N.

Step 3: choose one denominator. The finite A0A_0-module ΩA0/Db\Omega_{A_0/D_b} consequently becomes free of rank NN after inverting a nonzero element of A0A_0: choose lifts of a generic basis, then kill the finite torsion kernel and cokernel. Any nonzero element of A0A_0 becomes a unit after inverting a nonzero element of D0D_0, since A0⊗D0F0A_0\otimes_{D_0}F_0 is a field. Raising that latter element to its ppth power puts it in DbD_b. Combine this denominator with the one in Lemma 2.35.36, again taking a ppth power. We obtain 0≠d∈Db0\ne d\in D_b for which A0[1/d]A_0[1/d] has a triangular monic presentation over D0[1/d]D_0[1/d] and ΩA0[1/d]/Db[1/d]\Omega_{A_0[1/d]/D_b[1/d]} is free of rank NN.

Step 4: completed germs over countable scalar fields. Fix an analytically separable complete extension L/KL/K. Let F∗F_* be the closure of LpK0L^pK_0 in LL, and let HH range over the complete fields between F∗F_* and LL of countable type over F∗F_*. Lemma 2.35.39 gives H♯H^\sharp with H=K0⊗KbH♯H=K_0\otimes_{K_b}H^\sharp. Write DH=D0⊗^K0H,BH=Db⊗^KbH♯,AH=A0⊗D0DH.D_H=D_0\widehat\otimes_{K_0}H,\quad B_H=D_b\widehat\otimes_{K_b}H^\sharp,\quad A_H=A_0\otimes_{D_0}D_H . The finite monomial bases prove the actual algebra identity DH=D0⊗DbBHD_H=D_0\otimes_{D_b}B_H. Both DHD_H and BHB_H are algebras of affinoid domains in affine space: a product of a polydisc with polyannuli of exact radii rr and rpr^p, respectively. Here the Laurent construction remains a polyannulus even when a radius ceases to be free over HH.

Fix the image xHx_H of xx in M(AH)\mathcal M(A_H), and let yy and zz be its images in M(DH)\mathcal M(D_H) and M(BH)\mathcal M(B_H). The finite radicial map M(DH)→M(BH)\mathcal M(D_H)\to\mathcal M(B_H) has a unique point over zz: in a radicial finite fibre there is one residue field, purely inseparable over H(z)\mathcal H(z), and its absolute value is forced by ∣u∣pm=∣upm∣|u|^{p^m}=|u^{p^m}|. The finite branch and completion formulas of Lemma 2.35.1 therefore give, with S=OM(BH),z^S=\widehat{\mathcal O_{\mathcal M(B_H),z}} and B=OM(DH),y^B=\widehat{\mathcal O_{\mathcal M(D_H),y}}, B=D0⊗DbS,A0⊗D0B=∏t↦yOM(AH),t^.B=D_0\otimes_{D_b}S,\qquad A_0\otimes_{D_0}B =\prod_{t\mapsto y}\widehat{\mathcal O_{\mathcal M(A_H),t}} . The ring BB is regular by Theorem 2.35.23 and Tag 07NY. After inverting dd, its differentials over S[1/d]S[1/d] are free of rank NN, and those of (A0⊗D0B)[1/d](A_0\otimes_{D_0}B)[1/d] have the same rank, by finite algebraic base change of the differential presentations in Steps 2 and 3. The latter algebra also retains its triangular monic presentation over B[1/d]B[1/d]. Lemma 2.35.37 thus proves that it is regular. Each factor, in particular OM(AH),xH^[1/d]\widehat{\mathcal O_{\mathcal M(A_H),x_H}}[1/d], is regular.

Step 5: pass to LL. Lemma 2.35.40 identifies OXL,x^[1/d]\widehat{\mathcal O_{X_L,x}}[1/d] with the filtered colimit of the regular rings in Step 4. Localize at any prime and its contractions. The resulting system consists of regular local rings with local transition maps and has a Noetherian colimit. By [StacksAlg] (Tag 07DX), that colimit is regular local. This proves regularity of the completed localized germ. Completion is faithfully flat, also after inverting dd, so Tags 00HQ and 00OF descend regularity to OXL,x[1/d]\mathcal O_{X_L,x}[1/d]. Finally, every prime of ALA_L avoiding dd is the kernel of an analytic point by Lemma 2.52.1. Its faithfully flat local map into the ordinary germ at that point descends regularity once more, proving that AL[1/d]A_L[1/d] is regular. ◻

References. The proof reconstructs the positive-characteristic argument of Ducros, Les espaces de Berkovich sont excellents, Proposition 2.2(b) and Lemmas 2.6–2.8, using the explicit pp-basis construction and the two finite algebra lemmas above.

Theorem 2.35.42. NAG-001779 (Ordinary regular comparison for strict affinoids). Let KK be nontrivially valued, let AA be strict KK-affinoid, let x∈X=M(A)x\in X=\mathcal M(A), and put p=ker⁡(x)\mathfrak p=\ker(x) and R=OX,xR=\mathcal O_{X,x}. The homomorphisms Ap⟶R,R⟶R^A_{\mathfrak p}\longrightarrow R, \qquad R\longrightarrow\widehat R are regular: they are flat, and every algebraic fibre is geometrically regular.

Proof. The characteristic-zero case is Theorem 2.35.26. Assume that KK has characteristic p>0p>0. Noetherianity and flatness are exactly as in the first paragraph of that proof. By Tag 0381, it is enough to test regularity after every finite purely inseparable extension of the residue field of every prime. We perform those tests explicitly.

First choose q⊂p\mathfrak q\subset\mathfrak p in AA, put A0=A/qA_0=A/\mathfrak q, and let E/Frac⁡(A0)E/\operatorname{Frac}(A_0) be finite and purely inseparable. Choose generators αj\alpha_j of this field extension and powers pejp^{e_j} with αjpej∈Frac⁡(A0)\alpha_j^{p^{e_j}}\in\operatorname{Frac}(A_0). Clearing denominators gives nonzero cj∈A0c_j\in A_0 such that βj=cjαj\beta_j=c_j\alpha_j has βjpej∈A0\beta_j^{p^{e_j}}\in A_0. Thus B=A0[βj]B=A_0[\beta_j] is a finite integral domain, has fraction field EE, and is radicial over A0A_0. Equip it with its finite affinoid norm; it is strict by Proposition 1.14.4. There is a unique analytic point zz of Z=M(B)Z=\mathcal M(B) above xx: the finite fibre has one residue field, purely inseparable over H(x)\mathcal H(x), whose absolute value is determined by its pp-power equations. Finite branch separation gives R⊗AB=OZ,z,R⊗AE=OZ,z⊗BE.R\otimes_A B=\mathcal O_{Z,z},\qquad R\otimes_A E=\mathcal O_{Z,z}\otimes_B E . Theorem 1.12.2 gives a finite inclusion K{T1,…,Tl}⊂BK\{T_1,\ldots,T_l\}\subset B. Apply Theorem 2.35.41 with empty free polyradius and the identity scalar extension. It supplies 0≠d∈B0\ne d\in B such that OZ,z[1/d]\mathcal O_{Z,z}[1/d] is regular. The displayed EE-fibre is a further localization of this regular ring, since every nonzero element of BB is invertible in EE. Hence it is regular. This treats every finite purely inseparable test of every fibre of Ap→RA_{\mathfrak p}\to R.

For the completion map, choose any prime P⊂R\mathfrak P\subset R and any finite purely inseparable extension E/Frac⁡(R/P)E/\operatorname{Frac}(R/\mathfrak P). Clearing denominators as in Proposition 2.35.22 produces a finite domain CC over R/PR/\mathfrak P with fraction field EE. Theorem 2.35.20 and Lemma 2.35.25 realize it as B↪C=OZ,z,Z=M(B),B\hookrightarrow C=\mathcal O_{Z,z}, \qquad Z=\mathcal M(B), with BB a strict integral affinoid algebra. In particular CC is local and finite over RR. Its maximal-ideal topology equals its mRC\mathfrak m_R C-adic topology: the finite algebra C/mRCC/\mathfrak m_RC is local Artinian, so a power of its maximal ideal is zero. The finite-module completion formula, Tag 00MA, therefore gives R^⊗RC=C^,R^⊗RE=C^⊗CE.\widehat R\otimes_R C=\widehat C, \qquad \widehat R\otimes_R E =\widehat C\otimes_C E . Apply the same identity-extension case of Theorem 2.35.41 to a strict finite normalization of BB. It gives 0≠d∈B0\ne d\in B with C^[1/d]\widehat C[1/d] regular. The injection B↪CB\hookrightarrow C ensures that dd remains nonzero in CC, and hence becomes invertible in EE. Thus the last displayed ring is a localization of a regular ring and is regular. Tag 0381 now proves that every algebraic fibre of R→R^R\to\widehat R is geometrically regular, completing both assertions. ◻

Proposition 2.35.43. NAG-001780 (A regular open preserved by analytically separable extensions). Let KK be any complete field and let AA be an integral generalized KK-affinoid algebra. There is 0≠a∈A0\ne a\in A such that, for every analytically separable complete L/KL/K, the ring AL[1/a]A_L[1/a] is regular and all ordinary local rings of M(AL)\mathcal M(A_L) at points where aa is nonzero are regular.

Proof. In characteristic zero this is Proposition 2.35.27. In positive characteristic, choose a finite KK-free polyradius rr such that KrK_r is nontrivially valued and ArA_r is strict. The kernel-zero point and its Gauss lift used in the proof of that proposition also show, in every characteristic, that ArA_r is integral: the Gauss value of a nonzero Laurent coefficient family is positive. By Theorem 1.12.2 there is a finite inclusion Kr{T1,…,Tl}⊂ArK_r\{T_1,\ldots,T_l\}\subset A_r. Theorem 2.35.41, applied over the original field KK with this polyradius rr, gives a nonzero d∈Ard\in A_r having its stated regularity properties after every analytically separable extension of KK. Write d=∑ν∈Z∣r∣aνUν,a=aν0≠0d=\sum_{\nu\in\mathbb Z^{|r|}}a_\nu U^\nu, \qquad a=a_{\nu_0}\ne0 for one nonzero coefficient. Fix such an extension L/KL/K and y∈M(AL)y\in\mathcal M(A_L) with ∣a(y)∣>0|a(y)|>0. Associativity of completed tensor products and the coefficient description identify (Ar)L=AL{r−1U,rU−1}.(A_r)_L=A_L\{r^{-1}U,rU^{-1}\}. The right side is the algebra of the relative polyannulus P→M(AL)P\to\mathcal M(A_L) on which ∣Ui∣=ri|U_i|=r_i. Evaluate its coefficients in H(y)\mathcal H(y) and then in the completion of H(y)(U)\mathcal H(y)(U) for the weighted Gauss absolute value. This defines a point zz of PP above yy and gives ∣d(z)∣=max⁡ν∣aν(y)∣rν>0.|d(z)|=\max_\nu |a_\nu(y)|r^\nu>0. The multiplicativity and extension of this weighted Gauss norm were proved in Lemma 3.21.7; no freedom assumption over LL is used. The ring OP,z\mathcal O_{P,z} is regular by the choice of dd. Moreover PP is the Laurent domain ∣Ui∣≥ri|U_i|\ge r_i in the relative polydisc ∣Ui∣≤ri|U_i|\le r_i. Corollary 2.48.3 therefore makes OM(AL),y→OP,z\mathcal O_{\mathcal M(A_L),y}\to\mathcal O_{P,z} a flat local map. Tag 00OF descends regularity to the source.

Finally, every prime of ALA_L avoiding aa is the kernel of an analytic point yy, by Lemma 2.52.1. Its local ring maps faithfully flatly to the ordinary local ring at yy, which is regular by the preceding paragraph. Another application of Tag 00OF proves regularity at that prime. This proves regularity of AL[1/a]A_L[1/a] and all the assertions with a single coefficient aa independent of LL. ◻

Proposition 2.35.44. NAG-001781 (Generic regularity of arbitrary analytic germs). Let AA be an integral generalized affinoid algebra over any complete field KK. There is 0≠a∈A0\ne a\in A such that, for every x∈X=M(A)x\in X=\mathcal M(A), the rings OX,x[1/a]\mathcal O_{X,x}[1/a] and OX,x^[1/a]\widehat{\mathcal O_{X,x}}[1/a] are regular.

Proof. Choose aa by Proposition 2.35.43, and choose a nontrivial free strictification KrK_r of AA. Lemma 2.35.39 shows that this scalar extension is analytically separable, so B=ArB=A_r satisfies B[1/a]B[1/a] regular. For z∈M(B)z\in\mathcal M(B) above xx, put R=OX,xR=\mathcal O_{X,x} and S=OM(B),zS=\mathcal O_{\mathcal M(B),z}. Theorem 2.35.42 makes Bker⁡(z)→SB_{\ker(z)}\to S and S→S^S\to\widehat S regular. At a prime of SS avoiding aa, its contracted localization in BB is regular and its closed fibre is regular; Tag 031E therefore proves regularity of that localization of SS. The same argument for S→S^S\to\widehat S proves that both S[1/a]S[1/a] and S^[1/a]\widehat S[1/a] are regular.

The map R→SR\to S is faithfully flat by Theorem 2.35.4, and its completed map is faithfully flat by Tag 0C4G. These properties persist after inverting aa. Every prime of either source has a prime above it in the target by Tag 00HQ; the corresponding localized flat local map has regular target. Tag 00OF descends regularity in each case. Thus both R[1/a]R[1/a] and R^[1/a]\widehat R[1/a] are regular, including at algebraic primes of these rings that arise when a(x)=0a(x)=0. ◻

Theorem 2.35.45. NAG-001782 (Ordinary regular comparison in every characteristic). Let KK be any complete non-Archimedean field, let AA be a generalized KK-affinoid algebra, and let x∈X=M(A)x\in X=\mathcal M(A). Put p=ker⁡(x)\mathfrak p=\ker(x) and R=OX,xR=\mathcal O_{X,x}. Then Ap⟶R,R⟶R^A_{\mathfrak p}\longrightarrow R, \qquad R\longrightarrow\widehat R are regular homomorphisms of Noetherian rings.

Proof. The characteristic-zero case was established in Theorem 2.35.29. In positive characteristic, the flatness and Noetherianity argument of Theorem 2.35.42 is unchanged: the ordinary-neighborhood colimit uses arbitrary rational domains, and completion is faithfully flat. We retain its finite purely inseparable fibre tests, but replace strict finite normalization by Proposition 2.35.44. Here are the two substitutions in full.

For q⊂p\mathfrak q\subset\mathfrak p and a finite purely inseparable extension E/Frac⁡(A/q)E/\operatorname{Frac}(A/\mathfrak q), construct the finite radicial domain B⊂EB\subset E over A/qA/\mathfrak q by the same denominator-clearing power equations. The finite affinoid norm exists by Theorem 2.33.18. With Z=M(B)Z=\mathcal M(B), there is a unique zz above xx, and R⊗AE=OZ,z⊗BE.R\otimes_A E =\mathcal O_{Z,z}\otimes_B E. The new generic-germ proposition gives 0≠b∈B0\ne b\in B with OZ,z[1/b]\mathcal O_{Z,z}[1/b] regular. Since Frac⁡(B)=E\operatorname{Frac}(B)=E, the displayed fibre is its further localization and is regular. Tag 0381 proves geometric regularity of every fibre of the first homomorphism.

For P⊂R\mathfrak P\subset R and a finite purely inseparable extension E/Frac⁡(R/P)E/\operatorname{Frac}(R/\mathfrak P), construct a finite domain CC over R/PR/\mathfrak P with fraction field EE, and realize it using Theorem 2.35.20 as B↪C=OZ,zB\hookrightarrow C=\mathcal O_{Z,z}, where BB is now an arbitrary integral generalized affinoid algebra. The finite-local completion calculation in the strict proof applies without change and gives R^⊗RE=C^⊗CE.\widehat R\otimes_R E =\widehat C\otimes_C E. Proposition 2.35.44 supplies 0≠b∈B0\ne b\in B with C^[1/b]\widehat C[1/b] regular. Its image in CC is nonzero by the injection B↪CB\hookrightarrow C, hence invertible in EE. Again the fibre is a localization of a regular ring. Tag 0381 treats all primes P\mathfrak P and proves geometric regularity of every completion fibre. This completes the proof over arbitrary radii and with either trivial or nontrivial valuation. ◻

References. The last four results reconstruct the regular-open and ordinary-local comparison in Ducros, Les espaces de Berkovich sont excellents, Propositions 2.2 and 2.9. The reduction to finite purely inseparable field tests is retained explicitly, while completed-germ regularity is supplied by the preceding topological pp-basis argument.

Theorem 2.35.46. NAG-001783 (Excellence of affinoid algebras and ordinary analytic local rings). Every generalized affinoid algebra over a complete non-Archimedean field is excellent. Every ordinary local ring of a good analytic space over such a field is excellent. In particular, these rings are Nagata and have finite algebraic normalization.

Proof. Let AA be affinoid and let R=OM(A),xR=\mathcal O_{\mathcal M(A),x}. Both are Noetherian. Theorem 2.35.45 proves that R→R^R\to\widehat R is regular. The maximal-ideal criterion [StacksAlg] (Tag 07PT) therefore makes RR a GG-ring. For any prime p⊂A\mathfrak p\subset A, choose an analytic point xx with kernel p\mathfrak p and put D=ApD=A_{\mathfrak p}. The same theorem makes D→R→R^D\to R\to\widehat R a composite of regular maps. Since the target is Noetherian, the composite is regular by [StacksAlg] (Tag 07QI). The completed map D^→R^\widehat D\to\widehat R is a flat local map by Tag 0C4G, hence faithfully flat. Descent of regularity of ring maps, [StacksAlg] (Tag 07NT), applied to D→D^→R^D\to\widehat D\to\widehat R, makes D→D^D\to\widehat D regular. This holds for every p\mathfrak p, so AA is a GG-ring as well.

We verify the JJ-2 property with Tag 07PC(4). For a prime p⊂A\mathfrak p\subset A and a finite purely inseparable extension E/κ(p)E/\kappa(\mathfrak p), clearing monic-equation denominators gives a finite domain BB over A/pA/\mathfrak p with fraction field EE. Its canonical finite affinoid structure is supplied by Theorem 2.33.18. The identity-extension case of Proposition 2.35.43 gives a nonempty regular principal open in Spec⁡(B)\operatorname{Spec}(B). Thus BB is JJ-0 and the criterion applies to AA. For a prime P⊂R\mathfrak P\subset R and a finite purely inseparable extension E/κ(P)E/\kappa(\mathfrak P), take a finite domain CC over R/PR/\mathfrak P with fraction field EE. Realize it by Theorem 2.35.20 as B↪C=OZ,zB\hookrightarrow C=\mathcal O_{Z,z}, with BB integral affinoid. Proposition 2.35.44 gives 0≠b∈B0\ne b\in B for which C[1/b]C[1/b] is regular. Since B↪CB\hookrightarrow C, this is a nonempty principal open. Thus CC is JJ-0, and the same criterion proves that RR is JJ-2. Both AA and RR are now quasi-excellent.

It remains to prove universal catenarity. Present AA as a quotient of a generalized Tate algebra D=K{r−1T}D=K\{r^{-1}T\}. The ring DD is regular by Theorem 2.35.23. Its local rings are Cohen–Macaulay by Tag 00NQ; hence DD is universally catenary by [StacksAlg] (Tag 00NM). Quotients preserve universal catenarity by [StacksAlg] (Tag 0ECE), so AA is universally catenary. For RR, the closed immersion M(A)↪M(D)\mathcal M(A)\hookrightarrow\mathcal M(D) identifies RR with a quotient of the ordinary local ring OM(D),x\mathcal O_{\mathcal M(D),x} by the extended defining ideal; this is the finite closed-immersion case of Lemma 2.35.1. That local ring is regular by Theorem 2.35.23. The same two catenarity results therefore apply to its quotient RR. A Noetherian, universally catenary, quasi-excellent ring is excellent by definition. Nagata’s property and finite normalization now follow from Tags 07QV and 035S. Ordinary affinoid neighborhoods identify every ordinary local ring of a good space with an RR of this form. ◻

Corollary 2.35.47. NAG-001784 (Normal loci over arbitrary complete fields). Let X=M(A)X=\mathcal M(A) over an arbitrary complete field, and put p=ker⁡(x)\mathfrak p=\ker(x) for x∈Xx\in X. Then ApA_{\mathfrak p} is normal if and only if OX,x\mathcal O_{X,x} is normal. These points form a Zariski open subset of XX. Consequently the ordinary normal-point set of every good analytic space is ordinarily open.

Proof. Put D=ApD=A_{\mathfrak p} and R=OX,xR=\mathcal O_{X,x}. By Theorem 2.35.45, D→RD\to R is a faithfully flat regular local map. Its fibres are normal by [StacksAlg] (Tag 0567), so normality of DD implies normality of RR by [StacksAlg] (Tag 0C22). Conversely, if RR is normal, faithful flatness embeds DD into the domain RR. For u/v∈Frac⁡(D)u/v\in\operatorname{Frac}(D) integral over DD, normality of RR gives u/v∈Ru/v\in R, hence u∈vRu\in vR. Faithful flatness contracts vRvR to vDvD, so u/v∈Du/v\in D. This proves the equivalence.

Let NN be the product of the integral closures of the minimal-prime quotients of AA in their fraction fields. It is finite by Theorem 2.35.46. Integral closure commutes with localization, by clearing the finitely many denominators in a monic equation. Thus ApA_{\mathfrak p} is normal exactly when Ap→NpA_{\mathfrak p}\to N_{\mathfrak p} is an isomorphism: in the normal case the localization is reduced with one integral component and already integrally closed, and in the converse case it is a localization of the normal ring NN. The kernel and cokernel of A→NA\to N are finite, so their supports are closed. Their complement is precisely the normal locus in Spec⁡(A)\operatorname{Spec}(A). Its pullback by x↦ker⁡(x)x\mapsto\ker(x) is Zariski open in XX, since a principal open pulls back to ∣a(x)∣>0|a(x)|>0. The equivalence already proved identifies it with the normal-point set; ordinary affinoid neighborhoods give the assertion for every good space. ◻

References. These arguments reconstruct Ducros, Les espaces de Berkovich sont excellents, Theorem 2.13 and the ensuing normality comparison. The finite-prime field tests supply the JJ-2 condition, and universal catenarity is reduced to the checked Stacks results for regular rings and their quotients.

Proposition 2.35.48. NAG-001785 (Regular algebraic scalar extension). Let AA be a generalized KK-affinoid algebra and let L/KL/K be an analytically separable complete extension. The ordinary ring map A→ALA\to A_L is regular. Consequently, if BB is a finite-type AA-algebra, then B→B⊗AALB\to B\otimes_A A_L is regular, and regular localizations of BB have regular inverse images under this base change.

Proof. The map is flat by Theorem 2.33.6. For any p⊂A\mathfrak p\subset A and any finite purely inseparable extension E/κ(p)E/\kappa(\mathfrak p), clear denominators to find a finite domain CC over A/pA/\mathfrak p with Frac⁡(C)=E\operatorname{Frac}(C)=E. Give CC its canonical finite affinoid structure. Finite-module scalar extension identifies CLC_L with C⊗AALC\otimes_A A_L. Proposition 2.35.43 gives 0≠c∈C0\ne c\in C such that CL[1/c]C_L[1/c] is regular. Therefore AL⊗AE=CL⊗CEA_L\otimes_A E =C_L\otimes_C E is a localization of a regular ring and is regular. Tag 0381 proves geometric regularity of every algebraic fibre, giving the first assertion. Finite-type base change of regular ring maps is [StacksAlg] (Tag 07C1); its Noetherian fibre hypothesis holds since ALA_L and its finite-type algebra B⊗AALB\otimes_A A_L are Noetherian. At a prime above a regular localization of BB, apply Tag 031E to the flat local map and its regular closed fibre. This proves the last assertion. ◻

Lemma 2.35.49. NAG-001786 (Analytic domains around rigid points). Let XX be good and let x∈Xx\in X have [H(x):K]<∞[\mathcal H(x):K]<\infty. Every analytic domain V⊂XV\subset X containing xx contains an ordinary neighborhood of xx.

Proof. Choose an ordinary affinoid neighborhood UU of xx in XX, and then an affinoid domain W⊂V∩UW\subset V\cap U containing xx, using the affinoid domain basis. Choose a finite extension L/KL/K in which the finitely many minimal polynomials of generators of H(x)/K\mathcal H(x)/K split. The unique extension of the complete absolute value makes LL a complete valued field. Every residue field of H(x)⊗KL\mathcal H(x)\otimes_K L is then LL, even when the tensor product has nilpotents. Thus all points of the finite fibre of π:UL→U\pi:U_L\to U over xx are LL-rational. They all lie in WLW_L, by the domain base-change theorem. Lemma 2.35.9 gives each of them an ordinary neighborhood in ULU_L contained in WLW_L. Let OO be the union of these neighborhoods.

The space ULU_L is compact and UU is Hausdorff. Hence the continuous image π(UL∖O)\pi(U_L\setminus O) is closed in UU and does not contain xx. Its complement is an ordinary neighborhood of xx. Scalar-extension spectral surjectivity shows that this complement is contained in WW: a point outside WW has every lift outside WLW_L, and hence outside OO. The complement therefore lies in VV, proving the assertion. ◻

Proposition 2.35.50. NAG-001787 (Polynomial-to-Tate flatness over an affinoid base). For a generalized KK-affinoid algebra AA and any positive polyradius rr, the ordinary ring homomorphism A[T1,…,Tn]⟶A{r−1T}A[T_1,\ldots,T_n] \longrightarrow A\{r^{-1}T\} is flat. At a KK-rational point of the target, the induced map of local rings identifies their maximal-ideal completions.

Proof. We first prove the completion assertion, without strictness assumptions. Write A=K{s−1S}/IA=K\{s^{-1}S\}/I, and let the rational point have coordinates S=aS=a, T=bT=b. Translation by these coordinates is a bounded automorphism of the relevant Tate algebras, since ∣ai∣≤si|a_i|\le s_i and ∣bj∣≤rj|b_j|\le r_j. After translation the point is the origin. Grouping every monomial of total degree at least NN by one of the finitely many degree-NN monomials proves that reduction modulo (S,T)N(S,T)^N is precisely Taylor truncation. The grouped series remain convergent: division by a fixed monomial multiplies their weighted coefficient norms by a fixed positive constant. Consequently the completion of the localized target is K[[S,T]]/IK[[S,T]].K[[S,T]]/I K[[S,T]]. For the polynomial source, its maximal ideal is generated by the maximal ideal (S)(S) of AA together with the TjT_j. The same finite Taylor truncations, first in TT and then in SS, give exactly the same quotients by every power of that maximal ideal. Finite-module exactness of completion, Tag 00MA, passes to the quotient by II. This proves the claimed canonical isomorphism. Both local rings are Noetherian; completion is faithfully flat by Tag 00MB. The kernel argument in Lemma 2.33.32 therefore proves flatness of the local map at every rational point.

Suppose next that KK is nontrivially valued, AA is strict, and r=(1,…,1)r=(1,\ldots,1). Put P=A[T]P=A[T] and B=A{T}B=A\{T\}. For a maximal ideal m⊂B\mathfrak m\subset B, Theorem 1.12.3 gives the finite complete residue field E=B/mE=B/\mathfrak m. Evaluation defines an EE-rational maximal ideal n\mathfrak n of BEB_E, lying over m\mathfrak m. The ordinary extension B→BE=B⊗KEB\to B_E=B\otimes_K E is finite free, so Bm→(BE)nB_{\mathfrak m}\to(B_E)_{\mathfrak n} is faithfully flat. The first paragraph shows that (BE)n(B_E)_{\mathfrak n} is flat over the corresponding localization of AE[T]A_E[T]. Since AE[T]=P⊗KEA_E[T]=P\otimes_K E is flat over PP, the same kernel descent proves that BmB_{\mathfrak m} is flat over PP. Testing the kernel of J⊗PB→BJ\otimes_P B\to B at every maximal ideal of BB, for a finitely generated ideal J⊂PJ\subset P, proves that it vanishes by Tag 00HN. The finite-ideal criterion gives flatness of P→BP\to B.

Finally choose a complete extension L/KL/K containing elements whose absolute values are all the radii in a fixed presentation of AA and all the rjr_j, and one element of absolute value different from one. Such a field is obtained by completing a rational function field for the weighted Gauss absolute value, as in Lemma 3.21.7. Rescaling coordinates makes ALA_L strict and BL=AL{r−1T}B_L=A_L\{r^{-1}T\} a unit-radius Tate algebra over it. The preceding case makes BLB_L flat over AL[T]A_L[T], hence over A[T]A[T], because A→ALA\to A_L is flat. The map B→BLB\to B_L is faithfully flat by Theorem 2.33.6. Descent of the same finite-ideal kernels proves that BB is flat over A[T]A[T]. ◻

Proposition 2.35.51. NAG-001788 (Analytification over an affinoid algebra). Let AA be a generalized KK-affinoid algebra and let RR be a finite-type AA-algebra. There is a good Hausdorff analytic space ZanZ^{\mathrm{an}} over M(A)\mathcal M(A), where Z=Spec⁡(R)Z=\operatorname{Spec}(R), constructed by the radius pieces R=A[T]/I,Zr=M(A{r−1T}/I),Zan=⋃rZr.R=A[T]/I,\qquad Z_r=\mathcal M(A\{r^{-1}T\}/I), \qquad Z^{\mathrm{an}}=\bigcup_r Z_r. The construction is independent of the presentation, is compatible with finite morphisms, closed immersions, principal open subsets and complete scalar extensions, and glues on finite-type AA-schemes. If V=M(C)V=\mathcal M(C) is an affinoid domain in ZanZ^{\mathrm{an}}, the ring map R→CR\to C is flat. The map from an algebraic local ring to the corresponding ordinary analytic local ring is faithfully flat.

Proof. The radius-piece construction and its gluing are those of Theorem 2.33.23, with coefficients in AA. More explicitly, the intersection of two pieces is the piece obtained by taking the coordinatewise minimum of their radii, and is a Weierstrass domain in both. Every piece is contained in the ordinary interior of a larger piece, by strict coordinate inequalities. Hence the union is good. Two distinct points are separated either by their image in M(A)\mathcal M(A) or by the absolute value of a polynomial coordinate expression, so the union is Hausdorff. A morphism from an affinoid space over M(A)\mathcal M(A), together with an AA-algebra map RR to its analytic functions, has bounded images of the finitely many TiT_i. It therefore factors through a radius piece. The affinoid universal property gives the unique factorization. Localizing this assertion and gluing morphisms gives the same universal property for arbitrary analytic sources. It also proves presentation independence.

For a finite RR-algebra SS, its inverse image over a radius piece with algebra BB is M(B⊗RS)\mathcal M(B\otimes_R S). Indeed this is a finite BB-algebra with its canonical complete finite-module norm, and the bounded universal property identifies its maps with the maps to SS above that piece. Each of the finitely many integral generators satisfies a monic equation, so its value is bounded there; thus every algebraic valuation point above the piece is included. This proves the finite assertion, and quotient algebras give closed immersions. The principal open D(g)D(g) is the union of the Laurent domains ∣g∣≥ϵ>0|g|\ge\epsilon>0, with the function g−1g^{-1} supplied by their localization algebras. This proves the open assertion and gives the cocycle identifications required for scheme gluing. Completed tensor associativity identifies every radius piece after L/KL/K with the radius piece for R⊗AALR\otimes_A A_L, proving scalar compatibility.

For flatness, the finitely many coordinate functions are bounded on the compact affinoid domain VV. Choosing larger radii places all of VV in a single radius piece ZrZ_r, as an affinoid domain. Proposition 2.35.50, followed by the quotient by II, makes R→B=A{r−1T}/IR\to B=A\{r^{-1}T\}/I flat. Proposition 2.38.2 makes B→CB\to C flat. Their composite is flat. Ordinary stalks are filtered colimits over affinoid neighborhoods; the finite-relation argument in Theorem 2.33.34 makes their maps from RR flat. Localizing at the image prime gives a flat local map, hence a faithfully flat map. ◻

Lemma 2.35.52. NAG-001789 (Rigid completions in relative analytification). In the preceding construction, let x∈Zanx\in Z^{\mathrm{an}} be rigid over KK, let ξ∈Z\xi\in Z be its image, and let V=M(C)V=\mathcal M(C) be any affinoid domain containing xx, with m=ker⁡C(x)\mathfrak m=\ker_C(x). The natural maps identify all four complete local rings OZ,ξ^,OZan,x^,OV,x^,Cm^.\widehat{\mathcal O_{Z,\xi}},\qquad \widehat{\mathcal O_{Z^{\mathrm{an}},x}},\qquad \widehat{\mathcal O_{V,x}},\qquad \widehat{C_{\mathfrak m}}.

Proof. Assume first that xx is KK-rational. For a presentation A=K{s−1S}/JA=K\{s^{-1}S\}/J and R=A[T]/IR=A[T]/I, translate the coordinates to their values at xx. The finite Taylor calculation of Proposition 2.35.50, followed by the finite-ideal completion formula, identifies the first ring and the local completion of any radius piece containing xx with K[[S,T]]/(J,I).K[[S,T]]/(J,I). Proposition 2.35.10 gives the same ring for the ordinary analytic germ. Lemma 2.35.49 identifies the two ordinary germs, since VV contains an ordinary neighborhood of xx. Applying the rational Taylor calculation to a presentation of CC identifies its algebraic local completion with its ordinary germ completion. This proves all four comparisons in the rational case.

In general E=H(x)E=\mathcal H(x) is finite over KK. The residue fields of both RR and CC at the indicated ideals are EE: their valued fraction fields are dense in EE, while every KK-subspace of the finite-dimensional complete field EE is closed. Choose a finite extension L/KL/K splitting E/KE/K. All residue fields of E⊗KLE\otimes_K L are LL, so all points over xx are LL-rational. Tensoring any of the four complete local rings with LL decomposes it into the product of the complete local rings at these points. For the ordinary analytic rings this is Lemma 2.35.1. For the algebraic rings, the same assertion follows by decomposing the finite algebra modulo the maximal ideal into its local Artinian factors and lifting the resulting idempotents through the complete adic tower; the geometric-series correction solves an approximate idempotent modulo each next power, and Tag 00MA identifies the completed finite algebra. In each factor the maximal-ideal and original-ideal topologies agree because the residue algebra is local Artinian.

The maps in every factor are exactly the rational comparison maps already proved, since finite scalar extension commutes with the relative analytification and domain constructions. They are therefore isomorphisms. Tensoring with the finite field extension L/KL/K is faithfully flat; the kernels and cokernels of each original comparison map vanish after this tensor product and hence vanish. This descends all four canonical isomorphisms. ◻

Lemma 2.35.53. NAG-001790 (Regular localizations on relative analytic domains). Let AA be generalized affinoid, let RR be a finite-type AA-algebra, and let V=M(C)V=\mathcal M(C) be an affinoid domain in (Spec⁡R)an(\operatorname{Spec}R)^{\mathrm{an}}. For a prime q⊂C\mathfrak q\subset C, write p=q∩R\mathfrak p=\mathfrak q\cap R. If RpR_{\mathfrak p} is regular, then CqC_{\mathfrak q} is regular.

Proof. Choose an analytic point v∈Vv\in V with kernel q\mathfrak q, by Lemma 2.52.1. Theorem 2.35.46 makes AA excellent, so its JJ-2 property says that the regular locus of Spec⁡(R)\operatorname{Spec}(R) is open. Choose g∉pg\notin\mathfrak p such that R[1/g]R[1/g] is regular. Choose a nontrivially valued free extension L=KrL=K_r making both ALA_L and CLC_L strict, and lift vv to a point ww of VLV_L. Lemma 2.35.39 and Proposition 2.35.48 show that RL[1/g]R_L[1/g], where RL=R⊗AALR_L=R\otimes_A A_L, is regular.

By Lemma 2.35.25, take a strict affinoid ordinary neighborhood W=M(D)W=\mathcal M(D) of ww in the open set VL∩D(g)anV_L\cap D(g)^{\mathrm{an}}. For every maximal ideal n⊂D\mathfrak n\subset D, the strict point theorem gives an LL-rigid point t∈Wt\in W with that kernel. Lemma 2.35.52 identifies Dn^\widehat{D_{\mathfrak n}} with the completion of the local ring of Spec⁡(RL[1/g])\operatorname{Spec}(R_L[1/g]) at the image of tt. This algebraic local ring is regular. The regularity/completion criterion, Tag 07NY, shows first that its completion is regular and then that DnD_{\mathfrak n} is regular. Since every prime lies under a maximal ideal and localization preserves regularity, DD is regular.

The composite C→CL→DC\to C_L\to D is flat by scalar flatness and domain flatness. At the kernel nw\mathfrak n_w of ww on DD, its contraction to CC is q\mathfrak q. Thus Cq→DnwC_{\mathfrak q}\to D_{\mathfrak n_w} is a flat local map with regular target. Tag 00OF descends regularity and proves the assertion. ◻

Theorem 2.35.54. NAG-001791 (Regular comparison over an affinoid base). Let RR be a finite-type algebra over a generalized affinoid algebra AA, let Z=Spec⁡(R)Z=\operatorname{Spec}(R), and let V=M(C)V=\mathcal M(C) be an affinoid domain in ZanZ^{\mathrm{an}}. For x∈Vx\in V, put ξ=ker⁡R(x)\xi=\ker_R(x) and c=ker⁡C(x)\mathfrak c=\ker_C(x). Every arrow in the following ring diagram is regular: Rξ⟶Cc↓↓OZan,x⟶OV,x,Rξ⟶OV,x.\begin{array}{ccc} R_{\xi}&\longrightarrow&C_{\mathfrak c}\\ \big\downarrow&&\big\downarrow\\ \mathcal O_{Z^{\mathrm{an}},x}&\longrightarrow&\mathcal O_{V,x}, \end{array} \qquad R_{\xi}\longrightarrow\mathcal O_{V,x}.

Proof. All four rings are Noetherian. The relative analytification construction proves flatness of the upper and left arrows, and the right arrow is regular by Theorem 2.35.45. The bottom arrow is a flat local map by analytic-domain stalk flatness. It remains to verify geometric regularity; throughout we test the residue field at every algebraic prime, not only primes supplied by analytic fibres.

First consider the diagonal map Rξ→S=OV,xR_{\xi}\to S=\mathcal O_{V,x}. Choose p⊂ξ\mathfrak p\subset\xi in RR and a finite purely inseparable extension E/Frac⁡(R/p)E/\operatorname{Frac}(R/\mathfrak p). Clearing denominators in the power equations gives a finite radicial domain BB over R/pR/\mathfrak p with fraction field EE. The finite analytification of Spec⁡(B)→Z\operatorname{Spec}(B)\to Z has a unique point yy above xx; the inverse image WW of VV is affinoid, and finite branch separation gives S⊗RE=OW,y⊗BE.S\otimes_R E =\mathcal O_{W,y}\otimes_B E. Write W=M(D)W=\mathcal M(D) and let Q\mathfrak Q be any prime of OW,y\mathcal O_{W,y} surviving this localization. Its contraction to BB is zero. Its contraction q\mathfrak q to DD therefore lies above the regular generic point of the integral scheme Spec⁡(B)\operatorname{Spec}(B). Lemma 2.35.53 proves that DqD_{\mathfrak q} is regular. The regular map from DD to the ordinary local ring at yy, localized at Q\mathfrak Q, has regular source and regular closed fibre. Tag 031E shows that (OW,y)Q(\mathcal O_{W,y})_{\mathfrak Q} is regular. Thus the displayed EE-fibre is regular at every prime. Tag 0381 proves geometric regularity of all fibres of the diagonal map.

Next consider the bottom arrow. Replace ZanZ^{\mathrm{an}} by an ordinary affinoid neighborhood X0=M(A0)X_0=\mathcal M(A_0) of xx, and replace VV by an ordinary affinoid neighborhood of xx inside V∩X0V\cap X_0. These replacements do not change either ordinary stalk. Put R0=OX0,xR_0=\mathcal O_{X_0,x}. For a prime P⊂R0\mathfrak P\subset R_0 and a finite purely inseparable extension E/κ(P)E/\kappa(\mathfrak P), Proposition 2.35.22, applied to the domain inclusion, realizes the test as S⊗R0E=(∏iOT,ti)⊗C0E,B0↪C0=OY,z,Y=M(B0),S\otimes_{R_0}E =\left(\prod_i\mathcal O_{T,t_i}\right)\otimes_{C_0}E, \qquad B_0\hookrightarrow C_0=\mathcal O_{Y,z}, \qquad Y=\mathcal M(B_0), where B0B_0 and C0C_0 are domains, C0C_0 is finite over R0R_0, Frac⁡(C0)=E\operatorname{Frac}(C_0)=E, and TT is the inverse-image domain of VV in YY. A prime in any factor surviving the displayed localization contracts to zero in B0B_0. The diagonal comparison already proved, now with the affinoid base B0B_0 and its finite-type algebra B0B_0 itself, makes B0→OT,tiB_0\to\mathcal O_{T,t_i} regular at the relevant localizations. Their localizations at primes above zero are regular, because the source localization is the field Frac⁡(B0)\operatorname{Frac}(B_0) and Tag 031E applies. Hence every local ring of S⊗R0ES\otimes_{R_0}E is regular. Tag 0381 proves that the bottom arrow is regular.

The upper arrow is regular by descent, Tag 07NT: its composite with the faithfully flat right arrow is the regular diagonal. Finally, the diagonal assertion applied to an ordinary affinoid neighborhood of xx in ZanZ^{\mathrm{an}} proves regularity of the left arrow. This establishes every map of the diagram. ◻

Corollary 2.35.55. NAG-001792 (Regular ordinary local maps of analytic domains). If VV is a good analytic domain in a good analytic space XX, then OX,x→OV,x\mathcal O_{X,x}\to\mathcal O_{V,x} is regular for every x∈Vx\in V. The same holds after every complete extension and every analytic base change for which the spaces under consideration are good.

Proof. Restrict XX to an ordinary affinoid neighborhood of xx, and VV to an ordinary affinoid neighborhood of xx inside its inverse image. For the corresponding affinoid algebra AA, apply the bottom-arrow assertion of Theorem 2.35.54 to R=AR=A. Domain base change remains an analytic domain, so the same argument applies after each of the stated base changes. ◻

References. The preceding three comparisons reconstruct Ducros, Les espaces de Berkovich sont excellents, Lemma 3.2 and Theorem 3.3. The rigid-completion input and the finite-type-over-affinoid analytification used there have been supplied explicitly above.

Proposition 2.35.56. NAG-001793 (Ordinary regularity of relative affine space). Let XX be good and let YY be a good analytic domain in AXn,an\mathbb A_X^{n,\mathrm{an}}. At every y∈Yy\in Y, with image x∈Xx\in X, the ordinary local homomorphism OX,x→OY,y\mathcal O_{X,x}\to\mathcal O_{Y,y} is regular.

Proof. Flatness follows from relative polydisc flatness and analytic-domain flatness, since one can choose a polydisc ordinary neighborhood of yy with radii larger than its coordinate values. Put R=OX,xR=\mathcal O_{X,x} and S=OY,yS=\mathcal O_{Y,y}. For every prime p⊂R\mathfrak p\subset R and finite purely inseparable extension E/κ(p)E/\kappa(\mathfrak p), apply Proposition 2.35.22. After ordinary restrictions it supplies an integral affinoid space Z=M(B)Z=\mathcal M(B), an inclusion B↪C=OZ,zB\hookrightarrow C=\mathcal O_{Z,z} with Frac⁡(C)=E\operatorname{Frac}(C)=E, and an inverse-image domain TT of YY in AZn,an\mathbb A_Z^{n,\mathrm{an}} such that S⊗RE=(∏iOT,ti)⊗CE.S\otimes_R E =\left(\prod_i\mathcal O_{T,t_i}\right) \otimes_C E. Fix any prime Q\mathfrak Q in a factor surviving this localization. It contracts to zero in BB. On an ordinary affinoid neighborhood of tit_i in TT, the polynomial coordinates define a ring map B[T1,…,Tn]→OT,tiB[T_1,\ldots,T_n]\to\mathcal O_{T,t_i}. Let q\mathfrak q be the contraction of Q\mathfrak Q to this polynomial ring. Since q∩B=0\mathfrak q\cap B=0, its localization B[T]qB[T]_{\mathfrak q} is a localization of a polynomial algebra over Frac⁡(B)\operatorname{Frac}(B), and is regular. Theorem 2.35.54 makes its map to (OT,ti)Q(\mathcal O_{T,t_i})_{\mathfrak Q} regular. Tag 031E proves that this target is regular. Every prime of the displayed EE-fibre has been treated. Tag 0381 therefore gives geometric regularity of every algebraic fibre of R→SR\to S. ◻

Lemma 2.35.57. NAG-001794 (Ordinary local rings of étale morphisms). For an étale morphism of good analytic spaces, every induced ordinary local homomorphism is regular.

Proof. Use the ordinary finite neighborhoods in the definition of étaleness. For the corresponding finite étale affinoid algebra A→BA\to B and R=OX,xR=\mathcal O_{X,x}, Lemma 2.35.1 expresses each source stalk as a local factor of B⊗ARB\otimes_A R. Finite flatness and vanishing of algebraic relative differentials survive this base change and passage to a factor, as in Proposition 2.50.2. Hence the local map is flat. At every algebraic prime of RR, its fibre is a factor of a finite étale algebra over the residue field. After any field extension it remains finite étale: finite flatness and the standard étale presentations are preserved by tensor product. A finite étale algebra over a field is a product of finite separable fields. Indeed its standard presentations have an invertible polynomial derivative, so after an algebraic closure all their local factors are copies of that field, with no nilpotents; descent gives the asserted finite separable factors. Thus all the algebraic fibres and all their finite field extensions are regular. This proves geometric regularity. ◻

NAG-001795 · Open at full scope

Theorem 2.35.58. NAG-001795 (Ordinary regularity of quasi-smooth morphisms). Let f:Y→Xf:Y\to X be a morphism of good analytic spaces over any complete non-Archimedean field, quasi-smooth at yy, and put x=f(y)x=f(y). Then OX,x→OY,y\mathcal O_{X,x}\to\mathcal O_{Y,y} is a regular homomorphism of Noetherian local rings. If X=M(A)X=\mathcal M(A) and Y=M(B)Y=\mathcal M(B) are affinoid, then Aker⁡(x)→Bker⁡(y)A_{\ker(x)}\to B_{\ker(y)} is regular as well.

Proof. First suppose that ff is smooth. Locally on ordinary neighborhoods it factors as an étale map to an ordinary open subset of AXd,an\mathbb A_X^{d,\mathrm{an}} followed by projection. Proposition 2.35.56 and Lemma 2.35.57 make the two induced ordinary local homomorphisms regular. Their composite is regular by Tag 07QI; all target rings and fibres are Noetherian.

In the quasi-smooth case, Theorem 2.59.4 identifies an ordinary neighborhood VV of yy in YY with an analytic domain in a smooth XX-space TT, after an ordinary restriction around xx. The map on ordinary local rings factors as OX,x⟶OT,y⟶OV,y=OY,y.\mathcal O_{X,x}\longrightarrow \mathcal O_{T,y}\longrightarrow \mathcal O_{V,y}=\mathcal O_{Y,y}. The first arrow is regular by the smooth case; the second is regular by Corollary 2.35.55. Tag 07QI again proves regularity of the composite. This reasoning applies in every characteristic and tests all algebraic fibres through the preceding propositions.

For the affinoid assertion, put p=ker⁡(x)\mathfrak p=\ker(x) and q=ker⁡(y)\mathfrak q=\ker(y). Theorem 2.35.45 and the first assertion make the composite Ap→OX,x→OY,yA_{\mathfrak p}\to\mathcal O_{X,x}\to\mathcal O_{Y,y} regular. It factors through BqB_{\mathfrak q}, and the latter map Bq→OY,yB_{\mathfrak q}\to\mathcal O_{Y,y} is faithfully flat. Descent of regular ring maps, Tag 07NT, proves that Ap→BqA_{\mathfrak p}\to B_{\mathfrak q} is regular. ◻

NAG-001796 · Open at full scope

Corollary 2.35.59. NAG-001796 (Normality and regularity along quasi-smooth maps). Under the hypotheses of the preceding theorem, OY,y\mathcal O_{Y,y} is normal if and only if OX,x\mathcal O_{X,x} is normal, and it is regular if and only if OX,x\mathcal O_{X,x} is regular. In particular a quasi-smooth good space over a normal good analytic space is normal. These assertions hold after every analytic base change and complete extension for which the spaces are good.

Proof. The local map is regular and hence faithfully flat. Regularity ascends by Tag 031E and descends by Tag 00OF. Its geometrically regular fibres are geometrically normal. Normality therefore ascends by Tags 0C22 and 0567. For descent, if the target normal local ring SS is a domain, faithful flatness embeds the source RR into SS, so RR is a domain. If a/b∈Frac⁡(R)a/b\in\operatorname{Frac}(R) is integral over RR, its image is integral over SS and lies in SS. Thus a∈bS∩R=bRa\in bS\cap R=bR, where the equality follows by faithful-flat contraction of the ideal bRbR. Hence a/b∈Ra/b\in R, proving normality. Quasi-smoothness and good-domain restrictions are stable under the stated base changes, so the theorem applies again after each of them. ◻

References. The ordinary-ring argument reconstructs the regular-fibre comparison of [Duc18] (5.5.1–5.5.5). It uses the smooth-envelope factorization together with finite algebraic-prime tests, so its geometric regularity concerns the ordinary local rings, not merely the analytic fibre spaces.

Normality ascends along smooth maps. Let f:Y→Xf:Y\to X be a morphism of good analytic spaces over a complete non-Archimedean field, smooth at yy, and let x=f(y)x=f(y). If OX,x\mathcal O_{X,x} is normal, then OY,y\mathcal O_{Y,y} is normal.

Use the ordinary open étale factorization of a smooth morphism through a relative analytic affine space. On the actual local rings, Proposition 2.35.56 makes the projection regular and Lemma 2.35.57 makes the étale arrow regular. The composite is regular by [StacksAlg] (Tag 07QI); all of its fibre rings are Noetherian since the target ordinary local ring is Noetherian. In particular the composite is flat and its ordinary algebraic fibres are regular, hence normal by [StacksAlg] (Tag 0567). Apply [StacksAlg] (Tag 0C22) to the actual Noetherian local rings to obtain normality.

2.36 Rational refinement over the original field

The strict Gerritzen–Grauert theorem and the free-extension calculations now give rational refinements for general affinoid domains. The essential step is to describe the inverse image of a rational set under the Gauss section. Its defining functions have Laurent expansions, and a unit-ideal identity permits those expansions to be truncated uniformly. Throughout this section KK is complete with its fixed real-valued non-Archimedean absolute value, which may be trivial. Affinoid algebras carry norm-admissible Banach norms, and the domain property of Definition 2.1.1 is tested against all generalized affinoid algebras over KK.

Lemma 2.36.1. NAG-001838 (Affinoid tests over a free extension). Let rr be a finite tuple free over KK, and put L=KrL=K_r. Every norm-admissible LL-affinoid algebra CC is a norm-admissible generalized KK-affinoid algebra. Its bounded multiplicative seminorms over KK are exactly those over LL: MK(C)=ML(C).\mathcal M_K(C)=\mathcal M_L(C).

Proof. Theorem 1.17.2 presents LL as a quotient of a generalized Tate algebra over KK. Apply Proposition 1.17.3 to a presentation C=L{s−1Z}/JC=L\{s^{-1}Z\}/J. It presents L{s−1Z}L\{s^{-1}Z\} over KK, and its further quotient by the closed ideal JJ is again norm-admissible over KK. Uniformly equivalent admissible norms give the same conclusion.

A bounded multiplicative seminorm zz on CC restricts along the bounded map L→CL\to C to a bounded multiplicative seminorm on the valued field LL. For a∈L×a\in L^\times, a bound DD gives ∣a∣zn≤D∣a∣n|a|_z^n\leq D|a|^n for all nn, hence ∣a∣z≤∣a∣|a|_z\leq|a|. Applying the same argument to a−1a^{-1} gives the reverse inequality. Thus zz extends the specified absolute value of LL. Conversely an LL-point is a KK-point. Both spectra have the topology generated by evaluation on the same algebra CC, so the identification is topological as well. If C=0C=0, both spaces are empty. ◻

Lemma 2.36.2. NAG-001839 (Spectral injectivity of strict epimorphisms). Let LL be complete and nontrivially valued. A bounded categorical epimorphism D→ED\to E of strict LL-affinoid algebras induces an injective map M(E)→M(D)\mathcal M(E)\to\mathcal M(D).

Proof. Theorem 2.27.3 supplies a finite rational covering (Wj)(W_j) of M(D)\mathcal M(D) such that Dj→Ej=E⊗^DDjD_j\to E_j=E\widehat\otimes_DD_j has dense image. Suppose z1,z2∈M(E)z_1,z_2\in\mathcal M(E) have the same image xx, and choose jj with x∈Wjx\in W_j. By Proposition 1.14.3, EjE_j is the actual strict rational quotient over EE for the pulled tuple. The strict rational point comparison, Theorem 2.19.3, lifts each ziz_i uniquely to a point of EjE_j. Their restrictions to DjD_j are both the unique rational lift of xx. Thus these two seminorms agree on the dense image of DjD_j. Bounded multiplicative seminorms are continuous, by Lemma 1.6.1, so they agree on EjE_j, and hence z1=z2z_1=z_2. The claim is immediate if the source spectrum is empty. ◻

Proposition 2.36.3. NAG-001840 (Strictification of a domain presentation). Let A→BA\to B represent an affinoid domain U⊆MK(A)U\subseteq\mathcal M_K(A). There is a finite free tuple rr such that L=KrL=K_r is nontrivially valued, Ar=A⊗^KLA_r=A\widehat\otimes_K L and Br=B⊗^KLB_r=B\widehat\otimes_K L are strict LL-affinoid, and Ar→BrA_r\to B_r is a strict-test affinoid-domain presentation of its spectral image UrU_r. Writing φr:ML(Br)→ML(Ar)\varphi_r:\mathcal M_L(B_r)\to\mathcal M_L(A_r) and φ:MK(B)→MK(A)\varphi:\mathcal M_K(B)\to\mathcal M_K(A), one has φr∘σr,B=σr,A∘φ,σr,A−1(Ur)=U.\varphi_r\circ\sigma_{r,B}=\sigma_{r,A}\circ\varphi, \qquad \sigma_{r,A}^{-1}(U_r)=U.

Proof. Choose rr by Theorem 1.17.6 for presentations of AA and BB. If KK is trivially valued and the chosen tuple is empty, append one radius greater than one. It is free over KK, and the original presentation radii remain in ∣L×∣\sqrt{|L^\times|}, so the same strictification applies. All these algebras are also generalized KK-affinoid by Proposition 1.17.4.

The map Ar→BrA_r\to B_r is an epimorphism among strict LL-affinoid algebras. Indeed, two bounded LL-maps Br→CB_r\to C agreeing on ArA_r restrict to maps B→CB\to C agreeing on AA. Lemma 2.36.1 makes CC a permitted generalized KK-affinoid test, and both restrictions factor the same map with spectral image in UU. Domain uniqueness makes them equal. The original BB and the scalar field LL generate a dense subalgebra of BrB_r, so the two maps agree on BrB_r. Lemma 2.36.2 now gives injectivity on spectra. Compactness and Hausdorffness, Proposition 2.24.1, identify ML(Br)\mathcal M_L(B_r) homeomorphically with UrU_r.

Suppose a bounded LL-map Ar→CA_r\to C, with CC strict over LL, has spectral image in UrU_r. Its restriction A→CA\to C has image in UU, since restricting any point of BrB_r to BB gives a point over UU. By the same lemma, the original domain property provides a unique bounded map B→CB\to C. Together with the given scalar map L→CL\to C, its bounded balanced product extends to B⊗^KL→CB\widehat\otimes_K L\to C. This is the required LL-factorization; its uniqueness follows by restriction to BB and LL and density. Thus Ar→BrA_r\to B_r has the strict-test domain property.

The first displayed identity is the naturality in Theorem 1.19.1. It implies σr,A(U)⊆Ur\sigma_{r,A}(U)\subseteq U_r. Conversely the projection of UrU_r to MK(A)\mathcal M_K(A) is contained in UU, and πrσr,A\pi_r\sigma_{r,A} is the identity. These two inclusions prove the second displayed equality. ◻

Lemma 2.36.4. NAG-001841 (Rational inverse images under a Gauss section). Let AA be a generalized KK-affinoid algebra, let rr be a finite free tuple, and put Ar=A⊗^KKrA_r=A\widehat\otimes_K K_r. For every rational domain R′R' of MKr(Ar)\mathcal M_{K_r}(A_r), the subset σr,A−1(R′)\sigma_{r,A}^{-1}(R') is a finite union of rational domains defined over the original algebra AA, with positive real radii.

Proof. The zero algebra gives the empty case. Otherwise choose a rational presentation R′={z:∣fi(z)∣≤ρi∣g(z)∣ (1≤i≤m)},h0g+∑i=1mhifi=1R'=\{z:|f_i(z)|\leq\rho_i|g(z)|\ (1\leq i\leq m)\}, \qquad h_0g+\sum_{i=1}^m h_if_i=1 in ArA_r, with ρi>0\rho_i>0. Use the coefficient norm of Theorem 1.16.2 and set C=max⁡(∥h0∥,ρ1∥h1∥,…,ρm∥hm∥)>0,δ=C−1.C=\max\bigl(\|h_0\|,\rho_1\|h_1\|,\ldots,\rho_m\|h_m\|\bigr)>0, \qquad \delta=C^{-1}. For x∈σr,A−1(R′)x\in\sigma_{r,A}^{-1}(R'), evaluating the unit identity gives 1≤C σr,A(x)(g),σr,A(x)(g)≥δ.1\leq C\,\sigma_{r,A}(x)(g), \qquad \sigma_{r,A}(x)(g)\geq\delta. Write g=∑νgνTνg=\sum_{\nu}g_\nu T^\nu and fi=∑νfiνTνf_i=\sum_\nu f_{i\nu}T^\nu. Choose finite index sets G,Fi⊂ZsG,F_i\subset\mathbf Z^s such that the omitted coefficient weights satisfy ∥gν∥rν<δ/2(ν∉G),∥fiν∥rν<ρiδ/2(ν∉Fi).\|g_\nu\|r^\nu<\delta/2\quad(\nu\notin G),\qquad \|f_{i\nu}\|r^\nu<\rho_i\delta/2\quad(\nu\notin F_i). Such sets exist by coefficient convergence. For each ν∈G\nu\in G, define Rν⊆MK(A)R_\nu\subseteq\mathcal M_K(A) by δ≤rν∣gν(x)∣,rμ∣gμ(x)∣≤rν∣gν(x)∣(μ∈G),rμ∣fiμ(x)∣≤ρirν∣gν(x)∣(1≤i≤m, μ∈Fi).\begin{split} \delta&\leq r^\nu|g_\nu(x)|,\\ r^\mu|g_\mu(x)|&\leq r^\nu|g_\nu(x)|\quad(\mu\in G),\\ r^\mu|f_{i\mu}(x)|&\leq\rho_i r^\nu|g_\nu(x)| \quad(1\leq i\leq m,\ \mu\in F_i). \end{split} This is an original-AA rational presentation with denominator gνg_\nu: its numerators are 11, the gμg_\mu, and the fiμf_{i\mu}, with respective radii rν/δr^\nu/\delta, rν−μr^{\nu-\mu}, and ρirν−μ\rho_i r^{\nu-\mu}. The numerator 11 guarantees the unit-ideal condition. Empty members, including those with gν=0g_\nu=0, are allowed.

At a point of σr,A−1(R′)\sigma_{r,A}^{-1}(R'), the denominator value is at least δ\delta. The omitted gg-coefficients have smaller value, so some retained coefficient gνg_\nu attains its Gauss maximum. The rational inequalities and the same Gauss formula then put the point in RνR_\nu. Conversely, on RνR_\nu, the first two conditions and the omitted-coefficient bound give σr,A(x)(g)=rν∣gν(x)∣≥δ\sigma_{r,A}(x)(g)=r^\nu|g_\nu(x)|\geq\delta. The retained fif_i-coefficients satisfy the required bound by the third condition, while the omitted coefficients are smaller than ρiδ\rho_i\delta. Therefore σr,A(x)(fi)≤ρiσr,A(x)(g)\sigma_{r,A}(x)(f_i)\leq\rho_i\sigma_{r,A}(x)(g) for every ii. This proves the exact set equality σr,A−1(R′)=⋃ν∈GRν.\sigma_{r,A}^{-1}(R')=\bigcup_{\nu\in G}R_\nu. The proof also allows an empty tuple rr, zero retained coefficients and an empty inverse image. ◻

Lemma 2.36.5. NAG-001842 (Restriction to a rational subdomain). Let A→BA\to B represent an affinoid domain UU. If A→CA\to C represents an affinoid domain V⊆UV\subseteq U, there is a unique bounded AA-map B→CB\to C, representing VV as a domain of M(B)\mathcal M(B). These restriction maps compose, and compositions of domain embeddings are domain embeddings. If V=RV=R is rational over AA, with algebra ARA_R, its defining tuple in BB presents a rational domain with the same algebra, and B⊗^AAR≃ARB\widehat\otimes_A A_R\simeq A_R by canonical bounded inverse maps. These identifications commute with restriction to finite intersections of original-AA rational subsets contained in UU.

Proof. Test the domain property of A→BA\to B on CC. It supplies the unique map B→CB\to C, whose spectral map is V⊆UV\subseteq U. For a generalized affinoid BB-test EE with image in VV, the AA-domain property of CC supplies C→EC\to E. This is a BB-map because its two maps from BB agree over AA and the uniqueness for UU applies to EE. The same uniqueness proves composition of restrictions. Factoring a test map successively proves transitivity of domains.

For RR, write its actual rational algebra as AR=A{ρ−1T}/(gTi−fi)A_R=A\{\rho^{-1}T\}/(gT_i-f_i). The pulled-back tuple defines the actual quotient D=B{ρ−1T}/(φ(g)Ti−φ(fi))D=B\{\rho^{-1}T\}/(\varphi(g)T_i-\varphi(f_i)). It is affinoid, and its spectrum is the inverse image of RR in M(B)\mathcal M(B), by Theorem 2.31.1. The two bounded rational universal properties identify DD with the completed pushout B⊗^AARB\widehat\otimes_A A_R: the variables have the prescribed powers and images, and their finite polynomials are dense.

The restriction B→ARB\to A_R and the identity on ARA_R induce D→ARD\to A_R, inverse to the canonical map AR→DA_R\to D. One composite fixes ARA_R. For the other, its two maps from BB agree over AA and the spectral image of DD is contained in R⊆UR\subseteq U; hence domain uniqueness makes them equal. The pushout property gives equality on DD. The same argument on an intersection, which is original-AA rational by Proposition 2.31.3, proves compatibility. Every comparison is bounded in both directions. ◻

Theorem 2.36.6. NAG-001843 (Generalized Gerritzen–Grauert). Every affinoid domain U⊆MK(A)U\subseteq\mathcal M_K(A) has a finite covering by rational domains defined over the original AA, with positive real radii. If BB represents UU, each member RiR_i and each finite intersection RR of these members is rational over BB with algebra canonically boundedly isomorphic to ARA_R. In particular B⊗^AAR≃AR,B\widehat\otimes_A A_R\simeq A_R, compatibly with all restriction maps. The empty domain admits the empty covering and has algebra zero.

Proof. If UU is empty, the point identification and Theorem 1.20.1 give B=0B=0. Otherwise choose the simultaneous free strictification of Proposition 2.36.3. Corollary 2.29.4, over L=KrL=K_r, gives a finite covering Ur=⋃jRj′U_r=\bigcup_j R'_j, where the Rj′R'_j are rational in ML(Ar)\mathcal M_L(A_r). By that proposition and Lemma 2.36.4, U=σr,A−1(Ur)=⋃jσr,A−1(Rj′),U=\sigma_{r,A}^{-1}(U_r) =\bigcup_j\sigma_{r,A}^{-1}(R'_j), and each inverse image is a finite union of original-AA rational domains. Their combined family is the required finite covering. Lemma 2.36.5 supplies all algebra identifications, their bounded inverses and their compatibility on intersections. ◻

Theorem 2.36.7. NAG-001844 (Complete-extension tests for an affinoid domain). Let A→BA\to B represent an affinoid domain UU using generalized tests over KK. For every complete isometric extension L/KL/K and every norm-admissible LL-affinoid algebra CC, a bounded KK-map α:A→C\alpha:A\to C whose spectral image lies in UU has a unique bounded KK-factorization through BB. Thus the same-KK generalized domain property and the domain property using all complete-extension affinoid tests are equivalent, with the same representing algebra.

Proof. Choose the finite original-AA rational covering U=⋃iRiU=\bigcup_iR_i from Theorem 2.36.6, and write Bi=ARiB_i=A_{R_i} and Bij=ARi∩RjB_{ij}=A_{R_i\cap R_j}. They form a rational covering of M(B)\mathcal M(B) by Lemma 2.36.5. Pull the original rational tuples into CC. Their actual quotient algebras CiC_i and CijC_{ij} form a finite rational covering of ML(C)\mathcal M_L(C) and its intersections, by Theorem 2.31.1 and Proposition 2.31.3. The covering assertion follows from the spectral-image hypothesis.

Theorem 2.31.2, with its complete-extension test clause, gives bounded maps Bi→CiB_i\to C_i. After restricting to CijC_{ij}, the same theorem extends each map through the rational overlap algebra BijB_{ij}. The two resulting maps Bij=ARi∩Rj→CijB_{ij}=A_{R_i\cap R_j}\to C_{ij} agree: they extend the same map from AA and send every rational variable to its prescribed ratio. Composing with B→BiB\to B_i consequently gives a bounded map from BB to the matching kernel of ∏iCi⇉∏i,jCij\prod_i C_i\rightrightarrows\prod_{i,j}C_{ij}. Degree-zero norm-strict exactness in Theorem 2.32.2, applied over LL to the finite CC-module CC, identifies this kernel with CC, with bounded inverse. The resulting map B→CB\to C is bounded, preserves multiplication and the unit, and extends α\alpha, because these assertions hold after the injective restriction into ∏iCi\prod_iC_i.

For uniqueness, let β:B→C\beta:B\to C be any bounded factorization. After restricting to CiC_i, its spectral image in MK(B)\mathcal M_K(B) lies over RiR_i, hence in the rational BB-domain represented by BiB_i. The same rational universal property factors B→CiB\to C_i through BiB_i. This map Bi=ARi→CiB_i=A_{R_i}\to C_i is the unique map determined by α\alpha and the original rational ratios. It is therefore the map used in the construction. Injectivity of C→∏iCiC\to\prod_iC_i proves uniqueness. If the spectral image is empty, Theorem 1.20.1 over LL gives C=0C=0, for which the factorization is unique. The converse implication follows by taking L=KL=K. ◻

Corollary 2.36.8. NAG-001845 (Comparing tests for strict presentations). Suppose A,BA,B are strict norm-admissible KK-affinoid algebras, with trivial valuation on KK allowed. A bounded map A→BA\to B is a strict-test affinoid-domain presentation if and only if it is a generalized same-KK affinoid-domain presentation, and if and only if it has the affinoid-domain property for all complete-extension affinoid tests. The represented subset and algebra are unchanged.

Proof. Only the implication from strict tests to all complete-extension tests requires proof. Let UU be the represented subset. Corollary 2.29.4 gives a finite covering U=⋃iRiU=\bigcup_iR_i by original-AA rational domains with unit-radius presentations. Lemma 2.30.1 identifies each ARiA_{R_i} and each intersection algebra as an actual rational quotient of BB. Thus we have exactly the rational-cover data used in the proof of Theorem 2.36.7, without first enlarging the test class. For a bounded A→CA\to C into any complete-extension affinoid whose spectral image lies in UU, pull back these tuples. The rational universal property gives ARi→CiA_{R_i}\to C_i and the maps on their overlaps. Their composites from BB match. The norm-strict rational Tate equalizer glues them to a bounded B→CB\to C. Any other factorization has the same restrictions, by uniqueness for the actual rational quotients of BB; injectivity of the equalizer therefore gives equality. This proves the implication. The other implications restrict the test class. ◻

Theorem 2.36.9. NAG-001846 (Affinoid domains under arbitrary affinoid base change). Let A→BA\to B represent a generalized KK-affinoid domain UU. Let L/KL/K be any complete isometric extension, let CC be a generalized norm-admissible LL-affinoid algebra, and let α:A→C\alpha:A\to C be bounded over KK. Write p:ML(C)→MK(A)p:\mathcal M_L(C)\to\mathcal M_K(A). Then D=B⊗^ACD=B\widehat\otimes_A C is norm-admissible LL-affinoid and C→DC\to D represents the affinoid domain p−1(U)p^{-1}(U). For every original-AA rational domain R⊆UR\subseteq U, its pullback to ML(D)\mathcal M_L(D) is rational and has algebra D⊗^BAR≃C⊗^AAR,D\widehat\otimes_B A_R \simeq C\widehat\otimes_A A_R, the actual rational quotient of CC defined by the pulled-back tuple. These are canonical bounded isomorphisms and commute with finite intersections, restriction and further complete-affinoid base change. Finite affinoid-domain coverings pull back to finite affinoid-domain coverings.

Proof. Proposition 2.33.8 gives the actual generalized LL-affinoid algebra DD and its bounded pushout property. Choose U=⋃iRiU=\bigcup_iR_i by Theorem 2.36.6. Each RiR_i is rational over BB with algebra ARiA_{R_i} by Lemma 2.36.5. Pull its tuple into DD. By the rational point theorem these rational subsets cover ML(D)\mathcal M_L(D), since the RiR_i cover MK(B)\mathcal M_K(B). Their actual algebras have the displayed bounded identifications: compatible bounded maps out of either side are precisely compatible maps from CC and ARiA_{R_i} over AA, because the intervening map from BB is the restriction B→ARiB\to A_{R_i}. The rational universal properties, or the actual joint quotient presentations, therefore give bounded inverse maps.

The right-hand algebra is the rational quotient of CC for p−1(Ri)p^{-1}(R_i). Every point of DD maps into p−1(U)p^{-1}(U) by restricting to BB. Conversely, for x∈p−1(U)x\in p^{-1}(U) choose ii with p(x)∈Rip(x)\in R_i. Theorem 2.31.1 gives a unique point above xx in this rational quotient of CC, hence a point of DD. If two points of DD lie over the same xx, both belong to that iith rational piece; its identification with the rational quotient of CC makes them equal. Thus ML(D)→ML(C)\mathcal M_L(D)\to\mathcal M_L(C) is bijective onto p−1(U)p^{-1}(U); compactness and Hausdorffness in Proposition 2.24.1 make this a homeomorphism.

For the domain property, let C→EC\to E be a bounded LL-map to a generalized LL-affinoid algebra whose spectral image lies in p−1(U)p^{-1}(U). The composite A→EA\to E has image in UU. Theorem 2.36.7 supplies a unique bounded KK-map B→EB\to E over AA. Together with C→EC\to E, the pushout gives the required bounded LL-map D→ED\to E. Conversely any such factorization restricts to these two maps, so it is unique. This proves the generalized LL-domain property.

The algebra identification for any rational R⊆UR\subseteq U follows by the same pushout argument. Uniqueness of the bounded factorization makes these identifications compatible with restriction, intersections and a further base change. For successive base changes, the two parenthesizations of the completed pushout represent the same bounded-map problem, so their canonical maps are bounded inverses. A finite domain covering remains a covering because inverse images preserve finite unions. Empty pullbacks have zero algebra by Theorem 1.20.1 over LL. ◻

Corollary 2.36.10. NAG-001847 (Intersections and restrictions of affinoid domains). For affinoid domains U,V⊆MK(A)U,V\subseteq\mathcal M_K(A) represented by B,CB,C, their intersection is affinoid and is represented by B⊗^ACB\widehat\otimes_A C. All finite intersections have the corresponding iterated completed pushout algebras, independently of parenthesization. If U⊆VU\subseteq V, the canonical restriction C→BC\to B represents an affinoid domain and B⊗^AC≃B.B\widehat\otimes_A C\simeq B. All these identifications are canonical and bounded in both directions.

Proof. Apply Theorem 2.36.9 with L=KL=K to the domain UU and the map A→CA\to C representing VV. Its inverse image is U∩VU\cap V, viewed as a domain of VV. Domain transitivity in Lemma 2.36.5 makes it a domain of MK(A)\mathcal M_K(A). Iteration and uniqueness give the finite-intersection and parenthesization assertions. If U⊆VU\subseteq V, the restriction part of that lemma and the same uniqueness identify the pushout with BB. ◻

References. The generalized finite rational-refinement theorem is due to Gerritzen–Grauert in the strict setting and to its generalizations; compare [Tem05] (Theorem 3.1 and the following paragraph, p. 8). The proof here reuses the internal strict theorem through the free Gauss section of [Ber93] (Lemma 2.2.5). The complete-extension comparison follows the rational-cover gluing method of [Tem05] (Corollary 3.2, pp. 8–9).

2.37 Finite-module descent at arbitrary radii

The finite-module descent theorem is proved first for a finite rational covering. A single free-radius extension makes that entire diagram strict. Its matching module is then finite by the strict Kiehl theorem. Extracting coefficients descends finite generation and identifies the restrictions over the original field. This uses the strict descent argument already reconstructed from [BGR84] (§9.4.3); only the change of radii is added here.

Throughout this section, KK is complete for its specified real-valued non-Archimedean absolute value, which may be trivial. Affinoid algebras have arbitrary positive presentation radii and norm-admissible norms. Finite modules carry canonical norms. Rational algebras and their intersections mean the actual quotients of Proposition 2.31.3.

Lemma 2.37.1. NAG-001826 (Coefficient extension of matching kernels). Let d:E→Fd:E\to F be a bounded map of Banach KK-spaces, let N=ker⁡dN=\ker d with its induced norm, and let rr be a finite KK-free tuple. Under the coefficient identifications for scalar extension to KrK_r, the natural map identifies N⊗^KKr → ∼  ker⁡(E⊗^KKr⟶F⊗^KKr)N\widehat\otimes_KK_r\ \xrightarrow{\ \sim\ }\ \ker\bigl(E\widehat\otimes_KK_r\longrightarrow F\widehat\otimes_KK_r\bigr) isometrically. Coefficient extension also commutes isometrically with finite products equipped with their maximum norms. These identifications respect the actions and maps of Banach modules when the original maps are module-linear.

Proof. The kernel NN is closed, hence complete. By Theorem 1.16.2, the extended map applies dd to every coefficient. A family (eν)(e_\nu) belongs to its kernel exactly when every eνe_\nu belongs to NN. The induced norm on NN makes its weighted vanishing condition and supremum norm exactly those inherited from EE. This proves the kernel assertion without a closed-image assumption on dd. For finite products, interchange the finite maximum with the coefficient supremum; the vanishing condition is preserved by finite unions of exceptional index sets. Module compatibility follows first on finite Laurent families and then on their limits by Lemma 1.18.1. ◻

Lemma 2.37.2. NAG-001827 (Module sheaves on a rational slice). Let X=M(A)X=\mathcal M(A) and let RX\mathcal R_X be its arbitrary-radius finite rational site. For V∈RXV\in\mathcal R_X and a finite AVA_V-module PP, put P~V(W)=AW⊗AVP(W∈RX,W⊆V).\widetilde P_V(W)=A_W\otimes_{A_V}P \qquad(W\in\mathcal R_X, W\subseteq V). This is a sheaf on the slice RX/V\mathcal R_X/V. The functor P↦P~VP\mapsto\widetilde P_V is exact and fully faithful, its sections over VV are PP, and it commutes with restriction to smaller rational objects. The augmented Čech complex for every finite covering in this slice is norm-strict exact in all degrees.

Proof. A rational object W⊆VW\subseteq V is rational over AVA_V with algebra AWA_W: pull back its original defining tuple and apply uniqueness in Theorem 2.31.2. Finite intersections have the same identifications. Applying Theorem 2.32.2 over AVA_V, and then over each AWA_W, proves the sheaf and Čech assertions. Proposition 2.32.3 proves exactness. At VV the value is PP. A sheaf homomorphism is determined by its map on PP, since its value on b⊗pb\otimes p must be bb times the restriction of that map on pp. Conversely every AVA_V-linear map defines these maps on all objects. Ordinary tensor associativity proves the restriction assertion, with the bounded finite-module comparisons of Proposition 1.18.7. ◻

Theorem 2.37.3. NAG-001828 (Effective finite-module descent for a rational covering). Let V1,…,VnV_1,\ldots,V_n be a finite rational covering of X=M(A)X=\mathcal M(A), with algebras AiA_i and joint algebras Aij,AijkA_{ij},A_{ijk}. Suppose finite AiA_i-modules NiN_i are given, together with isomorphisms αij:Aij⊗AiNi→ ∼ Aij⊗AjNj\alpha_{ij}: A_{ij}\otimes_{A_i}N_i \xrightarrow{\ \sim\ } A_{ij}\otimes_{A_j}N_j satisfying the identity and cocycle conditions on all triple intersections. Define the matching AA-module N={(ni)∈∏iNi:αij(1⊗ni)=1⊗njin Aij⊗AjNj}.N=\left\{(n_i)\in\prod_iN_i: \alpha_{ij}(1\otimes n_i)=1\otimes n_j \text{in }A_{ij}\otimes_{A_j}N_j\right\}. Then NN is finite over AA. Its subspace norm from ∏iNi\prod_iN_i is canonical up to uniform equivalence, and the maps θi:Ai⊗AN⟶Ni,b⊗(nj)j⟼bni,\theta_i:A_i\otimes_A N\longrightarrow N_i, \qquad b\otimes(n_j)_j\longmapsto b n_i, are isomorphisms with bounded inverses and recover the prescribed overlap identifications. The construction is functorial and gives an equivalence between finite AA-modules and these finite-module gluing data. It also identifies NN with the global sections of the sheaf obtained by gluing the Ni~Vi\widetilde{N_i}_{V_i}.

Proof. Step 1: the matching Banach module. Choose the defining rational quotient norms, so that A→AiA\to A_i is contractive, and quotient canonical norms on the NiN_i. The module actions are contractive. The overlap modules are finite Banach modules and the αij\alpha_{ij} are bounded by Theorem 1.18.3. Thus NN is the kernel of a bounded AA-linear map between finite products of Banach spaces. It is a Banach AA-module with contractive action; finite generation is not used here. Uniformly equivalent choices of the finite local norms give uniformly equivalent norms on NN.

Step 2: strictify the finite gluing diagram. Choose one free extension L=KrL=K_r by Theorem 1.17.6 for the presentation of AA and the finitely many covering radii. We may require LL to be nontrivially valued: if KK is trivial, include any additional radius different from 11 in the finite list. The extended cover is a unit-radius rational cover after the bounded comparisons of Proposition 2.26.2. Proposition 2.32.1 identifies its actual intersection algebras and the extended finite modules, including the overlap maps. The cocycle conditions persist. Lemma 2.37.1 gives a bounded identification Nr≃ker⁡(∏i(Ni)r⟶∏i,j(Aij⊗AjNj)r).N_r\simeq \ker\left(\prod_i (N_i)_r\longrightarrow \prod_{i,j}(A_{ij}\otimes_{A_j}N_j)_r\right).

The strict local module sheaves glue to a sheaf Fr\mathcal F_r on the strict rational site. Explicitly, on a rational object WW take the matching families of sections on W∩(Vi)rW\cap (V_i)_r, using the prescribed overlap maps. The sheaf axiom follows by gluing in each local module sheaf and then checking the overlap identities after restriction to a cover. On a slice contained in (Vi)r(V_i)_r, projection to the iith component is an isomorphism: the other components are its restrictions transported by αij\alpha_{ij}, and the cocycle identity is precisely their compatibility. Consequently Fr\mathcal F_r has finite local models, and its global sections are the displayed kernel. The strict Kiehl theorem, Theorem 2.28.12, now proves that NrN_r is finite over ArA_r and that its restriction maps recover (Ni)r(N_i)_r.

Step 3: descend finite generation and the restriction maps. Since LL is nontrivially valued, the last clause of Theorem 1.18.9 applies to the Banach module NN. It proves finite generation over AA and equivalence of its matching norm with a canonical norm. The maps θi\theta_i are therefore bounded maps between finite canonical AiA_i-modules. By Proposition 2.32.1, their free extensions are exactly the isomorphisms obtained in Step 2. The reflection assertion of Theorem 1.16.3 implies that each θi\theta_i is an isomorphism with bounded inverse. Their overlap identities hold by the definition of NN.

A morphism of gluing data maps matching families to matching families, which gives functoriality. Starting with a finite AA-module MM, the degree-zero equalizer in Theorem 2.32.2 identifies its matching module with MM. Starting with gluing data, the maps θi\theta_i identify their reconstruction with the original data, compatibly with morphisms. These are the two inverse natural isomorphisms. The same explicit sheaf gluing construction as in Step 2, now over KK, and Lemma 2.37.2 identify its global sections with NN. Empty components contribute zero modules. If XX is empty, Theorem 1.20.1 gives A=0A=0 and every assertion is immediate. ◻

Theorem 2.37.4. NAG-001829 (Kiehl equivalence on the arbitrary-radius rational site). On (RX,OXrat)(\mathcal R_X,\mathcal O_X^{\mathrm{rat}}) the following conditions on a module sheaf F\mathcal F are equivalent: coherence in the sense of [StacksAlg] (Tag 03DL, Definition 18.23.1(8)); local finite presentation; possession of finite module models on some finite rational covering; and existence of a finite AA-module MM with F≃M~\mathcal F\simeq\widetilde M. Thus Mod⁡fin(A) ≃ Coh⁡(OXrat),M⟼M~,F⟼F(X)\operatorname{Mod}^{\mathrm{fin}}(A) \ \simeq\ \operatorname{Coh}(\mathcal O_X^{\mathrm{rat}}), \qquad M\longmapsto\widetilde M, \qquad\mathcal F\longmapsto\mathcal F(X) are inverse exact equivalences. Every coherent sheaf has a global finite free presentation. Its augmented Čech complex for every finite rational covering is norm-strict exact, with canonical norms on the section modules.

Proof. A coherent sheaf has finite local generators, and the kernels of the resulting maps from finite free sheaves are locally of finite type. Refining the finite cover supplies finite presentations. Each presentation on a slice is a matrix over AVA_V, since its columns are the images of the global basis sections. Lemma 2.37.2 identifies its cokernel sheaf with the sheaf of the finite matrix cokernel. Thus local finite presentation gives finite local module models. Their overlap sheaf isomorphisms give module isomorphisms by full faithfulness in the same lemma, with the required cocycle condition. Theorem 2.37.3 produces a global finite module MM and compatible local isomorphisms with M~\widetilde M. Applying the sheaf equalizer on the covers W∩ViW\cap V_i glues these to an isomorphism on every object WW.

Conversely, a finite AA-module has a finite presentation because AA is Noetherian by Theorem 1.18.3. Exactness of the module-sheaf functor gives a global finite free presentation of M~\widetilde M. For any VV and any map OVt→M~∣V\mathcal O_V^t\to\widetilde M|_V, full faithfulness identifies it with a map AVt→AV⊗AMA_V^t\to A_V\otimes_A M. Its kernel is finite by Noetherianity of AVA_V, and exactness identifies the sheaf kernel with its associated finite-module sheaf. This proves coherence. The same lemma gives exactness and full faithfulness, and global sections recover MM. The Čech assertion follows from Theorem 2.32.2. ◻

2.38 Affinoid coverings and coherent sheaves

We use the actual intersections of Corollary 2.36.10, the base-change identifications of Theorem 2.36.9, and the restriction maps of Lemma 2.36.5. All domains and intersection algebras in this section carry these identifications.

The rational refinement theorem allows the preceding module theory to be used on all affinoid domains. We retain the complete valued field KK and its original absolute value; presentation radii are arbitrary positive real numbers. Domains are those of Definition 2.1.1, whose complete-extension test property has now been proved in Theorem 2.36.7.

Theorem 2.38.1. NAG-001831 (Tate acyclicity for affinoid coverings). Let AA be a generalized KK-affinoid algebra, MM a finite AA-module, and (Vi)i∈I(V_i)_{i\in I} a finite affinoid-domain covering of X=M(A)X=\mathcal M(A). Write AJA_J for the actual affinoid algebra of VJ=⋂j∈JVjV_J=\bigcap_{j\in J}V_j. With canonical norms on the finite section modules, the augmented complex 0⟶M⟶∏iAi⊗AM⟶∏i<jAij⊗AM⟶⋯0\longrightarrow M\longrightarrow\prod_i A_i\otimes_A M \longrightarrow\prod_{i<j}A_{ij}\otimes_A M \longrightarrow\cdots is norm-strict exact. The all-tuples complex has the same property. Ordinary finite-module tensors in these formulas agree boundedly with their separated completed tensors.

Proof. Theorem 2.36.6 gives a finite original-AA rational covering of each ViV_i. Combining them gives a finite rational covering W\mathcal W of XX, with a chosen containing ViV_i for each member. Its augmented complex is norm-strict exact by Theorem 2.32.2.

For a nonempty tuple of coarse indices put Y=Vi0∩⋯∩VipY=V_{i_0}\cap\cdots\cap V_{i_p}. Corollary 2.36.10 gives its actual iterated completed pushout AYA_Y and the canonical domain restrictions. If YY is empty, AY=0A_Y=0 by the nonemptiness criterion of Theorem 1.20.1; the mixed intersections are then also represented by zero algebras. Otherwise use Theorem 2.36.9 for the whole domain XX represented by A→AA\to A, with L=KL=K and C=AYC=A_Y. It identifies each rational member WW of W\mathcal W over YY with the rational domain defined by its pulled-back tuple, with actual algebra AY⊗^AAWA_Y\widehat\otimes_A A_W. The restrictions cover YY, since the inverse images of a cover cover the source. Apply Theorem 2.32.2 over AYA_Y to this covering and the finite module MY=AY⊗AMM_Y=A_Y\otimes_A M. The canonical restriction maps and joint algebras agree with those of the mixed intersections by the same base-change and intersection identifications. Proposition 1.18.7 and the ordinary map b⊗(c⊗m)↦bc⊗mb\otimes(c\otimes m)\mapsto bc\otimes m give the bounded coefficient identifications, compatible with the augmentation and each restriction. Thus the ordered restricted complex is norm-strict exact on every coarse intersection, with the zero terms included.

For X≠∅X\ne\varnothing, apply Proposition 2.21.5(1) to the coarse and rational families, using on each actual intersection DD the finite canonical module N(D)=AD⊗AMN(D)=A_D\otimes_A M and the canonical bounded restriction maps. Each rational member keeps its chosen containing ViV_i from the first paragraph. The restricted complexes are those just identified, including the zero complexes for empty intersections. This transfers norm-strict exactness and gives the all-tuples assertion. Proposition 1.18.7 gives the final ordinary-to-completed comparison. If XX is empty, Theorem 1.20.1 gives A=0A=0 and hence M=0M=0, which includes the empty covering. ◻

Proposition 2.38.2. NAG-001832 (Flatness of affinoid-domain restriction). Every affinoid-domain map A→BA\to B is flat as a homomorphism of ordinary rings. A finite affinoid-domain covering induces a faithfully flat map A→∏iAViA\to\prod_i A_{V_i}. For a short exact sequence of finite canonical AA-modules, restriction to BB is norm-strict exact.

Proof.

Cover M(B)\mathcal M(B) by the original-AA rational domains RjR_j of Theorem 2.36.6, and put S=∏jARjS=\prod_j A_{R_j}. Lemma 2.36.5 makes the same family rational over BB, with the same algebras. By Proposition 2.32.3, B→SB\to S is faithfully flat and SS is flat over AA. For an injection N′→NN'\to N of arbitrary AA-modules, let QQ be the kernel of B⊗AN′→B⊗ANB\otimes_A N'\to B\otimes_A N. Flatness over BB identifies S⊗BQS\otimes_B Q with the corresponding kernel after tensoring with SS over AA; that kernel is zero. Faithfulness gives Q=0Q=0, proving flatness of BB over AA. The zero algebra handles an empty domain.

For a finite covering, refine each member by these rational covers. Their product SS is faithfully flat over AA and is an algebra over B′=∏iAViB'=\prod_i A_{V_i}. The latter is flat over AA by the result just proved. If B′⊗AN=0B'\otimes_A N=0, then S⊗AN=0S\otimes_A N=0, so N=0N=0; thus B′B' is faithfully flat. Finally the finite restricted modules carry canonical norms, and every finite-module map is norm-strict by Theorem 1.18.3. Algebraic exactness therefore gives the last assertion, with the completed-tensor interpretation supplied by Proposition 1.18.7. ◻

Definition 2.38.3. NAG-001833 (The finite affinoid-domain site). For X=M(A)X=\mathcal M(A), let DX\mathcal D_X be the category of all affinoid domains in XX, with inclusions as arrows. Give it finite set-theoretic coverings by its objects, including the empty covering of the empty domain. Choose representing algebras AVA_V, with the unique bounded restriction maps, and define OXaff(V)=AV,M~aff(V)=AV⊗AM(M finite over A).\mathcal O_X^{\mathrm{aff}}(V)=A_V, \qquad \widetilde M^{\mathrm{aff}}(V)=A_V\otimes_A M \quad(M\text{ finite over }A).

Theorem 2.38.4. NAG-001834 (Structure sheaf and rational comparison). The finite coverings define a pretopology on DX\mathcal D_X. The displayed assignments are a ring sheaf and module sheaves. The functor M↦M~affM\mapsto\widetilde M^{\mathrm{aff}} is exact and fully faithful, with Γ(X,M~aff)=M\Gamma(X,\widetilde M^{\mathrm{aff}})=M. Inclusion of the arbitrary-radius rational site induces an equivalence of ringed topoi (RX,OXrat)≃(DX,OXaff),(\mathcal R_X,\mathcal O_X^{\mathrm{rat}}) \simeq (\mathcal D_X,\mathcal O_X^{\mathrm{aff}}), identifying the sheaves associated to the same finite module.

Proof. Affinoid-domain intersections and restrictions exist and are compatible. Finite covers are stable under restriction and composition, proving the pretopology axioms. A finite cover of an object VV is a finite affinoid-domain cover over AVA_V. Apply Theorem 2.38.1 to AV⊗AMA_V\otimes_A M; tensor associativity gives exactly the equalizer for the displayed presheaf. This proves the sheaf axiom, and M=AM=A gives the ring sheaf. Restriction flatness proves exactness. Evaluation at XX gives MM. A module-sheaf map is determined on VV by b⊗m↦b⊗u(m)b\otimes m\mapsto b\otimes u(m), where uu is its map at XX; conversely this formula defines a sheaf map for every AA-linear uu. Thus the functor is fully faithful.

For the comparison apply [StacksAlg] (Tag 03A0, Lemma 7.29.1). The inclusion is fully faithful and preserves covers and finite intersections, so restriction of sheaves is continuous. Every object of DX\mathcal D_X has a finite covering by original-AA rational objects, by Theorem 2.36.6. For a cover of a rational object by affinoid domains, refine each member by such a cover; their finite union is a rational refinement. This verifies cocontinuity as well as the covering condition in the comparison lemma. Local fullness and faithfulness follow from full faithfulness of the inclusion. On rational objects the representing algebras and restrictions agree by uniqueness, so the resulting equivalence identifies the ring sheaves and the finite-module sheaves as stated. ◻

Theorem 2.38.5. NAG-001835 (Kiehl equivalence and effective affinoid gluing). Finite AA-modules and coherent OXaff\mathcal O_X^{\mathrm{aff}}-modules are equivalent through M↦M~affM\mapsto\widetilde M^{\mathrm{aff}} and global sections. The equivalence is exact, and coherence is equivalent to finite presentation. Every coherent sheaf has a global finite free presentation and norm-strict exact augmented Čech complexes for all finite affinoid-domain coverings. Finite modules with compatible gluing isomorphisms on such a covering descend effectively to a finite AA-module, given by their matching kernel.

Proof. The comparison just proved satisfies the cocontinuous comparison hypotheses of [StacksAlg] (Tag 03DM, Lemma 18.23.2), so it preserves coherence and finite presentation. Theorem 2.37.4 identifies the coherent sheaves on the rational site with finite AA-modules. Under comparison their extensions are exactly M~aff\widetilde M^{\mathrm{aff}}; hence the asserted equivalence, exactness and inverse follow from Theorem 2.38.4. A finite presentation over the Noetherian ring AA gives the global presentation. Apply Theorem 2.38.1 to obtain the Čech assertion.

For the gluing statement, attach to each local finite module its sheaf on the corresponding domain slice. On an arbitrary object take matching sections on its intersections with the covering members. Local sheaf gluing and the cocycle identity give a sheaf whose restriction to each member is the prescribed local sheaf. Thus it is locally finitely presented and hence coherent. The equivalence gives a finite global module, and the sheaf equalizer identifies that module with the matching kernel. The identifications on each member recover the given gluing maps by construction. All section-module comparisons and inverses are bounded by Theorem 1.18.3. ◻

Corollary 2.38.6. NAG-001836 (Affinoid Theorem B). For every coherent module F\mathcal F on DX\mathcal D_X and every object VV, one has Hq(V,F)=0H^q(V,\mathcal F)=0 for q>0q>0. The same holds on RX\mathcal R_X. These are the right derived functors of sections in the corresponding categories of abelian sheaves.

Proof. Apply [StacksAlg] (Tag 03F9, Lemma 21.10.9). Take its basis to be all objects and its covering system to be all finite coverings of the chosen site. Intersections remain in the site. Every covering of every object has vanishing higher Čech cohomology, by Theorem 2.38.1 and the Kiehl identification; for the rational site use Lemma 2.37.2. The covering system is cofinal by the definition of the generated topology. These verify the hypotheses of the cited lemma, which gives the asserted derived vanishing. ◻

References. This reconstruction combines the internally proved rational Tate theorem, finite-module descent and Gerritzen–Grauert refinement. The classical affinoid theory is developed in [Ber90] (§§2.1–2.2) and [BGR84] (§9.4.3); the site-theoretic comparisons used in the proofs are the explicitly indicated Stacks Project results.

2.39 The analytic G-site of an affinoid space

We now pass from compact affinoid pieces to arbitrary analytic domains. The covering condition is topological, but it allows a point to lie on the boundary of every member containing it. This is the reason for using quasi-nets. The construction follows the atlas language of [Ber93] (§1.2); its algebraic gluing has already been proved in the preceding section.

Definition 2.39.1. NAG-001619 NAG-001620 (Quasi-nets and analytic domains). A family (Vi)(V_i) of subsets of a topological space VV is a quasi-net if, for every x∈Vx\in V, some finite subfamily consists of sets containing xx and has union containing a neighborhood of xx in VV. Let X=M(A)X=\mathcal M(A). A subset V⊆XV\subseteq X is an analytic domain if it has a quasi-net covering by affinoid domains of XX, with the induced topology on VV. A family of analytic subdomains of VV is an admissible covering if it is a quasi-net on VV. The empty family covers the empty domain.

Lemma 2.39.2. (Refining quasi-nets). Quasi-nets restrict to subsets and are transitive under quasi-net refinements. A finite covering by closed subsets is a quasi-net. If a compact space has a quasi-net covering, some finite collection of its members covers the space.

Proof. Restriction follows by intersecting a neighborhood and its finitely many covering members with the subset. For transitivity, fix xx and choose finitely many ViV_i containing xx whose union contains a neighborhood NN of xx. For each chosen ii, select finitely many refining members containing xx whose union contains a relative neighborhood NiN_i of xx in ViV_i. Write Ni⊇Oi∩ViN_i\supseteq O_i\cap V_i with OiO_i a neighborhood of xx in the ambient space. Then N∩⋂iOiN\cap\bigcap_iO_i is a neighborhood covered by the finitely many selected refining members, all of which contain xx.

For a finite closed covering, delete the members not containing xx; their union is closed and misses xx, so the remaining members cover a neighborhood of xx. Finally, the neighborhoods supplied by a quasi-net cover a compact space. A finite subcover of these neighborhoods uses only finitely many of the original members and proves the compactness assertion. ◻

Proposition 2.39.3. NAG-001621 (Analytic domains and their coverings). Analytic domains in XX are closed under finite intersections. Their admissible coverings define a pretopology, denoted XGX_G. Every analytic domain is locally compact and Hausdorff. A subset is a compact analytic domain exactly when it is a finite union of affinoid domains of XX. Every ordinary open subset of XX is an analytic domain. If VV is affinoid, the analytic domains and coverings inside VV are precisely those obtained by applying Definition 2.39.1 to M(AV)\mathcal M(A_V).

Proof. Take affinoid quasi-net coverings (Ui)(U_i) of UU and (Vj)(V_j) of VV. At x∈U∩Vx\in U\cap V, choose the finite subfamilies containing xx and covering relative neighborhoods. The finitely many intersections Ui∩VjU_i\cap V_j contain xx and cover a neighborhood in U∩VU\cap V. They are affinoid domains by Corollary 2.36.10; hence U∩VU\cap V is analytic. The same argument, together with Lemma 2.39.2, proves stability of admissible covers under pullback and composition, and the identity cover is admissible.

An affinoid quasi-net gives every point a compact neighborhood, namely a finite union of affinoid pieces. These pieces are compact and closed in the Hausdorff space XX, so every analytic domain is locally compact and Hausdorff. If the domain is compact, the last assertion of Lemma 2.39.2 gives a finite affinoid cover. Conversely a finite union of affinoid domains is compact, and its finite closed covering is a quasi-net by that lemma.

For an open set OO and x∈Ox\in O, the evaluation topology gives finitely many strict upper or lower bounds on values of elements of AA whose common locus contains xx and is contained in OO. Choose weak bounds strictly between the value at xx and each prescribed strict bound. They define finitely many Weierstrass or Laurent domains; their intersection is rational by Proposition 2.31.3, is contained in OO, and contains a neighborhood of xx. These rational neighborhoods form a quasi-net covering of OO. Finally, affinoid domains of XX contained in VV are affinoid over AVA_V, and affinoid domains over AVA_V are affinoid over AA, by restriction and transitivity in Lemma 2.36.5. The induced topologies and quasi-net conditions coincide. This proves the last assertion. ◻

Theorem 2.39.4. NAG-001622 (Comparison with the full analytic G-site). The inclusions of the finite affinoid-domain site and the arbitrary-radius rational site into XGX_G induce equivalences of topoi. The structure sheaf OXaff\mathcal O_X^{\mathrm{aff}} therefore extends uniquely, up to its unique comparison isomorphism, to a ring sheaf OXG\mathcal O_{X_G}. For an analytic domain VV, its sections are compatible families of sections on affinoid domains contained in VV. On an affinoid object VV they equal AVA_V, with the original restriction homomorphisms.

Proof. A finite affinoid-domain cover is a finite closed cover, so it is admissible by Lemma 2.39.2. Conversely, take an admissible cover (Vi)(V_i) of an affinoid object VV. Choose an affinoid quasi-net covering of each ViV_i. Transitivity makes their combined family a quasi-net of VV, and compactness gives a finite affinoid refinement of the original cover. Every object of XGX_G is covered by affinoid objects by definition. Intersections of such objects remain affinoid.

These statements verify the comparison hypotheses of [StacksAlg] (Tag 03A0, Lemma 7.29.1): full faithfulness of the inclusion, continuity, cocontinuity by the finite refinement just constructed, and covering of every target object. Thus restriction gives an equivalence of topoi. Combining it with Theorem 2.38.4 gives the rational comparison. The inverse equivalence extends the ring sheaf. Concretely, its value on VV is the compatible-family limit over all affinoid domains contained in VV. Indeed, for an affinoid cover, compatibility on all intersections is the sheaf equalizer condition; refining two such covers compares them on their affinoid intersections. On an affinoid VV the family has final section datum at VV itself, recovering AVA_V. These identifications determine the restriction maps and the unique comparison isomorphism. ◻

Theorem 2.39.5. NAG-001623 (Banach sections on compact analytic domains). On compact analytic domains, OXG\mathcal O_{X_G} is a sheaf of Banach KK-algebras with bounded restrictions for finite domain coverings. For a finite affinoid covering V=⋃iViV=\bigcup_i V_i, its Banach algebra of sections is OXG(V)=ker⁡(∏iAVi⇉∏i,jAVi∩Vj),\mathcal O_{X_G}(V)= \ker\left(\prod_i A_{V_i}\rightrightarrows \prod_{i,j}A_{V_i\cap V_j}\right), with the induced maximum norm. Different finite affinoid covers and admissible norms give canonical bounded isomorphisms and uniformly equivalent section norms.

Proof. Proposition 2.39.3 supplies a finite affinoid cover. The preceding theorem identifies the section ring with the displayed matching kernel. The difference of the two restriction maps is bounded, so its kernel is closed in a finite product of Banach algebras and is complete. Multiplication is componentwise and is continuous.

For two covers (Vi)(V_i) and (Wj)(W_j), their intersections give a common finite affinoid cover. Restriction from the first matching kernel to this refinement is bounded. On each ViV_i, Theorem 2.38.1 applied to (Vi∩Wj)j(V_i\cap W_j)_j identifies AViA_{V_i} with its matching kernel by a bounded inverse. Taking the maximum of the finitely many inverse bounds shows that the map between the global matching kernels has bounded inverse. The same argument for (Wj)(W_j) proves uniform equivalence and cover independence. Changing finitely many initial norms to equivalent ones does not affect this conclusion.

For compact domains W⊆VW\subseteq V, choose finite affinoid covers of both. Restrict a matching family on VV to all pairwise intersections of these covers, and on each affinoid member of the cover of WW use the same bounded gluing inverse. There are only finitely many restrictions and inverses, proving that O(V)→O(W)\mathcal O(V)\to\mathcal O(W) is bounded. Every finite covering of a compact domain by compact analytic domains has finite affinoid refinements. The sheaf equalizer is therefore the one already established on XGX_G, with its bounded Banach realization. ◻

Theorem 2.39.6. NAG-001624 (Kiehl and Theorem B on the analytic G-site). The functors Mod⁡fin(A) ⇄ Coh⁡(XG),M⟼M~,F⟼Γ(X,F)\operatorname{Mod}^{\mathrm{fin}}(A) \ \rightleftarrows\ \operatorname{Coh}(X_G), \qquad M\longmapsto\widetilde M, \qquad\mathcal F\longmapsto\Gamma(X,\mathcal F) are inverse exact equivalences, where M~(V)=AV⊗AM\widetilde M(V)=A_V\otimes_A M on affinoid domains. Every coherent sheaf has a global finite free presentation. For every affinoid domain VV and every q>0q>0, Hq(VG,F∣VG)=0.H^q(V_G,\mathcal F|_{V_G})=0. The cohomology is derived sheaf cohomology on the full analytic G-site.

Proof. Theorem 2.39.4 is the cocontinuous ringed-site comparison of [StacksAlg] (Tag 03A0), so [StacksAlg] (Tag 03DM) preserves coherence and finite presentation. Theorem 2.38.5 therefore gives the equivalence and the global presentation. The equivalence preserves evaluation on each common affinoid object. On the slice at VV, Proposition 2.39.3 identifies the site with VGV_G. The corresponding equivalence of abelian sheaf categories is exact and has an exact inverse. It preserves injective objects: apply the inverse equivalence to a monomorphism and use the defining extension property of an injective. Thus an injective resolution may be transported across the equivalence, and its section complexes agree. Their cohomology groups agree as well. Corollary 2.38.6 now gives the asserted vanishing. ◻

2.40 Analytic spaces from affinoid nets

Affinoid domains also serve as charts when no single chart is a neighborhood of a point. We use the net construction of [Ber93] (§§1.1–1.4), with the overlap arguments expressed through the affinoid gluing results proved above. All affinoid algebras in this section may have arbitrary positive real radii.

Definition 2.40.1. NAG-001605 (An affinoid atlas). Let XX be a locally Hausdorff topological space. A net τ\tau of compact Hausdorff subsets is a quasi-net such that τ∣U∩V={W∈τ:W⊆U∩V}\tau|_{U\cap V}=\{W\in\tau:W\subseteq U\cap V\} is a quasi-net of U∩VU\cap V for every U,V∈τU,V\in\tau. An affinoid atlas consists of such a net, affinoid algebras AVA_V, homeomorphisms V≃M(AV)V\simeq\mathcal M(A_V), and bounded homomorphisms AV→AUA_V\to A_U for inclusions U⊆VU\subseteq V in τ\tau. The latter identify UU with an affinoid domain of VV, and are compatible with composition.

Lemma 2.40.2. NAG-001606 (Topology and bounded gluing). A quasi-net determines the topology: a subset O⊆XO\subseteq X is open if and only if O∩VO\cap V is relatively open for every member VV. Let Z=M(B)Z=\mathcal M(B) and let Z1,…,ZmZ_1,\ldots,Z_m be an affinoid-domain cover. Compatible bounded homomorphisms C→BZiC\to B_{Z_i} from an affinoid algebra CC glue uniquely to a bounded homomorphism C→BC\to B. Their induced maps on spectra glue to the induced map on ZZ.

Proof. Suppose all O∩VO\cap V are open and take x∈Ox\in O. Choose V1,…,VmV_1,\ldots,V_m containing xx whose union contains a neighborhood NN of xx. Write O∩Vi=Oi∩ViO\cap V_i=O_i\cap V_i near xx, with OiO_i open in XX and containing xx. An open neighborhood of xx contained in N∩⋂iOiN\cap\bigcap_iO_i lies in OO. This proves the topological assertion. For bounded gluing, the degree-zero statement of Theorem 2.38.1 identifies BB, by a bounded isomorphism, with the matching kernel of the finite diagram of the BZiB_{Z_i}. The given homomorphisms take values in that kernel. Composition with its bounded inverse gives the required map; addition, multiplication and the unit can be checked after the injective restriction to the product. Evaluation on each ZiZ_i proves the assertion about spectra. ◻

Proposition 2.40.3. NAG-001607 (Transport and saturation of charts). If U,V∈τU,V\in\tau and WW is an affinoid domain of UU contained in VV, then its affinoid algebra has a canonical bounded AVA_V-algebra structure representing the same subset as an affinoid domain of VV. These structures are transitive. Consequently the family τˉ\bar\tau of affinoid domains in members of τ\tau is a net with a unique extending affinoid atlas.

Proof. Choose finitely many Ci∈τ∣U∩VC_i\in\tau|_{U\cap V} whose intersections with WW cover WW. This is possible by the quasi-net condition and compactness of WW. Put Di=W∩CiD_i=W\cap C_i, using the intersection algebra inside UU. Each DiD_i is an affinoid domain of WW and of CiC_i, hence of VV by transitivity. The maps AV→ACi→ADiA_V\to A_{C_i}\to A_{D_i} agree on Di∩DjD_i\cap D_j: cover this compact intersection by sets W∩EW\cap E with E∈τ∣Ci∩CjE\in\tau|_{C_i\cap C_j}. On these sets the two maps agree by the original atlas functoriality. Affinoid Tate injectivity on the finite refinement proves their agreement on the intersection. Lemma 2.40.2 now gives AV→AWA_V\to A_W, whose spectrum map is the inclusion of WW.

We verify the domain property. Let TT be affinoid and let T→VT\to V have image in WW. The inverse images TiT_i of the domains Di⊆VD_i\subseteq V form a finite affinoid cover of TT. Each gives a map Ti→Di→WT_i\to D_i\to W. On Ti∩TjT_i\cap T_j, refine by inverse images of the common atlas pieces E⊆Ci∩CjE\subseteq C_i\cap C_j used above. These pieces cover because the common point image lies in Di∩DjD_i\cap D_j. On every refined piece, both maps to UU are the map through E→UE\to U prescribed by the original atlas. They therefore agree, and their factorizations through the original affinoid domain W⊆UW\subseteq U agree by its universal property. Tate gluing yields a bounded map T→WT\to W. Its composite with W→VW\to V is the given map, as can be checked on the TiT_i. Any other factorization restricts uniquely to Ti→DiT_i\to D_i, since DiD_i is an affinoid domain both of WW and of VV; hence it is equal to the constructed factorization.

The same universal property proves independence of the finite cover and transitivity under inclusions. For the net assertion take W⊆UW\subseteq U and W′⊆VW'\subseteq V in τˉ\bar\tau and a point x∈W∩W′x\in W\cap W'. Choose finitely many Ci∈τ∣U∩VC_i\in\tau|_{U\cap V} containing xx whose union contains a relative neighborhood of xx. The domains W∩W′∩CiW\cap W'\cap C_i, formed inside CiC_i, belong to τˉ\bar\tau, contain xx, and cover a neighborhood in W∩W′W\cap W'. This proves the net condition. The established transport gives all inclusion maps in τˉ\bar\tau and proves their uniqueness. ◻

Definition 2.40.4. NAG-001608 (Analytic domains and the analytic site). An analytic domain Y⊆XY\subseteq X is a subset for which τˉ∣Y\bar\tau|_Y is a quasi-net, equipped with the induced topology and the restricted atlas. A covering by analytic domains is admissible if it is a quasi-net. The associated site is denoted XGX_G.

Proposition 2.40.5. NAG-001609 (Local calculus of analytic domains). Analytic domains are stable under finite intersections and under passage to analytic domains inside them. Admissible coverings define a pretopology. Every open subset is an analytic domain. Every point has a compact Hausdorff analytic neighborhood contained in any prescribed open Hausdorff neighborhood. On a chart V∈τˉV\in\bar\tau, the domain structure and site are precisely those of M(AV)\mathcal M(A_V) in Proposition 2.39.3.

Proof. For intersections, intersect the finitely many chart members supplying the quasi-net condition at a point, and refine each resulting chart intersection by the net τˉ\bar\tau. Restriction and transitivity of quasi-nets, proved in Lemma 2.39.2, give the assertions about subdomains, pullbacks of covers, and composition of covers. Chart transport in Proposition 2.40.3 identifies each local refinement with an affinoid domain in the chart under consideration; this proves the assertion about its site.

Let x∈Ox\in O with OO open. Choose finitely many charts ViV_i containing xx whose union contains a neighborhood of xx. Inside each ViV_i, the rational-neighborhood construction in Proposition 2.39.3 gives an affinoid domain Wi⊆O∩ViW_i\subseteq O\cap V_i containing a relative neighborhood of xx. Their union contains a neighborhood of xx in OO. Thus OO is analytic. If OO is Hausdorff, the same union D=⋃iWiD=\bigcup_iW_i is compact Hausdorff, is an analytic domain by its finite closed covering, and is a neighborhood of xx in XX. ◻

Theorem 2.40.6. NAG-001610 (The structure sheaf of an atlas). There is a canonical sheaf of rings OXG\mathcal O_{X_G} whose restriction to every chart is its affinoid structure sheaf. Explicitly, for an analytic domain YY, OXG(Y)={(sV)V∈τˉ, V⊆Y:sV∈AV,sV∣W=sW whenever W⊆V}.\mathcal O_{X_G}(Y)= \left\{(s_V)_{V\in\bar\tau,\,V\subseteq Y}: s_V\in A_V,\quad s_V|_W=s_W\text{ whenever }W\subseteq V\right\}. Given any chart quasi-net σ⊆τˉ\sigma\subseteq\bar\tau, the family σˉ\bar\sigma of all affinoid domains in members of σ\sigma is a net. Replacing τ\tau by this net gives the same analytic domains, coverings and ring sheaf, with canonical comparison isomorphisms.

Proof. The displayed expression defines a ring presheaf. Given compatible sections on an admissible covering (Yi)(Y_i) of YY, fix a chart V⊆YV\subseteq Y. The intersections V∩YiV\cap Y_i are analytic domains of VV and form an admissible covering. Refining by affinoid domains and using compactness yields a finite affinoid cover of VV subordinate to the YiY_i. The given sections agree on its overlaps, so Theorem 2.38.1 gives a unique element of AVA_V. It is independent of the chosen finite refinement, since two refinements have a common affinoid refinement. The elements obtained for different VV restrict compatibly by the same uniqueness. This proves existence and uniqueness of the glued section. On a chart VV, the family is determined by its member sVs_V, so the sheaf recovers AVA_V and all its restrictions.

For a chart quasi-net σ⊆τˉ\sigma\subseteq\bar\tau, chart transitivity gives σˉ⊆τˉ\bar\sigma\subseteq\bar\tau. For two members of σˉ\bar\sigma, refine their intersection by the net τˉ\bar\tau. Each refining chart is contained in a member of σ\sigma and is an affinoid domain there by chart transport, hence belongs to σˉ\bar\sigma. Thus σˉ\bar\sigma is a net. Each chart V∈τˉV\in\bar\tau has an admissible cover by members of σˉ\bar\sigma: restrict σ\sigma to VV by intersections, then refine each intersection by chart domains. Quasi-net transitivity proves equality of the analytic domains and coverings. The just-proved gluing construction supplies the inverse to restriction of compatible section families. Its local uniqueness makes all comparison isomorphisms transitive. ◻

Definition 2.40.7. NAG-001611 (Locally bounded morphisms). For two affinoid atlases, a morphism consists of a continuous map f:X→X′f:X\to X' and, for every pair of charts U∈τˉU\in\bar\tau, V∈τˉ′V\in\bar\tau' with f(U)⊆Vf(U)\subseteq V, a bounded homomorphism fU,V:AV→AUf_{U,V}:A_V\to A_U inducing f∣Uf|_U. These maps commute with restrictions in both charts. We require that the charts UU admitting such a target chart VV form a quasi-net of XX. An analytic space is an affinoid atlas considered up to the canonical identifications under chart refinement of Theorem 2.40.6.

Theorem 2.40.8. NAG-001612 (Local extension and gluing of morphisms). Locally bounded morphisms form a category independent of atlas refinement. It is enough to specify compatible bounded affinoid maps on a chart quasi-net. Morphisms glue uniquely over admissible coverings. The inverse image of an analytic domain is analytic. The inclusion of an analytic domain Y⊆XY\subseteq X has the universal property Hom⁡(Z,Y)={f∈Hom⁡(Z,X):f(∣Z∣)⊆Y}.\operatorname{Hom}(Z,Y)= \{f\in\operatorname{Hom}(Z,X):f(|Z|)\subseteq Y\}.

Proof. Here compatibility of a collection of local maps means that, after refining a source overlap by common affinoid charts and refining the common target image by affinoid charts in the target overlap, the induced bounded homomorphisms coincide. Such refinements exist: target overlaps have their chart net, and an affinoid map pulls back each chart domain by Theorem 2.36.9. Pulling back a quasi-net along a continuous map preserves the quasi-net condition on its inverse image. Intersecting source chart nets then gives the required common source refinements.

Suppose the local maps have been specified, and take arbitrary charts U,VU,V with f(U)⊆Vf(U)\subseteq V. Intersect UU with the source charts on which the maps are given. On each piece, refine the target overlap between its original target chart and VV, then pull back those refinements. These are affinoid domains of UU and form a quasi-net of UU. Compactness gives a finite subcover. The resulting maps from AVA_V agree on all overlaps by the prescribed compatibility, so Lemma 2.40.2 gives fU,V:AV→AUf_{U,V}:A_V\to A_U. Any second finite refinement gives the same map after their common refinement. This proves existence, uniqueness and restriction compatibility of all maps required in Definition 2.40.7. The topological part glues by Lemma 2.40.2.

The inverse image assertion is local on a presenting source chart and a target chart. Cover the target domain by chart domains and pull them back using the affinoid-domain base-change theorem. The resulting quasi-net of affinoid domains proves the assertion. For a composite g∘fg\circ f, refine source charts by inverse images of presenting charts for gg, compose the bounded algebra maps there, and use the preceding extension construction. Common refinement and associativity of algebra homomorphisms prove associativity. The identity homomorphisms give identities. This construction also shows that replacing either atlas by a chart quasi-net does not change the morphism set.

For gluing along an admissible covering, present each local morphism on chart pieces contained in the relevant covering member. Their union is a quasi-net by transitivity. Agreement on pairwise intersections is exactly the compatibility used above, so extension gives a unique global morphism. Finally, if f(∣Z∣)⊆Yf(|Z|)\subseteq Y, pull back the chart net of YY. The local maps factor through its chart domains by the affinoid-domain universal property and glue to Z→YZ\to Y. Local uniqueness proves the displayed universal property. ◻

Corollary 2.40.9. NAG-001613 (Intrinsic affinoid domains). The analytic domains isomorphic to affinoid spaces form a canonical maximal affinoid atlas. If two such domains satisfy U⊆VU\subseteq V, then UU is an affinoid domain of VV in the original affinoid-algebra sense. Their algebras are OXG(U)\mathcal O_{X_G}(U), and all restriction maps are bounded. The analytic site and structure sheaf constructed from this maximal atlas equal those already constructed.

Proof. The inclusion U→VU\to V is a morphism between affinoid spaces and hence is induced by a bounded homomorphism. Its factorization property for affinoid tests follows from the analytic-domain universal property in Theorem 2.40.8; it is therefore an affinoid-domain map. Every original chart belongs to this family. The intersections of any two members are analytic domains and have affinoid chart quasi-nets, so the family is a net. Sections on each member are its affinoid algebra by the structure-sheaf theorem applied after refinement. Finally, every new chart already has an admissible covering by old chart pieces. The refinement argument of Theorem 2.40.6 gives equality of the sites and ring sheaves. ◻

Theorem 2.40.10. NAG-001614 (Gluing analytic spaces). Let (Xi)(X_i) be analytic spaces with analytic domains Xij⊆XiX_{ij}\subseteq X_i and isomorphisms vij:Xij≃Xjiv_{ij}:X_{ij}\simeq X_{ji} satisfying the identity and cocycle conditions. Their analytic gluing exists and is unique in each of the following cases:

  1. all XijX_{ij} are open in XiX_i;

  2. each XiX_i is Hausdorff, every XijX_{ij} is closed in XiX_i, and for each ii only finitely many XijX_{ij} are nonempty.

The images of the XiX_i are analytic domains and form an admissible cover. They are open in the first case. In the second case they are closed and locally finite, and the glued space is Hausdorff. The gluing has the usual mapping universal property.

Proof. Form the quotient XX of the disjoint union S=∐iXiS=\coprod_iX_i by the prescribed identifications. In the open case, the saturation of an open subset of one XiX_i is open in every XjX_j, so the quotient map is open and each XiX_i embeds as an open subspace. Hence XX is locally Hausdorff.

In the second case, the saturation of a closed subset of SS meets each XiX_i in a finite union of closed sets. Thus the quotient map qq is closed, and each XiX_i embeds as a closed subspace. Each equivalence class is finite, since all its representatives meet any one of its representatives’ components. Distinct classes are disjoint finite subsets of the Hausdorff space SS, so they have disjoint open neighborhoods P,QP,Q in SS. The complements of q(S∖P)q(S\setminus P) and q(S∖Q)q(S\setminus Q) are disjoint open neighborhoods of the corresponding quotient points. Thus XX is Hausdorff. Fix ii and let JiJ_i be the finite set of indices meeting XiX_i. The union of the images of XjX_j for j∉Jij\notin J_i is closed: on each component its inverse image is either the entire component or a finite union of closed overlaps. Its complement is an open neighborhood of the image of XiX_i meeting only the finitely many components indexed by JiJ_i. This proves local finiteness. A locally finite closed cover is a quasi-net, by applying the finite closed-cover assertion locally.

Take all images of affinoid chart domains in the XiX_i. Their family is a quasi-net by transitivity. On a pairwise intersection, the given analytic-domain isomorphism identifies the chart nets and their bounded restriction maps; refining inside the overlap gives the net axiom. If a chart from XiX_i is contained in a chart from XjX_j, transport it through Xij≃XjiX_{ij}\simeq X_{ji}. It is an affinoid analytic domain in XjX_j, so Corollary 2.40.9 supplies the required bounded affinoid-domain inclusion map. The cocycle identity gives composition compatibility. This is an affinoid atlas on XX. It identifies each XiX_i with its analytic-domain image. Theorem 2.40.8 gives the mapping universal property and uniqueness. ◻

Theorem 2.40.11. NAG-001615 (Fibre products of analytic spaces). The category of analytic spaces over KK has fibre products. If Y→X←ZY\to X\leftarrow Z are morphisms, the product P=Y×XZP=Y\times_XZ has an affinoid chart net consisting of V×UW=M(AV⊗^AUAW),V\times_UW= \mathcal M\bigl(A_V\widehat\otimes_{A_U}A_W\bigr), where U,V,WU,V,W are affinoid analytic domains of X,Y,ZX,Y,Z and both V,WV,W map into UU. Each such product is the analytic domain pY−1(V)∩pZ−1(W)p_Y^{-1}(V)\cap p_Z^{-1}(W) in PP. The canonical map ∣P∣→∣Y∣×∣X∣∣Z∣|P|\to |Y|\times_{|X|}|Z| is surjective. These constructions are compatible with restricting any of the three spaces to analytic domains.

Proof. Step 1: compact Hausdorff spaces. Suppose first that X,Y,ZX,Y,Z are compact Hausdorff. Choose a finite affinoid cover (Ua)(U_a) of XX. Its inverse images in YY and ZZ are compact analytic domains: they are analytic by Theorem 2.40.8 and closed because UaU_a is compact in the Hausdorff space XX. Choose finite affinoid covers (Vai)(V_{ai}) and (Waj)(W_{aj}) of these inverse images. Form the finitely many affinoid products Paij=Vai×UaWajP_{aij}=V_{ai}\times_{U_a}W_{aj} using Proposition 2.33.8.

For two triples α=(a,i,j)\alpha=(a,i,j) and β=(b,i′,j′)\beta=(b,i',j'), define Pαβ⊆PαP_{\alpha\beta}\subseteq P_\alpha by requiring its projections to lie in Vai∩Vbi′V_{ai}\cap V_{bi'} and Waj∩Wbj′W_{aj}\cap W_{bj'}. It is a compact analytic domain, since these are closed analytic domains of the Hausdorff spaces Y,ZY,Z. To compare it with PβαP_{\beta\alpha}, cover Ua∩UbU_a\cap U_b by finitely many affinoid domains RlR_l. Cover the inverse images of each RlR_l in Vai∩Vbi′V_{ai}\cap V_{bi'} and Waj∩Wbj′W_{aj}\cap W_{bj'} by finite affinoid families (Elm)(E_{lm}) and (Fln)(F_{ln}). The spaces Qlmn=Elm×RlFlnQ_{lmn}=E_{lm}\times_{R_l}F_{ln} are affinoid domains in both PαP_\alpha and PβP_\beta, by repeated use of Theorem 2.36.9. They cover the two designated overlaps: for a point of either overlap, first choose RlR_l containing its base point and then pieces containing its two projections. Their affinoid algebras are the same completed tensor products on both sides. On two such pieces, refine the base and both source intersections by common affinoid domains; associativity of the bounded pushout, Corollary 2.33.9, identifies the resulting tensor products. Thus the identifications agree on overlaps and give Pαβ≃PβαP_{\alpha\beta}\simeq P_{\beta\alpha} by Theorem 2.40.8. The same argument on triple intersections proves the cocycle condition.

The finite closed gluing of Theorem 2.40.10 now produces a compact Hausdorff analytic space PP. Its projections glue. Given any analytic TT with compatible maps to Y,ZY,Z, pull back the finite covers just used and refine by affinoid charts of TT. On each resulting affinoid piece the bounded-pushout property gives the unique map to an appropriate PαP_\alpha. On overlaps these maps agree after the common refinements QlmnQ_{lmn}, again by the bounded-pushout property. They therefore glue uniquely. This proves the product universal property in the compact Hausdorff case, with arbitrary analytic test spaces TT.

Step 2: a compact Hausdorff base. Keep XX compact Hausdorff and allow arbitrary Y,ZY,Z. By Proposition 2.40.5, cover Y,ZY,Z by open domains Oi,Oj′O_i,O'_j contained respectively in compact Hausdorff analytic domains Di,Dj′D_i,D'_j. Step 1 constructs Di×XDj′D_i\times_XD'_j. Restrict it to the open domain where the projections lie in Oi,Oj′O_i,O'_j. The result represents the product of these two open domains over XX, by the analytic-domain universal property. For two index pairs, the corresponding open subdomains over Oi∩Oi′O_i\cap O_{i'} and Oj′∩Oj′′O'_j\cap O'_{j'} have the same universal property. Their unique comparison isomorphisms satisfy the cocycle condition. Open gluing constructs Y×XZY\times_XZ. The universal property follows by restricting a test space to the inverse images of these open domains and gluing its local maps.

Step 3: an arbitrary base. Cover XX by open domains SlS_l contained in compact Hausdorff analytic domains ClC_l. Put Yl=f−1(Sl)Y_l=f^{-1}(S_l) and Zl=g−1(Sl)Z_l=g^{-1}(S_l). Step 2 constructs Yl×ClZlY_l\times_{C_l}Z_l. Because both maps have image in SlS_l, its universal property is also that of Yl×SlZlY_l\times_{S_l}Z_l. The products over two base opens agree on the open subdomains over their intersection. Open gluing therefore gives the required PP for arbitrary X,Y,ZX,Y,Z.

Finally take affinoid U,V,WU,V,W as in the statement. The analytic domain pY−1(V)∩pZ−1(W)p_Y^{-1}(V)\cap p_Z^{-1}(W) has the same universal property as the affinoid bounded pushout V×UWV\times_UW, since two maps into UU agree whenever their composites into XX agree. Thus they are canonically isomorphic. The construction shows that these domains form a chart quasi-net, and the same observation proves compatibility with arbitrary analytic-domain restriction. Surjectivity on underlying fibre-product sets is local on these charts and follows from Proposition 2.33.14. ◻

Theorem 2.40.12. NAG-001616 (Arbitrary complete ground-field extension). Let L/KL/K be any complete valued extension, with its specified isometric embedding. There is a scalar-extension functor X↦XLX\mapsto X_L on analytic spaces, characterized locally by M(A)L=M(A⊗^KL).\mathcal M(A)_L=\mathcal M(A\widehat\otimes_KL). The canonical continuous map π:XL→X\pi:X_L\to X is surjective. For every analytic domain Y⊆XY\subseteq X, its inverse image is canonically YLY_L. Scalar extension commutes with fibre products and with successive complete field extensions.

Proof. We first make the local mapping characterization explicit. A bounded map over KK from an LL-analytic space to a KK-analytic space is defined by the chart condition in Definition 2.40.7, using bounded KK-homomorphisms from the target KK-affinoid algebra to the source LL-affinoid algebra. The proof of Theorem 2.40.8 applies to these maps as well: all pullbacks of chart domains use Theorem 2.36.7 and Theorem 2.36.9. For an affinoid target, the bounded-pushout property gives Hom⁡L(T,M(A⊗^KL))≃Hom⁡K,loc(T,M(A)),\operatorname{Hom}_L(T,\mathcal M(A\widehat\otimes_KL)) \simeq \operatorname{Hom}_{K,\mathrm{loc}}(T,\mathcal M(A)), first on affinoid charts of TT and then by gluing.

If XX is compact Hausdorff, take a finite affinoid cover (Ui)(U_i). Each intersection Ui∩UjU_i\cap U_j is a compact analytic domain and has a finite affinoid cover. Its inverse image in (Ui)L(U_i)_L is therefore a finite union of the corresponding scalar-extended affinoid domains, hence a compact closed analytic domain. The analogous inverse image in (Uj)L(U_j)_L is canonically isomorphic to it: identify the scalar extensions of common affinoid pieces, and use the completed base-change identity on their intersections. These isomorphisms satisfy the cocycle condition by local uniqueness. Finite closed gluing produces XLX_L. The displayed local mapping characterization glues to Hom⁡L(T,XL)≃Hom⁡K,loc(T,X).\operatorname{Hom}_L(T,X_L) \simeq\operatorname{Hom}_{K,\mathrm{loc}}(T,X).

For general XX, cover it by open domains OiO_i contained in compact Hausdorff analytic domains DiD_i. Construct (Di)L(D_i)_L and restrict to the open inverse image of OiO_i. On the inverse images of overlaps, the same mapping characterization gives unique comparison isomorphisms satisfying the cocycle condition. Open gluing constructs XLX_L and proves the displayed characterization for all XX. It also makes X↦XLX\mapsto X_L functorial.

The inverse image of an analytic domain is analytic by the mixed-base version of the local pullback argument. Its mapping characterization is that of YLY_L, so the two spaces are canonically isomorphic. In particular, the inverse image of any affinoid chart is its affinoid scalar extension. Theorem 2.33.13 now proves surjectivity of π\pi. For successive extensions, the canonical bounded isomorphism of the iterated completed tensor products in Corollary 2.33.9 gives the local comparison, and uniqueness glues it. The same corollary identifies the scalar extension of each affinoid product chart with the product of the extended charts. Since such charts form a quasi-net by Theorem 2.40.11, these local identifications glue and prove compatibility with fibre products. ◻

Theorem 2.40.13. NAG-001617 (Points and fibres on general analytic spaces). For a point xx of an analytic space, the completed residue fields computed in affinoid domains containing xx identify canonically and isometrically with a field H(x)\mathcal H(x). For a morphism f:Y→Xf:Y\to X, the fibre at xx is a canonical H(x)\mathcal H(x)-analytic space YxY_x. On affinoid domains V⊆YV\subseteq Y, U⊆XU\subseteq X with f(V)⊆Uf(V)\subseteq U and x∈Ux\in U, it has the affinoid chart Vx=M(AV⊗^AUH(x)).V_x=\mathcal M\bigl(A_V\widehat\otimes_{A_U}\mathcal H(x)\bigr). The canonical map Yx→YY_x\to Y identifies its underlying topological space with f−1(x)f^{-1}(x). It preserves the completed residue field at every point.

Proof. Any two affinoid domains containing xx have a common affinoid domain containing xx, by the net condition. Corollary 2.33.15 identifies the fields along each inclusion. The isomorphisms satisfy the cocycle condition because they agree on evaluations, so they define F=H(x)F=\mathcal H(x). A morphism induces an isometric embedding of these fields, checked on any presenting affinoid chart.

Evaluation on affinoid domains containing xx defines a bounded map over KK from the FF-affinoid point M(F)\mathcal M(F) to XX. By Theorem 2.40.12, it corresponds to an FF-analytic point map ix:M(F)→XFi_x:\mathcal M(F)\to X_F. Define Yx=YF×XF,ixM(F).Y_x=Y_F\times_{X_F,i_x}\mathcal M(F). Theorem 2.40.11 and the affinoid pushout identity give the displayed chart formula. Theorem 2.33.11 identifies each VxV_x homeomorphically with V∩f−1(x)V\cap f^{-1}(x) and preserves the completed residue fields. The charts VxV_x form a quasi-net of YxY_x, while the corresponding subsets of f−1(x)f^{-1}(x) are the restriction of a presenting chart quasi-net of YY. These local homeomorphisms agree on overlaps. The topology criterion of Lemma 2.40.2, applied on both sides, makes their glued bijection a homeomorphism. The field identifications glue by the same restriction compatibility. ◻

2.41 Relative differentials and Jacobian presentations

We construct analytic differentials from convergent presentations and then express the Jacobian condition through a finite matrix. This is the presentation-theoretic part of the approach in [Duc18] (§§5.1–5.2). All finite modules below carry their canonical complete norms.

Lemma 2.41.1. NAG-001641 (Inner rational neighborhoods). Let x∈X=M(A)x\in X=\mathcal M(A) and let OO be an open neighborhood of xx. There is a rational domain U⊆OU\subseteq O which is a neighborhood of xx and for which x∈Int⁡(U/X)x\in\operatorname{Int}(U/X). It can be chosen as a finite intersection of Weierstrass and Laurent domains whose defining inequalities are all strict at xx.

Proof. The evaluation topology gives finitely many open conditions ∣fi∣<ai|f_i|<a_i and ∣gj∣>bj|g_j|>b_j whose common locus contains xx and is contained in OO. Choose radii ∣fi(x)∣<ri<ai|f_i(x)|<r_i<a_i and bj<sj<∣gj(x)∣b_j<s_j<|g_j(x)|, omitting lower conditions with no effective restriction. The domain U={∣fi∣≤ri, ∣gj∣≥sj},AU=A{r−1T,sS}/(Ti−fi, gjSj−1).U=\{|f_i|\le r_i,\ |g_j|\ge s_j\},\qquad A_U=A\{r^{-1}T,sS\}/(T_i-f_i,\ g_jS_j-1). The generalized rational-domain theorem identifies this actual quotient with the domain algebra. The joint quotient is norm-admissible, and its variables satisfy ∣Ti(x)∣<ri|T_i(x)|<r_i and ∣Sj(x)∣<sj−1|S_j(x)|<s_j^{-1}. It therefore witnesses the claimed relative interior. The same strict inequalities define an open neighborhood of xx contained in UU. ◻

Lemma 2.41.2. NAG-001642 (Lifting rational domains through a closed immersion). Let B=R/IB=R/I be a norm-admissible quotient of affinoid algebras. Every rational domain W⊆M(B)W\subseteq\mathcal M(B) is the inverse image of a rational domain V⊆M(R)V\subseteq\mathcal M(R). Moreover BW=RV/IRVB_W=R_V/IR_V as Banach algebras.

Proof. Write W={∣bi∣≤ρi∣b0∣}W=\{|b_i|\le\rho_i|b_0|\}, with (b0,…,bm)=B(b_0,\ldots,b_m)=B. Choose lifts gi∈Rg_i\in R and coefficients ai∈Ra_i\in R such that h=1−∑iaigih=1-\sum_i a_ig_i belongs to II. For any τ>0\tau>0, let VV be defined by ∣gi∣≤ρi∣g0∣(1≤i≤m),∣h∣≤τ∣g0∣.|g_i|\le\rho_i|g_0|\quad(1\le i\le m),\qquad |h|\le\tau|g_0|. The denominator and all numerators generate the unit ideal, because 1=h+∑iaigi1=h+\sum_i a_ig_i. On M(B)\mathcal M(B) the final numerator vanishes and the other conditions become precisely those defining WW. The base-change formula of Theorem 2.36.9 gives the displayed quotient; the ideal IRVIR_V is finitely generated and closed. ◻

Theorem 2.41.3. NAG-001643 (Bounded universal differentials). Let A→BA\to B be a bounded affinoid homomorphism and choose a norm-admissible relative presentation P=A{r−1T}↠B=P/I,I=(f1,…,fm).P=A\{r^{-1}T\}\twoheadrightarrow B=P/I,\qquad I=(f_1,\ldots,f_m). The finite BB-module ΩB/Aan=(⨁j=1nB dTj)/(∑j=1n∂fi∂Tj dTj:1≤i≤m)\Omega^{\mathrm{an}}_{B/A} =\left(\bigoplus_{j=1}^n B\,dT_j\right) \big/\left(\sum_{j=1}^n \frac{\partial f_i}{\partial T_j}\,dT_j: 1\le i\le m\right) has a bounded AA-derivation d:B→ΩB/Aand:B\to\Omega^{\mathrm{an}}_{B/A} which represents bounded AA-derivations from BB to every complete Banach BB-module whose action B×N→NB\times N\to N is bounded. In particular this applies to finite BB-modules, and its canonical pair (ΩB/Aan,d)(\Omega^{\mathrm{an}}_{B/A},d) is independent of the presentation.

Proof. Termwise differentiation is bounded on PP: since every integer has absolute value at most one, ∥∂F/∂Tj∥≤rj−1∥F∥\|\partial F/\partial T_j\|\le r_j^{-1}\|F\|. For a complete Banach BB-module NN with bounded action and elements v1,…,vn∈Nv_1,\ldots,v_n\in N, the formula Dv(F)=∑j(∂F∂Tj mod I)vjD_v(F)=\sum_j\left(\frac{\partial F}{\partial T_j}\bmod I\right)v_j defines a bounded AA-derivation. The Leibniz rule holds first for polynomials and then for series by continuity. Conversely, polynomial density shows that every bounded AA-derivation on PP is this one with vj=D(Tj)v_j=D(T_j). It kills II exactly when Dv(fi)=0D_v(f_i)=0 for every ii: the Leibniz rule and the fact that II acts trivially on NN then give Dv(∑hifi)=∑hiDv(fi)=0D_v(\sum h_if_i)=\sum h_iD_v(f_i)=0.

A bounded map killing II induces a bounded map on BB because the quotient is norm-admissible. The relation module in the displayed definition is closed by Theorem 1.18.2. Thus the quotient is a finite complete module. Every BB-linear map from it to NN is bounded: take controlled coefficient lifts along a finite free presentation, and bound the images of the finitely many generators using the bounded action on NN. We have therefore proved the natural bijection Hom⁡B(ΩB/Aan,N)≃Der⁡Abd(B,N).\operatorname{Hom}_B(\Omega^{\mathrm{an}}_{B/A},N) \simeq\operatorname{Der}^{\mathrm{bd}}_A(B,N). Taking NN to be the two differential modules constructed from two presentations gives mutually inverse maps commuting with dd. Uniqueness also proves compatibility with every further presentation change. ◻

Proposition 2.41.4. NAG-001644 (Base change and transitivity of differentials). Let L/KL/K be complete and let A→CA\to C be a bounded map to an LL-affinoid algebra. Put B′=B⊗^ACB'=B\widehat\otimes_A C. Then B′⊗BΩB/Aan≃ΩB′/Can.B'\otimes_B\Omega^{\mathrm{an}}_{B/A} \simeq\Omega^{\mathrm{an}}_{B'/C}. For bounded affinoid maps A→B→DA\to B\to D there is a right-exact sequence of finite DD-modules D⊗BΩB/Aan⟶ΩD/Aan⟶ΩD/Ban⟶0.D\otimes_B\Omega^{\mathrm{an}}_{B/A} \longrightarrow\Omega^{\mathrm{an}}_{D/A} \longrightarrow\Omega^{\mathrm{an}}_{D/B}\longrightarrow0.

Proof. Proposition 2.33.8 gives B′=C{r−1T}/(f1,…,fm)B'=C\{r^{-1}T\}/(f_1,\ldots,f_m). The Jacobian matrix of the base-changed equations is the image of the original matrix. Tensoring a finite cokernel is right exact, so Theorem 2.41.3 proves the first formula. These are ordinary tensor products of finite modules: Proposition 1.18.7 shows that completion does not change them.

For transitivity let QQ be the cokernel of the first map. It is finite and complete. A DD-linear map Q→NQ\to N is the same as a bounded AA-derivation D→ND\to N that vanishes on BB, hence the same as a bounded BB-derivation. The universal property identifies QQ with ΩD/Ban\Omega^{\mathrm{an}}_{D/B} and identifies the quotient map with the canonical differential map. This proves the exact sequence. ◻

Theorem 2.41.5. NAG-001645 (Differentials on analytic domains). If CC is the algebra of an affinoid domain in M(B)\mathcal M(B), then for every bounded affinoid map A→BA\to B the natural map is an isomorphism C⊗BΩB/Aan→ ∼ ΩC/Aan.C\otimes_B\Omega^{\mathrm{an}}_{B/A} \xrightarrow{\ \sim\ }\Omega^{\mathrm{an}}_{C/A}. In particular ΩC/Ban=0\Omega^{\mathrm{an}}_{C/B}=0. For an arbitrary analytic morphism f:Y→Xf:Y\to X, these modules determine a coherent sheaf ΩY/X\Omega_{Y/X} on YGY_G, compatible with restriction to analytic domains and with every analytic base change over a complete ground-field extension.

Proof. First suppose C=B{ρ−1S}/(gSi−fi)C=B\{\rho^{-1}S\}/(gS_i-f_i) is rational. The unit-ideal condition makes gg invertible in CC. For a finite CC-module NN, every bounded AA-derivation D:B→ND:B\to N has a unique possible extension to CC, namely D(Si)=g−1(D(fi)−SiD(g)).D(S_i)=g^{-1}\bigl(D(f_i)-S_iD(g)\bigr). These values do give an extension. On a series ∑νbνSν\sum_\nu b_\nu S^\nu, differentiate coefficients using DD and variables using the displayed values. The coefficientwise derivative term converges because DD is bounded; the variable term converges by the weighted derivative estimate used above and the uniform power bounds on the images of the SiS_i. Both estimates are uniform in the series norm. The resulting bounded derivation kills gSi−figS_i-f_i and hence their ideal, so it descends through the norm-admissible quotient. Thus restriction induces a natural bijection on derivations, which proves the differential isomorphism by the universal property. Taking A=BA=B proves the vanishing.

For a general domain, Theorem 2.36.6 supplies a finite cover by domains rational in M(B)\mathcal M(B). They are also rational in M(C)\mathcal M(C): pull back their defining inequalities and use the domain universal property. On each covering algebra CiC_i, the rational result identifies both sides of the proposed map with ΩCi/Aan\Omega^{\mathrm{an}}_{C_i/A}. The finite affinoid cover is faithfully flat on finite modules, by Theorem 2.38.1 and Proposition 2.38.2; applying it to the kernel and cokernel proves that the original map is an isomorphism.

For sheaf gluing, choose affinoid chart pairs V→UV\to U of ff. The finite modules just constructed restrict compatibly under source-chart refinement. If U′⊆UU'\subseteq U is an affinoid domain and V→UV\to U factors through U′U', then V=V×UU′V=V\times_UU' by the domain property; Proposition 2.41.4 gives the same differential module with base U′U'. Thus target-chart refinement is compatible as well. Common affinoid refinements and uniqueness of the universal derivation give the cocycle identities. Coherent-module gluing from Theorem 2.40.6 and the affinoid Kiehl equivalence produces the coherent sheaf. The base-change formula holds on every affinoid chart pair by Proposition 2.41.4; these canonical isomorphisms glue. ◻

Proposition 2.41.6. NAG-001646 (The conormal descriptions). For an affinoid morphism A→BA\to B, let JJ be the kernel of multiplication B⊗^AB→BB\widehat\otimes_A B\to B. Then J/J2J/J^2 is canonically isomorphic to ΩB/Aan\Omega^{\mathrm{an}}_{B/A}, with dbdb corresponding to 1⊗b−b⊗11\otimes b-b\otimes 1. If W↪VW\hookrightarrow V is a closed immersion over an analytic space XX, with coherent ideal I\mathcal I, there is an exact sequence I/I2→ d ΩV/X∣W⟶ΩW/X⟶0.\mathcal I/\mathcal I^2\xrightarrow{\ d\ } \Omega_{V/X}|_W\longrightarrow\Omega_{W/X} \longrightarrow0. In particular ΩAXn,an/X\Omega_{\mathbb A_X^{n,\mathrm{an}}/X} is free with basis dT1,…,dTndT_1,\ldots,dT_n, and the same basis restricts to every analytic domain of this relative affine space.

Proof. Choose relative topological generators b1,…,bnb_1,\ldots,b_n of BB. The ideal generated by δbi=1⊗bi−bi⊗1\delta b_i=1\otimes b_i-b_i\otimes 1 is closed. In its quotient the two bounded maps from BB agree on AA and on the bib_i, hence on all of BB by polynomial density. Multiplication and the first-factor map then give inverse bounded maps between this quotient and BB. Thus J=(δb1,…,δbn)J=(\delta b_1,\ldots,\delta b_n).

For a bounded AA-derivation D:B→ND:B\to N, the bounded bilinear rule b⊗c↦bD(c)b\otimes c\mapsto bD(c) extends to the completed tensor product; call the extension FF. The identity F(uv)=μ(u)F(v)+μ(v)F(u)F(uv)=\mu(u)F(v)+\mu(v)F(u) holds on algebraic tensors and then everywhere by continuity. Hence FF induces a BB-linear map J/J2→NJ/J^2\to N sending δb\delta b to D(b)D(b). It is unique because the δbi\delta b_i generate J/J2J/J^2. Conversely composition with b↦δbb\mapsto\delta b gives a bounded derivation. This is the same universal property as in Theorem 2.41.3, proving the first assertion. It also identifies the sheaf constructed above with the diagonal conormal sheaf.

For a quotient C=B/IC=B/I, a derivation from BB to a finite CC-module factors through CC precisely when it kills II. Since D(ab)=aD(b)D(ab)=aD(b) for b∈Ib\in I, the obstruction is the induced map I/I2→NI/I^2\to N. This gives the stated cokernel on affinoid chart pairs. The constructions commute with domain restriction, so the sequences glue on the analytic site. Finally relative Tate algebras have free differential module by their presentation with no equations. These free bases agree as the polydisc radii and base charts vary; relative affine space is their union, and Theorem 2.41.5 gives the assertion for its domains. ◻

Theorem 2.41.7. NAG-001647 (The matrix criterion for a Jacobian presentation). Let VV be an affinoid domain of AXn,an\mathbb A_X^{n,\mathrm{an}}, and let W↪VW\hookrightarrow V be the closed immersion defined by the ideal (f1,…,fm)(f_1,\ldots,f_m). Write dfi=∑j=1nJij dTj.df_i=\sum_{j=1}^n J_{ij}\,dT_j. For every y∈Wy\in W, dim⁡H(y)(ΩW/X⊗H(y))=n−rank⁡H(y)J(y).\dim_{\mathcal H(y)}(\Omega_{W/X}\otimes\mathcal H(y)) =n-\operatorname{rank}_{\mathcal H(y)}J(y). Consequently this presentation is Jacobian at yy, with m=n−rk⁡yΩW/Xm=n-\operatorname{rk}_y\Omega_{W/X}, if and only if an m×mm\times m minor is nonzero at yy. In that case mm is the minimum number of generators of the ideal. There is an affinoid neighborhood V′⊆VV'\subseteq V of yy on which such a minor is invertible; on W∩V′W\cap V' the differential sheaf is free of rank n−mn-m and the restricted presentation is Jacobian at every point.

Proof. Proposition 2.41.6 gives a presentation OWm→JOWn→ΩW/X→0\mathcal O_W^m\xrightarrow{J}\mathcal O_W^n\to\Omega_{W/X}\to 0. Tensoring with H(y)\mathcal H(y) is right exact, so elementary linear algebra gives the rank formula. Full row rank is equivalent to the existence of the indicated nonzero minor. If the same ideal had q<mq<m generators, the same cokernel computation would give rk⁡yΩW/X≥n−q>n−m\operatorname{rk}_y\Omega_{W/X}\ge n-q>n-m, a contradiction.

Let Δ\Delta be a nonzero minor at yy and choose 0<c<∣Δ(y)∣0<c<|\Delta(y)|. The Laurent domain V′={∣Δ∣≥c}V'=\{|\Delta|\ge c\} is a neighborhood of yy, and Δ\Delta is a unit on it. Permuting the coordinate columns, its invertible block expresses the first mm basis vectors modulo the relations as linear combinations of the remaining n−mn-m. Those remaining vectors are independent, since a relation among them would have zero first mm coordinates and invertibility forces every relation coefficient to vanish. They form a free basis of the cokernel. The same invertible minor at every point proves the final assertion. If m=0m=0, take Δ=1\Delta=1 and V′=VV'=V. ◻

Theorem 2.41.8. NAG-001648 (Domain locality and base change of quasi-smoothness). Use the Jacobian-presentation definition of quasi-smoothness in Section 2.9. Let f:Y→Xf:Y\to X, let y∈Yy\in Y, let DD be an analytic domain of YY containing yy, and let UU be an analytic domain of XX containing f(D)f(D). Then ff is quasi-smooth at yy if and only if D→UD\to U is quasi-smooth there. Quasi-smoothness is preserved by every analytic base change, including complete ground-field extension. Analytic-domain inclusions are quasi-smooth and have zero relative differential sheaf.

Proof. Start with a Jacobian presentation W↪V⊆AXn,anW\hookrightarrow V\subseteq\mathbb A_X^{n,\mathrm{an}} at yy, with mm equations. To restrict the target, choose an affinoid domain V0V_0 in V∩AUn,anV\cap\mathbb A_U^{n,\mathrm{an}} containing the image of yy. Its inverse image W0W_0 is affinoid and closed in V0V_0, with the restricted mm equations. The differential comparison of Theorem 2.41.5 identifies the relevant fibre ranks, so the same minor is nonzero at yy.

To restrict the source, choose an affinoid domain E⊆D∩W0E\subseteq D\cap W_0 containing yy. Theorem 2.36.6 gives a finite cover of EE by domains rational in W0W_0; choose a member E0E_0 containing yy. Lemma 2.41.2 lifts E0E_0 to a rational domain V1⊆V0V_1\subseteq V_0 with E0=W0×V0V1E_0=W_0\times_{V_0}V_1. Thus E0↪V1⊆AUn,anE_0\hookrightarrow V_1\subseteq\mathbb A_U^{n,\mathrm{an}} has the same equations and the same nonzero minor at yy, and is the required presentation. Conversely a Jacobian presentation for D→UD\to U is also one for Y→XY\to X: domain inclusions are transitive, AUn\mathbb A_U^n is a domain of AXn\mathbb A_X^n, and the differential fibres agree.

For base change by X′→XX'\to X, let y′∈Y×XX′y'\in Y\times_XX' lie over yy. Pull back W↪VW\hookrightarrow V and take an affinoid domain V′V' in V×XX′V\times_XX' containing the image of y′y'. The inverse image W′W' is an affinoid domain of Y×XX′Y\times_XX' closed in V′V', and V′V' is a domain of AX′n\mathbb A_{X'}^n. The mm pulled-back equations generate its ideal by the completed-pushout calculation. Proposition 2.41.4 identifies the differential fibres by extension from H(y)\mathcal H(y) to H(y′)\mathcal H(y'); the nonzero minor stays nonzero under this isometric field embedding. This proves base-change stability. The argument works verbatim over any complete extension. Finally an analytic-domain inclusion has locally the presentation W=W⊆AX0=XW=W\subseteq\mathbb A_X^0=X, with no equations, and its differential sheaf vanishes by Theorem 2.41.5. ◻

Theorem 2.41.9. NAG-001649 (Composition of Jacobian presentations). Let Z→gY→fXZ\xrightarrow{g}Y\xrightarrow{f}X be analytic morphisms and put y=g(z)y=g(z). If gg is quasi-smooth at zz and ff is quasi-smooth at yy, then fgfg is quasi-smooth at zz. The differential fibre ranks satisfy rk⁡zΩZ/X=rk⁡yΩY/X+rk⁡zΩZ/Y.\operatorname{rk}_z\Omega_{Z/X} =\operatorname{rk}_y\Omega_{Y/X} +\operatorname{rk}_z\Omega_{Z/Y}.

Proof. Write d=rk⁡yΩY/Xd=\operatorname{rk}_y\Omega_{Y/X} and e=rk⁡zΩZ/Ye=\operatorname{rk}_z\Omega_{Z/Y}. Choose W↪V⊆AXnW\hookrightarrow V\subseteq\mathbb A_X^n at yy with n−dn-d equations. By domain locality replace the part of ZZ under consideration by its inverse image over WW, and choose a Jacobian presentation T↪S⊆AWmT\hookrightarrow S\subseteq\mathbb A_W^m at zz with m−em-e equations.

Since SS is affinoid, its mm coordinate functions are bounded. Hence it lies in a closed relative polydisc PWP_W over WW, the inverse image of a closed relative polydisc PP over VV. It is an affinoid domain of PWP_W by the domain mapping property. Cover SS by rational domains of PWP_W and choose one S0S_0 containing the image of zz. Lemma 2.41.2 gives a rational domain P0⊆PP_0\subseteq P with P0∩PW=S0P_0\cap P_W=S_0. In particular P0P_0 is an affinoid domain of AXn+m\mathbb A_X^{n+m}. Put T0=T×SS0T_0=T\times_SS_0.

The ideal of S0S_0 in P0P_0 is generated by the original n−dn-d equations. The ideal of T0T_0 in S0S_0 is generated by the restricted m−em-e equations; lift these through the surjection of affinoid algebras O(P0)→O(S0)\mathcal O(P_0)\to\mathcal O(S_0). The union of the two lists generates the ideal of T0T_0 in P0P_0. Thus the rank formula of Theorem 2.41.7 gives rk⁡zΩZ/X=rk⁡zΩT0/X≥(n+m)−((n−d)+(m−e))=d+e.\operatorname{rk}_z\Omega_{Z/X} =\operatorname{rk}_z\Omega_{T_0/X} \ge(n+m)-((n-d)+(m-e))=d+e. The transitivity sequence of Proposition 2.41.4, tensored with H(z)\mathcal H(z), gives the opposite inequality. The first term has dimension dd, because its fibre is obtained from the fibre at yy by a field extension, and the final term has dimension ee. Hence equality holds. The immersion T0↪P0⊆AXn+mT_0\hookrightarrow P_0\subseteq\mathbb A_X^{n+m} has exactly the required number of equations, so is a Jacobian presentation at zz. ◻

Proposition 2.41.10. NAG-001650 (Algebraic differentials under analytification). Let Y→XY\to X be a morphism of finite-type KK-schemes. The pullback of the algebraic differential sheaf to YanY^{\mathrm{an}} is canonically isomorphic to ΩYan/Xan\Omega_{Y^{\mathrm{an}}/X^{\mathrm{an}}}, compatibly with universal derivations. More concretely, if S=R[T1,…,Tn]/(f1,…,fm)S=R[T_1,\ldots,T_n]/(f_1,\ldots,f_m), V=M(A)⊆(Spec⁡R)anV=\mathcal M(A)\subseteq(\operatorname{Spec}R)^{\mathrm{an}}, and D=A{r−1T}/(f1,…,fm)D=A\{r^{-1}T\}/(f_1,\ldots,f_m) is a relative radius piece, then D⊗SΩS/R≃ΩD/Aan.D\otimes_S\Omega_{S/R}\simeq\Omega^{\mathrm{an}}_{D/A}.

Proof. For polynomial rings the Leibniz rule shows directly that an RR-derivation is determined arbitrarily by its values on the TjT_j. It descends to SS precisely when it kills the fif_i. Thus ΩS/R\Omega_{S/R} is the cokernel of the polynomial Jacobian matrix (∂fi/∂Tj)(\partial f_i/\partial T_j). Tensoring this finite presentation with DD gives exactly the presentation of Theorem 2.41.3, proving the formula. Both maps send the differential of a polynomial to the differential of its analytic image. Uniqueness makes them compatible with algebraic localization, changes of generators and analytic domain restriction. The relative radius pieces cover the inverse images of affine scheme charts by Proposition 2.33.27; hence these isomorphisms glue to the asserted sheaf isomorphism. ◻

2.42 Cohomology of analytic projective space

The standard finite affinoid cover of projective space makes its line-bundle cohomology an explicit calculation with convergent Laurent series. The monomial contraction used in the algebraic computation [StacksAlg] (Tag 01XT) has norm at most one, and therefore acts on the analytic coefficient spaces as well.

Proposition 2.42.1. NAG-001891 (Cohomology on a finite acyclic cover). Let XX be an analytic space with an admissible finite cover (Ui)0≤i≤n(U_i)_{0\leq i\leq n} such that all nonempty finite intersections are affinoid domains. For a coherent OXG\mathcal O_{X_G}-module F\mathcal F, the ordered Čech complex of this cover computes Hq(XG,F)H^q(X_G,\mathcal F). In particular these groups vanish for q>nq>n.

Proof. The restriction of F\mathcal F to every intersection is coherent. Theorem 2.39.6 gives zero higher cohomology on that intersection. Thus the hypotheses of [StacksAlg] (Tag 03F7, Lemma 21.10.7) hold for the covering on XGX_G, including repeated-index intersections, and the all-tuples Čech complex computes derived section cohomology. The ordered-to-all-tuples homotopy comparison used in Theorem 2.21.3 applies to this finite intersection coefficient diagram. It identifies the same groups with the ordered complex, which has no terms above degree nn. ◻

Lemma 2.42.2. NAG-001892 (Laurent coefficient normal form). For 0≤ℓ≤s0\leq \ell\leq s, the quotient Banach algebra K⟨u1,…,us,v1,…,vℓ⟩/(u1v1−1,…,uℓvℓ−1)K\langle u_1,\ldots,u_s,v_1,\ldots,v_\ell\rangle/ (u_1v_1-1,\ldots,u_\ell v_\ell-1) is isometrically the space of series ∑α∈Zℓ×Ns−ℓcαuα\sum_{\alpha\in\mathbf Z^\ell\times\mathbf N^{s-\ell}} c_\alpha u^\alpha whose coefficients tend to zero, with the maximum norm and Laurent-series multiplication.

Proof. Replace a monomial uavbu^av^b by the monomial with exponent (a1−b1,…,aℓ−bℓ,aℓ+1,…,as)(a_1-b_1,\ldots,a_\ell-b_\ell,a_{\ell+1},\ldots,a_s). The coefficient attached to any one exponent is a convergent sum: for each positive ε\varepsilon, the original Tate series has only finitely many coefficients of size at least ε\varepsilon. The same observation proves that the resulting coefficients tend to zero and that this reduction is contractive. Conversely represent a negative exponent by the corresponding power of viv_i, using no simultaneous positive powers of ui,viu_i,v_i. This gives an isometric linear section of the reduction map.

For a monomial, its difference from its normal form is a sum of multiples of uivi−1u_iv_i-1, by successively using (uivi)d−1=(uivi−1)∑a=0d−1(uivi)a(u_iv_i)^d-1=(u_iv_i-1)\sum_{a=0}^{d-1}(u_iv_i)^a. The multipliers have coefficient norms at most one. Apply this identity to each term of a Tate series. For any ε>0\varepsilon>0, only finitely many original terms have coefficients of size at least ε\varepsilon, and each of them produces finitely many terms in its multipliers. Consequently the multiplier series converge in the Tate algebra. This proves that the kernel is the actual ideal displayed above. The contractive reduction and its isometric section identify its quotient norm with the maximum coefficient norm. Laurent convolution is well-defined by the same finite-large-coefficient argument and agrees with polynomial multiplication on a dense subalgebra, which proves the algebra assertion. ◻

Theorem 2.42.3. NAG-001893 (Twists on analytic projective space). Let P=(PKn)anP=(\mathbf P^n_K)^{\mathrm{an}}, and let OP(m)\mathcal O_P(m) be the line bundle obtained by analytifying the usual homogeneous transition functions. For every integer mm, the canonical map Hq(PKn,O(m))⟶Hq(PG,OP(m))H^q(\mathbf P^n_K,\mathcal O(m)) \longrightarrow H^q(P_G,\mathcal O_P(m)) is an isomorphism for every q≥0q\geq 0, compatibly with complete extensions of KK. If n≥1n\geq 1, the only possible nonzero groups have bases H0(PG,OP(m))=⨁νj≥0∑jνj=mKXν,Hn(PG,OP(m))=⨁νj<0∑jνj=mKXν.\begin{aligned} H^0(P_G,\mathcal O_P(m)) &=\bigoplus_{\substack{\nu_j\geq0\\\sum_j\nu_j=m}}KX^\nu,\\ H^n(P_G,\mathcal O_P(m)) &=\bigoplus_{\substack{\nu_j<0\\\sum_j\nu_j=m}}KX^\nu. \end{aligned} For n=0n=0, H0=KH^0=K for all mm, and all higher groups are zero. In the standard finite affinoid Čech complex, the projection onto these finite-dimensional representatives and a homotopy on its complement can be chosen with norm at most one.

Proof. Use the finite affinoid covering Ui={∣Xj∣≤∣Xi∣ for all j}U_i=\{|X_j|\leq|X_i|\text{ for all }j\}, constructed in Proposition 2.33.30. On UI=⋂i∈IUiU_I=\bigcap_{i\in I}U_i, fix i∈Ii\in I. The ratios Xj/XiX_j/X_i have norm at most one, and the ratios with j∈Ij\in I are units with inverse of norm at most one. Lemma 2.42.2, using the frame XimX_i^m, identifies the section space of OP(m)\mathcal O_P(m) with the maximum-norm space of coefficient families tending to zero on ΛI(m)={ν∈Zn+1:∑νj=m, νj≥0 if j∉I}.\Lambda_I(m)= \{\nu\in\mathbf Z^{n+1}:\textstyle\sum\nu_j=m,\ \nu_j\geq0\text{ if }j\notin I\}. Restrictions are the inclusions of these coefficient spaces. Every UIU_I is affinoid, so Proposition 2.42.1 computes the required cohomology by this ordered complex.

For a fixed exponent vector ν\nu, put N(ν)={j:νj<0}N(\nu)=\{j:\nu_j<0\}. Its coefficient occurs exactly in the components indexed by nonempty II containing N(ν)N(\nu). The differential is the alternating simplex differential. If N(ν)N(\nu) is nonempty and proper, choose the least index k∉N(ν)k\notin N(\nu). The insertion operator (hc)I=(−1)#{i∈I:i<k}cI∪{k}(hc)_I=(-1)^{\#\{i\in I:i<k\}}c_{I\cup\{k\}} when k∉Ik\notin I, and zero otherwise, satisfies dh+hd=1dh+hd=1. The identity follows by cancelling the two terms deleting and inserting each index different from kk; the term deleting kk has positive sign and equals cIc_I. The condition on N(ν)N(\nu) is preserved on both sides because k∉N(ν)k\notin N(\nu).

If N(ν)N(\nu) is empty, the full simplex complex has just the constant degree-zero class; the same insertion at k=0k=0 contracts the kernel of evaluation at its zeroth vertex. If N(ν)N(\nu) is the entire vertex set, only the degree-nn component remains. All the insertion maps have norm at most one. Applied simultaneously to the coefficient vectors, they preserve convergence to zero and therefore define bounded maps on the completed analytic complex. Their homotopy identities hold first on finite Laurent sums and then by continuity.

The surviving exponent vectors are exactly the two finite sets in the statement. For n=0n=0, the single exponent is mm and the complex has one term. The algebraic computation [StacksAlg] (Tag 01XT, Lemma 30.8.1) uses the same coefficient complex with finite-support families. The natural map is its inclusion in the analytic complex. Since the surviving finite representatives and their projections are identical, it induces the asserted canonical isomorphism. The formulas have coefficients 0,1,−10,1,-1; scalar extension therefore preserves all projections, representatives and comparison maps. ◻

Theorem 2.42.4. NAG-001894 (Cohomology comparison for algebraic coherent sheaves). For every coherent sheaf F\mathcal F on PKn\mathbf P^n_K, analytification induces natural isomorphisms Hq(PKn,F) ≃ Hq((PKn)Gan,Fan)(q≥0).H^q(\mathbf P^n_K,\mathcal F) \ \simeq\ H^q((\mathbf P^n_K)^{\mathrm{an}}_G,\mathcal F^{\mathrm{an}}) \qquad(q\geq0). They respect connecting homomorphisms of short exact sequences.

Proof. Use the algebraic standard affine cover D+(Xi)D_+(X_i) and the analytic unit cover Ui⊂D+(Xi)anU_i\subset D_+(X_i)^{\mathrm{an}}. Restriction and scalar extension on each intersection give a map of their ordered Čech complexes. The algebraic complex computes cohomology by affine acyclicity, and the analytic complex does so by Proposition 2.42.1. On every intersection, sections preserve short exact sequences of coherent sheaves: on the algebraic side this is the module description on an affine; on the analytic side it follows from Theorem 2.33.34 and affinoid Theorem B. The resulting comparison therefore commutes with connecting homomorphisms. These constructions give the usual cohomology map induced by analytification, since in degree zero they are restriction of sections and the acyclic resolutions give their derived maps.

We prove bijectivity by descending induction on qq, simultaneously for all coherent F\mathcal F. Both groups vanish for q>nq>n, by [StacksAlg] (Tag 01YS, Lemma 30.14.1(2)) and Proposition 2.42.1. Choose a surjection L↠F\mathcal L\twoheadrightarrow\mathcal F, where L\mathcal L is a finite direct sum of twists, using [StacksAlg] (Tag 01YS, Lemma 30.14.1(1)), and let R\mathcal R be its coherent kernel. Analytification preserves this short exact sequence. In its two long exact sequences, the comparison maps for Hq(L)H^q(\mathcal L), Hq+1(R)H^{q+1}(\mathcal R), and Hq+1(L)H^{q+1}(\mathcal L) are isomorphisms, by Theorem 2.42.3 and the induction hypothesis.

Here is the required diagram chase. Given an analytic class aa in degree qq for F\mathcal F, lift its connecting class to the algebraic Hq+1(R)H^{q+1}(\mathcal R). Its image in Hq+1(L)H^{q+1}(\mathcal L) is zero by injectivity there, so it comes from an algebraic class bb of F\mathcal F. The difference between aa and the image of bb comes from analytic Hq(L)H^q(\mathcal L), hence from algebraic Hq(L)H^q(\mathcal L). This proves surjectivity for every F\mathcal F. Apply that surjectivity also to R\mathcal R. If an algebraic class of F\mathcal F maps to zero analytically, injectivity in degree q+1q+1 makes it the image of a class of Hq(L)H^q(\mathcal L). Analytically this latter class comes from Hq(R)H^q(\mathcal R). Lift that class by the surjectivity just proved and subtract it; injectivity for L\mathcal L then proves that the original class is zero. This completes the induction. ◻

Corollary 2.42.5. NAG-001895 (Full faithfulness on projective space). For coherent algebraic sheaves F,G\mathcal F,\mathcal G on PKn\mathbf P^n_K, the canonical map Hom⁡(F,G)⟶Hom⁡(Fan,Gan)\operatorname{Hom}(\mathcal F,\mathcal G) \longrightarrow \operatorname{Hom}(\mathcal F^{\mathrm{an}},\mathcal G^{\mathrm{an}}) is bijective.

Proof. The sheaf H ⁣om(F,G)\mathcal H\!om(\mathcal F,\mathcal G) is coherent: locally a finite presentation of F\mathcal F expresses it as the kernel of a map between finite sums of G\mathcal G, over a Noetherian ring. Its analytification is H ⁣om(Fan,Gan)\mathcal H\!om(\mathcal F^{\mathrm{an}},\mathcal G^{\mathrm{an}}) by Theorem 2.33.34. Apply Theorem 2.42.4 in degree zero. Global sections of these Hom sheaves are precisely the stated sheaf morphisms, by the definition of the sheaf Hom. ◻

Proposition 2.42.6. NAG-001896 (Complete base change for projective coherent cohomology). Let P=(PKn)anP=(\mathbf P^n_K)^{\mathrm{an}}, let F\mathcal F be coherent on PGP_G, and let L/KL/K be any complete valued extension. For the coherent pullback FL\mathcal F_L there are natural isomorphisms of finite-dimensional LL-spaces L⊗KHq(PG,F)≃Hq((PL)G,FL).L\otimes_K H^q(P_G,\mathcal F) \simeq H^q((P_L)_G,\mathcal F_L).

Proof. Use the standard finite unit cover and its Banach Čech complex C∙C^\bullet. Put Zq=ker⁡dqZ^q=\ker d^q and Bq=im⁡dq−1B^q=\operatorname{im}d^{q-1}. By Theorem 2.33.38, every BqB^q is closed and Zq/BqZ^q/B^q is finite dimensional. Thus all these spaces are Banach. If KK is nontrivially valued, Lemma 1.14.1 makes the surjections Cq→Bq+1C^q\to B^{q+1} norm-strict. If KK is trivially valued, the coefficient algebras and their finite modules on this cover have the discrete 0/10/1 quotient norms; the same surjections are norm-strict directly. The sequences 0⟶Zq⟶Cq⟶Bq+1⟶0,0⟶Bq⟶Zq⟶Hq⟶0.0\longrightarrow Z^q\longrightarrow C^q \longrightarrow B^{q+1}\longrightarrow0,\qquad 0\longrightarrow B^q\longrightarrow Z^q \longrightarrow H^q\longrightarrow0 . Apply Theorem 2.33.3 to both sequences. The completed tensor complex is the Čech complex on PLP_L: this follows on each intersection from Proposition 2.33.5 and finite-module base change, Proposition 1.18.7. The two exact sequences identify its cohomology with Hq⊗^KLH^q\widehat\otimes_K L. Since HqH^q is finite dimensional, its completed tensor product is its ordinary tensor product: choose a basis and use equivalence of finite-dimensional norms. The intersection identifications respect restrictions, so the resulting isomorphism is natural. ◻

Lemma 2.42.7. NAG-001897 (Coherent modules on a closed analytic subspace). Let i:Z↪Xi:Z\hookrightarrow X be a closed analytic subspace defined by a coherent ideal I\mathcal I. Direct image identifies coherent modules on ZGZ_G with coherent modules on XGX_G annihilated by I\mathcal I, and is exact. If XX has a finite affinoid cover with affinoid finite intersections, then Hq(ZG,E)=Hq(XG,i∗E).H^q(Z_G,\mathcal E)=H^q(X_G,i_*\mathcal E).

Proof. On an affinoid domain V=M(A)V=\mathcal M(A) the inverse image is M(A/J)\mathcal M(A/J), where JJ is the finite ideal representing I∣V\mathcal I|_V. By the affinoid Kiehl equivalence, a coherent module on it is a finite A/JA/J-module MM. Restriction of scalars identifies such modules, with their morphisms and exact sequences, with finite AA-modules killed by JJ. On a smaller affinoid domain with algebra A′A', both restrictions are A′⊗AMA'\otimes_A M, viewed over A′/JA′A'/JA'. Thus these identifications agree on overlaps and glue uniquely. For the cohomology assertion, the inverse images of the given affinoid cover and its intersections are affinoid, and the two Čech coefficient complexes are literally the same. Proposition 2.42.1 computes both derived cohomology groups from that complex. ◻

Theorem 2.42.8. NAG-001898 (Global generation and GAGA on projective space). For every complete non-Archimedean field KK, analytification is an equivalence Coh⁡(PKn) ≃ Coh⁡((PKn)Gan).\operatorname{Coh}(\mathbf P^n_K) \ \simeq\ \operatorname{Coh}((\mathbf P^n_K)^{\mathrm{an}}_G). For every coherent analytic F\mathcal F, the twist F(m)\mathcal F(m) is generated by finitely many global sections for all sufficiently large mm, and Hq(PG,F(m))=0H^q(P_G,\mathcal F(m))=0 for every q>0q>0 and all sufficiently large mm.

Proof. We prove the three assertions together by induction on nn, for all complete fields simultaneously. For n=0n=0 they are the finite-module description of the one-point affinoid M(K)\mathcal M(K) and its Theorem B. Suppose n>0n>0 and that the assertions hold in dimension n−1n-1.

Generation at a rational point. Fix a KK-rational point xx and a hyperplane HH through it, defined by a nonzero linear form hh. Set C=ker⁡(F(−1)→hF)\mathcal C=\ker(\mathcal F(-1)\xrightarrow{h}\mathcal F) and B=F/hF(−1)\mathcal B=\mathcal F/h\mathcal F(-1). Both are coherent and killed by the ideal of HH. Lemma 2.42.7 identifies them with coherent sheaves on HanH^{\mathrm{an}}. Writing Rm=hF(m−1)\mathcal R_m=h\mathcal F(m-1), we have short exact sequences 0⟶C(m)⟶F(m−1)⟶Rm⟶0,0⟶Rm⟶F(m)⟶B(m)⟶0.\begin{split} 0&\longrightarrow\mathcal C(m)\longrightarrow\mathcal F(m-1) \longrightarrow\mathcal R_m\longrightarrow0,\\ 0&\longrightarrow\mathcal R_m\longrightarrow\mathcal F(m) \longrightarrow\mathcal B(m)\longrightarrow0 . \end{split} By induction, H2(C(m))=H1(B(m))=0H^2(\mathcal C(m))=H^1(\mathcal B(m))=0 for large mm. The cohomology groups here can be computed either on HanH^{\mathrm{an}} or after direct image on PP, by the preceding lemma. Their long exact sequences give surjections H1(F(m−1))↠H1(Rm)↠H1(F(m)).H^1(\mathcal F(m-1))\twoheadrightarrow H^1(\mathcal R_m) \twoheadrightarrow H^1(\mathcal F(m)). All these spaces are finite dimensional by Theorem 2.33.38. The nonincreasing sequence dim⁡KH1(F(m))\dim_K H^1(\mathcal F(m)) is eventually constant. For those mm, both displayed surjections are isomorphisms. The second short exact sequence then gives a surjection H0(F(m))↠H0(B(m))H^0(\mathcal F(m))\twoheadrightarrow H^0(\mathcal B(m)). By induction the latter sections generate B(m)\mathcal B(m) for large mm, so their lifts generate the fibre of F(m)\mathcal F(m) at xx.

All analytic points and a uniform twist. For an arbitrary x∈Px\in P, set L=H(x)L=\mathcal H(x). The canonical LL-rational point of PLP_L above xx is defined by the valued residue-field map. Apply the rational-point argument over LL. By Proposition 2.42.6, all global sections upstairs are LL-linear combinations of global sections downstairs. The fibre of the evaluation cokernel upstairs is exactly its downstairs fibre tensored with L=H(x)L=\mathcal H(x). It is zero. Thus a sufficiently large twist is generated at every analytic point xx, with the twist initially depending on xx.

We justify the passage from fibres to open generation loci. For a finite module MM over an affinoid algebra AA, its fibre at a point with prime kernel p\mathfrak p vanishes exactly when Mp=0M_{\mathfrak p}=0: extension from κ(p)\kappa(\mathfrak p) to H(x)\mathcal H(x) is faithful, and the finite-module Nakayama lemma applies to ApA_{\mathfrak p}. This lemma follows directly from the determinant argument: if finitely many generators are linear combinations of themselves with coefficients in the maximal ideal, the determinant of one minus that coefficient matrix is a unit annihilating the module. If Mp=0M_{\mathfrak p}=0, clearing denominators of a finite generating list gives a∉pa\notin\mathfrak p with aM=0aM=0. Thus the zero-fibre locus is open.

The evaluation map uses a finite basis of H0(PG,F(m))H^0(P_G,\mathcal F(m)), so its cokernel is coherent. On each member of the finite closed unit cover its nonzero-fibre locus is a closed subset; the finite union is closed in PP. Therefore the generation locus GmG_m is open in PP. Moreover Gm⊆Gm+1G_m\subseteq G_{m+1}: at any point choose XiX_i nonzero, and multiply generating sections by XiX_i, which is a unit after trivializing the twist there. The GmG_m cover the compact space PP. A finite subcover and this monotonicity give one m0m_0 with Gm=PG_m=P for every m≥m0m\geq m_0. The evaluation cokernel is then zero on each affinoid, since a nonzero finite module has a nonzero localization at some maximal ideal. This proves global generation.

Algebraization and vanishing. Global generation gives a surjection OP(−a)s↠F\mathcal O_P(-a)^s\twoheadrightarrow\mathcal F. Its kernel is coherent by the affinoid finite-module theorem. Apply global generation again to that kernel, obtaining a presentation OP(−b)t→uOP(−a)s⟶F⟶0.\mathcal O_P(-b)^t\xrightarrow{u} \mathcal O_P(-a)^s\longrightarrow\mathcal F \longrightarrow0 . By Corollary 2.42.5, the map uu is the analytification of a unique algebraic map between the corresponding sums of twists. Its algebraic cokernel E\mathcal E is coherent and satisfies Ean≃F\mathcal E^{\mathrm{an}}\simeq\mathcal F by Theorem 2.33.34. This proves essential surjectivity; full faithfulness was already proved in that corollary. Finally, apply [StacksAlg] (Tag 01YS, Lemma 30.14.1(4)) to E\mathcal E, and transport the resulting vanishing by Theorem 2.42.4. This proves the remaining assertion and completes the simultaneous induction. ◻

Corollary 2.42.9. NAG-001899 (GAGA for projective schemes). Let ZZ be a projective scheme over a complete non-Archimedean field KK. Analytification induces an exact equivalence between coherent algebraic sheaves on ZZ and coherent sheaves on ZGanZ^{\mathrm{an}}_G. For every coherent algebraic sheaf E\mathcal E, the canonical maps Hq(Z,E)⟶Hq(ZGan,Ean)H^q(Z,\mathcal E)\longrightarrow H^q(Z^{\mathrm{an}}_G,\mathcal E^{\mathrm{an}}) are isomorphisms in every degree.

Proof. Choose a closed immersion i:Z↪Q=PKni:Z\hookrightarrow Q=\mathbf P^n_K with coherent ideal I\mathcal I, and write P=QanP=Q^{\mathrm{an}}. On each affine chart, modules over the quotient ring are exactly modules over the original ring annihilated by the ideal. This identification, and its analogue for finite affinoid modules, commute with restriction. Thus algebraic closed pushforward and the analytic closed pushforward of Lemma 2.42.7 identify coherent sheaves on ZZ and ZGanZ^{\mathrm{an}}_G with the respective ideal-annihilated subcategories on QQ and PGP_G.

For an analytic coherent sheaf F\mathcal F on ZGanZ^{\mathrm{an}}_G, Theorem 2.42.8 algebraizes i∗anFi^{\mathrm{an}}_*\mathcal F to a coherent G\mathcal G on QQ. The multiplication map I⊗G→G\mathcal I\otimes\mathcal G\to\mathcal G becomes zero after analytification. Analytification commutes with this tensor product, as follows from finite presentations on affine and affinoid charts. Full faithfulness on QQ therefore makes the algebraic multiplication map zero as well. Hence G=i∗E\mathcal G=i_*\mathcal E for a coherent E\mathcal E on ZZ, and Ean≃F\mathcal E^{\mathrm{an}}\simeq\mathcal F. Full faithfulness restricts to these subcategories, and exactness is Theorem 2.33.34.

Finally, closed pushforward does not change the relevant cohomology. Analytically this is Lemma 2.42.7. Algebraically, use the standard finite affine cover of QQ and its pullback to ZZ; both have affine intersections. By [StacksAlg] (Tag 01XB) and [StacksAlg] (Tag 03F7), their ordered Čech complexes compute cohomology. The two complexes are identical for i∗Ei_*\mathcal E and E\mathcal E. The comparison maps commute with this identification because all maps are induced by restriction of the same sections. Theorem 2.42.4 on QQ now gives the asserted comparisons on ZZ. ◻

References. The organization by cohomological comparison, generation by twists, and finite presentations follows [Ser56] (§§13–17). The analytic input is proved here by bounded Laurent contractions, compact restriction operators, and completed-field base change.

2.43 Relative compact operators and affinoid finiteness

A restriction to a smaller relative polydisc has finite-rank approximations over the coefficient algebra. To retain these approximations after passing to a closed module of matching sections, one needs a uniform integral bound. We establish that bound first. Throughout the first five results, KK is nontrivially valued, O=K∘O=K^\circ, and AA is a strict KK-affinoid algebra with a defining quotient norm. Write A0=A≤1A_0=A_{\leq 1}. Proposition 3.14.7 identifies A0A_0 with a quotient of O⟨X1,…,Xn⟩O\langle X_1,\ldots,X_n\rangle. Banach module actions are taken to be contractive. An initially bounded action can be made contractive by the equivalent norm ∥m∥∗=sup⁡a≠0∥am∥/∥a∥\|m\|_* =\sup_{a\ne 0}\|am\|/\|a\|.

Lemma 2.43.1. NAG-001990 (Almost finite generation over the integral coefficient ring). Let EE be a finite A0A_0-module and N⊂EN\subset E any submodule. For every real 0<α<10<\alpha<1 there is a finite A0A_0-submodule N0⊂NN_0\subset N such that aN⊂N0aN\subset N_0 for every a∈Oa\in O with ∣a∣≤α|a|\leq\alpha.

Proof. It suffices to prove the assertion over Pn=O⟨X1,…,Xn⟩P_n=O\langle X_1,\ldots,X_n\rangle: take a finite free presentation above EE and then pass to its image. We use the Gauss norm inherited from K⟨X1,…,Xn⟩K\langle X_1,\ldots,X_n\rangle. The coordinate changes in Lemma 1.11.3 preserve PnP_n. The norm bounds in Theorem 1.10.3 therefore give division within PnP_n by a normalized distinguished element.

First suppose that the valuation is discrete. We prove that PnP_n is Noetherian by induction on nn, the case n=0n=0 following from the uniformizer description of ideals of OO. For a nonzero ideal J⊂PnJ\subset P_n, the norms of its nonzero elements have a maximum β\beta. Choose g∈Jg\in J attaining it. After an integral isometric coordinate change and multiplication by a norm-one unit, write g=cfg=cf with ∣c∣=β|c|=\beta and ff normalized distinguished of degree dd in XnX_n. For h∈Jh\in J, division gives h=qf+rh=qf+r with ∥q∥,∥r∥≤∥h∥≤∣c∣\|q\|,\|r\|\leq\|h\|\leq|c|. Hence q/c∈Pnq/c\in P_n, so r=h−(q/c)g∈Jr=h-(q/c)g\in J. These remainders lie in J∩Pn−1[Xn]<dJ\cap P_{n-1}[X_n]_{<d}, a submodule of a finite free module over the Noetherian ring Pn−1P_{n-1}. They have finitely many generators, which together with gg generate JJ. This proves Noetherianity and the assertion in the discrete case.

Now suppose that the value group is dense in R>0\mathbf R_{>0}. We record an elementary extension argument. In an exact sequence 0→E1→E→E2→00\to E_1\to E\to E_2\to 0, suppose that the assertion is known for the submodules N∩E1N\cap E_1 and the image of NN in E2E_2. Choose b∈Ob\in O with α<∣b∣<1\sqrt\alpha<|b|<1. Finite submodules containing their respective bb-multiples, together with lifts to NN of the chosen quotient generators, generate a finite submodule of NN containing b2Nb^2N. Since ∣a∣≤α<∣b∣2|a|\leq\alpha<|b|^2 implies a/b2∈Oa/b^2\in O, it contains every required aNaN. Thus the assertion passes from these two submodules to NN. This also proves passage from ideals to finite free modules by induction on the rank, and then to finite quotients.

We now induct on nn, proving the assertion for all finite PnP_n-modules. When n=0n=0, for a nonzero ideal J⊂OJ\subset O choose g∈Jg\in J with ∣g∣>αsup⁡h∈J∣h∣|g|>\alpha\sup_{h\in J}|h|; then aJ⊂gOaJ\subset gO for all ∣a∣≤α|a|\leq\alpha. The preceding rank and quotient argument finishes this case. Suppose the assertion is known over Pn−1P_{n-1}, and let J⊂PnJ\subset P_n be a nonzero ideal. Put β=sup⁡h∈J∥h∥\beta=\sup_{h\in J}\|h\| and choose bb as above. Choose g∈Jg\in J with ∥g∥>∣b∣β\|g\|>|b|\beta. Normalization and an integral isometric coordinate change give g=cfg=cf, with ∣c∣=∥g∥|c|=\|g\| and ff normalized distinguished, after absorbing a norm-one unit. The quotient Pn/(f)P_n/(f) is finite free over Pn−1P_{n-1} by integral division.

For h=fq∈J∩(f)h=fq\in J\cap(f), multiplicativity of the Gauss norm gives ∥q∥=∥h∥≤β\|q\|=\|h\|\leq\beta. Therefore bh=(bq/c)g∈(g)bh=(bq/c)g\in(g), since ∣b∣β<∣c∣|b|\beta<|c|. Thus b(J∩(f))b(J\cap(f)) lies in the single-generated submodule (g)⊂J∩(f)(g)\subset J\cap(f). By induction, the bb-multiple of the image of JJ in Pn/(f)P_n/(f) lies in a finite Pn−1P_{n-1}-submodule of that image. Lift its generators to JJ and allow PnP_n-coefficients. The extension argument gives a finite submodule of JJ containing b2Jb^2J, hence every aJaJ with ∣a∣≤α|a|\leq\alpha. Rank induction and passage to quotients complete the proof. ◻

Definition 2.43.2. NAG-001991 (Relative complete continuity). A bounded AA-linear map g:E→Fg:E\to F between Banach modules is completely continuous if it is an operator-norm limit of bounded AA-linear maps hih_i with finite AA-module image. It is strictly completely continuous if these approximants can be chosen with a fixed c∈O∖{0}c\in O\setminus\{0\} such that chi(E≤1)c h_i(E_{\leq 1}) is contained in a finite A0A_0-submodule of F≤1F_{\leq 1} for every ii. The finite integral submodule may depend on ii.

Theorem 2.43.3. NAG-001992 (The relative Schwarz perturbation lemma). Let f,g:E→Ff,g:E\to F be bounded AA-linear maps of Banach modules, with ff surjective and gg completely continuous. Then (f+g)(E)(f+g)(E) is closed in FF, and the quotient is a finite AA-module whose quotient norm is canonical.

Proof. Lemma 1.14.1 gives a lift bound CC for ff. If a bounded map uu satisfies C∥u∥<1C\|u\|<1, successive lifts of the errors show that f+uf+u remains surjective with a uniform lift bound: starting with y0=yy_0=y, choose f(xj)=yjf(x_j)=y_j, ∥xj∥≤C∥yj∥\|x_j\|\leq C\|y_j\|, and put yj+1=yj−(f+u)(xj)y_{j+1}=y_j-(f+u)(x_j). The norms decrease geometrically and ∑jxj\sum_jx_j is a lift of yy. Choose a finite-image approximant hh of gg with C∥g−h∥<1C\|g-h\|<1, and set f′=f+(g−h)f'=f+(g-h). Thus f′f' is a norm-strict surjection.

Let N=ker⁡hN=\ker h. It is closed, and Q=E/NQ=E/N is a complete finite AA-module. Its norm is canonical: an algebraic finite generating set defines a bounded surjection Am→QA^m\to Q, which is norm-strict by the same open-mapping lemma. The image of ker⁡f′\ker f' in QQ is an algebraic submodule of a finite module, hence closed by Theorem 1.11.5. Consequently N+ker⁡f′N+\ker f' is closed in EE. Since f′f' is a quotient map, f′(N)f'(N) is closed in FF, and F/f′(N)F/f'(N) is a finite module with its canonical norm. The image of f′+hf'+h contains f′(N)f'(N); its image modulo that closed submodule is a submodule of the finite module F/f′(N)F/f'(N), hence closed and finite. Its inverse image is therefore closed in FF, and its cokernel is finite with its canonical quotient norm. Since f′+h=f+gf'+h=f+g, this proves the assertion. ◻

Lemma 2.43.4. NAG-001993 (Strict complete continuity into a closed target). Let E=c0(I,A)E=c_0(I,A) with its maximum norm and standard basis (eλ)λ∈I(e_\lambda)_{\lambda\in I}. Let j:N→Fj:N\to F identify a Banach module with a closed submodule, with equivalent induced norm. If g:E→Ng:E\to N is bounded and jgjg is strictly completely continuous, then gg is strictly completely continuous.

Proof. Use the induced norm on NN. Multiplication of the maps by one fixed nonzero scalar, followed by division by that scalar at the end, reduces to the case that gg is contractive and approximants h:E→Fh:E\to F satisfy h(E≤1)⊂L⊂F≤1h(E_{\leq 1})\subset L\subset F_{\leq 1} for a finite A0A_0-module LL. This rescaling also shows directly that equivalent norms and bounded pre- or postcomposition preserve strict complete continuity: multiply the fixed scalar by a sufficiently small scalar to absorb each uniform operator bound.

Fix a∈Oa\in O with 0<∣a∣<10<|a|<1. Given ε>0\varepsilon>0, choose such an hh with ∥jg−h∥≤ε\|jg-h\|\leq\varepsilon. Apply Lemma 2.43.1 to h(E≤1)⊂Lh(E_{\leq 1})\subset L. It gives generators y1,…,ys∈h(E≤1)y_1,\ldots,y_s\in h(E_{\leq 1}) of a finite submodule containing ah(E≤1)a h(E_{\leq 1}). Choose xj∈E≤1x_j\in E_{\leq 1} with h(xj)=yjh(x_j)=y_j and put zj=g(xj)∈N≤1z_j=g(x_j)\in N_{\leq 1}. Then ∥jzj−yj∥≤ε\|jz_j-y_j\|\leq\varepsilon. For each λ\lambda, choose coefficients bjλ∈Ab_{j\lambda}\in A satisfying h(eλ)=∑jbjλyj,∥bjλ∥≤∣a∣−1.h(e_\lambda)=\sum_j b_{j\lambda}y_j, \qquad \|b_{j\lambda}\|\leq |a|^{-1}. Define g0(eλ)=∑jbjλzjg_0(e_\lambda)=\sum_jb_{j\lambda}z_j. The uniform coefficient bound extends this rule to a bounded AA-linear map on c0(I,A)c_0(I,A), with image in the finite module generated by the zjz_j. The basis estimate gives ∥g−g0∥≤∣a∣−1ε\|g-g_0\|\leq |a|^{-1}\varepsilon. If x=∑λdλeλx=\sum_\lambda d_\lambda e_\lambda has norm at most one, the series ∑λadλbjλ\sum_\lambda a d_\lambda b_{j\lambda} converge in A0A_0, so ag0(E≤1)⊂∑jA0zj⊂N≤1.a g_0(E_{\leq1})\subset\sum_j A_0z_j\subset N_{\leq1}. The same fixed aa works for every ε\varepsilon, proving strict complete continuity. ◻

Theorem 2.43.5. NAG-001994 (Relative Schwarz with a presentation and a closed target). Let f,g:E→Ff,g:E\to F be bounded AA-linear maps, with ff surjective. Suppose there are a bounded surjection p:E♭→Ep:E^\flat\to E and a closed embedding j:F→F♯j:F\to F^\sharp of Banach modules such that jgpjgp is strictly completely continuous. Then (f+g)(E)(f+g)(E) is closed and its cokernel is a finite AA-module with its canonical quotient norm.

Proof. Every contractive Banach AA-module MM admits a bounded norm-strict surjection q:c0(I,A)→Mq:c_0(I,A)\to M. Take I=M≤1I=M_{\leq 1} and send its basis element indexed by mm to mm. The map is contractive. For fixed π∈K\pi\in K with 0<∣π∣<10<|\pi|<1, any nonzero x∈Mx\in M can be divided by a power of π\pi to have norm in (∣π∣,1](|\pi|,1]; the corresponding single basis vector gives a lift of xx of norm at most ∣π∣−1∥x∥|\pi|^{-1}\|x\|.

Apply this to M=E♭M=E^\flat. The composite jgpqjgpq is strictly completely continuous by bounded precomposition. Lemma 2.43.4 makes gpqgpq completely continuous into FF. Theorem 2.43.3, applied to the surjection fpqfpq and this map, proves the conclusion, because pqpq is surjective and hence does not change the image of f+gf+g. ◻

Proposition 2.43.6. NAG-001995 (Strictly compact relative restriction). Let BB be a generalized KK-affinoid algebra and let u:A⟨T1,…,Tn⟩→Bu:A\langle T_1,\ldots,T_n\rangle\to B be bounded. Suppose that the spectral radius of each ti=u(Ti)t_i=u(T_i) is less than one. Then uu, regarded as an AA-linear map, is strictly completely continuous. The same holds for finite direct sums and for the restriction to any closed source submodule with its induced norm.

Proof. Choose 0<θ<10<\theta<1 larger than all spectral radii. The spectral-radius formula supplies a constant CC such that ∥tν∥≤Cθ∣ν∣\|t^\nu\|\leq C\theta^{|\nu|} for every multi-index. Replace the target norm by an equivalent norm making its AA-action contractive. The degree-dd truncation udu_d has image in the finite AA-module generated by tνt^\nu, ∣ν∣<d|\nu|<d, and ∥u−ud∥≤C′θd\|u-u_d\|\leq C'\theta^d for one fixed C′C'. Choose c∈O∖{0}c\in O\setminus\{0\} so small that ∥ctν∥≤1\|ct^\nu\|\leq 1 for every ν\nu. Then cud(A⟨T⟩≤1)⊂∑∣ν∣<dA0 ctν⊂B≤1.c u_d\bigl(A\langle T\rangle_{\leq1}\bigr) \subset \sum_{|\nu|<d}A_0\,ct^\nu\subset B_{\leq1}. This proves the required uniform integral bound. For finitely many maps take a common small cc; restriction of the approximants to a source submodule preserves the displayed bound and the operator convergence. ◻

Theorem 2.43.7. NAG-001996 (Finiteness from finitely many inner relative presentations). Assume that KK is nontrivially valued and AA is strict. Let A→BA\to B be bounded, with BB generalized affinoid. Suppose that M(B)\mathcal M(B) has a finite affinoid-domain covering (Vi)(V_i), and for each ii there is a norm-admissible surjection Pi=A⟨Ti⟩↠BP_i=A\langle T_i\rangle\twoheadrightarrow B whose variables have spectral radius less than one on ViV_i. Then BB is finite over AA, and its given norm is equivalent to its canonical finite-AA-module norm.

Proof. Write Bi=O(Vi)B_i=\mathcal O(V_i), and let Z⊂∏iBiZ\subset\prod_iB_i be the kernel of the first Čech differential. Theorem 2.38.1 identifies BB boundedly with ZZ, with bounded inverse. For comparison, take the outer covering consisting of one copy of M(B)\mathcal M(B) for each ii; its matching module is the diagonal Z0⊂BsZ_0\subset B^s, also boundedly isomorphic to BB. Restriction therefore induces a bounded isomorphism r:Z0→Zr:Z_0\to Z.

The product presentation q:∏iPi↠Bsq:\prod_iP_i\twoheadrightarrow B^s is norm-strict. Its inverse image E♭=q−1(Z0)E^\flat=q^{-1}(Z_0) is closed, and p=q∣E♭:E♭↠Z0p=q|_{E^\flat}:E^\flat\twoheadrightarrow Z_0 is norm-strict: any controlled lift of an element of Z0Z_0 automatically belongs to this inverse image. With j:Z↪∏iBij:Z\hookrightarrow\prod_iB_i, the map jrpjrp is the restriction of the product of the maps Pi→BiP_i\to B_i. Proposition 2.43.6 makes it strictly completely continuous. Apply Theorem 2.43.5 with f=rf=r and g=−rg=-r. The zero map then has finite cokernel ZZ, with its canonical norm. Transporting this norm through B≃ZB\simeq Z proves the theorem. ◻

Theorem 2.43.8. NAG-001997 (Boundaryless affinoid morphisms are finite). Let KK be any complete non-Archimedean field and A→BA\to B a bounded homomorphism of generalized KK-affinoid algebras. The following conditions are equivalent:

  1. Int⁡(M(B)/M(A))=M(B)\operatorname{Int}(\mathcal M(B)/\mathcal M(A))=\mathcal M(B).

  2. BB is a finite AA-module and its given norm is equivalent to its canonical finite-module norm.

Proof. The zero algebra is immediate. Suppose first that every point is interior. For y∈M(B)y\in\mathcal M(B) choose a witnessing norm-admissible relative presentation A{ry−1Ty}↠BA\{r_y^{-1}T_y\}\twoheadrightarrow B. Choose radii syjs_{yj} with ∣Tyj(y)∣<syj<ryj|T_{yj}(y)|<s_{yj}<r_{yj}. The Weierstrass domain Vy={z∈M(B):∣Tyj(z)∣≤syj for every j}V_y=\{z\in\mathcal M(B): |T_{yj}(z)|\leq s_{yj}\text{ for every }j\} contains an ordinary neighborhood of yy. Compactness gives a finite family of these domains covering M(B)\mathcal M(B). Only finitely many presentation radii and witness radii now occur. Apply Theorem 1.17.6 with all these radii, adjoining one additional nonunit radius if KK is trivially valued. We obtain a finite free-radius extension F=KuF=K_u, nontrivially valued, over which AFA_F is strict and all witness radii ryjr_{yj} belong to ∣F×∣\sqrt{|F^\times|}.

For each such radius choose an integer m>0m>0 and c∈F×c\in F^\times with ryjm=∣c∣r_{yj}^m=|c|. Adjoin a root of Xm−cX^m-c, successively for the finitely many radii. Lemma 1.12.1 makes the resulting finite field extension L/FL/F complete, and multiplicativity gives an element of absolute value ryjr_{yj} in LL. After this base change each witness presentation can be rescaled to unit radii. The inverse-image domains still cover by Theorem 2.36.9; on each such domain the rescaled variables have spectral radius at most syj/ryj<1s_{yj}/r_{yj}<1. Relative presentations commute with this base change by Proposition 2.33.8. Thus Theorem 2.43.7 shows that BLB_L is finite over ALA_L.

We descend this finiteness in two steps. Choose an FF-basis of LL containing 11. Equivalence of finite-dimensional norms shows directly that, for every Banach FF-space MM, the completed tensor M⊗^FLM\widehat\otimes_F L is a finite direct sum of copies of MM, with equivalent coefficient norms. Consequently ALA_L is finite over AFA_F, and the coefficient projection L→FL\to F taking 11 to 11 induces a bounded AFA_F-linear retraction BL→BFB_L\to B_F. Hence BFB_F is a direct summand of the finite AFA_F-module BLB_L and is finite. Since FF is nontrivially valued, its given Banach norm is canonical by the open-mapping argument. Theorem 1.18.9, applied to the original Banach AA-module BB, now gives finite generation and the canonical norm over the original field KK. A bounded action can first be made contractive by the equivalent norm specified at the start of this section; the conclusion is unchanged.

Conversely, if BB is finite with its canonical norm, Theorem 2.33.18 supplies one relative affinoid presentation whose generator radii strictly exceed their norms. It witnesses interiority at every point. ◻

References. The relative operator argument reconstructs the method of Schwarz and Kiehl in [Bosch14] (§6.4, Theorems 3–4 and Lemmas 5–6). The integral approximation, the closed-target passage, and the free-radius descent are proved above.

2.44 Composition and topological interior

Theorem 2.44.1. NAG-001626 (The affinoid composition formula for interior). For composable affinoid morphisms Z→gY→fXZ\xrightarrow{g}Y\xrightarrow{f}X over any complete non-Archimedean field, Int⁡(Z/X)=Int⁡(Z/Y)∩g−1Int⁡(Y/X).\operatorname{Int}(Z/X)= \operatorname{Int}(Z/Y)\cap g^{-1}\operatorname{Int}(Y/X).

Proof. The inclusion from right to left is the composition implication of Proposition 2.33.21. Write the algebras as A→B→CA\to B\to C and fix z∈Zz\in Z, with images yy and xx. The completed residue-field embeddings identify the images of their graded reductions with graded subrings RA⊂RB⊂RC⊂H(z)~gr.R_A\subset R_B\subset R_C\subset\widetilde{\mathcal H(z)}^{\rm gr}. If z∈Int⁡(Z/X)z\in\operatorname{Int}(Z/X), the relative point criterion says that RCR_C is finite over RAR_A. It is then finite over RBR_B, using the same generating set. Each element of RBR_B is integral over RAR_A: multiplication by this element on the finite RAR_A-module RCR_C, which contains 11, gives a monic relation by the determinant argument. Furthermore RBR_B is finitely generated as an RAR_A-algebra, because the fixed absolute presentation result makes B~gr\widetilde B^{\rm gr} finitely generated over K~gr\widetilde K^{\rm gr}, whose image is contained in RAR_A. Adjoining these finitely many integral generators successively shows that RBR_B is finite over RAR_A: the bounded powers below the respective monic degrees form a finite spanning family. Apply the reverse implication of the relative point criterion to A→BA\to B at yy and to B→CB\to C at zz. It gives y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X) and z∈Int⁡(Z/Y)z\in\operatorname{Int}(Z/Y), as required. ◻

Lemma 2.44.2. NAG-001627 (The image of a finite flat affinoid map). Let A→BA\to B be finite, norm-admissible and flat between generalized affinoid algebras. Its image on Berkovich spectra is open and closed.

Proof. The algebra AA is Noetherian, so the finite module BB is finitely presented. By [StacksAlg] (Tag 00NX, Lemma 10.78.2(1),(8)), its rank function on Spec⁡A\operatorname{Spec}A is locally constant. Hence its positive-rank locus is open and closed. For x∈M(A)x\in\mathcal M(A) with kernel p\mathfrak p, the analytic fibre has algebra B⊗AH(x)B\otimes_A\mathcal H(x), with its complete finite-dimensional norm, by Proposition 2.33.19 and Theorem 2.33.11. This algebra is nonzero exactly when B⊗Aκ(p)B\otimes_A\kappa(\mathfrak p) has positive dimension, since κ(p)→H(x)\kappa(\mathfrak p)\to\mathcal H(x) is a field embedding. A nonzero finite-dimensional algebra has a bounded character after a finite residue-field extension, so its spectrum is nonempty. Therefore the image is the inverse image of the positive-rank locus. Inverse images of D(a)D(a) are {∣a∣>0}\{|a|>0\}, and inverse images of Zariski closed subsets are common zero loci of continuous evaluations. The inverse image in question is consequently open and closed. ◻

Theorem 2.44.3. NAG-001628 (Interior of an affinoid domain). If UU is an affinoid domain of an affinoid space XX, then Int⁡(U/X)\operatorname{Int}(U/X) is the ordinary topological interior of UU in XX.

Proof. Suppose xx belongs to the ordinary interior. Lemma 2.41.1 gives a rational domain W⊂UW\subset U containing a neighborhood of xx and satisfying x∈Int⁡(W/X)x\in\operatorname{Int}(W/X). By Lemma 2.36.5, WW is also an affinoid domain of UU. The composition formula for W→U→XW\to U\to X then gives x∈Int⁡(U/X)x\in\operatorname{Int}(U/X).

Conversely, let x∈Int⁡(U/X)x\in\operatorname{Int}(U/X). The relative boundary D=U∖Int⁡(U/X)D=U\setminus\operatorname{Int}(U/X) is compact, by openness of affinoid relative interior. The inclusion U↪XU\hookrightarrow X is a topological embedding, so its compact image DD is closed in the Hausdorff space XX and does not contain xx. Choose a rational domain VV that is a neighborhood of xx and is contained in X∖DX\setminus D. The base change UV=U∩VU_V=U\cap V is an affinoid domain of VV. Every point of UVU_V was interior in U/XU/X, so the forward base-change inclusion in Proposition 2.33.21 shows that UV→VU_V\to V is boundaryless. By Theorem 2.43.8, its algebra is finite and norm-admissible over that of VV. It is flat by Proposition 2.38.2. Lemma 2.44.2 now says that U∩VU\cap V is open in VV. As VV is a neighborhood of xx in XX, this provides an ordinary neighborhood of xx contained in UU. ◻

Definition 2.44.4. (Relative interior on analytic atlases). For a morphism f:Y→Xf:Y\to X of general analytic spaces, a point yy is interior if, for every affinoid domain U⊂XU\subset X containing f(y)f(y), there is an affinoid domain V⊂f−1(U)V\subset f^{-1}(U) which is a neighborhood of yy in f−1(U)f^{-1}(U) and satisfies y∈Int⁡(V/U)y\in\operatorname{Int}(V/U).

Proposition 2.44.5. NAG-001630 (Affinoid compatibility and arbitrary analytic domains). Definition 2.44.4 agrees with the original affinoid definition. For any analytic domain V⊂XV\subset X, with no goodness or global Hausdorff hypothesis, it gives Int⁡(V/X)=Int⁡top(V,X).\operatorname{Int}(V/X)=\operatorname{Int}_{\rm top}(V,X).

Proof. For an affinoid map, the forward base-change inclusion proves that an affinoid interior point satisfies the new definition, taking VV to be the entire affinoid inverse image of each UU. Conversely, take U=XU=X in the definition. A witnessing affinoid neighborhood V⊂YV\subset Y has y∈Int⁡(V/X)y\in\operatorname{Int}(V/X). Apply Proposition 2.44.6 first with the affinoid pair (V,X)(V,X) and then with the whole pair (Y,X)(Y,X). Both tests compute the same general relative interior at yy, so y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X) in the original affinoid sense.

For a domain inclusion, an ordinary neighborhood of yy contained in VV meets each affinoid U∋yU\ni y in an open neighborhood in UU. Choose a strict-inequality rational neighborhood inside that intersection, using Lemma 2.41.1. It is the required witness. Conversely, if the definition holds, its witness WW over any affinoid U∋yU\ni y is an affinoid domain of UU. Theorem 2.44.3 makes WW, and hence V∩UV\cap U, a neighborhood of yy in UU.

To pass from these neighborhoods to XX, use its analytic net. Finitely many affinoid domains U1,…,UsU_1,\ldots,U_s containing yy have union containing a neighborhood of yy. For each ii choose an open subset Oi⊂XO_i\subset X with y∈Oiy\in O_i and Oi∩Ui⊂VO_i\cap U_i\subset V. Intersect the finitely many OiO_i with an open neighborhood contained in ⋃iUi\bigcup_iU_i. The resulting open neighborhood lies in VV. This proves the assertion for arbitrary analytic domains and also shows that the construction is unchanged by saturated atlas refinements. ◻

Proposition 2.44.6. NAG-001631 (Ordinary local testing for good spaces). Let f:Y→Xf:Y\to X be a morphism of good analytic spaces. Let X0X_0 be an affinoid neighborhood of x=f(y)x=f(y) and Y0⊂f−1(X0)Y_0\subset f^{-1}(X_0) an affinoid neighborhood of yy in YY. Then y∈Int⁡(Y/X)⟺y∈Int⁡(Y0/X0).y\in\operatorname{Int}(Y/X) \quad\Longleftrightarrow\quad y\in\operatorname{Int}(Y_0/X_0). In particular, relative interior is open for morphisms of good spaces.

Proof. If yy satisfies the general definition, take a witnessing affinoid neighborhood V⊂f−1(X0)V\subset f^{-1}(X_0) with y∈Int⁡(V/X0)y\in\operatorname{Int}(V/X_0). Choose an affinoid neighborhood WW of yy inside V∩Y0V\cap Y_0, for example a rational neighborhood in Y0Y_0 contained in this ordinary neighborhood. It is an affinoid domain of both VV and Y0Y_0. Since WW is a neighborhood in VV, the domain-interior theorem and the composition formula give y∈Int⁡(W/X0)y\in\operatorname{Int}(W/X_0), and the same formula for W→Y0→X0W\to Y_0\to X_0 gives the desired condition.

For the converse, fix any affinoid domain U⊂XU\subset X containing xx. Choose a rational neighborhood U′⊂UU'\subset U of xx contained in the open set U∩Int⁡top(X0,X)U\cap\operatorname{Int}_{\rm top}(X_0,X). Then U′U' is also an affinoid domain of X0X_0. The inverse image V′=Y0×X0U′V'=Y_0\times_{X_0}U' is affinoid and is a neighborhood of yy in f−1(U)f^{-1}(U). The base-change inclusion gives y∈Int⁡(V′/U′)y\in\operatorname{Int}(V'/U'); the domain-interior theorem gives x∈Int⁡(U′/U)x\in\operatorname{Int}(U'/U). Composition proves y∈Int⁡(V′/U)y\in\operatorname{Int}(V'/U), which is the required witness. Finally, on the ordinary interiors of Y0Y_0 and X0X_0, the displayed criterion identifies the relative interior with an open affinoid-interior locus. These neighborhoods cover YY, proving openness. ◻

Theorem 2.44.7. NAG-001632 (Base-domain locality for affinoid morphisms). Let f:Y→Xf:Y\to X be an affinoid morphism, let (Xi)(X_i) be a finite affinoid-domain covering of XX, and put Yi=f−1(Xi)Y_i=f^{-1}(X_i). Then ∂(Y/X)=⋃i∂(Yi/Xi).\partial(Y/X)=\bigcup_i\partial(Y_i/X_i).

Proof. The forward base-change inclusion for interior proves that the union on the right is contained in the left side. Conversely, fix y∈Yy\in Y with image xx, and suppose y∈Int⁡(Yi/Xi)y\in\operatorname{Int}(Y_i/X_i) for every ii with x∈Xix\in X_i. Write F=H(x)~grF=\widetilde{\mathcal H(x)}^{\rm gr}, G=H(y)~grG=\widetilde{\mathcal H(y)}^{\rm gr}, and denote by RX,RY,RXi,RYiR_X,R_Y,R_{X_i},R_{Y_i} the images of the corresponding graded affinoid reductions in these fields. We regard FF as a graded subfield of GG.

Each YiY_i is the actual inverse-image domain, and restriction to XiX_i and YiY_i preserves the relevant completed residue fields. Take any graded valuation ring V⊂GV\subset G containing RXR_X. Its intersection with FF is a graded valuation ring containing RXR_X. Proposition 3.19.6 applied to the cover of the germ (X,x)(X,x) supplies an index ii such that V∩FV\cap F contains RXiR_{X_i}. By the assumed interiority and the relative point criterion, RYiR_{Y_i} is finite over RXiR_{X_i}. A valuation ring containing the latter contains the former, by integral closedness on homogeneous elements. In particular VV contains RY⊂RYiR_Y\subset R_{Y_i}. Thus every graded valuation of GG containing RXR_X also contains RYR_Y. Lemma 3.19.1 and finite generation of RYR_Y from the fixed absolute presentation imply that RYR_Y is finite over RXR_X. The reverse implication of the relative point criterion now gives y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X), proving the reverse inclusion of the boundary sets. ◻

Proposition 2.44.8. NAG-001633 (Jacobian presentations of algebraically smooth morphisms). Let f:Y→Xf:Y\to X be a smooth morphism of finite-type KK-schemes. Its analytification admits a Jacobian presentation at every point and is boundaryless. The analytic differential rank is the algebraic relative differential rank at the corresponding scheme point.

Proof. By [StacksAlg] (Tag 01V7, Lemma 29.35.11), near each scheme point the morphism has a standard smooth presentation S=R[T1,…,Tn]/(f1,…,fc),det⁡(∂fi∂Tj)1≤i,j≤c∈S×.S=R[T_1,\ldots,T_n]/(f_1,\ldots,f_c), \qquad \det\left(\frac{\partial f_i}{\partial T_j}\right)_{1\le i,j\le c} \in S^\times. The determinant condition is the definition in [StacksAlg] (Tag 00T6). Choose an affinoid neighborhood of the image point in the analytification of Spec⁡R\operatorname{Spec}R, and choose finite radii for the coordinates strictly larger than their values at the selected analytic point. The relative analytification construction identifies the resulting affinoid source piece with the quotient of the corresponding relative polydisc by these same cc equations. The determinant remains invertible in its quotient algebra. Theorem 2.41.7 therefore gives a Jacobian presentation, with differential rank n−cn-c. The equality with algebraic rank follows directly from Proposition 2.41.10, or from the same invertible matrix. Proposition 2.33.27 supplies boundarylessness, with its all-target-domain interpretation. ◻

References. The composition and domain-interior assertions are classical; compare [TemIntro] (3.4.2.3 and 4.2.4.1–4.2.4.3). Their proofs here use the internal graded innerness criterion and relative compact-operator finiteness.

2.45 Gluing across a Laurent circle

The affinoid-neighborhood criterion for analytic germs rests on a two-chart calculation. The estimates below use arbitrary positive radii and the actual quotient norm on the overlap.

Lemma 2.45.1. NAG-001873 (Small Laurent coordinate changes). Let r1,…,rn,p,q>0r_1,\ldots,r_n,p,q>0, with p≤qp\le q, and set E=K{r−1T,q−1S,pS−1}.E=K\{r^{-1}T,q^{-1}S,pS^{-1}\}. Give EE its Laurent Gauss norm. Suppose Hi,H∈EH_i,H\in E satisfy, for a real number 0<ε<10<\varepsilon<1, ∥Hi−Ti∥≤εri,H=S(1+u),∥u∥≤ε.\|H_i-T_i\|\le\varepsilon r_i, \qquad H=S(1+u),\quad \|u\|\le\varepsilon. Then Ti↦HiT_i\mapsto H_i, S↦HS\mapsto H defines an isometric KK-algebra automorphism of EE.

Proof. The coefficient normal form in Theorem 2.13.2, with coefficients in K{r−1T}K\{r^{-1}T\}, identifies EE isometrically with its weighted convergent Laurent series; the weights are those displayed below. The geometric series gives (1+u)−1(1+u)^{-1} and ∥(1+u)−1−1∥≤ε\|(1+u)^{-1}-1\|\le\varepsilon. Consequently HH and H−1H^{-1} satisfy the defining radius bounds, so the substitution defines a bounded algebra endomorphism Φ\Phi. The Laurent monomials TνSjT^\nu S^j, ν∈Nn\nu\in\mathbb N^n, j∈Zj\in\mathbb Z, have weights wν,j=rν{qjj≥0,pjj<0.w_{\nu,j}=r^\nu \begin{cases}q^j&j\ge0,\\p^j&j<0.\end{cases} Expanding one factor at a time and using ∣m∣≤1|m|\le 1 for integers mm shows ∥Φ(TνSj)−TνSj∥≤εwν,j.\|\Phi(T^\nu S^j)-T^\nu S^j\| \le\varepsilon w_{\nu,j}. For negative jj use the already established estimate for (1+u)−1−1(1+u)^{-1}-1. Passing from finite sums to convergent Laurent series yields ∥Φ−1∥≤ε\|\Phi-1\|\le\varepsilon. The operator series ∑m≥0(1−Φ)m\sum_{m\ge 0}(1-\Phi)^m is therefore a continuous inverse to Φ\Phi. A bijective algebra homomorphism has an algebra-homomorphic inverse. Finally, ∥Φ(a)−a∥<∥a∥\|\Phi(a)-a\|<\|a\| for a≠0a\ne 0 proves ∥Φ(a)∥=∥a∥\|\Phi(a)\|=\|a\|. ◻

Theorem 2.45.2. NAG-001874 (Two Laurent charts form an affinoid space). Let X=Y∪ZX=Y\cup Z be an analytic space covered by two affinoid domains Y=M(B)Y=\mathcal M(B) and Z=M(C)Z=\mathcal M(C), with affinoid intersection Y∩Z=M(D)Y\cap Z=\mathcal M(D). Suppose 0<p<t<q0<p<t<q, ri>0r_i>0, and bounded norm-strict epimorphisms K{r−1T,t−1S,pS−1}⟶B,Ti⟼fi,S⟼f,K{r−1T,q−1S,tS−1}⟶C,Ti⟼gi,S⟼g.\begin{aligned} K\{r^{-1}T,t^{-1}S,pS^{-1}\}&\longrightarrow B, &T_i&\longmapsto f_i,&S&\longmapsto f,\\ K\{r^{-1}T,q^{-1}S,tS^{-1}\}&\longrightarrow C, &T_i&\longmapsto g_i,&S&\longmapsto g. \end{aligned} Assume Y∩Z=Y{∣f∣≥t}=Z{∣g∣≤t}Y\cap Z=Y\{|f|\ge t\}=Z\{|g|\le t\}. Give DD the quotient norm induced by K{r−1T,t−1S,tS−1}⟶D,Ti⟼fi, S⟼f.K\{r^{-1}T,t^{-1}S,tS^{-1}\}\longrightarrow D, \qquad T_i\longmapsto f_i,\ S\longmapsto f. If ∥fi−gi∥D<ri\|f_i-g_i\|_D<r_i and ∥f−g∥D<t\|f-g\|_D<t, then XX is affinoid. More precisely, for every δ>0\delta>0 there are an affinoid algebra AA, an isomorphism X≃M(A)X\simeq\mathcal M(A), and a bounded norm-strict epimorphism K{r−1T,q−1S,pS−1}⟶A,Ti⟼hi, S⟼h,K\{r^{-1}T,q^{-1}S,pS^{-1}\}\longrightarrow A, \qquad T_i\longmapsto h_i,\ S\longmapsto h, such that, for the quotient norms of the displayed presentations of BB and CC, ∥hi−fi∥B, ∥hi−gi∥C<∥fi−gi∥D+δ,∥h−f∥B, ∥h−g∥C<∥f−g∥D+δ.\begin{split} \|h_i-f_i\|_B,\ \|h_i-g_i\|_C &<\|f_i-g_i\|_D+\delta,\\ \|h-f\|_B,\ \|h-g\|_C &<\|f-g\|_D+\delta. \end{split}

Proof. First suppose the two epimorphisms are isomorphisms. The restrictions identify BB and CC with subalgebras of DD. In the ff-coordinates the latter is the Laurent Gauss algebra of the circle ∣S∣=t|S|=t. The hypotheses and Lemma 2.45.1, applied with equal inner and outer radius tt, identify its gg-coordinates isometrically as well. Choose ε<1\varepsilon<1 exceeding all ratios ∥fi−gi∥D/ri\|f_i-g_i\|_D/r_i and ∥f−g∥D/t\|f-g\|_D/t.

Write a Laurent series a=∑aν,jfνfja=\sum a_{\nu,j}f^\nu f^j on the circle, where fν=f1ν1⋯fnνnf^\nu=f_1^{\nu_1}\cdots f_n^{\nu_n}. Retain its terms with j>0j>0 in the ff-coordinates, and write the terms with j≤0j\le 0 in the gg-coordinates. The first sum b0b_0 converges in BB, the second sum c0c_0 converges in CC, and ∥b0∥B,∥c0∥C≤∥a∥D,∥a−b0−c0∥D≤ε∥a∥D.\|b_0\|_B,\|c_0\|_C\le\|a\|_D, \qquad \|a-b_0-c_0\|_D\le\varepsilon\|a\|_D. Iterating on the error gives a=b+ca=b+c, where bb is a series in strictly positive powers of ff and cc is a series in nonpositive powers of gg, with ∥b∥B,∥c∥C≤∥a∥D,∥b/f∥B,∥c/g∥C≤t−1∥a∥D.\|b\|_B,\|c\|_C\le\|a\|_D, \qquad \|b/f\|_B,\|c/g\|_C\le t^{-1}\|a\|_D. The divided estimates hold term by term: in BB the largest radius of ff is tt, whereas in CC the smallest radius of gg is tt.

Apply this splitting to fi−gi=bi+cif_i-g_i=b_i+c_i and f−g=b+cf-g=b+c. The matching pairs hi=fi−bi=gi+ci,h=f−b=g+ch_i=f_i-b_i=g_i+c_i,\qquad h=f-b=g+c define sections on XX. The inequalities ∥b/f∥B<1\|b/f\|_B<1 and ∥c/g∥C<1\|c/g\|_C<1 show that hh is invertible and has the same modulus as ff on YY and as gg on ZZ. On each chart, the change to the hi,hh_i,h coordinates satisfies Lemma 2.45.1; hence it is an isomorphism of the corresponding Laurent Gauss algebra. Therefore YY and ZZ, with these coordinates, are the standard two Laurent domains ∣S∣≤t|S|\le t and ∣S∣≥t|S|\ge t in M(K{r−1T,q−1S,pS−1}).\mathcal M\bigl(K\{r^{-1}T,q^{-1}S,pS^{-1}\}\bigr). Their identifications agree on the overlap. Theorem 2.40.10 identifies their union with this affinoid space. In this free case the coordinate errors are bounded above by the corresponding overlap errors, with no added δ\delta.

For the general case put E=K{r−1T,t−1S,tS−1}E=K\{r^{-1}T,t^{-1}S,tS^{-1}\}. By the defining quotient norm on DD, choose lifts Gi,G∈EG_i,G\in E of gi,gg_i,g such that ∥Ti−Gi∥E<min⁡{ri,∥fi−gi∥D+δ},∥S−G∥E<min⁡{t,∥f−g∥D+δ}.\begin{split} \|T_i-G_i\|_E&<\min\{r_i,\|f_i-g_i\|_D+\delta\},\\ \|S-G\|_E&<\min\{t,\|f-g\|_D+\delta\}. \end{split} Because SS has constant modulus tt on EE, the substitution Ti↦GiT_i\mapsto G_i, S↦GS\mapsto G is an isometric automorphism by the lemma. Let B′B' be the lower Laurent algebra in the Ti,ST_i,S coordinates and C′C' the upper Laurent algebra in the Gi,GG_i,G coordinates, both regarded as subalgebras of EE. Their spectra glue along M(E)\mathcal M(E) to an affinoid space X′=M(A′)X'=\mathcal M(A') by the free case. There are compatible epimorphisms B′→BB'\to B, C′→CC'\to C, E→DE\to D.

Let IB,IC,IDI_B,I_C,I_D be their kernels. The stated descriptions of the overlap and the quotient formula for rational localization give IBE=ID=ICEI_BE=I_D=I_CE. The ideals are finite modules, so they glue to a coherent ideal sheaf on X′X'. Theorem 2.38.5 identifies this sheaf with I~\widetilde I for an ideal I⊂A′I\subset A', and identifies its two restrictions with IB,ICI_B,I_C. It follows that M(A′/I)\mathcal M(A'/I) has affinoid domains with algebras B,CB,C and overlap DD. Theorem 2.40.10 therefore identifies XX with M(A′/I)\mathcal M(A'/I). Finally the free-case coordinates on A′A' descend to A′/IA'/I; quotient maps do not increase the displayed quotient norms. The chosen lifting estimates give all required bounds. ◻

References. The two-chart argument reconstructs [Tem00] (Lemma 3.2) with arbitrary positive radii, as required in [Tem04] (Theorem 5.1). The proof uses the internal rational localization, finite-module descent, and analytic gluing results.

2.46 Relative interior on arbitrary analytic spaces

The reduction of a germ records all the affinoid directions through its distinguished point. The good-germ criterion now makes it possible to pass from the affinoid interior calculation to arbitrary analytic spaces.

NAG-001808 · Open at full scope

Theorem 2.46.1. NAG-001808 (The intrinsic interior criterion). Let f:Y→Xf:Y\to X be a morphism of analytic spaces, let y∈Yy\in Y, and put x=f(y)x=f(y). With the all-target-domain definition of relative interior, y∈Int⁡(Y/X)⟺R(Y,y)⟶R(X,x) is proper.y\in\operatorname{Int}(Y/X) \quad\Longleftrightarrow\quad \mathscr R(Y,y)\longrightarrow\mathscr R(X,x) \text{ is proper}.

Proof. Suppose first that yy is interior. Choose finitely many affinoid domains Xi⊂XX_i\subset X, all containing xx, whose union is a neighborhood of xx. Write Yi=f−1(Xi)Y_i=f^{-1}(X_i). By Definition 2.44.4, each YiY_i has an affinoid neighborhood ViV_i of yy with y∈Int⁡(Vi/Xi)y\in\operatorname{Int}(V_i/X_i). Proposition 3.19.14 says that the reduction of (Vi,y)→(Xi,x)(V_i,y)\to(X_i,x) is proper. Since ViV_i is an ordinary neighborhood in YiY_i, it represents the same germ. Theorem 3.19.9 identifies these morphisms with the restrictions of the original reduction over the open cover R(Xi,x)\mathscr R(X_i,x) of R(X,x)\mathscr R(X,x). Target locality in Corollary 3.19.13 proves properness.

Conversely, suppose the reduction is proper and let U⊂XU\subset X be any affinoid domain containing xx. Put V=f−1(U)V=f^{-1}(U). The inverse-image formula for germ reductions and Corollary 3.19.13 show that R(V,y)→R(U,x)\mathscr R(V,y)\to\mathscr R(U,x) is proper. The latter target is affine. By the definition of properness for birational spaces, its source is the pullback of that affine object to H(y)~gr\widetilde{\mathcal H(y)}^{\rm gr}, hence is affine. Theorem 3.19.19 therefore makes (V,y)(V,y) a good germ. Choose an affinoid neighborhood WW of yy in VV. Its reduction agrees with that of VV, so Proposition 3.19.14 gives y∈Int⁡(W/U)y\in\operatorname{Int}(W/U). This is the witness required for the arbitrary target domain UU. ◻

NAG-001809 · Open at full scope

Theorem 2.46.2. NAG-001809 (Openness of relative interior). For every morphism of analytic spaces, Int⁡(Y/X)\operatorname{Int}(Y/X) is open in the ordinary topology of YY.

Proof. Fix an interior point yy and write x=f(y)x=f(y). Work inside Hausdorff open neighborhoods of xx and yy; these do not alter their germs or the criterion in Theorem 2.46.1. Choose finitely many affinoid domains XiX_i containing xx whose union contains an open neighborhood NN of xx. As in the proof of that theorem, for Yi=f−1(Xi)Y_i=f^{-1}(X_i) choose affinoid neighborhoods ViV_i of yy in YiY_i such that y∈Int⁡(Vi/Xi)y\in\operatorname{Int}(V_i/X_i). Affinoid interior is open. Consequently there is an open neighborhood ZiZ_i of yy in YiY_i, contained in ViV_i, on which the reduction of (Yi,z)→(Xi,f(z))(Y_i,z)\to(X_i,f(z)) is proper for every z∈Ziz\in Z_i. Here we first restrict to the ordinary interior of ViV_i in YiY_i, so the two source germs agree.

For each ii, choose an ordinary open neighborhood OiO_i of yy in YY with Oi∩Yi⊂ZiO_i\cap Y_i\subset Z_i, and put O=f−1(N)∩⋂iOiO=f^{-1}(N)\cap\bigcap_iO_i. This is an open neighborhood of yy. For z∈Oz\in O, put x′=f(z)x'=f(z). The members XiX_i containing x′x' cover the germ (X,x′)(X,x'): the other finitely many compact domains are closed in the chosen Hausdorff neighborhood and can be removed near x′x', while N⊂⋃iXiN\subset\bigcup_iX_i. For every remaining member, z∈Ziz\in Z_i, so the reduced morphism over R(Xi,x′)\mathscr R(X_i,x') is proper. Target locality proves properness of R(Y,z)→R(X,x′)\mathscr R(Y,z)\to\mathscr R(X,x'). The intrinsic criterion yields z∈Int⁡(Y/X)z\in\operatorname{Int}(Y/X). Thus OO is contained in the relative interior. ◻

NAG-001810 · Open at full scope

Theorem 2.46.3. NAG-001810 (Domain locality and analytic base change). Let (Xi)(X_i) be an admissible analytic-domain covering of XX and put Yi=f−1(Xi)Y_i=f^{-1}(X_i). Then ∂(Y/X)=⋃i∂(Yi/Xi).\partial(Y/X)=\bigcup_i\partial(Y_i/X_i). If X′→XX'\to X is an analytic morphism over any complete valued extension L/KL/K, Y′=Y×XX′Y'=Y\times_X X', and y′∈Y′y'\in Y' maps to y∈Yy\in Y, then y∈Int⁡(Y/X)⟹y′∈Int⁡(Y′/X′).y\in\operatorname{Int}(Y/X) \quad\Longrightarrow\quad y'\in\operatorname{Int}(Y'/X').

Proof. At each x∈Xx\in X, admissibility supplies a finite subdomain-germ covering subordinate to the given family. Theorem 3.19.9 turns it into an open covering of R(X,x)\mathscr R(X,x) and identifies the reductions of the inverse-image germs with the corresponding inverse images. A reduction morphism is proper if and only if all these restrictions are proper, by Corollary 3.19.13. Theorem 2.46.1 gives the displayed boundary identity. For an arbitrary analytic base change over LL, the same corollary preserves properness of the reduction. Applying the intrinsic criterion on Y′Y' proves the second assertion. ◻

NAG-001811 · Open at full scope

Corollary 2.46.4. NAG-001811 (The universal good-space test). A morphism f:Y→Xf:Y\to X is boundaryless if and only if, for every complete valued extension L/KL/K and every morphism X′→XX'\to X from a good LL-analytic space, the space Y′=Y×XX′Y'=Y\times_X X' is good and the induced good-space morphism Y′→X′Y'\to X' is boundaryless.

Proof. Suppose ff is boundaryless. The preceding theorem makes every point of Y′Y' interior over X′X'. At y′∈Y′y'\in Y', the germ of X′X' at its image is good and has affine reduction. Its proper inverse image is affine, so Theorem 3.19.19 makes the source germ good. Every point thus has an affinoid neighborhood, proving that Y′Y' is good. Conversely, apply the stated test with L=KL=K to every target affinoid domain U⊂XU\subset X. Its inverse image is good and boundaryless over UU, so every point admits an affinoid neighborhood with an inner presentation. Definition 2.44.4 gives that every point of YY is interior. ◻

References. These arguments reconstruct [Tem04] (Lemma 5.5, Theorem 5.2 and Corollary 5.6). The good-germ, graded product, and scalar-extension theorems used in the proof are established internally.

Forward interior over an affinoid target. Let f:Y→Xf:Y\to X be an analytic morphism with XX affinoid. Let X′→XX'\to X be an analytic morphism over a complete valued extension L/KL/K. Put Y′=Y×XX′Y'=Y\times_X X', and let y′∈Y′y'\in Y' map to y∈Yy\in Y. If y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X), then y′∈Int⁡(Y′/X′)y'\in\operatorname{Int}(Y'/X').

Proof. Use XX itself as the target test in Definition 2.44.4. It gives an affinoid domain VV of YY which is an ordinary neighbourhood of yy, with y∈Int⁡(V/X)y\in\operatorname{Int}(V/X). For an arbitrary affinoid domain U′U' of X′X' containing the image of y′y', put P′=Y′×X′U′P'=Y'\times_{X'}U'. The inverse image of VV in P′P' is V′=V×XU′V'=V\times_X U'. The analytic-domain and affinoid product assertions of Theorems 2.40.11 and 2.40.12 identify this with the affinoid chart for the actual completed tensor product. The ordinary open neighbourhood of yy contained in VV pulls back to one of y′y' contained in V′V', so V′V' is an affinoid neighbourhood of y′y' in the test space P′P'.

The forward affinoid base-change inclusion in Proposition 2.33.21, applied to V→X←U′V\to X\leftarrow U', puts y′y' in Int⁡(V′/U′)\operatorname{Int}(V'/U'). This works for every target test U′U', which is the required relative interior condition. ◻

2.47 Universal flatness of analytic domains

The affinoid stalk calculation in Theorem 2.33.17 already contains the analytic argument: finite tensor relations descend to sufficiently small affinoid neighborhoods. We now transfer that result to arbitrary domains and good-space tests.

Theorem 2.47.1. NAG-001665 (All analytic-domain inclusions are analytically flat). Every analytic-domain inclusion j:U↪Xj:U\hookrightarrow X is analytically flat, over any complete non-Archimedean ground field. If UU and XX are good, the ordinary local homomorphism OX,x⟶OU,x(x∈U)\mathcal O_{X,x}\longrightarrow\mathcal O_{U,x} \qquad(x\in U) is faithfully flat. In this good case, after every base change from a good space over every complete valued extension, the resulting domain is good and the same faithful-flat assertion holds.

Proof. First assume UU and XX good. Choose an affinoid neighborhood X0X_0 of xx in XX, and then an affinoid neighborhood U0U_0 of xx in UU contained in X0X_0. By Corollary 2.40.9, U0U_0 is an affinoid domain of X0X_0. Ordinary neighborhoods do not change stalks, so Theorem 2.33.17 gives the claimed flatness. The homomorphism is local by Proposition 2.33.16 and residue-field compatibility.

For clarity, a flat local homomorphism R→SR\to S of nonzero local rings is faithful for the following elementary reason. If N≠0N\ne 0, a nonzero element of NN generates a submodule R/IR/I with II proper. Flatness embeds S/ISS/IS into S⊗RNS\otimes_RN, and ISIS is contained in the maximal ideal of SS, so this tensor product is nonzero. Apply this to the two stalks.

Let X′→XX'\to X be a morphism from a good LL-analytic space, with L/KL/K complete. Its inverse image U′U' is an analytic domain by Theorem 2.40.8 and the construction of analytic products. To verify that U′U' is good in the current good-source case, fix u′∈U′u'\in U', with images u,x′u,x'. Choose affinoid neighborhoods X0X_0 of uu in XX, U0U_0 of uu in UU contained in X0X_0, and X0′X'_0 of x′x' in X′X' contained in the inverse image of the ordinary interior of X0X_0. Then U0′=U0×X0X0′U'_0=U_0\times_{X_0}X'_0 is an affinoid domain of X0′X'_0, by Theorem 2.36.9, and contains an ordinary neighborhood of u′u' in U′U': intersect the inverse images of the ordinary interiors of U0U_0 and X0′X'_0. Thus U′U' is good. The first part, applied over LL, proves faithful flatness on its ordinary stalks. This verifies every good base-change test in Definition 2.9.1.

For a general domain inclusion, choose an affinoid chart of the target and an affinoid chart of its inverse-image domain. The latter is an affinoid domain of the former. What was just proved supplies all universal good-space flatness tests for this local map. Such charts form the analytic nets on source and target. The general GG-local definition of analytic flatness therefore proves the assertion without assuming that an arbitrary domain of a good space is itself good. ◻

Proposition 2.47.2. NAG-001666 (Coherent restriction on ordinary stalks). Let U⊂YU\subset Y be an analytic domain, with U,YU,Y good, and let F\mathcal F be coherent on YGY_G. For every y∈Uy\in U, the canonical map gives (F∣UG)y≃Fy⊗OY,yOU,y.(\mathcal F|_{U_G})_y \simeq\mathcal F_y\otimes_{\mathcal O_{Y,y}}\mathcal O_{U,y}.

Proof. Choose affinoid neighborhoods Y0Y_0 in YY and U0U_0 in UU with U0⊂Y0U_0\subset Y_0. By Theorem 2.39.6, F∣Y0\mathcal F|_{Y_0} comes from a finite module MM over A=O(Y0)A=\mathcal O(Y_0). On every affinoid domain V⊂Y0V\subset Y_0 its sections are M⊗AAVM\otimes_AA_V. Ordinary stalks are filtered limits over affinoid neighborhoods, by Proposition 2.33.16; the same proof applies to these module sheaves. Tensor products commute with filtered colimits, so the two stalks are respectively M⊗AOY,yM\otimes_A\mathcal O_{Y,y} and M⊗AOU,yM\otimes_A\mathcal O_{U,y}. Associativity of tensor products gives the displayed isomorphism. ◻

Corollary 2.47.3. NAG-001667 (Flatness detected on source domains). Let f:Y→Xf:Y\to X be a morphism of good spaces, let F\mathcal F be coherent on YGY_G, and let U⊂YU\subset Y be a good analytic domain containing yy. Then F\mathcal F is naively XX-flat at yy if and only if its restriction to UU is. The same equivalence holds for analytic XX-flatness.

Proof. Put A=OX,f(y)A=\mathcal O_{X,f(y)}, R=OY,yR=\mathcal O_{Y,y}, S=OU,yS=\mathcal O_{U,y}, and M=FyM=\mathcal F_y. By the preceding results R→SR\to S is faithfully flat and the restricted module is M⊗RSM\otimes_RS. If MM is AA-flat, tensoring any injective map of AA-modules first with MM and then with the flat RR-module SS preserves injectivity; hence M⊗RSM\otimes_RS is AA-flat. Conversely, for an injective AA-linear map N1→N2N_1\to N_2, the kernel of N1⊗AM→N2⊗AMN_1\otimes_AM\to N_2\otimes_AM is an RR-module. Its tensor product with SS vanishes if M⊗RSM\otimes_RS is AA-flat, and faithfulness makes the kernel zero. This proves naive-flatness equivalence.

A fibre product of good spaces is good: at a product point choose affinoid neighborhoods of its three images, with the two source neighborhoods mapping into the target neighborhood. Their affinoid product contains the intersection of the inverse images of the two ordinary source interiors, hence is a neighborhood of the product point. After any good base change over a complete extension, the inverse image of UU is therefore a good analytic domain in a good source, by Theorem 2.47.1. Coherent pullback commutes with domain restriction, as follows from its local finite-module presentations. Apply the naive equivalence at every point above yy. These are exactly the tests defining analytic flatness. ◻

2.48 Flatness of relative polydiscs

The coefficient description of finite modules supplies the algebraic flatness of relative Tate algebras. Passing to ordinary analytic stalks requires two filtered systems of neighborhoods, but no Noetherianity assertion about their limit.

Proposition 2.48.1. NAG-001887 (Flat relative Tate algebras). Let AA be a generalized affinoid algebra over a complete non-Archimedean field, and let r=(r1,…,rn)r=(r_1,\ldots,r_n) be positive real radii. Then A→B=A{r−1T}A\to B=A\{r^{-1}T\} is faithfully flat as a map of rings.

Proof. For every finite AA-module MM, Theorem 1.18.6 identifies B⊗AMB\otimes_A M with the convergent weighted coefficient module Cr(Nn,M)C_r(\mathbf N^n,M), functorially in MM. The same theorem preserves injections of finite modules. Since AA is Noetherian by Theorem 1.18.2, every finitely generated ideal II is a finite module; applying the assertion to I↪AI\hookrightarrow A proves that I⊗AB→BI\otimes_A B\to B is injective. The finite-ideal criterion [StacksAlg] (Tag 00HD) gives flatness.

The constant coefficient map p0:B→Ap_0:B\to A is an AA-linear retraction of the coefficient inclusion. For every AA-module NN, the map N→B⊗ANN\to B\otimes_A N, m↦1⊗mm\mapsto 1\otimes m, is consequently split by p0⊗1p_0\otimes 1. Thus tensoring with BB detects nonzero modules, which proves faithfulness. Only the linear retraction is used here; multiplicativity is unnecessary. ◻

Theorem 2.48.2. NAG-001888 (Relative polydisc flatness on ordinary stalks). Let P=M(A{r−1T})→X=M(A)P=\mathcal M(A\{r^{-1}T\})\to X=\mathcal M(A) be the relative closed polydisc, and let y↦xy\mapsto x. Then the local map OX,x→OP,y\mathcal O_{X,x}\to\mathcal O_{P,y} is faithfully flat.

Proof. Write R=OX,xR=\mathcal O_{X,x} and S=OP,yS=\mathcal O_{P,y}. Fix a rational domain VV in XX containing an ordinary neighborhood of xx. By the bounded tensor universal property, the inverse image PVP_V has algebra AV{r−1T}A_V\{r^{-1}T\}. It contains an ordinary neighborhood of yy. The affinoid neighborhoods WW of yy in PP contained in PVP_V therefore form a cofinal system. Each is an affinoid domain of PVP_V, by the domain restriction calculus. The preceding proposition and Proposition 2.38.2 give flat maps AV⟶AV{r−1T}⟶AW.A_V\longrightarrow A_V\{r^{-1}T\}\longrightarrow A_W. Their composite is flat because the two consecutive tensor functors preserve injections. Ordinary stalks are the filtered colimits of these AWA_W by Proposition 2.33.16. For a finitely generated ideal J⊂AVJ\subset A_V, any element of J⊗AVSJ\otimes_{A_V}S is a finite sum represented at one stage. If its image in SS is zero, that image vanishes after shrinking WW once more. Flatness of the new stage then makes the tensor itself zero. The finite-ideal criterion proves that SS is flat over each fixed AVA_V.

Finally, take a finitely generated ideal I⊂RI\subset R and z∈I⊗RSz\in I\otimes_R S whose image in SS is zero. Choose generators a1,…,ama_1,\ldots,a_m of II and write z=∑iai⊗siz=\sum_i a_i\otimes s_i with si∈Ss_i\in S. Lift the aia_i to elements a~i\widetilde a_i of one AVA_V, and put J=(a~1,…,a~m)J=(\widetilde a_1,\ldots,\widetilde a_m). Then zV=∑ia~i⊗si∈J⊗AVSz_V=\sum_i\widetilde a_i\otimes s_i\in J\otimes_{A_V}S maps to zz and has zero image in SS. The flatness just proved gives zV=0z_V=0, hence z=0z=0. Again [StacksAlg] (Tag 00HD) gives RR-flatness. Evaluations at xx and yy are compatible through the isometric map H(x)↪H(y)\mathcal H(x)\hookrightarrow\mathcal H(y), so the map is local. A nonzero module contains a cyclic submodule R/IR/I with II proper; its tensor S/ISS/IS is nonzero because ISIS lies in the maximal ideal of SS. Flatness preserves the cyclic injection and therefore proves faithful flatness. ◻

Corollary 2.48.3. NAG-001889 (Analytic flatness of polydisc and affine-space projections). Over a good analytic space XX, a relative closed polydisc of any positive radii and relative analytic affine space are good, and their projections to XX are analytically flat. The same holds for the restriction of either projection to a good analytic domain of its source. These assertions include every good base change over every complete valued extension of the ground field.

Proof. At a polydisc point over xx, choose an affinoid neighborhood of xx. Its inverse image is an affinoid polydisc containing a neighborhood of the chosen point. Thus the total space is good, and the preceding theorem gives flatness of its ordinary local map. After a good base change over a complete extension, the bounded tensor universal property identifies the pullback with the polydisc of the same radii. Its ordinary stalk maps are flat by the same proof over that extension. This is analytic flatness.

Every point of relative analytic affine space has a closed polydisc neighborhood: take each radius strictly larger than the value of its coordinate, and an affinoid neighborhood in the base. The interior inequalities give an ordinary neighborhood, so stalks agree with those of that polydisc. The construction works after every complete-extension good base change, proving the affine-space assertion. Finally, Theorem 2.47.1 supplies flat ordinary domain-stalk maps after all such base changes, with good pulled-back domains. Composing them with the projection-stalk maps proves the last assertion. ◻

2.49 Isolated fibres and finite neighbourhoods

The finite criterion for affinoid morphisms also has a pointwise form. Its proof separates two compact sets: the relative analytic boundary, which detects finiteness, and the topological boundary of the chosen source neighbourhood, which allows one to work on ordinary open sets.

Theorem 2.49.1. NAG-001876 (The local finite criterion). Let f:Y→Xf:Y\to X be a morphism of good analytic spaces, and let y∈Yy\in Y, x=f(y)x=f(y). The following conditions are equivalent:

  1. yy is isolated in the analytic fibre YxY_x and y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X);

  2. there are ordinary open neighbourhoods W⊂YW\subset Y of yy and N⊂XN\subset X of xx such that f(W)⊂Nf(W)\subset N and W→NW\to N is norm-admissible finite.

When these conditions hold, WW and NN may be required to lie in any prescribed ordinary neighbourhoods of yy and xx.

Proof. Assume the first condition. By the topological identification of analytic fibres, there is an ordinary open neighbourhood of yy meeting f−1(x)f^{-1}(x) only in yy. Choose an affinoid neighbourhood X0X_0 of xx inside the prescribed target neighbourhood. Goodness supplies an affinoid neighbourhood VV of yy contained in this isolating source neighbourhood, in the prescribed source neighbourhood, and in f−1(Int⁡top(X0,X))f^{-1}(\operatorname{Int}_{\rm top}(X_0,X)). Thus Vx={y}V_x=\{y\} as a set. Proposition 2.44.6 gives y∈Int⁡(V/X0)y\in\operatorname{Int}(V/X_0).

The relative boundary D=∂(V/X0)D=\partial(V/X_0) is closed in the compact space VV. Its image in the Hausdorff affinoid space X0X_0 is therefore compact and closed. This image does not contain xx: the only point of VxV_x is yy, which is interior. Choose a rational neighbourhood UU of xx in X0∖f(D)X_0\setminus f(D) and inside Int⁡top(X0,X)\operatorname{Int}_{\rm top}(X_0,X). The inverse image VU=V×X0UV_U=V\times_{X_0}U is affinoid. Every one of its points comes from Int⁡(V/X0)\operatorname{Int}(V/X_0), so interior stability under base change makes VU→UV_U\to U boundaryless. Theorem 2.43.8 now makes this morphism norm-admissible finite.

It remains to obtain ordinary open neighbourhoods. Since both VV and UU were neighbourhoods of the chosen points, VUV_U is a neighbourhood of yy in YY. Set E=VU∖Int⁡top(VU,Y).E=V_U\setminus\operatorname{Int}_{\rm top}(V_U,Y). This is a closed subset of the compact space VUV_U, and its image in UU is closed. The image avoids xx, because (VU)x={y}(V_U)_x=\{y\} and y∉Ey\notin E. Choose an ordinary open neighbourhood NN of xx in XX, contained in Int⁡top(U,X)\operatorname{Int}_{\rm top}(U,X) and disjoint from f(E)f(E). Put W=VU∩f−1(N)W=V_U\cap f^{-1}(N). Then WW is contained in Int⁡top(VU,Y)\operatorname{Int}_{\rm top}(V_U,Y), and hence W=Int⁡top(VU,Y)∩f−1(N)W=\operatorname{Int}_{\rm top}(V_U,Y)\cap f^{-1}(N) is open in YY. The map W→NW\to N is the base change of the finite map VU→UV_U\to U, so it is finite by Proposition 2.33.19 and analytic-domain gluing. It contains yy and has all the required neighbourhood restrictions.

Conversely, suppose the finite map is given on ordinary open neighbourhoods W→NW\to N. Choose an affinoid neighbourhood X0⊆NX_0\subseteq N of xx on a finite chart, so its full inverse image VV in WW is affinoid and V→X0V\to X_0 is norm-admissible finite. As X0X_0 is a neighbourhood of xx and WW is open in YY, VV is a neighbourhood of yy in YY. Theorem 2.43.8 gives y∈Int⁡(V/X0)y\in\operatorname{Int}(V/X_0), and Proposition 2.44.6 gives y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X). By Lemma 2.35.1, the finite fibre in WW is discrete. Since WW is open, yy is isolated in the original fibre. ◻

Proposition 2.49.2. NAG-001877 (An ordinary local criterion for étaleness). Let f:Y→Xf:Y\to X be a morphism of good analytic spaces. Suppose y∈Yy\in Y is isolated in its analytic fibre and belongs to Int⁡(Y/X)\operatorname{Int}(Y/X). Then ff is étale at yy if and only if the local map OX,x→OY,y\mathcal O_{X,x}\to\mathcal O_{Y,y} is flat and (ΩY/X1)y=0(\Omega^1_{Y/X})_y=0, where x=f(y)x=f(y). In this case one can choose ordinary neighbourhoods on which ff is finite locally free and has zero relative differential sheaf.

Proof. The local finite criterion and Lemma 2.35.1 reduce the converse to a finite morphism M(B)→M(A)\mathcal M(B)\to\mathcal M(A) with fibre over xx consisting only of yy, where these affinoids are neighbourhoods of the original points. Put R=OX,xR=\mathcal O_{X,x}, S=OY,yS=\mathcal O_{Y,y}, and p=ker⁡(A→H(x))\mathfrak p=\ker(A\to\mathcal H(x)). The finite-branch stalk identity is B⊗AR≃S.B\otimes_A R\simeq S. The local map Ap→RA_{\mathfrak p}\to R is faithfully flat. Indeed, Proposition 2.33.16 writes RR as a filtered colimit of rational-neighbourhood algebras, all flat over AA by Proposition 2.38.2. It is consequently flat over ApA_{\mathfrak p}. The maximal ideal contracts to pAp\mathfrak p A_{\mathfrak p} by evaluation, so faithful flatness follows from [StacksAlg] (Tag 00HP).

If SS is RR-flat, then BpB_{\mathfrak p} is ApA_{\mathfrak p}-flat. For any injection of ApA_{\mathfrak p}-modules, the kernel after tensoring with BpB_{\mathfrak p} becomes zero after the faithfully flat scalar extension to RR, because the extended module is SS. Thus that kernel is zero. As BB is finite over the Noetherian ring AA, [StacksAlg] (Tag 00NX) makes BpB_{\mathfrak p} free of finite rank. Lift a basis and its inverse finite presentation to a localization AaA_a with a∉pa\notin\mathfrak p; clearing the finitely many kernel and cokernel generators after a further such localization gives Ba≃AamB_a\simeq A_a^m.

Let P=ΩB/A1P=\Omega^1_{B/A}, a finite BB-module and hence a finite AA-module. Here algebraic and analytic differentials agree. Indeed an algebraic AA-derivation from the finite AA-module BB into a finite BB-module is AA-linear and automatically bounded for the canonical finite-module topologies. Thus the two finite differential modules represent the same derivations, by Theorem 2.41.3. The finite-branch stalk argument, applied to a finite presentation of PP, gives P⊗AR≃(ΩY/X1)yP\otimes_A R\simeq(\Omega^1_{Y/X})_y. The assumed vanishing and faithful flatness imply Pp=0P_{\mathfrak p}=0. After enlarging a∉pa\notin\mathfrak p, finite generation gives Pa=0P_a=0. On the ordinary open neighbourhood {∣a∣>0}\{|a|>0\} of xx, the finite morphism is therefore locally free and has zero relative differential sheaf. To verify the first assertion analytically, restrict to a rational neighbourhood where ∣a∣|a| is bounded below. Its algebra contains a−1a^{-1}, so tensoring Ba≃AamB_a\simeq A_a^m gives a free finite algebra module there. Its ordinary local rings are the factors in the finite-branch decomposition; those factors are direct summands of a free module after passage to the base stalk. They are thus flat. The same argument applies after every analytic base change, since finite-module base change is ordinary tensor product. This proves the claimed étaleness and analytic flatness.

The forward implication follows from the local finite flat definition of an étale morphism and the same finite-branch stalk identity. ◻

References. The isolated-fibre criterion follows the compact-boundary method of [Ber93] (Proposition 3.1.4). The proof above supplies the passage from affinoid pieces to ordinary open neighbourhoods explicitly. The second result combines this criterion with finite-branch separation and finite-module flatness.

2.50 Quasi-smoothness at interior points

Lemma 2.50.1. NAG-002007 (Rank-zero quasi-smooth fibres). Let ZZ be a good analytic space over a complete non-Archimedean field FF. If Z→M(F)Z\to\mathcal M(F) is quasi-smooth at zz and rk⁡zΩZ/F=0\operatorname{rk}_z\Omega_{Z/F}=0, then zz has an ordinary open affinoid neighbourhood finite over FF. In particular, zz is isolated in ZZ.

Proof. Choose an affinoid neighbourhood Z0Z_0 of zz. By domain locality of Jacobian presentations, take an affinoid domain U=M(C)U=\mathcal M(C) contained in Int⁡top(Z0,Z)\operatorname{Int}_{\rm top}(Z_0,Z), containing zz, with a closed presentation in a domain of AFn\mathbb A_F^n by nn equations whose square Jacobian determinant is invertible. Such a domain is obtained from a rank-zero presentation by the Laurent restriction in Theorem 2.41.7. All points of UU have the same rank-zero Jacobian description.

Choose a nontrivially valued complete extension L/FL/F such that CLC_L is strict, using Theorem 1.17.6 and, if necessary, one further free-radius extension. Let m\mathfrak m be any maximal ideal of CLC_L. By Theorem 1.12.3, E=CL/mE=C_L/\mathfrak m is finite over LL with its extending complete absolute value. Evaluation gives an EE-rational point u∈UEu\in U_E. The pulled-back equations retain their invertible determinant. Proposition 2.35.13, with n−r=0n-r=0, identifies the completion of S=OUE,uS=\mathcal O_{U_E,u} with EE. The ring SS is Noetherian by Theorem 2.35.11. Since its completion is faithfully flat by [StacksAlg] (Tag 00MC), its maximal ideal is zero: that ideal becomes zero in its completion, and faithful flatness detects the zero module. Thus SS is a field.

The local map (CL)m→S(C_L)_{\mathfrak m}\to S is faithfully flat. Indeed CL→CEC_L\to C_E is finite free by finite-extension base change, and the rational-neighbourhood description of SS expresses it as a filtered colimit of algebras flat over CEC_E. The composite is flat, and it is local after localization at m\mathfrak m; use [StacksAlg] (Tag 00HP). A local ring with a faithfully flat local map to a field is itself a field: its maximal ideal extends to zero, so is zero. Thus (CL)m(C_L)_{\mathfrak m} is a field for every maximal m\mathfrak m. Every prime of CLC_L is maximal, since a strict inclusion of a prime in a maximal ideal would remain a strict inclusion after localization there. Therefore dim⁡CL=0\dim C_L=0. Finite strict normalization and Proposition 3.21.1 give a finite injection from a Tate algebra on zero variables. Consequently CLC_L is finite-dimensional over LL.

Every finite FF-linearly independent family in CC remains LL-linearly independent in CLC_L: algebraic scalar extension preserves linear independence, and Theorem 2.33.2 embeds the algebraic tensor product in its completion. Thus dim⁡FC≤dim⁡LCL<∞\dim_F C\le\dim_L C_L<\infty. The canonical finite-dimensional topology is the affinoid topology by Lemma 1.12.1, so U→M(F)U\to\mathcal M(F) is norm-admissible finite. It is boundaryless by Theorem 2.43.8. The composition formula for U→Z0→M(F)U\to Z_0\to\mathcal M(F) implies Int⁡(U/Z0)=U\operatorname{Int}(U/Z_0)=U, and Theorem 2.44.3 makes UU open in Z0Z_0. Since U⊂Int⁡top(Z0,Z)U\subset\operatorname{Int}_{\rm top}(Z_0,Z), it is open in ZZ. A finite-dimensional affinoid algebra has finitely many spectral points, a discrete set by Lemma 2.35.1. This proves the assertion. ◻

Proposition 2.50.2. NAG-002008 (Finite étale maps are quasi-smooth). A norm-admissible finite flat affinoid morphism with zero relative differential module is analytically flat and has geometrically regular analytic fibres. Consequently it is quasi-smooth, and these assertions hold after every complete-field extension and analytic base change. The same conclusions hold locally for étale analytic morphisms.

Proof. Finite-module base change preserves finite flatness. On ordinary local rings, the finite-branch product decomposition makes each source stalk a direct summand of the finite flat algebra tensored with the base stalk. These stalks are therefore flat. This reasoning applies after every analytic base change, proving analytic flatness. Algebraic and bounded analytic differentials agree for finite algebras, by the finite-module argument in Proposition 2.49.2; their vanishing is preserved by base change.

Consider an analytic fibre and any complete extension EE of its ground field. Its algebra DD is finite-dimensional over EE and satisfies ΩD/E1=0\Omega^1_{D/E}=0. Fix a point vv and put L=H(v)L=\mathcal H(v), a finite complete extension of EE. Lift vv to an LL-rational point uu of M(D⊗EL)\mathcal M(D\otimes_E L). The local ring SS at uu is a finite-dimensional local LL-algebra with residue field LL, by the finite-branch stalk product. Its ordinary algebraic differential module over LL is zero. If n\mathfrak n is its maximal ideal, the augmentation S→LS\to L identifies n/n2≃ΩS/L1⊗SL=0.\mathfrak n/\mathfrak n^2 \simeq\Omega^1_{S/L}\otimes_S L=0. For completeness, the isomorphism sends the class of bb to dbdb for b∈nb\in\mathfrak n. Its inverse is induced by the derivation s↦s−s‾s\mapsto s-\overline s modulo n2\mathfrak n^2, with s‾∈L\overline s\in L the residue; the product rule is immediate modulo n2\mathfrak n^2. The finite module n\mathfrak n satisfies n=n2\mathfrak n=\mathfrak n^2, so the determinant proof of Nakayama gives n=0\mathfrak n=0. Hence S=LS=L is a field. The scalar-stalk map from OM(D),v\mathcal O_{\mathcal M(D),v} to SS is faithfully flat by Theorem 2.35.4. As in the previous lemma, its source is a field as well, and is therefore regular. Since EE and vv were arbitrary, the fibres are geometrically regular. Theorem 2.35.14 proves quasi-smoothness. Ordinary-local restriction gives the final assertion. ◻

Theorem 2.50.3. NAG-002009 (Smoothness and interior quasi-smoothness). For a morphism f:Y→Xf:Y\to X of good analytic spaces and a point y∈Yy\in Y, ff is smooth at yy if and only if it is quasi-smooth at yy and y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X). Here smoothness is the ordinary-open local étale factorization through a relative analytic affine space.

Proof. Suppose first that ff is quasi-smooth and interior at yy. The strengthened presentation in Theorem 2.35.14 supplies affinoid neighbourhoods Y0Y_0 of yy and X0X_0 of x=f(y)x=f(y), together with a closed immersion Y0↪D⊂AX0nY_0\hookrightarrow D\subset\mathbb A_{X_0}^n whose ideal is generated by rr functions and whose Jacobian has an invertible r×rr\times r minor near yy. Apply the ordinary Laurent-neighbourhood restriction to make the minor invertible throughout the chosen source. Permute coordinates so that it uses T1,…,TrT_1,\ldots,T_r, and put d=n−rd=n-r. The remaining coordinates define g:Y0⟶AX0d,g=(f,Tr+1,…,Tn).g:Y_0\longrightarrow\mathbb A_{X_0}^{d},\qquad g=(f,T_{r+1},\ldots,T_n). Choose a closed relative polydisc P⊂AX0dP\subset\mathbb A_{X_0}^{d} whose radii are strictly larger than the corresponding coordinate bounds on the affinoid DD. Then g(Y0)⊂Pg(Y_0)\subset P, and PP is a neighbourhood of g(y)g(y). The domain DD lies in APr\mathbb A_P^r and remains an affinoid domain there by the domain mapping property. The same rr equations with their invertible square Jacobian make Y0→PY_0\to P quasi-smooth with zero relative differential sheaf.

Proposition 2.44.6 gives y∈Int⁡(Y0/X0)y\in\operatorname{Int}(Y_0/X_0). The affinoid composition formula for Y0→P→X0Y_0\to P\to X_0 implies y∈Int⁡(Y0/P)y\in\operatorname{Int}(Y_0/P). The analytic fibre of Y0→PY_0\to P through yy is good and quasi-smooth of differential rank zero, by base change. Lemma 2.50.1 makes yy isolated in that fibre. Theorem 2.35.14 gives the ordinary local flatness of Y0→PY_0\to P. Proposition 2.49.2 now makes it étale near yy. Shrink its ordinary finite neighbourhoods inside the ordinary interiors of Y0Y_0 in YY and PP in AX0d\mathbb A_{X_0}^d; the small-neighbourhood assertion of Theorem 2.49.1 permits this. Restrict also inside the ordinary interior of X0X_0 in XX. We obtain the required ordinary-open étale factorization of ff, proving smoothness.

Conversely, a local étale map is quasi-smooth by Proposition 2.50.2. A relative analytic affine-space projection is quasi-smooth by its presentation with no equations. The composition theorem for Jacobian presentations therefore makes every smooth morphism quasi-smooth. The finite local maps defining étaleness are boundaryless by the affinoid finite criterion. Relative affine-space projections are boundaryless because, at each point, one may choose coordinate radii strictly larger than the coordinate values. The composition inclusion for relative interior, applied on good affinoid neighbourhoods, and ordinary-local testing then make the composite boundaryless at yy. ◻

References. The smoothness comparison is [Duc18] (Corollary 5.4.8). The proof here combines the strengthened Jacobian presentation, the completed rank-zero calculation and the local finite criterion; these are all proved internally.

2.51 Finite fibre projections and original coefficients

A finite normalization of a fibre can be moved to functions on the original space. The argument first preserves finiteness under a small perturbation, and then realizes the finitely many coefficients on an affinoid neighbourhood of the base point.

Lemma 2.51.1. NAG-001674 (Small perturbations of a finite normalization). Let CC be a strict FF-affinoid algebra over a complete non-Archimedean field. Suppose that u1,…,ud∈Cu_1,\ldots,u_d\in C define a bounded finite homomorphism F⟨U1,…,Ud⟩→CF\langle U_1,\ldots,U_d\rangle\to C. If vi∈Cv_i\in C satisfy ρC(vi−ui)<1\rho_C(v_i-u_i)<1 for every ii, then Ui↦viU_i\mapsto v_i also defines a bounded finite homomorphism.

Proof. Boundedness of the original map gives ρC(ui)≤1\rho_C(u_i)\leq 1 by applying it to powers. Hence ρC(vi)≤1\rho_C(v_i)\leq 1. The pointwise power-boundedness theorem, Theorem 1.20.2, shows that the powers of each viv_i are bounded. The finitely many commuting elements therefore define the stated bounded Tate-algebra homomorphism. In the canonical reduction of CC, the images of viv_i and uiu_i coincide. Thus the two homomorphisms from F~[U1,…,Ud]\widetilde F[U_1,\ldots,U_d] into C~\widetilde C are identical. Theorem 3.16.5 applies to both maps: finiteness of the first reduction map proves finiteness of the second. ◻

Lemma 2.51.2. NAG-001675 (Approximating fibre functions by original coefficients). Let A→BA\to B be a bounded homomorphism of generalized KK-affinoid algebras, let x∈M(A)x\in\mathcal M(A), put F=H(x)F=\mathcal H(x) and C=B⊗^AFC=B\widehat\otimes_A F. For finitely many c1,…,cm∈Cc_1,\ldots,c_m\in C and ε>0\varepsilon>0, there are a rational neighbourhood UU of xx in M(A)\mathcal M(A) and elements b1,…,bm∈B⊗^AAUb_1,\ldots,b_m\in B\widehat\otimes_A A_U whose images in CC differ from the cic_i by norm less than ε\varepsilon. The domain UU may be chosen inside any prescribed ordinary neighbourhood of xx.

Proof. The image of B⊗AFB\otimes_A F is dense in its completed tensor product. Approximate each cic_i first by a finite sum ∑jβij⊗λij\sum_j\beta_{ij}\otimes\lambda_{ij}. By the definition of the completed residue field, Frac⁡(A/ker⁡x)\operatorname{Frac}(A/\ker x) is dense in FF. Approximate all the finitely many λij\lambda_{ij} by fractions aij(x)/dij(x)a_{ij}(x)/d_{ij}(x) with aij,dij∈Aa_{ij},d_{ij}\in A and dij(x)≠0d_{ij}(x)\ne 0. Their errors can be chosen small enough that the finite sums have error less than ε\varepsilon, since the completed tensor seminorm is bounded by the product of the two factor norms.

Choose positive numbers rij<∣dij(x)∣r_{ij}<|d_{ij}(x)|. The finitely many Laurent conditions ∣dij∣≥rij|d_{ij}|\geq r_{ij} define a rational neighbourhood of xx on which all denominators are invertible. Intersect with a rational neighbourhood inside the prescribed ordinary neighbourhood, using Lemma 2.41.1 and rational transitivity. On the resulting UU, set bi=∑jβijaij/dijb_i=\sum_j\beta_{ij}a_{ij}/d_{ij}. Completed base-change associativity identifies the fibre algebra over xx with the original CC, and the preceding estimates give the assertion. ◻

Proposition 2.51.3. NAG-001676 (Strict fibres near an inner point). Let f:Y→Xf:Y\to X be a morphism of good KK-analytic spaces, let x=f(y)x=f(y), and assume y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X) and ∣H(x)×∣≠{1}|\mathcal H(x)^\times|\ne\{1\}. There are affinoid neighbourhoods X0X_0 of xx and VV of yy, with V⊂f−1(X0)V\subset f^{-1}(X_0), such that VxV_x is strictly H(x)\mathcal H(x)-affinoid. The source neighbourhood may be chosen inside any prescribed ordinary neighbourhood of yy.

Proof. Choose affinoid neighbourhoods Y0→X0Y_0\to X_0 within the prescribed neighbourhoods. Proposition 2.44.6 makes yy an inner point of this affinoid morphism. There is therefore a norm-admissible relative presentation B=A{r−1T}/IB=A\{r^{-1}T\}/I with ∣Ti(y)∣<ri|T_i(y)|<r_i. Put F=H(x)F=\mathcal H(x). Since FF is nontrivially valued, ∣F×∣\sqrt{|F^\times|} is dense in R>0\mathbb R_{>0}: the rational powers of any fixed element of absolute value different from one already form a dense subgroup. Choose ∣Ti(y)∣<si<ri,si∈∣F×∣.|T_i(y)|<s_i<r_i,\qquad s_i\in\sqrt{|F^\times|}. The Weierstrass domain V={∣Ti∣≤si for all i}V=\{|T_i|\leq s_i\text{ for all }i\} contains an ordinary neighbourhood of yy. Relative presentations and base change identify its fibre with a quotient of F{s−1T}F\{s^{-1}T\}. It remains to exhibit this last algebra as strict.

Write siei=∣ai∣s_i^{e_i}=|a_i| with ei>0e_i>0 and ai∈F×a_i\in F^\times, and choose ci∈F×c_i\in F^\times with ∣ci∣≥si|c_i|\geq s_i. Then F{s−1T}≃F⟨U1,W1,…,Um,Wm⟩/(cieiUiei−aiWi)i=1m,Ti⟷ciUi.F\{s^{-1}T\}\simeq F\langle U_1,W_1,\ldots,U_m,W_m\rangle/ (c_i^{e_i}U_i^{e_i}-a_iW_i)_{i=1}^m, \quad T_i\longleftrightarrow c_iU_i. Indeed the map to the left sends UiU_i to Ti/ciT_i/c_i and WiW_i to Tiei/aiT_i^{e_i}/a_i, both power bounded by one. Conversely the relation gives a uniform bound for (ciUi)n/sin(c_iU_i)^n/s_i^n: divide nn by eie_i and use the finitely many possible remainders. The generalized Tate universal property therefore defines the map to the right. The maps are inverse on the polynomial generators and then on the completed algebras by density. This also proves norm admissibility. Quotients of a strict affinoid are strict, proving the assertion. ◻

Theorem 2.51.4. NAG-001677 (Finite projection from a strict fibre). Let Y=M(B)→X=M(A)Y=\mathcal M(B)\to X=\mathcal M(A) be an affinoid morphism, let y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X), and put x=f(y)x=f(y) and F=H(x)F=\mathcal H(x). Assume C=B⊗^AFC=B\widehat\otimes_A F is strict, and set d=dim⁡Cd=\dim C. After replacing XX by an affinoid neighbourhood of xx, there are dd analytic functions on YY defining an XX-morphism g:Y→AXd,ang:Y\to\mathbb A_X^{d,\mathrm{an}} which restricts to a finite morphism from an ordinary neighbourhood of yy to an ordinary neighbourhood of g(y)g(y). Consequently there are affinoid neighbourhoods VV of yy and TT of g(y)g(y), with TT an affinoid domain of AXd,an\mathbb A_X^{d,\mathrm{an}}, for which V→TV\to T is finite.

Proof. Theorem 1.12.2 and Proposition 3.21.1 provide a bounded finite injection F⟨U1,…,Ud⟩↪CF\langle U_1,\ldots,U_d\rangle\hookrightarrow C. Approximate its coordinate images, by Lemma 2.51.2, with images viv_i of functions on YU=Y×XUY_U=Y\times_X U, where UU is an affinoid neighbourhood of xx. Choose the errors of norm less than one. Lemma 2.51.1 makes the resulting homomorphism F⟨U⟩→CF\langle U\rangle\to C finite. Replace X,YX,Y by U,YUU,Y_U; these replacements are ordinary neighbourhoods of the points under consideration.

Let gg be the morphism defined by the chosen functions. Its fibre over g(y)g(y) is, near yy, a fibre of the finite FF-affinoid morphism just constructed, and is therefore a finite set. In particular yy is isolated in this fibre. To verify interior, choose an affinoid polydisc over XX containing g(Y)g(Y), with coordinate bounds strictly larger than the norms of the chosen functions. It gives a factorization Y→P→XY\to P\to X. The affinoid composition formula, Theorem 2.44.1, implies y∈Int⁡(Y/P)y\in\operatorname{Int}(Y/P) because y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X). Since g(y)g(y) is in the ordinary interior of PP in affine space, ordinary local testing gives y∈Int⁡(Y/AXd,an)y\in\operatorname{Int}(Y/\mathbb A_X^{d,\mathrm{an}}).

Theorem 2.49.1 now produces ordinary open neighbourhoods on which gg is finite. Choose an affinoid neighbourhood TT of g(y)g(y) inside the target open neighbourhood. Its inverse image VV is affinoid and finite over TT, and contains an ordinary neighbourhood of yy. This proves the final assertion. ◻

References. These coefficient and perturbation steps reconstruct the finite-projection mechanism underlying [Duc07] (Theorem 4.6 and Corollary 4.7), using internal strict normalization and the local finite criterion.

2.52 Generic fibre tests for universal injectivity

The proof of universal injectivity uses generic points of ordinary analytic open sets. Such points can be obtained from the finite spectral boundary and domain flatness; no dimension formula is needed for this step.

Lemma 2.52.1. NAG-001860 (Generic analytic points in ordinary open sets). Let CC be an integral generalized FF-affinoid algebra. Every nonempty ordinary open subset of M(C)\mathcal M(C) contains a point zz for which C→H(z)C\to\mathcal H(z) is injective. More generally every prime ideal of a generalized affinoid algebra is the kernel of an analytic point.

Proof. First let DD be any nonzero integral affinoid algebra. By Theorem 3.12.3, choose a finite spectral boundary S⊂M(D)S\subset\mathcal M(D). If every point of SS had nonzero kernel, choose a nonzero element in each kernel and take their product aa. The algebra DD is integral, so a≠0a\ne 0, whereas aa vanishes on the whole finite boundary. Thus ρD(a)=0\rho_D(a)=0, contradicting Proposition 3.21.4. At least one point has zero kernel. Applied to D=C/pD=C/\mathfrak p, this also proves the last assertion for every prime p\mathfrak p.

Now choose a nonempty rational domain V=M(B)V=\mathcal M(B) inside the given ordinary open set. The algebra BB is nonzero and flat over CC by Proposition 2.38.2. Let q\mathfrak q be a minimal prime of BB. Its contraction to CC is zero. Indeed, a nonzero a∈Ca\in C acts injectively on BB by flatness and hence on the nonzero ring BqB_{\mathfrak q}. If a∈qa\in\mathfrak q, its image in BqB_{\mathfrak q} is nilpotent, since the only prime ideal of that ring is its maximal ideal; a nilpotent element cannot act injectively on a nonzero ring. The preceding paragraph supplies a point of M(B/q)\mathcal M(B/\mathfrak q) with zero kernel. Its image in VV has kernel q\mathfrak q on BB and therefore zero kernel on CC, as required. ◻

Lemma 2.52.2. NAG-001861 (Regular affinoids and their components). Let Z=M(C)Z=\mathcal M(C) be a generalized affinoid space whose ordinary local rings are regular. Then CC is a regular Noetherian ring and a finite product of integral affinoid algebras. The corresponding analytic components are open and closed. In each component, the analytic points with kernel equal to its minimal prime are ordinarily dense.

Proof. For every prime p⊂C\mathfrak p\subset C, the preceding lemma supplies z∈Zz\in Z with kernel p\mathfrak p. The map Cp→OZ,zC_{\mathfrak p}\to\mathcal O_{Z,z} is faithfully flat: the ordinary stalk is a filtered colimit of flat rational-neighbourhood algebras by Proposition 2.33.16, and the map is local with maximal-ideal contraction p\mathfrak p. Both rings are Noetherian. The regularity descent theorem, [StacksAlg] (Tag 00OF), therefore makes CpC_{\mathfrak p} regular.

By [StacksAlg] (Tag 00NP), every such local ring is a domain. In particular CC is reduced, and its distinct minimal primes are pairwise comaximal: a maximal ideal containing two of them would yield a local ring with two distinct minimal primes. The Chinese remainder theorem and the finite number of minimal primes give C≃∏iC/piC\simeq\prod_i C/\mathfrak p_i. The idempotents of this product give precisely the open-and-closed analytic decomposition. Each factor is integral, and Lemma 2.52.1 proves the density assertion. ◻

Lemma 2.52.3. NAG-001862 (Fibrewise density after complete scalar extension). Let Z=M(C)Z=\mathcal M(C) be geometrically regular over a complete field FF, let S⊂ZS\subset Z be Zariski open, and suppose zz belongs to the ordinary closure of SS. For every complete extension E/FE/F, every point z′∈ZEz'\in Z_E above zz belongs to the ordinary closure of the inverse image SES_E.

Proof. Use the preceding lemma to select the integral factor C0C_0 containing zz. It is open and closed, so SS meets this factor. Write SS as the complement of the zero-locus of an ideal and choose, on C0C_0, a nonzero element aa of that ideal. Then D(a)⊂S∩M(C0)D(a)\subset S\cap\mathcal M(C_0). The homomorphism C0→(C0)EC_0\to(C_0)_E is flat by Theorem 2.33.6; multiplication by aa therefore remains injective. Geometric regularity and the preceding lemma make (C0)E(C_0)_E a product of integral affinoid algebras. The element aa is nonzero on each factor, since it acts injectively. The generic-point density lemma shows that D(a)D(a) is ordinarily dense in each factor, and hence in their union. The point z′z' belongs to this union, proving the assertion. ◻

Lemma 2.52.4. NAG-001863 (Lifting a fibrewise non-zero-divisor). Let (R,m)→(S,n)(R,\mathfrak m)\to(S,\mathfrak n) be a flat local homomorphism of Noetherian local rings. If multiplication by a∈Sa\in S is injective on S/mSS/\mathfrak mS, then it is injective on SS.

Proof. Flatness identifies the successive quotients with mjS/mj+1S≃(mj/mj+1)⊗R/m(S/mS).\mathfrak m^jS/\mathfrak m^{j+1}S \simeq (\mathfrak m^j/\mathfrak m^{j+1}) \otimes_{R/\mathfrak m}(S/\mathfrak mS). The first factor is a vector space, so multiplication by aa is injective on every displayed quotient. If ab=0ab=0, induction gives b∈mjSb\in\mathfrak m^jS for every j≥0j\geq 0. Since the map is local, these submodules lie in njS\mathfrak n^jS. Corollary 2.15.1, applied to the local Noetherian ring SS, gives b=0b=0. ◻

Theorem 2.52.5. NAG-001864 (Universal injectivity from a generic fibre isomorphism). Let T→XT\to X be quasi-smooth between good analytic spaces, let t∈Tt\in T and x=f(t)x=f(t), and let u:L→Nu:L\to N be a morphism of coherent sheaves with LL free of finite rank. Suppose that t∈Bij⁡(uTx)‾ Tx,t\in\overline{\operatorname{Bij}(u_{T_x})}^{\,T_x}, where closure means ordinary topological closure. Then the following are equivalent:

  1. uu is injective at every point over tt after every good analytic base change over every complete extension of the ground field;

  2. uu is injective at tt;

  3. t∈Bij⁡(u)∩Tx‾ Txt\in\overline{\operatorname{Bij}(u)\cap T_x}^{\,T_x}.

Proof. All assertions are ordinary-local near the points, so take T=M(B)T=\mathcal M(B) and X=M(A)X=\mathcal M(A) affinoid. Put F=H(x)F=\mathcal H(x) and C=B⊗^AFC=B\widehat\otimes_A F. By Theorem 2.35.14, the fibre is geometrically regular. Lemma 2.52.2 gives an integral component C0=C/qC_0=C/\mathfrak q containing tt. Choose an analytic point zz of this component whose kernel on CC is q\mathfrak q. If p,p′⊂B\mathfrak p,\mathfrak p'\subset B are the kernels of t,zt,z, then p′⊂p\mathfrak p'\subset\mathfrak p. The assumption says that uCu_C is an isomorphism at q\mathfrak q: a Zariski-open subset of this integral component is ordinarily dense exactly when it is nonempty, by Lemma 2.52.1.

The ordinary stalk at any analytic point is faithfully flat over the localization of its affinoid algebra at the point’s kernel, by the same rational-neighbourhood argument used in Lemma 2.52.2. Therefore injectivity and bijectivity of our finite-module maps can be tested at these algebraic localizations. Their bijectivity locus is Zariski open: its complement is the union of the supports of the finite kernel and cokernel.

Assume (2). The map uBpu_{B_{\mathfrak p}} is injective, hence so is uBp′u_{B_{\mathfrak p'}}. At zz the fibre map is bijective, so the cokernel of uu tensored with H(z)\mathcal H(z) vanishes. Nakayama’s lemma at Bp′B_{\mathfrak p'} implies that this cokernel is zero there. Thus uBp′u_{B_{\mathfrak p'}} is an isomorphism. Its Zariski-open bijectivity locus contains the generic point of the fibre component, hence is ordinarily dense in that component. It contains tt in its closure, proving (3).

For the converse, both the hypothesis and (3) remain valid after a complete scalar extension at every point above tt. Indeed Lemma 2.52.3 preserves the relevant fibre closures; scalar extension is flat on finite affinoid modules, so it identifies the fibre bijectivity locus with its pullback, and pullback of the total bijectivity locus is contained in the new one. Extend the ground field to F=H(x)F=\mathcal H(x), choose the canonical FF-rational lift of xx and a compatible lift of tt, and work there. Injectivity descends back by Theorem 2.35.4. We may therefore assume xx is rational.

Write m=ker⁡(A→F)\mathfrak m=\ker(A\to F). Then C=B/mBC=B/\mathfrak mB. The map Am→BpA_{\mathfrak m}\to B_{\mathfrak p} is flat. To see this directly from analytic flatness, the local ring OT,t\mathcal O_{T,t} is flat over OX,x\mathcal O_{X,x} and hence over AmA_{\mathfrak m}; it is faithfully flat over BpB_{\mathfrak p}. Tensoring kernels with this latter faithfully flat algebra descends the asserted flatness. The ring Bp/mBp=Cp/mBB_{\mathfrak p}/\mathfrak mB_{\mathfrak p}=C_{\mathfrak p/\mathfrak mB} is a regular local domain. Each element of Bp∖p′BpB_{\mathfrak p}\setminus\mathfrak p'B_{\mathfrak p} has nonzero image in that domain. Lemma 2.52.4 shows that it is a non-zero-divisor in BpB_{\mathfrak p}. Hence Bp⟶Bp′andLp⟶Lp′B_{\mathfrak p}\longrightarrow B_{\mathfrak p'} \quad\hbox{and}\quad L_{\mathfrak p}\longrightarrow L_{\mathfrak p'} are injective. By (3), up′u_{\mathfrak p'} is bijective. The commutative localization square then makes upu_{\mathfrak p} injective, and faithful flat comparison with the ordinary stalk proves (2).

Finally (1) implies (2) by the identity base change. Conversely assume (2), hence (3), and make any good analytic base change, over any complete extension. At an image x′x' the new fibre is the complete scalar extension of TxT_x to H(x′)\mathcal H(x'). Lemma 2.52.3 preserves the initial fibre hypothesis and the closure in (3); bijectivity on the total space is preserved by pullback. Quasi-smoothness is stable under this base change by Theorem 2.41.8. The implication (3)⇒\Rightarrow(2), already proved, therefore gives injectivity at each point above tt. This is (1). ◻

Corollary 2.52.6. NAG-001865 (Analytic flatness and the generic-fibre cokernel). In the setting of Theorem 2.52.5, put P=Coker⁡(u)P=\operatorname{Coker}(u). The sheaf NN is analytically XX-flat at tt if and only if uu is injective at tt and PP is analytically XX-flat at tt.

Proof. Suppose first that NN is analytically flat. Make an arbitrary good analytic base change over a complete field, and choose a point t′t' above tt, with image x′x'. The fibre hypothesis persists by Lemma 2.52.3. Enlarge the coefficient field once more to a nontrivially valued complete extension EE of H(t′)\mathcal H(t'), and choose the EE-rational point t′′t'' induced by the embedding H(t′)↪E\mathcal H(t')\hookrightarrow E. Its image x′′x'' is also EE-rational. If H(t′)\mathcal H(t') was trivially valued, a free-radius extension supplies such an EE.

Set R=OX′′,x′′R=\mathcal O_{X'',x''} and S=OT′′,t′′S=\mathcal O_{T'',t''}, and let m\mathfrak m be the maximal ideal of RR. By Lemma 2.35.12, the fibre stalk is S/mSS/\mathfrak mS. On its regular integral component the fibre map from the free sheaf LL is injective: its kernel is a submodule of a free module over a domain and becomes zero over that domain’s fraction field, because the fibre map is generically bijective. Thus the map of stalks Lt′′→Nt′′L_{t''}\to N_{t''} is injective modulo m\mathfrak m. The ring SS is Noetherian, and Nt′′N_{t''} is flat over RR by analytic flatness. Apply [StacksAlg] (Tag 00ME), with its finite source module equal to Lt′′L_{t''} and its flat target module equal to Nt′′N_{t''}. It gives both injectivity and RR-flatness of the cokernel Pt′′P_{t''}.

The two scalar-extension maps on the base and source ordinary local rings are faithfully flat by Theorem 2.35.4; coherent stalks pull back by ordinary tensor product. Injectivity therefore descends to t′t', and Lemma 2.35.6 descends the naive flatness of PP to t′t'. Since the original test and t′t' were arbitrary, PP is analytically flat at tt. Taking the identity test also proves that uu is injective there.

Conversely, suppose uu is injective and PP analytically flat at tt. Theorem 2.52.5 makes uu injective after every required base change at every point above tt. At such a point the short exact sequence 0→L→N→P→00\to L\to N\to P\to 0 has outer terms flat over the base local ring: this holds for PP by assumption, and for LL because it is free over the quasi-smooth, hence analytically flat, source. The ordinary-module argument in the proof of Theorem 2.35.14, applied to this sequence of modules over the base ordinary local ring, shows that the middle term is flat. This holds after every test, proving analytic flatness of NN. ◻

Definition 2.52.7. (An analytic dévissage). Let Y→XY\to X be a morphism of good spaces, let F\mathcal F be coherent on YY, and let y∈Supp⁡Fy\in\operatorname{Supp}\mathcal F map to xx. An XX-dévissage of F\mathcal F at yy consists of an affinoid ordinary neighbourhood VV of yy, a sequence n1>⋯>nr≥0n_1>\cdots>n_r\geq 0, and the following finite list of data. Put T0=VT_0=V, t0=yt_0=y and Q0=F∣VQ_0=\mathcal F|_V. For 1≤i≤r1\leq i\leq r one specifies:

  1. an affinoid domain TiT_i of a smooth XX-space of pure relative dimension nin_i, and ti∈Tit_i\in T_i above xx;

  2. a finite XX-morphism πi:Supp⁡(Qi−1)→Ti\pi_i:\operatorname{Supp}(Q_{i-1})\to T_i whose fibre over tit_i is the singleton {ti−1}\{t_{i-1}\};

  3. a finite free sheaf LiL_i and a morphism ui:Li→Ni=(πi)∗Qi−1u_i:L_i\to N_i=(\pi_i)_*Q_{i-1} with coherent cokernel QiQ_i, such that ti∈Bij⁡((ui)(Ti)x)‾ (Ti)xt_i\in\overline{\operatorname{Bij}((u_i)_{(T_i)_x})}^{\,(T_i)_x}.

One requires ti∈Supp⁡(Qi)t_i\in\operatorname{Supp}(Q_i) for i<ri<r and Qr=0Q_r=0. The support carries its closed analytic structure defined by the annihilator; Qi−1Q_{i-1} is regarded as a coherent sheaf on that closed subspace. All radii are allowed.

Theorem 2.52.8. NAG-001867 (Flatness along a given dévissage). For an XX-dévissage as above, F\mathcal F is analytically XX-flat at yy if and only if uiu_i is injective at tit_i for every 1≤i≤r1\leq i\leq r. When these conditions hold, every uiu_i is universally injective at tit_i in the sense of Theorem 2.52.5, and uru_r is an isomorphism at trt_r.

Proof. Restriction to the ordinary neighbourhood VV preserves all pointwise flatness tests. Regarding a sheaf as a sheaf on its closed support also preserves these tests: the closed immersion is finite, its point fibre is a singleton, and Theorem 2.35.17 applies. The same theorem, now applied to πi\pi_i, gives Qi−1 is analytically X-flat at ti−1⟺Ni is analytically X-flat at ti.Q_{i-1}\text{ is analytically }X\text{-flat at }t_{i-1} \quad\Longleftrightarrow\quad N_i\text{ is analytically }X\text{-flat at }t_i. The spaces TiT_i are quasi-smooth over XX by smoothness and domain invariance. Corollary 2.52.6 therefore identifies the condition on the right with the conjunction of injectivity of uiu_i at tit_i and analytic XX-flatness of QiQ_i there. Start with the zero sheaf QrQ_r, which is flat after every base change, and iterate these equivalences backwards. This proves the stated equivalence. The universal injectivity follows separately for each arrow from Theorem 2.52.5. Finally Qr=0Q_r=0 makes uru_r surjective; its injectivity gives the last assertion. ◻

References. The injectivity and cokernel results reconstruct [Duc18] (Propositions 8.1.8 and 8.1.10(1)). Generic-point density follows from the internal spectral boundary and flat affinoid domains. For the cokernel criterion, rational scalar lifting and the Stacks flat-map lemma give a direct proof of the universally flat formulation. The dévissage definition and its flatness test follow [Duc18] (8.2.4 and Theorem 8.3.4(2)).

2.53 Compactness and the separatedness of proper morphisms

Definition 2.53.1. NAG-001652 (Compact analytic spaces and morphisms). In the properness discussion, a compact analytic space or domain means a quasi-compact Hausdorff space; quasi-compact alone means the finite-subcover property. A morphism f:Y→Xf:Y\to X is compact if the inverse image of every compact analytic domain of XX is compact in this sense. A proper morphism is a compact morphism with empty relative analytic boundary.

This makes explicit the compactness convention in Section 2.7. The separatedness assertion there follows from compactness together with boundarylessness, by the theorem below.

Lemma 2.53.2. NAG-001653 (Closed immersions are target GG-local). Let h:W→Ph:W\to P be an analytic morphism and let (Pi)(P_i) be an admissible covering of PP by analytic domains. If every pullback hi:W×PPi→Pih_i:W\times_PP_i\to P_i is a closed immersion, then hh is a closed immersion.

Proof. The underlying map is injective: two points with the same image belong to a common inverse-image domain, where hih_i is injective. Write its image as S⊂PS\subset P. On each PiP_i the intersection S∩PiS\cap P_i is closed. Refine the covering by affinoid domains. The net topology, Lemma 2.40.2, tests a set’s closedness on this chart quasi-net, so SS is closed in PP. The same test proves that W→SW\to S is a homeomorphism. Indeed the pullbacks of the chart quasi-net form a quasi-net of WW, and intersecting it on the target with SS gives a quasi-net of SS. On every such pair the map is a homeomorphism.

Let Ii\mathcal I_i be the coherent ideal defining hih_i. Choose an affinoid chart U=M(A)U=\mathcal M(A) in PiP_i on which the inverse image is the actual quotient M(B)\mathcal M(B), where B=A/IB=A/I and I=Γ(U,Ii)I=\Gamma(U,\mathcal I_i). For any affinoid domain V=M(C)V=\mathcal M(C) of UU, the map A→CA\to C is a bounded domain restriction. By Corollary 1.18.4, II is finite and closed and the quotient norm on BB is the canonical finite-AA-module norm for the cyclic presentation A→BA\to B. With canonical finite-module norms on the terms, Proposition 2.38.2 restricts 0→I→A→B→00\to I\to A\to B\to 0 to a norm-strict exact sequence over CC. Its left map identifies I⊗ACI\otimes_A C with the ideal ICIC in CC, since its image consists of sums of products of images of elements of II with elements of CC. The canonical identification B⊗AC=C/ICB\otimes_A C=C/IC is the algebraic quotient identity. The same corollary over CC gives closedness of ICIC and the actual complete quotient norm; this is the canonical norm for the cyclic CC-module presentation. Applying Proposition 2.33.19 with L=KL=K identifies B⊗^ACB\widehat\otimes_A C with this actual quotient by a bounded isomorphism with bounded inverse. If VV or the quotient is empty, the corresponding zero algebra gives the same assertions directly.

The bounded pushout property of the completed affinoid tensor product, Proposition 2.33.8, represents the analytic inverse image over VV: on each affinoid chart of a test space the compatible maps are exactly the compatible bounded algebra maps, and these identifications agree on refinements and glue. This uses the given quotient chart over UU, not closed-immersion locality for the morphism being proved. On a common affinoid refinement of Pi∩PjP_i\cap P_j, the two restricted quotients thus represent the same inverse image over the refinement. Their canonical isomorphism over CC makes their kernels in CC equal. Consequently the ideals glue to a coherent ideal I⊂OPG\mathcal I\subset\mathcal O_{P_G}. One may construct this sheaf directly as compatible families of local sections, as in Theorem 2.40.6; finite presentation and coherence are local on the given admissible covering.

Fix an arbitrary affinoid domain Q=M(A)Q=\mathcal M(A) in PP. Theorem 2.39.6 identifies I∣Q\mathcal I|_Q with the finite ideal I=Γ(Q,I)⊂AI=\Gamma(Q,\mathcal I)\subset A. By Corollary 1.18.4, A/IA/I is an affinoid quotient with its actual quotient norm. The space M(A/I)\mathcal M(A/I) and W×PQW\times_PQ become the same closed subspace after an affinoid refinement subordinate to the PiP_i. Their quotient maps and their inverses agree on overlaps, so Theorem 2.40.8 glues those identifications. Hence hh is represented on every such QQ by the admissible quotient A↠A/IA\twoheadrightarrow A/I. Together with the closed topological embedding, this is a closed immersion. ◻

Lemma 2.53.3. NAG-001654 (The diagonal for a good Hausdorff source). If XX is affinoid and YY is good and Hausdorff, every analytic morphism f:Y→Xf:Y\to X is separated.

Proof. Put P=Y×XYP=Y\times_XY, with projections p1,p2p_1,p_2. The diagonal is a topological embedding onto its image because p1Δp_1\Delta is the identity. We describe its closed-subspace structure on an ordinary open covering of PP. Fix t∈Pt\in P and let yi=pi(t)y_i=p_i(t).

If y1≠y2y_1\ne y_2, choose disjoint ordinary open neighbourhoods N1,N2N_1,N_2 in the Hausdorff space YY. The open neighbourhood p1−1(N1)∩p2−1(N2)p_1^{-1}(N_1)\cap p_2^{-1}(N_2) of tt is disjoint from the diagonal image; there its closed-subspace ideal is the unit ideal.

If y1=y2=yy_1=y_2=y, goodness supplies an affinoid neighbourhood V=M(B)V=\mathcal M(B) of yy. Then V×XVV\times_XV is an affinoid domain of PP containing the open neighbourhood p1−1(Int⁡topV)∩p2−1(Int⁡topV)p_1^{-1}(\operatorname{Int}_{\rm top}V)\cap p_2^{-1}(\operatorname{Int}_{\rm top}V) of tt. For X=M(A)X=\mathcal M(A), the diagonal over this domain is exactly the multiplication quotient B⊗^AB⟶B.B\widehat\otimes_A B\longrightarrow B. It is bounded and has the bounded section b↦b⊗1b\mapsto b\otimes 1, so it is norm-admissible. Its kernel is a finite closed ideal by Theorem 1.18.2. The bounded pushout property identifies its spectrum with the diagonal, including the analytic points with their residue-field data; no identification of the analytic product with an ordinary topological product is used.

The two cases cover every point of PP by ordinary open neighbourhoods on which the diagonal is a closed subspace. Its image is consequently closed, and the ideals agree on overlaps since they are the kernels of the same restriction map. Lemma 2.53.2, applied to this ordinary open covering, proves that the full diagonal is a closed immersion. Thus ff is separated. ◻

Theorem 2.53.4. NAG-001655 (Proper analytic morphisms are separated). A compact boundaryless morphism of arbitrary analytic spaces is separated.

Proof. Choose an affinoid chart covering (Xi)(X_i) of the target and put Yi=f−1(Xi)Y_i=f^{-1}(X_i). Since each XiX_i is compact Hausdorff, compactness makes YiY_i compact Hausdorff. For each y∈Yiy\in Y_i, test the boundarylessness of ff on the actual affinoid domain XiX_i in Definition 2.44.4. It gives an affinoid neighbourhood of yy in YiY_i, so YiY_i is good. Lemma 2.53.3 therefore makes Yi→XiY_i\to X_i separated.

Let P=Y×XYP=Y\times_XY. The domains Pi=Yi×XiYiP_i=Y_i\times_{X_i}Y_i form an admissible covering of PP: they are the inverse images of the target chart covering under P→XP\to X, by Theorem 2.40.11 and analytic-domain base change. The pullback of the full diagonal to PiP_i is the closed diagonal of Yi→XiY_i\to X_i. Lemma 2.53.2 glues these closed immersions over the whole product, proving the assertion. ◻

References. The convention and conclusion follow the formulation of [TemIntro] (Definition 4.2.4.1 and Exercise 4.2.4.2(iv)). The proof is internal: boundarylessness provides good inverse images, Hausdorffness separates distinct projected points, and coherent quotient ideals glue over the analytic site.

Simple roots and finite étale germs

Lemma 2.54.1. NAG-001669 (A simple completed-residue root lifts analytically). Let XX be good, x∈Xx\in X, R=OX,xR=\mathcal O_{X,x}, and F=H(x)F=\mathcal H(x). If P∈R[T]P\in R[T] and α∈F\alpha\in F satisfy Px(α)=0P_x(\alpha)=0 and Px′(α)≠0P_x'(\alpha)\ne 0, there is a unique germ b∈Rb\in R with P(b)=0P(b)=0 and b(x)=αb(x)=\alpha. Consequently the residue field κ(x)\kappa(x) is separably algebraically closed in FF.

Proof. The image of RR in FF is dense. Indeed, on an affinoid neighborhood M(A)\mathcal M(A) it contains Frac⁡(A/ker⁡x)\operatorname{Frac}(A/\ker x): a denominator nonzero at xx is invertible on a rational ordinary neighborhood. This fraction field is dense in FF by the definition of the completed residue field. Choose a∈Ra\in R whose value is sufficiently close to α\alpha. Then P′(a)P'(a) is a unit of RR. Write P(a+Z)P′(a)=c0+Z+∑i=2dciZi.\frac{P(a+Z)}{P'(a)}=c_0+Z+\sum_{i=2}^{d}c_iZ^i. One can choose ρ>0\rho>0 and 0<q<10<q<1 so that ∣α−a(x)∣<ρ,∣c0(x)∣<qρ,∣ci(x)∣ρi−1<q(2≤i≤d).|\alpha-a(x)|<\rho,\qquad |c_0(x)|<q\rho, \qquad |c_i(x)|\rho^{i-1}<q\quad(2\leq i\leq d). To justify the simultaneous choice, first use the Taylor coefficients of PxP_x at its simple root α\alpha to choose a sufficiently small ρ\rho, and then approximate α\alpha by a(x)a(x) closely enough; all the finitely many normalized coefficients vary continuously and c0(x)c_0(x) tends to zero. Represent the coefficients on an ordinary affinoid neighborhood and impose the weak inequalities ∣c0∣≤qρ|c_0|\leq q\rho and ∣ci∣≤qρ1−i|c_i|\leq q\rho^{1-i}. They define a Weierstrass domain UU that is still an ordinary neighborhood of xx, because every inequality is strict at xx. In its defining quotient Banach norm the coefficients satisfy the same weak bounds: each one is the image of the corresponding scaled Weierstrass variable.

In B=O(U)B=\mathcal O(U) consider the closed ball ∥z∥≤ρ\|z\|\leq\rho and the map Φ(z)=−c0−∑i≥2cizi\Phi(z)=-c_0-\sum_{i\geq 2}c_iz^i. The displayed bounds give ∥Φ(z)∥≤qρ\|\Phi(z)\|\leq q\rho and ∥Φ(z)−Φ(w)∥≤q∥z−w∥.\|\Phi(z)-\Phi(w)\|\leq q\|z-w\|. Starting with z0=0z_0=0, the iterates satisfy ∥zn+1−zn∥≤qnρ\|z_{n+1}-z_n\|\leq q^n\rho, so completeness gives a unique fixed point zz in this ball. The element b=a+zb=a+z satisfies P(b)=0P(b)=0. Evaluation at xx commutes with the convergent iteration, and α−a(x)\alpha-a(x) is a fixed point in the same ball over FF; uniqueness gives b(x)=αb(x)=\alpha. If two germs b,b′b,b' have the stated properties, factor P(b)−P(b′)=(b−b′)Q(b,b′)P(b)-P(b')=(b-b')Q(b,b'). The value of Q(b,b′)Q(b,b') at xx is Px′(α)≠0P'_x(\alpha)\ne 0, so Q(b,b′)Q(b,b') is a unit and b=b′b=b'. Finally, lift the coefficients of the separable minimal polynomial of any element of FF algebraic separable over κ(x)\kappa(x) to RR and apply the first assertion. Its root is the value of a germ, hence belongs to κ(x)\kappa(x). ◻

Lemma 2.54.2. NAG-001670 (Separable extensions and dense coefficient fields). Let FF be a complete valued field and k0⊂Fk_0\subset F a dense subfield. Every finite separable extension E/FE/F has a presentation E≃F[T]/(Q)E\simeq F[T]/(Q) with Q∈k0[T]Q\in k_0[T] monic and separable.

Proof. By [StacksAlg] (Tag 030N), choose α\alpha with E=F(α)E=F(\alpha), and let P∈F[T]P\in F[T] be its monic minimal polynomial of degree dd. The powers 1,α,…,αd−11,\alpha,\ldots,\alpha^{d-1} form an FF-basis. By Lemma 1.12.1, its coordinate maximum norm is equivalent to the absolute value of EE. Hence, if β\beta is sufficiently close to α\alpha, the coordinate matrix of 1,β,…,βd−11,\beta,\ldots,\beta^{d-1} relative to this basis has the form I+MI+M with ∥M∥<1\|M\|<1. The convergent matrix series ∑j≥0(−M)j\sum_{j\geq 0}(-M)^j makes it invertible, so these powers are also a basis.

Approximate the coefficients of PP by elements of k0k_0, retaining the leading coefficient 11, to obtain QQ. Since P′(α)≠0P'(\alpha)\ne 0, sufficiently close approximation makes Q(α)Q(\alpha) arbitrarily small and keeps Q′(α)Q'(\alpha) nonzero. The contraction argument in the preceding lemma, now in the complete field EE and centered at α\alpha, produces a root β\beta of QQ as close to α\alpha as desired. The basis conclusion gives F(β)=EF(\beta)=E. The degree of QQ is dd, so F[T]/(Q)→EF[T]/(Q)\to E is an isomorphism. In particular QQ is irreducible and separable over FF, and therefore is separable over k0k_0. ◻

Theorem 2.54.3. NAG-001671 (Finite étale germs and completed residue fields). For a good analytic space XX and x∈Xx\in X, passage to the analytic fibre induces an equivalence between finite étale germs over (X,x)(X,x) and finite étale H(x)\mathcal H(x)-algebras, with arrows reversed on the algebra side. In particular, any finite separable extension of H(x)\mathcal H(x) is realized by a finite étale morphism over an ordinary neighborhood of xx, with exactly one point above xx.

Proof. Put R=OX,xR=\mathcal O_{X,x}, k0=κ(x)k_0=\kappa(x) and F=H(x)F=\mathcal H(x). Finite analytic maps near xx correspond to finite RR-algebras: existence is Theorem 2.35.20, and morphisms descend after shrinking because the algebras have finite presentations. Explicitly, represent the images of finitely many algebra generators on a common neighborhood and shrink again to make their finitely many relations hold. All the resulting maps are bounded by finite-module automatic boundedness. Equality of two maps is tested on the same finite list. Finite étaleness corresponds to finite flatness and the vanishing of relative differentials, by Proposition 2.49.2; these conditions spread from the germ by finite presentations and finite branch separation.

We describe maps explicitly. A finite étale RR-algebra is a finite product of local algebras by Corollary 2.35.21. A local factor CC is finite free over RR, and its residue algebra is a finite separable field over k0k_0 by Stacks 02GL(1). Choose a primitive residue element, using [StacksAlg] (Tag 030N), and lift it to c∈Cc\in C. If the residue degree is dd, the elements 1,c,…,cd−11,c,\ldots,c^{d-1} are an RR-basis: Nakayama gives a surjection from RdR^d, and flatness makes its finite kernel vanish after reduction and hence vanish. The expression of cdc^d in this basis gives C=R[T]/(P),P monic,P′(c)∈C×.C=R[T]/(P),\qquad P\ \text{monic}, \qquad P'(c)\in C^\times. For a second local factor DD, realized as an ordinary local ring OZ,z\mathcal O_{Z,z}, an FF-algebra map C⊗RF→D⊗RF=H(z)C\otimes_R F\to D\otimes_R F=\mathcal H(z) is determined by a simple root of PP in H(z)\mathcal H(z). The last equality follows from the finite étale fibre description in Stacks 02GL(1) and the fact that this local germ has a single point over xx; its finite reduced fibre algebra has one field factor. Lemma 2.54.1, applied on ZZ, lifts this root uniquely to DD. Thus it lifts the given homomorphism uniquely. Products cause no difficulty: an algebra map into a local factor selects exactly one source factor, as its orthogonal idempotents must map to zero or one. This proves full faithfulness.

For essential surjectivity it suffices to realize a finite separable field E/FE/F. The preceding lemma supplies a monic Q∈k0[T]Q\in k_0[T] with F[T]/(Q)=EF[T]/(Q)=E. Lift its finitely many coefficients to RR, obtaining Q^∈R[T]\widehat Q\in R[T], and set C=R[T]/(Q^)C=R[T]/(\widehat Q). The algebra CC is finite free. Its derivative is a unit: modulo the maximal ideal of RR, the coprime pair Q,Q′Q,Q' has a polynomial Bezout identity; lifting this identity shows that multiplication by Q^′\widehat Q' is surjective modulo that maximal ideal, hence surjective on the finite module CC by Nakayama. Thus CC is finite étale. Realize it on an ordinary neighborhood by the first paragraph. Its fibre is exactly EE, so there is one point above xx with completed residue field EE. Finite products realize all finite étale algebras. ◻

Corollary 2.54.4. NAG-001672 (An étale map with unchanged completed residue field). Let f:Y→Xf:Y\to X be étale between good spaces, and let y↦xy\mapsto x. If H(x)→H(y)\mathcal H(x)\to\mathcal H(y) is an isomorphism, then ff induces an isomorphism between ordinary neighborhoods of yy and xx.

Proof. An étale morphism is locally finite at yy by Proposition 2.49.2. Separate the branch at yy by Lemma 2.35.1. Its finite étale germ has fibre H(x)\mathcal H(x), so Theorem 2.54.3 identifies it with the identity germ. The inverse morphism and the two inverse equalities involve finitely many affinoid algebra generators; representing them and shrinking the ordinary neighborhoods makes these equalities hold as analytic morphisms. ◻

References. The construction reconstructs the good-space case of [Ber93] (Theorem 3.4.1). The simple-root argument supplies the valued-field step explicitly; finite-dimensional basis stability replaces a separate invocation of Krasner’s lemma in the realization of finite separable extensions.

2.55 Product neighborhoods in smooth families

Lemma 2.55.1. NAG-001999 (Product neighborhoods of free Gauss points). Let XX be good, x∈Xx\in X, and let r=(r1,…,rm)r=(r_1,\ldots,r_m) be H(x)\mathcal H(x)-free, with 0≤m≤n0\leq m\leq n. In AXn,an\mathbb A_X^{n,\mathrm{an}} let ξ\xi be the fibre point that is Gauss of radii rr in the first mm coordinates and has its remaining coordinates zero. Every ordinary neighborhood WW of ξ\xi contains a product U×KD1×K⋯×KDnU\times_K D_1\times_K\cdots\times_K D_n that contains ξ\xi, where UU is an ordinary neighborhood of xx, the DiD_i for i≤mi\leq m are open annuli around radius rir_i, and the remaining DiD_i are open discs centered at zero.

Proof. Take an ordinary affinoid neighborhood X0=M(A)X_0=\mathcal M(A) of xx. The topology on relative affine space is generated locally by finitely many conditions ∣Pj∣∈Ij|P_j|\in I_j with Pj∈A[T1,…,Tn]P_j\in A[T_1,\ldots,T_n] and open intervals Ij⊂R≥0I_j\subset\mathbb R_{\geq 0}. This follows directly on each closed polydisc from polynomial density in its Tate algebra: approximate a finite list of functions sufficiently closely to preserve the strict open inequalities at the point. Choose such conditions defining a neighborhood contained in WW.

Write T=(T′,T′′)T=(T',T''), separating the first mm coordinates. In each specialization Pj(T′,0)P_j(T',0), freeness makes the values of distinct nonzero monomials at ξ\xi distinct. If the specialization is nonzero at ξ\xi, it therefore has a unique dominant monomial. The finitely many strict inequalities expressing its dominance, and the condition that its value lies in IjI_j, persist when the coefficients vary over a sufficiently small ordinary neighborhood of xx and the values ∣Ti∣|T_i| vary in sufficiently small intervals around rir_i. Every term divisible by a coordinate of T′′T'' can simultaneously be made strictly smaller by taking those coordinates in sufficiently small zero-centered discs; there are only finitely many such terms. If the specialization vanishes at ξ\xi, all its coefficients have value zero at xx; bound them on a small ordinary base neighborhood and make every term have value in a sufficiently small interval [0,ε)[0,\varepsilon) contained in IjI_j. Intersecting these finitely many coefficient neighborhoods and choosing the annulus and disc radii proves the assertion. ◻

Proposition 2.55.2. NAG-002000 (A product patch from a rational fibre point). Let Y→XY\to X be smooth of pure relative dimension nn between good spaces. If the fibre over xx has a point yy with H(y)=H(x)\mathcal H(y)=\mathcal H(x), then YY has a nonempty ordinary open subset, over an ordinary neighborhood UU of xx, that is XX-isomorphic to U×KDU\times_K D, where DD is an nn-dimensional open polydisc. The subset can be chosen inside any prescribed ordinary neighborhood of yy, but need not contain yy.

Proof. Restrict to the prescribed neighborhood and take an étale coordinate map Y→AXn,anY\to\mathbb A_X^{n,\mathrm{an}} near yy, as supplied by Theorem 2.50.3. If zz is the image of yy, the inclusions H(x)⊂H(z)⊂H(y)\mathcal H(x)\subset\mathcal H(z)\subset\mathcal H(y) force equality. Corollary 2.54.4 therefore identifies an ordinary neighborhood of yy with an ordinary open set WW in relative affine space. The fibre WxW_x contains a point with coordinates in F=H(x)F=\mathcal H(x). Since κ(x)\kappa(x) is dense in FF, there is a point z′z' in this same fibre open set all of whose coordinates belong to κ(x)\kappa(x): only finitely many strict polynomial-value conditions need to be preserved. Represent these coordinates by germs on XX, shrink to an ordinary neighborhood where they are defined, and translate the corresponding relative affine coordinates. The translated open set contains the fibre origin. Lemma 2.55.1, with m=0m=0, now supplies the desired product. Translating back and using the étale local isomorphism completes the proof. ◻

Lemma 2.55.3. NAG-002001 (Separable rigid points in smooth spaces). Over a nontrivially valued complete field FF, every nonempty ordinary open subset of a good smooth FF-analytic space contains a point whose completed residue field is finite separable over FF.

Proof. First consider a nonempty ordinary open set in AFn,an\mathbb A_F^{n,\mathrm{an}}. Finite polynomial-value inequalities describe a smaller nonempty open neighborhood of any chosen point. The group ∣F×∣\sqrt{|F^\times|} is dense in R>0\mathbb R_{>0}: powers with rational exponents of any fixed value different from one already are dense. Choose all upper and lower bounds for these inequalities, and the ambient polydisc radii, in this group, strictly outside the values at the point. This produces a nonempty affinoid domain inside the open set. After a finite extension realizing the finitely many radii as absolute values, its presentation is strict. A maximal ideal of this strict algebra gives, by Theorem 1.12.3, a point whose coordinates are algebraic over FF. Surjectivity of complete scalar extension ensures that the chosen domain remains nonempty.

Separable algebraic elements are dense among algebraic elements over FF. To see the only nontrivial case, let aa have inseparable monic minimal polynomial PP of degree dd. Then P′=0P'=0. For arbitrarily small ε∈F×\varepsilon\in F^\times, the polynomial P(T)+εTP(T)+\varepsilon T is separable, as its derivative is the nonzero constant ε\varepsilon. In a finite splitting extension write its roots as b1,…,bdb_1,\ldots,b_d. Evaluation at aa gives ∏i∣a−bi∣=∣εa∣\prod_i|a-b_i|=|\varepsilon a|, so one root approaches aa as ε→0\varepsilon\to 0; the case a=0a=0 is already separable. Lemma 1.12.1 supplies compatible absolute values in the finite splitting extensions. Approximating the finitely many coordinates simultaneously preserves the finitely many strict polynomial inequalities. Their joint field is finite separable over FF. This proves the assertion for affine space.

Now take an étale coordinate chart of the given smooth open set. An étale map is ordinary-open. Indeed, near any point it is finite flat; separate the finite fibre into branches over a small ordinary target neighborhood. A source ordinary neighborhood contains the relevant branch after a further target shrinking, by compactness and the closed-map property. The finite flat branch has open-and-closed image by Lemma 2.44.2, so the image contains a neighborhood of the image point. Apply the affine-space assertion to this nonempty open image. Over the resulting finite separable point the étale fibre is a nonempty finite union of finite separable fields. Any point of it gives a finite separable extension of FF, as required. ◻

Theorem 2.55.4. NAG-002002 (Product patches after a finite étale base change). Let Y→XY\to X be smooth of pure relative dimension nn between good spaces, let x∈Xx\in X, and suppose H(x)\mathcal H(x) is nontrivially valued. For every nonempty ordinary open subset W⊂YxW\subset Y_x, there exist a finite étale morphism X′→UX'\to U over an ordinary neighborhood UU of xx, its unique point x′x' above xx, and a nonempty ordinary open subset V⊂Y×XX′V\subset Y\times_X X' such that Vx′V_{x'} maps into WW and V≃U′×KDV\simeq U'\times_K D over X′X', for an ordinary neighborhood U′U' of x′x' and an nn-dimensional open polydisc DD.

Proof. The preceding lemma gives y∈Wy\in W with E=H(y)E=\mathcal H(y) finite separable over F=H(x)F=\mathcal H(x). By Theorem 2.54.3, realize E/FE/F by a finite étale morphism X′→UX'\to U with one point x′x' above xx and H(x′)=E\mathcal H(x')=E. In the fibre product the evaluation at yy gives an EE-rational point y′y' over x′x'. The ordinary topology of an analytic fibre is the subspace topology by Theorem 2.33.11 and its gluing form. Hence one can choose an ordinary neighborhood of y′y' whose intersection with the fibre maps into WW. Restrict to that neighborhood and apply Proposition 2.55.2. Its product patch has precisely the required properties. ◻

References. These proofs reconstruct [Duc18] (Lemma 6.2.1, the rational-point case of Lemma 6.2.2, and the nontrivially valued completed-residue-field case of Lemma 6.2.4).

Theorem 2.55.5. NAG-002003 (A product patch from a free Gauss residue field). Let Y→XY\to X be smooth of pure relative dimension nn between good analytic spaces over any complete field KK. Suppose that y∈Yxy\in Y_x has an isometric H(x)\mathcal H(x)-identification H(y)=H(x)r\mathcal H(y)=\mathcal H(x)_r, where r=(r1,…,rm)r=(r_1,\ldots,r_m) is free and 0≤m≤n0\leq m\leq n. Inside any prescribed ordinary neighborhood of yy there is a nonempty ordinary open subset which, over XX, is isomorphic to U×KD×KΔ,U\times_K D\times_K\Delta, where UU is an ordinary neighborhood of xx, DD is an mm-dimensional open polyannulus, and Δ\Delta is an (n−m)(n-m)-dimensional open polydisc.

Proof. Restrict to the prescribed neighborhood. The field κ(y)\kappa(y) is dense in H(y)\mathcal H(y), so each free coordinate value of norm rir_i can be approximated by an element of κ(y)\kappa(y) closely enough to preserve its norm. Represent these elements by analytic germs fif_i near yy; a denominator with nonzero value is a unit on a smaller ordinary neighborhood. Proposition 3.24.5, applied on the analytic fibre, makes (df1)(y),…,(dfm)(y)(df_1)(y),\ldots,(df_m)(y) independent in the relative differential fibre. Fibre base change of analytic differentials identifies this with the corresponding subspace of ΩY/Xan⊗H(y)\Omega^{\rm an}_{Y/X}\otimes\mathcal H(y). Since the latter has dimension nn and is generated by differentials of local analytic functions, add fm+1,…,fnf_{m+1},\ldots,f_n so that their differentials form a basis.

The map g=(f1,…,fn):Y→AXn,ang=(f_1,\ldots,f_n):Y\to\mathbb A_X^{n,\mathrm{an}} is quasi-smooth of relative dimension zero at yy. Explicitly, take an ordinary Jacobian presentation for Y/XY/X, lift the finitely many fif_i to its ambient affinoid algebra using the surjective closed-immersion presentation, and adjoin the equations Zi−fiZ_i-f_i for the graph of gg. Independence of the displayed differential basis says exactly that the original relation rows together with these nn new rows have full rank; Theorem 2.35.14 gives the assertion. Because Y→XY\to X is inner at yy, the interior composition formula makes gg inner there. Its fibre through yy is isolated by Lemma 2.50.1; it is ordinarily flat by the Jacobian theorem and its relative differential module vanishes. Proposition 2.49.2 therefore makes gg étale near yy.

Put z=g(y)z=g(y). The completed field H(z)\mathcal H(z) contains H(x)\mathcal H(x) and the first mm coordinate values. Lemma 3.24.4 says that these generate all of H(y)\mathcal H(y) as a complete field, so H(z)=H(y)\mathcal H(z)=\mathcal H(y). Corollary 2.54.4 identifies an ordinary neighborhood of yy with an ordinary open subset W⊂AXn,anW\subset\mathbb A_X^{n,\mathrm{an}}. Let pp be projection onto the first mm coordinates and put a=p(z)a=p(z). Freeness implies that aa is the Gauss point of polyradius rr over xx, and zz is rational over H(a)\mathcal H(a).

The fibre open set WaW_a contains a rational point whose last n−mn-m coordinates lie in the dense subfield κ(a)⊂H(a)\kappa(a)\subset\mathcal H(a), by preserving the finitely many polynomial-value conditions of a neighborhood of zz. Extend these coordinates to germs on the good space AXm,an\mathbb A_X^{m,\mathrm{an}} and translate the last coordinates over an ordinary neighborhood of aa. The translated WW contains the relative fibre origin. Lemma 2.55.1 with no free coordinates, applied over this new base, supplies U1×KΔ⊂WU_1\times_K\Delta\subset W with U1U_1 an ordinary neighborhood of aa. Apply the same lemma to AXm,an→X\mathbb A_X^{m,\mathrm{an}}\to X at the free Gauss point aa: inside U1U_1 it gives U×KDU\times_KD. Restrict the first product to this second one, and undo the coordinate translation and the local étale isomorphism. The result is the stated product patch. ◻

References. The last theorem reconstructs the full free-polyradius case of [Duc18] (Lemma 6.2.2); its valued-field and differential inputs are proved in Section 3.24.

Lemma 2.55.6. NAG-002004 (Lifting a finite extension of a trivial residue field). Let XX be good, let x∈Xx\in X have trivially valued F=H(x)F=\mathcal H(x), and let E/FE/F be finite. There are an ordinary neighborhood UU of xx and a finite flat morphism X′→UX'\to U whose fibre at xx is M(E)\mathcal M(E). In particular it has a unique point x′x' over xx, with H(x′)=E\mathcal H(x')=E, and the closed fibre of Spec⁡OX′,x′→Spec⁡OX,x\operatorname{Spec}\mathcal O_{X',x'}\to\operatorname{Spec}\mathcal O_{X,x} is reduced. If E/FE/F is separable, the morphism may be chosen finite étale.

Proof. Put R=OX,xR=\mathcal O_{X,x} and let m\mathfrak m be its maximal ideal. The injection κ(x)↪F\kappa(x)\hookrightarrow F has dense image. As FF is trivially valued and hence discrete, it is surjective, so R/m=FR/\mathfrak m=F. Write the finite extension as a tower generated by finitely many elements. At each step lift the coefficients of the monic minimal polynomial from the preceding residue field to the preceding ring, starting with RR. Successive monic quotients give a finite free RR-algebra CC with C/mC=EC/\mathfrak mC=E. Every maximal ideal of the integral finite algebra CC lies over m\mathfrak m, and this residue quotient is a field, so CC is local.

Theorem 2.35.20 realizes CC by a finite analytic map over an ordinary neighborhood of xx. Its ordinary direct-image stalk is CC and its analytic fibre algebra is C⊗RF=EC\otimes_R F=E. To obtain flatness on a neighborhood, lift a basis of the free RR-module CC to finitely many sections of the finite direct-image module. The resulting map from a free coherent module is an isomorphism at xx. Its coherent kernel and cokernel vanish on a common ordinary neighborhood, since finitely many local generators vanish after one shrink. On that neighborhood the direct-image algebra is finite free. The finite-module base-change theorem then gives finite flatness, and the residue quotient proves the closed-fibre assertion. If E/FE/F is separable, one may instead use Theorem 2.54.3, which gives the required finite étale realization directly. ◻

Theorem 2.55.7. NAG-002005 (Product patches over arbitrary completed residue fields). Let Y→XY\to X be smooth of pure relative dimension nn between good analytic spaces over any complete field. Given x∈Xx\in X and a nonempty ordinary open subset W⊂YxW\subset Y_x, there are an ordinary neighborhood UU of xx, a finite flat map X′→UX'\to U with unique point x′x' above xx and reduced ordinary closed fibre there, and a nonempty ordinary open subset V⊂Y×XX′V\subset Y\times_X X' such that Vx′V_{x'} maps into WW and V≃U′×KD×KΔV\simeq U'\times_K D\times_K\Delta over X′X'. Here U′U' is an ordinary neighborhood of x′x', DD is either absent or a one-dimensional open annulus, and Δ\Delta is an open polydisc making the total relative dimension nn. If H(x)\mathcal H(x) is nontrivially valued, DD may be omitted and X′→UX'\to U may be chosen étale. If the ground field has characteristic zero, the map may also be chosen étale when H(x)\mathcal H(x) is trivially valued.

Proof. The nontrivially valued case is Theorem 2.55.4. Suppose that F=H(x)F=\mathcal H(x) is trivially valued. Proposition 3.24.9, applied to the smooth open fibre subset WW, gives a finite extension E/FE/F and a point z∈WEz\in W_E whose completed residue field is EE or ErE_r for some 0<r<10<r<1. Lemma 2.55.6 realizes EE by a finite flat X′→UX'\to U with the prescribed unique point and reduced closed fibre. Its fibre product with YY has analytic fibre Yx⊗^FEY_x\widehat\otimes_F E, by associativity of complete base change, so it contains the actual point zz.

Choose an ordinary neighborhood of zz in the fibre product whose intersection with the fibre is contained in WEW_E, using the subspace topology of analytic fibres. In the rational case apply Proposition 2.55.2. In the other case rr is free over the trivially valued EE; apply Theorem 2.55.5 with m=1m=1. Independence of the free differential also implies n≥1n\geq 1 in this case. These applications give exactly the stated ordinary product patches. In characteristic zero the finite extension E/FE/F is separable, so the last clause of the lifting lemma makes the base change étale. ◻

References. The finite-flat lifting and the rational-or-Gauss point construction complete the product-patch mechanism of [Duc18] (Proposition 6.2.4) for arbitrary completed residue fields.

2.56 Ordinary germs along a generic analytic fibre

Lemma 2.56.1. NAG-001849 (Normal affinoid components and the identity principle). Let X=M(A)X=\mathcal M(A) be a generalized affinoid space whose ordinary local rings are normal. Then AA is a finite product of normal domains. The spectra of these factors are exactly the connected components of XX, and they are open and closed. On each component, an analytic function that vanishes on a nonempty ordinary open set is zero.

Proof. For each prime p⊂A\mathfrak p\subset A, Lemma 2.52.1 gives an analytic point with kernel p\mathfrak p. Its ordinary local ring RR is faithfully flat over ApA_{\mathfrak p}, by rational-domain flatness and the ordinary-neighborhood colimit. Since RR is a normal local ring, it is a domain. Faithful flatness makes Ap→RA_{\mathfrak p}\to R injective, so ApA_{\mathfrak p} is also a domain. If u/vu/v in its fraction field is integral, normality of RR gives u∈vRu\in vR; faithful flatness contracts vRvR to vApvA_{\mathfrak p}, proving that u/v∈Apu/v\in A_{\mathfrak p}. Thus every ApA_{\mathfrak p} is normal.

The Noetherian ring AA is reduced, and distinct minimal primes are pairwise comaximal: otherwise their localizations at a maximal ideal containing both would contradict the domain property of that local ring. The Chinese remainder theorem gives a finite product of normal domains. Its idempotents supply open-and-closed affinoid factors. An integral factor has connected analytic spectrum: a partition into two nonempty open-and-closed sets would glue the functions zero and one to a nontrivial idempotent in its global-section ring, which is that integral factor by the structure-sheaf theorem. Finally, a nonempty ordinary open subset of an integral affinoid contains a kernel-zero point by Lemma 2.52.1. A function vanishing there is therefore zero in the algebra. ◻

Proposition 2.56.2. NAG-001850 (Spreading a closed fibre neighborhood in characteristic zero). Let KK be complete of characteristic zero, let Y→XY\to X be a morphism of good KK-analytic spaces smooth at yy, and put x=f(y)x=f(y). Suppose that OX,x\mathcal O_{X,x} is Artinian and H(x)\mathcal H(x) is nontrivially valued. If a Zariski-closed subset Z⊂YZ\subset Y contains an ordinary neighborhood of yy in YxY_x, then ZZ contains an ordinary neighborhood of yy in YY.

Proof. The assertion concerns underlying sets. We first reduce the base near xx. On an ordinary affinoid neighborhood M(A)\mathcal M(A) put p=ker⁡(x)\mathfrak p=\ker(x), D=ApD=A_{\mathfrak p} and R=OX,xR=\mathcal O_{X,x}. The map D→RD\to R is faithfully flat, so its map on spectra is surjective. Since RR is local Artinian, DD has only one prime and is zero-dimensional. Its reduction is therefore a field: its unique prime is the nilradical, which is nilpotent because the ring is Noetherian.

Let N⊂AN\subset A be the nilradical. The ordinary local ring of XredX_{\mathrm{red}} at xx is R/NRR/NR, by coherent closed-subspace restriction. The regular homomorphism D→RD\to R of Theorem 2.35.29, after quotienting by the nilradical of DD, shows that R/NRR/NR is geometrically regular over the field DredD_{\mathrm{red}}. It is also local Artinian, being a quotient of RR. A zero-dimensional regular local ring is a field, since its maximal ideal has zero-dimensional cotangent space and Nakayama makes that ideal zero. The base change Y×XXred→YY\times_X X_{\mathrm{red}}\to Y is a homeomorphism on underlying spaces, since it is cut out by a nilpotent ideal. It preserves smoothness and the analytic fibre. Thus we may replace the base by its reduction and assume RR is a field.

By Corollary 2.35.32, shrink to an ordinary neighborhood of xx on which XX is normal. Smoothness and the same corollary make YY normal near yy. Choose a normal affinoid ordinary neighborhood Y0=M(B)Y_0=\mathcal M(B) of yy and select its component containing yy. Lemma 2.56.1 makes its algebra a normal domain. Write Z∩Y0Z\cap Y_0 as the common zero set of finitely many functions h1,…,hs∈Bh_1,\ldots,h_s\in B. It suffices to show that these functions vanish on some nonempty ordinary open subset of Y0Y_0, for the identity principle will then make them zero on all of Y0Y_0.

Work in an ordinary open neighborhood of yy inside Y0Y_0, small enough that the morphism is smooth of constant relative dimension there. The hypothesis supplies a nonempty fibre open subset contained in ZZ. Theorem 2.55.4 gives a finite étale map X′→UX'\to U near xx, one point x′x' above xx, and a nonempty ordinary open product patch in Y′=Y×XX′Y'=Y\times_X X' whose fibre at x′x' lies in Z′Z'. Its base is an ordinary neighborhood of x′x' and its other factor is an open polydisc DD. The local ring R′=OX′,x′R'=\mathcal O_{X',x'} is a field: finite branch separation identifies it with a local factor of a finite étale algebra over the field RR. The space X′X' is normal near x′x' by quasi-smooth ascent.

Inside the base of the product patch choose a normal affinoid ordinary neighborhood X0′=M(A′)X'_0=\mathcal M(A') of x′x' and retain its component through x′x'. Then A′A' is a domain, and A′→R′↪H(x′)A'\to R'\hookrightarrow\mathcal H(x') is injective. Indeed A′→R′A'\to R' is a nonzero flat map from a domain, so multiplication by every nonzero element remains injective. Restrict the pulled-back hjh_j to X0′×KDX'_0\times_KD. Exhausting DD by nested closed polydiscs identifies each such function with a power series hj=∑IaI,jTI,aI,j∈A′.h_j=\sum_I a_{I,j}T^I,\qquad a_{I,j}\in A'. The coefficients agree on the exhaustion because Tate-series expansions are unique, and restrictions to smaller polydiscs preserve their coefficients. On the fibre at x′x' this function vanishes at every point. In particular it vanishes at every Gauss point of a sufficiently small closed polydisc in DD, so every coefficient aI,j(x′)a_{I,j}(x') is zero by the Gauss norm formula. Injectivity of A′→H(x′)A'\to\mathcal H(x') gives aI,j=0a_{I,j}=0 for all I,jI,j. Thus every hjh_j vanishes on X0′×KDX'_0\times_KD, which contains a nonempty ordinary open subset of the original product patch because X0′X'_0 is an ordinary neighborhood of x′x'.

The map Y′→YY'\to Y is étale over the chosen base neighborhood and is ordinary-open by the finite-branch argument in Lemma 2.55.3. Hence the image of that open subset is a nonempty ordinary open subset of Y0Y_0 contained in ZZ. The identity principle proves Z⊃Y0Z\supset Y_0. Since Y0Y_0 was an ordinary neighborhood of yy, this is the required assertion. ◻

References. The proposition reconstructs [Duc18] (Proposition 6.3.1) in characteristic zero with nontrivially valued completed residue field, using the internally established normal-locus and product-patch results. Its conclusion is topological containment of a Zariski-closed subset, not equality of nonreduced closed analytic subspaces.

Lemma 2.56.3. NAG-001851 (Dimension of an affinoid domain in affine space). Let FF be a complete non-Archimedean field and let V=M(C)V=\mathcal M(C) be a generalized affinoid domain of AFd,an\mathbb A_F^{d,\mathrm{an}}. Then dim⁡C≤d\dim C\leq d.

Proof. Fix a maximal ideal q\mathfrak q of CC. Lemma 2.52.1 realizes it as the kernel of a point z∈Vz\in V. Choose a nontrivially valued complete extension E/H(z)E/\mathcal H(z), using E=H(z)E=\mathcal H(z) when possible and otherwise adjoining a free radius. Evaluation defines an EE-rational point w∈VEw\in V_E above zz. Rational-domain flatness and scalar-stalk faithful flatness give faithfully flat local maps Cq⟶OV,z⟶OVE,w.C_{\mathfrak q}\longrightarrow\mathcal O_{V,z} \longrightarrow\mathcal O_{V_E,w}. Since VEV_E is an analytic domain of affine dd-space and ww is rational, Lemma 2.35.9 identifies the last ring with the ordinary affine-space germ at ww. After translating the coordinates it is the ring of convergent power series in dd variables of Lemma 2.35.8; its completion is E[[T1,…,Td]]E[[T_1,\ldots,T_d]]. It is a regular local ring of dimension dd: regularity follows by faithful-flat descent from that formal power-series ring, and its maximal ideal and cotangent space are generated freely to first order by the dd coordinate germs. Finally, a chain of primes of CqC_{\mathfrak q} lifts, successively downward from the maximal ideal, along the displayed flat local maps by [StacksAlg] (Tag 00HS). Its length is therefore at most dd. This holds at every maximal ideal of CC, and every prime chain is contained in one, proving the assertion. ◻

Lemma 2.56.4. NAG-001852 (Cutting out a fibre component over the original field). Let Y=M(B)→X=M(A)Y=\mathcal M(B)\to X=\mathcal M(A) be an affinoid morphism, let x∈Xx\in X, and set F=H(x)F=\mathcal H(x) and C=B⊗^AFC=B\widehat\otimes_A F. If e∈Ce\in C is idempotent and y∈Yxy\in Y_x satisfies e(y)=1e(y)=1, then, after replacing XX by an ordinary affinoid neighborhood of xx, a Weierstrass domain WW of the resulting YY is an ordinary neighborhood of yy and satisfies Wx=M(eC)W_x=\mathcal M(eC). In particular a strict fibre remains strict after this operation.

Proof. Choose 0<δ<ε<10<\delta<\varepsilon<1. Lemma 2.51.2 supplies, after the stated base restriction, a function bb on YY whose image in CC differs from ee by norm less than δ\delta. Define WW by ∣b−1∣≤ε|b-1|\leq\varepsilon. At yy the inequality is strict, so WW is an ordinary neighborhood there. On M(eC)\mathcal M(eC) one has ∣b−1∣<δ|b-1|<\delta, whereas on M((1−e)C)\mathcal M((1-e)C) one has ∣b−1∣=1|b-1|=1. Thus its fibre has exactly the first factor as underlying domain. Its domain algebra is also that factor: on eCeC the Weierstrass condition is automatic by the norm bound, while on (1−e)C(1-e)C the element b−1b-1 has inverse of spectral norm one, so the condition ∣b−1∣≤ε<1|b-1|\leq\varepsilon<1 gives the zero algebra. Equivalently, the affinoid-domain universal property identifies WxW_x with the open-and-closed affinoid factor M(eC)\mathcal M(eC). This factor is a quotient of CC, so it is strict when CC is strict. ◻

Lemma 2.56.5. NAG-001853 (Dimension lowering by a nonzero ordinary ideal). Let KK have characteristic zero, let XX be good, and let VV be an ordinary affinoid neighborhood of a point tt in AXd,an\mathbb A_X^{d,\mathrm{an}}. Put x=f(t)x=f(t), assume that OX,x\mathcal O_{X,x} is a field and that F=H(x)F=\mathcal H(x) is nontrivially valued, and let 0≠I⊊OV,t0\ne I\subsetneq\mathcal O_{V,t} be an ideal. The closed-subspace germ defined by II has an ordinary affinoid representative ZZ whose fibre ZxZ_x is strict and has dimension at most d−1d-1. The maps Z→XZ\to X are inner at tt, and their ordinary local rings are OZ,t=OV,t/I,OZx,t=OVx,t/IOVx,t.\mathcal O_{Z,t}=\mathcal O_{V,t}/I, \qquad \mathcal O_{Z_x,t}=\mathcal O_{V_x,t}/I\mathcal O_{V_x,t}.

Proof. The ordinary local ring is Noetherian, so finitely many generators of II extend to functions on an ordinary affinoid neighborhood of tt. Choose a nonzero germ h∈Ih\in I and represent it there as well. Its image in the fibre germ is nonzero. Indeed, otherwise its zero set would contain an ordinary fibre neighborhood; Proposition 2.56.2, applied to the smooth map from the ordinary interior of VV to XX, would make hh zero on an ordinary neighborhood of tt, contrary to its choice.

Proposition 2.51.3 now gives an ordinary affinoid neighborhood with strict fibre algebra CC. This fibre is an affinoid domain of AFd,an\mathbb A_F^{d,\mathrm{an}}. Its ordinary local rings are regular by Theorem 2.35.23. Lemma 2.52.2 therefore writes CC as a finite product of regular domains. Use Lemma 2.56.4 to retain the factor C0C_0 containing tt, over the original field and without changing the ordinary germs. The image of hh in C0C_0 is nonzero, since its fibre germ was nonzero. Lemma 2.56.3 gives dim⁡C0≤d\dim C_0\leq d.

Cut out ZZ by the chosen generators of II. Its fibre algebra is the strict quotient C0/ICC_0/I_C, where ICI_C contains the nonzero image of hh. Every chain of primes containing ICI_C can be extended strictly at its lower end by the zero prime of the domain C0C_0. Hence dim⁡(C0/IC)≤d−1\dim(C_0/I_C)\leq d-1. Coherent closed-subspace restriction gives the two displayed local-ring identifications. A closed immersion is finite and inner; composition with the inner map from the chosen ordinary neighborhood to XX proves the final assertion. ◻

Theorem 2.56.6. NAG-001854 (Regularity along an inner generic fibre in characteristic zero). Let KK be a complete field of characteristic zero and let f:Y→Xf:Y\to X be a morphism of good analytic spaces. Let y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X) and put x=f(y)x=f(y). If OX,x\mathcal O_{X,x} is a field and H(x)\mathcal H(x) is nontrivially valued, then the canonical local homomorphism OY,y⟶OYx,y\mathcal O_{Y,y}\longrightarrow\mathcal O_{Y_x,y} is regular: it is flat and all its algebraic prime fibres are geometrically regular.

Proof. We induct on an upper bound nn for the dimension of a strict affinoid fibre of an ordinary affinoid representative of the source germ. Such a bound exists by Proposition 2.51.3. The induction assertion concerns every morphism satisfying the stated hypotheses that has such a representative with fibre dimension at most nn. Thus later changes of representatives do not require a comparison with a minimal local-dimension invariant.

For such a representative let dd be its strict fibre dimension, so d≤nd\leq n. Theorem 2.51.4 gives, on ordinary neighborhoods, a finite map to an affinoid ordinary neighborhood VV of tt in AXd,an\mathbb A_X^{d,\mathrm{an}}. Finite branch separation, followed by shrinking VV, selects the branch containing yy and makes yy the only point above tt. Put R0=OV,t,T0=OVx,t,S=OY,y,T=OYx,y.R_0=\mathcal O_{V,t},\quad T_0=\mathcal O_{V_x,t},\quad S=\mathcal O_{Y,y},\quad T=\mathcal O_{Y_x,y}. The finite direct-image stalk formulas identify SS with a finite R0R_0-algebra and give T=S⊗R0T0T=S\otimes_{R_0}T_0. To check the latter identity, apply finite affinoid base change and then take ordinary stalks. The unique point above tt is still the unique point in the corresponding analytic fibre, so both direct-image stalk products have just one factor. It is enough to prove that R0→T0R_0\to T_0 is regular: flatness is preserved by this base change, and the fibre at a prime of SS is a field extension of the corresponding geometrically regular fibre of R0→T0R_0\to T_0.

Both R0R_0 and T0T_0 are regular local rings. For R0R_0, use the smooth affine-space projection on the ordinary interior of VV, Theorem 2.35.30, and regularity of the base field OX,x\mathcal O_{X,x}. For T0T_0, use Theorem 2.35.23. The regularity ascent used here is [StacksAlg] (Tag 031E). If R0R_0 is a field, its map to the regular ring T0T_0 is flat and geometrically regular by characteristic zero and Stacks 0381; this finishes that case. In particular it proves the induction when n=0n=0, since then d=0d=0 and VV is an ordinary neighborhood of xx in XX, so R0=OX,xR_0=\mathcal O_{X,x} is a field.

Suppose now that R0R_0 is not a field, and write m\mathfrak m for its maximal ideal. Every power mq\mathfrak m^q, q≥1q\geq 1, is nonzero because a regular local ring is a domain by Stacks 00NP. Lemma 2.56.5, applied to I=mqI=\mathfrak m^q, represents the resulting closed-subspace germ by an inner affinoid with strict fibre dimension at most d−1≤n−1d-1\leq n-1. The induction hypothesis makes R0/mq⟶T0/mqT0R_0/\mathfrak m^q\longrightarrow T_0/\mathfrak m^qT_0 regular, and in particular flat. The rings R0,T0R_0,T_0 are Noetherian and mT0\mathfrak mT_0 lies in the maximal ideal of T0T_0. Apply the adic flatness criterion [StacksAlg] (Tag 0523), with module T0T_0 over itself, to all these quotients. It follows that R0→T0R_0\to T_0 is flat.

It remains to examine its algebraic prime fibres. At the zero prime, the fibre is a localization of the regular ring T0T_0, hence regular and, by characteristic zero and Stacks 0381, geometrically regular. If p≠0\mathfrak p\ne 0 is a prime of R0R_0, apply the dimension-lowering lemma to I=pI=\mathfrak p. Induction makes R0/p→T0/pT0R_0/\mathfrak p\to T_0/\mathfrak pT_0 regular. Its fibre at the zero prime is exactly T0⊗R0κ(p)T_0\otimes_{R_0}\kappa(\mathfrak p), and is geometrically regular. This proves regularity of R0→T0R_0\to T_0 and then of S→TS\to T, completing the induction. ◻

Corollary 2.56.7. NAG-001855 (Ordinary closed fibres and analytic fibres). Let KK have characteristic zero, let f:Y→Xf:Y\to X be a morphism of good analytic spaces, and let y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X) with x=f(y)x=f(y) and nontrivially valued H(x)\mathcal H(x). Put R=OX,xR=\mathcal O_{X,x}, S=OY,yS=\mathcal O_{Y,y} and let m\mathfrak m be the maximal ideal of RR. Then S/mS⟶OYx,yS/\mathfrak mS\longrightarrow\mathcal O_{Y_x,y} is a regular faithfully flat local homomorphism.

Proof. The Noetherian ideal m\mathfrak m has finitely many generators. Represent them on an ordinary affinoid neighborhood of xx and let X0X_0 be the closed analytic subspace they define. It contains xx, and coherent closed-subspace restriction gives OX0,x=R/m\mathcal O_{X_0,x}=R/\mathfrak m. Set Y0=Y×XX0Y_0=Y\times_X X_0. The corresponding point has ordinary local ring S/mSS/\mathfrak mS, while its analytic fibre is unchanged because all the chosen generators vanish in H(x)\mathcal H(x). Interior is preserved by base change. Theorem 2.56.6 applies to Y0→X0Y_0\to X_0, giving the claimed regular map. It is local and flat between nonzero local rings, hence faithfully flat. ◻

References. The induction follows the mechanism of [Duc18] (Theorem 6.3.3). The finite-projection input is reconstructed with the global dimension of a strict fibre, and the intervening component cut and dimension bound are proved above. The adic flatness input is the exact algebraic statement [StacksAlg] (Tag 0523).

Lemma 2.56.8. NAG-001856 (A field germ at an auxiliary Gauss point). Let XX be good over a complete field of characteristic zero, let OX,x\mathcal O_{X,x} be a field, and suppose that F=H(x)F=\mathcal H(x) is trivially valued. Choose 0<r<10<r<1 and let x∗x^* be the radius-rr Gauss point above xx in X∗=X×KAK1,anX^*=X\times_K\mathbb A_K^{1,\mathrm{an}}. Then H(x∗)=Fr\mathcal H(x^*)=F_r is nontrivially valued and OX∗,x∗\mathcal O_{X^*,x^*} is a field.

Proof. The completed residue field is FrF_r by the Gauss evaluation and the completed-residue-field universal property. To prove the local-ring assertion, take a germ hh with h(x∗)=0h(x^*)=0. Lemma 2.55.1 gives an ordinary product neighborhood U×KDU\times_KD of x∗x^* on which hh is represented, with DD an open annulus containing radius rr. Since OX,x\mathcal O_{X,x} is a field, it is normal. Corollary 2.35.32 and Lemma 2.56.1 give a normal integral affinoid ordinary neighborhood M(A)⊂U\mathcal M(A)\subset U of xx. The map A→OX,x↪FA\to\mathcal O_{X,x}\hookrightarrow F is injective: it is a nonzero flat map from a domain followed by the residue-field embedding.

Choose a closed annulus D0⊂DD_0\subset D with rr strictly between its two radii. The restriction of hh to M(A)×KD0\mathcal M(A)\times_KD_0 has a unique convergent Laurent expansion ∑j∈ZajTj\sum_{j\in\mathbf Z}a_jT^j, with aj∈Aa_j\in A. This is the weighted Laurent description of the annulus algebra; uniqueness is preserved under restriction to smaller annuli. At the Gauss point x∗x^* its value is ∣h(x∗)∣=max⁡j∣aj(x)∣rj.|h(x^*)|=\max_j |a_j(x)|r^j. Thus every aj(x)a_j(x) is zero. Injectivity of A→FA\to F makes all aja_j zero simultaneously, so hh vanishes on this product, which is an ordinary neighborhood of x∗x^*. The maximal ideal of OX∗,x∗\mathcal O_{X^*,x^*} is exactly the kernel of evaluation in its completed residue field. We have proved that kernel zero, so this local ring is a field. ◻

Theorem 2.56.9. NAG-001857 (Generic-fibre comparison for all characteristic-zero residue fields). Let KK be any complete field of characteristic zero and let f:Y→Xf:Y\to X be a morphism of good analytic spaces. If y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X) and OX,f(y)\mathcal O_{X,f(y)} is a field, then OY,y⟶OYf(y),y\mathcal O_{Y,y}\longrightarrow \mathcal O_{Y_{f(y)},y} is a regular local homomorphism.

Proof. Put x=f(y)x=f(y) and F=H(x)F=\mathcal H(x). If FF is nontrivially valued, use Theorem 2.56.6. Suppose that FF is trivially valued. Form X∗=X×KAK1,anX^*=X\times_K\mathbb A_K^{1,\mathrm{an}} and its point x∗x^* as in Lemma 2.56.8; its ordinary local ring is a field and its completed residue field is FrF_r. Put Y∗=Y×XX∗=Y×KAK1,anY^*=Y\times_X X^*=Y\times_K\mathbb A_K^{1,\mathrm{an}}, and choose the radius-rr Gauss point y∗y^* above yy. Here the Gauss seminorm on H(y)[T]\mathcal H(y)[T] is multiplicative for any radius; rr need not be free over H(y)\mathcal H(y). Its restriction to F[T]F[T] is the chosen point x∗x^*.

Interior survives the base change. Write S=OY,y,S∗=OY∗,y∗,T=OYx,y,T∗=OYx∗∗,y∗.S=\mathcal O_{Y,y},\quad S^*=\mathcal O_{Y^*,y^*},\quad T=\mathcal O_{Y_x,y},\quad T^*=\mathcal O_{Y^*_{x^*},y^*}. The smooth projection Y∗→YY^*\to Y makes S→S∗S\to S^* regular by Theorem 2.35.30. The nontrivially valued comparison just proved makes S∗→T∗S^*\to T^* regular. Moreover Yx∗∗=(Yx)FrY^*_{x^*}=(Y_x)_{F_r} by complete base-change associativity, so T→T∗T\to T^* is faithfully flat by Theorem 2.35.4. These are maps of Noetherian local rings and form the expected commutative square with S→TS\to T.

We spell out composition and descent of regularity in this square. The composite S→T∗S\to T^* is flat. For a prime p⊂S\mathfrak p\subset S and a finite field extension E/κ(p)E/\kappa(\mathfrak p), the algebra S∗⊗SES^*\otimes_SE is regular. Base change of the regular map S∗→T∗S^*\to T^* is flat with geometrically regular fibres; localization and [StacksAlg] (Tag 031E) make T∗⊗SET^*\otimes_SE regular. Thus S→T∗S\to T^* is regular. Faithful flatness of T→T∗T\to T^* descends SS-flatness: tensor any injection of SS-modules first with TT and then with the faithfully flat TT-module T∗T^*, and detect its kernel. Finally T⊗SE⟶T∗⊗SET\otimes_SE\longrightarrow T^*\otimes_SE is faithfully flat and its target is regular. At every prime of the source choose a prime above it and apply regular-local-ring descent, [StacksAlg] (Tag 00OF). Hence every T⊗SET\otimes_SE is regular. This proves all geometrically regular prime fibres of S→TS\to T, and completes the proof. ◻

Corollary 2.56.10. NAG-001858 (Closed-fibre comparison in arbitrary characteristic-zero valuation). For a morphism f:Y→Xf:Y\to X of good analytic spaces over any complete characteristic-zero field and an inner point yy with image xx, put R=OX,xR=\mathcal O_{X,x} and S=OY,yS=\mathcal O_{Y,y}. If m\mathfrak m is the maximal ideal of RR, then S/mS⟶OYx,yS/\mathfrak mS\longrightarrow\mathcal O_{Y_x,y} is regular and faithfully flat.

Proof. Realize a finite generating family of m\mathfrak m on an ordinary base neighborhood and form the associated closed analytic base change. As in Corollary 2.56.7, the new ordinary base ring is the field R/mR/\mathfrak m, the ordinary source ring is S/mSS/\mathfrak mS, and the analytic fibre and the inner point are unchanged. Apply Theorem 2.56.9. Its regular map is local and flat, hence faithfully flat. ◻

2.57 Coherent flatness at an inner point

Finite projection replaces a coherent sheaf by a finite module on a relative affine-space domain. A basis at the generic point of an integral fibre then separates a free part from a quotient of smaller fibre dimension. We begin with the two module statements used in this induction.

Lemma 2.57.1. NAG-001869 (Support under analytic pullback). Let A→BA\to B be a bounded homomorphism of generalized affinoid algebras, allowing a complete extension of the ground field for BB, and let MM be a finite AA-module. For z∈M(B)z\in\mathcal M(B) with image x∈M(A)x\in\mathcal M(A), z∈Supp⁡(M⊗AB)⟺x∈Supp⁡(M).z\in\operatorname{Supp}(M\otimes_A B) \quad\Longleftrightarrow\quad x\in\operatorname{Supp}(M). Consequently the support of the pullback of a coherent sheaf is the inverse image of its support. The closed support defined by the annihilator and its analytic base change have the same underlying set as the support of the pulled-back sheaf.

Proof. Put p=ker⁡x\mathfrak p=\ker x and q=ker⁡z\mathfrak q=\ker z. A finite module over a local ring is zero exactly when its residue vector space is zero, by Nakayama’s determinant argument. Hence xx belongs to the support precisely when M⊗Aκ(p)≠0M\otimes_A\kappa(\mathfrak p)\ne 0. The bounded map supplies an injection κ(p)↪H(z)\kappa(\mathfrak p)\hookrightarrow\mathcal H(z), and extension of vector-space scalars is faithful. The canonical identifications (M⊗AB)⊗BH(z)≃M⊗AH(z)≃(M⊗Aκ(p))⊗κ(p)H(z)(M\otimes_A B)\otimes_B\mathcal H(z) \simeq M\otimes_A\mathcal H(z) \simeq (M\otimes_A\kappa(\mathfrak p)) \otimes_{\kappa(\mathfrak p)}\mathcal H(z) therefore prove the equivalence. The same argument at q\mathfrak q identifies the left side with membership in the support over BB. The internal finite-module correspondence identifies analytic pullback with the indicated ordinary tensor product of finite modules, by Theorem 2.39.6 and finite presentation. This gives the sheaf assertion on affinoid domains and hence globally. Finally the analytic base change of the annihilator-defined closed subspace has point set exactly the inverse image of its original point set, by the universal property of a closed immersion. That point set is the support just computed; no equality of the two annihilator ideals is asserted or needed. ◻

Lemma 2.57.2. NAG-001870 (A generic basis lifted from original sections). Let T=M(B)→X=M(A)T=\mathcal M(B)\to X=\mathcal M(A) be affinoid, let x∈Xx\in X, and suppose that C=B⊗^AH(x)C=B\widehat\otimes_A\mathcal H(x) is a nonzero integral strict affinoid algebra. For every finite BB-module NN, there are a nonnegative integer rr and a BB-linear map u:Br→Nu:B^r\to N such that uCu_C is injective and is an isomorphism on a nonempty Zariski-open subset of M(C)\mathcal M(C). If Q=Coker⁡(u)Q=\operatorname{Coker}(u) and d=dim⁡Cd=\dim C, then QC=0Q_C=0 when d=0d=0, and otherwise dim⁡Supp⁡(QC)≤d−1whenever QC≠0.\dim\operatorname{Supp}(Q_C)\leq d-1 \qquad\text{whenever }Q_C\ne0. The bijectivity locus of uCu_C is ordinarily dense in the fibre. The columns of uu may be chosen from any fixed finite generating family of NN.

Proof. Let E=Frac⁡(C)E=\operatorname{Frac}(C) and r=dim⁡E(NC⊗CE)r=\dim_E(N_C\otimes_C E). The images of the given generators span this vector space. Choose rr of them which form a basis and use the corresponding original elements of NN to define uu; if r=0r=0, use the zero module as source. The map uCu_C becomes an isomorphism over EE. Its kernel is a submodule of the free module CrC^r and becomes zero over EE. Since CC is a domain, a submodule of CrC^r has no nonzero element killed by a nonzero element of CC. Thus that kernel is zero.

Right exactness gives QC=Coker⁡(uC)Q_C=\operatorname{Coker}(u_C). This is a finite torsion CC-module. Kill a finite generating family by nonzero elements and multiply those elements, obtaining a∈C∖{0}a\in C\setminus\{0\} with aQC=0aQ_C=0; when QC=0Q_C=0, take a=1a=1. Thus uCu_C is an isomorphism on D(a)D(a), which is nonempty and ordinarily dense by Lemma 2.52.1. If d=0d=0, the Noetherian domain CC is a field and every finite torsion module is zero. If d>0d>0 and QC≠0Q_C\ne 0, every prime in its support contains aa. Prepending the zero prime to a chain of primes containing aa shows that its Krull dimension is at most d−1d-1. The algebra C/Ann⁡(QC)C/\operatorname{Ann}(Q_C) is strict; its analytic dimension equals its Krull dimension by Proposition 3.21.1 and strict finite normalization. This proves the asserted fibre-dimension bound. ◻

Lemma 2.57.3. NAG-001871 (Lifting fibre injectivity at an inner point). Let KK have characteristic zero, let T→XT\to X be a morphism of good analytic spaces, and let t∈Int⁡(T/X)t\in\operatorname{Int}(T/X) with image xx and nontrivially valued H(x)\mathcal H(x). Let u:L→Nu:L\to N be a morphism of coherent sheaves. If NtN_t is flat over OX,x\mathcal O_{X,x} and the induced map on the analytic fibre is injective at tt, then utu_t is injective and Coker⁡(u)t\operatorname{Coker}(u)_t is flat over OX,x\mathcal O_{X,x}.

Proof. Put R=OX,xR=\mathcal O_{X,x}, S=OT,tS=\mathcal O_{T,t}, and let m\mathfrak m be the maximal ideal of RR. Coherent pullback identifies the fibre stalk map with ut⊗SOTx,tu_t\otimes_S\mathcal O_{T_x,t}. Corollary 2.56.7 says that S/mS→OTx,tS/\mathfrak mS\to\mathcal O_{T_x,t} is faithfully flat. Thus fibre injectivity descends to Lt/mLt⟶Nt/mNtinjective.L_t/\mathfrak m L_t\longrightarrow N_t/\mathfrak m N_t \quad\text{injective}. The ring SS is Noetherian, LtL_t is a finite SS-module, and NtN_t is flat over RR. The local flat-map criterion [StacksAlg] (Tag 00ME), applied with these modules, gives both conclusions. Formation of the coherent cokernel commutes with ordinary stalks, identifying its module with the quotient in that criterion. ◻

Theorem 2.57.4. NAG-001872 (Coherent flatness at an inner point in characteristic zero). Let KK be a complete non-Archimedean field of characteristic zero, let f:Y→Xf:Y\to X be a morphism of good KK-analytic spaces, and let F\mathcal F be coherent on YGY_G. Suppose that y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X) and that H(f(y))\mathcal H(f(y)) is nontrivially valued. Then F\mathcal F is naively XX-flat at yy if and only if it is analytically XX-flat there. In particular this holds for every complete nontrivially valued characteristic-zero ground field and every relative inner point.

Proof. Analytic flatness implies naive flatness by the identity test. For the converse, put x=f(y)x=f(y) and F=H(x)F=\mathcal H(x). If the coherent stalk is zero, it is zero on an ordinary neighborhood: finitely many generators vanish after a common shrink. Every pullback is then zero near every point above yy, proving analytic flatness. Assume henceforth that yy belongs to the support.

Step 1: the induction parameter. Proposition 2.51.3 supplies an ordinary affinoid representative with strict FF-fibre. Let ZZ be the closed support of F\mathcal F on this representative. It is inner over XX at yy, since a closed immersion is finite and inner and interior is preserved under composition. Regard F\mathcal F as a coherent sheaf on ZZ. This changes neither kind of flatness, by Theorem 2.35.17. The algebra of ZxZ_x is a strict quotient, and Lemma 2.57.1 identifies its point set with the support of Fx\mathcal F_x. We prove the assertion by induction on an upper bound n≥0n\geq 0 for this strict fibre dimension. Restrictions to smaller ordinary affinoid neighborhoods cannot increase the analytic dimension of the support, by Corollary 3.22.8; on a strict fibre this is its Krull dimension.

Step 2: a finite projection with integral fibre target. Write d=dim⁡Zx≤nd=\dim Z_x\leq n. Theorem 2.51.4 gives, on ordinary neighborhoods, a finite map from ZZ to an affinoid ordinary neighborhood TT of tt in AXd,an\mathbb A_X^{d,\mathrm{an}}. Separate its finite branches and retain the one through yy, so that the fibre over tt is the singleton {y}\{y\}. The point tt is inner over XX, since it is in the ordinary interior of TT in relative affine space. Shrink TT by Proposition 2.51.3 so that TxT_x is strict. Its ordinary local rings are regular by Theorem 2.35.23; Lemma 2.52.2 therefore decomposes its algebra into integral factors. Lemma 2.56.4 retains the factor through tt by an ordinary domain over the original field. Pull the finite map back at each restriction. We obtain a finite map π:Z→T\pi:Z\to T with singleton distinguished fibre, whose fibre algebra C=O(Tx)C=\mathcal O(T_x) is integral and strict and satisfies dim⁡C≤d\dim C\leq d by Lemma 2.56.3. All distinguished ordinary germs remain unchanged.

Step 3: the quotient has smaller fibre support. Put N=π∗FN=\pi_*\mathcal F. It is coherent, and NtN_t is naively XX-flat by Theorem 2.35.17. The finite-module description of NN and Lemma 2.57.2 give a morphism u:L→Nu:L\to N with LL finite free, whose fibre map is injective and generically bijective on TxT_x. Set Q=Coker⁡(u)Q=\operatorname{Coker}(u). Lemma 2.57.3 makes utu_t injective and QtQ_t naively XX-flat. The strict fibre support of QQ is either empty or has dimension at most dim⁡C−1≤n−1\dim C-1\leq n-1. If Qt=0Q_t=0, the initial vanishing argument makes it analytically flat at tt. Otherwise apply the induction hypothesis to QQ on its closed support in TT: that support is still inner over XX, and its fibre has the stated strictly smaller bound. Thus QQ is analytically XX-flat at tt. This also covers the induction start n=0n=0, when the generic-basis lemma forces Qx=0Q_x=0 and Nakayama gives Qt=0Q_t=0.

Step 4: recover the original sheaf. The space TT is quasi-smooth over XX, and the bijectivity locus of uxu_x is ordinarily dense. Corollary 2.52.6, applied to the injectivity of utu_t and the analytic flatness of QtQ_t, proves analytic XX-flatness of NN at tt. The finite-pushforward equivalence then gives analytic flatness of F\mathcal F at yy. That equivalence and the generic-fibre criterion include every good analytic base change over every complete extension; thus the conclusion is exactly the universal analytic-flatness condition. Finally a completed residue field over a nontrivially valued ground field is automatically nontrivially valued, proving the last sentence. ◻

References. The support statement is finite-module algebra. The generic-basis and dimension-induction steps reorganize the mechanism of analytic dévissage in [Duc18] (§8.2 and Theorem 8.3.4). The ordinary-to-analytic fibre comparison is proved internally in Corollary 2.56.7; the sole additional algebraic criterion is [StacksAlg] (Tag 00ME).

2.58 Gauss lifting and flatness in trivial valuation

Lemma 2.58.1. NAG-001804 (The ordinary germ at a free point of a trivial-field line). Let FF be trivially valued and let 0<r<10<r<1. At the radius-rr Gauss point ξr\xi_r of AF1,an\mathbb A_F^{1,\mathrm{an}}, evaluation identifies the ordinary local ring with the underlying field of Fr=F((T))F_r=F((T)).

Proof. Lemma 2.55.1, with base M(F)\mathcal M(F), shows that annuli a≤∣T∣≤ba\leq|T|\leq b with 0<a<r<b<10<a<r<b<1 form an ordinary affinoid neighborhood basis. Their analytic functions have unique Laurent expansions ∑j∈ZcjTj\sum_{j\in\mathbf Z}c_jT^j, with ∣cj∣bj→0|c_j|b^j\to 0 as j→+∞j\to+\infty and ∣cj∣aj→0|c_j|a^j\to 0 as j→−∞j\to-\infty. This is the Laurent coefficient normal form of Theorem 2.13.2, applied to the annulus algebra. Since every nonzero coefficient has absolute value one, the positive coefficients are arbitrary and only finitely many negative coefficients can be nonzero. Thus the underlying algebra is exactly F((T))F((T)). Restriction between these annuli preserves every coefficient and is the identity on that field. The filtered colimit giving the ordinary stalk is therefore F((T))F((T)) as well. Gauss evaluation gives its original radius-rr absolute value and is injective. ◻

Lemma 2.58.2. NAG-001805 (Naive flatness lifts to an auxiliary Gauss point). Let f:Y→Xf:Y\to X be a morphism of good spaces over a complete characteristic-zero field, and let y∈Yy\in Y with image xx have trivially valued F=H(x)F=\mathcal H(x). Let F\mathcal F be coherent and naively XX-flat at yy. For 0<r<10<r<1, form X∗=X×KAK1,anX^*=X\times_K\mathbb A_K^{1,\mathrm{an}} and Y∗=Y×KAK1,anY^*=Y\times_K\mathbb A_K^{1,\mathrm{an}}, and let x∗,y∗x^*,y^* be their radius-rr Gauss points above x,yx,y. Then the pullback F∗\mathcal F^* is naively X∗X^*-flat at y∗y^*.

Proof. Write R=OX,xR=\mathcal O_{X,x}, S=OY,yS=\mathcal O_{Y,y}, R∗=OX∗,x∗R^*=\mathcal O_{X^*,x^*}, S∗=OY∗,y∗S^*=\mathcal O_{Y^*,y^*} and M=FyM=\mathcal F_y. The two affine-line projections are smooth, hence their ordinary local maps R→R∗R\to R^* and S→S∗S\to S^* are flat. Coherent pullback gives M∗=Fy∗∗=M⊗SS∗M^*=\mathcal F^*_{y^*}=M\otimes_SS^*. Since MM is RR-flat and S∗S^* is SS-flat, M∗M^* is RR-flat: tensoring an injection of RR-modules with MM gives an injection of SS-modules, which remains injective after tensoring with S∗S^*.

Let m\mathfrak m be the maximal ideal of RR and put I=mR∗I=\mathfrak mR^*. Apply Corollary 2.56.10 to the inner smooth projection X∗→XX^*\to X at x∗x^*. Its analytic fibre is AF1,an\mathbb A_F^{1,\mathrm{an}}, so the preceding lemma gives a faithfully flat local map R∗/I⟶F((T)).R^*/I\longrightarrow F((T)). Its source is a field. Indeed its maximal ideal extends to the zero ideal of the target field, and faithful flatness contracts that zero ideal to zero. Hence M∗/IM∗M^*/IM^* is flat over R∗/IR^*/I.

There is also no relative Tor obstruction. Choose a free RR-resolution of R/mR/\mathfrak m. Since R∗R^* is RR-flat, its tensor with R∗R^* is a free R∗R^*-resolution of R∗/IR^*/I. Tensoring that resolution with M∗M^* is the same as tensoring the original resolution over RR with M∗M^*, and the latter has zero positive homology by RR-flatness. Thus Tor⁡1R∗(M∗,R∗/I)=0.\operatorname{Tor}^{R^*}_1(M^*,R^*/I)=0. The local rings R∗,S∗R^*,S^* are Noetherian and M∗M^* is finite over S∗S^*. The local criterion [StacksAlg] (Tag 00ML) now makes M∗M^* flat over R∗R^*, proving the assertion. ◻

Lemma 2.58.3. NAG-001806 (Analytic flatness descends from a Gauss lift). Let f:Y→Xf:Y\to X be a morphism of good analytic spaces over any complete field, let F\mathcal F be coherent, and let r>0r>0. Form the simultaneous affine-line base change Y∗→X∗Y^*\to X^* and let y∗y^* be the radius-rr Gauss point above yy. Then F\mathcal F is analytically XX-flat at yy if and only if its pullback F∗\mathcal F^* is analytically X∗X^*-flat at y∗y^*.

Proof. The forward implication is a base-change test in the definition. Conversely, fix an arbitrary complete extension L/KL/K, a good LL-analytic test X1→XX_1\to X, and a point y1∈Y1=Y×XX1y_1\in Y_1=Y\times_X X_1 above yy, with image x1x_1. Form X1∗=X1×LAL1,anX_1^*=X_1\times_L\mathbb A_L^{1,\mathrm{an}} and Y1∗=Y1×LAL1,anY_1^*=Y_1\times_L\mathbb A_L^{1,\mathrm{an}}. The radius-rr Gauss point y1∗y_1^* maps to y∗y^*: the completed-residue embeddings are isometric and the coefficientwise Gauss norm restricts to the same norm on every polynomial over the smaller field. Its base point x1∗x_1^* likewise maps to the chosen Gauss point in X∗X^*.

The assumed analytic flatness at y∗y^* therefore makes the pulled-back coherent stalk at y1∗y_1^* flat over OX1∗,x1∗\mathcal O_{X_1^*,x_1^*}. The horizontal maps in the local square OX1,x1⟶OX1∗,x1∗↓↓OY1,y1⟶OY1∗,y1∗\begin{matrix} \mathcal O_{X_1,x_1}&\longrightarrow&\mathcal O_{X_1^*,x_1^*}\\ \big\downarrow&&\big\downarrow\\ \mathcal O_{Y_1,y_1}&\longrightarrow&\mathcal O_{Y_1^*,y_1^*} \end{matrix} are flat local maps by Theorem 2.48.3: take relative closed discs of radius greater than rr, whose ordinary interiors contain the Gauss points and have the same ordinary stalks as the affine-line products. In particular the lower map is faithfully flat. Coherent pullback is ordinary tensor product on these stalks. Lemma 2.35.6 proves naive flatness at y1y_1. Since the complete extension, the good test and the point above yy were arbitrary, this is analytic flatness at yy. ◻

Theorem 2.58.4. NAG-001807 (Inner coherent flatness in arbitrary characteristic-zero valuation). Let KK be any complete field of characteristic zero, let f:Y→Xf:Y\to X be a morphism of good analytic spaces, and let F\mathcal F be coherent. At every y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X), naive XX-flatness of F\mathcal F is equivalent to analytic XX-flatness.

Proof. Only the implication from naive to analytic flatness requires proof. Put x=f(y)x=f(y). If H(x)\mathcal H(x) is nontrivially valued, this is Theorem 2.57.4. Otherwise choose 0<r<10<r<1 and make the simultaneous affine-line base change. Lemma 2.58.2 gives naive flatness of the pullback at y∗y^*. The point y∗y^* is inner over X∗X^* because interior is stable under base change, and H(x∗)=H(x)r\mathcal H(x^*)=\mathcal H(x)_r is nontrivially valued. The same nontrivially valued theorem makes the pullback analytically flat at y∗y^*. Lemma 2.58.3 then descends analytic flatness to yy. ◻

References. These Gauss-point arguments extend the internally reconstructed characteristic-zero case of [Duc18] (Theorem 8.3.4) to trivial valuations. The local flatness criterion is the exact Stacks statement [StacksAlg] (Tag 00ML); the Tor comparison is proved by the explicit free-resolution calculation above.

2.59 Smooth ambient spaces for quasi-smooth germs

Lemma 2.59.1. NAG-002011 (Residue fields of rank-zero quasi-smooth morphisms). Let f:Y→Xf:Y\to X be a morphism of good analytic spaces over any complete field. If ff is quasi-smooth at yy and rk⁡yΩY/X=0\operatorname{rk}_y\Omega_{Y/X}=0, then H(y)/H(f(y))\mathcal H(y)/\mathcal H(f(y)) is finite separable.

Proof. Put x=f(y)x=f(y) and F=H(x)F=\mathcal H(x). Quasi-smoothness and its differential rank survive passage to the analytic fibre. Lemma 2.50.1 therefore gives an ordinary open affinoid neighborhood of yy in YxY_x whose algebra is finite-dimensional over FF. Its local factor DD at yy is the ordinary local ring of this finite space. By Theorem 2.35.14 that ring is regular. A zero-dimensional regular local ring is a field, so D=E=H(y)D=E=\mathcal H(y), and E/FE/F is finite.

For every finite field extension L/FL/F, the extending absolute value makes LL complete, and finite base change identifies the corresponding algebra with E⊗FLE\otimes_F L. All its local factors are regular by geometric regularity of the analytic fibre, hence fields. Thus E⊗FLE\otimes_F L is reduced. If some a∈Ea\in E were inseparable over FF, its minimal polynomial would acquire a repeated root over a finite splitting field L/FL/F. The algebra F[a]⊗FLF[a]\otimes_F L would then contain a nonzero nilpotent. Its injection into E⊗FLE\otimes_F L, obtained by tensoring the injection F[a]↪EF[a]\hookrightarrow E with the field LL, contradicts reducedness. Every element of EE is therefore separable over FF. ◻

NAG-002012 · Open at full scope

Theorem 2.59.2. NAG-002012 (An étale-domain factorization in rank zero). Let f:Y→Xf:Y\to X be a morphism of good analytic spaces, quasi-smooth of differential rank zero at yy, and put x=f(y)x=f(y). There are an ordinary affinoid neighborhood UU of xx, a finite étale morphism T→UT\to U, and an ordinary affinoid neighborhood VV of yy such that f∣Vf|_V factors as an affinoid-domain inclusion V↪TV\hookrightarrow T followed by T→U↪XT\to U\hookrightarrow X.

Proof. Work on good affinoid ordinary neighborhoods on which the rank-zero Jacobian presentation holds. Write F=H(x)F=\mathcal H(x) and E=H(y)E=\mathcal H(y). By the preceding lemma E/FE/F is finite separable. Theorem 2.54.3 realizes it by a finite étale morphism T→UT\to U with a unique point tt above xx and a specified identification H(t)=E\mathcal H(t)=E over FF. In Y×XTY\times_XT the multiplication map E⊗FE→EE\otimes_FE\to E defines a point y′y' above (y,t)(y,t) with completed residue field EE. The projection Y×XT→YY\times_XT\to Y is étale. Corollary 2.54.4 identifies its germ at y′y' with the germ of YY at yy. Thus, after restricting ordinary neighborhoods, the second projection gives a lift g:Y→Tg:Y\to T whose completed-residue-field map at yy is the identity of EE. This lifted map is quasi-smooth of rank zero: before the local identification it is the base change of ff, and the property is stable under such base change and ordinary restriction.

Consider the germ reductions of gg. Both source and target germs are good, so their reductions are affine open subspaces of the same graded valuation space PK~gr(E~gr)\mathscr P_{\widetilde K^{\rm gr}}(\widetilde E^{\rm gr}). Their structural maps are open embeddings, by Theorem 3.19.19 and the good-germ construction. Compatibility with the identity on E~gr\widetilde E^{\rm gr} identifies R(Y,y)→R(T,t)\mathscr R(Y,y)\to\mathscr R(T,t) with an inclusion of a nonempty quasi-compact affine open subspace. Theorem 3.19.8 supplies a subdomain germ (W,t)(W,t) of (T,t)(T,t) whose reduction is that open subspace. Its reduction is affine, so (W,t)(W,t) is good by Theorem 3.19.19.

The lift factors through this actual subdomain germ. Indeed the inverse-image domain Y×TWY\times_TW has the same reduction as (Y,y)(Y,y), by the inverse-image formula of Theorem 3.19.9. The order-reflecting domain classification therefore makes it the entire source germ. Restricting an ordinary neighborhood of yy gives g:Y→Wg:Y\to W. Its reduction is an isomorphism over the same graded field, hence proper. Theorem 2.46.1 makes yy an interior point of this map. Also Y→WY\to W remains quasi-smooth of rank zero, since it is the preceding map base-changed by the domain inclusion W↪TW\hookrightarrow T.

The fibre point yy is isolated by Lemma 2.50.1; the ordinary local map is flat and its relative differentials vanish by Theorem 2.35.14. Proposition 2.49.2 consequently makes Y→WY\to W étale near yy. Its completed residue fields agree, because domain inclusions preserve completed residue fields. Corollary 2.54.4 now gives an isomorphism of ordinary neighborhoods in YY and WW. Choose an affinoid ordinary neighborhood VV of yy within this isomorphism. Its image is an affinoid domain of WW, and hence of TT by transitivity of analytic domains. All the choices were made over ordinary restrictions of UU; the original T→UT\to U remains finite étale. This is the required factorization. ◻

Lemma 2.59.3. NAG-002013 (Independent functions as relative coordinates). Let Y→XY\to X be quasi-smooth of differential rank dd at a point yy of good analytic spaces. If analytic germs f1,…,fℓf_1,\ldots,f_\ell have linearly independent differentials in ΩY/X⊗H(y)\Omega_{Y/X}\otimes\mathcal H(y), then (f,f1,…,fℓ):Y→AXℓ,an(f,f_1,\ldots,f_\ell):Y\to\mathbb A_X^{\ell,\mathrm{an}} is quasi-smooth of differential rank d−ℓd-\ell at yy.

Proof. Use the ordinary-neighborhood presentation in Theorem 2.35.14: after restrictions, YY is cut out by rr functions g1,…,grg_1,\ldots,g_r in an affinoid domain DD of AXn\mathbb A_X^n, with independent Jacobian rows and n−r=dn-r=d. Represent the fif_i on this YY and lift them to the affinoid algebra of DD, using the surjection defining the closed immersion. Choose an affinoid polydisc neighborhood PP of their values in AXℓ\mathbb A_X^\ell. The graph of the new map is cut out, in the inverse image of DD inside APn\mathbb A_P^n, by g1,…,gr,f1−Z1,…,fℓ−Zℓ.g_1,\ldots,g_r,\quad f_1-Z_1,\ldots,f_\ell-Z_\ell. The original rr Jacobian rows together with the ℓ\ell rows dfidf_i are linearly independent at yy: this is precisely independence of the latter classes modulo the relation rows. Some (r+ℓ)(r+\ell)-minor is therefore nonzero at yy, and it is invertible on an ordinary Laurent neighborhood. The graph ideal is generated by the displayed functions, so the Jacobian criterion makes the graph quasi-smooth over PP with rank n−r−ℓ=d−ℓn-r-\ell=d-\ell. As PP is an ordinary neighborhood of the target point, this is the asserted property over AXℓ\mathbb A_X^\ell. ◻

NAG-002014 · Open at full scope

Theorem 2.59.4. NAG-002014 (Smooth envelopes of good quasi-smooth germs). Let f:Y→Xf:Y\to X be a morphism of good analytic spaces over any complete field. It is quasi-smooth at yy if and only if an ordinary affinoid neighborhood VV of yy is, over an ordinary neighborhood of x=f(y)x=f(y), an affinoid domain of a smooth analytic XX-space. If the differential rank at yy is dd, the smooth ambient space may be chosen of relative dimension dd.

Proof. Assume first that ff is quasi-smooth at yy. On an ordinary affinoid representative, the coordinate differentials generate the finite module ΩY/X\Omega_{Y/X}. Choose dd analytic functions whose differentials form a basis after tensoring with H(y)\mathcal H(y). Lemma 2.59.3 makes the resulting map g:Y→AXd,ang:Y\to\mathbb A_X^{d,\mathrm{an}} quasi-smooth of rank zero. Theorem 2.59.2 gives an ordinary affinoid neighborhood VV of yy, an affinoid ordinary neighborhood UU of g(y)g(y) in AXd\mathbb A_X^d, and a finite étale map T→UT\to U in which VV is an affinoid domain. Replace UU by its ordinary interior near g(y)g(y), replace TT by its inverse image, and shrink VV within the resulting source neighborhood. Then T→U↪AXd→XT\to U\hookrightarrow\mathbb A_X^d\to X is smooth of relative dimension dd, while VV remains an affinoid domain of TT.

Conversely, smooth maps are quasi-smooth by Theorem 2.50.3, and restriction to an analytic domain preserves the Jacobian presentation and quasi-smoothness. Thus V→XV\to X is quasi-smooth. Since VV is an ordinary neighborhood of yy in YY, the same holds for the original germ. The differential rank is unchanged by domain restriction, giving the dimension assertion. ◻

References. The coordinate lemma and the smooth-envelope theorem reconstruct [Duc18] (Proposition 5.4.5 and Theorem 5.4.6). The rank-zero factorization uses the germ-reduction mechanism of the analytic Zariski main theorem, but here finite separability allows the full completed residue field to be realized by an étale germ. All domain and finite-map steps are proved in the preceding chapters.

2.60 A graph at its exact convergence radius

Throughout this section KK is complete and nontrivially valued. No restriction on its characteristic is imposed. The argument follows the geometric continuation proof of [Duc18] (4.4.2–4.4.6); the curve local-ring step is derived directly from the strict ordinary comparison already established in this book.

Lemma 2.60.1. NAG-001879 (Reduced strict domains and ordinary stalks). Let AA be a reduced strict KK-affinoid algebra. Every ordinary local ring of M(A)\mathcal M(A) is reduced. Every strict affinoid domain in M(A)\mathcal M(A) has a reduced affinoid algebra.

Proof. Fix xx, set D=Aker⁡(x)D=A_{\ker(x)} and R=OM(A),xR=\mathcal O_{\mathcal M(A),x}. Theorem 2.35.42 makes D→RD\to R a regular, faithfully flat homomorphism. The reduced Noetherian ring DD embeds into the product of the fraction fields of its finitely many minimal-prime quotients. Tensoring this injection with the flat DD-module RR embeds RR into the product of the corresponding algebraic fibres. These fibres are geometrically regular by the cited comparison, hence reduced: every regular local ring is a domain by [StacksAlg] (Tag 00NP). Therefore RR is reduced.

Let V=M(B)V=\mathcal M(B) be a strict affinoid domain and let m\mathfrak m be any maximal ideal of BB. Theorem 1.12.3 realizes it as a rigid point vv. Its residue field is finite over KK, so its residue field in degree one is finite over K~\widetilde K by Lemma 3.24.2. Its canonical reduction in B~\widetilde B is consequently a closed point, and Theorem 3.25.1 makes vv an absolute inner point of VV. Theorem 2.44.1 implies that it is inner for V↪M(A)V\hookrightarrow\mathcal M(A). Proposition 2.44.5 therefore makes VV an ordinary neighborhood of vv in M(A)\mathcal M(A). Thus OV,v\mathcal O_{V,v} is the reduced local ring just considered. The map Bm→OV,vB_{\mathfrak m}\to\mathcal O_{V,v} is faithfully flat by Proposition 2.33.16 and domain flatness. It is injective, so BmB_{\mathfrak m} is reduced. This holds for all maximal ideals of BB, and hence BB is reduced. ◻

Lemma 2.60.2. NAG-001880 (Nonrigid ordinary germs on strict curves). Let TT be a reduced good strictly KK-analytic space of dimension at most one. At every nonrigid point tt, OT,t\mathcal O_{T,t} is a field.

Proof. Choose a reduced strict affinoid neighborhood of tt. By Lemmas 2.35.25 and 2.60.1, every germ at tt can be represented on a reduced strict affinoid neighborhood V=M(B)V=\mathcal M(B). By Corollary 3.22.8, dim⁡B≤1\dim B\leq 1. Its point kernel p=ker⁡(t)\mathfrak p=\ker(t) cannot be maximal, since every strict affinoid maximal point has finite residue field over KK, whereas tt is not rigid. It is therefore a minimal prime: a strictly smaller prime and a strictly larger maximal prime would give a chain of length two. The localization BpB_{\mathfrak p} is a reduced ring with exactly one prime, and hence is a field.

Let a germ hh vanish at tt, and represent it by b∈Bb\in B on such a neighborhood. Then b∈pb\in\mathfrak p and b/1=0b/1=0 in the field BpB_{\mathfrak p}. There exists c∉pc\notin\mathfrak p with cb=0cb=0. The germ of cc is a unit, since c(t)≠0c(t)\ne 0, so h=0h=0. Thus the maximal ideal of OT,t\mathcal O_{T,t} is zero, as required. ◻

Lemma 2.60.3. NAG-001881 (An outer collar at the Gauss point). Let D={∣T∣≤1}⊂AK1,anD=\{|T|\leq 1\}\subset\mathbb A_K^{1,\mathrm{an}} and let η\eta be its Gauss point. For every ordinary open neighborhood UU of η\eta, there is R>1R>1 with DR={∣T∣≤R}⊂D∪UD_R=\{|T|\leq R\}\subset D\cup U. If ff is analytic on DD and an analytic function on UU agrees with it on U∩DU\cap D, then ff extends analytically to some such DRD_R.

Proof. Choose R0>1R_0>1 and work inside the closed disc DR0D_{R_0}. Polynomials are uniformly dense in its affinoid algebra. The evaluation topology and this uniform approximation imply that finitely many strict inequalities involving polynomial absolute values define a neighborhood of η\eta contained in U∩DR0U\cap D_{R_0}. For a nonzero polynomial P=∑i=0maiTiP=\sum_{i=0}^m a_iT^i, put M=max⁡i∣ai∣M=\max_i|a_i| and choose the largest index jj with ∣aj∣=M|a_j|=M. For every s>1s>1 sufficiently close to one, ∣ai∣si<Msj(i≠j).|a_i|s^i<Ms^j\qquad(i\ne j). Consequently every analytic point zz with ∣T(z)∣=s|T(z)|=s has ∣P(z)∣=Msj|P(z)|=Ms^j. This tends to M=∣P(η)∣M=|P(\eta)| as s↓1s\downarrow 1, uniformly over all such zz. Apply this observation to the finitely many polynomials in the chosen inequalities. For some 1<R<R01<R<R_0, all points with 1<∣T∣≤R1<|T|\leq R belong to UU, proving the first assertion.

The disc DD is a pure one-dimensional strict affinoid, and its Shilov boundary is {η}\{\eta\}. By Theorem 3.25.4, its absolute boundary is also {η}\{\eta\}. The affine line is boundaryless over KK: a disc of radius strictly greater than ∣T(z)∣|T(z)| is an inner presentation at each of its points zz. Theorem 2.44.1 and Proposition 2.44.5 show that every point of D∖{η}D\setminus\{\eta\} has an ordinary neighborhood in the affine line contained in DD. Together with UU this proves that the two analytic domains DD and UU form an admissible covering of D∪UD\cup U. The structure sheaf glues the given functions, and restriction to DRD_R gives the required extension. ◻

Theorem 2.60.4. NAG-001882 (The ambient germ of an exact-radius graph). Let f(T)=∑n≥0anTn∈K⟨T⟩f(T)=\sum_{n\geq 0}a_nT^n\in K\langle T\rangle have convergence radius exactly one. Put D=M(K⟨T⟩)D=\mathcal M(K\langle T\rangle), let η\eta be its Gauss point, and set x=(η,f(η))x=(\eta,f(\eta)) in AK2,an\mathbb A_K^{2,\mathrm{an}}. Then OAK2,an,x\mathcal O_{\mathbb A_K^{2,\mathrm{an}},x} is a field.

Proof. Step 1: a hypothetical vanishing germ gives a curve. Write pp for the first projection, put X=p−1(D)X=p^{-1}(D), and let Z⊂XZ\subset X be the graph of ff, with isomorphism p∣Z:Z≃Dp|_Z:Z\simeq D. Its point xx has H(x)=H(η)\mathcal H(x)=\mathcal H(\eta): the evaluated first coordinate already has dense fraction field in H(η)\mathcal H(\eta), and the second coordinate belongs to that field. Thus dK(x)=1d_K(x)=1, by Lemma 3.22.4, and xx is not rigid.

Suppose a nonzero ordinary germ hh at xx vanishes there. Choose a strict affinoid neighborhood V=M(A)V=\mathcal M(A) representing it. By Theorem 2.35.23 and Lemma 2.52.2, we may retain the open-and-closed integral component containing xx. The element h∈Ah\in A remains nonzero. This domain has dimension at most two, by Corollary 3.22.8 applied inside a sufficiently large strict bidisc. Therefore the reduced closed subspace W=M(A/(h))W=\mathcal M(A/\sqrt{(h)}) has dimension at most one: every prime containing hh is nonzero, and a chain in that quotient can be preceded by the zero prime of AA. It contains xx, so its dimension at xx is one by the Abhyankar inequality.

The pullback of hh to the curve DD vanishes at its nonrigid point η\eta. By Lemma 2.60.2 its germ there is zero. The same is true of each generator of (h)\sqrt{(h)}, since its suitable power is a multiple of hh and the graph is reduced. After shrinking VV, we therefore have a closed immersion Z∩V⊂W∩XZ\cap V\subset W\cap X. The latter is a reduced strict domain of the curve WW, by Lemma 2.60.1; both its ordinary local ring at xx and that of ZZ are fields by the curve lemma. Their quotient map is an isomorphism. The defining coherent ideal consequently has zero stalk at xx, so its finitely many local generators vanish on a smaller ordinary neighborhood in XX. Restricting VV to a strict ambient neighborhood with that intersection gives W∩X=Z∩VW\cap X=Z\cap V as closed analytic subspaces of V∩XV\cap X.

Step 2: the projection is locally finite of rank one. For every point ww of WW lying above η\eta, the residue-field tower formula and the affinoid dimension bound give 1+dH(η)(w)=dK(w)≤1.1+d_{\mathcal H(\eta)}(w)=d_K(w)\leq1. Here the tower formula is Lemma 3.22.2 and the last inequality is Theorem 3.22.7. The fibre WηW_\eta is a strict H(η)\mathcal H(\eta)-affinoid: in a strict presentation of WW, scalar extension and the equation fixing its first coordinate give such a presentation. The same maximum formula makes it zero-dimensional. Theorem 1.12.2 and Proposition 3.21.1 then make its algebra finite-dimensional over H(η)\mathcal H(\eta), so its analytic point set is finite and discrete by Lemma 2.35.1. In particular xx is isolated in this fibre.

The projection AK2,an→AK1,an\mathbb A^{2,\mathrm{an}}_K\to \mathbb A^{1,\mathrm{an}}_K is boundaryless: near each point use a relative disc in the second coordinate with radius strictly larger than its value. The inclusion of VV is inner at xx because VV is an ordinary neighborhood, and the closed immersion of WW is finite and boundaryless. The forward composition rule applied on these affinoid neighborhoods therefore makes p∣Wp|_W inner at xx. Theorem 2.49.1 gives ordinary neighborhoods on which p∣Wp|_W is finite. Separate the branch through xx and shrink the target about η\eta as in Lemma 2.35.1. The ordinary target ring at η\eta is a field by the curve lemma. The finite direct image is therefore flat at that point; Lemma 2.35.16 makes it free of some positive rank mm after shrinking. We obtain affinoid neighborhoods W0W_0 of xx in WW and UU of η\eta in the affine line for which p:W0→Up:W_0\to U is finite free of rank mm.

Base change to U∩DU\cap D gives a finite free map of the same rank from W0∩XW_0\cap X. This source is an analytic domain of ZZ, by the closed-subspace equality proved above. Under Z≃DZ\simeq D it maps isomorphically onto its image. A positive-rank finite free algebra has nonzero fibres, so the image is all of U∩DU\cap D. Consequently this base change is an isomorphism and m=1m=1. A unital finite free algebra of rank one is the base algebra: its unit is a basis after every residue-field specialization, so the unit map is an isomorphism by Nakayama and finite presentation. Thus p:W0≃Up:W_0\simeq U.

Step 3: contradict the exact radius. The second coordinate of the inverse section U→W0U\to W_0 is an analytic function agreeing with ff on U∩DU\cap D. Restrict to the ordinary interior of UU at η\eta and apply Lemma 2.60.3. This extends ff to DRD_R for some R>1R>1. Restriction K{R−1T}→K⟨T⟩K\{R^{-1}T\}\to K\langle T\rangle preserves the power-series coefficients, so such an extension implies ∣an∣Rn→0|a_n|R^n\to 0, contrary to the convergence-radius hypothesis. Hence no nonzero germ at xx can vanish there. The maximal ideal of the ordinary local ring is zero, proving the theorem. ◻

Corollary 2.60.5. NAG-001883 (Domain base change can destroy naive flatness). Choose π∈K\pi\in K with 0<∣π∣<10<|\pi|<1 and set f(T)=∑n≥0π⌊n⌋Tn,ϕ:D⟶AK2,an,t⟼(t,f(t)).f(T)=\sum_{n\geq0}\pi^{\lfloor\sqrt n\rfloor}T^n, \qquad\phi:D\longrightarrow\mathbb A_K^{2,\mathrm{an}}, \qquad t\longmapsto(t,f(t)). The morphism ϕ\phi is naively flat at the unit Gauss point η\eta. Its base change by the analytic domain X={∣T1∣≤1}X=\{|T_1|\leq 1\} is not naively flat at that point.

Proof. The coefficients tend to zero, whereas their nnth-root absolute values tend to one. Thus f∈K⟨T⟩f\in K\langle T\rangle has convergence radius exactly one. Theorem 2.60.4 makes the ordinary target local ring at x=ϕ(η)x=\phi(\eta) a field. Every module over a field is flat, so ϕ\phi is naively flat at η\eta.

Because ϕ\phi factors through the domain XX, its base change to XX is its closed graph immersion, with equation g=T2−f(T1)g=T_2-f(T_1). The ordinary local ring R=OX,xR=\mathcal O_{X,x} is a domain by Theorem 2.35.23 and [StacksAlg] (Tag 00NP), using an affinoid neighborhood of xx in XX. The germ gg is nonzero: differentiation in the second coordinate gives ∂g/∂T2=1\partial g/\partial T_2=1. This derivation on convergent series restricts to the germ by Theorem 2.41.3 and Theorem 2.41.5. It vanishes at xx, so R/(g)R/(g) is nonzero. Tensoring the injective multiplication map R→gRR\xrightarrow{g}R with R/(g)R/(g) gives the zero map on the nonzero module R/(g)R/(g). This quotient is therefore not RR-flat. The graph local ring is exactly R/(g)R/(g) by the coherent closed-subspace restriction formula, proving the assertion. ◻

2.61 Flat comparison with analytic fibres

Lemma 2.61.1. NAG-001813 (Finite flat maps are ordinary-open). A finite flat morphism between good analytic spaces is open for the ordinary topology.

Proof. Let f:Y→Xf:Y\to X be finite flat, let y∈Oy\in O with OO ordinary open in YY, and put x=f(y)x=f(y). Lemma 2.35.1 gives disjoint clopen branches over an ordinary neighborhood UU of xx, with the branch through yy contained in OO: in its construction choose the separating neighborhood of yy inside OO before removing the compact image of the complement of the separating neighborhoods. Choose an affinoid ordinary neighborhood V=M(A)V=\mathcal M(A) of xx in UU, and write f−1(V)=M(B)f^{-1}(V)=\mathcal M(B). Set R=OX,xR=\mathcal O_{X,x}. By the same lemma, B⊗AR≃∏f(yi)=xOY,yiB\otimes_A R\simeq\prod_{f(y_i)=x}\mathcal O_{Y,y_i} as RR-algebras. The identity base change in the definition of analytic flatness gives RR-flatness at each yiy_i. The finite product is therefore flat over RR. The affinoid algebra AA is Noetherian by Theorem 1.18.2, so the finite AA-module BB has a finite presentation. Tensoring the presentation with RR gives a finite presentation of B⊗ARB\otimes_A R. Since RR is local, [StacksAlg] (Tag 00NX(1) and (5)) makes B⊗ARB\otimes_A R free over RR.

Fix such a finite AA-presentation of BB. Proposition 2.33.16 expresses RR as the filtered colimit of the actual algebras AWA_W of rational ordinary neighborhoods of xx in VV. Proposition 2.33.19 identifies the finite inverse-image algebra there with BW=B⊗AAWB_W=B\otimes_A A_W. Represent a basis of B⊗ARB\otimes_A R in one BWB_W, and represent the inverse map on the finitely many generators of the chosen presentation over the same AWA_W, after shrinking. The finitely many relation equations hold after a further shrinking, so the inverse data define an AWA_W-linear map. Both inverse identities on the two finite lists of generators also hold after shrinking. Thus BWB_W is free over AWA_W.

The branch through yy gives an idempotent factor of BWB_W by Lemma 2.35.1. This factor is finite flat over AWA_W, with its canonical finite-module topology. Lemma 2.44.2 makes its image open and closed in WW. The image contains xx and is contained in f(O)f(O). Intersecting it with the ordinary interior of WW gives an ordinary neighborhood of xx in XX contained in f(O)f(O). ◻

Proposition 2.61.2. NAG-001814 (Spreading a closed neighborhood of a generic fibre). Let Y→XY\to X be a morphism of good analytic spaces over any complete field, smooth at yy, and put x=f(y)x=f(y). If OX,x\mathcal O_{X,x} is Artinian, then every Zariski-closed subset of YY containing an ordinary neighborhood of yy in YxY_x contains an ordinary neighborhood of yy in YY.

Proof. First reduce the base. On an ordinary affinoid neighborhood M(A)\mathcal M(A) of xx, put p=ker⁡(x)\mathfrak p=\ker(x), D=ApD=A_{\mathfrak p} and R=OX,xR=\mathcal O_{X,x}. The faithfully flat map D→RD\to R and the Artinian hypothesis imply that DD is zero-dimensional. Its reduction is a field. If NN is the nilradical of AA, Theorem 2.35.45, after quotienting by the nilradical of DD, makes R/NRR/NR geometrically regular over that field. It is also local Artinian, hence is a field. The nilpotent closed base change Xred→XX_{\rm red}\to X and its pullback to YY are homeomorphisms and do not alter the analytic fibre. Thus it suffices to treat the case in which RR is a field.

Corollary 2.35.47 gives a normal ordinary neighborhood of xx. Normality ascends along the smooth map by smooth normality ascent. Choose an ordinary affinoid neighborhood Y0Y_0 of yy and its normal integral component through yy, using Lemma 2.56.1. Write the closed subset there as V(h1,…,hs)V(h_1,\ldots,h_s). The identity principle reduces the assertion to finding a nonempty ordinary open subset of Y0Y_0 on which every hjh_j vanishes. Restrict inside the ordinary interior of Y0Y_0 and the locus where the relative dimension of the smooth map is constant.

Theorem 2.55.7 applies to the nonempty ordinary fibre open set in the hypothesis. It gives a finite flat base change X′→UX'\to U, with a unique point x′x' above xx and reduced ordinary closed fibre there, and a nonempty ordinary open product patch in Y′=Y×XX′Y'=Y\times_XX'. Its factors are an ordinary base neighborhood U′U' of x′x', an open annulus DD or a point, and an open polydisc Δ\Delta; its fibre at x′x' lies in the given closed subset. The ordinary ring R′=OX′,x′R'=\mathcal O_{X',x'} is finite over the field RR, by finite branch separation and the uniqueness of the fibre point. Its ordinary closed fibre is itself, and is reduced. Thus R′R' is a field. Apply Corollary 2.35.47 at x′x' and then Lemma 2.56.1 to choose a normal integral affinoid ordinary neighborhood M(A′)⊂U′\mathcal M(A')\subset U'. The map A′⟶R′⟶H(x′)A'\longrightarrow R'\longrightarrow\mathcal H(x') is injective: its first map is flat and nonzero with integral source, and its second is the evaluation of a field germ.

Restrict each hjh_j to M(A′)×KD×KΔ\mathcal M(A')\times_KD\times_K\Delta. The weighted Laurent and Tate expansions on closed subannuli and closed subpolydiscs give unique coefficients in A′A'. They agree under all restrictions, so we may write hj=∑(a,I)∈Zm×Nn−mca,I,jSaTI,m∈{0,1}.h_j=\sum_{(a,I)\in\mathbf Z^m\times\mathbf N^{n-m}} c_{a,I,j}S^aT^I,\qquad m\in\{0,1\}. The fibre function vanishes at every point of DH(x′)×ΔH(x′)D_{\mathcal H(x')}\times\Delta_{\mathcal H(x')}. Evaluation at its Gauss points, with radii in the corresponding intervals, makes every coefficient ca,I,j(x′)c_{a,I,j}(x') zero. The displayed injection therefore makes all coefficients zero in A′A' simultaneously. Every hjh_j vanishes on the whole product, and hence on a nonempty ordinary open subset of Y′Y'. By Lemma 2.61.1, its image in Y0Y_0 is nonempty and ordinary-open. The functions vanish on this image, and the identity principle makes them zero throughout Y0Y_0. This proves the assertion. ◻

Lemma 2.61.3. NAG-001815 (A nonzero ideal lowers the strict fibre dimension). Let XX be good over any complete field and let VV be an ordinary affinoid neighborhood of tt in AXd,an\mathbb A_X^{d,\mathrm{an}}. Put x=f(t)x=f(t) and assume that OX,x\mathcal O_{X,x} is a field and F=H(x)F=\mathcal H(x) is nontrivially valued. For every nonzero proper ideal I⊂OV,tI\subset\mathcal O_{V,t}, its closed-subspace germ has an inner ordinary affinoid representative over XX with strict fibre algebra of Krull dimension at most d−1d-1. Its ordinary and fibre stalks are respectively OV,t/I,OVx,t/IOVx,t.\mathcal O_{V,t}/I, \qquad \mathcal O_{V_x,t}/I\mathcal O_{V_x,t}.

Proof. Represent a finite generating family of II and a nonzero h∈Ih\in I on an ordinary affinoid neighborhood. Its fibre germ is nonzero: otherwise its zero set contains an ordinary fibre neighborhood, and Proposition 2.61.2, applied on the ordinary smooth interior of VV, would then give an ordinary neighborhood on which hh vanishes at every point. By smooth normality ascent and Corollary 2.35.47, take a reduced ordinary affinoid neighborhood of tt. The spectral-radius and nilradical formulas in Proposition 1.20.3 and Proposition 3.21.4 make the germ of hh zero. Proposition 2.51.3 gives an ordinary representative with strict fibre CC. This fibre is an affinoid domain of AFd\mathbb A_F^d, so Theorem 2.35.23 and Lemma 2.52.2 make CC a finite product of regular domains. Lemma 2.56.4 retains the factor C0C_0 through tt over the original ground field. The image of hh in C0C_0 is nonzero, and Lemma 2.56.3 gives dim⁡C0≤d\dim C_0\leq d.

The chosen generators define a closed subspace with strict fibre algebra C0/ICC_0/I_C. Every prime chain in this quotient lifts to a chain containing the nonzero ideal ICI_C in the domain C0C_0; adjoining its zero prime increases the length by one. Thus the quotient has dimension at most d−1d-1. Finite closed-subspace base change gives the two stalk identifications. The closed immersion is inner, and the chosen ambient ordinary neighborhood is inner over XX, so their composite is inner as well. ◻

Theorem 2.61.4. NAG-001816 (Flatness along a nontrivially valued inner generic fibre). Let f:Y→Xf:Y\to X be a morphism of good analytic spaces over any complete field, let y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X), and put x=f(y)x=f(y). If OX,x\mathcal O_{X,x} is a field and H(x)\mathcal H(x) is nontrivially valued, then OY,y→OYx,y\mathcal O_{Y,y}\to\mathcal O_{Y_x,y} is faithfully flat.

Proof. Induct on an upper bound nn for the Krull dimension of a strict fibre algebra of an ordinary affinoid representative of the source germ. Such a representative exists by Proposition 2.51.3. For a representative of fibre dimension d≤nd\leq n, Theorem 2.51.4 gives, on ordinary neighborhoods, a finite map to an ordinary affinoid neighborhood VV of a point tt in AXd\mathbb A_X^d. Separate the finite branches and retain the one through yy, so that yy is the only point above tt. Put R0=OV,t,T0=OVx,t,S=OY,y,T=OYx,y.R_0=\mathcal O_{V,t},\quad T_0=\mathcal O_{V_x,t},\quad S=\mathcal O_{Y,y},\quad T=\mathcal O_{Y_x,y}. The finite direct-image stalk formulas give SS finite over R0R_0 and T=S⊗R0T0T=S\otimes_{R_0}T_0. This is an ordinary tensor identity: take finite affinoid base change, where completed and ordinary tensor products agree, and then take ordinary stalks. The unique point over tt means that the two finite stalk products each have one factor. Therefore flatness of R0→T0R_0\to T_0 suffices.

The map from the ordinary interior of VV to XX is smooth. Since the base ordinary ring is a field, smooth normality ascent makes R0R_0 a normal local domain. If it is a field, its map to T0T_0 is flat. This includes n=0n=0, for then d=0d=0 and R0=OX,xR_0=\mathcal O_{X,x}. Otherwise every power mq\mathfrak m^q of its maximal ideal is nonzero. Lemma 2.61.3 represents the closed germ defined by mq\mathfrak m^q by an inner affinoid with strict fibre dimension at most d−1≤n−1d-1\leq n-1. Induction gives flatness of R0/mq⟶T0/mqT0R_0/\mathfrak m^q\longrightarrow T_0/\mathfrak m^qT_0 for all q≥1q\geq 1. Both rings R0,T0R_0,T_0 are Noetherian and the extended ideal lies in the maximal ideal of T0T_0. The adic criterion [StacksAlg] (Tag 0523) makes T0T_0 flat over R0R_0. Base change gives flatness of S→TS\to T. This is a local map of nonzero local rings, so it is faithfully flat. The argument applies to every representative with strict fibre dimension at most nn, completing the induction. ◻

Lemma 2.61.5. NAG-001817 (A generic Gauss field germ in arbitrary characteristic). Let XX be good over any complete field, with OX,x\mathcal O_{X,x} a field and F=H(x)F=\mathcal H(x) trivially valued. If 0<r<10<r<1 and x∗x^* is the radius-rr Gauss point above xx in X×KAK1X\times_K\mathbb A_K^1, then H(x∗)=Fr\mathcal H(x^*)=F_r and OX×KAK1,x∗\mathcal O_{X\times_K\mathbb A_K^1,x^*} is a field.

Proof. The Gauss completed-residue calculation gives H(x∗)=Fr\mathcal H(x^*)=F_r. To test the kernel of ordinary evaluation, represent a germ hh on a product neighborhood U×DU\times D supplied by Lemma 2.55.1, with DD an open annulus around radius rr. Corollary 2.35.47 and Lemma 2.56.1 give a normal integral affinoid ordinary neighborhood M(A)⊂U\mathcal M(A)\subset U of xx. Its flat nonzero map to the field OX,x\mathcal O_{X,x} and then to FF is injective. On a closed subannulus about rr, write h=∑j∈ZajTjh=\sum_{j\in\mathbf Z}a_jT^j with aj∈Aa_j\in A. If h(x∗)=0h(x^*)=0, the Gauss norm formula forces all aj(x)=0a_j(x)=0, so the single injection A↪FA\hookrightarrow F makes every coefficient zero. Thus hh is the zero germ. The kernel of ordinary evaluation is the maximal ideal, so that maximal ideal is zero and the ordinary local ring is a field. This coefficient argument uses no restriction on characteristic. ◻

Theorem 2.61.6. NAG-001818 (Faithful flatness along every inner generic fibre). Let f:Y→Xf:Y\to X be a morphism of good analytic spaces over any complete field. If y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X) and OX,f(y)\mathcal O_{X,f(y)} is a field, then OY,y⟶OYf(y),y\mathcal O_{Y,y}\longrightarrow\mathcal O_{Y_{f(y)},y} is faithfully flat.

Proof. The case of nontrivially valued H(x)\mathcal H(x), where x=f(y)x=f(y), is Theorem 2.61.4. Otherwise choose an affinoid domain UU which is an ordinary neighborhood of xx. The test UU in Definition 2.44.4 provides an affinoid domain VV which is an ordinary neighborhood of yy in Y×XUY\times_X U, with y∈Int⁡(V/U)y\in\operatorname{Int}(V/U). Since xx is in the ordinary interior of UU, the source and base ordinary germs and the analytic fibre germ are unchanged. Replace Y,XY,X by V,UV,U. Choose 0<r<10<r<1 and form their simultaneous affine-line products Y∗→X∗Y^*\to X^*. These spaces are good by Theorem 2.48.3. Take the radius-rr Gauss points y∗,x∗y^*,x^*; the Gauss seminorms are compatible under the isometric inclusion of completed residue fields. Lemma 2.61.5 makes OX∗,x∗\mathcal O_{X^*,x^*} a field and H(x∗)\mathcal H(x^*) nontrivially valued. For every affinoid domain D⊂X∗D\subset X^* containing x∗x^*, the test inverse image Y∗×X∗D=V×UDY^*\times_{X^*}D=V\times_U D is affinoid by Theorem 2.40.11. Proposition 2.33.21, applied to the affinoid diagram V→U←DV\to U\leftarrow D, puts y∗y^* in the relative interior of that entire test over DD. Thus y∗∈Int⁡(Y∗/X∗)y^*\in\operatorname{Int}(Y^*/X^*) by the same definition, and the nontrivial case applies. With S=OY,y,S∗=OY∗,y∗,T=OYx,y,T∗=OYx∗∗,y∗,S=\mathcal O_{Y,y},\quad S^*=\mathcal O_{Y^*,y^*},\quad T=\mathcal O_{Y_x,y},\quad T^*=\mathcal O_{Y^*_{x^*},y^*}, the map S∗→T∗S^*\to T^* is flat. Around the Gauss point, the projection Y∗→YY^*\to Y is the projection from the ordinary interior of a relative closed disc of radius greater than rr. Theorem 2.48.3 therefore makes S→S∗S\to S^* flat. Finally Yx∗∗=(Yx)H(x)rY^*_{x^*}=(Y_x)_{\mathcal H(x)_r}, and Theorem 2.35.4 makes T→T∗T\to T^* faithfully flat. Thus the composite S→T∗S\to T^* is flat, and faithful-flat descent along T→T∗T\to T^* gives flatness of S→TS\to T. All maps are local and the rings are nonzero, proving faithful flatness. ◻

Corollary 2.61.7. NAG-001819 (Ordinary closed fibres map faithfully flat to analytic fibres). Let f:Y→Xf:Y\to X be a morphism of good analytic spaces over any complete field and let y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X) with image xx. Write R=OX,xR=\mathcal O_{X,x}, S=OY,yS=\mathcal O_{Y,y} and let m\mathfrak m be the maximal ideal of RR. Then the canonical local homomorphism S/mS⟶OYx,yS/\mathfrak mS\longrightarrow\mathcal O_{Y_x,y} is faithfully flat.

Proof. Choose an ordinary affinoid neighborhood U=M(A)U=\mathcal M(A) of xx on which a finite generating family of m\mathfrak m is represented by a1,…,aqa_1,\dots,a_q. Use UU in Definition 2.44.4 to choose an ordinary affinoid neighborhood V=M(B)V=\mathcal M(B) of yy in Y×XUY\times_X U with y∈Int⁡(V/U)y\in\operatorname{Int}(V/U). The ordinary stalks on U,VU,V at these points are R,SR,S. Put J=(a1,…,aq)J=(a_1,\dots,a_q), U0=M(A/J)U_0=\mathcal M(A/J) and V0=M(B/JB)V_0=\mathcal M(B/JB), with their quotient norms. The finite quotient and base-change formulas in Propositions 2.33.18 and 2.33.19 identify V0V_0 with V×UU0V\times_U U_0. Each closed immersion has just one point above its indicated point, since the point seminorm kills the defining ideal and factors through the quotient. Write these points as x0,y0x_0,y_0. Lemma 2.35.1 gives their ordinary stalks R/mR/\mathfrak m and S/mSS/\mathfrak m S. The analytic fibre germ is unchanged: H(x0)=H(x)\mathcal H(x_0)=\mathcal H(x) by quotient evaluation, and aia_i has zero image in this field for every ii, so the two affinoid fibre algebras identified by Theorem 2.33.11 coincide under finite base change. For each affinoid domain D⊂U0D\subset U_0 containing x0x_0, its test inverse image is the affinoid V0×U0D=V×UDV_0\times_{U_0}D=V\times_U D. The forward inclusion of Proposition 2.33.21 on V→U←DV\to U\leftarrow D makes y0y_0 interior over DD in that entire test. Thus y0∈Int⁡(V0/U0)y_0\in\operatorname{Int}(V_0/U_0) by the same definition. Apply Theorem 2.61.6 to V0→U0V_0\to U_0 at y0y_0, with the field stalk R/mR/\mathfrak m. Under the stated stalk and fibre identifications its faithfully flat map is the canonical map in the assertion. ◻

Lemma 2.61.8. NAG-001820 (Naive flatness and a trivial-residue Gauss lift). For a morphism Y→XY\to X of good analytic spaces over any complete field, let F\mathcal F be coherent and naively XX-flat at yy, and suppose H(x)\mathcal H(x) is trivially valued for x=f(y)x=f(y). Its pullback under the simultaneous affine-line base change is naively X∗X^*-flat at the radius-rr Gauss lift y∗y^* for any 0<r<10<r<1.

Proof. Put R=OX,xR=\mathcal O_{X,x}, S=OY,yS=\mathcal O_{Y,y}, R∗=OX∗,x∗R^*=\mathcal O_{X^*,x^*}, S∗=OY∗,y∗S^*=\mathcal O_{Y^*,y^*} and M=FyM=\mathcal F_y. Choose relative closed discs of radius greater than rr around the two Gauss points. Their ordinary stalks agree with those of the affine-line products, so Theorem 2.48.3 makes R∗R^* flat over RR and S∗S^* flat over SS. Coherent pullback gives M∗=M⊗SS∗M^*=M\otimes_SS^*; tensoring injections twice shows that M∗M^* is RR-flat. Put I=mRR∗I=\mathfrak m_RR^*. Represent a finite generating family of mR\mathfrak m_R on an ordinary affinoid neighbourhood of xx, and let X0X_0 be the closed quotient it defines. The ordinary quotient stalk is OX0,x=R/mR\mathcal O_{X_0,x}=R/\mathfrak m_R. Its completed residue field at xx is still H(x)\mathcal H(x), since all generators vanish at xx. On the simultaneous affine-line space the pullback closed quotient is X0∗=X0×KAK1,anX_0^*=X_0\times_K\mathbb A_K^{1,\mathrm{an}}. On the actual affinoid product charts its ideal is the extension of the original ideal, by Theorem 2.40.11 and Corollary 3.27.2. Since the generators evaluate to zero at xx, their pullbacks also evaluate to zero at x∗x^*. The Gauss formula on the quotient coefficients gives that same point in X0∗X_0^*. The local ring of this quotient at the radius-rr Gauss point is R∗/IR^*/I. These ordinary quotient stalk identities are the single-point finite stalk formula of Lemma 2.35.1, applied to the affinoid quotient maps with their quotient norms. By Lemma 2.61.5 on the good space X0X_0, this ordinary ring is a field. Thus II is proper and M∗/IM∗M^*/IM^* is R∗/IR^*/I-flat.

Tensor a free RR-resolution of R/mRR/\mathfrak m_R with the flat algebra R∗R^*. It is a free resolution of R∗/IR^*/I, and its tensor with M∗M^* has no positive homology because M∗M^* is RR-flat. Hence Tor⁡1R∗(M∗,R∗/I)=0\operatorname{Tor}^{R^*}_1(M^*,R^*/I)=0. The rings R∗,S∗R^*,S^* are Noetherian and M∗M^* is finite over S∗S^*. The exact local criterion [StacksAlg] (Tag 00ML) makes M∗M^* flat over R∗R^*, as required. ◻

References. The closed-neighborhood argument and the flat generic-fibre comparison reconstruct the mechanisms of [Duc18] (Proposition 6.3.1 and Theorem 6.3.3). The induction needs faithful flatness, and uses the exact adic criterion [StacksAlg] (Tag 0523). The trivial-valuation passage is the auxiliary Gauss construction established above; its module version uses [StacksAlg] (Tag 00ML).

2.62 Inner coherent flatness over arbitrary complete fields

Lemma 2.62.1. NAG-001822 (Lifting analytic-fibre injectivity at an inner point). Let T→XT\to X be a morphism of good analytic spaces over any complete field, let t∈Int⁡(T/X)t\in\operatorname{Int}(T/X) have image xx, and let u:L→Nu:L\to N be a morphism of coherent sheaves. If NtN_t is flat over OX,x\mathcal O_{X,x} and the analytic-fibre map is injective at tt, then utu_t is injective and Coker⁡(u)t\operatorname{Coker}(u)_t is flat over OX,x\mathcal O_{X,x}.

Proof. Write R=OX,xR=\mathcal O_{X,x}, S=OT,tS=\mathcal O_{T,t} and m=mR\mathfrak m=\mathfrak m_R. Coherent pullback identifies the fibre map with ut⊗SOTx,tu_t\otimes_S\mathcal O_{T_x,t}. Corollary 2.61.7 gives a faithfully flat local map S/mS→OTx,tS/\mathfrak mS\to\mathcal O_{T_x,t}. Its faithful flatness detects the kernel of the finite map Lt/mLt→Nt/mNtL_t/\mathfrak mL_t\to N_t/\mathfrak mN_t, so this map is injective. The ring SS is Noetherian, LtL_t is finite over SS, and NtN_t is RR-flat. Apply [StacksAlg] (Tag 00ME): it gives injectivity of utu_t and RR-flatness of its cokernel. Ordinary stalks commute with coherent cokernels, proving the statement. ◻

NAG-001823 · Open at full scope

Theorem 2.62.2. NAG-001823 (Inner flatness with nontrivially valued completed residue field). Let f:Y→Xf:Y\to X be a morphism of good analytic spaces over any complete field and let F\mathcal F be coherent on YGY_G. At an inner point yy, with H(f(y))\mathcal H(f(y)) nontrivially valued, naive XX-flatness of F\mathcal F is equivalent to analytic XX-flatness.

Proof. The identity base-change test proves the analytic-to-naive implication. Assume that Fy\mathcal F_y is naively flat, and put x=f(y)x=f(y) and F=H(x)F=\mathcal H(x). If the stalk is zero, a finite set of local generators vanishes on one ordinary neighborhood of yy. Every pullback then vanishes near every point above yy, giving the conclusion. We may assume the stalk is nonzero.

Step 1: a strict fibre-support induction. Choose an ordinary affinoid representative with strict FF-fibre, using Proposition 2.51.3. Replace its source by the closed support of F\mathcal F, defined by its annihilator. The sheaf is a finite module over this closed subspace, and its pushforward is the original sheaf. Theorem 2.35.17 identifies both naive and analytic flatness under this replacement. The closed immersion is finite and inner, so the new source is still inner over XX at yy. Its fibre is a strict affinoid quotient. Lemma 2.57.1 identifies its point set with the support of Fx\mathcal F_x. Induct on an upper bound nn for this strict fibre dimension, allowing every such representative with dimension at most nn. By Corollary 3.22.8, later restrictions to ordinary affinoid domains cannot increase this analytic support dimension. For a strict affinoid fibre it equals the Krull dimension.

Step 2: a finite projection to an integral fibre. Write the chosen closed support as ZZ and put d=dim⁡Zx≤nd=\dim Z_x\leq n. Theorem 2.51.4 gives a finite map from an ordinary neighborhood of yy in ZZ to an ordinary affinoid neighborhood TT of tt in AXd,an\mathbb A_X^{d,\mathrm{an}}. Finite branch separation retains the branch through yy and, after shrinking the target, makes yy its only point above tt. Since tt is in the ordinary interior of TT in relative affine space, it is inner over XX. Shrink by Proposition 2.51.3 to make TxT_x strict. This fibre is an affinoid domain of AFd\mathbb A_F^d; its ordinary rings are regular by Theorem 2.35.23. Lemma 2.52.2 decomposes its algebra into integral factors. Lemma 2.56.4 retains the factor containing tt by an actual ordinary domain over the original field. Pull the finite map back after each restriction. We obtain π:Z⟶T=M(B),C=B⊗^AF\pi:Z\longrightarrow T=\mathcal M(B), \qquad C=B\widehat\otimes_A F with singleton distinguished fibre, with CC integral and strict, and with dim⁡C≤d\dim C\leq d by Lemma 2.56.3. Here M(A)\mathcal M(A) is the chosen ordinary affinoid base neighborhood. The source and target ordinary germs at the distinguished points have not changed.

Step 3: an original generic basis and a smaller quotient. Put N=π∗FN=\pi_*\mathcal F and represent it by a finite BB-module. The singleton finite-pushforward formula makes NtN_t naively XX-flat. Fix any finite generating family of this BB-module. Its images span NC⊗CFrac⁡(C)N_C\otimes_C\operatorname{Frac}(C). Choose a subset which is a basis, allowing the empty subset when its rank is zero, and use the corresponding original generators to define u:Br→Nu:B^r\to N. Lemma 2.57.2 proves that uCu_C is injective, has ordinarily dense bijectivity locus, and that its cokernel QCQ_C has support of dimension at most dim⁡C−1\dim C-1 unless it is zero. Here injection follows because its kernel is a torsion submodule of the free module CrC^r over a domain. The dimension drop follows by annihilating the finite torsion cokernel by one nonzero element of CC and prepending the zero prime to each chain in its support. Thus the generators and the map uu are defined over the original affinoid algebra, not merely over the completed residue field.

Let L=OTrL=\mathcal O_T^r and Q=Coker⁡(u:L→N)Q=\operatorname{Coker}(u:L\to N). Lemma 2.62.1 gives injectivity of utu_t and naive XX-flatness of QtQ_t. By support compatibility under fibre base change, the closed support of QQ has strict fibre dimension at most dim⁡C−1≤n−1\dim C-1\leq n-1 whenever its fibre is nonempty. That closed support is inner at tt. If Qt≠0Q_t\ne 0, the induction hypothesis applied on this support makes QQ analytically XX-flat at tt; if Qt=0Q_t=0, local vanishing gives the same conclusion. When n=0n=0, the integral zero-dimensional fibre algebra CC is a field, so QC=0Q_C=0. Tensoring at the distinguished point gives Qt⊗OT,tH(t)=0Q_t\otimes_{\mathcal O_{T,t}}\mathcal H(t)=0, and Nakayama gives Qt=0Q_t=0. This supplies the induction start.

Step 4: all analytic tests. The target T→XT\to X is quasi-smooth, being an analytic-domain restriction of relative affine space. The map uxu_x is bijective on an ordinarily dense subset of its fibre and utu_t is injective. Corollary 2.52.6 therefore upgrades the analytic flatness of QQ to analytic flatness of NN. Explicitly, its universal-injectivity statement preserves L→NL\to N at every point above tt under every good-space base change over every complete extension. The pulled-back cokernel is the pullback of QQ by right exactness. Both that cokernel and the pullback of the free sheaf LL are flat over the new base, so their extension, the pullback of NN, is flat. This verifies all tests, not just those over the original field. The analytic version of Theorem 2.35.17 then gives analytic flatness of F\mathcal F at yy, because the distinguished finite fibre has one point; it includes all later fibre points after arbitrary tests. The induction is complete. ◻

NAG-001824 · Open at full scope

Theorem 2.62.3. NAG-001824 (Naive and analytic flatness at every inner point). Let KK be any complete non-Archimedean field, let f:Y→Xf:Y\to X be a morphism of good KK-analytic spaces, and let F\mathcal F be coherent on YGY_G. For every y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X), the following are equivalent: Fy\mathcal F_y is flat over OX,f(y)\mathcal O_{X,f(y)}; and F\mathcal F is analytically XX-flat at yy, tested after every good analytic base change over every complete valued extension of KK.

Proof. Only the naive-to-analytic implication requires proof. Put x=f(y)x=f(y). If H(x)\mathcal H(x) is nontrivially valued, use Theorem 2.62.2. Otherwise choose 0<r<10<r<1 and form the simultaneous affine-line product Y∗→X∗Y^*\to X^* with the radius-rr Gauss points y∗,x∗y^*,x^*. Lemma 2.61.8 proves naive flatness of the coherent pullback at y∗y^*. The point remains inner by analytic base-change stability of interior, and H(x∗)=H(x)r\mathcal H(x^*)=\mathcal H(x)_r is nontrivially valued. The preceding theorem makes this pullback analytically flat. Finally Lemma 2.58.3 gives analytic flatness downstairs. Its proof tests an arbitrary complete extension and good analytic base change and chooses the same-radius Gauss point above each test point; the two smooth projection maps on ordinary stalks are flat, and the source one is faithfully flat. Thus the descent covers exactly all tests in the stated condition. ◻

References. This reconstructs the inner-point flatness theorem [Duc18] (Theorem 8.3.4). It reuses the internal support and original-generic-basis lemmas, and the all-characteristic ordinary-to-analytic fibre comparison. The dimension induction is written for strict fibre-support bounds; the final Gauss step handles trivially valued completed residue fields.

Closed-fibre injection on the ordinary projection chart. Let TT be an affinoid domain of AXd,an\mathbb A_X^{d,\mathrm{an}}, where XX is a good analytic space over a complete field. Suppose tt lies in the ordinary interior of TT in relative affine space, with image x∈Xx\in X, and suppose H(x)\mathcal H(x) is nontrivially valued. In the notation of the proof, the canonical map S/mRS⟶OTx,tS/\mathfrak m_R S\longrightarrow\mathcal O_{T_x,t} is injective.

Proof. Put R=OX,xR=\mathcal O_{X,x}, S=OT,tS=\mathcal O_{T,t}, and m=mR\mathfrak m=\mathfrak m_R. Realize finite generators of m\mathfrak m on an ordinary affinoid neighborhood of xx, defining the closed base change X0→XX_0\to X. The resulting base stalk is R/mR/\mathfrak m. Over X0X_0, restrict to the ordinary open part of the relative affine space containing tt, where the projection is smooth. The source stalk is D=S/mSD=S/\mathfrak mS, and the analytic fibre is unchanged. The ordinary finite-branch stalk formula of Lemma 2.35.1 identifies the quotient stalk on the base and on the source. The actual affinoid fibre quotient follows from Proposition 2.33.11 and the finite base-change formula of Proposition 2.33.19.

Let h∈Dh\in D have zero germ in OTx,t\mathcal O_{T_x,t}. Represent hh on an ordinary affinoid neighborhood of tt in that smooth part. Its zero locus contains an ordinary neighborhood in the analytic fibre. Proposition 2.61.2 applies over the field stalk R/mR/\mathfrak m and gives pointwise vanishing on some ordinary neighborhood of tt in the source. Take a reduced ordinary affinoid neighborhood inside it. Such a neighborhood exists by regularity of the actual relative affine-space map on ordinary local rings over the field stalk (Proposition 2.35.56), normality of the resulting regular local ring [StacksAlg] (Tag 0567), and Corollary 2.35.47. On this reduced affinoid, pointwise vanishing gives spectral radius zero and hence h=0h=0, by Propositions 1.20.3 and 3.21.4. Therefore D→OTx,tD\to\mathcal O_{T_x,t} is injective. ◻

Inner coherent flatness in characteristic zero or at a rational base point. Let f:Y→Xf:Y\to X be a morphism of good KK-analytic spaces, let F\mathcal F be coherent on YGY_G, and let y∈Int⁡(Y/X)y\in\operatorname{Int}(Y/X) have image xx. Suppose F=H(x)F=\mathcal H(x) is nontrivially valued and that either char⁡K=0\operatorname{char}K=0 or xx is KK-rational. Then Fy\mathcal F_y is flat over OX,x\mathcal O_{X,x} if and only if F\mathcal F is analytically XX-flat at yy, with tests over every complete valued extension of KK.

Proof. The identity test gives naive flatness from analytic flatness. Assume naive flatness. A zero coherent stalk vanishes on an ordinary neighborhood and remains zero under every pullback. Otherwise take an ordinary affinoid neighborhood with strict FF-fibre by Proposition 2.51.3. Pass to the closed subspace cut out by the annihilator and use Theorem 2.35.17 to transport both flatness conditions. The immersion is finite and inner, and its fibre is a strict quotient. Lemma 2.57.1 identifies its fibre with the support on point sets. Induct on an upper bound nn on the dimension of the strict fibre of such a representative. Ordinary shrinking cannot increase that support dimension by Corollary 3.22.8.

For the resulting space ZZ, put d=dim⁡Zxd=\dim Z_x. Theorem 2.51.4 and Lemma 2.35.1 give an ordinary finite projection π:Z→T=M(B)\pi:Z\to T=\mathcal M(B) with only yy over t=π(y)t=\pi(y); here tt belongs to the ordinary interior of the target domain in AXd,an\mathbb A_X^{d,\mathrm{an}}. We shrink the target to have strict fibre, and use Theorem 2.35.23 and Lemma 2.52.2 to split its fibre algebra into integral factors. Lemma 2.56.4 retains the factor at tt by an ordinary restriction over KK. Pull the finite map back along these restrictions. Write C=B⊗^AFC=B\widehat\otimes_A F for the resulting integral strict fibre algebra, where XX has been restricted to an ordinary affinoid neighborhood M(A)\mathcal M(A) of xx. By Lemma 2.56.3, dim⁡C≤d\dim C\leq d. The point xx and its germ have remained fixed.

The stalk of N=π∗FN=\pi_*\mathcal F is naively XX-flat by the singleton formula in Theorem 2.35.17. Represent NN by a finite BB-module and choose a subset of its original generators whose images form a basis over Frac⁡(C)\operatorname{Frac}(C). Use these elements to define u:L=OTr→Nu:L=\mathcal O_T^r\to N, including r=0r=0 when the rank is zero. Lemma 2.57.2 gives an injective uxu_x, an ordinarily dense fibre bijectivity locus and a cokernel QxQ_x whose support has dimension at most dim⁡C−1\dim C-1 if nonempty.

Here put R=OX,xR=\mathcal O_{X,x}, S=OT,tS=\mathcal O_{T,t} and m=mR\mathfrak m=\mathfrak m_R. The canonical homomorphism D=S/mS→E=OTx,tD=S/\mathfrak m S\to E=\mathcal O_{T_x,t} is injective. In characteristic zero this follows from Corollary 2.56.10. In the rational case F=KF=K, so KK is nontrivially valued and Lemma 2.35.12 identifies this map with an isomorphism. Coherent pullback identifies the map on the analytic fibre with (ut mod m)⊗DE(u_t\bmod \mathfrak m)\otimes_D E. A vector in the kernel of Dr→Nt/mNtD^r\to N_t/\mathfrak m N_t has zero image in ErE^r, since the analytic-fibre map is injective. The map Dr→ErD^r\to E^r is injective because D→ED\to E is injective. Thus ut mod mu_t\bmod\mathfrak m is injective. The homomorphism R→SR\to S is local, SS is Noetherian, LtL_t is finite over SS, and NtN_t is flat over RR. Apply [StacksAlg] (Tag 00ME) to obtain injectivity of utu_t and RR-flatness of Qt=Coker⁡(ut)Q_t=\operatorname{Coker}(u_t).

If n=0n=0, the integral zero-dimensional fibre algebra CC is a field and Qx=0Q_x=0. Tensoring at tt gives Qt⊗SH(t)=0Q_t\otimes_S\mathcal H(t)=0. The residue field of SS embeds in H(t)\mathcal H(t), so Qt/mSQt=0Q_t/\mathfrak m_S Q_t=0, and [StacksAlg] (Tag 00DV(2)) gives Qt=0Q_t=0. For n>0n>0 and Qt≠0Q_t\ne0, the closed support of QQ is inner at tt, with strict FF-fibre of dimension at most n−1n-1 by the support and dimension statements above. Apply induction on this closed support and the finite pushforward equivalence to obtain analytic XX-flatness of QQ at tt. If Qt=0Q_t=0, local vanishing gives the same flatness.

The map T→XT\to X is quasi-smooth near tt, and tt lies in the ordinary closure of the fibre bijectivity locus. Apply the cokernel-to-target implication of Corollary 2.52.6. Its universal injectivity preserves uu at every point over tt in every good analytic base change over every complete valued extension. The pulled-back cokernel is the pullback of QQ, and the free source and that cokernel are flat over the new base. Thus NN is analytically XX-flat at tt. The analytic finite pushforward equivalence then proves the assertion for F\mathcal F at yy. ◻