F · Descent of an algebraized polarized tower

13 NAG Tags

Corollary F.0.1. NAG-002987 (A descended polarization in the exact tame Kummer class). Let KK be complete and nontrivially real-valued, with valuation ring OO, residue field kk, and value group Γ\Gamma. Assume the exact coefficient class of : e≥2e\geq2 is invertible in OO, ζ∈O\zeta\in O and its residue have order ee, and L=K(s),se=c∈K×,γ=v(s)>0,iγ∉Γ(0<i<e).(K)L=K(s),\quad s^e=c\in K^\times,\quad \gamma=v(s)>0, \qquad i\gamma\notin\Gamma\quad(0<i<e). \tag{K} Put O′=L∘O'=L^\circ and G=⟨σ⟩G=\langle\sigma\rangle, where σ(s)=ζs\sigma(s)=\zeta s. Choose 0≠π∈mO0\ne\pi\in\mathfrak m_O. Suppose a cartesian proper flat finitely presented tower over O′/πnO′O'/\pi^nO' with compatible semilinear GG-actions satisfies the nodal and full special-fibre identity hypotheses of Theorem D.0.3. Then there exist a projective, flat, finitely presented nodal OO-scheme YY of relative dimension one and a relatively ample invertible sheaf P\mathcal P on YY, with compatible equivariant polarized isomorphisms (Y,P)×O(O′/πnO′)≃(Xn,Nn).(6)(Y,\mathcal P)\times_O(O'/\pi^nO') \simeq (X_n,\mathcal N_n). \tag{6} Here Nn\mathcal N_n is the chosen orbit-tensor tower (4). The algebraized bundle N′\mathcal N' itself descends to P\mathcal P; no further tensor power of N′\mathcal N' is needed. In particular P∣Yk\mathcal P|_{Y_k} identifies with AC⊗e\mathcal A_C^{\otimes e} under the induced residue identification. The reductions (Y,P)×OO/πnO(Y,\mathcal P)\times_O O/\pi^nO form the corresponding descended polarized tower.

Proof. The coefficient criterion identifies the residue fields of OO and O′O' with kk, gives identity coefficient action on kk, and uses the actual complete valued field LL and its valuation ring. Thus Theorem D.0.3, applied over O′O', first supplies an algebraic projective nodal pair (Y′,N′)(Y',\mathcal N') with the specified semilinear action and full residual identities. This step uses no ramified descent. For clarity, with bg=Spec⁡(g−1)b_g=\operatorname{Spec}(g^{-1}) the induced left action on functions is g⋅a=ρg−1∗ag\cdot a=\rho_{g^{-1}}^*a, and on linearized sections it is g⋅t=λg−1(ρg−1∗t)g\cdot t=\lambda_{g^{-1}}(\rho_{g^{-1}}^*t). Thus its coefficient action on O′O' is exactly gg, as in the cited tame descent results.

Theorem E.0.1 applies to Y′/O′Y'/O'. It supplies a projective flat finitely presented nodal Y/OY/O and an equivariant isomorphism Y×OO′≃Y′Y\times_O O'\simeq Y', with the prescribed coefficient action on the base change. Properness is retained: Y′Y' is projective and proper, and the cited result descends properness before concluding projectivity.

Under this identification, N′\mathcal N' is an already existing GG-linearized invertible sheaf on Y×OO′Y\times_O O' whose action on the entire special-fibre module is the identity. Apply to this bundle itself. For p:Y×OO′→Yp:Y\times_O O'\to Y, it gives P=(p∗N′)G,p∗P≃N′,(7)\mathcal P=(p_*\mathcal N')^G, \qquad p^*\mathcal P\simeq\mathcal N', \tag{7} and makes P\mathcal P relatively ample. There is no need to replace N′\mathcal N' by a second orbit tensor. The scheme and bundle descents are unique with their specified identifications, by the two cited results.

Reduce the algebraic isomorphisms just constructed modulo πn\pi^n and use the level comparisons (5) of Theorem D.0.3. This gives (6), with all transition maps and linearizations, and then the asserted residue polarization. The O/πnOO/\pi^nO reductions of the actual pair (Y,P)(Y,\mathcal P) are automatically cartesian, proper, flat and finitely presented. Their base changes to O′/πnO′O'/\pi^nO' are exactly the supplied polarized levels. In particular no valuation-ring descent theorem has been applied directly to the non-valuation quotient rings O′/πnO′O'/\pi^nO'. ◻

Prescribed subdivisions of an existing model

Let KK be a complete non-Archimedean field with a specified nontrivial real-valued absolute value. Write O=K∘,k=K~,v(c)=−log⁡∣c∣,Γ=v(K×)⊂R.O=K^\circ,\qquad k=\widetilde K,\qquad v(c)=-\log|c|,\qquad \Gamma=v(K^\times)\subset\mathbb R. Let C/KC/K be a smooth projective geometrically connected curve. A model means a projective flat finitely presented nodal OO-curve with generic fibre identified with CC and geometrically connected, geometrically reduced special fibre, as in 4.45.3. A marking DD, when present, is a fixed finite etale OO-scheme embedded in the relative smooth locus. All model morphisms below induce the identity on CC.

Fix an actual model X\mathcal X. Choose a finite unramified Galois residue lift L/KL/K splitting its nodes. Put l=L~l=\widetilde L and G=Gal⁡(L/K)G=\operatorname{Gal}(L/K). Here unramified means the actual finite free etale coefficient extension of 4.43.1; in particular v(L×)=Γv(L^\times)=\Gamma. The splitting field and rational pointed etale nodal charts exist by 4.18.2.

For each node ee of Xl\mathcal X_l, choose an ordering of its two branches and a pointed chart xeye=aex_e y_e=a_e, with 0<∣ae∣<10<|a_e|<1. Put ℓe=v(ae)\ell_e=v(a_e). By 4.39.1 and 4.39.3, its entire tube is the standard open annulus ∣ae∣<∣xe∣<1,s=v(xe)∈(0,ℓe).|a_e|<|x_e|<1,\qquad s=v(x_e)\in(0,\ell_e). The coordinate ss increases from the chosen first end to the second. These are two distinct branch ends even if their component vertices coincide.

Use an algebraically closed complete isometric comparison field Ω⊃L\Omega\supset L with residue field k‾\overline k, as in 4.45.1, and write p:CΩan→Canp:C_\Omega^{\mathrm{an}}\to C^{\mathrm{an}}. The geometric component vertices of a model are exactly p−1(V(X))p^{-1}(V(\mathcal X)), by 4.45.2. Write ζe(r)\zeta_e(r) for the Gauss point of radius rr in the selected annular tube over Ω\Omega.

Definition F.1.1. (Compatible cut data). A cut datum consists of a finite set Se⊂(0,ℓe)∩ΓS_e\subset(0,\ell_e)\cap\Gamma for each split node ee. The canonical semilinear action of g∈Gg\in G takes the skeleton coordinate ss either to ss or to ℓe−s\ell_e-s on the gege-tube, according as it preserves or exchanges the chosen branch orderings. Require respectively Sge=SeorSge=ℓe−Se.S_{ge}=S_e \quad\hbox{or}\quad S_{ge}=\ell_e-S_e . Empty sets are permitted.

These conditions concern labelled tubes of the given model. The coordinate-change result 4.19.1 makes them independent of normalized annular coordinates. A stabilizer exchanging the two branches forces reflection invariance of SeS_e; a stabilizer preserving both branches permits asymmetric cuts. A self-loop alone imposes no reflection condition: the issue is the branch action, not equality of endpoint vertices.

Theorem. NAG-002600 (Exact arithmetic subdivisions). Every compatible cut datum S=(Se)S=(S_e) has a model X[S]\mathcal X[S], preserving DD, with geometric vertices exactly p−1(V(X)) ∪ ⋃e{ζe(exp⁡(−s)):s∈Se}.(*)p^{-1}(V(\mathcal X))\ \cup\ \bigcup_e\{\zeta_e(\exp(-s)):s\in S_e\}. \tag{*} It dominates X\mathcal X and is unique up to the unique marked generic-identity isomorphism. It is independent of the splitting field and coordinates when the prescribed analytic points are fixed.

If Se⊂TeS_e\subset T_e for all ee, there is a unique compatible morphism X[T]→X[S]\mathcal X[T]\to\mathcal X[S]. The model X[S∪T]\mathcal X[S\cup T] is the least common dominating model of X[S]\mathcal X[S] and X[T]\mathcal X[T], even among models with vertices outside the original skeleton.

Proof. First form the reflection closure Se#=Se∪(ℓe−Se)S_e^\#=S_e\cup(\ell_e-S_e). For nodes above one original closed node, these are the same finite numerical set independently of branch ordering: GG acts transitively above that node, and reflection no longer changes the set.

We construct an original-ring model X#\mathcal X^\# with exactly these extra cuts. For a nonempty reflection-invariant set in (0,ℓ)(0,\ell), let dd be its smallest member. Then d∈Γd\in\Gamma and d≤ℓ/2d\leq\ell/2. Apply Lemma C.1.84. If d<ℓ/2d<\ell/2, all remaining cuts lie in the central annulus, of length ℓ−2d\ell-2d; their new coordinates s−ds-d again form a reflection-invariant set. If d=ℓ/2d=\ell/2, there was just one cut.

Over ll, the new central nodes are rational and split by the displayed charts. They form a GG-invariant collection, hence are the full inverse image of a collection of closed nodes of the actual current kk-model. Perform the same original-ring operation at these nodes. At each nonempty step at least one member of the finite list is removed. All new distances are differences in Γ\Gamma, so the required scalars belong to KK. This finite process, applied at the finitely many original nodes, constructs X#\mathcal X^\# and its actual map to X\mathcal X. It uses no least positive value and no discreteness assumption.

Retain all old geometric components and exactly the new components labelled by SeS_e, discarding those in Se#∖SeS_e^\#\setminus S_e. This selection is GG-invariant. The new components and their attachment points are split over ll: they are the exceptional projective lines in the accepted chart calculation, and the pointed etale maps induce isomorphisms on the exceptional fibres. Thus the action of Aut⁡(k‾/k)\operatorname{Aut}(\overline k/k) on the new components factors through Gal⁡(l/k)\operatorname{Gal}(l/k). All old geometric components are retained. The selection is therefore invariant under the full residual action. By 4.49.1, it is the entire inverse image of a subset V⊂V(X#)V\subset V(\mathcal X^\#) of actual original-field component vertices.

Each connected collection of discarded geometric components is a chain of smooth projective lines with no marking and two attaching branches on retained components. It cannot be a cycle: every original edge has retained endpoints, including the retained old vertex at both ends of a self-loop. If the two attaching branches meet the same retained component, they are counted twice. This is exactly the marked forest condition 4.45.4, with b=2,m=0b=2,m=0. The actual original-ring theorem 4.45.5 now gives a projective model X[S]\mathcal X[S] with precisely (*).

All old vertices remain, so 4.46.2 gives X[S]→X\mathcal X[S]\to\mathcal X. That accepted theorem and 4.46.3 also give uniqueness, marking compatibility and the morphisms for S⊂TS\subset T. They prove independence of auxiliary choices, using 4.45.2 to identify original vertex sets from their entire geometric inverse images.

The union datum is compatible. Every model dominating both X[S]\mathcal X[S] and X[T]\mathcal X[T] contains both vertex sets, hence their union, which is V(X[S∪T])V(\mathcal X[S\cup T]). Applying 4.46.2 gives its unique morphism to X[S∪T]\mathcal X[S\cup T], as required. ◻

Proposition. NAG-002601 (The full tubes of a subdivision). In a geometric node tube AeA_e, list the cuts 0<s1<⋯<sm<ℓe0<s_1<\cdots<s_m<\ell_e, and set r0=1,rj=exp⁡(−sj),rm+1=∣ae∣.r_0=1,\qquad r_j=\exp(-s_j),\qquad r_{m+1}=|a_e|. The closed-point tubes of X[S]Ω∘\mathcal X[S]_{\Omega^\circ} inside AeA_e are precisely the open annuli {rj+1<∣xe∣<rj},0≤j≤m,(1)\{r_{j+1}<|x_e|<r_j\},\quad 0\leq j\leq m, \tag{1} and the open discs, for 1≤j≤m1\leq j\leq m, {∣xe/bj−α^∣<1},α∈k‾×,bj∈K×,v(bj)=sj.(2)\{|x_e/b_j-\widehat\alpha|<1\}, \quad \alpha\in\overline k^\times,\quad b_j\in K^\times,\quad v(b_j)=s_j. \tag{2} Here α^∈Ω∘\widehat\alpha\in\Omega^\circ is any lift of α\alpha. The omitted point on ∣xe∣=rj|x_e|=r_j is its Gauss vertex. The old smooth tubes are unchanged as analytic subsets.

The two ends of (1) carry consecutive component labels, allowing coincident endpoint labels at an original self-loop. Each disc (2) has the inserted vertex as its sole boundary point. On an inserted geometric component the two annular directions are the normalization points 0,∞0,\infty; the other residual points give (2).

Proof. On ∣xe∣=rj|x_e|=r_j, the coordinate z=xe/bjz=x_e/b_j gives M(Ω⟨z,z−1⟩)\mathcal M(\Omega\langle z,z^{-1}\rangle). Its canonical reduction is k‾[z,z−1]\overline k[z,z^{-1}], with one component point, by 3.20.3. The inverse images of its closed points are exactly (2), by 3.20.1; these are open unit discs with coordinate z−α^z-\widehat\alpha. The intervals between cut circles give the standard open annuli (1). These open connected sets partition the complement of the inserted vertices in AeA_e, so they are its connected components.

No new vertex lies in an old smooth tube. The specialization partition 4.39.5, applied over Ω\Omega, identifies precisely these components with the closed-point tubes of X[S]\mathcal X[S]. The boundary assertions follow from 4.39.3. The two ends are retained before any identification of component labels, so this includes self-loops. The accepted contraction theorem only removes chains between these retained endpoints. ◻

Over the original field, specialization commutes with coefficient extension by 4.28.1. The inverse image of an original closed-point tube is therefore the union of the geometric tubes above that residual closed point. A finite unramified field splitting the nodes need not split every smooth residual closed point. An abstract edge and a list of radii do not record this information. The actual blowups and contractions supply the residual schemes and their morphisms. In a nonsplit situation a stabilizer can exchange 0,∞0,\infty; its action on these points and on the other residual points must be retained. No assertion that the descended component or all its tubes are split is used.

Corollary. NAG-002602 (A literal-union common refinement). Let X,Y\mathcal X,\mathcal Y be two given models of CC, with the same DD when marked. Let ΣX,Ω\Sigma_{\mathcal X,\Omega} be the embedded compact geometric skeleton of X\mathcal X, including its component vertices. Assume V(YΩ∘)⊂ΣX,Ω.(3)V(\mathcal Y_{\Omega^\circ}) \subset\Sigma_{\mathcal X,\Omega}. \tag{3} There is then a least common dominating model Z\mathcal Z with V(Z)=V(X)∪V(Y).V(\mathcal Z) =V(\mathcal X)\cup V(\mathcal Y). It is the exact subdivision of X\mathcal X in Theorem F.1.2, with the full tube description of Proposition F.1.3. Condition (3) holds, in particular, if the two embedded geometric skeleta agree.

Proof. Choose a common finite unramified Galois splitting field for the nodes of both models. Every geometric vertex of Y\mathcal Y not already a geometric vertex of X\mathcal X lies on the radial skeleton of one of the split X\mathcal X-annuli. Its image w∈CLanw\in C_L^{\mathrm{an}} is a component vertex of YL∘\mathcal Y_{L^\circ}, by 4.45.2. The annular coordinate is analytic on its entire LL-tube, so its evaluation lies in H(w)\mathcal H(w). Consequently v(xe(w))∈v(H(w)×)=v(L×)=Γ.v(x_e(w))\in v(\mathcal H(w)^\times) =v(L^\times)=\Gamma . The finite cut sets obey the branch compatibility because both models are defined over KK. The theorem constructs X[S]\mathcal X[S] whose geometric vertices are their union. The entire-inverse-image assertion 4.45.2 identifies its original vertex set with the asserted original union. The two domination maps and least-common property follow from 4.46.2. Marking compatibility follows from 4.46.3. ◻

Residual action at the ramified midpoint

Use the field and model of Proposition C.1.85. Its quadratic extension is L=K(b)L=K(b) with b2=tb^2=t, and O′=L∘O'=L^\circ. The coefficient ring is the actual ring of Proposition C.1.86.

Proposition. NAG-002616 (The midpoint’s residual action is nontrivial). Let Y/O′\mathcal Y/O' be the three-vertex projective model constructed in the audited midpoint proof by symmetric insertion in TU=t=b2TU=t=b^2. The canonical action fixing T,UT,U and sending bb to −b-b extends to Y\mathcal Y. On the new residual component it acts as u‾⟼−u‾,u=T/b.(14)\overline u\longmapsto-\overline u, \qquad u=T/b . \tag{14} Its two branch ends 0,∞0,\infty are fixed. For α∈k×\alpha\in k^\times, the geometric residual-direction disc labelled α\alpha is carried to that labelled −α-\alpha. Thus the invariant midpoint and its fixed branch ends do not carry trivial residual-scheme or tube action. This gives a residual-action obstruction to effective descent of this given model with its fixed generic identification.

Proof. The audited symmetric insertion uses the ideal (T,U,b)(T,U,b) on the node chart. This ideal is invariant under the stated action, so its Rees algebra and blowup inherit the action; on the rest of the projective model the original coefficient action already exists. These actions agree on overlaps. The accepted symmetric blowup charts, used in the passing proof, give its middle chart as O′[u,v]/(uv−1),T=bu,U=bv.(15)O'[u,v]/(uv-1),\qquad T=bu,\quad U=bv. \tag{15} Here σ(u)=−u\sigma(u)=-u, σ(v)=−v\sigma(v)=-v. The two adjacent node charts add the missing ends 0,∞0,\infty of this residual Pk1\mathbb P^1_k, proving (14) on the whole component. Since char⁡(k)=0\operatorname{char}(k)=0, it is a nontrivial scheme automorphism.

The full tube description in F.1.3 identifies the smooth residual direction α≠0,∞\alpha\ne0,\infty with ∣T/b−α^∣<1.|T/b-\widehat\alpha|<1 . Here one may use the constant lift α^=α∈k\widehat\alpha=\alpha\in k. The action fixes this lift and changes T/bT/b to −T/b-T/b. It therefore carries this disc to the disc for −α-\alpha, which is distinct because α≠0\alpha\ne0 and the residue characteristic is zero.

Every actual base change from OO has identity inertia on its special fibre, because O′/m′=O/m=kO'/\mathfrak m'=O/\mathfrak m=k and the coefficient action on kk is the identity. An isomorphism giving descent with the prescribed generic identification would transport that action to the displayed action on Y\mathcal Y. Indeed extensions of a fixed generic action to a flat separated model are unique: their equalizer is closed by separatedness, and its defining ideal vanishes after inverting the nonzero elements of O′O', so vanishes by flatness. This contradicts (14). This argument diagnoses the residual mechanism of the already audited obstruction; it does not replace that earlier proof or assert a new model-existence theorem. ◻

Ordinary moment support and finite freeness

Moment norms near a supplied retraction

Let KK be a complete non-Archimedean field with its given real-valued absolute value. We use tf=(−log⁡∣fj∣)jt_f=(-\log|f_j|)_j and write Int⁡top(V,U)\operatorname{Int}_{\mathrm{top}}(V,U) for the ordinary interior of VV in UU.

Lemma. NAG-002478 (Local equality of moment norms). Let UU and YY be good Hausdorff KK-analytic spaces, let i:Y↪Ui:Y\hookrightarrow U be a closed immersion, and let r:U→Yr:U\to Y be an analytic morphism with r∘i=id⁡Yr\circ i=\operatorname{id}_Y. Assume that every affinoid domain of YY under consideration has analytic dimension at most dd. Let z=i(y)z=i(y), let V⊂UV\subset U be an ordinary affinoid neighbourhood of zz, and let f=(f1,…,fN)f=(f_1,\ldots,f_N) be analytic units on VV. There is an ordinary affinoid neighbourhood W⊂VW\subset V of zz such that dim⁡tf(W)≤d.\dim t_f(W)\leq d. The choice of WW depends on the tuple, but not on any ambient superform subsequently pulled back by that tuple. Thus all such forms of bidegree (p,q)(p,q) with max⁡(p,q)>d\max(p,q)>d induce zero ordinary sections on Int⁡top(W,U)\operatorname{Int}_{\mathrm{top}}(W,U).

Proof. Choose an ordinary affinoid neighbourhood BB of yy in YY contained in i−1(Int⁡top(V,U))i^{-1}(\operatorname{Int}_{\mathrm{top}}(V,U)). Write gj=i∗fj∣Bg_j=i^*f_j|_B, which is a unit on BB. On the ordinary open neighbourhood O0=Int⁡top(V,U)∩r−1(Int⁡top(B,Y))O_0=\operatorname{Int}_{\mathrm{top}}(V,U)\cap r^{-1}(\operatorname{Int}_{\mathrm{top}}(B,Y)) of zz, the functions aj=r∗gja_j=r^*g_j are units. The analytic-domain restriction and neighbourhood basis used here are those in Theorem 5.52.3. Since ri=id⁡r i=\operatorname{id}, each fj−ajf_j-a_j has value zero at zz, whereas ∣aj(z)∣>0|a_j(z)|>0. By continuity, the open set O1={x∈O0:∣fj(x)−aj(x)∣<∣aj(x)∣ for every j}O_1=\{x\in O_0: |f_j(x)-a_j(x)|<|a_j(x)| \text{ for every }j\} contains zz. Choose an ordinary affinoid neighbourhood WW of zz contained in O1O_1. At every x∈Wx\in W, the strict ultrametric inequality gives ∣fj(x)∣=∣aj(x)∣|f_j(x)|=|a_j(x)|: one inequality follows from fj=aj+(fj−aj)f_j=a_j+(f_j-a_j), and a strict reverse inequality would contradict the same triangle inequality applied to aj=fj−(fj−aj)a_j=f_j-(f_j-a_j). Consequently tf(W)=tg(r(W))⊆tg(B).t_f(W)=t_g(r(W))\subseteq t_g(B). By Theorem 5.39.3, the right-hand side is a finite polyhedral set of dimension at most dd, and the left-hand side is also finite polyhedral. Hence its dimension is at most dd. On every face its tangent space has dimension at most dd, so one of the exterior factors of a form with max⁡(p,q)>d\max(p,q)>d restricts to zero. The ordinary moment definition then gives the asserted zero sections. ◻

Avoiding the regular zero space

Let XX be good, Hausdorff, and pure of analytic dimension n≥1n\geq1 over a complete real-valued non-Archimedean field KK, possibly trivially valued. Let η\eta be an ordinary smooth compactly supported (n−1,n−1)(n-1,n-1)-form whose actual support is contained in the absolute analytic interior Int⁡(X/K)\operatorname{Int}(X/K). Let ff be a regular analytic function on XX, with D=V(f)D=V(f). On the affinoid neighbourhoods at issue its representative is a non-zero-divisor in the actual affinoid algebra.

Lemma F.1.7. (Support avoidance from the exact F7 support lemma; F8F7-S). For the test η\eta above, supp⁡(d′′η)∩D=∅.(2.1)\operatorname{supp}(d''\eta)\cap D=\varnothing. \tag{2.1} In particular the extension by zero ext⁡0(log⁡∣f∣ d′d′′η∣X∖D)(2.2)\operatorname{ext}_0\bigl(\log|f|\,d'd''\eta|_{X\setminus D}\bigr) \tag{2.2} is an ordinary smooth compactly supported top form on XX.

Proof. At a point xx of supp⁡(d′′η)\operatorname{supp}(d''\eta), choose an ordinary affinoid neighbourhood V=M(A)V=\mathcal M(A) carrying one unit-moment representative of d′′ηd''\eta, and on which f∈Af\in A is a non-zero-divisor. It can be chosen inside the analytic interior around xx. The exact F7 lemma NAG-002578, lines 165–225, applies to this (n−1,n)(n-1,n)-form: its germ is nonzero, and the evaluation kernel ker⁡(x)\ker(x) is a minimal prime of AA.

A non-zero-divisor in a Noetherian ring avoids every minimal prime. Indeed, localization at a minimal prime is a nonzero zero-dimensional Noetherian local ring, hence Artinian by Stacks 00KH, forward direction; an element of its maximal ideal is nilpotent and cannot act injectively. Localization preserves injectivity. Thus f∉ker⁡(x)f\notin\ker(x), so ∣f(x)∣>0|f(x)|>0, proving (2.1). Here Noetherianity is the same NAG-001560 used in the F7 source.

Locality of d′d' gives supp⁡(d′d′′η)⊆supp⁡(d′′η)\operatorname{supp}(d'd''\eta)\subseteq\operatorname{supp}(d''\eta). This compact support is disjoint from DD. On X∖DX\setminus D, the logarithm is a smooth function of the unit ff; near DD the tested top form is identically zero. Gluing gives the smooth extension (2.2). ◻

Freeness at defect at most one

Use the given complete field and the same convention for analytic dimension. Write dK(s)d_K(s) for the Abhyankar dimension of H(s)\mathcal H(s) over KK.

Lemma. NAG-002800 (Finite maps become free along the section; F8RC-F). lem:finiteflatlem:avoid Let F:Y→ZF:Y\to Z be finite between good Hausdorff spaces pure of analytic dimension nn. Suppose the ordinary local rings of ZZ are regular, and every affinoid domain of YY has S1S_1 algebra. At every s∈Zs\in Z with dK(s)≥n−1d_K(s)\geq n-1, the finite coherent module F∗OYF_*\mathcal O_Y is free on an ordinary neighbourhood of ss. Rank zero is allowed when the fibre is empty.

Proof. Put R=OZ,sR=\mathcal O_{Z,s} and M=(F∗OY)sM=(F_*\mathcal O_Y)_s. By Lemma C.1.58, RR is regular local of dimension zero or one. In dimension zero it is a field, and the finite module MM is free. Suppose dim⁡R=1\dim R=1, and choose a generator π\pi of its maximal ideal. A regular local ring is a domain by Stacks 00NP, so π≠0\pi\ne0.

Represent π\pi by aa on a sufficiently small ordinary affinoid neighbourhood M(C)\mathcal M(C) of ss. Write U=M(C)U=\mathcal M(C) for this chosen domain. At every x∈Ux\in U, including its domain boundary, the regular ordinary-stalk comparison makes OZ,x→OU,x\mathcal O_{Z,x}\to\mathcal O_{U,x} a flat local map of Noetherian local rings with regular closed fibre. The assumed regularity of OZ,x\mathcal O_{Z,x} then gives regularity of OU,x\mathcal O_{U,x} by Stacks 031E. Thus all ordinary stalks of this affinoid are regular. By NAG-001861, we may retain the integral, open-and-closed component containing ss. Its inverse image is affinoid, say M(B)\mathcal M(B), finite over CC, and is either empty or pure of dimension nn. The hypothesis makes BB an S1S_1 ring. We claim that every minimal prime qq of BB contracts to zero in CC.

Indeed, if p=q∩Cp=q\cap C, then C/p→B/qC/p\to B/q is injective and finite. Such a map is surjective on analytic spectra: its finite module has support all of Spec⁡(C/p)\operatorname{Spec}(C/p), so every residue-field fibre is nonzero by Stacks 00DV(2) by contraposition at the corresponding local ring; extending to the complete residue field gives a nonzero finite-dimensional affinoid algebra and hence an analytic point. Completed residue extensions at points of a finite map are finite, by NAG-001590. The maximum formula NAG-001591 therefore gives dim⁡KM(C/p)=dim⁡KM(B/q)=n.\dim_K\mathcal M(C/p)=\dim_K\mathcal M(B/q)=n. If p≠0p\ne0, dimension drop (NAG-002374) in the integral pure nn-dimensional algebra CC contradicts this equality. Thus p=0p=0.

The element aa is nonzero in CC, since its germ is π≠0\pi\ne0. It avoids every minimal prime of BB, hence every associated prime, because BB is S1S_1 by Stacks 031Q, (2) implies (1). Multiplication by aa on BB is injective. The ordinary comparison C→RC\to R is flat (NAG-001698, or 2.35.45 after localization), and M=B⊗CRM=B\otimes_C R by finite coherent restriction. Thus π\pi acts injectively on MM.

Lift a basis of M/πMM/\pi M to obtain Rr→MR^r\to M, surjective by Nakayama (Stacks 00DV(2) on the finite cokernel). Its kernel NN is finite. The resolution 0→R→πR→R/(π)→00\to R\xrightarrow{\pi}R\to R/(\pi)\to0 and injectivity of π\pi on MM give Tor⁡1R(M,R/(π))=0\operatorname{Tor}_1^R(M,R/(\pi))=0. Tensoring the presentation of MM therefore gives N/πN=0N/\pi N=0. Nakayama gives N=0N=0 by Stacks 00DV(2) on NN, proving that MM is free. Finally NAG-001753 spreads the free stalk to an ordinary neighbourhood. This proves actual local freeness, including its analytic base changes; mere torsion-freeness over an unspecified higher-dimensional ring has not been used. ◻

Serre comparison and Cartier coefficients

Fix a complete non-Archimedean field with its specified real-valued absolute value, possibly trivial. Use actual generalized affinoids with their canonical quotient and finite-module norms. All dimensions of analytic spaces in this section are analytic dimensions.

Serre S1S_1 for analytic stalks

Theorem. NAG-002765 (Analytic Serre comparison; F8ST-A). thm:analyticserrelem:avoid Let AA be a generalized affinoid algebra, X=M(A)X=\mathcal M(A), MM a finite AA-module, and M~\widetilde M its coherent sheaf. For x∈Xx\in X, put p=ker⁡A(x)\mathfrak p=\ker_A(x). Then Mp is S1⟺M~x is S1 as a module over OX,x.\begin{equation} M_{\mathfrak p}\text{ is }S_1 \quad\Longleftrightarrow\quad \widetilde M_x\text{ is }S_1 \text{ as a module over }\mathcal O_{X,x}. \end{equation} Thus MM is S1S_1 if and only if all these analytic stalk modules are S1S_1. If VV is a good analytic domain in a good analytic space XX, and F\mathcal F is coherent, then at every x∈Vx\in V Fx is S1⟺(F∣V)x is S1.\begin{equation} \mathcal F_x\text{ is }S_1 \quad\Longleftrightarrow\quad (\mathcal F|_V)_x\text{ is }S_1. \end{equation} The same comparison applies to a finite-type scheme over an affinoid base, its analytification, and an affinoid domain therein, with the corresponding algebraic and analytic stalks.

Proof. By NAG-001624/2.47.2, M~x=M⊗AOX,x\widetilde M_x=M\otimes_A\mathcal O_{X,x}. The map Ap→OX,xA_{\mathfrak p}\to\mathcal O_{X,x} is regular by NAG-001782 and is local, hence faithfully flat. Lemma C.1.15 proves (F.1.1). Every prime of AA is the kernel of an analytic point by NAG-001860, proving the assertion for all of MM. For (F.1.2), use the tensor identity NAG-001666 and the regular local map NAG-001792. For the last assertion use the regular comparison square NAG-001791 and the same finite-presentation stalk calculation. All localizations required in Lemma C.1.15 are Noetherian. ◻

The ordinary stalk identifications are the finite-coherent equivalence and coherent restriction. The regular maps used above are affinoid comparison, good-domain restriction, and comparison over an affinoid base. The passage to all primes uses realization by analytic points.

Let XX be pure of analytic dimension n≥1n\geq1, and let f∈Af\in A be regular. Write fi=f∣Xif_i=f|_{X_i} for its component restrictions. For an affinoid X=M(A)X=\mathcal M(A) pure of dimension nn, write Min⁡A={q1,…,qs},Xi=M(A/qi),ii:Xi↪X,mi=length⁡AqiAqi.\begin{equation} \operatorname{Min}A=\{\mathfrak q_1,\ldots,\mathfrak q_s\},\qquad X_i=\mathcal M(A/\mathfrak q_i),\quad i_i:X_i\hookrightarrow X,\quad m_i=\operatorname{length}_{A_{\mathfrak q_i}}A_{\mathfrak q_i}. \end{equation}

Proposition. NAG-002966 (Component scope and support; F8CR-S). fnd:nag002966:component-supportlem:avoid In (F.1.3), every XiX_i is integral and pure of dimension nn, every fif_i is regular, and every actual good affinoid domain in XiX_i has an S1S_1 coordinate ring. For an ordinary compactly supported form η\eta on XX, Supp⁡(ii∗η)⊆ii−1Supp⁡η.\operatorname{Supp}(i_i^*\eta)\subseteq i_i^{-1}\operatorname{Supp}\eta. If Supp⁡η⊂Int⁡(X/K)\operatorname{Supp}\eta\subset\operatorname{Int}(X/K), its pullback has compact support in Int⁡(Xi/K)\operatorname{Int}(X_i/K). Moreover ii∗(d′d′′η)=d′d′′(ii∗η)i_i^*(d'd''\eta)=d'd''(i_i^*\eta). The support and functoriality assertions also hold for an arbitrary closed immersion of good Hausdorff analytic spaces.

Proof. The quotients by minimal primes have their actual affinoid structures by NAG-001561 and NAG-001562. Purity of XX gives their analytic dimension nn. Proposition C.1.14’s avoidance argument makes fi≠0f_i\ne0 in the domain A/qiA/\mathfrak q_i. Every domain is S1S_1: at a nonzero prime its local ring has a nonzero element of the maximal ideal which is regular, and the zero-prime localization is a field. The finite coherent comparison NAG-002765 applies to the structure module of this affinoid and then to its actual good analytic domains. It gives the asserted analytic stalk and affinoid S1S_1 statements. It is never applied to infer S1S_1 for the original possibly embedded-component ring.

Closed immersions are finite locally with the canonical quotient norm. By NAG-001701, every point is relatively interior; NAG-001704 and the atlas definition of relative interior give ii−1Int⁡(X/K)⊆Int⁡(Xi/K).i_i^{-1}\operatorname{Int}(X/K)\subseteq\operatorname{Int}(X_i/K). Their underlying maps identify the source with a closed subspace. The preimage of the compact support is therefore compact. Pullback of a zero germ is zero, proving the support inclusion. Finally NAG-002424 gives pullback commuting with both differentials, with their fixed order unchanged. These statements are about the true analytic interior; compactness alone does not remove an analytic boundary. ◻

Proposition. NAG-002769 (Points computing full generic lengths; F8ST-G). fnd:nag002769:generic-lengthlem:avoid Let Y=M(B)Y=\mathcal M(B) be nonempty and pure of analytic dimension dd, p∈Min⁡B\mathfrak p\in\operatorname{Min}B, and MM a finite BB-module. There is x∈M(B/p)x\in\mathcal M(B/\mathfrak p) with dK(x)=dd_K(x)=d. Every such point has kernel p\mathfrak p on BB, belongs to no other irreducible component, and has the following property. For every affinoid domain V⊂YV\subset Y containing xx, the ordinary local ring SV=OV,xS_V=\mathcal O_{V,x} is Artinian, and length⁡SV(M⊗BSV)=length⁡BpMp.\begin{equation} \operatorname{length}_{S_V}(M\otimes_B S_V)=\operatorname{length}_{B_{\mathfrak p}}M_{\mathfrak p}. \end{equation} The same equality holds for the pointed analytic-domain colimit OYG,x\mathcal O_{Y_G,x} and the coherent module colimit. These colimit modules have finite length. Thus the analytic generic multiplicity can be computed by the algebraic generic length, including nilpotents.

Proof. The reduced component M(B/p)\mathcal M(B/\mathfrak p) is pure of dimension dd. The maximum formula gives the required point. A nonzero element of B/pB/\mathfrak p has zero locus of dimension at most d−1d-1, or empty when d=0d=0, by NAG-002374. Such a zero locus cannot contain a point of Abhyankar dimension dd. Hence the kernel on BB is exactly p\mathfrak p. For another minimal prime p′\mathfrak p', an element of p′∖p\mathfrak p'\setminus\mathfrak p proves that xx is not on that component.

For any affinoid domain W=M(C)W=\mathcal M(C) containing xx, dim⁡KW≤d\dim_K W\le d by NAG-001592. Its evaluation kernel q\mathfrak q is minimal in CC. The prime choice below it uses Stacks 00E0(4). Indeed, if r⊊q\mathfrak r\subsetneq\mathfrak q were a minimal prime, a nonzero element of q/r\mathfrak q/\mathfrak r would cut the integral affinoid M(C/r)\mathcal M(C/\mathfrak r) down to dimension at most d−1d-1, contradicting dK(x)=dd_K(x)=d. This also covers d=0d=0 using the empty-locus assertion of NAG-002374.

Now let hh be in the maximal ideal of SVS_V. By NAG-001698, it is represented by an analytic section on an affinoid ordinary neighbourhood WW of xx in VV. Its image lies in the minimal prime ker⁡C(x)\ker_C(x). The localization Cker⁡C(x)C_{\ker_C(x)} is Artinian. Its maximal ideal is nil by Stacks 00E0(7), so some power of hh is annihilated in CC by an element nonzero at xx. That element is a unit in the ordinary stalk. Thus hh is nilpotent in SVS_V. The stalk is Noetherian by NAG-001747; a Noetherian local ring with nil maximal ideal is Artinian.

The natural local map Bp→SVB_{\mathfrak p}\to S_V is regular by NAG-001791; both rings are Artinian. The coherent restriction identity and Lemma C.1.12 prove (F.1.4). In particular pSV\mathfrak pS_V is the maximal ideal of SVS_V.

For completeness, use the definition of the pointed G-local ring as the filtered colimit over analytic domains containing xx. Taking ordinary stalks on those domains gives the same colimit: a representative section restricts to an ordinary neighbourhood in its domain, and that neighbourhood has an affinoid refinement containing xx. Finite intersections admit such refinements (NAG-002411). Tensor a composition series of MpM_{\mathfrak p} with every SVS_V, then pass to this filtered colimit. Each factor is the filtered colimit of the fields SV/pSVS_V/\mathfrak pS_V. The transition maps of these fields are unital and injective, so this colimit is a field. It is precisely the residue field of the colimit local ring, because every stage has maximal ideal pSV\mathfrak pS_V. The resulting series has exactly length⁡BpMp\operatorname{length}_{B_{\mathfrak p}}M_{\mathfrak p} nonzero simple factors. Taking M=BM=B also shows directly that the colimit ring is Artinian. No general Noetherianity theorem for G-local rings is needed here. ◻

fnd:nag002772:mixed-neighbourhood-puritylem:avoid

Purity of the neighborhood used for an effective coefficient.

Let Z=M(B)Z=\mathcal M(B), let p\mathfrak p be the minimal prime of the chosen reduced component Y=M(B/p)Y=\mathcal M(B/\mathfrak p), and let x∈Yx\in Y be away from the other components. Take an ordinary affinoid neighborhood W=M(C)W=\mathcal M(C) of xx in ZZ disjoint from those components. Actual domain base change and the finite quotient tensor identify the inverse image of WW in YY with the affinoid domain M(C/pC)\mathcal M(C/\mathfrak pC) of YY. Every analytic point of WW maps into YY, hence its kernel contains pC\mathfrak pC. By realization of every prime of CC as an analytic kernel, pC\mathfrak pC is contained in every prime of CC. Thus pC\mathfrak pC is a nil ideal by Stacks 00E0(7). Consequently CC and C/pCC/\mathfrak pC have the same minimal-prime quotients. Domain purity on YY gives purity of the latter domain of dimension dim⁡KY\dim_KY, so the full WW is pure of that dimension. Apply the generic Artin-length argument on the full WW. Passing to C/pCC/\mathfrak pC in the purity check changes none of the nilpotent structure used in computing the actual coefficient on WW.

Theorem. NAG-002772 (Cartier coefficient transport with scoped gluing; F8ST-C-R1). fnd:nag002772:cartierlem:avoid Let AA be generalized affinoid and f∈Af\in A a regular element. Put B=A/(f)B=A/(f), with its actual quotient analytic structure. For a component Y=M(A/p)Y=\mathcal M(A/\mathfrak p) of the Cartier zero space, p\mathfrak p minimal over (f)(f), its coefficient is cY(f)=length⁡Ap(Ap/fAp)=length⁡OX,x(OX,x/fOX,x),X=M(A),\begin{equation} c_Y(f)=\operatorname{length}_{A_{\mathfrak p}}(A_{\mathfrak p}/fA_{\mathfrak p}) =\operatorname{length}_{\mathcal O_{X,x}}(\mathcal O_{X,x}/f\mathcal O_{X,x}), \qquad X=\mathcal M(A), \end{equation} where xx can be chosen on YY, away from the other zero-space components, of maximal Abhyankar dimension in YY. Equivalently, the right side is the generic length of the analytic quotient sheaf. The lengths are finite positive integers. The assertion applies in particular to every codimension-one Cartier component. It makes no claim about an arbitrary closed integral subspace of the zero locus.

The effective coefficients are compatible with actual affinoid-domain restriction and ordinary gluing. For regular a,ba,b, product additivity is asserted when Min⁡A(a)∪Min⁡A(b)\operatorname{Min}_A(a)\cup\operatorname{Min}_A(b) is an antichain: no two distinct primes in this union contain one another. Here Min⁡A(a)\operatorname{Min}_A(a) means primes of the actual common ring AA minimal over (a)(a), with the empty set for a unit. A sufficient condition is that the two actual zero spaces are empty or pure of one common analytic dimension.

For regular meromorphic presentations, the differences div⁡(a)−div⁡(b)\operatorname{div}(a)-\operatorname{div}(b) are independent and glue among atlases for which this antichain condition holds for the cross-products on every actual common affinoid refinement. In particular this holds when all local numerator and denominator zero spaces are empty or pure of one fixed analytic dimension. The effective coefficient formula above needs no such purity assumption. No reduction of AA, BB, or their finite fibres is permitted. The previously unrestricted product and meromorphic clauses are not claimed closed in this induced revision.

Proof. The prime p/(f)\mathfrak p/(f) is minimal in BB, so Bp/(f)=Ap/(f)B_{\mathfrak p/(f)}=A_{\mathfrak p}/(f) is Artinian. Its length as a module over itself is its length as an ApA_{\mathfrak p}-module. Choose xx on the integral component YY of maximal Abhyankar dimension. It has kernel p/(f)\mathfrak p/(f) on BB and is on no other component, by the dimension-drop argument in Proposition F.1.11. Shrink to an affinoid ordinary neighbourhood in the zero space avoiding the other components. It is pure of dimension dim⁡KY\dim_KY by the mixed-neighbourhood argument above and domain purity NAG-002413. The proof of Proposition F.1.11 makes its ordinary stalk at xx, hence OM(B),x\mathcal O_{\mathcal M(B),x}, Artinian. The regular comparison NAG-001782 for BB, followed by Lemma C.1.12, gives its length as length⁡Bp/(f)\operatorname{length}B_{\mathfrak p/(f)}. The finite closed-immersion case of NAG-001735 identifies OM(B),x=OX,x⊗AB=OX,x/fOX,x.\mathcal O_{\mathcal M(B),x}=\mathcal O_{X,x}\otimes_A B =\mathcal O_{X,x}/f\mathcal O_{X,x}. Length over a local ring is unchanged when computed over a quotient annihilating the module, proving (F.1.5). The same argument on the pointed domain colimit proves the generic-length interpretation.

fnd:nag002772:effective-domain-transportlem:avoid For a domain M(C)⊂M(B)\mathcal M(C)\subset\mathcal M(B) and q∈Min⁡C\mathfrak q\in\operatorname{Min}C, its contraction r\mathfrak r is minimal in BB by flat going down (Stacks 00HS). Choose an analytic point of M(C)\mathcal M(C) with kernel q\mathfrak q, using NAG-001860. The upper arrow of NAG-001791 gives a regular local map Br→CqB_{\mathfrak r}\to C_{\mathfrak q}. Both rings are Artinian, so their lengths agree. The same calculation holds for any finite coherent module. Actual domain/quotient base change, NAG-001702/2.36.9, identifies this quotient with the restriction of A/(f)A/(f). Flatness preserves the injectivity of multiplication by ff, so the restricted equation is still regular. This proves compatibility on every actual domain and hence on affinoid refinements of overlaps.

We first supply the missing component inclusion under the stated antichain hypothesis. In any Noetherian ring, Min⁡A(ab)\operatorname{Min}_A(ab) is the set of inclusion-minimal members of Min⁡A(a)∪Min⁡A(b)\operatorname{Min}_A(a)\cup\operatorname{Min}_A(b). Indeed a prime containing abab contains a factor and therefore a prime minimal over that factor by Stacks 00E0(4). A prime minimal over abab must equal that smaller prime. Conversely, if pp is inclusion-minimal in the union and q⊂pq\subset p contains abab, choose by Stacks 00E0(4) a factor-minimal prime r⊂qr\subset q. Then r⊂q⊂pr\subset q\subset p forces r=q=pr=q=p. Thus, for an antichain, every factor component occurs in the product: Min⁡A(ab)=Min⁡A(a)∪Min⁡A(b).\operatorname{Min}_A(ab)=\operatorname{Min}_A(a)\cup\operatorname{Min}_A(b).

The analytic purity criterion for this hypothesis uses only frozen NAG-002374. If the factor zero spaces are empty or pure of one common dimension dd, all primes in the union define integral affinoids of dimension dd. A strict inclusion q⊊pq\subsetneq p would make the nonzero ideal p/qp/q in A/qA/q cut out M(A/p)\mathcal M(A/p) of dimension at most d−1d-1, or the empty space when d=0d=0. This is impossible. Their union is therefore an antichain, and the product zero space is also empty or pure of dimension dd.

For regular a,ba,b, multiplication by bb gives the exact sequence 0⟶A/(a)→ b A/(ab)⟶A/(b)⟶0.0\longrightarrow A/(a)\xrightarrow{\,b\,} A/(ab) \longrightarrow A/(b)\longrightarrow0. At each prime in the now identified component union, each factor quotient is either zero (a unit after localization) or Artinian. A prime in the union containing a factor is minimal over that factor; otherwise a smaller factor-minimal prime would violate the antichain condition. Add localized finite lengths to get the coefficient identity on every component, including all factor components.

For meromorphic representatives in the stated scope, equality a/b=c/ea/b=c/e gives ae=bcae=bc on an actual common affinoid restriction: localization at regular elements is injective. The cross-product antichain hypotheses hold there by assumption. In the common pure-zero-space case they hold by the preceding dimension-drop argument: quotient/domain base change identifies the restricted zero spaces with actual domains in the original zero spaces, and NAG-002413 preserves their purity. Additivity yields div⁡(a)−div⁡(b)=div⁡(c)−div⁡(e)\operatorname{div}(a)-\operatorname{div}(b)=\operatorname{div}(c)-\operatorname{div}(e). fnd:nag002772:ordinary-quotient-shrinkinglem:avoid On a good space, local quotient representatives may be taken on ordinary neighbourhoods. If regularity is initially known only at a germ, the kernel of multiplication is coherent, so it vanishes after shrinking. For a unit of the local total quotient ring, a representative numerator is regular as well: multiplying by an inverse shows that it divides a regular element. These observations justify regular numerator and denominator representatives after the necessary restrictions. Domain compatibility just proved makes their local cycles glue. ◻

Theorem. NAG-002790 (Additivity in the verified scope; F8CA-C). fnd:nag002790:antichain-additivitylem:avoid Let AA be Noetherian and let a,ba,b be non-zero-divisors. Assume Min⁡A(a)∪Min⁡A(b)\operatorname{Min}_A(a)\cup\operatorname{Min}_A(b) is an antichain. Define the effective cycle of a regular equation by its actual components and the ambient lengths div⁡(a)=∑p∈Min⁡A(a)length⁡Ap(Ap/aAp)[Spec⁡(A/p)].\operatorname{div}(a)=\sum_{p\in\operatorname{Min}_A(a)} \operatorname{length}_{A_p}(A_p/aA_p)[\operatorname{Spec}(A/p)]. Then div⁡(ab)=div⁡(a)+div⁡(b)\operatorname{div}(ab)=\operatorname{div}(a)+\operatorname{div}(b). For generalized affinoids, the same assertion holds for the actual analytic cycles and their verified effective coefficient transport. In particular it holds under Lemma C.1.57’s hypotheses.

Proof. Lemma C.1.29 identifies all components on both sides, including the converse inclusion missing from the original proof. At a prime pp in this union, each of Ap/(a)A_p/(a) and Ap/(b)A_p/(b) is either zero (when the factor is a unit in ApA_p) or Artinian (when pp is minimal over that factor). It is minimal over every factor it contains: otherwise a smaller minimal prime of that factor would contradict the antichain condition. The regularity of bb makes the sequence 0⟶A/(a)→ b A/(ab)⟶A/(b)⟶00\longrightarrow A/(a)\xrightarrow{\,b\,}A/(ab) \longrightarrow A/(b)\longrightarrow0 exact. Localize at pp and add finite lengths. This gives the coefficient equality on the now identical component sets, the selected algebraic antichain clause. The effective analytic comparison and actual-domain compatibility are the unchanged, independently verified portions of NAG-002772 effective coefficients and actual domain transport; no new length comparison is being inferred from flatness. ◻

Proposition. NAG-003099 (Ambient regular representatives; F8NM-R). fnd:nag003099:regular-quotient-representativeslem:avoid Ambient S1S_1 is unnecessary for local regular numerator and denominator representatives of a unit in the total quotient sheaf. Such representatives remain regular after every actual affinoid restriction, and their pullbacks to the reduced components are nonzero and regular. Equality of two representations a/b=c/da/b=c/d gives ad=bcad=bc after an actual common ordinary refinement. Componentwise nonvanishing alone is not a substitute for ambient regularity.

Proof. For a ring AA, localization at its regular elements is injective. If a/ba/b is a unit, write its inverse as c/dc/d, where b,db,d are regular. Equality in the localization gives ac=bdac=bd in AA. The product bdbd is regular. If az=0az=0, then bdz=acz=0bdz=acz=0, so z=0z=0. Thus aa is regular; the same argument applies to cc. No associated-prime purity hypothesis is used.

The good-space construction at the end of the proof of frozen NAG-002772 takes these quotients on ordinary neighbourhoods. When injectivity is first known at a germ, the multiplication kernel is coherent and vanishes after shrinking. A finite collection of numerators, denominators, inverse representatives and equalities can be handled on one smaller neighbourhood. This is the exact coherent shrinking scope of that input, not a claim about arbitrary meromorphic objects on a nongood space. Affinoid-domain restriction is flat (NAG-001832), so it preserves the regular multiplication maps.

A regular element avoids every minimal prime, as checked in reviewed F8CR-L. Its image in the domain A/qiA/\mathfrak q_i is therefore nonzero and regular. F8CR-S and flat actual restrictions supply the corresponding analytic component regularity. The same injection of regular localization proves the cross-product identity. The failure of the last proposed weakening is shown in the algebraic example below. ◻

fnd:nag003099:algebraic-obstructionlem:avoid In A=K⟨t,e⟩/(e2,te)A=K\langle t,e\rangle/(e^2,te), every class has the unique module form a(t)+eca(t)+e c, with a∈K⟨t⟩a\in K\langle t\rangle and c∈Kc\in K. The nilradical is (e)(e), the underlying space is the pure disc, and ann⁡(e)=(t,e)\operatorname{ann}(e)=(t,e) is an embedded associated prime. At the minimal prime (e)(e), tt is invertible and kills ee, so the generic length is one. But A/(t)=K[e]/(e2)A/(t)=K[e]/(e^2) has length two at the rigid origin; the reduced component quotient is KK. Since te=0te=0 with e≠0e\ne0, tt is not a Cartier equation. It also cannot be a unit of the ambient total quotient ring: if t(c/d)=1t(c/d)=1 with dd regular, then tc=dtc=d, and multiplication by ee contradicts regularity of dd. Componentwise meromorphic invertibility does not solve this.