G · Stable inertia and original-field models

25 NAG Tags

Let KK be a complete nontrivially real-valued field, let O=K∘O=K^\circ, and write m\mathfrak m and kk for its maximal ideal and residue field. The valuation has rank one and need not be discrete. An F5 model of a smooth projective geometrically connected curve C/KC/K means a projective, flat, finitely presented nodal OO-curve, together with its generic-fibre identification with CC, whose special fibre is geometrically connected and reduced. The marking is empty throughout this note. Nodal means at worst nodal of pure relative dimension one; branches need not split over kk.

Stable families and their generic-fibre identifications have the conventions of the admitted source 4.14. We use its exact results 4.14.4 and 4.14.5. The example reuses the valued field and extension from C.1.85 and C.1.86.

Lemma. NAG-002913 (A necessary inertia condition for original-field existence). Let C/KC/K be as above, of genus g≥2g\geq2. Let L/KL/K be finite Galois, with group GG, and give LL its unique extending absolute value. Put O′=L∘O'=L^\circ, let k′k' be its residue field, and set I=ker⁡(G⟶Aut⁡(k′/k)).I=\ker\bigl(G\longrightarrow\operatorname{Aut}(k'/k)\bigr). Suppose a stable model Z/O′\mathcal Z/O' of CLC_L is supplied with its generic-fibre identification. Equip it with the unique semilinear GG-action of 4.14.5. If CC admits an F5 model over OO, then every element of II induces the identity morphism on the entire k′k'-scheme Zk′\mathcal Z_{k'}. No tameness or residue-separability hypothesis is needed for this necessary condition.

Proof. Suppose f:X→Spec⁡Of:\mathcal X\to\operatorname{Spec}O is an F5 model. A nontrivial rank-one valuation ring has only its generic and closed points. Hence all geometric fibres of ff are reduced, connected curves of dimension one: the generic fibre is smooth and geometrically connected, and the special fibre has the stated properties. The criterion in Stacks Tag 0E6T, with Stacks Tag 0E6L, makes ff a prestable family. The same lemma gives a finite locally free R1f∗OXR^1f_*\mathcal O_{\mathcal X} with arbitrary base change. Its rank is constant on the connected spectrum of OO and equals gg at the generic point. Thus this is a prestable family of genus gg in the sense of Stacks Tag 0E6V.

Apply the arbitrary-base stabilization theorem Stacks Tag 0E8A. It gives a stable family Y/O\mathcal Y/O and a morphism X→Y\mathcal X\to\mathcal Y whose fibres contract rational tails and bridges. The smooth geometrically connected generic curve of genus at least two has none, so this morphism identifies YK\mathcal Y_K with CC. This use of the stack of stable families requires no new projectivity theorem for Y\mathcal Y.

For the scheme convention, the absence of rational tails and bridges on the stable closed fibre gives an ample relative dualizing sheaf on a neighbourhood of the closed point by Stacks Tag 0E74. As the base is local, that neighbourhood is the whole base. The ample sheaf makes Y\mathcal Y a scheme by Stacks Tag 0D32.

Both YO′\mathcal Y_{O'} and Z\mathcal Z are stable families of genus gg. Their specified generic identifications give a unique isomorphism α:YO′→ ∼ Z\alpha:\mathcal Y_{O'}\xrightarrow{\ \sim\ }\mathcal Z by 4.14.4, which applies to arbitrary valuation rings. The canonical semilinear action on the base change YO′\mathcal Y_{O'} extends the canonical action on CLC_L. Transporting it by α\alpha must therefore give the specified action on Z\mathcal Z, by the uniqueness in 4.14.5.

On the full special fibre Yk′=Yk×kSpec⁡k′\mathcal Y_{k'}=\mathcal Y_k\times_k\operatorname{Spec}k', the canonical action of σ∈G\sigma\in G is id⁡Yk×Spec⁡(σ‾−1)\operatorname{id}_{\mathcal Y_k}\times \operatorname{Spec}(\overline{\sigma}^{-1}). For σ∈I\sigma\in I, this is the identity morphism, including the structure sheaf. The equivariant isomorphism αk′\alpha_{k'} transfers that identity to Zk′\mathcal Z_{k'}, as required. ◻

Proposition. NAG-002914 (A tame nondiscrete curve with no original-field F5 model). Use the admitted field and quadratic extension K=⋃r≥0C((t1/3r))^,v(t)=1,L=K(s),s2=t.K=\widehat{\bigcup_{r\geq0}\mathbf C((t^{1/3^r}))},\qquad v(t)=1,\qquad L=K(s),\quad s^2=t. Then v(K×)=Z[1/3]v(K^\times)=\mathbf Z[1/3] is dense, L/KL/K is tame Galois of degree two, and both residue fields are C\mathbf C. The curve C:Z4=t2(X4+Y4) ⊂ PK2C:\quad Z^4=t^2(X^4+Y^4)\ \subset\ \mathbf P^2_K is smooth, projective and geometrically connected of genus three. It admits a smooth projective stable model over L∘L^\circ, but admits no F5 model over K∘K^\circ. The geometric dual graph of the supplied stable model has a single vertex and no edges, and is fixed by the Galois action. Thus invariant stable vertices do not imply even existence of some original-field F5 model.

Proof. The field, value group, completeness and extension data are exactly those of C.1.85 and C.1.86, with k=Ck=\mathbf C and ss in place of bb. We reuse these facts. In particular G={1,σ}G=\{1,\sigma\} with σ(s)=−s\sigma(s)=-s, the residue field does not change, and I=GI=G. No assertion that O′/OO'/O is finite or etale is being used; the admitted coefficient result in fact proves that O′O' is not finite over OO.

For the homogeneous polynomial F=Z4−t2(X4+Y4)F=Z^4-t^2(X^4+Y^4) the partial derivatives are −4t2X3,−4t2Y3,4Z3.-4t^2X^3,\qquad -4t^2Y^3,\qquad 4Z^3. They cannot vanish simultaneously at a projective point over any field extension of KK. On each affine chart the dehomogenized partials likewise have no common zero on the curve, by the displayed formula and the degree-four Euler identity. Inverting a nonzero partial gives a standard smooth presentation as in Stacks Tag 00T6; Stacks Tag 00T7 gives smoothness and relative dimension one. Thus CC is a geometrically smooth plane quartic. These smooth hypersurfaces are pure one-dimensional and have no embedded points, so Stacks Tag 0BYD applies over KK and over every field extension. It gives H0(CE,OCE)=EH^0(C_E,\mathcal O_{C_E})=E and genus (4−1)(4−2)/2=3(4-1)(4-2)/2=3. In particular CC is geometrically connected: a disconnection over an algebraic closure would give a nontrivial idempotent in that field of global functions. Projectivity is part of its displayed construction.

Let D/CD/\mathbf C be the plane quartic W4=X4+Y4W^4=X^4+Y^4 and put D=D×CSpec⁡O′.\mathcal D=D\times_{\mathbf C}\operatorname{Spec}O'. The same computation makes DD a smooth projective geometrically connected curve of genus three. Hence D\mathcal D is a projective smooth family with these geometric fibres. It is stable by Stacks Tag 0E6T and Stacks Tag 0E75: a smooth connected genus-three curve has no rational tails or bridges. There is an LL-isomorphism ϕ:CL→ ∼ DL,[X:Y:Z]⟼[X:Y:Z/s].\phi:C_L\xrightarrow{\ \sim\ }D_L, \qquad [X:Y:Z]\longmapsto[X:Y:Z/s]. Indeed s4=t2s^4=t^2, and multiplication of the third coordinate by ss gives its inverse. This supplies the required stable model with generic-fibre identification.

Under ϕ\phi, the canonical semilinear action of σ\sigma is the coefficient action of σ\sigma together with τ:[X:Y:W]⟼[X:Y:−W].\tau:[X:Y:W]\longmapsto[X:Y:-W]. For example, the pullback of W=Z/sW=Z/s is Z/(−s)=−WZ/(-s)=-W. This action extends to D\mathcal D, since τ\tau is defined over C\mathbf C and preserves its equation. Uniqueness in 4.14.5 identifies it with the specified action on the stable model. On the special fibre it is τ\tau, which sends [1:0:1][1:0:1] to the distinct point [1:0:−1][1:0:-1]. It is therefore not the identity on the special fibre.

If an F5 model of CC over OO existed, Lemma G.0.1 would force this action to be the identity, a contradiction. Finally, the geometric special fibre of D\mathcal D is a smooth connected curve, hence has one irreducible component and no nodes. Its geometric dual graph therefore has one vertex and no edges; the action fixes that graph. This graph statement does not assert that its geometric component vertex descends to an original-field F5 vertex. ◻

Finite model measures and logarithmic integrals

Let KK be complete and nontrivially discretely valued, with its specified absolute value, valuation ring RR, residue field kk, uniformizer π\pi, and λ=−log⁡∣π∣>0.\lambda=-\log|\pi|>0. We do not renormalize λ\lambda to one or pass to an algebraic closure. Let XX be an integral proper KK-variety of dimension d≥2d\geq2. Take a proper flat integral normal finite-type RR-model X\mathcal X with identified generic fibre XX. Retain explicitly the following local data: every height-one local ring used below is a DVR, and the special fibre has finitely many irreducible components VV, all of dimension dd. These are hypotheses, not assertions about arbitrary analytic local rings. Smoothness or strict semistability of the model is not required.

Write AV=OX,ηVA_V=\mathcal O_{\mathcal X,\eta_V} and normalize ord⁡V\operatorname{ord}_V by a uniformizer of this DVR. Then bV=ord⁡V(π)=length⁡AV(AV/πAV)>0,[Xk]=∑VbV[V].\begin{equation} b_V=\operatorname{ord}_V(\pi)=\operatorname{length}_{A_V}(A_V/\pi A_V)>0, \qquad [\mathcal X_k]=\sum_V b_V[V]. \end{equation} Thus a nonreduced special fibre is allowed. Its bVb_V is not discarded. The point ξV∈Xan\xi_V\in X^\mathrm{an} is specified on K(X)K(X) by ∣a(ξV)∣=∣π∣ord⁡V(a)/bV(a≠0).\begin{equation} |a(\xi_V)|=|\pi|^{\operatorname{ord}_V(a)/b_V} \quad(a\ne0). \end{equation} Distinct components give distinct points: the bounded valuation ring in (G.0.2) has centre ηV\eta_V on X\mathcal X, so equality of points would give equality of their centres. If the points were equal, their valuations would give the same valuation ring. The two maps from its spectrum to the separated model agree at the generic point. The closed diagonal and integrality of that spectrum force them to agree everywhere, including at the centre. The model ring AVA_V, the ordinary analytic stalk OXan,ξV\mathcal O_{X^\mathrm{an},\xi_V}, and the completed residue field H(ξV)\mathcal H(\xi_V) are different objects. Only the first is a DVR hypothesis here.

Choose line bundles L1,…,Ld,ML_1,\ldots,L_d,M on XX and model lines Li∣X≃Li⊗ei,M∣X≃M⊗e0,ei,e0∈Z>0.\mathcal L_i|_X\simeq L_i^{\otimes e_i},\qquad \mathcal M|_X\simeq M^{\otimes e_0},\qquad e_i,e_0\in\mathbf Z_{>0}. Use their continuous root model metrics: a model frame has norm one, and if s⊗e0=aes^{\otimes e_0}=a e in a model frame ee, then ∥s∥=∣a∣1/e0\|s\|=|a|^{1/e_0} on the domain reducing to that frame. The model metrics are supplied as part of these data. Put E=∏i=1dei,δV=deg⁡L1∣V,…,Ld∣V,k(V),μX=∑VbVδVE δξV.\begin{equation} E=\prod_{i=1}^d e_i,\quad \delta_V=\deg_{\mathcal L_1|_V,\ldots,\mathcal L_d|_V,k}(V),\quad \mu_{\mathcal X}=\sum_V\frac{b_V\delta_V}{E}\,\delta_{\xi_V}. \end{equation} The degree is the mixed algebraic degree over kk, with its usual full residue-field dimensions. It is not a degree over the field of constants of VV followed by omission of that field degree. We use the specified integers δV\delta_V and the finite measure above. The measure may be signed. Positivity requires δV≥0\delta_V\geq0. Its mass is exactly E−1∑VbVδVE^{-1}\sum_V b_V\delta_V; identifying this with a generic-fibre degree would use an additional algebraic specialization identity. The mixed convention contains no extra d!d!. For a proper dd-dimensional integral kk-scheme WW and invertible sheaves M1,…,MdM_1,\ldots,M_d, use the mixed degree δ(W;M1,…,Md)=deg⁡k(c1(M1)⋯c1(Md)∩[W]).\delta(W;M_1,\ldots,M_d) =\deg_k\bigl(c_1(M_1)\cdots c_1(M_d)\cap[W]\bigr). The selected supported operations in NAG-003112 compute the actual multiples of the coordinate divisor cycles, and their push into the ambient support gives the first Chern operation. The degree convention is the one used with NAG-003117.

Let ss be a nonzero rational section of MM. A model frame at ηV\eta_V writes s⊗e0=aVeVs^{\otimes e_0}=a_V e_V, and put cV=ord⁡V(aV)c_V=\operatorname{ord}_V(a_V). A change of model frame multiplies aVa_V by a DVR unit, so cVc_V is well-defined. Write gs=−log⁡∥s∥g_s=-\log\|s\| on the dense open where ss is regular and nonvanishing.

Theorem. NAG-002830 (Exact logarithmic integral on a proper model). Under the standing finite-model data, every ξV\xi_V lies in the analytification of every nonempty Zariski open subset of XX. In particular gsg_s is finite at all the atoms of μX\mu_{\mathcal X} and is integrable with respect to its total variation. The exact formulas are gs(ξV)=λcVe0bV,∫Xangs dμX=λe0E∑VcVδV,∫Xan∣gs∣ d∣μX∣=λe0E∑V∣cVδV∣.\begin{align} g_s(\xi_V)&=\frac{\lambda c_V}{e_0b_V},\\ \int_{X^\mathrm{an}}g_s\,d\mu_{\mathcal X} &=\frac{\lambda}{e_0E}\sum_V c_V\delta_V,\\ \int_{X^\mathrm{an}}|g_s|\,d|\mu_{\mathcal X}| &=\frac{\lambda}{e_0E}\sum_V|c_V\delta_V|. \end{align} Every proper algebraic closed subset of XX has ∣μX∣|\mu_{\mathcal X}|-measure zero. For a∈K×a\in K^\times, replacing ss by asas changes (G.0.5) by −log⁡∣a∣ μX(Xan)-\log|a|\,\mu_{\mathcal X}(X^\mathrm{an}).

Direct inputs NAG-001707, the standing finite-model data.

Proof. On an affine model neighbourhood of ηV\eta_V, the valuation in (G.0.2) is nonnegative on the model coordinate ring and extends the given absolute value on KK. Indeed, ord⁡V(u)=0\operatorname{ord}_V(u)=0 for u∈R×u\in R^\times, and ord⁡V(π)=bV\operatorname{ord}_V(\pi)=b_V. The valuation inequality and multiplicativity therefore give an analytic point by NAG-001707. Its kernel in the generic-fibre coordinate ring is zero, since the valuation of each nonzero rational function is finite. Its underlying scheme point is the generic point of XX. Hence it belongs to every nonempty Zariski open, including the locus where ss is regular and nonvanishing. Formula (G.0.2) and the root metric give (G.0.4). This also shows that an arbitrary value assigned on the omitted divisor does not affect the integral.

A finite sum of distinct Dirac masses has total variation obtained by taking the absolute values of its coefficients. Substituting (G.0.4) into (G.0.3) proves both integral formulas. The positive integer bVb_V cancels between the mass and the value; it has not been set to one. The zero-mass assertion follows from the generic scheme point of every atom. Finally cV(as)=cV(s)+e0ord⁡V(a)=cV(s)+e0bVord⁡R(a).c_V(as)=c_V(s)+e_0\operatorname{ord}_V(a) =c_V(s)+e_0b_V\operatorname{ord}_R(a). Since −log⁡∣a∣=λord⁡R(a)-\log|a|=\lambda\operatorname{ord}_R(a), substitution proves the last assertion with its stated sign. ◻

Proposition. NAG-002851 (Weak convergence does not control the logarithm). For every d≥2d\geq2 and KK as in the standing finite-model data there are genuine semipositive product-model measures μj\mu_j of mixed mass one on X=(PK1)dX=(\mathbf P^1_K)^d and a fixed model-metrized line (M,h)(M,h) with a nonzero regular section ss such that μj⟹δo,∫−log⁡∥s∥h dμj=jλ⟶+∞,o∈Z(s).\mu_j\Longrightarrow\delta_o,\qquad \int-\log\|s\|_h\,d\mu_j=j\lambda\longrightarrow+\infty, \qquad o\in Z(s). Thus finite-model integrability and weak measure convergence, even with an unchanged metric on MM, do not imply logarithmic integrability of the limit. This does not contradict convergence theorems that also require uniform convergence of the semipositive metrics defining the measures.

Direct inputs NAG-000182, NAG-001707, NAG-001709, NAG-001714, the standing finite-model data, the mixed algebraic degree convention.

Proof. For each j≥1j\geq1 use the model (PR1)d(\mathbf P^1_R)^d with generic-fibre identification Ti=πjSiT_i=\pi^jS_i, and the model lines pr⁡i∗O(1)\operatorname{pr}_i^*\mathcal O(1). They are semipositive: on any special-fibre curve their degrees are nonnegative, since they are generated by global sections. The special fibre has one reduced component, b=1b=1, and its mixed degree is one, by intersecting the dd coordinate divisors at one kk-rational product point. Its model measure is therefore μj=δξj\mu_j=\delta_{\xi_j}, where ∣∑m∈NdamTm∣(ξj)=max⁡m∣am∣ ∣π∣j(m1+⋯+md).\left|\sum_{m\in\mathbf N^d}a_mT^m\right|(\xi_j) =\max_m|a_m|\,|\pi|^{j(m_1+\cdots+m_d)}. Gauss multiplicativity gives the point, and its polynomial evaluations converge to ∣a0∣|a_0|, the evaluations at the origin oo. The evaluation topology in NAG-001707 gives ξj→o\xi_j\to o in the common affine chart and thus in XanX^\mathrm{an}. Consequently the Dirac masses converge against every continuous analytic test, not just radial tests.

Use M=pr⁡1∗O(∞)M=\operatorname{pr}_1^*\mathcal O(\infty) with its standard model metric and its global section s=T1s=T_1 relative to the frame 11 on the finite chart. On that chart ∥s(x)∥h=∣T1(x)∣max⁡(1,∣T1(x)∣).\|s(x)\|_h=\frac{|T_1(x)|}{\max(1,|T_1(x)|)}. This extends as a regular section nonzero at infinity, with zero divisor {T1=0}\{T_1=0\}. At ξj\xi_j the norm is ∣π∣j|\pi|^j, so the integral is jλj\lambda. At oo it is zero and its negative logarithm is infinite. For fixed RR, the tails are (jλ−R)+(j\lambda-R)_+ and are unbounded in jj. The curvature metrics defining μj\mu_j are moving; their uniform convergence is not claimed. ◻

Coherent tensor projection for finite analytic maps

Fix a complete field KK with its specified real-valued non-Archimedean absolute value. Trivial valuation and generalized affinoid radii are allowed. Spaces in this supplement are good and Hausdorff. Coherent modules live on the analytic GG-site. A line bundle means a coherent module which is free of rank one on ordinary open neighbourhoods. The symbol q∗q^* on coherent modules denotes ordinary tensor pullback; it is not a derived functor.

For a norm-admissible finite morphism q:Y→Xq:Y\to X, the full inverse image of V=M(A)V=\mathcal M(A) is M(B)\mathcal M(B), where BB is finite over AA with its canonical finite-module norm. If NN is finite over BB, compose controlled finite free presentations Aa↠BA^a\twoheadrightarrow B and Bb↠NB^b\twoheadrightarrow N. Their lifting bounds compare the iterated quotient norm with the canonical finite-AA-module norm. Thus restriction of scalars also preserves the canonical topology on NN. Every tensor used below between finite modules has its canonical topology. The comparison with the completed analytic tensor is provided by NAG-001565 and NAG-001702. Algebraic finiteness with an unrelated Banach norm would not suffice.

Proposition. (Coherent tensor projection). Let q:Y→Xq:Y\to X be norm-admissible and finite. For F∈Coh⁡(YG)\mathcal F\in\operatorname{Coh}(Y_G) and E∈Coh⁡(XG)\mathcal E\in\operatorname{Coh}(X_G) there is a canonical isomorphism q∗(F⊗OYq∗E)≃q∗F⊗OXE.\begin{equation} q_*(\mathcal F\otimes_{\mathcal O_Y}q^*\mathcal E) \simeq q_*\mathcal F\otimes_{\mathcal O_X}\mathcal E. \end{equation} It commutes with affinoid-domain restriction and analytic base change. Neither F\mathcal F nor E\mathcal E is required to be locally free.

NAG-001565, NAG-001624, NAG-001702, and NAG-001752

Proof. On V=M(A)V=\mathcal M(A) and its full inverse image M(B)\mathcal M(B), let MM and NN represent E\mathcal E and F\mathcal F. The map N⊗B(B⊗AM)⟶N⊗AM,n⊗(b⊗m)⟼bn⊗mN\otimes_B(B\otimes_A M)\longrightarrow N\otimes_A M,\qquad n\otimes(b\otimes m)\longmapsto bn\otimes m has inverse n⊗m↦n⊗(1⊗m)n\otimes m\mapsto n\otimes(1\otimes m). These are maps of finite AA-modules, so they are bounded, with bounded inverses, for the canonical norms. The finite-tensor theorem identifies the modules with the analytic completed constructions.

For a bounded affinoid base change A→CA\to C, put D=B⊗ACD=B\otimes_A C. The same formulas identify both base changes with N⊗AM⊗ACN\otimes_A M\otimes_A C. In particular they respect domain restrictions. The finite coherent direct-image theorem and affinoid coherent equivalence therefore glue them to (G.0.7). This argument concerns an isomorphism of tensors, and asserts no exactness for tensoring with an arbitrary E\mathcal E. ◻

Norm integrals and logarithmic limits for finite models

Theorem. NAG-002837 (Norm integration under a component-degree criterion). Let q:Y→Xq:\mathcal Y\to\mathcal X be finite locally free of constant rank N≥1N\geq1, with both models in the class of the standing finite-model data and with integral generic fibres Y,XY,X. Give q∗Liq^*L_i and q∗Mq^*M their pulled-back model metrics. For each VV, set BV=(q∗OY)ηV,AV=OX,ηV.B_V=(q_*\mathcal O_{\mathcal Y})_{\eta_V},\qquad A_V=\mathcal O_{\mathcal X,\eta_V}. Retain the ordinary algebraic local descriptions: BVB_V is the whole finite free rank-NN algebra, and all its maximal localizations are the DVRs OY,ηW\mathcal O_{\mathcal Y,\eta_W} for the components WW over VV. Let tVt_V be a uniformizer of AVA_V, and put rW=ord⁡W(tV),fW=[κ(W):κ(V)],δW=deg⁡q∗L1∣W,…,q∗Ld∣W,k(W).r_W=\operatorname{ord}_W(t_V),\quad f_W=[\kappa(W):\kappa(V)],\quad \delta_W=\deg_{q^*\mathcal L_1|_W,\ldots,q^*\mathcal L_d|_W,k}(W). Assume the finite list of component-degree identities δW=fWδV(W∣V).\begin{equation} \delta_W=f_W\delta_V\quad(W\mid V). \end{equation} For a nonzero rational section ss of q∗Mq^*M, use the determinant section Nm⁡q(s)\operatorname{Nm}_q(s) of Nm⁡q(q∗M)≃M⊗N\operatorname{Nm}_q(q^*M)\simeq M^{\otimes N}, with the metric induced by the model line M⊗N\mathcal M^{\otimes N} and root exponent e0e_0. Then q∗anμY=NμX,∫Xan−log⁡∥Nm⁡q(s)∥ dμX=∫Yan−log⁡∥s∥ dμY.\begin{align} q^\mathrm{an}_*\mu_{\mathcal Y}&=N\mu_{\mathcal X},\\ \int_{X^\mathrm{an}}-\log\|\operatorname{Nm}_q(s)\|\,d\mu_{\mathcal X} &=\int_{Y^\mathrm{an}}-\log\|s\|\,d\mu_{\mathcal Y}. \end{align} Both logarithmic integrals are absolutely convergent with respect to total variation. All components, full residue degrees and unreduced DVR quotients are retained, including in inseparable cases. The criterion holds for the coordinate power map Ti=SiniT_i=S_i^{n_i} on (PR1)d(\mathbf P^1_R)^d, with ni≥1n_i\geq1, Li=pr⁡i∗O(1)\mathcal L_i=\operatorname{pr}_i^*\mathcal O(1), and ei=1e_i=1.

Direct inputs NAG-002543, NAG-002552, Theorem G.0.3, the mixed algebraic degree convention.

Proof. Each local DVR extension gives bW=rWbV,∑W∣VfWrW=N.\begin{equation} b_W=r_Wb_V, \qquad \sum_{W\mid V}f_Wr_W=N. \end{equation} The first equality follows by writing π=utVbV\pi=u t_V^{b_V} with uu a unit. For the second, apply NAG-002543 to the regular sequence (tV)(t_V) in AVA_V and the full algebra BVB_V. Its summand at WW is fWrWf_Wr_W because length⁡OY,ηW(OY,ηW/tV)=rW\operatorname{length}_{\mathcal O_{\mathcal Y,\eta_W}} (\mathcal O_{\mathcal Y,\eta_W}/t_V)=r_W. This proof does not require a separable residue extension. Equations (G.0.2) and (G.0.11) give qan(ξW)=ξVq^\mathrm{an}(\xi_W)=\xi_V. Thus the mass pushed to ξV\xi_V is ∑W∣VbWδWE=bVδVE∑W∣VrWfW=NbVδVE,\sum_{W\mid V}\frac{b_W\delta_W}{E} =\frac{b_V\delta_V}{E}\sum_{W\mid V}r_Wf_W =N\frac{b_V\delta_V}{E}, which proves (G.0.9).

Let F=K(X)F=K(X) and L=K(Y)L=K(Y). The generic algebra BV⊗AVFB_V\otimes_{A_V}F is LL, a field extension of full degree NN. For a∈L×a\in L^\times, NAG-002552 gives ord⁡VNm⁡L/F(a)=∑W∣VfWord⁡W(a).\begin{equation} \operatorname{ord}_V\operatorname{Nm}_{L/F}(a)=\sum_{W\mid V}f_W\operatorname{ord}_W(a). \end{equation} To include a rational aa, choose a nonzero z∈AVz\in A_V with za∈BVza\in B_V, possible because BVB_V spans LL over FF. Apply that lemma to zaza and to zz, and subtract. Multiplication by aa on LL is invertible even for an inseparable extension; its determinant, rather than a product over distinct embeddings, is the norm used here.

In a local frame ee of M\mathcal M, write s⊗e0=aq∗es^{\otimes e_0}=a q^*e. The coefficient of (Nm⁡qs)⊗e0(\operatorname{Nm}_qs)^{\otimes e_0} in e⊗Ne^{\otimes N} is Nm⁡L/F(a)\operatorname{Nm}_{L/F}(a). Indeed, exterior multiplication on the rank-NN module gives det⁡(ma)\det(m_a), while changing ee by a unit uu multiplies the determinant frame by uNu^N. These changes are exactly those of M⊗N\mathcal M^{\otimes N}, so the construction glues and has the stated pulled-back tensor-power metric. Consequently, if cW=ord⁡W(a)c_W=\operatorname{ord}_W(a), then the downstairs coefficient is ∑W∣VfWcW\sum_{W\mid V}f_Wc_W. Theorem G.0.3 now gives ∫−log⁡∥Nm⁡q(s)∥ dμX=λe0E∑VδV∑W∣VfWcW=λe0E∑WδWcW.\int-\log\|\operatorname{Nm}_q(s)\|\,d\mu_{\mathcal X} =\frac{\lambda}{e_0E}\sum_V\delta_V\sum_{W\mid V}f_Wc_W =\frac{\lambda}{e_0E}\sum_W\delta_Wc_W. The last expression is the upstairs integral. Absolute convergence is part of the same finite-sum theorem.

For the coordinate power map, the exponent classes ∏iSiai\prod_iS_i^{a_i}, 0≤ai<ni0\leq a_i<n_i, are a free basis on every coordinate chart. They give finite flat rank N=∏iniN=\prod_i n_i and the full special-fibre field degree NN, including inseparable degrees. The single reduced component has mixed degree one downstairs. Upstairs the pullback of the iith coordinate line is pr⁡i∗O(ni)\operatorname{pr}_i^*\mathcal O(n_i). Choose the coordinate section with divisor ni{Si=0}n_i\{S_i=0\} for each ii. Their proper intersection at the rational product origin has local algebra k[S1,…,Sd]/(S1n1,…,Sdnd),k[S_1,\ldots,S_d]/(S_1^{n_1},\ldots,S_d^{n_d}), whose monomial basis has length NN. This computes the mixed degree as NN and verifies (G.0.8) in the stated example. ◻

Assume in this section that XanX^\mathrm{an} is compact Hausdorff; the projective case is supplied by NAG-001714. Let ss be a nonzero global regular section of a line bundle MM, D=Z(s)D=Z(s), and hh a continuous metric. Extend g=−log⁡∥s∥hg=-\log\|s\|_h by +∞+\infty on DanD^\mathrm{an}. It is bounded below. For each jj, let hjh_j be a continuous metric and μj\mu_j a positive finite model measure on XanX^\mathrm{an}, with its generic divisorial atoms as above. Let gj=−log⁡∥s∥hjg_j=-\log\|s\|_{h_j} at those atoms, and assume εj=sup⁡x∈Xan∣log⁡∥⋅∥hj,x∥⋅∥h,x∣⟶0,μj⟹μ,\begin{equation} \varepsilon_j=\sup_{x\in X^\mathrm{an}} \left|\log\frac{\|\cdot\|_{h_j,x}}{\|\cdot\|_{h,x}}\right| \longrightarrow0,\qquad \mu_j\Longrightarrow\mu, \end{equation} where μ\mu is a finite positive Radon measure and convergence is against all real continuous functions. The fibrewise ratio uses any nonzero vector and is independent of that vector.

Theorem. NAG-002844 (Uniform logarithmic tails suffice). In the preceding setup write Tj(R)=∫(gj−R)+ dμj=∑Vaj,V(gj(ξj,V)−R)+,aj,V=μj({ξj,V})≥0.T_j(R)=\int(g_j-R)_+\,d\mu_j =\sum_V a_{j,V}\bigl(g_j(\xi_{j,V})-R\bigr)_+, \qquad a_{j,V}=\mu_j(\{\xi_{j,V}\})\geq0. Suppose lim⁡R→∞sup⁡jTj(R)=0.\begin{equation} \lim_{R\to\infty}\sup_j T_j(R)=0. \end{equation} Then μ(Dan)=0\mu(D^\mathrm{an})=0, g∈L1(μ)g\in L^1(\mu) and ∫g dμ=lim⁡j→∞∫gj dμj.\begin{equation} \int g\,d\mu =\lim_{j\to\infty}\int g_j\,d\mu_j. \end{equation} If hjh_j and the metrics defining μj\mu_j have the common model data of the standing finite-model data, the hypothesis is the explicit coefficient condition lim⁡R→∞sup⁡j∑Vδj,Ve0,jEj(λcj,V−e0,jbj,VR)+=0.\begin{equation} \lim_{R\to\infty}\sup_j \sum_V\frac{\delta_{j,V}}{e_{0,j}E_j} \bigl(\lambda c_{j,V}-e_{0,j}b_{j,V}R\bigr)_+=0. \end{equation} In particular a common upper bound on cj,V/(e0,jbj,V)c_{j,V}/(e_{0,j}b_{j,V}) suffices. More generally, if gj≥Cg_j\geq C at all atoms and for some α>0\alpha>0 and A<∞A<\infty one has sup⁡j∑Vaj,V(gj(ξj,V)−C)1+α≤A,\begin{equation} \sup_j\sum_V a_{j,V} (g_j(\xi_{j,V})-C)^{1+\alpha}\leq A, \end{equation} then Tj(R)≤A/(R−C)αT_j(R)\leq A/(R-C)^\alpha for R>CR>C, so the conclusion holds.

Direct inputs Theorem G.0.3, NAG-001714.

Proof. Because ∥s∥h\|s\|_h is continuous and bounded on the compact space, gg is bounded below. Each function gR=min⁡(g,R)g_R=\min(g,R) is a finite continuous function, including at DanD^\mathrm{an}: it is the continuous capped function of the nonnegative quantity ∥s∥h\|s\|_h. Metric convergence implies ∣gj−g∣≤εj|g_j-g|\leq\varepsilon_j off DanD^\mathrm{an} and hence at every model atom by Theorem G.0.3. The sequence εj\varepsilon_j is bounded. Thus all these atom values have a common lower bound CC, and gg can also be taken at least CC. Let mj=μj(Xan)m_j=\mu_j(X^\mathrm{an}). Weak convergence tested on 11 makes mjm_j bounded. The map u↦min⁡(u,R)u\mapsto\min(u,R) is 11-Lipschitz, so ∣∫min⁡(gj,R) dμj−∫gR dμj∣≤εjmj.\begin{equation} \left|\int\min(g_j,R)\,d\mu_j-\int g_R\,d\mu_j\right| \leq\varepsilon_jm_j. \end{equation} For fixed RR, its left integrals therefore have common limit IR=∫gR dμI_R=\int g_R\,d\mu.

Put Ij=∫gj dμjI_j=\int g_j\,d\mu_j and T(R)=sup⁡jTj(R)T(R)=\sup_jT_j(R). The exact identity Ij=∫min⁡(gj,R) dμj+Tj(R)I_j=\int\min(g_j,R)\,d\mu_j+T_j(R) gives IR≤lim inf⁡jIj≤lim sup⁡jIj≤IR+T(R).\begin{equation} I_R\leq\liminf_j I_j\leq\limsup_j I_j\leq I_R+T(R). \end{equation} For some fixed sufficiently large R0R_0, T(R0)T(R_0) is finite. Since gR0g_{R_0} is bounded and the mjm_j are bounded, the IjI_j are bounded above; their lower bound follows from CmjC m_j. Hence IRI_R is bounded above as RR grows. By the definition of the positive integral, sup⁡R∫(gR−C) dμ=∫(g−C) dμ\sup_R\int(g_R-C)\,d\mu=\int(g-C)\,d\mu: every bounded nonnegative simple function below g−Cg-C lies below gR−Cg_R-C once R−CR-C exceeds its largest value. Thus gg has finite integral. The simple functions n 1Dan≤g−Cn\,1_{D^\mathrm{an}}\leq g-C then force μ(Dan)=0\mu(D^\mathrm{an})=0, and the lower bound gives g∈L1(μ)g\in L^1(\mu). Letting R→∞R\to\infty in (G.0.19) proves (G.0.15).

On a common model, substitute aj,V=bj,Vδj,V/Eja_{j,V}=b_{j,V}\delta_{j,V}/E_j and gj(ξj,V)=λcj,V/(e0,jbj,V)g_j(\xi_{j,V})=\lambda c_{j,V}/(e_{0,j}b_{j,V}). Since bj,V,e0,j>0b_{j,V},e_{0,j}>0 and δj,V≥0\delta_{j,V}\geq0, the expression for Tj(R)T_j(R) becomes exactly the sum in (G.0.16). A uniform upper bound on the indicated ratios makes it zero for large RR. Finally, for x≥Cx\geq C and R>CR>C, (x−R)+≤(x−C)1+α(R−C)α.(x-R)_+\leq\frac{(x-C)^{1+\alpha}}{(R-C)^\alpha}. Summing proves the final sufficient condition. ◻

Lemma. NAG-003007 (A finite-group orbit tensor with trivial inertia). Let SS be affine, let X/SX/S be separated, and let a finite group GG act on XX by automorphisms covering automorphisms of SS. Let I◃GI\triangleleft G, and let C↪XC\hookrightarrow X be a GG-stable closed subscheme on which every element of II acts identically. If A\mathcal A is relatively ample on X/SX/S, then N=⨂g∈Gρg∗A(11)\mathcal N=\bigotimes_{g\in G}\rho_g^*\mathcal A \tag{11} is relatively ample and has its canonical permutation GG-linearization. Its restricted II-action on N∣C\mathcal N|_C is the identity. If H=G/IH=G/I, and ρˉh\bar\rho_h denotes its induced action on CC, the underlying restricted invertible sheaf is N∣C≃⨂h∈H(ρˉh∗(A∣C))⊗∣I∣.(12)\mathcal N|_C\simeq \bigotimes_{h\in H}(\bar\rho_h^*(\mathcal A|_C))^{\otimes |I|}. \tag{12} The construction commutes with equivariant pullbacks and specified transition isomorphisms of the input sheaves. It therefore applies to compatible towers. No group order is inverted.

Proof. For h∈Gh\in G, the factors of ρh∗N\rho_h^*\mathcal N are ρgh∗A\rho_{gh}^*\mathcal A, indexed by g∈Gg\in G. Reindexing by g↦ghg\mapsto gh gives λh\lambda_h. Associativity of pullback and the usual symmetry of tensor products show that the two maps for any composite agree: both reindex the same factors in the same way. The identity gives the identity map. This proves the linearization, including all noncommutative group relations.

On CC, the action factors through HH. The factors indexed by a right coset gIgI are canonically the same invertible sheaf ρˉgI∗(A∣C)\bar\rho_{gI}^*(\mathcal A|_C). Grouping these factors gives (12). For i∈Ii\in I, right multiplication by ii permutes factors within each such coset. Locally trivialize each of the finitely many coset sheaves. A permutation of copies of a rank-one module fixes the tensor of their chosen generator, and fixes every coefficient because ρi∣C\rho_i|_C is the identity. Hence it is the identity of the full restricted module. This includes nilpotents; no conclusion is being drawn merely from pointwise characters.

Ampleness is unchanged by an isomorphism of schemes. Over the affine base, relative and absolute ampleness agree by Stacks Tag 01VK. Finite tensor products of these ample sheaves are ample: use the external-product assertion Stacks Tag 0DNK over the affine, hence quasi-separated, base, and pull back by the closed diagonal using Stacks Tag 0B3E; then induct on the number of factors. In the section argument for the external-product assertion, sections of powers nn and mm are first raised to powers mm and nn, so their product is a section of the common power nmnm. This is the same ampleness argument used in the audited cyclic orbit construction, with a finite group as index set.

Finally, an equivariant pullback identifies each indexed factor with the corresponding pulled-back factor. Tensoring the given maps on these factors commutes with every permutation above. This proves the asserted transition compatibility and the tower statement. ◻

Let AA be a nonzero Noetherian ring. Set S=A∖⋃p∈Min⁡(A)p,T(A)=ker⁡ ⁣(A⟶∏p∈Min⁡(A)Ap),Au=A/T(A).(1.1)S=A\setminus\bigcup_{\mathfrak p\in\operatorname{Min}(A)}\mathfrak p,\qquad T(A)=\ker\!\left(A\longrightarrow \prod_{\mathfrak p\in\operatorname{Min}(A)}A_{\mathfrak p}\right), \qquad A^{\mathrm u}=A/T(A). \tag{1.1} We call T(A)T(A) the embedded torsion ideal. No reduction of the generic Artinian local rings is performed.

Supplied hypothesis (SC). For a finite module over a generalized affinoid algebra, the S1S_1 property of the algebraic localizations is equivalent to the corresponding analytic property. The property is preserved on analytic domains.

Proposition. NAG-002671 (Coherent embedded torsion quotient; F8U-G). Assume (SC) and the stated v0.38 coherent-domain interfaces. On a good Hausdorff pure nn-dimensional analytic XX, the ideals T(A)T(A) on affinoid domains glue to a coherent ideal T\mathcal T. The closed subspace i:Xu=(X,OX/T)⟶X(3.1)i:X^{\mathrm u}=(X,\mathcal O_X/\mathcal T)\longrightarrow X \tag{3.1} is finite, is a homeomorphism on underlying points, is pure of analytic dimension nn, and satisfies S1S_1. Every regular analytic local function on XX remains regular on XuX^{\mathrm u}. Pullback sends Int⁡(X/K)\operatorname{Int}(X/K) into Int⁡(Xu/K)\operatorname{Int}(X^{\mathrm u}/K).

Proof. Let A→BA\to B represent an actual nonempty affinoid domain of M(A)\mathcal M(A). Restriction is exact on coherent sheaves, and NAG-001624 identifies it with B⊗A−B\otimes_A- on finite modules. In particular TB=T(A)⊗ABTB=T(A)\otimes_A B injects into BB. The lower-dimensional support bound of NAG-002669 applies to the original torsion ideal. Its support is the restriction of Supp⁡T(A)\operatorname{Supp}T(A), so has analytic dimension at most n−1n-1, by domain base change NAG-001846 and dimension locality. The domain is pure of dimension nn by NAG-002413. Therefore TB⊆T(B)TB\subseteq T(B): a module with smaller-dimensional support has zero localization at every minimal prime of BB.

The quotient B/TBB/TB is the actual domain algebra obtained by base change from AuA^{\mathrm u}, using NAG-001565 and NAG-001846. By Lemma C.1.36, AuA^{\mathrm u} is S1S_1; (SC) makes B/TBB/TB S1S_1 as well. Its underlying space is unchanged by the nilpotent ideal TBTB and remains pure of dimension nn. The image of T(B)T(B) in B/TBB/TB has support of dimension at most n−1n-1. A nonzero submodule of a Noetherian S1S_1 pure structure module has an associated prime among its minimal primes, so cannot have smaller-dimensional support. The image is zero, proving T(B)=T(A)⊗AB.(3.2)T(B)=T(A)\otimes_A B. \tag{3.2} Thus the coherent ideals coincide on all actual affinoid intersections and satisfy the cocycle condition. The coherent sheaf equivalence NAG-001624 glues them, without assuming a finite global cover.

Each local quotient is by a finite nilpotent ideal, as in the actual closed quotient of NAG-001653, so (3.1) is a finite closed immersion and a homeomorphism. Its analytic dimension and purity agree with those of XX; (SC) gives the asserted S1S_1 property. Regularity follows locally from Lemma C.1.36. Finite morphisms are relatively interior by NAG-001701; composition of interiors, NAG-001704, proves the last claim. In particular a boundaryless XX gives boundaryless XuX^{\mathrm u}. ◻

Finite logarithmic graph: weights and persistence.

Use the finite logarithmic graph and its notation from the setup preceding the actual source-image lemma. In particular the field, the tuples h,(g0,f0)h,(g_0,f_0), and the domains V0,V0,SV_0,V_{0,S} are those specified there.

Lemma. NAG-002572 (The whole-fibre weight is one; F8GP-4). For every SS, the hh-moment chain of V0,SV_{0,S} is Trop⁡h,V0,S(OV0,S)=P∣{t∘F≤S}.(3.2)\operatorname{Trop}_{h,V_{0,S}}(\mathcal O_{V_{0,S}}) =\mathcal P|_{\{t\circ F\leq S\}}. \tag{3.2} Each full-dimensional product cell that survives truncation has weight one. Lower-dimensional pieces contribute no nn-chain.

Proof. Apply the frozen compact-chain theorem to the pure space V0,SV_{0,S} and tuple hh. Its construction subdivides the actual image and its boundary image, the latter of dimension at most n−1n-1. Choose ww in the relative interior of an nn-cell of PSP_S, away from that boundary image and from the product’s proper faces. Thus the entire hh-fibre over ww is interior over KK, and the local image is in one product cell’s affine nn-plane. These are precisely the two flat-germ hypotheses required by NAG-002362. This use of the frozen theorem is only its existing compact-domain scope. For a factor of type DiD_i or EiE_i, set qi=Tiq_i=T_i; for a factor of type HiH_i, set qi=Ti−1q_i=T_i-1. Adjoin the ZνZ_\nu as the remaining coordinates. The tangent lattice of this product cell is carried isomorphically onto Zn\mathbb Z^n by these selected moment coordinates: its basis vectors are respectively (1,1)(1,1), (0,1)(0,1), (1,0)(1,0), and the torus coordinate vectors in disjoint coordinate blocks.

The algebra of the entire hh-fibre over ww is the exact-radius Laurent algebra Rq=AK,r(w)R_q=A_{K,r(w)}. To check the assertion as an algebra statement, not just a set equality, express the omitted coordinates and their inverses in RqR_q. On a DiD_i-factor, ∣qi∣>1|q_i|>1, and (Ti−1)−1=qi−1∑ℓ≥0qi−ℓ.(T_i-1)^{-1}=q_i^{-1}\sum_{\ell\geq0}q_i^{-\ell}. On an EiE_i-factor, ∣qi∣<1|q_i|<1, and (Ti−1)−1=−∑ℓ≥0qiℓ.(T_i-1)^{-1}=-\sum_{\ell\geq0}q_i^\ell. On an HiH_i-factor, ∣qi∣<1|q_i|<1, and Ti−1=(1+qi)−1=∑ℓ≥0(−qi)ℓ.T_i^{-1}=(1+q_i)^{-1}=\sum_{\ell\geq0}(-q_i)^\ell. These series converge in the exact-radius norm. The simultaneous level relations give bounded homomorphisms in both directions; they fix all coordinate generators and hence are inverse. All inequalities defining V0,SV_{0,S} have fixed values on this whole fibre, so no factor or sheet is removed by them.

The choice of ww does not require t∘F<St\circ F<S. If t∘Ft\circ F is constant on an nn-cell and equals SS, that whole cell is part of the actual image PSP_S. The frozen boundary-image dimension bound still permits the same generic interior-fibre choice on it. No equality-level cell is removed.

For the selected transverse coordinates, the lattice divisor is one and the actual Gauss algebra is Rq⊗^RqHq=Hq.R_q\widehat\otimes_{R_q}H_q=H_q. Its dimension over HqH_q is one. The frozen module-weight theorem therefore gives weight one at ww. This determines the coefficient of every surviving nn-cell. The finite-chain theorem, subdivision independence, and foundations C.1.39 give (3.2). ◻

Theorem. NAG-002568 (Persistent logarithmic graph; F8GP-1). For every real SS, put V0,S={z∈V0:∣f0(z)∣≥e−S},PS=P∩{λ+∑imisi+b⋅x≤S}.(1.7)V_{0,S}=\{z\in V_0:|f_0(z)|\geq e^{-S}\},\qquad P_S=P\cap\left\{\lambda+\sum_i m_i s_i+b\cdot x\leq S\right\}. \tag{1.7} Every nonempty V0,SV_{0,S} is pure of analytic dimension nn. The following chain is defined once, independently of SS: C0=F∗P.(1.8)C_0=F_*\mathcal P. \tag{1.8} The pushforward counts each full-rank cell with its lattice index and discards cells on which FF has rank below nn. It is a finite nonnegative integral nn-chain, and Trop⁡(g0,f0),V0,S(OV0,S)=C0∣{t≤S}(1.9)\boxed{\quad \operatorname{Trop}_{(g_0,f_0),V_{0,S}}(\mathcal O_{V_{0,S}}) =C_0|_{\{t\leq S\}} \quad} \tag{1.9} for every SS, with zero chain for the empty domain. Its support lies in (κ+M ⁣(∏i[−Ai,Bi]×Q))×[tmin⁡,∞),tmin⁡=λ−∑imiAi+min⁡Qb⋅x.(1.10)\left(\kappa+M\!\left(\prod_i[-A_i,B_i]\times Q\right)\right) \times[t_{\min},\infty), \quad t_{\min}=\lambda-\sum_i m_iA_i+\min_Q b\cdot x. \tag{1.10} Thus the same finite weighted chain works at all levels, including levels where a full-dimensional cell lies in {t=S}\{t=S\}.

Proof of foundations G.0.11. Purity is foundations C.1.40. On V0,SV_{0,S}, all entries of hh and (g0,f0)(g_0,f_0) are units. Their actual moment maps satisfy t(g0,f0)=F∘tht_{(g_0,f_0)}=F\circ t_h. If V0,SV_{0,S} is empty, the actual source image makes P∣{t∘F≤S}\mathcal P|_{\{t\circ F\leq S\}} zero. The closed-cutoff pushforward then gives the required zero chain. Otherwise adjoin the actual constant units a,c1,…,cra,c_1,\ldots,c_r to hh. The whole level algebras and the transverse Gauss dimensions are unchanged, and the saturated tangent lattices embed with zero constant coordinates and index one. The translated formula is now a literal character map for the identity finite map of V0,SV_{0,S}. Use whole-interior projection off the target absolute boundary image. That lower-dimensional image does not contain any top-dimensional cell of a full-rank image, even at t=St=S. On a common subdivision the finite-module chain and weights identify the closed chains from their generic coefficients. This gives Trop⁡(g0,f0),V0,S(O)=F∗Trop⁡h,V0,S(O).\operatorname{Trop}_{(g_0,f_0),V_{0,S}}(\mathcal O) =F_*\operatorname{Trop}_{h,V_{0,S}}(\mathcal O). Substitute (3.2), then (3.3), to obtain (1.9). The bounds in (1.10) follow directly from yi∈[−Ai,Bi]y_i\in[-A_i,B_i], x∈Qx\in Q, and si≥−Ais_i\geq-A_i. The finite chain is constructed before any SS is chosen. ◻

Write A0=O(V0)A_0=\mathcal O(V_0) for the algebra of the specified graph domain.

Theorem. NAG-002575 (Persistence under covers and perturbations; F8GP-7). Let p:V=M(B)→V0=M(A0)p:V=\mathcal M(B)\to V_0=\mathcal M(A_0) be finite with its canonical module norm, and suppose BB is a locally free A0A_0-module of constant positive rank ee. Assume VV is pure of dimension nn. Let u0,u1,…,ur∈B×u_0,u_1,\ldots,u_r\in B^\times satisfy ∣uj(z)∣=1for every z∈V.(5.1)|u_j(z)|=1 \quad\text{for every }z\in V. \tag{5.1} For f=p∗f0 u0,gℓ=p∗g0,ℓ uℓ,VS={∣f∣≥e−S},f=p^*f_0\,u_0,\qquad g_\ell=p^*g_{0,\ell}\,u_\ell,\qquad V_S=\{|f|\geq e^{-S}\}, the function ff is regular, every nonempty VSV_S is pure of dimension nn, and one single chain C=eC0(5.2)C=eC_0 \tag{5.2} satisfies Trop⁡(g,f),VS(OVS)=C∣{t≤S}(5.3)\boxed{\quad \operatorname{Trop}_{(g,f),V_S}(\mathcal O_{V_S}) =C|_{\{t\leq S\}} \quad} \tag{5.3} for every real SS. The same bound (4.2) applies. Neither separability, reducedness of VV, nor global freeness of BB is required.

A concrete sufficient condition for (5.1) is uj=1+εju_j=1+\varepsilon_j with ∥εj∥<1\left\lVert\varepsilon_j\right\rVert<1 in a specified complete submultiplicative affinoid norm on BB.

Proof. The norm identities give, on all points, ∣f∣=∣p∗f0∣,∣gℓ∣=∣p∗g0,ℓ∣,VS=p−1(V0,S).(5.4)|f|=|p^*f_0|,\qquad |g_\ell|=|p^*g_{0,\ell}|,\qquad V_S=p^{-1}(V_{0,S}). \tag{5.4} Thus no source sheets are selected or omitted. The map on VSV_S is the actual finite base change; its direct-image module is still locally free of rank ee. Purity of VSV_S follows from domain purity applied to VV. Flatness of BB over A0A_0 preserves injectivity of multiplication by f0f_0. Multiplication by u0u_0 is invertible, proving regularity of ff.

If V0,SV_{0,S} is empty, (5.4) makes VSV_S empty and (1.9) gives (5.3). Otherwise apply whole-interior projection to p∣VSp|_{V_S} over the original field, using the literal pullbacks of the target tuple and the identity map on the moment coordinates. By (5.4) the norm-one twists have the same moment map and the same whole analytic level domains as those literal pullbacks. The relative interiors of target top cells avoid the target absolute boundary image after a common subdivision. The finite-module chain and weights give equality of the closed finite chains from the generic equalities. This gives Trop⁡(g,f),VS(OVS)=Trop⁡(g0,f0),V0,S(p∗OVS).(5.5)\operatorname{Trop}_{(g,f),V_S}(\mathcal O_{V_S}) =\operatorname{Trop}_{(g_0,f_0),V_{0,S}}(p_*\mathcal O_{V_S}). \tag{5.5} At a generic target cell, let EQE_Q be its complete transverse Gauss algebra. The tensor of the locally free module p∗OVSp_*\mathcal O_{V_S} with EQE_Q has constant rank ee. The finite-dimensional Artinian algebra EQE_Q is a product of local factors; on each factor a finite locally free module of constant rank ee is free of that rank. Consequently dim⁡HQ(p∗OVS⊗EQ)=e dim⁡HQEQ.\dim_{H_Q}\bigl(p_*\mathcal O_{V_S}\otimes E_Q\bigr) =e\,\dim_{H_Q}E_Q. The same lattice divisor occurs in both module-weight formulas. Therefore the right side of (5.5) is e Trop⁡(g0,f0),V0,S(O)e\,\operatorname{Trop}_{(g_0,f_0),V_{0,S}}(\mathcal O). Equation (1.9) proves (5.3). This dimension calculation retains all Artinian lengths and full residue degrees, including inseparable degrees.

A positive-rank locally free algebra has zero structural kernel, and the finite-image theorem makes pp surjective. Thus the bounded unit-moment base in (1.10) is still the actual moment image of VV. The scalar ee changes weights, not that base or the height bound in (4.2).

Finally, if ∥ε∥<1\left\lVert\varepsilon\right\rVert<1, the Neumann series ∑ℓ≥0(−ε)ℓ\sum_{\ell\geq0}(-\varepsilon)^\ell converges to the inverse of 1+ε1+\varepsilon. Every bounded multiplicative seminorm satisfies ∣ε(z)∣≤∥ε∥<1|\varepsilon(z)|\leq\left\lVert\varepsilon\right\rVert<1; this follows directly by applying boundedness to powers and taking roots. The strong triangle inequality then gives ∣1+ε(z)∣=1|1+\varepsilon(z)|=1, proving the sufficient condition. ◻

Fix the original complete non-Archimedean field KK and its absolute value; trivial valuation and generalized real radii are allowed. Let XX be good, Hausdorff, and pure of analytic dimension n≥1n\geq1. Fix C⋐Int⁡(X/K)C\Subset\operatorname{Int}(X/K), an ordinary open neighbourhood U⊃CU\supset C, and f∈O(U)f\in\mathcal O(U). Require:

  1. Every ordinary local ring of UU satisfies S1S_1.

  2. Multiplication by ff is injective on OUG\mathcal O_{U_G}. Thus “regular” means an actual non-zero-divisor, not merely a nonzero function on one component.

  3. The actual zero space D=V(f)⊂UD=V(f)\subset U, with its quotient structure, is empty or pure of analytic dimension n−1n-1.

The test η\eta is an ordinary smooth moment form of bidegree (n−1,n−1)(n-1,n-1), with Supp⁡η⊂C\operatorname{Supp}\eta\subset C. We use the structure sheaf coefficients in the scalar PL assertion. An arbitrary coherent coefficient version is not silently substituted. All dimensions here are analytic dimensions. In particular S3 is not a statement about ordinary Krull dimension alone.

Chart condition (IP). For a pure nn-dimensional affinoid VV, a regular analytic ff, nonempty pure (n−1)(n-1)-dimensional DVD_V, and a unit moment tuple gg, assume every actual affinoid-domain algebra in VV is S1S_1. Then for some HH, all H<a<bH<a<b satisfy T(g,f)(O{e−b≤∣f∣≤e−a})=Tg(ODV)×[a,b]\mathcal T_{(g,f)}\bigl(\mathcal O_{\{e^{-b}\leq|f|\leq e^{-a}\}}\bigr) =\mathcal T_g(\mathcal O_{D_V})\times[a,b] in the last coordinate −log⁡∣f∣-\log|f|, with full actual quotient weights and the standard interval lattice.

Lemma. NAG-002954 (Actual chart hypotheses; F8SA-C). Under S1–S3, choose finitely many affinoid domains Vi⊂UV_i\subset U whose ordinary interiors cover CC, and unit moment representatives η∣Vi=tgi∗αi\eta|_{V_i}=t_{g_i}^*\alpha_i. For every nonempty intersection VI=⋂i∈IViV_I=\bigcap_{i\in I}V_i, the hypotheses of IP hold whenever DI=D∩VID_I=D\cap V_I is nonempty. If DI=∅D_I=\varnothing, a sufficiently small ff-tube in VIV_I is empty.

Proof. Goodness and the ordinary local definition of a smooth moment form [NAG-002424] supply such neighbourhoods; compactness gives a finite family. NAG-002411/5.50.3 make each nonempty VIV_I an actual pure nn-dimensional affinoid domain. Restricting the injective sheaf map in S2 and taking sections proves that ff is a non-zero-divisor in its algebra. Quotient/domain base change NAG-001846 identifies DID_I with the actual algebra quotient by ff, not its reduction. It is an affinoid domain of DD, hence is pure of dimension n−1n-1 by S3 and domain purity. R-S1 [NAG-002765] applied to every affinoid domain of VIV_I supplies IP’s full affinoid-domain S1S_1 condition. A chosen gig_i, i∈Ii\in I, remains a tuple of units upon restriction. If DID_I is empty, continuity on compact VIV_I gives min⁡VI∣f∣>0\min_{V_I}|f|>0. ◻

The ordinary cutoff presentations and the final topologies are those of the two generating families and their comparison. For an open Ω⊂RN\Omega\subset\mathbb R^N, let Ep,q(Ω)E^{p,q}(\Omega) be the space of smooth ambient superforms with the coefficient seminorms pL,m(η)=max⁡I,J, ∣a∣≤msup⁡u∈L∣∂aηI,J(u)∣,L⋐Ω.p_{L,m}(\eta)=\max_{I,J,\,|a|\leq m}\sup_{u\in L} |\partial^a\eta_{I,J}(u)|, \qquad L\Subset\Omega. Empty coefficient slots give the zero space.

Definition G.0.14. (Local retraction condition). For a closed subset Z⊂XZ\subset X, condition (Rd)(\mathrm R_d) means that for every z∈Zz\in Z there are an ordinary neighbourhood Uz⊂XU_z\subset X, a good Hausdorff space YzY_z of local analytic dimension at most dd, a closed immersion iz:Yz↪Uzi_z:Y_z\hookrightarrow U_z whose image contains zz, and an analytic retraction rz:Uz→Yzr_z:U_z\to Y_z. The chosen image need only contain zz; it need not contain all of Z∩UzZ\cap U_z. The dimension bound means that the smaller affinoid neighbourhoods used in Lemma F.1.6 have dimension at most dd.

Theorem. NAG-002480 (A logarithmic current in the retraction scope). Let XX be good, Hausdorff and pure of analytic dimension n≥1n\geq1, and let F\mathcal F be a coherent OXG\mathcal O_{X_G}-module. Let g,h∈OX(X)g,h\in\mathcal O_X(X) be non-zero-divisors in every local ring and put Z={∣g∣=0}∪{∣h∣=0},u=log⁡∣g/h∣on X∖Z.Z=\{|g|=0\}\cup\{|h|=0\},\qquad u=\log|g/h|\quad\text{on }X\setminus Z. Assume (Rn−1)(\mathrm R_{n-1}) for ZZ. Then Tg/h,F(α)=∫X∖Z, F∣X∖Zuα\begin{equation} T_{g/h,\mathcal F}(\alpha)= \int_{X\setminus Z,\,\mathcal F|_{X\setminus Z}}u\alpha \end{equation} is well-defined and continuous on (En,n(X),τcut)=(En,n(X),τord)(\mathcal E^{n,n}(X),\tau_{\mathrm{cut}})= (\mathcal E^{n,n}(X),\tau_{\mathrm{ord}}). The displayed functional is independent of the representatives of the ordinary test form. It is a current of bidimension (n,n)(n,n) for this fixed regular quotient.

For each cutoff presentation QiQ_i, there is Ci<∞C_i<\infty with ∣Tg/h,F(Qiη)∣≤Cimax⁡I,Jsup⁡tfi(Vi)∣ηI,J∣.\begin{equation} |T_{g/h,\mathcal F}(Q_i\eta)| \leq C_i\max_{I,J}\sup_{t_{f_i}(V_i)}|\eta_{I,J}|. \end{equation} There are smooth functions uM∈A0,0(X)u_M\in\mathcal A^{0,0}(X), M≥0M\geq0, formed by constant smooth truncation of log⁡∣g∣\log|g| and log⁡∣h∣\log|h| near their zeros, such that for each ii one number MiM_i satisfies ∫X,FuMQiη=Tg/h,F(Qiη)(M≥Mi, η∈En,n(Ωi)).\begin{equation} \int_{X,\mathcal F}u_M Q_i\eta=T_{g/h,\mathcal F}(Q_i\eta) \quad(M\geq M_i,\ \eta\in E^{n,n}(\Omega_i)). \end{equation} The threshold is uniform over the whole fixed presentation, with no bound on η\eta required.

Proof. Fix a cutoff presentation QiQ_i and put K=supp⁡χiK=\operatorname{supp}\chi_i. For each z∈K∩Zz\in K\cap Z, Lemma F.1.6 applied to its fixed tuple makes every pulled-back (n,n)(n,n)-form zero on one ordinary neighbourhood of zz. A finite union OO of these neighbourhoods covers K∩ZK\cap Z. Therefore Ki=K∖OK_i=K\setminus O is compact in X∖ZX\setminus Z and contains supp⁡Qiη\operatorname{supp}Q_i\eta for every η\eta. The cutoff presentations span the test space, so every test has compact support in the complement. There gg and hh are locally analytic units, using 1.20.1 on smaller affinoids as in 5.53.1. Thus uu is an ordinary smooth moment function. The product uαu\alpha has compact support in the complement. The integral exists and is independent of its chart representatives by 5.52.4. Open restrictions are pure of dimension nn by 5.50.3. These facts also prove linearity of (G.0.20).

For the fixed ii and KiK_i, by C.1.60, choose θ∈Ac0,0(X)\theta\in\mathcal A_c^{0,0}(X) supported in X∖ZX\setminus Z and equal to one near KiK_i. The function b=ext⁡0(θu)b=\operatorname{ext}_0(\theta u) is ordinary smooth on XX, and locality of the baseline integral gives Tg/h,F(Qiη)=∫X,Fb Qiη.T_{g/h,\mathcal F}(Q_i\eta)=\int_{X,\mathcal F}b\,Q_i\eta. Compute this integral on a finite affinoid cover of the fixed support of χi\chi_i inside Int⁡top(Vi,X)\operatorname{Int}_{\mathrm{top}}(V_i,X), adjoining fif_i to representatives of bχib\chi_i. The finite inclusion–exclusion formula 5.50.1 and NAG-002405, finite module-chain and weights clause give fixed cells, actual coherent module weights and fixed coefficient multipliers. On each cell, the absolute integral is at most its lattice-normalized volume times the absolute weight times a fixed constant times the right-hand coefficient supremum in (G.0.21), by 5.46.2. The projected coefficient arguments lie in tfi(Vi)t_{f_i}(V_i). Summing these finitely many bounds proves (G.0.21) and continuity by the definition of the final topology. No positivity or integration by parts enters this estimate.

Choose a smooth real function ρ\rho with ρ=0\rho=0 on (−∞,0](-\infty,0] and ρ=1\rho=1 on [1,∞)[1,\infty), as explicitly constructed in 5.53.1. Put λ(s)=sρ(s+1),λM(t)=−M+λ(t+M).\lambda(s)=s\rho(s+1),\qquad \lambda_M(t)=-M+\lambda(t+M). Thus λM(t)=t\lambda_M(t)=t for t≥−Mt\geq-M and is −M-M for t≤−M−1t\leq-M-1. For an analytic function aa set ℓM(a)(x)={λM(log⁡∣a(x)∣),∣a(x)∣>0,−M,∣a(x)∣=0.\ell_M(a)(x)= \begin{cases} \lambda_M(\log|a(x)|),& |a(x)|>0,\\ -M,& |a(x)|=0. \end{cases} This is an ordinary smooth function: near a zero it is constant on {∣a∣<e−M−1}\{|a|<e^{-M-1}\}, and on the nonzero locus it is a smooth function of a unit moment coordinate. The scalar sheaf in 5.53.1 glues these representatives. Set uM=ℓM(g)−ℓM(h)u_M=\ell_M(g)-\ell_M(h).

If KiK_i is empty, every QiηQ_i\eta is zero and stationarity is immediate. Otherwise, the positive continuous functions ∣g∣|g| and ∣h∣|h| have positive minima on KiK_i. Choose Mi≥0M_i\geq0 strictly larger than both negatives of the minimum logarithms there. For M≥MiM\geq M_i, uM=uu_M=u on a neighbourhood of KiK_i. Since every QiηQ_i\eta is supported in KiK_i, the two products agree as ordinary forms after extension by zero. This proves (G.0.22). It also gives absolute finiteness on every finite cell presentation: all coefficients and the logarithms are bounded on the relevant compact supports. No separate representation by Radon measures is required. ◻

Proposition. NAG-003141 (A unit logarithm need not annihilate a positive current). For every n≥2n\geq2, let z1,…,znz_1,\ldots,z_n be the coordinates on the analytic split torus over KK, write xj=−log⁡∣zj∣x_j=-\log|z_j| and t=(x1,…,xn)t=(x_1,\ldots,x_n), and set Ω=(1,2)×(0,1)n−1,U=t−1(Ω),h=x1=log⁡∣z1−1∣.\Omega=(1,2)\times(0,1)^{n-1},\qquad U=t^{-1}(\Omega),\qquad h=x_1=\log|z_1^{-1}|. Then T(η)=∫U,OUx1η,η∈En,n(U),T(\eta)=\int_{U,\mathcal O_U}x_1\eta, \qquad \eta\in\mathcal E^{n,n}(U), is a positive native current with an order-zero bound on every fixed coefficient stage. Nevertheless d′d′′h=0d'd''h=0 and d′d′′(hT)≠0d'd''(hT)\ne0. Thus positivity and these order-zero bounds do not suffice for harmonic frame invariance.

Proof. Closed polyannuli with boxes inside Ω\Omega are ordinary affinoid neighborhoods in UU. They are integral and have analytic dimension nn by NAG-002294; The affine analytification construction NAG-001707 and separated-space theorem NAG-001713 make the ambient torus good and Hausdorff. Thus UU is good, Hausdorff and pure of dimension nn. For the global units g=z1−1g=z_1^{-1} and h0=1h_0=1, the zero set in NAG-002480 is empty. Its retraction condition is therefore vacuous. That frozen theorem constructs exactly TT and its fixed-stage order-zero estimates, on all native top tests. Since x1>0x_1>0, positivity follows from NAG-002425. The unit calculation in Theorem C.1.64 gives d′d′′h=0d'd''h=0 as a smooth form; its current conclusion is unavailable without closedness of TT.

Choose a nonnegative nonzero ψ∈Cc∞(Ω)\psi\in C_c^\infty(\Omega) and use the native test α=ψ(x)⋀j=2n(d′xj∧d′′xj)∈En−1,n−1(U).\alpha=\psi(x)\bigwedge_{j=2}^n(d'x_j\wedge d''x_j) \in\mathcal E^{n-1,n-1}(U). Its support is compact: the inverse image of supp⁡ψ\operatorname{supp}\psi is closed in a closed polyannulus whose box is compactly contained in Ω\Omega. For the coordinate moment map the actual whole-fibre Gauss module is the Gauss field itself, and the tangent lattice map is the identity. Thus its weight is one by NAG-002362, consistently with the degree-one identity case of NAG-002294. The global chain formula NAG-002405 and the lattice convention identify coordinate top integrals with ordinary real integrals of their coefficients, without a factorial or a valuation-normalization factor.

Only the x1x_1 derivative survives in d′d′′αd'd''\alpha. Therefore Theorem C.1.63 gives (d′d′′(hT))(α)=∫Ωx12 ∂12ψ dx1⋯dxn=2∫Ωψ dx1⋯dxn>0.(d'd''(hT))(\alpha) =\int_\Omega x_1^2\,\partial_1^2\psi\,dx_1\cdots dx_n =2\int_\Omega\psi\,dx_1\cdots dx_n>0. The second equality is the one-variable fundamental theorem applied twice, with vanishing boundary terms because ψ\psi is compactly supported; iterated integration is the elementary calculation proved in NAG-002389. No analytic Stokes or Poincaré–Lelong theorem enters this computation. In particular TT cannot be closed, by Theorem C.1.64. ◻

In the local Cartier setting of NAG-002772, use the full quotient B=A/(f)B=A/(f) by the given regular equation on the actual affinoid chart. The effective Cartier coefficients are those of that quotient. The purity of the zero space is supplied as stated below.

Theorem. NAG-002778 (Cartier integration comparison in the pure-zero-space scope; F8ST-I-R1). Let D=M(B)D=\mathcal M(B) be a nonempty pure dd-dimensional Cartier zero space as above, and let g=(g1,…,gr)g=(g_1,\ldots,g_r) be analytic units on a neighbourhood of the chart. Put tg=−log⁡∣g∣t_g=-\log|g|. For p∈Min⁡B\mathfrak p\in\operatorname{Min}B, let Dp=M(B/p)D_{\mathfrak p}=\mathcal M(B/\mathfrak p) and cp=length⁡BpBpc_{\mathfrak p}=\operatorname{length}_{B_{\mathfrak p}}B_{\mathfrak p}. Then the frozen integral chains and all their smooth top-form pairings satisfy Tg(B)=∑p∈Min⁡BcpTg∣Dp(B/p),IB,g(α)=∑p∈Min⁡BcpIB/p,g(α).\begin{align} \mathcal T_g(B)&=\sum_{\mathfrak p\in\operatorname{Min}B} c_{\mathfrak p}\mathcal T_{g|D_{\mathfrak p}}(B/\mathfrak p),\\ \mathcal I_{B,g}(\alpha)&=\sum_{\mathfrak p\in\operatorname{Min}B} c_{\mathfrak p}\mathcal I_{B/\mathfrak p,g}(\alpha). \end{align} Here α\alpha is a smooth (d,d)(d,d)-superform near the compact moment image. The same equality holds for ordinary compactly supported forms after chart localization and a finite partition. The coefficient cpc_{\mathfrak p} is exactly the analytic Cartier coefficient of Theorem F.1.12.

In particular, on a good Hausdorff pure nn-dimensional space, for a regular meromorphic ff whose local numerator and denominator zero spaces are empty or pure of dimension n−1n-1, the cycles formed from these presentations are independent among such presentations and glue. Their analytic Cartier-cycle pairing agrees with the frozen coherent-cycle pairing on every ordinary (n−1,n−1)(n-1,n-1)-test with compact support in Int⁡(X/K)\operatorname{Int}(X/K). This purity condition is retained explicitly to match the frozen integrator; it is not deduced here from a relative-dimension theorem. The local coefficient result in Theorem F.1.12 does not require it.

Proof. Take a common finite subdivision of the actual moment images and their boundary images for DD and the DpD_{\mathfrak p}, using the finite-module chain and weights. At a relative interior point ww of a top-dimensional cell, the full analytic moment fibre is interior to DD. The local moment image is contained in the affine plane w+NRw+N_{\mathbb R} for the saturated rank-dd tangent lattice NN. The boundary images have dimension at most d−1d-1, so they miss these relative interiors. For an integral projection QQ injective on the tangent lattice NN, put EQ=Dw⊗^RQHQ,eQ=[Zd:Q(N)].E_Q=D_w\widehat\otimes_{R_Q}H_Q,\qquad e_Q=[\mathbb Z^d:Q(N)]. Here DwD_w is the actual algebra of the entire moment fibre and HQH_Q is the projected polytorus Gauss field, precisely as in NAG-002362. That theorem, with structure module BB, gives mD,g(w)=dim⁡HQEQeQ=∑p∈Min⁡Bcp mDp,g(w).\begin{equation} m_{D,g}(w)=\frac{\dim_{H_Q}E_Q}{e_Q} =\sum_{\mathfrak p\in\operatorname{Min}B}c_{\mathfrak p}\, m_{D_{\mathfrak p},g}(w). \end{equation} For an empty component fibre its weight is zero. This formula includes all local Artin lengths and residue degrees in EQE_Q; no separate residue-degree multiplier is attached to cpc_{\mathfrak p}. The lattice index is divided out exactly once.

The equality holds on every top cell. Lower-dimensional cells carry no dd-chain term, so the finite-module chain and weights give (G.0.23). By the finite-cell functional, integration of the same tangential smooth coefficients on a common subdivision gives (G.0.24). This is a finite sum; no limiting or measurable-integration extension occurs. The ordinary compact integral and compact partitions give the ordinary-form assertion. Actual domain restriction of the coefficients is the effective domain transport in the Cartier theorem, so the local formulas agree on overlaps. For a regular meromorphic function in the stated pure-zero-space scope, subtract the two effective formulas. On each actual common affinoid refinement, quotient/domain base change and domain purity preserve the empty-or-pure dimension n−1n-1 condition for all four equations. The antichain clause of the Cartier theorem therefore applies to both cross-products and proves representation independence. A compact test needs finitely many such representatives and partition terms, whose sum is consequently independent of these choices. Closed immersions have compact inverse support on compact sets and are finite and everywhere relatively interior (NAG-001701); interior composition (NAG-001704) therefore shows that a test supported in Int⁡(X/K)\operatorname{Int}(X/K) has the required interior support on its zero spaces. ◻

Field steps in tame descent

PDF-derived transcription. Historical English reader, Section 1.55, printed pages 216–220 and 223. The review record identifies the printed statements, proofs, and cited inputs.

Throughout, a valued field is complete and nontrivially real-valued, with additive valuation vv, actual valuation ring F∘F^\circ, maximal ideal mF\mathfrak m_F, value group ΓF\Gamma_F, and residue field F~\widetilde F. Finite extensions carry their unique extending absolute values. Their existence, uniqueness and completeness are NAG-001450. Simple residue roots lift by NAG-002049.

For a finite separable extension E/KE/K, the explicit tame hypotheses in this patch mean all three conditions e=[ΓE:ΓK],f=[E~:K~],[E:K]=ef,e∈(K∘)×,E~/K~ separable.\begin{equation} \begin{gathered} e=[\Gamma_E:\Gamma_K],\qquad f=[\widetilde E:\widetilde K],\qquad [E:K]=ef,\\ e\in(K^\circ)^\times,\qquad \widetilde E/\widetilde K\text{ separable}. \end{gathered} \end{equation} Thus defectlessness is an assumption, not a consequence of the other two conditions. An actual unramified lift means the finite separable residue lift constructed by NAG-002050; in the Galois case its valuation rings have the finite free etale torsor property of NAG-002262.

An exact Kummer step F⊂F(s)F\subset F(s) of degree n≥2n\ge2 means the precise coefficient class of NAG-002610: n∈(F∘)×,ζ∈F∘,ord⁡(ζ)=ord⁡(ζˉ)=n,sn∈F×,v(s)>0,jv(s)∉ΓF(0<j<n).\begin{equation} \begin{gathered} n\in(F^\circ)^\times,\qquad \zeta\in F^\circ,\qquad \operatorname{ord}(\zeta)=\operatorname{ord}(\bar\zeta)=n,\\ s^n\in F^\times,\qquad v(s)>0,\qquad jv(s)\notin\Gamma_F\quad(0<j<n). \end{gathered} \end{equation}

Lemma. NAG-003143 (Corrected radicals and a value basis). Let E/FE/F be finite with equal residue fields and [E:F]=[ΓE:ΓF]=e[E:F]=[\Gamma_E:\Gamma_F]=e, where e∈(F∘)×e\in(F^\circ)^\times. There is a tower F=F0⊂F1⊂⋯⊂Fr=E,Fi=Fi−1(si),sini∈Fi−1×,\begin{equation} F=F_0\subset F_1\subset\cdots\subset F_r=E,\qquad F_i=F_{i-1}(s_i),\qquad s_i^{n_i}\in F_{i-1}^{\times}, \end{equation} with ni≥2n_i\ge2, v(si)>0v(s_i)>0, and [Fi:Fi−1]=[ΓFi:ΓFi−1]=ni[F_i:F_{i-1}]=[\Gamma_{F_i}:\Gamma_{F_{i-1}}]=n_i. The residue fields are unchanged, ∏ini=e\prod_i n_i=e, and the monomials ∏isiai\prod_i s_i^{a_i}, 0≤ai<ni0\le a_i<n_i, have distinct value classes modulo ΓF\Gamma_F and form an FF-basis of EE. The empty tower is allowed when e=1e=1.

Every step is exact Kummer if either E/FE/F is Galois or FF contains all ee-th roots of unity. Moreover, if N/FN/F is finite with ΓN=ΓF\Gamma_N=\Gamma_F in a common valued overfield, then E⊗FNE\otimes_F N is the field ENEN, of degree and value index ee over NN and with residue field N~\widetilde N. If NN contains all ee-th roots of unity, the base-changed tower (G.0.28) consists of exact Kummer steps of the same degrees.

Direct inputs NAG-001450, NAG-002049, NAG-002610.

Proof. Suppose Fi−1F_{i-1} has been constructed and its value group is not ΓE\Gamma_E. Choose a∈E×a\in E^\times whose value has order ni>1n_i>1 in ΓE/ΓFi−1\Gamma_E/\Gamma_{F_{i-1}}. Choose c∈Fi−1×c\in F_{i-1}^\times with v(c)=niv(a)v(c)=n_iv(a), and choose a unit u∈Fi−1∘u\in F_{i-1}^\circ lifting the residue of ani/ca^{n_i}/c. This is possible because every intermediate residue field is the common residue field. The integer nin_i divides ee and is a unit. The simple root 11 of the reduction of Tni−ani/(cu)T^{n_i}-a^{n_i}/(cu) therefore lifts to d∈1+mEd\in1+\mathfrak m_E. Set si=a/ds_i=a/d. Then sini=cus_i^{n_i}=cu and v(si)=v(a)v(s_i)=v(a). Multiplying sis_i by a sufficiently large positive-value power of any fixed element of FF makes its value positive and preserves all these assertions about its class and its power.

The powers 1,si,…,sini−11,s_i,\ldots,s_i^{n_i-1} have distinct value classes over Fi−1F_{i-1}. A nonzero linear combination has a unique term of smallest value and cannot vanish. Thus its degree is at least nin_i and, by the displayed polynomial, at most nin_i. The fundamental inequality Stacks 0ASH now shows that its value index is exactly nin_i and its residue degree is one. In particular ΓFi\Gamma_{F_i} is exactly the subgroup generated by ΓFi−1\Gamma_{F_{i-1}} and v(si)v(s_i).

These subgroups strictly increase in the finite quotient ΓE/ΓF\Gamma_E/\Gamma_F, so the process stops. Products of the listed powers give distinct representatives of successive group cosets, hence distinct classes modulo ΓF\Gamma_F. At termination their number is ee. Their degrees multiply to e=[E:F]e=[E:F], so the terminal field is EE and they are the required basis. The unique-minimum argument also gives the value of every nonzero basis expansion.

If E/FE/F is Galois, it is normal over every intermediate field, by Stacks 09DW. The minimal polynomial Tni−siniT^{n_i}-s_i^{n_i} therefore splits in EE; it has degree nin_i and is separable. All nin_i-th roots of unity belong to EE. Reduction is injective on them: two with the same reduction coincide by uniqueness of simple-root lifts, Stacks 06RR. Their residues lie in Fi−1~\widetilde{F_{i-1}}, and Hensel lifting there, with the same uniqueness in EE, places all these roots in Fi−1F_{i-1}. The group of roots is cyclic by Stacks 09HX, since a degree-dd polynomial has at most dd roots. Hence a primitive root and its reduction both have order nin_i. If instead μe⊂F\mu_e\subset F, the same conclusion follows because ni∣en_i\mid e, with reduction injective by Stacks 06RR. In either case the polynomial splits in FiF_i, and its root multiplications give the cyclic Galois group. This proves (G.0.27).

Finally the same ee monomials remain linearly independent over NN, since ΓN=ΓF\Gamma_N=\Gamma_F. The multiplication map E⊗FN→ENE\otimes_F N\to EN is therefore injective and has image of NN-dimension ee; its image is a finite-dimensional domain, hence a field and thus all of ENEN. Its value group contains ΓE\Gamma_E, of index ee over ΓN\Gamma_N. Another application of Stacks 0ASH forces equality of the value index with ee and equality of residue fields with N~\widetilde N. Applying the same basis argument to each initial segment of (G.0.28) preserves its degree and value index. If μe⊂N\mu_e\subset N, each segment has the required roots of unity with unchanged residue order, so is exact Kummer. ◻

Proposition. NAG-003144 (An actual unramified base change supplies the steps). Let E/KE/K satisfy (G.0.26), without assuming it Galois. There is an actual unramified subextension M/KM/K in EE with M~=E~\widetilde M=\widetilde E and [M:K]=f[M:K]=f. There is a finite Galois actual unramified extension N/KN/K, containing this MM in a common valued overfield, such that EN/NEN/N is a tower of exact Kummer steps, of total degree and value index ee and with residue field N~\widetilde N. For e=1e=1 this tower is empty. Neither a Galois action on a supplied EE-model nor a Galois extension EN/KEN/K is asserted by this conclusion.

Direct inputs NAG-002049, NAG-002050, NAG-002052, NAG-002262, NAG-003143.

Proof. Write k=K~k=\widetilde K and l=E~l=\widetilde E. Use the primitive residue generator and its lifted monic polynomial in the proof of NAG-002050, but lift its simple residue root inside EE. This gives α∈E∘\alpha\in E^\circ and M=K(α)M=K(\alpha). The degree is at most ff; the residue degree is at least ff. The fundamental inequality Stacks 0ASH forces [M:K]=f[M:K]=f, ΓM=ΓK\Gamma_M=\Gamma_K and M~=l\widetilde M=l. This is the same actual residue lift, with its chosen embedding in EE. By (G.0.26), E/ME/M has degree and value index ee and unchanged residue field.

Choose a finite Galois residue extension k′/kk'/k containing ll and all ee-th roots of unity. It exists by taking the splitting field of a separable defining polynomial for l/kl/k together with Te−1T^e-1, using Stacks 09HU; the latter polynomial is separable because ee is a unit. First lift k′/lk'/l over MM by NAG-002050, obtaining N/MN/M in a common algebraic overfield of EE. Thus [N:K]=[k′:k][N:K]=[k':k], N~=k′\widetilde N=k' and ΓN=ΓK\Gamma_N=\Gamma_K.

To identify N/KN/K as the actual Galois unramified lift, lift a primitive generator of k′/kk'/k inside NN by Hensel. Its field over KK has degree at least [k′:k][k':k], by its residue, and at most that degree, by its lifted polynomial. It is therefore NN. All roots of the residue polynomial lift inside NN, since its residue is k′k'; they are distinct and exhaust the roots of the lifted polynomial. Thus N/KN/K is Galois and is precisely the lift of NAG-002052. Its valuation rings have the torsor property in NAG-002262. Hensel also lifts all roots of Te−1T^e-1 from k′k' to NN. Lemma G.0.18, applied to E/ME/M and N/MN/M, proves all the assertions about EN/NEN/N. The construction supplies field inclusions; it does not supply action maps on a model or assert normality over KK. ◻

Theorem. NAG-003145 (A Galois-normal sequence of exact Kummer steps). Let L/KL/K be finite Galois satisfying (G.0.26), with group GG. Put k=K~k=\widetilde K and l=L~l=\widetilde L. Then l/kl/k is Galois, and its actual unramified lift M⊂LM\subset L satisfies 1⟶I=Gal⁡(L/M)⟶G⟶Gal⁡(l/k)⟶1,∣I∣=e.\begin{equation} 1\longrightarrow I=\operatorname{Gal}(L/M)\longrightarrow G \longrightarrow\operatorname{Gal}(l/k)\longrightarrow1,\qquad |I|=e. \end{equation} Here II is exactly the kernel of the action on ll. It is abelian, and every subgroup of II is normal in GG. There is consequently a chain 1=J0⊂J1⊂⋯⊂Jr=I,Fj=LJj,L=F0⊃⋯⊃Fr=M,\begin{equation} 1=J_0\subset J_1\subset\cdots\subset J_r=I,\qquad F_j=L^{J_j},\qquad L=F_0\supset\cdots\supset F_r=M, \end{equation} whose nontrivial factors Jj/Jj−1J_j/J_{j-1} have prime order and whose steps Fj−1/FjF_{j-1}/F_j are exact Kummer. All fields in (G.0.30) have residue field ll. No roots of unity are assumed in KK, and no cyclicity of II or splitting of (G.0.29) is assumed.

Direct inputs NAG-001450, NAG-002052, NAG-003143, NAG-003144.

Proof. Take the embedded M=K(α)M=K(\alpha) constructed in Proposition G.0.19; let PP be its lifted polynomial of degree ff. It is minimal by the degree calculation there. Since L/KL/K is normal, PP splits in LL. Every root is integral, by the unique extending absolute value and conjugacy. Reducing this factorization shows that the separable minimal polynomial of αˉ\bar\alpha splits in ll. Thus l/kl/k is Galois. Every root of its polynomial lifts inside MM, since M~=l\widetilde M=l; these lifts exhaust the roots of PP. Consequently M/KM/K is Galois, with reduction identifying its group with Gal⁡(l/k)\operatorname{Gal}(l/k), just as in NAG-002052. Restriction and Stacks 09DW give (G.0.29). An element fixing MM fixes ll. Conversely an element acting trivially on ll fixes α\alpha by Stacks 06RR, hence fixes MM. This identifies inertia and proves ∣I∣=[L:M]=e|I|=[L:M]=e.

We give the needed ramification argument to avoid importing a DVR-only cyclicity theorem. For i∈Ii\in I define θi:ΓL/ΓM⟶l×,θi(v(a))=i(a)/a‾.\begin{equation} \theta_i:\Gamma_L/\Gamma_M\longrightarrow l^\times,\qquad \theta_i(v(a))=\overline{i(a)/a}. \end{equation} This is well defined: changing aa by an element of the same value changes the ratio by i(u)/ui(u)/u for a unit uu, whose residue is one; changing the value by v(b)v(b), b∈M×b\in M^\times, has no effect. Multiplication shows that θi\theta_i is a character. Since II fixes ll, composition gives a homomorphism I→Hom⁡(ΓL/ΓM,l×)I\to\operatorname{Hom}(\Gamma_L/\Gamma_M,l^\times). It is injective. Indeed use the radical tower of Lemma G.0.18 for L/ML/M. If θi=1\theta_i=1 and ii fixes the previous field, then i(sj)/sji(s_j)/s_j is an njn_j-th root of unity with residue one. It equals one by Stacks 06RR. Induction fixes every radical and hence all of LL. The target of (G.0.31) is abelian, so II is abelian.

All values of (G.0.31) lie in μe(l)\mu_e(l), since the value quotient has order ee. This finite group is cyclic by Stacks 09HX. For g∈Gg\in G, valuation preservation gives θgig−1(γ)=gˉ(θi(γ)).\begin{equation} \theta_{gig^{-1}}(\gamma)=\bar g(\theta_i(\gamma)). \end{equation} To see this directly, represent γ\gamma by aa and apply gg to i(g−1a)/(g−1a)i(g^{-1}a)/(g^{-1}a); the value of g−1ag^{-1}a is still γ\gamma. If t=∣μe(l)∣t=|\mu_e(l)|, the automorphism gˉ\bar g of this cyclic group is exponentiation by some integer qgq_g coprime to tt. Thus injectivity of (G.0.31) implies gig−1=iqggig^{-1}=i^{q_g}. Also (G.0.31) implies it=1i^t=1 for every ii. Exponentiation by qgq_g therefore preserves every subgroup of II, proving its normality in GG.

A finite abelian group has a series with prime cyclic factors: take a maximal proper subgroup and induct on its order; the quotient is simple abelian, so any nonidentity element generates it and its order is prime. All groups in the resulting series are GG-normal by the preceding paragraph. The correspondence Stacks 09DW gives (G.0.30) and its cyclic prime-degree extensions. Their residue fields all equal ll, since they lie between MM and LL. The product of their degrees and the product of their value indices both equal ee. Each index is at most its degree by Stacks 0ASH, so equality holds at every step. For a prime-degree step its value quotient has prime order, hence is cyclic. The first radical in Lemma G.0.18 already generates that step; its Galois clause supplies a primitive root with primitive reduction. This is exactly (G.0.27) for each step. ◻

Proposition. NAG-003149 (A noncyclic obstruction surviving unramified base change). Give Q(x,y)\mathbb Q(x,y) the valuation trivial on Q\mathbb Q with v(x)=1v(x)=1 and v(y)=2v(y)=\sqrt2, defined on polynomials by the least monomial value, and let KK be its completion. Then ΓK=Z+2Z\Gamma_K=\mathbb Z+\sqrt2\mathbb Z and K~=Q\widetilde K=\mathbb Q. For L=K(u,w),u2=x,w2=y,\begin{equation} L=K(u,w),\qquad u^2=x,\qquad w^2=y, \end{equation} the field extension is tame Galois, of degree and value index four, with unchanged residue field, and Gal⁡(L/K)=I≃(Z/2)2,ΓL/ΓK≃(Z/2)2.\begin{equation} \operatorname{Gal}(L/K)=I\simeq(\mathbb Z/2)^2,\qquad \Gamma_L/\Gamma_K\simeq(\mathbb Z/2)^2. \end{equation} The same assertions hold over any actual finite unramified extension N/KN/K after replacing LL by LNLN, with residue field N~\widetilde N and degree four over NN. Consequently no such base change converts this into one full-value-index cyclic Kummer step. Two exact quadratic steps suffice. This is a complete rank-one nondiscrete example.

Direct inputs NAG-001450, NAG-002050, NAG-003143.

Proof. Irrationality makes the values of distinct monomials distinct. A polynomial has a unique lowest term, whose product with the lowest term of another polynomial is again uniquely lowest. This proves multiplicativity and extends the valuation to fractions. Its group is Z+2Z\mathbb Z+\sqrt2\mathbb Z. A value-zero fraction has numerator and denominator with the same leading exponents; its residue is their rational leading coefficient ratio. Hence its residue field is Q\mathbb Q. Completion changes neither value group nor residue field: for a nonzero completed element choose an approximation whose difference has larger value; the two values and their leading residues agree. For value-zero elements this also identifies ordinary residues. The value group is nondiscrete. Indeed the positive numbers (2−1)m(\sqrt2-1)^m lie in Z+2Z\mathbb Z+\sqrt2\mathbb Z and tend to zero. It is a nonzero ordered subgroup of R\mathbb R, hence has rank one.

Take the two roots in a finite algebraic extension of KK, with its unique extending valuation. The four monomials 1,u,w,uw1,u,w,uw have values in four distinct cosets modulo ΓK\Gamma_K: their values are 0,1/2,2/2,(1+2)/20,1/2,\sqrt2/2,(1+\sqrt2)/2. They are linearly independent by the unique-minimum argument. Their span is closed under multiplication using (G.0.33), and the extension has degree at most four, so they are a basis and [L:K]=4[L:K]=4. Their values generate 12ΓK\frac12\Gamma_K, of index four. The fundamental inequality Stacks 0ASH forces this to be the whole value group and forces residue degree one. Independent sign changes of u,wu,w are four automorphisms, giving (G.0.34); they are all inertia. Residue characteristic zero, equality of degree with value index, and unchanged residue verify all of (G.0.26).

The tower K⊂K(u)⊂LK\subset K(u)\subset L has value index and degree two at each step: the new value classes are represented by 1/21/2 and 2/2\sqrt2/2, respectively. Their positive generators and the root −1-1, of residue order two, satisfy (G.0.27). For any actual unramified N/KN/K, its value group is still ΓK\Gamma_K. Lemma G.0.18 proves L⊗KN=LNL\otimes_K N=LN, degree and value index four, and unchanged residue over NN. The four sign changes extend on this tensor product, giving the same noncyclic inertia and value quotient. A single (G.0.27) step of degree four has a cyclic value quotient of order four and a cyclic Galois group. Both conflict with (G.0.34). This proves the asserted obstruction, even if NN contains a primitive fourth root of unity. ◻

The toric Gauss section and smooth tests

Let KK be algebraically closed, complete and nontrivially valued, with its fixed absolute value and v=−log⁡∣⋅∣v=-\log|\cdot|. Throughout, d≥2,X=(PK1)d,T=(Gm,K)d,Li=pr⁡i∗O(mi∞),mi≥1.d\geq2,\qquad X=(\mathbf P^1_K)^d,\qquad T=(\mathbf G_{m,K})^d, \qquad L_i=\operatorname{pr}_i^*\mathcal O(m_i\infty),\quad m_i\geq1. We use the usual good Hausdorff analytification, pure of dimension dd. Put ti=−log⁡∣Ti∣t_i=-\log|T_i| on TanT^\mathrm{an}, and let tˉ:Xan→R‾,d\bar t:X^\mathrm{an}\to\overline{\mathbf R}^{,d} be its continuous extension, where R‾=[−∞,+∞]\overline{\mathbf R}=[-\infty,+\infty]; zero maps to +∞+\infty and infinity to −∞-\infty. The lattices are N=M=ZdN=M=\mathbf Z^d in the coordinate dual bases, and dtdt has covolume one.

Choose ϕi∈C∞(R)\phi_i\in C^\infty(\mathbf R) convex, with exact tails ϕi(t)=−mit+ci−(t≤Ai),ϕi(t)=ci+(t≥Bi),Ai<Bi.\begin{equation} \phi_i(t)=-m_i t+c_i^-\quad(t\leq A_i),\qquad \phi_i(t)=c_i^+\quad(t\geq B_i),\qquad A_i<B_i. \end{equation} The frame ei,0=1e_{i,0}=1 of LiL_i has weight ϕi(ti)\phi_i(t_i), and ei,∞=Timiei,0e_{i,\infty}=T_i^{m_i}e_{i,0} has weight ϕi(ti)+miti\phi_i(t_i)+m_it_i. Both are constant near their respective boundary points. These define global metrics hih_i in both the ordinary moment-smooth convention and the primary algebraic tropical smooth convention. Explicit examples are ϕi(t)=mi∫t∞(s−t)ρi(s) ds\phi_i(t)=m_i\int_t^\infty(s-t)\rho_i(s)\,ds, with ρi∈Cc∞((Ai,Bi))\rho_i\in C_c^\infty((A_i,B_i)), ρi≥0\rho_i\geq0, ∫ρi=1\int\rho_i=1.

Set θi=d′d′′ϕi(ti),θ=⋀i=1dθi,W(t)=∏iϕi′′(ti),D=∏imi,S=∏isupp⁡(ϕi′′).\begin{equation} \theta_i=d'd''\phi_i(t_i),\quad \theta=\bigwedge_{i=1}^d\theta_i, \quad W(t)=\prod_i\phi_i''(t_i),\quad D=\prod_i m_i, \quad S=\prod_i\operatorname{supp}(\phi_i''). \end{equation} Use the positive generator ωd=⋀i(d′ti∧d′′ti)\omega_d=\bigwedge_i(d't_i\wedge d''t_i), so that θ=t∗(Wωd)\theta=t^*(W\omega_d) and ∫aωd=∫a(t)dt\int a\omega_d=\int a(t)dt. There is no d!d! in this mixed wedge and no rescaling of the valuation. The curvature is zero near Xan∖TanX^\mathrm{an}\setminus T^\mathrm{an}. If a finite box BB contains SS in its interior, then the whole compact polyannulus Q=t−1(B)Q=t^{-1}(B) is an ordinary neighbourhood of its support. All frozen chart weights use full fibres and actual canonical Banach algebras.

For w∈Rdw\in\mathbf R^d, let ρ(w)\rho(w) be the Gauss point ∣∑m∈ZdamTm∣ρ(w)=max⁡m∣am∣e−⟨m,w⟩on finite Laurent polynomials.\begin{equation} \left|\sum_{m\in\mathbf Z^d}a_mT^m\right|_{\rho(w)} =\max_m |a_m|e^{-\langle m,w\rangle} \quad\text{on finite Laurent polynomials.} \end{equation} Write Γ=ρ(Rd)\Gamma=\rho(\mathbf R^d).

In the primary algebraic tropical convention used below, the germ is read by tangential restriction on the faces of the actual tropical chart, after subdivision. The ordinary convention uses the actual smooth moment germs.

Lemma. NAG-002744 (Local cancellation forces rank drop). The map ρ\rho extends continuously to a section ρˉ:R‾,d→Xan\bar\rho:\overline{\mathbf R}^{,d}\to X^\mathrm{an} of tˉ\bar t. Its image Γˉ\bar\Gamma is closed, and Γ=Γˉ∩Tan\Gamma=\bar\Gamma\cap T^\mathrm{an}. For every x∈Tan∖Γx\in T^\mathrm{an}\setminus\Gamma there is an ordinary open neighbourhood Ux⊂TanU_x\subset T^\mathrm{an} and a finite union HxH_x of proper integral-normal affine hyperplanes in Rd\mathbf R^d such that t(Ux)⊂Hx.t(U_x)\subset H_x. Consequently any smooth moment form t∗αt^*\alpha with α\alpha of bidegree (d,d)(d,d) has zero germ at xx, in both the frozen ordinary moment convention and the primary algebraic tropical convention. In particular the two versions of θ\theta in (G.0.36) satisfy supp⁡θ⊂ρ(S).\begin{equation} \operatorname{supp}\theta\subset\rho(S). \end{equation}

Proof. The weighted Gauss norm is multiplicative by NAG-000182. On a finite coordinate chart of XX, allow some radii to be zero and first quotient by those coordinate variables. The Gauss norm in the remaining variables gives a point on that chart. These constructions agree under inversion of nonzero coordinates and give ρˉ\bar\rho. Evaluation on a polynomial is a finite maximum of continuous monomial functions of the radii, including zero radii. This proves continuity in each analytic coordinate chart, hence globally. The equality tˉρˉ=1\bar t\bar\rho=1 follows directly. The domain is compact and XanX^\mathrm{an} is Hausdorff; thus ρˉ\bar\rho is a homeomorphism onto its closed image. A finite tropical coordinate has positive finite radius, so its image is in TanT^\mathrm{an}, proving the assertion about Γ\Gamma.

Fix x∉Γx\notin\Gamma, and put w=t(x)w=t(x). Distinct points of the torus analytification are distinct multiplicative seminorms on its Laurent coordinate ring. The ultrametric inequality always gives ∣f(x)∣≤∣f(ρ(w))∣|f(x)|\leq|f(\rho(w))|. Hence some Laurent polynomial f=∑mamTmf=\sum_m a_mT^m, with like terms collected, satisfies ∣f(x)∣<Mf(t(x)),Mf(u)=max⁡m∣am∣e−⟨m,u⟩.|f(x)|<M_f(t(x)),\qquad M_f(u)=\max_m |a_m|e^{-\langle m,u\rangle}. Both sides are continuous and MfM_f is positive. The same strict inequality defines an ordinary open neighbourhood UxU_x. At every y∈Uxy\in U_x the largest term in ff cannot be unique: a unique largest summand has the absolute value of the entire sum in a valued field, and the same statement holds in the completed residue field at yy. At least two distinct exponents m,nm,n therefore satisfy v(am)+⟨m,t(y)⟩=v(an)+⟨n,t(y)⟩.v(a_m)+\langle m,t(y)\rangle =v(a_n)+\langle n,t(y)\rangle. These are finitely many proper affine hyperplanes with nonzero integer normal m−nm-n. Their union is the required HxH_x.

For the frozen germs, choose an ordinary affinoid neighbourhood of xx inside UxU_x, using the neighbourhood construction in NAG-002424. Its actual TT-moment image is finite polyhedral. Subdivision along HxH_x puts every face in dimension at most d−1d-1. Restriction of an alternating dd-covector to each tangent space is zero. Thus the representative t∗αt^*\alpha is tangentially zero on every face, which is exactly a zero germ in the frozen definition.

For the primary convention the same argument can be placed on an explicit algebraic tropical chart. If f(x)≠0f(x)\ne0, put g=fg=f. If f(x)=0f(x)=0, choose a∈K×a\in K^\times with ∣a∣<Mf(w)|a|<M_f(w) and put g=f−ag=f-a. Then g(x)≠0g(x)\ne0, Mg(w)=Mf(w)M_g(w)=M_f(w) and ∣g(x)∣<Mg(w)|g(x)|<M_g(w). The very affine open T∖V(g)T\setminus V(g) has unit coordinates Ti,gT_i,g, possibly augmented to its canonical unit tuple. On its tropical open subset defined by ∣g∣<Mg(t)|g|<M_g(t), projection to the TiT_i coordinates lands in the finite union of hyperplanes just proved. Subdivide its polyhedral faces accordingly. On every face the projected tangent space has rank at most d−1d-1, so the pullback of α\alpha is zero by polyhedral restriction. This gives the zero primary smooth germ without an approximation of analytic moment maps.

Finally θ\theta is already zero off t−1(S)t^{-1}(S) in the torus and near the toric boundary. The argument just given removes every remaining point outside Γ\Gamma. Since ρ(S)\rho(S) is compact and closed, the support inclusion (G.0.38) follows. ◻

Let A\mathcal A be the real algebra of global ordinary smooth moment functions on XanX^\mathrm{an}. Let Acom\mathcal A_{\rm com} be its subalgebra of functions also smooth on primary algebraic tropical charts.

Lemma. NAG-002749 (A common smooth algebra is uniformly dense). The algebra Acom\mathcal A_{\rm com} separates points of XanX^\mathrm{an} and contains the constants. Both Acom\mathcal A_{\rm com} and A\mathcal A are uniformly dense in C(Xan)C(X^\mathrm{an}).

Proof. For any positive multi-degree ℓ=(ℓ1,…,ℓd)\ell=(\ell_1,\ldots,\ell_d), the line L(ℓ)=⨂ipr⁡i∗O(ℓi∞)L(\ell)=\bigotimes_i\operatorname{pr}_i^*\mathcal O(\ell_i\infty) has a common smooth metric with weight ∑i(ℓi/mi)ϕi(ti)\sum_i(\ell_i/m_i)\phi_i(t_i) on the torus. If ss is a nonzero global algebraic section and χ∈C∞(R)\chi\in C^\infty(\mathbf R) is constant at both sufficiently large positive and negative arguments, then us,χ(x)=χ(−log⁡∥s(x)∥)\begin{equation} u_{s,\chi}(x)=\chi(-\log\|s(x)\|) \end{equation} is a common smooth function. At a zero of ss, use the constant value χ(+∞)\chi(+\infty). This extends continuously and is locally constant there. Off the zero set, in an algebraic frame s=fes=f e its argument is −log⁡∣f∣−log⁡∥e∥-\log|f|-\log\|e\|, a smooth function of algebraic unit moments. Near a zero, one may make a tropical neighbourhood explicit by replacing the vanishing regular coefficient ff with f−af-a, where a≠0a\ne0 is small, and imposing a strict upper bound on ∣f−a∣|f-a| larger than ∣a∣|a|. After shrinking the other moment bounds, this lies in the region on which (G.0.39) is constant. At a coordinate boundary the same construction with Ti−bT_i-b or Ti−1−bT_i^{-1}-b describes the region where the metric weight has its constant tail. Removing the zeros of these finitely many nonvanishing algebraic units gives a very affine chart. Thus these functions have the required representatives in both conventions. Sums and products are computed using joint moment tuples.

Take distinct x,y∈Xanx,y\in X^\mathrm{an}. For each coordinate choose ai∈Ka_i\in K different from any rigid projection of xx or yy to that factor. This is possible since at most two values are excluded and KK is infinite. Then both points lie in U=∏i(P1∖{ai})≃AdU=\prod_i(\mathbf P^1\setminus\{a_i\})\simeq\mathbf A^d, with affine coordinates zi=(Ti−ai)−1z_i=(T_i-a_i)^{-1}. Some f∈K[z1,…,zd]f\in K[z_1,\ldots,z_d] has ∣f(x)∣≠∣f(y)∣|f(x)|\ne|f(y)|, because points of this analytification are polynomial seminorms. Choose positive ℓi\ell_i at least the degree of ff in ziz_i. There is a section s0s_0 of L(ℓ)L(\ell) with divisor ∑iℓi{Ti=ai}\sum_i\ell_i\{T_i=a_i\}, and sf=fs0s_f=f s_0 extends to a global section by the degree bound. The section s0s_0 is nonvanishing at xx and yy. If its norms there differ, use it in (G.0.39), with χ\chi taking different values at the two logarithmic norms. If its norms agree, the norms of sfs_f differ and the same construction with sfs_f separates the points. This also covers a zero value of ff, using the constant tail of χ\chi.

Here is a complete compact-space density argument. Let YY be a compact Hausdorff space and A⊂C(Y,R)A\subset C(Y,\mathbf R) a unital point-separating algebra. Its uniform closure BB is an algebra. It is closed under absolute values: on a compact interval, polynomial approximation of ∣r∣|r| follows from the Bernstein polynomials Bnh(t)=∑k=0nh(k/n)(nk)tk(1−t)n−k.B_n h(t)=\sum_{k=0}^n h(k/n){n\choose k}t^k(1-t)^{n-k}. Indeed, for continuous hh on [0,1][0,1], the binomial mean and variance give, for every δ>0\delta>0, ∣Bnh(t)−h(t)∣≤ωh(δ)+∥h∥∞2nδ2,|B_nh(t)-h(t)|\leq\omega_h(\delta) +\frac{\|h\|_\infty}{2n\delta^2}, uniformly in tt. Choose δ=n−1/4\delta=n^{-1/4} and use uniform continuity. Rescaling proves the required approximation of ∣r∣|r|. Polynomial composition therefore puts ∣b∣|b| in BB for every b∈Bb\in B. Thus BB is closed under finite maxima and minima.

Given u∈C(Y)u\in C(Y) and ϵ>0\epsilon>0, for any x,y∈Yx,y\in Y a suitable affine function of a separating element of AA takes the values u(x),u(y)u(x),u(y) at x,yx,y; for x=yx=y use a constant. Denote it by axya_{xy}. For fixed xx, the open sets where axy>u−ϵa_{xy}>u-\epsilon cover YY. Choose finitely many yy and let bxb_x be the maximum of the corresponding axya_{xy}. Then bx∈Bb_x\in B, bx>u−ϵb_x>u-\epsilon everywhere and bx(x)=u(x)b_x(x)=u(x). Hence bx<u+ϵb_x<u+\epsilon on a neighbourhood of xx. Finitely many such neighbourhoods cover YY. The minimum bb of their bxb_x satisfies u−ϵ<b<u+ϵu-\epsilon<b<u+\epsilon on YY and belongs to BB. Thus u∈Bu\in B. Apply this argument to the algebra generated by the functions (G.0.39) and the constants. It is contained in Acom⊂A\mathcal A_{\rm com}\subset\mathcal A, proving both density claims. ◻

Radon determination on the toric Gauss section

Use the same standing product, curvature and chart conventions.

Theorem. NAG-002746 (When radial Radon data determines the full measure). Let ν\nu be a finite positive Radon measure on R‾,d\overline{\mathbf R}^{,d} and μ\mu a finite positive Radon measure on XanX^\mathrm{an} with tˉ∗μ=ν\bar t_*\mu=\nu. Then μ=ρˉ∗ν⟺μ(Xan∖Γˉ)=0.\begin{equation} \mu=\bar\rho_*\nu \quad\Longleftrightarrow\quad \mu(X^\mathrm{an}\setminus\bar\Gamma)=0. \end{equation} If ν\nu gives zero mass to the tropical boundary and to every proper affine hyperplane in Rd\mathbf R^d with nonzero integer normal, the right-hand condition follows automatically. In particular it follows for ν=W(t)dt\nu=W(t)dt from (G.0.36). In this case for every u∈C(Xan)u\in C(X^\mathrm{an}), ∫u dμ=∫u(ρ(t))W(t) dt.\begin{equation} \int u\,d\mu=\int u(\rho(t))W(t)\,dt. \end{equation} No countability condition on the value group or the set of analytic points is needed. Positivity and Radon regularity are hypotheses.

Proof. On Γˉ\bar\Gamma the map ρˉtˉ\bar\rho\bar t is the identity. If μ\mu is concentrated there, then for every continuous uu, ∫u dμ=∫u∘ρˉ∘tˉ dμ=∫u∘ρˉ dν.\int u\,d\mu=\int u\circ\bar\rho\circ\bar t\,d\mu =\int u\circ\bar\rho\,d\nu. This proves the forward identification of measures; the converse concentration is immediate from the image of ρˉ\bar\rho.

Now impose the additional hypotheses on ν\nu. The analytic toric boundary has μ\mu-mass zero because its image is in the tropical boundary. For each x∈Tan∖Γx\in T^\mathrm{an}\setminus\Gamma, the neighbourhood in Lemma G.0.22 has mass zero: positivity gives 0≤μ(Ux)≤μ(t−1(Hx))=ν(Hx)=0.0\leq\mu(U_x)\leq\mu(t^{-1}(H_x))=\nu(H_x)=0. Every compact subset of the open set Tan∖ΓT^\mathrm{an}\setminus\Gamma is covered by finitely many such UxU_x, hence has mass zero. Inner regularity therefore makes the entire open set have mass zero. This finite-cover argument is essential; it does not take a countable union over all Laurent polynomials. Equation (G.0.40) now applies. An absolutely continuous measure WdtWdt assigns zero mass to all the specified hyperplanes, which proves (G.0.41). ◻

Normalizers in exact Kummer descent

PDF-derived transcription. Historical English reader, Section 1.55, Lemma 1.55.4, printed pages 220–221. The standing context is from printed pages 216–217. The historical original TeX remains unlocated.

The lemma uses the valued-field convention, the exact Kummer step (K*) and the action convention below. The remaining definitions record the surrounding historical terminology.

Standing context

Throughout, a valued field is complete and nontrivially real-valued, with additive valuation vv, actual valuation ring F∘F^\circ, maximal ideal mF\mathfrak m_F, value group ΓF\Gamma_F, and residue field F~\widetilde F. Finite extensions carry their unique extending absolute values. Their existence, uniqueness and completeness are NAG-001450. Simple residue roots lift by NAG-002049.

For a finite separable extension E/KE/K, the explicit tame hypotheses in this patch mean all three conditions e=[ΓE:ΓK],f=[E~:K~],[E:K]=ef,e∈(K∘)×,E~/K~ separable.\begin{equation} \tag{Tg} \begin{gathered} e=[\Gamma_E:\Gamma_K],\qquad f=[\widetilde E:\widetilde K],\qquad [E:K]=ef,\\ e\in(K^\circ)^\times,\qquad \widetilde E/\widetilde K\text{ separable}. \end{gathered} \end{equation} Thus defectlessness is an assumption, not a consequence of the other two conditions. An actual unramified lift means the finite separable residue lift constructed by NAG-002050; in the Galois case its valuation rings have the finite free etale torsor property of NAG-002262.

An exact Kummer step F⊂F(s)F\subset F(s) of degree n≥2n\ge2 means the precise coefficient class of NAG-002610: n∈(F∘)×,ζ∈F∘,ord⁡(ζ)=ord⁡(ζˉ)=n,sn∈F×,v(s)>0,jv(s)∉ΓF(0<j<n).\begin{equation} \tag{K*} \begin{gathered} n\in(F^\circ)^\times,\qquad \zeta\in F^\circ,\qquad \operatorname{ord}(\zeta)=\operatorname{ord}(\bar\zeta)=n,\\ s^n\in F^\times,\qquad v(s)>0,\qquad jv(s)\notin\Gamma_F\quad(0<j<n). \end{gathered} \end{equation} Its integral extension is faithfully flat but need not be finite. We use left scheme actions ρg\rho_g covering Spec⁡(g−1)\operatorname{Spec}(g^{-1}), and linearizations ρg∗L≃L\rho_g^*\mathcal L\simeq\mathcal L, as in the frozen mixed packet. Identity on a special fibre always means identity on the entire scheme or restricted module, including its structure sheaf.

Lemma. NAG-003146 (Retaining a normalizer action at a Kummer step). Let F/EF/E be an exact Kummer step with group QQ. Let a finite group PP act on FF by isometric field automorphisms preserving EE, with Q◃PQ\triangleleft P, and suppose its restriction to QQ is the specified Galois action. Let Z/F∘Z/F^\circ be projective, flat, finitely presented and nodal of relative dimension one, with a compatible semilinear PP-action. Suppose QQ acts identically on the entire special scheme. Then the canonical QQ-descent ZE/E∘Z_E/E^\circ carries a canonical semilinear P/QP/Q-action, and its base-change comparison is PP-equivariant.

A PP-linearized relatively ample invertible sheaf with identity QQ-action on its full residue module descends to a relatively ample P/QP/Q-linearized invertible sheaf on ZEZ_E. Both descents are functorial for equivariant isomorphisms with their specified comparisons. The action of P/QP/Q on EE need not be faithful; there is no unramified hypothesis on any lower field.

Proof. The exact input NAG-002908 supplies ZEZ_E and its projective nodal structure. Its construction is that of NAG-002683, from the invariant rings of a finite QQ-invariant affine cover. Adjoin the finitely many PP-translates and all their nonempty finite intersections. Normality of QQ preserves invariance; separatedness over the affine base makes the intersections affine. Write DU=Γ(U,OZ),CU=DUQ.D_U=\Gamma(U,\mathcal O_Z),\qquad C_U=D_U^Q. The cited gluing theorem gives F∘⊗E∘CU≃DUF^\circ\otimes_{E^\circ}C_U\simeq D_U by multiplication and identifies invariant overlap rings with the actual overlaps of the descended charts.

Pullback by ρp:U→ρpU\rho_p:U\to\rho_pU maps DρpUD_{\rho_pU} to DUD_U and CρpUC_{\rho_pU} to CUC_U, because conjugation preserves QQ. It restricts on E∘E^\circ to (p∣E)−1(p|_E)^{-1} and commutes with restriction maps. These maps glue to an action on ZEZ_E; its group laws hold already on the invariant rings. Every q∈Qq\in Q fixes each UU and its invariant ring, so the action factors through P/QP/Q. The multiplication comparisons commute with these maps and are therefore PP-equivariant.

The input NAG-002906 descends the bundle and its relative ampleness. On the charts its modules are Γ(U,L)Q\Gamma(U,\mathcal L)^Q. Linearized pullbacks transport these modules by the same conjugation argument and commute with restriction. They glue to a linearization factoring through P/QP/Q, with the required comparison. An equivariant isomorphism transports both invariant rings and modules; intersecting covers gives the same maps, by the canonical overlap comparisons. This proves functoriality and uniqueness with a specified comparison.

Only invariant descent in the exact Kummer class was used; no integral ramified torsor was presumed. ◻