H · Quotient action in the mixed tame tower

6 NAG Tags

All valuation rings below are the actual valuation rings, with the specified extending absolute values. Fix the following supplied field configuration, denoted (T): K⊂M⊂L,O=K∘,A=M∘,B=L∘,k=K~,l=M~.(T)K\subset M\subset L,\qquad O=K^\circ,\quad A=M^\circ, \quad B=L^\circ,\qquad k=\widetilde K,\quad l=\widetilde M. \tag{T} The field KK is complete and nontrivially real-valued. The extension M/KM/K is the finite unramified Galois lift of a finite Galois extension l/kl/k, in the exact sense of . The extension L/KL/K is finite Galois, with group GG, and restriction gives the exact sequence 1⟶I=Gal⁡(L/M)⟶G⟶H=Gal⁡(M/K)⟶1.(8)1\longrightarrow I=\operatorname{Gal}(L/M) \longrightarrow G\longrightarrow H=\operatorname{Gal}(M/K) \longrightarrow1. \tag{8} The extension L/ML/M satisfies the exact class of : for e=∣I∣≥2e=|I|\geq2, e∈A×,ζ∈A,ord⁡(ζ)=ord⁡(ζˉ)=e,L=M(s),se=c∈M×,γ=v(s)>0,jγ∉v(M×)(0<j<e).(9)\begin{gathered} e\in A^\times,\quad \zeta\in A,\quad \operatorname{ord}(\zeta)=\operatorname{ord}(\bar\zeta)=e,\qquad L=M(s),\quad s^e=c\in M^\times,\\ \gamma=v(s)>0,\qquad j\gamma\notin v(M^\times)\quad(0<j<e). \end{gathered} \tag{9} Thus II is generated by s↦ζss\mapsto\zeta s, L~=l\widetilde L=l, and A→BA\to B is faithfully flat, possibly not finite. In contrast, O→AO\to A is finite free, faithfully flat and etale, and A⊗OA≃∏h∈HA,a⊗b⟼(a h(b))h.(10)A\otimes_O A\simeq\prod_{h\in H}A, \qquad a\otimes b\longmapsto(a\,h(b))_h. \tag{10} These facts are outputs of the two exact coefficient results just cited. No splitting of (8), invertibility of ∣H∣|H|, or discreteness of the value group is assumed.

Scheme actions are left actions ρg\rho_g covering Spec⁡(g−1)\operatorname{Spec}(g^{-1}). Linearizations are isomorphisms λg:ρg∗L≃L\lambda_g:\rho_g^*\mathcal L\simeq\mathcal L compatible with composition. The induced left action on functions and sections is formed using ρg−1∗\rho_{g^{-1}}^* and λg−1\lambda_{g^{-1}}, so its coefficient action is gg. Identity on a special scheme or module always means scheme-theoretic identity on the entire restriction.

Lemma. NAG-003008 (The quotient action on the inertia descent). Assume (T). Let YB/BY_B/B be projective, flat, finitely presented, and at-worst-nodal of relative dimension one, with a semilinear GG-action. Suppose II acts identically on C=(YB)lC=(Y_B)_l. The canonical II-descent YA/AY_A/A carries a canonical semilinear HH-action, and its comparison YA×AB≃YB(13)Y_A\times_A B\simeq Y_B \tag{13} is GG-equivariant, where gg acts on YAY_A through gIgI and on the coefficient ring BB as specified. The model YAY_A is projective, flat, finitely presented and nodal of relative dimension one.

If LB\mathcal L_B is a GG-linearized relatively ample invertible sheaf whose restricted II-action on CC is the identity, then its inertia descent LA\mathcal L_A carries a canonical HH-linearization, is relatively ample, and (13) identifies its pullback equivariantly with LB\mathcal L_B. Both constructions are functorial for equivariant isomorphisms and retain their specified comparison maps.

Proof.

Apply to the II-action over B/AB/A. Its descent is the invariant-affine gluing of , so it gives the asserted projective model and the II-equivariant comparison (13). We prove the extra equivariance rather than presuming an integral Galois torsor for B/AB/A.

Choose the finite II-invariant affine cover furnished by that gluing theorem. Adjoin all its GG-translates and all nonempty finite intersections. Normality of II makes every translated open II-invariant; separatedness makes the intersections affine. For a member UU, put DU=Γ(U,OYB),EU=DUI,UA=Spec⁡EU.(14)D_U=\Gamma(U,\mathcal O_{Y_B}),\qquad E_U=D_U^I, \qquad U_A=\operatorname{Spec}E_U. \tag{14} The gluing theorem identifies B⊗AEUB\otimes_A E_U with DUD_U by multiplication and identifies invariant overlap rings with the actual overlaps of the UAU_A.

For g∈Gg\in G, pullback along ρg:U→ρgU\rho_g:U\to\rho_gU maps DρgUD_{\rho_gU} to DUD_U and maps EρgUE_{\rho_gU} to EUE_U: conjugation by gg preserves II. On AA this map is (g∣M)−1(g|_M)^{-1}. It commutes with every restriction map and hence glues to an automorphism of YAY_A covering Spec⁡((g∣M)−1)\operatorname{Spec}((g|_M)^{-1}). The multiplication comparisons commute with these pullbacks, which proves the claimed GG-equivariance of (13). The group law follows already on the invariant coordinate rings. For i∈Ii\in I, the map on each EUE_U is the identity and ρiU=U\rho_iU=U, so the action factors through HH, independently of a choice of coset lifts.

The given bundle has the inertia descent LA=(p∗LB)I\mathcal L_A=(p_*\mathcal L_B)^I by , with p:YB→YAp:Y_B\to Y_A. That result supplies its invertibility, ampleness and comparison. On (14), its sections are Γ(U,LB)I\Gamma(U,\mathcal L_B)^I. The linearized pullbacks send these invariant sections to invariant sections by the same conjugation calculation, and commute with restriction. They therefore define a GG-linearization of LA\mathcal L_A over the just constructed action. Every i∈Ii\in I is the identity on these invariant modules, so the linearization factors through HH as well.

An equivariant isomorphism transports the invariant-affine data and their invariant modules. Intersecting either cover with the transported other cover gives a common invariant affine refinement; the canonical gluing comparisons agree there. This proves the stated functoriality and independence of the cover. Uniqueness of a comparison can also be checked after the faithfully flat base change, using the faithful part of 02W0 for schemes and the faithful part of 023T for their sheaves. ◻

Theorem. NAG-003009 (The mixed inertia criterion). Assume (T), and let YB/BY_B/B be a supplied projective, flat, finitely presented nodal curve of relative dimension one with a semilinear GG-action. Then the following are equivalent:

  1. II acts identically on the entire scheme (YB)l(Y_B)_l.

  2. There is a projective, flat, finitely presented nodal OO-scheme YY of relative dimension one with a GG-equivariant isomorphism Y×OB≃YBY\times_O B\simeq Y_B, for the coefficient action.

This descent, with its comparison, is unique up to the unique compatible isomorphism. If a GG-linearized relatively ample invertible sheaf LB\mathcal L_B is supplied, it descends with its linearization to a relatively ample invertible sheaf on YY if and only if its entire special-fibre II-action is the identity. There is no requirement of identity for the residual HH-action.

Proof. Assume condition 1. Lemma H.0.1 constructs YAY_A and its semilinear HH-action. Equation (10) turns this action into an actual descent datum for Spec⁡A→Spec⁡O\operatorname{Spec}A\to\operatorname{Spec}O: the two pullbacks are identified on the components indexed by HH, and the group law is exactly the cocycle condition on the triple product. Since this base map is surjective finite locally free and YA/AY_A/A is projective, Stacks Tag 0CCJ makes the datum effective as an ordinary scheme Y/OY/O.

Properness and finite presentation descend by Stacks Tag 02L1 and Stacks Tag 02L0. Flatness and nodality of relative dimension one descend by Stacks Tag 0C5E. To retain projectivity, note that q:YA→Yq:Y_A\to Y is finite locally free and surjective, as a base change of Spec⁡A→Spec⁡O\operatorname{Spec}A\to\operatorname{Spec}O. The affine base and projectivity give an ample invertible sheaf on YAY_A. The finite locally free case of Stacks Tag 0BD4 gives an ample invertible sheaf on YY. Properness over the affine base now gives projectivity by Stacks Tag 0B45. No division by ∣H∣|H| occurs.

Combining Y×OA≃YAY\times_O A\simeq Y_A with (13) gives the required GG-equivariant comparison over BB. Conversely, in any such comparison the special fibre is Yk×klY_k\times_k l. The subgroup II acts trivially on ll by (9), and acts on the coefficient factor only. Hence it is the identity on the whole special scheme. This proves necessity.

Suppose now that LB\mathcal L_B has the stated residual II-identity. The preceding lemma gives an HH-linearized relatively ample LA\mathcal L_A on YAY_A. Equation (10), after base change to YY, makes this a genuine fpqc sheaf descent datum. Effectivity and full faithfulness in Stacks Tag 023T give a quasi-coherent sheaf L\mathcal L on YY with the specified pullback. By Stacks Tag 05B2 it is finite locally free; its rank is one because rank is unchanged by residue-field extension and qq is surjective. It is therefore invertible. Relative ampleness descends by Stacks Tag 0D2P. Conversely, a bundle pulled back from YY has identity II-action on its full special-fibre module, for the same coefficient reason as above. Thus the actual supplied bundle descends, with no further power.

The inertia comparison is unique by the cited invariant descent and isomorphism functoriality; unramified descent is fully faithful by Stacks Tag 02W0, and sheaf descent by Stacks Tag 023T. Any two final descents therefore have a unique compatible comparison. In particular the special scheme CC has the resulting full semilinear HH-descent datum over l/kl/k, and Yk×kl≃CY_k\times_k l\simeq C respects it. This datum, rather than identity of the HH-action or mere invariance of vertices, is what is used in the unramified step. ◻

Lemma. NAG-002794 (Closed subspace neighbourhood; F8RC-N). Let XX be an analytic space, Z↪XZ\hookrightarrow X a closed analytic subspace, and W⊂XW\subset X an analytic domain containing ∣Z∣|Z|. Then WW contains an ordinary open neighbourhood of ∣Z∣|Z|.

Proof. For z∈Zz\in Z, the domain universal property identifies Z×XWZ\times_XW with ZZ, since the entire map Z→XZ\to X factors through WW.

For the closed immersion into WW, one may check the pullback on actual affinoid domains of WW subordinate to affinoid charts of XX. On these charts the canonical finite tensor of the closed ideal quotient is the actual closed quotient. This verifies the closed immersion by the target-local criterion NAG-001653.

Both maps (Z,z)→(X,z)(Z,z)\to(X,z) and (Z,z)→(W,z)(Z,z)\to(W,z) are closed immersions of analytic germs. By NAG-001978, their maps on reductions are isomorphisms. Functoriality makes R(W,z)→R(X,z)\mathcal R(W,z)\to\mathcal R(X,z) an isomorphism.

The reduction functor NAG-001972 identifies the reduction of (W,z)(W,z) with an open subspace of the reduction of (X,z)(X,z). The isomorphism above makes this subspace the whole reduction. By the subdomain classification NAG-001973, (W,z)(W,z) and (X,z)(X,z) are the same subdomain germ of (X,z)(X,z). Thus some ordinary open neighbourhood of zz in XX lies in WW. Taking the union over zz proves the assertion. ◻

Lemma. (The ordinary moment-image tube). Let V=M(A)V=\mathcal M(A) be affinoid over the specified complete non-Archimedean field, let f∈Af\in A, and let gg be a finite unit tuple on VV. Suppose the nonempty actual zero space D=M(A/(f))D=\mathcal M(A/(f)) has analytic dimension at most dd, and put P0=tg(D)P_0=t_g(D). Then for some ρ>0\rho>0, tg({∣f∣<ρ})⊆P0.\begin{equation} t_g(\{|f|<\rho\})\subseteq P_0. \end{equation} If DD is empty, a sufficiently small tube is empty. Arbitrary positive affinoid radii and the trivial absolute value are allowed.

Proof. By NAG-002358, P0P_0 is a compact finite integral affine polyhedral set of dimension at most dd. The inverse image of each compact cell is an actual affinoid Laurent domain: integral affine inequalities become bounds on monomials in the units gig_i, with positive real radii. Their finite union W0=tg−1(P0)W_0=t_g^{-1}(P_0) is an analytic domain, obtained from its finite quasi-net of affinoid pieces along their actual intersections by the domain calculus. It contains the closed subspace DD. Lemma H.0.3 gives an ordinary neighbourhood OO of DD contained in W0W_0. On the compact complement V∖OV\setminus O, ∣f∣|f| has a positive minimum, if the complement is nonempty. This proves (H.0.1) for a sufficiently small positive ρ\rho.

If DD is empty and VV is nonempty, ∣f∣|f| has a positive minimum on the compact VV. If VV is empty, every positive ρ\rho works. ◻

Proposition. NAG-002806 (Fixed-test logarithmic integrability; F8RC-L). Let XX be good, Hausdorff, and pure of dimension nn, and let ff be a regular analytic function on an open neighbourhood of the compact support of an ordinary smooth form β\beta. If one of its bidegrees is nn, then β\beta vanishes on an ordinary neighbourhood of {f=0}\{f=0\}. In particular, for an ordinary compact (n,n)(n,n)-test α\alpha, the pairing Lf(α)=∫X∖{f=0}log⁡∣f∣ αL_f(\alpha)=\int_{X\setminus\{f=0\}}\log|f|\,\alpha is well-defined by the frozen ordinary compact integral. For every fixed such test, bounded continuous truncations agreeing with log⁡∣f∣\log|f| above −R-R give the same value for all sufficiently large RR. If η\eta is ordinary smooth of bidegree (n−1,n−1)(n-1,n-1), then d′′ηd''\eta and d′d′′ηd'd''\eta vanish near {f=0}\{f=0\}.

Proof. Work on the open set where ff is given. At a zero outside supp⁡β\operatorname{supp}\beta the form already vanishes nearby. At a zero in the support, choose a common ordinary affinoid neighbourhood VV on which ff is regular and β\beta has one unit-moment representative tg∗γt_g^*\gamma, using the ordinary moment sheaf. The affinoid is pure of dimension nn by domain purity. Regularity avoids every minimal prime. Apply dimension drop to the nonzero image of ff on each integral ambient component. The component zero spaces cover the support of the full actual zero quotient D=M(O(V)/(f))D=\mathcal M(\mathcal O(V)/(f)), so its dimension is at most n−1n-1; for n=0n=0 it is empty. The ordinary moment-image tube for this actual quotient puts the gg-image of a sufficiently small strict sublevel of ∣f∣|f| in tg(D)t_g(D), of dimension at most n−1n-1 when nonempty. Intersect that sublevel with the ordinary interior of VV in XX. The restriction of γ\gamma to every face is zero when one exterior degree is nn. Thus its ordinary pullback is zero on this neighbourhood. This is an assertion about the actual ordinary germ and its representative, not just about a vanishing integral. It is local at every point of the zero scheme.

For compactly supported α\alpha, its closed support is therefore disjoint from the zero scheme. On this compact support ∣f∣|f| has a positive lower bound and a finite upper bound. The logarithm is bounded there and is an ordinary smooth function where ff is a unit. Adjoin the unit ff to the chart tuple on the unit locus near the compact support. Multiplication gives a compact ordinary form, extended by zero across the neighbourhood where α\alpha is zero. Its integral is the displayed value by the ordinary compact integral. Choose RR so that log⁡∣f∣>−R\log|f|>-R on this support. Then the truncation agrees with the smooth unit logarithm on a neighbourhood of the support. The product agrees as an ordinary compact form and is zero where the fixed test is zero. The final assertions follow by applying the same argument to d′′ηd''\eta and d′d′′ηd'd''\eta, whose bidegrees are (n−1,n)(n-1,n) and (n,n)(n,n), on the open set where they and ff are defined. Neither S1S_1 nor purity of the zero scheme is needed for this proposition. ◻